From: Jane Chu <jane.chu@oracle.com>
To: Miaohe Lin <linmiaohe@huawei.com>,
nao.horiguchi@gmail.com, akpm@linux-foundation.org,
linux-mm@kvack.org, linux-kernel@vger.kernel.org
Subject: Re: [PATCH 3/3] mm/memory-failure: send SIGBUS in the event of thp split fail
Date: Thu, 9 May 2024 20:18:46 -0700 [thread overview]
Message-ID: <d6b7c197-2b3c-442d-9b0f-ff7db3bd1ed9@oracle.com> (raw)
In-Reply-To: <32b88a87-8edc-12eb-1fd7-2a028b8f9fb3@huawei.com>
On 5/9/2024 7:59 PM, Miaohe Lin wrote:
> On 2024/5/9 23:34, Jane Chu wrote:
>> On 5/9/2024 1:30 AM, Miaohe Lin wrote:
>>> On 2024/5/9 1:45, Jane Chu wrote:
>>>> On 5/8/2024 1:08 AM, Miaohe Lin wrote:
>>>>
>>>>> On 2024/5/7 4:26, Jane Chu wrote:
>>>>>> On 5/5/2024 12:00 AM, Miaohe Lin wrote:
>>>>>>
>>>>>>> On 2024/5/2 7:24, Jane Chu wrote:
>>>>>>>> When handle hwpoison in a GUP longterm pin'ed thp page,
>>>>>>>> try_to_split_thp_page() will fail. And at this point, there is little else
>>>>>>>> the kernel could do except sending a SIGBUS to the user process, thus
>>>>>>>> give it a chance to recover.
>>>>>>>>
>>>>>>>> Signed-off-by: Jane Chu <jane.chu@oracle.com>
>>>>>>> Thanks for your patch. Some comments below.
>>>>>>>
>>>>>>>> ---
>>>>>>>> mm/memory-failure.c | 36 ++++++++++++++++++++++++++++++++++++
>>>>>>>> 1 file changed, 36 insertions(+)
>>>>>>>>
>>>>>>>> diff --git a/mm/memory-failure.c b/mm/memory-failure.c
>>>>>>>> index 7fcf182abb96..67f4d24a98e7 100644
>>>>>>>> --- a/mm/memory-failure.c
>>>>>>>> +++ b/mm/memory-failure.c
>>>>>>>> @@ -2168,6 +2168,37 @@ static int memory_failure_dev_pagemap(unsigned long pfn, int flags,
>>>>>>>> return rc;
>>>>>>>> }
>>>>>>>> +/*
>>>>>>>> + * The calling condition is as such: thp split failed, page might have
>>>>>>>> + * been GUP longterm pinned, not much can be done for recovery.
>>>>>>>> + * But a SIGBUS should be delivered with vaddr provided so that the user
>>>>>>>> + * application has a chance to recover. Also, application processes'
>>>>>>>> + * election for MCE early killed will be honored.
>>>>>>>> + */
>>>>>>>> +static int kill_procs_now(struct page *p, unsigned long pfn, int flags,
>>>>>>>> + struct page *hpage)
>>>>>>>> +{
>>>>>>>> + struct folio *folio = page_folio(hpage);
>>>>>>>> + LIST_HEAD(tokill);
>>>>>>>> + int res = -EHWPOISON;
>>>>>>>> +
>>>>>>>> + /* deal with user pages only */
>>>>>>>> + if (PageReserved(p) || PageSlab(p) || PageTable(p) || PageOffline(p))
>>>>>>>> + res = -EBUSY;
>>>>>>>> + if (!(PageLRU(hpage) || PageHuge(p)))
>>>>>>>> + res = -EBUSY;
>>>>>>> Above checks seems unneeded. We already know it's thp?
>>>>>> Agreed.
>>>>>>
>>>>>> I lifted these checks from hwpoison_user_mapping() with a hope to make kill_procs_now() more generic,
>>>>>>
>>>>>> such as, potentially replacing kill_accessing_processes() for re-accessing hwpoisoned page.
>>>>>>
>>>>>> But I backed out at last, due to concerns that my tests might not have covered sufficient number of scenarios.
>>>>>>
>>>>>>>> +
>>>>>>>> + if (res == -EHWPOISON) {
>>>>>>>> + collect_procs(folio, p, &tokill, flags & MF_ACTION_REQUIRED);
>>>>>>>> + kill_procs(&tokill, true, pfn, flags);
>>>>>>>> + }
>>>>>>>> +
>>>>>>>> + if (flags & MF_COUNT_INCREASED)
>>>>>>>> + put_page(p);
>>>>>>> This if block is broken. put_page() has been done when try_to_split_thp_page() fails?
>>>>>> put_page() has not been done if try_to_split_thp_page() fails, and I think it should.
>>>>> In try_to_split_thp_page(), if split_huge_page fails, i.e. ret != 0, put_page() is called. See below:
>>>>>
>>>>> static int try_to_split_thp_page(struct page *page)
>>>>> {
>>>>> int ret;
>>>>>
>>>>> lock_page(page);
>>>>> ret = split_huge_page(page);
>>>>> unlock_page(page);
>>>>>
>>>>> if (unlikely(ret))
>>>>> put_page(page);
>>>>> ^^^^^^^^^^^^^^^^^^^^^^^
>>>>> return ret;
>>>>> }
>>>>>
>>>>> Or am I miss something?
>>>> I think you caught a bug in my code, thanks!
>>>>
>>>> How about moving put_page() outside try_to_split_thp_page() ?
>>> If you want to send SIGBUS in the event of thp split fail, it might be required to do so.
>>> I think kill_procs_now() needs extra thp refcnt to do its work.
>> Agreed. I added an boolean to try_to_split_thp_page(),the boolean indicates whether to put_page().
> IMHO, it might be too complicated to add an extra boolean to indicate whether to put_page(). It might be
> more straightforward to always put_page outside try_to_split_thp_page?
Looks okay to me, let's see. Will send out v2 in a while.
thanks,
-jane
> Thanks.
> .
>
next prev parent reply other threads:[~2024-05-10 3:19 UTC|newest]
Thread overview: 26+ messages / expand[flat|nested] mbox.gz Atom feed top
2024-05-01 23:24 [PATCH 0/3] Enhance soft hwpoison handling and injection Jane Chu
2024-05-01 23:24 ` [PATCH 1/3] mm/memory-failure: try to send SIGBUS even if unmap failed Jane Chu
2024-05-07 9:02 ` Oscar Salvador
2024-05-07 17:54 ` Jane Chu
2024-05-08 12:06 ` Oscar Salvador
2024-05-08 16:51 ` Jane Chu
2024-05-08 7:47 ` Miaohe Lin
2024-05-08 16:58 ` Jane Chu
2024-05-09 2:54 ` Miaohe Lin
2024-05-09 16:40 ` Jane Chu
2024-05-01 23:24 ` [PATCH 2/3] mm/madvise: Add MF_ACTION_REQUIRED to madvise(MADV_HWPOISON) Jane Chu
2024-05-05 7:02 ` Miaohe Lin
2024-05-06 19:54 ` Jane Chu
2024-05-08 7:58 ` Miaohe Lin
2024-05-01 23:24 ` [PATCH 3/3] mm/memory-failure: send SIGBUS in the event of thp split fail Jane Chu
2024-05-05 7:00 ` Miaohe Lin
2024-05-06 20:26 ` Jane Chu
2024-05-08 8:08 ` Miaohe Lin
2024-05-08 17:45 ` Jane Chu
2024-05-09 8:30 ` Miaohe Lin
2024-05-09 15:34 ` Jane Chu
2024-05-10 2:59 ` Miaohe Lin
2024-05-10 3:18 ` Jane Chu [this message]
2024-05-08 9:03 ` Miaohe Lin
2024-05-08 16:56 ` Jane Chu
2024-05-09 8:52 ` Miaohe Lin
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=d6b7c197-2b3c-442d-9b0f-ff7db3bd1ed9@oracle.com \
--to=jane.chu@oracle.com \
--cc=akpm@linux-foundation.org \
--cc=linmiaohe@huawei.com \
--cc=linux-kernel@vger.kernel.org \
--cc=linux-mm@kvack.org \
--cc=nao.horiguchi@gmail.com \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).