Re: deadlock in "modprobe -r ohci1394" shortly after "modprobe ohci1394"

[Stefan Richter]

Alan Stern wrote:
> Wait a minute.  Above you agreed that the problem was caused by knodemgrd 
> attempting to rescan the host's _parent_.  So which is the focus of the 
> deadlock: the host or its parent?

The parent of the hpsb_host.

ohci1394 works on a pci_dev which contains a dev, let's call it A.

ieee1394 has a hpsb_host which contains a dev, let's call it B. B's
parent is A. Then there is one or more node_entry with dev C whose
parent is B, end unit_directory with dev D whose parent is C.

The bus of devices B, C, D is set to be ieee1394_bus_type, and that's
what knodemgrd is scanning.

knodemgrd blocks on the semaphore of the parent of B because
driver_detach took the semaphore of A (and of the parent of A if there
is one).

[...]
> If I understand all this, you've got an hpsb_host, directly below which 
> are one or more node_entry's, below each of which may be some 
> unit_directory's.  Right?

Right.

> But how is this relevant if the problem is caused by knodemgrd trying to 
> rescan the hpsb_host's parent?

It's rescanning just the hpsb_host, but you surely got the picture now.

> Is the problem caused by the fact that some of these struct device's 
> aren't bound to a driver?  Remember, bus_rescan_devices() will skip over 
> anything that already has a driver.  Could you solve your problem by 
> adding a do_nothing driver that would bind to these otherwise unused 
> devices?

Excellently, that's what I will try in a minute. It is surely intended
that the hpsb_host can get a driver bound too, but as I mentioned, we
don't have a driver which needs this capability and I don't foresee any
such driver.

Thanks for the directions.
-- 
Stefan Richter
-=====-=-==- =-== =-=--
http://arcgraph.de/sr/
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/