| Age | Commit message (Collapse) | Author |
|---|
|
Add new seamaphore support,
Add uvd instance info
Signed-off-by: Sanjay R Mehta <sanju.mehta@amd.com>
|
|
Signed-off-by: Sanjay R Mehta <sanju.mehta@amd.com>
|
|
These configurations conflict with KGDB debugging in
some cases and generate stack dumps in various other
situations.
The stack dumps related to mwaitx are properly fixed
through
https://patchwork.kernel.org/patch/8549611/
but the mentioned patch is only available for Yocto
kernel 4.8 and we don't use these so it is much
better to avoid the problem with this for the time
being rather than backporting the patch.
Signed-off-by: Awais Belal <awais_belal@mentor.com>
|
|
mesa: Adding bug fixes for tesselation and unigine heaven issues
|
|
This fixes the kernel and amdgpu hang when I2S
audio is enabled through the BIOS.
Signed-off-by: Awais Belal <awais_belal@mentor.com>
|
|
Signed-off-by: Sanjay R Mehta <sanju.mehta@amd.com>
|
|
These are installed as needed through the initramfs
bits where these are actually required.
Signed-off-by: Awais Belal <awais_belal@mentor.com>
|
|
codexl: Output versioned libs
|
|
We've moved on to LLVM 3.9.1 so this is no
longer needed.
Signed-off-by: Awais Belal <awais_belal@mentor.com>
|
|
to resolve shlib dependency warnings
|
|
GStreamer VAAPI updates
|
|
This is no longer required after the upgrade to the latest
gstreamer-vaapi component.
Signed-off-by: Awais Belal <awais_belal@mentor.com>
|
|
Upgrade gstreamer1.0-vaapi to 1.8.3 as all the other
plugins are on the same version and this contains
several new fixes.
Signed-off-by: Awais Belal <awais_belal@mentor.com>
|
|
The recipe naming for different gstreamer components is pretty
well defined upstream, where gstreamer1.0- is used as a
prefix for all components. This is now corrected for our
in house gstreamer vaapi recipe.
Signed-off-by: Awais Belal <awais_belal@mentor.com>
|
|
llvm3.9.1: move to Ninja based builds
|
|
We'll have to delete the libLLVM.so here because poky/meta-oe allows
multiple versions of LLVM available at the same moment so it will
clash. Also, in rpm packaging -<version> is appended to every
package. This creates a clashing target between libllvm3.9.1-llvm and
libllvm3.9.1-llvm-3.9.1 when manual package splitting is done through
the recipe.
Signed-off-by: Awais Belal <awais_belal@mentor.com>
|
|
v4l-utils: add dependency on libgl and libglu when opengl is enabled
|
|
codexl: Resolve race conditions with mkdir.
|
|
The utility v4l2gl requires gl.h and libgl + libglu libraries
for building cleanly. We need to add an explicit dependency
here to avoid any sort of races although the configure
script checks for availability of the libraries by default
but this can race on the header (gl.h) availability on the
sysroot causing:
| ../../../v4l-utils-1.10.1/contrib/test/v4l2gl.c:27:19: fatal error: GL/gl.h: No such file or directory
| #include <GL/gl.h>
| ^
| compilation terminated.
during do_compile.
Signed-off-by: Awais Belal <awais_belal@mentor.com>
|
|
LLVM wikis cite Ninja based builds more widely
than CMake based builds and there have been
instances where CMake builds were thought of
being broken and not working as expected.
e.g.
https://llvm.org/bugs/show_bug.cgi?format=multiple&id=23328
The move to Ninja should help us to resolve all
sorts of intermittent build failures we're
seeing with LLVM builds.
Signed-off-by: Awais Belal <awais_belal@mentor.com>
|
|
There is an apparent race condition with the mechanism used
for making directories. There are corner cases where the
compile step is invoked before the output objdir is created.
This forces all compile steps to be prepended with a mkdir
command. This may be more than is needed but does seem to
resolve the current races.
Signed-off-by: Drew Moseley <drew_moseley@mentor.com>
|
|
This ensures that ISO images will not be included
in the prebuilt binaries when they are deployes to
the installer even when iso is present in
IMAGE_FSTYPES. The earlier mechanism only handled
IMAGE_EXTENSION variable.
Signed-off-by: Awais Belal <awais_belal@mentor.com>
|
|
The Toaster build artifacts collection mechanism depends
on the IMAGE_FSTYPES variable to identify if required
build artifacts are present. The Toaster GUI only shows
the build artifacts if they are detected properly.
We now add the required image types explicitly for
AMD BSPs so Toaster shows the build artifacts as
intended.
Signed-off-by: Awais Belal <awais_belal@mentor.com>
|
|
mpv: disable lua support
|
|
We support MEL builds on 32 bit hosts as well but luajit
which is required when lua support is enabled in mpv
cannot be built on a 32 bit host when the target is
64 bit so we disable the lua support.
Signed-off-by: Awais Belal <awais_belal@mentor.com>
|
|
Signed-off-by: Sanjay R Mehta <sanju.mehta@amd.com>
|
|
Upgraded llvm from 3.7.1 to 3.9.1 for AMD BSPs.
The recipe has been created from scratch taking and
merging changes from meta-oe because there are things
that cannot be overriden through appends or any other
cleaner mechanism. mesa has been adjusted accordingly
as well.
Signed-off-by: Awais Belal <awais_belal@mentor.com>
Signed-off-by: Sanjay R Mehta <sanju.mehta@amd.com>
|
|
This upgrades mesa to 12.0.3 from 11.0.8.
Signed-off-by: Awais Belal <awais_belal@mentor.com>
Signed-off-by: Sanjay R Mehta <sanju.mehta@amd.com>
|
|
libffi headers are checked by packages which depend
on it so we need to provide these on the sysroot.
Signed-off-by: Awais Belal <awais_belal@mentor.com>
|
|
This fixes the do_rootfs issues that come up while using
the extrausers functionality. The extrausers class
requires shadow as well as base-passwd to leverage
the functionality of the class. In case of initramfs
images the PACKAGE_INSTALL variable is overriden in
a way that the additions done through extrausers
bbclass are rendered useless.
Signed-off-by: Awais Belal <awais_belal@mentor.com>
|
|
Signed-off-by: Ahsan Hussain <ahsan_hussain@mentor.com>
|
|
Update grub and grub-efi
|
|
cml1-config: Remove unused bbclass file
|
|
Enable automated strip of debug symbols from the binaries and
shared objects. Couple of Workarounds applied;
1- shared objects(.so) mode is set to executable, so that these
get strip as well, otherwise auto strip will leave these as it is.
2- example program binaries mode is set to non executable in
do_install to skip the auto strip because these symbols will be
required during the debugging of the examples, after do_package
their mode is set back to executable.
It reduces codexl remote agent footprint from 160MB to 40MB.
Signed-off-by: Adeel Arshad <adeel_arshad@mentor.com>
|
|
Signed-off-by: Drew Moseley <drew_moseley@mentor.com>
|
|
This enables git based grub recipes to be used
for AMD BSPs.
Signed-off-by: Awais Belal <awais_belal@mentor.com>
|
|
This actually copies over the grub-efi_2.00 recipe
and makes simple modifications so it can use git as
source. This will give us the chance to use the
latest and greatest grub/grub-efi on our platforms.
Signed-off-by: Awais Belal <awais_belal@mentor.com>
|
|
grub2: fix some quirks and div by zero
|
|
Rather than erroring out on a single attempt while
terminating EFI services, make a few retries because
such quirks are found in a few implementations.
Also fix a div by zero issue in the same framework
which causes an infinite reboot on the target.
Both patches included here are backports.
Signed-off-by: Awais Belal <awais_belal@mentor.com>
|
|
Fix var expansion and override
|
|
If a variable is not defined at all bitbake tends to use
the literal value which causes failures and unintended
settings in various cases. This now uses a very weak
default to cope with this problem.
Signed-off-by: Awais Belal <awais_belal@mentor.com>
|
|
Revert "python3: fix for multilib environment"
|
|
VULKAN_COMPONENTS, CODEXL_COMPONENTS needs to be assign
the default values for the machine other than amdfalconx86,
otherwise it generates a parsing error.
Also only populate CODEXL_COMPONENTS for amdfalconx86 machine,
as we do not support it for other machines.
Signed-off-by: Adeel Arshad <adeel_arshad@mentor.com>
|
|
|
|
bb.utils.to_boolean generates the parsing error when the
variable it is evaluating is not defined, so replace it with
bb.utils.contains which handles this case correctly.
Signed-off-by: Adeel Arshad <adeel_arshad@mentor.com>
|
|
We have analyzed that the configuration variable INCLUDE_GAMING_COMPONENTS,
is no longer valid as the components it include are not particularly
specific to gaming. So we have splitted it into two configuration variables
INCLUDE_VULKAN and INCLUDE_CODEXL.
Signed-off-by: Adeel Arshad <adeel_arshad@mentor.com>
|
|
codexl: fix build with gcc5.x
|
|
This enables the chacha20 scheme of random number
generator which is way more optimized and
efficient from the legacy non-blocking pool. This
helps providing a steady and more robust random
mechanism which allows for better boot times and
other positives.
The included patch has been back ported from
linux-yocto 4.8 tree and adjusted accordingly.
Ref:
https://git.yoctoproject.org/cgit/cgit.cgi/linux-yocto-4.8/commit/drivers/char/random.c?h=standard/base&id=e192be9d9a30555aae2ca1dc3aad37cba484cd4a
Signed-off-by: Awais Belal <awais_belal@mentor.com>
|
|
|
|
codexl: fix build with gcc6
|
Sanjay R Mehta
Awais Belal
Wade Farnsworth
Drew Moseley
Michael Powell
Drew Moseley
Ahsan Hussain
Adeel Arshad