Re: Fw: Re: [mm PATCH 4/6] RCU: (now) CPU hotplug

[Pavel Machek]

Hi!

> > > This is needed so that the _resume_ works, when it's handled from the user land
> > > by our resume tool.  Currently, the resume code calls
> > > freeze_processes() too.
> > 
> > I do not understand... freeze_processes() always leaves curent process
> > running... why is it needed?
> 
> IIRC, the do_linuxrc thread cannot be frozen (doesn't call try_to_freeze()),
> so the freeze_processes() during the resume fails and the resume fails as a
> result.

Aha, ok. (We still may want to add try_to_freeze there; there's no
reason to have that running while resuming).

> Still, I have an idea:
> 
> Instead of hunting for PF_NOFREEZE and wondering if the suspend/resume fails
> when we remove them or replace them with try_to_freeze(), why don't we add
> an "ignore_pf_nofreeze" argument to freeze_processes() and make it regard
> _all_ tasks as if they haven't set PF_NOFREEZE when this "ignore_pf_nofreeze"
> is set?  Of course, additionally we'll have to make everyone call
> try_to_freeze(), even if they set PF_NOFREEZE anyway.
> 
> Then, if freeze_processes() is called with "ignore_pf_nofreeze = 0", it will
> work just as it does now.  However, if it's called with
> "ignore_pf_nofreeze = 1", it will try to make all prcesses enter the
> refrigerator.  The "ignore_pf_nofreeze = 0" version will be suitable for us
> (ie. suspend etc.) and the "ignore_pf_nofreeze = 1" version will be suitable
> for the CPU hotplug and such things.

Yep, something like that will be needed. Probably more finegrained
(with flags), because CPU hotplug and kprobes may want their own sets
of unfreezeable processes.
								Pavel
-- 
(english) http://www.livejournal.com/~pavelmachek
(cesky, pictures) http://atrey.karlin.mff.cuni.cz/~pavel/picture/horses/blog.html
-
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/