Update Corstone-1000 U-Boot device tree for the Cortex-A320 variant
and enable GICv3/GIC-600, while keeping compatibility with the
existing GIC-400 setup. A single DT image now supports either
configuration via Kconfig guards.
**Device-tree updates (Cortex-A320)**
* Map Ethos-U85 NPU registers at `0x1A050000` (16 KiB) and its SRAM at
`0x02400000` (2 MiB, no-map), plus a 32 MiB DDR carve-out for DMA.
* Add `/ethosu@1a050000` with interrupts, `dma-ranges`, `cs-region`,
and `ethosu-mem-config` for driver probe.
* Guard the NPU node behind `CONFIG_ETHOS_U85`.
* Add a Cortex-A320 compatible string to the Corstone-1000 DTS
downstream.
**GICv3/GIC-600 selection**
* Introduce `CONFIG_GIC_V3` to select the new interrupt controller.
* Add a full GICv3/GIC-600 node guarded by `#ifdef CONFIG_GIC_V3`.
* When GICv3 is enabled, set `cpu@1..3` `reg` to `0x100/0x200/0x300`
(retain `0x1/0x2/0x3` for GIC-400).
* Update the Ethos-U85 interrupt to **SPI 16** to match the interrupt
map.
Signed-off-by: Frazer Carsley <frazer.carsley@arm.com>
Signed-off-by: Harsimran Singh Tungal <harsimransingh.tungal@arm.com>
Signed-off-by: Jon Mason <jon.mason@arm.com>
Enable Trusted Firmware-A for Corstone-1000 platforms with Cortex-A320 and
switch the interrupt controller from GICv2/GIC-400 to GICv3/GIC-600.
**Platform/feature enablement**
* Map Ethos-U85 NPU registers (`0x1A050000`, 16 KiB) and its SRAM region
(`0x02400000`, 4 MiB) into Normal World
(`MT_DEVICE | MT_RW | MT_NS` / `MT_MEMORY | MT_RW | MT_NS`).
* Force Cortex-A320 feature selection: enable Armv9 features, disable
Cortex-A35 errata, and select the `cortexa320` override in
`trusted-firmware-a-corstone1000.inc`.
* Build TF-A-Tests with `CORSTONE1000_CORTEX_A320=1` to skip non-applicable
FF-A, PSCI, and CPU-extension tests on Cortex-A320.
**GICv3/GIC-600 transition (A320 builds)**
* Update `plat_my_core_pos()` and `plat_arm_calc_core_pos()` to compute the
linear core position using the Cortex-A320 MPIDR_EL1 affinity layout.
* Add an A320-specific core-position routine in assembly, guarded by
`CORSTONE1000_CORTEX_A320`.
* Switch to the GICv3 driver with GIC-600 extensions:
* Update platform GIC base addresses to the GIC-600 layout.
* Use GICv3 APIs; set `USE_GIC_DRIVER=3`, `GICV3_SUPPORT_GIC600=1`,
and `GIC_ENABLE_V4_EXTN=1`.
* Keep conditional GIC versioning so Cortex-A35 continues to use GICv2/GIC-400.
These changes ensure correct GIC configuration and reliable secondary-core
bring-up on Cortex-A320 while preserving existing Cortex-A35 behavior.
Signed-off-by: Harsimran Singh Tungal <harsimransingh.tungal@arm.com>
Signed-off-by: Frazer Carsley <frazer.carsley@arm.com>
Signed-off-by: Michael Safwat <michael.safwat@arm.com>
Signed-off-by: Jon Mason <jon.mason@arm.com>
Enable full Corstone‑1000 Cortex‑A320 DSU‑120T platform support in TF‑M:
- Reserve a 4 MiB Host SRAM region at 0x0240_0000 for the
Cortex‑A320 normal world and open it in the CVM firewall
(region 2), gated by `CORSTONE1000_CORTEX_A320``.
- Introduce a DSU‑120T Power-Policy Unit driver plus a
`CORSTONE1000_DSU_120T` CMake option to power on the Cortex‑A320
host cluster with proper secure-enclave firewall and memory-map
setup.
- Add a CMake platform define that auto‑activates when the
`cortexa320` machine feature is present, injecting
DSU‑120T‑specific compile definitions.
Signed-off-by: Harsimran Singh Tungal <harsimransingh.tungal@arm.com>
Signed-off-by: Jon Mason <jon.mason@arm.com>
Update the OP-TEE OS build logic to detect `MACHINE_FEATURES` and
append the appropriate `arm64-platform-cpuarch` value to
`EXTRA_OEMAKE`, instead of hard-coding `cortex-a35`.
This change ensures that when `MACHINE_FEATURES` includes
`cortexa320`, the OP-TEE build receives the matching `core-arch` flag,
while maintaining `cortex-a35` as the default.
The new Corstone-1000 variant with Cortex-A320 replaces the original
GIC-400 (v2) interrupt controller with a GIC-600, which is
architecturally compliant with GICv3. Since OP-TEE already provides
a generic GICv3 driver, only minimal platform changes are needed
to expose the updated register map and initialize the GICv3 interface.
Signed-off-by: Hugues KAMBA MPIANA <hugues.kambampiana@arm.com>
Signed-off-by: Harsimran Singh Tungal <harsimransingh.tungal@arm.com>
Signed-off-by: Jon Mason <jon.mason@arm.com>
Enable the Corstone‑1000 Cortex‑A320 variant by:
- Introduce `machine/include/corstone1000-a320.inc` to configure the
default Ethos‑U MAC count when `cortexa320` is in MACHINE_FEATURES,
and allow override via `ETHOSU_NUM_MACS`.
- Add a KAS profile at `kas/corstone1000-a320.yml` for Cortex‑A320
FVP-based builds.
- Extend corstone1000.inc to detect MACHINE_FEATURES (cortexa320) and
pull in the matching tune-<core>.inc (default still Cortex-A35).
- Add the `meta-ethos` layer as a dependency of `meta-arm-bsp` for
Cortex‑A320 builds and define a new KMachine override to pull in
the Ethos‑U driver recipe.
- In `conf/machine/corstone1000-fvp.conf`, inspect `MACHINE_FEATURES`
and set `FVP_EXE` to `FVP_Corstone-1000_with_Cortex-A320` when
`cortexa320` is enabled, otherwise fall back to `FVP_Corstone-1000`.
- In `recipes-devtools/fvp/fvp-corstone1000.bb`, add a
`SRC_URI:cortexa320` entry (with checksums) for the Cortex‑A320 FVP
build archive.
- Disable the rootfs CPIO file compression so it is not compressed
twice when bundled with the kernel
Signed-off-by: Michael Safwat <michael.safwat@arm.com>
Signed-off-by: Frazer Carsley <frazer.carsley@arm.com>
Signed-off-by: Hugues KAMBA MPIANA <hugues.kambampiana@arm.com>
Signed-off-by: Jon Mason <jon.mason@arm.com>
When a recipe uses the externalsrc class, the do_patch task is
skipped entirely as specified in SRCTREECOVEREDTASKS.
Since do_apply_local_src_patches function is registered as a postfuncs,
it would never run in that specific case.
This cause recipes relying on do_apply_local_src_patches to miss the
local source patching when built from external source tree.
To address the issue, schedule a new task after the do_patch and before
the do_configure, ensuring the local patching executes regardless of
whether do_patch was skipped by externalsrc.
Signed-off-by: Clement Faure <clement.faure@arm.com>
Signed-off-by: Jon Mason <jon.mason@arm.com>
Whilst TF-M 2.1.1 doesn't use this, setting the variable doesn't have
any negative effects and consolidates the external module assignments.
Signed-off-by: Ross Burton <ross.burton@arm.com>
Signed-off-by: Jon Mason <jon.mason@arm.com>
We now use Ninja to build TF-M[1], so setting CMAKE_VERBOSE_MAKEFILES
doesn't do anything.
We have arm-none-eabi-gcc 13.3[2], so there's no need to remove options
that <13 don't support.
[1] meta-arm 018fd6aecf ("arm/trusted-firmware-m: use Ninja to build")
[2] meta-arm f646ee4507 ("arm-toolchain: update to 13.3")
Signed-off-by: Ross Burton <ross.burton@arm.com>
Signed-off-by: Jon Mason <jon.mason@arm.com>
As the Poky layer will no longer be updated following
the integration of `bitbake-setup`, developers are advised to
use a combination of the `bitbake` and `openembedded-core`
layers instead of the `poky` layer.
Note that the `poky` layer is a combination of these two
layers glued into a single repository for convenience.
Signed-off-by: Hugues KAMBA MPIANA <hugues.kambampiana@arm.com>
Signed-off-by: Jon Mason <jon.mason@arm.com>
We don't run or package the tests, so there's no point to building them.
Signed-off-by: Ross Burton <ross.burton@arm.com>
Signed-off-by: Jon Mason <jon.mason@arm.com>
We depend on native tools to provide these binaries, so we can delete
them to ensure that our tools are always used and never the prebuilt.
Signed-off-by: Ross Burton <ross.burton@arm.com>
Signed-off-by: Jon Mason <jon.mason@arm.com>
Split configure/compile and invoke ninja directly so that we can
control parallelisation.
Signed-off-by: Ross Burton <ross.burton@arm.com>
Signed-off-by: Jon Mason <jon.mason@arm.com>
GICC registers are not defined for GICv3. Trusted-Firmware-A throws error when
GICC register address is not defined even for GICv3. Adding patch
to handle this in Trusted-Firmware-A.
Signed-off-by: Harsimran Singh Tungal <harsimransingh.tungal@arm.com>
Signed-off-by: Jon Mason <jon.mason@arm.com>
ARMmbed/mbedtls is the old name and redirects to Mbed-TLS/mbedtls, use
the correct name to avoid the redirection.
Signed-off-by: Ross Burton <ross.burton@arm.com>
Signed-off-by: Jon Mason <jon.mason@arm.com>
- Introduce new file `corstone1000-extsys.inc` to define variables related
to the external system.
- Ensure this file is included only when MACHINE_FEATURES do contain
corstone1000-extsys.
This change makes external system configuration modular and only applied
when explicitly enabled through machine features.
Signed-off-by: Michael Safwat <michael.safwat@arm.com>
Signed-off-by: Jon Mason <jon.mason@arm.com>
The image wasn’t autoloading kernel modules even though
`KERNEL_MODULE_AUTOLOAD` populated `/etc/modules-load.d/`. In this
configuration `/etc/init.d/rcS` only executes runlevel scripts from
`/etc/rcS.d` (and `/etc/rc5.d`), and `modutils.sh` was also missing.
This change:
* Includes the loader by adding `modutils-initscripts` to
`CORE_IMAGE_EXTRA_INSTALL`.
* Enables SysV init by appending `sysvinit` to `DISTRO_FEATURES`,
ensuring the `S*` start links in `/etc/rcS.d` (and `/etc/rc5.d`)
call `/etc/init.d/modutils.sh start` during boot.
**Result:** entries in `/etc/modules-load.d/*.conf` now load
automatically at boot.
**Verification**
* Before: `lsmod` empty after boot; manual `modprobe` needed.
* After: `lsmod` shows target modules; `dmesg` contains module init logs.
Signed-off-by: Hugues KAMBA MPIANA <hugues.kambampiana@arm.com>
Signed-off-by: Jon Mason <jon.mason@arm.com>
Switch the *reserved-memory* node from two-cell (64-bit) encoding to
one-cell (32-bit) encoding and adjust the `reg` property accordingly
to make reserved-memory node format compatible with rest of the dts.
Signed-off-by: Harsimran Singh Tungal <harsimransingh.tungal@arm.com>
Signed-off-by: Jon Mason <jon.mason@arm.com>
The Fedora 39 artifacts have been moved to an /archive/ directory, so
update the SRC_URI to match.
Signed-off-by: Ross Burton <ross.burton@arm.com>
Signed-off-by: Jon Mason <jon.mason@arm.com>
git.trustedfirmware.org is an alias for review.trustedfirmware.org. We
moved the main recipe to use review.trustedfirmware.org last year[1]
but not all other recipes that fetch the source followed, which means
that we have to fetch TF-A multiple times.
This commit ensures that all the recipes are using the same SRC_URI, so
we just fetch TF-A once.
[1] a6a4952e ("arm/trusted-firmware-a: use correct git URL")
Signed-off-by: Ross Burton <ross.burton@arm.com>
Signed-off-by: Jon Mason <jon.mason@arm.com>
tfm_sign_image.bbclass hard codes the image security counter, which is
generated from the image version x.y.z.
The generated image security counter value is huge if x or y > 0.
Platform security counter store may not support such a huge counter
value.
Introduce a variable RE_WRAPPER_SECURITY_COUNTER to enable platforms to
specify the actual image security counter.
Signed-off-by: David Hu <david.hu2@arm.com>
Signed-off-by: Jon Mason <jon.mason@arm.com>
This job takes a few minutes and isn't useful unless it's being ran for
master, or is being actively worked on.
Signed-off-by: Ross Burton <ross.burton@arm.com>
Signed-off-by: Jon Mason <jon.mason@arm.com>
Reduce the number of tests being run in CI to reduce the amount of time
it takes to complete, while providing the same code coverage. Internal
CI runs went from 2.5h to 1.5h.
Signed-off-by: Jon Mason <jon.mason@arm.com>
It turns out that the base SRCREV for trusted-services is a tag name,
which meant it was hitting the network on every build. Use the SHA
instead.
Signed-off-by: Ross Burton <ross.burton@arm.com>
Signed-off-by: Jon Mason <jon.mason@arm.com>
The SMAIL_GPL license in oe-core was renamed SMAIL-GPL to match SPDX.
Update the recipe to match this.
Signed-off-by: Ross Burton <ross.burton@arm.com>
Signed-off-by: Jon Mason <jon.mason@arm.com>
This bbappend was _appending_ to FILESEXTRAPATHS but putting the colon
separator _after_, so it actually constructed an invalid path.
Change the assignment to be prepend, so the separator is in the right
place.
Signed-off-by: Ross Burton <ross.burton@arm.com>
Signed-off-by: Jon Mason <jon.mason@arm.com>
trusted-firmware a has a compile error when building with clang. Since
this platform is EOL'ed and we're not currently building this platform
with clang in CI, the best option is to force GCC for it.
Signed-off-by: Jon Mason <jon.mason@arm.com>
Hopefully this issue can be fixed in a newer release. Move the
toolchain forcing to the versioned so that it can be tracked easier.
Signed-off-by: Jon Mason <jon.mason@arm.com>
Now that clang is in core, we don't need to use meta-clang anymore.
Also, use PREFERRED_TOOLCHAIN_TARGET to specify the toolchain to use.
Signed-off-by: Jon Mason <jon.mason@arm.com>
The u-boot recipe now uses unique build directories per config that is
being built, to ensure that there is no cross-contamination.
Handle this by moving the do_configure and do_install appends to
uboot_configure_config and uboot_install_config so that we can simply
use $builddir.
[1] oe-core 22e96b32b0b ("u-boot: Make sure the build dir is unique for each UBOOT_CONFIG")
Signed-off-by: Ross Burton <ross.burton@arm.com>
Since TF-M v2.2.1, the new crypto driver used does not support RSA
algorithms, so these tests are no longer valid.
Signed-off-by: Frazer Carsley <frazer.carsley@arm.com>
Signed-off-by: Ross Burton <ross.burton@arm.com>
Re-adds a patch aligning the Trusted Services PSA Crypto
structure with its equivalent definition in TF-M v2.1.1.
The patch was previously removed during the upgrade to
Trusted Services v1.2, as it was believed to be included in that
version. However, the alignment is still required to maintain
consistency with TF-M v2.1.1.
Signed-off-by: Frazer Carsley <frazer.carsley@arm.com>
Signed-off-by: Ross Burton <ross.burton@arm.com>
We have 2.2.1, so there's no need for 2.2.0. This removal was accidentally
omitted from the 2.2.1 upgrade.
Signed-off-by: Ross Burton <ross.burton@arm.com>
The OESSHTarget object now has a ignore_ssh_fails keyword argument[1],
so update this subclass to match.
As the implementation of run() here simply forwards the arguments, we
can use *args, **kwargs so that future changes don't cause problems.
[1] oe-core afe118d4f2d ("oeqa: target: ssh: Fail on SSH error even when errors are ignored")
Signed-off-by: Ross Burton <ross.burton@arm.com>
Moving forwards, it's expected that the poky repository will no longer be
updated as the integration of bitbake-setup means that users are
encouraged to use bitbake+oe-core separately instead.
We also need to fetch meta-yocto as our CI is currently explicitly based
on the poky distribution.
This is effectively a no-op change, as poky is simply these component
repositories glued into a single repository for convenience.
Signed-off-by: Ross Burton <ross.burton@arm.com>
Signed-off-by: Jon Mason <jon.mason@arm.com>
Backport a patch from upstream to fix the build of antlr with GCC 15 on
the host.
The build still fails with clang, so update the message with the current
error message.
Signed-off-by: Ross Burton <ross.burton@arm.com>
Signed-off-by: Jon Mason <jon.mason@arm.com>
Backport Trusted Firmware-A patch to re-enable the BL31
console during early boot.
Signed-off-by: Michael Safwat <michael.safwat@arm.com>
Signed-off-by: Jon Mason <jon.mason@arm.com>
Now that clang is part of oe-core we can't use meta-clang being present
as an indicator of clang being available.
This does mean we can clean up the logic and just use TOOLCHAIN, as that
is always set now.
Signed-off-by: Ross Burton <ross.burton@arm.com>
Signed-off-by: Jon Mason <jon.mason@arm.com>
Upstream appear to have resolved the libgcc linkage issues as these
variables are not used in upstream nor our patches.
Signed-off-by: Ross Burton <ross.burton@arm.com>
Signed-off-by: Jon Mason <jon.mason@arm.com>
Kernel builds are logging the following issue:
WARNING: linux-yocto-6.16.8+git-r0 do_kernel_configcheck: [kernel config]: specified values did not make it into the kernel's final configuration:
[NOTE]: 'CONFIG_BT_LEDS' last val (y) and .config val (n) do not match
This could be enabled by setting:
CONFIG_LEDS_CLASS=y
CONFIG_NEW_LEDS=y
CONFIG_SND_SOC_HDA=y
But this isn't really useful on the juno platform. So, better to
disable the BT LEDs.
Signed-off-by: Jon Mason <jon.mason@arm.com>
