drivers/sx127x: optimize power consumption on nss pin

[aabadie]

Contribution description

With this PR the power consumption drops from 6mA to 40uA when a stm32l0 CPU is sleeping.
I'm not sure it's the best fix and perhaps there's still for improving power consumption.

cc @fjmolinas who might be interested.

Testing procedure

• LoRaWAN application should still work
• Measure power consumption when running the examples/lorawan application (you might need to test this with examples/lorawan: add the possibility to use ABP activation procedure #11237). I used a Nucleo-l073rz + an mbed lora shield + an x-nucleo-lpm01a shield.

Issues/PRs references

Required by #11237

[fjmolinas]

@aabadie I think handling the NSS pin should be done but this doesn't seem to me like the right way to do so.

The SPI interface is active LOW. With this PR the device would always think it is active since after read or write actions to a register your are setting the pin as input with a pull down. This could interfere with other SPI devices on the same line.

I think the correct way would be using spi_init_cs() and passing the cs_pin when calling spi_acquire(). The NSS mode should be OUT_PP with a pull-up resistor. I thinks that should work for minimizing current consumption for the NSS pin.

[fjmolinas]

@aabadie I looked into this again... I came to the conclusion the reason we were seeing the excessive power consumption was because of the X-NUCLEO-LPM01A. I connected a multi-meter directly between the NSS_PIN and the board and I get 0uA of current draw (same for all I/O pins going to sx1276).

Doing some further measurements I can Identify sources of excessive current draw threw MISO, MOSI and SCK. NSS isn't a problem since it's inactive state is by default HIGH.

To minimize the current draw on MOSI and SCK they must be pulled low before going to sleep. From measurements after releasing the SPI device there state isn't 0 but close to 0. Two options for handling this:

• initiate them with pull-downs

    gpio_init(spi_config[bus].mosi_pin, GPIO_OUT_PD);
    gpio_init(spi_config[bus].sclk_pin, GPIO_OUT_PD);

• clear the gpio's before entering sleep

    gpio_init(spi_config[bus].mosi_pin, GPIO_OUT); 
    gpio_clear(spi_config[bus].mosi_pin);
    gpio_init(spi_config[bus].sclk_pin, GPIO_OUT);
    gpio_clear(spi_config[bus].sclk_pin,);

I think option 2 is better since it can be handled by the driver. This has to be done since according to reference manual (at least for stm32) "If software configures a GPIO pin as Alternate Function Output, but peripheral is not activated, its output is not specified".

In the case of MISO the SPI device should be initialized with the corresponding pull-up (in this case since it would be the inactive state).

gpio_init(spi_config[bus].miso_pin, GPIO_IN_PU);

I tested this using a nucleo-l1073rz and an mbed sx1272 sheild runing examples/lorawan. When in sleep the shield consumes 8uA and the node around 2uA.

DIOs are tied to VDD upon initialization and RESET is already in proper inactive state by default.

[aabadie]

Closing in favor of #11542