From: David Rientjes <rientjes@google.com>
To: "Kirill A. Shutemov" <kirill@shutemov.name>
Cc: Andrew Morton <akpm@linux-foundation.org>,
Linus Torvalds <torvalds@linux-foundation.org>,
"Kirill A. Shutemov" <kirill.shutemov@linux.intel.com>,
Andrea Arcangeli <aarcange@redhat.com>,
Suleiman Souhlal <suleiman@google.com>,
stable@vger.kernel.org, linux-kernel@vger.kernel.org,
linux-mm@kvack.org
Subject: Re: [patch for-3.17] mm, thp: fix collapsing of hugepages on madvise
Date: Mon, 6 Oct 2014 02:42:19 -0700 (PDT) [thread overview]
Message-ID: <alpine.DEB.2.02.1410060237580.12568@chino.kir.corp.google.com> (raw)
In-Reply-To: <20141005184115.GA21713@node.dhcp.inet.fi>
On Sun, 5 Oct 2014, Kirill A. Shutemov wrote:
> On Sat, Oct 04, 2014 at 07:48:04PM -0700, David Rientjes wrote:
> > If an anonymous mapping is not allowed to fault thp memory and then
> > madvise(MADV_HUGEPAGE) is used after fault, khugepaged will never
> > collapse this memory into thp memory.
> >
> > This occurs because the madvise(2) handler for thp, hugepage_advise(),
> > clears VM_NOHUGEPAGE on the stack and it isn't stored in vma->vm_flags
> > until the final action of madvise_behavior(). This causes the
> > khugepaged_enter_vma_merge() to be a no-op in hugepage_advise() when the
This should be hugepage_madvise().
> > vma had previously had VM_NOHUGEPAGE set.
> >
> > Fix this by passing the correct vma flags to the khugepaged mm slot
> > handler. There's no chance khugepaged can run on this vma until after
> > madvise_behavior() returns since we hold mm->mmap_sem.
> >
> > It would be possible to clear VM_NOHUGEPAGE directly from vma->vm_flags
> > in hugepage_advise(), but I didn't want to introduce special case
> > behavior into madvise_behavior(). I think it's best to just let it
> > always set vma->vm_flags itself.
> >
> > Cc: <stable@vger.kernel.org>
> > Reported-by: Suleiman Souhlal <suleiman@google.com>
> > Signed-off-by: David Rientjes <rientjes@google.com>
>
> Look like rather complex fix for a not that complex bug.
> What about untested patch below?
>
> From: "Kirill A. Shutemov" <kirill.shutemov@linux.intel.com>
> Date: Sun, 5 Oct 2014 21:22:43 +0300
> Subject: [PATCH] thp: fix registering VMA into khugepaged on
> madvise(MADV_HUGEPAGE)
>
> hugepage_madvise() tries to register VMA into khugepaged with
> khugepaged_enter_vma_merge() on madvise(MADV_HUGEPAGE). Unfortunately
> it's effectevely nop, since khugepaged_enter_vma_merge() rely on
> vma->vm_flags which has not yet updated by the time of
> hugepage_madvise().
>
> Let's move khugepaged_enter_vma_merge() to the end of madvise_behavior().
> Now we also have chance to catch VMAs which become good for THP after
> vma_merge().
>
> Signed-off-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
> ---
> mm/huge_memory.c | 8 +++-----
> mm/madvise.c | 6 ++++++
> 2 files changed, 9 insertions(+), 5 deletions(-)
>
> diff --git a/mm/huge_memory.c b/mm/huge_memory.c
> index f8ffd9412ec5..f84d52158a66 100644
> --- a/mm/huge_memory.c
> +++ b/mm/huge_memory.c
> @@ -1966,12 +1966,10 @@ int hugepage_madvise(struct vm_area_struct *vma,
> *vm_flags &= ~VM_NOHUGEPAGE;
> *vm_flags |= VM_HUGEPAGE;
> /*
> - * If the vma become good for khugepaged to scan,
> - * register it here without waiting a page fault that
> - * may not happen any time soon.
> + * vma->vm_flags is not yet updated here. madvise_behavior()
> + * will take care to register it in khugepaged once flags
> + * updated.
> */
> - if (unlikely(khugepaged_enter_vma_merge(vma)))
> - return -ENOMEM;
> break;
> case MADV_NOHUGEPAGE:
> /*
> diff --git a/mm/madvise.c b/mm/madvise.c
> index 0938b30da4ab..60effd2c5e9c 100644
> --- a/mm/madvise.c
> +++ b/mm/madvise.c
> @@ -128,6 +128,12 @@ success:
> */
> vma->vm_flags = new_flags;
>
> + /*
> + * If the vma become good for khugepaged to scan, register it here
> + * without waiting a page fault that may not happen any time soon.
> + */
> + if (unlikely(khugepaged_enter_vma_merge(vma)))
> + error = -ENOMEM;
> out:
> if (error == -ENOMEM)
> error = -EAGAIN;
I'm pretty sure this won't compile, but I'm also pretty sure it's easy to
come up with an madvise() bit for anon vmas that would cause the BUG_ON()
to trigger for CONFIG_DEBUG_VM and unnecessarily do alloc_mm_slot() for
madvise() calls that aren't MADV_HUGEPAGE with this that go through the
madvise_behavior() path, and for that reason it's probably not as
extendable as we'd like. I can verify this tomorrow if you'd like. This
is the point of the last paragraph of my changelog to isolate all thp
behavior changes to MADV_HUGEPAGE and MADV_NOHUGEPAGE in one place as it's
currently done and not add any special handling in madvise_behavior().
next prev parent reply other threads:[~2014-10-06 9:42 UTC|newest]
Thread overview: 8+ messages / expand[flat|nested] mbox.gz Atom feed top
2014-10-05 2:48 [patch for-3.17] mm, thp: fix collapsing of hugepages on madvise David Rientjes
2014-10-05 17:15 ` Linus Torvalds
2014-10-05 18:41 ` Kirill A. Shutemov
2014-10-05 18:51 ` Linus Torvalds
2014-10-06 9:42 ` David Rientjes [this message]
2014-10-06 15:03 ` Kirill A. Shutemov
2014-10-06 20:53 ` David Rientjes
2014-10-15 21:13 ` [patch resend] " David Rientjes
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=alpine.DEB.2.02.1410060237580.12568@chino.kir.corp.google.com \
--to=rientjes@google.com \
--cc=aarcange@redhat.com \
--cc=akpm@linux-foundation.org \
--cc=kirill.shutemov@linux.intel.com \
--cc=kirill@shutemov.name \
--cc=linux-kernel@vger.kernel.org \
--cc=linux-mm@kvack.org \
--cc=stable@vger.kernel.org \
--cc=suleiman@google.com \
--cc=torvalds@linux-foundation.org \
/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).