systemtap-common 1.6-1ubuntu1 / usr / share / systemtap / tapset / scheduler.stp

// scheduler tapset
// Copyright (C) 2006 Intel Corporation.
// Copyright (C) 2005, 2006 IBM Corp.
// Copyright (C) 2010 Red Hat Inc.
//
// This file is part of systemtap, and is free software.  You can
// redistribute it and/or modify it under the terms of the GNU General
// Public License (GPL); either version 2, or (at your option) any
// later version.


function __is_idle:long()
%{ /* pure */
    /* Ways to detect idle-ness:
     * - pid() or tid() == 0
     * - current == current->parent
     * - current == this_rq()->idle
     * - others?
     */
    THIS->__retvalue = (current->pid == 0);
%}


/** probe scheduler.cpu_off - Process is about to stop running on a cpu
 *
 * @name: name of the probe point
 * @task_prev: the process leaving the cpu (same as current)
 * @task_next: the process replacing current
 * @idle: boolean indicating whether current is the idle process
 *
 * Context: The process leaving the cpu.
 *
 */
probe scheduler.cpu_off =
	kernel.trace("sched_switch") !,
	kernel.function("context_switch")
{
    name = "cpu_off"
    task_prev = $prev
    task_next = $next
    idle = __is_idle()
}


/** probe scheduler.cpu_on - Process is beginning execution on a cpu
 *
 * @name: name of the probe point
 * @task_prev: the process that was previously running on this cpu
 * @idle:- boolean indicating whether current is the idle process
 *
 * Context: The resuming process.
 */
probe scheduler.cpu_on = kernel.function("finish_task_switch") ?
{
    name = "cpu_on"
    task_prev = $prev
    idle = __is_idle()
}


/* probe scheduler.tick - Schedulers internal tick, a processes timeslice accounting is updated
 *
 *  @name: name of the probe point
 *  @idle: boolean indicating whether current is the idle process
 *
 * Context: The process whose accounting will be updated.
 */
probe scheduler.tick = kernel.function("scheduler_tick")
{
    name = "tick"
    idle = __is_idle()
}

/* probe scheduler.balance - A cpu attemptint to find more work.
 * 
 * @name: name of the probe point
 *
 * Context: The cpu looking for more work.
 */
probe scheduler.balance = kernel.function("idle_balance") ?
{
    name = "balance"
}


/** 
 * probe scheduler.ctxswitch - A context switch is occuring.
 *
 * @name: name of the probe point
 * @prev_pid: The PID of the process to be switched out
 * @next_pid: The PID of the process to be switched in
 * @prev_tid: The TID of the process to be switched out
 * @next_tid: The TID of the process to be switched in	 
 * @prev_task_name: The name of the process to be switched out
 * @next_task_name: The name of the process to be switched in
 * @prev_priority: The priority of the process to be switched out
 * @next_priority: The priority of the process to be switched in
 * @prevtsk_state: the state of the process to be switched out
 * @nexttsk_state: the state of the process to be switched in
 *
 * Currently, SystemTap can't access arguments of inline 
 * functions. So we choose to probe __switch_to instead 
 * of context_switch()       
 */

probe __scheduler.ctxswitch.tp = kernel.trace("sched_switch") 
{
	next_pid =  $next->tgid
	next_tid = $next->pid
	next_task = $next
	next_task_name = task_execname($next)
	nexttsk_state = $next->state
	next_priority = $next->prio
	prev_priority = $prev->prio
	prev_pid = $prev->tgid
	prev_tid = $prev->pid
	prev_task	=	$prev
	prev_task_name	=	task_execname($prev)
	prevtsk_state = $prev->state
}

probe __scheduler.ctxswitch.kp =
%( arch != "x86_64" && arch != "ia64" %?
	kernel.function("__switch_to")
%:
	kernel.function("context_switch")
%)
{
	/*
	 * Note that we prefer '$prev_p' here because on RHEL5
	 * (2.6.18-238.1.1.el5) the '__switch_to()' function has both
	 * a '$prev_p' and a '$prev' argument.  Since '$prev_p' is of
	 * the correct type (struct task_struct *), we need to look
	 * for it first.
	 */
	if (@defined($prev_p)) {
		prev_priority = $prev_p->prio
		prev_pid = $prev_p->tgid
		prev_tid = $prev_p->pid
		prev_task = $prev_p
		prev_task_name = task_execname($prev_p)
		prevtsk_state = $prev_p->state
	}
	else {
		prev_priority = $prev->prio
		prev_pid = $prev->tgid
		prev_tid = $prev->pid
		prev_task = $prev
		prev_task_name = task_execname($prev)
		prevtsk_state = $prev->state
	}

	if (@defined($next)) {
		next_priority = $next->prio
		next_pid = $next->tgid
		next_tid = $next->pid
		next_task = $next
		next_task_name = task_execname($next)
		nexttsk_state = $next->state
	}
	else if (@defined($next_p)) {
		next_priority = $next_p->prio
		next_pid = $next_p->tgid
		next_tid = $next_p->pid
		next_task = $next_p
		next_task_name = task_execname($next_p)
		nexttsk_state = $next_p->state
	}
	else {
		next_priority = $new->prio
		next_pid = $new->tgid
		next_tid = $new->pid
		next_task = $new
		next_task_name = task_execname($new)
		nexttsk_state = $new->state
	}
}

probe scheduler.ctxswitch =
	__scheduler.ctxswitch.tp !,
	__scheduler.ctxswitch.kp
{
	name = "ctxswitch"
}


/**
 * probe scheduler.kthread_stop - A thread created by kthread_create is being stopped
 * @thread_pid: PID of the thread being stopped
 * @thread_priority: priority of the thread
 */
probe __scheduler.kthread_stop.kp = kernel.function("kthread_stop")
{
	thread_pid = $k->tgid
	thread_priority = $k->prio
}
probe __scheduler.kthread_stop.tp = kernel.trace("sched_kthread_stop") 
{
	thread_pid = $t->tgid
	thread_priority = $t->prio
}
probe scheduler.kthread_stop =
	__scheduler.kthread_stop.tp !,
	__scheduler.kthread_stop.kp
{
	name = "kthread_stop"
}


/**
 * probe scheduler.kthread_stop.return - A kthread is stopped and gets the return value
 * @name: name of the probe point
 * @return_value: return value after stopping the thread
 */

probe __scheduler.kthread_stop.return.kp =
	kernel.function("kthread_stop").return
{
	return_value = $k->exit_code
}
probe __scheduler.kthread_stop.return.tp =
	kernel.trace("sched_kthread_stop_ret")
{
	return_value = $ret
}

probe scheduler.kthread_stop.return =
	__scheduler.kthread_stop.return.tp !,
	__scheduler.kthread_stop.return.kp
{
	name = "kthread_stop"
}

/**
 * probe scheduler.wait_task - Waiting on a task to unschedule (become inactive)
 * @name: name of the probe point
 * @task_pid: PID of the task the scheduler is waiting on
 * @task_priority: priority of the task
 */

probe scheduler.wait_task =
	kernel.trace("sched_wait_task") !,
	kernel.function("wait_task_inactive") ?
{
	name = "wait_task"
	task_pid = $p->tgid
	task_priority = $p->prio
}

/**
 * probe scheduler.wakeup - Task is woken up 
 * @name: name of the probe point
 * @task_pid: PID of the task being woken up
 * @task_priority: priority of the task being woken up
 * @task_cpu: cpu of the task being woken up
 * @task_state: state of the task being woken up
 * @task_tid: tid of the task being woken up
 */

probe scheduler.wakeup =
	kernel.trace("sched_wakeup") !,
	kernel.function("try_to_wake_up")
{
	name = "wakeup"
	task = $p
	task_pid = $p->tgid
	task_tid = $p->pid
	task_priority = $p->prio
	task_cpu = task_cpu($p)
	task_state = task_state($p)
}

/**
 * probe scheduler.wakeup_new - Newly created task is woken up for the first time
 * @name: name of the probe point
 * @task_pid: PID of the new task woken up
 * @task_priority: priority of the new task
 * @task_tid: TID of the new task woken up
 * @task_state: state of the task woken up
 * @task_cpu: cpu of the task woken up
 */
probe scheduler.wakeup_new =
	kernel.trace("sched_wakeup_new") !,
	kernel.function("wake_up_new_task")
{
	name = "wakeup_new"
	task_pid = $p->tgid
	task_priority = $p->prio
	task_cpu = task_cpu($p)
	task_state = task_state($p)
	task = $p
	task_tid = $p->pid
}

/**
 * probe scheduler.migrate - Task migrating across cpus
 * @name: name of the probe point
 * @task: the process that is being migrated
 * @pid: PID of the task being migrated
 * @priority: priority of the task being migrated
 * @cpu_from: the original cpu
 * @cpu_to: the destination cpu
 */
probe __scheduler.migrate.kp =
	kernel.function("set_task_cpu") !,
	kernel.function("pull_task") ?
{
	cpu_to = (@defined($new_cpu) ? $new_cpu
		  : @defined($cpu) ? $cpu : $this_cpu)
}
probe __scheduler.migrate.tp = kernel.trace("sched_migrate_task")
{
	cpu_to = $dest_cpu
}
probe scheduler.migrate =
	__scheduler.migrate.tp !,
	__scheduler.migrate.kp ?
{
	name = "migrate"
	task = $p
	pid = $p->tgid
	priority = $p->prio
	cpu_from = task_cpu($p)
}

function container_of_task_rcu:long(rcu_ptr:long)
{
	if (@defined(&@cast(0, "task_struct", "kernel<linux/sched.h>")->rcu)) {
		/*
		 * The following is the script language equivalent of:
		 *
		 *    return container_of(rcu_ptr, struct task_struct, rcu);
		 *
		 * More spelled out, that would look like:
		 *
		 *    return (rcu_ptr - offsetof(struct task_struct, rcu));
		 * 
		 * Notice we're casting 0 here on purpose to find the
		 * offset of the 'rcu' member of 'struct task_struct'.
		 */
		offset = &@cast(0, "task_struct",
				"kernel<linux/sched.h>")->rcu
		if (rcu_ptr > offset)
			return (rcu_ptr - offset)
	}
	return 0
}

/**
 * probe scheduler.process_free - Scheduler freeing a data structure for a process
 * @name: name of the probe point
 * @pid: PID of the process getting freed
 * @priority: priority of the process getting freed
 */
probe __scheduler.process_free.kp =
	kernel.function("delayed_put_task_struct") !,
	kernel.function("__put_task_struct")
{
	if (@defined($rhp)) {
		__tsk = container_of_task_rcu($rhp)
		pid = @cast(__tsk, "task_struct",
			    "kernel<linux/sched.h>")->tgid
		priority = @cast(__tsk, "task_struct",
				 "kernel<linux/sched.h>")->prio
	}
	else {
		pid = $tsk->tgid
		priority = $tsk->prio
	}
}
probe __scheduler.process_free.tp = kernel.trace("sched_process_free")
{
	pid = $p->tgid
	priority = $p->prio
}
probe scheduler.process_free =
	__scheduler.process_free.tp !,
	__scheduler.process_free.kp
{
	name = "process_free"
}

/**
 * probe scheduler.process_exit - Process exiting
 * @name: name of the probe point
 * @pid: PID of the process exiting
 * @priority: priority of the process exiting
 */
probe __scheduler.process_exit.kp = kernel.function("do_exit")
{
	if (@defined($tsk)) {
		pid = $tsk->tgid
		priority = $tsk->prio
	}
	else {
		__tsk = task_current()
		pid = @cast(__tsk, "task_struct",
			    "kernel<linux/sched.h>")->tgid
		priority = @cast(__tsk, "task_struct",
				 "kernel<linux/sched.h>")->prio
	}
}
probe __scheduler.process_exit.tp = kernel.trace("sched_process_exit")
{
	pid = $p->tgid
	priority = $p->prio
}

probe scheduler.process_exit =
	__scheduler.process_exit.tp !,
	__scheduler.process_exit.kp
{
	name = "process_exit"
}

/**
 * probe scheduler.process_wait - Scheduler starting to wait on a process
 * @name: name of the probe point
 * @pid: PID of the process scheduler is waiting on
 */
probe __scheduler.process_wait.kp = kernel.function("do_wait")
{
	if (@defined($wo)) {
		pid = $wo->wo_pid
	}
	else {
		pid = $pid
	}
}
probe __scheduler.process_wait.tp = kernel.trace("sched_process_wait")
{
	pid = $pid
}
probe scheduler.process_wait =
	__scheduler.process_wait.tp !,
	__scheduler.process_wait.kp
{
	name = "process_wait"
}

/**
 * probe scheduler.process_fork - Process forked
 * @name: name of the probe point
 * @parent_pid: PID of the parent process
 * @child_pid: PID of the child process
 */
probe __scheduler.process_fork.kp = kernel.function("do_fork").return
{
	parent_pid = @cast(task_current(), "task_struct",
			   "kernel<linux/sched.h>")->tgid
	child_pid = $return
}
probe __scheduler.process_fork.tp = kernel.trace("sched_process_fork")
{
	parent_pid = $parent->tgid
	child_pid = $child->tgid
}

probe scheduler.process_fork =
	__scheduler.process_fork.tp !,
	__scheduler.process_fork.kp
{
	name = "process_fork"
}

/**
 * probe scheduler.signal_send - Sending a signal
 * @name: name of the probe point
 * @pid: pid of the process sending signal
 * @signal_number: signal number
 */
probe __scheduler.signal_send.kp =
	kernel.function("__send_signal") !,
	kernel.function("send_signal")
{
	pid = $t->tgid
}
probe __scheduler.signal_send.tp = kernel.trace("sched_signal_send")
{
	pid = $p->tgid
}
probe scheduler.signal_send =
	__scheduler.signal_send.tp !,
	__scheduler.signal_send.kp
{
	name = "signal_send"
	signal_number = $sig
}