bridge problems

[David Clemmer]

Hello all,

I'm trying to figure out what's going on with an odd box we've got here.
It has 4 TI 1225 cardbus bridges on it (8 total slots).  When I start it
up, the bridges seem to get configured correctly (insofar as memory
allocations and bus allocations seem sane, at least).  The only sign of
anything wrong with the configuration is that 3 of them get set to IRQ9,
and the other 5 get set to IRQ11.  What makes that noteworthy (I think) is
that both IRQs are also assigned to other hardware, and that there are only
3 IRQs assigned (err... that show up with lspci, that is; /proc/interrupts
lists 0,1,2,8,9,10,11,14 in use).

The key problem, though, is that none of the ports accept cards.  If I
start it up with a card inserted (or use 'cardctl insert'), I get an
'unable to apply power' message for the occupied ports.  Also, the socket
statuses (stati? :) reported by yenta when starting are (in order)
30000006, 30000006, 30000047, 30000061, 30000006, 30000047, 30000006, and
30000047.  The only one of those that changes with a card inserted is the
first one, which becomes 30000020 (oh, and the '61 always has a card
inserted; I can't remove it).  cardctl ident and status equally report
nothing being present, regardless of whether or not that is true.

Let's see... cardmgr also reports that it is watching 8 sockets, and that
it succeeded starting pcmcia.

Unfortunately, since the only network interface is cardbus, the only output
device on the machine is the monitor, so it's (to put it kindly) difficult
to send reams of debugging data...  However, if specific data will help,
let me know, and I can see about transcribing it.

Anyway, if anyone can give me any leads about which way to go, or what to
look for, or ways to isolate the problem, I'd greatly appreciate it.  My
knowledge of kernel internals is pretty much what I've been able to teach
myself over the last month or so, however.

Some other notes of interest.  I'm using fedora core 1.91 (core 1 had
different problems; mostly, bus allocations were badly hosed).  The
hardware is not guaranteed to be non-faulty, although it is likely to be
good (the ports, at least, were known to be good as of a month or so ago).

Also, please cc any replies to me, as I'm not on the list.  Thanks a lot,

Dave Clemmer