Key points of the C memory model that are relevant:
1) Memory writes can be seen in a different order in different threads.
2) When one thread (A) reads a value with acquire memory ordering that
another thread (B) has written with release memory ordering, then all
data written in thread (B) before the write that set this value will
be visible to thread (A) after that read.
3) This point requires that the read and write operate on the same
value. The guarantee is one-way: It specifies that thread (A) will
see the writes that thread (B) has performed before the specified
write. It does not specify that thread (B) will see writes that
thread (A) has performed before reading this value.
Outline of the issue:
1) While there is a memory sync at entry to the barrier, user code can
be ran after threads have all entered the barrier.
2) There are various points where a memory sync can occur after entry to
the barrier:
- One thread getting the `task_lock` mutex that another thread has
released.
- Last thread incrementing `bar->generation` with `MEMMODEL_RELEASE`
and some other thread reading it with `MEMMODEL_ACQUIRE`.
However there are code paths that can avoid these points.
3) On the code-paths that can avoid these points we could have no memory
synchronisation between a write to user data that happened in a task
executed after entry to the barrier, and some other thread running
the implicit task after the barrier. Hence that "other thread" may
read a stale value that should have been overwritten in the explicit
task.
There are two code-paths that I believe I've identified:
1) The last thread sees `task_count == 0` and increments the generation
with `MEMMODEL_RELEASE` before continuing on to the next implicit
task.
If some other thread had executed a task that wrote user data I
don't see any way in which an acquire-release ordering *from* the
thread writing user data *to* the last thread would have been formed.
2) After all threads have entered the barrier. Some thread (A) is
waiting in `do_wait`. Some other thread (B) completes a task writing
user data. Thread (B) increments the generation using
`gomp_team_barrier_done` (non atomically -- hence not allowing the
formation of any acquire-release ordering with this write). Thread
(A) reads that data with `MEMMODEL_ACQUIRE`, but since the write was
not atomic that does not form an ordering.
This patch makes two changes:
1) The write of `task_count == 0` in `gomp_barrier_handle_tasks` is done
atomically while the read of `task_count` in
`gomp_team_barrier_wait_end` is also made atomic. This addresses the
first case by forming an acquire-release ordering *from* the thread
executing tasks *to* the thread that will increment the generation
and continue.
2) The write of `bar->generation` via `gomp_team_barrier_done` called
from `gomp_barrier_handle_tasks` is done atomically. This means that
it will form an acquire-release synchronisation with the existing
atomic read of `bar->generation` in the main loop of
`gomp_team_barrier_wait_end`.
Concerns I have with this patch:
1) I've made the change as "seems correct" from reading the code and
have checked that it builds, but can't be that certain in my
modifications.
- Should I remove changes outside of any targets that I can't check
myself instead?
2) I don't understand the nvptx backend. Looks like it might be correct
since it has some kind of atomic operation on `task_count` to check
for zero. I've left it alone rather than try to understand.
3) I believe that we technically still have UB in this area (that was
around before). TSAN pointed it out in some experiments I was
running. If we are writing the value non-atomically for the
task_count != 1 case and reading the value atomically in the primary
thread then I believe this is UB simply because there is a read and a
write happening to the same location where at least one is not
atomic.
- Have not attempted to fix this because this kind of UB is already
quite widespread in libgomp (according to TSAN) -- e.g.
`task_count` is read in GOMP_taskloop and GOMP_task outside of the
`task_lock` mutex and that's technically UB because there's no
synchronisation between those reads and the writes under the mutex.
- I figure that fixing this kind of UB is a much larger job to be
taken on at a later time.
Testing done:
- Bootstrap & regtest on aarch64 and x86_64.
- With & without _LIBGOMP_CHECKING_.
- Testsuite with & without OMP_WAIT_POLICY=passive
- Cross compilation & regtest on arm.
- TSAN done on this as part of all my upstream patches.
libgomp/ChangeLog:
PR libgomp/122356
* config/gcn/bar.c (gomp_team_barrier_wait_end): Atomically read
team->task_count.
(gomp_team_barrier_wait_cancel_end): Likewise.
* config/gcn/bar.h (gomp_team_barrier_done): Atomically write
bar->generation.
* config/linux/bar.c (gomp_team_barrier_wait_end): Atomically
read team->task_count.
(gomp_team_barrier_wait_cancel_end): Likewise.
* config/linux/bar.h (gomp_team_barrier_done): Atomically write
bar->generation.
* config/posix/bar.c (gomp_team_barrier_wait_end): Atomically
read team->task_count.
(gomp_team_barrier_wait_cancel_end): Likewise.
* config/posix/bar.h (gomp_team_barrier_done): Atomically write
bar->generation.
* config/rtems/bar.h (gomp_team_barrier_done): Atomically write
bar->generation.
* task.c (gomp_barrier_handle_tasks): Atomically write
team->task_count when decrementing to zero.
* testsuite/libgomp.c/pr122356.c: New test.
Signed-off-by: Matthew Malcomson <[email protected]>
---
libgomp/config/gcn/bar.c | 8 +++++--
libgomp/config/gcn/bar.h | 5 +++-
libgomp/config/linux/bar.c | 8 +++++--
libgomp/config/linux/bar.h | 5 +++-
libgomp/config/posix/bar.c | 8 +++++--
libgomp/config/posix/bar.h | 5 +++-
libgomp/config/rtems/bar.h | 5 +++-
libgomp/task.c | 8 ++++++-
libgomp/testsuite/libgomp.c/pr122356.c | 33 ++++++++++++++++++++++++++
9 files changed, 74 insertions(+), 11 deletions(-)
create mode 100644 libgomp/testsuite/libgomp.c/pr122356.c
diff --git a/libgomp/config/gcn/bar.c b/libgomp/config/gcn/bar.c
index 05daa8fcbbc..d7e2d755685 100644
--- a/libgomp/config/gcn/bar.c
+++ b/libgomp/config/gcn/bar.c
@@ -89,7 +89,9 @@ gomp_team_barrier_wait_end (gomp_barrier_t *bar,
gomp_barrier_state_t state)
bar->awaited = bar->total;
team->work_share_cancelled = 0;
- if (__builtin_expect (team->task_count, 0))
+ unsigned task_count
+ = __atomic_load_n (&team->task_count, MEMMODEL_ACQUIRE);
+ if (__builtin_expect (task_count, 0))
{
gomp_barrier_handle_tasks (state);
state &= ~BAR_WAS_LAST;
@@ -164,7 +166,9 @@ gomp_team_barrier_wait_cancel_end (gomp_barrier_t *bar,
bar->awaited = bar->total;
team->work_share_cancelled = 0;
- if (__builtin_expect (team->task_count, 0))
+ unsigned task_count
+ = __atomic_load_n (&team->task_count, MEMMODEL_ACQUIRE);
+ if (__builtin_expect (task_count, 0))
{
gomp_barrier_handle_tasks (state);
state &= ~BAR_WAS_LAST;
diff --git a/libgomp/config/gcn/bar.h b/libgomp/config/gcn/bar.h
index 8fdd6465822..1f28f579092 100644
--- a/libgomp/config/gcn/bar.h
+++ b/libgomp/config/gcn/bar.h
@@ -162,7 +162,10 @@ gomp_team_barrier_cancelled (gomp_barrier_t *bar)
static inline void
gomp_team_barrier_done (gomp_barrier_t *bar, gomp_barrier_state_t state)
{
- bar->generation = (state & -BAR_INCR) + BAR_INCR;
+ /* Need the atomic store for acquire-release synchronisation with the
+ load in `gomp_team_barrier_wait_{cancel_,}end`. See PR112356 */
+ __atomic_store_n (&bar->generation, (state & -BAR_INCR) + BAR_INCR,
+ MEMMODEL_RELEASE);
}
static inline bool
diff --git a/libgomp/config/linux/bar.c b/libgomp/config/linux/bar.c
index 2c9d1ce894d..9eaec0e5f23 100644
--- a/libgomp/config/linux/bar.c
+++ b/libgomp/config/linux/bar.c
@@ -90,7 +90,9 @@ gomp_team_barrier_wait_end (gomp_barrier_t *bar,
gomp_barrier_state_t state)
bar->awaited = bar->total;
team->work_share_cancelled = 0;
- if (__builtin_expect (team->task_count, 0))
+ unsigned task_count
+ = __atomic_load_n (&team->task_count, MEMMODEL_ACQUIRE);
+ if (__builtin_expect (task_count, 0))
{
gomp_barrier_handle_tasks (state);
state &= ~BAR_WAS_LAST;
@@ -154,7 +156,9 @@ gomp_team_barrier_wait_cancel_end (gomp_barrier_t *bar,
bar->awaited = bar->total;
team->work_share_cancelled = 0;
- if (__builtin_expect (team->task_count, 0))
+ unsigned task_count
+ = __atomic_load_n (&team->task_count, MEMMODEL_ACQUIRE);
+ if (__builtin_expect (task_count, 0))
{
gomp_barrier_handle_tasks (state);
state &= ~BAR_WAS_LAST;
diff --git a/libgomp/config/linux/bar.h b/libgomp/config/linux/bar.h
index 9fb514526c4..faa03746d8f 100644
--- a/libgomp/config/linux/bar.h
+++ b/libgomp/config/linux/bar.h
@@ -162,7 +162,10 @@ gomp_team_barrier_cancelled (gomp_barrier_t *bar)
static inline void
gomp_team_barrier_done (gomp_barrier_t *bar, gomp_barrier_state_t state)
{
- bar->generation = (state & -BAR_INCR) + BAR_INCR;
+ /* Need the atomic store for acquire-release synchronisation with the
+ load in `gomp_team_barrier_wait_{cancel_,}end`. See PR112356 */
+ __atomic_store_n (&bar->generation, (state & -BAR_INCR) + BAR_INCR,
+ MEMMODEL_RELEASE);
}
static inline bool
diff --git a/libgomp/config/posix/bar.c b/libgomp/config/posix/bar.c
index d728c9c0afe..a86b2f38c2d 100644
--- a/libgomp/config/posix/bar.c
+++ b/libgomp/config/posix/bar.c
@@ -123,7 +123,9 @@ gomp_team_barrier_wait_end (gomp_barrier_t *bar,
gomp_barrier_state_t state)
struct gomp_team *team = thr->ts.team;
team->work_share_cancelled = 0;
- if (team->task_count)
+ unsigned task_count
+ = __atomic_load_n (&team->task_count, MEMMODEL_ACQUIRE);
+ if (task_count)
{
gomp_barrier_handle_tasks (state);
if (n > 0)
@@ -185,7 +187,9 @@ gomp_team_barrier_wait_cancel_end (gomp_barrier_t *bar,
struct gomp_team *team = thr->ts.team;
team->work_share_cancelled = 0;
- if (team->task_count)
+ unsigned task_count
+ = __atomic_load_n (&team->task_count, MEMMODEL_ACQUIRE);
+ if (task_count)
{
gomp_barrier_handle_tasks (state);
if (n > 0)
diff --git a/libgomp/config/posix/bar.h b/libgomp/config/posix/bar.h
index a48e99488bb..35b94e43ce2 100644
--- a/libgomp/config/posix/bar.h
+++ b/libgomp/config/posix/bar.h
@@ -152,7 +152,10 @@ gomp_team_barrier_cancelled (gomp_barrier_t *bar)
static inline void
gomp_team_barrier_done (gomp_barrier_t *bar, gomp_barrier_state_t state)
{
- bar->generation = (state & -BAR_INCR) + BAR_INCR;
+ /* Need the atomic store for acquire-release synchronisation with the
+ load in `gomp_team_barrier_wait_{cancel_,}end`. See PR112356 */
+ __atomic_store_n (&bar->generation, (state & -BAR_INCR) + BAR_INCR,
+ MEMMODEL_RELEASE);
}
static inline bool
diff --git a/libgomp/config/rtems/bar.h b/libgomp/config/rtems/bar.h
index ea39679e36d..5c8c074c08f 100644
--- a/libgomp/config/rtems/bar.h
+++ b/libgomp/config/rtems/bar.h
@@ -164,7 +164,10 @@ gomp_team_barrier_cancelled (gomp_barrier_t *bar)
static inline void
gomp_team_barrier_done (gomp_barrier_t *bar, gomp_barrier_state_t state)
{
- bar->generation = (state & -BAR_INCR) + BAR_INCR;
+ /* Need the atomic store for acquire-release synchronisation with the
+ load in `gomp_team_barrier_wait_{cancel_,}end`. See PR112356 */
+ __atomic_store_n (&bar->generation, (state & -BAR_INCR) + BAR_INCR,
+ MEMMODEL_RELEASE);
}
static inline bool
diff --git a/libgomp/task.c b/libgomp/task.c
index b7f4b8220c2..5965e781f7e 100644
--- a/libgomp/task.c
+++ b/libgomp/task.c
@@ -1702,7 +1702,13 @@ gomp_barrier_handle_tasks (gomp_barrier_state_t state)
if (do_wake > new_tasks)
do_wake = new_tasks;
}
- --team->task_count;
+ /* Need to use RELEASE to sync with barrier read outside of the
+ tasking code (See PR122356). Only care when decrementing to zero
+ because that's what the barrier cares about. */
+ if (team->task_count == 1)
+ __atomic_store_n (&team->task_count, 0, MEMMODEL_RELEASE);
+ else
+ team->task_count--;
}
}
}
diff --git a/libgomp/testsuite/libgomp.c/pr122356.c
b/libgomp/testsuite/libgomp.c/pr122356.c
new file mode 100644
index 00000000000..5dbf418619c
--- /dev/null
+++ b/libgomp/testsuite/libgomp.c/pr122356.c
@@ -0,0 +1,33 @@
+#include <omp.h>
+
+void abort ();
+
+#define NUM_THREADS 8
+unsigned full_data[NUM_THREADS] = {0};
+void
+test ()
+{
+#pragma omp parallel num_threads(8)
+ {
+#pragma omp for
+ for (int i = 0; i < 10; i++)
+#pragma omp task
+ {
+ full_data[omp_get_thread_num ()] += 1;
+ }
+#pragma omp barrier
+
+ unsigned total = 0;
+ for (int i = 0; i < NUM_THREADS; i++)
+ total += full_data[i];
+
+ if (total != 10)
+ abort ();
+ }
+}
+
+int
+main ()
+{
+ test ();
+}
