In the Linux kernel, the following vulnerability has been resolved:
IB/core: Fix a nested dead lock as part of ODP flow
Fix a nested dead lock as part of ODP flow by using mmput_async().
From the below call trace [1] can see that calling mmput() once we have the umemodp->umemmutex locked as required by ibumemodpmapdmaandlock() might trigger in the same task the exitmmap()->mmunotifierrelease()->mlx5ibinvalidaterange() which may dead lock when trying to lock the same mutex.
Moving to use mmputasync() will solve the problem as the above exitmmap() flow will be called in other task and will be executed once the lock will be available.
[1] [64843.077665] task:kworker/u133:2 state:D stack: 0 pid:80906 ppid: 2 flags:0x00004000 [64843.077672] Workqueue: mlx5ibpagefault mlx5ibeqepfaction [mlx5ib] [64843.077719] Call Trace: [64843.077722] <TASK> [64843.077724] _schedule+0x23d/0x590 [64843.077729] schedule+0x4e/0xb0 [64843.077735] schedulepreemptdisabled+0xe/0x10 [64843.077740] _mutexlock.constprop.0+0x263/0x490 [64843.077747] _mutexlockslowpath+0x13/0x20 [64843.077752] mutexlock+0x34/0x40 [64843.077758] mlx5ibinvalidaterange+0x48/0x270 [mlx5ib] [64843.077808] _mmunotifierrelease+0x1a4/0x200 [64843.077816] exitmmap+0x1bc/0x200 [64843.077822] ? walkpagerange+0x9c/0x120 [64843.077828] ? _condresched+0x1a/0x50 [64843.077833] ? mutexlock+0x13/0x40 [64843.077839] ? uprobeclearstate+0xac/0x120 [64843.077860] mmput+0x5f/0x140 [64843.077867] ibumemodpmapdmaandlock+0x21b/0x580 [ibcore] [64843.077931] pagefaultrealmr+0x9a/0x140 [mlx5ib] [64843.077962] pagefaultmr+0xb4/0x550 [mlx5ib] [64843.077992] pagefaultsingledatasegment.constprop.0+0x2ac/0x560 [mlx5ib] [64843.078022] mlx5ibeqepfaction+0x528/0x780 [mlx5ib] [64843.078051] processonework+0x22b/0x3d0 [64843.078059] workerthread+0x53/0x410 [64843.078065] ? processonework+0x3d0/0x3d0 [64843.078073] kthread+0x12a/0x150 [64843.078079] ? setkthreadstruct+0x50/0x50 [64843.078085] retfromfork+0x22/0x30 [64843.078093] </TASK>