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:   Mon, 26 Nov 2018 09:40:42 +0100
From:   Peter Zijlstra <peterz@infradead.org>
To:     Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Cc:     Andrea Parri <andrea.parri@amarulasolutions.com>,
        zhe.he@windriver.com, catalin.marinas@arm.com, tglx@linutronix.de,
        rostedt@goodmis.org, linux-mm@kvack.org,
        linux-kernel@vger.kernel.org, linux-rt-users@vger.kernel.org,
        boqun.feng@gmail.com
Subject: Re: [PATCH v2] kmemleak: Turn kmemleak_lock to raw spinlock on RT
Message-ID: <20181126084042.GK2113@hirez.programming.kicks-ass.net>
References: <1542877459-144382-1-git-send-email-zhe.he@windriver.com>
 <20181123095314.hervxkxtqoixovro@linutronix.de>
 <20181123110226.GA5125@andrea>
 <20181123110611.s2gmd237j7docrxt@linutronix.de>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <20181123110611.s2gmd237j7docrxt@linutronix.de>
User-Agent: Mutt/1.10.1 (2018-07-13)
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

On Fri, Nov 23, 2018 at 12:06:11PM +0100, Sebastian Andrzej Siewior wrote:
> On 2018-11-23 12:02:55 [+0100], Andrea Parri wrote:
> > > is this an RT-only problem? Because mainline should not allow read->read
> > > locking or read->write locking for reader-writer locks. If this only
> > > happens on v4.18 and not on v4.19 then something must have fixed it.
> > 
> > Probably misunderstanding, but I'd say that read->read locking is "the
> > norm"...?
> > 
> > If you don't use qrwlock, readers are also "recursive", in part.,
> > 
> >   P0			P1
> >   read_lock(l)
> > 			write_lock(l)
> >   read_lock(l)
> > 
> > won't block P0 on the second read_lock().  (qrwlock somehow complicate
> > the analysis; IIUC, they are recursive if and only if in_interrupt().).
> 
> ehm, peterz, is that true? My memory on that is that all readers will
> block if there is a writer pending.

Since qrwlock is the more strict, all users should use its semantics.
Just like we cannot 'rely' on the unfairness of some lock
implementations.