Re: [PATCH mm-unstable v3 6/6] mm/mglru: rework workingset protection

[Yu Zhao <yuzhao@google.com>]

On Mon, Dec 23, 2024 at 04:44:44PM +0800, kernel test robot wrote:
> 
> 
> Hello,
> 
> kernel test robot noticed a 5.7% regression of will-it-scale.per_process_ops on:

Thanks, Oliver!

> commit: 3b7734aa8458b62ecbfd785ca7918e831565006e ("[PATCH mm-unstable v3 6/6] mm/mglru: rework workingset protection")
> url: https://github.com/intel-lab-lkp/linux/commits/Yu-Zhao/mm-mglru-clean-up-workingset/20241208-061714
> base: v6.13-rc1
> patch link: https://lore.kernel.org/all/20241207221522.2250311-7-yuzhao@google.com/
> patch subject: [PATCH mm-unstable v3 6/6] mm/mglru: rework workingset protection
> 
> testcase: will-it-scale
> config: x86_64-rhel-9.4
> compiler: gcc-12
> test machine: 104 threads 2 sockets (Skylake) with 192G memory
> parameters:
> 
> 	nr_task: 100%
> 	mode: process
> 	test: pread2
> 	cpufreq_governor: performance

I think this is very likely caused by my change to folio_mark_accessed()
that unncessarily dirties cache lines shared between different cores.

Could you try the following fix please?

diff --git a/mm/swap.c b/mm/swap.c
index 062c8565b899..54bce14fef30 100644
--- a/mm/swap.c
+++ b/mm/swap.c
@@ -395,7 +395,8 @@ static void lru_gen_inc_refs(struct folio *folio)
 
 	do {
 		if ((old_flags & LRU_REFS_MASK) == LRU_REFS_MASK) {
-			folio_set_workingset(folio);
+			if (!folio_test_workingset(folio))
+				folio_set_workingset(folio);
 			return;
 		}