Re: [Intel-gfx] [PATCH 1/2] drm/ttm: prevent moving of pinned BOs

[Matthew Auld <matthew.william.auld@gmail.com>]

On Tue, 24 Jan 2023 at 09:51, Christian König
<ckoenig.leichtzumerken@gmail.com> wrote:
>
> Am 11.01.23 um 14:17 schrieb Matthew Auld:
> > On Wed, 11 Jan 2023 at 11:43, Christian König
> > <ckoenig.leichtzumerken@gmail.com> wrote:
> >> We have checks for this in the individual drivers move callback, but
> >> it's probably better to generally forbit that on a higher level.
> >>
> >> Also stops exporting ttm_resource_compat() since that's not necessary
> >> any more after removing the extra checks in vmwgfx.
> >>
> >> Signed-off-by: Christian König <christian.koenig@amd.com>
> >> ---
> >>   drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c |  4 ----
> >>   drivers/gpu/drm/nouveau/nouveau_bo.c    |  3 ---
> >>   drivers/gpu/drm/radeon/radeon_ttm.c     |  4 ----
> >>   drivers/gpu/drm/ttm/ttm_bo.c            | 20 ++++++++++++--------
> >>   drivers/gpu/drm/ttm/ttm_resource.c      |  1 -
> >>   drivers/gpu/drm/vmwgfx/vmwgfx_bo.c      | 19 ++-----------------
> >>   6 files changed, 14 insertions(+), 37 deletions(-)
> >>
> >> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
> >> index 068c2d8495fd..677cd7d91687 100644
> >> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
> >> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
> >> @@ -466,11 +466,7 @@ static int amdgpu_bo_move(struct ttm_buffer_object *bo, bool evict,
> >>                          return r;
> >>          }
> >>
> >> -       /* Can't move a pinned BO */
> >>          abo = ttm_to_amdgpu_bo(bo);
> >> -       if (WARN_ON_ONCE(abo->tbo.pin_count > 0))
> >> -               return -EINVAL;
> >> -
> >>          adev = amdgpu_ttm_adev(bo->bdev);
> >>
> >>          if (!old_mem || (old_mem->mem_type == TTM_PL_SYSTEM &&
> >> diff --git a/drivers/gpu/drm/nouveau/nouveau_bo.c b/drivers/gpu/drm/nouveau/nouveau_bo.c
> >> index 288eebc70a67..c2ec91cc845d 100644
> >> --- a/drivers/gpu/drm/nouveau/nouveau_bo.c
> >> +++ b/drivers/gpu/drm/nouveau/nouveau_bo.c
> >> @@ -1015,9 +1015,6 @@ nouveau_bo_move(struct ttm_buffer_object *bo, bool evict,
> >>          if (ret)
> >>                  goto out_ntfy;
> >>
> >> -       if (nvbo->bo.pin_count)
> >> -               NV_WARN(drm, "Moving pinned object %p!\n", nvbo);
> >> -
> >>          if (drm->client.device.info.family < NV_DEVICE_INFO_V0_TESLA) {
> >>                  ret = nouveau_bo_vm_bind(bo, new_reg, &new_tile);
> >>                  if (ret)
> >> diff --git a/drivers/gpu/drm/radeon/radeon_ttm.c b/drivers/gpu/drm/radeon/radeon_ttm.c
> >> index 1e8e287e113c..67075c85f847 100644
> >> --- a/drivers/gpu/drm/radeon/radeon_ttm.c
> >> +++ b/drivers/gpu/drm/radeon/radeon_ttm.c
> >> @@ -211,11 +211,7 @@ static int radeon_bo_move(struct ttm_buffer_object *bo, bool evict,
> >>          if (r)
> >>                  return r;
> >>
> >> -       /* Can't move a pinned BO */
> >>          rbo = container_of(bo, struct radeon_bo, tbo);
> >> -       if (WARN_ON_ONCE(rbo->tbo.pin_count > 0))
> >> -               return -EINVAL;
> >> -
> >>          rdev = radeon_get_rdev(bo->bdev);
> >>          if (old_mem->mem_type == TTM_PL_SYSTEM && bo->ttm == NULL) {
> >>                  ttm_bo_move_null(bo, new_mem);
> >> diff --git a/drivers/gpu/drm/ttm/ttm_bo.c b/drivers/gpu/drm/ttm/ttm_bo.c
> >> index 326a3d13a829..9baccb2f6e99 100644
> >> --- a/drivers/gpu/drm/ttm/ttm_bo.c
> >> +++ b/drivers/gpu/drm/ttm/ttm_bo.c
> >> @@ -894,14 +894,18 @@ int ttm_bo_validate(struct ttm_buffer_object *bo,
> >>          if (!placement->num_placement && !placement->num_busy_placement)
> >>                  return ttm_bo_pipeline_gutting(bo);
> >>
> >> -       /*
> >> -        * Check whether we need to move buffer.
> >> -        */
> >> -       if (!bo->resource || !ttm_resource_compat(bo->resource, placement)) {
> >> -               ret = ttm_bo_move_buffer(bo, placement, ctx);
> >> -               if (ret)
> >> -                       return ret;
> >> -       }
> >> +       /* Check whether we need to move buffer. */
> >> +       if (bo->resource && ttm_resource_compat(bo->resource, placement))
> >> +               return 0;
> > Note this now skips the tt create below (intentional?). I think i915
> > needed that, since it creates a dummy system resource initially for
> > all objects, and then relies on ZERO_ALLOC being set for certain
> > objects to know if the memory needs to be cleared or not when later
> > doing the dummy -> vram. Thoughts?
>
> That's unproblematic. On initial allocation bo->resource is NULL so we
> never branch out here.

Here is what I see in drm-tip, when first creating an object with ttm:

ttm_bo_init_reserved() -> ttm_resource_alloc(PL_SYSTEM, &bo->resource)
-> ttm_bo_validate()

So bo->resource is for sure not NULL on initial allocation, and is
pointing to PL_SYSTEM. And in i915 we initially stuff everything into
SYSTEM as a dummy placement.

IIRC you had a series that tried to address that (allowing NULL
resource or so), but it never landed:
https://patchwork.freedesktop.org/patch/500698/?series=107680&rev=2

>
> Christian.
>
> >
> >> +
> >> +       /* Moving of pinned BOs is forbidden */
> >> +       if (bo->pin_count)
> >> +               return -EINVAL;
> >> +
> >> +       ret = ttm_bo_move_buffer(bo, placement, ctx);
> >> +       if (ret)
> >> +               return ret;
> >> +
> >>          /*
> >>           * We might need to add a TTM.
> >>           */
> >> diff --git a/drivers/gpu/drm/ttm/ttm_resource.c b/drivers/gpu/drm/ttm/ttm_resource.c
> >> index b8a826a24fb2..7333f7a87a2f 100644
> >> --- a/drivers/gpu/drm/ttm/ttm_resource.c
> >> +++ b/drivers/gpu/drm/ttm/ttm_resource.c
> >> @@ -361,7 +361,6 @@ bool ttm_resource_compat(struct ttm_resource *res,
> >>
> >>          return false;
> >>   }
> >> -EXPORT_SYMBOL(ttm_resource_compat);
> >>
> >>   void ttm_resource_set_bo(struct ttm_resource *res,
> >>                           struct ttm_buffer_object *bo)
> >> diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_bo.c b/drivers/gpu/drm/vmwgfx/vmwgfx_bo.c
> >> index 321c551784a1..dbcef460c452 100644
> >> --- a/drivers/gpu/drm/vmwgfx/vmwgfx_bo.c
> >> +++ b/drivers/gpu/drm/vmwgfx/vmwgfx_bo.c
> >> @@ -87,12 +87,7 @@ int vmw_bo_pin_in_placement(struct vmw_private *dev_priv,
> >>          if (unlikely(ret != 0))
> >>                  goto err;
> >>
> >> -       if (buf->base.pin_count > 0)
> >> -               ret = ttm_resource_compat(bo->resource, placement)
> >> -                       ? 0 : -EINVAL;
> >> -       else
> >> -               ret = ttm_bo_validate(bo, placement, &ctx);
> >> -
> >> +       ret = ttm_bo_validate(bo, placement, &ctx);
> >>          if (!ret)
> >>                  vmw_bo_pin_reserved(buf, true);
> >>
> >> @@ -128,12 +123,6 @@ int vmw_bo_pin_in_vram_or_gmr(struct vmw_private *dev_priv,
> >>          if (unlikely(ret != 0))
> >>                  goto err;
> >>
> >> -       if (buf->base.pin_count > 0) {
> >> -               ret = ttm_resource_compat(bo->resource, &vmw_vram_gmr_placement)
> >> -                       ? 0 : -EINVAL;
> >> -               goto out_unreserve;
> >> -       }
> >> -
> >>          ret = ttm_bo_validate(bo, &vmw_vram_gmr_placement, &ctx);
> >>          if (likely(ret == 0) || ret == -ERESTARTSYS)
> >>                  goto out_unreserve;
> >> @@ -218,11 +207,7 @@ int vmw_bo_pin_in_start_of_vram(struct vmw_private *dev_priv,
> >>                  (void) ttm_bo_validate(bo, &vmw_sys_placement, &ctx);
> >>          }
> >>
> >> -       if (buf->base.pin_count > 0)
> >> -               ret = ttm_resource_compat(bo->resource, &placement)
> >> -                       ? 0 : -EINVAL;
> >> -       else
> >> -               ret = ttm_bo_validate(bo, &placement, &ctx);
> >> +       ret = ttm_bo_validate(bo, &placement, &ctx);
> >>
> >>          /* For some reason we didn't end up at the start of vram */
> >>          WARN_ON(ret == 0 && bo->resource->start != 0);
> >> --
> >> 2.34.1
> >>
>