We've been lax about matching END() macros in asm code for some time. This

is used to set the ELF size attribute for functions.  It isn't normally
critical but some things can make use of it (gdb for stack traces).
Valgrind needs it so I'm adding it in.  The problem is present on all
branches and on both i386 and amd64.

lib/libc/amd64/SYS.h

@@ -43,13 +43,15 @@
 			.weak CNAME(__CONCAT(_,x));			\
 			.set CNAME(__CONCAT(_,x)),CNAME(__CONCAT(__sys_,x)); \
 			mov __CONCAT($SYS_,x),%rax; KERNCALL; jb 2f; ret; \
-			2: movq PIC_GOT(HIDENAME(cerror)),%rcx; jmp *%rcx
+			2: movq PIC_GOT(HIDENAME(cerror)),%rcx; jmp *%rcx; \
+			END(__CONCAT(__sys_,x))
 
 #define	PSEUDO(x)	ENTRY(__CONCAT(__sys_,x));			\
 			.weak CNAME(__CONCAT(_,x));			\
 			.set CNAME(__CONCAT(_,x)),CNAME(__CONCAT(__sys_,x)); \
 			mov __CONCAT($SYS_,x),%rax; KERNCALL; jb 2f; ret ; \
-			2: movq PIC_GOT(HIDENAME(cerror)),%rcx; jmp *%rcx
+			2: movq PIC_GOT(HIDENAME(cerror)),%rcx; jmp *%rcx; \
+			END(__CONCAT(__sys_,x))
 #else
 #define	RSYSCALL(x)	ENTRY(__CONCAT(__sys_,x));			\
 			.weak CNAME(x);					\
@@ -57,13 +59,15 @@
 			.weak CNAME(__CONCAT(_,x));			\
 			.set CNAME(__CONCAT(_,x)),CNAME(__CONCAT(__sys_,x)); \
 			mov __CONCAT($SYS_,x),%rax; KERNCALL; jb 2f; ret; \
-			2: jmp HIDENAME(cerror)
+			2: jmp HIDENAME(cerror);			\
+			END(__CONCAT(__sys_,x))
 
 #define	PSEUDO(x)	ENTRY(__CONCAT(__sys_,x));			\
 			.weak CNAME(__CONCAT(_,x));			\
 			.set CNAME(__CONCAT(_,x)),CNAME(__CONCAT(__sys_,x)); \
 			mov __CONCAT($SYS_,x),%rax; KERNCALL; jb 2f; ret; \
-			2: jmp HIDENAME(cerror)
+			2: jmp HIDENAME(cerror);			\
+			END(__CONCAT(__sys_,x))
 #endif
 
 #define KERNCALL	movq %rcx, %r10; syscall

lib/libc/amd64/gen/_setjmp.S

@@ -61,6 +61,7 @@ ENTRY(_setjmp)
 	stmxcsr	68(%rax)		/*    and mxcsr */
 	xorq	%rax,%rax
 	ret
+END(_setjmp)
 
 	.weak	CNAME(_longjmp)
 	.set	CNAME(_longjmp),CNAME(___longjmp)
@@ -90,3 +91,4 @@ ENTRY(___longjmp)
 	incq	%rax
 1:	movq	%rcx,0(%rsp)
 	ret
+END(___longjmp)

lib/libc/amd64/gen/fabs.S

@@ -37,6 +37,7 @@ ENTRY(fabs)
 	movsd	signbit(%rip), %xmm0
 	andnpd  %xmm1, %xmm0
 	ret
+END(fabs)
 
 	.data
 signbit:

lib/libc/amd64/gen/modf.S

@@ -86,3 +86,4 @@ ENTRY(modf)
 	movsd   -8(%rsp),%xmm0
 
 	ret
+END(modf)

lib/libc/amd64/gen/rfork_thread.S

@@ -99,3 +99,4 @@ ENTRY(rfork_thread)
 #else
 	jmp	HIDENAME(cerror)
 #endif
+END(rfork_thread)

lib/libc/amd64/gen/setjmp.S

@@ -70,6 +70,7 @@ ENTRY(setjmp)
 	stmxcsr	68(%rcx)		/*    and mxcsr */
 	xorq	%rax,%rax
 	ret
+END(setjmp)
 
 	.weak CNAME(longjmp)
 	.set CNAME(longjmp),CNAME(__longjmp)
@@ -108,3 +109,4 @@ ENTRY(__longjmp)
 	incq	%rax
 1:	movq	%rcx,0(%rsp)
 	ret
+END(__longjmp)

lib/libc/amd64/gen/sigsetjmp.S

@@ -77,6 +77,7 @@ ENTRY(sigsetjmp)
 	fnstcw	64(%rcx)		/* 8; fpu cw */
 	xorq	%rax,%rax
 	ret
+END(sigsetjmp)
 
 	.weak CNAME(siglongjmp)
 	.set CNAME(siglongjmp),CNAME(__siglongjmp)
@@ -109,3 +110,4 @@ ENTRY(__siglongjmp)
 	incq	%rax
 1:	movq	%rcx,0(%rsp)
 	ret
+END(__siglongjmp)

lib/libc/amd64/stdlib/div.S

@@ -15,3 +15,4 @@ ENTRY(div)
 	salq	$32,%rdx
 	orq	%rdx,%rax
 	ret
+END(div)

lib/libc/amd64/stdlib/ldiv.S

@@ -13,3 +13,4 @@ ENTRY(ldiv)
 	cqto
 	idivq	%rsi
 	ret
+END(ldiv)

lib/libc/amd64/stdlib/lldiv.S

@@ -13,3 +13,4 @@ ENTRY(lldiv)
 	cqto
 	idivq	%rsi
 	ret
+END(lldiv)

lib/libc/amd64/string/bcmp.S

@@ -22,3 +22,4 @@ L1:
 	setne	%al
 	movsbl	%al,%eax
 	ret
+END(bcmp)

lib/libc/amd64/string/bcopy.S

@@ -86,3 +86,12 @@ ENTRY(bcopy)
 	movsq
 	cld
 	ret
+#ifdef MEMCOPY
+END(memcpy)
+#else
+#ifdef MEMMOVE
+END(memmove)
+#else
+END(bcopy)
+#endif
+#endif

lib/libc/amd64/string/bzero.S

@@ -41,3 +41,4 @@ L1:	movq	%rsi,%rcx		/* zero remainder by bytes */
 	stosb
 
 	ret
+END(bzero)

lib/libc/amd64/string/memcmp.S

@@ -39,3 +39,4 @@ L6:	xorl	%eax,%eax		/* Perform unsigned comparison	*/
 	movb	-1(%rsi),%dl
 	subl    %edx,%eax
 	ret
+END(memcmp)

lib/libc/amd64/string/memset.S

@@ -58,3 +58,4 @@ L1:	rep
 	movq	%r11,%rax
 
 	ret
+END(memset)

lib/libc/amd64/string/strcat.S

@@ -163,3 +163,4 @@ ENTRY(strcat)
 
 .Ldone:
 	ret
+END(strcat)

lib/libc/amd64/string/strcmp.S

@@ -71,3 +71,4 @@ ENTRY(strcmp)
 	movzbq	%dl,%rdx
 	subq	%rdx,%rax
 	ret
+END(strcmp)

lib/libc/amd64/string/strcpy.S

@@ -109,3 +109,4 @@ ENTRY(strcpy)
 
 .Ldone:
 	ret
+END(strcpy)

lib/libc/amd64/sys/brk.S

@@ -43,6 +43,7 @@ __FBSDID("$FreeBSD$");
 ENTRY(_brk)
 	pushq	%rdi
 	jmp	ok
+END(_brk)
 
 ENTRY(brk)
 	pushq	%rdi
@@ -81,3 +82,4 @@ err:
 #else
 	jmp	HIDENAME(cerror)
 #endif
+END(brk)

lib/libc/amd64/sys/exect.S

@@ -53,3 +53,4 @@ ENTRY(exect)
 #else
 	jmp	HIDENAME(cerror)
 #endif
+END(exect)

lib/libc/amd64/sys/getcontext.S

@@ -52,3 +52,4 @@ ENTRY(__sys_getcontext)
 #else
 	jmp	HIDENAME(cerror)
 #endif
+END(__sys_getcontext)

lib/libc/amd64/sys/pipe.S

@@ -57,3 +57,4 @@ ENTRY(__sys_pipe)
 #else
 	jmp	HIDENAME(cerror)
 #endif
+END(__sys_pipe)

lib/libc/amd64/sys/ptrace.S

@@ -57,3 +57,4 @@ err:
 #else
 	jmp	HIDENAME(cerror)
 #endif
+END(ptrace)

lib/libc/amd64/sys/reboot.S

@@ -54,3 +54,4 @@ ENTRY(__sys_reboot)
 #else
 	jmp	HIDENAME(cerror)
 #endif
+END(__sys_reboot)

lib/libc/amd64/sys/sbrk.S

@@ -85,3 +85,4 @@ err:
 #else
 	jmp	HIDENAME(cerror)
 #endif
+END(sbrk)

lib/libc/amd64/sys/setlogin.S

@@ -62,3 +62,4 @@ ENTRY(__sys_setlogin)
 #else
 	jmp	HIDENAME(cerror)
 #endif
+END(__sys_setlogin)

lib/libc/amd64/sys/vfork.S

@@ -56,3 +56,4 @@ ENTRY(__sys_vfork)
 #else
 	jmp	HIDENAME(cerror)
 #endif
+END(__sys_vfork)