17810: drivers/slipdev: implement sleep states r=benpicco a=benpicco
18348: sys/net/gnrc/pktbuf_static: make use of alignas() r=maribu a=maribu
### Contribution description
Since we are now using C11, we can make use of `alignas()` provided by `<stdalign.h>` to make the alignment code easier to read.
### Testing procedure
I didn't expect this to change binaries, but is safes 4 bytes. `elf_diff` shows that the compiler (at least GCC 11.3.0) was not able to detect that `gnrc_pktbuf_static_buf` was just an alias for `_pktbuf_buf`. That makes sense since it would be hard without LTO to rule out external writes to `gnrc_pktbuf_static_buf`, unless one would have added a `const` (to the pointer, not to the data the pointer points to).
The [output of `elf_diff`](https://mari-bu.de/pr_18348_gnrc_pktbuf_static_elf_diff.html) looks otherwise quite unscary.
Also:
```
$ make BOARD=nucleo-f767zi -C tests/unittests/ tests-pktbuf flash test
make: Entering directory '/home/maribu/Repos/software/RIOT/tests/unittests'
Building application "tests_unittests" for "nucleo-f767zi" with MCU "stm32".
[...]
Welcome to pyterm!
Type '/exit' to exit.
READY
s
START
.............................................
OK (45 tests)
make: Leaving directory '/home/maribu/Repos/software/RIOT/tests/unittests'
```
### Issues/PRs references
None
19120: CI: seperate check-labels and check-commits workflows r=maribu a=kaspar030
Co-authored-by: Benjamin Valentin <benjamin.valentin@ml-pa.com>
Co-authored-by: Marian Buschsieweke <marian.buschsieweke@ovgu.de>
Co-authored-by: Kaspar Schleiser <kaspar@schleiser.de>
19048: drivers/sht2x: some small improvements r=benpicco a=gschorcht
### Contribution description
This PR provides the following improvements for the SHT2x driver:
- migration to `ztimer` 8a605517f5 and 367549940de3bd8910052334c34af028d4992741
- floating point arithmetics replaced by integer arithmetics f424caebbfec9f2be56aa2337c6cc09dba5b97d6
- fix of sleep times (typical times replaced by maximum times as recommended by the datasheet) 13615c72094b8541ee62c3e8ed5a36dc4d725fd0
- release of the I2C bus during sleep 9415c216cfab734520ef98dd00b350c22f342c60
- Kconfig configuration of sensor parameters added dadbbcb4c328350893db53ba6743d03cb34ecc1c
- no-hold mode is now the default mode instead of the hold mode
Regarding the sleep times: The typical measurement times were used as sleep times. According to the datasheet, typical measurements are only recommended for calculating energy consumption, while maximum values should be used for calculating waiting times in communication. Therefore, the typical time values were replaced by maximum time values for the sleep in no-hold mode.
Regarding the hold mode: In hold mode, the sensor uses clock stretching until the measurement results can be read by the MCU. This blocks both the I2C bus and the MCU during the entire measurement, which can take up to 85 ms if I2C is not interrupt-driven. Therefore, the no-hold mode is now used by default, where the calling thread sleeps during the measurement, but the MCU is not completely blocked. Furthermore, the hold mode requires that the MCU supports clock stretching. Even if the MCU supports clock stretching, the hold mode with clock stretching doesn't seem to work with different MCUs. I couldn't get it working for STM32 and ESP32.
Regarding the I2C bus during sleep: If the no-hold mode is used and the calling thread sleeps up to 85 ms, it makes sense to release the I2C bus until the measurement results are available.
### Testing procedure
`tests/driver_sht2x` should still work.
### Issues/PRs references
Co-authored-by: Gunar Schorcht <gunar@schorcht.net>
According to the data sheet, typical times are recommended for calculating energy consumption, while maximum values should be used for calculating waiting times in communication. Therefore, the typical time values are replaced by maximum time values for sleep.
There are many modem commands for which you get a line of response followed
by an OK. Take for example the AT+CGSN command to get the IMEI of a Ublox
G350.
>> AT+CGSN
<< 004999010640000
<< OK
- most were trivial
- missing group close or open
- extra space
- no doxygen comment
- name commad might open an implicit group
this hould also be implicit cosed but does not happen somtimes
- crazy: internal declared groups have to be closed internal
- replace all `int`s and `unsigned`s with integers with fixed width
- replaced all signed integers of sizes with unsigned ones (sizes
cannot be negative)
- made bitshifts 8-bit safe (e.g. `1 << 24` is valid on 32-bit, but
undefined behavior on 8-bit, as a 16 bit wide `int` would be shifted
by more than the type width)
- use `void *` / `const void *` for data buffers to ease use
This makes life easier when calling e.g. `saul_reg_write()` with data
stored in flash.
As now the signatures for reading and writing differ (in that `const`
qualifier only), `saul_notsup()` is split into `saul_write_notsup()`
and `saul_read_notsup()`. However, one is implemented as a symbol alias
of the other, so that ROM consumption remains unchanged.
To benefit from the chunked ringbuffer if large frames are being sent,
we need to allocate more than one ethernet frame length to it.
Rename the define and make it overwriteable by the user.
`DOSE_TIMER_DEV` is defined in `board.h`, so we have to include it.
It's not included by the header (and should not), so move the check
to the .c file.
The RSSI values reported by LoRa transceiver can be less than -127.
Therefore, `int8_t` is not enough. This commit defines the RSSI of
`netdev_lora_rx_info` as `int16_t` and adapt the drivers accordingly
(sx126x, sx127x).
This moves the following parts of ethos' state machine out of ISR
context:
- Sending and replying to HELLO messages
- Byte-unstuffing
Some escape handling is still needed in the ISR handler, due to ethos'
protocol design, to determine if a received byte must go into the
netdev queue (tsrb) or the STDIO queue (isrpipe), but the actual
unstuffing is now done in the STDIO and netdev handler threads,
respectively.
When the measurement results are read from the `ALG_RESULT_DATA` register set including the STATUS register, the `DATA_RDY` flag in the STATUS register is already cleared during reading. Therefore it is not possible to check this flag after the `ALG_RESULT_DATA` has been read. Therefore the function `ccs811_read_iaq` always returned `CCS811_ERROR_NO_NEW_DATA` although the data were valid either after checking for new data with the function `ccs811_data_ready` or after triggering the Data Ready interrupt.
This is a temporary fix for Issue #17060. It allows to disable
auto inclusion of `ztimer_periph_rtt` in cases where another
module or application requires direct access.
Limitations:
- as ifeq are involved order of inclusion matters, therefore
these modules should be included early in the build at application
level and not in modules `Makefile.dep`
- this does not disallow direct inclusions of `ztimer_periph_rtt`,
since this only disables auto inclusion of these modules
This is a temporary solution since this is already possible with
Kconfig, but not in make.
Since all implementations simply return 0 and most drivers do not check the return value, it is better to return void and use an assert to ensure that the given device identifier and given device parameters are correct.
The default driver type is just an index into a device array defined
by the board.
If a platform wants to encode additional information in the device type,
it can define a custom type.
This means we can just set the default type to whatever fits the target
CPU best.
On ARM this will still be a 32 bit word, but on AVR it will by a 8 bit byte.
This API change refactors the usbdev API to supply buffers via the
usbdev_ep_xmit function. This changes from the usbdev_ep_ready call to allow
separate buffers per call. An usbdev_ep_buf_t pseudotype is available and must
be used when defining buffers used for endpoints to adhere to the DMA alignment
restrictions often required with usb peripherals.
Main advantage is that the usbdev peripherals no longer have to allocate
oversized buffers for the endpoint data, potentially saving multiple KiB
of unused buffer space. These allocations are now the responsibility of
the individual USB interfaces in the firmware
This API change refactors the usbdev API to supply buffers via the
usbdev_ep_xmit function. This changes from the usbdev_ep_ready call to allow
separate buffers per call. An usbdev_ep_buf_t pseudotype is available and must
be used when defining buffers used for endpoints to adhere to the DMA alignment
restrictions often required with usb peripherals.
Main advantage is that the usbdev peripherals no longer have to allocate
oversized buffers for the endpoint data, potentially saving multiple KiB
of unused buffer space. These allocations are now the responsibility of
the individual USB interfaces in the firmware
By moving all the single byte struct elements to the end, we can reduce
padding inside `dose_t` and ensure that `recv_buf` is always aligned.
This saves some RAM:
master
------
text data bss dec hex filename
36384 136 12944 49464 c138 tests/driver_dose/bin/samr21-xpro/tests_driver_dose.e
this patch
----------
text data bss dec hex filename
36484 136 12936 49556 c194 tests/driver_dose/bin/samr21-xpro/tests_driver_dose.elf
Some CAN transceivers have a standby pin that has to be pulled low
in order to use it.
If the interface is disabled we can set it to high again to save some
power.
BOARDs with RF switch might only support one of the TX modes, and
on init the BOARD needs to be configured accordingly and the correct
mode selected on TX.
The bogus packed attribute added to lis3dh_data_t resulted in the
structure being aligned two 1 byte. It was later casted to an
uint16_t pointer (which is aligned two 2 bytes). By dropping the
packed attribute the alignment mismatch is fixed.
A fixed timeout is either too long for high symbol rates or too short
for low symbol rates.
To fix this, calculate the timeout based on the symbol rate.
For this, the old 5ms timeout is equivalent to 58 bytes being transmitted
at 115200 baud (8 data bit + start & stop bit).
I rounded this to 50 bytes which should yield 4340 µs.
There is no way to properly handle incorrect SPI parameters at run time, so
just having an assertion blow up is the better choice here.
As most instances of `spi_acquire()` don't check the return value anyway, this
will improve debugging experience quite a bit. Some implementation of
spi_acquire() don't even check parameters anyway.
