Quoting Christian König (2018-01-22 20:00:03)
> We need to set shared_count even if we already have a fence to wait for.
>
> v2: init i to -1 as well
>
> Signed-off-by: Christian König <christian.koenig@xxxxxxx>
> Cc: stable@xxxxxxxxxxxxxxx
> Tested-by: Lyude Paul <lyude@xxxxxxxxxx>
> Reviewed-by: Lyude Paul <lyude@xxxxxxxxxx>
> ---
> drivers/dma-buf/reservation.c | 8 +++++---
> 1 file changed, 5 insertions(+), 3 deletions(-)
>
> diff --git a/drivers/dma-buf/reservation.c b/drivers/dma-buf/reservation.c
> index 461afa9febd4..314eb1071cce 100644
> --- a/drivers/dma-buf/reservation.c
> +++ b/drivers/dma-buf/reservation.c
> @@ -484,13 +484,15 @@ long reservation_object_wait_timeout_rcu(struct reservation_object *obj,
> unsigned long timeout)
> {
> struct dma_fence *fence;
> - unsigned seq, shared_count, i = 0;
> + unsigned seq, shared_count;
> long ret = timeout ? timeout : 1;
> + int i;
>
> retry:
> shared_count = 0;
> seq = read_seqcount_begin(&obj->seq);
> rcu_read_lock();
> + i = -1;
Could be before the seqlock, but
Reviewed-by: Chris Wilson <chris@xxxxxxxxxxxxxxxxxx>
Are we at the point where just grabbing the snapshot of fences with your
new get_fences_rcu() returning a single array will be simpler? (It also
has the change in behaviour of not updating the snapshot across the long
lived wait.)
If you want to avoid the kmalloc, we could teach it populate the
caller's stack allocated array first.
-Chris
_______________________________________________
dri-devel mailing list
dri-devel@xxxxxxxxxxxxxxxxxxxxx
https://lists.freedesktop.org/mailman/listinfo/dri-devel
