Re: [PATCH v3 1/1] drm/amdgpu: flush gart changes after all BO recovery

["Das, Nirmoy" <nirmoy.das@xxxxxxx>]

On 5/28/2021 5:59 PM, Christian König wrote:
> Am 28.05.21 um 17:54 schrieb Nirmoy Das:
> > Don't flush gart changes after recovering each BO instead
> > do it after recovering all the BOs. Flishing gart also needed
> > for amdgpu_ttm_alloc_gart().
> >
> > v3: rename amdgpu_gart_tlb_flush() -> amdgpu_gart_invalidate_tlb()
> > v2: abstract out gart tlb flushing logic to amdgpu_gart.c
> >
> > Signed-off-by: Nirmoy Das <nirmoy.das@xxxxxxx>
> > ---
> >   drivers/gpu/drm/amd/amdgpu/amdgpu_gart.c    | 20 +++++++++++++++-----
> >   drivers/gpu/drm/amd/amdgpu/amdgpu_gart.h    |  2 +-
> >   drivers/gpu/drm/amd/amdgpu/amdgpu_gtt_mgr.c |  6 ++++++
> >   drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c     |  1 +
> >   4 files changed, 23 insertions(+), 6 deletions(-)
> >
> > diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gart.c 
> > b/drivers/gpu/drm/amd/amdgpu/amdgpu_gart.c
> > index 5562b5c90c03..992b59c29dd7 100644
> > --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_gart.c
> > +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gart.c
> > @@ -322,16 +322,26 @@ int amdgpu_gart_bind(struct amdgpu_device 
> > *adev, uint64_t offset,
> >       if (!adev->gart.ptr)
> >           return 0;
> >
> > -    r = amdgpu_gart_map(adev, offset, pages, dma_addr, flags,
> > -            adev->gart.ptr);
> > -    if (r)
> > -        return r;
> > +    return amdgpu_gart_map(adev, offset, pages, dma_addr, flags,
> > +                   adev->gart.ptr);
> > +}
> > +
> > +/**
> > + * amdgpu_gart_invalidate_tlb - invalidate gart TLB
> > + *
> > + * @adev: amdgpu device driver pointer
> > + *
> > + * Invalidate gart TLB which can be use as a way to flush gart changes
> > + *
> > + */
> > +void amdgpu_gart_invalidate_tlb(struct amdgpu_device *adev)
> > +{
> > +    int i;
> >
> >       mb();
> >       amdgpu_asic_flush_hdp(adev, NULL);
> >       for (i = 0; i < adev->num_vmhubs; i++)
> >           amdgpu_gmc_flush_gpu_tlb(adev, 0, i, 0);
> > -    return 0;
> >   }
> >
> >   /**
> > diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gart.h 
> > b/drivers/gpu/drm/amd/amdgpu/amdgpu_gart.h
> > index a25fe97b0196..e104022197ae 100644
> > --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_gart.h
> > +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gart.h
> > @@ -66,5 +66,5 @@ int amdgpu_gart_map(struct amdgpu_device *adev, 
> > uint64_t offset,
> >   int amdgpu_gart_bind(struct amdgpu_device *adev, uint64_t offset,
> >                int pages, struct page **pagelist,
> >                dma_addr_t *dma_addr, uint64_t flags);
> > -
> > +void amdgpu_gart_invalidate_tlb(struct amdgpu_device *adev);
> >   #endif
> > diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gtt_mgr.c 
> > b/drivers/gpu/drm/amd/amdgpu/amdgpu_gtt_mgr.c
> > index 8860545344c7..b01bc2346082 100644
> > --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_gtt_mgr.c
> > +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gtt_mgr.c
> > @@ -205,6 +205,7 @@ uint64_t amdgpu_gtt_mgr_usage(struct 
> > ttm_resource_manager *man)
> >    */
> >   int amdgpu_gtt_mgr_recover(struct ttm_resource_manager *man)
> >   {
> > +    struct amdgpu_device *adev = NULL;
> >       struct amdgpu_gtt_mgr *mgr = to_gtt_mgr(man);
> >       struct amdgpu_gtt_node *node;
> >       struct drm_mm_node *mm_node;
> > @@ -216,9 +217,14 @@ int amdgpu_gtt_mgr_recover(struct 
> > ttm_resource_manager *man)
> >           r = amdgpu_ttm_recover_gart(node->tbo);
> >           if (r)
> >               break;
> > +        if (!adev)
> > +            adev = amdgpu_ttm_adev(node->tbo->bdev);
>
> container_of(mgr, typeof(*adev), mman.gtt_mgr) is probably a bit 
> cleaner than this.


This was bothering me as well, thanks for this!


Nirmoy

> Christian.
>
> >       }
> >       spin_unlock(&mgr->lock);
> >
> > +    if (adev)
> > +        amdgpu_gart_invalidate_tlb(adev);
> > +
> >       return r;
> >   }
> >
> > diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c 
> > b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
> > index c0aef327292a..786650a4a493 100644
> > --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
> > +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
> > @@ -1014,6 +1014,7 @@ int amdgpu_ttm_alloc_gart(struct 
> > ttm_buffer_object *bo)
> >               return r;
> >           }
> >
> > +        amdgpu_gart_invalidate_tlb(adev);
> >           ttm_resource_free(bo, &bo->mem);
> >           bo->mem = tmp;
> >       }
> > --
> > 2.31.1
_______________________________________________
amd-gfx mailing list
amd-gfx@xxxxxxxxxxxxxxxxxxxxx
https://lists.freedesktop.org/mailman/listinfo/amd-gfx