On Sat, Sep 18, 2021 at 2:18 AM Michael Stapelberg <michael@xxxxxxxxxxxxx> wrote: > > torvalds at linux-foundation.org (Linus Torvalds) writes: > > Did I fix it up correctly? Who knows. The code makes more sense to me > > now and seems valid. But I really *really* want to stress how locking > > is important. > > As far as I can tell, this merge conflict resolution made my Raspberry > Pi 3 hang on boot. Ok, that's a different merge issue than the locking one (which is about the amd ttm code). But the VC4 driver did have changes close to each other in the hdmi detection and clock setting code. And it doesn't seem to be just RPi3, there was a report back a couple of weeks ago about RPi4 also having regressed (with an Ubuntu install). That one was an oops in vc4_hdmi_audio_prepare(). I don't know if that got resolved, I heard nothing about it after the report. So there's something seriously wrong in VC4 space. The main issue seems to be the runtime power management changes. As far as I can tell, the commits that didn't come in through that drm pull were these two 9984d6664ce9 ("drm/vc4: hdmi: Make sure the controller is powered in detect") 411efa18e4b0 ("drm/vc4: hdmi: Move the HSM clock enable to runtime_pm") Michael - do things work if you revert those two (sadly, they don't revert cleanly exactly _because_ of the other changes in the same area)? Maxime? Linus