linux-kernel - Re: [PATCH] sched: schedule_raw_spin_unlock() and schedule_spin

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite for Android: free password hash cracker in your pocket

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Date:	Mon, 17 Jun 2013 10:29:02 -0400
From:	Steven Rostedt <rostedt@...dmis.org>
To:	Kirill Tkhai <tkhai@...dex.ru>
Cc:	"linux-kernel@...r.kernel.org" <linux-kernel@...r.kernel.org>,
	Ingo Molnar <mingo@...hat.com>,
	Peter Zijlstra <peterz@...radead.org>
Subject: Re: [PATCH] sched: schedule_raw_spin_unlock() and
 schedule_spin_unlock()

On Fri, 2013-06-14 at 18:40 +0400, Kirill Tkhai wrote:
>  
> diff --git a/kernel/sched/core.c b/kernel/sched/core.c
> index 58453b8..381e493 100644
> --- a/kernel/sched/core.c
> +++ b/kernel/sched/core.c
> @@ -3125,6 +3125,30 @@ asmlinkage void __sched preempt_schedule_irq(void)
>  	exception_exit(prev_state);
>  }
>  
> +/*
> + * schedule_raw_spin_unlock - unlock raw_spinlock and call schedule()
> + * 
> + * Should be used instead of the constructions
> + *   1) raw_spin_unlock_irq(lock);
> + *      schedule();
> + *   or
> + *   2) raw_spin_unlock_irqrestore(lock, flags);
> + *      schedule();

Is there a place that does #2? If interrupts were disabled, the flags
would keep them disabled and that would not be good when calling
schedule.

> + * where they have to be.
> + *
> + * It helps to prevent excess preempt_schedule() during the unlocking,
> + * which can be called on preemptible kernel.
> + * Returns with irqs enabled.
> + */
> +void __sched schedule_raw_spin_unlock(raw_spinlock_t *lock)
> +{

I agree with the idea of adding this, but I don't like this
implementation. Also, if this is to enable interrupts, the name must
represent that:

  schedule_raw_spin_unlock_irq()

You can't just enable them if they were not disabled. That will break
things like lockdep.

> +	preempt_disable();
> +	raw_spin_unlock_irq(lock);
> +	sched_preempt_enable_no_resched();

This is the easy way of implementing this, but it does add a slight
overhead here. Adding and subtracting preempt count just to prevent the
disable does add a bit more computation. I've done this in the tracing
code, but its within the tracing and in an unlikely path. The overhead
is not common. But I can see this being in a fast path and not something
that we want to add overhead to.

The ideal solution is not the easy one. It is to introduce a real
schedule_raw_spin_unlock() that is a copy of raw_spin_unlock() that does
not call preempt_enable, but calls preempt_enable_no_resched() instead,
and then does the schedule.

-- Steve

> +	schedule();
> +}
> +EXPORT_SYMBOL(schedule_raw_spin_unlock);
> +
>  #endif /* CONFIG_PREEMPT */
>  
>  int default_wake_function(wait_queue_t *curr, unsigned mode, int wake_flags,

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/