frv: Rewrite atomic implementation

Mostly complete rewrite of the FRV atomic implementation, instead of
using assembly files, use inline assembler.

The out-of-line CONFIG option makes a bit of a mess of things, but a
little CPP trickery gets that done too.

FRV already had the atomic logic ops but under a non standard name,
the reimplementation provides the generic names and provides the
intermediate form required for the bitops implementation.

The slightly inconsistent __atomic32_fetch_##op naming is because
__atomic_fetch_##op conlicts with GCC builtin functions.

The 64bit atomic ops use the inline assembly %Ln construct to access
the low word register (r+1), afaik this construct was not previously
used in the kernel and is completely undocumented, but I found it in
the FRV GCC code and it seems to work.

FRV had a non-standard definition of atomic_{clear,set}_mask() which
would work types other than atomic_t, the one user relying on that
(arch/frv/kernel/dma.c) got converted to use the new intermediate
form.

Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
diff --git a/arch/frv/include/asm/atomic.h b/arch/frv/include/asm/atomic.h
index 102190a..74d2245 100644
--- a/arch/frv/include/asm/atomic.h
+++ b/arch/frv/include/asm/atomic.h
@@ -15,7 +15,6 @@
 #define _ASM_ATOMIC_H
 
 #include <linux/types.h>
-#include <asm/spr-regs.h>
 #include <asm/cmpxchg.h>
 #include <asm/barrier.h>
 
@@ -23,6 +22,8 @@
 #error not SMP safe
 #endif
 
+#include <asm/atomic_defs.h>
+
 /*
  * Atomic operations that C can't guarantee us.  Useful for
  * resource counting etc..
@@ -34,56 +35,26 @@
 #define atomic_read(v)		ACCESS_ONCE((v)->counter)
 #define atomic_set(v, i)	(((v)->counter) = (i))
 
-#ifndef CONFIG_FRV_OUTOFLINE_ATOMIC_OPS
+static inline int atomic_inc_return(atomic_t *v)
+{
+	return __atomic_add_return(1, &v->counter);
+}
+
+static inline int atomic_dec_return(atomic_t *v)
+{
+	return __atomic_sub_return(1, &v->counter);
+}
+
 static inline int atomic_add_return(int i, atomic_t *v)
 {
-	unsigned long val;
-
-	asm("0:						\n"
-	    "	orcc		gr0,gr0,gr0,icc3	\n"	/* set ICC3.Z */
-	    "	ckeq		icc3,cc7		\n"
-	    "	ld.p		%M0,%1			\n"	/* LD.P/ORCR must be atomic */
-	    "	orcr		cc7,cc7,cc3		\n"	/* set CC3 to true */
-	    "	add%I2		%1,%2,%1		\n"
-	    "	cst.p		%1,%M0		,cc3,#1	\n"
-	    "	corcc		gr29,gr29,gr0	,cc3,#1	\n"	/* clear ICC3.Z if store happens */
-	    "	beq		icc3,#0,0b		\n"
-	    : "+U"(v->counter), "=&r"(val)
-	    : "NPr"(i)
-	    : "memory", "cc7", "cc3", "icc3"
-	    );
-
-	return val;
+	return __atomic_add_return(i, &v->counter);
 }
 
 static inline int atomic_sub_return(int i, atomic_t *v)
 {
-	unsigned long val;
-
-	asm("0:						\n"
-	    "	orcc		gr0,gr0,gr0,icc3	\n"	/* set ICC3.Z */
-	    "	ckeq		icc3,cc7		\n"
-	    "	ld.p		%M0,%1			\n"	/* LD.P/ORCR must be atomic */
-	    "	orcr		cc7,cc7,cc3		\n"	/* set CC3 to true */
-	    "	sub%I2		%1,%2,%1		\n"
-	    "	cst.p		%1,%M0		,cc3,#1	\n"
-	    "	corcc		gr29,gr29,gr0	,cc3,#1	\n"	/* clear ICC3.Z if store happens */
-	    "	beq		icc3,#0,0b		\n"
-	    : "+U"(v->counter), "=&r"(val)
-	    : "NPr"(i)
-	    : "memory", "cc7", "cc3", "icc3"
-	    );
-
-	return val;
+	return __atomic_sub_return(i, &v->counter);
 }
 
-#else
-
-extern int atomic_add_return(int i, atomic_t *v);
-extern int atomic_sub_return(int i, atomic_t *v);
-
-#endif
-
 static inline int atomic_add_negative(int i, atomic_t *v)
 {
 	return atomic_add_return(i, v) < 0;
@@ -101,17 +72,14 @@
 
 static inline void atomic_inc(atomic_t *v)
 {
-	atomic_add_return(1, v);
+	atomic_inc_return(v);
 }
 
 static inline void atomic_dec(atomic_t *v)
 {
-	atomic_sub_return(1, v);
+	atomic_dec_return(v);
 }
 
-#define atomic_dec_return(v)		atomic_sub_return(1, (v))
-#define atomic_inc_return(v)		atomic_add_return(1, (v))
-
 #define atomic_sub_and_test(i,v)	(atomic_sub_return((i), (v)) == 0)
 #define atomic_dec_and_test(v)		(atomic_sub_return(1, (v)) == 0)
 #define atomic_inc_and_test(v)		(atomic_add_return(1, (v)) == 0)
@@ -120,18 +88,19 @@
  * 64-bit atomic ops
  */
 typedef struct {
-	volatile long long counter;
+	long long counter;
 } atomic64_t;
 
 #define ATOMIC64_INIT(i)	{ (i) }
 
-static inline long long atomic64_read(atomic64_t *v)
+static inline long long atomic64_read(const atomic64_t *v)
 {
 	long long counter;
 
 	asm("ldd%I1 %M1,%0"
 	    : "=e"(counter)
 	    : "m"(v->counter));
+
 	return counter;
 }
 
@@ -142,10 +111,25 @@
 		     : "e"(i));
 }
 
-extern long long atomic64_inc_return(atomic64_t *v);
-extern long long atomic64_dec_return(atomic64_t *v);
-extern long long atomic64_add_return(long long i, atomic64_t *v);
-extern long long atomic64_sub_return(long long i, atomic64_t *v);
+static inline long long atomic64_inc_return(atomic64_t *v)
+{
+	return __atomic64_add_return(1, &v->counter);
+}
+
+static inline long long atomic64_dec_return(atomic64_t *v)
+{
+	return __atomic64_sub_return(1, &v->counter);
+}
+
+static inline long long atomic64_add_return(long long i, atomic64_t *v)
+{
+	return __atomic64_add_return(i, &v->counter);
+}
+
+static inline long long atomic64_sub_return(long long i, atomic64_t *v)
+{
+	return __atomic64_sub_return(i, &v->counter);
+}
 
 static inline long long atomic64_add_negative(long long i, atomic64_t *v)
 {
@@ -176,6 +160,7 @@
 #define atomic64_dec_and_test(v)	(atomic64_dec_return((v)) == 0)
 #define atomic64_inc_and_test(v)	(atomic64_inc_return((v)) == 0)
 
+
 #define atomic_cmpxchg(v, old, new)	(cmpxchg(&(v)->counter, old, new))
 #define atomic_xchg(v, new)		(xchg(&(v)->counter, new))
 #define atomic64_cmpxchg(v, old, new)	(__cmpxchg_64(old, new, &(v)->counter))
@@ -196,5 +181,33 @@
 	return c;
 }
 
+#define ATOMIC_OP(op)							\
+static inline void atomic_##op(int i, atomic_t *v)			\
+{									\
+	(void)__atomic32_fetch_##op(i, &v->counter);			\
+}									\
+									\
+static inline void atomic64_##op(long long i, atomic64_t *v)		\
+{									\
+	(void)__atomic64_fetch_##op(i, &v->counter);			\
+}
+
+#define CONFIG_ARCH_HAS_ATOMIC_OR
+
+ATOMIC_OP(or)
+ATOMIC_OP(and)
+ATOMIC_OP(xor)
+
+#undef ATOMIC_OP
+
+static inline __deprecated void atomic_clear_mask(unsigned int mask, atomic_t *v)
+{
+	atomic_and(~mask, v);
+}
+
+static inline __deprecated void atomic_set_mask(unsigned int mask, atomic_t *v)
+{
+	atomic_or(mask, v);
+}
 
 #endif /* _ASM_ATOMIC_H */
diff --git a/arch/frv/include/asm/atomic_defs.h b/arch/frv/include/asm/atomic_defs.h
new file mode 100644
index 0000000..36e126d
--- /dev/null
+++ b/arch/frv/include/asm/atomic_defs.h
@@ -0,0 +1,172 @@
+
+#include <asm/spr-regs.h>
+
+#ifdef __ATOMIC_LIB__
+
+#ifdef CONFIG_FRV_OUTOFLINE_ATOMIC_OPS
+
+#define ATOMIC_QUALS
+#define ATOMIC_EXPORT(x)	EXPORT_SYMBOL(x)
+
+#else /* !OUTOFLINE && LIB */
+
+#define ATOMIC_OP_RETURN(op)
+#define ATOMIC_FETCH_OP(op)
+
+#endif /* OUTOFLINE */
+
+#else /* !__ATOMIC_LIB__ */
+
+#define ATOMIC_EXPORT(x)
+
+#ifdef CONFIG_FRV_OUTOFLINE_ATOMIC_OPS
+
+#define ATOMIC_OP_RETURN(op)						\
+extern int __atomic_##op##_return(int i, int *v);			\
+extern long long __atomic64_##op##_return(long long i, long long *v);
+
+#define ATOMIC_FETCH_OP(op)						\
+extern int __atomic32_fetch_##op(int i, int *v);			\
+extern long long __atomic64_fetch_##op(long long i, long long *v);
+
+#else /* !OUTOFLINE && !LIB */
+
+#define ATOMIC_QUALS	static inline
+
+#endif /* OUTOFLINE */
+#endif /* __ATOMIC_LIB__ */
+
+
+/*
+ * Note on the 64 bit inline asm variants...
+ *
+ * CSTD is a conditional instruction and needs a constrained memory reference.
+ * Normally 'U' provides the correct constraints for conditional instructions
+ * and this is used for the 32 bit version, however 'U' does not appear to work
+ * for 64 bit values (gcc-4.9)
+ *
+ * The exact constraint is that conditional instructions cannot deal with an
+ * immediate displacement in the memory reference, so what we do is we read the
+ * address through a volatile cast into a local variable in order to insure we
+ * _have_ to compute the correct address without displacement. This allows us
+ * to use the regular 'm' for the memory address.
+ *
+ * Furthermore, the %Ln operand, which prints the low word register (r+1),
+ * really only works for registers, this means we cannot allow immediate values
+ * for the 64 bit versions -- like we do for the 32 bit ones.
+ *
+ */
+
+#ifndef ATOMIC_OP_RETURN
+#define ATOMIC_OP_RETURN(op)						\
+ATOMIC_QUALS int __atomic_##op##_return(int i, int *v)			\
+{									\
+	int val;							\
+									\
+	asm volatile(							\
+	    "0:						\n"		\
+	    "	orcc		gr0,gr0,gr0,icc3	\n"		\
+	    "	ckeq		icc3,cc7		\n"		\
+	    "	ld.p		%M0,%1			\n"		\
+	    "	orcr		cc7,cc7,cc3		\n"		\
+	    "   "#op"%I2	%1,%2,%1		\n"		\
+	    "	cst.p		%1,%M0		,cc3,#1	\n"		\
+	    "	corcc		gr29,gr29,gr0	,cc3,#1	\n"		\
+	    "	beq		icc3,#0,0b		\n"		\
+	    : "+U"(*v), "=&r"(val)					\
+	    : "NPr"(i)							\
+	    : "memory", "cc7", "cc3", "icc3"				\
+	    );								\
+									\
+	return val;							\
+}									\
+ATOMIC_EXPORT(__atomic_##op##_return);					\
+									\
+ATOMIC_QUALS long long __atomic64_##op##_return(long long i, long long *v)	\
+{									\
+	long long *__v = READ_ONCE(v);					\
+	long long val;							\
+									\
+	asm volatile(							\
+	    "0:						\n"		\
+	    "	orcc		gr0,gr0,gr0,icc3	\n"		\
+	    "	ckeq		icc3,cc7		\n"		\
+	    "	ldd.p		%M0,%1			\n"		\
+	    "	orcr		cc7,cc7,cc3		\n"		\
+	    "   "#op"cc		%L1,%L2,%L1,icc0	\n"		\
+	    "   "#op"x		%1,%2,%1,icc0		\n"		\
+	    "	cstd.p		%1,%M0		,cc3,#1	\n"		\
+	    "	corcc		gr29,gr29,gr0	,cc3,#1	\n"		\
+	    "	beq		icc3,#0,0b		\n"		\
+	    : "+m"(*__v), "=&e"(val)					\
+	    : "e"(i)							\
+	    : "memory", "cc7", "cc3", "icc0", "icc3"			\
+	    );								\
+									\
+	return val;							\
+}									\
+ATOMIC_EXPORT(__atomic64_##op##_return);
+#endif
+
+#ifndef ATOMIC_FETCH_OP
+#define ATOMIC_FETCH_OP(op)						\
+ATOMIC_QUALS int __atomic32_fetch_##op(int i, int *v)			\
+{									\
+	int old, tmp;							\
+									\
+	asm volatile(							\
+		"0:						\n"	\
+		"	orcc		gr0,gr0,gr0,icc3	\n"	\
+		"	ckeq		icc3,cc7		\n"	\
+		"	ld.p		%M0,%1			\n"	\
+		"	orcr		cc7,cc7,cc3		\n"	\
+		"	"#op"%I3	%1,%3,%2		\n"	\
+		"	cst.p		%2,%M0		,cc3,#1	\n"	\
+		"	corcc		gr29,gr29,gr0	,cc3,#1	\n"	\
+		"	beq		icc3,#0,0b		\n"	\
+		: "+U"(*v), "=&r"(old), "=r"(tmp)			\
+		: "NPr"(i)						\
+		: "memory", "cc7", "cc3", "icc3"			\
+		);							\
+									\
+	return old;							\
+}									\
+ATOMIC_EXPORT(__atomic32_fetch_##op);					\
+									\
+ATOMIC_QUALS long long __atomic64_fetch_##op(long long i, long long *v)	\
+{									\
+	long long *__v = READ_ONCE(v);					\
+	long long old, tmp;						\
+									\
+	asm volatile(							\
+		"0:						\n"	\
+		"	orcc		gr0,gr0,gr0,icc3	\n"	\
+		"	ckeq		icc3,cc7		\n"	\
+		"	ldd.p		%M0,%1			\n"	\
+		"	orcr		cc7,cc7,cc3		\n"	\
+		"	"#op"		%L1,%L3,%L2		\n"	\
+		"	"#op"		%1,%3,%2		\n"	\
+		"	cstd.p		%2,%M0		,cc3,#1	\n"	\
+		"	corcc		gr29,gr29,gr0	,cc3,#1	\n"	\
+		"	beq		icc3,#0,0b		\n"	\
+		: "+m"(*__v), "=&e"(old), "=e"(tmp)			\
+		: "e"(i)						\
+		: "memory", "cc7", "cc3", "icc3"			\
+		);							\
+									\
+	return old;							\
+}									\
+ATOMIC_EXPORT(__atomic64_fetch_##op);
+#endif
+
+ATOMIC_FETCH_OP(or)
+ATOMIC_FETCH_OP(and)
+ATOMIC_FETCH_OP(xor)
+
+ATOMIC_OP_RETURN(add)
+ATOMIC_OP_RETURN(sub)
+
+#undef ATOMIC_FETCH_OP
+#undef ATOMIC_OP_RETURN
+#undef ATOMIC_QUALS
+#undef ATOMIC_EXPORT
diff --git a/arch/frv/include/asm/bitops.h b/arch/frv/include/asm/bitops.h
index 96de220..0df8e95 100644
--- a/arch/frv/include/asm/bitops.h
+++ b/arch/frv/include/asm/bitops.h
@@ -25,109 +25,30 @@
 
 #include <asm-generic/bitops/ffz.h>
 
-#ifndef CONFIG_FRV_OUTOFLINE_ATOMIC_OPS
-static inline
-unsigned long atomic_test_and_ANDNOT_mask(unsigned long mask, volatile unsigned long *v)
-{
-	unsigned long old, tmp;
-
-	asm volatile(
-		"0:						\n"
-		"	orcc		gr0,gr0,gr0,icc3	\n"	/* set ICC3.Z */
-		"	ckeq		icc3,cc7		\n"
-		"	ld.p		%M0,%1			\n"	/* LD.P/ORCR are atomic */
-		"	orcr		cc7,cc7,cc3		\n"	/* set CC3 to true */
-		"	and%I3		%1,%3,%2		\n"
-		"	cst.p		%2,%M0		,cc3,#1	\n"	/* if store happens... */
-		"	corcc		gr29,gr29,gr0	,cc3,#1	\n"	/* ... clear ICC3.Z */
-		"	beq		icc3,#0,0b		\n"
-		: "+U"(*v), "=&r"(old), "=r"(tmp)
-		: "NPr"(~mask)
-		: "memory", "cc7", "cc3", "icc3"
-		);
-
-	return old;
-}
-
-static inline
-unsigned long atomic_test_and_OR_mask(unsigned long mask, volatile unsigned long *v)
-{
-	unsigned long old, tmp;
-
-	asm volatile(
-		"0:						\n"
-		"	orcc		gr0,gr0,gr0,icc3	\n"	/* set ICC3.Z */
-		"	ckeq		icc3,cc7		\n"
-		"	ld.p		%M0,%1			\n"	/* LD.P/ORCR are atomic */
-		"	orcr		cc7,cc7,cc3		\n"	/* set CC3 to true */
-		"	or%I3		%1,%3,%2		\n"
-		"	cst.p		%2,%M0		,cc3,#1	\n"	/* if store happens... */
-		"	corcc		gr29,gr29,gr0	,cc3,#1	\n"	/* ... clear ICC3.Z */
-		"	beq		icc3,#0,0b		\n"
-		: "+U"(*v), "=&r"(old), "=r"(tmp)
-		: "NPr"(mask)
-		: "memory", "cc7", "cc3", "icc3"
-		);
-
-	return old;
-}
-
-static inline
-unsigned long atomic_test_and_XOR_mask(unsigned long mask, volatile unsigned long *v)
-{
-	unsigned long old, tmp;
-
-	asm volatile(
-		"0:						\n"
-		"	orcc		gr0,gr0,gr0,icc3	\n"	/* set ICC3.Z */
-		"	ckeq		icc3,cc7		\n"
-		"	ld.p		%M0,%1			\n"	/* LD.P/ORCR are atomic */
-		"	orcr		cc7,cc7,cc3		\n"	/* set CC3 to true */
-		"	xor%I3		%1,%3,%2		\n"
-		"	cst.p		%2,%M0		,cc3,#1	\n"	/* if store happens... */
-		"	corcc		gr29,gr29,gr0	,cc3,#1	\n"	/* ... clear ICC3.Z */
-		"	beq		icc3,#0,0b		\n"
-		: "+U"(*v), "=&r"(old), "=r"(tmp)
-		: "NPr"(mask)
-		: "memory", "cc7", "cc3", "icc3"
-		);
-
-	return old;
-}
-
-#else
-
-extern unsigned long atomic_test_and_ANDNOT_mask(unsigned long mask, volatile unsigned long *v);
-extern unsigned long atomic_test_and_OR_mask(unsigned long mask, volatile unsigned long *v);
-extern unsigned long atomic_test_and_XOR_mask(unsigned long mask, volatile unsigned long *v);
-
-#endif
-
-#define atomic_clear_mask(mask, v)	atomic_test_and_ANDNOT_mask((mask), (v))
-#define atomic_set_mask(mask, v)	atomic_test_and_OR_mask((mask), (v))
+#include <asm/atomic.h>
 
 static inline int test_and_clear_bit(unsigned long nr, volatile void *addr)
 {
-	volatile unsigned long *ptr = addr;
-	unsigned long mask = 1UL << (nr & 31);
+	unsigned int *ptr = (void *)addr;
+	unsigned int mask = 1UL << (nr & 31);
 	ptr += nr >> 5;
-	return (atomic_test_and_ANDNOT_mask(mask, ptr) & mask) != 0;
+	return (__atomic32_fetch_and(~mask, ptr) & mask) != 0;
 }
 
 static inline int test_and_set_bit(unsigned long nr, volatile void *addr)
 {
-	volatile unsigned long *ptr = addr;
-	unsigned long mask = 1UL << (nr & 31);
+	unsigned int *ptr = (void *)addr;
+	unsigned int mask = 1UL << (nr & 31);
 	ptr += nr >> 5;
-	return (atomic_test_and_OR_mask(mask, ptr) & mask) != 0;
+	return (__atomic32_fetch_or(mask, ptr) & mask) != 0;
 }
 
 static inline int test_and_change_bit(unsigned long nr, volatile void *addr)
 {
-	volatile unsigned long *ptr = addr;
-	unsigned long mask = 1UL << (nr & 31);
+	unsigned int *ptr = (void *)addr;
+	unsigned int mask = 1UL << (nr & 31);
 	ptr += nr >> 5;
-	return (atomic_test_and_XOR_mask(mask, ptr) & mask) != 0;
+	return (__atomic32_fetch_xor(mask, ptr) & mask) != 0;
 }
 
 static inline void clear_bit(unsigned long nr, volatile void *addr)