Re: [PATCH 13/15] x86: add cmpxchg_flag() variant

[Jeremy Fitzhardinge <jeremy@goop.org>]

On 08/23/2011 12:01 PM, Christoph Lameter wrote:
> On Mon, 22 Aug 2011, Jeremy Fitzhardinge wrote:
>
>> Most callers of cmpxchg() direcly compare RETURN with OLD to see if it was
>> successful.  This results in unnecessary conditional comparisons
>> and conditionals since the cmpxchg instruction directly sets the flags
>> to indicate success/failure.
>> Add cmpxchg_flag() variants which return a boolean flag directly indicating
>> success.  Unfortunately an asm() statement can't directly export status
>> status flags, but sete isn't too bad.
> And so what happens through this patch is that a cmp with a value that is
> likely in a register is replaced by a sete. Is there really a benefit?

There's definitely a benefit for functions which return a bool resulting
from cmpxchg, which occurs in a few places.

> What I wish we would have is the actual use of the processor flag.
>
> if (cmpxchg_flags(....)) {
> }
>
> where the cmpxchg is followed immediately by a jump. I tried in the past
> to pass a goto label to cmpxchg but that did not work.

Yes, that would ideal.  The closest you can get is asm goto(), but the
syntax for that would be awful; something like:

#define cmpxchg_jump(ptr, old, new, fail)\
	asm goto (...)


:
:

again:
	old = *thingp;
	new = frobulate(old);
	cmpxchg_jump(thingp, old, new, again);
	/* worked */


Would this be useful enough?

Also, cmpxchg_flag() is limited to arch/x86; I hadn't looked into what
it would take to make it a generic part of the cmpxchg API for the rest
of the kernel.

    J