runtime, syscall: use the vdso page on linux x86 for faster syscalls instead of int $0x80.

8l: fix handling CALL $(constant) code generated by 8a.
8a,8l: add indirect call instruction: CALL *data(SB).

R=rsc, iant
CC=golang-dev
https://golang.org/cl/4817054

src/cmd/8a/a.y

@@ -210,6 +210,13 @@ spec3:	/* JMP/CALL */
 		$$.from = nullgen;
 		$$.to = $1;
 	}
+|	'*' nam
+	{
+		$$.from = nullgen;
+		$$.to = $2;
+		$$.to.index = $2.type;
+		$$.to.type = D_INDIR+D_ADDR;
+	}
 
 spec4:	/* NOP */
 	nonnon

src/cmd/8l/l.h

@@ -209,6 +209,7 @@ enum
 	Zbr,
 	Zcall,
 	Zcallcon,
+	Zcallind,
 	Zib_,
 	Zib_rp,
 	Zibo_m,

src/cmd/8l/optab.c

@@ -260,6 +260,7 @@ uchar	yloop[] =
 uchar	ycall[] =
 {
 	Ynone,	Yml,	Zo_m,	2,
+	Ynone,	Ycol,	Zcallind,	2,
 	Ynone,	Ybr,	Zcall,	0,
 	Ynone,	Yi32,	Zcallcon,	1,
 	0
@@ -383,7 +384,7 @@ Optab optab[] =
 	{ ABTSL,	yml_rl,	Pm, 0xab },
 	{ ABTSW,	yml_rl,	Pq, 0xab },
 	{ ABYTE,	ybyte,	Px, 1 },
-	{ ACALL,	ycall,	Px, 0xff,(02),0xe8 },
+	{ ACALL,	ycall,	Px, 0xff,(02),0xff,(0x15),0xe8 },
 	{ ACLC,		ynone,	Px, 0xf8 },
 	{ ACLD,		ynone,	Px, 0xfc },
 	{ ACLI,		ynone,	Px, 0xfa },

src/cmd/8l/pass.c

@@ -307,9 +307,12 @@ patch(void)
 					p->from.offset = 0;
 				}
 			}
-			if(p->as == ACALL || (p->as == AJMP && p->to.type != D_BRANCH)) {
+			if((p->as == ACALL && p->to.type != D_BRANCH) || (p->as == AJMP && p->to.type != D_BRANCH)) {
 				s = p->to.sym;
-				if(s) {
+				if(p->to.type == D_INDIR+D_ADDR) {
+					 /* skip check if this is an indirect call (CALL *symbol(SB)) */
+					 continue;
+				} else if(s) {
 					if(debug['c'])
 						Bprint(&bso, "%s calls %s\n", TNAME, s->name);
 					if((s->type&~SSUB) != STEXT) {

src/cmd/8l/span.c

@@ -285,6 +285,8 @@ oclass(Adr *a)
 				}
 				return Yxxx;
 			}
+			//if(a->type == D_INDIR+D_ADDR)
+			//	print("*Ycol\n");
 			return Ycol;
 		}
 		return Ym;
@@ -1143,6 +1145,18 @@ found:
 		r->add = p->to.offset;
 		put4(0);
 		break;
+	
+	case Zcallind:
+		*andptr++ = op;
+		*andptr++ = o->op[z+1];
+		r = addrel(cursym);
+		r->off = p->pc + andptr - and;
+		r->type = D_ADDR;
+		r->siz = 4;
+		r->add = p->to.offset;
+		r->sym = p->to.sym;
+		put4(0);
+		break;
 
 	case Zbyte:
 		v = vaddr(&p->from, &rel);

src/cmd/ld/data.c

@@ -182,7 +182,11 @@ relocsym(Sym *s)
 			o = symaddr(r->sym) + r->add;
 			break;
 		case D_PCREL:
-			o = symaddr(r->sym) + r->add - (s->value + r->off + r->siz);
+			// r->sym can be null when CALL $(constant) is transformed from absoulte PC to relative PC call.
+			o = 0;
+			if(r->sym)
+				o += symaddr(r->sym);
+			o += r->add - (s->value + r->off + r->siz);
 			break;
 		case D_SIZE:
 			o = r->sym->size + r->add;

src/pkg/runtime/linux/386/rt0.s

@@ -5,5 +5,13 @@
 // Darwin and Linux use the same linkage to main
 
 TEXT _rt0_386_linux(SB),7,$0
-	JMP	_rt0_386(SB)
+	CALL	runtime·linux_setup_vdso(SB)
+	JMP		_rt0_386(SB)
+
+TEXT _fallback_vdso(SB),7,$0
+	INT	$0x80
+	RET
+
+DATA	runtime·_vdso(SB)/4, $_fallback_vdso(SB)
+GLOBL	runtime·_vdso(SB), $4
 

src/pkg/runtime/linux/386/signal.c

@@ -182,3 +182,27 @@ os·sigpipe(void)
 	sigaction(SIGPIPE, SIG_DFL, false);
 	runtime·raisesigpipe();
 }
+
+#define AT_NULL		0
+#define AT_SYSINFO	32
+extern uint32 runtime·_vdso;
+
+#pragma textflag 7
+void runtime·linux_setup_vdso(int32 argc, void *argv_list)
+{
+	byte **argv = &argv_list;
+	byte **envp;
+	uint32 *auxv;
+
+	// skip envp to get to ELF auxiliary vector.
+	for(envp = &argv[argc+1]; *envp != nil; envp++)
+		;
+	envp++;
+	
+	for(auxv=(uint32*)envp; auxv[0] != AT_NULL; auxv += 2) {
+		if(auxv[0] == AT_SYSINFO) {
+			runtime·_vdso = auxv[1];
+			break;
+		}
+	}		
+}

src/pkg/runtime/linux/386/sys.s

@@ -11,14 +11,14 @@
 TEXT runtime·exit(SB),7,$0
 	MOVL	$252, AX	// syscall number
 	MOVL	4(SP), BX
-	INT	$0x80
+	CALL	*runtime·_vdso(SB)
 	INT $3	// not reached
 	RET
 
 TEXT runtime·exit1(SB),7,$0
 	MOVL	$1, AX	// exit - exit the current os thread
 	MOVL	4(SP), BX
-	INT	$0x80
+	CALL	*runtime·_vdso(SB)
 	INT $3	// not reached
 	RET
 
@@ -27,13 +27,13 @@ TEXT runtime·open(SB),7,$0
 	MOVL	4(SP), BX
 	MOVL	8(SP), CX
 	MOVL	12(SP), DX
-	INT	$0x80
+	CALL	*runtime·_vdso(SB)
 	RET
 
 TEXT runtime·close(SB),7,$0
 	MOVL	$6, AX		// syscall - close
 	MOVL	4(SP), BX
-	INT	$0x80
+	CALL	*runtime·_vdso(SB)
 	RET
 
 TEXT runtime·write(SB),7,$0
@@ -41,7 +41,7 @@ TEXT runtime·write(SB),7,$0
 	MOVL	4(SP), BX
 	MOVL	8(SP), CX
 	MOVL	12(SP), DX
-	INT	$0x80
+	CALL	*runtime·_vdso(SB)
 	RET
 
 TEXT runtime·read(SB),7,$0
@@ -49,16 +49,16 @@ TEXT runtime·read(SB),7,$0
 	MOVL	4(SP), BX
 	MOVL	8(SP), CX
 	MOVL	12(SP), DX
-	INT	$0x80
+	CALL	*runtime·_vdso(SB)
 	RET
 
 TEXT runtime·raisesigpipe(SB),7,$12
 	MOVL	$224, AX	// syscall - gettid
-	INT	$0x80
+	CALL	*runtime·_vdso(SB)
 	MOVL	AX, 0(SP)	// arg 1 tid
 	MOVL	$13, 4(SP)	// arg 2 SIGPIPE
 	MOVL	$238, AX	// syscall - tkill
-	INT	$0x80
+	CALL	*runtime·_vdso(SB)
 	RET
 
 TEXT runtime·setitimer(SB),7,$0-24
@@ -66,7 +66,7 @@ TEXT runtime·setitimer(SB),7,$0-24
 	MOVL	4(SP), BX
 	MOVL	8(SP), CX
 	MOVL	12(SP), DX
-	INT	$0x80
+	CALL	*runtime·_vdso(SB)
 	RET
 
 TEXT runtime·mincore(SB),7,$0-24
@@ -74,7 +74,7 @@ TEXT runtime·mincore(SB),7,$0-24
 	MOVL	4(SP), BX
 	MOVL	8(SP), CX
 	MOVL	12(SP), DX
-	INT	$0x80
+	CALL	*runtime·_vdso(SB)
 	RET
 
 TEXT runtime·gettime(SB), 7, $32
@@ -82,7 +82,7 @@ TEXT runtime·gettime(SB), 7, $32
 	LEAL	8(SP), BX
 	MOVL	$0, CX
 	MOVL	$0, DX
-	INT	$0x80
+	CALL	*runtime·_vdso(SB)
 
 	MOVL	8(SP), BX	// sec
 	MOVL	sec+0(FP), DI
@@ -100,7 +100,7 @@ TEXT runtime·rt_sigaction(SB),7,$0
 	MOVL	8(SP), CX
 	MOVL	12(SP), DX
 	MOVL	16(SP), SI
-	INT	$0x80
+	CALL	*runtime·_vdso(SB)
 	RET
 
 TEXT runtime·sigtramp(SB),7,$44
@@ -138,7 +138,9 @@ TEXT runtime·sigignore(SB),7,$0
 
 TEXT runtime·sigreturn(SB),7,$0
 	MOVL	$173, AX	// rt_sigreturn
-	INT $0x80
+	// Sigreturn expects same SP as signal handler,
+	// so cannot CALL *runtime._vsdo(SB) here.
+	INT	$0x80
 	INT $3	// not reached
 	RET
 
@@ -151,7 +153,7 @@ TEXT runtime·mmap(SB),7,$0
 	MOVL	20(SP), DI
 	MOVL	24(SP), BP
 	SHRL	$12, BP
-	INT	$0x80
+	CALL	*runtime·_vdso(SB)
 	CMPL	AX, $0xfffff001
 	JLS	3(PC)
 	NOTL	AX
@@ -162,7 +164,7 @@ TEXT runtime·munmap(SB),7,$0
 	MOVL	$91, AX	// munmap
 	MOVL	4(SP), BX
 	MOVL	8(SP), CX
-	INT	$0x80
+	CALL	*runtime·_vdso(SB)
 	CMPL	AX, $0xfffff001
 	JLS	2(PC)
 	INT $3
@@ -178,7 +180,7 @@ TEXT runtime·futex(SB),7,$0
 	MOVL	16(SP), SI
 	MOVL	20(SP), DI
 	MOVL	24(SP), BP
-	INT	$0x80
+	CALL	*runtime·_vdso(SB)
 	RET
 
 // int32 clone(int32 flags, void *stack, M *m, G *g, void (*fn)(void));
@@ -199,6 +201,10 @@ TEXT runtime·clone(SB),7,$0
 	MOVL	SI, 8(CX)
 	MOVL	$1234, 12(CX)
 
+	// cannot use CALL *runtime·_vdso(SB) here, because
+	// the stack changes during the system call (after 
+	// CALL *runtime·_vdso(SB), the child is still using
+	// the parent's stack when executing its RET instruction).
 	INT	$0x80
 
 	// In parent, return.
@@ -214,7 +220,7 @@ TEXT runtime·clone(SB),7,$0
 
 	// Initialize AX to Linux tid
 	MOVL	$224, AX
-	INT	$0x80
+	CALL	*runtime·_vdso(SB)
 
 	// In child on new stack.  Reload registers (paranoia).
 	MOVL	0(SP), BX	// m
@@ -262,7 +268,7 @@ TEXT runtime·sigaltstack(SB),7,$-8
 	MOVL	$186, AX	// sigaltstack
 	MOVL	new+4(SP), BX
 	MOVL	old+8(SP), CX
-	INT	$0x80
+	CALL	*runtime·_vdso(SB)
 	CMPL	AX, $0xfffff001
 	JLS	2(PC)
 	INT	$3
@@ -323,7 +329,7 @@ TEXT runtime·setldt(SB),7,$32
 	MOVL	AX, CX	// user_desc
 	MOVL	$16, DX	// sizeof(user_desc)
 	MOVL	$123, AX	// syscall - modify_ldt
-	INT	$0x80
+	CALL	*runtime·_vdso(SB)
 
 	// breakpoint on error
 	CMPL AX, $0xfffff001
@@ -340,5 +346,5 @@ TEXT runtime·setldt(SB),7,$32
 
 TEXT runtime·osyield(SB),7,$0
 	MOVL	$158, AX
-	INT	$0x80
+	CALL	*runtime·_vdso(SB)
 	RET

src/pkg/syscall/asm_linux_386.s

@@ -17,7 +17,7 @@ TEXT	·Syscall(SB),7,$0
 	MOVL	16(SP), DX
 	MOVL	$0, SI
 	MOVL	$0,  DI
-	INT	$0x80
+	CALL	*runtime·_vdso(SB)
 	CMPL	AX, $0xfffff001
 	JLS	ok
 	MOVL	$-1, 20(SP)	// r1
@@ -43,7 +43,7 @@ TEXT	·Syscall6(SB),7,$0
 	MOVL	20(SP), SI
 	MOVL	24(SP), DI
 	MOVL	28(SP), BP
-	INT	$0x80
+	CALL	*runtime·_vdso(SB)
 	CMPL	AX, $0xfffff001
 	JLS	ok6
 	MOVL	$-1, 32(SP)	// r1
@@ -67,7 +67,7 @@ TEXT ·RawSyscall(SB),7,$0
 	MOVL	16(SP), DX
 	MOVL	$0, SI
 	MOVL	$0,  DI
-	INT	$0x80
+	CALL	*runtime·_vdso(SB)
 	CMPL	AX, $0xfffff001
 	JLS	ok1
 	MOVL	$-1, 20(SP)	// r1
@@ -90,7 +90,7 @@ TEXT	·RawSyscall6(SB),7,$0
 	MOVL	20(SP), SI
 	MOVL	24(SP), DI
 	MOVL	28(SP), BP
-	INT	$0x80
+	CALL	*runtime·_vdso(SB)
 	CMPL	AX, $0xfffff001
 	JLS	ok2
 	MOVL	$-1, 32(SP)	// r1
@@ -116,7 +116,7 @@ TEXT ·socketcall(SB),7,$0
 	MOVL	$0, DX
 	MOVL	$0, SI
 	MOVL	$0,  DI
-	INT	$0x80
+	CALL	*runtime·_vdso(SB)
 	CMPL	AX, $0xfffff001
 	JLS	oksock
 	MOVL	$-1, 32(SP)	// n
@@ -139,7 +139,7 @@ TEXT ·rawsocketcall(SB),7,$0
 	MOVL	$0, DX
 	MOVL	$0, SI
 	MOVL	$0,  DI
-	INT	$0x80
+	CALL	*runtime·_vdso(SB)
 	CMPL	AX, $0xfffff001
 	JLS	oksock1
 	MOVL	$-1, 32(SP)	// n
@@ -165,7 +165,7 @@ TEXT ·Seek(SB),7,$0
 	MOVL	8(SP), DX	// offset-low
 	LEAL	20(SP), SI	// result pointer
 	MOVL	16(SP),  DI	// whence
-	INT	$0x80
+	CALL	*runtime·_vdso(SB)
 	CMPL	AX, $0xfffff001
 	JLS	okseek
 	MOVL	$-1, 20(SP)	// newoffset low