Address review comments

Author: kpamnany

base/timing.jl

@@ -68,8 +68,11 @@ function gc_alloc_count(diff::GC_Diff)
     diff.malloc + diff.realloc + diff.poolalloc + diff.bigalloc
 end
 
-lock_waiting_time(tid::Int16) = ccall(:jl_get_thread_lock_waiting_time, UInt64, (Int16,), tid)
-lock_waiting_time() = ccall(:jl_get_lock_waiting_time, UInt64, ())
+# 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() = ccall(:jl_get_lock_waiting_time, UInt64, ())
 
 # cumulative total time spent on compilation and recompilation, in nanoseconds
 function cumulative_compile_time_ns()

src/threading.c

@@ -949,8 +949,10 @@ 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_gc_threads);
+    int nmutatorthreads = nthreads - ngcthreads;
     jl_ptls_t *all_tls_states = jl_atomic_load_relaxed(&jl_all_tls_states);
-    for (int i = 0; i < nthreads; i++) {
+    for (int i = 0; i < nmutatorthreads; i++) {
         jl_ptls_t ptls = all_tls_states[i];
         if (ptls) {
             waiting_time += jl_atomic_load_relaxed(&ptls->lock_waiting_time);