[prefix] Add .text16.early section

Add a section .text16.early which is always kept inline with the
prefix.  This will allow for some code sharing between the .prefix and
.text16 sections.

Note that the simple solution of just prepending the .prefix section
to the .text16 section will not work, because a bug in Wyse Streaming
Manager server (WLDRM13.BIN) requires us to place a dummy PXENV+ entry
point at the start of .text16.

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

src/arch/i386/prefix/libprefix.S

@@ -406,6 +406,7 @@ copy_bytes:
  *   %edx : total length of block (including any uninitialised data portion)
  * Returns:
  *   %esi : next source physical address (will be a multiple of 16)
+ *   %edi : next destination physical address (will be a multiple of 16)
  * Corrupts:
  *   none
  ****************************************************************************
@@ -417,7 +418,6 @@ install_block:
 	pushw	%ds
 	pushw	%es
 	pushl	%ecx
-	pushl	%edi
 	
 	/* Convert %esi and %edi to %ds:esi and %es:edi */
 	shrl	$4, %esi
@@ -445,18 +445,23 @@ install_block:
 	rep addr32 stosb
 	popw	%ax
 
-	/* Round up %esi to start of next source block */
+	/* Round up %esi and %edi to start of next blocks */
 	addl	$0xf, %esi
 	andl	$~0xf, %esi
+	addl	$0xf, %edi
+	andl	$~0xf, %edi
 
-	/* Convert %ds:esi back to a physical address */
+	/* Convert %ds:esi and %es:edi back to physical addresses */
 	xorl	%ecx, %ecx
 	movw	%ds, %cx
 	shll	$4, %ecx
 	addl	%ecx, %esi
+	xorl	%ecx, %ecx
+	movw	%es, %cx
+	shll	$4, %ecx
+	addl	%ecx, %edi
 
 	/* Restore registers and return */
-	popl	%edi
 	popl	%ecx
 	popw	%es
 	popw	%ds
@@ -626,6 +631,23 @@ install_prealloc:
 	/* Sanity: clear the direction flag asap */
 	cld
 
+	/* Copy decompression temporary area physical address to %ebp */
+	movl	%edi, %ebp
+
+	/* Install .text16.early */
+	pushl	%esi
+	xorl	%esi, %esi
+	movw	%cs, %si
+	shll	$4, %esi
+	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	%esi
+
+	/* Open up access to payload */
 #ifndef KEEP_IT_REAL
 	/* Flatten real mode */
 	call	flatten_real_mode
@@ -636,21 +658,17 @@ install_prealloc:
 	jnz	1f
 	movw	%cs, %si
 	shll	$4, %esi
-1:	addl	$_payload_lma, %esi
+1:	addl	%cs:payload_lma, %esi
 
-	/* Install .text16 and .data16 */
-	pushl	%edi
-	movzwl	%ax, %edi
-	shll	$4, %edi
-	movl	$_text16_memsz, %ecx
-	movl	%ecx, %edx
-	call	install_block		/* .text16 */
+	/* Install .text16.late and .data16 */
+	movl	$_text16_late_filesz, %ecx
+	movl	$_text16_late_memsz, %edx
+	call	install_block		/* .text16.late */
 	movzwl	%bx, %edi
 	shll	$4, %edi
 	movl	$_data16_filesz, %ecx
 	movl	$_data16_memsz, %edx
 	call	install_block		/* .data16 */
-	popl	%edi
 
 	/* Set up %ds for access to .data16 */
 	movw	%bx, %ds
@@ -664,12 +682,14 @@ install_prealloc:
 	 * prior to reading the E820 memory map and relocating
 	 * properly.
 	 */
+	movl	%ebp, %edi
 	movl	$_textdata_filesz, %ecx
 	movl	$_textdata_memsz, %edx
 	call	install_block
 
 	/* Initialise librm at current location */
 	movw	%ax, (init_librm_vector+2)
+	movl	%ebp, %edi
 	lcall	*init_librm_vector
 
 	/* Call relocate() to determine target address for relocation.
@@ -682,13 +702,13 @@ install_prealloc:
 	popl	%edx /* discard */
 
 	/* Copy code to new location */
-	pushl	%edi
 	xorw	%ax, %ax
 	movw	%ax, %es
+	movl	%ebp, %edi
 	es rep addr32 movsb
-	popl	%edi
 
 	/* Initialise librm at new location */
+	movl	%ebp, %edi
 	lcall	*init_librm_vector
 #endif
 
@@ -719,6 +739,17 @@ prot_call_vector:
 	.size prot_call_vector, . - prot_call_vector
 #endif
 
+	/* 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
+
 /****************************************************************************
  * uninstall
  *
@@ -745,27 +776,32 @@ uninstall:
 
 	/* File split information for the compressor */
 #if COMPRESS
+#define PACK_OR_COPY	"PACK"
+#else
+#define PACK_OR_COPY	"COPY"
+#endif
 	.section ".zinfo", "a", @progbits
 	.ascii	"COPY"
 	.long	_prefix_lma
 	.long	_prefix_filesz
 	.long	_max_align
-	.ascii	"PACK"
-	.long	_text16_lma
-	.long	_text16_filesz
+	.ascii	PACK_OR_COPY
+	.long	_text16_early_lma
+	.long	_text16_early_filesz
 	.long	_max_align
-	.ascii	"PACK"
+	.ascii	"PAYL"
+	.long	0
+	.long	0
+	.long	_max_align
+	.ascii	PACK_OR_COPY
+	.long	_text16_late_lma
+	.long	_text16_late_filesz
+	.long	_max_align
+	.ascii	PACK_OR_COPY
 	.long	_data16_lma
 	.long	_data16_filesz
 	.long	_max_align
-	.ascii	"PACK"
+	.ascii	PACK_OR_COPY
 	.long	_textdata_lma
 	.long	_textdata_filesz
 	.long	_max_align
-#else /* COMPRESS */
-	.section ".zinfo", "a", @progbits
-	.ascii	"COPY"
-	.long	_prefix_lma
-	.long	_filesz
-	.long	_max_align
-#endif /* COMPRESS */

src/arch/i386/prefix/romprefix.S

@@ -362,7 +362,8 @@ got_pmm: /* PMM allocation succeeded */
 	addr32 rep movsb	/* PMM presence implies flat real mode */
 	movl	%edi, decompress_to
 	/* Shrink ROM */
-	movb	$_prefix_memsz_sect, romheader_size
+	movb	shrunk_rom_size, %al
+	movb	%al, romheader_size
 pmm_fail:
 	/* Restore upper register halves */
 	popal
@@ -488,6 +489,19 @@ image_source:
 	.long	0
 	.size	image_source, . - image_source
 
+/* 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
+
 /* Temporary decompression area
  *
  * May be either at HIGHMEM_LOADPOINT, or within PMM-allocated block.

src/arch/i386/scripts/i386.lds

@@ -45,17 +45,25 @@ SECTIONS {
      *
      */
 
-    .text16 0x0 : AT ( _text16_lma ) {
+    .text16.early 0x0 : AT ( _text16_early_lma ) {
 	_text16 = .;
 	*(.text16.null)
 	. += 1;				/* Prevent NULL being valid */
+	*(.text16.early)
+	*(.text16.early.*)
+	_etext16_early = .;
+    } .text16.late ALIGN ( _max_align ) : AT ( _text16_late_lma ) {
+	_text16_late = .;
 	*(.text16)
 	*(.text16.*)
 	_mtext16 = .;
     } .bss.text16 (NOLOAD) : AT ( _end_lma ) {
 	_etext16 = .;
     }
-    _text16_filesz	= ABSOLUTE ( _mtext16 - _text16 );
+    _text16_early_filesz = ABSOLUTE ( _etext16_early - _text16 );
+    _text16_early_memsz	= ABSOLUTE ( _etext16_early - _text16 );
+    _text16_late_filesz	= ABSOLUTE ( _mtext16 - _text16_late );
+    _text16_late_memsz	= ABSOLUTE ( _etext16 - _text16_late );
     _text16_memsz	= ABSOLUTE ( _etext16 - _text16 );
 
     /*
@@ -168,10 +176,14 @@ SECTIONS {
     _prefix_lma		= .;
     .			+= _prefix_filesz;
 
+    .			= ALIGN ( _max_align );
+    _text16_early_lma	= .;
+    .			+= _text16_early_filesz;
+
     .			= ALIGN ( _max_align );
     _payload_lma	= .;
-    _text16_lma		= .;
-    .			+= _text16_filesz;
+    _text16_late_lma	= .;
+    .			+= _text16_late_filesz;
 
     .			= ALIGN ( _max_align );
     _data16_lma		= .;
@@ -194,8 +206,6 @@ SECTIONS {
      * Values calculated to save code from doing it
      *
      */
-    _prefix_memsz_pgh	= ( ( _prefix_memsz + 15 ) / 16 );
-    _prefix_memsz_sect	= ( ( _prefix_memsz + 511 ) / 512 );
     _text16_memsz_pgh	= ( ( _text16_memsz + 15 ) / 16 );
     _data16_memsz_pgh	= ( ( _data16_memsz + 15 ) / 16 );
 }

src/util/zbin.c

@@ -16,6 +16,7 @@ struct input_file {
 struct output_file {
 	void *buf;
 	size_t len;
+	size_t hdr_len;
 	size_t max_len;
 };
 
@@ -38,6 +39,13 @@ struct zinfo_pack {
 	uint32_t align;
 };
 
+struct zinfo_payload {
+	char type[4];
+	uint32_t pad1;
+	uint32_t pad2;
+	uint32_t align;
+};
+
 struct zinfo_add {
 	char type[4];
 	uint32_t offset;
@@ -49,6 +57,7 @@ union zinfo_record {
 	struct zinfo_common common;
 	struct zinfo_copy copy;
 	struct zinfo_pack pack;
+	struct zinfo_payload payload;
 	struct zinfo_add add;
 };
 
@@ -209,8 +218,22 @@ static int process_zinfo_pack ( struct input_file *input,
 	return 0;
 }
 
+static int process_zinfo_payl ( struct input_file *input,
+				struct output_file *output,
+				union zinfo_record *zinfo ) {
+	struct zinfo_payload *payload = &zinfo->payload;
+
+	output->len = align ( output->len, payload->align );
+	output->hdr_len = output->len;
+
+	if ( DEBUG ) {
+		fprintf ( stderr, "PAYL at %#zx\n", output->hdr_len );
+	}
+}
+
 static int process_zinfo_add ( struct input_file *input,
 			       struct output_file *output,
+			       size_t len,
 			       struct zinfo_add *add,
 			       size_t datasize ) {
 	size_t offset = add->offset;
@@ -227,7 +250,7 @@ static int process_zinfo_add ( struct input_file *input,
 	}
 
 	target = ( output->buf + offset );
-	size = ( align ( output->len, add->divisor ) / add->divisor );
+	size = ( align ( len, add->divisor ) / add->divisor );
 
 	switch ( datasize ) {
 	case 1:
@@ -283,7 +306,7 @@ static int process_zinfo_add ( struct input_file *input,
 		fprintf ( stderr, "ADDx [%#zx,%#zx) (%s%#x+(%#zx/%#x)) = "
 			  "%#lx\n", offset, ( offset + datasize ),
 			  ( ( addend < 0 ) ? "-" : "" ), abs ( addend ),
-			  output->len, add->divisor, val );
+			  len, add->divisor, val );
 	}
 
 	return 0;
@@ -292,19 +315,43 @@ static int process_zinfo_add ( struct input_file *input,
 static int process_zinfo_addb ( struct input_file *input,
 				struct output_file *output,
 				union zinfo_record *zinfo ) {
-	return process_zinfo_add ( input, output, &zinfo->add, 1 );
+	return process_zinfo_add ( input, output, output->len,
+				   &zinfo->add, 1 );
 }
 
 static int process_zinfo_addw ( struct input_file *input,
 				struct output_file *output,
 				union zinfo_record *zinfo ) {
-	return process_zinfo_add ( input, output, &zinfo->add, 2 );
+	return process_zinfo_add ( input, output, output->len,
+				   &zinfo->add, 2 );
 }
 
 static int process_zinfo_addl ( struct input_file *input,
 				struct output_file *output,
 				union zinfo_record *zinfo ) {
-	return process_zinfo_add ( input, output, &zinfo->add, 4 );
+	return process_zinfo_add ( input, output, output->len,
+				   &zinfo->add, 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 );
+}
+
+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 );
+}
+
+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 );
 }
 
 struct zinfo_processor {
@@ -317,9 +364,13 @@ struct zinfo_processor {
 static struct zinfo_processor zinfo_processors[] = {
 	{ "COPY", process_zinfo_copy },
 	{ "PACK", process_zinfo_pack },
+	{ "PAYL", process_zinfo_payl },
 	{ "ADDB", process_zinfo_addb },
 	{ "ADDW", process_zinfo_addw },
 	{ "ADDL", process_zinfo_addl },
+	{ "ADHB", process_zinfo_adhb },
+	{ "ADHW", process_zinfo_adhw },
+	{ "ADHL", process_zinfo_adhl },
 };
 
 static int process_zinfo ( struct input_file *input,