On Fri, 26 Mar 2004, Jeff Garzik wrote:

Sridhar Samudrala wrote:

quoted

On Fri, 26 Mar 2004, Edgar Toernig wrote:

quoted

Sridhar Samudrala wrote:

quoted

The following patch to 2.6.5-rc2 consolidates 6 different implementations
of msecs to jiffies and 3 different implementation of jiffies to msecs.
All of them now use the generic msecs_to_jiffies() and jiffies_to_msecs()
that are added to include/linux/time.h
[...]
-#define MSECS(ms)  (((ms)*HZ/1000)+1)
-return (((ms)*HZ+999)/1000);
+return (msecs / 1000) * HZ + (msecs % 1000) * HZ / 1000;

Did you check that all users of the new version will work correctly
with your rounding?  Explicit round-up of delays is often required,
especially when talking to hardware...

I don't see any issues with the 2.6 default HZ value of 1000 as they become
no-ops and there is no need for any rounding.
I guess you are referring to cases when HZ < 1000(ex: 100) and msecs is
less than 10. In those cases, the new version returns 0, whereas some of the
older versions return 1.

We'll definitely want to return 1 rather than zero, for the uses in my
drivers, at least...

I have modified msecs_to_jiffies() to do the proper rounding when HZ=100.
Do these work better?

static inline unsigned long msecs_to_jiffies(unsigned long msecs)
{
#if 1000 % HZ == 0
        return (msecs + ((1000 / HZ) - 1)) / (1000 / HZ);
#elif HZ % 1000 == 0
        return msecs * (HZ / 1000);
#else
        return (msecs / 1000) * HZ + (msecs % 1000) * HZ / 1000;
#endif
}

static inline unsigned long jiffies_to_msecs(unsigned long jiffs)
{
#if 1000 % HZ == 0
        return jiffs * (1000 / HZ);
#elif HZ % 1000 == 0
        return jiffs / (HZ / 1000);
#else
        return (jiffs / HZ) * 1000 + (jiffs % HZ) * 1000 / HZ;
#endif
}

Thanks
Sridhar