Pthreads & Signals.

[gerald.feldman at hp.com (Jerry Feldman)]

On Wednesday 17 November 2004 11:02 am, Anthony Gabrielson wrote:
> Okay,
> 	With the help of several I got it working and clarifed my
> inderstanding on how pthreads work.  So anyway I have it working and if
> anyone is curious I attached it.
I just want to followup with a final note. 
WRT: pthrsignal_V2.c
In Main line 43 youn try to lock the mutex after the condition wait:
	result = pthread_cond_wait( &got_data, &guard_revc );
		:
		:
	result = pthread_mutex_lock( &guard_revc ); BAD 

pthread_cond_wait() is called with a locked mutex. It unlocks the mutex 
allowing other threads to proceed, and then when it returns the mutex is 
relocked. So in this case you are locking the mutex twice. If the mutex was 
a recursive mutex, that would cause some interesting problems. 


In function thr:
	result = pthread_mutex_lock( &guard_revc );			a. 
	printf("thr: Got Mutex %d\n",result);
	usleep(1);
	result = pthread_mutex_unlock( &guard_revc );               b. 
	printf("thr: Release Mutex %d\n",result);
	result = pthread_cond_signal( &got_data );                     c.

a. Good place to acquire a lock except that you don't want to acquire it 
before it is necessary. There is no reason to acquire a lock before 
usleep() or printf().

b. Very bad. The mutex should be locked BEFORE calling pthread_cond_wait. 
See my explanation above. 

WRT: While loops. They accomplsh nothing. 
Lock is initialized to 0. You loop on it, call pthread_cond_wait(), then you 
unconditionally set lock to 1. It appears you are trying to synchronize 
with main, but if you had set your mutexes correctly, like in the code I 
sent you, you would not need that. 

And, last of all, after you complete pthread_cond_wait() you FAIL to unlock 
the mutex. This exits the thread with a locked mutex causing a potential 
deadlock. As I mentioned yesterday, in main(), lock the mutex before 
calling pthread_create. This will cause thr() to block on entry before it 
calls pthread_cond_signal(). Then you enter the pthread_cond_wait() which 
unlocks the mutex allowing thr() to run. Thr() then signals the condition 
causing main() to wake up and block on the mutex (before exiting its 
cond_wait). Thr then enters its cond_wait, unlocking the mutex as it 
enters. Main now returns from cond_wait. It unlocks its mutex, signals the 
condition and joins. Thr() now wakes up, reacquires the mutex (implicitly 
by cond_wait), then releases the mutex and exits. 


Here is some pseudo code (with printfs et. al eliminated)
Main:										child:
	 lock mutex
	create child								locks mutex - is blocked.
	wait on condition (mutex is implicitly unlocked)		child now can run
	condition complete, blocks on mutex			signal condition
	(cond reacquires mutex implicitly)				waits on condition 
											(mutex is unlocked)
	Unlock mutex (no affect on child)
	issues signal to cond.						returns from cond
	Waits on join.							unlocks mutex
										return (pthread_exit)
	program completes/

Some other things to remember. You should maintain a locked mutex for as 
short a time as possible. In this case, your lock variable is useless. But, 
let's way you have a worker thread that wants to get data from a queue:

worker:
	lock mutex.
	Are there elements on the queue    <<<<<--------+
		yes -- Remove the next element		     ^
			unlock the mutex	                             |
			process that element.                        |
			continue.                                         |
                                                                             |
		no -- Wait on condition.                              |
			---------------------------------->>>>> +
	done:
	unlock mutex. 
	return.
	
--
Jerry Feldman <gerald.feldman at hp.com>
Partner Technology Access Center (contractor) (PTAC-MA)
Hewlett-Packard Co.
550 King Street LKG2a-X2
Littleton, Ma. 01460
(978)506-5243