Re: delayed switching

[Paul M. <le...@li...>]

On Thu, May 10, 2007 at 02:02:15PM +0900, Hideo Saito wrote:
> I hope to place the code looking at whether needs to reschedule in the
> same function as follows.
> 
> --- arch/sh/kernel/process.c.org	Thu Apr 26 12:08:32 2007
> +++ arch/sh/kernel/process.c	Thu May 10 13:05:44 2007
> +		if (pm_idle) {
> +			while (!need_resched())
> +				pm_idle();
> +		} else if (hlt_counter) {
> +			while (!need_resched())
> +				cpu_relax();
> +		} else {
> +			clear_thread_flag(TIF_POLLING_NRFLAG);
> +			smp_mb__after_clear_bit();
> +			set_bl_bit();
> +			while (!need_resched())
> +				cpu_sleep();
> +			clear_bl_bit();
> +			set_thread_flag(TIF_POLLING_NRFLAG);
> +		}
> +			

No, consider the use case where a platform implements pm_idle() and
depends on certain conditions to be met to determine what sort of
power-save state can be entered. It's still necessary to have the logic
in default_idle() and to have that exported so it can be called back in
to from pm_idle(), should the latter not have any meaningful work it can
do at the time.

In any event, if the previous patches fixes the problems you've been
seeing, I'll queue it up. It works fine with dynamic ticks on SH7785 at
least (both with and without hlt)..