From: Oleg Nesterov <oleg@redhat.com>
To: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Hugh Dickins <hughd@google.com>, Michal Hocko <mhocko@kernel.org>,
Linux-MM <linux-mm@kvack.org>,
LKML <linux-kernel@vger.kernel.org>,
Andrew Morton <akpm@linux-foundation.org>,
Tim Chen <tim.c.chen@linux.intel.com>,
Michal Hocko <mhocko@suse.com>
Subject: Re: [RFC PATCH] mm: silence soft lockups from unlock_page
Date: Thu, 23 Jul 2020 14:47:50 +0200 [thread overview]
Message-ID: <20200723124749.GA7428@redhat.com> (raw)
In-Reply-To: <CAHk-=whEjnsANEhTA3aqpNLZ3vv7huP7QAmcAEd-GUxm2YMo-Q@mail.gmail.com>
On 07/22, Linus Torvalds wrote:
>
> Comments? Oleg, this should fix the race you talked about too.
Yes.
I still can't convince myself thatI fully understand this patch but I see
nothing really wrong after a quick glance...
> + * We can no longer use 'wait' after we've done the
> + * list_del_init(&wait->entry),
Yes, but see below,
> + * the target may decide it's all done with no
> + * other locking, and 'wait' has been allocated on
> + * the stack of the target.
> */
> - if (test_bit(key->bit_nr, &key->page->flags))
> - return -1;
> + target = wait->private;
> + smp_mb();
>
> - return autoremove_wake_function(wait, mode, sync, key);
> + /*
> + * Ok, we have successfully done what we're waiting for.
> + *
> + * Now unconditionally remove the wait entry, so that the
> + * waiter can use that to see success or not.
> + *
> + * We _really_ should have a "list_del_init_careful()"
> + * to properly pair with an unlocked "list_empty_careful()".
> + */
> + list_del_init(&wait->entry);
> +
> + /*
> + * Theres's another memory barrier in the wakup path, that
> + * makes sure the wakup happens after the above is visible
> + * to the target.
> + */
> + wake_up_state(target, mode);
We can no longer use 'target'. If it was already woken up it can notice
list_empty_careful(), return without taking q->lock, and exit.
Of course, this is purely theoretical... rcu_read_lock() should help
but perhaps we can avoid it somehow?
Say, can't we abuse WQ_FLAG_WOKEN?
wake_page_function:
wait->flags |= WQ_FLAG_WOKEN;
wmb();
autoremove_wake_function(...);
wait_on_page_bit_common:
for (;;) {
set_current_state();
if (wait.flags & WQ_FLAG_WOKEN)
break;
schedule();
}
finish_wait();
rmb();
return wait.flags & WQ_FLAG_WOKEN ? 0 : -EINTR;
Another (cosmetic) problem is that wake_up_state(mode) looks confusing.
It is correct but only because we know that mode == TASK_NORMAL and thus
wake_up_state() can'fail if the target is still blocked.
> + spin_lock_irq(&q->lock);
> + SetPageWaiters(page);
> + if (!trylock_page_bit_common(page, bit_nr, behavior))
> + __add_wait_queue_entry_tail(q, wait);
do we need SetPageWaiters() if trylock() succeeds ?
Oleg.
next prev parent reply other threads:[~2020-07-23 12:48 UTC|newest]
Thread overview: 54+ messages / expand[flat|nested] mbox.gz Atom feed top
2020-07-21 6:32 [RFC PATCH] mm: silence soft lockups from unlock_page Michal Hocko
[not found] ` <FCC3EB2D-9F11-4E9E-88F4-40B2926B35CC@lca.pw>
2020-07-21 11:25 ` Michal Hocko
[not found] ` <664A07B6-DBCD-4520-84F1-241A4E7A339F@lca.pw>
2020-07-21 12:17 ` Michal Hocko
[not found] ` <20200721132343.GA4261@lca.pw>
2020-07-21 13:38 ` Michal Hocko
2020-07-21 14:17 ` Chris Down
2020-07-21 15:00 ` Michal Hocko
2020-07-21 15:33 ` Linus Torvalds
2020-07-21 15:49 ` Michal Hocko
2020-07-22 18:29 ` Linus Torvalds
2020-07-22 21:29 ` Hugh Dickins
2020-07-22 22:10 ` Linus Torvalds
2020-07-22 23:42 ` Linus Torvalds
2020-07-23 0:23 ` Linus Torvalds
2020-07-23 12:47 ` Oleg Nesterov [this message]
2020-07-23 17:32 ` Linus Torvalds
2020-07-23 18:01 ` Oleg Nesterov
2020-07-23 18:22 ` Linus Torvalds
2020-07-23 19:03 ` Linus Torvalds
2020-07-24 14:45 ` Oleg Nesterov
2020-07-23 20:03 ` Linus Torvalds
2020-07-23 23:11 ` Hugh Dickins
2020-07-23 23:43 ` Linus Torvalds
2020-07-24 0:07 ` Hugh Dickins
2020-07-24 0:46 ` Linus Torvalds
2020-07-24 3:45 ` Hugh Dickins
2020-07-24 15:24 ` Oleg Nesterov
2020-07-24 17:32 ` Linus Torvalds
2020-07-24 23:25 ` Linus Torvalds
2020-07-25 2:08 ` Hugh Dickins
2020-07-25 2:46 ` Linus Torvalds
2020-07-25 10:14 ` Oleg Nesterov
2020-07-25 18:48 ` Linus Torvalds
2020-07-25 19:27 ` Oleg Nesterov
2020-07-25 19:51 ` Linus Torvalds
2020-07-26 13:57 ` Oleg Nesterov
2020-07-25 21:19 ` Hugh Dickins
2020-07-26 4:22 ` Hugh Dickins
2020-07-26 20:30 ` Hugh Dickins
2020-07-26 20:41 ` Linus Torvalds
2020-07-26 22:09 ` Hugh Dickins
2020-07-27 19:35 ` Greg KH
2020-08-06 5:46 ` Hugh Dickins
2020-08-18 13:50 ` Greg KH
2020-08-06 5:21 ` Hugh Dickins
2020-08-06 17:07 ` Linus Torvalds
2020-08-06 18:00 ` Matthew Wilcox
2020-08-06 18:32 ` Linus Torvalds
2020-08-07 18:41 ` Hugh Dickins
2020-08-07 19:07 ` Linus Torvalds
2020-08-07 19:35 ` Matthew Wilcox
2020-08-03 13:14 ` Michal Hocko
2020-08-03 17:56 ` Linus Torvalds
2020-07-25 9:39 ` Oleg Nesterov
2020-07-23 8:03 ` Michal Hocko
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=20200723124749.GA7428@redhat.com \
--to=oleg@redhat.com \
--cc=akpm@linux-foundation.org \
--cc=hughd@google.com \
--cc=linux-kernel@vger.kernel.org \
--cc=linux-mm@kvack.org \
--cc=mhocko@kernel.org \
--cc=mhocko@suse.com \
--cc=tim.c.chen@linux.intel.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).