Subject: Re: Multia PCMCIA
To: Bernard Gardner <>
From: Bernard Gardner <>
List: port-alpha
Date: 04/30/2000 22:41:40
I'm talking to myself again...

On Thu, Apr 27, 2000 at 04:57:48PM +1000, Bernard Gardner wrote:
> I've been trying on and off (mostly off) for the past few weeks to
> get the PCMCIA support in NetBSD to work on my Multia.

Well, I've had a bit more of a play with this over the weekend.

By disabling the interrupt probing code in dev/isa/i82365_isasubr.c
(I just set pcic_irq_probe = 0) I've been able to get a kernel that
assigns IRQs to the pcic and to a card (a lucent WaveLAN) that's
inserted in (either of) the PC Card slot(s). The kernel seems to
like to choose IRQ 5 for the pcic (card insert/removal events) and
IRQ 14 for the card itself. I've also tried a kernel config that
assigns IRQ 14 to the pcic, and this kernel assigns IRQ 5 to the

With these kernels, tha card is detected, an IRQ is assigned, and
I can configure the interface and bring it up. The card is then powered,
and configured, and I can use wiconfig(8) to change settings on the
card. Using a second WaveLAN card in a windows laptop, I can send and
receive link quality packets, and the windows machine can see the
paramaters I configured for the card in the multia.

But (there's always a but) the interrupts still don't seem to be right.
Any attempt to use the interface on the multia results in console
messages "wi0: device timeout". As far as I can tell from watching
the activity lights on both cards, no data is ever sent over the
air from the multia. I have noticed that if I run tcpdump on the
multia, listening to the wi0 interface, the link quality monitor
app on the windows box stops working (maybe I should build a protocol
on top of that :). It's so close... if only I knew a little more about
the multia's ISA & PCMCIA workings... guess I need to do some more

BTW, can someone tell me what the pre-processor macros are that I
should use to make the disabling of IRQ probing conditional on the
alpha (or multia) build? This is already done for the hpcmips build,
but they seem to be overloading the irq with something else (or there
are a helluva lot of interrupt lines in those little handheld boxes).