Re: [PATCH] mm: Optimize TLB flushes during page reclaim

[SeongJae Park <sj@kernel.org>]

On Wed, 22 Jan 2025 07:28:56 -0600 Vinay Banakar <vny@google.com> wrote:

> On Wed, Jan 22, 2025 at 2:59 AM Bharata B Rao <bharata@amd.com> wrote:
> > While that may be true for MADV_PAGEOUT path, does the same assumption
> > hold good for other paths from which shrink_folio_list() gets called?
> 
> shrink_folio_list() is called by three other functions, each with
> different batching behavior:
> - reclaim_clean_pages_from_list(): Doesn't do PMD batching but only
> processes clean pages, so it won't take the path affected by this
> patch. This is called from the contiguous memory allocator
> (cma_alloc#alloc_contig_range)
> - shrink_inactive_list(): Reclaims inactive pages at SWAP_CLUSTER_MAX
> (default 32) at a time. With this patch, we will reduce IPIs for TLB
> flushes by a factor of 32 in kswapd
> - evict_folios(): In the MGLRU case, the number of pages
> shrink_folio_list() processes can vary between 64 (MIN_LRU_BATCH) and
> 4096 (MAX_LRU_BATCH). The reduction in IPIs will vary accordingly

damon_pa_pageout() from mm/damon/paddr.c also calls shrink_folio_list() similar
to madvise.c, but it doesn't aware such batching behavior.  Have you checked
that path?


Thanks,
SJ

> 
> Thanks!
> Vinay