Re: [PATCH] Fix failure to resume from initrds.

["Rafael J. Wysocki" <rjw@sisk.pl>]

On Tuesday, 11 September 2007 15:41, Nigel Cunningham wrote:
> Hi.
> 
> On Tuesday 11 September 2007 23:23:32 Rafael J. Wysocki wrote:
> > On Tuesday, 11 September 2007 15:12, Rafael J. Wysocki wrote:
> > > On Tuesday, 11 September 2007 13:55, Rafael J. Wysocki wrote:
> > > > On Tuesday, 11 September 2007 13:27, Nigel Cunningham wrote:
> > > > > Hi.
> > > > > 
> > > > > On Tuesday 11 September 2007 21:04:22 Rafael J. Wysocki wrote:
> > > > > > On Tuesday, 11 September 2007 05:54, Nigel Cunningham wrote:
> > > > > > > Hi all.
> > > > > > > 
> > > > > > > Commit 831441862956fffa17b9801db37e6ea1650b0f69 (Freezer: make 
> kernel 
> > > > > threads
> > > > > > > nonfreezable by default) breaks freezing when attempting to resume 
> from an
> > > > > > > initrd, because the init (which is freezeable) spins while waiting 
> for 
> > > > > another
> > > > > > > thread to run /linuxrc, but doesn't check whether it has been told 
> to 
> > > > > enter
> > > > > > > the refrigerator.
> > > > > > 
> > > > > > Hm.
> > > > > > 
> > > > > > I use a resume from an initrd on a regular basis and it works 
> without the 
> > > > > patch
> > > > > > below.
> > > > > > 
> > > > > > I think we need to investigate what happens in your test case a bit.
> > > > > 
> > > > > Ah. That makes me realise that I see that too - my AMD64 uniprocessor 
> laptop 
> > > > > didn't need the patch (guess that's why I didn't notice the need and 
> ack'd 
> > > > > the patch). But my x86 SMP machine... it needs this. I'll see if 
> they're 
> > > > > running on different processors.
> > > > 
> > > > Well, strange.  My x86_64 SMP machines don't need the patch too.
> > > 
> > > Anyway, yes, init is freezable, but should it be?
> > > 
> > > I mean, shouldn't we rather add PF_NOFREEZE to kernel_init()?
> > 
> > Argh, no.  PF_NOFREEZE is inherited by the children.
> > 
> > So, I think that your patch is correct, but there's some suspend2-specific
> > stuff in it.  I've rediffed it against 2.6.23-rc6 and moved try_to_freeze()
> > before yield().
> 
> Ah yeah. Sorry about that. Is there some reason I've forgotten that makes the 
> order of try_to_freeze & yield in a loop like this matter?

Technically it's not that important, but conceptually try_to_freeze() should
happen right after we've been woken up.

> By the way, I had a go at getting fuse processes frozen today. Seems to be 
> doable if you take a freeze filesystems prior to processes approach. I've got 
> a lot more testing and a bit of cleaning up to do before I'd want to show it 
> to anyone, but did successfully do cycles with sshfs, fuseiso and curlftpfs.

Sounds promising. :-)
 
> Of course I don't seriously expect it to get merged - everyone's too much in 
> love with kexec at the moment :)

Well, I guess it'll take some time to get that work reliable and it would be
nice to have a fix before it's ready.

Greetings,
Rafael