Compressed ROM images now work.

9 files changed, 439 insertions, 11 deletions

diff --git a/src/Makefile b/src/Makefile
index 4f50456a..d3280702 100644
--- a/src/Makefile
+++ b/src/Makefile
@@ -90,6 +90,7 @@ MKCONFIG	?= $(PERL) ./util/mkconfig.pl
 SYMCHECK	?= $(PERL) ./util/symcheck.pl
 SORTOBJDUMP	?= $(PERL) ./util/sortobjdump.pl
 NRV2B		?= ./util/nrv2b
+ZBIN		?= ./util/zbin
 DOXYGEN		?= doxygen
 
 # Location to place generated files
diff --git a/src/Makefile.housekeeping b/src/Makefile.housekeeping
index 12041ce6..ff52de1d 100644
--- a/src/Makefile.housekeeping
+++ b/src/Makefile.housekeeping
@@ -307,11 +307,23 @@ $(BIN)/%.tmp : $(BLIB) $(MAKEDEPS) $(LDSCRIPT)
 $(BIN)/%.map : $(BIN)/%.tmp
 	@less $(BIN)/$*.tmp.map
 
+# Extract compression information from intermediate object file
+#
+$(BIN)/%.zinfo : $(BIN)/%.tmp
+	$(QM)echo "  [ZINFO] $@"
+	$(Q)$(OBJCOPY) -O binary -j .zinfo $< $@
+
 # Build raw binary file from intermediate object file
 #
 $(BIN)/%.bin : $(BIN)/%.tmp
 	$(QM)echo "  [BIN] $@"
-	$(Q)$(OBJCOPY) -O binary $< $@
+	$(Q)$(OBJCOPY) -O binary -R .zinfo $< $@
+
+# Compress raw binary file
+#
+$(BIN)/%.zbin : $(BIN)/%.bin $(BIN)/%.zinfo $(ZBIN)
+	$(QM)echo "  [ZBIN] $@"
+	$(Q)$(ZBIN) $(BIN)/$*.bin $(BIN)/$*.zinfo > $@
 
 # Build bochs symbol table
 $(BIN)/%.bxs : $(BIN)/%.tmp
@@ -356,7 +368,7 @@ define media_template
 	@$(MKDIR) -p $(dir $(2))
 	@$(RM) $(2)
 	@$(TOUCH) $(2)
-	@echo -e '$$(BIN)/%$(1) : $$(BIN)/%$(1).bin' \
+	@echo -e '$$(BIN)/%$(1) : $$(BIN)/%$(1).zbin' \
 		  '\n\t$$(QM)echo "  [FINISH] $$@"' \
 		  '\n\t$$(Q)$$(CP) $$< $$@' \
 		  '\n\t$$(Q)$$(FINALISE_$(1))' \
@@ -383,13 +395,19 @@ include $(MEDIA_DEPS)
 allroms allzroms : all%s : $(foreach ROM,$(ROMS),$(BIN)/$(ROM).%)
 all%s : $(foreach DRIVER,$(DRIVERS),$(BIN)/$(DRIVER).%)
 
-# The compressor utility
+# The compression utilities
 #
 $(NRV2B) : util/nrv2b.c $(MAKEDEPS)
-	$(HOST_CC) -O2 -DENCODE -DDECODE -DMAIN -DVERBOSE -DNDEBUG \
+	$(QM)echo "  [HOSTCC] $@"
+	$(Q)$(HOST_CC) -O2 -DENCODE -DDECODE -DMAIN -DVERBOSE -DNDEBUG \
 		       -DBITSIZE=32 -DENDIAN=0 -o $@ $<
 CLEANUP	+= $(NRV2B)
 
+$(ZBIN) : util/zbin.c util/nrv2b.c $(MAKEDEPS)
+	$(QM)echo "  [HOSTCC] $@"
+	$(Q)$(HOST_CC) -O2 -o $@ $<
+CLEANUP += $(ZBIN)
+
 # Auto-incrementing build serial number.  Append "bs" to your list of
 # build targets to get a serial number printed at the end of the
 # build.  Enable -DBUILD_SERIAL in order to see it when the code runs.
diff --git a/src/arch/i386/prefix/libprefix.S b/src/arch/i386/prefix/libprefix.S
index d1554513..0966dcbd 100644
--- a/src/arch/i386/prefix/libprefix.S
+++ b/src/arch/i386/prefix/libprefix.S
@@ -37,6 +37,9 @@
  */
 #define HIGHMEM_LOADPOINT ( 4 << 20 )
 
+/* Image compression enabled */
+#define COMPRESS 1
+
 #define CR0_PE 1
 
 	.arch i386
@@ -81,7 +84,11 @@ install_block:
 	
 	/* Do the copy */
 	cld
-	rep addr32 movsb /* or "call decompress16" */
+#if COMPRESS
+	call	decompress16
+#else
+	call	nodecompress16
+#endif
 
 	/* Zero remaining space */
 	movl	%eax, %edi
diff --git a/src/arch/i386/prefix/nonrv2b.S b/src/arch/i386/prefix/nonrv2b.S
new file mode 100644
index 00000000..d40396b3
--- /dev/null
+++ b/src/arch/i386/prefix/nonrv2b.S
@@ -0,0 +1,18 @@
+/* Placeholder for decompress16 in non-compressed images */
+
+	.text
+	.arch i386
+	.section ".prefix.lib", "ax", @progbits
+
+	.code16
+	.globl	nodecompress16
+nodecompress16:
+	rep addr32 movsb
+	ret
+
+	/* File split information for the compressor */
+	.section ".zinfo", "a"
+	.ascii	"COPY"
+	.long	_prefix_load_offset
+	.long	_load_size
+	.long	_max_align
diff --git a/src/arch/i386/prefix/romprefix.S b/src/arch/i386/prefix/romprefix.S
index f68c14e0..3e6cd2d0 100644
--- a/src/arch/i386/prefix/romprefix.S
+++ b/src/arch/i386/prefix/romprefix.S
@@ -14,7 +14,7 @@
 	.org	0x00
 romheader:
 	.word	0xAA55			/* BIOS extension signature */
-	.byte	_rom_size		/* Size in 512-byte blocks */
+romheader_size:	.byte	_rom_size	/* Size in 512-byte blocks */
 	jmp	init_vector		/* Initialisation vector */
 	.org	0x16
 	.word	undiheader
@@ -34,7 +34,7 @@ pciheader:
 	.byte	0x02			/* Device Base Type code */
 	.byte	0x00			/* Device Sub-Type code */
 	.byte	0x00			/* Device Interface Type code */
-	.word	_rom_size		/* Image length same as offset 02h */
+pciheader_size:	.word	_rom_size	/* Image length same as offset 02h */
 	.word	0x0001			/* revision level of code/data */
 	.byte	0x00			/* code type */
 	.byte	0x80			/* Flags (last PCI data structure) */
@@ -215,3 +215,15 @@ print_message:
 	popw	%ax
 	ret
 	.size print_message, . - print_message
+
+
+	/* Data update information for the compressor */
+	.section ".zinfo.fixup", "a"
+	.ascii	"SUBB"
+	.long	romheader_size
+	.long	512
+	.long	0
+	.ascii	"SUBW"
+	.long	pciheader_size
+	.long	512
+	.long	0
diff --git a/src/arch/i386/prefix/unnrv2b.S b/src/arch/i386/prefix/unnrv2b.S
index 4e3090e2..1cf570d0 100644
--- a/src/arch/i386/prefix/unnrv2b.S
+++ b/src/arch/i386/prefix/unnrv2b.S
@@ -35,7 +35,7 @@
 
 	.text
 	.arch i386
-	.section ".prefix", "ax", @progbits
+	.section ".prefix.lib", "ax", @progbits
 
 #ifdef CODE16
 /****************************************************************************
@@ -54,6 +54,7 @@
  * NOTE: It would be possible to build a smaller version of the
  * decompression code for -DKEEP_IT_REAL by using
  *    #define REG(x) x
+ *    #define MOVSB movsb
  * to use 16-bit registers where possible.  This would impose limits
  * that the compressed data size must be in the range [1,65533-%si]
  * and the uncompressed data size must be in the range [1,65536-%di]
@@ -66,6 +67,7 @@
  */
 
 #define REG(x) e ## x
+#define MOVSB addr32 movsb
 
 	.code16
 	.globl	decompress16
@@ -109,11 +111,10 @@ decompress:
 	cld
 	xor	%xBP, %xBP
 	dec	%xBP		/* last_m_off = -1 */
-	add	$4, %xSI	/* Skip "file length" field */
 	jmp	dcl1_n2b
 	
 decompr_literals_n2b:
-	movsb
+	MOVSB
 decompr_loop_n2b:
 	addl	%ebx, %ebx
 	jnz	dcl2_n2b
@@ -157,7 +158,7 @@ decompr_got_mlen_n2b:
 	push	%xSI
 	lea	(%xBP,%xDI), %xSI	/* m_pos = dst + olen + -m_off  */
 	rep
-	es movsb		/* dst[olen++] = *m_pos++ while(m_len > 0) */
+	es MOVSB		/* dst[olen++] = *m_pos++ while(m_len > 0) */
 	pop	%xSI
 	jmp	decompr_loop_n2b
 
@@ -179,3 +180,23 @@ decompr_end_n2b:
 	popl	%ebx
 	pop	%xAX
 	ret
+
+
+	/* File split information for the compressor */
+	.section ".zinfo", "a"
+	.ascii	"COPY"
+	.long	_prefix_load_offset
+	.long	_prefix_progbits_size
+	.long	_max_align
+	.ascii	"PACK"
+	.long	_text16_load_offset
+	.long	_text16_progbits_size
+	.long	_max_align
+	.ascii	"PACK"
+	.long	_data16_load_offset
+	.long	_data16_progbits_size
+	.long	_max_align
+	.ascii	"PACK"
+	.long	_textdata_load_offset
+	.long	_textdata_progbits_size
+	.long	_max_align
diff --git a/src/arch/i386/scripts/i386.lds b/src/arch/i386/scripts/i386.lds
index 7b7b95da..8ff7c45a 100644
--- a/src/arch/i386/scripts/i386.lds
+++ b/src/arch/i386/scripts/i386.lds
@@ -165,6 +165,24 @@ SECTIONS {
     _end = .;
 
     /*
+     * Compressor information block
+     */
+
+    _zinfo_link_addr = 0;
+    . = _zinfo_link_addr;
+    _zinfo = .;
+
+    .zinfo : AT ( _zinfo_load_offset + __zinfo ) {
+	__zinfo = .;
+	_entry = .;
+	*(.zinfo)
+	*(.zinfo.*)
+	_ezinfo_progbits = .;
+    }
+    
+    _ezinfo = .;
+
+    /*
      * Dispose of the comment and note sections to make the link map
      * easier to read
      */
@@ -215,6 +233,13 @@ SECTIONS {
 
     _load_size		    = . - _load_addr;
 
+    .			   -= _zinfo_link_addr;
+    _zinfo_load_offset	    = ALIGN ( _max_align );
+    _zinfo_load_addr	    = _zinfo_link_addr + _zinfo_load_offset;
+    _zinfo_size		    = _ezinfo - _zinfo;
+    _zinfo_progbits_size    = _ezinfo_progbits - _zinfo;
+    .			    = _zinfo_load_addr + _zinfo_progbits_size;
+
     _payload_offset	    = _text16_load_offset;
 
     /*
diff --git a/src/util/.gitignore b/src/util/.gitignore
index 9550102a..98adc2df 100644
--- a/src/util/.gitignore
+++ b/src/util/.gitignore
@@ -1,3 +1,4 @@
 nrv2b
+zbin
 hijack
 prototester
diff --git a/src/util/zbin.c b/src/util/zbin.c
new file mode 100644
index 00000000..f47fa36b
--- /dev/null
+++ b/src/util/zbin.c
@@ -0,0 +1,325 @@
+#include <stdio.h>
+#include <sys/stat.h>
+
+#define ENCODE
+#define VERBOSE
+#include "nrv2b.c"
+FILE *infile, *outfile;
+
+struct input_file {
+	void *buf;
+	size_t len;
+};
+
+struct output_file {
+	void *buf;
+	size_t len;
+	size_t max_len;
+};
+
+struct zinfo_common {
+	char type[4];
+	char pad[12];
+};
+
+struct zinfo_copy {
+	char type[4];
+	uint32_t offset;
+	uint32_t len;
+	uint32_t align;
+};
+
+struct zinfo_pack {
+	char type[4];
+	uint32_t offset;
+	uint32_t len;
+	uint32_t align;
+};
+
+struct zinfo_subtract {
+	char type[4];
+	uint32_t offset;
+	uint32_t divisor;
+	uint32_t pad;
+};
+
+union zinfo_record {
+	struct zinfo_common common;
+	struct zinfo_copy copy;
+	struct zinfo_pack pack;
+	struct zinfo_subtract subtract;
+};
+
+struct zinfo_file {
+	union zinfo_record *zinfo;
+	unsigned int num_entries;
+};
+
+static int read_file ( const char *filename, void **buf, size_t *len ) {
+	FILE *file;
+	struct stat stat;
+
+	file = fopen ( filename, "r" );
+	if ( ! file ) {
+		fprintf ( stderr, "Could not open %s: %s\n", filename,
+			  strerror ( errno ) );
+		goto err;
+	}
+
+	if ( fstat ( fileno ( file ), &stat ) < 0 ) {
+		fprintf ( stderr, "Could not stat %s: %s\n", filename,
+			  strerror ( errno ) );
+		goto err;
+	}
+
+	*len = stat.st_size;
+	*buf = malloc ( *len );
+	if ( ! *buf ) {
+		fprintf ( stderr, "Could not malloc() %d bytes for %s: %s\n",
+			  *len, filename, strerror ( errno ) );
+		goto err;
+	}
+
+	if ( fread ( *buf, 1, *len, file ) != *len ) {
+		fprintf ( stderr, "Could not read %d bytes from %s: %s\n",
+			  *len, filename, strerror ( errno ) );
+		goto err;
+	}
+
+	fclose ( file );
+	return 0;
+
+ err:
+	fclose ( file );
+	return -1;
+}
+
+static int read_input_file ( const char *filename,
+			     struct input_file *input ) {
+	return read_file ( filename, &input->buf, &input->len );
+}
+
+static int read_zinfo_file ( const char *filename,
+			     struct zinfo_file *zinfo ) {
+	void *buf;
+	size_t len;
+
+	if ( read_file ( filename, &buf, &len ) < 0 )
+		return -1;
+
+	if ( ( len % sizeof ( *(zinfo->zinfo) ) ) != 0 ) {
+		fprintf ( stderr, ".zinfo file %s has invalid length %d\n",
+			  filename, len );
+		return -1;
+	}
+
+	zinfo->zinfo = buf;
+	zinfo->num_entries = ( len / sizeof ( *(zinfo->zinfo) ) );
+	return 0;
+}
+
+static int alloc_output_file ( size_t max_len, struct output_file *output ) {
+	output->len = 0;
+	output->max_len = ( max_len );
+	output->buf = malloc ( max_len );
+	if ( ! output->buf ) {
+		fprintf ( stderr, "Could not allocate %d bytes for output\n",
+			  max_len );
+		return -1;
+	}
+	memset ( output->buf, 0xff, sizeof ( output->buf ) );
+	return 0;
+}
+
+static int process_zinfo_copy ( struct input_file *input,
+				struct output_file *output,
+				union zinfo_record *zinfo ) {
+	struct zinfo_copy *copy = &zinfo->copy;
+	size_t offset = copy->offset;
+	size_t len = copy->len;
+	unsigned int align = copy->align;
+
+	if ( ( offset + len ) > input->len ) {
+		fprintf ( stderr, "Input buffer overrun on copy\n" );
+		return -1;
+	}
+
+	output->len = ( ( output->len + align - 1 ) & ~( align - 1 ) );
+	if ( ( output->len + len ) > output->max_len ) {
+		fprintf ( stderr, "Output buffer overrun on copy\n" );
+		return -1;
+	}
+
+	memcpy ( ( output->buf + output->len ),
+		 ( input->buf + offset ), len );
+	output->len += len;
+	return 0;
+}
+
+static int process_zinfo_pack ( struct input_file *input,
+				struct output_file *output,
+				union zinfo_record *zinfo ) {
+	struct zinfo_pack *pack = &zinfo->pack;
+	size_t offset = pack->offset;
+	size_t len = pack->len;
+	unsigned int align = pack->align;
+	unsigned long packed_len;
+
+	if ( ( offset + len ) > input->len ) {
+		fprintf ( stderr, "Input buffer overrun on pack\n" );
+		return -1;
+	}
+
+	output->len = ( ( output->len + align - 1 ) & ~( align - 1 ) );
+	if ( output->len > output->max_len ) {
+		fprintf ( stderr, "Output buffer overrun on pack\n" );
+		return -1;
+	}
+
+	if ( ucl_nrv2b_99_compress ( ( input->buf + offset ), len,
+				     ( output->buf + output->len ),
+				     &packed_len, 0 ) != UCL_E_OK ) {
+		fprintf ( stderr, "Compression failure\n" );
+		return -1;
+	}
+
+	output->len += packed_len;
+	if ( output->len > output->max_len ) {
+		fprintf ( stderr, "Output buffer overrun on pack\n" );
+		return -1;
+	}
+
+	return 0;
+}
+
+static int process_zinfo_subtract ( struct input_file *input,
+				    struct output_file *output,
+				    struct zinfo_subtract *subtract,
+				    size_t datasize ) {
+	size_t offset = subtract->offset;
+	void *target;
+	long delta;
+
+	if ( ( offset + datasize ) > output->len ) {
+		fprintf ( stderr, "Subtract at %#zx outside output buffer\n",
+			  offset );
+		return -1;
+	}
+
+	target = ( output->buf + offset );
+	delta = ( ( output->len / subtract->divisor ) -
+		  ( input->len / subtract->divisor ) );
+
+	switch ( datasize ) {
+	case 1: {
+		uint8_t *byte = target;
+		*byte += delta;
+		break; }
+	case 2: {
+		uint16_t *word = target;
+		*word += delta;
+		break; }
+	case 4: {
+		uint32_t *dword = target;
+		*dword += delta;
+		break; }
+	default:
+		fprintf ( stderr, "Unsupported subtract datasize %d\n",
+			  datasize );
+		return -1;
+	}
+	return 0;
+}
+
+static int process_zinfo_subb ( struct input_file *input,
+				struct output_file *output,
+				union zinfo_record *zinfo ) {
+	return process_zinfo_subtract ( input, output, &zinfo->subtract, 1 );
+}
+
+static int process_zinfo_subw ( struct input_file *input,
+				struct output_file *output,
+				union zinfo_record *zinfo ) {
+	return process_zinfo_subtract ( input, output, &zinfo->subtract, 2 );
+}
+
+static int process_zinfo_subl ( struct input_file *input,
+				struct output_file *output,
+				union zinfo_record *zinfo ) {
+	return process_zinfo_subtract ( input, output, &zinfo->subtract, 4 );
+}
+
+struct zinfo_processor {
+	char *type;
+	int ( * process ) ( struct input_file *input,
+			    struct output_file *output,
+			    union zinfo_record *zinfo );
+};
+
+static struct zinfo_processor zinfo_processors[] = {
+	{ "COPY", process_zinfo_copy },
+	{ "PACK", process_zinfo_pack },
+	{ "SUBB", process_zinfo_subb },
+	{ "SUBW", process_zinfo_subw },
+	{ "SUBL", process_zinfo_subl },
+};
+
+static int process_zinfo ( struct input_file *input,
+			   struct output_file *output,
+			   union zinfo_record *zinfo ) {
+	struct zinfo_common *common = &zinfo->common;
+	struct zinfo_processor *processor;
+	char type[ sizeof ( common->type ) + 1 ] = "";
+	unsigned int i;
+
+	strncat ( type, common->type, sizeof ( type ) - 1 );
+	for ( i = 0 ; i < ( sizeof ( zinfo_processors ) /
+			    sizeof ( zinfo_processors[0] ) ) ; i++ ) {
+		processor = &zinfo_processors[i];
+		if ( strcmp ( processor->type, type ) == 0 )
+			return processor->process ( input, output, zinfo );
+	}
+
+	fprintf ( stderr, "Unknown zinfo record type \"%s\"\n", &type[0] );
+	return -1;
+}
+
+static int write_output_file ( struct output_file *output ) {
+	if ( fwrite ( output->buf, 1, output->len, stdout ) != output->len ) {
+		fprintf ( stderr, "Could not write %d bytes of output: %s\n",
+			  output->len, strerror ( errno ) );
+		return -1;
+	}
+	return 0;
+}
+
+int main ( int argc, char **argv ) {
+	struct input_file input;
+	struct output_file output;
+	struct zinfo_file zinfo;
+	unsigned int i;
+
+	if ( argc != 3 ) {
+		fprintf ( stderr, "Syntax: %s file.bin file.zinfo "
+			  "> file.zbin\n", argv[0] );
+		exit ( 1 );
+	}
+
+	if ( read_input_file ( argv[1], &input ) < 0 )
+		exit ( 1 );
+	if ( read_zinfo_file ( argv[2], &zinfo ) < 0 )
+		exit ( 1 );
+	if ( alloc_output_file ( ( input.len * 4 ), &output ) < 0 )
+		exit ( 1 );
+
+	for ( i = 0 ; i < zinfo.num_entries ; i++ ) {
+		if ( process_zinfo ( &input, &output,
+				     &zinfo.zinfo[i] ) < 0 )
+			exit ( 1 );
+	}
+
+	if ( write_output_file ( &output ) < 0 )
+		exit ( 1 );
+
+	return 0;
+}