Re: IBM Microdrive PCCard on TiBook (ide-cs module)

[David Blythe <blythe@routefree.com>]

Ira Weiny wrote:
>
> I have spent a little more time tracking down why the PCCard adapter for
> my Microdrive causes the TiBook to shutdown.  But I need some help in
> interpreting the information I have found.


Your symptoms sound similar to the ones i was seeing.
I recently did some work to get a PCMCIA ATA Flash card running on a
405GP (using a circa 2.4.0 kernel with some later pcmcia updates from
the 2.4-devel tree).  I ran into a large number of problems with
interrupts that would hang the system, but haven't figured out what to
do with them.  The basic issue i had was that the ide interrupt is
shared with the pc card controller and during drive probing the prober
issues a command which generates an interrupt with no ide interrupt
handler in place but the pccard interrupt handler gets flooded since it
is sharing the same line.  I 'fixed' the problem by disabling the
specific interrupt from ide-cs.c right before the call to ide_register()
and then (ughh) re-enabled it from inside init_irq() in ide-probe.c
(rather than re-enabling it right after the call to ide_register() to
avoid missing interrupts :(.  Additionally i did something to force the
SHIRQ flag when the irq was requested.  ide-cs differs from other card
service drivers in that the others usually install the interrupt handler
themselves, but ide-cs lets the rest of the ide susbsystem do it.
	david

>
> 1) I was able to catch with xmon a stack trace before it shutdown.  This
> showed a stack trace as follows:
>
> ide-cs.c:ide_config
> ide.c:ide_register
> ide.c:ide_register_hw
> ide.c:ide_probe_module
> ide-probe.c:ideprobe_init
> ide-probe.c:probe_hwif
> ide-probe.c:do_probe
> ide-probe.c:try_to_identify
> ide-probe.c:actual_try_to_identify  <== shutdown somewhere here??
>
> 2) I am assuming at this point it is writing to registers which are
> mapped incorrectly (probably to the PMU) which cause it to shutdown.
>
> 3) I have found that the io_base addresses is 0x1000 (as set in my
> config.opts file) and this is what is passed to the ide_register
> function.  Changing this address does not seem to change it's behavior
> though.  (??)
>
> 4) ide_register calls ide_init_hwif_ports with these addresses which
> have a "pmac_" specific call.  So I am assuming that it would "ioremap"
> these as appropriate.  (However, I don't see it modifying this address
> at all?)
>
> 5) ide_register_hw calls "ide_probe_module" which takes me into the
> ide-probe.c file.  But I see a specific "pmac_ide_probe" which is called
> at init time.
>
> My conclusion is that I should either be calling this pmac_ide_probe
> rather than the generic ide_probe module (and making it _not_ an init
> function).  OR I need to modify the ide_probe module to handle the
> proper maping of the addresses such that I don't write to the PMU.
>
> Unfortunatly I have not found any examples of the later in the pcmcia
> drivers I have looked at so I am worried that I am totally off base.  If
> I am "barking up the wrong tree" I hope someone can tell me such that I
> can start looking elsewhere.
>
> Thanks in advance,
> Ira Weiny
> iweiny@acm.org
>

** Sent via the linuxppc-dev mail list. See http://lists.linuxppc.org/