Re: [PATCH 0/9] mm/madvise: batch tlb flushes for MADV_DONTNEED and MADV_FREE

[SeongJae Park <sj@kernel.org>]

On Mon, 10 Mar 2025 15:39:21 -0700 Andrew Morton <akpm@linux-foundation.org> wrote:

> On Mon, 10 Mar 2025 10:23:09 -0700 SeongJae Park <sj@kernel.org> wrote:
> 
> >  It is unclear if such use case
> > is common and the inefficiency is significant. 
> 
> Well, we could conduct a survey,
> 
> Can you add some logging to detect when userspace performs such an
> madvise() call, then run that kernel on some "typical" machines which
> are running "typical" workloads?  That should give us a feeling for how
> often userspace does this,

I agree that could make this patch series more informative.

> and hence will help us understand the usefulness
> of this patchset.

Nevertheless, what this patchset is really trying to optimize is not the
madvise() use case, but process_madvise() use.  I believe the usage is
apparently common, given the vectorization based semantic of process_madvise().
Jemalloc is also adding[1] that kind of use case.  And the benefit is clear,
given the microbenchmark results that I shared.

Also, this patchset shouldn't introduce new regression to madvise().

Hence, I think the survey can be interestign and helpful, but shouldn't be a
blocker for this patch series.  Coudl you please let me know if I'm missing
something and if you still want the survey?

[1] https://github.com/jemalloc/jemalloc/pull/2794


Thanks,
SJ