On Mon, Jul 07, 2008 at 07:30:58PM +0200, Matthias Drochner wrote:
>
> [catch master abort in cardbus probe]
> jakllsch%kollasch.net@localhost said:
> > Though the machine hard-hangs when i remove
> > that dual-function card, I'm not sure if this is a regression
> > though.
>
> It is unlikely to be related to the probe change, but there
> was the device/softc split shortly before which could have
> introduced a bug. At a first glance, I don't see a mistake
> in cardbus_detach_card()/cardbus_childdetached(), in particular
> nothing which could explain why multifunction cards could
> cause worse behaviour than singlefunctions.
> Could tou try a kernel with CARDBUS_DEBUG, or place some
> printfs into the two functions mentioned, to find out where
> the hang happens?
I think this is the result of a absent mutex_spin_enter()
at the beginning of com_activate(). Or, at least, adding
one seemed to help.
a log of inserting and removing the card with a
DEBUG,DIAGNOSTIC,CARDBUS_DEBUG,CBB_DEBUG mostly-4.99.64
kernel before adding that mutex_spin_enter line:
http://user.kollasch.net/jakllsch/2008/0708/com-lock-bug
Jonathan Kollasch
Attachment:
pgpI7Ii79PZrU.pgp
Description: PGP signature