diff options
| author | Chris Wilson <chris@chris-wilson.co.uk> | 2019-08-12 16:42:47 +0100 |
|---|---|---|
| committer | Chris Wilson <chris@chris-wilson.co.uk> | 2019-08-13 07:57:51 +0100 |
| commit | d3c6dd1fb30d3853c2012549affe75c930f4a2f9 (patch) | |
| tree | 53867d8f173a886038684e45c27de5e155ba7c49 /drivers/dma-buf | |
| parent | 1c2b93902faf8654d13eac8c86876365d6f0da6f (diff) | |
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
Diffstat (limited to 'drivers/dma-buf')
| -rw-r--r-- | drivers/dma-buf/sw_sync.c | 16 |
1 files changed, 7 insertions, 9 deletions
diff --git a/drivers/dma-buf/sw_sync.c b/drivers/dma-buf/sw_sync.c index 051f6c2873c7..6713cfb1995c 100644 --- a/drivers/dma-buf/sw_sync.c +++ b/drivers/dma-buf/sw_sync.c @@ -132,17 +132,14 @@ static void timeline_fence_release(struct dma_fence *fence) { struct sync_pt *pt = dma_fence_to_sync_pt(fence); struct sync_timeline *parent = dma_fence_parent(fence); + unsigned long flags; + spin_lock_irqsave(fence->lock, flags); if (!list_empty(&pt->link)) { - unsigned long flags; - - spin_lock_irqsave(fence->lock, flags); - if (!list_empty(&pt->link)) { - list_del(&pt->link); - rb_erase(&pt->node, &parent->pt_tree); - } - spin_unlock_irqrestore(fence->lock, flags); + list_del(&pt->link); + rb_erase(&pt->node, &parent->pt_tree); } + spin_unlock_irqrestore(fence->lock, flags); sync_timeline_put(parent); dma_fence_free(fence); @@ -265,7 +262,8 @@ static struct sync_pt *sync_pt_create(struct sync_timeline *obj, p = &parent->rb_left; } else { if (dma_fence_get_rcu(&other->base)) { - dma_fence_put(&pt->base); + sync_timeline_put(obj); + kfree(pt); pt = other; goto unlock; } |