[PATCH] drm/etnaviv: fix another potential dma_resv DAG violation

[Lucas Stach]

Am Mittwoch, dem 22.09.2021 um 13:41 +0200 schrieb Christian König:
> Setting the exclusive fence without waiting for the shared
> fences violates the DAG and is illegal.
> 
> We really need to get away from this ASAP and make the
> driver interface more bullet prove.

Uh, what happened to Daniels patch "drm/etnaviv: Don't break exclusive
fence ordering", which fixed this without totally doing away with the
no-implicit handling? I have to admit that I haven't followed up with
that patch series due to ETOOMUCHOTHERSTUFF, but from a quick glance
this patch looked better than what you propose here.

Regards,
Lucas

> 
> Signed-off-by: Christian König <christian.koenig at amd.com>
> Cc: <stable at vger.kernel.org>
> ---
>  drivers/gpu/drm/etnaviv/etnaviv_gem_submit.c | 6 +-----
>  1 file changed, 1 insertion(+), 5 deletions(-)
> 
> diff --git a/drivers/gpu/drm/etnaviv/etnaviv_gem_submit.c b/drivers/gpu/drm/etnaviv/etnaviv_gem_submit.c
> index 7e17bc2b5df1..4db40e88f67f 100644
> --- a/drivers/gpu/drm/etnaviv/etnaviv_gem_submit.c
> +++ b/drivers/gpu/drm/etnaviv/etnaviv_gem_submit.c
> @@ -185,19 +185,15 @@ static int submit_fence_sync(struct etnaviv_gem_submit *submit)
>  				return ret;
>  		}
>  
> -		if (submit->flags & ETNA_SUBMIT_NO_IMPLICIT)
> -			continue;
> -
>  		if (bo->flags & ETNA_SUBMIT_BO_WRITE) {
>  			ret = dma_resv_get_fences(robj, &bo->excl,
>  						  &bo->nr_shared,
>  						  &bo->shared);
>  			if (ret)
>  				return ret;
> -		} else {
> +		} else if (!(submit->flags & ETNA_SUBMIT_NO_IMPLICIT)) {
>  			bo->excl = dma_fence_get(dma_resv_excl_fence(robj));
>  		}
> -
>  	}
>  
>  	return ret;