Re: [PATCH v8 4/5] locking/qspinlock: Introduce starvation avoidance into CNA

[Waiman Long]

On 1/24/20 1:51 PM, Waiman Long wrote:
>> You can use the in_task() macro in include/linux/preempt.h. This is
>> just a percpu preempt_count read and test. If in_task() is false, it
>> is in a {soft|hard}irq or nmi context. If it is true, you can check
>> the rt_task() macro to see if it is an RT task. That will access to
>> the current task structure. So it may cost a little bit more if you
>> want to handle the RT task the same way.
>>
> We may not need to do that for softIRQ context. If that is the case, you
> can use in_irq() which checks for hardirq and nmi only. Peter, what is
> your thought on that?

In second thought, we should do that for softIRQ as well. Also, we may
want to also check if irqs_disabled() is true as well by calls like
spin_lock_irq() or spin_lock_irqsave(). We do not want to unnecessarily
prolong the irq off period.

Cheers,
Longman