fix(marvell): provide a default compile target

The DDR drivers require the `CROSS_COMPILE` prefix to be set. Platform
builds break if a user builds for Marvell without explicitly providing
this. Thi

fix(marvell): provide a default compile target

The DDR drivers require the `CROSS_COMPILE` prefix to be set. Platform
builds break if a user builds for Marvell without explicitly providing
this. This is hidden in CI because we always set the target. Provide a
default as a workaround.

Change-Id: Id93d71876a81e442e6612eae2b0e7cb12e0fd55a
Signed-off-by: Harrison Mutai <harrison.mutai@arm.com>

show more ...

fix(a8k): declare amb_memory_map as an array

The DDR files used in CI were recently updated. They introduce a
requriement for GCC 14 and new flags into CFLAGS, including
`-Wsizeof-pointer-div`. This

fix(a8k): declare amb_memory_map as an array

The DDR files used in CI were recently updated. They introduce a
requriement for GCC 14 and new flags into CFLAGS, including
`-Wsizeof-pointer-div`. This results in the following compiler warning
when building for this platform:

```
plat/marvell/armada/a8k/a70x0_amc/board/marvell_plat_config.c: In
function 'marvell_get_amb_memory_map': include/lib/utils_def.h:15:20:
error: division 'sizeof (struct addr_map_win *) / sizeof (struct
addr_map_win)' does not compute the number of array elements
[-Werror=sizeof-pointer-div] 15 | (sizeof(a) / sizeof((a)[0])) | ^
plat/marvell/armada/a8k/a70x0_amc/board/marvell_plat_config.c:30:25:
note: in expansion of macro 'ARRAY_SIZE'
```

The warning occurs because ARRAY_SIZE() was used on a pointer
(amb_memory_map) instead of an actual array. In that case the expression
becomes sizeof(pointer) / sizeof(element), which does not represent the
number of elements and therefore triggers the compiler warning
-Wsizeof-pointer-div (treated as an error due to -Werror). Fix the
declaration of amb_memory_map to pacify the compiler.

Change-Id: I5608be6401bfbc1291ded85a8fbce8abe169dbe2
Signed-off-by: Harrison Mutai <harrison.mutai@arm.com>

show more ...

fix(marvell): work around uutils coreutils truncate -s %SIZE bug

The boot-image.bin assembly uses "truncate -s %128K" to pad bl1.bin
to a 128 KiB boundary before appending the FIP. This ensures the

fix(marvell): work around uutils coreutils truncate -s %SIZE bug

The boot-image.bin assembly uses "truncate -s %128K" to pad bl1.bin
to a 128 KiB boundary before appending the FIP. This ensures the FIP
starts at offset 0x20000, which is where BL1 expects it at runtime
(PLAT_MARVELL_FIP_BASE = PLAT_MARVELL_ATF_LOAD_ADDR + 0x20000).

uutils coreutils (Rust rewrite of GNU coreutils), shipped as the
default coreutils on Ubuntu 25.04+, has a bug in its truncate
implementation of the "%" (round up to multiple) operator.

The RoundUp formula at line 71 of src/uu/truncate/src/truncate.rs
in uutils coreutils 0.5.0 is:

fsize + fsize % size

The correct formula (used by GNU coreutils and fixed on uutils main
branch) is:

checked_next_multiple_of(fsize, size)

which is equivalent to: fsize + (size - fsize % size) % size

Example with bl1.bin (24696 bytes) and "truncate -s %128K":

GNU coreutils: 24696 + (131072 - 24696 % 131072) % 131072
= 24696 + 106376 = 131072 (0x20000) -- correct

uutils 0.5.0: 24696 + 24696 % 131072
= 24696 + 24696 = 49392 (0xC0F0) -- wrong

As a result, the FIP is placed at offset 0xC0F0 instead of 0x20000
in boot-image.bin. BL1 then fails at runtime with:

WARNING: Firmware Image Package header check failed.
WARNING: Failed to obtain reference to image id=1 (-2)
ERROR: Failed to load BL2 firmware.

The workaround detects the bug at make parse time by creating a
1-byte test file and checking whether "truncate -s %128K" produces
131072 bytes. If not, it falls back to explicit numeric sizes.

Reference: uutils coreutils 0.5.0 buggy source:
https: //github.com/uutils/coreutils/blob/0.5.0/src/uu/truncate/src/truncate.rs#L71
Change-Id: I208a7ca7ca4c113817969935c26bd0c7f0207d7d
Signed-off-by: Vincent Jardin <vjardin@free.fr>

show more ...

fix(a8k): add XFI params for NBX SFI 10G

Add XFI static values for CP0 and CP1 Comphy4 lanes which connect to
the SFP+ cages on the NBX board. The tuning values for a80x0
configures the SerDes equal

fix(a8k): add XFI params for NBX SFI 10G

Add XFI static values for CP0 and CP1 Comphy4 lanes which connect to
the SFP+ cages on the NBX board. The tuning values for a80x0
configures the SerDes equalizer, TX amplitude, pre-emphasis and RX filter
coefficients for 10.3125 Gbps operation of both 10G ports of the CPU.

Change-Id: Ic4f0fb6331c504f98cb7e832848dd3481931dbb7
Signed-off-by: Vincent Jardin <vjardin@free.fr>

show more ...

fix(a8k): mv_ddr path may not be a git repo

When the mv-ddr-marvell folder is not a git tree the build fails
complaining that it "does not contain valid mv-ddr-marvell git
repository".

For instance

fix(a8k): mv_ddr path may not be a git repo

When the mv-ddr-marvell folder is not a git tree the build fails
complaining that it "does not contain valid mv-ddr-marvell git
repository".

For instance, it shall be removed when Building using Buildroot.

Change-Id: Ieaff6281785993873c7cb26ddc1d90d0896ad1b7
Signed-off-by: Vincent Jardin <vjardin@free.fr>

show more ...

feat(a8k): add a80x0_nbx Free Mobile board

Add TF-A platform support for the a80x0_nbx board (Free Mobile
Nodebox10G), a network appliance based on the Marvell Armada 8040
SoC with dual CP110 compan

feat(a8k): add a80x0_nbx Free Mobile board

Add TF-A platform support for the a80x0_nbx board (Free Mobile
Nodebox10G), a network appliance based on the Marvell Armada 8040
SoC with dual CP110 companion processors.

Hardware configuration:
- Quad-core ARM Cortex-A72 @ 1.3GHz
- DDR4 memory with ECC support (single channel, 32-bit)
- Dual CP110 companion processors (CP0 and CP1)
- SGMII 1G Ethernet on CP0 lane 5
- I2C buses for peripheral access (CP0: 100kHz, CP1: 400kHz)
- NS16550 UART console at 115200 baud
- eMMC boot via Xenon SDHCI controller

Key features implemented:

1. Ramoops buffer preservation across cold boot (ramoopsies driver)
On ECC-enabled DDR configurations, the memory controller must
scrub all memory during initialization to establish valid parity
bits. This would normally destroy the Linux kernel ramoops buffer
containing crash logs from the previous boot. The ramoopsies
driver intercepts the DDR scrubbing function using the GNU linker
--wrap feature, saving the 32KB ramoops buffer (at 0x3FFF8000) to
SRAM before scrubbing and restoring it afterward. The driver also
handles pending SError exceptions that occur when reading
uninitialized ECC memory by installing a minimal exception vector
that acknowledges and clears them.

2. UART-based skip image recovery mechanism
The platform uses the USER_DEFINED skip image detection callback
to implement software-based boot recovery without requiring a
dedicated GPIO button. Users can trigger recovery mode by holding
the 's' key during early boot, causing the bootloader to load
from the secondary/recovery image instead of the primary firmware.

3. SerDes/ComPhy configuration
PHY porting layer configured with default tuning values for
XFI (10G) and SATA interfaces, following the principle of minimal
bootloader configuration. Advanced SerDes tuning is deferred to
the Linux kernel where it can be more easily adjusted.

The implementation includes comprehensive documentation and a build
verification script to validate proper symbol exports, DDR driver
integration, and flash image format compliance.

Based on original work by Nicolas Schichan <nschichan@freebox.fr>
for the ramoops preservation mechanism and USER_DEFINED skip image
detection concept.

Change-Id: Iaacbb29631f27b47fbf5cc300d8c63aaf1e89e51
Signed-off-by: Vincent Jardin <vjardin@free.fr>

show more ...

feat(a8k): add user callback for skip_image

Add a user_defined.check function pointer to struct skip_image,
allowing platforms to implement custom skip image detection logic.
When set and returning

feat(a8k): add user callback for skip_image

Add a user_defined.check function pointer to struct skip_image,
allowing platforms to implement custom skip image detection logic.
When set and returning non-zero, BLE signals the bootloader to load
the recovery image.

This enables boards without dedicated recovery GPIOs to implement
software-based recovery triggers (e.g., UART key press detection).

Original-concept-by: Nicolas Schichan <nschichan@freebox.fr>
Change-Id: Ib58033fb1440f701dab027d77505f2582e3d98da
Signed-off-by: Vincent Jardin <vjardin@free.fr>

show more ...

fix(build): prevent races on the build directory

The tools (eg fiptool) don't depend on the build directory so it's
possible that make tries to build them before it exists. Doing that
leads to rando

fix(build): prevent races on the build directory

The tools (eg fiptool) don't depend on the build directory so it's
possible that make tries to build them before it exists. Doing that
leads to random and unpredictable errors. Almost always, they are built
after a BL image which always has a build directory dependency, but
when building MANY tf-a builds over MANY threads concurrently this could
be observed (I suspect the high load of the system increases latency
just enough that this race is lost).

The fix is simple - have an explicit dependency on the build directory.

This is the same problem as 9855568cc and 25cde5f81.

Change-Id: I769ac07f8882f82ea9d72f3b976337284d697310
Signed-off-by: Boyan Karatotev <boyan.karatotev@arm.com>

show more ...

refactor(platforms): remove usage of psci_power_down_wfi

The callers in PSCI will now call wfi() after calling the
pwr_domain_pwr_down() platform hook (previously they were mutually
exclusive). As s

refactor(platforms): remove usage of psci_power_down_wfi

The callers in PSCI will now call wfi() after calling the
pwr_domain_pwr_down() platform hook (previously they were mutually
exclusive). As such, there is no need for platforms to do this
themselves. In fact, it is strongly advisable for platforms not to do
this themselves so that the PSCI can apply any relevant errata
mitigations.

Change-Id: Ic9702edef7503996be24d984e29dfcc13bc3fcba
Signed-off-by: Boyan Karatotev <boyan.karatotev@arm.com>

show more ...

build(marvell): avoid using recursive expansion for BLE_INCLUDES

Commit 559ab2df4 specifies that there should be a PLAT_INCLUDES
specifically for the BLE image. This is done with a quirky rule that

build(marvell): avoid using recursive expansion for BLE_INCLUDES

Commit 559ab2df4 specifies that there should be a PLAT_INCLUDES
specifically for the BLE image. This is done with a quirky rule that
triggers only on the BLE files and relies on PLAT_INCLUDES expanding
recursively to work correctly.

Recursively expanded variables are a brittle part of the build system,
so this patch removes that aspect but keeps the BLE_INCLUDES principle
via the BLE_INCLUDE_DIRS variable that MAKE_C will correctly pick up.

Change-Id: Ibd10904f6a005fef9ed242b65e96662ce002eb18
Signed-off-by: Boyan Karatotev <boyan.karatotev@arm.com>

show more ...

feat(marvell): add trng driver

Armada-7K/8K and CN913x integrated the Rambus EIP-97 IP on CP11x die. It
supports to generate up to 4 32-bit random number in one shot.

This trivial driver provisions

feat(marvell): add trng driver

Armada-7K/8K and CN913x integrated the Rambus EIP-97 IP on CP11x die. It
supports to generate up to 4 32-bit random number in one shot.

This trivial driver provisions a simple API to read the random numbers
from hardware. It allows the bootloader to get one 32-bit or 64-bit
random number via SMC call to support KASLR.

Change-Id: I1707a85512ca163b8c7ab1644ff0f7e2fcf57344
Signed-off-by: Wilson Ding <dingwei@marvell.com>

show more ...

revert(rambus-trng): remove ip-76 driver

The Rambus TRNG IP-76 driver was ported from Linux kernel (omap-rng.c),
which was initially licensed under GPL-2.0. In term of the license
violation, remove

revert(rambus-trng): remove ip-76 driver

The Rambus TRNG IP-76 driver was ported from Linux kernel (omap-rng.c),
which was initially licensed under GPL-2.0. In term of the license
violation, remove this driver and the related SMC call that originally
added by the following two commits:

commit 57660d9d7945 ("plat/marvell/armada/a8k: support HW RNG by SMC")
commit 6aa9f5d164e8 ("drivers/rambus: add TRNG-IP-76 driver")

Change-Id: Id8c99db2e51b49623b3b034106c989a46f690b60
Signed-off-by: Wilson Ding <dingwei@marvell.com>

show more ...

Merge changes I32bd0c71,I167e7398 into integration

* changes:
fix(arm): don't race on the build directory
fix(armada): don't race on the UART_IMAGE

fix(armada): don't race on the UART_IMAGE

UART_IMAGE is not set when WTP isn't. The error rules will then provide
a recipe for $(BUILD_PLAT). When building with a lot of cores (64) this
rule might b

fix(armada): don't race on the UART_IMAGE

UART_IMAGE is not set when WTP isn't. The error rules will then provide
a recipe for $(BUILD_PLAT). When building with a lot of cores (64) this
rule might be called before the directory is made, causing a build
failure.

Hoist the definition so that the depended path is correct.

Change-Id: I167e7398e576e667d0c5c1fc0f07ab8c8ef939a8
Signed-off-by: Boyan Karatotev <boyan.karatotev@arm.com>

show more ...