Disable JIT when pthread protection is unavailable

[realFlowControl]

Follow up to #22712 (comment)

This moves the runtime capability check into zend_jit_check_support(). When pthread_jit_write_protect_supported_np() reports that per-thread JIT write protection is unavailable, OPcache now logs an ACCEL_LOG_WARNING, disables JIT, and continues using the interpreter. This matches the existing unsupported-JIT behavior.

JIT tests that modify opcache.jit at runtime now do so only when OPcache is inactive or its runtime status reports JIT as enabled. This preserves their original coverage on supported platforms without attempting to re-enable a JIT that the platform capability check disabled.

Testing

• Clean Apple Silicon ZTS build.
• Supported Apple Silicon JIT fork test passes.
• Forced-unsupported OPcache suites matching the CI Test and Test OpCache configurations: 928 tests each, 0 failures.

[iluuu1994]

Thank you Florian! Looks correct to me. The PR is still marked as draft, let me know when you think it's ok to merge.

[iluuu1994]

I don't quite understand why this is a new problem? 🤔

Edit: Oh, I guess the JIT was just fully disabled at compile time and the INI setting didn't do anything.

[realFlowControl]

Exactly. After #22712 this behavior changed, but CI did not expose this particular follow-up problem because the initial test step printed a fatal error and failed. Consequently, the job never reached the later tracing/function JIT stages. See https://github.com/php/php-src/actions/runs/29793770728/job/88520997977

So the problem was already there with #22712, we just did not see it because CI failed before those stages were executed. 😉

[realFlowControl]

I am also unsure whether this warning represents a practical user-facing problem. It can occur when the configure-time check confirms that the pthread_jit_write_protect_np() symbol is available, but the runtime capability probe pthread_jit_write_protect_supported_np() returns false.

So far, I have only observed this on GitHub’s virtualized macOS runners. On a physical M4 Mac, the runtime capability is available and JIT works as expected.

[iluuu1994]

Okay, given pthread_jit_write_protect_supported_np() returns false in CI, I don't think actually running it makes sense. I think fa47ae1 should just be reverted. @arnaud-lb@realFlowControl Are you ok with that?

[arnaud-lb]

Ok for me

[realFlowControl]

Commited the revert in d84ae5b2375a52aa125059b7f97eae22a56aca3d

[iluuu1994]

Thanks, but I need to commit this to 8.2+ separately anyway.

[iluuu1994]

I reverted the change directly. I believe, once you rebase, the changes of the test files directly in ext/opcache/tests/ are no longer necessary.

[iluuu1994]

Thank you Florian! Will merge once CI has finished.

[iluuu1994]

Shouldn't this be:

if (($status = opcache_get_status()) !== false && $status['jit']['enabled']) {

? I guess it's fine to try to set it even with opcache disabled.

[iluuu1994]

Thanks again!