runtime: use 4 byte writes in amd64p32 memmove/memclr

Currently, amd64p32's memmove and memclr use 8 byte writes as much as
possible and 1 byte writes for the tail of the object. However, if an
object ends with a 4 byte pointer at an 8 byte aligned offset, this
may copy/zero the pointer field one byte at a time, allowing the
garbage collector to observe a partially copied pointer.

Fix this by using 4 byte writes instead of 8 byte writes.

Updates #12552.

Change-Id: I13324fd05756fb25ae57e812e836f0a975b5595c
Reviewed-on: https://go-review.googlesource.com/15370
Run-TryBot: Austin Clements <[email protected]>
TryBot-Result: Gobot Gobot <[email protected]>
Reviewed-by: Brad Fitzpatrick <[email protected]>
Reviewed-by: Keith Randall <[email protected]>

Author: aclements

src/runtime/asm_amd64p32.s

@@ -620,12 +620,12 @@ TEXT runtime·memclr(SB),NOSPLIT,$0-8
 	MOVL	ptr+0(FP), DI
 	MOVL	n+4(FP), CX
 	MOVQ	CX, BX
-	ANDQ	$7, BX
-	SHRQ	$3, CX
+	ANDQ	$3, BX
+	SHRQ	$2, CX
 	MOVQ	$0, AX
 	CLD
 	REP
-	STOSQ
+	STOSL
 	MOVQ	BX, CX
 	REP
 	STOSB

src/runtime/memmove_nacl_amd64p32.s

@@ -4,6 +4,9 @@
 
 #include "textflag.h"
 
+// This could use MOVSQ, but we use MOVSL so that if an object ends in
+// a 4 byte pointer, we copy it as a unit instead of byte by byte.
+
 TEXT runtime·memmove(SB), NOSPLIT, $0-12
 	MOVL	to+0(FP), DI
 	MOVL	from+4(FP), SI
@@ -14,9 +17,9 @@ TEXT runtime·memmove(SB), NOSPLIT, $0-12
 
 forward:
 	MOVL	BX, CX
-	SHRL	$3, CX
-	ANDL	$7, BX
-	REP; MOVSQ
+	SHRL	$2, CX
+	ANDL	$3, BX
+	REP; MOVSL
 	MOVL	BX, CX
 	REP; MOVSB
 	RET
@@ -32,13 +35,13 @@ back:
 	STD
 
 	MOVL	BX, CX
-	SHRL	$3, CX
-	ANDL	$7, BX
-	SUBL	$8, DI
-	SUBL	$8, SI
-	REP; MOVSQ
-	ADDL	$7, DI
-	ADDL	$7, SI
+	SHRL	$2, CX
+	ANDL	$3, BX
+	SUBL	$4, DI
+	SUBL	$4, SI
+	REP; MOVSL
+	ADDL	$3, DI
+	ADDL	$3, SI
 	MOVL	BX, CX
 	REP; MOVSB
 	CLD