Re: [BUG] 3.7-rc regression bisected: s2disk fails to resume image: Processes could not be frozen, cannot continue resuming

[Al Viro <viro@ZenIV.linux.org.uk>]

On Wed, Sep 18, 2013 at 02:52:39PM +0100, Al Viro wrote:
> On Tue, Aug 27, 2013 at 07:48:43AM +0400, Andrew Savchenko wrote:
> > > Additional (but probably useless) information on this bug may be found
> > > here: https://forums.gentoo.org/viewtopic-p-7371120.html
> 
> Something's very fishy there:
> 
> [quote]
> Digging into suspend-utils code shows that the following ioctl fails on      
> "/dev/snapshot":                                                             
>                                                                                 
>                                     Code:                                       
>                           ioctl(dev, _IO(3, 1), 0);                          
> [end quote]
> 
> but that's _not_ anything freeze-related - that's HDIO_GETGEO, and with zero
> as last argument it will fail, no matter what.  With EFAULT, if nothing
> else...
> 
> Which ioctl() it really is?  A bit further down you write "I modified suspend
> code to see errno, so freeze on /dev/snapshot fails [with EAGAIN]", so you
> have isolated the call in question.  Could you quote the actual code?

*scratches head*  _IO('3', 1), perhaps?  At least that would make sense in
such context...  Assuming that's the case, slap
	printk(KERN_INFO "freeze_process() => %d", error);
after the call of freeze_process() in kernel/power/user.c along with
	printk(KERN_INFO "__usermodehelper_disable() => %d", error);
and
	printk(KERN_INFO "try_to_freeze_tasks() => %d", error);
in kernel/power/process.c:freeze_process(), after the calls of
__usermodehelper_disable() and try_to_freeze_tasks() resp.

FWIW, I suspect that it's __usermodehelper_disable() - it does
        retval = wait_event_timeout(running_helpers_waitq,
                                        atomic_read(&running_helpers) == 0,
                                        RUNNING_HELPERS_TIMEOUT);
and returns -EAGAIN on timeout.  I'm not familiar with swsusp code, but
it smells like we end up waiting for linuxrc itself to finish.

Pavel, any suggestions?  If SNAPSHOT_FREEZE really wants everything run
via usermodehelper gone for some reason, what makes /linuxrc different
from e.g /sbin/modprobe?