'conf' is part of the zephyr SDK or need to be built and installed in
the path. We now using python for Kconfig processing, so this is not a
strict requirement and we should be able to build without it. If the
binary is not found, just go on with our business.
Signed-off-by: Anas Nashif <anas.nashif@intel.com>
Due to a chicken-and-egg problem, it's not possible to call
board_runner_args() directly from an application CMakeLists.txt and
have it take effect:
- if the board_runner_args() call appears before including the
boilerplate file, then the function is undefined and an error
occurs.
- if the call appears after including the boilerplate file, then
board_finalize_runner_args() has already been called by the
relevant runner.board.cmake file, so it won't take effect.
Similarly to the way set_conf_file() works, allow applications to
declare additional runner args by checking if they have provided a
macro named app_set_runner_args(), which contains calls to
board_runner_args(). Using a macro means that we can invoke it with
the definition of board_runner_args() in scope.
Signed-off-by: Marti Bolivar <marti@opensourcefoundries.com>
Use two different variables for the distinct values LIBGCC_FILE_NAME
and LIBGCC_DIR and assert that each of them exists.
Signed-off-by: Sebastian Bøe <sebastian.boe@nordicsemi.no>
CROSS_COMPILE is a KBuild feature that was dropped during the CMake
migration. It is now re-introduced. Documentation for it is still
lacking, but at least it now behaves as expected.
Signed-off-by: Sebastian Bøe <sebastian.boe@nordicsemi.no>
Sync the outdated comments with the code. E.g. there is no
toolchain-clang.cmake, but there is a toolchain-host-gcc.cmake.
Signed-off-by: Sebastian Bøe <sebastian.boe@nordicsemi.no>
When Kconfiglib was introduced it caused a significant performance
issue. This patch uses pruning to mitigate the performance issue.
The pruning exploits the fact that before the Kconfig database is
parsed we already know what ARCH and BOARD has been selected. So in
theory we could prune away all Kconfig sources that are not related to
the current ARCH or BOARD. In practice, it is only the Kconfig sources
in zephyr/arch/$ARCH and zephyr/board/$ARCH/ that are easy to prune.
Still, that is quite a few Kconfig sources. For qemu_x86 this patch
reduced the number of parsed Kconfig source files from 632 to
272. This pruning resulted in a incremental reconfiguration (time
cmake ..) speedup of 21% (0.56s to 0.46) and a clean build speedup of
4% (Using board qemu_x86 and sample hello_world).
Furthermore, it should be easier to maintain ARCH's and BOARD's
out-of-tree since the user now has a mechanism to redirect where
Kconfig sources are found. But this has not been explored.
Signed-off-by: Sebastian Bøe <sebastian.boe@nordicsemi.no>
Kconfiglib does not support merging fragments without parsing the
entire Kconfig "database". For Zephyr this means we no longer get a
performance gain from splitting up the fragment-merging and the
.config generation.
This patch removes the seperated merge step and it's associated
caching mechanism. Now we do a full Kconfig execution on every
reconfiguration.
Signed-off-by: Sebastian Bøe <sebastian.boe@nordicsemi.no>
We have been using a fork of the Linux kernel's Kconfig system to
configure the Zephyr tree. The issue is that this is a native tool
written in C that is not easy to compile for Windows. This patch
replaces the use of the conf executable with kconfig.py, a script that
uses Kconfiglib to generate the .config and autoconf.h files required to
compile Zephyr.
Signed-off-by: Carles Cufi <carles.cufi@nordicsemi.no>
Instead of hard-coding the path use find_program(), since that allows
CMake to sort out platform-specific details such as .exe extensions on
Windows.
Signed-off-by: Carles Cufi <carles.cufi@nordicsemi.no>
Instead of accessing the environment variable ZEPHYR_BASE every time we
require accessing the source code root, use an intermediate variable
that has OS path separators correctly set to '/' to avoid issues on
Windows.
Note: This removes the ZEPHYR_SOURCE_DIR CMake variable. External
applications using that will need to change to use the new ZEPHYR_BASE
variable.
Signed-off-by: Carles Cufi <carles.cufi@nordicsemi.no>
Windows users have on multiple occasions cloned Zephyr using a Windows
git client. It seems that the windows git client defaults to
converting line endings from LF to CRLF when cloning repo's. This
breaks at least one of Zephyr's tools (Kconfig).
This patch introduces a sanity check of the environment for MSYS
users.
Signed-off-by: Sebastian Bøe <sebastian.boe@nordicsemi.no>
Now that MSYS2 ships with CMake 3.9.6 there's no need anymore to
downgrade the minimum required CMake version for it.
Signed-off-by: Carles Cufi <carles.cufi@nordicsemi.no>
Depending on a path inside the Zephyr tree to determine if we are a test
does not scale. Also some samples were marked as TEST while they are
not, just to get some options defined for tests.
Idenitfying a test will be addressed in another patch introducing
CONFIG_TEST.
Signed-off-by: Anas Nashif <anas.nashif@intel.com>
Usually the zephyr_get_* API returns values prefixed with tokens like
-I -system, -D. But sometimes we need the values without these
prefixes, so we introduce the optional SKIP_PREFIX argument.
Signed-off-by: Sebastian Bøe <sebastian.boe@nordicsemi.no>
Introduce Zephyr interface libraries to fix#5351.
From the documentation:
A Zephyr interface library is a thin wrapper over a CMake INTERFACE
library. The most important responsibility of this abstraction is to
ensure that when a user KConfig-enables a library then the header
files of this library will be accessible to the 'app' library.
This is done because when a user uses Kconfig to enable a library he
expects to be able to include it's header files and call it's
functions out-of-the box.
A Zephyr interface library should be used when there exists some
build information (include directories, defines, compiler flags,
etc.) that should be applied to a set of Zephyr libraries and 'app'
might be one of these libraries.
Zephyr libraries must explicitly call
zephyr_library_link_libraries(<interface_library>) to use this build
information. 'app' is treated as a special case for usability
reasons; a Kconfig option (CONFIG_APP_LINK_WITH_<interface_library>)
should exist for each interface_library and will determine if 'app'
links with the interface_library.
This API has a constructor like the zephyr_library API has, but it
Signed-off-by: Sebastian Bøe <sebastian.boe@nordicsemi.no>
I don't know why we have been passing "-serial none" to qemu but it
does not seem to be doing anything useful. So I propose we remove it.
According to the qemu documenation it should be disabling all serial
ports;
"-serial dev
Redirect the virtual serial port to host character device dev. The
default device is vc in graphical mode and stdio in non graphical
mode.
This option can be used several times to simulate up to 4 serial
ports.
Use -serial none to disable all serial ports."
But when we use "-serial none", we always combine it with "-serial
pipe", or "-serial unix", to redirect the serial port, so clearly it
is not disabling all serial ports as it attempting to do.
"-serial none" was first introduced to Zephyr in this commit:
0b54f984b8
As far I can tell, pinging an echo_server works fine without it, so I
suspect that our second -serial flag is overriding it and that it has
no effect.
Signed-off-by: Sebastian Bøe <sebastian.boe@nordicsemi.no>
Using the '\$' technique to accept make variables corrupts Ninja build
systems, so avoid using it when generating Ninja files. Not using it
with Ninja means we need to come up with some other mechanism to
support build-time configuration of the QEMU CLI with Ninja.
Signed-off-by: Sebastian Bøe <sebastian.boe@nordicsemi.no>
The generate_inc_file_for_target() extension is a useful wrapper
around the more low-level generate_inc_file(). It takes care of
ensuring that the generated file correctly depends on it's source.
To do so it needs to name and define a custom_target, the name must be
unique, so it was constructed based on the path to the generated file,
but this caused multiple issues.
The intention was that the path of the file would help identify the
target, in hindsight, identifying the target has proven to not be
important.
This patch fixes#5466 by using a random string to construct the
target name instead of the path to the generated file.
Signed-off-by: Sebastian Bøe <sebastian.boe@nordicsemi.no>
target_ld_options() was only testing whether a flag could be passed to
the compiler driver when it was compiling an object file. Not whether
a flag could be passed to the compiler driver when it was linking an
elf file.
For most flags, these tests would have the same result, but it does
not for flags like -Wl,--print-memory-usage or -Wl,notaflag.
This patch fixes#5488 by re-using the method from #5459.
Signed-off-by: Sebastian Bøe <sebastian.boe@nordicsemi.no>
Added posix to the list of arch's with boards. Usage will now show
native_posix as one of the available boards.
Signed-off-by: Sebastian Bøe <sebastian.boe@nordicsemi.no>
To indicate the generated binary is executable on the host, add .exe
extension to the generated ELF file.
Signed-off-by: Anas Nashif <anas.nashif@intel.com>
All runner logic was implemented in qemu.cmake, remove the generic stuff
and make qemu.cmake qemu specific.
Signed-off-by: Anas Nashif <anas.nashif@intel.com>
This adds support of 'make run' to the native port allowing us to run
applications natively on the host instead of qemu.
Signed-off-by: Anas Nashif <anas.nashif@intel.com>
Right now we are hardcoded to only qemu, with the native port, we make
this more generic and support this in a plugin mode where a running has
its own cmake definitons implementing the various targets.
Signed-off-by: Anas Nashif <anas.nashif@intel.com>
Signed-off-by: Alberto Escolar Piedras <alpi@oticon.com>
A new arch (posix) which relies on pthreads to emulate the context
switching
A new soc for it (inf_clock) which emulates a CPU running at an
infinely high clock (so when the CPU is awaken it runs till completion
in 0 time)
A new board, which provides a trivial system tick timer and
irq generation.
Origin: Original
Fixes#1891
Signed-off-by: Alberto Escolar Piedras <alpi@oticon.com>
Signed-off-by: Anas Nashif <anas.nashif@intel.com>
To avoid warnings from cmake such as:
'path' is reserved or not valid for certain CMake features, such as
generator expressions, and may result in undefined behavior.
This can be seen in Jenkins for example where @ is used when multiple
jobs are run on the same machine.
Signed-off-by: Anas Nashif <anas.nashif@intel.com>
The properties definitions were not doing much because they were
defined after they were used. They were actually overriding the true
properties with "".
Moving them earlier ensures that the properties behave as expected, as
documented global mutable variables.
Signed-off-by: Sebastian Bøe <sebastian.boe@nordicsemi.no>
The various runners (flash/debug scripts) use environment variables to
take arguments. This is legacy behavior which is not desirable.
Use command line arguments instead.
Note: this leaves more general environment variables with publicly
documented behavior in place for now, for compatibility, e.g.:
ZEPHYR_FLASH_OVER_DFU, OPENSDA_FW, ESP_IDF_PATH, PYOCD_DAPARG
For example, when using dfu-util to flash arduino_101, instead of
setting DFUUTIL_PID, DFUUTIL_ALT, and DFUUTIL_IMG environment
variables, have the script invocation look like this:
python3 .../zephyr_flash_debug.py dfu-util flash \
[common arguments omitted] \
--pid=8087:0aba --alt=x86_app \
--img=.../build/zephyr/zephyr.bin
Make similar changes for other runners (openocd, etc.) and
targets (debug, debugserver).
To implement this in the scripts:
- have the individual scripts/support/runner/some-runner.py files
register their own command line arguments
- teach them to construct instances from arguments, not the
environment
- have zephyr_flash_debug.py request runners to register command
line argument parsers, and handle arguments
In the build system:
- add a new board_runner_args() extension function that board.cmake
files can use to add to the zephyr_flash_debug.py command line
- adjust cmake/flash/CMakeLists.txt to invoke with arguments
- add new helper include files for each runner (like
boards/common/dfu-util.board.cmake, etc.), which add default
options as needed and then add on overrides from
board_runner_args() calls
- update board.cmake files to use the new includes and extension
This implied some tweaking when using openocd to make the CMake string
escaping and unescaping work properly.
Signed-off-by: Marti Bolivar <marti@opensourcefoundries.com>
I keep tripping over not being able to use set_ifndef to set a value
to a list, like this:
set_ifndef(VARIABLE value1 value2 value3...)
Allow that to work without changing its existing behavior when called
with one argument.
Signed-off-by: Marti Bolivar <marti@opensourcefoundries.com>
The Python-based runners have replaced the old shell scripts. Refactor
the build system accordingly:
- FLASH_SCRIPT is now BOARD_FLASH_RUNNER
- DEBUG_SCRIPT is now BOARD_DEBUG_RUNNER
The values, rather than being the names of files, are now the names of
runners in scripts/support/runner. They are still short, descriptive
names like "openocd", "jlink", "em-starterkit", etc.
Adjust the zephyr_flash_debug.py call and runner internals
accordingly. Have each runner class report a name and the commands it
can handle. This lets us move some boilerplate from each do_run()
method into the common run() routine, and enables further improvements
in future patches.
The handles_command() method is temporary, and will be replaced by a
more general mechanism for describing runner capabilities in a
subsequent patch. The initial use case for extending this is to add
device tree awareness to the runners.
To try to avoid user confusion, abort the configuration if an
xxx_SCRIPT is defined.
Signed-off-by: Marti Bolivar <marti@opensourcefoundries.com>
Support installing toolchains in "Program Files", or more generally,
in paths that contain whitespace. This is a common bug for new users
to discover.
Signed-off-by: Sebastian Bøe <sebastian.boe@nordicsemi.no>
size_report was using the environment variables NM and OBJDUMP to find
it's toolchain dependencies. It is not clear how well this will work
on different platforms; OSX, Windows. So we now pass the paths of
these dependencies from the build system to the script.
PS: This ensures that size_report uses the cross-compiler's GNU bin
tool instead of the host GNU bin tools. This is presumably beneficial
as it has been required for other GNU bin tools like GDB.
Signed-off-by: Sebastian Bøe <sebastian.boe@nordicsemi.no>
This is needed by platforms like jailhouse.
Signed-off-by: Sebastian Bøe <sebastian.boe@nordicsemi.no>
Signed-off-by: Gustavo Lima Chaves <gustavo.lima.chaves@intel.com>
We don't want to pollute core-infrastructure with board-specific code,
so we move the qemu discovery to after board.cmake has been executed,
and we introduce the overridable QEMU_binary_suffix CMake variable.
Signed-off-by: Sebastian Bøe <sebastian.boe@nordicsemi.no>
In that case, QEMU is emulating the base (root system) that will be
partitioned to receive Zephyr as one of the cells/guests, not the direct
system Zephyr will run on. We need this one more level of indirection,
otherwise it's impossible to attest Zephyr is running in a virtualized
environment in QEMU.
Fix QEMU binary to point to the x86_64 version, then, once Jailhouse is
targeted at that architecture.
Finally, don't pass QEMU the -kernel flag in this case, since Zephyr has
to loaded manually, from inside QEMU, for Jailhouse's make run target.
Signed-off-by: Gustavo Lima Chaves <gustavo.lima.chaves@intel.com>
Print the usage when an invalid board is given, this somewhat
allievates the catch-22 where you need to run CMake to know what
boards exist, but you need a board to run CMake.
Signed-off-by: Sebastian Bøe <sebastian.boe@nordicsemi.no>
There is a catch-22 in that you need to run CMake to get a build
system that can show you usage, but you need to know the usage to be
able to get a build system.
This assert could be used to improve the usability somewhat. When
invalid usage is detected it can be used to print the usage.
Signed-off-by: Sebastian Bøe <sebastian.boe@nordicsemi.no>
I can't explain why "make run" worked before, but after this patch it
should definitely work.
Signed-off-by: Sebastian Bøe <sebastian.boe@nordicsemi.no>
When exporting flags to an external build system we need to deal with
the fact that we sometimes use generator expressions. Specifically, we
use generator expressions that look like this:
$<$<COMPILE_LANGUAGE:CXX>:-fno-exceptions>
This patch replaces the old API with a new one where users can ask for
compile options for specific languages, like this:
zephyr_get_compile_options_for_lang_as_string(CXX x)
The existing API would have either crashed or silently omitted flags
when a COMPILE_LANG generator expression was present.
Signed-off-by: Sebastian Bøe <sebastian.boe@nordicsemi.no>
It is supported to add give extra flags to the linker from the
commandline like this:
cmake -DEXTRA_LDFLAGS=-Lmy_dir path
But unfortunately this was broken during the CMake
migration. Interestingly, the reason that it was broken is that KBuild
was also partially broken. KBuild would pass on EXTRA_LDFLAGS when
object files were linked together into built-in.o files, but it would
not use EXTRA_LDFLAGS for the final link into an elf file.
This patch fixes EXTRA_LDFLAGS.
Signed-off-by: Sebastian Bøe <sebastian.boe@nordicsemi.no>
This commit fixes
https://github.com/zephyrproject-rtos/zephyr/issues/5008.
It does so by splitting up gen_syscalls.py into two scripts with a
json metadata file to communicate syscall metadata between them. The
parsing script parses header files from include/ and writes syscall
metadata to a file if the contents changed. The generation script
reads from the json file and generates syscall code.
The build system DAG now looks like this:
always_rebuild -> json -> syscalls -> offset.o
The script for generating json will do so only if the content changes,
this ensures that the entire DAG does not always do a full rebuild.
Signed-off-by: Sebastian Bøe <sebastian.boe@nordicsemi.no>
This fixes https://github.com/zephyrproject-rtos/zephyr/issues/5186
The script that generates syscall_macros.h is moved from Configuration
time to build time. This allows us to express to the build system that
syscall_macros.h depends on the script that generates it.
Signed-off-by: Sebastian Bøe <sebastian.boe@nordicsemi.no>
Support adding options without checking them with the compiler. In some
cases the simple check routine fails due to missing symbols.
Signed-off-by: Anas Nashif <anas.nashif@intel.com>
Both options were originally parameters to *make*, not cmake, so people
can augment QEMU ibocation from run to run, not from complete rebuild
to complete rebuild. Make them such again.
However, just in case, ability to set "default" value for those options
on CMake level is preserved.
Signed-off-by: Paul Sokolovsky <paul.sokolovsky@linaro.org>
