Re: use mutex instead of semaphore in RocketPort driver

[Simon Arlott <simon@fire.lp0.eu>]

On 22/05/07 21:06, Matthias Kaehlcke wrote:
> El Tue, May 22, 2007 at 09:59:01AM -0700 Arjan van de Ven ha dit:
> 
>>>  

Please provide context when quoting a patch, git grep takes a while...

>>> -	down_interruptible(&info->write_sem);
>>> +	mutex_lock_interruptible(&info->write_mtx);
>>>  
>>>  #ifdef ROCKET_DEBUG_WRITE
>>>  	printk(KERN_INFO "rp_write %d chars...", count);
>>> @@ -1773,7 +1776,7 @@ end:
>>>  		wake_up_interruptible(&tty->poll_wait);
>>>  #endif
>>>  	}
>>> -	up(&info->write_sem);
>>> +	mutex_unlock(&info->write_mtx);
>>>  	return retval
>> this code is very very buggy.
> 
> more buggy than with the use of a semaphore?
>  
>> mutex_lock_interruptible() may not get the mutex in case a signal
>> happens... and yet you unlox the mutex unconditionally!!!
> 
> as far as i understand only the thread that locked the mutex can
> unlock it (as opposed to semaphores, which can be released by any
> thread/process). obviously this doesn't make the code be more
> correct. what i don't know is how the kernel behaves when
> trying to unlock a mutex the thread doesn't own. another and possibly
> more important problem of the code is that in case of being
> interrupted by a signal the data that should be protected by the
> mutex/semaphore can be accessed/changed by two threads at the same
> time.
> 
> would the following resolve the problem?
> 
> if(mutex_lock_interruptible(&info->write_mtx)) 
>         return -ERESTARTSYS
> 
> thanks for your comments
> 

No. At least one user of tty_operations/tty_driver's write function 
doesn't check the return value so it would never be retried, mutex_lock 
should be used instead.

All of the _interruptible and functions that return -ERESTARTSYS should 
probably use __must_check...

-- 
Simon Arlott