Re: [RFC][PATCH] PM / Sleep: Freeze filesystems during system suspend/hibernation

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

On 01/30/2012 01:23 AM, Rafael J. Wysocki wrote:

> On Sunday, January 29, 2012, Srivatsa S. Bhat wrote:

[...]

>>> Index: linux/kernel/power/suspend.c
>>> ===================================================================
>>> --- linux.orig/kernel/power/suspend.c
>>> +++ linux/kernel/power/suspend.c
>>> @@ -29,6 +29,48 @@
>>>
>>>  #include "power.h"
>>>
>>> +#ifdef CONFIG_SUSPEND_FREEZER
>>> +
>>> +static inline int suspend_freeze_processes(void)
>>> +{
>>> +	int error;
>>> +
>>> +	error = freeze_processes();
>>> +	if (error)
>>> +		return error;
>>> +
>>> +	error = freeze_supers();
>>> +	if (error) {
>>> +		thaw_processes();
>>> +		return error;
>>> +	}
>>> +
>>> +	error = freeze_kernel_threads();
>>> +	if (error)
>>> +		thaw_supers();
>>> +
>>
>>
>> If freezing of kernel threads fails, freeze_kernel_threads() calls
>> thaw_processes(), which means, even userspace processes get thawed.
>> So, there would be a time-window in which userspace is thawed but the
>> filesystems are still frozen. That is not very desirable right?
> 
> No, it is not.  I overlooked that, thanks!
> 
>> If that is right, then modifying freeze_kernel_threads() to call
>> thaw_kernel_threads() instead of thaw_processes() would fix it
>> (and of course, we would need to explicitly call thaw_processes
>> above).
>>
>> BTW, after your patch posted at https://lkml.org/lkml/2012/1/27/501,
>> I very much wanted to write a patch to convert the semantics of
>> freeze/thaw to something like:
>>
>> freeze_processes() calls thaw_processes on error.
>> //Both touch only userspace processes.
>>
>> freeze_kernel_threads() calls thaw_kernel_threads() on error.
>> //Both touch only kernel threads.
>>
>> Of course, such a patch would need to do a lot of fixing up at several
>> places, but IMHO, it would really help make the overall code more logical
>> and easier to understand.
>>
>> I can write it up and post it soon, but then you'll have to rebase
>> your patch (this one) on top of that. What do you say?
> 
> Please do that, it wouldn't be any problem for me to rebase the $subject
> patch.
> 


Hi Rafael,
I have posted the patchset (v2) at https://lkml.org/lkml/2012/1/30/530

 

Regards,
Srivatsa S. Bhat