aboutsummaryrefslogtreecommitdiffstats
path: root/drivers/infiniband/hw/bnxt_re
diff options
context:
space:
mode:
authorLinus Torvalds <torvalds@linux-foundation.org>2017-07-21 17:22:05 -0400
committerLinus Torvalds <torvalds@linux-foundation.org>2017-07-21 17:22:05 -0400
commitbb236dbeea8181c6733e6d34a53bfef9c8ef4e95 (patch)
tree2d71f98dcc5676beb4e6029462d60be0ff6ebbc1 /drivers/infiniband/hw/bnxt_re
parent24a1635a41bccb5cc426eaef8b88c7e0961ef6bb (diff)
parenta62ab66b13a0f9bcb17b7b761f6670941ed5cd62 (diff)
Merge tag 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/dledford/rdma
Pull more rdma fixes from Doug Ledford: "As per my previous pull request, there were two drivers that each had a rather large number of legitimate fixes still to be sent. As it turned out, I also missed a reasonably large set of fixes from one person across the stack that are all important fixes. All in all, the bnxt_re, i40iw, and Dan Carpenter are 3/4 to 2/3rds of this pull request. There were some other random fixes that I didn't send in the last pull request that I added to this one. This catches the rdma stack up to the fixes from up to about the beginning of this week. Any more fixes I'll wait and batch up later in the -rc cycle. This will give us a good base to start with for basing a for-next branch on -rc2. Summary: - i40iw fixes - bnxt_re fixes - Dan Carpenter bugfixes across stack - ten more random fixes, no more than two from any one person" * tag 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/dledford/rdma: (37 commits) RDMA/core: Initialize port_num in qp_attr RDMA/uverbs: Fix the check for port number IB/cma: Fix reference count leak when no ipv4 addresses are set RDMA/iser: don't send an rkey if all data is written as immadiate-data rxe: fix broken receive queue draining RDMA/qedr: Prevent memory overrun in verbs' user responses iw_cxgb4: don't use WR keys/addrs for 0 byte reads IB/mlx4: Fix CM REQ retries in paravirt mode IB/rdmavt: Setting of QP timeout can overflow jiffies computation IB/core: Fix sparse warnings RDMA/bnxt_re: Fix the value reported for local ack delay RDMA/bnxt_re: Report MISSED_EVENTS in req_notify_cq RDMA/bnxt_re: Fix return value of poll routine RDMA/bnxt_re: Enable atomics only if host bios supports RDMA/bnxt_re: Specify RDMA component when allocating stats context RDMA/bnxt_re: Fixed the max_rd_atomic support for initiator and destination QP RDMA/bnxt_re: Report supported value to IB stack in query_device RDMA/bnxt_re: Do not free the ctx_tbl entry if delete GID fails RDMA/bnxt_re: Fix WQE Size posted to HW to prevent it from throwing error RDMA/bnxt_re: Free doorbell page index (DPI) during dealloc ucontext ...
Diffstat (limited to 'drivers/infiniband/hw/bnxt_re')
-rw-r--r--drivers/infiniband/hw/bnxt_re/bnxt_re.h9
-rw-r--r--drivers/infiniband/hw/bnxt_re/ib_verbs.c119
-rw-r--r--drivers/infiniband/hw/bnxt_re/ib_verbs.h3
-rw-r--r--drivers/infiniband/hw/bnxt_re/main.c1
-rw-r--r--drivers/infiniband/hw/bnxt_re/qplib_fp.c29
-rw-r--r--drivers/infiniband/hw/bnxt_re/qplib_fp.h1
-rw-r--r--drivers/infiniband/hw/bnxt_re/qplib_sp.c16
-rw-r--r--drivers/infiniband/hw/bnxt_re/qplib_sp.h3
8 files changed, 132 insertions, 49 deletions
diff --git a/drivers/infiniband/hw/bnxt_re/bnxt_re.h b/drivers/infiniband/hw/bnxt_re/bnxt_re.h
index 08772836fded..85527532c49d 100644
--- a/drivers/infiniband/hw/bnxt_re/bnxt_re.h
+++ b/drivers/infiniband/hw/bnxt_re/bnxt_re.h
@@ -51,6 +51,8 @@
51#define BNXT_RE_PAGE_SIZE_8M BIT(23) 51#define BNXT_RE_PAGE_SIZE_8M BIT(23)
52#define BNXT_RE_PAGE_SIZE_1G BIT(30) 52#define BNXT_RE_PAGE_SIZE_1G BIT(30)
53 53
54#define BNXT_RE_MAX_MR_SIZE BIT(30)
55
54#define BNXT_RE_MAX_QPC_COUNT (64 * 1024) 56#define BNXT_RE_MAX_QPC_COUNT (64 * 1024)
55#define BNXT_RE_MAX_MRW_COUNT (64 * 1024) 57#define BNXT_RE_MAX_MRW_COUNT (64 * 1024)
56#define BNXT_RE_MAX_SRQC_COUNT (64 * 1024) 58#define BNXT_RE_MAX_SRQC_COUNT (64 * 1024)
@@ -60,6 +62,13 @@
60 62
61#define BNXT_RE_RQ_WQE_THRESHOLD 32 63#define BNXT_RE_RQ_WQE_THRESHOLD 32
62 64
65/*
66 * Setting the default ack delay value to 16, which means
67 * the default timeout is approx. 260ms(4 usec * 2 ^(timeout))
68 */
69
70#define BNXT_RE_DEFAULT_ACK_DELAY 16
71
63struct bnxt_re_work { 72struct bnxt_re_work {
64 struct work_struct work; 73 struct work_struct work;
65 unsigned long event; 74 unsigned long event;
diff --git a/drivers/infiniband/hw/bnxt_re/ib_verbs.c b/drivers/infiniband/hw/bnxt_re/ib_verbs.c
index c7bd68311d0c..f0e01b3ac711 100644
--- a/drivers/infiniband/hw/bnxt_re/ib_verbs.c
+++ b/drivers/infiniband/hw/bnxt_re/ib_verbs.c
@@ -145,10 +145,8 @@ int bnxt_re_query_device(struct ib_device *ibdev,
145 ib_attr->fw_ver = (u64)(unsigned long)(dev_attr->fw_ver); 145 ib_attr->fw_ver = (u64)(unsigned long)(dev_attr->fw_ver);
146 bnxt_qplib_get_guid(rdev->netdev->dev_addr, 146 bnxt_qplib_get_guid(rdev->netdev->dev_addr,
147 (u8 *)&ib_attr->sys_image_guid); 147 (u8 *)&ib_attr->sys_image_guid);
148 ib_attr->max_mr_size = ~0ull; 148 ib_attr->max_mr_size = BNXT_RE_MAX_MR_SIZE;
149 ib_attr->page_size_cap = BNXT_RE_PAGE_SIZE_4K | BNXT_RE_PAGE_SIZE_8K | 149 ib_attr->page_size_cap = BNXT_RE_PAGE_SIZE_4K;
150 BNXT_RE_PAGE_SIZE_64K | BNXT_RE_PAGE_SIZE_2M |
151 BNXT_RE_PAGE_SIZE_8M | BNXT_RE_PAGE_SIZE_1G;
152 150
153 ib_attr->vendor_id = rdev->en_dev->pdev->vendor; 151 ib_attr->vendor_id = rdev->en_dev->pdev->vendor;
154 ib_attr->vendor_part_id = rdev->en_dev->pdev->device; 152 ib_attr->vendor_part_id = rdev->en_dev->pdev->device;
@@ -174,9 +172,11 @@ int bnxt_re_query_device(struct ib_device *ibdev,
174 ib_attr->max_mr = dev_attr->max_mr; 172 ib_attr->max_mr = dev_attr->max_mr;
175 ib_attr->max_pd = dev_attr->max_pd; 173 ib_attr->max_pd = dev_attr->max_pd;
176 ib_attr->max_qp_rd_atom = dev_attr->max_qp_rd_atom; 174 ib_attr->max_qp_rd_atom = dev_attr->max_qp_rd_atom;
177 ib_attr->max_qp_init_rd_atom = dev_attr->max_qp_rd_atom; 175 ib_attr->max_qp_init_rd_atom = dev_attr->max_qp_init_rd_atom;
178 ib_attr->atomic_cap = IB_ATOMIC_HCA; 176 if (dev_attr->is_atomic) {
179 ib_attr->masked_atomic_cap = IB_ATOMIC_HCA; 177 ib_attr->atomic_cap = IB_ATOMIC_HCA;
178 ib_attr->masked_atomic_cap = IB_ATOMIC_HCA;
179 }
180 180
181 ib_attr->max_ee_rd_atom = 0; 181 ib_attr->max_ee_rd_atom = 0;
182 ib_attr->max_res_rd_atom = 0; 182 ib_attr->max_res_rd_atom = 0;
@@ -201,7 +201,7 @@ int bnxt_re_query_device(struct ib_device *ibdev,
201 ib_attr->max_fast_reg_page_list_len = MAX_PBL_LVL_1_PGS; 201 ib_attr->max_fast_reg_page_list_len = MAX_PBL_LVL_1_PGS;
202 202
203 ib_attr->max_pkeys = 1; 203 ib_attr->max_pkeys = 1;
204 ib_attr->local_ca_ack_delay = 0; 204 ib_attr->local_ca_ack_delay = BNXT_RE_DEFAULT_ACK_DELAY;
205 return 0; 205 return 0;
206} 206}
207 207
@@ -390,15 +390,17 @@ int bnxt_re_del_gid(struct ib_device *ibdev, u8 port_num,
390 return -EINVAL; 390 return -EINVAL;
391 ctx->refcnt--; 391 ctx->refcnt--;
392 if (!ctx->refcnt) { 392 if (!ctx->refcnt) {
393 rc = bnxt_qplib_del_sgid 393 rc = bnxt_qplib_del_sgid(sgid_tbl,
394 (sgid_tbl, 394 &sgid_tbl->tbl[ctx->idx],
395 &sgid_tbl->tbl[ctx->idx], true); 395 true);
396 if (rc) 396 if (rc) {
397 dev_err(rdev_to_dev(rdev), 397 dev_err(rdev_to_dev(rdev),
398 "Failed to remove GID: %#x", rc); 398 "Failed to remove GID: %#x", rc);
399 ctx_tbl = sgid_tbl->ctx; 399 } else {
400 ctx_tbl[ctx->idx] = NULL; 400 ctx_tbl = sgid_tbl->ctx;
401 kfree(ctx); 401 ctx_tbl[ctx->idx] = NULL;
402 kfree(ctx);
403 }
402 } 404 }
403 } else { 405 } else {
404 return -EINVAL; 406 return -EINVAL;
@@ -588,10 +590,10 @@ static int bnxt_re_create_fence_mr(struct bnxt_re_pd *pd)
588 590
589 /* Create a fence MW only for kernel consumers */ 591 /* Create a fence MW only for kernel consumers */
590 mw = bnxt_re_alloc_mw(&pd->ib_pd, IB_MW_TYPE_1, NULL); 592 mw = bnxt_re_alloc_mw(&pd->ib_pd, IB_MW_TYPE_1, NULL);
591 if (!mw) { 593 if (IS_ERR(mw)) {
592 dev_err(rdev_to_dev(rdev), 594 dev_err(rdev_to_dev(rdev),
593 "Failed to create fence-MW for PD: %p\n", pd); 595 "Failed to create fence-MW for PD: %p\n", pd);
594 rc = -EINVAL; 596 rc = PTR_ERR(mw);
595 goto fail; 597 goto fail;
596 } 598 }
597 fence->mw = mw; 599 fence->mw = mw;
@@ -612,30 +614,13 @@ int bnxt_re_dealloc_pd(struct ib_pd *ib_pd)
612 int rc; 614 int rc;
613 615
614 bnxt_re_destroy_fence_mr(pd); 616 bnxt_re_destroy_fence_mr(pd);
615 if (ib_pd->uobject && pd->dpi.dbr) {
616 struct ib_ucontext *ib_uctx = ib_pd->uobject->context;
617 struct bnxt_re_ucontext *ucntx;
618 617
619 /* Free DPI only if this is the first PD allocated by the 618 if (pd->qplib_pd.id) {
620 * application and mark the context dpi as NULL 619 rc = bnxt_qplib_dealloc_pd(&rdev->qplib_res,
621 */ 620 &rdev->qplib_res.pd_tbl,
622 ucntx = container_of(ib_uctx, struct bnxt_re_ucontext, ib_uctx); 621 &pd->qplib_pd);
623
624 rc = bnxt_qplib_dealloc_dpi(&rdev->qplib_res,
625 &rdev->qplib_res.dpi_tbl,
626 &pd->dpi);
627 if (rc) 622 if (rc)
628 dev_err(rdev_to_dev(rdev), "Failed to deallocate HW DPI"); 623 dev_err(rdev_to_dev(rdev), "Failed to deallocate HW PD");
629 /* Don't fail, continue*/
630 ucntx->dpi = NULL;
631 }
632
633 rc = bnxt_qplib_dealloc_pd(&rdev->qplib_res,
634 &rdev->qplib_res.pd_tbl,
635 &pd->qplib_pd);
636 if (rc) {
637 dev_err(rdev_to_dev(rdev), "Failed to deallocate HW PD");
638 return rc;
639 } 624 }
640 625
641 kfree(pd); 626 kfree(pd);
@@ -667,23 +652,22 @@ struct ib_pd *bnxt_re_alloc_pd(struct ib_device *ibdev,
667 if (udata) { 652 if (udata) {
668 struct bnxt_re_pd_resp resp; 653 struct bnxt_re_pd_resp resp;
669 654
670 if (!ucntx->dpi) { 655 if (!ucntx->dpi.dbr) {
671 /* Allocate DPI in alloc_pd to avoid failing of 656 /* Allocate DPI in alloc_pd to avoid failing of
672 * ibv_devinfo and family of application when DPIs 657 * ibv_devinfo and family of application when DPIs
673 * are depleted. 658 * are depleted.
674 */ 659 */
675 if (bnxt_qplib_alloc_dpi(&rdev->qplib_res.dpi_tbl, 660 if (bnxt_qplib_alloc_dpi(&rdev->qplib_res.dpi_tbl,
676 &pd->dpi, ucntx)) { 661 &ucntx->dpi, ucntx)) {
677 rc = -ENOMEM; 662 rc = -ENOMEM;
678 goto dbfail; 663 goto dbfail;
679 } 664 }
680 ucntx->dpi = &pd->dpi;
681 } 665 }
682 666
683 resp.pdid = pd->qplib_pd.id; 667 resp.pdid = pd->qplib_pd.id;
684 /* Still allow mapping this DBR to the new user PD. */ 668 /* Still allow mapping this DBR to the new user PD. */
685 resp.dpi = ucntx->dpi->dpi; 669 resp.dpi = ucntx->dpi.dpi;
686 resp.dbr = (u64)ucntx->dpi->umdbr; 670 resp.dbr = (u64)ucntx->dpi.umdbr;
687 671
688 rc = ib_copy_to_udata(udata, &resp, sizeof(resp)); 672 rc = ib_copy_to_udata(udata, &resp, sizeof(resp));
689 if (rc) { 673 if (rc) {
@@ -960,7 +944,7 @@ static int bnxt_re_init_user_qp(struct bnxt_re_dev *rdev, struct bnxt_re_pd *pd,
960 qplib_qp->rq.nmap = umem->nmap; 944 qplib_qp->rq.nmap = umem->nmap;
961 } 945 }
962 946
963 qplib_qp->dpi = cntx->dpi; 947 qplib_qp->dpi = &cntx->dpi;
964 return 0; 948 return 0;
965rqfail: 949rqfail:
966 ib_umem_release(qp->sumem); 950 ib_umem_release(qp->sumem);
@@ -1530,13 +1514,24 @@ int bnxt_re_modify_qp(struct ib_qp *ib_qp, struct ib_qp_attr *qp_attr,
1530 if (qp_attr_mask & IB_QP_MAX_QP_RD_ATOMIC) { 1514 if (qp_attr_mask & IB_QP_MAX_QP_RD_ATOMIC) {
1531 qp->qplib_qp.modify_flags |= 1515 qp->qplib_qp.modify_flags |=
1532 CMDQ_MODIFY_QP_MODIFY_MASK_MAX_RD_ATOMIC; 1516 CMDQ_MODIFY_QP_MODIFY_MASK_MAX_RD_ATOMIC;
1533 qp->qplib_qp.max_rd_atomic = qp_attr->max_rd_atomic; 1517 /* Cap the max_rd_atomic to device max */
1518 qp->qplib_qp.max_rd_atomic = min_t(u32, qp_attr->max_rd_atomic,
1519 dev_attr->max_qp_rd_atom);
1534 } 1520 }
1535 if (qp_attr_mask & IB_QP_SQ_PSN) { 1521 if (qp_attr_mask & IB_QP_SQ_PSN) {
1536 qp->qplib_qp.modify_flags |= CMDQ_MODIFY_QP_MODIFY_MASK_SQ_PSN; 1522 qp->qplib_qp.modify_flags |= CMDQ_MODIFY_QP_MODIFY_MASK_SQ_PSN;
1537 qp->qplib_qp.sq.psn = qp_attr->sq_psn; 1523 qp->qplib_qp.sq.psn = qp_attr->sq_psn;
1538 } 1524 }
1539 if (qp_attr_mask & IB_QP_MAX_DEST_RD_ATOMIC) { 1525 if (qp_attr_mask & IB_QP_MAX_DEST_RD_ATOMIC) {
1526 if (qp_attr->max_dest_rd_atomic >
1527 dev_attr->max_qp_init_rd_atom) {
1528 dev_err(rdev_to_dev(rdev),
1529 "max_dest_rd_atomic requested%d is > dev_max%d",
1530 qp_attr->max_dest_rd_atomic,
1531 dev_attr->max_qp_init_rd_atom);
1532 return -EINVAL;
1533 }
1534
1540 qp->qplib_qp.modify_flags |= 1535 qp->qplib_qp.modify_flags |=
1541 CMDQ_MODIFY_QP_MODIFY_MASK_MAX_DEST_RD_ATOMIC; 1536 CMDQ_MODIFY_QP_MODIFY_MASK_MAX_DEST_RD_ATOMIC;
1542 qp->qplib_qp.max_dest_rd_atomic = qp_attr->max_dest_rd_atomic; 1537 qp->qplib_qp.max_dest_rd_atomic = qp_attr->max_dest_rd_atomic;
@@ -2403,7 +2398,7 @@ struct ib_cq *bnxt_re_create_cq(struct ib_device *ibdev,
2403 } 2398 }
2404 cq->qplib_cq.sghead = cq->umem->sg_head.sgl; 2399 cq->qplib_cq.sghead = cq->umem->sg_head.sgl;
2405 cq->qplib_cq.nmap = cq->umem->nmap; 2400 cq->qplib_cq.nmap = cq->umem->nmap;
2406 cq->qplib_cq.dpi = uctx->dpi; 2401 cq->qplib_cq.dpi = &uctx->dpi;
2407 } else { 2402 } else {
2408 cq->max_cql = min_t(u32, entries, MAX_CQL_PER_POLL); 2403 cq->max_cql = min_t(u32, entries, MAX_CQL_PER_POLL);
2409 cq->cql = kcalloc(cq->max_cql, sizeof(struct bnxt_qplib_cqe), 2404 cq->cql = kcalloc(cq->max_cql, sizeof(struct bnxt_qplib_cqe),
@@ -2905,6 +2900,7 @@ int bnxt_re_poll_cq(struct ib_cq *ib_cq, int num_entries, struct ib_wc *wc)
2905 2900
2906 spin_lock_irqsave(&cq->cq_lock, flags); 2901 spin_lock_irqsave(&cq->cq_lock, flags);
2907 budget = min_t(u32, num_entries, cq->max_cql); 2902 budget = min_t(u32, num_entries, cq->max_cql);
2903 num_entries = budget;
2908 if (!cq->cql) { 2904 if (!cq->cql) {
2909 dev_err(rdev_to_dev(cq->rdev), "POLL CQ : no CQL to use"); 2905 dev_err(rdev_to_dev(cq->rdev), "POLL CQ : no CQL to use");
2910 goto exit; 2906 goto exit;
@@ -3031,6 +3027,11 @@ int bnxt_re_req_notify_cq(struct ib_cq *ib_cq,
3031 else if (ib_cqn_flags & IB_CQ_SOLICITED) 3027 else if (ib_cqn_flags & IB_CQ_SOLICITED)
3032 type = DBR_DBR_TYPE_CQ_ARMSE; 3028 type = DBR_DBR_TYPE_CQ_ARMSE;
3033 3029
3030 /* Poll to see if there are missed events */
3031 if ((ib_cqn_flags & IB_CQ_REPORT_MISSED_EVENTS) &&
3032 !(bnxt_qplib_is_cq_empty(&cq->qplib_cq)))
3033 return 1;
3034
3034 bnxt_qplib_req_notify_cq(&cq->qplib_cq, type); 3035 bnxt_qplib_req_notify_cq(&cq->qplib_cq, type);
3035 3036
3036 return 0; 3037 return 0;
@@ -3245,6 +3246,12 @@ struct ib_mr *bnxt_re_reg_user_mr(struct ib_pd *ib_pd, u64 start, u64 length,
3245 struct scatterlist *sg; 3246 struct scatterlist *sg;
3246 int entry; 3247 int entry;
3247 3248
3249 if (length > BNXT_RE_MAX_MR_SIZE) {
3250 dev_err(rdev_to_dev(rdev), "MR Size: %lld > Max supported:%ld\n",
3251 length, BNXT_RE_MAX_MR_SIZE);
3252 return ERR_PTR(-ENOMEM);
3253 }
3254
3248 mr = kzalloc(sizeof(*mr), GFP_KERNEL); 3255 mr = kzalloc(sizeof(*mr), GFP_KERNEL);
3249 if (!mr) 3256 if (!mr)
3250 return ERR_PTR(-ENOMEM); 3257 return ERR_PTR(-ENOMEM);
@@ -3388,8 +3395,26 @@ int bnxt_re_dealloc_ucontext(struct ib_ucontext *ib_uctx)
3388 struct bnxt_re_ucontext *uctx = container_of(ib_uctx, 3395 struct bnxt_re_ucontext *uctx = container_of(ib_uctx,
3389 struct bnxt_re_ucontext, 3396 struct bnxt_re_ucontext,
3390 ib_uctx); 3397 ib_uctx);
3398
3399 struct bnxt_re_dev *rdev = uctx->rdev;
3400 int rc = 0;
3401
3391 if (uctx->shpg) 3402 if (uctx->shpg)
3392 free_page((unsigned long)uctx->shpg); 3403 free_page((unsigned long)uctx->shpg);
3404
3405 if (uctx->dpi.dbr) {
3406 /* Free DPI only if this is the first PD allocated by the
3407 * application and mark the context dpi as NULL
3408 */
3409 rc = bnxt_qplib_dealloc_dpi(&rdev->qplib_res,
3410 &rdev->qplib_res.dpi_tbl,
3411 &uctx->dpi);
3412 if (rc)
3413 dev_err(rdev_to_dev(rdev), "Deallocte HW DPI failed!");
3414 /* Don't fail, continue*/
3415 uctx->dpi.dbr = NULL;
3416 }
3417
3393 kfree(uctx); 3418 kfree(uctx);
3394 return 0; 3419 return 0;
3395} 3420}
diff --git a/drivers/infiniband/hw/bnxt_re/ib_verbs.h b/drivers/infiniband/hw/bnxt_re/ib_verbs.h
index 6c160f6a5398..a0bb7e33d7ca 100644
--- a/drivers/infiniband/hw/bnxt_re/ib_verbs.h
+++ b/drivers/infiniband/hw/bnxt_re/ib_verbs.h
@@ -59,7 +59,6 @@ struct bnxt_re_pd {
59 struct bnxt_re_dev *rdev; 59 struct bnxt_re_dev *rdev;
60 struct ib_pd ib_pd; 60 struct ib_pd ib_pd;
61 struct bnxt_qplib_pd qplib_pd; 61 struct bnxt_qplib_pd qplib_pd;
62 struct bnxt_qplib_dpi dpi;
63 struct bnxt_re_fence_data fence; 62 struct bnxt_re_fence_data fence;
64}; 63};
65 64
@@ -127,7 +126,7 @@ struct bnxt_re_mw {
127struct bnxt_re_ucontext { 126struct bnxt_re_ucontext {
128 struct bnxt_re_dev *rdev; 127 struct bnxt_re_dev *rdev;
129 struct ib_ucontext ib_uctx; 128 struct ib_ucontext ib_uctx;
130 struct bnxt_qplib_dpi *dpi; 129 struct bnxt_qplib_dpi dpi;
131 void *shpg; 130 void *shpg;
132 spinlock_t sh_lock; /* protect shpg */ 131 spinlock_t sh_lock; /* protect shpg */
133}; 132};
diff --git a/drivers/infiniband/hw/bnxt_re/main.c b/drivers/infiniband/hw/bnxt_re/main.c
index 1fce5e73216b..ceae2d92fb08 100644
--- a/drivers/infiniband/hw/bnxt_re/main.c
+++ b/drivers/infiniband/hw/bnxt_re/main.c
@@ -333,6 +333,7 @@ static int bnxt_re_net_stats_ctx_alloc(struct bnxt_re_dev *rdev,
333 bnxt_re_init_hwrm_hdr(rdev, (void *)&req, HWRM_STAT_CTX_ALLOC, -1, -1); 333 bnxt_re_init_hwrm_hdr(rdev, (void *)&req, HWRM_STAT_CTX_ALLOC, -1, -1);
334 req.update_period_ms = cpu_to_le32(1000); 334 req.update_period_ms = cpu_to_le32(1000);
335 req.stats_dma_addr = cpu_to_le64(dma_map); 335 req.stats_dma_addr = cpu_to_le64(dma_map);
336 req.stat_ctx_flags = STAT_CTX_ALLOC_REQ_STAT_CTX_FLAGS_ROCE;
336 bnxt_re_fill_fw_msg(&fw_msg, (void *)&req, sizeof(req), (void *)&resp, 337 bnxt_re_fill_fw_msg(&fw_msg, (void *)&req, sizeof(req), (void *)&resp,
337 sizeof(resp), DFLT_HWRM_CMD_TIMEOUT); 338 sizeof(resp), DFLT_HWRM_CMD_TIMEOUT);
338 rc = en_dev->en_ops->bnxt_send_fw_msg(en_dev, BNXT_ROCE_ULP, &fw_msg); 339 rc = en_dev->en_ops->bnxt_send_fw_msg(en_dev, BNXT_ROCE_ULP, &fw_msg);
diff --git a/drivers/infiniband/hw/bnxt_re/qplib_fp.c b/drivers/infiniband/hw/bnxt_re/qplib_fp.c
index f05500bcdcf1..9af1514e5944 100644
--- a/drivers/infiniband/hw/bnxt_re/qplib_fp.c
+++ b/drivers/infiniband/hw/bnxt_re/qplib_fp.c
@@ -1128,6 +1128,11 @@ int bnxt_qplib_post_send(struct bnxt_qplib_qp *qp,
1128 } 1128 }
1129 /* Each SGE entry = 1 WQE size16 */ 1129 /* Each SGE entry = 1 WQE size16 */
1130 wqe_size16 = wqe->num_sge; 1130 wqe_size16 = wqe->num_sge;
1131 /* HW requires wqe size has room for atleast one SGE even if
1132 * none was supplied by ULP
1133 */
1134 if (!wqe->num_sge)
1135 wqe_size16++;
1131 } 1136 }
1132 1137
1133 /* Specifics */ 1138 /* Specifics */
@@ -1364,6 +1369,11 @@ int bnxt_qplib_post_recv(struct bnxt_qplib_qp *qp,
1364 rqe->flags = wqe->flags; 1369 rqe->flags = wqe->flags;
1365 rqe->wqe_size = wqe->num_sge + 1370 rqe->wqe_size = wqe->num_sge +
1366 ((offsetof(typeof(*rqe), data) + 15) >> 4); 1371 ((offsetof(typeof(*rqe), data) + 15) >> 4);
1372 /* HW requires wqe size has room for atleast one SGE even if none
1373 * was supplied by ULP
1374 */
1375 if (!wqe->num_sge)
1376 rqe->wqe_size++;
1367 1377
1368 /* Supply the rqe->wr_id index to the wr_id_tbl for now */ 1378 /* Supply the rqe->wr_id index to the wr_id_tbl for now */
1369 rqe->wr_id[0] = cpu_to_le32(sw_prod); 1379 rqe->wr_id[0] = cpu_to_le32(sw_prod);
@@ -1885,6 +1895,25 @@ flush_rq:
1885 return rc; 1895 return rc;
1886} 1896}
1887 1897
1898bool bnxt_qplib_is_cq_empty(struct bnxt_qplib_cq *cq)
1899{
1900 struct cq_base *hw_cqe, **hw_cqe_ptr;
1901 unsigned long flags;
1902 u32 sw_cons, raw_cons;
1903 bool rc = true;
1904
1905 spin_lock_irqsave(&cq->hwq.lock, flags);
1906 raw_cons = cq->hwq.cons;
1907 sw_cons = HWQ_CMP(raw_cons, &cq->hwq);
1908 hw_cqe_ptr = (struct cq_base **)cq->hwq.pbl_ptr;
1909 hw_cqe = &hw_cqe_ptr[CQE_PG(sw_cons)][CQE_IDX(sw_cons)];
1910
1911 /* Check for Valid bit. If the CQE is valid, return false */
1912 rc = !CQE_CMP_VALID(hw_cqe, raw_cons, cq->hwq.max_elements);
1913 spin_unlock_irqrestore(&cq->hwq.lock, flags);
1914 return rc;
1915}
1916
1888static int bnxt_qplib_cq_process_res_raweth_qp1(struct bnxt_qplib_cq *cq, 1917static int bnxt_qplib_cq_process_res_raweth_qp1(struct bnxt_qplib_cq *cq,
1889 struct cq_res_raweth_qp1 *hwcqe, 1918 struct cq_res_raweth_qp1 *hwcqe,
1890 struct bnxt_qplib_cqe **pcqe, 1919 struct bnxt_qplib_cqe **pcqe,
diff --git a/drivers/infiniband/hw/bnxt_re/qplib_fp.h b/drivers/infiniband/hw/bnxt_re/qplib_fp.h
index 36b7b7db0e3f..19176e06c98a 100644
--- a/drivers/infiniband/hw/bnxt_re/qplib_fp.h
+++ b/drivers/infiniband/hw/bnxt_re/qplib_fp.h
@@ -449,6 +449,7 @@ int bnxt_qplib_create_cq(struct bnxt_qplib_res *res, struct bnxt_qplib_cq *cq);
449int bnxt_qplib_destroy_cq(struct bnxt_qplib_res *res, struct bnxt_qplib_cq *cq); 449int bnxt_qplib_destroy_cq(struct bnxt_qplib_res *res, struct bnxt_qplib_cq *cq);
450int bnxt_qplib_poll_cq(struct bnxt_qplib_cq *cq, struct bnxt_qplib_cqe *cqe, 450int bnxt_qplib_poll_cq(struct bnxt_qplib_cq *cq, struct bnxt_qplib_cqe *cqe,
451 int num, struct bnxt_qplib_qp **qp); 451 int num, struct bnxt_qplib_qp **qp);
452bool bnxt_qplib_is_cq_empty(struct bnxt_qplib_cq *cq);
452void bnxt_qplib_req_notify_cq(struct bnxt_qplib_cq *cq, u32 arm_type); 453void bnxt_qplib_req_notify_cq(struct bnxt_qplib_cq *cq, u32 arm_type);
453void bnxt_qplib_free_nq(struct bnxt_qplib_nq *nq); 454void bnxt_qplib_free_nq(struct bnxt_qplib_nq *nq);
454int bnxt_qplib_alloc_nq(struct pci_dev *pdev, struct bnxt_qplib_nq *nq); 455int bnxt_qplib_alloc_nq(struct pci_dev *pdev, struct bnxt_qplib_nq *nq);
diff --git a/drivers/infiniband/hw/bnxt_re/qplib_sp.c b/drivers/infiniband/hw/bnxt_re/qplib_sp.c
index fde18cf0e406..ef91ab786dd4 100644
--- a/drivers/infiniband/hw/bnxt_re/qplib_sp.c
+++ b/drivers/infiniband/hw/bnxt_re/qplib_sp.c
@@ -51,6 +51,19 @@ const struct bnxt_qplib_gid bnxt_qplib_gid_zero = {{ 0, 0, 0, 0, 0, 0, 0, 0,
51 0, 0, 0, 0, 0, 0, 0, 0 } }; 51 0, 0, 0, 0, 0, 0, 0, 0 } };
52 52
53/* Device */ 53/* Device */
54
55static bool bnxt_qplib_is_atomic_cap(struct bnxt_qplib_rcfw *rcfw)
56{
57 int rc;
58 u16 pcie_ctl2;
59
60 rc = pcie_capability_read_word(rcfw->pdev, PCI_EXP_DEVCTL2,
61 &pcie_ctl2);
62 if (rc)
63 return false;
64 return !!(pcie_ctl2 & PCI_EXP_DEVCTL2_ATOMIC_REQ);
65}
66
54int bnxt_qplib_get_dev_attr(struct bnxt_qplib_rcfw *rcfw, 67int bnxt_qplib_get_dev_attr(struct bnxt_qplib_rcfw *rcfw,
55 struct bnxt_qplib_dev_attr *attr) 68 struct bnxt_qplib_dev_attr *attr)
56{ 69{
@@ -81,6 +94,8 @@ int bnxt_qplib_get_dev_attr(struct bnxt_qplib_rcfw *rcfw,
81 94
82 /* Extract the context from the side buffer */ 95 /* Extract the context from the side buffer */
83 attr->max_qp = le32_to_cpu(sb->max_qp); 96 attr->max_qp = le32_to_cpu(sb->max_qp);
97 /* max_qp value reported by FW for PF doesn't include the QP1 for PF */
98 attr->max_qp += 1;
84 attr->max_qp_rd_atom = 99 attr->max_qp_rd_atom =
85 sb->max_qp_rd_atom > BNXT_QPLIB_MAX_OUT_RD_ATOM ? 100 sb->max_qp_rd_atom > BNXT_QPLIB_MAX_OUT_RD_ATOM ?
86 BNXT_QPLIB_MAX_OUT_RD_ATOM : sb->max_qp_rd_atom; 101 BNXT_QPLIB_MAX_OUT_RD_ATOM : sb->max_qp_rd_atom;
@@ -129,6 +144,7 @@ int bnxt_qplib_get_dev_attr(struct bnxt_qplib_rcfw *rcfw,
129 attr->tqm_alloc_reqs[i * 4 + 3] = *(++tqm_alloc); 144 attr->tqm_alloc_reqs[i * 4 + 3] = *(++tqm_alloc);
130 } 145 }
131 146
147 attr->is_atomic = bnxt_qplib_is_atomic_cap(rcfw);
132bail: 148bail:
133 bnxt_qplib_rcfw_free_sbuf(rcfw, sbuf); 149 bnxt_qplib_rcfw_free_sbuf(rcfw, sbuf);
134 return rc; 150 return rc;
diff --git a/drivers/infiniband/hw/bnxt_re/qplib_sp.h b/drivers/infiniband/hw/bnxt_re/qplib_sp.h
index a543f959098b..2ce7e2a32cf0 100644
--- a/drivers/infiniband/hw/bnxt_re/qplib_sp.h
+++ b/drivers/infiniband/hw/bnxt_re/qplib_sp.h
@@ -42,6 +42,8 @@
42 42
43#define BNXT_QPLIB_RESERVED_QP_WRS 128 43#define BNXT_QPLIB_RESERVED_QP_WRS 128
44 44
45#define PCI_EXP_DEVCTL2_ATOMIC_REQ 0x0040
46
45struct bnxt_qplib_dev_attr { 47struct bnxt_qplib_dev_attr {
46 char fw_ver[32]; 48 char fw_ver[32];
47 u16 max_sgid; 49 u16 max_sgid;
@@ -70,6 +72,7 @@ struct bnxt_qplib_dev_attr {
70 u32 max_inline_data; 72 u32 max_inline_data;
71 u32 l2_db_size; 73 u32 l2_db_size;
72 u8 tqm_alloc_reqs[MAX_TQM_ALLOC_REQ]; 74 u8 tqm_alloc_reqs[MAX_TQM_ALLOC_REQ];
75 bool is_atomic;
73}; 76};
74 77
75struct bnxt_qplib_pd { 78struct bnxt_qplib_pd {