Address review comment

And small tweaks to conform to standard API.

Author: kpamnany

base/timing.jl

@@ -71,7 +71,7 @@ end
 # Retrieve the specified thread's, or the aggregated value across all threads
 # of wait time on all the locks for which wait-time measurement has been
 # enabled (for RAI, currently only the codegen lock).
-lock_waiting_time_ns(tid::Int16) = ccall(:jl_get_thread_lock_waiting_time, UInt64, (Int16,), tid)
+lock_waiting_time_ns(tid::Integer) = ccall(:jl_get_thread_lock_waiting_time, UInt64, (Cint,), tid-1)
 lock_waiting_time_ns() = ccall(:jl_get_lock_waiting_time, UInt64, ())
 
 # cumulative total time spent on compilation and recompilation, in nanoseconds

src/threading.c

@@ -932,9 +932,9 @@ void _jl_mutex_unlock(jl_task_t *self, jl_mutex_t *lock)
     }
 }
 
-JL_DLLEXPORT uint64_t jl_get_thread_lock_waiting_time(int64_t tid)
+JL_DLLEXPORT uint64_t jl_get_thread_lock_waiting_time(int16_t tid)
 {
-    int nthreads = jl_atomic_load_acquire(&jl_n_threads);
+    int nthreads = jl_atomic_load_relaxed(&jl_n_threads);
     jl_ptls_t *all_tls_states = jl_atomic_load_relaxed(&jl_all_tls_states);
     if (tid < nthreads) {
         jl_ptls_t ptls = all_tls_states[tid];
@@ -948,8 +948,8 @@ JL_DLLEXPORT uint64_t jl_get_thread_lock_waiting_time(int64_t tid)
 JL_DLLEXPORT uint64_t jl_get_lock_waiting_time(void)
 {
     uint64_t waiting_time = 0;
-    int nthreads = jl_atomic_load_acquire(&jl_n_threads);
-    int ngcthreads = jl_atomic_load_acquire(&jl_n_gcthreads);
+    int nthreads = jl_atomic_load_relaxed(&jl_n_threads);
+    int ngcthreads = jl_atomic_load_relaxed(&jl_n_gcthreads);
     int nmutatorthreads = nthreads - ngcthreads;
     jl_ptls_t *all_tls_states = jl_atomic_load_relaxed(&jl_all_tls_states);
     for (int i = 0; i < nmutatorthreads; i++) {