Move per-thread RCU setup into isc_thread

All the per-loop `libuv` setup remains in `isc_loop`, but the per-thread
RCU setup is moved to `isc_thread` alongside the other per-thread setup.
This avoids repeating the per-thread setup for `call_rcu()` helpers,
and explains a little better why some parts of the per-thread setup
is missing for `call_rcu()` helpers.

This also removes the per-loop `call_rcu()` helpers as we refactored the
isc__random_initialize() in the previous commit.

lib/isc/include/isc/thread.h

@@ -35,6 +35,13 @@ ISC_LANG_BEGINDECLS
 typedef pthread_t isc_thread_t;
 typedef void *(*isc_threadfunc_t)(void *);
 
+/*%
+ * like isc_thread_create(), but run the function on the current
+ * thread which must be the main thread.
+ */
+void
+isc_thread_main(isc_threadfunc_t, void *);
+
 void
 isc_thread_create(isc_threadfunc_t, void *, isc_thread_t *);
 

lib/isc/include/isc/util.h

@@ -247,9 +247,19 @@
 #endif /* if __has_feature(thread_sanitizer) */
 
 #if __SANITIZE_THREAD__
+/*
+ * We should rather be including <sanitizer/tsan_interface.h>, but GCC 10
+ * header is broken, so we just make the declarations by hand.
+ */
+void
+__tsan_acquire(void *addr);
+void
+__tsan_release(void *addr);
 #define ISC_NO_SANITIZE_THREAD __attribute__((no_sanitize("thread")))
 #else /* if __SANITIZE_THREAD__ */
 #define ISC_NO_SANITIZE_THREAD
+#define __tsan_acquire(addr)
+#define __tsan_release(addr)
 #endif /* if __SANITIZE_THREAD__ */
 
 #if __GNUC__ > 4 || (__GNUC__ == 4 && __GNUC_MINOR >= 6)

lib/isc/loop.c

@@ -103,6 +103,7 @@ resume_loop(isc_loop_t *loop) {
 
 	(void)isc_barrier_wait(&loopmgr->resuming);
 	loop->paused = false;
+
 	rcu_thread_online();
 }
 
@@ -268,54 +269,17 @@ static void
 quiescent_cb(uv_prepare_t *handle) {
 	isc__qsbr_quiescent_cb(handle);
 
-#ifndef RCU_QSBR
-	INSIST(!rcu_read_ongoing());
-#else
+#if defined(RCU_QSBR)
 	/* safe memory reclamation */
 	rcu_quiescent_state();
 
 	/* mark the thread offline when polling */
 	rcu_thread_offline();
+#else
+	INSIST(!rcu_read_ongoing());
 #endif
 }
 
-static void
-loop_call_rcu_init(struct rcu_head *rcu_head __attribute__((__unused__))) {
-	/* Work around the jemalloc bug, see trampoline.c for details */
-	void *ptr = malloc(8);
-	free(ptr);
-
-	/* Initialize the random generator in the call_rcu thread */
-	isc__random_initialize();
-}
-
-static void
-loop_run(isc_loop_t *loop) {
-	int r = uv_prepare_start(&loop->quiescent, quiescent_cb);
-	UV_RUNTIME_CHECK(uv_prepare_start, r);
-
-	isc_barrier_wait(&loop->loopmgr->starting);
-
-	isc_async_run(loop, setup_jobs_cb, loop);
-
-	rcu_register_thread();
-
-	struct call_rcu_data *crdp = create_call_rcu_data(0, -1);
-	set_thread_call_rcu_data(crdp);
-
-	call_rcu(&loop->rcu_head, loop_call_rcu_init);
-
-	r = uv_run(&loop->loop, UV_RUN_DEFAULT);
-	UV_RUNTIME_CHECK(uv_run, r);
-
-	rcu_unregister_thread();
-
-	set_thread_call_rcu_data(NULL);
-	call_rcu_data_free(crdp);
-
-	isc_barrier_wait(&loop->loopmgr->stopping);
-}
-
 static void
 loop_close(isc_loop_t *loop) {
 	int r = uv_loop_close(&loop->loop);
@@ -337,7 +301,17 @@ loop_thread(void *arg) {
 
 	isc__tid_init(loop->tid);
 
-	loop_run(loop);
+	int r = uv_prepare_start(&loop->quiescent, quiescent_cb);
+	UV_RUNTIME_CHECK(uv_prepare_start, r);
+
+	isc_barrier_wait(&loop->loopmgr->starting);
+
+	isc_async_run(loop, setup_jobs_cb, loop);
+
+	r = uv_run(&loop->loop, UV_RUN_DEFAULT);
+	UV_RUNTIME_CHECK(uv_run, r);
+
+	isc_barrier_wait(&loop->loopmgr->stopping);
 
 	return (NULL);
 }
@@ -511,7 +485,7 @@ isc_loopmgr_run(isc_loopmgr_t *loopmgr) {
 		isc_thread_setname(loop->thread, name);
 	}
 
-	loop_thread(&loopmgr->loops[0]);
+	isc_thread_main(loop_thread, &loopmgr->loops[0]);
 }
 
 void

lib/isc/loop_p.h

@@ -28,7 +28,6 @@
 #include <isc/stack.h>
 #include <isc/thread.h>
 #include <isc/types.h>
-#include <isc/urcu.h>
 #include <isc/uv.h>
 #include <isc/work.h>
 
@@ -80,8 +79,6 @@ struct isc_loop {
 	uv_async_t wakeup_trigger;
 	uv_prepare_t quiescent;
 	isc_qsbr_phase_t qsbr_phase;
-
-	struct rcu_head rcu_head;
 };
 
 /*

lib/isc/thread.c

@@ -30,9 +30,12 @@
 
 #include <stdlib.h>
 
+#include <isc/atomic.h>
 #include <isc/iterated_hash.h>
 #include <isc/strerr.h>
 #include <isc/thread.h>
+#include <isc/tid.h>
+#include <isc/urcu.h>
 #include <isc/util.h>
 
 #ifndef THREAD_MINSTACKSIZE
@@ -46,18 +49,15 @@
  */
 
 struct thread_wrap {
+	struct rcu_head rcu_head;
 	isc_threadfunc_t func;
 	void *arg;
-	void (*free)(void *);
 };
 
 static struct thread_wrap *
 thread_wrap(isc_threadfunc_t func, void *arg) {
 	struct thread_wrap *wrap = malloc(sizeof(*wrap));
-
-	RUNTIME_CHECK(wrap != NULL);
 	*wrap = (struct thread_wrap){
-		.free = free, /* from stdlib */
 		.func = func,
 		.arg = arg,
 	};
@@ -66,30 +66,56 @@ thread_wrap(isc_threadfunc_t func, void *arg) {
 }
 
 static void *
-thread_run(void *arg) {
-	struct thread_wrap *wrap = arg;
-	isc_threadfunc_t wrap_func = wrap->func;
-	void *wrap_arg = wrap->arg;
-	void *result = NULL;
+thread_body(struct thread_wrap *wrap) {
+	isc_threadfunc_t func = wrap->func;
+	void *arg = wrap->arg;
+	void *ret = NULL;
 
 	/*
 	 * Every thread starts with a malloc() call to prevent memory bloat
-	 * caused by a jemalloc quirk. To stop an optimizing compiler from
-	 * stripping out free(malloc(1)), we call free via a function pointer.
+	 * caused by a jemalloc quirk.  We use CMM_ACCESS_ONCE() To stop an
+	 * optimizing compiler from stripping out free(malloc(1)).
 	 */
-	wrap->free(malloc(1));
-	wrap->free(wrap);
+	void *jemalloc_enforce_init = NULL;
+	CMM_ACCESS_ONCE(jemalloc_enforce_init) = malloc(1);
+	free(jemalloc_enforce_init);
 
-	/* Get a thread-local digest context. */
+	/* Reassure Thread Sanitizer that it is safe to free the wrapper */
+	__tsan_acquire(wrap);
+	free(wrap);
+
+	rcu_register_thread();
+
+	ret = func(arg);
+
+	rcu_unregister_thread();
+
+	return (ret);
+}
+
+static void *
+thread_run(void *wrap) {
+	/*
+	 * Get a thread-local digest context only in new threads.
+	 * The main thread is handled by isc__initialize().
+	 */
 	isc__iterated_hash_initialize();
 
-	/* Run the main function */
-	result = wrap_func(wrap_arg);
+	void *ret = thread_body(wrap);
 
-	/* Cleanup */
 	isc__iterated_hash_shutdown();
 
-	return (result);
+	return (ret);
+}
+
+void
+isc_thread_main(isc_threadfunc_t func, void *arg) {
+	/*
+	 * Either this thread has not yet been started, so it can become the
+	 * main thread, or it has already been annointed as the chosen zero
+	 */
+	REQUIRE(isc_tid() == ISC_TID_UNKNOWN || isc_tid() == 0);
+	thread_body(thread_wrap(func, arg));
 }
 
 void