These patches backport bug fixes for the GPT library in TF-M.
Signed-off-by: Frazer Carsley <frazer.carsley@arm.com>
Signed-off-by: Jon Mason <jon.mason@arm.com>
Returning from the finally block at the end of the start_fvp() function
is discarding exceptions. Since start_fvp() is near to the top off the
call tree, this hides uncaught exceptions thrown by most of the code,
which makes detecting and debugging issues hard.
This is resolved by removing the return statement from the finally
block, allowing exceptions to propagate normally.
Signed-off-by: Gyorgy Szing <gyorgy.szing@arm.com>
Signed-off-by: Jon Mason <jon.mason@arm.com>
Add support for Screen the GNU terminal multiplexer. The -t/--terminal
option now accepts "screen" as a terminal type. When selected, the tool
must be run from within an existing Screen session, and each FVP
terminal is opened in a new Screen window.
Signed-off-by: Gyorgy Szing <gyorgy.szing@arm.com>
Signed-off-by: Jon Mason <jon.mason@arm.com>
Improve usability by detecting which terminal types are available on the
system.
Extend the terminal abstraction to support checking whether a terminal
can be executed, and add basic validation for all terminal types. Update
the documentation to reflect the new behavior.
Signed-off-by: Gyorgy Szing <gyorgy.szing@arm.com>
Signed-off-by: Jon Mason <jon.mason@arm.com>
The initramfs for Corstone1000 is based on core-image-minimal, which has
GRUB included as a package. Corstone1000 uses U-Boot's efiloader to load
the initramfs and kernel and therefore does not require GRUB for boot.
Signed-off-by: Frazer Carsley <frazer.carsley@arm.com>
Signed-off-by: Jon Mason <jon.mason@arm.com>
Print an error message if the FVP binary cannot be found, instead of
failing silently.
Signed-off-by: Roger Knecht <roger.knecht@arm.com>
Signed-off-by: Jon Mason <jon.mason@arm.com>
When --selected_components is not specified, the script should include
all components as documented in the help text. However, the current
implementation skips all components when the parameter is empty,
resulting in an error.
Fix the filtering logic to only apply when --selected_components is
explicitly provided with values.
Signed-off-by: Kevin Hao <kexin.hao@windriver.com>
Signed-off-by: Jon Mason <jon.mason@arm.com>
Now that Xen has been removed, clean up the remaining references to
the meta-virtualization layer.
This was done as a separate commit to make reverting easier in the
future in case we need to add back meta-virtualization for some other
builds.
Signed-off-by: Ross Burton <ross.burton@arm.com>
Signed-off-by: Jon Mason <jon.mason@arm.com>
We originally added Xen builds to our CI to exercise Xen on Arm. Now
that the Yocto autobuilder has jobs that test Xen on qemuarm/qemuarm64
we no longer need the builds in meta-arm as they don't provide any more
coverage.
Signed-off-by: Ross Burton <ross.burton@arm.com>
Signed-off-by: Jon Mason <jon.mason@arm.com>
Remove the user-guide warning and rebuild steps for the
disable_module_autoloading workaround on Corstone-1000 with Cortex-A320.
With the NPU reset issue fixed, this workaround is no longer
needed.
Signed-off-by: Michael Safwat <michael.safwat@arm.com>
Signed-off-by: Jon Mason <jon.mason@arm.com>
- Derive host base addresses from offsets and add ext sys reset bitfields
- Deassert CPUWAIT with a bitmask
- Enable ext sys boot handling for Corstone1000 with Cortex-A320
- Add the new downstream TF‑M patches to the recipe
Signed-off-by: Michael Safwat <michael.safwat@arm.com>
Signed-off-by: Jon Mason <jon.mason@arm.com>
- Replace `systemready-patch` with `iot-platform-assets` in the guide.
- Update the Corstone-1000 asset paths and U85 kas command examples.
Align the user guide with the renamed asset repository, avoiding
stale commands and broken paths.
Signed-off-by: Alex Chapman <alex.chapman@arm.com>
Signed-off-by: Jon Mason <jon.mason@arm.com>
To workaround the CPU topology issue with edk2-firmware, we can specify
the number of processors being used, which seems to setup the CPU
topology mask correctly (and thus no errors). With this addressed, we
can use the latest edk2 firmware, which allows us to use the latest TF-A
version.
Signed-off-by: Jon Mason <jon.mason@arm.com>
rdv1 has been EOLed, and requires keeping around some recipes because
of its removal. Also, the FVP is x86 only, thus limiting CI.
Signed-off-by: Jon Mason <jon.mason@arm.com>
sgi575 has been EOLed, and requires keeping around some recipes because
of its removal. Also, the FVP is x86 only, thus limiting CI.
Signed-off-by: Jon Mason <jon.mason@arm.com>
Some of the soon-to-be-added FVPs have the execute issue still. So,
re-add this until it can be resolved. Also, address S issue that
changed upstream between the original removal and now.
Signed-off-by: Jon Mason <jon.mason@arm.com>
Update the Corstone-1000 user guide to reflect the current Ethos-U85
test and workaround flow.
As the Corstone-1000 software stack moves to the in-tree ethosu
driver, switch the Ethos-U85 test instructions from delegate_runner to
Mesa's test_teflon application. Also update the guide to use the
renamed ethos-u85-test kas fragment and to apply the Mesa patch needed
to package test_teflon into the image.
Also replace the disable_module_autoloading kas fragment in the A320
workaround instructions with the disable-ethosu patch, to align with
the workaround flow used in systemready-patch.
Signed-off-by: Michael Safwat <michael.safwat@arm.com>
Signed-off-by: Jon Mason <jon.mason@arm.com>
Corstone-1000 with Cortex-A320 FVP now supports SMP, so update the
documentation to reflect the current status.
Update the user guide to state that SMP is supported on Corstone-1000 with
Cortex-A35 FVP and on Corstone-1000 with Cortex-A320 FVP.
Add the Cortex-A320 multicore build and run commands.
Signed-off-by: Michael Safwat <michael.safwat@arm.com>
Signed-off-by: Jon Mason <jon.mason@arm.com>
Corstone1000 with Cortex A320 now uses the in-tree ethosu driver, so the
external meta-ethos layer (and its meta-sca dependency) is no longer needed.
Remove meta-ethos and meta-sca from the Corstone-1000 A320 kas
config, drop the layer dependency on meta-ethos, and stop
installing arm-npu-ethosu. Update the Corstone-1000
change log to reflect the removed layers.
Signed-off-by: Michael Safwat <michael.safwat@arm.com>
Signed-off-by: Jon Mason <jon.mason@arm.com>
Align the Corstone1000 Ethos-U85 device tree to match the upstream
bindings used by the in-tree Ethos-U DRM accel driver.
- Rework the Corstone1000 U-Boot patch to replace the legacy arm,ethosu-direct
node with an upstream-style Ethos-U85 node (arm,ethos-u85), add the
required clocks/clock-names, and switch the SRAM description to
mmio-sram.
- Drop meta-ethos specific properties (reserved-memory/dma-ranges,
/region-cfgs/mem-config) from the U-Boot DT.
- Enable required kernel options for the in-tree driver and SRAM provider:
- CONFIG_SRAM
- CONFIG_DRM
- CONFIG_DRM_ACCEL
- CONFIG_DRM_ACCEL_ARM_ETHOSU
Signed-off-by: Michael Safwat <michael.safwat@arm.com>
Signed-off-by: Frazer Carsley <frazer.carsley@arm.com>
Signed-off-by: Jon Mason <jon.mason@arm.com>
linux-yocto does not currently ship a 6.19 recipe, so add a
meta-arm-bsp linux-yocto_6.19.bb that tracks linux-yocto-dev v6.19/base.
Pin SRCREV_machine to fixed revision to keep builds reproducible.
Update Corstone-1000 to prefer linux-yocto 6.19 and align
the user guide pointers accordingly.
Signed-off-by: Michael Safwat <michael.safwat@arm.com>
Signed-off-by: Jon Mason <jon.mason@arm.com>
The U-Boot maintainers rejected the previous patch [1] for two primary
reasons:
1. The Cortex-A320 changes should be considered a separate platform
2. The NPU node bindings do not match those in the Linux kernel
The former is handled by this commit. The latter point has not been
resolved, hence marking the newly added patch as Inappropriate. This is
simply the first step in resolving the comments.
[1] https://lore.kernel.org/all/20251127154752.589691-1-frazer.carsley@arm.com
Signed-off-by: Frazer Carsley <frazer.carsley@arm.com>
Signed-off-by: Jon Mason <jon.mason@arm.com>
Rework the patch that enables the OF_UPSTREAM config option to split
off the extra device tree nodes into their own dtsi files, making it
easier to combine them in different ways. The rest of the patches have
changed only so that they can be applied cleanly.
Signed-off-by: Frazer Carsley <frazer.carsley@arm.com>
Signed-off-by: Jon Mason <jon.mason@arm.com>
Corstone1000 used to use its own implementation of a GPT parser for use
in a capsule update. The patches in this commit replace what exists with
the generic GPT library, as well as other minor self-explanatory fixes.
The wic file has undergone two changes as a result of this:
1. The partition type GUID of each of the four partitions that can be
updated by a capsule update must match the GUID of each capsule. This
allows for the existing partition to be matched with its updated
image. Different machines (e.g. MPS3 vs FVP) have different GUIDs for
these images hence the need for separate files.
2. The second bank has been removed from provisioning. Because the
library supports dynamic creation of partitions, the second bank no
longer needs to be provisioned at build time. However, a small
reserved partition is still created above the 32KiB mark to force
wic to size the disk as 64KiB and write this into the GPT header
for the library to read on initialisation.
Finally, the size of bl1_1 is reduced by one of the patches in this
commit, so this is also reflected in the recipe.
Signed-off-by: Frazer Carsley <frazer.carsley@arm.com>
Signed-off-by: Jon Mason <jon.mason@arm.com>
The patches added in this commit add a generic GPT library for use in
flash devices. Corstone1000 could use these to manage partitions during
a firmware update.
The patches are all backports from trusted-firmware-m (TF-M) main
branch and can be removed if Corstone1000 upgrades when the next version
of TF-M is released.
Signed-off-by: Frazer Carsley <frazer.carsley@arm.com>
Signed-off-by: Jon Mason <jon.mason@arm.com>
Split off the main portion of the SCP firmware recipe into an include
file and create a git versioned recipe. This allows for building and
testing the latest sources.
Signed-off-by: Jon Mason <jon.mason@arm.com>
New versions of the FVPs have been released. Update the recipes to use
them. License SHA changes caused by 3rd party software versions used
being modified and white space changes.
Signed-off-by: Jon Mason <jon.mason@arm.com>
Massive layout changes inside the FVP tarball. The files are not
located in *_FVP anymore, including the license files, which were in 2
locations in previous releases. The tarball changed from
Linux64/Linux64_armv8l to Linux_x86/Linux_armv8. It hchanged extention
from tgz to tar.gz
Also, adding skip of dev-so, due to libstdc++.so and libsystemc.so being
symlinks and fvps being a binary.
NOTE: the license file changes are due to version bumps for sdl, ffmpeg,
libvpx, and python. The license changes in the managements utilities
was version changes in clang and golang, and removal of wxWidgets,
libstdc++-6.dll, libgcc_s_seh-1.dll, and libwinpthread-1.dll
Signed-off-by: Jon Mason <jon.mason@arm.com>
To improve portability, testing coverage, and future platform enablement.
- Replace FVP-only multicore checks with platform-generic checks.
- Add the corresponding TF-M patch to the Corstone-1000 recipe.
Signed-off-by: Alex Chapman <alex.chapman@arm.com>
Signed-off-by: Jon Mason <jon.mason@arm.com>
To improve portability, testing coverage, and future platform enablement.
- Replace FVP-only multicore guards with platform-generic guards.
- Add the corresponding TF-A patch to the Corstone-1000 recipe.
Signed-off-by: Alex Chapman <alex.chapman@arm.com>
Signed-off-by: Jon Mason <jon.mason@arm.com>
To improve portability, testing coverage, and future platform enablement.
- Gate multicore on `MACHINE_FEATURES += "corstone1000_smp"`.
- Change recipe overrides from `:corstone1000-fvp` to `:corstone1000`.
- Update the Corstone-1000 multicore kas/doc references.
Signed-off-by: Alex Chapman <alex.chapman@arm.com>
Signed-off-by: Jon Mason <jon.mason@arm.com>
The recipe previously only copied the Python BaseTools into the native
sysroot, which was sufficient for capsule signing but not for Firmware
Volume (FV) creation. Downstream recipes that build UEFI capsule
update images for embedded platforms need GenFfs and GenFv to assemble
FV images - a key component of the UEFI Capsule Update format used for
system firmware updates.
Add the brotli submodule source (build-time dependency of both tools),
build the C BaseTools (Common, BrotliCompress, GenFfs, GenFv) and
install the binaries into ${bindir} together with GenerateCapsule.py
and the Common Python library so that capsule-generating recipes can
consume them from the native sysroot.
Signed-off-by: Igor Opaniuk <igor.opaniuk@foundries.io>
Signed-off-by: Jon Mason <jon.mason@arm.com>
The oe-core master branch just switched from whinlatter to wrynose[1] so
follow this change in our layers.
[1] oe-core cd1179544d7 ("layer.conf: Update to wrynose")
Signed-off-by: Ross Burton <ross.burton@arm.com>
Add the corstone1000 patch that inserts a compiler barrier before the
first HOST_AO_LOCK_BITS write in CC_LibInit().
On corstone1000-mps3 with GCC 15.x, TF-M can HardFault on that first
AO lock register update. Adding the barrier avoids the fault.
Signed-off-by: Michael Safwat <michael.safwat@arm.com>
Signed-off-by: Jon Mason <jon.mason@arm.com>
Compiles of edk2-firmware with GCC 15 on the host will fail:
main.c: In function ‘ProcessArgs’:
| main.c:163:42: error: too many arguments to function ‘p->process’; expected 0, have 2
| 163 | (*p->process)( *argv, *(argv+1) );
| | ~^~~~~~~~~~~~ ~~~~~
We actually already had the fix for this, but were doing SRC_URI +=
_before_ the include file did SRC_URI =, so the patch was never applied.
Move the require to the top of the recipe so this ordering problem does
not happen, and fix the line-endings in the patch that never got applied.
[ YOCTO #16116 ]
Signed-off-by: Ross Burton <ross.burton@arm.com>
Signed-off-by: Jon Mason <jon.mason@arm.com>
If the sstate is being mounted into the container then the hashserver
also needs to be shared, and not stored inside the build tree. Mark
BB_HASHSERVE as being set via an environment variable so the GitLab
runner environment can set the correct location.
This fixes a long-standing problem causing substandard reuse, which now
is detected and causes a warning in oe-core[1].
[1] oe-core 491de0db64a ("sanity.bbclass: warn when sstate is outside of
build dir, but hash equiv database is inside it")
Signed-off-by: Ross Burton <ross.burton@arm.com>
Signed-off-by: Jon Mason <jon.mason@arm.com>
When the binary toolchain support was removed[1] we no longer used this
variable, but a few instances of it was left behind.
[1] meta-arm 03af0c72f1 ("arm-toolchain: remove external-arm-toolchain")
Signed-off-by: Ross Burton <ross.burton@arm.com>
Signed-off-by: Jon Mason <jon.mason@arm.com>
Do the changes necessary to get qemuarm64-secureboot to work with edk2
firmware, and add it to CI. The CI changes needed to make it dynamic
based on edk2.yml or u-boot.yml required moving the relevant parts into
inc files.
Signed-off-by: Jon Mason <jon.mason@arm.com>
edk2 has been broken on qemuarm64 for an unknown amount of time. Add it
to CI to prevent this from happening (until edk2 works on
qemuarm64-secureboot).
Signed-off-by: Jon Mason <jon.mason@arm.com>
Make the relevant correction for the following warning in qemu:
warning: short-form boolean option 'readonly' deprecated
Please use readonly=on instead
Signed-off-by: Jon Mason <jon.mason@arm.com>
Frazer Carsley
Gyorgy Szing
Roger Knecht
Kevin Hao
Ross Burton
Michael Safwat
Jose Quaresma
Alex Chapman
Jon Mason
Igor Opaniuk