From d6d142cb7f79bec6051c5ecf744b7a5309c5a0ee Mon Sep 17 00:00:00 2001 From: Sergey Gorenko Date: Fri, 5 Aug 2022 09:01:35 +0300 Subject: IB/iser: Fix login with authentication The iSER Initiator uses two types of receive buffers: - one big login buffer posted by iser_post_recvl(); - several small message buffers posted by iser_post_recvm(). The login buffer is used at the login phase and full feature phase in the discovery session. It may take a few requests and responses to complete the login phase. The message buffers are only used in the normal operational session at the full feature phase. After the commit referred in the fixes line, the login operation fails if the authentication is enabled. That happens because the Initiator posts a small receive buffer after the first response from Target. So, the next send operation fails because Target's second response does not fit into the small receive buffer. This commit adds additional checks to prevent posting small receive buffers until the full feature phase. Fixes: 39b169ea0d36 ("IB/iser: Fix RNR errors") Link: https://lore.kernel.org/r/20220805060135.18493-1-sergeygo@nvidia.com Signed-off-by: Sergey Gorenko Reviewed-by: Max Gurtovoy Signed-off-by: Leon Romanovsky --- drivers/infiniband/ulp/iser/iser_initiator.c | 7 ++++++- 1 file changed, 6 insertions(+), 1 deletion(-) (limited to 'drivers') diff --git a/drivers/infiniband/ulp/iser/iser_initiator.c b/drivers/infiniband/ulp/iser/iser_initiator.c index bd5f3b5e1727..7b83f48f60c5 100644 --- a/drivers/infiniband/ulp/iser/iser_initiator.c +++ b/drivers/infiniband/ulp/iser/iser_initiator.c @@ -537,6 +537,7 @@ void iser_login_rsp(struct ib_cq *cq, struct ib_wc *wc) struct iscsi_hdr *hdr; char *data; int length; + bool full_feature_phase; if (unlikely(wc->status != IB_WC_SUCCESS)) { iser_err_comp(wc, "login_rsp"); @@ -550,6 +551,9 @@ void iser_login_rsp(struct ib_cq *cq, struct ib_wc *wc) hdr = desc->rsp + sizeof(struct iser_ctrl); data = desc->rsp + ISER_HEADERS_LEN; length = wc->byte_len - ISER_HEADERS_LEN; + full_feature_phase = ((hdr->flags & ISCSI_FULL_FEATURE_PHASE) == + ISCSI_FULL_FEATURE_PHASE) && + (hdr->flags & ISCSI_FLAG_CMD_FINAL); iser_dbg("op 0x%x itt 0x%x dlen %d\n", hdr->opcode, hdr->itt, length); @@ -560,7 +564,8 @@ void iser_login_rsp(struct ib_cq *cq, struct ib_wc *wc) desc->rsp_dma, ISER_RX_LOGIN_SIZE, DMA_FROM_DEVICE); - if (iser_conn->iscsi_conn->session->discovery_sess) + if (!full_feature_phase || + iser_conn->iscsi_conn->session->discovery_sess) return; /* Post the first RX buffer that is skipped in iser_post_rx_bufs() */ -- cgit v1.2.3 From 4b83c3caf289b80acecc539c79f10a6937cc42dd Mon Sep 17 00:00:00 2001 From: Mark Bloch Date: Mon, 8 Aug 2022 10:48:06 +0300 Subject: RDMA/mlx5: Use the proper number of ports The cited commit allowed the driver to operate over HCAs that have 4 physical ports. Use the number of ports of the RDMA device in the for loop instead of using the struct size. Fixes: 4cd14d44b11d ("net/mlx5: Support devices with more than 2 ports") Link: https://lore.kernel.org/r/a54a56c2ede16044a29d119209b35189c662ac72.1659944855.git.leonro@nvidia.com Signed-off-by: Mark Bloch Signed-off-by: Leon Romanovsky --- drivers/infiniband/hw/mlx5/main.c | 34 ++++++++++++++++------------------ 1 file changed, 16 insertions(+), 18 deletions(-) (limited to 'drivers') diff --git a/drivers/infiniband/hw/mlx5/main.c b/drivers/infiniband/hw/mlx5/main.c index a174a0eee8dc..fc94a1b25485 100644 --- a/drivers/infiniband/hw/mlx5/main.c +++ b/drivers/infiniband/hw/mlx5/main.c @@ -2738,26 +2738,24 @@ static int set_has_smi_cap(struct mlx5_ib_dev *dev) int err; int port; - for (port = 1; port <= ARRAY_SIZE(dev->port_caps); port++) { - dev->port_caps[port - 1].has_smi = false; - if (MLX5_CAP_GEN(dev->mdev, port_type) == - MLX5_CAP_PORT_TYPE_IB) { - if (MLX5_CAP_GEN(dev->mdev, ib_virt)) { - err = mlx5_query_hca_vport_context(dev->mdev, 0, - port, 0, - &vport_ctx); - if (err) { - mlx5_ib_err(dev, "query_hca_vport_context for port=%d failed %d\n", - port, err); - return err; - } - dev->port_caps[port - 1].has_smi = - vport_ctx.has_smi; - } else { - dev->port_caps[port - 1].has_smi = true; - } + if (MLX5_CAP_GEN(dev->mdev, port_type) != MLX5_CAP_PORT_TYPE_IB) + return 0; + + for (port = 1; port <= dev->num_ports; port++) { + if (!MLX5_CAP_GEN(dev->mdev, ib_virt)) { + dev->port_caps[port - 1].has_smi = true; + continue; } + err = mlx5_query_hca_vport_context(dev->mdev, 0, port, 0, + &vport_ctx); + if (err) { + mlx5_ib_err(dev, "query_hca_vport_context for port=%d failed %d\n", + port, err); + return err; + } + dev->port_caps[port - 1].has_smi = vport_ctx.has_smi; } + return 0; } -- cgit v1.2.3 From ef0162298abf46b881e4a4d0c604d1a066228647 Mon Sep 17 00:00:00 2001 From: Potnuri Bharat Teja Date: Wed, 10 Aug 2022 00:11:18 +0530 Subject: RDMA/cxgb4: fix accept failure due to increased cpl_t5_pass_accept_rpl size Commit 'c2ed5611afd7' has increased the cpl_t5_pass_accept_rpl{} structure size by 8B to avoid roundup. cpl_t5_pass_accept_rpl{} is a HW specific structure and increasing its size will lead to unwanted adapter errors. Current commit reverts the cpl_t5_pass_accept_rpl{} back to its original and allocates zeroed skb buffer there by avoiding the memset for iss field. Reorder code to minimize chip type checks. Fixes: c2ed5611afd7 ("iw_cxgb4: Use memset_startat() for cpl_t5_pass_accept_rpl") Link: https://lore.kernel.org/r/20220809184118.2029-1-rahul.lakkireddy@chelsio.com Signed-off-by: Potnuri Bharat Teja Signed-off-by: Rahul Lakkireddy Reviewed-by: Kees Cook Signed-off-by: Leon Romanovsky --- drivers/infiniband/hw/cxgb4/cm.c | 25 +++++++++---------------- drivers/net/ethernet/chelsio/cxgb4/t4_msg.h | 2 +- 2 files changed, 10 insertions(+), 17 deletions(-) (limited to 'drivers') diff --git a/drivers/infiniband/hw/cxgb4/cm.c b/drivers/infiniband/hw/cxgb4/cm.c index c16017f6e8db..14392c942f49 100644 --- a/drivers/infiniband/hw/cxgb4/cm.c +++ b/drivers/infiniband/hw/cxgb4/cm.c @@ -2468,31 +2468,24 @@ static int accept_cr(struct c4iw_ep *ep, struct sk_buff *skb, opt2 |= CCTRL_ECN_V(1); } - skb_get(skb); - rpl = cplhdr(skb); if (!is_t4(adapter_type)) { - BUILD_BUG_ON(sizeof(*rpl5) != roundup(sizeof(*rpl5), 16)); - skb_trim(skb, sizeof(*rpl5)); - rpl5 = (void *)rpl; - INIT_TP_WR(rpl5, ep->hwtid); - } else { - skb_trim(skb, sizeof(*rpl)); - INIT_TP_WR(rpl, ep->hwtid); - } - OPCODE_TID(rpl) = cpu_to_be32(MK_OPCODE_TID(CPL_PASS_ACCEPT_RPL, - ep->hwtid)); - - if (CHELSIO_CHIP_VERSION(adapter_type) > CHELSIO_T4) { u32 isn = (prandom_u32() & ~7UL) - 1; + + skb = get_skb(skb, roundup(sizeof(*rpl5), 16), GFP_KERNEL); + rpl5 = __skb_put_zero(skb, roundup(sizeof(*rpl5), 16)); + rpl = (void *)rpl5; + INIT_TP_WR_CPL(rpl5, CPL_PASS_ACCEPT_RPL, ep->hwtid); opt2 |= T5_OPT_2_VALID_F; opt2 |= CONG_CNTRL_V(CONG_ALG_TAHOE); opt2 |= T5_ISS_F; - rpl5 = (void *)rpl; - memset_after(rpl5, 0, iss); if (peer2peer) isn += 4; rpl5->iss = cpu_to_be32(isn); pr_debug("iss %u\n", be32_to_cpu(rpl5->iss)); + } else { + skb = get_skb(skb, sizeof(*rpl), GFP_KERNEL); + rpl = __skb_put_zero(skb, sizeof(*rpl)); + INIT_TP_WR_CPL(rpl, CPL_PASS_ACCEPT_RPL, ep->hwtid); } rpl->opt0 = cpu_to_be64(opt0); diff --git a/drivers/net/ethernet/chelsio/cxgb4/t4_msg.h b/drivers/net/ethernet/chelsio/cxgb4/t4_msg.h index 26433a62d7f0..fed5f93bf620 100644 --- a/drivers/net/ethernet/chelsio/cxgb4/t4_msg.h +++ b/drivers/net/ethernet/chelsio/cxgb4/t4_msg.h @@ -497,7 +497,7 @@ struct cpl_t5_pass_accept_rpl { __be32 opt2; __be64 opt0; __be32 iss; - __be32 rsvd[3]; + __be32 rsvd; }; struct cpl_act_open_req { -- cgit v1.2.3 From 1374901b850c8f2ff1ee0e62ffafbc4030572fc6 Mon Sep 17 00:00:00 2001 From: Cheng Xu Date: Wed, 10 Aug 2022 09:43:20 +0800 Subject: RDMA/erdma: Using the key in FMR WR instead of MR structure RDMA driver should get the MR key from FMR WR, not the MR structure passed in. Fixes: 155055771704 ("RDMA/erdma: Add verbs implementation") Link: https://lore.kernel.org/r/20220810014320.88026-2-chengyou@linux.alibaba.com Signed-off-by: Cheng Xu Signed-off-by: Leon Romanovsky --- drivers/infiniband/hw/erdma/erdma_qp.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'drivers') diff --git a/drivers/infiniband/hw/erdma/erdma_qp.c b/drivers/infiniband/hw/erdma/erdma_qp.c index 72f08171a28a..bc3ec22a62c5 100644 --- a/drivers/infiniband/hw/erdma/erdma_qp.c +++ b/drivers/infiniband/hw/erdma/erdma_qp.c @@ -407,7 +407,7 @@ static int erdma_push_one_sqe(struct erdma_qp *qp, u16 *pi, to_erdma_access_flags(reg_wr(send_wr)->access); regmr_sge->addr = cpu_to_le64(mr->ibmr.iova); regmr_sge->length = cpu_to_le32(mr->ibmr.length); - regmr_sge->stag = cpu_to_le32(mr->ibmr.lkey); + regmr_sge->stag = cpu_to_le32(reg_wr(send_wr)->key); attrs = FIELD_PREP(ERDMA_SQE_MR_MODE_MASK, 0) | FIELD_PREP(ERDMA_SQE_MR_ACCESS_MASK, mr->access) | FIELD_PREP(ERDMA_SQE_MR_MTT_CNT_MASK, -- cgit v1.2.3 From 6cd8351c4ede5b3335c7eb13d009c3456e90b6df Mon Sep 17 00:00:00 2001 From: Cheng Xu Date: Wed, 10 Aug 2022 09:43:19 +0800 Subject: RDMA/erdma: Correct the max_qp and max_cq capacities of the device QP0 in HW is used for CMDQ, and the rest is for RDMA QPs. So the actual max_qp capacity reported to core should be max_qp (reported by HW) - 1. So does max_cq. Fixes: 155055771704 ("RDMA/erdma: Add verbs implementation") Link: https://lore.kernel.org/all/20220810014320.88026-1-chengyou@linux.alibaba.com Signed-off-by: Cheng Xu Signed-off-by: Leon Romanovsky --- drivers/infiniband/hw/erdma/erdma_verbs.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) (limited to 'drivers') diff --git a/drivers/infiniband/hw/erdma/erdma_verbs.c b/drivers/infiniband/hw/erdma/erdma_verbs.c index a7a3d42e2016..699bd3f59cd3 100644 --- a/drivers/infiniband/hw/erdma/erdma_verbs.c +++ b/drivers/infiniband/hw/erdma/erdma_verbs.c @@ -280,7 +280,7 @@ int erdma_query_device(struct ib_device *ibdev, struct ib_device_attr *attr, attr->vendor_id = PCI_VENDOR_ID_ALIBABA; attr->vendor_part_id = dev->pdev->device; attr->hw_ver = dev->pdev->revision; - attr->max_qp = dev->attrs.max_qp; + attr->max_qp = dev->attrs.max_qp - 1; attr->max_qp_wr = min(dev->attrs.max_send_wr, dev->attrs.max_recv_wr); attr->max_qp_rd_atom = dev->attrs.max_ord; attr->max_qp_init_rd_atom = dev->attrs.max_ird; @@ -291,7 +291,7 @@ int erdma_query_device(struct ib_device *ibdev, struct ib_device_attr *attr, attr->max_send_sge = dev->attrs.max_send_sge; attr->max_recv_sge = dev->attrs.max_recv_sge; attr->max_sge_rd = dev->attrs.max_sge_rd; - attr->max_cq = dev->attrs.max_cq; + attr->max_cq = dev->attrs.max_cq - 1; attr->max_cqe = dev->attrs.max_cqe; attr->max_mr = dev->attrs.max_mr; attr->max_pd = dev->attrs.max_pd; -- cgit v1.2.3 From b16de8b9e7d1aae169d059c3a0dd9a881a3c0d1d Mon Sep 17 00:00:00 2001 From: Jason Gunthorpe Date: Tue, 16 Aug 2022 11:03:20 -0300 Subject: RDMA: Handle the return code from dma_resv_wait_timeout() properly ib_umem_dmabuf_map_pages() returns 0 on success and -ERRNO on failure. dma_resv_wait_timeout() uses a different scheme: * Returns -ERESTARTSYS if interrupted, 0 if the wait timed out, or * greater than zero on success. This results in ib_umem_dmabuf_map_pages() being non-functional as a positive return will be understood to be an error by drivers. Fixes: f30bceab16d1 ("RDMA: use dma_resv_wait() instead of extracting the fence") Cc: stable@kernel.org Link: https://lore.kernel.org/r/0-v1-d8f4e1fa84c8+17-rdma_dmabuf_fix_jgg@nvidia.com Tested-by: Maor Gottlieb Signed-off-by: Jason Gunthorpe Signed-off-by: Leon Romanovsky --- drivers/infiniband/core/umem_dmabuf.c | 8 +++++++- 1 file changed, 7 insertions(+), 1 deletion(-) (limited to 'drivers') diff --git a/drivers/infiniband/core/umem_dmabuf.c b/drivers/infiniband/core/umem_dmabuf.c index fce80a4a5147..04c04e6d24c3 100644 --- a/drivers/infiniband/core/umem_dmabuf.c +++ b/drivers/infiniband/core/umem_dmabuf.c @@ -18,6 +18,7 @@ int ib_umem_dmabuf_map_pages(struct ib_umem_dmabuf *umem_dmabuf) struct scatterlist *sg; unsigned long start, end, cur = 0; unsigned int nmap = 0; + long ret; int i; dma_resv_assert_held(umem_dmabuf->attach->dmabuf->resv); @@ -67,9 +68,14 @@ wait_fence: * may be not up-to-date. Wait for the exporter to finish * the migration. */ - return dma_resv_wait_timeout(umem_dmabuf->attach->dmabuf->resv, + ret = dma_resv_wait_timeout(umem_dmabuf->attach->dmabuf->resv, DMA_RESV_USAGE_KERNEL, false, MAX_SCHEDULE_TIMEOUT); + if (ret < 0) + return ret; + if (ret == 0) + return -ETIMEDOUT; + return 0; } EXPORT_SYMBOL(ib_umem_dmabuf_map_pages); -- cgit v1.2.3