Summarized this squashed PR replaces the hal/ folder in the port. This has been replaced the official
HAL layer from Nordic Semiconductor; https://github.com/NordicSemiconductor/nrfx.

A Git submodule has been added under lib/nrfx, for the nrfx dependency.

The drivers / modules has been updated to use this new HAL layer; nrfx at v1.0.0.

Also, header files and system files for nrf51/nrf52x chip variants has been deleted from the device/ folder, only keeping back the startup files written in C. All other files are now fetched from nrfx.

3 new header files in the ports/nrf/ folder has been added to configure nrfx (nrfx_config.h), logging (nrfx_log.h) and glue nrfx together with the drivers and modules from micropython (nrfx_glue.h).

The PR has been a joint effort from @aykevl (Ayke van Laethem) and @glennrub.

For reference, the commit log will be kept to get an overview of the changes done:

* ports/nrf: Initial commit for moving hal to Nordic Semiconductor BSD-3 licensed nrfx-hal.

* ports/nrf: Adding nrfx, Nordic Semiconductor BSD-3 hal layer, as git submodule checked out at lib/nrfx.

* ports/nrf/modules/machine/uart: Fixing bug which set hwfc to parity excluded, always resulting in no flow control, hence corrupted output. Also adding an extra loop on uart_tx_char to prevent any tx when any ongoing tx is in progress.

* ports/nrf/i2c: Moving I2C over to nrfx driver.

* ports/nrf/modules/machine/i2c: Alignment. Renaming print function param 'o' to 'self_in'

* ports/nrf/spi: Updating SPI machine module to use nrfx drivers.

* ports/nrf: Renaming modules/machine/rtc.c/.h to rtcounter.c/.h to not confuse the peripheral with Real-Time Clock:

* ports/nrf: Updating various files after renaming machine module RTC to RTCounter.

* ports/nrf: Renaming RTC to RTCounter in modmachine globals dict table. Also updating object type name to reflect new module name.

* ports/nrf: Fixing leftovers after renaming rtc to rtcounter.

* ports/nrf: Early untested adoption of nrfx_rtc in RTCounter. Untested.

* nrf/modules/machine/i2c: Improve keyword argument handling

* ports/nrf/modules/temp: Updating Temp machine module to use nrfx defined hal nrf_temp.h. Moving logic of BLE stack awareness to machine module.

* ports/nrf/boards/pca10040: Enable machine Temp module.

* nrf/modules/machine/rtcounter: Remove magic constants.

* ports/nrf: Adding base support for nrfx module logging. Adding option to disable logging of UART as it might log its own setup over UART while the peripheral is not yet set up. Logging of UART could make sense if other transport of log is used.

* ports/nrf: updating nrfx_log.h with more correct parenthisis on macro grouping.

* ports/nrf: Updating nrfx logging with configuration to disable logging of UART module. The pattern can be used to turn off other modules as well. However, for now UART is the only module locking itself by logging before the peripheral is configured. Logging is turned off by default, can be enabled in nrfx_config.h by setting NRFX_LOG_ENABLED=1.

* ports/nrf/modules/random: Updating modrandom to use nrfx hal for rng. Not using nrfx-driver for this peripheral as its blocking mode would do the trick on RNG. Moving softdevice aware code from legacy hal to modrandom.c.

* nrf: Enable Peripheral Resource Sharing.

This enables TWI and SPI to be enabled at the same time.

* nrf/Makefile: Define MCU sub variant (e.g. NRF51822/NRF51422)

* nrf: Port TIMER peripheral to nrfx HAL.

* nrf/modules/machine/uart: Optimize UART module

For a nRF51, this results in a size reduction of:
.text: -68 bytes
.data: -56 bytes

* nrf/modules/machine/uart: Don't use magic index numbers.

* nrf/modules/machine/uart: Fix off-by-one error.

For nrf51:
.text: -40 bytes

* nrf/modules/machine/rtcounter: Update for nrfx HAL.

* nrf/modules/machine/i2c: Reduce RAM consumption.

Reductions for the nrf51:
flash: -108 bytes
RAM:   -72 bytes

* nrf/mpconfigport: Avoid unnecessary root pointers.

This saves 92 bytes of RAM.

* nrf: Support SoftDevice with nrfx HAL.

* nrf: Add NVMC peripheral (microbitfs) support.

There is no support yet for a SoftDevice.

It also fixes a potentially serious bug in start_index generation.

* nrf/modules/machine/spi: Optimize SPI peripheral.

nrf51:
text: -340 bytes
data: -72  bytes

nrf52:
text: -352 bytes
data: -108 bytes

* nrf/modules/random: Forgot to commit header file.

* nrf: Make nrfx_config.h universal for all boards.

* nrf: Use SoftDevice API for flash access when built for SD

* nrf/drivers/bluetooth: Remove legacy HAL driver includes.

These were not used anymore so can be removed.

* ports/nrf/microbit: Port microbit targets to nrfx HAL

Initial port of microbit modules to use nrfx HAL layer.
Tested display/image and modmusic on micro:bit to verify that
softpwm and ticker for nrf51 is working as expected.

Changing IRQ priority on timer to priority 2, as 1 might collide if
used side by side of SD110 BLE stack.

The patch reserves Timer1 peripheral compile time. This is not ideal
and should be resolved in seperate task.

* nrf/boards/microbit: Remove custom nrfx_config.h from microbit target, adding disablement of timer1 if softpwm is enabled.

* nrf/adc: Update ADC module to use nrfx

* nrf/modules/machine/pwm: Updating machine PWM module to use nrfx HAL driver.

examples/nrf52_pwm.py and examples/nrf52_servo.py tested on pca10040.

* nrf: Removing hal folder and boards nrf5x_hal_conf.h headers.

* nrf/nrfx_glue: Adding direct NVIC access for S110 BLE stack

If SoftDevice s110 has not yet been initialized, the IRQ will not be forwarded to
the application using the sd_nvic* function calls. Hence, direct access to cmsi
nvic functions are used instead if SoftDevice is not enabled.

* nrf/drivers/ticker: Setting IRQ priority 3 on Timer1

SoftDevice fails to initilize if Timer1 has been configured to priority
level 2 before enabling the SD. The timer is set to priority 1, higher than BLE
stack in order to provide better quality of music rendering when used with the
music module. This might be too high, time will show.

* nrf/examples: Updating ubluepy_temp after moving RTCounter to nrfx.

* nrf: delete duplicate files from device folder which can be located in nrfx/mdk.

* nrf/Makefile: Fetch system files from nrfx.

Testing on each device sub-variant to figure out which system file to
use. Reason for this is that nrf52.c is actually defining nrf52832.

Removing NRF_DEFINES parameter setting the device in use into the
same sub-variant test, as NRF52 is unique to nrf52832 when using nrfx.
Without this exclusion of -DNRF52 in compilation for nrf52840, the
device will be interpreted as a nrf52, hence nrf52832.

Also, changing name on variable SRC_NRF_HAL to SRC_NRFX_HAL to
explicitly tell the origin of the file.

* nrf: Updating device #ifdefs to be more open to non-nrf51 targets.

* nrf/modules/machine/uart: Removing second instance of UART for nrf52840 as it only has one non-DMA variant.

* nrf/device: Removing system files as these are now used from nrfx/mdk

* nrf: Moving startup files in device one level up as there is no need for deep hierarchy.

* nrf: Use NRF52_SERIES defined in nrfx/mdk/nrf.h as define value when testing for both nrf52(832) and nrf52840 variants.

* nrf/modules/machine/uart: Enable UART RX by default

Enable rx by default after intiialization of the peripheral.
Else, the nrfx driver will re-enable rx for each byte read
on uart REPL, clearing the EVENT_RXDRDY before second byte,
which again will make second byte get lost and read will get stuck.

This happens if the bytes are transmitted nrf(51) while still
processing the previous byte. Not seen on nrf52, but should
also become an issue at higher speeds.

This patch sets rx to always be enabled. Hence, not clearing the event
between read bytes, and it will be able to detect next byte recieved
upon finishing the first.

* nrf/modules/machine/timer: Fixing defines excluding Timer1 if ticker/softpwm is used.

* nrf: Switching import form mpconfigboard.h to mpconfigport.h in nrfx_config.h as mpconfigboard.h might define default values for defines not set by board specific header.

* nrf/modules/machine/i2c: nrfx integration fixes

Increasing speed to 400K.

Returning Address NACK's as MP error code; MP_ENODEV.

Returning MP_ETIMEOUT on all other error codes from TWI nrfx driver
except the ANACK.

Enabling and disabling the TWI peripheral before and after each transaction.

* nrf/examples: Updating ssd1306_mod.py to split framebuffer transfer into multiple chunks

* nrf/modules/machine/i2c: Return MP_EIO error if Data NACK occurs.

* nrf: Addressing review comments.

* nrf: Updating git submodule and users to nrfx v1.0.0.

* nrf/modules/machine/adc: Update adc module to follow v1.0.0 nrfx API.

* nrf/modules/machine/spi: Implement init and deinit functions

Extending SPI objects with a config member such that
configuration can be kept between new() and init().

Moving initialization done in new() to common init
function shared between the module functions.

If SPI is already configured, the SPI peripheral will
be uninitialized before initalized again.

Adding logic to handle initialization of polarity and
phase. As well, updating default speed to 1M from 500K.

* nrf/modules/machine: Removing unused nrfx includes in machine module header files

When compiling for microbit with LTO=0, a compiler error occurs due to
'ms' variable in the microbit_sleep function has not been initialized.

This patch initialize the variable to 0.

LTO messes up debuggability and may cause some other issues.
Additionally, it does not always result in reduced code size.

Removing shell commands for checking if boardmodules.mk exists
under boards/<BOARD>/modules folder before including it.

This patch does the equivalent to previous test without using
shell commands. Hence, including the .mk if it exists.

Reference:
https://stackoverflow.com/questions/8346118/check-if-a-makefile-exists-before-including-it

The nrf51x22_256k_16k_s110_8.0.0.ld had a stack size of only 1kB, which
is way too low. Additionally, the indicated _minimum_stack_size (set at
2kB for that chip) isn't respected.

This commit sets the heap end based on the stack size (heap end = RAM
end - stack size) making it much easier to configure.

Additionally, the stack/heap size of nrf52 chips has been set to a more
sane value of 8kB.

Leave it enabled by default on all targets.

This is only possible when using UART-over-BLE (NUS) instead of the
default hardware peripheral. The flash area saved is quite substantial
(about 2.2KB) so this is useful for custom builds that do not need UART.

Feather52 target which is using SD s132 v.2.0.1 cannot compile
due to variable containing RAM start address is not used.

This patch enables the correct sd_ble_enable variant for this SD.

Similar commit to this one:
https://github.com/tralamazza/micropython/commit/6e56e6269f467e59316b5e4cb04ea37ab6a0dfe3

When .text + .data oveflow available flash, the linker may not show an
error. This change makes sure .data is included in the size calculation.

Saves for the nrf51:
flash: 336 bytes
RAM:     4 bytes

This frees 128 bytes of .bss RAM on the nRF51, at the cost of possibly
more expensive GC cycles. Leave it as-is on the nRF52 as that chip has a
lot more RAM.

This is also done in the micro:bit:
https://github.com/bbcmicrobit/micropython/blob/a7544718a7138a04168e8e6b283e14e500ffbe8b/inc/microbit/mpconfigport.h#L6

* ports/nrf/boards: Adding linker script for nrf52832 using BLE stack s132 v.5.0.0.

* ports/nrf/drivers/bluetooth: Updating makefile to add BLE_API_VERSION=4 if s132 v5.0.0 is used.

* ports/nrf/drivers/bluetooth: Updating BLE stack download script to also download S132 v5.0.0.

* ports/nrf/drivers/bluetooth: Updating ble_drv.c to handle BLE_API_VERSION=4 (s132 v5.0.0).

* ports/nrf/boards: Updating linker script for nrf52832 with s132 v.5.0.0 bluetooth stack.

* ports/nrf/drivers/bluetooth: Removing commented out code in ble_drv.c

* ports/nrf/drivers/bluetooth: Updating define of GATT_MTU_SIZE_DEFAULT for SD132v5 to be defined using the new name defined in the SD headers in a more generic way.

* ports/nrf/drivers/bluetooth: Cleaning up use of BLE_API_VERSION in the ble_drv.c. Also considering s140v6 API, so not all has been changed to >= if API version 3 and 4 in combo is used. New s140v6 will differ on these, and add a new API not compatible with the API for 3 and 4.

For each iteration of polling BLE events from the Bluetooth LE stack.

Enabled by default on microbit targets, with or without BLE stack.  Also
enable softpwm to make display and music module compile.

Not all targets are using bluetooth le.

Applicable for targets with board specific modules.

Increases size by 68 bytes.  Should be considered to be removed as temp
module is already providing this functionality.

To use if BLE stack is enabled.  The custom linker script also set off
space enough to compile in microbitfs+hal_nvmc.

If the board has these drivers then they will be initialized:
- softpwm (implicit ticker)
- music module
- board specific module

Also update microbit_music_init0 to register low priority ticker callback
for the music module.

- Rename microbit_module_init to board_module_init0 which is the generic
  board module init function.
- Add low priority callback registration of display tick handler in the
  module init function.

- Rename init function to ticker_init0.
- Implement ticker_register_low_pri_callback (recycle of unused
  set_low_priority_callback function which was unimplemented).
- Add support for registering 2 low pri callbacks.  For now, one intended
  for microbit display, and one for modmusic.

So that users of the board module can find the init function of the module
implicitly.

And update the API to align with new unary/binary function callback
structures.

Makefile and mpconfigport.h update is generic, and could be used by other
boards to give extra modules which are only for a selected board.

This aligns implementation with new style structures.

Also rename modmicrobit.h to microbitconstimage.h.

Source: https://github.com/lancaster-university/microbit-dal.git

From micro:bit port repository, https://github.com/bbcmicrobit/micropython

Costs 136 bytes on a nRF51822.

Saves 448 bytes of flash. Can still be enabled using:

    make FROZEN_MPY_DIR=freeze BOARD=foo

Disconnecting after a connect would not restart advertising, so
reconnecting may get harder.

The ampy tool expects a "soft reboot" line when it does a soft reset.

Using libm from micropython free up about 5.5kb flash on nrf52
targets which have floating point enabled.