services/threads/index.rst

.. _threads_v2:

Threads
#######

.. note::
   There is also limited support for using :ref:`nothread`.

.. contents::
    :local:
    :depth: 2

This section describes kernel services for creating, scheduling, and deleting
independently executable threads of instructions.

A :dfn:`thread` is a kernel object that is used for application processing
that is too lengthy or too complex to be performed by an ISR.

Any number of threads can be defined by an application (limited only by
available RAM). Each thread is referenced by a :dfn:`thread id` that is assigned
when the thread is spawned.

A thread has the following key properties:

* A **stack area**, which is a region of memory used for the thread's stack.
  The **size** of the stack area can be tailored to conform to the actual needs
  of the thread's processing. Special macros exist to create and work with
  stack memory regions.

* A **thread control block** for private kernel bookkeeping of the thread's
  metadata. This is an instance of type :c:struct:`k_thread`.

* An **entry point function**, which is invoked when the thread is started.
  Up to 3 **argument values** can be passed to this function.

* A **scheduling priority**, which instructs the kernel's scheduler how to
  allocate CPU time to the thread. (See :ref:`scheduling_v2`.)

* A set of **thread options**, which allow the thread to receive special
  treatment by the kernel under specific circumstances.
  (See :ref:`thread_options_v2`.)

* A **start delay**, which specifies how long the kernel should wait before
  starting the thread.

* An **execution mode**, which can either be supervisor or user mode.
  By default, threads run in supervisor mode and allow access to
  privileged CPU instructions, the entire memory address space, and
  peripherals. User mode threads have a reduced set of privileges.
  This depends on the :kconfig:option:`CONFIG_USERSPACE` option. See :ref:`usermode_api`.

.. _lifecycle_v2:

Lifecycle
***********

.. _spawning_thread:

Thread Creation
===============

A thread must be created before it can be used. The kernel initializes
the thread control block as well as one end of the stack portion. The remainder
of the thread's stack is typically left uninitialized.

Specifying a start delay of :c:macro:`K_NO_WAIT` instructs the kernel
to start thread execution immediately. Alternatively, the kernel can be
instructed to delay execution of the thread by specifying a timeout
value -- for example, to allow device hardware used by the thread
to become available.

The kernel allows a delayed start to be canceled before the thread begins
executing. A cancellation request has no effect if the thread has already
started. A thread whose delayed start was successfully canceled must be
re-spawned before it can be used.

Thread Termination
===================

Once a thread is started it typically executes forever. However, a thread may
synchronously end its execution by returning from its entry point function.
This is known as **termination**.

A thread that terminates is responsible for releasing any shared resources
it may own (such as mutexes and dynamically allocated memory)
prior to returning, since the kernel does *not* reclaim them automatically.

In some cases a thread may want to sleep until another thread terminates.
This can be accomplished with the :c:func:`k_thread_join` API. This
will block the calling thread until either the timeout expires, the target
thread self-exits, or the target thread aborts (either due to a
:c:func:`k_thread_abort` call or triggering a fatal error).

Once a thread has terminated, the kernel guarantees that no use will
be made of the thread struct.  The memory of such a struct can then be
re-used for any purpose, including spawning a new thread.  Note that
the thread must be fully terminated, which presents race conditions
where a thread's own logic signals completion which is seen by another
thread before the kernel processing is complete.  Under normal
circumstances, application code should use :c:func:`k_thread_join` or
:c:func:`k_thread_abort` to synchronize on thread termination state
and not rely on signaling from within application logic.

Thread Aborting
===============

A thread may asynchronously end its execution by **aborting**. The kernel
automatically aborts a thread if the thread triggers a fatal error condition,
such as dereferencing a null pointer.

A thread can also be aborted by another thread (or by itself)
by calling :c:func:`k_thread_abort`. However, it is typically preferable
to signal a thread to terminate itself gracefully, rather than aborting it.

As with thread termination, the kernel does not reclaim shared resources
owned by an aborted thread.

.. note::
    The kernel does not currently make any claims regarding an application's
    ability to respawn a thread that aborts.

Thread Suspension
==================

A thread can be prevented from executing for an indefinite period of time
if it becomes **suspended**. The function :c:func:`k_thread_suspend`
can be used to suspend any thread, including the calling thread.
Suspending a thread that is already suspended has no additional effect.

Once suspended, a thread cannot be scheduled until another thread calls
:c:func:`k_thread_resume` to remove the suspension.

.. note::
   A thread can prevent itself from executing for a specified period of time
   using :c:func:`k_sleep`. However, this is different from suspending
   a thread since a sleeping thread becomes executable automatically when the
   time limit is reached.

.. _thread_states:

Thread States
*************

A thread that has no factors that prevent its execution is deemed
to be **ready**, and is eligible to be selected as the current thread.

A thread that has one or more factors that prevent its execution
is deemed to be **unready**, and cannot be selected as the current thread.

The following factors make a thread unready:

* The thread has not been started.
* The thread is waiting for a kernel object to complete an operation.
  (For example, the thread is taking a semaphore that is unavailable.)
* The thread is waiting for a timeout to occur.
* The thread has been suspended.
* The thread has terminated or aborted.

  .. image:: thread_states.svg
     :align: center

.. note::

  Although the diagram above may appear to suggest that both **Ready** and
  **Running** are distinct thread states, that is not the correct
  interpretation. **Ready** is a thread state, and **Running** is a
  schedule state that only applies to **Ready** threads.

Thread Stack objects
********************

Every thread requires its own stack buffer for the CPU to push context.
Depending on configuration, there are several constraints that must be
met:

- There may need to be additional memory reserved for memory management
  structures
- If guard-based stack overflow detection is enabled, a small write-
  protected memory management region must immediately precede the stack buffer
  to catch overflows.
- If userspace is enabled, a separate fixed-size privilege elevation stack must
  be reserved to serve as a private kernel stack for handling system calls.
- If userspace is enabled, the thread's stack buffer must be appropriately
  sized and aligned such that a memory protection region may be programmed
  to exactly fit.

The alignment constraints can be quite restrictive, for example some MPUs
require their regions to be of some power of two in size, and aligned to its
own size.

Because of this, portable code can't simply pass an arbitrary character buffer
to :c:func:`k_thread_create`. Special macros exist to instantiate stacks,
prefixed with ``K_KERNEL_STACK`` and ``K_THREAD_STACK``.

Kernel-only Stacks
==================

If it is known that a thread will never run in user mode, or the stack is
being used for special contexts like handling interrupts, it is best to
define stacks using the ``K_KERNEL_STACK`` macros.

These stacks save memory because an MPU region will never need to be
programmed to cover the stack buffer itself, and the kernel will not need
to reserve additional room for the privilege elevation stack, or memory
management data structures which only pertain to user mode threads.

Attempts from user mode to use stacks declared in this way will result in
a fatal error for the caller.

If ``CONFIG_USERSPACE`` is not enabled, the set of ``K_THREAD_STACK`` macros
have an identical effect to the ``K_KERNEL_STACK`` macros.

Thread stacks
=============

If it is known that a stack will need to host user threads, or if this
cannot be determined, define the stack with ``K_THREAD_STACK`` macros.
This may use more memory but the stack object is suitable for hosting
user threads.

If ``CONFIG_USERSPACE`` is not enabled, the set of ``K_THREAD_STACK`` macros
have an identical effect to the ``K_KERNEL_STACK`` macros.

.. _thread_priorities:

Thread Priorities
******************

A thread's priority is an integer value, and can be either negative or
non-negative.
Numerically lower priorities takes precedence over numerically higher values.
For example, the scheduler gives thread A of priority 4 *higher* priority
over thread B of priority 7; likewise thread C of priority -2 has higher
priority than both thread A and thread B.

The scheduler distinguishes between two classes of threads,
based on each thread's priority.

* A :dfn:`cooperative thread` has a negative priority value.
  Once it becomes the current thread, a cooperative thread remains
  the current thread until it performs an action that makes it unready.

* A :dfn:`preemptible thread` has a non-negative priority value.
  Once it becomes the current thread, a preemptible thread may be supplanted
  at any time if a cooperative thread, or a preemptible thread of higher
  or equal priority, becomes ready.


A thread's initial priority value can be altered up or down after the thread
has been started. Thus it is possible for a preemptible thread to become
a cooperative thread, and vice versa, by changing its priority.

.. note::
    The scheduler does not make heuristic decisions to re-prioritize threads.
    Thread priorities are set and changed only at the application's request.

The kernel supports a virtually unlimited number of thread priority levels.
The configuration options :kconfig:option:`CONFIG_NUM_COOP_PRIORITIES` and
:kconfig:option:`CONFIG_NUM_PREEMPT_PRIORITIES` specify the number of priority
levels for each class of thread, resulting in the following usable priority
ranges:

* cooperative threads: (-:kconfig:option:`CONFIG_NUM_COOP_PRIORITIES`) to -1
* preemptive threads: 0 to (:kconfig:option:`CONFIG_NUM_PREEMPT_PRIORITIES` - 1)

.. image:: priorities.svg
   :align: center

For example, configuring 5 cooperative priorities and 10 preemptive priorities
results in the ranges -5 to -1 and 0 to 9, respectively.

.. _metairq_priorities:

Meta-IRQ Priorities
===================

When enabled (see :kconfig:option:`CONFIG_NUM_METAIRQ_PRIORITIES`), there is a special
subclass of cooperative priorities at the highest (numerically lowest)
end of the priority space: meta-IRQ threads.  These are scheduled
according to their normal priority, but also have the special ability
to preempt all other threads (and other meta-IRQ threads) at lower
priorities, even if those threads are cooperative and/or have taken a
scheduler lock. Meta-IRQ threads are still threads, however,
and can still be interrupted by any hardware interrupt.

This behavior makes the act of unblocking a meta-IRQ thread (by any
means, e.g. creating it, calling k_sem_give(), etc.) into the
equivalent of a synchronous system call when done by a lower
priority thread, or an ARM-like "pended IRQ" when done from true
interrupt context.  The intent is that this feature will be used to
implement interrupt "bottom half" processing and/or "tasklet" features
in driver subsystems.  The thread, once woken, will be guaranteed to
run before the current CPU returns into application code.

Unlike similar features in other OSes, meta-IRQ threads are true
threads and run on their own stack (which must be allocated normally),
not the per-CPU interrupt stack. Design work to enable the use of the
IRQ stack on supported architectures is pending.

Note that because this breaks the promise made to cooperative
threads by the Zephyr API (namely that the OS won't schedule other
thread until the current thread deliberately blocks), it should be
used only with great care from application code.  These are not simply
very high priority threads and should not be used as such.

.. _thread_options_v2:

Thread Options
***************

The kernel supports a small set of :dfn:`thread options` that allow a thread
to receive special treatment under specific circumstances. The set of options
associated with a thread are specified when the thread is spawned.

A thread that does not require any thread option has an option value of zero.
A thread that requires a thread option specifies it by name, using the
:literal:`|` character as a separator if multiple options are needed
(i.e. combine options using the bitwise OR operator).

The following thread options are supported.

:c:macro:`K_ESSENTIAL`
    This option tags the thread as an :dfn:`essential thread`. This instructs
    the kernel to treat the termination or aborting of the thread as a fatal
    system error.

    By default, the thread is not considered to be an essential thread.

:c:macro:`K_SSE_REGS`
    This x86-specific option indicate that the thread uses the CPU's
    SSE registers. Also see :c:macro:`K_FP_REGS`.

    By default, the kernel does not attempt to save and restore the contents
    of these registers when scheduling the thread.

:c:macro:`K_FP_REGS`
    This option indicate that the thread uses the CPU's floating point
    registers. This instructs the kernel to take additional steps to save
    and restore the contents of these registers when scheduling the thread.
    (For more information see :ref:`float_v2`.)

    By default, the kernel does not attempt to save and restore the contents
    of this register when scheduling the thread.

:c:macro:`K_USER`
    If :kconfig:option:`CONFIG_USERSPACE` is enabled, this thread will be created in
    user mode and will have reduced privileges. See :ref:`usermode_api`. Otherwise
    this flag does nothing.

:c:macro:`K_INHERIT_PERMS`
    If :kconfig:option:`CONFIG_USERSPACE` is enabled, this thread will inherit all
    kernel object permissions that the parent thread had, except the parent
    thread object.  See :ref:`usermode_api`.


.. _custom_data_v2:

Thread Custom Data
******************

Every thread has a 32-bit :dfn:`custom data` area, accessible only by
the thread itself, and may be used by the application for any purpose
it chooses. The default custom data value for a thread is zero.

.. note::
   Custom data support is not available to ISRs because they operate
   within a single shared kernel interrupt handling context.

By default, thread custom data support is disabled. The configuration option
:kconfig:option:`CONFIG_THREAD_CUSTOM_DATA` can be used to enable support.

The :c:func:`k_thread_custom_data_set` and
:c:func:`k_thread_custom_data_get` functions are used to write and read
a thread's custom data, respectively. A thread can only access its own
custom data, and not that of another thread.

The following code uses the custom data feature to record the number of times
each thread calls a specific routine.

.. note::
    Obviously, only a single routine can use this technique,
    since it monopolizes the use of the custom data feature.

.. code-block:: c

    int call_tracking_routine(void)
    {
        uint32_t call_count;

        if (k_is_in_isr()) {
	    /* ignore any call made by an ISR */
        } else {
            call_count = (uint32_t)k_thread_custom_data_get();
            call_count++;
            k_thread_custom_data_set((void *)call_count);
	}

        /* do rest of routine's processing */
        ...
    }

Use thread custom data to allow a routine to access thread-specific information,
by using the custom data as a pointer to a data structure owned by the thread.

Implementation
**************

Spawning a Thread
=================

A thread is spawned by defining its stack area and its thread control block,
and then calling :c:func:`k_thread_create`.

The stack area must be defined using :c:macro:`K_THREAD_STACK_DEFINE` or
:c:macro:`K_KERNEL_STACK_DEFINE` to ensure it is properly set up in memory.

The size parameter for the stack must be one of three values:

- The original requested stack size passed to
  ``K_THREAD_STACK`` or ``K_KERNEL_STACK`` family of stack instantiation
  macros.
- For a stack object defined with the ``K_THREAD_STACK`` family of
  macros, the return value of :c:macro:`K_THREAD_STACK_SIZEOF()` for that'
  object.
- For a stack object defined with the ``K_KERNEL_STACK`` family of
  macros, the return value of :c:macro:`K_KERNEL_STACK_SIZEOF()` for that
  object.

The thread spawning function returns its thread id, which can be used
to reference the thread.

The following code spawns a thread that starts immediately.

.. code-block:: c

    #define MY_STACK_SIZE 500
    #define MY_PRIORITY 5

    extern void my_entry_point(void *, void *, void *);

    K_THREAD_STACK_DEFINE(my_stack_area, MY_STACK_SIZE);
    struct k_thread my_thread_data;

    k_tid_t my_tid = k_thread_create(&my_thread_data, my_stack_area,
                                     K_THREAD_STACK_SIZEOF(my_stack_area),
                                     my_entry_point,
                                     NULL, NULL, NULL,
                                     MY_PRIORITY, 0, K_NO_WAIT);

Alternatively, a thread can be declared at compile time by calling
:c:macro:`K_THREAD_DEFINE`. Observe that the macro defines
the stack area, control block, and thread id variables automatically.

The following code has the same effect as the code segment above.

.. code-block:: c

    #define MY_STACK_SIZE 500
    #define MY_PRIORITY 5

    extern void my_entry_point(void *, void *, void *);

    K_THREAD_DEFINE(my_tid, MY_STACK_SIZE,
                    my_entry_point, NULL, NULL, NULL,
                    MY_PRIORITY, 0, 0);

.. note::
   The delay parameter to :c:func:`k_thread_create` is a
   :c:type:`k_timeout_t` value, so :c:macro:`K_NO_WAIT` means to start the
   thread immediately. The corresponding parameter to :c:macro:`K_THREAD_DEFINE`
   is a duration in integral milliseconds, so the equivalent argument is 0.

User Mode Constraints
---------------------

This section only applies if :kconfig:option:`CONFIG_USERSPACE` is enabled, and a user
thread tries to create a new thread. The :c:func:`k_thread_create` API is
still used, but there are additional constraints which must be met or the
calling thread will be terminated:

* The calling thread must have permissions granted on both the child thread
  and stack parameters; both are tracked by the kernel as kernel objects.

* The child thread and stack objects must be in an uninitialized state,
  i.e. it is not currently running and the stack memory is unused.

* The stack size parameter passed in must be equal to or less than the
  bounds of the stack object when it was declared.

* The :c:macro:`K_USER` option must be used, as user threads can only create
  other user threads.

* The :c:macro:`K_ESSENTIAL` option must not be used, user threads may not be
  considered essential threads.

* The priority of the child thread must be a valid priority value, and equal to
  or lower than the parent thread.

Dropping Permissions
====================

If :kconfig:option:`CONFIG_USERSPACE` is enabled, a thread running in supervisor mode
may perform a one-way transition to user mode using the
:c:func:`k_thread_user_mode_enter` API. This is a one-way operation which
will reset and zero the thread's stack memory. The thread will be marked
as non-essential.

Terminating a Thread
====================

A thread terminates itself by returning from its entry point function.

The following code illustrates the ways a thread can terminate.

.. code-block:: c

    void my_entry_point(int unused1, int unused2, int unused3)
    {
        while (1) {
            ...
	    if (<some condition>) {
	        return; /* thread terminates from mid-entry point function */
	    }
	    ...
        }

        /* thread terminates at end of entry point function */
    }

If :kconfig:option:`CONFIG_USERSPACE` is enabled, aborting a thread will additionally
mark the thread and stack objects as uninitialized so that they may be re-used.

Runtime Statistics
******************

Thread runtime statistics can be gathered and retrieved if
:kconfig:option:`CONFIG_THREAD_RUNTIME_STATS` is enabled, for example, total number of
execution cycles of a thread.

By default, the runtime statistics are gathered using the default kernel
timer. For some architectures, SoCs or boards, there are timers with higher
resolution available via timing functions. Using of these timers can be
enabled via :kconfig:option:`CONFIG_THREAD_RUNTIME_STATS_USE_TIMING_FUNCTIONS`.

Here is an example:

.. code-block:: c

   k_thread_runtime_stats_t rt_stats_thread;

   k_thread_runtime_stats_get(k_current_get(), &rt_stats_thread);

   printk("Cycles: %llu\n", rt_stats_thread.execution_cycles);

Suggested Uses
**************

Use threads to handle processing that cannot be handled in an ISR.

Use separate threads to handle logically distinct processing operations
that can execute in parallel.


Configuration Options
**********************

Related configuration options:

* :kconfig:option:`CONFIG_MAIN_THREAD_PRIORITY`
* :kconfig:option:`CONFIG_MAIN_STACK_SIZE`
* :kconfig:option:`CONFIG_IDLE_STACK_SIZE`
* :kconfig:option:`CONFIG_THREAD_CUSTOM_DATA`
* :kconfig:option:`CONFIG_NUM_COOP_PRIORITIES`
* :kconfig:option:`CONFIG_NUM_PREEMPT_PRIORITIES`
* :kconfig:option:`CONFIG_TIMESLICING`
* :kconfig:option:`CONFIG_TIMESLICE_SIZE`
* :kconfig:option:`CONFIG_TIMESLICE_PRIORITY`
* :kconfig:option:`CONFIG_USERSPACE`


API Reference
**************

.. doxygengroup:: thread_apis

.. doxygengroup:: thread_stack_api