[PATCH] drm/ttm: Work around performance regression with VM_PFNMAP

Alex Deucher alexdeucher at gmail.com
Wed Mar 12 05:49:11 PDT 2014


On Wed, Mar 12, 2014 at 8:12 AM, Thomas Hellstrom <thellstrom at vmware.com> wrote:
> A performance regression was introduced in TTM in linux 3.13 when we started using
> VM_PFNMAP for shared mappings. In theory this should've been faster due to
> less page book-keeping but it appears like VM_PFNMAP + x86 PAT + write-combine
> is a particularly cpu-hungry combination, as seen by largely increased
> cpu-usage on r200 GL video playback.
>
> Until we've sorted out why, revert to always use VM_MIXEDMAP.
> Reference: freedesktop.org bugzilla bug #75719
>
> Reported-and-tested-by: <smoki00790 at gmail.com>
> Signed-off-by: Thomas Hellstrom <thellstrom at vmware.com>
> Cc: stable at vger.kernel.org

Acked-by: Alex Deucher <alexander.deucher at amd.com>

> ---
>  drivers/gpu/drm/ttm/ttm_bo_vm.c |   12 +++++++-----
>  1 file changed, 7 insertions(+), 5 deletions(-)
>
> diff --git a/drivers/gpu/drm/ttm/ttm_bo_vm.c b/drivers/gpu/drm/ttm/ttm_bo_vm.c
> index 801231c..0ce48e5 100644
> --- a/drivers/gpu/drm/ttm/ttm_bo_vm.c
> +++ b/drivers/gpu/drm/ttm/ttm_bo_vm.c
> @@ -339,11 +339,13 @@ int ttm_bo_mmap(struct file *filp, struct vm_area_struct *vma,
>         vma->vm_private_data = bo;
>
>         /*
> -        * PFNMAP is faster than MIXEDMAP due to reduced page
> -        * administration. So use MIXEDMAP only if private VMA, where
> -        * we need to support COW.
> +        * We'd like to use VM_PFNMAP on shared mappings, where
> +        * (vma->vm_flags & VM_SHARED) != 0, for performance reasons,
> +        * but for some reason VM_PFNMAP + x86 PAT + write-combine is very
> +        * bad for performance. Until that has been sorted out, use
> +        * VM_MIXEDMAP on all mappings. See freedesktop.org bug #75719
>          */
> -       vma->vm_flags |= (vma->vm_flags & VM_SHARED) ? VM_PFNMAP : VM_MIXEDMAP;
> +       vma->vm_flags |= VM_MIXEDMAP;
>         vma->vm_flags |= VM_IO | VM_DONTEXPAND | VM_DONTDUMP;
>         return 0;
>  out_unref:
> @@ -359,7 +361,7 @@ int ttm_fbdev_mmap(struct vm_area_struct *vma, struct ttm_buffer_object *bo)
>
>         vma->vm_ops = &ttm_bo_vm_ops;
>         vma->vm_private_data = ttm_bo_reference(bo);
> -       vma->vm_flags |= (vma->vm_flags & VM_SHARED) ? VM_PFNMAP : VM_MIXEDMAP;
> +       vma->vm_flags |= VM_MIXEDMAP;
>         vma->vm_flags |= VM_IO | VM_DONTEXPAND;
>         return 0;
>  }
> --
> 1.7.10.4
> _______________________________________________
> dri-devel mailing list
> dri-devel at lists.freedesktop.org
> http://lists.freedesktop.org/mailman/listinfo/dri-devel


More information about the dri-devel mailing list