Date:   Fri, 17 Dec 2021 10:00:52 +0100
From:   Sebastian Andrzej Siewior <bigeasy@linutronix.de>
To:     Herbert Xu <herbert@gondor.apana.org.au>
Cc:     Jason@zx2c4.com, linux-kernel@vger.kernel.org, tytso@mit.edu,
        tglx@linutronix.de, peterz@infradead.org
Subject: Re: [PATCH 5/5] random: Defer processing of randomness on PREEMPT_RT.
Message-ID: <YbxRxPdBDMu8KIy6@linutronix.de>
References: <20211207201037.h46573oa5nfj33xq@linutronix.de>
 <20211217022338.GA19411@gondor.apana.org.au>
MIME-Version: 1.0
Content-Type: text/plain; charset=utf-8
Content-Disposition: inline
In-Reply-To: <20211217022338.GA19411@gondor.apana.org.au>
Precedence: bulk

On 2021-12-17 13:23:38 [+1100], Herbert Xu wrote:
> Sebastian Andrzej Siewior <bigeasy@linutronix.de> wrote:
> >
> > Even the IOCTL path must spin with disabled interrupts to avoid dead
> > locks. Therefore it makes no sense if attempt acquire the lock from
> > process or IRQ context. Something like
> >        while (!raw_spin_trylock_irqsave())
> >                cpu_relax()
> 
> What about the TCP socket locking model?
> 
> IOW, the user-space slow path will add itself to a backlog queue
> during contention, and the interrupt fast path will schedule work
> to process any user-space backlog on exit.

I'm sorry, I can't connect the dots here. I was trying to explain that
for the lock in question it is not possible to spin-wait without
disabling interrupts first.

> Cheers,

Sebastian