o New memset. Fastest in town for size > 6 bytes.

 o New clear_user.
 o Memcpy now efficiently copies the (src^dest)&3 != 0.
 o Memmove new correctly deals with overlaps
 o Rewrite csum_partial in assembler.
 o Rewrte csum_partial_from_user in assembler.
 o __copy_user is now integrated with memcpy.
 o get_user now returns a zero value on error.
 o copy_from_user now clears the destination buffer on error.
 o strncpy_user now has a more efficient caller routine.
 o strlen_user now has a more efficient caller routines and is faster.
 o The unaligned handler is now much cleaner.  It's now also save from
   interrupt.  Some more esotheric bugs fixed as well.
 o Don't export bcopy anymore, it's now a inline function.
 o Delete ancient junk from the first days of Linux/MIPS.
 o Delete dead code in indy_sc.c.
 o Including the IDE driver doesn't crash an Indy anymore.
 o Eleminate active_ds.  We now use current_ds directly in the thread
   structure which is faster and threadsafe.  Saves almost 2kb on the
   kernel.
 o Serial console should work again.

8 files changed, 309 insertions, 208 deletions

diff --git a/include/asm-mips/branch.h b/include/asm-mips/branch.h
index 08dda213e..d8882cfb1 100644
--- a/include/asm-mips/branch.h
+++ b/include/asm-mips/branch.h
@@ -5,7 +5,9 @@
  * License.  See the file "COPYING" in the main directory of this archive
  * for more details.
  *
- * Copyright (C) 1996, 1997 by Ralf Baechle
+ * Copyright (C) 1996, 1997, 1998 by Ralf Baechle
+ *
+ * $Id: branch.h,v 1.2 1998/04/28 19:37:46 ralf Exp $
  */
 #include <asm/ptrace.h>
 
@@ -15,12 +17,13 @@ extern inline int delay_slot(struct pt_regs *regs)
 }
 
 extern int __compute_return_epc(struct pt_regs *regs);
+
 extern inline int compute_return_epc(struct pt_regs *regs)
 {
-	if (delay_slot(regs)) {
-		return __compute_return_epc(regs);
+	if (!delay_slot(regs)) {
+		regs->cp0_epc += 4;
+		return 0;
 	}
 
-	regs->cp0_epc += 4;
-	return 0;
+	return __compute_return_epc(regs);
 }
diff --git a/include/asm-mips/ide.h b/include/asm-mips/ide.h
index bda27206d..b54a7657c 100644
--- a/include/asm-mips/ide.h
+++ b/include/asm-mips/ide.h
@@ -21,40 +21,36 @@ typedef unsigned short ide_ioreg_t;
 
 #define ide_sti()	sti()
 
+struct ide_ops {
+	int (*ide_default_irq)(ide_ioreg_t base);
+	ide_ioreg_t (*ide_default_io_base)(int index);
+	void (*ide_init_hwif_ports)(ide_ioreg_t *p, ide_ioreg_t base, int *irq);
+	int (*ide_request_irq)(unsigned int irq, void (*handler)(int, void *,
+	                       struct pt_regs *), unsigned long flags,
+	                       const char *device, void *dev_id);
+	void (*ide_free_irq)(unsigned int irq, void *dev_id);
+	int (*ide_check_region) (ide_ioreg_t from, unsigned int extent);
+	void (*ide_request_region)(ide_ioreg_t from, unsigned int extent,
+	                        const char *name);
+	void (*ide_release_region)(ide_ioreg_t from, unsigned int extent);
+};
+
+extern struct ide_ops *ide_ops;
+
 static __inline__ int ide_default_irq(ide_ioreg_t base)
 {
-	switch (base) {
-		case 0x1f0: return 14;
-		case 0x170: return 15;
-		case 0x1e8: return 11;
-		case 0x168: return 10;
-		default:
-			return 0;
-	}
+	return ide_ops->ide_default_irq(base);
 }
 
 static __inline__ ide_ioreg_t ide_default_io_base(int index)
 {
-	switch (index) {
-		case 0:	return 0x1f0;
-		case 1:	return 0x170;
-		case 2: return 0x1e8;
-		case 3: return 0x168;
-		default:
-			return 0;
-	}
+	return ide_ops->ide_default_io_base(index);
 }
 
-static __inline__ void ide_init_hwif_ports (ide_ioreg_t *p, ide_ioreg_t base, int *irq)
+static __inline__ void ide_init_hwif_ports(ide_ioreg_t *p, ide_ioreg_t base,
+                                           int *irq)
 {
-	ide_ioreg_t port = base;
-	int i = 8;
-
-	while (i--)
-		*p++ = port++;
-	*p++ = base + 0x206;
-	if (irq != NULL)
-		*irq = 0;
+	ide_ops->ide_init_hwif_ports(p, base, irq);
 }
 
 typedef union {
@@ -68,38 +64,41 @@ typedef union {
 	} b;
 	} select_t;
 
-static __inline__ int ide_request_irq(unsigned int irq, void (*handler)(int, void *, struct pt_regs *),
+static __inline__ int ide_request_irq(unsigned int irq, void (*handler)(int,void *, struct pt_regs *),
 			unsigned long flags, const char *device, void *dev_id)
 {
-	return request_irq(irq, handler, flags, device, dev_id);
-}			
+	return ide_ops->ide_request_irq(irq, handler, flags, device, dev_id);
+}
 
 static __inline__ void ide_free_irq(unsigned int irq, void *dev_id)
 {
-	free_irq(irq, dev_id);
+	ide_ops->ide_free_irq(irq, dev_id);
 }
 
 static __inline__ int ide_check_region (ide_ioreg_t from, unsigned int extent)
 {
-	return check_region(from, extent);
+	return ide_ops->ide_check_region(from, extent);
 }
 
-static __inline__ void ide_request_region (ide_ioreg_t from, unsigned int extent, const char *name)
+static __inline__ void ide_request_region(ide_ioreg_t from,
+                                          unsigned int extent, const char *name)
 {
-	request_region(from, extent, name);
+	ide_ops->ide_request_region(from, extent, name);
 }
 
-static __inline__ void ide_release_region (ide_ioreg_t from, unsigned int extent)
+static __inline__ void ide_release_region(ide_ioreg_t from,
+                                          unsigned int extent)
 {
-	release_region(from, extent);
+	ide_ops->ide_release_region(from, extent);
 }
 
 /*
  * The following are not needed for the non-m68k ports
  */
-static __inline__ int ide_ack_intr (ide_ioreg_t status_port, ide_ioreg_t irq_port)
+static __inline__ int ide_ack_intr (ide_ioreg_t status_port,
+                                    ide_ioreg_t irq_port)
 {
-	return(1);
+	return 1;
 }
 
 static __inline__ void ide_fix_driveid(struct hd_driveid *id)
@@ -110,7 +109,10 @@ static __inline__ void ide_release_lock (int *ide_lock)
 {
 }
 
-static __inline__ void ide_get_lock (int *ide_lock, void (*handler)(int, void *, struct pt_regs *), void *data)
+static __inline__ void ide_get_lock (int *ide_lock,
+                                     void (*handler)(int, void *,
+                                                    struct pt_regs *),
+                                     void *data)
 {
 }
 
diff --git a/include/asm-mips/offset.h b/include/asm-mips/offset.h
index e211de3cc..32deafe7b 100644
--- a/include/asm-mips/offset.h
+++ b/include/asm-mips/offset.h
@@ -50,7 +50,7 @@
 #define TASK_PRIORITY      56
 #define TASK_FLAGS         4
 #define TASK_SIGPENDING    8
-#define TASK_MM            920
+#define TASK_MM            928
 
 /* MIPS specific thread_struct offsets. */
 #define THREAD_REG16   568
@@ -67,13 +67,14 @@
 #define THREAD_STATUS  612
 #define THREAD_FPU     616
 #define THREAD_BVADDR  880
-#define THREAD_ECODE   884
-#define THREAD_TRAPNO  888
-#define THREAD_PGDIR   892
-#define THREAD_MFLAGS  896
-#define THREAD_CURDS   900
-#define THREAD_TRAMP   904
-#define THREAD_OLDCTX  908
+#define THREAD_BUADDR  884
+#define THREAD_ECODE   888
+#define THREAD_TRAPNO  892
+#define THREAD_PGDIR   896
+#define THREAD_MFLAGS  900
+#define THREAD_CURDS   904
+#define THREAD_TRAMP   908
+#define THREAD_OLDCTX  912
 
 /* Linux mm_struct offsets. */
 #define MM_COUNT      12
diff --git a/include/asm-mips/processor.h b/include/asm-mips/processor.h
index 765208d3e..98d5e6a86 100644
--- a/include/asm-mips/processor.h
+++ b/include/asm-mips/processor.h
@@ -5,7 +5,7 @@
  * written by Ralf Baechle
  * Modified further for R[236]000 compatibility by Paul M. Antoine
  *
- * $Id: processor.h,v 1.8 1998/03/27 04:47:59 ralf Exp $
+ * $Id: processor.h,v 1.13 1998/04/25 05:35:15 ralf Exp $
  */
 #ifndef __ASM_MIPS_PROCESSOR_H
 #define __ASM_MIPS_PROCESSOR_H
@@ -90,7 +90,7 @@ typedef struct {
  */
 struct thread_struct {
         /* Saved main processor registers. */
-        unsigned long reg16 __attribute__ ((aligned (8)));
+        unsigned long reg16;
 	unsigned long reg17, reg18, reg19, reg20, reg21, reg22, reg23;
         unsigned long reg29, reg30, reg31;
 
@@ -98,10 +98,11 @@ struct thread_struct {
 	unsigned long cp0_status;
 
 	/* Saved fpu/fpu emulator stuff. */
-	union mips_fpu_union fpu __attribute__ ((aligned (8)));
+	union mips_fpu_union fpu;
 
 	/* Other stuff associated with the thread. */
-	unsigned long cp0_badvaddr;
+	unsigned long cp0_badvaddr;	/* Last user fault */
+	unsigned long cp0_baduaddr;	/* Last kernel fault accessing USEG */
 	unsigned long error_code;
 	unsigned long trap_no;
 	unsigned long pg_dir;                   /* used in tlb refill    */
@@ -135,7 +136,7 @@ struct thread_struct {
 	/* \
 	 * Other stuff associated with the process \
 	 */ \
-	0, 0, 0, (unsigned long) swapper_pg_dir, \
+	0, 0, 0, 0, (unsigned long) swapper_pg_dir, \
 	/* \
 	 * For now the default is to fix address errors \
 	 */ \
diff --git a/include/asm-mips/stackframe.h b/include/asm-mips/stackframe.h
index e925dcd50..726e82f23 100644
--- a/include/asm-mips/stackframe.h
+++ b/include/asm-mips/stackframe.h
@@ -3,7 +3,7 @@
  *
  *  Copyright (C) 1994, 1995, 1996 by Ralf Baechle and Paul M. Antoine.
  *
- * $Id: stackframe.h,v 1.6 1998/03/26 07:39:21 ralf Exp $
+ * $Id: stackframe.h,v 1.7 1998/04/28 19:39:15 ralf Exp $
  */
 #ifndef __ASM_MIPS_STACKFRAME_H
 #define __ASM_MIPS_STACKFRAME_H
@@ -174,10 +174,6 @@
 /*
  * Move to kernel mode and enable interrupts.
  * Set cp0 enable bit as sign that we're running on the kernel stack
- *
- * Note that the mtc0 will be effective on R4000 pipeline stage 7. This
- * means that another three instructions will be executed with interrupts
- * disabled.  Arch/mips/mips3/r4xx0.S makes use of this fact.
  */
 #define STI                                             \
 		mfc0	t0,CP0_STATUS;                  \
@@ -186,4 +182,15 @@
 		xori	t0,0x1e;                        \
 		mtc0	t0,CP0_STATUS
 
+/*
+ * Just move to kernel mode and leave interrupts as they are.
+ * Set cp0 enable bit as sign that we're running on the kernel stack
+ */
+#define KMODE                                           \
+		mfc0	t0,CP0_STATUS;                  \
+		li	t1,ST0_CU0|0x1e;                \
+		or	t0,t1;                          \
+		xori	t0,0x1e;                        \
+		mtc0	t0,CP0_STATUS
+
 #endif /* __ASM_MIPS_STACKFRAME_H */
diff --git a/include/asm-mips/string.h b/include/asm-mips/string.h
index 39e6b445c..7f4bded24 100644
--- a/include/asm-mips/string.h
+++ b/include/asm-mips/string.h
@@ -7,7 +7,7 @@
  *
  * Copyright (c) 1994, 1995, 1996, 1997 by Ralf Baechle
  *
- * $Id: string.h,v 1.4 1998/03/21 19:31:09 ralf Exp $
+ * $Id: string.h,v 1.7 1998/03/25 00:24:10 ralf Exp $
  */
 #ifndef __ASM_MIPS_STRING_H
 #define __ASM_MIPS_STRING_H
@@ -128,7 +128,10 @@ extern void *memcpy(void *__to, __const__ void *__from, size_t __n);
 extern void *memmove(void *__dest, __const__ void *__src, size_t __n);
 
 #define __HAVE_ARCH_BCOPY
-extern char * bcopy(const char * src, char * dest, int count);
+extern __inline__ char * bcopy(const char * src, char * dest, int count)
+{
+	memmove(dest, src, count);
+}
 
 #define __HAVE_ARCH_MEMSCAN
 extern __inline__ void *memscan(void *__addr, int __c, size_t __size)
diff --git a/include/asm-mips/system.h b/include/asm-mips/system.h
index abeaa3343..ab8faf278 100644
--- a/include/asm-mips/system.h
+++ b/include/asm-mips/system.h
@@ -7,6 +7,8 @@
  *
  * Copyright (C) 1994, 1995 by Ralf Baechle
  * Modified further for R[236]000 by Paul M. Antoine, 1996
+ *
+ * $Id: system.h,v 1.6 1998/05/03 11:13:54 ralf Exp $
  */
 #ifndef __ASM_MIPS_SYSTEM_H
 #define __ASM_MIPS_SYSTEM_H
@@ -17,18 +19,18 @@
 extern __inline__ void
 __sti(void)
 {
-    __asm__ __volatile__(
-	".set\tnoreorder\n\t"
-	".set\tnoat\n\t"
-	"mfc0\t$1,$12\n\t"
-	"ori\t$1,0x1f\n\t"
-	"xori\t$1,0x1e\n\t"
-	"mtc0\t$1,$12\n\t"
-	".set\tat\n\t"
-	".set\treorder"
-	: /* no outputs */
-	: /* no inputs */
-	: "$1", "memory");
+	__asm__ __volatile__(
+		".set\tnoreorder\n\t"
+		".set\tnoat\n\t"
+		"mfc0\t$1,$12\n\t"
+		"ori\t$1,0x1f\n\t"
+		"xori\t$1,0x1e\n\t"
+		"mtc0\t$1,$12\n\t"
+		".set\tat\n\t"
+		".set\treorder"
+		: /* no outputs */
+		: /* no inputs */
+		: "$1", "memory");
 }
 
 /*
@@ -41,21 +43,21 @@ __sti(void)
 extern __inline__ void
 __cli(void)
 {
-    __asm__ __volatile__(
-	".set\tnoreorder\n\t"
-	".set\tnoat\n\t"
-	"mfc0\t$1,$12\n\t"
-	"ori\t$1,1\n\t"
-	"xori\t$1,1\n\t"
-	"mtc0\t$1,$12\n\t"
-	"nop\n\t"
-	"nop\n\t"
-	"nop\n\t"
-	".set\tat\n\t"
-	".set\treorder"
-	: /* no outputs */
-	: /* no inputs */
-	: "$1", "memory");
+	__asm__ __volatile__(
+		".set\tnoreorder\n\t"
+		".set\tnoat\n\t"
+		"mfc0\t$1,$12\n\t"
+		"ori\t$1,1\n\t"
+		"xori\t$1,1\n\t"
+		"mtc0\t$1,$12\n\t"
+		"nop\n\t"
+		"nop\n\t"
+		"nop\n\t"
+		".set\tat\n\t"
+		".set\treorder"
+		: /* no outputs */
+		: /* no inputs */
+		: "$1", "memory");
 }
 
 #define __save_flags(x)                  \
@@ -87,16 +89,16 @@ __asm__ __volatile__(                    \
 extern void __inline__
 __restore_flags(int flags)
 {
-    __asm__ __volatile__(
-	".set\tnoreorder\n\t"
-	"mtc0\t%0,$12\n\t"
-	"nop\n\t"
-	"nop\n\t"
-	"nop\n\t"
-	".set\treorder"
-	: /* no output */
-	: "r" (flags)
-	: "memory");
+	__asm__ __volatile__(
+		".set\tnoreorder\n\t"
+		"mtc0\t%0,$12\n\t"
+		"nop\n\t"
+		"nop\n\t"
+		"nop\n\t"
+		".set\treorder"
+		: /* no output */
+		: "r" (flags)
+		: "memory");
 }
 
 /*
@@ -113,7 +115,7 @@ __asm__ __volatile__(					\
 	"# prevent instructions being moved around\n\t"	\
 	".set\tnoreorder\n\t"				\
 	".set\treorder"					\
-        : /* no output */				\
+	: /* no output */				\
 	: /* no input */				\
 	: "memory")
 
@@ -125,47 +127,12 @@ __asm__ __volatile__(					\
 extern asmlinkage void (*resume)(void *tsk);
 #endif /* !defined (__LANGUAGE_ASSEMBLY__) */
 
-/*
- * FIXME: resume() assumes current == prev
- */
 #define switch_to(prev,next) \
 do { \
-	prev->tss.current_ds = active_ds; \
-        active_ds = next->tss.current_ds; \
-        resume(next); \
+	resume(next); \
 } while(0)
 
 /*
- * The 8 and 16 bit variants have to disable interrupts temporarily.
- * Both are currently unused.
- */
-extern __inline__ unsigned long xchg_u8(volatile char * m, unsigned long val)
-{
-	unsigned long flags, retval;
-
-	save_flags(flags);
-	cli();
-	retval = *m;
-	*m = val;
-	restore_flags(flags);
-
-	return retval;
-}
-
-extern __inline__ unsigned long xchg_u16(volatile short * m, unsigned long val)
-{
-	unsigned long flags, retval;
-
-	save_flags(flags);
-	cli();
-	retval = *m;
-	*m = val;
-	restore_flags(flags);
-
-	return retval;
-}
-
-/*
  * For 32 and 64 bit operands we can take advantage of ll and sc.
  * FIXME: This doesn't work for R3000 machines.
  */
@@ -186,8 +153,9 @@ extern __inline__ unsigned long xchg_u32(volatile int * m, unsigned long val)
 		".set\tat\n\t"
 		".set\treorder"
 		: "=r" (val), "=r" (m), "=r" (dummy)
-		: "1" (m), "2" (val));
-#else /* FIXME: Brain-dead approach, but then again, I AM hacking - PMA */
+		: "1" (m), "2" (val)
+		: "memory");
+#else
 	unsigned long flags, retval;
 
 	save_flags(flags);
@@ -218,7 +186,8 @@ extern __inline__ unsigned long xchg_u64(volatile long * m, unsigned long val)
 		".set\tat\n\t"
 		".set\treorder"
 		: "=r" (val), "=r" (m), "=r" (dummy)
-		: "1" (m), "2" (val));
+		: "1" (m), "2" (val)
+		: "memory");
 
 	return val;
 }
@@ -239,14 +208,12 @@ extern void __xchg_called_with_bad_pointer(void);
 static __inline__ unsigned long __xchg(unsigned long x, volatile void * ptr, int size)
 {
 	switch (size) {
-		case 1:
-			return xchg_u8(ptr, x);
-		case 2:
-			return xchg_u16(ptr, x);
 		case 4:
 			return xchg_u32(ptr, x);
+#if defined(__mips64)
 		case 8:
 			return xchg_u64(ptr, x);
+#endif
 	}
 	__xchg_called_with_bad_pointer();
 	return x;
diff --git a/include/asm-mips/uaccess.h b/include/asm-mips/uaccess.h
index 42a7df686..8b139fe23 100644
--- a/include/asm-mips/uaccess.h
+++ b/include/asm-mips/uaccess.h
@@ -5,16 +5,15 @@
  * License.  See the file "COPYING" in the main directory of this archive
  * for more details.
  *
- * Copyright (C) 1996, 1997 by Ralf Baechle
+ * Copyright (C) 1996, 1997, 1998 by Ralf Baechle
  *
- * $Id: uaccess.h,v 1.7 1998/03/21 08:04:33 ralf Exp $
+ * $Id: uaccess.h,v 1.15 1998/05/03 11:13:54 ralf Exp $
  */
 #ifndef __ASM_MIPS_UACCESS_H
 #define __ASM_MIPS_UACCESS_H
 
 #include <linux/errno.h>
 #include <linux/sched.h>
-#include <asm/asm.h>
 
 #define STR(x)  __STR(x)
 #define __STR(x)  #x
@@ -32,11 +31,9 @@
 #define VERIFY_READ    0
 #define VERIFY_WRITE   1
 
-extern mm_segment_t active_ds;
-
-#define get_fs()        (active_ds)
+#define get_fs()        (current->tss.current_ds)
 #define get_ds()	(KERNEL_DS)
-#define set_fs(x)       (active_ds=(x))
+#define set_fs(x)       (current->tss.current_ds=(x))
 
 #define segment_eq(a,b)	((a).seg == (b).seg)
 
@@ -107,24 +104,6 @@ if (__get_user(x,ptr)) return ret; })
 struct __large_struct { unsigned long buf[100]; };
 #define __m(x) (*(struct __large_struct *)(x))
 
-#define copy_to_user(to,from,n)   __copy_tofrom_user((to),(from),(n),__cu_to)
-#define copy_from_user(to,from,n) __copy_tofrom_user((to),(from),(n),__cu_from)
-
-extern size_t __copy_user(void *__to, const void *__from, size_t __n);
-
-#define __copy_tofrom_user(to,from,n,v) ({ \
-	void * __cu_to; \
-	const void * __cu_from; \
-	long __cu_len; \
-	\
-	__cu_to = (to); \
-	__cu_from = (from); \
-	__cu_len = (n); \
-	if (__access_ok(((unsigned long)(v)),__cu_len,__access_mask)) \
-		__cu_len = __copy_user(__cu_to, __cu_from, __cu_len); \
-	__cu_len; \
-})
-
 /*
  * Yuck.  We need two variants, one for 64bit operation and one
  * for 32 bit mode and old iron.
@@ -174,10 +153,11 @@ __asm__ __volatile__( \
 	"2:\n\t" \
 	".section\t.fixup,\"ax\"\n" \
 	"3:\tli\t%0,%3\n\t" \
+	"move\t%1,$0\n\t" \
 	"j\t2b\n\t" \
 	".previous\n\t" \
 	".section\t__ex_table,\"a\"\n\t" \
-	STR(PTR)"\t1b,3b\n\t" \
+	".word\t1b,3b\n\t" \
 	".previous" \
 	:"=r" (__gu_err), "=r" (__gu_val) \
 	:"o" (__m(__gu_addr)), "i" (-EFAULT)); })
@@ -193,11 +173,13 @@ __asm__ __volatile__( \
 	"move\t%0,$0\n" \
 	"3:\t.section\t.fixup,\"ax\"\n" \
 	"4:\tli\t%0,%4\n\t" \
+	"move\t%1,$0\n\t" \
+	"move\t%D1,$0\n\t" \
 	"j\t3b\n\t" \
 	".previous\n\t" \
 	".section\t__ex_table,\"a\"\n\t" \
-	STR(PTR)"\t1b,4b\n\t" \
-	STR(PTR)"\t2b,4b\n\t" \
+	".word\t1b,4b\n\t" \
+	".word\t2b,4b\n\t" \
 	".previous" \
 	:"=r" (__gu_err), "=&r" (__gu_val) \
 	:"o" (__m(__gu_addr)), "o" (__m(__gu_addr + 4)), \
@@ -257,7 +239,7 @@ __asm__ __volatile__( \
 	"j\t2b\n\t" \
 	".previous\n\t" \
 	".section\t__ex_table,\"a\"\n\t" \
-	STR(PTR)"\t1b,3b\n\t" \
+	".word\t1b,3b\n\t" \
 	".previous" \
 	:"=r" (__pu_err) \
 	:"r" (__pu_val), "o" (__m(__pu_addr)), "i" (-EFAULT)); })
@@ -274,8 +256,8 @@ __asm__ __volatile__( \
 	"j\t3b\n\t" \
 	".previous\n\t" \
 	".section\t__ex_table,\"a\"\n\t" \
-	STR(PTR)"\t1b,4b\n\t" \
-	STR(PTR)"\t2b,4b\n\t" \
+	".word\t1b,4b\n\t" \
+	".word\t2b,4b\n\t" \
 	".previous" \
 	:"=r" (__pu_err) \
 	:"r" (__pu_val), "o" (__m(__pu_addr)), "o" (__m(__pu_addr + 4)), \
@@ -293,60 +275,195 @@ if (copy_from_user(to,from,n)) \
         return retval; \
 })
 
-#define __copy_to_user(to,from,n)                       \
-         __copy_user((to),(from),(n))
+extern size_t __copy_user(void *__to, const void *__from, size_t __n);
 
-#define __copy_from_user(to,from,n)                     \
-         __copy_user((to),(from),(n))
+#define __copy_to_user(to,from,n) ({ \
+	void *__cu_to; \
+	const void *__cu_from; \
+	long __cu_len; \
+	\
+	__cu_to = (to); \
+	__cu_from = (from); \
+	__cu_len = (n); \
+	__asm__ __volatile__( \
+		"move\t$4, %1\n\t" \
+		"move\t$5, %2\n\t" \
+		"move\t$6, %3\n\t" \
+		"jal\t__copy_user\n\t" \
+		"move\t%0, $6" \
+		: "=r" (__cu_len) \
+		: "r" (__cu_to), "r" (__cu_from), "r" (__cu_len) \
+		: "$4", "$5", "$6", "$8", "$9", "$10", "$11", "$12", "$15", \
+		  "$24", "$31","memory"); \
+	__cu_len; \
+})
 
-#define __clear_user(addr,size) \
-({ \
-	void *__cu_end; \
+#define __copy_from_user(to,from,n) ({ \
+	void *__cu_to; \
+	const void *__cu_from; \
+	long __cu_len; \
+	\
+	__cu_to = (to); \
+	__cu_from = (from); \
+	__cu_len = (n); \
 	__asm__ __volatile__( \
-		".set\tnoreorder\n\t" \
-		"1:\taddiu\t%0,1\n" \
-		"bne\t%0,%1,1b\n\t" \
-		"sb\t$0,-1(%0)\n\t" \
-		"2:\t.set\treorder\n\t" \
-		".section\t.fixup,\"ax\"\n" \
-		"3:\tsubu\t%0,1\n\t" \
-		"j\t2b\n\t" \
-		".previous\n\t" \
-		".section\t__ex_table,\"a\"\n\t" \
-		STR(PTR)"\t1b,3b\n\t" \
-		".previous" \
-		:"=r" (addr), "=r" (__cu_end) \
-		:"0" (addr), "1" (addr + size), "i" (-EFAULT) \
-		:"memory"); \
-		size = __cu_end - (addr); \
+		"move\t$4, %1\n\t" \
+		"move\t$5, %2\n\t" \
+		"move\t$6, %3\n\t" \
+		".set\tnoat\n\t" \
+		"addu\t$1, %2, %3\n\t" \
+		".set\tat\n\t" \
+		"jal\t__copy_user\n\t" \
+		"move\t%0, $6" \
+		: "=r" (__cu_len) \
+		: "r" (__cu_to), "r" (__cu_from), "r" (__cu_len) \
+		: "$4", "$5", "$6", "$8", "$9", "$10", "$11", "$12", "$15", \
+		  "$24", "$31","memory"); \
+	__cu_len; \
+})
+
+#define copy_to_user(to,from,n) ({ \
+	void *__cu_to; \
+	const void *__cu_from; \
+	long __cu_len; \
+	\
+	__cu_to = (to); \
+	__cu_from = (from); \
+	__cu_len = (n); \
+	if (access_ok(VERIFY_WRITE, __cu_to, __cu_len)) \
+		__asm__ __volatile__( \
+			"move\t$4, %1\n\t" \
+			"move\t$5, %2\n\t" \
+			"move\t$6, %3\n\t" \
+			"jal\t__copy_user\n\t" \
+			"move\t%0, $6" \
+			: "=r" (__cu_len) \
+			: "r" (__cu_to), "r" (__cu_from), "r" (__cu_len) \
+			: "$4", "$5", "$6", "$8", "$9", "$10", "$11", "$12", \
+			  "$15", "$24", "$31","memory"); \
+	__cu_len; \
+})
+
+#define copy_from_user(to,from,n) ({ \
+	void *__cu_to; \
+	const void *__cu_from; \
+	long __cu_len; \
+	\
+	__cu_to = (to); \
+	__cu_from = (from); \
+	__cu_len = (n); \
+	if (access_ok(VERIFY_READ, __cu_from, __cu_len)) \
+		__asm__ __volatile__( \
+			"move\t$4, %1\n\t" \
+			"move\t$5, %2\n\t" \
+			"move\t$6, %3\n\t" \
+			".set\tnoat\n\t" \
+			"addu\t$1, %2, %3\n\t" \
+			".set\tat\n\t" \
+			"jal\t__copy_user\n\t" \
+			"move\t%0, $6" \
+			: "=r" (__cu_len) \
+			: "r" (__cu_to), "r" (__cu_from), "r" (__cu_len) \
+			: "$4", "$5", "$6", "$8", "$9", "$10", "$11", "$12", \
+			  "$15", "$24", "$31","memory"); \
+	__cu_len; \
 })
 
+extern inline __kernel_size_t
+__clear_user(void *addr, __kernel_size_t size)
+{
+	__kernel_size_t res;
+
+	__asm__ __volatile__(
+		"move\t$4, %1\n\t"
+		"move\t$5, $0\n\t"
+		"move\t$6, %2\n\t"
+		"jal\t__bzero\n\t"
+		"move\t%0, $6"
+		: "=r" (res)
+		: "r" (addr), "r" (size)
+		: "$4", "$5", "$6", "$8", "$9", "$31");
+
+	return res;
+}
+
 #define clear_user(addr,n) ({ \
 void * __cl_addr = (addr); \
 unsigned long __cl_size = (n); \
 if (__cl_size && __access_ok(VERIFY_WRITE, ((unsigned long)(__cl_addr)), __cl_size)) \
-__clear_user(__cl_addr, __cl_size); \
+__cl_size = __clear_user(__cl_addr, __cl_size); \
 __cl_size; })
 
 /*
  * Returns: -EFAULT if exception before terminator, N if the entire
  * buffer filled, else strlen.
  */
-extern long __strncpy_from_user(char *__to, const char *__from, long __to_len);
+extern inline long
+__strncpy_from_user(char *__to, const char *__from, long __len)
+{
+	long res;
+
+	__asm__ __volatile__(
+		"move\t$4, %1\n\t"
+		"move\t$5, %2\n\t"
+		"move\t$6, %3\n\t"
+		"jal\t__strncpy_from_user_nocheck_asm\n\t"
+		"move\t%0, $2"
+		: "=r" (res)
+		: "r" (__to), "r" (__from), "r" (__len)
+		: "$2", "$3", "$4", "$5", "$6", "$8", "$31", "memory");
+
+	return res;
+}
+
+extern inline long
+strncpy_from_user(char *__to, const char *__from, long __len)
+{
+	long res;
+
+	__asm__ __volatile__(
+		"move\t$4, %1\n\t"
+		"move\t$5, %2\n\t"
+		"move\t$6, %3\n\t"
+		"jal\t__strncpy_from_user_asm\n\t"
+		"move\t%0, $2"
+		: "=r" (res)
+		: "r" (__to), "r" (__from), "r" (__len)
+		: "$2", "$3", "$4", "$5", "$6", "$8", "$31", "memory");
+
+	return res;
+}
 
-#define strncpy_from_user(dest,src,count) ({                            \
-	const void * __sc_src = (src);                                  \
-	long __sc_res = -EFAULT;                                        \
-	if (access_ok(VERIFY_READ, __sc_src, 0)) {                      \
-		__sc_res = __strncpy_from_user(dest, __sc_src, count);  \
-} __sc_res; })
 
 /* Returns: 0 if bad, string length+1 (memory size) of string if ok */
-extern long __strlen_user(const char *);
+extern inline long __strlen_user(const char *s)
+{
+	long res;
 
-extern inline long strlen_user(const char *str)
+	__asm__ __volatile__(
+		"move\t$4, %1\n\t"
+		"jal\t__strlen_user_nocheck_asm\n\t"
+		"move\t%0, $2"
+		: "=r" (res)
+		: "r" (s)
+		: "$2", "$4", "$8", "$31");
+
+	return res;
+}
+
+extern inline long strlen_user(const char *s)
 {
-	return access_ok(VERIFY_READ,str,0) ? __strlen_user(str) : 0;
+	long res;
+
+	__asm__ __volatile__(
+		"move\t$4, %1\n\t"
+		"jal\t__strlen_user_asm\n\t"
+		"move\t%0, $2"
+		: "=r" (res)
+		: "r" (s)
+		: "$2", "$4", "$8", "$31");
+
+	return res;
 }
 
 struct exception_table_entry