Linux RDMA and InfiniBand development
 help / color / mirror / Atom feed
From: Cheng Xu <chengyou@linux.alibaba.com>
To: jgg@ziepe.ca, leon@kernel.org
Cc: linux-rdma@vger.kernel.org, KaiShen@linux.alibaba.com
Subject: [PATCH 1/2] RDMA/erdma: Hold CQ references when processing EQ events
Date: Thu, 30 Jul 2026 20:43:53 +0800	[thread overview]
Message-ID: <20260730124357.12976-1-chengyou@linux.alibaba.com> (raw)

EQ handlers look up CQs from dev->cq_xa and invoke CQ completion or
error callbacks outside the xarray lock. erdma_destroy_cq() can erase the
CQ from the xarray and free its queue buffer and doorbell record while a
previously scheduled EQ handler is still using the CQ.

Add a CQ refcount and take a reference under the xarray lock with
refcount_inc_not_zero(). Remove the CQ from the xarray before dropping
the destroy-path reference, then wait for in-flight EQ users before
releasing CQ resources.

Fixes: 155055771704 ("RDMA/erdma: Add verbs implementation")
Signed-off-by: Cheng Xu <chengyou@linux.alibaba.com>
---
 drivers/infiniband/hw/erdma/erdma_eq.c    |  6 ++++--
 drivers/infiniband/hw/erdma/erdma_verbs.c | 12 +++++++++--
 drivers/infiniband/hw/erdma/erdma_verbs.h | 25 +++++++++++++++++++++--
 3 files changed, 37 insertions(+), 6 deletions(-)

diff --git a/drivers/infiniband/hw/erdma/erdma_eq.c b/drivers/infiniband/hw/erdma/erdma_eq.c
index a8784e07acd6..96e74fcf417c 100644
--- a/drivers/infiniband/hw/erdma/erdma_eq.c
+++ b/drivers/infiniband/hw/erdma/erdma_eq.c
@@ -52,7 +52,7 @@ void erdma_aeq_event_handler(struct erdma_dev *dev)
 		if (FIELD_GET(ERDMA_AEQE_HDR_TYPE_MASK,
 			      le32_to_cpu(aeqe->hdr)) == ERDMA_AE_TYPE_CQ_ERR) {
 			cqn = le32_to_cpu(aeqe->event_data0);
-			cq = find_cq_by_cqn(dev, cqn);
+			cq = erdma_cq_get_by_cqn(dev, cqn);
 			if (!cq)
 				continue;
 
@@ -62,6 +62,7 @@ void erdma_aeq_event_handler(struct erdma_dev *dev)
 			if (cq->ibcq.event_handler)
 				cq->ibcq.event_handler(&event,
 						       cq->ibcq.cq_context);
+			erdma_cq_put(cq);
 		} else {
 			qpn = le32_to_cpu(aeqe->event_data0);
 			qp = find_qp_by_qpn(dev, qpn);
@@ -157,7 +158,7 @@ void erdma_ceq_completion_handler(struct erdma_eq_cb *ceq_cb)
 		poll_cnt++;
 		cqn = FIELD_GET(ERDMA_CEQE_HDR_CQN_MASK, READ_ONCE(*ceqe));
 
-		cq = find_cq_by_cqn(dev, cqn);
+		cq = erdma_cq_get_by_cqn(dev, cqn);
 		if (!cq)
 			continue;
 
@@ -166,6 +167,7 @@ void erdma_ceq_completion_handler(struct erdma_eq_cb *ceq_cb)
 
 		if (cq->ibcq.comp_handler)
 			cq->ibcq.comp_handler(&cq->ibcq, cq->ibcq.cq_context);
+		erdma_cq_put(cq);
 	}
 
 	notify_eq(&ceq_cb->eq);
diff --git a/drivers/infiniband/hw/erdma/erdma_verbs.c b/drivers/infiniband/hw/erdma/erdma_verbs.c
index ab6abbab029e..2adcd00c4199 100644
--- a/drivers/infiniband/hw/erdma/erdma_verbs.c
+++ b/drivers/infiniband/hw/erdma/erdma_verbs.c
@@ -1326,6 +1326,7 @@ int erdma_destroy_cq(struct ib_cq *ibcq, struct ib_udata *udata)
 	struct erdma_dev *dev = to_edev(ibcq->device);
 	struct erdma_ucontext *ctx = rdma_udata_to_drv_context(
 		udata, struct erdma_ucontext, ibucontext);
+	unsigned long flags;
 	int err;
 	struct erdma_cmdq_destroy_cq_req req;
 
@@ -1340,6 +1341,13 @@ int erdma_destroy_cq(struct ib_cq *ibcq, struct ib_udata *udata)
 				       "failed to destroy CQ %u: %d\n",
 				       cq->cqn, err);
 
+	xa_lock_irqsave(&dev->cq_xa, flags);
+	__xa_erase(&dev->cq_xa, cq->cqn);
+	xa_unlock_irqrestore(&dev->cq_xa, flags);
+
+	erdma_cq_put(cq);
+	wait_for_completion(&cq->free);
+
 	if (rdma_is_kernel_res(&cq->ibcq.res)) {
 		dma_free_coherent(&dev->pdev->dev, cq->depth << CQE_SHIFT,
 				  cq->kern_cq.qbuf, cq->kern_cq.qbuf_dma_addr);
@@ -1350,8 +1358,6 @@ int erdma_destroy_cq(struct ib_cq *ibcq, struct ib_udata *udata)
 		put_mtt_entries(dev, &cq->user_cq.qbuf_mem);
 	}
 
-	xa_erase(&dev->cq_xa, cq->cqn);
-
 	return 0;
 }
 
@@ -1980,6 +1986,8 @@ int erdma_create_cq(struct ib_cq *ibcq, const struct ib_cq_init_attr *attr,
 	cq->ibcq.cqe = depth;
 	cq->depth = depth;
 	cq->assoc_eqn = attr->comp_vector + 1;
+	refcount_set(&cq->refcount, 1);
+	init_completion(&cq->free);
 
 	ret = xa_alloc_cyclic(&dev->cq_xa, &cq->cqn, cq,
 			      XA_LIMIT(1, dev->attrs.max_cq - 1),
diff --git a/drivers/infiniband/hw/erdma/erdma_verbs.h b/drivers/infiniband/hw/erdma/erdma_verbs.h
index 7d8d3fe501d5..894e080435fb 100644
--- a/drivers/infiniband/hw/erdma/erdma_verbs.h
+++ b/drivers/infiniband/hw/erdma/erdma_verbs.h
@@ -7,6 +7,9 @@
 #ifndef __ERDMA_VERBS_H__
 #define __ERDMA_VERBS_H__
 
+#include <linux/completion.h>
+#include <linux/refcount.h>
+
 #include "erdma.h"
 
 /* RDMA Capability. */
@@ -341,6 +344,8 @@ struct erdma_cq {
 
 	u32 depth;
 	u32 assoc_eqn;
+	refcount_t refcount;
+	struct completion free;
 
 	union {
 		struct erdma_kcq_info kern_cq;
@@ -355,9 +360,25 @@ static inline struct erdma_qp *find_qp_by_qpn(struct erdma_dev *dev, int id)
 	return (struct erdma_qp *)xa_load(&dev->qp_xa, id);
 }
 
-static inline struct erdma_cq *find_cq_by_cqn(struct erdma_dev *dev, int id)
+static inline struct erdma_cq *erdma_cq_get_by_cqn(struct erdma_dev *dev,
+						   int id)
+{
+	struct erdma_cq *cq;
+	unsigned long flags;
+
+	xa_lock_irqsave(&dev->cq_xa, flags);
+	cq = xa_load(&dev->cq_xa, id);
+	if (cq && !refcount_inc_not_zero(&cq->refcount))
+		cq = NULL;
+	xa_unlock_irqrestore(&dev->cq_xa, flags);
+
+	return cq;
+}
+
+static inline void erdma_cq_put(struct erdma_cq *cq)
 {
-	return (struct erdma_cq *)xa_load(&dev->cq_xa, id);
+	if (refcount_dec_and_test(&cq->refcount))
+		complete(&cq->free);
 }
 
 void erdma_qp_get(struct erdma_qp *qp);
-- 
2.31.1


             reply	other threads:[~2026-07-30 12:49 UTC|newest]

Thread overview: 2+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2026-07-30 12:43 Cheng Xu [this message]
2026-07-30 12:43 ` [PATCH 2/2] RDMA/erdma: Hold QP references for AE and CM processing Cheng Xu

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=20260730124357.12976-1-chengyou@linux.alibaba.com \
    --to=chengyou@linux.alibaba.com \
    --cc=KaiShen@linux.alibaba.com \
    --cc=jgg@ziepe.ca \
    --cc=leon@kernel.org \
    --cc=linux-rdma@vger.kernel.org \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox