From 08c7f09356e45d093d1867c7a3c6ac6526e2f98b Mon Sep 17 00:00:00 2001 From: Selvin Xavier Date: Sun, 7 May 2023 11:29:29 -0700 Subject: [PATCH 01/14] RDMA/bnxt_re: Fix the page_size used during the MR creation Driver populates the list of pages used for Memory region wrongly when page size is more than system page size. This is causing a failure when some of the applications that creates MR with page size as 2M. Since HW can support multiple page sizes, pass the correct page size while creating the MR. Also, driver need not adjust the number of pages when HW Queues are created with user memory. It should work with the number of dma blocks returned by ib_umem_num_dma_blocks. Fix this calculation also. Fixes: 0c4dcd602817 ("RDMA/bnxt_re: Refactor hardware queue memory allocation") Fixes: f6919d56388c ("RDMA/bnxt_re: Code refactor while populating user MRs") Link: https://lore.kernel.org/r/1683484169-9539-1-git-send-email-selvin.xavier@broadcom.com Signed-off-by: Kalesh AP Signed-off-by: Kashyap Desai Signed-off-by: Selvin Xavier Signed-off-by: Jason Gunthorpe --- drivers/infiniband/hw/bnxt_re/qplib_res.c | 12 ++---------- drivers/infiniband/hw/bnxt_re/qplib_sp.c | 7 +++---- 2 files changed, 5 insertions(+), 14 deletions(-) diff --git a/drivers/infiniband/hw/bnxt_re/qplib_res.c b/drivers/infiniband/hw/bnxt_re/qplib_res.c index 126d4f26f75a..81b0c5e879f9 100644 --- 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) { diff --git a/drivers/infiniband/hw/bnxt_re/qplib_sp.c b/drivers/infiniband/hw/bnxt_re/qplib_sp.c index 1714a1e23113..b967a17a44be 100644 --- 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, From 78b6a9a3f445e121ca08195084206cb8faa6999f Mon Sep 17 00:00:00 2001 From: Haoyue Xu Date: Sat, 6 May 2023 15:06:04 +0800 Subject: [PATCH 02/14] MAINTAINERS: Update maintainers of HiSilicon RoCE Wenpeng has moved to other technical areas, and Junxian will take over his responsibilities in maintaining this module. Link: https://lore.kernel.org/r/20230506070604.2982542-1-xuhaoyue1@hisilicon.com Signed-off-by: Haoyue Xu Signed-off-by: Jason Gunthorpe --- MAINTAINERS | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/MAINTAINERS b/MAINTAINERS index 7e0b87d5aa2e..c56043737848 100644 --- a/MAINTAINERS +++ b/MAINTAINERS @@ -9360,7 +9360,7 @@ F: drivers/crypto/hisilicon/zip/ HISILICON ROCE DRIVER M: Haoyue Xu -M: Wenpeng Liang +M: Junxian Huang L: linux-rdma@vger.kernel.org S: Maintained F: Documentation/devicetree/bindings/infiniband/hisilicon-hns-roce.txt From 17eabd6a044b57b2c1b5459eb9d11af3ea909e63 Mon Sep 17 00:00:00 2001 From: Bob Pearson Date: Mon, 15 May 2023 15:10:57 -0500 Subject: [PATCH 03/14] RDMA/rxe: Fix double unlock in rxe_qp.c A recent patch can cause a double spin_unlock_bh() in rxe_qp_to_attr() at line 715 in rxe_qp.c. Move the 2nd unlock into the if statement. Fixes: f605f26ea196 ("RDMA/rxe: Protect QP state with qp->state_lock") Link: https://lore.kernel.org/r/20230515201056.1591140-1-rpearsonhpe@gmail.com Reported-by: Dan Carpenter Closes: https://lore.kernel.org/r/27773078-40ce-414f-8b97-781954da9f25@kili.mountain Signed-off-by: Bob Pearson Signed-off-by: Jason Gunthorpe --- drivers/infiniband/sw/rxe/rxe_qp.c | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-) diff --git a/drivers/infiniband/sw/rxe/rxe_qp.c b/drivers/infiniband/sw/rxe/rxe_qp.c index c5451a4488ca..245dd36638c7 100644 --- a/drivers/infiniband/sw/rxe/rxe_qp.c +++ b/drivers/infiniband/sw/rxe/rxe_qp.c @@ -712,8 +712,9 @@ int rxe_qp_to_attr(struct rxe_qp *qp, struct ib_qp_attr *attr, int mask) if (qp->attr.sq_draining) { spin_unlock_bh(&qp->state_lock); cond_resched(); + } else { + spin_unlock_bh(&qp->state_lock); } - spin_unlock_bh(&qp->state_lock); return 0; } From b5f3fe27c54b1fe77d4dd834ccd48a6d694f872e Mon Sep 17 00:00:00 2001 From: Guoqing Jiang Date: Wed, 10 May 2023 11:50:56 +0800 Subject: [PATCH 04/14] RDMA/rxe: Convert spin_{lock_bh,unlock_bh} to spin_{lock_irqsave,unlock_irqrestore} We need to call spin_lock_irqsave()/spin_unlock_irqrestore() for state_lock in rxe, otherwsie the callchain: ib_post_send_mad -> spin_lock_irqsave -> ib_post_send -> rxe_post_send -> spin_lock_bh -> spin_unlock_bh -> spin_unlock_irqrestore Causes below traces during run block nvmeof-mp/001 test due to mismatched spinlock nesting: WARNING: CPU: 0 PID: 94794 at kernel/softirq.c:376 __local_bh_enable_ip+0xc2/0x140 [ ... ] CPU: 0 PID: 94794 Comm: kworker/u4:1 Tainted: G E 6.4.0-rc1 #9 Hardware name: QEMU Standard PC (Q35 + ICH9, 2009), BIOS rel-1.15.0-0-g2dd4b9b-rebuilt.opensuse.org 04/01/2014 Workqueue: rdma_cm cma_work_handler [rdma_cm] RIP: 0010:__local_bh_enable_ip+0xc2/0x140 Code: 48 85 c0 74 72 5b 41 5c 5d 31 c0 89 c2 89 c1 89 c6 89 c7 41 89 c0 e9 bd 0e 11 01 65 8b 05 f2 65 72 48 85 c0 0f 85 76 ff ff ff <0f> 0b e9 6f ff ff ff e8 d2 39 1c 00 eb 80 4c 89 e7 e8 68 ad 0a 00 RSP: 0018:ffffb7cf818539f0 EFLAGS: 00010046 RAX: 0000000000000000 RBX: 0000000000000201 RCX: 0000000000000000 RDX: 0000000000000000 RSI: 0000000000000201 RDI: ffffffffc0f25f79 RBP: ffffb7cf81853a00 R08: 0000000000000000 R09: 0000000000000000 R10: 0000000000000000 R11: 0000000000000000 R12: ffffffffc0f25f79 R13: ffff8db1f0fa6000 R14: ffff8db2c63ff000 R15: 00000000000000e8 FS: 0000000000000000(0000) GS:ffff8db33bc00000(0000) knlGS:0000000000000000 CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033 CR2: 0000559758db0f20 CR3: 0000000105124000 CR4: 00000000003506f0 Call Trace: _raw_spin_unlock_bh+0x31/0x40 rxe_post_send+0x59/0x8b0 [rdma_rxe] ib_send_mad+0x26b/0x470 [ib_core] ib_post_send_mad+0x150/0xb40 [ib_core] ? cm_form_tid+0x5b/0x90 [ib_cm] ib_send_cm_req+0x7c8/0xb70 [ib_cm] rdma_connect_locked+0x433/0x940 [rdma_cm] nvme_rdma_cm_handler+0x5d7/0x9c0 [nvme_rdma] cma_cm_event_handler+0x4f/0x170 [rdma_cm] cma_work_handler+0x6a/0xe0 [rdma_cm] process_one_work+0x2a9/0x580 worker_thread+0x52/0x3f0 ? __pfx_worker_thread+0x10/0x10 kthread+0x109/0x140 ? __pfx_kthread+0x10/0x10 ret_from_fork+0x2c/0x50 raw_local_irq_restore() called with IRQs enabled WARNING: CPU: 0 PID: 94794 at kernel/locking/irqflag-debug.c:10 warn_bogus_irq_restore+0x37/0x60 [ ... ] CPU: 0 PID: 94794 Comm: kworker/u4:1 Tainted: G W E 6.4.0-rc1 #9 Hardware name: QEMU Standard PC (Q35 + ICH9, 2009), BIOS rel-1.15.0-0-g2dd4b9b-rebuilt.opensuse.org 04/01/2014 Workqueue: rdma_cm cma_work_handler [rdma_cm] RIP: 0010:warn_bogus_irq_restore+0x37/0x60 Code: fb 01 77 36 83 e3 01 74 0e 48 8b 5d f8 c9 31 f6 89 f7 e9 ac ea 01 00 48 c7 c7 e0 52 33 b9 c6 05 bb 1c 69 01 01 e8 39 24 f0 fe <0f> 0b 48 8b 5d f8 c9 31 f6 89 f7 e9 89 ea 01 00 0f b6 f3 48 c7 c7 RSP: 0018:ffffb7cf81853a58 EFLAGS: 00010246 RAX: 0000000000000000 RBX: 0000000000000000 RCX: 0000000000000000 RDX: 0000000000000000 RSI: 0000000000000000 RDI: 0000000000000000 RBP: ffffb7cf81853a60 R08: 0000000000000000 R09: 0000000000000000 R10: 0000000000000000 R11: 0000000000000000 R12: ffff8db2cfb1a9e8 R13: ffff8db2cfb1a9d8 R14: ffff8db2c63ff000 R15: 0000000000000000 FS: 0000000000000000(0000) GS:ffff8db33bc00000(0000) knlGS:0000000000000000 CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033 CR2: 0000559758db0f20 CR3: 0000000105124000 CR4: 00000000003506f0 Call Trace: _raw_spin_unlock_irqrestore+0x91/0xa0 ib_send_mad+0x1e3/0x470 [ib_core] ib_post_send_mad+0x150/0xb40 [ib_core] ? cm_form_tid+0x5b/0x90 [ib_cm] ib_send_cm_req+0x7c8/0xb70 [ib_cm] rdma_connect_locked+0x433/0x940 [rdma_cm] nvme_rdma_cm_handler+0x5d7/0x9c0 [nvme_rdma] cma_cm_event_handler+0x4f/0x170 [rdma_cm] cma_work_handler+0x6a/0xe0 [rdma_cm] process_one_work+0x2a9/0x580 worker_thread+0x52/0x3f0 ? __pfx_worker_thread+0x10/0x10 kthread+0x109/0x140 ? __pfx_kthread+0x10/0x10 ret_from_fork+0x2c/0x50 Fixes: f605f26ea196 ("RDMA/rxe: Protect QP state with qp->state_lock") Link: https://lore.kernel.org/r/20230510035056.881196-1-guoqing.jiang@linux.dev Signed-off-by: Guoqing Jiang Reviewed-by: Bob Pearson Signed-off-by: Jason Gunthorpe --- drivers/infiniband/sw/rxe/rxe_comp.c | 26 +++++++++++-------- drivers/infiniband/sw/rxe/rxe_net.c | 7 +++--- drivers/infiniband/sw/rxe/rxe_qp.c | 36 +++++++++++++++++---------- drivers/infiniband/sw/rxe/rxe_recv.c | 9 ++++--- drivers/infiniband/sw/rxe/rxe_req.c | 30 ++++++++++++---------- drivers/infiniband/sw/rxe/rxe_resp.c | 14 ++++++----- drivers/infiniband/sw/rxe/rxe_verbs.c | 25 ++++++++++--------- 7 files changed, 86 insertions(+), 61 deletions(-) diff --git a/drivers/infiniband/sw/rxe/rxe_comp.c b/drivers/infiniband/sw/rxe/rxe_comp.c index db18ace74d2b..f46c5a5fd0ae 100644 --- 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; diff --git a/drivers/infiniband/sw/rxe/rxe_net.c b/drivers/infiniband/sw/rxe/rxe_net.c index 2bc7361152ea..a38fab19bed1 100644 --- 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); diff --git a/drivers/infiniband/sw/rxe/rxe_qp.c b/drivers/infiniband/sw/rxe/rxe_qp.c index 245dd36638c7..61a2eb77d999 100644 --- 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 +717,12 @@ 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_bh(&qp->state_lock); + spin_unlock_irqrestore(&qp->state_lock, flags); } return 0; @@ -737,10 +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) { diff --git a/drivers/infiniband/sw/rxe/rxe_recv.c b/drivers/infiniband/sw/rxe/rxe_recv.c index 2f953cc74256..5861e4244049 100644 --- 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; } diff --git a/drivers/infiniband/sw/rxe/rxe_req.c b/drivers/infiniband/sw/rxe/rxe_req.c index 65134a9aefe7..5fe7cbae3031 100644 --- 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 diff --git a/drivers/infiniband/sw/rxe/rxe_resp.c b/drivers/infiniband/sw/rxe/rxe_resp.c index 68f6cd188d8e..1da044f6b7d4 100644 --- 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; diff --git a/drivers/infiniband/sw/rxe/rxe_verbs.c b/drivers/infiniband/sw/rxe/rxe_verbs.c index dea605b7f683..4d8f6b8051ff 100644 --- 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; } From 866422cdddcdf59d8c68e9472d49ba1be29b5fcf Mon Sep 17 00:00:00 2001 From: Yonatan Nachum Date: Thu, 11 May 2023 11:51:03 +0000 Subject: [PATCH 05/14] RDMA/efa: Fix unsupported page sizes in device Device uses 4KB size blocks for user pages indirect list while the driver creates those blocks with the size of PAGE_SIZE of the kernel. On kernels with PAGE_SIZE different than 4KB (ARM RHEL), this leads to a failure on register MR with indirect list because of the miss communication between driver and device. Fixes: 40909f664d27 ("RDMA/efa: Add EFA verbs implementation") Link: https://lore.kernel.org/r/20230511115103.13876-1-ynachum@amazon.com Reviewed-by: Firas Jahjah Reviewed-by: Michael Margolin Signed-off-by: Yonatan Nachum Signed-off-by: Jason Gunthorpe --- drivers/infiniband/hw/efa/efa_verbs.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/drivers/infiniband/hw/efa/efa_verbs.c b/drivers/infiniband/hw/efa/efa_verbs.c index 8eca6c14d0cf..2a195c4b0f17 100644 --- 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; From 58caa2a51ad4fd21763696cc6c4defc9fc1b4b4f Mon Sep 17 00:00:00 2001 From: Chengchang Tang Date: Fri, 12 May 2023 17:22:43 +0800 Subject: [PATCH 06/14] RDMA/hns: Fix timeout attr in query qp for HIP08 On HIP08, the queried timeout attr is different from the timeout attr configured by the user. It is found by rdma-core testcase test_rdmacm_async_traffic: ====================================================================== FAIL: test_rdmacm_async_traffic (tests.test_rdmacm.CMTestCase) ---------------------------------------------------------------------- Traceback (most recent call last): File "./tests/test_rdmacm.py", line 33, in test_rdmacm_async_traffic self.two_nodes_rdmacm_traffic(CMAsyncConnection, self.rdmacm_traffic, File "./tests/base.py", line 382, in two_nodes_rdmacm_traffic raise(res) AssertionError Fixes: 926a01dc000d ("RDMA/hns: Add QP operations support for hip08 SoC") Link: https://lore.kernel.org/r/20230512092245.344442-2-huangjunxian6@hisilicon.com Signed-off-by: Chengchang Tang Signed-off-by: Junxian Huang Signed-off-by: Jason Gunthorpe --- drivers/infiniband/hw/hns/hns_roce_hw_v2.c | 17 ++++++++++++++--- drivers/infiniband/hw/hns/hns_roce_hw_v2.h | 2 ++ 2 files changed, 16 insertions(+), 3 deletions(-) diff --git a/drivers/infiniband/hw/hns/hns_roce_hw_v2.c b/drivers/infiniband/hw/hns/hns_roce_hw_v2.c index 84f1167de1d9..3a1c90406ed9 100644 --- a/drivers/infiniband/hw/hns/hns_roce_hw_v2.c +++ b/drivers/infiniband/hw/hns/hns_roce_hw_v2.c @@ -5012,7 +5012,6 @@ 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 +5020,7 @@ 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 +5306,18 @@ 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 +5395,7 @@ 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); diff --git a/drivers/infiniband/hw/hns/hns_roce_hw_v2.h b/drivers/infiniband/hw/hns/hns_roce_hw_v2.h index 1b44d2434ab4..7033eae2407c 100644 --- 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 From 7f3969b14f356dd65fa95b3528eb05c32e68bc06 Mon Sep 17 00:00:00 2001 From: Chengchang Tang Date: Fri, 12 May 2023 17:22:44 +0800 Subject: [PATCH 07/14] RDMA/hns: Fix base address table allocation For hns, the specification of an entry like resource (E.g. WQE/CQE/EQE) depends on BT page size, buf page size and hopnum. For user mode, the buf page size depends on UMEM. Therefore, the actual specification is controlled by BT page size and hopnum. The current BT page size and hopnum are obtained from firmware. This makes the driver inflexible and introduces unnecessary constraints. Resource allocation failures occur in many scenarios. This patch will calculate whether the BT page size set by firmware is sufficient before allocating BT, and increase the BT page size if it is insufficient. Fixes: 1133401412a9 ("RDMA/hns: Optimize base address table config flow for qp buffer") Link: https://lore.kernel.org/r/20230512092245.344442-3-huangjunxian6@hisilicon.com Signed-off-by: Chengchang Tang Signed-off-by: Junxian Huang Signed-off-by: Jason Gunthorpe --- drivers/infiniband/hw/hns/hns_roce_mr.c | 43 +++++++++++++++++++++++++ 1 file changed, 43 insertions(+) diff --git a/drivers/infiniband/hw/hns/hns_roce_mr.c b/drivers/infiniband/hw/hns/hns_roce_mr.c index 37a5cf62f88b..14376490ac22 100644 --- a/drivers/infiniband/hw/hns/hns_roce_mr.c +++ b/drivers/infiniband/hw/hns/hns_roce_mr.c @@ -33,6 +33,7 @@ #include #include +#include #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); From 56518a603fd2bf74762d176ac980572db84a3e14 Mon Sep 17 00:00:00 2001 From: Yangyang Li Date: Fri, 12 May 2023 17:22:45 +0800 Subject: [PATCH 08/14] RDMA/hns: Modify the value of long message loopback slice Long message loopback slice is used for achieving traffic balance between QPs. It prevents the problem that QPs with large traffic occupying the hardware pipeline for a long time and QPs with small traffic cannot be scheduled. Currently, its maximum value is set to 16K, which means only after a QP sends 16K will the second QP be scheduled. This value is too large, which will lead to unbalanced traffic scheduling, and thus it needs to be modified. The setting range of the long message loopback slice is modified to be from 1024 (the lower limit supported by hardware) to mtu. Actual testing shows that this value can significantly reduce error in hardware traffic scheduling. This solution is compatible with both HIP08 and HIP09. The modified lp_pktn_ini has a maximum value of 2 (when mtu is 256), so the range checking code for lp_pktn_ini is no longer necessary and needs to be deleted. Fixes: 0e60778efb07 ("RDMA/hns: Modify the value of MAX_LP_MSG_LEN to meet hardware compatibility") Link: https://lore.kernel.org/r/20230512092245.344442-4-huangjunxian6@hisilicon.com Signed-off-by: Yangyang Li Signed-off-by: Junxian Huang Signed-off-by: Jason Gunthorpe --- drivers/infiniband/hw/hns/hns_roce_hw_v2.c | 8 +++----- 1 file changed, 3 insertions(+), 5 deletions(-) diff --git a/drivers/infiniband/hw/hns/hns_roce_hw_v2.c b/drivers/infiniband/hw/hns/hns_roce_hw_v2.c index 3a1c90406ed9..d4c6b9bc0a4e 100644 --- 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); From 349e3c0cf239cc01d58a1e6c749e171de014cd6a Mon Sep 17 00:00:00 2001 From: Kalesh AP Date: Thu, 18 May 2023 01:10:59 -0700 Subject: [PATCH 09/14] RDMA/bnxt_re: Fix a possible memory leak Inside bnxt_qplib_create_cq(), when the check for NULL DPI fails, driver returns directly without freeing the memory allocated inside bnxt_qplib_alloc_init_hwq() routine. Fixed this by moving the check for NULL DPI before invoking bnxt_qplib_alloc_init_hwq(). Fixes: 1ac5a4047975 ("RDMA/bnxt_re: Add bnxt_re RoCE driver") Link: https://lore.kernel.org/r/1684397461-23082-2-git-send-email-selvin.xavier@broadcom.com Reviewed-by: Kashyap Desai Signed-off-by: Kalesh AP Signed-off-by: Selvin Xavier Signed-off-by: Jason Gunthorpe --- drivers/infiniband/hw/bnxt_re/qplib_fp.c | 11 ++++++----- 1 file changed, 6 insertions(+), 5 deletions(-) diff --git a/drivers/infiniband/hw/bnxt_re/qplib_fp.c b/drivers/infiniband/hw/bnxt_re/qplib_fp.c index f139d4cd1712..8974f6235cfa 100644 --- 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); From 0fa0d520e2a878cb4c94c4dc84395905d3f14f54 Mon Sep 17 00:00:00 2001 From: Kalesh AP Date: Thu, 18 May 2023 01:11:00 -0700 Subject: [PATCH 10/14] RDMA/bnxt_re: Fix return value of bnxt_re_process_raw_qp_pkt_rx bnxt_re_process_raw_qp_pkt_rx() always return 0 and ignores the return value of bnxt_re_post_send_shadow_qp(). Fixes: 1ac5a4047975 ("RDMA/bnxt_re: Add bnxt_re RoCE driver") Link: https://lore.kernel.org/r/1684397461-23082-3-git-send-email-selvin.xavier@broadcom.com Reviewed-by: Hongguang Gao Reviewed-by: Ajit Khaparde Signed-off-by: Kalesh AP Signed-off-by: Selvin Xavier Signed-off-by: Jason Gunthorpe --- drivers/infiniband/hw/bnxt_re/ib_verbs.c | 4 +--- 1 file changed, 1 insertion(+), 3 deletions(-) diff --git a/drivers/infiniband/hw/bnxt_re/ib_verbs.c b/drivers/infiniband/hw/bnxt_re/ib_verbs.c index e86afecfbe46..b1c36412025f 100644 --- 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, From dd5fb04857d7346c9ea4901ec3ebe5b90b0e8868 Mon Sep 17 00:00:00 2001 From: Kalesh AP Date: Thu, 18 May 2023 01:11:01 -0700 Subject: [PATCH 11/14] RDMA/bnxt_re: Do not enable congestion control on VFs Congestion control needs to be enabled only on the PFs. FW fails the command if issued on VFs. Avoid sending the command on VFs. Fixes: f13bcef04ba0 ("RDMA/bnxt_re: Enable congestion control by default") Link: https://lore.kernel.org/r/1684397461-23082-4-git-send-email-selvin.xavier@broadcom.com Signed-off-by: Kalesh AP Reviewed-by: Selvin Thyparampil Xavier Signed-off-by: Selvin Xavier Signed-off-by: Jason Gunthorpe --- drivers/infiniband/hw/bnxt_re/main.c | 4 ++++ 1 file changed, 4 insertions(+) diff --git a/drivers/infiniband/hw/bnxt_re/main.c b/drivers/infiniband/hw/bnxt_re/main.c index b9e2f89337e8..e34eccd178d0 100644 --- 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; From ffe14de983252862c91ad23bd5ca72fd9398d0e6 Mon Sep 17 00:00:00 2001 From: Michael Margolin Date: Thu, 25 May 2023 09:44:44 +0000 Subject: [PATCH 12/14] MAINTAINERS: Update maintainer of Amazon EFA driver Change EFA driver maintainer from Gal Pressman to myself. Keep Gal as a reviewer at his request. Link: https://lore.kernel.org/r/20230525094444.12570-1-mrgolin@amazon.com Signed-off-by: Michael Margolin Acked-by: Gal Pressman Signed-off-by: Jason Gunthorpe --- MAINTAINERS | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-) diff --git a/MAINTAINERS b/MAINTAINERS index c56043737848..dc8c4708733b 100644 --- a/MAINTAINERS +++ b/MAINTAINERS @@ -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 +M: Michael Margolin +R: Gal Pressman R: Yossi Leybovich L: linux-rdma@vger.kernel.org S: Supported From c8f304d75f6c6cc679a73f89591f9a915da38f09 Mon Sep 17 00:00:00 2001 From: Mustafa Ismail Date: Mon, 22 May 2023 10:56:53 -0500 Subject: [PATCH 13/14] RDMA/irdma: Prevent QP use after free There is a window where the poll cq may use a QP that has been freed. This can happen if a CQE is polled before irdma_clean_cqes() can clear the CQE's related to the QP and the destroy QP races to free the QP memory. then the QP structures are used in irdma_poll_cq. Fix this by moving the clearing of CQE's before the reference is removed and the QP is destroyed. Fixes: b48c24c2d710 ("RDMA/irdma: Implement device supported verb APIs") Link: https://lore.kernel.org/r/20230522155654.1309-3-shiraz.saleem@intel.com Signed-off-by: Mustafa Ismail Signed-off-by: Shiraz Saleem Signed-off-by: Jason Gunthorpe --- drivers/infiniband/hw/irdma/verbs.c | 11 ++++++----- 1 file changed, 6 insertions(+), 5 deletions(-) diff --git a/drivers/infiniband/hw/irdma/verbs.c b/drivers/infiniband/hw/irdma/verbs.c index ab5cdf782785..68ce3cd40002 100644 --- 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); From 5842d1d9c1b0d17e0c29eae65ae1f245f83682dd Mon Sep 17 00:00:00 2001 From: Mustafa Ismail Date: Mon, 22 May 2023 10:56:54 -0500 Subject: [PATCH 14/14] RDMA/irdma: Fix Local Invalidate fencing If the local invalidate fence is indicated in the WR, only the read fence is currently being set in WQE. Fix this to set both the read and local fence in the WQE. Fixes: b48c24c2d710 ("RDMA/irdma: Implement device supported verb APIs") Link: https://lore.kernel.org/r/20230522155654.1309-4-shiraz.saleem@intel.com Signed-off-by: Mustafa Ismail Signed-off-by: Shiraz Saleem Signed-off-by: Jason Gunthorpe --- drivers/infiniband/hw/irdma/verbs.c | 1 + 1 file changed, 1 insertion(+) diff --git a/drivers/infiniband/hw/irdma/verbs.c b/drivers/infiniband/hw/irdma/verbs.c index 68ce3cd40002..eaa12c124598 100644 --- a/drivers/infiniband/hw/irdma/verbs.c +++ b/drivers/infiniband/hw/irdma/verbs.c @@ -3292,6 +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;