Re: [PATCH v2] mm: Don't overwrite user min_free_kbytes, consider THP when adjusting

[Andrew Morton]

On Mon, 10 Feb 2020 11:01:21 -0800 Mike Kravetz <mike.kravetz@xxxxxxxxxx> wrote:

> The value of min_free_kbytes is calculated in two routines:
> 1) init_per_zone_wmark_min based on available memory
> 2) set_recommended_min_free_kbytes may reserve extra space for
>    THP allocations
> 
> In both of these routines, a user defined min_free_kbytes value will
> be overwritten if the value calculated in the code is larger. No message
> is logged if the user value is overwritten.

Could we provide a detailed description of why this is considered to be
a problem?  This is fairly easily guessable, but is there a real
in-field bad user experience we can point at?

> Change code to never overwrite user defined value.  However, do log a
> message (once per value) showing the value calculated in code.
> 
> At system initialization time, both init_per_zone_wmark_min and
> set_recommended_min_free_kbytes are called to set the initial value
> for min_free_kbytes.  When memory is offlined or onlined, min_free_kbytes
> is recalculated and adjusted based on the amount of memory.  However,
> the adjustment for THP is not considered.  Here is an example from a 2
> node system with 8GB of memory.
> 
>  # cat /proc/sys/vm/min_free_kbytes
>  90112
>  # echo 0 > /sys/devices/system/node/node1/memory56/online
>  # cat /proc/sys/vm/min_free_kbytes
>  11243
>  # echo 1 > /sys/devices/system/node/node1/memory56/online
>  # cat /proc/sys/vm/min_free_kbytes
>  11412
> 
> One would expect that min_free_kbytes would return to it's original
> value after the offline/online operations.
> 
> Create a simple interface for THP/khugepaged based adjustment and
> call this whenever min_free_kbytes is adjusted.
> 
> ...
>
>  include/linux/khugepaged.h |  5 ++++
>  mm/internal.h              |  2 ++
>  mm/khugepaged.c            | 56 ++++++++++++++++++++++++++++++++------
>  mm/page_alloc.c            | 35 ++++++++++++++++--------

min_free_kbytes gets a few mentions in Documentation/.  Should we make
the appropriate updates there to bring this behavior to people's
attention?