On 09.09.22 02:27, Miaohe Lin wrote:
On 2022/9/8 15:07, Michal Hocko wrote:
On Thu 08-09-22 11:25:54, Miaohe Lin wrote:
On 2022/9/8 11:06, HORIGUCHI NAOYA(堀口 直也) wrote:
On Thu, Sep 08, 2022 at 10:19:03AM +0800, Miaohe Lin wrote:
On 2022/9/7 20:11, Naoya Horiguchi wrote:
...
>From 8a5c284df732943065d23838090d15c94cd10395 Mon Sep 17 00:00:00 2001
From: Naoya Horiguchi <naoya.horiguchi@xxxxxxx>
Date: Wed, 7 Sep 2022 20:58:44 +0900
Subject: [PATCH] mm/huge_memory: use pfn_to_online_page() in
split_huge_pages_all()
NULL pointer dereference is triggered when calling thp split via debugfs
on the system with offlined memory blocks. With debug option enabled,
the following kernel messages are printed out:
page:00000000467f4890 refcount:1 mapcount:0 mapping:0000000000000000 index:0x0 pfn:0x121c000
flags: 0x17fffc00000000(node=0|zone=2|lastcpupid=0x1ffff)
raw: 0017fffc00000000 0000000000000000 dead000000000122 0000000000000000
raw: 0000000000000000 0000000000000000 00000001ffffffff 0000000000000000
page dumped because: unmovable page
page:000000007d7ab72e is uninitialized and poisoned
page dumped because: VM_BUG_ON_PAGE(PagePoisoned(p))
------------[ cut here ]------------
kernel BUG at include/linux/mm.h:1248!
invalid opcode: 0000 [#1] PREEMPT SMP PTI
CPU: 16 PID: 20964 Comm: bash Tainted: G I 6.0.0-rc3-foll-numa+ #41
...
RIP: 0010:split_huge_pages_write+0xcf4/0xe30
This shows that page_to_nid() in page_zone() is unexpectedly called for an
offlined memmap.
Use pfn_to_online_page() to get struct page in PFN walker.
With changes proposed by David, this patch looks good to me.
Reviewed-by: Miaohe Lin <linmiaohe@xxxxxxxxxx>
Thank you.
BTW: IMHO, there might be many similar code places need to take care of memory hotremove where
*pfn is operated directly* while it's not protected against memory hotremove.
I had the similar concern, but there seems many place doing PFN walk,
so checking them one-by-one that offlined memory can be walked over
requires much effort.
Yes, that will be a heavy work. We could fix them one by one if they ever occur. ;)
Most of those whic are directly triggerable should be taken care of. It
would be still good to go through `git grep -w pfn_to_page' and evaluate
all callers. Still more than 400 callsites so not a trivial task.
Agree. Even if pfn_to_online_page() is used, it might still races with the memory hotremove if
there's no way to guard against it, e.g. holding an extra page refcnt. So code audit should also
apply to pfn_to_online_page().
Racing with offlining+unplug a long known problem and so far we decided
to ignore it, because it never appeared in practice.
There once were ideas of using RCU as protection.
--
Thanks,
David / dhildenb
