From: Claudio Imbrenda <imbrenda@linux.ibm.com>
To: David Hildenbrand <david@redhat.com>
Cc: ultrachin@163.com, akpm@linux-foundation.org, linux-mm@kvack.org,
linux-kernel@vger.kernel.org, brookxu.cn@gmail.com,
chen xiaoguang <xiaoggchen@tencent.com>,
zeng jingxiang <linuszeng@tencent.com>,
lu yihui <yihuilu@tencent.com>
Subject: Re: [PATCH] mm: Free per cpu pages async to shorten program exit time
Date: Fri, 8 Oct 2021 11:22:35 +0200 [thread overview]
Message-ID: <20211008112235.1d37f2db@p-imbrenda> (raw)
In-Reply-To: <1db11d75-d2d8-ef71-471a-ddad5c90a733@redhat.com>
On Fri, 8 Oct 2021 11:15:25 +0200
David Hildenbrand <david@redhat.com> wrote:
> On 08.10.21 10:52, Claudio Imbrenda wrote:
> > On Fri, 8 Oct 2021 10:17:50 +0200
> > David Hildenbrand <david@redhat.com> wrote:
> >
> >> On 08.10.21 08:39, ultrachin@163.com wrote:
> >>> From: chen xiaoguang <xiaoggchen@tencent.com>
> >>>
> >>> The exit time is long when program allocated big memory and
> >>> the most time consuming part is free memory which takes 99.9%
> >>> of the total exit time. By using async free we can save 25% of
> >>> exit time.
> >>>
> >>> Signed-off-by: chen xiaoguang <xiaoggchen@tencent.com>
> >>> Signed-off-by: zeng jingxiang <linuszeng@tencent.com>
> >>> Signed-off-by: lu yihui <yihuilu@tencent.com>
> >>
> >> I recently discussed with Claudio if it would be possible to tear down
> >> the process MM deferred, because for some use cases (secure/encrypted
> >> virtualization, very large mmaps) tearing down the page tables is
> >> already the much more expensive operation.
> >>
> >> There is mmdrop_async(), and I wondered if one could reuse that concept
> >> when tearing down a process -- I didn't look into feasibility, however,
> >> so it's just some very rough idea.
> >
> > I have done some experiments by unconditionally replacing mmdrop with
> > mmdrop_async in exit.c and nothing broke, and exit time of large
> > processes was almost instant (with the actual cleanup being performed in
> > background)
> >
> > my approach is probably simpler/cleaner:
> >
> > diff --git a/include/asm-generic/mmu_context.h b/include/asm-generic/mmu_context.h
> > index 91727065bacb..900931a6a105 100644
> > --- a/include/asm-generic/mmu_context.h
> > +++ b/include/asm-generic/mmu_context.h
> > @@ -73,4 +73,8 @@ static inline void deactivate_mm(struct task_struct *tsk,
> > }
> > #endif
> >
> > +#ifndef arch_exit_mm_mmput
> > +#define arch_exit_mm_mmput mmput
> > +#endif
> > +
> > #endif /* __ASM_GENERIC_MMU_CONTEXT_H */
> > diff --git a/kernel/exit.c b/kernel/exit.c
> > index 9a89e7f36acb..604cb9c740fa 100644
> > --- a/kernel/exit.c
> > +++ b/kernel/exit.c
> > @@ -498,7 +498,7 @@ static void exit_mm(void)
> > task_unlock(current);
> > mmap_read_unlock(mm);
> > mm_update_next_owner(mm);
> > - mmput(mm);
> > + arch_exit_mm_mmput(mm);
> > if (test_thread_flag(TIF_MEMDIE))
> > exit_oom_victim();
> > }
> >
> > these are the minimal changes to common code, then each architecture can
> > define their own arch_exit_mm_mmput as they see fit (for example, to free
> > asynchronously only for certain classes of mm, like big ones, VMs, or so).
> >
> > Another option is to simply always replace mmput with mmput_async, which I
> > expect will raise more eyebrows.
>
> Thanks Claudio.
>
> I guess we'd use some heuristic to keep the eyebrows down. Having
> something like
>
> if (should_mput_async_on_exit(mm))
> mmput_async(mm);
> else
> mmput(mm);
>
> whereby the heuristic can optionally consult the arch/config-knobs/...
> doesn't sound too wrong to me if it works.
>
yes, that is one of the possible solutions I had thought of.
although probably the small patch I posted above is even less intrusive
and should hopefully raise even fewer eyebrows, while also leaving the
door open to arch-specific code to do more than just mmput_async, if
needed.
in the end I really do not have any preference, I simply want something
everybody can agree on :)
next prev parent reply other threads:[~2021-10-08 9:22 UTC|newest]
Thread overview: 16+ messages / expand[flat|nested] mbox.gz Atom feed top
[not found] <20211008063933.331989-1-ultrachin@163.com>
2021-10-08 8:17 ` [PATCH] mm: Free per cpu pages async to shorten program exit time David Hildenbrand
2021-10-08 8:52 ` Claudio Imbrenda
2021-10-08 9:15 ` David Hildenbrand
2021-10-08 9:22 ` Claudio Imbrenda [this message]
2021-10-08 9:24 ` David Hildenbrand
2021-10-08 12:38 ` Vlastimil Babka
2021-10-08 12:54 ` Claudio Imbrenda
2021-10-08 12:55 ` David Hildenbrand
2021-10-10 22:42 ` Andrew Morton
2021-10-11 9:28 ` Michal Hocko
2021-10-11 9:40 ` David Hildenbrand
2021-10-11 11:23 ` Michal Hocko
2021-10-13 17:38 ` Daniel Jordan
2021-10-08 11:27 ` kernel test robot
2021-10-08 12:29 ` kernel test robot
2021-10-11 8:20 ` [mm] 3e55b36596: kernel_BUG_at_mm/page_alloc.c kernel test robot
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=20211008112235.1d37f2db@p-imbrenda \
--to=imbrenda@linux.ibm.com \
--cc=akpm@linux-foundation.org \
--cc=brookxu.cn@gmail.com \
--cc=david@redhat.com \
--cc=linuszeng@tencent.com \
--cc=linux-kernel@vger.kernel.org \
--cc=linux-mm@kvack.org \
--cc=ultrachin@163.com \
--cc=xiaoggchen@tencent.com \
--cc=yihuilu@tencent.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).