On Sun, Aug 5, 2012 at 6:36 PM, Ben Hutchings
<ben@decadent.org.uk> wrote:
On Sun, 2012-08-05 at 17:50 -0500, hugo vanwoerkom wrote:
[...]
> Yes, 3.2.19-1 still takes 12s. But as to problems
> disappearing/appearing I could have sworn that 3.2.21-3 took 12s. last
> night but now it takes 35s. I think the problem comes from the kernel,
> because that is the only thing that I am varying and I have been
> hibernating this box for years with up to now predictable speeds.
> Don't use wicd though.
If the change was made between 3.2.19 and 3.2.21 then it might be due
to:
commit d006ab31cd818f5e4dda2453fd09767063f49933
Author: Michal Hocko <mhocko@suse.cz>
Date: Tue May 29 15:06:45 2012 -0700
mm: consider all swapped back pages in used-once logic
So you could test with the reverse of that patch (attached), following
I tested all the snapshot kernels again and the change occurred with 3.2.21-1. That is the first one to have a hibernate of 35s. The one previous to that 3.2.20-1 is fine with 10s. These results are consistent. It could not be the mm vmscan patch because both 3.2.20-1 and 3.2.21-1 have that set the same way:
if (referenced_ptes) {
if (PageSwapBacked(page))
return PAGEREF_ACTIVATE;
Hugo