Re: Question about ISA2+pooncelock+pooncelock+pombonce litmus

[Paul E. McKenney]

On Mon, Mar 18, 2024 at 01:47:43AM +0800, Alan Huang wrote:
> Hi,
> 
> I’m playing with the LKMM, then I saw the ISA2+pooncelock+pooncelock+pombonce.
> 
> The original litmus is as follows:
> ------------------------------------------------------
> P0(int *x, int *y, spinlock_t *mylock)
> {
> 	spin_lock(mylock);
> 	WRITE_ONCE(*x, 1);
> 	WRITE_ONCE(*y, 1);
> 	spin_unlock(mylock);
> }
> 
> P1(int *y, int *z, spinlock_t *mylock)
> {
> 	int r0;
> 
> 	spin_lock(mylock);
> 	r0 = READ_ONCE(*y);
> 	WRITE_ONCE(*z, 1);
> 	spin_unlock(mylock);
> }
> 
> P2(int *x, int *z)
> {
> 	int r1;
> 	int r2;
> 
> 	r2 = READ_ONCE(*z);
> 	smp_mb();
> 	r1 = READ_ONCE(*x);
> }
> 
> exists (1:r0=1 /\ 2:r2=1 /\ 2:r1=0)
> ------------------------------------------------------
> Of course, the result is Never. 
> 
> But when I delete P0’s spin_lock and P1’s spin_unlock:
> -------------------------------------------------------
> P0(int *x, int *y, spinlock_t *mylock)
> {
> 	WRITE_ONCE(*x, 1);
> 	WRITE_ONCE(*y, 1);
> 	spin_unlock(mylock);
> }
> 
> P1(int *y, int *z, spinlock_t *mylock)
> {
> 	int r0;
> 
> 	spin_lock(mylock);
> 	r0 = READ_ONCE(*y);
> 	WRITE_ONCE(*z, 1);
> }
> 
> P2(int *x, int *z)
> {
> 	int r1;
> 	int r2;
> 
> 	r2 = READ_ONCE(*z);
> 	smp_mb();
> 	r1 = READ_ONCE(*x);
> }
> 
> exists (1:r0=1 /\ 2:r2=1 /\ 2:r1=0)
> ------------------------------------------------------
> Then herd told me the result is Sometimes.

You mean like this?

Test ISA2+pooncelock+pooncelock+pombonce Allowed
States 8
1:r0=0; 2:r1=0; 2:r2=0;
1:r0=0; 2:r1=0; 2:r2=1;
1:r0=0; 2:r1=1; 2:r2=0;
1:r0=0; 2:r1=1; 2:r2=1;
1:r0=1; 2:r1=0; 2:r2=0;
1:r0=1; 2:r1=0; 2:r2=1;
1:r0=1; 2:r1=1; 2:r2=0;
1:r0=1; 2:r1=1; 2:r2=1;
Ok
Witnesses
Positive: 1 Negative: 7
Flag unmatched-unlock
Condition exists (1:r0=1 /\ 2:r2=1 /\ 2:r1=0)
Observation ISA2+pooncelock+pooncelock+pombonce Sometimes 1 7
Time ISA2+pooncelock+pooncelock+pombonce 0.01
Hash=f55b8515e48310f812aa676084f2cc88

> Is this expected? 

There are no locks held initially, so why can't the following
sequence of events unfold:

o	P1() acquires the lock.

o	P0() does WRITE_ONCE(*y, 1). (Yes, out of order)

o	P1() does READ_ONCE(*y), and gets 1.

o	P1() does WRITE_ONCE(*z, 1).

o	P2() does READ_ONCE(*z) and gets 1.

o	P2() does smp_mb(), but there is nothing to order with.

o	P2() does READ_ONCE(*x) and gets 0.

o	P0() does WRITE_ONCE(*x, 1), but too late to affect P2().

o	P0() releases the lock that is does not hold, which is why you see
	the "Flag unmatched-unlock" in the output.  LKMM is complaining
	that the litmus test is not legitimate, and rightly so!

Or am I missing your point?

							Thanx, Paul