[tsan] Fix CallocOverflow unit test on OS X

[kubamracek]

Summary

On OS X, interceptors don't work in unit tests, so calloc() calls the system allocator. We need to use user_calloc() instead.

[dvyukov]

Can we use __interceptor_calloc instead like in other tests? Behavior in such case will be closer to user programs as the interceptor can also contain some logic.

[kubamracek]

In D14918#294890, @dvyukov wrote:

Can we use __interceptor_calloc instead like in other tests? Behavior in such case will be closer to user programs as the interceptor can also contain some logic.

Sure, I'll update the patch. (I thought user_calloc is more consistent with the rest of this file, which uses user_alloc and user_free.)

[dvyukov]

Sure, I'll update the patch. (I thought user_calloc is more consistent with the rest of this file, which uses user_alloc and user_free.)

Then, you can submit it as is.

The story behind these tests is somewhat moot. We used them when output tests did not exist. Ideally, they are rewritten to output tests. For output tests the idea is clear: write what user will write, test what matters for user.

compiler-rt/trunk/lib/tsan/tests/unit/tsan_mman_test.cc

 }
 
 TEST(Mman, CallocOverflow) {
 #if SANITIZER_DEBUG
   // EXPECT_DEATH clones a thread with 4K stack,
   // which is overflown by tsan memory accesses functions in debug mode.
   return;
 #endif
+  ThreadState *thr = cur_thread();
+  uptr pc = 0;
   size_t kArraySize = 4096;
   volatile size_t kMaxSizeT = std::numeric_limits<size_t>::max();
   volatile size_t kArraySize2 = kMaxSizeT / kArraySize + 10;
   volatile void *p = NULL;
-  EXPECT_DEATH(p = calloc(kArraySize, kArraySize2),
+  EXPECT_DEATH(p = user_calloc(thr, pc, kArraySize, kArraySize2),
                "allocator is terminating the process instead of returning 0");
   EXPECT_EQ(0L, p);
 }
 
 }  // namespace __tsan