Re: [Intel-gfx] [PATCH v9 32/70] drm/i915: Prepare for obj->mm.lock removal, v2.

["Thomas Hellström" <thomas.hellstrom@linux.intel.com>]

On Tue, 2021-03-23 at 16:18 +0000, Matthew Auld wrote:
> On Tue, 23 Mar 2021 at 15:52, Maarten Lankhorst
> <maarten.lankhorst@linux.intel.com> wrote:
> > 
> > From: Thomas Hellström <thomas.hellstrom@intel.com>
> > 
> > Stolen objects need to lock, and we may call put_pages when
> > refcount drops to 0, ensure all calls are handled correctly.
> > 
> > Changes since v1:
> > - Rebase on top of upstream changes.
> > 
> > Idea-from: Thomas Hellström <thomas.hellstrom@intel.com>
> > Signed-off-by: Maarten Lankhorst
> > <maarten.lankhorst@linux.intel.com>
> > Signed-off-by: Thomas Hellström <thomas.hellstrom@linux.intel.com>
> > ---
> >  drivers/gpu/drm/i915/gem/i915_gem_object.h | 14 ++++++++++++++
> >  drivers/gpu/drm/i915/gem/i915_gem_pages.c  | 14 ++++++++++++--
> >  drivers/gpu/drm/i915/gem/i915_gem_stolen.c | 12 +++++++-----
> >  3 files changed, 33 insertions(+), 7 deletions(-)
> > 
> > diff --git a/drivers/gpu/drm/i915/gem/i915_gem_object.h
> > b/drivers/gpu/drm/i915/gem/i915_gem_object.h
> > index 983f2d4b2a85..74de195b57de 100644
> > --- a/drivers/gpu/drm/i915/gem/i915_gem_object.h
> > +++ b/drivers/gpu/drm/i915/gem/i915_gem_object.h
> > @@ -144,6 +144,20 @@ i915_gem_object_put(struct drm_i915_gem_object
> > *obj)
> > 
> >  #define assert_object_held(obj) dma_resv_assert_held((obj)-
> > >base.resv)
> > 
> > +/*
> > + * If more than one potential simultaneous locker, assert held.
> > + */
> > +static inline void assert_object_held_shared(struct
> > drm_i915_gem_object *obj)
> > +{
> > +       /*
> > +        * Note mm list lookup is protected by
> 
> What is meant with mm list here? Maybe just a stale comment?

That would be the i915->mm lists, (shrink and purge).

> 
> > +        * kref_get_unless_zero().
> > +        */
> > +       if (IS_ENABLED(CONFIG_LOCKDEP) &&
> > +           kref_read(&obj->base.refcount) > 0)
> > +               lockdep_assert_held(&obj->mm.lock);
> > +}
> > +
> >  static inline int __i915_gem_object_lock(struct
> > drm_i915_gem_object *obj,
> >                                          struct i915_gem_ww_ctx
> > *ww,
> >                                          bool intr)
> > diff --git a/drivers/gpu/drm/i915/gem/i915_gem_pages.c
> > b/drivers/gpu/drm/i915/gem/i915_gem_pages.c
> > index a24617af3c93..2d0065fa6e80 100644
> > --- a/drivers/gpu/drm/i915/gem/i915_gem_pages.c
> > +++ b/drivers/gpu/drm/i915/gem/i915_gem_pages.c
> > @@ -19,7 +19,7 @@ void __i915_gem_object_set_pages(struct
> > drm_i915_gem_object *obj,
> >         bool shrinkable;
> >         int i;
> > 
> > -       lockdep_assert_held(&obj->mm.lock);
> > +       assert_object_held_shared(obj);
> > 
> >         if (i915_gem_object_is_volatile(obj))
> >                 obj->mm.madv = I915_MADV_DONTNEED;
> > @@ -70,6 +70,7 @@ void __i915_gem_object_set_pages(struct
> > drm_i915_gem_object *obj,
> >                 struct list_head *list;
> >                 unsigned long flags;
> > 
> > +               lockdep_assert_held(&obj->mm.lock);
> >                 spin_lock_irqsave(&i915->mm.obj_lock, flags);
> > 
> >                 i915->mm.shrink_count++;
> > @@ -91,6 +92,8 @@ int ____i915_gem_object_get_pages(struct
> > drm_i915_gem_object *obj)
> >         struct drm_i915_private *i915 = to_i915(obj->base.dev);
> >         int err;
> > 
> > +       assert_object_held_shared(obj);
> > +
> >         if (unlikely(obj->mm.madv != I915_MADV_WILLNEED)) {
> >                 drm_dbg(&i915->drm,
> >                         "Attempting to obtain a purgeable
> > object\n");
> > @@ -118,6 +121,8 @@ int __i915_gem_object_get_pages(struct
> > drm_i915_gem_object *obj)
> >         if (err)
> >                 return err;
> > 
> > +       assert_object_held_shared(obj);
> > +
> >         if (unlikely(!i915_gem_object_has_pages(obj))) {
> >                 GEM_BUG_ON(i915_gem_object_has_pinned_pages(obj));
> > 
> > @@ -145,7 +150,7 @@ void i915_gem_object_truncate(struct
> > drm_i915_gem_object *obj)
> >  /* Try to discard unwanted pages */
> >  void i915_gem_object_writeback(struct drm_i915_gem_object *obj)
> >  {
> > -       lockdep_assert_held(&obj->mm.lock);
> > +       assert_object_held_shared(obj);
> >         GEM_BUG_ON(i915_gem_object_has_pages(obj));
> > 
> >         if (obj->ops->writeback)
> > @@ -176,6 +181,8 @@ __i915_gem_object_unset_pages(struct
> > drm_i915_gem_object *obj)
> >  {
> >         struct sg_table *pages;
> > 
> > +       assert_object_held_shared(obj);
> > +
> >         pages = fetch_and_zero(&obj->mm.pages);
> >         if (IS_ERR_OR_NULL(pages))
> >                 return pages;
> > @@ -203,6 +210,9 @@ int __i915_gem_object_put_pages_locked(struct
> > drm_i915_gem_object *obj)
> >         if (i915_gem_object_has_pinned_pages(obj))
> >                 return -EBUSY;
> > 
> > +       /* May be called by shrinker from within get_pages() (on
> > another bo) */
> > +       assert_object_held_shared(obj);
> > +
> >         i915_gem_object_release_mmap_offset(obj);
> > 
> >         /*
> > diff --git a/drivers/gpu/drm/i915/gem/i915_gem_stolen.c
> > b/drivers/gpu/drm/i915/gem/i915_gem_stolen.c
> > index 7cdb32d881d9..b0597de206de 100644
> > --- a/drivers/gpu/drm/i915/gem/i915_gem_stolen.c
> > +++ b/drivers/gpu/drm/i915/gem/i915_gem_stolen.c
> > @@ -637,13 +637,15 @@ static int
> > __i915_gem_object_create_stolen(struct intel_memory_region *mem,
> >         cache_level = HAS_LLC(mem->i915) ? I915_CACHE_LLC :
> > I915_CACHE_NONE;
> >         i915_gem_object_set_cache_coherency(obj, cache_level);
> > 
> > -       err = i915_gem_object_pin_pages(obj);
> > -       if (err)
> > -               return err;
> > +       if (WARN_ON(!i915_gem_object_trylock(obj)))
> > +               return -EBUSY;
> > 
> > -       i915_gem_object_init_memory_region(obj, mem);
> > +       err = i915_gem_object_pin_pages(obj);
> > +       if (!err)
> > +               i915_gem_object_init_memory_region(obj, mem);
> 
> Probably more consistent to call init_memory_region() before calling
> pin_pages(), but I guess it was already like that, plus it doesn't
> seem to matter for stolen memory.

Yes, I agree, Probably a rebasing error? In the DG1 enabling code we
add the object to the relevant region mm lists during get_pages(), and
to do that the region needs to be initialized.

/Thomas

> 
> Reviewed-by: Matthew Auld <matthew.auld@intel.com>


_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/intel-gfx