如何创建一个新的Linux内核调度程序

Question

查看调度程序源代码（2.6.34，kernel/sched.c），我可以看到如何使用“可插拔”调度程序，并且我相信我理解的接口是实现。我不明白的是如何让我的代码内置到内核中。至少，其他网站的指针将不胜感激。

现在，我grepping为SCHED_FIFO，SCHED_RR，并在内核源代码树SCHED_NORMAL，所以我真的正在寻找一种更深入的方式来看待它:-)

编辑：由于一些背景，我对FreeBSD调度程序（以及一般的FreeBSD内核）非常熟悉，所以我不在寻找如何执行进程/线程级调度的指针。我正在寻找一种方法来将我自己的调度程序与普通的Linux调度程序（类似于SCHED_FIFO）一起添加。编辑＃2：下面的BFS指针是一个很好的开始，但它仍然将CFS从内核中剥离出来。 sched.c中现在看起来像：

#ifdef CONFIG_SCHED_BFS 
#include "sched_bfs.c" 
#else 
    // original sched.c 
#endif // CONFIG_SCHED_BFS 

我很想看到一个答案或就如何做到这一点了更好的指针（即保持CFS，至少现在）。

编辑＃3：我已经在下面回答了我自己的问题，因为我想我已经知道了。

Answer 1

我已经想出了我的问题的答案，所以我想我会在这里添加它。下面是将为2.6.34 vanilla内核添加一个新调度器的补丁。现在，我只编译了内核。我完全相信运行此确切补丁的系统将导致其崩溃 - 所以需要您自担风险使用:-)

diff --git a/include/linux/sched.h b/include/linux/sched.h 
index 2b7b81d..a2a2b21 100644 
--- a/include/linux/sched.h 
+++ b/include/linux/sched.h 
@@ -37,6 +37,7 @@ 
#define SCHED_RR  2 
#define SCHED_BATCH  3 
/* SCHED_ISO: reserved but not implemented yet */ 
+#define SCHED_NEW    4 /* Stealing from SCHED_ISO */ 
#define SCHED_IDLE  5 
/* Can be ORed in to make sure the process is reverted back to SCHED_NORMAL on fork */ 
#define SCHED_RESET_ON_FORK  0x40000000 
diff --git a/init/Kconfig b/init/Kconfig 
index eb77e8c..0055d26 100644 
--- a/init/Kconfig 
+++ b/init/Kconfig 
@@ -23,6 +23,11 @@ config CONSTRUCTORS 

menu "General setup" 

+config SCHED_NEW 
+  bool "NEW cpu scheduler" 
+  ---help--- 
+   Brand new scheduler 
+ 
config EXPERIMENTAL 
    bool "Prompt for development and/or incomplete code/drivers" 
    ---help--- 
diff --git a/kernel/sched.c b/kernel/sched.c 
index 3c2a54f..588960d 100644 
--- a/kernel/sched.c 
+++ b/kernel/sched.c 
@@ -1931,6 +1931,7 @@ static void deactivate_task(struct rq *rq, struct task_struct *p, int sleep) 

#include "sched_idletask.c" 
#include "sched_fair.c" 
+#include "sched_new.c" 
#include "sched_rt.c" 
#ifdef CONFIG_SCHED_DEBUG 
# include "sched_debug.c" 
diff --git a/kernel/sched_new.c b/kernel/sched_new.c 
new file mode 100644 
index 0000000..c2e269e 
--- /dev/null 
+++ b/kernel/sched_new.c 
@@ -0,0 +1,140 @@ 
+#ifdef CONFIG_SCHED_NEW 
+ 
+/* 
+ * Starting with a simple, 1 runq per cpu scheduler. Don't care 
+ * about fairness for right now. Just get it up and running to 
+ * verify that we have the interface correct 
+ */ 
+ 
+static void 
+enqueue_task_new(struct rq *rq, struct task_struct *p, int wakeup, bool head) 
+{ 
+} 
+ 
+static void dequeue_task_new(struct rq *rq, struct task_struct *p, int sleep) 
+{ 
+} 
+ 
+static void yield_task_new(struct rq *rq) 
+{ 
+} 
+ 
+static void check_preempt_curr_new(struct rq *rq, struct task_struct *p, int flags) 
+{ 
+} 
+ 
+static struct task_struct *pick_next_task_new(struct rq *rq) 
+{ 
+} 
+ 
+static void put_prev_task_new(struct rq *rq, struct task_struct *p) 
+{ 
+} 
+ 
+#ifdef CONFIG_SMP 
+static int select_task_rq_new(struct task_struct *p, int sd_flag, int flags) 
+{ 
+} 
+static void pre_schedule_new(struct rq *rq, struct task_struct *prev) 
+{ 
+} 
+ 
+static void post_schedule_new(struct rq *rq) 
+{ 
+} 
+ 
+static void task_woken_new(struct rq *rq, struct task_struct *p) 
+{ 
+} 
+ 
+static void task_waking_new(struct rq *this_rq, struct task_struct *task) 
+{ 
+} 
+static void set_cpus_allowed_new(struct task_struct *p, 
+    const struct cpumask *new_mask) 
+{ 
+} 
+/* Assumes rq->lock is held */ 
+static void rq_online_new(struct rq *rq) 
+{ 
+} 
+ 
+/* Assumes rq->lock is held */ 
+static void rq_offline_new(struct rq *rq) 
+{ 
+} 
+#endif /* COMFIG_SMP */ 
+ 
+static void set_curr_task_new(struct rq *rq) 
+{ 
+} 
+ 
+ 
+static void task_tick_new(struct rq *rq, struct task_struct *p, int queued) 
+{ 
+} 
+ 
+static void task_fork_new(struct task_struct *p) 
+{ 
+} 
+static void switched_from_new(struct rq *rq, struct task_struct *p, 
+    int running) 
+{ 
+} 
+static void switched_to_new(struct rq *this_rq, struct task_struct *task, 
+    int running) 
+{ 
+} 
+static void prio_changed_new(struct rq *rq, struct task_struct *p, 
+    int oldprio, int running) 
+{ 
+} 
+static unsigned int get_rr_interval_new(struct rq *rq, struct task_struct *task) 
+{ 
+} 
+ 
+ 
+ 
+static const struct sched_class new_sched_class = { 
+ .next   = &fair_sched_class, 
+ .enqueue_task  = enqueue_task_new, 
+ .dequeue_task  = dequeue_task_new, 
+ .yield_task  = yield_task_new, 
+ 
+ .check_preempt_curr = check_preempt_curr_new, 
+ 
+ .pick_next_task  = pick_next_task_new, 
+ .put_prev_task  = put_prev_task_new, 
+ 
+#ifdef CONFIG_SMP 
+ .select_task_rq  = select_task_rq_new, 
+ 
+ .pre_schedule  = pre_schedule_new, 
+ .post_schedule  = post_schedule_new, 
+ 
+ .task_waking   = task_waking_new, 
+ .task_woken  = task_woken_new, 
+ 
+ .set_cpus_allowed  = set_cpus_allowed_new, 
+ 
+ .rq_online    = rq_online_new, 
+ .rq_offline    = rq_offline_new, 
+#endif 
+ 
+ .set_curr_task   = set_curr_task_new, 
+ .task_tick  = task_tick_new, 
+ .task_fork    = task_fork_new, 
+ 
+ .switched_from   = switched_from_new, 
+ .switched_to  = switched_to_new, 
+ 
+ .prio_changed  = prio_changed_new, 
+ 
+ .get_rr_interval = get_rr_interval_new, 
+#ifdef CONFIG_FAIR_GROUP_SCHED 
+ .moved_group   = NULL 
+#endif 
+}; 
+ 
+#endif /* CONFIG_SCHED_NEW */ 
diff --git a/kernel/sched_rt.c b/kernel/sched_rt.c 
index b5b920a..aaf4beb 100644 
--- a/kernel/sched_rt.c 
+++ b/kernel/sched_rt.c 
@@ -1731,7 +1731,11 @@ static unsigned int get_rr_interval_rt(struct rq *rq, struct task_struct *task) 
} 

static const struct sched_class rt_sched_class = { 
+#ifdef CONFIG_SCHED_NEW 
+ .next   = &new_sched_class, 
+#else 
    .next   = &fair_sched_class, 
+#endif /* CONFIG_SCHED_NEW */ 
    .enqueue_task  = enqueue_task_rt, 
    .dequeue_task  = dequeue_task_rt, 
    .yield_task  = yield_task_rt, 

Answer 2

Con Kolivas Brainfuck Scheduler。 我刚刚通过谷歌搜索发现。这可能是一个例子。

Answer 3

Embedded.com具有通过实现一个简单的实走3个部分进入时间调度：

• part 1
• part 2
• part 3

不像其他的答案，这一个被创建为一个教程：

[...]在文献中，我们没有发现，解释如何 实现对Linux新的调度策略文件。

[...]

在本文中，我们在[原文]深入的描述已经提出的所有步骤 实施新的调度策略需要。

[...]

这是一个简单的实现，调度算法的。但是，诸如中断，定时器和多处理器系统之类的高级问题，仅仅提到了一些，超出了本文的范围。