[PATCH] drm/etnaviv: take GPU lock later in the submit process

[Christian Gmeiner]

2016-08-08 13:00 GMT+02:00 Lucas Stach <l.stach at pengutronix.de>:
> Both the fence and event alloc are safe to be done without holding the GPU
> lock, as they either don't need any locking (fences) or are protected by
> their own lock (events).
>
> This solves a bad locking interaction between the submit path and the
> recover worker. If userspace manages to exhaust all available events while
> the GPU is hung, the submit will wait for events to become available
> holding the GPU lock. The recover worker waits for this lock to become
> available before trying to recover the GPU which frees up the allocated
> events. Essentially both paths are deadlocked until the submit path
> times out waiting for available events, failing the submit that could
> otherwise be handled just fine if the recover worker had the chance to
> bring the GPU back in a working state.
>
> Signed-off-by: Lucas Stach <l.stach at pengutronix.de>
Reviewed-by: Christian Gmeiner <christian.gmeiner at gmail.com>
> ---
>  drivers/gpu/drm/etnaviv/etnaviv_gpu.c | 10 +++++-----
>  1 file changed, 5 insertions(+), 5 deletions(-)
>
> diff --git a/drivers/gpu/drm/etnaviv/etnaviv_gpu.c b/drivers/gpu/drm/etnaviv/etnaviv_gpu.c
> index 87ef34150d46..b382cf505262 100644
> --- a/drivers/gpu/drm/etnaviv/etnaviv_gpu.c
> +++ b/drivers/gpu/drm/etnaviv/etnaviv_gpu.c
> @@ -1333,8 +1333,6 @@ int etnaviv_gpu_submit(struct etnaviv_gpu *gpu,
>         if (ret < 0)
>                 return ret;
>
> -       mutex_lock(&gpu->lock);
> -
>         /*
>          * TODO
>          *
> @@ -1348,16 +1346,18 @@ int etnaviv_gpu_submit(struct etnaviv_gpu *gpu,
>         if (unlikely(event == ~0U)) {
>                 DRM_ERROR("no free event\n");
>                 ret = -EBUSY;
> -               goto out_unlock;
> +               goto out_pm_put;
>         }
>
>         fence = etnaviv_gpu_fence_alloc(gpu);
>         if (!fence) {
>                 event_free(gpu, event);
>                 ret = -ENOMEM;
> -               goto out_unlock;
> +               goto out_pm_put;
>         }
>
> +       mutex_lock(&gpu->lock);
> +
>         gpu->event[event].fence = fence;
>         submit->fence = fence->seqno;
>         gpu->active_fence = submit->fence;
> @@ -1395,9 +1395,9 @@ int etnaviv_gpu_submit(struct etnaviv_gpu *gpu,
>         hangcheck_timer_reset(gpu);
>         ret = 0;
>
> -out_unlock:
>         mutex_unlock(&gpu->lock);
>
> +out_pm_put:
>         etnaviv_gpu_pm_put(gpu);
>
>         return ret;
> --
> 2.8.1
>