Re: [PATCH kcsan 6/9] tools/memory-model: Expand the cheatsheet.txt notion of relaxed

[Boqun Feng]

On Wed, Sep 02, 2020 at 12:14:12PM +0200, peterz@xxxxxxxxxxxxx wrote:
> On Wed, Sep 02, 2020 at 11:54:48AM +0800, Boqun Feng wrote:
> > On Mon, Aug 31, 2020 at 11:20:34AM -0700, paulmck@xxxxxxxxxx wrote:
> > > From: "Paul E. McKenney" <paulmck@xxxxxxxxxx>
> > > 
> > > This commit adds a key entry enumerating the various types of relaxed
> > > operations.
> > > 
> > > Signed-off-by: Paul E. McKenney <paulmck@xxxxxxxxxx>
> > > ---
> > >  tools/memory-model/Documentation/cheatsheet.txt | 27 ++++++++++++++-----------
> > >  1 file changed, 15 insertions(+), 12 deletions(-)
> > > 
> > > diff --git a/tools/memory-model/Documentation/cheatsheet.txt b/tools/memory-model/Documentation/cheatsheet.txt
> > > index 33ba98d..31b814d 100644
> > > --- a/tools/memory-model/Documentation/cheatsheet.txt
> > > +++ b/tools/memory-model/Documentation/cheatsheet.txt
> > > @@ -5,7 +5,7 @@
> > >  
> > >  Store, e.g., WRITE_ONCE()            Y                                       Y
> > >  Load, e.g., READ_ONCE()              Y                          Y   Y        Y
> > > -Unsuccessful RMW operation           Y                          Y   Y        Y
> > > +Relaxed operation                    Y                          Y   Y        Y
> > >  rcu_dereference()                    Y                          Y   Y        Y
> > >  Successful *_acquire()               R                   Y  Y   Y   Y    Y   Y
> > >  Successful *_release()         C        Y  Y    Y     W                      Y
> > > @@ -17,14 +17,17 @@ smp_mb__before_atomic()       CP        Y  Y    Y        a  a   a   a    Y
> > >  smp_mb__after_atomic()        CP        a  a    Y        Y  Y   Y   Y    Y
> > >  
> > >  
> > > -Key:	C:	Ordering is cumulative
> > > -	P:	Ordering propagates
> > > -	R:	Read, for example, READ_ONCE(), or read portion of RMW
> > > -	W:	Write, for example, WRITE_ONCE(), or write portion of RMW
> > > -	Y:	Provides ordering
> > > -	a:	Provides ordering given intervening RMW atomic operation
> > > -	DR:	Dependent read (address dependency)
> > > -	DW:	Dependent write (address, data, or control dependency)
> > > -	RMW:	Atomic read-modify-write operation
> > > -	SELF:	Orders self, as opposed to accesses before and/or after
> > > -	SV:	Orders later accesses to the same variable
> > > +Key:	Relaxed:  A relaxed operation is either a *_relaxed() RMW
> > > +		  operation, an unsuccessful RMW operation, or one of
> > > +		  the atomic_read() and atomic_set() family of operations.
> > 
> > To be accurate, atomic_set() doesn't return any value, so it cannot be
> > ordered against DR and DW ;-)
> 
> Surely DW is valid for any store.
> 

IIUC, the DW colomn stands for whether the corresponding operation (in
this case, it's atomic_set()) is ordered any write that depends on this
operation. I don't think there is a write->write dependency, so DW for
atomic_set() should not be Y, just as the DW for WRITE_ONCE().

> > I think we can split the Relaxed family into two groups:
> > 
> > void Relaxed: atomic_set() or atomic RMW operations that don't return
> >               any value (e.g atomic_inc())
> > 
> > non-void Relaxed: a *_relaxed() RMW operation, an unsuccessful RMW
> >                   operation, or atomic_read().
> > 
> > And "void Relaxed" is similar to WRITE_ONCE(), only has "Self" and "SV"
> > equal "Y", while "non-void Relaxed" plays the same rule as "Relaxed"
> > in this patch.
> > 
> > Thoughts?
> 
> I get confused by the mention of all this atomic_read() atomic_set()
> crud in the first place, why are they called out specifically from any
> other regular load/store ?

Agreed. Probably we should fold those two operations into "Load" and
"Store" cases.

Regards,
Boqun