Re: Lenovo ThinkPad X13s regerssions (was Re: Linux regressions report for mainline [2024-02-25])

[Johan Hovold <johan@kernel.org>]

On Tue, Mar 05, 2024 at 08:19:47PM -0800, Bjorn Andersson wrote:
> On Tue, Mar 05, 2024 at 02:51:44PM +0100, Johan Hovold wrote:
> > On Tue, Mar 05, 2024 at 10:33:39AM +0100, Linux regression tracking (Thorsten Leemhuis) wrote:

> > > Any decision yet if they are going to be reverted for now?
> > >
> > > Am I right assuming those would fix
> > > https://lore.kernel.org/lkml/Zd3kvD02Qvsh2Sid@hovoldconsulting.com/
> > > which did not get even a single reply?
> > 
> > That was the hope, but I've managed to trigger a reset on disconnect
> > once also with the runtime PM series reverted.

I have not been able to reproduce the reset with the series reverted,
and after reviewing the code in question it seems unlikely that I ever
did so.

> > One of the patches from that series has already been reverted (to fix
> > the VT console hotplug regression) and there is some indication that
> > that was sufficient to address the issue with hotplug not being detected
> > in X/Wayland too. I'm waiting for confirmation from some users that have
> > not been able to use their external displays at all since 6.8-rc1, but
> > it does seem to fix the X/Wayland issues I could reproduce here.

> I bumped my X13s to v6.8-rc7 earlier today and took it for a spin.
> 
> I was successfully able to plug/unplug my main display both in fbcon and
> Wayland (sway) a number of times, I was able to boot with external
> display connected and have it show up in fbcon and then survive into
> sway. I tried suspending (echo mem > /sys/power/state) and got back from
> that state a few times without problems.
> 
> Mixing connection/disconnection with being in suspended state was less
> successful and I was able to crash the machine twice here - but I can't
> say this worked before... (As previously we would not have eDP after
> suspending with external display).
> 
> So, things are looking much better with -rc7, but of course, my test
> scope is limited.

Thanks for confirming. The revert in rc7 seems to help with the hotplug
detect issues I could reproduce too. And for some reason, I can no
longer seem to reproduce the reset either, possibly due to unrelated
changes in timing (e.g. as I don't see it after reapplying the reverted
patch either). 

I just spent some more time on this driver and sent a follow-up report
here:

	https://lore.kernel.org/lkml/ZesH21DcfOldRD9g@hovoldconsulting.com/

It seems quite clear to me that the reset-on-disconnect regression have
been introduced by the runtime PM series and I don't currently see how
the hotplug notification revert in rc7 could have fixed it.

Johan