Documentation/litmus-tests: Demonstrate unordered failing cmpxchg
This commit adds four litmus tests showing that a failing cmpxchg()
operation is unordered unless followed by an smp_mb__after_atomic()
operation.
Suggested-by: Frederic Weisbecker <frederic@kernel.org>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
Cc: Alan Stern <stern@rowland.harvard.edu>
Cc: Will Deacon <will@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Boqun Feng <boqun.feng@gmail.com>
Cc: Nicholas Piggin <npiggin@gmail.com>
Cc: David Howells <dhowells@redhat.com>
Cc: Jade Alglave <j.alglave@ucl.ac.uk>
Cc: Luc Maranget <luc.maranget@inria.fr>
Cc: "Paul E. McKenney" <paulmck@kernel.org>
Cc: Akira Yokosawa <akiyks@gmail.com>
Cc: Daniel Lustig <dlustig@nvidia.com>
Cc: Joel Fernandes <joel@joelfernandes.org>
Cc: Mark Rutland <mark.rutland@arm.com>
Cc: Jonathan Corbet <corbet@lwn.net>
Cc: <linux-arch@vger.kernel.org>
Cc: <linux-doc@vger.kernel.org>
diff --git a/Documentation/litmus-tests/README b/Documentation/litmus-tests/README
index 5c8915e..6c666f3 100644
--- a/Documentation/litmus-tests/README
+++ b/Documentation/litmus-tests/README
@@ -21,34 +21,50 @@
Test that atomic_set() cannot break the atomicity of atomic RMWs.
NOTE: Require herd7 7.56 or later which supports "(void)expr".
+cmpxchg-fail-ordered-1.litmus
+ Demonstrate that a failing cmpxchg() operation acts as a full barrier
+ when followed by smp_mb__after_atomic().
+
+cmpxchg-fail-ordered-2.litmus
+ Demonstrate that a failing cmpxchg() operation acts as an acquire
+ operation when followed by smp_mb__after_atomic().
+
+cmpxchg-fail-unordered-1.litmus
+ Demonstrate that a failing cmpxchg() operation does not act as a
+ full barrier.
+
+cmpxchg-fail-unordered-2.litmus
+ Demonstrate that a failing cmpxchg() operation does not act as an
+ acquire operation.
+
locking (/locking directory)
----------------------------
DCL-broken.litmus
- Demonstrates that double-checked locking needs more than just
- the obvious lock acquisitions and releases.
+ Demonstrates that double-checked locking needs more than just
+ the obvious lock acquisitions and releases.
DCL-fixed.litmus
- Demonstrates corrected double-checked locking that uses
- smp_store_release() and smp_load_acquire() in addition to the
- obvious lock acquisitions and releases.
+ Demonstrates corrected double-checked locking that uses
+ smp_store_release() and smp_load_acquire() in addition to the
+ obvious lock acquisitions and releases.
RM-broken.litmus
- Demonstrates problems with "roach motel" locking, where code is
- freely moved into lock-based critical sections. This example also
- shows how to use the "filter" clause to discard executions that
- would be excluded by other code not modeled in the litmus test.
- Note also that this "roach motel" optimization is emulated by
- physically moving P1()'s two reads from x under the lock.
+ Demonstrates problems with "roach motel" locking, where code is
+ freely moved into lock-based critical sections. This example also
+ shows how to use the "filter" clause to discard executions that
+ would be excluded by other code not modeled in the litmus test.
+ Note also that this "roach motel" optimization is emulated by
+ physically moving P1()'s two reads from x under the lock.
- What is a roach motel? This is from an old advertisement for
- a cockroach trap, much later featured in one of the "Men in
- Black" movies. "The roaches check in. They don't check out."
+ What is a roach motel? This is from an old advertisement for
+ a cockroach trap, much later featured in one of the "Men in
+ Black" movies. "The roaches check in. They don't check out."
RM-fixed.litmus
- The counterpart to RM-broken.litmus, showing P0()'s two loads from
- x safely outside of the critical section.
+ The counterpart to RM-broken.litmus, showing P0()'s two loads from
+ x safely outside of the critical section.
RCU (/rcu directory)
diff --git a/Documentation/litmus-tests/atomic/cmpxchg-fail-ordered-1.litmus b/Documentation/litmus-tests/atomic/cmpxchg-fail-ordered-1.litmus
new file mode 100644
index 0000000..3df1d14
--- /dev/null
+++ b/Documentation/litmus-tests/atomic/cmpxchg-fail-ordered-1.litmus
@@ -0,0 +1,34 @@
+C cmpxchg-fail-ordered-1
+
+(*
+ * Result: Never
+ *
+ * Demonstrate that a failing cmpxchg() operation will act as a full
+ * barrier when followed by smp_mb__after_atomic().
+ *)
+
+{}
+
+P0(int *x, int *y, int *z)
+{
+ int r0;
+ int r1;
+
+ WRITE_ONCE(*x, 1);
+ r1 = cmpxchg(z, 1, 0);
+ smp_mb__after_atomic();
+ r0 = READ_ONCE(*y);
+}
+
+P1(int *x, int *y, int *z)
+{
+ int r0;
+
+ WRITE_ONCE(*y, 1);
+ r1 = cmpxchg(z, 1, 0);
+ smp_mb__after_atomic();
+ r0 = READ_ONCE(*x);
+}
+
+locations[0:r1;1:r1]
+exists (0:r0=0 /\ 1:r0=0)
diff --git a/Documentation/litmus-tests/atomic/cmpxchg-fail-ordered-2.litmus b/Documentation/litmus-tests/atomic/cmpxchg-fail-ordered-2.litmus
new file mode 100644
index 0000000..5414604
--- /dev/null
+++ b/Documentation/litmus-tests/atomic/cmpxchg-fail-ordered-2.litmus
@@ -0,0 +1,30 @@
+C cmpxchg-fail-ordered-2
+
+(*
+ * Result: Never
+ *
+ * Demonstrate use of smp_mb__after_atomic() to make a failing cmpxchg
+ * operation have acquire ordering.
+ *)
+
+{}
+
+P0(int *x, int *y)
+{
+ int r0;
+ int r1;
+
+ WRITE_ONCE(*x, 1);
+ r1 = cmpxchg(y, 0, 1);
+}
+
+P1(int *x, int *y)
+{
+ int r0;
+
+ r1 = cmpxchg(y, 0, 1);
+ smp_mb__after_atomic();
+ r2 = READ_ONCE(*x);
+}
+
+exists (0:r1=0 /\ 1:r1=1 /\ 1:r2=0)
diff --git a/Documentation/litmus-tests/atomic/cmpxchg-fail-unordered-1.litmus b/Documentation/litmus-tests/atomic/cmpxchg-fail-unordered-1.litmus
new file mode 100644
index 0000000..a727ce2
--- /dev/null
+++ b/Documentation/litmus-tests/atomic/cmpxchg-fail-unordered-1.litmus
@@ -0,0 +1,33 @@
+C cmpxchg-fail-unordered-1
+
+(*
+ * Result: Sometimes
+ *
+ * Demonstrate that a failing cmpxchg() operation does not act as a
+ * full barrier. (In contrast, a successful cmpxchg() does act as a
+ * full barrier.)
+ *)
+
+{}
+
+P0(int *x, int *y, int *z)
+{
+ int r0;
+ int r1;
+
+ WRITE_ONCE(*x, 1);
+ r1 = cmpxchg(z, 1, 0);
+ r0 = READ_ONCE(*y);
+}
+
+P1(int *x, int *y, int *z)
+{
+ int r0;
+
+ WRITE_ONCE(*y, 1);
+ r1 = cmpxchg(z, 1, 0);
+ r0 = READ_ONCE(*x);
+}
+
+locations[0:r1;1:r1]
+exists (0:r0=0 /\ 1:r0=0)
diff --git a/Documentation/litmus-tests/atomic/cmpxchg-fail-unordered-2.litmus b/Documentation/litmus-tests/atomic/cmpxchg-fail-unordered-2.litmus
new file mode 100644
index 0000000..a245bac
--- /dev/null
+++ b/Documentation/litmus-tests/atomic/cmpxchg-fail-unordered-2.litmus
@@ -0,0 +1,30 @@
+C cmpxchg-fail-unordered-2
+
+(*
+ * Result: Sometimes
+ *
+ * Demonstrate that a failing cmpxchg() operation does not act as either
+ * an acquire release operation. (In contrast, a successful cmpxchg()
+ * does act as both an acquire and a release operation.)
+ *)
+
+{}
+
+P0(int *x, int *y)
+{
+ int r0;
+ int r1;
+
+ WRITE_ONCE(*x, 1);
+ r1 = cmpxchg(y, 0, 1);
+}
+
+P1(int *x, int *y)
+{
+ int r0;
+
+ r1 = cmpxchg(y, 0, 1);
+ r2 = READ_ONCE(*x);
+}
+
+exists (0:r1=0 /\ 1:r1=1 /\ 1:r2=0)
