Hey, On Wed, 2 Oct 2019, Chris Wilson wrote: > > + if (INTEL_GEN(dev_priv) == 11 || INTEL_GEN(dev_priv) == 10) > > + I915_WRITE(AUD_PIN_BUF_CTL, > > + (I915_READ(AUD_PIN_BUF_CTL) | > > + AUD_PIN_BUF_ENABLE)); > > From the observation that this reduces the module reload time from 196s > to 2s, I'd say this works. yes indeed, the CI results look very promising. The problem does not occur every time, so this has been bugging us for a long time, but in scenarios and systems where this happens, the patch seems to help. Upon further investigation, I believe this needs to be applied on GLK as well. This would explain some of the other sightings. E.g. here's one from from audio/SOF CI from this week: https://sof-ci.01.org/sofpr/PR1864/build3366/devicetest/GLK_BOB_DA7219/verify-pcm-list.sh/dmesg.txt I'll update both patches to include GLK. > Do you need to disable the bit later? Now that is trickier. I'm still digging more info about this. Based on info I have, to avoid unsolicited wakes from i915 audio codec, we should clear the bit after we have turned off i915 audio power domain. In practise onsystems I have for testing, this results in frequent probe failures, so I left this out from the patch. I will try to search for more info on this (especially if we get still some failures with the patch). But enabling it every time does seem to help and so far has not caused any regressions elsewhere, so at least this part seems good. Br, Kai _______________________________________________ Intel-gfx mailing list Intel-gfx@xxxxxxxxxxxxxxxxxxxxx https://lists.freedesktop.org/mailman/listinfo/intel-gfx
