Re: [PATCH] mm/memory_hotplug: Cache auto_movable stats to optimize online check

[David Hildenbrand (Red Hat)]

On 12/18/25 07:41, Swaraj Gaikwad wrote:
> Hi David,

Hi!

> I’m just checking in on this patch to see if you’ve had a chance to review
> the benchmark results I shared last week. 

Not really, still traveling and this is veeeery low priority :)

> To recap, the caching reduced the
> execution time for the NUMA_NO_NODE case from ~2402 ns to ~453 ns in my test
> environment.

Yeah, but these micro-benchmarks don't really matter ... at all. What 
would be interesting is what happens when you hotplug a lot of memory to 
a system with a lot of nodes. I suspect it won't really be a problem.

> Please let me know if the performance gain justifies the change in your view,or
> if you would prefer I send a v2 that simply updates the TODO comment as you suggested.

I'd prefer if things that are not a real problem would not consume my 
bandwidth while traveling ;)

Anyhow, there was a kernel bot complaint that you are using "struct 
auto_movable_stats" before the compiler knows about it (and its size).

When you resend, make sure to better describe the "why" we are doing it. 
It cleans up the code a little, so that could be used as an argument.

--
Cheers

David