cvs commit: src/sys/conf files src/sys/dev/uart uart_cpu.h uart_cpu_alpha.c uart_cpu_amd64.c uart_cpu_i386.c uart_cpu_ia64.c uart_cpu_pc98.c uart_cpu_sparc64.c uart_subr.c

[Marcel Moolenaar]

On Sat, Mar 20, 2004 at 12:53:56PM -0800, John-Mark Gurney wrote:
> Marcel Moolenaar wrote this message on Sat, Mar 20, 2004 at 12:16 -0800:
> > > > Probe order is irrelevant if you specify the device by its I/O port
> > > > or memory mapped I/O address. When you need to pinpoint a device
> > > > prior to any form of bus enumeration, the unit number is meaningless
> > > > and any mechanism that includes a unit number is bogus. That's why
> > > > hints are the wrong tool for pinpointing a console or debug port.
> > > 
> > > I hope that you don't have to specify the console via io/memory port..
> > 
> > Yes you have. This is no different than hints, except that you set
> > flags in that case to tell uart(4) that the device is to be used as
> > a console or debug port. But you still need to specify the I/O port
> > because without it your hint is bogus.
> 
> My objection is to Yet Another Hint Place to stick IO/Memory wiring down
> addresses...

You're missing the point. The hw.uart.console and hw.uart.dbgport variables
do not wire down anything. They simply tell uart(4) that the serial device
at that particular address is to be used as a console or debug port. We're
not programming any chipsets to put a serial device at that address, nor do
we reserve that address in some form or another and automagically have a
serial device show up during bus enumeration. Hence, they are not hints
and they do not depend on hints. Hints are a thing of the past and useful
only on certain platforms. One should not gratuitously and artificially
add a dependency to hints. Yes, this also means SCSI busses...

-- 
 Marcel Moolenaar	  USPA: A-39004		 marcel at xcllnt.net