try_to_unmap_one() should be able to handle hugetlb, but page_remove_rmap()
in that function takes false as a compound flag, which breaks hugepage migration
with the following message.
Soft offlining page 0x1d4a00 at 0x7ff634a00000
soft offline: 0x1d4a00: migration failed 1, type 60000000000401c
BUG: Bad page state in process sysctl pfn:1d4a00
page:ffffea0007528000 count:0 mapcount:0 mapping: (null) index:0x0 compound_mapcount: 10
flags: 0x600000000004008(uptodate|head)
page dumped because: nonzero mapcount
Modules linked in: cfg80211 rfkill crc32c_intel virtio_balloon serio_raw i2c_piix4 virtio_blk virtio_net ata_generic pata_acpi
CPU: 3 PID: 11882 Comm: sysctl Tainted: G W 4.3.0-mmotm-2015-11-10-15-53-151112-1812-00015-53+ #240
Hardware name: Bochs Bochs, BIOS Bochs 01/01/2011
ffffffff81a1a124 ffff8800da047bd0 ffffffff81340adf ffffea0007528000
ffff8800da047bf8 ffffffff8116e206 0000000000000009 ffffffff81f72538
ffffea0007528000 ffff8800da047c48 ffffffff8116e6d6 0000020000000000
Call Trace:
[<ffffffff81340adf>] dump_stack+0x44/0x55
[<ffffffff8116e206>] bad_page+0xc6/0x110
[<ffffffff8116e6d6>] free_pages_prepare+0x486/0x500
[<ffffffff81170f9a>] __free_pages_ok+0x1a/0xb0
[<ffffffff81171af1>] __free_pages+0x21/0x40
[<ffffffff811b5d79>] update_and_free_page+0xa9/0x230
[<ffffffff811b6641>] free_pool_huge_page+0xc1/0xe0
[<ffffffff811b6d02>] set_max_huge_pages+0x382/0x4d0
[<ffffffff811b7002>] hugetlb_sysctl_handler_common+0xa2/0xd0
[<ffffffff811b823e>] hugetlb_sysctl_handler+0x1e/0x20
[<ffffffff812535fe>] proc_sys_call_handler+0xae/0xc0
[<ffffffff81253624>] proc_sys_write+0x14/0x20
[<ffffffff811e8758>] __vfs_write+0x28/0xe0
[<ffffffff810d1864>] ? percpu_down_read+0x14/0x60
[<ffffffff811e8d29>] vfs_write+0xa9/0x190
[<ffffffff811e9646>] SyS_write+0x46/0xb0
[<ffffffff8163d797>] entry_SYSCALL_64_fastpath+0x12/0x6a
This patch simply fixes this by giving the compound flag via PageHuge.
---
# This patch is against mmotm-2015-11-10-15-53 as a fix for "rmap: add argument
# to charge compound page".
Signed-off-by: Naoya Horiguchi <[email protected]>
---
mm/rmap.c | 2 +-
1 files changed, 1 insertions(+), 1 deletions(-)
diff --git mmotm-2015-11-10-15-53/mm/rmap.c mmotm-2015-11-10-15-53_patched/mm/rmap.c
index e90b81f..e045fea 100644
--- mmotm-2015-11-10-15-53/mm/rmap.c
+++ mmotm-2015-11-10-15-53_patched/mm/rmap.c
@@ -1508,7 +1508,7 @@ static int try_to_unmap_one(struct page *page, struct vm_area_struct *vma,
} else
dec_mm_counter(mm, MM_FILEPAGES);
- page_remove_rmap(page, false);
+ page_remove_rmap(page, PageHuge(page));
page_cache_release(page);
out_unmap:
--
1.7.1
On Fri, Nov 13, 2015 at 09:44:29AM +0900, Naoya Horiguchi wrote:
> try_to_unmap_one() should be able to handle hugetlb, but page_remove_rmap()
> in that function takes false as a compound flag, which breaks hugepage migration
> with the following message.
>
> Soft offlining page 0x1d4a00 at 0x7ff634a00000
> soft offline: 0x1d4a00: migration failed 1, type 60000000000401c
> BUG: Bad page state in process sysctl pfn:1d4a00
> page:ffffea0007528000 count:0 mapcount:0 mapping: (null) index:0x0 compound_mapcount: 10
> flags: 0x600000000004008(uptodate|head)
> page dumped because: nonzero mapcount
> Modules linked in: cfg80211 rfkill crc32c_intel virtio_balloon serio_raw i2c_piix4 virtio_blk virtio_net ata_generic pata_acpi
> CPU: 3 PID: 11882 Comm: sysctl Tainted: G W 4.3.0-mmotm-2015-11-10-15-53-151112-1812-00015-53+ #240
> Hardware name: Bochs Bochs, BIOS Bochs 01/01/2011
> ffffffff81a1a124 ffff8800da047bd0 ffffffff81340adf ffffea0007528000
> ffff8800da047bf8 ffffffff8116e206 0000000000000009 ffffffff81f72538
> ffffea0007528000 ffff8800da047c48 ffffffff8116e6d6 0000020000000000
> Call Trace:
> [<ffffffff81340adf>] dump_stack+0x44/0x55
> [<ffffffff8116e206>] bad_page+0xc6/0x110
> [<ffffffff8116e6d6>] free_pages_prepare+0x486/0x500
> [<ffffffff81170f9a>] __free_pages_ok+0x1a/0xb0
> [<ffffffff81171af1>] __free_pages+0x21/0x40
> [<ffffffff811b5d79>] update_and_free_page+0xa9/0x230
> [<ffffffff811b6641>] free_pool_huge_page+0xc1/0xe0
> [<ffffffff811b6d02>] set_max_huge_pages+0x382/0x4d0
> [<ffffffff811b7002>] hugetlb_sysctl_handler_common+0xa2/0xd0
> [<ffffffff811b823e>] hugetlb_sysctl_handler+0x1e/0x20
> [<ffffffff812535fe>] proc_sys_call_handler+0xae/0xc0
> [<ffffffff81253624>] proc_sys_write+0x14/0x20
> [<ffffffff811e8758>] __vfs_write+0x28/0xe0
> [<ffffffff810d1864>] ? percpu_down_read+0x14/0x60
> [<ffffffff811e8d29>] vfs_write+0xa9/0x190
> [<ffffffff811e9646>] SyS_write+0x46/0xb0
> [<ffffffff8163d797>] entry_SYSCALL_64_fastpath+0x12/0x6a
>
> This patch simply fixes this by giving the compound flag via PageHuge.
> ---
> # This patch is against mmotm-2015-11-10-15-53 as a fix for "rmap: add argument
> # to charge compound page".
>
> Signed-off-by: Naoya Horiguchi <[email protected]>
Acked-by: Kirill A. Shutemov <[email protected]>
--
Kirill A. Shutemov