Re: [patch V2 4/9] softirq: Make softirq control and processing RT aware

From: Thomas Gleixner <tglx@linutronix.de>
To: Peter Zijlstra <peterz@infradead.org>
Cc: LKML <linux-kernel@vger.kernel.org>,
	Frederic Weisbecker <frederic@kernel.org>,
	Paul McKenney <paulmck@kernel.org>,
	Sebastian Andrzej Siewior <bigeasy@linutronix.de>,
	Boqun Feng <boqun.feng@gmail.com>, Ingo Molnar <mingo@kernel.org>,
	Will Deacon <will@kernel.org>
Subject: Re: [patch V2 4/9] softirq: Make softirq control and processing RT aware
Date: Wed, 09 Dec 2020 13:36:54 +0100	[thread overview]
Message-ID: <878sa717h5.fsf@nanos.tec.linutronix.de> (raw)
In-Reply-To: <20201209101102.GJ2414@hirez.programming.kicks-ass.net>

On Wed, Dec 09 2020 at 11:11, Peter Zijlstra wrote:
> On Fri, Dec 04, 2020 at 06:01:55PM +0100, Thomas Gleixner wrote:
>> From: Thomas Gleixner <tglx@linutronix.de>
>> +	/* First entry of a task into a BH disabled section? */
>> +	if (!current->softirq_disable_cnt) {
>> +		if (preemptible()) {
>> +			local_lock(&softirq_ctrl.lock);
>
> AFAICT this significantly changes the locking rules.
>
> Where previously we could do:
>
> 	spin_lock(&ponies)
> 	spin_lock_bh(&foo);
>
> vs
>
> 	spin_lock_bh(&bar);
> 	spin_lock(&ponies)
>
> provided the rest of the code observed: bar -> ponies -> foo
> and never takes ponies from in-softirq.
>
> This is now a genuine deadlock on RT.

I know, but making this work is trying to square the circle.

Any approach we tried before going this way had worse problems than
this particular limitation.

> Also see:
>
>   https://lkml.kernel.org/r/X9CheYjuXWc75Spa@hirez.programming.kicks-ass.net

I'm aware of that and it's fortunately not that many instances of this.

Thanks,

        tglx