Quentin Schulz <foss+uboot@0leil.net> says:
I have a couple of products whose U-Boot FIT is signed via a proprietary
OpenSSL engine which only expects the name of a "slot" to select the key
to sign data with.
Currently mkimage fit support expects either a key-dir (-k) or a
key-file (-G) as a toggle for signing, however this doesn't apply to our
usecase because we use an OpenSSL engine (so no key-file to provide)
which doesn't mimic a directory layout like key-dir implies. Moreover,
binman really expects private keys (.key extension) to be available in
this key-dir directory, which we of course cannot provide.
This series allows to sign a FIT image with mkimage (and binman) with
an OpenSSL engine, including PKCS11 and custom engines. If a key-dir
needs to be passed (which is typical for PKCS11), one can do so by using
fit,engine-keydir.
Note that the public key (.crt extension) still needs to be available if
one wants to embed it for signature verification (which is probably what
one wants to do :) ). It is probably possible to use the engine for
getting the public key instead of storing it on disk, but this needs to
be added to fdt_add_pubkey and then binman, through a mechanism
different from fit,engine*.
One issue though is that since binman resolves key paths absolutely and
that I don't believe an OpenSSL engine would happen to have the exact
same key_id value than a local absolute path, fit,encrypt and
fit,engine cannot cohabit. An issue for the next person who wants
an OpenSSL engine AND encrypt the same FIT image, I don't.
Note that LibreSSL supports neither engines nor providers as far as I
could tell (engine support has been explicitly removed).
Note that OpenSSL engines have been deprecated since 3.0 (Q3-2021),
however note that OpenSSL 3.5 still seems to support engines (git grep)
and is EOL end of Q1 2030.
If anyone has an idea on how to test PKCS11 with SOftHSMv2 with id=
passed in fit,engine-keydir, I'm all ears.
I'm also wondering if the explanation around fit,engine-keydir aren't
too much. After all, they are passed verbatim to mkimage as -k argument
and the special cases are all specific to mkimage and not binman.
Link: https://lore.kernel.org/r/20251121-binman-engine-v3-0-b80180aaa783@cherry.de
This adds a test that signs a FIT and verifies the signature with
fit_check_sign.
OpenSSL engines are typically for signing with external HW so it's not
that straight-forward to simulate.
For a simple RSA OpenSSL engine, a dummy engine with a hardcoded RSA
4096 private key is made available. It can be selected by setting the
OpenSSL engine argument to dummy-rsa-engine. This can only be done if
the engine is detected by OpenSSL, which works by setting the
OPENSSL_ENGINES environment variable. I have no clue if dummy-rsa-engine
is properly implementing what is expected from an RSA engine, but it
seems to be enough for testing.
For a simple PKCS11 engine, SoftHSMv2 is used, which allows to do PKCS11
without specific hardware. The keypairs and tokens are generated on the
fly. The "prod" token is generated with a different PIN (1234 instead of
1111) to also test MKIMAGE_SIGN_PIN env variable while we're at it.
Binman will not mess with the local SoftHSMv2 setup as it will only use
tokens from a per-test temporary directory enforced via the temporary
configuration file set via SOFTHSM2_CONF env variable in the tests. The
files created in the input dir should NOT be named the same as it is
shared between all tests in the same process (which is all tests when
running binman with -P 1 or with -T).
Once signed, it's checked with fit_check_sign with the associated
certificate.
Finally, a new softhsm2_util bintool is added so that we can initialize
the token and import keypairs. On Debian, the package also brings
libsofthsm2 which is required for OpenSSL to interact with SoftHSMv2. It
is not the only package required though, as it also needs p11-kit and
libengine-pkcs11-openssl (the latter bringing the former). We can detect
if it's properly installed by running openssl engine dynamic -c pkcs11.
If that fails, we simply skip the test.
The package is installed in the CI container by default.
Signed-off-by: Quentin Schulz <quentin.schulz@cherry.de>
This adds support for using an OpenSSL engine for signing a FIT image.
To use it, one should set the fit,engine property at the FIT node level
with the engine to use. This will in turn call mkimage with the -N
option.
The -k argument to mkimage can be specified via fit,engine-keydir. If
not specified, -k is not passed to mkimage. This property is especially
useful for pkcs11 engine to specify slots, token label, etc...
As far as I could tell, mkimage encrypts and signs a FIT in one go, thus
the -k argument applies to both signing and encrypting. Considering we
reuse the -k argument for two different meanings (info to pass to the
engine when using an engine otherwise the directory where keys are
stored), we cannot reasonably encrypt using local keys and signing with
an engine, hence the enforced check. I believe it should be possible to
support encrypting and signing with the same engine (using different
key pairs of course, via different key-name-hint likely), but this is
left for the next person to implement.
This is why the property is named fit,engine and not fit,sign-engine.
Ditto for fit,engine-keydir.
The public key (with .crt extension) is still required if it needs to be
embedded in the SPL DTB for example. We could probably support
retrieving the public key from an engine, but this is a change to make
to fdt_add_pubkey.c.
Signed-off-by: Quentin Schulz <quentin.schulz@cherry.de>
mkimage has support for OpenSSL engines but binman currently doesn't for
direct callers of mkimage (e.g. the fit etype). This prepares for adding
support for OpenSSL engines for signing elements of a FIT image, which
will done in the next commit.
Reviewed-by: Wolfgang Wallner <wolfgang.wallner@br-automation.com>
Reviewed-by: Simon Glass <sjg@chromium.org>
Signed-off-by: Quentin Schulz <quentin.schulz@cherry.de>
Currently, when one wants to use an OpenSSL engine to sign a FIT image,
one needs to pass a keydir (via -k) to mkimage which will then be
prepended to the value of the key-name-hint before being passed as
key_id argument to the OpenSSL Engine API, or pass a keyfile (via -G) to
mkimage.
My OpenSSL engine only has "slots" which are not mapped like
directories, so using keydir is not proper, though I could simply have
-k '' I guess but this won't work currently with binman anyway.
Additionally, passing a keyfile (-G) when using an engine doesn't make
sense as the key is stored in the engine.
Let simply allow FIT images be signed if both keydir and keyfile are
missing but an engine is to be used.
The keyname member is already filled by looking at key-name-hint
property in the FIT and passed to the engine, which is exactly what is
needed here.
Reviewed-by: Wolfgang Wallner <wolfgang.wallner@br-automation.com>
Reviewed-by: Simon Glass <sjg@chromium.org>
Signed-off-by: Quentin Schulz <quentin.schulz@cherry.de>
While adding CPSW device support to enable Ethernet boot for J722S,
dev-data and clk-data for eMMC was removed by mistake, which leads to eMMC
boot failure. Update the dev-data and clk-data to fix that.
Fixes: a02009f3a8 ("arm: mach-k3: j722s: Update SoC autogenerated data to enable Ethernet boot")
Reported-by: Michael Walle <mwalle@kernel.org>
Signed-off-by: Chintan Vankar <c-vankar@ti.com>
Tested-by: Michael Walle <mwalle@kernel.org>
Reviewed-by: Udit Kumar <u-kumar1@ti.com>
ARM GCC tool check is not up to date,
while issues are reported such as file truncated linker errors.
Using ARM official cross tools shows that 10.0.1 is a safe
version to support latest kbuild bump properly.
Signed-off-by: Brian Sune <briansune@gmail.com>
I am moving over to using my email address at kernel.org.
Change this in all the affected MAINTAINERS files.
Reviewed-by: Quentin Schulz <quentin.schulz@cherry.de>
Signed-off-by: Linus Walleij <linusw@kernel.org>
Reviewed-by: Neil Armstrong <neil.armstrong@linaro.org>
Pull request efi-2026-01-rc4
CI: https://source.denx.de/u-boot/custodians/u-boot-efi/-/pipelines/28652
Documentation:
* Correct and add missing pytest hook script environment variable names
* board: verdin-am62p: Fix boot log output
* Add a page for downloading the U-Boot logo with and without text
UEFI:
* Fix a memory leak when retrieving device paths from boot vars
Fix boot log output not being rendered correctly, fix
it doing the same as done in verdin-am62.rst.
Fixes: b8c43968b8 ("board: toradex: add verdin am62p support")
Signed-off-by: Francesco Dolcini <francesco.dolcini@toradex.com>
get_dp_device() is used to derive the device path from a boot variable.
However, if the last efi_get_variable_int() call fails, we return an
error without freeing 'buf'.
There's no need to call efi_get_variable_int() for variables we don't
know the size since we have the efi_get_var() wrapper.
Replace that in the two instances we use it. The first one will also
fix the memory leak.
A nice sideeffect is that the code size is also reduced, since we are
re-using functions instead of open coding them
$~ bloat-o-meter u-boot u-boot.new
add/remove: 0/0 grow/shrink: 1/2 up/down: 6/-196 (-190)
Function old new delta
version_string 70 76 +6
efi_launch_capsules 2288 2196 -92
get_dp_device 244 140 -104
Total: Before=1222331, After=1222141, chg -0.02%
Fixes: c74cd8bd08 ("efi_loader: capsule: add capsule_on_disk support")
Signed-off-by: Ilias Apalodimas <ilias.apalodimas@linaro.org>
Reviewed-by: Heinrich Schuchardt <heinrich.schuchardt@canonical.com>
Add the _EXTRA variants of U_BOOT_BUILD_DIR and U_BOOT_RESULT_DIR to the
list of environment variables set for hook scripts.
These were added in commit 8f2a9fa7d6 ("test: Support testing with two
board-builds") but were not documented.
Signed-off-by: David Lechner <dlechner@baylibre.com>
Fix the names of environment variables set for hook scripts. These
have a U_BOOT_ prefix, not UBOOT_.
Signed-off-by: David Lechner <dlechner@baylibre.com>
Andrew Goodbody <andrew.goodbody@linaro.org> says:
Smatch reported an error where a value calculated by ERR_PTR was not
used. Fixing this to return the generated value led to a test failure
which meant updating the sandbox clock code so that it would still cause
the tests to pass with the above correction.
Debugging this problem led to a SIGSEGV which is addressed in 1/3.
Possible memory leaks noticed are addressed in 3/3.
Link: https://lore.kernel.org/r/20251121-clk_uclass_fix-v2-0-74f4ea10e194@linaro.org
In clk_set_default_rates() memory is allocated to store the clock rates
that are read. Direct returns fail to free this memory leading to a
memory leak so instead use 'goto fail;' which will then perform the free
before exiting the function.
Signed-off-by: Andrew Goodbody <andrew.goodbody@linaro.org>
In clk_set_default_get_by_id ret is passed to ERR_PTR but nothing is
done with the value that this calculates which is obviously not the
intention of the code. This is confirmed by the code around where this
function is called.
Instead return the value from ERR_PTR.
Then fixup the sandbox code so that the test dm_test_clk does not fail
as it relied on the broken behaviour.
Finally disable part of the test that does not work correctly with
CLK_AUTO_ID
This issue found by Smatch.
Signed-off-by: Andrew Goodbody <andrew.goodbody@linaro.org>
If LOG_DEBUG is defined and a NULL clk is passed to clk_enable or
clk_disable then an attempt is made to dereference NULL in the debug
statement. Guard against this.
Signed-off-by: Andrew Goodbody <andrew.goodbody@linaro.org>
Reviewed-by: Patrick Delaunay <patrick.delaunay@foss.st.com>
The macro U_BOOT_DEVICE has been renamed to U_BOOT_DRVINFO.
This patch updates the reference in reboot-mode-gpio.h.
Signed-off-by: Cibil Pankiras <cibil.pankiras@egym.com>
If noffset is negative, do not pass it to fit_get_name() and then further to
libfdt, this will crash sandbox with SIGSEGV because libfdt can not handle
negative node offsets without full tree check, which U-Boot inhibits to keep
size lower.
Instead, always check noffset before use, and if the return value indicates
failure, exit right away.
Signed-off-by: Marek Vasut <marek.vasut+renesas@mailbox.org>
Acked-by: Heinrich Schuchardt <xypron.glpk@gmx.de>
The DA8xx GPIO driver was not being built as part of the A53 U-Boot
image on AM64x and AM65x. This meant only i2c GPIO expanders were
accessible to the users from the U-Boot prompt.
This patch fixes it by setting CONFIG_DA8XX_GPIO.
Signed-off-by: Anshul Dalal <anshuld@ti.com>
Otherwise the custom-cape eeprom (at address 57) reports NACK which
results into "i2c_write: error waiting for data ACK (status=0x116)" and
terminates further scanning.
Signed-off-by: Marian Cingel <cingel.marian@gmail.com>
Commit 5f70be08b0 ("Fix autoboot countdown printing wrong") introduces
inconsistency in how the countdown is displayed. For example, in case
when BOOTDELAY=5, the next output is observed during the boot:
Hit any key to stop autoboot: 5
Hit any key to stop autoboot: 4
Hit any key to stop autoboot: 3
That happens due to different printf format (%2d vs %1d). Moreover, the
mentioned commit fails to handle the case when the user is holding some
key before the countdown is shown. E.g. if BOOTDELAY=101, the next
malformed output is being produced:
Hit any key to stop autoboot: 1 0
That's because the fast path code wasn't modified accordingly, and still
tries to erase the number using '\b\b\b' format.
Fix both issues by introducing a dedicated routine for printing the
whole countdown line.
Fixes: 5f70be08b0 ("Fix autoboot countdown printing wrong")
Signed-off-by: Sam Protsenko <semen.protsenko@linaro.org>
Reviewed-by: Tom Rini <trini@konsulko.com>
Acked-by: David Zang <davidzangcs@gmail.com>
lthreads is of size MAX_THREADS, hence id must be lower than
MAX_THREADS to avoid any potential buffer overflow in
thread_start function.
Signed-off-by: Francois Berder <fberder@outlook.fr>
Attempting to build with "allyesconfig" means that we try and build all
available options for the sandbox platforms. Doing so exposes that the
drivers under drivers/clk/ti/ can only be compiled or linked on
ARCH_OMAP2PLUS platforms as some drivers require platform specific
headers while other drivers depend on these first drivers to link.
Express those requirements in Kconfig as well.
Reviewed-by: Manorit Chawdhry <m-chawdhry@ti.com>
Signed-off-by: Tom Rini <trini@konsulko.com>
The ALTERA_QSPI driver conflicts with the regular FLASH_CFI_DRIVER as
both implement the same high level functionality and so use the same
global namespace. In a similar fashion, all NAND drivers are mutually
exclusive due to namespace collisions. For the remaining drivers which
did not already have some architecture specific dependency, add them.
Signed-off-by: Tom Rini <trini@konsulko.com>
Quentin Schulz <quentin.schulz@cherry.de> says:
Only the Siemens corvus board seems to be using these two status LEDs
from the legacy LED API.
Since we're trying to get rid of the last users of the legacy LED API,
let's migrate Corvus to the modern LED API instead, which uses DM. For
Corvus's case, it also uses DM_GPIO (already enabled in defconfig).
Since there was no use for the green status_led (not compiled in), it
simply is removed without migrating it to the modern API. If need be, we
can always add a new gpio-led in the FDT.
Note that I do not own a Siemens Corvus board so it's a bit of a shot in
the dark whether it'll work on the first try, only build tested.
The red LED should be on whenever reaching U-Boot proper CLI, if not we
have an issue.
The LED should be controllable with the led command from U-Boot proper
CLI.
Link: https://lore.kernel.org/r/20251119-corvus-led-red-green-v1-0-ce86b8d59dfc@cherry.de
To the exception of red_led_on in the arm-specific assembly code, all
code interacting with the red status LED was guarded by the
CONFIG_LED_STATUS_RED symbol, which is enabled in none of the upstream
defconfigs.
Since the last board which overrode the weak red_led_on function got
migrated to the new LED mechanism, there's also no user of the
arm-specific assembly code anymore, therefore it can be removed along
the other unreachable code sections.
Signed-off-by: Quentin Schulz <quentin.schulz@cherry.de>
Reviewed-by: Heiko Schocher <hs@nabladev.com>
red_led_on is either called from the legacy LED shell command (which is
disabled for corvus) or from arm-specific assembly code right before
jumping into board_init_r() in U-Boot proper.
Let's migrate to use the more modern LED subsystem by migrating to DM.
The default-state is set to on to mimic red_led_on() from the
arm-specific assembly code as a missing default-state FDT property
currently means the LED is not probed except if explicitly done via the
led shell command. Note though that this is running much later in the
boot process, once DM is started.
Signed-off-by: Quentin Schulz <quentin.schulz@cherry.de>
Reviewed-by: Heiko Schocher <hs@nabladev.com>
The last user of it was removed in a previous commit so let's remove its
support entirely.
Signed-off-by: Quentin Schulz <quentin.schulz@cherry.de>
Reviewed-by: Heiko Schocher <hs@nabladev.com>
green_led_on and green_led_off are only called by the legacy LED command
(CONFIG_LED_STATUS_CMD) when CONFIG_LED_STATUS_GREEN is enabled, both of
which aren't enabled for corvus, so let's simply remove it as it's dead
code.
Signed-off-by: Quentin Schulz <quentin.schulz@cherry.de>
Reviewed-by: Heiko Schocher <hs@nabladev.com>
No PPC upstream defconfig actually enables CONFIG_LED_STATUS and we're
trying to get rid of the legacy LED API, so let's remove one of its last
users.
Signed-off-by: Quentin Schulz <quentin.schulz@cherry.de>
The code is guarded by a condition none of the defconfigs meet (that is
CONFIG_SYS_FAULT_ECHO_LINK_DOWN and CONFIG_LED_STATUS_RED both enabled),
so we can remove the unreachable code sections.
When doing that, there's no caller for miiphy_link anymore, so it can be
removed.
This in turns makes CONFIG_SYS_FAULT_ECHO_LINK_DOWN and
CONFIG_SYS_FAULT_MII_ADDR unused so they are removed as well.
Signed-off-by: Quentin Schulz <quentin.schulz@cherry.de>
Reviewed-by: Tom Rini <trini@konsulko.com>
Commit 76386d6195 ("arm: Remove cm_t35 board") removed support for the
board that was built when TARGET_CM_T35 is selected, but removal of the
symbol was forgotten, so let's fix this oversight.
While at it, update the README for omap3 to remove the last mention of
cm_t35.
Fixes: 76386d6195 ("arm: Remove cm_t35 board")
Signed-off-by: Quentin Schulz <quentin.schulz@cherry.de>
Reviewed-by: Tom Rini <trini@konsulko.com>
The last user of coloured_LED_init has been recently removed, so we can
remove all places it's called and defined as it does nothing now.
Nobody makes use of the yellow and blue status LEDs from the legacy API,
so let's remove all references to it.
Signed-off-by: Quentin Schulz <quentin.schulz@cherry.de>
The last CONFIG_MVS in code was removed almost 12 years ago in commit
3b98b57fa7 ("include: delete unused header files").
STATUS_LED_PAR was last seen 8 years ago when removed in commit
5b8e76c35e ("powerpc, 8xx: remove support for 8xx").
If CONFIG_LED_STATUS_BOARD_SPECIFIC is not defined, the build will fail
so we won't even reach this part of the code.
Let's simplify the if block to actually possible configurations.
Signed-off-by: Quentin Schulz <quentin.schulz@cherry.de>
Reviewed-by: Tom Rini <trini@konsulko.com>
Newer versions of libfdt strictly check whether the FDT blob
passed to them is at 8-byte aligned offset, if it is not, then
the library fails checks with -FDT_ERR_ALIGNMENT . Currently,
'abootimg get dtb --index=1 addr size' may return non 8-byte
aligned FDT address which points directly into the abootimg.
Copy the result into temporary location before validation to
avoid FDT alignment check failure.
Signed-off-by: Marek Vasut <marek.vasut+renesas@mailbox.org>
Reviewed-by: Mattijs Korpershoek <mkorpershoek@kernel.org>
Link: https://lore.kernel.org/r/20251119193311.127633-2-marek.vasut+renesas@mailbox.org
Signed-off-by: Mattijs Korpershoek <mkorpershoek@kernel.org>
Newer versions of libfdt strictly check whether the FDT blob
passed to them is at 8-byte aligned offset, if it is not, then
the library fails checks with -FDT_ERR_ALIGNMENT . Currently,
android_image_print_dtb_contents() passed FDT directly mapped
from abootimg to libfdt, and this FDT is not always aligned to
8-byte offset. Specifically, the FDTs are somewhat packed in
the abootimg, therefore if the first FDT blob is e.g. 0xfd bytes
long, then the next FDT blob ends up at 0xfd offset, which is
not 8-byte aligned.
Fix this by first extracting the header into 8-byte aligned buffer,
checking only the header for validity, and then by copying the
entire FDT into newly allocated 8-byte aligned buffer. While this
is not efficient, it is the correct way to handle DTs, which must
be at 8-byte aligned offsets. Mitigate the inefficiency for the
common case by checking whether the DT might be 8-byte aligned and
if it is, map it directly.
Signed-off-by: Marek Vasut <marek.vasut+renesas@mailbox.org>
Reviewed-by: Mattijs Korpershoek <mkorpershoek@kernel.org>
Reviewed-by: Tom Rini <trini@konsulko.com>
Link: https://lore.kernel.org/r/20251119193311.127633-1-marek.vasut+renesas@mailbox.org
Signed-off-by: Mattijs Korpershoek <mkorpershoek@kernel.org>
Heinrich Schuchardt <heinrich.schuchardt@canonical.com> says:
Many C unit tests are not executed if CONFIG_CONSOLE_RECORD is not set.
Hence Tom suggested to let UNIT_TEST imply CONSOLE_RECORD.
The first patch makes the skipped C unit tests visible.
The rest of the series deals with hidden bugs in our tests.
The 'fdt get value' command returned incorrect values on low-endian
systems. So this needed fixing too.
Link: https://lore.kernel.org/r/20251123225711.227016-1-heinrich.schuchardt@canonical.com
The cmd and the log test suites rely on CONFIG_CONSOLE_RECORD.
The log test suite is always built if CONFIG_UNIT_TEST=y.
The print_do_hex_dump test relies on CONFIG_HEXDUMP. Imply it too.
Signed-off-by: Heinrich Schuchardt <heinrich.schuchardt@canonical.com>
The 32bit cells of a device-tree property are big-endian. When printing
them via 0x08x we must first convert to the host endianness.
Remove the restriction to 20 bytes length. This would not allow to read an
SHA256 value.
Signed-off-by: Heinrich Schuchardt <heinrich.schuchardt@canonical.com>
On the power architecture the bdinfo command prints architecture specific
information. The test needs to accept these output lines.
Signed-off-by: Heinrich Schuchardt <heinrich.schuchardt@canonical.com>
The original value of the first variable msr (0x200) is not controlled by
U-Boot. Don't make any assumption.
Signed-off-by: Heinrich Schuchardt <heinrich.schuchardt@canonical.com>
Hexdumps for types other then byte look different in dependence of the
endianness.
Signed-off-by: Heinrich Schuchardt <heinrich.schuchardt@canonical.com>
