aarch64架构下运行unittest报错

[leaf-potato]

Describe the bug (描述bug)
在aarch64架构下运行brpc_grpc_protocol_unittest报错bthread=4294968320 sched_to itself!并异常退出，具体日志信息如下：

[==========] Running 7 tests from 1 test case.
[----------] Global test environment set-up.
[----------] 7 tests from GrpcTest
[ RUN      ] GrpcTest.percent_encode
I0716 16:53:28.192583 1732065 /home/leaf-potato/Open-Source/brpc/src/brpc/server.cpp:1127] Server[(anonymous namespace)::MyGrpcService] is serving on port=8011.
I0716 16:53:28.337017 1732065 /home/leaf-potato/Open-Source/brpc/src/brpc/server.cpp:1130] Check out http://dev-ubuntu:8011 in web browser.
I0716 16:53:28.337832 1732065 /home/leaf-potato/Open-Source/brpc/src/brpc/server.cpp:1187] Server[(anonymous namespace)::MyGrpcService] is going to quit
F0716 16:53:28.338519 1732076 /home/leaf-potato/Open-Source/brpc/src/bthread/task_group.cpp:623] bthread=4294968320 sched_to itself!
[       OK ] GrpcTest.percent_encode (177 ms)
[ RUN      ] GrpcTest.percent_decode
[2]    1732065 abort      ./brpc_grpc_protocol_unittest 

To Reproduce (复现方法)

1. 拉取master版本最新代码&安装依赖
2. mkdir build && cd build && cmake -DBUILD_UNIT_TESTS=ON .. && make

Expected behavior (期望行为)
正常运行单元测试

Versions (各种版本)
OS: Ubuntu 22.04 | aarch64
Compiler: g++ 11.3.0
brpc: master | 2729272
protobuf: 3.12.4

Additional context/screenshots (更多上下文/截图)

[leaf-potato]

其他UT也有类似问题 ，brpc_ssl_unittest`运行错误：

[==========] Running 5 tests from 1 test case.
[----------] Global test environment set-up.
[----------] 5 tests from SSLTest
[ RUN      ] SSLTest.sanity
W0716 17:00:42.435997 1733130 /home/leaf-potato/Open-Source/brpc/src/brpc/server.cpp:1968] Duplicate certificate hostname=cert1
W0716 17:00:42.436084 1733130 /home/leaf-potato/Open-Source/brpc/src/brpc/server.cpp:1968] Duplicate certificate hostname=cert1
I0716 17:00:42.442796 1733130 /home/leaf-potato/Open-Source/brpc/src/brpc/server.cpp:1127] Server[EchoServiceImpl] is serving on port=8613.
I0716 17:00:42.599956 1733130 /home/leaf-potato/Open-Source/brpc/src/brpc/server.cpp:1130] Check out http://dev-ubuntu:8613 in web browser.
F0716 17:00:42.608247 1733141 /home/leaf-potato/Open-Source/brpc/src/bthread/task_group.cpp:574] Recursively(1) call sched_to(0xaaaad7326d80)
I0716 17:00:42.608291 1733136 /home/leaf-potato/Open-Source/brpc/src/bthread/fd.cpp:330] Use DEL+ADD instead of EPOLLONESHOT+MOD due to kernel bug. Performance will be much lower.
F0716 17:00:42.609952 1733141 /home/leaf-potato/Open-Source/brpc/src/bthread/task_group.cpp:574] Recursively(1) call sched_to(0xaaaad7326d80)
F0716 17:00:42.614055 1733141 /home/leaf-potato/Open-Source/brpc/src/bthread/task_group.cpp:574] Recursively(1) call sched_to(0xaaaad7326d80)
F0716 17:00:42.615524 1733137 /home/leaf-potato/Open-Source/brpc/src/bthread/task_group.cpp:574] Recursively(1) call sched_to(0xaaaad7326780)
F0716 17:00:42.615918 1733140 /home/leaf-potato/Open-Source/brpc/src/bthread/task_group.cpp:574] Recursively(1) call sched_to(0xaaaad7326c00)
F0716 17:00:42.617769 1733135 /home/leaf-potato/Open-Source/brpc/src/bthread/task_group.cpp:574] Recursively(1) call sched_to(0xaaaad7326480)
F0716 17:00:42.618183 1733138 /home/leaf-potato/Open-Source/brpc/src/bthread/task_group.cpp:574] Recursively(1) call sched_to(0xaaaad7326900)
F0716 17:00:42.618773 1733139 /home/leaf-potato/Open-Source/brpc/src/bthread/task_group.cpp:574] Recursively(1) call sched_to(0xaaaad7326a80)
F0716 17:00:42.619078 1733139 /home/leaf-potato/Open-Source/brpc/src/bthread/task_group.cpp:574] Recursively(2) call sched_to(0xaaaad7326a80)

相同的版本使用clang编译后运行没有问题，比较奇怪。不知道跟之前这个#845 问题是否类似

[wwbmmm]

clang没有问题可能是因为这里走了不同的分支
https://github.com/apache/brpc/blob/master/src/butil/thread_local.h#L49
试试改下这里让gcc+aarch64也走上面的分支呢

[leaf-potato]

https://github.com/apache/brpc/blob/master/src/butil/thread_local.h#L49
试试改下这里让gcc+aarch64也走上面的分支呢

改了后单测可以过了，这是啥原因呀？方便讲下嘛

[Huixxi]

应该就是编译器的volatile thread local access这个优化在aarch64下有问题，需要关闭了才能正常执行。