Add assembly CRYPTO_memcmp.

GH: #102

Reviewed-by: Richard Levitte <levitte@openssl.org>

crypto/alphacpuid.pl

@@ -134,6 +134,34 @@ OPENSSL_cleanse:
 	bne	$17,.Little
 .Ldone: ret	($26)
 .end	OPENSSL_cleanse
+
+.globl	CRYPTO_memcmp
+.ent	CRYPTO_memcmp
+CRYPTO_memcmp:
+	.frame	$30,0,$26
+	.prologue 0
+	xor	$0,$0,$0
+	beq	$18,.Lno_data
+
+	xor	$1,$1,$1
+	nop
+.Loop_cmp:
+	ldq_u	$2,0($16)
+	subq	$18,1,$18
+	ldq_u	$3,0($17)
+	extbl	$2,$16,$2
+	lda	$16,1($16)
+	extbl	$3,$17,$3
+	lda	$17,1($17)
+	xor	$3,$2,$2
+	or	$2,$0,$0
+	bne	$18,.Loop_cmp
+
+	subq	$31,$0,$0
+	srl	$0,63,$0
+.Lno_data:
+	ret	($26)
+.end	CRYPTO_memcmp
 ___
 {
 my ($out,$cnt,$max)=("\$16","\$17","\$18");

crypto/arm64cpuid.pl

@@ -100,6 +100,26 @@ OPENSSL_cleanse:
 	cbnz	x1,.Little	// len!=0?
 	ret
 .size	OPENSSL_cleanse,.-OPENSSL_cleanse
+
+.globl	CRYPTO_memcmp
+.type	CRYPTO_memcmp,%function
+.align	4
+CRYPTO_memcmp:
+	eor	w3,w3,w3
+	cbz	x2,.Lno_data	// len==0?
+.Loop_cmp:
+	ldrb	w4,[x0],#1
+	ldrb	w5,[x1],#1
+	eor	w4,w4,w5
+	orr	w3,w3,w4
+	subs	x2,x2,#1
+	b.ne	.Loop_cmp
+
+.Lno_data:
+	neg	w0,w3
+	lsr	w0,w0,#31
+	ret
+.size	CRYPTO_memcmp,.-CRYPTO_memcmp
 ___
 
 print $code;

crypto/armv4cpuid.pl

@@ -105,6 +105,36 @@ OPENSSL_cleanse:
 #endif
 .size	OPENSSL_cleanse,.-OPENSSL_cleanse
 
+.global	CRYPTO_memcmp
+.type	CRYPTO_memcmp,%function
+.align	4
+CRYPTO_memcmp:
+	eor	ip,ip,ip
+	cmp	r2,#0
+	beq	.Lno_data
+	stmdb	sp!,{r4,r5}
+
+.Loop_cmp:
+	ldrb	r4,[r0],#1
+	ldrb	r5,[r1],#1
+	eor	r4,r4,r5
+	orr	ip,ip,r4
+	subs	r2,r2,#1
+	bne	.Loop_cmp
+
+	ldmia	sp!,{r4,r5}
+.Lno_data:
+	neg	r0,ip
+	mov	r0,r0,lsr#31
+#if __ARM_ARCH__>=5
+	bx	lr
+#else
+	tst	lr,#1
+	moveq	pc,lr
+	.word	0xe12fff1e	@ bx	lr
+#endif
+.size	CRYPTO_memcmp,.-CRYPTO_memcmp
+
 #if __ARM_MAX_ARCH__>=7
 .arch	armv7-a
 .fpu	neon

crypto/c64xpluscpuid.pl

@@ -18,6 +18,7 @@ $code.=<<___;
 	.if	__TI_EABI__
 	.asg	OPENSSL_rdtsc,_OPENSSL_rdtsc
 	.asg	OPENSSL_cleanse,_OPENSSL_cleanse
+	.asg	CRYPTO_memcmp,_CRYPTO_memcmp
 	.asg	OPENSSL_atomic_add,_OPENSSL_atomic_add
 	.asg	OPENSSL_wipe_cpu,_OPENSSL_wipe_cpu
 	.asg	OPENSSL_instrument_bus,_OPENSSL_instrument_bus
@@ -87,6 +88,29 @@ _OPENSSL_cleanse:
    [A1]	STB	A2,*A4++[2]
 	.endasmfunc
 
+	.global	_CRYPTO_memcmp
+_CRYPTO_memcmp:
+	.asmfunc
+	MV	A6,B0
+  [!B0]	BNOP	RA
+||[!B0]	ZERO	A4
+   [B0]	MVC	B0,ILC
+|| [B0]	ZERO	A0
+	NOP	4
+
+	SPLOOP	1
+	LDBU	*A4++,A1
+||	LDBU	*B4++,B1
+	NOP	4
+	XOR.L	B1,A1,A2
+	SPKERNEL 1,0
+||	OR.S	A2,A0,A0
+
+	BNOP	RA,3
+	ZERO.L	A4
+  [A0]	MVK	1,A4
+	.endasmfunc
+
 	.global	_OPENSSL_atomic_add
 _OPENSSL_atomic_add:
 	.asmfunc

crypto/cryptlib.c

@@ -316,6 +316,7 @@ void OPENSSL_die(const char *message, const char *file, int line)
 #endif
 }
 
+#if !defined(OPENSSL_CPUID_OBJ)
 /* volatile unsigned char* pointers are there because
  * 1. Accessing a variable declared volatile via a pointer
  *    that lacks a volatile qualifier causes undefined behavior.
@@ -347,3 +348,4 @@ int CRYPTO_memcmp(const volatile void * volatile in_a,
 
     return x;
 }
+#endif

crypto/ia64cpuid.S

@@ -2,6 +2,12 @@
 // On Win64i compile with ias.exe.
 .text
 
+#if defined(_HPUX_SOURCE) && !defined(_LP64)
+#define	ADDP	addp4
+#else
+#define	ADDP	add
+#endif
+
 .global	OPENSSL_cpuid_setup#
 .proc	OPENSSL_cpuid_setup#
 OPENSSL_cpuid_setup:
@@ -131,9 +137,7 @@ OPENSSL_wipe_cpu:
 .proc	OPENSSL_cleanse#
 OPENSSL_cleanse:
 { .mib;	cmp.eq		p6,p0=0,r33	    // len==0
-#if defined(_HPUX_SOURCE) && !defined(_LP64)
-	addp4		r32=0,r32
-#endif
+	ADDP		r32=0,r32
 (p6)	br.ret.spnt	b0		};;
 { .mib;	and		r2=7,r32
 	cmp.leu		p6,p0=15,r33	    // len>=15
@@ -166,14 +170,51 @@ OPENSSL_cleanse:
 (p6)	br.ret.sptk.many	b0	};;
 .endp	OPENSSL_cleanse#
 
+.global	CRYPTO_memcmp#
+.proc	CRYPTO_memcmp#
+.align	32
+.skip	16
+CRYPTO_memcmp:
+	.prologue
+{ .mib;	mov		r8=0
+	cmp.eq		p6,p0=0,r34	    // len==0?
+(p6)	br.ret.spnt	b0		};;
+	.save		ar.pfs,r2
+{ .mib;	alloc		r2=ar.pfs,3,5,0,8
+	.save		ar.lc,r3
+	mov		r3=ar.lc
+	brp.loop.imp	.Loop_cmp_ctop,.Loop_cmp_cend-16
+					}
+{ .mib;	sub		r10=r34,r0,1
+	.save		pr,r9
+	mov		r9=pr		};;
+{ .mii;	ADDP		r16=0,r32
+	mov		ar.lc=r10
+	mov		ar.ec=4		}
+{ .mib;	ADDP		r17=0,r33
+	mov		pr.rot=1<<16	};;
+
+.Loop_cmp_ctop:
+{ .mib;	(p16)	ld1	r32=[r16],1
+	(p18)	xor	r34=r34,r38	}
+{ .mib;	(p16)	ld1	r36=[r17],1
+	(p19)	or	r8=r8,r35
+	br.ctop.sptk	.Loop_cmp_ctop	};;
+.Loop_cmp_cend:
+
+{ .mib;	cmp.ne		p6,p0=0,r8
+	mov		ar.lc=r3	};;
+{ .mib;
+(p6)	mov		r8=1
+	mov		pr=r9,0x1ffff
+	br.ret.sptk.many	b0	};;
+.endp	CRYPTO_memcmp#
+
 .global	OPENSSL_instrument_bus#
 .proc	OPENSSL_instrument_bus#
 OPENSSL_instrument_bus:
 { .mmi;	mov		r2=r33
-#if defined(_HPUX_SOURCE) && !defined(_LP64)
-	addp4		r32=0,r32
-#endif
-					}
+	ADDP		r32=0,r32	}
 { .mmi;	mov		r8=ar.itc;;
 	mov		r10=r0
 	mov		r9=r8		};;
@@ -208,10 +249,7 @@ OPENSSL_instrument_bus:
 .proc	OPENSSL_instrument_bus2#
 OPENSSL_instrument_bus2:
 { .mmi;	mov		r2=r33			// put aside cnt
-#if defined(_HPUX_SOURCE) && !defined(_LP64)
-	addp4		r32=0,r32
-#endif
-					}
+	ADDP		r32=0,r32	}
 { .mmi;	mov		r8=ar.itc;;
 	mov		r10=r0
 	mov		r9=r8		};;

crypto/pariscid.pl

@@ -138,6 +138,37 @@ L\$done
 ___
 }
 {
+my ($in1,$in2,$len)=("%r26","%r25","%r24");
+
+$code.=<<___;
+	.EXPORT	CRYPTO_memcmp,ENTRY,ARGW0=GR,ARGW1=GR,ARGW1=GR
+	.ALIGN	8
+CRYPTO_memcmp
+	.PROC
+	.CALLINFO	NO_CALLS
+	.ENTRY
+	cmpib,*=	0,$len,L\$no_data
+	xor		$rv,$rv,$rv
+
+L\$oop_cmp
+	ldb		0($in1),%r19
+	ldb		0($in2),%r20
+	ldo		1($in1),$in1
+	ldo		1($in2),$in2
+	xor		%r19,%r20,%r29
+	addib,*<>	-1,$len,L\$oop_cmp
+	or		%r29,$rv,$rv
+
+	sub		%r0,$rv,%r29
+	extru		%r29,31,1,$rv
+L\$no_data
+	bv		($rp)
+	.EXIT
+	nop
+	.PROCEND
+___
+}
+{
 my ($out,$cnt,$max)=("%r26","%r25","%r24");
 my ($tick,$lasttick)=("%r23","%r22");
 my ($diff,$lastdiff)=("%r21","%r20");

crypto/ppccpuid.pl

@@ -177,6 +177,32 @@ Laligned:
 	.byte	0,12,0x14,0,0,0,2,0
 	.long	0
 .size	.OPENSSL_cleanse,.-.OPENSSL_cleanse
+
+globl	.CRYPTO_memcmp
+.align	4
+.CRYPTO_memcmp:
+	$CMPLI	r5,0
+	li	r0,0
+	beq	Lno_data
+	mtctr	r5
+Loop_cmp:
+	lbz	r6,0(r3)
+	addi	r3,r3,1
+	lbz	r7,0(r4)
+	addi	r4,r4,1
+	xor	r6,r6,r7
+	or	r0,r0,r6
+	bdnz	Loop_cmp
+
+Lno_data:
+	li	r3,0
+	sub	r3,r3,r0
+	extrwi	r3,r3,1,0
+	blr
+	.long	0
+	.byte	0,12,0x14,0,0,0,3,0
+	.long	0
+.size	.CRYPTO_memcmp,.-.CRYPTO_memcmp
 ___
 {
 my ($out,$cnt,$max)=("r3","r4","r5");

crypto/s390xcpuid.S

@@ -125,6 +125,33 @@ OPENSSL_cleanse:
 	br	%r14
 .size	OPENSSL_cleanse,.-OPENSSL_cleanse
 
+.globl	CRYPTO_memcmp
+.type	CRYPTO_memcmp,@function
+.align	16
+CRYPTO_memcmp:
+#if !defined(__s390x__) && !defined(__s390x)
+	llgfr	%r4,%r4
+#endif
+	lghi	%r5,0
+	clgr	%r4,%r5
+	je	.Lno_data
+
+.Loop_cmp:
+	llc	%r0,0(%r2)
+	la	%r2,1(%r2)
+	llc	%r1,0(%r3)
+	la	%r3,1(%r3)
+	xr	%r1,%r0
+	or	%r5,%r1
+	brctg	%r4,.Loop_cmp
+
+	lnr	%r5,%r5
+	srl	%r5,31
+.Lno_data:
+	lgr	%r2,%r5
+	br	%r14
+.size	CRYPTO_memcmp,.-CRYPTO_memcmp
+
 .globl	OPENSSL_instrument_bus
 .type	OPENSSL_instrument_bus,@function
 .align	16

crypto/sparccpuid.S

@@ -440,6 +440,40 @@ OPENSSL_cleanse:
 .type	OPENSSL_cleanse,#function
 .size	OPENSSL_cleanse,.-OPENSSL_cleanse
 
+.global	CRYPTO_memcmp
+.align	16
+CRYPTO_memcmp:
+	cmp	%o2,0
+#ifdef ABI64
+	beq,pn	%xcc,.Lno_data
+#else
+	beq	.Lno_data
+#endif
+	xor	%g1,%g1,%g1
+	nop
+
+.Loop_cmp:
+	ldub	[%o0],%o3
+	add	%o0,1,%o0
+	ldub	[%o1],%o4
+	add	%o1,1,%o1
+	subcc	%o2,1,%o2
+	xor	%o3,%o4,%o4
+#ifdef ABI64
+	bnz	%xcc,.Loop_cmp
+#else
+	bnz	.Loop_cmp
+#endif
+	or	%o4,%g1,%g1
+
+	sub	%g0,%g1,%g1
+	srl	%g1,31,%g1
+.Lno_data:
+	retl
+	mov	%g1,%o0
+.type	CRYPTO_memcmp,#function
+.size	CRYPTO_memcmp,.-CRYPTO_memcmp
+
 .global	_sparcv9_vis1_instrument_bus
 .align	8
 _sparcv9_vis1_instrument_bus:

crypto/x86_64cpuid.pl

@@ -224,6 +224,28 @@ OPENSSL_cleanse:
 	jne	.Little
 	ret
 .size	OPENSSL_cleanse,.-OPENSSL_cleanse
+
+.globl  CRYPTO_memcmp
+.type   CRYPTO_memcmp,\@abi-omnipotent
+.align  16
+CRYPTO_memcmp:
+	xor	%rax,%rax
+	xor	%r10,%r10
+	cmp	\$0,$arg3
+	je	.Lno_data
+.Loop_cmp:
+	mov	($arg1),%r10b
+	lea	1($arg1),$arg1
+	xor	($arg2),%r10b
+	lea	1($arg2),$arg2
+	or	%r10b,%al
+	dec	$arg3
+	jnz	.Loop_cmp
+	neg	%rax
+	shr	\$63,%rax
+.Lno_data:
+	ret
+.size	CRYPTO_memcmp,.-CRYPTO_memcmp
 ___
 
 print<<___ if (!$win64);

crypto/x86cpuid.pl

@@ -365,6 +365,31 @@ for (@ARGV) { $sse2=1 if (/-DOPENSSL_IA32_SSE2/); }
 	&ret	();
 &function_end_B("OPENSSL_cleanse");
 
+&function_begin_B("CRYPTO_memcmp");
+	&push	("esi");
+	&push	("edi");
+	&mov	("esi",&wparam(0));
+	&mov	("edi",&wparam(1));
+	&mov	("ecx",&wparam(2));
+	&xor	("eax","eax");
+	&xor	("edx","edx");
+	&cmp	("ecx",0);
+	&je	(&label("no_data"));
+&set_label("loop");
+	&mov	("dl",&BP(0,"esi"));
+	&lea	("esi",&DWP(1,"esi"));
+	&xor	("dl",&BP(0,"edi"));
+	&lea	("edi",&DWP(1,"edi"));
+	&or	("al","dl");
+	&dec	("ecx");
+	&jnz	(&label("loop"));
+	&neg	("eax");
+	&shr	("eax",31);
+&set_label("no_data");
+	&pop	("edi");
+	&pop	("esi");
+	&ret	();
+&function_end_B("CRYPTO_memcmp");
 {
 my $lasttick = "esi";
 my $lastdiff = "ebx";