PCMCIA (ray_cs) and CONFIG_ISA

[Andrew Pimlott <andrew@pimlott.net>]

I just tried to use the in-kernel ray_cs driver in 2.4.22 (after
having used the stand-alone PCMCIA driver for years).  I encountered
much frustration, and while I don't have a complete understanding of
the issues, it seems that there is a tangled relationship between
CONFIG_ISA and allocating interrupts for PCMCIA devices.  In short,
everything works with CONFIG_ISA enabled (even only in cs.c and
rsrc_mgr.c) and fails otherwise.  I'm pretty sure my notebook
doesn't have any ISA, and there is nothing in kconfig to require
CONFIG_ISA, leaving a nasty trap.

Looking at pcmcia_request_irq, it seems that there is no chance to
pick an irq if CONFIG_ISA is disabled and the IRQ_HANDLE_PRESENT
flag is set.  The only chance is if the socket irq_mask is zero, but
this doesn't seem to be the case for PC sockets (I'm using
yenta_socket).  So it ends up requesting irq 0, which doesn't work
out so well (CS_IN_USE).

The code is basically the same in the stand-alone drivers, but I
noticed that the Debian build scripts for the stand-alone drivers
force CONFIG_ISA on.  I didn't find any documentation of why.  I
guess that's why I never had a problem before.

So how are things supposed to work?  Is this driver doing something
wrong (I observed the pcnet_cs driver works fine without CONFIG_ISA:
it doesn't get its own irq and is happy with that)?  Pilot error?
Or just an undocumented pitfall?

Here are the kernel messages for a failed attempt to use the
in-kernel ray_cs driver:

    Linux Kernel Card Services 3.1.22
      options:  [pci] [cardbus] [pm]
    cs.c 1.279 2001/10/13 00:08:28 (David Hinds)
    PCI: Found IRQ 11 for device 00:0b.0
    PCI: Found IRQ 11 for device 00:0b.1
    Yenta IRQ list 06b0, PCI irq11
    Socket status: 30000007
    cs: pcmcia_register_socket(0xd0b9f360)
    Yenta IRQ list 06b0, PCI irq11
    ray_cs Detected: WebGear PC Card WLAN Adapter Version 4.88 Jan 1999
    ray_cs: AccessConfigurationRegister: Resource in use

Andrew

Cc: Rusty King since he seems to have submitted some maybe-related
patches for 2.5.