I always define a disk layout with two separate partitions for the two
copies of the U-Boot environment and, being the one who introduced the
type GUID for such partitions, of course also set those partitions'
type GUID appropriately.
This has worked just fine, but, it turns out, only because I've never
had CONFIG_PARTITION_TYPE_GUID enabled on any of my boards; I've
always just set the offsets of the two partitions via the config
variables CONFIG_ENV_OFFSET(,_REDUND).
I didn't even know that env/mmc.c had learnt to look for the env
partition based on the type GUID, or that that would overrule the
ENV_OFFSET config variables, until I experienced weird random
corruption while doing bringup for an stm32 board, where
PARTITION_TYPE_GUID is automatically set because it is select'ed by
CMD_STM32PROG.
These patches try to fix the code to fit my scheme, while not changing
anything for existing setups that use the two-copies-one-partition
scheme, other than complaining loudly if the system is misconfigured
and avoiding such random corruption of neighbouring partitions.
env: mmc: rework mmc_env_partition_by_guid() to work with two separate partitions
Having two separate partitions for use in a redundant environment
setup works just fine, if one only relies on setting CONFIG_ENV_OFFSET
and CONFIG_ENV_OFFSET_REDUND. However, if CONFIG_PARTITION_TYPE_GUID
is enabled, the current logic in mmc_env_partition_by_guid() means
that only the first partition will ever be considered, and prior to
the previous commit, lead to silent data corruption.
Extend the logic so that, when we are looking for the location for the
second copy of the environment, we keep track of whether we have
already found one matching partition. If a second match is found,
return that, but also modify *copy so that the logic in the caller
will use the last ENV_SIZE bytes of that second partition - in my
case, and I suppose that would be typical, both partitions have been
created with a size of exactly the desired ENV_SIZE.
When only a single matching partition exists, the behaviour is
unchanged: We return that single partition, and *copy is left as-is,
so the logic in the caller will either use the last (copy==0) or
second-to-last (copy==1) ENV_SIZE bytes.
and have set CONFIG_ENV_OFFSET=0x1fc000,
CONFIG_ENV_OFFSET_REDUND=0x1fe000 and CCONFIG_ENV_SIZE=0x2000.
This usually works just fine, but on an stm32mp, I was seeing weird
behaviour. It turns out that can be tracked down to that board setting
CONFIG_PARTITION_TYPE_GUID, so the logic in mmc.c ends up only finding
the first of the two partitions, but then in the copy=1 case ends up
computing 0x1fa000 as the *val returned (that is, the end of the
partition minus two times the environment size). That is of course
outside the found partition and leads to random corruption of the
partition preceding U-Boot-env-1.
Add a sanity check that the partition found is at least as large as
needed for the "one or two copies from the end of the partition" logic
to at least produce something within that partition. That will also
catch a bug where the partition is too small for even one copy of the
environment.
In preparation for fixing the handling of a the case of redundant
environment defined in two separate partitions with the U-Boot env
GUID, refactor the
for ()
if (str)
...
#ifdef CONFIG_FOO
if (!str)
..
#endif
to
if (str)
for ()
else if (CONFIG_FOO && !str)
for ()
and put those for loops in separate functions.
No functional change intended, but I did change the direct access of
info.type_guid into using the disk_partition_type_guid() helper, so
that I could avoid the #ifdef and use IS_ENABLED() in the if() statement.
Marek Vasut [Thu, 26 Sep 2024 23:14:13 +0000 (01:14 +0200)]
power: regulator: Convert regulators_enable_boot_on/off() to regulator_post_probe
Turn regulators_enable_boot_on() and regulators_enable_boot_off() into
empty functions. Implement matching functionality in regulator_post_probe()
instead. The regulator_post_probe() is called for all regulators after they
probe, and regulators that have regulator-always-on or regulator-boot-on DT
properties now always probe due to DM_FLAG_PROBE_AFTER_BIND being set on
such regulators in regulator_post_bind().
Finally, fold regulator_unset() functionality into regulator_autoset().
Marek Vasut [Thu, 26 Sep 2024 23:14:12 +0000 (01:14 +0200)]
power: regulator: Trigger probe of regulators which are always-on or boot-on
In case a regulator DT node contains regulator-always-on or regulator-boot-on
property, make sure the regulator gets correctly configured by U-Boot on start
up. Unconditionally probe such regulator drivers. This is a preparatory patch
for introduction of .regulator_post_probe() which would trigger the regulator
configuration.
Parsing of regulator-always-on and regulator-boot-on DT property has been
moved to regulator_post_bind() as the information is required early, the
rest of the DT parsing has been kept in regulator_pre_probe() to avoid
slowing down the boot process.
usb: gadget: cdns3: Fix missing cache operations for non-zero Endpoints
Transfer initiation and completion for the non-zero Endpoints are
handled by cdns3_ep_run_transfer() and cdns3_transfer_completed()
respectively.
Failing to flush the cache associated with the TRB Pool within
cdns3_ep_run_transfer() results in the transfers never being initiated.
Similarly, failing to invalidate the cache associated with the TRB pool
within cdns3_transfer_completed() results in the transfers never being
completed.
Fix this.
Fixes: 7e91f6ccdc84 ("usb: Add Cadence USB3 host and gadget driver") Signed-off-by: Siddharth Vadapalli <s-vadapalli@ti.com> Reviewed-by: Roger Quadros <rogerq@kernel.org> Reviewed-by: Marek Vasut <marex@denx.de>
Jonas Karlman [Tue, 17 Sep 2024 20:59:17 +0000 (20:59 +0000)]
regulator: rk8xx: Fix buck get and set enabled state on RK806
Wrong POWER_EN reg is used to get and set enabled state for the RK806
buck 4 and 8 regulators, also wrong POWER_SLP_EN0 bit is used for
suspend state for the RK806 buck 1-8 regulators.
Fix this by not adding one to the zero based buck variable.
Fixes: f172575d92cd ("power: rk8xx: add support for RK806") Signed-off-by: Jonas Karlman <jonas@kwiboo.se> Reviewed-by: Kever Yang <kever.yang@rock-chips.com> Reviewed-by: Quentin Schulz <quentin.schulz@cherry.de>
Simon Glass [Mon, 26 Aug 2024 19:11:40 +0000 (13:11 -0600)]
binman: Provide a way to set the symbol base address
The base address of the ELF containing symbols is normally added to
any symbols written, so that the value points to the correct address in
memory when everything is loaded. When the binary resides on disk, a
different offset may be needed, typically 0. Provide a way to specify
this.
Simon Glass [Mon, 26 Aug 2024 19:11:39 +0000 (13:11 -0600)]
binman: Add minor improvements to symbol-writing
Add a clarification to the documentation and add a missing comment. Also
update the test so that when it fails it is easier to see what is going
on, rather than having to decode hex strings.
Simon Glass [Mon, 26 Aug 2024 19:11:38 +0000 (13:11 -0600)]
binman: Adjust naming for reading symbols
These functions get the value of a symbol. The reference to ELF files
is confusing since they are reading the position/size of entries, not
ELF symbols. Rename the functions and adjust the comments also.
Simon Glass [Mon, 26 Aug 2024 19:11:36 +0000 (13:11 -0600)]
binman: fit: Set the image_pos attributes only once
The section etype has its own implementation of SetImagePos(), most of
which is not useful since the code is included here. So call
Entry.SetImagePos() which has the only piece of this which we actually
want.
Simon Glass [Thu, 15 Aug 2024 19:57:45 +0000 (13:57 -0600)]
buildman: Allow skipping the dtc build
For most boards, the device-tree compiler is built in-tree, ignoring the
system version. Add a special option to skip this build. This can be
useful when the system dtc is up-to-date, as it speeds up the build.
Simon Glass [Sun, 23 Jun 2024 17:55:14 +0000 (11:55 -0600)]
buildman: Retry the build for current source
Buildman retries a failed build when processing a branch, but does not
do this when building current source. It is useful to do this retry in
both cases, so add the logic for it.
Marek Vasut [Tue, 24 Sep 2024 22:08:23 +0000 (00:08 +0200)]
arm: imx: Enable always-on regulators using board specific workaround on DHCOM SoM
U-Boot is currently unable to automatically enable regulators which
contain DT property regulator-always-on or regulator-boot-on. There
is an ongoing work to add this functionality to regulator core code,
but until the proper solution lands, add this awful workaround code
to force these regulators on. This is specifically needed to assure
the VIO regulator used to supply ethernet magnetics is enabled.
Marek Vasut [Mon, 8 Jul 2024 11:43:23 +0000 (13:43 +0200)]
ARM: stm32: Fix secure_waitbits() mask check
Do not apply bitwise AND to register value and expected value, only
apply bitwise AND to register value and mask, and only then compare
the result with expected value that the function polls for.
Fixes: b49105320a5b ("stm32mp: psci: Implement PSCI system suspend and DRAM SSR") Signed-off-by: Marek Vasut <marex@denx.de>
Marek Vasut [Tue, 18 Jun 2024 22:57:25 +0000 (00:57 +0200)]
ARM: stm32: Fix TAMP_SMCR BKP..PROT fields on STM32MP15xx
Update the TAMP_SMCR BKP..PROT fields to put first 10 registers
into protection zone 1 and next 5 into zone 2. This fixes use of
boot counter which is often in zone 3 and has to be updated from
Linux, which runs in NS.
Fixes: 73f7fc944cf6 ("ARM: stm32: Initialize TAMP_SMCR BKP..PROT fields on STM32MP15xx") Signed-off-by: Marek Vasut <marex@denx.de> Reviewed-by: Patrice Chotard <patrice.chotard@foss.st.com>
Jonas Karlman [Wed, 24 Jul 2024 22:47:11 +0000 (22:47 +0000)]
power: regulator: Consistently return -ENOSYS when ops is unimplemented
dev_get_driver_ops() may return NULL when the udevice is invalid.
Move the ops check to top of functions to consistently return -ENOSYS
when ops is unimplemented and prevent trying to access uclass plat data,
also add missing NULL checks to suspend ops.
This result in suspend_uV having the value -ENODATA after probe.
This negative voltage, -ENODATA, gets missinterpreted and result in an
unexpected voltage being set by autoset.
E.g. on Radxa ROCK 3A the vcc_ddr regulator by default have a normal and
suspend voltage value of 0.5v. However, due to this missinterpretation
the suspend voltage end up beind set to 0.5625v instead.
Fix this by skip calling regulator_set_suspend_value() in autoset and
also protect calling set value ops when input value is -ENODATA.
Ye Li [Tue, 24 Sep 2024 07:32:00 +0000 (15:32 +0800)]
net: fec_mxc: Skip recv packet process when fec is halted
After FEC is halted by calling fec_halt callback, we should not continue
receiving packet. Otherwise it will process previous pending interrupts
on EIR register and uses wrong rbd index as this has been reset to 0.
The GRA interrupt which is triggered by issuing graceful stop command to
FEC transmitter in fec_halt is processed in this case. It causes wrong
receive buffer descriptors be used by FEC in next time.
Signed-off-by: Ye Li <ye.li@nxp.com> Reviewed-by: Peng Fan <peng.fan@nxp.com> Signed-off-by: Peng Fan <peng.fan@nxp.com>
Peng Fan [Tue, 24 Sep 2024 07:31:59 +0000 (15:31 +0800)]
net: fec_mxc: Fix clk_ref rate on iMX93
i.MX93 FEC ENET port supports two mode: RGMII and RMII. For RGMII,
there is an internal /2 divider, so the freq needs to set with (*2),
otherwise the speed will not reach 1G and cause communication error
in some network environments. For RMII, the clk path is
ccm -> enet tx_clk pin -> pad loop back to enet, no /2 divider.
So fix for RGMII mode with freq multiplied by 2.
Fixes: 09de565f76b ("net: fec_mxc: support i.MX93") Signed-off-by: Ye Li <ye.li@nxp.com> Signed-off-by: Peng Fan <peng.fan@nxp.com>
Marek Vasut [Tue, 24 Sep 2024 13:02:09 +0000 (15:02 +0200)]
arm64: dts: imx8m: Always name the generated fitImage u-boot.itb
Maintain backward compatibility with pre-binman u-boot file naming,
the U-Boot fitImage used to be named u-boot.itb before, restore the
file name after binman conversion.
Signed-off-by: Marek Vasut <marex@denx.de> Reviewed-by: Peng Fan <peng.fan@nxp.com>
Miscellaneous fixes made when developing the lwIP series [1]. They are
posted separately since they make sense on their own. Subsequent
versions of the lwIP series will contain a squashed version of this one.
common/update.c (UPDATE_COMMON) depends on tftp_timeout_ms which is
defined in cmd/tftp.c (CMD_TFTPBOOT). Therefore add the appropriate
imply statement to Kconfig.
Signed-off-by: Jerome Forissier <jerome.forissier@linaro.org> Reviewed-by: Peter Robinson <pbrobinson@gmail.com>
net: fec_mxc_init(): do not ignore return status of fec_open()
The fec_mxc_init() function currently always returns 0. This does not
allow the callers to detect when for instance the PHY initialization
failed due to the port being unconnected. Fix that by returning the
status of fec_open().
wget followed by bootefi currently fails as follows:
U-Boot> wget 200000 192.168.0.30:helloworld.efi
Waiting for Ethernet connection... done.
HTTP/1.0 200 OK
Packets received 13, Transfer Successful
Bytes transferred = 12720 (31b0 hex)
U-Boot> bootefi 200000
No UEFI binary known at 200000
U-Boot>
Fix the problem by adding the missing efi_set_bootdev() call.
Signed-off-by: Jerome Forissier <jerome.forissier@linaro.org> Reviewed-by: Peter Robinson <pbrobinson@gmail.com> Reviewed-by: Ilias Apalodimas <ilias.apalodimas@linaro.org> Reviewed-by: Simon Glass <sjg@chromium.org>
FTGMAC100 enables drivers/net/ftgmac100.c which uses
PHY_INTERFACE_MODE_NCSI, which is defined only when PHY_NCSI is enabled.
Therefore FTGMAC100 depends on PHY_NCSI. However adding such a
dependency causes a "recursive dependency detected!" message, so
add a dependency on NET instead (PHY_NCSI depends on NET).
All in all, either the stack is NET and FTGMAC100 can be enabled,
or it is NET_LWIP (or NO_NET) and it cannot.
Signed-off-by: Jerome Forissier <jerome.forissier@linaro.org> Reviewed-by: Peter Robinson <pbrobinson@gmail.com> Reviewed-by: Simon Glass <sjg@chromium.org>
PHY_NCSI enables drivers/net/phy/ncsi.c which calls net_loop() and
net_set_timeout_handler(). That's the legacy NET stack (as opposed to
NET_LWIP). Therefore add the dependency to Kconfig.
The AT91-based platforms have a mem_init() function declared in
arch/arm/mach-at91/include/mach/at91_common.h and implemented in various
places. In preparation of the introduction of the lwIP networking library
which also has a global mem_init() function, rename the AT91 one to
at91_mem_init().
Prefix the flash status codes (ERR_*) with FL_ in order to avoid clashes
with third-party libraries. Case in point: including the lwIP library
header file <lwip/err.h> which defines err_enum_t as an enum with values
being ERR_*.
Signed-off-by: Jerome Forissier <jerome.forissier@linaro.org> Reviewed-by: Tom Rini <trini@konsulko.com> Reviewed-by: Ilias Apalodimas <ilias.apalodimas@linaro.org> Reviewed-by: Peter Robinson <pbrobinson@gmail.com>
buildman/toolchain.py: do not set CROSS_COMPILE for sandbox
When building for sandbox, self.cross is empty.
In MakeEnvironment(), CROSS_COMPILE is defined to be self.cross (with
or without a full path), optionally prefixed by the toolchain wrapper
defined in ~/.buildman. This is fine when self.cross is not empty, but
it doesn't make sense when it is:
- Either there is no wrapper and we end up with an empty CROSS_COMPILE
which is the same as not defining it (the host compiler will be used),
- Or there is a wrapper and CROSS_COMPILE will contain only the wrapper
which obviously is not a valid compiler, hence an error.
Test case:
$ sudo apt install ccache
$ grep -q toolchain-wrapper ~/.buildman || \
printf "[toolchain-wrapper]\nwrapper = ccache\n" >>~/.buildman
$ make mrproper
$ ./tools/buildman/buildman sandbox_noinst
$ ./tools/buildman/buildman sandbox_noinst
Building current source for 1 boards (1 thread, 24 jobs per thread)
sandbox: + sandbox_noinst
+arch/sandbox/lib/reloc_sandbox_efi.c:10:15: error: operator '==' has no left operand
+ 10 | #if HOST_ARCH == HOST_ARCH_X86_64
+ | ^~
[...]
The GetEnvArgs function is modified too, since the VAR_CROSS_COMPILE
case has the same issue.
In tools/buildman/test.py, testGetEnvArgs is extended and
testMakeEnvironment is added. They check the 'arm' and 'sandbox'
toolchains, with and without a wrapper.
Signed-off-by: Jerome Forissier <jerome.forissier@linaro.org> Reviewed-by: Simon Glass <sjg@chromium.org>
When running the bootstage tests currently we get a warning like:
tests/test_bootstage.py::test_bootstage_stash
...PytestReturnNotNoneWarning: Expected None, but tests/test_bootstage.py::test_bootstage_stash returned (37748736, 4096), which will be an error in a future version of pytest. Did you mean to use `assert` in stead of `return`?
This is because the unstash test will run the stash test and fetch the
addr / size from that. Rework the test to be stash and unstash and then
run the unstash command at the end of the current stash test.
Acked-by: Michal Simek <michal.simek@amd.com> Reviewed-by: Love Kumar <love.kumar@amd.com> Reviewed-by: Simon Glass <sjg@chromium.org> Signed-off-by: Tom Rini <trini@konsulko.com>
Make SPL_RAM_SUPPORT a hidden Kconfig symbol, automatically selected
by SPL_RAM_DEVICE or SPL_DFU. Avoids the situation where SPL_RAM_SUPPORT
may be enabled without the other two being enabled, which results in the
following build warning:
common/spl/spl_ram.c:19:14: warning: ‘spl_ram_load_read’ defined but not used [-Wunused-function]
19 | static ulong spl_ram_load_read(struct spl_load_info *load, ulong sector,
| ^~~~~~~~~~~~~~~~~
Signed-off-by: Jerome Forissier <jerome.forissier@linaro.org> Reviewed-by: Sean Anderson <seanga2@gmail.com> Reviewed-by: Tom Rini <trini@konsulko.com> Reviewed-by: Mattijs Korpershoek <mkorpershoek@baylibre.com>
Udit Kumar [Thu, 29 Aug 2024 08:34:53 +0000 (14:04 +0530)]
ram: k3-ddrss: Handle error properly in lpddr4_start
In lpddr4_start function error returned by lpddr4_enablepiinitiator
may go undetected due to overwrite of return code.
Add support to handle error in above case.
Marek Vasut [Fri, 6 Sep 2024 17:10:42 +0000 (19:10 +0200)]
mmc: Poll CD in case cyclic framework is enabled
In case the cyclic framework is enabled, poll the card detect of already
initialized cards and deinitialize them in case they are removed. Since
the card initialization is a longer process and card initialization is
done on first access to an uninitialized card anyway, avoid initializing
newly detected uninitialized cards in the cyclic callback.
Signed-off-by: Marek Vasut <marek.vasut+renesas@mailbox.org> Reviewed-by: Simon Glass <sjg@chromium.org>
Added LMB API to prevent SF command from overwriting reserved
memory areas. The current SPI code does not use LMB APIs for
loading data into memory addresses. To resolve this, LMB APIs
were added to check the load address of an SF command and ensure it
does not overwrite reserved memory addresses. Similar checks are
used in TFTP, serial load, and boot code to prevent overwriting
reserved memory.
wget: rework the logic to validate the load address
Use the lmb_read_check() function to verify if it is safe to use a
region of memory to load data from the wget command. The current logic
checks the amount of free memory available, starting from the 'load
address'. This call fails if the same region of memory has been used
earlier. This used to work earlier as the LMB memory map had a local
scope and was not persistent. Fix this issue by using the
lmb_read_check() call instead which only returns an error in case the
memory region has been marked for not allowing re-use.
tftp: rework the logic to validate the load address
Use the lmb_read_check() function to verify if it is safe to use a
region of memory to load data from a tftp command. The current logic
checks the amount of free memory available, starting from the 'load
address'. This call fails if the same region of memory has been used
earlier. This used to work earlier as the LMB memory map had a local
scope and was not persistent. Fix this issue by using the
lmb_read_check() call instead which only returns an error in case the
memory region has been marked for not allowing re-use.
Popup the bootmenu for 5 sec with default boot options to CC on AMD CCs and
default to SOM on others.
Users can anytime disable the bootmenu by setting the variable
enable_bootmenu=0 in zynqmp_kria.env or setup it up at run time and save
variables to NV memory.
The patch is also fixing issue created by commit 4c7363068651 ("cmd:
setexpr: fix no matching string in gsub return empty value") which has
changed return value from setexpr command (Before this commit when
matching string wasn't found command return 1 that's why was possible to
use with if).
Michal Simek [Fri, 13 Sep 2024 09:28:48 +0000 (11:28 +0200)]
arm64: zynqmp: Align mini-qspi DT with dt-schema
fixed-clock can't be described on the bus because it is missing reg
property. Also remove additional compatible string for flash. Mini qspi
configuration is used with multiple different flashes that's why describing
only one is not correct but also not required based on DT schema.
Michal Simek [Fri, 13 Sep 2024 09:28:44 +0000 (11:28 +0200)]
xilinx: Fix axi and mmc node names in mini configuration
amba is not approved node name for simple-bus that's why use axi instead to
be aligned with other xilinx boards. Node reference is not changed that's
why there is no impact but also mini configuration will never gets to OS
that's why nothing should be affected from OS perspective (paths in /proc/
for example).
Michal Simek [Fri, 13 Sep 2024 07:37:38 +0000 (09:37 +0200)]
net: gem: Remove undocumented is-internal-pcspma dt flag
Generic understanding/consideration is that phy-mode as sgmi means that the
internal PCS(Physical Coding Sublayer) should be enabled by default.
Xilinx GEM implementation allows configuration GEM (gmii mode) + PL PCS PMA
(sgmii mode, Physical Medum Attachment) but in this case phy-mode should be
setup as gmii.
The reason for this assumption is that phy-mode should be described based
on GEM configuration not based on mode coming out of PHY.
Also Linux kernel automatically setting up PCSSEL bit when phy mode is
sgmii without a need to specified additional DT propety.
All our DTSes with sgmii phy mode have this flag enabled that's why there
is no need/reason to just duplicate information.
board: xilinx: Remove conditional check for Microblaze
U-Boot is not picking boot.scr script address from device tree
rather it's using default address for Microblaze platform,
and it's picking for other platforms. Remove conditional check
for Microblaze platform, so that u-boot pick up boot.scr script
address for all platforms from device tree.
mmc: zynq_sdhci: Remove device_is_compatible() function
There are lot of device_is_compatible() present in the driver.
Remove them and replace with a variables "SDHCI_COMPATIBLE_SDHCI_89A"
and "SDHCI_COMPATIBLE_VERSAL_NET_EMMC" with match data.
This change saves the space and reduce the execution time.
arm64: zynqmp: Remove overlays and add new dtb entries for ZynqMP
Remove device tree overlay (DTBO) entries for the ZynqMP target
from the Makefile. Add new device tree binaries (DTBs) for the
zynqmp-sm-k24-revA and zynqmp-smk-k24-revA configurations.
kbuild: cherry-pick kbuild fdtoverlay changes from linux
Linux commits: 15d16d6dadf6 kbuild: Add generic rule to apply fdtoverlay 44f87191d105 kbuild: parameterize the .o part of suffix-search
The Linux commit 15d16d6dadf6 adds a generic rule in Makefile.lib
to automatically apply fdtoverlay, so that each platform doesn't
need to include a complex rule. This also automatically appends
DTC_FLAGS_foo_base += -@ to all base files
The platform's Makefile only needs to have this now:
projects / u-boot.git / log