From 00baeade709fb66da647e8327e8398bb532e30f7 Mon Sep 17 00:00:00 2001 From: Cheng Xu Date: Fri, 28 Aug 2026 11:03:44 +0800 Subject: [PATCH] RDMA/erdma: Use IRQ-safe XArray helpers for QP and CQ tables Locked QP and CQ lookups from EQ interrupts can deadlock with create-path XArray updates. If an interrupt arrives while the create path holds the plain xa_lock, the lookup spins forever trying to acquire the same lock. Use IRQ-safe XArray helpers for all QP and CQ create-path updates, including the GSI QP store and error paths. Initialize both arrays with XA_FLAGS_LOCK_IRQ so sleeping allocations preserve interrupt state. Fixes: 98df2aee1459 ("RDMA/erdma: Hold CQ references when processing EQ events") Fixes: a52eeff32024 ("RDMA/erdma: Hold QP references for AE and CM processing") Signed-off-by: Cheng Xu Link: https://patch.msgid.link/20260828030344.88021-1-chengyou@linux.alibaba.com Signed-off-by: Leon Romanovsky --- drivers/infiniband/hw/erdma/erdma_main.c | 4 ++-- drivers/infiniband/hw/erdma/erdma_verbs.c | 18 +++++++++--------- 2 files changed, 11 insertions(+), 11 deletions(-) diff --git a/drivers/infiniband/hw/erdma/erdma_main.c b/drivers/infiniband/hw/erdma/erdma_main.c index 7e87a815e853..445182c6bc5d 100644 --- a/drivers/infiniband/hw/erdma/erdma_main.c +++ b/drivers/infiniband/hw/erdma/erdma_main.c @@ -572,8 +572,8 @@ static int erdma_ib_device_add(struct pci_dev *pdev) INIT_LIST_HEAD(&dev->cep_list); spin_lock_init(&dev->lock); - xa_init_flags(&dev->qp_xa, XA_FLAGS_ALLOC1); - xa_init_flags(&dev->cq_xa, XA_FLAGS_ALLOC1); + xa_init_flags(&dev->qp_xa, XA_FLAGS_ALLOC1 | XA_FLAGS_LOCK_IRQ); + xa_init_flags(&dev->cq_xa, XA_FLAGS_ALLOC1 | XA_FLAGS_LOCK_IRQ); dev->next_alloc_cqn = 1; dev->next_alloc_qpn = 1; diff --git a/drivers/infiniband/hw/erdma/erdma_verbs.c b/drivers/infiniband/hw/erdma/erdma_verbs.c index 65b1af1e6623..f18b88bba281 100644 --- a/drivers/infiniband/hw/erdma/erdma_verbs.c +++ b/drivers/infiniband/hw/erdma/erdma_verbs.c @@ -1021,15 +1021,15 @@ int erdma_create_qp(struct ib_qp *ibqp, struct ib_qp_init_attr *attrs, init_completion(&qp->safe_free); if (qp->ibqp.qp_type == IB_QPT_GSI) { - old_entry = xa_store(&dev->qp_xa, 1, qp, GFP_KERNEL); + old_entry = xa_store_irq(&dev->qp_xa, 1, qp, GFP_KERNEL); if (xa_is_err(old_entry)) ret = xa_err(old_entry); else qp->ibqp.qp_num = 1; } else { - ret = xa_alloc_cyclic(&dev->qp_xa, &qp->ibqp.qp_num, qp, - XA_LIMIT(1, dev->attrs.max_qp - 1), - &dev->next_alloc_qpn, GFP_KERNEL); + ret = xa_alloc_cyclic_irq(&dev->qp_xa, &qp->ibqp.qp_num, qp, + XA_LIMIT(1, dev->attrs.max_qp - 1), + &dev->next_alloc_qpn, GFP_KERNEL); } if (ret < 0) { @@ -1089,7 +1089,7 @@ int erdma_create_qp(struct ib_qp *ibqp, struct ib_qp_init_attr *attrs, else free_kernel_qp(qp); err_out_xa: - xa_erase(&dev->qp_xa, QP_ID(qp)); + xa_erase_irq(&dev->qp_xa, QP_ID(qp)); err_out: return ret; } @@ -1993,9 +1993,9 @@ int erdma_create_cq(struct ib_cq *ibcq, const struct ib_cq_init_attr *attr, 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), - &dev->next_alloc_cqn, GFP_KERNEL); + ret = xa_alloc_cyclic_irq(&dev->cq_xa, &cq->cqn, cq, + XA_LIMIT(1, dev->attrs.max_cq - 1), + &dev->next_alloc_cqn, GFP_KERNEL); if (ret < 0) return ret; @@ -2041,7 +2041,7 @@ int erdma_create_cq(struct ib_cq *ibcq, const struct ib_cq_init_attr *attr, } err_out_xa: - xa_erase(&dev->cq_xa, cq->cqn); + xa_erase_irq(&dev->cq_xa, cq->cqn); return ret; }