Re: [PATCH] misc. kernel preemption bits

[Andrew Morton <akpm@zip.com.au>]

Robert Love wrote:
> 
> ...
>         - we have a debug check in preempt_schedule that, even
>           on detecting a schedule with irqs disabled, still goes
>           ahead and reschedules.  We should return. (me)
> 

OK, but that warning will still come out of the mess in mm/slab.c.

Reminder:

CPU0:				CPU1:
	local_irq_disable();		resched_task(task on CPU0)
	spin_lock();
	...				p->need_resched = 1;
	spin_unlock(); // reschedules
	local_irq_enable();

There is one code path in slab which fixes this with _raw_spin_unlock()
and a subsequent preempt_disable(), but there are quite a lot of other
code paths which need the same treatment.  Fixing them is messy.

So unless you guys hit me with a brick, I'll create:

#ifdef CONFIG_SMP
#define preempt_disable_irqsave(flags)
	do {
		preempt_disable();
		local_irq_save(flags);
	} while (0)
#define preempt_enable_irqrestore(flags)
	do {
		local_irq_restore(flags);
		preempt_enable();
	} while (0)
#else
#define preempt_disable_irqsave(flags)
	do {
		local_irq_save(flags);
	} while (0)
#define preempt_enable_irqrestore(flags)
	do {
		local_irq_restore(flags);
	} while (0)
#endif

and use that in slab.