refactor(profiling): use module globals for ZMM state

[morrisonlevi]

Description

• Introduced ProfilerGlobals struct which for now just contains zend_mm_state: Cell<ZendMMState>.
• Created module_globals.rs to centralize PHP globals management.
  • Moved GLOBALS (NTS) into module_globals.rs.
  • Moved GLOBALS_ID_PTR (ZTS) into module_globals.rs and renamed it GLOBALS_ID (because it's not a pointer, it's the thing the pointer points at).
  • Moved ginit() and gshutdown() lifecycle functions into module_globals.rs, and gave them export names of ddog_php_prof_{ginit,gshutdown).
• Updated zend::ModuleEntry to use ProfilerGlobals and pointers to module globals. Previously, this was faked just so we could observe the ginit/gshutdown phases but now we have real module globals.
• Moved some tls macros in the allocation profiler into mod.rs because they are the same for ge84 and le83.

Motivation

In #3542 we worked around a compatibility issue with ext/grpc by mimicking the PHP module globals pattern. Why mimic it when we can use it?

For accessing module globals, we do have to translate the C macro ZEND_MODULE_GLOBALS_BULK to Rust, though. Here are its expansions:

(ZTS) ZEND_MODULE_GLOBALS_BULK(module_name)
→ TSRMG_BULK(module_name##_globals_id, zend_##module_name##_globals *)
→ ((zend_##module_name##_globals *) (*((void ***) tsrm_get_ls_cache()))[TSRM_UNSHUFFLE_RSRC_ID(module_name##_globals_id)])
→ ((zend_##module_name##_globals *) (*((void ***) tsrm_get_ls_cache()))[module_name##_globals_id - 1])

(NTS) ZEND_MODULE_GLOBALS_BULK(module_name)
→ (&module_name##_globals)

So for us, module_name##_globals_id is just GLOBALS_ID and module_name##_globals is just GLOBALS. The rest matches C directly.

Reviewer checklist

• Test coverage seems ok.
• Appropriate labels assigned.

[datadog-datadog-prod-us1]

⚠️ Tests

✨ Fix all issues with Cursor

⚠️ Warnings

🧪 1026 Tests failed

    testSearchPhpBinaries from integration.DDTrace\Tests\Integration\PHPInstallerTest (Fix with Cursor)

testSimplePushAndProcess from laravel-58-test.DDTrace\Tests\Integrations\Laravel\V5_8\QueueTest (Datadog) (Fix with Cursor)

DDTrace\Tests\Integrations\Laravel\V5_8\QueueTest::testSimplePushAndProcess
Test code or tested code printed unexpected output: spanLinksTraceId: 697cde8f000000005596de911bb29288
tid: 697cde8f00000000
hexProcessTraceId: 5596de911bb29288
hexProcessSpanId: c975a2947e84e912
processTraceId: 6167361454546784904
processSpanId: 14516687732560161042

phpvfscomposer://tests/vendor/phpunit/phpunit/phpunit:106

testSimplePushAndProcess from laravel-8x-test.DDTrace\Tests\Integrations\Laravel\V8_x\QueueTest (Datadog) (Fix with Cursor)

DDTrace\Tests\Integrations\Laravel\V8_x\QueueTest::testSimplePushAndProcess
Test code or tested code printed unexpected output: spanLinksTraceId: 697cde8c00000000d066843fc28ee4d6
tid: 697cde8c00000000
hexProcessTraceId: d066843fc28ee4d6
hexProcessSpanId: 737d0ef296ebbefb
processTraceId: 15016835416895448278
processSpanId: 8321824121527451387

View all

ℹ️ Info

❄️ No new flaky tests detected

_{This comment will be updated automatically if new data arrives.
🔗 Commit SHA: 4c14aa4 | Docs | Datadog PR Page | Was this helpful? Give us feedback!}

[codecov-commenter]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 62.14%. Comparing base (e380043) to head (4c14aa4).

Additional details and impacted files

@@            Coverage Diff             @@
##           master    #3608      +/-   ##
==========================================
+ Coverage   62.12%   62.14%   +0.02%     
==========================================
  Files         141      141              
  Lines       13387    13387              
  Branches     1753     1753              
==========================================
+ Hits         8317     8320       +3     
  Misses       4270     4270              
+ Partials      800      797       -3     

see 3 files with indirect coverage changes

---

Continue to review full report in Codecov by Sentry.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update e380043...4c14aa4. Read the comment docs.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.

[pr-commenter]

Benchmarks [ profiler ]

Benchmark execution time: 2026-01-30 16:42:46

Comparing candidate commit 4c14aa4 in PR branch levi/module-globals with baseline commit e380043 in branch master.

Found 0 performance improvements and 0 performance regressions! Performance is the same for 28 metrics, 8 unstable metrics.

[morrisonlevi]

I've had AI help me set up a container that uses PHP 8.4 ZTS with FrankenPHP as well as NTS with php-fpm. I used PHP 8.4 because that's what the current release of FrankenPHP embeds. Things look good so far.

[realFlowControl]

We should (in a follow up PR) move all the other "global state" we have in other modules to PHP globals, but so far this is looking good.
Can you do a short run with DoE? I'd not expect this to show any performance problems, but just to be sure and I've added two comments.

[realFlowControl]

Suggested change

	// TODO: Florian, do we need this?
	// let globals = globals_ptr.cast::<ProfilerGlobals>();
	// (*globals).zend_mm_state = ZendMMState::new();

I think we do not, we did not do so earlier also. Also we unregister our hooks in RSHUTDOWN, nothing in allocation profiling touches the global state after RSHUTDOWN and PHP itself takes care of freeing the memory, I guess it is fine the way it is.

[morrisonlevi]

One of our internal performance tests is showing extreme improvements, which seems suspect, like maybe it's failing to collect the amount of data it should. I will investigate this more before committing.

[morrisonlevi]

Subsequent tests have been "normal." Going to proceed.