cpu/atmega_common: Fix atmega_port_addr()

[maribu]

Contribution description

In 04ab5a7 a bug was introduced in
the calculation of the GPIO port address by refactoring code. This
fixes the issue by extracting the GPIO port first from the pin.

Testing procedure

See #18245 (comment)

Issues/PRs references

Reported in #18245 (comment)

[maribu]

@valentinpi: Care to test?

[krzysztof-cabaj]

Hi! I confirm that now everything works well!

[maribu]

It seems that the compiler's optimizer trips over the volatile qualifier in the GPIO struct, despite the struct only being used to calculate address offsets. The new version has the same .text size as the old gpio_init() (the only difference to the stable branch left is from a change in assert()).

For me, the code still works. Care to test it again?

[maribu]

For reference: This models the memory layout of the three registers per GPIO port:

RIOT/cpu/atmega_common/include/periph_cpu_common.h

Lines 103 to 130 in d9fc082

	/**
	* @brief Structure describing the memory layout of the registers of a
	* GPIO port on ATmega MCUs.
	*/
	typedef struct {
	/**
	* @brief Toggle bits in the port register
	* @note The bits in the port register will be also toggled for inputs.
	* This can be both a footgun as well as an efficient way to toggle
	* the pull up resistor on inputs
	*
	* Referred to as "Input Pins Address" in the datasheet.
	*/
	volatile uint8_t pin;
	/**
	* @brief Configure pins as output (1) or input (0) using the Data
	* Direction Register
	*/
	volatile uint8_t ddr;
	/**
	* @brief Read/write the state of GPIO pins using the Port Data Register
	*
	* @note When in input mode (see @ref atmega_gpio_port_t::ddr) writing a
	* 1 will enable the pull up resistor, writing a 0 will put the
	* pin in floating mode.
	*/
	volatile uint8_t port;
	} atmega_gpio_port_t;

That is why I reordered the functions now to match the order they come in memory. Interestingly, the datasheet lists the registers in reverse order (in regard to their address) as well:

[valentinpi]

The current version is working for me.

[krzysztof-cabaj]

The current version is working for me.

For me too. This probably resolve these BUGs.

[benpicco]

So then squash to get this through CI

[maribu]

@krzysztof-cabaj and @valentinpi: Thanks for reporting the issue and testing!

@benpicco: Thx for the quick review :)