Index: stand/defs.mk
===================================================================
--- stand/defs.mk
+++ stand/defs.mk
@@ -111,6 +111,7 @@
 .if ${MACHINE_CPUARCH} == "i386" || (${MACHINE_CPUARCH} == "amd64" && ${DO32:U0} == 1)
 CFLAGS+=	-march=i386
 CFLAGS.gcc+=	-mpreferred-stack-boundary=2
+CFLAGS.clang+=	-Oz -mstack-alignment=8
 .endif
 .if ${MACHINE_CPUARCH} == "amd64" && ${DO32:U0} == 0
 CFLAGS+=	-fPIC -mno-red-zone
Index: stand/efi/boot1/ufs_module.c
===================================================================
--- stand/efi/boot1/ufs_module.c
+++ stand/efi/boot1/ufs_module.c
@@ -38,12 +38,15 @@
 #include <sys/param.h>
 #include <efi.h>
 
+#include <ufs/ffs/fs.h>
+
 #include "boot_module.h"
+#include "ufsread.h"
 
 static dev_info_t *devinfo;
 static dev_info_t *devices;
 
-static int
+int
 dskread(void *buf, u_int64_t lba, int nblk)
 {
 	int size;
@@ -66,8 +69,6 @@
 	return (0);
 }
 
-#include "ufsread.c"
-
 static struct dmadat __dmadat;
 
 static int
Index: stand/i386/boot2/Makefile
===================================================================
--- stand/i386/boot2/Makefile
+++ stand/i386/boot2/Makefile
@@ -81,7 +81,7 @@
 	${OBJCOPY} -S -O binary boot2.out ${.TARGET}
 
 boot2.out: ${BTXCRT} boot2.o sio.o
-	${LD} ${LD_FLAGS} -Ttext ${ORG2} -o ${.TARGET} ${.ALLSRC}
+	${LD} ${LD_FLAGS} -Ttext ${ORG2} -o ${.TARGET} ${.ALLSRC} ${LIBSA32}
 
 SRCS=	boot2.c boot2.h
 
Index: stand/i386/boot2/boot2.c
===================================================================
--- stand/i386/boot2/boot2.c
+++ stand/i386/boot2/boot2.c
@@ -25,6 +25,8 @@
 #include <machine/bootinfo.h>
 #include <machine/elf.h>
 
+#include <ufs/ffs/fs.h>
+
 #include <stdarg.h>
 
 #include <a.out.h>
@@ -35,6 +37,7 @@
 #include "lib.h"
 #include "paths.h"
 #include "rbx.h"
+#include "ufsread.h"
 
 /* Define to 0 to omit serial support */
 #ifndef SERIAL
@@ -109,21 +112,23 @@
 static uint8_t ioctrl = IO_KEYBOARD;
 #endif
 
-int main(void);
-void exit(int);
-static void load(void);
-static int parse(void);
-static int dskread(void *, unsigned, unsigned);
-static void printf(const char *,...);
-static void putchar(int);
-static int drvread(void *, unsigned, unsigned);
-static int keyhit(unsigned);
-static int xputc(int);
-static int xgetc(int);
-static inline int getc(int);
-
-static void memcpy(void *, const void *, int);
-static void
+int			main(void);
+void			exit(int);
+void			memcpy(void *, const void *, int);
+void			printf(const char *, ...);
+int			strcmp(const char *, const char *);
+int			dskread(void *, unsigned, unsigned);
+
+static void		load(void);
+static int		parse(void);
+static void		putchar(int);
+static int		drvread(void *, unsigned, unsigned);
+static int		keyhit(unsigned);
+static int		xputc(int);
+static int		xgetc(int);
+static inline int	getc(int);
+
+void
 memcpy(void *dst, const void *src, int len)
 {
     const char *s = src;
@@ -133,16 +138,13 @@
         *d++ = *s++;
 }
 
-static inline int
+int
 strcmp(const char *s1, const char *s2)
 {
     for (; *s1 == *s2 && *s1; s1++, s2++);
     return (unsigned char)*s1 - (unsigned char)*s2;
 }
 
-#define	UFS_SMALL_CGBASE
-#include "ufsread.c"
-
 static int
 xfsread(ufs_ino_t inode, void *buf, size_t nbyte)
 {
@@ -454,7 +456,7 @@
     return 0;
 }
 
-static int
+int
 dskread(void *buf, unsigned lba, unsigned nblk)
 {
     struct dos_partition *dp;
@@ -521,8 +523,8 @@
     return -1;
 }
 
-static void
-printf(const char *fmt,...)
+void
+printf(const char *fmt, ...)
 {
     va_list ap;
     static char buf[10];
Index: stand/i386/gptboot/gptboot.c
===================================================================
--- stand/i386/gptboot/gptboot.c
+++ stand/i386/gptboot/gptboot.c
@@ -26,6 +26,8 @@
 #include <machine/pc/bios.h>
 #include <machine/psl.h>
 
+#include <ufs/ffs/fs.h>
+
 #include <stdarg.h>
 
 #include <a.out.h>
@@ -41,6 +43,7 @@
 #include "cons.h"
 #include "gpt.h"
 #include "paths.h"
+#include "ufsread.h"
 
 #define ARGS		0x900
 #define NOPT		14
@@ -102,15 +105,15 @@
 static char *heap_next;
 static char *heap_end;
 
+int dskread(void *, daddr_t, unsigned);
+
 static void load(void);
 static int parse_cmds(char *, int *);
-static int dskread(void *, daddr_t, unsigned);
 #ifdef LOADER_GELI_SUPPORT
 static int vdev_read(void *vdev __unused, void *priv, off_t off, void *buf,
 	size_t bytes);
 #endif
 
-#include "ufsread.c"
 #include "gpt.c"
 #ifdef LOADER_GELI_SUPPORT
 #include "geliboot.c"
@@ -574,7 +577,7 @@
     return 0;
 }
 
-static int
+int
 dskread(void *buf, daddr_t lba, unsigned nblk)
 {
 	int err;
Index: stand/libsa/Makefile
===================================================================
--- stand/libsa/Makefile
+++ stand/libsa/Makefile
@@ -145,6 +145,9 @@
 .PATH: ${SRCTOP}/sys/ufs/ffs
 SRCS+=ffs_subr.c ffs_tables.c
 
+# simple filesystem read support
+SRCS+=	ufsread.c
+
 CFLAGS.bzipfs.c+= -I${SRCTOP}/contrib/bzip2
 
 # explicit_bzero
Index: stand/libsa/ufsread.h
===================================================================
--- /dev/null
+++ stand/libsa/ufsread.h
@@ -0,0 +1,40 @@
+#ifndef	UFSREAD_H
+#define	UFSREAD_H
+
+typedef	uint32_t	ufs_ino_t;
+
+/*
+ * We use 4k `virtual' blocks for filesystem data, whatever the actual
+ * filesystem block size. FFS blocks are always a multiple of 4k.
+ */
+#define VBLKSHIFT	12
+#define VBLKSIZE	(1 << VBLKSHIFT)
+#define VBLKMASK	(VBLKSIZE - 1)
+#define DBPERVBLK	(VBLKSIZE / DEV_BSIZE)
+#define INDIRPERVBLK(fs) (NINDIR(fs) / ((fs)->fs_bsize >> VBLKSHIFT))
+#define IPERVBLK(fs)	(INOPB(fs) / ((fs)->fs_bsize >> VBLKSHIFT))
+#define INO_TO_VBA(fs, ipervblk, x) \
+    (fsbtodb(fs, cgimin(fs, ino_to_cg(fs, x))) + \
+    (((x) % (fs)->fs_ipg) / (ipervblk) * DBPERVBLK))
+#define INO_TO_VBO(ipervblk, x) ((x) % ipervblk)
+#define FS_TO_VBA(fs, fsb, off) (fsbtodb(fs, fsb) + \
+    ((off) / VBLKSIZE) * DBPERVBLK)
+#define FS_TO_VBO(fs, fsb, off) ((off) & VBLKMASK)
+
+/* Buffers that must not span a 64k boundary. */
+struct dmadat {
+	char blkbuf[VBLKSIZE];	/* filesystem blocks */
+	char indbuf[VBLKSIZE];	/* indir blocks */
+	char sbbuf[SBLOCKSIZE];	/* superblock */
+	char secbuf[DEV_BSIZE];	/* for MBR/disklabel */
+};
+extern struct dmadat *dmadat;
+
+extern uint8_t ls, dsk_meta;
+extern uint32_t fs_off;
+
+ufs_ino_t   lookup(const char *);
+ssize_t     fsread(ufs_ino_t, void *, size_t);
+ssize_t		fsread_size(ufs_ino_t, void *, size_t, size_t *);
+
+#endif	/* UFSREAD_H */
Index: stand/libsa/ufsread.c
===================================================================
--- stand/libsa/ufsread.c
+++ stand/libsa/ufsread.c
@@ -46,10 +46,20 @@
 #include <sys/cdefs.h>
 __FBSDID("$FreeBSD$");
 
+#include <sys/param.h>
+
 #include <ufs/ufs/dinode.h>
 #include <ufs/ufs/dir.h>
 #include <ufs/ffs/fs.h>
 
+#include "ufsread.h"
+
+/* These routines must be provided by other code. */
+void	memcpy(void *, const void *, int);
+void	printf(const char *, ...);
+int	strcmp(const char *, const char *);
+int	dskread(void *, unsigned, unsigned);
+
 #ifdef UFS_SMALL_CGBASE
 /* XXX: Revert to old (broken for over 1.5Tb filesystems) version of cgbase
    (see sys/ufs/ffs/fs.h rev 1.39) so that small boot loaders (e.g. boot2) can
@@ -58,40 +68,10 @@
 #define cgbase(fs, c)   ((ufs2_daddr_t)((fs)->fs_fpg * (c)))
 #endif
 
-typedef	uint32_t	ufs_ino_t;
-
-/*
- * We use 4k `virtual' blocks for filesystem data, whatever the actual
- * filesystem block size. FFS blocks are always a multiple of 4k.
- */
-#define VBLKSHIFT	12
-#define VBLKSIZE	(1 << VBLKSHIFT)
-#define VBLKMASK	(VBLKSIZE - 1)
-#define DBPERVBLK	(VBLKSIZE / DEV_BSIZE)
-#define INDIRPERVBLK(fs) (NINDIR(fs) / ((fs)->fs_bsize >> VBLKSHIFT))
-#define IPERVBLK(fs)	(INOPB(fs) / ((fs)->fs_bsize >> VBLKSHIFT))
-#define INO_TO_VBA(fs, ipervblk, x) \
-    (fsbtodb(fs, cgimin(fs, ino_to_cg(fs, x))) + \
-    (((x) % (fs)->fs_ipg) / (ipervblk) * DBPERVBLK))
-#define INO_TO_VBO(ipervblk, x) ((x) % ipervblk)
-#define FS_TO_VBA(fs, fsb, off) (fsbtodb(fs, fsb) + \
-    ((off) / VBLKSIZE) * DBPERVBLK)
-#define FS_TO_VBO(fs, fsb, off) ((off) & VBLKMASK)
-
-/* Buffers that must not span a 64k boundary. */
-struct dmadat {
-	char blkbuf[VBLKSIZE];	/* filesystem blocks */
-	char indbuf[VBLKSIZE];	/* indir blocks */
-	char sbbuf[SBLOCKSIZE];	/* superblock */
-	char secbuf[DEV_BSIZE];	/* for MBR/disklabel */
-};
-static struct dmadat *dmadat;
-
-static ufs_ino_t lookup(const char *);
-static ssize_t fsread(ufs_ino_t, void *, size_t);
+struct dmadat *dmadat;
 
-static uint8_t ls, dsk_meta;
-static uint32_t fs_off;
+uint8_t ls, dsk_meta;
+uint32_t fs_off;
 
 static __inline uint8_t
 fsfind(const char *name, ufs_ino_t * ino)
@@ -118,7 +98,7 @@
 	return 0;
 }
 
-static ufs_ino_t
+ufs_ino_t
 lookup(const char *path)
 {
 	static char name[UFS_MAXNAMLEN + 1];
@@ -164,7 +144,7 @@
 #define DIP(field) fs.fs_magic == FS_UFS1_MAGIC ? dp1.field : dp2.field
 #endif
 
-static ssize_t
+ssize_t
 fsread_size(ufs_ino_t inode, void *buf, size_t nbyte, size_t *fsizep)
 {
 #ifndef UFS2_ONLY
@@ -317,7 +297,7 @@
 	return nbyte;
 }
 
-static ssize_t
+ssize_t
 fsread(ufs_ino_t inode, void *buf, size_t nbyte)
 {
 
Index: stand/mips/beri/boot2/Makefile
===================================================================
--- stand/mips/beri/boot2/Makefile
+++ stand/mips/beri/boot2/Makefile
@@ -59,7 +59,8 @@
 .PATH:		${BOOTSRC}/mips/beri/common
 CFLAGS+=	-I${BOOTSRC}/mips/beri/common
 
-flashboot.elf: relocate.o start.o boot2.o altera_jtag_uart.o cfi.o sdcard.o
+flashboot.elf: relocate.o start.o boot2.o altera_jtag_uart.o cfi.o sdcard.o \
+	       ${LIBSA}
 	${CC} ${LDFLAGS} -T ${.CURDIR}/flashboot.ldscript -o ${.TARGET}	\
 	    ${.ALLSRC} ${LIBSA}
 flashboot: flashboot.elf
@@ -67,7 +68,7 @@
 flashboot.md5: flashboot
 	md5 flashboot > flashboot.md5
 
-jtagboot: start.o boot2.o altera_jtag_uart.o cfi.o sdcard.o
+jtagboot: start.o boot2.o altera_jtag_uart.o cfi.o sdcard.o ${LIBSA}
 	${CC} ${LDFLAGS} -T ${.CURDIR}/jtagboot.ldscript -o ${.TARGET}	\
 	    ${.ALLSRC} ${LIBSA}
 jtagboot.md5: jtagboot
Index: stand/mips/beri/boot2/boot2.c
===================================================================
--- stand/mips/beri/boot2/boot2.c
+++ stand/mips/beri/boot2/boot2.c
@@ -54,6 +54,8 @@
 #include <machine/bootinfo.h>
 #include <machine/elf.h>
 
+#include <ufs/ffs/fs.h>
+
 #include <stand.h>
 #include <stdarg.h>
 #include <string.h>
@@ -66,6 +68,7 @@
 
 #include "paths.h"
 #include "rbx.h"
+#include "ufsread.h"
 
 static int		 beri_argc;
 static const char	**beri_argv, **beri_envv;
@@ -138,17 +141,15 @@
 static uint8_t ioctrl = IO_KEYBOARD;
 
 void putchar(int);
+int dskread(void *, unsigned, unsigned);
+
 static void boot_fromdram(void);
 static void boot_fromfs(void);
 static void load(void);
 static int parse(void);
-static int dskread(void *, unsigned, unsigned);
 static int xputc(int);
 static int xgetc(int);
 
-#define	UFS_SMALL_CGBASE
-#include "ufsread.c"
-
 static struct dmadat __dmadat;
 
 static inline int
@@ -543,7 +544,7 @@
 	}
 }
 
-static int
+int
 dskread(void *buf, unsigned lba, unsigned nblk)
 {
 #if 0
Index: stand/powerpc/boot1.chrp/Makefile
===================================================================
--- stand/powerpc/boot1.chrp/Makefile
+++ stand/powerpc/boot1.chrp/Makefile
@@ -12,7 +12,10 @@
 CFLAGS+=-I${LDRSRC}
 LDFLAGS=-nostdlib -static -Wl,-N
 
-.PATH:  ${SYSDIR}/libkern ${SRCTOP}/lib/libc/powerpc/gen ${.CURDIR}
+DPADD+=		${LIBSA}
+LDADD+=		${LIBSA}
+
+.PATH:	${SYSDIR}/libkern ${SRCTOP}/lib/libc/powerpc/gen ${.CURDIR}
 
 # The following inserts out objects into a template HFS 
 # created by generate-hfs.sh
Index: stand/powerpc/boot1.chrp/boot1.c
===================================================================
--- stand/powerpc/boot1.chrp/boot1.c
+++ stand/powerpc/boot1.chrp/boot1.c
@@ -24,7 +24,10 @@
 #include <machine/stdarg.h>
 #include <machine/md_var.h>
 
+#include <ufs/ffs/fs.h>
+
 #include "paths.h"
+#include "ufsread.h"
 
 #define BSIZEMAX	16384
 
@@ -48,13 +51,14 @@
 static char blkbuf[BSIZEMAX];
 static unsigned int fsblks;
 
-static uint32_t fs_off;
-
 int main(int ac, char **av);
+int dskread(void *, u_int64_t, int);
+void memcpy(void *dst, const void *src, size_t len);
+int printf(const char *fmt, ...);
+int strcmp(const char *s1, const char *s2);
 
 static void exit(int) __dead2;
 static void load(const char *);
-static int dskread(void *, u_int64_t, int);
 
 static void usage(void);
 
@@ -64,7 +68,6 @@
 static int domount(const char *device, int quiet);
 
 static void panic(const char *fmt, ...) __dead2;
-static int printf(const char *fmt, ...);
 static int putchar(char c, void *arg);
 static int vprintf(const char *fmt, va_list ap);
 static int vsnprintf(char *str, size_t sz, const char *fmt, va_list ap);
@@ -364,7 +367,7 @@
 		*d++ = *s++;
 }
 
-static void
+void
 memcpy(void *dst, const void *src, size_t len)
 {
 	bcopy(src, dst, len);
@@ -379,7 +382,7 @@
 		*p++ = 0;
 }
 
-static int
+int
 strcmp(const char *s1, const char *s2)
 {
 	for (; *s1 == *s2 && *s1; s1++, s2++)
@@ -387,8 +390,6 @@
 	return ((u_char)*s1 - (u_char)*s2);
 }
 
-#include "ufsread.c"
-
 int
 main(int ac, char **av)
 {
@@ -540,7 +541,7 @@
 	    ofw,NULL,0);
 }
 
-static int
+int
 dskread(void *buf, u_int64_t lba, int nblk)
 {
 	/*
@@ -567,7 +568,7 @@
 	exit(1);
 }
 
-static int
+int
 printf(const char *fmt, ...)
 {
 	va_list ap;
Index: stand/sparc64/boot1/Makefile
===================================================================
--- stand/sparc64/boot1/Makefile
+++ stand/sparc64/boot1/Makefile
@@ -8,6 +8,9 @@
 SRCS=	_start.s boot1.c
 CLEANFILES+=${FILES} boot1.aout
 
+DPADD+=	${LIBSA}
+LDADD+=	${LIBSA}
+
 BOOTBLOCKBASE= 0x4000
 
 CFLAGS.clang+=-mcmodel=small
Index: stand/sparc64/boot1/boot1.c
===================================================================
--- stand/sparc64/boot1/boot1.c
+++ stand/sparc64/boot1/boot1.c
@@ -24,7 +24,10 @@
 #include <machine/elf.h>
 #include <machine/stdarg.h>
 
+#include <ufs/ffs/fs.h>
+
 #include "paths.h"
+#include "ufsread.h"
 
 #define	READ_BUF_SIZE	8192
 
@@ -44,9 +47,12 @@
 
 static ofwh_t bootdev;
 
-static uint32_t fs_off;
-
 int main(int ac, char **av);
+int dskread(void *buf, u_int64_t lba, int nblk);
+void memcpy(void *dst, const void *src, size_t len);
+int printf(const char *fmt, ...);
+int strcmp(const char *s1, const char *s2);
+
 static void exit(int) __dead2;
 static void usage(void);
 
@@ -61,10 +67,8 @@
 static void bzero(void *b, size_t len);
 
 static int domount(const char *device);
-static int dskread(void *buf, u_int64_t lba, int nblk);
 
 static void panic(const char *fmt, ...) __dead2;
-static int printf(const char *fmt, ...);
 static int putchar(char c, void *arg);
 static int vprintf(const char *fmt, va_list ap);
 static int vsnprintf(char *str, size_t sz, const char *fmt, va_list ap);
@@ -293,7 +297,7 @@
 		*d++ = *s++;
 }
 
-static void
+void
 memcpy(void *dst, const void *src, size_t len)
 {
 
@@ -309,7 +313,7 @@
 		*p++ = 0;
 }
 
-static int
+int
 strcmp(const char *s1, const char *s2)
 {
 
@@ -448,8 +452,6 @@
 
 #else
 
-#include "ufsread.c"
-
 static struct dmadat __dmadat;
 
 static void
@@ -514,7 +516,7 @@
 	return (0);
 }
 
-static int
+int
 dskread(void *buf, u_int64_t lba, int nblk)
 {
 
@@ -542,7 +544,7 @@
 	exit(1);
 }
 
-static int
+int
 printf(const char *fmt, ...)
 {
 	va_list ap;