Re: 2.6.35 suspend regression (bisected)

[Roberto Oppedisano <roberto.oppedisano@gmail.com>]

  On 28/07/2010 00:37, Rafael J. Wysocki wrote:
> On Tuesday, July 27, 2010, Roberto Oppedisano wrote:
>>    On 27/07/2010 21:03, Rafael J. Wysocki wrote:
>>>> I did bisect again and I did get a different result:
>>>>
>>>> c1518f12bab97a6d409a25aaccb02dc8895800f3 is the first bad commit
>>>>
>>>> Trying to revert that commit gives:
>>>>
>>>> $ git bisect reset
>>>> $ git revert c1518f12bab97a6d409a25aaccb02dc8895800f3
>>>> fatal: Commit c1518f12bab97a6d409a25aaccb02dc8895800f3 is a merge but no
>>>> -m option was given.
>>>>
>>>> How am I supposed to proceed now ? (sorry I'm a git newbie)
>>> The bisect turned up a merge, which means it probably didn't really converge
>>> for you.
>>>
>>> Do you use a camera driver?
>> Yes, this laptop has an integrated webcam.
>>
>> Bus 001 Device 004: ID 04f2:b059 Chicony Electronics Co., Ltd
>>
>>
>> which i think is driven by uvcvideo.ko. I've tryed to rmmod uvcvideo,
>> videodev and
>>
>> v4l1_compat but on current git suspend still hangs in the same way
>> (screen black,
>> backlight on, fans at full speed, no response to ping and... nothing in
>> the logs).
> Please try the current mainline with acpi_sleep=nonvs in the command line.

With:

	acpi_sleep=nonvs console=tty0

(kubuntu defaults to quiet spash) I was able to complete 3 suspend/resume cycles before hitting an hang. After a power cycle I was able to suspend again, so it seems that, albeit racy, this cmdline has a positive effect.

With:
	acpi_sleep=nonvs

or

	console=tty0

alone I've never been able to suspend (only the combination of the two makes a difference).

R