Re: [PATCH v5 17/17] mm/mglru: make folio_test_workingset() work based on folio LRU refs

From: Kairui Song

Date: Sat Nov 21 2026 - 15:13:45 EST


On Fri, Nov 20, 2026 at 05:53:36PM +0800, KunWu Chan wrote:
> On Thu, Nov 19, 2026 at 5:04 AM Kairui Song via B4 Relay
> <devnull+kasong.tencent.com@xxxxxxxxxx> wrote:
> >
> > From: Kairui Song <kasong@xxxxxxxxxxx>
> >
> > folio_test_workingset() currently tests the PG_workingset bit, which
> > is only the second lowest access bit of the folio LRU refs count now.
> > Folios with refs 4 or 5 carry it in the LRU_REFS_MASK field
> > and no longer have PG_workingset set, so the bit test misses them.
> >
> > folio_set_workingset() currently sets the PG_workingset bit with a
> > plain set_bit(), which also over-promotes folios at a higher refs
> > count: e.g. a folio at refs 4 is bumped to refs 6, advancing it
> > toward LRU_REFS_MAX and an earlier generation promotion.
> >
> > Move the test into mm_inline.h as an inline helper based on
> > folio_lru_refs(), checking refs >= LRU_REFS_WORKINGSET. The set
> > stays a plain PG_workingset bit operation reserved for the
> > classical LRU: under MGLRU the refs count is maintained by
> > folio_inc_lru_refs(), and now triggers a debug WARN when called
> > while MGLRU is fully on (the switching window is exempt, as the
> > classical paths legitimately run alongside MGLRU then). The one
> > PageWorkingset() user in erofs is converted to the folio helper.
> >
> > Under the classical LRU the refs count is not maintained, so the
> > test is equivalent to the PG_workingset bit test, which is the old
> > behavior.
> >
> > Tested-by: Zhao Zhenzhuo <zhaozhengzhuo@xxxxxxxxxxxxx>
> > Signed-off-by: Kairui Song <kasong@xxxxxxxxxxx>
> > ---

...

> > +/*
> > + * For the classical LRU only: with MGLRU the PG_workingset bit is
> > + * part of the folio refs count maintained by folio_inc_lru_refs(),
> > + * and a raw set would corrupt it. The switching window is exempt
> > + * because the classical paths legitimately run alongside MGLRU then.
> > + */
> > +static __always_inline void folio_set_workingset(struct folio *folio)
> > +{
> > + VM_WARN_ON_ONCE(lru_gen_enabled() && !lru_gen_switching());
>
> Could shrink_active_list() still hit this WARN after the switch
> to MGLRU completes? It can reschedule after isolating folios and
> reach this setter after lru_switch is cleared.

Hi Kunwu, thanks for the very careful review!

Right I thought a lot about it after your reply, first your concern
is right, thanks! I should just remove and replace this WARN with
something else.

And in theory, we should only call this for isolated folio (for CLRU
it means a folio moving out of active workingset, isolation is always
the first step), so this helper is fine - except, for the MADV cold
case, it will call it on for a on-list folio, and will could skew
the counter and if it raced with switching. Switching is so painful
I hope we can get rid of it soon :/

But, good thing is it's easy to fix, we just need (part of this belong
to another patch, a trimmed diff just for reference):

diff --git a/include/linux/mm_inline.h b/include/linux/mm_inline.h
index 4f65867446dc..c936d1fadfbd 100644
--- a/include/linux/mm_inline.h
+++ b/include/linux/mm_inline.h
@@ -261,8 +262,9 @@ static inline void __folio_set_lru_refs(struct folio *folio, unsigned int refs)

do {
new_flags = old_flags;
- VM_WARN_ON_ONCE(lru_get_gen_flags(old_flags) != -1);
lru_set_refs_flags(&new_flags, refs);
+ if (lru_get_gen_flags(old_flags) != -1)
+ break;
if (new_flags == old_flags)
break;
} while (!try_cmpxchg(folio_flags(folio, 0), &old_flags, new_flags));
@@ -569,16 +571,16 @@ static inline void folio_clear_lru_refs(struct folio *folio)
*/
static __always_inline void folio_set_workingset(struct folio *folio)
{
- VM_WARN_ON_ONCE(lru_gen_enabled() && !lru_gen_switching());
- set_bit(PG_workingset, folio_flags(folio, FOLIO_HEAD_PAGE));
+ unsigned long flags;
+
+ flags = set_mask_bits(folio_flags(folio, 0), BIT(PG_workingset),
+ BIT(PG_workingset));
+ VM_WARN_ON_ONCE(lru_get_gen_flags(flags) != -1);
}

static __always_inline
diff --git a/mm/madvise.c b/mm/madvise.c
index bbbb729ed647..a73d89a2ecca 100644
--- a/mm/madvise.c
+++ b/mm/madvise.c
@@ -374,11 +374,12 @@ static void madvise_cold_prep_folio(struct folio *folio)
* as a workingset refault, and the referenced bit always needs to
* be cleared even for inactive folios. For MGLRU, MADV_COLD clears
* refs in folio_deactivate(); MADV_PAGEOUT handles refs after isolation.
*/
if (!lru_gen_enabled()) {
folio_clear_lru_refs(folio);
if (folio_test_active(folio))
- folio_set_workingset(folio);
+ __folio_set_lru_refs(folio, LRU_REFS_WORKINGSET);
}
}