The final patch removes the unplug/replug deadlock by not holding the device
mutex during dvb_init(). However, this mutex has already been locked during
device initialisation by em28xx_usb_probe() and is not released again until all
extensions have been initialised successfully.
The device mutex is not held during either em28xx_register_extension() or
em28xx_unregister_extension() any more. More importantly, I don't believe it can
safely be held by these functions because they must both - by their nature -
acquire the device list mutex before they can iterate through the device list.
In other words, while usb_probe() and usb_disconnect() acquire the device mutex
followed by the device list mutex, the register/unregister_extension() functions
would need to acquire these mutexes in the opposite order. And that sounds like
a potential deadlock.
On the other hand, the new situation is a definite improvement :-).
Signed-off-by: Chris Rankin <rankincj@xxxxxxxxx>
--- linux-3.0/drivers/media/video/em28xx/em28xx-dvb.c.orig 2011-08-19 00:50:41.000000000 +0100
+++ linux-3.0/drivers/media/video/em28xx/em28xx-dvb.c 2011-08-19 00:51:03.000000000 +0100
@@ -542,7 +542,6 @@
dev->dvb = dvb;
dvb->fe[0] = dvb->fe[1] = NULL;
- mutex_lock(&dev->lock);
em28xx_set_mode(dev, EM28XX_DIGITAL_MODE);
/* init frontend */
switch (dev->model) {
@@ -711,7 +710,6 @@
em28xx_info("Successfully loaded em28xx-dvb\n");
ret:
em28xx_set_mode(dev, EM28XX_SUSPEND);
- mutex_unlock(&dev->lock);
return result;
out_free:
