X-Git-Url: http://git.liburcu.org/?a=blobdiff_plain;f=liblttng-ust%2Flttng-ust-comm.c;h=86cce18d57fe658588db907ed8016f39e95f975f;hb=efe0de097e8e1cd372a7eb926e0ec68b97e6ee50;hp=d0641c5c6e510b01d0538d9c36b07c5c430644c6;hpb=48621a4272bdeb1e8fced511ca6bf3c4c2240c15;p=lttng-ust.git

diff --git a/liblttng-ust/lttng-ust-comm.c b/liblttng-ust/lttng-ust-comm.c
index d0641c5c..86cce18d 100644
--- a/liblttng-ust/lttng-ust-comm.c
+++ b/liblttng-ust/lttng-ust-comm.c
@@ -383,7 +383,7 @@ error:
 }
 
 static
-void cleanup_sock_info(struct sock_info *sock_info)
+void cleanup_sock_info(struct sock_info *sock_info, int exiting)
 {
 	int ret;
 
@@ -402,7 +402,13 @@ void cleanup_sock_info(struct sock_info *sock_info)
 		sock_info->root_handle = -1;
 	}
 	sock_info->constructor_sem_posted = 0;
-	if (sock_info->wait_shm_mmap) {
+	/*
+	 * wait_shm_mmap is used by listener threads outside of the
+	 * ust lock, so we cannot tear it down ourselves, because we
+	 * cannot join on these threads. Leave this task to the OS
+	 * process exit.
+	 */
+	if (!exiting && sock_info->wait_shm_mmap) {
 		ret = munmap(sock_info->wait_shm_mmap, sysconf(_SC_PAGE_SIZE));
 		if (ret) {
 			ERR("Error unmapping wait shm");
@@ -876,10 +882,17 @@ void __attribute__((constructor)) lttng_ust_init(void)
 static
 void lttng_ust_cleanup(int exiting)
 {
-	cleanup_sock_info(&global_apps);
+	cleanup_sock_info(&global_apps, exiting);
 	if (local_apps.allowed) {
-		cleanup_sock_info(&local_apps);
+		cleanup_sock_info(&local_apps, exiting);
 	}
+	/*
+	 * The teardown in this function all affect data structures
+	 * accessed under the UST lock by the listener thread. This
+	 * lock, along with the lttng_ust_comm_should_quit flag, ensure
+	 * that none of these threads are accessing this data at this
+	 * point.
+	 */
 	lttng_ust_abi_exit();
 	lttng_ust_events_exit();
 	ltt_ring_buffer_client_discard_exit();
@@ -913,6 +926,7 @@ void __attribute__((destructor)) lttng_ust_exit(void)
 	lttng_ust_comm_should_quit = 1;
 	ust_unlock();
 
+	/* cancel threads */
 	ret = pthread_cancel(global_apps.ust_listener);
 	if (ret) {
 		ERR("Error cancelling global ust listener thread");
@@ -923,6 +937,14 @@ void __attribute__((destructor)) lttng_ust_exit(void)
 			ERR("Error cancelling local ust listener thread");
 		}
 	}
+	/*
+	 * Do NOT join threads: use of sys_futex makes it impossible to
+	 * join the threads without using async-cancel, but async-cancel
+	 * is delivered by a signal, which could hit the target thread
+	 * anywhere in its code path, including while the ust_lock() is
+	 * held, causing a deadlock for the other thread. Let the OS
+	 * cleanup the threads if there are stalled in a syscall.
+	 */
 	lttng_ust_cleanup(1);
 }