From: Mike Kravetz <mike.kravetz@oracle.com>
To: Mina Almasry <almasrymina@google.com>
Cc: Axel Rasmussen <axelrasmussen@google.com>,
Peter Xu <peterx@redhat.com>, Linux-MM <linux-mm@kvack.org>,
Andrew Morton <akpm@linux-foundation.org>,
open list <linux-kernel@vger.kernel.org>
Subject: Re: [PATCH] mm, hugetlb: fix resv_huge_pages underflow on UFFDIO_COPY
Date: Thu, 20 May 2021 13:00:25 -0700 [thread overview]
Message-ID: <61c78897-27a2-768b-f4fe-04e24b617ab6@oracle.com> (raw)
In-Reply-To: <CAHS8izNECN6U4KDLS-jrtni6WMszK6rF1XjcWwwGAByb3pEr8A@mail.gmail.com>
On 5/20/21 12:21 PM, Mina Almasry wrote:
> On Thu, May 20, 2021 at 12:18 PM Mina Almasry <almasrymina@google.com> wrote:
>>
>> On Thu, May 13, 2021 at 5:14 PM Mike Kravetz <mike.kravetz@oracle.com> wrote:
>>>
>>> How about this approach?
>>> - Keep the check for hugetlbfs_pagecache_present in hugetlb_mcopy_atomic_pte
>>> that you added. That will catch the race where the page was added to
>>> the cache before entering the routine.
>>> - With the above check in place, we only need to worry about the case
>>> where copy_huge_page_from_user fails and we must drop locks. In this
>>> case we:
>>> - Free the page previously allocated.
>>> - Allocate a 'temporary' huge page without consuming reserves. I'm
>>> thinking of something similar to page migration.
>>> - Drop the locks and let the copy_huge_page_from_user be done to the
>>> temporary page.
>>> - When reentering hugetlb_mcopy_atomic_pte after dropping locks (the
>>> *pagep case) we need to once again check
>>> hugetlbfs_pagecache_present.
>>> - We then try to allocate the huge page which will consume the
>>> reserve. If successful, copy contents of temporary page to newly
>>> allocated page. Free temporary page.
>>>
>>> There may be issues with this, and I have not given it deep thought. It
>>> does abuse the temporary huge page concept, but perhaps no more than
>>> page migration. Things do slow down if the extra page allocation and
>>> copy is required, but that would only be the case if copy_huge_page_from_user
>>> needs to be done without locks. Not sure, but hoping that is rare.
>>
>> Just following up this a bit: I've implemented this approach locally,
>> and with it it's passing the test as-is. When I hack the code such
>> that the copy in hugetlb_mcopy_atomic_pte() always fails, I run into
>> this edge case, which causes resv_huge_pages to underflow again (this
>> time permemantly):
>>
>> - hugetlb_no_page() is called on an index and a page is allocated and
>> inserted into the cache consuming the reservation.
>> - remove_huge_page() is called on this index and the page is removed from cache.
>> - hugetlb_mcopy_atomic_pte() is called on this index, we do not find
>> the page in the cache and we trigger this code patch and the copy
>> fails.
>> - The allocations in this code path seem to double consume the
>> reservation and resv_huge_pages underflows.
>>
>> I'm looking at this edge case to understand why a prior
>> remove_huge_page() causes my code to underflow resv_huge_pages.
>>
>
> I should also mention, without a prior remove_huge_page() this code
> path works fine, so it seems the fact that the reservation is consumed
> before causes trouble, but I'm not sure why yet.
>
Hi Mina,
How about quickly posting the code? I may be able to provide more
suggestions if I can see the actual code.
--
Mike Kravetz
next prev parent reply other threads:[~2021-05-20 20:00 UTC|newest]
Thread overview: 19+ messages / expand[flat|nested] mbox.gz Atom feed top
2021-05-13 23:43 [PATCH] mm, hugetlb: fix resv_huge_pages underflow on UFFDIO_COPY Mina Almasry
2021-05-13 23:49 ` Mina Almasry
2021-05-14 0:14 ` Mike Kravetz
2021-05-14 0:23 ` Mina Almasry
2021-05-14 4:02 ` Mike Kravetz
2021-05-14 12:31 ` Peter Xu
2021-05-14 17:56 ` Mike Kravetz
2021-05-14 18:30 ` Axel Rasmussen
2021-05-14 19:16 ` Peter Xu
2021-05-20 19:18 ` Mina Almasry
2021-05-20 19:21 ` Mina Almasry
2021-05-20 20:00 ` Mike Kravetz [this message]
2021-05-20 20:31 ` Mina Almasry
2021-05-21 2:05 ` Mina Almasry
-- strict thread matches above, loose matches on Subject: below --
2021-05-12 3:06 resv_huge_page underflow with userfaultfd test Mike Kravetz
[not found] ` <20210512065813.89270-1-almasrymina@google.com>
2021-05-12 7:44 ` [PATCH] mm, hugetlb: fix resv_huge_pages underflow on UFFDIO_COPY Mina Almasry
[not found] ` <CAJHvVch0ZMapPVEc0Ge5V4KDgNDNhECbqwDi0y9XxsxFXQZ-gg@mail.gmail.com>
[not found] ` <c455d241-11f6-95a6-eb29-0ddd94eedbd7@oracle.com>
2021-05-12 19:42 ` Mina Almasry
2021-05-12 20:14 ` Peter Xu
2021-05-12 21:31 ` Mike Kravetz
2021-05-12 21:52 ` Mina Almasry
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=61c78897-27a2-768b-f4fe-04e24b617ab6@oracle.com \
--to=mike.kravetz@oracle.com \
--cc=akpm@linux-foundation.org \
--cc=almasrymina@google.com \
--cc=axelrasmussen@google.com \
--cc=linux-kernel@vger.kernel.org \
--cc=linux-mm@kvack.org \
--cc=peterx@redhat.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 NNTP newsgroup(s).