[prefix] Cope with image source addresses outside base memory

When PMM is used, the gPXE image source will no longer be in base memory.
Decompression of .text16 and .data16 can therefore no longer be done in
real mode.

src/arch/i386/prefix/libprefix.S

@@ -175,13 +175,14 @@ print_hex_nibble:
 #ifndef KEEP_IT_REAL
 
 	/* GDT for protected-mode calls */
-	.section ".data16"
+	.section ".prefix.lib"
 	.align 16
+pm_call_vars:
 gdt:
 gdt_limit:		.word gdt_length - 1
 gdt_base:		.long 0
 			.word 0 /* padding */
-pm_cs:		/* 16-bit protected-mode code segment */
+pm_cs:		/* 16-bit protected-mode code segment */	
 	.equ    PM_CS, pm_cs - gdt
 	.word   0xffff, 0
 	.byte   0, 0x9b, 0x00, 0
@@ -197,18 +198,24 @@ gdt_end:
 	.equ	gdt_length, . - gdt
 	.size	gdt, . - gdt
 
-	.section ".data16"
+	.section ".prefix.lib"
 	.align 16
 pm_saved_gdt:	
 	.long	0, 0
 	.size	pm_saved_gdt, . - pm_saved_gdt
 
+	.equ	pm_call_vars_size, . - pm_call_vars
+#define PM_CALL_VAR(x) ( -pm_call_vars_size + ( (x) - pm_call_vars ) )
+
 	.section ".prefix.lib"
 	.code16
 pm_call:
-	/* Preserve registers, flags, GDT, and RM return point */
+	/* Preserve registers, flags, and RM return point */
+	pushw	%bp
+	movw	%sp, %bp
+	subw	$pm_call_vars_size, %sp
+	andw	$0xfff0, %sp
 	pushfl
-	sgdt	pm_saved_gdt
 	pushw	%gs
 	pushw	%fs
 	pushw	%es
@@ -217,27 +224,43 @@ pm_call:
 	pushw	%cs
 	pushw	$99f
 
+	/* Set up local variable block, and preserve GDT */
+	pushw	%cx
+	pushw	%si
+	pushw	%di
+	pushw	%ss
+	popw	%es
+	movw	$pm_call_vars, %si
+	leaw	PM_CALL_VAR(pm_call_vars)(%bp), %di
+	movw	$pm_call_vars_size, %cx
+	cs rep movsb
+	popw	%di
+	popw	%si
+	popw	%cx
+	sgdt	PM_CALL_VAR(pm_saved_gdt)(%bp)
+
 	/* Set up GDT bases */
 	pushl	%eax
-	pushw	%bx
+	pushl	%edi
 	xorl	%eax, %eax
-	movw	%ds, %ax
+	movw	%ss, %ax
 	shll	$4, %eax
-	addl	$gdt, %eax
-	movl	%eax, gdt_base
+	movzwl	%bp, %edi
+	leal	PM_CALL_VAR(gdt)(%eax, %edi), %eax
+	movl	%eax, PM_CALL_VAR(gdt_base)(%bp)
 	movw	%cs, %ax
-	movw	$pm_cs, %bx
+	movw	$PM_CALL_VAR(pm_cs), %di
 	call	set_seg_base
 	movw	%ss, %ax
-	movw	$pm_ss, %bx
+	movw	$PM_CALL_VAR(pm_ss), %di
 	call	set_seg_base
-	popw	%bx
+	popl	%edi
 	popl	%eax
 
 	/* Switch CPU to protected mode and load up segment registers */
 	pushl	%eax
 	cli
-	lgdt	gdt
+	lgdt	PM_CALL_VAR(gdt)(%bp)
 	movl	%cr0, %eax
 	orb	$CR0_PE, %al
 	movl	%eax, %cr0
@@ -273,18 +296,19 @@ pm_call:
 	popw	%es
 	popw	%fs
 	popw	%gs
-	lgdt	pm_saved_gdt
+	lgdt	PM_CALL_VAR(pm_saved_gdt)(%bp)
 	popfl
-
+	movw	%bp, %sp
+	popw	%bp
 	ret
 	.size pm_call, . - pm_call
 
 set_seg_base:
 	rolw	$4, %ax
-	movw	%ax, 2(%bx)
-	andw	$0xfff0, 2(%bx)
-	movb	%al, 4(%bx)
-	andb	$0x0f, 4(%bx)
+	movw	%ax, 2(%bp,%di)
+	andw	$0xfff0, 2(%bp,%di)
+	movb	%al, 4(%bp,%di)
+	andb	$0x0f, 4(%bp,%di)
 	ret
 	.size set_seg_base, . - set_seg_base
 
@@ -301,7 +325,7 @@ set_seg_base:
  *   %ecx : length
  * Returns:
  *   %ds:esi : next source address
- *   %ds:esi : next destination address
+ *   %es:edi : next destination address
  * Corrupts:
  *   None
  ****************************************************************************
@@ -316,27 +340,57 @@ copy_bytes:
 	.size copy_bytes, . - copy_bytes
 
 /****************************************************************************
- * install_block (real-mode or 16-bit protected-mode near call)
+ * install_block (real-mode near call)
  *
  * Install block to specified address
  *
  * Parameters:
- *   %ds:esi : source address (must be a multiple of 16)
- *   %es:edi : destination address
+ *   %esi : source physical address (must be a multiple of 16)
+ *   %edi : destination physical address (must be a multiple of 16)
  *   %ecx : length of (decompressed) data
  *   %edx : total length of block (including any uninitialised data portion)
  * Returns:
- *   %ds:esi : next source address (will be a multiple of 16)
+ *   %esi : next source physical address (will be a multiple of 16)
  * Corrupts:
- *   %ecx, %edx
+ *   none
  ****************************************************************************
  */
 	.section ".prefix.lib"
 	.code16
 install_block:
+	
+#ifdef KEEP_IT_REAL
+
+	/* Preserve registers */
+	pushw	%ds
+	pushw	%es
+	pushl	%ecx
+	pushl	%edi
+	
+	/* Convert %esi and %edi to segment registers */
+	shrl	$4, %esi
+	movw	%si, %ds
+	xorw	%si, %si
+	shrl	$4, %edi
+	movw	%di, %es
+	xorw	%di, %di
+
+#else /* KEEP_IT_REAL */
+
+	/* Call self in protected mode */
+	pushw	%ax
+	movw	$1f, %ax
+	call	pm_call
+	popw	%ax
+	ret
+1:
 	/* Preserve registers */
+	pushl	%ecx
 	pushl	%edi
 	
+#endif /* KEEP_IT_REAL */
+
+	
 #if COMPRESS
 	/* Decompress source to destination */
 	call	decompress16
@@ -357,8 +411,28 @@ install_block:
 	addl	$0xf, %esi
 	andl	$~0xf, %esi
 
-	/* Restore registers and return */
+
+#ifdef KEEP_IT_REAL
+
+	/* Convert %ds:esi back to a physical address */
+	movzwl	%ds, %cx
+	shll	$4, %ecx
+	addl	%ecx, %esi
+
+	/* Restore registers */
 	popl	%edi
+	popl	%ecx
+	popw	%es
+	popw	%ds
+
+#else /* KEEP_IT_REAL */
+
+	/* Restore registers */
+	popl	%edi
+	popl	%ecx
+
+#endif
+
 	ret
 	.size install_block, . - install_block
 	
@@ -406,87 +480,6 @@ alloc_basemem:
 	ret
 	.size alloc_basemem, . - alloc_basemem
 
-/****************************************************************************
- * install_basemem (real-mode near call)
- *
- * Install source block into base memory
- *
- * Parameters:
- *   %esi : source physical address (must be a multiple of 16)
- *   %es : destination segment address
- *   %cx : length of (decompressed) data
- *   %dx : total length of block (including any uninitialised data portion)
- * Returns:
- *   %esi : next source physical address (will be a multiple of 16)
- * Corrupts:
- *   %ecx, %edx
- ****************************************************************************
- */
-	.section ".prefix.lib"
-	.code16
-install_basemem:
-	/* Preserve registers */
-	pushl	%edi
-	pushw	%ds
-
-	/* Preserve original %esi */
-	pushl	%esi
-
-	/* Install to specified address */
-	shrl	$4, %esi
-	movw	%si, %ds
-	xorw	%si, %si
-	xorl	%edi, %edi
-	movzwl	%cx, %ecx
-	movzwl	%dx, %edx
-	call	install_block
-
-	/* Fix up %esi for return */
-	popl	%ecx
-	addl	%ecx, %esi
-
-	/* Restore registers */
-	popw	%ds
-	popl	%edi
-	ret
-	.size install_basemem, . - install_basemem
-
-/****************************************************************************
- * install_highmem (real-mode near call)
- *
- * Install source block into high memory
- *
- * Parameters:
- *   %esi : source physical address (must be a multiple of 16)
- *   %edi : destination physical address
- *   %ecx : length of (decompressed) data
- *   %edx : total length of block (including any uninitialised data portion)
- * Returns:
- *   %esi : next source physical address (will be a multiple of 16)
- * Corrupts:
- *   %ecx, %edx
- ****************************************************************************
- */
-
-#ifndef KEEP_IT_REAL
-
-	.section ".prefix.lib"
-	.code16
-install_highmem:
-	/* Preserve registers */
-	pushw	%ax
-
-	/* Install to specified address */
-	movw	$install_block, %ax
-	call	pm_call
-
-	/* Restore registers */
-	popw	%ax
-	ret
-	.size install_highmem, . - install_highmem
-	
-#endif /* KEEP_IT_REAL */
-	
 /****************************************************************************
  * install (real-mode near call)
  *
@@ -555,17 +548,19 @@ install_prealloc:
 	shll	$4, %esi
 1:	addl	$_payload_offset, %esi
 
-	/* Install .text16 */
-	movw	%ax, %es
-	movw	$_text16_size, %cx
-	movw	%cx, %dx
-	call	install_basemem
-
-	/* Install .data16 */
-	movw	%bx, %es
-	movw	$_data16_progbits_size, %cx
-	movw	$_data16_size, %dx
-	call	install_basemem
+	/* Install .text16 and .data16 */
+	pushl	%edi
+	movzwl	%ax, %edi
+	shll	$4, %edi
+	movl	$_text16_size, %ecx
+	movl	%ecx, %edx
+	call	install_block		/* .text16 */
+	movzwl	%bx, %edi
+	shll	$4, %edi
+	movl	$_data16_progbits_size, %ecx
+	movl	$_data16_size, %edx
+	call	install_block		/* .data16 */
+	popl	%edi
 
 	/* Set up %ds for access to .data16 */
 	movw	%bx, %ds
@@ -581,9 +576,7 @@ install_prealloc:
 	 */
 	movl	$_textdata_progbits_size, %ecx
 	movl	$_textdata_size, %edx
-	pushl	%edi
-	call	install_highmem
-	popl	%edi
+	call	install_block
 
 	/* Initialise librm at current location */
 	movw	%ax, (init_librm_vector+2)