[Dynamicpower-devel] RE: patch for omap for dpm-2.6.11 (change read_mputimer1() to sched_clock())

[Amit C. <Ami...@mi...>]

Hi Todd,

The sched_clock fucntion returns time in nanoseconds. It is defined for =
both
32K timer and MPU timer. Please let me know whether the the bitwise =
negation
should be removed because I didn't see any "0 - omap_mpu_timer_read(0)".

If we want the resolution in microseconds or milliseconds, we can =
multiply
the value returned by sched_clock(). Please let me know.

From arch/arm/kernel/time.c

unsigned long long __attribute__((weak)) sched_clock(void)
{
        return (unsigned long long)jiffies * (1000000000 / HZ);
}


Regards,
Amit

-----Original Message-----
From: Todd Poynor [mailto:tp...@mv...]
Sent: Thursday, March 17, 2005 8:52 AM
To: Amit Choudhary
Cc: dyn...@li...
Subject: Re: patch for omap for dpm-2.6.11 (change read_mputimer1() to
sched_clock())


Amit Choudhary wrote:

> If I have not missed something then the call to read_mputimer1() =
should be
> modified to sched_clock().
> I am attaching a patch to change the definition of dpm_md_time()
> [include/asm-arm/arch-omap/dpm.h] to take care of that.

Hello Amit -- Sounds right; the copy of MPU timer1 reading code into DPM =

was a temporary mod while sched_clock was jiffie-based instead of a=20
microsecond-level timer in the OMAP tree (if I remember correctly).  Now =

that it's (normally) based on MPU timer 1 we should be able to kill most =

or all of the code for that in the DPM patch (the definitions of MPU=20
timer register addresses and bits, etc.).

It's slightly more complicated than I've been able to keep up with=20
recently, for example, use of MPU timer 1 is dependent on=20
CONFIG_OMAP_MPU_TIMER, and there's also CONFIG_OMAP_32K_TIMER available=20
on most OMAPs, which seems not to provide a sched_clock() (I suppose we=20
could make DPM timing dependent on CONFIG_OMAP_MPU_TIMER).

Since all platforms supported so far by DPM on sourceforge now have=20
microsecond-resolution sched_clock(), we can actually drop dpm_md_time() =

entirely and just use sched_clock().

A couple brief comments on your patch:

+extern unsigned long long sched_clock(void);

This should come from include/linux/sched.h.

+#define dpm_md_time() ~sched_clock()

I believe the bitwise negation (in order to convert the MPU timer value=20
to an up counter) can be dropped, sched_clock is already doing "0 -=20
omap_mpu_timer_read(0)" for the same purpose.

Feel free to send an updated patch or one way or another I'll get to=20
this soon, thanks,

--=20
Todd