Re: [PATCH v2] mm/memory-failure: fix VM_BUG_ON_PAGE(PagePoisoned(page)) when unpoison memory

[David Hildenbrand <david@redhat.com>]

On 20.12.24 03:35, Miaohe Lin wrote:
> On 2024/12/19 20:18, David Hildenbrand wrote:
>> On 19.12.24 12:52, Miaohe Lin wrote:
>>> When I did memory failure tests recently, below panic occurs:
>>>
>>> page dumped because: VM_BUG_ON_PAGE(PagePoisoned(page))
>>> kernel BUG at include/linux/page-flags.h:616!
>>> Oops: invalid opcode: 0000 [#1] PREEMPT SMP NOPTI
>>> CPU: 3 PID: 720 Comm: bash Not tainted 6.10.0-rc1-00195-g148743902568 #40
>>> RIP: 0010:unpoison_memory+0x2f3/0x590
>>> RSP: 0018:ffffa57fc8787d60 EFLAGS: 00000246
>>> RAX: 0000000000000037 RBX: 0000000000000009 RCX: ffff9be25fcdc9c8
>>> RDX: 0000000000000000 RSI: 0000000000000027 RDI: ffff9be25fcdc9c0
>>> RBP: 0000000000300000 R08: ffffffffb4956f88 R09: 0000000000009ffb
>>> R10: 0000000000000284 R11: ffffffffb4926fa0 R12: ffffe6b00c000000
>>> R13: ffff9bdb453dfd00 R14: 0000000000000000 R15: fffffffffffffffe
>>> FS:  00007f08f04e4740(0000) GS:ffff9be25fcc0000(0000) knlGS:0000000000000000
>>> CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
>>> CR2: 0000564787a30410 CR3: 000000010d4e2000 CR4: 00000000000006f0
>>> Call Trace:
>>>    <TASK>
>>>    unpoison_memory+0x2f3/0x590
>>>    simple_attr_write_xsigned.constprop.0.isra.0+0xb3/0x110
>>>    debugfs_attr_write+0x42/0x60
>>>    full_proxy_write+0x5b/0x80
>>>    vfs_write+0xd5/0x540
>>>    ksys_write+0x64/0xe0
>>>    do_syscall_64+0xb9/0x1d0
>>>    entry_SYSCALL_64_after_hwframe+0x77/0x7f
>>> RIP: 0033:0x7f08f0314887
>>> RSP: 002b:00007ffece710078 EFLAGS: 00000246 ORIG_RAX: 0000000000000001
>>> RAX: ffffffffffffffda RBX: 0000000000000009 RCX: 00007f08f0314887
>>> RDX: 0000000000000009 RSI: 0000564787a30410 RDI: 0000000000000001
>>> RBP: 0000564787a30410 R08: 000000000000fefe R09: 000000007fffffff
>>> R10: 0000000000000000 R11: 0000000000000246 R12: 0000000000000009
>>> R13: 00007f08f041b780 R14: 00007f08f0417600 R15: 00007f08f0416a00
>>>    </TASK>
>>> Modules linked in: hwpoison_inject
>>> ---[ end trace 0000000000000000 ]---
>>> RIP: 0010:unpoison_memory+0x2f3/0x590
>>> RSP: 0018:ffffa57fc8787d60 EFLAGS: 00000246
>>> RAX: 0000000000000037 RBX: 0000000000000009 RCX: ffff9be25fcdc9c8
>>> RDX: 0000000000000000 RSI: 0000000000000027 RDI: ffff9be25fcdc9c0
>>> RBP: 0000000000300000 R08: ffffffffb4956f88 R09: 0000000000009ffb
>>> R10: 0000000000000284 R11: ffffffffb4926fa0 R12: ffffe6b00c000000
>>> R13: ffff9bdb453dfd00 R14: 0000000000000000 R15: fffffffffffffffe
>>> FS:  00007f08f04e4740(0000) GS:ffff9be25fcc0000(0000) knlGS:0000000000000000
>>> CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
>>> CR2: 0000564787a30410 CR3: 000000010d4e2000 CR4: 00000000000006f0
>>> Kernel panic - not syncing: Fatal exception
>>> Kernel Offset: 0x31c00000 from 0xffffffff81000000 (relocation range: 0xffffffff80000000-0xffffffffbfffffff)
>>> ---[ end Kernel panic - not syncing: Fatal exception ]---
>>>
>>> The root cause is that unpoison_memory() tries to check the PG_HWPoison
>>> flags of an uninitialized page. So VM_BUG_ON_PAGE(PagePoisoned(page)) is
>>> triggered. This can be reproduced by below steps:
>>> 1.Offline memory block:
>>>    echo offline > /sys/devices/system/memory/memory12/state
>>> 2.Get offlined memory pfn:
>>>    page-types -b n -rlN
>>> 3.Write pfn to unpoison-pfn
>>>    echo <pfn> > /sys/kernel/debug/hwpoison/unpoison-pfn
>>>
>>> Signed-off-by: Miaohe Lin <linmiaohe@huawei.com>
>>> ---
>>> v2: Use pfn_to_online_page per David. Thanks.
>>> ---
>>>    mm/memory-failure.c | 14 +++++++++++---
>>>    1 file changed, 11 insertions(+), 3 deletions(-)
>>>
>>> diff --git a/mm/memory-failure.c b/mm/memory-failure.c
>>> index a7b8ccd29b6f..02be0596ce67 100644
>>> --- a/mm/memory-failure.c
>>> +++ b/mm/memory-failure.c
>>> @@ -2556,10 +2556,18 @@ int unpoison_memory(unsigned long pfn)
>>>        static DEFINE_RATELIMIT_STATE(unpoison_rs, DEFAULT_RATELIMIT_INTERVAL,
>>>                        DEFAULT_RATELIMIT_BURST);
>>>    -    if (!pfn_valid(pfn))
>>> -        return -ENXIO;
>>> +    p = pfn_to_online_page(pfn);
>>> +    if (!p) {
>>> +        struct dev_pagemap *pgmap;
>>>    -    p = pfn_to_page(pfn);
>>> +        if (!pfn_valid(pfn))
>>> +            return -ENXIO;
>>> +        pgmap = get_dev_pagemap(pfn, NULL);
>>> +        if (!pgmap)
>>> +            return -ENXIO;
>>> +        put_dev_pagemap(pgmap);
>>> +        p = pfn_to_page(pfn);
>>> +    }
>>
>> Hm, I wonder if we can do anything reasonable with ZONE_DEVICE pages here?
> 
> All I can see in unpoison_memory() is folio_test_clear_hwpoison() for ZONE_DEVICE pages.

IIRC, it can only be triggered via debugfs in special kernel configs. So 
chances are this was never ever actually run against a ZONE_DEVICE page.

> 
>>
>> CCing Dan, maybe he knows if this interface used to do something reasonable with ZONE_DEVICE pages.
>>
>> Also, I'm not sure about using the page after doing the put_dev_pagemap(). Likely we would have to do that at the before exiting from this function.
> 
> IMHO, the page can be used after doing put_dev_pagemap(). The page should still be
> available while it has HWPoison set.

1) Why do you think it can be used afterwards? :)

2) At that point in time you don't even know yet if the page is
    HWPoisoned! That check is performed later

3) From GUP code, I recall that we must keep the pagemap referenced
    until we grabbed a page/folio reference.

Or are you worrying about memory offline? That
> might not be the scope of this problem. But I might be miss something.

I suspect unpoison_memory() doesn't do with ZONE_DEVICE pages what it 
should be doing. Maybe I'm wrong, it would be great to get feedback from 
Dan.


-- 
Cheers,

David / dhildenb