sched/semaphore: change semcount type to int

[zyfeier]

Summary

Fix the issue introduced by bug #14465.
Some memory on the ESP32 requires aligned access,
so the semcount type has been changed to int.

Impact

semaphore

Testing

bes board test pass

[nuttxpr]

[Experimental Bot, please feedback here]

No, this PR does not fully meet the NuttX requirements. While it provides some information, it lacks crucial details. Here's a breakdown:

• Insufficient Summary: While it mentions the bug fix and the affected code, it's too brief. It needs to explain how changing semcount to int resolves the alignment issue. What was the original type? Why did that type cause a problem on the ESP32? What specific memory areas require aligned access? A link to bug sem: change sem wait to atomic operation #14465 is needed.

• Incomplete Impact Assessment: Simply stating "semaphore" is not enough. Think about the broader implications:

  • Is new feature added? Is existing feature changed?: Existing feature changed (semaphore implementation).
  • Impact on user: Potentially YES if user code relies on the previous size of semcount. This needs investigation and explanation.
  • Impact on build: Potentially YES if the change affects structure packing or memory layout. This needs to be addressed.
  • Impact on hardware: YES, specifically ESP32.
  • Impact on documentation: Potentially YES if the API changes or behavior is modified.
  • Impact on security: Potentially YES if changing the size of semcount leads to overflows or other vulnerabilities. This needs careful consideration.
  • Impact on compatibility: Potentially YES, backward compatibility might be affected. Needs investigation and explicit mention.

• Inadequate Testing Information: "bes board test pass" is too vague.

  • Build Host(s): Missing. Provide details about the development environment.
  • Target(s): While "bes board" is mentioned, the specific configuration is missing. Provide the full board:config string.
  • Testing logs before change: Missing entirely. Include logs demonstrating the issue caused by bug sem: change sem wait to atomic operation #14465.
  • Testing logs after change: Missing any meaningful detail. Show logs demonstrating that the problem is resolved and that the semaphore functionality still works correctly.

In short, the PR needs significantly more detail and evidence to meet the NuttX requirements. It needs to thoroughly explain the problem, the solution, and the impact of the change, supported by clear testing evidence.

[tmedicci]

Testing...

[tmedicci]

Testing...

Unfortunately, the tests' results are the same... still failing to boot...

[zyfeier]

Testing...

Unfortunately, the tests' results are the same... still failing to boot...

Hi, @tmedicci, we only have the ESP32-S3 boards. Can we reproduce this issue using the ESP32S3? Just to confirm, this issue only occurs when using the IRAM heap, correct?

[tmedicci]

Hi @zyfeier ,

Just to confirm, this issue only occurs when using the IRAM heap, correct?
Yes, exactly.

Hi, @tmedicci, we only have the ESP32-S3 boards. Can we reproduce this issue using the ESP32S3?

Unfortunately not. ESP32-S3 uses SRAM1 (which is accessible using either the instruction and data bus): there isn't a separate heap for loading apps (we just need to access the heap using its data address when copying data to avoid non-aligned access).

For ESP32, the text heap is used from SRAM0 (here). This memory bank is only accessible using the instruction bus and any non-aligned word would cause an exception. In fact, we capture this exception and repeat the operation using a aligned access. Check esp32_user.c: something in this PR broke this logic...