[PATCH 3/3] drm/amdgpu: stop evicting busy PDs/PTs

[Christian König]

No, that would create a nice deny of service.

An application could map the same tiny bit of memory all over the 
address space and so force the system to run out of memory.

Christian.

Am 08.03.19 um 21:46 schrieb Zeng, Oak:
> Should we even consider not evicting all PDs/PTs, busy or not. See PDs/PTs only use a tiny bit (roughly ~0.05%) of the total memory. You really don't get much memory when you evict them. But I doubt the evict/restore performance penalty is big.
>
> Thanks,
> Oak
>
> -----Original Message-----
> From: amd-gfx <amd-gfx-bounces at lists.freedesktop.org> On Behalf Of Christian König
> Sent: Friday, March 8, 2019 9:32 AM
> To: amd-gfx at lists.freedesktop.org
> Subject: [PATCH 3/3] drm/amdgpu: stop evicting busy PDs/PTs
>
> Otherwise we won't be able to cleanly handle page faults.
>
> Signed-off-by: Christian König <christian.koenig at amd.com>
> ---
>   drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c | 7 +++++++
>   1 file changed, 7 insertions(+)
>
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
> index cd0ccfbbcb84..c1240bf243ba 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
> @@ -1371,6 +1371,13 @@ static bool amdgpu_ttm_bo_eviction_valuable(struct ttm_buffer_object *bo,
>   	struct dma_fence *f;
>   	int i;
>   
> +	/* Don't evict VM page tables while they are busy, otherwise we can't
> +	 * cleanly handle page faults.
> +	 */
> +	if (bo->type == ttm_bo_type_kernel &&
> +	    !reservation_object_test_signaled_rcu(bo->resv, true))
> +		return false;
> +
>   	/* If bo is a KFD BO, check if the bo belongs to the current process.
>   	 * If true, then return false as any KFD process needs all its BOs to
>   	 * be resident to run successfully