howdy,
> this code already breaks even without a non-preemptible kernel.....
> your use of *_sleep_on() is defective.
oops sorry.
> You, probably, want wait_event_interruptible().
I'll try :)
> Is this code supposed to be race-condition-free and to be working on
> multiprocessing environments? If so, you should consider using atomic
> instructions like mutexes and semaphores....
well I was not thinking about the SMP part (apologies).
previous code obviously doesn't work if under SMP.
let me try rephrase one more time:
1) is exclusive access policy basically okay like the below
under SMP and/or PREEMPTIVE kernel? [I also want to sleep
until some conditions are met]
2) how does the kernel support existing old code (eg., drivers)
which assumes non-preemptive kernel behavior even under
preemptive option enabled?
Or, ...said differently, does being 'SMP aware' automatically
imply 'preemptive-kernel ready'? (if so, plase disregard
rest of my message)
--
static wait_queue_head_t q;
static int some_flag = 0; /* condition variable of some sort */
static int lock = 0; /* lock for critical section */
static ssize_t
my_write(struct file * file, const char * buf,
size_t count, loff_t *ppos)
{
while (1) {
if ((error = wait_event_interruptible(q, !some_flag)) < 0) {
retval = error;
goto out;
}
#if SMP_OR_PREEMPTIVE /* = !(UNI_processor && NONPREEMPTIVE) */
if (!test_and_set_bit(0, &lock)) { /* samaphore equally okay, atomic_inc_and_test() and such, too */
if (!some_flag) {
some_flag = 1;
break;
}
else
test_and_clear_bit(0, &lock);
}
#else
some_flag = 1; break;
#endif
}
<...access shared resource assuming exclusive access...>
#if SMP_OR_PREEMPTIVE /* = !(UNI_processor && NONPREEMPTIVE) */
test_and_clear_bit(0, &lock); /* just clear_bit() instead? */
#endif
wake_up_interruptible_all(&q); /* tell some_flag has changed */
out:
return retval;
}
--
regads,
kenji
--
Kernelnewbies: Help each other learn about the Linux kernel.
Archive: http://mail.nl.linux.org/kernelnewbies/
FAQ: http://kernelnewbies.org/faq/
