Received-SPF: pass (google.com: best guess record for domain of linux-kernel-owner@vger.kernel.org designates 209.132.180.67 as permitted sender) client-ip=209.132.180.67;
From:   NeilBrown <neilb@suse.com>
To:     Jeff Layton <jlayton@kernel.org>,
        Alexander Viro <viro@zeniv.linux.org.uk>
Date:   Tue, 14 Aug 2018 13:56:51 +1000
Subject: [PATCH 0/5 v2] locks: avoid thundering-herd wake-ups
Cc:     "J. Bruce Fields" <bfields@fieldses.org>,
        Martin Wilck <mwilck@suse.de>, linux-fsdevel@vger.kernel.org,
        Frank Filz <ffilzlnx@mindspring.com>,
        linux-kernel@vger.kernel.org
Message-ID: <153421852728.24426.2111161640156686201.stgit@noble>
User-Agent: StGit/0.17.1-dirty
MIME-Version: 1.0
Content-Type: text/plain; charset="utf-8"
Content-Transfer-Encoding: 7bit
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk


V2, which added wake_non_conflicts() was more broken than V1 - as
Bruce explained there is no transitivity in the blocking relation
between locks.
So this series takes a simpler approach.
It still attached waiters between other waiters as necessary to ensure
that:
  - a waiter is blocked by it's parent (fl->blocker) and all further
    ancestors, and
  - the list of waiters on fl_blocked are mutually non-conflicting.

When a lock (the root of a tree of requests) is released, only its
immediate children (fl_blocked) are woken.
When any lock is woken (either because its fl_blocker was released
to due to a signal or similar) it with either:
 - be granted
 - be aborted
 - be re-queued beneath some other lock.

In the first case tree of blocked locks is moved across to the newly
created lock, and the invariants still hold.
In the order two cases, the tree or blocked waiters are all detached
and woken.

Note that this series has not received much testing yet.

Original description:
If you have a many-core machine, and have many threads all wanting to
briefly lock a give file (udev is known to do this), you can get quite
poor performance.

When one thread releases a lock, it wakes up all other threads that
are waiting (classic thundering-herd) - one will get the lock and the
others go to sleep.
When you have few cores, this is not very noticeable: by the time the
4th or 5th thread gets enough CPU time to try to claim the lock, the
earlier threads have claimed it, done what was needed, and released.
With 50+ cores, the contention can easily be measured.

This patchset creates a tree of pending lock request in which siblings
don't conflict and each lock request does conflict with its parent.
When a lock is released, only requests which don't conflict with each
other a woken.

Testing shows that lock-acquisitions-per-second is now fairly stable even
as number of contending process goes to 1000.  Without this patch,
locks-per-second drops off steeply after a few 10s of processes.

There is a small cost to this extra complexity.
At 20 processes running a particular test on 72 cores, the lock
acquisitions per second drops from 1.8 million to 1.4 million with
this patch.  For 100 processes, this patch still provides 1.4 million
while without this patch there are about 700,000.

NeilBrown

---

NeilBrown (5):
      fs/locks: rename some lists and pointers.
      fs/locks: split out __locks_wake_up_blocks().
      fs/locks: allow a lock request to block other requests.
      fs/locks: change all *_conflict() functions to return bool.
      fs/locks: create a tree of dependent requests.


 fs/cifs/file.c                  |    2 -
 fs/locks.c                      |  156 ++++++++++++++++++++++++++-------------
 include/linux/fs.h              |    7 +-
 include/trace/events/filelock.h |   16 ++--
 4 files changed, 119 insertions(+), 62 deletions(-)

--
Signature