Re: pcic kthread creation timing

[Rafal Boni <rafal%pobox.com@localhost>]

KIYOHARA Takashi wrote:
> I saw neither wdc nor NIC at pcmcia (at pcic) attached on hpcmips.  And,
> if the neither is attached on hpcmips, root device cannot be found. X-<

Hello Kiyohara-san!

hpcarm had this same problem, and I think Peter Postma and I came up 
with pretty much the same fix, which used config_interrupts().  See also 
PR port-hpcarm/37227 for more detail.

I think at least on hpcarm, the calling sequence before the vmlocking 
merge was something like:

pcic_attach (I forget the exact method here)
  +----> kthread_create(create_event_thread())
        ...
(on new thread) create_event_thread()
                      +----> check for card presence
                      +----> kthread_create(pcic_event_thread())

which was then flattened to (note, no extra thread creation):

pcic_attach (I forget the exact method here)
  +----> create_event_thread()
              +----> check for card presence
              +----> kthread_create(pcic_event_thread())

The result being that the 'check for card presence'  was done before the 
PCIC was fully set up and thus no PCMCIA / CF cards were detected at boot.

> I saw pcic_event_thread() called earlier than pcic_attach_socket_finish().

Right, exactly.

> Perhaps, pcic_attach_socket_finish() should call
> kthread_create(pcic_event_thread()) in it if config_interrupts() is slower
> than kthread_create().

I think you do want to use config_interrupts() instead of relying on 
kthread creation timing... one of the side effects is that you'll be 
guaranteed to be called before autoconfiguration finishes and you need 
the boot device.

--rafal