Re: [PATCH v2 2/2] mm: add docs for per-order mTHP split counters

[Ryan Roberts]

On 28/06/2024 14:07, Lance Yang wrote:
> This commit introduces documentation for mTHP split counters in
> transhuge.rst.
> 
> Signed-off-by: Mingzhe Yang <mingzhe.yang@xxxxxx>
> Signed-off-by: Lance Yang <ioworker0@xxxxxxxxx>
> ---
>  Documentation/admin-guide/mm/transhuge.rst | 16 ++++++++++++++++
>  1 file changed, 16 insertions(+)
> 
> diff --git a/Documentation/admin-guide/mm/transhuge.rst b/Documentation/admin-guide/mm/transhuge.rst
> index 1f72b00af5d3..709fe10b60f4 100644
> --- a/Documentation/admin-guide/mm/transhuge.rst
> +++ b/Documentation/admin-guide/mm/transhuge.rst
> @@ -514,6 +514,22 @@ file_fallback_charge
>  	falls back to using small pages even though the allocation was
>  	successful.


I note at the top of this section there is a note:

Monitoring usage
================

.. note::
   Currently the below counters only record events relating to
   PMD-sized THP. Events relating to other THP sizes are not included.

Which is out of date, now that we support mTHP stats. Perhaps it should be removed?

>  
> +split
> +	is incremented every time a huge page is successfully split into
> +	base pages. This can happen for a variety of reasons but a common
> +	reason is that a huge page is old and is being reclaimed.
> +	This action implies splitting any block mappings into PTEs.

Now that I'm reading this, I'm reminded that Yang Shi suggested at LSFMM that a
potential aid so solving the swap-out fragmentation problem is to split high
orders to lower (but not 0) orders. I don't know if we would take that route,
but in principle it sounds like splitting mTHP to smaller mTHP might be
something we want some day. I wonder if we should spec this counter to also
include splits to smaller orders and not just splits to base pages?

Actually looking at the code, I think split_huge_page_to_list_to_order(order>0)
would already increment this counter without actually splitting to base pages.
So the documantation should probably just reflect that.

> +
> +split_failed
> +	is incremented if kernel fails to split huge
> +	page. This can happen if the page was pinned by somebody.
> +
> +split_deferred
> +	is incremented when a huge page is put onto split
> +	queue. This happens when a huge page is partially unmapped and
> +	splitting it would free up some memory. Pages on split queue are
> +	going to be split under memory pressure.
> +
>  As the system ages, allocating huge pages may be expensive as the
>  system uses memory compaction to copy data around memory to free a
>  huge page for use. There are some counters in ``/proc/vmstat`` to help