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;
Date:   Tue, 10 Dec 2019 20:30:11 +0100
From:   Peter Zijlstra <peterz@infradead.org>
To:     David Howells <dhowells@redhat.com>
Cc:     linux-afs@lists.infradead.org, Ingo Molnar <mingo@redhat.com>,
        Will Deacon <will@kernel.org>, linux-fsdevel@vger.kernel.org,
        linux-kernel@vger.kernel.org, Thomas Gleixner <tglx@linutronix.de>,
        Davidlohr Bueso <dave@stgolabs.net>
Subject: Re: [PATCH] rxrpc: Mutexes are unusable from softirq context, so use
 rwsem instead
Message-ID: <20191210193011.GA11802@worktop.programming.kicks-ass.net>
References: <157599917879.6327.69195741890962065.stgit@warthog.procyon.org.uk>
 <20191210191009.GA11457@worktop.programming.kicks-ass.net>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <20191210191009.GA11457@worktop.programming.kicks-ass.net>
User-Agent: Mutt/1.10.1 (2018-07-13)
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

On Tue, Dec 10, 2019 at 08:10:09PM +0100, Peter Zijlstra wrote:
> On Tue, Dec 10, 2019 at 05:32:58PM +0000, David Howells wrote:
> > rxrpc_call::user_mutex is of type struct mutex, but it's required to start
> > off locked on an incoming call as it is being set up in softirq context to
> > prevent sendmsg and recvmsg interfering with it until it is ready.  It is
> > then unlocked in rxrpc_input_packet() to make the call live.
> > 
> > Unfortunately, commit a0855d24fc22d49cdc25664fb224caee16998683
> > ("locking/mutex: Complain upon mutex API misuse in IRQ contexts") causes
> > big warnings to be splashed in dmesg for each a new call that comes in from
> > the server.
> > 
> > It *seems* like it should be okay, since the accept path trylocks the mutex
> > when no one else can see it and drops the mutex before it leaves softirq
> > context.
> > 
> > Fix this by switching to using an rw_semaphore instead as that is permitted
> > to be used in softirq context.
> 
> This really has the very same problem. It just avoids the WARN. We do PI
> boosting for rwsem write side identical to what we do for mutexes.
> 
> I would rather we revert David's patch for now and more carefully
> consider what to do about this.

To clarify (I only just reliazed David is a bit ambiguous here), take
this patch out for now:

  a0855d24fc22 ("locking/mutex: Complain upon mutex API misuse in IRQ contexts")

The RXRPC code has been there for a while... and like I wrote, both
mutex and rwsem have the exact same issue, the rwsem code just doesn't
have a WARN on it.