Re: 2.6.24-rc5-mm1 -- inconsistent {in-hardirq-W} -> {hardirq-on-W} usage -- pm-hibernate/9940 [HC0[0]:SC0[0]:HE1:SE1]

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

On Wednesday, 19 of December 2007, Christoph Lameter wrote:
> On Wed, 19 Dec 2007, Daniel Walker wrote:
> 
> > > It looks like the swsusp_save() calls drain_all_pages() , which calls
> > > on_each_cpu() .. On return on_each_cpu() unconditionally enables
> > > interrupts so the rest of the resume process has interrupt enable
> > > (which , it looks like, shouldn't happen) and then you get the lockdep()
> > > warning due to the above..
> > > 
> > > Not sure if this has been found already, or not?
> 
> Hmmm... It will unconditionally enable interrupts regardless how we call 
> this. We could explicity save and restore interrrupts in 
> swsusp_save() I guess. Why is swsusp_save() disabling interrupts?

Actually, it's called with interrupts disabled, because it's job is to create
the hibernation image.  At this point everything is off except for the CPU
running swsusp_save().
 
> > > Should drain_all_pages() really be drain_local_pages() ?
> > 
> > It looks like it was drain_local_pages, but the following patch
> > 
> > page-allocator-clean-up-pcp-draining-functions.patch
> > 
> > Changes that in -mm .. I added Christoph Lameter to the CC since it's
> > his patch ..
> 
> We could reexport drain_local_pages() again but then I do not understand 
> why we would only drain the pages of this processor and not of all other
> processors as well. It seems that software suspend intend was to flush 
> them all right?

Well, not exactly.  We are on one CPU at this point, the others have been
disabled.