From: Willy Tarreau <w@1wt.eu>
Date: Wed, 22 Jun 2022 07:00:08 +0000 (+0200)
Subject: MINOR: thread: move the flags to the shared cache line
X-Git-Tag: v2.7-dev2~137
X-Git-Url: http://git.ipfire.org/?a=commitdiff_plain;h=8e079cdd4432946abd7d22d44b5a1090b35871c4;p=thirdparty%2Fhaproxy.git

MINOR: thread: move the flags to the shared cache line

The thread flags were once believed to be local to the thread, but as
it stands, even the STUCK flag is shared since it's looked at by the
watchdog. As such we'll need to use atomic ops to manipulate them, and
likely to move them into the shared area.

This patch only moves the flag into the shared area so that we can later
decide whether it's best to leave them there or to move them back to the
local area. Interestingly, some tests have shown a 3% better performance
on dequeuing with this, while they're not used by other threads yet, so
there are definitely alignment effects that might change over time.
---

diff --git a/include/haproxy/tinfo-t.h b/include/haproxy/tinfo-t.h
index f9534d1111..81a246c604 100644
--- a/include/haproxy/tinfo-t.h
+++ b/include/haproxy/tinfo-t.h
@@ -36,7 +36,10 @@ enum {
 	TL_CLASSES       /* must be last */
 };
 
-/* thread_ctx flags, for ha_thread_ctx[].flags */
+/* thread_ctx flags, for ha_thread_ctx[].flags. These flags describe the
+ * thread's state and are visible to other threads, so they must be used
+ * with atomic ops.
+ */
 #define TH_FL_STUCK             0x00000001
 #define TH_FL_TASK_PROFILING    0x00000002
 
@@ -82,7 +85,6 @@ struct thread_ctx {
 	struct task *current;               /* current task (not tasklet) */
 	int current_queue;                  /* points to current tasklet list being run, -1 if none */
 	unsigned int nb_tasks;              /* number of tasks allocated on this thread */
-	uint flags;                         /* thread flags, TH_FL_* */
 	uint8_t tl_class_mask;              /* bit mask of non-empty tasklets classes */
 
 	// 7 bytes hole here
@@ -100,6 +102,8 @@ struct thread_ctx {
 	unsigned int rq_total;              /* total size of the run queue, prio_tree + tasklets */
 	int tasks_in_list;                  /* Number of tasks in the per-thread tasklets list */
 	uint idle_pct;                      /* idle to total ratio over last sample (percent) */
+	uint flags;                         /* thread flags, TH_FL_*, atomic! */
+	/* 32-bit hole here */
 
 	uint64_t prev_cpu_time;             /* previous per thread CPU time */
 	uint64_t prev_mono_time;            /* previous system wide monotonic time  */