libc/atomic: decoupling atomic and spinlock to avoid recursion

[anchao]

Summary

libc/atomic: decoupling atomic and spinlock to avoid recursion

1. use irq save in AMP mode
2. use critical section in SMP mode

Impact

N/A

Testing

gcc14 with sim/smp

[nuttxpr]

[Experimental Bot, please feedback here]

NuttX PR Requirements Review

Meets Requirements? Partially

Breakdown:

• Summary:
  • Adequate: Provides a brief overview of the change (decoupling atomic and spinlock) and the methods used.
  • Missing: Lacks context on why this change is necessary. What problems were caused by the previous coupling? This is crucial for reviewers to understand the PR's purpose.
• Impact:
  • Insufficient: "N/A" is rarely accurate for code changes. This section needs careful consideration:
    • Feature Changes: Even if no new features are directly exposed, internal changes can impact existing functionality.
    • Build/Hardware/Documentation: While the impact might be minimal, it's unlikely to be truly "N/A". Be specific. Does this affect certain architectures more than others?
    • Security/Compatibility: These are critical and cannot be dismissed outright. Atomic operations and spinlocks are fundamental to concurrency; any change has potential implications that MUST be addressed, even if the answer is "no impact, and here's why...".
• Testing:
  • Insufficient:
    • Build Host: Provide OS and compiler versions for reproducibility.
    • Targets: "sim/smp" is vague. Specify the exact simulator/emulator and configurations used.
    • Logs: "your testing logs here" is unhelpful. Include ACTUAL logs demonstrating the issue BEFORE the change, and the improved behavior AFTER.

Recommendation:

Do not submit this PR in its current state. Address the missing information:

1. Expand the Summary: Explain the problem the change solves.
2. Thoroughly Analyze Impact: For each point, either state "No impact" with a brief justification OR describe the specific impact.
3. Provide Detailed Testing Information: Include full build host details, specific target configurations, and relevant log snippets that demonstrate the change's effect.

[xiaoxiang781216]

hardware must support atomic operation in SMP, so we.don't.need this.codd.block

[anchao]

but some architectures may not fully support all atomic operations, eg: -march=armv7e-m does not support atomic8(64 bit) OpenAMP/libmetal#301

[xiaoxiang781216]

OK