| Message ID | 1449669373-8588-1-git-send-email-joonas.lahtinen@linux.intel.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
On ke, 2015-12-09 at 15:56 +0200, Joonas Lahtinen wrote: > In order to avoid accessing GPU registers while GPU is suspended > cancel > the hangcheck work before calling intel_suspend_complete which > actually > puts the GPU to suspend. Otherwise hangcheck might do MMIO reads to a > suspended GPU. > > Placement before intel_guc_suspend is imitated from i915_drm_suspend > which cancels the work at i915_gem_suspend, to keep the functions > similar. > > On VLV systems, namely BYT, this was causing an error during runtime > suspend cycle: > > [drm:vlv_check_no_gt_access [i915]] *ERROR* GT register access while > GT waking disabled > > Testcase: igt/pm_rpm/basic-rte > Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=93121 > > Signed-off-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com> > Cc: Imre Deak <imre.deak@intel.com> > Cc: Chris Wilson <chris@chris-wilson.co.uk> Yep, looks good to me: Reviewed-by: Imre Deak <imre.deak@intel.com> > --- > drivers/gpu/drm/i915/i915_drv.c | 3 ++- > 1 file changed, 2 insertions(+), 1 deletion(-) > > diff --git a/drivers/gpu/drm/i915/i915_drv.c > b/drivers/gpu/drm/i915/i915_drv.c > index e6935f1..81862d5 100644 > --- a/drivers/gpu/drm/i915/i915_drv.c > +++ b/drivers/gpu/drm/i915/i915_drv.c > @@ -1462,6 +1462,8 @@ static int intel_runtime_suspend(struct device > *device) > i915_gem_release_all_mmaps(dev_priv); > mutex_unlock(&dev->struct_mutex); > > + cancel_delayed_work_sync(&dev_priv- > >gpu_error.hangcheck_work); > + > intel_guc_suspend(dev); > > intel_suspend_gt_powersave(dev); > @@ -1475,7 +1477,6 @@ static int intel_runtime_suspend(struct device > *device) > return ret; > } > > - cancel_delayed_work_sync(&dev_priv- > >gpu_error.hangcheck_work); > intel_uncore_forcewake_reset(dev, false); > dev_priv->pm.suspended = true; >
On Wed, Dec 09, 2015 at 04:10:03PM +0200, Imre Deak wrote: > On ke, 2015-12-09 at 15:56 +0200, Joonas Lahtinen wrote: > > In order to avoid accessing GPU registers while GPU is suspended > > cancel > > the hangcheck work before calling intel_suspend_complete which > > actually > > puts the GPU to suspend. Otherwise hangcheck might do MMIO reads to a > > suspended GPU. > > > > Placement before intel_guc_suspend is imitated from i915_drm_suspend > > which cancels the work at i915_gem_suspend, to keep the functions > > similar. > > > > On VLV systems, namely BYT, this was causing an error during runtime > > suspend cycle: > > > > [drm:vlv_check_no_gt_access [i915]] *ERROR* GT register access while > > GT waking disabled > > > > Testcase: igt/pm_rpm/basic-rte > > Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=93121 > > > > Signed-off-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com> > > Cc: Imre Deak <imre.deak@intel.com> > > Cc: Chris Wilson <chris@chris-wilson.co.uk> > > Yep, looks good to me: > Reviewed-by: Imre Deak <imre.deak@intel.com> Well Imre also shotdown the simple tryget plan, so the only thing I can say is that I'd rather move the cancel_sync(hangcheck) out of the intel_runtime_suspend() into the mark_idle to match it's counterpart in mark_busy. Patch in the post, so for now Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk> -Chris
On Wed, 2015-12-09 at 16:19 +0000, Chris Wilson wrote: > On Wed, Dec 09, 2015 at 04:10:03PM +0200, Imre Deak wrote: > > On ke, 2015-12-09 at 15:56 +0200, Joonas Lahtinen wrote: > > > In order to avoid accessing GPU registers while GPU is suspended > > > cancel > > > the hangcheck work before calling intel_suspend_complete which > > > actually > > > puts the GPU to suspend. Otherwise hangcheck might do MMIO reads > > > to a > > > suspended GPU. > > > > > > Placement before intel_guc_suspend is imitated from > > > i915_drm_suspend > > > which cancels the work at i915_gem_suspend, to keep the functions > > > similar. > > > > > > On VLV systems, namely BYT, this was causing an error during > > > runtime > > > suspend cycle: > > > > > > [drm:vlv_check_no_gt_access [i915]] *ERROR* GT register access > > > while > > > GT waking disabled > > > > > > Testcase: igt/pm_rpm/basic-rte > > > Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=93121 > > > > > > Signed-off-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com> > > > Cc: Imre Deak <imre.deak@intel.com> > > > Cc: Chris Wilson <chris@chris-wilson.co.uk> > > > > Yep, looks good to me: > > Reviewed-by: Imre Deak <imre.deak@intel.com> > > Well Imre also shotdown the simple tryget plan, so the only thing I > can > say is that I'd rather move the cancel_sync(hangcheck) out of the > intel_runtime_suspend() into the mark_idle to match it's counterpart > in > mark_busy. > > Patch in the post, so for now > Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk> Pushed to dinq, thanks for the patch and review. > -Chris >
diff --git a/drivers/gpu/drm/i915/i915_drv.c b/drivers/gpu/drm/i915/i915_drv.c index e6935f1..81862d5 100644 --- a/drivers/gpu/drm/i915/i915_drv.c +++ b/drivers/gpu/drm/i915/i915_drv.c @@ -1462,6 +1462,8 @@ static int intel_runtime_suspend(struct device *device) i915_gem_release_all_mmaps(dev_priv); mutex_unlock(&dev->struct_mutex); + cancel_delayed_work_sync(&dev_priv->gpu_error.hangcheck_work); + intel_guc_suspend(dev); intel_suspend_gt_powersave(dev); @@ -1475,7 +1477,6 @@ static int intel_runtime_suspend(struct device *device) return ret; } - cancel_delayed_work_sync(&dev_priv->gpu_error.hangcheck_work); intel_uncore_forcewake_reset(dev, false); dev_priv->pm.suspended = true;
In order to avoid accessing GPU registers while GPU is suspended cancel the hangcheck work before calling intel_suspend_complete which actually puts the GPU to suspend. Otherwise hangcheck might do MMIO reads to a suspended GPU. Placement before intel_guc_suspend is imitated from i915_drm_suspend which cancels the work at i915_gem_suspend, to keep the functions similar. On VLV systems, namely BYT, this was causing an error during runtime suspend cycle: [drm:vlv_check_no_gt_access [i915]] *ERROR* GT register access while GT waking disabled Testcase: igt/pm_rpm/basic-rte Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=93121 Signed-off-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com> Cc: Imre Deak <imre.deak@intel.com> Cc: Chris Wilson <chris@chris-wilson.co.uk> --- drivers/gpu/drm/i915/i915_drv.c | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-)