Re: [PATCH] WorkStruct: Implement generic UP cmpxchg() where an arch doesn't support it

[Matthew Wilcox <matthew@wil.cx>]

On Wed, Dec 06, 2006 at 11:29:42AM -0800, Christoph Lameter wrote:
> On Wed, 6 Dec 2006, Matthew Wilcox wrote:
> 
> > It's just been pointed out to me that the parisc one isn't safe.
> > 
> > <dhowells> imagine variable X is set to 3
> > <dhowells> CPU A issues cmpxchg(&X, 3, 5)
> > <dhowells> you'd expect that to change X to 5
> > <dhowells> but what if CPU B assigns 6 to X between cmpxchg reading X
> > and it setting X?
> 
> The same could happen with a regular cmpxchg. Cmpxchg changes it to 5 and 
> then other cpu performs a store before the next instruction.

For someone who's advocating use of cmpxchg, it seems you don't
understand its semantics!  In the scenario dhowells pointed out, X would
be left set to 5.  X should have the value 6 under any legitimate
implementation:

CPU A		CPU B
cmpxchg(3,5)
		X = 6


CPU A		CPU B
		X = 6
cmpxhcg(3,5)


CPU A
cmpxchg(3,
		X = 6
5)


Given that even yourself got confused about how to use it, perhaps it's
not a good idea to expose this primitive to most programmers anyway?