Update the context state parameters when the context is already locked.
This allows changing the callback information for asynchronous
operations while the context is locked.
Signed-off-by: Jordan Yates <jordan@embeint.com>
Make clear in the include/ header that data frame size
is the same thing as word size for the context of this API.
Also, add some comments to the spi_context to make it easier
for driver writers to understand how to use the functions,
by noting the meaning of the dfs and len parameters to the update
functions. Otherwise it takes some time to understand what they mean.
Signed-off-by: Declan Snyder <declan.snyder@nxp.com>
spi_context_get_next_buf must not divide `len` by `dfs` because, in SPI,
buffer lengths are given in units of data (in this case, 16 bits), not in
bytes.
Signed-off-by: Daniel Gaston Ochoa <dgastonochoa@gmail.com>
Add spi_cs_is_gpio(_dt) helpers to check whether SPI CS is controlled by
GPIO or not. This both improves code readability and isolates SPI
internals.
Signed-off-by: Gerard Marull-Paretas <gerard.marull@nordicsemi.no>
As of today it is not possible to use SPI dt-spec macros in C++,
something known and documented. The main reason is because `cs` property
is initialized using a compound literal, something not supported in C++.
This PR takes another approach, that is to not make `cs` a pointer but a
struct member. This way, we can perform a regular initialization, at the
cost of using extra memory for unused delay/pin/flags if `cs` is not
used.
Fixes#56572
Signed-off-by: Gerard Marull-Paretas <gerard.marull@nordicsemi.no>
When estimating the time that a given SPI transfer will take, whole
buffer sets for TX and RX need to be taken into account, not only their
first parts. Correct `spi_context_wait_for_completion()` accordingly.
Signed-off-by: Andrzej Głąbek <andrzej.glabek@nordicsemi.no>
This casts the char * buffers to void * before giving them to the
LOG_DBG function to fix a warning at runtime.
Signed-off-by: Gaël PORTAY <gael.portay@gmail.com>
Adds a new spi_transcieve_cb API which enables asynchronous
SPI transactions with callback notification.
The exist spi_transcieve_async API remains and uses the new
spi_transcieve_cb API to provide a k_poll_signal notifier.
The driver API changes to provide a callback and userdata
parameter to async transcieve. All drivers in the tree
have been updated to the change.
Signed-off-by: Tom Burdick <thomas.burdick@intel.com>
In order to bring consistency in-tree, migrate all drivers to the new
prefix <zephyr/...>. Note that the conversion has been scripted, refer
to #45388 for more details.
Signed-off-by: Gerard Marull-Paretas <gerard.marull@nordicsemi.no>
commit 54907c7014
("drivers: spi: spi_context: improve support of multiple cs gpios")
added function to initialize all CS GPIOs defined in devicetree.
This function, spi_context_cs_configure_all, is intended to be
called during SPI driver initialization (POST_KERNEL init level).
It is also obvious that a SPI driver was not used at that time,
and no bus configuration (struct spi_config) is assigned to
SPI bus (spi_context.config).
The spi_context_cs_configure_all function has a homeopathic
ASSERT to validate CS levels, which causes a null pointer
dereferencing by ctx->config->operation if asserts are
enabled.
Signed-off-by: Johann Fischer <johann.fischer@nordicsemi.no>
In some cases, it is quite useful to have the possibility to also
include zero-length buffers in a buffer set used in transfers
(for example, when frames in a protocol consist of several parts,
of which some are optional). So far, the behavior of spi_context
update functions was that the transfer in a given direction was
finished when a zero-length buffer was encountered in the buffer
set. Change those functions to simply skip such buffers. Correct
in the same way also the spi_context_buffers_setup() function.
Signed-off-by: Andrzej Głąbek <andrzej.glabek@nordicsemi.no>
Do not use any timeout in the slave mode, as in this case it is not
known when the transfer will actually start and what the frequency
will be.
Signed-off-by: Andrzej Głąbek <andrzej.glabek@nordicsemi.no>
Since cs gpios are initialized during driver initialization
remove spi_context_cs_configure that is not longer need.
Signed-off-by: Bartosz Bilas <bartosz.bilas@hotmail.com>
Add extra cs_gpios and num_cs_gpios members into
spi_context structure that will be used to
initialize all defined cs gpios during the driver
initialization using SPI_CONTEXT_CS_GPIOS_INITIALIZE macro.
While at it add a new spi_context_cs_configure_all
function that allows the user to configure
all available cs gpios in inactive mode.
Signed-off-by: Bartosz Bilas <bartosz.bilas@hotmail.com>
Convert all CS control logic to be based on the `gpio_dt_spec` member
instead of the standalone `port`, `pin` and `flags` members.
Signed-off-by: Jordan Yates <jordan.yates@data61.csiro.au>
Co-authored-by: Jordan Yates <jordan.yates@data61.csiro.au>
Signed-off-by: Bartosz Bilas <bartosz.bilas@hotmail.com>
Co-authored-by: Bartosz Bilas <bartosz.bilas@hotmail.com>
Instead of waiting forever for the SPI transfer to complete, let's use
a timeout value and bail out if elapsed. The timeout value logic is,
xfer_len/frequency + tolerance
Tolerance value can be modified using a Kconfig symbol,
CONFIG_SPI_COMPLETION_TIMEOUT_TOLERANCE. It defaults to 200ms.
Fixes: #33192
Signed-off-by: Manivannan Sadhasivam <manivannan.sadhasivam@linaro.org>
Keep locking for SPI_LOCK_ON from the first call of transceive until
spi_release release the lock. Use owner parameter to in the spi_context
to store the owner of the lock.
The locking is in line with the SPI_HOLD_ON_CS
Signed-off-by: Stefan Bigler <stefan@bigler.io>
Function spi_context_longest_current_buf() has been introduced in
commit ddef35c1da for the purpose of
getting the longest possible (potentially partial) SPI transfer
for which all currently active directions have a continuous buffer.
Such transfer can be done with taking advantage of a DMA that cannot
use scattered buffers (and this is the case for nRF SPI drivers with
which this function has been introduced).
Unfortunately, because of its inadequate name, later on this function
has been incorrectly used in other SPI drivers for getting the longer
of TX/RX buffers. And commit afc480f12b
recently "fixed" the implementation of this function, assumably to
adjust it to those incorrect uses, but this way it has also broken
the nRF SPI drivers.
Instead of restoring the original implementation of the function in
question, this commit adds a new one with functionality equivalent
to that original but with a hopefully less misleading name, and this
function is used in the nRF SPI drivers.
Signed-off-by: Andrzej Głąbek <andrzej.glabek@nordicsemi.no>
- Fix GPIO CS timing when using DMA. When using GPIO CS the
CS select was enabled after the DMA started the transfer,
resulting in the first few bits being transfered while
CS was still disabled.
- Fix TX or RX only DMA transfers. When only a RX or only
a TX transfer was requested the DMA never finished.
For the RX only cause the size on the transfer was
calculated by taking the TX buffer length (0), this
caused problems.
For the TX only transfer the RX buffer was set to NULL,
this caused the DMA to acctually writing data to the
adress 0x00000000. By using the dummy destination buffer
it now only writes to valid memory.
- Add semaphore to signal that DMA is ready, instead of
just busy waiting.
Signed-off-by: Erwin Rol <erwin@erwinrol.com>
Add an additional option to the spi_cs_control struct that records how
the pin has been configured in devicetree. For drivers that are not
updated, the CS behaviour is the same as before (Push-Pull).
Use the devicetree knowledge with the GPIO subsystem so that the correct
physical pin levels for the CS pin are automatically selected.
Fixes#26267
Signed-off-by: Jordan Yates <jordan.yates@data61.csiro.au>
move spi.h to drivers/spi.h and
create a shim for backward-compatibility.
No functional changes to the headers.
A warning in the shim can be controlled with CONFIG_COMPAT_INCLUDES.
Related to #16539
Signed-off-by: Anas Nashif <anas.nashif@intel.com>
move gpio.h to drivers/gpio.h and
create a shim for backward-compatibility.
No functional changes to the headers.
A warning in the shim can be controlled with CONFIG_COMPAT_INCLUDES.
Related to #16539
Signed-off-by: Anas Nashif <anas.nashif@intel.com>
When some header are included into C++ source file, this kind of
compilations errors are generated:
error: invalid conversion from 'void*'
to 'u32_t*' {aka 'unsigned int*'} [-fpermissive]
Signed-off-by: Benoit Leforestier <benoit.leforestier@gmail.com>
Rename reserved function names in drivers/ subdirectory. Update
function macros concatenatenating function names with '##'. As
there is a conflict between the existing gpio_sch_manage_callback()
and _gpio_sch_manage_callback() names, leave the latter unmodified.
Signed-off-by: Patrik Flykt <patrik.flykt@intel.com>
In order to avoid changing the signature of spi_context_cs_control
function, which is used in every driver, let's just make it an alias to
a new version.
Fixes#10344
Signed-off-by: Tomasz Bursztyka <tomasz.bursztyka@linux.intel.com>
k_poll_signal was being used by both, struct and function. Besides
this being extremely error prone it is also a MISRA-C violation.
Changing the function to contain a verb, since it performs an action
and the struct will be a noun. This pattern must be formalized and
followed and across the project.
MISRA-C rules 5.7 and 5.9
Signed-off-by: Flavio Ceolin <flavio.ceolin@intel.com>
Any word started with underscore followed by and uppercase letter or a
second underscore is a reserved word according with C99.
Signed-off-by: Flavio Ceolin <flavio.ceolin@intel.com>
Synchronous call return them on success, asynchronous one - with proper
signal - require to get this info through the status.
Signed-off-by: Tomasz Bursztyka <tomasz.bursztyka@linux.intel.com>
- removing a left over from previous SPI slave attempt
- unlock async calls on error
Signed-off-by: Tomasz Bursztyka <tomasz.bursztyka@linux.intel.com>
Some device may need to be put up on CS high logic. The active low logic
is the default as usual, but it is now possible to request the active
high logic.
Signed-off-by: Tomasz Bursztyka <tomasz.bursztyka@linux.intel.com>
Adding Kconfig options to set supported modes by the controller
(master, slave or both)
Signed-off-by: Tomasz Bursztyka <tomasz.bursztyka@linux.intel.com>
Instead of using CONFIG_POLL, which is not directly related to SPI and
is a kernel option, let's have SPI_ASYNC instead. When enabled, it will
select POLL automatically.
Fixes#5839
Signed-off-by: Tomasz Bursztyka <tomasz.bursztyka@linux.intel.com>
tx_bufs/tx_count and rx_bufs/rx_count can be hold in another dedicated
structure, thus reducing the number of parameters to transceive. This
permits to avoid using the stack when calling transceive.
Since we saved parameters, we can expose back the struct device pointer,
to stay consistent with other device driver APIs.
Fixes#5839
Signed-off-by: Tomasz Bursztyka <tomasz.bursztyka@linux.intel.com>
As the content of this struct will not be modified by drivers, it's
better to pass it as constant. Also, if someday struct device can be
made contant too, this change will make spi ready for registering the
spi_config into ROM directly.
Fixes#5839
Signed-off-by: Tomasz Bursztyka <tomasz.bursztyka@linux.intel.com>
Current buffers might be configured to skip data, thus only len will be
set, buf will be NULL. Buffer should be used if only len is > 0 and
buffer is valid as well.
tx/rx are "on" if len is > 0
tx/rx buf should be touched if only len is > 0 _and_ buf != NULL.
Signed-off-by: Tomasz Bursztyka <tomasz.bursztyka@linux.intel.com>
Previous approach allowed only single word update for single
function call. Updating context in ISR was inefficient for
controllers supporting automatic multiple data packets transaction.
Signed-off-by: Michał Kruszewski <michal.kruszewski@nordicsemi.no>
Added function helps setting the longest possible rx and tx buffers for
single SPI transfer. Each of these buffers is a continuous memory
region. It is useful for example when peripheral supports easyDMA.
Signed-off-by: Michał Kruszewski <michal.kruszewski@nordicsemi.no>
The SPI API allows waiters to block until SPI I/O has completed. The
asynchronous subset of the API allows waiters to learn a status value
for the result of the I/O. However, the synchronous API does not allow
this.
Due to this limitation, synchronous API users cannot learn when
interrupt-driven I/O fails, which precludes proper error handling.
Resolve this limitation by adding a sync_status field to struct
spi_context, and using it to return operation results to the waiter.
Since there is only one status field, reduce the maximum number of
supported waiters from UINT_MAX to 1. This is not a problem for
current users, which all wait with the entire context locked.
Signed-off-by: Marti Bolivar <marti.bolivar@linaro.org>
Void pointer arithmetic is undefined behavior (UB).
It's OK for struct spi_buf to contain a void *, because those values
are only ever stored, read, and compared. However, pointer arithmetic
is done on the tx_buf and rx_buf fields in struct spi_context, so
those need to be u8_t * to avoid UB.
Signed-off-by: Marti Bolivar <marti.bolivar@linaro.org>
