In message <3DC46E6B.2010005@mac.com>, Andrew Cagney writes:
>Just an update on this.  Went back to a 1.6 stable kernel (which 
>hopefully doesn't have smp) and am still seeing a problem.  The thing 
>that is striking is the collision rate:

I have more updates on this too.

As I may have commented, I can produce this on two different wi devices:


1.
wi0 at pci2 dev 2 function 0: Intersil Prism2.5 Wireless Lan
wi0: interrupting at irq 11
wi0: 802.11 address 00:20:e0:8d:15:f2
wi0: using RF:PRISM2.5 MAC:ISL3874A(Mini-PCI)
wi0: Intersil Firmware: Primary (1.1.0), Station (1.4.2)
wi0: supported rates: 1Mbps 2Mbps 5.5Mbps 11Mbps

2.
wi1 at pcmcia0 function 0: Instant Wireless ,  Network PC CARD, Version 01.02
wi1: 802.11 address 00:03:2f:00:02:40
wi1: using RF:PRISM2 MAC:HFA3841 CARD:HWB3163 rev.A
wi1: Intersil Firmware: Primary (0.3.0), Station (0.7.6)
wi1: supported rates: 1Mbps 2Mbps 5.5Mbps 11Mbps

(wi1 was the wi0 in the machine it was in when I was using it, but it's the
same piece of hardware I had the problem on when I first reported it.)

Here's the neat part:  I was obliged to try to network with my wife's iBook,
and I had no base station, so I did an adhoc network.

No problems.

Hypothesis:  The driver is not handling loaded networks and/or base stations
as well as it used to.

The next test would be to set up the iBook using "software base station",
connect to it, and see how things go.  The failures when I had my hardware
base station were fairly repeatable... and a huge collision rate would fit
very well.

I was gonna send this update in yesterday, but I'm in Denver using a 19.2kbps
cell phone connection right now...  I'm only logged in because I needed to
update a couple of files in my source tree (like all of /sys/dev/ic, which
I deleted.  I am *SUCH* a moron.)

-s