diff options
Diffstat (limited to 'drivers/staging/tidspbridge/core')
| -rw-r--r-- | drivers/staging/tidspbridge/core/_deh.h | 5 | ||||
| -rw-r--r-- | drivers/staging/tidspbridge/core/_tiomap.h | 19 | ||||
| -rw-r--r-- | drivers/staging/tidspbridge/core/dsp-mmu.c | 317 | ||||
| -rw-r--r-- | drivers/staging/tidspbridge/core/io_sm.c | 180 | ||||
| -rw-r--r-- | drivers/staging/tidspbridge/core/tiomap3430.c | 1083 | ||||
| -rw-r--r-- | drivers/staging/tidspbridge/core/tiomap3430_pwr.c | 4 | ||||
| -rw-r--r-- | drivers/staging/tidspbridge/core/tiomap_io.c | 17 | ||||
| -rw-r--r-- | drivers/staging/tidspbridge/core/ue_deh.c | 115 |
8 files changed, 1275 insertions, 465 deletions
diff --git a/drivers/staging/tidspbridge/core/_deh.h b/drivers/staging/tidspbridge/core/_deh.h index 8ae263387a87..16723cd34831 100644 --- a/drivers/staging/tidspbridge/core/_deh.h +++ b/drivers/staging/tidspbridge/core/_deh.h @@ -27,8 +27,9 @@ struct deh_mgr { struct bridge_dev_context *hbridge_context; /* Bridge context. */ struct ntfy_object *ntfy_obj; /* NTFY object */ -}; -int mmu_fault_isr(struct iommu *mmu); + /* MMU Fault DPC */ + struct tasklet_struct dpc_tasklet; +}; #endif /* _DEH_ */ diff --git a/drivers/staging/tidspbridge/core/_tiomap.h b/drivers/staging/tidspbridge/core/_tiomap.h index e0a801c1cb98..1c1f157e167a 100644 --- a/drivers/staging/tidspbridge/core/_tiomap.h +++ b/drivers/staging/tidspbridge/core/_tiomap.h @@ -23,8 +23,8 @@ #include <plat/clockdomain.h> #include <mach-omap2/prm-regbits-34xx.h> #include <mach-omap2/cm-regbits-34xx.h> -#include <dspbridge/dsp-mmu.h> #include <dspbridge/devdefs.h> +#include <hw_defs.h> #include <dspbridge/dspioctl.h> /* for bridge_ioctl_extproc defn */ #include <dspbridge/sync.h> #include <dspbridge/clk.h> @@ -306,18 +306,6 @@ static const struct bpwr_clk_t bpwr_clks[] = { #define CLEAR_BIT_INDEX(reg, index) (reg &= ~(1 << (index))) -struct shm_segs { - u32 seg0_da; - u32 seg0_pa; - u32 seg0_va; - u32 seg0_size; - u32 seg1_da; - u32 seg1_pa; - u32 seg1_va; - u32 seg1_size; -}; - - /* This Bridge driver's device context: */ struct bridge_dev_context { struct dev_object *hdev_obj; /* Handle to Bridge device object. */ @@ -328,6 +316,7 @@ struct bridge_dev_context { */ u32 dw_dsp_ext_base_addr; /* See the comment above */ u32 dw_api_reg_base; /* API mem map'd registers */ + void __iomem *dw_dsp_mmu_base; /* DSP MMU Mapped registers */ u32 dw_api_clk_base; /* CLK Registers */ u32 dw_dsp_clk_m2_base; /* DSP Clock Module m2 */ u32 dw_public_rhea; /* Pub Rhea */ @@ -339,8 +328,7 @@ struct bridge_dev_context { u32 dw_internal_size; /* Internal memory size */ struct omap_mbox *mbox; /* Mail box handle */ - struct iommu *dsp_mmu; /* iommu for iva2 handler */ - struct shm_segs sh_s; + struct cfg_hostres *resources; /* Host Resources */ /* @@ -353,6 +341,7 @@ struct bridge_dev_context { /* TC Settings */ bool tc_word_swap_on; /* Traffic Controller Word Swap */ + struct pg_table_attrs *pt_attrs; u32 dsp_per_clks; }; diff --git a/drivers/staging/tidspbridge/core/dsp-mmu.c b/drivers/staging/tidspbridge/core/dsp-mmu.c deleted file mode 100644 index 983c95adc8ff..000000000000 --- a/drivers/staging/tidspbridge/core/dsp-mmu.c +++ /dev/null @@ -1,317 +0,0 @@ -/* - * dsp-mmu.c - * - * DSP-BIOS Bridge driver support functions for TI OMAP processors. - * - * DSP iommu. - * - * Copyright (C) 2010 Texas Instruments, Inc. - * - * This package is free software; you can redistribute it and/or modify - * it under the terms of the GNU General Public License version 2 as - * published by the Free Software Foundation. - * - * THIS PACKAGE IS PROVIDED ``AS IS'' AND WITHOUT ANY EXPRESS OR - * IMPLIED WARRANTIES, INCLUDING, WITHOUT LIMITATION, THE IMPLIED - * WARRANTIES OF MERCHANTIBILITY AND FITNESS FOR A PARTICULAR PURPOSE. - */ - -#include <dspbridge/host_os.h> -#include <plat/dmtimer.h> -#include <dspbridge/dbdefs.h> -#include <dspbridge/dev.h> -#include <dspbridge/io_sm.h> -#include <dspbridge/dspdeh.h> -#include "_tiomap.h" - -#include <dspbridge/dsp-mmu.h> - -#define MMU_CNTL_TWL_EN (1 << 2) - -static struct tasklet_struct mmu_tasklet; - -#ifdef CONFIG_TIDSPBRIDGE_BACKTRACE -static void mmu_fault_print_stack(struct bridge_dev_context *dev_context) -{ - void *dummy_addr; - u32 fa, tmp; - struct iotlb_entry e; - struct iommu *mmu = dev_context->dsp_mmu; - dummy_addr = (void *)__get_free_page(GFP_ATOMIC); - - /* - * Before acking the MMU fault, let's make sure MMU can only - * access entry #0. Then add a new entry so that the DSP OS - * can continue in order to dump the stack. - */ - tmp = iommu_read_reg(mmu, MMU_CNTL); - tmp &= ~MMU_CNTL_TWL_EN; - iommu_write_reg(mmu, tmp, MMU_CNTL); - fa = iommu_read_reg(mmu, MMU_FAULT_AD); - e.da = fa & PAGE_MASK; - e.pa = virt_to_phys(dummy_addr); - e.valid = 1; - e.prsvd = 1; - e.pgsz = IOVMF_PGSZ_4K & MMU_CAM_PGSZ_MASK; - e.endian = MMU_RAM_ENDIAN_LITTLE; - e.elsz = MMU_RAM_ELSZ_32; - e.mixed = 0; - - load_iotlb_entry(mmu, &e); - - dsp_clk_enable(DSP_CLK_GPT8); - - dsp_gpt_wait_overflow(DSP_CLK_GPT8, 0xfffffffe); - - /* Clear MMU interrupt */ - tmp = iommu_read_reg(mmu, MMU_IRQSTATUS); - iommu_write_reg(mmu, tmp, MMU_IRQSTATUS); - - dump_dsp_stack(dev_context); - dsp_clk_disable(DSP_CLK_GPT8); - - iopgtable_clear_entry(mmu, fa); - free_page((unsigned long)dummy_addr); -} -#endif - - -static void fault_tasklet(unsigned long data) -{ - struct iommu *mmu = (struct iommu *)data; - struct bridge_dev_context *dev_ctx; - struct deh_mgr *dm; - u32 fa; - dev_get_deh_mgr(dev_get_first(), &dm); - dev_get_bridge_context(dev_get_first(), &dev_ctx); - - if (!dm || !dev_ctx) - return; - - fa = iommu_read_reg(mmu, MMU_FAULT_AD); - -#ifdef CONFIG_TIDSPBRIDGE_BACKTRACE - print_dsp_trace_buffer(dev_ctx); - dump_dl_modules(dev_ctx); - mmu_fault_print_stack(dev_ctx); -#endif - - bridge_deh_notify(dm, DSP_MMUFAULT, fa); -} - -/* - * ======== mmu_fault_isr ======== - * ISR to be triggered by a DSP MMU fault interrupt. - */ -static int mmu_fault_callback(struct iommu *mmu) -{ - if (!mmu) - return -EPERM; - - iommu_write_reg(mmu, 0, MMU_IRQENABLE); - tasklet_schedule(&mmu_tasklet); - return 0; -} - -/** - * dsp_mmu_init() - initialize dsp_mmu module and returns a handle - * - * This function initialize dsp mmu module and returns a struct iommu - * handle to use it for dsp maps. - * - */ -struct iommu *dsp_mmu_init() -{ - struct iommu *mmu; - - mmu = iommu_get("iva2"); - - if (!IS_ERR(mmu)) { - tasklet_init(&mmu_tasklet, fault_tasklet, (unsigned long)mmu); - mmu->isr = mmu_fault_callback; - } - - return mmu; -} - -/** - * dsp_mmu_exit() - destroy dsp mmu module - * @mmu: Pointer to iommu handle. - * - * This function destroys dsp mmu module. - * - */ -void dsp_mmu_exit(struct iommu *mmu) -{ - if (mmu) - iommu_put(mmu); - tasklet_kill(&mmu_tasklet); -} - -/** - * user_va2_pa() - get physical address from userspace address. - * @mm: mm_struct Pointer of the process. - * @address: Virtual user space address. - * - */ -static u32 user_va2_pa(struct mm_struct *mm, u32 address) -{ - pgd_t *pgd; - pmd_t *pmd; - pte_t *ptep, pte; - - pgd = pgd_offset(mm, address); - if (!(pgd_none(*pgd) || pgd_bad(*pgd))) { - pmd = pmd_offset(pgd, address); - if (!(pmd_none(*pmd) || pmd_bad(*pmd))) { - ptep = pte_offset_map(pmd, address); - if (ptep) { - pte = *ptep; - if (pte_present(pte)) - return pte & PAGE_MASK; - } - } - } - - return 0; -} - -/** - * get_io_pages() - pin and get pages of io user's buffer. - * @mm: mm_struct Pointer of the process. - * @uva: Virtual user space address. - * @pages Pages to be pined. - * @usr_pgs struct page array pointer where the user pages will be stored - * - */ -static int get_io_pages(struct mm_struct *mm, u32 uva, unsigned pages, - struct page **usr_pgs) -{ - u32 pa; - int i; - struct page *pg; - - for (i = 0; i < pages; i++) { - pa = user_va2_pa(mm, uva); - - if (!pfn_valid(__phys_to_pfn(pa))) - break; - - pg = phys_to_page(pa); - usr_pgs[i] = pg; - get_page(pg); - } - return i; -} - -/** - * user_to_dsp_map() - maps user to dsp virtual address - * @mmu: Pointer to iommu handle. - * @uva: Virtual user space address. - * @da DSP address - * @size Buffer size to map. - * @usr_pgs struct page array pointer where the user pages will be stored - * - * This function maps a user space buffer into DSP virtual address. - * - */ -u32 user_to_dsp_map(struct iommu *mmu, u32 uva, u32 da, u32 size, - struct page **usr_pgs) -{ - int res, w; - unsigned pages; - int i; - struct vm_area_struct *vma; - struct mm_struct *mm = current->mm; - struct sg_table *sgt; - struct scatterlist *sg; - - if (!size || !usr_pgs) - return -EINVAL; - - pages = size / PG_SIZE4K; - - down_read(&mm->mmap_sem); - vma = find_vma(mm, uva); - while (vma && (uva + size > vma->vm_end)) - vma = find_vma(mm, vma->vm_end + 1); - - if (!vma) { - pr_err("%s: Failed to get VMA region for 0x%x (%d)\n", - __func__, uva, size); - up_read(&mm->mmap_sem); - return -EINVAL; - } - if (vma->vm_flags & (VM_WRITE | VM_MAYWRITE)) - w = 1; - - if (vma->vm_flags & VM_IO) - i = get_io_pages(mm, uva, pages, usr_pgs); - else - i = get_user_pages(current, mm, uva, pages, w, 1, - usr_pgs, NULL); - up_read(&mm->mmap_sem); - - if (i < 0) - return i; - - if (i < pages) { - res = -EFAULT; - goto err_pages; - } - - sgt = kzalloc(sizeof(*sgt), GFP_KERNEL); - if (!sgt) { - res = -ENOMEM; - goto err_pages; - } - - res = sg_alloc_table(sgt, pages, GFP_KERNEL); - - if (res < 0) - goto err_sg; - - for_each_sg(sgt->sgl, sg, sgt->nents, i) - sg_set_page(sg, usr_pgs[i], PAGE_SIZE, 0); - - da = iommu_vmap(mmu, da, sgt, IOVMF_ENDIAN_LITTLE | IOVMF_ELSZ_32); - - if (!IS_ERR_VALUE(da)) - return da; - res = (int)da; - - sg_free_table(sgt); -err_sg: - kfree(sgt); - i = pages; -err_pages: - while (i--) - put_page(usr_pgs[i]); - return res; -} - -/** - * user_to_dsp_unmap() - unmaps DSP virtual buffer. - * @mmu: Pointer to iommu handle. - * @da DSP address - * - * This function unmaps a user space buffer into DSP virtual address. - * - */ -int user_to_dsp_unmap(struct iommu *mmu, u32 da) -{ - unsigned i; - struct sg_table *sgt; - struct scatterlist *sg; - - sgt = iommu_vunmap(mmu, da); - if (!sgt) - return -EFAULT; - - for_each_sg(sgt->sgl, sg, sgt->nents, i) - put_page(sg_page(sg)); - sg_free_table(sgt); - kfree(sgt); - - return 0; -} diff --git a/drivers/staging/tidspbridge/core/io_sm.c b/drivers/staging/tidspbridge/core/io_sm.c index 194badaba0ed..571864555ddd 100644 --- a/drivers/staging/tidspbridge/core/io_sm.c +++ b/drivers/staging/tidspbridge/core/io_sm.c @@ -39,6 +39,10 @@ #include <dspbridge/ntfy.h> #include <dspbridge/sync.h> +/* Hardware Abstraction Layer */ +#include <hw_defs.h> +#include <hw_mmu.h> + /* Bridge Driver */ #include <dspbridge/dspdeh.h> #include <dspbridge/dspio.h> @@ -287,7 +291,6 @@ int bridge_io_on_loaded(struct io_mgr *hio_mgr) struct cod_manager *cod_man; struct chnl_mgr *hchnl_mgr; struct msg_mgr *hmsg_mgr; - struct shm_segs *sm_sg; u32 ul_shm_base; u32 ul_shm_base_offset; u32 ul_shm_limit; @@ -310,9 +313,18 @@ int bridge_io_on_loaded(struct io_mgr *hio_mgr) struct bridge_ioctl_extproc ae_proc[BRDIOCTL_NUMOFMMUTLB]; struct cfg_hostres *host_res; struct bridge_dev_context *pbridge_context; + u32 map_attrs; u32 shm0_end; u32 ul_dyn_ext_base; u32 ul_seg1_size = 0; + u32 pa_curr = 0; + u32 va_curr = 0; + u32 gpp_va_curr = 0; + u32 num_bytes = 0; + u32 all_bits = 0; + u32 page_size[] = { HW_PAGE_SIZE16MB, HW_PAGE_SIZE1MB, + HW_PAGE_SIZE64KB, HW_PAGE_SIZE4KB + }; status = dev_get_bridge_context(hio_mgr->hdev_obj, &pbridge_context); if (!pbridge_context) { @@ -325,8 +337,6 @@ int bridge_io_on_loaded(struct io_mgr *hio_mgr) status = -EFAULT; goto func_end; } - sm_sg = &pbridge_context->sh_s; - status = dev_get_cod_mgr(hio_mgr->hdev_obj, &cod_man); if (!cod_man) { status = -EFAULT; @@ -461,14 +471,129 @@ int bridge_io_on_loaded(struct io_mgr *hio_mgr) if (status) goto func_end; - sm_sg->seg1_pa = ul_gpp_pa; - sm_sg->seg1_da = ul_dyn_ext_base; - sm_sg->seg1_va = ul_gpp_va; - sm_sg->seg1_size = ul_seg1_size; - sm_sg->seg0_pa = ul_gpp_pa + ul_pad_size + ul_seg1_size; - sm_sg->seg0_da = ul_dsp_va; - sm_sg->seg0_va = ul_gpp_va + ul_pad_size + ul_seg1_size; - sm_sg->seg0_size = ul_seg_size; + pa_curr = ul_gpp_pa; + va_curr = ul_dyn_ext_base * hio_mgr->word_size; + gpp_va_curr = ul_gpp_va; + num_bytes = ul_seg1_size; + + /* + * Try to fit into TLB entries. If not possible, push them to page + * tables. It is quite possible that if sections are not on + * bigger page boundary, we may end up making several small pages. + * So, push them onto page tables, if that is the case. + */ + map_attrs = 0x00000000; + map_attrs = DSP_MAPLITTLEENDIAN; + map_attrs |= DSP_MAPPHYSICALADDR; + map_attrs |= DSP_MAPELEMSIZE32; + map_attrs |= DSP_MAPDONOTLOCK; + + while (num_bytes) { + /* + * To find the max. page size with which both PA & VA are + * aligned. + */ + all_bits = pa_curr | va_curr; + dev_dbg(bridge, "all_bits %x, pa_curr %x, va_curr %x, " + "num_bytes %x\n", all_bits, pa_curr, va_curr, + num_bytes); + for (i = 0; i < 4; i++) { + if ((num_bytes >= page_size[i]) && ((all_bits & + (page_size[i] - + 1)) == 0)) { + status = + hio_mgr->intf_fxns-> + pfn_brd_mem_map(hio_mgr->hbridge_context, + pa_curr, va_curr, + page_size[i], map_attrs, + NULL); + if (status) + goto func_end; + pa_curr += page_size[i]; + va_curr += page_size[i]; + gpp_va_curr += page_size[i]; + num_bytes -= page_size[i]; + /* + * Don't try smaller sizes. Hopefully we have + * reached an address aligned to a bigger page + * size. + */ + break; + } + } + } + pa_curr += ul_pad_size; + va_curr += ul_pad_size; + gpp_va_curr += ul_pad_size; + + /* Configure the TLB entries for the next cacheable segment */ + num_bytes = ul_seg_size; + va_curr = ul_dsp_va * hio_mgr->word_size; + while (num_bytes) { + /* + * To find the max. page size with which both PA & VA are + * aligned. + */ + all_bits = pa_curr | va_curr; + dev_dbg(bridge, "all_bits for Seg1 %x, pa_curr %x, " + "va_curr %x, num_bytes %x\n", all_bits, pa_curr, + va_curr, num_bytes); + for (i = 0; i < 4; i++) { + if (!(num_bytes >= page_size[i]) || + !((all_bits & (page_size[i] - 1)) == 0)) + continue; + if (ndx < MAX_LOCK_TLB_ENTRIES) { + /* + * This is the physical address written to + * DSP MMU. + */ + ae_proc[ndx].ul_gpp_pa = pa_curr; + /* + * This is the virtual uncached ioremapped + * address!!! + */ + ae_proc[ndx].ul_gpp_va = gpp_va_curr; + ae_proc[ndx].ul_dsp_va = + va_curr / hio_mgr->word_size; + ae_proc[ndx].ul_size = page_size[i]; + ae_proc[ndx].endianism = HW_LITTLE_ENDIAN; + ae_proc[ndx].elem_size = HW_ELEM_SIZE16BIT; + ae_proc[ndx].mixed_mode = HW_MMU_CPUES; + dev_dbg(bridge, "shm MMU TLB entry PA %x" + " VA %x DSP_VA %x Size %x\n", + ae_proc[ndx].ul_gpp_pa, + ae_proc[ndx].ul_gpp_va, + ae_proc[ndx].ul_dsp_va * + hio_mgr->word_size, page_size[i]); + ndx++; + } else { + status = + hio_mgr->intf_fxns-> + pfn_brd_mem_map(hio_mgr->hbridge_context, + pa_curr, va_curr, + page_size[i], map_attrs, + NULL); + dev_dbg(bridge, + "shm MMU PTE entry PA %x" + " VA %x DSP_VA %x Size %x\n", + ae_proc[ndx].ul_gpp_pa, + ae_proc[ndx].ul_gpp_va, + ae_proc[ndx].ul_dsp_va * + hio_mgr->word_size, page_size[i]); + if (status) + goto func_end; + } + pa_curr += page_size[i]; + va_curr += page_size[i]; + gpp_va_curr += page_size[i]; + num_bytes -= page_size[i]; + /* + * Don't try smaller sizes. Hopefully we have reached + * an address aligned to a bigger page size. + */ + break; + } + } /* * Copy remaining entries from CDB. All entries are 1 MB and @@ -509,12 +634,38 @@ int bridge_io_on_loaded(struct io_mgr *hio_mgr) "DSP_VA 0x%x\n", ae_proc[ndx].ul_gpp_pa, ae_proc[ndx].ul_dsp_va); ndx++; + } else { + status = hio_mgr->intf_fxns->pfn_brd_mem_map + (hio_mgr->hbridge_context, + hio_mgr->ext_proc_info.ty_tlb[i]. + ul_gpp_phys, + hio_mgr->ext_proc_info.ty_tlb[i]. + ul_dsp_virt, 0x100000, map_attrs, + NULL); } } if (status) goto func_end; } + map_attrs = 0x00000000; + map_attrs = DSP_MAPLITTLEENDIAN; + map_attrs |= DSP_MAPPHYSICALADDR; + map_attrs |= DSP_MAPELEMSIZE32; + map_attrs |= DSP_MAPDONOTLOCK; + + /* Map the L4 peripherals */ + i = 0; + while (l4_peripheral_table[i].phys_addr) { + status = hio_mgr->intf_fxns->pfn_brd_mem_map + (hio_mgr->hbridge_context, l4_peripheral_table[i].phys_addr, + l4_peripheral_table[i].dsp_virt_addr, HW_PAGE_SIZE4KB, + map_attrs, NULL); + if (status) + goto func_end; + i++; + } + for (i = ndx; i < BRDIOCTL_NUMOFMMUTLB; i++) { ae_proc[i].ul_dsp_va = 0; ae_proc[i].ul_gpp_pa = 0; @@ -537,12 +688,12 @@ int bridge_io_on_loaded(struct io_mgr *hio_mgr) status = -EFAULT; goto func_end; } else { - if (sm_sg->seg0_da > ul_shm_base) { + if (ae_proc[0].ul_dsp_va > ul_shm_base) { status = -EPERM; goto func_end; } /* ul_shm_base may not be at ul_dsp_va address */ - ul_shm_base_offset = (ul_shm_base - sm_sg->seg0_da) * + ul_shm_base_offset = (ul_shm_base - ae_proc[0].ul_dsp_va) * hio_mgr->word_size; /* * bridge_dev_ctrl() will set dev context dsp-mmu info. In @@ -566,7 +717,8 @@ int bridge_io_on_loaded(struct io_mgr *hio_mgr) goto func_end; } /* Register SM */ - status = register_shm_segs(hio_mgr, cod_man, sm_sg->seg0_pa); + status = + register_shm_segs(hio_mgr, cod_man, ae_proc[0].ul_gpp_pa); } hio_mgr->shared_mem = (struct shm *)ul_shm_base; diff --git a/drivers/staging/tidspbridge/core/tiomap3430.c b/drivers/staging/tidspbridge/core/tiomap3430.c index f22bc12bc0d3..1be081f917a7 100644 --- a/drivers/staging/tidspbridge/core/tiomap3430.c +++ b/drivers/staging/tidspbridge/core/tiomap3430.c @@ -23,7 +23,6 @@ #include <dspbridge/host_os.h> #include <linux/mm.h> #include <linux/mmzone.h> -#include <plat/control.h> /* ----------------------------------- DSP/BIOS Bridge */ #include <dspbridge/dbdefs.h> @@ -35,6 +34,10 @@ #include <dspbridge/drv.h> #include <dspbridge/sync.h> +/* ------------------------------------ Hardware Abstraction Layer */ +#include <hw_defs.h> +#include <hw_mmu.h> + /* ----------------------------------- Link Driver */ #include <dspbridge/dspdefs.h> #include <dspbridge/dspchnl.h> @@ -47,6 +50,7 @@ /* ----------------------------------- Platform Manager */ #include <dspbridge/dev.h> #include <dspbridge/dspapi.h> +#include <dspbridge/dmm.h> #include <dspbridge/wdt.h> /* ----------------------------------- Local */ @@ -67,6 +71,20 @@ #define MMU_SMALL_PAGE_MASK 0xFFFFF000 #define OMAP3_IVA2_BOOTADDR_MASK 0xFFFFFC00 #define PAGES_II_LVL_TABLE 512 +#define PHYS_TO_PAGE(phys) pfn_to_page((phys) >> PAGE_SHIFT) + +/* + * This is a totally ugly layer violation, but needed until + * omap_ctrl_set_dsp_boot*() are provided. + */ +#define OMAP3_IVA2_BOOTMOD_IDLE 1 +#define OMAP2_CONTROL_GENERAL 0x270 +#define OMAP343X_CONTROL_IVA2_BOOTADDR (OMAP2_CONTROL_GENERAL + 0x0190) +#define OMAP343X_CONTROL_IVA2_BOOTMOD (OMAP2_CONTROL_GENERAL + 0x0194) + +#define OMAP343X_CTRL_REGADDR(reg) \ + OMAP2_L4_IO_ADDRESS(OMAP343X_CTRL_BASE + (reg)) + /* Forward Declarations: */ static int bridge_brd_monitor(struct bridge_dev_context *dev_ctxt); @@ -91,6 +109,12 @@ static int bridge_brd_mem_copy(struct bridge_dev_context *dev_ctxt, static int bridge_brd_mem_write(struct bridge_dev_context *dev_ctxt, u8 *host_buff, u32 dsp_addr, u32 ul_num_bytes, u32 mem_type); +static int bridge_brd_mem_map(struct bridge_dev_context *dev_ctxt, + u32 ul_mpu_addr, u32 virt_addr, + u32 ul_num_bytes, u32 ul_map_attr, + struct page **mapped_pages); +static int bridge_brd_mem_un_map(struct bridge_dev_context *dev_ctxt, + u32 virt_addr, u32 ul_num_bytes); static int bridge_dev_create(struct bridge_dev_context **dev_cntxt, struct dev_object *hdev_obj, @@ -98,8 +122,57 @@ static int bridge_dev_create(struct bridge_dev_context static int bridge_dev_ctrl(struct bridge_dev_context *dev_context, u32 dw_cmd, void *pargs); static int bridge_dev_destroy(struct bridge_dev_context *dev_ctxt); +static u32 user_va2_pa(struct mm_struct *mm, u32 address); +static int pte_update(struct bridge_dev_context *dev_ctxt, u32 pa, + u32 va, u32 size, + struct hw_mmu_map_attrs_t *map_attrs); +static int pte_set(struct pg_table_attrs *pt, u32 pa, u32 va, + u32 size, struct hw_mmu_map_attrs_t *attrs); +static int mem_map_vmalloc(struct bridge_dev_context *dev_context, + u32 ul_mpu_addr, u32 virt_addr, + u32 ul_num_bytes, + struct hw_mmu_map_attrs_t *hw_attrs); + bool wait_for_start(struct bridge_dev_context *dev_context, u32 dw_sync_addr); +/* ----------------------------------- Globals */ + +/* Attributes of L2 page tables for DSP MMU */ +struct page_info { + u32 num_entries; /* Number of valid PTEs in the L2 PT */ +}; + +/* Attributes used to manage the DSP MMU page tables */ +struct pg_table_attrs { + spinlock_t pg_lock; /* Critical section object handle */ + + u32 l1_base_pa; /* Physical address of the L1 PT */ + u32 l1_base_va; /* Virtual address of the L1 PT */ + u32 l1_size; /* Size of the L1 PT */ + u32 l1_tbl_alloc_pa; + /* Physical address of Allocated mem for L1 table. May not be aligned */ + u32 l1_tbl_alloc_va; + /* Virtual address of Allocated mem for L1 table. May not be aligned */ + u32 l1_tbl_alloc_sz; + /* Size of consistent memory allocated for L1 table. + * May not be aligned */ + + u32 l2_base_pa; /* Physical address of the L2 PT */ + u32 l2_base_va; /* Virtual address of the L2 PT */ + u32 l2_size; /* Size of the L2 PT */ + u32 l2_tbl_alloc_pa; + /* Physical address of Allocated mem for L2 table. May not be aligned */ + u32 l2_tbl_alloc_va; + /* Virtual address of Allocated mem for L2 table. May not be aligned */ + u32 l2_tbl_alloc_sz; + /* Size of consistent memory allocated for L2 table. + * May not be aligned */ + + u32 l2_num_pages; /* Number of allocated L2 PT */ + /* Array [l2_num_pages] of L2 PT info structs */ + struct page_info *pg_info; +}; + /* * This Bridge driver's function interface table. */ @@ -119,6 +192,8 @@ static struct bridge_drv_interface drv_interface_fxns = { bridge_brd_set_state, bridge_brd_mem_copy, bridge_brd_mem_write, + bridge_brd_mem_map, + bridge_brd_mem_un_map, /* The following CHNL functions are provided by chnl_io.lib: */ bridge_chnl_create, bridge_chnl_destroy, @@ -148,6 +223,27 @@ static struct bridge_drv_interface drv_interface_fxns = { bridge_msg_set_queue_id, }; +static inline void flush_all(struct bridge_dev_context *dev_context) +{ + if (dev_context->dw_brd_state == BRD_DSP_HIBERNATION || + dev_context->dw_brd_state == BRD_HIBERNATION) + wake_dsp(dev_context, NULL); + + hw_mmu_tlb_flush_all(dev_context->dw_dsp_mmu_base); +} + +static void bad_page_dump(u32 pa, struct page *pg) +{ + pr_emerg("DSPBRIDGE: MAP function: COUNT 0 FOR PA 0x%x\n", pa); + pr_emerg("Bad page state in process '%s'\n" + "page:%p flags:0x%0*lx mapping:%p mapcount:%d count:%d\n" + "Backtrace:\n", + current->comm, pg, (int)(2 * sizeof(unsigned long)), + (unsigned long)pg->flags, pg->mapping, + page_mapcount(pg), page_count(pg)); + dump_stack(); +} + /* * ======== bridge_drv_entry ======== * purpose: @@ -203,7 +299,8 @@ static int bridge_brd_monitor(struct bridge_dev_context *dev_ctxt) (*pdata->dsp_cm_write)(OMAP34XX_CLKSTCTRL_DISABLE_AUTO, OMAP3430_IVA2_MOD, OMAP2_CM_CLKSTCTRL); } - + (*pdata->dsp_prm_rmw_bits)(OMAP3430_RST2_IVA2_MASK, 0, + OMAP3430_IVA2_MOD, OMAP2_RM_RSTCTRL); dsp_clk_enable(DSP_CLK_IVA2); /* set the device state to IDLE */ @@ -274,17 +371,14 @@ static int bridge_brd_start(struct bridge_dev_context *dev_ctxt, { int status = 0; struct bridge_dev_context *dev_context = dev_ctxt; - struct iommu *mmu = NULL; - struct shm_segs *sm_sg; - int l4_i = 0, tlb_i = 0; - u32 sg0_da = 0, sg1_da = 0; - struct bridge_ioctl_extproc *tlb = dev_context->atlb_entry; u32 dw_sync_addr = 0; u32 ul_shm_base; /* Gpp Phys SM base addr(byte) */ u32 ul_shm_base_virt; /* Dsp Virt SM base addr */ u32 ul_tlb_base_virt; /* Base of MMU TLB entry */ /* Offset of shm_base_virt from tlb_base_virt */ u32 ul_shm_offset_virt; + s32 entry_ndx; + s32 itmp_entry_ndx = 0; /* DSP-MMU TLB entry base address */ struct cfg_hostres *resources = NULL; u32 temp; u32 ul_dsp_clk_rate; @@ -305,12 +399,12 @@ static int bridge_brd_start(struct bridge_dev_context *dev_ctxt, ul_shm_base_virt *= DSPWORDSIZE; DBC_ASSERT(ul_shm_base_virt != 0); /* DSP Virtual address */ - ul_tlb_base_virt = dev_context->sh_s.seg0_da; + ul_tlb_base_virt = dev_context->atlb_entry[0].ul_dsp_va; DBC_ASSERT(ul_tlb_base_virt <= ul_shm_base_virt); ul_shm_offset_virt = ul_shm_base_virt - (ul_tlb_base_virt * DSPWORDSIZE); /* Kernel logical address */ - ul_shm_base = dev_context->sh_s.seg0_va + ul_shm_offset_virt; + ul_shm_base = dev_context->atlb_entry[0].ul_gpp_va + ul_shm_offset_virt; DBC_ASSERT(ul_shm_base != 0); /* 2nd wd is used as sync field */ @@ -345,83 +439,78 @@ static int bridge_brd_start(struct bridge_dev_context *dev_ctxt, OMAP343X_CONTROL_IVA2_BOOTMOD)); } } - if (!status) { + /* Reset and Unreset the RST2, so that BOOTADDR is copied to + * IVA2 SYSC register */ + (*pdata->dsp_prm_rmw_bits)(OMAP3430_RST2_IVA2_MASK, + OMAP3430_RST2_IVA2_MASK, OMAP3430_IVA2_MOD, OMAP2_RM_RSTCTRL); + udelay(100); (*pdata->dsp_prm_rmw_bits)(OMAP3430_RST2_IVA2_MASK, 0, OMAP3430_IVA2_MOD, OMAP2_RM_RSTCTRL); - mmu = dev_context->dsp_mmu; - if (mmu) - dsp_mmu_exit(mmu); - mmu = dsp_mmu_init(); - if (IS_ERR(mmu)) { - dev_err(bridge, "dsp_mmu_init failed!\n"); - dev_context->dsp_mmu = NULL; - status = (int)mmu; - } - } - if (!status) { - dev_context->dsp_mmu = mmu; - sm_sg = &dev_context->sh_s; - sg0_da = iommu_kmap(mmu, sm_sg->seg0_da, sm_sg->seg0_pa, - sm_sg->seg0_size, IOVMF_ENDIAN_LITTLE | IOVMF_ELSZ_32); - if (IS_ERR_VALUE(sg0_da)) { - status = (int)sg0_da; - sg0_da = 0; - } - } - if (!status) { - sg1_da = iommu_kmap(mmu, sm_sg->seg1_da, sm_sg->seg1_pa, - sm_sg->seg1_size, IOVMF_ENDIAN_LITTLE | IOVMF_ELSZ_32); - if (IS_ERR_VALUE(sg1_da)) { - status = (int)sg1_da; - sg1_da = 0; - } - } - if (!status) { - u32 da; - for (tlb_i = 0; tlb_i < BRDIOCTL_NUMOFMMUTLB; tlb_i++) { - if (!tlb[tlb_i].ul_gpp_pa) + udelay(100); + + /* Disbale the DSP MMU */ + hw_mmu_disable(resources->dw_dmmu_base); + /* Disable TWL */ + hw_mmu_twl_disable(resources->dw_dmmu_base); + + /* Only make TLB entry if both addresses are non-zero */ + for (entry_ndx = 0; entry_ndx < BRDIOCTL_NUMOFMMUTLB; + entry_ndx++) { + struct bridge_ioctl_extproc *e = &dev_context->atlb_entry[entry_ndx]; + struct hw_mmu_map_attrs_t map_attrs = { + .endianism = e->endianism, + .element_size = e->elem_size, + .mixed_size = e->mixed_mode, + }; + + if (!e->ul_gpp_pa || !e->ul_dsp_va) continue; - dev_dbg(bridge, "IOMMU %d GppPa: 0x%x DspVa 0x%x Size" - " 0x%x\n", tlb_i, tlb[tlb_i].ul_gpp_pa, - tlb[tlb_i].ul_dsp_va, tlb[tlb_i].ul_size); - - da = iommu_kmap(mmu, tlb[tlb_i].ul_dsp_va, - tlb[tlb_i].ul_gpp_pa, PAGE_SIZE, - IOVMF_ENDIAN_LITTLE | IOVMF_ELSZ_32); - if (IS_ERR_VALUE(da)) { - status = (int)da; - break; - } - } - } - if (!status) { - u32 da; - l4_i = 0; - while (l4_peripheral_table[l4_i].phys_addr) { - da = iommu_kmap(mmu, l4_peripheral_table[l4_i]. - dsp_virt_addr, l4_peripheral_table[l4_i]. - phys_addr, PAGE_SIZE, - IOVMF_ENDIAN_LITTLE | IOVMF_ELSZ_32); - if (IS_ERR_VALUE(da)) { - status = (int)da; - break; - } - l4_i++; + dev_dbg(bridge, + "MMU %d, pa: 0x%x, va: 0x%x, size: 0x%x", + itmp_entry_ndx, + e->ul_gpp_pa, + e->ul_dsp_va, + e->ul_size); + + hw_mmu_tlb_add(dev_context->dw_dsp_mmu_base, + e->ul_gpp_pa, + e->ul_dsp_va, + e->ul_size, + itmp_entry_ndx, + &map_attrs, 1, 1); + + itmp_entry_ndx++; } } /* Lock the above TLB entries and get the BIOS and load monitor timer * information */ if (!status) { + hw_mmu_num_locked_set(resources->dw_dmmu_base, itmp_entry_ndx); + hw_mmu_victim_num_set(resources->dw_dmmu_base, itmp_entry_ndx); + hw_mmu_ttb_set(resources->dw_dmmu_base, + dev_context->pt_attrs->l1_base_pa); + hw_mmu_twl_enable(resources->dw_dmmu_base); + /* Enable the SmartIdle and AutoIdle bit for MMU_SYSCONFIG */ + + temp = __raw_readl((resources->dw_dmmu_base) + 0x10); + temp = (temp & 0xFFFFFFEF) | 0x11; + __raw_writel(temp, (resources->dw_dmmu_base) + 0x10); + + /* Let the DSP MMU run */ + hw_mmu_enable(resources->dw_dmmu_base); + /* Enable the BIOS clock */ (void)dev_get_symbol(dev_context->hdev_obj, BRIDGEINIT_BIOSGPTIMER, &ul_bios_gp_timer); (void)dev_get_symbol(dev_context->hdev_obj, BRIDGEINIT_LOADMON_GPTIMER, &ul_load_monitor_timer); + } + if (!status) { if (ul_load_monitor_timer != 0xFFFF) { clk_cmd = (BPWR_ENABLE_CLOCK << MBX_PM_CLK_CMDSHIFT) | ul_load_monitor_timer; @@ -430,7 +519,9 @@ static int bridge_brd_start(struct bridge_dev_context *dev_ctxt, dev_dbg(bridge, "Not able to get the symbol for Load " "Monitor Timer\n"); } + } + if (!status) { if (ul_bios_gp_timer != 0xFFFF) { clk_cmd = (BPWR_ENABLE_CLOCK << MBX_PM_CLK_CMDSHIFT) | ul_bios_gp_timer; @@ -439,7 +530,9 @@ static int bridge_brd_start(struct bridge_dev_context *dev_ctxt, dev_dbg(bridge, "Not able to get the symbol for BIOS Timer\n"); } + } + if (!status) { /* Set the DSP clock rate */ (void)dev_get_symbol(dev_context->hdev_obj, "_BRIDGEINIT_DSP_FREQ", &ul_dsp_clk_addr); @@ -492,6 +585,9 @@ static int bridge_brd_start(struct bridge_dev_context *dev_ctxt, /* Let DSP go */ dev_dbg(bridge, "%s Unreset\n", __func__); + /* Enable DSP MMU Interrupts */ + hw_mmu_event_enable(resources->dw_dmmu_base, + HW_MMU_ALL_INTERRUPTS); /* release the RST1, DSP starts executing now .. */ (*pdata->dsp_prm_rmw_bits)(OMAP3430_RST1_IVA2_MASK, 0, OMAP3430_IVA2_MOD, OMAP2_RM_RSTCTRL); @@ -521,23 +617,11 @@ static int bridge_brd_start(struct bridge_dev_context *dev_ctxt, /* update board state */ dev_context->dw_brd_state = BRD_RUNNING; - return 0; + /* (void)chnlsm_enable_interrupt(dev_context); */ } else { dev_context->dw_brd_state = BRD_UNKNOWN; } } - - while (tlb_i--) { - if (!tlb[tlb_i].ul_gpp_pa) - continue; - iommu_kunmap(mmu, tlb[tlb_i].ul_gpp_va); - } - while (l4_i--) - iommu_kunmap(mmu, l4_peripheral_table[l4_i].dsp_virt_addr); - if (sg0_da) - iommu_kunmap(mmu, sg0_da); - if (sg1_da) - iommu_kunmap(mmu, sg1_da); return status; } @@ -553,9 +637,8 @@ static int bridge_brd_stop(struct bridge_dev_context *dev_ctxt) { int status = 0; struct bridge_dev_context *dev_context = dev_ctxt; + struct pg_table_attrs *pt_attrs; u32 dsp_pwr_state; - int i; - struct bridge_ioctl_extproc *tlb = dev_context->atlb_entry; struct omap_dsp_platform_data *pdata = omap_dspbridge_dev->dev.platform_data; @@ -591,37 +674,23 @@ static int bridge_brd_stop(struct bridge_dev_context *dev_ctxt) dsp_wdt_enable(false); - /* Reset DSP */ - (*pdata->dsp_prm_rmw_bits)(OMAP3430_RST1_IVA2_MASK, - OMAP3430_RST1_IVA2_MASK, OMAP3430_IVA2_MOD, OMAP2_RM_RSTCTRL); - + /* This is a good place to clear the MMU page tables as well */ + if (dev_context->pt_attrs) { + pt_attrs = dev_context->pt_attrs; + memset((u8 *) pt_attrs->l1_base_va, 0x00, pt_attrs->l1_size); + memset((u8 *) pt_attrs->l2_base_va, 0x00, pt_attrs->l2_size); + memset((u8 *) pt_attrs->pg_info, 0x00, + (pt_attrs->l2_num_pages * sizeof(struct page_info))); + } /* Disable the mailbox interrupts */ if (dev_context->mbox) { omap_mbox_disable_irq(dev_context->mbox, IRQ_RX); omap_mbox_put(dev_context->mbox); dev_context->mbox = NULL; } - if (dev_context->dsp_mmu) { - pr_err("Proc stop mmu if statement\n"); - for (i = 0; i < BRDIOCTL_NUMOFMMUTLB; i++) { - if (!tlb[i].ul_gpp_pa) - continue; - iommu_kunmap(dev_context->dsp_mmu, tlb[i].ul_gpp_va); - } - i = 0; - while (l4_peripheral_table[i].phys_addr) { - iommu_kunmap(dev_context->dsp_mmu, - l4_peripheral_table[i].dsp_virt_addr); - i++; - } - iommu_kunmap(dev_context->dsp_mmu, dev_context->sh_s.seg0_da); - iommu_kunmap(dev_context->dsp_mmu, dev_context->sh_s.seg1_da); - dsp_mmu_exit(dev_context->dsp_mmu); - dev_context->dsp_mmu = NULL; - } - /* Reset IVA IOMMU*/ - (*pdata->dsp_prm_rmw_bits)(OMAP3430_RST2_IVA2_MASK, - OMAP3430_RST2_IVA2_MASK, OMAP3430_IVA2_MOD, OMAP2_RM_RSTCTRL); + /* Reset IVA2 clocks*/ + (*pdata->dsp_prm_write)(OMAP3430_RST1_IVA2_MASK | OMAP3430_RST2_IVA2_MASK | + OMAP3430_RST3_IVA2_MASK, OMAP3430_IVA2_MOD, OMAP2_RM_RSTCTRL); dsp_clock_disable_all(dev_context->dsp_per_clks); dsp_clk_disable(DSP_CLK_IVA2); @@ -681,6 +750,10 @@ static int bridge_dev_create(struct bridge_dev_context struct bridge_dev_context *dev_context = NULL; s32 entry_ndx; struct cfg_hostres *resources = config_param; + struct pg_table_attrs *pt_attrs; + u32 pg_tbl_pa; + u32 pg_tbl_va; + u32 align_size; struct drv_data *drv_datap = dev_get_drvdata(bridge); /* Allocate and initialize a data structure to contain the bridge driver @@ -711,8 +784,97 @@ static int bridge_dev_create(struct bridge_dev_context if (!dev_context->dw_dsp_base_addr) status = -EPERM; + pt_attrs = kzalloc(sizeof(struct pg_table_attrs), GFP_KERNEL); + if (pt_attrs != NULL) { + /* Assuming that we use only DSP's memory map + * until 0x4000:0000 , we would need only 1024 + * L1 enties i.e L1 size = 4K */ + pt_attrs->l1_size = 0x1000; + align_size = pt_attrs->l1_size; + /* Align sizes are expected to be power of 2 */ + /* we like to get aligned on L1 table size */ + pg_tbl_va = (u32) mem_alloc_phys_mem(pt_attrs->l1_size, + align_size, &pg_tbl_pa); + + /* Check if the PA is aligned for us */ + if ((pg_tbl_pa) & (align_size - 1)) { + /* PA not aligned to page table size , + * try with more allocation and align */ + mem_free_phys_mem((void *)pg_tbl_va, pg_tbl_pa, + pt_attrs->l1_size); + /* we like to get aligned on L1 table size */ + pg_tbl_va = + (u32) mem_alloc_phys_mem((pt_attrs->l1_size) * 2, + align_size, &pg_tbl_pa); + /* We should be able to get aligned table now */ + pt_attrs->l1_tbl_alloc_pa = pg_tbl_pa; + pt_attrs->l1_tbl_alloc_va = pg_tbl_va; + pt_attrs->l1_tbl_alloc_sz = pt_attrs->l1_size * 2; + /* Align the PA to the next 'align' boundary */ + pt_attrs->l1_base_pa = + ((pg_tbl_pa) + + (align_size - 1)) & (~(align_size - 1)); + pt_attrs->l1_base_va = + pg_tbl_va + (pt_attrs->l1_base_pa - pg_tbl_pa); + } else { + /* We got aligned PA, cool */ + pt_attrs->l1_tbl_alloc_pa = pg_tbl_pa; + pt_attrs->l1_tbl_alloc_va = pg_tbl_va; + pt_attrs->l1_tbl_alloc_sz = pt_attrs->l1_size; + pt_attrs->l1_base_pa = pg_tbl_pa; + pt_attrs->l1_base_va = pg_tbl_va; + } + if (pt_attrs->l1_base_va) + memset((u8 *) pt_attrs->l1_base_va, 0x00, + pt_attrs->l1_size); + + /* number of L2 page tables = DMM pool used + SHMMEM +EXTMEM + + * L4 pages */ + pt_attrs->l2_num_pages = ((DMMPOOLSIZE >> 20) + 6); + pt_attrs->l2_size = HW_MMU_COARSE_PAGE_SIZE * + pt_attrs->l2_num_pages; + align_size = 4; /* Make it u32 aligned */ + /* we like to get aligned on L1 table size */ + pg_tbl_va = (u32) mem_alloc_phys_mem(pt_attrs->l2_size, + align_size, &pg_tbl_pa); + pt_attrs->l2_tbl_alloc_pa = pg_tbl_pa; + pt_attrs->l2_tbl_alloc_va = pg_tbl_va; + pt_attrs->l2_tbl_alloc_sz = pt_attrs->l2_size; + pt_attrs->l2_base_pa = pg_tbl_pa; + pt_attrs->l2_base_va = pg_tbl_va; + + if (pt_attrs->l2_base_va) + memset((u8 *) pt_attrs->l2_base_va, 0x00, + pt_attrs->l2_size); + + pt_attrs->pg_info = kzalloc(pt_attrs->l2_num_pages * + sizeof(struct page_info), GFP_KERNEL); + dev_dbg(bridge, + "L1 pa %x, va %x, size %x\n L2 pa %x, va " + "%x, size %x\n", pt_attrs->l1_base_pa, + pt_attrs->l1_base_va, pt_attrs->l1_size, + pt_attrs->l2_base_pa, pt_attrs->l2_base_va, + pt_attrs->l2_size); + dev_dbg(bridge, "pt_attrs %p L2 NumPages %x pg_info %p\n", + pt_attrs, pt_attrs->l2_num_pages, pt_attrs->pg_info); + } + if ((pt_attrs != NULL) && (pt_attrs->l1_base_va != 0) && + (pt_attrs->l2_base_va != 0) && (pt_attrs->pg_info != NULL)) + dev_context->pt_attrs = pt_attrs; + else + status = -ENOMEM; + if (!status) { + spin_lock_init(&pt_attrs->pg_lock); dev_context->tc_word_swap_on = drv_datap->tc_wordswapon; + + /* Set the Clock Divisor for the DSP module */ + udelay(5); + /* MMU address is obtained from the host + * resources struct */ + dev_context->dw_dsp_mmu_base = resources->dw_dmmu_base; + } + if (!status) { dev_context->hdev_obj = hdev_obj; /* Store current board state. */ dev_context->dw_brd_state = BRD_UNKNOWN; @@ -722,6 +884,23 @@ static int bridge_dev_create(struct bridge_dev_context /* Return ptr to our device state to the DSP API for storage */ *dev_cntxt = dev_context; } else { + if (pt_attrs != NULL) { + kfree(pt_attrs->pg_info); + + if (pt_attrs->l2_tbl_alloc_va) { + mem_free_phys_mem((void *) + pt_attrs->l2_tbl_alloc_va, + pt_attrs->l2_tbl_alloc_pa, + pt_attrs->l2_tbl_alloc_sz); + } + if (pt_attrs->l1_tbl_alloc_va) { + mem_free_phys_mem((void *) + pt_attrs->l1_tbl_alloc_va, + pt_attrs->l1_tbl_alloc_pa, + pt_attrs->l1_tbl_alloc_sz); + } + } + kfree(pt_attrs); kfree(dev_context); } func_end: @@ -789,6 +968,7 @@ static int bridge_dev_ctrl(struct bridge_dev_context *dev_context, */ static int bridge_dev_destroy(struct bridge_dev_context *dev_ctxt) { + struct pg_table_attrs *pt_attrs; int status = 0; struct bridge_dev_context *dev_context = (struct bridge_dev_context *) dev_ctxt; @@ -802,6 +982,23 @@ static int bridge_dev_destroy(struct bridge_dev_context *dev_ctxt) /* first put the device to stop state */ bridge_brd_stop(dev_context); + if (dev_context->pt_attrs) { + pt_attrs = dev_context->pt_attrs; + kfree(pt_attrs->pg_info); + + if (pt_attrs->l2_tbl_alloc_va) { + mem_free_phys_mem((void *)pt_attrs->l2_tbl_alloc_va, + pt_attrs->l2_tbl_alloc_pa, + pt_attrs->l2_tbl_alloc_sz); + } + if (pt_attrs->l1_tbl_alloc_va) { + mem_free_phys_mem((void *)pt_attrs->l1_tbl_alloc_va, + pt_attrs->l1_tbl_alloc_pa, + pt_attrs->l1_tbl_alloc_sz); + } + kfree(pt_attrs); + + } if (dev_context->resources) { host_res = dev_context->resources; @@ -832,6 +1029,8 @@ static int bridge_dev_destroy(struct bridge_dev_context *dev_ctxt) iounmap((void *)host_res->dw_mem_base[3]); if (host_res->dw_mem_base[4]) iounmap((void *)host_res->dw_mem_base[4]); + if (host_res->dw_dmmu_base) + iounmap(host_res->dw_dmmu_base); if (host_res->dw_per_base) iounmap(host_res->dw_per_base); if (host_res->dw_per_pm_base) @@ -845,6 +1044,7 @@ static int bridge_dev_destroy(struct bridge_dev_context *dev_ctxt) host_res->dw_mem_base[2] = (u32) NULL; host_res->dw_mem_base[3] = (u32) NULL; host_res->dw_mem_base[4] = (u32) NULL; + host_res->dw_dmmu_base = NULL; host_res->dw_sys_ctrl_base = NULL; kfree(host_res); @@ -928,6 +1128,673 @@ static int bridge_brd_mem_write(struct bridge_dev_context *dev_ctxt, } /* + * ======== bridge_brd_mem_map ======== + * This function maps MPU buffer to the DSP address space. It performs + * linear to physical address translation if required. It translates each + * page since linear addresses can be physically non-contiguous + * All address & size arguments are assumed to be page aligned (in proc.c) + * + * TODO: Disable MMU while updating the page tables (but that'll stall DSP) + */ +static int bridge_brd_mem_map(struct bridge_dev_context *dev_ctxt, + u32 ul_mpu_addr, u32 virt_addr, + u32 ul_num_bytes, u32 ul_map_attr, + struct page **mapped_pages) +{ + u32 attrs; + int status = 0; + struct bridge_dev_context *dev_context = dev_ctxt; + struct hw_mmu_map_attrs_t hw_attrs; + struct vm_area_struct *vma; + struct mm_struct *mm = current->mm; + u32 write = 0; + u32 num_usr_pgs = 0; + struct page *mapped_page, *pg; + s32 pg_num; + u32 va = virt_addr; + struct task_struct *curr_task = current; + u32 pg_i = 0; + u32 mpu_addr, pa; + + dev_dbg(bridge, + "%s hDevCtxt %p, pa %x, va %x, size %x, ul_map_attr %x\n", + __func__, dev_ctxt, ul_mpu_addr, virt_addr, ul_num_bytes, + ul_map_attr); + if (ul_num_bytes == 0) + return -EINVAL; + + if (ul_map_attr & DSP_MAP_DIR_MASK) { + attrs = ul_map_attr; + } else { + /* Assign default attributes */ + attrs = ul_map_attr | (DSP_MAPVIRTUALADDR | DSP_MAPELEMSIZE16); + } + /* Take mapping properties */ + if (attrs & DSP_MAPBIGENDIAN) + hw_attrs.endianism = HW_BIG_ENDIAN; + else + hw_attrs.endianism = HW_LITTLE_ENDIAN; + + hw_attrs.mixed_size = (enum hw_mmu_mixed_size_t) + ((attrs & DSP_MAPMIXEDELEMSIZE) >> 2); + /* Ignore element_size if mixed_size is enabled */ + if (hw_attrs.mixed_size == 0) { + if (attrs & DSP_MAPELEMSIZE8) { + /* Size is 8 bit */ + hw_attrs.element_size = HW_ELEM_SIZE8BIT; + } else if (attrs & DSP_MAPELEMSIZE16) { + /* Size is 16 bit */ + hw_attrs.element_size = HW_ELEM_SIZE16BIT; + } else if (attrs & DSP_MAPELEMSIZE32) { + /* Size is 32 bit */ + hw_attrs.element_size = HW_ELEM_SIZE32BIT; + } else if (attrs & DSP_MAPELEMSIZE64) { + /* Size is 64 bit */ + hw_attrs.element_size = HW_ELEM_SIZE64BIT; + } else { + /* + * Mixedsize isn't enabled, so size can't be + * zero here + */ + return -EINVAL; + } + } + if (attrs & DSP_MAPDONOTLOCK) + hw_attrs.donotlockmpupage = 1; + else + hw_attrs.donotlockmpupage = 0; + + if (attrs & DSP_MAPVMALLOCADDR) { + return mem_map_vmalloc(dev_ctxt, ul_mpu_addr, virt_addr, + ul_num_bytes, &hw_attrs); + } + /* + * Do OS-specific user-va to pa translation. + * Combine physically contiguous regions to reduce TLBs. + * Pass the translated pa to pte_update. + */ + if ((attrs & DSP_MAPPHYSICALADDR)) { + status = pte_update(dev_context, ul_mpu_addr, virt_addr, + ul_num_bytes, &hw_attrs); + goto func_cont; + } + + /* + * Important Note: ul_mpu_addr is mapped from user application process + * to current process - it must lie completely within the current + * virtual memory address space in order to be of use to us here! + */ + down_read(&mm->mmap_sem); + vma = find_vma(mm, ul_mpu_addr); + if (vma) + dev_dbg(bridge, + "VMAfor UserBuf: ul_mpu_addr=%x, ul_num_bytes=%x, " + "vm_start=%lx, vm_end=%lx, vm_flags=%lx\n", ul_mpu_addr, + ul_num_bytes, vma->vm_start, vma->vm_end, + vma->vm_flags); + + /* + * It is observed that under some circumstances, the user buffer is + * spread across several VMAs. So loop through and check if the entire + * user buffer is covered + */ + while ((vma) && (ul_mpu_addr + ul_num_bytes > vma->vm_end)) { + /* jump to the next VMA region */ + vma = find_vma(mm, vma->vm_end + 1); + dev_dbg(bridge, + "VMA for UserBuf ul_mpu_addr=%x ul_num_bytes=%x, " + "vm_start=%lx, vm_end=%lx, vm_flags=%lx\n", ul_mpu_addr, + ul_num_bytes, vma->vm_start, vma->vm_end, + vma->vm_flags); + } + if (!vma) { + pr_err("%s: Failed to get VMA region for 0x%x (%d)\n", + __func__, ul_mpu_addr, ul_num_bytes); + status = -EINVAL; + up_read(&mm->mmap_sem); + goto func_cont; + } + + if (vma->vm_flags & VM_IO) { + num_usr_pgs = ul_num_bytes / PG_SIZE4K; + mpu_addr = ul_mpu_addr; + + /* Get the physical addresses for user buffer */ + for (pg_i = 0; pg_i < num_usr_pgs; pg_i++) { + pa = user_va2_pa(mm, mpu_addr); + if (!pa) { + status = -EPERM; + pr_err("DSPBRIDGE: VM_IO mapping physical" + "address is invalid\n"); + break; + } + if (pfn_valid(__phys_to_pfn(pa))) { + pg = PHYS_TO_PAGE(pa); + get_page(pg); + if (page_count(pg) < 1) { + pr_err("Bad page in VM_IO buffer\n"); + bad_page_dump(pa, pg); + } + } + status = pte_set(dev_context->pt_attrs, pa, + va, HW_PAGE_SIZE4KB, &hw_attrs); + if (status) + break; + + va += HW_PAGE_SIZE4KB; + mpu_addr += HW_PAGE_SIZE4KB; + pa += HW_PAGE_SIZE4KB; + } + } else { + num_usr_pgs = ul_num_bytes / PG_SIZE4K; + if (vma->vm_flags & (VM_WRITE | VM_MAYWRITE)) + write = 1; + + for (pg_i = 0; pg_i < num_usr_pgs; pg_i++) { + pg_num = get_user_pages(curr_task, mm, ul_mpu_addr, 1, + write, 1, &mapped_page, NULL); + if (pg_num > 0) { + if (page_count(mapped_page) < 1) { + pr_err("Bad page count after doing" + "get_user_pages on" + "user buffer\n"); + bad_page_dump(page_to_phys(mapped_page), + mapped_page); + } + status = pte_set(dev_context->pt_attrs, + page_to_phys(mapped_page), va, + HW_PAGE_SIZE4KB, &hw_attrs); + if (status) + break; + + if (mapped_pages) + mapped_pages[pg_i] = mapped_page; + + va += HW_PAGE_SIZE4KB; + ul_mpu_addr += HW_PAGE_SIZE4KB; + } else { + pr_err("DSPBRIDGE: get_user_pages FAILED," + "MPU addr = 0x%x," + "vma->vm_flags = 0x%lx," + "get_user_pages Err" + "Value = %d, Buffer" + "size=0x%x\n", ul_mpu_addr, + vma->vm_flags, pg_num, ul_num_bytes); + status = -EPERM; + break; + } + } + } + up_read(&mm->mmap_sem); +func_cont: + if (status) { + /* + * Roll out the mapped pages incase it failed in middle of + * mapping + */ + if (pg_i) { + bridge_brd_mem_un_map(dev_context, virt_addr, + (pg_i * PG_SIZE4K)); + } + status = -EPERM; + } + /* + * In any case, flush the TLB + * This is called from here instead from pte_update to avoid unnecessary + * repetition while mapping non-contiguous physical regions of a virtual + * region + */ + flush_all(dev_context); + dev_dbg(bridge, "%s status %x\n", __func__, status); + return status; +} + +/* + * ======== bridge_brd_mem_un_map ======== + * Invalidate the PTEs for the DSP VA block to be unmapped. + * + * PTEs of a mapped memory block are contiguous in any page table + * So, instead of looking up the PTE address for every 4K block, + * we clear consecutive PTEs until we unmap all the bytes + */ +static int bridge_brd_mem_un_map(struct bridge_dev_context *dev_ctxt, + u32 virt_addr, u32 ul_num_bytes) +{ + u32 l1_base_va; + u32 l2_base_va; + u32 l2_base_pa; + u32 l2_page_num; + u32 pte_val; + u32 pte_size; + u32 pte_count; + u32 pte_addr_l1; + u32 pte_addr_l2 = 0; + u32 rem_bytes; + u32 rem_bytes_l2; + u32 va_curr; + struct page *pg = NULL; + int status = 0; + struct bridge_dev_context *dev_context = dev_ctxt; + struct pg_table_attrs *pt = dev_context->pt_attrs; + u32 temp; + u32 paddr; + u32 numof4k_pages = 0; + + va_curr = virt_addr; + rem_bytes = ul_num_bytes; + rem_bytes_l2 = 0; + l1_base_va = pt->l1_base_va; + pte_addr_l1 = hw_mmu_pte_addr_l1(l1_base_va, va_curr); + dev_dbg(bridge, "%s dev_ctxt %p, va %x, NumBytes %x l1_base_va %x, " + "pte_addr_l1 %x\n", __func__, dev_ctxt, virt_addr, + ul_num_bytes, l1_base_va, pte_addr_l1); + + while (rem_bytes && !status) { + u32 va_curr_orig = va_curr; + /* Find whether the L1 PTE points to a valid L2 PT */ + pte_addr_l1 = hw_mmu_pte_addr_l1(l1_base_va, va_curr); + pte_val = *(u32 *) pte_addr_l1; + pte_size = hw_mmu_pte_size_l1(pte_val); + + if (pte_size != HW_MMU_COARSE_PAGE_SIZE) + goto skip_coarse_page; + + /* + * Get the L2 PA from the L1 PTE, and find + * corresponding L2 VA + */ + l2_base_pa = hw_mmu_pte_coarse_l1(pte_val); + l2_base_va = l2_base_pa - pt->l2_base_pa + pt->l2_base_va; + l2_page_num = + (l2_base_pa - pt->l2_base_pa) / HW_MMU_COARSE_PAGE_SIZE; + /* + * Find the L2 PTE address from which we will start + * clearing, the number of PTEs to be cleared on this + * page, and the size of VA space that needs to be + * cleared on this L2 page + */ + pte_addr_l2 = hw_mmu_pte_addr_l2(l2_base_va, va_curr); + pte_count = pte_addr_l2 & (HW_MMU_COARSE_PAGE_SIZE - 1); + pte_count = (HW_MMU_COARSE_PAGE_SIZE - pte_count) / sizeof(u32); + if (rem_bytes < (pte_count * PG_SIZE4K)) + pte_count = rem_bytes / PG_SIZE4K; + rem_bytes_l2 = pte_count * PG_SIZE4K; + + /* + * Unmap the VA space on this L2 PT. A quicker way + * would be to clear pte_count entries starting from + * pte_addr_l2. However, below code checks that we don't + * clear invalid entries or less than 64KB for a 64KB + * entry. Similar checking is done for L1 PTEs too + * below + */ + while (rem_bytes_l2 && !status) { + pte_val = *(u32 *) pte_addr_l2; + pte_size = hw_mmu_pte_size_l2(pte_val); + /* va_curr aligned to pte_size? */ + if (pte_size == 0 || rem_bytes_l2 < pte_size || + va_curr & (pte_size - 1)) { + status = -EPERM; + break; + } + + /* Collect Physical addresses from VA */ + paddr = (pte_val & ~(pte_size - 1)); + if (pte_size == HW_PAGE_SIZE64KB) + numof4k_pages = 16; + else + numof4k_pages = 1; + temp = 0; + while (temp++ < numof4k_pages) { + if (!pfn_valid(__phys_to_pfn(paddr))) { + paddr += HW_PAGE_SIZE4KB; + continue; + } + pg = PHYS_TO_PAGE(paddr); + if (page_count(pg) < 1) { + pr_info("DSPBRIDGE: UNMAP function: " + "COUNT 0 FOR PA 0x%x, size = " + "0x%x\n", paddr, ul_num_bytes); + bad_page_dump(paddr, pg); + } else { + set_page_dirty(pg); + page_cache_release(pg); + } + paddr += HW_PAGE_SIZE4KB; + } + if (hw_mmu_pte_clear(pte_addr_l2, va_curr, pte_size)) { + status = -EPERM; + goto EXIT_LOOP; + } + + status = 0; + rem_bytes_l2 -= pte_size; + va_curr += pte_size; + pte_addr_l2 += (pte_size >> 12) * sizeof(u32); + } + spin_lock(&pt->pg_lock); + if (rem_bytes_l2 == 0) { + pt->pg_info[l2_page_num].num_entries -= pte_count; + if (pt->pg_info[l2_page_num].num_entries == 0) { + /* + * Clear the L1 PTE pointing to the L2 PT + */ + if (!hw_mmu_pte_clear(l1_base_va, va_curr_orig, + HW_MMU_COARSE_PAGE_SIZE)) + status = 0; + else { + status = -EPERM; + spin_unlock(&pt->pg_lock); + goto EXIT_LOOP; + } + } + rem_bytes -= pte_count * PG_SIZE4K; + } else + status = -EPERM; + + spin_unlock(&pt->pg_lock); + continue; +skip_coarse_page: + /* va_curr aligned to pte_size? */ + /* pte_size = 1 MB or 16 MB */ + if (pte_size == 0 || rem_bytes < pte_size || + va_curr & (pte_size - 1)) { + status = -EPERM; + break; + } + + if (pte_size == HW_PAGE_SIZE1MB) + numof4k_pages = 256; + else + numof4k_pages = 4096; + temp = 0; + /* Collect Physical addresses from VA */ + paddr = (pte_val & ~(pte_size - 1)); + while (temp++ < numof4k_pages) { + if (pfn_valid(__phys_to_pfn(paddr))) { + pg = PHYS_TO_PAGE(paddr); + if (page_count(pg) < 1) { + pr_info("DSPBRIDGE: UNMAP function: " + "COUNT 0 FOR PA 0x%x, size = " + "0x%x\n", paddr, ul_num_bytes); + bad_page_dump(paddr, pg); + } else { + set_page_dirty(pg); + page_cache_release(pg); + } + } + paddr += HW_PAGE_SIZE4KB; + } + if (!hw_mmu_pte_clear(l1_base_va, va_curr, pte_size)) { + status = 0; + rem_bytes -= pte_size; + va_curr += pte_size; + } else { + status = -EPERM; + goto EXIT_LOOP; + } + } + /* + * It is better to flush the TLB here, so that any stale old entries + * get flushed + */ +EXIT_LOOP: + flush_all(dev_context); + dev_dbg(bridge, + "%s: va_curr %x, pte_addr_l1 %x pte_addr_l2 %x rem_bytes %x," + " rem_bytes_l2 %x status %x\n", __func__, va_curr, pte_addr_l1, + pte_addr_l2, rem_bytes, rem_bytes_l2, status); + return status; +} + +/* + * ======== user_va2_pa ======== + * Purpose: + * This function walks through the page tables to convert a userland + * virtual address to physical address + */ +static u32 user_va2_pa(struct mm_struct *mm, u32 address) +{ + pgd_t *pgd; + pmd_t *pmd; + pte_t *ptep, pte; + + pgd = pgd_offset(mm, address); + if (!(pgd_none(*pgd) || pgd_bad(*pgd))) { + pmd = pmd_offset(pgd, address); + if (!(pmd_none(*pmd) || pmd_bad(*pmd))) { + ptep = pte_offset_map(pmd, address); + if (ptep) { + pte = *ptep; + if (pte_present(pte)) + return pte & PAGE_MASK; + } + } + } + + return 0; +} + +/* + * ======== pte_update ======== + * This function calculates the optimum page-aligned addresses and sizes + * Caller must pass page-aligned values + */ +static int pte_update(struct bridge_dev_context *dev_ctxt, u32 pa, + u32 va, u32 size, + struct hw_mmu_map_attrs_t *map_attrs) +{ + u32 i; + u32 all_bits; + u32 pa_curr = pa; + u32 va_curr = va; + u32 num_bytes = size; + struct bridge_dev_context *dev_context = dev_ctxt; + int status = 0; + u32 page_size[] = { HW_PAGE_SIZE16MB, HW_PAGE_SIZE1MB, + HW_PAGE_SIZE64KB, HW_PAGE_SIZE4KB + }; + + while (num_bytes && !status) { + /* To find the max. page size with which both PA & VA are + * aligned */ + all_bits = pa_curr | va_curr; + + for (i = 0; i < 4; i++) { + if ((num_bytes >= page_size[i]) && ((all_bits & + (page_size[i] - + 1)) == 0)) { + status = + pte_set(dev_context->pt_attrs, pa_curr, + va_curr, page_size[i], map_attrs); + pa_curr += page_size[i]; + va_curr += page_size[i]; + num_bytes -= page_size[i]; + /* Don't try smaller sizes. Hopefully we have + * reached an address aligned to a bigger page + * size */ + break; + } + } + } + + return status; +} + +/* + * ======== pte_set ======== + * This function calculates PTE address (MPU virtual) to be updated + * It also manages the L2 page tables + */ +static int pte_set(struct pg_table_attrs *pt, u32 pa, u32 va, + u32 size, struct hw_mmu_map_attrs_t *attrs) +{ + u32 i; + u32 pte_val; + u32 pte_addr_l1; + u32 pte_size; + /* Base address of the PT that will be updated */ + u32 pg_tbl_va; + u32 l1_base_va; + /* Compiler warns that the next three variables might be used + * uninitialized in this function. Doesn't seem so. Working around, + * anyways. */ + u32 l2_base_va = 0; + u32 l2_base_pa = 0; + u32 l2_page_num = 0; + int status = 0; + + l1_base_va = pt->l1_base_va; + pg_tbl_va = l1_base_va; + if ((size == HW_PAGE_SIZE64KB) || (size == HW_PAGE_SIZE4KB)) { + /* Find whether the L1 PTE points to a valid L2 PT */ + pte_addr_l1 = hw_mmu_pte_addr_l1(l1_base_va, va); + if (pte_addr_l1 <= (pt->l1_base_va + pt->l1_size)) { + pte_val = *(u32 *) pte_addr_l1; + pte_size = hw_mmu_pte_size_l1(pte_val); + } else { + return -EPERM; + } + spin_lock(&pt->pg_lock); + if (pte_size == HW_MMU_COARSE_PAGE_SIZE) { + /* Get the L2 PA from the L1 PTE, and find + * corresponding L2 VA */ + l2_base_pa = hw_mmu_pte_coarse_l1(pte_val); + l2_base_va = + l2_base_pa - pt->l2_base_pa + pt->l2_base_va; + l2_page_num = + (l2_base_pa - + pt->l2_base_pa) / HW_MMU_COARSE_PAGE_SIZE; + } else if (pte_size == 0) { + /* L1 PTE is invalid. Allocate a L2 PT and + * point the L1 PTE to it */ + /* Find a free L2 PT. */ + for (i = 0; (i < pt->l2_num_pages) && + (pt->pg_info[i].num_entries != 0); i++) + ;; + if (i < pt->l2_num_pages) { + l2_page_num = i; + l2_base_pa = pt->l2_base_pa + (l2_page_num * + HW_MMU_COARSE_PAGE_SIZE); + l2_base_va = pt->l2_base_va + (l2_page_num * + HW_MMU_COARSE_PAGE_SIZE); + /* Endianness attributes are ignored for + * HW_MMU_COARSE_PAGE_SIZE */ + status = + hw_mmu_pte_set(l1_base_va, l2_base_pa, va, + HW_MMU_COARSE_PAGE_SIZE, + attrs); + } else { + status = -ENOMEM; + } + } else { + /* Found valid L1 PTE of another size. + * Should not overwrite it. */ + status = -EPERM; + } + if (!status) { + pg_tbl_va = l2_base_va; + if (size == HW_PAGE_SIZE64KB) + pt->pg_info[l2_page_num].num_entries += 16; + else + pt->pg_info[l2_page_num].num_entries++; + dev_dbg(bridge, "PTE: L2 BaseVa %x, BasePa %x, PageNum " + "%x, num_entries %x\n", l2_base_va, + l2_base_pa, l2_page_num, + pt->pg_info[l2_page_num].num_entries); + } + spin_unlock(&pt->pg_lock); + } + if (!status) { + dev_dbg(bridge, "PTE: pg_tbl_va %x, pa %x, va %x, size %x\n", + pg_tbl_va, pa, va, size); + dev_dbg(bridge, "PTE: endianism %x, element_size %x, " + "mixed_size %x\n", attrs->endianism, + attrs->element_size, attrs->mixed_size); + status = hw_mmu_pte_set(pg_tbl_va, pa, va, size, attrs); + } + + return status; +} + +/* Memory map kernel VA -- memory allocated with vmalloc */ +static int mem_map_vmalloc(struct bridge_dev_context *dev_context, + u32 ul_mpu_addr, u32 virt_addr, + u32 ul_num_bytes, + struct hw_mmu_map_attrs_t *hw_attrs) +{ + int status = 0; + struct page *page[1]; + u32 i; + u32 pa_curr; + u32 pa_next; + u32 va_curr; + u32 size_curr; + u32 num_pages; + u32 pa; + u32 num_of4k_pages; + u32 temp = 0; + + /* + * Do Kernel va to pa translation. + * Combine physically contiguous regions to reduce TLBs. + * Pass the translated pa to pte_update. + */ + num_pages = ul_num_bytes / PAGE_SIZE; /* PAGE_SIZE = OS page size */ + i = 0; + va_curr = ul_mpu_addr; + page[0] = vmalloc_to_page((void *)va_curr); + pa_next = page_to_phys(page[0]); + while (!status && (i < num_pages)) { + /* + * Reuse pa_next from the previous iteraion to avoid + * an extra va2pa call + */ + pa_curr = pa_next; + size_curr = PAGE_SIZE; + /* + * If the next page is physically contiguous, + * map it with the current one by increasing + * the size of the region to be mapped + */ + while (++i < num_pages) { + page[0] = + vmalloc_to_page((void *)(va_curr + size_curr)); + pa_next = page_to_phys(page[0]); + + if (pa_next == (pa_curr + size_curr)) + size_curr += PAGE_SIZE; + else + break; + + } + if (pa_next == 0) { + status = -ENOMEM; + break; + } + pa = pa_curr; + num_of4k_pages = size_curr / HW_PAGE_SIZE4KB; + while (temp++ < num_of4k_pages) { + get_page(PHYS_TO_PAGE(pa)); + pa += HW_PAGE_SIZE4KB; + } + status = pte_update(dev_context, pa_curr, virt_addr + + (va_curr - ul_mpu_addr), size_curr, + hw_attrs); + va_curr += size_curr; + } + /* + * In any case, flush the TLB + * This is called from here instead from pte_update to avoid unnecessary + * repetition while mapping non-contiguous physical regions of a virtual + * region + */ + flush_all(dev_context); + dev_dbg(bridge, "%s status %x\n", __func__, status); + return status; +} + +/* * ======== wait_for_start ======== * Wait for the singal from DSP that it has started, or time out. */ diff --git a/drivers/staging/tidspbridge/core/tiomap3430_pwr.c b/drivers/staging/tidspbridge/core/tiomap3430_pwr.c index b57a9fd5e757..fb9026e1403c 100644 --- a/drivers/staging/tidspbridge/core/tiomap3430_pwr.c +++ b/drivers/staging/tidspbridge/core/tiomap3430_pwr.c @@ -31,6 +31,10 @@ #include <dspbridge/dev.h> #include <dspbridge/iodefs.h> +/* ------------------------------------ Hardware Abstraction Layer */ +#include <hw_defs.h> +#include <hw_mmu.h> + #include <dspbridge/pwr_sh.h> /* ----------------------------------- Bridge Driver */ diff --git a/drivers/staging/tidspbridge/core/tiomap_io.c b/drivers/staging/tidspbridge/core/tiomap_io.c index 66dbf02549e4..ba2961049dad 100644 --- a/drivers/staging/tidspbridge/core/tiomap_io.c +++ b/drivers/staging/tidspbridge/core/tiomap_io.c @@ -134,16 +134,17 @@ int read_ext_dsp_data(struct bridge_dev_context *dev_ctxt, if (!status) { ul_tlb_base_virt = - dev_context->sh_s.seg0_da * DSPWORDSIZE; + dev_context->atlb_entry[0].ul_dsp_va * DSPWORDSIZE; DBC_ASSERT(ul_tlb_base_virt <= ul_shm_base_virt); - dw_ext_prog_virt_mem = dev_context->sh_s.seg0_va; + dw_ext_prog_virt_mem = + dev_context->atlb_entry[0].ul_gpp_va; if (!trace_read) { ul_shm_offset_virt = ul_shm_base_virt - ul_tlb_base_virt; ul_shm_offset_virt += PG_ALIGN_HIGH(ul_ext_end - ul_dyn_ext_base + - 1, PAGE_SIZE * 16); + 1, HW_PAGE_SIZE64KB); dw_ext_prog_virt_mem -= ul_shm_offset_virt; dw_ext_prog_virt_mem += (ul_ext_base - ul_dyn_ext_base); @@ -317,9 +318,8 @@ int write_ext_dsp_data(struct bridge_dev_context *dev_context, ret = -EPERM; if (!ret) { - ul_tlb_base_virt = dev_context->sh_s.seg0_da * - DSPWORDSIZE; - + ul_tlb_base_virt = + dev_context->atlb_entry[0].ul_dsp_va * DSPWORDSIZE; DBC_ASSERT(ul_tlb_base_virt <= ul_shm_base_virt); if (symbols_reloaded) { @@ -337,7 +337,7 @@ int write_ext_dsp_data(struct bridge_dev_context *dev_context, ul_shm_base_virt - ul_tlb_base_virt; if (trace_load) { dw_ext_prog_virt_mem = - dev_context->sh_s.seg0_va; + dev_context->atlb_entry[0].ul_gpp_va; } else { dw_ext_prog_virt_mem = host_res->dw_mem_base[1]; dw_ext_prog_virt_mem += @@ -393,6 +393,7 @@ int sm_interrupt_dsp(struct bridge_dev_context *dev_context, u16 mb_val) omap_dspbridge_dev->dev.platform_data; struct cfg_hostres *resources = dev_context->resources; int status = 0; + u32 temp; if (!dev_context->mbox) return 0; @@ -436,7 +437,7 @@ int sm_interrupt_dsp(struct bridge_dev_context *dev_context, u16 mb_val) omap_mbox_restore_ctx(dev_context->mbox); /* Access MMU SYS CONFIG register to generate a short wakeup */ - iommu_read_reg(dev_context->dsp_mmu, MMU_SYSCONFIG); + temp = readl(resources->dw_dmmu_base + 0x10); dev_context->dw_brd_state = BRD_RUNNING; } else if (dev_context->dw_brd_state == BRD_RETENTION) { diff --git a/drivers/staging/tidspbridge/core/ue_deh.c b/drivers/staging/tidspbridge/core/ue_deh.c index e24ea0c73914..3430418190da 100644 --- a/drivers/staging/tidspbridge/core/ue_deh.c +++ b/drivers/staging/tidspbridge/core/ue_deh.c @@ -31,6 +31,57 @@ #include <dspbridge/drv.h> #include <dspbridge/wdt.h> +static u32 fault_addr; + +static void mmu_fault_dpc(unsigned long data) +{ + struct deh_mgr *deh = (void *)data; + + if (!deh) + return; + + bridge_deh_notify(deh, DSP_MMUFAULT, 0); +} + +static irqreturn_t mmu_fault_isr(int irq, void *data) +{ + struct deh_mgr *deh = data; + struct cfg_hostres *resources; + u32 event; + + if (!deh) + return IRQ_HANDLED; + + resources = deh->hbridge_context->resources; + if (!resources) { + dev_dbg(bridge, "%s: Failed to get Host Resources\n", + __func__); + return IRQ_HANDLED; + } + + hw_mmu_event_status(resources->dw_dmmu_base, &event); + if (event == HW_MMU_TRANSLATION_FAULT) { + hw_mmu_fault_addr_read(resources->dw_dmmu_base, &fault_addr); + dev_dbg(bridge, "%s: event=0x%x, fault_addr=0x%x\n", __func__, + event, fault_addr); + /* + * Schedule a DPC directly. In the future, it may be + * necessary to check if DSP MMU fault is intended for + * Bridge. + */ + tasklet_schedule(&deh->dpc_tasklet); + + /* Disable the MMU events, else once we clear it will + * start to raise INTs again */ + hw_mmu_event_disable(resources->dw_dmmu_base, + HW_MMU_TRANSLATION_FAULT); + } else { + hw_mmu_event_disable(resources->dw_dmmu_base, + HW_MMU_ALL_INTERRUPTS); + } + return IRQ_HANDLED; +} + int bridge_deh_create(struct deh_mgr **ret_deh, struct dev_object *hdev_obj) { @@ -58,9 +109,18 @@ int bridge_deh_create(struct deh_mgr **ret_deh, } ntfy_init(deh->ntfy_obj); + /* Create a MMUfault DPC */ + tasklet_init(&deh->dpc_tasklet, mmu_fault_dpc, (u32) deh); + /* Fill in context structure */ deh->hbridge_context = hbridge_context; + /* Install ISR function for DSP MMU fault */ + status = request_irq(INT_DSP_MMU_IRQ, mmu_fault_isr, 0, + "DspBridge\tiommu fault", deh); + if (status < 0) + goto err; + *ret_deh = deh; return 0; @@ -80,6 +140,11 @@ int bridge_deh_destroy(struct deh_mgr *deh) ntfy_delete(deh->ntfy_obj); kfree(deh->ntfy_obj); } + /* Disable DSP MMU fault */ + free_irq(INT_DSP_MMU_IRQ, deh); + + /* Free DPC object */ + tasklet_kill(&deh->dpc_tasklet); /* Deallocate the DEH manager object */ kfree(deh); @@ -101,6 +166,48 @@ int bridge_deh_register_notify(struct deh_mgr *deh, u32 event_mask, return ntfy_unregister(deh->ntfy_obj, hnotification); } +#ifdef CONFIG_TIDSPBRIDGE_BACKTRACE +static void mmu_fault_print_stack(struct bridge_dev_context *dev_context) +{ + struct cfg_hostres *resources; + struct hw_mmu_map_attrs_t map_attrs = { + .endianism = HW_LITTLE_ENDIAN, + .element_size = HW_ELEM_SIZE16BIT, + .mixed_size = HW_MMU_CPUES, + }; + void *dummy_va_addr; + + resources = dev_context->resources; + dummy_va_addr = (void*)__get_free_page(GFP_ATOMIC); + + /* + * Before acking the MMU fault, let's make sure MMU can only + * access entry #0. Then add a new entry so that the DSP OS + * can continue in order to dump the stack. + */ + hw_mmu_twl_disable(resources->dw_dmmu_base); + hw_mmu_tlb_flush_all(resources->dw_dmmu_base); + + hw_mmu_tlb_add(resources->dw_dmmu_base, + virt_to_phys(dummy_va_addr), fault_addr, + HW_PAGE_SIZE4KB, 1, + &map_attrs, HW_SET, HW_SET); + + dsp_clk_enable(DSP_CLK_GPT8); + + dsp_gpt_wait_overflow(DSP_CLK_GPT8, 0xfffffffe); + + /* Clear MMU interrupt */ + hw_mmu_event_ack(resources->dw_dmmu_base, + HW_MMU_TRANSLATION_FAULT); + dump_dsp_stack(dev_context); + dsp_clk_disable(DSP_CLK_GPT8); + + hw_mmu_disable(resources->dw_dmmu_base); + free_page((unsigned long)dummy_va_addr); +} +#endif + static inline const char *event_to_string(int event) { switch (event) { @@ -133,7 +240,13 @@ void bridge_deh_notify(struct deh_mgr *deh, int event, int info) #endif break; case DSP_MMUFAULT: - dev_err(bridge, "%s: %s, addr=0x%x", __func__, str, info); + dev_err(bridge, "%s: %s, addr=0x%x", __func__, + str, fault_addr); +#ifdef CONFIG_TIDSPBRIDGE_BACKTRACE + print_dsp_trace_buffer(dev_context); + dump_dl_modules(dev_context); + mmu_fault_print_stack(dev_context); +#endif break; default: dev_err(bridge, "%s: %s", __func__, str); |