调度器24—CFS任务选核
作者：互联网
一、select_task_rq_fair()函数
CFS任务选核最终都是要走 select_task_rq_fair() 函数，三种CFS选核路径如下：
try_to_wake_up //core.c
select_task_rq(p, p->wake_cpu, SD_BALANCE_WAKE, wake_flags); //唤醒选核路径
wake_up_new_task //core.c
select_task_rq(p, task_cpu(p), SD_BALANCE_FORK, 0); //fork选核路径
sched_exec //core.c
select_task_rq(p, task_cpu(p), SD_BALANCE_EXEC, 0); //exec选核路径
对比可以看到，只有任务唤醒选核才传了wake_flags，可能选择sync唤醒。sched_domain 的flag中是否包含 SD_BALANCE_XX 可以通过下面方法查看：
/proc/sys/kernel/sched_domain # find ./ -name flags | xargs cat SD_BALANCE_NEWIDLE SD_BALANCE_EXEC SD_BALANCE_FORK SD_WAKE_AFFINE SD_SHARE_PKG_RESOURCES //MC SD_BALANCE_NEWIDLE SD_BALANCE_EXEC SD_BALANCE_FORK SD_WAKE_AFFINE SD_ASYM_CPUCAPACITY SD_PREFER_SIBLING //DIE
可以看出对于嵌入式这种大小核的Arm64架构中，MC和DIE都没有 SD_BALANCE_WAKE 这个标志。
函数解读：
/*
* select_task_rq_fair: Select target runqueue for the waking task in domains
* that have the 'sd_flag' flag set. In practice, this is SD_BALANCE_WAKE,
* SD_BALANCE_FORK, or SD_BALANCE_EXEC.
*
* Balances load by selecting the idlest CPU in the idlest group, or under
* certain conditions an idle sibling CPU if the domain has SD_WAKE_AFFINE set.
*
* Returns the target CPU number.
*
* preempt must be disabled.
*/
/*
* 传参：
* p: 待选核任务
* prev_cpu：任务之前运行的cpu
* sd_flag: 在包含此标志的sd中为任务选核，可取值 SD_BALANCE_WAKE(唤醒任务的选核)、SD_BALANCE_FORK(fork的新任务选核)、SD_BALANCE_EXEC(exec的任务选核)
* wake_flags: 可取值有 WF_SYNC、WF_FORK、WF_MIGRATED、WF_ON_CPU、WF_ANDROID_VENDOR，但是只对WF_SYNC进行了特殊处理，表示同步唤醒，其它标志被此函数忽略。
*
* 返回值：选出的目标cpu
*
* 执行环境：必须关抢占
*/
static int select_task_rq_fair(struct task_struct *p, int prev_cpu, int sd_flag, int wake_flags)
{
struct sched_domain *tmp, *sd = NULL;
int cpu = smp_processor_id();
int new_cpu = prev_cpu; //目标cpu先默认取prev_cpu
int want_affine = 0;
int sync = (wake_flags & WF_SYNC) && !(current->flags & PF_EXITING);
int target_cpu = -1;
//注册了这个HOOK且不是对新fork出来的任务进行选核，就更新任务的负载
if (trace_android_rvh_select_task_rq_fair_enabled() && !(sd_flag & SD_BALANCE_FORK))
sync_entity_load_avg(&p->se);
//Vendor厂商可以更改选核路径，若是选到了核，就直接退出了
trace_android_rvh_select_task_rq_fair(p, prev_cpu, sd_flag, wake_flags, &target_cpu); probe_android_rvh_select_task_rq_fair
if (target_cpu >= 0)
return target_cpu;
//只有对唤醒任务的选核才有可能走EAS选核路径
if (sd_flag & SD_BALANCE_WAKE) {
//记录当前任何和被唤醒任务p之间是否有固定的唤醒关系
record_wakee(p);
//全局使能控制/proc/sys/kernel/sched_energy_aware是否开启EAS
if (sched_energy_enabled()) {
//进行EAS选核，EAS选到核了，就直接退出。里面判断了rd->overutilized，若overutilized则直接失败退出
new_cpu = find_energy_efficient_cpu(p, prev_cpu, sync);
if (new_cpu >= 0)
return new_cpu;
//若EAS没有选到核，就回退为初始状态，继续走传统路径进行选核
new_cpu = prev_cpu;
}
//若p和current之间比较有相关性且任务p允许运行在当前cpu上，就认为是want_affine
want_affine = !wake_wide(p) && cpumask_test_cpu(cpu, p->cpus_ptr);
}
rcu_read_lock();
for_each_domain(cpu, tmp) { //MC-->DIE
/* If both 'cpu' and 'prev_cpu' are part of this domain, cpu is a valid SD_WAKE_AFFINE target.*/
/* SD_WAKE_AFFINE 标志MC和DIE都定义了，恒成立, 因此可以解释为只要满足want_affine, 就肯定会进入 */
if (want_affine && (tmp->flags & SD_WAKE_AFFINE) && cpumask_test_cpu(prev_cpu, sched_domain_span(tmp))) {
//若当前cpu就是任务p之前运行的prev_cpu, 不用调用wake_affine()选核了，若是唤醒选核直接选idle的兄弟cpu,非唤醒选核直接选prev_cpu
if (cpu != prev_cpu)
new_cpu = wake_affine(tmp, p, cpu, prev_cpu, sync); //若是既不走快速路径，又不走慢速路径，就返回它。
sd = NULL; /* Prefer wake_affine over balance flags */
break;
}
/*
* 注意 tmp->flags 不会包含 SD_BALANCE_WAKE, 因为MC和DIE都没设置这个标志。只有 SD_BALANCE_FORK、SD_BALANCE_EXEC 选核才有可能赋值
* 对于want_affine=0的情况，这个循环退出后，sd就是当前cpu对应的DIE层级的sd. 这里的作用是找到支持sd_flag的最高层级的sd
*/
if (tmp->flags & sd_flag)
sd = tmp;
else if (!want_affine)
break;
}
if (unlikely(sd)) { //SD_BALANCE_WAKE这类休眠唤醒的任务选核不可能走慢速路径，因为sd恒为NULL
/* Slow path */
new_cpu = find_idlest_cpu(sd, p, cpu, prev_cpu, sd_flag);
} else if (sd_flag & SD_BALANCE_WAKE) { //对唤醒的任务进行选核
/* Fast path */
new_cpu = select_idle_sibling(p, prev_cpu, new_cpu);
if (want_affine)
current->recent_used_cpu = cpu;
}
rcu_read_unlock();
return new_cpu;
}
1. 主要逻辑
(1) Vendor可以注册HOOK来决定是否执行这个原生的选核流程，若注册了且选到了核，则原生选核流程不再执行。否则执行原生选核流程。
(2) 判断是否进行EAS选核，若EAS选到和核则直接退出，否则继续往下走选核流程。EAS选核的条件为：
a. 必须是唤醒场景的选核，fork和exec场景的选核都不会走EAS。
b. EAS全局控制开关/proc/sys/kernel/sched_energy_aware必须是开启的。
c. 系统没有overutilized，若是overutilized则直接退出EAS选核。
(3) 满足wake_affine选出的核会作为一个候选结果，若是唤醒场景，还有走快速路径进行选核。
(4) 慢速路径选核只有fork和exec两类选核流程会走，唤醒选核是不会走慢速路径的。且慢速路径传参sd只可能是当前cpu对应的DIE层级的sd。
(5) 快速路径选核只有唤醒选核流程会走，fork和exec两类选核流程不会走快速选核流程。虽然名字叫select_idle_sibling，但是在同cluster中选不到idle核时，还是会在所有cluster中进行选核。
二、wake affine 属性
wake affine 只支持唤醒场景的选核。主要判断当前任务与被唤醒任务之间是否有比较固定的唤醒关系，来帮助选出一个候选cpu。
1. task_struct 中有三个成员用来线程唤醒信息
struct task_struct {
...
unsigned int wakee_flips; //该线程唤醒不同wakee的次数，值越大说明越偏向于唤醒不同的线程且唤醒越频繁
unsigned long wakee_flip_decay_ts; //记录上次wakee_flips衰减的时间点，其每秒衰减为原来的50%
struct task_struct *last_wakee; //该线程上次唤醒的线程
...
}
2. record_wakee()
/*
* 若current对p有比较固定的唤醒关系的话，current->wakee_flips将非常的小。
* 若经常唤醒其它不同的线程，其值将比较大
*/
static void record_wakee(struct task_struct *p)
{
/*
* Only decay a single time; tasks that have less then 1 wakeup per
* jiffy will not have built up many flips.
*/
//超过1秒衰减为原来的1/2
if (time_after(jiffies, current->wakee_flip_decay_ts + HZ)) {
current->wakee_flips >>= 1;
current->wakee_flip_decay_ts = jiffies;
}
//当前任务上下文中，若上次唤醒的不是任务p, 当前任务的wakee_flips加1
if (current->last_wakee != p) {
current->last_wakee = p;
current->wakee_flips++;
}
}
Waker唤醒wakee的场景中，有两种选核思路：一种是聚合的思路，即让waker和wakee尽量的close，从而提高cache hit。另外一种思考是分散，即让load尽量平均分配在多个cpu上。不同的唤醒模型使用不同的放置策略，两种简单的唤醒模型：
(1) 在1:N模型中，一个server会不断的唤醒多个不同的client。
(2) 在1:1模型中，线程A和线程B不断的唤醒对方。
在1:N模型中，如果N是一个较大的数值，那么让waker和wakee尽量的close会导致负荷的极度不平均，这会waker所在的sd会承担太多的task，从而引起性能下降。
在1:1模型中，让waker和wakee尽量的close不存在这样的问题，还能提高性能。
实际的程序中，唤醒关系可能没有那么简单，一个wakee可能是另外一个关系中的waker，交互可能是M:N的形式。考虑这样一个场景：waker把wakee拉近，而wakee自身的wakee flips比较大，由于wakee也会做waker，那么更多的线程也会拉近waker所在的sd，从而进一步加剧CPU资源的竞争。因此waker和wakee的wakee flips的数值都不能太大，太大的时候应该禁止wake affine。内核中通过 wake_wide() 来判断是否使能wake affine。
3. wake_wide()
/*
* 通过开关频率启发式检测 M:N 唤醒者/被唤醒者关系。
* 许多唤醒者唤醒的是与上次不同的任务，其频率大约是唤醒某一任务的 N 倍。
* 为了确定我们是否应该让负载分散与合并到共享缓存，我们在一对伙伴的一个中寻找最小为 llc_size 的“翻转”频率，在另一个中寻找大于 lls_size 频率的因子。
* 满足这两个条件，我们可以相对确定这种关系是非一夫一妻制的，伙伴数量超过套接字大小。
* Waker/wakee 是client/server、worker/dispatcher、中断源或任何无关紧要的东西，传播标准是明显的合作伙伴数量超过套接字大小。
*
* 对于有比较固定唤醒关系的，返回0，否则返回1
*/
static int wake_wide(struct task_struct *p)
{
unsigned int master = current->wakee_flips;
unsigned int slave = p->wakee_flips;
int factor = __this_cpu_read(sd_llc_size); //表示此cpu所在cluster中cpu的个数
if (master < slave)
swap(master, slave);
//这里是用的是或
if (slave < factor || master < slave * factor)
return 0;
return 1;
}
这个函数返回0，并且当前cpu在任务p的亲和性里面，就判断为want_affine。这里的or逻辑是存疑的，因为master和slave其一的wakee_flips比较小就满足wake affine，这会使得任务太容易在LLC domain堆积了。在1:N模型中（例如手机转屏的时候，一个线程会唤醒非常非常多的线程来处理configChange消息），master的wakee_flips巨大无比，slave的wakee_flips非常小，如果仍然wake affine是不合理的。
4. wake_affine()
/*
* select_task_rq_fair传参：sd为当前cpu对应的MC或DIE层级的sd; p为待选核任务; this_cpu为当前正在执行的cpu;
* prev_cpu为任务之前运行的cpu; sync表示选核时是否指定了同步WF_SYNC标志
*/
static int wake_affine(struct sched_domain *sd, struct task_struct *p, int this_cpu, int prev_cpu, int sync)
{
int target = nr_cpumask_bits;
if (sched_feat(WA_IDLE))
target = wake_affine_idle(this_cpu, prev_cpu, sync);
if (sched_feat(WA_WEIGHT) && target == nr_cpumask_bits) //若上面没有选到cpu
target = wake_affine_weight(sd, p, this_cpu, prev_cpu, sync);
schedstat_inc(p->se.statistics.nr_wakeups_affine_attempts); //增加统计计数
if (target == nr_cpumask_bits)
return prev_cpu; //若是还是没选到核，就返回prev_cpu，否则返回选到的cpu
schedstat_inc(sd->ttwu_move_affine);
schedstat_inc(p->se.statistics.nr_wakeups_affine);
return target;
}
/*
* The purpose of wake_affine() is to quickly determine on which CPU we can run
* soonest. For the purpose of speed we only consider the waking and previous CPU.
*
* wake_affine_idle() - only considers 'now', it check if the waking CPU is
* cache-affine and is (or will be) idle.
*
* wake_affine_weight() - considers the weight to reflect the average
* scheduling latency of the CPUs. This seems to work
* for the overloaded case.
*
* wake_affine传参：this_cpu是当前正在执行的cpu; prev_cpu是任务上次运行的cpu; sync是是
* 否指定了AF_SYNC同步唤醒标志
*/
static int wake_affine_idle(int this_cpu, int prev_cpu, int sync)
{
/*
* If this_cpu is idle, it implies the wakeup is from interrupt
* context. Only allow the move if cache is shared. Otherwise an
* interrupt intensive(密集型) workload could force all tasks onto one
* node depending on the IO topology or IRQ affinity settings.
*
* If the prev_cpu is idle and cache affine then avoid a migration.
* There is no guarantee that the cache hot data from an interrupt
* is more important than cache hot data on the prev_cpu and from
* a cpufreq perspective, it's better to have higher utilisation
* on one CPU.
*/
//若正在执行的cpu是idle的(中断唤醒)，且当前cpu和任务上次运行的cpu属于同一个cluster
if (available_idle_cpu(this_cpu) && cpus_share_cache(this_cpu, prev_cpu))
//若prev_cpu也idle就返回prev_cpu,否则返回当前正在运行的idle cpu
return available_idle_cpu(prev_cpu) ? prev_cpu : this_cpu;
/*
* 若同步唤醒且当前cpu上只有一个任务(也就是waker)在运行，就返回当前cpu，因为sync唤醒，
* 当前waker马上就要睡眠了。
*/
if (sync && cpu_rq(this_cpu)->nr_running == 1)
return this_cpu;
return nr_cpumask_bits; //没选到cpu返回它
}
/*
* wake_affine传参：sd是当前正在运行的cpu对应的MC或DIE层级的sd; p是待选核的任务; this_cpu是当前正在执行的cpu; prev_cpu是任务上次运行的cpu;
* sync是是否指定了AF_SYNC同步唤醒标志。
*/
static int wake_affine_weight(struct sched_domain *sd, struct task_struct *p, int this_cpu, int prev_cpu, int sync)
{
s64 this_eff_load, prev_eff_load;
unsigned long task_load;
//当前正在运行cpu上的CFS任务的负载
this_eff_load = cpu_load(cpu_rq(this_cpu)); //return cfs_rq->avg.load_avg;
if (sync) {
unsigned long current_load = task_h_load(current); //约为 current->se.avg.load_avg;
//当前任务的负载比当前cpu上cfs任务的负载之和还大？应该是安全的容错处理
if (current_load > this_eff_load)
return this_cpu;
this_eff_load -= current_load;
}
task_load = task_h_load(p); //约为 p->se.avg.load_avg;
this_eff_load += task_load; //计算若p把current从当前cpu上挤下去后，cpu上CFS任务的总负载
if (sched_feat(WA_BIAS))
this_eff_load *= 100; //乘以100
this_eff_load *= capacity_of(prev_cpu); //再乘以cpu算力中可用于cfs任务的算力(出去温限/irq/rt/dl占用后的算力)，乘以的竟然是prev_cpu的！
prev_eff_load = cpu_load(cpu_rq(prev_cpu)); //prev_cpu::rq.cfs_rq->avg.load_avg
prev_eff_load -= task_load; //这里为什么要减去？
if (sched_feat(WA_BIAS))
prev_eff_load *= 100 + (sd->imbalance_pct - 100) / 2; //MC和DIE都是乘以 100+(117-100)/2
prev_eff_load *= capacity_of(this_cpu); //乘以的竟然是this_cpu可用于cfs任务的算力！
/*
* If sync, adjust the weight of prev_eff_load such that if
* prev_eff == this_eff that select_idle_sibling() will consider
* stacking the wakee on top of the waker if no other CPU is
* idle.
* 如果同步唤醒，则调整 prev_eff_load 的权重，以便如果 prev_eff == this_eff 则 select_idle_sibling() 将考虑在没有其它空闲CPU的情况下将
* 被唤醒的任务放在唤醒它的CPU上。
*/
if (sync)
prev_eff_load += 1; //前面乘以100又乘以CPU可用于CFS任务的算力了，这里加1有啥用？
/*公式整理一下为：this_cpu_load/this_cpu_cap < 108.5% * prev_cpu_load/prev_cpu_cap 成立就返回this_cpu*/
return this_eff_load < prev_eff_load ? this_cpu : nr_cpumask_bits;
}
三、慢速路径
只有fork和exec类型的选核才会走慢速路径，慢速路径选核函数为 find_idlest_cpu()，走慢速路径选核时传参sd只可能是DIE层级的。
/*
* 只有fork或exec的选核才可能走这个函数。
*
* 传参：sd当前cpu对应的DIE层级的sd(只可能是DIE层级); p是待选核任务; cpu为当前cpu; prev_cpu为任务p之前运行的cpu; sd_flag表示哪种选核类型。
*
* 作用：在最idle的sg中找出最idle的cpu, 若是没找到就返回当前cpu.
*/
static inline int find_idlest_cpu(struct sched_domain *sd, struct task_struct *p, int cpu, int prev_cpu, int sd_flag)
{
int new_cpu = cpu;
//这个sd中的所有cpu都不在任务p的cpu亲和性里面
if (!cpumask_intersects(sched_domain_span(sd), p->cpus_ptr))
return prev_cpu;
/*
* We need task's util for cpu_util_without, sync it up to prev_cpu's last_update_time.
*/
//仅对于exec类型的选核更新任务的负载
if (!(sd_flag & SD_BALANCE_FORK))
sync_entity_load_avg(&p->se);
while (sd) {
struct sched_group *group;
struct sched_domain *tmp;
int weight;
if (!(sd->flags & sd_flag)) { //恒不成立,因为fork或exec类型的选核MC/DIE都支持
sd = sd->child;
continue;
}
//若是返回NULL,直接选的是当前cpu
group = find_idlest_group(sd, p, cpu);
if (!group) {
sd = sd->child;
continue;
}
//若是在d中找到了最忙的sg，就在其中选退出延迟最小且最后进idle的cpu,若是没有idle的cpu就选load最小的cpu
new_cpu = find_idlest_group_cpu(group, p, cpu);
if (new_cpu == cpu) {
/* Now try balancing at a lower domain level of 'cpu': */
sd = sd->child;
continue;
}
/* Now try balancing at a lower domain level of 'new_cpu': */
cpu = new_cpu;
weight = sd->span_weight;
sd = NULL;
for_each_domain(cpu, tmp) {
if (weight <= tmp->span_weight) //一进来就退出了，相当于没有执行
break;
if (tmp->flags & sd_flag)
sd = tmp;
}
}
return new_cpu;
}
1. find_idlest_group()
在当前cpu对应的sd中找出最空闲的sg
/*
* find_idlest_group() finds and returns the least busy CPU group within the domain.
*
* Assumes p is allowed on at least one CPU in sd.
*
* find_idlest_cpu传参：sd为当前cpu DIE层级对应的sd; p为待选核任务; cpu为当前cpu;
*/
static struct sched_group *find_idlest_group(struct sched_domain *sd, struct task_struct *p, int this_cpu)
{
struct sched_group *idlest = NULL, *local = NULL, *group = sd->groups;
struct sg_lb_stats local_sgs, tmp_sgs;
struct sg_lb_stats *sgs;
unsigned long imbalance;
struct sg_lb_stats idlest_sgs = {
.avg_load = UINT_MAX,
.group_type = group_overloaded, //初始化为最大的不均衡
};
//1024 * (sd->imbalance_pct-100) / 100
imbalance = scale_load_down(NICE_0_LOAD) * (sd->imbalance_pct-100) / 100;
//遍历MC层级先的每个cpu
do {
int local_group;
/* Skip over this group if it has no CPUs allowed */
//跳过任务p亲和性不允许的cpu
if (!cpumask_intersects(sched_group_span(group), p->cpus_ptr))
continue;
//当前正在运行的cpu是否在这个group中
local_group = cpumask_test_cpu(this_cpu, sched_group_span(group));
if (local_group) {
sgs = &local_sgs;
local = group;
} else {
sgs = &tmp_sgs;
}
update_sg_wakeup_stats(sd, group, sgs, p);
//group和idlest比较，看group是否比idlest更闲
if (!local_group && update_pick_idlest(idlest, &idlest_sgs, group, sgs)) {
idlest = group;
idlest_sgs = *sgs;
}
} while (group = group->next, group != sd->groups);
/* There is no idlest group to push tasks to */
//若没找到idlest sg，选当前cpu
if (!idlest)
return NULL;
/* The local group has been skipped because of CPU affinity */
//此sd中若没有local group就直接返回idlest，否则还要和local group再PK一下，此例中，local group恒存在的
if (!local)
return idlest;
/*
* If the local group is idler than the selected idlest group
* don't try and push the task.
*/
//若local sg更闲，选当前cpu
if (local_sgs.group_type < idlest_sgs.group_type)
return NULL;
/*
* If the local group is busier than the selected idlest group
* try and push the task.
*/
if (local_sgs.group_type > idlest_sgs.group_type)
return idlest;
//local group和idlest group一样闲的情况了：
switch (local_sgs.group_type) {
case group_overloaded:
case group_fully_busy:
/*
* When comparing groups across NUMA domains, it's possible for
* the local domain to be very lightly loaded relative to the
* remote domains but "imbalance" skews the comparison making
* remote CPUs look much more favourable. When considering
* cross-domain, add imbalance to the load on the remote node
* and consider staying local.
*/
//大小核架构没有SD_NUMA标志，恒不执行
if ((sd->flags & SD_NUMA) && ((idlest_sgs.avg_load + imbalance) >= local_sgs.avg_load))
return NULL;
/*
* If the local group is less loaded than the selected
* idlest group don't try and push any tasks.
*/
//最闲的sg和local sg差值大于阈值17*1024，选当前cpu
if (idlest_sgs.avg_load >= (local_sgs.avg_load + imbalance))
return NULL;
//local_avg_load/idlest_avg_load <= sd->imbalance_pct%，选当前cpu
if (100 * local_sgs.avg_load <= sd->imbalance_pct * idlest_sgs.avg_load)
return NULL;
break;
case group_imbalanced:
case group_asym_packing:
/* Those type are not used in the slow wakeup path */
return NULL; //选当前cpu
case group_misfit_task:
/* Select group with the highest max capacity */
if (local->sgc->max_capacity >= idlest->sgc->max_capacity)
return NULL; //选当前cpu
break;
case group_has_spare:
if (sd->flags & SD_NUMA) {
#ifdef CONFIG_NUMA_BALANCING //没使能，不执行
int idlest_cpu;
/*
* If there is spare capacity at NUMA, try to select
* the preferred node
*/
if (cpu_to_node(this_cpu) == p->numa_preferred_nid)
return NULL;
idlest_cpu = cpumask_first(sched_group_span(idlest));
if (cpu_to_node(idlest_cpu) == p->numa_preferred_nid)
return idlest;
#endif
/*
* Otherwise, keep the task on this node to stay close
* its wakeup source and improve locality. If there is
* a real need of migration, periodic load balance will
* take care of it.
*/
if (local_sgs.idle_cpus)
return NULL; //若local sg有idle cpu存在，选当前cpu
}
/*
* Select group with highest number of idle CPUs. We could also
* compare the utilization which is more stable but it can end
* up that the group has less spare capacity but finally more
* idle CPUs which means more opportunity to run task.
*/
//若local sg的idle cpu个数比idlest sg更多，选当前cpu
if (local_sgs.idle_cpus >= idlest_sgs.idle_cpus)
return NULL;
break;
}
return idlest;
}
update_sg_wakeup_stats 更新 sgs：
/*
* update_sg_wakeup_stats - Update sched_group's statistics for wakeup.
* @sd: The sched_domain level to look for idlest group.
* @group: sched_group whose statistics are to be updated.
* @sgs: variable to hold the statistics for this group.
* @p: The task for which we look for the idlest group/CPU.
*
* find_idlest_group传参：sd为当前cpu对应的MC层级的sd; group为MC层级的一个sg; sgs为未初始化的参数; p为待选核任务
* 作用：更新 sgs
*/
static inline void update_sg_wakeup_stats(struct sched_domain *sd,
struct sched_group *group, struct sg_lb_stats *sgs, struct task_struct *p)
{
int i, nr_running;
memset(sgs, 0, sizeof(*sgs));
//MC层级一个sg中只有一个cpu
for_each_cpu(i, sched_group_span(group)) {
struct rq *rq = cpu_rq(i);
unsigned int local;
sgs->group_load += cpu_load_without(rq, p);
sgs->group_util += cpu_util_without(i, p);
sgs->group_runnable += cpu_runnable_without(rq, p);
local = task_running_on_cpu(i, p); //判断p是否queue在这个cpu的rq上，此case下应该return 0
sgs->sum_h_nr_running += rq->cfs.h_nr_running - local;
nr_running = rq->nr_running - local;
sgs->sum_nr_running += nr_running;
/*
* No need to call idle_cpu_without() if nr_running is not 0
*/
if (!nr_running && idle_cpu_without(i, p))
sgs->idle_cpus++;
}
/* Check if task fits in the group */
//只有DIE层级有这个标志,恒不成立
if (sd->flags & SD_ASYM_CPUCAPACITY && !task_fits_capacity(p, group->sgc->max_capacity)) {
sgs->group_misfit_task_load = 1;
}
sgs->group_capacity = group->sgc->capacity;
sgs->group_weight = group->group_weight;
sgs->group_type = group_classify(sd->imbalance_pct, group, sgs);
/*
* Computing avg_load makes sense only when group is fully busy or overloaded
*/
if (sgs->group_type == group_fully_busy || sgs->group_type == group_overloaded)
sgs->avg_load = (sgs->group_load * SCHED_CAPACITY_SCALE) / sgs->group_capacity;
}
//根据imbalance_pct和sgs来确认group_type
static inline enum group_type group_classify(unsigned int imbalance_pct, struct sched_group *group, struct sg_lb_stats *sgs)
{
/*
* sgs->sum_nr_running <= sgs->group_weight 返回fase, 只要cpu个数大于任务个数，就不认为group_overloaded
* group_util/group_capacity > imbalance_pct/100=117% 或 group_runnable/group_capacity > imbalance_pct/100=117% 返回true
*/
if (group_is_overloaded(imbalance_pct, sgs))
return group_overloaded;
if (sg_imbalanced(group)) //return group->sgc->imbalance;
return group_imbalanced;
if (sgs->group_asym_packing)
return group_asym_packing;
if (sgs->group_misfit_task_load)
return group_misfit_task;
if (!group_has_capacity(imbalance_pct, sgs))
return group_fully_busy;
return group_has_spare;
}
后续判断的group和idlest比较，看group是否比idlest更闲：
static bool update_pick_idlest(struct sched_group *idlest, struct sg_lb_stats *idlest_sgs,
struct sched_group *group, struct sg_lb_stats *sgs)
{
if (sgs->group_type < idlest_sgs->group_type)
return true;
if (sgs->group_type > idlest_sgs->group_type)
return false;
/*
* The candidate and the current idlest group are the same type of
* group. Let check which one is the idlest according to the type.
*/
//group_type相等的情况下比较谁更闲
switch (sgs->group_type) {
case group_overloaded:
case group_fully_busy: //哪个sg的avg_load，哪个sg更闲
/* Select the group with lowest avg_load. */
if (idlest_sgs->avg_load <= sgs->avg_load)
return false;
break;
case group_imbalanced:
case group_asym_packing:
/* Those types are not used in the slow wakeup path */
return false;
case group_misfit_task: //哪个sg的最大算力大，哪个sg更闲
/* Select group with the highest max capacity */
if (idlest->sgc->max_capacity >= group->sgc->max_capacity)
return false;
break;
case group_has_spare: //哪个sg的idle cpu个数多，哪个sg更闲
/* Select group with most idle CPUs */
if (idlest_sgs->idle_cpus > sgs->idle_cpus)
return false;
/* Select group with lowest group_util */
//若idle cpu个数相等，哪个group_util小哪个更闲
if (idlest_sgs->idle_cpus == sgs->idle_cpus && idlest_sgs->group_util <= sgs->group_util)
return false;
break;
}
return true;
}
2. find_idlest_group_cpu()
/*
* find_idlest_group_cpu - find the idlest CPU among the CPUs in the group.
*
* find_idlest_cpu传参：group为在当前cpu对应的MC层级sd中找出的最闲的sg; p为待选核任务; this_cpu为当前正在执行的cpu
*/
static int find_idlest_group_cpu(struct sched_group *group, struct task_struct *p, int this_cpu)
{
unsigned long load, min_load = ULONG_MAX;
unsigned int min_exit_latency = UINT_MAX;
u64 latest_idle_timestamp = 0;
int least_loaded_cpu = this_cpu;
int shallowest_idle_cpu = -1;
int i;
/* Check if we have any choice: */
//对于MC层级的sg这里就直接就返回了，因为sg中只有一个cpu，DIE层级继续往下走。
if (group->group_weight == 1)
return cpumask_first(sched_group_span(group));
/* Traverse only the allowed CPUs */
//DIE层级的sg,遍历这个cluster中任务p亲和性允许的cpu
for_each_cpu_and(i, sched_group_span(group), p->cpus_ptr) {
if (sched_idle_cpu(i)) //若此cpu rq中只有SCHED_IDLE类型的任务，就选此cpu
return i;
//判断此cpu是否是idle cpu: rq->curr==rq->idle && rq->nr_running==0 && rq->ttwu_pending==0
if (available_idle_cpu(i)) {
struct rq *rq = cpu_rq(i);
struct cpuidle_state *idle = idle_get_state(rq); //return rq->idle_state;
if (idle && idle->exit_latency < min_exit_latency) {
/*
* We give priority to a CPU whose idle state has the smallest exit latency irrespective
* of any idle timestamp.
*/
min_exit_latency = idle->exit_latency; //记录最小退出延迟
latest_idle_timestamp = rq->idle_stamp;
shallowest_idle_cpu = i; //记录最小的退出延迟的cpu
} else if ((!idle || idle->exit_latency == min_exit_latency) && rq->idle_stamp > latest_idle_timestamp) { //idle退出延迟相等，且进入idle的时间更晚
/*
* If equal or no active idle state, then the most recently idled CPU might have
* a warmer cache.
*/
latest_idle_timestamp = rq->idle_stamp;
shallowest_idle_cpu = i;
}
} else if (shallowest_idle_cpu == -1) {//对于非idle类型的cpu,并且之前也没有遍历到idle cpu
load = cpu_load(cpu_rq(i));
if (load < min_load) {
min_load = load; //记录load最小的cpu
least_loaded_cpu = i;
}
}
}
//若是有idle cpu就选idle cpu,若是没有idle cpu就选load最小的cpu
return shallowest_idle_cpu != -1 ? shallowest_idle_cpu : least_loaded_cpu;
}
同 load_balance()寻找最忙的cpu类似。这里的慢速路径，选择最空闲的cpu，先在sd中找最空闲的sg，然后再sg中找最空闲的cpu。这里最空闲指的是当前cpu所在cluster内，有idle的cpu就算idle的cpu，若是没有idle的cpu就选负载最小的cpu。
四、快速路径
快速选核路径只有唤醒选核流程才会走。虽然函数名中是选兄弟cpu，但是在选不到idle cpu的时候，也会在其它Cluster进行选核
/*
* Try and locate an idle core/thread in the LLC cache domain.
*
* select_task_rq_fair传参：p为待选核任务; prev是任务上次运行的cpu; target是备选目标cpu，可能是prev_cpu也可能是wake_affine选出的cpu
*/
static int select_idle_sibling(struct task_struct *p, int prev, int target)
{
struct sched_domain *sd;
unsigned long task_util;
int i, recent_used_cpu;
/*
* On asymmetric system, update task utilization because we will check
* that the task fits with cpu's capacity.
*/
if (static_branch_unlikely(&sched_asym_cpucapacity)) {
sync_entity_load_avg(&p->se);
task_util = uclamp_task_util(p); //将util_est clamp 在MIN和MAX之间
}
/*
* 传入的target是一个idle cpu且可以容纳的了uclamp后的任务的util，就直接选之前选的cpu。
* Wake affine的场景下，target cpu是通过wake_affine函数寻找的target cpu，其他场景下，
* target CPU其实等于prev CPU。
*/
if ((available_idle_cpu(target) || sched_idle_cpu(target)) && asym_fits_capacity(task_util, target))
return target;
/*
* If the previous CPU is cache affine and idle, don't be stupid:
*
* 若传入的参数是由wake_affine选出的cpu，且和prev_cpu属于同一个cluster(共享L2-Cache), 且prev_cpu是idle的, 且prev_cpu能容纳的下uclamp后的任务p的util
*/
if (prev != target && cpus_share_cache(prev, target) && (available_idle_cpu(prev) || sched_idle_cpu(prev)) && asym_fits_capacity(task_util, prev))
return prev;
/*
* Allow a per-cpu kthread to stack with the wakee if the kworker thread and the tasks previous CPUs are the same.
* The assumption is that the wakee queued work for the per-cpu kthread that is now complete and the wakeup is
* essentially a sync wakeup. An obvious example of this pattern is IO completions.
* 翻译：
* 如果 kworker 线程和任务prev_cpu是同一个，则允许 per-cpu kthread 与被唤醒线程堆叠。假设是被唤醒任务往per-cpu的kthread
* 上queue work, 唤醒本质上是同步唤醒。这种模式的一个明显例子是 IO 完成。
*
* 当前任务是一个per-cpu的内核线程，且任务的prev_cpu就是当前cpu，且当前cpu上最多只有一个任务在运行，就选prev_cpu.
*/
if (is_per_cpu_kthread(current) && prev == smp_processor_id() && this_rq()->nr_running <= 1) {
return prev;
}
/* Check a recently used CPU as a potential idle candidate: */
/*
* recent_used_cpu的更新位置：
* select_task_rq_fair: 更新 current->recent_used_cpu 为当前正在运行的cpu
* select_idle_sibling: 将待选核任务 p->recent_used_cpu 设置为 prev_cpu
*
* 若 p->recent_used_cpu 既不是prev_cpu也不是target_cpu, 但是是和target_cpu共cluster的，且是idle的, 且任务的亲和性允许，
* 且算力能够容纳任务，那么就选 p->recent_used_cpu。
*/
recent_used_cpu = p->recent_used_cpu;
if (recent_used_cpu != prev && recent_used_cpu != target && cpus_share_cache(recent_used_cpu, target) &&
(available_idle_cpu(recent_used_cpu) || sched_idle_cpu(recent_used_cpu)) && cpumask_test_cpu(p->recent_used_cpu, p->cpus_ptr) &&
asym_fits_capacity(task_util, recent_used_cpu)) {
/*
* Replace recent_used_cpu with prev as it is a potential candidate for the next wake:
*/
p->recent_used_cpu = prev;
return recent_used_cpu; //还将p放在它最近使用过的cpu上
}
/*
* For asymmetric CPU capacity systems, our domain of interest is sd_asym_cpucapacity rather than sd_llc.
*/
if (static_branch_unlikely(&sched_asym_cpucapacity)) {
sd = rcu_dereference(per_cpu(sd_asym_cpucapacity, target)); //target_cpu对应的DIE层级的sd
/*
* On an asymmetric CPU capacity system where an exclusive cpuset defines a symmetric island (i.e. one unique
* capacity_orig value through the cpuset), the key will be set but the CPUs within that cpuset will not have
* a domain with SD_ASYM_CPUCAPACITY. These should follow the usual symmetric capacity path.
*/
if (sd) {
//这个函数还可能返回-1,但是并没有兼容这个错误的结果！后续的使用也没有做判断，是个BUG!!
i = select_idle_capacity(p, sd, target);
return ((unsigned)i < nr_cpumask_bits) ? i : target; //这里会返回，下面的部分不会被执行到
}
}
/* 下面就是非大小核架构会执行的路径了 */
sd = rcu_dereference(per_cpu(sd_llc, target)); //target_cpu对应的MC层级的sd
if (!sd)
return target;
i = select_idle_core(p, sd, target); //CONFIG_SCHED_SMT没定义为空函数
if ((unsigned)i < nr_cpumask_bits)
return i;
i = select_idle_cpu(p, sd, target); //CONFIG_SCHED_SMT没定义为空函数
if ((unsigned)i < nr_cpumask_bits)
return i;
i = select_idle_smt(p, sd, target); //CONFIG_SCHED_SMT没定义为空函数
if ((unsigned)i < nr_cpumask_bits)
return i;
return target;
}
/*
* select_idle_sibling传参：p为待选核任务; sd为参数target_cpu对应的DIE层级的sd; target为候选的target_cpu
*
* 作用：从target_cpu开始遍历，找一个能容纳下任务p的idle cpu, 若所有idle cpu都不能容纳下任务，就返回算力最大的idle cpu,
* 否则返回-1;
*/
static int select_idle_capacity(struct task_struct *p, struct sched_domain *sd, int target)
{
unsigned long task_util, best_cap = 0;
int cpu, best_cpu = -1;
struct cpumask *cpus;
cpus = this_cpu_cpumask_var_ptr(select_idle_mask);
//先赋值后使用
cpumask_and(cpus, sched_domain_span(sd), p->cpus_ptr);
task_util = uclamp_task_util(p);
//这里从target_cpu开始遍历，利用了之前判断的结果target_cpu的算力是能容纳任务p的
for_each_cpu_wrap(cpu, cpus, target) {
unsigned long cpu_cap = capacity_of(cpu);
if (!available_idle_cpu(cpu) && !sched_idle_cpu(cpu)) //若此cpu非idle则跳过这个cpu
continue;
if (fits_capacity(task_util, cpu_cap)) //返回一个能容纳任务的idle cpu
return cpu;
if (cpu_cap > best_cap) { //若是没有能容纳的，则返回一个算力最大的
best_cap = cpu_cap;
best_cpu = cpu;
}
}
return best_cpu;
}
1. 此函数中要使用到p的util, 所以需要先更新其负载。
2. 快速路径选核依次判断：
(1) 传入的target是一个idle cpu, 且可以容纳的了uclamp后的任务的util，就直接选之前选的这个 target_cpu
(2) 若传入的参数target是由wake_affine选出的cpu，且和prev_cpu属于同一个cluster(共享L2-Cache), 且prev_cpu是idle的, 且prev_cpu能容纳的下uclamp后的任务p的util, 那么选择 prev_cpu。
(3) 当前任务是一个per-cpu的内核线程，且任务的prev_cpu就是当前cpu，且当前cpu上最多只有一个任务在运行，就选 prev_cpu。一个典型的例子是IO操作场景。
(4) 若 p->recent_used_cpu 既不是prev_cpu也不是target_cpu, 但是是和target_cpu共cluster的，且是idle的, 且任务的亲和性允许，且算力能够容纳任务，那么就选 p->recent_used_cpu。
(5) 从target_cpu开始遍历，找一个能容纳下任务p的idle cpu, 若所有idle cpu都不能容纳下任务，就返回算力最大的idle cpu, 否则返回-1;
3. 注意有个BUG: select_idle_capacity()返回-1的话没有做容错处理。
五、总结
1. 有三种任务选核路径，sd_flag分别对应为 SD_BALANCE_WAKE(唤醒场景)、SD_BALANCE_FORK(fork新任务场景)、SD_BALANCE_EXEC(exec场景)。其中EAS选核和wake-affine选核只适用于唤醒场景，EAS选到核后直接退出，wake-affine选到核后还需要走快速路径继续选核，希望在当前cpu的同cluster中选一个空闲cpu或prev_cpu，若选不到，最终也会尝试在所有cluster中选核。fork新任务场景和exec场景走慢速选核路径，找出系统中最闲的cpu作为目标cpu。
参考：https://blog.csdn.net/feelabclihu/article/details/122007603?spm=1001.2014.3001.5501
标签：24,选核,group,load,cpu,idle,prev,CFS,sd 来源： https://www.cnblogs.com/hellokitty2/p/15750931.html