From: Alistair Popple <apopple@nvidia.com>
To: Miaohe Lin <linmiaohe@huawei.com>
Cc: akpm@linux-foundation.org, mike.kravetz@oracle.com,
shy828301@gmail.com, willy@infradead.org, ying.huang@intel.com,
ziy@nvidia.com, minchan@kernel.org, ave.hansen@linux.intel.com,
o451686892@gmail.com, almasrymina@google.com,
jhubbard@nvidia.com, rcampbell@nvidia.com, peterx@redhat.com,
naoya.horiguchi@nec.com, mhocko@suse.com, riel@redhat.com,
linux-mm@kvack.org, linux-kernel@vger.kernel.org
Subject: Re: [PATCH 16/16] mm/migration: fix potential pte_unmap on an not mapped pte
Date: Mon, 07 Mar 2022 18:35:51 +1100 [thread overview]
Message-ID: <87y21mrpww.fsf@nvdebian.thelocal> (raw)
In-Reply-To: <20220304093409.25829-17-linmiaohe@huawei.com>
[-- Attachment #1: Type: text/plain, Size: 4648 bytes --]
Miaohe Lin <linmiaohe@huawei.com> writes:
> __migration_entry_wait and migration_entry_wait_on_locked assume pte is
> always mapped from caller. But this is not the case when it's called from
> migration_entry_wait_huge and follow_huge_pmd. And a parameter unmap to
> indicate whether pte needs to be unmapped to fix this issue.
[...]
> diff --git a/mm/filemap.c b/mm/filemap.c
> index 8f7e6088ee2a..18c353d52aae 100644
> --- a/mm/filemap.c
> +++ b/mm/filemap.c
> @@ -1389,6 +1389,7 @@ static inline int folio_wait_bit_common(struct folio *folio, int bit_nr,
> * @ptep: mapped pte pointer. Will return with the ptep unmapped. Only required
> * for pte entries, pass NULL for pmd entries.
> * @ptl: already locked ptl. This function will drop the lock.
> + * @unmap: indicating whether ptep need to be unmapped.
> *
> * Wait for a migration entry referencing the given page to be removed. This is
> * equivalent to put_and_wait_on_page_locked(page, TASK_UNINTERRUPTIBLE) except
> @@ -1402,7 +1403,7 @@ static inline int folio_wait_bit_common(struct folio *folio, int bit_nr,
> * there.
> */
> void migration_entry_wait_on_locked(swp_entry_t entry, pte_t *ptep,
> - spinlock_t *ptl)
> + spinlock_t *ptl, bool unmap)
It's a pity we have to pass unmap all the way down to
migration_entry_wait_on_locked().
> {
> struct wait_page_queue wait_page;
> wait_queue_entry_t *wait = &wait_page.wait;
> @@ -1439,10 +1440,9 @@ void migration_entry_wait_on_locked(swp_entry_t entry, pte_t *ptep,
> * a valid reference to the page, and it must take the ptl to remove the
> * migration entry. So the page is valid until the ptl is dropped.
> */
> - if (ptep)
> - pte_unmap_unlock(ptep, ptl);
> - else
> - spin_unlock(ptl);
> + spin_unlock(ptl);
> + if (unmap && ptep)
> + pte_unmap(ptep);
However we might not have to - afaict this is the only usage of ptep so callers
could do the pte_unmap() prior to calling migration_entry_wait_on_locked(). We
could then remove both the `ptep` and `unmap` parameters. Ie:
migration_entry_wait_on_locked(swp_entry_t entry, spinlock_t *ptl)
This does assume it's ok to change the order of pte unmap/ptl unlock operations.
I'm not terribly familiar with CONFIG_HIGHPTE systems, but it seems like that
should be ok.
- Alistair
>
> for (;;) {
> unsigned int flags;
> diff --git a/mm/hugetlb.c b/mm/hugetlb.c
> index 07668781c246..8088128c25db 100644
> --- a/mm/hugetlb.c
> +++ b/mm/hugetlb.c
> @@ -6713,7 +6713,7 @@ follow_huge_pmd(struct mm_struct *mm, unsigned long address,
> } else {
> if (is_hugetlb_entry_migration(pte)) {
> spin_unlock(ptl);
> - __migration_entry_wait(mm, (pte_t *)pmd, ptl);
> + __migration_entry_wait(mm, (pte_t *)pmd, ptl, false);
> goto retry;
> }
> /*
> diff --git a/mm/migrate.c b/mm/migrate.c
> index 98a968e6f465..5519261f54fe 100644
> --- a/mm/migrate.c
> +++ b/mm/migrate.c
> @@ -281,7 +281,7 @@ void remove_migration_ptes(struct folio *src, struct folio *dst, bool locked)
> * When we return from this function the fault will be retried.
> */
> void __migration_entry_wait(struct mm_struct *mm, pte_t *ptep,
> - spinlock_t *ptl)
> + spinlock_t *ptl, bool unmap)
> {
> pte_t pte;
> swp_entry_t entry;
> @@ -295,10 +295,12 @@ void __migration_entry_wait(struct mm_struct *mm, pte_t *ptep,
> if (!is_migration_entry(entry))
> goto out;
>
> - migration_entry_wait_on_locked(entry, ptep, ptl);
> + migration_entry_wait_on_locked(entry, ptep, ptl, unmap);
> return;
> out:
> - pte_unmap_unlock(ptep, ptl);
> + spin_unlock(ptl);
> + if (unmap)
> + pte_unmap(ptep);
> }
>
> void migration_entry_wait(struct mm_struct *mm, pmd_t *pmd,
> @@ -306,14 +308,14 @@ void migration_entry_wait(struct mm_struct *mm, pmd_t *pmd,
> {
> spinlock_t *ptl = pte_lockptr(mm, pmd);
> pte_t *ptep = pte_offset_map(pmd, address);
> - __migration_entry_wait(mm, ptep, ptl);
> + __migration_entry_wait(mm, ptep, ptl, true);
> }
>
> void migration_entry_wait_huge(struct vm_area_struct *vma,
> struct mm_struct *mm, pte_t *pte)
> {
> spinlock_t *ptl = huge_pte_lockptr(hstate_vma(vma), mm, pte);
> - __migration_entry_wait(mm, pte, ptl);
> + __migration_entry_wait(mm, pte, ptl, false);
> }
>
> #ifdef CONFIG_ARCH_ENABLE_THP_MIGRATION
> @@ -324,7 +326,7 @@ void pmd_migration_entry_wait(struct mm_struct *mm, pmd_t *pmd)
> ptl = pmd_lock(mm, pmd);
> if (!is_pmd_migration_entry(*pmd))
> goto unlock;
> - migration_entry_wait_on_locked(pmd_to_swp_entry(*pmd), NULL, ptl);
> + migration_entry_wait_on_locked(pmd_to_swp_entry(*pmd), NULL, ptl, false);
> return;
> unlock:
> spin_unlock(ptl);
next prev parent reply other threads:[~2022-03-07 11:24 UTC|newest]
Thread overview: 73+ messages / expand[flat|nested] mbox.gz Atom feed top
2022-03-04 9:33 [PATCH 00/16] A few cleanup and fixup patches for migration Miaohe Lin
2022-03-04 9:33 ` [PATCH 01/16] mm/migration: remove unneeded local variable mapping_locked Miaohe Lin
2022-03-04 13:48 ` Muchun Song
2022-03-07 2:00 ` Huang, Ying
2022-03-08 11:41 ` Miaohe Lin
2022-03-04 9:33 ` [PATCH 02/16] mm/migration: remove unneeded out label Miaohe Lin
2022-03-04 12:12 ` Muchun Song
2022-03-07 2:03 ` Huang, Ying
2022-03-08 11:44 ` Miaohe Lin
2022-03-04 9:33 ` [PATCH 03/16] mm/migration: remove unneeded local variable page_lru Miaohe Lin
2022-03-07 10:58 ` Alistair Popple
2022-03-08 11:29 ` Miaohe Lin
2022-03-04 9:33 ` [PATCH 04/16] mm/migration: reduce the rcu lock duration Miaohe Lin
2022-03-04 12:16 ` Muchun Song
2022-03-07 2:32 ` Huang, Ying
2022-03-08 12:09 ` Miaohe Lin
2022-03-09 1:02 ` Huang, Ying
2022-03-09 8:28 ` Miaohe Lin
2022-03-04 9:33 ` [PATCH 05/16] mm/migration: fix the confusing PageTransHuge check Miaohe Lin
2022-03-04 12:20 ` Muchun Song
2022-03-04 9:33 ` [PATCH 06/16] mm/migration: use helper function vma_lookup() in add_page_for_migration Miaohe Lin
2022-03-04 9:34 ` [PATCH 07/16] mm/migration: use helper macro min_t in do_pages_stat Miaohe Lin
2022-03-04 13:51 ` Muchun Song
2022-03-07 1:14 ` Andrew Morton
2022-03-07 11:51 ` Miaohe Lin
2022-03-04 9:34 ` [PATCH 08/16] mm/migration: avoid unneeded nodemask_t initialization Miaohe Lin
2022-03-04 13:57 ` Muchun Song
2022-03-07 2:31 ` Baolin Wang
2022-03-04 9:34 ` [PATCH 09/16] mm/migration: remove some duplicated codes in migrate_pages Miaohe Lin
2022-03-04 15:16 ` Zi Yan
2022-03-07 1:44 ` Baolin Wang
2022-03-04 9:34 ` [PATCH 10/16] mm/migration: remove PG_writeback handle in folio_migrate_flags Miaohe Lin
2022-03-07 1:21 ` Andrew Morton
2022-03-07 12:44 ` Miaohe Lin
2022-03-04 9:34 ` [PATCH 11/16] mm/migration: remove unneeded lock page and PageMovable check Miaohe Lin
2022-03-04 9:34 ` [PATCH 12/16] mm/migration: fix potential page refcounts leak in migrate_pages Miaohe Lin
2022-03-04 15:21 ` Zi Yan
2022-03-07 1:57 ` Baolin Wang
2022-03-07 5:02 ` Huang, Ying
2022-03-07 6:00 ` Baolin Wang
2022-03-07 12:03 ` Miaohe Lin
2022-03-07 12:01 ` Miaohe Lin
2022-03-07 5:01 ` Huang, Ying
2022-03-07 12:11 ` Miaohe Lin
2022-03-04 9:34 ` [PATCH 13/16] mm/migration: return errno when isolate_huge_page failed Miaohe Lin
2022-03-05 2:23 ` Muchun Song
2022-03-07 11:46 ` Miaohe Lin
2022-03-07 2:14 ` Baolin Wang
2022-03-07 12:20 ` Miaohe Lin
2022-03-08 1:32 ` Baolin Wang
2022-03-08 6:34 ` Miaohe Lin
2022-03-07 5:07 ` Huang, Ying
2022-03-08 12:12 ` Miaohe Lin
2022-03-09 1:00 ` Huang, Ying
2022-03-09 8:29 ` Miaohe Lin
2022-03-04 9:34 ` [PATCH 14/16] mm/migration: fix potential invalid node access for reclaim-based migration Miaohe Lin
2022-03-07 2:25 ` Baolin Wang
2022-03-07 5:14 ` Huang, Ying
2022-03-07 7:04 ` Baolin Wang
2022-03-08 11:46 ` Miaohe Lin
2022-03-07 5:14 ` Huang, Ying
2022-03-04 9:34 ` [PATCH 15/16] mm/migration: fix possible do_pages_stat_array racing with memory offline Miaohe Lin
2022-03-07 5:21 ` Huang, Ying
2022-03-07 7:01 ` Muchun Song
2022-03-07 7:42 ` Huang, Ying
2022-03-08 11:33 ` Miaohe Lin
2022-03-04 9:34 ` [PATCH 16/16] mm/migration: fix potential pte_unmap on an not mapped pte Miaohe Lin
2022-03-07 5:37 ` Huang, Ying
2022-03-08 12:19 ` Miaohe Lin
2022-03-09 0:56 ` Huang, Ying
2022-03-09 8:48 ` Miaohe Lin
2022-03-07 7:35 ` Alistair Popple [this message]
2022-03-08 11:55 ` 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=87y21mrpww.fsf@nvdebian.thelocal \
--to=apopple@nvidia.com \
--cc=akpm@linux-foundation.org \
--cc=almasrymina@google.com \
--cc=ave.hansen@linux.intel.com \
--cc=jhubbard@nvidia.com \
--cc=linmiaohe@huawei.com \
--cc=linux-kernel@vger.kernel.org \
--cc=linux-mm@kvack.org \
--cc=mhocko@suse.com \
--cc=mike.kravetz@oracle.com \
--cc=minchan@kernel.org \
--cc=naoya.horiguchi@nec.com \
--cc=o451686892@gmail.com \
--cc=peterx@redhat.com \
--cc=rcampbell@nvidia.com \
--cc=riel@redhat.com \
--cc=shy828301@gmail.com \
--cc=willy@infradead.org \
--cc=ying.huang@intel.com \
--cc=ziy@nvidia.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).