| Commit message (Collapse) | Author | Age | Files | Lines |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Remove support for the v6.1 TI and Beagle BSPs. This was originally
split into meta-ti-bsp and meta-beagle changes, but there are some
hidden interdependencies that made that more painful than needed.
Remove v6.1 support from both BSP layers here.
This includes packages specific to those BSPs:
- TI and BB.org v6.1 Linux kernel
- TI and BB.org 2023.04 U-Boot
- Rogue GPU driver 23.3 and Mesa 22.3.5
Signed-off-by: Andrew Davis <afd@ti.com>
Signed-off-by: Ryan Eatmon <reatmon@ti.com>
Signed-off-by: Denys Dmytriyenko <denys@konsulko.com>
|
|
|
|
|
|
|
|
|
| |
If a recipe is fetching an entire git repo and then just building a
single dir, then we need to update S to lock into the sub directory.
Easiest fix is to simply append the subdir to whatever the system wants
to set S to by default.
Signed-off-by: Ryan Eatmon <reatmon@ti.com>
|
|
|
|
|
|
|
|
|
|
| |
BB_GIT_DEFAULT_DESTSUFFIX
Removing all the S = ${WORKDIR}/git assignments works because BB_GIT_DEFAULT_DESTSUFFIX
is set to match S from bitbake.conf (which itself is set to match typical tarball
releases).
Signed-off-by: Ryan Eatmon <reatmon@ti.com>
|
|
|
|
|
|
|
|
| |
Every recipe should have a SUMMARY and that can only be 72 characters at
most. Some of these recipes had DESCRIPTIONS and not SUMMARY, some were
missing the SUMMARY, and some had slightly too long values for the SUMMARY.
Signed-off-by: Ryan Eatmon <reatmon@ti.com>
|
|
|
|
|
|
|
|
|
|
| |
Fix the package config params to make it obvious what combinations are
allowed. Also, rework the way we are inheriting init related classes so
that we don't accidentally do any additional postinstall steps. While
we're at it, don't execute anything for the compile or configure stage.
Signed-off-by: Randolph Sapp <rs@ti.com>
Signed-off-by: Ryan Eatmon <reatmon@ti.com>
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Well, this is the evil vendor driver. Time to start acting like it.
Blacklist the upstream powervr module if this driver is present.
This patch also does away with the bin_package components. It was cool
and found some interesting bugs, but we might as well start leveraging
that Makefile more now that we're starting to get some more common
files between devices.
Signed-off-by: Randolph Sapp <rs@ti.com>
Signed-off-by: Ryan Eatmon <reatmon@ti.com>
|
|
|
|
|
|
|
|
|
| |
No reason to do this pre install now that my python function can
evaluate file paths. Also, this may have been causing issues with state
when attempting to switch between opencl enabled/disabled builds.
Signed-off-by: Randolph Sapp <rs@ti.com>
Signed-off-by: Ryan Eatmon <reatmon@ti.com>
|
|
|
|
|
|
|
|
| |
Bump the package version for DDK 24.2, this includes a few fixups from
IMG and support for a newer version of mesa.
Signed-off-by: Randolph Sapp <rs@ti.com>
Signed-off-by: Ryan Eatmon <reatmon@ti.com>
|
|
|
|
|
|
|
|
|
|
|
| |
Resolve some issues with J721E and compressed NPOT (Non-Power-Of-Two)
textures.
Also, add some additional allocation tracking in render process specific
tasks.
Signed-off-by: Randolph Sapp <rs@ti.com>
Signed-off-by: Ryan Eatmon <reatmon@ti.com>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
We got a bit more time to do a proper migration this time. A discovery
period to see what actually works and boy, it wasn't much.
This update resolves issues with:
- Macro redefinition
- Sparse NPOT texture allocations
- vec4 reductions
- Mesa image extensions
- Buffer allocations under weston
- Devices with both a RENDER and PRIMARY display node
- Underprivileged users being unable to allocate buffers
- Removed call to drm_map_legacy
- Replace deprecated kernel calls
- Fix some cache attributes for userspace allocations
- About 40 other various errors and warnings
This core was never fully GLES2 conformant. In fact, if you try to run
cts on any of the older releases you'll find yourself in a hardware
recovery loop pretty quickly. We've gotten it to about 94.2% conformance
with these patches.
There are still dragons here, but they are at least a little more
manageable.
Signed-off-by: Randolph Sapp <rs@ti.com>
Signed-off-by: Ryan Eatmon <reatmon@ti.com>
|
|
|
|
|
|
|
| |
j742s2 has the same GPU as j784s4. Add j742s2 support to the recipes.
Signed-off-by: Aniket Limaye <a-limaye@ti.com>
Signed-off-by: Ryan Eatmon <reatmon@ti.com>
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Adding this sanity check here preemptively. Since these libraries were
compiled with an older version of GCC than what is currently in use on
scarthgap and master they won't trip the sanity check just yet, but if
someone attempts to update them and uses a newer compiler (for some
reason) it will trigger the sanity checks again.
This also keeps the DDK 23.3 and 24.1 recipe diff to a minimum.
Signed-off-by: Randolph Sapp <rs@ti.com>
Signed-off-by: Ryan Eatmon <reatmon@ti.com>
|
|
|
|
|
|
|
|
|
|
|
|
| |
Add ldflags to INSANE_SKIP for the libopencl-rogue-tools package. For
some reason I didn't catch this sooner, but this falls victim to the
same issues the parent package faced earlier this year.
Just enough options enabled that Yocto wants to do a ldflags check, but
the DDK will not be able to utilize all the build flags Yocto uses.
Signed-off-by: Randolph Sapp <rs@ti.com>
Signed-off-by: Ryan Eatmon <reatmon@ti.com>
|
|
|
|
|
|
|
|
|
|
|
| |
Bump this recipe to add a change to help with GCC 14 compatibility in
the check-cc script used to check compiler flags before use.
Reset this back to the proper branches for kernel 6.1 since this recipe
is only being used in 6.6 bsp providers.
Signed-off-by: Randolph Sapp <rs@ti.com>
Signed-off-by: Ryan Eatmon <reatmon@ti.com>
|
|
|
|
|
|
|
|
| |
Bump the SRCREV to pick up some new binaries for AM62P. The allocation
patch for Chromium was not included for this device previously.
Signed-off-by: Randolph Sapp <rs@ti.com>
Signed-off-by: Ryan Eatmon <reatmon@ti.com>
|
|
|
|
|
|
|
|
|
|
| |
Bump the SRCREV to pick up rebuilt binaries. Upstream release notes:
Regenerate all binaries with a fix for exportable buffer allocations
that should improve Chromium compatibility.
Signed-off-by: Randolph Sapp <rs@ti.com>
Signed-off-by: Ryan Eatmon <reatmon@ti.com>
|
|
|
|
|
|
|
|
| |
Now that are past the inital stages of our LTS migration, we can remove
the DEFAULT_PREFERENCE settings that prevented these recipes from being
the default.
Signed-off-by: Ryan Eatmon <reatmon@ti.com>
|
|
|
|
|
|
|
|
|
| |
Recipes for DDK 24.1, but lock them behind ltsprep tags because they
require device tree changes that do not work on the current default
kernel for this layer.
Signed-off-by: Randolph Sapp <rs@ti.com>
Signed-off-by: Ryan Eatmon <reatmon@ti.com>
|
|
|
|
|
|
|
|
| |
Switch the branches to the new LTS names. These are still experimental,
hence the isolation from kirkstone.
Signed-off-by: Randolph Sapp <rs@ti.com>
Signed-off-by: Ryan Eatmon <reatmon@ti.com>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Migrate to version 23.3.6512818, this includes the stability updates
released in 23.2 as well as some additional modifications to help with
userspace application support, namely Chromium.
We're also dropping the x11 window system extensions from the Vulkan
libraries for now. Ideally, both x11 and wayland extensions will be
provided by mesa's vulkan-wsi-layer in the future, but currently x11
isn't supported and this just added extra complexity to the package in
Yocto.
Signed-off-by: Randolph Sapp <rs@ti.com>
Tested-by: Darren Etheridge <detheridge@ti.com>
Signed-off-by: Ryan Eatmon <reatmon@ti.com>
|
|
|
|
|
|
|
|
|
| |
If Yocto even sees a "-dev" package in the PACKAGES string it will
automatically forward all dev-so links to that package. Remove that
entry from the PACKAGES string so we get the required links again.
Signed-off-by: Randolph Sapp <rs@ti.com>
Signed-off-by: Ryan Eatmon <reatmon@ti.com>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Bump SRCREV to include the following changes in UM and KM:
- Rebuild all releases with external memory wrap disabled due to
some incompatibility with certain Qt use cases.
- Add a watchdog change that prevents an occasional false
timeout when servicing long DM processes.
- Rely on snooping to perform certain cache operations and
results in decreased overhead for device -> cpu
communications.
Signed-off-by: Randolph Sapp <rs@ti.com>
Signed-off-by: Ryan Eatmon <reatmon@ti.com>
|
|
|
|
|
|
| |
Signed-off-by: Darren Etheridge <detheridge@ti.com>
Signed-off-by: Aniket Limaye <a-limaye@ti.com>
Signed-off-by: Ryan Eatmon <reatmon@ti.com>
|
|
|
|
|
|
| |
Signed-off-by: Darren Etheridge <detheridge@ti.com>
Signed-off-by: Aniket Limaye <a-limaye@ti.com>
Signed-off-by: Ryan Eatmon <reatmon@ti.com>
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Add the configuration options for am62pxx-evm.
This also adds support for new devicetree properties to override the
clock value of the device. See the kernel module modifications [1] and
dt bindings for more information.
[1] https://git.ti.com/cgit/graphics/ti-img-rogue-driver/log/?h=linuxws/kirkstone/k6.1/23.2.6460340
Signed-off-by: Randolph Sapp <rs@ti.com>
Signed-off-by: Ryan Eatmon <reatmon@ti.com>
|
|
|
|
|
|
|
|
|
|
| |
Move to the 23.2 release branch and bump SRCREV accordingly. This branch
has the proper implementation of the cache coherency fix along with some
other checks that address some system stability issues on Jacinto
devices.
Signed-off-by: Randolph Sapp <rs@ti.com>
Signed-off-by: Ryan Eatmon <reatmon@ti.com>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Bump ti-sgx-ddk-um and ti-sgx-ddk-km to the new branch for Kirkstone and
mesa independent builds. This also sets up SGX to use the latest mesa
revision rogue uses.
Changelog:
- IMG Support has officially ended. This repo will not be getting
regular updates after this unless things are really broken.
- The Mesa patches which were previously being relicensed internally
were opened and combined with the unofficial rogue efforts on
freedesktop.
- Toolchains were upgraded and multiple warnings and issues were
addressed. Hardware recoveries work now, but there are still some
longstanding issues with SGX that remain to be addressed if further
resources are set aside for that.
- We moved to an lws-generic build target to avoid rebuilding external
components. As such, these external components must be patched as
needed. Libdrm is an example of this and the required patches can be
found in meta-ti.
Known bugs:
- Buffer allocations under Weston may misbehave
- GLSL macro redefinitions will cause errors regardless of whether or
not they are identical
Signed-off-by: Randolph Sapp <rs@ti.com>
Signed-off-by: Ryan Eatmon <reatmon@ti.com>
|
|
|
|
|
|
|
|
|
|
| |
Only move directories which haven't already been moved.
This prevents a build error should a subsequent bitbake execution
rerun do_install where the sources to 'mv' no longer exist.
| mv: cannot stat 'ti-img-rogue-umlibs/23.1.6404501/git/targetfs/am62_linux/lws-generic/release/lib/firmware': No such file or directory
Signed-off-by: Max Krummenacher <max.krummenacher@toradex.com>
Signed-off-by: Ryan Eatmon <reatmon@ti.com>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
A recent change to oe-core [1], has caused a pile of QA errors during
packaging for this binary package. The effect of the change in oe-core
was to remove the dependency on the compiler and other compiler adjacent
libraries. But the binary packages need some of the compiler tools as
part of the packaging flow (ie. objdump).
Until the above gets reverted, this patch will fix the issue and can be
reverted itself once oe-core is fixed.
[1] https://git.openembedded.org/openembedded-core/commit/?h=master&id=d1d09bd4d7be88f0e341d5fccbfbefeb98d4b727
Signed-off-by: Ryan Eatmon <reatmon@ti.com>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Bump the SRCREV for the umlibs and driver package to include some fixes
for Jacinto devices. These patches do the following:
- Fix virtual memory addressing
- Resolve an issue with APM and debug requests
- Attempt to resolve most of the cache coherency related issues on
Jacinto devices
- Introduce some extra cache maintenance operations in the kernel
module
It also introduces the following known issues:
- Visual artifacting on the second-to-last EGL/GLES context on BXS
based devices (J721S2/J784S4)
We're still working on that last one, but the stability improvements
make this worth while for now.
Signed-off-by: Randolph Sapp <rs@ti.com>
Signed-off-by: Ryan Eatmon <reatmon@ti.com>
|
|
|
|
|
|
|
|
|
| |
Edit the file paths so they point at files in the source directory
explicitly. For some reason shell glob were previously only respected by
some machines.
Signed-off-by: Randolph Sapp <rs@ti.com>
Signed-off-by: Ryan Eatmon <reatmon@ti.com>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Use RRECOMMENDS for components. Apparently I made an resolvable
dependency cycle that Yocto silently clipped. My best guess at the
culprit was: mesa-megadriver (RR) -> rogue-km (RR) -> rogue-um (RD) ->
libgles1-rogue (RD) -> mesa-megadriver
That killed the ti-img-rogue-umlibs recipe during do_rootfs. The rogue
GLES libs do depend on mesa-megadriver at runtime but I guess the UM
components should only really recommend the rogue GLES components be
present at runtime to make the cycle resolvable by opkg.
Fixes:
1705f6d3 (ti-img-rogue-umlibs: make all components optional, 2023-06-21)
Signed-off-by: Randolph Sapp <rs@ti.com>
Signed-off-by: Ryan Eatmon <reatmon@ti.com>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Bumping into the limitations of the binary_package class. This change
allows all rogue components to be optionally packaged without blowing
out the dependency chain.
Add a python function to fetch file paths from the FILE vars for each
component so we don't end up double tracking files. Add variables to
track component packages if things split up more later. Add a package
config string to set runtime dependencies and recommendations. Change
all logic to revolve around the PACKAGECONFIG instead of distro
features.
Signed-off-by: Randolph Sapp <rs@ti.com>
Signed-off-by: Ryan Eatmon <reatmon@ti.com>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Well, flew a bit too close to the sun on that last patch. Apparently
there are a few hidden runtime dependencies on the API devlinks that
make things like EGL upset when they aren't present.
We may circle back to this at some point but for now we're fine shipping
the devlinks in the base API packages as these links don't step on any
other package.
Fixes: 1899f5ea (ti-img-rogue-umlibs: bump version and explicitly
package, 2023-06-01)
Signed-off-by: Randolph Sapp <rs@ti.com>
Signed-off-by: Ryan Eatmon <reatmon@ti.com>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Bump the version and explicitly package all Khronos APIs separately.
This should be nice for those folks that want a headless opencl system
for tinkering and such.
This version includes experimental GEM support (for applications like
Chromium that like to poke at the render node) and a new Vulkan ICD
loader interface for better support with the Vulkan ICD loaders present
in Yocto.
This lifts a method from mesa used to force a common naming convention
for Vulkan, OpenCL, and OpenGLES libs to make things a little easier to
follow. Without it the target package names get scrambled based on the
shared object name.
Signed-off-by: Randolph Sapp <rs@ti.com>
Signed-off-by: Ryan Eatmon <reatmon@ti.com>
|
|
Move recipes to new subdir since we offer more than just libgles now.
Reusing the naming convention of the subdirectory used for the km
components for now.
Signed-off-by: Randolph Sapp <rs@ti.com>
Acked-by: Andrew Davis <afd@ti.com>
Signed-off-by: Ryan Eatmon <reatmon@ti.com>
|