[PATCH] mm: disable fault around on emulated access bit architecture

[Minchan Kim]

The fault_around aims for reducing minor fault of file-backed pages
via speculative ahead pte mapping with relying on readahead logic.
However, on non-HW access bit architecture, the benefit is highly
limited because they should emulate young bit with minor fault
for page aging algorithm of reclaim. IOW, we cannot reduce minor fault
on those architectures.

I did quick test in my ARM machine.

512M file mmap sequential every word read on eSATA drive with 4 times.
stdev is stable.

= fault_around 4096 =
elapsed time(usec): 6747645

= fault_around 65536 =
elapsed time(usec): 6709263

0.5% gain.

Even, when I tested it with eMMC, there is no gain because I guess
with slow storage, major fault is more dominant factor.

As well, fault_around has side effect to shrink slab more aggressively
and higher vmpressure so if such speculation fails, it can evict slab
more which can result in page I/O(e.g., inode cache), in the end,
it would make void benefit of fault_around.

So let's make default disable on those architectures.

Cc: Kirill A. Shutemov <kirill@xxxxxxxxxxxxx>
Cc: linux-arch@xxxxxxxxxxxxxxx
Signed-off-by: Minchan Kim <minchan@xxxxxxxxxx>
---
 mm/memory.c | 8 ++++++++
 1 file changed, 8 insertions(+)

diff --git a/mm/memory.c b/mm/memory.c
index b762b17aa4c5..9f652fdc0295 100644
--- a/mm/memory.c
+++ b/mm/memory.c
@@ -2897,8 +2897,16 @@ void do_set_pte(struct vm_area_struct *vma, unsigned long address,
 	update_mmu_cache(vma, address, pte);
 }
 
+/*
+ * If architecture emulates "accessed" or "young" bit without HW support,
+ * there is no much gain with fault_around.
+ */
 static unsigned long fault_around_bytes __read_mostly =
+#ifndef __HAVE_ARCH_PTEP_SET_ACCESS_FLAGS
+	PAGE_SIZE;
+#else
 	rounddown_pow_of_two(65536);
+#endif
 
 #ifdef CONFIG_DEBUG_FS
 static int fault_around_bytes_get(void *data, u64 *val)
-- 
1.9.1