Re: swsusp initialized after SATA (was Re: swsusp APIC oopsen (was Re: swsusp ooms))

[Andrew Morton <akpm@osdl.org>]

> On Wed, 25 Oct 2006 12:43:18 +0200 Pavel Machek <pavel@ucw.cz> wrote:
> Hi!
> 
> > > > (cc-ed to public list)
> > > > 
> > > > > Andrew Morton <akpm@osdl.org> wrote:
> > > > > 
> > > > > > and I'm not having much luck.  See 
> > > > > > 
> > > > > > http://userweb.kernel.org/~akpm/s5000340.jpg and
> > > > > > http://userweb.kernel.org/~akpm/s5000339.jpg
> > > > > 
> > > > > Running an UP kernel and disabling local APIC avoided the oopses and
> > > > > allowed me to confirm that it was leaking.  whoops.
> > > > 
> > > > I wonder why everyone but me sees those APIC problems?
> > > > 
> > > > Anyway, there's one more problem in -rc1: boot order changed, and (at
> > > > least with paralel boot options), swsusp gets initialized *after*
> > > > swsusp => bad, but should be easy to fix.
> > > 
> > > Sorry, I meant:
> > > 
> > > "sata is initialized *after* swsusp => bad".
> > 
> > Which patch made this change, and why?
> 
> CONFIG_PCI_MULTITHREAD_PROBE is the setting responsible, and IIRC
> that's Greg's code.
> 
> Now... what is the recommended way to wait for hard disks to become
> online?

The multithreaded probing is breaking (or at least altering) the initcall
ordering guarantees.  We should wait for all the probing kernel threads to
terminate after processing each initcall level.