2.1.8 broke setserial - my CSLIP connection didn't work anymore, since
setserial believed (autodetected) both serial ports having IRQ 0.
Until the problem is fixed,
setserial -b /dev/cua0 autoconfig irq 4 spd_vhi
setserial -b /dev/cua1 autoconfig irq 3 spd_vhi
Automatic IRQ detection is one of those things which I implemented
because I thought it was a cool thing, and then later regretted because
it's so hard to get it right. There are a lot of different, nonstandard
UART's out there. Some pretend (and fail) to be 16550A compatible.
Others (like the StarTech) try to be upwards compatible, but trying to
support them and other UARTs can be devilishly difficult.
In this case, my attempt to support the StarTech UART's is probably what
broke automatic IRQ detection. I'll see if I can fix that; if I can't,
I may very well likely turn it off completely, and simply make every
specify the irq explicitly in their rc.serial file.
So I would call this "until the problem is fixed" --- I'd strongly
suggest explicitly using the irq in your rc.serial permanently, because
some day I may just decide that automatic IRQ detection was a mistake
(like the callout/dial in devices was a mistake), and that the 2.1
development release would be a fine time to yank support for these
features.
Heck, it would make the serial driver smaller and simpler, and I've
gotten a lot of complaints about how big the serial driver is......
- Ted
P.S. It looks like my attempt to detect TI 16750's break on some
no-name internal modems that try to emulate a 16550. In that case,
since I haven't seen anyone with hardware that actually *uses* the
16750, I'll probably turn off the 16750 detection, and let people just
configure the uart explicitly in rc.serial. Again, automatic detection
is nice, but it has its limits.