Swap the write and read data paths within the GSPI controller to ensure
correct data is seen on the SPI lines during 16-bit transfers.
Signed-off-by: Sai Santhosh Malae <Santhosh.Malae@silabs.com>
Updated mosi_overrun alignment from 4 bytes to 32 bytes to
ensure compatibility with the GPDMA driver. This prevents
alignment-related errors during DMA transfers.
Signed-off-by: Sai Santhosh Malae <Santhosh.Malae@silabs.com>
Enable higher SPI burst sizes when using GPDMA to improve throughput.
Bursts are only enabled when both TX and RX channels use GPDMA and
the SPI clock frequency is at least 10 MHz, as DMA flow control is
unreliable at lower SPI rates.
Signed-off-by: Sai Santhosh Malae <Santhosh.Malae@silabs.com>
Add gspi_siwx91x_pick_lower_freq() to select the nearest lower
supported GSPI clock frequency based on the requested value.
The GSPI clock can operate only at discrete frequencies such as
80 MHz, 40 MHz, 26.6 MHz, 20 MHz, and so on (80 MHz divided by
integer factors). If the requested frequency does not match one
of these valid steps, it is rounded down to the nearest lower
supported frequency.
The driver now logs both the requested and the actual programmed
frequency along with the divider value, helping users verify the
effective SPI clock configuration.
Signed-off-by: Sai Santhosh Malae <Santhosh.Malae@silabs.com>
Some STM32 SPI have the ability to bypass the baudrate prescaler, which is
equivalent to having a prescaler of 1.
Adds this new value in the prescaler calculation if available.
Signed-off-by: Guillaume Gautier <guillaume.gautier-ext@st.com>
The `ll` part of driver's file name indicates that the implementation uses
the STM32Cube Low Level ("LL") library, which is not interesting because
there is only *one* STM32 driver implementation anyways. Addtionnally, the
presence of `ll` among STM32 drivers is not consistent.
Get rid of it for this class's driver to match most other STM32 drivers.
Signed-off-by: Mathieu Choplain <mathieu.choplain-ext@st.com>
The burst length unit is bytes, not number of transfers. This had not been
an issue since the DMA driver historically ignored the values, but has now
become one since they are used and (most importantly for us) validated.
Fixes broken 16-bit on STM32N6 series.
Signed-off-by: Mathieu Choplain <mathieu.choplain-ext@st.com>
Enable clock control by default for siwx91x SoCs. Moreover, most
drivers for siwx91x soc depend on clock control, but didn't declare
it.
Signed-off-by: Martin Hoff <martin.hoff@silabs.com>
Added calls to pm_device_runtime_put() to ensure proper device runtime
management when configuration or DMA transceive operations fail.
Signed-off-by: Martin Hoff <martin.hoff@silabs.com>
Most drivers for Series 2 depend on clock control, but didn't
declare it. Enable clock control by default for Series 2 SoCs.
Signed-off-by: Aksel Skauge Mellbye <aksel.mellbye@silabs.com>
New nrfx release contains major rework of nrfx drivers
instantiation making it easier to integrate with dts nodes.
Now, nrfx driver instances can no longer be `const`
because they contain driver runtime state.
Additionally, all nrfx drivers return `errno` error codes
instead of deprecated `nrfx_err_t`.
Signed-off-by: Nikodem Kastelik <nikodem.kastelik@nordicsemi.no>
Removing references to infineon,cat1-spi-pdl from the device tree files.
Updates the driver bindings and driver file to look at the
infineon,cat1-spi binding and uses a Kconfig option to select Legacy HAL
implementation instead.
See https://github.com/zephyrproject-rtos/zephyr/pull/98035 for context.
Signed-off-by: John Batch <john.batch@infineon.com>
This driver had a lot of regressions introduced by commits that were
clearly mixing up what values had units of bytes and what values had
units of words. Fix the issues in the DMA path of the driver.
Signed-off-by: Declan Snyder <declan.snyder@nxp.com>
The asserts are hard to debug and unrecoverable, they just say kernel
panicked but difficult to realize where and why, it's better to return
an error and log message clearly from spi driver.
Signed-off-by: Declan Snyder <declan.snyder@nxp.com>
It is not illegal to provide an RX len with no buffer.
It just means the controller should clock out that many
words but not store them in memory. So remove the assert
and just avoid null dereference.
Signed-off-by: Declan Snyder <declan.snyder@nxp.com>
Refactor this function to return an error code, instead of assert.
Put the data into a pointed argument instead.
Refactor other functions to make sure to
propogate this error all the way up the call stack.
Signed-off-by: Declan Snyder <declan.snyder@nxp.com>
Fixes an issue with an unused function being defined if
CONFIG_DEVICE_DEINIT_SUPPORT was disabled
Signed-off-by: Jamie McCrae <jamie.mccrae@nordicsemi.no>
Release the PM constraint at the end of the context in which it was
requested (`transceive`), instead of in a semi-related function
(`finalize_spi_transaction`) which is also called by other API functions
(`spi_release`). Asynchronous usage is released in the transaction
complete callback.
Signed-off-by: Jordan Yates <jordan@embeint.com>
When SPI spec is reconfigured, avoid reallocating another sm and properly
free the loaded sm programs before attempting to load new programs.
Fix#94416.
Signed-off-by: Terry Geng <terry@terriex.com>
Enabled intel LPSS DMA interface using dw common to support
usage of internal DMA in LPSS SPI to transfer and
receive data.
Signed-off-by: Bindu S <bindu.s@intel.com>
This patch adds a check for the EOT (End Of Transfer) flag
inside the polling loop in spi_stm32_half_duplex_switch_to_receive().
If EOT is set, the transfer is already complete and the loop
is exited, preventing the lockup.
Signed-off-by: Khaoula Bidani <khaoula.bidani-ext@st.com>
Signed-off-by: Julien Racki <julien.racki-ext@st.com>
The current implementation checking for the validity of the spi slave
number is wrong. First of all, there is an off by one error where it
allows SPI_CHIP_SELECT_COUNT value as a valid slave, when really it
should be that value minus one. Secondly, it doesn't take into account
the fact that having software controlled GPIO CS can technically have
any number of slaves on the bus. So fix by finding the max of these two
values and fixing the off by one mistake. Also, for RW612, only one HW
native CS is available due to SOC pinmux limitations.
Signed-off-by: Declan Snyder <declan.snyder@nxp.com>
The version register is not accessible through aux-reg interface,
So skip it entirely on platforms with aux-reg device tree property.
Signed-off-by: Mohamed Moawad <moawad@synopsys.com>
Because of limitation of gpdma, DMA Rx won't run if user doesn't specify
destination buffer.
Thus, the DMA Rx may stop before the end of the full transaction. So, wait
on DMA Tx instead.
Then, the SPI data won't be consumed by the DMA. We need to properly reset
the fifo before to start a new transaction (it is better to ensure we start
with a clean state before every transaction).
Signed-off-by: Jérôme Pouiller <jerome.pouiller@silabs.com>
For all STM32 drivers and SoC, replace the READ_REG macro and the
LL_xxx_ReadReg functions (defined in the STM32 HAL) by
stm32_reg_read defined in Zephyr.
Signed-off-by: Guillaume Gautier <guillaume.gautier-ext@st.com>
This commit enhances the SPI shell in a number of different ways:
- Previously only spi buses could be specified in the shell.
Now both spi buses and spi devices (physical device on the spi bus,
not the device struct) can be specified.
- Both spi buses and spi devices can be specified by their node name or
by any of their node labels.
- Instead of having a single instance of a spi config struct, each spi
bus/device has it's own dedicated config. This means that
users only have to configure spi config once for each device and use
them freely, without needing to reconfigure them every time after
switching between different spi buses/devices.
- Spi devices get their spi configs automatically from the devicetree at
compile time. This means that once a good working configuration is
found for a spi device, that can be specified as a "default" in the
devicetree.
- Spi devices don't need to have their status set to "okay" or be
initialized by their driver to be accessible from the spi shell. As
long as the spi bus that they are on has status="okay", they can
be accessed.
- When changing a cs pin with spi cs command, the spi device/bus
needs to first be specified before specifying the gpio device.
The nested dynamic subcommands enable users to autocomplete both
of the arguments.
Implementation details
- It was possible to gather all the information about the spi devices
from the devicetree at the compile time, but not for the spi buses.
For the spi devices we have a way to get spi device nodes that are
on an okay spi bus with a devicetree macros and extract data from
those nodes. For the spi buses that is not possible,
since there is no macro to get all spi bus nodes at once.
- So, for the spi buses this discovery step has to be done at the
runtime, with the help of SYS_INIT macro and shell_device_filter
function.
- Since we now allow users to either specify a spi bus or a spi device
by their node label or their node name, we needed to have a way to map
those strings to the correct spi config struct, so we can later pass
it to the spi API functions or do some configuration on it. That is
made possibly by the find_spec_by_label and the two arrays,
spi_things and maps.
Signed-off-by: Marko Sagadin <marko.sagadin42@gmail.com>
Remove the handling of cross domain pins from nrf drivers. To use
cross domain in tests, force on constlat and disable power domains
for the test.
Signed-off-by: Bjarki Arge Andreasen <bjarki.andreasen@nordicsemi.no>
This patch fixes an issue in the SPI driver that caused the chip select
line to remain low after a transaction completed (when using a GPIO and
operating in non-interrupt mode).
Signed-off-by: Ian Morris <ian.d.morris@outlook.com>
Silabs siwx91x offer two DMA hardware block: GPDMA and UDMA.
While UDMA has some benefit when running in low power modes, GPDMA offer
better performances. So GDMA is probably better suited for SPI device.
Ideally, we would like to leave the ability to switch back to UDMA.
Unfortunately, UDMA and GPDMA are not configured in the same way:
- the maximum length of the block are different (1024 or 2048 for UDMA
and 4096 for GPDMA)
- the burst length is different
So, we only support GPDMA.
Signed-off-by: Jérôme Pouiller <jerome.pouiller@silabs.com>
Since various macros don't end with a colon, the code formatters give bad
results. Let's fix that.
Signed-off-by: Jérôme Pouiller <jerome.pouiller@silabs.com>
The siwx91x platform provides two DMA block: UDMA and GPDMA. GPDMA require
the dma_slot attribute.
Signed-off-by: Jérôme Pouiller <jerome.pouiller@silabs.com>
Replace a #ifndef directive with a if(!IS_ENABLED()) instrcution
in transceive() function. This change makes later integration of RTIO
support in this function smoother, polluting a bit less this function
with #if based directives.
Signed-off-by: Etienne Carriere <etienne.carriere@st.com>
Fix a missing braces pair around a conditioned instruction in
STM32 SPI driver. Fix that by aggregating the 2 if() instructions
into a single ANDed one.
Signed-off-by: Etienne Carriere <etienne.carriere@st.com>
Correct some indentation issues, a few useless line escapes,
double space characters or parentheses pair in STM32 SPI driver.
Signed-off-by: Etienne Carriere <etienne.carriere@st.com>
Remove bits2byte() helper function than was not always used. Replace it
with a division by BITS_PER_BYTE that is explicit enough.
Signed-off-by: Etienne Carriere <etienne.carriere@st.com>
Reorder inclusions of header files to clarify and simplify later changes.
By the way, remove #ifdef CONFIG_SPI_STM32_DMA condition to include
DMA header files as its not required.
Keep zephyr/log.h with use of LOG_MODULE_REGISTER() first since included
local spi_context.h depends on log resources to be defined.
Signed-off-by: Etienne Carriere <etienne.carriere@st.com>
