Message ID | 1453406585-10233-1-git-send-email-paulo.r.zanoni@intel.com (mailing list archive) |
---|---|
State | New, archived |
Headers | show |
Em Qui, 2016-01-21 às 18:03 -0200, Paulo Zanoni escreveu: > Instead of waiting for 50ms, just wait until the next vblank, since > it's the minimum requirement. The whole infrastructure of FBC is > based > on vblanks, so waiting for X vblanks instead of X milliseconds sounds > like the correct way to go. Besides, 50ms may be less than a vblank > on > super slow modes that may or may not exist. > > There are some small improvements in PC state residency (due to the > fact that we're now using 16ms for the common modes instead of 50ms), > but the biggest advantage is still the correctness of being > vblank-based instead of time-based. > > v2: > - Rebase after changing the patch order. > - Update the commit message. > v3: > - Fix bogus vblank_get() instead of vblank_count() (Ville). > - Don't forget to call drm_crtc_vblank_{get,put} (Chris, Ville) > - Adjust the performance details on the commit message. > v4: > - Don't grab the FBC mutex just to grab the vblank (Maarten) CCing Chris and Ville per Maarten's request on IRC: "would be nice if someone who reviewed the earlier version would ack it too". > > Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com> > --- > drivers/gpu/drm/i915/i915_drv.h | 2 +- > drivers/gpu/drm/i915/intel_fbc.c | 39 +++++++++++++++++++++++++++++- > --------- > 2 files changed, 30 insertions(+), 11 deletions(-) > > diff --git a/drivers/gpu/drm/i915/i915_drv.h > b/drivers/gpu/drm/i915/i915_drv.h > index 204661f..d8f21f0 100644 > --- a/drivers/gpu/drm/i915/i915_drv.h > +++ b/drivers/gpu/drm/i915/i915_drv.h > @@ -921,9 +921,9 @@ struct i915_fbc { > > struct intel_fbc_work { > bool scheduled; > + u32 scheduled_vblank; > struct work_struct work; > struct drm_framebuffer *fb; > - unsigned long enable_jiffies; > } work; > > const char *no_fbc_reason; > diff --git a/drivers/gpu/drm/i915/intel_fbc.c > b/drivers/gpu/drm/i915/intel_fbc.c > index a1988a4..3993b43 100644 > --- a/drivers/gpu/drm/i915/intel_fbc.c > +++ b/drivers/gpu/drm/i915/intel_fbc.c > @@ -381,7 +381,17 @@ static void intel_fbc_work_fn(struct work_struct > *__work) > container_of(__work, struct drm_i915_private, > fbc.work.work); > struct intel_fbc_work *work = &dev_priv->fbc.work; > struct intel_crtc *crtc = dev_priv->fbc.crtc; > - int delay_ms = 50; > + struct drm_vblank_crtc *vblank = &dev_priv->dev- > >vblank[crtc->pipe]; > + > + if (drm_crtc_vblank_get(&crtc->base)) { > + DRM_ERROR("vblank not available for FBC on pipe > %c\n", > + pipe_name(crtc->pipe)); > + > + mutex_lock(&dev_priv->fbc.lock); > + work->scheduled = false; > + mutex_unlock(&dev_priv->fbc.lock); > + return; > + } > > retry: > /* Delay the actual enabling to let pageflipping cease and > the > @@ -390,14 +400,16 @@ retry: > * vblank to pass after disabling the FBC before we attempt > * to modify the control registers. > * > - * A more complicated solution would involve tracking > vblanks > - * following the termination of the page-flipping sequence > - * and indeed performing the enable as a co-routine and not > - * waiting synchronously upon the vblank. > - * > * WaFbcWaitForVBlankBeforeEnable:ilk,snb > + * > + * It is also worth mentioning that since work- > >scheduled_vblank can be > + * updated multiple times by the other threads, hitting the > timeout is > + * not an error condition. We'll just end up hitting the > "goto retry" > + * case below. > */ > - wait_remaining_ms_from_jiffies(work->enable_jiffies, > delay_ms); > + wait_event_timeout(vblank->queue, > + drm_crtc_vblank_count(&crtc->base) != work- > >scheduled_vblank, > + msecs_to_jiffies(50)); > > mutex_lock(&dev_priv->fbc.lock); > > @@ -406,8 +418,7 @@ retry: > goto out; > > /* Were we delayed again while this function was sleeping? > */ > - if (time_after(work->enable_jiffies + > msecs_to_jiffies(delay_ms), > - jiffies)) { > + if (drm_crtc_vblank_count(&crtc->base) == work- > >scheduled_vblank) { > mutex_unlock(&dev_priv->fbc.lock); > goto retry; > } > @@ -419,6 +430,7 @@ retry: > > out: > mutex_unlock(&dev_priv->fbc.lock); > + drm_crtc_vblank_put(&crtc->base); > } > > static void intel_fbc_cancel_work(struct drm_i915_private *dev_priv) > @@ -434,13 +446,20 @@ static void > intel_fbc_schedule_activation(struct intel_crtc *crtc) > > WARN_ON(!mutex_is_locked(&dev_priv->fbc.lock)); > > + if (drm_crtc_vblank_get(&crtc->base)) { > + DRM_ERROR("vblank not available for FBC on pipe > %c\n", > + pipe_name(crtc->pipe)); > + return; > + } > + > /* It is useless to call intel_fbc_cancel_work() in this > function since > * we're not releasing fbc.lock, so it won't have an > opportunity to grab > * it to discover that it was cancelled. So we just update > the expected > * jiffy count. */ > work->fb = crtc->base.primary->fb; > work->scheduled = true; > - work->enable_jiffies = jiffies; > + work->scheduled_vblank = drm_crtc_vblank_count(&crtc->base); > + drm_crtc_vblank_put(&crtc->base); > > schedule_work(&work->work); > }
On Thu, Jan 21, 2016 at 12:03 PM Paulo Zanoni <paulo.r.zanoni@intel.com> wrote: > Instead of waiting for 50ms, just wait until the next vblank, since > it's the minimum requirement. The whole infrastructure of FBC is based > on vblanks, so waiting for X vblanks instead of X milliseconds sounds > like the correct way to go. Besides, 50ms may be less than a vblank on > super slow modes that may or may not exist. > > There are some small improvements in PC state residency (due to the > fact that we're now using 16ms for the common modes instead of 50ms), > but the biggest advantage is still the correctness of being > vblank-based instead of time-based. > > v2: > - Rebase after changing the patch order. > - Update the commit message. > v3: > - Fix bogus vblank_get() instead of vblank_count() (Ville). > - Don't forget to call drm_crtc_vblank_{get,put} (Chris, Ville) > - Adjust the performance details on the commit message. > v4: > - Don't grab the FBC mutex just to grab the vblank (Maarten) > > Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com> > --- > drivers/gpu/drm/i915/i915_drv.h | 2 +- > drivers/gpu/drm/i915/intel_fbc.c | 39 > +++++++++++++++++++++++++++++---------- > 2 files changed, 30 insertions(+), 11 deletions(-) > > diff --git a/drivers/gpu/drm/i915/i915_drv.h > b/drivers/gpu/drm/i915/i915_drv.h > index 204661f..d8f21f0 100644 > --- a/drivers/gpu/drm/i915/i915_drv.h > +++ b/drivers/gpu/drm/i915/i915_drv.h > @@ -921,9 +921,9 @@ struct i915_fbc { > > struct intel_fbc_work { > bool scheduled; > + u32 scheduled_vblank; > struct work_struct work; > struct drm_framebuffer *fb; > - unsigned long enable_jiffies; > } work; > > const char *no_fbc_reason; > diff --git a/drivers/gpu/drm/i915/intel_fbc.c > b/drivers/gpu/drm/i915/intel_fbc.c > index a1988a4..3993b43 100644 > --- a/drivers/gpu/drm/i915/intel_fbc.c > +++ b/drivers/gpu/drm/i915/intel_fbc.c > @@ -381,7 +381,17 @@ static void intel_fbc_work_fn(struct work_struct > *__work) > container_of(__work, struct drm_i915_private, > fbc.work.work); > struct intel_fbc_work *work = &dev_priv->fbc.work; > struct intel_crtc *crtc = dev_priv->fbc.crtc; > - int delay_ms = 50; > + struct drm_vblank_crtc *vblank = > &dev_priv->dev->vblank[crtc->pipe]; > + > + if (drm_crtc_vblank_get(&crtc->base)) { > + DRM_ERROR("vblank not available for FBC on pipe %c\n", > + pipe_name(crtc->pipe)); > + > + mutex_lock(&dev_priv->fbc.lock); > + work->scheduled = false; > I couldn't understand this here... doesn't look related to s/time/vblank... could you please explain? + mutex_unlock(&dev_priv->fbc.lock); > + return; > + } > > retry: > /* Delay the actual enabling to let pageflipping cease and the > @@ -390,14 +400,16 @@ retry: > * vblank to pass after disabling the FBC before we attempt > * to modify the control registers. > * > - * A more complicated solution would involve tracking vblanks > - * following the termination of the page-flipping sequence > - * and indeed performing the enable as a co-routine and not > - * waiting synchronously upon the vblank. > - * > * WaFbcWaitForVBlankBeforeEnable:ilk,snb > hm... is it still valid for newer platforms or we should put a if gen <=6 on these checks? > + * > + * It is also worth mentioning that since work->scheduled_vblank > can be > + * updated multiple times by the other threads, hitting the > timeout is > + * not an error condition. We'll just end up hitting the "goto > retry" > + * case below. > */ > - wait_remaining_ms_from_jiffies(work->enable_jiffies, delay_ms); > + wait_event_timeout(vblank->queue, > + drm_crtc_vblank_count(&crtc->base) != > work->scheduled_vblank, > + msecs_to_jiffies(50)); > > mutex_lock(&dev_priv->fbc.lock); > > @@ -406,8 +418,7 @@ retry: > goto out; > > /* Were we delayed again while this function was sleeping? */ > - if (time_after(work->enable_jiffies + msecs_to_jiffies(delay_ms), > - jiffies)) { > + if (drm_crtc_vblank_count(&crtc->base) == work->scheduled_vblank) { > mutex_unlock(&dev_priv->fbc.lock); > goto retry; > } > @@ -419,6 +430,7 @@ retry: > > out: > mutex_unlock(&dev_priv->fbc.lock); > + drm_crtc_vblank_put(&crtc->base); > } > > static void intel_fbc_cancel_work(struct drm_i915_private *dev_priv) > @@ -434,13 +446,20 @@ static void intel_fbc_schedule_activation(struct > intel_crtc *crtc) > > WARN_ON(!mutex_is_locked(&dev_priv->fbc.lock)); > > + if (drm_crtc_vblank_get(&crtc->base)) { > + DRM_ERROR("vblank not available for FBC on pipe %c\n", > + pipe_name(crtc->pipe)); > + return; > + } > + > /* It is useless to call intel_fbc_cancel_work() in this function > since > * we're not releasing fbc.lock, so it won't have an opportunity > to grab > * it to discover that it was cancelled. So we just update the > expected > * jiffy count. */ > work->fb = crtc->base.primary->fb; > work->scheduled = true; > - work->enable_jiffies = jiffies; > + work->scheduled_vblank = drm_crtc_vblank_count(&crtc->base); > + drm_crtc_vblank_put(&crtc->base); > > schedule_work(&work->work); > } > -- > 2.6.4 > > _______________________________________________ > Intel-gfx mailing list > Intel-gfx@lists.freedesktop.org > http://lists.freedesktop.org/mailman/listinfo/intel-gfx >
Em Ter, 2016-01-26 às 17:44 +0000, Rodrigo Vivi escreveu: > > > On Thu, Jan 21, 2016 at 12:03 PM Paulo Zanoni <paulo.r.zanoni@intel.c > om> wrote: > > Instead of waiting for 50ms, just wait until the next vblank, since > > it's the minimum requirement. The whole infrastructure of FBC is > > based > > on vblanks, so waiting for X vblanks instead of X milliseconds > > sounds > > like the correct way to go. Besides, 50ms may be less than a vblank > > on > > super slow modes that may or may not exist. > > > > There are some small improvements in PC state residency (due to the > > fact that we're now using 16ms for the common modes instead of > > 50ms), > > but the biggest advantage is still the correctness of being > > vblank-based instead of time-based. > > > > v2: > > - Rebase after changing the patch order. > > - Update the commit message. > > v3: > > - Fix bogus vblank_get() instead of vblank_count() (Ville). > > - Don't forget to call drm_crtc_vblank_{get,put} (Chris, Ville) > > - Adjust the performance details on the commit message. > > v4: > > - Don't grab the FBC mutex just to grab the vblank (Maarten) > > > > Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com> > > --- > > drivers/gpu/drm/i915/i915_drv.h | 2 +- > > drivers/gpu/drm/i915/intel_fbc.c | 39 > > +++++++++++++++++++++++++++++---------- > > 2 files changed, 30 insertions(+), 11 deletions(-) > > > > diff --git a/drivers/gpu/drm/i915/i915_drv.h > > b/drivers/gpu/drm/i915/i915_drv.h > > index 204661f..d8f21f0 100644 > > --- a/drivers/gpu/drm/i915/i915_drv.h > > +++ b/drivers/gpu/drm/i915/i915_drv.h > > @@ -921,9 +921,9 @@ struct i915_fbc { > > > > struct intel_fbc_work { > > bool scheduled; > > + u32 scheduled_vblank; > > struct work_struct work; > > struct drm_framebuffer *fb; > > - unsigned long enable_jiffies; > > } work; > > > > const char *no_fbc_reason; > > diff --git a/drivers/gpu/drm/i915/intel_fbc.c > > b/drivers/gpu/drm/i915/intel_fbc.c > > index a1988a4..3993b43 100644 > > --- a/drivers/gpu/drm/i915/intel_fbc.c > > +++ b/drivers/gpu/drm/i915/intel_fbc.c > > @@ -381,7 +381,17 @@ static void intel_fbc_work_fn(struct > > work_struct *__work) > > container_of(__work, struct drm_i915_private, > > fbc.work.work); > > struct intel_fbc_work *work = &dev_priv->fbc.work; > > struct intel_crtc *crtc = dev_priv->fbc.crtc; > > - int delay_ms = 50; > > + struct drm_vblank_crtc *vblank = &dev_priv->dev- > > >vblank[crtc->pipe]; > > + > > + if (drm_crtc_vblank_get(&crtc->base)) { > > + DRM_ERROR("vblank not available for FBC on pipe > > %c\n", > > + pipe_name(crtc->pipe)); > > + > > + mutex_lock(&dev_priv->fbc.lock); > > + work->scheduled = false; > I couldn't understand this here... doesn't look related to > s/time/vblank... > could you please explain? Previously we were just dealing with "wait a certain amount of time/jiffies", and for that we can just call the delay/sleep/wait/etc calls without needing any get/put calls. Now we'll wait for a certain number of vblanks, and we need to have the vblank interrupts enabled before we can wait for them. That's why we have the vblank get/put calls. And since get() can fail, we need an error path. Under normal FBC operation, every vblank get/put call should succeed because the pipe's supposed to be running. But in case we actually fail to get vblanks, we just exit from the work function. The way we signal "the work function is not running" is by setting work->scheduled to false. > > > + mutex_unlock(&dev_priv->fbc.lock); > > + return; > > + } > > > > retry: > > /* Delay the actual enabling to let pageflipping cease and > > the > > @@ -390,14 +400,16 @@ retry: > > * vblank to pass after disabling the FBC before we attempt > > * to modify the control registers. > > * > > - * A more complicated solution would involve tracking > > vblanks > > - * following the termination of the page-flipping sequence > > - * and indeed performing the enable as a co-routine and not > > - * waiting synchronously upon the vblank. > > - * > > * WaFbcWaitForVBlankBeforeEnable:ilk,snb > hm... is it still valid for newer platforms or we should put a if gen > <=6 on these checks? I tested this on BDW some time ago, and it seems we don't actually need the vblank wait anymore (although I didn't check the docs if we still need the wait). But I wanted to convert the code to use vblanks before optimizing it more. And the residency impact won't be big. > > > + * > > + * It is also worth mentioning that since work- > > >scheduled_vblank can be > > + * updated multiple times by the other threads, hitting the > > timeout is > > + * not an error condition. We'll just end up hitting the > > "goto retry" > > + * case below. > > */ > > - wait_remaining_ms_from_jiffies(work->enable_jiffies, > > delay_ms); > > + wait_event_timeout(vblank->queue, > > + drm_crtc_vblank_count(&crtc->base) != work- > > >scheduled_vblank, > > + msecs_to_jiffies(50)); > > > > mutex_lock(&dev_priv->fbc.lock); > > > > @@ -406,8 +418,7 @@ retry: > > goto out; > > > > /* Were we delayed again while this function was sleeping? > > */ > > - if (time_after(work->enable_jiffies + > > msecs_to_jiffies(delay_ms), > > - jiffies)) { > > + if (drm_crtc_vblank_count(&crtc->base) == work- > > >scheduled_vblank) { > > mutex_unlock(&dev_priv->fbc.lock); > > goto retry; > > } > > @@ -419,6 +430,7 @@ retry: > > > > out: > > mutex_unlock(&dev_priv->fbc.lock); > > + drm_crtc_vblank_put(&crtc->base); > > } > > > > static void intel_fbc_cancel_work(struct drm_i915_private > > *dev_priv) > > @@ -434,13 +446,20 @@ static void > > intel_fbc_schedule_activation(struct intel_crtc *crtc) > > > > WARN_ON(!mutex_is_locked(&dev_priv->fbc.lock)); > > > > + if (drm_crtc_vblank_get(&crtc->base)) { > > + DRM_ERROR("vblank not available for FBC on pipe > > %c\n", > > + pipe_name(crtc->pipe)); > > + return; > > + } > > + > > /* It is useless to call intel_fbc_cancel_work() in this > > function since > > * we're not releasing fbc.lock, so it won't have an > > opportunity to grab > > * it to discover that it was cancelled. So we just update > > the expected > > * jiffy count. */ > > work->fb = crtc->base.primary->fb; > > work->scheduled = true; > > - work->enable_jiffies = jiffies; > > + work->scheduled_vblank = drm_crtc_vblank_count(&crtc- > > >base); > > + drm_crtc_vblank_put(&crtc->base); > > > > schedule_work(&work->work); > > } > > -- > > 2.6.4 > > > > _______________________________________________ > > Intel-gfx mailing list > > Intel-gfx@lists.freedesktop.org > > http://lists.freedesktop.org/mailman/listinfo/intel-gfx > >
Thanks for all the explanation. Makes sense now and everything looks fine for me. Reviewed-by: Rodrigo Vivi <rodrigo.vivi@intel.com> On Tue, Jan 26, 2016 at 10:08 AM Zanoni, Paulo R <paulo.r.zanoni@intel.com> wrote: > Em Ter, 2016-01-26 às 17:44 +0000, Rodrigo Vivi escreveu: > > > > > > On Thu, Jan 21, 2016 at 12:03 PM Paulo Zanoni <paulo.r.zanoni@intel.c > > om> wrote: > > > Instead of waiting for 50ms, just wait until the next vblank, since > > > it's the minimum requirement. The whole infrastructure of FBC is > > > based > > > on vblanks, so waiting for X vblanks instead of X milliseconds > > > sounds > > > like the correct way to go. Besides, 50ms may be less than a vblank > > > on > > > super slow modes that may or may not exist. > > > > > > There are some small improvements in PC state residency (due to the > > > fact that we're now using 16ms for the common modes instead of > > > 50ms), > > > but the biggest advantage is still the correctness of being > > > vblank-based instead of time-based. > > > > > > v2: > > > - Rebase after changing the patch order. > > > - Update the commit message. > > > v3: > > > - Fix bogus vblank_get() instead of vblank_count() (Ville). > > > - Don't forget to call drm_crtc_vblank_{get,put} (Chris, Ville) > > > - Adjust the performance details on the commit message. > > > v4: > > > - Don't grab the FBC mutex just to grab the vblank (Maarten) > > > > > > Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com> > > > --- > > > drivers/gpu/drm/i915/i915_drv.h | 2 +- > > > drivers/gpu/drm/i915/intel_fbc.c | 39 > > > +++++++++++++++++++++++++++++---------- > > > 2 files changed, 30 insertions(+), 11 deletions(-) > > > > > > diff --git a/drivers/gpu/drm/i915/i915_drv.h > > > b/drivers/gpu/drm/i915/i915_drv.h > > > index 204661f..d8f21f0 100644 > > > --- a/drivers/gpu/drm/i915/i915_drv.h > > > +++ b/drivers/gpu/drm/i915/i915_drv.h > > > @@ -921,9 +921,9 @@ struct i915_fbc { > > > > > > struct intel_fbc_work { > > > bool scheduled; > > > + u32 scheduled_vblank; > > > struct work_struct work; > > > struct drm_framebuffer *fb; > > > - unsigned long enable_jiffies; > > > } work; > > > > > > const char *no_fbc_reason; > > > diff --git a/drivers/gpu/drm/i915/intel_fbc.c > > > b/drivers/gpu/drm/i915/intel_fbc.c > > > index a1988a4..3993b43 100644 > > > --- a/drivers/gpu/drm/i915/intel_fbc.c > > > +++ b/drivers/gpu/drm/i915/intel_fbc.c > > > @@ -381,7 +381,17 @@ static void intel_fbc_work_fn(struct > > > work_struct *__work) > > > container_of(__work, struct drm_i915_private, > > > fbc.work.work); > > > struct intel_fbc_work *work = &dev_priv->fbc.work; > > > struct intel_crtc *crtc = dev_priv->fbc.crtc; > > > - int delay_ms = 50; > > > + struct drm_vblank_crtc *vblank = &dev_priv->dev- > > > >vblank[crtc->pipe]; > > > + > > > + if (drm_crtc_vblank_get(&crtc->base)) { > > > + DRM_ERROR("vblank not available for FBC on pipe > > > %c\n", > > > + pipe_name(crtc->pipe)); > > > + > > > + mutex_lock(&dev_priv->fbc.lock); > > > + work->scheduled = false; > > I couldn't understand this here... doesn't look related to > > s/time/vblank... > > could you please explain? > > Previously we were just dealing with "wait a certain amount of > time/jiffies", and for that we can just call the delay/sleep/wait/etc > calls without needing any get/put calls. > > Now we'll wait for a certain number of vblanks, and we need to have the > vblank interrupts enabled before we can wait for them. That's why we > have the vblank get/put calls. And since get() can fail, we need an > error path. > > Under normal FBC operation, every vblank get/put call should succeed > because the pipe's supposed to be running. But in case we actually > fail to get vblanks, we just exit from the work function. The way we > signal "the work function is not running" is by setting work->scheduled > to false. > > > > > > + mutex_unlock(&dev_priv->fbc.lock); > > > + return; > > > + } > > > > > > retry: > > > /* Delay the actual enabling to let pageflipping cease and > > > the > > > @@ -390,14 +400,16 @@ retry: > > > * vblank to pass after disabling the FBC before we attempt > > > * to modify the control registers. > > > * > > > - * A more complicated solution would involve tracking > > > vblanks > > > - * following the termination of the page-flipping sequence > > > - * and indeed performing the enable as a co-routine and not > > > - * waiting synchronously upon the vblank. > > > - * > > > * WaFbcWaitForVBlankBeforeEnable:ilk,snb > > hm... is it still valid for newer platforms or we should put a if gen > > <=6 on these checks? > > I tested this on BDW some time ago, and it seems we don't actually need > the vblank wait anymore (although I didn't check the docs if we still > need the wait). But I wanted to convert the code to use vblanks before > optimizing it more. And the residency impact won't be big. > > > > > > + * > > > + * It is also worth mentioning that since work- > > > >scheduled_vblank can be > > > + * updated multiple times by the other threads, hitting the > > > timeout is > > > + * not an error condition. We'll just end up hitting the > > > "goto retry" > > > + * case below. > > > */ > > > - wait_remaining_ms_from_jiffies(work->enable_jiffies, > > > delay_ms); > > > + wait_event_timeout(vblank->queue, > > > + drm_crtc_vblank_count(&crtc->base) != work- > > > >scheduled_vblank, > > > + msecs_to_jiffies(50)); > > > > > > mutex_lock(&dev_priv->fbc.lock); > > > > > > @@ -406,8 +418,7 @@ retry: > > > goto out; > > > > > > /* Were we delayed again while this function was sleeping? > > > */ > > > - if (time_after(work->enable_jiffies + > > > msecs_to_jiffies(delay_ms), > > > - jiffies)) { > > > + if (drm_crtc_vblank_count(&crtc->base) == work- > > > >scheduled_vblank) { > > > mutex_unlock(&dev_priv->fbc.lock); > > > goto retry; > > > } > > > @@ -419,6 +430,7 @@ retry: > > > > > > out: > > > mutex_unlock(&dev_priv->fbc.lock); > > > + drm_crtc_vblank_put(&crtc->base); > > > } > > > > > > static void intel_fbc_cancel_work(struct drm_i915_private > > > *dev_priv) > > > @@ -434,13 +446,20 @@ static void > > > intel_fbc_schedule_activation(struct intel_crtc *crtc) > > > > > > WARN_ON(!mutex_is_locked(&dev_priv->fbc.lock)); > > > > > > + if (drm_crtc_vblank_get(&crtc->base)) { > > > + DRM_ERROR("vblank not available for FBC on pipe > > > %c\n", > > > + pipe_name(crtc->pipe)); > > > + return; > > > + } > > > + > > > /* It is useless to call intel_fbc_cancel_work() in this > > > function since > > > * we're not releasing fbc.lock, so it won't have an > > > opportunity to grab > > > * it to discover that it was cancelled. So we just update > > > the expected > > > * jiffy count. */ > > > work->fb = crtc->base.primary->fb; > > > work->scheduled = true; > > > - work->enable_jiffies = jiffies; > > > + work->scheduled_vblank = drm_crtc_vblank_count(&crtc- > > > >base); > > > + drm_crtc_vblank_put(&crtc->base); > > > > > > schedule_work(&work->work); > > > } > > > -- > > > 2.6.4 > > > > > > _______________________________________________ > > > Intel-gfx mailing list > > > Intel-gfx@lists.freedesktop.org > > > http://lists.freedesktop.org/mailman/listinfo/intel-gfx > > >
Em Sex, 2016-01-29 às 00:07 +0000, Rodrigo Vivi escreveu: > Thanks for all the explanation. > > Makes sense now and everything looks fine for me. > > Reviewed-by: Rodrigo Vivi <rodrigo.vivi@intel.com> Thanks Rodrigo and Maarten for the reviews. The CI part was discussed offline with Daniel, so I just pushed the patches. Thanks, Paulo > > > On Tue, Jan 26, 2016 at 10:08 AM Zanoni, Paulo R <paulo.r.zanoni@inte > l.com> wrote: > > Em Ter, 2016-01-26 às 17:44 +0000, Rodrigo Vivi escreveu: > > > > > > > > > On Thu, Jan 21, 2016 at 12:03 PM Paulo Zanoni <paulo.r.zanoni@int > > el.c > > > om> wrote: > > > > Instead of waiting for 50ms, just wait until the next vblank, > > since > > > > it's the minimum requirement. The whole infrastructure of FBC > > is > > > > based > > > > on vblanks, so waiting for X vblanks instead of X milliseconds > > > > sounds > > > > like the correct way to go. Besides, 50ms may be less than a > > vblank > > > > on > > > > super slow modes that may or may not exist. > > > > > > > > There are some small improvements in PC state residency (due to > > the > > > > fact that we're now using 16ms for the common modes instead of > > > > 50ms), > > > > but the biggest advantage is still the correctness of being > > > > vblank-based instead of time-based. > > > > > > > > v2: > > > > - Rebase after changing the patch order. > > > > - Update the commit message. > > > > v3: > > > > - Fix bogus vblank_get() instead of vblank_count() (Ville). > > > > - Don't forget to call drm_crtc_vblank_{get,put} (Chris, > > Ville) > > > > - Adjust the performance details on the commit message. > > > > v4: > > > > - Don't grab the FBC mutex just to grab the vblank (Maarten) > > > > > > > > Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com> > > > > --- > > > > drivers/gpu/drm/i915/i915_drv.h | 2 +- > > > > drivers/gpu/drm/i915/intel_fbc.c | 39 > > > > +++++++++++++++++++++++++++++---------- > > > > 2 files changed, 30 insertions(+), 11 deletions(-) > > > > > > > > diff --git a/drivers/gpu/drm/i915/i915_drv.h > > > > b/drivers/gpu/drm/i915/i915_drv.h > > > > index 204661f..d8f21f0 100644 > > > > --- a/drivers/gpu/drm/i915/i915_drv.h > > > > +++ b/drivers/gpu/drm/i915/i915_drv.h > > > > @@ -921,9 +921,9 @@ struct i915_fbc { > > > > > > > > struct intel_fbc_work { > > > > bool scheduled; > > > > + u32 scheduled_vblank; > > > > struct work_struct work; > > > > struct drm_framebuffer *fb; > > > > - unsigned long enable_jiffies; > > > > } work; > > > > > > > > const char *no_fbc_reason; > > > > diff --git a/drivers/gpu/drm/i915/intel_fbc.c > > > > b/drivers/gpu/drm/i915/intel_fbc.c > > > > index a1988a4..3993b43 100644 > > > > --- a/drivers/gpu/drm/i915/intel_fbc.c > > > > +++ b/drivers/gpu/drm/i915/intel_fbc.c > > > > @@ -381,7 +381,17 @@ static void intel_fbc_work_fn(struct > > > > work_struct *__work) > > > > container_of(__work, struct drm_i915_private, > > > > fbc.work.work); > > > > struct intel_fbc_work *work = &dev_priv->fbc.work; > > > > struct intel_crtc *crtc = dev_priv->fbc.crtc; > > > > - int delay_ms = 50; > > > > + struct drm_vblank_crtc *vblank = &dev_priv->dev- > > > > >vblank[crtc->pipe]; > > > > + > > > > + if (drm_crtc_vblank_get(&crtc->base)) { > > > > + DRM_ERROR("vblank not available for FBC on pipe > > > > %c\n", > > > > + pipe_name(crtc->pipe)); > > > > + > > > > + mutex_lock(&dev_priv->fbc.lock); > > > > + work->scheduled = false; > > > I couldn't understand this here... doesn't look related to > > > s/time/vblank... > > > could you please explain? > > > > Previously we were just dealing with "wait a certain amount of > > time/jiffies", and for that we can just call the > > delay/sleep/wait/etc > > calls without needing any get/put calls. > > > > Now we'll wait for a certain number of vblanks, and we need to have > > the > > vblank interrupts enabled before we can wait for them. That's why > > we > > have the vblank get/put calls. And since get() can fail, we need an > > error path. > > > > Under normal FBC operation, every vblank get/put call should > > succeed > > because the pipe's supposed to be running. But in case we actually > > fail to get vblanks, we just exit from the work function. The way > > we > > signal "the work function is not running" is by setting work- > > >scheduled > > to false. > > > > > > > > > + mutex_unlock(&dev_priv->fbc.lock); > > > > + return; > > > > + } > > > > > > > > retry: > > > > /* Delay the actual enabling to let pageflipping cease > > and > > > > the > > > > @@ -390,14 +400,16 @@ retry: > > > > * vblank to pass after disabling the FBC before we > > attempt > > > > * to modify the control registers. > > > > * > > > > - * A more complicated solution would involve tracking > > > > vblanks > > > > - * following the termination of the page-flipping > > sequence > > > > - * and indeed performing the enable as a co-routine and > > not > > > > - * waiting synchronously upon the vblank. > > > > - * > > > > * WaFbcWaitForVBlankBeforeEnable:ilk,snb > > > hm... is it still valid for newer platforms or we should put a if > > gen > > > <=6 on these checks? > > > > I tested this on BDW some time ago, and it seems we don't actually > > need > > the vblank wait anymore (although I didn't check the docs if we > > still > > need the wait). But I wanted to convert the code to use vblanks > > before > > optimizing it more. And the residency impact won't be big. > > > > > > > > > + * > > > > + * It is also worth mentioning that since work- > > > > >scheduled_vblank can be > > > > + * updated multiple times by the other threads, hitting > > the > > > > timeout is > > > > + * not an error condition. We'll just end up hitting > > the > > > > "goto retry" > > > > + * case below. > > > > */ > > > > - wait_remaining_ms_from_jiffies(work->enable_jiffies, > > > > delay_ms); > > > > + wait_event_timeout(vblank->queue, > > > > + drm_crtc_vblank_count(&crtc->base) != work- > > > > >scheduled_vblank, > > > > + msecs_to_jiffies(50)); > > > > > > > > mutex_lock(&dev_priv->fbc.lock); > > > > > > > > @@ -406,8 +418,7 @@ retry: > > > > goto out; > > > > > > > > /* Were we delayed again while this function was > > sleeping? > > > > */ > > > > - if (time_after(work->enable_jiffies + > > > > msecs_to_jiffies(delay_ms), > > > > - jiffies)) { > > > > + if (drm_crtc_vblank_count(&crtc->base) == work- > > > > >scheduled_vblank) { > > > > mutex_unlock(&dev_priv->fbc.lock); > > > > goto retry; > > > > } > > > > @@ -419,6 +430,7 @@ retry: > > > > > > > > out: > > > > mutex_unlock(&dev_priv->fbc.lock); > > > > + drm_crtc_vblank_put(&crtc->base); > > > > } > > > > > > > > static void intel_fbc_cancel_work(struct drm_i915_private > > > > *dev_priv) > > > > @@ -434,13 +446,20 @@ static void > > > > intel_fbc_schedule_activation(struct intel_crtc *crtc) > > > > > > > > WARN_ON(!mutex_is_locked(&dev_priv->fbc.lock)); > > > > > > > > + if (drm_crtc_vblank_get(&crtc->base)) { > > > > + DRM_ERROR("vblank not available for FBC on pipe > > > > %c\n", > > > > + pipe_name(crtc->pipe)); > > > > + return; > > > > + } > > > > + > > > > /* It is useless to call intel_fbc_cancel_work() in > > this > > > > function since > > > > * we're not releasing fbc.lock, so it won't have an > > > > opportunity to grab > > > > * it to discover that it was cancelled. So we just > > update > > > > the expected > > > > * jiffy count. */ > > > > work->fb = crtc->base.primary->fb; > > > > work->scheduled = true; > > > > - work->enable_jiffies = jiffies; > > > > + work->scheduled_vblank = drm_crtc_vblank_count(&crtc- > > > > >base); > > > > + drm_crtc_vblank_put(&crtc->base); > > > > > > > > schedule_work(&work->work); > > > > } > > > > -- > > > > 2.6.4 > > > > > > > > _______________________________________________ > > > > Intel-gfx mailing list > > > > Intel-gfx@lists.freedesktop.org > > > > http://lists.freedesktop.org/mailman/listinfo/intel-gfx > > > > > _______________________________________________ > Intel-gfx mailing list > Intel-gfx@lists.freedesktop.org > http://lists.freedesktop.org/mailman/listinfo/intel-gfx
diff --git a/drivers/gpu/drm/i915/i915_drv.h b/drivers/gpu/drm/i915/i915_drv.h index 204661f..d8f21f0 100644 --- a/drivers/gpu/drm/i915/i915_drv.h +++ b/drivers/gpu/drm/i915/i915_drv.h @@ -921,9 +921,9 @@ struct i915_fbc { struct intel_fbc_work { bool scheduled; + u32 scheduled_vblank; struct work_struct work; struct drm_framebuffer *fb; - unsigned long enable_jiffies; } work; const char *no_fbc_reason; diff --git a/drivers/gpu/drm/i915/intel_fbc.c b/drivers/gpu/drm/i915/intel_fbc.c index a1988a4..3993b43 100644 --- a/drivers/gpu/drm/i915/intel_fbc.c +++ b/drivers/gpu/drm/i915/intel_fbc.c @@ -381,7 +381,17 @@ static void intel_fbc_work_fn(struct work_struct *__work) container_of(__work, struct drm_i915_private, fbc.work.work); struct intel_fbc_work *work = &dev_priv->fbc.work; struct intel_crtc *crtc = dev_priv->fbc.crtc; - int delay_ms = 50; + struct drm_vblank_crtc *vblank = &dev_priv->dev->vblank[crtc->pipe]; + + if (drm_crtc_vblank_get(&crtc->base)) { + DRM_ERROR("vblank not available for FBC on pipe %c\n", + pipe_name(crtc->pipe)); + + mutex_lock(&dev_priv->fbc.lock); + work->scheduled = false; + mutex_unlock(&dev_priv->fbc.lock); + return; + } retry: /* Delay the actual enabling to let pageflipping cease and the @@ -390,14 +400,16 @@ retry: * vblank to pass after disabling the FBC before we attempt * to modify the control registers. * - * A more complicated solution would involve tracking vblanks - * following the termination of the page-flipping sequence - * and indeed performing the enable as a co-routine and not - * waiting synchronously upon the vblank. - * * WaFbcWaitForVBlankBeforeEnable:ilk,snb + * + * It is also worth mentioning that since work->scheduled_vblank can be + * updated multiple times by the other threads, hitting the timeout is + * not an error condition. We'll just end up hitting the "goto retry" + * case below. */ - wait_remaining_ms_from_jiffies(work->enable_jiffies, delay_ms); + wait_event_timeout(vblank->queue, + drm_crtc_vblank_count(&crtc->base) != work->scheduled_vblank, + msecs_to_jiffies(50)); mutex_lock(&dev_priv->fbc.lock); @@ -406,8 +418,7 @@ retry: goto out; /* Were we delayed again while this function was sleeping? */ - if (time_after(work->enable_jiffies + msecs_to_jiffies(delay_ms), - jiffies)) { + if (drm_crtc_vblank_count(&crtc->base) == work->scheduled_vblank) { mutex_unlock(&dev_priv->fbc.lock); goto retry; } @@ -419,6 +430,7 @@ retry: out: mutex_unlock(&dev_priv->fbc.lock); + drm_crtc_vblank_put(&crtc->base); } static void intel_fbc_cancel_work(struct drm_i915_private *dev_priv) @@ -434,13 +446,20 @@ static void intel_fbc_schedule_activation(struct intel_crtc *crtc) WARN_ON(!mutex_is_locked(&dev_priv->fbc.lock)); + if (drm_crtc_vblank_get(&crtc->base)) { + DRM_ERROR("vblank not available for FBC on pipe %c\n", + pipe_name(crtc->pipe)); + return; + } + /* It is useless to call intel_fbc_cancel_work() in this function since * we're not releasing fbc.lock, so it won't have an opportunity to grab * it to discover that it was cancelled. So we just update the expected * jiffy count. */ work->fb = crtc->base.primary->fb; work->scheduled = true; - work->enable_jiffies = jiffies; + work->scheduled_vblank = drm_crtc_vblank_count(&crtc->base); + drm_crtc_vblank_put(&crtc->base); schedule_work(&work->work); }
Instead of waiting for 50ms, just wait until the next vblank, since it's the minimum requirement. The whole infrastructure of FBC is based on vblanks, so waiting for X vblanks instead of X milliseconds sounds like the correct way to go. Besides, 50ms may be less than a vblank on super slow modes that may or may not exist. There are some small improvements in PC state residency (due to the fact that we're now using 16ms for the common modes instead of 50ms), but the biggest advantage is still the correctness of being vblank-based instead of time-based. v2: - Rebase after changing the patch order. - Update the commit message. v3: - Fix bogus vblank_get() instead of vblank_count() (Ville). - Don't forget to call drm_crtc_vblank_{get,put} (Chris, Ville) - Adjust the performance details on the commit message. v4: - Don't grab the FBC mutex just to grab the vblank (Maarten) Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com> --- drivers/gpu/drm/i915/i915_drv.h | 2 +- drivers/gpu/drm/i915/intel_fbc.c | 39 +++++++++++++++++++++++++++++---------- 2 files changed, 30 insertions(+), 11 deletions(-)