fix: add acquire fence in bthread_join for ARM memory visibility

[guoliushui]

Ensure memory visibility on ARM architecture after thread join.

What problem does this PR solve?

Issue Number: fix #3274

Problem Summary:

What is changed and the side effects?

Changed:

Side effects:

• Performance effects:

• Breaking backward compatibility:

---

Check List:

• Please make sure your changes are compilable.
• When providing us with a new feature, it is best to add related tests.
• Please follow Contributor Covenant Code of Conduct.

[guoliushui]

已使用测试程序验证生效

[Copilot]

Pull request overview

This PR addresses weak-memory-ordering visibility issues where bthread_join() may return before the joining thread can reliably observe the joined bthread’s prior memory writes (reported on ARM/aarch64).

Changes:

• Add an acquire fence after TaskGroup::join() observes the joined bthread has advanced version_butex.
• Gate the fence with ARM architecture preprocessor checks (__aarch64__ / __arm__).

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[altman08]

我们最近也在用arm，想验证一下这个bug，可以提供一下复现bug的代码吗

[chenBright]

butex_wait 有 acquire 语义。是不是只有没有进入循环的情况下才有问题？

brpc/src/bthread/task_group.cpp

Lines 632 to 637 in 35682ff

	while (*m->version_butex == expected_version) {
	if (butex_wait(m->version_butex, expected_version, NULL) < 0 &&
	errno != EWOULDBLOCK && errno != EINTR) {
	return errno;
	}
	}

[guoliushui]

butex_wait 有 acquire 语义。是不是只有没有进入循环的情况下才有问题？

brpc/src/bthread/task_group.cpp

Lines 632 to 637 in 35682ff

	while (*m->version_butex == expected_version) {
	if (butex_wait(m->version_butex, expected_version, NULL) < 0 &&
	errno != EWOULDBLOCK && errno != EINTR) {
	return errno;
	}
	}

当时也有过这方面的考虑，最后未做区分是否touch到了butex_wait原因如下：

1. 即使之 butex_wait 内部有 acquire，从那个 acquire 到 while 退出的普通读之间，invalidate queue 可能又积累了新条目。只有在 while 退出后、读用户数据前加 fence，才能保证用户数据可见
2. 区分逻辑增加复杂度，收益可忽略：多了变量和条件判断，只省下偶尔一条 DMB ISHLD（10~几十ns），在阻塞等待函数中意义不大

[chenBright]

LGTM

[condy0919]

@guoliushui 之前我在鲲鹏 950 上测试时也发现，bthread_join(..) 子任务时会卡住，导致客户端无法退出

[guoliushui]

@guoliushui 之前我在鲲鹏 950 上测试时也发现，bthread_join(..) 子任务时会卡住，导致客户端无法退出

目前在鲲鹏920上，还未遇到过卡住的现象

[guoliushui]

我们最近也在用arm，想验证一下这个bug，可以提供一下复现bug的代码吗

可以看下这里 https://www.cnblogs.com/guoliushui/p/19902091

[altman08]

我们最近也在用arm，想验证一下这个bug，可以提供一下复现bug的代码吗

可以看下这里 https://www.cnblogs.com/guoliushui/p/19902091

感谢，我们机器也是鲲鹏920的，复现了这个问题