DTRACE_SCHED(4)

📦 version 15.1, in freebsd 14.5, freebsd 15.1, nixpkgs unstable

🌸 NAME 🌸

dtrace_sched — a DTrace provider for tracing CPU scheduling events. This is the DTRACE_SCHED(4) page from the Kernel Interfaces Manual, and bestie, it is exactly the kind of kernel-scheduler content that makes my brain go ✨.

✨ SYNOPSIS ✨

sched:::change-pri(struct thread *, struct proc *, uint8_t);

sched:::dequeue(struct thread *, struct proc *, void *);

sched:::enqueue(struct thread *, struct proc *, void *, int);

sched:::lend-pri(struct thread *, struct proc *, uint8_t,
    struct thread *);

sched:::load-change(int, int);

sched:::off-cpu(struct thread *, struct proc *);

sched:::on-cpu();

sched:::preempt();

sched:::remain-cpu();

sched:::surrender(struct thread *, struct proc *);

sched:::sleep();

sched:::tick(struct thread *, struct proc *);

sched:::wakeup(struct thread *, struct proc *);

💅 DESCRIPTION 💅

The DTrace sched provider lets you trace events related to CPU scheduling in the 4BSD and ULE schedulers. It's giving observability, but make it scheduler-core 💖.

  • sched:::change-pri() — fires when a thread's active scheduling priority is about to be updated. The first two arguments are the thread whose priority is about to be changed, and the corresponding process. The third argument is the new absolute priority for the thread, while the current value is given by args[0]->td_priority.
  • sched:::lend-pri() — fires when the currently-running thread elevates the priority of another thread via priority lending. The first two arguments are the thread whose priority is about to be changed, and the corresponding process. The third argument is the new absolute priority for the thread. The fourth argument is the currently-running thread.
  • sched:::dequeue() — fires immediately before a runnable thread is removed from a scheduler run queue. This may occur when the thread is about to begin execution on a CPU, or because the thread is being migrated to a different run queue. The latter event may occur in several circumstances: the scheduler may be attempting to rebalance load between multiple CPUs, the thread's scheduling priority may have changed, or the thread's CPU affinity settings may have changed. The first two arguments to sched:::dequeue() are the thread and corresponding process. The third argument is currently always NULL.
  • sched:::enqueue() — fires when a runnable thread is about to be added to a scheduler run queue. Its first two arguments are the thread and corresponding process. The third argument is currently always NULL. The fourth argument is a boolean value that is non-zero if the thread is enqueued at the beginning of its run queue slot, and zero if the thread is instead enqueued at the end.
  • sched:::load-change() — fires after the load of a thread queue is adjusted. The first argument is the cpuid for the CPU associated with the thread queue, and the second argument is the adjusted load of the thread queue, i.e., the number of elements in the queue.
  • sched:::off-cpu() — triggered by the scheduler suspending execution of the currently-running thread. The arguments are the thread and corresponding process selected to run following the currently-running thread.
  • sched:::on-cpu() — fires when the current thread has been selected to run on a CPU and is about to begin or resume execution.
  • sched:::remain-cpu() — if the two threads from the sched:::off-cpu() situation are the same, this probe fires instead. Plot twist, but make it scheduler-lawful 💀.
  • sched:::surrender() — fires when the scheduler is called upon to make a scheduling decision by a thread running on a different CPU, via an interprocessor interrupt. The arguments to this probe are the interrupted thread and its corresponding process. This probe currently always fires in the context of the interrupted thread.
  • sched:::preempt() — will fire immediately before the currently-running thread is preempted. When this occurs, the scheduler will select a new thread to run, and one of the sched:::off-cpu() or sched:::remain-cpu() probes will subsequently fire, depending on whether or not the scheduler selects the preempted thread.
  • sched:::sleep() — fires immediately before the currently-running thread is about to suspend execution and begin waiting for a condition to be met.
  • sched:::wakeup() — fires when a thread is set up to resume execution after having gone to sleep. Its arguments are the thread being awoken, and the corresponding process.
  • sched:::tick() — fires before each scheduler clock tick. Its arguments are the currently-running thread and its corresponding process.

🧚 ARGUMENTS 🧚

The sched provider probes use the kernel types struct proc and struct thread to represent processes and threads, respectively. These structures have many fields and are defined in sys/proc.h. In a probe body, the currently-running thread can always be obtained with the curthread global variable, which has type struct thread *. For example, when a running thread is about to sleep, the sched:::sleep() probe fires in the context of that thread, which can be accessed using curthread. The curcpu global variable contains the cpuid of the CPU on which the currently-running thread is executing.

🍓 EXAMPLES 🍓

The following script gives a breakdown of CPU utilization by process name:

sched:::on-cpu
{
        self->ts = timestamp;
}

sched:::off-cpu
/self->ts != 0/
{
        @[execname] = sum((timestamp - self->ts) / 1000);
        self->ts = 0;
}

Here, DTrace stores a timestamp each time a thread is scheduled to run, and computes the time elapsed in microseconds when it is descheduled. The results are summed by process name. Literally iconic accounting behavior 👑.

🦋 COMPATIBILITY 🦋

This provider is not compatible with the sched provider found in Solaris. In particular, the probe argument types are native FreeBSD types, and the sched:::cpucaps-sleep(), sched:::cpucaps-wakeup(), sched:::schedctl-nopreempt(), sched:::schedctl-preempt(), and sched:::schedctl-yield() probes are not available in FreeBSD.

The sched:::lend-pri() and sched:::load-change() probes are specific to FreeBSD.

🔥 SEE ALSO 🔥

👑 HISTORY 👑

The sched provider first appeared in FreeBSD 8.4 and 9.1. History really said “let there be scheduling probes” ✨.

💜 AUTHORS 💜

This manual page was written by Mark Johnston <markj@FreeBSD.org>.

signed, sealed, and scheduled by Debian on April 18, 2015 — page name DTRACE_SCHED(4) 🫶💖💅

📦 where this page comes from