[0/2] thp nodereclaim fixes

Message

Michal Hocko Sept. 25, 2018, 12:03 p.m. UTC

Hi,
this has been brought up by Andrea [1] and he proposed two different
fixes for the regression. I have proposed an alternative fix [2]. I have
changed my mind in the end because whatever fix we end up with it should
be backported to the stable trees so going with a minimalistic one is
preferred so I have got back to the Andrea's second proposed solution
[3] in the end. I have just reworded the changelog to reflect other bug
report with the stall information.

My primary concern about [3] was that the __GFP_THISNODE logic should be
placed in alloc_hugepage_direct_gfpmask which I've done on top of the
fix as a cleanup (patch 2) and it doesn't need to be backported to the
stable tree.

I am still not happy that the David's workload will regress as a result
but we should really focus on the default behavior and come with a more
robust solution for specialized one for those who have more restrictive
NUMA preferences. I am thinking about a new numa policy that would mimic
node reclaim behavior and I am willing to work on that but we really
have to fix the regression first and that is the patch 1.

Thoughts, alternative patches?

[1] http://lkml.kernel.org/r/[e-mail]
[2] http://lkml.kernel.org/r/[e-mail]
[3] http://lkml.kernel.org/r/[e-mail]

Comments

Kirill A. Shutemov Sept. 26, 2018, 1:08 p.m. UTC | #1

On Tue, Sep 25, 2018 at 02:03:24PM +0200, Michal Hocko wrote:
> Thoughts, alternative patches?
> 
> [1] http://lkml.kernel.org/r/[e-mail]
> [2] http://lkml.kernel.org/r/[e-mail]
> [3] http://lkml.kernel.org/r/[e-mail]

All these links are broken. lore.kernel.org doesn't have linux-mm@ archive.

Can we get it added?

Does anybody have good enough archive of the list to get started?

Michal Hocko Sept. 26, 2018, 1:14 p.m. UTC | #2

On Wed 26-09-18 16:08:50, Kirill A. Shutemov wrote:
> On Tue, Sep 25, 2018 at 02:03:24PM +0200, Michal Hocko wrote:
> > Thoughts, alternative patches?
> > 
> > [1] http://lkml.kernel.org/r/[e-mail]
> > [2] http://lkml.kernel.org/r/[e-mail]
> > [3] http://lkml.kernel.org/r/[e-mail]
> 
> All these links are broken. lore.kernel.org doesn't have linux-mm@ archive.

Yeah, this is really sad. But one good thing is that it gives possible
and usually correct references to marc.info so I keep using it and
hoping this will improve in the future.

> Can we get it added?

I would love to have it covered. You need to talk to Konstantin Ryabitsev.
 
> Does anybody have good enough archive of the list to get started?

Maybe Andrew has. Mine is incomplete for sure.

Konstantin Ryabitsev Sept. 26, 2018, 3:25 p.m. UTC | #3

On Wed, Sep 26, 2018 at 04:08:50PM +0300, Kirill A. Shutemov wrote:
>On Tue, Sep 25, 2018 at 02:03:24PM +0200, Michal Hocko wrote:
>> Thoughts, alternative patches?
>>
>> [1] http://lkml.kernel.org/r/[e-mail]
>> [2] http://lkml.kernel.org/r/[e-mail]
>> [3] http://lkml.kernel.org/r/[e-mail]
>
>All these links are broken. lore.kernel.org doesn't have linux-mm@ archive.
>
>Can we get it added?

Adding linux-mm to lore.kernel.org certainly should happen, but it will 
not fix the above problem, because lkml.kernel.org/r/<foo> links only 
work for messages on LKML, not for all messages passing through vger 
lists (hence the word "lkml" in the name).

Once linux-mm is added, you should link to those discussions using 
lore.kernel.org/linux-mm/<msgid> links.

-K

Andrew Morton Sept. 26, 2018, 10:22 p.m. UTC | #4

On Wed, 26 Sep 2018 15:14:45 +0200 Michal Hocko <[e-mail]> wrote:

> > Does anybody have good enough archive of the list to get started?
> 
> Maybe Andrew has. Mine is incomplete for sure.

fwiw, I have everything going back to 2001.   For some reason.

Mel Gorman Sept. 26, 2018, 11:08 p.m. UTC | #5

On Wed, Sep 26, 2018 at 03:22:00PM -0700, Andrew Morton wrote:
> On Wed, 26 Sep 2018 15:14:45 +0200 Michal Hocko <[e-mail]> wrote:
> 
> > > Does anybody have good enough archive of the list to get started?
> > 
> > Maybe Andrew has. Mine is incomplete for sure.
> 
> fwiw, I have everything going back to 2001.   For some reason.

This is far better than mine and the only person I think could potentially
improve upon the archives is Rik. 2001 would cover the most interesting
periods when things like rmap was introduced. Back then the VM was relatively
simple and the concepts were generally covered by academic textbooks AFAIK.

Konstantin Ryabitsev Sept. 27, 2018, 12:47 a.m. UTC | #6

On Thu, Sep 27, 2018 at 12:08:38AM +0100, Mel Gorman wrote:
> > fwiw, I have everything going back to 2001.   For some reason.
> 
> This is far better than mine and the only person I think could potentially
> improve upon the archives is Rik. 2001 would cover the most interesting
> periods when things like rmap was introduced. Back then the VM was relatively
> simple and the concepts were generally covered by academic textbooks AFAIK.

Now you just need to have a point person who would do the requesting and
archive-providing, per the procedure described on
https://korg.wiki.kernel.org/userdoc/lore.

-K

Kirill A. Shutemov Sept. 27, 2018, 11:30 a.m. UTC | #7

On Wed, Sep 26, 2018 at 11:25:23AM -0400, Konstantin Ryabitsev wrote:
> On Wed, Sep 26, 2018 at 04:08:50PM +0300, Kirill A. Shutemov wrote:
> > On Tue, Sep 25, 2018 at 02:03:24PM +0200, Michal Hocko wrote:
> > > Thoughts, alternative patches?
> > > 
> > > [1] http://lkml.kernel.org/r/[e-mail]
> > > [2] http://lkml.kernel.org/r/[e-mail]
> > > [3] http://lkml.kernel.org/r/[e-mail]
> > 
> > All these links are broken. lore.kernel.org doesn't have linux-mm@ archive.
> > 
> > Can we get it added?
> 
> Adding linux-mm to lore.kernel.org certainly should happen, but it will not
> fix the above problem, because lkml.kernel.org/r/<foo> links only work for
> messages on LKML, not for all messages passing through vger lists (hence the
> word "lkml" in the name).

What is the reason for the separation? From my POV it's beneficial to have
single url scheme to refer any message in the archive regardless of the
list.