driver crash on card insert

[David Seery]

Hi,

I have a problem with pcmcia wireless I was hoping someone could offer
advice about.

I'm running Debian sid with a custom 2.6.0-test9 kernel on a hp/compaq
nx9005 AMD Athlon 2400+ laptop, and I'm trying to use an Adaptec
AWN-8030 pcmcia card.  As far as I can make out from various deposits
left in syslog, this is a Prism2 card with (secondary) firmware version
1.5.6.

The system hangs on card insert.  I think this is a hang rather than
kernel panic because there is nothing left in syslog, and (for example)
the caps lock light isn't illuminated.  The problem is exhibited with
the hostap driver (which I'd prefer to use because of the wireless
extensions support) on 2.6.0-test9 and 2.4.22, the orinoco driver on
both kernels, and the linux-wlan-ng driver on 2.4.22.  The linux-wlan-ng
driver on 2.6.0-test9 doesn't crash but doesn't work, either.  Therefore
this doesn't seem to be a driver problem, so I'm aware that this list
might not be the most appropriate venue.  However I'd like to ask in
case someone here has encountered a similar difficulty.

There are some known irq problems on this laptop such that one has to
exclude the irqs assigned to various built-in ports.  This is not a
problem.  When running the linux-wlan-ng driver on 2.4.22, the only
instance where the machine doesn't hang, wlan0 is assigned to irq 10
despite eth0 and the yenta_socket driver this machine uses apparently
running on irq 11.  To judge from the dmesg spew at boot-time, the PCI
code is disabling irq sharing.  Might I ask if this is a sane irq
assignment?  The result is that no traffic appears to get sent or
received on wlan0, although the access point ssid seems to be set
correctly.  WEP is disabled.  For all other combinations of driver and
kernel, as I said, the system hangs.

It was necessary to disable symbol version information for modules on
the 2.4 series kernels to prevent a crash on card insert, apparently
before the wireless drivers were loaded.  I gather this is a known
problem with the AMD MMX.  It doesn't occur with 2.6 series kernels. 
I'm also supposed to have yenta_socket loaded before (at least) the usb
and eth0 drivers on this laptop.  Although I'm not sure what problem
this resolves, it doesn't seem to make any difference on the 2.6 series
kernels.  As far as I know, it isn't possible to tell yenta_socket which
irq to use.  On the other hand, the whole irq scenario may not be the
culprit.  I'm not very knowledgeable about hardware :/

Thanks in advance

David