Merge tag 'rdma-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/roland/infiniband

Pull InfiniBand/RDMA changes from Roland Dreier:
 - AF_IB (native IB addressing) for CMA from Sean Hefty
 - new mlx5 driver for Mellanox Connect-IB adapters (including post
   merge request fixes)
 - SRP fixes from Bart Van Assche (including fix to first merge request)
 - qib HW driver updates
 - resurrection of ocrdma HW driver development
 - uverbs conversion to create fds with O_CLOEXEC set
 - other small changes and fixes

* tag 'rdma-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/roland/infiniband: (66 commits)
  mlx5: Return -EFAULT instead of -EPERM
  IB/qib: Log all SDMA errors unconditionally
  IB/qib: Fix module-level leak
  mlx5_core: Adjust hca_cap.uar_page_sz to conform to Connect-IB spec
  IB/srp: Let srp_abort() return FAST_IO_FAIL if TL offline
  IB/uverbs: Use get_unused_fd_flags(O_CLOEXEC) instead of get_unused_fd()
  mlx5_core: Fixes for sparse warnings
  IB/mlx5: Make profile[] static in main.c
  mlx5: Fix parameter type of health_handler_t
  mlx5: Add driver for Mellanox Connect-IB adapters
  IB/core: Add reserved values to enums for low-level driver use
  IB/srp: Bump driver version and release date
  IB/srp: Make HCA completion vector configurable
  IB/srp: Maintain a single connection per I_T nexus
  IB/srp: Fail I/O fast if target offline
  IB/srp: Skip host settle delay
  IB/srp: Avoid skipping srp_reset_host() after a transport error
  IB/srp: Fix remove_one crash due to resource exhaustion
  IB/qib: New transmitter tunning settings for Dell 1.1 backplane
  IB/core: Fix error return code in add_port()
  ...

1 files changed, 128 insertions, 48 deletions

diff --git a/drivers/infiniband/hw/qib/qib_file_ops.c b/drivers/infiniband/hw/qib/qib_file_ops.c
index 9dd0bc89c3aa..b51a51486cb8 100644
--- a/drivers/infiniband/hw/qib/qib_file_ops.c
+++ b/drivers/infiniband/hw/qib/qib_file_ops.c
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2012 Intel Corporation. All rights reserved.
+ * Copyright (c) 2012, 2013 Intel Corporation. All rights reserved.
  * Copyright (c) 2006 - 2012 QLogic Corporation. All rights reserved.
  * Copyright (c) 2003, 2004, 2005, 2006 PathScale, Inc. All rights reserved.
  *
@@ -1155,6 +1155,49 @@ static unsigned int qib_poll(struct file *fp, struct poll_table_struct *pt)
 	return pollflag;
 }
 
+static void assign_ctxt_affinity(struct file *fp, struct qib_devdata *dd)
+{
+	struct qib_filedata *fd = fp->private_data;
+	const unsigned int weight = cpumask_weight(&current->cpus_allowed);
+	const struct cpumask *local_mask = cpumask_of_pcibus(dd->pcidev->bus);
+	int local_cpu;
+
+	/*
+	 * If process has NOT already set it's affinity, select and
+	 * reserve a processor for it on the local NUMA node.
+	 */
+	if ((weight >= qib_cpulist_count) &&
+		(cpumask_weight(local_mask) <= qib_cpulist_count)) {
+		for_each_cpu(local_cpu, local_mask)
+			if (!test_and_set_bit(local_cpu, qib_cpulist)) {
+				fd->rec_cpu_num = local_cpu;
+				return;
+			}
+	}
+
+	/*
+	 * If process has NOT already set it's affinity, select and
+	 * reserve a processor for it, as a rendevous for all
+	 * users of the driver.  If they don't actually later
+	 * set affinity to this cpu, or set it to some other cpu,
+	 * it just means that sooner or later we don't recommend
+	 * a cpu, and let the scheduler do it's best.
+	 */
+	if (weight >= qib_cpulist_count) {
+		int cpu;
+		cpu = find_first_zero_bit(qib_cpulist,
+					  qib_cpulist_count);
+		if (cpu == qib_cpulist_count)
+			qib_dev_err(dd,
+			"no cpus avail for affinity PID %u\n",
+			current->pid);
+		else {
+			__set_bit(cpu, qib_cpulist);
+			fd->rec_cpu_num = cpu;
+		}
+	}
+}
+
 /*
  * Check that userland and driver are compatible for subcontexts.
  */
@@ -1259,12 +1302,20 @@ bail:
 static int setup_ctxt(struct qib_pportdata *ppd, int ctxt,
 		      struct file *fp, const struct qib_user_info *uinfo)
 {
+	struct qib_filedata *fd = fp->private_data;
 	struct qib_devdata *dd = ppd->dd;
 	struct qib_ctxtdata *rcd;
 	void *ptmp = NULL;
 	int ret;
+	int numa_id;
+
+	assign_ctxt_affinity(fp, dd);
 
-	rcd = qib_create_ctxtdata(ppd, ctxt);
+	numa_id = qib_numa_aware ? ((fd->rec_cpu_num != -1) ?
+		cpu_to_node(fd->rec_cpu_num) :
+		numa_node_id()) : dd->assigned_node_id;
+
+	rcd = qib_create_ctxtdata(ppd, ctxt, numa_id);
 
 	/*
 	 * Allocate memory for use in qib_tid_update() at open to
@@ -1296,6 +1347,9 @@ static int setup_ctxt(struct qib_pportdata *ppd, int ctxt,
 	goto bail;
 
 bailerr:
+	if (fd->rec_cpu_num != -1)
+		__clear_bit(fd->rec_cpu_num, qib_cpulist);
+
 	dd->rcd[ctxt] = NULL;
 	kfree(rcd);
 	kfree(ptmp);
@@ -1485,6 +1539,57 @@ static int qib_open(struct inode *in, struct file *fp)
 	return fp->private_data ? 0 : -ENOMEM;
 }
 
+static int find_hca(unsigned int cpu, int *unit)
+{
+	int ret = 0, devmax, npresent, nup, ndev;
+
+	*unit = -1;
+
+	devmax = qib_count_units(&npresent, &nup);
+	if (!npresent) {
+		ret = -ENXIO;
+		goto done;
+	}
+	if (!nup) {
+		ret = -ENETDOWN;
+		goto done;
+	}
+	for (ndev = 0; ndev < devmax; ndev++) {
+		struct qib_devdata *dd = qib_lookup(ndev);
+		if (dd) {
+			if (pcibus_to_node(dd->pcidev->bus) < 0) {
+				ret = -EINVAL;
+				goto done;
+			}
+			if (cpu_to_node(cpu) ==
+				pcibus_to_node(dd->pcidev->bus)) {
+				*unit = ndev;
+				goto done;
+			}
+		}
+	}
+done:
+	return ret;
+}
+
+static int do_qib_user_sdma_queue_create(struct file *fp)
+{
+	struct qib_filedata *fd = fp->private_data;
+	struct qib_ctxtdata *rcd = fd->rcd;
+	struct qib_devdata *dd = rcd->dd;
+
+	if (dd->flags & QIB_HAS_SEND_DMA)
+
+		fd->pq = qib_user_sdma_queue_create(&dd->pcidev->dev,
+						    dd->unit,
+						    rcd->ctxt,
+						    fd->subctxt);
+		if (!fd->pq)
+			return -ENOMEM;
+
+	return 0;
+}
+
 /*
  * Get ctxt early, so can set affinity prior to memory allocation.
  */
@@ -1517,61 +1622,36 @@ static int qib_assign_ctxt(struct file *fp, const struct qib_user_info *uinfo)
 	if (qib_compatible_subctxts(swmajor, swminor) &&
 	    uinfo->spu_subctxt_cnt) {
 		ret = find_shared_ctxt(fp, uinfo);
-		if (ret) {
-			if (ret > 0)
-				ret = 0;
-			goto done_chk_sdma;
+		if (ret > 0) {
+			ret = do_qib_user_sdma_queue_create(fp);
+			if (!ret)
+				assign_ctxt_affinity(fp, (ctxt_fp(fp))->dd);
+			goto done_ok;
 		}
 	}
 
 	i_minor = iminor(file_inode(fp)) - QIB_USER_MINOR_BASE;
 	if (i_minor)
 		ret = find_free_ctxt(i_minor - 1, fp, uinfo);
-	else
+	else {
+		int unit;
+		const unsigned int cpu = cpumask_first(&current->cpus_allowed);
+		const unsigned int weight =
+			cpumask_weight(&current->cpus_allowed);
+
+		if (weight == 1 && !test_bit(cpu, qib_cpulist))
+			if (!find_hca(cpu, &unit) && unit >= 0)
+				if (!find_free_ctxt(unit, fp, uinfo)) {
+					ret = 0;
+					goto done_chk_sdma;
+				}
 		ret = get_a_ctxt(fp, uinfo, alg);
-
-done_chk_sdma:
-	if (!ret) {
-		struct qib_filedata *fd = fp->private_data;
-		const struct qib_ctxtdata *rcd = fd->rcd;
-		const struct qib_devdata *dd = rcd->dd;
-		unsigned int weight;
-
-		if (dd->flags & QIB_HAS_SEND_DMA) {
-			fd->pq = qib_user_sdma_queue_create(&dd->pcidev->dev,
-							    dd->unit,
-							    rcd->ctxt,
-							    fd->subctxt);
-			if (!fd->pq)
-				ret = -ENOMEM;
-		}
-
-		/*
-		 * If process has NOT already set it's affinity, select and
-		 * reserve a processor for it, as a rendezvous for all
-		 * users of the driver.  If they don't actually later
-		 * set affinity to this cpu, or set it to some other cpu,
-		 * it just means that sooner or later we don't recommend
-		 * a cpu, and let the scheduler do it's best.
-		 */
-		weight = cpumask_weight(tsk_cpus_allowed(current));
-		if (!ret && weight >= qib_cpulist_count) {
-			int cpu;
-			cpu = find_first_zero_bit(qib_cpulist,
-						  qib_cpulist_count);
-			if (cpu != qib_cpulist_count) {
-				__set_bit(cpu, qib_cpulist);
-				fd->rec_cpu_num = cpu;
-			}
-		} else if (weight == 1 &&
-			test_bit(cpumask_first(tsk_cpus_allowed(current)),
-				 qib_cpulist))
-			qib_devinfo(dd->pcidev,
-				"%s PID %u affinity set to cpu %d; already allocated\n",
-				current->comm, current->pid,
-				cpumask_first(tsk_cpus_allowed(current)));
 	}
 
+done_chk_sdma:
+	if (!ret)
+		ret = do_qib_user_sdma_queue_create(fp);
+done_ok:
 	mutex_unlock(&qib_mutex);
 
 done: