Added interception of _os_nospin_lock_lock to fix large atomic storage on macOS

Author: drowaudio

src/rtcheck.cpp

@@ -260,7 +260,7 @@ struct check_flags_tests
         assert(are_all_bits_enabled (to_underlying (check_flags::syscall) + to_underlying (check_flags::openat), 0b0));
         auto a1 = to_underlying (check_flags::syscall);
         auto a2 = to_underlying (check_flags::openat);
-        assert(! are_all_bits_enabled (to_underlying (check_flags::syscall) + to_underlying (check_flags::openat), 0b10001000000000000000000000000000000000));
+        assert(! are_all_bits_enabled (to_underlying (check_flags::syscall) + to_underlying (check_flags::openat), 0b100010000000000000000000000000000000000));
     }
 };
 
@@ -687,6 +687,22 @@ INTERCEPTOR(void, os_unfair_lock_lock, os_unfair_lock_t lock)
 
 #pragma clang diagnostic pop
 
+// Newer macOS versions use an internal _os_nospin_lock_lock which is interecpted as per LLVM's RTSan in this commit:
+// https://code.ornl.gov/llvm-doe/llvm-project/-/commit/481a55a3d9645a6bc1540d326319b78ad8ed8db1
+extern "C" {
+    // A pointer to this type is in the interface for `_os_nospin_lock_lock`, but
+    // it's an internal implementation detail of `os/lock.c` on Darwin, and
+    // therefore not available in any headers. As a workaround, we forward declare
+    // it here, which is enough to facilitate interception of _os_nospin_lock_lock.
+    struct _os_nospin_lock_s;
+    using _os_nospin_lock_t = _os_nospin_lock_s *;
+}
+
+INTERCEPTOR(void, _os_nospin_lock_lock, _os_nospin_lock_t lock) {
+    log_function_if_realtime_context_and_enabled (rtc::check_flags::os_unfair_lock_lock, __func__);
+    return REAL(_os_nospin_lock_lock)(lock);
+}
+
 #endif
 
 //==============================================================================

src/rtcheck.h

@@ -108,40 +108,42 @@ namespace rtc
         futex                           = 1 << 23,      // linux only
         OSSpinLockLock                  = 1 << 24,      // apple only
         os_unfair_lock_lock             = 1 << 25,      // apple only
+        _os_nospin_lock_lock            = 1 << 26,      // apple only
 
         threads = pthread_create | pthread_mutex_lock | pthread_mutex_unlock | pthread_join
                     | pthread_cond_signal | pthread_cond_broadcast | pthread_cond_wait
                     | pthread_rwlock_init | pthread_rwlock_destroy | pthread_cond_timedwait
                     | pthread_rwlock_rdlock | pthread_rwlock_unlock | pthread_rwlock_wrlock
-                    | pthread_spin_lock | futex | OSSpinLockLock | os_unfair_lock_lock,
+                    | pthread_spin_lock | futex | OSSpinLockLock | os_unfair_lock_lock
+                    | _os_nospin_lock_lock,
 
         //==============================================================================
         // sleeping
         //==============================================================================
-        sleep                           = 1 << 26,
-        usleep                          = 1 << 27,
-        nanosleep                       = 1 << 28,
+        sleep                           = 1 << 27,
+        usleep                          = 1 << 28,
+        nanosleep                       = 1 << 29,
 
         sleeping                        = sleep | usleep | nanosleep,
 
         //==============================================================================
         // files
         //==============================================================================
-        stat                            = 1 << 29,
-        fstat                           = 1 << 30,
-        open                            = 1ull << 31,
-        fopen                           = 1ull << 32,
-        openat                          = 1ull << 33,
-        fcntl                           = 1ull << 34,
+        stat                            = 1 << 30,
+        fstat                           = 1ull << 31,
+        open                            = 1ull << 32,
+        fopen                           = 1ull << 33,
+        openat                          = 1ull << 34,
+        fcntl                           = 1ull << 35,
 
         files                           = stat | fstat | open | fopen | openat,
 
         //==============================================================================
         // system
         //==============================================================================
-        schedule                        = 1ull << 35,  // linux only
-        context_switch                  = 1ull << 36,  // linux only
-        syscall                         = 1ull << 37,
+        schedule                        = 1ull << 36,  // linux only
+        context_switch                  = 1ull << 37,  // linux only
+        syscall                         = 1ull << 38,
 
         sys                             = schedule | context_switch | syscall
     };

tests/fail_large_atomic.cpp

@@ -1,4 +1,6 @@
 #include <atomic>
+#include <cassert>
+#include <array>
 #include <rtcheck.h>
 
 struct doubles
@@ -9,7 +11,9 @@ struct doubles
 int main()
 {
     std::atomic<doubles> a;
-    doubles d;
+    assert (! a.is_lock_free());
+
+    doubles d { 1.0, 2.0, 3.0, 4.0 };
 
     rtc::realtime_context rc;
     a.store (d);