Real-time operating systems (RTOS) are designed to handle time-critical tasks with predictable interrupt latency. High interrupt latency can cause problems in real-time applications, resulting in missed deadlines or other timing issues. There are several techniques that can be used to reduce interrupt latency in an RTOS:
Use Hardware Interrupts Judiciously
Hardware interrupts are the main source of unpredictable latency in an RTOS. Each hardware interrupt triggers a context switch which can add significant overhead. Hardware interrupts should be disabled during critical sections of code. Enable interrupts only around operations like blocking on queues, semaphores, or other synchronization primitives where context switches are allowed. Minimize the number of different interrupt sources to only those that are absolutely required.
Assign Priorities to Interrupts
If multiple interrupt sources are present, assign priorities so that the most time-critical interrupts will preempt less critical ones. This prevents lower priority interrupts from delaying higher priority ones. The RTOS scheduler can also assign thread priorities so that interrupt handlers run before application threads when necessary.
Use Interrupt Grouping
Grouping related interrupts into a single combined handler can reduce context switch overhead. For example, a UART driver may group transmit, receive and error interrupts into one UART handler rather than treating each one separately. The combined handler would check which flags are set and handle them together.
Implement Fast Interrupt Handlers
Interrupt handlers themselves should have very short execution times. Perform only minimal critical work in the handler and defer longer tasks to a thread. Context switching between threads is less time-critical than interrupting a running thread. Move common operations into library functions to keep handlers small.
Use an RTOS Tickless Mode
RTOS kernels periodically interrupt threads to tick counters, check if scheduling is needed, etc. Disabling the kernel tick timer can eliminate these periodic interrupts entirely. However, this requires the application to handle tickless timer management. It also requires interrupt handlers to force a context switch when scheduling is needed.
Minimize RTOS Kernel Objects
The RTOS may pause interrupts while accessing internal kernel objects like message queues and semaphores. Usage of these objects should be minimized, and they should not be accessed from within critical sections. For thread synchronization, consider disabling preemption just for the threads involved rather than using kernel objects.
Set Scheduler Time Slice Longer
The RTOS scheduler can preempt threads after a certain time slice expires to give other threads a chance to run. Setting a longer time slice means less frequent scheduler interrupts. However, this comes at the expense of reduced responsiveness and longer worst-case latency for other threads.
Assign Threads to Cores
On multicore systems, related threads can be assigned to the same CPU core rather than time-slicing across cores. This eliminates latency of cache misses and inter-core communication during context switches. For example, an application thread and its interrupt handler can stay on the same core.
Use Compiler Optimization Options
Enabling compiler optimizations such as -O3 can produce more efficient code, reducing interrupt handler execution times. Loop unrolling, function inlining and unused code elimination helps keep handlers small. Frame pointers and symbol information should also be omitted to reduce context switch stack usage.
Use an RTOS Optimization Mode
Some commercial RTOS kernels offer an optimization mode that disables unnecessary runtime error checks and statistics gathering. While this reduces the RTOS feature set, it eliminates internal housekeeping that can add latency. The tradeoff is accepted to meet real-time requirements.
Minimize ISR to Thread Signal Latency
Signalling from an interrupt handler to a thread adds latency if the thread is forced to wait until the ISR completes. Priority inheritance, priority boosting, and custom wake locks allow threads to preempt ISRs by inheriting a boosted priority when signaling occurs.
Tune RTOS Configuration
RTOS thread stacks, kernel tick rates, and other configuration parameters can be tuned to minimize overhead for a given application. Stack sizes directly impact context switch time, and smaller is better as long as stacks do not overflow. The kernel tick rate can also affect periodic timer latency.
Use RTOS-Aware Debugging
Debug and instrumentation features like printf() can add significant overhead in ISRs. RTOS-aware debugging allows the kernel state to be inspected without instrumentation. Debug builds still add latency so should not be used for final testing.
Measure and Profile Latency
Interrupt latency should be measured under worst-case conditions to validate real-time performance. Hardware tracing provides an accurate picture to identify problem areas. Profilers and debug builds also help characterize ISR and thread timing.
In summary, reducing interrupt latency requires hardware-specific RTOS configuration tuning, minimizing the use of interrupts, restricting kernel services in critical code, and designing applications to avoid accessing shared resources during time-critical moments. When done properly, RTOS interrupt latency can be reduced from microseconds down to tens or hundreds of nanoseconds to meet real-time requirements.
