Re: RCU question

[George Anzinger <george@mvista.com>]

Manfred Spraul wrote:
> George Anzinger wrote:
> 
>>
>> The "normal" idle loop just looks at the need_resched flag and goes 
>> right back to the hlt,
> 
> 
> That's the problem: If a the tasklet does a wakeup then the reschedule 
> is delayed until the next interrupt. 

Not so.  On the interrupt that runs the tasklet, on the way out via entry.S, the 
need_resched flag is checked and acted on.  Thus the task switch is done prio to 
getting back to the hlt.

> Testing need_resched and executing 
> hlt must be atomic, but it isn't - NMIs break the atomicity.

Actually this is not required, especially if preemption is turned on.

> Not a big deal, except if someone implements a tickless kernel. 

Well, it is not tickless, but VST that I am working on :).  The notion is to 
turn off the ticks when in idle and there are not time events in the list.

I think
> we can ignore it for now [or was the thread started by someone who 
> want's to disable the hardware timer when the system is really idle?]

Yep, me!  But still, I keep a timer around to exit, it is just way more than a 
tick later (depending on what the next entry in the time list needs).
> 

-- 
George Anzinger   george@mvista.com
High-res-timers:  http://sourceforge.net/projects/high-res-timers/