Chris Wilson <chris@xxxxxxxxxxxxxxxxxx> writes: > On Mon, Dec 14, 2015 at 07:14:23PM +0200, Mika Kuoppala wrote: >> When we drop caches, this debugfs entry does hardware access later in >> the chain, when fences are updated, so it needs a runtime pm ref. >> >> Dropping caches is used by some igt/bat tests, so this fixes >> some unclaimed register access traces. >> >> Signed-off-by: Mika Kuoppala <mika.kuoppala@xxxxxxxxx> >> --- >> drivers/gpu/drm/i915/i915_debugfs.c | 3 +++ >> 1 file changed, 3 insertions(+) >> >> diff --git a/drivers/gpu/drm/i915/i915_debugfs.c b/drivers/gpu/drm/i915/i915_debugfs.c >> index 24318b7..bd8ba7e 100644 >> --- a/drivers/gpu/drm/i915/i915_debugfs.c >> +++ b/drivers/gpu/drm/i915/i915_debugfs.c >> @@ -4839,6 +4839,8 @@ i915_drop_caches_set(void *data, u64 val) >> if (ret) >> return ret; >> >> + intel_runtime_pm_get(dev_priv); > > The current idea of the very coarse granularity of rpm_get() is to do it > before struct_mutex (since rpm_get resume may try to acquire the mutex > iirc). > > Ok, fixing that may be bolting the stable door after the horse bolted, > but we should nevertheless. In my opinion, it would be more productive > to work with Imre on making rpm fine grained so that we don't so many > and can actually place the wakelock around the hardware access itself, > not every single path that *may* touch hardware. Please consider 1/6 v2 as it is needed to avoid random unclaimed accesses during igt/bat if the drop caches is used in wrong spot. We can forget the rest. -Mika > -Chris > > -- > Chris Wilson, Intel Open Source Technology Centre _______________________________________________ Intel-gfx mailing list Intel-gfx@xxxxxxxxxxxxxxxxxxxxx http://lists.freedesktop.org/mailman/listinfo/intel-gfx
