SUMMARY: Re: serial i/o, bangs, buffers, bleary eyes

From: Bob Makowski (
Date: Wed Feb 03 1993 - 02:37:37 CST

I posted these original questions, and I received a series of e-mails
that I will try to summarize.

In article <> writes:
>It's time for a bleary eyed dump of some questions I'm interested in
>hearing answered. These focus on some thought experiments about the
>(Sun/sparc) serial ports. I'm assuming outbound modem i/o in all cases below.
>(I hope these questions aren't too cryptically worded.)
> * stty says: raw, clocal, cread. Will this CPU interrupt per character?

Always get interrupts per char on the built in serial ports. In fact when
continuous at 19.2, the perfmeter's interrupt display is quite convincing

> * tty interrupt priority vs. SCSI bus interrupts. How do they
> compare logically?

Both h/w and s/w interrupt on the built ins are much higher than the SCSI.

> * how much write(2) i/o before get flow control from serial device,
> not the DCE.

I personally did not find the limit. But I was not looking too hard. I
know I threw a lot of data at 19.2 DTE (each approx. 1k writes) when the
modem speed was in 9.6. In general, I'm told that the STREAMS driver can
get full, as a function both of messages and their aggregate size. But
with such a slow device, and such small buffers used, I couldn't see
flowcontrol in my meager little tests.

> * does the device driver dma (written) data onto the serial line
> driver/chip.

Not the built in serial ports, of course. I asked that question because I
was confused about some includes I found. The Sbus MCP (e.g.,) 8::1
probably does DMA. Those aforementioned include files show dma buffers and
sizes. There are other SCSI terminal servers, that i learned about in this
process. These servers can interrupt upon SCSI packet "xmit", but
don't interrupt otherwise.

= Mak

P.S. Thanks and gratitude to Guy Harris, and to Cendata's support people,
including Dennis Cronin and Vic Serbe.

This archive was generated by hypermail 2.1.2 : Fri Sep 28 2001 - 23:07:27 CDT