Yao Zi [Tue, 2 Sep 2025 08:19:30 +0000 (08:19 +0000)]
riscv: Add Kconfig options to distinguish Zaamo and Zalrsc
Ratified on Apr. 2024, the original RISC-V "A" extension is now split
into two separate extensions, "Zaamo" for atomic operations and "Zalrsc"
for load-reserved/store-conditional instructions.
For now, we've already seen real-world designs implement the Zalrsc
extension only[2]. As U-Boot mainly runs with only one HART, we could
easily support these designs by not using AMO instructions in the
hard-written assembly if necessary, for which this patch introduces two
new Kconfig options to indicate the availability of "Zaamo" and "Zalrsc".
Note that even with this patch, "A" extension is specified in the ISA
string passed to the compiler as long as one of "Zaamo" or "Zalrsc" is
available, since they're only recognized with a quite recent version of
GCC/Clang. The compiler usually doesn't automatically generate atomic
instructions unless the source explicitly instructs it to do so, thus
this should be safe.
Yao Zi [Wed, 13 Aug 2025 10:09:26 +0000 (10:09 +0000)]
pinctrl: th1520: Mark driver as DM_FLAG_PRE_RELOC
It's common that UARTs are bound and probed before U-Boot relocation, in
which case the UART's pincontroller and pinconfig must be probed first.
Let's apply DM_FLAG_PRE_RELOC to the driver, allow it to bind before
relocation.
Signed-off-by: Yao Zi <ziyao@disroot.org> Reviewed-by: Leo Yu-Chi Liang <ycliang@andestech.com>
Yao Zi [Wed, 13 Aug 2025 10:09:25 +0000 (10:09 +0000)]
clk: thead: th1520-ap: Mark drivers as DM_FLAG_PRE_RELOC
It's common that UARTs are bound and probed before U-Boot relocation,
in which case the clocks of UART and UART's pincontroller must be
registered first. Let's apply DM_FLAG_PRE_RELOC to the driver, allowing
it to bind before relocation.
Signed-off-by: Yao Zi <ziyao@disroot.org> Reviewed-by: Leo Yu-Chi Liang <ycliang@andestech.com>
Yao Zi [Wed, 13 Aug 2025 10:09:24 +0000 (10:09 +0000)]
configs: th1520_lpi4a: Enlarge SYS_MALLOC_F_LEN to 0x10000
For TH1520, we want clock and pinctrl drivers to bind before relocation
along with the UART which makes use of them, since upstream devicetree
specifies pinctrl properties for the UART.
This requires a large malloc pool before relocation, let's enlarge it.
Signed-off-by: Yao Zi <ziyao@disroot.org> Reviewed-by: Leo Yu-Chi Liang <ycliang@andestech.com>
Marek Vasut [Sun, 7 Sep 2025 01:00:47 +0000 (03:00 +0200)]
board: dhelectronics: Use isascii() before isprint() in dh_read_eeprom_id_page()
The isprint() checks printability across all 256 characters, some of the
upper 128 characters are printable and produce artifacts on UART. Call
isascii() first to only consider the bottom 7bit ASCII characters as
printable, and then check their printability using isprint(). This fixes
a rare misprint in case the ID page content is uninitialized or corrupted.
Signed-off-by: Marek Vasut <marek.vasut@mailbox.org> Reviewed-by: Christoph Niedermaier <cniedermaier@dh-electronics.com>
Marek Vasut [Sun, 7 Sep 2025 01:00:46 +0000 (03:00 +0200)]
board: dhelectronics: Check pointer before access in dh_get_value_from_eeprom_buffer()
The eip pointer in dh_get_value_from_eeprom_buffer() might be NULL.
The current NULL pointer check happens too late, after the eip was
accessed in variable assignment. Reorder the two, so the NULL pointer
check happens first, and any access second, otherwise the access may
trigger a hang or other undefined behavior.
Signed-off-by: Marek Vasut <marek.vasut@mailbox.org> Reviewed-by: Christoph Niedermaier <cniedermaier@dh-electronics.com>
Marek Vasut [Sat, 6 Sep 2025 23:00:44 +0000 (01:00 +0200)]
phy: Reset init count on phy exit failure
In case the PHY exit callback reports failure, reset init_count to 0 anyway,
so the next attempt at PHY initialization might try to reinitialize the PHY
and restore it to normal operation.
Signed-off-by: Marek Vasut <marek.vasut@mailbox.org> Reviewed-by: Siddharth Vadapalli <s-vadapalli@ti.com>
Tom Rini [Tue, 16 Sep 2025 19:45:08 +0000 (13:45 -0600)]
Merge patch series "Modernize U-Boot code formatting with clang-format"
Javier Tia <javier.tia@linaro.org> says:
This patch series modernizes U-Boot's code formatting infrastructure by
with current Linux kernel practices and providing a more robust and
comprehensive formatting solution.
Javier Tia [Fri, 5 Sep 2025 20:53:47 +0000 (14:53 -0600)]
doc: Update U-Boot coding style guide with clang-format usage
The U-Boot coding style guide has been updated to include information
about using the `.clang-format` configuration file for automatic code
formatting. This ensures consistent formatting across the entire
codebase and aligns with Linux kernel coding standards. The goal with
introducing a predefined coding style is consistency rather than
personal preference.
The .clang-format file is copied directly from the Linux kernel without
any modifications, ensuring complete compatibility with kernel coding
standards.
Include comprehensive best practices for using clang-format,
specifically guidance on formatting only changed blocks versus entire
files, creating separate formatting-only commits for better code review,
and leveraging git clang-format for targeted formatting. Add examples of
editor integrations. This enhancement will help maintainers and
contributors to easily adhere to U-Boot coding standards.
Javier Tia [Fri, 5 Sep 2025 20:53:45 +0000 (14:53 -0600)]
tools: zynqmp_psu_init_minimize.sh: Switch to clang-format
Replace the use of scripts/Lindent with clang-format in
zynqmp_psu_init_minimize.sh. This change is made to align with the rest
of the codebase that uses clang-format for code formatting. This ensures
consistency across all scripts in terms of code style and formatting.
Tested-by: Michal Simek <michal.simek@amd.com> Signed-off-by: Javier Tia <javier.tia@linaro.org>
Javier Tia [Fri, 5 Sep 2025 20:53:44 +0000 (14:53 -0600)]
clang-format: Add configuration file
Introduce .clang-format configuration file to U-Boot, providing
significant improvements over the existing scripts/Lindent approach for
C code formatting.
Benefits of clang-format over scripts/Lindent:
- More comprehensive formatting rules: While Lindent relies on the basic
'indent' tool with limited options, clang-format provides extensive
control over code formatting with 800+ configuration options
- Better handling of complex code structures: clang-format understands
modern C constructs and handles nested structures, macros, and complex
expressions more intelligently
- IDE and editor integration: Native support in major development
environments (VS Code, Vim, Emacs, etc.) enables real-time formatting
- Consistent results across environments: Eliminates variations between
different versions of 'indent' tool and system configurations
- Active maintenance: clang-format is actively developed and updated,
unlike the aging 'indent' tool
Alignment with Linux kernel practices:
Continues U-Boot alignment with Linux kernel development practices,
maintaining consistency between these closely related projects. The
Linux kernel adopted clang-format to modernize its code formatting
infrastructure and improve developer experience.
The .clang-format file is based on the Linux kernel configuration,
specifically copied from Linux kernel v6.16 tag, which itself builds
upon the initial introduction in commit d4ef8d3ff005c ("clang-format:
add configuration file").
Tom Rini [Tue, 16 Sep 2025 19:44:39 +0000 (13:44 -0600)]
Merge patch series "Fix ADI driver header dependencies"
Greg Malysa <malysagreg@gmail.com> says:
Between 2025.07 and 2025.10 many header dependency chains were improved,
but this exposed implicit header usage in several of our drivers. This
wasn't discovered before or included in the original fixes because our
drivers are not yet used by any mainline-supported boards, so build
tests did not find them. This series addresses the two build failures
I've encountered while rebasing our work onto 2025.10 and continuing to
prepare the next submission of our board files.
As part of the header dependency cleanup between 2025.07 and 2025.10, an
implicit route to obtain SZ_128M from linux/sizes.h was removed. This
adds an explicit reference to linux/sizes.h to fix build failures for
this driver.
Signed-off-by: Greg Malysa <malysagreg@gmail.com> Reviewed-by: Peng Fan <peng.fan@nxp.com>
Following header dependency cleanups, an implicit dependence on env.h
was exposed in dwc_eth_qos_adi. However because this driver is not (yet)
enabled in any defconfigs, build tests did not identify the missing
header. This adds the missing #include so that the driver builds
correctly when enabled.
Philip Molloy [Fri, 5 Sep 2025 11:25:13 +0000 (11:25 +0000)]
gpio: adp5588: Add ADP5587 as compatible
The ADP5587 is a simpler version of the ADP5588. The ADP5588 can
configure two pins, C8 and C9, as GPIOs or light sensors. The ADP5587
does not include the light sensors.
Signed-off-by: Philip Molloy <philip@philipmolloy.com>
Marek Vasut [Wed, 3 Sep 2025 14:20:52 +0000 (16:20 +0200)]
env: Remove usb_ignorelist and env_fdt_path from ifdef CONFIG_ENV_VARS_UBOOT_CONFIG
The CONFIG_ENV_VARS_UBOOT_CONFIG should protect only U-Boot
configuration variables in environment, those are arch, cpu,
board, board_name, vendor, soc. It should certainly not hide
usb_ignorelist or env_fdt_path from the environment. Fix it.
Signed-off-by: Marek Vasut <marek.vasut+renesas@mailbox.org> Reviewed-by: Tom Rini <trini@konsulko.com>
fdtdec_get_addr_size uses architecture dependent datatypes which causes
the 32-bit R5 to fail when reading the 64-bit size and addr fields of
reg nodes from the fdt.
Therefore change it to a common api for both 64 and 32 bit platforms to
allow for fdt fixups from R5.
Add CMD_MEMINFO and CMD_MEMINFO_MAP to list of configs implied by
TI_COMMON_CMD_OPTIONS. This allows users to easily view the memory
configuration and the memory maps at runtime.
Marek Vasut [Tue, 2 Sep 2025 22:01:55 +0000 (00:01 +0200)]
mkimage: Add support for bundling TFA BL31 in mkimage -f auto
Introduce two new parameters to be used with mkimage -f auto to bundle
TFA BL31 image into fitImage, using auto-generated fitImage. Add -y to
specify TFA BL31 file name and -Y to specify TFA BL31 load and entry
point address. This is meant to be used with systems which boot all of
TFA BL31, Linux and its DT from a single fitImage, all booted by U-Boot.
Example invocation:
"
$ mkimage -E -A arm64 -C none -e 0x50200000 -a 0x50200000 -f auto \
-d arch/arm64/boot/Image \
-b arch/arm64/boot/dts/renesas/r8a779g3-sparrow-hawk.dtb \
-y ../tfa/build/rcar_gen4/release/bl31.bin -Y 0x46400000 \
/path/to/output/fitImage
"
Documentation update and test are also included, the test validates
both positive and negative test cases, where fitImage does not include
TFA BL31 and does include TFA BL31 blobs.
Signed-off-by: Marek Vasut <marek.vasut+renesas@mailbox.org>
bloblist: adjust default bloblist size after reloc
If neither CONFIG_BLOBLIST_FIXED NOR CONFIG_BLOBLIST_ALLOC is set,
currently CONFIG_BLOBLIST_SIZE_RELOC defaults to 0 except if
* CONFIG_ARM=y && CONFIG_EFI_LOADER=y && GENERATE_ACPI_TABLE=y.
A size of zero never makes sense for a bloblist.
When using QFW we need more than 64 KiB to host the ACPI table.
In this case CONFIG_BLOBLIST_ALLOC is used.
Set a reasonable default.
Remove the CONFIG_BLOBLIST_SIZE_RELOC in ARM QEMU defconfigs which are
not compatible with ACPI tables passed from QEMU.
Reported-by: Emil Renner Berthing <emil.renner.berthing@canonical.com> Fixes: 6f9b015c138b ("common: Enable BLOBLIST_TABLES on arm") Signed-off-by: Heinrich Schuchardt <heinrich.schuchardt@canonical.com> Acked-by: Ilias Apalodimas <ilias.apalodimas@linaro.org>
Tom Rini [Fri, 12 Sep 2025 18:30:15 +0000 (12:30 -0600)]
Merge patch series "arm: mach-k3: am64x: Add reset cause for cpuinfo"
Steffen Kothe <steffen.kothe@skothe.net> says:
AM64X hereby receives support for reset reason printing during boot.
Unfortunately does the AM64X register mapping slighlty differ from
the AM62X. WDT and PORZ are not part of the reset source register, but
the mapping remains the same for all other causes.
To prevent ifdef/else constructs, I decided to follow a simple
copy/paste approach and adjusted the logic accordingly.
board: ti: fdt_ops: make ti_set_fdt_env() const-correct
Make the fdt_map parameter a pointer to const, since the function only
reads the mapping table. This improves API correctness and allows maps
to live in read-only data.
Judith Mendez [Mon, 25 Aug 2025 15:23:10 +0000 (10:23 -0500)]
board: ti: am65x: Overwrite get_overlay_mmc
Unlike other K3 SoC's, am65 SoC has the capability to detect daughter
cards and automatically generate a list of white-space separated overlays
in name_overlays environment variable.
When applied during boot with get_overlay_mmc, the path to overlays with
default distribution is incorrect where path is currently: boot/dtb and
the overlays exist in boot/dtb/ti.
Fix the path in get_overlay_mmc so that overlays are automatically applied
correctly during boot time.
Judith Mendez [Wed, 20 Aug 2025 21:56:34 +0000 (16:56 -0500)]
mach-k3: am62*: Fix backup from eMMC boot mode
Currently logic in spl_mmc_boot_mode only lookes at main devstat
to determine the bootmode to return. Thus, when using: 'eMMC boot'
as primary boot mode and 'MMCSD boot from eMMC UDA' as backup
boot mode, 'eMMC boot' is always selected. Add check for bootindex
to determine if ROM boot via backup boot mode and return MMCSD_MODE_FS
which is the only supported backup bootmode with eMMC device.
virtio: blk: support block sizes exceeding 512 bytes
QEMU allows to specify the logical block size via parameter
logical_block_size of a virtio-blk-device.
The communication channel via virtqueues remains based on 512 byte blocks
even if the logical_block_size is larger.
Consider the logical block size in the block device driver.
Reported-by: Emil Renner Berthing <emil.renner.berthing@canonical.com> Signed-off-by: Heinrich Schuchardt <heinrich.schuchardt@canonical.com> Tested-by: Emil Renner Berthing <emil.renner.berthing@canonical.com>
Tom Rini [Wed, 27 Aug 2025 18:38:36 +0000 (12:38 -0600)]
checkpatch.pl: Ignore mdelay instead of udelay
Whereas in Linux, on ARM there is the notion of delay operations and
mdelay and udelay are not the same, here we just have udelay and mdelay
is a trivial wrapper. Tell checkpatch to not complain here.
Reported-by: Bhimeswararao Matsa <bhimeswararao.matsa@gmail.com> Signed-off-by: Tom Rini <trini@konsulko.com>
Tony Dinh [Tue, 26 Aug 2025 19:45:37 +0000 (12:45 -0700)]
fs: ext4fs: add CONFIG_EXT4_MAX_JOURNAL_ENTRIES to Kconfig
Add maximum ext4 journal entries to Kconfig. It is necessary since the
number of journal entries is proportional to disk capacity. For example,
an ext4 4TB HDD partition could require approximately 500 entries.
Signed-off-by: Tony Dinh <mibodhi@gmail.com> Reviewed-by: Tom Rini <trini@konsulko.com>
Tony Dinh [Tue, 26 Aug 2025 19:37:17 +0000 (12:37 -0700)]
fs: ext4fs: Add initialization failure recovery path in ext4fs_write
Don't invoke ext4fs_deinit() in ext4fs_write() if the failure occurs
during initialization. It would result in a crash since ext4fs_init()
has already done that.
Signed-off-by: Tony Dinh <mibodhi@gmail.com> Reviewed-by: Tom Rini <trini@konsulko.com>
Tom Rini [Tue, 9 Sep 2025 18:44:23 +0000 (12:44 -0600)]
Merge patch series "tools: fdtgrep: Mark util_version() as static"
This series from Ilias Apalodimas <ilias.apalodimas@linaro.org> lays
some of the groundwork for being able to enable the -Wmissing-prototypes
compiler flag while building U-Boot by fixing the obvious problems in
the tools directory.
Brock reports a breakage on an RK3568 SoC. His patch is
correct but he never followed up on the requested changes.
We currently use ldr to calculate the address of __bss_start and
__bss_end. However the absolute addresses of the literal pool are never
relocated and we end up clearing the wrong memory section. Use
PC-relative addressing instead.
Mark Kettenis [Sat, 23 Aug 2025 14:03:56 +0000 (16:03 +0200)]
pci: apple: Fix use of uninitialized variable
Replace use of uninitialized variable with the PCI device number
in an error message as this is what we use elsewhere to derive
the PCIe port number. Use ofnode_read_pci_addr() to read the
PCI address of the node and derive the device number from that.
Signed-off-by: Mark Kettenis <kettenis@openbsd.org> Reported-by: Andrew Goodbody <andrew.goodbody@linaro.org>
Rasmus Villemoes [Tue, 19 Aug 2025 10:55:54 +0000 (12:55 +0200)]
common/spl: use memmove() in load_simple_fit()
I had trouble booting some am335x boards (both
beagleboneblack and a custom board). SPL would start just fine, and
apparently load U-Boot proper, but it would hang when jumping to
U-Boot.
While debugging, I stumbled on this memcpy() which from code
inspection very much looked to have overlapping src and dst, and
indeed a simple printf revealed
calling memcpy(0x8087bf68, 0x8087bf80, 0xf7f8)
Now, it will always be with src > dst, our memcpy()
implementations "most likely" do forward-copying, and in the end it
turned out that this wasn't the culprit after all [*].
But to avoid me or others barking up the wrong tree in the future, and
because this use of memcpy() is technically undefined, use memmove()
instead.
[*] That was 358d1cc232c ("spl: Align FDT load address"), which has
since been fixed in master but not the v2025.07 I worked of by 52caad0d14a ("ARM: Align image end to 8 bytes to fit DT alignment").
Signed-off-by: Rasmus Villemoes <ravi@prevas.dk> Reviewed-by: Heinrich Schuchardt <heinrich.schuchardt@gmx.de>
Tom Rini [Mon, 18 Aug 2025 16:51:19 +0000 (10:51 -0600)]
boot: Add LEGACY_IMAGE_FORMAT to DISTRO_DEFAULTS
At this time there are still major Linux distributions which by default
boot using LEGACY_IMAGE_FORMAT type scripts. Add this option to
DISTRO_DEFAULTS to ensure these platforms can still boot.
Fixes: d780965927d4 ("Drop the special am335x_boneblack_vboot target") Reported-by: Sascha Silbe <sascha-pgp@silbe.org> Tested-By: Sascha Silbe <sascha-pgp@silbe.org> Signed-off-by: Tom Rini <trini@konsulko.com>
Initialize the actread variable to prevent undefined behavior
that can occur if the variable is used before being assigned a
value.
This will help to prevent potential issues, especially if
actread is used (e.g., read, incremented, or returned) before
being explicitly set elsewhere in the code.
Ben Hoelker [Sun, 17 Aug 2025 22:30:00 +0000 (10:30 +1200)]
drivers: rtc: max313xx: Add delay after setting date
The MAX31331 was not correctly updating the seconds when
setting the time and would return the seconds previously set.
Like the MAX31343, a delay needs to be added after setting the
time. Wait one second after writing so that the date command shows the
correct time.
Reviewed-by: Chris Packham <chris.packham@alliedtelesis.co.nz> Reviewed-by: Bruce Adams <bruce.adams@alliedtelesis.co.nz> Signed-off-by: Ben Hoelker <ben.hoelker@alliedtelesis.co.nz>
include: env: ti: Use .env for environment variables
Add omap common environment variables to .env. We retain the old-style C
environment .h files to maintain compatibility with other omap devices that
have not moved to using .env yet.
Marek Vasut [Wed, 3 Sep 2025 11:23:23 +0000 (13:23 +0200)]
arm64: renesas: r8a779g3: Use $loadaddr in bootcmd on Retronix R-Car V4H Sparrow Hawk board
Avoid use of hard-coded address in boot command, instead use $loadaddr
which is the default load address. This improves consistency of the
environment on this board.
Signed-off-by: Marek Vasut <marek.vasut+renesas@mailbox.org>
Marek Vasut [Wed, 3 Sep 2025 11:23:57 +0000 (13:23 +0200)]
ARM: renesas: Enable CONFIG_ENV_VARS_UBOOT_CONFIG on all boards
The CONFIG_ENV_VARS_UBOOT_CONFIG extends U-Boot environment with
variables arch/board/board_name/soc/vendor, which can be used to
discern different devices from each other based purely on U-Boot
environment variables.
Signed-off-by: Marek Vasut <marek.vasut+renesas@mailbox.org>
- Several improvements for kontron-sl-mx6ul.
- Add Phytec imx8mp-libra-fpsc board
- Add redundant environment support for imx8m evk boards.
- Several improvements for phycore-imx93.
Benjamin Hahn [Tue, 2 Sep 2025 06:07:32 +0000 (08:07 +0200)]
Add imx8mp-libra-fpsc board
Add new imx8mp-libra-fpsc board.
Bootph tags as well as USB device tree nodes are in u-boot.dtsi for now
and will be removed when upstreamed.
The Libra i.MX 8M Plus FPSC is a single board computer. It uses an i.MX
8M Plus FPSC [1] System on Module which utilizes the FPSC standard [2].
Primoz Fiser [Thu, 28 Aug 2025 08:20:08 +0000 (10:20 +0200)]
board: phytec: phycore-imx93: Use CONFIG_PHYTEC_EEPROM_BUS
Switch to use CONFIG_PHYTEC_EEPROM_BUS instead of the hard coded value
of 2 for the EEPROM I2C bus for both SPL and U-Boot proper. Possible
since commit 88a1816a9b4e ("board: phytec: common: Add PHYTEC_EEPROM_BUS
to Kconfig").
By default, lets disable configuration option CONFIG_AHAB_BOOT=y on the
phyCORE-i.MX93 based boards. This option is only used in the secureboot
context which is not provided by default anyway. Lets remove it from the
defconfig to not give false impressions it is supported out of the box
for this board.
On the other hand, in the context of PHYTEC secureboot, this option is
selected by the distro which enables CONFIG_AHAB_BOOT among other secure
boot related options and tweaks needed to properly support it.
Signed-off-by: Primoz Fiser <primoz.fiser@norik.com> Reviewed-by: Martin Schwan <m.schwan@phytec.de>
Clean-up "imx93-phyboard-segin-u-boot.dtsi" internal device-tree from
nodes already part of the upstream device-tree since commit 79f3e77133bd
("Subtree merge tag 'v6.16-dts' of dts repo [1] into dts/upstream"). No
functional change is made with this commit.
Eberhard Stoll [Mon, 25 Aug 2025 13:54:34 +0000 (15:54 +0200)]
imx: kontron-sl-mx6ul: Force default environment for serial loader boot
In case of booting from serial loader (USB) we want to always use the
default environment in order to get a defined state that is
independent of any environment stored in persistent memory.
This way we can avoid corruption of the boot process during
development and manufacturing by existing environment settings in
flash.
For some cases it is beneficial to not store the environment in
persistent memory, but instead use the default environment and
keep it in volatile RAM only. Allow this by enabling
CONFIG_ENV_IS_NOWHERE.
Andrew Goodbody [Mon, 1 Sep 2025 15:13:14 +0000 (16:13 +0100)]
spi: exynos: Remove extra term from test
In spi_rx_tx there comes a test for execution of a code block that
allows execution if rxp is not NULL or stopping is true. However all the
code in this block relies on rxp being valid so allowing entry just if
stopping is true does not make sense. So remove this from the test
expression leaving just a NULL check for rxp.
This issue was found by Smatch.
Signed-off-by: Andrew Goodbody <andrew.goodbody@linaro.org> Signed-off-by: Minkyu Kang <mk7.kang@samsung.com>
Andrew Goodbody [Thu, 7 Aug 2025 14:41:18 +0000 (15:41 +0100)]
pinctrl: rzg2l: Variable may not have been assigned to
In rzg2l_pinconf_set and rzg2l_get_pin_muxing if the call to
rzg2l_selector_decode fails then the variable pin may not have been
assigned to. Remove the use of pin from the error message. Also update
the error message to show the invalid selector used instead of port
which will be the error code returned.
This issue was found by Smatch.
Signed-off-by: Andrew Goodbody <andrew.goodbody@linaro.org> Reviewed-by: Paul Barker <paul@pbarker.dev> Reviewed-by: Marek Vasut <marek.vasut+renesas@mailbox.org>
Tom Rini [Tue, 2 Sep 2025 20:11:45 +0000 (14:11 -0600)]
Merge patch series "memtest performance improvements"
Rasmus Villemoes <ravi@prevas.dk> says:
The first two patches make memtest run ~40x faster (when, as it should
be, dcache is disabled), with the second patch being responsible for
most of that. At least on the beagleboneblack which I used for
testing; other boards and configurations will likely see different
numbers.
This is for CONFIG_SYS_ALT_MEMTEST=y and
CONFIG_SYS_ALT_MEMTEST_BITFLIP=n; one could probably get a similar
improvement in the bitflip case since that also has a schedule() call
in the inner loop.
Rasmus Villemoes [Fri, 22 Aug 2025 18:18:48 +0000 (20:18 +0200)]
memtest: remove use of vu_long typedef in mem_test_alt
Hiding a qualifier such as "volatile" inside a typedef makes the code
much harder to understand. Since addr and dummy being
volatile-qualified are important for the correctness of the test code,
make it more obvious by spelling it out as "volatile ulong".
Rasmus Villemoes [Fri, 22 Aug 2025 18:18:47 +0000 (20:18 +0200)]
memtest: only call schedule() once for every 256 words
A function call itself for every word written or read+written in these
loops is bad enough. But since the memory test must be run with dcache
disabled, the schedule() call, traversing the linked list of
registered cyclic clients, and accessing the 'struct cyclic_info' for
each to see if any are due for a callback, is quite expensive. On a
beagleboneblack, testing a modest 16MiB region takes 2.5 minutes:
=> dcache off
=> time mtest 0x81000000 0x82000000 0 1
Testing 81000000 ... 82000000:
Iteration: 1
Tested 1 iteration(s) with 0 errors.
time: 2 minutes, 28.946 seconds
There is really no need for calling schedule() so frequently. It is
quite easy to limit the calls to once for every 256 words by using a
u8 variable. With that, the same test as above becomes 37 times
faster:
=> dcache off
=> time mtest 0x81000000 0x82000000 0 1
Testing 81000000 ... 82000000:
Iteration: 1
Tested 1 iteration(s) with 0 errors.
during those ~4000 milliseconds, so the schedule() calls are still
done less than 0.1ms apart.
These numbers are just for a beagleboneblack, other boards may have a
slower memory, but we are _two orders of magnitude_ away from
schedule() "only" being called at 100Hz, which is still more than
enough to ensure any watchdog is kept happy.
projects / thirdparty / u-boot.git / log
