Merge ../linux-2.6/

6 files changed, 233 insertions, 482 deletions

diff --git a/include/asm-s390/atomic.h b/include/asm-s390/atomic.h
index d5a05cf4716..9d86ba6f12d 100644
--- a/include/asm-s390/atomic.h
+++ b/include/asm-s390/atomic.h
@@ -123,19 +123,19 @@ typedef struct {
 #define atomic64_read(v)          ((v)->counter)
 #define atomic64_set(v,i)         (((v)->counter) = (i))
 
-static __inline__ void atomic64_add(int i, atomic64_t * v)
+static __inline__ void atomic64_add(long long i, atomic64_t * v)
 {
 	       __CSG_LOOP(v, i, "agr");
 }
-static __inline__ long long atomic64_add_return(int i, atomic64_t * v)
+static __inline__ long long atomic64_add_return(long long i, atomic64_t * v)
 {
 	return __CSG_LOOP(v, i, "agr");
 }
-static __inline__ long long atomic64_add_negative(int i, atomic64_t * v)
+static __inline__ long long atomic64_add_negative(long long i, atomic64_t * v)
 {
 	return __CSG_LOOP(v, i, "agr") < 0;
 }
-static __inline__ void atomic64_sub(int i, atomic64_t * v)
+static __inline__ void atomic64_sub(long long i, atomic64_t * v)
 {
 	       __CSG_LOOP(v, i, "sgr");
 }
diff --git a/include/asm-s390/bitops.h b/include/asm-s390/bitops.h
index 16bb08499c7..8651524217f 100644
--- a/include/asm-s390/bitops.h
+++ b/include/asm-s390/bitops.h
@@ -527,13 +527,64 @@ __constant_test_bit(unsigned long nr, const volatile unsigned long *addr) {
  __constant_test_bit((nr),(addr)) : \
  __test_bit((nr),(addr)) )
 
-#ifndef __s390x__
+/*
+ * ffz = Find First Zero in word. Undefined if no zero exists,
+ * so code should check against ~0UL first..
+ */
+static inline unsigned long ffz(unsigned long word)
+{
+        unsigned long bit = 0;
+
+#ifdef __s390x__
+	if (likely((word & 0xffffffff) == 0xffffffff)) {
+		word >>= 32;
+		bit += 32;
+	}
+#endif
+	if (likely((word & 0xffff) == 0xffff)) {
+		word >>= 16;
+		bit += 16;
+	}
+	if (likely((word & 0xff) == 0xff)) {
+		word >>= 8;
+		bit += 8;
+	}
+	return bit + _zb_findmap[word & 0xff];
+}
+
+/*
+ * __ffs = find first bit in word. Undefined if no bit exists,
+ * so code should check against 0UL first..
+ */
+static inline unsigned long __ffs (unsigned long word)
+{
+	unsigned long bit = 0;
+
+#ifdef __s390x__
+	if (likely((word & 0xffffffff) == 0)) {
+		word >>= 32;
+		bit += 32;
+	}
+#endif
+	if (likely((word & 0xffff) == 0)) {
+		word >>= 16;
+		bit += 16;
+	}
+	if (likely((word & 0xff) == 0)) {
+		word >>= 8;
+		bit += 8;
+	}
+	return bit + _sb_findmap[word & 0xff];
+}
 
 /*
  * Find-bit routines..
  */
+
+#ifndef __s390x__
+
 static inline int
-find_first_zero_bit(const unsigned long * addr, unsigned int size)
+find_first_zero_bit(const unsigned long * addr, unsigned long size)
 {
 	typedef struct { long _[__BITOPS_WORDS(size)]; } addrtype;
 	unsigned long cmp, count;
@@ -548,7 +599,7 @@ find_first_zero_bit(const unsigned long * addr, unsigned int size)
                 "   srl  %2,5\n"
                 "0: c    %1,0(%0,%4)\n"
                 "   jne  1f\n"
-                "   ahi  %0,4\n"
+                "   la   %0,4(%0)\n"
                 "   brct %2,0b\n"
                 "   lr   %0,%3\n"
                 "   j    4f\n"
@@ -574,7 +625,7 @@ find_first_zero_bit(const unsigned long * addr, unsigned int size)
 }
 
 static inline int
-find_first_bit(const unsigned long * addr, unsigned int size)
+find_first_bit(const unsigned long * addr, unsigned long size)
 {
 	typedef struct { long _[__BITOPS_WORDS(size)]; } addrtype;
 	unsigned long cmp, count;
@@ -589,7 +640,7 @@ find_first_bit(const unsigned long * addr, unsigned int size)
                 "   srl  %2,5\n"
                 "0: c    %1,0(%0,%4)\n"
                 "   jne  1f\n"
-                "   ahi  %0,4\n"
+                "   la   %0,4(%0)\n"
                 "   brct %2,0b\n"
                 "   lr   %0,%3\n"
                 "   j    4f\n"
@@ -614,89 +665,8 @@ find_first_bit(const unsigned long * addr, unsigned int size)
         return (res < size) ? res : size;
 }
 
-static inline int
-find_next_zero_bit (const unsigned long * addr, int size, int offset)
-{
-        unsigned long * p = ((unsigned long *) addr) + (offset >> 5);
-        unsigned long bitvec, reg;
-        int set, bit = offset & 31, res;
-
-        if (bit) {
-                /*
-                 * Look for zero in first word
-                 */
-                bitvec = (*p) >> bit;
-                __asm__("   slr  %0,%0\n"
-                        "   lhi  %2,0xff\n"
-                        "   tml  %1,0xffff\n"
-                        "   jno  0f\n"
-                        "   ahi  %0,16\n"
-                        "   srl  %1,16\n"
-                        "0: tml  %1,0x00ff\n"
-                        "   jno  1f\n"
-                        "   ahi  %0,8\n"
-                        "   srl  %1,8\n"
-                        "1: nr   %1,%2\n"
-                        "   ic   %1,0(%1,%3)\n"
-                        "   alr  %0,%1"
-                        : "=&d" (set), "+a" (bitvec), "=&d" (reg)
-                        : "a" (&_zb_findmap) : "cc" );
-                if (set < (32 - bit))
-                        return set + offset;
-                offset += 32 - bit;
-                p++;
-        }
-        /*
-         * No zero yet, search remaining full words for a zero
-         */
-        res = find_first_zero_bit (p, size - 32 * (p - (unsigned long *) addr));
-        return (offset + res);
-}
-
-static inline int
-find_next_bit (const unsigned long * addr, int size, int offset)
-{
-        unsigned long * p = ((unsigned long *) addr) + (offset >> 5);
-        unsigned long bitvec, reg;
-        int set, bit = offset & 31, res;
-
-        if (bit) {
-                /*
-                 * Look for set bit in first word
-                 */
-                bitvec = (*p) >> bit;
-                __asm__("   slr  %0,%0\n"
-                        "   lhi  %2,0xff\n"
-                        "   tml  %1,0xffff\n"
-                        "   jnz  0f\n"
-                        "   ahi  %0,16\n"
-                        "   srl  %1,16\n"
-                        "0: tml  %1,0x00ff\n"
-                        "   jnz  1f\n"
-                        "   ahi  %0,8\n"
-                        "   srl  %1,8\n"
-                        "1: nr   %1,%2\n"
-                        "   ic   %1,0(%1,%3)\n"
-                        "   alr  %0,%1"
-                        : "=&d" (set), "+a" (bitvec), "=&d" (reg)
-                        : "a" (&_sb_findmap) : "cc" );
-                if (set < (32 - bit))
-                        return set + offset;
-                offset += 32 - bit;
-                p++;
-        }
-        /*
-         * No set bit yet, search remaining full words for a bit
-         */
-        res = find_first_bit (p, size - 32 * (p - (unsigned long *) addr));
-        return (offset + res);
-}
-
 #else /* __s390x__ */
 
-/*
- * Find-bit routines..
- */
 static inline unsigned long
 find_first_zero_bit(const unsigned long * addr, unsigned long size)
 {
@@ -712,7 +682,7 @@ find_first_zero_bit(const unsigned long * addr, unsigned long size)
                 "   srlg  %2,%2,6\n"
                 "0: cg    %1,0(%0,%4)\n"
                 "   jne   1f\n"
-                "   aghi  %0,8\n"
+                "   la    %0,8(%0)\n"
                 "   brct  %2,0b\n"
                 "   lgr   %0,%3\n"
                 "   j     5f\n"
@@ -785,143 +755,66 @@ find_first_bit(const unsigned long * addr, unsigned long size)
         return (res < size) ? res : size;
 }
 
-static inline unsigned long
-find_next_zero_bit (const unsigned long * addr, unsigned long size, unsigned long offset)
-{
-        unsigned long * p = ((unsigned long *) addr) + (offset >> 6);
-        unsigned long bitvec, reg;
-        unsigned long set, bit = offset & 63, res;
-
-        if (bit) {
-                /*
-                 * Look for zero in first word
-                 */
-                bitvec = (*p) >> bit;
-                __asm__("   lhi  %2,-1\n"
-                        "   slgr %0,%0\n"
-                        "   clr  %1,%2\n"
-                        "   jne  0f\n"
-                        "   aghi %0,32\n"
-                        "   srlg %1,%1,32\n"
-			"0: lghi %2,0xff\n"
-                        "   tmll %1,0xffff\n"
-                        "   jno  1f\n"
-                        "   aghi %0,16\n"
-                        "   srlg %1,%1,16\n"
-                        "1: tmll %1,0x00ff\n"
-                        "   jno  2f\n"
-                        "   aghi %0,8\n"
-                        "   srlg %1,%1,8\n"
-                        "2: ngr  %1,%2\n"
-                        "   ic   %1,0(%1,%3)\n"
-                        "   algr %0,%1"
-                        : "=&d" (set), "+a" (bitvec), "=&d" (reg)
-                        : "a" (&_zb_findmap) : "cc" );
-                if (set < (64 - bit))
-                        return set + offset;
-                offset += 64 - bit;
-                p++;
-        }
-        /*
-         * No zero yet, search remaining full words for a zero
-         */
-        res = find_first_zero_bit (p, size - 64 * (p - (unsigned long *) addr));
-        return (offset + res);
-}
-
-static inline unsigned long
-find_next_bit (const unsigned long * addr, unsigned long size, unsigned long offset)
-{
-        unsigned long * p = ((unsigned long *) addr) + (offset >> 6);
-        unsigned long bitvec, reg;
-        unsigned long set, bit = offset & 63, res;
-
-        if (bit) {
-                /*
-                 * Look for zero in first word
-                 */
-                bitvec = (*p) >> bit;
-                __asm__("   slgr %0,%0\n"
-                        "   ltr  %1,%1\n"
-                        "   jnz  0f\n"
-                        "   aghi %0,32\n"
-                        "   srlg %1,%1,32\n"
-			"0: lghi %2,0xff\n"
-                        "   tmll %1,0xffff\n"
-                        "   jnz  1f\n"
-                        "   aghi %0,16\n"
-                        "   srlg %1,%1,16\n"
-                        "1: tmll %1,0x00ff\n"
-                        "   jnz  2f\n"
-                        "   aghi %0,8\n"
-                        "   srlg %1,%1,8\n"
-                        "2: ngr  %1,%2\n"
-                        "   ic   %1,0(%1,%3)\n"
-                        "   algr %0,%1"
-                        : "=&d" (set), "+a" (bitvec), "=&d" (reg)
-                        : "a" (&_sb_findmap) : "cc" );
-                if (set < (64 - bit))
-                        return set + offset;
-                offset += 64 - bit;
-                p++;
-        }
-        /*
-         * No set bit yet, search remaining full words for a bit
-         */
-        res = find_first_bit (p, size - 64 * (p - (unsigned long *) addr));
-        return (offset + res);
-}
-
 #endif /* __s390x__ */
 
-/*
- * ffz = Find First Zero in word. Undefined if no zero exists,
- * so code should check against ~0UL first..
- */
-static inline unsigned long ffz(unsigned long word)
+static inline int
+find_next_zero_bit (const unsigned long * addr, unsigned long size,
+		    unsigned long offset)
 {
-        unsigned long bit = 0;
-
-#ifdef __s390x__
-	if (likely((word & 0xffffffff) == 0xffffffff)) {
-		word >>= 32;
-		bit += 32;
-	}
-#endif
-	if (likely((word & 0xffff) == 0xffff)) {
-		word >>= 16;
-		bit += 16;
+        const unsigned long *p;
+	unsigned long bit, set;
+
+	if (offset >= size)
+		return size;
+	bit = offset & (__BITOPS_WORDSIZE - 1);
+	offset -= bit;
+	size -= offset;
+	p = addr + offset / __BITOPS_WORDSIZE;
+	if (bit) {
+		/*
+		 * s390 version of ffz returns __BITOPS_WORDSIZE
+		 * if no zero bit is present in the word.
+		 */
+		set = ffz(*p >> bit) + bit;
+		if (set >= size)
+			return size + offset;
+		if (set < __BITOPS_WORDSIZE)
+			return set + offset;
+		offset += __BITOPS_WORDSIZE;
+		size -= __BITOPS_WORDSIZE;
+		p++;
 	}
-	if (likely((word & 0xff) == 0xff)) {
-		word >>= 8;
-		bit += 8;
-	}
-	return bit + _zb_findmap[word & 0xff];
+	return offset + find_first_zero_bit(p, size);
 }
 
-/*
- * __ffs = find first bit in word. Undefined if no bit exists,
- * so code should check against 0UL first..
- */
-static inline unsigned long __ffs (unsigned long word)
+static inline int
+find_next_bit (const unsigned long * addr, unsigned long size,
+	       unsigned long offset)
 {
-	unsigned long bit = 0;
-
-#ifdef __s390x__
-	if (likely((word & 0xffffffff) == 0)) {
-		word >>= 32;
-		bit += 32;
+        const unsigned long *p;
+	unsigned long bit, set;
+
+	if (offset >= size)
+		return size;
+	bit = offset & (__BITOPS_WORDSIZE - 1);
+	offset -= bit;
+	size -= offset;
+	p = addr + offset / __BITOPS_WORDSIZE;
+	if (bit) {
+		/*
+		 * s390 version of __ffs returns __BITOPS_WORDSIZE
+		 * if no one bit is present in the word.
+		 */
+		set = __ffs(*p & (~0UL << bit));
+		if (set >= size)
+			return size + offset;
+		if (set < __BITOPS_WORDSIZE)
+			return set + offset;
+		offset += __BITOPS_WORDSIZE;
+		size -= __BITOPS_WORDSIZE;
+		p++;
 	}
-#endif
-	if (likely((word & 0xffff) == 0)) {
-		word >>= 16;
-		bit += 16;
-	}
-	if (likely((word & 0xff) == 0)) {
-		word >>= 8;
-		bit += 8;
-	}
-	return bit + _sb_findmap[word & 0xff];
+	return offset + find_first_bit(p, size);
 }
 
 /*
@@ -1031,49 +924,6 @@ ext2_find_first_zero_bit(void *vaddr, unsigned int size)
         return (res < size) ? res : size;
 }
 
-static inline int 
-ext2_find_next_zero_bit(void *vaddr, unsigned int size, unsigned offset)
-{
-        unsigned long *addr = vaddr;
-        unsigned long *p = addr + (offset >> 5);
-        unsigned long word, reg;
-        unsigned int bit = offset & 31UL, res;
-
-        if (offset >= size)
-                return size;
-
-        if (bit) {
-                __asm__("   ic   %0,0(%1)\n"
-                        "   icm  %0,2,1(%1)\n"
-                        "   icm  %0,4,2(%1)\n"
-                        "   icm  %0,8,3(%1)"
-                        : "=&a" (word) : "a" (p) : "cc" );
-		word >>= bit;
-                res = bit;
-                /* Look for zero in first longword */
-                __asm__("   lhi  %2,0xff\n"
-                        "   tml  %1,0xffff\n"
-                	"   jno  0f\n"
-                	"   ahi  %0,16\n"
-                	"   srl  %1,16\n"
-                	"0: tml  %1,0x00ff\n"
-                	"   jno  1f\n"
-                	"   ahi  %0,8\n"
-                	"   srl  %1,8\n"
-                	"1: nr   %1,%2\n"
-                	"   ic   %1,0(%1,%3)\n"
-                	"   alr  %0,%1"
-                	: "+&d" (res), "+&a" (word), "=&d" (reg)
-                  	: "a" (&_zb_findmap) : "cc" );
-                if (res < 32)
-			return (p - addr)*32 + res;
-                p++;
-        }
-        /* No zero yet, search remaining full bytes for a zero */
-        res = ext2_find_first_zero_bit (p, size - 32 * (p - addr));
-        return (p - addr) * 32 + res;
-}
-
 #else /* __s390x__ */
 
 static inline unsigned long
@@ -1120,56 +970,46 @@ ext2_find_first_zero_bit(void *vaddr, unsigned long size)
         return (res < size) ? res : size;
 }
 
-static inline unsigned long
+#endif /* __s390x__ */
+
+static inline int
 ext2_find_next_zero_bit(void *vaddr, unsigned long size, unsigned long offset)
 {
-        unsigned long *addr = vaddr;
-        unsigned long *p = addr + (offset >> 6);
-        unsigned long word, reg;
-        unsigned long bit = offset & 63UL, res;
+        unsigned long *addr = vaddr, *p;
+	unsigned long word, bit, set;
 
         if (offset >= size)
                 return size;
-
+	bit = offset & (__BITOPS_WORDSIZE - 1);
+	offset -= bit;
+	size -= offset;
+	p = addr + offset / __BITOPS_WORDSIZE;
         if (bit) {
-                __asm__("   lrvg %0,%1" /* load reversed, neat instruction */
-                        : "=a" (word) : "m" (*p) );
-                word >>= bit;
-                res = bit;
-                /* Look for zero in first 8 byte word */
-                __asm__("   lghi %2,0xff\n"
-			"   tmll %1,0xffff\n"
-			"   jno  2f\n"
-			"   ahi  %0,16\n"
-			"   srlg %1,%1,16\n"
-                	"0: tmll %1,0xffff\n"
-                        "   jno  2f\n"
-                        "   ahi  %0,16\n"
-                        "   srlg %1,%1,16\n"
-                        "1: tmll %1,0xffff\n"
-                        "   jno  2f\n"
-                        "   ahi  %0,16\n"
-                        "   srl  %1,16\n"
-                        "2: tmll %1,0x00ff\n"
-                	"   jno  3f\n"
-                	"   ahi  %0,8\n"
-                	"   srl  %1,8\n"
-                	"3: ngr  %1,%2\n"
-                	"   ic   %1,0(%1,%3)\n"
-                	"   alr  %0,%1"
-                	: "+&d" (res), "+a" (word), "=&d" (reg)
-                  	: "a" (&_zb_findmap) : "cc" );
-                if (res < 64)
-			return (p - addr)*64 + res;
-                p++;
+#ifndef __s390x__
+                asm("   ic   %0,0(%1)\n"
+		    "   icm  %0,2,1(%1)\n"
+		    "   icm  %0,4,2(%1)\n"
+		    "   icm  %0,8,3(%1)"
+		    : "=&a" (word) : "a" (p), "m" (*p) : "cc" );
+#else
+                asm("   lrvg %0,%1" : "=a" (word) : "m" (*p) );
+#endif
+		/*
+		 * s390 version of ffz returns __BITOPS_WORDSIZE
+		 * if no zero bit is present in the word.
+		 */
+		set = ffz(word >> bit) + bit;
+		if (set >= size)
+			return size + offset;
+		if (set < __BITOPS_WORDSIZE)
+			return set + offset;
+		offset += __BITOPS_WORDSIZE;
+		size -= __BITOPS_WORDSIZE;
+		p++;
         }
-        /* No zero yet, search remaining full bytes for a zero */
-        res = ext2_find_first_zero_bit (p, size - 64 * (p - addr));
-        return (p - addr) * 64 + res;
+	return offset + ext2_find_first_zero_bit(p, size);
 }
 
-#endif /* __s390x__ */
-
 /* Bitmap functions for the minix filesystem.  */
 /* FIXME !!! */
 #define minix_test_and_set_bit(nr,addr) \
diff --git a/include/asm-s390/emergency-restart.h b/include/asm-s390/emergency-restart.h
new file mode 100644
index 00000000000..108d8c48e42
--- /dev/null
+++ b/include/asm-s390/emergency-restart.h
@@ -0,0 +1,6 @@
+#ifndef _ASM_EMERGENCY_RESTART_H
+#define _ASM_EMERGENCY_RESTART_H
+
+#include <asm-generic/emergency-restart.h>
+
+#endif /* _ASM_EMERGENCY_RESTART_H */
diff --git a/include/asm-s390/lowcore.h b/include/asm-s390/lowcore.h
index 76b5b19c0ae..afe6a9f9b0a 100644
--- a/include/asm-s390/lowcore.h
+++ b/include/asm-s390/lowcore.h
@@ -90,7 +90,6 @@
 #define __LC_SYSTEM_TIMER		0x278
 #define __LC_LAST_UPDATE_CLOCK		0x280
 #define __LC_STEAL_CLOCK		0x288
-#define __LC_DIAG44_OPCODE		0x290
 #define __LC_KERNEL_STACK               0xD40
 #define __LC_THREAD_INFO		0xD48
 #define __LC_ASYNC_STACK                0xD50
@@ -286,8 +285,7 @@ struct _lowcore
 	__u64        system_timer;             /* 0x278 */
 	__u64        last_update_clock;        /* 0x280 */
 	__u64        steal_clock;              /* 0x288 */
-	__u32        diag44_opcode;            /* 0x290 */
-        __u8         pad8[0xc00-0x294];        /* 0x294 */
+        __u8         pad8[0xc00-0x290];        /* 0x290 */
         /* System info area */
 	__u64        save_area[16];            /* 0xc00 */
         __u8         pad9[0xd40-0xc80];        /* 0xc80 */
diff --git a/include/asm-s390/processor.h b/include/asm-s390/processor.h
index 8bd14de69e3..4ec652ebb3b 100644
--- a/include/asm-s390/processor.h
+++ b/include/asm-s390/processor.h
@@ -203,7 +203,10 @@ unsigned long get_wchan(struct task_struct *p);
 # define cpu_relax()	asm volatile ("diag 0,0,68" : : : "memory")
 #else /* __s390x__ */
 # define cpu_relax() \
-	asm volatile ("ex 0,%0" : : "i" (__LC_DIAG44_OPCODE) : "memory")
+	do { \
+		if (MACHINE_HAS_DIAG44) \
+			asm volatile ("diag 0,0,68" : : : "memory"); \
+	} while (0)
 #endif /* __s390x__ */
 
 /*
diff --git a/include/asm-s390/spinlock.h b/include/asm-s390/spinlock.h
index 53cc736b982..8ff10300f7e 100644
--- a/include/asm-s390/spinlock.h
+++ b/include/asm-s390/spinlock.h
@@ -11,21 +11,16 @@
 #ifndef __ASM_SPINLOCK_H
 #define __ASM_SPINLOCK_H
 
-#ifdef __s390x__
-/*
- * Grmph, take care of %&#! user space programs that include
- * asm/spinlock.h. The diagnose is only available in kernel
- * context.
- */
-#ifdef __KERNEL__
-#include <asm/lowcore.h>
-#define __DIAG44_INSN "ex"
-#define __DIAG44_OPERAND __LC_DIAG44_OPCODE
-#else
-#define __DIAG44_INSN "#"
-#define __DIAG44_OPERAND 0
-#endif
-#endif /* __s390x__ */
+static inline int
+_raw_compare_and_swap(volatile unsigned int *lock,
+		      unsigned int old, unsigned int new)
+{
+	asm volatile ("cs %0,%3,0(%4)"
+		      : "=d" (old), "=m" (*lock)
+		      : "0" (old), "d" (new), "a" (lock), "m" (*lock)
+		      : "cc", "memory" );
+	return old;
+}
 
 /*
  * Simple spin lock operations.  There are two variants, one clears IRQ's
@@ -41,58 +36,35 @@ typedef struct {
 #endif
 } __attribute__ ((aligned (4))) spinlock_t;
 
-#define SPIN_LOCK_UNLOCKED (spinlock_t) { 0 }
-#define spin_lock_init(lp) do { (lp)->lock = 0; } while(0)
+#define SPIN_LOCK_UNLOCKED	(spinlock_t) { 0 }
+#define spin_lock_init(lp)	do { (lp)->lock = 0; } while(0)
 #define spin_unlock_wait(lp)	do { barrier(); } while(((volatile spinlock_t *)(lp))->lock)
-#define spin_is_locked(x) ((x)->lock != 0)
+#define spin_is_locked(x)	((x)->lock != 0)
 #define _raw_spin_lock_flags(lock, flags) _raw_spin_lock(lock)
 
-extern inline void _raw_spin_lock(spinlock_t *lp)
+extern void _raw_spin_lock_wait(spinlock_t *lp, unsigned int pc);
+extern int _raw_spin_trylock_retry(spinlock_t *lp, unsigned int pc);
+
+static inline void _raw_spin_lock(spinlock_t *lp)
 {
-#ifndef __s390x__
-	unsigned int reg1, reg2;
-        __asm__ __volatile__("    bras  %0,1f\n"
-                           "0:  diag  0,0,68\n"
-                           "1:  slr   %1,%1\n"
-                           "    cs    %1,%0,0(%3)\n"
-                           "    jl    0b\n"
-                           : "=&d" (reg1), "=&d" (reg2), "=m" (lp->lock)
-			   : "a" (&lp->lock), "m" (lp->lock)
-			   : "cc", "memory" );
-#else /* __s390x__ */
-	unsigned long reg1, reg2;
-        __asm__ __volatile__("    bras  %1,1f\n"
-                           "0:  " __DIAG44_INSN " 0,%4\n"
-                           "1:  slr   %0,%0\n"
-                           "    cs    %0,%1,0(%3)\n"
-                           "    jl    0b\n"
-                           : "=&d" (reg1), "=&d" (reg2), "=m" (lp->lock)
-			   : "a" (&lp->lock), "i" (__DIAG44_OPERAND),
-			     "m" (lp->lock) : "cc", "memory" );
-#endif /* __s390x__ */
+	unsigned long pc = (unsigned long) __builtin_return_address(0);
+
+	if (unlikely(_raw_compare_and_swap(&lp->lock, 0, pc) != 0))
+		_raw_spin_lock_wait(lp, pc);
 }
 
-extern inline int _raw_spin_trylock(spinlock_t *lp)
+static inline int _raw_spin_trylock(spinlock_t *lp)
 {
-	unsigned long reg;
-	unsigned int result;
-
-	__asm__ __volatile__("    basr  %1,0\n"
-			   "0:  cs    %0,%1,0(%3)"
-			   : "=d" (result), "=&d" (reg), "=m" (lp->lock)
-			   : "a" (&lp->lock), "m" (lp->lock), "0" (0)
-			   : "cc", "memory" );
-	return !result;
+	unsigned long pc = (unsigned long) __builtin_return_address(0);
+
+	if (likely(_raw_compare_and_swap(&lp->lock, 0, pc) == 0))
+		return 1;
+	return _raw_spin_trylock_retry(lp, pc);
 }
 
-extern inline void _raw_spin_unlock(spinlock_t *lp)
+static inline void _raw_spin_unlock(spinlock_t *lp)
 {
-	unsigned int old;
-
-	__asm__ __volatile__("cs %0,%3,0(%4)"
-			   : "=d" (old), "=m" (lp->lock)
-			   : "0" (lp->lock), "d" (0), "a" (lp)
-			   : "cc", "memory" );
+	_raw_compare_and_swap(&lp->lock, lp->lock, 0);
 }
 		
 /*
@@ -106,7 +78,7 @@ extern inline void _raw_spin_unlock(spinlock_t *lp)
  * read-locks.
  */
 typedef struct {
-	volatile unsigned long lock;
+	volatile unsigned int lock;
 	volatile unsigned long owner_pc;
 #ifdef CONFIG_PREEMPT
 	unsigned int break_lock;
@@ -129,123 +101,55 @@ typedef struct {
  */
 #define write_can_lock(x) ((x)->lock == 0)
 
-#ifndef __s390x__
-#define _raw_read_lock(rw)   \
-        asm volatile("   l     2,0(%1)\n"   \
-                     "   j     1f\n"     \
-                     "0: diag  0,0,68\n" \
-                     "1: la    2,0(2)\n"     /* clear high (=write) bit */ \
-                     "   la    3,1(2)\n"     /* one more reader */ \
-                     "   cs    2,3,0(%1)\n"  /* try to write new value */ \
-                     "   jl    0b"       \
-                     : "=m" ((rw)->lock) : "a" (&(rw)->lock), \
-		       "m" ((rw)->lock) : "2", "3", "cc", "memory" )
-#else /* __s390x__ */
-#define _raw_read_lock(rw)   \
-        asm volatile("   lg    2,0(%1)\n"   \
-                     "   j     1f\n"     \
-                     "0: " __DIAG44_INSN " 0,%2\n" \
-                     "1: nihh  2,0x7fff\n" /* clear high (=write) bit */ \
-                     "   la    3,1(2)\n"   /* one more reader */  \
-                     "   csg   2,3,0(%1)\n" /* try to write new value */ \
-                     "   jl    0b"       \
-                     : "=m" ((rw)->lock) \
-		     : "a" (&(rw)->lock), "i" (__DIAG44_OPERAND), \
-		       "m" ((rw)->lock) : "2", "3", "cc", "memory" )
-#endif /* __s390x__ */
-
-#ifndef __s390x__
-#define _raw_read_unlock(rw) \
-        asm volatile("   l     2,0(%1)\n"   \
-                     "   j     1f\n"     \
-                     "0: diag  0,0,68\n" \
-                     "1: lr    3,2\n"    \
-                     "   ahi   3,-1\n"    /* one less reader */ \
-                     "   cs    2,3,0(%1)\n" \
-                     "   jl    0b"       \
-                     : "=m" ((rw)->lock) : "a" (&(rw)->lock), \
-		       "m" ((rw)->lock) : "2", "3", "cc", "memory" )
-#else /* __s390x__ */
-#define _raw_read_unlock(rw) \
-        asm volatile("   lg    2,0(%1)\n"   \
-                     "   j     1f\n"     \
-                     "0: " __DIAG44_INSN " 0,%2\n" \
-                     "1: lgr   3,2\n"    \
-                     "   bctgr 3,0\n"    /* one less reader */ \
-                     "   csg   2,3,0(%1)\n" \
-                     "   jl    0b"       \
-                     : "=m" ((rw)->lock) \
-		     : "a" (&(rw)->lock), "i" (__DIAG44_OPERAND), \
-		       "m" ((rw)->lock) : "2", "3", "cc", "memory" )
-#endif /* __s390x__ */
-
-#ifndef __s390x__
-#define _raw_write_lock(rw) \
-        asm volatile("   lhi   3,1\n"    \
-                     "   sll   3,31\n"    /* new lock value = 0x80000000 */ \
-                     "   j     1f\n"     \
-                     "0: diag  0,0,68\n" \
-                     "1: slr   2,2\n"     /* old lock value must be 0 */ \
-                     "   cs    2,3,0(%1)\n" \
-                     "   jl    0b"       \
-                     : "=m" ((rw)->lock) : "a" (&(rw)->lock), \
-		       "m" ((rw)->lock) : "2", "3", "cc", "memory" )
-#else /* __s390x__ */
-#define _raw_write_lock(rw) \
-        asm volatile("   llihh 3,0x8000\n" /* new lock value = 0x80...0 */ \
-                     "   j     1f\n"       \
-                     "0: " __DIAG44_INSN " 0,%2\n"   \
-                     "1: slgr  2,2\n"      /* old lock value must be 0 */ \
-                     "   csg   2,3,0(%1)\n" \
-                     "   jl    0b"         \
-                     : "=m" ((rw)->lock) \
-		     : "a" (&(rw)->lock), "i" (__DIAG44_OPERAND), \
-		       "m" ((rw)->lock) : "2", "3", "cc", "memory" )
-#endif /* __s390x__ */
-
-#ifndef __s390x__
-#define _raw_write_unlock(rw) \
-        asm volatile("   slr   3,3\n"     /* new lock value = 0 */ \
-                     "   j     1f\n"     \
-                     "0: diag  0,0,68\n" \
-                     "1: lhi   2,1\n"    \
-                     "   sll   2,31\n"    /* old lock value must be 0x80000000 */ \
-                     "   cs    2,3,0(%1)\n" \
-                     "   jl    0b"       \
-                     : "=m" ((rw)->lock) : "a" (&(rw)->lock), \
-		       "m" ((rw)->lock) : "2", "3", "cc", "memory" )
-#else /* __s390x__ */
-#define _raw_write_unlock(rw) \
-        asm volatile("   slgr  3,3\n"      /* new lock value = 0 */ \
-                     "   j     1f\n"       \
-                     "0: " __DIAG44_INSN " 0,%2\n"   \
-                     "1: llihh 2,0x8000\n" /* old lock value must be 0x8..0 */\
-                     "   csg   2,3,0(%1)\n"   \
-                     "   jl    0b"         \
-                     : "=m" ((rw)->lock) \
-		     : "a" (&(rw)->lock), "i" (__DIAG44_OPERAND), \
-		       "m" ((rw)->lock) : "2", "3", "cc", "memory" )
-#endif /* __s390x__ */
-
-#define _raw_read_trylock(lock) generic_raw_read_trylock(lock)
-
-extern inline int _raw_write_trylock(rwlock_t *rw)
+extern void _raw_read_lock_wait(rwlock_t *lp);
+extern int _raw_read_trylock_retry(rwlock_t *lp);
+extern void _raw_write_lock_wait(rwlock_t *lp);
+extern int _raw_write_trylock_retry(rwlock_t *lp);
+
+static inline void _raw_read_lock(rwlock_t *rw)
+{
+	unsigned int old;
+	old = rw->lock & 0x7fffffffU;
+	if (_raw_compare_and_swap(&rw->lock, old, old + 1) != old)
+		_raw_read_lock_wait(rw);
+}
+
+static inline void _raw_read_unlock(rwlock_t *rw)
+{
+	unsigned int old, cmp;
+
+	old = rw->lock;
+	do {
+		cmp = old;
+		old = _raw_compare_and_swap(&rw->lock, old, old - 1);
+	} while (cmp != old);
+}
+
+static inline void _raw_write_lock(rwlock_t *rw)
+{
+	if (unlikely(_raw_compare_and_swap(&rw->lock, 0, 0x80000000) != 0))
+		_raw_write_lock_wait(rw);
+}
+
+static inline void _raw_write_unlock(rwlock_t *rw)
+{
+	_raw_compare_and_swap(&rw->lock, 0x80000000, 0);
+}
+
+static inline int _raw_read_trylock(rwlock_t *rw)
+{
+	unsigned int old;
+	old = rw->lock & 0x7fffffffU;
+	if (likely(_raw_compare_and_swap(&rw->lock, old, old + 1) == old))
+		return 1;
+	return _raw_read_trylock_retry(rw);
+}
+
+static inline int _raw_write_trylock(rwlock_t *rw)
 {
-	unsigned long result, reg;
-	
-	__asm__ __volatile__(
-#ifndef __s390x__
-			     "   lhi  %1,1\n"
-			     "   sll  %1,31\n"
-			     "   cs   %0,%1,0(%3)"
-#else /* __s390x__ */
-			     "   llihh %1,0x8000\n"
-			     "0: csg %0,%1,0(%3)\n"
-#endif /* __s390x__ */
-			     : "=d" (result), "=&d" (reg), "=m" (rw->lock)
-			     : "a" (&rw->lock), "m" (rw->lock), "0" (0UL)
-			     : "cc", "memory" );
-	return result == 0;
+	if (likely(_raw_compare_and_swap(&rw->lock, 0, 0x80000000) == 0))
+		return 1;
+	return _raw_write_trylock_retry(rw);
 }
 
 #endif /* __ASM_SPINLOCK_H */