[PATCH v3 2/5] drm/amdgpu: don't sync the user queue eviction fence

[Liang, Prike]

[Public]

> -----Original Message-----
> From: Koenig, Christian <Christian.Koenig at amd.com>
> Sent: Tuesday, May 6, 2025 4:23 PM
> To: Liang, Prike <Prike.Liang at amd.com>; amd-gfx at lists.freedesktop.org
> Cc: Deucher, Alexander <Alexander.Deucher at amd.com>
> Subject: Re: [PATCH v3 2/5] drm/amdgpu: don't sync the user queue eviction
> fence
>
> On 5/6/25 04:09, Liang, Prike wrote:
> > [Public]
> >
> >> From: Koenig, Christian <Christian.Koenig at amd.com>
> >> Sent: Wednesday, April 30, 2025 7:57 PM
> >> To: Liang, Prike <Prike.Liang at amd.com>; amd-gfx at lists.freedesktop.org
> >> Cc: Deucher, Alexander <Alexander.Deucher at amd.com>
> >> Subject: Re: [PATCH v3 2/5] drm/amdgpu: don't sync the user queue
> >> eviction fence
> >>
> >> On 4/30/25 04:40, Prike Liang wrote:
> >>> Don't return and sync the user queue eviction fence; otherwise, the
> >>> eviction fence will be returned as a dependent fence during VM
> >>> update and refer to the fence result in leakage.
> >>
> >> Please drop that patch, it shouldn't be needed any more after the
> >> changes in patch #1.
> >
> > Yes, may I get patch#1( drm/amdgpu: promote the implicit sync to the
> dependent read fences) reviewed?
>
> Sorry, I thought I've already done that.
>
> Feel free to add Reviewed-by: Christian König <christian.koenig at amd.com> to
> that patch.
>
> Regards,
> Christian.

Thanks, I have pushed the eviction fence release fixes.

> >
> >
> >> Regards,
> >> Christian.
> >>
> >>>
> >>> Signed-off-by: Prike Liang <Prike.Liang at amd.com>
> >>> ---
> >>>  drivers/gpu/drm/amd/amdgpu/amdgpu_eviction_fence.c | 11 +++++++++++
> >>> drivers/gpu/drm/amd/amdgpu/amdgpu_eviction_fence.h |  1 +
> >>>  drivers/gpu/drm/amd/amdgpu/amdgpu_sync.c           |  4 ++++
> >>>  3 files changed, 16 insertions(+)
> >>>
> >>> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_eviction_fence.c
> >>> b/drivers/gpu/drm/amd/amdgpu/amdgpu_eviction_fence.c
> >>> index d86e611a9ff4..6c9b2b43a929 100644
> >>> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_eviction_fence.c
> >>> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_eviction_fence.c
> >>> @@ -224,6 +224,17 @@ void amdgpu_eviction_fence_detach(struct
> >> amdgpu_eviction_fence_mgr *evf_mgr,
> >>>     dma_fence_put(stub);
> >>>  }
> >>>
> >>> +bool amdgpu_eviction_fence_valid(struct dma_fence *f) {
> >>> +
> >>> +   if(!f)
> >>> +           return false;
> >>> +   if (f->ops == &amdgpu_eviction_fence_ops)
> >>> +           return true;
> >>> +
> >>> +   return false;
> >>> +}
> >>> +
> >>>  int amdgpu_eviction_fence_init(struct amdgpu_eviction_fence_mgr
> >>> *evf_mgr)  {
> >>>     /* This needs to be done one time per open */ diff --git
> >>> a/drivers/gpu/drm/amd/amdgpu/amdgpu_eviction_fence.h
> >>> b/drivers/gpu/drm/amd/amdgpu/amdgpu_eviction_fence.h
> >>> index fcd867b7147d..d4e1975cac71 100644
> >>> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_eviction_fence.h
> >>> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_eviction_fence.h
> >>> @@ -42,6 +42,7 @@ struct amdgpu_eviction_fence_mgr {  };
> >>>
> >>>  /* Eviction fence helper functions */
> >>> +bool amdgpu_eviction_fence_valid(struct dma_fence *f);
> >>>  struct amdgpu_eviction_fence *
> >>>  amdgpu_eviction_fence_create(struct amdgpu_eviction_fence_mgr
> >>> *evf_mgr);
> >>>
> >>> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_sync.c
> >>> b/drivers/gpu/drm/amd/amdgpu/amdgpu_sync.c
> >>> index d6ae9974c952..8ac685eb1be1 100644
> >>> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_sync.c
> >>> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_sync.c
> >>> @@ -185,6 +185,10 @@ static bool amdgpu_sync_test_fence(struct
> >>> amdgpu_device *adev,  {
> >>>     void *fence_owner = amdgpu_sync_get_owner(f);
> >>>
> >>> +   /* don't sync the kgd userq eviction fence*/
> >>> +   if(amdgpu_eviction_fence_valid(f))
> >>> +           return false;
> >>> +
> >>>     /* Always sync to moves, no matter what */
> >>>     if (fence_owner == AMDGPU_FENCE_OWNER_UNDEFINED)
> >>>             return true;
> >