Skip to content
Calcrivo

Kernel Thread Calculator

Estimate kernel thread count and per-core distribution from kthreadd children observed via ps.

Inputs

Expected Total Kernel Threads

45

Consistency Check

Observed kthreadd child count is consistent with the expected per-CPU + global thread model.

Expected Per-CPU Threads (all cores)

32

Avg Per-CPU Threads per Core

4.0

Difference From Observed Count

0

Step by step

  1. Values used

    Total kthreadd Children (from ps --ppid 2) = 45; Per-CPU Thread Types (e.g. kworker, migration, ksoftirqd) = 4; CPU Cores = 8; Global (non-per-CPU) Kernel Threads = 13

  2. Expected thread count

    expected_total = per_cpu_thread_types × cores + global_threads

  3. Expected Total Kernel Threads

    = 45

  4. Consistency Check

    = Observed kthreadd child count is consistent with the expected per-CPU + global thread model.

  5. Expected Per-CPU Threads (all cores)

    = 32

  6. Avg Per-CPU Threads per Core

    = 4.0

  7. Difference From Observed Count

    = 0

How it works

kthreadd (PID 2) is the kernel's thread spawner — every kernel thread (kworker pools, migration/N, ksoftirqd/N, rcu_sched, writeback workers, etc.) is a descendant of it, visible via `ps --ppid 2` or `ps -ef | grep '\[.*\]'` (kernel threads are shown in square brackets). Many kernel thread types are per-CPU (one instance per core, like ksoftirqd/N and migration/N), so total kernel thread count scales with core count for that subset, plus a smaller set of global (single-instance) threads that don't multiply per core.

Formula

Expected thread count

expected_total = per_cpu_thread_types × cores + global_threads

t_cpu
per-CPU thread types
c
cores
t_global
global (non-per-CPU) threads

Frequently Asked Questions

How do I list all kernel threads on a running system?

`ps --ppid 2 -o pid,comm` lists direct kthreadd children (kernel threads shown in brackets), while `ps -ef | awk '$3==2'` gives a similar view; `ps -eLf | wc -l` includes all threads (kernel + userspace) for a total system thread count.

Why do kworker thread counts fluctuate over time?

Since Linux 4.10+, unbound kworker pools are created and destroyed dynamically based on workqueue demand (replacing the older fixed-per-CPU model), so the exact kworker/N:M count you observe varies with recent I/O and workqueue activity rather than being a fixed number tied only to core count.

Do kernel threads consume the same resources as user processes?

Kernel threads share the kernel's address space (no separate page tables) and never execute user-mode code, so they're lighter weight than typical user processes, but they still consume a task_struct, a kernel stack, and appear in scheduler accounting exactly like any other schedulable task.

You might also need

Related Calculators

Kernel Parameter CalculatorCPU Scheduling CalculatorModule Dependency Calculator