> I was playing with nanosleep() and SCHED_FIFO, precisely with 500ns
> delays on a P100. As I suspected some problem I inspected the code.
>
> nanosleep() calls udelay() which in turn only has 1 microsecond
> resolution.
>
> Therefore I tried to find udelay(). Was a bit tricky as it's in a
> include file. Well I saw that udelay uses delay() in turn, and that
> uses loops_per_sec. All old news you'd say, so far.
>
> Then I wondered who sets loops_per_sec, and I found out that a _lot_
> of code uses loops_per_sec directly to call delay(). As I think
> udelay() is rather new in Linux, I'd suggest to modify the obsolete
> code to use udelay() (unless a delay shorter than 1 microsecond is
> used).
I don't understand why udelay() is inline -- IMHO it would be useful
to make it a standard function using loops_per_sec on [34]86's and
the cycle counter on 586+. Another possibility is to read current
value of the hardware timer instead. This would eliminate all the
udelay alignment problems discussed some time ago.
Have a nice fortnight
-- Martin `MJ' Mares <mj@gts.cz> http://atrey.karlin.mff.cuni.cz/~mj/ Faculty of Math and Physics, Charles University, Prague, Czech Rep., Earth "Quote of the day: '"