Re: [PATCH v2 2/3] drm/amdgpu: introduce kfd user flag for amdgpu_bo

[Christian König <christian.koenig@xxxxxxx>]

Am 03.03.21 um 13:41 schrieb Nirmoy:
> [SNIP]
> > > +             struct amdgpu_bo_param *bp,
> > > +             struct amdgpu_bo **bo_ptr)
> > > +{
> > > +    u64 flags = bp->flags;
> > > +    int r;
> > > +
> > > +    bp->flags = bp->flags & ~AMDGPU_GEM_CREATE_SHADOW;
> > > +    bp->flags = bp->flags | AMDGPU_GEM_CREATE_KFD;
> > > +    r = amdgpu_bo_do_create(adev, bp, bo_ptr);
> > > +    if (r)
> > > +        return r;
> > > +
> > > +    if ((flags & AMDGPU_GEM_CREATE_SHADOW) && !(adev->flags & 
> > > AMD_IS_APU)) {
> > > +        if (!bp->resv)
> > > + WARN_ON(dma_resv_lock((*bo_ptr)->tbo.base.resv,
> > > +                            NULL));
> > > +
> > > +        r = amdgpu_bo_create_shadow(adev, bp->size, *bo_ptr);
> > > +
> > > +        if (!bp->resv)
> > > +            dma_resv_unlock((*bo_ptr)->tbo.base.resv);
> > > +
> > > +        if (r)
> > > +            amdgpu_bo_unref(bo_ptr);
> > > +    }
> >
> > I don't think the KFD should ever have a reason to use the shadow 
> > buffer functionality.
>
>
> This is interesting, I didn't know. I will remove 
> amdgpu_bo_create_shadow().

Well the VM code is using the shadow stuff.

> > > +
> > > +    return r;
> > > +}
> > > +
> > > +
> > >   /**
> > >    * amdgpu_bo_validate - validate an &amdgpu_bo buffer object
> > >    * @bo: pointer to the buffer object
> > > @@ -1309,7 +1355,7 @@ void amdgpu_bo_release_notify(struct 
> > > ttm_buffer_object *bo)
> > >
> > >       abo = ttm_to_amdgpu_bo(bo);
> > >
> > > -    if (abo->kfd_bo)
> > > +    if (abo->flags & AMDGPU_GEM_CREATE_KFD)
> > >           amdgpu_amdkfd_unreserve_memory_limit(abo);
> > >
> > >       /* We only remove the fence if the resv has individualized. */
> > > diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h 
> > > b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h
> > > index 8cd96c9330dd..665ee0015f06 100644
> > > --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h
> > > +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h
> > > @@ -245,6 +245,9 @@ void amdgpu_bo_placement_from_domain(struct 
> > > amdgpu_bo *abo, u32 domain);
> > >   int amdgpu_bo_create(struct amdgpu_device *adev,
> > >                struct amdgpu_bo_param *bp,
> > >                struct amdgpu_bo **bo_ptr);
> > > +int amdgpu_kfd_bo_create(struct amdgpu_device *adev,
> > > +             struct amdgpu_bo_param *bp,
> > > +             struct amdgpu_bo **bo_ptr);
> > >   int amdgpu_bo_create_reserved(struct amdgpu_device *adev,
> > >                     unsigned long size, int align,
> > >                     u32 domain, struct amdgpu_bo **bo_ptr,
> > > diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c 
> > > b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
> > > index 7b2db779f313..030bec382f54 100644
> > > --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
> > > +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
> > > @@ -164,7 +164,7 @@ static int amdgpu_verify_access(struct 
> > > ttm_buffer_object *bo, struct file *filp)
> > >        * Don't verify access for KFD BOs. They don't have a GEM
> > >        * object associated with them.
> > >        */
> > > -    if (abo->kfd_bo)
> > > +    if (abo->flags & AMDGPU_GEM_CREATE_KFD)
> > >           return 0;
> > >
> > >       if (amdgpu_ttm_tt_get_usermm(bo->ttm))
> > > diff --git a/include/uapi/drm/amdgpu_drm.h 
> > > b/include/uapi/drm/amdgpu_drm.h
> > > index 8b832f7458f2..f510e8302228 100644
> > > --- a/include/uapi/drm/amdgpu_drm.h
> > > +++ b/include/uapi/drm/amdgpu_drm.h
> > > @@ -142,6 +142,11 @@ extern "C" {
> > >    */
> > >   #define AMDGPU_GEM_CREATE_ENCRYPTED        (1 << 10)
> > >
> > > +/* Flag that the allocating BO's user is KFD. It should never be 
> > > used by
> > > + * user space applications
> > > + */
> > > +#define AMDGPU_GEM_CREATE_KFD            (1 << 20)
> >
> > Why 20? 11 is the next one here.
>
>
> I feel BO owner flag is different than others so wanted to reserve 
> some bits for grouping.
>
> I can assign it to 11 if that makes more sense.

We already have other flags which userspace is not allowed to use, so 
this is nothing special here.

Christian.

> Thanks,
>
> Nirmoy
>
>
>
> > Christian.
> >
> > > +
> > >   struct drm_amdgpu_gem_create_in  {
> > >       /** the requested memory size */
> > >       __u64 bo_size;
> > > --
> > > 2.30.1
> > >
> > > _______________________________________________
> > > amd-gfx mailing list
> > > amd-gfx@xxxxxxxxxxxxxxxxxxxxx
> > > https://nam11.safelinks.protection.outlook.com/?url=https%3A%2F%2Flists.freedesktop.org%2Fmailman%2Flistinfo%2Famd-gfx&amp;data=04%7C01%7Cnirmoy.das%40amd.com%7C5c41ee9032df45e36f1508d8de3c0c57%7C3dd8961fe4884e608e11a82d994e183d%7C0%7C0%7C637503696776437773%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C1000&amp;sdata=1Lbaor5CuBUsnxr%2BQgB6zDYbRQVPWogth7gpIOhYRFI%3D&amp;reserved=0 

_______________________________________________
amd-gfx mailing list
amd-gfx@xxxxxxxxxxxxxxxxxxxxx
https://lists.freedesktop.org/mailman/listinfo/amd-gfx