🎯 Sync smart & scheduler codes #8537

polarvid · 2024-02-18T02:46:17Z

拉取/合并请求描述：(PR description)

[

为什么提交这份PR (why to submit this PR)

sync rt-smart sources
fixup of issue SMP下中断与互斥锁同时使用存在死锁的场景 #8493

你的解决方案是什么 (what is your solution)

请提供验证的bsp和config (provide the config and bsp)

BSP:

.config:

action:

]

当前拉取/合并请求的状态 Intent for your PR

必须选择一项 Choose one (Mandatory):

本拉取/合并请求是一个草稿版本 This PR is for a code-review and is intended to get feedback
本拉取/合并请求是一个成熟版本 This PR is mature, and ready to be integrated into the repo

代码质量 Code Quality：

我在这个拉取/合并请求中已经考虑了 As part of this pull request, I've considered the following:

已经仔细查看过代码改动的对比 Already check the difference between PR and old code
代码风格正确，包括缩进空格，命名及其他风格 Style guide is adhered to, including spacing, naming and other styles
没有垃圾代码，代码尽量精简，不包含#if 0代码，不包含已经被注释了的代码 All redundant code is removed and cleaned up
所有变更均有原因及合理的，并且不会影响到其他软件组件代码或BSP All modifications are justified and not affect other components or BSP
对难懂代码均提供对应的注释 I've commented appropriately where code is tricky
代码是高质量的 Code in this PR is of high quality
已经使用formatting 等源码格式化工具确保格式符合RT-Thread代码规范 This PR complies with RT-Thread code specification

include/rtdef.h

Guozhanxin · 2024-02-19T02:23:16Z

改动这么大，是不是有些测试结果给出来，是不是确实优化了，性能提高了多少😂

components/drivers/ipc/completion.c

src/ipc.c

src/kservice.c

components/drivers/ipc/dataqueue.c

include/rtdef.h

mysterywolf

我阻塞一下合入按钮，这个准备好了麻烦让我review一下

components/drivers/ipc/SConscript

polarvid · 2024-02-20T13:35:30Z

新增 UTEST 集

测试项目	配置	状态 (patch 后)	状态 (patch 前)
scheduler.mtx	qemu aa64, RT_USING_SMP, RT_CPUS_NR=4	PASSED	内核崩溃
scheduler.mtx	qemu a9, RT_USING_SMP, RT_CPUS_NR=2	PASSED	内核死锁
scheduler.mtx	qemu a9, RT_USING_SMP, RT_CPUS_NR=1	PASSED
scheduler.mtx	qemu rv64, RT_CPUS_NR=1	PASSED
scheduler.mtx	rt-spark	PASSED
scheduler.mtx	rpi4b 64	PASSED
scheduler.sem	qemu aa64, RT_USING_SMP, RT_CPUS_NR=4	PASSED
scheduler.sem	qemu a9, RT_USING_SMP, RT_CPUS_NR=2	PASSED
scheduler.sem	qemu a9, RT_USING_SMP, RT_CPUS_NR=1	PASSED
scheduler.sem	qemu a9, RT_CPUS_NR=1	PASSED
scheduler.sem	qemu rv64, RT_CPUS_NR=1	PASSED
scheduler.sem	rt-spark	PASSED
scheduler.sem	rpi4b 64	PASSED
scheduler.timed_mtx	qemu aa64, RT_USING_SMP, RT_CPUS_NR=4	PASSED	内核崩溃
scheduler.timed_mtx	qemu a9, RT_USING_SMP, RT_CPUS_NR=2	PASSED	内核死锁
scheduler.timed_mtx	qemu a9, RT_USING_SMP, RT_CPUS_NR=1	PASSED
scheduler.timed_mtx	qemu rv64, RT_CPUS_NR=1	PASSED
scheduler.timed_mtx	rt-spark	PASSED
scheduler.timed_mtx	rpi4b 64	PASSED
scheduler.timed_sem	qemu aa64, RT_USING_SMP, RT_CPUS_NR=4	PASSED
scheduler.timed_sem	qemu a9, RT_USING_SMP, RT_CPUS_NR=2	PASSED
scheduler.timed_sem	qemu a9, RT_USING_SMP, RT_CPUS_NR=1	PASSED
scheduler.timed_sem	qemu rv64, RT_CPUS_NR=1	PASSED
scheduler.timed_sem	rt-spark	PASSED
scheduler.timed_sem	rpi4b 64	PASSED
scheduler.thread	qemu aa64, RT_USING_SMP, RT_CPUS_NR=4	PASSED
scheduler.thread	qemu a9, RT_USING_SMP, RT_CPUS_NR=2	PASSED
scheduler.thread	qemu a9, RT_USING_SMP, RT_CPUS_NR=1	PASSED
scheduler.thread	qemu rv64, RT_CPUS_NR=1	PASSED
scheduler.thread	rt-spark	PASSED
scheduler.thread	rpi4b 64	PASSED

mysterywolf · 2024-02-21T05:18:06Z

麻烦基于最新基线重新rebase一下并通过CI

components/drivers/ipc/completion.c

components/drivers/ipc/dataqueue.c

components/vbus/prio_queue.c

src/kservice.c

src/scheduler_comm.c

components/libc/compilers/common/include/sys/time.h

src/cpu.c

Add Support for ~FUTEX_PRIVATE, and futex requeue and futex pi feature. Co-authored-by: xqyjlj <[email protected]> Signed-off-by: Shell <[email protected]>

In the previous implementation, each call will create a new timer object for the process. But according to the manual, the itimer is a singleton object per-process. This un-compatible semantic will corrupt the API by a call to cancel previous setup timer. Signed-off-by: Shell <[email protected]>

This patch adds an abstraction layer of scheduler(rt_sched). Following features are added: - The scheduler class (rt_sched) is added for rtthread source and ipc primitives. Critical region of scheduler is also redesigned to simplify the synchronization and acquire better performance. - Support nested lock/unlock detection for scheduler - Suspended thread and completed status tracking of rt_completion is modified for smaller footprint and easier locking. - Add suspend list - A basic building block for IPC primitives which interacts with scheduler directly. Its API is similar to a FIFO list. This new object simplify the synchronization between IPC primitives and asynchronization event like timeout, signals. And it fixed the data racing commonly found on IPC objects currently. - add RT_TIMER_FLAG_THREAD_TIMER as new type of rt_timer for rt_thread, so the timer is notified to do the necessary co-op with scheduler, as the legacy semantics required. Besides, following bugs are fixed: - _scheduler_get_highest_priority_thread() may access an empty list, and cause the codes to corrupt. - rt_thread_control(RT_THREAD_CTRL_CHANGE_PRIORITY) has data racing with scheduler codes, by the leaving of the critical region for a while which corrupts the data coherency of ready queue. - data racing on rt_completion. - in rt_timer_check(): rt_hw_cpu_id() should be called with IRQ masked. - fixup of mutex_delete(). While mutex owner did not release the mutex but delete it after taken, the prio inheritance algorithm may cause the owner thread to lose its original priority. This patch fix this issue by adding a thread priority restore on rt_mutex_delete()/ rt_mutex_detach() API. - For qemu-vexpress-a9, fixup boot-up failure while RT_CPUS_NR is config as 1. Signed-off-by: Shell <[email protected]>

Signed-off-by: Shell <[email protected]>

mysterywolf

我关心的问题已经都解决。

geniusgogo · 2024-02-23T01:48:08Z

bsp/qemu-vexpress-a9/applications/main.c


 int main(void)
 {
-    printf("Hello RT-Thread!\n");
+    rt_kprintf("Hello RT-Thread!\n");


现在内核里用printf和rt_kprintf的讲究是啥？

尽量用rt_kprintf吧这个只是编程习惯~ printf里边乱七八糟一堆东西不可控

musl 工具链会串到 stdio 那个地方，然后跑飞了。总之内核态还是不要用用户态的 print 为好，两边实现思路都不一样。

如果是我我也不用printf，这玩意完全依赖第三方libc。

后面mlibc需要提上日程了，smart最终还得依赖于mlibc的

现在还有个头文件的问题。仓库里面不同的 BSP 用的工具链自带的头文件，因为跨架构，有些甚至是直接拿的 x86 Linux ，存在很多不兼容的定义。比如 time.h 里面的 struct timeval。

geniusgogo · 2024-02-23T01:53:53Z

components/dfs/dfs_v2/src/dfs_pcache.c

-    RT_ASSERT(_dfs_page_insert(aspace, page) == 0);
+    if (_dfs_page_insert(aspace, page))
+    {
+        RT_ASSERT(0);


不如将RT_ASSERT在不开debug的选项的时候让他求表达式，例如:

#define RT_ASSERT(EX) \ do { \ rt_base_t val; \ val = (exp); \ } while (0)

因为ASSERT是对值进行断言，如果调用者使用了复合表达式，那应该要执行他的符合表达式。

这个变量不使用的话，编译器还是可以优化掉的吧

不过感觉这样改会有兼容性问题，还是先别动了，可以做为一个讨论。

还是建议不要在ASSERT内写太复杂的计算，而更多是判断。所以，最好也用不ASSERT的方式编译过代码，运行过代码。

未来也可以加入，RT_ASSERT_RET(condition, ret)的方式来处理condition不满足时应该返回，跳出函数的处理。

RT_ASSERT_RET(condition, ret) 好建议~ 我创建了一个issue 后续可以处理 #8553

components/lwp/lwp_internal.h

include/rtsched.h

src/SConscript

components/drivers/ipc/completion.c

…ad-mirror into shell/sched-dev

polarvid · 2024-02-23T09:24:08Z

改动这么大，是不是有些测试结果给出来，是不是确实优化了，性能提高了多少😂

这个 PR 主要是功能修复和 smart 的 IPC 研发分支同步。功能在 qemu aa64, a9, rv64 以及星火一号，rpi4b 64 平台上面测试过所有的内核 UTEST。新加入的 UTEST 结果参考 #8537 (comment) 。

性能方面，在 arm64 多核平台下测试，PR 前后 cyclictest 的 MIN/AVG 指标分别有 42.31%/41.94% 的优化，MAX 降低了 50.77%。iperf 带宽抖动明显减小，多核性能和单核性能基本一致。

polarvid force-pushed the shell/sched-dev branch 5 times, most recently from 67fdcbb to 00bb5b9 Compare February 18, 2024 10:28

Guozhanxin reviewed Feb 19, 2024

View reviewed changes

include/rtdef.h Outdated Show resolved Hide resolved

Guozhanxin reviewed Feb 19, 2024

View reviewed changes

include/rtdef.h Outdated Show resolved Hide resolved

BernardXiong changed the title ~~Sync smart & scheduler codes~~ 🎯 Sync smart & scheduler codes Feb 19, 2024

geniusgogo reviewed Feb 19, 2024

View reviewed changes

components/drivers/ipc/completion.c Outdated Show resolved Hide resolved

polarvid force-pushed the shell/sched-dev branch 3 times, most recently from bbac067 to 8519468 Compare February 19, 2024 09:18

geniusgogo reviewed Feb 19, 2024

View reviewed changes

src/ipc.c Outdated Show resolved Hide resolved

src/kservice.c Show resolved Hide resolved

components/drivers/ipc/dataqueue.c Outdated Show resolved Hide resolved

include/rtdef.h Show resolved Hide resolved

mysterywolf linked an issue Feb 20, 2024 that may be closed by this pull request

SMP下中断与互斥锁同时使用存在死锁的场景 #8493

Closed

mysterywolf requested changes Feb 20, 2024

View reviewed changes

mysterywolf reviewed Feb 20, 2024

View reviewed changes

components/drivers/ipc/SConscript Outdated Show resolved Hide resolved

mysterywolf mentioned this pull request Feb 20, 2024

🎯 [libcpu/arm/cortex-a]修改arm/cortex-a启动方式，解决cortex-a开启SMP工作异常问题 #8517

Merged

9 tasks

polarvid force-pushed the shell/sched-dev branch from b616303 to 83ab91c Compare February 20, 2024 13:34