overhaul sh atomics for new atomics framework, add j-core cas.l backend

sh needs runtime-selected atomic backends since there are a number of
supported models that use non-forwards-compatible (non-smp-compatible)
atomic mechanisms. previously, the code paths for this were highly
inefficient since they involved C function calls with multiple
branches in the callee and heavy spills in the caller. the new code
performs calls the runtime-selected asm fragment from inline asm with
extremely minimal clobbers, rather than using a function call.

for the sh4a case where the atomic mechanism is known and there is no
forward-compatibility issue, the movli.l and movco.l instructions are
provided as a_ll and a_sc, allowing the new shared atomic.h to
generate efficient inline versions of all the basic atomic operations
without needing a cas loop.

1 files changed, 30 insertions, 80 deletions

diff --git a/arch/sh/atomic_arch.h b/arch/sh/atomic_arch.h
index 2ac77246..74444d5d 100644
--- a/arch/sh/atomic_arch.h
+++ b/arch/sh/atomic_arch.h
@@ -1,96 +1,46 @@
-#define LLSC_CLOBBERS "r0", "t", "memory"
-#define LLSC_START(mem) "synco\n"  \
-	"0:	movli.l @" mem ", r0\n"
-#define LLSC_END(mem)              \
-	"1:	movco.l r0, @" mem "\n"    \
-	"	bf 0b\n"                   \
-	"	synco\n"
+#if defined(__SH4A__)
 
-static inline int __sh_cas_llsc(volatile int *p, int t, int s)
+#define a_ll a_ll
+static inline int a_ll(volatile int *p)
 {
-	int old;
-	__asm__ __volatile__(
-		LLSC_START("%1")
-		"	mov r0, %0\n"
-		"	cmp/eq %0, %2\n"
-		"	bf 1f\n"
-		"	mov %3, r0\n"
-		LLSC_END("%1")
-		: "=&r"(old) : "r"(p), "r"(t), "r"(s) : LLSC_CLOBBERS);
-	return old;
+	int v;
+	__asm__ __volatile__ ("movli.l @%1, %0" : "=z"(v) : "r"(p), "m"(*p));
+	return v;
 }
 
-static inline int __sh_swap_llsc(volatile int *x, int v)
+#define a_sc a_sc
+static inline int a_sc(volatile int *p, int v)
 {
-	int old;
-	__asm__ __volatile__(
-		LLSC_START("%1")
-		"	mov r0, %0\n"
-		"	mov %2, r0\n"
-		LLSC_END("%1")
-		: "=&r"(old) : "r"(x), "r"(v) : LLSC_CLOBBERS);
-	return old;
+	int r;
+	__asm__ __volatile__ (
+		"movco.l %2, @%3 ; movt %0"
+		: "=r"(r), "=m"(*p) : "z"(v), "r"(p) : "memory", "cc");
+	return r;
 }
 
-static inline int __sh_fetch_add_llsc(volatile int *x, int v)
+#define a_barrier a_barrier
+static inline void a_barrier()
 {
-	int old;
-	__asm__ __volatile__(
-		LLSC_START("%1")
-		"	mov r0, %0\n"
-		"	add %2, r0\n"
-		LLSC_END("%1")
-		: "=&r"(old) : "r"(x), "r"(v) : LLSC_CLOBBERS);
-	return old;
+	__asm__ __volatile__ ("synco" : : "memory");
 }
 
-static inline void __sh_store_llsc(volatile int *p, int x)
-{
-	__asm__ __volatile__(
-		"	synco\n"
-		"	mov.l %1, @%0\n"
-		"	synco\n"
-		: : "r"(p), "r"(x) : "memory");
-}
+#define a_pre_llsc a_barrier
+#define a_post_llsc a_barrier
 
-static inline void __sh_and_llsc(volatile int *x, int v)
-{
-	__asm__ __volatile__(
-		LLSC_START("%0")
-		"	and %1, r0\n"
-		LLSC_END("%0")
-		: : "r"(x), "r"(v) : LLSC_CLOBBERS);
-}
+#else
 
-static inline void __sh_or_llsc(volatile int *x, int v)
+#define a_cas a_cas
+__attribute__((__visibility__("hidden"))) extern const void *__sh_cas_ptr;
+static inline int a_cas(volatile int *p, int t, int s)
 {
-	__asm__ __volatile__(
-		LLSC_START("%0")
-		"	or %1, r0\n"
-		LLSC_END("%0")
-		: : "r"(x), "r"(v) : LLSC_CLOBBERS);
+	register int r1 __asm__("r1");
+	register int r2 __asm__("r2") = t;
+	register int r3 __asm__("r3") = s;
+	__asm__ __volatile__ (
+		"jsr @%4 ; nop"
+		: "=r"(r1), "+r"(r3) : "z"(p), "r"(r2), "r"(__sh_cas_ptr)
+		: "memory", "pr", "cc");
+	return r3;
 }
 
-#ifdef __SH4A__
-#define a_cas(p,t,s)     __sh_cas_llsc(p,t,s)
-#define a_swap(x,v)      __sh_swap_llsc(x,v)
-#define a_fetch_add(x,v) __sh_fetch_add_llsc(x, v)
-#define a_store(x,v)     __sh_store_llsc(x, v)
-#define a_and(x,v)       __sh_and_llsc(x, v)
-#define a_or(x,v)        __sh_or_llsc(x, v)
-#else
-
-int  __sh_cas(volatile int *, int, int);
-int  __sh_swap(volatile int *, int);
-int  __sh_fetch_add(volatile int *, int);
-void __sh_store(volatile int *, int);
-void __sh_and(volatile int *, int);
-void __sh_or(volatile int *, int);
-
-#define a_cas(p,t,s)     __sh_cas(p,t,s)
-#define a_swap(x,v)      __sh_swap(x,v)
-#define a_fetch_add(x,v) __sh_fetch_add(x, v)
-#define a_store(x,v)     __sh_store(x, v)
-#define a_and(x,v)       __sh_and(x, v)
-#define a_or(x,v)        __sh_or(x, v)
 #endif