Content-Type: text/plain; charset="utf-8"
MIME-Version: 1.0
Content-Transfer-Encoding: 8BIT
To: Mel Gorman <mgorman@suse.de>, Thomas Gleixner <tglx@linutronix.de>,
        Peter Zijlstra <peterz@infradead.org>
From: Chris Mason <chris.mason@fusionio.com>
In-Reply-To: <20131029173814.GH2400@suse.de>
CC: LKML <linux-kernel@vger.kernel.org>
References: <20131029173814.GH2400@suse.de>
Message-ID: <20131029184827.10719.27487@localhost.localdomain>
User-Agent: alot/0.3.4
Subject: Re: [RFC PATCH] futex: Remove requirement for lock_page in get_futex_key
Date: Tue, 29 Oct 2013 14:48:27 -0400
Sender: linux-kernel-owner@vger.kernel.org
Content-Length: 2415
Lines: 52

Quoting Mel Gorman (2013-10-29 13:38:14)
> Thomas Gleixner and Peter Zijlstra discussed off-list that real-time users
> currently have a problem with the page lock being contended for unbounded
> periods of time during futex operations. The three of us discussed the
> possibiltity that the page lock is unnecessary in this case because we are
> not concerned with the usual races with reclaim and page cache updates. For
> anonymous pages, the associated futex object is the mm_struct which does
> not require the page lock. For inodes, we should be able to check under
> RCU read lock if the page mapping is still valid to take a reference to
> the inode.  This just leaves one rare race that requires the page lock
> in the slow path. This patch does not completely eliminate the page lock
> but it should reduce contention in the majority of cases.
> 
> Patch boots and futextest did not explode but I did no comparison
> performance tests. Thomas, do you have details of the workload that
> drove you to examine this problem? Alternatively, can you test it and
> see does it help you? I added Chris to the To list because he mentioned
> that some filesystems might already be doing tricks similar to this
> patch that are worth copying.

Unfortunately, all the special cases I see in the filesystems either
have an inode ref or are trylocking the page to safety.

XFS is a special case because they have their own inode cache, but by my
reading they are still using i_count and free by rcu.

The iput in here is a little tricky:

> 
> +               /* Should be impossible but lets be paranoid for now */
> +               if (WARN_ON(inode->i_mapping != mapping)) {
> +                       rcu_read_unlock();
> +                       iput(inode);
> +                       put_page(page_head);
> +                       goto again;
> +               }
> +

Once you call iput, you add the potential to call the filesystem unlink
operation if i_nlink had gone to zero.  This shouldn't be a problem
since you've dropped the rcu lock, but just for fun I'd move the
put_page up a line.

Or, change it to a BUG_ON instead, it really should be impossible.

-chris

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/