Re: [PATCH] mm: Set min_free_kbytes with user_min_free_kbytes when user_min_free_kbytes is preferred

[Fan Liangcai]

Andrew Morton <akpm@xxxxxxxxxxxxxxxxxxxx> 于2021年9月29日周三 上午6:29写道：
>
> On Tue, 28 Sep 2021 20:23:17 +0800 Liangcai Fan <liangcaifan19@xxxxxxxxx> wrote:
>
> > The 'min_free_kbytes' and 'user_min_free_kbytes' maybe inconsistent
> > after a few times of memory hotplug.
>
> What does "inconsistent" mean here?
>
> Please describe the problem in more detail, perhaps with examples.
>
example:
1. Turn on the device with 1G memory, and the calculated 'new_min_free_kbytes'
is 3452kb, at this time 'min_free_kbytes' = 'new_min_free_kbytes' = 3452kb,
'user_min_free_kbytes' is the initial value -1.

2. The user sets 'min_free_kbytes' to 4096kb through the node,
at this time 'user_min_free_kbytes' = 'min_free_kbytes' = 4096kb.

3. The memory has increased by 2G, and the memory has become 3G.
Recalculate 'new_min_free_kbytes' to be 6568kb, which is greater than
'user_min_free_kbytes'(4096kb), at this time
'min_free_kbytes' = 'new_min_free_kbytes' = 6568kb,
'user_min_free_kbytes' = 4096kb.

4. The memory has been reduced by 2G, and now the memory has become 1G.
Recalculate 'new_min_free_kbytes' to be 3452kb, which is less than
'user_min_free_kbytes'(4096kb). The original logic does not modify
'min_free_kbytes'.
At this time, 'min_free_kbytes' is the last 'new_min_free_kbytes'(6568kb).
The memory is 1G ,'min_free_kbytes' should be equal to
'user_min_free_kbytes'(4096kb)
set by the user.

> > When 'new_min_free_kbytes' is not larger than 'user_min_free_kbytes',
> > set 'min_free_kbytes' with 'user_min_free_kbytes' rather than leave
> > it as the 'new_min_free_kbytes' calculated for the last time.
> >
> > Signed-off-by: Liangcai Fan <liangcaifan19@xxxxxxxxx>
> > Cc: Chunyan Zhang <zhang.lyra@xxxxxxxxx>
> > ---
> >  mm/page_alloc.c | 6 ++++++
> >  1 file changed, 6 insertions(+)
> >
> > diff --git a/mm/page_alloc.c b/mm/page_alloc.c
> > index b37435c..ddf9dc1 100644
> > --- a/mm/page_alloc.c
> > +++ b/mm/page_alloc.c
> > @@ -8467,6 +8467,12 @@ int __meminit init_per_zone_wmark_min(void)
> >               if (min_free_kbytes > 262144)
> >                       min_free_kbytes = 262144;
> >       } else {
> > +             /*
> > +              * Set 'min_free_kbytes' with 'user_min_free_kbytes' rather than
> > +              * leave it as the 'new_min_free_kbytes' calculated for the last
> > +              * time.
> > +              */
>
> This comment explains what the code is doing, which is almost always
> obvious from reading the code!  A better comment will describe *why*
> the code is doing whatever is does.   "why, not what", please.
>
I think this comment is redundant and will be deleted.
> > +             min_free_kbytes = user_min_free_kbytes;
> >               pr_warn("min_free_kbytes is not updated to %d because user defined value %d is preferred\n",
> >                               new_min_free_kbytes, user_min_free_kbytes);
> >       }
>