Re: 2.6.24 regression: hibernation hangs on "Suspending console" in low-battery condition

[Andrey Borzenkov <arvidjaar@mail.ru>]

[-- Attachment #1: Type: text/plain, Size: 2916 bytes --]

On Sunday 11 November 2007, Rafael J. Wysocki wrote:
> On Sunday, 11 of November 2007, Andrey Borzenkov wrote:
> > On Sunday 11 November 2007, Rafael J. Wysocki wrote:
> > > On Sunday, 11 of November 2007, Andrey Borzenkov wrote:
> > > > On Monday 05 November 2007, Andrey Borzenkov wrote:
> > > > > Notice "hung" not "hangs". This happened so far only once - when low battery 
> > > > > condition triggered suspend to disk. I was not able to reproduce it after 
> > > > > this running on AC.
> > > > > 
> > > > > Just in case it rings the bell for someone. This is not suspend regression 
> > > > > reported earlier by Jens - I do not even have SATA nor is using libata 
> > > > > drivers.
> > > > > 
> > > > 
> > > > This happened second time, now with 2.6.24-rc2. I have kpowersave configured
> > > > to initiate suspend-to-disk on low battery. Suspend-to-disk is initiated,
> > > > but system stops after Suspending console line. Suspend image is not created.
> > > > 
> > > > This did work before although I do not name exact version.
> > > > 
> > > > I am rather at loss how to proceed further ... any idea? dmesg and config
> > > > attached just in case. The problem is that hibernation works when I initiate
> > > > it manually ...
> > > 
> > > Have you tried it with the battery removed?
> > > 
> > > 
> > 
> > It works. Subjectively I have relatively long pause after first Suspending console
> > message (where it hangs otherwise), according to dmesg timestamp it is about
> > 1 second before next messages appear. Also last two times I tried it writeout
> > of suspend image was really slow; it was both with and without battery. Reading
> > on resume in really in order of magnitude faster.
> > 
> > But manual STD so far works both on AC (with or without battery) or on battery.
> > 
> > Of course it is possible that battery gets exhausted during STD, but system
> > is not switched off. Rather I would suspect something like ACPI notification
> > that comes during freeze ...
> 
> Frankly, I'm suspecting a deadlock in the ACPI code.
> 
> Please try to make kpowersave start hibernation a bit earlier and see what
> happens.
> 
> What's your critical time now?
> 
> 

it was set to 5%. Setting it to 10% makes it work, but it just confirms what I
said before - suspend-to-RAM in normal battery conditions (or without battery)
works. With 5% threshold I have visible indication of alarm from *firmware* -
battery LED flashing.

Something happens differently when battery is in dangerous state. I tried patch
4c41d3ad6544f1c9aec37c441af04f5d0ad3a731 (got from ACPI tree) but it did not
change anything. It worked once when I left system on battery with artificial
load (find . -exec md5sum in linux tree in a loop) but I suspect it is just
because it just triggered limit earlier. (Frankly I do not know which percentage
kpowersave computes)

[-- Attachment #2: This is a digitally signed message part. --]
[-- Type: application/pgp-signature, Size: 189 bytes --]