[romprefix] Add a dummy ROM header to cover the .mrom payload

The header of a .mrom image declares its length to be only a few
kilobytes; the remainder is accessed via a sideband mechanism.  This
makes it difficult to append an additional ROM image, such as an EFI
ROM.

Add a second, dummy ROM header covering the payload portion of the
.mrom image, allowing consumers to locate any appended ROM images in
the usual way.

Signed-off-by: Michael Brown <mcb30@ipxe.org>

src/arch/i386/prefix/libprefix.S

 	/* Save decompression temporary area physical address */
 	pushl	%edi
 
-	/* Install .text16.early */
+	/* Install .text16.early and calculate %ecx as offset to next block */
 	progress "  .text16.early\n"
 	pushl	%esi
 	xorl	%esi, %esi
 	movw	%cs, %si
 	shll	$4, %esi
+	pushl	%esi			/* Save original %cs:0000 */
 	addl	$_text16_early_lma, %esi
 	movzwl	%ax, %edi
 	shll	$4, %edi
 	movl	$_text16_early_filesz, %ecx
 	movl	$_text16_early_memsz, %edx
 	call	install_block		/* .text16.early */
+	popl	%ecx			/* Calculate offset to next block */
+	subl	%esi, %ecx
+	negl	%ecx
 	popl	%esi
 
 #ifndef KEEP_IT_REAL
 	jnz	1f
 	movw	%cs, %si
 	shll	$4, %esi
-1:	addl	payload_lma, %esi
+1:	addl	%ecx, %esi
 
 	/* Install .text16.late and .data16 */
 	progress "  .text16.late\n"
 	.word 0
 	.size close_payload_vector, . - close_payload_vector
 
-	/* Payload address */
-	.section ".prefix.lib", "awx", @progbits
-payload_lma:
-	.long 0
-	.section ".zinfo.fixup", "a", @progbits	/* Compressor fixups */
-	.ascii	"ADHL"
-	.long	payload_lma
-	.long	1
-	.long	0
-	.previous
-
 	/* Dummy routines to open and close payload */
 	.section ".text16.early.data", "aw", @progbits
 	.weak	open_payload
 	.ascii	"PAYL"
 	.long	0
 	.long	0
+	.long	_payload_align
+	.ascii	"COPY"
+	.long	_pprefix_lma
+	.long	_pprefix_filesz
 	.long	_max_align
 	.ascii	PACK_OR_COPY
 	.long	_text16_late_lma
 	.long	_textdata_lma
 	.long	_textdata_filesz
 	.long	_max_align
+
+	.weak	_payload_align
+	.equ	_payload_align, 1

src/arch/i386/prefix/mromprefix.S

 #define PCI_BAR_EXPROM			0x30
 
 #define ROMPREFIX_EXCLUDE_PAYLOAD 1
+#define ROMPREFIX_MORE_IMAGES 1
 #define _rom_start _mrom_start
 #include "romprefix.S"
 
  * Parameters:
  *   %ds:0000 : Prefix
  *   %esi : Buffer for copy of image source (or zero if no buffer available)
+ *   %ecx : Expected offset within buffer of first payload block
  * Returns:
  *   %esi : Valid image source address (buffered or unbuffered)
+ *   %ecx : Actual offset within buffer of first payload block
  *   CF set on error
  */
 	.section ".text16.early", "awx", @progbits
 	/* Preserve registers */
 	pushl	%eax
 	pushw	%bx
-	pushl	%ecx
 	pushl	%edx
 	pushl	%edi
 	pushw	%bp
+	pushw	%es
 	pushw	%ds
 
-	/* Retrieve bus:dev.fn and image source length from .prefix */
+	/* Retrieve bus:dev.fn from .prefix */
 	movw	init_pci_busdevfn, %bx
-	movl	image_source_len_dword, %ecx
 
 	/* Set up %ds for access to .text16.early */
 	pushw	%cs
 	popw	%ds
 
-	/* Store bus:dev.fn and image source length to .text16.early */
+	/* Set up %es for access to flat address space */
+	xorw	%ax, %ax
+	movw	%ax, %es
+
+	/* Store bus:dev.fn to .text16.early */
 	movw	%bx, payload_pci_busdevfn
-	movl	%ecx, rom_bar_copy_len_dword
 
 	/* Get expansion ROM BAR current value */
 	movw	$PCI_BAR_EXPROM, %di
 	 * properly support flat real mode, it will die horribly.)
 	 */
 	pushl	%esi
-	pushw	%es
 	movl	%esi, %edi
 	movl	%eax, %esi
-	movl	rom_bar_copy_len_dword, %ecx
-	xorw	%ax, %ax
-	movw	%ax, %es
+	addr32 es movzbl 2(%esi), %ecx
+	shll	$7, %ecx
+	addr32 es movzbl 2(%esi,%ecx,4), %edx
+	shll	$7, %edx
+	addl	%edx, %ecx
 	addr32 es rep movsl
-	popw	%es
 	popl	%esi
 	jmp	2f
 1:	/* We have no buffer; set %esi to the BAR address */
 	movl	%eax, %esi
 2:
 
+	/* Locate first payload block (after the dummy ROM header) */
+	addr32 es movzbl 2(%esi), %ecx
+	shll	$9, %ecx
+	addl	$_pprefix_skip, %ecx
+
 	clc
 	/* Restore registers and return */
 99:	popw	%ds
+	popw	%es
 	popw	%bp
 	popl	%edi
 	popl	%edx
-	popl	%ecx
 	popw	%bx
 	popl	%eax
 	lret
 	.long	0
 	.size	rom_bar_size, . - rom_bar_size
 
-	.section ".text16.early.data", "aw", @progbits
-rom_bar_copy_len_dword:
-	.long	0
-	.size	rom_bar_copy_len_dword, . - rom_bar_copy_len_dword
-
 	.section ".text16.early.data", "aw", @progbits
 stolen_bar_register:
 	.word	0
 	ret
 	.size	pci_set_mem_access, . - pci_set_mem_access
 
-/* Image source area length (in dwords)
+/* Payload prefix
  *
+ * We include a dummy ROM header to cover the "hidden" portion of the
+ * overall ROM image.
  */
-	.section ".prefix", "ax", @progbits
-image_source_len_dword:
+	.globl	_payload_align
+	.equ	_payload_align, 512
+	.section ".pprefix", "ax", @progbits
+	.org	0x00
+mromheader:
+	.word	0xaa55			/* BIOS extension signature */
+mromheader_size: .byte 0		/* Size in 512-byte blocks */
+	.org	0x18
+	.word	mpciheader
+	.org	0x1a
+	.word	0
+	.size	mromheader, . - mromheader
+
+	.section ".zinfo.fixup", "a", @progbits	/* Compressor fixups */
+	.ascii	"APPB"
+	.long	mromheader_size
+	.long	512
+	.long	0
+	.previous
+
+mpciheader:
+	.ascii	"PCIR"			/* Signature */
+	.word	pci_vendor_id		/* Vendor identification */
+	.word	pci_device_id		/* Device identification */
+	.word	0x0000			/* Device list pointer */
+	.word	mpciheader_len		/* PCI data structure length */
+	.byte	0x03			/* PCI data structure revision */
+	.byte	0x02, 0x00, 0x00	/* Class code */
+mpciheader_image_length:
+	.word	0			/* Image length */
+	.word	0x0001			/* Revision level */
+	.byte	0xff			/* Code type */
+	.byte	0x80			/* Last image indicator */
+mpciheader_runtime_length:
+	.word	0			/* Maximum run-time image length */
+	.word	0x0000			/* Configuration utility code header */
+	.word	0x0000			/* DMTF CLP entry point */
+	.equ	mpciheader_len, . - mpciheader
+	.size	mpciheader, . - mpciheader
+
+	.section ".zinfo.fixup", "a", @progbits	/* Compressor fixups */
+	.ascii	"APPW"
+	.long	mpciheader_image_length
+	.long	512
 	.long	0
-	.size	image_source_len_dword, . - image_source_len_dword
+	.ascii	"APPW"
+	.long	mpciheader_runtime_length
+	.long	512
+	.long	0
+	.previous
+
+/* Fix up additional image source size
+ *
+ */
 	.section ".zinfo.fixup", "a", @progbits	/* Compressor fixups */
-	.ascii	"ADDL"
-	.long	image_source_len_dword
-	.long	4
+	.ascii	"ADPW"
+	.long	extra_size
+	.long	512
 	.long	0
 	.previous

src/arch/i386/prefix/romprefix.S

 #else
 #define	ZINFO_TYPE_ADxB "ADDB"
 #define	ZINFO_TYPE_ADxW "ADDW"
+#endif
+
+/* Allow ROM to be marked as containing multiple images
+ */
+#if ROMPREFIX_MORE_IMAGES
+#define INDICATOR 0x00
+#else
+#define INDICATOR 0x80
 #endif
 
 	.text
 	.word	0			/* Image length */
 	.word	0x0001			/* Revision level */
 	.byte	0x00			/* Code type */
-	.byte	0x80			/* Last image indicator */
+	.byte	INDICATOR		/* Last image indicator */
 pciheader_runtime_length:
 	.word	0			/* Maximum run-time image length */
 	.word	0x0000			/* Configuration utility code header */
 	.long	pciheader_image_length
 	.long	512
 	.long	0
-	.ascii	ZINFO_TYPE_ADxW
+	.ascii	"ADHW"
 	.long	pciheader_runtime_length
 	.long	512
 	.long	0
 	/* We have PMM and so a 1kB stack: preserve whole registers */
 	pushal
 	/* Allocate image source PMM block */
-	movzwl	image_source_size, %ecx
+	movzbl	romheader_size, %ecx
+	addw	extra_size, %cx
 	shll	$5, %ecx
 	movl	$PMM_HANDLE_BASE_IMAGE_SOURCE, %ebx
 	movw	$get_pmm_image_source, %bp
 	movl	%esi, %edi
 	xorl	%esi, %esi
 	movzbl	romheader_size, %ecx
-	shll	$9, %ecx
-	addr32 rep movsb	/* PMM presence implies flat real mode */
+	shll	$7, %ecx
+	addr32 rep movsl	/* PMM presence implies flat real mode */
 	popw	%es
 	/* Shrink ROM */
-	movb	shrunk_rom_size, %al
+	movw	pciheader_runtime_length, %ax
 	movb	%al, romheader_size
 1:	/* Allocate decompression PMM block.  Round up the size to the
 	 * nearest 128kB and use the size within the PMM handle; this
 	.long	0
 	.size	image_source, . - image_source
 
-/* Image source size (in 512-byte sectors)
+/* Additional image source size (in 512-byte sectors)
  *
  */
-image_source_size:
+extra_size:
 	.word	0
-	.size	image_source_size, . - image_source_size
-	.section ".zinfo.fixup", "a", @progbits	/* Compressor fixups */
-	.ascii	"ADDW"
-	.long	image_source_size
-	.long	512
-	.long	0
-	.previous
-
-/* Shrunk ROM size (in 512-byte sectors)
- *
- */
-shrunk_rom_size:
-	.byte	0
-	.size	shrunk_rom_size, . - shrunk_rom_size
-	.section ".zinfo.fixup", "a", @progbits	/* Compressor fixups */
-	.ascii	"ADHB"
-	.long	shrunk_rom_size
-	.long	512
-	.long	0
-	.previous
+	.size	extra_size, . - extra_size
 
 /* Temporary decompression area
  *

src/arch/i386/scripts/i386.lds

-/* -*- sh -*- */
+/* -*- ld-script -*- */
 
 /*
  * Linker script for i386 images
     _textdata_filesz	= ABSOLUTE ( _mtextdata ) - ABSOLUTE ( _textdata );
     _textdata_memsz	= ABSOLUTE ( _etextdata ) - ABSOLUTE ( _textdata );
 
+    /*
+     * Payload prefix
+     *
+     * If present, this will be placed between .text16.early and .text16.late.
+     *
+     */
+    .pprefix 0x0 : AT ( _pprefix_lma ) {
+	_pprefix = .;
+	KEEP(*(.pprefix))
+	KEEP(*(.pprefix.*))
+	_mpprefix = .;
+    } .bss.pprefix (NOLOAD) : AT ( _end_lma ) {
+	_epprefix = .;
+    }
+    _pprefix_filesz	= ABSOLUTE ( _mpprefix ) - ABSOLUTE ( _pprefix );
+    _pprefix_memsz	= ABSOLUTE ( _epprefix ) - ABSOLUTE ( _pprefix );
+
     /*
      * Compressor information block
      *
     _text16_early_lma	= .;
     .			+= _text16_early_filesz;
 
+    .			= ALIGN ( _max_align );
+    .			= ALIGN ( _payload_align );
+    _pprefix_lma	= .;
+    .			+= _pprefix_filesz;
+
     .			= ALIGN ( _max_align );
     _payload_lma	= .;
+    _pprefix_skip	= ABSOLUTE ( _payload_lma ) - ABSOLUTE ( _pprefix_lma );
     _text16_late_lma	= .;
     .			+= _text16_late_filesz;
 

src/util/zbin.c

 					__attribute__ (( unused )),
 			       struct output_file *output,
 			       size_t len,
-			       struct zinfo_add *add,
+			       struct zinfo_add *add, size_t offset,
 			       size_t datasize ) {
-	size_t offset = add->offset;
 	void *target;
 	signed long addend;
 	unsigned long size;
 	signed long val;
 	unsigned long mask;
 
+	offset += add->offset;
 	if ( ( offset + datasize ) > output->len ) {
 		fprintf ( stderr, "Add at %#zx outside output buffer\n",
 			  offset );
 				struct output_file *output,
 				union zinfo_record *zinfo ) {
 	return process_zinfo_add ( input, output, output->len,
-				   &zinfo->add, 1 );
+				   &zinfo->add, 0, 1 );
 }
 
 static int process_zinfo_addw ( struct input_file *input,
 				struct output_file *output,
 				union zinfo_record *zinfo ) {
 	return process_zinfo_add ( input, output, output->len,
-				   &zinfo->add, 2 );
+				   &zinfo->add, 0, 2 );
 }
 
 static int process_zinfo_addl ( struct input_file *input,
 				struct output_file *output,
 				union zinfo_record *zinfo ) {
 	return process_zinfo_add ( input, output, output->len,
-				   &zinfo->add, 4 );
+				   &zinfo->add, 0, 4 );
 }
 
 static int process_zinfo_adhb ( struct input_file *input,
 				struct output_file *output,
 				union zinfo_record *zinfo ) {
 	return process_zinfo_add ( input, output, output->hdr_len,
-				   &zinfo->add, 1 );
+				   &zinfo->add, 0, 1 );
 }
 
 static int process_zinfo_adhw ( struct input_file *input,
 				struct output_file *output,
 				union zinfo_record *zinfo ) {
 	return process_zinfo_add ( input, output, output->hdr_len,
-				   &zinfo->add, 2 );
+				   &zinfo->add, 0, 2 );
 }
 
 static int process_zinfo_adhl ( struct input_file *input,
 				struct output_file *output,
 				union zinfo_record *zinfo ) {
 	return process_zinfo_add ( input, output, output->hdr_len,
-				   &zinfo->add, 4 );
+				   &zinfo->add, 0, 4 );
+}
+
+static int process_zinfo_adpb ( struct input_file *input,
+				struct output_file *output,
+				union zinfo_record *zinfo ) {
+	return process_zinfo_add ( input, output,
+				   ( output->len - output->hdr_len ),
+				   &zinfo->add, 0, 1 );
+}
+
+static int process_zinfo_adpw ( struct input_file *input,
+				struct output_file *output,
+				union zinfo_record *zinfo ) {
+	return process_zinfo_add ( input, output,
+				   ( output->len - output->hdr_len ),
+				   &zinfo->add, 0, 2 );
+}
+
+static int process_zinfo_adpl ( struct input_file *input,
+				struct output_file *output,
+				union zinfo_record *zinfo ) {
+	return process_zinfo_add ( input, output,
+				   ( output->len - output->hdr_len ),
+				   &zinfo->add, 0, 4 );
+}
+
+static int process_zinfo_appb ( struct input_file *input,
+				struct output_file *output,
+				union zinfo_record *zinfo ) {
+	return process_zinfo_add ( input, output,
+				   ( output->len - output->hdr_len ),
+				   &zinfo->add, output->hdr_len, 1 );
+}
+
+static int process_zinfo_appw ( struct input_file *input,
+				struct output_file *output,
+				union zinfo_record *zinfo ) {
+	return process_zinfo_add ( input, output,
+				   ( output->len - output->hdr_len ),
+				   &zinfo->add, output->hdr_len, 2 );
+}
+
+static int process_zinfo_appl ( struct input_file *input,
+				struct output_file *output,
+				union zinfo_record *zinfo ) {
+	return process_zinfo_add ( input, output,
+				   ( output->len - output->hdr_len ),
+				   &zinfo->add, output->hdr_len, 4 );
 }
 
 struct zinfo_processor {
 	{ "ADHB", process_zinfo_adhb },
 	{ "ADHW", process_zinfo_adhw },
 	{ "ADHL", process_zinfo_adhl },
+	{ "ADPB", process_zinfo_adpb },
+	{ "ADPW", process_zinfo_adpw },
+	{ "ADPL", process_zinfo_adpl },
+	{ "APPB", process_zinfo_appb },
+	{ "APPW", process_zinfo_appw },
+	{ "APPL", process_zinfo_appl },
 };
 
 static int process_zinfo ( struct input_file *input,