[OT] volatile keyword

From: Vadim Lobanov
Date: Thu Aug 25 2005 - 15:45:21 EST


Hi,

The recent discussion on the list concerning memory barriers and write
ordering took a side-trip to the volatile keyword, especially its
correct / incorrect usage. Someone posted a link to the LKML archives,
in which the argument is made that it is best to keep 'volatile' _out_
of variable and structure definitions, and _into_ the code that uses
them. I was curious, so I decided to try this out (looking at
kernel/posix-timers.c for inspiration)...

Here's sample userland program number one, written one way:
===========================
#include <pthread.h>
#include <stdio.h>
#include <unistd.h>

struct sync {
volatile unsigned long lock;
volatile unsigned long value;
};

struct sync data;

void * thread (void * arg) {
sleep(5);
data.value = 0;
data.lock = 0;

return NULL;
}

int main (void) {
pthread_t other;

data.lock = 1;
data.value = 1;
pthread_create(&other, NULL, thread, NULL);
while (data.lock);
printf("Value is %lu.\n", data.value);
pthread_join(other, NULL);

return 0;
}
===========================

And here's what should be the same program, written the "suggested" way:
===========================
#include <pthread.h>
#include <stdio.h>
#include <unistd.h>

struct sync {
unsigned long lock;
unsigned long value;
};

struct sync data;

void * thread (void * arg) {
sleep(5);
data.value = 0;
data.lock = 0;

return NULL;
}

int main (void) {
pthread_t other;

data.lock = 1;
data.value = 1;
pthread_create(&other, NULL, thread, NULL);
while ((volatile unsigned long)(data.lock));
printf("Value is %lu.\n", data.value);
pthread_join(other, NULL);

return 0;
}
===========================

The first program works exactly as expected. The second program,
however, never syncs with the sleeping thread. In fact, for the second
program, gcc compiles the while loop down to an infinite loop.

I'm positive I'm doing something wrong here. In fact, I bet it's the
volatile cast within the loop that's wrong; but I'm not sure how to do
it correctly. Any help / pointers / discussion would be appreciated.

Thanks. :-)
-VadimL
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/