Re: [PATCH 3/3] [media] cx88: use a mutex to protect cx8802_devlist

[Jonathan Nieder <jrnieder@gmail.com>]

Hi Andreas,

(please turn off HTML mail.)
Andreas Huber wrote:

> There is a reference count bug in the driver code. The driver's
> active_ref count may become negative which leads to unpredictable
> behavior. (mpeg video device inaccessible, etc ...)

Hmm, the patchset didn't touch active_ref handling.

active_ref was added by v2.6.25-rc3~132^2~7 (V4L/DVB (7194):
cx88-mpeg: Allow concurrent access to cx88-mpeg devices, 2008-02-11)
and relies on three assumptions:

 * (successful) calls to cx8802_driver::request_acquire are balanced
   with calls to cx8802_driver::request_release;

 * cx8802_driver::advise_acquire is non-null if and only if
   cx8802_driver::advise_release is (since both are NULL for
   blackbird, non-NULL for dvb);

 * no data races.

I suppose it would be more idiomatic to use an atomic_t, but access to
active_ref was previously protected by the BKL and now it is protected
by core->lock.  So it's not clear to me why this doesn't work.

Any hints?  (e.g., a detailed reproduction recipe, or a log after
adding a printk to find out when exactly active_ref becomes negative)

Thanks for reporting.
Jonathan