Atmega8 initial support

[hugueslarrive]

Contribution description

The €1 board!

• boards/atmega8: copied from atmega328p
• cpu/atmega8: copied from atmega328p
• cpu/atmega_common: encountered some surprises in register organization...
• cpu/avr8_common: no system clock prescaler, no avr4.ld, no jmp instruction

Testing procedure

The big challenge of this port! With only 8 Kbyte of flash and 1 Kbyte of RAM, hardly any test fits without modification, except periph/timer.

Therefore, in a second commit, I provide the streamlined test battery that I used (not mergeable, I believe).

Surprising things can be found, such as THREAD_STACKSIZE_IDLE=64 and THREAD_STACKSIZE_MAIN=128 for the SPI!

For the I2C, it's a bit particular. It's a use case of a 4x20 LCD screen connected through an I/O port expander (the only I2C thing I had on hand). It uses a slightly hacked version of the hd44780 driver because xtimer didn't fit...

There's also the shell, reduced as much as possible, which I ended up keeping only for the gpio test.

Regarding the hardware, I used this kind of board:
https://external-content.duckduckgo.com/iu/?u=https%3A%2F%2Ftse4.mm.bing.net%2Fth%3Fid%3DOIP.CMKk4Bbtbj7ZMmNHdHx74wHaHa%26pid%3DApi&f=1&ipt=26e53f5395ddeb5f231a2f9a6cdd1de7dc97e3e4e41f1fe6ead63fe8b8b56bfd&ipo=images

a usbasp programmer, and an stlink as a serial adapter. I used an oscilloscope to visualize the state of the ports for the gpio test, the SPI output signals (MISO not tested), and the PWM. For the ADC, I simply connected each input to ground to see if the value dropped to 0.

Issues/PRs references

Split into 5:

• cpu/avr8 common: added avr4.ld script #19751
• cpu/atmega_common: checking features instead of CPU models #19752
• cpu/atmega8: new cpu #19753
• boards/atmega8: new board #19754
• tests/atmega8: streamlined test battery #19755

[nandojve]

Hi @hugueslarrive ,

I look at a bit on avr8_common and atmega_common folders. I think there are many improvements that can be made to make changes generic.

The GPIO in specific is trick to understand if address is decedent or not. In this case, the PORTB is always available and it is a 100% match.

I would like review after you apply the idea behind my comments on the parts that you need apply patches.

I would like to see commits for main folders avr8_common and atmega_common if possible.

[hugueslarrive]

Hi @nandojve, I've started applying the idea behind your comments but I'm not finished yet. I'm a bit concerned that it might break something for other CPUs, so I'll test it on the ones I have (328p and 1284).

I took the opportunity to create a specific cpu_conf.h file to fine-tune other values. The only thing I'm not sure how to handle is adjusting THREAD_STACKSIZE_IDLE when using xtimer or ztimer because I can't make them fit without using the stdio_null module, which means I can no longer retrieve the output of ps()...

The test for PORTB is causing a compilation issue. I tried adding #include <avr/io.h>, but it doesn't resolve it.

I really like the idea of testing functionalities rather than CPU models; that's what I usually do. But in this case, it wasn't done that way, and I wanted to make sure it works first.

[nandojve]

@maribu Do you have any recommendation in the case of THREAD_STACKSIZE_IDLE ?

[maribu]

THREAD_STACKSIZE_IDLE is mostly architecture specific. Still, when e.g. the return address is larger (e.g. on more than 128 KiB addressable program memory for AVR) or additional registers need to be safed on function calls (e.g. when external memory is supported on AVR there typically is an additional byte to address data RAM beyond 64 KiB), or e.g. FPUs (not relevant for AVR) require space to save registers in case of context switches.

All of this doesn't really apply on ATmega8. So the default value should really be large enough. It might be possible to reduce it by a handful of bytes. But effort would be better spent on supporting threadless idle, so that the idle stack could be dropped completely instead.