drm/i915: fix RPS on runtime suspend

From: Paulo Zanoni <[email protected]>

With this patch, the RPS sequence for runtime suspend/resume is exactly like the sequence for S3 suspend/resume:
- flush_delayed_work(&dev_priv->rps.delayed_resume_work)
- intel_runtime_pm_disable_interrupts()
- intel_suspend_gt_powersave()
(suspended)
- intel_runtime_pm_enable_interrupts()
- intel_enable_gt_powersave()

With this, we get rid of WARNs that are currently intermittently triggered by the system-suspend-execbuf subtest of runtime PM. Notice that these WARNs could also be triggered in other ways that involved doing lots of RPM suspend/resume cycles just after a system S3 resume.

Testcase: igt/pm_rpm/system-suspend-execbuf

Signed-off-by Paulo Zanoni <[email protected]>

---
 drivers/gpu/drm/i915/i915_drv.c | 10 +++-------
 1 file changed, 3 insertions(+), 7 deletions(-)

Also, with this patch + the sprite planes regression fix, this will be the first time ever that my BDW machine passes every single test of pm_rpm without any WARNs! \o/

drivers/gpu/drm/i915/i915_drv.c +2 -6

index 0c7cf48..2404b2b 100644
@@ -1395,13 +1395,9 @@ static int intel_runtime_suspend(struct device *device)
 	i915_gem_release_all_mmaps(dev_priv);
 	mutex_unlock(&dev->struct_mutex);
 
-	/*
-	 * rps.work can't be rearmed here, since we get here only after making
-	 * sure the GPU is idle and the RPS freq is set to the minimum. See
-	 * intel_mark_idle().
-	 */
-	cancel_work_sync(&dev_priv->rps.work);
+	flush_delayed_work(&dev_priv->rps.delayed_resume_work);
 	intel_runtime_pm_disable_interrupts(dev_priv);
+	intel_suspend_gt_powersave(dev);
 
 	ret = intel_suspend_complete(dev_priv);
 	if (ret) {

Quoted 16 lines · Paulo Zanoni

From: Paulo Zanoni <[email protected]>

With this patch, the RPS sequence for runtime suspend/resume is exactly like the sequence for S3 suspend/resume:
- flush_delayed_work(&dev_priv->rps.delayed_resume_work)
- intel_runtime_pm_disable_interrupts()
- intel_suspend_gt_powersave()
(suspended)
- intel_runtime_pm_enable_interrupts()
- intel_enable_gt_powersave()

With this, we get rid of WARNs that are currently intermittently triggered by the system-suspend-execbuf subtest of runtime PM. Notice that these WARNs could also be triggered in other ways that involved doing lots of RPM suspend/resume cycles just after a system S3 resume.

Quoted 2 lines

Testcase: igt/pm_rpm/system-suspend-execbuf

Signed-off-by Paulo Zanoni <[email protected]>

This also fixes some parts of pm_rpm subtest failures on VLV so you could add:
Reference: https://bugs.freedesktop.org/show_bug.cgi?id=82939
and

Reviewed-by Imre Deak <[email protected]>

While fixing the same WARN in the above bug I also cleaned up the rps code a bit and made the disabling part more robust against rearming the rps work. But that's a separate issue which can't normally happen here, so I can send a version rebased on top of your changes.