Re: [PATCH v2] drm/amdgpu: implement workaround for sdma dcc corruption

[Christian König]

On 9/24/26 16:06, Pierre-Eric Pelloux-Prayer wrote:
> For unknown reasons, on gfx12 using multiple entities can causes
> random corruption of BOs with DCC.
> This workaround seems to prevent the issue until the root cause
> is understood and fixed.
> 
> Link: https://gitlab.freedesktop.org/drm/amd/-/work_items/5663
> Fixes: 3a6f6eeb3db5 ("drm/amdgpu: give ttm entities access to all the sdma scheds")
> Signed-off-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@xxxxxxx>
> Reviewed-by: Alex Deucher <alexander.deucher@xxxxxxx>

Reviewed-by: Christian König <christian.koenig@xxxxxxx>

> ---
>  drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c | 8 +++++++-
>  1 file changed, 7 insertions(+), 1 deletion(-)
> 
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
> index 3d620ec2937f..e9de5c4b7f63 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
> @@ -2367,7 +2367,7 @@ void amdgpu_ttm_fini(struct amdgpu_device *adev)
>  void amdgpu_ttm_enable_buffer_funcs(struct amdgpu_device *adev)
>  {
>  	struct ttm_resource_manager *man = ttm_manager_type(&adev->mman.bdev, TTM_PL_VRAM);
> -	u32 num_clear_entities, num_move_entities;
> +	u32 num_clear_entities, num_move_entities, sdma_ip_version;
>  	int r, i, j;
>  
>  	if (!adev->mman.initialized || amdgpu_in_reset(adev) ||
> @@ -2392,6 +2392,12 @@ void amdgpu_ttm_enable_buffer_funcs(struct amdgpu_device *adev)
>  
>  	num_clear_entities = MIN(adev->mman.num_buffer_funcs_scheds, TTM_NUM_MOVE_FENCES);
>  	num_move_entities = MIN(adev->mman.num_buffer_funcs_scheds, TTM_NUM_MOVE_FENCES);
> +        /* TODO: workaround for DCC corruption when moving BOs from multiple queues at
> +         * the same time: use a single queue until the root cause is identified and fixed.
> +         */
> +	sdma_ip_version = amdgpu_ip_version(adev, SDMA0_HWIP, 0);
> +	if (sdma_ip_version == IP_VERSION(7, 0, 0) || sdma_ip_version == IP_VERSION(7, 0, 1))
> +		num_move_entities = 1;
>  
>  	adev->mman.clear_entities = kcalloc(num_clear_entities,
>  						sizeof(struct amdgpu_ttm_buffer_entity),