CVS: cvs.openbsd.org: src

[Jason Wright <jason_(_at_)_cvs_(_dot_)_openbsd_(_dot_)_org>]

CVSROOT:	/cvs
Module name:	src
Changes by:	jason_(_at_)_cvs_(_dot_)_openbsd_(_dot_)_org	2000/11/17 13:20:09

Modified files:
	sys/arch/sparc/conf: Tag: OPENBSD_2_8 files.sparc 
	sys/arch/sparc/dev: Tag: OPENBSD_2_8 qe.c qereg.h qevar.h 

Log message:
Pull in patches from current:
Errata (003):
Configuring a qec+qe causes a NMI panic.
Fix (jason):
- Fix the NMI problem: it seems the qec channel reset also resets the MACE, and
a race condition existed where after a qec channel reset, accesses to the
MACE would cause an NMI: so wait for the MACE to come out of reset before
accessing.

- add ifmedia support (mainly so I can get at the link status from other code)
- move some of the spl handling around (more correct)
- rewrite qe_mcreset() again so that all it does is set the multicast filter
- use bzero to initialize the buffers and descriptors
- rearrange the MACE setup to conform to the suggestions in the datasheet
----------------------------
wait for the MACE to figure out whether it has media or not, and then
clear interrupt status and such before enabling rx and tx.
----------------------------
- remove other media types, the MACE isn't as bright as one would hope
- always set it to autoselect, tp
- remove the wait for media stuff: appears to be a local issue on
- simplify the ifmedia callbacks appropriately
----------------------------
seems the qe/mace combination needs a delay between accesses to the
two chips.  Better fix for the NMI problem.