![]() Recent doxygen versions generate references for constants that are named in the text. Since we do not produce a constant table in the manual (neither markdown nor tex), these references point to nowhere and produce a KeyError in the ref_dict on lookup. The corresponding xml nodes are of this form: <ref kindref="member" refid="include_2sel4_2constants_8h_1a9a3...">seL4_TCBFlag</ref> Ignore such dangling references and return the content (seL4_TCBFlag in the example) instead. Run text escape on the content to cover underscores etc in LaTeX. Does not produce a warning since we expect this to be normal behaviour. Signed-off-by: Gerwin Klein <gerwin.klein@proofcraft.systems> |
||
---|---|---|
.. | ||
figs | ||
logos | ||
parts | ||
tools | ||
Doxyfile | ||
Makefile | ||
README.md | ||
VERSION | ||
export.bst | ||
extra.bib | ||
manual.tex | ||
references.bib | ||
sel4.sty |
README.md
seL4 Reference Manual
Build
To build a PDF use
make
You will need recent version of LaTeX
and doxygen
as well as the seL4 python
dependencies.
The main source file is manual.tex
, most of the text is in parts/
,
and most of the API reference is generated with doxygen
.
API Generation
The documentation of the seL4 API is automatically generated from comments in the source code. This section documents the process.
Types of API
seL4 has two types of API, system calls and object invocations.
- Top-level system calls are generally for message passing, for example
Call
,Send
, andRecv
. In addition there are also debugging and benchmarking system calls which can be enabled with a build flag. - Object invocations are regular message-passing system calls, usually
Call
, where the recipient is effectively the kernel itself and the message encodes some operation on a kernel object. Some examples areTCB_Resume
andCNode_Copy
. Some kernel objects and object invocations are specific to a particular processor architecture, for instanceX86_Page_Map
, andARM_VCPU_InjectIrq
.
Almost all system calls, including Call
, Send
, and Recv
, are in fact
object invocations under the hood, for instance on endpoint or notification
objects -- they just provide a top-level API for message passing. The only
"true" system calls are Yield
, debug and benchmark operations.
The API documentation in the manual is divided into the following hierarchy:
- System Calls
- General System Calls
- Debugging System Calls
- Benchmarking System Calls
- Architecture-Independent Object Methods
- x86-Specific Object Methods
- General x86 Object Methods
- IA32 Object Methods
- x86_64 Object Methods
- ARM-Specific Object Methods
- General ARM Object Methods
- aarch32 Object Methods
- aarch64 Object Methods
- RISC-V-Specific Object Methods
The process of generating API docs is different between System Calls and Object Invocations, though each process has some parts in common.
Common
Approach
Documentation for each seL4 API is written in the form of doxygen comments in C header files. The rest of the seL4 manual is written in LaTeX. To generate API documentation for the manual, we generate LaTeX files from doxygen comments which are then included in the manual.
Rather than using doxygen's LaTeX output directly, we use doxygen to generate XML files. A custom script then parses the XML and produces the final LaTeX output. This is because we already have an established style for API documentation in our manual, and it was easier to generate LaTeX in this style ourselves from some simple (ie. easy to parse) intermediate format (ie. XML) rather than try to coerce doxygen into generating perfectly-styled LaTeX.
The script that translates doxygen-generated xml into LaTeX is in: manual/tools/parse_doxygen_xml.py.
Custom Notation in Doxygen Comments
Some parts of the API documentation reference other parts of the manual. Additionally, there are some custom formatting rules we'd like to apply to the API docs in the manual that aren't understood by doxygen. To achieve both these goals, we introduce some additional XML tags which we explicitly add to doxygen comments inside @xmlonly ... @endxmlonly blocks.
Here's a description of all the custom tags:
<!--
Introduces documentation for a new function.
The title of the section documenting the function will be NAME.
Other parts of the manual can refer to this function's documentation with \autoref{sec:LABEL}
-->
<manual name="NAME" label="LABEL"/>
<!-- Translated to the latex \autoref{sec:SEC} -->
<autoref sec="SEC"/>
<!-- Translated to the latex \ref{sec:SEC} -->
<shortref sec="SEC"/>
<!-- Translated to the latex \errorenumdesc, a custom command defined in manual/parts/api.tex <https://github.com/seL4/seL4/blob/master/manual/parts/api.tex> -->
<errorenumdesc/>
<!-- Translated to the latex \obj{NAME}, a custom command defined in manual/manual.tex <https://github.com/seL4/seL4/blob/master/manual/manual.tex> -->
<obj name="NAME"/>
<!-- Translated to the latex \texttt{TEXT} -->
<texttt text="TEXT"/>
Note that these must appear within @xmlonly
... @endxmlonly
blocks in
order to function.
Required Documentation
Each function in the API must have the following documentation:
- a
@xmlonly <manual name="..." label=".../> @endxmlonly
tag with thename
for use in the manual's text, and alabel
for creating references within the manual - a
@brief
description - a detailed description
- a
@param
description of each argument - a
@return
description of the return value, unless the function isvoid
Detecting Missing Documentation
If a required part of a function's documentation is empty, the translation
script will insert the LaTeX command \todo
, defined in
manual/parts/api.tex.
It generates the text "TODO" to help readers of the manual identify which parts
of the API are undocumented.
Generated LaTeX Files
The LaTeX files generated by the translation script are placed in
manual/generated
. Doxygen-generated XML files are placed in
manual/doxygen-output/xml
.
Each API is documented in a separate file. The separation of APIs is implemented
using doxygen's @defgroup
directive. Specifically, each API is defined in a
separate named group. In the doxygen-generated XML, each group's documentation
is in a separate file. The XML-to-LaTeX script then transforms each relevant XML
file into the corresponding LaTeX file.
Doxygen Configuration
The correct behaviour of the manual build system depends on a specific doxygen
configuration. A Doxyfile
containing this configuration is checked into the
kernel repo at
manual/Doxyfile
System Calls
A prototype for each system call is declared in libsel4/include/sel4/syscalls.h. Each function is documented with a comment of the form described above.
Object Invocations
Stub Generation
These are more complicated, as the C source code implementing the user-level object invocations functions is generated from interface descriptions in XML. The following XML files contain object invocation interface descriptions:
- libsel4/include/interfaces/object-api.xml
- libsel4/arch_include/x86/interfaces/object-api-arch.xml
- libsel4/arch_include/arm/interfaces/object-api-arch.xml.xml
- libsel4/sel4_arch_include/ia32/interfaces/object-api-sel4-arch.xml
- libsel4/sel4_arch_include/x86_64/interfaces/object-api-sel4-arch.xml
- libsel4/sel4_arch_include/aarch32/interfaces/object-api-sel4-arch.xml
- libsel4/sel4_arch_include/aarch64/interfaces/object-api-sel4-arch.xml
There is a script in the sel4 repo for generating C header files from a given interface description: libsel4/tools/syscall_stub_gen.py. Note that despite its name, the script generates object invocation stubs - not syscall stubs.
Documentation
Object invocation API documentation is located inline with the interface descriptions in the XML files listed above. The XML language with which the interfaces are described contains tags for documenting functions. For the purpose of validation, the XML schema defining this language is in: libsel4/tools/sel4_idl.xsd. This is a superset of the XML tags used in doxygen comments. The doxygen comment tags described above have the same meaning in the interface description files. Additional tags are used for the description and documentation of object invocation interfaces.
When C code is generated from an interface description file, the documentation inlined in that file is converted into doxygen comments of the form described above.