Posix/Linux Simulator for FreeRTOS
GCC and Eclipse
The Linux FreeRTOS simulator and accompanying documentation was kindly provided by William Davy. The simulator source code is included as a FreeRTOS contributed port. The documentation is included in a readme file within the source directory, and reproduced below.
Posix GCC Eclipse FreeRTOS SimulatorContributed by William Davy.
- 10/04/2010 – Added code for the Asynchronous Serial example. Refactored the generic Asynchronous module. Added support for the Run-time Statistics. Integrated the latest FreeRTOS code base. (Fixed the Eclipse GDB debug issues, see .gdbinit settings).
- 09/10/2009 – Added code and example tasks for Asynchronous IO specifically for Posix Message Queues (IPC) and UDP packets.
- 22/06/2009 – Fixed a bug regarding re-enabling of interrupts when a task previously had them disabled.
- Implemented vPortSetInterruptMaskFromISR and vPortClearInterruptMaskFromISR to enable the use of FreeRTOS system calls from interrupt handlers (Signal handlers which interact with queues).
- Changed the Signal Handler masks so that other signals can be used without having to change port layer code and signals will act more like interrupts.
- Changed the tick to be the real-time tick and added code in main.c that makes the process far more agreeable. The process will no longer steal all of the host OS idle time but will still run in pseudo real-time.
This is a port that allows FreeRTOS to act as a scheduler for pthreads within a process. It is designed to allow for development and testing of code in a Posix environment. It is considered a simulator because it will not keep real-time but it will retain the same deterministic task switching.
- make (tested with GNU Make 3.81)
- gcc (tested with gcc 4.4.3)
- Eclipse Galileo
- CDT 6.0
- Eclipse STATEVIEWER Plug-in v1.0.10
If you have Eclipse and the CDT installed then you can simply import the project (copying into the workspace or not). The project has Debug and Release build configurations and is a Managed-make C executable project so the Project Properties define the contents of the Makefile.
Alternatively, to build the executable without Eclipse and the CDT, simply cd into the Debug or Release directory and type make all.
The compilation and execution have been tested in an IA64 Ubuntu 8.10 and i386 Debian 4.0 build environments (using standard packages). No changes are necessary between the 64 bit and 32 bit builds. and 32 bit builds. The Debian machine is a 667MHz and it is able to run the simulator successfully, even with the 50 or so tasks within the demo.
Debugging using GDB
The port layer makes use of process signals. The four process signals that are used are SIGUSR1, SIGUSR2, SIGIO and SIGALRM. If a pthread is not waiting on the signal then GDB will pause the process on receipt of the signal. GDB must be told to ignore (and not print) the signal SIGUSR1 because it is received asynchronously by each thread. In GDB type handle SIGUSR1 nostop noprint pass to ensure that the debugging is not interrupted by the signals. The equivalent in Eclipse is achieved by using the ‘Signals’ view and getting the properties of SIGUSR1 and de-selecting the suspend option. See man signal for more information.
Alternatively, create a file in your home directory called .gdbinit and place the following two lines in it:
handle SIGUSR1 nostop noignore noprint
handle SIG34 nostop noignore noprint
Adding the two lines above to the .gdbinit file will tell GDB to not break on those signals. It may be necessary to change the Eclipse Debug configuration to point at the file ~/.gdbinit
If necessary, SIGUSR1 and SIGUSR2 can be re-defined to any other unused signals (such as the real-time signals).
There are three different timers available for use as the System tick; ITIMER_REAL, ITIMER_VIRTUAL and ITIMER_PROF. The default timer is ITIMER_VIRTUAL because it only counts when the process is executing in user space, therefore, the timer will stop when a break-point is hit. ITIMER_PROF is equivalent to ITIMER_VIRTUAL but it includes time executing system calls. ITIMER_REAL continues counting even when the process is not executing at all, therefore, it represents real-time. ITIMER_REAL is the only sensible option because the other timers don’t tick unless the process is actually running, hence, if nanosleep is called in the IDLE task hook, the time will hardly ever increase with the non-real timers.
One of the big advantages of debugging using Eclipse and the simulator is that you get a separate thread/task listing for each task you create. This means that you can inspect the call stack of any task when paused in the debugger, even if it is not the currently excuting task. If you also use the Eclipse STATEVIEWER Plug-in available from the Downloads section of www.HighIntegritySystems.com then you get a little bit of FreeRTOS Task scheduler information and Posix Threads call stacks. The STATEVIEWER Plug-in is shipped in a Windows Installer but that installer works in WINE and you simply need to ensure that the rtos.openrtos.viewer jar file is placed in your Eclipse/plugins directory.
Please note that the demo includes 50 or so tasks so it is quite processor intensive. The projects where I use this port for testing, have approximately 10 tasks and all of the IO and subsequent processing is event driven so the process as a whole spends 99% of its time in the IDLE task performing the nanosleep (which doesn’t consume any processing time). For this version I decided to mask out some of the most intensive tasks, set #define mainCPU_INTENSIVE_TASKS 1 to enable the intensive tasks or set it to 0 (default) to disable them.
Provided with the demo are two examples of Asynchronous IO that can be used for implementing event driven communicaiton. Asynchronous IO is important because it means that the processing time is not consumed by polling for the next message (especially important in this simulator because you don’t want to steal processing time from the other processes unless there is some work to be done). The three AsyncIO examples that have been provided are Posix Message Queues, UDP sockets and Serial communication (/dev/ttyS0).
The Posix Message Queues can be used for Inter-Process Communication on a single host. A queue is created by passing a Queue name. Processes communicate by opening a handle to the Queue by passing the same string name for that queue. Packets are then sent and received via the file descriptor. When a packet is written to a queue and there is a task waiting to receive the packet, a signal is sent to the task and the registered signal handler picks up the packet and delivers it via a FreeRTOS Queue to the task that is waiting on that Queue. See man mq_overview to find out more about Message Queues.
The UDP AsyncIO simply uses the BSD sockets interface but registers a signal handler that handles SIGIO signals which indicate that a packet is waiting to be read from a particular socket. When opening a socket, a callback function can be registered which is called when there is a packet waiting to be read. An example callback function is provided which takes a FreeRTOS queue as a parameter and delivers the received packet to that queue. Sockets can also be opened that are for sending only, which is done by passing NULL parameters to the open function.
The Serial communication example is very simple. It configures /dev/ttyS0 to be a RAW 38400 serial pipe and for each character that is received, it is echoed to the local console. The code should be enhanced with error correction and packet transmission so that it works in a similar method to the UDP example. Note that it is possible to add software flow control for better results, see man termios for more details.
The code provided for the AsyncIO can be extended to provide asynchronous IO on all file descriptors that support it. For each IO device, simply provide a method for configuring and opening that device and another method as a callback when data is received. Finally, register the opened file descriptor and the callback function with the AsyncIO module to start using it.
I use the two AsyncIO mechanisms as alternatives to the CAN bus. I can execute a CAN-Master and CAN-Slave on the same machine as two different instances (processes) of the simulator and they communicate via two Posix Message Queues. I can easily debug the CAN-Master/Slave using Eclipse and gdb. On top of that, I can use gprof and gcov(lcov), to profile and find code coverage of the tasks. Finally, I use the UDP sockets to provide a broadcast mechanism by sending packets to the local subnet, and all simulators on the all machines receive those and broadcast their replies. As such, the UDP broadcast provides the bus in CAN bus.
The send/receive tasks interact using FreeRTOS queues with CAN message objects so it is simple to replace the communication driver without any impact on the application code. In fact, the real CAN bus driver, on the embedded platform, is interrupt driven and delivers its messages to a FreeRTOS Queue. The use of signals in this Simulator mirrors the use of interrupts in the real world applications.
Port-Layer Design Methodology Justification
A simple implementation of a FreeRTOS Simulator would simply wrap the platform native threads and calls to switch Task contexts would call the OS suspend and resume thread API. This simulator uses the Posix signals as a method of controlling the execution of the underlying Posix threads. Signals can be delivered to the threads asynchronously so they interrupt the execution of the target thread, whilst suspended threads wait for synchronous signals to resume.
Typically, when designing a multi-threaded process, the multiple threads are used to allow for concurrent execution and to implement a degree of non-blocking on IO tasks. This simulator uses the threads not for their concurrent execution but solely to store the context of the execution. Signals and mutexes are used to synchronise the switching of the context but ultimately, the choice to change the context is driven by the FreeRTOS scheduler.
When a new Task is created, a pthread is created as the context for the execution of that Task. The pthread immediately suspends itself, returning the execution to the creator. When a pthread is suspended it is waiting in a call to ‘sigwait’ which is blocked until it receives a resume signal.
FreeRTOS Tasks can be switched in two manners, co-operatively by calling ‘taskYIELD()’ or pre-emptively as part of the System Tick. In this simulator, the Task contexts are switched by resuming the next task context (decided by the FreeRTOS Scheduler) and suspending the current context (with a brief handshake between the two).
The System tick is generated using an ITIMER and the signal is delivered (only to) the currently executing pthread. The System Tick Signal Handler increments the tick and selects the next context, resuming that thread and sending a signal to itself to suspend (which is only processed when exiting the System Tick Signal Handler as signals are queued).
The pthread_create and pthread_exit/cancel are system intensive calls which can rapidly saturate the processing time. The Common Demo code includes a Suicidal Tasks test which executes successfully, however, if the test is the only test which is being executed the process slowly grinds to a halt.
To prevent the process from stealing all of the Idle excution time of the Host OS, nano_sleep() can be used because it doesn’t use any signals in its implementation but will abort from the sleep/suspending the process immediately to service a signal. Therefore, the best way to use it is to set a sleep time longer than a FreeRTOS execution time slice and call it from the Idle task so that the process suspends until the next tick.
All feedback is welcome.