* Update nvidia-container-toolkit, libnvidia-container to v1.16.2
* Refresh nvidia-container-toolkit alt-roots patch
* Drop obsolete backported patch and config.toml template file
* Use nvidia-modprobe patch in the libnvidia-container sources
instead of maintaining our own copy
* Other minor cleanup of obsolete settings in libnvidia-container
recipe
* Fix PACKAGE_ARCH of libnvidia-container for tegra platforms
* Add missing BSD-3-Clause license to nvidia-container-toolkit recipe
* Update license checksums in nvidia-container-toolkit recipe;
no license-affecting changes.
Signed-off-by: Matt Madison <matt@madison.systems>
* Python package files moved to a subdirectory
* Need to use gitsm:// to pick up the onnx parser submodule
* Refresh the build-related changes and patch
Signed-off-by: Matt Madison <matt@madison.systems>
* Switch git:// to gitsm:// to pick up the ONNX submodule
* Refresh build patch, dropping some long-obsolete changes
* Replace second build patch with the correct fix via CUDAFLAGS setting
Signed-off-by: Matt Madison <matt@madison.systems>
to add -trimpath to the build of libnvidia-container-go.so, so it doesn't
trigger buildpaths QA warnings.
Signed-off-by: Matt Madison <matt@madison.systems>
to fix a problem with inconsistent results from the symlink chain resolver
used to set up the hooks to create symlinks.
Signed-off-by: Matt Madison <matt@madison.systems>
With the updated toolkit/runtime setup, the toolkit is run at boot time to generate the
configuration file used by the container runtime. To allow for a read-only rootfs setup,
we place the generated config.toml under /run. We need to install a symlink in the
normal /etc/nvidia-container-runtime directory to point to that generated config, so the
container runtime can find it.
Signed-off-by: Matt Madison <matt@madison.systems>
* Added patch to support location of files in the container-passthrough tree
* Added systemd service to generate the config.toml file for the container
runtime during boot, so we can add the location of the passthrough tree.
Signed-off-by: Ilies CHERGUI <ichergui@nvidia.com>
Signed-off-by: Matt Madison <matt@madison.systems>
* Dropped libnvidia-container-jetson recipe, which is now obsoloete
Signed-off-by: Ilies CHERGUI <ichergui@nvidia.com>
Signed-off-by: Matt Madison <matt@madison.systems>
* No more 'S = "${WORKDIR}"', which is not permitted
* All ${WORKDIR} references for local files updated to ${UNPACKDIR}
* Custom do_unpack for the recipes that are in the public_sources
archive in the BSP adjusted for the new unpacking approach
* Some SRC_URIs updated with 'destsuffix=' parameters where needed
Signed-off-by: Matt Madison <matt@madison.systems>
The original problem with compatibility with newer version of libtirpc was
resolved, so we don't need the workaround any longer.
Also drop runtime dependencies that have been moved to the main libnvidia-container
recipe. This package can't be used on its own, anyway, as it's called on
by the nvidia-container-cli tool built in the libnvidia-container recipe.
Signed-off-by: Matt Madison <matt@madison.systems>
* Refresh the build patch to have nvcgo built, so the library can be used
with cgroupsv2 (for 'auto' and 'csv' modes)
* Move some runtime dependencies from libnvidia-container-jetson to
this recipe, as they're needed here
* Patch the toolkit to implement 'alternate roots' for locating files to be
passed through from the host, to support our tegra-container-passthrough tree
* Provide our own config file, rather than modifying the one from the source tree
on the fly
This enables us to support 'csv' mode in the toolkit for our builds, instead
of requring 'legacy' mode. With 'csv' mode, host passthrough mounts are injected
into the container when it is created, rather than using the legacy method of
adding a prestart hook to run nvidia-container-cli to set them up.
The 'legacy' mode can still be used, if needed, as long as your target system
is configured to support cgroupsv1.
Signed-off-by: Matt Madison <matt@madison.systems>
Previously the 17 standard was needed just for some plugins, and leaving
14 was fine for the rest of the project. With recent changes to absl, now
we need the 17 standard for other parts as well.
This change builds on the do_configure:prepend sed fixup to include the
CMakeLists.txt CMAKE_CXX_STANDARD being forced in a couple more places.
Signed-off-by: Kurt Kiefer <kekiefer@gmail.com>
Due to updates to protobuf and gcc, make some updates to the build. Namely,
this means using C++17 as required by protobuf and making sure we link to
the system protobuf.
Signed-off-by: Kurt Kiefer <kekiefer@gmail.com>
This brings container support recipes up to the versions used in
L4T R35.4.1/JetPack 5.1.2.
* Rename libnvidia-container-tools recipe to the more accurate 'libnvidia-container', and
update to v1.10.0 (with patch consolidiation and refresh)
* Fix the makefiles for libnvidia-container and libnvidia-container-jetson to eliminate
the build paths that were getting embedded in the RPC files
* Include the full set of DEBUG_PREFIX_MAP flags when building libnvidia-container, which fixes
more embedded build paths
* Update nvidia-container-toolkit to v1.11.0
* Switch to static linking for nvidia-container-toolkit to work around panic on startup
in nvidia-ctk
* Hard-code the toolkit configuration to "legacy" mode, which uses the older prestart hook
method for mapping the paths in the CSV file, until our hacks for the passthrough tree
and for mapping the Debian/Ubuntu multi-arch library paths to the normal host paths
get migrated to the Go code in nvidia-container-toolkit.
* Update nvidia-docker to v2.11.0
Signed-off-by: Matt Madison <matt@madison.systems>
nvidia-container-cli requires access to the host ldconfig at runtime, so
have it as a runtime dependency explicitly.
Signed-off-by: Ricardo Salveti <ricardo@foundries.io>
Add separate bbappends for Qt 6 (https://code.qt.io/cgit/yocto/meta-qt6.git).
Enabled EGLFS EglDevice backend plugin to be used with tegra libraries
and add extra CMake arguments to select the preferred backend plugin
based on the preferred gbm provider.
Signed-off-by: Samuli Piippo <samuli.piippo@qt.io>
This reverts commit 35dab0f604.
Commit 225ce6a14a8110ab6b573b4dc9f5297a03d17e0f in meta-oe added in
a patch for opencv that fixes the C++11/C++17 protobuf issue for the
dnn module, so we can put dnn back in the PACKAGECONFIG.
Signed-off-by: Matt Madison <matt@madison.systems>
from the EXTRA_OECMAKE setting, as the cmake build there uses CUDA
language support instead of the obsolete FindCUDA module.
Signed-off-by: Matt Madison <matt@madison.systems>
docker-ce is deprecated, with docker-moby now set as the default
docker recipe, so switch over the bbappend to fix up the kernel
module dependencies for that recipe.
Signed-off-by: Matt Madison <matt@madison.systems>
systemd's random-seed service units/targets have changed with
recent updates, and the haveged service was causing a dependency
loop. Disable the default dependencies, and change the WantedBy
to sysinit.target to fix this.
Signed-off-by: Matt Madison <matt@madison.systems>
* Update OpenCVDetectCUDA.cmake patch to prevent insertion of
build paths into the generated opencv4.pc file
* Disable cmake build checks/regeneration so the build isn't
regenerated during the install step, which is causing
incorrect recompilations
Signed-off-by: Matt Madison <matt@madison.systems>
to add a patch that supports the use of tree of files to be directly
exported to containers, overriding any files that reside in the normal
host rootfs. This allows us to export NVIDIA's prebuilt binary userland
libraries to NVIDIA's containers, to ensure compatibility with the library
versions (notably glibc's) that are inside the container.
Signed-off-by: Matt Madison <matt@madison.systems>
