Chapter 1. Overview
Table of Contents
- 1.1. Terminology
- 1.2. Why Configurability?
- 1.3. Approaches to Configurability
- 1.4. Degrees of Configurability
- 1.5. Warnings
eCos® was designed from the very beginning as a configurable component architecture. The core eCos system consists of a number of different components such as the kernel, the C library, an infrastructure package. Each of these provides a large number of configuration options, allowing application developers to build a system that matches the requirements of their particular application. To manage the potential complexity of multiple components and lots of configuration options, eCos comes with a component framework: a collection of tools specifically designed to support configuring multiple components. Furthermore this component framework is extensible, allowing additional components to be added to the system at any time.
The eCos component architecture involves a number of key concepts.
1.1.1. Component Framework
The phrase component framework is used to describe the collection of tools that allow users to configure a system and administer a component repository. This includes the ecosconfig command line tool, the graphical configuration tool, and the package administration tool. Both the command line and graphical tools are based on a single underlying library, the CDL library.
1.1.2. Configuration Option
The option is the basic unit of configurability. Typically each option
corresponds to a single choice that a user can make. For example there
is an option to control whether or not assertions are enabled, and the
kernel provides an option corresponding to the number of scheduling
priority levels in the system. Options can control very small amounts
of code such as whether or not the C library's
strtok gets inlined. They can also control quite
large amounts of code, for example whether or not the
printf supports floating point conversions.
Many options are straightforward, and the user only gets to choose whether the option is enabled or disabled. Some options are more complicated, for example the number of scheduling priority levels is a number that should be within a certain range. Options should always start off with a sensible default setting, so that it is not necessary for users to make hundreds of decisions before any work can start on developing the application. Once the application is running the various configuration options can be used to tune the system for the specific needs of the application.
The component framework allows for options that are not directly user-modifiable. Consider the case of processor endianness: some processors are always big-endian or always little-endian, while with other processors there is a choice. Depending on the user's choice of target hardware, endianness may or may not be user-modifiable.
A component is a unit of functionality such as a particular kernel scheduler or a device driver for a specific device. A component is also a configuration option in that users may want to enable or disable all the functionality in a component. For example, if a particular device on the target hardware is not going to be used by the application, directly or indirectly, then there is no point in having a device driver for it. Furthermore disabling the device driver should reduce the memory requirements for both code and data.
Components may contain further configuration options. In the case of a device driver, there may be options to control the exact behavior of that driver. These will of course be irrelevant if the driver as a whole is disabled. More generally options and components live in a hierarchy, where any component can contain options specific to that component and further sub-components. It is possible to view the entire eCos kernel as one big component, containing sub-components for scheduling, exception handling, synchronization primitives, and so on. The synchronization primitives component can contain further sub-components for mutexes, semaphores, condition variables, event flags, and so on. The mutex component can contain configuration options for issues like priority inversion support.
A package is a special type of component. Specifically, a package is the unit of distribution of components. It is possible to create a distribution file for a package containing all of the source code, header files, documentation, and other relevant files. This distribution file can then be installed using the appropriate tool. Afterwards it is possible to uninstall that package, or to install a later version. The core eCos distribution comes with a number of packages such as the kernel and the infrastructure. Other packages such as network stacks can come from various different sources and can be installed alongside the core distribution.
Packages can be enabled or disabled, but the user experience is a little bit different. Generally it makes no sense for the tools to load the details of every single package that has been installed. For example, if the target hardware uses an ARM processor then there is no point in loading the HAL packages for other architectures and displaying choices to the user which are not relevant. Therefore enabling a package means loading its configuration data into the appropriate tool, and disabling a package is an unload operation. In addition, packages are not just enabled or disabled: it is also possible to select the particular version of a package that should be used.
A configuration is a collection of user choices. The various
tools that make up the component framework deal with entire
configurations. Users can create a new configuration, output a
savefile (by default
ecos.ecc), manipulate a
configuration, and use a configuration to generate a build tree prior
to building eCos and any other packages that have been selected.
A configuration includes details such as which packages have been
selected, in addition to finer-grained information such as which
options in those packages have been enabled or disabled by the user.
The target is the specific piece of hardware on which the application is expected to run. This may be an off-the-shelf evaluation board, a piece of custom hardware intended for a specific application, or it could be something like a simulator. One of the steps when creating a new configuration is need to specify the target. The component framework will map this on to a set of packages that are used to populate the configuration, typically HAL and device driver packages, and in addition it may cause certain options to be changed from their default settings to something more appropriate for the specified target.
A template is a partial configuration, aimed at providing users with an appropriate starting point. eCos is shipped with a small number of templates, which correspond closely to common ways of using the system. There is a minimal template which provides very little functionality, just enough to bootstrap the hardware and then jump directly to application code. The default template adds additional functionality, for example it causes the kernel and C library packages to be loaded as well. The uitron template adds further functionality in the form of a µITRON compatibility layer. Creating a new configuration typically involves specifying a template as well as a target, resulting in a configuration that can be built and linked with the application code and that will run on the actual hardware. It is then possible to fine-tune configuration options to produce something that better matches the specific requirements of the application.
The component framework needs a certain amount of information about each option. For example it needs to know what the legal values are, what the default should be, where to find the on-line documentation if the user needs to consult that in order to make a decision, and so on. These are all properties of the option. Every option (including components and packages) consists of a name and a set of properties.
Choices must have consequences. For an eCos configuration the main
end product is a library that can be linked with application code, so
the consequences of a user choice must affect the build process. This
happens in two main ways. First, options can affect which files get
built and end up in the library. Second, details of the current option
settings get written into various configuration header files using C
#define directives, and package source
#include these configuration headers and
adapt accordingly. This allows options to affect a package at a very
fine grain, at the level of individual lines in a source file if
desired. There may be other consequences as well, for example there
are options to control the compiler flags that get used during the
Configuration choices are not independent. The C library can provide
thread-safe implementations of functions like
rand, but only if the kernel provides support for
per-thread data. This is a constraint: the C library option has a
requirement on the kernel. A typical configuration involves a
considerable number of constraints, of varying complexity: many
constraints are straightforward, option
B, or option
D. Other constraints can be more
complicated, for example option
E may require the
presence of a kernel scheduler but does not care whether it is the
bitmap scheduler, the mlqueue scheduler, or something else.
Another type of constraint involves the values that can be used for certain options. For example there is a kernel option related to the number of scheduling levels, and there is a legal values constraint on this option: specifying zero or a negative number for the number of scheduling levels makes no sense.
As the user manipulates options it is possible to end up with an invalid configuration, where one or more constraints are not satisfied. For example if kernel per-thread data is disabled but the C library's thread-safety options are left enabled then there are unsatisfied constraints, also known as conflicts. Such conflicts will be reported by the configuration tools. The presence of conflicts does not prevent users from attempting to build eCos, but the consequences are undefined: there may be compile-time failures, there may be link-time failures, the application may completely fail to run, or the application may run most of the time but once in a while there will be a strange failure… Typically users will want to resolve all conflicts before continuing.
To make things easier for the user, the configuration tools contain an inference engine. This can examine a conflict in a particular configuration and try to figure out some way of resolving the conflict. Depending on the particular tool being used, the inference engine may get invoked automatically at certain times or the user may need to invoke it explicitly. Also depending on the tool, the inference engine may apply any solutions it finds automatically or it may request user confirmation.
The configuration tools require information about the various options provided by each package, their consequences and constraints, and other properties such as the location of on-line documentation. This information has to be provided in the form of CDL scripts. CDL is short for Component Definition Language, and is specifically designed as a way of describing configuration options.
A typical package contains the following:
Some number of source files which will end up in a library. The application code will be linked with this library to produce an executable. Some source files may serve other purposes, for example to provide a linker script.
Exported header files which define the interface provided by the package.
On-line documentation, for example reference pages for each exported function.
Some number of test cases, shipped in source format, allowing users to check that the package is working as expected on their particular hardware and in their specific configuration.
One or more CDL scripts describing the package to the configuration system.
Not all packages need to contain all of these. For example some packages such as device drivers may not provide a new interface, instead they just provide another implementation of an existing interface. However all packages must contain a CDL script that describes the package to the configuration tools.
1.1.13. Component Repository
All eCos installations include a component repository. This is a directory structure where all the packages get installed. The component framework comes with an administration tool that allows new packages or new versions of a package to be installed, old packages to be removed, and so on. The component repository includes a simple database, maintained by the administration tool, which contains details of the various packages.
Generally application developers do not need to modify anything inside the component repository, except by means of the administration tool. Instead their work involves separate build and install trees. This allows the component repository to be treated as a read-only resource that can be shared by multiple projects and multiple users. Component writers modifying one of the packages do need to manipulate files in the component repository.
1.2. Why Configurability?
The eCos component framework places a great deal of emphasis on configurability. The fundamental goal is to allow large parts of embedded applications to be constructed from re-usable software components, which does not a priori require that those components be highly configurable. However embedded application development often involves some serious constraints.
Many embedded applications have to work with very little memory, to keep down manufacturing costs. The final application image that will get blown into EPROM's or used to manufacture ROMs should contain only the code that is absolutely necessary for the application to work, and nothing else. If a few tens of kilobytes are added unnecessarily to a typical desktop application then this is regrettable, but is quite likely to go unnoticed. If an embedded application does not fit on the target hardware then the problem is much more serious. The component framework must allow users to configure the components so that any unnecessary functionality gets removed.
Many embedded applications need deterministic behavior so that they can meet real-time requirements. Such deterministic behavior can often be provided, but at a cost in terms of code size, slower algorithms, and so on. Other applications have no such real-time requirements, or only for a small part of the overall system, and the bulk of the system should not suffer any penalties. Again the component framework must allow the users control over the timing behavior of components.
Embedded systems tend to be difficult to debug. Even when it is possible to get information out of the target hardware by means other than flashing an LED, the more interesting debugging problems are likely to be timing-related and hence very hard to reproduce and track down. The re-usable components can provide debugging assistance in various ways. They can provide functionality that can be exploited by source level debuggers such as gdb, for example per-thread debugging information. They can also contain various assertions so that problems can be detected early on, tracing mechanisms to figure out what happened before the assertion failure, and so on. Of course all of these involve overheads, especially code size, and affect the timing. Allowing users to control which debugging features are enabled for any given application build is very desirable.
However, although it is desirable for re-usable components to provide appropriate configuration options this is not required. It is possible to produce a package which does not provide a single configuration option — although the user still gets to choose whether or not to use the package. In such cases it is still necessary to provide a minimal CDL script, but its main purpose would be to integrate the package with the component framework's build system.
1.3. Approaches to Configurability
The purpose of configurability is to control the behavior of components. A scheduler component may or may not support time slicing; it may or may not support multiple priorities; it may or may not perform error checking on arguments passed to the scheduler routines. In the context of a desktop application a button widget may contain some text or it may contain a picture; the text may be displayed in a variety of fonts; the foreground and background color may vary. When an application uses a component there must be some way of specifying the desired behavior. The component writer has no way of knowing in advance exactly how a particular component will end up being used.
One way to control the behavior is at run time. The application creates an instance of a button object, and then instructs this object to display either text or a picture. No special effort by the application developer is required, since a button can always support all desired behavior. There is of course a major disadvantage in terms of the size of the final application image: the code that gets linked with the application has to provide support for all possible behavior, even if the application does not require it.
Another approach is to control the behavior at link-time, typically
by using inheritance in an object-oriented language. The button
library provides an abstract base class
and derived classes
PictureButton. If an application only uses text
buttons then it will only create objects of type
TextButton, and the code for the
PictureButton class does not get used. In
many cases this approach works rather well and reduces the final image
size, but there are limitations. The main one is that you can only
have so many derived classes before the system gets unmanageable: a
is not particularly sensible as far as most application developers are
The eCos component framework allows the behavior of components to be controlled at an even earlier time: when the component source code gets compiled and turned into a library. The button component could provide options, for example an option that only text buttons need to be supported. The component gets built and becomes part of a library intended specifically for the application, and the library will contain only the code that is required by this application and nothing else. A different application with different requirements would need its own version of the library, configured separately.
In theory compile-time configurability should give the best possible
results in terms of code size, because it allows code to be controlled
at the individual statement level rather than at the function or
object level. Consider an example more closely related to embedded
systems, a package to support multi-threading. A standard routine
within such a package allows applications to kill threads
asynchronously: the POSIX routine for this is
pthread_cancel; the equivalent routine in µITRON
ter_tsk. These routines themselves tend to
involve a significant amount of code, but that is not the real
problem: other parts of the system require extra code and data for the
kill routine to be able to function correctly. For example if a thread
is blocked while waiting on a mutex and is killed off by another
thread then the kill operation may have to do two things: remove the
thread from the mutex's queue of waiting threads; and undo the
effects, if any, of priority inheritance. The implementation requires
extra fields in the thread data structure so that the kill routine
knows about the thread's current state, and extra code in the mutex
routines to fill in and clear these extra fields correctly.
Most embedded applications do not require the ability to kill off a thread asynchronously, and hence the kill routine will not get linked into the final application image. Without compile-time configurability this would still mean that the mutex code and similar parts of the system contain code and data that serve no useful purpose in this application. The eCos approach allows the user to select that the thread kill functionality is not required, and all the components can adapt to this at compile-time. For example the code in the mutex lock routine contains statements to support the killing of threads, but these statements will only get compiled in if that functionality is required. The overall result is that the final application image contains only the code and data that is really needed for the application to work, and nothing else.
Of course there are complications. To return to the button example, the application code might only use text buttons directly, but it might also use some higher-level widget such as a file selector and this file selector might require buttons with pictures. Therefore the button code must still be compiled to support pictures as well as text. The configuration tools must be aware of the dependencies between components and ensure that the internal constraints are met, as well as the external requirements of the application code. An area of particular concern is conflicting requirements: a button component might be written in such a way that it can only support either text buttons or picture buttons, but not both in one application; this would represent a weakness in the component itself rather than in the component framework as a whole.
Compile-time configurability is not intended to replace the other
approaches but rather to complement them. There will be times when
run-time selection of behavior is desirable: for example an
application may need to be able to change the baud rate of a serial
line, and the system must then provide a way of doing this at
run-time. There will also be times when link-time selection is
desirable: for example a C library might provide two different random
lrand48; these do not affect other code so there
is no good reason for the C library component not to provide both of
these, and allow the application code to use none, one, or both of
them as appropriate; any unused functions will just get eliminated at
link-time. Compile-time selection of behavior is another option, and
it can be the most powerful one of the three and the best suited to
embedded systems development.
1.4. Degrees of Configurability
Components can support configurability in varying degrees. It is not necessary to have any configuration options at all, and the only user choice is whether or not to load a particular package. Alternatively it is possible to implement highly-configurable code. As an example consider a typical facility that is provided by many real-time kernels, mutex locks. The possible configuration options include:
If no part of the application and no other component requires mutexes then there is no point in having the mutex code compiled into a library at all. This saves having to compile the code. In addition there will never be any need for the user to configure the detailed behavior of mutexes. Therefore the presence of mutexes is a configuration option in itself.
Even if the application does make use of mutexes directly or indirectly, this does not mean that all mutex functions have to be included. The minimum functionality consists of lock and unlock functions. However there are variants of the locking primitive such as try-lock and try-with-timeout which may or may not be needed.
Generally it will be harmless to compile the try-lock function even if it is not actually required, because the function will get eliminated at link-time. Some users might take the view that the try-lock function should never get compiled in unless it is actually needed, to reduce compile-time and disk usage. Other users might argue that there are very few valid uses for a try-lock function and it should not be compiled by default to discourage incorrect uses. The presence of a try-lock function is a possible configuration option, although it may be sensible to default it to true.
The try-with-timeout variant is more complicated because it adds a dependency: the mutex code will now rely on some other component to provide a timer facility. To make things worse the presence of this timer might impact other components, for example it may now be necessary to guard against timer interrupts, and thus have an insidious effect on code size. The presence of a lock-with-timeout function is clearly a sensible configuration option, but the default value is less obvious. If the option is enabled by default then the final application image may end up with code that is not actually essential. If the option is disabled by default then users will have to enable the option somehow in order to use the function, implying more effort on the part of the user. One possible approach is to calculate the default value based on whether or not a timer component is present anyway.
The application may or may not require the ability to create and destroy mutexes dynamically. For most embedded systems it is both less error-prone and more efficient to create objects like mutexes statically. Dynamic creation of mutexes can be implemented using a pre-allocated pool of mutex objects, involving some extra code to manipulate the pool and an additional configuration option to define the size of the pool. Alternatively it can be implemented using a general-purpose memory allocator, involving quite a lot of extra code and configuration options. However this general-purpose memory allocator may be present anyway to support the application itself or some other component. The ability to create and destroy mutexes dynamically is a configuration option, and there may not be a sensible default that is appropriate for all applications.
An important issue for mutex locks is the handling of priority inversion, where a high priority thread is prevented from running because it needs a lock owned by a lower priority thread. This is only an issue if there is a scheduler with multiple priorities: some systems may need multi-threading and hence synchronization primitives, but a single priority level may suffice. If priority inversion is a theoretical possibility then the application developer may still want to ignore it because the application has been designed such that the problem cannot arise in practice. Alternatively the developer may want some sort of exception raised if priority inversion does occur, because it should not happen but there may still be bugs in the code. If priority inversion can occur legally then there are three main ways of handling it: priority ceilings, priority inheritance, and ignoring the problem. Priority ceilings require little code but extra effort on the part of the application developer. Priority inheritance requires more code but is automatic. Ignoring priority inversion may or may not be acceptable, depending on the application and exactly when priority inversion can occur. Some of these choices involve additional configuration options, for example there are different ways of raising an exception, and priority inheritance may or may not be applied recursively.
As a further complication some mutexes may be hidden inside a component rather than being an explicit part of the application. For example, if the C library is configured to provide a
malloccall then there may be an associated mutex to make the function automatically thread-safe, with no need for external locking. In such cases the memory allocation component of the C library can impose a constraint on the kernel, requiring that mutexes be provided. If the user attempts to disable mutexes anyway then the configuration tools will report a conflict.
The mutex code should contain some general debugging code such as assertions and tracing. Usually such debug support will be enabled or disabled at a coarse level such as the entire system or everything inside the kernel, but sometimes it will be desirable to enable the support more selectively. One reason would be memory requirements: the target may not have enough memory to hold the system if all debugging is enabled. Another reason is if most of the system is working but there are a few problems still to resolved; enabling debugging in the entire system might change the system's timing behavior too much, but enabling some debug options selectively can still be useful. There should be configuration options to allow specific types of debugging to be enabled at a fine-grain, but with default settings inherited from an enclosing component or from global settings.
The mutex code may contain specialized code to interact with a debugging tool running on the host. It should be possible to enable or disable this debugging code, and there may be additional configuration options controlling the detailed behavior.
Altogether there may be something like ten to twenty configuration options that are specific to the mutex code. There may be a similar number of additional options related to assertions and other debug facilities. All of the options should have sensible default values, possibly fixed, possibly calculated depending on what is happening elsewhere in the configuration. For example the default setting for an assertion option should generally inherit from a kernel-wide assertion control option, which in turn inherits from a global option. This allows users to enable or disable assertions globally or at a more fine-grained level, as desired.
Different components may be configurable to different degrees, ranging from no options at all to the fine-grained configurability of the above mutex example (or possibly even further). It is up to component writers to decide what options should be provided and how best to serve the needs of application developers who want to use that component.
Large parts of eCos were developed concurrently with the development of the configuration technology, or in some cases before design work on that technology was complete. As a consequence the various eCos packages often make only limited use of the available functionality. This situation is expected to change over time. It does mean that many of the descriptions in this guide will not correspond exactly to how the eCos packages work right now, but rather to how they could work. Some of the more extreme discrepancies such as the location of on-line documentation in the component repository will be mentioned in the appropriate places in the guide.
A consequence of this is that developers of new components can look at existing CDL scripts for examples, and discover discrepancies between what is recommended in this guide and what actually happens at present. In such cases this guide should be treated as authoritative.
It is also worth noting that the current component framework is not finished. Various parts of this guide will refer to possible changes and enhancements in future versions. Examining the source code of the configuration tools may reveal hints about other likely developments, and there are many more possible enhancements which only exist at a conceptual level right now.
|2021-04-29||Open Publication License|