lkml.org 
[lkml]   [2023]   [Nov]   [17]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
SubjectRe: [PATCH V3 3/3] mm: page_alloc: drain pcp lists before oom kill
From
Thanks Michal!!

On 11/16/2023 6:25 PM, Michal Hocko wrote:
>> May be other way to look at that patch is comment is really not being
>> reflected in the code. It says, " Limit the number reserved to 1
>> pageblock or roughly 1% of a zone.", but the current code is making it 2
>> pageblocks. So, for 4M block size, it is > 1%.
>>
>> A second patch, that I will post, like not reserving the high atomic
>> page blocks on small systems -- But how to define the meaning of small
>> systems is not sure. Instead will let the system administrators chose
>> this through either:
>> a) command line param, high_atomic_reserves=off, on by default --
>> Another knob, so admins may really not like this?
>> b) CONFIG_HIGH_ATOMIC_RESERVES, which if not defined, will not reserve.
> Please don't! I do not see any admin wanting to care about this at all.
> It just takes a lot of understanding of internal MM stuff to make an
> educated guess. This should really be auto-tuned. And as responded in
> other reply my take would be to reserve a page block on if it doesn't
> consume more than 1% of memory to preserve the existing behavior yet not
> overconsume on small systems.

This idea of auto tune, by reserving a pageblock only if it doesn't
consume more than 1% of memory, seems cleaner to me. For a page block
size of 4MB, this will turnout to be upto 400MB of RAM.

If it is fine, I can post a patch with suggested-by: you.

>

\
 
 \ /
  Last update: 2023-11-20 14:02    [W:7.032 / U:1.100 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site