ffi 1.15.5 → 1.16.0

data/ext/ffi_c/libffi/src/arm/ffi.c

@@ -34,9 +34,10 @@
 #include <ffi_common.h>
 #include <stdint.h>
 #include <stdlib.h>
+#include <tramp.h>
 #include "internal.h"
 
-#if defined(_MSC_VER) && defined(_M_ARM)
+#if defined(_WIN32)
 #define WIN32_LEAN_AND_MEAN
 #include <windows.h>
 #endif
@@ -48,10 +49,13 @@
 #endif
 
 #else
-#ifndef _M_ARM
+#ifndef _WIN32
 extern unsigned int ffi_arm_trampoline[2] FFI_HIDDEN;
 #else
-extern unsigned int ffi_arm_trampoline[3] FFI_HIDDEN;
+// Declare this as an array of char, instead of array of int,
+// otherwise Clang optimizes out the "& 0xFFFFFFFE" for clearing
+// the thumb bit.
+extern unsigned char ffi_arm_trampoline[12] FFI_HIDDEN;
 #endif
 #endif
 
@@ -60,6 +64,10 @@ extern unsigned int ffi_arm_trampoline[3] FFI_HIDDEN;
 #include <machine/sysarch.h>
 #endif
 
+#if defined(__QNX__)
+#include <sys/mman.h>
+#endif
+
 /* Forward declares. */
 static int vfp_type_p (const ffi_type *);
 static void layout_vfp_args (ffi_cif *);
@@ -103,13 +111,13 @@ ffi_put_arg (ffi_type *ty, void *src, void *dst)
     case FFI_TYPE_SINT32:
     case FFI_TYPE_UINT32:
     case FFI_TYPE_POINTER:
-#ifndef _MSC_VER
+#ifndef _WIN32
     case FFI_TYPE_FLOAT:
 #endif
       *(UINT32 *)dst = *(UINT32 *)src;
       break;
 
-#ifdef _MSC_VER
+#ifdef _WIN32
     // casting a float* to a UINT32* doesn't work on Windows
     case FFI_TYPE_FLOAT:
         *(uintptr_t *)dst = 0;
@@ -536,6 +544,8 @@ ffi_prep_incoming_args_VFP (ffi_cif *cif, void *rvalue, char *stack,
   return rvalue;
 }
 
+#if FFI_CLOSURES
+
 struct closure_frame
 {
   char vfp_space[8*8] __attribute__((aligned(8)));
@@ -571,6 +581,10 @@ ffi_closure_inner_VFP (ffi_cif *cif,
 
 void ffi_closure_SYSV (void) FFI_HIDDEN;
 void ffi_closure_VFP (void) FFI_HIDDEN;
+#if defined(FFI_EXEC_STATIC_TRAMP)
+void ffi_closure_SYSV_alt (void) FFI_HIDDEN;
+void ffi_closure_VFP_alt (void) FFI_HIDDEN;
+#endif
 
 #ifdef FFI_GO_CLOSURES
 void ffi_go_closure_SYSV (void) FFI_HIDDEN;
@@ -612,27 +626,44 @@ ffi_prep_closure_loc (ffi_closure * closure,
   config[1] = closure_func;
 #else
 
-#ifndef _M_ARM
+# if defined(FFI_EXEC_STATIC_TRAMP)
+  if (ffi_tramp_is_present(closure))
+    {
+      /* Initialize the static trampoline's parameters. */
+      if (closure_func == ffi_closure_SYSV)
+        closure_func = ffi_closure_SYSV_alt;
+      else
+        closure_func = ffi_closure_VFP_alt;
+      ffi_tramp_set_parms (closure->ftramp, closure_func, closure);
+      goto out;
+    }
+# endif
+
+  /* Initialize the dynamic trampoline. */
+# ifndef _WIN32
   memcpy(closure->tramp, ffi_arm_trampoline, 8);
-#else
+# else
   // cast away function type so MSVC doesn't set the lower bit of the function pointer
   memcpy(closure->tramp, (void*)((uintptr_t)ffi_arm_trampoline & 0xFFFFFFFE), FFI_TRAMPOLINE_CLOSURE_OFFSET);
-#endif
+# endif
 
-#if defined (__QNX__)
-  msync(closure->tramp, 8, 0x1000000);	/* clear data map */
-  msync(codeloc, 8, 0x1000000);	/* clear insn map */
-#elif defined(_MSC_VER)
+# if defined(__QNX__)
+  msync (closure->tramp, 8, MS_INVALIDATE_ICACHE);	/* clear data map */
+  msync (codeloc, 8, MS_INVALIDATE_ICACHE);		/* clear insn map */
+# elif defined(_WIN32)
   FlushInstructionCache(GetCurrentProcess(), closure->tramp, FFI_TRAMPOLINE_SIZE);
-#else
+# else
   __clear_cache(closure->tramp, closure->tramp + 8);	/* clear data map */
   __clear_cache(codeloc, codeloc + 8);			/* clear insn map */
-#endif
-#ifdef _M_ARM
+# endif
+# ifdef _WIN32
   *(void(**)(void))(closure->tramp + FFI_TRAMPOLINE_CLOSURE_FUNCTION) = closure_func;
-#else
+# else
   *(void (**)(void))(closure->tramp + 8) = closure_func;
-#endif
+# endif
+# if defined(FFI_EXEC_STATIC_TRAMP)
+out:
+# endif
 #endif
 
   closure->cif = cif;
@@ -666,6 +697,8 @@ ffi_prep_go_closure (ffi_go_closure *closure, ffi_cif *cif,
 }
 #endif
 
+#endif /* FFI_CLOSURES */
+
 /* Below are routines for VFP hard-float support. */
 
 /* A subroutine of vfp_type_p.  Given a structure type, return the type code
@@ -873,4 +906,16 @@ layout_vfp_args (ffi_cif * cif)
     }
 }
 
+#if defined(FFI_EXEC_STATIC_TRAMP)
+void *
+ffi_tramp_arch (size_t *tramp_size, size_t *map_size)
+{
+  extern void *trampoline_code_table;
+
+  *tramp_size = ARM_TRAMP_SIZE;
+  *map_size = ARM_TRAMP_MAP_SIZE;
+  return &trampoline_code_table;
+}
+#endif
+
 #endif /* __arm__ or _M_ARM */

data/ext/ffi_c/libffi/src/arm/ffitarget.h

@@ -43,7 +43,7 @@ typedef enum ffi_abi {
   FFI_SYSV,
   FFI_VFP,
   FFI_LAST_ABI,
-#if defined(__ARM_PCS_VFP) || defined(_M_ARM)
+#if defined(__ARM_PCS_VFP) || defined(_WIN32)
   FFI_DEFAULT_ABI = FFI_VFP,
 #else
   FFI_DEFAULT_ABI = FFI_SYSV,
@@ -57,7 +57,7 @@ typedef enum ffi_abi {
   signed char vfp_args[16]			\
 
 #define FFI_TARGET_SPECIFIC_VARIADIC
-#ifndef _M_ARM
+#ifndef _WIN32
 #define FFI_TARGET_HAS_COMPLEX_TYPE
 #endif
 
@@ -77,7 +77,7 @@ typedef enum ffi_abi {
 #endif
 
 #else
-#ifdef _MSC_VER
+#ifdef _WIN32
 #define FFI_TRAMPOLINE_SIZE 16
 #define FFI_TRAMPOLINE_CLOSURE_FUNCTION 12
 #else

data/ext/ffi_c/libffi/src/arm/internal.h

@@ -5,3 +5,13 @@
 #define ARM_TYPE_INT	4
 #define ARM_TYPE_VOID	5
 #define ARM_TYPE_STRUCT	6
+
+#if defined(FFI_EXEC_STATIC_TRAMP)
+/*
+ * For the trampoline table mapping, a mapping size of 4K (base page size)
+ * is chosen.
+ */
+#define ARM_TRAMP_MAP_SHIFT	12
+#define ARM_TRAMP_MAP_SIZE	(1 << ARM_TRAMP_MAP_SHIFT)
+#define ARM_TRAMP_SIZE		20
+#endif

data/ext/ffi_c/libffi/src/arm/sysv.S

@@ -92,9 +92,25 @@
 #define ARM_FUNC_END(name) \
 	SIZE(name)
 
+	.text
+	.syntax unified
+#if defined(_WIN32)
+	/* Windows on ARM is thumb-only */
+	.thumb
+#else
+	/* Keep the assembly in ARM mode in other cases, for simplicity
+	 * (to avoid interworking issues). */
+#undef __thumb__
+	.arm
+#endif
+
 /* Aid in defining a jump table with 8 bytes between entries.  */
+#ifdef __thumb__
+/* In thumb mode, instructions can be shorter than expected in arm mode, so
+ * we need to align the start of each case. */
+# define E(index) .align 3
+#elif defined(__clang__)
 /* ??? The clang assembler doesn't handle .if with symbolic expressions.  */
-#ifdef __clang__
 # define E(index)
 #else
 # define E(index)				\
@@ -103,9 +119,6 @@
 	.endif
 #endif
 
-	.text
-	.syntax unified
-	.arm
 
 #ifndef __clang__
 	/* We require interworking on LDM, which implies ARMv5T,
@@ -128,13 +141,9 @@ ARM_FUNC_START(ffi_call_VFP)
 	cfi_startproc
 
 	cmp	r3, #3			@ load only d0 if possible
-#ifdef __clang__
-	vldrle d0, [r0]
-	vldmgt r0, {d0-d7}
-#else
+	ite	le
 	ldcle	p11, cr0, [r0]		@ vldrle d0, [r0]
 	ldcgt	p11, cr0, [r0], {16}	@ vldmgt r0, {d0-d7}
-#endif
 	add	r0, r0, #64		@ discard the vfp register args
 	/* FALLTHRU */
 ARM_FUNC_END(ffi_call_VFP)
@@ -167,30 +176,25 @@ ARM_FUNC_START(ffi_call_SYSV)
 	cfi_def_cfa_register(sp)
 
 	@ Store values stored in registers.
+#ifndef __thumb__
 	.align	3
 	add	pc, pc, r3, lsl #3
 	nop
+#else
+	adr	ip, 0f
+	add	ip, ip, r3, lsl #3
+	mov	pc, ip
+	.align	3
+#endif
 0:
 E(ARM_TYPE_VFP_S)
-#ifdef __clang__
-	vstr s0, [r2]
-#else
 	stc	p10, cr0, [r2]		@ vstr s0, [r2]
-#endif
 	pop	{fp,pc}
 E(ARM_TYPE_VFP_D)
-#ifdef __clang__
-	vstr d0, [r2]
-#else
 	stc	p11, cr0, [r2]		@ vstr d0, [r2]
-#endif
 	pop	{fp,pc}
 E(ARM_TYPE_VFP_N)
-#ifdef __clang__
-	vstm r2, {d0-d3}
-#else
 	stc	p11, cr0, [r2], {8}	@ vstm r2, {d0-d3}
-#endif
 	pop	{fp,pc}
 E(ARM_TYPE_INT64)
 	str	r1, [r2, #4]
@@ -208,6 +212,7 @@ E(ARM_TYPE_STRUCT)
 	UNWIND(.fnend)
 ARM_FUNC_END(ffi_call_SYSV)
 
+#if FFI_CLOSURES
 
 /*
 	int ffi_closure_inner_* (cif, fun, user_data, frame)
@@ -227,6 +232,9 @@ ARM_FUNC_END(ffi_go_closure_SYSV)
 ARM_FUNC_START(ffi_closure_SYSV)
 	UNWIND(.fnstart)
 	cfi_startproc
+#ifdef _WIN32
+	ldmfd	sp!, {r0, ip}			@ restore fp (r0 is used for stack alignment)
+#endif
 	stmdb	sp!, {r0-r3}			@ save argument regs
 	cfi_adjust_cfa_offset(16)
 
@@ -255,7 +263,12 @@ ARM_FUNC_START(ffi_closure_SYSV)
 	@ Load values returned in registers.
 	add	r2, sp, #8+64			@ load result
 	adr	r3, CNAME(ffi_closure_ret)
+#ifndef __thumb__
 	add	pc, r3, r0, lsl #3
+#else
+	add	r3, r3, r0, lsl #3
+	mov	pc, r3
+#endif
 	cfi_endproc
 	UNWIND(.fnend)
 ARM_FUNC_END(ffi_closure_SYSV)
@@ -274,6 +287,9 @@ ARM_FUNC_END(ffi_go_closure_VFP)
 ARM_FUNC_START(ffi_closure_VFP)
 	UNWIND(.fnstart)
 	cfi_startproc
+#ifdef _WIN32
+	ldmfd	sp!, {r0, ip}			@ restore fp (r0 is used for stack alignment)
+#endif
 	stmdb	sp!, {r0-r3}			@ save argument regs
 	cfi_adjust_cfa_offset(16)
 
@@ -287,11 +303,7 @@ ARM_FUNC_START(ffi_closure_VFP)
 	add	ip, sp, #16
 	sub	sp, sp, #64+32			@ allocate frame
 	cfi_adjust_cfa_offset(64+32)
-#ifdef __clang__
-	vstm sp, {d0-d7}
-#else
 	stc	p11, cr0, [sp], {16}		@ vstm sp, {d0-d7}
-#endif
 	stmdb	sp!, {ip,lr}
 
 	/* See above.  */
@@ -305,7 +317,12 @@ ARM_FUNC_START(ffi_closure_VFP)
 	@ Load values returned in registers.
 	add	r2, sp, #8+64			@ load result
 	adr	r3, CNAME(ffi_closure_ret)
+#ifndef __thumb__
 	add	pc, r3, r0, lsl #3
+#else
+	add	r3, r3, r0, lsl #3
+	mov	pc, r3
+#endif
 	cfi_endproc
 	UNWIND(.fnend)
 ARM_FUNC_END(ffi_closure_VFP)
@@ -320,40 +337,85 @@ ARM_FUNC_START_LOCAL(ffi_closure_ret)
 	cfi_rel_offset(lr, 4)
 0:
 E(ARM_TYPE_VFP_S)
-#ifdef __clang__
-	vldr s0, [r2]
-#else
 	ldc	p10, cr0, [r2]			@ vldr s0, [r2]
-#endif
-	ldm	sp, {sp,pc}
+	b	call_epilogue
 E(ARM_TYPE_VFP_D)
-#ifdef __clang__
-	vldr d0, [r2]
-#else
 	ldc	p11, cr0, [r2]			@ vldr d0, [r2]
-#endif
-	ldm	sp, {sp,pc}
+	b	call_epilogue
 E(ARM_TYPE_VFP_N)
-#ifdef __clang__
-	vldm r2, {d0-d3}
-#else
 	ldc	p11, cr0, [r2], {8}		@ vldm r2, {d0-d3}
-#endif
-	ldm	sp, {sp,pc}
+	b	call_epilogue
 E(ARM_TYPE_INT64)
 	ldr	r1, [r2, #4]
 	nop
 E(ARM_TYPE_INT)
 	ldr	r0, [r2]
-	ldm	sp, {sp,pc}
+	b	call_epilogue
 E(ARM_TYPE_VOID)
-	ldm	sp, {sp,pc}
+	b	call_epilogue
 	nop
 E(ARM_TYPE_STRUCT)
+	b	call_epilogue
+call_epilogue:
+#ifndef __thumb__
 	ldm	sp, {sp,pc}
+#else
+	ldm	sp, {ip,lr}
+	mov	sp, ip
+	bx	lr
+#endif
 	cfi_endproc
 ARM_FUNC_END(ffi_closure_ret)
 
+#if defined(FFI_EXEC_STATIC_TRAMP)
+ARM_FUNC_START(ffi_closure_SYSV_alt)
+	/* See the comments above trampoline_code_table. */
+	ldr	ip, [sp, #4]			/* Load closure in ip */
+	add	sp, sp, 8			/* Restore the stack */
+	b	CNAME(ffi_closure_SYSV)
+ARM_FUNC_END(ffi_closure_SYSV_alt)
+
+ARM_FUNC_START(ffi_closure_VFP_alt)
+	/* See the comments above trampoline_code_table. */
+	ldr	ip, [sp, #4]			/* Load closure in ip */
+	add	sp, sp, 8			/* Restore the stack */
+	b	CNAME(ffi_closure_VFP)
+ARM_FUNC_END(ffi_closure_VFP_alt)
+
+/*
+ * Below is the definition of the trampoline code table. Each element in
+ * the code table is a trampoline.
+ */
+/*
+ * The trampoline uses register ip (r12). It saves the original value of ip
+ * on the stack.
+ *
+ * The trampoline has two parameters - target code to jump to and data for
+ * the target code. The trampoline extracts the parameters from its parameter
+ * block (see tramp_table_map()). The trampoline saves the data address on
+ * the stack. Finally, it jumps to the target code.
+ *
+ * The target code can choose to:
+ *
+ * - restore the value of ip
+ * - load the data address in a register
+ * - restore the stack pointer to what it was when the trampoline was invoked.
+ */
+	.align	ARM_TRAMP_MAP_SHIFT
+ARM_FUNC_START(trampoline_code_table)
+	.rept	ARM_TRAMP_MAP_SIZE / ARM_TRAMP_SIZE
+	sub	sp, sp, #8		/* Make space on the stack */
+	str	ip, [sp]		/* Save ip on stack */
+	ldr	ip, [pc, #4080]		/* Copy data into ip */
+	str	ip, [sp, #4]		/* Save data on stack */
+	ldr	pc, [pc, #4076]		/* Copy code into PC */
+	.endr
+ARM_FUNC_END(trampoline_code_table)
+	.align	ARM_TRAMP_MAP_SHIFT
+#endif /* FFI_EXEC_STATIC_TRAMP */
+
+#endif /* FFI_CLOSURES */
+
 #if FFI_EXEC_TRAMPOLINE_TABLE
 
 #ifdef __MACH__
@@ -369,6 +431,15 @@ ARM_FUNC_START(ffi_closure_trampoline_table_page)
 ARM_FUNC_END(ffi_closure_trampoline_table_page)
 #endif
 
+#elif defined(_WIN32)
+
+ARM_FUNC_START(ffi_arm_trampoline)
+0:	adr	ip, 0b
+	stmdb	sp!, {r0, ip}
+	ldr	pc, 1f
+1:	.long	0
+ARM_FUNC_END(ffi_arm_trampoline)
+
 #else
 
 ARM_FUNC_START(ffi_arm_trampoline)