Fix regressions in arch_cpu_user_strlcpy().

- repnz movsb turns out to not actually be a legal instruction,
  resulting in various strings being copied incorrectly, leading to
  random crashes in various places. Rework to use loop instead.
  Thanks to Alex Smith for helping review changes and offering
  improvements.
- Minor cleanups.
- Fixes #8650 properly.

src/system/kernel/arch/x86/arch_x86.S

@@ -276,19 +276,32 @@ FUNCTION(arch_cpu_user_strlcpy):
 	/* Copy at most count - 1 bytes */
 	dec		%ecx
 
-	/* move data by bytes */
+	/* If count is now 0, skip straight to null terminating
+	   as our loop will otherwise overflow */
+	cmp		$0,%ecx
+	jne		.L_user_strlcpy_copy_begin
+	movb		$0,(%edi)
+	jmp		.L_user_strlcpy_source_count
+
+.L_user_strlcpy_copy_begin:
 	cld
-	repnz
+.L_user_strlcpy_copy_loop:
+	/* move data by bytes */
 	movsb
+	cmpb $0,-1(%esi)
+	je .L_user_strlcpy_copy_loop_done
+	loop .L_user_strlcpy_copy_loop
 
-	/* null terminate string */
-	movb	$0,(%edi)
-	dec		%esi
-
+.L_user_strlcpy_copy_loop_done:
 	/* check if we copied the entire source string */
 	cmp		$0,%ecx
 	jne		.L_user_strlcpy_source_done
 
+.L_user_strlcpy_zero_terminate:
+	/* null terminate string */
+	movb	$0,(%edi)
+	dec	%esi
+
 	/* count remaining bytes in src */
 .L_user_strlcpy_source_count:
 	not		%ecx
@@ -297,11 +310,9 @@ FUNCTION(arch_cpu_user_strlcpy):
 	scasb
 
 .L_user_strlcpy_source_done:
-
 	movl	%esi,%eax
 	subl	20(%esp),%eax
 	subl	$1,%eax
-
 	/* restore the old fault handler */
 	movl	%ebx,(%edx)