runtime: fix cputicks on x86

See the following issue for context:
#9729 (comment)
In short, RDTSC can produce skewed results without preceding LFENCE/MFENCE.
Information on this matter is very scrappy in the internet.
But this is what linux kernel does (see rdtsc_barrier).
It also fixes the test program on my machine.

Update #9729

Change-Id: I3c1ffbf129fdfdd388bd5b7911b392b319248e68
Reviewed-on: https://go-review.googlesource.com/5033
Reviewed-by: Ian Lance Taylor <[email protected]>

Author: dvyukov

src/runtime/asm_386.s

@@ -30,6 +30,19 @@ TEXT runtime·rt0_go(SB),NOSPLIT,$0
 	CPUID
 	CMPL	AX, $0
 	JE	nocpuinfo
+
+	// Figure out how to serialize RDTSC.
+	// On Intel processors LFENCE is enough. AMD requires MFENCE.
+	// Don't know about the rest, so let's do MFENCE.
+	CMPL	BX, $0x756E6547  // "Genu"
+	JNE	notintel
+	CMPL	DX, $0x49656E69  // "ineI"
+	JNE	notintel
+	CMPL	CX, $0x6C65746E  // "ntel"
+	JNE	notintel
+	MOVB	$1, runtime·lfenceBeforeRdtsc(SB)
+notintel:
+
 	MOVL	$1, AX
 	CPUID
 	MOVL	CX, runtime·cpuid_ecx(SB)
@@ -868,9 +881,17 @@ TEXT runtime·gogetcallersp(SB),NOSPLIT,$0-8
 	MOVL	AX, ret+4(FP)
 	RET
 
-// int64 runtime·cputicks(void), so really
-// void runtime·cputicks(int64 *ticks)
+// func cputicks() int64
 TEXT runtime·cputicks(SB),NOSPLIT,$0-8
+	TESTL	$0x4000000, runtime·cpuid_edx(SB) // no sse2, no mfence
+	JEQ	done
+	CMPB	runtime·lfenceBeforeRdtsc(SB), $1
+	JNE	mfence
+	BYTE	$0x0f; BYTE $0xae; BYTE $0xe8 // LFENCE
+	JMP	done
+mfence:
+	BYTE	$0x0f; BYTE $0xae; BYTE $0xf0 // MFENCE
+done:
 	RDTSC
 	MOVL	AX, ret_lo+0(FP)
 	MOVL	DX, ret_hi+4(FP)

src/runtime/asm_amd64.s

@@ -30,6 +30,19 @@ TEXT runtime·rt0_go(SB),NOSPLIT,$0
 	CPUID
 	CMPQ	AX, $0
 	JE	nocpuinfo
+
+	// Figure out how to serialize RDTSC.
+	// On Intel processors LFENCE is enough. AMD requires MFENCE.
+	// Don't know about the rest, so let's do MFENCE.
+	CMPL	BX, $0x756E6547  // "Genu"
+	JNE	notintel
+	CMPL	DX, $0x49656E69  // "ineI"
+	JNE	notintel
+	CMPL	CX, $0x6C65746E  // "ntel"
+	JNE	notintel
+	MOVB	$1, runtime·lfenceBeforeRdtsc(SB)
+notintel:
+
 	MOVQ	$1, AX
 	CPUID
 	MOVL	CX, runtime·cpuid_ecx(SB)
@@ -865,8 +878,15 @@ TEXT runtime·gogetcallersp(SB),NOSPLIT,$0-16
 	MOVQ	AX, ret+8(FP)
 	RET
 
-// int64 runtime·cputicks(void)
+// func cputicks() int64
 TEXT runtime·cputicks(SB),NOSPLIT,$0-0
+	CMPB	runtime·lfenceBeforeRdtsc(SB), $1
+	JNE	mfence
+	BYTE	$0x0f; BYTE $0xae; BYTE $0xe8 // LFENCE
+	JMP	done
+mfence:
+	BYTE	$0x0f; BYTE $0xae; BYTE $0xf0 // MFENCE
+done:
 	RDTSC
 	SHLQ	$32, DX
 	ADDQ	DX, AX

src/runtime/runtime1.go

@@ -58,13 +58,6 @@ var (
 	iswindows int32
 )
 
-// Information about what cpu features are available.
-// Set on startup in asm_{x86/amd64}.s.
-var (
-//cpuid_ecx uint32
-//cpuid_edx uint32
-)
-
 func goargs() {
 	if GOOS == "windows" {
 		return

src/runtime/runtime2.go

@@ -562,12 +562,16 @@ var (
 	goos        *int8
 	ncpu        int32
 	iscgo       bool
-	cpuid_ecx   uint32
-	cpuid_edx   uint32
 	signote     note
 	forcegc     forcegcstate
 	sched       schedt
 	newprocs    int32
+
+	// Information about what cpu features are available.
+	// Set on startup in asm_{x86,amd64}.s.
+	cpuid_ecx         uint32
+	cpuid_edx         uint32
+	lfenceBeforeRdtsc bool
 )
 
 /*