[PATCH v2 2/5] drm/amdgpu: move GTT to shmem after eviction for hibernation

[Christian König]

On 04.07.25 12:12, Samuel Zhang wrote:
> When hibernate with data center dGPUs, huge number of VRAM BOs evicted
> to GTT and takes too much system memory. This will cause hibernation
> fail due to insufficient memory for creating the hibernation image.
> 
> Move GTT BOs to shmem in KMD, then shmem to swap disk in kernel
> hibernation code to make room for hibernation image.
> 
> Signed-off-by: Samuel Zhang <guoqing.zhang at amd.com>
> ---
>  drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c | 13 ++++++++++++-
>  1 file changed, 12 insertions(+), 1 deletion(-)
> 
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
> index 27ab4e754b2a..a0b0682236e3 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
> @@ -2414,6 +2414,7 @@ int amdgpu_fill_buffer(struct amdgpu_bo *bo,
>  int amdgpu_ttm_evict_resources(struct amdgpu_device *adev, int mem_type)
>  {
>  	struct ttm_resource_manager *man;
> +	int r;
>  
>  	switch (mem_type) {
>  	case TTM_PL_VRAM:
> @@ -2428,7 +2429,17 @@ int amdgpu_ttm_evict_resources(struct amdgpu_device *adev, int mem_type)
>  		return -EINVAL;
>  	}
>  
> -	return ttm_resource_manager_evict_all(&adev->mman.bdev, man);
> +	r = ttm_resource_manager_evict_all(&adev->mman.bdev, man);
> +	if (r) {
> +		DRM_ERROR("Failed to evict memory type %d\n", mem_type);
> +		return r;
> +	}
> +	if (adev->in_s4 && mem_type == TTM_PL_VRAM) {
> +		r = ttm_device_prepare_hibernation();
> +		if (r)
> +			DRM_ERROR("Failed to swap out, %d\n", r);
> +	}
> +	return r;

That call needs to go into a separate amdgpu_ttm_* function and only be called from amdgpu_device_evict_resources().

Otherwise the debugfs tests will trigger it as well which is undesirable.

Regards,
Christian.

>  }
>  
>  #if defined(CONFIG_DEBUG_FS)