From a45bc17b360d75fac9ced85e99fda14bf38b4dc3 Mon Sep 17 00:00:00 2001 From: Devesh Sharma Date: Mon, 26 Feb 2018 01:51:37 -0800 Subject: RDMA/bnxt_re: Unconditionly fence non wire memory operations HW requires an unconditonal fence for all non-wire memory operations through SQ. This guarantees the completions of these memory operations. Signed-off-by: Devesh Sharma Signed-off-by: Selvin Xavier Signed-off-by: Jason Gunthorpe --- drivers/infiniband/hw/bnxt_re/ib_verbs.c | 15 +++++++++++---- 1 file changed, 11 insertions(+), 4 deletions(-) (limited to 'drivers/infiniband/hw/bnxt_re/ib_verbs.c') diff --git a/drivers/infiniband/hw/bnxt_re/ib_verbs.c b/drivers/infiniband/hw/bnxt_re/ib_verbs.c index 643174d949a8..755f1ccd82bb 100644 --- a/drivers/infiniband/hw/bnxt_re/ib_verbs.c +++ b/drivers/infiniband/hw/bnxt_re/ib_verbs.c @@ -2227,10 +2227,13 @@ static int bnxt_re_build_inv_wqe(struct ib_send_wr *wr, wqe->type = BNXT_QPLIB_SWQE_TYPE_LOCAL_INV; wqe->local_inv.inv_l_key = wr->ex.invalidate_rkey; + /* Need unconditional fence for local invalidate + * opcode to work as expected. + */ + wqe->flags |= BNXT_QPLIB_SWQE_FLAGS_UC_FENCE; + if (wr->send_flags & IB_SEND_SIGNALED) wqe->flags |= BNXT_QPLIB_SWQE_FLAGS_SIGNAL_COMP; - if (wr->send_flags & IB_SEND_FENCE) - wqe->flags |= BNXT_QPLIB_SWQE_FLAGS_UC_FENCE; if (wr->send_flags & IB_SEND_SOLICITED) wqe->flags |= BNXT_QPLIB_SWQE_FLAGS_SOLICIT_EVENT; @@ -2251,8 +2254,12 @@ static int bnxt_re_build_reg_wqe(struct ib_reg_wr *wr, wqe->frmr.levels = qplib_frpl->hwq.level + 1; wqe->type = BNXT_QPLIB_SWQE_TYPE_REG_MR; - if (wr->wr.send_flags & IB_SEND_FENCE) - wqe->flags |= BNXT_QPLIB_SWQE_FLAGS_UC_FENCE; + /* Need unconditional fence for reg_mr + * opcode to function as expected. + */ + + wqe->flags |= BNXT_QPLIB_SWQE_FLAGS_UC_FENCE; + if (wr->wr.send_flags & IB_SEND_SIGNALED) wqe->flags |= BNXT_QPLIB_SWQE_FLAGS_SIGNAL_COMP; -- cgit v1.2.1 From 942c9b6ca8de5b7ad675e9b2e0e964449c10c18a Mon Sep 17 00:00:00 2001 From: Selvin Xavier Date: Mon, 5 Mar 2018 21:49:28 -0800 Subject: RDMA/bnxt_re: Avoid Hard lockup during error CQE processing Hitting the following hardlockup due to a race condition in error CQE processing. [26146.879798] bnxt_en 0000:04:00.0: QPLIB: FP: CQ Processed Req [26146.886346] bnxt_en 0000:04:00.0: QPLIB: wr_id[1251] = 0x0 with status 0xa [26156.350935] NMI watchdog: Watchdog detected hard LOCKUP on cpu 4 [26156.357470] Modules linked in: nfsd auth_rpcgss nfs_acl lockd grace [26156.447957] CPU: 4 PID: 3413 Comm: kworker/4:1H Kdump: loaded [26156.457994] Hardware name: Dell Inc. PowerEdge R430/0CN7X8, [26156.466390] Workqueue: ib-comp-wq ib_cq_poll_work [ib_core] [26156.472639] Call Trace: [26156.475379] [] dump_stack+0x19/0x1b [26156.481833] [] watchdog_overflow_callback+0x135/0x140 [26156.489341] [] __perf_event_overflow+0x57/0x100 [26156.496256] [] perf_event_overflow+0x14/0x20 [26156.502887] [] intel_pmu_handle_irq+0x220/0x510 [26156.509813] [] perf_event_nmi_handler+0x31/0x50 [26156.516738] [] nmi_handle.isra.0+0x8c/0x150 [26156.523273] [] do_nmi+0x218/0x460 [26156.528834] [] end_repeat_nmi+0x1e/0x7e [26156.534980] [] ? native_queued_spin_lock_slowpath+0x1d0/0x200 [26156.543268] [] ? native_queued_spin_lock_slowpath+0x1d0/0x200 [26156.551556] [] ? native_queued_spin_lock_slowpath+0x1d0/0x200 [26156.559842] [] queued_spin_lock_slowpath+0xb/0xf [26156.567555] [] _raw_spin_lock+0x20/0x30 [26156.573696] [] bnxt_qplib_lock_buddy_cq+0x31/0x40 [bnxt_re] [26156.581789] [] bnxt_qplib_poll_cq+0x43a/0xf10 [bnxt_re] [26156.589493] [] bnxt_re_poll_cq+0x9b/0x760 [bnxt_re] The issue happens if RQ poll_cq or SQ poll_cq or Async error event tries to put the error QP in flush list. Since SQ and RQ of each error qp are added to two different flush list, we need to protect it using locks of corresponding CQs. Difference in order of acquiring the lock in SQ poll_cq and RQ poll_cq can cause a hard lockup. Revisits the locking strategy and removes the usage of qplib_cq.hwq.lock. Instead of this lock, introduces qplib_cq.flush_lock to handle addition/deletion of QPs in flush list. Also, always invoke the flush_lock in order (SQ CQ lock first and then RQ CQ lock) to avoid any potential deadlock. Other than the poll_cq context, the movement of QP to/from flush list can be done in modify_qp context or from an async error event from HW. Synchronize these operations using the bnxt_re verbs layer CQ locks. To achieve this, adds a call back to the HW abstraction layer(qplib) to bnxt_re ib_verbs layer in case of async error event. Also, removes the buddy cq functions as it is no longer required. Signed-off-by: Sriharsha Basavapatna Signed-off-by: Somnath Kotur Signed-off-by: Devesh Sharma Signed-off-by: Selvin Xavier Signed-off-by: Jason Gunthorpe --- drivers/infiniband/hw/bnxt_re/ib_verbs.c | 11 ++++++++--- 1 file changed, 8 insertions(+), 3 deletions(-) (limited to 'drivers/infiniband/hw/bnxt_re/ib_verbs.c') diff --git a/drivers/infiniband/hw/bnxt_re/ib_verbs.c b/drivers/infiniband/hw/bnxt_re/ib_verbs.c index 755f1ccd82bb..0dd75f449872 100644 --- a/drivers/infiniband/hw/bnxt_re/ib_verbs.c +++ b/drivers/infiniband/hw/bnxt_re/ib_verbs.c @@ -785,7 +785,7 @@ int bnxt_re_query_ah(struct ib_ah *ib_ah, struct rdma_ah_attr *ah_attr) return 0; } -static unsigned long bnxt_re_lock_cqs(struct bnxt_re_qp *qp) +unsigned long bnxt_re_lock_cqs(struct bnxt_re_qp *qp) __acquires(&qp->scq->cq_lock) __acquires(&qp->rcq->cq_lock) { unsigned long flags; @@ -799,8 +799,8 @@ static unsigned long bnxt_re_lock_cqs(struct bnxt_re_qp *qp) return flags; } -static void bnxt_re_unlock_cqs(struct bnxt_re_qp *qp, - unsigned long flags) +void bnxt_re_unlock_cqs(struct bnxt_re_qp *qp, + unsigned long flags) __releases(&qp->scq->cq_lock) __releases(&qp->rcq->cq_lock) { if (qp->rcq != qp->scq) @@ -1606,6 +1606,7 @@ int bnxt_re_modify_qp(struct ib_qp *ib_qp, struct ib_qp_attr *qp_attr, int status; union ib_gid sgid; struct ib_gid_attr sgid_attr; + unsigned int flags; u8 nw_type; qp->qplib_qp.modify_flags = 0; @@ -1634,14 +1635,18 @@ int bnxt_re_modify_qp(struct ib_qp *ib_qp, struct ib_qp_attr *qp_attr, dev_dbg(rdev_to_dev(rdev), "Move QP = %p to flush list\n", qp); + flags = bnxt_re_lock_cqs(qp); bnxt_qplib_add_flush_qp(&qp->qplib_qp); + bnxt_re_unlock_cqs(qp, flags); } if (!qp->sumem && qp->qplib_qp.state == CMDQ_MODIFY_QP_NEW_STATE_RESET) { dev_dbg(rdev_to_dev(rdev), "Move QP = %p out of flush list\n", qp); + flags = bnxt_re_lock_cqs(qp); bnxt_qplib_clean_qp(&qp->qplib_qp); + bnxt_re_unlock_cqs(qp, flags); } } if (qp_attr_mask & IB_QP_EN_SQD_ASYNC_NOTIFY) { -- cgit v1.2.1 From bd8602ca42f6d8b25392a8f2a0e7be34e1206618 Mon Sep 17 00:00:00 2001 From: Arnd Bergmann Date: Tue, 20 Feb 2018 21:56:27 +0100 Subject: infiniband: bnxt_re: use BIT_ULL() for 64-bit bit masks On 32-bit targets, we otherwise get a warning about an impossible constant integer expression: In file included from include/linux/kernel.h:11, from include/linux/interrupt.h:6, from drivers/infiniband/hw/bnxt_re/ib_verbs.c:39: drivers/infiniband/hw/bnxt_re/ib_verbs.c: In function 'bnxt_re_query_device': include/linux/bitops.h:7:24: error: left shift count >= width of type [-Werror=shift-count-overflow] #define BIT(nr) (1UL << (nr)) ^~ drivers/infiniband/hw/bnxt_re/bnxt_re.h:61:34: note: in expansion of macro 'BIT' #define BNXT_RE_MAX_MR_SIZE_HIGH BIT(39) ^~~ drivers/infiniband/hw/bnxt_re/bnxt_re.h:62:30: note: in expansion of macro 'BNXT_RE_MAX_MR_SIZE_HIGH' #define BNXT_RE_MAX_MR_SIZE BNXT_RE_MAX_MR_SIZE_HIGH ^~~~~~~~~~~~~~~~~~~~~~~~ drivers/infiniband/hw/bnxt_re/ib_verbs.c:149:25: note: in expansion of macro 'BNXT_RE_MAX_MR_SIZE' ib_attr->max_mr_size = BNXT_RE_MAX_MR_SIZE; ^~~~~~~~~~~~~~~~~~~ Fixes: 872f3578241d ("RDMA/bnxt_re: Add support for MRs with Huge pages") Signed-off-by: Arnd Bergmann Signed-off-by: Jason Gunthorpe --- drivers/infiniband/hw/bnxt_re/ib_verbs.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'drivers/infiniband/hw/bnxt_re/ib_verbs.c') diff --git a/drivers/infiniband/hw/bnxt_re/ib_verbs.c b/drivers/infiniband/hw/bnxt_re/ib_verbs.c index 0dd75f449872..8301d7e5fa8c 100644 --- a/drivers/infiniband/hw/bnxt_re/ib_verbs.c +++ b/drivers/infiniband/hw/bnxt_re/ib_verbs.c @@ -3598,7 +3598,7 @@ struct ib_mr *bnxt_re_reg_user_mr(struct ib_pd *ib_pd, u64 start, u64 length, int umem_pgs, page_shift, rc; if (length > BNXT_RE_MAX_MR_SIZE) { - dev_err(rdev_to_dev(rdev), "MR Size: %lld > Max supported:%ld\n", + dev_err(rdev_to_dev(rdev), "MR Size: %lld > Max supported:%lld\n", length, BNXT_RE_MAX_MR_SIZE); return ERR_PTR(-ENOMEM); } -- cgit v1.2.1