| Message ID | 20220406075132.3263-9-christian.koenig@amd.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | [01/16] dma-buf/drivers: make reserving a shared slot mandatory v4 | expand |
On Wed, Apr 06, 2022 at 09:51:24AM +0200, Christian König wrote: > Even for explicit synchronization we should wait for kernel fences. Yeah I don't think this patch makes much sense, because aside from etnaviv there's also msm and lima which allow you to ignore all dma_resv fences completely. But it's also not an issue because these drivers don't move buffers, don't have any other kernel fences and also don't do dynamic importing. I think the real fix is replacing the write argument to drm_sched_job_add_implicit_dependencies with dma_resv_usage and rolling that out. I'd just drop this for now, seems like a detour. -Daniel > > Signed-off-by: Christian König <christian.koenig@amd.com> > --- > drivers/gpu/drm/etnaviv/etnaviv_gem_submit.c | 27 ++++++++++++++++++-- > 1 file changed, 25 insertions(+), 2 deletions(-) > > diff --git a/drivers/gpu/drm/etnaviv/etnaviv_gem_submit.c b/drivers/gpu/drm/etnaviv/etnaviv_gem_submit.c > index 98bb5c9239de..3fedd29732d5 100644 > --- a/drivers/gpu/drm/etnaviv/etnaviv_gem_submit.c > +++ b/drivers/gpu/drm/etnaviv/etnaviv_gem_submit.c > @@ -171,6 +171,26 @@ static int submit_lock_objects(struct etnaviv_gem_submit *submit, > return ret; > } > > +/* TODO: This should be moved into the GPU scheduler if others need it */ > +static int submit_fence_kernel_sync(struct etnaviv_gem_submit *submit, > + struct dma_resv *resv) > +{ > + struct dma_resv_iter cursor; > + struct dma_fence *fence; > + int ret; > + > + dma_resv_for_each_fence(&cursor, resv, DMA_RESV_USAGE_KERNEL, fence) { > + /* Make sure to grab an additional ref on the added fence */ > + dma_fence_get(fence); > + ret = drm_sched_job_add_dependency(&submit->sched_job, fence); > + if (ret) { > + dma_fence_put(fence); > + return ret; > + } > + } > + return 0; > +} > + > static int submit_fence_sync(struct etnaviv_gem_submit *submit) > { > int i, ret = 0; > @@ -183,8 +203,11 @@ static int submit_fence_sync(struct etnaviv_gem_submit *submit) > if (ret) > return ret; > > - if (submit->flags & ETNA_SUBMIT_NO_IMPLICIT) > - continue; > + if (submit->flags & ETNA_SUBMIT_NO_IMPLICIT) { > + ret = submit_fence_kernel_sync(submit, robj); > + if (ret) > + return ret; > + } > > ret = drm_sched_job_add_implicit_dependencies(&submit->sched_job, > &bo->obj->base, > -- > 2.25.1 >
diff --git a/drivers/gpu/drm/etnaviv/etnaviv_gem_submit.c b/drivers/gpu/drm/etnaviv/etnaviv_gem_submit.c index 98bb5c9239de..3fedd29732d5 100644 --- a/drivers/gpu/drm/etnaviv/etnaviv_gem_submit.c +++ b/drivers/gpu/drm/etnaviv/etnaviv_gem_submit.c @@ -171,6 +171,26 @@ static int submit_lock_objects(struct etnaviv_gem_submit *submit, return ret; } +/* TODO: This should be moved into the GPU scheduler if others need it */ +static int submit_fence_kernel_sync(struct etnaviv_gem_submit *submit, + struct dma_resv *resv) +{ + struct dma_resv_iter cursor; + struct dma_fence *fence; + int ret; + + dma_resv_for_each_fence(&cursor, resv, DMA_RESV_USAGE_KERNEL, fence) { + /* Make sure to grab an additional ref on the added fence */ + dma_fence_get(fence); + ret = drm_sched_job_add_dependency(&submit->sched_job, fence); + if (ret) { + dma_fence_put(fence); + return ret; + } + } + return 0; +} + static int submit_fence_sync(struct etnaviv_gem_submit *submit) { int i, ret = 0; @@ -183,8 +203,11 @@ static int submit_fence_sync(struct etnaviv_gem_submit *submit) if (ret) return ret; - if (submit->flags & ETNA_SUBMIT_NO_IMPLICIT) - continue; + if (submit->flags & ETNA_SUBMIT_NO_IMPLICIT) { + ret = submit_fence_kernel_sync(submit, robj); + if (ret) + return ret; + } ret = drm_sched_job_add_implicit_dependencies(&submit->sched_job, &bo->obj->base,
Even for explicit synchronization we should wait for kernel fences. Signed-off-by: Christian König <christian.koenig@amd.com> --- drivers/gpu/drm/etnaviv/etnaviv_gem_submit.c | 27 ++++++++++++++++++-- 1 file changed, 25 insertions(+), 2 deletions(-)