I originally reported about the fact that the vc switching is done in the
interrupt response logic which meant that it can disturb a busy loop in
another driver.
I since have seen similar (long) delays due to simple disk activity.
I had my stats program run for a day, and during the morning 'updatedb'
I measured delays as long as 3.9ms. This may be from the NCR PCI scsi
controller logic.
I wonder how much can be done to minimize these delays. Of course, my
original concern with the joystick driver was to do with it WANTING to do
a busy wait for up to 1ms while reading the stick, and I see no way around
this here (except if I could insert a 'reschedule' request in the busy
loop in the driver without breaking the kernel).
My test program is on csc.canberra.edu.au/pub/ise/eyal/misc and is called
timestat-xxxx.tgz (xxxx is version).
-- Regards Eyal Lebedinsky (eyal@ise.canberra.edu.au, eyal@pcug.org.au)