OSDN Git Service

dma-buf/sw_sync: Synchronize signal vs syncpt free
authorChris Wilson <chris@chris-wilson.co.uk>
Mon, 12 Aug 2019 15:42:47 +0000 (16:42 +0100)
committerChris Wilson <chris@chris-wilson.co.uk>
Tue, 13 Aug 2019 06:57:51 +0000 (07:57 +0100)
commitd3c6dd1fb30d3853c2012549affe75c930f4a2f9
tree53867d8f173a886038684e45c27de5e155ba7c49
parent1c2b93902faf8654d13eac8c86876365d6f0da6f
dma-buf/sw_sync: Synchronize signal vs syncpt free

During release of the syncpt, we remove it from the list of syncpt and
the tree, but only if it is not already been removed. However, during
signaling, we first remove the syncpt from the list. So, if we
concurrently free and signal the syncpt, the free may decide that it is
not part of the tree and immediately free itself -- meanwhile the
signaler goes on to use the now freed datastructure.

In particular, we get struck by commit 0e2f733addbf ("dma-buf: make
dma_fence structure a bit smaller v2") as the cb_list is immediately
clobbered by the kfree_rcu.

v2: Avoid calling into timeline_fence_release() from under the spinlock

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=111381
Fixes: d3862e44daa7 ("dma-buf/sw-sync: Fix locking around sync_timeline lists")
References: 0e2f733addbf ("dma-buf: make dma_fence structure a bit smaller v2")
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Sumit Semwal <sumit.semwal@linaro.org>
Cc: Sean Paul <seanpaul@chromium.org>
Cc: Gustavo Padovan <gustavo@padovan.org>
Cc: Christian König <christian.koenig@amd.com>
Cc: <stable@vger.kernel.org> # v4.14+
Acked-by: Christian König <christian.koenig@amd.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190812154247.20508-1-chris@chris-wilson.co.uk
drivers/dma-buf/sw_sync.c