As am57xx.inc is now the only user of omap-a15.inc, merge the content of
omap-a15.inc into am57xx.inc and remove omap-a15.inc. As we no longer have
this common include, also remove "omap-a15" and use "am57xx" in its place.
Signed-off-by: Andrew Davis <afd@ti.com>
Reviewed-by: Denys Dmytriyenko <denys@konsulko.com>
Signed-off-by: Ryan Eatmon <reatmon@ti.com>
New recipes for GPU DDK 25.2. Update BSP configuration files with new
preferred recipe versions.
For the driver recipe, specifically unset the CC variable due to yocto
appending CROSS_COMPILE and CC, resulting in a warning for the
do_compile task:
make[1]: aarch64-oe-linux-aarch64-oe-linux-gcc: No such file or directory
Signed-off-by: Antonios Christidis <a-christidis@ti.com>
Signed-off-by: Ryan Eatmon <reatmon@ti.com>
Now that display controller aliases are not set using PACKAGECONFIG,
instead of using PREFERRED_PROVIDER to add either "sgx" or "pvr" to
PACKAGECONFIG, followed by reading it back out, simply use
PREFERRED_PROVIDER directly.
Signed-off-by: Andrew Davis <afd@ti.com>
Signed-off-by: Ryan Eatmon <reatmon@ti.com>
Update mesa-pvr which includes a default set of display controller
aliases for the supported display controllers. This means we do not
have to manually set the alias for each device here in the integration
layer. Remove that here.
Signed-off-by: Andrew Davis <afd@ti.com>
Signed-off-by: Ryan Eatmon <reatmon@ti.com>
Seems having hyphens in the revision tag causes issues with
RPM packaging. Remove this hyphen.
Reported-by: Daniel Schultz <D.Schultz@phytec.de>
Signed-off-by: Andrew Davis <afd@ti.com>
Signed-off-by: Ryan Eatmon <reatmon@ti.com>
This change is needed due to the SGX driver, any distro, not just Arago,
using a TI SoC with SGX will need this fix, so it should go in meta-ti.
Signed-off-by: Andrew Davis <afd@ti.com>
Acked-by: Denys Dmytriyenko <denys@konsulko.com>
Signed-off-by: Ryan Eatmon <reatmon@ti.com>
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>
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>
OE-core updated mesa to 25.0.2 [1]. The resulting mesa.inc file no
longer matches mesa recipes building older versions.
Copy the mesa.inc version from before the update renamed to
mesa-pvr.inc to work around the issue.
| ../git/meson.build:21:0: ERROR: Value "softpipe" for option "gallium-drivers" is not in allowed choices: "auto, kmsro, radeonsi, r300, r600, nouveau, freedreno, swrast, v3d, vc4, etnaviv, tegra, i915, svga, virgl, panfrost, iris, lima, zink, d3d12, asahi, crocus, pvr, sgx, all"
[1] https://git.openembedded.org/openembedded-core/commit/?id=f9eb0468e45ee5a6a3b3195ef5e78c328c4347c9
Signed-off-by: Max Krummenacher <max.krummenacher@toradex.com>
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>
The user mode libraries needed recompiling with the scarthgap revision of gcc to
remove all the warnings about 32bit API usage during the linking stage, new
SRCREV includes this changes. Versions of the libraries remain the same.
Signed-off-by: Antonios Christidis <a-christidis@ti>
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>
Relocate the patches for 22.3.5 to the mesa-pvr-22.3.5 subdirectory so
it's more clear what version of mesa they apply to.
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>
The gfx library provided by rogue can support this never version of mesa
out of the box. Pull it in now that we have ported the SGX side of
things.
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>
Bump versions of mesa-pvr and ti-sgx-ddk-km to support SGX with kernel
6.6 and the same version of mesa rogue is using.
Enable SGX by default in the 6.6 BSP provider section logic.
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>
openembedded-core commit f5cfb3e23603 ("mesa: set PV from the .inc
file and not via filenames") now sets PV in the mesa.inc file
explicitely to whatever version oe-core is using.
Override that setting to the actual version here to prevent the
following parse error and whatever else would go wrong afterwards.
| ERROR: .../mesa-pvr_23.2.1.bb: Unable to get checksum for mesa-pvr SRC_URI entry 0001-meson-Disable-cmake-dependency-detector-for-llvm.patch: file could not be found
| The following paths were searched:
| .../meta-ti-bsp/recipes-graphics/mesa/mesa-pvr-24.0.7/bsp-ti-6_1/0001-meson-Disable-cmake-dependency-detector-for-llvm.patch
| ...
Signed-off-by: Max Krummenacher <max.krummenacher@toradex.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>
Add a recipe for mesa-pvr to use a newer branch based on mesa 23.2.1.
This branch does not currently contain SGX patches so it is also being
masked by the ltsprep branding to prevent regressing those devices
currently.
Please note that may run into sstate errors when switching to or away
from the ltsprep tag. Please issue a clean to mesa and
ti-img-rogue-umlibs if you get warnings from either package's
do_create_runtime_spdx step.
Signed-off-by: Randolph Sapp <rs@ti.com>
Signed-off-by: Ryan Eatmon <reatmon@ti.com>
Locally overlay verbatim copies of mesa 22.3.5 patches from
oe-core/master, so we get whatever QoL patches from core relevant to the
current release.
Signed-off-by: Randolph Sapp <rs@ti.com>
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>
Bump the SRCREV to include some useful patches that:
- Fix the ROGUE segfault issue introduced by SGX (this was what
the old conditional logic was addressing)
- Fix destruction of event queue with proxies still attached
Signed-off-by: Randolph Sapp <rs@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>
RPM does not accept '-' as a character in the PR/version and fails with:
| error: line 4: Illegal char '-' (0x2d) in: Release: sgxrgx-54fd9d7dea098b6f11c2a244b0c6763dc8c5690c.1
The correct approach is to reference SRCPV and not SRCREV and also add it
to PV and not PR. But PV is used in the branch to fetch the sources and
cannot be modified, so use SRCPV in PR w/o any illegal characters.
Signed-off-by: Denys Dmytriyenko <denys@konsulko.com>
Signed-off-by: Ryan Eatmon <reatmon@ti.com>
A bug was introduced while trying to unify the Mesa version that is used
on SGX GPU's and Rogue GPU's. It manifests itself as an immediate segfault
whenever you try and run something like glmark2-es2-wayland or
weston-simple-egl on certain Rogue based platforms (specifically am62x).
Use different srcrev for Rogue GPU and SGX GPU platforms. The SGX GPU platforms
will use the latest commit. Rogue GPU platforms will use the last good working
commit before SGX related changes were introduced.
This change will be reverted once the issue is rootcaused and a common solution
is found for both Rogue and SGX based GPU platforms.
Signed-off-by: Darren Etheridge <detheridge@ti.com>
Signed-off-by: Anand Balagopalakrishnan <anandb@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 SRCREV to include some patches to better isolate SGX and PVR
components.
Signed-off-by: Randolph Sapp <rs@ti.com>
Signed-off-by: Ryan Eatmon <reatmon@ti.com>
