🚀 go-pugleaf

(this should probably be put in the FAQ)

why does an unused serial port consume cpu?

because you've got a (probably fairly long) unshielded serial cable
attached to it, with either nothing at the other end or a powered-off
device at the other end.  EE's call this an antenna.  it's probably
picking up most of the radio stations in your area, which the serial
chip is interpreting as a continuous stream of garbage bytes, which
it feeds to getty, which tries to interpret them as login attempts.

how do you avoid this problem?

   - leave the device at the other end switched on (even when it's not
     transmitting, it will assert a voltage that overrides the noise)

   - unplug the cable from the next when you're not using it

   - use 'kill -STOP' & 'kill -CONT' to stop & resume the getty process
     as needed

   - buy an adequately shielded serial cable

                                             .mike.

uli@kino.intellektik.informatik.th-darmstadt.de (Ulrich Grepel) writes:
>this is a summary of the hints I've got for my problem with my unused but
>enabled serial ports eating up my CPU:

Just disable them. - see 3a

>2. It seems to be the case that the ports generate interrupts.

They need to work correctly. Where other should the device driver know
from that the fifo needs to be emptied? It could busy wait, but this
would even nned more CPU.

>3. You can decrease CPU usage in several ways:
>   a) don't enable the ports
>   c) use the dsp together with additional hardware to get a high speed
>      serial port that doesn't use up your CPU

Would be interesting, but there are still many things to go on the
DSP-Port.

>4. The whole thing seems to be irrelevant of the OS version. I have 3.0, but
>   it was that bad much earlier. That's easy to understand since it is a
>   hardware problem.

It is not a problem! That's the way a device driver works. If NeXT had
used serial chips with 16 byte fifo for each serial port (and enabled
this 16 bytes fifo), interrupt load would have decreased.


-- 
Heiko W.Rupp  Gerwigstr.5  7500 Kh'e 1  hwr@pilhuhn.ka.sub.org  +49 721 693642
Mustgo, n.:
	Any item of food that has been sitting in the refrigerator so
long it has become a science project.
		-- Sniglets, "Rich Hall & Friends"

In article <1992Nov18.110002.17869@pilhuhn.ka.sub.org> hwr@pilhuhn.ka.sub.org (Heiko W.Rupp) writes:
>It is not a problem! That's the way a device driver works. If NeXT had
>used serial chips with 16 byte fifo for each serial port (and enabled
>this 16 bytes fifo), interrupt load would have decreased.

NeXT uses one of the Zilog serial controllers, which is a good serial
controller, but I think it's a combination of the device-driver and
the apparent lack of DMA (or use of it) in moving bits around.


andrew

hwr@pilhuhn.ka.sub.org (Heiko W.Rupp) writes:

    [ ... ]

you've misunderstood the original posting.  the problem isn't that serial
port communications generate interrupts and consume cpu time when they're
being used; it's that they generate interrupts and consume *more* cpu time
when they're not being used (under some circumstances; see my previous
posting for details).
                                             .mike.