[PATCH] dma-buf/sync_file: Always increment refcount when merging fences.

[Gustavo Padovan]

Hi Rafael,

2016-09-13 Rafael Antognolli <rafael.antognolli at intel.com>:

> The refcount of a fence should be increased whenever it is added to a merged
> fence, since it will later be decreased when the merged fence is destroyed.
> Failing to do so will cause the original fence to be freed if the merged fence
> gets freed, but other places still referencing won't know about it.
> 
> This patch fixes a kernel panic that can be triggered by creating a fence that
> is expired (or increasing the timeline until it expires), then creating a
> merged fence out of it, and deleting the merged fence. This will make the
> original expired fence's refcount go to zero.
> 
> Signed-off-by: Rafael Antognolli <rafael.antognolli at intel.com>
> ---
> 
> Sample code to trigger the mentioned kernel panic (might need to be executed a
> couple times before it actually breaks everything):
> 
> static void test_sync_expired_merge(void)
> {
>        int iterations = 1 << 20;
>        int timeline;
>        int i;
>        int fence_expired, fence_merged;
> 
>        timeline = sw_sync_timeline_create();
> 
>        sw_sync_timeline_inc(timeline, 100);
>        fence_expired = sw_sync_fence_create(timeline, 1);
>        fence_merged = sw_sync_merge(fence_expired, fence_expired);
>        sw_sync_fence_destroy(fence_merged);
> 
>        for (i = 0; i < iterations; i++) {
>                int fence = sw_sync_merge(fence_expired, fence_expired);
> 
>                igt_assert_f(sw_sync_wait(fence, -1) > 0,
>                                     "Failure waiting on fence\n");
>                sw_sync_fence_destroy(fence);
>        }
> 
>        sw_sync_fence_destroy(fence_expired);
> }
> 
>  drivers/dma-buf/sync_file.c | 7 ++-----
>  1 file changed, 2 insertions(+), 5 deletions(-)


Thanks for spotting this.

Reviewed-by: Gustavo Padovan <gustavo.padovan at collabora.co.uk>

Gustavo