Re: swsusp intermittent failures in 2.6.15-rc3-mm1

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

Hi,

On Friday, 2 of December 2005 19:13, Pavel Machek wrote:
> > > > > My Thinkpad X40 (1.25 GB, ipw2200) has had fairly reliable swsusp since
> > > > > 2.6.13 or thereabouts, and as recently as 2.6.15-rc1-mm1 I had about 20
> > > > > successful suspend/resume cycles.  But now that I'm running
> > > > > 2.6.15-rc3-mm1 I'm seeing intermittent failures like this:
> > > > 
> > > > Thanks a lot for the report.
> > > > 
> > > > The problem is apparently caused by my recent patch intended to speed up
> > > > suspend.  Could you please apply the appended debug patch, try to reproduce
> > > > the problem and send full dmesg output back to me?
> > > > 
> > > > Index: linux-2.6.15-rc3-mm1/kernel/power/swsusp.c
> > > > ===================================================================
> > > > --- linux-2.6.15-rc3-mm1.orig/kernel/power/swsusp.c	2005-12-01 22:13:17.000000000 +0100
> > > > +++ linux-2.6.15-rc3-mm1/kernel/power/swsusp.c	2005-12-01 22:24:56.000000000 +0100
> > > > @@ -635,12 +635,18 @@
> > > >  	printk("Shrinking memory...  ");
> > > >  	do {
> > > >  #ifdef FAST_FREE
> > > > -		tmp = count_data_pages() + count_highmem_pages();
> > > > +		tmp = count_data_pages();
> > > > +		printk("Data pages: %ld\n", tmp);
> > > > +		tmp += count_highmem_pages();
> > > 
> > > You need at least count_data_pages() + 2*count_highmem_pages() free
> > > pages (in lowmem).
> > 
> > Do you mean a separate page is needed for each kmalloc() in
> > save_highmem_zone()?
> 
> No, but notice get_zeroed_page() there, and that still needs to be
> atomically copied.
> 
> So each higmem page takes:
> 
> 1 get_zeroed_page()
> 1 kmalloc(struct(highmem_page))
> + copies of those into snapshot.

Yeah, right.  I forgot to count them twice and I didn't take the kmalloc()s
into account.  I'll do my best to produce a patch for it ASAP.

BTW, every kmalloc() in there seems to take 32 bytes (ie. the smallest generic
slab object size), which is a bit wasteful.  The Andy's numbers indicate it
can take more than 700 pages just for that, and they have to be taken twice,
which gives more than 1400 pages total.  It could be as little as about 300
pages _total_ if we did it more carefully.  I think I'll place this on my todo
list for after 2.6.15.

Greetings,
Rafael


-- 
Beer is proof that God loves us and wants us to be happy - Benjamin Franklin