Add a new status word bit USER_REGULATOR_NOT_SUPPORTED_STSBIT and set it
if the user regulator support is not compiled in.

Signed-off-by: Marek Behún <marek.behun@nic.cz>

It can be useful for the userspace to be able to determine whether the
MCU is STM32 or GD32, without depending on determining this from
firmware version (git commit ID).

Add new status word bit GD32_MCU_STSBIT and set it to 1 in GD32
firmware.

Signed-off-by: Marek Behún <marek.behun@nic.cz>

The reset button does not currenlty work when in bootloader.

Also after flashing MCU firmware with mcutool and rebooting, the reboot
does not work.

This is because in bootloader's RESET_MANAGER code, we need to read
SYSRES_OUT_PIN's input, not output.

In STM32's version we are reading input.

I guess this was a mistake when converting from STM32 to GD32.

Signed-off-by: Marek Behún <marek.behun@nic.cz>

This new tool compiles bootloader and application, and creates one
output image containing both, padding them with 0xFF bytes.

Signed-off-by: Marek Behún <marek.behun@nic.cz>

GD32 offers HW CRC32 circuit. When compiling application binary, store
its size and crc32 checksum at offsets 0x110 and 0x114 of the binary
(immediately after ISR vectors).

Then in bootloader if the crc32 checksum does not check out, do not boot
the application and stay in bootloader, but power the device on.

Signed-off-by: Marek Behún <marek.behun@nic.cz>

- generate dissassembled output (useful for debugging)
- some cleanups

Signed-off-by: Marek Behún <marek.behun@nic.cz>

Similarly to how we have rewritten application's slave_i2c_device code,
rewrite also boot_i2c.

With correct handling of interrupts we can get rid of the weird "missing
129th byte" issue.

Also by disabling ACK when erasing/writing, we achieve that mcutool
retries sending packets.

Signed-off-by: Marek Behún <marek.behun@nic.cz>

For all interrupts we are using pre-emption priority 0 and
sub-priorities 1-5, but GD32's implementation of nvic_enable_irq()
allows for only 4 bits of priority, and the default setting is 2 bits
for pre-emption priority and 2 bits for subpriority.

This means that upper bits of the subpriorities we are using are throws
away, and so our prioritization

  handler     prio                       handler     prio
  =======     ====                       =======     ====
  i2c         1                          leds        0
  delay       2       ==============>    i2c         1
  debounce    3       gets changed to    led effect  1
  leds        4                          power       1
  led effect  5                          delay       2
  power       5                          debounce    3

Fix this by allowing one more bit for subpriority.

Signed-off-by: Marek Behún <marek.behun@nic.cz>

Signed-off-by: Marek Behún <marek.behun@nic.cz>

Now that we can detect whether a transaction was successful, move
deleting button status and counter bit to AERR handler, so that it is
deleted only after the information is successfuly transmitted to master.

Signed-off-by: Marek Behún <marek.behun@nic.cz>

In case we somehow miss the end of tx transaction, we need to reset tx
buffer when new rx transaction is being started.

Signed-off-by: Marek Behún <marek.behun@nic.cz>

Signed-off-by: Marek Behún <marek.behun@nic.cz>

Improvements:
- enable TBE/RBNE interrupt only after address is received
- enable ERR interrupt after when slave is trasmitting, so that we can
  detect STOP condition via AERR interrupt
- handle BERR and OUERR interrupts to avoid interrupt storms in case
  something bad occurs when ERR interrupt is enabled
- send 0xFF byte if master wants data and we do not have any (this is
  the only sensible way to do this on GD32)

Signed-off-by: Marek Behún <marek.behun@nic.cz>

User configurable regulator was never used and is not put on boards.
The corresponding code takes a lot of place. Build it only if
USER_REGULATOR_ENABLED is set to 1 during build:
  make USER_REGULATOR_ENABLED=1

Signed-off-by: Marek Behún <marek.behun@nic.cz>

Allow choosing baudrate for serial debug output via DBG_BAUDRATE make
option, for example:
  make DBG_ENABLE=1 DBG_BAUDRATE=921600

If not used, 115200 Bd is used as default.

Signed-off-by: Marek Behún <marek.behun@nic.cz>

Instead of redefining the DBG_ENABLE macro in debug_serial.h header,
make it so that we can enable serial debug output by passing
DBG_ENABLE=1 to make:
  make app DBG_ENABLE=1

Signed-off-by: Marek Behún <marek.behun@nic.cz>

Signed-off-by: Marek Behún <marek.behun@nic.cz>

When debugging is enabled, the CARD_DET pin is used as UART-TX and
PCI_PLED1 pin as UART-RX. Don't configure them as GPIOs when debugging
is enabled.

Signed-off-by: Marek Behún <marek.behun@nic.cz>

It can also happen that a new ADDSEND interrupt is received while we
still haven't handled receiving/transmitting data from previous
transaction. In such a case we receive both ADDSEND and RBNE/TBE
interrupt. We need to prefer RBNE/TBE before ADDSEND. (It always means
that the data not yet received/transmitted belong to previous
transaction, because I2C only starts receiving/transmitting after
ADDSEND is cleared.)

Signed-off-by: Marek Behún <marek.behun@nic.cz>

This way when receiving both STPDET and ADDSEND interrupt, we first stop
the state machine and only then start receive new transaction.

Signed-off-by: Marek Behún <marek.behun@nic.cz>

On GD32 sometimes when the I2C interrupt handler is invoked, multiple
interrupts can be set.

If these happen to be STPDET and ADDSEND because a transaction was
completed (STPDET) and new trasaction immediately started (ADDSEND), we
need to somehow enforce handling of STPDET first.

Prepare a handler state machine for this.

Signed-off-by: Marek Behún <marek.behun@nic.cz>

Signed-off-by: Marek Behún <marek.behun@nic.cz>

Signed-off-by: Marek Behún <marek.behun@nic.cz>

Signed-off-by: Marek Behún <marek.behun@nic.cz>

Reset rx counters on ADDR event and TX counters on STPDET event.

Signed-off-by: Marek Behún <marek.behun@nic.cz>

Rename the variable to tx_data_len, similar to tx_data_ctr, and move it
into the i2c_status structure.

Signed-off-by: Marek Behún <marek.behun@nic.cz>

Resetting rx buffer size in this case makes it so that the following
bytes are interpreted as next command, although they should be ignored
because this I2C transaction is already bad.

Signed-off-by: Marek Behún <marek.behun@nic.cz>

Signed-off-by: Marek Behún <marek.behun@nic.cz>

There is currently a bug wherein the access to the rx_buf[] array
overflows when too much bytes are received. This was not a problem
because in the i2c_state structure there is tx_buf[] array after
rx_buf[], but fix this anyway.

Signed-off-by: Marek Behún <marek.behun@nic.cz>

There is a bug wherein on too much received bytes we return from the I2C
interrupt handler without enabling interrupts.

Signed-off-by: Marek Behún <marek.behun@nic.cz>

Error interrupts are disabled and this check is commented, remove it.

Signed-off-by: Marek Behún <marek.behun@nic.cz>

This includes STM32 specific code and also ACK bit experiments.

Signed-off-by: Marek Behún <marek.behun@nic.cz>

Writing 1 into the ACKEN bit in I2C control 0 register does not trigger
sending ACK.

Instead it makes is so that when next byte is received, the controller
sends ACK if ACKEN=1 or NACK if ACKEN=0.

Thus calling
  i2c_ack_config(I2C_PERIPH_NAME, I2C_ACK_ENABLE);
after every received byte is unnecessary.

In current code we never disable ACKEN, thus remove the redundant
enables.

NOTE: THIS ALSO RESULTS IN STPDET INTERRUPT NOT BEING THROWN AWAY!
Previously Jirka moved code from STPDET handling into
CMD_GET_STATUS_WORD because he didn't get STPDET interrupt. That was
because this rewriting of ACKEN bit clears STPDET bit!

Signed-off-by: Marek Behún <marek.behun@nic.cz>

The i2c_interrupt_flag_get() has a rather complex code for what it does,
and the reason is simply bad design of macro constants in the GD32
library.

Read the I2C status 0 register directly in the interrupt handler, which
makes the handler faster and less complicated (even for reading the
code!).

Simplify also clearing of the ADDSEND bit.

Signed-off-by: Marek Behún <marek.behun@nic.cz>

Fix LED support

See merge request !1