if (idx < 10)
		__uart_tx_byte('0' + (char)idx);
	else if (idx < 16)
		__uart_tx_byte('A' - 10 + (char)idx);
	__uart_tx_byte('!');
	UART_DIRECT_PRINT(raw_read_esr(), 32)
	__uart_tx_byte('!');

Sounds good. But I will have to write it like this (even if it is taking a bit more space):
```
__uart_tx_byte('\r');
__uart_tx_byte('\n');
__uart_tx_byte('!');
__uart_tx_byte('E');
__uart_tx_byte('X');
__uart_tx_byte('C');
__uart_tx_byte('E');
__uart_tx_byte('P');
__uart_tx_byte('T');
__uart_tx_byte('!');
```
The reason is the following:
The issue that I hit was linking the bootblock at the wrong memory address (in memlayout.ld). During that time quite some stuff worked fine. But printk didn't work and my only clue was the `__uart_tx_byte` output. The reason why printk didn't work and that is also why this code didn't work:
```
const char *msg = "\r\n!EXCPT!";
while (*msg)
  __uart_tx_byte(*msg++);
```
is that the compiler will reference `msg` variable using an address relative to where I linked the bootblock. If I on the other hand write all the `__uart_tx_byte` seperate than the compiler doesn't need to access something using a memory address, but only needs to write a single byte character in a register as argument to `__uart_tx_byte`. I hope my explanation was good enough. I added an explanation into the commit-msg in case someone tries to update this code in the future and wonders why it was written this way.