Sleeping for <1 jiffie in kernel module (cont'd)

[Sven Schwyn <echo37@xxxxxxxxxxx>]

Ooops, messed up the units. I'm not in the nanosecond but in the 
microsecond range. Here's the mail again with the correct numbers.

--- --- ---

I'm working on a kernel module to drive a PCI card. The module reads a 
clock port (frequency: 1187 Hz) and sends a bit to the data port every 
time the clock port rises. This stream of bits has to flow as long as 
the PCI card is used – for hours maybe.

The module has to check the clock port to detect a rise. This looks 
something like this:

while (...) {
    while (port_get() != 0) { }	// wait for the clock to fall
    while (port_get() == 0) { }	// wait for the clock to rise
    *** send bit ***
}

This effectively locks up the computer as it eats all the CPU cycles do 
do... nothing ;-) Given the clock frequency of 1187Hz, the time between 
two clock port rises is about 842 microseconds. So a better code that's 
supposed NOT to eat up all CPU cycles should look something like this:

while (...) {
    usleep(800);
    while (port_get() == 0) { usleep(20); }	// wait for the clock to 
rise
    *** send bit ***
}

Last but not least: the question!

I can't use usleep() in kernel modules and the resolution of 
schedule_timeout() is way too large. What else can I use instead of 
usleep()?

Many thanks for your help,
-sven


--
Kernelnewbies: Help each other learn about the Linux kernel.
Archive:       http://mail.nl.linux.org/kernelnewbies/
FAQ:           http://kernelnewbies.org/faq/