On Tue, Feb 23, 2021 at 02:50:11PM -0800, Andrew Morton wrote:
> On Tue, 26 Jan 2021 15:14:38 -0700 Yu Zhao <yuz...@google.com> wrote:
>
> > On Tue, Jan 26, 2021 at 10:01:11PM +0000, Matthew Wilcox wrote:
> > > On Fri, Jan 22, 2021 at 03:05:53PM -0700, Yu Zhao wrote:
> > > > +++ b/mm/swap.c
> > > > @@ -231,7 +231,7 @@ static void pagevec_move_tail_fn(struct page *page,
> > > > struct lruvec *lruvec)
> > > > if (!PageUnevictable(page)) {
> > > > del_page_from_lru_list(page, lruvec, page_lru(page));
> > > > ClearPageActive(page);
> > > > - add_page_to_lru_list_tail(page, lruvec, page_lru(page));
> > > > + add_page_to_lru_list_tail(page, lruvec);
> > > > __count_vm_events(PGROTATED, thp_nr_pages(page));
> > > > }
> > >
> > > Is it profitable to do ...
> > >
> > > - del_page_from_lru_list(page, lruvec, page_lru(page));
> > > + enum lru_list lru = page_lru(page);
> > > + del_page_from_lru_list(page, lruvec, lru);
> > > ClearPageActive(page);
> > > - add_page_to_lru_list_tail(page, lruvec, page_lru(page));
> > > + lru &= ~LRU_ACTIVE;
> > > + add_page_to_lru_list_tail(page, lruvec, lru);
> >
> > Ok, now we want to trade readability for size. Sure, I'll see how
> > much we could squeeze.
>
> As nothing has happened here and the code bloat issue remains, I'll
> hold this series out of 5.12-rc1.
Sorry for the slow response. I was trying to ascertain why
page_lru(), a tiny helper, could bloat vmlinux by O(KB). It turned out
compound_head() included in Page{Active,Unevictable} is a nuisance in
our case. Testing PG_{active,unevictable} against
compound_head(page)->flags is really unnecessary because all lru
operations are eventually done on page->lru not
compound_head(page)->lru. With the following change, which sacrifices
the readability a bit, we gain 998 bytes with Clang but lose 227 bytes
with GCC, which IMO is a win. (We use Clang by default.)
diff --git a/include/linux/mm_inline.h b/include/linux/mm_inline.h
index 355ea1ee32bd..ec0878a3cdfe 100644
--- a/include/linux/mm_inline.h
+++ b/include/linux/mm_inline.h
@@ -46,14 +46,12 @@ static __always_inline void __clear_page_lru_flags(struct
page *page)
{
VM_BUG_ON_PAGE(!PageLRU(page), page);
- __ClearPageLRU(page);
-
/* this shouldn't happen, so leave the flags to bad_page() */
- if (PageActive(page) && PageUnevictable(page))
+ if ((page->flags & (BIT(PG_active) | BIT(PG_unevictable))) ==
+ (BIT(PG_active) | BIT(PG_unevictable)))
return;
- __ClearPageActive(page);
- __ClearPageUnevictable(page);
+ page->flags &= ~(BIT(PG_lru) | BIT(PG_active) | BIT(PG_unevictable));
}
/**
@@ -65,18 +63,12 @@ static __always_inline void __clear_page_lru_flags(struct
page *page)
*/
static __always_inline enum lru_list page_lru(struct page *page)
{
- enum lru_list lru;
+ unsigned long flags = READ_ONCE(page->flags);
VM_BUG_ON_PAGE(PageActive(page) && PageUnevictable(page), page);
- if (PageUnevictable(page))
- return LRU_UNEVICTABLE;
-
- lru = page_is_file_lru(page) ? LRU_INACTIVE_FILE : LRU_INACTIVE_ANON;
- if (PageActive(page))
- lru += LRU_ACTIVE;
-
- return lru;
+ return (flags & BIT(PG_unevictable)) ? LRU_UNEVICTABLE :
+ (LRU_FILE * !(flags & BIT(PG_swapbacked)) + !!(flags &
BIT(PG_active)));
}
static __always_inline void add_page_to_lru_list(struct page *page,
I'll post this as a separate patch. Below the bloat-o-meter collected
on top of c03c21ba6f4e.
$ ./scripts/bloat-o-meter ../vmlinux.clang.orig ../vmlinux.clang
add/remove: 0/1 grow/shrink: 7/10 up/down: 191/-1189 (-998)
Function old new delta
lru_lazyfree_fn 848 893 +45
lru_deactivate_file_fn 1037 1075 +38
perf_trace_mm_lru_insertion 515 548 +33
check_move_unevictable_pages 983 1006 +23
__activate_page 706 729 +23
trace_event_raw_event_mm_lru_insertion 476 497 +21
lru_deactivate_fn 691 699 +8
__bpf_trace_mm_lru_insertion 13 11 -2
__traceiter_mm_lru_insertion 67 62 -5
move_pages_to_lru 964 881 -83
__pagevec_lru_add_fn 665 581 -84
isolate_lru_page 524 419 -105
__munlock_pagevec 1609 1481 -128
isolate_migratepages_block 3370 3237 -133
__page_cache_release 556 413 -143
lruvec_lru_size 151 - -151
release_pages 1025 866 -159
pagevec_move_tail_fn 805 609 -196
Total: Before=19502982, After=19501984, chg -0.01%
$ ./scripts/bloat-o-meter ../vmlinux.gcc.orig ../vmlinux.gcc
add/remove: 0/1 grow/shrink: 9/9 up/down: 1010/-783 (227)
Function old new delta
shrink_lruvec 1690 1950 +260
lru_deactivate_file_fn 961 1128 +167
isolate_migratepages_block 3286 3427 +141
check_move_unevictable_pages 1042 1170 +128
lru_lazyfree_fn 709 822 +113
lru_deactivate_fn 665 724 +59
__activate_page 703 760 +57
trace_event_raw_event_mm_lru_insertion 432 478 +46
perf_trace_mm_lru_insertion 464 503 +39
__bpf_trace_mm_lru_insertion 13 11 -2
__traceiter_mm_lru_insertion 66 57 -9
isolate_lru_page 472 405 -67
__munlock_pagevec 1282 1212 -70
__pagevec_lru_add 976 893 -83
__page_cache_release 508 418 -90
release_pages 978 887 -91
move_pages_to_lru 954 853 -101
lruvec_lru_size 131 - -131
pagevec_move_tail_fn 770 631 -139
Total: Before=19237248, After=19237475, chg +0.00%
