[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Bug#595711: linux-image-2.6.26-2-xen-686: Wrong free memory values for xen kernel with xenballoon use



On Mon, 2010-09-06 at 22:36 +0400, George Shuklin wrote:
> В Пнд, 06/09/2010 в 07:59 +0100, Ian Campbell пишет:
> > On Mon, 2010-09-06 at 04:03 +0400, George Shuklin wrote:
> > > 
> > > I belive, this somehow related to difference in drivers/xen/balloon.c,
> > > gentoo/SUSE version have some lines like 
> > > 
> > >         totalram_pages--;
> > > (in balloon_append() function)
> > > 
> > > and 
> > >       totalram_pages++;
> > > (in balloon_retrieve() function)
> > > 
> > > Those lines are in Gentoo kernel  (linux-2.6.34-xen) but absent in
> > > Debian (linux-image-2.6.26-2-xen-686) 
> > 
> > Is the Gentoo kernel pvops or classic-Xen patches?
> > 
> > Are you sure you are using the classic-Xen
> > (linux-image-...-xen-{686,amd}) kernel rather than the pvops one
> > (linux-image-...-686-bigmem, no 64 bit pvops in Lenny)?
> > 
> > I wasn't aware that it applied to the classic-Xen case but the
> > difference you describe sounds like this commit from upstream pvops:
> > 
> > http://git.kernel.org/?p=linux/kernel/git/jeremy/xen.git;a=commit;h=3d65c9488cadd2f11bd4d60c7266e639ece5d0d6
> > 
> > Probably for a 2.6.26 classic-Xen kernel you'd need to adjust the path
> > to drivers/xen/balloon/balloon.c and perhaps tweak some context a bit
> > but the concept should still apply.
> 
> Classic Xen (xen-686). pv_ops does not support preinflated ballooning
> (maxmem >mem), so I stuck with classic xen.
> 
> I think, balloon code does not differ (principally) in pv_ops and
> classic xen.
> 
> And, in any way, this bug exists. And it really suck: application ask
> memory, one part of kernel (or libc?) thinking of free memory, kernel
> found 'no memory left' and start OOM killer. It walking and killing some
> random applications (include init).
> 
> I'm testing kernel with with patch right now and it seems work fine -
> memory reporting correctly and oom kills only most bloated application
> requesting tons of memory.

Great, thanks for testing, this confirms that the issue is resolved by
this patch.

Dann, Is it OK to commit a backport of this changeset as the start of
2.6.26-26?

Ian.

-- 
Ian Campbell

Keep on keepin' on.

Attachment: signature.asc
Description: This is a digitally signed message part


Reply to: