My scn driver has code to run the chip at it's maximum (57.6K), which
with compressed PPP should be pretty good (multi-link might be even
better), BUT I've never gotten it tuned to run well at 57.6, and
I haven't even tested my lastest sources (I was still running 1.0BETA
(a year old), but I updated to 1.1alpha last night.

I also support the ioctl's for "ttyflags" (as well as flags set from "config")
and use of the RxRDY interrupt lines...  Here is the latest README;

$Id: scn.README,v 1.5 1995/11/02 20:53:01 phil Exp $

Phil Budne's PC532 SCN driver

scn.c features;
        implements modem control:
                TIOCM{GET,SET,BIS,BIC} ioctl's
                CRTSCTS flow control (in hardware)!!
        obeys TIOCxFLAGS ioctls (used by "ttyflags" utility)
        honors "flags" passed by config.new;
                SCN_SW_{SOFTCAR,CLOCAL,CRTSCTS,MDMBUF}
        "sun" style carrier support
                no longer requires tweaks to tty.c!!
                open on regular minor will sleep until carrier comes up
                        for use by "getty"
                open on minor+128 (/dev/cuaN) will open even if carrier is low,
                        and locks out "regular" opens 
        allows one alternate speed which is always available (using C/T)
                same speed on all 4 chips; default is 57600 (max rate possible)
        Duart rxrdy output lines used for fast input interrupt
                using ringbuffer code from LBL SPARC zs driver
        Full ansi prototyping

messages;

scnN: ring overflow

        ring buffer was overrun; increase SCN_RING_SIZE 
        or decrease SCN_RING_THRESH.  

scnN: fifo overflow

        UART fifo was overrun;
        Use CRTSCTS flow control
        make scnrxintr faster; reduce system load;
        make sure a miniumum of time spent at splhigh()

Neither overflow message should be output more than once a second
for a channel.

TODO:
        get rid of all those ifdefs!!

        implement/debug/test BREAK interrupts with ring buffer

        use DUART counter/timer better;
                allow different fixed rate per-chip?

                allow dynamicly determined rate for each chip. It
                would be easiest to have C/T stay in use by a channel
                until it closes (avoid playing games switching use of
                C/T on open channels)

        make speed group bits a seperate (bit)field in struct?

        ignore inactive channel speed? (see comments in scn_config)

        keep channel rx interrupt mask bits in softc (like for hw int mask)??

        honor CREAD bit in termios?

        implement MDMBUF flow????

        include MAKEDEV diffs in shar