On Sun, Jul 12, 2020 at 01:26:41PM +0300, Leon Romanovsky wrote:
> From: Maor Gottlieb <maorg@xxxxxxxxxxxx>
>
> SRQ table is accessed both from interrupt and process context,
> therefore we must use xa_lock_irq.
>
> [ 9878.321379] --------------------------------
> [ 9878.322349] inconsistent {IN-HARDIRQ-W} -> {HARDIRQ-ON-W} usage.
> [ 9878.323667] kworker/u17:9/8573 [HC0[0]:SC0[0]:HE1:SE1] takes:
> [ 9878.324894] ffff8883e3503d30 (&xa->xa_lock#13){?...}-{2:2}, at:
> mlx5_cmd_get_srq+0x18/0x70 [mlx5_ib]
> [ 9878.326816] {IN-HARDIRQ-W} state was registered at:
> [ 9878.327905] lock_acquire+0xb9/0x3a0
> [ 9878.328720] _raw_spin_lock+0x25/0x30
> [ 9878.329475] srq_event_notifier+0x2b/0xc0 [mlx5_ib]
> [ 9878.330433] notifier_call_chain+0x45/0x70
> [ 9878.331393] __atomic_notifier_call_chain+0x69/0x100
> [ 9878.332530] forward_event+0x36/0xc0 [mlx5_core]
> [ 9878.333558] notifier_call_chain+0x45/0x70
> [ 9878.334418] __atomic_notifier_call_chain+0x69/0x100
> [ 9878.335498] mlx5_eq_async_int+0xc5/0x160 [mlx5_core]
> [ 9878.336543] notifier_call_chain+0x45/0x70
> [ 9878.337354] __atomic_notifier_call_chain+0x69/0x100
> [ 9878.338337] mlx5_irq_int_handler+0x19/0x30 [mlx5_core]
> [ 9878.339369] __handle_irq_event_percpu+0x43/0x2a0
> [ 9878.340382] handle_irq_event_percpu+0x30/0x70
> [ 9878.341252] handle_irq_event+0x34/0x60
> [ 9878.342020] handle_edge_irq+0x7c/0x1b0
> [ 9878.342788] do_IRQ+0x60/0x110
> [ 9878.343482] ret_from_intr+0x0/0x2a
> [ 9878.344251] default_idle+0x34/0x160
> [ 9878.344996] do_idle+0x1ec/0x220
> [ 9878.345682] cpu_startup_entry+0x19/0x20
> [ 9878.346511] start_secondary+0x153/0x1a0
> [ 9878.347328] secondary_startup_64+0xa4/0xb0
> [ 9878.348226] irq event stamp: 20907
> [ 9878.348953] hardirqs last enabled at (20907): [<ffffffff819f0eb4>]
> _raw_spin_unlock_irq+0x24/0x30
> [ 9878.350599] hardirqs last disabled at (20906): [<ffffffff819f0cbf>]
> _raw_spin_lock_irq+0xf/0x40
> [ 9878.352300] softirqs last enabled at (20746): [<ffffffff81c002c9>]
> __do_softirq+0x2c9/0x436
> [ 9878.353859] softirqs last disabled at (20681): [<ffffffff81139543>]
> irq_exit+0xb3/0xc0
> [ 9878.355365]
> [ 9878.355365] other info that might help us debug this:
> [ 9878.356703] Possible unsafe locking scenario:
> [ 9878.356703]
> [ 9878.357941] CPU0
> [ 9878.358522] ----
> [ 9878.359109] lock(&xa->xa_lock#13);
> [ 9878.359875] <Interrupt>
> [ 9878.360504] lock(&xa->xa_lock#13);
> [ 9878.361315]
> [ 9878.361315] *** DEADLOCK ***
> [ 9878.361315]
> [ 9878.362632] 2 locks held by kworker/u17:9/8573:
> [ 9878.374883] #0: ffff888295218d38
> ((wq_completion)mlx5_ib_page_fault){+.+.}-{0:0}, at:
> process_one_work+0x1f1/0x5f0
> [ 9878.376728] #1: ffff888401647e78
> ((work_completion)(&pfault->work)){+.+.}-{0:0}, at:
> process_one_work+0x1f1/0x5f0
> [ 9878.378550]
> [ 9878.378550] stack backtrace:
> [ 9878.379489] CPU: 0 PID: 8573 Comm: kworker/u17:9 Tainted: G
> O 5.7.0_for_upstream_min_debug_2020_06_14_11_31_46_41 #1
> [ 9878.381730] Hardware name: QEMU Standard PC (Q35 + ICH9, 2009), BIOS
> rel-1.12.1-0-ga5cab58e9a3f-prebuilt.qemu.org 04/01/2014
> [ 9878.383940] Workqueue: mlx5_ib_page_fault mlx5_ib_eqe_pf_action
> [mlx5_ib]
> [ 9878.385239] Call Trace:
> [ 9878.385822] dump_stack+0x71/0x9b
> [ 9878.386519] mark_lock+0x4f2/0x590
> [ 9878.387263] ? print_shortest_lock_dependencies+0x200/0x200
> [ 9878.388362] __lock_acquire+0xa00/0x1eb0
> [ 9878.389133] lock_acquire+0xb9/0x3a0
> [ 9878.389854] ? mlx5_cmd_get_srq+0x18/0x70 [mlx5_ib]
> [ 9878.390796] _raw_spin_lock+0x25/0x30
> [ 9878.391533] ? mlx5_cmd_get_srq+0x18/0x70 [mlx5_ib]
> [ 9878.392455] mlx5_cmd_get_srq+0x18/0x70 [mlx5_ib]
> [ 9878.393351] mlx5_ib_eqe_pf_action+0x257/0xa30 [mlx5_ib]
> [ 9878.394337] ? process_one_work+0x209/0x5f0
> [ 9878.395150] process_one_work+0x27b/0x5f0
> [ 9878.395939] ? __schedule+0x280/0x7e0
> [ 9878.396683] worker_thread+0x2d/0x3c0
> [ 9878.397424] ? process_one_work+0x5f0/0x5f0
> [ 9878.398249] kthread+0x111/0x130
> [ 9878.398926] ? kthread_park+0x90/0x90
> [ 9878.399709] ret_from_fork+0x24/0x30
>
> Fixes: b02a29eb8841 ("mlx5: Convert mlx5_srq_table to XArray")
> Signed-off-by: Maor Gottlieb <maorg@xxxxxxxxxxxx>
> Signed-off-by: Leon Romanovsky <leonro@xxxxxxxxxxxx>
Applied to for-rc, I updated the fixes line, this bug has been present
since day 1 apparently.
Jason
