re-enable JIT for ZTS builds on Apple Silicon #22712
There are no files selected for viewing
| Original file line number | Diff line number | Diff line change |
|---|---|---|
|
|
@@ -17,6 +17,13 @@ | |
| #ifndef HAVE_JIT_H | ||
| #define HAVE_JIT_H | ||
|
|
||
| #if defined(__APPLE__) && defined(__aarch64__) && defined(ZTS) | ||
| # ifndef HAVE_PTHREAD_JIT_WRITE_PROTECT_NP | ||
| # error "Apple Silicon ZTS JIT requires pthread_jit_write_protect_np()" | ||
| # endif | ||
| # define ZEND_JIT_USE_APPLE_MAP_JIT 1 | ||
| #endif | ||
|
Member
There was a problem hiding this comment. Would it be feasible to unify all
Contributor
Author
There was a problem hiding this comment. Technically the split could be done for all macOS builds, but the protection path cannot be unified.
Member
There was a problem hiding this comment. This makes sense. Can we merge the
Contributor
Author
There was a problem hiding this comment. Nice finding, I cleaned it up in 8f51da2 . I also change the
Member
There was a problem hiding this comment. Thank you for the changes. There are also cases where we call php-src/ext/opcache/jit/zend_jit.c Lines 3528 to 3536 in 8f51da2
Contributor
Author
There was a problem hiding this comment. I did a bit of research and found that those where added in #8382 and adjusted in #9371. They are effectively obsolete: Apple Silicon ZTS now takes the dedicated |
||
|
|
||
| #if defined(__x86_64__) || defined(i386) || defined(ZEND_WIN32) | ||
| # define ZEND_JIT_TARGET_X86 1 | ||
| # define ZEND_JIT_TARGET_ARM64 0 | ||
|
|
||
| Original file line number | Diff line number | Diff line change |
|---|---|---|
| @@ -0,0 +1,55 @@ | ||
| --TEST-- | ||
| GH-13400: JIT code generated after fork is visible to the parent | ||
| --DESCRIPTION-- | ||
| OPcache metadata and generated JIT code are shared by forked workers, so the | ||
| JIT mapping and its allocation pointer must remain shared after fork. This test | ||
| records the available JIT space before forking and makes only the child call a | ||
| function often enough to generate code. The parent then verifies that it sees | ||
| the reduced free space and can execute the generated function. | ||
|
|
||
| Without shared inheritance, a private JIT mapping becomes copy-on-write in the | ||
| child. The parent's free-space check would print bool(false), while shared | ||
| OPcache metadata could point the parent at code bytes that exist only in the | ||
| child and cause a crash instead of printing int(42). A startup-only test would | ||
| not detect this failure in fork-based SAPIs such as FPM or Apache. | ||
| --EXTENSIONS-- | ||
| opcache | ||
| pcntl | ||
| --INI-- | ||
| opcache.enable=1 | ||
| opcache.enable_cli=1 | ||
| opcache.file_update_protection=0 | ||
| opcache.jit=tracing | ||
| opcache.jit_buffer_size=16M | ||
| opcache.jit_hot_func=1 | ||
| --SKIPIF-- | ||
| <?php | ||
| if (!function_exists('pcntl_fork')) die('skip pcntl_fork() not available'); | ||
| if (!(opcache_get_status()['jit']['on'] ?? false)) die('skip JIT is not available'); | ||
| ?> | ||
| --FILE-- | ||
| <?php | ||
| function value(): int { | ||
| return 42; | ||
| } | ||
|
|
||
| $bufferFree = opcache_get_status()['jit']['buffer_free']; | ||
| $pid = pcntl_fork(); | ||
| if ($pid === 0) { | ||
| for ($i = 0; $i < 100; $i++) { | ||
| value(); | ||
| } | ||
| exit(0); | ||
| } | ||
| if ($pid === -1) { | ||
| echo "pcntl_fork() failed\n"; | ||
| exit(1); | ||
| } | ||
|
|
||
| pcntl_waitpid($pid, $status); | ||
| var_dump(opcache_get_status()['jit']['buffer_free'] < $bufferFree); | ||
| var_dump(value()); | ||
| ?> | ||
| --EXPECT-- | ||
| bool(true) | ||
| int(42) |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I don't understand this change, could you comment on why this is necessary?