cpu/cc2538: fix RSSI computation

[maribu]

Contribution description

• use int instead of int8_t to avoid loosing precision
• the offset is already negative, so add it instead of subtracting it.

Testing procedure

One could send to an CC2538 board with another board next to it using different TX power settings and check if the RSSI is reasonable.

Issues/PRs references

Split out of #22463

Declaration of AI-Tools / LLMs usage:

AI-Tools / LLMs that were used are:

• none

[riot-ci]

Murdock results

✔️ PASSED

390554d Update cpu/cc2538/radio/cc2538_rf_radio_ops.c

Success	Failures	Total	Runtime
11128	0	11128	11m:31s

Artifacts

• Documentation preview

[AnnsAnns]

One could send to an CC2538 board with another board next to it using different TX power settings and check if the RSSI is reasonable.

This sounds like a "I will simply trust maribu on this one" situation 😛

[maribu]

This sounds like a "I will simply trust maribu on this one" situation 😛

Hmm, we might want to actually test this. I just guessed the change from the code context. (So basically did Vibe-code this change, just with the guessing done by me rather than a GPU.)

The code in master does however indeed not only look fishy to the LLVM's integer overflow checker, but also to me.

[AnnsAnns]

This sounds like a "I will simply trust maribu on this one" situation 😛

Hmm, we might want to actually test this. I just guessed the change from the code context. (So basically did Vibe-code this change, just with the guessing done by me rather than a GPU.)

The code in master does however indeed not only look fishy to the LLVM's integer overflow checker, but also to me.

Agentic vibe based merging, Microsoft better pay me one trillion for that

[kfessel]

unobfuscate please

[kfessel]

int8_t rssi_val is read from the device
‎IEEE802154_RADIO_RSSI_OFFSET is (-174)
CC2538_RSSI_OFFSET is (-73)
->
const int hw_rssi_min = −247

so if the signed 8 bit value ( rage [-128 127] is bigger than −247)
its a rssi is a valid value
else
rssi is 0

is that correct?

might that register be unsigned?

this is addressed

[kfessel]

unobfuscated code made flawes visible

see the suggestion in other comment - thread

addressed

[kfessel]

squash is missing