The AMD Versal Gen 2 PCIe host controller places its DBI registers
(0x100000000000, 1 MB) and config space (0x100000100000, 255 MB)
above 1 TB. Without MMU entries covering these ranges, U-Boot faults
when the PCIe driver accesses them.
The two regions are merged into a single entry as these are
contiguous and share identical MMU attributes. Add this
entry under a CONFIG_IS_ENABLED(PCIE_DW_AMD) guard so it is only
included when the PCIe driver is configured. VERSAL2_MEM_MAP_USED is
adjusted from 5 to 6 accordingly, keeping the DRAM bank index correct.
Signed-off-by: Pranav Sanwal <pranav.sanwal@amd.com>
Link: https://lore.kernel.org/r/20260327121015.996806-3-pranav.sanwal@amd.com
Signed-off-by: Michal Simek <michal.simek@amd.com>
Add support for the DesignWare-based PCIe host controller found in
AMD Versal2 SoCs. This enables PCIe functionality (e.g. NVMe storage)
on boards such as the VEK385.
The driver builds on the existing pcie_dw_common infrastructure and
adds Versal2-specific handling: it maps the SLCR register region to
mask and clear TLP interrupt status bits, parses dbi/config/atu/slcr
register regions from device tree, and supports an optional PERST#
GPIO on child nodes for endpoint reset sequencing. The outbound iATU
is programmed for the non-prefetchable memory window from device tree
ranges.
Signed-off-by: Pranav Sanwal <pranav.sanwal@amd.com>
Link: https://lore.kernel.org/r/20260327121015.996806-2-pranav.sanwal@amd.com
Signed-off-by: Michal Simek <michal.simek@amd.com>
When a clock parent entry is DUMMY_PARENT (0xFFFFFFFE), masking it
with CLK_PARENTS_ID_MASK (0xFFFF) produces the value 0xFFFE (65534).
This value is stored in parent->id and later used as a clock array
index in versal_clock_get_parentid(). Since clock_max_idx is
typically 228, accessing clock[65534] is out-of-bounds, and the
garbage value read is used as a clock ID in subsequent clock rate
calculations, eventually causing U-Boot to crash. This is observed
as a crash during "clk dump" on AMD Versal Gen 2.
Fix this by setting parent->id = 0 for DUMMY_PARENT entries.
Fixes: 95105089af ("clk: versal: Add clock driver support")
Signed-off-by: Padmarao Begari <padmarao.begari@amd.com>
Signed-off-by: Michal Simek <michal.simek@amd.com>
Link: https://lore.kernel.org/r/20260327101053.900154-1-padmarao.begari@amd.com
Set the XAE_RAF_BCSTREJ_MASK bit in the Reset and Address Filter (RAF)
register during hardware initialization to make the MAC drop incoming
frames with broadcast destination address. This avoids unnecessary
broadcast traffic processing by the CPU.
Additionally, when IPv6 is not enabled (CONFIG_IPV6), also set the
XAE_RAF_MCSTREJ_MASK bit to reject multicast frames. Multicast
rejection is skipped when IPv6 is configured because IPv6 Neighbor
Discovery and DHCPv6 rely on multicast.
Expose the RAF register (offset 0x0) in struct axi_regs which was
previously hidden in a reserved array.
Signed-off-by: Michal Simek <michal.simek@amd.com>
Link: https://lore.kernel.org/r/07ad94eb8a3a9d80273a16a7ac8c7caaba607fe2.1774282153.git.michal.simek@amd.com
Simon Glass <sjg@chromium.org> says:
This series reworks the FIT test (test_fit.py) to make it easier to
maintain and extend. It fixes the buildconfigspec so the test actually
runs on sandbox, avoids unnecessary U-Boot restarts, renames the main
test for easier selection, adds a missing-kernel check, fixes Python
warnings, converts the test to use a class, splits into separate test
functions, and adds Sphinx documentation.
Link: https://lore.kernel.org/r/20260413130007.1987326-1-sjg@chromium.org
Add a Sphinx automodule page for test_fit.py so it appears in the pytest
documentation alongside the other test modules. The index already uses a
glob pattern, so this is picked up automatically.
Signed-off-by: Simon Glass <sjg@chromium.org>
Create an 'fsetup' fixture which sets up files and parameters, and a
prepare() helper which builds a FIT with given parameter overrides.
Update check_equal() and check_not_equal() to look up filenames from a
params dict by key, reducing the number of local variables needed.
Split the single test_fit_operations() into individual test functions so
that each appears separately in the results.
Signed-off-by: Simon Glass <sjg@chromium.org>
Move this test over to use a class instead of a function, so we can
easily split it into some related tests.
Move the TODO so it is part of the comment for the class.
Tidy up some function comments while we are here.
Signed-off-by: Simon Glass <sjg@chromium.org>
The current name is uses as a root name by a few other tests, such as
test_fit_ecdsa() which makes it hard to run just this test.
Rename it to test_fit_base()
Signed-off-by: Simon Glass <sjg@chromium.org>
We don't actually need to use the test FDT as the control FDT. It slows
down the test since U-Boot needs to be restarted each time.
Instead of restarting, update the test to clear memory before it loads
the FIT.
Rename the data-variable to fdt_data since is it no-longer the control
FDT.
Signed-off-by: Simon Glass <sjg@chromium.org>
The test_fit test requires buildconfigspec('fit_signature') but does
not exercise signature verification. The ITS template includes a
signature-1 node, but mkimage only needs TOOLS_FIT_SIGNATURE (a host
tool option) to handle it, not CONFIG_FIT_SIGNATURE in the U-Boot
binary.
Since sandbox does not enable CONFIG_FIT_SIGNATURE, the test is silently
skipped on every run. Change the marker to buildconfigspec('fit') which
is all the test actually needs.
Signed-off-by: Simon Glass <sjg@chromium.org>
Currently there is no possibility to flush stdin after autocommands are
executed. If in the bootcmd the stdin is changed, e.g. from nulldev to
serial, it could happen that junk characters sit in the fifo and appear
on the cli.
Add a option to clear stdin before starting the CLI.
Signed-off-by: Gregor Herburger <gregor.herburger@linutronix.de>
Reviewed-by: Simon Glass <sjg@chromium.org>
Add a common helper console_flush_stdin() to drain all pending
characters from stdin. This consolidates the open-coded
while (tstc()) getchar() pattern that appeared in multiple places
across the tree.
Signed-off-by: Gregor Herburger <gregor.herburger@linutronix.de>
Reviewed-by: Alexander Sverdlin <alexander.sverdlin@siemens.com>
Reviewed-by: Quentin Schulz <quentin.schulz@cherry.de>
Reviewed-by: Simon Glass <sjg@chromium.org>
Some extension boards have EEPROM part numbers that do not directly match
their devicetree overlay filenames. Introduce a static name_mapping table
and a set_cape_overlay() helper that translates the part number and version
strings read from the EEPROM into the correct overlay filename.
When no entry matches, fall back to the existing behavior of constructing
the overlay name as "<part_number>-<version>.dtbo" directly from the
EEPROM content.
Add an initial entry mapping BB-GREEN-HDMI revision 00A0 to
am335x-bone-hdmi-00a0.dtbo.
Signed-off-by: Kory Maincent (TI) <kory.maincent@bootlin.com>
Set CONFIG_ENV_MMC_EMMC_HW_PARTITION=1 to store the environment in the
eMMC first hardware boot partition instead of defaulting to 0 (user
data area), where the rootfs and data partitions normally reside,
risking corruption. This keeps the full boot stack self-contained in
the eMMC hardware boot partition.
Enable CONFIG_ENV_RELOC_GD_ENV_ADDR so that gd->env_addr is adjusted by
the relocation offset during U-Boot relocation. Without it, gd->env_addr
keeps pointing to the pre-relocation address of the default
environment, leading to incorrect environment access before env_load()
completes.
Fixes: 3f0528882c ("board: toradex: add aquila am69 support")
Signed-off-by: Vitor Soares <vitor.soares@toradex.com>
Acked-by: Francesco Dolcini <francesco.dolcini@toradex.com>
Although less common, ADSP SoM boards can be operated standalone without
a carrier board. The SoM can be powered over USB-C by moving the jumper
on JP1.
The board configuration needs to be heavily refactored to reduce the
number of defconfigs and leverage the devicetree. The TI DP83867
support, which is dependent on that configuration, needs to be removed.
Until then this change allows the same U-Boot binary to work correctly
both with carrier boards and the standalone SoM.
The carrier board initialization code in somcrr_ezkit.c and
somcrr_ezlite.c calls gpio_hog_lookup_name() to find ethernet
control GPIOs, but does not check if the lookup succeeds before
using the returned pointers.
When running on a standalone SoM without a carrier board, these
GPIO hogs do not exist in the device tree. The gpio_hog_lookup_name()
function returns -ENODEV and sets the pointer to NULL, but the code
continues to call dm_gpio_set_value() with uninitialized pointers,
causing a synchronous abort exception:
"Synchronous Abort" handler, esr 0x96000004
elr: 00000000960326c4 lr : 0000000096038d5c (reloc)
Signed-off-by: Philip Molloy <philip.molloy@analog.com>
Reviewed-by: Greg Malysa <malysagreg@gmail.com>
The current gzwrite() implementation is limited to 4 GiB compressed
input buffer size due to struct z_stream_s { uInt avail_in } member,
which is of type unsigned int. Current gzwrite() implementation sets
the entire input buffer size as avail_in and performs decompression
of the whole compressed input buffer in one round, which limits the
size of input buffer to 4 GiB.
Rework the decompression loop to use chunked approach, and decompress
the input buffer in up to 4 GiB - 1 kiB avail_in chunks, possibly in
multiple decompression rounds. This way, the compressed input buffer
size is limited by gzwrite() function 'len' parameter type, which is
unsigned long.
In case of sandbox build, include parsing of 'gzwrite_chunk'
environment variable, so the chunked approach can be thoroughly tested
with non default chunk size. For non-sandbox builds, the chunk size is
4 GiB - 1 kiB.
The gzwrite test case is extended to test various chunk sizes during
gzwrite decompression test.
Signed-off-by: Marek Vasut <marek.vasut+renesas@mailbox.org>
Currently, virtio_mmio_probe() returns 0 when it encounters an invalid
magic value, an unsupported version, or a dummy device (ID 0). In
U-Boot's driver model, returning 0 indicates a successful probe. This
causes the system to incorrectly register and activate invalid or
placeholder devices, potentially leading to undefined behavior or
crashes later on.
Update the probe function to return appropriate error codes (-ENODEV
for invalid magic values and dummy devices, and -ENXIO for unsupported
versions). This fix correctly instructs the DM to abort the binding
process.
Fixes: fdc4aca89e ("virtio: Add virtio over mmio transport driver")
Signed-off-by: Kuan-Wei Chiu <visitorckw@gmail.com>
Acked-by: Daniel Palmer <daniel@thingy.jp>
CI: https://source.denx.de/u-boot/custodians/u-boot-imx/-/pipelines/29880
- Convert TQMA7 boards to SPL.
- Use second Ethernet MAC also from fuse on i.MX dhelectronics boards.
- Add support for Toradex Verdin iMX95.
- Drop SPL_POWER_LEGACY from imx8m boards.
- Increase the Aquantia Ethernet PHY reset timeout.
Previously, Amlogic SoCs defined their own reset_cpu() implementation
that essentially just did what the sysreset PSCI driver already
supports. To reduce duplication (and allow for EFI Capsule on Disk
support), make use of it instead of a board-specific reset_cpu.
This does not affect SPL, as in that stage we do not actually have PSCI
(BL31 is not loaded yet at that point). Also, boards that don't select
PSCI support in their config will not have it enabled.
Reviewed-by: Neil Armstrong <neil.armstrong@linaro.org>
Signed-off-by: Ferass El Hafidi <funderscore@postmarketos.org>
Link: https://patch.msgid.link/20260422-b4-amlogic-sysreset-capsules-v3-2-07b5ccc8f552@postmarketos.org
Signed-off-by: Neil Armstrong <neil.armstrong@linaro.org>
This is in preparation of enabling SYSRESET across all Amlogic boards by
default. SYSRESET brings its own reset_cpu implementation which will
conflict with the one defined in board-common.c.
SPL however will not have SYSRESET enabled (due to overhead in the final
image, as we have very little space to work with), and thus will still
need that reset_cpu to be defined. Furthermore, some boards do not enable
PSCI support in U-Boot, so they may still need this implementation.
Guard reset_cpu inside #ifdef to only compile it in when SYSRESET is not
enabled.
Signed-off-by: Ferass El Hafidi <funderscore@postmarketos.org>
Reviewed-by: Neil Armstrong <neil.armstrong@linaro.org>
Link: https://patch.msgid.link/20260422-b4-amlogic-sysreset-capsules-v3-1-07b5ccc8f552@postmarketos.org
Signed-off-by: Neil Armstrong <neil.armstrong@linaro.org>
The ufshcd_read_string_desc() can perform out of bounds write and
corrupt heap in case the input utf-16 string contains code points
which convert to anything more than plain 7-bit ASCII string.
This occurs because utf16_to_utf8(dst, src, size) in U-Boot behaves
differently than Linux utf16s_to_utf8s(..., maxlen), but the porting
process did not take that into consideration. The U-Boot variant of
the function converts up to $size utf-16 fixed-length 16-bit input
characters into as many 1..4 Byte long variable-length utf-8 output
characters. That means for 16 Byte input, the output can be up to 64
Bytes long. The Linux variant converts up utf-16 input into up to
$maxlen Bytes worth of utf-8 output, but stops at the $maxlen limit.
That means for 16 Byte input with maxlen=32, the processing will stop
after writing 32 output Bytes.
In case of U-Boot, use of utf16_to_utf8() leads to potential corruption
of data past the $size Bytes and therefore corruption of surrounding
content on the heap.
The fix is as simple, allocate buffer that is sufficient to fit the
utf-8 string. The rest of the code in ufshcd_read_string_desc() does
correctly limit the buffer to fit into the DMA descriptor afterward.
Signed-off-by: Marek Vasut <marek.vasut+renesas@mailbox.org>
Link: https://patch.msgid.link/20260329231151.332108-1-marek.vasut+renesas@mailbox.org
Signed-off-by: Neil Armstrong <neil.armstrong@linaro.org>
Switch the Phanbell i.MX8MQ SPL to full Driver Model (DM) boot flow by
moving early device initialization into devicetree and enabling the
required SPL DM subsystems.
Mark GPIO, USDHC, pinctrl, and regulator nodes with bootph-pre-ram so
they are available during SPL. With DM handling MMC and power rails,
remove legacy board-specific USDHC, GPIO, and pad setup code from SPL.
Update the SPL initialization sequence to use spl_early_init(), clears
BSS earlier, and explicitly enables USDHC clocks before handing off to
board_init_r().
Signed-off-by: Peng Fan <peng.fan@nxp.com>
Migrate the Kontron PITX-IMX8M board from the legacy power framework
to the DM PMIC infrastructure.
The SPL power initialization is converted to use DM_PMIC_PFUZE100,
dropping all legacy I2C, GPIO, pinmux, and MMC setup code that is
now handled through the device tree and driver model instead.
Key changes:
- Switch PFUZE100 access to DM PMIC APIs
- Enable SPL OF_CONTROL and mark required nodes with bootph*
- Remove legacy I2C and power configuration
- Enable SPL DM, pinctrl, regulator, and PMIC support
- Adjust SPL stack placement for DM usage
Signed-off-by: Peng Fan <peng.fan@nxp.com>
This board does not have any code to call legacy power API, so drop the
selection of CONFIG_SPL_POWER_LEGACY.
Signed-off-by: Peng Fan <peng.fan@nxp.com>
This board does not have any code to call legacy power API, so drop the
selection of CONFIG_SPL_POWER_LEGACY.
Signed-off-by: Peng Fan <peng.fan@nxp.com>
This board does not have any code to call legacy power API, so drop the
selection of CONFIG_SPL_POWER_LEGACY.
Signed-off-by: Peng Fan <peng.fan@nxp.com>
Migrate the i.MX8MQ EVK board to the Driver Model (DM) PMIC framework.
This replaces legacy PMIC handling with DM_PMIC and enables proper
device-model support during SPL.
As part of this transition, enable CONFIG_SPL_DM and the required
DM subsystems such as I2C, PINCTRL, MMC, GPIO, and regulators so the
SPL boot flow continues to work correctly.
Board-specific SPL code is simplified accordingly by removing legacy
I2C, USDHC, and PMIC handling, relying instead on device tree
descriptions and DM drivers.
Signed-off-by: Peng Fan <peng.fan@nxp.com>
Switch PFUZE100 object linkage to be phase-aware and fix build issues
when using driver model PMIC support.
The PFUZE100 helper code is reworked to:
- Build pfuze.o only when CONFIG_(SPL_)DM_PMIC_PFUZE100 is enabled
- Use CONFIG_IS_ENABLED(DM_PMIC_PFUZE100) for proper DM/non-DM selection
- Align function signatures and implementations with DM PMIC APIs
- Use udevice-based pmic access for DM and legacy pmic for non-DM
- Avoid mixing struct pmic and struct udevice in the same build
configuration
No functional change intended beyond fixing DM support and build
consistency.
Signed-off-by: Peng Fan <peng.fan@nxp.com>
To mx6sabreauto and mx6sabresd, power_init_board is called in board_r phase.
After CONFIG_SPL_DM_PMIC_PFUZE100 included, there will be build error to build
power_init_board() when CONFIG_SPL_DM_PMIC_PFUZE100 is not enabled.
Signed-off-by: Peng Fan <peng.fan@nxp.com>
signed_hdmi_imx8m.bin is already signed and has a IVT header. It should not
be put in u-boot-spl-mkimage.signed.bin. Move it to head of flash.bin
following NXP imx-mkimage. Keeping it in u-boot-spl-mkimage.signed.bin also
consumes a lot of TCM space which is not expected.
While moving it to head of flash.bin, other changes are required,
u-boot.itb is put at sector 768 per defconfig, so u-boot.itb binman
offset should be updated and it should be moved out from binman section.
Also binman symbol address are updated, so need to subtract u-boot-spl
image_pos + CONFIG_SPL_TEXT_BASE to find the correct location of ddr phy
firmware.
Because there is 1KB padding in HDMI firmware, use 32KB when burning
flash.bin to sd card.
Signed-off-by: Peng Fan <peng.fan@nxp.com>
The global data (gd) area is already zeroed in
board_init_f_init_reserve() before board_init_f() is called.
Remove the duplicate gd clearing from board_init_f().
Signed-off-by: Peng Fan <peng.fan@nxp.com>
The global data (gd) area is already zeroed in
board_init_f_init_reserve() before board_init_f() is called.
Remove the duplicate gd clearing from board_init_f().
Signed-off-by: Peng Fan <peng.fan@nxp.com>
The global data (gd) area is already zeroed in
board_init_f_init_reserve() before board_init_f() is called.
Remove the duplicate gd clearing from board_init_f().
Signed-off-by: Peng Fan <peng.fan@nxp.com>
The global data (gd) area is already zeroed in
board_init_f_init_reserve() before board_init_f() is called.
Remove the duplicate gd clearing from board_init_f().
Signed-off-by: Peng Fan <peng.fan@nxp.com>
The global data (gd) area is already zeroed in
board_init_f_init_reserve() before board_init_f() is called.
Remove the duplicate gd clearing from board_init_f().
Signed-off-by: Peng Fan <peng.fan@nxp.com>
The global data (gd) area is already zeroed in
board_init_f_init_reserve() before board_init_f() is called.
Remove the duplicate gd clearing from board_init_f().
Signed-off-by: Peng Fan <peng.fan@nxp.com>
Add probe_ram_size_by_alias() to detect RAM size by checking whether a
write to one address aliases to another address.
Compared to get_ram_size(), this function allows the caller to:
- limit probing to a small set of required accesses
- avoid touching reserved or already used memory regions
- handle non-linear alias patterns
On the iMX95 SoC, when used with LPDDR5, accesses beyond the end of an 8GB DDR
configuration do not alias to the expected linear wrap-around addresses.
Instead, the aliased addresses appear to follow a pattern related to the
DDRC bank and bank-group addresses mapping. Experimentally, the observed
pattern is:
Write Read
y00000000 -> x0001c000
y00004000 -> x00018000
y00008000 -> x00014000
y0000c000 -> x00010000
y00010000 -> x0000c000
y00014000 -> x00008000
y00018000 -> x00004000
y0001c000 -> x00000000
This helper makes it possible to probe RAM size by explicitly specifying
the probed address and the expected alias address for each size check.
Signed-off-by: Emanuele Ghidoli <emanuele.ghidoli@toradex.com>
Reviewed-by: Simon Glass <sjg@chromium.org>
Instead of adding select ARCH_MISC_INIT for each i.MX 9x SoC, select it
at the i.MX9 family Kconfig level. The corresponding arch_misc_init() is
already defined in the common arch/arm/mach-imx/imx9/soc.c file and will
print out ELE firmware BuildInfo for all i.MX9 SoCs now instead of just
i.MX91 and i.MX95.
Signed-off-by: Primoz Fiser <primoz.fiser@norik.com>
Reviewed-by: Peng Fan <peng.fan@nxp.com>
Currently, the board specific code evaluates only the first Ethernet
MAC address fuse, regardless of whether the first or second MAC
address is requested. When the function to determine the second
Ethernet MAC address is looking for the fused MAC address, it only
reads the first MAC address fuse and increment it by one to set
the second Ethernet MAC address. That is not the expected behavior
when two MAC addresses are fused, because this causes the second
fused MAC address to be ignored. Change this so that the second
fused MAC address will be used.
Reviewed-by: Peng Fan <peng.fan@nxp.com>
Signed-off-by: Christoph Niedermaier <cniedermaier@dh-electronics.com>
Reviewed-by: Marek Vasut <marex@nabladev.com>
