v6.4 first rc RDMA pull request

Small rc bug fixes: - Fix 64K ARM page size support in bnxt_re and efa - bnxt_re fixes for a memory leak, incorrect error handling and a remove a bogus FW failure when running on a VF - Update MAINTAINERS for hns and efa - Fix two rxe regressions added this merge window in error unwind and incorrect spinlock primitives - hns gets a better algorithm for allocating page tables to avoid running out of resources, and a timeout adjustment - Fix a text case failure in hns - Use after free in irdma and fix incorrect construction of a WQE causing mis-execution -----BEGIN PGP SIGNATURE----- iHUEABYIAB0WIQRRRCHOFoQz/8F5bUaFwuHvBreFYQUCZHd8IwAKCRCFwuHvBreF YYOFAQDkEWwKF1+L+vvI0Do6gTPEb1D5KDRl3uasl4AAVl66UgEA8UulKNiZF/iJ L9NS8M4C/7WPMtPNFty3QjtSK5BXkwg= =FCJZ -----END PGP SIGNATURE----- Merge tag 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/rdma/rdma Pull rdma fixes from Jason Gunthorpe: - Fix 64K ARM page size support in bnxt_re and efa - bnxt_re fixes for a memory leak, incorrect error handling and a remove a bogus FW failure when running on a VF - Update MAINTAINERS for hns and efa - Fix two rxe regressions added this merge window in error unwind and incorrect spinlock primitives - hns gets a better algorithm for allocating page tables to avoid running out of resources, and a timeout adjustment - Fix a text case failure in hns - Use after free in irdma and fix incorrect construction of a WQE causing mis-execution * tag 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/rdma/rdma: RDMA/irdma: Fix Local Invalidate fencing RDMA/irdma: Prevent QP use after free MAINTAINERS: Update maintainer of Amazon EFA driver RDMA/bnxt_re: Do not enable congestion control on VFs RDMA/bnxt_re: Fix return value of bnxt_re_process_raw_qp_pkt_rx RDMA/bnxt_re: Fix a possible memory leak RDMA/hns: Modify the value of long message loopback slice RDMA/hns: Fix base address table allocation RDMA/hns: Fix timeout attr in query qp for HIP08 RDMA/efa: Fix unsupported page sizes in device RDMA/rxe: Convert spin_{lock_bh,unlock_bh} to spin_{lock_irqsave,unlock_irqrestore} RDMA/rxe: Fix double unlock in rxe_qp.c MAINTAINERS: Update maintainers of HiSilicon RoCE RDMA/bnxt_re: Fix the page_size used during the MR creation
2024-11-30 07:34:12 +08:00 · 2023-05-31 14:13:09 -04:00 · 2023-05-31 14:13:09 -04:00 · 884fe9da1b
commit 884fe9da1b
parent fd2186d1c7 5842d1d9c1
18 changed files with 176 additions and 99 deletions
--- a/5
+++ b/5
@ -956,7 +956,8 @@ F:	Documentation/networking/device_drivers/ethernet/amazon/ena.rst
 F:	drivers/net/ethernet/amazon/

 AMAZON RDMA EFA DRIVER
-M:	Gal Pressman <galpress@amazon.com>
+M:	Michael Margolin <mrgolin@amazon.com>
+R:	Gal Pressman <gal.pressman@linux.dev>
 R:	Yossi Leybovich <sleybo@amazon.com>
 L:	linux-rdma@vger.kernel.org
 S:	Supported
@ -9343,7 +9344,7 @@ F:	include/linux/hisi_acc_qm.h

 HISILICON ROCE DRIVER
 M:	Haoyue Xu <xuhaoyue1@hisilicon.com>
-M:	Wenpeng Liang <liangwenpeng@huawei.com>
+M:	Junxian Huang <huangjunxian6@hisilicon.com>
 L:	linux-rdma@vger.kernel.org
 S:	Maintained
 F:	Documentation/devicetree/bindings/infiniband/hisilicon-hns-roce.txt
--- a/drivers/infiniband/hw/bnxt_re/ib_verbs.c
+++ b/drivers/infiniband/hw/bnxt_re/ib_verbs.c
@ -3341,9 +3341,7 @@ static int bnxt_re_process_raw_qp_pkt_rx(struct bnxt_re_qp *gsi_qp,
 	udwr.remote_qkey = gsi_sqp->qplib_qp.qkey;

 	/* post data received  in the send queue */
-	rc = bnxt_re_post_send_shadow_qp(rdev, gsi_sqp, swr);
-
-	return 0;
+	return bnxt_re_post_send_shadow_qp(rdev, gsi_sqp, swr);
 }

 static void bnxt_re_process_res_rawqp1_wc(struct ib_wc *wc,
--- a/drivers/infiniband/hw/bnxt_re/main.c
+++ b/drivers/infiniband/hw/bnxt_re/main.c
@ -1336,6 +1336,10 @@ static void bnxt_re_setup_cc(struct bnxt_re_dev *rdev, bool enable)
 {
 	struct bnxt_qplib_cc_param cc_param = {};

+	/* Do not enable congestion control on VFs */
+	if (rdev->is_virtfn)
+		return;
+
 	/* Currently enabling only for GenP5 adapters */
 	if (!bnxt_qplib_is_chip_gen_p5(rdev->chip_ctx))
 		return;
--- a/drivers/infiniband/hw/bnxt_re/qplib_fp.c
+++ b/drivers/infiniband/hw/bnxt_re/qplib_fp.c
@ -2056,6 +2056,12 @@ int bnxt_qplib_create_cq(struct bnxt_qplib_res *res, struct bnxt_qplib_cq *cq)
 	u32 pg_sz_lvl;
 	int rc;

+	if (!cq->dpi) {
+		dev_err(&rcfw->pdev->dev,
+			"FP: CREATE_CQ failed due to NULL DPI\n");
+		return -EINVAL;
+	}
+
 	hwq_attr.res = res;
 	hwq_attr.depth = cq->max_wqe;
 	hwq_attr.stride = sizeof(struct cq_base);
@ -2069,11 +2075,6 @@ int bnxt_qplib_create_cq(struct bnxt_qplib_res *res, struct bnxt_qplib_cq *cq)
 				 CMDQ_BASE_OPCODE_CREATE_CQ,
 				 sizeof(req));

-	if (!cq->dpi) {
-		dev_err(&rcfw->pdev->dev,
-			"FP: CREATE_CQ failed due to NULL DPI\n");
-		return -EINVAL;
-	}
 	req.dpi = cpu_to_le32(cq->dpi->dpi);
 	req.cq_handle = cpu_to_le64(cq->cq_handle);
 	req.cq_size = cpu_to_le32(cq->hwq.max_elements);
--- a/drivers/infiniband/hw/bnxt_re/qplib_res.c
+++ b/drivers/infiniband/hw/bnxt_re/qplib_res.c
@ -215,17 +215,9 @@ int bnxt_qplib_alloc_init_hwq(struct bnxt_qplib_hwq *hwq,
 			return -EINVAL;
 		hwq_attr->sginfo->npages = npages;
 	} else {
-		unsigned long sginfo_num_pages = ib_umem_num_dma_blocks(
-			hwq_attr->sginfo->umem, hwq_attr->sginfo->pgsize);
-
+		npages = ib_umem_num_dma_blocks(hwq_attr->sginfo->umem,
+						hwq_attr->sginfo->pgsize);
 		hwq->is_user = true;
-		npages = sginfo_num_pages;
-		npages = (npages * PAGE_SIZE) /
-			  BIT_ULL(hwq_attr->sginfo->pgshft);
-		if ((sginfo_num_pages * PAGE_SIZE) %
-		     BIT_ULL(hwq_attr->sginfo->pgshft))
-			if (!npages)
-				npages++;
 	}

 	if (npages == MAX_PBL_LVL_0_PGS && !hwq_attr->sginfo->nopte) {
--- a/drivers/infiniband/hw/bnxt_re/qplib_sp.c
+++ b/drivers/infiniband/hw/bnxt_re/qplib_sp.c
@ -617,16 +617,15 @@ int bnxt_qplib_reg_mr(struct bnxt_qplib_res *res, struct bnxt_qplib_mrw *mr,
 		/* Free the hwq if it already exist, must be a rereg */
 		if (mr->hwq.max_elements)
 			bnxt_qplib_free_hwq(res, &mr->hwq);
-		/* Use system PAGE_SIZE */
 		hwq_attr.res = res;
 		hwq_attr.depth = pages;
-		hwq_attr.stride = buf_pg_size;
+		hwq_attr.stride = sizeof(dma_addr_t);
 		hwq_attr.type = HWQ_TYPE_MR;
 		hwq_attr.sginfo = &sginfo;
 		hwq_attr.sginfo->umem = umem;
 		hwq_attr.sginfo->npages = pages;
-		hwq_attr.sginfo->pgsize = PAGE_SIZE;
-		hwq_attr.sginfo->pgshft = PAGE_SHIFT;
+		hwq_attr.sginfo->pgsize = buf_pg_size;
+		hwq_attr.sginfo->pgshft = ilog2(buf_pg_size);
 		rc = bnxt_qplib_alloc_init_hwq(&mr->hwq, &hwq_attr);
 		if (rc) {
 			dev_err(&res->pdev->dev,
--- a/drivers/infiniband/hw/efa/efa_verbs.c
+++ b/drivers/infiniband/hw/efa/efa_verbs.c
@ -1403,7 +1403,7 @@ static int pbl_continuous_initialize(struct efa_dev *dev,
 */
 static int pbl_indirect_initialize(struct efa_dev *dev, struct pbl_context *pbl)
 {
-	u32 size_in_pages = DIV_ROUND_UP(pbl->pbl_buf_size_in_bytes, PAGE_SIZE);
+	u32 size_in_pages = DIV_ROUND_UP(pbl->pbl_buf_size_in_bytes, EFA_CHUNK_PAYLOAD_SIZE);
 	struct scatterlist *sgl;
 	int sg_dma_cnt, err;

--- a/drivers/infiniband/hw/hns/hns_roce_hw_v2.c
+++ b/drivers/infiniband/hw/hns/hns_roce_hw_v2.c
@ -4583,11 +4583,9 @@ static int modify_qp_init_to_rtr(struct ib_qp *ibqp,
 	mtu = ib_mtu_enum_to_int(ib_mtu);
 	if (WARN_ON(mtu <= 0))
 		return -EINVAL;
-#define MAX_LP_MSG_LEN 16384
-	/* MTU * (2 ^ LP_PKTN_INI) shouldn't be bigger than 16KB */
-	lp_pktn_ini = ilog2(MAX_LP_MSG_LEN / mtu);
-	if (WARN_ON(lp_pktn_ini >= 0xF))
-		return -EINVAL;
+#define MIN_LP_MSG_LEN 1024
+	/* mtu * (2 ^ lp_pktn_ini) should be in the range of 1024 to mtu */
+	lp_pktn_ini = ilog2(max(mtu, MIN_LP_MSG_LEN) / mtu);

 	if (attr_mask & IB_QP_PATH_MTU) {
 		hr_reg_write(context, QPC_MTU, ib_mtu);
@ -5012,7 +5010,6 @@ static int hns_roce_v2_set_abs_fields(struct ib_qp *ibqp,
 static bool check_qp_timeout_cfg_range(struct hns_roce_dev *hr_dev, u8 *timeout)
 {
 #define QP_ACK_TIMEOUT_MAX_HIP08 20
-#define QP_ACK_TIMEOUT_OFFSET 10
 #define QP_ACK_TIMEOUT_MAX 31

 	if (hr_dev->pci_dev->revision == PCI_REVISION_ID_HIP08) {
@ -5021,7 +5018,7 @@ static bool check_qp_timeout_cfg_range(struct hns_roce_dev *hr_dev, u8 *timeout)
 				   "local ACK timeout shall be 0 to 20.\n");
 			return false;
 		}
-		*timeout += QP_ACK_TIMEOUT_OFFSET;
+		*timeout += HNS_ROCE_V2_QP_ACK_TIMEOUT_OFS_HIP08;
 	} else if (hr_dev->pci_dev->revision > PCI_REVISION_ID_HIP08) {
 		if (*timeout > QP_ACK_TIMEOUT_MAX) {
 			ibdev_warn(&hr_dev->ib_dev,
@ -5307,6 +5304,18 @@ out:
 	return ret;
 }

+static u8 get_qp_timeout_attr(struct hns_roce_dev *hr_dev,
+			      struct hns_roce_v2_qp_context *context)
+{
+	u8 timeout;
+
+	timeout = (u8)hr_reg_read(context, QPC_AT);
+	if (hr_dev->pci_dev->revision == PCI_REVISION_ID_HIP08)
+		timeout -= HNS_ROCE_V2_QP_ACK_TIMEOUT_OFS_HIP08;
+
+	return timeout;
+}
+
 static int hns_roce_v2_query_qp(struct ib_qp *ibqp, struct ib_qp_attr *qp_attr,
 				int qp_attr_mask,
 				struct ib_qp_init_attr *qp_init_attr)
@ -5384,7 +5393,7 @@ static int hns_roce_v2_query_qp(struct ib_qp *ibqp, struct ib_qp_attr *qp_attr,
 	qp_attr->max_dest_rd_atomic = 1 << hr_reg_read(&context, QPC_RR_MAX);

 	qp_attr->min_rnr_timer = (u8)hr_reg_read(&context, QPC_MIN_RNR_TIME);
-	qp_attr->timeout = (u8)hr_reg_read(&context, QPC_AT);
+	qp_attr->timeout = get_qp_timeout_attr(hr_dev, &context);
 	qp_attr->retry_cnt = hr_reg_read(&context, QPC_RETRY_NUM_INIT);
 	qp_attr->rnr_retry = hr_reg_read(&context, QPC_RNR_NUM_INIT);

--- a/drivers/infiniband/hw/hns/hns_roce_hw_v2.h
+++ b/drivers/infiniband/hw/hns/hns_roce_hw_v2.h
@ -44,6 +44,8 @@
 #define HNS_ROCE_V2_MAX_XRCD_NUM		0x1000000
 #define HNS_ROCE_V2_RSV_XRCD_NUM		0

+#define HNS_ROCE_V2_QP_ACK_TIMEOUT_OFS_HIP08    10
+
 #define HNS_ROCE_V3_SCCC_SZ			64
 #define HNS_ROCE_V3_GMV_ENTRY_SZ		32

--- a/drivers/infiniband/hw/hns/hns_roce_mr.c
+++ b/drivers/infiniband/hw/hns/hns_roce_mr.c
@ -33,6 +33,7 @@

 #include <linux/vmalloc.h>
 #include <rdma/ib_umem.h>
+#include <linux/math.h>
 #include "hns_roce_device.h"
 #include "hns_roce_cmd.h"
 #include "hns_roce_hem.h"
@ -909,6 +910,44 @@ static int mtr_init_buf_cfg(struct hns_roce_dev *hr_dev,
 	return page_cnt;
 }

+static u64 cal_pages_per_l1ba(unsigned int ba_per_bt, unsigned int hopnum)
+{
+	return int_pow(ba_per_bt, hopnum - 1);
+}
+
+static unsigned int cal_best_bt_pg_sz(struct hns_roce_dev *hr_dev,
+				      struct hns_roce_mtr *mtr,
+				      unsigned int pg_shift)
+{
+	unsigned long cap = hr_dev->caps.page_size_cap;
+	struct hns_roce_buf_region *re;
+	unsigned int pgs_per_l1ba;
+	unsigned int ba_per_bt;
+	unsigned int ba_num;
+	int i;
+
+	for_each_set_bit_from(pg_shift, &cap, sizeof(cap) * BITS_PER_BYTE) {
+		if (!(BIT(pg_shift) & cap))
+			continue;
+
+		ba_per_bt = BIT(pg_shift) / BA_BYTE_LEN;
+		ba_num = 0;
+		for (i = 0; i < mtr->hem_cfg.region_count; i++) {
+			re = &mtr->hem_cfg.region[i];
+			if (re->hopnum == 0)
+				continue;
+
+			pgs_per_l1ba = cal_pages_per_l1ba(ba_per_bt, re->hopnum);
+			ba_num += DIV_ROUND_UP(re->count, pgs_per_l1ba);
+		}
+
+		if (ba_num <= ba_per_bt)
+			return pg_shift;
+	}
+
+	return 0;
+}
+
 static int mtr_alloc_mtt(struct hns_roce_dev *hr_dev, struct hns_roce_mtr *mtr,
 			 unsigned int ba_page_shift)
 {
@ -917,6 +956,10 @@ static int mtr_alloc_mtt(struct hns_roce_dev *hr_dev, struct hns_roce_mtr *mtr,

 	hns_roce_hem_list_init(&mtr->hem_list);
 	if (!cfg->is_direct) {
+		ba_page_shift = cal_best_bt_pg_sz(hr_dev, mtr, ba_page_shift);
+		if (!ba_page_shift)
+			return -ERANGE;
+
 		ret = hns_roce_hem_list_request(hr_dev, &mtr->hem_list,
 						cfg->region, cfg->region_count,
 						ba_page_shift);
--- a/drivers/infiniband/hw/irdma/verbs.c
+++ b/drivers/infiniband/hw/irdma/verbs.c
@ -522,11 +522,6 @@ static int irdma_destroy_qp(struct ib_qp *ibqp, struct ib_udata *udata)
 	if (!iwqp->user_mode)
 		cancel_delayed_work_sync(&iwqp->dwork_flush);

-	irdma_qp_rem_ref(&iwqp->ibqp);
-	wait_for_completion(&iwqp->free_qp);
-	irdma_free_lsmm_rsrc(iwqp);
-	irdma_cqp_qp_destroy_cmd(&iwdev->rf->sc_dev, &iwqp->sc_qp);
-
 	if (!iwqp->user_mode) {
 		if (iwqp->iwscq) {
 			irdma_clean_cqes(iwqp, iwqp->iwscq);
@ -534,6 +529,12 @@ static int irdma_destroy_qp(struct ib_qp *ibqp, struct ib_udata *udata)
 				irdma_clean_cqes(iwqp, iwqp->iwrcq);
 		}
 	}
+
+	irdma_qp_rem_ref(&iwqp->ibqp);
+	wait_for_completion(&iwqp->free_qp);
+	irdma_free_lsmm_rsrc(iwqp);
+	irdma_cqp_qp_destroy_cmd(&iwdev->rf->sc_dev, &iwqp->sc_qp);
+
 	irdma_remove_push_mmap_entries(iwqp);
 	irdma_free_qp_rsrc(iwqp);

@ -3291,6 +3292,7 @@ static int irdma_post_send(struct ib_qp *ibqp,
 			break;
 		case IB_WR_LOCAL_INV:
 			info.op_type = IRDMA_OP_TYPE_INV_STAG;
+			info.local_fence = info.read_fence;
 			info.op.inv_local_stag.target_stag = ib_wr->ex.invalidate_rkey;
 			err = irdma_uk_stag_local_invalidate(ukqp, &info, true);
 			break;
--- a/drivers/infiniband/sw/rxe/rxe_comp.c
+++ b/drivers/infiniband/sw/rxe/rxe_comp.c
@ -115,15 +115,16 @@ static enum ib_wc_opcode wr_to_wc_opcode(enum ib_wr_opcode opcode)
 void retransmit_timer(struct timer_list *t)
 {
 	struct rxe_qp *qp = from_timer(qp, t, retrans_timer);
+	unsigned long flags;

 	rxe_dbg_qp(qp, "retransmit timer fired\n");

-	spin_lock_bh(&qp->state_lock);
+	spin_lock_irqsave(&qp->state_lock, flags);
 	if (qp->valid) {
 		qp->comp.timeout = 1;
 		rxe_sched_task(&qp->comp.task);
 	}
-	spin_unlock_bh(&qp->state_lock);
+	spin_unlock_irqrestore(&qp->state_lock, flags);
 }

 void rxe_comp_queue_pkt(struct rxe_qp *qp, struct sk_buff *skb)
@ -481,11 +482,13 @@ static void do_complete(struct rxe_qp *qp, struct rxe_send_wqe *wqe)

 static void comp_check_sq_drain_done(struct rxe_qp *qp)
 {
-	spin_lock_bh(&qp->state_lock);
+	unsigned long flags;
+
+	spin_lock_irqsave(&qp->state_lock, flags);
 	if (unlikely(qp_state(qp) == IB_QPS_SQD)) {
 		if (qp->attr.sq_draining && qp->comp.psn == qp->req.psn) {
 			qp->attr.sq_draining = 0;
-			spin_unlock_bh(&qp->state_lock);
+			spin_unlock_irqrestore(&qp->state_lock, flags);

 			if (qp->ibqp.event_handler) {
 				struct ib_event ev;
@ -499,7 +502,7 @@ static void comp_check_sq_drain_done(struct rxe_qp *qp)
 			return;
 		}
 	}
-	spin_unlock_bh(&qp->state_lock);
+	spin_unlock_irqrestore(&qp->state_lock, flags);
 }

 static inline enum comp_state complete_ack(struct rxe_qp *qp,
@ -625,13 +628,15 @@ static void free_pkt(struct rxe_pkt_info *pkt)
 */
 static void reset_retry_timer(struct rxe_qp *qp)
 {
+	unsigned long flags;
+
 	if (qp_type(qp) == IB_QPT_RC && qp->qp_timeout_jiffies) {
-		spin_lock_bh(&qp->state_lock);
+		spin_lock_irqsave(&qp->state_lock, flags);
 		if (qp_state(qp) >= IB_QPS_RTS &&
 		    psn_compare(qp->req.psn, qp->comp.psn) > 0)
 			mod_timer(&qp->retrans_timer,
 				  jiffies + qp->qp_timeout_jiffies);
-		spin_unlock_bh(&qp->state_lock);
+		spin_unlock_irqrestore(&qp->state_lock, flags);
 	}
 }

@ -643,18 +648,19 @@ int rxe_completer(struct rxe_qp *qp)
 	struct rxe_pkt_info *pkt = NULL;
 	enum comp_state state;
 	int ret;
+	unsigned long flags;

-	spin_lock_bh(&qp->state_lock);
+	spin_lock_irqsave(&qp->state_lock, flags);
 	if (!qp->valid || qp_state(qp) == IB_QPS_ERR ||
 			  qp_state(qp) == IB_QPS_RESET) {
 		bool notify = qp->valid && (qp_state(qp) == IB_QPS_ERR);

 		drain_resp_pkts(qp);
 		flush_send_queue(qp, notify);
-		spin_unlock_bh(&qp->state_lock);
+		spin_unlock_irqrestore(&qp->state_lock, flags);
 		goto exit;
 	}
-	spin_unlock_bh(&qp->state_lock);
+	spin_unlock_irqrestore(&qp->state_lock, flags);

 	if (qp->comp.timeout) {
 		qp->comp.timeout_retry = 1;
--- a/drivers/infiniband/sw/rxe/rxe_net.c
+++ b/drivers/infiniband/sw/rxe/rxe_net.c
@ -412,15 +412,16 @@ int rxe_xmit_packet(struct rxe_qp *qp, struct rxe_pkt_info *pkt,
 	int err;
 	int is_request = pkt->mask & RXE_REQ_MASK;
 	struct rxe_dev *rxe = to_rdev(qp->ibqp.device);
+	unsigned long flags;

-	spin_lock_bh(&qp->state_lock);
+	spin_lock_irqsave(&qp->state_lock, flags);
 	if ((is_request && (qp_state(qp) < IB_QPS_RTS)) ||
 	    (!is_request && (qp_state(qp) < IB_QPS_RTR))) {
-		spin_unlock_bh(&qp->state_lock);
+		spin_unlock_irqrestore(&qp->state_lock, flags);
 		rxe_dbg_qp(qp, "Packet dropped. QP is not in ready state\n");
 		goto drop;
 	}
-	spin_unlock_bh(&qp->state_lock);
+	spin_unlock_irqrestore(&qp->state_lock, flags);

 	rxe_icrc_generate(skb, pkt);

--- a/drivers/infiniband/sw/rxe/rxe_qp.c
+++ b/drivers/infiniband/sw/rxe/rxe_qp.c
@ -300,6 +300,7 @@ int rxe_qp_from_init(struct rxe_dev *rxe, struct rxe_qp *qp, struct rxe_pd *pd,
 	struct rxe_cq *rcq = to_rcq(init->recv_cq);
 	struct rxe_cq *scq = to_rcq(init->send_cq);
 	struct rxe_srq *srq = init->srq ? to_rsrq(init->srq) : NULL;
+	unsigned long flags;

 	rxe_get(pd);
 	rxe_get(rcq);
@ -325,10 +326,10 @@ int rxe_qp_from_init(struct rxe_dev *rxe, struct rxe_qp *qp, struct rxe_pd *pd,
 	if (err)
 		goto err2;

-	spin_lock_bh(&qp->state_lock);
+	spin_lock_irqsave(&qp->state_lock, flags);
 	qp->attr.qp_state = IB_QPS_RESET;
 	qp->valid = 1;
-	spin_unlock_bh(&qp->state_lock);
+	spin_unlock_irqrestore(&qp->state_lock, flags);

 	return 0;

@ -492,24 +493,28 @@ static void rxe_qp_reset(struct rxe_qp *qp)
 /* move the qp to the error state */
 void rxe_qp_error(struct rxe_qp *qp)
 {
-	spin_lock_bh(&qp->state_lock);
+	unsigned long flags;
+
+	spin_lock_irqsave(&qp->state_lock, flags);
 	qp->attr.qp_state = IB_QPS_ERR;

 	/* drain work and packet queues */
 	rxe_sched_task(&qp->resp.task);
 	rxe_sched_task(&qp->comp.task);
 	rxe_sched_task(&qp->req.task);
-	spin_unlock_bh(&qp->state_lock);
+	spin_unlock_irqrestore(&qp->state_lock, flags);
 }

 static void rxe_qp_sqd(struct rxe_qp *qp, struct ib_qp_attr *attr,
 		       int mask)
 {
-	spin_lock_bh(&qp->state_lock);
+	unsigned long flags;
+
+	spin_lock_irqsave(&qp->state_lock, flags);
 	qp->attr.sq_draining = 1;
 	rxe_sched_task(&qp->comp.task);
 	rxe_sched_task(&qp->req.task);
-	spin_unlock_bh(&qp->state_lock);
+	spin_unlock_irqrestore(&qp->state_lock, flags);
 }

 /* caller should hold qp->state_lock */
@ -555,14 +560,16 @@ int rxe_qp_from_attr(struct rxe_qp *qp, struct ib_qp_attr *attr, int mask,
 		qp->attr.cur_qp_state = attr->qp_state;

 	if (mask & IB_QP_STATE) {
-		spin_lock_bh(&qp->state_lock);
+		unsigned long flags;
+
+		spin_lock_irqsave(&qp->state_lock, flags);
 		err = __qp_chk_state(qp, attr, mask);
 		if (!err) {
 			qp->attr.qp_state = attr->qp_state;
 			rxe_dbg_qp(qp, "state -> %s\n",
 					qps2str[attr->qp_state]);
 		}
-		spin_unlock_bh(&qp->state_lock);
+		spin_unlock_irqrestore(&qp->state_lock, flags);

 		if (err)
 			return err;
@ -688,6 +695,8 @@ int rxe_qp_from_attr(struct rxe_qp *qp, struct ib_qp_attr *attr, int mask,
 /* called by the query qp verb */
 int rxe_qp_to_attr(struct rxe_qp *qp, struct ib_qp_attr *attr, int mask)
 {
+	unsigned long flags;
+
 	*attr = qp->attr;

 	attr->rq_psn				= qp->resp.psn;
@ -708,12 +717,13 @@ int rxe_qp_to_attr(struct rxe_qp *qp, struct ib_qp_attr *attr, int mask)
 	/* Applications that get this state typically spin on it.
 	 * Yield the processor
 	 */
-	spin_lock_bh(&qp->state_lock);
+	spin_lock_irqsave(&qp->state_lock, flags);
 	if (qp->attr.sq_draining) {
-		spin_unlock_bh(&qp->state_lock);
+		spin_unlock_irqrestore(&qp->state_lock, flags);
 		cond_resched();
+	} else {
+		spin_unlock_irqrestore(&qp->state_lock, flags);
 	}
-	spin_unlock_bh(&qp->state_lock);

 	return 0;
 }
@ -736,10 +746,11 @@ int rxe_qp_chk_destroy(struct rxe_qp *qp)
 static void rxe_qp_do_cleanup(struct work_struct *work)
 {
 	struct rxe_qp *qp = container_of(work, typeof(*qp), cleanup_work.work);
+	unsigned long flags;

-	spin_lock_bh(&qp->state_lock);
+	spin_lock_irqsave(&qp->state_lock, flags);
 	qp->valid = 0;
-	spin_unlock_bh(&qp->state_lock);
+	spin_unlock_irqrestore(&qp->state_lock, flags);
 	qp->qp_timeout_jiffies = 0;

 	if (qp_type(qp) == IB_QPT_RC) {
--- a/drivers/infiniband/sw/rxe/rxe_recv.c
+++ b/drivers/infiniband/sw/rxe/rxe_recv.c
@ -14,6 +14,7 @@ static int check_type_state(struct rxe_dev *rxe, struct rxe_pkt_info *pkt,
 			    struct rxe_qp *qp)
 {
 	unsigned int pkt_type;
+	unsigned long flags;

 	if (unlikely(!qp->valid))
 		return -EINVAL;
@ -38,19 +39,19 @@ static int check_type_state(struct rxe_dev *rxe, struct rxe_pkt_info *pkt,
 		return -EINVAL;
 	}

-	spin_lock_bh(&qp->state_lock);
+	spin_lock_irqsave(&qp->state_lock, flags);
 	if (pkt->mask & RXE_REQ_MASK) {
 		if (unlikely(qp_state(qp) < IB_QPS_RTR)) {
-			spin_unlock_bh(&qp->state_lock);
+			spin_unlock_irqrestore(&qp->state_lock, flags);
 			return -EINVAL;
 		}
 	} else {
 		if (unlikely(qp_state(qp) < IB_QPS_RTS)) {
-			spin_unlock_bh(&qp->state_lock);
+			spin_unlock_irqrestore(&qp->state_lock, flags);
 			return -EINVAL;
 		}
 	}
-	spin_unlock_bh(&qp->state_lock);
+	spin_unlock_irqrestore(&qp->state_lock, flags);

 	return 0;
 }
--- a/drivers/infiniband/sw/rxe/rxe_req.c
+++ b/drivers/infiniband/sw/rxe/rxe_req.c
@ -99,17 +99,18 @@ static void req_retry(struct rxe_qp *qp)
 void rnr_nak_timer(struct timer_list *t)
 {
 	struct rxe_qp *qp = from_timer(qp, t, rnr_nak_timer);
+	unsigned long flags;

 	rxe_dbg_qp(qp, "nak timer fired\n");

-	spin_lock_bh(&qp->state_lock);
+	spin_lock_irqsave(&qp->state_lock, flags);
 	if (qp->valid) {
 		/* request a send queue retry */
 		qp->req.need_retry = 1;
 		qp->req.wait_for_rnr_timer = 0;
 		rxe_sched_task(&qp->req.task);
 	}
-	spin_unlock_bh(&qp->state_lock);
+	spin_unlock_irqrestore(&qp->state_lock, flags);
 }

 static void req_check_sq_drain_done(struct rxe_qp *qp)
@ -118,8 +119,9 @@ static void req_check_sq_drain_done(struct rxe_qp *qp)
 	unsigned int index;
 	unsigned int cons;
 	struct rxe_send_wqe *wqe;
+	unsigned long flags;

-	spin_lock_bh(&qp->state_lock);
+	spin_lock_irqsave(&qp->state_lock, flags);
 	if (qp_state(qp) == IB_QPS_SQD) {
 		q = qp->sq.queue;
 		index = qp->req.wqe_index;
@ -140,7 +142,7 @@ static void req_check_sq_drain_done(struct rxe_qp *qp)
 				break;

 			qp->attr.sq_draining = 0;
-			spin_unlock_bh(&qp->state_lock);
+			spin_unlock_irqrestore(&qp->state_lock, flags);

 			if (qp->ibqp.event_handler) {
 				struct ib_event ev;
@ -154,7 +156,7 @@ static void req_check_sq_drain_done(struct rxe_qp *qp)
 			return;
 		} while (0);
 	}
-	spin_unlock_bh(&qp->state_lock);
+	spin_unlock_irqrestore(&qp->state_lock, flags);
 }

 static struct rxe_send_wqe *__req_next_wqe(struct rxe_qp *qp)
@ -173,6 +175,7 @@ static struct rxe_send_wqe *__req_next_wqe(struct rxe_qp *qp)
 static struct rxe_send_wqe *req_next_wqe(struct rxe_qp *qp)
 {
 	struct rxe_send_wqe *wqe;
+	unsigned long flags;

 	req_check_sq_drain_done(qp);

@ -180,13 +183,13 @@ static struct rxe_send_wqe *req_next_wqe(struct rxe_qp *qp)
 	if (wqe == NULL)
 		return NULL;

-	spin_lock_bh(&qp->state_lock);
+	spin_lock_irqsave(&qp->state_lock, flags);
 	if (unlikely((qp_state(qp) == IB_QPS_SQD) &&
 		     (wqe->state != wqe_state_processing))) {
-		spin_unlock_bh(&qp->state_lock);
+		spin_unlock_irqrestore(&qp->state_lock, flags);
 		return NULL;
 	}
-	spin_unlock_bh(&qp->state_lock);
+	spin_unlock_irqrestore(&qp->state_lock, flags);

 	wqe->mask = wr_opcode_mask(wqe->wr.opcode, qp);
 	return wqe;
@ -676,16 +679,17 @@ int rxe_requester(struct rxe_qp *qp)
 	struct rxe_queue *q = qp->sq.queue;
 	struct rxe_ah *ah;
 	struct rxe_av *av;
+	unsigned long flags;

-	spin_lock_bh(&qp->state_lock);
+	spin_lock_irqsave(&qp->state_lock, flags);
 	if (unlikely(!qp->valid)) {
-		spin_unlock_bh(&qp->state_lock);
+		spin_unlock_irqrestore(&qp->state_lock, flags);
 		goto exit;
 	}

 	if (unlikely(qp_state(qp) == IB_QPS_ERR)) {
 		wqe = __req_next_wqe(qp);
-		spin_unlock_bh(&qp->state_lock);
+		spin_unlock_irqrestore(&qp->state_lock, flags);
 		if (wqe)
 			goto err;
 		else
@ -700,10 +704,10 @@ int rxe_requester(struct rxe_qp *qp)
 		qp->req.wait_psn = 0;
 		qp->req.need_retry = 0;
 		qp->req.wait_for_rnr_timer = 0;
-		spin_unlock_bh(&qp->state_lock);
+		spin_unlock_irqrestore(&qp->state_lock, flags);
 		goto exit;
 	}
-	spin_unlock_bh(&qp->state_lock);
+	spin_unlock_irqrestore(&qp->state_lock, flags);

 	/* we come here if the retransmit timer has fired
 	 * or if the rnr timer has fired. If the retransmit
--- a/drivers/infiniband/sw/rxe/rxe_resp.c
+++ b/drivers/infiniband/sw/rxe/rxe_resp.c
@ -1047,6 +1047,7 @@ static enum resp_states do_complete(struct rxe_qp *qp,
 	struct ib_uverbs_wc *uwc = &cqe.uibwc;
 	struct rxe_recv_wqe *wqe = qp->resp.wqe;
 	struct rxe_dev *rxe = to_rdev(qp->ibqp.device);
+	unsigned long flags;

 	if (!wqe)
 		goto finish;
@ -1137,12 +1138,12 @@ static enum resp_states do_complete(struct rxe_qp *qp,
 		return RESPST_ERR_CQ_OVERFLOW;

 finish:
-	spin_lock_bh(&qp->state_lock);
+	spin_lock_irqsave(&qp->state_lock, flags);
 	if (unlikely(qp_state(qp) == IB_QPS_ERR)) {
-		spin_unlock_bh(&qp->state_lock);
+		spin_unlock_irqrestore(&qp->state_lock, flags);
 		return RESPST_CHK_RESOURCE;
 	}
-	spin_unlock_bh(&qp->state_lock);
+	spin_unlock_irqrestore(&qp->state_lock, flags);

 	if (unlikely(!pkt))
 		return RESPST_DONE;
@ -1468,18 +1469,19 @@ int rxe_responder(struct rxe_qp *qp)
 	enum resp_states state;
 	struct rxe_pkt_info *pkt = NULL;
 	int ret;
+	unsigned long flags;

-	spin_lock_bh(&qp->state_lock);
+	spin_lock_irqsave(&qp->state_lock, flags);
 	if (!qp->valid || qp_state(qp) == IB_QPS_ERR ||
 			  qp_state(qp) == IB_QPS_RESET) {
 		bool notify = qp->valid && (qp_state(qp) == IB_QPS_ERR);

 		drain_req_pkts(qp);
 		flush_recv_queue(qp, notify);
-		spin_unlock_bh(&qp->state_lock);
+		spin_unlock_irqrestore(&qp->state_lock, flags);
 		goto exit;
 	}
-	spin_unlock_bh(&qp->state_lock);
+	spin_unlock_irqrestore(&qp->state_lock, flags);

 	qp->resp.aeth_syndrome = AETH_ACK_UNLIMITED;

--- a/drivers/infiniband/sw/rxe/rxe_verbs.c
+++ b/drivers/infiniband/sw/rxe/rxe_verbs.c
@ -904,10 +904,10 @@ static int rxe_post_send_kernel(struct rxe_qp *qp,
 	if (!err)
 		rxe_sched_task(&qp->req.task);

-	spin_lock_bh(&qp->state_lock);
+	spin_lock_irqsave(&qp->state_lock, flags);
 	if (qp_state(qp) == IB_QPS_ERR)
 		rxe_sched_task(&qp->comp.task);
-	spin_unlock_bh(&qp->state_lock);
+	spin_unlock_irqrestore(&qp->state_lock, flags);

 	return err;
 }
@ -917,22 +917,23 @@ static int rxe_post_send(struct ib_qp *ibqp, const struct ib_send_wr *wr,
 {
 	struct rxe_qp *qp = to_rqp(ibqp);
 	int err;
+	unsigned long flags;

-	spin_lock_bh(&qp->state_lock);
+	spin_lock_irqsave(&qp->state_lock, flags);
 	/* caller has already called destroy_qp */
 	if (WARN_ON_ONCE(!qp->valid)) {
-		spin_unlock_bh(&qp->state_lock);
+		spin_unlock_irqrestore(&qp->state_lock, flags);
 		rxe_err_qp(qp, "qp has been destroyed");
 		return -EINVAL;
 	}

 	if (unlikely(qp_state(qp) < IB_QPS_RTS)) {
-		spin_unlock_bh(&qp->state_lock);
+		spin_unlock_irqrestore(&qp->state_lock, flags);
 		*bad_wr = wr;
 		rxe_err_qp(qp, "qp not ready to send");
 		return -EINVAL;
 	}
-	spin_unlock_bh(&qp->state_lock);
+	spin_unlock_irqrestore(&qp->state_lock, flags);

 	if (qp->is_user) {
 		/* Utilize process context to do protocol processing */
@ -1008,22 +1009,22 @@ static int rxe_post_recv(struct ib_qp *ibqp, const struct ib_recv_wr *wr,
 	struct rxe_rq *rq = &qp->rq;
 	unsigned long flags;

-	spin_lock_bh(&qp->state_lock);
+	spin_lock_irqsave(&qp->state_lock, flags);
 	/* caller has already called destroy_qp */
 	if (WARN_ON_ONCE(!qp->valid)) {
-		spin_unlock_bh(&qp->state_lock);
+		spin_unlock_irqrestore(&qp->state_lock, flags);
 		rxe_err_qp(qp, "qp has been destroyed");
 		return -EINVAL;
 	}

 	/* see C10-97.2.1 */
 	if (unlikely((qp_state(qp) < IB_QPS_INIT))) {
-		spin_unlock_bh(&qp->state_lock);
+		spin_unlock_irqrestore(&qp->state_lock, flags);
 		*bad_wr = wr;
 		rxe_dbg_qp(qp, "qp not ready to post recv");
 		return -EINVAL;
 	}
-	spin_unlock_bh(&qp->state_lock);
+	spin_unlock_irqrestore(&qp->state_lock, flags);

 	if (unlikely(qp->srq)) {
 		*bad_wr = wr;
@ -1044,10 +1045,10 @@ static int rxe_post_recv(struct ib_qp *ibqp, const struct ib_recv_wr *wr,

 	spin_unlock_irqrestore(&rq->producer_lock, flags);

-	spin_lock_bh(&qp->state_lock);
+	spin_lock_irqsave(&qp->state_lock, flags);
 	if (qp_state(qp) == IB_QPS_ERR)
 		rxe_sched_task(&qp->resp.task);
-	spin_unlock_bh(&qp->state_lock);
+	spin_unlock_irqrestore(&qp->state_lock, flags);

 	return err;
 }