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

Pull rdma fixes from Jason Gunthorpe:
 "Several miscellaneous fixes. A lot of bnxt_re activity, there will be
  more rc patches there coming.

   - Many bnxt_re bug fixes - Memory leaks, kasn, NULL pointer deref,
     soft lockups, error unwinding and some small functional issues

   - Error unwind bug in rdma netlink

   - Two issues with incorrect VLAN detection for iWarp

   - skb_splice_from_iter() splat in siw

   - Give SRP slab caches unique names to resolve the merge window
     WARN_ON regression"

* tag 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/rdma/rdma:
  RDMA/bnxt_re: Fix the GID table length
  RDMA/bnxt_re: Fix a bug while setting up Level-2 PBL pages
  RDMA/bnxt_re: Change the sequence of updating the CQ toggle value
  RDMA/bnxt_re: Fix an error path in bnxt_re_add_device
  RDMA/bnxt_re: Avoid CPU lockups due fifo occupancy check loop
  RDMA/bnxt_re: Fix a possible NULL pointer dereference
  RDMA/bnxt_re: Return more meaningful error
  RDMA/bnxt_re: Fix incorrect dereference of srq in async event
  RDMA/bnxt_re: Fix out of bound check
  RDMA/bnxt_re: Fix the max CQ WQEs for older adapters
  RDMA/srpt: Make slab cache names unique
  RDMA/irdma: Fix misspelling of "accept*"
  RDMA/cxgb4: Fix RDMA_CM_EVENT_UNREACHABLE error for iWARP
  RDMA/siw: Add sendpage_ok() check to disable MSG_SPLICE_PAGES
  RDMA/core: Fix ENODEV error for iWARP test over vlan
  RDMA/nldev: Fix NULL pointer dereferences issue in rdma_nl_notify_event
  RDMA/bnxt_re: Fix the max WQEs used in Static WQE mode
  RDMA/bnxt_re: Add a check for memory allocation
  RDMA/bnxt_re: Fix incorrect AVID type in WQE structure
  RDMA/bnxt_re: Fix a possible memory leak

15 files changed, 133 insertions, 61 deletions

diff --git a/drivers/infiniband/core/addr.c b/drivers/infiniband/core/addr.c
index be0743dac3ff..c4cf26f1d149 100644
--- a/drivers/infiniband/core/addr.c
+++ b/drivers/infiniband/core/addr.c
@@ -269,6 +269,8 @@ rdma_find_ndev_for_src_ip_rcu(struct net *net, const struct sockaddr *src_in)
 		break;
 #endif
 	}
+	if (!ret && dev && is_vlan_dev(dev))
+		dev = vlan_dev_real_dev(dev);
 	return ret ? ERR_PTR(ret) : dev;
 }
 
diff --git a/drivers/infiniband/core/nldev.c b/drivers/infiniband/core/nldev.c
index 39f89a4b8649..7dc8e2ec62cc 100644
--- a/drivers/infiniband/core/nldev.c
+++ b/drivers/infiniband/core/nldev.c
@@ -2816,6 +2816,8 @@ int rdma_nl_notify_event(struct ib_device *device, u32 port_num,
 	nlh = nlmsg_put(skb, 0, 0,
 			RDMA_NL_GET_TYPE(RDMA_NL_NLDEV, RDMA_NLDEV_CMD_MONITOR),
 			0, 0);
+	if (!nlh)
+		goto err_free;
 
 	switch (type) {
 	case RDMA_REGISTER_EVENT:
diff --git a/drivers/infiniband/hw/bnxt_re/hw_counters.c b/drivers/infiniband/hw/bnxt_re/hw_counters.c
index 128651c01595..1e63f8091748 100644
--- a/drivers/infiniband/hw/bnxt_re/hw_counters.c
+++ b/drivers/infiniband/hw/bnxt_re/hw_counters.c
@@ -366,7 +366,7 @@ int bnxt_re_ib_get_hw_stats(struct ib_device *ibdev,
 				goto done;
 			}
 		}
-		if (rdev->pacing.dbr_pacing)
+		if (rdev->pacing.dbr_pacing && bnxt_qplib_is_chip_gen_p5_p7(rdev->chip_ctx))
 			bnxt_re_copy_db_pacing_stats(rdev, stats);
 	}
 
diff --git a/drivers/infiniband/hw/bnxt_re/ib_verbs.c b/drivers/infiniband/hw/bnxt_re/ib_verbs.c
index 460f33914825..e66ae9f22c71 100644
--- a/drivers/infiniband/hw/bnxt_re/ib_verbs.c
+++ b/drivers/infiniband/hw/bnxt_re/ib_verbs.c
@@ -1307,7 +1307,11 @@ static int bnxt_re_init_sq_attr(struct bnxt_re_qp *qp,
 			0 : BNXT_QPLIB_RESERVED_QP_WRS;
 		entries = bnxt_re_init_depth(entries + diff + 1, uctx);
 		sq->max_wqe = min_t(u32, entries, dev_attr->max_qp_wqes + diff + 1);
-		sq->max_sw_wqe = bnxt_qplib_get_depth(sq, qplqp->wqe_mode, true);
+		if (qplqp->wqe_mode == BNXT_QPLIB_WQE_MODE_VARIABLE)
+			sq->max_sw_wqe = bnxt_qplib_get_depth(sq, qplqp->wqe_mode, true);
+		else
+			sq->max_sw_wqe = sq->max_wqe;
+
 	}
 	sq->q_full_delta = diff + 1;
 	/*
diff --git a/drivers/infiniband/hw/bnxt_re/main.c b/drivers/infiniband/hw/bnxt_re/main.c
index 777068de4bbc..6715c96a3eee 100644
--- a/drivers/infiniband/hw/bnxt_re/main.c
+++ b/drivers/infiniband/hw/bnxt_re/main.c
@@ -188,8 +188,11 @@ static int bnxt_re_setup_chip_ctx(struct bnxt_re_dev *rdev)
 
 	bnxt_re_set_db_offset(rdev);
 	rc = bnxt_qplib_map_db_bar(&rdev->qplib_res);
-	if (rc)
+	if (rc) {
+		kfree(rdev->chip_ctx);
+		rdev->chip_ctx = NULL;
 		return rc;
+	}
 
 	if (bnxt_qplib_determine_atomics(en_dev->pdev))
 		ibdev_info(&rdev->ibdev,
@@ -531,6 +534,7 @@ static bool is_dbr_fifo_full(struct bnxt_re_dev *rdev)
 static void __wait_for_fifo_occupancy_below_th(struct bnxt_re_dev *rdev)
 {
 	struct bnxt_qplib_db_pacing_data *pacing_data = rdev->qplib_res.pacing_data;
+	u32 retry_fifo_check = 1000;
 	u32 fifo_occup;
 
 	/* loop shouldn't run infintely as the occupancy usually goes
@@ -544,6 +548,14 @@ static void __wait_for_fifo_occupancy_below_th(struct bnxt_re_dev *rdev)
 
 		if (fifo_occup < pacing_data->pacing_th)
 			break;
+		if (!retry_fifo_check--) {
+			dev_info_once(rdev_to_dev(rdev),
+				      "%s: fifo_occup = 0x%xfifo_max_depth = 0x%x pacing_th = 0x%x\n",
+				      __func__, fifo_occup, pacing_data->fifo_max_depth,
+					pacing_data->pacing_th);
+			break;
+		}
+
 	}
 }
 
@@ -957,7 +969,7 @@ static int bnxt_re_register_ib(struct bnxt_re_dev *rdev)
 	return ib_register_device(ibdev, "bnxt_re%d", &rdev->en_dev->pdev->dev);
 }
 
-static struct bnxt_re_dev *bnxt_re_dev_add(struct bnxt_aux_priv *aux_priv,
+static struct bnxt_re_dev *bnxt_re_dev_add(struct auxiliary_device *adev,
 					   struct bnxt_en_dev *en_dev)
 {
 	struct bnxt_re_dev *rdev;
@@ -973,6 +985,7 @@ static struct bnxt_re_dev *bnxt_re_dev_add(struct bnxt_aux_priv *aux_priv,
 	rdev->nb.notifier_call = NULL;
 	rdev->netdev = en_dev->net;
 	rdev->en_dev = en_dev;
+	rdev->adev = adev;
 	rdev->id = rdev->en_dev->pdev->devfn;
 	INIT_LIST_HEAD(&rdev->qp_list);
 	mutex_init(&rdev->qp_lock);
@@ -1025,12 +1038,15 @@ static int bnxt_re_handle_unaffi_async_event(struct creq_func_event
 static int bnxt_re_handle_qp_async_event(struct creq_qp_event *qp_event,
 					 struct bnxt_re_qp *qp)
 {
-	struct bnxt_re_srq *srq = container_of(qp->qplib_qp.srq, struct bnxt_re_srq,
-					       qplib_srq);
 	struct creq_qp_error_notification *err_event;
+	struct bnxt_re_srq *srq = NULL;
 	struct ib_event event = {};
 	unsigned int flags;
 
+	if (qp->qplib_qp.srq)
+		srq =  container_of(qp->qplib_qp.srq, struct bnxt_re_srq,
+				    qplib_srq);
+
 	if (qp->qplib_qp.state == CMDQ_MODIFY_QP_NEW_STATE_ERR &&
 	    rdma_is_kernel_res(&qp->ib_qp.res)) {
 		flags = bnxt_re_lock_cqs(qp);
@@ -1258,15 +1274,9 @@ static int bnxt_re_cqn_handler(struct bnxt_qplib_nq *nq,
 {
 	struct bnxt_re_cq *cq = container_of(handle, struct bnxt_re_cq,
 					     qplib_cq);
-	u32 *cq_ptr;
 
-	if (cq->ib_cq.comp_handler) {
-		if (cq->uctx_cq_page) {
-			cq_ptr = (u32 *)cq->uctx_cq_page;
-			*cq_ptr = cq->qplib_cq.toggle;
-		}
+	if (cq->ib_cq.comp_handler)
 		(*cq->ib_cq.comp_handler)(&cq->ib_cq, cq->ib_cq.cq_context);
-	}
 
 	return 0;
 }
@@ -1823,7 +1833,6 @@ static void bnxt_re_update_en_info_rdev(struct bnxt_re_dev *rdev,
 	 */
 	rtnl_lock();
 	en_info->rdev = rdev;
-	rdev->adev = adev;
 	rtnl_unlock();
 }
 
@@ -1840,7 +1849,7 @@ static int bnxt_re_add_device(struct auxiliary_device *adev, u8 op_type)
 	en_dev = en_info->en_dev;
 
 
-	rdev = bnxt_re_dev_add(aux_priv, en_dev);
+	rdev = bnxt_re_dev_add(adev, en_dev);
 	if (!rdev || !rdev_to_dev(rdev)) {
 		rc = -ENOMEM;
 		goto exit;
@@ -1865,12 +1874,14 @@ static int bnxt_re_add_device(struct auxiliary_device *adev, u8 op_type)
 		rdev->nb.notifier_call = NULL;
 		pr_err("%s: Cannot register to netdevice_notifier",
 		       ROCE_DRV_MODULE_NAME);
-		return rc;
+		goto re_dev_unreg;
 	}
 	bnxt_re_setup_cc(rdev, true);
 
 	return 0;
 
+re_dev_unreg:
+	ib_unregister_device(&rdev->ibdev);
 re_dev_uninit:
 	bnxt_re_update_en_info_rdev(NULL, en_info, adev);
 	bnxt_re_dev_uninit(rdev, BNXT_RE_COMPLETE_REMOVE);
@@ -2014,15 +2025,7 @@ static int bnxt_re_probe(struct auxiliary_device *adev,
 	auxiliary_set_drvdata(adev, en_info);
 
 	rc = bnxt_re_add_device(adev, BNXT_RE_COMPLETE_INIT);
-	if (rc)
-		goto err;
 	mutex_unlock(&bnxt_re_mutex);
-	return 0;
-
-err:
-	mutex_unlock(&bnxt_re_mutex);
-	bnxt_re_remove(adev);
-
 	return rc;
 }
 
diff --git a/drivers/infiniband/hw/bnxt_re/qplib_fp.c b/drivers/infiniband/hw/bnxt_re/qplib_fp.c
index 42e98e5f94cb..2ebcb2de962b 100644
--- a/drivers/infiniband/hw/bnxt_re/qplib_fp.c
+++ b/drivers/infiniband/hw/bnxt_re/qplib_fp.c
@@ -327,6 +327,7 @@ static void bnxt_qplib_service_nq(struct tasklet_struct *t)
 		case NQ_BASE_TYPE_CQ_NOTIFICATION:
 		{
 			struct nq_cn *nqcne = (struct nq_cn *)nqe;
+			struct bnxt_re_cq *cq_p;
 
 			q_handle = le32_to_cpu(nqcne->cq_handle_low);
 			q_handle |= (u64)le32_to_cpu(nqcne->cq_handle_high)
@@ -337,6 +338,10 @@ static void bnxt_qplib_service_nq(struct tasklet_struct *t)
 			cq->toggle = (le16_to_cpu(nqe->info10_type) &
 					NQ_CN_TOGGLE_MASK) >> NQ_CN_TOGGLE_SFT;
 			cq->dbinfo.toggle = cq->toggle;
+			cq_p = container_of(cq, struct bnxt_re_cq, qplib_cq);
+			if (cq_p->uctx_cq_page)
+				*((u32 *)cq_p->uctx_cq_page) = cq->toggle;
+
 			bnxt_qplib_armen_db(&cq->dbinfo,
 					    DBC_DBC_TYPE_CQ_ARMENA);
 			spin_lock_bh(&cq->compl_lock);
diff --git a/drivers/infiniband/hw/bnxt_re/qplib_fp.h b/drivers/infiniband/hw/bnxt_re/qplib_fp.h
index b62df8701950..820611a23943 100644
--- a/drivers/infiniband/hw/bnxt_re/qplib_fp.h
+++ b/drivers/infiniband/hw/bnxt_re/qplib_fp.h
@@ -170,7 +170,7 @@ struct bnxt_qplib_swqe {
 			};
 			u32		q_key;
 			u32		dst_qp;
-			u16		avid;
+			u32		avid;
 		} send;
 
 		/* Send Raw Ethernet and QP1 */
diff --git a/drivers/infiniband/hw/bnxt_re/qplib_rcfw.c b/drivers/infiniband/hw/bnxt_re/qplib_rcfw.c
index 3ffaef0c2651..7294221b3316 100644
--- a/drivers/infiniband/hw/bnxt_re/qplib_rcfw.c
+++ b/drivers/infiniband/hw/bnxt_re/qplib_rcfw.c
@@ -525,7 +525,7 @@ static int __bnxt_qplib_rcfw_send_message(struct bnxt_qplib_rcfw *rcfw,
 		/* failed with status */
 		dev_err(&rcfw->pdev->dev, "cmdq[%#x]=%#x status %#x\n",
 			cookie, opcode, evnt->status);
-		rc = -EFAULT;
+		rc = -EIO;
 	}
 
 	return rc;
diff --git a/drivers/infiniband/hw/bnxt_re/qplib_res.c b/drivers/infiniband/hw/bnxt_re/qplib_res.c
index dfc943fab87b..96ceec1e8199 100644
--- a/drivers/infiniband/hw/bnxt_re/qplib_res.c
+++ b/drivers/infiniband/hw/bnxt_re/qplib_res.c
@@ -244,6 +244,8 @@ int bnxt_qplib_alloc_init_hwq(struct bnxt_qplib_hwq *hwq,
 			sginfo.pgsize = npde * pg_size;
 			sginfo.npages = 1;
 			rc = __alloc_pbl(res, &hwq->pbl[PBL_LVL_0], &sginfo);
+			if (rc)
+				goto fail;
 
 			/* Alloc PBL pages */
 			sginfo.npages = npbl;
@@ -255,22 +257,9 @@ int bnxt_qplib_alloc_init_hwq(struct bnxt_qplib_hwq *hwq,
 			dst_virt_ptr =
 				(dma_addr_t **)hwq->pbl[PBL_LVL_0].pg_arr;
 			src_phys_ptr = hwq->pbl[PBL_LVL_1].pg_map_arr;
-			if (hwq_attr->type == HWQ_TYPE_MR) {
-			/* For MR it is expected that we supply only 1 contigous
-			 * page i.e only 1 entry in the PDL that will contain
-			 * all the PBLs for the user supplied memory region
-			 */
-				for (i = 0; i < hwq->pbl[PBL_LVL_1].pg_count;
-				     i++)
-					dst_virt_ptr[0][i] = src_phys_ptr[i] |
-						flag;
-			} else {
-				for (i = 0; i < hwq->pbl[PBL_LVL_1].pg_count;
-				     i++)
-					dst_virt_ptr[PTR_PG(i)][PTR_IDX(i)] =
-						src_phys_ptr[i] |
-						PTU_PDE_VALID;
-			}
+			for (i = 0; i < hwq->pbl[PBL_LVL_1].pg_count; i++)
+				dst_virt_ptr[0][i] = src_phys_ptr[i] | flag;
+
 			/* Alloc or init PTEs */
 			rc = __alloc_pbl(res, &hwq->pbl[PBL_LVL_2],
 					 hwq_attr->sginfo);
diff --git a/drivers/infiniband/hw/bnxt_re/qplib_sp.c b/drivers/infiniband/hw/bnxt_re/qplib_sp.c
index 4f75e7e5bcf7..e29fbbdab9fd 100644
--- a/drivers/infiniband/hw/bnxt_re/qplib_sp.c
+++ b/drivers/infiniband/hw/bnxt_re/qplib_sp.c
@@ -140,6 +140,8 @@ int bnxt_qplib_get_dev_attr(struct bnxt_qplib_rcfw *rcfw,
 			    min_t(u32, sb->max_sge_var_wqe, BNXT_VAR_MAX_SGE) : 6;
 	attr->max_cq = le32_to_cpu(sb->max_cq);
 	attr->max_cq_wqes = le32_to_cpu(sb->max_cqe);
+	if (!bnxt_qplib_is_chip_gen_p7(rcfw->res->cctx))
+		attr->max_cq_wqes = min_t(u32, BNXT_QPLIB_MAX_CQ_WQES, attr->max_cq_wqes);
 	attr->max_cq_sges = attr->max_qp_sges;
 	attr->max_mr = le32_to_cpu(sb->max_mr);
 	attr->max_mw = le32_to_cpu(sb->max_mw);
@@ -157,7 +159,14 @@ int bnxt_qplib_get_dev_attr(struct bnxt_qplib_rcfw *rcfw,
 	if (!bnxt_qplib_is_chip_gen_p7(rcfw->res->cctx))
 		attr->l2_db_size = (sb->l2_db_space_size + 1) *
 				    (0x01 << RCFW_DBR_BASE_PAGE_SHIFT);
-	attr->max_sgid = BNXT_QPLIB_NUM_GIDS_SUPPORTED;
+	/*
+	 * Read the max gid supported by HW.
+	 * For each entry in HW  GID in HW table, we consume 2
+	 * GID entries in the kernel GID table.  So max_gid reported
+	 * to stack can be up to twice the value reported by the HW, up to 256 gids.
+	 */
+	attr->max_sgid = le32_to_cpu(sb->max_gid);
+	attr->max_sgid = min_t(u32, BNXT_QPLIB_NUM_GIDS_SUPPORTED, 2 * attr->max_sgid);
 	attr->dev_cap_flags = le16_to_cpu(sb->dev_cap_flags);
 	attr->dev_cap_flags2 = le16_to_cpu(sb->dev_cap_ext_flags_2);
 
diff --git a/drivers/infiniband/hw/bnxt_re/qplib_sp.h b/drivers/infiniband/hw/bnxt_re/qplib_sp.h
index acd9c14a31c4..ecf3f45fea74 100644
--- a/drivers/infiniband/hw/bnxt_re/qplib_sp.h
+++ b/drivers/infiniband/hw/bnxt_re/qplib_sp.h
@@ -56,6 +56,7 @@ struct bnxt_qplib_dev_attr {
 	u32				max_qp_wqes;
 	u32				max_qp_sges;
 	u32				max_cq;
+#define BNXT_QPLIB_MAX_CQ_WQES          0xfffff
 	u32				max_cq_wqes;
 	u32				max_cq_sges;
 	u32				max_mr;
diff --git a/drivers/infiniband/hw/cxgb4/cm.c b/drivers/infiniband/hw/cxgb4/cm.c
index b3757c6a0457..8d753e6e0c71 100644
--- a/drivers/infiniband/hw/cxgb4/cm.c
+++ b/drivers/infiniband/hw/cxgb4/cm.c
@@ -2086,7 +2086,7 @@ static int import_ep(struct c4iw_ep *ep, int iptype, __u8 *peer_ip,
 	err = -ENOMEM;
 	if (n->dev->flags & IFF_LOOPBACK) {
 		if (iptype == 4)
-			pdev = ip_dev_find(&init_net, *(__be32 *)peer_ip);
+			pdev = __ip_dev_find(&init_net, *(__be32 *)peer_ip, false);
 		else if (IS_ENABLED(CONFIG_IPV6))
 			for_each_netdev(&init_net, pdev) {
 				if (ipv6_chk_addr(&init_net,
@@ -2101,12 +2101,12 @@ static int import_ep(struct c4iw_ep *ep, int iptype, __u8 *peer_ip,
 			err = -ENODEV;
 			goto out;
 		}
+		if (is_vlan_dev(pdev))
+			pdev = vlan_dev_real_dev(pdev);
 		ep->l2t = cxgb4_l2t_get(cdev->rdev.lldi.l2t,
 					n, pdev, rt_tos2priority(tos));
-		if (!ep->l2t) {
-			dev_put(pdev);
+		if (!ep->l2t)
 			goto out;
-		}
 		ep->mtu = pdev->mtu;
 		ep->tx_chan = cxgb4_port_chan(pdev);
 		ep->smac_idx = ((struct port_info *)netdev_priv(pdev))->smt_idx;
@@ -2119,7 +2119,6 @@ static int import_ep(struct c4iw_ep *ep, int iptype, __u8 *peer_ip,
 		ep->rss_qid = cdev->rdev.lldi.rxq_ids[
 			cxgb4_port_idx(pdev) * step];
 		set_tcp_window(ep, (struct port_info *)netdev_priv(pdev));
-		dev_put(pdev);
 	} else {
 		pdev = get_real_dev(n->dev);
 		ep->l2t = cxgb4_l2t_get(cdev->rdev.lldi.l2t,
diff --git a/drivers/infiniband/hw/irdma/cm.c b/drivers/infiniband/hw/irdma/cm.c
index 36bb7e5ce638..ce8d821bdad8 100644
--- a/drivers/infiniband/hw/irdma/cm.c
+++ b/drivers/infiniband/hw/irdma/cm.c
@@ -3631,7 +3631,7 @@ void irdma_free_lsmm_rsrc(struct irdma_qp *iwqp)
 /**
  * irdma_accept - registered call for connection to be accepted
  * @cm_id: cm information for passive connection
- * @conn_param: accpet parameters
+ * @conn_param: accept parameters
  */
 int irdma_accept(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
 {
diff --git a/drivers/infiniband/sw/siw/siw_qp_tx.c b/drivers/infiniband/sw/siw/siw_qp_tx.c
index 64ad9e0895bd..a034264c5669 100644
--- a/drivers/infiniband/sw/siw/siw_qp_tx.c
+++ b/drivers/infiniband/sw/siw/siw_qp_tx.c
@@ -331,6 +331,8 @@ static int siw_tcp_sendpages(struct socket *s, struct page **page, int offset,
 			msg.msg_flags &= ~MSG_MORE;
 
 		tcp_rate_check_app_limited(sk);
+		if (!sendpage_ok(page[i]))
+			msg.msg_flags &= ~MSG_SPLICE_PAGES;
 		bvec_set_page(&bvec, page[i], bytes, offset);
 		iov_iter_bvec(&msg.msg_iter, ITER_SOURCE, &bvec, 1, size);
 
diff --git a/drivers/infiniband/ulp/srpt/ib_srpt.c b/drivers/infiniband/ulp/srpt/ib_srpt.c
index 9632afbd727b..5dfb4644446b 100644
--- a/drivers/infiniband/ulp/srpt/ib_srpt.c
+++ b/drivers/infiniband/ulp/srpt/ib_srpt.c
@@ -68,6 +68,8 @@ MODULE_LICENSE("Dual BSD/GPL");
 static u64 srpt_service_guid;
 static DEFINE_SPINLOCK(srpt_dev_lock);	/* Protects srpt_dev_list. */
 static LIST_HEAD(srpt_dev_list);	/* List of srpt_device structures. */
+static DEFINE_MUTEX(srpt_mc_mutex);	/* Protects srpt_memory_caches. */
+static DEFINE_XARRAY(srpt_memory_caches); /* See also srpt_memory_cache_entry */
 
 static unsigned srp_max_req_size = DEFAULT_MAX_REQ_SIZE;
 module_param(srp_max_req_size, int, 0444);
@@ -105,6 +107,63 @@ static void srpt_recv_done(struct ib_cq *cq, struct ib_wc *wc);
 static void srpt_send_done(struct ib_cq *cq, struct ib_wc *wc);
 static void srpt_process_wait_list(struct srpt_rdma_ch *ch);
 
+/* Type of the entries in srpt_memory_caches. */
+struct srpt_memory_cache_entry {
+	refcount_t ref;
+	struct kmem_cache *c;
+};
+
+static struct kmem_cache *srpt_cache_get(unsigned int object_size)
+{
+	struct srpt_memory_cache_entry *e;
+	char name[32];
+	void *res;
+
+	guard(mutex)(&srpt_mc_mutex);
+	e = xa_load(&srpt_memory_caches, object_size);
+	if (e) {
+		refcount_inc(&e->ref);
+		return e->c;
+	}
+	snprintf(name, sizeof(name), "srpt-%u", object_size);
+	e = kmalloc(sizeof(*e), GFP_KERNEL);
+	if (!e)
+		return NULL;
+	refcount_set(&e->ref, 1);
+	e->c = kmem_cache_create(name, object_size, /*align=*/512, 0, NULL);
+	if (!e->c)
+		goto free_entry;
+	res = xa_store(&srpt_memory_caches, object_size, e, GFP_KERNEL);
+	if (xa_is_err(res))
+		goto destroy_cache;
+	return e->c;
+
+destroy_cache:
+	kmem_cache_destroy(e->c);
+
+free_entry:
+	kfree(e);
+	return NULL;
+}
+
+static void srpt_cache_put(struct kmem_cache *c)
+{
+	struct srpt_memory_cache_entry *e = NULL;
+	unsigned long object_size;
+
+	guard(mutex)(&srpt_mc_mutex);
+	xa_for_each(&srpt_memory_caches, object_size, e)
+		if (e->c == c)
+			break;
+	if (WARN_ON_ONCE(!e))
+		return;
+	if (!refcount_dec_and_test(&e->ref))
+		return;
+	WARN_ON_ONCE(xa_erase(&srpt_memory_caches, object_size) != e);
+	kmem_cache_destroy(e->c);
+	kfree(e);
+}
+
 /*
  * The only allowed channel state changes are those that change the channel
  * state into a state with a higher numerical value. Hence the new > prev test.
@@ -2119,13 +2178,13 @@ static void srpt_release_channel_work(struct work_struct *w)
 			     ch->sport->sdev, ch->rq_size,
 			     ch->rsp_buf_cache, DMA_TO_DEVICE);
 
-	kmem_cache_destroy(ch->rsp_buf_cache);
+	srpt_cache_put(ch->rsp_buf_cache);
 
 	srpt_free_ioctx_ring((struct srpt_ioctx **)ch->ioctx_recv_ring,
 			     sdev, ch->rq_size,
 			     ch->req_buf_cache, DMA_FROM_DEVICE);
 
-	kmem_cache_destroy(ch->req_buf_cache);
+	srpt_cache_put(ch->req_buf_cache);
 
 	kref_put(&ch->kref, srpt_free_ch);
 }
@@ -2245,8 +2304,7 @@ static int srpt_cm_req_recv(struct srpt_device *const sdev,
 	INIT_LIST_HEAD(&ch->cmd_wait_list);
 	ch->max_rsp_size = ch->sport->port_attrib.srp_max_rsp_size;
 
-	ch->rsp_buf_cache = kmem_cache_create("srpt-rsp-buf", ch->max_rsp_size,
-					      512, 0, NULL);
+	ch->rsp_buf_cache = srpt_cache_get(ch->max_rsp_size);
 	if (!ch->rsp_buf_cache)
 		goto free_ch;
 
@@ -2280,8 +2338,7 @@ static int srpt_cm_req_recv(struct srpt_device *const sdev,
 		alignment_offset = round_up(imm_data_offset, 512) -
 			imm_data_offset;
 		req_sz = alignment_offset + imm_data_offset + srp_max_req_size;
-		ch->req_buf_cache = kmem_cache_create("srpt-req-buf", req_sz,
-						      512, 0, NULL);
+		ch->req_buf_cache = srpt_cache_get(req_sz);
 		if (!ch->req_buf_cache)
 			goto free_rsp_ring;
 
@@ -2478,7 +2535,7 @@ free_recv_ring:
 			     ch->req_buf_cache, DMA_FROM_DEVICE);
 
 free_recv_cache:
-	kmem_cache_destroy(ch->req_buf_cache);
+	srpt_cache_put(ch->req_buf_cache);
 
 free_rsp_ring:
 	srpt_free_ioctx_ring((struct srpt_ioctx **)ch->ioctx_ring,
@@ -2486,7 +2543,7 @@ free_rsp_ring:
 			     ch->rsp_buf_cache, DMA_TO_DEVICE);
 
 free_rsp_cache:
-	kmem_cache_destroy(ch->rsp_buf_cache);
+	srpt_cache_put(ch->rsp_buf_cache);
 
 free_ch:
 	if (rdma_cm_id)
@@ -3055,7 +3112,7 @@ static void srpt_free_srq(struct srpt_device *sdev)
 	srpt_free_ioctx_ring((struct srpt_ioctx **)sdev->ioctx_ring, sdev,
 			     sdev->srq_size, sdev->req_buf_cache,
 			     DMA_FROM_DEVICE);
-	kmem_cache_destroy(sdev->req_buf_cache);
+	srpt_cache_put(sdev->req_buf_cache);
 	sdev->srq = NULL;
 }
 
@@ -3082,8 +3139,7 @@ static int srpt_alloc_srq(struct srpt_device *sdev)
 	pr_debug("create SRQ #wr= %d max_allow=%d dev= %s\n", sdev->srq_size,
 		 sdev->device->attrs.max_srq_wr, dev_name(&device->dev));
 
-	sdev->req_buf_cache = kmem_cache_create("srpt-srq-req-buf",
-						srp_max_req_size, 0, 0, NULL);
+	sdev->req_buf_cache = srpt_cache_get(srp_max_req_size);
 	if (!sdev->req_buf_cache)
 		goto free_srq;
 
@@ -3105,7 +3161,7 @@ static int srpt_alloc_srq(struct srpt_device *sdev)
 	return 0;
 
 free_cache:
-	kmem_cache_destroy(sdev->req_buf_cache);
+	srpt_cache_put(sdev->req_buf_cache);
 
 free_srq:
 	ib_destroy_srq(srq);