Hi,

I will re-test this problem.

Thanks,
Ken

quoted

smc91c92_cs:
    set smc->base to NULL before iounmap
    to avoid writing to smc->base in smc_interrupt.

Signed-off-by: Ken Kawasaki <redacted>

The device should be completely made quiet and all interrupts
synchronized before unmapping chip registers.  Using NULL checks
is merely a workaround for the real problem, even though you take
great pains to assign NULL before the iounmap() call.

Perhaps putting the pcmcia_disable_device() call before the iounmap()
will be sufficient to fix this bug.  Or perhaps that, plus a call to
synchronize_irq() or similar.