Re: klists and struct device semaphores

[Patrick Mochel]

On Wed, 6 Apr 2005, Alan Stern wrote:

> The patch looks good.  But isn't there still a problem with
> device_release_driver()?  It doesn't wait for the klist_node to be removed
> from the klist before unlocking the device and moving on.  As a result, if
> another driver was waiting to bind to the device you would corrupt the
> list pointers, by calling klist_add_tail() for the new driver before
> klist_release() had run for the old driver.
>
> I'll be interested to see how you manage to solve this.  The only way I
> can think of is to avoid using driver_for_each_device() in
> driver_detach().

I had implemented something along the lines of what you suggested in a
previous email:

  - Add flag to klist_node: n_attached.
  - Use that in klist_node_attached()
  - Check during klist_next(), and skip nodes that have been removed.
  - Reset flag during klist_del().

Based on the deadlock that occurs when using klist_remove() during an
iteration over that the list, and the common indirect usage of it (parents
and buses unregistering devices, drivers unbinding devices), I've just
removed klist_remove() and the embedded completion. I'll post updated
patches soon.

Thanks,


	Pat
-
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/