On Mon, 2022-09-12 at 15:43 +0300, Ville Syrjälä wrote:
> On Mon, Sep 12, 2022 at 02:19:57PM +0200, Thomas Hellström wrote:
> > Commit 39a2bd34c933 ("drm/i915: Use the vma resource as argument
> > for gtt
> > binding / unbinding") introduced a regression that due to the vma
> > resource
> > tracking of the binding state, dpt ptes were not correctly
> > repopulated.
>
> Doesn't this mean all ggtt ptes weren't repopulated? So I'm
> wondering how anything at all has been working?
ggtt ptes had a different check that was copy-pasted from the pre-vma-
resource code and that wasn't that strict. Hence why it worked.
/Thomas
>
> > Fix this by clearing the vma resource state before repopulating.
> > The state will subsequently be restored by the bind_vma operation.
> >
> > Fixes: 39a2bd34c933 ("drm/i915: Use the vma resource as argument
> > for gtt binding / unbinding")
> > Signed-off-by: Thomas Hellström <thomas.hellstrom@xxxxxxxxxxxxxxx>
> > ---
> > drivers/gpu/drm/i915/gt/intel_ggtt.c | 8 +++++++-
> > 1 file changed, 7 insertions(+), 1 deletion(-)
> >
> > diff --git a/drivers/gpu/drm/i915/gt/intel_ggtt.c
> > b/drivers/gpu/drm/i915/gt/intel_ggtt.c
> > index 30cf5c3369d9..2049a00417af 100644
> > --- a/drivers/gpu/drm/i915/gt/intel_ggtt.c
> > +++ b/drivers/gpu/drm/i915/gt/intel_ggtt.c
> > @@ -1275,10 +1275,16 @@ bool i915_ggtt_resume_vm(struct
> > i915_address_space *vm)
> > atomic_read(&vma->flags) &
> > I915_VMA_BIND_MASK;
> >
> > GEM_BUG_ON(!was_bound);
> > - if (!retained_ptes)
> > + if (!retained_ptes) {
> > + /*
> > + * Clear the bound flags of the vma
> > resource to allow
> > + * ptes to be repopulated.
> > + */
> > + vma->resource->bound_flags = 0;
> > vma->ops->bind_vma(vm, NULL, vma->resource,
> > obj ? obj->cache_level :
> > 0,
> > was_bound);
> > + }
> > if (obj) { /* only used during resume => exclusive
> > access */
> > write_domain_objs |= fetch_and_zero(&obj-
> > >write_domain);
> > obj->read_domains |= I915_GEM_DOMAIN_GTT;
> > --
> > 2.34.1
>
