Wiil,

On Fri, Jul 13, 2012 at 2:27 PM, Will Deacon [off-list ref] wrote:

On Fri, Jul 13, 2012 at 03:16:41AM +0100, Shinya Kuribayashi wrote:

quoted

On 7/13/2012 1:40 AM, Stephen Boyd wrote:

quoted

quoted

quoted

quoted

As a result, actual udelay()s may be toooo long than expected, in
particular udelay()s used between init_current_timer_delay() and
calibrate_delay().  It's unlikely be short, as the frequency of a
counter for read_current_timer is typically slower than CPU frequency.

Surely using udelay before calibrate_delay_loop has been called is a
fundamental error?

Got it.  I'm just not confident about disallowing early use of udelay().

I don't think it's an error. Instead you get a very large delay, similar
to what would happen if you called udelay() before calibrate_delay()
anyway (see the comment in init/main.c above loops_per_jiffy).

Interesting, I didn't notice it was initialised to 4k, so yes I suppose you
could make use of some sort of delay. I don't think it's necessarily `very
large' though -- anything ticking at over ~400KHz with HZ=100 would give you
a smaller delay.

quoted

Thanks, so I'd set up loops_per_jiffy early, along with lpj_fine in
init_current_timer_delay().

That should work, providing you can get a sensible initial estimate for
loops_per_jiffy.

Do you have an updated version of the patch ?

Regards,
Santosh