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;
MIME-Version: 1.0
References: <20200208193445.27421-1-ira.weiny@intel.com> <20200208193445.27421-8-ira.weiny@intel.com>
 <20200211080035.GI10776@dread.disaster.area> <20200211201430.GE12866@iweiny-DESK2.sc.intel.com>
In-Reply-To: <20200211201430.GE12866@iweiny-DESK2.sc.intel.com>
From:   Dan Williams <dan.j.williams@intel.com>
Date:   Tue, 11 Feb 2020 12:59:40 -0800
Message-ID: <CAPcyv4hLTg+1dx2bE2S7sLRK17EP_gT5kutwhyUfVB7Ad8khgw@mail.gmail.com>
Subject: Re: [PATCH v3 07/12] fs: Add locking for a dynamic DAX state
To:     Ira Weiny <ira.weiny@intel.com>
Cc:     Dave Chinner <david@fromorbit.com>,
        Linux Kernel Mailing List <linux-kernel@vger.kernel.org>,
        Alexander Viro <viro@zeniv.linux.org.uk>,
        "Darrick J. Wong" <darrick.wong@oracle.com>,
        Christoph Hellwig <hch@lst.de>,
        "Theodore Y. Ts'o" <tytso@mit.edu>, Jan Kara <jack@suse.cz>,
        linux-ext4 <linux-ext4@vger.kernel.org>,
        linux-xfs <linux-xfs@vger.kernel.org>,
        linux-fsdevel <linux-fsdevel@vger.kernel.org>
Content-Type: text/plain; charset="UTF-8"
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

On Tue, Feb 11, 2020 at 12:14 PM Ira Weiny <ira.weiny@intel.com> wrote:
>
> On Tue, Feb 11, 2020 at 07:00:35PM +1100, Dave Chinner wrote:
> > On Sat, Feb 08, 2020 at 11:34:40AM -0800, ira.weiny@intel.com wrote:
> > > From: Ira Weiny <ira.weiny@intel.com>
> > >
> > > DAX requires special address space operations but many other functions
> > > check the IS_DAX() state.
> > >
> > > While DAX is a property of the inode we perfer a lock at the super block
> > > level because of the overhead of a rwsem within the inode.
> > >
> > > Define a vfs per superblock percpu rs semaphore to lock the DAX state
> >
> > ????
>
> oops...  I must have forgotten to update the commit message when I did the
> global RW sem.  I implemented the per-SB, percpu rwsem first but it was
> suggested that the percpu nature of the lock combined with the anticipated
> infrequent use of the write side made using a global easier.
>
> But before I proceed on V4 I'd like to get consensus on which of the 2 locking
> models to go with.
>
>         1) percpu per superblock lock
>         2) per inode rwsem
>
> Depending on my mood I can convince myself of both being performant but the
> percpu is very attractive because I don't anticipate many changes of state
> during run time.  OTOH concurrent threads accessing the same file at run time
> may also be low so there is likely to be little read contention across CPU's on
> the per-inode lock?
>
> Opinions?

As one who thought a global lock would be reasonable relative to how
often dax address_space_operations are swapped out (on the order of
taking cgroup_threadgroup_rwsem for write), I think a per-superblock
lock is also an ok starting point. We can always go to finer grained
locking in the future if we see evidence that the benefits of percpu
are lost to stopping the world at the superblock level.