> It's not so simple: the problem seems to be related more to the PCMCIA
> *controller* than to the cards themselves. There are too many
> ``broken'' cards to consider the problem comes from them only.

I don't think so, and the evidence give is not consistent with current
marketing practices ;-) But there may be driver bugs (bus and/or card)
involved.

> This may be true. That's why we should investigate the idea of
> detecting the PCMCIA address bus size: 0x300 fits within 12 bits
> whereas 0x400 does not.

I'm not an expert in this area (nor would I like to be), but isn't this
a property of the card and noted in the CIS info? Or do you mean detecting
controllers with broken wiring that do not connect the upper four address
lines? Does such crap exist? Then either autodetecting or at least offering
a kernel option is a must.

The problem with typical PC pcmcia bus controllers is that they don't remap
the io space into mem space (or relocate it). This, and the traditional dense
population of IO ports in the lower 0x400 io ports on PCs makes this problem
hard. Cardbus controllers (even on the PC) do not have this problem, since
their bridge chip relocates/maps io and memory to free regions.

Cards that only decode very few address bits (i.e. all sane ones) do not
expose this problems.


Martin