[PATCH] drm/amdgpu: fix huge page handling on Vega10

[Kuehling, Felix]

On 2018-11-12 12:09 p.m., Christian König wrote:
> We accidentially set the huge flag on the parent instead of the childs.
> This caused some VM faults under memory pressure.

Reviewed-by: Felix Kuehling <Felix.Kuehling at amd.com>

I got a bit confused when re-reading this code. Maybe part of it is that
cursor.entry is not really an entry (as in PTE), but a page table.
cursor.pfn identifies the entry within the table. Maybe renaming
cursor.entry to something like cursor.table would make it slightly more
intuitive.

Regards,
  Felix


>
> Signed-off-by: Christian König <christian.koenig at amd.com>
> ---
>  drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c | 18 ++++++++++--------
>  1 file changed, 10 insertions(+), 8 deletions(-)
>
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
> index 352b30409060..dad0e2342df9 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
> @@ -1632,13 +1632,6 @@ static int amdgpu_vm_update_ptes(struct amdgpu_pte_update_params *params,
>  			continue;
>  		}
>  
> -		/* First check if the entry is already handled */
> -		if (cursor.pfn < frag_start) {
> -			cursor.entry->huge = true;
> -			amdgpu_vm_pt_next(adev, &cursor);
> -			continue;
> -		}
> -
>  		/* If it isn't already handled it can't be a huge page */
>  		if (cursor.entry->huge) {
>  			/* Add the entry to the relocated list to update it. */
> @@ -1701,8 +1694,17 @@ static int amdgpu_vm_update_ptes(struct amdgpu_pte_update_params *params,
>  			}
>  		} while (frag_start < entry_end);
>  
> -		if (frag >= shift)
> +		if (amdgpu_vm_pt_descendant(adev, &cursor)) {
> +			/* Mark all child entries as huge */
> +			while (cursor.pfn < frag_start) {
> +				cursor.entry->huge = true;
> +				amdgpu_vm_pt_next(adev, &cursor);
> +			}
> +
> +		} else if (frag >= shift) {
> +			/* or just move on to the next on the same level. */
>  			amdgpu_vm_pt_next(adev, &cursor);
> +		}
>  	}
>  
>  	return 0;