IB/srp: One FMR pool per SRP connection

Allocate one FMR pool per SRP connection instead of one SRP pool
per HCA. This improves scalability of the SRP initiator.

Only request the SCSI mid-layer to retry a SCSI command after a
temporary mapping failure (-ENOMEM) but not after a permanent
mapping failure. This avoids that SCSI commands are retried
indefinitely if a permanent memory mapping failure occurs.

Tell the SCSI mid-layer to reduce queue depth temporarily in the
unlikely case where an application is queuing many requests with
more than max_pages_per_fmr sg-list elements.

For FMR pool allocation, base the max_pages_per_fmr parameter on
the HCA memory registration limit. Only try to allocate an FMR
pool if FMR is supported.

Signed-off-by: Bart Van Assche <bvanassche@acm.org>
Signed-off-by: Roland Dreier <roland@purestorage.com>

2 files changed, 84 insertions, 52 deletions

diff --git a/drivers/infiniband/ulp/srp/ib_srp.c b/drivers/infiniband/ulp/srp/ib_srp.c
index 77ba96553b27..80dfe173deac 100644
--- a/drivers/infiniband/ulp/srp/ib_srp.c
+++ b/drivers/infiniband/ulp/srp/ib_srp.c
@@ -293,12 +293,31 @@ static int srp_new_cm_id(struct srp_target_port *target)
 	return 0;
 }
 
+static struct ib_fmr_pool *srp_alloc_fmr_pool(struct srp_target_port *target)
+{
+	struct srp_device *dev = target->srp_host->srp_dev;
+	struct ib_fmr_pool_param fmr_param;
+
+	memset(&fmr_param, 0, sizeof(fmr_param));
+	fmr_param.pool_size	    = target->scsi_host->can_queue;
+	fmr_param.dirty_watermark   = fmr_param.pool_size / 4;
+	fmr_param.cache		    = 1;
+	fmr_param.max_pages_per_fmr = dev->max_pages_per_fmr;
+	fmr_param.page_shift	    = ilog2(dev->fmr_page_size);
+	fmr_param.access	    = (IB_ACCESS_LOCAL_WRITE |
+				       IB_ACCESS_REMOTE_WRITE |
+				       IB_ACCESS_REMOTE_READ);
+
+	return ib_create_fmr_pool(dev->pd, &fmr_param);
+}
+
 static int srp_create_target_ib(struct srp_target_port *target)
 {
 	struct srp_device *dev = target->srp_host->srp_dev;
 	struct ib_qp_init_attr *init_attr;
 	struct ib_cq *recv_cq, *send_cq;
 	struct ib_qp *qp;
+	struct ib_fmr_pool *fmr_pool = NULL;
 	int ret;
 
 	init_attr = kzalloc(sizeof *init_attr, GFP_KERNEL);
@@ -341,6 +360,19 @@ static int srp_create_target_ib(struct srp_target_port *target)
 	if (ret)
 		goto err_qp;
 
+	if (dev->has_fmr) {
+		fmr_pool = srp_alloc_fmr_pool(target);
+		if (IS_ERR(fmr_pool)) {
+			ret = PTR_ERR(fmr_pool);
+			shost_printk(KERN_WARNING, target->scsi_host, PFX
+				     "FMR pool allocation failed (%d)\n", ret);
+			goto err_qp;
+		}
+		if (target->fmr_pool)
+			ib_destroy_fmr_pool(target->fmr_pool);
+		target->fmr_pool = fmr_pool;
+	}
+
 	if (target->qp)
 		ib_destroy_qp(target->qp);
 	if (target->recv_cq)
@@ -377,6 +409,8 @@ static void srp_free_target_ib(struct srp_target_port *target)
 {
 	int i;
 
+	if (target->fmr_pool)
+		ib_destroy_fmr_pool(target->fmr_pool);
 	ib_destroy_qp(target->qp);
 	ib_destroy_cq(target->send_cq);
 	ib_destroy_cq(target->recv_cq);
@@ -623,8 +657,8 @@ static int srp_alloc_req_data(struct srp_target_port *target)
 		req = &target->req_ring[i];
 		req->fmr_list = kmalloc(target->cmd_sg_cnt * sizeof(void *),
 					GFP_KERNEL);
-		req->map_page = kmalloc(SRP_FMR_SIZE * sizeof(void *),
-					GFP_KERNEL);
+		req->map_page = kmalloc(srp_dev->max_pages_per_fmr *
+					sizeof(void *), GFP_KERNEL);
 		req->indirect_desc = kmalloc(target->indirect_size, GFP_KERNEL);
 		if (!req->fmr_list || !req->map_page || !req->indirect_desc)
 			goto out;
@@ -936,11 +970,10 @@ static void srp_map_desc(struct srp_map_state *state, dma_addr_t dma_addr,
 static int srp_map_finish_fmr(struct srp_map_state *state,
 			      struct srp_target_port *target)
 {
-	struct srp_device *dev = target->srp_host->srp_dev;
 	struct ib_pool_fmr *fmr;
 	u64 io_addr = 0;
 
-	fmr = ib_fmr_pool_map_phys(dev->fmr_pool, state->pages,
+	fmr = ib_fmr_pool_map_phys(target->fmr_pool, state->pages,
 				   state->npages, io_addr);
 	if (IS_ERR(fmr))
 		return PTR_ERR(fmr);
@@ -1033,7 +1066,7 @@ static int srp_map_sg_entry(struct srp_map_state *state,
 		srp_map_update_start(state, sg, sg_index, dma_addr);
 
 	while (dma_len) {
-		if (state->npages == SRP_FMR_SIZE) {
+		if (state->npages == dev->max_pages_per_fmr) {
 			ret = srp_finish_mapping(state, target);
 			if (ret)
 				return ret;
@@ -1077,7 +1110,7 @@ static void srp_map_fmr(struct srp_map_state *state,
 	state->pages	= req->map_page;
 	state->next_fmr	= req->fmr_list;
 
-	use_fmr = dev->fmr_pool ? SRP_MAP_ALLOW_FMR : SRP_MAP_NO_FMR;
+	use_fmr = target->fmr_pool ? SRP_MAP_ALLOW_FMR : SRP_MAP_NO_FMR;
 
 	for_each_sg(scat, sg, count, i) {
 		if (srp_map_sg_entry(state, target, sg, i, use_fmr)) {
@@ -1555,7 +1588,7 @@ static int srp_queuecommand(struct Scsi_Host *shost, struct scsi_cmnd *scmnd)
 	struct srp_cmd *cmd;
 	struct ib_device *dev;
 	unsigned long flags;
-	int len, result;
+	int len, ret;
 	const bool in_scsi_eh = !in_interrupt() && current == shost->ehandler;
 
 	/*
@@ -1567,12 +1600,9 @@ static int srp_queuecommand(struct Scsi_Host *shost, struct scsi_cmnd *scmnd)
 	if (in_scsi_eh)
 		mutex_lock(&rport->mutex);
 
-	result = srp_chkready(target->rport);
-	if (unlikely(result)) {
-		scmnd->result = result;
-		scmnd->scsi_done(scmnd);
-		goto unlock_rport;
-	}
+	scmnd->result = srp_chkready(target->rport);
+	if (unlikely(scmnd->result))
+		goto err;
 
 	spin_lock_irqsave(&target->lock, flags);
 	iu = __srp_get_tx_iu(target, SRP_IU_CMD);
@@ -1587,7 +1617,6 @@ static int srp_queuecommand(struct Scsi_Host *shost, struct scsi_cmnd *scmnd)
 	ib_dma_sync_single_for_cpu(dev, iu->dma, target->max_iu_len,
 				   DMA_TO_DEVICE);
 
-	scmnd->result        = 0;
 	scmnd->host_scribble = (void *) req;
 
 	cmd = iu->buf;
@@ -1604,7 +1633,15 @@ static int srp_queuecommand(struct Scsi_Host *shost, struct scsi_cmnd *scmnd)
 	len = srp_map_data(scmnd, target, req);
 	if (len < 0) {
 		shost_printk(KERN_ERR, target->scsi_host,
-			     PFX "Failed to map data\n");
+			     PFX "Failed to map data (%d)\n", len);
+		/*
+		 * If we ran out of memory descriptors (-ENOMEM) because an
+		 * application is queuing many requests with more than
+		 * max_pages_per_fmr sg-list elements, tell the SCSI mid-layer
+		 * to reduce queue depth temporarily.
+		 */
+		scmnd->result = len == -ENOMEM ?
+			DID_OK << 16 | QUEUE_FULL << 1 : DID_ERROR << 16;
 		goto err_iu;
 	}
 
@@ -1616,11 +1653,13 @@ static int srp_queuecommand(struct Scsi_Host *shost, struct scsi_cmnd *scmnd)
 		goto err_unmap;
 	}
 
+	ret = 0;
+
 unlock_rport:
 	if (in_scsi_eh)
 		mutex_unlock(&rport->mutex);
 
-	return 0;
+	return ret;
 
 err_unmap:
 	srp_unmap_data(scmnd, target, req);
@@ -1640,10 +1679,15 @@ err_iu:
 err_unlock:
 	spin_unlock_irqrestore(&target->lock, flags);
 
-	if (in_scsi_eh)
-		mutex_unlock(&rport->mutex);
+err:
+	if (scmnd->result) {
+		scmnd->scsi_done(scmnd);
+		ret = 0;
+	} else {
+		ret = SCSI_MLQUEUE_HOST_BUSY;
+	}
 
-	return SCSI_MLQUEUE_HOST_BUSY;
+	goto unlock_rport;
 }
 
 /*
@@ -2647,7 +2691,8 @@ static ssize_t srp_create_target(struct device *dev,
 		container_of(dev, struct srp_host, dev);
 	struct Scsi_Host *target_host;
 	struct srp_target_port *target;
-	struct ib_device *ibdev = host->srp_dev->dev;
+	struct srp_device *srp_dev = host->srp_dev;
+	struct ib_device *ibdev = srp_dev->dev;
 	int ret;
 
 	target_host = scsi_host_alloc(&srp_template,
@@ -2692,8 +2737,8 @@ static ssize_t srp_create_target(struct device *dev,
 		goto err;
 	}
 
-	if (!host->srp_dev->fmr_pool && !target->allow_ext_sg &&
-				target->cmd_sg_cnt < target->sg_tablesize) {
+	if (!srp_dev->has_fmr && !target->allow_ext_sg &&
+	    target->cmd_sg_cnt < target->sg_tablesize) {
 		pr_warn("No FMR pool and no external indirect descriptors, limiting sg_tablesize to cmd_sg_cnt\n");
 		target->sg_tablesize = target->cmd_sg_cnt;
 	}
@@ -2832,9 +2877,9 @@ static void srp_add_one(struct ib_device *device)
 {
 	struct srp_device *srp_dev;
 	struct ib_device_attr *dev_attr;
-	struct ib_fmr_pool_param fmr_param;
 	struct srp_host *host;
-	int max_pages_per_fmr, fmr_page_shift, s, e, p;
+	int fmr_page_shift, s, e, p;
+	u64 max_pages_per_fmr;
 
 	dev_attr = kmalloc(sizeof *dev_attr, GFP_KERNEL);
 	if (!dev_attr)
@@ -2849,6 +2894,9 @@ static void srp_add_one(struct ib_device *device)
 	if (!srp_dev)
 		goto free_attr;
 
+	srp_dev->has_fmr = (device->alloc_fmr && device->dealloc_fmr &&
+			    device->map_phys_fmr && device->unmap_fmr);
+
 	/*
 	 * Use the smallest page size supported by the HCA, down to a
 	 * minimum of 4096 bytes. We're unlikely to build large sglists
@@ -2857,7 +2905,15 @@ static void srp_add_one(struct ib_device *device)
 	fmr_page_shift		= max(12, ffs(dev_attr->page_size_cap) - 1);
 	srp_dev->fmr_page_size	= 1 << fmr_page_shift;
 	srp_dev->fmr_page_mask	= ~((u64) srp_dev->fmr_page_size - 1);
-	srp_dev->fmr_max_size	= srp_dev->fmr_page_size * SRP_FMR_SIZE;
+	max_pages_per_fmr	= dev_attr->max_mr_size;
+	do_div(max_pages_per_fmr, srp_dev->fmr_page_size);
+	srp_dev->max_pages_per_fmr = min_t(u64, SRP_FMR_SIZE,
+					   max_pages_per_fmr);
+	srp_dev->fmr_max_size	= srp_dev->fmr_page_size *
+				   srp_dev->max_pages_per_fmr;
+	pr_debug("%s: fmr_page_shift = %d, dev_attr->max_mr_size = %#llx, max_pages_per_fmr = %d, fmr_max_size = %#x\n",
+		 device->name, fmr_page_shift, dev_attr->max_mr_size,
+		 srp_dev->max_pages_per_fmr, srp_dev->fmr_max_size);
 
 	INIT_LIST_HEAD(&srp_dev->dev_list);
 
@@ -2873,27 +2929,6 @@ static void srp_add_one(struct ib_device *device)
 	if (IS_ERR(srp_dev->mr))
 		goto err_pd;
 
-	for (max_pages_per_fmr = SRP_FMR_SIZE;
-			max_pages_per_fmr >= SRP_FMR_MIN_SIZE;
-			max_pages_per_fmr /= 2, srp_dev->fmr_max_size /= 2) {
-		memset(&fmr_param, 0, sizeof fmr_param);
-		fmr_param.pool_size	    = SRP_FMR_POOL_SIZE;
-		fmr_param.dirty_watermark   = SRP_FMR_DIRTY_SIZE;
-		fmr_param.cache		    = 1;
-		fmr_param.max_pages_per_fmr = max_pages_per_fmr;
-		fmr_param.page_shift	    = fmr_page_shift;
-		fmr_param.access	    = (IB_ACCESS_LOCAL_WRITE |
-					       IB_ACCESS_REMOTE_WRITE |
-					       IB_ACCESS_REMOTE_READ);
-
-		srp_dev->fmr_pool = ib_create_fmr_pool(srp_dev->pd, &fmr_param);
-		if (!IS_ERR(srp_dev->fmr_pool))
-			break;
-	}
-
-	if (IS_ERR(srp_dev->fmr_pool))
-		srp_dev->fmr_pool = NULL;
-
 	if (device->node_type == RDMA_NODE_IB_SWITCH) {
 		s = 0;
 		e = 0;
@@ -2956,8 +2991,6 @@ static void srp_remove_one(struct ib_device *device)
 		kfree(host);
 	}
 
-	if (srp_dev->fmr_pool)
-		ib_destroy_fmr_pool(srp_dev->fmr_pool);
 	ib_dereg_mr(srp_dev->mr);
 	ib_dealloc_pd(srp_dev->pd);
 
diff --git a/drivers/infiniband/ulp/srp/ib_srp.h b/drivers/infiniband/ulp/srp/ib_srp.h
index aad27b7b4a46..2d99e52f2f5c 100644
--- a/drivers/infiniband/ulp/srp/ib_srp.h
+++ b/drivers/infiniband/ulp/srp/ib_srp.h
@@ -67,9 +67,6 @@ enum {
 	SRP_TAG_TSK_MGMT	= 1U << 31,
 
 	SRP_FMR_SIZE		= 512,
-	SRP_FMR_MIN_SIZE	= 128,
-	SRP_FMR_POOL_SIZE	= 1024,
-	SRP_FMR_DIRTY_SIZE	= SRP_FMR_POOL_SIZE / 4,
 
 	SRP_MAP_ALLOW_FMR	= 0,
 	SRP_MAP_NO_FMR		= 1,
@@ -91,10 +88,11 @@ struct srp_device {
 	struct ib_device       *dev;
 	struct ib_pd	       *pd;
 	struct ib_mr	       *mr;
-	struct ib_fmr_pool     *fmr_pool;
 	u64			fmr_page_mask;
 	int			fmr_page_size;
 	int			fmr_max_size;
+	int			max_pages_per_fmr;
+	bool			has_fmr;
 };
 
 struct srp_host {
@@ -131,6 +129,7 @@ struct srp_target_port {
 	struct ib_cq	       *send_cq ____cacheline_aligned_in_smp;
 	struct ib_cq	       *recv_cq;
 	struct ib_qp	       *qp;
+	struct ib_fmr_pool     *fmr_pool;
 	u32			lkey;
 	u32			rkey;
 	enum srp_target_state	state;