| Message ID | 20220511115219.46507-1-maarten.lankhorst@linux.intel.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | drm/i915: Use i915_gem_object_ggtt_pin_ww for reloc_iomap | expand |
On 11/05/2022 12:52, Maarten Lankhorst wrote: > Instead of its own path, use the common path when it doesn't result > in evicting any vma. This fixes the case where we don't wait for > binding. > https://gitlab.freedesktop.org/drm/intel/-/issues/5806 If I'm reading that correctly waiting for the bind doesn't seem to help? > Fixes: b5cfe6f7a6e1 ("drm/i915: Remove short-term pins from execbuf, v6.") > Cc: Matthew Auld <matthew.auld@intel.com> > Reported-by: Mateusz Jończyk <mat.jonczyk@o2.pl> > Tested-by: Hans de Goede <hdegoede@redhat.com> > Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com> > --- > drivers/gpu/drm/i915/gem/i915_gem_execbuffer.c | 6 ++---- > 1 file changed, 2 insertions(+), 4 deletions(-) > > diff --git a/drivers/gpu/drm/i915/gem/i915_gem_execbuffer.c b/drivers/gpu/drm/i915/gem/i915_gem_execbuffer.c > index 498b458fd784..919d01082909 100644 > --- a/drivers/gpu/drm/i915/gem/i915_gem_execbuffer.c > +++ b/drivers/gpu/drm/i915/gem/i915_gem_execbuffer.c > @@ -1262,14 +1262,12 @@ static void *reloc_iomap(struct i915_vma *batch, > * Only attempt to pin the batch buffer to ggtt if the current batch > * is not inside ggtt, or the batch buffer is not misplaced. > */ > - if (!i915_is_ggtt(batch->vm)) { > + if (!i915_is_ggtt(batch->vm) || > + !i915_vma_misplaced(batch, 0, 0, PIN_MAPPABLE)) { > vma = i915_gem_object_ggtt_pin_ww(obj, &eb->ww, NULL, 0, 0, > PIN_MAPPABLE | > PIN_NONBLOCK /* NOWARN */ | > PIN_NOEVICT); > - } else if (i915_vma_is_map_and_fenceable(batch)) { > - __i915_vma_pin(batch); > - vma = batch; > } > > if (vma == ERR_PTR(-EDEADLK))
Op 11-05-2022 om 20:23 schreef Matthew Auld: > On 11/05/2022 12:52, Maarten Lankhorst wrote: >> Instead of its own path, use the common path when it doesn't result >> in evicting any vma. This fixes the case where we don't wait for >> binding. >> > > https://gitlab.freedesktop.org/drm/intel/-/issues/5806 > > If I'm reading that correctly waiting for the bind doesn't seem to help? I suspect the actual pinning there might do some stuff that we are not doing. It was working before the change, and manually calling pin caused the failure, so I reverted it back to what was working before. It was specifically the manual pin code that was failing. I can change the commit message if it helps. ~Maarten >> Fixes: b5cfe6f7a6e1 ("drm/i915: Remove short-term pins from execbuf, v6.") >> Cc: Matthew Auld <matthew.auld@intel.com> >> Reported-by: Mateusz Jończyk <mat.jonczyk@o2.pl> >> Tested-by: Hans de Goede <hdegoede@redhat.com> >> Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com> >> --- >> drivers/gpu/drm/i915/gem/i915_gem_execbuffer.c | 6 ++---- >> 1 file changed, 2 insertions(+), 4 deletions(-) >> >> diff --git a/drivers/gpu/drm/i915/gem/i915_gem_execbuffer.c b/drivers/gpu/drm/i915/gem/i915_gem_execbuffer.c >> index 498b458fd784..919d01082909 100644 >> --- a/drivers/gpu/drm/i915/gem/i915_gem_execbuffer.c >> +++ b/drivers/gpu/drm/i915/gem/i915_gem_execbuffer.c >> @@ -1262,14 +1262,12 @@ static void *reloc_iomap(struct i915_vma *batch, >> * Only attempt to pin the batch buffer to ggtt if the current batch >> * is not inside ggtt, or the batch buffer is not misplaced. >> */ >> - if (!i915_is_ggtt(batch->vm)) { >> + if (!i915_is_ggtt(batch->vm) || >> + !i915_vma_misplaced(batch, 0, 0, PIN_MAPPABLE)) { >> vma = i915_gem_object_ggtt_pin_ww(obj, &eb->ww, NULL, 0, 0, >> PIN_MAPPABLE | >> PIN_NONBLOCK /* NOWARN */ | >> PIN_NOEVICT); >> - } else if (i915_vma_is_map_and_fenceable(batch)) { >> - __i915_vma_pin(batch); >> - vma = batch; >> } >> if (vma == ERR_PTR(-EDEADLK))
On 11/05/2022 19:38, Maarten Lankhorst wrote: > Op 11-05-2022 om 20:23 schreef Matthew Auld: >> On 11/05/2022 12:52, Maarten Lankhorst wrote: >>> Instead of its own path, use the common path when it doesn't result >>> in evicting any vma. This fixes the case where we don't wait for >>> binding. >>> >> >> https://gitlab.freedesktop.org/drm/intel/-/issues/5806 >> >> If I'm reading that correctly waiting for the bind doesn't seem to help? > > I suspect the actual pinning there might do some stuff that we are not doing. > > It was working before the change, and manually calling pin caused the failure, so I reverted it back to what was working before. It was specifically the manual pin code that was failing. > > I can change the commit message if it helps. Hmm strange. With the commit message updated, Acked-by: Matthew Auld <matthew.auld@intel.com> > > ~Maarten > >>> Fixes: b5cfe6f7a6e1 ("drm/i915: Remove short-term pins from execbuf, v6.") >>> Cc: Matthew Auld <matthew.auld@intel.com> >>> Reported-by: Mateusz Jończyk <mat.jonczyk@o2.pl> >>> Tested-by: Hans de Goede <hdegoede@redhat.com> >>> Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com> >>> --- >>> drivers/gpu/drm/i915/gem/i915_gem_execbuffer.c | 6 ++---- >>> 1 file changed, 2 insertions(+), 4 deletions(-) >>> >>> diff --git a/drivers/gpu/drm/i915/gem/i915_gem_execbuffer.c b/drivers/gpu/drm/i915/gem/i915_gem_execbuffer.c >>> index 498b458fd784..919d01082909 100644 >>> --- a/drivers/gpu/drm/i915/gem/i915_gem_execbuffer.c >>> +++ b/drivers/gpu/drm/i915/gem/i915_gem_execbuffer.c >>> @@ -1262,14 +1262,12 @@ static void *reloc_iomap(struct i915_vma *batch, >>> * Only attempt to pin the batch buffer to ggtt if the current batch >>> * is not inside ggtt, or the batch buffer is not misplaced. >>> */ >>> - if (!i915_is_ggtt(batch->vm)) { >>> + if (!i915_is_ggtt(batch->vm) || >>> + !i915_vma_misplaced(batch, 0, 0, PIN_MAPPABLE)) { >>> vma = i915_gem_object_ggtt_pin_ww(obj, &eb->ww, NULL, 0, 0, >>> PIN_MAPPABLE | >>> PIN_NONBLOCK /* NOWARN */ | >>> PIN_NOEVICT); >>> - } else if (i915_vma_is_map_and_fenceable(batch)) { >>> - __i915_vma_pin(batch); >>> - vma = batch; >>> } >>> if (vma == ERR_PTR(-EDEADLK)) > >
diff --git a/drivers/gpu/drm/i915/gem/i915_gem_execbuffer.c b/drivers/gpu/drm/i915/gem/i915_gem_execbuffer.c index 498b458fd784..919d01082909 100644 --- a/drivers/gpu/drm/i915/gem/i915_gem_execbuffer.c +++ b/drivers/gpu/drm/i915/gem/i915_gem_execbuffer.c @@ -1262,14 +1262,12 @@ static void *reloc_iomap(struct i915_vma *batch, * Only attempt to pin the batch buffer to ggtt if the current batch * is not inside ggtt, or the batch buffer is not misplaced. */ - if (!i915_is_ggtt(batch->vm)) { + if (!i915_is_ggtt(batch->vm) || + !i915_vma_misplaced(batch, 0, 0, PIN_MAPPABLE)) { vma = i915_gem_object_ggtt_pin_ww(obj, &eb->ww, NULL, 0, 0, PIN_MAPPABLE | PIN_NONBLOCK /* NOWARN */ | PIN_NOEVICT); - } else if (i915_vma_is_map_and_fenceable(batch)) { - __i915_vma_pin(batch); - vma = batch; } if (vma == ERR_PTR(-EDEADLK))
Instead of its own path, use the common path when it doesn't result in evicting any vma. This fixes the case where we don't wait for binding. Fixes: b5cfe6f7a6e1 ("drm/i915: Remove short-term pins from execbuf, v6.") Cc: Matthew Auld <matthew.auld@intel.com> Reported-by: Mateusz Jończyk <mat.jonczyk@o2.pl> Tested-by: Hans de Goede <hdegoede@redhat.com> Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com> --- drivers/gpu/drm/i915/gem/i915_gem_execbuffer.c | 6 ++---- 1 file changed, 2 insertions(+), 4 deletions(-)