psi

先仔细看看文档

Monitors activate only when system enters stall state for the monitored psi metric and deactivates upon exit from the stall state. While system is in the stall state psi signal growth is monitored at a rate of 10 times per tracking window.

The kernel accepts window sizes ranging from 500ms to 10s, therefore min monitoring update interval is 50ms and max is 1s. Min limit is set to prevent overly frequent polling. Max limit is chosen as a high enough number after which monitors are most likely not needed and psi averages can be used instead.

psi_task_switch psi_group_change psi_rtpoll_work : rt 是什么含义? psi_cgroup_restart

使用接口

代码分析

/* Tracked task states */
enum psi_task_count {
	NR_IOWAIT,
	NR_MEMSTALL,
	NR_RUNNING,
	/*
	 * For IO and CPU stalls the presence of running/oncpu tasks
	 * in the domain means a partial rather than a full stall.
	 * For memory it's not so simple because of page reclaimers:
	 * they are running/oncpu while representing a stall. To tell
	 * whether a domain has productivity left or not, we need to
	 * distinguish between regular running (i.e. productive)
	 * threads and memstall ones.
	 */
	NR_MEMSTALL_RUNNING,
	NR_PSI_TASK_COUNTS = 4,
};

基本逻辑简单，检查这四个变量就可以找到标记的位置

基本的操作是在上下文的时间统计完成的:

每一个系统，或者 cgroup 中含有一个 psi_avgs_work

irq

仅仅在 psi_account_irqtime 中处理

其实就是借助 docs/kernel/sched-irq-time-accounting.md ，然后将之前 irq 占用的时间修改为 cgroup level 的

赞同这里的说法，这个其实用途不大: https://utcc.utoronto.ca/~cks/space/blog/linux/PSIIRQNumbersAndMeanings

cpu

io

#0  io_schedule_prepare () at kernel/sched/core.c:8777
#1  io_schedule () at kernel/sched/core.c:8810
#2  0xffffffff821ae17c in bit_wait_io (word=<optimized out>, mode=2) at kernel/sched/build_utility.c:4266
#3  0xffffffff821adb27 in __wait_on_bit (wq_head=wq_head@entry=0xffffffff82c06e88 <bit_wait_table+3336>, wbq_entry=wbq_entry@entry=0xffffc9004046bcf0, action=0xffffffff821ae170 <bit_wait_io>, mode=mode@entry=2) at kernel/sched/build_utility.c:4106
#4  0xffffffff821adc91 in out_of_line_wait_on_bit (word=word@entry=0xffff888100b4a000, bit=bit@entry=2, action=<optimized out>, mode=mode@entry=2) at kernel/sched/build_utility.c:4121
#5  0xffffffff81415c53 in wait_on_bit_io (bit=2, mode=2, word=0xffff888100b4a000) at ./include/linux/wait_bit.h:101
#6  0xffffffff814f1407 in wait_on_buffer (bh=0xffff888100b4a000) at ./include/linux/buffer_head.h:385
#7  journal_wait_on_commit_record (journal=0xffff888140d7e800, bh=0xffff888100b4a000) at fs/jbd2/commit.c:175
#8  jbd2_journal_commit_transaction (journal=journal@entry=0xffff888140d7e800) at fs/jbd2/commit.c:919
#9  0xffffffff814f80a8 in kjournald2 (arg=0xffff888140d7e800) at fs/jbd2/journal.c:210
#10 0xffffffff811556a4 in kthread (_create=0xffff88814066eb40) at kernel/kthread.c:376
#11 0xffffffff81002659 in ret_from_fork () at arch/x86/entry/entry_64.S:308

io_schedule_prepare 中设置

	current->in_iowait = 1;

之后在 psi_task_switch 采样 current::in_iowait

mem

memory 记录有，这个需要在被各种地方调用:

其实也就是标记一下，之后统计:

	current->in_memstall = 1;

TODO

本站所有文章转发 CSDN 将按侵权追究法律责任，其它情况随意。