From e15ce4b8d11227007577e6dc1364d288b8874fbe Mon Sep 17 00:00:00 2001 From: Jack Morgenstein Date: Tue, 12 Mar 2019 17:05:47 +0200 Subject: [PATCH 1/3] net/mlx4_core: Fix reset flow when in command polling mode As part of unloading a device, the driver switches from FW command event mode to FW command polling mode. Part of switching over to polling mode is freeing the command context array memory (unfortunately, currently, without NULLing the command context array pointer). The reset flow calls "complete" to complete all outstanding fw commands (if we are in event mode). The check for event vs. polling mode here is to test if the command context array pointer is NULL. If the reset flow is activated after the switch to polling mode, it will attempt (incorrectly) to complete all the commands in the context array -- because the pointer was not NULLed when the driver switched over to polling mode. As a result, we have a use-after-free situation, which results in a kernel crash. For example: BUG: unable to handle kernel NULL pointer dereference at (null) IP: [] __wake_up_common+0x2e/0x90 PGD 0 Oops: 0000 [#1] SMP Modules linked in: netconsole nfsv3 nfs_acl nfs lockd grace ... CPU: 2 PID: 940 Comm: kworker/2:3 Kdump: loaded Not tainted 3.10.0-862.el7.x86_64 #1 Hardware name: Microsoft Corporation Virtual Machine/Virtual Machine, BIOS 090006 04/28/2016 Workqueue: events hv_eject_device_work [pci_hyperv] task: ffff8d1734ca0fd0 ti: ffff8d17354bc000 task.ti: ffff8d17354bc000 RIP: 0010:[] [] __wake_up_common+0x2e/0x90 RSP: 0018:ffff8d17354bfa38 EFLAGS: 00010082 RAX: 0000000000000000 RBX: ffff8d17362d42c8 RCX: 0000000000000000 RDX: 0000000000000001 RSI: 0000000000000003 RDI: ffff8d17362d42c8 RBP: ffff8d17354bfa70 R08: 0000000000000000 R09: 0000000000000000 R10: 0000000000000298 R11: ffff8d173610e000 R12: ffff8d17362d42d0 R13: 0000000000000246 R14: 0000000000000000 R15: 0000000000000003 FS: 0000000000000000(0000) GS:ffff8d1802680000(0000) knlGS:0000000000000000 CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033 CR2: 0000000000000000 CR3: 00000000f16d8000 CR4: 00000000001406e0 Call Trace: [] complete+0x3c/0x50 [] mlx4_cmd_wake_completions+0x70/0x90 [mlx4_core] [] mlx4_enter_error_state+0xe1/0x380 [mlx4_core] [] mlx4_comm_cmd+0x29b/0x360 [mlx4_core] [] __mlx4_cmd+0x441/0x920 [mlx4_core] [] ? __slab_free+0x81/0x2f0 [] ? __radix_tree_lookup+0x84/0xf0 [] mlx4_free_mtt_range+0x5b/0xb0 [mlx4_core] [] mlx4_mtt_cleanup+0x17/0x20 [mlx4_core] [] mlx4_free_eq+0xa7/0x1c0 [mlx4_core] [] mlx4_cleanup_eq_table+0xde/0x130 [mlx4_core] [] mlx4_unload_one+0x118/0x300 [mlx4_core] [] mlx4_remove_one+0x91/0x1f0 [mlx4_core] The fix is to set the command context array pointer to NULL after freeing the array. Fixes: f5aef5aa3506 ("net/mlx4_core: Activate reset flow upon fatal command cases") Signed-off-by: Jack Morgenstein Signed-off-by: Tariq Toukan Signed-off-by: David S. Miller --- drivers/net/ethernet/mellanox/mlx4/cmd.c | 1 + 1 file changed, 1 insertion(+) diff --git a/drivers/net/ethernet/mellanox/mlx4/cmd.c b/drivers/net/ethernet/mellanox/mlx4/cmd.c index c19e74e6ac94d9..44081acd48fb10 100644 --- a/drivers/net/ethernet/mellanox/mlx4/cmd.c +++ b/drivers/net/ethernet/mellanox/mlx4/cmd.c @@ -2689,6 +2689,7 @@ void mlx4_cmd_use_polling(struct mlx4_dev *dev) down(&priv->cmd.event_sem); kfree(priv->cmd.context); + priv->cmd.context = NULL; up(&priv->cmd.poll_sem); up_write(&priv->cmd.switch_sem); From c07d27927f2f2e96fcd27bb9fb330c9ea65612d0 Mon Sep 17 00:00:00 2001 From: Jack Morgenstein Date: Tue, 12 Mar 2019 17:05:48 +0200 Subject: [PATCH 2/3] net/mlx4_core: Fix locking in SRIOV mode when switching between events and polling In procedures mlx4_cmd_use_events() and mlx4_cmd_use_polling(), we need to guarantee that there are no FW commands in progress on the comm channel (for VFs) or wrapped FW commands (on the PF) when SRIOV is active. We do this by also taking the slave_cmd_mutex when SRIOV is active. This is especially important when switching from event to polling, since we free the command-context array during the switch. If there are FW commands in progress (e.g., waiting for a completion event), the completion event handler will access freed memory. Since the decision to use comm_wait or comm_poll is taken before grabbing the event_sem/poll_sem in mlx4_comm_cmd_wait/poll, we must take the slave_cmd_mutex as well (to guarantee that the decision to use events or polling and the call to the appropriate cmd function are atomic). Fixes: a7e1f04905e5 ("net/mlx4_core: Fix deadlock when switching between polling and event fw commands") Signed-off-by: Jack Morgenstein Signed-off-by: Tariq Toukan Signed-off-by: David S. Miller --- drivers/net/ethernet/mellanox/mlx4/cmd.c | 8 ++++++++ 1 file changed, 8 insertions(+) diff --git a/drivers/net/ethernet/mellanox/mlx4/cmd.c b/drivers/net/ethernet/mellanox/mlx4/cmd.c index 44081acd48fb10..a5d5d6fc1da003 100644 --- a/drivers/net/ethernet/mellanox/mlx4/cmd.c +++ b/drivers/net/ethernet/mellanox/mlx4/cmd.c @@ -2645,6 +2645,8 @@ int mlx4_cmd_use_events(struct mlx4_dev *dev) if (!priv->cmd.context) return -ENOMEM; + if (mlx4_is_mfunc(dev)) + mutex_lock(&priv->cmd.slave_cmd_mutex); down_write(&priv->cmd.switch_sem); for (i = 0; i < priv->cmd.max_cmds; ++i) { priv->cmd.context[i].token = i; @@ -2670,6 +2672,8 @@ int mlx4_cmd_use_events(struct mlx4_dev *dev) down(&priv->cmd.poll_sem); priv->cmd.use_events = 1; up_write(&priv->cmd.switch_sem); + if (mlx4_is_mfunc(dev)) + mutex_unlock(&priv->cmd.slave_cmd_mutex); return err; } @@ -2682,6 +2686,8 @@ void mlx4_cmd_use_polling(struct mlx4_dev *dev) struct mlx4_priv *priv = mlx4_priv(dev); int i; + if (mlx4_is_mfunc(dev)) + mutex_lock(&priv->cmd.slave_cmd_mutex); down_write(&priv->cmd.switch_sem); priv->cmd.use_events = 0; @@ -2693,6 +2699,8 @@ void mlx4_cmd_use_polling(struct mlx4_dev *dev) up(&priv->cmd.poll_sem); up_write(&priv->cmd.switch_sem); + if (mlx4_is_mfunc(dev)) + mutex_unlock(&priv->cmd.slave_cmd_mutex); } struct mlx4_cmd_mailbox *mlx4_alloc_cmd_mailbox(struct mlx4_dev *dev) From 8511a653e9250ef36b95803c375a7be0e2edb628 Mon Sep 17 00:00:00 2001 From: Jack Morgenstein Date: Tue, 12 Mar 2019 17:05:49 +0200 Subject: [PATCH 3/3] net/mlx4_core: Fix qp mtt size calculation Calculation of qp mtt size (in function mlx4_RST2INIT_wrapper) ultimately depends on function roundup_pow_of_two. If the amount of memory required by the QP is less than one page, roundup_pow_of_two is called with argument zero. In this case, the roundup_pow_of_two result is undefined. Calling roundup_pow_of_two with a zero argument resulted in the following stack trace: UBSAN: Undefined behaviour in ./include/linux/log2.h:61:13 shift exponent 64 is too large for 64-bit type 'long unsigned int' CPU: 4 PID: 26939 Comm: rping Tainted: G OE 4.19.0-rc1 Hardware name: Supermicro X9DR3-F/X9DR3-F, BIOS 3.2a 07/09/2015 Call Trace: dump_stack+0x9a/0xeb ubsan_epilogue+0x9/0x7c __ubsan_handle_shift_out_of_bounds+0x254/0x29d ? __ubsan_handle_load_invalid_value+0x180/0x180 ? debug_show_all_locks+0x310/0x310 ? sched_clock+0x5/0x10 ? sched_clock+0x5/0x10 ? sched_clock_cpu+0x18/0x260 ? find_held_lock+0x35/0x1e0 ? mlx4_RST2INIT_QP_wrapper+0xfb1/0x1440 [mlx4_core] mlx4_RST2INIT_QP_wrapper+0xfb1/0x1440 [mlx4_core] Fix this by explicitly testing for zero, and returning one if the argument is zero (assuming that the next higher power of 2 in this case should be one). Fixes: c82e9aa0a8bc ("mlx4_core: resource tracking for HCA resources used by guests") Signed-off-by: Jack Morgenstein Signed-off-by: Tariq Toukan Signed-off-by: David S. Miller --- drivers/net/ethernet/mellanox/mlx4/resource_tracker.c | 6 +++--- 1 file changed, 3 insertions(+), 3 deletions(-) diff --git a/drivers/net/ethernet/mellanox/mlx4/resource_tracker.c b/drivers/net/ethernet/mellanox/mlx4/resource_tracker.c index eb13d361816248..4356f3a5800276 100644 --- a/drivers/net/ethernet/mellanox/mlx4/resource_tracker.c +++ b/drivers/net/ethernet/mellanox/mlx4/resource_tracker.c @@ -2719,13 +2719,13 @@ static int qp_get_mtt_size(struct mlx4_qp_context *qpc) int total_pages; int total_mem; int page_offset = (be32_to_cpu(qpc->params2) >> 6) & 0x3f; + int tot; sq_size = 1 << (log_sq_size + log_sq_sride + 4); rq_size = (srq|rss|xrc) ? 0 : (1 << (log_rq_size + log_rq_stride + 4)); total_mem = sq_size + rq_size; - total_pages = - roundup_pow_of_two((total_mem + (page_offset << 6)) >> - page_shift); + tot = (total_mem + (page_offset << 6)) >> page_shift; + total_pages = !tot ? 1 : roundup_pow_of_two(tot); return total_pages; }