[RFC] drm/nouveau: use tile_mode and pte_kind for VM_BIND bo allocations

Dave Airlie airlied at gmail.com
Sun Apr 28 20:22:51 UTC 2024


This looks like a good start, though I wonder can we just refactor out
the common code from the vma and non-vma paths easier.

Also I wonder should we add a new GETPARAM so userspace can know this
change is available.

Of course we also need a pointer to the userspace implementation to
move forward.

Thanks,
Dave.

On Sun, 28 Apr 2024 at 19:32, Mohamed Ahmed
<mohamedahmedegypt2001 at gmail.com> wrote:
>
> Allows PTE kind and tile mode on BO create with VM_BIND. This is needed to support modifiers in NVK and ensure correctness when dealing with the nouveau GL driver.
>
> Signed-off-by: Mohamed Ahmed <mohamedahmedegypt2001 at gmail.com>
> ---
>  drivers/gpu/drm/nouveau/nouveau_bo.c | 24 +++++++++++++++++++-----
>  1 file changed, 19 insertions(+), 5 deletions(-)
>
> diff --git a/drivers/gpu/drm/nouveau/nouveau_bo.c b/drivers/gpu/drm/nouveau/nouveau_bo.c
> index db8cbf615..0da0b5286 100644
> --- a/drivers/gpu/drm/nouveau/nouveau_bo.c
> +++ b/drivers/gpu/drm/nouveau/nouveau_bo.c
> @@ -304,11 +304,25 @@ nouveau_bo_alloc(struct nouveau_cli *cli, u64 *size, int *align, u32 domain,
>                 }
>                 nvbo->page = vmm->page[pi].shift;
>         } else {
> -               /* reject other tile flags when in VM mode. */
> -               if (tile_mode)
> -                       return ERR_PTR(-EINVAL);
> -               if (tile_flags & ~NOUVEAU_GEM_TILE_NONCONTIG)
> -                       return ERR_PTR(-EINVAL);
> +               if (cli->device.info.family >= NV_DEVICE_INFO_V0_FERMI) {
> +                       nvbo->kind = (tile_flags & 0x0000ff00) >> 8;
> +                       if (!nvif_mmu_kind_valid(mmu, nvbo->kind)) {
> +                               kfree(nvbo);
> +                               return ERR_PTR(-EINVAL);
> +                       }
> +
> +                       nvbo->comp = mmu->kind[nvbo->kind] != nvbo->kind;
> +               } else if (cli->device.info.family >= NV_DEVICE_INFO_V0_TESLA) {
> +                       nvbo->kind = (tile_flags & 0x00007f00) >> 8;
> +                       nvbo->comp = (tile_flags & 0x00030000) >> 16;
> +                       if (!nvif_mmu_kind_valid(mmu, nvbo->kind)) {
> +                               kfree(nvbo);
> +                               return ERR_PTR(-EINVAL);
> +                       }
> +               } else {
> +                       nvbo->zeta = (tile_flags & 0x00000007);
> +               }
> +               nvbo->mode = tile_mode;
>
>                 /* Determine the desirable target GPU page size for the buffer. */
>                 for (i = 0; i < vmm->page_nr; i++) {
> --
> 2.44.0
>


More information about the dri-devel mailing list