> I needed an atomic test-and-set for the VAX, and I believe I've got
> it right with the following -- [...]

> static int tas(int *lock) {
>   asm("movl 4(ap), r0");
>   asm("bbssi $0, (r0), 1f");
>   asm("clrl r0");
>   asm("1: ret");
> }

Your assumptions about bbssi are *supposed* to be valid.  I am not
experienced enough with any model of VAX to know whether they actually
are...but if not, I'm sure it would be considered a bug in the
implementation.  Indeed, the operation is *supposed* to be interlocked
even against other CPUs in a multiprocessor machine.

> While I'm asking, how can I make this secure against GCC
> optimization?  The damn thing inlines my assembly code at
> optimization levels above -O2, which breaks this stuff completely.

You might try using a single asm:

	asm("movl 4(ap),r0; bbssi $0,(r0),1f; clrl r0; 1:ret");

You really _ought_ to use gcc's syntax for referring to variables
instead of assuming 4(ap), which I think is probably the real source of
the breakage you're seeing when the code gets inlined.  Here's my guess
at the sort of thing you'd have to write; someone who actually knows
the extended-asm syntax should check it:

static int tas(int *lock) {
	int rv;
	asm(	"movl %1,r0; bbssi $0,(r0),1f; clrl r0; 1:movl r0,%0"
		: "=g" (rv)
		: "g" (lock)
		: "r0"	);
	return(rv);
}

					der Mouse

			       mouse@rodents.montreal.qc.ca
		     7D C8 61 52 5D E7 2D 39  4E F1 31 3E E8 B3 27 4B