1 files changed, 137 insertions, 143 deletions

diff --git a/drivers/nvme/host/rdma.c b/drivers/nvme/host/rdma.c
index 4f9bf2f815c3..2a0bba7f50cf 100644
--- a/drivers/nvme/host/rdma.c
+++ b/drivers/nvme/host/rdma.c
@@ -15,6 +15,7 @@
 #include <linux/module.h>
 #include <linux/init.h>
 #include <linux/slab.h>
+#include <rdma/mr_pool.h>
 #include <linux/err.h>
 #include <linux/string.h>
 #include <linux/atomic.h>
@@ -59,6 +60,9 @@ struct nvme_rdma_request {
 	struct nvme_request	req;
 	struct ib_mr		*mr;
 	struct nvme_rdma_qe	sqe;
+	union nvme_result	result;
+	__le16			status;
+	refcount_t		ref;
 	struct ib_sge		sge[1 + NVME_RDMA_MAX_INLINE_SEGMENTS];
 	u32			num_sge;
 	int			nents;
@@ -73,11 +77,11 @@ struct nvme_rdma_request {
 enum nvme_rdma_queue_flags {
 	NVME_RDMA_Q_ALLOCATED		= 0,
 	NVME_RDMA_Q_LIVE		= 1,
+	NVME_RDMA_Q_TR_READY		= 2,
 };
 
 struct nvme_rdma_queue {
 	struct nvme_rdma_qe	*rsp_ring;
-	atomic_t		sig_count;
 	int			queue_size;
 	size_t			cmnd_capsule_len;
 	struct nvme_rdma_ctrl	*ctrl;
@@ -258,32 +262,6 @@ static int nvme_rdma_create_qp(struct nvme_rdma_queue *queue, const int factor)
 	return ret;
 }
 
-static int nvme_rdma_reinit_request(void *data, struct request *rq)
-{
-	struct nvme_rdma_ctrl *ctrl = data;
-	struct nvme_rdma_device *dev = ctrl->device;
-	struct nvme_rdma_request *req = blk_mq_rq_to_pdu(rq);
-	int ret = 0;
-
-	if (WARN_ON_ONCE(!req->mr))
-		return 0;
-
-	ib_dereg_mr(req->mr);
-
-	req->mr = ib_alloc_mr(dev->pd, IB_MR_TYPE_MEM_REG,
-			ctrl->max_fr_pages);
-	if (IS_ERR(req->mr)) {
-		ret = PTR_ERR(req->mr);
-		req->mr = NULL;
-		goto out;
-	}
-
-	req->mr->need_inval = false;
-
-out:
-	return ret;
-}
-
 static void nvme_rdma_exit_request(struct blk_mq_tag_set *set,
 		struct request *rq, unsigned int hctx_idx)
 {
@@ -293,9 +271,6 @@ static void nvme_rdma_exit_request(struct blk_mq_tag_set *set,
 	struct nvme_rdma_queue *queue = &ctrl->queues[queue_idx];
 	struct nvme_rdma_device *dev = queue->device;
 
-	if (req->mr)
-		ib_dereg_mr(req->mr);
-
 	nvme_rdma_free_qe(dev->dev, &req->sqe, sizeof(struct nvme_command),
 			DMA_TO_DEVICE);
 }
@@ -317,21 +292,9 @@ static int nvme_rdma_init_request(struct blk_mq_tag_set *set,
 	if (ret)
 		return ret;
 
-	req->mr = ib_alloc_mr(dev->pd, IB_MR_TYPE_MEM_REG,
-			ctrl->max_fr_pages);
-	if (IS_ERR(req->mr)) {
-		ret = PTR_ERR(req->mr);
-		goto out_free_qe;
-	}
-
 	req->queue = queue;
 
 	return 0;
-
-out_free_qe:
-	nvme_rdma_free_qe(dev->dev, &req->sqe, sizeof(struct nvme_command),
-			DMA_TO_DEVICE);
-	return -ENOMEM;
 }
 
 static int nvme_rdma_init_hctx(struct blk_mq_hw_ctx *hctx, void *data,
@@ -428,10 +391,23 @@ out_err:
 
 static void nvme_rdma_destroy_queue_ib(struct nvme_rdma_queue *queue)
 {
-	struct nvme_rdma_device *dev = queue->device;
-	struct ib_device *ibdev = dev->dev;
+	struct nvme_rdma_device *dev;
+	struct ib_device *ibdev;
 
-	rdma_destroy_qp(queue->cm_id);
+	if (!test_and_clear_bit(NVME_RDMA_Q_TR_READY, &queue->flags))
+		return;
+
+	dev = queue->device;
+	ibdev = dev->dev;
+
+	ib_mr_pool_destroy(queue->qp, &queue->qp->rdma_mrs);
+
+	/*
+	 * The cm_id object might have been destroyed during RDMA connection
+	 * establishment error flow to avoid getting other cma events, thus
+	 * the destruction of the QP shouldn't use rdma_cm API.
+	 */
+	ib_destroy_qp(queue->qp);
 	ib_free_cq(queue->ib_cq);
 
 	nvme_rdma_free_ring(ibdev, queue->rsp_ring, queue->queue_size,
@@ -440,6 +416,12 @@ static void nvme_rdma_destroy_queue_ib(struct nvme_rdma_queue *queue)
 	nvme_rdma_dev_put(dev);
 }
 
+static int nvme_rdma_get_max_fr_pages(struct ib_device *ibdev)
+{
+	return min_t(u32, NVME_RDMA_MAX_SEGMENTS,
+		     ibdev->attrs.max_fast_reg_page_list_len);
+}
+
 static int nvme_rdma_create_queue_ib(struct nvme_rdma_queue *queue)
 {
 	struct ib_device *ibdev;
@@ -482,8 +464,24 @@ static int nvme_rdma_create_queue_ib(struct nvme_rdma_queue *queue)
 		goto out_destroy_qp;
 	}
 
+	ret = ib_mr_pool_init(queue->qp, &queue->qp->rdma_mrs,
+			      queue->queue_size,
+			      IB_MR_TYPE_MEM_REG,
+			      nvme_rdma_get_max_fr_pages(ibdev));
+	if (ret) {
+		dev_err(queue->ctrl->ctrl.device,
+			"failed to initialize MR pool sized %d for QID %d\n",
+			queue->queue_size, idx);
+		goto out_destroy_ring;
+	}
+
+	set_bit(NVME_RDMA_Q_TR_READY, &queue->flags);
+
 	return 0;
 
+out_destroy_ring:
+	nvme_rdma_free_ring(ibdev, queue->rsp_ring, queue->queue_size,
+			    sizeof(struct nvme_completion), DMA_FROM_DEVICE);
 out_destroy_qp:
 	rdma_destroy_qp(queue->cm_id);
 out_destroy_ib_cq:
@@ -510,7 +508,6 @@ static int nvme_rdma_alloc_queue(struct nvme_rdma_ctrl *ctrl,
 		queue->cmnd_capsule_len = sizeof(struct nvme_command);
 
 	queue->queue_size = queue_size;
-	atomic_set(&queue->sig_count, 0);
 
 	queue->cm_id = rdma_create_id(&init_net, nvme_rdma_cm_handler, queue,
 			RDMA_PS_TCP, IB_QPT_RC);
@@ -546,6 +543,7 @@ static int nvme_rdma_alloc_queue(struct nvme_rdma_ctrl *ctrl,
 
 out_destroy_cm_id:
 	rdma_destroy_id(queue->cm_id);
+	nvme_rdma_destroy_queue_ib(queue);
 	return ret;
 }
 
@@ -756,8 +754,7 @@ static int nvme_rdma_configure_admin_queue(struct nvme_rdma_ctrl *ctrl,
 
 	ctrl->device = ctrl->queues[0].device;
 
-	ctrl->max_fr_pages = min_t(u32, NVME_RDMA_MAX_SEGMENTS,
-		ctrl->device->dev->attrs.max_fast_reg_page_list_len);
+	ctrl->max_fr_pages = nvme_rdma_get_max_fr_pages(ctrl->device->dev);
 
 	if (new) {
 		ctrl->ctrl.admin_tagset = nvme_rdma_alloc_tagset(&ctrl->ctrl, true);
@@ -771,10 +768,6 @@ static int nvme_rdma_configure_admin_queue(struct nvme_rdma_ctrl *ctrl,
 			error = PTR_ERR(ctrl->ctrl.admin_q);
 			goto out_free_tagset;
 		}
-	} else {
-		error = nvme_reinit_tagset(&ctrl->ctrl, ctrl->ctrl.admin_tagset);
-		if (error)
-			goto out_free_queue;
 	}
 
 	error = nvme_rdma_start_queue(ctrl, 0);
@@ -854,10 +847,6 @@ static int nvme_rdma_configure_io_queues(struct nvme_rdma_ctrl *ctrl, bool new)
 			goto out_free_tag_set;
 		}
 	} else {
-		ret = nvme_reinit_tagset(&ctrl->ctrl, ctrl->ctrl.tagset);
-		if (ret)
-			goto out_free_io_queues;
-
 		blk_mq_update_nr_hw_queues(&ctrl->tag_set,
 			ctrl->ctrl.queue_count - 1);
 	}
@@ -985,12 +974,18 @@ static void nvme_rdma_error_recovery_work(struct work_struct *work)
 	blk_mq_unquiesce_queue(ctrl->ctrl.admin_q);
 	nvme_start_queues(&ctrl->ctrl);
 
+	if (!nvme_change_ctrl_state(&ctrl->ctrl, NVME_CTRL_RECONNECTING)) {
+		/* state change failure should never happen */
+		WARN_ON_ONCE(1);
+		return;
+	}
+
 	nvme_rdma_reconnect_or_remove(ctrl);
 }
 
 static void nvme_rdma_error_recovery(struct nvme_rdma_ctrl *ctrl)
 {
-	if (!nvme_change_ctrl_state(&ctrl->ctrl, NVME_CTRL_RECONNECTING))
+	if (!nvme_change_ctrl_state(&ctrl->ctrl, NVME_CTRL_RESETTING))
 		return;
 
 	queue_work(nvme_wq, &ctrl->err_work);
@@ -1018,8 +1013,18 @@ static void nvme_rdma_memreg_done(struct ib_cq *cq, struct ib_wc *wc)
 
 static void nvme_rdma_inv_rkey_done(struct ib_cq *cq, struct ib_wc *wc)
 {
-	if (unlikely(wc->status != IB_WC_SUCCESS))
+	struct nvme_rdma_request *req =
+		container_of(wc->wr_cqe, struct nvme_rdma_request, reg_cqe);
+	struct request *rq = blk_mq_rq_from_pdu(req);
+
+	if (unlikely(wc->status != IB_WC_SUCCESS)) {
 		nvme_rdma_wr_error(cq, wc, "LOCAL_INV");
+		return;
+	}
+
+	if (refcount_dec_and_test(&req->ref))
+		nvme_end_request(rq, req->status, req->result);
+
 }
 
 static int nvme_rdma_inv_rkey(struct nvme_rdma_queue *queue,
@@ -1030,7 +1035,7 @@ static int nvme_rdma_inv_rkey(struct nvme_rdma_queue *queue,
 		.opcode		    = IB_WR_LOCAL_INV,
 		.next		    = NULL,
 		.num_sge	    = 0,
-		.send_flags	    = 0,
+		.send_flags	    = IB_SEND_SIGNALED,
 		.ex.invalidate_rkey = req->mr->rkey,
 	};
 
@@ -1044,22 +1049,15 @@ static void nvme_rdma_unmap_data(struct nvme_rdma_queue *queue,
 		struct request *rq)
 {
 	struct nvme_rdma_request *req = blk_mq_rq_to_pdu(rq);
-	struct nvme_rdma_ctrl *ctrl = queue->ctrl;
 	struct nvme_rdma_device *dev = queue->device;
 	struct ib_device *ibdev = dev->dev;
-	int res;
 
 	if (!blk_rq_bytes(rq))
 		return;
 
-	if (req->mr->need_inval && test_bit(NVME_RDMA_Q_LIVE, &req->queue->flags)) {
-		res = nvme_rdma_inv_rkey(queue, req);
-		if (unlikely(res < 0)) {
-			dev_err(ctrl->ctrl.device,
-				"Queueing INV WR for rkey %#x failed (%d)\n",
-				req->mr->rkey, res);
-			nvme_rdma_error_recovery(queue->ctrl);
-		}
+	if (req->mr) {
+		ib_mr_pool_put(queue->qp, &queue->qp->rdma_mrs, req->mr);
+		req->mr = NULL;
 	}
 
 	ib_dma_unmap_sg(ibdev, req->sg_table.sgl,
@@ -1118,12 +1116,18 @@ static int nvme_rdma_map_sg_fr(struct nvme_rdma_queue *queue,
 	struct nvme_keyed_sgl_desc *sg = &c->common.dptr.ksgl;
 	int nr;
 
+	req->mr = ib_mr_pool_get(queue->qp, &queue->qp->rdma_mrs);
+	if (WARN_ON_ONCE(!req->mr))
+		return -EAGAIN;
+
 	/*
 	 * Align the MR to a 4K page size to match the ctrl page size and
 	 * the block virtual boundary.
 	 */
 	nr = ib_map_mr_sg(req->mr, req->sg_table.sgl, count, NULL, SZ_4K);
 	if (unlikely(nr < count)) {
+		ib_mr_pool_put(queue->qp, &queue->qp->rdma_mrs, req->mr);
+		req->mr = NULL;
 		if (nr < 0)
 			return nr;
 		return -EINVAL;
@@ -1142,8 +1146,6 @@ static int nvme_rdma_map_sg_fr(struct nvme_rdma_queue *queue,
 			     IB_ACCESS_REMOTE_READ |
 			     IB_ACCESS_REMOTE_WRITE;
 
-	req->mr->need_inval = true;
-
 	sg->addr = cpu_to_le64(req->mr->iova);
 	put_unaligned_le24(req->mr->length, sg->length);
 	put_unaligned_le32(req->mr->rkey, sg->key);
@@ -1163,7 +1165,7 @@ static int nvme_rdma_map_data(struct nvme_rdma_queue *queue,
 
 	req->num_sge = 1;
 	req->inline_data = false;
-	req->mr->need_inval = false;
+	refcount_set(&req->ref, 2); /* send and recv completions */
 
 	c->common.flags |= NVME_CMD_SGL_METABUF;
 
@@ -1200,25 +1202,24 @@ static int nvme_rdma_map_data(struct nvme_rdma_queue *queue,
 
 static void nvme_rdma_send_done(struct ib_cq *cq, struct ib_wc *wc)
 {
-	if (unlikely(wc->status != IB_WC_SUCCESS))
-		nvme_rdma_wr_error(cq, wc, "SEND");
-}
+	struct nvme_rdma_qe *qe =
+		container_of(wc->wr_cqe, struct nvme_rdma_qe, cqe);
+	struct nvme_rdma_request *req =
+		container_of(qe, struct nvme_rdma_request, sqe);
+	struct request *rq = blk_mq_rq_from_pdu(req);
 
-/*
- * We want to signal completion at least every queue depth/2.  This returns the
- * largest power of two that is not above half of (queue size + 1) to optimize
- * (avoid divisions).
- */
-static inline bool nvme_rdma_queue_sig_limit(struct nvme_rdma_queue *queue)
-{
-	int limit = 1 << ilog2((queue->queue_size + 1) / 2);
+	if (unlikely(wc->status != IB_WC_SUCCESS)) {
+		nvme_rdma_wr_error(cq, wc, "SEND");
+		return;
+	}
 
-	return (atomic_inc_return(&queue->sig_count) & (limit - 1)) == 0;
+	if (refcount_dec_and_test(&req->ref))
+		nvme_end_request(rq, req->status, req->result);
 }
 
 static int nvme_rdma_post_send(struct nvme_rdma_queue *queue,
 		struct nvme_rdma_qe *qe, struct ib_sge *sge, u32 num_sge,
-		struct ib_send_wr *first, bool flush)
+		struct ib_send_wr *first)
 {
 	struct ib_send_wr wr, *bad_wr;
 	int ret;
@@ -1227,31 +1228,12 @@ static int nvme_rdma_post_send(struct nvme_rdma_queue *queue,
 	sge->length = sizeof(struct nvme_command),
 	sge->lkey   = queue->device->pd->local_dma_lkey;
 
-	qe->cqe.done = nvme_rdma_send_done;
-
 	wr.next       = NULL;
 	wr.wr_cqe     = &qe->cqe;
 	wr.sg_list    = sge;
 	wr.num_sge    = num_sge;
 	wr.opcode     = IB_WR_SEND;
-	wr.send_flags = 0;
-
-	/*
-	 * Unsignalled send completions are another giant desaster in the
-	 * IB Verbs spec:  If we don't regularly post signalled sends
-	 * the send queue will fill up and only a QP reset will rescue us.
-	 * Would have been way to obvious to handle this in hardware or
-	 * at least the RDMA stack..
-	 *
-	 * Always signal the flushes. The magic request used for the flush
-	 * sequencer is not allocated in our driver's tagset and it's
-	 * triggered to be freed by blk_cleanup_queue(). So we need to
-	 * always mark it as signaled to ensure that the "wr_cqe", which is
-	 * embedded in request's payload, is not freed when __ib_process_cq()
-	 * calls wr_cqe->done().
-	 */
-	if (nvme_rdma_queue_sig_limit(queue) || flush)
-		wr.send_flags |= IB_SEND_SIGNALED;
+	wr.send_flags = IB_SEND_SIGNALED;
 
 	if (first)
 		first->next = &wr;
@@ -1301,6 +1283,12 @@ static struct blk_mq_tags *nvme_rdma_tagset(struct nvme_rdma_queue *queue)
 	return queue->ctrl->tag_set.tags[queue_idx - 1];
 }
 
+static void nvme_rdma_async_done(struct ib_cq *cq, struct ib_wc *wc)
+{
+	if (unlikely(wc->status != IB_WC_SUCCESS))
+		nvme_rdma_wr_error(cq, wc, "ASYNC");
+}
+
 static void nvme_rdma_submit_async_event(struct nvme_ctrl *arg)
 {
 	struct nvme_rdma_ctrl *ctrl = to_rdma_ctrl(arg);
@@ -1319,10 +1307,12 @@ static void nvme_rdma_submit_async_event(struct nvme_ctrl *arg)
 	cmd->common.flags |= NVME_CMD_SGL_METABUF;
 	nvme_rdma_set_sg_null(cmd);
 
+	sqe->cqe.done = nvme_rdma_async_done;
+
 	ib_dma_sync_single_for_device(dev, sqe->dma, sizeof(*cmd),
 			DMA_TO_DEVICE);
 
-	ret = nvme_rdma_post_send(queue, sqe, &sge, 1, NULL, false);
+	ret = nvme_rdma_post_send(queue, sqe, &sge, 1, NULL);
 	WARN_ON_ONCE(ret);
 }
 
@@ -1343,14 +1333,34 @@ static int nvme_rdma_process_nvme_rsp(struct nvme_rdma_queue *queue,
 	}
 	req = blk_mq_rq_to_pdu(rq);
 
-	if (rq->tag == tag)
-		ret = 1;
+	req->status = cqe->status;
+	req->result = cqe->result;
 
-	if ((wc->wc_flags & IB_WC_WITH_INVALIDATE) &&
-	    wc->ex.invalidate_rkey == req->mr->rkey)
-		req->mr->need_inval = false;
+	if (wc->wc_flags & IB_WC_WITH_INVALIDATE) {
+		if (unlikely(wc->ex.invalidate_rkey != req->mr->rkey)) {
+			dev_err(queue->ctrl->ctrl.device,
+				"Bogus remote invalidation for rkey %#x\n",
+				req->mr->rkey);
+			nvme_rdma_error_recovery(queue->ctrl);
+		}
+	} else if (req->mr) {
+		ret = nvme_rdma_inv_rkey(queue, req);
+		if (unlikely(ret < 0)) {
+			dev_err(queue->ctrl->ctrl.device,
+				"Queueing INV WR for rkey %#x failed (%d)\n",
+				req->mr->rkey, ret);
+			nvme_rdma_error_recovery(queue->ctrl);
+		}
+		/* the local invalidation completion will end the request */
+		return 0;
+	}
+
+	if (refcount_dec_and_test(&req->ref)) {
+		if (rq->tag == tag)
+			ret = 1;
+		nvme_end_request(rq, req->status, req->result);
+	}
 
-	nvme_end_request(rq, cqe->status, cqe->result);
 	return ret;
 }
 
@@ -1591,31 +1601,11 @@ nvme_rdma_timeout(struct request *rq, bool reserved)
  * We cannot accept any other command until the Connect command has completed.
  */
 static inline blk_status_t
-nvme_rdma_queue_is_ready(struct nvme_rdma_queue *queue, struct request *rq)
-{
-	if (unlikely(!test_bit(NVME_RDMA_Q_LIVE, &queue->flags))) {
-		struct nvme_command *cmd = nvme_req(rq)->cmd;
-
-		if (!blk_rq_is_passthrough(rq) ||
-		    cmd->common.opcode != nvme_fabrics_command ||
-		    cmd->fabrics.fctype != nvme_fabrics_type_connect) {
-			/*
-			 * reconnecting state means transport disruption, which
-			 * can take a long time and even might fail permanently,
-			 * fail fast to give upper layers a chance to failover.
-			 * deleting state means that the ctrl will never accept
-			 * commands again, fail it permanently.
-			 */
-			if (queue->ctrl->ctrl.state == NVME_CTRL_RECONNECTING ||
-			    queue->ctrl->ctrl.state == NVME_CTRL_DELETING) {
-				nvme_req(rq)->status = NVME_SC_ABORT_REQ;
-				return BLK_STS_IOERR;
-			}
-			return BLK_STS_RESOURCE; /* try again later */
-		}
-	}
-
-	return 0;
+nvme_rdma_is_ready(struct nvme_rdma_queue *queue, struct request *rq)
+{
+	if (unlikely(!test_bit(NVME_RDMA_Q_LIVE, &queue->flags)))
+		return nvmf_check_init_req(&queue->ctrl->ctrl, rq);
+	return BLK_STS_OK;
 }
 
 static blk_status_t nvme_rdma_queue_rq(struct blk_mq_hw_ctx *hctx,
@@ -1627,14 +1617,13 @@ static blk_status_t nvme_rdma_queue_rq(struct blk_mq_hw_ctx *hctx,
 	struct nvme_rdma_request *req = blk_mq_rq_to_pdu(rq);
 	struct nvme_rdma_qe *sqe = &req->sqe;
 	struct nvme_command *c = sqe->data;
-	bool flush = false;
 	struct ib_device *dev;
 	blk_status_t ret;
 	int err;
 
 	WARN_ON_ONCE(rq->tag < 0);
 
-	ret = nvme_rdma_queue_is_ready(queue, rq);
+	ret = nvme_rdma_is_ready(queue, rq);
 	if (unlikely(ret))
 		return ret;
 
@@ -1656,13 +1645,13 @@ static blk_status_t nvme_rdma_queue_rq(struct blk_mq_hw_ctx *hctx,
 		goto err;
 	}
 
+	sqe->cqe.done = nvme_rdma_send_done;
+
 	ib_dma_sync_single_for_device(dev, sqe->dma,
 			sizeof(struct nvme_command), DMA_TO_DEVICE);
 
-	if (req_op(rq) == REQ_OP_FLUSH)
-		flush = true;
 	err = nvme_rdma_post_send(queue, sqe, req->sge, req->num_sge,
-			req->mr->need_inval ? &req->reg_wr.wr : NULL, flush);
+			req->mr ? &req->reg_wr.wr : NULL);
 	if (unlikely(err)) {
 		nvme_rdma_unmap_data(queue, rq);
 		goto err;
@@ -1770,6 +1759,12 @@ static void nvme_rdma_reset_ctrl_work(struct work_struct *work)
 	nvme_stop_ctrl(&ctrl->ctrl);
 	nvme_rdma_shutdown_ctrl(ctrl, false);
 
+	if (!nvme_change_ctrl_state(&ctrl->ctrl, NVME_CTRL_RECONNECTING)) {
+		/* state change failure should never happen */
+		WARN_ON_ONCE(1);
+		return;
+	}
+
 	ret = nvme_rdma_configure_admin_queue(ctrl, false);
 	if (ret)
 		goto out_fail;
@@ -1810,7 +1805,6 @@ static const struct nvme_ctrl_ops nvme_rdma_ctrl_ops = {
 	.submit_async_event	= nvme_rdma_submit_async_event,
 	.delete_ctrl		= nvme_rdma_delete_ctrl,
 	.get_address		= nvmf_get_address,
-	.reinit_request		= nvme_rdma_reinit_request,
 };
 
 static inline bool