[Etherboot-developers] PCMCIA

[Anselm M. H. <an...@ho...>]

Hi list,

nearing LWE, I had planned to release PCMCIA support for i82365
chipset, but I'm not sure if I will do this in time. It's not in CVS
(yet) as there's more than one design issue to be handled, which I'd
prefer to have done before "public" (you :-) work on it.

This is not just a status announcement. I'd like advice on some points
in integration. Up to now, I defined CONFIG_PCMCIA (like CONFIG_ISA),
but propably the latter is needed for PCMCIA anyway. Not talking about
CARDBUS of course, but I don't have a CARDBUS card neither compatible
laptop anyway, I'm working on a 486 with some 10M really old NICs.
Where in code should I best place the pcmcia_init() call (which loads
support for the mainboard part of PCMCIA, configures and activates the
first found PCMCIA NIC)? Where best to place the pcmcia_deinit() that
puts everything in status 'off'? - Perhaps there should also be an
option to not deinit the pcmcia device. My laptop for example has no
DOS "point enabler" software, so that NICs cannot be used with DOS,
only with winblows (which knows about enabling PCMCIA). They seem -
except that PCMCIA stuff - to be NE2000 compatible. I'm quite sure
they will interface seemlessly to the etherboot NE2000 module, but am
not far enough to test this. And DOS could work after having them
point-enabled...

Is there any sense in activating more than the first found PCMCIA NIC?
Is there any reason to use non-NIC PCMCIA devices in etherboot?
Does someone know about special requirements to PCMCIA enabler stuff,
except giving a NIC the right power & io address? Can I skip the work
for interrupt activation or is there any NIC out there that needs it
(and is available as PCMCIA)? Do we need an option to configure cards
other than to their standard configuration*?

*Inside most (compliant PCMCIA2.x) card there is a Card Information
Structure holding several (at least 1) possible ioport/irq/... tuples
one of which can be selected. I'd like to just activate the first one,
without too cmplicated (or without any at all) checking for
plausibility - I expect the first tuple to be 0x300/irq5 or similar in
NE2000 clones anyway. However having written that config to console
would clarify this point. Is there any chance to detect which ports
could be used yet? I don't expect a laptop to use e.g. ports 0x300;
0x320; 0x240.

FYI - what is working yet:
o Detection of motherboard chipset (Intel 82365, compatible CIRRUS and
  VADEM chipsets, mainly those boards supported by the same linux kernel
  driver) - only tested with my laptop's VADEM VG-468 chip.
o Detection of "used" aka "card-inside" slots
o The logic behind CIS, tuple reading and so on

What needs work - and that probably is all:
o Write and read accesses to CIS

A pity that CIS is so f**ine complicated. I'm still investigating
where between all this multiplexing stuff the CIS hides, but I hope to
catch it this evening.

I'll surely release to CVS before leaving for a 6-day-holiday this
friday, so perhaps (well, it depends :-) you will have some working
demo stuff on LWE, but I cannot guarantee.

Best regards,
  Anselm Martin Hoffmeister
  Stockholm Projekt Computer-Service
  <an...@ho...>