VMM CPU Scheduling in Xen

http://hi.baidu.com/%B0%B5%D4%C2%C1%F7%B9%E2/blog/item/0f68f850e5af7215377abe5d.html

（有时间我会给原版打个patch，让其更完整：）

文件：xen-3.4.2/xen/common/schedule.c

0、Per-CPU schedule data

struct schedule_data {
spinlock_t schedule_lock; /* spinlock protecting curr */
struct vcpu *curr; /* current task */
struct vcpu *idle; /* idle task for this cpu */
void *sched_priv;
struct timer s_timer; /* scheduling timer */
} __cacheline_aligned; （和cache line对齐）

DECLARE_PER_CPU(struct schedule_data, schedule_data);

1、SCHEDULE_SOFTIRQ
这个软中断用于触发调度器的执行。在调度器初始化的时候初始化，中断服务例程是调度函数schedule。
代码：
/* Initialise the data structures. */
void __initscheduler_init(void)
{
....................................
open_softirq(SCHEDULE_SOFTIRQ, schedule);
....................................
使用：
raise_softirq(SCHEDULE_SOFTIRQ); 执行该代码之后会导致所在cpu执行调度

2、schedule
选择下一个待执行的vcpu，执行vcpu切换
/*
* The main function
* - deschedule the current domain (scheduler independent).
* - pick a new domain (scheduler dependent).
*/
static void schedule(void)
{
struct vcpu *prev = current, *next = NULL;
s_time_t now = NOW();
struct schedule_data *sd;
struct task_slice next_slice;
....................................
/* get policy-specific decision on scheduling... */
next_slice =ops.do_schedule(now); //对于credit调度算法，会执行credit调度器中注册的csched_schedule函数

next = next_slice.task;

sd->curr = next;

if ( next_slice.time >= 0 ) /* -ve means no limit */
set_timer(&sd->s_timer, now + next_slice.time); //设置next需要运行的时间

if ( unlikely(prev == next) )
{
spin_unlock_irq(&sd->schedule_lock);
trace_continue_running(next);
returncontinue_running(prev); //直接执行schedule_tail函数，跳过vcpu切换
}

..................................
if (test_bit(_VPF_blocked, &prev->pause_flags)) { // 原来那句太难懂啦，我弄个易懂的：）
vcpu_runstate_change(prev, RUNSTATE_blocked, now);
} else {
if (vcpu_runnable(prev))
vcpu_runstate_change(prev, RUNSTATE_runnable, now);
else
vcpu_runstate_change(prev, RUNSTATE_offline, now);
}
prev->last_run_time = now;

ASSERT(next->runstate.state != RUNSTATE_running);
vcpu_runstate_change(next, RUNSTATE_running, now);

ASSERT(!next->is_running);
next->is_running = 1; //设置正在运行标志位
.........................

context_switch(prev, next); //执行vcpu切换
}

----------------------------------------------------------------------------
Per-VCPU pause flags.All are for "bit" operation

>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>

3rd 2nd 1st 0th

v v v v

migration blk-xen down blocked

>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>

/* Domain is blocked waiting for an event. */
#define _VPF_blocked 0
#define VPF_blocked 1

/* VCPU is offline. */
#define _VPF_down 1
#define VPF_down 2

/* VCPU is blocked awaiting an event to be consumed by Xen. */
#define _VPF_blocked_in_xen 2
#define VPF_blocked_in_xen 4

/* VCPU affinity has changed: migrating to a new CPU. */
#define _VPF_migrating 3
#define VPF_migrating 8

----------------------------------------------------------------------------

3、block
/* Block the currently-executing domain until a pertinent event occurs. */
static long do_block(void)
{
struct vcpu *v = current;

local_event_delivery_enable();
set_bit(_VPF_blocked, &v->pause_flags); //设置block标志位 _VPF_blocked

/* Check for events /after/ blocking: avoids wakeup waiting race. */
if ( local_events_need_delivery() ) { //检测current vcpu的pending位是否被设置以及event delivery是否被允许
clear_bit(_VPF_blocked, &v->pause_flags);
}else{
raise_softirq(SCHEDULE_SOFTIRQ); //current vcpu被标识为block状态，调度其他vcpu运行
}

return 0;
}


4、wake
唤醒指定的vcpu，使其从block状态却换到可运行状态
void vcpu_wake(struct vcpu *v)
{
unsigned long flags;

vcpu_schedule_lock_irqsave(v, flags);

if ( likely(vcpu_runnable(v)) ){
if ( v->runstate.state >= RUNSTATE_blocked )
vcpu_runstate_change(v, RUNSTATE_runnable, NOW());
SCHED_OP(wake, v); //执行调度器中的wake函数
}else if ( !test_bit(_VPF_blocked, &v->pause_flags) ){
if ( v->runstate.state == RUNSTATE_blocked )
vcpu_runstate_change(v, RUNSTATE_offline, NOW());
}
vcpu_schedule_unlock_irqrestore(v, flags);
}



重要函数记录：
struct domain *d;
struct vcpu *v;
for_each_vcpu( d, v ) 遍历domain中的vcpu
{
if ( v != current )
vcpu_pause(v);
}


struct domain *rd;
domid_t dom;
rd = rcu_lock_domain_by_id(dom); 根据domid获取相应的数据结构