From 25d0e2db3d8f4f79fddd436dcc848d912e98b485 Mon Sep 17 00:00:00 2001 From: Zhu Yanjun Date: Sat, 14 Apr 2018 21:16:54 -0400 Subject: IB/mlx5: remove duplicate header file The header file fs_helpers.h is included twice. So it should be removed. Fixes: 802c2125689d ("IB/mlx5: Add IPsec support for egress and ingress") CC: Srinivas Eeda CC: Junxiao Bi Signed-off-by: Zhu Yanjun Acked-by: Leon Romanovsky Signed-off-by: Jason Gunthorpe --- drivers/infiniband/hw/mlx5/main.c | 1 - 1 file changed, 1 deletion(-) (limited to 'drivers/infiniband/hw') diff --git a/drivers/infiniband/hw/mlx5/main.c b/drivers/infiniband/hw/mlx5/main.c index daa919e5a442..6a749c02b14c 100644 --- a/drivers/infiniband/hw/mlx5/main.c +++ b/drivers/infiniband/hw/mlx5/main.c @@ -52,7 +52,6 @@ #include #include #include -#include #include #include #include -- cgit v1.2.3 From b3fe6c62bc66868c45b5bb16050e6bcb333af337 Mon Sep 17 00:00:00 2001 From: Randy Dunlap Date: Mon, 16 Apr 2018 18:51:50 -0700 Subject: infiniband: mlx5: fix build errors when INFINIBAND_USER_ACCESS=m Fix build errors when INFINIBAND_USER_ACCESS=m and MLX5_INFINIBAND=y. The build error occurs when the mlx5 driver code attempts to use USER_ACCESS interfaces, which are built as a loadable module. Fixes these build errors: drivers/infiniband/hw/mlx5/main.o: In function `populate_specs_root': ../drivers/infiniband/hw/mlx5/main.c:4982: undefined reference to `uverbs_default_get_objects' ../drivers/infiniband/hw/mlx5/main.c:4994: undefined reference to `uverbs_alloc_spec_tree' drivers/infiniband/hw/mlx5/main.o: In function `depopulate_specs_root': ../drivers/infiniband/hw/mlx5/main.c:5001: undefined reference to `uverbs_free_spec_tree' Build-tested with multiple config combinations. Fixes: 8c84660bb437 ("IB/mlx5: Initialize the parsing tree root without the help of uverbs") Cc: stable@vger.kernel.org # reported against 4.16 Reported-by: kbuild test robot Signed-off-by: Randy Dunlap Signed-off-by: Jason Gunthorpe --- drivers/infiniband/hw/mlx5/Kconfig | 1 + 1 file changed, 1 insertion(+) (limited to 'drivers/infiniband/hw') diff --git a/drivers/infiniband/hw/mlx5/Kconfig b/drivers/infiniband/hw/mlx5/Kconfig index bce263b92821..fb4d77be019b 100644 --- a/drivers/infiniband/hw/mlx5/Kconfig +++ b/drivers/infiniband/hw/mlx5/Kconfig @@ -1,6 +1,7 @@ config MLX5_INFINIBAND tristate "Mellanox Connect-IB HCA support" depends on NETDEVICES && ETHERNET && PCI && MLX5_CORE + depends on INFINIBAND_USER_ACCESS || INFINIBAND_USER_ACCESS=n ---help--- This driver provides low-level InfiniBand support for Mellanox Connect-IB PCI Express host channel adapters (HCAs). -- cgit v1.2.3 From b4bd701ac469075d94ed9699a28755f2862252b9 Mon Sep 17 00:00:00 2001 From: Leon Romanovsky Date: Mon, 23 Apr 2018 17:01:52 +0300 Subject: RDMA/mlx5: Fix multiple NULL-ptr deref errors in rereg_mr flow Failure in rereg MR releases UMEM but leaves the MR to be destroyed by the user. As a result the following scenario may happen: "create MR -> rereg MR with failure -> call to rereg MR again" and hit "NULL-ptr deref or user memory access" errors. Ensure that rereg MR is only performed on a non-dead MR. Cc: syzkaller Cc: # 4.5 Fixes: 395a8e4c32ea ("IB/mlx5: Refactoring register MR code") Reported-by: Noa Osherovich Signed-off-by: Leon Romanovsky Signed-off-by: Doug Ledford --- drivers/infiniband/hw/mlx5/mr.c | 32 +++++++++++++++++++++++--------- 1 file changed, 23 insertions(+), 9 deletions(-) (limited to 'drivers/infiniband/hw') diff --git a/drivers/infiniband/hw/mlx5/mr.c b/drivers/infiniband/hw/mlx5/mr.c index 1520a2f20f98..90a9c461cedc 100644 --- a/drivers/infiniband/hw/mlx5/mr.c +++ b/drivers/infiniband/hw/mlx5/mr.c @@ -866,25 +866,28 @@ static int mr_umem_get(struct ib_pd *pd, u64 start, u64 length, int *order) { struct mlx5_ib_dev *dev = to_mdev(pd->device); + struct ib_umem *u; int err; - *umem = ib_umem_get(pd->uobject->context, start, length, - access_flags, 0); - err = PTR_ERR_OR_ZERO(*umem); + *umem = NULL; + + u = ib_umem_get(pd->uobject->context, start, length, access_flags, 0); + err = PTR_ERR_OR_ZERO(u); if (err) { - *umem = NULL; - mlx5_ib_err(dev, "umem get failed (%d)\n", err); + mlx5_ib_dbg(dev, "umem get failed (%d)\n", err); return err; } - mlx5_ib_cont_pages(*umem, start, MLX5_MKEY_PAGE_SHIFT_MASK, npages, + mlx5_ib_cont_pages(u, start, MLX5_MKEY_PAGE_SHIFT_MASK, npages, page_shift, ncont, order); if (!*npages) { mlx5_ib_warn(dev, "avoid zero region\n"); - ib_umem_release(*umem); + ib_umem_release(u); return -EINVAL; } + *umem = u; + mlx5_ib_dbg(dev, "npages %d, ncont %d, order %d, page_shift %d\n", *npages, *ncont, *order, *page_shift); @@ -1458,13 +1461,12 @@ int mlx5_ib_rereg_user_mr(struct ib_mr *ib_mr, int flags, u64 start, int access_flags = flags & IB_MR_REREG_ACCESS ? new_access_flags : mr->access_flags; - u64 addr = (flags & IB_MR_REREG_TRANS) ? virt_addr : mr->umem->address; - u64 len = (flags & IB_MR_REREG_TRANS) ? length : mr->umem->length; int page_shift = 0; int upd_flags = 0; int npages = 0; int ncont = 0; int order = 0; + u64 addr, len; int err; mlx5_ib_dbg(dev, "start 0x%llx, virt_addr 0x%llx, length 0x%llx, access_flags 0x%x\n", @@ -1472,6 +1474,17 @@ int mlx5_ib_rereg_user_mr(struct ib_mr *ib_mr, int flags, u64 start, atomic_sub(mr->npages, &dev->mdev->priv.reg_pages); + if (!mr->umem) + return -EINVAL; + + if (flags & IB_MR_REREG_TRANS) { + addr = virt_addr; + len = length; + } else { + addr = mr->umem->address; + len = mr->umem->length; + } + if (flags != IB_MR_REREG_PD) { /* * Replace umem. This needs to be done whether or not UMR is @@ -1479,6 +1492,7 @@ int mlx5_ib_rereg_user_mr(struct ib_mr *ib_mr, int flags, u64 start, */ flags |= IB_MR_REREG_TRANS; ib_umem_release(mr->umem); + mr->umem = NULL; err = mr_umem_get(pd, addr, len, access_flags, &mr->umem, &npages, &page_shift, &ncont, &order); if (err) -- cgit v1.2.3 From 002bf2282b2d7318e444dca9ffcb994afc5d5f15 Mon Sep 17 00:00:00 2001 From: Leon Romanovsky Date: Mon, 23 Apr 2018 17:01:53 +0300 Subject: RDMA/mlx5: Protect from shift operand overflow Ensure that user didn't supply values too large that can cause overflow. UBSAN: Undefined behaviour in drivers/infiniband/hw/mlx5/qp.c:263:23 shift exponent -2147483648 is negative CPU: 0 PID: 292 Comm: syzkaller612609 Not tainted 4.16.0-rc1+ #131 Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS rel-1.11.0-0-g63451fca13-prebuilt.qemu-project.org 04/01/2014 Call Trace: dump_stack+0xde/0x164 ubsan_epilogue+0xe/0x81 set_rq_size+0x7c2/0xa90 create_qp_common+0xc18/0x43c0 mlx5_ib_create_qp+0x379/0x1ca0 create_qp.isra.5+0xc94/0x2260 ib_uverbs_create_qp+0x21b/0x2a0 ib_uverbs_write+0xc2c/0x1010 vfs_write+0x1b0/0x550 SyS_write+0xc7/0x1a0 do_syscall_64+0x1aa/0x740 entry_SYSCALL_64_after_hwframe+0x26/0x9b RIP: 0033:0x433569 RSP: 002b:00007ffc6e62f448 EFLAGS: 00000217 ORIG_RAX: 0000000000000001 RAX: ffffffffffffffda RBX: 00000000004002f8 RCX: 0000000000433569 RDX: 0000000000000070 RSI: 00000000200042c0 RDI: 0000000000000003 RBP: 00000000006d5018 R08: 00000000004002f8 R09: 00000000004002f8 R10: 00000000004002f8 R11: 0000000000000217 R12: 0000000000000000 R13: 000000000040c9f0 R14: 000000000040ca80 R15: 0000000000000006 Cc: # 3.10 Fixes: e126ba97dba9 ("mlx5: Add driver for Mellanox Connect-IB adapters") Cc: syzkaller Reported-by: Noa Osherovich Signed-off-by: Leon Romanovsky Signed-off-by: Doug Ledford --- drivers/infiniband/hw/mlx5/qp.c | 4 ++++ 1 file changed, 4 insertions(+) (limited to 'drivers/infiniband/hw') diff --git a/drivers/infiniband/hw/mlx5/qp.c b/drivers/infiniband/hw/mlx5/qp.c index 7ed4b70f6447..e6219a5f1f37 100644 --- a/drivers/infiniband/hw/mlx5/qp.c +++ b/drivers/infiniband/hw/mlx5/qp.c @@ -259,7 +259,11 @@ static int set_rq_size(struct mlx5_ib_dev *dev, struct ib_qp_cap *cap, } else { if (ucmd) { qp->rq.wqe_cnt = ucmd->rq_wqe_count; + if (ucmd->rq_wqe_shift > BITS_PER_BYTE * sizeof(ucmd->rq_wqe_shift)) + return -EINVAL; qp->rq.wqe_shift = ucmd->rq_wqe_shift; + if ((1 << qp->rq.wqe_shift) / sizeof(struct mlx5_wqe_data_seg) < qp->wq_sig) + return -EINVAL; qp->rq.max_gs = (1 << qp->rq.wqe_shift) / sizeof(struct mlx5_wqe_data_seg) - qp->wq_sig; qp->rq.max_post = qp->rq.wqe_cnt; } else { -- cgit v1.2.3 From 4f32ac2e452c2180cd2df581cbadac183e27ecd0 Mon Sep 17 00:00:00 2001 From: Danit Goldberg Date: Mon, 23 Apr 2018 17:01:54 +0300 Subject: IB/mlx5: Use unlimited rate when static rate is not supported Before the change, if the user passed a static rate value different than zero and the FW doesn't support static rate, it would end up configuring rate of 2.5 GBps. Fix this by using rate 0; unlimited, in cases where FW doesn't support static rate configuration. Cc: # 3.10 Fixes: e126ba97dba9 ("mlx5: Add driver for Mellanox Connect-IB adapters") Reviewed-by: Majd Dibbiny Signed-off-by: Danit Goldberg Signed-off-by: Leon Romanovsky Signed-off-by: Doug Ledford --- drivers/infiniband/hw/mlx5/qp.c | 18 +++++++++--------- 1 file changed, 9 insertions(+), 9 deletions(-) (limited to 'drivers/infiniband/hw') diff --git a/drivers/infiniband/hw/mlx5/qp.c b/drivers/infiniband/hw/mlx5/qp.c index e6219a5f1f37..87b7c1be2a11 100644 --- a/drivers/infiniband/hw/mlx5/qp.c +++ b/drivers/infiniband/hw/mlx5/qp.c @@ -2455,18 +2455,18 @@ enum { static int ib_rate_to_mlx5(struct mlx5_ib_dev *dev, u8 rate) { - if (rate == IB_RATE_PORT_CURRENT) { + if (rate == IB_RATE_PORT_CURRENT) return 0; - } else if (rate < IB_RATE_2_5_GBPS || rate > IB_RATE_300_GBPS) { + + if (rate < IB_RATE_2_5_GBPS || rate > IB_RATE_300_GBPS) return -EINVAL; - } else { - while (rate != IB_RATE_2_5_GBPS && - !(1 << (rate + MLX5_STAT_RATE_OFFSET) & - MLX5_CAP_GEN(dev->mdev, stat_rate_support))) - --rate; - } - return rate + MLX5_STAT_RATE_OFFSET; + while (rate != IB_RATE_PORT_CURRENT && + !(1 << (rate + MLX5_STAT_RATE_OFFSET) & + MLX5_CAP_GEN(dev->mdev, stat_rate_support))) + --rate; + + return rate ? rate + MLX5_STAT_RATE_OFFSET : rate; } static int modify_raw_packet_eth_prio(struct mlx5_core_dev *dev, -- cgit v1.2.3 From 84a6a7a99c0ac2f67366288c0625c9fba176b264 Mon Sep 17 00:00:00 2001 From: Parav Pandit Date: Mon, 23 Apr 2018 17:01:55 +0300 Subject: IB/mlx5: Fix represent correct netdevice in dual port RoCE In commit bcf87f1dbbec ("IB/mlx5: Listen to netdev register/unresiter events in switchdev mode") incorrectly mapped primary device's netdevice to 2nd port netdevice. It always represented primary port's netdevice for 2nd port netdevice when ib representors were not used. This results into failing to process CM request arriving on 2nd port due to incorrect mapping of netdevice. This fix corrects it by considering the right mdev. Cc: # 4.16 Fixes: bcf87f1dbbec ("IB/mlx5: Listen to netdev register/unresiter events in switchdev mode") Reviewed-by: Mark Bloch Signed-off-by: Parav Pandit Signed-off-by: Leon Romanovsky Signed-off-by: Doug Ledford --- drivers/infiniband/hw/mlx5/main.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'drivers/infiniband/hw') diff --git a/drivers/infiniband/hw/mlx5/main.c b/drivers/infiniband/hw/mlx5/main.c index 6a749c02b14c..78a4b2797057 100644 --- a/drivers/infiniband/hw/mlx5/main.c +++ b/drivers/infiniband/hw/mlx5/main.c @@ -179,7 +179,7 @@ static int mlx5_netdev_event(struct notifier_block *this, if (rep_ndev == ndev) roce->netdev = (event == NETDEV_UNREGISTER) ? NULL : ndev; - } else if (ndev->dev.parent == &ibdev->mdev->pdev->dev) { + } else if (ndev->dev.parent == &mdev->pdev->dev) { roce->netdev = (event == NETDEV_UNREGISTER) ? NULL : ndev; } -- cgit v1.2.3 From 444261ca6ff201fa03de97a5041237e67a9d8d31 Mon Sep 17 00:00:00 2001 From: Leon Romanovsky Date: Mon, 23 Apr 2018 17:01:56 +0300 Subject: RDMA/mlx5: Properly check return value of mlx5_get_uars_page Starting from commit 72f36be06138 ("net/mlx5: Fix mlx5_get_uars_page to return error code") the mlx5_get_uars_page() call returns error in case of failure, but it was mistakenly overlooked in the merge commit. Fixes: e7996a9a77fc ("Merge tag v4.15 of git://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux-2.6.git") Reported-by: Alaa Hleihel Signed-off-by: Leon Romanovsky Signed-off-by: Doug Ledford --- drivers/infiniband/hw/mlx5/main.c | 4 +--- 1 file changed, 1 insertion(+), 3 deletions(-) (limited to 'drivers/infiniband/hw') diff --git a/drivers/infiniband/hw/mlx5/main.c b/drivers/infiniband/hw/mlx5/main.c index 78a4b2797057..a42c6b1cdb5a 100644 --- a/drivers/infiniband/hw/mlx5/main.c +++ b/drivers/infiniband/hw/mlx5/main.c @@ -5426,9 +5426,7 @@ static void mlx5_ib_stage_cong_debugfs_cleanup(struct mlx5_ib_dev *dev) static int mlx5_ib_stage_uar_init(struct mlx5_ib_dev *dev) { dev->mdev->priv.uar = mlx5_get_uars_page(dev->mdev); - if (!dev->mdev->priv.uar) - return -ENOMEM; - return 0; + return PTR_ERR_OR_ZERO(dev->mdev->priv.uar); } static void mlx5_ib_stage_uar_cleanup(struct mlx5_ib_dev *dev) -- cgit v1.2.3 From 26bff1bd74a4f7417509a83295614e9dab995b2a Mon Sep 17 00:00:00 2001 From: Raju Rangoju Date: Mon, 23 Apr 2018 21:42:37 +0530 Subject: RDMA/cxgb4: release hw resources on device removal The c4iw_rdev_close() logic was not releasing all the hw resources (PBL and RQT memory) during the device removal event (driver unload / system reboot). This can cause panic in gen_pool_destroy(). The module remove function will wait for all the hw resources to be released during the device removal event. Fixes c12a67fe(iw_cxgb4: free EQ queue memory on last deref) Signed-off-by: Raju Rangoju Reviewed-by: Steve Wise Cc: stable@vger.kernel.org Signed-off-by: Doug Ledford --- drivers/infiniband/hw/cxgb4/device.c | 9 ++++++++- drivers/infiniband/hw/cxgb4/iw_cxgb4.h | 4 ++++ drivers/infiniband/hw/cxgb4/resource.c | 26 ++++++++++++++++++++++++-- 3 files changed, 36 insertions(+), 3 deletions(-) (limited to 'drivers/infiniband/hw') diff --git a/drivers/infiniband/hw/cxgb4/device.c b/drivers/infiniband/hw/cxgb4/device.c index feeb8ee6f4a2..44161ca4d2a8 100644 --- a/drivers/infiniband/hw/cxgb4/device.c +++ b/drivers/infiniband/hw/cxgb4/device.c @@ -875,6 +875,11 @@ static int c4iw_rdev_open(struct c4iw_rdev *rdev) rdev->status_page->db_off = 0; + init_completion(&rdev->rqt_compl); + init_completion(&rdev->pbl_compl); + kref_init(&rdev->rqt_kref); + kref_init(&rdev->pbl_kref); + return 0; err_free_status_page_and_wr_log: if (c4iw_wr_log && rdev->wr_log) @@ -893,13 +898,15 @@ destroy_resource: static void c4iw_rdev_close(struct c4iw_rdev *rdev) { - destroy_workqueue(rdev->free_workq); kfree(rdev->wr_log); c4iw_release_dev_ucontext(rdev, &rdev->uctx); free_page((unsigned long)rdev->status_page); c4iw_pblpool_destroy(rdev); c4iw_rqtpool_destroy(rdev); + wait_for_completion(&rdev->pbl_compl); + wait_for_completion(&rdev->rqt_compl); c4iw_ocqp_pool_destroy(rdev); + destroy_workqueue(rdev->free_workq); c4iw_destroy_resource(&rdev->resource); } diff --git a/drivers/infiniband/hw/cxgb4/iw_cxgb4.h b/drivers/infiniband/hw/cxgb4/iw_cxgb4.h index cc929002c05e..a60def23e9ef 100644 --- a/drivers/infiniband/hw/cxgb4/iw_cxgb4.h +++ b/drivers/infiniband/hw/cxgb4/iw_cxgb4.h @@ -185,6 +185,10 @@ struct c4iw_rdev { struct wr_log_entry *wr_log; int wr_log_size; struct workqueue_struct *free_workq; + struct completion rqt_compl; + struct completion pbl_compl; + struct kref rqt_kref; + struct kref pbl_kref; }; static inline int c4iw_fatal_error(struct c4iw_rdev *rdev) diff --git a/drivers/infiniband/hw/cxgb4/resource.c b/drivers/infiniband/hw/cxgb4/resource.c index 3cf25997ed2b..0ef25ae05e6f 100644 --- a/drivers/infiniband/hw/cxgb4/resource.c +++ b/drivers/infiniband/hw/cxgb4/resource.c @@ -260,12 +260,22 @@ u32 c4iw_pblpool_alloc(struct c4iw_rdev *rdev, int size) rdev->stats.pbl.cur += roundup(size, 1 << MIN_PBL_SHIFT); if (rdev->stats.pbl.cur > rdev->stats.pbl.max) rdev->stats.pbl.max = rdev->stats.pbl.cur; + kref_get(&rdev->pbl_kref); } else rdev->stats.pbl.fail++; mutex_unlock(&rdev->stats.lock); return (u32)addr; } +static void destroy_pblpool(struct kref *kref) +{ + struct c4iw_rdev *rdev; + + rdev = container_of(kref, struct c4iw_rdev, pbl_kref); + gen_pool_destroy(rdev->pbl_pool); + complete(&rdev->pbl_compl); +} + void c4iw_pblpool_free(struct c4iw_rdev *rdev, u32 addr, int size) { pr_debug("addr 0x%x size %d\n", addr, size); @@ -273,6 +283,7 @@ void c4iw_pblpool_free(struct c4iw_rdev *rdev, u32 addr, int size) rdev->stats.pbl.cur -= roundup(size, 1 << MIN_PBL_SHIFT); mutex_unlock(&rdev->stats.lock); gen_pool_free(rdev->pbl_pool, (unsigned long)addr, size); + kref_put(&rdev->pbl_kref, destroy_pblpool); } int c4iw_pblpool_create(struct c4iw_rdev *rdev) @@ -310,7 +321,7 @@ int c4iw_pblpool_create(struct c4iw_rdev *rdev) void c4iw_pblpool_destroy(struct c4iw_rdev *rdev) { - gen_pool_destroy(rdev->pbl_pool); + kref_put(&rdev->pbl_kref, destroy_pblpool); } /* @@ -331,12 +342,22 @@ u32 c4iw_rqtpool_alloc(struct c4iw_rdev *rdev, int size) rdev->stats.rqt.cur += roundup(size << 6, 1 << MIN_RQT_SHIFT); if (rdev->stats.rqt.cur > rdev->stats.rqt.max) rdev->stats.rqt.max = rdev->stats.rqt.cur; + kref_get(&rdev->rqt_kref); } else rdev->stats.rqt.fail++; mutex_unlock(&rdev->stats.lock); return (u32)addr; } +static void destroy_rqtpool(struct kref *kref) +{ + struct c4iw_rdev *rdev; + + rdev = container_of(kref, struct c4iw_rdev, rqt_kref); + gen_pool_destroy(rdev->rqt_pool); + complete(&rdev->rqt_compl); +} + void c4iw_rqtpool_free(struct c4iw_rdev *rdev, u32 addr, int size) { pr_debug("addr 0x%x size %d\n", addr, size << 6); @@ -344,6 +365,7 @@ void c4iw_rqtpool_free(struct c4iw_rdev *rdev, u32 addr, int size) rdev->stats.rqt.cur -= roundup(size << 6, 1 << MIN_RQT_SHIFT); mutex_unlock(&rdev->stats.lock); gen_pool_free(rdev->rqt_pool, (unsigned long)addr, size << 6); + kref_put(&rdev->rqt_kref, destroy_rqtpool); } int c4iw_rqtpool_create(struct c4iw_rdev *rdev) @@ -380,7 +402,7 @@ int c4iw_rqtpool_create(struct c4iw_rdev *rdev) void c4iw_rqtpool_destroy(struct c4iw_rdev *rdev) { - gen_pool_destroy(rdev->rqt_pool); + kref_put(&rdev->rqt_kref, destroy_rqtpool); } /* -- cgit v1.2.3 From c192a12ce82ca0a951bd5449a21d53c5f3f88697 Mon Sep 17 00:00:00 2001 From: Luc Van Oostenryck Date: Tue, 24 Apr 2018 15:15:45 +0200 Subject: IB/nes: fix nes_netdev_start_xmit()'s return type The method ndo_start_xmit() is defined as returning an 'netdev_tx_t', which is a typedef for an enum type, but the implementation in this driver returns an 'int'. Fix this by returning 'netdev_tx_t' in this driver too. Signed-off-by: Luc Van Oostenryck Reviewed-by: Leon Romanovsky Signed-off-by: Doug Ledford --- drivers/infiniband/hw/nes/nes_nic.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'drivers/infiniband/hw') diff --git a/drivers/infiniband/hw/nes/nes_nic.c b/drivers/infiniband/hw/nes/nes_nic.c index 0a75164cedea..007d5e8a0121 100644 --- a/drivers/infiniband/hw/nes/nes_nic.c +++ b/drivers/infiniband/hw/nes/nes_nic.c @@ -461,7 +461,7 @@ static bool nes_nic_send(struct sk_buff *skb, struct net_device *netdev) /** * nes_netdev_start_xmit */ -static int nes_netdev_start_xmit(struct sk_buff *skb, struct net_device *netdev) +static netdev_tx_t nes_netdev_start_xmit(struct sk_buff *skb, struct net_device *netdev) { struct nes_vnic *nesvnic = netdev_priv(netdev); struct nes_device *nesdev = nesvnic->nesdev; -- cgit v1.2.3 From 215a8c09e5e2aa6ae1fbcef87f8f27d65d5d1ca4 Mon Sep 17 00:00:00 2001 From: oulijun Date: Thu, 26 Apr 2018 14:46:15 +0800 Subject: RDMA/hns: Bugfix for init hem table During init hem table, type should be used instead of table->type which is finally initializaed with type. Signed-off-by: Lijun Ou Signed-off-by: Yixian Liu Signed-off-by: Doug Ledford --- drivers/infiniband/hw/hns/hns_roce_hem.c | 8 ++++---- 1 file changed, 4 insertions(+), 4 deletions(-) (limited to 'drivers/infiniband/hw') diff --git a/drivers/infiniband/hw/hns/hns_roce_hem.c b/drivers/infiniband/hw/hns/hns_roce_hem.c index 0eeabfbee192..0d8c113083ad 100644 --- a/drivers/infiniband/hw/hns/hns_roce_hem.c +++ b/drivers/infiniband/hw/hns/hns_roce_hem.c @@ -912,7 +912,7 @@ int hns_roce_init_hem_table(struct hns_roce_dev *hr_dev, obj_per_chunk = buf_chunk_size / obj_size; num_hem = (nobj + obj_per_chunk - 1) / obj_per_chunk; bt_chunk_num = bt_chunk_size / 8; - if (table->type >= HEM_TYPE_MTT) + if (type >= HEM_TYPE_MTT) num_bt_l0 = bt_chunk_num; table->hem = kcalloc(num_hem, sizeof(*table->hem), @@ -920,7 +920,7 @@ int hns_roce_init_hem_table(struct hns_roce_dev *hr_dev, if (!table->hem) goto err_kcalloc_hem_buf; - if (check_whether_bt_num_3(table->type, hop_num)) { + if (check_whether_bt_num_3(type, hop_num)) { unsigned long num_bt_l1; num_bt_l1 = (num_hem + bt_chunk_num - 1) / @@ -939,8 +939,8 @@ int hns_roce_init_hem_table(struct hns_roce_dev *hr_dev, goto err_kcalloc_l1_dma; } - if (check_whether_bt_num_2(table->type, hop_num) || - check_whether_bt_num_3(table->type, hop_num)) { + if (check_whether_bt_num_2(type, hop_num) || + check_whether_bt_num_3(type, hop_num)) { table->bt_l0 = kcalloc(num_bt_l0, sizeof(*table->bt_l0), GFP_KERNEL); if (!table->bt_l0) -- cgit v1.2.3 From 328d405b3d4c8dd1f06bfd77f498e23281ae348c Mon Sep 17 00:00:00 2001 From: oulijun Date: Thu, 26 Apr 2018 14:46:16 +0800 Subject: RDMA/hns: Intercept illegal RDMA operation when use inline data RDMA read operation is not supported inline data. If user cofigures issue a RDMA read and use inline data, it will happen a hardware error. Signed-off-by: Lijun Ou Signed-off-by: Doug Ledford --- drivers/infiniband/hw/hns/hns_roce_hw_v2.c | 5 +++++ 1 file changed, 5 insertions(+) (limited to 'drivers/infiniband/hw') diff --git a/drivers/infiniband/hw/hns/hns_roce_hw_v2.c b/drivers/infiniband/hw/hns/hns_roce_hw_v2.c index 8b84ab7800d8..aa5f9b3e1c10 100644 --- a/drivers/infiniband/hw/hns/hns_roce_hw_v2.c +++ b/drivers/infiniband/hw/hns/hns_roce_hw_v2.c @@ -71,6 +71,11 @@ static int set_rwqe_data_seg(struct ib_qp *ibqp, struct ib_send_wr *wr, return -EINVAL; } + if (wr->opcode == IB_WR_RDMA_READ) { + dev_err(hr_dev->dev, "Not support inline data!\n"); + return -EINVAL; + } + for (i = 0; i < wr->num_sge; i++) { memcpy(wqe, ((void *)wr->sg_list[i].addr), wr->sg_list[i].length); -- cgit v1.2.3 From 6e1a70943cecdca9bb13b601b1a9772a7bdcc2c3 Mon Sep 17 00:00:00 2001 From: oulijun Date: Thu, 26 Apr 2018 14:46:17 +0800 Subject: RDMA/hns: Fix the qp context state diagram According to RoCE protocol, it is possible to transition from error to error state for modifying qp in hip08. This patch fix it. Signed-off-by: Lijun Ou Signed-off-by: Doug Ledford --- drivers/infiniband/hw/hns/hns_roce_hw_v2.c | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-) (limited to 'drivers/infiniband/hw') diff --git a/drivers/infiniband/hw/hns/hns_roce_hw_v2.c b/drivers/infiniband/hw/hns/hns_roce_hw_v2.c index aa5f9b3e1c10..a4eea701d19d 100644 --- a/drivers/infiniband/hw/hns/hns_roce_hw_v2.c +++ b/drivers/infiniband/hw/hns/hns_roce_hw_v2.c @@ -3166,7 +3166,8 @@ static int hns_roce_v2_modify_qp(struct ib_qp *ibqp, (cur_state == IB_QPS_RTR && new_state == IB_QPS_ERR) || (cur_state == IB_QPS_RTS && new_state == IB_QPS_ERR) || (cur_state == IB_QPS_SQD && new_state == IB_QPS_ERR) || - (cur_state == IB_QPS_SQE && new_state == IB_QPS_ERR)) { + (cur_state == IB_QPS_SQE && new_state == IB_QPS_ERR) || + (cur_state == IB_QPS_ERR && new_state == IB_QPS_ERR)) { /* Nothing */ ; } else { -- cgit v1.2.3 From 6852af86627c7bd8de11c9ad3eb5cca7d99e5884 Mon Sep 17 00:00:00 2001 From: oulijun Date: Thu, 26 Apr 2018 14:46:18 +0800 Subject: RDMA/hns: Only assign mtu if IB_QP_PATH_MTU bit is set Only when the IB_QP_PATH_MTU flag of attr_mask is set it is valid to assign the mtu field of qp context when qp type is not GSI and UD. Signed-off-by: Lijun Ou Signed-off-by: Doug Ledford --- drivers/infiniband/hw/hns/hns_roce_hw_v2.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'drivers/infiniband/hw') diff --git a/drivers/infiniband/hw/hns/hns_roce_hw_v2.c b/drivers/infiniband/hw/hns/hns_roce_hw_v2.c index a4eea701d19d..9a3148103a5f 100644 --- a/drivers/infiniband/hw/hns/hns_roce_hw_v2.c +++ b/drivers/infiniband/hw/hns/hns_roce_hw_v2.c @@ -2850,7 +2850,7 @@ static int modify_qp_init_to_rtr(struct ib_qp *ibqp, if (ibqp->qp_type == IB_QPT_GSI || ibqp->qp_type == IB_QPT_UD) roce_set_field(context->byte_24_mtu_tc, V2_QPC_BYTE_24_MTU_M, V2_QPC_BYTE_24_MTU_S, IB_MTU_4096); - else + else if (attr_mask & IB_QP_PATH_MTU) roce_set_field(context->byte_24_mtu_tc, V2_QPC_BYTE_24_MTU_M, V2_QPC_BYTE_24_MTU_S, attr->path_mtu); -- cgit v1.2.3 From 734f38638d7b91c173ddfe46ceadb0382fae9b1f Mon Sep 17 00:00:00 2001 From: oulijun Date: Thu, 26 Apr 2018 14:46:19 +0800 Subject: RDMA/hns: Remove some unnecessary attr_mask judgement This patch deletes some unnecessary attr_mask if condition in hip08 according to the IB protocol. Signed-off-by: Lijun Ou Signed-off-by: Doug Ledford --- drivers/infiniband/hw/hns/hns_roce_hw_v2.c | 11 ++++------- 1 file changed, 4 insertions(+), 7 deletions(-) (limited to 'drivers/infiniband/hw') diff --git a/drivers/infiniband/hw/hns/hns_roce_hw_v2.c b/drivers/infiniband/hw/hns/hns_roce_hw_v2.c index 9a3148103a5f..97850f41a633 100644 --- a/drivers/infiniband/hw/hns/hns_roce_hw_v2.c +++ b/drivers/infiniband/hw/hns/hns_roce_hw_v2.c @@ -2655,8 +2655,7 @@ static int modify_qp_init_to_rtr(struct ib_qp *ibqp, return -EINVAL; } - if ((attr_mask & IB_QP_ALT_PATH) || (attr_mask & IB_QP_ACCESS_FLAGS) || - (attr_mask & IB_QP_PKEY_INDEX) || (attr_mask & IB_QP_QKEY)) { + if (attr_mask & IB_QP_ALT_PATH) { dev_err(dev, "INIT2RTR attr_mask (0x%x) error\n", attr_mask); return -EINVAL; } @@ -2927,11 +2926,9 @@ static int modify_qp_rtr_to_rts(struct ib_qp *ibqp, return -EINVAL; } - /* If exist optional param, return error */ - if ((attr_mask & IB_QP_ALT_PATH) || (attr_mask & IB_QP_ACCESS_FLAGS) || - (attr_mask & IB_QP_QKEY) || (attr_mask & IB_QP_PATH_MIG_STATE) || - (attr_mask & IB_QP_CUR_STATE) || - (attr_mask & IB_QP_MIN_RNR_TIMER)) { + /* Not support alternate path and path migration */ + if ((attr_mask & IB_QP_ALT_PATH) || + (attr_mask & IB_QP_PATH_MIG_STATE)) { dev_err(dev, "RTR2RTS attr_mask (0x%x)error\n", attr_mask); return -EINVAL; } -- cgit v1.2.3 From b6dd9b34838506f5307850d8d320d7b1ac6761d3 Mon Sep 17 00:00:00 2001 From: oulijun Date: Thu, 26 Apr 2018 14:46:20 +0800 Subject: RDMA/hns: Only assign dqpn if IB_QP_PATH_DEST_QPN bit is set Only when the IB_QP_PATH_DEST_QPN flag of attr_mask is set is it valid to assign the dqpn field of qp context Signed-off-by: Lijun Ou Signed-off-by: Doug Ledford --- drivers/infiniband/hw/hns/hns_roce_hw_v2.c | 20 ++++++++++++-------- 1 file changed, 12 insertions(+), 8 deletions(-) (limited to 'drivers/infiniband/hw') diff --git a/drivers/infiniband/hw/hns/hns_roce_hw_v2.c b/drivers/infiniband/hw/hns/hns_roce_hw_v2.c index 97850f41a633..51ec60a9cfd5 100644 --- a/drivers/infiniband/hw/hns/hns_roce_hw_v2.c +++ b/drivers/infiniband/hw/hns/hns_roce_hw_v2.c @@ -2597,10 +2597,12 @@ static void modify_qp_init_to_init(struct ib_qp *ibqp, roce_set_field(qpc_mask->byte_4_sqpn_tst, V2_QPC_BYTE_4_SQPN_M, V2_QPC_BYTE_4_SQPN_S, 0); - roce_set_field(context->byte_56_dqpn_err, V2_QPC_BYTE_56_DQPN_M, - V2_QPC_BYTE_56_DQPN_S, hr_qp->qpn); - roce_set_field(qpc_mask->byte_56_dqpn_err, V2_QPC_BYTE_56_DQPN_M, - V2_QPC_BYTE_56_DQPN_S, 0); + if (attr_mask & IB_QP_DEST_QPN) { + roce_set_field(context->byte_56_dqpn_err, V2_QPC_BYTE_56_DQPN_M, + V2_QPC_BYTE_56_DQPN_S, hr_qp->qpn); + roce_set_field(qpc_mask->byte_56_dqpn_err, + V2_QPC_BYTE_56_DQPN_M, V2_QPC_BYTE_56_DQPN_S, 0); + } roce_set_field(context->byte_168_irrl_idx, V2_QPC_BYTE_168_SQ_SHIFT_BAK_M, V2_QPC_BYTE_168_SQ_SHIFT_BAK_S, @@ -2804,10 +2806,12 @@ static int modify_qp_init_to_rtr(struct ib_qp *ibqp, V2_QPC_BYTE_140_RR_MAX_S, 0); } - roce_set_field(context->byte_56_dqpn_err, V2_QPC_BYTE_56_DQPN_M, - V2_QPC_BYTE_56_DQPN_S, attr->dest_qp_num); - roce_set_field(qpc_mask->byte_56_dqpn_err, V2_QPC_BYTE_56_DQPN_M, - V2_QPC_BYTE_56_DQPN_S, 0); + if (attr_mask & IB_QP_DEST_QPN) { + roce_set_field(context->byte_56_dqpn_err, V2_QPC_BYTE_56_DQPN_M, + V2_QPC_BYTE_56_DQPN_S, attr->dest_qp_num); + roce_set_field(qpc_mask->byte_56_dqpn_err, + V2_QPC_BYTE_56_DQPN_M, V2_QPC_BYTE_56_DQPN_S, 0); + } /* Configure GID index */ port_num = rdma_ah_get_port_num(&attr->ah_attr); -- cgit v1.2.3 From ae25db00285bf664fbd78222491b29aac6242902 Mon Sep 17 00:00:00 2001 From: oulijun Date: Thu, 26 Apr 2018 14:46:21 +0800 Subject: RDMA/hns: Adjust the order of cleanup hem table This patch update the order of cleaning hem table for trrl_table and irrl_table as well as mtt_cqe_table and mtt_table. Signed-off-by: Lijun Ou Signed-off-by: Doug Ledford --- drivers/infiniband/hw/hns/hns_roce_hem.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) (limited to 'drivers/infiniband/hw') diff --git a/drivers/infiniband/hw/hns/hns_roce_hem.c b/drivers/infiniband/hw/hns/hns_roce_hem.c index 0d8c113083ad..63b5b3edabcb 100644 --- a/drivers/infiniband/hw/hns/hns_roce_hem.c +++ b/drivers/infiniband/hw/hns/hns_roce_hem.c @@ -1039,14 +1039,14 @@ void hns_roce_cleanup_hem_table(struct hns_roce_dev *hr_dev, void hns_roce_cleanup_hem(struct hns_roce_dev *hr_dev) { hns_roce_cleanup_hem_table(hr_dev, &hr_dev->cq_table.table); - hns_roce_cleanup_hem_table(hr_dev, &hr_dev->qp_table.irrl_table); if (hr_dev->caps.trrl_entry_sz) hns_roce_cleanup_hem_table(hr_dev, &hr_dev->qp_table.trrl_table); + hns_roce_cleanup_hem_table(hr_dev, &hr_dev->qp_table.irrl_table); hns_roce_cleanup_hem_table(hr_dev, &hr_dev->qp_table.qp_table); hns_roce_cleanup_hem_table(hr_dev, &hr_dev->mr_table.mtpt_table); - hns_roce_cleanup_hem_table(hr_dev, &hr_dev->mr_table.mtt_table); if (hns_roce_check_whether_mhop(hr_dev, HEM_TYPE_CQE)) hns_roce_cleanup_hem_table(hr_dev, &hr_dev->mr_table.mtt_cqe_table); + hns_roce_cleanup_hem_table(hr_dev, &hr_dev->mr_table.mtt_table); } -- cgit v1.2.3 From 634f63902277700d0a02eaa101cc685958c35aa4 Mon Sep 17 00:00:00 2001 From: oulijun Date: Thu, 26 Apr 2018 14:46:22 +0800 Subject: RDMA/hns: Update assignment method for owner field of send wqe When posting a work reqeust, it need to update the owner bit of send wqe. This patch mainly fix the bug when posting multiply work request. Signed-off-by: Lijun Ou Signed-off-by: Doug Ledford --- drivers/infiniband/hw/hns/hns_roce_hw_v2.c | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-) (limited to 'drivers/infiniband/hw') diff --git a/drivers/infiniband/hw/hns/hns_roce_hw_v2.c b/drivers/infiniband/hw/hns/hns_roce_hw_v2.c index 51ec60a9cfd5..8ea5470303ee 100644 --- a/drivers/infiniband/hw/hns/hns_roce_hw_v2.c +++ b/drivers/infiniband/hw/hns/hns_roce_hw_v2.c @@ -187,7 +187,8 @@ static int hns_roce_v2_post_send(struct ib_qp *ibqp, struct ib_send_wr *wr, qp->sq.wrid[(qp->sq.head + nreq) & (qp->sq.wqe_cnt - 1)] = wr->wr_id; - owner_bit = ~(qp->sq.head >> ilog2(qp->sq.wqe_cnt)) & 0x1; + owner_bit = + ~(((qp->sq.head + nreq) >> ilog2(qp->sq.wqe_cnt)) & 0x1); /* Corresponding to the QP type, wqe process separately */ if (ibqp->qp_type == IB_QPT_GSI) { -- cgit v1.2.3 From 137ae3208416278aabef3b71e0ea1052940ca362 Mon Sep 17 00:00:00 2001 From: oulijun Date: Thu, 26 Apr 2018 14:46:23 +0800 Subject: RDMA/hns: Submit bad wr When generated bad work reqeust, it needs to report to user. This patch mainly fixes it. Signed-off-by: Lijun Ou Signed-off-by: Doug Ledford --- drivers/infiniband/hw/hns/hns_roce_hw_v2.c | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-) (limited to 'drivers/infiniband/hw') diff --git a/drivers/infiniband/hw/hns/hns_roce_hw_v2.c b/drivers/infiniband/hw/hns/hns_roce_hw_v2.c index 8ea5470303ee..38a0b6b3446e 100644 --- a/drivers/infiniband/hw/hns/hns_roce_hw_v2.c +++ b/drivers/infiniband/hw/hns/hns_roce_hw_v2.c @@ -153,7 +153,7 @@ static int hns_roce_v2_post_send(struct ib_qp *ibqp, struct ib_send_wr *wr, ibqp->qp_type != IB_QPT_GSI && ibqp->qp_type != IB_QPT_UD)) { dev_err(dev, "Not supported QP(0x%x)type!\n", ibqp->qp_type); - *bad_wr = NULL; + *bad_wr = wr; return -EOPNOTSUPP; } @@ -462,6 +462,7 @@ static int hns_roce_v2_post_send(struct ib_qp *ibqp, struct ib_send_wr *wr, } else { dev_err(dev, "Illegal qp_type(0x%x)\n", ibqp->qp_type); spin_unlock_irqrestore(&qp->sq.lock, flags); + *bad_wr = wr; return -EOPNOTSUPP; } } -- cgit v1.2.3 From ab17884903f97054c719a3c68017513b922efe43 Mon Sep 17 00:00:00 2001 From: oulijun Date: Thu, 26 Apr 2018 14:46:24 +0800 Subject: RDMA/hns: Fix a couple misspellings This patch fixes two spelling errors. Signed-off-by: Lijun Ou Signed-off-by: Doug Ledford --- drivers/infiniband/hw/hns/hns_roce_hw_v2.c | 2 +- drivers/infiniband/hw/hns/hns_roce_qp.c | 2 +- 2 files changed, 2 insertions(+), 2 deletions(-) (limited to 'drivers/infiniband/hw') diff --git a/drivers/infiniband/hw/hns/hns_roce_hw_v2.c b/drivers/infiniband/hw/hns/hns_roce_hw_v2.c index 38a0b6b3446e..25916e8522ed 100644 --- a/drivers/infiniband/hw/hns/hns_roce_hw_v2.c +++ b/drivers/infiniband/hw/hns/hns_roce_hw_v2.c @@ -4487,7 +4487,7 @@ static int hns_roce_v2_create_eq(struct hns_roce_dev *hr_dev, ret = hns_roce_cmd_mbox(hr_dev, mailbox->dma, 0, eq->eqn, 0, eq_cmd, HNS_ROCE_CMD_TIMEOUT_MSECS); if (ret) { - dev_err(dev, "[mailbox cmd] creat eqc failed.\n"); + dev_err(dev, "[mailbox cmd] create eqc failed.\n"); goto err_cmd_mbox; } diff --git a/drivers/infiniband/hw/hns/hns_roce_qp.c b/drivers/infiniband/hw/hns/hns_roce_qp.c index e289a924e789..d4aad34c21e2 100644 --- a/drivers/infiniband/hw/hns/hns_roce_qp.c +++ b/drivers/infiniband/hw/hns/hns_roce_qp.c @@ -620,7 +620,7 @@ static int hns_roce_create_qp_common(struct hns_roce_dev *hr_dev, to_hr_ucontext(ib_pd->uobject->context), ucmd.db_addr, &hr_qp->rdb); if (ret) { - dev_err(dev, "rp record doorbell map failed!\n"); + dev_err(dev, "rq record doorbell map failed!\n"); goto err_mtt; } } -- cgit v1.2.3 From 4f9ca2d8686ecfdd40ca4f0294a3d94f83f05cea Mon Sep 17 00:00:00 2001 From: Leon Romanovsky Date: Thu, 26 Apr 2018 15:37:48 +0300 Subject: RDMA/mlx4: Add missed RSS hash inner header flag Despite being advertised to user space application, the RSS inner header flag was filtered by checks at the beginning of QP creation routine. Cc: # 4.15 Fixes: 4d02ebd9bbbd ("IB/mlx4: Fix RSS hash fields restrictions") Fixes: 07d84f7b6adf ("IB/mlx4: Add support to RSS hash for inner headers") Signed-off-by: Leon Romanovsky Signed-off-by: Doug Ledford --- drivers/infiniband/hw/mlx4/qp.c | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-) (limited to 'drivers/infiniband/hw') diff --git a/drivers/infiniband/hw/mlx4/qp.c b/drivers/infiniband/hw/mlx4/qp.c index 50af8915e7ec..199648adac74 100644 --- a/drivers/infiniband/hw/mlx4/qp.c +++ b/drivers/infiniband/hw/mlx4/qp.c @@ -673,7 +673,8 @@ static int set_qp_rss(struct mlx4_ib_dev *dev, struct mlx4_ib_rss *rss_ctx, MLX4_IB_RX_HASH_SRC_PORT_TCP | MLX4_IB_RX_HASH_DST_PORT_TCP | MLX4_IB_RX_HASH_SRC_PORT_UDP | - MLX4_IB_RX_HASH_DST_PORT_UDP)) { + MLX4_IB_RX_HASH_DST_PORT_UDP | + MLX4_IB_RX_HASH_INNER)) { pr_debug("RX Hash fields_mask has unsupported mask (0x%llx)\n", ucmd->rx_hash_fields_mask); return (-EOPNOTSUPP); -- cgit v1.2.3 From 2df19e19ae90d94fd8724083f161f368a2797537 Mon Sep 17 00:00:00 2001 From: Bharat Potnuri Date: Fri, 27 Apr 2018 16:41:16 +0530 Subject: iw_cxgb4: Atomically flush per QP HW CQEs When a CQ is shared by multiple QPs, c4iw_flush_hw_cq() needs to acquire corresponding QP lock before moving the CQEs into its corresponding SW queue and accessing the SQ contents for completing a WR. Ignore CQEs if corresponding QP is already flushed. Cc: stable@vger.kernel.org Signed-off-by: Potnuri Bharat Teja Reviewed-by: Steve Wise Signed-off-by: Doug Ledford --- drivers/infiniband/hw/cxgb4/cq.c | 11 ++++++++++- drivers/infiniband/hw/cxgb4/iw_cxgb4.h | 2 +- drivers/infiniband/hw/cxgb4/qp.c | 4 ++-- 3 files changed, 13 insertions(+), 4 deletions(-) (limited to 'drivers/infiniband/hw') diff --git a/drivers/infiniband/hw/cxgb4/cq.c b/drivers/infiniband/hw/cxgb4/cq.c index 6f2b26126c64..2be2e1ac1b5f 100644 --- a/drivers/infiniband/hw/cxgb4/cq.c +++ b/drivers/infiniband/hw/cxgb4/cq.c @@ -315,7 +315,7 @@ static void advance_oldest_read(struct t4_wq *wq) * Deal with out-of-order and/or completions that complete * prior unsignalled WRs. */ -void c4iw_flush_hw_cq(struct c4iw_cq *chp) +void c4iw_flush_hw_cq(struct c4iw_cq *chp, struct c4iw_qp *flush_qhp) { struct t4_cqe *hw_cqe, *swcqe, read_cqe; struct c4iw_qp *qhp; @@ -339,6 +339,13 @@ void c4iw_flush_hw_cq(struct c4iw_cq *chp) if (qhp == NULL) goto next_cqe; + if (flush_qhp != qhp) { + spin_lock(&qhp->lock); + + if (qhp->wq.flushed == 1) + goto next_cqe; + } + if (CQE_OPCODE(hw_cqe) == FW_RI_TERMINATE) goto next_cqe; @@ -390,6 +397,8 @@ void c4iw_flush_hw_cq(struct c4iw_cq *chp) next_cqe: t4_hwcq_consume(&chp->cq); ret = t4_next_hw_cqe(&chp->cq, &hw_cqe); + if (qhp && flush_qhp != qhp) + spin_unlock(&qhp->lock); } } diff --git a/drivers/infiniband/hw/cxgb4/iw_cxgb4.h b/drivers/infiniband/hw/cxgb4/iw_cxgb4.h index a60def23e9ef..831027717121 100644 --- a/drivers/infiniband/hw/cxgb4/iw_cxgb4.h +++ b/drivers/infiniband/hw/cxgb4/iw_cxgb4.h @@ -1053,7 +1053,7 @@ u32 c4iw_pblpool_alloc(struct c4iw_rdev *rdev, int size); void c4iw_pblpool_free(struct c4iw_rdev *rdev, u32 addr, int size); u32 c4iw_ocqp_pool_alloc(struct c4iw_rdev *rdev, int size); void c4iw_ocqp_pool_free(struct c4iw_rdev *rdev, u32 addr, int size); -void c4iw_flush_hw_cq(struct c4iw_cq *chp); +void c4iw_flush_hw_cq(struct c4iw_cq *chp, struct c4iw_qp *flush_qhp); void c4iw_count_rcqes(struct t4_cq *cq, struct t4_wq *wq, int *count); int c4iw_ep_disconnect(struct c4iw_ep *ep, int abrupt, gfp_t gfp); int c4iw_flush_rq(struct t4_wq *wq, struct t4_cq *cq, int count); diff --git a/drivers/infiniband/hw/cxgb4/qp.c b/drivers/infiniband/hw/cxgb4/qp.c index de77b6027d69..ae167b686608 100644 --- a/drivers/infiniband/hw/cxgb4/qp.c +++ b/drivers/infiniband/hw/cxgb4/qp.c @@ -1343,12 +1343,12 @@ static void __flush_qp(struct c4iw_qp *qhp, struct c4iw_cq *rchp, qhp->wq.flushed = 1; t4_set_wq_in_error(&qhp->wq); - c4iw_flush_hw_cq(rchp); + c4iw_flush_hw_cq(rchp, qhp); c4iw_count_rcqes(&rchp->cq, &qhp->wq, &count); rq_flushed = c4iw_flush_rq(&qhp->wq, &rchp->cq, count); if (schp != rchp) - c4iw_flush_hw_cq(schp); + c4iw_flush_hw_cq(schp, qhp); sq_flushed = c4iw_flush_sq(qhp); spin_unlock(&qhp->lock); -- cgit v1.2.3 From f59fb9e05109b836230813e45f71c9ecc2d5dbe6 Mon Sep 17 00:00:00 2001 From: Mike Marciniszyn Date: Tue, 1 May 2018 05:35:36 -0700 Subject: IB/hfi1: Fix handling of FECN marked multicast packet The code for handling a marked UD packet unconditionally returns the dlid in the header of the FECN marked packet. This is not correct for multicast packets where the DLID is in the multicast range. The subsequent attempt to send the CNP with the multicast lid will cause the chip to halt the ack send context because the source lid doesn't match the chip programming. The send context will be halted and flush any other pending packets in the pio ring causing the CNP to not be sent. A part of investigating the fix, it was determined that the 16B work broke the FECN routine badly with inconsistent use of 16 bit and 32 bits types for lids and pkeys. Since the port's source lid was correctly 32 bits the type mixmatches need to be dealt with at the same time as fixing the CNP header issue. Fix these issues by: - Using the ports lid for as the SLID for responding to FECN marked UD packets - Insure pkey is always 16 bit in this and subordinate routines - Insure lids are 32 bits in this and subordinate routines Cc: # 4.14.x Fixes: 88733e3b8450 ("IB/hfi1: Add 16B UD support") Reviewed-by: Don Hiatt Reviewed-by: Michael J. Ruhl Signed-off-by: Mike Marciniszyn Signed-off-by: Dennis Dalessandro Signed-off-by: Doug Ledford --- drivers/infiniband/hw/hfi1/driver.c | 19 +++++++++++++++---- drivers/infiniband/hw/hfi1/hfi.h | 8 ++++---- drivers/infiniband/hw/hfi1/ud.c | 4 ++-- 3 files changed, 21 insertions(+), 10 deletions(-) (limited to 'drivers/infiniband/hw') diff --git a/drivers/infiniband/hw/hfi1/driver.c b/drivers/infiniband/hw/hfi1/driver.c index 46d1475b2154..bd837a048bf4 100644 --- a/drivers/infiniband/hw/hfi1/driver.c +++ b/drivers/infiniband/hw/hfi1/driver.c @@ -433,31 +433,43 @@ void hfi1_process_ecn_slowpath(struct rvt_qp *qp, struct hfi1_packet *pkt, bool do_cnp) { struct hfi1_ibport *ibp = to_iport(qp->ibqp.device, qp->port_num); + struct hfi1_pportdata *ppd = ppd_from_ibp(ibp); struct ib_other_headers *ohdr = pkt->ohdr; struct ib_grh *grh = pkt->grh; u32 rqpn = 0, bth1; - u16 pkey, rlid, dlid = ib_get_dlid(pkt->hdr); + u16 pkey; + u32 rlid, slid, dlid = 0; u8 hdr_type, sc, svc_type; bool is_mcast = false; + /* can be called from prescan */ if (pkt->etype == RHF_RCV_TYPE_BYPASS) { is_mcast = hfi1_is_16B_mcast(dlid); pkey = hfi1_16B_get_pkey(pkt->hdr); sc = hfi1_16B_get_sc(pkt->hdr); + dlid = hfi1_16B_get_dlid(pkt->hdr); + slid = hfi1_16B_get_slid(pkt->hdr); hdr_type = HFI1_PKT_TYPE_16B; } else { is_mcast = (dlid > be16_to_cpu(IB_MULTICAST_LID_BASE)) && (dlid != be16_to_cpu(IB_LID_PERMISSIVE)); pkey = ib_bth_get_pkey(ohdr); sc = hfi1_9B_get_sc5(pkt->hdr, pkt->rhf); + dlid = ib_get_dlid(pkt->hdr); + slid = ib_get_slid(pkt->hdr); hdr_type = HFI1_PKT_TYPE_9B; } switch (qp->ibqp.qp_type) { + case IB_QPT_UD: + dlid = ppd->lid; + rlid = slid; + rqpn = ib_get_sqpn(pkt->ohdr); + svc_type = IB_CC_SVCTYPE_UD; + break; case IB_QPT_SMI: case IB_QPT_GSI: - case IB_QPT_UD: - rlid = ib_get_slid(pkt->hdr); + rlid = slid; rqpn = ib_get_sqpn(pkt->ohdr); svc_type = IB_CC_SVCTYPE_UD; break; @@ -482,7 +494,6 @@ void hfi1_process_ecn_slowpath(struct rvt_qp *qp, struct hfi1_packet *pkt, dlid, rlid, sc, grh); if (!is_mcast && (bth1 & IB_BECN_SMASK)) { - struct hfi1_pportdata *ppd = ppd_from_ibp(ibp); u32 lqpn = bth1 & RVT_QPN_MASK; u8 sl = ibp->sc_to_sl[sc]; diff --git a/drivers/infiniband/hw/hfi1/hfi.h b/drivers/infiniband/hw/hfi1/hfi.h index 32c48265405e..cac2c62bc42d 100644 --- a/drivers/infiniband/hw/hfi1/hfi.h +++ b/drivers/infiniband/hw/hfi1/hfi.h @@ -1537,13 +1537,13 @@ void set_link_ipg(struct hfi1_pportdata *ppd); void process_becn(struct hfi1_pportdata *ppd, u8 sl, u32 rlid, u32 lqpn, u32 rqpn, u8 svc_type); void return_cnp(struct hfi1_ibport *ibp, struct rvt_qp *qp, u32 remote_qpn, - u32 pkey, u32 slid, u32 dlid, u8 sc5, + u16 pkey, u32 slid, u32 dlid, u8 sc5, const struct ib_grh *old_grh); void return_cnp_16B(struct hfi1_ibport *ibp, struct rvt_qp *qp, - u32 remote_qpn, u32 pkey, u32 slid, u32 dlid, + u32 remote_qpn, u16 pkey, u32 slid, u32 dlid, u8 sc5, const struct ib_grh *old_grh); typedef void (*hfi1_handle_cnp)(struct hfi1_ibport *ibp, struct rvt_qp *qp, - u32 remote_qpn, u32 pkey, u32 slid, u32 dlid, + u32 remote_qpn, u16 pkey, u32 slid, u32 dlid, u8 sc5, const struct ib_grh *old_grh); #define PKEY_CHECK_INVALID -1 @@ -2437,7 +2437,7 @@ static inline void hfi1_make_16b_hdr(struct hfi1_16b_header *hdr, ((slid >> OPA_16B_SLID_SHIFT) << OPA_16B_SLID_HIGH_SHIFT); lrh2 = (lrh2 & ~OPA_16B_DLID_MASK) | ((dlid >> OPA_16B_DLID_SHIFT) << OPA_16B_DLID_HIGH_SHIFT); - lrh2 = (lrh2 & ~OPA_16B_PKEY_MASK) | (pkey << OPA_16B_PKEY_SHIFT); + lrh2 = (lrh2 & ~OPA_16B_PKEY_MASK) | ((u32)pkey << OPA_16B_PKEY_SHIFT); lrh2 = (lrh2 & ~OPA_16B_L4_MASK) | l4; hdr->lrh[0] = lrh0; diff --git a/drivers/infiniband/hw/hfi1/ud.c b/drivers/infiniband/hw/hfi1/ud.c index bcf3b0bebac8..69c17a5ef038 100644 --- a/drivers/infiniband/hw/hfi1/ud.c +++ b/drivers/infiniband/hw/hfi1/ud.c @@ -628,7 +628,7 @@ int hfi1_lookup_pkey_idx(struct hfi1_ibport *ibp, u16 pkey) } void return_cnp_16B(struct hfi1_ibport *ibp, struct rvt_qp *qp, - u32 remote_qpn, u32 pkey, u32 slid, u32 dlid, + u32 remote_qpn, u16 pkey, u32 slid, u32 dlid, u8 sc5, const struct ib_grh *old_grh) { u64 pbc, pbc_flags = 0; @@ -687,7 +687,7 @@ void return_cnp_16B(struct hfi1_ibport *ibp, struct rvt_qp *qp, } void return_cnp(struct hfi1_ibport *ibp, struct rvt_qp *qp, u32 remote_qpn, - u32 pkey, u32 slid, u32 dlid, u8 sc5, + u16 pkey, u32 slid, u32 dlid, u8 sc5, const struct ib_grh *old_grh) { u64 pbc, pbc_flags = 0; -- cgit v1.2.3 From 5da9e742be44d9b7c68b1bf6e1aaf46a1aa7a52b Mon Sep 17 00:00:00 2001 From: "Michael J. Ruhl" Date: Tue, 1 May 2018 05:35:43 -0700 Subject: IB/hfi1 Use correct type for num_user_context The module parameter num_user_context is defined as 'int' and defaults to -1. The module_param_named() says that it is uint. Correct module_param_named() type information and update the modinfo text to reflect the default value. Reviewed-by: Dennis Dalessandro Signed-off-by: Michael J. Ruhl Signed-off-by: Dennis Dalessandro Signed-off-by: Doug Ledford --- drivers/infiniband/hw/hfi1/init.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) (limited to 'drivers/infiniband/hw') diff --git a/drivers/infiniband/hw/hfi1/init.c b/drivers/infiniband/hw/hfi1/init.c index 33eba2356742..c45cca556942 100644 --- a/drivers/infiniband/hw/hfi1/init.c +++ b/drivers/infiniband/hw/hfi1/init.c @@ -88,9 +88,9 @@ * pio buffers per ctxt, etc.) Zero means use one user context per CPU. */ int num_user_contexts = -1; -module_param_named(num_user_contexts, num_user_contexts, uint, S_IRUGO); +module_param_named(num_user_contexts, num_user_contexts, int, 0444); MODULE_PARM_DESC( - num_user_contexts, "Set max number of user contexts to use"); + num_user_contexts, "Set max number of user contexts to use (default: -1 will use the real (non-HT) CPU count)"); uint krcvqs[RXE_NUM_DATA_VL]; int krcvqsset; -- cgit v1.2.3 From 0a0bcb046b2f0c15b89f8c1b08ad3de601a83c66 Mon Sep 17 00:00:00 2001 From: Mike Marciniszyn Date: Tue, 1 May 2018 05:35:51 -0700 Subject: IB/hfi1: Fix loss of BECN with AHG AHG may be armed to use the stored header, which by design is limited to edits in the PSN/A 32 bit word (bth2). When the code is trying to send a BECN, the use of the stored header will lose the BECN bit. Fix by avoiding AHG when getting ready to send a BECN. This is accomplished by always claiming the packet is not a middle packet which is an AHG precursor. BECNs are not a normal case and this should not hurt AHG optimizations. Cc: # 4.14.x Reviewed-by: Michael J. Ruhl Signed-off-by: Mike Marciniszyn Signed-off-by: Dennis Dalessandro Signed-off-by: Doug Ledford --- drivers/infiniband/hw/hfi1/ruc.c | 50 ++++++++++++++++++++++++++++++++-------- 1 file changed, 40 insertions(+), 10 deletions(-) (limited to 'drivers/infiniband/hw') diff --git a/drivers/infiniband/hw/hfi1/ruc.c b/drivers/infiniband/hw/hfi1/ruc.c index 3daa94bdae3a..c0071ca4147a 100644 --- a/drivers/infiniband/hw/hfi1/ruc.c +++ b/drivers/infiniband/hw/hfi1/ruc.c @@ -733,6 +733,20 @@ static inline void hfi1_make_ruc_bth(struct rvt_qp *qp, ohdr->bth[2] = cpu_to_be32(bth2); } +/** + * hfi1_make_ruc_header_16B - build a 16B header + * @qp: the queue pair + * @ohdr: a pointer to the destination header memory + * @bth0: bth0 passed in from the RC/UC builder + * @bth2: bth2 passed in from the RC/UC builder + * @middle: non zero implies indicates ahg "could" be used + * @ps: the current packet state + * + * This routine may disarm ahg under these situations: + * - packet needs a GRH + * - BECN needed + * - migration state not IB_MIG_MIGRATED + */ static inline void hfi1_make_ruc_header_16B(struct rvt_qp *qp, struct ib_other_headers *ohdr, u32 bth0, u32 bth2, int middle, @@ -777,6 +791,12 @@ static inline void hfi1_make_ruc_header_16B(struct rvt_qp *qp, else middle = 0; + if (qp->s_flags & RVT_S_ECN) { + qp->s_flags &= ~RVT_S_ECN; + /* we recently received a FECN, so return a BECN */ + becn = true; + middle = 0; + } if (middle) build_ahg(qp, bth2); else @@ -784,11 +804,6 @@ static inline void hfi1_make_ruc_header_16B(struct rvt_qp *qp, bth0 |= pkey; bth0 |= extra_bytes << 20; - if (qp->s_flags & RVT_S_ECN) { - qp->s_flags &= ~RVT_S_ECN; - /* we recently received a FECN, so return a BECN */ - becn = true; - } hfi1_make_ruc_bth(qp, ohdr, bth0, bth1, bth2); if (!ppd->lid) @@ -806,6 +821,20 @@ static inline void hfi1_make_ruc_header_16B(struct rvt_qp *qp, pkey, becn, 0, l4, priv->s_sc); } +/** + * hfi1_make_ruc_header_9B - build a 9B header + * @qp: the queue pair + * @ohdr: a pointer to the destination header memory + * @bth0: bth0 passed in from the RC/UC builder + * @bth2: bth2 passed in from the RC/UC builder + * @middle: non zero implies indicates ahg "could" be used + * @ps: the current packet state + * + * This routine may disarm ahg under these situations: + * - packet needs a GRH + * - BECN needed + * - migration state not IB_MIG_MIGRATED + */ static inline void hfi1_make_ruc_header_9B(struct rvt_qp *qp, struct ib_other_headers *ohdr, u32 bth0, u32 bth2, int middle, @@ -839,6 +868,12 @@ static inline void hfi1_make_ruc_header_9B(struct rvt_qp *qp, else middle = 0; + if (qp->s_flags & RVT_S_ECN) { + qp->s_flags &= ~RVT_S_ECN; + /* we recently received a FECN, so return a BECN */ + bth1 |= (IB_BECN_MASK << IB_BECN_SHIFT); + middle = 0; + } if (middle) build_ahg(qp, bth2); else @@ -846,11 +881,6 @@ static inline void hfi1_make_ruc_header_9B(struct rvt_qp *qp, bth0 |= pkey; bth0 |= extra_bytes << 20; - if (qp->s_flags & RVT_S_ECN) { - qp->s_flags &= ~RVT_S_ECN; - /* we recently received a FECN, so return a BECN */ - bth1 |= (IB_BECN_MASK << IB_BECN_SHIFT); - } hfi1_make_ruc_bth(qp, ohdr, bth0, bth1, bth2); hfi1_make_ib_hdr(&ps->s_txreq->phdr.hdr.ibh, lrh0, -- cgit v1.2.3 From 45d924571a5e1329580811f2419da61b07ac3613 Mon Sep 17 00:00:00 2001 From: Sebastian Sanchez Date: Tue, 1 May 2018 05:35:58 -0700 Subject: IB/hfi1: Fix NULL pointer dereference when invalid num_vls is used When an invalid num_vls is used as a module parameter, the code execution follows an exception path where the macro dd_dev_err() expects dd->pcidev->dev not to be NULL in hfi1_init_dd(). This causes a NULL pointer dereference. Fix hfi1_init_dd() by initializing dd->pcidev and dd->pcidev->dev earlier in the code. If a dd exists, then dd->pcidev and dd->pcidev->dev always exists. BUG: unable to handle kernel NULL pointer dereference at 00000000000000f0 IP: __dev_printk+0x15/0x90 Workqueue: events work_for_cpu_fn RIP: 0010:__dev_printk+0x15/0x90 Call Trace: dev_err+0x6c/0x90 ? hfi1_init_pportdata+0x38d/0x3f0 [hfi1] hfi1_init_dd+0xdd/0x2530 [hfi1] ? pci_conf1_read+0xb2/0xf0 ? pci_read_config_word.part.9+0x64/0x80 ? pci_conf1_write+0xb0/0xf0 ? pcie_capability_clear_and_set_word+0x57/0x80 init_one+0x141/0x490 [hfi1] local_pci_probe+0x3f/0xa0 work_for_cpu_fn+0x10/0x20 process_one_work+0x152/0x350 worker_thread+0x1cf/0x3e0 kthread+0xf5/0x130 ? max_active_store+0x80/0x80 ? kthread_bind+0x10/0x10 ? do_syscall_64+0x6e/0x1a0 ? SyS_exit_group+0x10/0x10 ret_from_fork+0x35/0x40 Cc: # 4.9.x Reviewed-by: Mike Marciniszyn Reviewed-by: Michael J. Ruhl Signed-off-by: Sebastian Sanchez Signed-off-by: Dennis Dalessandro Signed-off-by: Doug Ledford --- drivers/infiniband/hw/hfi1/init.c | 2 ++ drivers/infiniband/hw/hfi1/pcie.c | 3 --- 2 files changed, 2 insertions(+), 3 deletions(-) (limited to 'drivers/infiniband/hw') diff --git a/drivers/infiniband/hw/hfi1/init.c b/drivers/infiniband/hw/hfi1/init.c index c45cca556942..b417e3b40e4a 100644 --- a/drivers/infiniband/hw/hfi1/init.c +++ b/drivers/infiniband/hw/hfi1/init.c @@ -1265,6 +1265,8 @@ struct hfi1_devdata *hfi1_alloc_devdata(struct pci_dev *pdev, size_t extra) return ERR_PTR(-ENOMEM); dd->num_pports = nports; dd->pport = (struct hfi1_pportdata *)(dd + 1); + dd->pcidev = pdev; + pci_set_drvdata(pdev, dd); INIT_LIST_HEAD(&dd->list); idr_preload(GFP_KERNEL); diff --git a/drivers/infiniband/hw/hfi1/pcie.c b/drivers/infiniband/hw/hfi1/pcie.c index 83d66e862207..c1c982908b4b 100644 --- a/drivers/infiniband/hw/hfi1/pcie.c +++ b/drivers/infiniband/hw/hfi1/pcie.c @@ -163,9 +163,6 @@ int hfi1_pcie_ddinit(struct hfi1_devdata *dd, struct pci_dev *pdev) resource_size_t addr; int ret = 0; - dd->pcidev = pdev; - pci_set_drvdata(pdev, dd); - addr = pci_resource_start(pdev, 0); len = pci_resource_len(pdev, 0); -- cgit v1.2.3 From e9777ad4399c26c70318c4945f94efac2ed95391 Mon Sep 17 00:00:00 2001 From: Sebastian Sanchez Date: Tue, 1 May 2018 05:36:06 -0700 Subject: IB/{hfi1, rdmavt}: Fix memory leak in hfi1_alloc_devdata() upon failure When allocating device data, if there's an allocation failure, the already allocated memory won't be freed such as per-cpu counters. Fix memory leaks in exception path by creating a common reentrant clean up function hfi1_clean_devdata() to be used at driver unload time and device data allocation failure. To accomplish this, free_platform_config() and clean_up_i2c() are changed to be reentrant to remove dependencies when they are called in different order. This helps avoid NULL pointer dereferences introduced by this patch if those two functions weren't reentrant. In addition, set dd->int_counter, dd->rcv_limit, dd->send_schedule and dd->tx_opstats to NULL after they're freed in hfi1_clean_devdata(), so that hfi1_clean_devdata() is fully reentrant. Reviewed-by: Mike Marciniszyn Reviewed-by: Michael J. Ruhl Signed-off-by: Sebastian Sanchez Signed-off-by: Dennis Dalessandro Signed-off-by: Doug Ledford --- drivers/infiniband/hw/hfi1/init.c | 37 +++++++++++++++++++++++++---------- drivers/infiniband/hw/hfi1/platform.c | 1 + drivers/infiniband/hw/hfi1/qsfp.c | 2 ++ 3 files changed, 30 insertions(+), 10 deletions(-) (limited to 'drivers/infiniband/hw') diff --git a/drivers/infiniband/hw/hfi1/init.c b/drivers/infiniband/hw/hfi1/init.c index b417e3b40e4a..6309edf811df 100644 --- a/drivers/infiniband/hw/hfi1/init.c +++ b/drivers/infiniband/hw/hfi1/init.c @@ -1209,30 +1209,49 @@ static void finalize_asic_data(struct hfi1_devdata *dd, kfree(ad); } -static void __hfi1_free_devdata(struct kobject *kobj) +/** + * hfi1_clean_devdata - cleans up per-unit data structure + * @dd: pointer to a valid devdata structure + * + * It cleans up all data structures set up by + * by hfi1_alloc_devdata(). + */ +static void hfi1_clean_devdata(struct hfi1_devdata *dd) { - struct hfi1_devdata *dd = - container_of(kobj, struct hfi1_devdata, kobj); struct hfi1_asic_data *ad; unsigned long flags; spin_lock_irqsave(&hfi1_devs_lock, flags); - idr_remove(&hfi1_unit_table, dd->unit); - list_del(&dd->list); + if (!list_empty(&dd->list)) { + idr_remove(&hfi1_unit_table, dd->unit); + list_del_init(&dd->list); + } ad = release_asic_data(dd); spin_unlock_irqrestore(&hfi1_devs_lock, flags); - if (ad) - finalize_asic_data(dd, ad); + + finalize_asic_data(dd, ad); free_platform_config(dd); rcu_barrier(); /* wait for rcu callbacks to complete */ free_percpu(dd->int_counter); free_percpu(dd->rcv_limit); free_percpu(dd->send_schedule); free_percpu(dd->tx_opstats); + dd->int_counter = NULL; + dd->rcv_limit = NULL; + dd->send_schedule = NULL; + dd->tx_opstats = NULL; sdma_clean(dd, dd->num_sdma); rvt_dealloc_device(&dd->verbs_dev.rdi); } +static void __hfi1_free_devdata(struct kobject *kobj) +{ + struct hfi1_devdata *dd = + container_of(kobj, struct hfi1_devdata, kobj); + + hfi1_clean_devdata(dd); +} + static struct kobj_type hfi1_devdata_type = { .release = __hfi1_free_devdata, }; @@ -1333,9 +1352,7 @@ struct hfi1_devdata *hfi1_alloc_devdata(struct pci_dev *pdev, size_t extra) return dd; bail: - if (!list_empty(&dd->list)) - list_del_init(&dd->list); - rvt_dealloc_device(&dd->verbs_dev.rdi); + hfi1_clean_devdata(dd); return ERR_PTR(ret); } diff --git a/drivers/infiniband/hw/hfi1/platform.c b/drivers/infiniband/hw/hfi1/platform.c index d486355880cb..cbf7faa5038c 100644 --- a/drivers/infiniband/hw/hfi1/platform.c +++ b/drivers/infiniband/hw/hfi1/platform.c @@ -199,6 +199,7 @@ void free_platform_config(struct hfi1_devdata *dd) { /* Release memory allocated for eprom or fallback file read. */ kfree(dd->platform_config.data); + dd->platform_config.data = NULL; } void get_port_type(struct hfi1_pportdata *ppd) diff --git a/drivers/infiniband/hw/hfi1/qsfp.c b/drivers/infiniband/hw/hfi1/qsfp.c index 1869f639c3ae..b5966991d647 100644 --- a/drivers/infiniband/hw/hfi1/qsfp.c +++ b/drivers/infiniband/hw/hfi1/qsfp.c @@ -204,6 +204,8 @@ static void clean_i2c_bus(struct hfi1_i2c_bus *bus) void clean_up_i2c(struct hfi1_devdata *dd, struct hfi1_asic_data *ad) { + if (!ad) + return; clean_i2c_bus(ad->i2c_bus0); ad->i2c_bus0 = NULL; clean_i2c_bus(ad->i2c_bus1); -- cgit v1.2.3 From 59482a14918b282ca2a98f38c69da5ebeb1107d2 Mon Sep 17 00:00:00 2001 From: Sebastian Sanchez Date: Tue, 1 May 2018 05:36:13 -0700 Subject: IB/hfi1: Fix memory leak in exception path in get_irq_affinity() When IRQ affinity is set and the interrupt type is unknown, a cpu mask allocated within the function is never freed. Fix this memory leak by allocating memory within the scope where it is used. Reviewed-by: Mike Marciniszyn Reviewed-by: Michael J. Ruhl Signed-off-by: Sebastian Sanchez Signed-off-by: Dennis Dalessandro Signed-off-by: Doug Ledford --- drivers/infiniband/hw/hfi1/affinity.c | 11 +++++------ 1 file changed, 5 insertions(+), 6 deletions(-) (limited to 'drivers/infiniband/hw') diff --git a/drivers/infiniband/hw/hfi1/affinity.c b/drivers/infiniband/hw/hfi1/affinity.c index a97055dd4fbd..b5fab55cc275 100644 --- a/drivers/infiniband/hw/hfi1/affinity.c +++ b/drivers/infiniband/hw/hfi1/affinity.c @@ -412,7 +412,6 @@ static void hfi1_cleanup_sdma_notifier(struct hfi1_msix_entry *msix) static int get_irq_affinity(struct hfi1_devdata *dd, struct hfi1_msix_entry *msix) { - int ret; cpumask_var_t diff; struct hfi1_affinity_node *entry; struct cpu_mask_set *set = NULL; @@ -424,10 +423,6 @@ static int get_irq_affinity(struct hfi1_devdata *dd, extra[0] = '\0'; cpumask_clear(&msix->mask); - ret = zalloc_cpumask_var(&diff, GFP_KERNEL); - if (!ret) - return -ENOMEM; - entry = node_affinity_lookup(dd->node); switch (msix->type) { @@ -458,6 +453,9 @@ static int get_irq_affinity(struct hfi1_devdata *dd, * finds its CPU here. */ if (cpu == -1 && set) { + if (!zalloc_cpumask_var(&diff, GFP_KERNEL)) + return -ENOMEM; + if (cpumask_equal(&set->mask, &set->used)) { /* * We've used up all the CPUs, bump up the generation @@ -469,6 +467,8 @@ static int get_irq_affinity(struct hfi1_devdata *dd, cpumask_andnot(diff, &set->mask, &set->used); cpu = cpumask_first(diff); cpumask_set_cpu(cpu, &set->used); + + free_cpumask_var(diff); } cpumask_set_cpu(cpu, &msix->mask); @@ -482,7 +482,6 @@ static int get_irq_affinity(struct hfi1_devdata *dd, hfi1_setup_sdma_notifier(msix); } - free_cpumask_var(diff); return 0; } -- cgit v1.2.3 From b03bcde962606d2ee59a4e9dd470db9ad53c5418 Mon Sep 17 00:00:00 2001 From: Jack Morgenstein Date: Wed, 2 May 2018 13:04:25 +0300 Subject: IB/mlx4: Fix integer overflow when calculating optimal MTT size When the kernel was compiled using the UBSAN option, we saw the following stack trace: [ 1184.827917] UBSAN: Undefined behaviour in drivers/infiniband/hw/mlx4/mr.c:349:27 [ 1184.828114] signed integer overflow: [ 1184.828247] -2147483648 - 1 cannot be represented in type 'int' The problem was caused by calling round_up in procedure mlx4_ib_umem_calc_optimal_mtt_size (on line 349, as noted in the stack trace) with the second parameter (1 << block_shift) (which is an int). The second parameter should have been (1ULL << block_shift) (which is an unsigned long long). (1 << block_shift) is treated by the compiler as an int (because 1 is an integer). Now, local variable block_shift is initialized to 31. If block_shift is 31, 1 << block_shift is 1 << 31 = 0x80000000=-214748368. This is the most negative int value. Inside the round_up macro, there is a cast applied to ((1 << 31) - 1). However, this cast is applied AFTER ((1 << 31) - 1) is calculated. Since (1 << 31) is treated as an int, we get the negative overflow identified by UBSAN in the process of calculating ((1 << 31) - 1). The fix is to change (1 << block_shift) to (1ULL << block_shift) on line 349. Fixes: 9901abf58368 ("IB/mlx4: Use optimal numbers of MTT entries") Signed-off-by: Jack Morgenstein Signed-off-by: Leon Romanovsky Signed-off-by: Doug Ledford --- drivers/infiniband/hw/mlx4/mr.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'drivers/infiniband/hw') diff --git a/drivers/infiniband/hw/mlx4/mr.c b/drivers/infiniband/hw/mlx4/mr.c index 17f4f151a97f..61d8b06375bb 100644 --- a/drivers/infiniband/hw/mlx4/mr.c +++ b/drivers/infiniband/hw/mlx4/mr.c @@ -346,7 +346,7 @@ int mlx4_ib_umem_calc_optimal_mtt_size(struct ib_umem *umem, u64 start_va, /* Add to the first block the misalignment that it suffers from. */ total_len += (first_block_start & ((1ULL << block_shift) - 1ULL)); last_block_end = current_block_start + current_block_len; - last_block_aligned_end = round_up(last_block_end, 1 << block_shift); + last_block_aligned_end = round_up(last_block_end, 1ULL << block_shift); total_len += (last_block_aligned_end - last_block_end); if (total_len & ((1ULL << block_shift) - 1ULL)) -- cgit v1.2.3