Re: MacPPC serial ports in 4.0.1 - not quite

[Donald Lee <MacPPC2%caution.icompute.com@localhost>]

>MI z8530 driver might have some problem because zs serial on cobalt
>also gets the similar hangup on heavy both TX and RX load
>with high baud rate (115200bps).
>
>MacPPC2%caution.icompute.com@localhost wrote:
>
>> It looks to me like the arch independent code disables ints on the
>> serial when the buffer fills, and the arch dependent code re-enables
>> them.  The token/bit in question is ZSWR1_TIE and ZSWR1_RIE (and friends).
>> I see where this happens 4 xmit in dev/ic/z8530tty.c, about line 1646.
>> The re-enabling of these interrupts, though I can't find.  Maybe
>> it was forgotten?
>
>RIE seems disabled when RX buffer is overflowed in zstty_rxint(),
>and re-enabled when buffer is okay in zstty_rxsoft(),
>which will be invoked via softint(9) scheduled in MD zshard().
>
>TIE is enabled when any data is put into TX buffer in zsstart(),
>and disabled when all queued data has been sent in zstty_txint()
>invoked via softint(9) as well as RX.
>
>I wonder if we should set ZSWR0_RESET_TXINT after TX
>in zstty_txint() rather than clearing ZSWR1_TIE,
>but I haven't tried it.
>(maybe it could have some other race condition)
>
>We might have to check Z85C30 manual anyway:
>http://www.zilog.com/docs/serial/scc_escc_um.pdf
>---
>Izumi Tsutsui

I looked at the code a little, and see clearly that I am not
going to find anything "easy" in only a few minutes (or hours) of
work.  This is a job for someone who knows this code better
than I do.

I am all set up with serial and two machines - one is a reliable
NetBSD 2.1 production machine with Cyclades serial that I use in
production for dial-in.  (I have extra ports)  The other is the
Beige G3, NetBSD 4.0.1, with built-in serial.  I have cables strung, with
a little mini-tester (LEDs on the rs-232 lines) in-line.

I'm set up to build kernels, and can test patches or run experiments.
If someone can provide some patches/experiments to run, I can
make a contribution.

-dgl-