bio_iov_iter_get_pages() + page_alloc.shuffle=1 migrating failures

From: Qian Cai
Date: Thu Apr 25 2019 - 01:07:59 EST


Memory offline [1] starts to fail on linux-next on ppc64le with
page_alloc.shuffle=1 where the "echo offline" command hangs with lots of
migrating failures below. It seems in migrate_page_move_mapping()

if (!mapping) {
/* Anonymous page without mapping */
if (page_count(page) != expected_count)
return -EAGAIN;

It expected count=1 but actual count=2.

There are two ways to make the problem go away. One is to remove this line in
__shuffle_free_memory(),

shuffle_zone(z);

The other is reverting some bio commits. Bisecting so far indicates the culprit
is in one of those (the 3rd commit looks more suspicious than the others).

block: only allow contiguous page structs in a bio_vec
block: don't allow multiple bio_iov_iter_get_pages calls per bio
block: change how we get page references in bio_iov_iter_get_pages

[ 446.578064] migrating pfn 2003d5eaa failed ret:22
[ 446.578066] page:c00a00800f57aa80 count:2 mapcount:0 mapping:c000001db4c827e9
index:0x13c08a
[ 446.578220] anon
[ 446.578222] flags: 0x83fffc00008002e(referenced|uptodate|dirty|active|swapbacked)
[ 446.578347] raw: 083fffc00008002e c00a00800f57f808 c00a00800f579f88
c000001db4c827e9
[ 446.944807] raw: 000000000013c08a 0000000000000000 00000002ffffffff
c00020141a738008
[ 446.944883] page dumped because: migration failure
[ 446.944948] page->mem_cgroup:c00020141a738008
[ 446.945024] page allocated via order 0, migratetype Movable, gfp_mask
0x100cca(GFP_HIGHUSER_MOVABLE)
[ 446.945148] prep_new_page+0x390/0x3a0
[ 446.945228] get_page_from_freelist+0xd9c/0x1bf0
[ 446.945292] __alloc_pages_nodemask+0x1cc/0x1780
[ 446.945335] alloc_pages_vma+0xc0/0x360
[ 446.945401] do_anonymous_page+0x244/0xb20
[ 446.945472] __handle_mm_fault+0xcf8/0xfb0
[ 446.945532] handle_mm_fault+0x1c0/0x2b0
[ 446.945615] __get_user_pages+0x3ec/0x690
[ 446.945652] get_user_pages_unlocked+0x104/0x2f0
[ 446.945693] get_user_pages_fast+0xb0/0x200
[ 446.945762] iov_iter_get_pages+0xf4/0x6a0
[ 446.945802] bio_iov_iter_get_pages+0xc0/0x450
[ 446.945876] blkdev_direct_IO+0x2e0/0x630
[ 446.945941] generic_file_read_iter+0xbc/0x230
[ 446.945990] blkdev_read_iter+0x50/0x80
[ 446.946031] aio_read+0x128/0x1d0
[ 446.946082] migrating pfn 2003d5fe0 failed ret:22
[ 446.946084] page:c00a00800f57f800 count:2 mapcount:0 mapping:c000001db4c827e9
index:0x13c19e
[ 446.946239] anon
[ 446.946241] flags: 0x83fffc00008002e(referenced|uptodate|dirty|active|swapbacked)
[ 446.946384] raw: 083fffc00008002e c000200deb3dfa28 c00a00800f57aa88
c000001db4c827e9
[ 446.946497] raw: 000000000013c19e 0000000000000000 00000002ffffffff
c00020141a738008
[ 446.946605] page dumped because: migration failure
[ 446.946662] page->mem_cgroup:c00020141a738008
[ 446.946724] page allocated via order 0, migratetype Movable, gfp_mask
0x100cca(GFP_HIGHUSER_MOVABLE)
[ 446.946846] prep_new_page+0x390/0x3a0
[ 446.946899] get_page_from_freelist+0xd9c/0x1bf0
[ 446.946959] __alloc_pages_nodemask+0x1cc/0x1780
[ 446.947047] alloc_pages_vma+0xc0/0x360
[ 446.947101] do_anonymous_page+0x244/0xb20
[ 446.947143] __handle_mm_fault+0xcf8/0xfb0
[ 446.947200] handle_mm_fault+0x1c0/0x2b0
[ 446.947256] __get_user_pages+0x3ec/0x690
[ 446.947306] get_user_pages_unlocked+0x104/0x2f0
[ 446.947366] get_user_pages_fast+0xb0/0x200
[ 446.947458] iov_iter_get_pages+0xf4/0x6a0
[ 446.947515] bio_iov_iter_get_pages+0xc0/0x450
[ 446.947588] blkdev_direct_IO+0x2e0/0x630
[ 446.947636] generic_file_read_iter+0xbc/0x230
[ 446.947703] blkdev_read_iter+0x50/0x80
[ 446.947758] aio_read+0x128/0x1d0

[1]
i=0
found=0
for mem in $(ls -d /sys/devices/system/memory/memory*); do
i=$((i + 1))
echo "iteration: $i"
echo offline > $mem/state
if [ $? -eq 0 ] && [ $found -eq 0 ]; then
found=1
continue
fi
echo online > $mem/state
done