Missing (?) data edge
hernanponcedeleon opened this issue · comments
I am running herd7 (version 7.56.2, Rev: exported) on the litmus test below using the attached files (coming from the latest dev branch in the tree of @paulmckrcu)
C dep+plain
(*
* Result: Never
*
* This litmus test demonstrates that in LKMM, plain accesses
* carry dependencies much like accesses to registers:
* the data stored to *z1 and *z2 by P0() originates from P0()'s
* READ_ONCE(), and therefore using that data to compute the
* conditional of P0()'s if-statement creates a control dependency
* from that READ_ONCE() to P0()'s WRITE_ONCE() which is inside
* the if-statement.
*
*)
{}
P0(int *x, int *y, int *z1, int *z2)
{
int a = READ_ONCE(*x);
*z1 = a;
*z2 = *z1;
if (*z2 == 1){
WRITE_ONCE(*y, 1);
}
}
P1(int *x, int *y)
{
int r = smp_load_acquire(y);
smp_store_release(x, r);
}
exists (x=1 /\ y=1)
As the comment states, the result should not be observable. However this is what herd7 reports
> herd7 -model linux-kernel.cat -I cat/ -bell linux-kernel.bell -macros linux-kernel.def litmus/LKMM/tree/dep_plain.litmus
Test dep+plain Allowed
States 2
x=0; y=0;
x=1; y=1;
Ok
Witnesses
Positive: 1 Negative: 2
Condition exists (x=1 /\ y=1)
Observation dep+plain Sometimes 1 2
Time dep+plain 0.02
Hash=76f5e4b940dcf810444eb27dcfffc9bf
The problem seems to be that herd7 does not recognizes the data dependency between events c and d in this execution graph
dep_plain.pdf. If I split *z2 = *z1; into int t = *z1; *z2 = t;, then I get the correct result.
What is very surprising to me is that if I use the web version of herd7 and just add the following to the cat file (before the include "lock.cat")
let carry-dep = (data ; rfi)*
let addr = carry-dep ; addr
let ctrl = carry-dep ; ctrl
let data = carry-dep ; data
I also get the correct result. This suggests there is some difference between the herd7 version executed in the web-server and the one I execute locally.
Hi @hernanponcedeleon, I could not reproduce with the latest dev branch of Paul's deposit. If I am not wrong, It may be that the linux-kernel.bell file you are using in your first experiment does not contain the fix that you have added to the cat file in your second experiment.
Hi @maranget . The changes are there
> ~/git/linux/tools/memory-model$ cat linux-kernel.bell | tail -5
(* Redefine dependencies to include those carried through plain accesses *)
let carry-dep = (data ; rfi)*
let addr = carry-dep ; addr
let ctrl = carry-dep ; ctrl
let data = carry-dep ; data
Am I passing the right flags to make herd pick up the files (specially the *.bell one) from the current directory?
> ~/git/linux/tools/memory-model$ herd7 -model linux-kernel.cat -I ./ -bell linux-kernel.bell -macros linux-kernel.def dep_plain.litmus
Test dep+plain Allowed
States 2
x=0; y=0;
x=1; y=1;
Ok
Witnesses
Positive: 1 Negative: 2
Condition exists (x=1 /\ y=1)
Observation dep+plain Sometimes 1 2
Time dep+plain 0.00
Hash=76f5e4b940dcf810444eb27dcfffc9bf
I also tried using the config file directly, but I get the same result
~/git/linux/tools/memory-model$ herd7 -conf ./linux-kernel.cfg dep_plain.litmus
Test dep+plain Allowed
States 2
x=0; y=0;
x=1; y=1;
Ok
Witnesses
Positive: 1 Negative: 2
Condition exists (x=1 /\ y=1)
Observation dep+plain Sometimes 1 2
Time dep+plain 0.02
Hash=76f5e4b940dcf810444eb27dcfffc9bf
From the execution graph in my first msg, relation carry-dep is defined (as it is shown in the graph) and thus it seems to be picking up the correct *.bell file.
Please add option -debug files to see what files herd7 actually loads.
Found and opened: './linux-kernel.cat'
Found and opened: './linux-kernel.bell'
Found and opened: '/home/ponce/.opam/default/share/herdtools7/herd/stdlib.cat'
Found and opened: '/home/ponce/git/Dat3M/litmus/LKMM/tree/dep_plain.litmus'
Found and opened: './linux-kernel.def'
Found and opened: './lock.cat'
Found and opened: '/home/ponce/.opam/default/share/herdtools7/herd/cross.cat'
Found and opened: '/home/ponce/.opam/default/share/herdtools7/herd/cos-opt.cat'
Warning: File "/home/ponce/.opam/default/share/herdtools7/herd/cos-opt.cat", line 25, characters 0-19: including cross.cat another time
Warning: File "/home/ponce/.opam/default/share/herdtools7/herd/cos-opt.cat", line 25, characters 0-19: including cross.cat another time
Warning: File "/home/ponce/.opam/default/share/herdtools7/herd/cos-opt.cat", line 25, characters 0-19: including cross.cat another time
The *.bell file seems to be the correct one
Ok I have reproduced the problem. You are using the opam installed herdtools7 suite. Using the head version from this very deposit yields the correct result. I guess we need a new opam release. Thanks for the report.
I confirm I get the correct result using version 7.56+03, Rev: b69b01a281e19808300b876c1deb3967c15a2b53. Thanks @maranget for taking a look to this.