Feature #5349
Updated by Vadim Mutilin about 10 years ago
*Summary*
# It’s not allowed to acquire spin_lock twice.
# It’s not allowed to release not acquired spin_lock.
# At the end all spin_lock should be released.
*Description*
Unlike other operating systems spin-locks in Linux are not recursive. So if a thread tries to acquire a lock it has already acquired then this thread begins a periodical check and waits till it releases the lock. This leads to a deadlocks. The situation is not often in its explicit form, but it is widely spread in an implicit form when some function (external function usually) tries to acquire a lock that has already been acquired by the module.
It’s not allowed to acquire spin_lock twice. It’s not allowed to release not acquired spin_lock. At the end all spin_lock should be released. It’s not allowed to re-release a lock by spin_unlock/spin_unlock_irqrestore functions.
*Links*
"Sample bugfix":https://git.kernel.org/cgit/linux/kernel/git/stable/linux-stable.git/commit/?id=83c7c693ed3e61535ad6a097ad991a88aafc54b8
*Example*
example of bug (2.6.24 -> 2.6.25, drivers/char/specialix.c):
<pre><code class="с">
@@ 2109
sx_out(bp, CD186x_CAR, port_No(port));
/* releasing a lock */
spin_unlock_irqrestore(&bp->lock, flags);
if (I_IXOFF(tty)) {
/* in some cases double unlock call is possible */
spin_unlock_irqrestore(&bp->lock, flags);
sx_wait_CCR(bp);
spin_lock_irqsave(&bp->lock, flags);
sx_out(bp, CD186x_CCR, CCR_SSCH2);
</code></pre>