[PATCH 3/3] [ARM] Implement a timer based __delay() loop

[Daniel Walker]

On Tue, 2010-10-05 at 20:36 -0700, Stephen Boyd wrote:
> On 10/05/2010 10:38 AM, Daniel Walker wrote:
> >>  
> >> +#ifdef ARCH_HAS_READ_CURRENT_TIMER
> >> +/*
> >> + * Assuming read_current_timer() is monotonically increasing
> >> + * across calls.
> > You should add more comments here. You assuming that it's monotonic over
> > a 2000us (2ms) period .. I'm not sure this is a good assumption this
> > timer may not be monotonically increasing all the time, what happens
> > then?
> 
> Ok I'll add that it shouldn't wrap more than once within 2000us (or
> should I say 5ms since mdelay uses udelay?). Is that what you're saying
> by it not being monotonically increasing? If a timer isn't increasing
> the tick count it's broken and this call will loop forever. If the timer
> wraps, we'll be safe due to unsigned maths as long as it wraps only once.

I was saying it could be near wrapping when this process starts, then
wrap in the middle. The unsigned math thing should work I think, but add
a comment .. Your basically saying it's OK if it wraps, but the comment
says it needs to be "monotonically increasing" which actually doesn't
appear to be true. It needs to not wrap twice during this process.

> >> +void read_current_timer_delay_loop(unsigned long loops)
> >> +{
> >> +	unsigned long bclock, now;
> >> +
> >> +	read_current_timer(&bclock);
> >> +	do {
> >> +		read_current_timer(&now);
> >> +	} while ((now - bclock) < loops);
> > Have you looked at time_before()/time_after() ?
> 
> Nope. Wouldn't that require an addition though to make it work? I'd
> rather just leave it like it is.

I'm not saying you should change it, I'm saying look at those macro's to
make sure your doing thinks correctly.

Daniel


-- 
Sent by an consultant of the Qualcomm Innovation Center, Inc.
The Qualcomm Innovation Center, Inc. is a member of the Code Aurora
Forum.