diff --git a/ChangeLog b/ChangeLog
index 7fd12f963..e531142ec 100644
--- a/ChangeLog
+++ b/ChangeLog
@@ -1,3 +1,33 @@
+2002-12-28  Yoshinori K. Okuji  <okuji@enbug.org>
+
+	Use -mrtd and -mregparm=3 to reduce the generated code sizes.
+	This means that any missing prototypes could be fatal. Also, you
+	must take care when writing assembly code. See the comments at
+	the beginning of startup.S, for more details.
+	
+	* kern/i386/pc/startup.S (pupa_halt): Modified for the new
+	compilation mechanism.
+	(pupa_chainloader_real_boot): Likewise.
+	(pupa_biosdisk_rw_int13_extensions): Likewise.
+	(pupa_biosdisk_rw_standard): Likewise.
+	(pupa_biosdisk_check_int13_extensions): Likewise.
+	(pupa_biosdisk_get_diskinfo_int13_extensions): Likewise.
+	(pupa_biosdisk_get_diskinfo_standard): Likewise.
+	(pupa_get_memsize): Likewise.
+	(pupa_get_mmap_entry): Likewise.
+	(pupa_console_putchar): Likewise.
+	(pupa_console_setcursor): Likewise.
+	(pupa_getrtsecs): Use pushl instead of push.
+
+	* kern/i386/pc/init.c (pupa_machine_init): Use the scratch
+	memory instead of the stack for a mmap entry, because some
+	BIOSes may ignore the maximum size and overflow.
+
+	* conf/i386-pc.rmk (COMMON_CFLAGS): Added -mrtd and -mregparm=3.
+
+	* genmk.rb (PModule#rule): Compile automatically generated
+	sources with module-specific CFLAGS as well as other sources.
+
 2002-12-27  Yoshinori K. Okuji  <okuji@enbug.org>
 
 	* configure.ac: Check ld.
diff --git a/NEWS b/NEWS
index 032e2dd75..e026778a4 100644
--- a/NEWS
+++ b/NEWS
@@ -2,6 +2,11 @@ New in 0.7:
 
 * Problems in cross-compiling PUPA are fixed.
 
+* Use -mrtd and -mregparm=3 to reduce the generated code sizes.	This
+  means that any missing prototypes could be fatal. Also, you must take
+  care when writing assembly code. See the comments at the beginning of
+  startup.S, for more details.
+
 New in 0.6 - 2002-12-27, Yoshinori K. Okuji:
 
 * The chainloader and the FAT filesystem are modularized.
diff --git a/conf/i386-pc.mk b/conf/i386-pc.mk
index ee4880b77..0f0575457 100644
--- a/conf/i386-pc.mk
+++ b/conf/i386-pc.mk
@@ -1,7 +1,7 @@
 # -*- makefile -*-
 
 COMMON_ASFLAGS = -nostdinc -fno-builtin
-COMMON_CFLAGS = -fno-builtin
+COMMON_CFLAGS = -fno-builtin -mrtd -mregparm=3
 
 # Images.
 pkgdata_IMAGES = boot.img diskboot.img kernel.img
@@ -312,7 +312,7 @@ pre-chain.o: chain_mod-loader_i386_pc_chainloader.o
 	$(LD) -r -o $@ $^
 
 mod-chain.o: mod-chain.c
-	$(CC) $(CPPFLAGS) $(CFLAGS) -c -o $@ $<
+	$(CC) $(CPPFLAGS) $(CFLAGS) $(chain_mod_CFLAGS) -c -o $@ $<
 
 mod-chain.c: moddep.lst genmodsrc.sh
 	sh $(srcdir)/genmodsrc.sh 'chain' $< > $@ || (rm -f $@; exit 1)
@@ -351,7 +351,7 @@ pre-fat.o: fat_mod-fs_fat.o
 	$(LD) -r -o $@ $^
 
 mod-fat.o: mod-fat.c
-	$(CC) $(CPPFLAGS) $(CFLAGS) -c -o $@ $<
+	$(CC) $(CPPFLAGS) $(CFLAGS) $(fat_mod_CFLAGS) -c -o $@ $<
 
 mod-fat.c: moddep.lst genmodsrc.sh
 	sh $(srcdir)/genmodsrc.sh 'fat' $< > $@ || (rm -f $@; exit 1)
diff --git a/conf/i386-pc.rmk b/conf/i386-pc.rmk
index 9f72c7e46..b5a503914 100644
--- a/conf/i386-pc.rmk
+++ b/conf/i386-pc.rmk
@@ -1,7 +1,7 @@
 # -*- makefile -*-
 
 COMMON_ASFLAGS = -nostdinc -fno-builtin
-COMMON_CFLAGS = -fno-builtin
+COMMON_CFLAGS = -fno-builtin -mrtd -mregparm=3
 
 # Images.
 pkgdata_IMAGES = boot.img diskboot.img kernel.img
diff --git a/genmk.rb b/genmk.rb
index 7c621b970..96516e7ba 100644
--- a/genmk.rb
+++ b/genmk.rb
@@ -125,7 +125,7 @@ UNDSYMFILES += #{undsym}
 	$(LD) -r -o $@ $^
 
 #{mod_obj}: #{mod_src}
-	$(CC) $(CPPFLAGS) $(CFLAGS) -c -o $@ $<
+	$(CC) $(CPPFLAGS) $(CFLAGS) $(#{prefix}_CFLAGS) -c -o $@ $<
 
 #{mod_src}: moddep.lst genmodsrc.sh
 	sh $(srcdir)/genmodsrc.sh '#{mod_name}' $< > $@ || (rm -f $@; exit 1)
diff --git a/kern/i386/pc/init.c b/kern/i386/pc/init.c
index 589696e69..012ec99a4 100644
--- a/kern/i386/pc/init.c
+++ b/kern/i386/pc/init.c
@@ -30,7 +30,8 @@ void
 pupa_machine_init (void)
 {
   pupa_uint32_t cont;
-  struct pupa_machine_mmap_entry entry;
+  struct pupa_machine_mmap_entry *entry
+    = (struct pupa_machine_mmap_entry *) PUPA_MEMORY_MACHINE_SCRATCH_ADDR;
   pupa_size_t lower_mem = (pupa_get_memsize (0) << 10);
   pupa_addr_t end_addr = pupa_get_end_addr ();
 
@@ -53,31 +54,31 @@ pupa_machine_init (void)
 		       PUPA_MEMORY_MACHINE_RESERVED_START - end_addr);
 
   /* Check if pupa_get_mmap_entry works.  */
-  cont = pupa_get_mmap_entry (&entry, 0);
+  cont = pupa_get_mmap_entry (entry, 0);
 
-  if (entry.size)
+  if (entry->size)
     do
       {
 	/* Avoid the lower memory.  */
-	if (entry.addr < 0x100000)
+	if (entry->addr < 0x100000)
 	  {
-	    if (entry.len <= 0x100000 - entry.addr)
+	    if (entry->len <= 0x100000 - entry->addr)
 	      goto next;
 
-	    entry.len -= 0x100000 - entry.addr;
-	    entry.addr = 0x100000;
+	    entry->len -= 0x100000 - entry->addr;
+	    entry->addr = 0x100000;
 	  }
 	
 	/* Ignore >4GB.  */
-	if (entry.addr <= 0xFFFFFFFF && entry.type == 1)
+	if (entry->addr <= 0xFFFFFFFF && entry->type == 1)
 	  {
 	    pupa_addr_t addr;
 	    pupa_size_t len;
 
-	    addr = (pupa_addr_t) entry.addr;
-	    len = ((addr + entry.len > 0xFFFFFFFF)
+	    addr = (pupa_addr_t) entry->addr;
+	    len = ((addr + entry->len > 0xFFFFFFFF)
 		   ? 0xFFFFFFFF - addr
-		   : (pupa_size_t) entry.len);
+		   : (pupa_size_t) entry->len);
 	    pupa_mm_init_region ((void *) addr, len);
 	  }
 	
@@ -85,9 +86,9 @@ pupa_machine_init (void)
 	if (! cont)
 	  break;
 	
-	cont = pupa_get_mmap_entry (&entry, cont);
+	cont = pupa_get_mmap_entry (entry, cont);
       }
-    while (entry.size);
+    while (entry->size);
   else
     {
       pupa_uint32_t eisa_mmap = pupa_get_eisa_mmap ();
diff --git a/kern/i386/pc/startup.S b/kern/i386/pc/startup.S
index ecc97f3f7..aaabdc307 100644
--- a/kern/i386/pc/startup.S
+++ b/kern/i386/pc/startup.S
@@ -35,6 +35,14 @@
 {  1, 1, 1, 0, 0, 0, 0, 1, 1,  1,  1,  1,  1,  1,  1,  1,  1 }
  */
 
+/*
+ * Note: PUPA is compiled with the options -mrtd and -mregparm=3.
+ *       So the first three arguments are passed in %eax, %edx, and %ecx,
+ *       respectively, and if a function has a fixed number of arguments
+ *       and the number if greater than three, the function must return
+ *       with "ret $N" where N is ((the number of arguments) - 3) * 4.
+ */
+
 #include <config.h>
 #include <pupa/symbol.h>
 #include <pupa/boot.h>
@@ -203,9 +211,6 @@ FUNCTION(pupa_reboot)
  * APM even if it is available.
  */
 FUNCTION(pupa_halt)
-	/* get the argument */
-	movl	4(%esp), %eax
-	
 	/* see if zero */
 	testl	%eax, %eax
 	jnz	EXT_C(pupa_stop)
@@ -256,16 +261,16 @@ FUNCTION(pupa_halt)
  */
 
 FUNCTION(pupa_chainloader_real_boot)
-	/* no need to save anything, just use %esp */
+	/* no need to save anything */
 
 	/* ESI must point to a partition table entry */
-	movl	8(%esp), %esi
+	movl	%edx, %esi
 
 	/* set up to pass boot drive */
-	movl	4(%esp), %edx
+	movl	%eax, %edx
 
 	/* Turn off Gate A20 */
-	pushl	$0
+	xorl	%eax, %eax
 	call	EXT_C(pupa_gate_a20)
 	
 	call	prot_to_real
@@ -398,21 +403,15 @@ realcseg:
 
 FUNCTION(pupa_biosdisk_rw_int13_extensions)
 	pushl	%ebp
-	movl	%esp, %ebp
-
 	pushl	%esi
 
 	/* compute the address of disk_address_packet */
-	movl	0x10(%ebp), %eax
-	movw	%ax, %si
-	xorw	%ax, %ax
-	shrl	$4, %eax
-	movw	%ax, %cx	/* save the segment to cx */
+	movw	%cx, %si
+	xorw	%cx, %cx
+	shrl	$4, %ecx	/* save the segment to cx */
 
-	/* drive */
-	movb	0xc(%ebp), %dl
 	/* ah */
-	movb	0x8(%ebp), %dh
+	movb	%al, %dh
 	/* enter real mode */
 	call	prot_to_real
 	
@@ -453,21 +452,24 @@ FUNCTION(pupa_biosdisk_rw_standard)
 	pushl	%esi
 
 	/* set up CHS information */
-	movl	0x10(%ebp), %eax
-	movb	%al, %ch
-	movb	0x18(%ebp), %al
-	shlb	$2, %al
-	shrw	$2, %ax
-	movb	%al, %cl
-	movb	0x14(%ebp), %dh
-	/* drive */
-	movb	0xc(%ebp), %dl
-	/* segment */
-	movw	0x20(%ebp), %bx
-	/* save nsec and ah to %di */
-	movb	0x8(%ebp), %ah
-	movb	0x1c(%ebp), %al
+
+	/* set %ch to low eight bits of cylinder */
+	xchgb	%cl, %ch
+	/* set bits 6-7 of %cl to high two bits of cylinder */
+	shlb	$6, %cl
+	/* set bits 0-5 of %cl to sector */
+	addb	0xc(%ebp), %cl
+	/* set %dh to head */
+	movb	0x8(%ebp), %dh
+	/* set %ah to AH */
+	movb	%al, %ah
+	/* set %al to NSEC */
+	movb	0x10(%ebp), %al
+	/* save %ax in %di */
 	movw	%ax, %di
+	/* save SEGMENT in %bx */
+	movw	0x14(%ebp), %bx
+		
 	/* enter real mode */
 	call	prot_to_real
 
@@ -503,7 +505,7 @@ FUNCTION(pupa_biosdisk_rw_standard)
 	popl	%ebx
 	popl	%ebp
 
-	ret
+	ret	$(4 * 4)
 
 
 /*
@@ -515,12 +517,10 @@ FUNCTION(pupa_biosdisk_rw_standard)
 
 FUNCTION(pupa_biosdisk_check_int13_extensions)
 	pushl	%ebp
-	movl	%esp, %ebp
-
 	pushl	%ebx
 
 	/* drive */
-	movb	0x8(%ebp), %dl
+	movb	%al, %dl
 	/* enter real mode */
 	call	prot_to_real
 
@@ -564,19 +564,16 @@ FUNCTION(pupa_biosdisk_check_int13_extensions)
 
 FUNCTION(pupa_biosdisk_get_diskinfo_int13_extensions)
 	pushl	%ebp
-	movl	%esp, %ebp
-
 	pushl	%ebx
 	pushl	%esi
 
 	/* compute the address of drive parameters */
-	movl	0xc(%ebp), %eax
-	movw	%ax, %si
-	xorw	%ax, %ax
-	shrl	$4, %eax
-	movw	%ax, %bx	/* save the segment into %bx */
+	movw	%dx, %si
+	xorw	%dx, %dx
+	shrl	$4, %edx
+	movw	%dx, %bx	/* save the segment into %bx */
 	/* drive */
-	movb	0x8(%ebp), %dl
+	movb	%al, %dl
 	/* enter real mode */
 	call	prot_to_real
 
@@ -613,13 +610,17 @@ FUNCTION(pupa_biosdisk_get_diskinfo_int13_extensions)
 
 FUNCTION(pupa_biosdisk_get_diskinfo_standard)
 	pushl	%ebp
-	movl	%esp, %ebp
-
 	pushl	%ebx
 	pushl	%edi
 
+	/* push CYLINDERS */
+	pushl	%edx
+	/* push HEADS */
+	pushl	%ecx
+	/* SECTORS is on the stack */
+	
 	/* drive */
-	movb	0x8(%ebp), %dl
+	movb	%al, %dl
 	/* enter real mode */
 	call	prot_to_real
 
@@ -640,31 +641,26 @@ FUNCTION(pupa_biosdisk_get_diskinfo_standard)
 	DATA32	call	real_to_prot
 	.code32
 
-	/* restore %ebp */
-	leal	0x8(%esp), %ebp
-	
-	/* heads */
+	/* pop HEADS */
+	popl	%edi
 	movb	%dh, %al
-	incl	%eax		/* the number of heads is counted from zero */
-	movl	0x10(%ebp), %edi
+	incl	%eax	/* the number of heads is counted from zero */
 	movl	%eax, (%edi)
 
-	/* sectors */
-	xorl	%eax, %eax
-	movb	%cl, %al
-	andb	$0x3f, %al
-	movl	0x14(%ebp), %edi
-	movl	%eax, (%edi)
-
-	/* cylinders */
-	shrb	$6, %cl
-	movb	%cl, %ah
+	/* pop CYLINDERS */
+	popl	%edi
 	movb	%ch, %al
-	incl	%eax		/* the number of cylinders is 
-				   counted from zero */
-	movl	0xc(%ebp), %edi
+	movb	%cl, %ah
+	shrb	$6, %ah	/* the number of cylinders is counted from zero */
+	incl	%eax
 	movl	%eax, (%edi)
 
+	/* SECTORS */
+	movl	0x10(%esp), %edi
+	andb	$0x3f, %cl
+	movzbl	%cl, %eax
+	movl	%eax, (%edi)
+	
 	xorl	%eax, %eax
 	movb	%bl, %al	/* return value in %eax */
 
@@ -672,7 +668,7 @@ FUNCTION(pupa_biosdisk_get_diskinfo_standard)
 	popl	%ebx
 	popl	%ebp
 
-	ret
+	ret	$4
 
 
 /*
@@ -724,14 +720,13 @@ FUNCTION(pupa_biosdisk_get_num_floppies)
 
 FUNCTION(pupa_get_memsize)
 	pushl	%ebp
-	pushl	%ebx
 
-	movl	0xc(%esp), %ebx
+	movl	%eax, %edx
 
 	call	prot_to_real	/* enter real mode */
 	.code16
 
-	testl	%ebx, %ebx
+	testl	%edx, %edx
 	jnz	xext
 
 	int	$0x12
@@ -742,13 +737,13 @@ xext:
 	int	$0x15
 
 xdone:
-	movw	%ax, %bx
+	movw	%ax, %dx
 
 	DATA32	call	real_to_prot
 	.code32
 
-	movw	%bx, %ax
-	popl	%ebx
+	movw	%dx, %ax
+
 	popl	%ebp
 	ret
 
@@ -802,13 +797,12 @@ xnoteisa:
  */
 
 FUNCTION(pupa_get_mmap_entry)
-	push	%ebp
-	push	%ebx
-	push	%edi
-	push	%esi
+	pusha
 
+	/* push ADDR */
+	pushl	%eax
+	
 	/* place address (+4) in ES:DI */
-	movl	0x14(%esp), %eax
 	addl	$4, %eax
 	movl	%eax, %edi
 	andl	$0xf, %edi
@@ -816,7 +810,7 @@ FUNCTION(pupa_get_mmap_entry)
 	movl	%eax, %esi
 
 	/* set continuation value */
-	movl	0x18(%esp), %ebx
+	movl	%edx, %ebx
 
 	/* set default maximum buffer size */
 	movl	$0x14, %ecx
@@ -851,17 +845,14 @@ xsmap:
 	DATA32	call	real_to_prot
 	.code32
 
-	/* write length of buffer (zero if error) into "addr" */
-	movl	0x14(%esp), %eax
+	/* write length of buffer (zero if error) into ADDR */
+	popl	%eax
 	movl	%ecx, (%eax)
 
 	/* set return value to continuation */
 	movl	%ebx, %eax
 
-	pop	%esi
-	pop	%edi
-	pop	%ebx
-	pop	%ebp
+	popa
 	ret
 
 	
@@ -937,7 +928,7 @@ gloop2ret:
  * support setting a background attribute.
  */
 FUNCTION(pupa_console_putchar)
-	movl	0x4(%esp), %edx
+	movl	%eax, %edx
 	pusha
 	movb	EXT_C(pupa_console_cur_color), %bl
 	
@@ -1122,8 +1113,8 @@ FUNCTION(pupa_console_gotoxy)
 	pushl	%ebp
 	pushl	%ebx                    /* save EBX */
 
-	movb	0xc(%esp), %dl           /* %dl = x */
-	movb	0x10(%esp), %dh          /* %dh = y */
+	movb	%dl, %dh	/* %dh = y */
+	movb	%al, %dl	/* %dl = x */
 
 	call	prot_to_real
 	.code16
@@ -1195,9 +1186,12 @@ console_cursor_shape:
 	.word	0
 	
 FUNCTION(pupa_console_setcursor)
-	push	%ebp
-	push	%ebx
+	pushl	%ebp
+	pushl	%ebx
 
+	/* push ON */
+	pushl	%eax
+	
 	/* check if the standard cursor shape has already been saved */
 	movw	console_cursor_shape, %ax
 	testw	%ax, %ax
@@ -1217,8 +1211,8 @@ FUNCTION(pupa_console_setcursor)
 1:
 	/* set %cx to the designated cursor shape */
 	movw	$0x2000, %cx
-	movl	0xc(%esp), %ebx
-	testl	%ebx, %ebx
+	popl	%eax
+	testl	%eax, %eax
 	jz	2f
 	movw	console_cursor_shape, %cx
 2:	
@@ -1254,7 +1248,7 @@ FUNCTION(pupa_console_setcursor)
  *                          that it isn't running)
  */
 FUNCTION(pupa_getrtsecs)
-	push	%ebp
+	pushl	%ebp
 
 	call	prot_to_real	/* enter real mode */
 	.code16
@@ -1272,7 +1266,7 @@ gottime:
 
 	movb	%dh, %al
 
-	pop	%ebp
+	popl	%ebp
 	ret