Re: [PATCH 0/8] Cpufreq, cpu hotplug, suspend/resume related fixes

["Srivatsa S. Bhat" <srivatsa.bhat@linux.vnet.ibm.com>]

On 07/17/2013 03:02 AM, Rafael J. Wysocki wrote:
> On Tuesday, July 16, 2013 05:15:14 PM Toralf Förster wrote:
>> On 07/12/2013 12:23 AM, Srivatsa S. Bhat wrote:
>>> On 07/12/2013 04:03 AM, Rafael J. Wysocki wrote:
>>>> On Friday, July 12, 2013 03:45:17 AM Srivatsa S. Bhat wrote:
>>>>>
>>>>> Hi,
>>>>
>>>> Hi,
>>>>
>>>>> Commit a66b2e (cpufreq: Preserve sysfs files across suspend/resume) caused
>>>>> some subtle regressions in the cpufreq subsystem during suspend/resume.
>>>>> This patchset is aimed at rectifying those problems, by fixing the regression
>>>>> as well as achieving the original goal of that commit in a proper way.
>>>>>
>>>>> Patch 1 reverts the above commit, and is CC'ed to stable.
>>>>>
>>>>> Patches 2 - 5 reorganize the code and have no functional impact, and can go
>>>>> in as general cleanups as well. This reorganization builds a base that the
>>>>> rest of the patches will make use of.
>>>>>
>>>>> Patch 6 and 7 add a mechanism to perform light-weight init/tear-down of CPUs
>>>>> in the cpufreq subsystem and finally patch 8 uses it to preserve sysfs files
>>>>> across suspend/resume.
>>>>>
>>>>> All the patches apply on current mainline.
>>>>>
>>>>>
>>>>> Robert, Durgadoss, it would be great if you could try it out and see if it works
>>>>> well for your usecase. I tested it locally and cpufreq related files did retain
>>>>> their permissions across suspend/resume. Let me know if it works fine in your
>>>>> setup too.
>>>>>
>>>>> And I'd of course appreciate to hear from Dirk, Tianyu and Toralf to know
>>>>> whether their systems work fine after:
>>>>> a. applying only the first commit (this is what gets backported to stable)
>>>>> b. applying all the commits
>>>>>
>>>>> (Note: I had to use Michael's fix[1] to avoid CPU hotplug deadlock while
>>>>> testing this patchset. Though that patch also touches cpufreq subsystem, it
>>>>> doesn't affect this patchset in any way and there is absolutely no dependency
>>>>> between the two in terms of code. That fix just makes basic CPU hotplug work
>>>>> without locking up on current mainline).
>>>>>
>>>>> [1]. https://lkml.org/lkml/2013/7/10/611
>>>>>
>>>>>
>>>>> Thank you very much!
>>>>
>>>> Thanks Srivatsa!
>>>>
>>>> I'm going to take [1/8] for 3.11 and queue up the rest for 3.12 if people don't
>>>> hate them.  This way we'll have some more testing coverage before they reach
>>>> the mainline hopefully.
>>>>
>>
>> On 07/16/2013 01:25 AM, Rafael J. Wysocki wrote:> On Monday, July 15, 2013 07:38:02 PM Toralf Förster wrote:
>>> Sorry, I have no idea what 1#8 means.
>>
>> sry - here again with full quote of the email :
>>
>> I applied patch [1/8] on top of v3.11-rc1-8-g47188d3 passes two s2ram/wakeup
>> cycles fine and crashed the system at the 3rd attempt / one times just at
>> the 4th (blinking power led, no sysrq, ...).
>>
>> Applying patch 1-8 on top of that tree differs in that way that it
>> crashes now the system even at the 1st attempt or at least at the 2nd
>>
>> My hardware is a ThinkPad T420 with latest BIOS and a 32 bit stable
>> Gentoo Linux - FWIW .config attached.
> 
> I think you'll need the fixes first, basically [1/8] from this series and
> this: https://patchwork.kernel.org/patch/2827512/ .
> 
> Please try to run with these two things applied only and see how that goes.
> 

In addition to what Rafael suggested above, also try running your kernel
with cpufreq completely turned off (CONFIG_CPU_FREQ=n). My patches touch only
cpufreq, so this experiment will tell us if your suspend/resume issues are
really related to cpufreq or not. If turning off cpufreq also breaks your
suspend/resume, then a fresh git-bisect might be the only way to go.
 
Regards,
Srivatsa S. Bhat