[OpenMP] Fix race condition in omp_init_lock

This is a partial fix for bug 34050.

This prevents callers of omp_set_lock (which does not hold __kmp_global_lock)
from ever seeing an uninitialized version of __kmp_i_lock_table.table.

It does not solve a use-after-free race condition if omp_set_lock obtains a
pointer to __kmp_i_lock_table.table before it is updated and then attempts to
dereference afterwards. That race is far less likely and can be handled in a
separate patch.

The unit test usually segfaults on the current trunk revision. It passes with
the patch.

Patch by Adam Azarchs

Differential Revision: https://reviews.llvm.org/D39439

llvm-svn: 317115

openmp/runtime/src/kmp_lock.cpp

@@ -3058,11 +3058,12 @@ kmp_indirect_lock_t *__kmp_allocate_indirect_lock(void **user_lock,
     if (idx == __kmp_i_lock_table.size) {
       // Double up the space for block pointers
       int row = __kmp_i_lock_table.size / KMP_I_LOCK_CHUNK;
-      kmp_indirect_lock_t **old_table = __kmp_i_lock_table.table;
-      __kmp_i_lock_table.table = (kmp_indirect_lock_t **)__kmp_allocate(
+      kmp_indirect_lock_t **new_table = (kmp_indirect_lock_t **)__kmp_allocate(
           2 * row * sizeof(kmp_indirect_lock_t *));
-      KMP_MEMCPY(__kmp_i_lock_table.table, old_table,
+      KMP_MEMCPY(new_table, __kmp_i_lock_table.table,
                  row * sizeof(kmp_indirect_lock_t *));
+      kmp_indirect_lock_t **old_table = __kmp_i_lock_table.table;
+      __kmp_i_lock_table.table = new_table;
       __kmp_free(old_table);
       // Allocate new objects in the new blocks
       for (int i = row; i < 2 * row; ++i)

openmp/runtime/test/lock/omp_init_lock.c

@@ -0,0 +1,42 @@
+// RUN: %libomp-compile-and-run
+#include "omp_testsuite.h"
+#include <stdio.h>
+
+// This should be slightly less than KMP_I_LOCK_CHUNK, which is 1024
+#define LOCKS_PER_ITER 1000
+#define ITERATIONS (REPETITIONS + 1)
+
+// This tests concurrently using locks on one thread while initializing new
+// ones on another thread.  This exercises the global lock pool.
+int test_omp_init_lock() {
+  int i;
+  omp_lock_t lcks[ITERATIONS * LOCKS_PER_ITER];
+#pragma omp parallel for schedule(static) num_threads(NUM_TASKS)
+  for (i = 0; i < ITERATIONS; i++) {
+    int j;
+    omp_lock_t *my_lcks = &lcks[i * LOCKS_PER_ITER];
+    for (j = 0; j < LOCKS_PER_ITER; j++) {
+      omp_init_lock(&my_lcks[j]);
+    }
+    for (j = 0; j < LOCKS_PER_ITER * 100; j++) {
+      omp_set_lock(&my_lcks[j % LOCKS_PER_ITER]);
+      omp_unset_lock(&my_lcks[j % LOCKS_PER_ITER]);
+    }
+  }
+  // Wait until all repititions are done.  The test is exercising growth of
+  // the global lock pool, which does not shrink when no locks are allocated.
+  {
+    int j;
+    for (j = 0; j < ITERATIONS * LOCKS_PER_ITER; j++) {
+      omp_destroy_lock(&lcks[j]);
+    }
+  }
+
+  return 0;
+}
+
+int main() {
+  // No use repeating this test, since it's exercising a private global pool
+  // which is not reset between test iterations.
+  return test_omp_init_lock();
+}