Re: [PATCH v2 2/2] tools/memory-model: Make ppo a subrelation of po

[Alan Stern]

On Sun, Jan 29, 2023 at 08:36:45PM -0800, Paul E. McKenney wrote:
> On Sun, Jan 29, 2023 at 09:39:17PM -0500, Alan Stern wrote:
> > On Sun, Jan 29, 2023 at 11:19:32PM +0100, Jonas Oberhauser wrote:
> > > I see now. Somehow I thought stores must execute in program order, but I
> > > guess it doesn't make sense.
> > > In that sense, W ->xbstar&int X always means W propagates to X's CPU before
> > > X executes.
> > 
> > It also means any write that propagates to W's CPU before W executes 
> > also propagates to X's CPU before X executes (because it's the same CPU 
> > and W executes before X).
> > 
> > > > Ideally we would fix this by changing the definition of po-rel to:
> > > > 
> > > > 	[M] ; (xbstar & int) ; [Release]
> > > > 
> > > > (This is closely related to the use of (xbstar & int) in the definition
> > > > of vis that you asked about.)
> > > 
> > > This misses the property of release stores that any po-earlier store must
> > > also execute before the release store.
> > 
> > I should have written:
> > 
> > 	[M] ; (po | (xbstar & int)) ; [Release]
> > 
> > > Perhaps it could be changed to the old  po-rel | [M] ; (xbstar & int) ;
> > > [Release] but then one could instead move this into the definition of
> > > cumul-fence.
> > > In fact you'd probably want this for all the propagation fences, so
> > > cumul-fence and pb should be the right place.
> > > 
> > > > Unfortunately we can't do this, because
> > > > po-rel has to be defined long before xbstar.
> > > 
> > > You could do it, by turning the relation into one massive recursive
> > > definition.
> > 
> > Which would make pretty much the entire memory model one big recursion.  
> > I do not want to do that.
> > 
> > > Thinking about what the options are:
> > > 1) accept the difference and run with it by making it consistent inside the
> > > axiomatic model
> > > 2) fix it through the recursive definition, which seems to be quite ugly but
> > > also consistent with the power operational model as far as I can tell
> > > 3) weaken the operational model... somehow
> > > 4) just ignore the anomaly
> > > 5) ???
> > > 
> > > Currently my least favorite option is 4) since it seems a bit off that the
> > > reasoning applies in one specific case of LKMM, more specifically the data
> > > race definition which should be equivalent to "the order of the two races
> > > isn't fixed", but here the order isn't fixed but it's a data race.
> > > I think the patch happens to almost do 1) because the xbstar&int at the end
> > > should already imply ordering through the prop&int <= hb rule.
> > > What would remain is to also exclude rcu-fence somehow.
> > 
> > IMO 1) is the best choice.
> > 
> > Alan
> > 
> > PS: For the record, here's a simpler litmus test to illustrates the 
> > failing.  The idea is that Wz=1 is reordered before the store-release, 
> > so it ought to propagate before Wy=1.  The LKMM does not require this.
> 
> In PowerPC terms, would this be like having the Wz=1 being reorders
> before the Wy=1, but not before the lwsync instruction preceding the
> Wy=1 that made it be a release store?

No, it would be like having the Wz=1 reordered before the Rx=1, 
therefore before the lwsync.  Obviously this can't ever happen on 
PowerPC.

Alan

> If so, we might have to keep this quirk.
> 
> 							Thanx, Paul
> 
> > C before-release
> > 
> > {}
> > 
> > P0(int *x, int *y, int *z)
> > {
> > 	int r1;
> > 
> > 	r1 = READ_ONCE(*x);
> > 	smp_store_release(y, 1);
> > 	WRITE_ONCE(*z, 1);
> > }
> > 
> > P1(int *x, int *y, int *z)
> > {
> > 	int r2;
> > 
> > 	r2 = READ_ONCE(*z);
> > 	WRITE_ONCE(*x, r2);
> > }
> > 
> > P2(int *x, int *y, int *z)
> > {
> > 	int r3;
> > 	int r4;
> > 
> > 	r3 = READ_ONCE(*y);
> > 	smp_rmb();
> > 	r4 = READ_ONCE(*z);
> > }
> > 
> > exists (0:r1=1 /\ 2:r3=1 /\ 2:r4=0)