TSRM: make CG, EG, SCNG and AG compile-time offsets

TSRM/TSRM.c

@@ -36,12 +36,11 @@ struct _tsrm_tls_entry {
 	tsrm_tls_entry *next;
 };
 
-
 typedef struct {
 	size_t size;
 	ts_allocate_ctor ctor;
 	ts_allocate_dtor dtor;
-	size_t fast_offset;
+	ptrdiff_t fast_offset;
 	int done;
 } tsrm_resource_type;
 
@@ -58,6 +57,7 @@ static int					resource_types_table_size;
 /* Reserved space for fast globals access */
 static size_t tsrm_reserved_pos  = 0;
 static size_t tsrm_reserved_size = 0;
+static size_t tsrm_reserved_front = 0;
 
 static MUTEX_T tsmm_mutex;	  /* thread-safe memory manager mutex */
 static MUTEX_T tsrm_env_mutex; /* tsrm environ mutex */
@@ -155,6 +155,7 @@ TSRM_API bool tsrm_startup(int expected_threads, int expected_resources, int deb
 
 	tsrm_reserved_pos  = 0;
 	tsrm_reserved_size = 0;
+	tsrm_reserved_front = 0;
 
 	tsrm_env_mutex = tsrm_mutex_alloc();
 
@@ -205,7 +206,7 @@ TSRM_API void tsrm_shutdown(void)
 			} else {
 				free(p->storage);
 			}
-			free(p);
+			free((char *) p - tsrm_reserved_front);
 			p = next_p;
 		}
 	}
@@ -232,6 +233,7 @@ TSRM_API void tsrm_shutdown(void)
 
 	tsrm_reserved_pos  = 0;
 	tsrm_reserved_size = 0;
+	tsrm_reserved_front = 0;
 }/*}}}*/
 
 /* {{{ */
@@ -319,17 +321,20 @@ TSRM_API void tsrm_reserve(size_t size)
 }/*}}}*/
 
 
+/* Carve a fixed-offset front region out of the reserved space. It is placed
+ * before the TLS entry, so the hot globals get compile-time-constant negative
+ * offsets from the cache pointer. */
+TSRM_API void tsrm_reserve_fast_front(size_t size)
+{
+	tsrm_reserved_front = TSRM_ALIGNED_SIZE(size);
+	tsrm_reserved_size -= tsrm_reserved_front;
+}
+
+
 /* allocates a new fast thread-safe-resource id */
 TSRM_API ts_rsrc_id ts_allocate_fast_id(ts_rsrc_id *rsrc_id, size_t *offset, size_t size, ts_allocate_ctor ctor, ts_allocate_dtor dtor)
 {/*{{{*/
-	TSRM_ERROR((TSRM_ERROR_LEVEL_CORE, "Obtaining a new fast resource id, %d bytes", size));
-
 	tsrm_mutex_lock(tsmm_mutex);
-
-	/* obtain a resource id */
-	*rsrc_id = TSRM_SHUFFLE_RSRC_ID(id_count++);
-	TSRM_ERROR((TSRM_ERROR_LEVEL_CORE, "Obtained resource id %d", *rsrc_id));
-
 	size = TSRM_ALIGNED_SIZE(size);
 	if (tsrm_reserved_size - tsrm_reserved_pos < size) {
 		TSRM_ERROR((TSRM_ERROR_LEVEL_ERROR, "Unable to allocate space for fast resource"));
@@ -338,9 +343,26 @@ TSRM_API ts_rsrc_id ts_allocate_fast_id(ts_rsrc_id *rsrc_id, size_t *offset, siz
 		tsrm_mutex_unlock(tsmm_mutex);
 		return 0;
 	}
-
-	*offset = TSRM_ALIGNED_SIZE(sizeof(tsrm_tls_entry)) + tsrm_reserved_pos;
+	ptrdiff_t fixed_offset = TSRM_ALIGNED_SIZE(sizeof(tsrm_tls_entry)) + tsrm_reserved_pos;
 	tsrm_reserved_pos += size;
+	tsrm_mutex_unlock(tsmm_mutex);
+
+	return ts_allocate_fast_id_at(rsrc_id, offset, fixed_offset, size, ctor, dtor);
+}/*}}}*/
+
+
+TSRM_API ts_rsrc_id ts_allocate_fast_id_at(ts_rsrc_id *rsrc_id, size_t *offset, ptrdiff_t fixed_offset, size_t size, ts_allocate_ctor ctor, ts_allocate_dtor dtor)
+{
+	TSRM_ERROR((TSRM_ERROR_LEVEL_CORE, "Obtaining a new fast resource id, %d bytes", size));
+
+	tsrm_mutex_lock(tsmm_mutex);
+
+	/* obtain a resource id */
+	*rsrc_id = TSRM_SHUFFLE_RSRC_ID(id_count++);
+	TSRM_ERROR((TSRM_ERROR_LEVEL_CORE, "Obtained resource id %d", *rsrc_id));
+
+	size = TSRM_ALIGNED_SIZE(size);
+	*offset = (size_t) fixed_offset;
 
 	/* store the new resource type in the resource sizes table */
 	if (resource_types_table_size < id_count) {
@@ -366,7 +388,7 @@ TSRM_API ts_rsrc_id ts_allocate_fast_id(ts_rsrc_id *rsrc_id, size_t *offset, siz
 
 	TSRM_ERROR((TSRM_ERROR_LEVEL_CORE, "Successfully allocated new resource id %d", *rsrc_id));
 	return *rsrc_id;
-}/*}}}*/
+}
 
 static void set_thread_local_storage_resource_to(tsrm_tls_entry *thread_resource)
 {
@@ -378,7 +400,10 @@ static void set_thread_local_storage_resource_to(tsrm_tls_entry *thread_resource
 static void allocate_new_resource(tsrm_tls_entry **thread_resources_ptr, THREAD_T thread_id)
 {/*{{{*/
 	TSRM_ERROR((TSRM_ERROR_LEVEL_CORE, "Creating data structures for thread %x", thread_id));
-	(*thread_resources_ptr) = (tsrm_tls_entry *) malloc(TSRM_ALIGNED_SIZE(sizeof(tsrm_tls_entry)) + tsrm_reserved_size);
+	/* The entry follows the fixed-offset front region.
+	 * hot globals live at negative offsets from the TLS cache pointer. */
+	char *block = (char *) malloc(tsrm_reserved_front + TSRM_ALIGNED_SIZE(sizeof(tsrm_tls_entry)) + tsrm_reserved_size);
+	(*thread_resources_ptr) = (tsrm_tls_entry *) (block + tsrm_reserved_front);
 	(*thread_resources_ptr)->storage = NULL;
 	if (id_count > 0) {
 		(*thread_resources_ptr)->storage = (void **) malloc(sizeof(void *)*id_count);
@@ -487,7 +512,7 @@ TSRM_API void *ts_resource_ex(ts_rsrc_id id, THREAD_T *th_id)
 		set_thread_local_storage_resource_to(thread_resources);
 		/* Free up the old resource from the old thread instance */
 		ts_free_resources(thread_resources);
-		free(thread_resources);
+		free((char *) thread_resources - tsrm_reserved_front);
 		/* Allocate a new resource at the same point in the linked list, and relink the next pointer */
 		allocate_new_resource(last_thread_resources, thread_id);
 		thread_resources = *last_thread_resources;
@@ -529,7 +554,7 @@ void ts_free_thread(void)
 				tsrm_tls_table[hash_value] = thread_resources->next;
 			}
 			tsrm_tls_set(0);
-			free(thread_resources);
+			free((char *) thread_resources - tsrm_reserved_front);
 			break;
 		}
 		if (thread_resources->next) {

TSRM/TSRM.h

@@ -20,6 +20,7 @@
 # include <main/php_config.h>
 #endif
 
+#include <stddef.h>
 #include <stdint.h>
 #include <stdbool.h>
 
@@ -94,6 +95,11 @@ TSRM_API ts_rsrc_id ts_allocate_id(ts_rsrc_id *rsrc_id, size_t size, ts_allocate
 TSRM_API void tsrm_reserve(size_t size);
 TSRM_API ts_rsrc_id ts_allocate_fast_id(ts_rsrc_id *rsrc_id, size_t *offset, size_t size, ts_allocate_ctor ctor, ts_allocate_dtor dtor);
 
+/* Fast resources at caller-chosen, compile-time-constant offsets. The fixed
+ * front region must be reserved after tsrm_reserve() and before any fast id. */
+TSRM_API void tsrm_reserve_fast_front(size_t size);
+TSRM_API ts_rsrc_id ts_allocate_fast_id_at(ts_rsrc_id *rsrc_id, size_t *offset, ptrdiff_t fixed_offset, size_t size, ts_allocate_ctor ctor, ts_allocate_dtor dtor);
+
 /* fetches the requested resource for the current thread */
 TSRM_API void *ts_resource_ex(ts_rsrc_id id, THREAD_T *th_id);
 #define ts_resource(id)			ts_resource_ex(id, NULL)

Zend/zend.c

@@ -1019,9 +1019,12 @@ void zend_startup(zend_utility_functions *utility_functions) /* {{{ */
 	zend_init_rsrc_list_dtors();
 
 #ifdef ZTS
-	ts_allocate_fast_id(&compiler_globals_id, &compiler_globals_offset, sizeof(zend_compiler_globals), (ts_allocate_ctor) compiler_globals_ctor, (ts_allocate_dtor) compiler_globals_dtor);
-	ts_allocate_fast_id(&executor_globals_id, &executor_globals_offset, sizeof(zend_executor_globals), (ts_allocate_ctor) executor_globals_ctor, (ts_allocate_dtor) executor_globals_dtor);
-	ts_allocate_fast_id(&language_scanner_globals_id, &language_scanner_globals_offset, sizeof(zend_php_scanner_globals), (ts_allocate_ctor) php_scanner_globals_ctor, NULL);
+	ts_allocate_fast_id_at(&compiler_globals_id, &compiler_globals_offset, ZEND_CG_OFFSET, sizeof(zend_compiler_globals), (ts_allocate_ctor) compiler_globals_ctor, (ts_allocate_dtor) compiler_globals_dtor);
+	ts_allocate_fast_id_at(&executor_globals_id, &executor_globals_offset, ZEND_EG_OFFSET, sizeof(zend_executor_globals), (ts_allocate_ctor) executor_globals_ctor, (ts_allocate_dtor) executor_globals_dtor);
+	ts_allocate_fast_id_at(&language_scanner_globals_id, &language_scanner_globals_offset, ZEND_SCNG_OFFSET, sizeof(zend_php_scanner_globals), (ts_allocate_ctor) php_scanner_globals_ctor, NULL);
+	ZEND_ASSERT(compiler_globals_offset == ZEND_CG_OFFSET);
+	ZEND_ASSERT(executor_globals_offset == ZEND_EG_OFFSET);
+	ZEND_ASSERT(language_scanner_globals_offset == ZEND_SCNG_OFFSET);
 	ts_allocate_fast_id(&ini_scanner_globals_id, &ini_scanner_globals_offset, sizeof(zend_ini_scanner_globals), (ts_allocate_ctor) ini_scanner_globals_ctor, NULL);
 	compiler_globals = ts_resource(compiler_globals_id);
 	executor_globals = ts_resource(executor_globals_id);

Zend/zend_alloc.c

@@ -2614,7 +2614,8 @@ typedef struct _zend_alloc_globals {
 #ifdef ZTS
 static int alloc_globals_id;
 static size_t alloc_globals_offset;
-# define AG(v) ZEND_TSRMG_FAST(alloc_globals_offset, zend_alloc_globals *, v)
+# define ZEND_AG_OFFSET (ZEND_SCNG_OFFSET - (ptrdiff_t) TSRM_ALIGNED_SIZE(sizeof(zend_alloc_globals)))
+# define AG(v) ZEND_TSRMG_FAST(ZEND_AG_OFFSET, zend_alloc_globals *, v)
 #else
 # define AG(v) (alloc_globals.v)
 static zend_alloc_globals alloc_globals;
@@ -3335,7 +3336,8 @@ ZEND_API void start_memory_manager(void)
 #  endif
 #endif
 #ifdef ZTS
-	ts_allocate_fast_id(&alloc_globals_id, &alloc_globals_offset, sizeof(zend_alloc_globals), (ts_allocate_ctor) alloc_globals_ctor, (ts_allocate_dtor) alloc_globals_dtor);
+	ts_allocate_fast_id_at(&alloc_globals_id, &alloc_globals_offset, ZEND_AG_OFFSET, sizeof(zend_alloc_globals), (ts_allocate_ctor) alloc_globals_ctor, (ts_allocate_dtor) alloc_globals_dtor);
+	ZEND_ASSERT(alloc_globals_offset == ZEND_AG_OFFSET);
 #else
 	alloc_globals_ctor(&alloc_globals);
 #endif

Zend/zend_globals.h

@@ -329,6 +329,14 @@ struct _zend_executor_globals {
 	void *reserved[ZEND_MAX_RESERVED_RESOURCES];
 };
 
+#ifdef ZTS
+/* Compile-time offsets of the hot globals, in a reserved region just before the
+ * cache pointer. ZEND_AG_OFFSET is furthest, in zend_alloc.c. */
+# define ZEND_CG_OFFSET   (-(ptrdiff_t) TSRM_ALIGNED_SIZE(sizeof(zend_compiler_globals)))
+# define ZEND_EG_OFFSET   (ZEND_CG_OFFSET - (ptrdiff_t) TSRM_ALIGNED_SIZE(sizeof(zend_executor_globals)))
+# define ZEND_SCNG_OFFSET (ZEND_EG_OFFSET - (ptrdiff_t) TSRM_ALIGNED_SIZE(sizeof(zend_php_scanner_globals)))
+#endif
+
 #define EG_FLAGS_INITIAL				(0)
 #define EG_FLAGS_IN_SHUTDOWN			(1<<0)
 #define EG_FLAGS_OBJECT_STORE_NO_REUSE	(1<<1)

Zend/zend_globals_macros.h

@@ -30,7 +30,7 @@ BEGIN_EXTERN_C()
 
 /* Compiler */
 #ifdef ZTS
-# define CG(v) ZEND_TSRMG_FAST(compiler_globals_offset, zend_compiler_globals *, v)
+# define CG(v) ZEND_TSRMG_FAST(ZEND_CG_OFFSET, zend_compiler_globals *, v)
 #else
 # define CG(v) (compiler_globals.v)
 extern ZEND_API struct _zend_compiler_globals compiler_globals;
@@ -40,15 +40,15 @@ ZEND_API int zendparse(void);
 
 /* Executor */
 #ifdef ZTS
-# define EG(v) ZEND_TSRMG_FAST(executor_globals_offset, zend_executor_globals *, v)
+# define EG(v) ZEND_TSRMG_FAST(ZEND_EG_OFFSET, zend_executor_globals *, v)
 #else
 # define EG(v) (executor_globals.v)
 extern ZEND_API zend_executor_globals executor_globals;
 #endif
 
 /* Language Scanner */
 #ifdef ZTS
-# define LANG_SCNG(v) ZEND_TSRMG_FAST(language_scanner_globals_offset, zend_php_scanner_globals *, v)
+# define LANG_SCNG(v) ZEND_TSRMG_FAST(ZEND_SCNG_OFFSET, zend_php_scanner_globals *, v)
 extern ZEND_API ts_rsrc_id language_scanner_globals_id;
 extern ZEND_API size_t language_scanner_globals_offset;
 #else

main/main.c

@@ -2842,6 +2842,12 @@ PHPAPI bool php_tsrm_startup_ex(int expected_threads)
 {
 	bool ret = tsrm_startup(expected_threads, 1, 0, NULL);
 	php_reserve_tsrm_memory();
+	/* Must cover the total size of every ZEND_*_OFFSET global, or the furthest underflows the block. */
+	tsrm_reserve_fast_front(
+		TSRM_ALIGNED_SIZE(sizeof(zend_compiler_globals)) +
+		TSRM_ALIGNED_SIZE(sizeof(zend_executor_globals)) +
+		TSRM_ALIGNED_SIZE(sizeof(zend_php_scanner_globals)) +
+		TSRM_ALIGNED_SIZE(zend_mm_globals_size())); // AG size, exposed through function call
 	(void)ts_resource(0);
 	return ret;
 }