Actions

EmSys

Introduction to Atomthreads

From EdWiki

Atomthreads is a free RTOS for embedded systems, released under the flexible, open source BSD license and is free to use for commercial or educational purposes without restriction.

It is targeted at systems that need only a lightweight scheduler and the usual RTOS primitives. No file system, IP stack or device drivers are included, but developers can bolt on their own as required. Atomthreads will always be a small number of C files which are easy to port to any platforms that require threading by adding a simple architecture-specific file.

Kernel library

atomkernel.c module implements the core kernel functionality of managing threads, context-switching and interrupt handlers. It also contains functions for managing queues of TCBs (task control blocks) which are used not only for the queue of ready threads, but also by other OS primitives (such as semaphores) for generically managing lists of TCBs.

Core kernel functionality such as managing the queue of ready threads and how context-switch decisions are made is described within the code. However a quick summary is as follows:

There is a ready queue of threads. There must always be at least one thread ready-to-run. If no application threads are ready, the internal kernel idle thread will be run. This ensures that there is a thread to run at all times.

Application code creates threads using atomThreadCreate(). These threads are added to the ready queue and eventually run when it is their turn (based on priority). When threads are currently-running they are taken off the ready queue. Threads continue to run until:

  • They schedule themselves out by calling an OS primitive which blocks, such as a timer delay or blocking on a semaphore. At this point they are placed on the queue of the OS primitive in which they are blocking (for example a timer delay or semaphore).
  • They are preempted by a higher priority thread. This could happen at any time if a kernel call from the currently-running thread or from an interrupt handler makes a higher priority thread ready-to-run. Generally this will occur immediately, and while the previously-running thread is still considered ready-to-run, it is no longer the currently-running thread so goes back on to the ready queue.
  • They are scheduled out after a time slice when another thread of the same priority is also ready. This happens on a timer tick, and ensures that threads of the same priority share time slices. In this case the previously running thread is still considered ready-to-run so is placed back on to the ready queue.


Thread scheduling decisions are made by atomSched(). This is called at several times, but should never be called by application code directly:

  • After interrupt handlers
    The scheduler is called after every interrupt handler has completed. This allows for any threads which have been made ready-to-run by the interrupt handler to be scheduled in. For example if an interrupt handler posts a semaphore which wakes up a thread of higher priority than the currently running thread, then the end of interrupt handler reschedule will schedule that thread in.
  • On timer ticks
    The timer tick is implemented as an interrupt handler so the end of interrupt call to the scheduler is made as normal, except that in this case round-robin rescheduling is allowed (where threads of the same priority are given a time slice each in round-robin fashion). This must only occur on timer ticks when the system tick count is incremented.
  • After any OS call changes ready states
    Any OS primitives which change the running state of a thread will call the scheduler to ensure that the change of thread state is noted. For example if a new thread is created using atomThreadCreate(), it will internally call the scheduler in case the newly created thread is higher priority than the currently running thread. Similarly OS primitives such as semaphores often make changes to a thread's running state. If a thread is going to sleep blocking on a semaphore then the scheduler will be run to ensure that some other thread is scheduled in its place. If a thread is woken by a semaphore post, the scheduler will also be called in case that thread should now be scheduled in (note that when semaphores are posted from an interrupt handler this is deferred to the end of interrupt scheduler call).

When a thread reschedule needs to take place, the scheduler calls out to the architecture specific port to perform the context switch, using archContextSwitch() which must be provided by each architecture port. This function carries out the low level saving and restoring of registers appropriate for the architecture. The thread being switched out must have a set of CPU registers saved, and the thread being scheduled in has a set of CPU registers restored (which were previously saved). In this fashion threads are rescheduled with the CPU registers in exactly the same state as when the thread was scheduled out.

New threads which have never been scheduled in have a pre-formatted stack area containing a set of CPU register values ready for restoring that appears exactly as if the thread had been previously scheduled out. In other words, the scheduler need not know when it restores registers to switch a thread in whether it has previously run or if it has never been run since the thread was created. The context-save area is formatted in exactly the same manner.

Reference

atomthreads Documentation