Merge with Linux 2.3.99-pre4.

15 files changed, 321 insertions, 192 deletions

diff --git a/include/asm-sh/bitops.h b/include/asm-sh/bitops.h
index bf0d00c0a..85462da1e 100644
--- a/include/asm-sh/bitops.h
+++ b/include/asm-sh/bitops.h
@@ -103,7 +103,7 @@ extern __inline__ unsigned long ffz(unsigned long word)
 {
 	unsigned long result;
 
-	__asm__("1:\n"
+	__asm__("1:\n\t"
 		"shlr	%1\n\t"
 		"bt/s	1b\n\t"
 		" add	#1, %0"
diff --git a/include/asm-sh/byteorder.h b/include/asm-sh/byteorder.h
index dcc9f8241..cddaeb58e 100644
--- a/include/asm-sh/byteorder.h
+++ b/include/asm-sh/byteorder.h
@@ -19,7 +19,7 @@ static __inline__ __const__ __u32 ___arch__swab32(__u32 x)
 
 static __inline__ __const__ __u16 ___arch__swab16(__u16 x)
 {
-	__asm__("swap.b %0,%0"
+	__asm__("swap.b %0, %0"
 		: "=r" (x)
 		:  "0" (x));
 	return x;
diff --git a/include/asm-sh/checksum.h b/include/asm-sh/checksum.h
index ae0272e9b..846e64509 100644
--- a/include/asm-sh/checksum.h
+++ b/include/asm-sh/checksum.h
@@ -55,24 +55,6 @@ unsigned int csum_partial_copy_from_user ( const char *src, char *dst,
 	return csum_partial_copy_generic ( src, dst, len, sum, err_ptr, NULL);
 }
 
-#if 0
-
-/* Not used at the moment. It is difficult to imagine for what purpose
-   it can be used :-) Please, do not forget to verify_area before it --ANK
- */
-
-/*
- * This combination is currently not used, but possible:
- */
-
-extern __inline__
-unsigned int csum_partial_copy_to_user ( const char *src, char *dst,
-					int len, int sum, int *err_ptr)
-{
-	return csum_partial_copy_generic ( src, dst, len, sum, NULL, err_ptr);
-}
-#endif
-
 /*
  * These are the old (and unsafe) way of doing checksums, a warning message will be
  * printed if they are used and an exeption occurs.
@@ -91,12 +73,12 @@ static __inline__ unsigned int csum_fold(unsigned int sum)
 {
 	unsigned int __dummy;
 	__asm__("clrt\n\t"
-		"mov	%0,%1\n\t"
+		"mov	%0, %1\n\t"
 		"shll16	%0\n\t"
-		"addc	%0,%1\n\t"
+		"addc	%0, %1\n\t"
 		"movt	%0\n\t"
 		"shlr16	%1\n\t"
-		"add	%1,%0"
+		"add	%1, %0"
 		: "=r" (sum), "=&r" (__dummy)
 		: "0" (sum));
 	return ~sum;
@@ -114,24 +96,24 @@ static __inline__ unsigned short ip_fast_csum(unsigned char * iph, unsigned int
 	unsigned int sum, __dummy;
 
 	__asm__ __volatile__(
-		"mov.l	@%1+,%0\n\t"
-		"add	#-4,%2\n\t"
+		"mov.l	@%1+, %0\n\t"
+		"add	#-4, %2\n\t"
 		"clrt\n\t"
-		"mov.l	@%1+,%3\n\t"
-		"addc	%3,%0\n\t"
-		"mov.l	@%1+,%3\n\t"
-		"addc	%3,%0\n\t"
-		"mov.l	@%1+,%3\n\t"
-		"addc	%3,%0\n"
+		"mov.l	@%1+, %3\n\t"
+		"addc	%3, %0\n\t"
+		"mov.l	@%1+, %3\n\t"
+		"addc	%3, %0\n\t"
+		"mov.l	@%1+, %3\n\t"
+		"addc	%3, %0\n"
 		"1:\t"
-		"mov.l	@%1+,%3\n\t"
-		"addc	%3,%0\n\t"
+		"mov.l	@%1+, %3\n\t"
+		"addc	%3, %0\n\t"
 		"movt	%3\n\t"
 		"dt	%2\n\t"
 		"bf/s	1b\n\t"
-		" cmp/eq #1,%3\n\t"
-		"mov	#0,%3\n\t"
-		"addc	%3,%0\n\t"
+		" cmp/eq #1, %3\n\t"
+		"mov	#0, %3\n\t"
+		"addc	%3, %0\n\t"
 	/* Since the input registers which are loaded with iph and ihl
 	   are modified, we must also specify them as outputs, or gcc
 	   will assume they contain their original values. */
@@ -153,11 +135,11 @@ static __inline__ unsigned long csum_tcpudp_nofold(unsigned long saddr,
 	unsigned long len_proto = (proto<<16)+len;
 #endif
 	__asm__("clrt\n\t"
-		"addc	%0,%1\n\t"
-		"addc	%2,%1\n\t"
-		"addc	%3,%1\n\t"
+		"addc	%0, %1\n\t"
+		"addc	%2, %1\n\t"
+		"addc	%3, %1\n\t"
 		"movt	%0\n\t"
-		"add	%1,%0"
+		"add	%1, %0"
 		: "=r" (sum), "=r" (len_proto)
 		: "r" (daddr), "r" (saddr), "1" (len_proto), "0" (sum));
 	return sum;
@@ -195,26 +177,26 @@ static __inline__ unsigned short int csum_ipv6_magic(struct in6_addr *saddr,
 {
 	unsigned int __dummy;
 	__asm__("clrt\n\t"
-		"mov.l	@(0,%2),%1\n\t"
-		"addc	%1,%0\n\t"
-		"mov.l	@(4,%2),%1\n\t"
-		"addc	%1,%0\n\t"
-		"mov.l	@(8,%2),%1\n\t"
-		"addc	%1,%0\n\t"
-		"mov.l	@(12,%2),%1\n\t"
-		"addc	%1,%0\n\t"
-		"mov.l	@(0,%3),%1\n\t"
-		"addc	%1,%0\n\t"
-		"mov.l	@(4,%3),%1\n\t"
-		"addc	%1,%0\n\t"
-		"mov.l	@(8,%3),%1\n\t"
-		"addc	%1,%0\n\t"
-		"mov.l	@(12,%3),%1\n\t"
-		"addc	%1,%0\n\t"
-		"addc	%4,%0\n\t"
-		"addc	%5,%0\n\t"
+		"mov.l	@(0,%2), %1\n\t"
+		"addc	%1, %0\n\t"
+		"mov.l	@(4,%2), %1\n\t"
+		"addc	%1, %0\n\t"
+		"mov.l	@(8,%2), %1\n\t"
+		"addc	%1, %0\n\t"
+		"mov.l	@(12,%2), %1\n\t"
+		"addc	%1, %0\n\t"
+		"mov.l	@(0,%3), %1\n\t"
+		"addc	%1, %0\n\t"
+		"mov.l	@(4,%3), %1\n\t"
+		"addc	%1, %0\n\t"
+		"mov.l	@(8,%3), %1\n\t"
+		"addc	%1, %0\n\t"
+		"mov.l	@(12,%3), %1\n\t"
+		"addc	%1, %0\n\t"
+		"addc	%4, %0\n\t"
+		"addc	%5, %0\n\t"
 		"movt	%1\n\t"
-		"add	%1,%0\n"
+		"add	%1, %0\n"
 		: "=r" (sum), "=&r" (__dummy)
 		: "r" (saddr), "r" (daddr), 
 		  "r" (htonl(len)), "r" (htonl(proto)), "0" (sum));
diff --git a/include/asm-sh/current.h b/include/asm-sh/current.h
index fe65bfcc9..3bd231173 100644
--- a/include/asm-sh/current.h
+++ b/include/asm-sh/current.h
@@ -12,8 +12,8 @@ static __inline__ struct task_struct * get_current(void)
 {
 	struct task_struct *current;
 
-	__asm__("stc	r4_bank,%0\n\t"
-		"add	%1,%0"
+	__asm__("stc	$r4_bank, %0\n\t"
+		"add	%1, %0"
 		:"=&r" (current)
 		:"r" (-8192));
 	return current;
diff --git a/include/asm-sh/delay.h b/include/asm-sh/delay.h
index 222561a3e..3e25684fe 100644
--- a/include/asm-sh/delay.h
+++ b/include/asm-sh/delay.h
@@ -8,7 +8,7 @@
 extern __inline__ void __delay(unsigned long loops)
 {
 	__asm__ __volatile__(
-		"tst	%0,%0\n\t"
+		"tst	%0, %0\n\t"
 		"1:\t"
 		"bf/s	1b\n\t"
 		" dt	%0"
@@ -19,8 +19,8 @@ extern __inline__ void __delay(unsigned long loops)
 extern __inline__ void __udelay(unsigned long usecs, unsigned long lps)
 {
 	usecs *= 0x000010c6;		/* 2**32 / 1000000 */
-	__asm__("dmulu.l	%0,%2\n\t"
-		"sts	mach,%0"
+	__asm__("dmulu.l	%0, %2\n\t"
+		"sts	$mach, %0"
 		: "=r" (usecs)
 		: "0" (usecs), "r" (lps)
 		: "macl", "mach");
diff --git a/include/asm-sh/mman.h b/include/asm-sh/mman.h
index 19b02f6da..49fecaed5 100644
--- a/include/asm-sh/mman.h
+++ b/include/asm-sh/mman.h
@@ -25,6 +25,12 @@
 #define MCL_CURRENT	1		/* lock all current mappings */
 #define MCL_FUTURE	2		/* lock all future mappings */
 
+#define MADV_NORMAL	0x0		/* default page-in behavior */
+#define MADV_RANDOM	0x1		/* page-in minimum required */
+#define MADV_SEQUENTIAL	0x2		/* read-ahead aggressively */
+#define MADV_WILLNEED	0x3		/* pre-fault pages */
+#define MADV_DONTNEED	0x4		/* discard these pages */
+
 /* compatibility flags */
 #define MAP_ANON	MAP_ANONYMOUS
 #define MAP_FILE	0
diff --git a/include/asm-sh/namei.h b/include/asm-sh/namei.h
index 582fc77e3..a51522ac1 100644
--- a/include/asm-sh/namei.h
+++ b/include/asm-sh/namei.h
@@ -12,7 +12,6 @@
  * Look at asm-sparc/namei.h for details.
  */
 
-#define __prefix_lookup_dentry(name, lookup_flags) \
-        do {} while (0)
+#define __emul_prefix() NULL
 
 #endif /* __ASM_SH_NAMEI_H */
diff --git a/include/asm-sh/processor.h b/include/asm-sh/processor.h
index 962461a33..1fcfb1300 100644
--- a/include/asm-sh/processor.h
+++ b/include/asm-sh/processor.h
@@ -45,8 +45,10 @@ extern struct sh_cpuinfo boot_cpu_data;
 
 /*
  * User space process size: 2GB.
+ *
+ * Since SH7709 and SH7750 have "area 7", we can't use 0x7c000000--0x7fffffff
  */
-#define TASK_SIZE	0x80000000
+#define TASK_SIZE	0x7c000000UL
 
 /* This decides where the kernel will search for a free chunk of vm
  * space during mmap's.
@@ -54,19 +56,25 @@ extern struct sh_cpuinfo boot_cpu_data;
 #define TASK_UNMAPPED_BASE	(TASK_SIZE / 3)
 
 /*
- * FPU structure and data
- */
-/* FD-bit of SR register.
- * When it's set, it means the processor doesn't have right to use FPU,
- * and it results exception when the floating operation is executed.
+ * Bit of SR register
+ *
+ * FD-bit:
+ *     When it's set, it means the processor doesn't have right to use FPU,
+ *     and it results exception when the floating operation is executed.
+ *
+ * IMASK-bit:
+ *     Interrupt level mask
  */
-#define SR_FD	0x00008000
+#define SR_FD    0x00008000
+#define SR_IMASK 0x000000f0
 
-#define NUM_FPU_REGS	16
+/*
+ * FPU structure and data
+ */
 
 struct sh_fpu_hard_struct {
-	unsigned long fp_regs[NUM_FPU_REGS];
-	unsigned long xf_regs[NUM_FPU_REGS];
+	unsigned long fp_regs[16];
+	unsigned long long xd_regs[8];
 	unsigned long fpscr;
 	unsigned long fpul;
 
@@ -75,13 +83,13 @@ struct sh_fpu_hard_struct {
 
 /* Dummy fpu emulator  */
 struct sh_fpu_soft_struct {
-	unsigned long fp_regs[NUM_FPU_REGS];
+	unsigned long fp_regs[16];
+	unsigned long long xd_regs[8];
 	unsigned long fpscr;
 	unsigned long fpul;
-	unsigned long xf_regs[NUM_FPU_REGS];
 
-	unsigned char	lookahead;
-	unsigned long	entry_pc;
+	unsigned char lookahead;
+	unsigned long entry_pc;
 };
 
 union sh_fpu_union {
@@ -120,7 +128,7 @@ struct thread_struct {
 	regs->pr = 0;   		 	 \
 	regs->sr = 0;		/* User mode. */ \
 	regs->pc = new_pc;			 \
-	regs->sp = new_sp
+	regs->regs[15] = new_sp
 
 /* Forward declaration, a strange C thing */
 struct task_struct;
@@ -189,6 +197,9 @@ extern void save_fpu(struct task_struct *__tsk);
 		(tsk)->flags &= ~PF_USEDFPU; 	\
 } while (0)
 
+/* Double presision, NANS as NANS, rounding to nearest, no exceptions */
+#define FPSCR_INIT  0x00080000
+
 /*
  * Return saved PC of a blocked thread.
  */
diff --git a/include/asm-sh/ptrace.h b/include/asm-sh/ptrace.h
index ad3ab8905..369fc23bb 100644
--- a/include/asm-sh/ptrace.h
+++ b/include/asm-sh/ptrace.h
@@ -2,19 +2,49 @@
 #define __ASM_SH_PTRACE_H
 
 /*
- * Copyright (C) 1999  Niibe Yutaka
+ * Copyright (C) 1999, 2000  Niibe Yutaka
  *
  */
 
 /*
+ * As GCC does:
+ *  0 - 15 are integer registers
+ * 17 - 22 are control/special registers
+ * 24 - 39 fp registers
+ * 40 - 47 xd registers
+ * 48 -    fpscr register
+ * -----------------------------
+ * Not as GCC:
+ * 16 --- program counter PC
+ * 23 --- syscall #
+ */
+#define REG_REG0	 0
+#define REG_REG15	15
+#define REG_PC		16
+
+#define REG_PR		17
+#define REG_SR		18
+#define REG_GBR      	19
+#define REG_MACH	20
+#define REG_MACL	21
+#define REG_FPUL	22
+
+#define REG_SYSCALL	23
+
+#define REG_FPREG0	24
+#define REG_FPREG15	39
+#define REG_XDREG0	40
+#define REG_XDREG14	47
+#define REG_FPSCR	48
+
+/*
  * This struct defines the way the registers are stored on the
  * kernel stack during a system call or other kernel entry.
  */
 struct pt_regs {
 	long syscall_nr;
 	unsigned long sr;
-	unsigned long sp;
-	unsigned long regs[15];
+	unsigned long regs[16];
 	unsigned long gbr;
 	unsigned long mach;
 	unsigned long macl;
@@ -26,6 +56,60 @@ struct pt_regs {
 #define user_mode(regs) (((regs)->sr & 0x40000000)==0)
 #define instruction_pointer(regs) ((regs)->pc)
 extern void show_regs(struct pt_regs *);
+
+/* User Break Controller */
+
+#if defined(__sh3__)
+/* The value is for sh4, please fix... */
+#define UBC_BARA		0xff200000
+#define UBC_BAMRA		0xff200004
+#define UBC_BBRA		0xff200008
+#define UBC_BASRA		0xff000014
+#define UBC_BARB		0xff20000c
+#define UBC_BAMRB		0xff200010
+#define UBC_BBRB		0xff200014
+#define UBC_BASRB		0xff000018
+#define UBC_BDRB		0xff200018
+#define UBC_BDMRB		0xff20001c
+#define UBC_BRCR		0xff200020
+#elif defined(__SH4__)
+#define UBC_BARA		0xff200000
+#define UBC_BAMRA		0xff200004
+#define UBC_BBRA		0xff200008
+#define UBC_BASRA		0xff000014
+#define UBC_BARB		0xff20000c
+#define UBC_BAMRB		0xff200010
+#define UBC_BBRB		0xff200014
+#define UBC_BASRB		0xff000018
+#define UBC_BDRB		0xff200018
+#define UBC_BDMRB		0xff20001c
+#define UBC_BRCR		0xff200020
+#endif
+
+#define BAMR_ASID		(1 << 2)
+#define BAMR_NONE		0
+#define BAMR_10		0x1
+#define BAMR_12		0x2
+#define BAMR_ALL		0x3
+#define BAMR_16		0x8
+#define BAMR_20		0x9
+
+#define BBR_INST		(1 << 4)
+#define BBR_DATA		(2 << 4)
+#define BBR_READ		(1 << 2)
+#define BBR_WRITE		(2 << 4)
+#define BBR_BYTE		0x1
+#define BBR_HALF		0x2
+#define BBR_LONG		0x3
+#define BBR_QUAD		(1 << 6)
+
+#define BRCR_CMFA		(1 << 15)
+#define BRCR_CMFB		(1 << 14)
+#define BRCR_PCBA		(1 << 10)	/* 1: after execution */
+#define BRCR_DBEB		(1 << 7)
+#define BRCR_PCBB		(1 << 6)
+#define BRCR_SEQ		(1 << 3)
+#define BRCR_UBDE		(1 << 0)
 #endif
 
 #endif /* __ASM_SH_PTRACE_H */
diff --git a/include/asm-sh/sigcontext.h b/include/asm-sh/sigcontext.h
index c1a77873b..6f61ed38c 100644
--- a/include/asm-sh/sigcontext.h
+++ b/include/asm-sh/sigcontext.h
@@ -5,14 +5,22 @@ struct sigcontext {
 	unsigned long	oldmask;
 
 	/* CPU registers */
-	unsigned long sc_regs[15];
+	unsigned long sc_regs[16];
 	unsigned long sc_gbr;
 	unsigned long sc_mach;
 	unsigned long sc_macl;
 	unsigned long sc_pr;
-	unsigned long sc_sp;
 	unsigned long sc_sr;
 	unsigned long sc_pc;
+
+#if defined(__SH4__)
+	/* FPU registers */
+	unsigned long sc_fpregs[16];
+	unsigned long long sc_xdregs[8];
+	unsigned int sc_fpscr;
+	unsigned int sc_fpul;
+	unsigned int sc_ownedfp;
+#endif
 };
 
 #endif /* __ASM_SH_SIGCONTEXT_H */
diff --git a/include/asm-sh/string.h b/include/asm-sh/string.h
index bcff30489..3eab1c123 100644
--- a/include/asm-sh/string.h
+++ b/include/asm-sh/string.h
@@ -13,11 +13,11 @@ extern __inline__ char *strcpy(char *__dest, const char *__src)
 	unsigned long __dummy;
 
 	__asm__ __volatile__("1:\n\t"
-			     "mov.b	@%1+,%2\n\t"
-			     "mov.b	%2,@%0\n\t"
-			     "cmp/eq	#0,%2\n\t"
+			     "mov.b	@%1+, %2\n\t"
+			     "mov.b	%2, @%0\n\t"
+			     "cmp/eq	#0, %2\n\t"
 			     "bf/s	1b\n\t"
-			     " add	#1,%0\n\t"
+			     " add	#1, %0\n\t"
 			     : "=r" (__dest), "=r" (__src), "=&z" (__dummy)
 			     : "0" (__dest), "1" (__src)
 			     : "memory");
@@ -36,13 +36,13 @@ extern __inline__ char *strncpy(char *__dest, const char *__src, size_t __n)
 
 	__asm__ __volatile__(
 		"1:\n"
-		"mov.b	@%1+,%2\n\t"
-		"mov.b	%2,@%0\n\t"
-		"cmp/eq	#0,%2\n\t"
+		"mov.b	@%1+, %2\n\t"
+		"mov.b	%2, @%0\n\t"
+		"cmp/eq	#0, %2\n\t"
 		"bt/s	2f\n\t"
 		" cmp/eq	%5,%1\n\t"
 		"bf/s	1b\n\t"
-		" add	#1,%0\n"
+		" add	#1, %0\n"
 		"2:"
 		: "=r" (__dest), "=r" (__src), "=&z" (__dummy)
 		: "0" (__dest), "1" (__src), "r" (__src+__n)
@@ -58,17 +58,17 @@ extern __inline__ int strcmp(const char *__cs, const char *__ct)
 	unsigned long __dummy;
 
 	__asm__ __volatile__(
-		"mov.b	@%1+,%3\n"
+		"mov.b	@%1+, %3\n"
 		"1:\n\t"
-		"mov.b	@%0+,%2\n\t"
-		"cmp/eq #0,%3\n\t"
+		"mov.b	@%0+, %2\n\t"
+		"cmp/eq #0, %3\n\t"
 		"bt	2f\n\t"
-		"cmp/eq %2,%3\n\t"
+		"cmp/eq %2, %3\n\t"
 		"bt/s	1b\n\t"
-		" mov.b	@%1+,%3\n\t"
-		"add	#-2,%1\n\t"
-		"mov.b	@%1,%3\n\t"
-		"sub	%3,%2\n"
+		" mov.b	@%1+, %3\n\t"
+		"add	#-2, %1\n\t"
+		"mov.b	@%1, %3\n\t"
+		"sub	%3, %2\n"
 		"2:"
 		: "=r" (__cs), "=r" (__ct), "=&r" (__res), "=&z" (__dummy)
 		: "0" (__cs), "1" (__ct));
@@ -83,20 +83,20 @@ extern __inline__ int strncmp(const char *__cs, const char *__ct, size_t __n)
 	unsigned long __dummy;
 
 	__asm__ __volatile__(
-		"mov.b	@%1+,%3\n"
+		"mov.b	@%1+, %3\n"
 		"1:\n\t"
-		"mov.b	@%0+,%2\n\t"
-		"cmp/eq %6,%0\n\t"
+		"mov.b	@%0+, %2\n\t"
+		"cmp/eq %6, %0\n\t"
 		"bt/s	2f\n\t"
-		" cmp/eq #0,%3\n\t"
+		" cmp/eq #0, %3\n\t"
 		"bt/s	3f\n\t"
-		" cmp/eq %3,%2\n\t"
+		" cmp/eq %3, %2\n\t"
 		"bt/s	1b\n\t"
-		" mov.b	@%1+,%3\n\t"
-		"add	#-2,%1\n\t"
-		"mov.b	@%1,%3\n"
+		" mov.b	@%1+, %3\n\t"
+		"add	#-2, %1\n\t"
+		"mov.b	@%1, %3\n"
 		"2:\n\t"
-		"sub	%3,%2\n"
+		"sub	%3, %2\n"
 		"3:"
 		:"=r" (__cs), "=r" (__ct), "=&r" (__res), "=&z" (__dummy)
 		: "0" (__cs), "1" (__ct), "r" (__cs+__n));
diff --git a/include/asm-sh/system.h b/include/asm-sh/system.h
index e76071bf2..130800b47 100644
--- a/include/asm-sh/system.h
+++ b/include/asm-sh/system.h
@@ -2,7 +2,7 @@
 #define __ASM_SH_SYSTEM_H
 
 /*
- * Copyright (C) 1999, 2000  Niibe Yutaka
+ * Copyright (C) 1999, 2000  Niibe Yutaka  &  Kaz Kojima
  */
 
 /*
@@ -93,13 +93,15 @@ extern void __xchg_called_with_bad_pointer(void);
 /* Interrupt Control */
 extern __inline__ void __sti(void)
 {
-	unsigned long __dummy;
+	unsigned long __dummy0, __dummy1;
 
 	__asm__ __volatile__("stc	$sr, %0\n\t"
 			     "and	%1, %0\n\t"
+			     "stc	$r5_bank, %1\n\t"
+			     "or	%1, %0\n\t"
 			     "ldc	%0, $sr"
-			     : "=&r" (__dummy)
-			     : "r" (0xefffffff)
+			     : "=&r" (__dummy0), "=r" (__dummy1)
+			     : "1" (~0xf0)
 			     : "memory");
 }
 
@@ -107,10 +109,10 @@ extern __inline__ void __cli(void)
 {
 	unsigned long __dummy;
 	__asm__ __volatile__("stc	$sr, %0\n\t"
-			     "or	%1, %0\n\t"
+			     "or	#0xf0, %0\n\t"
 			     "ldc	%0, $sr"
-			     : "=&r" (__dummy)
-			     : "r" (0x10000000)
+			     : "=&z" (__dummy)
+			     : /* no inputs */
 			     : "memory");
 }
 
@@ -118,35 +120,72 @@ extern __inline__ void __cli(void)
 x = (__extension__ ({	unsigned long __sr;	\
 	__asm__ __volatile__(			\
 		"stc	$sr, %0"		\
-		: "=r" (__sr)			\
+		: "=&r" (__sr)			\
 		: /* no inputs */		\
 		: "memory");			\
-	 (__sr & 0xffff7f0f);}))
+	 (__sr & 0x000000f0);}))
 
 #define __save_and_cli(x)    				\
 x = (__extension__ ({	unsigned long __dummy,__sr;	\
 	__asm__ __volatile__(                   	\
 		"stc	$sr, %1\n\t" 			\
-		"or	%0, %1\n\t" 			\
-		"stc	$sr, %0\n\t" 			\
-		"ldc	%1, $sr"     			\
-		: "=r" (__sr), "=&r" (__dummy) 		\
-		: "0" (0x10000000) 			\
-		: "memory"); (__sr & 0xffff7f0f); }))
+		"mov	%1, %0\n\t" 			\
+		"or	#0xf0, %0\n\t" 			\
+		"ldc	%0, $sr"     			\
+		: "=&z" (__dummy), "=&r" (__sr)		\
+		: /* no inputs */ 			\
+		: "memory"); (__sr & 0x000000f0); }))
 
 #define __restore_flags(x) do { 			\
-	unsigned long __dummy;				\
-	__asm__ __volatile__(				\
+	unsigned long __dummy0, __dummy1;		\
+	if (x != 0xf0)	/* not CLI-ed? */		\
+		__asm__ __volatile__(		       	\
 		"stc	$sr, %0\n\t"			\
 		"and	%1, %0\n\t"			\
-		"or	%2, %0\n\t"			\
+		"stc	$r5_bank, %1\n\t"		\
+		"or	%1, %0\n\t"			\
 		"ldc	%0, $sr"			\
-		: "=&r" (__dummy)			\
-		: "r" (0x000080f0), /* IMASK+FD */	\
-		  "r" (x)		 		\
+		: "=&r" (__dummy0), "=r" (__dummy1)	\
+		: "1" (0xffffff0f)	 		\
 		: "memory"); 				\
 } while (0)
 
+/*
+ * Jump to P2 area.
+ * When handling TLB or caches, we need to do it from P2 area.
+ */
+#define jump_to_P2()			\
+do {					\
+	unsigned long __dummy;		\
+	__asm__ __volatile__(		\
+		"mov.l	1f, %0\n\t"	\
+		"or	%1, %0\n\t"	\
+		"jmp	@%0\n\t"	\
+		" nop\n\t" 		\
+		".balign 4\n"		\
+		"1:	.long 2f\n"	\
+		"2:"			\
+		: "=&r" (__dummy)	\
+		: "r" (0x20000000));	\
+} while (0)
+
+/*
+ * Back to P1 area.
+ */
+#define back_to_P1()				\
+do {						\
+	unsigned long __dummy;			\
+	__asm__ __volatile__(			\
+		"nop;nop;nop;nop;nop;nop\n\t"	\
+		"mov.l	1f, %0\n\t"		\
+		"jmp	@%0\n\t"		\
+		" nop\n\t"			\
+		".balign 4\n"			\
+		"1:	.long 2f\n"		\
+		"2:"				\
+		: "=&r" (__dummy));		\
+} while (0)
+
 /* For spinlocks etc */
 #define local_irq_save(x)	__save_and_cli(x)
 #define local_irq_restore(x)	__restore_flags(x)
diff --git a/include/asm-sh/uaccess.h b/include/asm-sh/uaccess.h
index b64de2cd5..65ed58fd8 100644
--- a/include/asm-sh/uaccess.h
+++ b/include/asm-sh/uaccess.h
@@ -1,4 +1,4 @@
-/* $Id: uaccess.h,v 1.6 1999/10/29 13:10:44 gniibe Exp $
+/* $Id: uaccess.h,v 1.10 2000/03/24 13:53:45 gniibe Exp $
  *
  * User space memory access functions
  *
@@ -47,7 +47,7 @@
  */
 #define __range_ok(addr,size) ({ \
 	unsigned long flag,sum; \
-	__asm__("clrt; addc %3,%1; movt %0; cmp/hi %4,%1; rotcl %0" \
+	__asm__("clrt; addc %3, %1; movt %0; cmp/hi %4, %1; rotcl %0" \
 		:"=&r" (flag), "=r" (sum) \
 		:"1" (addr), "r" ((int)(size)), "r" (current->addr_limit.seg)); \
 	flag; })
@@ -137,17 +137,17 @@ default: __get_user_unknown(); break; \
 ({ \
 __asm__ __volatile__( \
 	"1:\n\t" \
-	"mov." insn "	%2,%1\n\t" \
-	"mov	#0,%0\n" \
-	"2:\n\t" \
+	"mov." insn "	%2, %1\n\t" \
+	"mov	#0, %0\n" \
+	"2:\n" \
 	".section	.fixup,\"ax\"\n" \
 	"3:\n\t" \
-	"mov	#0,%1\n\t" \
-	"mov.l	4f,%0\n\t" \
+	"mov	#0, %1\n\t" \
+	"mov.l	4f, %0\n\t" \
 	"jmp	@%0\n\t" \
-	" mov	%3,%0\n" \
+	" mov	%3, %0\n" \
 	"4:	.long	2b\n\t" \
-	".previous\n\t" \
+	".previous\n" \
 	".section	__ex_table,\"a\"\n\t" \
 	".long	1b, 3b\n\t" \
 	".previous" \
@@ -189,17 +189,17 @@ default: __put_user_unknown(); break; \
 ({ \
 __asm__ __volatile__( \
 	"1:\n\t" \
-	"mov." insn "	%1,%2\n\t" \
-	"mov	#0,%0\n" \
-	"2:\n\t" \
+	"mov." insn "	%1, %2\n\t" \
+	"mov	#0, %0\n" \
+	"2:\n" \
 	".section	.fixup,\"ax\"\n" \
 	"3:\n\t" \
 	"nop\n\t" \
-	"mov.l	4f,%0\n\t" \
+	"mov.l	4f, %0\n\t" \
 	"jmp	@%0\n\t" \
-	"mov	%3,%0\n" \
+	"mov	%3, %0\n" \
 	"4:	.long	2b\n\t" \
-	".previous\n\t" \
+	".previous\n" \
 	".section	__ex_table,\"a\"\n\t" \
 	".long	1b, 3b\n\t" \
 	".previous" \
@@ -218,18 +218,18 @@ __copy_user(void *__to, const void *__from, __kernel_size_t __n)
 
 	__asm__ __volatile__(
 		"9:\n\t"
-		"mov.b	@%2+,%1\n\t"
+		"mov.b	@%2+, %1\n\t"
 		"dt	%0\n"
 		"1:\n\t"
-		"mov.b	%1,@%3\n\t"
+		"mov.b	%1, @%3\n\t"
 		"bf/s	9b\n\t"
-		" add	#1,%3\n"
-		"2:"
+		" add	#1, %3\n"
+		"2:\n"
 		".section .fixup,\"ax\"\n"
 		"3:\n\t"
-		"mov.l	5f,%1\n\t"
+		"mov.l	5f, %1\n\t"
 		"jmp	@%1\n\t"
-		" mov	%7,%0\n\t"
+		" mov	%7, %0\n\t"
 		".balign 4\n"
 		"5:	.long 2b\n"
 		".previous\n"
@@ -238,7 +238,7 @@ __copy_user(void *__to, const void *__from, __kernel_size_t __n)
 		"	.long 9b,3b\n"
 		"	.long 1b,2b\n"
 		".previous"
-		: "=&r" (res), "=&z" (__dummy), "=&r" (_f), "=&r" (_t)
+		: "=r" (res), "=&z" (__dummy), "=r" (_f), "=r" (_t)
 		: "2" (__from), "3" (__to), "0" (__n), "i" (-EFAULT)
 		: "memory");
 
@@ -297,23 +297,21 @@ if (__copy_from_user(to,from,n)) \
 extern __inline__ __kernel_size_t
 __clear_user(void *addr, __kernel_size_t size)
 {
-	__kernel_size_t res;
-	unsigned long __a, __s;
+	unsigned long __a;
 
 	__asm__ __volatile__(
 		"9:\n\t"
-		"dt	%2\n"
+		"dt	%0\n"
 		"1:\n\t"
-		"mov.b	%5,@%1\n\t"
+		"mov.b	%4, @%1\n\t"
 		"bf/s	9b\n\t"
-		" add	#1,%1\n\t"
-		"sub	%2,%0\n"
+		" add	#1, %1\n"
 		"2:\n"
 		".section .fixup,\"ax\"\n"
 		"3:\n\t"
-		"mov.l	4f,%0\n\t"
-		"jmp	@%0\n\t"
-		" mov	%7,%0\n"
+		"mov.l	4f, %1\n\t"
+		"jmp	@%1\n\t"
+		" nop\n"
 		".balign 4\n"
 		"4:	.long 2b\n"
 		".previous\n"
@@ -321,10 +319,10 @@ __clear_user(void *addr, __kernel_size_t size)
 		"	.balign 4\n"
 		"	.long 1b,3b\n"
 		".previous"
-		: "=&r" (res), "=&r" (__a), "=&r" (__s)
-		: "1" (addr), "2" (size), "r" (0), "0" (size), "i" (-EFAULT));
+		: "=r" (size), "=r" (__a)
+		: "0" (size), "1" (addr), "r" (0));
 
-	return res;
+	return size;
 }
 
 #define clear_user(addr,n) ({ \
@@ -342,31 +340,31 @@ __strncpy_from_user(unsigned long __dest, unsigned long __src, int __count)
 
 	__asm__ __volatile__(
 		"9:\n"
-		"mov.b	@%2+,%1\n\t"
-		"cmp/eq	#0,%1\n\t"
+		"mov.b	@%2+, %1\n\t"
+		"cmp/eq	#0, %1\n\t"
 		"bt/s	2f\n"
 		"1:\n"
-		"mov.b	%1,@%3\n\t"
-		"dt	%0\n\t"
+		"mov.b	%1, @%3\n\t"
+		"dt	%7\n\t"
 		"bf/s	9b\n\t"
-		" add	#1,%3\n\t"
-		"sub	%6,%0\n"
-		"2:\n"
+		" add	#1, %3\n\t"
+		"2:\n\t"
+		"sub	%7, %0\n"
+		"3:\n"
 		".section .fixup,\"ax\"\n"
-		"3:\n\t"
-		"mov.l	4f,%1\n\t"
+		"4:\n\t"
+		"mov.l	5f, %1\n\t"
 		"jmp	@%1\n\t"
-		" mov	%8,%0\n\t"
+		" mov	%8, %0\n\t"
 		".balign 4\n"
-		"4:	.long 2b\n"
+		"5:	.long 3b\n"
 		".previous\n"
 		".section __ex_table,\"a\"\n"
 		"	.balign 4\n"
-		"	.long 9b,3b\n"
-		"	.long 1b,2b\n"
+		"	.long 9b,4b\n"
 		".previous"
-		: "=&r" (res), "=&z" (__dummy), "=&r" (_s), "=&r" (_d)
-		: "2" (__src), "3" (__dest), "r" (__count), "0" (__count),
+		: "=r" (res), "=&z" (__dummy), "=r" (_s), "=r" (_d)
+		: "0" (__count), "2" (__src), "3" (__dest), "r" (__count),
 		  "i" (-EFAULT)
 		: "memory");
 
@@ -393,19 +391,19 @@ extern __inline__ long __strnlen_user(const char *__s, long __n)
 
 	__asm__ __volatile__(
 		"9:\n"
-		"cmp/eq	%4,%0\n\t"
+		"cmp/eq	%4, %0\n\t"
 		"bt	2f\n"
 		"1:\t"
-		"mov.b	@(%0,%3),%1\n\t"
-		"tst	%1,%1\n\t"
+		"mov.b	@(%0,%3), %1\n\t"
+		"tst	%1, %1\n\t"
 		"bf/s	9b\n\t"
-		" add	#1,%0\n"
+		" add	#1, %0\n"
 		"2:\n"
 		".section .fixup,\"ax\"\n"
 		"3:\n\t"
-		"mov.l	4f,%1\n\t"
+		"mov.l	4f, %1\n\t"
 		"jmp	@%1\n\t"
-		" mov	%5,%0\n"
+		" mov	%5, %0\n"
 		".balign 4\n"
 		"4:	.long 2b\n"
 		".previous\n"
diff --git a/include/asm-sh/unistd.h b/include/asm-sh/unistd.h
index 20f5f659b..baaff1798 100644
--- a/include/asm-sh/unistd.h
+++ b/include/asm-sh/unistd.h
@@ -227,6 +227,8 @@
 #define __NR_setfsuid32		215
 #define __NR_setfsgid32		216
 #define __NR_pivot_root		217
+#define __NR_mincore		218
+#define __NR_madvise		219
 
 /* user-visible error numbers are in the range -1 - -125: see <asm-sh/errno.h> */
 
diff --git a/include/asm-sh/user.h b/include/asm-sh/user.h
index 243cc04fd..6888368aa 100644
--- a/include/asm-sh/user.h
+++ b/include/asm-sh/user.h
@@ -30,16 +30,16 @@
  */
 
 struct user_fpu_struct {
-	unsigned long fp_regs[NUM_FPU_REGS];
-	unsigned long xf_regs[NUM_FPU_REGS];
+	unsigned long fp_regs[16];
+	unsigned long long xd_regs[8];
 	unsigned long fpscr;
 	unsigned long fpul;
 };
 
 struct user {
 	struct pt_regs	regs;			/* entire machine state */
-	struct user_fpu_struct fpu;	/* Math Co-processor registers. */
-	int u_fpvalid;		/* True if math co-processor being used. */
+	struct user_fpu_struct fpu;	/* Math Co-processor registers  */
+	int u_fpvalid;		/* True if math co-processor being used */
 	size_t		u_tsize;		/* text size (pages) */
 	size_t		u_dsize;		/* data size (pages) */
 	size_t		u_ssize;		/* stack size (pages) */
@@ -48,7 +48,7 @@ struct user {
 	unsigned long	start_stack;		/* stack starting address */
 	long int	signal;			/* signal causing core dump */
 	struct regs *	u_ar0;			/* help gdb find registers */
-	struct user_fpu_struct* u_fpstate;	/* Math Co-processor pointer. */
+	struct user_fpu_struct* u_fpstate;	/* Math Co-processor pointer */
 	unsigned long	magic;			/* identifies a core file */
 	char		u_comm[32];		/* user command name */
 };