lockup on boot -- drm/i915/display: Force the state compute phase once to enable PSR

From: Dominik Brodowski
Date: Tue Feb 18 2020 - 06:48:39 EST


On Mon, Feb 17, 2020 at 10:08:52PM +0100, Boris Brezillon wrote:
> On Mon, 17 Feb 2020 21:09:42 +0100
> Dominik Brodowski <linux@xxxxxxxxxxxxxxxxxxxx> wrote:
>
> > On my old Dell XPS 13 laptop with
> >
> > 00:00.0 Host bridge: Intel Corporation Broadwell-U Host Bridge -OPI (rev 09)
> > 00:02.0 VGA compatible controller: Intel Corporation HD Graphics 5500 (rev 09) (prog-if 00 [VGA controller])
> >
> > booting 5.6-rc1 and -rc2 fails after the dmesg line
> >
> > fb0: switching to inteldrmfb from simple
> >
> > while the next lines should be something like (v5.5):
> >
> > Console: switching to colour dummy device 80x25
> > i915 0000:00:02.0: vgaarb: deactivate vga console
> > [drm] ACPI BIOS requests an excessive sleep of 25000 ms, using 1500 ms instead
> > [drm] Supports vblank timestamp caching Rev 2 (21.10.2013).
> > [drm] Driver supports precise vblank timestamp query.
> > i915 0000:00:02.0: vgaarb: changed VGA decodes: olddecodes=io+mem,decodes=io+mem:owns=io+mem
> >
> > A git bisect lead to
> >
> > commit b86d895524ab ("drm/bridge: Add an ->atomic_check() hook")
>
> This commit has been reverted: you should ignore any failures between
> b86d895524ab ("drm/bridge: Add an ->atomic_check() hook") and
> 099126352303 ("Revert "drm/bridge: Add a drm_bridge_state object").

A new bisect now points to

60c6a14b489b ("drm/i915/display: Force the state compute phase
once to enable PSR").

Please note that bisecting this is quite a hassle, in particular due to
various reverts in between and back-merges (such as ec027b33c8bb, which has
two parents in "bad" state). As 60c6a14b489b does not revert cleanly, I
can't test a revert on top of 5.6-rc2.

Thanks,
Dominik