sdhci_brcmstb_init_2712() reads host->mmc->host_caps to decide whether
to force card-detect for a non-removable eMMC, or to route the CD signal
for a removable SD card. At the time this function runs from
sdhci_bcmstb_probe(), however, host->mmc->host_caps is still zero, that
field is only populated later by the MMC uclass, after the driver's
probe returns. mmc_of_parse() has already filled plat->cfg.host_caps
from the device tree by this point, so check that field instead.
Without the fix, every BCM2712 SDHCI instance takes the else branch and
writes SDIO_CFG_SD_PIN_SEL = SDIO_CFG_SD_PIN_SEL_CARD (0x02), including
the non-removable eMMC on boards such as CM5 on Home Assistant Yellow.
The SDIO_CFG block lies outside the SDHCI core's reset scope, so this
value persists across SDHCI_RESET_ALL into the next stage. On the
BCM2712, having SD_PIN_SEL set to "SD" when the Linux kernel performs
its first set_power(MMC_POWER_UP) write racily prevents the SDHCI
POWER_ON bit from latching (see [1] for the whole backstory) - the
voltage bits stick but POWER_ON drops - which wedges the first CMD0 the
full 10 s software timeout. On Home Assistant Yellow this manifested as
a ~20 s eMMC probe delay on roughly one in two Linux boots when U-Boot
was the previous stage. Booting directly from the Pi firmware (no U-Boot
in between) left SD_PIN_SEL at its default and did not exhibit the race.
Reading plat->cfg.host_caps lets init_2712 see the "non-removable"
property and take the correct branch, leaving SD_PIN_SEL untouched for
the eMMC.
[1] https://github.com/home-assistant/operating-system/pull/3700#issuecomment-4430229511
Fixes: 10127cdbab ("mmc: bcmstb: Add support for bcm2712 SD controller")
Signed-off-by: Jan Čermák <sairon@sairon.cz>
Reviewed-by: Ivan T. Ivanov <iivanov@suse.de>
The bcm2835 SDHCI driver sets up the MMC host configuration via
sdhci_setup_cfg(), but does not parse generic MMC device tree
properties.
As a result, properties such as bus-width are ignored. On Raspberry Pi
Compute Module 4, the eMMC node describes an 8-bit bus, but U-Boot
initialized the device as 4-bit.
Call mmc_of_parse() before sdhci_setup_cfg() so that generic MMC
properties are folded into the host configuration before the MMC core
selects the bus width.
Before this change, mmc info reported:
Bus Speed: 52000000
Bus Width: 4-bit
After this change, mmc info reports:
Bus Speed: 52000000
Bus Width: 8-bit
Tested on Raspberry Pi Compute Module 4 with onboard eMMC.
Signed-off-by: Liel Harel <liel.harel@gmail.com>
Reviewed-by: Peter Robinson <pbrobinson@gmail.com>
Tested-by: Peter Robinson <pbrobinson@gmail.com> # on the CM4 as well
The 'brcm,bcm2712-hdmi0' compatible string is used on RPi5.
There appears to be no change that impacts early boot output
on the display controller so add the RPi5 compatible string.
Signed-off-by: Peter Robinson <pbrobinson@gmail.com>
Reviewed-by: Matthias Brugger <mbrugger@suse.com>
Our official domain is now u-boot-project.org, so update all in-tree
references to use the correct domain.
Reviewed-by: Tony Dinh <mibodhi@gmail.com>
Reviewed-by: Peter Robinson <pbrobinson@gmail.com>
Signed-off-by: Tom Rini <trini@konsulko.com>
When fpga load is called with a misaligned buffer address, the
versal_align_dma_buffer() function shifts the pointer forward to the
next aligned boundary and uses memcpy() to copy the data. Since the
destination is ahead of the source and the regions overlap, memcpy()
produces undefined behavior; in practice U-Boot's generic memcpy()
copies forward, repeating the first ARCH_DMA_MINALIGN-aligned chunk
throughout the buffer.
Replace memcpy() with memmove() which correctly handles overlapping
regions by copying backwards when the destination is ahead of the
source.
Fixes: 26e054c943 ("arm64: versal: fpga: Add PL bit stream load support")
Signed-off-by: Pranav Tilak <pranav.vinaytilak@amd.com>
Signed-off-by: Michal Simek <michal.simek@amd.com>
Link: https://lore.kernel.org/r/20260507113359.3665220-1-pranav.vinaytilak@amd.com
Add Renesas R-Car R8A78000 X5H MDLC power domain and reset driver,
which serves as a remap driver between DT power domain and reset IDs
and SCMI power domain and reset IDs in case U-Boot runs on Cortex-A,
and as a direct hardware access driver for RSIP.
The R-Car X5H SCP firmware uses different SCMI power domain and
reset IDs in different versions of the SCP firmware, which makes
this remapping necessary. The SCMI base protocol version is updated
for each new SCP firmware version, it is therefore possible to
determine which SCP firmware version is running on the platform
from the base protocol and then determine which remapping table to
use for DT power domain and reset ID to SCMI power domain and reset
ID remapping.
Currently supported versions are SCP 4.28, 4.31, 4.32 .
The DT power domain and reset ID to SCMI power domain and reset ID
remap and call mechanism is simple. Unlike SCMI clock protocol driver,
the SCMI reset and power domain protocol drivers register only a single
device. This driver looks up that single device, obtains its reset or
power domain ops, sets up struct reset_ctl or struct power_domain with
remapped SCMI ID, and invokes operations directly on the device.
In case of RSIP, all power domains are already enabled by BootROM or
early SoC initialization code, the driver therefore only acts as a
stub for the power domain part. The reset part operates as a direct
hardware access reset driver.
Signed-off-by: Marek Vasut <marek.vasut+renesas@mailbox.org>
Add Renesas R-Car R8A78000 X5H CPG clock driver, which serves as a
remap driver between DT clock IDs and SCMI clock IDs in case U-Boot
runs on the Cortex-A, and as a trivial clock driver for RSIP.
The R-Car X5H SCP firmware uses different SCMI clock IDs in different
versions of the SCP firmware, which makes this remapping necessary.
The SCMI base protocol version is updated for each new SCP firmware
version, it is therefore possible to determine which SCP firmware
version is running on the platform from the base protocol and then
determine which remapping table to use for DT clock ID to SCMI clock
ID remapping.
Currently supported versions are SCP 4.28, 4.31, 4.32 .
The DT clock ID to SCMI clock ID remap and call mechanism is a bit
complex. The driver looks up the SCMI clock protocol device on probe
and stores pointer to it in private data. On each clock request which
has to be remapped, the device sequence ID of this SCMI clock protocol
device is incremented by the remapped SCMI clock ID + 1 and used to
look up matching clock device by sequence number. If the device is
found, it is converted to clock, which can be used in regular clock
operations. This look up has to be done because the SCMI clock driver
registers a subdevice for each clock, and this look up is the only way
to find the correct SCMI clock subdevice. Since the SCMI device and
the clock subdevices are registered in the same function, we can depend
on the device sequence numbers to be monotonically incrementing, with
SCMI clock protocol device being sequence number N, the first SCMI
clock subdevice being sequence number N+1 and so on.
In case of RSIP, all clocks are already enabled by BootROM or early
SoC initialization code, the driver therefore only acts as a stub.
Signed-off-by: Marek Vasut <marek.vasut+renesas@mailbox.org>
Endpoints should not be disabled on bus reset inside UDC driver,
otherwise a race condition will happen between gadget driver. Gadget
driver will free the requests and disable endpoints in disconnect ops.
Also remove outdated comment about it in usba_ep_disable().
Signed-off-by: Zixun LI <admin@hifiphile.com>
Reviewed-by: Mattijs Korpershoek <mkorpershoek@kernel.org>
Fixes: 59310d1ecb ("usb: gadget: introduce 'enabled' flag in struct usb_ep")
Link: https://patch.msgid.link/20260515-udc_ep-v2-1-cd335b4e62e4@hifiphile.com
[mkorpershoek: removed empty newline between Fixes: and sob]
Signed-off-by: Mattijs Korpershoek <mkorpershoek@kernel.org>
The PRP setup code advances prp_pool using u64 pointer
arithmetic:
prp_pool += page_size;
This increments the pointer by page_size * sizeof(u64)
bytes instead of page_size bytes, resulting in invalid
PRP list addresses when multiple PRP list pages are
required.
The issue becomes visible for large transfers, typically
above 2 MiB when MDTS > 9.
Fix it by using byte-wise pointer arithmetic when
advancing to the next PRP list page.
Signed-off-by: Prashant Kamble <prashant.kamble223@gmail.com>
Reviewed-by: Neil Armstrong <neil.armstrong@linaro.org>
Link: https://patch.msgid.link/20260518022535.17197-1-prashant.kamble223@gmail.com
Signed-off-by: Neil Armstrong <neil.armstrong@linaro.org>
Set the ops structure as static const. The structure is not accessible
from outside of this driver and is not going to be modified at runtime.
Signed-off-by: Marek Vasut <marek.vasut+renesas@mailbox.org>
Reviewed-by: Peng Fan <peng.fan@nxp.com>
Signed-off-by: Peng Fan <peng.fan@nxp.com>
Since commit 7917c2e356 ("spi: fsl_espi: fix din offset") MTD is
basically broken because any read transaction will get wrong data. While
the commit in question will fix simple transfers (where both
SPI_XFER_BEGIN and SPI_XFER_END is set), it will break the most common
case, where opcode and address is send first and then data comes as a
second transfer.
This basically reverts commit 7917c2e356 ("spi: fsl_espi: fix din
offset") and make the fix particular for this simple case. Instead of
providing two buffers for reading and writing, just malloc one which is
used for both. This will work because the data is first written on the
SPI bus and then it will be read (and overwite the written data) into
the same buffer.
Suggested-by: Tomas Alvarez Vanoli <tomas.alvarez-vanoli@hitachienergy.com>
Fixes: 7917c2e356 ("spi: fsl_espi: fix din offset")
Signed-off-by: Michael Walle <mwalle@kernel.org>
Signed-off-by: Peng Fan <peng.fan@nxp.com>
The last users of this legacy i2c stack have been removed or converted
to a modern part of the stack instead. Remove this code and references
to it.
Signed-off-by: Tom Rini <trini@konsulko.com>
Reviewed-by: Heiko Schocher <hs@nabladev.com>
Set the ops structure as static. The structure is not accessible
from outside of this driver.
Reviewed-by: Heiko Schocher <hs@nabladev.com>
Signed-off-by: Marek Vasut <marek.vasut+renesas@mailbox.org>
The virtio_blk_do_single_req function returns ulong, which normally is
the processed size, but in an error case can be the actual error. Use
the special IS_ERR_VALUE macro to test for error.
Addresses-Coverity-ID: CID 645833 (DEADCODE) & CID 645834 (NO_EFFECT)
Signed-off-by: Christian Pötzsch <christian.poetzsch@kernkonzept.com>
Reviewed-by: Simon Glass <sjg@chromium.org>
The main use case for u-boot on Apple silicon based devices is to
provide an EFI based bootloader for operating systems. This uses a
generic u-boot image with DTBs passed from an earlier boot loader
(m1n1). Use the generic board name "mac" for this purpose.
Signed-off-by: Janne Grunau <j@jannau.net>
The Cadence SD6HC (SDHCI spec v4.20+) controller uses a soft PHY whose
DLL delay characteristics vary with PVT (Process, Voltage, Temperature)
and board-level trace routing.
A static delay value programmed via device tree for SD High Speed mode is
insufficient because the optimal sampling point varies per board, SD card,
and operating conditions. Runtime calibration is required.
While the SD Physical Layer Specification does not mandate tuning for
SD HS mode (only for UHS-I SDR50/SDR104), the Cadence SD6HC PHY
requires runtime calibration of its receive data delay line to find a
valid sampling window under constrained clock conditions.
The tuning is triggered from the set_ios_post callback because at that
moment hardware has committed the new bus width, clock frequency, and speed
mode to the controller registers. This ensuring the tuning sequence runs
at the correct SD HS operating conditions.
The tuning is gated by a device tree property "cdns,sd-hs-tuning" so
that only boards requiring runtime calibration opt in. When enabled,
the driver performs a 40-tap DLL sweep using CMD19 to find the largest
consecutive passing window, then programs the midpoint into
PHY_DLL_SLAVE_CTRL_REG.
To enable on a board, add to the MMC node in device tree:
&mmc {
cdns,sd-hs-tuning;
};
Signed-off-by: Tze Yee Ng <tze.yee.ng@altera.com>
The following compilation error occurs when environment variable
KBUILD_OUTPUT is not set :
drivers/reset/stm32/stm32-reset-mp21.c:8:10: fatal error: stm32-reset-core.h: No such file or directory
8 | #include <stm32-reset-core.h>
| ^~~~~~~~~~~~~~~~~~~~
As stm32-reset-core.h is located in same directory than stm32-reset-mp21.c,
we should use #include "stm32-reset-core.h".
Signed-off-by: Patrice Chotard <patrice.chotard@foss.st.com>
Reviewed-by: Raphaël Gallais-Pou <rgallaispou@gmail.com>
The dwc3_free_one_event_buffer() function incorrectly called free()
on event buffer structures allocated with devm_kzalloc(). This
caused heap corruption and a synchronous abort when exiting
fastboot mode via "fastboot continue".
Device-managed memory is automatically freed when the device is
removed, so manual deallocation causes the heap allocator to access
corrupted metadata.
Fixes: 884b10e86a ("usb: dwc3: core: fix memory leaks in event buffer cleanup")
Signed-off-by: Balaji Selvanathan <balaji.selvanathan@oss.qualcomm.com>
Reviewed-by: Marek Vasut <marek.vasut+usb@mailbox.org>
The symbol USB_EMUL is how sandbox has access to USB. It's
implementation however enforces a few other requirements. It must have
SCSI enabled, and in turn that means it must have BLK enabled. Finally,
we should not be using SANDBOX itself as a symbol to decide what to
build or not build here, as SANDBOX is selected for COMPILE_TEST builds
as well and so may not have enabled the sandbox specific USB support.
Signed-off-by: Tom Rini <trini@konsulko.com>
Reviewed-by: Marek Vasut <marek.vasut+usb@mailbox.org>
As exposed by "make randconfig", we have an issue with the dependencies
for RESET_RZG2L_USBPHY_CTRL. As this functionally depends on
REGULATOR_RZG2L_USBPHY, express this dependency directly in Kconfig as
well.
Signed-off-by: Tom Rini <trini@konsulko.com>
Reviewed-by: Marek Vasut <marek.vasut+renesas@mailbox.org>
The USB_RENESAS_USBHS functionality can only work with DM_USB_GADGET
enabled, so express this dependency in Kconfig.
Signed-off-by: Tom Rini <trini@konsulko.com>
Reviewed-by: Marek Vasut <marek.vasut+renesas@mailbox.org>
Commit 70101c3217 ("virtio: mmio: Return error codes on probe
failures") returns -ENODEV where it would return 0 before. That path is
apparently hit in qemu and breaks boot device discovery
(virtio_bootdev_hunt() expects only ENOENT). So return -ENOENT in
that path instead.
The remaining two error returns in the function are untouched as I
don't know where they play a role.
Signed-off-by: Ludwig Nussel <ludwig.nussel@siemens.com>
Reviewed-by: Simon Glass <sjg@chromium.org>
Acked-by: Kuan-Wei Chiu <visitorckw@gmail.com>
A fixed gigabit link on a non-gigabit controller is only rejected
during PHY init (even though there is no PHY to init), because, on
device-tree parsing, the controller is not probed, and it is still
unknown whether it is gigabit-capable.
This code was only tested on emulator with a full-duplex RGMII
interface, but is expected to work in GMII or half-duplex as well.
Signed-off-by: Christian DREHER <christian.dreher@nanoxplore.com>
Cadence Ethernet MAC has a feature named user_io, which provides
some input and some output signals for arbitrary purpose in the SoC.
From the driver code, I understand that, on Atmel SoC, it is used to
drive the PHY mode.
At least on Cadence IP7014 r1p12, this feature is optional, and I am
working on a SoC that does not instantiate it. The presence of this
feature is advertised in DCFG1, this patch merely disables the access
to the user_io register based on this information.
I did not apply this change to the non-gigabit capable versions of
the IP, as I do not have documentation for them, and a new non-gigabit
instance is unlikely to appear. I prefer avoiding regressions on old
systems.
Signed-off-by: Christian DREHER <christian.dreher@nanoxplore.com>
The MACB uses specific address registers (SA Top and Bottom) to
filter source or destination MAC addresses.
On the Gigabit Ethernet version, SA1B is @0x88.
On the non-GEM version, SA1B is @0x98.
Before this commit, the code was always writing 0x98. By chance,
on GEM, this is the address of SA3B, allowing the driver to work
anyway.
The motivation for this change is to be able to use the driver on
an instance of the GEM with less than 4 SA registers.
Signed-off-by: Christian DREHER <christian.dreher@nanoxplore.com>
It does not exist in my setup (an on-going arm64 SoC), and removing
it does not cause any missing declaration, but some code called when
CONFIG_CLK is missing calls get_macb_pclk_rate, which is only defined
in arch/arm/mach-at91/include/mach/clk.h
Signed-off-by: Christian DREHER <christian.dreher@nanoxplore.com>
Replace the proprietary airoha,pnswap-rx / airoha,pnswap-tx boolean
device tree properties with the standard rx-polarity and tx-polarity
properties defined in phy-common-props.yaml.
Backward compatibility is maintained by reading the legacy boolean
properties first and passing them as the default_pol argument to
phy_get_rx/tx_polarity(). If the standard properties are absent the
legacy values are used transparently, so existing device trees remain
functional without modification.
Link: https://git.kernel.org/linus/66d8a334b57e64e43810623b3d88f0ce9745270b
Signed-off-by: Lucien.Jheng <lucienzx159@gmail.com>
Add a new PHY_COMMON_PROPS library that provides helper functions for
PHY drivers to read standardized polarity properties from the device
tree node:
- phy_get_rx_polarity() / phy_get_tx_polarity()
- phy_get_manual_rx_polarity() / phy_get_manual_tx_polarity()
The dt-bindings/phy/phy.h header with PHY_POL_NORMAL, PHY_POL_INVERT,
and PHY_POL_AUTO constants is provided via dts/upstream/include, which
is already in the build include path.
Ported from Merge tag 'phy-for-7.0':
git://git.kernel.org/pub/scm/linux/kernel/git/phy/linux-phy
Link: https://git.kernel.org/linus/e7556b59ba65179612bce3fa56bb53d1b4fb20db
Signed-off-by: Lucien.Jheng <lucienzx159@gmail.com>
Upstream devicetrees use a newer DT binding using cpsw-switch
compatibles. The bindings are a bit different, so two functions are
introduced to capture the differences, cpsw_eth_of_to_plat_switch() and
cpsw_eth_of_to_plat_legacy().
Signed-off-by: Markus Schneider-Pargmann (TI) <msp@baylibre.com>
Use driver data to pass the correct gmii_sel function. This way new
compatibles don't need manual compatible matching as is done in
cpsw_phy_sel().
Signed-off-by: Markus Schneider-Pargmann (TI) <msp@baylibre.com>
Since Linux commit c360eb0c3ccb ("dt-bindings: net: ethernet-controller:
Add informative text about RGMII delays"), the interpretation of RGMII
delays has changed. Prior to the commit, the RGMII Variant among "rgmii",
"rgmii-id", "rgmii-rxid" and "rgmii-txid" clearly specified whether it is
the MAC or the PHY that "should" add the delay. However, post that commit,
the RGMII Variant only specifies whether or not there is a delay on the
PCB traces between the MAC and the PHY, leaving it open as to who adds the
delay.
Hence, instead of enforcing the existence of the device-tree properties
"ti,rx-internal-delay" and "ti,tx-internal-delay", default to a delay
of 2ns, while continuing to override this delay with the aforementioned
properties, if they exist in the device-tree.
This is in line with the Linux driver implementation updated by commit
6bf78849371d ("net: phy: dp83867: use 2ns delay if not specified in DTB").
Signed-off-by: Siddharth Vadapalli <s-vadapalli@ti.com>
Reviewed-by: Anshul Dalal <anshuld@ti.com>
