Re: Suspend/resume regression between 2.6.26 and 2.6.27-rc1

From: Alan Jenkins
Date: Thu Oct 16 2008 - 05:40:13 EST


Zhao Yakui wrote:
> On Fri, 2008-10-10 at 14:41 +0200, Rafael J. Wysocki wrote:
>> On Friday, 10 of October 2008, Zhang Rui wrote:
>>> Hi, len,
>>>
>>> this is the ACPI regression test result based on the latest ACPI test branch.
>>>
>>> 1. on ïAcer:(AMD CPU, VIA chipset. 64 bit kernel)
>>> ïWhen doing S3 test, after pressing the power button, the system
>>> reboots instead of resuming. But if S3 is done after S4, the system can
>>> resume very well after pressing power button or using the RTC
>>> alarm.
>>> Note that this is an upstream regression as it can be reproduced on
>>> linus' tree.
>>> Yakui is investigating this issue.
>> We had some reports of the second suspend (S3) failure too, where the second
>> attempt to suspend to RAM (or to resume from it) failed after a successful
>> one. I wonder if that's related.
> Some suspend/resume tests are done on one Acer laptop(AMD CPU, VIA
> chipset, 64-bit kernel).
> The system will be rebooted when pressing power button after the box
> enters S3 state. But if S3 is done after doing S4, the system can be
> resumed very well after pressing power button.This issue can be
> reproduced on the upstream kernel.
>
> After the further test we can confirm that this is a regression. The
> 2.6.26 kernel can work well on this box. But the 2.6.27-rc1 will fail.
>
> After using the git-bisect it is confirmed that the commit
> 736f12bff9d9e7b4e895c64f73b190c8383fc2a1 is good.
> >commit 736f12bff9d9e7b4e895c64f73b190c8383fc2a1
> >Author: Glauber Costa <gcosta@xxxxxxxxxx>
> > Date: Tue May 27 20:14:51 2008 -0700
> >x86: don't use gdt_page openly.
>
> And the commit
> 55f262391a2365d657a00ed68edd1a51bca66af5 is bad.
> >commit 55f262391a2365d657a00ed68edd1a51bca66af5
> >Author: Yinghai Lu <yhlu.kernel@xxxxxxxxx>
> >Date: Wed Jun 25 17:54:23 2008 -0700
> >x86: rename setup_32.c to setup.c
>
> The patches between the above two commits are related with X86. When
> using git-bisect between the above two commits, we will get the
> compiling errors(For example: some files don't exist) or the kernel
> panic. So we can't continue using git-bisect to identify which commit
> the regression is caused by.

There's a known fix for the kernel panic. It's referenced at <http://bugzilla.kernel.org/show_bug.cgi?id=11237#c25>. That should help you bisect down to a smaller range. Hopefully you can rule out the commit that caused^Wexposed Bug #11237, which is really a nasty BIOS bug.

HTH
Alan
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/