From 2e45c0f7ecca481cbec37ab8e8729926ecc68324 Mon Sep 17 00:00:00 2001 From: gatieme Date: Mon, 20 Jun 2016 16:39:15 +0800 Subject: [PATCH] =?UTF-8?q?=E8=BF=9B=E7=A8=8B=E8=B0=83=E5=BA=A6...?= MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit --- .../05-schedule/06-load_weight/README.md | 123 +++++++++++++++++- 1 file changed, 121 insertions(+), 2 deletions(-) diff --git a/study/kernel/01-process/05-schedule/06-load_weight/README.md b/study/kernel/01-process/05-schedule/06-load_weight/README.md index 5c13c01..c72a220 100644 --- a/study/kernel/01-process/05-schedule/06-load_weight/README.md +++ b/study/kernel/01-process/05-schedule/06-load_weight/README.md @@ -192,12 +192,131 @@ struct task_struct 因此我们就可以通过task_statuct->se.load获取负荷权重的信息, 而set_load_weight负责根据进程类型及其静态优先级计算符合权重. +#优先级和权重转换表 +------- + + +##优先级->权重转换表 +------- + + 内核不仅维护了负荷权重自身, 还保存另外一个数值, 用于击碎安被负荷权重重除的结果. +一般这个概念是这样的, 进程每降低一个nice值(优先级提升), 则多获得10%的CPU时间, 没升高一个nice值(优先级降低), 则放弃10%的CPU时间. + +为执行该策略, 内核需要将优先级转换为权重值, 并提供了一张优先级->权重转换表sched_prio_to_weight + +```c +// http://lxr.free-electrons.com/source/kernel/sched/sched.h?v=4.6#L1132 +/* + * To aid in avoiding the subversion of "niceness" due to uneven distribution + * of tasks with abnormal "nice" values across CPUs the contribution that + * each task makes to its run queue's load is weighted according to its + * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a + * scaled version of the new time slice allocation that they receive on time + * slice expiry etc. + */ + +#define WEIGHT_IDLEPRIO 3 +#define WMULT_IDLEPRIO 1431655765 + + +extern const int sched_prio_to_weight[40]; +extern const u32 sched_prio_to_wmult[40]; + + +// http://lxr.free-electrons.com/source/kernel/sched/core.c?v=4.6#L8484 +/* +* Nice levels are multiplicative, with a gentle 10% change for every +* nice level changed. I.e. when a CPU-bound task goes from nice 0 to +* nice 1, it will get ~10% less CPU time than another CPU-bound task +* that remained on nice 0. +* +* The "10% effect" is relative and cumulative: from _any_ nice level, +* if you go up 1 level, it's -10% CPU usage, if you go down 1 level +* it's +10% CPU usage. (to achieve that we use a multiplier of 1.25. +* If a task goes up by ~10% and another task goes down by ~10% then +* the relative distance between them is ~25%.) +*/ +const int sched_prio_to_weight[40] = { +/* -20 */ 88761, 71755, 56483, 46273, 36291, +/* -15 */ 29154, 23254, 18705, 14949, 11916, +/* -10 */ 9548, 7620, 6100, 4904, 3906, +/* -5 */ 3121, 2501, 1991, 1586, 1277, +/* 0 */ 1024, 820, 655, 526, 423, +/* 5 */ 335, 272, 215, 172, 137, +/* 10 */ 110, 87, 70, 56, 45, +/* 15 */ 36, 29, 23, 18, 15, +}; +``` + +对内核使用的范围[0, 39]中的每个nice级别, 该数组都有一个对应项. + + +##linux-4.4之前的prio_to_weight +------- + +>关于优先级->权重转换表sched_prio_to_weight +> +>在linux-4.4之前的内核中, 优先级->权重转换表用prio_to_weight表示, 定义在[kernel/sched/sched.h, line 1116](http://lxr.free-electrons.com/source/kernel/sched/sched.h?v=4.4#L1116), 与它一同定义的还有prio_to_wmult, 在[kernel/sched/sched.h, line 1139](http://lxr.free-electrons.com/source/kernel/sched/sched.h?v=4.4#L1139) +>均被定义为static const +> +>但是其实这种能够方式不太符合规范的编码风格, 因此常规来说, 我们的头文件中不应该存储结构的定义, 即为了是程序的模块结构更加清晰, 头文件中尽量只包含宏或者声明, 而将具体的定义， 需要分配存储空间的代码放在源文件中 +> +>否则如果在头文件中定义全局变量，并且将此全局变量赋初值，那么在多个引用此头文件的C文件中同样存在相同变量名的拷贝，关键是此变量被赋了初值，所以编译器就会将此变量放入DATA段，最终在连接阶段，会在DATA段中存在多个相同的变量，它无法将这些变量统一成一个变量，也就是仅为此变量分配一个空间，而不是多份空间，假定这个变量在头文件没有赋初值，编译器就会将之放入BSS段，连接器会对BSS段的多个同名变量仅分配一个存储空间 +> +>因此在新的内核中, 内核黑客们将这两个变量存放在了[kernel/sched/core.c](http://lxr.free-electrons.com/source/kernel/sched/core.c?v=4.6#L8472), 并加上了sched_前缀, 以表明这些变量是在进程调度的过程中使用的, 而在[kernel/sched/sched.h, line 1144](http://lxr.free-electrons.com/source/kernel/sched/sched.h?v=4.6#L1144)中则只包含了他们的声明. + + +下面我们列出其对比项 + +| 内核版本 | 实现 | 地址 | +| ------------- |:-------------:| +| <= linux-4.4 | static const int prio_to_weight[40] | [kernel/sched/sched.h, line 1116](http://lxr.free-electrons.com/source/kernel/sched/sched.h?v=4.4#L1116) | +| >=linux-4.5 | const int sched_prio_to_weight[40] | 声明在[kernel/sched/sched.h, line 1144](http://lxr.free-electrons.com/source/kernel/sched/sched.h?v=4.6#L1144), 定义在[kernel/sched/core.c](http://lxr.free-electrons.com/source/kernel/sched/core.c?v=4.6#L8472) + +其定义并没有发生变化, 依然是一个一对一NICE to WEIGHT的转换表 + +##1.25的乘积因子 +------- + +各数组之间的乘积因子是1.25. 要知道为何使用该因子, 可考虑下面的例子 + +两个进程A和B在nice级别0, 即静态优先级120运行, 因此两个进程的CPU份额相同, 都是50%, nice级别为0的进程, 查其权重表可知是1024. 每个进程的份额是1024/(1024+1024)=0.5, 即50% + +如果进程B的优先级+1(优先级降低), 成为nice=1, 那么其CPU份额应该减少10%, 换句话说进程A得到的总的CPU应该是55%, 而进程B应该是45%. 优先级增加1导致权重减少, 即1024/1.25=820, 而进程A仍旧是1024, 则进程A现在将得到的CPU份额是1024/(1024+820=0.55, 而进程B的CPU份额则是820/(1024+820)=0.45. 这样就正好产生了10%的差值. + + + +#进程负荷权重的计算 +------- + +set_load_weight负责根据进程类型极其静态优先级计算符合权重 + +执行转换的代码也需要实时进程. 实时进程的权重是普通进程的两倍, 另一方面, SCHED_IDLE进程的权值总是非常小 + +```c +/* +* Inverse (2^32/x) values of the sched_prio_to_weight[] array, precalculated. +* +* In cases where the weight does not change often, we can use the +* precalculated inverse to speed up arithmetics by turning divisions +* into multiplications: +*/ +const u32 sched_prio_to_wmult[40] = { +/* -20 */ 48388, 59856, 76040, 92818, 118348, +/* -15 */ 147320, 184698, 229616, 287308, 360437, +/* -10 */ 449829, 563644, 704093, 875809, 1099582, +/* -5 */ 1376151, 1717300, 2157191, 2708050, 3363326, +/* 0 */ 4194304, 5237765, 6557202, 8165337, 10153587, +/* 5 */ 12820798, 15790321, 19976592, 24970740, 31350126, +/* 10 */ 39045157, 49367440, 61356676, 76695844, 95443717, +/* 15 */ 119304647, 148102320, 186737708, 238609294, 286331153, +}; +``` + -http://blog.sina.com.cn/s/blog_9ca3f6e70102wkwp.html -http://lxr.free-electrons.com/source/kernel/sched/core.c#L3527 http://blog.chinaunix.net/uid-20671208-id-4909620.html http://blog.chinaunix.net/uid-20671208-id-4909623.html http://www.linuxidc.com/Linux/2016-05/131244.htm \ No newline at end of file