Re: Is wake_up safe on 2.4?

[Parag Warudkar <kernel-stuff@comcast.net>]

On Sunday 29 May 2005 10:28, manoj.sharma@wipro.com wrote:
> Is it safe to use wake_up() in 2.4 inside an interrupt handler or in a
> spin lock region?
>
> wake_up() uses reshedule_idle() to find an idle cpu for the woken up
> task. If it doesn't find any, it checks the current running tasks on all
> CPUs and uses goodness value to pick up the best cpu to schedule the
> woken up task.  Isn't possible to preempt the current task where
> reschedule_idle() is running?
>
> There are plenty of instances in the kernel (2.4) where wake_up() is
> being used inside interrupt handler or after taking spin locks. If it
> can preempt the task calling wake_up(), how safe is it to use then?

Looking at the code - On SMP, the woken up task will be always scheduled on a 
CPU other than the one executing the caller, in which case caller of 
wake_up() is not pre-empted. On UP, it just sets need_resched (different than 
calling schedule() ) in which case the caller of wake_up() will be 
rescheduled only when appropriate, when it is safe to preempt it.