reverting semaphore lock changes from sources (r41ccd6d221da, rb28756e5ba29)

semaphore locks have much higher overhead than initially presented
in the "Semaphores in Plan9" paper. until the reason for it has been
found out i will revert the changes.

3 files changed, 14 insertions, 145 deletions

diff --git a/sys/src/ape/lib/ap/386/atom.s b/sys/src/ape/lib/ap/386/atom.s
deleted file mode 100755
index 5f51daf81..000000000
--- a/sys/src/ape/lib/ap/386/atom.s
+++ /dev/null
@@ -1,121 +0,0 @@
-TEXT ainc(SB), 1, $-4				/* int ainc(int*); */
-	MOVL	arg+0(FP), BX
-	MOVL	$1, AX
-	LOCK; BYTE $0x0f; BYTE $0xc1; BYTE $0x03/* XADDL AX, (BX) */
-	ADDL	$1, AX				/* overflow if -ve or 0 */
-	RET
-
-TEXT adec(SB), 1, $-4				/* int adec(int*); */
-	MOVL	arg+0(FP), BX
-	MOVL	$-1, AX
-	LOCK; BYTE $0x0f; BYTE $0xc1; BYTE $0x03/* XADDL AX, (BX) */
-	SUBL	$1, AX				/* underflow if -ve */
-	RET
-
-/*
- * int cas32(u32int *p, u32int ov, u32int nv);
- * int cas(uint *p, int ov, int nv);
- * int casp(void **p, void *ov, void *nv);
- * int casl(ulong *p, ulong ov, ulong nv);
- */
-
-/*
- * CMPXCHG (CX), DX: 0000 1111 1011 000w oorr rmmm,
- * mmm = CX = 001; rrr = DX = 010
- */
-
-#define CMPXCHG		BYTE $0x0F; BYTE $0xB1; BYTE $0x11
-
-TEXT	cas32+0(SB),0,$0
-TEXT	cas+0(SB),0,$0
-TEXT	casp+0(SB),0,$0
-TEXT	casl+0(SB),0,$0
-	MOVL	p+0(FP), CX
-	MOVL	ov+4(FP), AX
-	MOVL	nv+8(FP), DX
-	LOCK
-	CMPXCHG
-	JNE	fail
-	MOVL	$1,AX
-	RET
-fail:
-	MOVL	$0,AX
-	RET
-
-/*
- * int cas64(u64int *p, u64int ov, u64int nv);
- */
-
-/*
- * CMPXCHG64 (DI): 0000 1111 1100 0111 0000 1110,
- */
-
-#define CMPXCHG64		BYTE $0x0F; BYTE $0xC7; BYTE $0x0F
-
-TEXT	cas64+0(SB),0,$0
-	MOVL	p+0(FP), DI
-	MOVL	ov+0x4(FP), AX
-	MOVL	ov+0x8(FP), DX
-	MOVL	nv+0xc(FP), BX
-	MOVL	nv+0x10(FP), CX
-	LOCK
-	CMPXCHG64
-	JNE	fail
-	MOVL	$1,AX
-	RET
-
-/*
- * Versions of compare-and-swap that return the old value
- * (i.e., the value of *p at the time of the operation
- * 	xcas(p, o, n) == o
- * yields the same value as
- *	cas(p, o, n)
- * xcas can be used in constructs like
- *	for(o = *p; (oo = xcas(p, o, o+1)) != o; o = oo)
- *		;
- * to avoid the extra dereference of *p (the example is a silly
- * way to increment *p atomically)
- *
- * u32int	xcas32(u32int *p, u32int ov, u32int nv);
- * u64int	xcas64(u64int *p, u64int ov, u64int nv);
- * int		xcas(int *p, int ov, int nv);
- * void*	xcasp(void **p, void *ov, void *nv);
- * ulong	xcasl(ulong *p, ulong ov, ulong nv);
- */
-
-TEXT	xcas32+0(SB),0,$0
-TEXT	xcas+0(SB),0,$0
-TEXT	xcasp+0(SB),0,$0
-TEXT	xcasl+0(SB),0,$0
-	MOVL	p+0(FP), CX
-	MOVL	ov+4(FP), AX	/* accumulator */
-	MOVL	nv+8(FP), DX
-	LOCK
-	CMPXCHG
-	RET
-	
-/*
- * The CMPXCHG8B instruction also requires three operands:
- * a 64-bit value in EDX:EAX, a 64-bit value in ECX:EBX,
- * and a destination operand in memory. The instruction compar
- * es the 64-bit value in the EDX:EAX registers with the
- * destination operand. If they are equal, the 64-bit value
- * in the ECX:EBX register is stored in the destination
- * operand. If the EDX:EAX register and the destination ar
- * e not equal, the destination is loaded in the EDX:EAX
- * register. The CMPXCHG8B instruction can be combined with
- * the LOCK prefix to perform the operation atomically
- */
-
-TEXT	xcas64+0(SB),0,$0
-	MOVL	p+4(FP), DI
-	MOVL	ov+0x8(FP), AX
-	MOVL	ov+0xc(FP), DX
-	MOVL	nv+0x10(FP), BX
-	MOVL	nv+0x14(FP), CX
-	LOCK
-	CMPXCHG64
-	MOVL	.ret+0x0(FP),CX	/* pointer to return value */
-	MOVL	AX,0x0(CX)
-	MOVL	DX,0x4(CX)
-	RET
diff --git a/sys/src/ape/lib/ap/386/lock.c b/sys/src/ape/lib/ap/386/lock.c
index 65a65df7b..91c0ba233 100644
--- a/sys/src/ape/lib/ap/386/lock.c
+++ b/sys/src/ape/lib/ap/386/lock.c
@@ -1,35 +1,26 @@
-#include "../plan9/lib.h"
-#include "../plan9/sys9.h"
 #define _LOCK_EXTENSION
+#include "../plan9/sys9.h"
 #include <lock.h>
 
+int	tas(int*);
+
 void
-lock(Lock *l)
+lock(Lock *lk)
 {
-	if(ainc(&l->key) == 1)
-		return;	/* changed from 0 -> 1: we hold lock */
-	/* otherwise wait in kernel */
-	while(_SEMACQUIRE(&l->sem, 1) < 0){
-		/* interrupted; try again */
-	}
+	while(tas(&lk->val))
+		_SLEEP(0);
 }
 
-void
-unlock(Lock *l)
+int
+canlock(Lock *lk)
 {
-	if(adec(&l->key) == 0)
-		return;	/* changed from 1 -> 0: no contention */
-	_SEMRELEASE(&l->sem, 1);
+	if(tas(&lk->val))
+		return 0;
+	return 1;
 }
 
-int
-canlock(Lock *l)
+void
+unlock(Lock *lk)
 {
-	if(ainc(&l->key) == 1)
-		return 1;	/* changed from 0 -> 1: success */
-	/* Undo increment (but don't miss wakeup) */
-	if(adec(&l->key) == 0)
-		return 0;	/* changed from 1 -> 0: no contention */
-	_SEMRELEASE(&l->sem, 1);
-	return 0;
+	lk->val = 0;
 }
diff --git a/sys/src/ape/lib/ap/386/mkfile b/sys/src/ape/lib/ap/386/mkfile
index 680c7b4e4..e19a4b852 100644
--- a/sys/src/ape/lib/ap/386/mkfile
+++ b/sys/src/ape/lib/ap/386/mkfile
@@ -2,7 +2,6 @@ APE=/sys/src/ape
 <$APE/config
 LIB=/$objtype/lib/ape/libap.a
 OFILES=\
-	atom.$O\
 	cycles.$O\
 	lock.$O\
 	main9.$O\