[Mesa-dev] [PATCH] radeonsi: try flushing unflushed fences in si_fence_finish even when timeout == 0

[Marek Olšák]

Reviewed-by: Marek Olšák <marek.olsak at amd.com>

Marek

On Wed, Nov 22, 2017 at 5:52 PM, Nicolai Hähnle <nhaehnle at gmail.com> wrote:
> From: Nicolai Hähnle <nicolai.haehnle at amd.com>
>
> Under certain conditions, waiting on a GL sync objects should act like
> a flush, regardless of the timeout.
>
> Portal 2, CS:GO, and presumably other Source engine games rely on this
> behavior and hang during loading without this fix.
>
> Fixes: bc65dcab3bc4 ("radeonsi: avoid syncing the driver thread in si_fence_finish")
> ---
>  src/gallium/drivers/radeonsi/si_fence.c | 6 +++---
>  1 file changed, 3 insertions(+), 3 deletions(-)
>
> diff --git a/src/gallium/drivers/radeonsi/si_fence.c b/src/gallium/drivers/radeonsi/si_fence.c
> index 9d6bcfe1027..b835ed649ee 100644
> --- a/src/gallium/drivers/radeonsi/si_fence.c
> +++ b/src/gallium/drivers/radeonsi/si_fence.c
> @@ -184,36 +184,36 @@ static void si_fine_fence_set(struct si_context *ctx,
>  static boolean si_fence_finish(struct pipe_screen *screen,
>                                struct pipe_context *ctx,
>                                struct pipe_fence_handle *fence,
>                                uint64_t timeout)
>  {
>         struct radeon_winsys *rws = ((struct r600_common_screen*)screen)->ws;
>         struct si_multi_fence *rfence = (struct si_multi_fence *)fence;
>         int64_t abs_timeout = os_time_get_absolute_timeout(timeout);
>
>         if (!util_queue_fence_is_signalled(&rfence->ready)) {
> -               if (!timeout)
> -                       return false;
> -
>                 if (rfence->tc_token) {
>                         /* Ensure that si_flush_from_st will be called for
>                          * this fence, but only if we're in the API thread
>                          * where the context is current.
>                          *
>                          * Note that the batch containing the flush may already
>                          * be in flight in the driver thread, so the fence
>                          * may not be ready yet when this call returns.
>                          */
>                         threaded_context_flush(ctx, rfence->tc_token,
>                                                timeout == 0);
>                 }
>
> +               if (!timeout)
> +                       return false;
> +
>                 if (timeout == PIPE_TIMEOUT_INFINITE) {
>                         util_queue_fence_wait(&rfence->ready);
>                 } else {
>                         if (!util_queue_fence_wait_timeout(&rfence->ready, abs_timeout))
>                                 return false;
>                 }
>
>                 if (timeout && timeout != PIPE_TIMEOUT_INFINITE) {
>                         int64_t time = os_time_get_nano();
>                         timeout = abs_timeout > time ? abs_timeout - time : 0;
> --
> 2.11.0
>
> _______________________________________________
> mesa-dev mailing list
> mesa-dev at lists.freedesktop.org
> https://lists.freedesktop.org/mailman/listinfo/mesa-dev