runtime: on windows, read nanotime with one instruction or issue barrier

On 64-bit, this is more efficient, and on ARM64, this prevents the time
from moving backwards due to the weaker memory model. On ARM32 due to
the weaker memory model, we issue a memory barrier.

Updates #48072.

Change-Id: If4695716c3039d8af14e14808af217f5c99fc93a
Reviewed-on: https://go-review.googlesource.com/c/go/+/361057
Trust: Jason A. Donenfeld <Jason@zx2c4.com>
Run-TryBot: Jason A. Donenfeld <Jason@zx2c4.com>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Austin Clements <austin@google.com>

Author: zx2c4

src/runtime/sys_windows_amd64.s

@@ -344,16 +344,9 @@ TEXT runtime·nanotime1(SB),NOSPLIT,$0-8
 	CMPB	runtime·useQPCTime(SB), $0
 	JNE	useQPC
 	MOVQ	$_INTERRUPT_TIME, DI
-loop:
-	MOVL	time_hi1(DI), AX
-	MOVL	time_lo(DI), BX
-	MOVL	time_hi2(DI), CX
-	CMPL	AX, CX
-	JNE	loop
-	SHLQ	$32, CX
-	ORQ	BX, CX
-	IMULQ	$100, CX
-	MOVQ	CX, ret+0(FP)
+	MOVQ	time_lo(DI), AX
+	IMULQ	$100, AX
+	MOVQ	AX, ret+0(FP)
 	RET
 useQPC:
 	JMP	runtime·nanotimeQPC(SB)

src/runtime/sys_windows_arm.s

@@ -350,7 +350,9 @@ TEXT runtime·nanotime1(SB),NOSPLIT|NOFRAME,$0-8
 	MOVW	$_INTERRUPT_TIME, R3
 loop:
 	MOVW	time_hi1(R3), R1
+	DMB	MB_ISH
 	MOVW	time_lo(R3), R0
+	DMB	MB_ISH
 	MOVW	time_hi2(R3), R2
 	CMP	R1, R2
 	BNE	loop

src/runtime/sys_windows_arm64.s

@@ -415,15 +415,7 @@ TEXT runtime·nanotime1(SB),NOSPLIT|NOFRAME,$0-8
 	CMP	$0, R0
 	BNE	useQPC
 	MOVD	$_INTERRUPT_TIME, R3
-loop:
-	MOVWU	time_hi1(R3), R1
-	MOVWU	time_lo(R3), R0
-	MOVWU	time_hi2(R3), R2
-	CMP	R1, R2
-	BNE	loop
-
-	// wintime = R1:R0, multiply by 100
-	ORR	R1<<32, R0
+	MOVD	time_lo(R3), R0
 	MOVD	$100, R1
 	MUL	R1, R0
 	MOVD	R0, ret+0(FP)

src/runtime/time_windows.h

@@ -9,6 +9,7 @@
 // http://web.archive.org/web/20210411000829/https://wrkhpi.wordpress.com/2007/08/09/getting-os-information-the-kuser_shared_data-structure/
 
 // Must read hi1, then lo, then hi2. The snapshot is valid if hi1 == hi2.
+// Or, on 64-bit, just read lo:hi1 all at once atomically.
 #define _INTERRUPT_TIME 0x7ffe0008
 #define _SYSTEM_TIME 0x7ffe0014
 #define time_lo 0

src/runtime/time_windows_amd64.s

@@ -12,27 +12,14 @@
 TEXT time·now(SB),NOSPLIT,$0-24
 	CMPB	runtime·useQPCTime(SB), $0
 	JNE	useQPC
+
 	MOVQ	$_INTERRUPT_TIME, DI
-loop:
-	MOVL	time_hi1(DI), AX
-	MOVL	time_lo(DI), BX
-	MOVL	time_hi2(DI), CX
-	CMPL	AX, CX
-	JNE	loop
-	SHLQ	$32, AX
-	ORQ	BX, AX
+	MOVQ	time_lo(DI), AX
 	IMULQ	$100, AX
 	MOVQ	AX, mono+16(FP)
 
 	MOVQ	$_SYSTEM_TIME, DI
-wall:
-	MOVL	time_hi1(DI), AX
-	MOVL	time_lo(DI), BX
-	MOVL	time_hi2(DI), CX
-	CMPL	AX, CX
-	JNE	wall
-	SHLQ	$32, AX
-	ORQ	BX, AX
+	MOVQ	time_lo(DI), AX
 	MOVQ	$116444736000000000, DI
 	SUBQ	DI, AX
 	IMULQ	$100, AX

src/runtime/time_windows_arm.s

@@ -17,7 +17,9 @@ TEXT time·now(SB),NOSPLIT|NOFRAME,$0-20
 	MOVW	$_INTERRUPT_TIME, R3
 loop:
 	MOVW	time_hi1(R3), R1
+	DMB	MB_ISH
 	MOVW	time_lo(R3), R0
+	DMB	MB_ISH
 	MOVW	time_hi2(R3), R2
 	CMP	R1, R2
 	BNE	loop
@@ -34,7 +36,9 @@ loop:
 	MOVW	$_SYSTEM_TIME, R3
 wall:
 	MOVW	time_hi1(R3), R1
+	DMB	MB_ISH
 	MOVW	time_lo(R3), R0
+	DMB	MB_ISH
 	MOVW	time_hi2(R3), R2
 	CMP	R1, R2
 	BNE	wall

src/runtime/time_windows_arm64.s

@@ -13,34 +13,18 @@ TEXT time·now(SB),NOSPLIT|NOFRAME,$0-24
 	MOVB    runtime·useQPCTime(SB), R0
 	CMP	$0, R0
 	BNE	useQPC
-	MOVD	$_INTERRUPT_TIME, R3
-loop:
-	MOVWU	time_hi1(R3), R1
-	MOVWU	time_lo(R3), R0
-	MOVWU	time_hi2(R3), R2
-	CMP	R1, R2
-	BNE	loop
 
-	// wintime = R1:R0, multiply by 100
-	ORR	R1<<32, R0
+	MOVD	$_INTERRUPT_TIME, R3
+	MOVD	time_lo(R3), R0
 	MOVD	$100, R1
 	MUL	R1, R0
 	MOVD	R0, mono+16(FP)
 
 	MOVD	$_SYSTEM_TIME, R3
-wall:
-	MOVWU	time_hi1(R3), R1
-	MOVWU	time_lo(R3), R0
-	MOVWU	time_hi2(R3), R2
-	CMP	R1, R2
-	BNE	wall
-
-	// w = R1:R0 in 100ns units
+	MOVD	time_lo(R3), R0
 	// convert to Unix epoch (but still 100ns units)
 	#define delta 116444736000000000
-	ORR	R1<<32, R0
 	SUB	$delta, R0
-
 	// Convert to nSec
 	MOVD	$100, R1
 	MUL	R1, R0