Improvements for Apple?

lib/i386/relocator_asm.S

@@ -27,30 +27,28 @@
 #ifdef __x86_64__
 #define RAX %rax
 #define RCX %rcx
+#define RDX %rdx
 #define RDI %rdi
 #define RSI %rdi
 #else
 #define RAX %eax
 #define RCX %ecx
+#define RDX %edx
 #define RDI %edi	
 #define RSI %esi	
 #endif
 
-/* Apple's linker has a problem with 64-bit relocations.  */
-#if defined (__apple__) || ! defined (__x86_64__)
-#define RSIA %esi
-#define RAXA %eax
-#else
-#define RSIA %rsi
-#define RAXA %rax
-#endif
+/* The code segment of the protected mode.  */
+#define CODE_SEGMENT	0x10
 
+/* The data segment of the protected mode.  */
+#define DATA_SEGMENT	0x18
 	
 	.p2align	4	/* force 16-byte alignment */
 
 RELOCATOR_VARIABLE(start)
 #ifdef BACKWARD
-base:
+L_base:
 #endif
 	cli
 
@@ -105,14 +103,14 @@ RELOCATOR_VARIABLE(size)
 #ifndef BACKWARD
 	add	RCX, RAX
 #endif
-	addq	$0x3, RCX
-	shrq	$2, RCX
+	add	$0x3, RCX
+	shr	$2, RCX
 
 	
 #ifdef BACKWARD
 	/* Backward movsl is implicitly off-by-four.  compensate that.  */
-	subq	$4,	RSI
-	subq	$4,	RDI
+	sub	$4,	RSI
+	sub	$4,	RDI
 
 	/* Backward copy.  */
 	std
@@ -129,27 +127,27 @@ RELOCATOR_VARIABLE(size)
 
 	/* %rax contains now our new 'base'.  */
 	mov	RAX, RSI
-	add	$(cont0 - base), RAXA
+	add	$(L_cont0 - L_base), RAX
 	jmp	*RAX
-cont0:
-	lea	(cont1 - base) (RSIA, 1), RAXA
-	movl	%eax, (jump_vector - base) (RSIA, 1)
+L_cont0:
+	lea	(L_cont1 - L_base) (RSI, 1), RAX
+	movl	%eax, (L_jump_vector - L_base) (RSI, 1)
 
-	lea	(gdt - base) (RSIA, 1), RAXA
-	mov	RAXA, (gdt_addr - base) (RSIA, 1)
+	lea	(L_gdt - L_base) (RSI, 1), RAX
+	mov	RAX, (L_gdt_addr - L_base) (RSI, 1)
 
 	/* Switch to compatibility mode. */
 
-	lgdt	(gdtdesc - base) (RSIA, 1)
+	lgdt	(L_gdtdesc - L_base) (RSI, 1)
 
 	/* Update %cs. Thanks to David Miller for pointing this mistake out. */
-	ljmp	*(jump_vector - base) (RSIA, 1)
+	ljmp	*(L_jump_vector - L_base) (RSI, 1)
 
-cont1:
+L_cont1:
 	.code32
 
 	/* Update other registers. */
-	movl	$GRUB_MEMORY_MACHINE_PROT_MODE_DSEG, %eax
+	movl	$DATA_SEGMENT, %eax
 	movl	%eax, %ds
 	movl	%eax, %es
 	movl	%eax, %fs
@@ -158,22 +156,22 @@ cont1:
 
 	/* Disable paging. */
 	movl	%cr0, %eax
-	andl	$0x7fffffff, %eax
+	andl	$(~GRUB_MEMORY_CPU_CR0_PAGING_ON), %eax
 	movl	%eax, %cr0
 
 	/* Disable amd64. */
-	movl	$0xc0000080, %ecx
+	movl	$GRUB_MEMORY_CPU_AMD64_MSR, %ecx
 	rdmsr
-	andl	$0xfffffeff, %eax
+	andl	$(~GRUB_MEMORY_CPU_AMD64_MSR_ON), %eax
 	wrmsr
 
 	/* Turn off PAE. */
 	movl	%cr4, %eax
-	andl	$0xffffffcf, %eax
+	andl	$GRUB_MEMORY_CPU_CR4_PAE_ON, %eax
 	movl	%eax, %cr4
 
-	jmp	cont2
-cont2:
+	jmp	L_cont2
+L_cont2:
 	.code32
 
 	/* mov imm32, %eax */
@@ -212,44 +210,25 @@ RELOCATOR_VARIABLE (eip)
 	.long	0
 	.word	0x08
 
-	/* GDT. The same as is used in 32-bit GRUB. */
+	/* GDT. Copied from loader/i386/linux.c. */
 	.p2align	4
-gdt:
-	.word	0, 0
-	.byte	0, 0, 0, 0
+L_gdt:
+	/* NULL.  */
+	.byte 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00
+	
+	/* Reserved.  */
+	.byte 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00
 
-	/* -- code segment --
-	 * base = 0x00000000, limit = 0xFFFFF (4 KiB Granularity), present
-	 * type = 32bit code execute/read, DPL = 0
-	 */
-	.word	0xFFFF, 0
-	.byte	0, 0x9A, 0xCF, 0
+	/* Code segment.  */
+	.byte 0xFF, 0xFF, 0x00, 0x00, 0x00, 0x9A, 0xCF, 0x00
 
-	/* -- data segment --
-	 * base = 0x00000000, limit 0xFFFFF (4 KiB Granularity), present
-	 * type = 32 bit data read/write, DPL = 0
-	 */
-	.word	0xFFFF, 0
-	.byte	0, 0x92, 0xCF, 0
-
-	/* -- 16 bit real mode CS --
-	 * base = 0x00000000, limit 0x0FFFF (1 B Granularity), present
-	 * type = 16 bit code execute/read only/conforming, DPL = 0
-	 */
-	.word	0xFFFF, 0
-	.byte	0, 0x9E, 0, 0
-
-	/* -- 16 bit real mode DS --
-	 * base = 0x00000000, limit 0x0FFFF (1 B Granularity), present
-	 * type = 16 bit data read/write, DPL = 0
-	 */
-	.word	0xFFFF, 0
-	.byte	0, 0x92, 0, 0
+	/* Data segment.  */
+	.byte 0xFF, 0xFF, 0x00, 0x00, 0x00, 0x92, 0xCF, 0x00
 
 	.p2align	4
-gdtdesc:
+L_gdtdesc:
 	.word	0x27
-gdt_addr:
+L_gdt_addr:
 #ifdef __x86_64__
 	/* Filled by the code. */
 	.quad	0
@@ -259,13 +238,13 @@ gdt_addr:
 #endif
 	
 	.p2align	4
-jump_vector:
+L_jump_vector:
 	/* Jump location. Is filled by the code */
 	.long	0
-	.long	GRUB_MEMORY_MACHINE_PROT_MODE_CSEG
+	.long	CODE_SEGMENT
 
 #ifndef BACKWARD
-base:
+L_base:
 #endif
 
 RELOCATOR_VARIABLE(end)