Re: [RFC] restore user defined min_free_kbytes when disabling thp
From: Mel Gorman
Date: Wed Jan 22 2014 - 04:41:11 EST
On Wed, Jan 22, 2014 at 02:05:06PM +0800, Han Pingtian wrote:
> On Tue, Jan 21, 2014 at 10:23:51AM +0000, Mel Gorman wrote:
> > On Tue, Jan 21, 2014 at 05:38:59PM +0800, Han Pingtian wrote:
> > > The testcase 'thp04' of LTP will enable THP, do some testing, then
> > > disable it if it wasn't enabled. But this will leave a different value
> > > of min_free_kbytes if it has been set by admin. So I think it's better
> > > to restore the user defined value after disabling THP.
> > >
> > Then have LTP record what min_free_kbytes was at the same time THP was
> > enabled by the test and restore both settings. It leaves a window where
> > an admin can set an alternative value during the test but that would also
> > invalidate the test in same cases and gets filed under "don't do that".
> Because the value is changed in kernel, so it would be better to
> restore it in kernel, right? :) I have a v2 patch which will restore
> the value only if it isn't set again by user after THP's initialization.
> This v2 patch is dependent on the patch 'mm: show message when updating
> min_free_kbytes in thp' which has been added to -mm tree, can be found
It still feels like the type of scenario that only shows up during tests
that modify kernel parameters as part of the test. I do not consider it
normal operation for THP to be enabled and disabled multiple types during
the lifetime of the system. If the system started with THP disabled, ran
for a long period of time then the benefit of having min_free_kbytes at
a higher value is already lost due to the system being potentially in a
fragmented state already.
I'm ok with the warning being displayed if min_free_kbytes is updated
but I'm not convinced that further trickery is necessary.
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/