Fix GH-11188: Error when building TSRM in ARM64

Although the issue mentioned FreeBSD, this is a broader problem:
the current ARM64 code to load the TLS offset assumes a setup with
the non-default TLS model. This problem can also apply on some
configurations on other platforms.

Closes GH-11236.

Author: nielsdos

NEWS

@@ -30,6 +30,7 @@ PHP                                                                        NEWS
   . Fixed bug GH-14267 (opcache.jit=off does not allow enabling JIT at runtime).
     (ilutov)
   . Fixed TLS access in JIT on FreeBSD/amd64. (Arnaud)
+  . Fixed bug GH-11188 (Error when building TSRM in ARM64). (nielsdos)
 
 - Soap:
   . Fixed bug #47925 (PHPClient can't decompress response). (nielsdos)

TSRM/TSRM.c

@@ -777,11 +777,20 @@ TSRM_API size_t tsrm_get_ls_cache_tcb_offset(void)
 	asm("adrp %0, #__tsrm_ls_cache@TLVPPAGE\n\t"
 	    "ldr %0, [%0, #__tsrm_ls_cache@TLVPPAGEOFF]"
 	     : "=r" (ret));
-# else
+# elif defined(TSRM_TLS_MODEL_DEFAULT)
+	/* Surplus Static TLS space isn't guaranteed. */
+	ret = 0;
+# elif defined(TSRM_TLS_MODEL_INITIAL_EXEC)
+	asm("adrp %0, :gottprel:_tsrm_ls_cache\n\t"
+		"ldr %0, [%0, #:gottprel_lo12:_tsrm_ls_cache]"
+		: "=r" (ret));
+# elif defined(TSRM_TLS_MODEL_LOCAL_EXEC)
 	asm("mov %0, xzr\n\t"
 	    "add %0, %0, #:tprel_hi12:_tsrm_ls_cache, lsl #12\n\t"
 	    "add %0, %0, #:tprel_lo12_nc:_tsrm_ls_cache"
 	     : "=r" (ret));
+# else
+#  error "TSRM TLS model not set"
 # endif
 	return ret;
 #else

TSRM/TSRM.h

@@ -150,10 +150,13 @@ TSRM_API const char *tsrm_api_name(void);
 
 #if !__has_attribute(tls_model) || defined(__FreeBSD__) || defined(__MUSL__) || defined(__HAIKU__)
 # define TSRM_TLS_MODEL_ATTR
+# define TSRM_TLS_MODEL_DEFAULT
 #elif __PIC__
 # define TSRM_TLS_MODEL_ATTR __attribute__((tls_model("initial-exec")))
+# define TSRM_TLS_MODEL_INITIAL_EXEC
 #else
 # define TSRM_TLS_MODEL_ATTR __attribute__((tls_model("local-exec")))
+# define TSRM_TLS_MODEL_LOCAL_EXEC
 #endif
 
 #define TSRM_SHUFFLE_RSRC_ID(rsrc_id)		((rsrc_id)+1)

ext/opcache/jit/zend_jit_arm64.dasc

@@ -110,13 +110,22 @@ static void* dasm_labels[zend_lb_MAX];
 
 #if ZTS
 static size_t tsrm_ls_cache_tcb_offset = 0;
+static size_t tsrm_tls_index = 0;
+static size_t tsrm_tls_offset = 0;
 # ifdef __APPLE__
 struct TLVDescriptor {
 	void*       (*thunk)(struct TLVDescriptor*);
 	uint64_t    key;
 	uint64_t    offset;
 };
 typedef struct TLVDescriptor TLVDescriptor;
+# elif defined(__FreeBSD__)
+/* https://github.com/freebsd/freebsd-src/blob/c52ca7dd09066648b1cc40f758289404d68ab886/libexec/rtld-elf/aarch64/reloc.c#L180-L184 */
+typedef struct TLSDescriptor {
+	void*   thunk;
+	int     index;
+	size_t  offset;
+} TLSDescriptor;
 # endif
 #endif
 
@@ -494,8 +503,14 @@ static bool logical_immediate_p(uint64_t value, uint32_t reg_size)
 |	MEM_ACCESS_64_WITH_UOFFSET_64 ldr, reg, TMP3, (((TLVDescriptor*)tsrm_ls_cache_tcb_offset)->offset), TMP1
 ||#else
 |	.long 0xd53bd051 // TODO: hard-coded: mrs TMP3, tpidr_el0
-||	ZEND_ASSERT(tsrm_ls_cache_tcb_offset <= LDR_STR_PIMM64);
-|	ldr reg, [TMP3, #tsrm_ls_cache_tcb_offset]
+||	if (tsrm_ls_cache_tcb_offset == 0) {
+|		ldr TMP3, [TMP3, #0]
+|		MEM_ACCESS_64_WITH_UOFFSET_64 ldr, TMP3, TMP3, tsrm_tls_index, TMP1
+|		MEM_ACCESS_64_WITH_UOFFSET_64 ldr, reg, TMP3, tsrm_tls_offset, TMP1
+||	} else {
+||		ZEND_ASSERT(tsrm_ls_cache_tcb_offset <= LDR_STR_PIMM64);
+|		ldr reg, [TMP3, #tsrm_ls_cache_tcb_offset]
+||	}
 ||#endif
 |.endmacro
 
@@ -2754,7 +2769,27 @@ static int zend_jit_setup(void)
 
 #if ZTS
 	tsrm_ls_cache_tcb_offset = tsrm_get_ls_cache_tcb_offset();
+# if defined(__FreeBSD__)
+	if (tsrm_ls_cache_tcb_offset == 0) {
+		TLSDescriptor **where;
+
+		__asm__(
+			"adrp %0, :tlsdesc:_tsrm_ls_cache\n"
+			"add %0, %0, :tlsdesc_lo12:_tsrm_ls_cache\n"
+			: "=r" (where));
+		/* See https://github.com/ARM-software/abi-aa/blob/2a70c42d62e9c3eb5887fa50b71257f20daca6f9/aaelf64/aaelf64.rst
+		 * section "Relocations for thread-local storage".
+		 * The first entry holds a pointer to the variable's TLS descriptor resolver function and the second entry holds
+		 * a platform-specific offset or pointer. */
+		TLSDescriptor *tlsdesc = where[1];
+
+		tsrm_tls_offset = tlsdesc->offset;
+		/* Index is offset by 1 on FreeBSD (https://github.com/freebsd/freebsd-src/blob/22ca6db50f4e6bd75a141f57cf953d8de6531a06/lib/libc/gen/tls.c#L88) */
+		tsrm_tls_index = (tlsdesc->index + 1) * 8;
+	}
+# else
 	ZEND_ASSERT(tsrm_ls_cache_tcb_offset != 0);
+# endif
 #endif
 
     memset(sp_adj, 0, sizeof(sp_adj));