runtime: update openbsd runtime to use new tfork syscall

Update OpenBSD runtime to use the new version of the sys___tfork
syscall and switch TLS initialisation from sys_arch to sys___set_tcb
(note that both of these syscalls are available in OpenBSD 5.2).

R=golang-dev, minux.ma
CC=golang-dev
https://golang.org/cl/6843058

src/pkg/runtime/defs_openbsd_386.h

@@ -85,7 +85,7 @@ typedef struct Itimerval Itimerval;
 struct Tfork {
 	byte	*tf_tcb;
 	int32	*tf_tid;
-	int32	tf_flags;
+	byte	*tf_stack;
 };
 
 struct Sigaltstack {

src/pkg/runtime/defs_openbsd_amd64.h

@@ -85,8 +85,7 @@ typedef struct Itimerval Itimerval;
 struct Tfork {
 	byte	*tf_tcb;
 	int32	*tf_tid;
-	int32	tf_flags;
-	byte	Pad_cgo_0[4];
+	byte	*tf_stack;
 };
 
 struct Sigaltstack {

src/pkg/runtime/sys_openbsd_386.s

@@ -190,12 +190,14 @@ TEXT runtime·sigtramp(SB),7,$44
 	MOVL	$0xf1, 0xf1		// crash
 	RET
 
-// int32 tfork_thread(void *param, void *stack, M *m, G *g, void (*fn)(void));
-TEXT runtime·tfork_thread(SB),7,$8
+// int32 tfork(void *param, uintptr psize, M *m, G *g, void (*fn)(void));
+TEXT runtime·tfork(SB),7,$12
 
-	// Copy m, g, fn off parent stack and onto the child stack.
-	MOVL	stack+8(FP), CX
+	// Copy m, g and fn from the parent stack onto the child stack.
+	MOVL	params+4(FP), AX
+	MOVL	8(AX), CX		// tf_stack
 	SUBL	$16, CX
+	MOVL	CX, 8(AX)
 	MOVL	mm+12(FP), SI
 	MOVL	SI, 0(CX)
 	MOVL	gg+16(FP), SI
@@ -203,12 +205,13 @@ TEXT runtime·tfork_thread(SB),7,$8
 	MOVL	fn+20(FP), SI
 	MOVL	SI, 8(CX)
 	MOVL	$1234, 12(CX)
-	MOVL	CX, SI
 
 	MOVL	$0, 0(SP)		// syscall gap
 	MOVL	params+4(FP), AX
 	MOVL	AX, 4(SP)		// arg 1 - param
-	MOVL	$328, AX		// sys___tfork
+	MOVL	psize+8(FP), AX
+	MOVL	AX, 8(SP)		// arg 2 - psize
+	MOVL	$8, AX			// sys___tfork
 	INT	$0x80
 
 	// Return if tfork syscall failed.
@@ -225,9 +228,6 @@ TEXT runtime·tfork_thread(SB),7,$8
 	MOVL	AX, 0(DX)
 	RET
 
-	// In child, switch to new stack.
-	MOVL    SI, SP
-
 	// Paranoia: check that SP is as we expect.
 	MOVL	12(SP), BP
 	CMPL	BP, $1234
@@ -278,22 +278,20 @@ TEXT runtime·sigaltstack(SB),7,$-8
 	INT	$3
 	RET
 
-TEXT runtime·setldt(SB),7,$8
+TEXT runtime·setldt(SB),7,$4
 	// Under OpenBSD we set the GS base instead of messing with the LDT.
-	MOVL	16(SP), AX		// tls0
+	MOVL	tls0+4(FP), AX
 	MOVL	AX, 0(SP)
 	CALL	runtime·settls(SB)
 	RET
 
-TEXT runtime·settls(SB),7,$16
+TEXT runtime·settls(SB),7,$8
 	// adjust for ELF: wants to use -8(GS) and -4(GS) for g and m
-	MOVL	20(SP), CX
+	MOVL	tlsbase+0(FP), CX
 	ADDL	$8, CX
-	MOVL	CX, 0(CX)
 	MOVL	$0, 0(SP)		// syscall gap
-	MOVL	$9, 4(SP)		// I386_SET_GSBASE (machine/sysarch.h)
-	MOVL	CX, 8(SP)		// pointer to base
-	MOVL	$165, AX		// sys_sysarch
+	MOVL	CX, 4(SP)		// arg 1 - tcb
+	MOVL	$329, AX		// sys___set_tcb
 	INT	$0x80
 	JCC	2(PC)
 	MOVL	$0xf1, 0xf1		// crash

src/pkg/runtime/sys_openbsd_amd64.s

@@ -8,17 +8,17 @@
 
 #include "zasm_GOOS_GOARCH.h"
 
-// int64 tfork_thread(void *param, void *stack, M *m, G *g, void (*fn)(void));
-TEXT runtime·tfork_thread(SB),7,$32
+// int64 tfork(void *param, uintptr psize, M *m, G *g, void (*fn)(void));
+TEXT runtime·tfork(SB),7,$32
 
-	// Copy stack, m, g and fn off parent stack for use by child.
-	MOVQ	stack+8(FP), SI
+	// Copy m, g and fn off parent stack for use by child.
 	MOVQ	mm+16(FP), R8
 	MOVQ	gg+24(FP), R9
 	MOVQ	fn+32(FP), R12
 
 	MOVQ	param+0(FP), DI
-	MOVL	$328, AX		// sys___tfork
+	MOVQ	psize+8(FP), SI
+	MOVL	$8, AX			// sys___tfork
 	SYSCALL
 
 	// Return if tfork syscall failed.
@@ -31,9 +31,6 @@ TEXT runtime·tfork_thread(SB),7,$32
 	JEQ	2(PC)
 	RET
 
-	// In child, switch to new stack.
-	MOVQ	SI, SP
-
 	// Set FS to point at m->tls.
 	LEAQ	m_tls(R8), DI
 	CALL	runtime·settls(SB)
@@ -54,7 +51,7 @@ TEXT runtime·tfork_thread(SB),7,$32
 	JMP	-3(PC)			// keep exiting
 
 TEXT runtime·osyield(SB),7,$0
-	MOVL $298, AX			// sys_sched_yield
+	MOVL	$298, AX		// sys_sched_yield
 	SYSCALL
 	RET
 
@@ -249,13 +246,10 @@ TEXT runtime·sigaltstack(SB),7,$-8
 	RET
 
 // set tls base to DI
-TEXT runtime·settls(SB),7,$8
+TEXT runtime·settls(SB),7,$0
 	// adjust for ELF: wants to use -16(FS) and -8(FS) for g and m
 	ADDQ	$16, DI
-	MOVQ	DI, 0(SP)
-	MOVQ	SP, SI
-	MOVQ	$12, DI			// AMD64_SET_FSBASE (machine/sysarch.h)
-	MOVQ	$165, AX		// sys_sysarch
+	MOVQ	$329, AX		// sys___settcb
 	SYSCALL
 	JCC	2(PC)
 	MOVL	$0xf1, 0xf1		// crash

src/pkg/runtime/thread_openbsd.c

@@ -23,7 +23,7 @@ extern SigTab runtime·sigtab[];
 static Sigset sigset_all = ~(Sigset)0;
 static Sigset sigset_none;
 
-extern int64 runtime·tfork_thread(void *param, void *stack, M *m, G *g, void (*fn)(void));
+extern int64 runtime·tfork(void *param, uintptr psize, M *m, G *g, void (*fn)(void));
 extern int32 runtime·thrsleep(void *ident, int32 clock_id, void *tsp, void *lock, const int32 *abort);
 extern int32 runtime·thrwakeup(void *ident, int32 n);
 
@@ -139,10 +139,10 @@ runtime·newosproc(M *m, G *g, void *stk, void (*fn)(void))
 
 	param.tf_tcb = (byte*)&m->tls[0];
 	param.tf_tid = (int32*)&m->procid;
-	param.tf_flags = (int32)0;
+	param.tf_stack = stk;
 
 	oset = runtime·sigprocmask(SIG_SETMASK, sigset_all);
-	ret = runtime·tfork_thread((byte*)&param, stk, m, g, fn);
+	ret = runtime·tfork((byte*)&param, sizeof(param), m, g, fn);
 	runtime·sigprocmask(SIG_SETMASK, oset);
 
 	if(ret < 0) {