36. Ada Support

36.1. Introduction

RTEMS has long had support for the Ada programming language by supporting the GNU Ada Compiler (GNAT). There are two primary components to this support:

  • Ada Programming Language Support

  • Classic API Ada Bindings

36.2. Ada Programming Language Support

The Ada programming natively supports multi-threaded programming with its own tasking and concurrency model. Native Ada multi-threaded applications should work using GNAT/RTEMS with no changes.

The application developer will have to account for the specific requirements of the GNAT Run-Time when configuring RTEMS. There are example Ada programs with RTEMS configuration and startup sequences.

36.3. Classic API Ada Bindings

An Ada language binding exists for a subset of the RTEMS Classic API. In the early 1990’s, there were C and Ada implementations of RTEMS which were functionally equivalent. The source structure was as similar as possible. In fact, the top level c/ directory at one point had a sibling ada/. The current Ada language bindings and test code was derived from that Ada implementation.

The Ada binding specifically excludes some methods which are either not safe or not intended for use from Ada programs. However, methods are generally only added to this binding when a user makes a requests. Thus some methods that could be supported are not. If in doubt, ask about a methods and contribute bindings.

The bindings are located in the c/src/ada directory of the RTEMS source tree. The tests are in c/src/ada-tests. The bindings following a simple pattern to map the C Classic API calls into Ada subprograms. The following rules are used:

  • All RTEMS interfaces are in the RTEMS Ada package. The rtems_ and RTEMS_ prefixes in the C version of the Classic API thus correspond to “RTEMS.” in Ada symbol nomenclature. For example, rtems_task_create() in C is RTEMS.Task_Create() in Ada.

  • Classic API directives tend to return an rtems_status_code. Some directives also have an output parameter such as an object id on a create operation. Ada subprograms are either pure functions with only a single return value or subprograms. For consistency, the returned status code is always the last parameter of the Ada calling sequence.

Caution should be exercised when writing programs which mix Ada tasks, Classic API tasks, and POSIX API threads. Ada tasks use a priority numbering scheme defined by the Ada programming language. Each Ada task is implemented in GNAT/RTEMS as a single POSIX thread. Thus Ada task priorities must be mapped onto POSIX thread priorities. Complicating matters, Classic API tasks and POSIX API threads use different numbering schemes for priority. Low numbers are high priority in the Classic API while indicating low priority in the POSIX threads API. Experience writing mixed threading model programs teaches that creating a table of the priorities used in the application with the value in all tasking models used is helpful.

The GNAT run-time uses a priority ceiling mutex to protect its data structures. The priority ceiling value is one priority more important than the most important Ada task priority (in POSIX API terms). Do not invoke any services implemented in Ada from a thread or task which is of greater priority. This will result in a priority ceiling violation error and lead to a failure in the Ada run-time.

Exercise extreme caution when considering writing code in Ada which will execute in the context of an interrupt handler. Hardware interrupts are processed outside the context of any thread in RTEMS and this can lead to violating assumptions in the GNAT run-time. Specifically a priority ceiling mutex should never be used from an ISR and it is difficult to predict when the Ada compiler or run-time will use a mutex.

RTEMS has two capabilities which can assist in avoiding this problem. The Classic API Timer Manager allows the creation of Timer Service Routines which execute in the context of a task rather than the clock tick Interrupt Service Routine. Similarly, there is support for Interrupt Tasks which is a mechanism to defer the processing of the event from the hardware interrupt level to a thread.