Willy Tarreau wrote:
> Hi Jan,
> 
> On Sun, Oct 08, 2006 at 11:42:07PM +0200, Jan Kiszka wrote:
>> Hi,
>>
>> after going through debugging hell with some out-of-tree code, I
>> realised that this patch
>>
>> http://www.kernel.org/git/?p=linux/kernel/git/torvalds/linux-2.6.git;a=commit;h=92934bcbf96bc9dc931c40ca5f1a57685b7b813b
>>
>> makes a difference: current 2.6 works with the following code sequence
>> as expected (printk is executed), 2.4 fails.
>>
>>
>> #include <asm/bitops.h>
>> #include <linux/module.h>
>>
>> unsigned long a = 1;
>>
>> int module_init(void)
>> {
>> 	unsigned long b = 0;
>> 	int x;
>>
>> 	x = __test_and_set_bit(0, &b);
>> 	if (__test_and_set_bit(0, &a))
>> 		printk("x = %d\n", x);
>>
>> 	return -1;
>> }
>>
>>
>> There will likely be a way to work around my issue. Nevertheless, I
>> wondered if that patch was already considered for 2.4 inclusion. Or is
>> there no risk that in-tree code is affected?
> 
> While I remember some discussion on the subject for 2.6, I don't
> recall anything similiar in 2.4. Wouldn't you happen to build with
> gcc-3.4 ? IIRC, the clobbering changed around this version. Could

I used gcc 3.3.6-13 (Debian).

> you confirm that the patch you pointed above fixes the problem for
> your case ?

Yes. I ported the x86 part back to 2.4.33 and my problem disappeared.

Jan