Merge git://git.kernel.org/pub/scm/linux/kernel/git/paulus/powerpc-merge

60 files changed, 13360 insertions, 1385 deletions

diff --git a/arch/powerpc/platforms/cell/Kconfig b/arch/powerpc/platforms/cell/Kconfig
new file mode 100644
index 000000000000..3157071e241c
--- /dev/null
+++ b/arch/powerpc/platforms/cell/Kconfig
@@ -0,0 +1,13 @@
+menu "Cell Broadband Engine options"
+	depends on PPC_CELL
+
+config SPU_FS
+	tristate "SPU file system"
+	default m
+	depends on PPC_CELL
+	help
+	  The SPU file system is used to access Synergistic Processing
+	  Units on machines implementing the Broadband Processor
+	  Architecture.
+
+endmenu
diff --git a/arch/powerpc/platforms/cell/Makefile b/arch/powerpc/platforms/cell/Makefile
index 55e094b96bc0..16031b565be4 100644
--- a/arch/powerpc/platforms/cell/Makefile
+++ b/arch/powerpc/platforms/cell/Makefile
@@ -1,2 +1,10 @@
 obj-y			+= interrupt.o iommu.o setup.o spider-pic.o
+obj-y			+= pervasive.o
+
 obj-$(CONFIG_SMP)	+= smp.o
+obj-$(CONFIG_SPU_FS)	+= spufs/ spu-base.o
+
+spu-base-y		+= spu_base.o spu_priv1.o
+
+builtin-spufs-$(CONFIG_SPU_FS)	+= spu_syscalls.o
+obj-y			+= $(builtin-spufs-m)
diff --git a/arch/powerpc/platforms/cell/interrupt.c b/arch/powerpc/platforms/cell/interrupt.c
index 7fbe78a9327d..63aa52acf441 100644
--- a/arch/powerpc/platforms/cell/interrupt.c
+++ b/arch/powerpc/platforms/cell/interrupt.c
@@ -23,6 +23,7 @@
 #include <linux/config.h>
 #include <linux/interrupt.h>
 #include <linux/irq.h>
+#include <linux/module.h>
 #include <linux/percpu.h>
 #include <linux/types.h>
 
@@ -55,6 +56,7 @@ struct iic_regs {
 
 struct iic {
 	struct iic_regs __iomem *regs;
+	u8 target_id;
 };
 
 static DEFINE_PER_CPU(struct iic, iic);
@@ -172,12 +174,11 @@ int iic_get_irq(struct pt_regs *regs)
 	return irq;
 }
 
-static struct iic_regs __iomem *find_iic(int cpu)
+static int setup_iic(int cpu, struct iic *iic)
 {
 	struct device_node *np;
 	int nodeid = cpu / 2;
 	unsigned long regs;
-	struct iic_regs __iomem *iic_regs;
 
 	for (np = of_find_node_by_type(NULL, "cpu");
 	     np;
@@ -188,20 +189,23 @@ static struct iic_regs __iomem *find_iic(int cpu)
 
 	if (!np) {
 		printk(KERN_WARNING "IIC: CPU %d not found\n", cpu);
-		iic_regs = NULL;
-	} else {
-		regs = *(long *)get_property(np, "iic", NULL);
-
-		/* hack until we have decided on the devtree info */
-		regs += 0x400;
-		if (cpu & 1)
-			regs += 0x20;
-
-		printk(KERN_DEBUG "IIC for CPU %d at %lx\n", cpu, regs);
-		iic_regs = __ioremap(regs, sizeof(struct iic_regs),
-						 _PAGE_NO_CACHE);
+		iic->regs = NULL;
+		iic->target_id = 0xff;
+		return -ENODEV;
 	}
-	return iic_regs;
+
+	regs = *(long *)get_property(np, "iic", NULL);
+
+	/* hack until we have decided on the devtree info */
+	regs += 0x400;
+	if (cpu & 1)
+		regs += 0x20;
+
+	printk(KERN_DEBUG "IIC for CPU %d at %lx\n", cpu, regs);
+	iic->regs = __ioremap(regs, sizeof(struct iic_regs),
+					 _PAGE_NO_CACHE);
+	iic->target_id = (nodeid << 4) + ((cpu & 1) ? 0xf : 0xe);
+	return 0;
 }
 
 #ifdef CONFIG_SMP
@@ -227,6 +231,12 @@ void iic_cause_IPI(int cpu, int mesg)
 	out_be64(&per_cpu(iic, cpu).regs->generate, (IIC_NUM_IPIS - 1 - mesg) << 4);
 }
 
+u8 iic_get_target_id(int cpu)
+{
+	return per_cpu(iic, cpu).target_id;
+}
+EXPORT_SYMBOL_GPL(iic_get_target_id);
+
 static irqreturn_t iic_ipi_action(int irq, void *dev_id, struct pt_regs *regs)
 {
 	smp_message_recv(iic_irq_to_ipi(irq), regs);
@@ -276,7 +286,7 @@ void iic_init_IRQ(void)
 	irq_offset = 0;
 	for_each_cpu(cpu) {
 		iic = &per_cpu(iic, cpu);
-		iic->regs = find_iic(cpu);
+		setup_iic(cpu, iic);
 		if (iic->regs)
 			out_be64(&iic->regs->prio, 0xff);
 	}
diff --git a/arch/powerpc/platforms/cell/interrupt.h b/arch/powerpc/platforms/cell/interrupt.h
index 37d58e6fd0c6..a14bd38791c0 100644
--- a/arch/powerpc/platforms/cell/interrupt.h
+++ b/arch/powerpc/platforms/cell/interrupt.h
@@ -54,6 +54,7 @@ extern void iic_setup_cpu(void);
 extern void iic_local_enable(void);
 extern void iic_local_disable(void);
 
+extern u8 iic_get_target_id(int cpu);
 
 extern void spider_init_IRQ(void);
 extern int spider_get_irq(unsigned long int_pending);
diff --git a/arch/powerpc/platforms/cell/iommu.c b/arch/powerpc/platforms/cell/iommu.c
index 74f999b4ac9e..46e7cb9c3e64 100644
--- a/arch/powerpc/platforms/cell/iommu.c
+++ b/arch/powerpc/platforms/cell/iommu.c
@@ -29,6 +29,8 @@
 #include <linux/bootmem.h>
 #include <linux/mm.h>
 #include <linux/dma-mapping.h>
+#include <linux/kernel.h>
+#include <linux/compiler.h>
 
 #include <asm/sections.h>
 #include <asm/iommu.h>
@@ -40,6 +42,7 @@
 #include <asm/abs_addr.h>
 #include <asm/system.h>
 #include <asm/ppc-pci.h>
+#include <asm/udbg.h>
 
 #include "iommu.h"
 
@@ -220,8 +223,6 @@ set_iopt_cache(void __iomem *base, unsigned long index,
 {
 	unsigned long __iomem *tags = base + IOC_PT_CACHE_DIR;
 	unsigned long __iomem *p = base + IOC_PT_CACHE_REG;
-	pr_debug("iopt %02lx was v%016lx/t%016lx, store v%016lx/t%016lx\n",
-		index, get_iopt_cache(base, index, &oldtag), oldtag, val, tag);
 
 	out_be64(p, val);
 	out_be64(&tags[index], tag);
@@ -248,67 +249,176 @@ set_iocmd_config(void __iomem *base)
 	out_be64(p, conf | IOCMD_CONF_TE);
 }
 
-/* FIXME: get these from the device tree */
-#define ioc_base	0x20000511000ull
-#define ioc_mmio_base	0x20000510000ull
-#define ioid		0x48a
-#define iopt_phys_offset (- 0x20000000) /* We have a 512MB offset from the SB */
-#define io_page_size	0x1000000
-
-static unsigned long map_iopt_entry(unsigned long address)
+static void enable_mapping(void __iomem *base, void __iomem *mmio_base)
 {
-	switch (address >> 20) {
-	case 0x600:
-		address = 0x24020000000ull; /* spider i/o */
-		break;
-	default:
-		address += iopt_phys_offset;
-		break;
-	}
-
-	return get_iopt_entry(address, ioid, IOPT_PROT_RW);
+	set_iocmd_config(base);
+	set_iost_origin(mmio_base);
 }
 
-static void iommu_bus_setup_null(struct pci_bus *b) { }
 static void iommu_dev_setup_null(struct pci_dev *d) { }
+static void iommu_bus_setup_null(struct pci_bus *b) { }
+
+struct cell_iommu {
+	unsigned long base;
+	unsigned long mmio_base;
+	void __iomem *mapped_base;
+	void __iomem *mapped_mmio_base;
+};
+
+static struct cell_iommu cell_iommus[NR_CPUS];
 
 /* initialize the iommu to support a simple linear mapping
  * for each DMA window used by any device. For now, we
  * happen to know that there is only one DMA window in use,
  * starting at iopt_phys_offset. */
-static void cell_map_iommu(void)
+static void cell_do_map_iommu(struct cell_iommu *iommu,
+			      unsigned int ioid,
+			      unsigned long map_start,
+			      unsigned long map_size)
 {
-	unsigned long address;
-	void __iomem *base;
+	unsigned long io_address, real_address;
+	void __iomem *ioc_base, *ioc_mmio_base;
 	ioste ioste;
 	unsigned long index;
 
-	base = __ioremap(ioc_base, 0x1000, _PAGE_NO_CACHE);
-	pr_debug("%lx mapped to %p\n", ioc_base, base);
-	set_iocmd_config(base);
-	iounmap(base);
+	/* we pretend the io page table was at a very high address */
+	const unsigned long fake_iopt = 0x10000000000ul;
+	const unsigned long io_page_size = 0x1000000; /* use 16M pages */
+	const unsigned long io_segment_size = 0x10000000; /* 256M */
+
+	ioc_base = iommu->mapped_base;
+	ioc_mmio_base = iommu->mapped_mmio_base;
+
+	for (real_address = 0, io_address = 0;
+	     io_address <= map_start + map_size;
+	     real_address += io_page_size, io_address += io_page_size) {
+		ioste = get_iost_entry(fake_iopt, io_address, io_page_size);
+		if ((real_address % io_segment_size) == 0) /* segment start */
+			set_iost_cache(ioc_mmio_base,
+				       io_address >> 28, ioste);
+		index = get_ioc_hash_1way(ioste, io_address);
+		pr_debug("addr %08lx, index %02lx, ioste %016lx\n",
+					 io_address, index, ioste.val);
+		set_iopt_cache(ioc_mmio_base,
+			get_ioc_hash_1way(ioste, io_address),
+			get_ioc_tag(ioste, io_address),
+			get_iopt_entry(real_address-map_start, ioid, IOPT_PROT_RW));
+	}
+}
 
-	base = __ioremap(ioc_mmio_base, 0x1000, _PAGE_NO_CACHE);
-	pr_debug("%lx mapped to %p\n", ioc_mmio_base, base);
+static void iommu_devnode_setup(struct device_node *d)
+{
+	unsigned int *ioid;
+	unsigned long *dma_window, map_start, map_size, token;
+	struct cell_iommu *iommu;
 
-	set_iost_origin(base);
+	ioid = (unsigned int *)get_property(d, "ioid", NULL);
+	if (!ioid)
+		pr_debug("No ioid entry found !\n");
 
-	for (address = 0; address < 0x100000000ul; address += io_page_size) {
-		ioste = get_iost_entry(0x10000000000ul, address, io_page_size);
-		if ((address & 0xfffffff) == 0) /* segment start */
-			set_iost_cache(base, address >> 28, ioste);
-		index = get_ioc_hash_1way(ioste, address);
-		pr_debug("addr %08lx, index %02lx, ioste %016lx\n",
-					 address, index, ioste.val);
-		set_iopt_cache(base,
-			get_ioc_hash_1way(ioste, address),
-			get_ioc_tag(ioste, address),
-			map_iopt_entry(address));
-	}
-	iounmap(base);
+	dma_window = (unsigned long *)get_property(d, "ibm,dma-window", NULL);
+	if (!dma_window)
+		pr_debug("No ibm,dma-window entry found !\n");
+
+	map_start = dma_window[1];
+	map_size = dma_window[2];
+	token = dma_window[0] >> 32;
+
+	iommu = &cell_iommus[token];
+
+	cell_do_map_iommu(iommu, *ioid, map_start, map_size);
+}
+
+static void iommu_bus_setup(struct pci_bus *b)
+{
+	struct device_node *d = (struct device_node *)b->sysdata;
+	iommu_devnode_setup(d);
+}
+
+
+static int cell_map_iommu_hardcoded(int num_nodes)
+{
+	struct cell_iommu *iommu = NULL;
+
+	pr_debug("%s(%d): Using hardcoded defaults\n", __FUNCTION__, __LINE__);
+
+	/* node 0 */
+	iommu = &cell_iommus[0];
+	iommu->mapped_base = __ioremap(0x20000511000, 0x1000, _PAGE_NO_CACHE);
+	iommu->mapped_mmio_base = __ioremap(0x20000510000, 0x1000, _PAGE_NO_CACHE);
+
+	enable_mapping(iommu->mapped_base, iommu->mapped_mmio_base);
+
+	cell_do_map_iommu(iommu, 0x048a,
+			  0x20000000ul,0x20000000ul);
+
+	if (num_nodes < 2)
+		return 0;
+
+	/* node 1 */
+	iommu = &cell_iommus[1];
+	iommu->mapped_base = __ioremap(0x30000511000, 0x1000, _PAGE_NO_CACHE);
+	iommu->mapped_mmio_base = __ioremap(0x30000510000, 0x1000, _PAGE_NO_CACHE);
+
+	enable_mapping(iommu->mapped_base, iommu->mapped_mmio_base);
+
+	cell_do_map_iommu(iommu, 0x048a,
+			  0x20000000,0x20000000ul);
+
+	return 0;
 }
 
 
+static int cell_map_iommu(void)
+{
+	unsigned int num_nodes = 0, *node_id;
+	unsigned long *base, *mmio_base;
+	struct device_node *dn;
+	struct cell_iommu *iommu = NULL;
+
+	/* determine number of nodes (=iommus) */
+	pr_debug("%s(%d): determining number of nodes...", __FUNCTION__, __LINE__);
+	for(dn = of_find_node_by_type(NULL, "cpu");
+	    dn;
+	    dn = of_find_node_by_type(dn, "cpu")) {
+		node_id = (unsigned int *)get_property(dn, "node-id", NULL);
+
+		if (num_nodes < *node_id)
+			num_nodes = *node_id;
+		}
+
+	num_nodes++;
+	pr_debug("%i found.\n", num_nodes);
+
+	/* map the iommu registers for each node */
+	pr_debug("%s(%d): Looping through nodes\n", __FUNCTION__, __LINE__);
+	for(dn = of_find_node_by_type(NULL, "cpu");
+	    dn;
+	    dn = of_find_node_by_type(dn, "cpu")) {
+
+		node_id = (unsigned int *)get_property(dn, "node-id", NULL);
+		base = (unsigned long *)get_property(dn, "ioc-cache", NULL);
+		mmio_base = (unsigned long *)get_property(dn, "ioc-translation", NULL);
+
+		if (!base || !mmio_base || !node_id)
+			return cell_map_iommu_hardcoded(num_nodes);
+
+		iommu = &cell_iommus[*node_id];
+		iommu->base = *base;
+		iommu->mmio_base = *mmio_base;
+
+		iommu->mapped_base = __ioremap(*base, 0x1000, _PAGE_NO_CACHE);
+		iommu->mapped_mmio_base = __ioremap(*mmio_base, 0x1000, _PAGE_NO_CACHE);
+
+		enable_mapping(iommu->mapped_base,
+			       iommu->mapped_mmio_base);
+
+		/* everything else will be done in iommu_bus_setup */
+	}
+
+	return 1;
+}
+
 static void *cell_alloc_coherent(struct device *hwdev, size_t size,
 			   dma_addr_t *dma_handle, gfp_t flag)
 {
@@ -365,11 +475,28 @@ static int cell_dma_supported(struct device *dev, u64 mask)
 
 void cell_init_iommu(void)
 {
-	cell_map_iommu();
-
-	/* Direct I/O, IOMMU off */
-	ppc_md.iommu_dev_setup = iommu_dev_setup_null;
-	ppc_md.iommu_bus_setup = iommu_bus_setup_null;
+	int setup_bus = 0;
+
+	if (of_find_node_by_path("/mambo")) {
+		pr_info("Not using iommu on systemsim\n");
+	} else {
+
+		if (!(of_chosen &&
+		      get_property(of_chosen, "linux,iommu-off", NULL)))
+			setup_bus = cell_map_iommu();
+
+		if (setup_bus) {
+			pr_debug("%s: IOMMU mapping activated\n", __FUNCTION__);
+			ppc_md.iommu_dev_setup = iommu_dev_setup_null;
+			ppc_md.iommu_bus_setup = iommu_bus_setup;
+		} else {
+			pr_debug("%s: IOMMU mapping activated, "
+				 "no device action necessary\n", __FUNCTION__);
+			/* Direct I/O, IOMMU off */
+			ppc_md.iommu_dev_setup = iommu_dev_setup_null;
+			ppc_md.iommu_bus_setup = iommu_bus_setup_null;
+		}
+	}
 
 	pci_dma_ops.alloc_coherent = cell_alloc_coherent;
 	pci_dma_ops.free_coherent = cell_free_coherent;
diff --git a/arch/powerpc/platforms/cell/pervasive.c b/arch/powerpc/platforms/cell/pervasive.c
new file mode 100644
index 000000000000..85152544c153
--- /dev/null
+++ b/arch/powerpc/platforms/cell/pervasive.c
@@ -0,0 +1,229 @@
+/*
+ * CBE Pervasive Monitor and Debug
+ *
+ * (C) Copyright IBM Corporation 2005
+ *
+ * Authors: Maximino Aguilar (maguilar@us.ibm.com)
+ *          Michael N. Day (mnday@us.ibm.com)
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; either version 2, or (at your option)
+ * any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software
+ * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
+ */
+
+#undef DEBUG
+
+#include <linux/config.h>
+#include <linux/interrupt.h>
+#include <linux/irq.h>
+#include <linux/percpu.h>
+#include <linux/types.h>
+#include <linux/kallsyms.h>
+
+#include <asm/io.h>
+#include <asm/machdep.h>
+#include <asm/prom.h>
+#include <asm/pgtable.h>
+#include <asm/reg.h>
+
+#include "pervasive.h"
+
+static DEFINE_SPINLOCK(cbe_pervasive_lock);
+struct cbe_pervasive {
+	struct pmd_regs __iomem *regs;
+	unsigned int thread;
+};
+
+/* can't use per_cpu from setup_arch */
+static struct cbe_pervasive cbe_pervasive[NR_CPUS];
+
+static void __init cbe_enable_pause_zero(void)
+{
+	unsigned long thread_switch_control;
+	unsigned long temp_register;
+	struct cbe_pervasive *p;
+	int thread;
+
+	spin_lock_irq(&cbe_pervasive_lock);
+	p = &cbe_pervasive[smp_processor_id()];
+
+	if (!cbe_pervasive->regs)
+		goto out;
+
+	pr_debug("Power Management: CPU %d\n", smp_processor_id());
+
+	 /* Enable Pause(0) control bit */
+	temp_register = in_be64(&p->regs->pm_control);
+
+	out_be64(&p->regs->pm_control,
+		 temp_register|PMD_PAUSE_ZERO_CONTROL);
+
+	/* Enable DEC and EE interrupt request */
+	thread_switch_control  = mfspr(SPRN_TSC_CELL);
+	thread_switch_control |= TSC_CELL_EE_ENABLE | TSC_CELL_EE_BOOST;
+
+	switch ((mfspr(SPRN_CTRLF) & CTRL_CT)) {
+	case CTRL_CT0:
+		thread_switch_control |= TSC_CELL_DEC_ENABLE_0;
+		thread = 0;
+		break;
+	case CTRL_CT1:
+		thread_switch_control |= TSC_CELL_DEC_ENABLE_1;
+		thread = 1;
+		break;
+	default:
+		printk(KERN_WARNING "%s: unknown configuration\n",
+			__FUNCTION__);
+		thread = -1;
+		break;
+	}
+
+	if (p->thread != thread)
+		printk(KERN_WARNING "%s: device tree inconsistant, "
+				     "cpu %i: %d/%d\n", __FUNCTION__,
+				     smp_processor_id(),
+				     p->thread, thread);
+
+	mtspr(SPRN_TSC_CELL, thread_switch_control);
+
+out:
+	spin_unlock_irq(&cbe_pervasive_lock);
+}
+
+static void cbe_idle(void)
+{
+	unsigned long ctrl;
+
+	cbe_enable_pause_zero();
+
+	while (1) {
+		if (!need_resched()) {
+			local_irq_disable();
+			while (!need_resched()) {
+				/* go into low thread priority */
+				HMT_low();
+
+				/*
+				 * atomically disable thread execution
+				 * and runlatch.
+				 * External and Decrementer exceptions
+				 * are still handled when the thread
+				 * is disabled but now enter in
+				 * cbe_system_reset_exception()
+				 */
+				ctrl = mfspr(SPRN_CTRLF);
+				ctrl &= ~(CTRL_RUNLATCH | CTRL_TE);
+				mtspr(SPRN_CTRLT, ctrl);
+			}
+			/* restore thread prio */
+			HMT_medium();
+			local_irq_enable();
+		}
+
+		/*
+		 * turn runlatch on again before scheduling the
+		 * process we just woke up
+		 */
+		ppc64_runlatch_on();
+
+		preempt_enable_no_resched();
+		schedule();
+		preempt_disable();
+	}
+}
+
+int cbe_system_reset_exception(struct pt_regs *regs)
+{
+	switch (regs->msr & SRR1_WAKEMASK) {
+	case SRR1_WAKEEE:
+		do_IRQ(regs);
+		break;
+	case SRR1_WAKEDEC:
+		timer_interrupt(regs);
+		break;
+	case SRR1_WAKEMT:
+		/* no action required */
+		break;
+	default:
+		/* do system reset */
+		return 0;
+	}
+	/* everything handled */
+	return 1;
+}
+
+static int __init cbe_find_pmd_mmio(int cpu, struct cbe_pervasive *p)
+{
+	struct device_node *node;
+	unsigned int *int_servers;
+	char *addr;
+	unsigned long real_address;
+	unsigned int size;
+
+	struct pmd_regs __iomem *pmd_mmio_area;
+	int hardid, thread;
+	int proplen;
+
+	pmd_mmio_area = NULL;
+	hardid = get_hard_smp_processor_id(cpu);
+	for (node = NULL; (node = of_find_node_by_type(node, "cpu"));) {
+		int_servers = (void *) get_property(node,
+				"ibm,ppc-interrupt-server#s", &proplen);
+		if (!int_servers) {
+			printk(KERN_WARNING "%s misses "
+				"ibm,ppc-interrupt-server#s property",
+				node->full_name);
+			continue;
+		}
+		for (thread = 0; thread < proplen / sizeof (int); thread++) {
+			if (hardid == int_servers[thread]) {
+				addr = get_property(node, "pervasive", NULL);
+				goto found;
+			}
+		}
+	}
+
+	printk(KERN_WARNING "%s: CPU %d not found\n", __FUNCTION__, cpu);
+	return -EINVAL;
+
+found:
+	real_address = *(unsigned long*) addr;
+	addr += sizeof (unsigned long);
+	size = *(unsigned int*) addr;
+
+	pr_debug("pervasive area for CPU %d at %lx, size %x\n",
+			cpu, real_address, size);
+	p->regs = __ioremap(real_address, size, _PAGE_NO_CACHE);
+	p->thread = thread;
+	return 0;
+}
+
+void __init cell_pervasive_init(void)
+{
+	struct cbe_pervasive *p;
+	int cpu;
+	int ret;
+
+	if (!cpu_has_feature(CPU_FTR_PAUSE_ZERO))
+		return;
+
+	for_each_cpu(cpu) {
+		p = &cbe_pervasive[cpu];
+		ret = cbe_find_pmd_mmio(cpu, p);
+		if (ret)
+			return;
+	}
+
+	ppc_md.idle_loop = cbe_idle;
+	ppc_md.system_reset_exception = cbe_system_reset_exception;
+}
diff --git a/arch/powerpc/platforms/cell/pervasive.h b/arch/powerpc/platforms/cell/pervasive.h
new file mode 100644
index 000000000000..da1fb85ca3e8
--- /dev/null
+++ b/arch/powerpc/platforms/cell/pervasive.h
@@ -0,0 +1,62 @@
+/*
+ * Cell Pervasive Monitor and Debug interface and HW structures
+ *
+ * (C) Copyright IBM Corporation 2005
+ *
+ * Authors: Maximino Aguilar (maguilar@us.ibm.com)
+ *          David J. Erb (djerb@us.ibm.com)
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; either version 2, or (at your option)
+ * any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software
+ * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
+ */
+
+
+#ifndef PERVASIVE_H
+#define PERVASIVE_H
+
+struct pmd_regs {
+	u8 pad_0x0000_0x0800[0x0800 - 0x0000];			/* 0x0000 */
+
+	/* Thermal Sensor Registers */
+	u64  ts_ctsr1;						/* 0x0800 */
+	u64  ts_ctsr2;						/* 0x0808 */
+	u64  ts_mtsr1;						/* 0x0810 */
+	u64  ts_mtsr2;						/* 0x0818 */
+	u64  ts_itr1;						/* 0x0820 */
+	u64  ts_itr2;						/* 0x0828 */
+	u64  ts_gitr;						/* 0x0830 */
+	u64  ts_isr;						/* 0x0838 */
+	u64  ts_imr;						/* 0x0840 */
+	u64  tm_cr1;						/* 0x0848 */
+	u64  tm_cr2;						/* 0x0850 */
+	u64  tm_simr;						/* 0x0858 */
+	u64  tm_tpr;						/* 0x0860 */
+	u64  tm_str1;						/* 0x0868 */
+	u64  tm_str2;						/* 0x0870 */
+	u64  tm_tsr;						/* 0x0878 */
+
+	/* Power Management */
+	u64  pm_control;					/* 0x0880 */
+#define PMD_PAUSE_ZERO_CONTROL		0x10000
+	u64  pm_status;						/* 0x0888 */
+
+	/* Time Base Register */
+	u64  tbr;						/* 0x0890 */
+
+	u8   pad_0x0898_0x1000 [0x1000 - 0x0898];		/* 0x0898 */
+};
+
+void __init cell_pervasive_init(void);
+
+#endif
diff --git a/arch/powerpc/platforms/cell/setup.c b/arch/powerpc/platforms/cell/setup.c
index 9a495634d0c2..18e25e65c04b 100644
--- a/arch/powerpc/platforms/cell/setup.c
+++ b/arch/powerpc/platforms/cell/setup.c
@@ -33,6 +33,7 @@
 #include <asm/mmu.h>
 #include <asm/processor.h>
 #include <asm/io.h>
+#include <asm/kexec.h>
 #include <asm/pgtable.h>
 #include <asm/prom.h>
 #include <asm/rtas.h>
@@ -48,6 +49,7 @@
 
 #include "interrupt.h"
 #include "iommu.h"
+#include "pervasive.h"
 
 #ifdef DEBUG
 #define DBG(fmt...) udbg_printf(fmt)
@@ -67,6 +69,77 @@ void cell_show_cpuinfo(struct seq_file *m)
 	of_node_put(root);
 }
 
+#ifdef CONFIG_SPARSEMEM
+static int __init find_spu_node_id(struct device_node *spe)
+{
+	unsigned int *id;
+#ifdef CONFIG_NUMA
+	struct device_node *cpu;
+	cpu = spe->parent->parent;
+	id = (unsigned int *)get_property(cpu, "node-id", NULL);
+#else
+	id = NULL;
+#endif
+	return id ? *id : 0;
+}
+
+static void __init cell_spuprop_present(struct device_node *spe,
+				       const char *prop, int early)
+{
+	struct address_prop {
+		unsigned long address;
+		unsigned int len;
+	} __attribute__((packed)) *p;
+	int proplen;
+
+	unsigned long start_pfn, end_pfn, pfn;
+	int node_id;
+
+	p = (void*)get_property(spe, prop, &proplen);
+	WARN_ON(proplen != sizeof (*p));
+
+	node_id = find_spu_node_id(spe);
+
+	start_pfn = p->address >> PAGE_SHIFT;
+	end_pfn = (p->address + p->len + PAGE_SIZE - 1) >> PAGE_SHIFT;
+
+	/* We need to call memory_present *before* the call to sparse_init,
+	   but we can initialize the page structs only *after* that call.
+	   Thus, we're being called twice. */
+	if (early)
+		memory_present(node_id, start_pfn, end_pfn);
+	else {
+		/* As the pages backing SPU LS and I/O are outside the range
+		   of regular memory, their page structs were not initialized
+		   by free_area_init. Do it here instead. */
+		for (pfn = start_pfn; pfn < end_pfn; pfn++) {
+			struct page *page = pfn_to_page(pfn);
+			set_page_links(page, ZONE_DMA, node_id, pfn);
+			set_page_count(page, 1);
+			reset_page_mapcount(page);
+			SetPageReserved(page);
+			INIT_LIST_HEAD(&page->lru);
+		}
+	}
+}
+
+static void __init cell_spumem_init(int early)
+{
+	struct device_node *node;
+	for (node = of_find_node_by_type(NULL, "spe");
+			node; node = of_find_node_by_type(node, "spe")) {
+		cell_spuprop_present(node, "local-store", early);
+		cell_spuprop_present(node, "problem", early);
+		cell_spuprop_present(node, "priv1", early);
+		cell_spuprop_present(node, "priv2", early);
+	}
+}
+#else
+static void __init cell_spumem_init(int early)
+{
+}
+#endif
+
 static void cell_progress(char *s, unsigned short hex)
 {
 	printk("*** %04x : %s\n", hex, s ? s : "");
@@ -93,11 +166,14 @@ static void __init cell_setup_arch(void)
 	init_pci_config_tokens();
 	find_and_init_phbs();
 	spider_init_IRQ();
+	cell_pervasive_init();
 #ifdef CONFIG_DUMMY_CONSOLE
 	conswitchp = &dummy_con;
 #endif
 
 	mmio_nvram_init();
+
+	cell_spumem_init(0);
 }
 
 /*
@@ -113,6 +189,8 @@ static void __init cell_init_early(void)
 
 	ppc64_interrupt_controller = IC_CELL_PIC;
 
+	cell_spumem_init(1);
+
 	DBG(" <- cell_init_early()\n");
 }
 
@@ -125,6 +203,15 @@ static int __init cell_probe(int platform)
 	return 1;
 }
 
+/*
+ * Cell has no legacy IO; anything calling this function has to
+ * fail or bad things will happen
+ */
+static int cell_check_legacy_ioport(unsigned int baseport)
+{
+	return -ENODEV;
+}
+
 struct machdep_calls __initdata cell_md = {
 	.probe			= cell_probe,
 	.setup_arch		= cell_setup_arch,
@@ -137,5 +224,11 @@ struct machdep_calls __initdata cell_md = {
 	.get_rtc_time		= rtas_get_rtc_time,
 	.set_rtc_time		= rtas_set_rtc_time,
 	.calibrate_decr		= generic_calibrate_decr,
+	.check_legacy_ioport	= cell_check_legacy_ioport,
 	.progress		= cell_progress,
+#ifdef CONFIG_KEXEC
+	.machine_kexec		= default_machine_kexec,
+	.machine_kexec_prepare	= default_machine_kexec_prepare,
+	.machine_crash_shutdown	= default_machine_crash_shutdown,
+#endif
 };
diff --git a/arch/powerpc/platforms/cell/spu_base.c b/arch/powerpc/platforms/cell/spu_base.c
new file mode 100644
index 000000000000..d75ae03df686
--- /dev/null
+++ b/arch/powerpc/platforms/cell/spu_base.c
@@ -0,0 +1,711 @@
+/*
+ * Low-level SPU handling
+ *
+ * (C) Copyright IBM Deutschland Entwicklung GmbH 2005
+ *
+ * Author: Arnd Bergmann <arndb@de.ibm.com>
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; either version 2, or (at your option)
+ * any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software
+ * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
+ */
+
+#undef DEBUG
+
+#include <linux/interrupt.h>
+#include <linux/list.h>
+#include <linux/module.h>
+#include <linux/poll.h>
+#include <linux/ptrace.h>
+#include <linux/slab.h>
+#include <linux/wait.h>
+
+#include <asm/io.h>
+#include <asm/prom.h>
+#include <asm/semaphore.h>
+#include <asm/spu.h>
+#include <asm/mmu_context.h>
+
+#include "interrupt.h"
+
+static int __spu_trap_invalid_dma(struct spu *spu)
+{
+	pr_debug("%s\n", __FUNCTION__);
+	force_sig(SIGBUS, /* info, */ current);
+	return 0;
+}
+
+static int __spu_trap_dma_align(struct spu *spu)
+{
+	pr_debug("%s\n", __FUNCTION__);
+	force_sig(SIGBUS, /* info, */ current);
+	return 0;
+}
+
+static int __spu_trap_error(struct spu *spu)
+{
+	pr_debug("%s\n", __FUNCTION__);
+	force_sig(SIGILL, /* info, */ current);
+	return 0;
+}
+
+static void spu_restart_dma(struct spu *spu)
+{
+	struct spu_priv2 __iomem *priv2 = spu->priv2;
+
+	if (!test_bit(SPU_CONTEXT_SWITCH_PENDING, &spu->flags))
+		out_be64(&priv2->mfc_control_RW, MFC_CNTL_RESTART_DMA_COMMAND);
+}
+
+static int __spu_trap_data_seg(struct spu *spu, unsigned long ea)
+{
+	struct spu_priv2 __iomem *priv2 = spu->priv2;
+	struct mm_struct *mm = spu->mm;
+	u64 esid, vsid;
+
+	pr_debug("%s\n", __FUNCTION__);
+
+	if (test_bit(SPU_CONTEXT_SWITCH_ACTIVE, &spu->flags)) {
+		/* SLBs are pre-loaded for context switch, so
+		 * we should never get here!
+		 */
+		printk("%s: invalid access during switch!\n", __func__);
+		return 1;
+	}
+	if (!mm || (REGION_ID(ea) != USER_REGION_ID)) {
+		/* Future: support kernel segments so that drivers
+		 * can use SPUs.
+		 */
+		pr_debug("invalid region access at %016lx\n", ea);
+		return 1;
+	}
+
+	esid = (ea & ESID_MASK) | SLB_ESID_V;
+	vsid = (get_vsid(mm->context.id, ea) << SLB_VSID_SHIFT) | SLB_VSID_USER;
+	if (in_hugepage_area(mm->context, ea))
+		vsid |= SLB_VSID_L;
+
+	out_be64(&priv2->slb_index_W, spu->slb_replace);
+	out_be64(&priv2->slb_vsid_RW, vsid);
+	out_be64(&priv2->slb_esid_RW, esid);
+
+	spu->slb_replace++;
+	if (spu->slb_replace >= 8)
+		spu->slb_replace = 0;
+
+	spu_restart_dma(spu);
+
+	return 0;
+}
+
+extern int hash_page(unsigned long ea, unsigned long access, unsigned long trap); //XXX
+static int __spu_trap_data_map(struct spu *spu, unsigned long ea, u64 dsisr)
+{
+	pr_debug("%s\n", __FUNCTION__);
+
+	/* Handle kernel space hash faults immediately.
+	   User hash faults need to be deferred to process context. */
+	if ((dsisr & MFC_DSISR_PTE_NOT_FOUND)
+	    && REGION_ID(ea) != USER_REGION_ID
+	    && hash_page(ea, _PAGE_PRESENT, 0x300) == 0) {
+		spu_restart_dma(spu);
+		return 0;
+	}
+
+	if (test_bit(SPU_CONTEXT_SWITCH_ACTIVE, &spu->flags)) {
+		printk("%s: invalid access during switch!\n", __func__);
+		return 1;
+	}
+
+	spu->dar = ea;
+	spu->dsisr = dsisr;
+	mb();
+	if (spu->stop_callback)
+		spu->stop_callback(spu);
+	return 0;
+}
+
+static int __spu_trap_mailbox(struct spu *spu)
+{
+	if (spu->ibox_callback)
+		spu->ibox_callback(spu);
+
+	/* atomically disable SPU mailbox interrupts */
+	spin_lock(&spu->register_lock);
+	spu_int_mask_and(spu, 2, ~0x1);
+	spin_unlock(&spu->register_lock);
+	return 0;
+}
+
+static int __spu_trap_stop(struct spu *spu)
+{
+	pr_debug("%s\n", __FUNCTION__);
+	spu->stop_code = in_be32(&spu->problem->spu_status_R);
+	if (spu->stop_callback)
+		spu->stop_callback(spu);
+	return 0;
+}
+
+static int __spu_trap_halt(struct spu *spu)
+{
+	pr_debug("%s\n", __FUNCTION__);
+	spu->stop_code = in_be32(&spu->problem->spu_status_R);
+	if (spu->stop_callback)
+		spu->stop_callback(spu);
+	return 0;
+}
+
+static int __spu_trap_tag_group(struct spu *spu)
+{
+	pr_debug("%s\n", __FUNCTION__);
+	/* wake_up(&spu->dma_wq); */
+	return 0;
+}
+
+static int __spu_trap_spubox(struct spu *spu)
+{
+	if (spu->wbox_callback)
+		spu->wbox_callback(spu);
+
+	/* atomically disable SPU mailbox interrupts */
+	spin_lock(&spu->register_lock);
+	spu_int_mask_and(spu, 2, ~0x10);
+	spin_unlock(&spu->register_lock);
+	return 0;
+}
+
+static irqreturn_t
+spu_irq_class_0(int irq, void *data, struct pt_regs *regs)
+{
+	struct spu *spu;
+
+	spu = data;
+	spu->class_0_pending = 1;
+	if (spu->stop_callback)
+		spu->stop_callback(spu);
+
+	return IRQ_HANDLED;
+}
+
+int
+spu_irq_class_0_bottom(struct spu *spu)
+{
+	unsigned long stat, mask;
+
+	spu->class_0_pending = 0;
+
+	mask = spu_int_mask_get(spu, 0);
+	stat = spu_int_stat_get(spu, 0);
+
+	stat &= mask;
+
+	if (stat & 1) /* invalid MFC DMA */
+		__spu_trap_invalid_dma(spu);
+
+	if (stat & 2) /* invalid DMA alignment */
+		__spu_trap_dma_align(spu);
+
+	if (stat & 4) /* error on SPU */
+		__spu_trap_error(spu);
+
+	spu_int_stat_clear(spu, 0, stat);
+
+	return (stat & 0x7) ? -EIO : 0;
+}
+EXPORT_SYMBOL_GPL(spu_irq_class_0_bottom);
+
+static irqreturn_t
+spu_irq_class_1(int irq, void *data, struct pt_regs *regs)
+{
+	struct spu *spu;
+	unsigned long stat, mask, dar, dsisr;
+
+	spu = data;
+
+	/* atomically read & clear class1 status. */
+	spin_lock(&spu->register_lock);
+	mask  = spu_int_mask_get(spu, 1);
+	stat  = spu_int_stat_get(spu, 1) & mask;
+	dar   = spu_mfc_dar_get(spu);
+	dsisr = spu_mfc_dsisr_get(spu);
+	if (stat & 2) /* mapping fault */
+		spu_mfc_dsisr_set(spu, 0ul);
+	spu_int_stat_clear(spu, 1, stat);
+	spin_unlock(&spu->register_lock);
+
+	if (stat & 1) /* segment fault */
+		__spu_trap_data_seg(spu, dar);
+
+	if (stat & 2) { /* mapping fault */
+		__spu_trap_data_map(spu, dar, dsisr);
+	}
+
+	if (stat & 4) /* ls compare & suspend on get */
+		;
+
+	if (stat & 8) /* ls compare & suspend on put */
+		;
+
+	return stat ? IRQ_HANDLED : IRQ_NONE;
+}
+EXPORT_SYMBOL_GPL(spu_irq_class_1_bottom);
+
+static irqreturn_t
+spu_irq_class_2(int irq, void *data, struct pt_regs *regs)
+{
+	struct spu *spu;
+	unsigned long stat;
+	unsigned long mask;
+
+	spu = data;
+	stat = spu_int_stat_get(spu, 2);
+	mask = spu_int_mask_get(spu, 2);
+
+	pr_debug("class 2 interrupt %d, %lx, %lx\n", irq, stat, mask);
+
+	stat &= mask;
+
+	if (stat & 1)  /* PPC core mailbox */
+		__spu_trap_mailbox(spu);
+
+	if (stat & 2) /* SPU stop-and-signal */
+		__spu_trap_stop(spu);
+
+	if (stat & 4) /* SPU halted */
+		__spu_trap_halt(spu);
+
+	if (stat & 8) /* DMA tag group complete */
+		__spu_trap_tag_group(spu);
+
+	if (stat & 0x10) /* SPU mailbox threshold */
+		__spu_trap_spubox(spu);
+
+	spu_int_stat_clear(spu, 2, stat);
+	return stat ? IRQ_HANDLED : IRQ_NONE;
+}
+
+static int
+spu_request_irqs(struct spu *spu)
+{
+	int ret;
+	int irq_base;
+
+	irq_base = IIC_NODE_STRIDE * spu->node + IIC_SPE_OFFSET;
+
+	snprintf(spu->irq_c0, sizeof (spu->irq_c0), "spe%02d.0", spu->number);
+	ret = request_irq(irq_base + spu->isrc,
+		 spu_irq_class_0, 0, spu->irq_c0, spu);
+	if (ret)
+		goto out;
+
+	snprintf(spu->irq_c1, sizeof (spu->irq_c1), "spe%02d.1", spu->number);
+	ret = request_irq(irq_base + IIC_CLASS_STRIDE + spu->isrc,
+		 spu_irq_class_1, 0, spu->irq_c1, spu);
+	if (ret)
+		goto out1;
+
+	snprintf(spu->irq_c2, sizeof (spu->irq_c2), "spe%02d.2", spu->number);
+	ret = request_irq(irq_base + 2*IIC_CLASS_STRIDE + spu->isrc,
+		 spu_irq_class_2, 0, spu->irq_c2, spu);
+	if (ret)
+		goto out2;
+	goto out;
+
+out2:
+	free_irq(irq_base + IIC_CLASS_STRIDE + spu->isrc, spu);
+out1:
+	free_irq(irq_base + spu->isrc, spu);
+out:
+	return ret;
+}
+
+static void
+spu_free_irqs(struct spu *spu)
+{
+	int irq_base;
+
+	irq_base = IIC_NODE_STRIDE * spu->node + IIC_SPE_OFFSET;
+
+	free_irq(irq_base + spu->isrc, spu);
+	free_irq(irq_base + IIC_CLASS_STRIDE + spu->isrc, spu);
+	free_irq(irq_base + 2*IIC_CLASS_STRIDE + spu->isrc, spu);
+}
+
+static LIST_HEAD(spu_list);
+static DECLARE_MUTEX(spu_mutex);
+
+static void spu_init_channels(struct spu *spu)
+{
+	static const struct {
+		 unsigned channel;
+		 unsigned count;
+	} zero_list[] = {
+		{ 0x00, 1, }, { 0x01, 1, }, { 0x03, 1, }, { 0x04, 1, },
+		{ 0x18, 1, }, { 0x19, 1, }, { 0x1b, 1, }, { 0x1d, 1, },
+	}, count_list[] = {
+		{ 0x00, 0, }, { 0x03, 0, }, { 0x04, 0, }, { 0x15, 16, },
+		{ 0x17, 1, }, { 0x18, 0, }, { 0x19, 0, }, { 0x1b, 0, },
+		{ 0x1c, 1, }, { 0x1d, 0, }, { 0x1e, 1, },
+	};
+	struct spu_priv2 __iomem *priv2;
+	int i;
+
+	priv2 = spu->priv2;
+
+	/* initialize all channel data to zero */
+	for (i = 0; i < ARRAY_SIZE(zero_list); i++) {
+		int count;
+
+		out_be64(&priv2->spu_chnlcntptr_RW, zero_list[i].channel);
+		for (count = 0; count < zero_list[i].count; count++)
+			out_be64(&priv2->spu_chnldata_RW, 0);
+	}
+
+	/* initialize channel counts to meaningful values */
+	for (i = 0; i < ARRAY_SIZE(count_list); i++) {
+		out_be64(&priv2->spu_chnlcntptr_RW, count_list[i].channel);
+		out_be64(&priv2->spu_chnlcnt_RW, count_list[i].count);
+	}
+}
+
+struct spu *spu_alloc(void)
+{
+	struct spu *spu;
+
+	down(&spu_mutex);
+	if (!list_empty(&spu_list)) {
+		spu = list_entry(spu_list.next, struct spu, list);
+		list_del_init(&spu->list);
+		pr_debug("Got SPU %x %d\n", spu->isrc, spu->number);
+	} else {
+		pr_debug("No SPU left\n");
+		spu = NULL;
+	}
+	up(&spu_mutex);
+
+	if (spu)
+		spu_init_channels(spu);
+
+	return spu;
+}
+EXPORT_SYMBOL_GPL(spu_alloc);
+
+void spu_free(struct spu *spu)
+{
+	down(&spu_mutex);
+	list_add_tail(&spu->list, &spu_list);
+	up(&spu_mutex);
+}
+EXPORT_SYMBOL_GPL(spu_free);
+
+static int spu_handle_mm_fault(struct spu *spu)
+{
+	struct mm_struct *mm = spu->mm;
+	struct vm_area_struct *vma;
+	u64 ea, dsisr, is_write;
+	int ret;
+
+	ea = spu->dar;
+	dsisr = spu->dsisr;
+#if 0
+	if (!IS_VALID_EA(ea)) {
+		return -EFAULT;
+	}
+#endif /* XXX */
+	if (mm == NULL) {
+		return -EFAULT;
+	}
+	if (mm->pgd == NULL) {
+		return -EFAULT;
+	}
+
+	down_read(&mm->mmap_sem);
+	vma = find_vma(mm, ea);
+	if (!vma)
+		goto bad_area;
+	if (vma->vm_start <= ea)
+		goto good_area;
+	if (!(vma->vm_flags & VM_GROWSDOWN))
+		goto bad_area;
+#if 0
+	if (expand_stack(vma, ea))
+		goto bad_area;
+#endif /* XXX */
+good_area:
+	is_write = dsisr & MFC_DSISR_ACCESS_PUT;
+	if (is_write) {
+		if (!(vma->vm_flags & VM_WRITE))
+			goto bad_area;
+	} else {
+		if (dsisr & MFC_DSISR_ACCESS_DENIED)
+			goto bad_area;
+		if (!(vma->vm_flags & (VM_READ | VM_EXEC)))
+			goto bad_area;
+	}
+	ret = 0;
+	switch (handle_mm_fault(mm, vma, ea, is_write)) {
+	case VM_FAULT_MINOR:
+		current->min_flt++;
+		break;
+	case VM_FAULT_MAJOR:
+		current->maj_flt++;
+		break;
+	case VM_FAULT_SIGBUS:
+		ret = -EFAULT;
+		goto bad_area;
+	case VM_FAULT_OOM:
+		ret = -ENOMEM;
+		goto bad_area;
+	default:
+		BUG();
+	}
+	up_read(&mm->mmap_sem);
+	return ret;
+
+bad_area:
+	up_read(&mm->mmap_sem);
+	return -EFAULT;
+}
+
+int spu_irq_class_1_bottom(struct spu *spu)
+{
+	u64 ea, dsisr, access, error = 0UL;
+	int ret = 0;
+
+	ea = spu->dar;
+	dsisr = spu->dsisr;
+	if (dsisr & MFC_DSISR_PTE_NOT_FOUND) {
+		access = (_PAGE_PRESENT | _PAGE_USER);
+		access |= (dsisr & MFC_DSISR_ACCESS_PUT) ? _PAGE_RW : 0UL;
+		if (hash_page(ea, access, 0x300) != 0)
+			error |= CLASS1_ENABLE_STORAGE_FAULT_INTR;
+	}
+	if ((error & CLASS1_ENABLE_STORAGE_FAULT_INTR) ||
+	    (dsisr & MFC_DSISR_ACCESS_DENIED)) {
+		if ((ret = spu_handle_mm_fault(spu)) != 0)
+			error |= CLASS1_ENABLE_STORAGE_FAULT_INTR;
+		else
+			error &= ~CLASS1_ENABLE_STORAGE_FAULT_INTR;
+	}
+	spu->dar = 0UL;
+	spu->dsisr = 0UL;
+	if (!error) {
+		spu_restart_dma(spu);
+	} else {
+		__spu_trap_invalid_dma(spu);
+	}
+	return ret;
+}
+
+void spu_irq_setaffinity(struct spu *spu, int cpu)
+{
+	u64 target = iic_get_target_id(cpu);
+	u64 route = target << 48 | target << 32 | target << 16;
+	spu_int_route_set(spu, route);
+}
+EXPORT_SYMBOL_GPL(spu_irq_setaffinity);
+
+static void __iomem * __init map_spe_prop(struct device_node *n,
+						 const char *name)
+{
+	struct address_prop {
+		unsigned long address;
+		unsigned int len;
+	} __attribute__((packed)) *prop;
+
+	void *p;
+	int proplen;
+
+	p = get_property(n, name, &proplen);
+	if (proplen != sizeof (struct address_prop))
+		return NULL;
+
+	prop = p;
+
+	return ioremap(prop->address, prop->len);
+}
+
+static void spu_unmap(struct spu *spu)
+{
+	iounmap(spu->priv2);
+	iounmap(spu->priv1);
+	iounmap(spu->problem);
+	iounmap((u8 __iomem *)spu->local_store);
+}
+
+static int __init spu_map_device(struct spu *spu, struct device_node *spe)
+{
+	char *prop;
+	int ret;
+
+	ret = -ENODEV;
+	prop = get_property(spe, "isrc", NULL);
+	if (!prop)
+		goto out;
+	spu->isrc = *(unsigned int *)prop;
+
+	spu->name = get_property(spe, "name", NULL);
+	if (!spu->name)
+		goto out;
+
+	prop = get_property(spe, "local-store", NULL);
+	if (!prop)
+		goto out;
+	spu->local_store_phys = *(unsigned long *)prop;
+
+	/* we use local store as ram, not io memory */
+	spu->local_store = (void __force *)map_spe_prop(spe, "local-store");
+	if (!spu->local_store)
+		goto out;
+
+	spu->problem= map_spe_prop(spe, "problem");
+	if (!spu->problem)
+		goto out_unmap;
+
+	spu->priv1= map_spe_prop(spe, "priv1");
+	/* priv1 is not available on a hypervisor */
+
+	spu->priv2= map_spe_prop(spe, "priv2");
+	if (!spu->priv2)
+		goto out_unmap;
+	ret = 0;
+	goto out;
+
+out_unmap:
+	spu_unmap(spu);
+out:
+	return ret;
+}
+
+static int __init find_spu_node_id(struct device_node *spe)
+{
+	unsigned int *id;
+	struct device_node *cpu;
+
+	cpu = spe->parent->parent;
+	id = (unsigned int *)get_property(cpu, "node-id", NULL);
+
+	return id ? *id : 0;
+}
+
+static int __init create_spu(struct device_node *spe)
+{
+	struct spu *spu;
+	int ret;
+	static int number;
+
+	ret = -ENOMEM;
+	spu = kmalloc(sizeof (*spu), GFP_KERNEL);
+	if (!spu)
+		goto out;
+
+	ret = spu_map_device(spu, spe);
+	if (ret)
+		goto out_free;
+
+	spu->node = find_spu_node_id(spe);
+	spu->stop_code = 0;
+	spu->slb_replace = 0;
+	spu->mm = NULL;
+	spu->ctx = NULL;
+	spu->rq = NULL;
+	spu->pid = 0;
+	spu->class_0_pending = 0;
+	spu->flags = 0UL;
+	spu->dar = 0UL;
+	spu->dsisr = 0UL;
+	spin_lock_init(&spu->register_lock);
+
+	spu_mfc_sdr_set(spu, mfspr(SPRN_SDR1));
+	spu_mfc_sr1_set(spu, 0x33);
+
+	spu->ibox_callback = NULL;
+	spu->wbox_callback = NULL;
+	spu->stop_callback = NULL;
+
+	down(&spu_mutex);
+	spu->number = number++;
+	ret = spu_request_irqs(spu);
+	if (ret)
+		goto out_unmap;
+
+	list_add(&spu->list, &spu_list);
+	up(&spu_mutex);
+
+	pr_debug(KERN_DEBUG "Using SPE %s %02x %p %p %p %p %d\n",
+		spu->name, spu->isrc, spu->local_store,
+		spu->problem, spu->priv1, spu->priv2, spu->number);
+	goto out;
+
+out_unmap:
+	up(&spu_mutex);
+	spu_unmap(spu);
+out_free:
+	kfree(spu);
+out:
+	return ret;
+}
+
+static void destroy_spu(struct spu *spu)
+{
+	list_del_init(&spu->list);
+
+	spu_free_irqs(spu);
+	spu_unmap(spu);
+	kfree(spu);
+}
+
+static void cleanup_spu_base(void)
+{
+	struct spu *spu, *tmp;
+	down(&spu_mutex);
+	list_for_each_entry_safe(spu, tmp, &spu_list, list)
+		destroy_spu(spu);
+	up(&spu_mutex);
+}
+module_exit(cleanup_spu_base);
+
+static int __init init_spu_base(void)
+{
+	struct device_node *node;
+	int ret;
+
+	ret = -ENODEV;
+	for (node = of_find_node_by_type(NULL, "spe");
+			node; node = of_find_node_by_type(node, "spe")) {
+		ret = create_spu(node);
+		if (ret) {
+			printk(KERN_WARNING "%s: Error initializing %s\n",
+				__FUNCTION__, node->name);
+			cleanup_spu_base();
+			break;
+		}
+	}
+	/* in some old firmware versions, the spe is called 'spc', so we
+	   look for that as well */
+	for (node = of_find_node_by_type(NULL, "spc");
+			node; node = of_find_node_by_type(node, "spc")) {
+		ret = create_spu(node);
+		if (ret) {
+			printk(KERN_WARNING "%s: Error initializing %s\n",
+				__FUNCTION__, node->name);
+			cleanup_spu_base();
+			break;
+		}
+	}
+	return ret;
+}
+module_init(init_spu_base);
+
+MODULE_LICENSE("GPL");
+MODULE_AUTHOR("Arnd Bergmann <arndb@de.ibm.com>");
diff --git a/arch/powerpc/platforms/cell/spu_priv1.c b/arch/powerpc/platforms/cell/spu_priv1.c
new file mode 100644
index 000000000000..b2656421c7b5
--- /dev/null
+++ b/arch/powerpc/platforms/cell/spu_priv1.c
@@ -0,0 +1,133 @@
+/*
+ * access to SPU privileged registers
+ */
+#include <linux/module.h>
+
+#include <asm/io.h>
+#include <asm/spu.h>
+
+void spu_int_mask_and(struct spu *spu, int class, u64 mask)
+{
+	u64 old_mask;
+
+	old_mask = in_be64(&spu->priv1->int_mask_RW[class]);
+	out_be64(&spu->priv1->int_mask_RW[class], old_mask & mask);
+}
+EXPORT_SYMBOL_GPL(spu_int_mask_and);
+
+void spu_int_mask_or(struct spu *spu, int class, u64 mask)
+{
+	u64 old_mask;
+
+	old_mask = in_be64(&spu->priv1->int_mask_RW[class]);
+	out_be64(&spu->priv1->int_mask_RW[class], old_mask | mask);
+}
+EXPORT_SYMBOL_GPL(spu_int_mask_or);
+
+void spu_int_mask_set(struct spu *spu, int class, u64 mask)
+{
+	out_be64(&spu->priv1->int_mask_RW[class], mask);
+}
+EXPORT_SYMBOL_GPL(spu_int_mask_set);
+
+u64 spu_int_mask_get(struct spu *spu, int class)
+{
+	return in_be64(&spu->priv1->int_mask_RW[class]);
+}
+EXPORT_SYMBOL_GPL(spu_int_mask_get);
+
+void spu_int_stat_clear(struct spu *spu, int class, u64 stat)
+{
+	out_be64(&spu->priv1->int_stat_RW[class], stat);
+}
+EXPORT_SYMBOL_GPL(spu_int_stat_clear);
+
+u64 spu_int_stat_get(struct spu *spu, int class)
+{
+	return in_be64(&spu->priv1->int_stat_RW[class]);
+}
+EXPORT_SYMBOL_GPL(spu_int_stat_get);
+
+void spu_int_route_set(struct spu *spu, u64 route)
+{
+	out_be64(&spu->priv1->int_route_RW, route);
+}
+EXPORT_SYMBOL_GPL(spu_int_route_set);
+
+u64 spu_mfc_dar_get(struct spu *spu)
+{
+	return in_be64(&spu->priv1->mfc_dar_RW);
+}
+EXPORT_SYMBOL_GPL(spu_mfc_dar_get);
+
+u64 spu_mfc_dsisr_get(struct spu *spu)
+{
+	return in_be64(&spu->priv1->mfc_dsisr_RW);
+}
+EXPORT_SYMBOL_GPL(spu_mfc_dsisr_get);
+
+void spu_mfc_dsisr_set(struct spu *spu, u64 dsisr)
+{
+	out_be64(&spu->priv1->mfc_dsisr_RW, dsisr);
+}
+EXPORT_SYMBOL_GPL(spu_mfc_dsisr_set);
+
+void spu_mfc_sdr_set(struct spu *spu, u64 sdr)
+{
+	out_be64(&spu->priv1->mfc_sdr_RW, sdr);
+}
+EXPORT_SYMBOL_GPL(spu_mfc_sdr_set);
+
+void spu_mfc_sr1_set(struct spu *spu, u64 sr1)
+{
+	out_be64(&spu->priv1->mfc_sr1_RW, sr1);
+}
+EXPORT_SYMBOL_GPL(spu_mfc_sr1_set);
+
+u64 spu_mfc_sr1_get(struct spu *spu)
+{
+	return in_be64(&spu->priv1->mfc_sr1_RW);
+}
+EXPORT_SYMBOL_GPL(spu_mfc_sr1_get);
+
+void spu_mfc_tclass_id_set(struct spu *spu, u64 tclass_id)
+{
+	out_be64(&spu->priv1->mfc_tclass_id_RW, tclass_id);
+}
+EXPORT_SYMBOL_GPL(spu_mfc_tclass_id_set);
+
+u64 spu_mfc_tclass_id_get(struct spu *spu)
+{
+	return in_be64(&spu->priv1->mfc_tclass_id_RW);
+}
+EXPORT_SYMBOL_GPL(spu_mfc_tclass_id_get);
+
+void spu_tlb_invalidate(struct spu *spu)
+{
+	out_be64(&spu->priv1->tlb_invalidate_entry_W, 0ul);
+}
+EXPORT_SYMBOL_GPL(spu_tlb_invalidate);
+
+void spu_resource_allocation_groupID_set(struct spu *spu, u64 id)
+{
+	out_be64(&spu->priv1->resource_allocation_groupID_RW, id);
+}
+EXPORT_SYMBOL_GPL(spu_resource_allocation_groupID_set);
+
+u64 spu_resource_allocation_groupID_get(struct spu *spu)
+{
+	return in_be64(&spu->priv1->resource_allocation_groupID_RW);
+}
+EXPORT_SYMBOL_GPL(spu_resource_allocation_groupID_get);
+
+void spu_resource_allocation_enable_set(struct spu *spu, u64 enable)
+{
+	out_be64(&spu->priv1->resource_allocation_enable_RW, enable);
+}
+EXPORT_SYMBOL_GPL(spu_resource_allocation_enable_set);
+
+u64 spu_resource_allocation_enable_get(struct spu *spu)
+{
+	return in_be64(&spu->priv1->resource_allocation_enable_RW);
+}
+EXPORT_SYMBOL_GPL(spu_resource_allocation_enable_get);
diff --git a/arch/powerpc/platforms/cell/spu_syscalls.c b/arch/powerpc/platforms/cell/spu_syscalls.c
new file mode 100644
index 000000000000..261b507a901a
--- /dev/null
+++ b/arch/powerpc/platforms/cell/spu_syscalls.c
@@ -0,0 +1,88 @@
+/*
+ * SPU file system -- system call stubs
+ *
+ * (C) Copyright IBM Deutschland Entwicklung GmbH 2005
+ *
+ * Author: Arnd Bergmann <arndb@de.ibm.com>
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; either version 2, or (at your option)
+ * any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software
+ * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
+ */
+#include <linux/file.h>
+#include <linux/module.h>
+#include <linux/syscalls.h>
+
+#include <asm/spu.h>
+
+struct spufs_calls spufs_calls = {
+	.owner = NULL,
+};
+
+/* These stub syscalls are needed to have the actual implementation
+ * within a loadable module. When spufs is built into the kernel,
+ * this file is not used and the syscalls directly enter the fs code */
+
+asmlinkage long sys_spu_create(const char __user *name,
+		unsigned int flags, mode_t mode)
+{
+	long ret;
+	struct module *owner = spufs_calls.owner;
+
+	ret = -ENOSYS;
+	if (owner && try_module_get(owner)) {
+		ret = spufs_calls.create_thread(name, flags, mode);
+		module_put(owner);
+	}
+	return ret;
+}
+
+asmlinkage long sys_spu_run(int fd, __u32 __user *unpc, __u32 __user *ustatus)
+{
+	long ret;
+	struct file *filp;
+	int fput_needed;
+	struct module *owner = spufs_calls.owner;
+
+	ret = -ENOSYS;
+	if (owner && try_module_get(owner)) {
+		ret = -EBADF;
+		filp = fget_light(fd, &fput_needed);
+		if (filp) {
+			ret = spufs_calls.spu_run(filp, unpc, ustatus);
+			fput_light(filp, fput_needed);
+		}
+		module_put(owner);
+	}
+	return ret;
+}
+
+int register_spu_syscalls(struct spufs_calls *calls)
+{
+	if (spufs_calls.owner)
+		return -EBUSY;
+
+	spufs_calls.create_thread = calls->create_thread;
+	spufs_calls.spu_run = calls->spu_run;
+	smp_mb();
+	spufs_calls.owner = calls->owner;
+	return 0;
+}
+EXPORT_SYMBOL_GPL(register_spu_syscalls);
+
+void unregister_spu_syscalls(struct spufs_calls *calls)
+{
+	BUG_ON(spufs_calls.owner != calls->owner);
+	spufs_calls.owner = NULL;
+}
+EXPORT_SYMBOL_GPL(unregister_spu_syscalls);
diff --git a/arch/powerpc/platforms/cell/spufs/Makefile b/arch/powerpc/platforms/cell/spufs/Makefile
new file mode 100644
index 000000000000..a7cddf40e3d9
--- /dev/null
+++ b/arch/powerpc/platforms/cell/spufs/Makefile
@@ -0,0 +1,54 @@
+obj-$(CONFIG_SPU_FS) += spufs.o
+spufs-y += inode.o file.o context.o switch.o syscalls.o
+spufs-y += sched.o backing_ops.o hw_ops.o run.o
+
+# Rules to build switch.o with the help of SPU tool chain
+SPU_CROSS	:= spu-
+SPU_CC		:= $(SPU_CROSS)gcc
+SPU_AS		:= $(SPU_CROSS)gcc
+SPU_LD		:= $(SPU_CROSS)ld
+SPU_OBJCOPY	:= $(SPU_CROSS)objcopy
+SPU_CFLAGS	:= -O2 -Wall -I$(srctree)/include -I$(objtree)/include2
+SPU_AFLAGS	:= -c -D__ASSEMBLY__ -I$(srctree)/include -I$(objtree)/include2
+SPU_LDFLAGS	:= -N -Ttext=0x0
+
+$(obj)/switch.o: $(obj)/spu_save_dump.h $(obj)/spu_restore_dump.h
+
+# Compile SPU files
+      cmd_spu_cc = $(SPU_CC) $(SPU_CFLAGS) -c -o $@ $<
+quiet_cmd_spu_cc = SPU_CC  $@
+$(obj)/spu_%.o: $(src)/spu_%.c
+	$(call if_changed,spu_cc)
+
+# Assemble SPU files
+      cmd_spu_as = $(SPU_AS) $(SPU_AFLAGS) -o $@ $<
+quiet_cmd_spu_as = SPU_AS  $@
+$(obj)/spu_%.o: $(src)/spu_%.S
+	$(call if_changed,spu_as)
+
+# Link SPU Executables
+      cmd_spu_ld = $(SPU_LD) $(SPU_LDFLAGS) -o $@ $^
+quiet_cmd_spu_ld = SPU_LD  $@
+$(obj)/spu_%: $(obj)/spu_%_crt0.o $(obj)/spu_%.o
+	$(call if_changed,spu_ld)
+
+# Copy into binary format
+      cmd_spu_objcopy = $(SPU_OBJCOPY) -O binary $< $@
+quiet_cmd_spu_objcopy = OBJCOPY $@
+$(obj)/spu_%.bin: $(src)/spu_%
+	$(call if_changed,spu_objcopy)
+
+# create C code from ELF executable
+cmd_hexdump   = ( \
+		echo "/*" ; \
+		echo " * $*_dump.h: Copyright (C) 2005 IBM." ; \
+		echo " * Hex-dump auto generated from $*.c." ; \
+		echo " * Do not edit!" ; \
+		echo " */" ; \
+		echo "static unsigned int $*_code[] __page_aligned = {" ; \
+		hexdump -v -e '"0x" 4/1 "%02x" "," "\n"' $< ; \
+		echo "};" ; \
+		) > $@
+quiet_cmd_hexdump = HEXDUMP $@
+$(obj)/%_dump.h: $(obj)/%.bin
+	$(call if_changed,hexdump)
diff --git a/arch/powerpc/platforms/cell/spufs/backing_ops.c b/arch/powerpc/platforms/cell/spufs/backing_ops.c
new file mode 100644
index 000000000000..a5c489a53c61
--- /dev/null
+++ b/arch/powerpc/platforms/cell/spufs/backing_ops.c
@@ -0,0 +1,308 @@
+/* backing_ops.c - query/set operations on saved SPU context.
+ *
+ * Copyright (C) IBM 2005
+ * Author: Mark Nutter <mnutter@us.ibm.com>
+ *
+ * These register operations allow SPUFS to operate on saved
+ * SPU contexts rather than hardware.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; either version 2, or (at your option)
+ * any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software
+ * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
+ */
+
+#include <linux/config.h>
+#include <linux/module.h>
+#include <linux/errno.h>
+#include <linux/sched.h>
+#include <linux/kernel.h>
+#include <linux/mm.h>
+#include <linux/vmalloc.h>
+#include <linux/smp.h>
+#include <linux/smp_lock.h>
+#include <linux/stddef.h>
+#include <linux/unistd.h>
+#include <linux/poll.h>
+
+#include <asm/io.h>
+#include <asm/spu.h>
+#include <asm/spu_csa.h>
+#include <asm/mmu_context.h>
+#include "spufs.h"
+
+/*
+ * Reads/writes to various problem and priv2 registers require
+ * state changes, i.e.  generate SPU events, modify channel
+ * counts, etc.
+ */
+
+static void gen_spu_event(struct spu_context *ctx, u32 event)
+{
+	u64 ch0_cnt;
+	u64 ch0_data;
+	u64 ch1_data;
+
+	ch0_cnt = ctx->csa.spu_chnlcnt_RW[0];
+	ch0_data = ctx->csa.spu_chnldata_RW[0];
+	ch1_data = ctx->csa.spu_chnldata_RW[1];
+	ctx->csa.spu_chnldata_RW[0] |= event;
+	if ((ch0_cnt == 0) && !(ch0_data & event) && (ch1_data & event)) {
+		ctx->csa.spu_chnlcnt_RW[0] = 1;
+	}
+}
+
+static int spu_backing_mbox_read(struct spu_context *ctx, u32 * data)
+{
+	u32 mbox_stat;
+	int ret = 0;
+
+	spin_lock(&ctx->csa.register_lock);
+	mbox_stat = ctx->csa.prob.mb_stat_R;
+	if (mbox_stat & 0x0000ff) {
+		/* Read the first available word.
+		 * Implementation note: the depth
+		 * of pu_mb_R is currently 1.
+		 */
+		*data = ctx->csa.prob.pu_mb_R;
+		ctx->csa.prob.mb_stat_R &= ~(0x0000ff);
+		ctx->csa.spu_chnlcnt_RW[28] = 1;
+		gen_spu_event(ctx, MFC_PU_MAILBOX_AVAILABLE_EVENT);
+		ret = 4;
+	}
+	spin_unlock(&ctx->csa.register_lock);
+	return ret;
+}
+
+static u32 spu_backing_mbox_stat_read(struct spu_context *ctx)
+{
+	return ctx->csa.prob.mb_stat_R;
+}
+
+static unsigned int spu_backing_mbox_stat_poll(struct spu_context *ctx,
+					  unsigned int events)
+{
+	int ret;
+	u32 stat;
+
+	ret = 0;
+	spin_lock_irq(&ctx->csa.register_lock);
+	stat = ctx->csa.prob.mb_stat_R;
+
+	/* if the requested event is there, return the poll
+	   mask, otherwise enable the interrupt to get notified,
+	   but first mark any pending interrupts as done so
+	   we don't get woken up unnecessarily */
+
+	if (events & (POLLIN | POLLRDNORM)) {
+		if (stat & 0xff0000)
+			ret |= POLLIN | POLLRDNORM;
+		else {
+			ctx->csa.priv1.int_stat_class0_RW &= ~0x1;
+			ctx->csa.priv1.int_mask_class2_RW |= 0x1;
+		}
+	}
+	if (events & (POLLOUT | POLLWRNORM)) {
+		if (stat & 0x00ff00)
+			ret = POLLOUT | POLLWRNORM;
+		else {
+			ctx->csa.priv1.int_stat_class0_RW &= ~0x10;
+			ctx->csa.priv1.int_mask_class2_RW |= 0x10;
+		}
+	}
+	spin_unlock_irq(&ctx->csa.register_lock);
+	return ret;
+}
+
+static int spu_backing_ibox_read(struct spu_context *ctx, u32 * data)
+{
+	int ret;
+
+	spin_lock(&ctx->csa.register_lock);
+	if (ctx->csa.prob.mb_stat_R & 0xff0000) {
+		/* Read the first available word.
+		 * Implementation note: the depth
+		 * of puint_mb_R is currently 1.
+		 */
+		*data = ctx->csa.priv2.puint_mb_R;
+		ctx->csa.prob.mb_stat_R &= ~(0xff0000);
+		ctx->csa.spu_chnlcnt_RW[30] = 1;
+		gen_spu_event(ctx, MFC_PU_INT_MAILBOX_AVAILABLE_EVENT);
+		ret = 4;
+	} else {
+		/* make sure we get woken up by the interrupt */
+		ctx->csa.priv1.int_mask_class2_RW |= 0x1UL;
+		ret = 0;
+	}
+	spin_unlock(&ctx->csa.register_lock);
+	return ret;
+}
+
+static int spu_backing_wbox_write(struct spu_context *ctx, u32 data)
+{
+	int ret;
+
+	spin_lock(&ctx->csa.register_lock);
+	if ((ctx->csa.prob.mb_stat_R) & 0x00ff00) {
+		int slot = ctx->csa.spu_chnlcnt_RW[29];
+		int avail = (ctx->csa.prob.mb_stat_R & 0x00ff00) >> 8;
+
+		/* We have space to write wbox_data.
+		 * Implementation note: the depth
+		 * of spu_mb_W is currently 4.
+		 */
+		BUG_ON(avail != (4 - slot));
+		ctx->csa.spu_mailbox_data[slot] = data;
+		ctx->csa.spu_chnlcnt_RW[29] = ++slot;
+		ctx->csa.prob.mb_stat_R = (((4 - slot) & 0xff) << 8);
+		gen_spu_event(ctx, MFC_SPU_MAILBOX_WRITTEN_EVENT);
+		ret = 4;
+	} else {
+		/* make sure we get woken up by the interrupt when space
+		   becomes available */
+		ctx->csa.priv1.int_mask_class2_RW |= 0x10;
+		ret = 0;
+	}
+	spin_unlock(&ctx->csa.register_lock);
+	return ret;
+}
+
+static u32 spu_backing_signal1_read(struct spu_context *ctx)
+{
+	return ctx->csa.spu_chnldata_RW[3];
+}
+
+static void spu_backing_signal1_write(struct spu_context *ctx, u32 data)
+{
+	spin_lock(&ctx->csa.register_lock);
+	if (ctx->csa.priv2.spu_cfg_RW & 0x1)
+		ctx->csa.spu_chnldata_RW[3] |= data;
+	else
+		ctx->csa.spu_chnldata_RW[3] = data;
+	ctx->csa.spu_chnlcnt_RW[3] = 1;
+	gen_spu_event(ctx, MFC_SIGNAL_1_EVENT);
+	spin_unlock(&ctx->csa.register_lock);
+}
+
+static u32 spu_backing_signal2_read(struct spu_context *ctx)
+{
+	return ctx->csa.spu_chnldata_RW[4];
+}
+
+static void spu_backing_signal2_write(struct spu_context *ctx, u32 data)
+{
+	spin_lock(&ctx->csa.register_lock);
+	if (ctx->csa.priv2.spu_cfg_RW & 0x2)
+		ctx->csa.spu_chnldata_RW[4] |= data;
+	else
+		ctx->csa.spu_chnldata_RW[4] = data;
+	ctx->csa.spu_chnlcnt_RW[4] = 1;
+	gen_spu_event(ctx, MFC_SIGNAL_2_EVENT);
+	spin_unlock(&ctx->csa.register_lock);
+}
+
+static void spu_backing_signal1_type_set(struct spu_context *ctx, u64 val)
+{
+	u64 tmp;
+
+	spin_lock(&ctx->csa.register_lock);
+	tmp = ctx->csa.priv2.spu_cfg_RW;
+	if (val)
+		tmp |= 1;
+	else
+		tmp &= ~1;
+	ctx->csa.priv2.spu_cfg_RW = tmp;
+	spin_unlock(&ctx->csa.register_lock);
+}
+
+static u64 spu_backing_signal1_type_get(struct spu_context *ctx)
+{
+	return ((ctx->csa.priv2.spu_cfg_RW & 1) != 0);
+}
+
+static void spu_backing_signal2_type_set(struct spu_context *ctx, u64 val)
+{
+	u64 tmp;
+
+	spin_lock(&ctx->csa.register_lock);
+	tmp = ctx->csa.priv2.spu_cfg_RW;
+	if (val)
+		tmp |= 2;
+	else
+		tmp &= ~2;
+	ctx->csa.priv2.spu_cfg_RW = tmp;
+	spin_unlock(&ctx->csa.register_lock);
+}
+
+static u64 spu_backing_signal2_type_get(struct spu_context *ctx)
+{
+	return ((ctx->csa.priv2.spu_cfg_RW & 2) != 0);
+}
+
+static u32 spu_backing_npc_read(struct spu_context *ctx)
+{
+	return ctx->csa.prob.spu_npc_RW;
+}
+
+static void spu_backing_npc_write(struct spu_context *ctx, u32 val)
+{
+	ctx->csa.prob.spu_npc_RW = val;
+}
+
+static u32 spu_backing_status_read(struct spu_context *ctx)
+{
+	return ctx->csa.prob.spu_status_R;
+}
+
+static char *spu_backing_get_ls(struct spu_context *ctx)
+{
+	return ctx->csa.lscsa->ls;
+}
+
+static void spu_backing_runcntl_write(struct spu_context *ctx, u32 val)
+{
+	spin_lock(&ctx->csa.register_lock);
+	ctx->csa.prob.spu_runcntl_RW = val;
+	if (val & SPU_RUNCNTL_RUNNABLE) {
+		ctx->csa.prob.spu_status_R |= SPU_STATUS_RUNNING;
+	} else {
+		ctx->csa.prob.spu_status_R &= ~SPU_STATUS_RUNNING;
+	}
+	spin_unlock(&ctx->csa.register_lock);
+}
+
+static void spu_backing_runcntl_stop(struct spu_context *ctx)
+{
+	spu_backing_runcntl_write(ctx, SPU_RUNCNTL_STOP);
+}
+
+struct spu_context_ops spu_backing_ops = {
+	.mbox_read = spu_backing_mbox_read,
+	.mbox_stat_read = spu_backing_mbox_stat_read,
+	.mbox_stat_poll = spu_backing_mbox_stat_poll,
+	.ibox_read = spu_backing_ibox_read,
+	.wbox_write = spu_backing_wbox_write,
+	.signal1_read = spu_backing_signal1_read,
+	.signal1_write = spu_backing_signal1_write,
+	.signal2_read = spu_backing_signal2_read,
+	.signal2_write = spu_backing_signal2_write,
+	.signal1_type_set = spu_backing_signal1_type_set,
+	.signal1_type_get = spu_backing_signal1_type_get,
+	.signal2_type_set = spu_backing_signal2_type_set,
+	.signal2_type_get = spu_backing_signal2_type_get,
+	.npc_read = spu_backing_npc_read,
+	.npc_write = spu_backing_npc_write,
+	.status_read = spu_backing_status_read,
+	.get_ls = spu_backing_get_ls,
+	.runcntl_write = spu_backing_runcntl_write,
+	.runcntl_stop = spu_backing_runcntl_stop,
+};
diff --git a/arch/powerpc/platforms/cell/spufs/context.c b/arch/powerpc/platforms/cell/spufs/context.c
new file mode 100644
index 000000000000..336f238102fd
--- /dev/null
+++ b/arch/powerpc/platforms/cell/spufs/context.c
@@ -0,0 +1,167 @@
+/*
+ * SPU file system -- SPU context management
+ *
+ * (C) Copyright IBM Deutschland Entwicklung GmbH 2005
+ *
+ * Author: Arnd Bergmann <arndb@de.ibm.com>
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; either version 2, or (at your option)
+ * any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software
+ * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
+ */
+
+#include <linux/fs.h>
+#include <linux/mm.h>
+#include <linux/slab.h>
+#include <asm/spu.h>
+#include <asm/spu_csa.h>
+#include "spufs.h"
+
+struct spu_context *alloc_spu_context(struct address_space *local_store)
+{
+	struct spu_context *ctx;
+	ctx = kmalloc(sizeof *ctx, GFP_KERNEL);
+	if (!ctx)
+		goto out;
+	/* Binding to physical processor deferred
+	 * until spu_activate().
+	 */
+	spu_init_csa(&ctx->csa);
+	if (!ctx->csa.lscsa) {
+		goto out_free;
+	}
+	spin_lock_init(&ctx->mmio_lock);
+	kref_init(&ctx->kref);
+	init_rwsem(&ctx->state_sema);
+	init_MUTEX(&ctx->run_sema);
+	init_waitqueue_head(&ctx->ibox_wq);
+	init_waitqueue_head(&ctx->wbox_wq);
+	init_waitqueue_head(&ctx->stop_wq);
+	ctx->ibox_fasync = NULL;
+	ctx->wbox_fasync = NULL;
+	ctx->state = SPU_STATE_SAVED;
+	ctx->local_store = local_store;
+	ctx->spu = NULL;
+	ctx->ops = &spu_backing_ops;
+	ctx->owner = get_task_mm(current);
+	goto out;
+out_free:
+	kfree(ctx);
+	ctx = NULL;
+out:
+	return ctx;
+}
+
+void destroy_spu_context(struct kref *kref)
+{
+	struct spu_context *ctx;
+	ctx = container_of(kref, struct spu_context, kref);
+	down_write(&ctx->state_sema);
+	spu_deactivate(ctx);
+	ctx->ibox_fasync = NULL;
+	ctx->wbox_fasync = NULL;
+	up_write(&ctx->state_sema);
+	spu_fini_csa(&ctx->csa);
+	kfree(ctx);
+}
+
+struct spu_context * get_spu_context(struct spu_context *ctx)
+{
+	kref_get(&ctx->kref);
+	return ctx;
+}
+
+int put_spu_context(struct spu_context *ctx)
+{
+	return kref_put(&ctx->kref, &destroy_spu_context);
+}
+
+/* give up the mm reference when the context is about to be destroyed */
+void spu_forget(struct spu_context *ctx)
+{
+	struct mm_struct *mm;
+	spu_acquire_saved(ctx);
+	mm = ctx->owner;
+	ctx->owner = NULL;
+	mmput(mm);
+	spu_release(ctx);
+}
+
+void spu_acquire(struct spu_context *ctx)
+{
+	down_read(&ctx->state_sema);
+}
+
+void spu_release(struct spu_context *ctx)
+{
+	up_read(&ctx->state_sema);
+}
+
+void spu_unmap_mappings(struct spu_context *ctx)
+{
+	unmap_mapping_range(ctx->local_store, 0, LS_SIZE, 1);
+}
+
+int spu_acquire_runnable(struct spu_context *ctx)
+{
+	int ret = 0;
+
+	down_read(&ctx->state_sema);
+	if (ctx->state == SPU_STATE_RUNNABLE) {
+		ctx->spu->prio = current->prio;
+		return 0;
+	}
+	up_read(&ctx->state_sema);
+
+	down_write(&ctx->state_sema);
+	/* ctx is about to be freed, can't acquire any more */
+	if (!ctx->owner) {
+		ret = -EINVAL;
+		goto out;
+	}
+
+	if (ctx->state == SPU_STATE_SAVED) {
+		ret = spu_activate(ctx, 0);
+		if (ret)
+			goto out;
+		ctx->state = SPU_STATE_RUNNABLE;
+	}
+
+	downgrade_write(&ctx->state_sema);
+	/* On success, we return holding the lock */
+
+	return ret;
+out:
+	/* Release here, to simplify calling code. */
+	up_write(&ctx->state_sema);
+
+	return ret;
+}
+
+void spu_acquire_saved(struct spu_context *ctx)
+{
+	down_read(&ctx->state_sema);
+
+	if (ctx->state == SPU_STATE_SAVED)
+		return;
+
+	up_read(&ctx->state_sema);
+	down_write(&ctx->state_sema);
+
+	if (ctx->state == SPU_STATE_RUNNABLE) {
+		spu_deactivate(ctx);
+		ctx->state = SPU_STATE_SAVED;
+	}
+
+	downgrade_write(&ctx->state_sema);
+}
diff --git a/arch/powerpc/platforms/cell/spufs/file.c b/arch/powerpc/platforms/cell/spufs/file.c
new file mode 100644
index 000000000000..dfa649c9b956
--- /dev/null
+++ b/arch/powerpc/platforms/cell/spufs/file.c
@@ -0,0 +1,794 @@
+/*
+ * SPU file system -- file contents
+ *
+ * (C) Copyright IBM Deutschland Entwicklung GmbH 2005
+ *
+ * Author: Arnd Bergmann <arndb@de.ibm.com>
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; either version 2, or (at your option)
+ * any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software
+ * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
+ */
+
+#include <linux/fs.h>
+#include <linux/ioctl.h>
+#include <linux/module.h>
+#include <linux/pagemap.h>
+#include <linux/poll.h>
+#include <linux/ptrace.h>
+
+#include <asm/io.h>
+#include <asm/semaphore.h>
+#include <asm/spu.h>
+#include <asm/uaccess.h>
+
+#include "spufs.h"
+
+
+static int
+spufs_mem_open(struct inode *inode, struct file *file)
+{
+	struct spufs_inode_info *i = SPUFS_I(inode);
+	file->private_data = i->i_ctx;
+	file->f_mapping = i->i_ctx->local_store;
+	return 0;
+}
+
+static ssize_t
+spufs_mem_read(struct file *file, char __user *buffer,
+				size_t size, loff_t *pos)
+{
+	struct spu_context *ctx = file->private_data;
+	char *local_store;
+	int ret;
+
+	spu_acquire(ctx);
+
+	local_store = ctx->ops->get_ls(ctx);
+	ret = simple_read_from_buffer(buffer, size, pos, local_store, LS_SIZE);
+
+	spu_release(ctx);
+	return ret;
+}
+
+static ssize_t
+spufs_mem_write(struct file *file, const char __user *buffer,
+					size_t size, loff_t *pos)
+{
+	struct spu_context *ctx = file->private_data;
+	char *local_store;
+	int ret;
+
+	size = min_t(ssize_t, LS_SIZE - *pos, size);
+	if (size <= 0)
+		return -EFBIG;
+	*pos += size;
+
+	spu_acquire(ctx);
+
+	local_store = ctx->ops->get_ls(ctx);
+	ret = copy_from_user(local_store + *pos - size,
+			     buffer, size) ? -EFAULT : size;
+
+	spu_release(ctx);
+	return ret;
+}
+
+#ifdef CONFIG_SPARSEMEM
+static struct page *
+spufs_mem_mmap_nopage(struct vm_area_struct *vma,
+		      unsigned long address, int *type)
+{
+	struct page *page = NOPAGE_SIGBUS;
+
+	struct spu_context *ctx = vma->vm_file->private_data;
+	unsigned long offset = address - vma->vm_start;
+	offset += vma->vm_pgoff << PAGE_SHIFT;
+
+	spu_acquire(ctx);
+
+	if (ctx->state == SPU_STATE_SAVED)
+		page = vmalloc_to_page(ctx->csa.lscsa->ls + offset);
+	else
+		page = pfn_to_page((ctx->spu->local_store_phys + offset)
+				   >> PAGE_SHIFT);
+
+	spu_release(ctx);
+
+	if (type)
+		*type = VM_FAULT_MINOR;
+
+	page_cache_get(page);
+	return page;
+}
+
+static struct vm_operations_struct spufs_mem_mmap_vmops = {
+	.nopage = spufs_mem_mmap_nopage,
+};
+
+static int
+spufs_mem_mmap(struct file *file, struct vm_area_struct *vma)
+{
+	if (!(vma->vm_flags & VM_SHARED))
+		return -EINVAL;
+
+	/* FIXME: */
+	vma->vm_page_prot = __pgprot(pgprot_val(vma->vm_page_prot)
+				     | _PAGE_NO_CACHE);
+
+	vma->vm_ops = &spufs_mem_mmap_vmops;
+	return 0;
+}
+#endif
+
+static struct file_operations spufs_mem_fops = {
+	.open	 = spufs_mem_open,
+	.read    = spufs_mem_read,
+	.write   = spufs_mem_write,
+	.llseek  = generic_file_llseek,
+#ifdef CONFIG_SPARSEMEM
+	.mmap    = spufs_mem_mmap,
+#endif
+};
+
+static int
+spufs_regs_open(struct inode *inode, struct file *file)
+{
+	struct spufs_inode_info *i = SPUFS_I(inode);
+	file->private_data = i->i_ctx;
+	return 0;
+}
+
+static ssize_t
+spufs_regs_read(struct file *file, char __user *buffer,
+		size_t size, loff_t *pos)
+{
+	struct spu_context *ctx = file->private_data;
+	struct spu_lscsa *lscsa = ctx->csa.lscsa;
+	int ret;
+
+	spu_acquire_saved(ctx);
+
+	ret = simple_read_from_buffer(buffer, size, pos,
+				      lscsa->gprs, sizeof lscsa->gprs);
+
+	spu_release(ctx);
+	return ret;
+}
+
+static ssize_t
+spufs_regs_write(struct file *file, const char __user *buffer,
+		 size_t size, loff_t *pos)
+{
+	struct spu_context *ctx = file->private_data;
+	struct spu_lscsa *lscsa = ctx->csa.lscsa;
+	int ret;
+
+	size = min_t(ssize_t, sizeof lscsa->gprs - *pos, size);
+	if (size <= 0)
+		return -EFBIG;
+	*pos += size;
+
+	spu_acquire_saved(ctx);
+
+	ret = copy_from_user(lscsa->gprs + *pos - size,
+			     buffer, size) ? -EFAULT : size;
+
+	spu_release(ctx);
+	return ret;
+}
+
+static struct file_operations spufs_regs_fops = {
+	.open	 = spufs_regs_open,
+	.read    = spufs_regs_read,
+	.write   = spufs_regs_write,
+	.llseek  = generic_file_llseek,
+};
+
+static ssize_t
+spufs_fpcr_read(struct file *file, char __user * buffer,
+		size_t size, loff_t * pos)
+{
+	struct spu_context *ctx = file->private_data;
+	struct spu_lscsa *lscsa = ctx->csa.lscsa;
+	int ret;
+
+	spu_acquire_saved(ctx);
+
+	ret = simple_read_from_buffer(buffer, size, pos,
+				      &lscsa->fpcr, sizeof(lscsa->fpcr));
+
+	spu_release(ctx);
+	return ret;
+}
+
+static ssize_t
+spufs_fpcr_write(struct file *file, const char __user * buffer,
+		 size_t size, loff_t * pos)
+{
+	struct spu_context *ctx = file->private_data;
+	struct spu_lscsa *lscsa = ctx->csa.lscsa;
+	int ret;
+
+	size = min_t(ssize_t, sizeof(lscsa->fpcr) - *pos, size);
+	if (size <= 0)
+		return -EFBIG;
+	*pos += size;
+
+	spu_acquire_saved(ctx);
+
+	ret = copy_from_user((char *)&lscsa->fpcr + *pos - size,
+			     buffer, size) ? -EFAULT : size;
+
+	spu_release(ctx);
+	return ret;
+}
+
+static struct file_operations spufs_fpcr_fops = {
+	.open = spufs_regs_open,
+	.read = spufs_fpcr_read,
+	.write = spufs_fpcr_write,
+	.llseek = generic_file_llseek,
+};
+
+/* generic open function for all pipe-like files */
+static int spufs_pipe_open(struct inode *inode, struct file *file)
+{
+	struct spufs_inode_info *i = SPUFS_I(inode);
+	file->private_data = i->i_ctx;
+
+	return nonseekable_open(inode, file);
+}
+
+static ssize_t spufs_mbox_read(struct file *file, char __user *buf,
+			size_t len, loff_t *pos)
+{
+	struct spu_context *ctx = file->private_data;
+	u32 mbox_data;
+	int ret;
+
+	if (len < 4)
+		return -EINVAL;
+
+	spu_acquire(ctx);
+	ret = ctx->ops->mbox_read(ctx, &mbox_data);
+	spu_release(ctx);
+
+	if (!ret)
+		return -EAGAIN;
+
+	if (copy_to_user(buf, &mbox_data, sizeof mbox_data))
+		return -EFAULT;
+
+	return 4;
+}
+
+static struct file_operations spufs_mbox_fops = {
+	.open	= spufs_pipe_open,
+	.read	= spufs_mbox_read,
+};
+
+static ssize_t spufs_mbox_stat_read(struct file *file, char __user *buf,
+			size_t len, loff_t *pos)
+{
+	struct spu_context *ctx = file->private_data;
+	u32 mbox_stat;
+
+	if (len < 4)
+		return -EINVAL;
+
+	spu_acquire(ctx);
+
+	mbox_stat = ctx->ops->mbox_stat_read(ctx) & 0xff;
+
+	spu_release(ctx);
+
+	if (copy_to_user(buf, &mbox_stat, sizeof mbox_stat))
+		return -EFAULT;
+
+	return 4;
+}
+
+static struct file_operations spufs_mbox_stat_fops = {
+	.open	= spufs_pipe_open,
+	.read	= spufs_mbox_stat_read,
+};
+
+/* low-level ibox access function */
+size_t spu_ibox_read(struct spu_context *ctx, u32 *data)
+{
+	return ctx->ops->ibox_read(ctx, data);
+}
+
+static int spufs_ibox_fasync(int fd, struct file *file, int on)
+{
+	struct spu_context *ctx = file->private_data;
+
+	return fasync_helper(fd, file, on, &ctx->ibox_fasync);
+}
+
+/* interrupt-level ibox callback function. */
+void spufs_ibox_callback(struct spu *spu)
+{
+	struct spu_context *ctx = spu->ctx;
+
+	wake_up_all(&ctx->ibox_wq);
+	kill_fasync(&ctx->ibox_fasync, SIGIO, POLLIN);
+}
+
+static ssize_t spufs_ibox_read(struct file *file, char __user *buf,
+			size_t len, loff_t *pos)
+{
+	struct spu_context *ctx = file->private_data;
+	u32 ibox_data;
+	ssize_t ret;
+
+	if (len < 4)
+		return -EINVAL;
+
+	spu_acquire(ctx);
+
+	ret = 0;
+	if (file->f_flags & O_NONBLOCK) {
+		if (!spu_ibox_read(ctx, &ibox_data))
+			ret = -EAGAIN;
+	} else {
+		ret = spufs_wait(ctx->ibox_wq, spu_ibox_read(ctx, &ibox_data));
+	}
+
+	spu_release(ctx);
+
+	if (ret)
+		return ret;
+
+	ret = 4;
+	if (copy_to_user(buf, &ibox_data, sizeof ibox_data))
+		ret = -EFAULT;
+
+	return ret;
+}
+
+static unsigned int spufs_ibox_poll(struct file *file, poll_table *wait)
+{
+	struct spu_context *ctx = file->private_data;
+	unsigned int mask;
+
+	poll_wait(file, &ctx->ibox_wq, wait);
+
+	spu_acquire(ctx);
+	mask = ctx->ops->mbox_stat_poll(ctx, POLLIN | POLLRDNORM);
+	spu_release(ctx);
+
+	return mask;
+}
+
+static struct file_operations spufs_ibox_fops = {
+	.open	= spufs_pipe_open,
+	.read	= spufs_ibox_read,
+	.poll	= spufs_ibox_poll,
+	.fasync	= spufs_ibox_fasync,
+};
+
+static ssize_t spufs_ibox_stat_read(struct file *file, char __user *buf,
+			size_t len, loff_t *pos)
+{
+	struct spu_context *ctx = file->private_data;
+	u32 ibox_stat;
+
+	if (len < 4)
+		return -EINVAL;
+
+	spu_acquire(ctx);
+	ibox_stat = (ctx->ops->mbox_stat_read(ctx) >> 16) & 0xff;
+	spu_release(ctx);
+
+	if (copy_to_user(buf, &ibox_stat, sizeof ibox_stat))
+		return -EFAULT;
+
+	return 4;
+}
+
+static struct file_operations spufs_ibox_stat_fops = {
+	.open	= spufs_pipe_open,
+	.read	= spufs_ibox_stat_read,
+};
+
+/* low-level mailbox write */
+size_t spu_wbox_write(struct spu_context *ctx, u32 data)
+{
+	return ctx->ops->wbox_write(ctx, data);
+}
+
+static int spufs_wbox_fasync(int fd, struct file *file, int on)
+{
+	struct spu_context *ctx = file->private_data;
+	int ret;
+
+	ret = fasync_helper(fd, file, on, &ctx->wbox_fasync);
+
+	return ret;
+}
+
+/* interrupt-level wbox callback function. */
+void spufs_wbox_callback(struct spu *spu)
+{
+	struct spu_context *ctx = spu->ctx;
+
+	wake_up_all(&ctx->wbox_wq);
+	kill_fasync(&ctx->wbox_fasync, SIGIO, POLLOUT);
+}
+
+static ssize_t spufs_wbox_write(struct file *file, const char __user *buf,
+			size_t len, loff_t *pos)
+{
+	struct spu_context *ctx = file->private_data;
+	u32 wbox_data;
+	int ret;
+
+	if (len < 4)
+		return -EINVAL;
+
+	if (copy_from_user(&wbox_data, buf, sizeof wbox_data))
+		return -EFAULT;
+
+	spu_acquire(ctx);
+
+	ret = 0;
+	if (file->f_flags & O_NONBLOCK) {
+		if (!spu_wbox_write(ctx, wbox_data))
+			ret = -EAGAIN;
+	} else {
+		ret = spufs_wait(ctx->wbox_wq, spu_wbox_write(ctx, wbox_data));
+	}
+
+	spu_release(ctx);
+
+	return ret ? ret : sizeof wbox_data;
+}
+
+static unsigned int spufs_wbox_poll(struct file *file, poll_table *wait)
+{
+	struct spu_context *ctx = file->private_data;
+	unsigned int mask;
+
+	poll_wait(file, &ctx->wbox_wq, wait);
+
+	spu_acquire(ctx);
+	mask = ctx->ops->mbox_stat_poll(ctx, POLLOUT | POLLWRNORM);
+	spu_release(ctx);
+
+	return mask;
+}
+
+static struct file_operations spufs_wbox_fops = {
+	.open	= spufs_pipe_open,
+	.write	= spufs_wbox_write,
+	.poll	= spufs_wbox_poll,
+	.fasync	= spufs_wbox_fasync,
+};
+
+static ssize_t spufs_wbox_stat_read(struct file *file, char __user *buf,
+			size_t len, loff_t *pos)
+{
+	struct spu_context *ctx = file->private_data;
+	u32 wbox_stat;
+
+	if (len < 4)
+		return -EINVAL;
+
+	spu_acquire(ctx);
+	wbox_stat = (ctx->ops->mbox_stat_read(ctx) >> 8) & 0xff;
+	spu_release(ctx);
+
+	if (copy_to_user(buf, &wbox_stat, sizeof wbox_stat))
+		return -EFAULT;
+
+	return 4;
+}
+
+static struct file_operations spufs_wbox_stat_fops = {
+	.open	= spufs_pipe_open,
+	.read	= spufs_wbox_stat_read,
+};
+
+static ssize_t spufs_signal1_read(struct file *file, char __user *buf,
+			size_t len, loff_t *pos)
+{
+	struct spu_context *ctx = file->private_data;
+	u32 data;
+
+	if (len < 4)
+		return -EINVAL;
+
+	spu_acquire(ctx);
+	data = ctx->ops->signal1_read(ctx);
+	spu_release(ctx);
+
+	if (copy_to_user(buf, &data, 4))
+		return -EFAULT;
+
+	return 4;
+}
+
+static ssize_t spufs_signal1_write(struct file *file, const char __user *buf,
+			size_t len, loff_t *pos)
+{
+	struct spu_context *ctx;
+	u32 data;
+
+	ctx = file->private_data;
+
+	if (len < 4)
+		return -EINVAL;
+
+	if (copy_from_user(&data, buf, 4))
+		return -EFAULT;
+
+	spu_acquire(ctx);
+	ctx->ops->signal1_write(ctx, data);
+	spu_release(ctx);
+
+	return 4;
+}
+
+static struct file_operations spufs_signal1_fops = {
+	.open = spufs_pipe_open,
+	.read = spufs_signal1_read,
+	.write = spufs_signal1_write,
+};
+
+static ssize_t spufs_signal2_read(struct file *file, char __user *buf,
+			size_t len, loff_t *pos)
+{
+	struct spu_context *ctx;
+	u32 data;
+
+	ctx = file->private_data;
+
+	if (len < 4)
+		return -EINVAL;
+
+	spu_acquire(ctx);
+	data = ctx->ops->signal2_read(ctx);
+	spu_release(ctx);
+
+	if (copy_to_user(buf, &data, 4))
+		return -EFAULT;
+
+	return 4;
+}
+
+static ssize_t spufs_signal2_write(struct file *file, const char __user *buf,
+			size_t len, loff_t *pos)
+{
+	struct spu_context *ctx;
+	u32 data;
+
+	ctx = file->private_data;
+
+	if (len < 4)
+		return -EINVAL;
+
+	if (copy_from_user(&data, buf, 4))
+		return -EFAULT;
+
+	spu_acquire(ctx);
+	ctx->ops->signal2_write(ctx, data);
+	spu_release(ctx);
+
+	return 4;
+}
+
+static struct file_operations spufs_signal2_fops = {
+	.open = spufs_pipe_open,
+	.read = spufs_signal2_read,
+	.write = spufs_signal2_write,
+};
+
+static void spufs_signal1_type_set(void *data, u64 val)
+{
+	struct spu_context *ctx = data;
+
+	spu_acquire(ctx);
+	ctx->ops->signal1_type_set(ctx, val);
+	spu_release(ctx);
+}
+
+static u64 spufs_signal1_type_get(void *data)
+{
+	struct spu_context *ctx = data;
+	u64 ret;
+
+	spu_acquire(ctx);
+	ret = ctx->ops->signal1_type_get(ctx);
+	spu_release(ctx);
+
+	return ret;
+}
+DEFINE_SIMPLE_ATTRIBUTE(spufs_signal1_type, spufs_signal1_type_get,
+					spufs_signal1_type_set, "%llu");
+
+static void spufs_signal2_type_set(void *data, u64 val)
+{
+	struct spu_context *ctx = data;
+
+	spu_acquire(ctx);
+	ctx->ops->signal2_type_set(ctx, val);
+	spu_release(ctx);
+}
+
+static u64 spufs_signal2_type_get(void *data)
+{
+	struct spu_context *ctx = data;
+	u64 ret;
+
+	spu_acquire(ctx);
+	ret = ctx->ops->signal2_type_get(ctx);
+	spu_release(ctx);
+
+	return ret;
+}
+DEFINE_SIMPLE_ATTRIBUTE(spufs_signal2_type, spufs_signal2_type_get,
+					spufs_signal2_type_set, "%llu");
+
+static void spufs_npc_set(void *data, u64 val)
+{
+	struct spu_context *ctx = data;
+	spu_acquire(ctx);
+	ctx->ops->npc_write(ctx, val);
+	spu_release(ctx);
+}
+
+static u64 spufs_npc_get(void *data)
+{
+	struct spu_context *ctx = data;
+	u64 ret;
+	spu_acquire(ctx);
+	ret = ctx->ops->npc_read(ctx);
+	spu_release(ctx);
+	return ret;
+}
+DEFINE_SIMPLE_ATTRIBUTE(spufs_npc_ops, spufs_npc_get, spufs_npc_set, "%llx\n")
+
+static void spufs_decr_set(void *data, u64 val)
+{
+	struct spu_context *ctx = data;
+	struct spu_lscsa *lscsa = ctx->csa.lscsa;
+	spu_acquire_saved(ctx);
+	lscsa->decr.slot[0] = (u32) val;
+	spu_release(ctx);
+}
+
+static u64 spufs_decr_get(void *data)
+{
+	struct spu_context *ctx = data;
+	struct spu_lscsa *lscsa = ctx->csa.lscsa;
+	u64 ret;
+	spu_acquire_saved(ctx);
+	ret = lscsa->decr.slot[0];
+	spu_release(ctx);
+	return ret;
+}
+DEFINE_SIMPLE_ATTRIBUTE(spufs_decr_ops, spufs_decr_get, spufs_decr_set,
+			"%llx\n")
+
+static void spufs_decr_status_set(void *data, u64 val)
+{
+	struct spu_context *ctx = data;
+	struct spu_lscsa *lscsa = ctx->csa.lscsa;
+	spu_acquire_saved(ctx);
+	lscsa->decr_status.slot[0] = (u32) val;
+	spu_release(ctx);
+}
+
+static u64 spufs_decr_status_get(void *data)
+{
+	struct spu_context *ctx = data;
+	struct spu_lscsa *lscsa = ctx->csa.lscsa;
+	u64 ret;
+	spu_acquire_saved(ctx);
+	ret = lscsa->decr_status.slot[0];
+	spu_release(ctx);
+	return ret;
+}
+DEFINE_SIMPLE_ATTRIBUTE(spufs_decr_status_ops, spufs_decr_status_get,
+			spufs_decr_status_set, "%llx\n")
+
+static void spufs_spu_tag_mask_set(void *data, u64 val)
+{
+	struct spu_context *ctx = data;
+	struct spu_lscsa *lscsa = ctx->csa.lscsa;
+	spu_acquire_saved(ctx);
+	lscsa->tag_mask.slot[0] = (u32) val;
+	spu_release(ctx);
+}
+
+static u64 spufs_spu_tag_mask_get(void *data)
+{
+	struct spu_context *ctx = data;
+	struct spu_lscsa *lscsa = ctx->csa.lscsa;
+	u64 ret;
+	spu_acquire_saved(ctx);
+	ret = lscsa->tag_mask.slot[0];
+	spu_release(ctx);
+	return ret;
+}
+DEFINE_SIMPLE_ATTRIBUTE(spufs_spu_tag_mask_ops, spufs_spu_tag_mask_get,
+			spufs_spu_tag_mask_set, "%llx\n")
+
+static void spufs_event_mask_set(void *data, u64 val)
+{
+	struct spu_context *ctx = data;
+	struct spu_lscsa *lscsa = ctx->csa.lscsa;
+	spu_acquire_saved(ctx);
+	lscsa->event_mask.slot[0] = (u32) val;
+	spu_release(ctx);
+}
+
+static u64 spufs_event_mask_get(void *data)
+{
+	struct spu_context *ctx = data;
+	struct spu_lscsa *lscsa = ctx->csa.lscsa;
+	u64 ret;
+	spu_acquire_saved(ctx);
+	ret = lscsa->event_mask.slot[0];
+	spu_release(ctx);
+	return ret;
+}
+DEFINE_SIMPLE_ATTRIBUTE(spufs_event_mask_ops, spufs_event_mask_get,
+			spufs_event_mask_set, "%llx\n")
+
+static void spufs_srr0_set(void *data, u64 val)
+{
+	struct spu_context *ctx = data;
+	struct spu_lscsa *lscsa = ctx->csa.lscsa;
+	spu_acquire_saved(ctx);
+	lscsa->srr0.slot[0] = (u32) val;
+	spu_release(ctx);
+}
+
+static u64 spufs_srr0_get(void *data)
+{
+	struct spu_context *ctx = data;
+	struct spu_lscsa *lscsa = ctx->csa.lscsa;
+	u64 ret;
+	spu_acquire_saved(ctx);
+	ret = lscsa->srr0.slot[0];
+	spu_release(ctx);
+	return ret;
+}
+DEFINE_SIMPLE_ATTRIBUTE(spufs_srr0_ops, spufs_srr0_get, spufs_srr0_set,
+			"%llx\n")
+
+struct tree_descr spufs_dir_contents[] = {
+	{ "mem",  &spufs_mem_fops,  0666, },
+	{ "regs", &spufs_regs_fops,  0666, },
+	{ "mbox", &spufs_mbox_fops, 0444, },
+	{ "ibox", &spufs_ibox_fops, 0444, },
+	{ "wbox", &spufs_wbox_fops, 0222, },
+	{ "mbox_stat", &spufs_mbox_stat_fops, 0444, },
+	{ "ibox_stat", &spufs_ibox_stat_fops, 0444, },
+	{ "wbox_stat", &spufs_wbox_stat_fops, 0444, },
+	{ "signal1", &spufs_signal1_fops, 0666, },
+	{ "signal2", &spufs_signal2_fops, 0666, },
+	{ "signal1_type", &spufs_signal1_type, 0666, },
+	{ "signal2_type", &spufs_signal2_type, 0666, },
+	{ "npc", &spufs_npc_ops, 0666, },
+	{ "fpcr", &spufs_fpcr_fops, 0666, },
+	{ "decr", &spufs_decr_ops, 0666, },
+	{ "decr_status", &spufs_decr_status_ops, 0666, },
+	{ "spu_tag_mask", &spufs_spu_tag_mask_ops, 0666, },
+	{ "event_mask", &spufs_event_mask_ops, 0666, },
+	{ "srr0", &spufs_srr0_ops, 0666, },
+	{},
+};
diff --git a/arch/powerpc/platforms/cell/spufs/hw_ops.c b/arch/powerpc/platforms/cell/spufs/hw_ops.c
new file mode 100644
index 000000000000..5445719bff79
--- /dev/null
+++ b/arch/powerpc/platforms/cell/spufs/hw_ops.c
@@ -0,0 +1,255 @@
+/* hw_ops.c - query/set operations on active SPU context.
+ *
+ * Copyright (C) IBM 2005
+ * Author: Mark Nutter <mnutter@us.ibm.com>
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; either version 2, or (at your option)
+ * any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software
+ * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
+ */
+
+#include <linux/config.h>
+#include <linux/module.h>
+#include <linux/errno.h>
+#include <linux/sched.h>
+#include <linux/kernel.h>
+#include <linux/mm.h>
+#include <linux/poll.h>
+#include <linux/smp.h>
+#include <linux/smp_lock.h>
+#include <linux/stddef.h>
+#include <linux/unistd.h>
+
+#include <asm/io.h>
+#include <asm/spu.h>
+#include <asm/spu_csa.h>
+#include <asm/mmu_context.h>
+#include "spufs.h"
+
+static int spu_hw_mbox_read(struct spu_context *ctx, u32 * data)
+{
+	struct spu *spu = ctx->spu;
+	struct spu_problem __iomem *prob = spu->problem;
+	u32 mbox_stat;
+	int ret = 0;
+
+	spin_lock_irq(&spu->register_lock);
+	mbox_stat = in_be32(&prob->mb_stat_R);
+	if (mbox_stat & 0x0000ff) {
+		*data = in_be32(&prob->pu_mb_R);
+		ret = 4;
+	}
+	spin_unlock_irq(&spu->register_lock);
+	return ret;
+}
+
+static u32 spu_hw_mbox_stat_read(struct spu_context *ctx)
+{
+	return in_be32(&ctx->spu->problem->mb_stat_R);
+}
+
+static unsigned int spu_hw_mbox_stat_poll(struct spu_context *ctx,
+					  unsigned int events)
+{
+	struct spu *spu = ctx->spu;
+	int ret = 0;
+	u32 stat;
+
+	spin_lock_irq(&spu->register_lock);
+	stat = in_be32(&spu->problem->mb_stat_R);
+
+	/* if the requested event is there, return the poll
+	   mask, otherwise enable the interrupt to get notified,
+	   but first mark any pending interrupts as done so
+	   we don't get woken up unnecessarily */
+
+	if (events & (POLLIN | POLLRDNORM)) {
+		if (stat & 0xff0000)
+			ret |= POLLIN | POLLRDNORM;
+		else {
+			spu_int_stat_clear(spu, 2, 0x1);
+			spu_int_mask_or(spu, 2, 0x1);
+		}
+	}
+	if (events & (POLLOUT | POLLWRNORM)) {
+		if (stat & 0x00ff00)
+			ret = POLLOUT | POLLWRNORM;
+		else {
+			spu_int_stat_clear(spu, 2, 0x10);
+			spu_int_mask_or(spu, 2, 0x10);
+		}
+	}
+	spin_unlock_irq(&spu->register_lock);
+	return ret;
+}
+
+static int spu_hw_ibox_read(struct spu_context *ctx, u32 * data)
+{
+	struct spu *spu = ctx->spu;
+	struct spu_problem __iomem *prob = spu->problem;
+	struct spu_priv2 __iomem *priv2 = spu->priv2;
+	int ret;
+
+	spin_lock_irq(&spu->register_lock);
+	if (in_be32(&prob->mb_stat_R) & 0xff0000) {
+		/* read the first available word */
+		*data = in_be64(&priv2->puint_mb_R);
+		ret = 4;
+	} else {
+		/* make sure we get woken up by the interrupt */
+		spu_int_mask_or(spu, 2, 0x1);
+		ret = 0;
+	}
+	spin_unlock_irq(&spu->register_lock);
+	return ret;
+}
+
+static int spu_hw_wbox_write(struct spu_context *ctx, u32 data)
+{
+	struct spu *spu = ctx->spu;
+	struct spu_problem __iomem *prob = spu->problem;
+	int ret;
+
+	spin_lock_irq(&spu->register_lock);
+	if (in_be32(&prob->mb_stat_R) & 0x00ff00) {
+		/* we have space to write wbox_data to */
+		out_be32(&prob->spu_mb_W, data);
+		ret = 4;
+	} else {
+		/* make sure we get woken up by the interrupt when space
+		   becomes available */
+		spu_int_mask_or(spu, 2, 0x10);
+		ret = 0;
+	}
+	spin_unlock_irq(&spu->register_lock);
+	return ret;
+}
+
+static u32 spu_hw_signal1_read(struct spu_context *ctx)
+{
+	return in_be32(&ctx->spu->problem->signal_notify1);
+}
+
+static void spu_hw_signal1_write(struct spu_context *ctx, u32 data)
+{
+	out_be32(&ctx->spu->problem->signal_notify1, data);
+}
+
+static u32 spu_hw_signal2_read(struct spu_context *ctx)
+{
+	return in_be32(&ctx->spu->problem->signal_notify1);
+}
+
+static void spu_hw_signal2_write(struct spu_context *ctx, u32 data)
+{
+	out_be32(&ctx->spu->problem->signal_notify2, data);
+}
+
+static void spu_hw_signal1_type_set(struct spu_context *ctx, u64 val)
+{
+	struct spu *spu = ctx->spu;
+	struct spu_priv2 __iomem *priv2 = spu->priv2;
+	u64 tmp;
+
+	spin_lock_irq(&spu->register_lock);
+	tmp = in_be64(&priv2->spu_cfg_RW);
+	if (val)
+		tmp |= 1;
+	else
+		tmp &= ~1;
+	out_be64(&priv2->spu_cfg_RW, tmp);
+	spin_unlock_irq(&spu->register_lock);
+}
+
+static u64 spu_hw_signal1_type_get(struct spu_context *ctx)
+{
+	return ((in_be64(&ctx->spu->priv2->spu_cfg_RW) & 1) != 0);
+}
+
+static void spu_hw_signal2_type_set(struct spu_context *ctx, u64 val)
+{
+	struct spu *spu = ctx->spu;
+	struct spu_priv2 __iomem *priv2 = spu->priv2;
+	u64 tmp;
+
+	spin_lock_irq(&spu->register_lock);
+	tmp = in_be64(&priv2->spu_cfg_RW);
+	if (val)
+		tmp |= 2;
+	else
+		tmp &= ~2;
+	out_be64(&priv2->spu_cfg_RW, tmp);
+	spin_unlock_irq(&spu->register_lock);
+}
+
+static u64 spu_hw_signal2_type_get(struct spu_context *ctx)
+{
+	return ((in_be64(&ctx->spu->priv2->spu_cfg_RW) & 2) != 0);
+}
+
+static u32 spu_hw_npc_read(struct spu_context *ctx)
+{
+	return in_be32(&ctx->spu->problem->spu_npc_RW);
+}
+
+static void spu_hw_npc_write(struct spu_context *ctx, u32 val)
+{
+	out_be32(&ctx->spu->problem->spu_npc_RW, val);
+}
+
+static u32 spu_hw_status_read(struct spu_context *ctx)
+{
+	return in_be32(&ctx->spu->problem->spu_status_R);
+}
+
+static char *spu_hw_get_ls(struct spu_context *ctx)
+{
+	return ctx->spu->local_store;
+}
+
+static void spu_hw_runcntl_write(struct spu_context *ctx, u32 val)
+{
+	eieio();
+	out_be32(&ctx->spu->problem->spu_runcntl_RW, val);
+}
+
+static void spu_hw_runcntl_stop(struct spu_context *ctx)
+{
+	spin_lock_irq(&ctx->spu->register_lock);
+	out_be32(&ctx->spu->problem->spu_runcntl_RW, SPU_RUNCNTL_STOP);
+	while (in_be32(&ctx->spu->problem->spu_status_R) & SPU_STATUS_RUNNING)
+		cpu_relax();
+	spin_unlock_irq(&ctx->spu->register_lock);
+}
+
+struct spu_context_ops spu_hw_ops = {
+	.mbox_read = spu_hw_mbox_read,
+	.mbox_stat_read = spu_hw_mbox_stat_read,
+	.mbox_stat_poll = spu_hw_mbox_stat_poll,
+	.ibox_read = spu_hw_ibox_read,
+	.wbox_write = spu_hw_wbox_write,
+	.signal1_read = spu_hw_signal1_read,
+	.signal1_write = spu_hw_signal1_write,
+	.signal2_read = spu_hw_signal2_read,
+	.signal2_write = spu_hw_signal2_write,
+	.signal1_type_set = spu_hw_signal1_type_set,
+	.signal1_type_get = spu_hw_signal1_type_get,
+	.signal2_type_set = spu_hw_signal2_type_set,
+	.signal2_type_get = spu_hw_signal2_type_get,
+	.npc_read = spu_hw_npc_read,
+	.npc_write = spu_hw_npc_write,
+	.status_read = spu_hw_status_read,
+	.get_ls = spu_hw_get_ls,
+	.runcntl_write = spu_hw_runcntl_write,
+	.runcntl_stop = spu_hw_runcntl_stop,
+};
diff --git a/arch/powerpc/platforms/cell/spufs/inode.c b/arch/powerpc/platforms/cell/spufs/inode.c
new file mode 100644
index 000000000000..1f3507c75e90
--- /dev/null
+++ b/arch/powerpc/platforms/cell/spufs/inode.c
@@ -0,0 +1,486 @@
+/*
+ * SPU file system
+ *
+ * (C) Copyright IBM Deutschland Entwicklung GmbH 2005
+ *
+ * Author: Arnd Bergmann <arndb@de.ibm.com>
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; either version 2, or (at your option)
+ * any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software
+ * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
+ */
+
+#include <linux/file.h>
+#include <linux/fs.h>
+#include <linux/backing-dev.h>
+#include <linux/init.h>
+#include <linux/ioctl.h>
+#include <linux/module.h>
+#include <linux/mount.h>
+#include <linux/namei.h>
+#include <linux/pagemap.h>
+#include <linux/poll.h>
+#include <linux/slab.h>
+#include <linux/parser.h>
+
+#include <asm/io.h>
+#include <asm/semaphore.h>
+#include <asm/spu.h>
+#include <asm/uaccess.h>
+
+#include "spufs.h"
+
+static kmem_cache_t *spufs_inode_cache;
+
+static struct inode *
+spufs_alloc_inode(struct super_block *sb)
+{
+	struct spufs_inode_info *ei;
+
+	ei = kmem_cache_alloc(spufs_inode_cache, SLAB_KERNEL);
+	if (!ei)
+		return NULL;
+	return &ei->vfs_inode;
+}
+
+static void
+spufs_destroy_inode(struct inode *inode)
+{
+	kmem_cache_free(spufs_inode_cache, SPUFS_I(inode));
+}
+
+static void
+spufs_init_once(void *p, kmem_cache_t * cachep, unsigned long flags)
+{
+	struct spufs_inode_info *ei = p;
+
+	if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
+	    SLAB_CTOR_CONSTRUCTOR) {
+		inode_init_once(&ei->vfs_inode);
+	}
+}
+
+static struct inode *
+spufs_new_inode(struct super_block *sb, int mode)
+{
+	struct inode *inode;
+
+	inode = new_inode(sb);
+	if (!inode)
+		goto out;
+
+	inode->i_mode = mode;
+	inode->i_uid = current->fsuid;
+	inode->i_gid = current->fsgid;
+	inode->i_blksize = PAGE_CACHE_SIZE;
+	inode->i_blocks = 0;
+	inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
+out:
+	return inode;
+}
+
+static int
+spufs_setattr(struct dentry *dentry, struct iattr *attr)
+{
+	struct inode *inode = dentry->d_inode;
+
+	if ((attr->ia_valid & ATTR_SIZE) &&
+	    (attr->ia_size != inode->i_size))
+		return -EINVAL;
+	return inode_setattr(inode, attr);
+}
+
+
+static int
+spufs_new_file(struct super_block *sb, struct dentry *dentry,
+		struct file_operations *fops, int mode,
+		struct spu_context *ctx)
+{
+	static struct inode_operations spufs_file_iops = {
+		.setattr = spufs_setattr,
+	};
+	struct inode *inode;
+	int ret;
+
+	ret = -ENOSPC;
+	inode = spufs_new_inode(sb, S_IFREG | mode);
+	if (!inode)
+		goto out;
+
+	ret = 0;
+	inode->i_op = &spufs_file_iops;
+	inode->i_fop = fops;
+	inode->u.generic_ip = SPUFS_I(inode)->i_ctx = get_spu_context(ctx);
+	d_add(dentry, inode);
+out:
+	return ret;
+}
+
+static void
+spufs_delete_inode(struct inode *inode)
+{
+	if (SPUFS_I(inode)->i_ctx)
+		put_spu_context(SPUFS_I(inode)->i_ctx);
+	clear_inode(inode);
+}
+
+static void spufs_prune_dir(struct dentry *dir)
+{
+	struct dentry *dentry, *tmp;
+	down(&dir->d_inode->i_sem);
+	list_for_each_entry_safe(dentry, tmp, &dir->d_subdirs, d_child) {
+		spin_lock(&dcache_lock);
+		spin_lock(&dentry->d_lock);
+		if (!(d_unhashed(dentry)) && dentry->d_inode) {
+			dget_locked(dentry);
+			__d_drop(dentry);
+			spin_unlock(&dentry->d_lock);
+			simple_unlink(dir->d_inode, dentry);
+			spin_unlock(&dcache_lock);
+			dput(dentry);
+		} else {
+			spin_unlock(&dentry->d_lock);
+			spin_unlock(&dcache_lock);
+		}
+	}
+	shrink_dcache_parent(dir);
+	up(&dir->d_inode->i_sem);
+}
+
+static int spufs_rmdir(struct inode *root, struct dentry *dir_dentry)
+{
+	struct spu_context *ctx;
+
+	/* remove all entries */
+	down(&root->i_sem);
+	spufs_prune_dir(dir_dentry);
+	up(&root->i_sem);
+
+	/* We have to give up the mm_struct */
+	ctx = SPUFS_I(dir_dentry->d_inode)->i_ctx;
+	spu_forget(ctx);
+
+	/* XXX Do we need to hold i_sem here ? */
+	return simple_rmdir(root, dir_dentry);
+}
+
+static int spufs_fill_dir(struct dentry *dir, struct tree_descr *files,
+			  int mode, struct spu_context *ctx)
+{
+	struct dentry *dentry;
+	int ret;
+
+	while (files->name && files->name[0]) {
+		ret = -ENOMEM;
+		dentry = d_alloc_name(dir, files->name);
+		if (!dentry)
+			goto out;
+		ret = spufs_new_file(dir->d_sb, dentry, files->ops,
+					files->mode & mode, ctx);
+		if (ret)
+			goto out;
+		files++;
+	}
+	return 0;
+out:
+	spufs_prune_dir(dir);
+	return ret;
+}
+
+static int spufs_dir_close(struct inode *inode, struct file *file)
+{
+	struct inode *dir;
+	struct dentry *dentry;
+	int ret;
+
+	dentry = file->f_dentry;
+	dir = dentry->d_parent->d_inode;
+
+	ret = spufs_rmdir(dir, dentry);
+	WARN_ON(ret);
+
+	return dcache_dir_close(inode, file);
+}
+
+struct inode_operations spufs_dir_inode_operations = {
+	.lookup = simple_lookup,
+};
+
+struct file_operations spufs_context_fops = {
+	.open		= dcache_dir_open,
+	.release	= spufs_dir_close,
+	.llseek		= dcache_dir_lseek,
+	.read		= generic_read_dir,
+	.readdir	= dcache_readdir,
+	.fsync		= simple_sync_file,
+};
+
+static int
+spufs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
+{
+	int ret;
+	struct inode *inode;
+	struct spu_context *ctx;
+
+	ret = -ENOSPC;
+	inode = spufs_new_inode(dir->i_sb, mode | S_IFDIR);
+	if (!inode)
+		goto out;
+
+	if (dir->i_mode & S_ISGID) {
+		inode->i_gid = dir->i_gid;
+		inode->i_mode &= S_ISGID;
+	}
+	ctx = alloc_spu_context(inode->i_mapping);
+	SPUFS_I(inode)->i_ctx = ctx;
+	if (!ctx)
+		goto out_iput;
+
+	inode->i_op = &spufs_dir_inode_operations;
+	inode->i_fop = &simple_dir_operations;
+	ret = spufs_fill_dir(dentry, spufs_dir_contents, mode, ctx);
+	if (ret)
+		goto out_free_ctx;
+
+	d_instantiate(dentry, inode);
+	dget(dentry);
+	dir->i_nlink++;
+	dentry->d_inode->i_nlink++;
+	goto out;
+
+out_free_ctx:
+	put_spu_context(ctx);
+out_iput:
+	iput(inode);
+out:
+	return ret;
+}
+
+static int spufs_context_open(struct dentry *dentry, struct vfsmount *mnt)
+{
+	int ret;
+	struct file *filp;
+
+	ret = get_unused_fd();
+	if (ret < 0) {
+		dput(dentry);
+		mntput(mnt);
+		goto out;
+	}
+
+	filp = dentry_open(dentry, mnt, O_RDONLY);
+	if (IS_ERR(filp)) {
+		put_unused_fd(ret);
+		ret = PTR_ERR(filp);
+		goto out;
+	}
+
+	filp->f_op = &spufs_context_fops;
+	fd_install(ret, filp);
+out:
+	return ret;
+}
+
+static struct file_system_type spufs_type;
+
+long spufs_create_thread(struct nameidata *nd,
+			 unsigned int flags, mode_t mode)
+{
+	struct dentry *dentry;
+	int ret;
+
+	/* need to be at the root of spufs */
+	ret = -EINVAL;
+	if (nd->dentry->d_sb->s_type != &spufs_type ||
+	    nd->dentry != nd->dentry->d_sb->s_root)
+		goto out;
+
+	dentry = lookup_create(nd, 1);
+	ret = PTR_ERR(dentry);
+	if (IS_ERR(dentry))
+		goto out_dir;
+
+	ret = -EEXIST;
+	if (dentry->d_inode)
+		goto out_dput;
+
+	mode &= ~current->fs->umask;
+	ret = spufs_mkdir(nd->dentry->d_inode, dentry, mode & S_IRWXUGO);
+	if (ret)
+		goto out_dput;
+
+	/*
+	 * get references for dget and mntget, will be released
+	 * in error path of *_open().
+	 */
+	ret = spufs_context_open(dget(dentry), mntget(nd->mnt));
+	if (ret < 0)
+		spufs_rmdir(nd->dentry->d_inode, dentry);
+
+out_dput:
+	dput(dentry);
+out_dir:
+	up(&nd->dentry->d_inode->i_sem);
+out:
+	return ret;
+}
+
+/* File system initialization */
+enum {
+	Opt_uid, Opt_gid, Opt_err,
+};
+
+static match_table_t spufs_tokens = {
+	{ Opt_uid, "uid=%d" },
+	{ Opt_gid, "gid=%d" },
+	{ Opt_err, NULL  },
+};
+
+static int
+spufs_parse_options(char *options, struct inode *root)
+{
+	char *p;
+	substring_t args[MAX_OPT_ARGS];
+
+	while ((p = strsep(&options, ",")) != NULL) {
+		int token, option;
+
+		if (!*p)
+			continue;
+
+		token = match_token(p, spufs_tokens, args);
+		switch (token) {
+		case Opt_uid:
+			if (match_int(&args[0], &option))
+				return 0;
+			root->i_uid = option;
+			break;
+		case Opt_gid:
+			if (match_int(&args[0], &option))
+				return 0;
+			root->i_gid = option;
+			break;
+		default:
+			return 0;
+		}
+	}
+	return 1;
+}
+
+static int
+spufs_create_root(struct super_block *sb, void *data)
+{
+	struct inode *inode;
+	int ret;
+
+	ret = -ENOMEM;
+	inode = spufs_new_inode(sb, S_IFDIR | 0775);
+	if (!inode)
+		goto out;
+
+	inode->i_op = &spufs_dir_inode_operations;
+	inode->i_fop = &simple_dir_operations;
+	SPUFS_I(inode)->i_ctx = NULL;
+
+	ret = -EINVAL;
+	if (!spufs_parse_options(data, inode))
+		goto out_iput;
+
+	ret = -ENOMEM;
+	sb->s_root = d_alloc_root(inode);
+	if (!sb->s_root)
+		goto out_iput;
+
+	return 0;
+out_iput:
+	iput(inode);
+out:
+	return ret;
+}
+
+static int
+spufs_fill_super(struct super_block *sb, void *data, int silent)
+{
+	static struct super_operations s_ops = {
+		.alloc_inode = spufs_alloc_inode,
+		.destroy_inode = spufs_destroy_inode,
+		.statfs = simple_statfs,
+		.delete_inode = spufs_delete_inode,
+		.drop_inode = generic_delete_inode,
+	};
+
+	sb->s_maxbytes = MAX_LFS_FILESIZE;
+	sb->s_blocksize = PAGE_CACHE_SIZE;
+	sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
+	sb->s_magic = SPUFS_MAGIC;
+	sb->s_op = &s_ops;
+
+	return spufs_create_root(sb, data);
+}
+
+static struct super_block *
+spufs_get_sb(struct file_system_type *fstype, int flags,
+		const char *name, void *data)
+{
+	return get_sb_single(fstype, flags, data, spufs_fill_super);
+}
+
+static struct file_system_type spufs_type = {
+	.owner = THIS_MODULE,
+	.name = "spufs",
+	.get_sb = spufs_get_sb,
+	.kill_sb = kill_litter_super,
+};
+
+static int spufs_init(void)
+{
+	int ret;
+	ret = -ENOMEM;
+	spufs_inode_cache = kmem_cache_create("spufs_inode_cache",
+			sizeof(struct spufs_inode_info), 0,
+			SLAB_HWCACHE_ALIGN, spufs_init_once, NULL);
+
+	if (!spufs_inode_cache)
+		goto out;
+	if (spu_sched_init() != 0) {
+		kmem_cache_destroy(spufs_inode_cache);
+		goto out;
+	}
+	ret = register_filesystem(&spufs_type);
+	if (ret)
+		goto out_cache;
+	ret = register_spu_syscalls(&spufs_calls);
+	if (ret)
+		goto out_fs;
+	return 0;
+out_fs:
+	unregister_filesystem(&spufs_type);
+out_cache:
+	kmem_cache_destroy(spufs_inode_cache);
+out:
+	return ret;
+}
+module_init(spufs_init);
+
+static void spufs_exit(void)
+{
+	spu_sched_exit();
+	unregister_spu_syscalls(&spufs_calls);
+	unregister_filesystem(&spufs_type);
+	kmem_cache_destroy(spufs_inode_cache);
+}
+module_exit(spufs_exit);
+
+MODULE_LICENSE("GPL");
+MODULE_AUTHOR("Arnd Bergmann <arndb@de.ibm.com>");
+
diff --git a/arch/powerpc/platforms/cell/spufs/run.c b/arch/powerpc/platforms/cell/spufs/run.c
new file mode 100644
index 000000000000..18ea8866c61a
--- /dev/null
+++ b/arch/powerpc/platforms/cell/spufs/run.c
@@ -0,0 +1,131 @@
+#include <linux/wait.h>
+#include <linux/ptrace.h>
+
+#include <asm/spu.h>
+
+#include "spufs.h"
+
+/* interrupt-level stop callback function. */
+void spufs_stop_callback(struct spu *spu)
+{
+	struct spu_context *ctx = spu->ctx;
+
+	wake_up_all(&ctx->stop_wq);
+}
+
+static inline int spu_stopped(struct spu_context *ctx, u32 * stat)
+{
+	struct spu *spu;
+	u64 pte_fault;
+
+	*stat = ctx->ops->status_read(ctx);
+	if (ctx->state != SPU_STATE_RUNNABLE)
+		return 1;
+	spu = ctx->spu;
+	pte_fault = spu->dsisr &
+	    (MFC_DSISR_PTE_NOT_FOUND | MFC_DSISR_ACCESS_DENIED);
+	return (!(*stat & 0x1) || pte_fault || spu->class_0_pending) ? 1 : 0;
+}
+
+static inline int spu_run_init(struct spu_context *ctx, u32 * npc,
+			       u32 * status)
+{
+	int ret;
+
+	if ((ret = spu_acquire_runnable(ctx)) != 0)
+		return ret;
+	ctx->ops->npc_write(ctx, *npc);
+	ctx->ops->runcntl_write(ctx, SPU_RUNCNTL_RUNNABLE);
+	return 0;
+}
+
+static inline int spu_run_fini(struct spu_context *ctx, u32 * npc,
+			       u32 * status)
+{
+	int ret = 0;
+
+	*status = ctx->ops->status_read(ctx);
+	*npc = ctx->ops->npc_read(ctx);
+	spu_release(ctx);
+
+	if (signal_pending(current))
+		ret = -ERESTARTSYS;
+	if (unlikely(current->ptrace & PT_PTRACED)) {
+		if ((*status & SPU_STATUS_STOPPED_BY_STOP)
+		    && (*status >> SPU_STOP_STATUS_SHIFT) == 0x3fff) {
+			force_sig(SIGTRAP, current);
+			ret = -ERESTARTSYS;
+		}
+	}
+	return ret;
+}
+
+static inline int spu_reacquire_runnable(struct spu_context *ctx, u32 *npc,
+				         u32 *status)
+{
+	int ret;
+
+	if ((ret = spu_run_fini(ctx, npc, status)) != 0)
+		return ret;
+	if (*status & (SPU_STATUS_STOPPED_BY_STOP |
+		       SPU_STATUS_STOPPED_BY_HALT)) {
+		return *status;
+	}
+	if ((ret = spu_run_init(ctx, npc, status)) != 0)
+		return ret;
+	return 0;
+}
+
+static inline int spu_process_events(struct spu_context *ctx)
+{
+	struct spu *spu = ctx->spu;
+	u64 pte_fault = MFC_DSISR_PTE_NOT_FOUND | MFC_DSISR_ACCESS_DENIED;
+	int ret = 0;
+
+	if (spu->dsisr & pte_fault)
+		ret = spu_irq_class_1_bottom(spu);
+	if (spu->class_0_pending)
+		ret = spu_irq_class_0_bottom(spu);
+	if (!ret && signal_pending(current))
+		ret = -ERESTARTSYS;
+	return ret;
+}
+
+long spufs_run_spu(struct file *file, struct spu_context *ctx,
+		   u32 * npc, u32 * status)
+{
+	int ret;
+
+	if (down_interruptible(&ctx->run_sema))
+		return -ERESTARTSYS;
+
+	ret = spu_run_init(ctx, npc, status);
+	if (ret)
+		goto out;
+
+	do {
+		ret = spufs_wait(ctx->stop_wq, spu_stopped(ctx, status));
+		if (unlikely(ret))
+			break;
+		if (unlikely(ctx->state != SPU_STATE_RUNNABLE)) {
+			ret = spu_reacquire_runnable(ctx, npc, status);
+			if (ret)
+				goto out;
+			continue;
+		}
+		ret = spu_process_events(ctx);
+
+	} while (!ret && !(*status & (SPU_STATUS_STOPPED_BY_STOP |
+				      SPU_STATUS_STOPPED_BY_HALT)));
+
+	ctx->ops->runcntl_stop(ctx);
+	ret = spu_run_fini(ctx, npc, status);
+	if (!ret)
+		ret = *status;
+	spu_yield(ctx);
+
+out:
+	up(&ctx->run_sema);
+	return ret;
+}
+
diff --git a/arch/powerpc/platforms/cell/spufs/sched.c b/arch/powerpc/platforms/cell/spufs/sched.c
new file mode 100644
index 000000000000..963182fbd1aa
--- /dev/null
+++ b/arch/powerpc/platforms/cell/spufs/sched.c
@@ -0,0 +1,461 @@
+/* sched.c - SPU scheduler.
+ *
+ * Copyright (C) IBM 2005
+ * Author: Mark Nutter <mnutter@us.ibm.com>
+ *
+ * SPU scheduler, based on Linux thread priority.  For now use
+ * a simple "cooperative" yield model with no preemption.  SPU
+ * scheduling will eventually be preemptive: When a thread with
+ * a higher static priority gets ready to run, then an active SPU
+ * context will be preempted and returned to the waitq.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; either version 2, or (at your option)
+ * any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software
+ * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
+ */
+
+#undef DEBUG
+
+#include <linux/config.h>
+#include <linux/module.h>
+#include <linux/errno.h>
+#include <linux/sched.h>
+#include <linux/kernel.h>
+#include <linux/mm.h>
+#include <linux/completion.h>
+#include <linux/vmalloc.h>
+#include <linux/smp.h>
+#include <linux/smp_lock.h>
+#include <linux/stddef.h>
+#include <linux/unistd.h>
+
+#include <asm/io.h>
+#include <asm/mmu_context.h>
+#include <asm/spu.h>
+#include <asm/spu_csa.h>
+#include "spufs.h"
+
+#define SPU_MIN_TIMESLICE 	(100 * HZ / 1000)
+
+#define SPU_BITMAP_SIZE (((MAX_PRIO+BITS_PER_LONG)/BITS_PER_LONG)+1)
+struct spu_prio_array {
+	atomic_t nr_blocked;
+	unsigned long bitmap[SPU_BITMAP_SIZE];
+	wait_queue_head_t waitq[MAX_PRIO];
+};
+
+/* spu_runqueue - This is the main runqueue data structure for SPUs. */
+struct spu_runqueue {
+	struct semaphore sem;
+	unsigned long nr_active;
+	unsigned long nr_idle;
+	unsigned long nr_switches;
+	struct list_head active_list;
+	struct list_head idle_list;
+	struct spu_prio_array prio;
+};
+
+static struct spu_runqueue *spu_runqueues = NULL;
+
+static inline struct spu_runqueue *spu_rq(void)
+{
+	/* Future: make this a per-NODE array,
+	 * and use cpu_to_node(smp_processor_id())
+	 */
+	return spu_runqueues;
+}
+
+static inline struct spu *del_idle(struct spu_runqueue *rq)
+{
+	struct spu *spu;
+
+	BUG_ON(rq->nr_idle <= 0);
+	BUG_ON(list_empty(&rq->idle_list));
+	/* Future: Move SPU out of low-power SRI state. */
+	spu = list_entry(rq->idle_list.next, struct spu, sched_list);
+	list_del_init(&spu->sched_list);
+	rq->nr_idle--;
+	return spu;
+}
+
+static inline void del_active(struct spu_runqueue *rq, struct spu *spu)
+{
+	BUG_ON(rq->nr_active <= 0);
+	BUG_ON(list_empty(&rq->active_list));
+	list_del_init(&spu->sched_list);
+	rq->nr_active--;
+}
+
+static inline void add_idle(struct spu_runqueue *rq, struct spu *spu)
+{
+	/* Future: Put SPU into low-power SRI state. */
+	list_add_tail(&spu->sched_list, &rq->idle_list);
+	rq->nr_idle++;
+}
+
+static inline void add_active(struct spu_runqueue *rq, struct spu *spu)
+{
+	rq->nr_active++;
+	rq->nr_switches++;
+	list_add_tail(&spu->sched_list, &rq->active_list);
+}
+
+static void prio_wakeup(struct spu_runqueue *rq)
+{
+	if (atomic_read(&rq->prio.nr_blocked) && rq->nr_idle) {
+		int best = sched_find_first_bit(rq->prio.bitmap);
+		if (best < MAX_PRIO) {
+			wait_queue_head_t *wq = &rq->prio.waitq[best];
+			wake_up_interruptible_nr(wq, 1);
+		}
+	}
+}
+
+static void prio_wait(struct spu_runqueue *rq, struct spu_context *ctx,
+		      u64 flags)
+{
+	int prio = current->prio;
+	wait_queue_head_t *wq = &rq->prio.waitq[prio];
+	DEFINE_WAIT(wait);
+
+	__set_bit(prio, rq->prio.bitmap);
+	atomic_inc(&rq->prio.nr_blocked);
+	prepare_to_wait_exclusive(wq, &wait, TASK_INTERRUPTIBLE);
+	if (!signal_pending(current)) {
+		up(&rq->sem);
+		up_write(&ctx->state_sema);
+		pr_debug("%s: pid=%d prio=%d\n", __FUNCTION__,
+			 current->pid, current->prio);
+		schedule();
+		down_write(&ctx->state_sema);
+		down(&rq->sem);
+	}
+	finish_wait(wq, &wait);
+	atomic_dec(&rq->prio.nr_blocked);
+	if (!waitqueue_active(wq))
+		__clear_bit(prio, rq->prio.bitmap);
+}
+
+static inline int is_best_prio(struct spu_runqueue *rq)
+{
+	int best_prio;
+
+	best_prio = sched_find_first_bit(rq->prio.bitmap);
+	return (current->prio < best_prio) ? 1 : 0;
+}
+
+static inline void mm_needs_global_tlbie(struct mm_struct *mm)
+{
+	/* Global TLBIE broadcast required with SPEs. */
+#if (NR_CPUS > 1)
+	__cpus_setall(&mm->cpu_vm_mask, NR_CPUS);
+#else
+	__cpus_setall(&mm->cpu_vm_mask, NR_CPUS+1); /* is this ok? */
+#endif
+}
+
+static inline void bind_context(struct spu *spu, struct spu_context *ctx)
+{
+	pr_debug("%s: pid=%d SPU=%d\n", __FUNCTION__, current->pid,
+		 spu->number);
+	spu->ctx = ctx;
+	spu->flags = 0;
+	ctx->flags = 0;
+	ctx->spu = spu;
+	ctx->ops = &spu_hw_ops;
+	spu->pid = current->pid;
+	spu->prio = current->prio;
+	spu->mm = ctx->owner;
+	mm_needs_global_tlbie(spu->mm);
+	spu->ibox_callback = spufs_ibox_callback;
+	spu->wbox_callback = spufs_wbox_callback;
+	spu->stop_callback = spufs_stop_callback;
+	mb();
+	spu_unmap_mappings(ctx);
+	spu_restore(&ctx->csa, spu);
+	spu->timestamp = jiffies;
+}
+
+static inline void unbind_context(struct spu *spu, struct spu_context *ctx)
+{
+	pr_debug("%s: unbind pid=%d SPU=%d\n", __FUNCTION__,
+		 spu->pid, spu->number);
+	spu_unmap_mappings(ctx);
+	spu_save(&ctx->csa, spu);
+	spu->timestamp = jiffies;
+	ctx->state = SPU_STATE_SAVED;
+	spu->ibox_callback = NULL;
+	spu->wbox_callback = NULL;
+	spu->stop_callback = NULL;
+	spu->mm = NULL;
+	spu->pid = 0;
+	spu->prio = MAX_PRIO;
+	ctx->ops = &spu_backing_ops;
+	ctx->spu = NULL;
+	ctx->flags = 0;
+	spu->flags = 0;
+	spu->ctx = NULL;
+}
+
+static void spu_reaper(void *data)
+{
+	struct spu_context *ctx = data;
+	struct spu *spu;
+
+	down_write(&ctx->state_sema);
+	spu = ctx->spu;
+	if (spu && test_bit(SPU_CONTEXT_PREEMPT, &ctx->flags)) {
+		if (atomic_read(&spu->rq->prio.nr_blocked)) {
+			pr_debug("%s: spu=%d\n", __func__, spu->number);
+			ctx->ops->runcntl_stop(ctx);
+			spu_deactivate(ctx);
+			wake_up_all(&ctx->stop_wq);
+		} else {
+			clear_bit(SPU_CONTEXT_PREEMPT, &ctx->flags);
+		}
+	}
+	up_write(&ctx->state_sema);
+	put_spu_context(ctx);
+}
+
+static void schedule_spu_reaper(struct spu_runqueue *rq, struct spu *spu)
+{
+	struct spu_context *ctx = get_spu_context(spu->ctx);
+	unsigned long now = jiffies;
+	unsigned long expire = spu->timestamp + SPU_MIN_TIMESLICE;
+
+	set_bit(SPU_CONTEXT_PREEMPT, &ctx->flags);
+	INIT_WORK(&ctx->reap_work, spu_reaper, ctx);
+	if (time_after(now, expire))
+		schedule_work(&ctx->reap_work);
+	else
+		schedule_delayed_work(&ctx->reap_work, expire - now);
+}
+
+static void check_preempt_active(struct spu_runqueue *rq)
+{
+	struct list_head *p;
+	struct spu *worst = NULL;
+
+	list_for_each(p, &rq->active_list) {
+		struct spu *spu = list_entry(p, struct spu, sched_list);
+		struct spu_context *ctx = spu->ctx;
+		if (!test_bit(SPU_CONTEXT_PREEMPT, &ctx->flags)) {
+			if (!worst || (spu->prio > worst->prio)) {
+				worst = spu;
+			}
+		}
+	}
+	if (worst && (current->prio < worst->prio))
+		schedule_spu_reaper(rq, worst);
+}
+
+static struct spu *get_idle_spu(struct spu_context *ctx, u64 flags)
+{
+	struct spu_runqueue *rq;
+	struct spu *spu = NULL;
+
+	rq = spu_rq();
+	down(&rq->sem);
+	for (;;) {
+		if (rq->nr_idle > 0) {
+			if (is_best_prio(rq)) {
+				/* Fall through. */
+				spu = del_idle(rq);
+				break;
+			} else {
+				prio_wakeup(rq);
+				up(&rq->sem);
+				yield();
+				if (signal_pending(current)) {
+					return NULL;
+				}
+				rq = spu_rq();
+				down(&rq->sem);
+				continue;
+			}
+		} else {
+			check_preempt_active(rq);
+			prio_wait(rq, ctx, flags);
+			if (signal_pending(current)) {
+				prio_wakeup(rq);
+				spu = NULL;
+				break;
+			}
+			continue;
+		}
+	}
+	up(&rq->sem);
+	return spu;
+}
+
+static void put_idle_spu(struct spu *spu)
+{
+	struct spu_runqueue *rq = spu->rq;
+
+	down(&rq->sem);
+	add_idle(rq, spu);
+	prio_wakeup(rq);
+	up(&rq->sem);
+}
+
+static int get_active_spu(struct spu *spu)
+{
+	struct spu_runqueue *rq = spu->rq;
+	struct list_head *p;
+	struct spu *tmp;
+	int rc = 0;
+
+	down(&rq->sem);
+	list_for_each(p, &rq->active_list) {
+		tmp = list_entry(p, struct spu, sched_list);
+		if (tmp == spu) {
+			del_active(rq, spu);
+			rc = 1;
+			break;
+		}
+	}
+	up(&rq->sem);
+	return rc;
+}
+
+static void put_active_spu(struct spu *spu)
+{
+	struct spu_runqueue *rq = spu->rq;
+
+	down(&rq->sem);
+	add_active(rq, spu);
+	up(&rq->sem);
+}
+
+/* Lock order:
+ *	spu_activate() & spu_deactivate() require the
+ *	caller to have down_write(&ctx->state_sema).
+ *
+ *	The rq->sem is breifly held (inside or outside a
+ *	given ctx lock) for list management, but is never
+ *	held during save/restore.
+ */
+
+int spu_activate(struct spu_context *ctx, u64 flags)
+{
+	struct spu *spu;
+
+	if (ctx->spu)
+		return 0;
+	spu = get_idle_spu(ctx, flags);
+	if (!spu)
+		return (signal_pending(current)) ? -ERESTARTSYS : -EAGAIN;
+	bind_context(spu, ctx);
+	/*
+	 * We're likely to wait for interrupts on the same
+	 * CPU that we are now on, so send them here.
+	 */
+	spu_irq_setaffinity(spu, raw_smp_processor_id());
+	put_active_spu(spu);
+	return 0;
+}
+
+void spu_deactivate(struct spu_context *ctx)
+{
+	struct spu *spu;
+	int needs_idle;
+
+	spu = ctx->spu;
+	if (!spu)
+		return;
+	needs_idle = get_active_spu(spu);
+	unbind_context(spu, ctx);
+	if (needs_idle)
+		put_idle_spu(spu);
+}
+
+void spu_yield(struct spu_context *ctx)
+{
+	struct spu *spu;
+	int need_yield = 0;
+
+	down_write(&ctx->state_sema);
+	spu = ctx->spu;
+	if (spu && (sched_find_first_bit(spu->rq->prio.bitmap) < MAX_PRIO)) {
+		pr_debug("%s: yielding SPU %d\n", __FUNCTION__, spu->number);
+		spu_deactivate(ctx);
+		ctx->state = SPU_STATE_SAVED;
+		need_yield = 1;
+	} else if (spu) {
+		spu->prio = MAX_PRIO;
+	}
+	up_write(&ctx->state_sema);
+	if (unlikely(need_yield))
+		yield();
+}
+
+int __init spu_sched_init(void)
+{
+	struct spu_runqueue *rq;
+	struct spu *spu;
+	int i;
+
+	rq = spu_runqueues = kmalloc(sizeof(struct spu_runqueue), GFP_KERNEL);
+	if (!rq) {
+		printk(KERN_WARNING "%s: Unable to allocate runqueues.\n",
+		       __FUNCTION__);
+		return 1;
+	}
+	memset(rq, 0, sizeof(struct spu_runqueue));
+	init_MUTEX(&rq->sem);
+	INIT_LIST_HEAD(&rq->active_list);
+	INIT_LIST_HEAD(&rq->idle_list);
+	rq->nr_active = 0;
+	rq->nr_idle = 0;
+	rq->nr_switches = 0;
+	atomic_set(&rq->prio.nr_blocked, 0);
+	for (i = 0; i < MAX_PRIO; i++) {
+		init_waitqueue_head(&rq->prio.waitq[i]);
+		__clear_bit(i, rq->prio.bitmap);
+	}
+	__set_bit(MAX_PRIO, rq->prio.bitmap);
+	for (;;) {
+		spu = spu_alloc();
+		if (!spu)
+			break;
+		pr_debug("%s: adding SPU[%d]\n", __FUNCTION__, spu->number);
+		add_idle(rq, spu);
+		spu->rq = rq;
+		spu->timestamp = jiffies;
+	}
+	if (!rq->nr_idle) {
+		printk(KERN_WARNING "%s: No available SPUs.\n", __FUNCTION__);
+		kfree(rq);
+		return 1;
+	}
+	return 0;
+}
+
+void __exit spu_sched_exit(void)
+{
+	struct spu_runqueue *rq = spu_rq();
+	struct spu *spu;
+
+	if (!rq) {
+		printk(KERN_WARNING "%s: no runqueues!\n", __FUNCTION__);
+		return;
+	}
+	while (rq->nr_idle > 0) {
+		spu = del_idle(rq);
+		if (!spu)
+			break;
+		spu_free(spu);
+	}
+	kfree(rq);
+}
diff --git a/arch/powerpc/platforms/cell/spufs/spu_restore.c b/arch/powerpc/platforms/cell/spufs/spu_restore.c
new file mode 100644
index 000000000000..0bf723dcd677
--- /dev/null
+++ b/arch/powerpc/platforms/cell/spufs/spu_restore.c
@@ -0,0 +1,336 @@
+/*
+ * spu_restore.c
+ *
+ * (C) Copyright IBM Corp. 2005
+ *
+ * SPU-side context restore sequence outlined in
+ * Synergistic Processor Element Book IV
+ *
+ * Author: Mark Nutter <mnutter@us.ibm.com>
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; either version 2, or (at your option)
+ * any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software
+ * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
+ *
+ */
+
+
+#ifndef LS_SIZE
+#define LS_SIZE                 0x40000	/* 256K (in bytes) */
+#endif
+
+typedef unsigned int u32;
+typedef unsigned long long u64;
+
+#include <spu_intrinsics.h>
+#include <asm/spu_csa.h>
+#include "spu_utils.h"
+
+#define BR_INSTR		0x327fff80	/* br -4         */
+#define NOP_INSTR		0x40200000	/* nop           */
+#define HEQ_INSTR		0x7b000000	/* heq $0, $0    */
+#define STOP_INSTR		0x00000000	/* stop 0x0      */
+#define ILLEGAL_INSTR		0x00800000	/* illegal instr */
+#define RESTORE_COMPLETE	0x00003ffc	/* stop 0x3ffc   */
+
+static inline void fetch_regs_from_mem(addr64 lscsa_ea)
+{
+	unsigned int ls = (unsigned int)&regs_spill[0];
+	unsigned int size = sizeof(regs_spill);
+	unsigned int tag_id = 0;
+	unsigned int cmd = 0x40;	/* GET */
+
+	spu_writech(MFC_LSA, ls);
+	spu_writech(MFC_EAH, lscsa_ea.ui[0]);
+	spu_writech(MFC_EAL, lscsa_ea.ui[1]);
+	spu_writech(MFC_Size, size);
+	spu_writech(MFC_TagID, tag_id);
+	spu_writech(MFC_Cmd, cmd);
+}
+
+static inline void restore_upper_240kb(addr64 lscsa_ea)
+{
+	unsigned int ls = 16384;
+	unsigned int list = (unsigned int)&dma_list[0];
+	unsigned int size = sizeof(dma_list);
+	unsigned int tag_id = 0;
+	unsigned int cmd = 0x44;	/* GETL */
+
+	/* Restore, Step 4:
+	 *    Enqueue the GETL command (tag 0) to the MFC SPU command
+	 *    queue to transfer the upper 240 kb of LS from CSA.
+	 */
+	spu_writech(MFC_LSA, ls);
+	spu_writech(MFC_EAH, lscsa_ea.ui[0]);
+	spu_writech(MFC_EAL, list);
+	spu_writech(MFC_Size, size);
+	spu_writech(MFC_TagID, tag_id);
+	spu_writech(MFC_Cmd, cmd);
+}
+
+static inline void restore_decr(void)
+{
+	unsigned int offset;
+	unsigned int decr_running;
+	unsigned int decr;
+
+	/* Restore, Step 6:
+	 *    If the LSCSA "decrementer running" flag is set
+	 *    then write the SPU_WrDec channel with the
+	 *    decrementer value from LSCSA.
+	 */
+	offset = LSCSA_QW_OFFSET(decr_status);
+	decr_running = regs_spill[offset].slot[0];
+	if (decr_running) {
+		offset = LSCSA_QW_OFFSET(decr);
+		decr = regs_spill[offset].slot[0];
+		spu_writech(SPU_WrDec, decr);
+	}
+}
+
+static inline void write_ppu_mb(void)
+{
+	unsigned int offset;
+	unsigned int data;
+
+	/* Restore, Step 11:
+	 *    Write the MFC_WrOut_MB channel with the PPU_MB
+	 *    data from LSCSA.
+	 */
+	offset = LSCSA_QW_OFFSET(ppu_mb);
+	data = regs_spill[offset].slot[0];
+	spu_writech(SPU_WrOutMbox, data);
+}
+
+static inline void write_ppuint_mb(void)
+{
+	unsigned int offset;
+	unsigned int data;
+
+	/* Restore, Step 12:
+	 *    Write the MFC_WrInt_MB channel with the PPUINT_MB
+	 *    data from LSCSA.
+	 */
+	offset = LSCSA_QW_OFFSET(ppuint_mb);
+	data = regs_spill[offset].slot[0];
+	spu_writech(SPU_WrOutIntrMbox, data);
+}
+
+static inline void restore_fpcr(void)
+{
+	unsigned int offset;
+	vector unsigned int fpcr;
+
+	/* Restore, Step 13:
+	 *    Restore the floating-point status and control
+	 *    register from the LSCSA.
+	 */
+	offset = LSCSA_QW_OFFSET(fpcr);
+	fpcr = regs_spill[offset].v;
+	spu_mtfpscr(fpcr);
+}
+
+static inline void restore_srr0(void)
+{
+	unsigned int offset;
+	unsigned int srr0;
+
+	/* Restore, Step 14:
+	 *    Restore the SPU SRR0 data from the LSCSA.
+	 */
+	offset = LSCSA_QW_OFFSET(srr0);
+	srr0 = regs_spill[offset].slot[0];
+	spu_writech(SPU_WrSRR0, srr0);
+}
+
+static inline void restore_event_mask(void)
+{
+	unsigned int offset;
+	unsigned int event_mask;
+
+	/* Restore, Step 15:
+	 *    Restore the SPU_RdEventMsk data from the LSCSA.
+	 */
+	offset = LSCSA_QW_OFFSET(event_mask);
+	event_mask = regs_spill[offset].slot[0];
+	spu_writech(SPU_WrEventMask, event_mask);
+}
+
+static inline void restore_tag_mask(void)
+{
+	unsigned int offset;
+	unsigned int tag_mask;
+
+	/* Restore, Step 16:
+	 *    Restore the SPU_RdTagMsk data from the LSCSA.
+	 */
+	offset = LSCSA_QW_OFFSET(tag_mask);
+	tag_mask = regs_spill[offset].slot[0];
+	spu_writech(MFC_WrTagMask, tag_mask);
+}
+
+static inline void restore_complete(void)
+{
+	extern void exit_fini(void);
+	unsigned int *exit_instrs = (unsigned int *)exit_fini;
+	unsigned int offset;
+	unsigned int stopped_status;
+	unsigned int stopped_code;
+
+	/* Restore, Step 18:
+	 *    Issue a stop-and-signal instruction with
+	 *    "good context restore" signal value.
+	 *
+	 * Restore, Step 19:
+	 *    There may be additional instructions placed
+	 *    here by the PPE Sequence for SPU Context
+	 *    Restore in order to restore the correct
+	 *    "stopped state".
+	 *
+	 *    This step is handled here by analyzing the
+	 *    LSCSA.stopped_status and then modifying the
+	 *    exit() function to behave appropriately.
+	 */
+
+	offset = LSCSA_QW_OFFSET(stopped_status);
+	stopped_status = regs_spill[offset].slot[0];
+	stopped_code = regs_spill[offset].slot[1];
+
+	switch (stopped_status) {
+	case SPU_STOPPED_STATUS_P_I:
+		/* SPU_Status[P,I]=1.  Add illegal instruction
+		 * followed by stop-and-signal instruction after
+		 * end of restore code.
+		 */
+		exit_instrs[0] = RESTORE_COMPLETE;
+		exit_instrs[1] = ILLEGAL_INSTR;
+		exit_instrs[2] = STOP_INSTR | stopped_code;
+		break;
+	case SPU_STOPPED_STATUS_P_H:
+		/* SPU_Status[P,H]=1.  Add 'heq $0, $0' followed
+		 * by stop-and-signal instruction after end of
+		 * restore code.
+		 */
+		exit_instrs[0] = RESTORE_COMPLETE;
+		exit_instrs[1] = HEQ_INSTR;
+		exit_instrs[2] = STOP_INSTR | stopped_code;
+		break;
+	case SPU_STOPPED_STATUS_S_P:
+		/* SPU_Status[S,P]=1.  Add nop instruction
+		 * followed by 'br -4' after end of restore
+		 * code.
+		 */
+		exit_instrs[0] = RESTORE_COMPLETE;
+		exit_instrs[1] = STOP_INSTR | stopped_code;
+		exit_instrs[2] = NOP_INSTR;
+		exit_instrs[3] = BR_INSTR;
+		break;
+	case SPU_STOPPED_STATUS_S_I:
+		/* SPU_Status[S,I]=1.  Add  illegal instruction
+		 * followed by 'br -4' after end of restore code.
+		 */
+		exit_instrs[0] = RESTORE_COMPLETE;
+		exit_instrs[1] = ILLEGAL_INSTR;
+		exit_instrs[2] = NOP_INSTR;
+		exit_instrs[3] = BR_INSTR;
+		break;
+	case SPU_STOPPED_STATUS_I:
+		/* SPU_Status[I]=1. Add illegal instruction followed
+		 * by infinite loop after end of restore sequence.
+		 */
+		exit_instrs[0] = RESTORE_COMPLETE;
+		exit_instrs[1] = ILLEGAL_INSTR;
+		exit_instrs[2] = NOP_INSTR;
+		exit_instrs[3] = BR_INSTR;
+		break;
+	case SPU_STOPPED_STATUS_S:
+		/* SPU_Status[S]=1. Add two 'nop' instructions. */
+		exit_instrs[0] = RESTORE_COMPLETE;
+		exit_instrs[1] = NOP_INSTR;
+		exit_instrs[2] = NOP_INSTR;
+		exit_instrs[3] = BR_INSTR;
+		break;
+	case SPU_STOPPED_STATUS_H:
+		/* SPU_Status[H]=1. Add 'heq $0, $0' instruction
+		 * after end of restore code.
+		 */
+		exit_instrs[0] = RESTORE_COMPLETE;
+		exit_instrs[1] = HEQ_INSTR;
+		exit_instrs[2] = NOP_INSTR;
+		exit_instrs[3] = BR_INSTR;
+		break;
+	case SPU_STOPPED_STATUS_P:
+		/* SPU_Status[P]=1. Add stop-and-signal instruction
+		 * after end of restore code.
+		 */
+		exit_instrs[0] = RESTORE_COMPLETE;
+		exit_instrs[1] = STOP_INSTR | stopped_code;
+		break;
+	case SPU_STOPPED_STATUS_R:
+		/* SPU_Status[I,S,H,P,R]=0. Add infinite loop. */
+		exit_instrs[0] = RESTORE_COMPLETE;
+		exit_instrs[1] = NOP_INSTR;
+		exit_instrs[2] = NOP_INSTR;
+		exit_instrs[3] = BR_INSTR;
+		break;
+	default:
+		/* SPU_Status[R]=1. No additonal instructions. */
+		break;
+	}
+	spu_sync();
+}
+
+/**
+ * main - entry point for SPU-side context restore.
+ *
+ * This code deviates from the documented sequence in the
+ * following aspects:
+ *
+ * 	1. The EA for LSCSA is passed from PPE in the
+ *	   signal notification channels.
+ *	2. The register spill area is pulled by SPU
+ *	   into LS, rather than pushed by PPE.
+ *	3. All 128 registers are restored by exit().
+ *	4. The exit() function is modified at run
+ *	   time in order to properly restore the
+ *	   SPU_Status register.
+ */
+int main()
+{
+	addr64 lscsa_ea;
+
+	lscsa_ea.ui[0] = spu_readch(SPU_RdSigNotify1);
+	lscsa_ea.ui[1] = spu_readch(SPU_RdSigNotify2);
+	fetch_regs_from_mem(lscsa_ea);
+
+	set_event_mask();		/* Step 1.  */
+	set_tag_mask();			/* Step 2.  */
+	build_dma_list(lscsa_ea);	/* Step 3.  */
+	restore_upper_240kb(lscsa_ea);	/* Step 4.  */
+					/* Step 5: done by 'exit'. */
+	restore_decr();			/* Step 6. */
+	enqueue_putllc(lscsa_ea);	/* Step 7. */
+	set_tag_update();		/* Step 8. */
+	read_tag_status();		/* Step 9. */
+	read_llar_status();		/* Step 10. */
+	write_ppu_mb();			/* Step 11. */
+	write_ppuint_mb();		/* Step 12. */
+	restore_fpcr();			/* Step 13. */
+	restore_srr0();			/* Step 14. */
+	restore_event_mask();		/* Step 15. */
+	restore_tag_mask();		/* Step 16. */
+					/* Step 17. done by 'exit'. */
+	restore_complete();		/* Step 18. */
+
+	return 0;
+}
diff --git a/arch/powerpc/platforms/cell/spufs/spu_restore_crt0.S b/arch/powerpc/platforms/cell/spufs/spu_restore_crt0.S
new file mode 100644
index 000000000000..2905949debe1
--- /dev/null
+++ b/arch/powerpc/platforms/cell/spufs/spu_restore_crt0.S
@@ -0,0 +1,116 @@
+/*
+ * crt0_r.S: Entry function for SPU-side context restore.
+ *
+ * Copyright (C) 2005 IBM
+ *
+ * Entry and exit function for SPU-side of the context restore
+ * sequence.  Sets up an initial stack frame, then branches to
+ * 'main'.  On return, restores all 128 registers from the LSCSA
+ * and exits.
+ *
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; either version 2, or (at your option)
+ * any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software
+ * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
+ */
+
+#include <asm/spu_csa.h>
+
+.data
+.align 7
+.globl regs_spill
+regs_spill:
+.space SIZEOF_SPU_SPILL_REGS, 0x0
+
+.text
+.global _start
+_start:
+	/* Initialize the stack pointer to point to 16368
+	 * (16kb-16). The back chain pointer is initialized
+	 * to NULL.
+	 */
+	il      $0, 0
+	il      $SP, 16368
+	stqd    $0, 0($SP)
+
+	/* Allocate a minimum stack frame for the called main.
+	 * This is needed so that main has a place to save the
+	 * link register when it calls another function.
+	 */
+	stqd    $SP, -160($SP)
+	ai      $SP, $SP, -160
+
+	/* Call the program's main function. */
+	brsl    $0, main
+
+.global exit
+.global	_exit
+exit:
+_exit:
+	/* SPU Context Restore, Step 5: Restore the remaining 112 GPRs. */
+	ila     $3, regs_spill + 256
+restore_regs:
+	lqr     $4, restore_reg_insts
+restore_reg_loop:
+	ai      $4, $4, 4
+	.balignl 16, 0x40200000
+restore_reg_insts:       /* must be quad-word aligned. */
+	lqd     $16, 0($3)
+	lqd     $17, 16($3)
+	lqd     $18, 32($3)
+	lqd     $19, 48($3)
+	andi    $5, $4, 0x7F
+	stqr    $4, restore_reg_insts
+	ai      $3, $3, 64
+	brnz    $5, restore_reg_loop
+
+	/* SPU Context Restore Step 17: Restore the first 16 GPRs. */
+	lqa $0, regs_spill + 0
+	lqa $1, regs_spill + 16
+	lqa $2, regs_spill + 32
+	lqa $3, regs_spill + 48
+	lqa $4, regs_spill + 64
+	lqa $5, regs_spill + 80
+	lqa $6, regs_spill + 96
+	lqa $7, regs_spill + 112
+	lqa $8, regs_spill + 128
+	lqa $9, regs_spill + 144
+	lqa $10, regs_spill + 160
+	lqa $11, regs_spill + 176
+	lqa $12, regs_spill + 192
+	lqa $13, regs_spill + 208
+	lqa $14, regs_spill + 224
+	lqa $15, regs_spill + 240
+
+	/* Under normal circumstances, the 'exit' function
+	 * terminates with 'stop SPU_RESTORE_COMPLETE',
+	 * indicating that the SPU-side restore code has
+	 * completed.
+	 *
+	 * However it is possible that instructions immediately
+	 * following the 'stop 0x3ffc' have been modified at run
+	 * time so as to recreate the exact SPU_Status settings
+	 * from the application, e.g. illegal instruciton, halt,
+	 * etc.
+	 */
+.global exit_fini
+.global	_exit_fini
+exit_fini:
+_exit_fini:
+	stop	SPU_RESTORE_COMPLETE
+	stop	0
+	stop	0
+	stop	0
+
+	/* Pad the size of this crt0.o to be multiple of 16 bytes. */
+.balignl 16, 0x0
diff --git a/arch/powerpc/platforms/cell/spufs/spu_restore_dump.h_shipped b/arch/powerpc/platforms/cell/spufs/spu_restore_dump.h_shipped
new file mode 100644
index 000000000000..1b2355ff7036
--- /dev/null
+++ b/arch/powerpc/platforms/cell/spufs/spu_restore_dump.h_shipped
@@ -0,0 +1,231 @@
+/*
+ * spu_restore_dump.h: Copyright (C) 2005 IBM.
+ * Hex-dump auto generated from spu_restore.c.
+ * Do not edit!
+ */
+static unsigned int spu_restore_code[] __page_aligned = {
+0x40800000, 0x409ff801, 0x24000080, 0x24fd8081,
+0x1cd80081, 0x33001180, 0x42030003, 0x33800284,
+0x1c010204, 0x40200000, 0x40200000, 0x40200000,
+0x34000190, 0x34004191, 0x34008192, 0x3400c193,
+0x141fc205, 0x23fffd84, 0x1c100183, 0x217ffa85,
+0x3080a000, 0x3080a201, 0x3080a402, 0x3080a603,
+0x3080a804, 0x3080aa05, 0x3080ac06, 0x3080ae07,
+0x3080b008, 0x3080b209, 0x3080b40a, 0x3080b60b,
+0x3080b80c, 0x3080ba0d, 0x3080bc0e, 0x3080be0f,
+0x00003ffc, 0x00000000, 0x00000000, 0x00000000,
+0x01a00182, 0x3ec00083, 0xb0a14103, 0x01a00204,
+0x3ec10082, 0x4202800e, 0x04000703, 0xb0a14202,
+0x21a00803, 0x3fbf028d, 0x3f20068d, 0x3fbe0682,
+0x3fe30102, 0x21a00882, 0x3f82028f, 0x3fe3078f,
+0x3fbf0784, 0x3f200204, 0x3fbe0204, 0x3fe30204,
+0x04000203, 0x21a00903, 0x40848002, 0x21a00982,
+0x40800003, 0x21a00a03, 0x40802002, 0x21a00a82,
+0x21a00083, 0x40800082, 0x21a00b02, 0x10002818,
+0x40a80002, 0x32800007, 0x4207000c, 0x18008208,
+0x40a0000b, 0x4080020a, 0x40800709, 0x00200000,
+0x42070002, 0x3ac30384, 0x1cffc489, 0x00200000,
+0x18008383, 0x38830382, 0x4cffc486, 0x3ac28185,
+0xb0408584, 0x28830382, 0x1c020387, 0x38828182,
+0xb0408405, 0x1802c408, 0x28828182, 0x217ff886,
+0x04000583, 0x21a00803, 0x3fbe0682, 0x3fe30102,
+0x04000106, 0x21a00886, 0x04000603, 0x21a00903,
+0x40803c02, 0x21a00982, 0x40800003, 0x04000184,
+0x21a00a04, 0x40802202, 0x21a00a82, 0x42028005,
+0x34208702, 0x21002282, 0x21a00804, 0x21a00886,
+0x3fbf0782, 0x3f200102, 0x3fbe0102, 0x3fe30102,
+0x21a00902, 0x40804003, 0x21a00983, 0x21a00a04,
+0x40805a02, 0x21a00a82, 0x40800083, 0x21a00b83,
+0x01a00c02, 0x01a00d83, 0x3420c282, 0x21a00e02,
+0x34210283, 0x21a00f03, 0x34200284, 0x77400200,
+0x3421c282, 0x21a00702, 0x34218283, 0x21a00083,
+0x34214282, 0x21a00b02, 0x4200480c, 0x00200000,
+0x1c010286, 0x34220284, 0x34220302, 0x0f608203,
+0x5c024204, 0x3b81810b, 0x42013c02, 0x00200000,
+0x18008185, 0x38808183, 0x3b814182, 0x21004e84,
+0x4020007f, 0x35000100, 0x000004e0, 0x000002a0,
+0x000002e8, 0x00000428, 0x00000360, 0x000002e8,
+0x000004a0, 0x00000468, 0x000003c8, 0x00000360,
+0x409ffe02, 0x30801203, 0x40800204, 0x3ec40085,
+0x10009c09, 0x3ac10606, 0xb060c105, 0x4020007f,
+0x4020007f, 0x20801203, 0x38810602, 0xb0408586,
+0x28810602, 0x32004180, 0x34204702, 0x21a00382,
+0x4020007f, 0x327fdc80, 0x409ffe02, 0x30801203,
+0x40800204, 0x3ec40087, 0x40800405, 0x00200000,
+0x40800606, 0x3ac10608, 0x3ac14609, 0x3ac1860a,
+0xb060c107, 0x20801203, 0x41004003, 0x38810602,
+0x4020007f, 0xb0408188, 0x4020007f, 0x28810602,
+0x41201002, 0x38814603, 0x10009c09, 0xb060c109,
+0x4020007f, 0x28814603, 0x41193f83, 0x38818602,
+0x60ffc003, 0xb040818a, 0x28818602, 0x32003080,
+0x409ffe02, 0x30801203, 0x40800204, 0x3ec40087,
+0x41201008, 0x10009c14, 0x40800405, 0x3ac10609,
+0x40800606, 0x3ac1460a, 0xb060c107, 0x3ac1860b,
+0x20801203, 0x38810602, 0xb0408409, 0x28810602,
+0x38814603, 0xb060c40a, 0x4020007f, 0x28814603,
+0x41193f83, 0x38818602, 0x60ffc003, 0xb040818b,
+0x28818602, 0x32002380, 0x409ffe02, 0x30801204,
+0x40800205, 0x3ec40083, 0x40800406, 0x3ac14607,
+0x3ac18608, 0xb0810103, 0x41004002, 0x20801204,
+0x4020007f, 0x38814603, 0x10009c0b, 0xb060c107,
+0x4020007f, 0x4020007f, 0x28814603, 0x38818602,
+0x4020007f, 0x4020007f, 0xb0408588, 0x28818602,
+0x4020007f, 0x32001780, 0x409ffe02, 0x1000640e,
+0x40800204, 0x30801203, 0x40800405, 0x3ec40087,
+0x40800606, 0x3ac10608, 0x3ac14609, 0x3ac1860a,
+0xb060c107, 0x20801203, 0x413d8003, 0x38810602,
+0x4020007f, 0x327fd780, 0x409ffe02, 0x10007f0c,
+0x40800205, 0x30801204, 0x40800406, 0x3ec40083,
+0x3ac14607, 0x3ac18608, 0xb0810103, 0x413d8002,
+0x20801204, 0x38814603, 0x4020007f, 0x327feb80,
+0x409ffe02, 0x30801203, 0x40800204, 0x3ec40087,
+0x40800405, 0x1000650a, 0x40800606, 0x3ac10608,
+0x3ac14609, 0x3ac1860a, 0xb060c107, 0x20801203,
+0x38810602, 0xb0408588, 0x4020007f, 0x327fc980,
+0x00400000, 0x40800003, 0x4020007f, 0x35000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+};
diff --git a/arch/powerpc/platforms/cell/spufs/spu_save.c b/arch/powerpc/platforms/cell/spufs/spu_save.c
new file mode 100644
index 000000000000..196033b8a579
--- /dev/null
+++ b/arch/powerpc/platforms/cell/spufs/spu_save.c
@@ -0,0 +1,195 @@
+/*
+ * spu_save.c
+ *
+ * (C) Copyright IBM Corp. 2005
+ *
+ * SPU-side context save sequence outlined in
+ * Synergistic Processor Element Book IV
+ *
+ * Author: Mark Nutter <mnutter@us.ibm.com>
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; either version 2, or (at your option)
+ * any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software
+ * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
+ *
+ */
+
+
+#ifndef LS_SIZE
+#define LS_SIZE                 0x40000	/* 256K (in bytes) */
+#endif
+
+typedef unsigned int u32;
+typedef unsigned long long u64;
+
+#include <spu_intrinsics.h>
+#include <asm/spu_csa.h>
+#include "spu_utils.h"
+
+static inline void save_event_mask(void)
+{
+	unsigned int offset;
+
+	/* Save, Step 2:
+	 *    Read the SPU_RdEventMsk channel and save to the LSCSA.
+	 */
+	offset = LSCSA_QW_OFFSET(event_mask);
+	regs_spill[offset].slot[0] = spu_readch(SPU_RdEventStatMask);
+}
+
+static inline void save_tag_mask(void)
+{
+	unsigned int offset;
+
+	/* Save, Step 3:
+	 *    Read the SPU_RdTagMsk channel and save to the LSCSA.
+	 */
+	offset = LSCSA_QW_OFFSET(tag_mask);
+	regs_spill[offset].slot[0] = spu_readch(MFC_RdTagMask);
+}
+
+static inline void save_upper_240kb(addr64 lscsa_ea)
+{
+	unsigned int ls = 16384;
+	unsigned int list = (unsigned int)&dma_list[0];
+	unsigned int size = sizeof(dma_list);
+	unsigned int tag_id = 0;
+	unsigned int cmd = 0x24;	/* PUTL */
+
+	/* Save, Step 7:
+	 *    Enqueue the PUTL command (tag 0) to the MFC SPU command
+	 *    queue to transfer the remaining 240 kb of LS to CSA.
+	 */
+	spu_writech(MFC_LSA, ls);
+	spu_writech(MFC_EAH, lscsa_ea.ui[0]);
+	spu_writech(MFC_EAL, list);
+	spu_writech(MFC_Size, size);
+	spu_writech(MFC_TagID, tag_id);
+	spu_writech(MFC_Cmd, cmd);
+}
+
+static inline void save_fpcr(void)
+{
+	// vector unsigned int fpcr;
+	unsigned int offset;
+
+	/* Save, Step 9:
+	 *    Issue the floating-point status and control register
+	 *    read instruction, and save to the LSCSA.
+	 */
+	offset = LSCSA_QW_OFFSET(fpcr);
+	regs_spill[offset].v = spu_mffpscr();
+}
+
+static inline void save_decr(void)
+{
+	unsigned int offset;
+
+	/* Save, Step 10:
+	 *    Read and save the SPU_RdDec channel data to
+	 *    the LSCSA.
+	 */
+	offset = LSCSA_QW_OFFSET(decr);
+	regs_spill[offset].slot[0] = spu_readch(SPU_RdDec);
+}
+
+static inline void save_srr0(void)
+{
+	unsigned int offset;
+
+	/* Save, Step 11:
+	 *    Read and save the SPU_WSRR0 channel data to
+	 *    the LSCSA.
+	 */
+	offset = LSCSA_QW_OFFSET(srr0);
+	regs_spill[offset].slot[0] = spu_readch(SPU_RdSRR0);
+}
+
+static inline void spill_regs_to_mem(addr64 lscsa_ea)
+{
+	unsigned int ls = (unsigned int)&regs_spill[0];
+	unsigned int size = sizeof(regs_spill);
+	unsigned int tag_id = 0;
+	unsigned int cmd = 0x20;	/* PUT */
+
+	/* Save, Step 13:
+	 *    Enqueue a PUT command (tag 0) to send the LSCSA
+	 *    to the CSA.
+	 */
+	spu_writech(MFC_LSA, ls);
+	spu_writech(MFC_EAH, lscsa_ea.ui[0]);
+	spu_writech(MFC_EAL, lscsa_ea.ui[1]);
+	spu_writech(MFC_Size, size);
+	spu_writech(MFC_TagID, tag_id);
+	spu_writech(MFC_Cmd, cmd);
+}
+
+static inline void enqueue_sync(addr64 lscsa_ea)
+{
+	unsigned int tag_id = 0;
+	unsigned int cmd = 0xCC;
+
+	/* Save, Step 14:
+	 *    Enqueue an MFC_SYNC command (tag 0).
+	 */
+	spu_writech(MFC_TagID, tag_id);
+	spu_writech(MFC_Cmd, cmd);
+}
+
+static inline void save_complete(void)
+{
+	/* Save, Step 18:
+	 *    Issue a stop-and-signal instruction indicating
+	 *    "save complete".  Note: This function will not
+	 *    return!!
+	 */
+	spu_stop(SPU_SAVE_COMPLETE);
+}
+
+/**
+ * main - entry point for SPU-side context save.
+ *
+ * This code deviates from the documented sequence as follows:
+ *
+ *      1. The EA for LSCSA is passed from PPE in the
+ *         signal notification channels.
+ *      2. All 128 registers are saved by crt0.o.
+ */
+int main()
+{
+	addr64 lscsa_ea;
+
+	lscsa_ea.ui[0] = spu_readch(SPU_RdSigNotify1);
+	lscsa_ea.ui[1] = spu_readch(SPU_RdSigNotify2);
+
+	/* Step 1: done by exit(). */
+	save_event_mask();	/* Step 2.  */
+	save_tag_mask();	/* Step 3.  */
+	set_event_mask();	/* Step 4.  */
+	set_tag_mask();		/* Step 5.  */
+	build_dma_list(lscsa_ea);	/* Step 6.  */
+	save_upper_240kb(lscsa_ea);	/* Step 7.  */
+	/* Step 8: done by exit(). */
+	save_fpcr();		/* Step 9.  */
+	save_decr();		/* Step 10. */
+	save_srr0();		/* Step 11. */
+	enqueue_putllc(lscsa_ea);	/* Step 12. */
+	spill_regs_to_mem(lscsa_ea);	/* Step 13. */
+	enqueue_sync(lscsa_ea);	/* Step 14. */
+	set_tag_update();	/* Step 15. */
+	read_tag_status();	/* Step 16. */
+	read_llar_status();	/* Step 17. */
+	save_complete();	/* Step 18. */
+
+	return 0;
+}
diff --git a/arch/powerpc/platforms/cell/spufs/spu_save_crt0.S b/arch/powerpc/platforms/cell/spufs/spu_save_crt0.S
new file mode 100644
index 000000000000..6659d6a66faa
--- /dev/null
+++ b/arch/powerpc/platforms/cell/spufs/spu_save_crt0.S
@@ -0,0 +1,102 @@
+/*
+ * crt0_s.S: Entry function for SPU-side context save.
+ *
+ * Copyright (C) 2005 IBM
+ *
+ * Entry function for SPU-side of the context save sequence.
+ * Saves all 128 GPRs, sets up an initial stack frame, then
+ * branches to 'main'.
+ *
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; either version 2, or (at your option)
+ * any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software
+ * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
+ */
+
+#include <asm/spu_csa.h>
+
+.data
+.align 7
+.globl regs_spill
+regs_spill:
+.space SIZEOF_SPU_SPILL_REGS, 0x0
+
+.text
+.global _start
+_start:
+	/* SPU Context Save Step 1: Save the first 16 GPRs. */
+	stqa $0, regs_spill + 0
+	stqa $1, regs_spill + 16
+	stqa $2, regs_spill + 32
+	stqa $3, regs_spill + 48
+	stqa $4, regs_spill + 64
+	stqa $5, regs_spill + 80
+	stqa $6, regs_spill + 96
+	stqa $7, regs_spill + 112
+	stqa $8, regs_spill + 128
+	stqa $9, regs_spill + 144
+	stqa $10, regs_spill + 160
+	stqa $11, regs_spill + 176
+	stqa $12, regs_spill + 192
+	stqa $13, regs_spill + 208
+	stqa $14, regs_spill + 224
+	stqa $15, regs_spill + 240
+
+	/* SPU Context Save, Step 8: Save the remaining 112 GPRs. */
+	ila     $3, regs_spill + 256
+save_regs:
+	lqr     $4, save_reg_insts
+save_reg_loop:
+	ai      $4, $4, 4
+	.balignl 16, 0x40200000
+save_reg_insts:       /* must be quad-word aligned. */
+	stqd    $16, 0($3)
+	stqd    $17, 16($3)
+	stqd    $18, 32($3)
+	stqd    $19, 48($3)
+	andi    $5, $4, 0x7F
+	stqr    $4, save_reg_insts
+	ai      $3, $3, 64
+	brnz    $5, save_reg_loop
+
+	/* Initialize the stack pointer to point to 16368
+	 * (16kb-16). The back chain pointer is initialized
+	 * to NULL.
+	 */
+	il	$0, 0
+	il	$SP, 16368
+	stqd	$0, 0($SP)
+
+	/* Allocate a minimum stack frame for the called main.
+	 * This is needed so that main has a place to save the
+	 * link register when it calls another function.
+	 */
+	stqd	$SP, -160($SP)
+	ai	$SP, $SP, -160
+
+	/* Call the program's main function. */
+	brsl	$0, main
+
+	/* In this case main should not return; if it does
+	 * there has been an error in the sequence.  Execute
+	 * stop-and-signal with code=0.
+	 */
+.global exit
+.global	_exit
+exit:
+_exit:
+	stop	0x0
+
+	/* Pad the size of this crt0.o to be multiple of 16 bytes. */
+.balignl 16, 0x0
+
diff --git a/arch/powerpc/platforms/cell/spufs/spu_save_dump.h_shipped b/arch/powerpc/platforms/cell/spufs/spu_save_dump.h_shipped
new file mode 100644
index 000000000000..39e54003f1df
--- /dev/null
+++ b/arch/powerpc/platforms/cell/spufs/spu_save_dump.h_shipped
@@ -0,0 +1,191 @@
+/*
+ * spu_save_dump.h: Copyright (C) 2005 IBM.
+ * Hex-dump auto generated from spu_save.c.
+ * Do not edit!
+ */
+static unsigned int spu_save_code[] __page_aligned = {
+0x20805000, 0x20805201, 0x20805402, 0x20805603,
+0x20805804, 0x20805a05, 0x20805c06, 0x20805e07,
+0x20806008, 0x20806209, 0x2080640a, 0x2080660b,
+0x2080680c, 0x20806a0d, 0x20806c0e, 0x20806e0f,
+0x4201c003, 0x33800184, 0x1c010204, 0x40200000,
+0x24000190, 0x24004191, 0x24008192, 0x2400c193,
+0x141fc205, 0x23fffd84, 0x1c100183, 0x217ffb85,
+0x40800000, 0x409ff801, 0x24000080, 0x24fd8081,
+0x1cd80081, 0x33000180, 0x00000000, 0x00000000,
+0x01a00182, 0x3ec00083, 0xb1c38103, 0x01a00204,
+0x3ec10082, 0x4201400d, 0xb1c38202, 0x01a00583,
+0x34218682, 0x3ed80684, 0xb0408184, 0x24218682,
+0x01a00603, 0x00200000, 0x34214682, 0x3ed40684,
+0xb0408184, 0x40800003, 0x24214682, 0x21a00083,
+0x40800082, 0x21a00b02, 0x4020007f, 0x1000251e,
+0x40a80002, 0x32800008, 0x4205c00c, 0x00200000,
+0x40a0000b, 0x3f82070f, 0x4080020a, 0x40800709,
+0x3fe3078f, 0x3fbf0783, 0x3f200183, 0x3fbe0183,
+0x3fe30187, 0x18008387, 0x4205c002, 0x3ac30404,
+0x1cffc489, 0x00200000, 0x18008403, 0x38830402,
+0x4cffc486, 0x3ac28185, 0xb0408584, 0x28830402,
+0x1c020408, 0x38828182, 0xb0408385, 0x1802c387,
+0x28828182, 0x217ff886, 0x04000582, 0x32800007,
+0x21a00802, 0x3fbf0705, 0x3f200285, 0x3fbe0285,
+0x3fe30285, 0x21a00885, 0x04000603, 0x21a00903,
+0x40803c02, 0x21a00982, 0x04000386, 0x21a00a06,
+0x40801202, 0x21a00a82, 0x73000003, 0x24200683,
+0x01a00404, 0x00200000, 0x34204682, 0x3ec40683,
+0xb0408203, 0x24204682, 0x01a00783, 0x00200000,
+0x3421c682, 0x3edc0684, 0xb0408184, 0x2421c682,
+0x21a00806, 0x21a00885, 0x3fbf0784, 0x3f200204,
+0x3fbe0204, 0x3fe30204, 0x21a00904, 0x40804002,
+0x21a00982, 0x21a00a06, 0x40805a02, 0x21a00a82,
+0x04000683, 0x21a00803, 0x21a00885, 0x21a00904,
+0x40848002, 0x21a00982, 0x21a00a06, 0x40801002,
+0x21a00a82, 0x21a00a06, 0x40806602, 0x00200000,
+0x35800009, 0x21a00a82, 0x40800083, 0x21a00b83,
+0x01a00c02, 0x01a00d83, 0x00003ffb, 0x40800003,
+0x4020007f, 0x35000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+0x00000000, 0x00000000, 0x00000000, 0x00000000,
+};
diff --git a/arch/powerpc/platforms/cell/spufs/spu_utils.h b/arch/powerpc/platforms/cell/spufs/spu_utils.h
new file mode 100644
index 000000000000..58359feb6c95
--- /dev/null
+++ b/arch/powerpc/platforms/cell/spufs/spu_utils.h
@@ -0,0 +1,160 @@
+/*
+ * utils.h: Utilities for SPU-side of the context switch operation.
+ *
+ * (C) Copyright IBM 2005
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; either version 2, or (at your option)
+ * any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software
+ * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
+ */
+
+#ifndef _SPU_CONTEXT_UTILS_H_
+#define _SPU_CONTEXT_UTILS_H_
+
+/*
+ * 64-bit safe EA.
+ */
+typedef union {
+	unsigned long long ull;
+	unsigned int ui[2];
+} addr64;
+
+/*
+ * 128-bit register template.
+ */
+typedef union {
+	unsigned int slot[4];
+	vector unsigned int v;
+} spu_reg128v;
+
+/*
+ * DMA list structure.
+ */
+struct dma_list_elem {
+	unsigned int size;
+	unsigned int ea_low;
+};
+
+/*
+ * Declare storage for 8-byte aligned DMA list.
+ */
+struct dma_list_elem dma_list[15] __attribute__ ((aligned(8)));
+
+/*
+ * External definition for storage
+ * declared in crt0.
+ */
+extern spu_reg128v regs_spill[NR_SPU_SPILL_REGS];
+
+/*
+ * Compute LSCSA byte offset for a given field.
+ */
+static struct spu_lscsa *dummy = (struct spu_lscsa *)0;
+#define LSCSA_BYTE_OFFSET(_field)  \
+	((char *)(&(dummy->_field)) - (char *)(&(dummy->gprs[0].slot[0])))
+#define LSCSA_QW_OFFSET(_field)  (LSCSA_BYTE_OFFSET(_field) >> 4)
+
+static inline void set_event_mask(void)
+{
+	unsigned int event_mask = 0;
+
+	/* Save, Step 4:
+	 * Restore, Step 1:
+	 *    Set the SPU_RdEventMsk channel to zero to mask
+	 *    all events.
+	 */
+	spu_writech(SPU_WrEventMask, event_mask);
+}
+
+static inline void set_tag_mask(void)
+{
+	unsigned int tag_mask = 1;
+
+	/* Save, Step 5:
+	 * Restore, Step 2:
+	 *    Set the SPU_WrTagMsk channel to '01' to unmask
+	 *    only tag group 0.
+	 */
+	spu_writech(MFC_WrTagMask, tag_mask);
+}
+
+static inline void build_dma_list(addr64 lscsa_ea)
+{
+	unsigned int ea_low;
+	int i;
+
+	/* Save, Step 6:
+	 * Restore, Step 3:
+	 *    Update the effective address for the CSA in the
+	 *    pre-canned DMA-list in local storage.
+	 */
+	ea_low = lscsa_ea.ui[1];
+	ea_low += LSCSA_BYTE_OFFSET(ls[16384]);
+
+	for (i = 0; i < 15; i++, ea_low += 16384) {
+		dma_list[i].size = 16384;
+		dma_list[i].ea_low = ea_low;
+	}
+}
+
+static inline void enqueue_putllc(addr64 lscsa_ea)
+{
+	unsigned int ls = 0;
+	unsigned int size = 128;
+	unsigned int tag_id = 0;
+	unsigned int cmd = 0xB4;	/* PUTLLC */
+
+	/* Save, Step 12:
+	 * Restore, Step 7:
+	 *    Send a PUTLLC (tag 0) command to the MFC using
+	 *    an effective address in the CSA in order to
+	 *    remove any possible lock-line reservation.
+	 */
+	spu_writech(MFC_LSA, ls);
+	spu_writech(MFC_EAH, lscsa_ea.ui[0]);
+	spu_writech(MFC_EAL, lscsa_ea.ui[1]);
+	spu_writech(MFC_Size, size);
+	spu_writech(MFC_TagID, tag_id);
+	spu_writech(MFC_Cmd, cmd);
+}
+
+static inline void set_tag_update(void)
+{
+	unsigned int update_any = 1;
+
+	/* Save, Step 15:
+	 * Restore, Step 8:
+	 *    Write the MFC_TagUpdate channel with '01'.
+	 */
+	spu_writech(MFC_WrTagUpdate, update_any);
+}
+
+static inline void read_tag_status(void)
+{
+	/* Save, Step 16:
+	 * Restore, Step 9:
+	 *    Read the MFC_TagStat channel data.
+	 */
+	spu_readch(MFC_RdTagStat);
+}
+
+static inline void read_llar_status(void)
+{
+	/* Save, Step 17:
+	 * Restore, Step 10:
+	 *    Read the MFC_AtomicStat channel data.
+	 */
+	spu_readch(MFC_RdAtomicStat);
+}
+
+#endif				/* _SPU_CONTEXT_UTILS_H_ */
diff --git a/arch/powerpc/platforms/cell/spufs/spufs.h b/arch/powerpc/platforms/cell/spufs/spufs.h
new file mode 100644
index 000000000000..db2601f0abd5
--- /dev/null
+++ b/arch/powerpc/platforms/cell/spufs/spufs.h
@@ -0,0 +1,163 @@
+/*
+ * SPU file system
+ *
+ * (C) Copyright IBM Deutschland Entwicklung GmbH 2005
+ *
+ * Author: Arnd Bergmann <arndb@de.ibm.com>
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; either version 2, or (at your option)
+ * any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software
+ * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
+ */
+#ifndef SPUFS_H
+#define SPUFS_H
+
+#include <linux/kref.h>
+#include <linux/rwsem.h>
+#include <linux/spinlock.h>
+#include <linux/fs.h>
+
+#include <asm/spu.h>
+#include <asm/spu_csa.h>
+
+/* The magic number for our file system */
+enum {
+	SPUFS_MAGIC = 0x23c9b64e,
+};
+
+struct spu_context_ops;
+
+#define SPU_CONTEXT_PREEMPT          0UL
+
+struct spu_context {
+	struct spu *spu;		  /* pointer to a physical SPU */
+	struct spu_state csa;		  /* SPU context save area. */
+	spinlock_t mmio_lock;		  /* protects mmio access */
+	struct address_space *local_store;/* local store backing store */
+
+	enum { SPU_STATE_RUNNABLE, SPU_STATE_SAVED } state;
+	struct rw_semaphore state_sema;
+	struct semaphore run_sema;
+
+	struct mm_struct *owner;
+
+	struct kref kref;
+	wait_queue_head_t ibox_wq;
+	wait_queue_head_t wbox_wq;
+	wait_queue_head_t stop_wq;
+	struct fasync_struct *ibox_fasync;
+	struct fasync_struct *wbox_fasync;
+	struct spu_context_ops *ops;
+	struct work_struct reap_work;
+	u64 flags;
+};
+
+/* SPU context query/set operations. */
+struct spu_context_ops {
+	int (*mbox_read) (struct spu_context * ctx, u32 * data);
+	 u32(*mbox_stat_read) (struct spu_context * ctx);
+	unsigned int (*mbox_stat_poll)(struct spu_context *ctx,
+					unsigned int events);
+	int (*ibox_read) (struct spu_context * ctx, u32 * data);
+	int (*wbox_write) (struct spu_context * ctx, u32 data);
+	 u32(*signal1_read) (struct spu_context * ctx);
+	void (*signal1_write) (struct spu_context * ctx, u32 data);
+	 u32(*signal2_read) (struct spu_context * ctx);
+	void (*signal2_write) (struct spu_context * ctx, u32 data);
+	void (*signal1_type_set) (struct spu_context * ctx, u64 val);
+	 u64(*signal1_type_get) (struct spu_context * ctx);
+	void (*signal2_type_set) (struct spu_context * ctx, u64 val);
+	 u64(*signal2_type_get) (struct spu_context * ctx);
+	 u32(*npc_read) (struct spu_context * ctx);
+	void (*npc_write) (struct spu_context * ctx, u32 data);
+	 u32(*status_read) (struct spu_context * ctx);
+	char*(*get_ls) (struct spu_context * ctx);
+	void (*runcntl_write) (struct spu_context * ctx, u32 data);
+	void (*runcntl_stop) (struct spu_context * ctx);
+};
+
+extern struct spu_context_ops spu_hw_ops;
+extern struct spu_context_ops spu_backing_ops;
+
+struct spufs_inode_info {
+	struct spu_context *i_ctx;
+	struct inode vfs_inode;
+};
+#define SPUFS_I(inode) \
+	container_of(inode, struct spufs_inode_info, vfs_inode)
+
+extern struct tree_descr spufs_dir_contents[];
+
+/* system call implementation */
+long spufs_run_spu(struct file *file,
+		   struct spu_context *ctx, u32 *npc, u32 *status);
+long spufs_create_thread(struct nameidata *nd,
+			 unsigned int flags, mode_t mode);
+extern struct file_operations spufs_context_fops;
+
+/* context management */
+struct spu_context * alloc_spu_context(struct address_space *local_store);
+void destroy_spu_context(struct kref *kref);
+struct spu_context * get_spu_context(struct spu_context *ctx);
+int put_spu_context(struct spu_context *ctx);
+void spu_unmap_mappings(struct spu_context *ctx);
+
+void spu_forget(struct spu_context *ctx);
+void spu_acquire(struct spu_context *ctx);
+void spu_release(struct spu_context *ctx);
+int spu_acquire_runnable(struct spu_context *ctx);
+void spu_acquire_saved(struct spu_context *ctx);
+
+int spu_activate(struct spu_context *ctx, u64 flags);
+void spu_deactivate(struct spu_context *ctx);
+void spu_yield(struct spu_context *ctx);
+int __init spu_sched_init(void);
+void __exit spu_sched_exit(void);
+
+/*
+ * spufs_wait
+ * 	Same as wait_event_interruptible(), except that here
+ *	we need to call spu_release(ctx) before sleeping, and
+ *	then spu_acquire(ctx) when awoken.
+ */
+
+#define spufs_wait(wq, condition)					\
+({									\
+	int __ret = 0;							\
+	DEFINE_WAIT(__wait);						\
+	for (;;) {							\
+		prepare_to_wait(&(wq), &__wait, TASK_INTERRUPTIBLE);	\
+		if (condition)						\
+			break;						\
+		if (!signal_pending(current)) {				\
+			spu_release(ctx);				\
+			schedule();					\
+			spu_acquire(ctx);				\
+			continue;					\
+		}							\
+		__ret = -ERESTARTSYS;					\
+		break;							\
+	}								\
+	finish_wait(&(wq), &__wait);					\
+	__ret;								\
+})
+
+size_t spu_wbox_write(struct spu_context *ctx, u32 data);
+size_t spu_ibox_read(struct spu_context *ctx, u32 *data);
+
+/* irq callback funcs. */
+void spufs_ibox_callback(struct spu *spu);
+void spufs_wbox_callback(struct spu *spu);
+void spufs_stop_callback(struct spu *spu);
+
+#endif
diff --git a/arch/powerpc/platforms/cell/spufs/switch.c b/arch/powerpc/platforms/cell/spufs/switch.c
new file mode 100644
index 000000000000..1061c12b2edb
--- /dev/null
+++ b/arch/powerpc/platforms/cell/spufs/switch.c
@@ -0,0 +1,2180 @@
+/*
+ * spu_switch.c
+ *
+ * (C) Copyright IBM Corp. 2005
+ *
+ * Author: Mark Nutter <mnutter@us.ibm.com>
+ *
+ * Host-side part of SPU context switch sequence outlined in
+ * Synergistic Processor Element, Book IV.
+ *
+ * A fully premptive switch of an SPE is very expensive in terms
+ * of time and system resources.  SPE Book IV indicates that SPE
+ * allocation should follow a "serially reusable device" model,
+ * in which the SPE is assigned a task until it completes.  When
+ * this is not possible, this sequence may be used to premptively
+ * save, and then later (optionally) restore the context of a
+ * program executing on an SPE.
+ *
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; either version 2, or (at your option)
+ * any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software
+ * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
+ */
+
+#include <linux/config.h>
+#include <linux/module.h>
+#include <linux/errno.h>
+#include <linux/sched.h>
+#include <linux/kernel.h>
+#include <linux/mm.h>
+#include <linux/vmalloc.h>
+#include <linux/smp.h>
+#include <linux/smp_lock.h>
+#include <linux/stddef.h>
+#include <linux/unistd.h>
+
+#include <asm/io.h>
+#include <asm/spu.h>
+#include <asm/spu_csa.h>
+#include <asm/mmu_context.h>
+
+#include "spu_save_dump.h"
+#include "spu_restore_dump.h"
+
+#if 0
+#define POLL_WHILE_TRUE(_c) {				\
+    do {						\
+    } while (_c);					\
+  }
+#else
+#define RELAX_SPIN_COUNT				1000
+#define POLL_WHILE_TRUE(_c) {				\
+    do {						\
+	int _i;						\
+	for (_i=0; _i<RELAX_SPIN_COUNT && (_c); _i++) { \
+	    cpu_relax();				\
+	}						\
+	if (unlikely(_c)) yield();			\
+	else break;					\
+    } while (_c);					\
+  }
+#endif				/* debug */
+
+#define POLL_WHILE_FALSE(_c) 	POLL_WHILE_TRUE(!(_c))
+
+static inline void acquire_spu_lock(struct spu *spu)
+{
+	/* Save, Step 1:
+	 * Restore, Step 1:
+	 *    Acquire SPU-specific mutual exclusion lock.
+	 *    TBD.
+	 */
+}
+
+static inline void release_spu_lock(struct spu *spu)
+{
+	/* Restore, Step 76:
+	 *    Release SPU-specific mutual exclusion lock.
+	 *    TBD.
+	 */
+}
+
+static inline int check_spu_isolate(struct spu_state *csa, struct spu *spu)
+{
+	struct spu_problem __iomem *prob = spu->problem;
+	u32 isolate_state;
+
+	/* Save, Step 2:
+	 * Save, Step 6:
+	 *     If SPU_Status[E,L,IS] any field is '1', this
+	 *     SPU is in isolate state and cannot be context
+	 *     saved at this time.
+	 */
+	isolate_state = SPU_STATUS_ISOLATED_STATE |
+	    SPU_STATUS_ISOLATED_LOAD_STAUTUS | SPU_STATUS_ISOLATED_EXIT_STAUTUS;
+	return (in_be32(&prob->spu_status_R) & isolate_state) ? 1 : 0;
+}
+
+static inline void disable_interrupts(struct spu_state *csa, struct spu *spu)
+{
+	/* Save, Step 3:
+	 * Restore, Step 2:
+	 *     Save INT_Mask_class0 in CSA.
+	 *     Write INT_MASK_class0 with value of 0.
+	 *     Save INT_Mask_class1 in CSA.
+	 *     Write INT_MASK_class1 with value of 0.
+	 *     Save INT_Mask_class2 in CSA.
+	 *     Write INT_MASK_class2 with value of 0.
+	 */
+	spin_lock_irq(&spu->register_lock);
+	if (csa) {
+		csa->priv1.int_mask_class0_RW = spu_int_mask_get(spu, 0);
+		csa->priv1.int_mask_class1_RW = spu_int_mask_get(spu, 1);
+		csa->priv1.int_mask_class2_RW = spu_int_mask_get(spu, 2);
+	}
+	spu_int_mask_set(spu, 0, 0ul);
+	spu_int_mask_set(spu, 1, 0ul);
+	spu_int_mask_set(spu, 2, 0ul);
+	eieio();
+	spin_unlock_irq(&spu->register_lock);
+}
+
+static inline void set_watchdog_timer(struct spu_state *csa, struct spu *spu)
+{
+	/* Save, Step 4:
+	 * Restore, Step 25.
+	 *    Set a software watchdog timer, which specifies the
+	 *    maximum allowable time for a context save sequence.
+	 *
+	 *    For present, this implementation will not set a global
+	 *    watchdog timer, as virtualization & variable system load
+	 *    may cause unpredictable execution times.
+	 */
+}
+
+static inline void inhibit_user_access(struct spu_state *csa, struct spu *spu)
+{
+	/* Save, Step 5:
+	 * Restore, Step 3:
+	 *     Inhibit user-space access (if provided) to this
+	 *     SPU by unmapping the virtual pages assigned to
+	 *     the SPU memory-mapped I/O (MMIO) for problem
+	 *     state. TBD.
+	 */
+}
+
+static inline void set_switch_pending(struct spu_state *csa, struct spu *spu)
+{
+	/* Save, Step 7:
+	 * Restore, Step 5:
+	 *     Set a software context switch pending flag.
+	 */
+	set_bit(SPU_CONTEXT_SWITCH_PENDING, &spu->flags);
+	mb();
+}
+
+static inline void save_mfc_cntl(struct spu_state *csa, struct spu *spu)
+{
+	struct spu_priv2 __iomem *priv2 = spu->priv2;
+
+	/* Save, Step 8:
+	 *     Read and save MFC_CNTL[Ss].
+	 */
+	if (csa) {
+		csa->priv2.mfc_control_RW = in_be64(&priv2->mfc_control_RW) &
+		    MFC_CNTL_SUSPEND_DMA_STATUS_MASK;
+	}
+}
+
+static inline void save_spu_runcntl(struct spu_state *csa, struct spu *spu)
+{
+	struct spu_problem __iomem *prob = spu->problem;
+
+	/* Save, Step 9:
+	 *     Save SPU_Runcntl in the CSA.  This value contains
+	 *     the "Application Desired State".
+	 */
+	csa->prob.spu_runcntl_RW = in_be32(&prob->spu_runcntl_RW);
+}
+
+static inline void save_mfc_sr1(struct spu_state *csa, struct spu *spu)
+{
+	/* Save, Step 10:
+	 *     Save MFC_SR1 in the CSA.
+	 */
+	csa->priv1.mfc_sr1_RW = spu_mfc_sr1_get(spu);
+}
+
+static inline void save_spu_status(struct spu_state *csa, struct spu *spu)
+{
+	struct spu_problem __iomem *prob = spu->problem;
+
+	/* Save, Step 11:
+	 *     Read SPU_Status[R], and save to CSA.
+	 */
+	if ((in_be32(&prob->spu_status_R) & SPU_STATUS_RUNNING) == 0) {
+		csa->prob.spu_status_R = in_be32(&prob->spu_status_R);
+	} else {
+		u32 stopped;
+
+		out_be32(&prob->spu_runcntl_RW, SPU_RUNCNTL_STOP);
+		eieio();
+		POLL_WHILE_TRUE(in_be32(&prob->spu_status_R) &
+				SPU_STATUS_RUNNING);
+		stopped =
+		    SPU_STATUS_INVALID_INSTR | SPU_STATUS_SINGLE_STEP |
+		    SPU_STATUS_STOPPED_BY_HALT | SPU_STATUS_STOPPED_BY_STOP;
+		if ((in_be32(&prob->spu_status_R) & stopped) == 0)
+			csa->prob.spu_status_R = SPU_STATUS_RUNNING;
+		else
+			csa->prob.spu_status_R = in_be32(&prob->spu_status_R);
+	}
+}
+
+static inline void save_mfc_decr(struct spu_state *csa, struct spu *spu)
+{
+	struct spu_priv2 __iomem *priv2 = spu->priv2;
+
+	/* Save, Step 12:
+	 *     Read MFC_CNTL[Ds].  Update saved copy of
+	 *     CSA.MFC_CNTL[Ds].
+	 */
+	if (in_be64(&priv2->mfc_control_RW) & MFC_CNTL_DECREMENTER_RUNNING) {
+		csa->priv2.mfc_control_RW |= MFC_CNTL_DECREMENTER_RUNNING;
+		csa->suspend_time = get_cycles();
+		out_be64(&priv2->spu_chnlcntptr_RW, 7ULL);
+		eieio();
+		csa->spu_chnldata_RW[7] = in_be64(&priv2->spu_chnldata_RW);
+		eieio();
+	}
+}
+
+static inline void halt_mfc_decr(struct spu_state *csa, struct spu *spu)
+{
+	struct spu_priv2 __iomem *priv2 = spu->priv2;
+
+	/* Save, Step 13:
+	 *     Write MFC_CNTL[Dh] set to a '1' to halt
+	 *     the decrementer.
+	 */
+	out_be64(&priv2->mfc_control_RW, MFC_CNTL_DECREMENTER_HALTED);
+	eieio();
+}
+
+static inline void save_timebase(struct spu_state *csa, struct spu *spu)
+{
+	/* Save, Step 14:
+	 *    Read PPE Timebase High and Timebase low registers
+	 *    and save in CSA.  TBD.
+	 */
+	csa->suspend_time = get_cycles();
+}
+
+static inline void remove_other_spu_access(struct spu_state *csa,
+					   struct spu *spu)
+{
+	/* Save, Step 15:
+	 *     Remove other SPU access to this SPU by unmapping
+	 *     this SPU's pages from their address space.  TBD.
+	 */
+}
+
+static inline void do_mfc_mssync(struct spu_state *csa, struct spu *spu)
+{
+	struct spu_problem __iomem *prob = spu->problem;
+
+	/* Save, Step 16:
+	 * Restore, Step 11.
+	 *     Write SPU_MSSync register. Poll SPU_MSSync[P]
+	 *     for a value of 0.
+	 */
+	out_be64(&prob->spc_mssync_RW, 1UL);
+	POLL_WHILE_TRUE(in_be64(&prob->spc_mssync_RW) & MS_SYNC_PENDING);
+}
+
+static inline void issue_mfc_tlbie(struct spu_state *csa, struct spu *spu)
+{
+	/* Save, Step 17:
+	 * Restore, Step 12.
+	 * Restore, Step 48.
+	 *     Write TLB_Invalidate_Entry[IS,VPN,L,Lp]=0 register.
+	 *     Then issue a PPE sync instruction.
+	 */
+	spu_tlb_invalidate(spu);
+	mb();
+}
+
+static inline void handle_pending_interrupts(struct spu_state *csa,
+					     struct spu *spu)
+{
+	/* Save, Step 18:
+	 *     Handle any pending interrupts from this SPU
+	 *     here.  This is OS or hypervisor specific.  One
+	 *     option is to re-enable interrupts to handle any
+	 *     pending interrupts, with the interrupt handlers
+	 *     recognizing the software Context Switch Pending
+	 *     flag, to ensure the SPU execution or MFC command
+	 *     queue is not restarted.  TBD.
+	 */
+}
+
+static inline void save_mfc_queues(struct spu_state *csa, struct spu *spu)
+{
+	struct spu_priv2 __iomem *priv2 = spu->priv2;
+	int i;
+
+	/* Save, Step 19:
+	 *     If MFC_Cntl[Se]=0 then save
+	 *     MFC command queues.
+	 */
+	if ((in_be64(&priv2->mfc_control_RW) & MFC_CNTL_DMA_QUEUES_EMPTY) == 0) {
+		for (i = 0; i < 8; i++) {
+			csa->priv2.puq[i].mfc_cq_data0_RW =
+			    in_be64(&priv2->puq[i].mfc_cq_data0_RW);
+			csa->priv2.puq[i].mfc_cq_data1_RW =
+			    in_be64(&priv2->puq[i].mfc_cq_data1_RW);
+			csa->priv2.puq[i].mfc_cq_data2_RW =
+			    in_be64(&priv2->puq[i].mfc_cq_data2_RW);
+			csa->priv2.puq[i].mfc_cq_data3_RW =
+			    in_be64(&priv2->puq[i].mfc_cq_data3_RW);
+		}
+		for (i = 0; i < 16; i++) {
+			csa->priv2.spuq[i].mfc_cq_data0_RW =
+			    in_be64(&priv2->spuq[i].mfc_cq_data0_RW);
+			csa->priv2.spuq[i].mfc_cq_data1_RW =
+			    in_be64(&priv2->spuq[i].mfc_cq_data1_RW);
+			csa->priv2.spuq[i].mfc_cq_data2_RW =
+			    in_be64(&priv2->spuq[i].mfc_cq_data2_RW);
+			csa->priv2.spuq[i].mfc_cq_data3_RW =
+			    in_be64(&priv2->spuq[i].mfc_cq_data3_RW);
+		}
+	}
+}
+
+static inline void save_ppu_querymask(struct spu_state *csa, struct spu *spu)
+{
+	struct spu_problem __iomem *prob = spu->problem;
+
+	/* Save, Step 20:
+	 *     Save the PPU_QueryMask register
+	 *     in the CSA.
+	 */
+	csa->prob.dma_querymask_RW = in_be32(&prob->dma_querymask_RW);
+}
+
+static inline void save_ppu_querytype(struct spu_state *csa, struct spu *spu)
+{
+	struct spu_problem __iomem *prob = spu->problem;
+
+	/* Save, Step 21:
+	 *     Save the PPU_QueryType register
+	 *     in the CSA.
+	 */
+	csa->prob.dma_querytype_RW = in_be32(&prob->dma_querytype_RW);
+}
+
+static inline void save_mfc_csr_tsq(struct spu_state *csa, struct spu *spu)
+{
+	struct spu_priv2 __iomem *priv2 = spu->priv2;
+
+	/* Save, Step 22:
+	 *     Save the MFC_CSR_TSQ register
+	 *     in the LSCSA.
+	 */
+	csa->priv2.spu_tag_status_query_RW =
+	    in_be64(&priv2->spu_tag_status_query_RW);
+}
+
+static inline void save_mfc_csr_cmd(struct spu_state *csa, struct spu *spu)
+{
+	struct spu_priv2 __iomem *priv2 = spu->priv2;
+
+	/* Save, Step 23:
+	 *     Save the MFC_CSR_CMD1 and MFC_CSR_CMD2
+	 *     registers in the CSA.
+	 */
+	csa->priv2.spu_cmd_buf1_RW = in_be64(&priv2->spu_cmd_buf1_RW);
+	csa->priv2.spu_cmd_buf2_RW = in_be64(&priv2->spu_cmd_buf2_RW);
+}
+
+static inline void save_mfc_csr_ato(struct spu_state *csa, struct spu *spu)
+{
+	struct spu_priv2 __iomem *priv2 = spu->priv2;
+
+	/* Save, Step 24:
+	 *     Save the MFC_CSR_ATO register in
+	 *     the CSA.
+	 */
+	csa->priv2.spu_atomic_status_RW = in_be64(&priv2->spu_atomic_status_RW);
+}
+
+static inline void save_mfc_tclass_id(struct spu_state *csa, struct spu *spu)
+{
+	/* Save, Step 25:
+	 *     Save the MFC_TCLASS_ID register in
+	 *     the CSA.
+	 */
+	csa->priv1.mfc_tclass_id_RW = spu_mfc_tclass_id_get(spu);
+}
+
+static inline void set_mfc_tclass_id(struct spu_state *csa, struct spu *spu)
+{
+	/* Save, Step 26:
+	 * Restore, Step 23.
+	 *     Write the MFC_TCLASS_ID register with
+	 *     the value 0x10000000.
+	 */
+	spu_mfc_tclass_id_set(spu, 0x10000000);
+	eieio();
+}
+
+static inline void purge_mfc_queue(struct spu_state *csa, struct spu *spu)
+{
+	struct spu_priv2 __iomem *priv2 = spu->priv2;
+
+	/* Save, Step 27:
+	 * Restore, Step 14.
+	 *     Write MFC_CNTL[Pc]=1 (purge queue).
+	 */
+	out_be64(&priv2->mfc_control_RW, MFC_CNTL_PURGE_DMA_REQUEST);
+	eieio();
+}
+
+static inline void wait_purge_complete(struct spu_state *csa, struct spu *spu)
+{
+	struct spu_priv2 __iomem *priv2 = spu->priv2;
+
+	/* Save, Step 28:
+	 *     Poll MFC_CNTL[Ps] until value '11' is read
+	 *     (purge complete).
+	 */
+	POLL_WHILE_FALSE(in_be64(&priv2->mfc_control_RW) &
+			 MFC_CNTL_PURGE_DMA_COMPLETE);
+}
+
+static inline void save_mfc_slbs(struct spu_state *csa, struct spu *spu)
+{
+	struct spu_priv2 __iomem *priv2 = spu->priv2;
+	int i;
+
+	/* Save, Step 29:
+	 *     If MFC_SR1[R]='1', save SLBs in CSA.
+	 */
+	if (spu_mfc_sr1_get(spu) & MFC_STATE1_RELOCATE_MASK) {
+		csa->priv2.slb_index_W = in_be64(&priv2->slb_index_W);
+		for (i = 0; i < 8; i++) {
+			out_be64(&priv2->slb_index_W, i);
+			eieio();
+			csa->slb_esid_RW[i] = in_be64(&priv2->slb_esid_RW);
+			csa->slb_vsid_RW[i] = in_be64(&priv2->slb_vsid_RW);
+			eieio();
+		}
+	}
+}
+
+static inline void setup_mfc_sr1(struct spu_state *csa, struct spu *spu)
+{
+	/* Save, Step 30:
+	 * Restore, Step 18:
+	 *     Write MFC_SR1 with MFC_SR1[D=0,S=1] and
+	 *     MFC_SR1[TL,R,Pr,T] set correctly for the
+	 *     OS specific environment.
+	 *
+	 *     Implementation note: The SPU-side code
+	 *     for save/restore is privileged, so the
+	 *     MFC_SR1[Pr] bit is not set.
+	 *
+	 */
+	spu_mfc_sr1_set(spu, (MFC_STATE1_MASTER_RUN_CONTROL_MASK |
+			      MFC_STATE1_RELOCATE_MASK |
+			      MFC_STATE1_BUS_TLBIE_MASK));
+}
+
+static inline void save_spu_npc(struct spu_state *csa, struct spu *spu)
+{
+	struct spu_problem __iomem *prob = spu->problem;
+
+	/* Save, Step 31:
+	 *     Save SPU_NPC in the CSA.
+	 */
+	csa->prob.spu_npc_RW = in_be32(&prob->spu_npc_RW);
+}
+
+static inline void save_spu_privcntl(struct spu_state *csa, struct spu *spu)
+{
+	struct spu_priv2 __iomem *priv2 = spu->priv2;
+
+	/* Save, Step 32:
+	 *     Save SPU_PrivCntl in the CSA.
+	 */
+	csa->priv2.spu_privcntl_RW = in_be64(&priv2->spu_privcntl_RW);
+}
+
+static inline void reset_spu_privcntl(struct spu_state *csa, struct spu *spu)
+{
+	struct spu_priv2 __iomem *priv2 = spu->priv2;
+
+	/* Save, Step 33:
+	 * Restore, Step 16:
+	 *     Write SPU_PrivCntl[S,Le,A] fields reset to 0.
+	 */
+	out_be64(&priv2->spu_privcntl_RW, 0UL);
+	eieio();
+}
+
+static inline void save_spu_lslr(struct spu_state *csa, struct spu *spu)
+{
+	struct spu_priv2 __iomem *priv2 = spu->priv2;
+
+	/* Save, Step 34:
+	 *     Save SPU_LSLR in the CSA.
+	 */
+	csa->priv2.spu_lslr_RW = in_be64(&priv2->spu_lslr_RW);
+}
+
+static inline void reset_spu_lslr(struct spu_state *csa, struct spu *spu)
+{
+	struct spu_priv2 __iomem *priv2 = spu->priv2;
+
+	/* Save, Step 35:
+	 * Restore, Step 17.
+	 *     Reset SPU_LSLR.
+	 */
+	out_be64(&priv2->spu_lslr_RW, LS_ADDR_MASK);
+	eieio();
+}
+
+static inline void save_spu_cfg(struct spu_state *csa, struct spu *spu)
+{
+	struct spu_priv2 __iomem *priv2 = spu->priv2;
+
+	/* Save, Step 36:
+	 *     Save SPU_Cfg in the CSA.
+	 */
+	csa->priv2.spu_cfg_RW = in_be64(&priv2->spu_cfg_RW);
+}
+
+static inline void save_pm_trace(struct spu_state *csa, struct spu *spu)
+{
+	/* Save, Step 37:
+	 *     Save PM_Trace_Tag_Wait_Mask in the CSA.
+	 *     Not performed by this implementation.
+	 */
+}
+
+static inline void save_mfc_rag(struct spu_state *csa, struct spu *spu)
+{
+	/* Save, Step 38:
+	 *     Save RA_GROUP_ID register and the
+	 *     RA_ENABLE reigster in the CSA.
+	 */
+	csa->priv1.resource_allocation_groupID_RW =
+		spu_resource_allocation_groupID_get(spu);
+	csa->priv1.resource_allocation_enable_RW =
+		spu_resource_allocation_enable_get(spu);
+}
+
+static inline void save_ppu_mb_stat(struct spu_state *csa, struct spu *spu)
+{
+	struct spu_problem __iomem *prob = spu->problem;
+
+	/* Save, Step 39:
+	 *     Save MB_Stat register in the CSA.
+	 */
+	csa->prob.mb_stat_R = in_be32(&prob->mb_stat_R);
+}
+
+static inline void save_ppu_mb(struct spu_state *csa, struct spu *spu)
+{
+	struct spu_problem __iomem *prob = spu->problem;
+
+	/* Save, Step 40:
+	 *     Save the PPU_MB register in the CSA.
+	 */
+	csa->prob.pu_mb_R = in_be32(&prob->pu_mb_R);
+}
+
+static inline void save_ppuint_mb(struct spu_state *csa, struct spu *spu)
+{
+	struct spu_priv2 __iomem *priv2 = spu->priv2;
+
+	/* Save, Step 41:
+	 *     Save the PPUINT_MB register in the CSA.
+	 */
+	csa->priv2.puint_mb_R = in_be64(&priv2->puint_mb_R);
+}
+
+static inline void save_ch_part1(struct spu_state *csa, struct spu *spu)
+{
+	struct spu_priv2 __iomem *priv2 = spu->priv2;
+	u64 idx, ch_indices[7] = { 0UL, 1UL, 3UL, 4UL, 24UL, 25UL, 27UL };
+	int i;
+
+	/* Save, Step 42:
+	 *     Save the following CH: [0,1,3,4,24,25,27]
+	 */
+	for (i = 0; i < 7; i++) {
+		idx = ch_indices[i];
+		out_be64(&priv2->spu_chnlcntptr_RW, idx);
+		eieio();
+		csa->spu_chnldata_RW[idx] = in_be64(&priv2->spu_chnldata_RW);
+		csa->spu_chnlcnt_RW[idx] = in_be64(&priv2->spu_chnlcnt_RW);
+		out_be64(&priv2->spu_chnldata_RW, 0UL);
+		out_be64(&priv2->spu_chnlcnt_RW, 0UL);
+		eieio();
+	}
+}
+
+static inline void save_spu_mb(struct spu_state *csa, struct spu *spu)
+{
+	struct spu_priv2 __iomem *priv2 = spu->priv2;
+	int i;
+
+	/* Save, Step 43:
+	 *     Save SPU Read Mailbox Channel.
+	 */
+	out_be64(&priv2->spu_chnlcntptr_RW, 29UL);
+	eieio();
+	csa->spu_chnlcnt_RW[29] = in_be64(&priv2->spu_chnlcnt_RW);
+	for (i = 0; i < 4; i++) {
+		csa->spu_mailbox_data[i] = in_be64(&priv2->spu_chnldata_RW);
+	}
+	out_be64(&priv2->spu_chnlcnt_RW, 0UL);
+	eieio();
+}
+
+static inline void save_mfc_cmd(struct spu_state *csa, struct spu *spu)
+{
+	struct spu_priv2 __iomem *priv2 = spu->priv2;
+
+	/* Save, Step 44:
+	 *     Save MFC_CMD Channel.
+	 */
+	out_be64(&priv2->spu_chnlcntptr_RW, 21UL);
+	eieio();
+	csa->spu_chnlcnt_RW[21] = in_be64(&priv2->spu_chnlcnt_RW);
+	eieio();
+}
+
+static inline void reset_ch(struct spu_state *csa, struct spu *spu)
+{
+	struct spu_priv2 __iomem *priv2 = spu->priv2;
+	u64 ch_indices[4] = { 21UL, 23UL, 28UL, 30UL };
+	u64 ch_counts[4] = { 16UL, 1UL, 1UL, 1UL };
+	u64 idx;
+	int i;
+
+	/* Save, Step 45:
+	 *     Reset the following CH: [21, 23, 28, 30]
+	 */
+	for (i = 0; i < 4; i++) {
+		idx = ch_indices[i];
+		out_be64(&priv2->spu_chnlcntptr_RW, idx);
+		eieio();
+		out_be64(&priv2->spu_chnlcnt_RW, ch_counts[i]);
+		eieio();
+	}
+}
+
+static inline void resume_mfc_queue(struct spu_state *csa, struct spu *spu)
+{
+	struct spu_priv2 __iomem *priv2 = spu->priv2;
+
+	/* Save, Step 46:
+	 * Restore, Step 25.
+	 *     Write MFC_CNTL[Sc]=0 (resume queue processing).
+	 */
+	out_be64(&priv2->mfc_control_RW, MFC_CNTL_RESUME_DMA_QUEUE);
+}
+
+static inline void invalidate_slbs(struct spu_state *csa, struct spu *spu)
+{
+	struct spu_priv2 __iomem *priv2 = spu->priv2;
+
+	/* Save, Step 45:
+	 * Restore, Step 19:
+	 *     If MFC_SR1[R]=1, write 0 to SLB_Invalidate_All.
+	 */
+	if (spu_mfc_sr1_get(spu) & MFC_STATE1_RELOCATE_MASK) {
+		out_be64(&priv2->slb_invalidate_all_W, 0UL);
+		eieio();
+	}
+}
+
+static inline void get_kernel_slb(u64 ea, u64 slb[2])
+{
+	slb[0] = (get_kernel_vsid(ea) << SLB_VSID_SHIFT) | SLB_VSID_KERNEL;
+	slb[1] = (ea & ESID_MASK) | SLB_ESID_V;
+
+	/* Large pages are used for kernel text/data, but not vmalloc.  */
+	if (cpu_has_feature(CPU_FTR_16M_PAGE)
+	    && REGION_ID(ea) == KERNEL_REGION_ID)
+		slb[0] |= SLB_VSID_L;
+}
+
+static inline void load_mfc_slb(struct spu *spu, u64 slb[2], int slbe)
+{
+	struct spu_priv2 __iomem *priv2 = spu->priv2;
+
+	out_be64(&priv2->slb_index_W, slbe);
+	eieio();
+	out_be64(&priv2->slb_vsid_RW, slb[0]);
+	out_be64(&priv2->slb_esid_RW, slb[1]);
+	eieio();
+}
+
+static inline void setup_mfc_slbs(struct spu_state *csa, struct spu *spu)
+{
+	u64 code_slb[2];
+	u64 lscsa_slb[2];
+
+	/* Save, Step 47:
+	 * Restore, Step 30.
+	 *     If MFC_SR1[R]=1, write 0 to SLB_Invalidate_All
+	 *     register, then initialize SLB_VSID and SLB_ESID
+	 *     to provide access to SPU context save code and
+	 *     LSCSA.
+	 *
+	 *     This implementation places both the context
+	 *     switch code and LSCSA in kernel address space.
+	 *
+	 *     Further this implementation assumes that the
+	 *     MFC_SR1[R]=1 (in other words, assume that
+	 *     translation is desired by OS environment).
+	 */
+	invalidate_slbs(csa, spu);
+	get_kernel_slb((unsigned long)&spu_save_code[0], code_slb);
+	get_kernel_slb((unsigned long)csa->lscsa, lscsa_slb);
+	load_mfc_slb(spu, code_slb, 0);
+	if ((lscsa_slb[0] != code_slb[0]) || (lscsa_slb[1] != code_slb[1]))
+		load_mfc_slb(spu, lscsa_slb, 1);
+}
+
+static inline void set_switch_active(struct spu_state *csa, struct spu *spu)
+{
+	/* Save, Step 48:
+	 * Restore, Step 23.
+	 *     Change the software context switch pending flag
+	 *     to context switch active.
+	 */
+	set_bit(SPU_CONTEXT_SWITCH_ACTIVE, &spu->flags);
+	clear_bit(SPU_CONTEXT_SWITCH_PENDING, &spu->flags);
+	mb();
+}
+
+static inline void enable_interrupts(struct spu_state *csa, struct spu *spu)
+{
+	unsigned long class1_mask = CLASS1_ENABLE_SEGMENT_FAULT_INTR |
+	    CLASS1_ENABLE_STORAGE_FAULT_INTR;
+
+	/* Save, Step 49:
+	 * Restore, Step 22:
+	 *     Reset and then enable interrupts, as
+	 *     needed by OS.
+	 *
+	 *     This implementation enables only class1
+	 *     (translation) interrupts.
+	 */
+	spin_lock_irq(&spu->register_lock);
+	spu_int_stat_clear(spu, 0, ~0ul);
+	spu_int_stat_clear(spu, 1, ~0ul);
+	spu_int_stat_clear(spu, 2, ~0ul);
+	spu_int_mask_set(spu, 0, 0ul);
+	spu_int_mask_set(spu, 1, class1_mask);
+	spu_int_mask_set(spu, 2, 0ul);
+	spin_unlock_irq(&spu->register_lock);
+}
+
+static inline int send_mfc_dma(struct spu *spu, unsigned long ea,
+			       unsigned int ls_offset, unsigned int size,
+			       unsigned int tag, unsigned int rclass,
+			       unsigned int cmd)
+{
+	struct spu_problem __iomem *prob = spu->problem;
+	union mfc_tag_size_class_cmd command;
+	unsigned int transfer_size;
+	volatile unsigned int status = 0x0;
+
+	while (size > 0) {
+		transfer_size =
+		    (size > MFC_MAX_DMA_SIZE) ? MFC_MAX_DMA_SIZE : size;
+		command.u.mfc_size = transfer_size;
+		command.u.mfc_tag = tag;
+		command.u.mfc_rclassid = rclass;
+		command.u.mfc_cmd = cmd;
+		do {
+			out_be32(&prob->mfc_lsa_W, ls_offset);
+			out_be64(&prob->mfc_ea_W, ea);
+			out_be64(&prob->mfc_union_W.all64, command.all64);
+			status =
+			    in_be32(&prob->mfc_union_W.by32.mfc_class_cmd32);
+			if (unlikely(status & 0x2)) {
+				cpu_relax();
+			}
+		} while (status & 0x3);
+		size -= transfer_size;
+		ea += transfer_size;
+		ls_offset += transfer_size;
+	}
+	return 0;
+}
+
+static inline void save_ls_16kb(struct spu_state *csa, struct spu *spu)
+{
+	unsigned long addr = (unsigned long)&csa->lscsa->ls[0];
+	unsigned int ls_offset = 0x0;
+	unsigned int size = 16384;
+	unsigned int tag = 0;
+	unsigned int rclass = 0;
+	unsigned int cmd = MFC_PUT_CMD;
+
+	/* Save, Step 50:
+	 *     Issue a DMA command to copy the first 16K bytes
+	 *     of local storage to the CSA.
+	 */
+	send_mfc_dma(spu, addr, ls_offset, size, tag, rclass, cmd);
+}
+
+static inline void set_spu_npc(struct spu_state *csa, struct spu *spu)
+{
+	struct spu_problem __iomem *prob = spu->problem;
+
+	/* Save, Step 51:
+	 * Restore, Step 31.
+	 *     Write SPU_NPC[IE]=0 and SPU_NPC[LSA] to entry
+	 *     point address of context save code in local
+	 *     storage.
+	 *
+	 *     This implementation uses SPU-side save/restore
+	 *     programs with entry points at LSA of 0.
+	 */
+	out_be32(&prob->spu_npc_RW, 0);
+	eieio();
+}
+
+static inline void set_signot1(struct spu_state *csa, struct spu *spu)
+{
+	struct spu_problem __iomem *prob = spu->problem;
+	union {
+		u64 ull;
+		u32 ui[2];
+	} addr64;
+
+	/* Save, Step 52:
+	 * Restore, Step 32:
+	 *    Write SPU_Sig_Notify_1 register with upper 32-bits
+	 *    of the CSA.LSCSA effective address.
+	 */
+	addr64.ull = (u64) csa->lscsa;
+	out_be32(&prob->signal_notify1, addr64.ui[0]);
+	eieio();
+}
+
+static inline void set_signot2(struct spu_state *csa, struct spu *spu)
+{
+	struct spu_problem __iomem *prob = spu->problem;
+	union {
+		u64 ull;
+		u32 ui[2];
+	} addr64;
+
+	/* Save, Step 53:
+	 * Restore, Step 33:
+	 *    Write SPU_Sig_Notify_2 register with lower 32-bits
+	 *    of the CSA.LSCSA effective address.
+	 */
+	addr64.ull = (u64) csa->lscsa;
+	out_be32(&prob->signal_notify2, addr64.ui[1]);
+	eieio();
+}
+
+static inline void send_save_code(struct spu_state *csa, struct spu *spu)
+{
+	unsigned long addr = (unsigned long)&spu_save_code[0];
+	unsigned int ls_offset = 0x0;
+	unsigned int size = sizeof(spu_save_code);
+	unsigned int tag = 0;
+	unsigned int rclass = 0;
+	unsigned int cmd = MFC_GETFS_CMD;
+
+	/* Save, Step 54:
+	 *     Issue a DMA command to copy context save code
+	 *     to local storage and start SPU.
+	 */
+	send_mfc_dma(spu, addr, ls_offset, size, tag, rclass, cmd);
+}
+
+static inline void set_ppu_querymask(struct spu_state *csa, struct spu *spu)
+{
+	struct spu_problem __iomem *prob = spu->problem;
+
+	/* Save, Step 55:
+	 * Restore, Step 38.
+	 *     Write PPU_QueryMask=1 (enable Tag Group 0)
+	 *     and issue eieio instruction.
+	 */
+	out_be32(&prob->dma_querymask_RW, MFC_TAGID_TO_TAGMASK(0));
+	eieio();
+}
+
+static inline void wait_tag_complete(struct spu_state *csa, struct spu *spu)
+{
+	struct spu_problem __iomem *prob = spu->problem;
+	u32 mask = MFC_TAGID_TO_TAGMASK(0);
+	unsigned long flags;
+
+	/* Save, Step 56:
+	 * Restore, Step 39.
+	 * Restore, Step 39.
+	 * Restore, Step 46.
+	 *     Poll PPU_TagStatus[gn] until 01 (Tag group 0 complete)
+	 *     or write PPU_QueryType[TS]=01 and wait for Tag Group
+	 *     Complete Interrupt.  Write INT_Stat_Class0 or
+	 *     INT_Stat_Class2 with value of 'handled'.
+	 */
+	POLL_WHILE_FALSE(in_be32(&prob->dma_tagstatus_R) & mask);
+
+	local_irq_save(flags);
+	spu_int_stat_clear(spu, 0, ~(0ul));
+	spu_int_stat_clear(spu, 2, ~(0ul));
+	local_irq_restore(flags);
+}
+
+static inline void wait_spu_stopped(struct spu_state *csa, struct spu *spu)
+{
+	struct spu_problem __iomem *prob = spu->problem;
+	unsigned long flags;
+
+	/* Save, Step 57:
+	 * Restore, Step 40.
+	 *     Poll until SPU_Status[R]=0 or wait for SPU Class 0
+	 *     or SPU Class 2 interrupt.  Write INT_Stat_class0
+	 *     or INT_Stat_class2 with value of handled.
+	 */
+	POLL_WHILE_TRUE(in_be32(&prob->spu_status_R) & SPU_STATUS_RUNNING);
+
+	local_irq_save(flags);
+	spu_int_stat_clear(spu, 0, ~(0ul));
+	spu_int_stat_clear(spu, 2, ~(0ul));
+	local_irq_restore(flags);
+}
+
+static inline int check_save_status(struct spu_state *csa, struct spu *spu)
+{
+	struct spu_problem __iomem *prob = spu->problem;
+	u32 complete;
+
+	/* Save, Step 54:
+	 *     If SPU_Status[P]=1 and SPU_Status[SC] = "success",
+	 *     context save succeeded, otherwise context save
+	 *     failed.
+	 */
+	complete = ((SPU_SAVE_COMPLETE << SPU_STOP_STATUS_SHIFT) |
+		    SPU_STATUS_STOPPED_BY_STOP);
+	return (in_be32(&prob->spu_status_R) != complete) ? 1 : 0;
+}
+
+static inline void terminate_spu_app(struct spu_state *csa, struct spu *spu)
+{
+	/* Restore, Step 4:
+	 *    If required, notify the "using application" that
+	 *    the SPU task has been terminated.  TBD.
+	 */
+}
+
+static inline void suspend_mfc(struct spu_state *csa, struct spu *spu)
+{
+	struct spu_priv2 __iomem *priv2 = spu->priv2;
+
+	/* Restore, Step 7:
+	 * Restore, Step 47.
+	 *     Write MFC_Cntl[Dh,Sc]='1','1' to suspend
+	 *     the queue and halt the decrementer.
+	 */
+	out_be64(&priv2->mfc_control_RW, MFC_CNTL_SUSPEND_DMA_QUEUE |
+		 MFC_CNTL_DECREMENTER_HALTED);
+	eieio();
+}
+
+static inline void wait_suspend_mfc_complete(struct spu_state *csa,
+					     struct spu *spu)
+{
+	struct spu_priv2 __iomem *priv2 = spu->priv2;
+
+	/* Restore, Step 8:
+	 * Restore, Step 47.
+	 *     Poll MFC_CNTL[Ss] until 11 is returned.
+	 */
+	POLL_WHILE_FALSE(in_be64(&priv2->mfc_control_RW) &
+			 MFC_CNTL_SUSPEND_COMPLETE);
+}
+
+static inline int suspend_spe(struct spu_state *csa, struct spu *spu)
+{
+	struct spu_problem __iomem *prob = spu->problem;
+
+	/* Restore, Step 9:
+	 *    If SPU_Status[R]=1, stop SPU execution
+	 *    and wait for stop to complete.
+	 *
+	 *    Returns       1 if SPU_Status[R]=1 on entry.
+	 *                  0 otherwise
+	 */
+	if (in_be32(&prob->spu_status_R) & SPU_STATUS_RUNNING) {
+		if (in_be32(&prob->spu_status_R) &
+		    SPU_STATUS_ISOLATED_EXIT_STAUTUS) {
+			POLL_WHILE_TRUE(in_be32(&prob->spu_status_R) &
+					SPU_STATUS_RUNNING);
+		}
+		if ((in_be32(&prob->spu_status_R) &
+		     SPU_STATUS_ISOLATED_LOAD_STAUTUS)
+		    || (in_be32(&prob->spu_status_R) &
+			SPU_STATUS_ISOLATED_STATE)) {
+			out_be32(&prob->spu_runcntl_RW, SPU_RUNCNTL_STOP);
+			eieio();
+			POLL_WHILE_TRUE(in_be32(&prob->spu_status_R) &
+					SPU_STATUS_RUNNING);
+			out_be32(&prob->spu_runcntl_RW, 0x2);
+			eieio();
+			POLL_WHILE_TRUE(in_be32(&prob->spu_status_R) &
+					SPU_STATUS_RUNNING);
+		}
+		if (in_be32(&prob->spu_status_R) &
+		    SPU_STATUS_WAITING_FOR_CHANNEL) {
+			out_be32(&prob->spu_runcntl_RW, SPU_RUNCNTL_STOP);
+			eieio();
+			POLL_WHILE_TRUE(in_be32(&prob->spu_status_R) &
+					SPU_STATUS_RUNNING);
+		}
+		return 1;
+	}
+	return 0;
+}
+
+static inline void clear_spu_status(struct spu_state *csa, struct spu *spu)
+{
+	struct spu_problem __iomem *prob = spu->problem;
+
+	/* Restore, Step 10:
+	 *    If SPU_Status[R]=0 and SPU_Status[E,L,IS]=1,
+	 *    release SPU from isolate state.
+	 */
+	if (!(in_be32(&prob->spu_status_R) & SPU_STATUS_RUNNING)) {
+		if (in_be32(&prob->spu_status_R) &
+		    SPU_STATUS_ISOLATED_EXIT_STAUTUS) {
+			spu_mfc_sr1_set(spu,
+					MFC_STATE1_MASTER_RUN_CONTROL_MASK);
+			eieio();
+			out_be32(&prob->spu_runcntl_RW, SPU_RUNCNTL_RUNNABLE);
+			eieio();
+			POLL_WHILE_TRUE(in_be32(&prob->spu_status_R) &
+					SPU_STATUS_RUNNING);
+		}
+		if ((in_be32(&prob->spu_status_R) &
+		     SPU_STATUS_ISOLATED_LOAD_STAUTUS)
+		    || (in_be32(&prob->spu_status_R) &
+			SPU_STATUS_ISOLATED_STATE)) {
+			spu_mfc_sr1_set(spu,
+					MFC_STATE1_MASTER_RUN_CONTROL_MASK);
+			eieio();
+			out_be32(&prob->spu_runcntl_RW, 0x2);
+			eieio();
+			POLL_WHILE_TRUE(in_be32(&prob->spu_status_R) &
+					SPU_STATUS_RUNNING);
+		}
+	}
+}
+
+static inline void reset_ch_part1(struct spu_state *csa, struct spu *spu)
+{
+	struct spu_priv2 __iomem *priv2 = spu->priv2;
+	u64 ch_indices[7] = { 0UL, 1UL, 3UL, 4UL, 24UL, 25UL, 27UL };
+	u64 idx;
+	int i;
+
+	/* Restore, Step 20:
+	 *     Reset the following CH: [0,1,3,4,24,25,27]
+	 */
+	for (i = 0; i < 7; i++) {
+		idx = ch_indices[i];
+		out_be64(&priv2->spu_chnlcntptr_RW, idx);
+		eieio();
+		out_be64(&priv2->spu_chnldata_RW, 0UL);
+		out_be64(&priv2->spu_chnlcnt_RW, 0UL);
+		eieio();
+	}
+}
+
+static inline void reset_ch_part2(struct spu_state *csa, struct spu *spu)
+{
+	struct spu_priv2 __iomem *priv2 = spu->priv2;
+	u64 ch_indices[5] = { 21UL, 23UL, 28UL, 29UL, 30UL };
+	u64 ch_counts[5] = { 16UL, 1UL, 1UL, 0UL, 1UL };
+	u64 idx;
+	int i;
+
+	/* Restore, Step 21:
+	 *     Reset the following CH: [21, 23, 28, 29, 30]
+	 */
+	for (i = 0; i < 5; i++) {
+		idx = ch_indices[i];
+		out_be64(&priv2->spu_chnlcntptr_RW, idx);
+		eieio();
+		out_be64(&priv2->spu_chnlcnt_RW, ch_counts[i]);
+		eieio();
+	}
+}
+
+static inline void setup_spu_status_part1(struct spu_state *csa,
+					  struct spu *spu)
+{
+	u32 status_P = SPU_STATUS_STOPPED_BY_STOP;
+	u32 status_I = SPU_STATUS_INVALID_INSTR;
+	u32 status_H = SPU_STATUS_STOPPED_BY_HALT;
+	u32 status_S = SPU_STATUS_SINGLE_STEP;
+	u32 status_S_I = SPU_STATUS_SINGLE_STEP | SPU_STATUS_INVALID_INSTR;
+	u32 status_S_P = SPU_STATUS_SINGLE_STEP | SPU_STATUS_STOPPED_BY_STOP;
+	u32 status_P_H = SPU_STATUS_STOPPED_BY_HALT |SPU_STATUS_STOPPED_BY_STOP;
+	u32 status_P_I = SPU_STATUS_STOPPED_BY_STOP |SPU_STATUS_INVALID_INSTR;
+	u32 status_code;
+
+	/* Restore, Step 27:
+	 *     If the CSA.SPU_Status[I,S,H,P]=1 then add the correct
+	 *     instruction sequence to the end of the SPU based restore
+	 *     code (after the "context restored" stop and signal) to
+	 *     restore the correct SPU status.
+	 *
+	 *     NOTE: Rather than modifying the SPU executable, we
+	 *     instead add a new 'stopped_status' field to the
+	 *     LSCSA.  The SPU-side restore reads this field and
+	 *     takes the appropriate action when exiting.
+	 */
+
+	status_code =
+	    (csa->prob.spu_status_R >> SPU_STOP_STATUS_SHIFT) & 0xFFFF;
+	if ((csa->prob.spu_status_R & status_P_I) == status_P_I) {
+
+		/* SPU_Status[P,I]=1 - Illegal Instruction followed
+		 * by Stop and Signal instruction, followed by 'br -4'.
+		 *
+		 */
+		csa->lscsa->stopped_status.slot[0] = SPU_STOPPED_STATUS_P_I;
+		csa->lscsa->stopped_status.slot[1] = status_code;
+
+	} else if ((csa->prob.spu_status_R & status_P_H) == status_P_H) {
+
+		/* SPU_Status[P,H]=1 - Halt Conditional, followed
+		 * by Stop and Signal instruction, followed by
+		 * 'br -4'.
+		 */
+		csa->lscsa->stopped_status.slot[0] = SPU_STOPPED_STATUS_P_H;
+		csa->lscsa->stopped_status.slot[1] = status_code;
+
+	} else if ((csa->prob.spu_status_R & status_S_P) == status_S_P) {
+
+		/* SPU_Status[S,P]=1 - Stop and Signal instruction
+		 * followed by 'br -4'.
+		 */
+		csa->lscsa->stopped_status.slot[0] = SPU_STOPPED_STATUS_S_P;
+		csa->lscsa->stopped_status.slot[1] = status_code;
+
+	} else if ((csa->prob.spu_status_R & status_S_I) == status_S_I) {
+
+		/* SPU_Status[S,I]=1 - Illegal instruction followed
+		 * by 'br -4'.
+		 */
+		csa->lscsa->stopped_status.slot[0] = SPU_STOPPED_STATUS_S_I;
+		csa->lscsa->stopped_status.slot[1] = status_code;
+
+	} else if ((csa->prob.spu_status_R & status_P) == status_P) {
+
+		/* SPU_Status[P]=1 - Stop and Signal instruction
+		 * followed by 'br -4'.
+		 */
+		csa->lscsa->stopped_status.slot[0] = SPU_STOPPED_STATUS_P;
+		csa->lscsa->stopped_status.slot[1] = status_code;
+
+	} else if ((csa->prob.spu_status_R & status_H) == status_H) {
+
+		/* SPU_Status[H]=1 - Halt Conditional, followed
+		 * by 'br -4'.
+		 */
+		csa->lscsa->stopped_status.slot[0] = SPU_STOPPED_STATUS_H;
+
+	} else if ((csa->prob.spu_status_R & status_S) == status_S) {
+
+		/* SPU_Status[S]=1 - Two nop instructions.
+		 */
+		csa->lscsa->stopped_status.slot[0] = SPU_STOPPED_STATUS_S;
+
+	} else if ((csa->prob.spu_status_R & status_I) == status_I) {
+
+		/* SPU_Status[I]=1 - Illegal instruction followed
+		 * by 'br -4'.
+		 */
+		csa->lscsa->stopped_status.slot[0] = SPU_STOPPED_STATUS_I;
+
+	}
+}
+
+static inline void setup_spu_status_part2(struct spu_state *csa,
+					  struct spu *spu)
+{
+	u32 mask;
+
+	/* Restore, Step 28:
+	 *     If the CSA.SPU_Status[I,S,H,P,R]=0 then
+	 *     add a 'br *' instruction to the end of
+	 *     the SPU based restore code.
+	 *
+	 *     NOTE: Rather than modifying the SPU executable, we
+	 *     instead add a new 'stopped_status' field to the
+	 *     LSCSA.  The SPU-side restore reads this field and
+	 *     takes the appropriate action when exiting.
+	 */
+	mask = SPU_STATUS_INVALID_INSTR |
+	    SPU_STATUS_SINGLE_STEP |
+	    SPU_STATUS_STOPPED_BY_HALT |
+	    SPU_STATUS_STOPPED_BY_STOP | SPU_STATUS_RUNNING;
+	if (!(csa->prob.spu_status_R & mask)) {
+		csa->lscsa->stopped_status.slot[0] = SPU_STOPPED_STATUS_R;
+	}
+}
+
+static inline void restore_mfc_rag(struct spu_state *csa, struct spu *spu)
+{
+	/* Restore, Step 29:
+	 *     Restore RA_GROUP_ID register and the
+	 *     RA_ENABLE reigster from the CSA.
+	 */
+	spu_resource_allocation_groupID_set(spu,
+			csa->priv1.resource_allocation_groupID_RW);
+	spu_resource_allocation_enable_set(spu,
+			csa->priv1.resource_allocation_enable_RW);
+}
+
+static inline void send_restore_code(struct spu_state *csa, struct spu *spu)
+{
+	unsigned long addr = (unsigned long)&spu_restore_code[0];
+	unsigned int ls_offset = 0x0;
+	unsigned int size = sizeof(spu_restore_code);
+	unsigned int tag = 0;
+	unsigned int rclass = 0;
+	unsigned int cmd = MFC_GETFS_CMD;
+
+	/* Restore, Step 37:
+	 *     Issue MFC DMA command to copy context
+	 *     restore code to local storage.
+	 */
+	send_mfc_dma(spu, addr, ls_offset, size, tag, rclass, cmd);
+}
+
+static inline void setup_decr(struct spu_state *csa, struct spu *spu)
+{
+	/* Restore, Step 34:
+	 *     If CSA.MFC_CNTL[Ds]=1 (decrementer was
+	 *     running) then adjust decrementer, set
+	 *     decrementer running status in LSCSA,
+	 *     and set decrementer "wrapped" status
+	 *     in LSCSA.
+	 */
+	if (csa->priv2.mfc_control_RW & MFC_CNTL_DECREMENTER_RUNNING) {
+		cycles_t resume_time = get_cycles();
+		cycles_t delta_time = resume_time - csa->suspend_time;
+
+		csa->lscsa->decr.slot[0] = delta_time;
+	}
+}
+
+static inline void setup_ppu_mb(struct spu_state *csa, struct spu *spu)
+{
+	/* Restore, Step 35:
+	 *     Copy the CSA.PU_MB data into the LSCSA.
+	 */
+	csa->lscsa->ppu_mb.slot[0] = csa->prob.pu_mb_R;
+}
+
+static inline void setup_ppuint_mb(struct spu_state *csa, struct spu *spu)
+{
+	/* Restore, Step 36:
+	 *     Copy the CSA.PUINT_MB data into the LSCSA.
+	 */
+	csa->lscsa->ppuint_mb.slot[0] = csa->priv2.puint_mb_R;
+}
+
+static inline int check_restore_status(struct spu_state *csa, struct spu *spu)
+{
+	struct spu_problem __iomem *prob = spu->problem;
+	u32 complete;
+
+	/* Restore, Step 40:
+	 *     If SPU_Status[P]=1 and SPU_Status[SC] = "success",
+	 *     context restore succeeded, otherwise context restore
+	 *     failed.
+	 */
+	complete = ((SPU_RESTORE_COMPLETE << SPU_STOP_STATUS_SHIFT) |
+		    SPU_STATUS_STOPPED_BY_STOP);
+	return (in_be32(&prob->spu_status_R) != complete) ? 1 : 0;
+}
+
+static inline void restore_spu_privcntl(struct spu_state *csa, struct spu *spu)
+{
+	struct spu_priv2 __iomem *priv2 = spu->priv2;
+
+	/* Restore, Step 41:
+	 *     Restore SPU_PrivCntl from the CSA.
+	 */
+	out_be64(&priv2->spu_privcntl_RW, csa->priv2.spu_privcntl_RW);
+	eieio();
+}
+
+static inline void restore_status_part1(struct spu_state *csa, struct spu *spu)
+{
+	struct spu_problem __iomem *prob = spu->problem;
+	u32 mask;
+
+	/* Restore, Step 42:
+	 *     If any CSA.SPU_Status[I,S,H,P]=1, then
+	 *     restore the error or single step state.
+	 */
+	mask = SPU_STATUS_INVALID_INSTR |
+	    SPU_STATUS_SINGLE_STEP |
+	    SPU_STATUS_STOPPED_BY_HALT | SPU_STATUS_STOPPED_BY_STOP;
+	if (csa->prob.spu_status_R & mask) {
+		out_be32(&prob->spu_runcntl_RW, SPU_RUNCNTL_RUNNABLE);
+		eieio();
+		POLL_WHILE_TRUE(in_be32(&prob->spu_status_R) &
+				SPU_STATUS_RUNNING);
+	}
+}
+
+static inline void restore_status_part2(struct spu_state *csa, struct spu *spu)
+{
+	struct spu_problem __iomem *prob = spu->problem;
+	u32 mask;
+
+	/* Restore, Step 43:
+	 *     If all CSA.SPU_Status[I,S,H,P,R]=0 then write
+	 *     SPU_RunCntl[R0R1]='01', wait for SPU_Status[R]=1,
+	 *     then write '00' to SPU_RunCntl[R0R1] and wait
+	 *     for SPU_Status[R]=0.
+	 */
+	mask = SPU_STATUS_INVALID_INSTR |
+	    SPU_STATUS_SINGLE_STEP |
+	    SPU_STATUS_STOPPED_BY_HALT |
+	    SPU_STATUS_STOPPED_BY_STOP | SPU_STATUS_RUNNING;
+	if (!(csa->prob.spu_status_R & mask)) {
+		out_be32(&prob->spu_runcntl_RW, SPU_RUNCNTL_RUNNABLE);
+		eieio();
+		POLL_WHILE_FALSE(in_be32(&prob->spu_status_R) &
+				 SPU_STATUS_RUNNING);
+		out_be32(&prob->spu_runcntl_RW, SPU_RUNCNTL_STOP);
+		eieio();
+		POLL_WHILE_TRUE(in_be32(&prob->spu_status_R) &
+				SPU_STATUS_RUNNING);
+	}
+}
+
+static inline void restore_ls_16kb(struct spu_state *csa, struct spu *spu)
+{
+	unsigned long addr = (unsigned long)&csa->lscsa->ls[0];
+	unsigned int ls_offset = 0x0;
+	unsigned int size = 16384;
+	unsigned int tag = 0;
+	unsigned int rclass = 0;
+	unsigned int cmd = MFC_GET_CMD;
+
+	/* Restore, Step 44:
+	 *     Issue a DMA command to restore the first
+	 *     16kb of local storage from CSA.
+	 */
+	send_mfc_dma(spu, addr, ls_offset, size, tag, rclass, cmd);
+}
+
+static inline void clear_interrupts(struct spu_state *csa, struct spu *spu)
+{
+	/* Restore, Step 49:
+	 *     Write INT_MASK_class0 with value of 0.
+	 *     Write INT_MASK_class1 with value of 0.
+	 *     Write INT_MASK_class2 with value of 0.
+	 *     Write INT_STAT_class0 with value of -1.
+	 *     Write INT_STAT_class1 with value of -1.
+	 *     Write INT_STAT_class2 with value of -1.
+	 */
+	spin_lock_irq(&spu->register_lock);
+	spu_int_mask_set(spu, 0, 0ul);
+	spu_int_mask_set(spu, 1, 0ul);
+	spu_int_mask_set(spu, 2, 0ul);
+	spu_int_stat_clear(spu, 0, ~0ul);
+	spu_int_stat_clear(spu, 1, ~0ul);
+	spu_int_stat_clear(spu, 2, ~0ul);
+	spin_unlock_irq(&spu->register_lock);
+}
+
+static inline void restore_mfc_queues(struct spu_state *csa, struct spu *spu)
+{
+	struct spu_priv2 __iomem *priv2 = spu->priv2;
+	int i;
+
+	/* Restore, Step 50:
+	 *     If MFC_Cntl[Se]!=0 then restore
+	 *     MFC command queues.
+	 */
+	if ((csa->priv2.mfc_control_RW & MFC_CNTL_DMA_QUEUES_EMPTY_MASK) == 0) {
+		for (i = 0; i < 8; i++) {
+			out_be64(&priv2->puq[i].mfc_cq_data0_RW,
+				 csa->priv2.puq[i].mfc_cq_data0_RW);
+			out_be64(&priv2->puq[i].mfc_cq_data1_RW,
+				 csa->priv2.puq[i].mfc_cq_data1_RW);
+			out_be64(&priv2->puq[i].mfc_cq_data2_RW,
+				 csa->priv2.puq[i].mfc_cq_data2_RW);
+			out_be64(&priv2->puq[i].mfc_cq_data3_RW,
+				 csa->priv2.puq[i].mfc_cq_data3_RW);
+		}
+		for (i = 0; i < 16; i++) {
+			out_be64(&priv2->spuq[i].mfc_cq_data0_RW,
+				 csa->priv2.spuq[i].mfc_cq_data0_RW);
+			out_be64(&priv2->spuq[i].mfc_cq_data1_RW,
+				 csa->priv2.spuq[i].mfc_cq_data1_RW);
+			out_be64(&priv2->spuq[i].mfc_cq_data2_RW,
+				 csa->priv2.spuq[i].mfc_cq_data2_RW);
+			out_be64(&priv2->spuq[i].mfc_cq_data3_RW,
+				 csa->priv2.spuq[i].mfc_cq_data3_RW);
+		}
+	}
+	eieio();
+}
+
+static inline void restore_ppu_querymask(struct spu_state *csa, struct spu *spu)
+{
+	struct spu_problem __iomem *prob = spu->problem;
+
+	/* Restore, Step 51:
+	 *     Restore the PPU_QueryMask register from CSA.
+	 */
+	out_be32(&prob->dma_querymask_RW, csa->prob.dma_querymask_RW);
+	eieio();
+}
+
+static inline void restore_ppu_querytype(struct spu_state *csa, struct spu *spu)
+{
+	struct spu_problem __iomem *prob = spu->problem;
+
+	/* Restore, Step 52:
+	 *     Restore the PPU_QueryType register from CSA.
+	 */
+	out_be32(&prob->dma_querytype_RW, csa->prob.dma_querytype_RW);
+	eieio();
+}
+
+static inline void restore_mfc_csr_tsq(struct spu_state *csa, struct spu *spu)
+{
+	struct spu_priv2 __iomem *priv2 = spu->priv2;
+
+	/* Restore, Step 53:
+	 *     Restore the MFC_CSR_TSQ register from CSA.
+	 */
+	out_be64(&priv2->spu_tag_status_query_RW,
+		 csa->priv2.spu_tag_status_query_RW);
+	eieio();
+}
+
+static inline void restore_mfc_csr_cmd(struct spu_state *csa, struct spu *spu)
+{
+	struct spu_priv2 __iomem *priv2 = spu->priv2;
+
+	/* Restore, Step 54:
+	 *     Restore the MFC_CSR_CMD1 and MFC_CSR_CMD2
+	 *     registers from CSA.
+	 */
+	out_be64(&priv2->spu_cmd_buf1_RW, csa->priv2.spu_cmd_buf1_RW);
+	out_be64(&priv2->spu_cmd_buf2_RW, csa->priv2.spu_cmd_buf2_RW);
+	eieio();
+}
+
+static inline void restore_mfc_csr_ato(struct spu_state *csa, struct spu *spu)
+{
+	struct spu_priv2 __iomem *priv2 = spu->priv2;
+
+	/* Restore, Step 55:
+	 *     Restore the MFC_CSR_ATO register from CSA.
+	 */
+	out_be64(&priv2->spu_atomic_status_RW, csa->priv2.spu_atomic_status_RW);
+}
+
+static inline void restore_mfc_tclass_id(struct spu_state *csa, struct spu *spu)
+{
+	/* Restore, Step 56:
+	 *     Restore the MFC_TCLASS_ID register from CSA.
+	 */
+	spu_mfc_tclass_id_set(spu, csa->priv1.mfc_tclass_id_RW);
+	eieio();
+}
+
+static inline void set_llr_event(struct spu_state *csa, struct spu *spu)
+{
+	u64 ch0_cnt, ch0_data;
+	u64 ch1_data;
+
+	/* Restore, Step 57:
+	 *    Set the Lock Line Reservation Lost Event by:
+	 *      1. OR CSA.SPU_Event_Status with bit 21 (Lr) set to 1.
+	 *      2. If CSA.SPU_Channel_0_Count=0 and
+	 *         CSA.SPU_Wr_Event_Mask[Lr]=1 and
+	 *         CSA.SPU_Event_Status[Lr]=0 then set
+	 *         CSA.SPU_Event_Status_Count=1.
+	 */
+	ch0_cnt = csa->spu_chnlcnt_RW[0];
+	ch0_data = csa->spu_chnldata_RW[0];
+	ch1_data = csa->spu_chnldata_RW[1];
+	csa->spu_chnldata_RW[0] |= MFC_LLR_LOST_EVENT;
+	if ((ch0_cnt == 0) && !(ch0_data & MFC_LLR_LOST_EVENT) &&
+	    (ch1_data & MFC_LLR_LOST_EVENT)) {
+		csa->spu_chnlcnt_RW[0] = 1;
+	}
+}
+
+static inline void restore_decr_wrapped(struct spu_state *csa, struct spu *spu)
+{
+	/* Restore, Step 58:
+	 *     If the status of the CSA software decrementer
+	 *     "wrapped" flag is set, OR in a '1' to
+	 *     CSA.SPU_Event_Status[Tm].
+	 */
+	if (csa->lscsa->decr_status.slot[0] == 1) {
+		csa->spu_chnldata_RW[0] |= 0x20;
+	}
+	if ((csa->lscsa->decr_status.slot[0] == 1) &&
+	    (csa->spu_chnlcnt_RW[0] == 0 &&
+	     ((csa->spu_chnldata_RW[2] & 0x20) == 0x0) &&
+	     ((csa->spu_chnldata_RW[0] & 0x20) != 0x1))) {
+		csa->spu_chnlcnt_RW[0] = 1;
+	}
+}
+
+static inline void restore_ch_part1(struct spu_state *csa, struct spu *spu)
+{
+	struct spu_priv2 __iomem *priv2 = spu->priv2;
+	u64 idx, ch_indices[7] = { 0UL, 1UL, 3UL, 4UL, 24UL, 25UL, 27UL };
+	int i;
+
+	/* Restore, Step 59:
+	 *     Restore the following CH: [0,1,3,4,24,25,27]
+	 */
+	for (i = 0; i < 7; i++) {
+		idx = ch_indices[i];
+		out_be64(&priv2->spu_chnlcntptr_RW, idx);
+		eieio();
+		out_be64(&priv2->spu_chnldata_RW, csa->spu_chnldata_RW[idx]);
+		out_be64(&priv2->spu_chnlcnt_RW, csa->spu_chnlcnt_RW[idx]);
+		eieio();
+	}
+}
+
+static inline void restore_ch_part2(struct spu_state *csa, struct spu *spu)
+{
+	struct spu_priv2 __iomem *priv2 = spu->priv2;
+	u64 ch_indices[3] = { 9UL, 21UL, 23UL };
+	u64 ch_counts[3] = { 1UL, 16UL, 1UL };
+	u64 idx;
+	int i;
+
+	/* Restore, Step 60:
+	 *     Restore the following CH: [9,21,23].
+	 */
+	ch_counts[0] = 1UL;
+	ch_counts[1] = csa->spu_chnlcnt_RW[21];
+	ch_counts[2] = 1UL;
+	for (i = 0; i < 3; i++) {
+		idx = ch_indices[i];
+		out_be64(&priv2->spu_chnlcntptr_RW, idx);
+		eieio();
+		out_be64(&priv2->spu_chnlcnt_RW, ch_counts[i]);
+		eieio();
+	}
+}
+
+static inline void restore_spu_lslr(struct spu_state *csa, struct spu *spu)
+{
+	struct spu_priv2 __iomem *priv2 = spu->priv2;
+
+	/* Restore, Step 61:
+	 *     Restore the SPU_LSLR register from CSA.
+	 */
+	out_be64(&priv2->spu_lslr_RW, csa->priv2.spu_lslr_RW);
+	eieio();
+}
+
+static inline void restore_spu_cfg(struct spu_state *csa, struct spu *spu)
+{
+	struct spu_priv2 __iomem *priv2 = spu->priv2;
+
+	/* Restore, Step 62:
+	 *     Restore the SPU_Cfg register from CSA.
+	 */
+	out_be64(&priv2->spu_cfg_RW, csa->priv2.spu_cfg_RW);
+	eieio();
+}
+
+static inline void restore_pm_trace(struct spu_state *csa, struct spu *spu)
+{
+	/* Restore, Step 63:
+	 *     Restore PM_Trace_Tag_Wait_Mask from CSA.
+	 *     Not performed by this implementation.
+	 */
+}
+
+static inline void restore_spu_npc(struct spu_state *csa, struct spu *spu)
+{
+	struct spu_problem __iomem *prob = spu->problem;
+
+	/* Restore, Step 64:
+	 *     Restore SPU_NPC from CSA.
+	 */
+	out_be32(&prob->spu_npc_RW, csa->prob.spu_npc_RW);
+	eieio();
+}
+
+static inline void restore_spu_mb(struct spu_state *csa, struct spu *spu)
+{
+	struct spu_priv2 __iomem *priv2 = spu->priv2;
+	int i;
+
+	/* Restore, Step 65:
+	 *     Restore MFC_RdSPU_MB from CSA.
+	 */
+	out_be64(&priv2->spu_chnlcntptr_RW, 29UL);
+	eieio();
+	out_be64(&priv2->spu_chnlcnt_RW, csa->spu_chnlcnt_RW[29]);
+	for (i = 0; i < 4; i++) {
+		out_be64(&priv2->spu_chnldata_RW, csa->spu_mailbox_data[i]);
+	}
+	eieio();
+}
+
+static inline void check_ppu_mb_stat(struct spu_state *csa, struct spu *spu)
+{
+	struct spu_problem __iomem *prob = spu->problem;
+	u32 dummy = 0;
+
+	/* Restore, Step 66:
+	 *     If CSA.MB_Stat[P]=0 (mailbox empty) then
+	 *     read from the PPU_MB register.
+	 */
+	if ((csa->prob.mb_stat_R & 0xFF) == 0) {
+		dummy = in_be32(&prob->pu_mb_R);
+		eieio();
+	}
+}
+
+static inline void check_ppuint_mb_stat(struct spu_state *csa, struct spu *spu)
+{
+	struct spu_priv2 __iomem *priv2 = spu->priv2;
+	u64 dummy = 0UL;
+
+	/* Restore, Step 66:
+	 *     If CSA.MB_Stat[I]=0 (mailbox empty) then
+	 *     read from the PPUINT_MB register.
+	 */
+	if ((csa->prob.mb_stat_R & 0xFF0000) == 0) {
+		dummy = in_be64(&priv2->puint_mb_R);
+		eieio();
+		spu_int_stat_clear(spu, 2, CLASS2_ENABLE_MAILBOX_INTR);
+		eieio();
+	}
+}
+
+static inline void restore_mfc_slbs(struct spu_state *csa, struct spu *spu)
+{
+	struct spu_priv2 __iomem *priv2 = spu->priv2;
+	int i;
+
+	/* Restore, Step 68:
+	 *     If MFC_SR1[R]='1', restore SLBs from CSA.
+	 */
+	if (csa->priv1.mfc_sr1_RW & MFC_STATE1_RELOCATE_MASK) {
+		for (i = 0; i < 8; i++) {
+			out_be64(&priv2->slb_index_W, i);
+			eieio();
+			out_be64(&priv2->slb_esid_RW, csa->slb_esid_RW[i]);
+			out_be64(&priv2->slb_vsid_RW, csa->slb_vsid_RW[i]);
+			eieio();
+		}
+		out_be64(&priv2->slb_index_W, csa->priv2.slb_index_W);
+		eieio();
+	}
+}
+
+static inline void restore_mfc_sr1(struct spu_state *csa, struct spu *spu)
+{
+	/* Restore, Step 69:
+	 *     Restore the MFC_SR1 register from CSA.
+	 */
+	spu_mfc_sr1_set(spu, csa->priv1.mfc_sr1_RW);
+	eieio();
+}
+
+static inline void restore_other_spu_access(struct spu_state *csa,
+					    struct spu *spu)
+{
+	/* Restore, Step 70:
+	 *     Restore other SPU mappings to this SPU. TBD.
+	 */
+}
+
+static inline void restore_spu_runcntl(struct spu_state *csa, struct spu *spu)
+{
+	struct spu_problem __iomem *prob = spu->problem;
+
+	/* Restore, Step 71:
+	 *     If CSA.SPU_Status[R]=1 then write
+	 *     SPU_RunCntl[R0R1]='01'.
+	 */
+	if (csa->prob.spu_status_R & SPU_STATUS_RUNNING) {
+		out_be32(&prob->spu_runcntl_RW, SPU_RUNCNTL_RUNNABLE);
+		eieio();
+	}
+}
+
+static inline void restore_mfc_cntl(struct spu_state *csa, struct spu *spu)
+{
+	struct spu_priv2 __iomem *priv2 = spu->priv2;
+
+	/* Restore, Step 72:
+	 *    Restore the MFC_CNTL register for the CSA.
+	 */
+	out_be64(&priv2->mfc_control_RW, csa->priv2.mfc_control_RW);
+	eieio();
+}
+
+static inline void enable_user_access(struct spu_state *csa, struct spu *spu)
+{
+	/* Restore, Step 73:
+	 *     Enable user-space access (if provided) to this
+	 *     SPU by mapping the virtual pages assigned to
+	 *     the SPU memory-mapped I/O (MMIO) for problem
+	 *     state. TBD.
+	 */
+}
+
+static inline void reset_switch_active(struct spu_state *csa, struct spu *spu)
+{
+	/* Restore, Step 74:
+	 *     Reset the "context switch active" flag.
+	 */
+	clear_bit(SPU_CONTEXT_SWITCH_ACTIVE, &spu->flags);
+	mb();
+}
+
+static inline void reenable_interrupts(struct spu_state *csa, struct spu *spu)
+{
+	/* Restore, Step 75:
+	 *     Re-enable SPU interrupts.
+	 */
+	spin_lock_irq(&spu->register_lock);
+	spu_int_mask_set(spu, 0, csa->priv1.int_mask_class0_RW);
+	spu_int_mask_set(spu, 1, csa->priv1.int_mask_class1_RW);
+	spu_int_mask_set(spu, 2, csa->priv1.int_mask_class2_RW);
+	spin_unlock_irq(&spu->register_lock);
+}
+
+static int quiece_spu(struct spu_state *prev, struct spu *spu)
+{
+	/*
+	 * Combined steps 2-18 of SPU context save sequence, which
+	 * quiesce the SPU state (disable SPU execution, MFC command
+	 * queues, decrementer, SPU interrupts, etc.).
+	 *
+	 * Returns      0 on success.
+	 *              2 if failed step 2.
+	 *              6 if failed step 6.
+	 */
+
+	if (check_spu_isolate(prev, spu)) {	/* Step 2. */
+		return 2;
+	}
+	disable_interrupts(prev, spu);	        /* Step 3. */
+	set_watchdog_timer(prev, spu);	        /* Step 4. */
+	inhibit_user_access(prev, spu);	        /* Step 5. */
+	if (check_spu_isolate(prev, spu)) {	/* Step 6. */
+		return 6;
+	}
+	set_switch_pending(prev, spu);	        /* Step 7. */
+	save_mfc_cntl(prev, spu);		/* Step 8. */
+	save_spu_runcntl(prev, spu);	        /* Step 9. */
+	save_mfc_sr1(prev, spu);	        /* Step 10. */
+	save_spu_status(prev, spu);	        /* Step 11. */
+	save_mfc_decr(prev, spu);	        /* Step 12. */
+	halt_mfc_decr(prev, spu);	        /* Step 13. */
+	save_timebase(prev, spu);		/* Step 14. */
+	remove_other_spu_access(prev, spu);	/* Step 15. */
+	do_mfc_mssync(prev, spu);	        /* Step 16. */
+	issue_mfc_tlbie(prev, spu);	        /* Step 17. */
+	handle_pending_interrupts(prev, spu);	/* Step 18. */
+
+	return 0;
+}
+
+static void save_csa(struct spu_state *prev, struct spu *spu)
+{
+	/*
+	 * Combine steps 19-44 of SPU context save sequence, which
+	 * save regions of the privileged & problem state areas.
+	 */
+
+	save_mfc_queues(prev, spu);	/* Step 19. */
+	save_ppu_querymask(prev, spu);	/* Step 20. */
+	save_ppu_querytype(prev, spu);	/* Step 21. */
+	save_mfc_csr_tsq(prev, spu);	/* Step 22. */
+	save_mfc_csr_cmd(prev, spu);	/* Step 23. */
+	save_mfc_csr_ato(prev, spu);	/* Step 24. */
+	save_mfc_tclass_id(prev, spu);	/* Step 25. */
+	set_mfc_tclass_id(prev, spu);	/* Step 26. */
+	purge_mfc_queue(prev, spu);	/* Step 27. */
+	wait_purge_complete(prev, spu);	/* Step 28. */
+	save_mfc_slbs(prev, spu);	/* Step 29. */
+	setup_mfc_sr1(prev, spu);	/* Step 30. */
+	save_spu_npc(prev, spu);	/* Step 31. */
+	save_spu_privcntl(prev, spu);	/* Step 32. */
+	reset_spu_privcntl(prev, spu);	/* Step 33. */
+	save_spu_lslr(prev, spu);	/* Step 34. */
+	reset_spu_lslr(prev, spu);	/* Step 35. */
+	save_spu_cfg(prev, spu);	/* Step 36. */
+	save_pm_trace(prev, spu);	/* Step 37. */
+	save_mfc_rag(prev, spu);	/* Step 38. */
+	save_ppu_mb_stat(prev, spu);	/* Step 39. */
+	save_ppu_mb(prev, spu);	        /* Step 40. */
+	save_ppuint_mb(prev, spu);	/* Step 41. */
+	save_ch_part1(prev, spu);	/* Step 42. */
+	save_spu_mb(prev, spu);	        /* Step 43. */
+	save_mfc_cmd(prev, spu);	/* Step 44. */
+	reset_ch(prev, spu);	        /* Step 45. */
+}
+
+static void save_lscsa(struct spu_state *prev, struct spu *spu)
+{
+	/*
+	 * Perform steps 46-57 of SPU context save sequence,
+	 * which save regions of the local store and register
+	 * file.
+	 */
+
+	resume_mfc_queue(prev, spu);	/* Step 46. */
+	setup_mfc_slbs(prev, spu);	/* Step 47. */
+	set_switch_active(prev, spu);	/* Step 48. */
+	enable_interrupts(prev, spu);	/* Step 49. */
+	save_ls_16kb(prev, spu);	/* Step 50. */
+	set_spu_npc(prev, spu);	        /* Step 51. */
+	set_signot1(prev, spu);		/* Step 52. */
+	set_signot2(prev, spu);		/* Step 53. */
+	send_save_code(prev, spu);	/* Step 54. */
+	set_ppu_querymask(prev, spu);	/* Step 55. */
+	wait_tag_complete(prev, spu);	/* Step 56. */
+	wait_spu_stopped(prev, spu);	/* Step 57. */
+}
+
+static void harvest(struct spu_state *prev, struct spu *spu)
+{
+	/*
+	 * Perform steps 2-25 of SPU context restore sequence,
+	 * which resets an SPU either after a failed save, or
+	 * when using SPU for first time.
+	 */
+
+	disable_interrupts(prev, spu);	        /* Step 2.  */
+	inhibit_user_access(prev, spu);	        /* Step 3.  */
+	terminate_spu_app(prev, spu);	        /* Step 4.  */
+	set_switch_pending(prev, spu);	        /* Step 5.  */
+	remove_other_spu_access(prev, spu);	/* Step 6.  */
+	suspend_mfc(prev, spu);	                /* Step 7.  */
+	wait_suspend_mfc_complete(prev, spu);	/* Step 8.  */
+	if (!suspend_spe(prev, spu))	        /* Step 9.  */
+		clear_spu_status(prev, spu);	/* Step 10. */
+	do_mfc_mssync(prev, spu);	        /* Step 11. */
+	issue_mfc_tlbie(prev, spu);	        /* Step 12. */
+	handle_pending_interrupts(prev, spu);	/* Step 13. */
+	purge_mfc_queue(prev, spu);	        /* Step 14. */
+	wait_purge_complete(prev, spu);	        /* Step 15. */
+	reset_spu_privcntl(prev, spu);	        /* Step 16. */
+	reset_spu_lslr(prev, spu);              /* Step 17. */
+	setup_mfc_sr1(prev, spu);	        /* Step 18. */
+	invalidate_slbs(prev, spu);	        /* Step 19. */
+	reset_ch_part1(prev, spu);	        /* Step 20. */
+	reset_ch_part2(prev, spu);	        /* Step 21. */
+	enable_interrupts(prev, spu);	        /* Step 22. */
+	set_switch_active(prev, spu);	        /* Step 23. */
+	set_mfc_tclass_id(prev, spu);	        /* Step 24. */
+	resume_mfc_queue(prev, spu);	        /* Step 25. */
+}
+
+static void restore_lscsa(struct spu_state *next, struct spu *spu)
+{
+	/*
+	 * Perform steps 26-40 of SPU context restore sequence,
+	 * which restores regions of the local store and register
+	 * file.
+	 */
+
+	set_watchdog_timer(next, spu);	        /* Step 26. */
+	setup_spu_status_part1(next, spu);	/* Step 27. */
+	setup_spu_status_part2(next, spu);	/* Step 28. */
+	restore_mfc_rag(next, spu);	        /* Step 29. */
+	setup_mfc_slbs(next, spu);	        /* Step 30. */
+	set_spu_npc(next, spu);	                /* Step 31. */
+	set_signot1(next, spu);	                /* Step 32. */
+	set_signot2(next, spu);	                /* Step 33. */
+	setup_decr(next, spu);	                /* Step 34. */
+	setup_ppu_mb(next, spu);	        /* Step 35. */
+	setup_ppuint_mb(next, spu);	        /* Step 36. */
+	send_restore_code(next, spu);	        /* Step 37. */
+	set_ppu_querymask(next, spu);	        /* Step 38. */
+	wait_tag_complete(next, spu);	        /* Step 39. */
+	wait_spu_stopped(next, spu);	        /* Step 40. */
+}
+
+static void restore_csa(struct spu_state *next, struct spu *spu)
+{
+	/*
+	 * Combine steps 41-76 of SPU context restore sequence, which
+	 * restore regions of the privileged & problem state areas.
+	 */
+
+	restore_spu_privcntl(next, spu);	/* Step 41. */
+	restore_status_part1(next, spu);	/* Step 42. */
+	restore_status_part2(next, spu);	/* Step 43. */
+	restore_ls_16kb(next, spu);	        /* Step 44. */
+	wait_tag_complete(next, spu);	        /* Step 45. */
+	suspend_mfc(next, spu);	                /* Step 46. */
+	wait_suspend_mfc_complete(next, spu);	/* Step 47. */
+	issue_mfc_tlbie(next, spu);	        /* Step 48. */
+	clear_interrupts(next, spu);	        /* Step 49. */
+	restore_mfc_queues(next, spu);	        /* Step 50. */
+	restore_ppu_querymask(next, spu);	/* Step 51. */
+	restore_ppu_querytype(next, spu);	/* Step 52. */
+	restore_mfc_csr_tsq(next, spu);	        /* Step 53. */
+	restore_mfc_csr_cmd(next, spu);	        /* Step 54. */
+	restore_mfc_csr_ato(next, spu);	        /* Step 55. */
+	restore_mfc_tclass_id(next, spu);	/* Step 56. */
+	set_llr_event(next, spu);	        /* Step 57. */
+	restore_decr_wrapped(next, spu);	/* Step 58. */
+	restore_ch_part1(next, spu);	        /* Step 59. */
+	restore_ch_part2(next, spu);	        /* Step 60. */
+	restore_spu_lslr(next, spu);	        /* Step 61. */
+	restore_spu_cfg(next, spu);	        /* Step 62. */
+	restore_pm_trace(next, spu);	        /* Step 63. */
+	restore_spu_npc(next, spu);	        /* Step 64. */
+	restore_spu_mb(next, spu);	        /* Step 65. */
+	check_ppu_mb_stat(next, spu);	        /* Step 66. */
+	check_ppuint_mb_stat(next, spu);	/* Step 67. */
+	restore_mfc_slbs(next, spu);	        /* Step 68. */
+	restore_mfc_sr1(next, spu);	        /* Step 69. */
+	restore_other_spu_access(next, spu);	/* Step 70. */
+	restore_spu_runcntl(next, spu);	        /* Step 71. */
+	restore_mfc_cntl(next, spu);	        /* Step 72. */
+	enable_user_access(next, spu);	        /* Step 73. */
+	reset_switch_active(next, spu);	        /* Step 74. */
+	reenable_interrupts(next, spu);	        /* Step 75. */
+}
+
+static int __do_spu_save(struct spu_state *prev, struct spu *spu)
+{
+	int rc;
+
+	/*
+	 * SPU context save can be broken into three phases:
+	 *
+	 *     (a) quiesce [steps 2-16].
+	 *     (b) save of CSA, performed by PPE [steps 17-42]
+	 *     (c) save of LSCSA, mostly performed by SPU [steps 43-52].
+	 *
+	 * Returns      0 on success.
+	 *              2,6 if failed to quiece SPU
+	 *              53 if SPU-side of save failed.
+	 */
+
+	rc = quiece_spu(prev, spu);	        /* Steps 2-16. */
+	switch (rc) {
+	default:
+	case 2:
+	case 6:
+		harvest(prev, spu);
+		return rc;
+		break;
+	case 0:
+		break;
+	}
+	save_csa(prev, spu);	                /* Steps 17-43. */
+	save_lscsa(prev, spu);	                /* Steps 44-53. */
+	return check_save_status(prev, spu);	/* Step 54.     */
+}
+
+static int __do_spu_restore(struct spu_state *next, struct spu *spu)
+{
+	int rc;
+
+	/*
+	 * SPU context restore can be broken into three phases:
+	 *
+	 *    (a) harvest (or reset) SPU [steps 2-24].
+	 *    (b) restore LSCSA [steps 25-40], mostly performed by SPU.
+	 *    (c) restore CSA [steps 41-76], performed by PPE.
+	 *
+	 * The 'harvest' step is not performed here, but rather
+	 * as needed below.
+	 */
+
+	restore_lscsa(next, spu);	        /* Steps 24-39. */
+	rc = check_restore_status(next, spu);	/* Step 40.     */
+	switch (rc) {
+	default:
+		/* Failed. Return now. */
+		return rc;
+		break;
+	case 0:
+		/* Fall through to next step. */
+		break;
+	}
+	restore_csa(next, spu);
+
+	return 0;
+}
+
+/**
+ * spu_save - SPU context save, with locking.
+ * @prev: pointer to SPU context save area, to be saved.
+ * @spu: pointer to SPU iomem structure.
+ *
+ * Acquire locks, perform the save operation then return.
+ */
+int spu_save(struct spu_state *prev, struct spu *spu)
+{
+	int rc;
+
+	acquire_spu_lock(spu);	        /* Step 1.     */
+	rc = __do_spu_save(prev, spu);	/* Steps 2-53. */
+	release_spu_lock(spu);
+	if (rc) {
+		panic("%s failed on SPU[%d], rc=%d.\n",
+		      __func__, spu->number, rc);
+	}
+	return rc;
+}
+
+/**
+ * spu_restore - SPU context restore, with harvest and locking.
+ * @new: pointer to SPU context save area, to be restored.
+ * @spu: pointer to SPU iomem structure.
+ *
+ * Perform harvest + restore, as we may not be coming
+ * from a previous succesful save operation, and the
+ * hardware state is unknown.
+ */
+int spu_restore(struct spu_state *new, struct spu *spu)
+{
+	int rc;
+
+	acquire_spu_lock(spu);
+	harvest(NULL, spu);
+	spu->stop_code = 0;
+	spu->dar = 0;
+	spu->dsisr = 0;
+	spu->slb_replace = 0;
+	spu->class_0_pending = 0;
+	rc = __do_spu_restore(new, spu);
+	release_spu_lock(spu);
+	if (rc) {
+		panic("%s failed on SPU[%d] rc=%d.\n",
+		       __func__, spu->number, rc);
+	}
+	return rc;
+}
+
+/**
+ * spu_harvest - SPU harvest (reset) operation
+ * @spu: pointer to SPU iomem structure.
+ *
+ * Perform SPU harvest (reset) operation.
+ */
+void spu_harvest(struct spu *spu)
+{
+	acquire_spu_lock(spu);
+	harvest(NULL, spu);
+	release_spu_lock(spu);
+}
+
+static void init_prob(struct spu_state *csa)
+{
+	csa->spu_chnlcnt_RW[9] = 1;
+	csa->spu_chnlcnt_RW[21] = 16;
+	csa->spu_chnlcnt_RW[23] = 1;
+	csa->spu_chnlcnt_RW[28] = 1;
+	csa->spu_chnlcnt_RW[30] = 1;
+	csa->prob.spu_runcntl_RW = SPU_RUNCNTL_STOP;
+}
+
+static void init_priv1(struct spu_state *csa)
+{
+	/* Enable decode, relocate, tlbie response, master runcntl. */
+	csa->priv1.mfc_sr1_RW = MFC_STATE1_LOCAL_STORAGE_DECODE_MASK |
+	    MFC_STATE1_MASTER_RUN_CONTROL_MASK |
+	    MFC_STATE1_PROBLEM_STATE_MASK |
+	    MFC_STATE1_RELOCATE_MASK | MFC_STATE1_BUS_TLBIE_MASK;
+
+	/* Set storage description.  */
+	csa->priv1.mfc_sdr_RW = mfspr(SPRN_SDR1);
+
+	/* Enable OS-specific set of interrupts. */
+	csa->priv1.int_mask_class0_RW = CLASS0_ENABLE_DMA_ALIGNMENT_INTR |
+	    CLASS0_ENABLE_INVALID_DMA_COMMAND_INTR |
+	    CLASS0_ENABLE_SPU_ERROR_INTR;
+	csa->priv1.int_mask_class1_RW = CLASS1_ENABLE_SEGMENT_FAULT_INTR |
+	    CLASS1_ENABLE_STORAGE_FAULT_INTR;
+	csa->priv1.int_mask_class2_RW = CLASS2_ENABLE_SPU_STOP_INTR |
+	    CLASS2_ENABLE_SPU_HALT_INTR;
+}
+
+static void init_priv2(struct spu_state *csa)
+{
+	csa->priv2.spu_lslr_RW = LS_ADDR_MASK;
+	csa->priv2.mfc_control_RW = MFC_CNTL_RESUME_DMA_QUEUE |
+	    MFC_CNTL_NORMAL_DMA_QUEUE_OPERATION |
+	    MFC_CNTL_DMA_QUEUES_EMPTY_MASK;
+}
+
+/**
+ * spu_alloc_csa - allocate and initialize an SPU context save area.
+ *
+ * Allocate and initialize the contents of an SPU context save area.
+ * This includes enabling address translation, interrupt masks, etc.,
+ * as appropriate for the given OS environment.
+ *
+ * Note that storage for the 'lscsa' is allocated separately,
+ * as it is by far the largest of the context save regions,
+ * and may need to be pinned or otherwise specially aligned.
+ */
+void spu_init_csa(struct spu_state *csa)
+{
+	struct spu_lscsa *lscsa;
+	unsigned char *p;
+
+	if (!csa)
+		return;
+	memset(csa, 0, sizeof(struct spu_state));
+
+	lscsa = vmalloc(sizeof(struct spu_lscsa));
+	if (!lscsa)
+		return;
+
+	memset(lscsa, 0, sizeof(struct spu_lscsa));
+	csa->lscsa = lscsa;
+	csa->register_lock = SPIN_LOCK_UNLOCKED;
+
+	/* Set LS pages reserved to allow for user-space mapping. */
+	for (p = lscsa->ls; p < lscsa->ls + LS_SIZE; p += PAGE_SIZE)
+		SetPageReserved(vmalloc_to_page(p));
+
+	init_prob(csa);
+	init_priv1(csa);
+	init_priv2(csa);
+}
+
+void spu_fini_csa(struct spu_state *csa)
+{
+	/* Clear reserved bit before vfree. */
+	unsigned char *p;
+	for (p = csa->lscsa->ls; p < csa->lscsa->ls + LS_SIZE; p += PAGE_SIZE)
+		ClearPageReserved(vmalloc_to_page(p));
+
+	vfree(csa->lscsa);
+}
diff --git a/arch/powerpc/platforms/cell/spufs/syscalls.c b/arch/powerpc/platforms/cell/spufs/syscalls.c
new file mode 100644
index 000000000000..d549aa7ebea6
--- /dev/null
+++ b/arch/powerpc/platforms/cell/spufs/syscalls.c
@@ -0,0 +1,101 @@
+#include <linux/file.h>
+#include <linux/fs.h>
+#include <linux/module.h>
+#include <linux/mount.h>
+#include <linux/namei.h>
+
+#include <asm/uaccess.h>
+
+#include "spufs.h"
+
+/**
+ * sys_spu_run - run code loaded into an SPU
+ *
+ * @unpc:    next program counter for the SPU
+ * @ustatus: status of the SPU
+ *
+ * This system call transfers the control of execution of a
+ * user space thread to an SPU. It will return when the
+ * SPU has finished executing or when it hits an error
+ * condition and it will be interrupted if a signal needs
+ * to be delivered to a handler in user space.
+ *
+ * The next program counter is set to the passed value
+ * before the SPU starts fetching code and the user space
+ * pointer gets updated with the new value when returning
+ * from kernel space.
+ *
+ * The status value returned from spu_run reflects the
+ * value of the spu_status register after the SPU has stopped.
+ *
+ */
+long do_spu_run(struct file *filp, __u32 __user *unpc, __u32 __user *ustatus)
+{
+	long ret;
+	struct spufs_inode_info *i;
+	u32 npc, status;
+
+	ret = -EFAULT;
+	if (get_user(npc, unpc) || get_user(status, ustatus))
+		goto out;
+
+	/* check if this file was created by spu_create */
+	ret = -EINVAL;
+	if (filp->f_op != &spufs_context_fops)
+		goto out;
+
+	i = SPUFS_I(filp->f_dentry->d_inode);
+	ret = spufs_run_spu(filp, i->i_ctx, &npc, &status);
+
+	if (put_user(npc, unpc) || put_user(status, ustatus))
+		ret = -EFAULT;
+out:
+	return ret;
+}
+
+#ifndef MODULE
+asmlinkage long sys_spu_run(int fd, __u32 __user *unpc, __u32 __user *ustatus)
+{
+	int fput_needed;
+	struct file *filp;
+	long ret;
+
+	ret = -EBADF;
+	filp = fget_light(fd, &fput_needed);
+	if (filp) {
+		ret = do_spu_run(filp, unpc, ustatus);
+		fput_light(filp, fput_needed);
+	}
+
+	return ret;
+}
+#endif
+
+asmlinkage long sys_spu_create(const char __user *pathname,
+					unsigned int flags, mode_t mode)
+{
+	char *tmp;
+	int ret;
+
+	tmp = getname(pathname);
+	ret = PTR_ERR(tmp);
+	if (!IS_ERR(tmp)) {
+		struct nameidata nd;
+
+		ret = path_lookup(tmp, LOOKUP_PARENT|
+				LOOKUP_OPEN|LOOKUP_CREATE, &nd);
+		if (!ret) {
+			ret = spufs_create_thread(&nd, flags, mode);
+			path_release(&nd);
+		}
+		putname(tmp);
+	}
+
+	return ret;
+}
+
+struct spufs_calls spufs_calls = {
+	.create_thread = sys_spu_create,
+	.spu_run = do_spu_run,
+	.owner = THIS_MODULE,
+};
diff --git a/arch/powerpc/platforms/chrp/setup.c b/arch/powerpc/platforms/chrp/setup.c
index dda5f2c72c25..4ec8ba737e7d 100644
--- a/arch/powerpc/platforms/chrp/setup.c
+++ b/arch/powerpc/platforms/chrp/setup.c
@@ -49,7 +49,6 @@
 #include <asm/hydra.h>
 #include <asm/sections.h>
 #include <asm/time.h>
-#include <asm/btext.h>
 #include <asm/i8259.h>
 #include <asm/mpic.h>
 #include <asm/rtas.h>
@@ -58,7 +57,6 @@
 #include "chrp.h"
 
 void rtas_indicator_progress(char *, unsigned short);
-void btext_progress(char *, unsigned short);
 
 int _chrp_type;
 EXPORT_SYMBOL(_chrp_type);
@@ -264,11 +262,6 @@ void __init chrp_setup_arch(void)
 		ppc_md.set_rtc_time	= rtas_set_rtc_time;
 	}
 
-#ifdef CONFIG_BOOTX_TEXT
-	if (ppc_md.progress == NULL && boot_text_mapped)
-		ppc_md.progress = btext_progress;
-#endif
-
 #ifdef CONFIG_BLK_DEV_INITRD
 	/* this is fine for chrp */
 	initrd_below_start_ok = 1;
@@ -522,12 +515,3 @@ void __init chrp_init(void)
 	smp_ops = &chrp_smp_ops;
 #endif /* CONFIG_SMP */
 }
-
-#ifdef CONFIG_BOOTX_TEXT
-void
-btext_progress(char *s, unsigned short hex)
-{
-	btext_drawstring(s);
-	btext_drawstring("\n");
-}
-#endif /* CONFIG_BOOTX_TEXT */
diff --git a/arch/powerpc/platforms/iseries/irq.c b/arch/powerpc/platforms/iseries/irq.c
index a58daa153686..42e978e4897a 100644
--- a/arch/powerpc/platforms/iseries/irq.c
+++ b/arch/powerpc/platforms/iseries/irq.c
@@ -35,161 +35,138 @@
 #include <linux/irq.h>
 #include <linux/spinlock.h>
 
+#include <asm/paca.h>
 #include <asm/iseries/hv_types.h>
 #include <asm/iseries/hv_lp_event.h>
 #include <asm/iseries/hv_call_xm.h>
+#include <asm/iseries/it_lp_queue.h>
 
 #include "irq.h"
 #include "call_pci.h"
 
-static long Pci_Interrupt_Count;
-static long Pci_Event_Count;
-
-enum XmPciLpEvent_Subtype {
-	XmPciLpEvent_BusCreated		= 0,	// PHB has been created
-	XmPciLpEvent_BusError		= 1,	// PHB has failed
-	XmPciLpEvent_BusFailed		= 2,	// Msg to Secondary, Primary failed bus
-	XmPciLpEvent_NodeFailed		= 4,	// Multi-adapter bridge has failed
-	XmPciLpEvent_NodeRecovered	= 5,	// Multi-adapter bridge has recovered
-	XmPciLpEvent_BusRecovered	= 12,	// PHB has been recovered
-	XmPciLpEvent_UnQuiesceBus	= 18,	// Secondary bus unqiescing
-	XmPciLpEvent_BridgeError	= 21,	// Bridge Error
-	XmPciLpEvent_SlotInterrupt	= 22	// Slot interrupt
-};
-
-struct XmPciLpEvent_BusInterrupt {
-	HvBusNumber	busNumber;
-	HvSubBusNumber	subBusNumber;
-};
+#if defined(CONFIG_SMP)
+extern void iSeries_smp_message_recv(struct pt_regs *);
+#endif
 
-struct XmPciLpEvent_NodeInterrupt {
-	HvBusNumber	busNumber;
-	HvSubBusNumber	subBusNumber;
-	HvAgentId	deviceId;
+enum pci_event_type {
+	pe_bus_created		= 0,	/* PHB has been created */
+	pe_bus_error		= 1,	/* PHB has failed */
+	pe_bus_failed		= 2,	/* Msg to Secondary, Primary failed bus */
+	pe_node_failed		= 4,	/* Multi-adapter bridge has failed */
+	pe_node_recovered	= 5,	/* Multi-adapter bridge has recovered */
+	pe_bus_recovered	= 12,	/* PHB has been recovered */
+	pe_unquiese_bus		= 18,	/* Secondary bus unqiescing */
+	pe_bridge_error		= 21,	/* Bridge Error */
+	pe_slot_interrupt	= 22	/* Slot interrupt */
 };
 
-struct XmPciLpEvent {
-	struct HvLpEvent hvLpEvent;
-
+struct pci_event {
+	struct HvLpEvent event;
 	union {
-		u64 alignData;			// Align on an 8-byte boundary
-
+		u64 __align;		/* Align on an 8-byte boundary */
 		struct {
 			u32		fisr;
-			HvBusNumber	busNumber;
-			HvSubBusNumber	subBusNumber;
-			HvAgentId	deviceId;
-		} slotInterrupt;
-
-		struct XmPciLpEvent_BusInterrupt busFailed;
-		struct XmPciLpEvent_BusInterrupt busRecovered;
-		struct XmPciLpEvent_BusInterrupt busCreated;
-
-		struct XmPciLpEvent_NodeInterrupt nodeFailed;
-		struct XmPciLpEvent_NodeInterrupt nodeRecovered;
-
-	} eventData;
-
+			HvBusNumber	bus_number;
+			HvSubBusNumber	sub_bus_number;
+			HvAgentId	dev_id;
+		} slot;
+		struct {
+			HvBusNumber	bus_number;
+			HvSubBusNumber	sub_bus_number;
+		} bus;
+		struct {
+			HvBusNumber	bus_number;
+			HvSubBusNumber	sub_bus_number;
+			HvAgentId	dev_id;
+		} node;
+	} data;
 };
 
-static void intReceived(struct XmPciLpEvent *eventParm,
-		struct pt_regs *regsParm)
+static DEFINE_SPINLOCK(pending_irqs_lock);
+static int num_pending_irqs;
+static int pending_irqs[NR_IRQS];
+
+static void int_received(struct pci_event *event, struct pt_regs *regs)
 {
 	int irq;
-#ifdef CONFIG_IRQSTACKS
-	struct thread_info *curtp, *irqtp;
-#endif
 
-	++Pci_Interrupt_Count;
-
-	switch (eventParm->hvLpEvent.xSubtype) {
-	case XmPciLpEvent_SlotInterrupt:
-		irq = eventParm->hvLpEvent.xCorrelationToken;
-		/* Dispatch the interrupt handlers for this irq */
-#ifdef CONFIG_IRQSTACKS
-		/* Switch to the irq stack to handle this */
-		curtp = current_thread_info();
-		irqtp = hardirq_ctx[smp_processor_id()];
-		if (curtp != irqtp) {
-			irqtp->task = curtp->task;
-			irqtp->flags = 0;
-			call___do_IRQ(irq, regsParm, irqtp);
-			irqtp->task = NULL;
-			if (irqtp->flags)
-				set_bits(irqtp->flags, &curtp->flags);
-		} else
-#endif
-			__do_IRQ(irq, regsParm);
-		HvCallPci_eoi(eventParm->eventData.slotInterrupt.busNumber,
-			eventParm->eventData.slotInterrupt.subBusNumber,
-			eventParm->eventData.slotInterrupt.deviceId);
+	switch (event->event.xSubtype) {
+	case pe_slot_interrupt:
+		irq = event->event.xCorrelationToken;
+		if (irq < NR_IRQS) {
+			spin_lock(&pending_irqs_lock);
+			pending_irqs[irq]++;
+			num_pending_irqs++;
+			spin_unlock(&pending_irqs_lock);
+		} else {
+			printk(KERN_WARNING "int_received: bad irq number %d\n",
+					irq);
+			HvCallPci_eoi(event->data.slot.bus_number,
+					event->data.slot.sub_bus_number,
+					event->data.slot.dev_id);
+		}
 		break;
 		/* Ignore error recovery events for now */
-	case XmPciLpEvent_BusCreated:
-		printk(KERN_INFO "intReceived: system bus %d created\n",
-			eventParm->eventData.busCreated.busNumber);
+	case pe_bus_created:
+		printk(KERN_INFO "int_received: system bus %d created\n",
+			event->data.bus.bus_number);
 		break;
-	case XmPciLpEvent_BusError:
-	case XmPciLpEvent_BusFailed:
-		printk(KERN_INFO "intReceived: system bus %d failed\n",
-			eventParm->eventData.busFailed.busNumber);
+	case pe_bus_error:
+	case pe_bus_failed:
+		printk(KERN_INFO "int_received: system bus %d failed\n",
+			event->data.bus.bus_number);
 		break;
-	case XmPciLpEvent_BusRecovered:
-	case XmPciLpEvent_UnQuiesceBus:
-		printk(KERN_INFO "intReceived: system bus %d recovered\n",
-			eventParm->eventData.busRecovered.busNumber);
+	case pe_bus_recovered:
+	case pe_unquiese_bus:
+		printk(KERN_INFO "int_received: system bus %d recovered\n",
+			event->data.bus.bus_number);
 		break;
-	case XmPciLpEvent_NodeFailed:
-	case XmPciLpEvent_BridgeError:
+	case pe_node_failed:
+	case pe_bridge_error:
 		printk(KERN_INFO
-			"intReceived: multi-adapter bridge %d/%d/%d failed\n",
-			eventParm->eventData.nodeFailed.busNumber,
-			eventParm->eventData.nodeFailed.subBusNumber,
-			eventParm->eventData.nodeFailed.deviceId);
+			"int_received: multi-adapter bridge %d/%d/%d failed\n",
+			event->data.node.bus_number,
+			event->data.node.sub_bus_number,
+			event->data.node.dev_id);
 		break;
-	case XmPciLpEvent_NodeRecovered:
+	case pe_node_recovered:
 		printk(KERN_INFO
-			"intReceived: multi-adapter bridge %d/%d/%d recovered\n",
-			eventParm->eventData.nodeRecovered.busNumber,
-			eventParm->eventData.nodeRecovered.subBusNumber,
-			eventParm->eventData.nodeRecovered.deviceId);
+			"int_received: multi-adapter bridge %d/%d/%d recovered\n",
+			event->data.node.bus_number,
+			event->data.node.sub_bus_number,
+			event->data.node.dev_id);
 		break;
 	default:
 		printk(KERN_ERR
-			"intReceived: unrecognized event subtype 0x%x\n",
-			eventParm->hvLpEvent.xSubtype);
+			"int_received: unrecognized event subtype 0x%x\n",
+			event->event.xSubtype);
 		break;
 	}
 }
 
-static void XmPciLpEvent_handler(struct HvLpEvent *eventParm,
-		struct pt_regs *regsParm)
+static void pci_event_handler(struct HvLpEvent *event, struct pt_regs *regs)
 {
-#ifdef CONFIG_PCI
-	++Pci_Event_Count;
-
-	if (eventParm && (eventParm->xType == HvLpEvent_Type_PciIo)) {
-		switch (eventParm->xFlags.xFunction) {
+	if (event && (event->xType == HvLpEvent_Type_PciIo)) {
+		switch (event->xFlags.xFunction) {
 		case HvLpEvent_Function_Int:
-			intReceived((struct XmPciLpEvent *)eventParm, regsParm);
+			int_received((struct pci_event *)event, regs);
 			break;
 		case HvLpEvent_Function_Ack:
 			printk(KERN_ERR
-				"XmPciLpEvent_handler: unexpected ack received\n");
+				"pci_event_handler: unexpected ack received\n");
 			break;
 		default:
 			printk(KERN_ERR
-				"XmPciLpEvent_handler: unexpected event function %d\n",
-				(int)eventParm->xFlags.xFunction);
+				"pci_event_handler: unexpected event function %d\n",
+				(int)event->xFlags.xFunction);
 			break;
 		}
-	} else if (eventParm)
+	} else if (event)
 		printk(KERN_ERR
-			"XmPciLpEvent_handler: Unrecognized PCI event type 0x%x\n",
-			(int)eventParm->xType);
+			"pci_event_handler: Unrecognized PCI event type 0x%x\n",
+			(int)event->xType);
 	else
-		printk(KERN_ERR "XmPciLpEvent_handler: NULL event received\n");
-#endif
+		printk(KERN_ERR "pci_event_handler: NULL event received\n");
 }
 
 /*
@@ -199,20 +176,21 @@ static void XmPciLpEvent_handler(struct HvLpEvent *eventParm,
 void __init iSeries_init_IRQ(void)
 {
 	/* Register PCI event handler and open an event path */
-	int xRc;
-
-	xRc = HvLpEvent_registerHandler(HvLpEvent_Type_PciIo,
-			&XmPciLpEvent_handler);
-	if (xRc == 0) {
-		xRc = HvLpEvent_openPath(HvLpEvent_Type_PciIo, 0);
-		if (xRc != 0)
-			printk(KERN_ERR "iSeries_init_IRQ: open event path "
-					"failed with rc 0x%x\n", xRc);
+	int ret;
+
+	ret = HvLpEvent_registerHandler(HvLpEvent_Type_PciIo,
+			&pci_event_handler);
+	if (ret == 0) {
+		ret = HvLpEvent_openPath(HvLpEvent_Type_PciIo, 0);
+		if (ret != 0)
+			printk(KERN_ERR "iseries_init_IRQ: open event path "
+					"failed with rc 0x%x\n", ret);
 	} else
-		printk(KERN_ERR "iSeries_init_IRQ: register handler "
-				"failed with rc 0x%x\n", xRc);
+		printk(KERN_ERR "iseries_init_IRQ: register handler "
+				"failed with rc 0x%x\n", ret);
 }
 
+#define REAL_IRQ_TO_SUBBUS(irq)	(((irq) >> 14) & 0xff)
 #define REAL_IRQ_TO_BUS(irq)	((((irq) >> 6) & 0xff) + 1)
 #define REAL_IRQ_TO_IDSEL(irq)	((((irq) >> 3) & 7) + 1)
 #define REAL_IRQ_TO_FUNC(irq)	((irq) & 7)
@@ -221,40 +199,40 @@ void __init iSeries_init_IRQ(void)
  * This will be called by device drivers (via enable_IRQ)
  * to enable INTA in the bridge interrupt status register.
  */
-static void iSeries_enable_IRQ(unsigned int irq)
+static void iseries_enable_IRQ(unsigned int irq)
 {
-	u32 bus, deviceId, function, mask;
-	const u32 subBus = 0;
+	u32 bus, dev_id, function, mask;
+	const u32 sub_bus = 0;
 	unsigned int rirq = virt_irq_to_real_map[irq];
 
 	/* The IRQ has already been locked by the caller */
 	bus = REAL_IRQ_TO_BUS(rirq);
 	function = REAL_IRQ_TO_FUNC(rirq);
-	deviceId = (REAL_IRQ_TO_IDSEL(rirq) << 4) + function;
+	dev_id = (REAL_IRQ_TO_IDSEL(rirq) << 4) + function;
 
 	/* Unmask secondary INTA */
 	mask = 0x80000000;
-	HvCallPci_unmaskInterrupts(bus, subBus, deviceId, mask);
+	HvCallPci_unmaskInterrupts(bus, sub_bus, dev_id, mask);
 }
 
-/* This is called by iSeries_activate_IRQs */
-static unsigned int iSeries_startup_IRQ(unsigned int irq)
+/* This is called by iseries_activate_IRQs */
+static unsigned int iseries_startup_IRQ(unsigned int irq)
 {
-	u32 bus, deviceId, function, mask;
-	const u32 subBus = 0;
+	u32 bus, dev_id, function, mask;
+	const u32 sub_bus = 0;
 	unsigned int rirq = virt_irq_to_real_map[irq];
 
 	bus = REAL_IRQ_TO_BUS(rirq);
 	function = REAL_IRQ_TO_FUNC(rirq);
-	deviceId = (REAL_IRQ_TO_IDSEL(rirq) << 4) + function;
+	dev_id = (REAL_IRQ_TO_IDSEL(rirq) << 4) + function;
 
 	/* Link the IRQ number to the bridge */
-	HvCallXm_connectBusUnit(bus, subBus, deviceId, irq);
+	HvCallXm_connectBusUnit(bus, sub_bus, dev_id, irq);
 
 	/* Unmask bridge interrupts in the FISR */
 	mask = 0x01010000 << function;
-	HvCallPci_unmaskFisr(bus, subBus, deviceId, mask);
-	iSeries_enable_IRQ(irq);
+	HvCallPci_unmaskFisr(bus, sub_bus, dev_id, mask);
+	iseries_enable_IRQ(irq);
 	return 0;
 }
 
@@ -279,78 +257,115 @@ void __init iSeries_activate_IRQs()
 }
 
 /*  this is not called anywhere currently */
-static void iSeries_shutdown_IRQ(unsigned int irq)
+static void iseries_shutdown_IRQ(unsigned int irq)
 {
-	u32 bus, deviceId, function, mask;
-	const u32 subBus = 0;
+	u32 bus, dev_id, function, mask;
+	const u32 sub_bus = 0;
 	unsigned int rirq = virt_irq_to_real_map[irq];
 
 	/* irq should be locked by the caller */
 	bus = REAL_IRQ_TO_BUS(rirq);
 	function = REAL_IRQ_TO_FUNC(rirq);
-	deviceId = (REAL_IRQ_TO_IDSEL(rirq) << 4) + function;
+	dev_id = (REAL_IRQ_TO_IDSEL(rirq) << 4) + function;
 
 	/* Invalidate the IRQ number in the bridge */
-	HvCallXm_connectBusUnit(bus, subBus, deviceId, 0);
+	HvCallXm_connectBusUnit(bus, sub_bus, dev_id, 0);
 
 	/* Mask bridge interrupts in the FISR */
 	mask = 0x01010000 << function;
-	HvCallPci_maskFisr(bus, subBus, deviceId, mask);
+	HvCallPci_maskFisr(bus, sub_bus, dev_id, mask);
 }
 
 /*
  * This will be called by device drivers (via disable_IRQ)
  * to disable INTA in the bridge interrupt status register.
  */
-static void iSeries_disable_IRQ(unsigned int irq)
+static void iseries_disable_IRQ(unsigned int irq)
 {
-	u32 bus, deviceId, function, mask;
-	const u32 subBus = 0;
+	u32 bus, dev_id, function, mask;
+	const u32 sub_bus = 0;
 	unsigned int rirq = virt_irq_to_real_map[irq];
 
 	/* The IRQ has already been locked by the caller */
 	bus = REAL_IRQ_TO_BUS(rirq);
 	function = REAL_IRQ_TO_FUNC(rirq);
-	deviceId = (REAL_IRQ_TO_IDSEL(rirq) << 4) + function;
+	dev_id = (REAL_IRQ_TO_IDSEL(rirq) << 4) + function;
 
 	/* Mask secondary INTA   */
 	mask = 0x80000000;
-	HvCallPci_maskInterrupts(bus, subBus, deviceId, mask);
+	HvCallPci_maskInterrupts(bus, sub_bus, dev_id, mask);
 }
 
-/*
- * This does nothing because there is not enough information
- * provided to do the EOI HvCall.  This is done by XmPciLpEvent.c
- */
-static void iSeries_end_IRQ(unsigned int irq)
+static void iseries_end_IRQ(unsigned int irq)
 {
+	unsigned int rirq = virt_irq_to_real_map[irq];
+
+	HvCallPci_eoi(REAL_IRQ_TO_BUS(rirq), REAL_IRQ_TO_SUBBUS(rirq),
+		(REAL_IRQ_TO_IDSEL(rirq) << 4) + REAL_IRQ_TO_FUNC(rirq));
 }
 
 static hw_irq_controller iSeries_IRQ_handler = {
 	.typename = "iSeries irq controller",
-	.startup = iSeries_startup_IRQ,
-	.shutdown = iSeries_shutdown_IRQ,
-	.enable = iSeries_enable_IRQ,
-	.disable = iSeries_disable_IRQ,
-	.end = iSeries_end_IRQ
+	.startup = iseries_startup_IRQ,
+	.shutdown = iseries_shutdown_IRQ,
+	.enable = iseries_enable_IRQ,
+	.disable = iseries_disable_IRQ,
+	.end = iseries_end_IRQ
 };
 
 /*
  * This is called out of iSeries_scan_slot to allocate an IRQ for an EADS slot
  * It calculates the irq value for the slot.
- * Note that subBusNumber is always 0 (at the moment at least).
+ * Note that sub_bus is always 0 (at the moment at least).
  */
-int __init iSeries_allocate_IRQ(HvBusNumber busNumber,
-		HvSubBusNumber subBusNumber, HvAgentId deviceId)
+int __init iSeries_allocate_IRQ(HvBusNumber bus,
+		HvSubBusNumber sub_bus, HvAgentId dev_id)
 {
 	int virtirq;
 	unsigned int realirq;
-	u8 idsel = (deviceId >> 4);
-	u8 function = deviceId & 7;
+	u8 idsel = (dev_id >> 4);
+	u8 function = dev_id & 7;
 
-	realirq = ((busNumber - 1) << 6) + ((idsel - 1) << 3) + function;
+	realirq = (((((sub_bus << 8) + (bus - 1)) << 3) + (idsel - 1)) << 3)
+		+ function;
 	virtirq = virt_irq_create_mapping(realirq);
 
 	irq_desc[virtirq].handler = &iSeries_IRQ_handler;
 	return virtirq;
 }
+
+/*
+ * Get the next pending IRQ.
+ */
+int iSeries_get_irq(struct pt_regs *regs)
+{
+	struct paca_struct *lpaca;
+	/* -2 means ignore this interrupt */
+	int irq = -2;
+
+	lpaca = get_paca();
+#ifdef CONFIG_SMP
+	if (lpaca->lppaca.int_dword.fields.ipi_cnt) {
+		lpaca->lppaca.int_dword.fields.ipi_cnt = 0;
+		iSeries_smp_message_recv(regs);
+	}
+#endif /* CONFIG_SMP */
+	if (hvlpevent_is_pending())
+		process_hvlpevents(regs);
+
+	if (num_pending_irqs) {
+		spin_lock(&pending_irqs_lock);
+		for (irq = 0; irq < NR_IRQS; irq++) {
+			if (pending_irqs[irq]) {
+				pending_irqs[irq]--;
+				num_pending_irqs--;
+				break;
+			}
+		}
+		spin_unlock(&pending_irqs_lock);
+		if (irq >= NR_IRQS)
+			irq = -2;
+	}
+
+	return irq;
+}
diff --git a/arch/powerpc/platforms/iseries/irq.h b/arch/powerpc/platforms/iseries/irq.h
index 5f643f16ecc0..b9c801ba5a47 100644
--- a/arch/powerpc/platforms/iseries/irq.h
+++ b/arch/powerpc/platforms/iseries/irq.h
@@ -4,5 +4,6 @@
 extern void iSeries_init_IRQ(void);
 extern int  iSeries_allocate_IRQ(HvBusNumber, HvSubBusNumber, HvAgentId);
 extern void iSeries_activate_IRQs(void);
+extern int iSeries_get_irq(struct pt_regs *);
 
 #endif /* _ISERIES_IRQ_H */
diff --git a/arch/powerpc/platforms/iseries/lpardata.c b/arch/powerpc/platforms/iseries/lpardata.c
index bb8c91537f35..ea72385aaf0a 100644
--- a/arch/powerpc/platforms/iseries/lpardata.c
+++ b/arch/powerpc/platforms/iseries/lpardata.c
@@ -225,3 +225,10 @@ struct ItVpdAreas itVpdAreas = {
 		0,0
 	}
 };
+
+struct ItLpRegSave iseries_reg_save[] = {
+	[0 ... (NR_CPUS-1)] = {
+		.xDesc = 0xd397d9e2,	/* "LpRS" */
+		.xSize = sizeof(struct ItLpRegSave),
+	},
+};
diff --git a/arch/powerpc/platforms/iseries/setup.c b/arch/powerpc/platforms/iseries/setup.c
index da26639190db..ad5ef80500ce 100644
--- a/arch/powerpc/platforms/iseries/setup.c
+++ b/arch/powerpc/platforms/iseries/setup.c
@@ -571,16 +571,6 @@ static void iSeries_show_cpuinfo(struct seq_file *m)
 
 /*
  * Document me.
- * and Implement me.
- */
-static int iSeries_get_irq(struct pt_regs *regs)
-{
-	/* -2 means ignore this interrupt */
-	return -2;
-}
-
-/*
- * Document me.
  */
 static void iSeries_restart(char *cmd)
 {
diff --git a/arch/powerpc/platforms/maple/setup.c b/arch/powerpc/platforms/maple/setup.c
index 7ece8983a105..dd73e38bfb7d 100644
--- a/arch/powerpc/platforms/maple/setup.c
+++ b/arch/powerpc/platforms/maple/setup.c
@@ -51,6 +51,7 @@
 #include <asm/pgtable.h>
 #include <asm/bitops.h>
 #include <asm/io.h>
+#include <asm/kexec.h>
 #include <asm/pci-bridge.h>
 #include <asm/iommu.h>
 #include <asm/machdep.h>
@@ -191,24 +192,10 @@ static void __init maple_init_early(void)
 	 */
 	hpte_init_native();
 
-	/* Find the serial port */
-	generic_find_legacy_serial_ports(&physport, &default_speed);
-
-	DBG("phys port addr: %lx\n", (long)physport);
-
-	if (physport) {
-		void *comport;
-		/* Map the uart for udbg. */
-		comport = (void *)ioremap(physport, 16);
-		udbg_init_uart(comport, default_speed);
-
-		DBG("Hello World !\n");
-	}
-
 	/* Setup interrupt mapping options */
 	ppc64_interrupt_controller = IC_OPEN_PIC;
 
-	iommu_init_early_u3();
+	iommu_init_early_dart();
 
 	DBG(" <- maple_init_early\n");
 }
@@ -270,7 +257,7 @@ static int __init maple_probe(int platform)
 	 * occupies having to be broken up so the DART itself is not
 	 * part of the cacheable linar mapping
 	 */
-	alloc_u3_dart_table();
+	alloc_dart_table();
 
 	return 1;
 }
@@ -292,4 +279,9 @@ struct machdep_calls __initdata maple_md = {
       	.calibrate_decr		= generic_calibrate_decr,
 	.progress		= maple_progress,
 	.idle_loop		= native_idle,
+#ifdef CONFIG_KEXEC
+	.machine_kexec		= default_machine_kexec,
+	.machine_kexec_prepare	= default_machine_kexec_prepare,
+	.machine_crash_shutdown	= default_machine_crash_shutdown,
+#endif
 };
diff --git a/arch/powerpc/platforms/powermac/Makefile b/arch/powerpc/platforms/powermac/Makefile
index c9df44fcf571..78093d7f97af 100644
--- a/arch/powerpc/platforms/powermac/Makefile
+++ b/arch/powerpc/platforms/powermac/Makefile
@@ -1,9 +1,14 @@
+CFLAGS_bootx_init.o  		+= -fPIC
+
 obj-y				+= pic.o setup.o time.o feature.o pci.o \
-				   sleep.o low_i2c.o cache.o
+				   sleep.o low_i2c.o cache.o pfunc_core.o \
+				   pfunc_base.o
 obj-$(CONFIG_PMAC_BACKLIGHT)	+= backlight.o
 obj-$(CONFIG_CPU_FREQ_PMAC)	+= cpufreq_32.o
 obj-$(CONFIG_CPU_FREQ_PMAC64)	+= cpufreq_64.o
 obj-$(CONFIG_NVRAM)		+= nvram.o
 # ppc64 pmac doesn't define CONFIG_NVRAM but needs nvram stuff
 obj-$(CONFIG_PPC64)		+= nvram.o
+obj-$(CONFIG_PPC32)		+= bootx_init.o
 obj-$(CONFIG_SMP)		+= smp.o
+obj-$(CONFIG_PPC_MERGE)		+= udbg_scc.o udbg_adb.o
diff --git a/arch/powerpc/platforms/powermac/bootx_init.c b/arch/powerpc/platforms/powermac/bootx_init.c
new file mode 100644
index 000000000000..fa8b4d7b5ded
--- /dev/null
+++ b/arch/powerpc/platforms/powermac/bootx_init.c
@@ -0,0 +1,547 @@
+/*
+ *  Early boot support code for BootX bootloader
+ *
+ *  Copyright (C) 2005 Ben. Herrenschmidt (benh@kernel.crashing.org)
+ *
+ *  This program is free software; you can redistribute it and/or
+ *  modify it under the terms of the GNU General Public License
+ *  as published by the Free Software Foundation; either version
+ *  2 of the License, or (at your option) any later version.
+ */
+
+#include <linux/config.h>
+#include <linux/kernel.h>
+#include <linux/string.h>
+#include <linux/init.h>
+#include <linux/version.h>
+#include <asm/sections.h>
+#include <asm/prom.h>
+#include <asm/page.h>
+#include <asm/bootx.h>
+#include <asm/bootinfo.h>
+#include <asm/btext.h>
+#include <asm/io.h>
+
+#undef DEBUG
+#define SET_BOOT_BAT
+
+#ifdef DEBUG
+#define DBG(fmt...) do { bootx_printf(fmt); } while(0)
+#else
+#define DBG(fmt...) do { } while(0)
+#endif
+
+extern void __start(unsigned long r3, unsigned long r4, unsigned long r5);
+
+static unsigned long __initdata bootx_dt_strbase;
+static unsigned long __initdata bootx_dt_strend;
+static unsigned long __initdata bootx_node_chosen;
+static boot_infos_t * __initdata bootx_info;
+static char __initdata bootx_disp_path[256];
+
+/* Is boot-info compatible ? */
+#define BOOT_INFO_IS_COMPATIBLE(bi) \
+	((bi)->compatible_version <= BOOT_INFO_VERSION)
+#define BOOT_INFO_IS_V2_COMPATIBLE(bi)	((bi)->version >= 2)
+#define BOOT_INFO_IS_V4_COMPATIBLE(bi)	((bi)->version >= 4)
+
+#ifdef CONFIG_BOOTX_TEXT
+static void __init bootx_printf(const char *format, ...)
+{
+	const char *p, *q, *s;
+	va_list args;
+	unsigned long v;
+
+	va_start(args, format);
+	for (p = format; *p != 0; p = q) {
+		for (q = p; *q != 0 && *q != '\n' && *q != '%'; ++q)
+			;
+		if (q > p)
+			btext_drawtext(p, q - p);
+		if (*q == 0)
+			break;
+		if (*q == '\n') {
+			++q;
+			btext_flushline();
+			btext_drawstring("\r\n");
+			btext_flushline();
+			continue;
+		}
+		++q;
+		if (*q == 0)
+			break;
+		switch (*q) {
+		case 's':
+			++q;
+			s = va_arg(args, const char *);
+			if (s == NULL)
+				s = "<NULL>";
+			btext_drawstring(s);
+			break;
+		case 'x':
+			++q;
+			v = va_arg(args, unsigned long);
+			btext_drawhex(v);
+			break;
+		}
+	}
+}
+#else /* CONFIG_BOOTX_TEXT */
+static void __init bootx_printf(const char *format, ...) {}
+#endif /* CONFIG_BOOTX_TEXT */
+
+static void * __init bootx_early_getprop(unsigned long base,
+					 unsigned long node,
+					 char *prop)
+{
+	struct bootx_dt_node *np = (struct bootx_dt_node *)(base + node);
+	u32 *ppp = &np->properties;
+
+	while(*ppp) {
+		struct bootx_dt_prop *pp =
+			(struct bootx_dt_prop *)(base + *ppp);
+
+		if (strcmp((char *)((unsigned long)pp->name + base),
+			   prop) == 0) {
+			return (void *)((unsigned long)pp->value + base);
+		}
+		ppp = &pp->next;
+	}
+	return NULL;
+}
+
+#define dt_push_token(token, mem) \
+	do { \
+		*(mem) = _ALIGN_UP(*(mem),4); \
+		*((u32 *)*(mem)) = token; \
+		*(mem) += 4; \
+	} while(0)
+
+static unsigned long __init bootx_dt_find_string(char *str)
+{
+	char *s, *os;
+
+	s = os = (char *)bootx_dt_strbase;
+	s += 4;
+	while (s <  (char *)bootx_dt_strend) {
+		if (strcmp(s, str) == 0)
+			return s - os;
+		s += strlen(s) + 1;
+	}
+	return 0;
+}
+
+static void __init bootx_dt_add_prop(char *name, void *data, int size,
+				  unsigned long *mem_end)
+{
+	unsigned long soff = bootx_dt_find_string(name);
+	if (data == NULL)
+		size = 0;
+	if (soff == 0) {
+		bootx_printf("WARNING: Can't find string index for <%s>\n",
+			     name);
+		return;
+	}
+	if (size > 0x20000) {
+		bootx_printf("WARNING: ignoring large property ");
+		bootx_printf("%s length 0x%x\n", name, size);
+		return;
+	}
+	dt_push_token(OF_DT_PROP, mem_end);
+	dt_push_token(size, mem_end);
+	dt_push_token(soff, mem_end);
+
+	/* push property content */
+	if (size && data) {
+		memcpy((void *)*mem_end, data, size);
+		*mem_end = _ALIGN_UP(*mem_end + size, 4);
+	}
+}
+
+static void __init bootx_add_chosen_props(unsigned long base,
+					  unsigned long *mem_end)
+{
+	u32 val = _MACH_Pmac;
+
+	bootx_dt_add_prop("linux,platform", &val, 4, mem_end);
+
+	if (bootx_info->kernelParamsOffset) {
+		char *args = (char *)((unsigned long)bootx_info) +
+			bootx_info->kernelParamsOffset;
+		bootx_dt_add_prop("bootargs", args, strlen(args) + 1, mem_end);
+	}
+	if (bootx_info->ramDisk) {
+		val = ((unsigned long)bootx_info) + bootx_info->ramDisk;
+		bootx_dt_add_prop("linux,initrd-start", &val, 4, mem_end);
+		val += bootx_info->ramDiskSize;
+		bootx_dt_add_prop("linux,initrd-end", &val, 4, mem_end);
+	}
+	if (strlen(bootx_disp_path))
+		bootx_dt_add_prop("linux,stdout-path", bootx_disp_path,
+				  strlen(bootx_disp_path) + 1, mem_end);
+}
+
+static void __init bootx_add_display_props(unsigned long base,
+					   unsigned long *mem_end)
+{
+	bootx_dt_add_prop("linux,boot-display", NULL, 0, mem_end);
+	bootx_dt_add_prop("linux,opened", NULL, 0, mem_end);
+}
+
+static void __init bootx_dt_add_string(char *s, unsigned long *mem_end)
+{
+	unsigned int l = strlen(s) + 1;
+	memcpy((void *)*mem_end, s, l);
+	bootx_dt_strend = *mem_end = *mem_end + l;
+}
+
+static void __init bootx_scan_dt_build_strings(unsigned long base,
+					       unsigned long node,
+					       unsigned long *mem_end)
+{
+	struct bootx_dt_node *np = (struct bootx_dt_node *)(base + node);
+	u32 *cpp, *ppp = &np->properties;
+	unsigned long soff;
+	char *namep;
+
+	/* Keep refs to known nodes */
+	namep = np->full_name ? (char *)(base + np->full_name) : NULL;
+       	if (namep == NULL) {
+		bootx_printf("Node without a full name !\n");
+		namep = "";
+	}
+	DBG("* strings: %s\n", namep);
+
+	if (!strcmp(namep, "/chosen")) {
+		DBG(" detected /chosen ! adding properties names !\n");
+		bootx_dt_add_string("linux,platform", mem_end);
+		bootx_dt_add_string("linux,stdout-path", mem_end);
+		bootx_dt_add_string("linux,initrd-start", mem_end);
+		bootx_dt_add_string("linux,initrd-end", mem_end);
+		bootx_dt_add_string("bootargs", mem_end);
+		bootx_node_chosen = node;
+	}
+	if (node == bootx_info->dispDeviceRegEntryOffset) {
+		DBG(" detected display ! adding properties names !\n");
+		bootx_dt_add_string("linux,boot-display", mem_end);
+		bootx_dt_add_string("linux,opened", mem_end);
+		strncpy(bootx_disp_path, namep, 255);
+	}
+
+	/* get and store all property names */
+	while (*ppp) {
+		struct bootx_dt_prop *pp =
+			(struct bootx_dt_prop *)(base + *ppp);
+
+		namep = pp->name ? (char *)(base + pp->name) : NULL;
+ 		if (namep == NULL || strcmp(namep, "name") == 0)
+ 			goto next;
+		/* get/create string entry */
+		soff = bootx_dt_find_string(namep);
+		if (soff == 0)
+			bootx_dt_add_string(namep, mem_end);
+	next:
+		ppp = &pp->next;
+	}
+
+	/* do all our children */
+	cpp = &np->child;
+	while(*cpp) {
+		np = (struct bootx_dt_node *)(base + *cpp);
+		bootx_scan_dt_build_strings(base, *cpp, mem_end);
+		cpp = &np->sibling;
+	}
+}
+
+static void __init bootx_scan_dt_build_struct(unsigned long base,
+					      unsigned long node,
+					      unsigned long *mem_end)
+{
+	struct bootx_dt_node *np = (struct bootx_dt_node *)(base + node);
+	u32 *cpp, *ppp = &np->properties;
+	char *namep, *p, *ep, *lp;
+	int l;
+
+	dt_push_token(OF_DT_BEGIN_NODE, mem_end);
+
+	/* get the node's full name */
+	namep = np->full_name ? (char *)(base + np->full_name) : NULL;
+	if (namep == NULL)
+		namep = "";
+	l = strlen(namep);
+
+	DBG("* struct: %s\n", namep);
+
+	/* Fixup an Apple bug where they have bogus \0 chars in the
+	 * middle of the path in some properties, and extract
+	 * the unit name (everything after the last '/').
+	 */
+	memcpy((void *)*mem_end, namep, l + 1);
+	namep = (char *)*mem_end;
+	for (lp = p = namep, ep = namep + l; p < ep; p++) {
+		if (*p == '/')
+			lp = namep;
+		else if (*p != 0)
+			*lp++ = *p;
+	}
+	*lp = 0;
+	*mem_end = _ALIGN_UP((unsigned long)lp + 1, 4);
+
+	/* get and store all properties */
+	while (*ppp) {
+		struct bootx_dt_prop *pp =
+			(struct bootx_dt_prop *)(base + *ppp);
+
+		namep = pp->name ? (char *)(base + pp->name) : NULL;
+		/* Skip "name" */
+ 		if (namep == NULL || !strcmp(namep, "name"))
+ 			goto next;
+		/* Skip "bootargs" in /chosen too as we replace it */
+		if (node == bootx_node_chosen && !strcmp(namep, "bootargs"))
+			goto next;
+
+		/* push property head */
+		bootx_dt_add_prop(namep,
+				  pp->value ? (void *)(base + pp->value): NULL,
+				  pp->length, mem_end);
+	next:
+		ppp = &pp->next;
+	}
+
+	if (node == bootx_node_chosen)
+		bootx_add_chosen_props(base, mem_end);
+	if (node == bootx_info->dispDeviceRegEntryOffset)
+		bootx_add_display_props(base, mem_end);
+
+	/* do all our children */
+	cpp = &np->child;
+	while(*cpp) {
+		np = (struct bootx_dt_node *)(base + *cpp);
+		bootx_scan_dt_build_struct(base, *cpp, mem_end);
+		cpp = &np->sibling;
+	}
+
+	dt_push_token(OF_DT_END_NODE, mem_end);
+}
+
+static unsigned long __init bootx_flatten_dt(unsigned long start)
+{
+	boot_infos_t *bi = bootx_info;
+	unsigned long mem_start, mem_end;
+	struct boot_param_header *hdr;
+	unsigned long base;
+	u64 *rsvmap;
+
+	/* Start using memory after the big blob passed by BootX, get
+	 * some space for the header
+	 */
+	mem_start = mem_end = _ALIGN_UP(((unsigned long)bi) + start, 4);
+	DBG("Boot params header at: %x\n", mem_start);
+	hdr = (struct boot_param_header *)mem_start;
+	mem_end += sizeof(struct boot_param_header);
+	rsvmap = (u64 *)(_ALIGN_UP(mem_end, 8));
+	hdr->off_mem_rsvmap = ((unsigned long)rsvmap) - mem_start;
+	mem_end = ((unsigned long)rsvmap) + 8 * sizeof(u64);
+
+	/* Get base of tree */
+	base = ((unsigned long)bi) + bi->deviceTreeOffset;
+
+	/* Build string array */
+	DBG("Building string array at: %x\n", mem_end);
+	DBG("Device Tree Base=%x\n", base);
+	bootx_dt_strbase = mem_end;
+	mem_end += 4;
+	bootx_dt_strend = mem_end;
+	bootx_scan_dt_build_strings(base, 4, &mem_end);
+	hdr->off_dt_strings = bootx_dt_strbase - mem_start;
+	hdr->dt_strings_size = bootx_dt_strend - bootx_dt_strbase;
+
+	/* Build structure */
+	mem_end = _ALIGN(mem_end, 16);
+	DBG("Building device tree structure at: %x\n", mem_end);
+	hdr->off_dt_struct = mem_end - mem_start;
+	bootx_scan_dt_build_struct(base, 4, &mem_end);
+	dt_push_token(OF_DT_END, &mem_end);
+
+	/* Finish header */
+	hdr->boot_cpuid_phys = 0;
+	hdr->magic = OF_DT_HEADER;
+	hdr->totalsize = mem_end - mem_start;
+	hdr->version = OF_DT_VERSION;
+	/* Version 16 is not backward compatible */
+	hdr->last_comp_version = 0x10;
+
+	/* Reserve the whole thing and copy the reserve map in, we
+	 * also bump mem_reserve_cnt to cause further reservations to
+	 * fail since it's too late.
+	 */
+	mem_end = _ALIGN(mem_end, PAGE_SIZE);
+	DBG("End of boot params: %x\n", mem_end);
+	rsvmap[0] = mem_start;
+	rsvmap[1] = mem_end;
+	rsvmap[2] = 0;
+	rsvmap[3] = 0;
+
+	return (unsigned long)hdr;
+}
+
+
+#ifdef CONFIG_BOOTX_TEXT
+static void __init btext_welcome(boot_infos_t *bi)
+{
+	unsigned long flags;
+	unsigned long pvr;
+
+	bootx_printf("Welcome to Linux, kernel " UTS_RELEASE "\n");
+	bootx_printf("\nlinked at        : 0x%x", KERNELBASE);
+	bootx_printf("\nframe buffer at  : 0x%x", bi->dispDeviceBase);
+	bootx_printf(" (phys), 0x%x", bi->logicalDisplayBase);
+	bootx_printf(" (log)");
+	bootx_printf("\nklimit           : 0x%x",(unsigned long)klimit);
+	bootx_printf("\nboot_info at     : 0x%x", bi);
+	__asm__ __volatile__ ("mfmsr %0" : "=r" (flags));
+	bootx_printf("\nMSR              : 0x%x", flags);
+	__asm__ __volatile__ ("mfspr %0, 287" : "=r" (pvr));
+	bootx_printf("\nPVR              : 0x%x", pvr);
+	pvr >>= 16;
+	if (pvr > 1) {
+	    __asm__ __volatile__ ("mfspr %0, 1008" : "=r" (flags));
+	    bootx_printf("\nHID0             : 0x%x", flags);
+	}
+	if (pvr == 8 || pvr == 12 || pvr == 0x800c) {
+	    __asm__ __volatile__ ("mfspr %0, 1019" : "=r" (flags));
+	    bootx_printf("\nICTC             : 0x%x", flags);
+	}
+#ifdef DEBUG
+	bootx_printf("\n\n");
+	bootx_printf("bi->deviceTreeOffset   : 0x%x\n",
+		     bi->deviceTreeOffset);
+	bootx_printf("bi->deviceTreeSize     : 0x%x\n",
+		     bi->deviceTreeSize);
+#endif
+	bootx_printf("\n\n");
+}
+#endif /* CONFIG_BOOTX_TEXT */
+
+void __init bootx_init(unsigned long r3, unsigned long r4)
+{
+	boot_infos_t *bi = (boot_infos_t *) r4;
+	unsigned long hdr;
+	unsigned long space;
+	unsigned long ptr, x;
+	char *model;
+	unsigned long offset = reloc_offset();
+
+	reloc_got2(offset);
+
+	bootx_info = bi;
+
+	/* We haven't cleared any bss at this point, make sure
+	 * what we need is initialized
+	 */
+	bootx_dt_strbase = bootx_dt_strend = 0;
+	bootx_node_chosen = 0;
+	bootx_disp_path[0] = 0;
+
+	if (!BOOT_INFO_IS_V2_COMPATIBLE(bi))
+		bi->logicalDisplayBase = bi->dispDeviceBase;
+
+#ifdef CONFIG_BOOTX_TEXT
+	btext_setup_display(bi->dispDeviceRect[2] - bi->dispDeviceRect[0],
+			    bi->dispDeviceRect[3] - bi->dispDeviceRect[1],
+			    bi->dispDeviceDepth, bi->dispDeviceRowBytes,
+			    (unsigned long)bi->logicalDisplayBase);
+	btext_clearscreen();
+	btext_flushscreen();
+#endif /* CONFIG_BOOTX_TEXT */
+
+	/*
+	 * Test if boot-info is compatible.  Done only in config
+	 * CONFIG_BOOTX_TEXT since there is nothing much we can do
+	 * with an incompatible version, except display a message
+	 * and eventually hang the processor...
+	 *
+	 * I'll try to keep enough of boot-info compatible in the
+	 * future to always allow display of this message;
+	 */
+	if (!BOOT_INFO_IS_COMPATIBLE(bi)) {
+		bootx_printf(" !!! WARNING - Incompatible version"
+			     " of BootX !!!\n\n\n");
+		for (;;)
+			;
+	}
+	if (bi->architecture != BOOT_ARCH_PCI) {
+		bootx_printf(" !!! WARNING - Usupported machine"
+			     " architecture !\n");
+		for (;;)
+			;
+	}
+
+#ifdef CONFIG_BOOTX_TEXT
+	btext_welcome(bi);
+#endif
+	/* New BootX enters kernel with MMU off, i/os are not allowed
+	 * here. This hack will have been done by the boostrap anyway.
+	 */
+	if (bi->version < 4) {
+		/*
+		 * XXX If this is an iMac, turn off the USB controller.
+		 */
+		model = (char *) bootx_early_getprop(r4 + bi->deviceTreeOffset,
+						     4, "model");
+		if (model
+		    && (strcmp(model, "iMac,1") == 0
+			|| strcmp(model, "PowerMac1,1") == 0)) {
+			bootx_printf("iMac,1 detected, shutting down USB \n");
+			out_le32((unsigned *)0x80880008, 1);	/* XXX */
+		}
+	}
+
+	/* Get a pointer that points above the device tree, args, ramdisk,
+	 * etc... to use for generating the flattened tree
+	 */
+	if (bi->version < 5) {
+		space = bi->deviceTreeOffset + bi->deviceTreeSize;
+		if (bi->ramDisk)
+			space = bi->ramDisk + bi->ramDiskSize;
+	} else
+		space = bi->totalParamsSize;
+
+	bootx_printf("Total space used by parameters & ramdisk: %x \n", space);
+
+	/* New BootX will have flushed all TLBs and enters kernel with
+	 * MMU switched OFF, so this should not be useful anymore.
+	 */
+	if (bi->version < 4) {
+		bootx_printf("Touching pages...\n");
+
+		/*
+		 * Touch each page to make sure the PTEs for them
+		 * are in the hash table - the aim is to try to avoid
+		 * getting DSI exceptions while copying the kernel image.
+		 */
+		for (ptr = ((unsigned long) &_stext) & PAGE_MASK;
+		     ptr < (unsigned long)bi + space; ptr += PAGE_SIZE)
+			x = *(volatile unsigned long *)ptr;
+	}
+
+	/* Ok, now we need to generate a flattened device-tree to pass
+	 * to the kernel
+	 */
+	bootx_printf("Preparing boot params...\n");
+
+	hdr = bootx_flatten_dt(space);
+
+#ifdef CONFIG_BOOTX_TEXT
+#ifdef SET_BOOT_BAT
+	bootx_printf("Preparing BAT...\n");
+	btext_prepare_BAT();
+#else
+	btext_unmap();
+#endif
+#endif
+
+	reloc_got2(-offset);
+
+	__start(hdr, KERNELBASE + offset, 0);
+}
diff --git a/arch/powerpc/platforms/powermac/cpufreq_64.c b/arch/powerpc/platforms/powermac/cpufreq_64.c
index 39150342c6f1..a4b50c4109c2 100644
--- a/arch/powerpc/platforms/powermac/cpufreq_64.c
+++ b/arch/powerpc/platforms/powermac/cpufreq_64.c
@@ -28,6 +28,7 @@
 #include <asm/cputable.h>
 #include <asm/time.h>
 #include <asm/smu.h>
+#include <asm/pmac_pfunc.h>
 
 #undef DEBUG
 
@@ -85,6 +86,10 @@ static u32 *g5_pmode_data;
 static int g5_pmode_max;
 static int g5_pmode_cur;
 
+static void (*g5_switch_volt)(int speed_mode);
+static int (*g5_switch_freq)(int speed_mode);
+static int (*g5_query_freq)(void);
+
 static DECLARE_MUTEX(g5_switch_mutex);
 
 
@@ -92,9 +97,11 @@ static struct smu_sdbp_fvt *g5_fvt_table;	/* table of op. points */
 static int g5_fvt_count;			/* number of op. points */
 static int g5_fvt_cur;				/* current op. point */
 
-/* ----------------- real hardware interface */
+/*
+ * SMU based voltage switching for Neo2 platforms
+ */
 
-static void g5_switch_volt(int speed_mode)
+static void g5_smu_switch_volt(int speed_mode)
 {
 	struct smu_simple_cmd	cmd;
 
@@ -105,26 +112,57 @@ static void g5_switch_volt(int speed_mode)
 	wait_for_completion(&comp);
 }
 
-static int g5_switch_freq(int speed_mode)
+/*
+ * Platform function based voltage/vdnap switching for Neo2
+ */
+
+static struct pmf_function *pfunc_set_vdnap0;
+static struct pmf_function *pfunc_vdnap0_complete;
+
+static void g5_vdnap_switch_volt(int speed_mode)
 {
-	struct cpufreq_freqs freqs;
-	int to;
+	struct pmf_args args;
+	u32 slew, done = 0;
+	unsigned long timeout;
 
-	if (g5_pmode_cur == speed_mode)
-		return 0;
+	slew = (speed_mode == CPUFREQ_LOW) ? 1 : 0;
+	args.count = 1;
+	args.u[0].p = &slew;
 
-	down(&g5_switch_mutex);
+	pmf_call_one(pfunc_set_vdnap0, &args);
 
-	freqs.old = g5_cpu_freqs[g5_pmode_cur].frequency;
-	freqs.new = g5_cpu_freqs[speed_mode].frequency;
-	freqs.cpu = 0;
+	/* It's an irq GPIO so we should be able to just block here,
+	 * I'll do that later after I've properly tested the IRQ code for
+	 * platform functions
+	 */
+	timeout = jiffies + HZ/10;
+	while(!time_after(jiffies, timeout)) {
+		args.count = 1;
+		args.u[0].p = &done;
+		pmf_call_one(pfunc_vdnap0_complete, &args);
+		if (done)
+			break;
+		msleep(1);
+	}
+	if (done == 0)
+		printk(KERN_WARNING "cpufreq: Timeout in clock slewing !\n");
+}
 
-	cpufreq_notify_transition(&freqs, CPUFREQ_PRECHANGE);
+
+/*
+ * SCOM based frequency switching for 970FX rev3
+ */
+static int g5_scom_switch_freq(int speed_mode)
+{
+	unsigned long flags;
+	int to;
 
 	/* If frequency is going up, first ramp up the voltage */
 	if (speed_mode < g5_pmode_cur)
 		g5_switch_volt(speed_mode);
 
+	local_irq_save(flags);
+
 	/* Clear PCR high */
 	scom970_write(SCOM_PCR, 0);
 	/* Clear PCR low */
@@ -147,6 +185,8 @@ static int g5_switch_freq(int speed_mode)
 		udelay(100);
 	}
 
+	local_irq_restore(flags);
+
 	/* If frequency is going down, last ramp the voltage */
 	if (speed_mode > g5_pmode_cur)
 		g5_switch_volt(speed_mode);
@@ -154,14 +194,10 @@ static int g5_switch_freq(int speed_mode)
 	g5_pmode_cur = speed_mode;
 	ppc_proc_freq = g5_cpu_freqs[speed_mode].frequency * 1000ul;
 
-	cpufreq_notify_transition(&freqs, CPUFREQ_POSTCHANGE);
-
-	up(&g5_switch_mutex);
-
 	return 0;
 }
 
-static int g5_query_freq(void)
+static int g5_scom_query_freq(void)
 {
 	unsigned long psr = scom970_read(SCOM_PSR);
 	int i;
@@ -173,7 +209,104 @@ static int g5_query_freq(void)
 	return i;
 }
 
-/* ----------------- cpufreq bookkeeping */
+/*
+ * Platform function based voltage switching for PowerMac7,2 & 7,3
+ */
+
+static struct pmf_function *pfunc_cpu0_volt_high;
+static struct pmf_function *pfunc_cpu0_volt_low;
+static struct pmf_function *pfunc_cpu1_volt_high;
+static struct pmf_function *pfunc_cpu1_volt_low;
+
+static void g5_pfunc_switch_volt(int speed_mode)
+{
+	if (speed_mode == CPUFREQ_HIGH) {
+		if (pfunc_cpu0_volt_high)
+			pmf_call_one(pfunc_cpu0_volt_high, NULL);
+		if (pfunc_cpu1_volt_high)
+			pmf_call_one(pfunc_cpu1_volt_high, NULL);
+	} else {
+		if (pfunc_cpu0_volt_low)
+			pmf_call_one(pfunc_cpu0_volt_low, NULL);
+		if (pfunc_cpu1_volt_low)
+			pmf_call_one(pfunc_cpu1_volt_low, NULL);
+	}
+	msleep(10); /* should be faster , to fix */
+}
+
+/*
+ * Platform function based frequency switching for PowerMac7,2 & 7,3
+ */
+
+static struct pmf_function *pfunc_cpu_setfreq_high;
+static struct pmf_function *pfunc_cpu_setfreq_low;
+static struct pmf_function *pfunc_cpu_getfreq;
+static struct pmf_function *pfunc_slewing_done;;
+
+static int g5_pfunc_switch_freq(int speed_mode)
+{
+	struct pmf_args args;
+	u32 done = 0;
+	unsigned long timeout;
+
+	/* If frequency is going up, first ramp up the voltage */
+	if (speed_mode < g5_pmode_cur)
+		g5_switch_volt(speed_mode);
+
+	/* Do it */
+	if (speed_mode == CPUFREQ_HIGH)
+		pmf_call_one(pfunc_cpu_setfreq_high, NULL);
+	else
+		pmf_call_one(pfunc_cpu_setfreq_low, NULL);
+
+	/* It's an irq GPIO so we should be able to just block here,
+	 * I'll do that later after I've properly tested the IRQ code for
+	 * platform functions
+	 */
+	timeout = jiffies + HZ/10;
+	while(!time_after(jiffies, timeout)) {
+		args.count = 1;
+		args.u[0].p = &done;
+		pmf_call_one(pfunc_slewing_done, &args);
+		if (done)
+			break;
+		msleep(1);
+	}
+	if (done == 0)
+		printk(KERN_WARNING "cpufreq: Timeout in clock slewing !\n");
+
+	/* If frequency is going down, last ramp the voltage */
+	if (speed_mode > g5_pmode_cur)
+		g5_switch_volt(speed_mode);
+
+	g5_pmode_cur = speed_mode;
+	ppc_proc_freq = g5_cpu_freqs[speed_mode].frequency * 1000ul;
+
+	return 0;
+}
+
+static int g5_pfunc_query_freq(void)
+{
+	struct pmf_args args;
+	u32 val = 0;
+
+	args.count = 1;
+	args.u[0].p = &val;
+	pmf_call_one(pfunc_cpu_getfreq, &args);
+	return val ? CPUFREQ_HIGH : CPUFREQ_LOW;
+}
+
+/*
+ * Fake voltage switching for platforms with missing support
+ */
+
+static void g5_dummy_switch_volt(int speed_mode)
+{
+}
+
+/*
+ * Common interface to the cpufreq core
+ */
 
 static int g5_cpufreq_verify(struct cpufreq_policy *policy)
 {
@@ -183,13 +316,30 @@ static int g5_cpufreq_verify(struct cpufreq_policy *policy)
 static int g5_cpufreq_target(struct cpufreq_policy *policy,
 	unsigned int target_freq, unsigned int relation)
 {
-	unsigned int    newstate = 0;
+	unsigned int newstate = 0;
+	struct cpufreq_freqs freqs;
+	int rc;
 
 	if (cpufreq_frequency_table_target(policy, g5_cpu_freqs,
 			target_freq, relation, &newstate))
 		return -EINVAL;
 
-	return g5_switch_freq(newstate);
+	if (g5_pmode_cur == newstate)
+		return 0;
+
+	down(&g5_switch_mutex);
+
+	freqs.old = g5_cpu_freqs[g5_pmode_cur].frequency;
+	freqs.new = g5_cpu_freqs[newstate].frequency;
+	freqs.cpu = 0;
+
+	cpufreq_notify_transition(&freqs, CPUFREQ_PRECHANGE);
+	rc = g5_switch_freq(newstate);
+	cpufreq_notify_transition(&freqs, CPUFREQ_POSTCHANGE);
+
+	up(&g5_switch_mutex);
+
+	return rc;
 }
 
 static unsigned int g5_cpufreq_get_speed(unsigned int cpu)
@@ -205,6 +355,7 @@ static int g5_cpufreq_cpu_init(struct cpufreq_policy *policy)
 	policy->governor = CPUFREQ_DEFAULT_GOVERNOR;
 	policy->cpuinfo.transition_latency = CPUFREQ_ETERNAL;
 	policy->cur = g5_cpu_freqs[g5_query_freq()].frequency;
+	policy->cpus = cpu_possible_map;
 	cpufreq_frequency_table_get_attr(g5_cpu_freqs, policy->cpu);
 
 	return cpufreq_frequency_table_cpuinfo(policy,
@@ -224,19 +375,39 @@ static struct cpufreq_driver g5_cpufreq_driver = {
 };
 
 
-static int __init g5_cpufreq_init(void)
+static int __init g5_neo2_cpufreq_init(struct device_node *cpus)
 {
 	struct device_node *cpunode;
 	unsigned int psize, ssize;
-	struct smu_sdbp_header *shdr;
 	unsigned long max_freq;
-	u32 *valp;
+	char *freq_method, *volt_method;
+	u32 *valp, pvr_hi;
+	int use_volts_vdnap = 0;
+	int use_volts_smu = 0;
 	int rc = -ENODEV;
 
-	/* Look for CPU and SMU nodes */
-	cpunode = of_find_node_by_type(NULL, "cpu");
-	if (!cpunode) {
-		DBG("No CPU node !\n");
+	/* Check supported platforms */
+	if (machine_is_compatible("PowerMac8,1") ||
+	    machine_is_compatible("PowerMac8,2") ||
+	    machine_is_compatible("PowerMac9,1"))
+		use_volts_smu = 1;
+	else if (machine_is_compatible("PowerMac11,2"))
+		use_volts_vdnap = 1;
+	else
+		return -ENODEV;
+
+	/* Get first CPU node */
+	for (cpunode = NULL;
+	     (cpunode = of_get_next_child(cpus, cpunode)) != NULL;) {
+		u32 *reg =
+			(u32 *)get_property(cpunode, "reg", NULL);
+		if (reg == NULL || (*reg) != 0)
+			continue;
+		if (!strcmp(cpunode->type, "cpu"))
+			break;
+	}
+	if (cpunode == NULL) {
+		printk(KERN_ERR "cpufreq: Can't find any CPU 0 node\n");
 		return -ENODEV;
 	}
 
@@ -246,8 +417,9 @@ static int __init g5_cpufreq_init(void)
 		DBG("No cpu-version property !\n");
 		goto bail_noprops;
 	}
-	if (((*valp) >> 16) != 0x3c) {
-		DBG("Wrong CPU version: %08x\n", *valp);
+	pvr_hi = (*valp) >> 16;
+	if (pvr_hi != 0x3c && pvr_hi != 0x44) {
+		printk(KERN_ERR "cpufreq: Unsupported CPU version\n");
 		goto bail_noprops;
 	}
 
@@ -259,18 +431,50 @@ static int __init g5_cpufreq_init(void)
 	}
 	g5_pmode_max = psize / sizeof(u32) - 1;
 
-	/* Look for the FVT table */
-	shdr = smu_get_sdb_partition(SMU_SDB_FVT_ID, NULL);
-	if (!shdr)
-		goto bail_noprops;
-	g5_fvt_table = (struct smu_sdbp_fvt *)&shdr[1];
-	ssize = (shdr->len * sizeof(u32)) - sizeof(struct smu_sdbp_header);
-	g5_fvt_count = ssize / sizeof(struct smu_sdbp_fvt);
-	g5_fvt_cur = 0;
-
-	/* Sanity checking */
-	if (g5_fvt_count < 1 || g5_pmode_max < 1)
-		goto bail_noprops;
+	if (use_volts_smu) {
+		struct smu_sdbp_header *shdr;
+
+		/* Look for the FVT table */
+		shdr = smu_get_sdb_partition(SMU_SDB_FVT_ID, NULL);
+		if (!shdr)
+			goto bail_noprops;
+		g5_fvt_table = (struct smu_sdbp_fvt *)&shdr[1];
+		ssize = (shdr->len * sizeof(u32)) -
+			sizeof(struct smu_sdbp_header);
+		g5_fvt_count = ssize / sizeof(struct smu_sdbp_fvt);
+		g5_fvt_cur = 0;
+
+		/* Sanity checking */
+		if (g5_fvt_count < 1 || g5_pmode_max < 1)
+			goto bail_noprops;
+
+		g5_switch_volt = g5_smu_switch_volt;
+		volt_method = "SMU";
+	} else if (use_volts_vdnap) {
+		struct device_node *root;
+
+		root = of_find_node_by_path("/");
+		if (root == NULL) {
+			printk(KERN_ERR "cpufreq: Can't find root of "
+			       "device tree\n");
+			goto bail_noprops;
+		}
+		pfunc_set_vdnap0 = pmf_find_function(root, "set-vdnap0");
+		pfunc_vdnap0_complete =
+			pmf_find_function(root, "slewing-done");
+		if (pfunc_set_vdnap0 == NULL ||
+		    pfunc_vdnap0_complete == NULL) {
+			printk(KERN_ERR "cpufreq: Can't find required "
+			       "platform function\n");
+			goto bail_noprops;
+		}
+
+		g5_switch_volt = g5_vdnap_switch_volt;
+		volt_method = "GPIO";
+	} else {
+		g5_switch_volt = g5_dummy_switch_volt;
+		volt_method = "none";
+	}
 
 	/*
 	 * From what I see, clock-frequency is always the maximal frequency.
@@ -286,19 +490,23 @@ static int __init g5_cpufreq_init(void)
 	g5_cpu_freqs[0].frequency = max_freq;
 	g5_cpu_freqs[1].frequency = max_freq/2;
 
-	/* Check current frequency */
-	g5_pmode_cur = g5_query_freq();
-	if (g5_pmode_cur > 1)
-		/* We don't support anything but 1:1 and 1:2, fixup ... */
-		g5_pmode_cur = 1;
+	/* Set callbacks */
+	g5_switch_freq = g5_scom_switch_freq;
+	g5_query_freq = g5_scom_query_freq;
+	freq_method = "SCOM";
 
 	/* Force apply current frequency to make sure everything is in
 	 * sync (voltage is right for example). Firmware may leave us with
 	 * a strange setting ...
 	 */
-	g5_switch_freq(g5_pmode_cur);
+	g5_switch_volt(CPUFREQ_HIGH);
+	msleep(10);
+	g5_pmode_cur = -1;
+	g5_switch_freq(g5_query_freq());
 
 	printk(KERN_INFO "Registering G5 CPU frequency driver\n");
+	printk(KERN_INFO "Frequency method: %s, Voltage method: %s\n",
+	       freq_method, volt_method);
 	printk(KERN_INFO "Low: %d Mhz, High: %d Mhz, Cur: %d MHz\n",
 		g5_cpu_freqs[1].frequency/1000,
 		g5_cpu_freqs[0].frequency/1000,
@@ -317,6 +525,200 @@ static int __init g5_cpufreq_init(void)
 	return rc;
 }
 
+static int __init g5_pm72_cpufreq_init(struct device_node *cpus)
+{
+	struct device_node *cpuid = NULL, *hwclock = NULL, *cpunode = NULL;
+	u8 *eeprom = NULL;
+	u32 *valp;
+	u64 max_freq, min_freq, ih, il;
+	int has_volt = 1, rc = 0;
+
+	/* Get first CPU node */
+	for (cpunode = NULL;
+	     (cpunode = of_get_next_child(cpus, cpunode)) != NULL;) {
+		if (!strcmp(cpunode->type, "cpu"))
+			break;
+	}
+	if (cpunode == NULL) {
+		printk(KERN_ERR "cpufreq: Can't find any CPU node\n");
+		return -ENODEV;
+	}
+
+	/* Lookup the cpuid eeprom node */
+        cpuid = of_find_node_by_path("/u3@0,f8000000/i2c@f8001000/cpuid@a0");
+	if (cpuid != NULL)
+		eeprom = (u8 *)get_property(cpuid, "cpuid", NULL);
+	if (eeprom == NULL) {
+		printk(KERN_ERR "cpufreq: Can't find cpuid EEPROM !\n");
+		rc = -ENODEV;
+		goto bail;
+	}
+
+	/* Lookup the i2c hwclock */
+	for (hwclock = NULL;
+	     (hwclock = of_find_node_by_name(hwclock, "i2c-hwclock")) != NULL;){
+		char *loc = get_property(hwclock, "hwctrl-location", NULL);
+		if (loc == NULL)
+			continue;
+		if (strcmp(loc, "CPU CLOCK"))
+			continue;
+		if (!get_property(hwclock, "platform-get-frequency", NULL))
+			continue;
+		break;
+	}
+	if (hwclock == NULL) {
+		printk(KERN_ERR "cpufreq: Can't find i2c clock chip !\n");
+		rc = -ENODEV;
+		goto bail;
+	}
+
+	DBG("cpufreq: i2c clock chip found: %s\n", hwclock->full_name);
+
+	/* Now get all the platform functions */
+	pfunc_cpu_getfreq =
+		pmf_find_function(hwclock, "get-frequency");
+	pfunc_cpu_setfreq_high =
+		pmf_find_function(hwclock, "set-frequency-high");
+	pfunc_cpu_setfreq_low =
+		pmf_find_function(hwclock, "set-frequency-low");
+	pfunc_slewing_done =
+		pmf_find_function(hwclock, "slewing-done");
+	pfunc_cpu0_volt_high =
+		pmf_find_function(hwclock, "set-voltage-high-0");
+	pfunc_cpu0_volt_low =
+		pmf_find_function(hwclock, "set-voltage-low-0");
+	pfunc_cpu1_volt_high =
+		pmf_find_function(hwclock, "set-voltage-high-1");
+	pfunc_cpu1_volt_low =
+		pmf_find_function(hwclock, "set-voltage-low-1");
+
+	/* Check we have minimum requirements */
+	if (pfunc_cpu_getfreq == NULL || pfunc_cpu_setfreq_high == NULL ||
+	    pfunc_cpu_setfreq_low == NULL || pfunc_slewing_done == NULL) {
+		printk(KERN_ERR "cpufreq: Can't find platform functions !\n");
+		rc = -ENODEV;
+		goto bail;
+	}
+
+	/* Check that we have complete sets */
+	if (pfunc_cpu0_volt_high == NULL || pfunc_cpu0_volt_low == NULL) {
+		pmf_put_function(pfunc_cpu0_volt_high);
+		pmf_put_function(pfunc_cpu0_volt_low);
+		pfunc_cpu0_volt_high = pfunc_cpu0_volt_low = NULL;
+		has_volt = 0;
+	}
+	if (!has_volt ||
+	    pfunc_cpu1_volt_high == NULL || pfunc_cpu1_volt_low == NULL) {
+		pmf_put_function(pfunc_cpu1_volt_high);
+		pmf_put_function(pfunc_cpu1_volt_low);
+		pfunc_cpu1_volt_high = pfunc_cpu1_volt_low = NULL;
+	}
+
+	/* Note: The device tree also contains a "platform-set-values"
+	 * function for which I haven't quite figured out the usage. It
+	 * might have to be called on init and/or wakeup, I'm not too sure
+	 * but things seem to work fine without it so far ...
+	 */
+
+	/* Get max frequency from device-tree */
+	valp = (u32 *)get_property(cpunode, "clock-frequency", NULL);
+	if (!valp) {
+		printk(KERN_ERR "cpufreq: Can't find CPU frequency !\n");
+		rc = -ENODEV;
+		goto bail;
+	}
+
+	max_freq = (*valp)/1000;
+
+	/* Now calculate reduced frequency by using the cpuid input freq
+	 * ratio. This requires 64 bits math unless we are willing to lose
+	 * some precision
+	 */
+	ih = *((u32 *)(eeprom + 0x10));
+	il = *((u32 *)(eeprom + 0x20));
+	min_freq = 0;
+	if (ih != 0 && il != 0)
+		min_freq = (max_freq * il) / ih;
+
+	/* Sanity check */
+	if (min_freq >= max_freq || min_freq < 1000) {
+		printk(KERN_ERR "cpufreq: Can't calculate low frequency !\n");
+		rc = -ENODEV;
+		goto bail;
+	}
+	g5_cpu_freqs[0].frequency = max_freq;
+	g5_cpu_freqs[1].frequency = min_freq;
+
+	/* Set callbacks */
+	g5_switch_volt = g5_pfunc_switch_volt;
+	g5_switch_freq = g5_pfunc_switch_freq;
+	g5_query_freq = g5_pfunc_query_freq;
+
+	/* Force apply current frequency to make sure everything is in
+	 * sync (voltage is right for example). Firmware may leave us with
+	 * a strange setting ...
+	 */
+	g5_switch_volt(CPUFREQ_HIGH);
+	msleep(10);
+	g5_pmode_cur = -1;
+	g5_switch_freq(g5_query_freq());
+
+	printk(KERN_INFO "Registering G5 CPU frequency driver\n");
+	printk(KERN_INFO "Frequency method: i2c/pfunc, "
+	       "Voltage method: %s\n", has_volt ? "i2c/pfunc" : "none");
+	printk(KERN_INFO "Low: %d Mhz, High: %d Mhz, Cur: %d MHz\n",
+		g5_cpu_freqs[1].frequency/1000,
+		g5_cpu_freqs[0].frequency/1000,
+		g5_cpu_freqs[g5_pmode_cur].frequency/1000);
+
+	rc = cpufreq_register_driver(&g5_cpufreq_driver);
+ bail:
+	if (rc != 0) {
+		pmf_put_function(pfunc_cpu_getfreq);
+		pmf_put_function(pfunc_cpu_setfreq_high);
+		pmf_put_function(pfunc_cpu_setfreq_low);
+		pmf_put_function(pfunc_slewing_done);
+		pmf_put_function(pfunc_cpu0_volt_high);
+		pmf_put_function(pfunc_cpu0_volt_low);
+		pmf_put_function(pfunc_cpu1_volt_high);
+		pmf_put_function(pfunc_cpu1_volt_low);
+	}
+	of_node_put(hwclock);
+	of_node_put(cpuid);
+	of_node_put(cpunode);
+
+	return rc;
+}
+
+static int __init g5_rm31_cpufreq_init(struct device_node *cpus)
+{
+	/* NYI */
+	return 0;
+}
+
+static int __init g5_cpufreq_init(void)
+{
+	struct device_node *cpus;
+	int rc;
+
+	cpus = of_find_node_by_path("/cpus");
+	if (cpus == NULL) {
+		DBG("No /cpus node !\n");
+		return -ENODEV;
+	}
+
+	if (machine_is_compatible("PowerMac7,2") ||
+	    machine_is_compatible("PowerMac7,3"))
+		rc = g5_pm72_cpufreq_init(cpus);
+	else if (machine_is_compatible("RackMac3,1"))
+		rc = g5_rm31_cpufreq_init(cpus);
+	else
+		rc = g5_neo2_cpufreq_init(cpus);
+
+	of_node_put(cpus);
+	return rc;
+}
+
 module_init(g5_cpufreq_init);
 
 
diff --git a/arch/powerpc/platforms/powermac/feature.c b/arch/powerpc/platforms/powermac/feature.c
index f6e22da2a5da..558dd0692092 100644
--- a/arch/powerpc/platforms/powermac/feature.c
+++ b/arch/powerpc/platforms/powermac/feature.c
@@ -58,12 +58,11 @@ extern int powersave_lowspeed;
 extern int powersave_nap;
 extern struct device_node *k2_skiplist[2];
 
-
 /*
  * We use a single global lock to protect accesses. Each driver has
  * to take care of its own locking
  */
-static DEFINE_SPINLOCK(feature_lock);
+DEFINE_SPINLOCK(feature_lock);
 
 #define LOCK(flags)	spin_lock_irqsave(&feature_lock, flags);
 #define UNLOCK(flags)	spin_unlock_irqrestore(&feature_lock, flags);
@@ -101,26 +100,17 @@ static const char *macio_names[] =
 	"Keylargo",
 	"Pangea",
 	"Intrepid",
-	"K2"
+	"K2",
+	"Shasta",
 };
 
 
+struct device_node *uninorth_node;
+u32 __iomem *uninorth_base;
 
-/*
- * Uninorth reg. access. Note that Uni-N regs are big endian
- */
-
-#define UN_REG(r)	(uninorth_base + ((r) >> 2))
-#define UN_IN(r)	(in_be32(UN_REG(r)))
-#define UN_OUT(r,v)	(out_be32(UN_REG(r), (v)))
-#define UN_BIS(r,v)	(UN_OUT((r), UN_IN(r) | (v)))
-#define UN_BIC(r,v)	(UN_OUT((r), UN_IN(r) & ~(v)))
-
-static struct device_node *uninorth_node;
-static u32 __iomem *uninorth_base;
 static u32 uninorth_rev;
-static int uninorth_u3;
-static void __iomem *u3_ht;
+static int uninorth_maj;
+static void __iomem *u3_ht_base;
 
 /*
  * For each motherboard family, we have a table of functions pointers
@@ -1399,8 +1389,15 @@ static long g5_fw_enable(struct device_node *node, long param, long value)
 static long g5_mpic_enable(struct device_node *node, long param, long value)
 {
 	unsigned long flags;
+	struct device_node *parent = of_get_parent(node);
+	int is_u3;
 
-	if (node->parent == NULL || strcmp(node->parent->name, "u3"))
+	if (parent == NULL)
+		return 0;
+	is_u3 = strcmp(parent->name, "u3") == 0 ||
+		strcmp(parent->name, "u4") == 0;
+	of_node_put(parent);
+	if (!is_u3)
 		return 0;
 
 	LOCK(flags);
@@ -1445,20 +1442,53 @@ static long g5_i2s_enable(struct device_node *node, long param, long value)
 	/* Very crude implementation for now */
 	struct macio_chip *macio = &macio_chips[0];
 	unsigned long flags;
-
-	if (value == 0)
-		return 0; /* don't disable yet */
+	int cell;
+	u32 fcrs[3][3] = {
+		{ 0,
+		  K2_FCR1_I2S0_CELL_ENABLE |
+		  K2_FCR1_I2S0_CLK_ENABLE_BIT | K2_FCR1_I2S0_ENABLE,
+		  KL3_I2S0_CLK18_ENABLE
+		},
+		{ KL0_SCC_A_INTF_ENABLE,
+		  K2_FCR1_I2S1_CELL_ENABLE |
+		  K2_FCR1_I2S1_CLK_ENABLE_BIT | K2_FCR1_I2S1_ENABLE,
+		  KL3_I2S1_CLK18_ENABLE
+		},
+		{ KL0_SCC_B_INTF_ENABLE,
+		  SH_FCR1_I2S2_CELL_ENABLE |
+		  SH_FCR1_I2S2_CLK_ENABLE_BIT | SH_FCR1_I2S2_ENABLE,
+		  SH_FCR3_I2S2_CLK18_ENABLE
+		},
+	};
+
+	if (macio->type != macio_keylargo2 && macio->type != macio_shasta)
+		return -ENODEV;
+	if (strncmp(node->name, "i2s-", 4))
+		return -ENODEV;
+	cell = node->name[4] - 'a';
+	switch(cell) {
+	case 0:
+	case 1:
+		break;
+	case 2:
+		if (macio->type == macio_shasta)
+			break;
+	default:
+		return -ENODEV;
+	}
 
 	LOCK(flags);
-	MACIO_BIS(KEYLARGO_FCR3, KL3_CLK45_ENABLE | KL3_CLK49_ENABLE |
-		  KL3_I2S0_CLK18_ENABLE);
-	udelay(10);
-	MACIO_BIS(KEYLARGO_FCR1, K2_FCR1_I2S0_CELL_ENABLE |
-		  K2_FCR1_I2S0_CLK_ENABLE_BIT | K2_FCR1_I2S0_ENABLE);
+	if (value) {
+		MACIO_BIC(KEYLARGO_FCR0, fcrs[cell][0]);
+		MACIO_BIS(KEYLARGO_FCR1, fcrs[cell][1]);
+		MACIO_BIS(KEYLARGO_FCR3, fcrs[cell][2]);
+	} else {
+		MACIO_BIC(KEYLARGO_FCR3, fcrs[cell][2]);
+		MACIO_BIC(KEYLARGO_FCR1, fcrs[cell][1]);
+		MACIO_BIS(KEYLARGO_FCR0, fcrs[cell][0]);
+	}
 	udelay(10);
-	MACIO_BIC(KEYLARGO_FCR1, K2_FCR1_I2S0_RESET);
 	UNLOCK(flags);
-	udelay(10);
 
 	return 0;
 }
@@ -1473,7 +1503,7 @@ static long g5_reset_cpu(struct device_node *node, long param, long value)
 	struct device_node *np;
 
 	macio = &macio_chips[0];
-	if (macio->type != macio_keylargo2)
+	if (macio->type != macio_keylargo2 && macio->type != macio_shasta)
 		return -ENODEV;
 
 	np = find_path_device("/cpus");
@@ -1512,14 +1542,17 @@ static long g5_reset_cpu(struct device_node *node, long param, long value)
  */
 void g5_phy_disable_cpu1(void)
 {
-	UN_OUT(U3_API_PHY_CONFIG_1, 0);
+	if (uninorth_maj == 3)
+		UN_OUT(U3_API_PHY_CONFIG_1, 0);
 }
 #endif /* CONFIG_POWER4 */
 
 #ifndef CONFIG_POWER4
 
-static void
-keylargo_shutdown(struct macio_chip *macio, int sleep_mode)
+
+#ifdef CONFIG_PM
+
+static void keylargo_shutdown(struct macio_chip *macio, int sleep_mode)
 {
 	u32 temp;
 
@@ -1572,8 +1605,7 @@ keylargo_shutdown(struct macio_chip *macio, int sleep_mode)
 	(void)MACIO_IN32(KEYLARGO_FCR0); mdelay(1);
 }
 
-static void
-pangea_shutdown(struct macio_chip *macio, int sleep_mode)
+static void pangea_shutdown(struct macio_chip *macio, int sleep_mode)
 {
 	u32 temp;
 
@@ -1606,8 +1638,7 @@ pangea_shutdown(struct macio_chip *macio, int sleep_mode)
 	(void)MACIO_IN32(KEYLARGO_FCR0); mdelay(1);
 }
 
-static void
-intrepid_shutdown(struct macio_chip *macio, int sleep_mode)
+static void intrepid_shutdown(struct macio_chip *macio, int sleep_mode)
 {
 	u32 temp;
 
@@ -1635,124 +1666,6 @@ intrepid_shutdown(struct macio_chip *macio, int sleep_mode)
 }
 
 
-void pmac_tweak_clock_spreading(int enable)
-{
-	struct macio_chip *macio = &macio_chips[0];
-
-	/* Hack for doing clock spreading on some machines PowerBooks and
-	 * iBooks. This implements the "platform-do-clockspreading" OF
-	 * property as decoded manually on various models. For safety, we also
-	 * check the product ID in the device-tree in cases we'll whack the i2c
-	 * chip to make reasonably sure we won't set wrong values in there
-	 *
-	 * Of course, ultimately, we have to implement a real parser for
-	 * the platform-do-* stuff...
-	 */
-
-	if (macio->type == macio_intrepid) {
-		struct device_node *clock =
-			of_find_node_by_path("/uni-n@f8000000/hw-clock");
-		if (clock && get_property(clock, "platform-do-clockspreading",
-					  NULL)) {
-			printk(KERN_INFO "%sabling clock spreading on Intrepid"
-			       " ASIC\n", enable ? "En" : "Dis");
-			if (enable)
-				UN_OUT(UNI_N_CLOCK_SPREADING, 2);
-			else
-				UN_OUT(UNI_N_CLOCK_SPREADING, 0);
-			mdelay(40);
-		}
-		of_node_put(clock);
-	}
-
-	while (machine_is_compatible("PowerBook5,2") ||
-	       machine_is_compatible("PowerBook5,3") ||
-	       machine_is_compatible("PowerBook6,2") ||
-	       machine_is_compatible("PowerBook6,3")) {
-		struct device_node *ui2c = of_find_node_by_type(NULL, "i2c");
-		struct device_node *dt = of_find_node_by_name(NULL, "device-tree");
-		u8 buffer[9];
-		u32 *productID;
-		int i, rc, changed = 0;
-
-		if (dt == NULL)
-			break;
-		productID = (u32 *)get_property(dt, "pid#", NULL);
-		if (productID == NULL)
-			break;
-		while(ui2c) {
-			struct device_node *p = of_get_parent(ui2c);
-			if (p && !strcmp(p->name, "uni-n"))
-				break;
-			ui2c = of_find_node_by_type(ui2c, "i2c");
-		}
-		if (ui2c == NULL)
-			break;
-		DBG("Trying to bump clock speed for PID: %08x...\n", *productID);
-		rc = pmac_low_i2c_open(ui2c, 1);
-		if (rc != 0)
-			break;
-		pmac_low_i2c_setmode(ui2c, pmac_low_i2c_mode_combined);
-		rc = pmac_low_i2c_xfer(ui2c, 0xd2 | pmac_low_i2c_read, 0x80, buffer, 9);
-		DBG("read result: %d,", rc);
-		if (rc != 0) {
-			pmac_low_i2c_close(ui2c);
-			break;
-		}
-		for (i=0; i<9; i++)
-			DBG(" %02x", buffer[i]);
-		DBG("\n");
-
-		switch(*productID) {
-		case 0x1182:	/* AlBook 12" rev 2 */
-		case 0x1183:	/* iBook G4 12" */
-			buffer[0] = (buffer[0] & 0x8f) | 0x70;
-			buffer[2] = (buffer[2] & 0x7f) | 0x00;
-			buffer[5] = (buffer[5] & 0x80) | 0x31;
-			buffer[6] = (buffer[6] & 0x40) | 0xb0;
-			buffer[7] = (buffer[7] & 0x00) | (enable ? 0xc0 : 0xba);
-			buffer[8] = (buffer[8] & 0x00) | 0x30;
-			changed = 1;
-			break;
-		case 0x3142:	/* AlBook 15" (ATI M10) */
-		case 0x3143:	/* AlBook 17" (ATI M10) */
-			buffer[0] = (buffer[0] & 0xaf) | 0x50;
-			buffer[2] = (buffer[2] & 0x7f) | 0x00;
-			buffer[5] = (buffer[5] & 0x80) | 0x31;
-			buffer[6] = (buffer[6] & 0x40) | 0xb0;
-			buffer[7] = (buffer[7] & 0x00) | (enable ? 0xd0 : 0xc0);
-			buffer[8] = (buffer[8] & 0x00) | 0x30;
-			changed = 1;
-			break;
-		default:
-			DBG("i2c-hwclock: Machine model not handled\n");
-			break;
-		}
-		if (!changed) {
-			pmac_low_i2c_close(ui2c);
-			break;
-		}
-		printk(KERN_INFO "%sabling clock spreading on i2c clock chip\n",
-		       enable ? "En" : "Dis");
-
-		pmac_low_i2c_setmode(ui2c, pmac_low_i2c_mode_stdsub);
-		rc = pmac_low_i2c_xfer(ui2c, 0xd2 | pmac_low_i2c_write, 0x80, buffer, 9);
-		DBG("write result: %d,", rc);
-		pmac_low_i2c_setmode(ui2c, pmac_low_i2c_mode_combined);
-		rc = pmac_low_i2c_xfer(ui2c, 0xd2 | pmac_low_i2c_read, 0x80, buffer, 9);
-		DBG("read result: %d,", rc);
-		if (rc != 0) {
-			pmac_low_i2c_close(ui2c);
-			break;
-		}
-		for (i=0; i<9; i++)
-			DBG(" %02x", buffer[i]);
-		pmac_low_i2c_close(ui2c);
-		break;
-	}
-}
-
-
 static int
 core99_sleep(void)
 {
@@ -1909,6 +1822,8 @@ core99_wake_up(void)
 	return 0;
 }
 
+#endif /* CONFIG_PM */
+
 static long
 core99_sleep_state(struct device_node *node, long param, long value)
 {
@@ -1930,10 +1845,13 @@ core99_sleep_state(struct device_node *node, long param, long value)
 	if ((pmac_mb.board_flags & PMAC_MB_CAN_SLEEP) == 0)
 		return -EPERM;
 
+#ifdef CONFIG_PM
 	if (value == 1)
 		return core99_sleep();
 	else if (value == 0)
 		return core99_wake_up();
+
+#endif /* CONFIG_PM */
 	return 0;
 }
 
@@ -2057,7 +1975,9 @@ static struct feature_table_entry core99_features[] = {
 	{ PMAC_FTR_USB_ENABLE,		core99_usb_enable },
 	{ PMAC_FTR_1394_ENABLE,		core99_firewire_enable },
 	{ PMAC_FTR_1394_CABLE_POWER,	core99_firewire_cable_power },
+#ifdef CONFIG_PM
 	{ PMAC_FTR_SLEEP_STATE,		core99_sleep_state },
+#endif
 #ifdef CONFIG_SMP
 	{ PMAC_FTR_RESET_CPU,		core99_reset_cpu },
 #endif /* CONFIG_SMP */
@@ -2427,6 +2347,14 @@ static struct pmac_mb_def pmac_mb_defs[] = {
 		PMAC_TYPE_POWERMAC_G5_U3L,	g5_features,
 		0,
 	},
+	{	"PowerMac11,2",			"PowerMac G5 Dual Core",
+		PMAC_TYPE_POWERMAC_G5_U3L,	g5_features,
+		0,
+	},
+	{	"PowerMac12,1",			"iMac G5 (iSight)",
+		PMAC_TYPE_POWERMAC_G5_U3L,	g5_features,
+		0,
+	},
 	{       "RackMac3,1",                   "XServe G5",
 		PMAC_TYPE_XSERVE_G5,		g5_features,
 		0,
@@ -2539,6 +2467,11 @@ static int __init probe_motherboard(void)
 		pmac_mb.model_name = "Unknown K2-based";
 		pmac_mb.features = g5_features;
 		break;
+	case macio_shasta:
+		pmac_mb.model_id = PMAC_TYPE_UNKNOWN_SHASTA;
+		pmac_mb.model_name = "Unknown Shasta-based";
+		pmac_mb.features = g5_features;
+		break;
 #endif /* CONFIG_POWER4 */
 	default:
 		return -ENODEV;
@@ -2607,6 +2540,8 @@ found:
  */
 static void __init probe_uninorth(void)
 {
+	u32 *addrp;
+	phys_addr_t address;
 	unsigned long actrl;
 
 	/* Locate core99 Uni-N */
@@ -2614,22 +2549,31 @@ static void __init probe_uninorth(void)
 	/* Locate G5 u3 */
 	if (uninorth_node == NULL) {
 		uninorth_node = of_find_node_by_name(NULL, "u3");
-		uninorth_u3 = 1;
-	}
-	if (uninorth_node && uninorth_node->n_addrs > 0) {
-		unsigned long address = uninorth_node->addrs[0].address;
-		uninorth_base = ioremap(address, 0x40000);
-		uninorth_rev = in_be32(UN_REG(UNI_N_VERSION));
-		if (uninorth_u3)
-			u3_ht = ioremap(address + U3_HT_CONFIG_BASE, 0x1000);
-	} else
-		uninorth_node = NULL;
-
-	if (!uninorth_node)
+		uninorth_maj = 3;
+	}
+	/* Locate G5 u4 */
+	if (uninorth_node == NULL) {
+		uninorth_node = of_find_node_by_name(NULL, "u4");
+		uninorth_maj = 4;
+	}
+	if (uninorth_node == NULL)
 		return;
 
-	printk(KERN_INFO "Found %s memory controller & host bridge, revision: %d\n",
-	       uninorth_u3 ? "U3" : "UniNorth", uninorth_rev);
+	addrp = (u32 *)get_property(uninorth_node, "reg", NULL);
+	if (addrp == NULL)
+		return;
+	address = of_translate_address(uninorth_node, addrp);
+	if (address == 0)
+		return;
+	uninorth_base = ioremap(address, 0x40000);
+	uninorth_rev = in_be32(UN_REG(UNI_N_VERSION));
+	if (uninorth_maj == 3 || uninorth_maj == 4)
+		u3_ht_base = ioremap(address + U3_HT_CONFIG_BASE, 0x1000);
+
+	printk(KERN_INFO "Found %s memory controller & host bridge"
+	       " @ 0x%08x revision: 0x%02x\n", uninorth_maj == 3 ? "U3" :
+	       uninorth_maj == 4 ? "U4" : "UniNorth",
+	       (unsigned int)address, uninorth_rev);
 	printk(KERN_INFO "Mapped at 0x%08lx\n", (unsigned long)uninorth_base);
 
 	/* Set the arbitrer QAck delay according to what Apple does
@@ -2637,7 +2581,8 @@ static void __init probe_uninorth(void)
 	if (uninorth_rev < 0x11) {
 		actrl = UN_IN(UNI_N_ARB_CTRL) & ~UNI_N_ARB_CTRL_QACK_DELAY_MASK;
 		actrl |= ((uninorth_rev < 3) ? UNI_N_ARB_CTRL_QACK_DELAY105 :
-			UNI_N_ARB_CTRL_QACK_DELAY) << UNI_N_ARB_CTRL_QACK_DELAY_SHIFT;
+			UNI_N_ARB_CTRL_QACK_DELAY) <<
+			UNI_N_ARB_CTRL_QACK_DELAY_SHIFT;
 		UN_OUT(UNI_N_ARB_CTRL, actrl);
 	}
 
@@ -2645,7 +2590,8 @@ static void __init probe_uninorth(void)
 	 * revs 1.5 to 2.O and Pangea. Seem to toggle the UniN Maxbus/PCI
 	 * memory timeout
 	 */
-	if ((uninorth_rev >= 0x11 && uninorth_rev <= 0x24) || uninorth_rev == 0xc0)
+	if ((uninorth_rev >= 0x11 && uninorth_rev <= 0x24) ||
+	    uninorth_rev == 0xc0)
 		UN_OUT(0x2160, UN_IN(0x2160) & 0x00ffffff);
 }
 
@@ -2653,18 +2599,17 @@ static void __init probe_one_macio(const char *name, const char *compat, int typ
 {
 	struct device_node*	node;
 	int			i;
-	volatile u32 __iomem *	base;
-	u32*			revp;
+	volatile u32 __iomem	*base;
+	u32			*addrp, *revp;
+	phys_addr_t		addr;
+	u64			size;
 
-	node = find_devices(name);
-	if (!node || !node->n_addrs)
-		return;
-	if (compat)
-		do {
-			if (device_is_compatible(node, compat))
-				break;
-			node = node->next;
-		} while (node);
+	for (node = NULL; (node = of_find_node_by_name(node, name)) != NULL;) {
+		if (!compat)
+			break;
+		if (device_is_compatible(node, compat))
+			break;
+	}
 	if (!node)
 		return;
 	for(i=0; i<MAX_MACIO_CHIPS; i++) {
@@ -2673,22 +2618,38 @@ static void __init probe_one_macio(const char *name, const char *compat, int typ
 		if (macio_chips[i].of_node == node)
 			return;
 	}
+
 	if (i >= MAX_MACIO_CHIPS) {
 		printk(KERN_ERR "pmac_feature: Please increase MAX_MACIO_CHIPS !\n");
 		printk(KERN_ERR "pmac_feature: %s skipped\n", node->full_name);
 		return;
 	}
-	base = ioremap(node->addrs[0].address, node->addrs[0].size);
+	addrp = of_get_pci_address(node, 0, &size, NULL);
+	if (addrp == NULL) {
+		printk(KERN_ERR "pmac_feature: %s: can't find base !\n",
+		       node->full_name);
+		return;
+	}
+	addr = of_translate_address(node, addrp);
+	if (addr == 0) {
+		printk(KERN_ERR "pmac_feature: %s, can't translate base !\n",
+		       node->full_name);
+		return;
+	}
+	base = ioremap(addr, (unsigned long)size);
 	if (!base) {
-		printk(KERN_ERR "pmac_feature: Can't map mac-io chip !\n");
+		printk(KERN_ERR "pmac_feature: %s, can't map mac-io chip !\n",
+		       node->full_name);
 		return;
 	}
-	if (type == macio_keylargo) {
+	if (type == macio_keylargo || type == macio_keylargo2) {
 		u32 *did = (u32 *)get_property(node, "device-id", NULL);
 		if (*did == 0x00000025)
 			type = macio_pangea;
 		if (*did == 0x0000003e)
 			type = macio_intrepid;
+		if (*did == 0x0000004f)
+			type = macio_shasta;
 	}
 	macio_chips[i].of_node	= node;
 	macio_chips[i].type	= type;
@@ -2787,7 +2748,8 @@ set_initial_features(void)
 	}
 
 #ifdef CONFIG_POWER4
-	if (macio_chips[0].type == macio_keylargo2) {
+	if (macio_chips[0].type == macio_keylargo2 ||
+	    macio_chips[0].type == macio_shasta) {
 #ifndef CONFIG_SMP
 		/* On SMP machines running UP, we have the second CPU eating
 		 * bus cycles. We need to take it off the bus. This is done
@@ -2896,12 +2858,6 @@ set_initial_features(void)
 		MACIO_BIC(HEATHROW_FCR, HRW_SOUND_POWER_N);
 	}
 
-	/* Some machine models need the clock chip to be properly setup for
-	 * clock spreading now. This should be a platform function but we
-	 * don't do these at the moment
-	 */
-	pmac_tweak_clock_spreading(1);
-
 #endif /* CONFIG_POWER4 */
 
 	/* On all machines, switch modem & serial ports off */
@@ -2929,9 +2885,6 @@ pmac_feature_init(void)
 		return;
 	}
 
-	/* Setup low-level i2c stuffs */
-	pmac_init_low_i2c();
-
 	/* Probe machine type */
 	if (probe_motherboard())
 		printk(KERN_WARNING "Unknown PowerMac !\n");
@@ -2942,26 +2895,6 @@ pmac_feature_init(void)
 	set_initial_features();
 }
 
-int __init pmac_feature_late_init(void)
-{
-#if 0
-	struct device_node *np;
-
-	/* Request some resources late */
-	if (uninorth_node)
-		request_OF_resource(uninorth_node, 0, NULL);
-	np = find_devices("hammerhead");
-	if (np)
-		request_OF_resource(np, 0, NULL);
-	np = find_devices("interrupt-controller");
-	if (np)
-		request_OF_resource(np, 0, NULL);
-#endif
-	return 0;
-}
-
-device_initcall(pmac_feature_late_init);
-
 #if 0
 static void dump_HT_speeds(char *name, u32 cfg, u32 frq)
 {
@@ -2984,9 +2917,9 @@ void __init pmac_check_ht_link(void)
 	u8	px_bus, px_devfn;
 	struct pci_controller *px_hose;
 
-	(void)in_be32(u3_ht + U3_HT_LINK_COMMAND);
-	ucfg = cfg = in_be32(u3_ht + U3_HT_LINK_CONFIG);
-	ufreq = freq = in_be32(u3_ht + U3_HT_LINK_FREQ);
+	(void)in_be32(u3_ht_base + U3_HT_LINK_COMMAND);
+	ucfg = cfg = in_be32(u3_ht_base + U3_HT_LINK_CONFIG);
+	ufreq = freq = in_be32(u3_ht_base + U3_HT_LINK_FREQ);
 	dump_HT_speeds("U3 HyperTransport", cfg, freq);
 
 	pcix_node = of_find_compatible_node(NULL, "pci", "pci-x");
diff --git a/arch/powerpc/platforms/powermac/low_i2c.c b/arch/powerpc/platforms/powermac/low_i2c.c
index f3f39e8e337a..535c802b369f 100644
--- a/arch/powerpc/platforms/powermac/low_i2c.c
+++ b/arch/powerpc/platforms/powermac/low_i2c.c
@@ -1,22 +1,34 @@
 /*
- *  arch/ppc/platforms/pmac_low_i2c.c
+ * arch/powerpc/platforms/powermac/low_i2c.c
  *
- *  Copyright (C) 2003 Ben. Herrenschmidt (benh@kernel.crashing.org)
+ *  Copyright (C) 2003-2005 Ben. Herrenschmidt (benh@kernel.crashing.org)
  *
  *  This program is free software; you can redistribute it and/or
  *  modify it under the terms of the GNU General Public License
  *  as published by the Free Software Foundation; either version
  *  2 of the License, or (at your option) any later version.
  *
- *  This file contains some low-level i2c access routines that
- *  need to be used by various bits of the PowerMac platform code
- *  at times where the real asynchronous & interrupt driven driver
- *  cannot be used. The API borrows some semantics from the darwin
- *  driver in order to ease the implementation of the platform
- *  properties parser
+ * The linux i2c layer isn't completely suitable for our needs for various
+ * reasons ranging from too late initialisation to semantics not perfectly
+ * matching some requirements of the apple platform functions etc...
+ *
+ * This file thus provides a simple low level unified i2c interface for
+ * powermac that covers the various types of i2c busses used in Apple machines.
+ * For now, keywest, PMU and SMU, though we could add Cuda, or other bit
+ * banging busses found on older chipstes in earlier machines if we ever need
+ * one of them.
+ *
+ * The drivers in this file are synchronous/blocking. In addition, the
+ * keywest one is fairly slow due to the use of msleep instead of interrupts
+ * as the interrupt is currently used by i2c-keywest. In the long run, we
+ * might want to get rid of those high-level interfaces to linux i2c layer
+ * either completely (converting all drivers) or replacing them all with a
+ * single stub driver on top of this one. Once done, the interrupt will be
+ * available for our use.
  */
 
 #undef DEBUG
+#undef DEBUG_LOW
 
 #include <linux/config.h>
 #include <linux/types.h>
@@ -25,66 +37,91 @@
 #include <linux/module.h>
 #include <linux/adb.h>
 #include <linux/pmu.h>
+#include <linux/delay.h>
+#include <linux/completion.h>
+#include <linux/platform_device.h>
+#include <linux/interrupt.h>
+#include <linux/completion.h>
+#include <linux/timer.h>
 #include <asm/keylargo.h>
 #include <asm/uninorth.h>
 #include <asm/io.h>
 #include <asm/prom.h>
 #include <asm/machdep.h>
+#include <asm/smu.h>
+#include <asm/pmac_pfunc.h>
 #include <asm/pmac_low_i2c.h>
 
-#define MAX_LOW_I2C_HOST	4
-
 #ifdef DEBUG
 #define DBG(x...) do {\
-		printk(KERN_DEBUG "KW:" x);	\
+		printk(KERN_DEBUG "low_i2c:" x);	\
 	} while(0)
 #else
 #define DBG(x...)
 #endif
 
-struct low_i2c_host;
-
-typedef int (*low_i2c_func_t)(struct low_i2c_host *host, u8 addr, u8 sub, u8 *data, int len);
+#ifdef DEBUG_LOW
+#define DBG_LOW(x...) do {\
+		printk(KERN_DEBUG "low_i2c:" x);	\
+	} while(0)
+#else
+#define DBG_LOW(x...)
+#endif
 
-struct low_i2c_host
-{
-	struct device_node	*np;		/* OF device node */
-	struct semaphore	mutex;		/* Access mutex for use by i2c-keywest */
-	low_i2c_func_t		func;		/* Access function */
-	unsigned int		is_open : 1;	/* Poor man's access control */
-	int			mode;		/* Current mode */
-	int			channel;	/* Current channel */
-	int			num_channels;	/* Number of channels */
-	void __iomem		*base;		/* For keywest-i2c, base address */
-	int			bsteps;		/* And register stepping */
-	int			speed;		/* And speed */
-};
 
-static struct low_i2c_host	low_i2c_hosts[MAX_LOW_I2C_HOST];
+static int pmac_i2c_force_poll = 1;
 
-/* No locking is necessary on allocation, we are running way before
- * anything can race with us
+/*
+ * A bus structure. Each bus in the system has such a structure associated.
  */
-static struct low_i2c_host *find_low_i2c_host(struct device_node *np)
+struct pmac_i2c_bus
 {
-	int i;
+	struct list_head	link;
+	struct device_node	*controller;
+	struct device_node	*busnode;
+	int			type;
+	int			flags;
+	struct i2c_adapter	*adapter;
+	void			*hostdata;
+	int			channel;	/* some hosts have multiple */
+	int			mode;		/* current mode */
+	struct semaphore	sem;
+	int			opened;
+	int			polled;		/* open mode */
+	struct platform_device	*platform_dev;
+
+	/* ops */
+	int (*open)(struct pmac_i2c_bus *bus);
+	void (*close)(struct pmac_i2c_bus *bus);
+	int (*xfer)(struct pmac_i2c_bus *bus, u8 addrdir, int subsize,
+		    u32 subaddr, u8 *data, int len);
+};
 
-	for (i = 0; i < MAX_LOW_I2C_HOST; i++)
-		if (low_i2c_hosts[i].np == np)
-			return &low_i2c_hosts[i];
-	return NULL;
-}
+static LIST_HEAD(pmac_i2c_busses);
 
 /*
- *
- * i2c-keywest implementation (UniNorth, U2, U3, Keylargo's)
- *
+ * Keywest implementation
  */
 
-/*
- * Keywest i2c definitions borrowed from drivers/i2c/i2c-keywest.h,
- * should be moved somewhere in include/asm-ppc/
- */
+struct pmac_i2c_host_kw
+{
+	struct semaphore	mutex;		/* Access mutex for use by
+						 * i2c-keywest */
+	void __iomem		*base;		/* register base address */
+	int			bsteps;		/* register stepping */
+	int			speed;		/* speed */
+	int			irq;
+	u8			*data;
+	unsigned		len;
+	int			state;
+	int			rw;
+	int			polled;
+	int			result;
+	struct completion	complete;
+	spinlock_t		lock;
+	struct timer_list	timeout_timer;
+};
+
 /* Register indices */
 typedef enum {
 	reg_mode = 0,
@@ -97,6 +134,8 @@ typedef enum {
 	reg_data
 } reg_t;
 
+/* The Tumbler audio equalizer can be really slow sometimes */
+#define KW_POLL_TIMEOUT		(2*HZ)
 
 /* Mode register */
 #define KW_I2C_MODE_100KHZ	0x00
@@ -140,8 +179,9 @@ enum {
 };
 
 #define WRONG_STATE(name) do {\
-		printk(KERN_DEBUG "KW: wrong state. Got %s, state: %s (isr: %02x)\n", \
-		       name, __kw_state_names[state], isr); \
+		printk(KERN_DEBUG "KW: wrong state. Got %s, state: %s " \
+		       "(isr: %02x)\n",	\
+		       name, __kw_state_names[host->state], isr); \
 	} while(0)
 
 static const char *__kw_state_names[] = {
@@ -153,120 +193,137 @@ static const char *__kw_state_names[] = {
 	"state_dead"
 };
 
-static inline u8 __kw_read_reg(struct low_i2c_host *host, reg_t reg)
+static inline u8 __kw_read_reg(struct pmac_i2c_host_kw *host, reg_t reg)
 {
 	return readb(host->base + (((unsigned int)reg) << host->bsteps));
 }
 
-static inline void __kw_write_reg(struct low_i2c_host *host, reg_t reg, u8 val)
+static inline void __kw_write_reg(struct pmac_i2c_host_kw *host,
+				  reg_t reg, u8 val)
 {
 	writeb(val, host->base + (((unsigned)reg) << host->bsteps));
 	(void)__kw_read_reg(host, reg_subaddr);
 }
 
-#define kw_write_reg(reg, val)	__kw_write_reg(host, reg, val) 
-#define kw_read_reg(reg)	__kw_read_reg(host, reg) 
-
+#define kw_write_reg(reg, val)	__kw_write_reg(host, reg, val)
+#define kw_read_reg(reg)	__kw_read_reg(host, reg)
 
-/* Don't schedule, the g5 fan controller is too
- * timing sensitive
- */
-static u8 kw_wait_interrupt(struct low_i2c_host* host)
+static u8 kw_i2c_wait_interrupt(struct pmac_i2c_host_kw *host)
 {
 	int i, j;
 	u8 isr;
 	
-	for (i = 0; i < 100000; i++) {
+	for (i = 0; i < 1000; i++) {
 		isr = kw_read_reg(reg_isr) & KW_I2C_IRQ_MASK;
 		if (isr != 0)
 			return isr;
 
 		/* This code is used with the timebase frozen, we cannot rely
-		 * on udelay ! For now, just use a bogus loop
+		 * on udelay nor schedule when in polled mode !
+		 * For now, just use a bogus loop....
 		 */
-		for (j = 1; j < 10000; j++)
-			mb();
+		if (host->polled) {
+			for (j = 1; j < 100000; j++)
+				mb();
+		} else
+			msleep(1);
 	}
 	return isr;
 }
 
-static int kw_handle_interrupt(struct low_i2c_host *host, int state, int rw, int *rc, u8 **data, int *len, u8 isr)
+static void kw_i2c_handle_interrupt(struct pmac_i2c_host_kw *host, u8 isr)
 {
 	u8 ack;
 
-	DBG("kw_handle_interrupt(%s, isr: %x)\n", __kw_state_names[state], isr);
+	DBG_LOW("kw_handle_interrupt(%s, isr: %x)\n",
+		__kw_state_names[host->state], isr);
+
+	if (host->state == state_idle) {
+		printk(KERN_WARNING "low_i2c: Keywest got an out of state"
+		       " interrupt, ignoring\n");
+		kw_write_reg(reg_isr, isr);
+		return;
+	}
 
 	if (isr == 0) {
-		if (state != state_stop) {
-			DBG("KW: Timeout !\n");
-			*rc = -EIO;
+		if (host->state != state_stop) {
+			DBG_LOW("KW: Timeout !\n");
+			host->result = -EIO;
 			goto stop;
 		}
-		if (state == state_stop) {
+		if (host->state == state_stop) {
 			ack = kw_read_reg(reg_status);
-			if (!(ack & KW_I2C_STAT_BUSY)) {
-				state = state_idle;
-				kw_write_reg(reg_ier, 0x00);
-			}
+			if (ack & KW_I2C_STAT_BUSY)
+				kw_write_reg(reg_status, 0);
+			host->state = state_idle;
+			kw_write_reg(reg_ier, 0x00);
+			if (!host->polled)
+				complete(&host->complete);
 		}
-		return state;
+		return;
 	}
 
 	if (isr & KW_I2C_IRQ_ADDR) {
 		ack = kw_read_reg(reg_status);
-		if (state != state_addr) {
+		if (host->state != state_addr) {
 			kw_write_reg(reg_isr, KW_I2C_IRQ_ADDR);
 			WRONG_STATE("KW_I2C_IRQ_ADDR"); 
-			*rc = -EIO;
+			host->result = -EIO;
 			goto stop;
 		}
-		if ((ack & KW_I2C_STAT_LAST_AAK) == 0) {			
-			*rc = -ENODEV;
-			DBG("KW: NAK on address\n");
-			return state_stop;		     
+		if ((ack & KW_I2C_STAT_LAST_AAK) == 0) {
+			host->result = -ENODEV;
+			DBG_LOW("KW: NAK on address\n");
+			host->state = state_stop;
+			return;
 		} else {
-			if (rw) {
-				state = state_read;
-				if (*len > 1)
-					kw_write_reg(reg_control, KW_I2C_CTL_AAK);
+			if (host->len == 0) {
+				kw_write_reg(reg_isr, KW_I2C_IRQ_ADDR);
+				goto stop;
+			}
+			if (host->rw) {
+				host->state = state_read;
+				if (host->len > 1)
+					kw_write_reg(reg_control,
+						     KW_I2C_CTL_AAK);
 			} else {
-				state = state_write;
-				kw_write_reg(reg_data, **data);
-				(*data)++; (*len)--;
+				host->state = state_write;
+				kw_write_reg(reg_data, *(host->data++));
+				host->len--;
 			}
 		}
 		kw_write_reg(reg_isr, KW_I2C_IRQ_ADDR);
 	}
 
 	if (isr & KW_I2C_IRQ_DATA) {
-		if (state == state_read) {
-			**data = kw_read_reg(reg_data);
-			(*data)++; (*len)--;
+		if (host->state == state_read) {
+			*(host->data++) = kw_read_reg(reg_data);
+			host->len--;
 			kw_write_reg(reg_isr, KW_I2C_IRQ_DATA);
-			if ((*len) == 0)
-				state = state_stop;
-			else if ((*len) == 1)
+			if (host->len == 0)
+				host->state = state_stop;
+			else if (host->len == 1)
 				kw_write_reg(reg_control, 0);
-		} else if (state == state_write) {
+		} else if (host->state == state_write) {
 			ack = kw_read_reg(reg_status);
 			if ((ack & KW_I2C_STAT_LAST_AAK) == 0) {
-				DBG("KW: nack on data write\n");
-				*rc = -EIO;
+				DBG_LOW("KW: nack on data write\n");
+				host->result = -EIO;
 				goto stop;
-			} else if (*len) {
-				kw_write_reg(reg_data, **data);
-				(*data)++; (*len)--;
+			} else if (host->len) {
+				kw_write_reg(reg_data, *(host->data++));
+				host->len--;
 			} else {
 				kw_write_reg(reg_control, KW_I2C_CTL_STOP);
-				state = state_stop;
-				*rc = 0;
+				host->state = state_stop;
+				host->result = 0;
 			}
 			kw_write_reg(reg_isr, KW_I2C_IRQ_DATA);
 		} else {
 			kw_write_reg(reg_isr, KW_I2C_IRQ_DATA);
 			WRONG_STATE("KW_I2C_IRQ_DATA"); 
-			if (state != state_stop) {
-				*rc = -EIO;
+			if (host->state != state_stop) {
+				host->result = -EIO;
 				goto stop;
 			}
 		}
@@ -274,98 +331,194 @@ static int kw_handle_interrupt(struct low_i2c_host *host, int state, int rw, int
 
 	if (isr & KW_I2C_IRQ_STOP) {
 		kw_write_reg(reg_isr, KW_I2C_IRQ_STOP);
-		if (state != state_stop) {
+		if (host->state != state_stop) {
 			WRONG_STATE("KW_I2C_IRQ_STOP");
-			*rc = -EIO;
+			host->result = -EIO;
 		}
-		return state_idle;
+		host->state = state_idle;
+		if (!host->polled)
+			complete(&host->complete);
 	}
 
 	if (isr & KW_I2C_IRQ_START)
 		kw_write_reg(reg_isr, KW_I2C_IRQ_START);
 
-	return state;
-
+	return;
  stop:
 	kw_write_reg(reg_control, KW_I2C_CTL_STOP);	
-	return state_stop;
+	host->state = state_stop;
+	return;
 }
 
-static int keywest_low_i2c_func(struct low_i2c_host *host, u8 addr, u8 subaddr, u8 *data, int len)
+/* Interrupt handler */
+static irqreturn_t kw_i2c_irq(int irq, void *dev_id, struct pt_regs *regs)
 {
+	struct pmac_i2c_host_kw *host = dev_id;
+	unsigned long flags;
+
+	spin_lock_irqsave(&host->lock, flags);
+	del_timer(&host->timeout_timer);
+	kw_i2c_handle_interrupt(host, kw_read_reg(reg_isr));
+	if (host->state != state_idle) {
+		host->timeout_timer.expires = jiffies + KW_POLL_TIMEOUT;
+		add_timer(&host->timeout_timer);
+	}
+	spin_unlock_irqrestore(&host->lock, flags);
+	return IRQ_HANDLED;
+}
+
+static void kw_i2c_timeout(unsigned long data)
+{
+	struct pmac_i2c_host_kw *host = (struct pmac_i2c_host_kw *)data;
+	unsigned long flags;
+
+	spin_lock_irqsave(&host->lock, flags);
+	kw_i2c_handle_interrupt(host, kw_read_reg(reg_isr));
+	if (host->state != state_idle) {
+		host->timeout_timer.expires = jiffies + KW_POLL_TIMEOUT;
+		add_timer(&host->timeout_timer);
+	}
+	spin_unlock_irqrestore(&host->lock, flags);
+}
+
+static int kw_i2c_open(struct pmac_i2c_bus *bus)
+{
+	struct pmac_i2c_host_kw *host = bus->hostdata;
+	down(&host->mutex);
+	return 0;
+}
+
+static void kw_i2c_close(struct pmac_i2c_bus *bus)
+{
+	struct pmac_i2c_host_kw *host = bus->hostdata;
+	up(&host->mutex);
+}
+
+static int kw_i2c_xfer(struct pmac_i2c_bus *bus, u8 addrdir, int subsize,
+		       u32 subaddr, u8 *data, int len)
+{
+	struct pmac_i2c_host_kw *host = bus->hostdata;
 	u8 mode_reg = host->speed;
-	int state = state_addr;
-	int rc = 0;
+	int use_irq = host->irq != NO_IRQ && !bus->polled;
 
 	/* Setup mode & subaddress if any */
-	switch(host->mode) {
-	case pmac_low_i2c_mode_dumb:
-		printk(KERN_ERR "low_i2c: Dumb mode not supported !\n");
+	switch(bus->mode) {
+	case pmac_i2c_mode_dumb:
 		return -EINVAL;
-	case pmac_low_i2c_mode_std:
+	case pmac_i2c_mode_std:
 		mode_reg |= KW_I2C_MODE_STANDARD;
+		if (subsize != 0)
+			return -EINVAL;
 		break;
-	case pmac_low_i2c_mode_stdsub:
+	case pmac_i2c_mode_stdsub:
 		mode_reg |= KW_I2C_MODE_STANDARDSUB;
+		if (subsize != 1)
+			return -EINVAL;
 		break;
-	case pmac_low_i2c_mode_combined:
+	case pmac_i2c_mode_combined:
 		mode_reg |= KW_I2C_MODE_COMBINED;
+		if (subsize != 1)
+			return -EINVAL;
 		break;
 	}
 
 	/* Setup channel & clear pending irqs */
 	kw_write_reg(reg_isr, kw_read_reg(reg_isr));
-	kw_write_reg(reg_mode, mode_reg | (host->channel << 4));
+	kw_write_reg(reg_mode, mode_reg | (bus->channel << 4));
 	kw_write_reg(reg_status, 0);
 
-	/* Set up address and r/w bit */
-	kw_write_reg(reg_addr, addr);
+	/* Set up address and r/w bit, strip possible stale bus number from
+	 * address top bits
+	 */
+	kw_write_reg(reg_addr, addrdir & 0xff);
 
 	/* Set up the sub address */
 	if ((mode_reg & KW_I2C_MODE_MODE_MASK) == KW_I2C_MODE_STANDARDSUB
 	    || (mode_reg & KW_I2C_MODE_MODE_MASK) == KW_I2C_MODE_COMBINED)
 		kw_write_reg(reg_subaddr, subaddr);
 
-	/* Start sending address & disable interrupt*/
-	kw_write_reg(reg_ier, 0 /*KW_I2C_IRQ_MASK*/);
+	/* Prepare for async operations */
+	host->data = data;
+	host->len = len;
+	host->state = state_addr;
+	host->result = 0;
+	host->rw = (addrdir & 1);
+	host->polled = bus->polled;
+
+	/* Enable interrupt if not using polled mode and interrupt is
+	 * available
+	 */
+	if (use_irq) {
+		/* Clear completion */
+		INIT_COMPLETION(host->complete);
+		/* Ack stale interrupts */
+		kw_write_reg(reg_isr, kw_read_reg(reg_isr));
+		/* Arm timeout */
+		host->timeout_timer.expires = jiffies + KW_POLL_TIMEOUT;
+		add_timer(&host->timeout_timer);
+		/* Enable emission */
+		kw_write_reg(reg_ier, KW_I2C_IRQ_MASK);
+	}
+
+	/* Start sending address */
 	kw_write_reg(reg_control, KW_I2C_CTL_XADDR);
 
-	/* State machine, to turn into an interrupt handler */
-	while(state != state_idle) {
-		u8 isr = kw_wait_interrupt(host);
-		state = kw_handle_interrupt(host, state, addr & 1, &rc, &data, &len, isr);
+	/* Wait for completion */
+	if (use_irq)
+		wait_for_completion(&host->complete);
+	else {
+		while(host->state != state_idle) {
+			unsigned long flags;
+
+			u8 isr = kw_i2c_wait_interrupt(host);
+			spin_lock_irqsave(&host->lock, flags);
+			kw_i2c_handle_interrupt(host, isr);
+			spin_unlock_irqrestore(&host->lock, flags);
+		}
 	}
 
-	return rc;
+	/* Disable emission */
+	kw_write_reg(reg_ier, 0);
+
+	return host->result;
 }
 
-static void keywest_low_i2c_add(struct device_node *np)
+static struct pmac_i2c_host_kw *__init kw_i2c_host_init(struct device_node *np)
 {
-	struct low_i2c_host	*host = find_low_i2c_host(NULL);
-	u32			*psteps, *prate, steps, aoffset = 0;
-	struct device_node	*parent;
+	struct pmac_i2c_host_kw *host;
+	u32			*psteps, *prate, *addrp, steps;
 
+	host = kzalloc(sizeof(struct pmac_i2c_host_kw), GFP_KERNEL);
 	if (host == NULL) {
 		printk(KERN_ERR "low_i2c: Can't allocate host for %s\n",
 		       np->full_name);
-		return;
+		return NULL;
 	}
-	memset(host, 0, sizeof(*host));
 
+	/* Apple is kind enough to provide a valid AAPL,address property
+	 * on all i2c keywest nodes so far ... we would have to fallback
+	 * to macio parsing if that wasn't the case
+	 */
+	addrp = (u32 *)get_property(np, "AAPL,address", NULL);
+	if (addrp == NULL) {
+		printk(KERN_ERR "low_i2c: Can't find address for %s\n",
+		       np->full_name);
+		kfree(host);
+		return NULL;
+	}
 	init_MUTEX(&host->mutex);
-	host->np = of_node_get(np);	
+	init_completion(&host->complete);
+	spin_lock_init(&host->lock);
+	init_timer(&host->timeout_timer);
+	host->timeout_timer.function = kw_i2c_timeout;
+	host->timeout_timer.data = (unsigned long)host;
+
 	psteps = (u32 *)get_property(np, "AAPL,address-step", NULL);
 	steps = psteps ? (*psteps) : 0x10;
 	for (host->bsteps = 0; (steps & 0x01) == 0; host->bsteps++)
 		steps >>= 1;
-	parent = of_get_parent(np);
-	host->num_channels = 1;
-	if (parent && parent->name[0] == 'u') {
-		host->num_channels = 2;
-		aoffset = 3;
-	}
 	/* Select interface rate */
-	host->speed = KW_I2C_MODE_100KHZ;
+	host->speed = KW_I2C_MODE_25KHZ;
 	prate = (u32 *)get_property(np, "AAPL,i2c-rate", NULL);
 	if (prate) switch(*prate) {
 	case 100:
@@ -378,146 +531,981 @@ static void keywest_low_i2c_add(struct device_node *np)
 		host->speed = KW_I2C_MODE_25KHZ;
 		break;
 	}	
+	if (np->n_intrs > 0)
+		host->irq = np->intrs[0].line;
+	else
+		host->irq = NO_IRQ;
+
+	host->base = ioremap((*addrp), 0x1000);
+	if (host->base == NULL) {
+		printk(KERN_ERR "low_i2c: Can't map registers for %s\n",
+		       np->full_name);
+		kfree(host);
+		return NULL;
+	}
+
+	/* Make sure IRA is disabled */
+	kw_write_reg(reg_ier, 0);
+
+	/* Request chip interrupt */
+	if (request_irq(host->irq, kw_i2c_irq, SA_SHIRQ, "keywest i2c", host))
+		host->irq = NO_IRQ;
+
+	printk(KERN_INFO "KeyWest i2c @0x%08x irq %d %s\n",
+	       *addrp, host->irq, np->full_name);
 
-	host->mode = pmac_low_i2c_mode_std;
-	host->base = ioremap(np->addrs[0].address + aoffset,
-						np->addrs[0].size);
-	host->func = keywest_low_i2c_func;
+	return host;
 }
 
+
+static void __init kw_i2c_add(struct pmac_i2c_host_kw *host,
+			      struct device_node *controller,
+			      struct device_node *busnode,
+			      int channel)
+{
+	struct pmac_i2c_bus *bus;
+
+	bus = kzalloc(sizeof(struct pmac_i2c_bus), GFP_KERNEL);
+	if (bus == NULL)
+		return;
+
+	bus->controller = of_node_get(controller);
+	bus->busnode = of_node_get(busnode);
+	bus->type = pmac_i2c_bus_keywest;
+	bus->hostdata = host;
+	bus->channel = channel;
+	bus->mode = pmac_i2c_mode_std;
+	bus->open = kw_i2c_open;
+	bus->close = kw_i2c_close;
+	bus->xfer = kw_i2c_xfer;
+	init_MUTEX(&bus->sem);
+	if (controller == busnode)
+		bus->flags = pmac_i2c_multibus;
+	list_add(&bus->link, &pmac_i2c_busses);
+
+	printk(KERN_INFO " channel %d bus %s\n", channel,
+	       (controller == busnode) ? "<multibus>" : busnode->full_name);
+}
+
+static void __init kw_i2c_probe(void)
+{
+	struct device_node *np, *child, *parent;
+
+	/* Probe keywest-i2c busses */
+	for (np = NULL;
+	     (np = of_find_compatible_node(np, "i2c","keywest-i2c")) != NULL;){
+		struct pmac_i2c_host_kw *host;
+		int multibus, chans, i;
+
+		/* Found one, init a host structure */
+		host = kw_i2c_host_init(np);
+		if (host == NULL)
+			continue;
+
+		/* Now check if we have a multibus setup (old style) or if we
+		 * have proper bus nodes. Note that the "new" way (proper bus
+		 * nodes) might cause us to not create some busses that are
+		 * kept hidden in the device-tree. In the future, we might
+		 * want to work around that by creating busses without a node
+		 * but not for now
+		 */
+		child = of_get_next_child(np, NULL);
+		multibus = !child || strcmp(child->name, "i2c-bus");
+		of_node_put(child);
+
+		/* For a multibus setup, we get the bus count based on the
+		 * parent type
+		 */
+		if (multibus) {
+			parent = of_get_parent(np);
+			if (parent == NULL)
+				continue;
+			chans = parent->name[0] == 'u' ? 2 : 1;
+			for (i = 0; i < chans; i++)
+				kw_i2c_add(host, np, np, i);
+		} else {
+			for (child = NULL;
+			     (child = of_get_next_child(np, child)) != NULL;) {
+				u32 *reg =
+					(u32 *)get_property(child, "reg", NULL);
+				if (reg == NULL)
+					continue;
+				kw_i2c_add(host, np, child, *reg);
+			}
+		}
+	}
+}
+
+
 /*
  *
  * PMU implementation
  *
  */
 
-
 #ifdef CONFIG_ADB_PMU
 
-static int pmu_low_i2c_func(struct low_i2c_host *host, u8 addr, u8 sub, u8 *data, int len)
+/*
+ * i2c command block to the PMU
+ */
+struct pmu_i2c_hdr {
+	u8	bus;
+	u8	mode;
+	u8	bus2;
+	u8	address;
+	u8	sub_addr;
+	u8	comb_addr;
+	u8	count;
+	u8	data[];
+};
+
+static void pmu_i2c_complete(struct adb_request *req)
 {
-	// TODO
-	return -ENODEV;
+	complete(req->arg);
 }
 
-static void pmu_low_i2c_add(struct device_node *np)
+static int pmu_i2c_xfer(struct pmac_i2c_bus *bus, u8 addrdir, int subsize,
+			u32 subaddr, u8 *data, int len)
 {
-	struct low_i2c_host	*host = find_low_i2c_host(NULL);
+	struct adb_request *req = bus->hostdata;
+	struct pmu_i2c_hdr *hdr = (struct pmu_i2c_hdr *)&req->data[1];
+	struct completion comp;
+	int read = addrdir & 1;
+	int retry;
+	int rc = 0;
 
-	if (host == NULL) {
-		printk(KERN_ERR "low_i2c: Can't allocate host for %s\n",
-		       np->full_name);
-		return;
+	/* For now, limit ourselves to 16 bytes transfers */
+	if (len > 16)
+		return -EINVAL;
+
+	init_completion(&comp);
+
+	for (retry = 0; retry < 16; retry++) {
+		memset(req, 0, sizeof(struct adb_request));
+		hdr->bus = bus->channel;
+		hdr->count = len;
+
+		switch(bus->mode) {
+		case pmac_i2c_mode_std:
+			if (subsize != 0)
+				return -EINVAL;
+			hdr->address = addrdir;
+			hdr->mode = PMU_I2C_MODE_SIMPLE;
+			break;
+		case pmac_i2c_mode_stdsub:
+		case pmac_i2c_mode_combined:
+			if (subsize != 1)
+				return -EINVAL;
+			hdr->address = addrdir & 0xfe;
+			hdr->comb_addr = addrdir;
+			hdr->sub_addr = subaddr;
+			if (bus->mode == pmac_i2c_mode_stdsub)
+				hdr->mode = PMU_I2C_MODE_STDSUB;
+			else
+				hdr->mode = PMU_I2C_MODE_COMBINED;
+			break;
+		default:
+			return -EINVAL;
+		}
+
+		INIT_COMPLETION(comp);
+		req->data[0] = PMU_I2C_CMD;
+		req->reply[0] = 0xff;
+		req->nbytes = sizeof(struct pmu_i2c_hdr) + 1;
+		req->done = pmu_i2c_complete;
+		req->arg = &comp;
+		if (!read && len) {
+			memcpy(hdr->data, data, len);
+			req->nbytes += len;
+		}
+		rc = pmu_queue_request(req);
+		if (rc)
+			return rc;
+		wait_for_completion(&comp);
+		if (req->reply[0] == PMU_I2C_STATUS_OK)
+			break;
+		msleep(15);
 	}
-	memset(host, 0, sizeof(*host));
+	if (req->reply[0] != PMU_I2C_STATUS_OK)
+		return -EIO;
 
-	init_MUTEX(&host->mutex);
-	host->np = of_node_get(np);	
-	host->num_channels = 3;
-	host->mode = pmac_low_i2c_mode_std;
-	host->func = pmu_low_i2c_func;
+	for (retry = 0; retry < 16; retry++) {
+		memset(req, 0, sizeof(struct adb_request));
+
+		/* I know that looks like a lot, slow as hell, but darwin
+		 * does it so let's be on the safe side for now
+		 */
+		msleep(15);
+
+		hdr->bus = PMU_I2C_BUS_STATUS;
+
+		INIT_COMPLETION(comp);
+		req->data[0] = PMU_I2C_CMD;
+		req->reply[0] = 0xff;
+		req->nbytes = 2;
+		req->done = pmu_i2c_complete;
+		req->arg = &comp;
+		rc = pmu_queue_request(req);
+		if (rc)
+			return rc;
+		wait_for_completion(&comp);
+
+		if (req->reply[0] == PMU_I2C_STATUS_OK && !read)
+			return 0;
+		if (req->reply[0] == PMU_I2C_STATUS_DATAREAD && read) {
+			int rlen = req->reply_len - 1;
+
+			if (rlen != len) {
+				printk(KERN_WARNING "low_i2c: PMU returned %d"
+				       " bytes, expected %d !\n", rlen, len);
+				return -EIO;
+			}
+			if (len)
+				memcpy(data, &req->reply[1], len);
+			return 0;
+		}
+	}
+	return -EIO;
+}
+
+static void __init pmu_i2c_probe(void)
+{
+	struct pmac_i2c_bus *bus;
+	struct device_node *busnode;
+	int channel, sz;
+
+	if (!pmu_present())
+		return;
+
+	/* There might or might not be a "pmu-i2c" node, we use that
+	 * or via-pmu itself, whatever we find. I haven't seen a machine
+	 * with separate bus nodes, so we assume a multibus setup
+	 */
+	busnode = of_find_node_by_name(NULL, "pmu-i2c");
+	if (busnode == NULL)
+		busnode = of_find_node_by_name(NULL, "via-pmu");
+	if (busnode == NULL)
+		return;
+
+	printk(KERN_INFO "PMU i2c %s\n", busnode->full_name);
+
+	/*
+	 * We add bus 1 and 2 only for now, bus 0 is "special"
+	 */
+	for (channel = 1; channel <= 2; channel++) {
+		sz = sizeof(struct pmac_i2c_bus) + sizeof(struct adb_request);
+		bus = kzalloc(sz, GFP_KERNEL);
+		if (bus == NULL)
+			return;
+
+		bus->controller = busnode;
+		bus->busnode = busnode;
+		bus->type = pmac_i2c_bus_pmu;
+		bus->channel = channel;
+		bus->mode = pmac_i2c_mode_std;
+		bus->hostdata = bus + 1;
+		bus->xfer = pmu_i2c_xfer;
+		init_MUTEX(&bus->sem);
+		bus->flags = pmac_i2c_multibus;
+		list_add(&bus->link, &pmac_i2c_busses);
+
+		printk(KERN_INFO " channel %d bus <multibus>\n", channel);
+	}
 }
 
 #endif /* CONFIG_ADB_PMU */
 
-void __init pmac_init_low_i2c(void)
+
+/*
+ *
+ * SMU implementation
+ *
+ */
+
+#ifdef CONFIG_PMAC_SMU
+
+static void smu_i2c_complete(struct smu_i2c_cmd *cmd, void *misc)
 {
-	struct device_node *np;
+	complete(misc);
+}
 
-	/* Probe keywest-i2c busses */
-	np = of_find_compatible_node(NULL, "i2c", "keywest-i2c");
-	while(np) {
-		keywest_low_i2c_add(np);
-		np = of_find_compatible_node(np, "i2c", "keywest-i2c");
+static int smu_i2c_xfer(struct pmac_i2c_bus *bus, u8 addrdir, int subsize,
+			u32 subaddr, u8 *data, int len)
+{
+	struct smu_i2c_cmd *cmd = bus->hostdata;
+	struct completion comp;
+	int read = addrdir & 1;
+	int rc = 0;
+
+	if ((read && len > SMU_I2C_READ_MAX) ||
+	    ((!read) && len > SMU_I2C_WRITE_MAX))
+		return -EINVAL;
+
+	memset(cmd, 0, sizeof(struct smu_i2c_cmd));
+	cmd->info.bus = bus->channel;
+	cmd->info.devaddr = addrdir;
+	cmd->info.datalen = len;
+
+	switch(bus->mode) {
+	case pmac_i2c_mode_std:
+		if (subsize != 0)
+			return -EINVAL;
+		cmd->info.type = SMU_I2C_TRANSFER_SIMPLE;
+		break;
+	case pmac_i2c_mode_stdsub:
+	case pmac_i2c_mode_combined:
+		if (subsize > 3 || subsize < 1)
+			return -EINVAL;
+		cmd->info.sublen = subsize;
+		/* that's big-endian only but heh ! */
+		memcpy(&cmd->info.subaddr, ((char *)&subaddr) + (4 - subsize),
+		       subsize);
+		if (bus->mode == pmac_i2c_mode_stdsub)
+			cmd->info.type = SMU_I2C_TRANSFER_STDSUB;
+		else
+			cmd->info.type = SMU_I2C_TRANSFER_COMBINED;
+		break;
+	default:
+		return -EINVAL;
 	}
+	if (!read && len)
+		memcpy(cmd->info.data, data, len);
+
+	init_completion(&comp);
+	cmd->done = smu_i2c_complete;
+	cmd->misc = &comp;
+	rc = smu_queue_i2c(cmd);
+	if (rc < 0)
+		return rc;
+	wait_for_completion(&comp);
+	rc = cmd->status;
+
+	if (read && len)
+		memcpy(data, cmd->info.data, len);
+	return rc < 0 ? rc : 0;
+}
+
+static void __init smu_i2c_probe(void)
+{
+	struct device_node *controller, *busnode;
+	struct pmac_i2c_bus *bus;
+	u32 *reg;
+	int sz;
+
+	if (!smu_present())
+		return;
+
+	controller = of_find_node_by_name(NULL, "smu-i2c-control");
+	if (controller == NULL)
+		controller = of_find_node_by_name(NULL, "smu");
+	if (controller == NULL)
+		return;
+
+	printk(KERN_INFO "SMU i2c %s\n", controller->full_name);
+
+	/* Look for childs, note that they might not be of the right
+	 * type as older device trees mix i2c busses and other thigns
+	 * at the same level
+	 */
+	for (busnode = NULL;
+	     (busnode = of_get_next_child(controller, busnode)) != NULL;) {
+		if (strcmp(busnode->type, "i2c") &&
+		    strcmp(busnode->type, "i2c-bus"))
+			continue;
+		reg = (u32 *)get_property(busnode, "reg", NULL);
+		if (reg == NULL)
+			continue;
+
+		sz = sizeof(struct pmac_i2c_bus) + sizeof(struct smu_i2c_cmd);
+		bus = kzalloc(sz, GFP_KERNEL);
+		if (bus == NULL)
+			return;
+
+		bus->controller = controller;
+		bus->busnode = of_node_get(busnode);
+		bus->type = pmac_i2c_bus_smu;
+		bus->channel = *reg;
+		bus->mode = pmac_i2c_mode_std;
+		bus->hostdata = bus + 1;
+		bus->xfer = smu_i2c_xfer;
+		init_MUTEX(&bus->sem);
+		bus->flags = 0;
+		list_add(&bus->link, &pmac_i2c_busses);
+
+		printk(KERN_INFO " channel %x bus %s\n",
+		       bus->channel, busnode->full_name);
+	}
+}
+
+#endif /* CONFIG_PMAC_SMU */
+
+/*
+ *
+ * Core code
+ *
+ */
 
-#ifdef CONFIG_ADB_PMU
-	/* Probe PMU busses */
-	np = of_find_node_by_name(NULL, "via-pmu");
-	if (np)
-		pmu_low_i2c_add(np);
-#endif /* CONFIG_ADB_PMU */
 
-	/* TODO: Add CUDA support as well */
+struct pmac_i2c_bus *pmac_i2c_find_bus(struct device_node *node)
+{
+	struct device_node *p = of_node_get(node);
+	struct device_node *prev = NULL;
+	struct pmac_i2c_bus *bus;
+
+	while(p) {
+		list_for_each_entry(bus, &pmac_i2c_busses, link) {
+			if (p == bus->busnode) {
+				if (prev && bus->flags & pmac_i2c_multibus) {
+					u32 *reg;
+					reg = (u32 *)get_property(prev, "reg",
+								  NULL);
+					if (!reg)
+						continue;
+					if (((*reg) >> 8) != bus->channel)
+						continue;
+				}
+				of_node_put(p);
+				of_node_put(prev);
+				return bus;
+			}
+		}
+		of_node_put(prev);
+		prev = p;
+		p = of_get_parent(p);
+	}
+	return NULL;
 }
+EXPORT_SYMBOL_GPL(pmac_i2c_find_bus);
+
+u8 pmac_i2c_get_dev_addr(struct device_node *device)
+{
+	u32 *reg = (u32 *)get_property(device, "reg", NULL);
+
+	if (reg == NULL)
+		return 0;
+
+	return (*reg) & 0xff;
+}
+EXPORT_SYMBOL_GPL(pmac_i2c_get_dev_addr);
+
+struct device_node *pmac_i2c_get_controller(struct pmac_i2c_bus *bus)
+{
+	return bus->controller;
+}
+EXPORT_SYMBOL_GPL(pmac_i2c_get_controller);
+
+struct device_node *pmac_i2c_get_bus_node(struct pmac_i2c_bus *bus)
+{
+	return bus->busnode;
+}
+EXPORT_SYMBOL_GPL(pmac_i2c_get_bus_node);
+
+int pmac_i2c_get_type(struct pmac_i2c_bus *bus)
+{
+	return bus->type;
+}
+EXPORT_SYMBOL_GPL(pmac_i2c_get_type);
+
+int pmac_i2c_get_flags(struct pmac_i2c_bus *bus)
+{
+	return bus->flags;
+}
+EXPORT_SYMBOL_GPL(pmac_i2c_get_flags);
+
+int pmac_i2c_get_channel(struct pmac_i2c_bus *bus)
+{
+	return bus->channel;
+}
+EXPORT_SYMBOL_GPL(pmac_i2c_get_channel);
+
+
+void pmac_i2c_attach_adapter(struct pmac_i2c_bus *bus,
+			     struct i2c_adapter *adapter)
+{
+	WARN_ON(bus->adapter != NULL);
+	bus->adapter = adapter;
+}
+EXPORT_SYMBOL_GPL(pmac_i2c_attach_adapter);
+
+void pmac_i2c_detach_adapter(struct pmac_i2c_bus *bus,
+			     struct i2c_adapter *adapter)
+{
+	WARN_ON(bus->adapter != adapter);
+	bus->adapter = NULL;
+}
+EXPORT_SYMBOL_GPL(pmac_i2c_detach_adapter);
+
+struct i2c_adapter *pmac_i2c_get_adapter(struct pmac_i2c_bus *bus)
+{
+	return bus->adapter;
+}
+EXPORT_SYMBOL_GPL(pmac_i2c_get_adapter);
+
+struct pmac_i2c_bus *pmac_i2c_adapter_to_bus(struct i2c_adapter *adapter)
+{
+	struct pmac_i2c_bus *bus;
+
+	list_for_each_entry(bus, &pmac_i2c_busses, link)
+		if (bus->adapter == adapter)
+			return bus;
+	return NULL;
+}
+EXPORT_SYMBOL_GPL(pmac_i2c_adapter_to_bus);
+
+extern int pmac_i2c_match_adapter(struct device_node *dev,
+				  struct i2c_adapter *adapter)
+{
+	struct pmac_i2c_bus *bus = pmac_i2c_find_bus(dev);
+
+	if (bus == NULL)
+		return 0;
+	return (bus->adapter == adapter);
+}
+EXPORT_SYMBOL_GPL(pmac_i2c_match_adapter);
 
 int pmac_low_i2c_lock(struct device_node *np)
 {
-	struct low_i2c_host *host = find_low_i2c_host(np);
+	struct pmac_i2c_bus *bus, *found = NULL;
 
-	if (!host)
+	list_for_each_entry(bus, &pmac_i2c_busses, link) {
+		if (np == bus->controller) {
+			found = bus;
+			break;
+		}
+	}
+	if (!found)
 		return -ENODEV;
-	down(&host->mutex);
-	return 0;
+	return pmac_i2c_open(bus, 0);
 }
-EXPORT_SYMBOL(pmac_low_i2c_lock);
+EXPORT_SYMBOL_GPL(pmac_low_i2c_lock);
 
 int pmac_low_i2c_unlock(struct device_node *np)
 {
-	struct low_i2c_host *host = find_low_i2c_host(np);
+	struct pmac_i2c_bus *bus, *found = NULL;
 
-	if (!host)
+	list_for_each_entry(bus, &pmac_i2c_busses, link) {
+		if (np == bus->controller) {
+			found = bus;
+			break;
+		}
+	}
+	if (!found)
 		return -ENODEV;
-	up(&host->mutex);
+	pmac_i2c_close(bus);
 	return 0;
 }
-EXPORT_SYMBOL(pmac_low_i2c_unlock);
+EXPORT_SYMBOL_GPL(pmac_low_i2c_unlock);
 
 
-int pmac_low_i2c_open(struct device_node *np, int channel)
+int pmac_i2c_open(struct pmac_i2c_bus *bus, int polled)
 {
-	struct low_i2c_host *host = find_low_i2c_host(np);
+	int rc;
+
+	down(&bus->sem);
+	bus->polled = polled || pmac_i2c_force_poll;
+	bus->opened = 1;
+	bus->mode = pmac_i2c_mode_std;
+	if (bus->open && (rc = bus->open(bus)) != 0) {
+		bus->opened = 0;
+		up(&bus->sem);
+		return rc;
+	}
+	return 0;
+}
+EXPORT_SYMBOL_GPL(pmac_i2c_open);
 
-	if (!host)
-		return -ENODEV;
+void pmac_i2c_close(struct pmac_i2c_bus *bus)
+{
+	WARN_ON(!bus->opened);
+	if (bus->close)
+		bus->close(bus);
+	bus->opened = 0;
+	up(&bus->sem);
+}
+EXPORT_SYMBOL_GPL(pmac_i2c_close);
 
-	if (channel >= host->num_channels)
+int pmac_i2c_setmode(struct pmac_i2c_bus *bus, int mode)
+{
+	WARN_ON(!bus->opened);
+
+	/* Report me if you see the error below as there might be a new
+	 * "combined4" mode that I need to implement for the SMU bus
+	 */
+	if (mode < pmac_i2c_mode_dumb || mode > pmac_i2c_mode_combined) {
+		printk(KERN_ERR "low_i2c: Invalid mode %d requested on"
+		       " bus %s !\n", mode, bus->busnode->full_name);
 		return -EINVAL;
-
-	down(&host->mutex);
-	host->is_open = 1;
-	host->channel = channel;
+	}
+	bus->mode = mode;
 
 	return 0;
 }
-EXPORT_SYMBOL(pmac_low_i2c_open);
+EXPORT_SYMBOL_GPL(pmac_i2c_setmode);
 
-int pmac_low_i2c_close(struct device_node *np)
+int pmac_i2c_xfer(struct pmac_i2c_bus *bus, u8 addrdir, int subsize,
+		  u32 subaddr, u8 *data, int len)
 {
-	struct low_i2c_host *host = find_low_i2c_host(np);
+	int rc;
 
-	if (!host)
-		return -ENODEV;
+	WARN_ON(!bus->opened);
 
-	host->is_open = 0;
-	up(&host->mutex);
+	DBG("xfer() chan=%d, addrdir=0x%x, mode=%d, subsize=%d, subaddr=0x%x,"
+	    " %d bytes, bus %s\n", bus->channel, addrdir, bus->mode, subsize,
+	    subaddr, len, bus->busnode->full_name);
 
-	return 0;
+	rc = bus->xfer(bus, addrdir, subsize, subaddr, data, len);
+
+#ifdef DEBUG
+	if (rc)
+		DBG("xfer error %d\n", rc);
+#endif
+	return rc;
 }
-EXPORT_SYMBOL(pmac_low_i2c_close);
+EXPORT_SYMBOL_GPL(pmac_i2c_xfer);
+
+/* some quirks for platform function decoding */
+enum {
+	pmac_i2c_quirk_invmask = 0x00000001u,
+};
 
-int pmac_low_i2c_setmode(struct device_node *np, int mode)
+static void pmac_i2c_devscan(void (*callback)(struct device_node *dev,
+					      int quirks))
 {
-	struct low_i2c_host *host = find_low_i2c_host(np);
+	struct pmac_i2c_bus *bus;
+	struct device_node *np;
+	static struct whitelist_ent {
+		char *name;
+		char *compatible;
+		int quirks;
+	} whitelist[] = {
+		/* XXX Study device-tree's & apple drivers are get the quirks
+		 * right !
+		 */
+		{ "i2c-hwclock", NULL, pmac_i2c_quirk_invmask },
+		{ "i2c-cpu-voltage", NULL, 0},
+		{  "temp-monitor", NULL, 0 },
+		{  "supply-monitor", NULL, 0 },
+		{ NULL, NULL, 0 },
+	};
+
+	/* Only some devices need to have platform functions instanciated
+	 * here. For now, we have a table. Others, like 9554 i2c GPIOs used
+	 * on Xserve, if we ever do a driver for them, will use their own
+	 * platform function instance
+	 */
+	list_for_each_entry(bus, &pmac_i2c_busses, link) {
+		for (np = NULL;
+		     (np = of_get_next_child(bus->busnode, np)) != NULL;) {
+			struct whitelist_ent *p;
+			/* If multibus, check if device is on that bus */
+			if (bus->flags & pmac_i2c_multibus)
+				if (bus != pmac_i2c_find_bus(np))
+					continue;
+			for (p = whitelist; p->name != NULL; p++) {
+				if (strcmp(np->name, p->name))
+					continue;
+				if (p->compatible &&
+				    !device_is_compatible(np, p->compatible))
+					continue;
+				callback(np, p->quirks);
+				break;
+			}
+		}
+	}
+}
 
-	if (!host)
-		return -ENODEV;
-	WARN_ON(!host->is_open);
-	host->mode = mode;
+#define MAX_I2C_DATA	64
+
+struct pmac_i2c_pf_inst
+{
+	struct pmac_i2c_bus	*bus;
+	u8			addr;
+	u8			buffer[MAX_I2C_DATA];
+	u8			scratch[MAX_I2C_DATA];
+	int			bytes;
+	int			quirks;
+};
+
+static void* pmac_i2c_do_begin(struct pmf_function *func, struct pmf_args *args)
+{
+	struct pmac_i2c_pf_inst *inst;
+	struct pmac_i2c_bus	*bus;
+
+	bus = pmac_i2c_find_bus(func->node);
+	if (bus == NULL) {
+		printk(KERN_ERR "low_i2c: Can't find bus for %s (pfunc)\n",
+		       func->node->full_name);
+		return NULL;
+	}
+	if (pmac_i2c_open(bus, 0)) {
+		printk(KERN_ERR "low_i2c: Can't open i2c bus for %s (pfunc)\n",
+		       func->node->full_name);
+		return NULL;
+	}
+
+	/* XXX might need GFP_ATOMIC when called during the suspend process,
+	 * but then, there are already lots of issues with suspending when
+	 * near OOM that need to be resolved, the allocator itself should
+	 * probably make GFP_NOIO implicit during suspend
+	 */
+	inst = kzalloc(sizeof(struct pmac_i2c_pf_inst), GFP_KERNEL);
+	if (inst == NULL) {
+		pmac_i2c_close(bus);
+		return NULL;
+	}
+	inst->bus = bus;
+	inst->addr = pmac_i2c_get_dev_addr(func->node);
+	inst->quirks = (int)(long)func->driver_data;
+	return inst;
+}
+
+static void pmac_i2c_do_end(struct pmf_function *func, void *instdata)
+{
+	struct pmac_i2c_pf_inst *inst = instdata;
+
+	if (inst == NULL)
+		return;
+	pmac_i2c_close(inst->bus);
+	if (inst)
+		kfree(inst);
+}
+
+static int pmac_i2c_do_read(PMF_STD_ARGS, u32 len)
+{
+	struct pmac_i2c_pf_inst *inst = instdata;
+
+	inst->bytes = len;
+	return pmac_i2c_xfer(inst->bus, inst->addr | pmac_i2c_read, 0, 0,
+			     inst->buffer, len);
+}
+
+static int pmac_i2c_do_write(PMF_STD_ARGS, u32 len, const u8 *data)
+{
+	struct pmac_i2c_pf_inst *inst = instdata;
+
+	return pmac_i2c_xfer(inst->bus, inst->addr | pmac_i2c_write, 0, 0,
+			     (u8 *)data, len);
+}
+
+/* This function is used to do the masking & OR'ing for the "rmw" type
+ * callbacks. Ze should apply the mask and OR in the values in the
+ * buffer before writing back. The problem is that it seems that
+ * various darwin drivers implement the mask/or differently, thus
+ * we need to check the quirks first
+ */
+static void pmac_i2c_do_apply_rmw(struct pmac_i2c_pf_inst *inst,
+				  u32 len, const u8 *mask, const u8 *val)
+{
+	int i;
+
+	if (inst->quirks & pmac_i2c_quirk_invmask) {
+		for (i = 0; i < len; i ++)
+			inst->scratch[i] = (inst->buffer[i] & mask[i]) | val[i];
+	} else {
+		for (i = 0; i < len; i ++)
+			inst->scratch[i] = (inst->buffer[i] & ~mask[i])
+				| (val[i] & mask[i]);
+	}
+}
+
+static int pmac_i2c_do_rmw(PMF_STD_ARGS, u32 masklen, u32 valuelen,
+			   u32 totallen, const u8 *maskdata,
+			   const u8 *valuedata)
+{
+	struct pmac_i2c_pf_inst *inst = instdata;
+
+	if (masklen > inst->bytes || valuelen > inst->bytes ||
+	    totallen > inst->bytes || valuelen > masklen)
+		return -EINVAL;
+
+	pmac_i2c_do_apply_rmw(inst, masklen, maskdata, valuedata);
+
+	return pmac_i2c_xfer(inst->bus, inst->addr | pmac_i2c_write, 0, 0,
+			     inst->scratch, totallen);
+}
+
+static int pmac_i2c_do_read_sub(PMF_STD_ARGS, u8 subaddr, u32 len)
+{
+	struct pmac_i2c_pf_inst *inst = instdata;
+
+	inst->bytes = len;
+	return pmac_i2c_xfer(inst->bus, inst->addr | pmac_i2c_read, 1, subaddr,
+			     inst->buffer, len);
+}
+
+static int pmac_i2c_do_write_sub(PMF_STD_ARGS, u8 subaddr, u32 len,
+				     const u8 *data)
+{
+	struct pmac_i2c_pf_inst *inst = instdata;
+
+	return pmac_i2c_xfer(inst->bus, inst->addr | pmac_i2c_write, 1,
+			     subaddr, (u8 *)data, len);
+}
 
+static int pmac_i2c_do_set_mode(PMF_STD_ARGS, int mode)
+{
+	struct pmac_i2c_pf_inst *inst = instdata;
+
+	return pmac_i2c_setmode(inst->bus, mode);
+}
+
+static int pmac_i2c_do_rmw_sub(PMF_STD_ARGS, u8 subaddr, u32 masklen,
+			       u32 valuelen, u32 totallen, const u8 *maskdata,
+			       const u8 *valuedata)
+{
+	struct pmac_i2c_pf_inst *inst = instdata;
+
+	if (masklen > inst->bytes || valuelen > inst->bytes ||
+	    totallen > inst->bytes || valuelen > masklen)
+		return -EINVAL;
+
+	pmac_i2c_do_apply_rmw(inst, masklen, maskdata, valuedata);
+
+	return pmac_i2c_xfer(inst->bus, inst->addr | pmac_i2c_write, 1,
+			     subaddr, inst->scratch, totallen);
+}
+
+static int pmac_i2c_do_mask_and_comp(PMF_STD_ARGS, u32 len,
+				     const u8 *maskdata,
+				     const u8 *valuedata)
+{
+	struct pmac_i2c_pf_inst *inst = instdata;
+	int i, match;
+
+	/* Get return value pointer, it's assumed to be a u32 */
+	if (!args || !args->count || !args->u[0].p)
+		return -EINVAL;
+
+	/* Check buffer */
+	if (len > inst->bytes)
+		return -EINVAL;
+
+	for (i = 0, match = 1; match && i < len; i ++)
+		if ((inst->buffer[i] & maskdata[i]) != valuedata[i])
+			match = 0;
+	*args->u[0].p = match;
 	return 0;
 }
-EXPORT_SYMBOL(pmac_low_i2c_setmode);
 
-int pmac_low_i2c_xfer(struct device_node *np, u8 addrdir, u8 subaddr, u8 *data, int len)
+static int pmac_i2c_do_delay(PMF_STD_ARGS, u32 duration)
 {
-	struct low_i2c_host *host = find_low_i2c_host(np);
+	msleep((duration + 999) / 1000);
+	return 0;
+}
 
-	if (!host)
-		return -ENODEV;
-	WARN_ON(!host->is_open);
 
-	return host->func(host, addrdir, subaddr, data, len);
+static struct pmf_handlers pmac_i2c_pfunc_handlers = {
+	.begin			= pmac_i2c_do_begin,
+	.end			= pmac_i2c_do_end,
+	.read_i2c		= pmac_i2c_do_read,
+	.write_i2c		= pmac_i2c_do_write,
+	.rmw_i2c		= pmac_i2c_do_rmw,
+	.read_i2c_sub		= pmac_i2c_do_read_sub,
+	.write_i2c_sub		= pmac_i2c_do_write_sub,
+	.rmw_i2c_sub		= pmac_i2c_do_rmw_sub,
+	.set_i2c_mode		= pmac_i2c_do_set_mode,
+	.mask_and_compare	= pmac_i2c_do_mask_and_comp,
+	.delay			= pmac_i2c_do_delay,
+};
+
+static void __init pmac_i2c_dev_create(struct device_node *np, int quirks)
+{
+	DBG("dev_create(%s)\n", np->full_name);
+
+	pmf_register_driver(np, &pmac_i2c_pfunc_handlers,
+			    (void *)(long)quirks);
+}
+
+static void __init pmac_i2c_dev_init(struct device_node *np, int quirks)
+{
+	DBG("dev_create(%s)\n", np->full_name);
+
+	pmf_do_functions(np, NULL, 0, PMF_FLAGS_ON_INIT, NULL);
+}
+
+static void pmac_i2c_dev_suspend(struct device_node *np, int quirks)
+{
+	DBG("dev_suspend(%s)\n", np->full_name);
+	pmf_do_functions(np, NULL, 0, PMF_FLAGS_ON_SLEEP, NULL);
+}
+
+static void pmac_i2c_dev_resume(struct device_node *np, int quirks)
+{
+	DBG("dev_resume(%s)\n", np->full_name);
+	pmf_do_functions(np, NULL, 0, PMF_FLAGS_ON_WAKE, NULL);
+}
+
+void pmac_pfunc_i2c_suspend(void)
+{
+	pmac_i2c_devscan(pmac_i2c_dev_suspend);
+}
+
+void pmac_pfunc_i2c_resume(void)
+{
+	pmac_i2c_devscan(pmac_i2c_dev_resume);
+}
+
+/*
+ * Initialize us: probe all i2c busses on the machine, instantiate
+ * busses and platform functions as needed.
+ */
+/* This is non-static as it might be called early by smp code */
+int __init pmac_i2c_init(void)
+{
+	static int i2c_inited;
+
+	if (i2c_inited)
+		return 0;
+	i2c_inited = 1;
+
+	/* Probe keywest-i2c busses */
+	kw_i2c_probe();
+
+#ifdef CONFIG_ADB_PMU
+	/* Probe PMU i2c busses */
+	pmu_i2c_probe();
+#endif
+
+#ifdef CONFIG_PMAC_SMU
+	/* Probe SMU i2c busses */
+	smu_i2c_probe();
+#endif
+
+	/* Now add plaform functions for some known devices */
+	pmac_i2c_devscan(pmac_i2c_dev_create);
+
+	return 0;
 }
-EXPORT_SYMBOL(pmac_low_i2c_xfer);
+arch_initcall(pmac_i2c_init);
+
+/* Since pmac_i2c_init can be called too early for the platform device
+ * registration, we need to do it at a later time. In our case, subsys
+ * happens to fit well, though I agree it's a bit of a hack...
+ */
+static int __init pmac_i2c_create_platform_devices(void)
+{
+	struct pmac_i2c_bus *bus;
+	int i = 0;
+
+	/* In the case where we are initialized from smp_init(), we must
+	 * not use the timer (and thus the irq). It's safe from now on
+	 * though
+	 */
+	pmac_i2c_force_poll = 0;
+
+	/* Create platform devices */
+	list_for_each_entry(bus, &pmac_i2c_busses, link) {
+		bus->platform_dev =
+			platform_device_alloc("i2c-powermac", i++);
+		if (bus->platform_dev == NULL)
+			return -ENOMEM;
+		bus->platform_dev->dev.platform_data = bus;
+		platform_device_add(bus->platform_dev);
+	}
+
+	/* Now call platform "init" functions */
+	pmac_i2c_devscan(pmac_i2c_dev_init);
 
+	return 0;
+}
+subsys_initcall(pmac_i2c_create_platform_devices);
diff --git a/arch/powerpc/platforms/powermac/nvram.c b/arch/powerpc/platforms/powermac/nvram.c
index 4042e2f06ee0..3ebd045a3350 100644
--- a/arch/powerpc/platforms/powermac/nvram.c
+++ b/arch/powerpc/platforms/powermac/nvram.c
@@ -514,7 +514,7 @@ static void core99_nvram_sync(void)
 #endif
 }
 
-static int __init core99_nvram_setup(struct device_node *dp)
+static int __init core99_nvram_setup(struct device_node *dp, unsigned long addr)
 {
 	int i;
 	u32 gen_bank0, gen_bank1;
@@ -528,7 +528,7 @@ static int __init core99_nvram_setup(struct device_node *dp)
 		printk(KERN_ERR "nvram: can't allocate ram image\n");
 		return -ENOMEM;
 	}
-	nvram_data = ioremap(dp->addrs[0].address, NVRAM_SIZE*2);
+	nvram_data = ioremap(addr, NVRAM_SIZE*2);
 	nvram_naddrs = 1; /* Make sure we get the correct case */
 
 	DBG("nvram: Checking bank 0...\n");
@@ -549,6 +549,7 @@ static int __init core99_nvram_setup(struct device_node *dp)
 	ppc_md.nvram_write	= core99_nvram_write;
 	ppc_md.nvram_size	= core99_nvram_size;
 	ppc_md.nvram_sync	= core99_nvram_sync;
+	ppc_md.machine_shutdown	= core99_nvram_sync;
 	/* 
 	 * Maybe we could be smarter here though making an exclusive list
 	 * of known flash chips is a bit nasty as older OF didn't provide us
@@ -569,34 +570,48 @@ static int __init core99_nvram_setup(struct device_node *dp)
 int __init pmac_nvram_init(void)
 {
 	struct device_node *dp;
+	struct resource r1, r2;
+	unsigned int s1 = 0, s2 = 0;
 	int err = 0;
 
 	nvram_naddrs = 0;
 
-	dp = find_devices("nvram");
+	dp = of_find_node_by_name(NULL, "nvram");
 	if (dp == NULL) {
 		printk(KERN_ERR "Can't find NVRAM device\n");
 		return -ENODEV;
 	}
-	nvram_naddrs = dp->n_addrs;
+
+	/* Try to obtain an address */
+	if (of_address_to_resource(dp, 0, &r1) == 0) {
+		nvram_naddrs = 1;
+		s1 = (r1.end - r1.start) + 1;
+		if (of_address_to_resource(dp, 1, &r2) == 0) {
+			nvram_naddrs = 2;
+			s2 = (r2.end - r2.start) + 1;
+		}
+	}
+
 	is_core_99 = device_is_compatible(dp, "nvram,flash");
-	if (is_core_99)
-		err = core99_nvram_setup(dp);
+	if (is_core_99) {
+		err = core99_nvram_setup(dp, r1.start);
+		goto bail;
+	}
+
 #ifdef CONFIG_PPC32
-	else if (_machine == _MACH_chrp && nvram_naddrs == 1) {
-		nvram_data = ioremap(dp->addrs[0].address + isa_mem_base,
-				     dp->addrs[0].size);
+	if (_machine == _MACH_chrp && nvram_naddrs == 1) {
+		nvram_data = ioremap(r1.start, s1);
 		nvram_mult = 1;
 		ppc_md.nvram_read_val	= direct_nvram_read_byte;
 		ppc_md.nvram_write_val	= direct_nvram_write_byte;
 	} else if (nvram_naddrs == 1) {
-		nvram_data = ioremap(dp->addrs[0].address, dp->addrs[0].size);
-		nvram_mult = (dp->addrs[0].size + NVRAM_SIZE - 1) / NVRAM_SIZE;
+		nvram_data = ioremap(r1.start, s1);
+		nvram_mult = (s1 + NVRAM_SIZE - 1) / NVRAM_SIZE;
 		ppc_md.nvram_read_val	= direct_nvram_read_byte;
 		ppc_md.nvram_write_val	= direct_nvram_write_byte;
 	} else if (nvram_naddrs == 2) {
-		nvram_addr = ioremap(dp->addrs[0].address, dp->addrs[0].size);
-		nvram_data = ioremap(dp->addrs[1].address, dp->addrs[1].size);
+		nvram_addr = ioremap(r1.start, s1);
+		nvram_data = ioremap(r2.start, s2);
 		ppc_md.nvram_read_val	= indirect_nvram_read_byte;
 		ppc_md.nvram_write_val	= indirect_nvram_write_byte;
 	} else if (nvram_naddrs == 0 && sys_ctrler == SYS_CTRLER_PMU) {
@@ -605,13 +620,15 @@ int __init pmac_nvram_init(void)
 		ppc_md.nvram_read_val	= pmu_nvram_read_byte;
 		ppc_md.nvram_write_val	= pmu_nvram_write_byte;
 #endif /* CONFIG_ADB_PMU */
-	}
-#endif
-	else {
+	} else {
 		printk(KERN_ERR "Incompatible type of NVRAM\n");
-		return -ENXIO;
+		err = -ENXIO;
 	}
-	lookup_partitions();
+#endif /* CONFIG_PPC32 */
+bail:
+	of_node_put(dp);
+	if (err == 0)
+		lookup_partitions();
 	return err;
 }
 
diff --git a/arch/powerpc/platforms/powermac/pci.c b/arch/powerpc/platforms/powermac/pci.c
index 443be526cde7..f671ed253901 100644
--- a/arch/powerpc/platforms/powermac/pci.c
+++ b/arch/powerpc/platforms/powermac/pci.c
@@ -1,7 +1,7 @@
 /*
  * Support for PCI bridges found on Power Macintoshes.
  *
- * Copyright (C) 2003 Benjamin Herrenschmuidt (benh@kernel.crashing.org)
+ * Copyright (C) 2003-2005 Benjamin Herrenschmuidt (benh@kernel.crashing.org)
  * Copyright (C) 1997 Paul Mackerras (paulus@samba.org)
  *
  * This program is free software; you can redistribute it and/or
@@ -25,7 +25,7 @@
 #include <asm/pmac_feature.h>
 #include <asm/grackle.h>
 #ifdef CONFIG_PPC64
-#include <asm/iommu.h>
+//#include <asm/iommu.h>
 #include <asm/ppc-pci.h>
 #endif
 
@@ -44,6 +44,7 @@ static int add_bridge(struct device_node *dev);
 static int has_uninorth;
 #ifdef CONFIG_PPC64
 static struct pci_controller *u3_agp;
+static struct pci_controller *u4_pcie;
 static struct pci_controller *u3_ht;
 #endif /* CONFIG_PPC64 */
 
@@ -97,11 +98,8 @@ static void __init fixup_bus_range(struct device_node *bridge)
 
 	/* Lookup the "bus-range" property for the hose */
 	bus_range = (int *) get_property(bridge, "bus-range", &len);
-	if (bus_range == NULL || len < 2 * sizeof(int)) {
-		printk(KERN_WARNING "Can't get bus-range for %s\n",
-			       bridge->full_name);
+	if (bus_range == NULL || len < 2 * sizeof(int))
 		return;
-	}
 	bus_range[1] = fixup_one_level_bus_range(bridge->child, bus_range[1]);
 }
 
@@ -128,14 +126,14 @@ static void __init fixup_bus_range(struct device_node *bridge)
  */
 
 #define MACRISC_CFA0(devfn, off)	\
-	((1 << (unsigned long)PCI_SLOT(dev_fn)) \
-	| (((unsigned long)PCI_FUNC(dev_fn)) << 8) \
-	| (((unsigned long)(off)) & 0xFCUL))
+	((1 << (unsigned int)PCI_SLOT(dev_fn)) \
+	| (((unsigned int)PCI_FUNC(dev_fn)) << 8) \
+	| (((unsigned int)(off)) & 0xFCUL))
 
 #define MACRISC_CFA1(bus, devfn, off)	\
-	((((unsigned long)(bus)) << 16) \
-	|(((unsigned long)(devfn)) << 8) \
-	|(((unsigned long)(off)) & 0xFCUL) \
+	((((unsigned int)(bus)) << 16) \
+	|(((unsigned int)(devfn)) << 8) \
+	|(((unsigned int)(off)) & 0xFCUL) \
 	|1UL)
 
 static unsigned long macrisc_cfg_access(struct pci_controller* hose,
@@ -168,7 +166,8 @@ static int macrisc_read_config(struct pci_bus *bus, unsigned int devfn,
 	hose = pci_bus_to_host(bus);
 	if (hose == NULL)
 		return PCIBIOS_DEVICE_NOT_FOUND;
-
+	if (offset >= 0x100)
+		return  PCIBIOS_BAD_REGISTER_NUMBER;
 	addr = macrisc_cfg_access(hose, bus->number, devfn, offset);
 	if (!addr)
 		return PCIBIOS_DEVICE_NOT_FOUND;
@@ -199,7 +198,8 @@ static int macrisc_write_config(struct pci_bus *bus, unsigned int devfn,
 	hose = pci_bus_to_host(bus);
 	if (hose == NULL)
 		return PCIBIOS_DEVICE_NOT_FOUND;
-
+	if (offset >= 0x100)
+		return  PCIBIOS_BAD_REGISTER_NUMBER;
 	addr = macrisc_cfg_access(hose, bus->number, devfn, offset);
 	if (!addr)
 		return PCIBIOS_DEVICE_NOT_FOUND;
@@ -234,12 +234,13 @@ static struct pci_ops macrisc_pci_ops =
 /*
  * Verify that a specific (bus, dev_fn) exists on chaos
  */
-static int
-chaos_validate_dev(struct pci_bus *bus, int devfn, int offset)
+static int chaos_validate_dev(struct pci_bus *bus, int devfn, int offset)
 {
 	struct device_node *np;
 	u32 *vendor, *device;
 
+	if (offset >= 0x100)
+		return  PCIBIOS_BAD_REGISTER_NUMBER;
 	np = pci_busdev_to_OF_node(bus, devfn);
 	if (np == NULL)
 		return PCIBIOS_DEVICE_NOT_FOUND;
@@ -285,15 +286,13 @@ static struct pci_ops chaos_pci_ops =
 };
 
 static void __init setup_chaos(struct pci_controller *hose,
-			       struct reg_property *addr)
+			       struct resource *addr)
 {
 	/* assume a `chaos' bridge */
 	hose->ops = &chaos_pci_ops;
-	hose->cfg_addr = ioremap(addr->address + 0x800000, 0x1000);
-	hose->cfg_data = ioremap(addr->address + 0xc00000, 0x1000);
+	hose->cfg_addr = ioremap(addr->start + 0x800000, 0x1000);
+	hose->cfg_data = ioremap(addr->start + 0xc00000, 0x1000);
 }
-#else
-#define setup_chaos(hose, addr)
 #endif /* CONFIG_PPC32 */
 
 #ifdef CONFIG_PPC64
@@ -326,7 +325,7 @@ static int u3_ht_skip_device(struct pci_controller *hose,
 	else
 		busdn = hose->arch_data;
 	for (dn = busdn->child; dn; dn = dn->sibling)
-		if (dn->data && PCI_DN(dn)->devfn == devfn)
+		if (PCI_DN(dn) && PCI_DN(dn)->devfn == devfn)
 			break;
 	if (dn == NULL)
 		return -1;
@@ -343,10 +342,10 @@ static int u3_ht_skip_device(struct pci_controller *hose,
 }
 
 #define U3_HT_CFA0(devfn, off)		\
-		((((unsigned long)devfn) << 8) | offset)
+		((((unsigned int)devfn) << 8) | offset)
 #define U3_HT_CFA1(bus, devfn, off)	\
 		(U3_HT_CFA0(devfn, off) \
-		+ (((unsigned long)bus) << 16) \
+		+ (((unsigned int)bus) << 16) \
 		+ 0x01000000UL)
 
 static unsigned long u3_ht_cfg_access(struct pci_controller* hose,
@@ -356,9 +355,11 @@ static unsigned long u3_ht_cfg_access(struct pci_controller* hose,
 		/* For now, we don't self probe U3 HT bridge */
 		if (PCI_SLOT(devfn) == 0)
 			return 0;
-		return ((unsigned long)hose->cfg_data) + U3_HT_CFA0(devfn, offset);
+		return ((unsigned long)hose->cfg_data) +
+			U3_HT_CFA0(devfn, offset);
 	} else
-		return ((unsigned long)hose->cfg_data) + U3_HT_CFA1(bus, devfn, offset);
+		return ((unsigned long)hose->cfg_data) +
+			U3_HT_CFA1(bus, devfn, offset);
 }
 
 static int u3_ht_read_config(struct pci_bus *bus, unsigned int devfn,
@@ -370,7 +371,8 @@ static int u3_ht_read_config(struct pci_bus *bus, unsigned int devfn,
 	hose = pci_bus_to_host(bus);
 	if (hose == NULL)
 		return PCIBIOS_DEVICE_NOT_FOUND;
-
+	if (offset >= 0x100)
+		return  PCIBIOS_BAD_REGISTER_NUMBER;
 	addr = u3_ht_cfg_access(hose, bus->number, devfn, offset);
 	if (!addr)
 		return PCIBIOS_DEVICE_NOT_FOUND;
@@ -419,7 +421,8 @@ static int u3_ht_write_config(struct pci_bus *bus, unsigned int devfn,
 	hose = pci_bus_to_host(bus);
 	if (hose == NULL)
 		return PCIBIOS_DEVICE_NOT_FOUND;
-
+	if (offset >= 0x100)
+		return  PCIBIOS_BAD_REGISTER_NUMBER;
 	addr = u3_ht_cfg_access(hose, bus->number, devfn, offset);
 	if (!addr)
 		return PCIBIOS_DEVICE_NOT_FOUND;
@@ -459,6 +462,112 @@ static struct pci_ops u3_ht_pci_ops =
 	u3_ht_read_config,
 	u3_ht_write_config
 };
+
+#define U4_PCIE_CFA0(devfn, off)	\
+	((1 << ((unsigned int)PCI_SLOT(dev_fn)))	\
+	 | (((unsigned int)PCI_FUNC(dev_fn)) << 8)	\
+	 | ((((unsigned int)(off)) >> 8) << 28) \
+	 | (((unsigned int)(off)) & 0xfcU))
+
+#define U4_PCIE_CFA1(bus, devfn, off)	\
+	((((unsigned int)(bus)) << 16) \
+	 |(((unsigned int)(devfn)) << 8)	\
+	 | ((((unsigned int)(off)) >> 8) << 28) \
+	 |(((unsigned int)(off)) & 0xfcU)	\
+	 |1UL)
+
+static unsigned long u4_pcie_cfg_access(struct pci_controller* hose,
+					u8 bus, u8 dev_fn, int offset)
+{
+	unsigned int caddr;
+
+	if (bus == hose->first_busno) {
+		caddr = U4_PCIE_CFA0(dev_fn, offset);
+	} else
+		caddr = U4_PCIE_CFA1(bus, dev_fn, offset);
+
+	/* Uninorth will return garbage if we don't read back the value ! */
+	do {
+		out_le32(hose->cfg_addr, caddr);
+	} while (in_le32(hose->cfg_addr) != caddr);
+
+	offset &= 0x03;
+	return ((unsigned long)hose->cfg_data) + offset;
+}
+
+static int u4_pcie_read_config(struct pci_bus *bus, unsigned int devfn,
+			       int offset, int len, u32 *val)
+{
+	struct pci_controller *hose;
+	unsigned long addr;
+
+	hose = pci_bus_to_host(bus);
+	if (hose == NULL)
+		return PCIBIOS_DEVICE_NOT_FOUND;
+	if (offset >= 0x1000)
+		return  PCIBIOS_BAD_REGISTER_NUMBER;
+	addr = u4_pcie_cfg_access(hose, bus->number, devfn, offset);
+	if (!addr)
+		return PCIBIOS_DEVICE_NOT_FOUND;
+	/*
+	 * Note: the caller has already checked that offset is
+	 * suitably aligned and that len is 1, 2 or 4.
+	 */
+	switch (len) {
+	case 1:
+		*val = in_8((u8 *)addr);
+		break;
+	case 2:
+		*val = in_le16((u16 *)addr);
+		break;
+	default:
+		*val = in_le32((u32 *)addr);
+		break;
+	}
+	return PCIBIOS_SUCCESSFUL;
+}
+
+static int u4_pcie_write_config(struct pci_bus *bus, unsigned int devfn,
+				int offset, int len, u32 val)
+{
+	struct pci_controller *hose;
+	unsigned long addr;
+
+	hose = pci_bus_to_host(bus);
+	if (hose == NULL)
+		return PCIBIOS_DEVICE_NOT_FOUND;
+	if (offset >= 0x1000)
+		return  PCIBIOS_BAD_REGISTER_NUMBER;
+	addr = u4_pcie_cfg_access(hose, bus->number, devfn, offset);
+	if (!addr)
+		return PCIBIOS_DEVICE_NOT_FOUND;
+	/*
+	 * Note: the caller has already checked that offset is
+	 * suitably aligned and that len is 1, 2 or 4.
+	 */
+	switch (len) {
+	case 1:
+		out_8((u8 *)addr, val);
+		(void) in_8((u8 *)addr);
+		break;
+	case 2:
+		out_le16((u16 *)addr, val);
+		(void) in_le16((u16 *)addr);
+		break;
+	default:
+		out_le32((u32 *)addr, val);
+		(void) in_le32((u32 *)addr);
+		break;
+	}
+	return PCIBIOS_SUCCESSFUL;
+}
+
+static struct pci_ops u4_pcie_pci_ops =
+{
+	u4_pcie_read_config,
+	u4_pcie_write_config
+};
+
 #endif /* CONFIG_PPC64 */
 
 #ifdef CONFIG_PPC32
@@ -532,7 +641,8 @@ static void __init init_p2pbridge(void)
 	}
 	if (early_read_config_word(hose, bus, devfn,
 				   PCI_BRIDGE_CONTROL, &val) < 0) {
-		printk(KERN_ERR "init_p2pbridge: couldn't read bridge control\n");
+		printk(KERN_ERR "init_p2pbridge: couldn't read bridge"
+		       " control\n");
 		return;
 	}
 	val &= ~PCI_BRIDGE_CTL_MASTER_ABORT;
@@ -576,36 +686,38 @@ static void __init fixup_nec_usb2(void)
 			continue;
 		early_read_config_dword(hose, bus, devfn, 0xe4, &data);
 		if (data & 1UL) {
-			printk("Found NEC PD720100A USB2 chip with disabled EHCI, fixing up...\n");
+			printk("Found NEC PD720100A USB2 chip with disabled"
+			       " EHCI, fixing up...\n");
 			data &= ~1UL;
 			early_write_config_dword(hose, bus, devfn, 0xe4, data);
-			early_write_config_byte(hose, bus, devfn | 2, PCI_INTERRUPT_LINE,
+			early_write_config_byte(hose, bus,
+						devfn | 2, PCI_INTERRUPT_LINE,
 				nec->intrs[0].line);
 		}
 	}
 }
 
 static void __init setup_bandit(struct pci_controller *hose,
-				struct reg_property *addr)
+				struct resource *addr)
 {
 	hose->ops = &macrisc_pci_ops;
-	hose->cfg_addr = ioremap(addr->address + 0x800000, 0x1000);
-	hose->cfg_data = ioremap(addr->address + 0xc00000, 0x1000);
+	hose->cfg_addr = ioremap(addr->start + 0x800000, 0x1000);
+	hose->cfg_data = ioremap(addr->start + 0xc00000, 0x1000);
 	init_bandit(hose);
 }
 
 static int __init setup_uninorth(struct pci_controller *hose,
-				 struct reg_property *addr)
+				 struct resource *addr)
 {
 	pci_assign_all_buses = 1;
 	has_uninorth = 1;
 	hose->ops = &macrisc_pci_ops;
-	hose->cfg_addr = ioremap(addr->address + 0x800000, 0x1000);
-	hose->cfg_data = ioremap(addr->address + 0xc00000, 0x1000);
+	hose->cfg_addr = ioremap(addr->start + 0x800000, 0x1000);
+	hose->cfg_data = ioremap(addr->start + 0xc00000, 0x1000);
 	/* We "know" that the bridge at f2000000 has the PCI slots. */
-	return addr->address == 0xf2000000;
+	return addr->start == 0xf2000000;
 }
-#endif
+#endif /* CONFIG_PPC32 */
 
 #ifdef CONFIG_PPC64
 static void __init setup_u3_agp(struct pci_controller* hose)
@@ -625,15 +737,36 @@ static void __init setup_u3_agp(struct pci_controller* hose)
 	hose->ops = &macrisc_pci_ops;
 	hose->cfg_addr = ioremap(0xf0000000 + 0x800000, 0x1000);
 	hose->cfg_data = ioremap(0xf0000000 + 0xc00000, 0x1000);
-
 	u3_agp = hose;
 }
 
+static void __init setup_u4_pcie(struct pci_controller* hose)
+{
+	/* We currently only implement the "non-atomic" config space, to
+	 * be optimised later.
+	 */
+	hose->ops = &u4_pcie_pci_ops;
+	hose->cfg_addr = ioremap(0xf0000000 + 0x800000, 0x1000);
+	hose->cfg_data = ioremap(0xf0000000 + 0xc00000, 0x1000);
+
+	/* The bus contains a bridge from root -> device, we need to
+	 * make it visible on bus 0 so that we pick the right type
+	 * of config cycles. If we didn't, we would have to force all
+	 * config cycles to be type 1. So we override the "bus-range"
+	 * property here
+	 */
+	hose->first_busno = 0x00;
+	hose->last_busno = 0xff;
+	u4_pcie = hose;
+}
+
 static void __init setup_u3_ht(struct pci_controller* hose)
 {
 	struct device_node *np = (struct device_node *)hose->arch_data;
+	struct pci_controller *other = NULL;
 	int i, cur;
 
+
 	hose->ops = &u3_ht_pci_ops;
 
 	/* We hard code the address because of the different size of
@@ -667,11 +800,20 @@ static void __init setup_u3_ht(struct pci_controller* hose)
 
 	u3_ht = hose;
 
-	if (u3_agp == NULL) {
-		DBG("U3 has no AGP, using full resource range\n");
+	if (u3_agp != NULL)
+		other = u3_agp;
+	else if (u4_pcie != NULL)
+		other = u4_pcie;
+
+	if (other == NULL) {
+		DBG("U3/4 has no AGP/PCIE, using full resource range\n");
 		return;
 	}
 
+	/* Fixup bus range vs. PCIE */
+	if (u4_pcie)
+		hose->last_busno = u4_pcie->first_busno - 1;
+
 	/* We "remove" the AGP resources from the resources allocated to HT,
 	 * that is we create "holes". However, that code does assumptions
 	 * that so far happen to be true (cross fingers...), typically that
@@ -679,7 +821,7 @@ static void __init setup_u3_ht(struct pci_controller* hose)
 	 */
 	cur = 0;
 	for (i=0; i<3; i++) {
-		struct resource *res = &u3_agp->mem_resources[i];
+		struct resource *res = &other->mem_resources[i];
 		if (res->flags != IORESOURCE_MEM)
 			continue;
 		/* We don't care about "fine" resources */
@@ -722,7 +864,7 @@ static void __init setup_u3_ht(struct pci_controller* hose)
 		hose->mem_resources[cur-1].end = res->start - 1;
 	}
 }
-#endif
+#endif /* CONFIG_PPC64 */
 
 /*
  * We assume that if we have a G3 powermac, we have one bridge called
@@ -733,24 +875,17 @@ static int __init add_bridge(struct device_node *dev)
 {
 	int len;
 	struct pci_controller *hose;
-#ifdef CONFIG_PPC32
-	struct reg_property *addr;
-#endif
+	struct resource rsrc;
 	char *disp_name;
 	int *bus_range;
-	int primary = 1;
+	int primary = 1, has_address = 0;
 
 	DBG("Adding PCI host bridge %s\n", dev->full_name);
 
-#ifdef CONFIG_PPC32
-	/* XXX fix this */
-	addr = (struct reg_property *) get_property(dev, "reg", &len);
-	if (addr == NULL || len < sizeof(*addr)) {
-		printk(KERN_WARNING "Can't use %s: no address\n",
-		       dev->full_name);
-		return -ENODEV;
-	}
-#endif
+	/* Fetch host bridge registers address */
+	has_address = (of_address_to_resource(dev, 0, &rsrc) == 0);
+
+	/* Get bus range if any */
 	bus_range = (int *) get_property(dev, "bus-range", &len);
 	if (bus_range == NULL || len < 2 * sizeof(int)) {
 		printk(KERN_WARNING "Can't get bus-range for %s, assume"
@@ -770,6 +905,8 @@ static int __init add_bridge(struct device_node *dev)
 	hose->last_busno = bus_range ? bus_range[1] : 0xff;
 
 	disp_name = NULL;
+
+	/* 64 bits only bridges */
 #ifdef CONFIG_PPC64
 	if (device_is_compatible(dev, "u3-agp")) {
 		setup_u3_agp(hose);
@@ -779,28 +916,37 @@ static int __init add_bridge(struct device_node *dev)
 		setup_u3_ht(hose);
 		disp_name = "U3-HT";
 		primary = 1;
+	} else if (device_is_compatible(dev, "u4-pcie")) {
+		setup_u4_pcie(hose);
+		disp_name = "U4-PCIE";
+		primary = 0;
 	}
-	printk(KERN_INFO "Found %s PCI host bridge.  Firmware bus number: %d->%d\n",
-		disp_name, hose->first_busno, hose->last_busno);
-#else
+	printk(KERN_INFO "Found %s PCI host bridge.  Firmware bus number:"
+	       " %d->%d\n", disp_name, hose->first_busno, hose->last_busno);
+#endif /* CONFIG_PPC64 */
+
+	/* 32 bits only bridges */
+#ifdef CONFIG_PPC32
 	if (device_is_compatible(dev, "uni-north")) {
-		primary = setup_uninorth(hose, addr);
+		primary = setup_uninorth(hose, &rsrc);
 		disp_name = "UniNorth";
 	} else if (strcmp(dev->name, "pci") == 0) {
 		/* XXX assume this is a mpc106 (grackle) */
 		setup_grackle(hose);
 		disp_name = "Grackle (MPC106)";
 	} else if (strcmp(dev->name, "bandit") == 0) {
-		setup_bandit(hose, addr);
+		setup_bandit(hose, &rsrc);
 		disp_name = "Bandit";
 	} else if (strcmp(dev->name, "chaos") == 0) {
-		setup_chaos(hose, addr);
+		setup_chaos(hose, &rsrc);
 		disp_name = "Chaos";
 		primary = 0;
 	}
-	printk(KERN_INFO "Found %s PCI host bridge at 0x%08lx. Firmware bus number: %d->%d\n",
-		disp_name, addr->address, hose->first_busno, hose->last_busno);
-#endif
+	printk(KERN_INFO "Found %s PCI host bridge at 0x%08lx. "
+	       "Firmware bus number: %d->%d\n",
+		disp_name, rsrc.start, hose->first_busno, hose->last_busno);
+#endif /* CONFIG_PPC32 */
+
 	DBG(" ->Hose at 0x%p, cfg_addr=0x%p,cfg_data=0x%p\n",
 		hose, hose->cfg_addr, hose->cfg_data);
 
@@ -814,8 +960,7 @@ static int __init add_bridge(struct device_node *dev)
 	return 0;
 }
 
-static void __init
-pcibios_fixup_OF_interrupts(void)
+static void __init pcibios_fixup_OF_interrupts(void)
 {
 	struct pci_dev* dev = NULL;
 
@@ -835,8 +980,7 @@ pcibios_fixup_OF_interrupts(void)
 	}
 }
 
-void __init
-pmac_pcibios_fixup(void)
+void __init pmac_pcibios_fixup(void)
 {
 	/* Fixup interrupts according to OF tree */
 	pcibios_fixup_OF_interrupts();
@@ -899,6 +1043,8 @@ void __init pmac_pci_init(void)
 		pci_setup_phb_io(u3_ht, 1);
 	if (u3_agp)
 		pci_setup_phb_io(u3_agp, 0);
+	if (u4_pcie)
+		pci_setup_phb_io(u4_pcie, 0);
 
 	/*
 	 * On ppc64, fixup the IO resources on our host bridges as
@@ -911,7 +1057,8 @@ void __init pmac_pci_init(void)
 
 	/* Fixup the PCI<->OF mapping for U3 AGP due to bus renumbering. We
 	 * assume there is no P2P bridge on the AGP bus, which should be a
-	 * safe assumptions hopefully.
+	 * safe assumptions for now. We should do something better in the
+	 * future though
 	 */
 	if (u3_agp) {
 		struct device_node *np = u3_agp->arch_data;
@@ -919,7 +1066,6 @@ void __init pmac_pci_init(void)
 		for (np = np->child; np; np = np->sibling)
 			PCI_DN(np)->busno = 0xf0;
 	}
-
 	/* pmac_check_ht_link(); */
 
 	/* Tell pci.c to not use the common resource allocation mechanism */
@@ -1126,7 +1272,8 @@ void pmac_pci_fixup_pciata(struct pci_dev* dev)
  good:
 	pci_read_config_byte(dev, PCI_CLASS_PROG, &progif);
 	if ((progif & 5) != 5) {
-		printk(KERN_INFO "Forcing PCI IDE into native mode: %s\n", pci_name(dev));
+		printk(KERN_INFO "Forcing PCI IDE into native mode: %s\n",
+		       pci_name(dev));
 		(void) pci_write_config_byte(dev, PCI_CLASS_PROG, progif|5);
 		if (pci_read_config_byte(dev, PCI_CLASS_PROG, &progif) ||
 		    (progif & 5) != 5)
@@ -1152,7 +1299,8 @@ static void fixup_k2_sata(struct pci_dev* dev)
 		for (i = 0; i < 6; i++) {
 			dev->resource[i].start = dev->resource[i].end = 0;
 			dev->resource[i].flags = 0;
-			pci_write_config_dword(dev, PCI_BASE_ADDRESS_0 + 4 * i, 0);
+			pci_write_config_dword(dev, PCI_BASE_ADDRESS_0 + 4 * i,
+					       0);
 		}
 	} else {
 		pci_read_config_word(dev, PCI_COMMAND, &cmd);
@@ -1161,7 +1309,8 @@ static void fixup_k2_sata(struct pci_dev* dev)
 		for (i = 0; i < 5; i++) {
 			dev->resource[i].start = dev->resource[i].end = 0;
 			dev->resource[i].flags = 0;
-			pci_write_config_dword(dev, PCI_BASE_ADDRESS_0 + 4 * i, 0);
+			pci_write_config_dword(dev, PCI_BASE_ADDRESS_0 + 4 * i,
+					       0);
 		}
 	}
 }
diff --git a/arch/powerpc/platforms/powermac/pfunc_base.c b/arch/powerpc/platforms/powermac/pfunc_base.c
new file mode 100644
index 000000000000..4ffd2a9832a0
--- /dev/null
+++ b/arch/powerpc/platforms/powermac/pfunc_base.c
@@ -0,0 +1,405 @@
+#include <linux/config.h>
+#include <linux/types.h>
+#include <linux/init.h>
+#include <linux/delay.h>
+#include <linux/kernel.h>
+#include <linux/interrupt.h>
+#include <linux/spinlock.h>
+
+#include <asm/pmac_feature.h>
+#include <asm/pmac_pfunc.h>
+
+#define DBG(fmt...)	printk(fmt)
+
+static irqreturn_t macio_gpio_irq(int irq, void *data, struct pt_regs *regs)
+{
+	pmf_do_irq(data);
+
+	return IRQ_HANDLED;
+}
+
+static int macio_do_gpio_irq_enable(struct pmf_function *func)
+{
+	if (func->node->n_intrs < 1)
+		return -EINVAL;
+
+	return request_irq(func->node->intrs[0].line, macio_gpio_irq, 0,
+			   func->node->name, func);
+}
+
+static int macio_do_gpio_irq_disable(struct pmf_function *func)
+{
+	if (func->node->n_intrs < 1)
+		return -EINVAL;
+
+	free_irq(func->node->intrs[0].line, func);
+	return 0;
+}
+
+static int macio_do_gpio_write(PMF_STD_ARGS, u8 value, u8 mask)
+{
+	u8 __iomem *addr = (u8 __iomem *)func->driver_data;
+	unsigned long flags;
+	u8 tmp;
+
+	/* Check polarity */
+	if (args && args->count && !args->u[0].v)
+		value = ~value;
+
+	/* Toggle the GPIO */
+	spin_lock_irqsave(&feature_lock, flags);
+	tmp = readb(addr);
+	tmp = (tmp & ~mask) | (value & mask);
+	DBG("Do write 0x%02x to GPIO %s (%p)\n",
+	    tmp, func->node->full_name, addr);
+	writeb(tmp, addr);
+	spin_unlock_irqrestore(&feature_lock, flags);
+
+	return 0;
+}
+
+static int macio_do_gpio_read(PMF_STD_ARGS, u8 mask, int rshift, u8 xor)
+{
+	u8 __iomem *addr = (u8 __iomem *)func->driver_data;
+	u32 value;
+
+	/* Check if we have room for reply */
+	if (args == NULL || args->count == 0 || args->u[0].p == NULL)
+		return -EINVAL;
+
+	value = readb(addr);
+	*args->u[0].p = ((value & mask) >> rshift) ^ xor;
+
+	return 0;
+}
+
+static int macio_do_delay(PMF_STD_ARGS, u32 duration)
+{
+	/* assume we can sleep ! */
+	msleep((duration + 999) / 1000);
+	return 0;
+}
+
+static struct pmf_handlers macio_gpio_handlers = {
+	.irq_enable	= macio_do_gpio_irq_enable,
+	.irq_disable	= macio_do_gpio_irq_disable,
+	.write_gpio	= macio_do_gpio_write,
+	.read_gpio	= macio_do_gpio_read,
+	.delay		= macio_do_delay,
+};
+
+static void macio_gpio_init_one(struct macio_chip *macio)
+{
+	struct device_node *gparent, *gp;
+
+	/*
+	 * Find the "gpio" parent node
+	 */
+
+	for (gparent = NULL;
+	     (gparent = of_get_next_child(macio->of_node, gparent)) != NULL;)
+		if (strcmp(gparent->name, "gpio") == 0)
+			break;
+	if (gparent == NULL)
+		return;
+
+	DBG("Installing GPIO functions for macio %s\n",
+	    macio->of_node->full_name);
+
+	/*
+	 * Ok, got one, we dont need anything special to track them down, so
+	 * we just create them all
+	 */
+	for (gp = NULL; (gp = of_get_next_child(gparent, gp)) != NULL;) {
+		u32 *reg = (u32 *)get_property(gp, "reg", NULL);
+		unsigned long offset;
+		if (reg == NULL)
+			continue;
+		offset = *reg;
+		/* Deal with old style device-tree. We can safely hard code the
+		 * offset for now too even if it's a bit gross ...
+		 */
+		if (offset < 0x50)
+			offset += 0x50;
+		offset += (unsigned long)macio->base;
+		pmf_register_driver(gp, &macio_gpio_handlers, (void *)offset);
+	}
+
+	DBG("Calling initial GPIO functions for macio %s\n",
+	    macio->of_node->full_name);
+
+	/* And now we run all the init ones */
+	for (gp = NULL; (gp = of_get_next_child(gparent, gp)) != NULL;)
+		pmf_do_functions(gp, NULL, 0, PMF_FLAGS_ON_INIT, NULL);
+
+	/* Note: We do not at this point implement the "at sleep" or "at wake"
+	 * functions. I yet to find any for GPIOs anyway
+	 */
+}
+
+static int macio_do_write_reg32(PMF_STD_ARGS, u32 offset, u32 value, u32 mask)
+{
+	struct macio_chip *macio = func->driver_data;
+	unsigned long flags;
+
+	spin_lock_irqsave(&feature_lock, flags);
+	MACIO_OUT32(offset, (MACIO_IN32(offset) & ~mask) | (value & mask));
+	spin_unlock_irqrestore(&feature_lock, flags);
+	return 0;
+}
+
+static int macio_do_read_reg32(PMF_STD_ARGS, u32 offset)
+{
+	struct macio_chip *macio = func->driver_data;
+
+	/* Check if we have room for reply */
+	if (args == NULL || args->count == 0 || args->u[0].p == NULL)
+		return -EINVAL;
+
+	*args->u[0].p = MACIO_IN32(offset);
+	return 0;
+}
+
+static int macio_do_write_reg8(PMF_STD_ARGS, u32 offset, u8 value, u8 mask)
+{
+	struct macio_chip *macio = func->driver_data;
+	unsigned long flags;
+
+	spin_lock_irqsave(&feature_lock, flags);
+	MACIO_OUT8(offset, (MACIO_IN8(offset) & ~mask) | (value & mask));
+	spin_unlock_irqrestore(&feature_lock, flags);
+	return 0;
+}
+
+static int macio_do_read_reg8(PMF_STD_ARGS, u32 offset)
+{
+	struct macio_chip *macio = func->driver_data;
+
+	/* Check if we have room for reply */
+	if (args == NULL || args->count == 0 || args->u[0].p == NULL)
+		return -EINVAL;
+
+	*((u8 *)(args->u[0].p)) = MACIO_IN8(offset);
+	return 0;
+}
+
+static int macio_do_read_reg32_msrx(PMF_STD_ARGS, u32 offset, u32 mask,
+				    u32 shift, u32 xor)
+{
+	struct macio_chip *macio = func->driver_data;
+
+	/* Check if we have room for reply */
+	if (args == NULL || args->count == 0 || args->u[0].p == NULL)
+		return -EINVAL;
+
+	*args->u[0].p = ((MACIO_IN32(offset) & mask) >> shift) ^ xor;
+	return 0;
+}
+
+static int macio_do_read_reg8_msrx(PMF_STD_ARGS, u32 offset, u32 mask,
+				   u32 shift, u32 xor)
+{
+	struct macio_chip *macio = func->driver_data;
+
+	/* Check if we have room for reply */
+	if (args == NULL || args->count == 0 || args->u[0].p == NULL)
+		return -EINVAL;
+
+	*((u8 *)(args->u[0].p)) = ((MACIO_IN8(offset) & mask) >> shift) ^ xor;
+	return 0;
+}
+
+static int macio_do_write_reg32_slm(PMF_STD_ARGS, u32 offset, u32 shift,
+				    u32 mask)
+{
+	struct macio_chip *macio = func->driver_data;
+	unsigned long flags;
+	u32 tmp, val;
+
+	/* Check args */
+	if (args == NULL || args->count == 0)
+		return -EINVAL;
+
+	spin_lock_irqsave(&feature_lock, flags);
+	tmp = MACIO_IN32(offset);
+	val = args->u[0].v << shift;
+	tmp = (tmp & ~mask) | (val & mask);
+	MACIO_OUT32(offset, tmp);
+	spin_unlock_irqrestore(&feature_lock, flags);
+	return 0;
+}
+
+static int macio_do_write_reg8_slm(PMF_STD_ARGS, u32 offset, u32 shift,
+				   u32 mask)
+{
+	struct macio_chip *macio = func->driver_data;
+	unsigned long flags;
+	u32 tmp, val;
+
+	/* Check args */
+	if (args == NULL || args->count == 0)
+		return -EINVAL;
+
+	spin_lock_irqsave(&feature_lock, flags);
+	tmp = MACIO_IN8(offset);
+	val = args->u[0].v << shift;
+	tmp = (tmp & ~mask) | (val & mask);
+	MACIO_OUT8(offset, tmp);
+	spin_unlock_irqrestore(&feature_lock, flags);
+	return 0;
+}
+
+static struct pmf_handlers macio_mmio_handlers = {
+	.write_reg32		= macio_do_write_reg32,
+	.read_reg32		= macio_do_read_reg32,
+	.write_reg8		= macio_do_write_reg8,
+	.read_reg32		= macio_do_read_reg8,
+	.read_reg32_msrx	= macio_do_read_reg32_msrx,
+	.read_reg8_msrx		= macio_do_read_reg8_msrx,
+	.write_reg32_slm	= macio_do_write_reg32_slm,
+	.write_reg8_slm		= macio_do_write_reg8_slm,
+	.delay			= macio_do_delay,
+};
+
+static void macio_mmio_init_one(struct macio_chip *macio)
+{
+	DBG("Installing MMIO functions for macio %s\n",
+	    macio->of_node->full_name);
+
+	pmf_register_driver(macio->of_node, &macio_mmio_handlers, macio);
+}
+
+static struct device_node *unin_hwclock;
+
+static int unin_do_write_reg32(PMF_STD_ARGS, u32 offset, u32 value, u32 mask)
+{
+	unsigned long flags;
+
+	spin_lock_irqsave(&feature_lock, flags);
+	/* This is fairly bogus in darwin, but it should work for our needs
+	 * implemeted that way:
+	 */
+	UN_OUT(offset, (UN_IN(offset) & ~mask) | (value & mask));
+	spin_unlock_irqrestore(&feature_lock, flags);
+	return 0;
+}
+
+
+static struct pmf_handlers unin_mmio_handlers = {
+	.write_reg32		= unin_do_write_reg32,
+	.delay			= macio_do_delay,
+};
+
+static void uninorth_install_pfunc(void)
+{
+	struct device_node *np;
+
+	DBG("Installing functions for UniN %s\n",
+	    uninorth_node->full_name);
+
+	/*
+	 * Install handlers for the bridge itself
+	 */
+	pmf_register_driver(uninorth_node, &unin_mmio_handlers, NULL);
+	pmf_do_functions(uninorth_node, NULL, 0, PMF_FLAGS_ON_INIT, NULL);
+
+
+	/*
+	 * Install handlers for the hwclock child if any
+	 */
+	for (np = NULL; (np = of_get_next_child(uninorth_node, np)) != NULL;)
+		if (strcmp(np->name, "hw-clock") == 0) {
+			unin_hwclock = np;
+			break;
+		}
+	if (unin_hwclock) {
+		DBG("Installing functions for UniN clock %s\n",
+		    unin_hwclock->full_name);
+		pmf_register_driver(unin_hwclock, &unin_mmio_handlers, NULL);
+		pmf_do_functions(unin_hwclock, NULL, 0, PMF_FLAGS_ON_INIT,
+				 NULL);
+	}
+}
+
+/* We export this as the SMP code might init us early */
+int __init pmac_pfunc_base_install(void)
+{
+	static int pfbase_inited;
+	int i;
+
+	if (pfbase_inited)
+		return 0;
+	pfbase_inited = 1;
+
+
+	DBG("Installing base platform functions...\n");
+
+	/*
+	 * Locate mac-io chips and install handlers
+	 */
+	for (i = 0 ; i < MAX_MACIO_CHIPS; i++) {
+		if (macio_chips[i].of_node) {
+			macio_mmio_init_one(&macio_chips[i]);
+			macio_gpio_init_one(&macio_chips[i]);
+		}
+	}
+
+	/*
+	 * Install handlers for northbridge and direct mapped hwclock
+	 * if any. We do not implement the config space access callback
+	 * which is only ever used for functions that we do not call in
+	 * the current driver (enabling/disabling cells in U2, mostly used
+	 * to restore the PCI settings, we do that differently)
+	 */
+	if (uninorth_node && uninorth_base)
+		uninorth_install_pfunc();
+
+	DBG("All base functions installed\n");
+
+	return 0;
+}
+
+arch_initcall(pmac_pfunc_base_install);
+
+#ifdef CONFIG_PM
+
+/* Those can be called by pmac_feature. Ultimately, I should use a sysdev
+ * or a device, but for now, that's good enough until I sort out some
+ * ordering issues. Also, we do not bother with GPIOs, as so far I yet have
+ * to see a case where a GPIO function has the on-suspend or on-resume bit
+ */
+void pmac_pfunc_base_suspend(void)
+{
+	int i;
+
+	for (i = 0 ; i < MAX_MACIO_CHIPS; i++) {
+		if (macio_chips[i].of_node)
+			pmf_do_functions(macio_chips[i].of_node, NULL, 0,
+					 PMF_FLAGS_ON_SLEEP, NULL);
+	}
+	if (uninorth_node)
+		pmf_do_functions(uninorth_node, NULL, 0,
+				 PMF_FLAGS_ON_SLEEP, NULL);
+	if (unin_hwclock)
+		pmf_do_functions(unin_hwclock, NULL, 0,
+				 PMF_FLAGS_ON_SLEEP, NULL);
+}
+
+void pmac_pfunc_base_resume(void)
+{
+	int i;
+
+	if (unin_hwclock)
+		pmf_do_functions(unin_hwclock, NULL, 0,
+				 PMF_FLAGS_ON_WAKE, NULL);
+	if (uninorth_node)
+		pmf_do_functions(uninorth_node, NULL, 0,
+				 PMF_FLAGS_ON_WAKE, NULL);
+	for (i = 0 ; i < MAX_MACIO_CHIPS; i++) {
+		if (macio_chips[i].of_node)
+			pmf_do_functions(macio_chips[i].of_node, NULL, 0,
+					 PMF_FLAGS_ON_WAKE, NULL);
+	}
+}
+
+#endif /* CONFIG_PM */
diff --git a/arch/powerpc/platforms/powermac/pfunc_core.c b/arch/powerpc/platforms/powermac/pfunc_core.c
new file mode 100644
index 000000000000..c32c623001dc
--- /dev/null
+++ b/arch/powerpc/platforms/powermac/pfunc_core.c
@@ -0,0 +1,989 @@
+/*
+ *
+ * FIXME: Properly make this race free with refcounting etc...
+ *
+ * FIXME: LOCKING !!!
+ */
+
+#include <linux/config.h>
+#include <linux/init.h>
+#include <linux/delay.h>
+#include <linux/kernel.h>
+#include <linux/spinlock.h>
+#include <linux/module.h>
+
+#include <asm/semaphore.h>
+#include <asm/prom.h>
+#include <asm/pmac_pfunc.h>
+
+/* Debug */
+#define LOG_PARSE(fmt...)
+#define LOG_ERROR(fmt...)	printk(fmt)
+#define LOG_BLOB(t,b,c)
+#define DBG(fmt...)		printk(fmt)
+
+/* Command numbers */
+#define PMF_CMD_LIST			0
+#define PMF_CMD_WRITE_GPIO		1
+#define PMF_CMD_READ_GPIO		2
+#define PMF_CMD_WRITE_REG32		3
+#define PMF_CMD_READ_REG32		4
+#define PMF_CMD_WRITE_REG16		5
+#define PMF_CMD_READ_REG16		6
+#define PMF_CMD_WRITE_REG8		7
+#define PMF_CMD_READ_REG8		8
+#define PMF_CMD_DELAY			9
+#define PMF_CMD_WAIT_REG32		10
+#define PMF_CMD_WAIT_REG16		11
+#define PMF_CMD_WAIT_REG8		12
+#define PMF_CMD_READ_I2C		13
+#define PMF_CMD_WRITE_I2C		14
+#define PMF_CMD_RMW_I2C			15
+#define PMF_CMD_GEN_I2C			16
+#define PMF_CMD_SHIFT_BYTES_RIGHT	17
+#define PMF_CMD_SHIFT_BYTES_LEFT	18
+#define PMF_CMD_READ_CFG		19
+#define PMF_CMD_WRITE_CFG		20
+#define PMF_CMD_RMW_CFG			21
+#define PMF_CMD_READ_I2C_SUBADDR	22
+#define PMF_CMD_WRITE_I2C_SUBADDR	23
+#define PMF_CMD_SET_I2C_MODE		24
+#define PMF_CMD_RMW_I2C_SUBADDR		25
+#define PMF_CMD_READ_REG32_MASK_SHR_XOR	26
+#define PMF_CMD_READ_REG16_MASK_SHR_XOR	27
+#define PMF_CMD_READ_REG8_MASK_SHR_XOR	28
+#define PMF_CMD_WRITE_REG32_SHL_MASK	29
+#define PMF_CMD_WRITE_REG16_SHL_MASK	30
+#define PMF_CMD_WRITE_REG8_SHL_MASK	31
+#define PMF_CMD_MASK_AND_COMPARE	32
+#define PMF_CMD_COUNT			33
+
+/* This structure holds the state of the parser while walking through
+ * a function definition
+ */
+struct pmf_cmd {
+	const void		*cmdptr;
+	const void		*cmdend;
+	struct pmf_function	*func;
+	void			*instdata;
+	struct pmf_args		*args;
+	int			error;
+};
+
+#if 0
+/* Debug output */
+static void print_blob(const char *title, const void *blob, int bytes)
+{
+	printk("%s", title);
+	while(bytes--) {
+		printk("%02x ", *((u8 *)blob));
+		blob += 1;
+	}
+	printk("\n");
+}
+#endif
+
+/*
+ * Parser helpers
+ */
+
+static u32 pmf_next32(struct pmf_cmd *cmd)
+{
+	u32 value;
+	if ((cmd->cmdend - cmd->cmdptr) < 4) {
+		cmd->error = 1;
+		return 0;
+	}
+	value = *((u32 *)cmd->cmdptr);
+	cmd->cmdptr += 4;
+	return value;
+}
+
+static const void* pmf_next_blob(struct pmf_cmd *cmd, int count)
+{
+	const void *value;
+	if ((cmd->cmdend - cmd->cmdptr) < count) {
+		cmd->error = 1;
+		return NULL;
+	}
+	value = cmd->cmdptr;
+	cmd->cmdptr += count;
+	return value;
+}
+
+/*
+ * Individual command parsers
+ */
+
+#define PMF_PARSE_CALL(name, cmd, handlers, p...) \
+	do { \
+		if (cmd->error) \
+			return -ENXIO; \
+		if (handlers == NULL) \
+			return 0; \
+		if (handlers->name)				      \
+			return handlers->name(cmd->func, cmd->instdata, \
+					      cmd->args, p);	      \
+		return -1; \
+	} while(0) \
+
+
+static int pmf_parser_write_gpio(struct pmf_cmd *cmd, struct pmf_handlers *h)
+{
+	u8 value = (u8)pmf_next32(cmd);
+	u8 mask = (u8)pmf_next32(cmd);
+
+	LOG_PARSE("pmf: write_gpio(value: %02x, mask: %02x)\n", value, mask);
+
+	PMF_PARSE_CALL(write_gpio, cmd, h, value, mask);
+}
+
+static int pmf_parser_read_gpio(struct pmf_cmd *cmd, struct pmf_handlers *h)
+{
+	u8 mask = (u8)pmf_next32(cmd);
+	int rshift = (int)pmf_next32(cmd);
+	u8 xor = (u8)pmf_next32(cmd);
+
+	LOG_PARSE("pmf: read_gpio(mask: %02x, rshift: %d, xor: %02x)\n",
+		  mask, rshift, xor);
+
+	PMF_PARSE_CALL(read_gpio, cmd, h, mask, rshift, xor);
+}
+
+static int pmf_parser_write_reg32(struct pmf_cmd *cmd, struct pmf_handlers *h)
+{
+	u32 offset = pmf_next32(cmd);
+	u32 value = pmf_next32(cmd);
+	u32 mask = pmf_next32(cmd);
+
+	LOG_PARSE("pmf: write_reg32(offset: %08x, value: %08x, mask: %08x)\n",
+		  offset, value, mask);
+
+	PMF_PARSE_CALL(write_reg32, cmd, h, offset, value, mask);
+}
+
+static int pmf_parser_read_reg32(struct pmf_cmd *cmd, struct pmf_handlers *h)
+{
+	u32 offset = pmf_next32(cmd);
+
+	LOG_PARSE("pmf: read_reg32(offset: %08x)\n", offset);
+
+	PMF_PARSE_CALL(read_reg32, cmd, h, offset);
+}
+
+
+static int pmf_parser_write_reg16(struct pmf_cmd *cmd, struct pmf_handlers *h)
+{
+	u32 offset = pmf_next32(cmd);
+	u16 value = (u16)pmf_next32(cmd);
+	u16 mask = (u16)pmf_next32(cmd);
+
+	LOG_PARSE("pmf: write_reg16(offset: %08x, value: %04x, mask: %04x)\n",
+		  offset, value, mask);
+
+	PMF_PARSE_CALL(write_reg16, cmd, h, offset, value, mask);
+}
+
+static int pmf_parser_read_reg16(struct pmf_cmd *cmd, struct pmf_handlers *h)
+{
+	u32 offset = pmf_next32(cmd);
+
+	LOG_PARSE("pmf: read_reg16(offset: %08x)\n", offset);
+
+	PMF_PARSE_CALL(read_reg16, cmd, h, offset);
+}
+
+
+static int pmf_parser_write_reg8(struct pmf_cmd *cmd, struct pmf_handlers *h)
+{
+	u32 offset = pmf_next32(cmd);
+	u8 value = (u16)pmf_next32(cmd);
+	u8 mask = (u16)pmf_next32(cmd);
+
+	LOG_PARSE("pmf: write_reg8(offset: %08x, value: %02x, mask: %02x)\n",
+		  offset, value, mask);
+
+	PMF_PARSE_CALL(write_reg8, cmd, h, offset, value, mask);
+}
+
+static int pmf_parser_read_reg8(struct pmf_cmd *cmd, struct pmf_handlers *h)
+{
+	u32 offset = pmf_next32(cmd);
+
+	LOG_PARSE("pmf: read_reg8(offset: %08x)\n", offset);
+
+	PMF_PARSE_CALL(read_reg8, cmd, h, offset);
+}
+
+static int pmf_parser_delay(struct pmf_cmd *cmd, struct pmf_handlers *h)
+{
+	u32 duration = pmf_next32(cmd);
+
+	LOG_PARSE("pmf: delay(duration: %d us)\n", duration);
+
+	PMF_PARSE_CALL(delay, cmd, h, duration);
+}
+
+static int pmf_parser_wait_reg32(struct pmf_cmd *cmd, struct pmf_handlers *h)
+{
+	u32 offset = pmf_next32(cmd);
+	u32 value = pmf_next32(cmd);
+	u32 mask = pmf_next32(cmd);
+
+	LOG_PARSE("pmf: wait_reg32(offset: %08x, comp_value: %08x,mask: %08x)\n",
+		  offset, value, mask);
+
+	PMF_PARSE_CALL(wait_reg32, cmd, h, offset, value, mask);
+}
+
+static int pmf_parser_wait_reg16(struct pmf_cmd *cmd, struct pmf_handlers *h)
+{
+	u32 offset = pmf_next32(cmd);
+	u16 value = (u16)pmf_next32(cmd);
+	u16 mask = (u16)pmf_next32(cmd);
+
+	LOG_PARSE("pmf: wait_reg16(offset: %08x, comp_value: %04x,mask: %04x)\n",
+		  offset, value, mask);
+
+	PMF_PARSE_CALL(wait_reg16, cmd, h, offset, value, mask);
+}
+
+static int pmf_parser_wait_reg8(struct pmf_cmd *cmd, struct pmf_handlers *h)
+{
+	u32 offset = pmf_next32(cmd);
+	u8 value = (u8)pmf_next32(cmd);
+	u8 mask = (u8)pmf_next32(cmd);
+
+	LOG_PARSE("pmf: wait_reg8(offset: %08x, comp_value: %02x,mask: %02x)\n",
+		  offset, value, mask);
+
+	PMF_PARSE_CALL(wait_reg8, cmd, h, offset, value, mask);
+}
+
+static int pmf_parser_read_i2c(struct pmf_cmd *cmd, struct pmf_handlers *h)
+{
+	u32 bytes = pmf_next32(cmd);
+
+	LOG_PARSE("pmf: read_i2c(bytes: %ud)\n", bytes);
+
+	PMF_PARSE_CALL(read_i2c, cmd, h, bytes);
+}
+
+static int pmf_parser_write_i2c(struct pmf_cmd *cmd, struct pmf_handlers *h)
+{
+	u32 bytes = pmf_next32(cmd);
+	const void *blob = pmf_next_blob(cmd, bytes);
+
+	LOG_PARSE("pmf: write_i2c(bytes: %ud) ...\n", bytes);
+	LOG_BLOB("pmf:   data: \n", blob, bytes);
+
+	PMF_PARSE_CALL(write_i2c, cmd, h, bytes, blob);
+}
+
+
+static int pmf_parser_rmw_i2c(struct pmf_cmd *cmd, struct pmf_handlers *h)
+{
+	u32 maskbytes = pmf_next32(cmd);
+	u32 valuesbytes = pmf_next32(cmd);
+	u32 totalbytes = pmf_next32(cmd);
+	const void *maskblob = pmf_next_blob(cmd, maskbytes);
+	const void *valuesblob = pmf_next_blob(cmd, valuesbytes);
+
+	LOG_PARSE("pmf: rmw_i2c(maskbytes: %ud, valuebytes: %ud, "
+		  "totalbytes: %d) ...\n",
+		  maskbytes, valuesbytes, totalbytes);
+	LOG_BLOB("pmf:   mask data: \n", maskblob, maskbytes);
+	LOG_BLOB("pmf:   values data: \n", valuesblob, valuesbytes);
+
+	PMF_PARSE_CALL(rmw_i2c, cmd, h, maskbytes, valuesbytes, totalbytes,
+		       maskblob, valuesblob);
+}
+
+static int pmf_parser_read_cfg(struct pmf_cmd *cmd, struct pmf_handlers *h)
+{
+	u32 offset = pmf_next32(cmd);
+	u32 bytes = pmf_next32(cmd);
+
+	LOG_PARSE("pmf: read_cfg(offset: %x, bytes: %ud)\n", offset, bytes);
+
+	PMF_PARSE_CALL(read_cfg, cmd, h, offset, bytes);
+}
+
+
+static int pmf_parser_write_cfg(struct pmf_cmd *cmd, struct pmf_handlers *h)
+{
+	u32 offset = pmf_next32(cmd);
+	u32 bytes = pmf_next32(cmd);
+	const void *blob = pmf_next_blob(cmd, bytes);
+
+	LOG_PARSE("pmf: write_cfg(offset: %x, bytes: %ud)\n", offset, bytes);
+	LOG_BLOB("pmf:   data: \n", blob, bytes);
+
+	PMF_PARSE_CALL(write_cfg, cmd, h, offset, bytes, blob);
+}
+
+static int pmf_parser_rmw_cfg(struct pmf_cmd *cmd, struct pmf_handlers *h)
+{
+	u32 offset = pmf_next32(cmd);
+	u32 maskbytes = pmf_next32(cmd);
+	u32 valuesbytes = pmf_next32(cmd);
+	u32 totalbytes = pmf_next32(cmd);
+	const void *maskblob = pmf_next_blob(cmd, maskbytes);
+	const void *valuesblob = pmf_next_blob(cmd, valuesbytes);
+
+	LOG_PARSE("pmf: rmw_cfg(maskbytes: %ud, valuebytes: %ud,"
+		  " totalbytes: %d) ...\n",
+		  maskbytes, valuesbytes, totalbytes);
+	LOG_BLOB("pmf:   mask data: \n", maskblob, maskbytes);
+	LOG_BLOB("pmf:   values data: \n", valuesblob, valuesbytes);
+
+	PMF_PARSE_CALL(rmw_cfg, cmd, h, offset, maskbytes, valuesbytes,
+		       totalbytes, maskblob, valuesblob);
+}
+
+
+static int pmf_parser_read_i2c_sub(struct pmf_cmd *cmd, struct pmf_handlers *h)
+{
+	u8 subaddr = (u8)pmf_next32(cmd);
+	u32 bytes = pmf_next32(cmd);
+
+	LOG_PARSE("pmf: read_i2c_sub(subaddr: %x, bytes: %ud)\n",
+		  subaddr, bytes);
+
+	PMF_PARSE_CALL(read_i2c_sub, cmd, h, subaddr, bytes);
+}
+
+static int pmf_parser_write_i2c_sub(struct pmf_cmd *cmd, struct pmf_handlers *h)
+{
+	u8 subaddr = (u8)pmf_next32(cmd);
+	u32 bytes = pmf_next32(cmd);
+	const void *blob = pmf_next_blob(cmd, bytes);
+
+	LOG_PARSE("pmf: write_i2c_sub(subaddr: %x, bytes: %ud) ...\n",
+		  subaddr, bytes);
+	LOG_BLOB("pmf:   data: \n", blob, bytes);
+
+	PMF_PARSE_CALL(write_i2c_sub, cmd, h, subaddr, bytes, blob);
+}
+
+static int pmf_parser_set_i2c_mode(struct pmf_cmd *cmd, struct pmf_handlers *h)
+{
+	u32 mode = pmf_next32(cmd);
+
+	LOG_PARSE("pmf: set_i2c_mode(mode: %d)\n", mode);
+
+	PMF_PARSE_CALL(set_i2c_mode, cmd, h, mode);
+}
+
+
+static int pmf_parser_rmw_i2c_sub(struct pmf_cmd *cmd, struct pmf_handlers *h)
+{
+	u8 subaddr = (u8)pmf_next32(cmd);
+	u32 maskbytes = pmf_next32(cmd);
+	u32 valuesbytes = pmf_next32(cmd);
+	u32 totalbytes = pmf_next32(cmd);
+	const void *maskblob = pmf_next_blob(cmd, maskbytes);
+	const void *valuesblob = pmf_next_blob(cmd, valuesbytes);
+
+	LOG_PARSE("pmf: rmw_i2c_sub(subaddr: %x, maskbytes: %ud, valuebytes: %ud"
+		  ", totalbytes: %d) ...\n",
+		  subaddr, maskbytes, valuesbytes, totalbytes);
+	LOG_BLOB("pmf:   mask data: \n", maskblob, maskbytes);
+	LOG_BLOB("pmf:   values data: \n", valuesblob, valuesbytes);
+
+	PMF_PARSE_CALL(rmw_i2c_sub, cmd, h, subaddr, maskbytes, valuesbytes,
+		       totalbytes, maskblob, valuesblob);
+}
+
+static int pmf_parser_read_reg32_msrx(struct pmf_cmd *cmd,
+				      struct pmf_handlers *h)
+{
+	u32 offset = pmf_next32(cmd);
+	u32 mask = pmf_next32(cmd);
+	u32 shift = pmf_next32(cmd);
+	u32 xor = pmf_next32(cmd);
+
+	LOG_PARSE("pmf: read_reg32_msrx(offset: %x, mask: %x, shift: %x,"
+		  " xor: %x\n", offset, mask, shift, xor);
+
+	PMF_PARSE_CALL(read_reg32_msrx, cmd, h, offset, mask, shift, xor);
+}
+
+static int pmf_parser_read_reg16_msrx(struct pmf_cmd *cmd,
+				      struct pmf_handlers *h)
+{
+	u32 offset = pmf_next32(cmd);
+	u32 mask = pmf_next32(cmd);
+	u32 shift = pmf_next32(cmd);
+	u32 xor = pmf_next32(cmd);
+
+	LOG_PARSE("pmf: read_reg16_msrx(offset: %x, mask: %x, shift: %x,"
+		  " xor: %x\n", offset, mask, shift, xor);
+
+	PMF_PARSE_CALL(read_reg16_msrx, cmd, h, offset, mask, shift, xor);
+}
+static int pmf_parser_read_reg8_msrx(struct pmf_cmd *cmd,
+				     struct pmf_handlers *h)
+{
+	u32 offset = pmf_next32(cmd);
+	u32 mask = pmf_next32(cmd);
+	u32 shift = pmf_next32(cmd);
+	u32 xor = pmf_next32(cmd);
+
+	LOG_PARSE("pmf: read_reg8_msrx(offset: %x, mask: %x, shift: %x,"
+		  " xor: %x\n", offset, mask, shift, xor);
+
+	PMF_PARSE_CALL(read_reg8_msrx, cmd, h, offset, mask, shift, xor);
+}
+
+static int pmf_parser_write_reg32_slm(struct pmf_cmd *cmd,
+				      struct pmf_handlers *h)
+{
+	u32 offset = pmf_next32(cmd);
+	u32 shift = pmf_next32(cmd);
+	u32 mask = pmf_next32(cmd);
+
+	LOG_PARSE("pmf: write_reg32_slm(offset: %x, shift: %x, mask: %x\n",
+		  offset, shift, mask);
+
+	PMF_PARSE_CALL(write_reg32_slm, cmd, h, offset, shift, mask);
+}
+
+static int pmf_parser_write_reg16_slm(struct pmf_cmd *cmd,
+				      struct pmf_handlers *h)
+{
+	u32 offset = pmf_next32(cmd);
+	u32 shift = pmf_next32(cmd);
+	u32 mask = pmf_next32(cmd);
+
+	LOG_PARSE("pmf: write_reg16_slm(offset: %x, shift: %x, mask: %x\n",
+		  offset, shift, mask);
+
+	PMF_PARSE_CALL(write_reg16_slm, cmd, h, offset, shift, mask);
+}
+
+static int pmf_parser_write_reg8_slm(struct pmf_cmd *cmd,
+				     struct pmf_handlers *h)
+{
+	u32 offset = pmf_next32(cmd);
+	u32 shift = pmf_next32(cmd);
+	u32 mask = pmf_next32(cmd);
+
+	LOG_PARSE("pmf: write_reg8_slm(offset: %x, shift: %x, mask: %x\n",
+		  offset, shift, mask);
+
+	PMF_PARSE_CALL(write_reg8_slm, cmd, h, offset, shift, mask);
+}
+
+static int pmf_parser_mask_and_compare(struct pmf_cmd *cmd,
+				       struct pmf_handlers *h)
+{
+	u32 bytes = pmf_next32(cmd);
+	const void *maskblob = pmf_next_blob(cmd, bytes);
+	const void *valuesblob = pmf_next_blob(cmd, bytes);
+
+	LOG_PARSE("pmf: mask_and_compare(length: %ud ...\n", bytes);
+	LOG_BLOB("pmf:   mask data: \n", maskblob, bytes);
+	LOG_BLOB("pmf:   values data: \n", valuesblob, bytes);
+
+	PMF_PARSE_CALL(mask_and_compare, cmd, h,
+		       bytes, maskblob, valuesblob);
+}
+
+
+typedef int (*pmf_cmd_parser_t)(struct pmf_cmd *cmd, struct pmf_handlers *h);
+
+static pmf_cmd_parser_t pmf_parsers[PMF_CMD_COUNT] =
+{
+	NULL,
+	pmf_parser_write_gpio,
+	pmf_parser_read_gpio,
+	pmf_parser_write_reg32,
+	pmf_parser_read_reg32,
+	pmf_parser_write_reg16,
+	pmf_parser_read_reg16,
+	pmf_parser_write_reg8,
+	pmf_parser_read_reg8,
+	pmf_parser_delay,
+	pmf_parser_wait_reg32,
+	pmf_parser_wait_reg16,
+	pmf_parser_wait_reg8,
+	pmf_parser_read_i2c,
+	pmf_parser_write_i2c,
+	pmf_parser_rmw_i2c,
+	NULL, /* Bogus command */
+	NULL, /* Shift bytes right: NYI */
+	NULL, /* Shift bytes left: NYI */
+	pmf_parser_read_cfg,
+	pmf_parser_write_cfg,
+	pmf_parser_rmw_cfg,
+	pmf_parser_read_i2c_sub,
+	pmf_parser_write_i2c_sub,
+	pmf_parser_set_i2c_mode,
+	pmf_parser_rmw_i2c_sub,
+	pmf_parser_read_reg32_msrx,
+	pmf_parser_read_reg16_msrx,
+	pmf_parser_read_reg8_msrx,
+	pmf_parser_write_reg32_slm,
+	pmf_parser_write_reg16_slm,
+	pmf_parser_write_reg8_slm,
+	pmf_parser_mask_and_compare,
+};
+
+struct pmf_device {
+	struct list_head	link;
+	struct device_node	*node;
+	struct pmf_handlers	*handlers;
+	struct list_head	functions;
+	struct kref		ref;
+};
+
+static LIST_HEAD(pmf_devices);
+static spinlock_t pmf_lock = SPIN_LOCK_UNLOCKED;
+
+static void pmf_release_device(struct kref *kref)
+{
+	struct pmf_device *dev = container_of(kref, struct pmf_device, ref);
+	kfree(dev);
+}
+
+static inline void pmf_put_device(struct pmf_device *dev)
+{
+	kref_put(&dev->ref, pmf_release_device);
+}
+
+static inline struct pmf_device *pmf_get_device(struct pmf_device *dev)
+{
+	kref_get(&dev->ref);
+	return dev;
+}
+
+static inline struct pmf_device *pmf_find_device(struct device_node *np)
+{
+	struct pmf_device *dev;
+
+	list_for_each_entry(dev, &pmf_devices, link) {
+		if (dev->node == np)
+			return pmf_get_device(dev);
+	}
+	return NULL;
+}
+
+static int pmf_parse_one(struct pmf_function *func,
+			 struct pmf_handlers *handlers,
+			 void *instdata, struct pmf_args *args)
+{
+	struct pmf_cmd cmd;
+	u32 ccode;
+	int count, rc;
+
+	cmd.cmdptr		= func->data;
+	cmd.cmdend		= func->data + func->length;
+	cmd.func       		= func;
+	cmd.instdata		= instdata;
+	cmd.args		= args;
+	cmd.error		= 0;
+
+	LOG_PARSE("pmf: func %s, %d bytes, %s...\n",
+		  func->name, func->length,
+		  handlers ? "executing" : "parsing");
+
+	/* One subcommand to parse for now */
+	count = 1;
+
+	while(count-- && cmd.cmdptr < cmd.cmdend) {
+		/* Get opcode */
+		ccode = pmf_next32(&cmd);
+		/* Check if we are hitting a command list, fetch new count */
+		if (ccode == 0) {
+			count = pmf_next32(&cmd) - 1;
+			ccode = pmf_next32(&cmd);
+		}
+		if (cmd.error) {
+			LOG_ERROR("pmf: parse error, not enough data\n");
+			return -ENXIO;
+		}
+		if (ccode >= PMF_CMD_COUNT) {
+			LOG_ERROR("pmf: command code %d unknown !\n", ccode);
+			return -ENXIO;
+		}
+		if (pmf_parsers[ccode] == NULL) {
+			LOG_ERROR("pmf: no parser for command %d !\n", ccode);
+			return -ENXIO;
+		}
+		rc = pmf_parsers[ccode](&cmd, handlers);
+		if (rc != 0) {
+			LOG_ERROR("pmf: parser for command %d returned"
+				  " error %d\n", ccode, rc);
+			return rc;
+		}
+	}
+
+	/* We are doing an initial parse pass, we need to adjust the size */
+	if (handlers == NULL)
+		func->length = cmd.cmdptr - func->data;
+
+	return 0;
+}
+
+static int pmf_add_function_prop(struct pmf_device *dev, void *driverdata,
+				 const char *name, u32 *data,
+				 unsigned int length)
+{
+	int count = 0;
+	struct pmf_function *func = NULL;
+
+	DBG("pmf: Adding functions for platform-do-%s\n", name);
+
+	while (length >= 12) {
+		/* Allocate a structure */
+		func = kzalloc(sizeof(struct pmf_function), GFP_KERNEL);
+		if (func == NULL)
+			goto bail;
+		kref_init(&func->ref);
+		INIT_LIST_HEAD(&func->irq_clients);
+		func->node = dev->node;
+		func->driver_data = driverdata;
+		func->name = name;
+		func->phandle = data[0];
+		func->flags = data[1];
+		data += 2;
+		length -= 8;
+		func->data = data;
+		func->length = length;
+		func->dev = dev;
+		DBG("pmf: idx %d: flags=%08x, phandle=%08x "
+		    " %d bytes remaining, parsing...\n",
+		    count+1, func->flags, func->phandle, length);
+		if (pmf_parse_one(func, NULL, NULL, NULL)) {
+			kfree(func);
+			goto bail;
+		}
+		length -= func->length;
+		data = (u32 *)(((u8 *)data) + func->length);
+		list_add(&func->link, &dev->functions);
+		pmf_get_device(dev);
+		count++;
+	}
+ bail:
+	DBG("pmf: Added %d functions\n", count);
+
+	return count;
+}
+
+static int pmf_add_functions(struct pmf_device *dev, void *driverdata)
+{
+	struct property *pp;
+#define PP_PREFIX "platform-do-"
+	const int plen = strlen(PP_PREFIX);
+	int count = 0;
+
+	for (pp = dev->node->properties; pp != 0; pp = pp->next) {
+		char *name;
+		if (strncmp(pp->name, PP_PREFIX, plen) != 0)
+			continue;
+		name = pp->name + plen;
+		if (strlen(name) && pp->length >= 12)
+			count += pmf_add_function_prop(dev, driverdata, name,
+						       (u32 *)pp->value,
+						       pp->length);
+	}
+	return count;
+}
+
+
+int pmf_register_driver(struct device_node *np,
+			struct pmf_handlers *handlers,
+			void *driverdata)
+{
+	struct pmf_device *dev;
+	unsigned long flags;
+	int rc = 0;
+
+	if (handlers == NULL)
+		return -EINVAL;
+
+	DBG("pmf: registering driver for node %s\n", np->full_name);
+
+	spin_lock_irqsave(&pmf_lock, flags);
+	dev = pmf_find_device(np);
+	spin_unlock_irqrestore(&pmf_lock, flags);
+	if (dev != NULL) {
+		DBG("pmf: already there !\n");
+		pmf_put_device(dev);
+		return -EBUSY;
+	}
+
+	dev = kzalloc(sizeof(struct pmf_device), GFP_KERNEL);
+	if (dev == NULL) {
+		DBG("pmf: no memory !\n");
+		return -ENOMEM;
+	}
+	kref_init(&dev->ref);
+	dev->node = of_node_get(np);
+	dev->handlers = handlers;
+	INIT_LIST_HEAD(&dev->functions);
+
+	rc = pmf_add_functions(dev, driverdata);
+	if (rc == 0) {
+		DBG("pmf: no functions, disposing.. \n");
+		of_node_put(np);
+		kfree(dev);
+		return -ENODEV;
+	}
+
+	spin_lock_irqsave(&pmf_lock, flags);
+	list_add(&dev->link, &pmf_devices);
+	spin_unlock_irqrestore(&pmf_lock, flags);
+
+	return 0;
+}
+EXPORT_SYMBOL_GPL(pmf_register_driver);
+
+struct pmf_function *pmf_get_function(struct pmf_function *func)
+{
+	if (!try_module_get(func->dev->handlers->owner))
+		return NULL;
+	kref_get(&func->ref);
+	return func;
+}
+EXPORT_SYMBOL_GPL(pmf_get_function);
+
+static void pmf_release_function(struct kref *kref)
+{
+	struct pmf_function *func =
+		container_of(kref, struct pmf_function, ref);
+	pmf_put_device(func->dev);
+	kfree(func);
+}
+
+static inline void __pmf_put_function(struct pmf_function *func)
+{
+	kref_put(&func->ref, pmf_release_function);
+}
+
+void pmf_put_function(struct pmf_function *func)
+{
+	if (func == NULL)
+		return;
+	module_put(func->dev->handlers->owner);
+	__pmf_put_function(func);
+}
+EXPORT_SYMBOL_GPL(pmf_put_function);
+
+void pmf_unregister_driver(struct device_node *np)
+{
+	struct pmf_device *dev;
+	unsigned long flags;
+
+	DBG("pmf: unregistering driver for node %s\n", np->full_name);
+
+	spin_lock_irqsave(&pmf_lock, flags);
+	dev = pmf_find_device(np);
+	if (dev == NULL) {
+		DBG("pmf: not such driver !\n");
+		spin_unlock_irqrestore(&pmf_lock, flags);
+		return;
+	}
+	list_del(&dev->link);
+
+	while(!list_empty(&dev->functions)) {
+		struct pmf_function *func =
+			list_entry(dev->functions.next, typeof(*func), link);
+		list_del(&func->link);
+		__pmf_put_function(func);
+	}
+
+	pmf_put_device(dev);
+	spin_unlock_irqrestore(&pmf_lock, flags);
+}
+EXPORT_SYMBOL_GPL(pmf_unregister_driver);
+
+struct pmf_function *__pmf_find_function(struct device_node *target,
+					 const char *name, u32 flags)
+{
+	struct device_node *actor = of_node_get(target);
+	struct pmf_device *dev;
+	struct pmf_function *func, *result = NULL;
+	char fname[64];
+	u32 *prop, ph;
+
+	/*
+	 * Look for a "platform-*" function reference. If we can't find
+	 * one, then we fallback to a direct call attempt
+	 */
+	snprintf(fname, 63, "platform-%s", name);
+	prop = (u32 *)get_property(target, fname, NULL);
+	if (prop == NULL)
+		goto find_it;
+	ph = *prop;
+	if (ph == 0)
+		goto find_it;
+
+	/*
+	 * Ok, now try to find the actor. If we can't find it, we fail,
+	 * there is no point in falling back there
+	 */
+	of_node_put(actor);
+	actor = of_find_node_by_phandle(ph);
+	if (actor == NULL)
+		return NULL;
+ find_it:
+	dev = pmf_find_device(actor);
+	if (dev == NULL)
+		return NULL;
+
+	list_for_each_entry(func, &dev->functions, link) {
+		if (name && strcmp(name, func->name))
+			continue;
+		if (func->phandle && target->node != func->phandle)
+			continue;
+		if ((func->flags & flags) == 0)
+			continue;
+		result = func;
+		break;
+	}
+	of_node_put(actor);
+	pmf_put_device(dev);
+	return result;
+}
+
+
+int pmf_register_irq_client(struct device_node *target,
+			    const char *name,
+			    struct pmf_irq_client *client)
+{
+	struct pmf_function *func;
+	unsigned long flags;
+
+	spin_lock_irqsave(&pmf_lock, flags);
+	func = __pmf_find_function(target, name, PMF_FLAGS_INT_GEN);
+	if (func == NULL) {
+		spin_unlock_irqrestore(&pmf_lock, flags);
+		return -ENODEV;
+	}
+	list_add(&client->link, &func->irq_clients);
+	spin_unlock_irqrestore(&pmf_lock, flags);
+
+	return 0;
+}
+EXPORT_SYMBOL_GPL(pmf_register_irq_client);
+
+void pmf_unregister_irq_client(struct device_node *np,
+			      const char *name,
+			      struct pmf_irq_client *client)
+{
+	unsigned long flags;
+
+	spin_lock_irqsave(&pmf_lock, flags);
+	list_del(&client->link);
+	spin_unlock_irqrestore(&pmf_lock, flags);
+}
+EXPORT_SYMBOL_GPL(pmf_unregister_irq_client);
+
+
+void pmf_do_irq(struct pmf_function *func)
+{
+	unsigned long flags;
+	struct pmf_irq_client *client;
+
+	/* For now, using a spinlock over the whole function. Can be made
+	 * to drop the lock using 2 lists if necessary
+	 */
+	spin_lock_irqsave(&pmf_lock, flags);
+	list_for_each_entry(client, &func->irq_clients, link) {
+		if (!try_module_get(client->owner))
+			continue;
+		client->handler(client->data);
+		module_put(client->owner);
+	}
+	spin_unlock_irqrestore(&pmf_lock, flags);
+}
+EXPORT_SYMBOL_GPL(pmf_do_irq);
+
+
+int pmf_call_one(struct pmf_function *func, struct pmf_args *args)
+{
+	struct pmf_device *dev = func->dev;
+	void *instdata = NULL;
+	int rc = 0;
+
+	DBG(" ** pmf_call_one(%s/%s) **\n", dev->node->full_name, func->name);
+
+	if (dev->handlers->begin)
+		instdata = dev->handlers->begin(func, args);
+	rc = pmf_parse_one(func, dev->handlers, instdata, args);
+	if (dev->handlers->end)
+		dev->handlers->end(func, instdata);
+
+	return rc;
+}
+EXPORT_SYMBOL_GPL(pmf_call_one);
+
+int pmf_do_functions(struct device_node *np, const char *name,
+		     u32 phandle, u32 fflags, struct pmf_args *args)
+{
+	struct pmf_device *dev;
+	struct pmf_function *func, *tmp;
+	unsigned long flags;
+	int rc = -ENODEV;
+
+	spin_lock_irqsave(&pmf_lock, flags);
+
+	dev = pmf_find_device(np);
+	if (dev == NULL) {
+		spin_unlock_irqrestore(&pmf_lock, flags);
+		return -ENODEV;
+	}
+	list_for_each_entry_safe(func, tmp, &dev->functions, link) {
+		if (name && strcmp(name, func->name))
+			continue;
+		if (phandle && func->phandle && phandle != func->phandle)
+			continue;
+		if ((func->flags & fflags) == 0)
+			continue;
+		if (pmf_get_function(func) == NULL)
+			continue;
+		spin_unlock_irqrestore(&pmf_lock, flags);
+		rc = pmf_call_one(func, args);
+		pmf_put_function(func);
+		spin_lock_irqsave(&pmf_lock, flags);
+	}
+	pmf_put_device(dev);
+	spin_unlock_irqrestore(&pmf_lock, flags);
+
+	return rc;
+}
+EXPORT_SYMBOL_GPL(pmf_do_functions);
+
+
+struct pmf_function *pmf_find_function(struct device_node *target,
+				       const char *name)
+{
+	struct pmf_function *func;
+	unsigned long flags;
+
+	spin_lock_irqsave(&pmf_lock, flags);
+	func = __pmf_find_function(target, name, PMF_FLAGS_ON_DEMAND);
+	if (func)
+		func = pmf_get_function(func);
+	spin_unlock_irqrestore(&pmf_lock, flags);
+	return func;
+}
+EXPORT_SYMBOL_GPL(pmf_find_function);
+
+int pmf_call_function(struct device_node *target, const char *name,
+		      struct pmf_args *args)
+{
+	struct pmf_function *func = pmf_find_function(target, name);
+	int rc;
+
+	if (func == NULL)
+		return -ENODEV;
+
+	rc = pmf_call_one(func, args);
+	pmf_put_function(func);
+	return rc;
+}
+EXPORT_SYMBOL_GPL(pmf_call_function);
+
diff --git a/arch/powerpc/platforms/powermac/pic.c b/arch/powerpc/platforms/powermac/pic.c
index 90040c49494d..18bf3011d1e3 100644
--- a/arch/powerpc/platforms/powermac/pic.c
+++ b/arch/powerpc/platforms/powermac/pic.c
@@ -5,8 +5,8 @@
  *  in a separate file
  *
  *  Copyright (C) 1997 Paul Mackerras (paulus@samba.org)
- *
- *  Maintained by Benjamin Herrenschmidt (benh@kernel.crashing.org)
+ *  Copyright (C) 2005 Benjamin Herrenschmidt (benh@kernel.crashing.org)
+ *                     IBM, Corp.
  *
  *  This program is free software; you can redistribute it and/or
  *  modify it under the terms of the GNU General Public License
@@ -54,12 +54,7 @@ struct pmac_irq_hw {
 };
 
 /* Default addresses */
-static volatile struct pmac_irq_hw *pmac_irq_hw[4] = {
-        (struct pmac_irq_hw *) 0xf3000020,
-        (struct pmac_irq_hw *) 0xf3000010,
-        (struct pmac_irq_hw *) 0xf4000020,
-        (struct pmac_irq_hw *) 0xf4000010,
-};
+static volatile struct pmac_irq_hw __iomem *pmac_irq_hw[4];
 
 #define GC_LEVEL_MASK		0x3ff00000
 #define OHARE_LEVEL_MASK	0x1ff00000
@@ -82,8 +77,7 @@ static unsigned long ppc_lost_interrupts[NR_MASK_WORDS];
  * since it can lose interrupts (see pmac_set_irq_mask).
  * -- Cort
  */
-void
-__set_lost(unsigned long irq_nr, int nokick)
+void __set_lost(unsigned long irq_nr, int nokick)
 {
 	if (!test_and_set_bit(irq_nr, ppc_lost_interrupts)) {
 		atomic_inc(&ppc_n_lost_interrupts);
@@ -92,8 +86,7 @@ __set_lost(unsigned long irq_nr, int nokick)
 	}
 }
 
-static void
-pmac_mask_and_ack_irq(unsigned int irq_nr)
+static void pmac_mask_and_ack_irq(unsigned int irq_nr)
 {
         unsigned long bit = 1UL << (irq_nr & 0x1f);
         int i = irq_nr >> 5;
@@ -224,8 +217,7 @@ static irqreturn_t gatwick_action(int cpl, void *dev_id, struct pt_regs *regs)
 	return IRQ_NONE;
 }
 
-int
-pmac_get_irq(struct pt_regs *regs)
+static int pmac_get_irq(struct pt_regs *regs)
 {
 	int irq;
 	unsigned long bits = 0;
@@ -256,34 +248,40 @@ pmac_get_irq(struct pt_regs *regs)
 
 /* This routine will fix some missing interrupt values in the device tree
  * on the gatwick mac-io controller used by some PowerBooks
+ *
+ * Walking of OF nodes could use a bit more fixing up here, but it's not
+ * very important as this is all boot time code on static portions of the
+ * device-tree.
+ *
+ * However, the modifications done to "intrs" will have to be removed and
+ * replaced with proper updates of the "interrupts" properties or
+ * AAPL,interrupts, yet to be decided, once the dynamic parsing is there.
  */
-static void __init
-pmac_fix_gatwick_interrupts(struct device_node *gw, int irq_base)
+static void __init pmac_fix_gatwick_interrupts(struct device_node *gw,
+					       int irq_base)
 {
 	struct device_node *node;
 	int count;
 
 	memset(gatwick_int_pool, 0, sizeof(gatwick_int_pool));
-	node = gw->child;
 	count = 0;
-	while(node)
-	{
+	for (node = NULL; (node = of_get_next_child(gw, node)) != NULL;) {
 		/* Fix SCC */
-		if (strcasecmp(node->name, "escc") == 0)
-			if (node->child) {
-				if (node->child->n_intrs < 3) {
-					node->child->intrs = &gatwick_int_pool[count];
-					count += 3;
-				}
-				node->child->n_intrs = 3;
-				node->child->intrs[0].line = 15+irq_base;
-				node->child->intrs[1].line =  4+irq_base;
-				node->child->intrs[2].line =  5+irq_base;
-				printk(KERN_INFO "irq: fixed SCC on second controller (%d,%d,%d)\n",
-					node->child->intrs[0].line,
-					node->child->intrs[1].line,
-					node->child->intrs[2].line);
+		if ((strcasecmp(node->name, "escc") == 0) && node->child) {
+			if (node->child->n_intrs < 3) {
+				node->child->intrs = &gatwick_int_pool[count];
+				count += 3;
 			}
+			node->child->n_intrs = 3;
+			node->child->intrs[0].line = 15+irq_base;
+			node->child->intrs[1].line =  4+irq_base;
+			node->child->intrs[2].line =  5+irq_base;
+			printk(KERN_INFO "irq: fixed SCC on gatwick"
+			       " (%d,%d,%d)\n",
+			       node->child->intrs[0].line,
+			       node->child->intrs[1].line,
+			       node->child->intrs[2].line);
+		}
 		/* Fix media-bay & left SWIM */
 		if (strcasecmp(node->name, "media-bay") == 0) {
 			struct device_node* ya_node;
@@ -292,12 +290,11 @@ pmac_fix_gatwick_interrupts(struct device_node *gw, int irq_base)
 				node->intrs = &gatwick_int_pool[count++];
 			node->n_intrs = 1;
 			node->intrs[0].line = 29+irq_base;
-			printk(KERN_INFO "irq: fixed media-bay on second controller (%d)\n",
-					node->intrs[0].line);
+			printk(KERN_INFO "irq: fixed media-bay on gatwick"
+			       " (%d)\n", node->intrs[0].line);
 
 			ya_node = node->child;
-			while(ya_node)
-			{
+			while(ya_node) {
 				if (strcasecmp(ya_node->name, "floppy") == 0) {
 					if (ya_node->n_intrs < 2) {
 						ya_node->intrs = &gatwick_int_pool[count];
@@ -323,7 +320,6 @@ pmac_fix_gatwick_interrupts(struct device_node *gw, int irq_base)
 				ya_node = ya_node->sibling;
 			}
 		}
-		node = node->sibling;
 	}
 	if (count > 10) {
 		printk("WARNING !! Gatwick interrupt pool overflow\n");
@@ -338,45 +334,41 @@ pmac_fix_gatwick_interrupts(struct device_node *gw, int irq_base)
  * controller.  If we find this second ohare, set it up and fix the
  * interrupt value in the device tree for the ethernet chip.
  */
-static int __init enable_second_ohare(void)
+static void __init enable_second_ohare(struct device_node *np)
 {
 	unsigned char bus, devfn;
 	unsigned short cmd;
-        unsigned long addr;
-	struct device_node *irqctrler = find_devices("pci106b,7");
 	struct device_node *ether;
 
-	if (irqctrler == NULL || irqctrler->n_addrs <= 0)
-		return -1;
-	addr = (unsigned long) ioremap(irqctrler->addrs[0].address, 0x40);
-	pmac_irq_hw[1] = (volatile struct pmac_irq_hw *)(addr + 0x20);
-	max_irqs = 64;
-	if (pci_device_from_OF_node(irqctrler, &bus, &devfn) == 0) {
-		struct pci_controller* hose = pci_find_hose_for_OF_device(irqctrler);
-		if (!hose)
-		    printk(KERN_ERR "Can't find PCI hose for OHare2 !\n");
-		else {
-		    early_read_config_word(hose, bus, devfn, PCI_COMMAND, &cmd);
-		    cmd |= PCI_COMMAND_MEMORY | PCI_COMMAND_MASTER;
-	  	    cmd &= ~PCI_COMMAND_IO;
-		    early_write_config_word(hose, bus, devfn, PCI_COMMAND, cmd);
+	/* This code doesn't strictly belong here, it could be part of
+	 * either the PCI initialisation or the feature code. It's kept
+	 * here for historical reasons.
+	 */
+	if (pci_device_from_OF_node(np, &bus, &devfn) == 0) {
+		struct pci_controller* hose =
+			pci_find_hose_for_OF_device(np);
+		if (!hose) {
+			printk(KERN_ERR "Can't find PCI hose for OHare2 !\n");
+			return;
 		}
+		early_read_config_word(hose, bus, devfn, PCI_COMMAND, &cmd);
+		cmd |= PCI_COMMAND_MEMORY | PCI_COMMAND_MASTER;
+		cmd &= ~PCI_COMMAND_IO;
+		early_write_config_word(hose, bus, devfn, PCI_COMMAND, cmd);
 	}
 
 	/* Fix interrupt for the modem/ethernet combo controller. The number
-	   in the device tree (27) is bogus (correct for the ethernet-only
-	   board but not the combo ethernet/modem board).
-	   The real interrupt is 28 on the second controller -> 28+32 = 60.
-	*/
-	ether = find_devices("pci1011,14");
+	 * in the device tree (27) is bogus (correct for the ethernet-only
+	 * board but not the combo ethernet/modem board).
+	 * The real interrupt is 28 on the second controller -> 28+32 = 60.
+	 */
+	ether = of_find_node_by_name(NULL, "pci1011,14");
 	if (ether && ether->n_intrs > 0) {
 		ether->intrs[0].line = 60;
 		printk(KERN_INFO "irq: Fixed ethernet IRQ to %d\n",
 		       ether->intrs[0].line);
 	}
-
-	/* Return the interrupt number of the cascade */
-	return irqctrler->intrs[0].line;
+	of_node_put(ether);
 }
 
 #ifdef CONFIG_XMON
@@ -394,189 +386,251 @@ static struct irqaction gatwick_cascade_action = {
 	.mask		= CPU_MASK_NONE,
 	.name		= "cascade",
 };
-#endif /* CONFIG_PPC32 */
 
-static int pmac_u3_cascade(struct pt_regs *regs, void *data)
+static void __init pmac_pic_probe_oldstyle(void)
 {
-	return mpic_get_one_irq((struct mpic *)data, regs);
-}
-
-void __init pmac_pic_init(void)
-{
-        struct device_node *irqctrler  = NULL;
-        struct device_node *irqctrler2 = NULL;
-	struct device_node *np;
-#ifdef CONFIG_PPC32
         int i;
-        unsigned long addr;
 	int irq_cascade = -1;
-#endif
-	struct mpic *mpic1, *mpic2;
+        struct device_node *master = NULL;
+	struct device_node *slave = NULL;
+	u8 __iomem *addr;
+	struct resource r;
 
-	/* We first try to detect Apple's new Core99 chipset, since mac-io
-	 * is quite different on those machines and contains an IBM MPIC2.
-	 */
-	np = find_type_devices("open-pic");
-	while (np) {
-		if (np->parent && !strcmp(np->parent->name, "u3"))
-			irqctrler2 = np;
-		else
-			irqctrler = np;
-		np = np->next;
-	}
-	if (irqctrler != NULL && irqctrler->n_addrs > 0) {
-		unsigned char senses[128];
-
-		printk(KERN_INFO "PowerMac using OpenPIC irq controller at 0x%08x\n",
-		       (unsigned int)irqctrler->addrs[0].address);
-		pmac_call_feature(PMAC_FTR_ENABLE_MPIC, irqctrler, 0, 0);
-
-		prom_get_irq_senses(senses, 0, 128);
-		mpic1 = mpic_alloc(irqctrler->addrs[0].address,
-				   MPIC_PRIMARY | MPIC_WANTS_RESET,
-				   0, 0, 128, 252, senses, 128, " OpenPIC  ");
-		BUG_ON(mpic1 == NULL);
-		mpic_init(mpic1);		
-
-		if (irqctrler2 != NULL && irqctrler2->n_intrs > 0 &&
-		    irqctrler2->n_addrs > 0) {
-			printk(KERN_INFO "Slave OpenPIC at 0x%08x hooked on IRQ %d\n",
-			       (u32)irqctrler2->addrs[0].address,
-			       irqctrler2->intrs[0].line);
-
-			pmac_call_feature(PMAC_FTR_ENABLE_MPIC, irqctrler2, 0, 0);
-			prom_get_irq_senses(senses, 128, 128 + 124);
-
-			/* We don't need to set MPIC_BROKEN_U3 here since we don't have
-			 * hypertransport interrupts routed to it
-			 */
-			mpic2 = mpic_alloc(irqctrler2->addrs[0].address,
-					   MPIC_BIG_ENDIAN | MPIC_WANTS_RESET,
-					   0, 128, 124, 0, senses, 124,
-					   " U3-MPIC  ");
-			BUG_ON(mpic2 == NULL);
-			mpic_init(mpic2);
-			mpic_setup_cascade(irqctrler2->intrs[0].line,
-					   pmac_u3_cascade, mpic2);
-		}
-#if defined(CONFIG_XMON) && defined(CONFIG_PPC32)
-		{
-			struct device_node* pswitch;
-			int nmi_irq;
-
-			pswitch = find_devices("programmer-switch");
-			if (pswitch && pswitch->n_intrs) {
-				nmi_irq = pswitch->intrs[0].line;
-				mpic_irq_set_priority(nmi_irq, 9);
-				setup_irq(nmi_irq, &xmon_action);
-			}
-		}
-#endif	/* CONFIG_XMON */
-		return;
-	}
-	irqctrler = NULL;
+	/* Set our get_irq function */
+	ppc_md.get_irq = pmac_get_irq;
 
-#ifdef CONFIG_PPC32
-	/* Get the level/edge settings, assume if it's not
-	 * a Grand Central nor an OHare, then it's an Heathrow
-	 * (or Paddington).
+	/*
+	 * Find the interrupt controller type & node
 	 */
-	ppc_md.get_irq = pmac_get_irq;
-	if (find_devices("gc"))
+
+	if ((master = of_find_node_by_name(NULL, "gc")) != NULL) {
+		max_irqs = max_real_irqs = 32;
 		level_mask[0] = GC_LEVEL_MASK;
-	else if (find_devices("ohare")) {
+	} else if ((master = of_find_node_by_name(NULL, "ohare")) != NULL) {
+		max_irqs = max_real_irqs = 32;
 		level_mask[0] = OHARE_LEVEL_MASK;
+
 		/* We might have a second cascaded ohare */
-		level_mask[1] = OHARE_LEVEL_MASK;
-	} else {
+		slave = of_find_node_by_name(NULL, "pci106b,7");
+		if (slave) {
+			max_irqs = 64;
+			level_mask[1] = OHARE_LEVEL_MASK;
+			enable_second_ohare(slave);
+		}
+	} else if ((master = of_find_node_by_name(NULL, "mac-io")) != NULL) {
+		max_irqs = max_real_irqs = 64;
 		level_mask[0] = HEATHROW_LEVEL_MASK;
 		level_mask[1] = 0;
+
 		/* We might have a second cascaded heathrow */
-		level_mask[2] = HEATHROW_LEVEL_MASK;
-		level_mask[3] = 0;
-	}
+		slave = of_find_node_by_name(master, "mac-io");
+
+		/* Check ordering of master & slave */
+		if (device_is_compatible(master, "gatwick")) {
+			struct device_node *tmp;
+			BUG_ON(slave == NULL);
+			tmp = master;
+			master = slave;
+			slave = tmp;
+		}
 
-	/*
-	 * G3 powermacs and 1999 G3 PowerBooks have 64 interrupts,
-	 * 1998 G3 Series PowerBooks have 128,
-	 * other powermacs have 32.
-	 * The combo ethernet/modem card for the Powerstar powerbooks
-	 * (2400/3400/3500, ohare based) has a second ohare chip
-	 * effectively making a total of 64.
-	 */
-	max_irqs = max_real_irqs = 32;
-	irqctrler = find_devices("mac-io");
-	if (irqctrler)
-	{
-		max_real_irqs = 64;
-		if (irqctrler->next)
+		/* We found a slave */
+		if (slave) {
 			max_irqs = 128;
-		else
-			max_irqs = 64;
+			level_mask[2] = HEATHROW_LEVEL_MASK;
+			level_mask[3] = 0;
+			pmac_fix_gatwick_interrupts(slave, max_real_irqs);
+		}
 	}
+	BUG_ON(master == NULL);
+
+	/* Set the handler for the main PIC */
 	for ( i = 0; i < max_real_irqs ; i++ )
 		irq_desc[i].handler = &pmac_pic;
 
-	/* get addresses of first controller */
-	if (irqctrler) {
-		if  (irqctrler->n_addrs > 0) {
-			addr = (unsigned long)
-				ioremap(irqctrler->addrs[0].address, 0x40);
-			for (i = 0; i < 2; ++i)
-				pmac_irq_hw[i] = (volatile struct pmac_irq_hw*)
-					(addr + (2 - i) * 0x10);
-		}
+	/* Get addresses of first controller if we have a node for it */
+	BUG_ON(of_address_to_resource(master, 0, &r));
 
-		/* get addresses of second controller */
-		irqctrler = irqctrler->next;
-		if (irqctrler && irqctrler->n_addrs > 0) {
-			addr = (unsigned long)
-				ioremap(irqctrler->addrs[0].address, 0x40);
-			for (i = 2; i < 4; ++i)
-				pmac_irq_hw[i] = (volatile struct pmac_irq_hw*)
-					(addr + (4 - i) * 0x10);
-			irq_cascade = irqctrler->intrs[0].line;
-			if (device_is_compatible(irqctrler, "gatwick"))
-				pmac_fix_gatwick_interrupts(irqctrler, max_real_irqs);
-		}
-	} else {
-		/* older powermacs have a GC (grand central) or ohare at
-		   f3000000, with interrupt control registers at f3000020. */
-		addr = (unsigned long) ioremap(0xf3000000, 0x40);
-		pmac_irq_hw[0] = (volatile struct pmac_irq_hw *) (addr + 0x20);
+	/* Map interrupts of primary controller */
+	addr = (u8 __iomem *) ioremap(r.start, 0x40);
+	i = 0;
+	pmac_irq_hw[i++] = (volatile struct pmac_irq_hw __iomem *)
+		(addr + 0x20);
+	if (max_real_irqs > 32)
+		pmac_irq_hw[i++] = (volatile struct pmac_irq_hw __iomem *)
+			(addr + 0x10);
+	of_node_put(master);
+
+	printk(KERN_INFO "irq: Found primary Apple PIC %s for %d irqs\n",
+	       master->full_name, max_real_irqs);
+
+	/* Map interrupts of cascaded controller */
+	if (slave && !of_address_to_resource(slave, 0, &r)) {
+		addr = (u8 __iomem *)ioremap(r.start, 0x40);
+		pmac_irq_hw[i++] = (volatile struct pmac_irq_hw __iomem *)
+			(addr + 0x20);
+		if (max_irqs > 64)
+			pmac_irq_hw[i++] =
+				(volatile struct pmac_irq_hw __iomem *)
+				(addr + 0x10);
+		irq_cascade = slave->intrs[0].line;
+
+		printk(KERN_INFO "irq: Found slave Apple PIC %s for %d irqs"
+		       " cascade: %d\n", slave->full_name,
+		       max_irqs - max_real_irqs, irq_cascade);
 	}
-
-	/* PowerBooks 3400 and 3500 can have a second controller in a second
-	   ohare chip, on the combo ethernet/modem card */
-	if (machine_is_compatible("AAPL,3400/2400")
-	     || machine_is_compatible("AAPL,3500"))
-		irq_cascade = enable_second_ohare();
+	of_node_put(slave);
 
 	/* disable all interrupts in all controllers */
 	for (i = 0; i * 32 < max_irqs; ++i)
 		out_le32(&pmac_irq_hw[i]->enable, 0);
+
 	/* mark level interrupts */
 	for (i = 0; i < max_irqs; i++)
 		if (level_mask[i >> 5] & (1UL << (i & 0x1f)))
 			irq_desc[i].status = IRQ_LEVEL;
 
-	/* get interrupt line of secondary interrupt controller */
-	if (irq_cascade >= 0) {
-		printk(KERN_INFO "irq: secondary controller on irq %d\n",
-			(int)irq_cascade);
+	/* Setup handlers for secondary controller and hook cascade irq*/
+	if (slave) {
 		for ( i = max_real_irqs ; i < max_irqs ; i++ )
 			irq_desc[i].handler = &gatwick_pic;
 		setup_irq(irq_cascade, &gatwick_cascade_action);
 	}
-	printk("System has %d possible interrupts\n", max_irqs);
-	if (max_irqs != max_real_irqs)
-		printk(KERN_DEBUG "%d interrupts on main controller\n",
-			max_real_irqs);
-
+	printk(KERN_INFO "irq: System has %d possible interrupts\n", max_irqs);
 #ifdef CONFIG_XMON
 	setup_irq(20, &xmon_action);
-#endif	/* CONFIG_XMON */
-#endif	/* CONFIG_PPC32 */
+#endif
+}
+#endif /* CONFIG_PPC32 */
+
+static int pmac_u3_cascade(struct pt_regs *regs, void *data)
+{
+	return mpic_get_one_irq((struct mpic *)data, regs);
+}
+
+static void __init pmac_pic_setup_mpic_nmi(struct mpic *mpic)
+{
+#if defined(CONFIG_XMON) && defined(CONFIG_PPC32)
+	struct device_node* pswitch;
+	int nmi_irq;
+
+	pswitch = of_find_node_by_name(NULL, "programmer-switch");
+	if (pswitch && pswitch->n_intrs) {
+		nmi_irq = pswitch->intrs[0].line;
+		mpic_irq_set_priority(nmi_irq, 9);
+		setup_irq(nmi_irq, &xmon_action);
+	}
+	of_node_put(pswitch);
+#endif	/* defined(CONFIG_XMON) && defined(CONFIG_PPC32) */
+}
+
+static struct mpic * __init pmac_setup_one_mpic(struct device_node *np,
+						int master)
+{
+	unsigned char senses[128];
+	int offset = master ? 0 : 128;
+	int count = master ? 128 : 124;
+	const char *name = master ? " MPIC 1   " : " MPIC 2   ";
+	struct resource r;
+	struct mpic *mpic;
+	unsigned int flags = master ? MPIC_PRIMARY : 0;
+	int rc;
+
+	rc = of_address_to_resource(np, 0, &r);
+	if (rc)
+		return NULL;
+
+	pmac_call_feature(PMAC_FTR_ENABLE_MPIC, np, 0, 0);
+
+	prom_get_irq_senses(senses, offset, offset + count);
+
+	flags |= MPIC_WANTS_RESET;
+	if (get_property(np, "big-endian", NULL))
+		flags |= MPIC_BIG_ENDIAN;
+
+	/* Primary Big Endian means HT interrupts. This is quite dodgy
+	 * but works until I find a better way
+	 */
+	if (master && (flags & MPIC_BIG_ENDIAN))
+		flags |= MPIC_BROKEN_U3;
+
+	mpic = mpic_alloc(r.start, flags, 0, offset, count, master ? 252 : 0,
+			  senses, count, name);
+	if (mpic == NULL)
+		return NULL;
+
+	mpic_init(mpic);
+
+	return mpic;
+ }
+
+static int __init pmac_pic_probe_mpic(void)
+{
+	struct mpic *mpic1, *mpic2;
+	struct device_node *np, *master = NULL, *slave = NULL;
+
+	/* We can have up to 2 MPICs cascaded */
+	for (np = NULL; (np = of_find_node_by_type(np, "open-pic"))
+		     != NULL;) {
+		if (master == NULL &&
+		    get_property(np, "interrupts", NULL) == NULL)
+			master = of_node_get(np);
+		else if (slave == NULL)
+			slave = of_node_get(np);
+		if (master && slave)
+			break;
+	}
+
+	/* Check for bogus setups */
+	if (master == NULL && slave != NULL) {
+		master = slave;
+		slave = NULL;
+	}
+
+	/* Not found, default to good old pmac pic */
+	if (master == NULL)
+		return -ENODEV;
+
+	/* Set master handler */
+	ppc_md.get_irq = mpic_get_irq;
+
+	/* Setup master */
+	mpic1 = pmac_setup_one_mpic(master, 1);
+	BUG_ON(mpic1 == NULL);
+
+	/* Install NMI if any */
+	pmac_pic_setup_mpic_nmi(mpic1);
+
+	of_node_put(master);
+
+	/* No slave, let's go out */
+	if (slave == NULL || slave->n_intrs < 1)
+		return 0;
+
+	mpic2 = pmac_setup_one_mpic(slave, 0);
+	if (mpic2 == NULL) {
+		printk(KERN_ERR "Failed to setup slave MPIC\n");
+		of_node_put(slave);
+		return 0;
+	}
+	mpic_setup_cascade(slave->intrs[0].line, pmac_u3_cascade, mpic2);
+
+	of_node_put(slave);
+	return 0;
+}
+
+
+void __init pmac_pic_init(void)
+{
+	/* We first try to detect Apple's new Core99 chipset, since mac-io
+	 * is quite different on those machines and contains an IBM MPIC2.
+	 */
+	if (pmac_pic_probe_mpic() == 0)
+		return;
+
+#ifdef CONFIG_PPC32
+	pmac_pic_probe_oldstyle();
+#endif
 }
 
 #if defined(CONFIG_PM) && defined(CONFIG_PPC32)
diff --git a/arch/powerpc/platforms/powermac/pmac.h b/arch/powerpc/platforms/powermac/pmac.h
index 2ad25e13423e..21c7b0f8f329 100644
--- a/arch/powerpc/platforms/powermac/pmac.h
+++ b/arch/powerpc/platforms/powermac/pmac.h
@@ -42,10 +42,6 @@ extern void pmac_ide_init_hwif_ports(hw_regs_t *hw,
 	unsigned long data_port, unsigned long ctrl_port, int *irq);
 
 extern int pmac_nvram_init(void);
-
-extern struct hw_interrupt_type pmac_pic;
-
-void pmac_pic_init(void);
-int pmac_get_irq(struct pt_regs *regs);
+extern void pmac_pic_init(void);
 
 #endif /* __PMAC_H__ */
diff --git a/arch/powerpc/platforms/powermac/setup.c b/arch/powerpc/platforms/powermac/setup.c
index 7acb0546671f..3b1a9d4fcbc6 100644
--- a/arch/powerpc/platforms/powermac/setup.c
+++ b/arch/powerpc/platforms/powermac/setup.c
@@ -60,6 +60,7 @@
 #include <asm/system.h>
 #include <asm/pgtable.h>
 #include <asm/io.h>
+#include <asm/kexec.h>
 #include <asm/pci-bridge.h>
 #include <asm/ohare.h>
 #include <asm/mediabay.h>
@@ -74,8 +75,8 @@
 #include <asm/iommu.h>
 #include <asm/smu.h>
 #include <asm/pmc.h>
-#include <asm/mpic.h>
 #include <asm/lmb.h>
+#include <asm/udbg.h>
 
 #include "pmac.h"
 
@@ -321,16 +322,6 @@ void __init pmac_setup_arch(void)
 	l2cr_init();
 #endif /* CONFIG_PPC32 */
 
-#ifdef CONFIG_PPC64
-	/* Probe motherboard chipset */
-	/* this is done earlier in setup_arch for 32-bit */
-	pmac_feature_init();
-
-	/* We can NAP */
-	powersave_nap = 1;
-	printk(KERN_INFO "Using native/NAP idle loop\n");
-#endif
-
 #ifdef CONFIG_KGDB
 	zs_kgdb_hook(0);
 #endif
@@ -354,7 +345,7 @@ void __init pmac_setup_arch(void)
 
 #ifdef CONFIG_SMP
 	/* Check for Core99 */
-	if (find_devices("uni-n") || find_devices("u3"))
+	if (find_devices("uni-n") || find_devices("u3") || find_devices("u4"))
 		smp_ops = &core99_smp_ops;
 #ifdef CONFIG_PPC32
 	else
@@ -621,35 +612,31 @@ static void __init pmac_init_early(void)
 	 * and call ioremap
 	 */
 	hpte_init_native();
+#endif
 
-	/* Init SCC */
-	if (strstr(cmd_line, "sccdbg")) {
-		sccdbg = 1;
-		udbg_init_scc(NULL);
+	/* Enable early btext debug if requested */
+	if (strstr(cmd_line, "btextdbg")) {
+		udbg_adb_init_early();
+		register_early_udbg_console();
 	}
 
-	/* Setup interrupt mapping options */
-	ppc64_interrupt_controller = IC_OPEN_PIC;
+	/* Probe motherboard chipset */
+	pmac_feature_init();
 
-	iommu_init_early_u3();
-#endif
-}
+	/* We can NAP */
+	powersave_nap = 1;
+	printk(KERN_INFO "Using native/NAP idle loop\n");
+
+	/* Initialize debug stuff */
+	udbg_scc_init(!!strstr(cmd_line, "sccdbg"));
+	udbg_adb_init(!!strstr(cmd_line, "btextdbg"));
 
-static void __init pmac_progress(char *s, unsigned short hex)
-{
 #ifdef CONFIG_PPC64
-	if (sccdbg) {
-		udbg_puts(s);
-		udbg_puts("\n");
-		return;
-	}
+	/* Setup interrupt mapping options */
+	ppc64_interrupt_controller = IC_OPEN_PIC;
+
+	iommu_init_early_dart();
 #endif
-#ifdef CONFIG_BOOTX_TEXT
-	if (boot_text_mapped) {
-		btext_drawstring(s);
-		btext_drawchar('\n');
-	}
-#endif /* CONFIG_BOOTX_TEXT */
 }
 
 /*
@@ -663,35 +650,14 @@ static int pmac_check_legacy_ioport(unsigned int baseport)
 
 static int __init pmac_declare_of_platform_devices(void)
 {
-	struct device_node *np, *npp;
-
-	np = find_devices("uni-n");
-	if (np) {
-		for (np = np->child; np != NULL; np = np->sibling)
-			if (strncmp(np->name, "i2c", 3) == 0) {
-				of_platform_device_create(np, "uni-n-i2c",
-							  NULL);
-				break;
-			}
-	}
-	np = find_devices("valkyrie");
+	struct device_node *np;
+
+	np = of_find_node_by_name(NULL, "valkyrie");
 	if (np)
 		of_platform_device_create(np, "valkyrie", NULL);
-	np = find_devices("platinum");
+	np = of_find_node_by_name(NULL, "platinum");
 	if (np)
 		of_platform_device_create(np, "platinum", NULL);
-
-	npp = of_find_node_by_name(NULL, "u3");
-	if (npp) {
-		for (np = NULL; (np = of_get_next_child(npp, np)) != NULL;) {
-			if (strncmp(np->name, "i2c", 3) == 0) {
-				of_platform_device_create(np, "u3-i2c", NULL);
-				of_node_put(np);
-				break;
-			}
-		}
-		of_node_put(npp);
-	}
         np = of_find_node_by_type(NULL, "smu");
         if (np) {
 		of_platform_device_create(np, "smu", NULL);
@@ -718,7 +684,7 @@ static int __init pmac_probe(int platform)
 	 * occupies having to be broken up so the DART itself is not
 	 * part of the cacheable linar mapping
 	 */
-	alloc_u3_dart_table();
+	alloc_dart_table();
 #endif
 
 #ifdef CONFIG_PMAC_SMU
@@ -734,15 +700,17 @@ static int __init pmac_probe(int platform)
 }
 
 #ifdef CONFIG_PPC64
-static int pmac_probe_mode(struct pci_bus *bus)
+/* Move that to pci.c */
+static int pmac_pci_probe_mode(struct pci_bus *bus)
 {
 	struct device_node *node = bus->sysdata;
 
 	/* We need to use normal PCI probing for the AGP bus,
-	   since the device for the AGP bridge isn't in the tree. */
-	if (bus->self == NULL && device_is_compatible(node, "u3-agp"))
+	 * since the device for the AGP bridge isn't in the tree.
+	 */
+	if (bus->self == NULL && (device_is_compatible(node, "u3-agp") ||
+				  device_is_compatible(node, "u4-pcie")))
 		return PCI_PROBE_NORMAL;
-
 	return PCI_PROBE_DEVTREE;
 }
 #endif
@@ -756,7 +724,7 @@ struct machdep_calls __initdata pmac_md = {
 	.init_early		= pmac_init_early,
 	.show_cpuinfo		= pmac_show_cpuinfo,
 	.init_IRQ		= pmac_pic_init,
-	.get_irq		= mpic_get_irq,	/* changed later */
+	.get_irq		= NULL,	/* changed later */
 	.pcibios_fixup		= pmac_pcibios_fixup,
 	.restart		= pmac_restart,
 	.power_off		= pmac_power_off,
@@ -768,12 +736,17 @@ struct machdep_calls __initdata pmac_md = {
 	.calibrate_decr		= pmac_calibrate_decr,
 	.feature_call		= pmac_do_feature_call,
 	.check_legacy_ioport	= pmac_check_legacy_ioport,
-	.progress		= pmac_progress,
+	.progress		= udbg_progress,
 #ifdef CONFIG_PPC64
-	.pci_probe_mode		= pmac_probe_mode,
+	.pci_probe_mode		= pmac_pci_probe_mode,
 	.idle_loop		= native_idle,
 	.enable_pmcs		= power4_enable_pmcs,
+#ifdef CONFIG_KEXEC
+	.machine_kexec		= default_machine_kexec,
+	.machine_kexec_prepare	= default_machine_kexec_prepare,
+	.machine_crash_shutdown	= default_machine_crash_shutdown,
 #endif
+#endif /* CONFIG_PPC64 */
 #ifdef CONFIG_PPC32
 	.pcibios_enable_device_hook = pmac_pci_enable_device_hook,
 	.pcibios_after_init	= pmac_pcibios_after_init,
diff --git a/arch/powerpc/platforms/powermac/smp.c b/arch/powerpc/platforms/powermac/smp.c
index fb2a7c798e82..0df2cdcd805c 100644
--- a/arch/powerpc/platforms/powermac/smp.c
+++ b/arch/powerpc/platforms/powermac/smp.c
@@ -52,8 +52,9 @@
 #include <asm/cacheflush.h>
 #include <asm/keylargo.h>
 #include <asm/pmac_low_i2c.h>
+#include <asm/pmac_pfunc.h>
 
-#undef DEBUG
+#define DEBUG
 
 #ifdef DEBUG
 #define DBG(fmt...) udbg_printf(fmt)
@@ -62,6 +63,7 @@
 #endif
 
 extern void __secondary_start_pmac_0(void);
+extern int pmac_pfunc_base_install(void);
 
 #ifdef CONFIG_PPC32
 
@@ -361,7 +363,6 @@ static void __init psurge_dual_sync_tb(int cpu_nr)
 	set_dec(tb_ticks_per_jiffy);
 	/* XXX fixme */
 	set_tb(0, 0);
-	last_jiffy_stamp(cpu_nr) = 0;
 
 	if (cpu_nr > 0) {
 		mb();
@@ -429,15 +430,62 @@ struct smp_ops_t psurge_smp_ops = {
 };
 #endif /* CONFIG_PPC32 - actually powersurge support */
 
+/*
+ * Core 99 and later support
+ */
+
+static void (*pmac_tb_freeze)(int freeze);
+static unsigned long timebase;
+static int tb_req;
+
+static void smp_core99_give_timebase(void)
+{
+	unsigned long flags;
+
+	local_irq_save(flags);
+
+	while(!tb_req)
+		barrier();
+	tb_req = 0;
+	(*pmac_tb_freeze)(1);
+	mb();
+	timebase = get_tb();
+	mb();
+	while (timebase)
+		barrier();
+	mb();
+	(*pmac_tb_freeze)(0);
+	mb();
+
+	local_irq_restore(flags);
+}
+
+
+static void __devinit smp_core99_take_timebase(void)
+{
+	unsigned long flags;
+
+	local_irq_save(flags);
+
+	tb_req = 1;
+	mb();
+	while (!timebase)
+		barrier();
+	mb();
+	set_tb(timebase >> 32, timebase & 0xffffffff);
+	timebase = 0;
+	mb();
+	set_dec(tb_ticks_per_jiffy/2);
+
+	local_irq_restore(flags);
+}
+
 #ifdef CONFIG_PPC64
 /*
  * G5s enable/disable the timebase via an i2c-connected clock chip.
  */
-static struct device_node *pmac_tb_clock_chip_host;
+static struct pmac_i2c_bus *pmac_tb_clock_chip_host;
 static u8 pmac_tb_pulsar_addr;
-static void (*pmac_tb_freeze)(int freeze);
-static DEFINE_SPINLOCK(timebase_lock);
-static unsigned long timebase;
 
 static void smp_core99_cypress_tb_freeze(int freeze)
 {
@@ -447,19 +495,20 @@ static void smp_core99_cypress_tb_freeze(int freeze)
 	/* Strangely, the device-tree says address is 0xd2, but darwin
 	 * accesses 0xd0 ...
 	 */
-	pmac_low_i2c_setmode(pmac_tb_clock_chip_host, pmac_low_i2c_mode_combined);
-	rc = pmac_low_i2c_xfer(pmac_tb_clock_chip_host,
-			       0xd0 | pmac_low_i2c_read,
-			       0x81, &data, 1);
+	pmac_i2c_setmode(pmac_tb_clock_chip_host,
+			 pmac_i2c_mode_combined);
+	rc = pmac_i2c_xfer(pmac_tb_clock_chip_host,
+			   0xd0 | pmac_i2c_read,
+			   1, 0x81, &data, 1);
 	if (rc != 0)
 		goto bail;
 
 	data = (data & 0xf3) | (freeze ? 0x00 : 0x0c);
 
-       	pmac_low_i2c_setmode(pmac_tb_clock_chip_host, pmac_low_i2c_mode_stdsub);
-	rc = pmac_low_i2c_xfer(pmac_tb_clock_chip_host,
-			       0xd0 | pmac_low_i2c_write,
-			       0x81, &data, 1);
+       	pmac_i2c_setmode(pmac_tb_clock_chip_host, pmac_i2c_mode_stdsub);
+	rc = pmac_i2c_xfer(pmac_tb_clock_chip_host,
+			   0xd0 | pmac_i2c_write,
+			   1, 0x81, &data, 1);
 
  bail:
 	if (rc != 0) {
@@ -475,19 +524,20 @@ static void smp_core99_pulsar_tb_freeze(int freeze)
 	u8 data;
 	int rc;
 
-	pmac_low_i2c_setmode(pmac_tb_clock_chip_host, pmac_low_i2c_mode_combined);
-	rc = pmac_low_i2c_xfer(pmac_tb_clock_chip_host,
-			       pmac_tb_pulsar_addr | pmac_low_i2c_read,
-			       0x2e, &data, 1);
+	pmac_i2c_setmode(pmac_tb_clock_chip_host,
+			 pmac_i2c_mode_combined);
+	rc = pmac_i2c_xfer(pmac_tb_clock_chip_host,
+			   pmac_tb_pulsar_addr | pmac_i2c_read,
+			   1, 0x2e, &data, 1);
 	if (rc != 0)
 		goto bail;
 
 	data = (data & 0x88) | (freeze ? 0x11 : 0x22);
 
-	pmac_low_i2c_setmode(pmac_tb_clock_chip_host, pmac_low_i2c_mode_stdsub);
-	rc = pmac_low_i2c_xfer(pmac_tb_clock_chip_host,
-			       pmac_tb_pulsar_addr | pmac_low_i2c_write,
-			       0x2e, &data, 1);
+	pmac_i2c_setmode(pmac_tb_clock_chip_host, pmac_i2c_mode_stdsub);
+	rc = pmac_i2c_xfer(pmac_tb_clock_chip_host,
+			   pmac_tb_pulsar_addr | pmac_i2c_write,
+			   1, 0x2e, &data, 1);
  bail:
 	if (rc != 0) {
 		printk(KERN_ERR "Pulsar Timebase %s rc: %d\n",
@@ -496,54 +546,14 @@ static void smp_core99_pulsar_tb_freeze(int freeze)
 	}
 }
 
-
-static void smp_core99_give_timebase(void)
-{
-	/* Open i2c bus for synchronous access */
-	if (pmac_low_i2c_open(pmac_tb_clock_chip_host, 0))
-		panic("Can't open i2c for TB sync !\n");
-
-	spin_lock(&timebase_lock);
-	(*pmac_tb_freeze)(1);
-	mb();
-	timebase = get_tb();
-	spin_unlock(&timebase_lock);
-
-	while (timebase)
-		barrier();
-
-	spin_lock(&timebase_lock);
-	(*pmac_tb_freeze)(0);
-	spin_unlock(&timebase_lock);
-
-	/* Close i2c bus */
-	pmac_low_i2c_close(pmac_tb_clock_chip_host);
-}
-
-
-static void __devinit smp_core99_take_timebase(void)
-{
-	while (!timebase)
-		barrier();
-	spin_lock(&timebase_lock);
-	set_tb(timebase >> 32, timebase & 0xffffffff);
-	timebase = 0;
-	spin_unlock(&timebase_lock);
-}
-
-static void __init smp_core99_setup(int ncpus)
+static void __init smp_core99_setup_i2c_hwsync(int ncpus)
 {
 	struct device_node *cc = NULL;	
 	struct device_node *p;
+	const char *name = NULL;
 	u32 *reg;
 	int ok;
 
-	/* HW sync only on these platforms */
-	if (!machine_is_compatible("PowerMac7,2") &&
-	    !machine_is_compatible("PowerMac7,3") &&
-	    !machine_is_compatible("RackMac3,1"))
-		return;
-
 	/* Look for the clock chip */
 	while ((cc = of_find_node_by_name(cc, "i2c-hwclock")) != NULL) {
 		p = of_get_parent(cc);
@@ -552,124 +562,86 @@ static void __init smp_core99_setup(int ncpus)
 		if (!ok)
 			continue;
 
+		pmac_tb_clock_chip_host = pmac_i2c_find_bus(cc);
+		if (pmac_tb_clock_chip_host == NULL)
+			continue;
 		reg = (u32 *)get_property(cc, "reg", NULL);
 		if (reg == NULL)
 			continue;
-
 		switch (*reg) {
 		case 0xd2:
-			if (device_is_compatible(cc, "pulsar-legacy-slewing")) {
+			if (device_is_compatible(cc,"pulsar-legacy-slewing")) {
 				pmac_tb_freeze = smp_core99_pulsar_tb_freeze;
 				pmac_tb_pulsar_addr = 0xd2;
-				printk(KERN_INFO "Timebase clock is Pulsar chip\n");
+				name = "Pulsar";
 			} else if (device_is_compatible(cc, "cy28508")) {
 				pmac_tb_freeze = smp_core99_cypress_tb_freeze;
-				printk(KERN_INFO "Timebase clock is Cypress chip\n");
+				name = "Cypress";
 			}
 			break;
 		case 0xd4:
 			pmac_tb_freeze = smp_core99_pulsar_tb_freeze;
 			pmac_tb_pulsar_addr = 0xd4;
-			printk(KERN_INFO "Timebase clock is Pulsar chip\n");
+			name = "Pulsar";
 			break;
 		}
-		if (pmac_tb_freeze != NULL) {
-			pmac_tb_clock_chip_host = of_get_parent(cc);
-			of_node_put(cc);
+		if (pmac_tb_freeze != NULL)
 			break;
-		}
 	}
-	if (pmac_tb_freeze == NULL) {
-		smp_ops->give_timebase = smp_generic_give_timebase;
-		smp_ops->take_timebase = smp_generic_take_timebase;
+	if (pmac_tb_freeze != NULL) {
+		/* Open i2c bus for synchronous access */
+		if (pmac_i2c_open(pmac_tb_clock_chip_host, 1)) {
+			printk(KERN_ERR "Failed top open i2c bus for clock"
+			       " sync, fallback to software sync !\n");
+			goto no_i2c_sync;
+		}
+		printk(KERN_INFO "Processor timebase sync using %s i2c clock\n",
+		       name);
+		return;
 	}
+ no_i2c_sync:
+	pmac_tb_freeze = NULL;
+	pmac_tb_clock_chip_host = NULL;
 }
 
-/* nothing to do here, caches are already set up by service processor */
-static inline void __devinit core99_init_caches(int cpu)
+
+
+/*
+ * Newer G5s uses a platform function
+ */
+
+static void smp_core99_pfunc_tb_freeze(int freeze)
 {
+	struct device_node *cpus;
+	struct pmf_args args;
+
+	cpus = of_find_node_by_path("/cpus");
+	BUG_ON(cpus == NULL);
+	args.count = 1;
+	args.u[0].v = !freeze;
+	pmf_call_function(cpus, "cpu-timebase", &args);
+	of_node_put(cpus);
 }
 
 #else /* CONFIG_PPC64 */
 
 /*
- * SMP G4 powermacs use a GPIO to enable/disable the timebase.
+ * SMP G4 use a GPIO to enable/disable the timebase.
  */
 
 static unsigned int core99_tb_gpio;	/* Timebase freeze GPIO */
 
-static unsigned int pri_tb_hi, pri_tb_lo;
-static unsigned int pri_tb_stamp;
-
-/* not __init, called in sleep/wakeup code */
-void smp_core99_give_timebase(void)
+static void smp_core99_gpio_tb_freeze(int freeze)
 {
-	unsigned long flags;
-	unsigned int t;
-
-	/* wait for the secondary to be in take_timebase */
-	for (t = 100000; t > 0 && !sec_tb_reset; --t)
-		udelay(10);
-	if (!sec_tb_reset) {
-		printk(KERN_WARNING "Timeout waiting sync on second CPU\n");
-		return;
-	}
-
-	/* freeze the timebase and read it */
-	/* disable interrupts so the timebase is disabled for the
-	   shortest possible time */
-	local_irq_save(flags);
-	pmac_call_feature(PMAC_FTR_WRITE_GPIO, NULL, core99_tb_gpio, 4);
+	if (freeze)
+		pmac_call_feature(PMAC_FTR_WRITE_GPIO, NULL, core99_tb_gpio, 4);
+	else
+		pmac_call_feature(PMAC_FTR_WRITE_GPIO, NULL, core99_tb_gpio, 0);
 	pmac_call_feature(PMAC_FTR_READ_GPIO, NULL, core99_tb_gpio, 0);
-	mb();
-	pri_tb_hi = get_tbu();
-	pri_tb_lo = get_tbl();
-	pri_tb_stamp = last_jiffy_stamp(smp_processor_id());
-	mb();
-
-	/* tell the secondary we're ready */
-	sec_tb_reset = 2;
-	mb();
-
-	/* wait for the secondary to have taken it */
-	/* note: can't use udelay here, since it needs the timebase running */
-	for (t = 10000000; t > 0 && sec_tb_reset; --t)
-		barrier();
-	if (sec_tb_reset)
-		/* XXX BUG_ON here? */
-		printk(KERN_WARNING "Timeout waiting sync(2) on second CPU\n");
-
-	/* Now, restart the timebase by leaving the GPIO to an open collector */
-       	pmac_call_feature(PMAC_FTR_WRITE_GPIO, NULL, core99_tb_gpio, 0);
-        pmac_call_feature(PMAC_FTR_READ_GPIO, NULL, core99_tb_gpio, 0);
-	local_irq_restore(flags);
 }
 
-/* not __init, called in sleep/wakeup code */
-void smp_core99_take_timebase(void)
-{
-	unsigned long flags;
-
-	/* tell the primary we're here */
-	sec_tb_reset = 1;
-	mb();
-
-	/* wait for the primary to set pri_tb_hi/lo */
-	while (sec_tb_reset < 2)
-		mb();
-
-	/* set our stuff the same as the primary */
-	local_irq_save(flags);
-	set_dec(1);
-	set_tb(pri_tb_hi, pri_tb_lo);
-	last_jiffy_stamp(smp_processor_id()) = pri_tb_stamp;
-	mb();
 
-	/* tell the primary we're done */
-       	sec_tb_reset = 0;
-	mb();
-	local_irq_restore(flags);
-}
+#endif /* !CONFIG_PPC64 */
 
 /* L2 and L3 cache settings to pass from CPU0 to CPU1 on G4 cpus */
 volatile static long int core99_l2_cache;
@@ -677,6 +649,7 @@ volatile static long int core99_l3_cache;
 
 static void __devinit core99_init_caches(int cpu)
 {
+#ifndef CONFIG_PPC64
 	if (!cpu_has_feature(CPU_FTR_L2CR))
 		return;
 
@@ -702,30 +675,76 @@ static void __devinit core99_init_caches(int cpu)
 		_set_L3CR(core99_l3_cache);
 		printk("CPU%d: L3CR set to %lx\n", cpu, core99_l3_cache);
 	}
+#endif /* !CONFIG_PPC64 */
 }
 
 static void __init smp_core99_setup(int ncpus)
 {
-	struct device_node *cpu;
-	u32 *tbprop = NULL;
-	int i;
+#ifdef CONFIG_PPC64
+
+	/* i2c based HW sync on some G5s */
+	if (machine_is_compatible("PowerMac7,2") ||
+	    machine_is_compatible("PowerMac7,3") ||
+	    machine_is_compatible("RackMac3,1"))
+		smp_core99_setup_i2c_hwsync(ncpus);
 
-	core99_tb_gpio = KL_GPIO_TB_ENABLE;	/* default value */
-	cpu = of_find_node_by_type(NULL, "cpu");
-	if (cpu != NULL) {
-		tbprop = (u32 *)get_property(cpu, "timebase-enable", NULL);
-		if (tbprop)
-			core99_tb_gpio = *tbprop;
-		of_node_put(cpu);
+	/* pfunc based HW sync on recent G5s */
+	if (pmac_tb_freeze == NULL) {
+		struct device_node *cpus =
+			of_find_node_by_path("/cpus");
+		if (cpus &&
+		    get_property(cpus, "platform-cpu-timebase", NULL)) {
+			pmac_tb_freeze = smp_core99_pfunc_tb_freeze;
+			printk(KERN_INFO "Processor timebase sync using"
+			       " platform function\n");
+		}
 	}
 
-	/* XXX should get this from reg properties */
-	for (i = 1; i < ncpus; ++i)
-		smp_hw_index[i] = i;
-	powersave_nap = 0;
-}
+#else /* CONFIG_PPC64 */
+
+	/* GPIO based HW sync on ppc32 Core99 */
+	if (pmac_tb_freeze == NULL && !machine_is_compatible("MacRISC4")) {
+		struct device_node *cpu;
+		u32 *tbprop = NULL;
+
+		core99_tb_gpio = KL_GPIO_TB_ENABLE;	/* default value */
+		cpu = of_find_node_by_type(NULL, "cpu");
+		if (cpu != NULL) {
+			tbprop = (u32 *)get_property(cpu, "timebase-enable",
+						     NULL);
+			if (tbprop)
+				core99_tb_gpio = *tbprop;
+			of_node_put(cpu);
+		}
+		pmac_tb_freeze = smp_core99_gpio_tb_freeze;
+		printk(KERN_INFO "Processor timebase sync using"
+		       " GPIO 0x%02x\n", core99_tb_gpio);
+	}
+
+#endif /* CONFIG_PPC64 */
+
+	/* No timebase sync, fallback to software */
+	if (pmac_tb_freeze == NULL) {
+		smp_ops->give_timebase = smp_generic_give_timebase;
+		smp_ops->take_timebase = smp_generic_take_timebase;
+		printk(KERN_INFO "Processor timebase sync using software\n");
+	}
+
+#ifndef CONFIG_PPC64
+	{
+		int i;
+
+		/* XXX should get this from reg properties */
+		for (i = 1; i < ncpus; ++i)
+			smp_hw_index[i] = i;
+	}
 #endif
 
+	/* 32 bits SMP can't NAP */
+	if (!machine_is_compatible("MacRISC4"))
+		powersave_nap = 0;
+}
+
 static int __init smp_core99_probe(void)
 {
 	struct device_node *cpus;
@@ -743,8 +762,19 @@ static int __init smp_core99_probe(void)
 	if (ncpus <= 1)
 		return 1;
 
+	/* We need to perform some early initialisations before we can start
+	 * setting up SMP as we are running before initcalls
+	 */
+	pmac_pfunc_base_install();
+	pmac_i2c_init();
+
+	/* Setup various bits like timebase sync method, ability to nap, ... */
 	smp_core99_setup(ncpus);
+
+	/* Install IPIs */
 	mpic_request_ipis();
+
+	/* Collect l2cr and l3cr values from CPU 0 */
 	core99_init_caches(0);
 
 	return ncpus;
@@ -753,14 +783,15 @@ static int __init smp_core99_probe(void)
 static void __devinit smp_core99_kick_cpu(int nr)
 {
 	unsigned int save_vector;
-	unsigned long new_vector;
-	unsigned long flags;
+	unsigned long target, flags;
 	volatile unsigned int *vector
 		 = ((volatile unsigned int *)(KERNELBASE+0x100));
 
 	if (nr < 0 || nr > 3)
 		return;
-	if (ppc_md.progress) ppc_md.progress("smp_core99_kick_cpu", 0x346);
+
+	if (ppc_md.progress)
+		ppc_md.progress("smp_core99_kick_cpu", 0x346);
 
 	local_irq_save(flags);
 	local_irq_disable();
@@ -768,14 +799,11 @@ static void __devinit smp_core99_kick_cpu(int nr)
 	/* Save reset vector */
 	save_vector = *vector;
 
-	/* Setup fake reset vector that does	
+	/* Setup fake reset vector that does
 	 *   b __secondary_start_pmac_0 + nr*8 - KERNELBASE
 	 */
-	new_vector = (unsigned long) __secondary_start_pmac_0 + nr * 8;
-	*vector = 0x48000002 + new_vector - KERNELBASE;
-
-	/* flush data cache and inval instruction cache */
-	flush_icache_range((unsigned long) vector, (unsigned long) vector + 4);
+	target = (unsigned long) __secondary_start_pmac_0 + nr * 8;
+	create_branch((unsigned long)vector, target, BRANCH_SET_LINK);
 
 	/* Put some life in our friend */
 	pmac_call_feature(PMAC_FTR_RESET_CPU, NULL, nr, 0);
@@ -805,17 +833,25 @@ static void __devinit smp_core99_setup_cpu(int cpu_nr)
 	mpic_setup_this_cpu();
 
 	if (cpu_nr == 0) {
-#ifdef CONFIG_POWER4
+#ifdef CONFIG_PPC64
 		extern void g5_phy_disable_cpu1(void);
 
+		/* Close i2c bus if it was used for tb sync */
+		if (pmac_tb_clock_chip_host) {
+			pmac_i2c_close(pmac_tb_clock_chip_host);
+			pmac_tb_clock_chip_host	= NULL;
+		}
+
 		/* If we didn't start the second CPU, we must take
 		 * it off the bus
 		 */
 		if (machine_is_compatible("MacRISC4") &&
 		    num_online_cpus() < 2)		
 			g5_phy_disable_cpu1();
-#endif /* CONFIG_POWER4 */
-		if (ppc_md.progress) ppc_md.progress("core99_setup_cpu 0 done", 0x349);
+#endif /* CONFIG_PPC64 */
+
+		if (ppc_md.progress)
+			ppc_md.progress("core99_setup_cpu 0 done", 0x349);
 	}
 }
 
diff --git a/arch/powerpc/platforms/powermac/time.c b/arch/powerpc/platforms/powermac/time.c
index feb0a94e7819..5d9afa1fa02d 100644
--- a/arch/powerpc/platforms/powermac/time.c
+++ b/arch/powerpc/platforms/powermac/time.c
@@ -258,15 +258,20 @@ int __init via_calibrate_decr(void)
 	volatile unsigned char __iomem *via;
 	int count = VIA_TIMER_FREQ_6 / 100;
 	unsigned int dstart, dend;
+	struct resource rsrc;
 
-	vias = find_devices("via-cuda");
+	vias = of_find_node_by_name(NULL, "via-cuda");
 	if (vias == 0)
-		vias = find_devices("via-pmu");
+		vias = of_find_node_by_name(NULL, "via-pmu");
 	if (vias == 0)
-		vias = find_devices("via");
-	if (vias == 0 || vias->n_addrs == 0)
+		vias = of_find_node_by_name(NULL, "via");
+	if (vias == 0 || of_address_to_resource(vias, 0, &rsrc))
 		return 0;
-	via = ioremap(vias->addrs[0].address, vias->addrs[0].size);
+	via = ioremap(rsrc.start, rsrc.end - rsrc.start + 1);
+	if (via == NULL) {
+		printk(KERN_ERR "Failed to map VIA for timer calibration !\n");
+		return 0;
+	}
 
 	/* set timer 1 for continuous interrupts */
 	out_8(&via[ACR], (via[ACR] & ~T1MODE) | T1MODE_CONT);
diff --git a/arch/powerpc/platforms/powermac/udbg_adb.c b/arch/powerpc/platforms/powermac/udbg_adb.c
new file mode 100644
index 000000000000..06c8265c2baf
--- /dev/null
+++ b/arch/powerpc/platforms/powermac/udbg_adb.c
@@ -0,0 +1,221 @@
+#include <linux/config.h>
+#include <linux/string.h>
+#include <linux/kernel.h>
+#include <linux/errno.h>
+#include <linux/bitops.h>
+#include <linux/ptrace.h>
+#include <linux/adb.h>
+#include <linux/pmu.h>
+#include <linux/cuda.h>
+#include <asm/machdep.h>
+#include <asm/io.h>
+#include <asm/page.h>
+#include <asm/xmon.h>
+#include <asm/prom.h>
+#include <asm/bootx.h>
+#include <asm/machdep.h>
+#include <asm/errno.h>
+#include <asm/pmac_feature.h>
+#include <asm/processor.h>
+#include <asm/delay.h>
+#include <asm/btext.h>
+#include <asm/time.h>
+#include <asm/udbg.h>
+
+/*
+ * This implementation is "special", it can "patch" the current
+ * udbg implementation and work on top of it. It must thus be
+ * initialized last
+ */
+
+static void (*udbg_adb_old_putc)(char c);
+static int (*udbg_adb_old_getc)(void);
+static int (*udbg_adb_old_getc_poll)(void);
+
+static enum {
+	input_adb_none,
+	input_adb_pmu,
+	input_adb_cuda,
+} input_type = input_adb_none;
+
+int xmon_wants_key, xmon_adb_keycode;
+
+static inline void udbg_adb_poll(void)
+{
+#ifdef CONFIG_ADB_PMU
+	if (input_type == input_adb_pmu)
+		pmu_poll_adb();
+#endif /* CONFIG_ADB_PMU */
+#ifdef CONFIG_ADB_CUDA
+	if (input_type == input_adb_cuda)
+		cuda_poll();
+#endif /* CONFIG_ADB_CUDA */
+}
+
+#ifdef CONFIG_BOOTX_TEXT
+
+static int udbg_adb_use_btext;
+static int xmon_adb_shiftstate;
+
+static unsigned char xmon_keytab[128] =
+	"asdfhgzxcv\000bqwer"				/* 0x00 - 0x0f */
+	"yt123465=97-80]o"				/* 0x10 - 0x1f */
+	"u[ip\rlj'k;\\,/nm."				/* 0x20 - 0x2f */
+	"\t `\177\0\033\0\0\0\0\0\0\0\0\0\0"		/* 0x30 - 0x3f */
+	"\0.\0*\0+\0\0\0\0\0/\r\0-\0"			/* 0x40 - 0x4f */
+	"\0\0000123456789\0\0\0";			/* 0x50 - 0x5f */
+
+static unsigned char xmon_shift_keytab[128] =
+	"ASDFHGZXCV\000BQWER"				/* 0x00 - 0x0f */
+	"YT!@#$^%+(&_*)}O"				/* 0x10 - 0x1f */
+	"U{IP\rLJ\"K:|<?NM>"				/* 0x20 - 0x2f */
+	"\t ~\177\0\033\0\0\0\0\0\0\0\0\0\0"		/* 0x30 - 0x3f */
+	"\0.\0*\0+\0\0\0\0\0/\r\0-\0"			/* 0x40 - 0x4f */
+	"\0\0000123456789\0\0\0";			/* 0x50 - 0x5f */
+
+static int udbg_adb_local_getc(void)
+{
+	int k, t, on;
+
+	xmon_wants_key = 1;
+	for (;;) {
+		xmon_adb_keycode = -1;
+		t = 0;
+		on = 0;
+		k = -1;
+		do {
+			if (--t < 0) {
+				on = 1 - on;
+				btext_drawchar(on? 0xdb: 0x20);
+				btext_drawchar('\b');
+				t = 200000;
+			}
+			udbg_adb_poll();
+			if (udbg_adb_old_getc_poll)
+				k = udbg_adb_old_getc_poll();
+		} while (k == -1 && xmon_adb_keycode == -1);
+		if (on)
+			btext_drawstring(" \b");
+		if (k != -1)
+			return k;
+		k = xmon_adb_keycode;
+
+		/* test for shift keys */
+		if ((k & 0x7f) == 0x38 || (k & 0x7f) == 0x7b) {
+			xmon_adb_shiftstate = (k & 0x80) == 0;
+			continue;
+		}
+		if (k >= 0x80)
+			continue;	/* ignore up transitions */
+		k = (xmon_adb_shiftstate? xmon_shift_keytab: xmon_keytab)[k];
+		if (k != 0)
+			break;
+	}
+	xmon_wants_key = 0;
+	return k;
+}
+#endif /* CONFIG_BOOTX_TEXT */
+
+static int udbg_adb_getc(void)
+{
+#ifdef CONFIG_BOOTX_TEXT
+	if (udbg_adb_use_btext && input_type != input_adb_none)
+		return udbg_adb_local_getc();
+#endif
+	if (udbg_adb_old_getc)
+		return udbg_adb_old_getc();
+	return -1;
+}
+
+/* getc_poll() is not really used, unless you have the xmon-over modem
+ * hack that doesn't quite concern us here, thus we just poll the low level
+ * ADB driver to prevent it from timing out and call back the original poll
+ * routine.
+ */
+static int udbg_adb_getc_poll(void)
+{
+	udbg_adb_poll();
+
+	if (udbg_adb_old_getc_poll)
+		return udbg_adb_old_getc_poll();
+	return -1;
+}
+
+static void udbg_adb_putc(char c)
+{
+#ifdef CONFIG_BOOTX_TEXT
+	if (udbg_adb_use_btext)
+		btext_drawchar(c);
+#endif
+	if (udbg_adb_old_putc)
+		return udbg_adb_old_putc(c);
+}
+
+void udbg_adb_init_early(void)
+{
+#ifdef CONFIG_BOOTX_TEXT
+	if (btext_find_display(1) == 0) {
+		udbg_adb_use_btext = 1;
+		udbg_putc = udbg_adb_putc;
+	}
+#endif
+}
+
+int udbg_adb_init(int force_btext)
+{
+	struct device_node *np;
+
+	/* Capture existing callbacks */
+	udbg_adb_old_putc = udbg_putc;
+	udbg_adb_old_getc = udbg_getc;
+	udbg_adb_old_getc_poll = udbg_getc_poll;
+
+	/* Check if our early init was already called */
+	if (udbg_adb_old_putc == udbg_adb_putc)
+		udbg_adb_old_putc = NULL;
+#ifdef CONFIG_BOOTX_TEXT
+	if (udbg_adb_old_putc == btext_drawchar)
+		udbg_adb_old_putc = NULL;
+#endif
+
+	/* Set ours as output */
+	udbg_putc = udbg_adb_putc;
+	udbg_getc = udbg_adb_getc;
+	udbg_getc_poll = udbg_adb_getc_poll;
+
+#ifdef CONFIG_BOOTX_TEXT
+	/* Check if we should use btext output */
+	if (btext_find_display(force_btext) == 0)
+		udbg_adb_use_btext = 1;
+#endif
+
+	/* See if there is a keyboard in the device tree with a parent
+	 * of type "adb". If not, we return a failure, but we keep the
+	 * bext output set for now
+	 */
+	for (np = NULL; (np = of_find_node_by_name(np, "keyboard")) != NULL;) {
+		struct device_node *parent = of_get_parent(np);
+		int found = (parent && strcmp(parent->type, "adb") == 0);
+		of_node_put(parent);
+		if (found)
+			break;
+	}
+	if (np == NULL)
+		return -ENODEV;
+	of_node_put(np);
+
+#ifdef CONFIG_ADB_PMU
+	if (find_via_pmu())
+		input_type = input_adb_pmu;
+#endif
+#ifdef CONFIG_ADB_CUDA
+	if (find_via_cuda())
+		input_type = input_adb_cuda;
+#endif
+
+	/* Same as above: nothing found, keep btext set for output */
+	if (input_type == input_adb_none)
+		return -ENODEV;
+
+	return 0;
+}
diff --git a/arch/powerpc/platforms/powermac/udbg_scc.c b/arch/powerpc/platforms/powermac/udbg_scc.c
new file mode 100644
index 000000000000..e87d53acfb61
--- /dev/null
+++ b/arch/powerpc/platforms/powermac/udbg_scc.c
@@ -0,0 +1,165 @@
+/*
+ * udbg for for zilog scc ports as found on Apple PowerMacs
+ *
+ * Copyright (C) 2001-2005 PPC 64 Team, IBM Corp
+ *
+ *      This program is free software; you can redistribute it and/or
+ *      modify it under the terms of the GNU General Public License
+ *      as published by the Free Software Foundation; either version
+ *      2 of the License, or (at your option) any later version.
+ */
+#include <linux/config.h>
+#include <linux/types.h>
+#include <asm/udbg.h>
+#include <asm/processor.h>
+#include <asm/io.h>
+#include <asm/prom.h>
+#include <asm/pmac_feature.h>
+
+extern u8 real_readb(volatile u8 __iomem  *addr);
+extern void real_writeb(u8 data, volatile u8 __iomem *addr);
+
+#define	SCC_TXRDY	4
+#define SCC_RXRDY	1
+
+static volatile u8 __iomem *sccc;
+static volatile u8 __iomem *sccd;
+
+static void udbg_scc_putc(char c)
+{
+	if (sccc) {
+		while ((in_8(sccc) & SCC_TXRDY) == 0)
+			;
+		out_8(sccd,  c);
+		if (c == '\n')
+			udbg_scc_putc('\r');
+	}
+}
+
+static int udbg_scc_getc_poll(void)
+{
+	if (sccc) {
+		if ((in_8(sccc) & SCC_RXRDY) != 0)
+			return in_8(sccd);
+		else
+			return -1;
+	}
+	return -1;
+}
+
+static int udbg_scc_getc(void)
+{
+	if (sccc) {
+		while ((in_8(sccc) & SCC_RXRDY) == 0)
+			;
+		return in_8(sccd);
+	}
+	return -1;
+}
+
+static unsigned char scc_inittab[] = {
+    13, 0,		/* set baud rate divisor */
+    12, 0,
+    14, 1,		/* baud rate gen enable, src=rtxc */
+    11, 0x50,		/* clocks = br gen */
+    5,  0xea,		/* tx 8 bits, assert DTR & RTS */
+    4,  0x46,		/* x16 clock, 1 stop */
+    3,  0xc1,		/* rx enable, 8 bits */
+};
+
+void udbg_scc_init(int force_scc)
+{
+	u32 *reg;
+	unsigned long addr;
+	struct device_node *stdout = NULL, *escc = NULL, *macio = NULL;
+	struct device_node *ch, *ch_def = NULL, *ch_a = NULL;
+	char *path;
+	int i, x;
+
+	escc = of_find_node_by_name(NULL, "escc");
+	if (escc == NULL)
+		goto bail;
+	macio = of_get_parent(escc);
+	if (macio == NULL)
+		goto bail;
+	path = (char *)get_property(of_chosen, "linux,stdout-path", NULL);
+	if (path != NULL)
+		stdout = of_find_node_by_path(path);
+	for (ch = NULL; (ch = of_get_next_child(escc, ch)) != NULL;) {
+		if (ch == stdout)
+			ch_def = of_node_get(ch);
+		if (strcmp(ch->name, "ch-a") == 0)
+			ch_a = of_node_get(ch);
+	}
+	if (ch_def == NULL && !force_scc)
+		goto bail;
+
+	ch = ch_def ? ch_def : ch_a;
+
+	/* Get address within mac-io ASIC */
+	reg = (u32 *)get_property(escc, "reg", NULL);
+	if (reg == NULL)
+		goto bail;
+	addr = reg[0];
+
+	/* Get address of mac-io PCI itself */
+	reg = (u32 *)get_property(macio, "assigned-addresses", NULL);
+	if (reg == NULL)
+		goto bail;
+	addr += reg[2];
+
+	/* Lock the serial port */
+	pmac_call_feature(PMAC_FTR_SCC_ENABLE, ch,
+			  PMAC_SCC_ASYNC | PMAC_SCC_FLAG_XMON, 1);
+
+
+	/* Setup for 57600 8N1 */
+	if (ch == ch_a)
+		addr += 0x20;
+	sccc = (volatile u8 * __iomem) ioremap(addr & PAGE_MASK, PAGE_SIZE) ;
+	sccc += addr & ~PAGE_MASK;
+	sccd = sccc + 0x10;
+
+	mb();
+
+	for (i = 20000; i != 0; --i)
+		x = in_8(sccc);
+	out_8(sccc, 0x09);		/* reset A or B side */
+	out_8(sccc, 0xc0);
+	for (i = 0; i < sizeof(scc_inittab); ++i)
+		out_8(sccc, scc_inittab[i]);
+
+	udbg_putc = udbg_scc_putc;
+	udbg_getc = udbg_scc_getc;
+	udbg_getc_poll = udbg_scc_getc_poll;
+
+	udbg_puts("Hello World !\n");
+
+ bail:
+	of_node_put(macio);
+	of_node_put(escc);
+	of_node_put(stdout);
+	of_node_put(ch_def);
+	of_node_put(ch_a);
+}
+
+#ifdef CONFIG_PPC64
+static void udbg_real_scc_putc(char c)
+{
+	while ((real_readb(sccc) & SCC_TXRDY) == 0)
+		;
+	real_writeb(c, sccd);
+	if (c == '\n')
+		udbg_real_scc_putc('\r');
+}
+
+void udbg_init_pmac_realmode(void)
+{
+	sccc = (volatile u8 __iomem *)0x80013020ul;
+	sccd = (volatile u8 __iomem *)0x80013030ul;
+
+	udbg_putc = udbg_real_scc_putc;
+	udbg_getc = NULL;
+	udbg_getc_poll = NULL;
+}
+#endif /* CONFIG_PPC64 */
diff --git a/arch/powerpc/platforms/pseries/Makefile b/arch/powerpc/platforms/pseries/Makefile
index 06d5ef501218..6accdd155505 100644
--- a/arch/powerpc/platforms/pseries/Makefile
+++ b/arch/powerpc/platforms/pseries/Makefile
@@ -1,5 +1,5 @@
 obj-y			:= pci.o lpar.o hvCall.o nvram.o reconfig.o \
-			   setup.o iommu.o ras.o rtasd.o
+			   setup.o iommu.o ras.o rtasd.o pci_dlpar.o
 obj-$(CONFIG_SMP)	+= smp.o
 obj-$(CONFIG_IBMVIO)	+= vio.o
 obj-$(CONFIG_XICS)	+= xics.o
diff --git a/arch/powerpc/platforms/pseries/eeh.c b/arch/powerpc/platforms/pseries/eeh.c
index c8d2a40dc5b4..7fbfd16d72b7 100644
--- a/arch/powerpc/platforms/pseries/eeh.c
+++ b/arch/powerpc/platforms/pseries/eeh.c
@@ -1093,6 +1093,15 @@ void eeh_add_device_early(struct device_node *dn)
 }
 EXPORT_SYMBOL_GPL(eeh_add_device_early);
 
+void eeh_add_device_tree_early(struct device_node *dn)
+{
+	struct device_node *sib;
+	for (sib = dn->child; sib; sib = sib->sibling)
+		eeh_add_device_tree_early(sib);
+	eeh_add_device_early(dn);
+}
+EXPORT_SYMBOL_GPL(eeh_add_device_tree_early);
+
 /**
  * eeh_add_device_late - perform EEH initialization for the indicated pci device
  * @dev: pci device for which to set up EEH
@@ -1147,6 +1156,23 @@ void eeh_remove_device(struct pci_dev *dev)
 }
 EXPORT_SYMBOL_GPL(eeh_remove_device);
 
+void eeh_remove_bus_device(struct pci_dev *dev)
+{
+	eeh_remove_device(dev);
+	if (dev->hdr_type == PCI_HEADER_TYPE_BRIDGE) {
+		struct pci_bus *bus = dev->subordinate;
+		struct list_head *ln;
+		if (!bus)
+			return; 
+		for (ln = bus->devices.next; ln != &bus->devices; ln = ln->next) {
+			struct pci_dev *pdev = pci_dev_b(ln);
+			if (pdev)
+				eeh_remove_bus_device(pdev);
+		}
+	}
+}
+EXPORT_SYMBOL_GPL(eeh_remove_bus_device);
+
 static int proc_eeh_show(struct seq_file *m, void *v)
 {
 	unsigned int cpu;
diff --git a/arch/powerpc/platforms/pseries/iommu.c b/arch/powerpc/platforms/pseries/iommu.c
index 2043659ea7b1..169f9148789c 100644
--- a/arch/powerpc/platforms/pseries/iommu.c
+++ b/arch/powerpc/platforms/pseries/iommu.c
@@ -436,7 +436,7 @@ static void iommu_bus_setup_pSeriesLP(struct pci_bus *bus)
 		return;
 	}
 
-	ppci = pdn->data;
+	ppci = PCI_DN(pdn);
 	if (!ppci->iommu_table) {
 		/* Bussubno hasn't been copied yet.
 		 * Do it now because iommu_table_setparms_lpar needs it.
@@ -483,10 +483,10 @@ static void iommu_dev_setup_pSeries(struct pci_dev *dev)
 	 * an already allocated iommu table is found and use that.
 	 */
 
-	while (dn && dn->data && PCI_DN(dn)->iommu_table == NULL)
+	while (dn && PCI_DN(dn) && PCI_DN(dn)->iommu_table == NULL)
 		dn = dn->parent;
 
-	if (dn && dn->data) {
+	if (dn && PCI_DN(dn)) {
 		PCI_DN(mydn)->iommu_table = PCI_DN(dn)->iommu_table;
 	} else {
 		DBG("iommu_dev_setup_pSeries, dev %p (%s) has no iommu table\n", dev, pci_name(dev));
@@ -497,7 +497,7 @@ static int iommu_reconfig_notifier(struct notifier_block *nb, unsigned long acti
 {
 	int err = NOTIFY_OK;
 	struct device_node *np = node;
-	struct pci_dn *pci = np->data;
+	struct pci_dn *pci = PCI_DN(np);
 
 	switch (action) {
 	case PSERIES_RECONFIG_REMOVE:
@@ -533,7 +533,7 @@ static void iommu_dev_setup_pSeriesLP(struct pci_dev *dev)
 	 */
 	dn = pci_device_to_OF_node(dev);
 
-	for (pdn = dn; pdn && pdn->data && !PCI_DN(pdn)->iommu_table;
+	for (pdn = dn; pdn && PCI_DN(pdn) && !PCI_DN(pdn)->iommu_table;
 	     pdn = pdn->parent) {
 		dma_window = (unsigned int *)
 			get_property(pdn, "ibm,dma-window", NULL);
@@ -552,7 +552,7 @@ static void iommu_dev_setup_pSeriesLP(struct pci_dev *dev)
 		DBG("Found DMA window, allocating table\n");
 	}
 
-	pci = pdn->data;
+	pci = PCI_DN(pdn);
 	if (!pci->iommu_table) {
 		/* iommu_table_setparms_lpar needs bussubno. */
 		pci->bussubno = pci->phb->bus->number;
diff --git a/arch/powerpc/platforms/pseries/lpar.c b/arch/powerpc/platforms/pseries/lpar.c
index cf1bc11b3346..1fe445ab78a6 100644
--- a/arch/powerpc/platforms/pseries/lpar.c
+++ b/arch/powerpc/platforms/pseries/lpar.c
@@ -24,6 +24,7 @@
 #include <linux/config.h>
 #include <linux/kernel.h>
 #include <linux/dma-mapping.h>
+#include <linux/console.h>
 #include <asm/processor.h>
 #include <asm/mmu.h>
 #include <asm/page.h>
@@ -60,7 +61,7 @@ extern void pSeries_find_serial_port(void);
 int vtermno;	/* virtual terminal# for udbg  */
 
 #define __ALIGNED__ __attribute__((__aligned__(sizeof(long))))
-static void udbg_hvsi_putc(unsigned char c)
+static void udbg_hvsi_putc(char c)
 {
 	/* packet's seqno isn't used anyways */
 	uint8_t packet[] __ALIGNED__ = { 0xff, 5, 0, 0, c };
@@ -111,7 +112,7 @@ static int udbg_hvsi_getc_poll(void)
 	return ch;
 }
 
-static unsigned char udbg_hvsi_getc(void)
+static int udbg_hvsi_getc(void)
 {
 	int ch;
 	for (;;) {
@@ -127,7 +128,7 @@ static unsigned char udbg_hvsi_getc(void)
 	}
 }
 
-static void udbg_putcLP(unsigned char c)
+static void udbg_putcLP(char c)
 {
 	char buf[16];
 	unsigned long rc;
@@ -172,7 +173,7 @@ static int udbg_getc_pollLP(void)
 	return ch;
 }
 
-static unsigned char udbg_getcLP(void)
+static int udbg_getcLP(void)
 {
 	int ch;
 	for (;;) {
@@ -191,7 +192,7 @@ static unsigned char udbg_getcLP(void)
 /* call this from early_init() for a working debug console on
  * vterm capable LPAR machines
  */
-void udbg_init_debug_lpar(void)
+void __init udbg_init_debug_lpar(void)
 {
 	vtermno = 0;
 	udbg_putc = udbg_putcLP;
@@ -200,63 +201,54 @@ void udbg_init_debug_lpar(void)
 }
 
 /* returns 0 if couldn't find or use /chosen/stdout as console */
-int find_udbg_vterm(void)
+void __init find_udbg_vterm(void)
 {
 	struct device_node *stdout_node;
 	u32 *termno;
 	char *name;
-	int found = 0;
+	int add_console;
 
 	/* find the boot console from /chosen/stdout */
 	if (!of_chosen)
-		return 0;
+		return;
 	name = (char *)get_property(of_chosen, "linux,stdout-path", NULL);
 	if (name == NULL)
-		return 0;
+		return;
 	stdout_node = of_find_node_by_path(name);
 	if (!stdout_node)
-		return 0;
-
-	/* now we have the stdout node; figure out what type of device it is. */
+		return;
 	name = (char *)get_property(stdout_node, "name", NULL);
 	if (!name) {
 		printk(KERN_WARNING "stdout node missing 'name' property!\n");
 		goto out;
 	}
+	/* The user has requested a console so this is already set up. */
+	add_console = !strstr(cmd_line, "console=");
 
-	if (strncmp(name, "vty", 3) == 0) {
-		if (device_is_compatible(stdout_node, "hvterm1")) {
-			termno = (u32 *)get_property(stdout_node, "reg", NULL);
-			if (termno) {
-				vtermno = termno[0];
-				udbg_putc = udbg_putcLP;
-				udbg_getc = udbg_getcLP;
-				udbg_getc_poll = udbg_getc_pollLP;
-				found = 1;
-			}
-		} else if (device_is_compatible(stdout_node, "hvterm-protocol")) {
-			termno = (u32 *)get_property(stdout_node, "reg", NULL);
-			if (termno) {
-				vtermno = termno[0];
-				udbg_putc = udbg_hvsi_putc;
-				udbg_getc = udbg_hvsi_getc;
-				udbg_getc_poll = udbg_hvsi_getc_poll;
-				found = 1;
-			}
-		}
-	} else if (strncmp(name, "serial", 6)) {
-		/* XXX fix ISA serial console */
-		printk(KERN_WARNING "serial stdout on LPAR ('%s')! "
-				"can't print udbg messages\n",
-		       stdout_node->full_name);
-	} else {
-		printk(KERN_WARNING "don't know how to print to stdout '%s'\n",
-		       stdout_node->full_name);
+	/* Check if it's a virtual terminal */
+	if (strncmp(name, "vty", 3) != 0)
+		goto out;
+	termno = (u32 *)get_property(stdout_node, "reg", NULL);
+	if (termno == NULL)
+		goto out;
+	vtermno = termno[0];
+
+	if (device_is_compatible(stdout_node, "hvterm1")) {
+		udbg_putc = udbg_putcLP;
+		udbg_getc = udbg_getcLP;
+		udbg_getc_poll = udbg_getc_pollLP;
+		if (add_console)
+			add_preferred_console("hvc", termno[0] & 0xff, NULL);
+	} else if (device_is_compatible(stdout_node, "hvterm-protocol")) {
+		vtermno = termno[0];
+		udbg_putc = udbg_hvsi_putc;
+		udbg_getc = udbg_hvsi_getc;
+		udbg_getc_poll = udbg_hvsi_getc_poll;
+		if (add_console)
+			add_preferred_console("hvsi", termno[0] & 0xff, NULL);
 	}
-
 out:
 	of_node_put(stdout_node);
-	return found;
 }
 
 void vpa_init(int cpu)
diff --git a/arch/powerpc/platforms/pseries/pci_dlpar.c b/arch/powerpc/platforms/pseries/pci_dlpar.c
new file mode 100644
index 000000000000..21934784f936
--- /dev/null
+++ b/arch/powerpc/platforms/pseries/pci_dlpar.c
@@ -0,0 +1,174 @@
+/*
+ * PCI Dynamic LPAR, PCI Hot Plug and PCI EEH recovery code
+ * for RPA-compliant PPC64 platform.
+ * Copyright (C) 2003 Linda Xie <lxie@us.ibm.com>
+ * Copyright (C) 2005 International Business Machines
+ *
+ * Updates, 2005, John Rose <johnrose@austin.ibm.com>
+ * Updates, 2005, Linas Vepstas <linas@austin.ibm.com>
+ *
+ * All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; either version 2 of the License, or (at
+ * your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful, but
+ * WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE, GOOD TITLE or
+ * NON INFRINGEMENT.  See the GNU General Public License for more
+ * details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software
+ * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
+ */
+
+#include <linux/pci.h>
+#include <asm/pci-bridge.h>
+
+static struct pci_bus *
+find_bus_among_children(struct pci_bus *bus,
+                        struct device_node *dn)
+{
+	struct pci_bus *child = NULL;
+	struct list_head *tmp;
+	struct device_node *busdn;
+
+	busdn = pci_bus_to_OF_node(bus);
+	if (busdn == dn)
+		return bus;
+
+	list_for_each(tmp, &bus->children) {
+		child = find_bus_among_children(pci_bus_b(tmp), dn);
+		if (child)
+			break;
+	};
+	return child;
+}
+
+struct pci_bus *
+pcibios_find_pci_bus(struct device_node *dn)
+{
+	struct pci_dn *pdn = dn->data;
+
+	if (!pdn  || !pdn->phb || !pdn->phb->bus)
+		return NULL;
+
+	return find_bus_among_children(pdn->phb->bus, dn);
+}
+
+/**
+ * pcibios_remove_pci_devices - remove all devices under this bus
+ *
+ * Remove all of the PCI devices under this bus both from the
+ * linux pci device tree, and from the powerpc EEH address cache.
+ */
+void
+pcibios_remove_pci_devices(struct pci_bus *bus)
+{
+	struct pci_dev *dev, *tmp;
+
+	list_for_each_entry_safe(dev, tmp, &bus->devices, bus_list) {
+		eeh_remove_bus_device(dev);
+		pci_remove_bus_device(dev);
+	}
+}
+
+/* Must be called before pci_bus_add_devices */
+void
+pcibios_fixup_new_pci_devices(struct pci_bus *bus, int fix_bus)
+{
+	struct pci_dev *dev;
+
+	list_for_each_entry(dev, &bus->devices, bus_list) {
+		/*
+		 * Skip already-present devices (which are on the
+		 * global device list.)
+		 */
+		if (list_empty(&dev->global_list)) {
+			int i;
+
+			/* Need to setup IOMMU tables */
+			ppc_md.iommu_dev_setup(dev);
+
+			if(fix_bus)
+				pcibios_fixup_device_resources(dev, bus);
+			pci_read_irq_line(dev);
+			for (i = 0; i < PCI_NUM_RESOURCES; i++) {
+				struct resource *r = &dev->resource[i];
+
+				if (r->parent || !r->start || !r->flags)
+					continue;
+				pci_claim_resource(dev, i);
+			}
+		}
+	}
+}
+
+static int
+pcibios_pci_config_bridge(struct pci_dev *dev)
+{
+	u8 sec_busno;
+	struct pci_bus *child_bus;
+	struct pci_dev *child_dev;
+
+	/* Get busno of downstream bus */
+	pci_read_config_byte(dev, PCI_SECONDARY_BUS, &sec_busno);
+
+	/* Add to children of PCI bridge dev->bus */
+	child_bus = pci_add_new_bus(dev->bus, dev, sec_busno);
+	if (!child_bus) {
+		printk (KERN_ERR "%s: could not add second bus\n", __FUNCTION__);
+		return -EIO;
+	}
+	sprintf(child_bus->name, "PCI Bus #%02x", child_bus->number);
+
+	pci_scan_child_bus(child_bus);
+
+	list_for_each_entry(child_dev, &child_bus->devices, bus_list) {
+		eeh_add_device_late(child_dev);
+	}
+
+	/* Fixup new pci devices without touching bus struct */
+	pcibios_fixup_new_pci_devices(child_bus, 0);
+
+	/* Make the discovered devices available */
+	pci_bus_add_devices(child_bus);
+	return 0;
+}
+
+/**
+ * pcibios_add_pci_devices - adds new pci devices to bus
+ *
+ * This routine will find and fixup new pci devices under
+ * the indicated bus. This routine presumes that there
+ * might already be some devices under this bridge, so
+ * it carefully tries to add only new devices.  (And that
+ * is how this routine differs from other, similar pcibios
+ * routines.)
+ */
+void
+pcibios_add_pci_devices(struct pci_bus * bus)
+{
+	int slotno, num;
+	struct pci_dev *dev;
+	struct device_node *dn = pci_bus_to_OF_node(bus);
+
+	eeh_add_device_tree_early(dn);
+
+	/* pci_scan_slot should find all children */
+	slotno = PCI_SLOT(PCI_DN(dn->child)->devfn);
+	num = pci_scan_slot(bus, PCI_DEVFN(slotno, 0));
+	if (num) {
+		pcibios_fixup_new_pci_devices(bus, 1);
+		pci_bus_add_devices(bus);
+	}
+
+	list_for_each_entry(dev, &bus->devices, bus_list) {
+		eeh_add_device_late (dev);
+		if (dev->hdr_type == PCI_HEADER_TYPE_BRIDGE)
+			pcibios_pci_config_bridge(dev);
+	}
+}
diff --git a/arch/powerpc/platforms/pseries/ras.c b/arch/powerpc/platforms/pseries/ras.c
index fbd214d68b07..b046bcf7443d 100644
--- a/arch/powerpc/platforms/pseries/ras.c
+++ b/arch/powerpc/platforms/pseries/ras.c
@@ -49,14 +49,14 @@
 #include <asm/machdep.h>
 #include <asm/rtas.h>
 #include <asm/udbg.h>
+#include <asm/firmware.h>
+
+#include "ras.h"
 
 static unsigned char ras_log_buf[RTAS_ERROR_LOG_MAX];
 static DEFINE_SPINLOCK(ras_log_buf_lock);
 
-char mce_data_buf[RTAS_ERROR_LOG_MAX]
-;
-/* This is true if we are using the firmware NMI handler (typically LPAR) */
-extern int fwnmi_active;
+char mce_data_buf[RTAS_ERROR_LOG_MAX];
 
 static int ras_get_sensor_state_token;
 static int ras_check_exception_token;
@@ -280,7 +280,7 @@ static void fwnmi_release_errinfo(void)
 		printk("FWNMI: nmi-interlock failed: %d\n", ret);
 }
 
-void pSeries_system_reset_exception(struct pt_regs *regs)
+int pSeries_system_reset_exception(struct pt_regs *regs)
 {
 	if (fwnmi_active) {
 		struct rtas_error_log *errhdr = fwnmi_get_errinfo(regs);
@@ -289,6 +289,7 @@ void pSeries_system_reset_exception(struct pt_regs *regs)
 		}
 		fwnmi_release_errinfo();
 	}
+	return 0; /* need to perform reset */
 }
 
 /*
diff --git a/arch/powerpc/platforms/pseries/ras.h b/arch/powerpc/platforms/pseries/ras.h
new file mode 100644
index 000000000000..0e66b0da55e2
--- /dev/null
+++ b/arch/powerpc/platforms/pseries/ras.h
@@ -0,0 +1,9 @@
+#ifndef _PSERIES_RAS_H
+#define _PSERIES_RAS_H
+
+struct pt_regs;
+
+extern int pSeries_system_reset_exception(struct pt_regs *regs);
+extern int pSeries_machine_check_exception(struct pt_regs *regs);
+
+#endif /* _PSERIES_RAS_H */
diff --git a/arch/powerpc/platforms/pseries/setup.c b/arch/powerpc/platforms/pseries/setup.c
index 4a465f067ede..8903cf63236a 100644
--- a/arch/powerpc/platforms/pseries/setup.c
+++ b/arch/powerpc/platforms/pseries/setup.c
@@ -56,6 +56,7 @@
 #include <asm/dma.h>
 #include <asm/machdep.h>
 #include <asm/irq.h>
+#include <asm/kexec.h>
 #include <asm/time.h>
 #include <asm/nvram.h>
 #include "xics.h"
@@ -68,6 +69,7 @@
 #include <asm/smp.h>
 
 #include "plpar_wrappers.h"
+#include "ras.h"
 
 #ifdef DEBUG
 #define DBG(fmt...) udbg_printf(fmt)
@@ -76,16 +78,9 @@
 #endif
 
 extern void find_udbg_vterm(void);
-extern void system_reset_fwnmi(void);	/* from head.S */
-extern void machine_check_fwnmi(void);	/* from head.S */
-extern void generic_find_legacy_serial_ports(u64 *physport,
-		unsigned int *default_speed);
 
 int fwnmi_active;  /* TRUE if an FWNMI handler is present */
 
-extern void pSeries_system_reset_exception(struct pt_regs *regs);
-extern int pSeries_machine_check_exception(struct pt_regs *regs);
-
 static void pseries_shared_idle(void);
 static void pseries_dedicated_idle(void);
 
@@ -105,18 +100,22 @@ void pSeries_show_cpuinfo(struct seq_file *m)
 
 /* Initialize firmware assisted non-maskable interrupts if
  * the firmware supports this feature.
- *
  */
 static void __init fwnmi_init(void)
 {
-	int ret;
+	unsigned long system_reset_addr, machine_check_addr;
+
 	int ibm_nmi_register = rtas_token("ibm,nmi-register");
 	if (ibm_nmi_register == RTAS_UNKNOWN_SERVICE)
 		return;
-	ret = rtas_call(ibm_nmi_register, 2, 1, NULL,
-			__pa((unsigned long)system_reset_fwnmi),
-			__pa((unsigned long)machine_check_fwnmi));
-	if (ret == 0)
+
+	/* If the kernel's not linked at zero we point the firmware at low
+	 * addresses anyway, and use a trampoline to get to the real code. */
+	system_reset_addr  = __pa(system_reset_fwnmi) - PHYSICAL_START;
+	machine_check_addr = __pa(machine_check_fwnmi) - PHYSICAL_START;
+
+	if (0 == rtas_call(ibm_nmi_register, 2, 1, NULL, system_reset_addr,
+				machine_check_addr))
 		fwnmi_active = 1;
 }
 
@@ -323,15 +322,18 @@ static  void __init pSeries_discover_pic(void)
 	ppc64_interrupt_controller = IC_INVALID;
 	for (np = NULL; (np = of_find_node_by_name(np, "interrupt-controller"));) {
 		typep = (char *)get_property(np, "compatible", NULL);
-		if (strstr(typep, "open-pic"))
+		if (strstr(typep, "open-pic")) {
 			ppc64_interrupt_controller = IC_OPEN_PIC;
-		else if (strstr(typep, "ppc-xicp"))
+			break;
+		} else if (strstr(typep, "ppc-xicp")) {
 			ppc64_interrupt_controller = IC_PPC_XIC;
-		else
-			printk("pSeries_discover_pic: failed to recognize"
-			       " interrupt-controller\n");
-		break;
+			break;
+		}
 	}
+	if (ppc64_interrupt_controller == IC_INVALID)
+		printk("pSeries_discover_pic: failed to recognize"
+			" interrupt-controller\n");
+
 }
 
 static void pSeries_mach_cpu_die(void)
@@ -365,10 +367,7 @@ static int pseries_set_xdabr(unsigned long dabr)
  */
 static void __init pSeries_init_early(void)
 {
-	void *comport;
 	int iommu_off = 0;
-	unsigned int default_speed;
-	u64 physport;
 
 	DBG(" -> pSeries_init_early()\n");
 
@@ -382,17 +381,8 @@ static void __init pSeries_init_early(void)
 			     get_property(of_chosen, "linux,iommu-off", NULL));
 	}
 
-	generic_find_legacy_serial_ports(&physport, &default_speed);
-
 	if (platform_is_lpar())
 		find_udbg_vterm();
-	else if (physport) {
-		/* Map the uart for udbg. */
-		comport = (void *)ioremap(physport, 16);
-		udbg_init_uart(comport, default_speed);
-
-		DBG("Hello World !\n");
-	}
 
 	if (firmware_has_feature(FW_FEATURE_DABR))
 		ppc_md.set_dabr = pseries_set_dabr;
@@ -638,5 +628,8 @@ struct machdep_calls __initdata pSeries_md = {
 	.machine_check_exception = pSeries_machine_check_exception,
 #ifdef CONFIG_KEXEC
 	.kexec_cpu_down		= pseries_kexec_cpu_down,
+	.machine_kexec		= default_machine_kexec,
+	.machine_kexec_prepare	= default_machine_kexec_prepare,
+	.machine_crash_shutdown	= default_machine_crash_shutdown,
 #endif
 };
diff --git a/arch/powerpc/platforms/pseries/xics.c b/arch/powerpc/platforms/pseries/xics.c
index 0377decc0719..0c0cfa32eb58 100644
--- a/arch/powerpc/platforms/pseries/xics.c
+++ b/arch/powerpc/platforms/pseries/xics.c
@@ -407,7 +407,7 @@ irqreturn_t xics_ipi_action(int irq, void *dev_id, struct pt_regs *regs)
 			smp_message_recv(PPC_MSG_MIGRATE_TASK, regs);
 		}
 #endif
-#ifdef CONFIG_DEBUGGER
+#if defined(CONFIG_DEBUGGER) || defined(CONFIG_KEXEC)
 		if (test_and_clear_bit(PPC_MSG_DEBUGGER_BREAK,
 				       &xics_ipi_message[cpu].value)) {
 			mb();