[PATCH 1/1] drm/amdgpu: flush gpu tlb after a gart allocation

[Christian König]

Am 28.05.21 um 14:54 schrieb Nirmoy Das:
> Flush gpu tlb in amdgpu_ttm_alloc_gart(). Also
> don't flush gpu tlb after recovering each BO.
>
> Signed-off-by: Nirmoy Das <nirmoy.das at amd.com>
> ---
>   drivers/gpu/drm/amd/amdgpu/amdgpu_gart.c | 2 --
>   drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c  | 5 ++++-
>   2 files changed, 4 insertions(+), 3 deletions(-)
>
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gart.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_gart.c
> index 5562b5c90c03..68d41063e120 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_gart.c
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gart.c
> @@ -329,8 +329,6 @@ int amdgpu_gart_bind(struct amdgpu_device *adev, uint64_t offset,
>   
>   	mb();
>   	amdgpu_asic_flush_hdp(adev, NULL);

The memory barrier and HDP flush can be moved as well.

> -	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_ttm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
> index c0aef327292a..e68b5dab84a4 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
> @@ -977,7 +977,7 @@ int amdgpu_ttm_alloc_gart(struct ttm_buffer_object *bo)
>   	struct ttm_placement placement;
>   	struct ttm_place placements;
>   	uint64_t addr, flags;
> -	int r;
> +	int r, i;
>   
>   	if (bo->mem.start != AMDGPU_BO_INVALID_OFFSET)
>   		return 0;
> @@ -1014,6 +1014,9 @@ int amdgpu_ttm_alloc_gart(struct ttm_buffer_object *bo)
>   			return r;
>   		}
>   
> +		for (i = 0; i < adev->num_vmhubs; i++)
> +			amdgpu_gmc_flush_gpu_tlb(adev, 0, i, 0);
> +

Probably best to make that a function in amdgpu_gart.c to keep the GART 
functionality properly abstracted.

This can then be called from both amdgpu_gtt_mgr_recover() after 
dropping the lock and here.

Regards,
Christian.

>   		ttm_resource_free(bo, &bo->mem);
>   		bo->mem = tmp;
>   	}