| Message ID | 20220518130716.10936-8-anshuman.gupta@intel.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | DG2 VRAM_SR Support | expand |
On Wed, May 18, 2022 at 06:37:16PM +0530, Anshuman Gupta wrote: > Intel Client DGFX card supports D3Cold with two option. > D3Cold-off zero watt, D3Cold-VRAM Self Refresh. > > i915 requires to evict the lmem objects to smem in order to > support D3Cold-Off, which increases i915 the suspend/resume > latency. Enabling VRAM Self Refresh feature optimize the > latency with additional power cost which required to retain > the lmem. > > Adding intel_runtime_idle (runtime_idle callback) to enable > VRAM_SR, it will be used for policy to choose > between D3Cold-off vs D3Cold-VRAM_SR. > > Since we have introduced i915 runtime_idle callback. > It need to be warranted that Runtime PM Core invokes runtime_idle > callback when runtime usages count becomes zero. That requires > to use pm_runtime_put instead of pm_runtime_put_autosuspend. > > Cc: Rodrigo Vivi <rodrigo.vivi@intel.com> > Cc: Chris Wilson <chris.p.wilson@intel.com> > Signed-off-by: Anshuman Gupta <anshuman.gupta@intel.com> > --- > drivers/gpu/drm/i915/i915_driver.c | 26 +++++++++++++++++++++++++ > drivers/gpu/drm/i915/intel_runtime_pm.c | 3 +-- > 2 files changed, 27 insertions(+), 2 deletions(-) > > diff --git a/drivers/gpu/drm/i915/i915_driver.c b/drivers/gpu/drm/i915/i915_driver.c > index 5a9d5529fc90..bbb11c632799 100644 > --- a/drivers/gpu/drm/i915/i915_driver.c > +++ b/drivers/gpu/drm/i915/i915_driver.c > @@ -1541,6 +1541,31 @@ static int i915_pm_restore(struct device *kdev) > return i915_pm_resume(kdev); > } > > +static int intel_runtime_idle(struct device *kdev) > +{ > + struct drm_i915_private *dev_priv = kdev_to_i915(kdev); > + int ret = 1; > + > + if (!HAS_LMEM_SR(dev_priv)) { > + /*TODO: Prepare for D3Cold-Off */ > + goto out; > + } > + > + disable_rpm_wakeref_asserts(&dev_priv->runtime_pm); > + > + ret = intel_pm_vram_sr(dev_priv, true) I don't get why this idle callback is here. Why aren't you just calling that directly from the suspend handler? > + if (!ret) > + drm_dbg(&dev_priv->drm, "VRAM Self Refresh enabled\n"); > + > + enable_rpm_wakeref_asserts(&dev_priv->runtime_pm); > + > +out: > + pm_runtime_mark_last_busy(kdev); > + pm_runtime_autosuspend(kdev); > + > + return ret; > +} > + > static int intel_runtime_suspend(struct device *kdev) > { > struct drm_i915_private *dev_priv = kdev_to_i915(kdev); > @@ -1726,6 +1751,7 @@ const struct dev_pm_ops i915_pm_ops = { > .restore = i915_pm_restore, > > /* S0ix (via runtime suspend) event handlers */ > + .runtime_idle = intel_runtime_idle, > .runtime_suspend = intel_runtime_suspend, > .runtime_resume = intel_runtime_resume, > }; > diff --git a/drivers/gpu/drm/i915/intel_runtime_pm.c b/drivers/gpu/drm/i915/intel_runtime_pm.c > index 6ed5786bcd29..4dade7e8a795 100644 > --- a/drivers/gpu/drm/i915/intel_runtime_pm.c > +++ b/drivers/gpu/drm/i915/intel_runtime_pm.c > @@ -492,8 +492,7 @@ static void __intel_runtime_pm_put(struct intel_runtime_pm *rpm, > > intel_runtime_pm_release(rpm, wakelock); > > - pm_runtime_mark_last_busy(kdev); > - pm_runtime_put_autosuspend(kdev); > + pm_runtime_put(kdev); > } > > /** > -- > 2.26.2
> -----Original Message----- > From: Ville Syrjälä <ville.syrjala@linux.intel.com> > Sent: Wednesday, May 18, 2022 7:46 PM > To: Gupta, Anshuman <anshuman.gupta@intel.com> > Cc: intel-gfx@lists.freedesktop.org; Nikula, Jani <jani.nikula@intel.com>; > Wilson, Chris P <chris.p.wilson@intel.com>; Vivi, Rodrigo > <rodrigo.vivi@intel.com> > Subject: Re: [Intel-gfx] [PATCH 7/7] drm/i915/rpm: Enable D3Cold VRAM SR > Support > > On Wed, May 18, 2022 at 06:37:16PM +0530, Anshuman Gupta wrote: > > Intel Client DGFX card supports D3Cold with two option. > > D3Cold-off zero watt, D3Cold-VRAM Self Refresh. > > > > i915 requires to evict the lmem objects to smem in order to support > > D3Cold-Off, which increases i915 the suspend/resume latency. Enabling > > VRAM Self Refresh feature optimize the latency with additional power > > cost which required to retain the lmem. > > > > Adding intel_runtime_idle (runtime_idle callback) to enable VRAM_SR, > > it will be used for policy to choose between D3Cold-off vs > > D3Cold-VRAM_SR. > > > > Since we have introduced i915 runtime_idle callback. > > It need to be warranted that Runtime PM Core invokes runtime_idle > > callback when runtime usages count becomes zero. That requires to use > > pm_runtime_put instead of pm_runtime_put_autosuspend. > > > > Cc: Rodrigo Vivi <rodrigo.vivi@intel.com> > > Cc: Chris Wilson <chris.p.wilson@intel.com> > > Signed-off-by: Anshuman Gupta <anshuman.gupta@intel.com> > > --- > > drivers/gpu/drm/i915/i915_driver.c | 26 +++++++++++++++++++++++++ > > drivers/gpu/drm/i915/intel_runtime_pm.c | 3 +-- > > 2 files changed, 27 insertions(+), 2 deletions(-) > > > > diff --git a/drivers/gpu/drm/i915/i915_driver.c > > b/drivers/gpu/drm/i915/i915_driver.c > > index 5a9d5529fc90..bbb11c632799 100644 > > --- a/drivers/gpu/drm/i915/i915_driver.c > > +++ b/drivers/gpu/drm/i915/i915_driver.c > > @@ -1541,6 +1541,31 @@ static int i915_pm_restore(struct device *kdev) > > return i915_pm_resume(kdev); > > } > > > > +static int intel_runtime_idle(struct device *kdev) { > > + struct drm_i915_private *dev_priv = kdev_to_i915(kdev); > > + int ret = 1; > > + > > + if (!HAS_LMEM_SR(dev_priv)) { > > + /*TODO: Prepare for D3Cold-Off */ > > + goto out; > > + } > > + > > + disable_rpm_wakeref_asserts(&dev_priv->runtime_pm); > > + > > + ret = intel_pm_vram_sr(dev_priv, true) > > I don't get why this idle callback is here. Why aren't you just calling that directly > from the suspend handler? We will be having a D3Cold policy in future to decide between D3Cold-VRAM_SR and D3Cold-Off based upon lmem usages also we need to evict the lmem content if D3cold-off option is used. It will be better to keep it in separate runtime idle call back to prepare the actual suspend. Thanks, Anshuman Gupta. > > > + if (!ret) > > + drm_dbg(&dev_priv->drm, "VRAM Self Refresh enabled\n"); > > + > > + enable_rpm_wakeref_asserts(&dev_priv->runtime_pm); > > + > > +out: > > + pm_runtime_mark_last_busy(kdev); > > + pm_runtime_autosuspend(kdev); > > + > > + return ret; > > +} > > + > > static int intel_runtime_suspend(struct device *kdev) { > > struct drm_i915_private *dev_priv = kdev_to_i915(kdev); @@ -1726,6 > > +1751,7 @@ const struct dev_pm_ops i915_pm_ops = { > > .restore = i915_pm_restore, > > > > /* S0ix (via runtime suspend) event handlers */ > > + .runtime_idle = intel_runtime_idle, > > .runtime_suspend = intel_runtime_suspend, > > .runtime_resume = intel_runtime_resume, }; diff --git > > a/drivers/gpu/drm/i915/intel_runtime_pm.c > > b/drivers/gpu/drm/i915/intel_runtime_pm.c > > index 6ed5786bcd29..4dade7e8a795 100644 > > --- a/drivers/gpu/drm/i915/intel_runtime_pm.c > > +++ b/drivers/gpu/drm/i915/intel_runtime_pm.c > > @@ -492,8 +492,7 @@ static void __intel_runtime_pm_put(struct > > intel_runtime_pm *rpm, > > > > intel_runtime_pm_release(rpm, wakelock); > > > > - pm_runtime_mark_last_busy(kdev); > > - pm_runtime_put_autosuspend(kdev); > > + pm_runtime_put(kdev); > > } > > > > /** > > -- > > 2.26.2 > > -- > Ville Syrjälä > Intel
diff --git a/drivers/gpu/drm/i915/i915_driver.c b/drivers/gpu/drm/i915/i915_driver.c index 5a9d5529fc90..bbb11c632799 100644 --- a/drivers/gpu/drm/i915/i915_driver.c +++ b/drivers/gpu/drm/i915/i915_driver.c @@ -1541,6 +1541,31 @@ static int i915_pm_restore(struct device *kdev) return i915_pm_resume(kdev); } +static int intel_runtime_idle(struct device *kdev) +{ + struct drm_i915_private *dev_priv = kdev_to_i915(kdev); + int ret = 1; + + if (!HAS_LMEM_SR(dev_priv)) { + /*TODO: Prepare for D3Cold-Off */ + goto out; + } + + disable_rpm_wakeref_asserts(&dev_priv->runtime_pm); + + ret = intel_pm_vram_sr(dev_priv, true); + if (!ret) + drm_dbg(&dev_priv->drm, "VRAM Self Refresh enabled\n"); + + enable_rpm_wakeref_asserts(&dev_priv->runtime_pm); + +out: + pm_runtime_mark_last_busy(kdev); + pm_runtime_autosuspend(kdev); + + return ret; +} + static int intel_runtime_suspend(struct device *kdev) { struct drm_i915_private *dev_priv = kdev_to_i915(kdev); @@ -1726,6 +1751,7 @@ const struct dev_pm_ops i915_pm_ops = { .restore = i915_pm_restore, /* S0ix (via runtime suspend) event handlers */ + .runtime_idle = intel_runtime_idle, .runtime_suspend = intel_runtime_suspend, .runtime_resume = intel_runtime_resume, }; diff --git a/drivers/gpu/drm/i915/intel_runtime_pm.c b/drivers/gpu/drm/i915/intel_runtime_pm.c index 6ed5786bcd29..4dade7e8a795 100644 --- a/drivers/gpu/drm/i915/intel_runtime_pm.c +++ b/drivers/gpu/drm/i915/intel_runtime_pm.c @@ -492,8 +492,7 @@ static void __intel_runtime_pm_put(struct intel_runtime_pm *rpm, intel_runtime_pm_release(rpm, wakelock); - pm_runtime_mark_last_busy(kdev); - pm_runtime_put_autosuspend(kdev); + pm_runtime_put(kdev); } /**
Intel Client DGFX card supports D3Cold with two option. D3Cold-off zero watt, D3Cold-VRAM Self Refresh. i915 requires to evict the lmem objects to smem in order to support D3Cold-Off, which increases i915 the suspend/resume latency. Enabling VRAM Self Refresh feature optimize the latency with additional power cost which required to retain the lmem. Adding intel_runtime_idle (runtime_idle callback) to enable VRAM_SR, it will be used for policy to choose between D3Cold-off vs D3Cold-VRAM_SR. Since we have introduced i915 runtime_idle callback. It need to be warranted that Runtime PM Core invokes runtime_idle callback when runtime usages count becomes zero. That requires to use pm_runtime_put instead of pm_runtime_put_autosuspend. Cc: Rodrigo Vivi <rodrigo.vivi@intel.com> Cc: Chris Wilson <chris.p.wilson@intel.com> Signed-off-by: Anshuman Gupta <anshuman.gupta@intel.com> --- drivers/gpu/drm/i915/i915_driver.c | 26 +++++++++++++++++++++++++ drivers/gpu/drm/i915/intel_runtime_pm.c | 3 +-- 2 files changed, 27 insertions(+), 2 deletions(-)