This repository contains non-configured version of the kernel represented as a set of components. It is useful if you want to contribute to OS.
In case if you just need a kernel to use in your embedded application consider using preconfigured kernels available for [ARM](https://github.com/Eremex/fxrtos-lite-armv7m) and for [RISC-V](https://github.com/Eremex/fxrtos-lite-riscv32).
Please note that Lite version is a soft-realtime (or best-effort) kernel. It is NOT intended for deterministic hard-realtime operation.
Advanced features such as deterministic timers, low-latency deferred interrupt processing, multiprocessing support, privilege levels separation and security
are available only in full version. Please, contact EREMEX sales department for further details.
When developing latency-critical applications using Lite edition the following limitations should been taken into account:
- Broadcasting (or "notify all") synchronization objects such as condvar or event. More waiting threads results in longer latency since notification process is non-preemptive. Possible solutions: do not use broadcasting objects or limit maximum number of waiting threads.
- Priority-based notification policy with synchronization objects (i.e. posting semaphore and releasing the most prioritized waiting thread) uses linear search with scheduling disabled. This means that N waiting threads results in unbounded O(n) scheduling and interrupt latency. Possible solutions: use FIFO notification policy or limit maximum number of waiting threads for any synchronization object.
- Message queue flush releases up to N threads where N is queue length. Possible solutions: do not use queue flushing or limit queue length to reasonable value.
- Timers implementation uses sorted queues and linear search resulting in O(n) latency depending on number of active timers in the system. Possible solutions: Limit maximum number of software timers and do not use timeslicing.
