MEDIUM: mux-fcgi: use task_kill() during fcgi_takeover() instead of task_wakeup()
task_wakeup() passes the task through the global run queue under the
global RQ lock, which is expensive when dealing with large amounts of
fcgi_takeover() calls. Let's use the new task_kill() instead to kill the
task.
diff --git a/src/mux_fcgi.c b/src/mux_fcgi.c
index b09ad56..bd27a63 100644
--- a/src/mux_fcgi.c
+++ b/src/mux_fcgi.c
@@ -4082,6 +4082,7 @@
static int fcgi_takeover(struct connection *conn)
{
struct fcgi_conn *fcgi = conn->ctx;
+ struct task *task;
if (fd_takeover(conn->handle.fd, conn) != 0)
return -1;
@@ -4093,10 +4094,13 @@
*/
fcgi->wait_event.tasklet->context = NULL;
tasklet_wakeup(fcgi->wait_event.tasklet);
- if (fcgi->task) {
- fcgi->task->context = NULL;
- /* Wake the task, to let it free itself */
- task_wakeup(fcgi->task, TASK_WOKEN_OTHER);
+
+ task = fcgi->task;
+ if (task) {
+ task->context = NULL;
+ fcgi->task = NULL;
+ __ha_barrier_store();
+ task_kill(task);
fcgi->task = task_new(tid_bit);
if (!fcgi->task) {