Kill more FIPS tentacles by removing the private_AES_set_{enc,dec}rypt_key()

internal interfaces, and promoting them to being the public
AES_set_{enc,dec}rypt_key() interfaces. In non-FIPS mode, these public
interfaces were directly calling the private ones.

ok guenther@ jsing@

4 files changed, 1306 insertions, 298 deletions

diff --git a/lib/libcrypto/aes/asm/aes-586.pl b/lib/libcrypto/aes/asm/aes-586.pl
index 687ed811be4..aab40e6f1cf 100644
--- a/lib/libcrypto/aes/asm/aes-586.pl
+++ b/lib/libcrypto/aes/asm/aes-586.pl
@@ -39,7 +39,7 @@
 # but exhibits up to 10% improvement on other cores.
 #
 # Second version is "monolithic" replacement for aes_core.c, which in
-# addition to AES_[de|en]crypt implements private_AES_set_[de|en]cryption_key.
+# addition to AES_[de|en]crypt implements AES_set_[de|en]cryption_key.
 # This made it possible to implement little-endian variant of the
 # algorithm without modifying the base C code. Motivating factor for
 # the undertaken effort was that it appeared that in tight IA-32
@@ -2854,12 +2854,12 @@ sub enckey()
     &set_label("exit");
 &function_end("_x86_AES_set_encrypt_key");
 
-# int private_AES_set_encrypt_key(const unsigned char *userKey, const int bits,
+# int AES_set_encrypt_key(const unsigned char *userKey, const int bits,
 #                        AES_KEY *key)
-&function_begin_B("private_AES_set_encrypt_key");
+&function_begin_B("AES_set_encrypt_key");
 	&call	("_x86_AES_set_encrypt_key");
 	&ret	();
-&function_end_B("private_AES_set_encrypt_key");
+&function_end_B("AES_set_encrypt_key");
 
 sub deckey()
 { my ($i,$key,$tp1,$tp2,$tp4,$tp8) = @_;
@@ -2916,9 +2916,9 @@ sub deckey()
 	&mov	(&DWP(4*$i,$key),$tp1);
 }
 
-# int private_AES_set_decrypt_key(const unsigned char *userKey, const int bits,
+# int AES_set_decrypt_key(const unsigned char *userKey, const int bits,
 #                        AES_KEY *key)
-&function_begin_B("private_AES_set_decrypt_key");
+&function_begin_B("AES_set_decrypt_key");
 	&call	("_x86_AES_set_encrypt_key");
 	&cmp	("eax",0);
 	&je	(&label("proceed"));
@@ -2974,7 +2974,7 @@ sub deckey()
 	&jb	(&label("permute"));
 
 	&xor	("eax","eax");			# return success
-&function_end("private_AES_set_decrypt_key");
+&function_end("AES_set_decrypt_key");
 &asciz("AES for x86, CRYPTOGAMS by <appro\@openssl.org>");
 
 &asm_finish();
diff --git a/lib/libcrypto/aes/asm/aes-armv4.pl b/lib/libcrypto/aes/asm/aes-armv4.pl
index 15742c1ec54..717cc1ed7f0 100644
--- a/lib/libcrypto/aes/asm/aes-armv4.pl
+++ b/lib/libcrypto/aes/asm/aes-armv4.pl
@@ -16,12 +16,25 @@
 # allows to merge logical or arithmetic operation with shift or rotate
 # in one instruction and emit combined result every cycle. The module
 # is endian-neutral. The performance is ~42 cycles/byte for 128-bit
-# key.
+# key [on single-issue Xscale PXA250 core].
 
 # May 2007.
 #
 # AES_set_[en|de]crypt_key is added.
 
+# July 2010.
+#
+# Rescheduling for dual-issue pipeline resulted in 12% improvement on
+# Cortex A8 core and ~25 cycles per byte processed with 128-bit key.
+
+# February 2011.
+#
+# Profiler-assisted and platform-specific optimization resulted in 16%
+# improvement on Cortex A8 core and ~21.5 cycles per byte.
+
+while (($output=shift) && ($output!~/^\w[\w\-]*\.\w+$/)) {}
+open STDOUT,">$output";
+
 $s0="r0";
 $s1="r1";
 $s2="r2";
@@ -38,6 +51,7 @@ $key="r11";
 $rounds="r12";
 
 $code=<<___;
+#include "arm_arch.h"
 .text
 .code	32
 
@@ -158,219 +172,243 @@ AES_encrypt:
 	mov	$rounds,r0		@ inp
 	mov	$key,r2
 	sub	$tbl,r3,#AES_encrypt-AES_Te	@ Te
-
+#if __ARM_ARCH__<7
 	ldrb	$s0,[$rounds,#3]	@ load input data in endian-neutral
 	ldrb	$t1,[$rounds,#2]	@ manner...
 	ldrb	$t2,[$rounds,#1]
 	ldrb	$t3,[$rounds,#0]
 	orr	$s0,$s0,$t1,lsl#8
-	orr	$s0,$s0,$t2,lsl#16
-	orr	$s0,$s0,$t3,lsl#24
 	ldrb	$s1,[$rounds,#7]
+	orr	$s0,$s0,$t2,lsl#16
 	ldrb	$t1,[$rounds,#6]
+	orr	$s0,$s0,$t3,lsl#24
 	ldrb	$t2,[$rounds,#5]
 	ldrb	$t3,[$rounds,#4]
 	orr	$s1,$s1,$t1,lsl#8
-	orr	$s1,$s1,$t2,lsl#16
-	orr	$s1,$s1,$t3,lsl#24
 	ldrb	$s2,[$rounds,#11]
+	orr	$s1,$s1,$t2,lsl#16
 	ldrb	$t1,[$rounds,#10]
+	orr	$s1,$s1,$t3,lsl#24
 	ldrb	$t2,[$rounds,#9]
 	ldrb	$t3,[$rounds,#8]
 	orr	$s2,$s2,$t1,lsl#8
-	orr	$s2,$s2,$t2,lsl#16
-	orr	$s2,$s2,$t3,lsl#24
 	ldrb	$s3,[$rounds,#15]
+	orr	$s2,$s2,$t2,lsl#16
 	ldrb	$t1,[$rounds,#14]
+	orr	$s2,$s2,$t3,lsl#24
 	ldrb	$t2,[$rounds,#13]
 	ldrb	$t3,[$rounds,#12]
 	orr	$s3,$s3,$t1,lsl#8
 	orr	$s3,$s3,$t2,lsl#16
 	orr	$s3,$s3,$t3,lsl#24
-
+#else
+	ldr	$s0,[$rounds,#0]
+	ldr	$s1,[$rounds,#4]
+	ldr	$s2,[$rounds,#8]
+	ldr	$s3,[$rounds,#12]
+#ifdef __ARMEL__
+	rev	$s0,$s0
+	rev	$s1,$s1
+	rev	$s2,$s2
+	rev	$s3,$s3
+#endif
+#endif
 	bl	_armv4_AES_encrypt
 
 	ldr	$rounds,[sp],#4		@ pop out
+#if __ARM_ARCH__>=7
+#ifdef __ARMEL__
+	rev	$s0,$s0
+	rev	$s1,$s1
+	rev	$s2,$s2
+	rev	$s3,$s3
+#endif
+	str	$s0,[$rounds,#0]
+	str	$s1,[$rounds,#4]
+	str	$s2,[$rounds,#8]
+	str	$s3,[$rounds,#12]
+#else
 	mov	$t1,$s0,lsr#24		@ write output in endian-neutral
 	mov	$t2,$s0,lsr#16		@ manner...
 	mov	$t3,$s0,lsr#8
 	strb	$t1,[$rounds,#0]
 	strb	$t2,[$rounds,#1]
-	strb	$t3,[$rounds,#2]
-	strb	$s0,[$rounds,#3]
 	mov	$t1,$s1,lsr#24
+	strb	$t3,[$rounds,#2]
 	mov	$t2,$s1,lsr#16
+	strb	$s0,[$rounds,#3]
 	mov	$t3,$s1,lsr#8
 	strb	$t1,[$rounds,#4]
 	strb	$t2,[$rounds,#5]
-	strb	$t3,[$rounds,#6]
-	strb	$s1,[$rounds,#7]
 	mov	$t1,$s2,lsr#24
+	strb	$t3,[$rounds,#6]
 	mov	$t2,$s2,lsr#16
+	strb	$s1,[$rounds,#7]
 	mov	$t3,$s2,lsr#8
 	strb	$t1,[$rounds,#8]
 	strb	$t2,[$rounds,#9]
-	strb	$t3,[$rounds,#10]
-	strb	$s2,[$rounds,#11]
 	mov	$t1,$s3,lsr#24
+	strb	$t3,[$rounds,#10]
 	mov	$t2,$s3,lsr#16
+	strb	$s2,[$rounds,#11]
 	mov	$t3,$s3,lsr#8
 	strb	$t1,[$rounds,#12]
 	strb	$t2,[$rounds,#13]
 	strb	$t3,[$rounds,#14]
 	strb	$s3,[$rounds,#15]
-
+#endif
+#if __ARM_ARCH__>=5
+	ldmia	sp!,{r4-r12,pc}
+#else
 	ldmia   sp!,{r4-r12,lr}
 	tst	lr,#1
 	moveq	pc,lr			@ be binary compatible with V4, yet
 	bx	lr			@ interoperable with Thumb ISA:-)
+#endif
 .size	AES_encrypt,.-AES_encrypt
 
 .type   _armv4_AES_encrypt,%function
 .align	2
 _armv4_AES_encrypt:
 	str	lr,[sp,#-4]!		@ push lr
-	ldr	$t1,[$key],#16
-	ldr	$t2,[$key,#-12]
-	ldr	$t3,[$key,#-8]
-	ldr	$i1,[$key,#-4]
-	ldr	$rounds,[$key,#240-16]
+	ldmia	$key!,{$t1-$i1}
 	eor	$s0,$s0,$t1
+	ldr	$rounds,[$key,#240-16]
 	eor	$s1,$s1,$t2
 	eor	$s2,$s2,$t3
 	eor	$s3,$s3,$i1
 	sub	$rounds,$rounds,#1
 	mov	lr,#255
 
-.Lenc_loop:
+	and	$i1,lr,$s0
 	and	$i2,lr,$s0,lsr#8
 	and	$i3,lr,$s0,lsr#16
-	and	$i1,lr,$s0
 	mov	$s0,$s0,lsr#24
+.Lenc_loop:
 	ldr	$t1,[$tbl,$i1,lsl#2]	@ Te3[s0>>0]
-	ldr	$s0,[$tbl,$s0,lsl#2]	@ Te0[s0>>24]
-	ldr	$t2,[$tbl,$i2,lsl#2]	@ Te2[s0>>8]
-	ldr	$t3,[$tbl,$i3,lsl#2]	@ Te1[s0>>16]
-
 	and	$i1,lr,$s1,lsr#16	@ i0
+	ldr	$t2,[$tbl,$i2,lsl#2]	@ Te2[s0>>8]
 	and	$i2,lr,$s1
+	ldr	$t3,[$tbl,$i3,lsl#2]	@ Te1[s0>>16]
 	and	$i3,lr,$s1,lsr#8
+	ldr	$s0,[$tbl,$s0,lsl#2]	@ Te0[s0>>24]
 	mov	$s1,$s1,lsr#24
+
 	ldr	$i1,[$tbl,$i1,lsl#2]	@ Te1[s1>>16]
-	ldr	$s1,[$tbl,$s1,lsl#2]	@ Te0[s1>>24]
 	ldr	$i2,[$tbl,$i2,lsl#2]	@ Te3[s1>>0]
 	ldr	$i3,[$tbl,$i3,lsl#2]	@ Te2[s1>>8]
 	eor	$s0,$s0,$i1,ror#8
-	eor	$s1,$s1,$t1,ror#24
-	eor	$t2,$t2,$i2,ror#8
-	eor	$t3,$t3,$i3,ror#8
-
+	ldr	$s1,[$tbl,$s1,lsl#2]	@ Te0[s1>>24]
 	and	$i1,lr,$s2,lsr#8	@ i0
+	eor	$t2,$t2,$i2,ror#8
 	and	$i2,lr,$s2,lsr#16	@ i1
+	eor	$t3,$t3,$i3,ror#8
 	and	$i3,lr,$s2
-	mov	$s2,$s2,lsr#24
 	ldr	$i1,[$tbl,$i1,lsl#2]	@ Te2[s2>>8]
+	eor	$s1,$s1,$t1,ror#24
 	ldr	$i2,[$tbl,$i2,lsl#2]	@ Te1[s2>>16]
-	ldr	$s2,[$tbl,$s2,lsl#2]	@ Te0[s2>>24]
+	mov	$s2,$s2,lsr#24
+
 	ldr	$i3,[$tbl,$i3,lsl#2]	@ Te3[s2>>0]
 	eor	$s0,$s0,$i1,ror#16
-	eor	$s1,$s1,$i2,ror#8
-	eor	$s2,$s2,$t2,ror#16
-	eor	$t3,$t3,$i3,ror#16
-
+	ldr	$s2,[$tbl,$s2,lsl#2]	@ Te0[s2>>24]
 	and	$i1,lr,$s3		@ i0
+	eor	$s1,$s1,$i2,ror#8
 	and	$i2,lr,$s3,lsr#8	@ i1
+	eor	$t3,$t3,$i3,ror#16
 	and	$i3,lr,$s3,lsr#16	@ i2
-	mov	$s3,$s3,lsr#24
 	ldr	$i1,[$tbl,$i1,lsl#2]	@ Te3[s3>>0]
+	eor	$s2,$s2,$t2,ror#16
 	ldr	$i2,[$tbl,$i2,lsl#2]	@ Te2[s3>>8]
+	mov	$s3,$s3,lsr#24
+
 	ldr	$i3,[$tbl,$i3,lsl#2]	@ Te1[s3>>16]
-	ldr	$s3,[$tbl,$s3,lsl#2]	@ Te0[s3>>24]
 	eor	$s0,$s0,$i1,ror#24
+	ldr	$i1,[$key],#16
 	eor	$s1,$s1,$i2,ror#16
+	ldr	$s3,[$tbl,$s3,lsl#2]	@ Te0[s3>>24]
 	eor	$s2,$s2,$i3,ror#8
+	ldr	$t1,[$key,#-12]
 	eor	$s3,$s3,$t3,ror#8
 
-	ldr	$t1,[$key],#16
-	ldr	$t2,[$key,#-12]
-	ldr	$t3,[$key,#-8]
-	ldr	$i1,[$key,#-4]
-	eor	$s0,$s0,$t1
-	eor	$s1,$s1,$t2
-	eor	$s2,$s2,$t3
-	eor	$s3,$s3,$i1
+	ldr	$t2,[$key,#-8]
+	eor	$s0,$s0,$i1
+	ldr	$t3,[$key,#-4]
+	and	$i1,lr,$s0
+	eor	$s1,$s1,$t1
+	and	$i2,lr,$s0,lsr#8
+	eor	$s2,$s2,$t2
+	and	$i3,lr,$s0,lsr#16
+	eor	$s3,$s3,$t3
+	mov	$s0,$s0,lsr#24
 
 	subs	$rounds,$rounds,#1
 	bne	.Lenc_loop
 
 	add	$tbl,$tbl,#2
 
-	and	$i1,lr,$s0
-	and	$i2,lr,$s0,lsr#8
-	and	$i3,lr,$s0,lsr#16
-	mov	$s0,$s0,lsr#24
 	ldrb	$t1,[$tbl,$i1,lsl#2]	@ Te4[s0>>0]
-	ldrb	$s0,[$tbl,$s0,lsl#2]	@ Te4[s0>>24]
-	ldrb	$t2,[$tbl,$i2,lsl#2]	@ Te4[s0>>8]
-	ldrb	$t3,[$tbl,$i3,lsl#2]	@ Te4[s0>>16]
-
 	and	$i1,lr,$s1,lsr#16	@ i0
+	ldrb	$t2,[$tbl,$i2,lsl#2]	@ Te4[s0>>8]
 	and	$i2,lr,$s1
+	ldrb	$t3,[$tbl,$i3,lsl#2]	@ Te4[s0>>16]
 	and	$i3,lr,$s1,lsr#8
+	ldrb	$s0,[$tbl,$s0,lsl#2]	@ Te4[s0>>24]
 	mov	$s1,$s1,lsr#24
+
 	ldrb	$i1,[$tbl,$i1,lsl#2]	@ Te4[s1>>16]
-	ldrb	$s1,[$tbl,$s1,lsl#2]	@ Te4[s1>>24]
 	ldrb	$i2,[$tbl,$i2,lsl#2]	@ Te4[s1>>0]
 	ldrb	$i3,[$tbl,$i3,lsl#2]	@ Te4[s1>>8]
 	eor	$s0,$i1,$s0,lsl#8
-	eor	$s1,$t1,$s1,lsl#24
-	eor	$t2,$i2,$t2,lsl#8
-	eor	$t3,$i3,$t3,lsl#8
-
+	ldrb	$s1,[$tbl,$s1,lsl#2]	@ Te4[s1>>24]
 	and	$i1,lr,$s2,lsr#8	@ i0
+	eor	$t2,$i2,$t2,lsl#8
 	and	$i2,lr,$s2,lsr#16	@ i1
+	eor	$t3,$i3,$t3,lsl#8
 	and	$i3,lr,$s2
-	mov	$s2,$s2,lsr#24
 	ldrb	$i1,[$tbl,$i1,lsl#2]	@ Te4[s2>>8]
+	eor	$s1,$t1,$s1,lsl#24
 	ldrb	$i2,[$tbl,$i2,lsl#2]	@ Te4[s2>>16]
-	ldrb	$s2,[$tbl,$s2,lsl#2]	@ Te4[s2>>24]
+	mov	$s2,$s2,lsr#24
+
 	ldrb	$i3,[$tbl,$i3,lsl#2]	@ Te4[s2>>0]
 	eor	$s0,$i1,$s0,lsl#8
-	eor	$s1,$s1,$i2,lsl#16
-	eor	$s2,$t2,$s2,lsl#24
-	eor	$t3,$i3,$t3,lsl#8
-
+	ldrb	$s2,[$tbl,$s2,lsl#2]	@ Te4[s2>>24]
 	and	$i1,lr,$s3		@ i0
+	eor	$s1,$s1,$i2,lsl#16
 	and	$i2,lr,$s3,lsr#8	@ i1
+	eor	$t3,$i3,$t3,lsl#8
 	and	$i3,lr,$s3,lsr#16	@ i2
-	mov	$s3,$s3,lsr#24
 	ldrb	$i1,[$tbl,$i1,lsl#2]	@ Te4[s3>>0]
+	eor	$s2,$t2,$s2,lsl#24
 	ldrb	$i2,[$tbl,$i2,lsl#2]	@ Te4[s3>>8]
+	mov	$s3,$s3,lsr#24
+
 	ldrb	$i3,[$tbl,$i3,lsl#2]	@ Te4[s3>>16]
-	ldrb	$s3,[$tbl,$s3,lsl#2]	@ Te4[s3>>24]
 	eor	$s0,$i1,$s0,lsl#8
+	ldr	$i1,[$key,#0]
+	ldrb	$s3,[$tbl,$s3,lsl#2]	@ Te4[s3>>24]
 	eor	$s1,$s1,$i2,lsl#8
+	ldr	$t1,[$key,#4]
 	eor	$s2,$s2,$i3,lsl#16
+	ldr	$t2,[$key,#8]
 	eor	$s3,$t3,$s3,lsl#24
+	ldr	$t3,[$key,#12]
 
-	ldr	lr,[sp],#4		@ pop lr
-	ldr	$t1,[$key,#0]
-	ldr	$t2,[$key,#4]
-	ldr	$t3,[$key,#8]
-	ldr	$i1,[$key,#12]
-	eor	$s0,$s0,$t1
-	eor	$s1,$s1,$t2
-	eor	$s2,$s2,$t3
-	eor	$s3,$s3,$i1
+	eor	$s0,$s0,$i1
+	eor	$s1,$s1,$t1
+	eor	$s2,$s2,$t2
+	eor	$s3,$s3,$t3
 
 	sub	$tbl,$tbl,#2
-	mov	pc,lr			@ return
+	ldr	pc,[sp],#4		@ pop and return
 .size	_armv4_AES_encrypt,.-_armv4_AES_encrypt
 
 .global AES_set_encrypt_key
 .type   AES_set_encrypt_key,%function
 .align	5
 AES_set_encrypt_key:
+_armv4_AES_set_encrypt_key:
 	sub	r3,pc,#8		@ AES_set_encrypt_key
 	teq	r0,#0
 	moveq	r0,#-1
@@ -388,44 +426,61 @@ AES_set_encrypt_key:
 	bne	.Labrt
 
 .Lok:	stmdb   sp!,{r4-r12,lr}
-	sub	$tbl,r3,#AES_set_encrypt_key-AES_Te-1024	@ Te4
+	sub	$tbl,r3,#_armv4_AES_set_encrypt_key-AES_Te-1024	@ Te4
 
 	mov	$rounds,r0		@ inp
 	mov	lr,r1			@ bits
 	mov	$key,r2			@ key
 
+#if __ARM_ARCH__<7
 	ldrb	$s0,[$rounds,#3]	@ load input data in endian-neutral
 	ldrb	$t1,[$rounds,#2]	@ manner...
 	ldrb	$t2,[$rounds,#1]
 	ldrb	$t3,[$rounds,#0]
 	orr	$s0,$s0,$t1,lsl#8
-	orr	$s0,$s0,$t2,lsl#16
-	orr	$s0,$s0,$t3,lsl#24
 	ldrb	$s1,[$rounds,#7]
+	orr	$s0,$s0,$t2,lsl#16
 	ldrb	$t1,[$rounds,#6]
+	orr	$s0,$s0,$t3,lsl#24
 	ldrb	$t2,[$rounds,#5]
 	ldrb	$t3,[$rounds,#4]
 	orr	$s1,$s1,$t1,lsl#8
-	orr	$s1,$s1,$t2,lsl#16
-	orr	$s1,$s1,$t3,lsl#24
 	ldrb	$s2,[$rounds,#11]
+	orr	$s1,$s1,$t2,lsl#16
 	ldrb	$t1,[$rounds,#10]
+	orr	$s1,$s1,$t3,lsl#24
 	ldrb	$t2,[$rounds,#9]
 	ldrb	$t3,[$rounds,#8]
 	orr	$s2,$s2,$t1,lsl#8
-	orr	$s2,$s2,$t2,lsl#16
-	orr	$s2,$s2,$t3,lsl#24
 	ldrb	$s3,[$rounds,#15]
+	orr	$s2,$s2,$t2,lsl#16
 	ldrb	$t1,[$rounds,#14]
+	orr	$s2,$s2,$t3,lsl#24
 	ldrb	$t2,[$rounds,#13]
 	ldrb	$t3,[$rounds,#12]
 	orr	$s3,$s3,$t1,lsl#8
+	str	$s0,[$key],#16
 	orr	$s3,$s3,$t2,lsl#16
+	str	$s1,[$key,#-12]
 	orr	$s3,$s3,$t3,lsl#24
+	str	$s2,[$key,#-8]
+	str	$s3,[$key,#-4]
+#else
+	ldr	$s0,[$rounds,#0]
+	ldr	$s1,[$rounds,#4]
+	ldr	$s2,[$rounds,#8]
+	ldr	$s3,[$rounds,#12]
+#ifdef __ARMEL__
+	rev	$s0,$s0
+	rev	$s1,$s1
+	rev	$s2,$s2
+	rev	$s3,$s3
+#endif
 	str	$s0,[$key],#16
 	str	$s1,[$key,#-12]
 	str	$s2,[$key,#-8]
 	str	$s3,[$key,#-4]
+#endif
 
 	teq	lr,#128
 	bne	.Lnot128
@@ -437,48 +492,58 @@ AES_set_encrypt_key:
 .L128_loop:
 	and	$t2,lr,$s3,lsr#24
 	and	$i1,lr,$s3,lsr#16
-	and	$i2,lr,$s3,lsr#8
-	and	$i3,lr,$s3
 	ldrb	$t2,[$tbl,$t2]
+	and	$i2,lr,$s3,lsr#8
 	ldrb	$i1,[$tbl,$i1]
+	and	$i3,lr,$s3
 	ldrb	$i2,[$tbl,$i2]
-	ldrb	$i3,[$tbl,$i3]
-	ldr	$t1,[$t3],#4			@ rcon[i++]
 	orr	$t2,$t2,$i1,lsl#24
+	ldrb	$i3,[$tbl,$i3]
 	orr	$t2,$t2,$i2,lsl#16
+	ldr	$t1,[$t3],#4			@ rcon[i++]
 	orr	$t2,$t2,$i3,lsl#8
 	eor	$t2,$t2,$t1
 	eor	$s0,$s0,$t2			@ rk[4]=rk[0]^...
 	eor	$s1,$s1,$s0			@ rk[5]=rk[1]^rk[4]
-	eor	$s2,$s2,$s1			@ rk[6]=rk[2]^rk[5]
-	eor	$s3,$s3,$s2			@ rk[7]=rk[3]^rk[6]
 	str	$s0,[$key],#16
+	eor	$s2,$s2,$s1			@ rk[6]=rk[2]^rk[5]
 	str	$s1,[$key,#-12]
+	eor	$s3,$s3,$s2			@ rk[7]=rk[3]^rk[6]
 	str	$s2,[$key,#-8]
-	str	$s3,[$key,#-4]
-
 	subs	$rounds,$rounds,#1
+	str	$s3,[$key,#-4]
 	bne	.L128_loop
 	sub	r2,$key,#176
 	b	.Ldone
 
 .Lnot128:
+#if __ARM_ARCH__<7
 	ldrb	$i2,[$rounds,#19]
 	ldrb	$t1,[$rounds,#18]
 	ldrb	$t2,[$rounds,#17]
 	ldrb	$t3,[$rounds,#16]
 	orr	$i2,$i2,$t1,lsl#8
-	orr	$i2,$i2,$t2,lsl#16
-	orr	$i2,$i2,$t3,lsl#24
 	ldrb	$i3,[$rounds,#23]
+	orr	$i2,$i2,$t2,lsl#16
 	ldrb	$t1,[$rounds,#22]
+	orr	$i2,$i2,$t3,lsl#24
 	ldrb	$t2,[$rounds,#21]
 	ldrb	$t3,[$rounds,#20]
 	orr	$i3,$i3,$t1,lsl#8
 	orr	$i3,$i3,$t2,lsl#16
+	str	$i2,[$key],#8
 	orr	$i3,$i3,$t3,lsl#24
+	str	$i3,[$key,#-4]
+#else
+	ldr	$i2,[$rounds,#16]
+	ldr	$i3,[$rounds,#20]
+#ifdef __ARMEL__
+	rev	$i2,$i2
+	rev	$i3,$i3
+#endif
 	str	$i2,[$key],#8
 	str	$i3,[$key,#-4]
+#endif
 
 	teq	lr,#192
 	bne	.Lnot192
@@ -491,27 +556,26 @@ AES_set_encrypt_key:
 .L192_loop:
 	and	$t2,lr,$i3,lsr#24
 	and	$i1,lr,$i3,lsr#16
-	and	$i2,lr,$i3,lsr#8
-	and	$i3,lr,$i3
 	ldrb	$t2,[$tbl,$t2]
+	and	$i2,lr,$i3,lsr#8
 	ldrb	$i1,[$tbl,$i1]
+	and	$i3,lr,$i3
 	ldrb	$i2,[$tbl,$i2]
-	ldrb	$i3,[$tbl,$i3]
-	ldr	$t1,[$t3],#4			@ rcon[i++]
 	orr	$t2,$t2,$i1,lsl#24
+	ldrb	$i3,[$tbl,$i3]
 	orr	$t2,$t2,$i2,lsl#16
+	ldr	$t1,[$t3],#4			@ rcon[i++]
 	orr	$t2,$t2,$i3,lsl#8
 	eor	$i3,$t2,$t1
 	eor	$s0,$s0,$i3			@ rk[6]=rk[0]^...
 	eor	$s1,$s1,$s0			@ rk[7]=rk[1]^rk[6]
-	eor	$s2,$s2,$s1			@ rk[8]=rk[2]^rk[7]
-	eor	$s3,$s3,$s2			@ rk[9]=rk[3]^rk[8]
 	str	$s0,[$key],#24
+	eor	$s2,$s2,$s1			@ rk[8]=rk[2]^rk[7]
 	str	$s1,[$key,#-20]
+	eor	$s3,$s3,$s2			@ rk[9]=rk[3]^rk[8]
 	str	$s2,[$key,#-16]
-	str	$s3,[$key,#-12]
-
 	subs	$rounds,$rounds,#1
+	str	$s3,[$key,#-12]
 	subeq	r2,$key,#216
 	beq	.Ldone
 
@@ -524,22 +588,33 @@ AES_set_encrypt_key:
 	b	.L192_loop
 
 .Lnot192:
+#if __ARM_ARCH__<7
 	ldrb	$i2,[$rounds,#27]
 	ldrb	$t1,[$rounds,#26]
 	ldrb	$t2,[$rounds,#25]
 	ldrb	$t3,[$rounds,#24]
 	orr	$i2,$i2,$t1,lsl#8
-	orr	$i2,$i2,$t2,lsl#16
-	orr	$i2,$i2,$t3,lsl#24
 	ldrb	$i3,[$rounds,#31]
+	orr	$i2,$i2,$t2,lsl#16
 	ldrb	$t1,[$rounds,#30]
+	orr	$i2,$i2,$t3,lsl#24
 	ldrb	$t2,[$rounds,#29]
 	ldrb	$t3,[$rounds,#28]
 	orr	$i3,$i3,$t1,lsl#8
 	orr	$i3,$i3,$t2,lsl#16
+	str	$i2,[$key],#8
 	orr	$i3,$i3,$t3,lsl#24
+	str	$i3,[$key,#-4]
+#else
+	ldr	$i2,[$rounds,#24]
+	ldr	$i3,[$rounds,#28]
+#ifdef __ARMEL__
+	rev	$i2,$i2
+	rev	$i3,$i3
+#endif
 	str	$i2,[$key],#8
 	str	$i3,[$key,#-4]
+#endif
 
 	mov	$rounds,#14
 	str	$rounds,[$key,#240-32]
@@ -550,52 +625,51 @@ AES_set_encrypt_key:
 .L256_loop:
 	and	$t2,lr,$i3,lsr#24
 	and	$i1,lr,$i3,lsr#16
-	and	$i2,lr,$i3,lsr#8
-	and	$i3,lr,$i3
 	ldrb	$t2,[$tbl,$t2]
+	and	$i2,lr,$i3,lsr#8
 	ldrb	$i1,[$tbl,$i1]
+	and	$i3,lr,$i3
 	ldrb	$i2,[$tbl,$i2]
-	ldrb	$i3,[$tbl,$i3]
-	ldr	$t1,[$t3],#4			@ rcon[i++]
 	orr	$t2,$t2,$i1,lsl#24
+	ldrb	$i3,[$tbl,$i3]
 	orr	$t2,$t2,$i2,lsl#16
+	ldr	$t1,[$t3],#4			@ rcon[i++]
 	orr	$t2,$t2,$i3,lsl#8
 	eor	$i3,$t2,$t1
 	eor	$s0,$s0,$i3			@ rk[8]=rk[0]^...
 	eor	$s1,$s1,$s0			@ rk[9]=rk[1]^rk[8]
-	eor	$s2,$s2,$s1			@ rk[10]=rk[2]^rk[9]
-	eor	$s3,$s3,$s2			@ rk[11]=rk[3]^rk[10]
 	str	$s0,[$key],#32
+	eor	$s2,$s2,$s1			@ rk[10]=rk[2]^rk[9]
 	str	$s1,[$key,#-28]
+	eor	$s3,$s3,$s2			@ rk[11]=rk[3]^rk[10]
 	str	$s2,[$key,#-24]
-	str	$s3,[$key,#-20]
-
 	subs	$rounds,$rounds,#1
+	str	$s3,[$key,#-20]
 	subeq	r2,$key,#256
 	beq	.Ldone
 
 	and	$t2,lr,$s3
 	and	$i1,lr,$s3,lsr#8
-	and	$i2,lr,$s3,lsr#16
-	and	$i3,lr,$s3,lsr#24
 	ldrb	$t2,[$tbl,$t2]
+	and	$i2,lr,$s3,lsr#16
 	ldrb	$i1,[$tbl,$i1]
+	and	$i3,lr,$s3,lsr#24
 	ldrb	$i2,[$tbl,$i2]
-	ldrb	$i3,[$tbl,$i3]
 	orr	$t2,$t2,$i1,lsl#8
+	ldrb	$i3,[$tbl,$i3]
 	orr	$t2,$t2,$i2,lsl#16
+	ldr	$t1,[$key,#-48]
 	orr	$t2,$t2,$i3,lsl#24
 
-	ldr	$t1,[$key,#-48]
 	ldr	$i1,[$key,#-44]
 	ldr	$i2,[$key,#-40]
-	ldr	$i3,[$key,#-36]
 	eor	$t1,$t1,$t2			@ rk[12]=rk[4]^...
+	ldr	$i3,[$key,#-36]
 	eor	$i1,$i1,$t1			@ rk[13]=rk[5]^rk[12]
-	eor	$i2,$i2,$i1			@ rk[14]=rk[6]^rk[13]
-	eor	$i3,$i3,$i2			@ rk[15]=rk[7]^rk[14]
 	str	$t1,[$key,#-16]
+	eor	$i2,$i2,$i1			@ rk[14]=rk[6]^rk[13]
 	str	$i1,[$key,#-12]
+	eor	$i3,$i3,$i2			@ rk[15]=rk[7]^rk[14]
 	str	$i2,[$key,#-8]
 	str	$i3,[$key,#-4]
 	b	.L256_loop
@@ -612,7 +686,7 @@ AES_set_encrypt_key:
 .align	5
 AES_set_decrypt_key:
 	str	lr,[sp,#-4]!            @ push lr
-	bl	AES_set_encrypt_key
+	bl	_armv4_AES_set_encrypt_key
 	teq	r0,#0
 	ldrne	lr,[sp],#4              @ pop lr
 	bne	.Labrt
@@ -691,10 +765,14 @@ $code.=<<___;
 	bne	.Lmix
 
 	mov	r0,#0
+#if __ARM_ARCH__>=5
+	ldmia	sp!,{r4-r12,pc}
+#else
 	ldmia   sp!,{r4-r12,lr}
 	tst	lr,#1
 	moveq	pc,lr			@ be binary compatible with V4, yet
 	bx	lr			@ interoperable with Thumb ISA:-)
+#endif
 .size	AES_set_decrypt_key,.-AES_set_decrypt_key
 
 .type	AES_Td,%object
@@ -810,221 +888,247 @@ AES_decrypt:
 	mov	$rounds,r0		@ inp
 	mov	$key,r2
 	sub	$tbl,r3,#AES_decrypt-AES_Td		@ Td
-
+#if __ARM_ARCH__<7
 	ldrb	$s0,[$rounds,#3]	@ load input data in endian-neutral
 	ldrb	$t1,[$rounds,#2]	@ manner...
 	ldrb	$t2,[$rounds,#1]
 	ldrb	$t3,[$rounds,#0]
 	orr	$s0,$s0,$t1,lsl#8
-	orr	$s0,$s0,$t2,lsl#16
-	orr	$s0,$s0,$t3,lsl#24
 	ldrb	$s1,[$rounds,#7]
+	orr	$s0,$s0,$t2,lsl#16
 	ldrb	$t1,[$rounds,#6]
+	orr	$s0,$s0,$t3,lsl#24
 	ldrb	$t2,[$rounds,#5]
 	ldrb	$t3,[$rounds,#4]
 	orr	$s1,$s1,$t1,lsl#8
-	orr	$s1,$s1,$t2,lsl#16
-	orr	$s1,$s1,$t3,lsl#24
 	ldrb	$s2,[$rounds,#11]
+	orr	$s1,$s1,$t2,lsl#16
 	ldrb	$t1,[$rounds,#10]
+	orr	$s1,$s1,$t3,lsl#24
 	ldrb	$t2,[$rounds,#9]
 	ldrb	$t3,[$rounds,#8]
 	orr	$s2,$s2,$t1,lsl#8
-	orr	$s2,$s2,$t2,lsl#16
-	orr	$s2,$s2,$t3,lsl#24
 	ldrb	$s3,[$rounds,#15]
+	orr	$s2,$s2,$t2,lsl#16
 	ldrb	$t1,[$rounds,#14]
+	orr	$s2,$s2,$t3,lsl#24
 	ldrb	$t2,[$rounds,#13]
 	ldrb	$t3,[$rounds,#12]
 	orr	$s3,$s3,$t1,lsl#8
 	orr	$s3,$s3,$t2,lsl#16
 	orr	$s3,$s3,$t3,lsl#24
-
+#else
+	ldr	$s0,[$rounds,#0]
+	ldr	$s1,[$rounds,#4]
+	ldr	$s2,[$rounds,#8]
+	ldr	$s3,[$rounds,#12]
+#ifdef __ARMEL__
+	rev	$s0,$s0
+	rev	$s1,$s1
+	rev	$s2,$s2
+	rev	$s3,$s3
+#endif
+#endif
 	bl	_armv4_AES_decrypt
 
 	ldr	$rounds,[sp],#4		@ pop out
+#if __ARM_ARCH__>=7
+#ifdef __ARMEL__
+	rev	$s0,$s0
+	rev	$s1,$s1
+	rev	$s2,$s2
+	rev	$s3,$s3
+#endif
+	str	$s0,[$rounds,#0]
+	str	$s1,[$rounds,#4]
+	str	$s2,[$rounds,#8]
+	str	$s3,[$rounds,#12]
+#else
 	mov	$t1,$s0,lsr#24		@ write output in endian-neutral
 	mov	$t2,$s0,lsr#16		@ manner...
 	mov	$t3,$s0,lsr#8
 	strb	$t1,[$rounds,#0]
 	strb	$t2,[$rounds,#1]
-	strb	$t3,[$rounds,#2]
-	strb	$s0,[$rounds,#3]
 	mov	$t1,$s1,lsr#24
+	strb	$t3,[$rounds,#2]
 	mov	$t2,$s1,lsr#16
+	strb	$s0,[$rounds,#3]
 	mov	$t3,$s1,lsr#8
 	strb	$t1,[$rounds,#4]
 	strb	$t2,[$rounds,#5]
-	strb	$t3,[$rounds,#6]
-	strb	$s1,[$rounds,#7]
 	mov	$t1,$s2,lsr#24
+	strb	$t3,[$rounds,#6]
 	mov	$t2,$s2,lsr#16
+	strb	$s1,[$rounds,#7]
 	mov	$t3,$s2,lsr#8
 	strb	$t1,[$rounds,#8]
 	strb	$t2,[$rounds,#9]
-	strb	$t3,[$rounds,#10]
-	strb	$s2,[$rounds,#11]
 	mov	$t1,$s3,lsr#24
+	strb	$t3,[$rounds,#10]
 	mov	$t2,$s3,lsr#16
+	strb	$s2,[$rounds,#11]
 	mov	$t3,$s3,lsr#8
 	strb	$t1,[$rounds,#12]
 	strb	$t2,[$rounds,#13]
 	strb	$t3,[$rounds,#14]
 	strb	$s3,[$rounds,#15]
-
+#endif
+#if __ARM_ARCH__>=5
+	ldmia	sp!,{r4-r12,pc}
+#else
 	ldmia   sp!,{r4-r12,lr}
 	tst	lr,#1
 	moveq	pc,lr			@ be binary compatible with V4, yet
 	bx	lr			@ interoperable with Thumb ISA:-)
+#endif
 .size	AES_decrypt,.-AES_decrypt
 
 .type   _armv4_AES_decrypt,%function
 .align	2
 _armv4_AES_decrypt:
 	str	lr,[sp,#-4]!		@ push lr
-	ldr	$t1,[$key],#16
-	ldr	$t2,[$key,#-12]
-	ldr	$t3,[$key,#-8]
-	ldr	$i1,[$key,#-4]
-	ldr	$rounds,[$key,#240-16]
+	ldmia	$key!,{$t1-$i1}
 	eor	$s0,$s0,$t1
+	ldr	$rounds,[$key,#240-16]
 	eor	$s1,$s1,$t2
 	eor	$s2,$s2,$t3
 	eor	$s3,$s3,$i1
 	sub	$rounds,$rounds,#1
 	mov	lr,#255
 
-.Ldec_loop:
 	and	$i1,lr,$s0,lsr#16
 	and	$i2,lr,$s0,lsr#8
 	and	$i3,lr,$s0
 	mov	$s0,$s0,lsr#24
+.Ldec_loop:
 	ldr	$t1,[$tbl,$i1,lsl#2]	@ Td1[s0>>16]
-	ldr	$s0,[$tbl,$s0,lsl#2]	@ Td0[s0>>24]
-	ldr	$t2,[$tbl,$i2,lsl#2]	@ Td2[s0>>8]
-	ldr	$t3,[$tbl,$i3,lsl#2]	@ Td3[s0>>0]
-
 	and	$i1,lr,$s1		@ i0
+	ldr	$t2,[$tbl,$i2,lsl#2]	@ Td2[s0>>8]
 	and	$i2,lr,$s1,lsr#16
+	ldr	$t3,[$tbl,$i3,lsl#2]	@ Td3[s0>>0]
 	and	$i3,lr,$s1,lsr#8
+	ldr	$s0,[$tbl,$s0,lsl#2]	@ Td0[s0>>24]
 	mov	$s1,$s1,lsr#24
+
 	ldr	$i1,[$tbl,$i1,lsl#2]	@ Td3[s1>>0]
-	ldr	$s1,[$tbl,$s1,lsl#2]	@ Td0[s1>>24]
 	ldr	$i2,[$tbl,$i2,lsl#2]	@ Td1[s1>>16]
 	ldr	$i3,[$tbl,$i3,lsl#2]	@ Td2[s1>>8]
 	eor	$s0,$s0,$i1,ror#24
-	eor	$s1,$s1,$t1,ror#8
-	eor	$t2,$i2,$t2,ror#8
-	eor	$t3,$i3,$t3,ror#8
-
+	ldr	$s1,[$tbl,$s1,lsl#2]	@ Td0[s1>>24]
 	and	$i1,lr,$s2,lsr#8	@ i0
+	eor	$t2,$i2,$t2,ror#8
 	and	$i2,lr,$s2		@ i1
+	eor	$t3,$i3,$t3,ror#8
 	and	$i3,lr,$s2,lsr#16
-	mov	$s2,$s2,lsr#24
 	ldr	$i1,[$tbl,$i1,lsl#2]	@ Td2[s2>>8]
+	eor	$s1,$s1,$t1,ror#8
 	ldr	$i2,[$tbl,$i2,lsl#2]	@ Td3[s2>>0]
-	ldr	$s2,[$tbl,$s2,lsl#2]	@ Td0[s2>>24]
+	mov	$s2,$s2,lsr#24
+
 	ldr	$i3,[$tbl,$i3,lsl#2]	@ Td1[s2>>16]
 	eor	$s0,$s0,$i1,ror#16
-	eor	$s1,$s1,$i2,ror#24
-	eor	$s2,$s2,$t2,ror#8
-	eor	$t3,$i3,$t3,ror#8
-
+	ldr	$s2,[$tbl,$s2,lsl#2]	@ Td0[s2>>24]
 	and	$i1,lr,$s3,lsr#16	@ i0
+	eor	$s1,$s1,$i2,ror#24
 	and	$i2,lr,$s3,lsr#8	@ i1
+	eor	$t3,$i3,$t3,ror#8
 	and	$i3,lr,$s3		@ i2
-	mov	$s3,$s3,lsr#24
 	ldr	$i1,[$tbl,$i1,lsl#2]	@ Td1[s3>>16]
+	eor	$s2,$s2,$t2,ror#8
 	ldr	$i2,[$tbl,$i2,lsl#2]	@ Td2[s3>>8]
+	mov	$s3,$s3,lsr#24
+
 	ldr	$i3,[$tbl,$i3,lsl#2]	@ Td3[s3>>0]
-	ldr	$s3,[$tbl,$s3,lsl#2]	@ Td0[s3>>24]
 	eor	$s0,$s0,$i1,ror#8
+	ldr	$i1,[$key],#16
 	eor	$s1,$s1,$i2,ror#16
+	ldr	$s3,[$tbl,$s3,lsl#2]	@ Td0[s3>>24]
 	eor	$s2,$s2,$i3,ror#24
-	eor	$s3,$s3,$t3,ror#8
 
-	ldr	$t1,[$key],#16
-	ldr	$t2,[$key,#-12]
-	ldr	$t3,[$key,#-8]
-	ldr	$i1,[$key,#-4]
-	eor	$s0,$s0,$t1
-	eor	$s1,$s1,$t2
-	eor	$s2,$s2,$t3
-	eor	$s3,$s3,$i1
+	ldr	$t1,[$key,#-12]
+	eor	$s0,$s0,$i1
+	ldr	$t2,[$key,#-8]
+	eor	$s3,$s3,$t3,ror#8
+	ldr	$t3,[$key,#-4]
+	and	$i1,lr,$s0,lsr#16
+	eor	$s1,$s1,$t1
+	and	$i2,lr,$s0,lsr#8
+	eor	$s2,$s2,$t2
+	and	$i3,lr,$s0
+	eor	$s3,$s3,$t3
+	mov	$s0,$s0,lsr#24
 
 	subs	$rounds,$rounds,#1
 	bne	.Ldec_loop
 
 	add	$tbl,$tbl,#1024
 
-	ldr	$t1,[$tbl,#0]		@ prefetch Td4
-	ldr	$t2,[$tbl,#32]
-	ldr	$t3,[$tbl,#64]
-	ldr	$i1,[$tbl,#96]
-	ldr	$i2,[$tbl,#128]
-	ldr	$i3,[$tbl,#160]
-	ldr	$t1,[$tbl,#192]
-	ldr	$t2,[$tbl,#224]
+	ldr	$t2,[$tbl,#0]		@ prefetch Td4
+	ldr	$t3,[$tbl,#32]
+	ldr	$t1,[$tbl,#64]
+	ldr	$t2,[$tbl,#96]
+	ldr	$t3,[$tbl,#128]
+	ldr	$t1,[$tbl,#160]
+	ldr	$t2,[$tbl,#192]
+	ldr	$t3,[$tbl,#224]
 
-	and	$i1,lr,$s0,lsr#16
-	and	$i2,lr,$s0,lsr#8
-	and	$i3,lr,$s0
-	ldrb	$s0,[$tbl,$s0,lsr#24]	@ Td4[s0>>24]
+	ldrb	$s0,[$tbl,$s0]		@ Td4[s0>>24]
 	ldrb	$t1,[$tbl,$i1]		@ Td4[s0>>16]
-	ldrb	$t2,[$tbl,$i2]		@ Td4[s0>>8]
-	ldrb	$t3,[$tbl,$i3]		@ Td4[s0>>0]
-
 	and	$i1,lr,$s1		@ i0
+	ldrb	$t2,[$tbl,$i2]		@ Td4[s0>>8]
 	and	$i2,lr,$s1,lsr#16
+	ldrb	$t3,[$tbl,$i3]		@ Td4[s0>>0]
 	and	$i3,lr,$s1,lsr#8
+
 	ldrb	$i1,[$tbl,$i1]		@ Td4[s1>>0]
 	ldrb	$s1,[$tbl,$s1,lsr#24]	@ Td4[s1>>24]
 	ldrb	$i2,[$tbl,$i2]		@ Td4[s1>>16]
-	ldrb	$i3,[$tbl,$i3]		@ Td4[s1>>8]
 	eor	$s0,$i1,$s0,lsl#24
+	ldrb	$i3,[$tbl,$i3]		@ Td4[s1>>8]
 	eor	$s1,$t1,$s1,lsl#8
-	eor	$t2,$t2,$i2,lsl#8
-	eor	$t3,$t3,$i3,lsl#8
-
 	and	$i1,lr,$s2,lsr#8	@ i0
+	eor	$t2,$t2,$i2,lsl#8
 	and	$i2,lr,$s2		@ i1
-	and	$i3,lr,$s2,lsr#16
 	ldrb	$i1,[$tbl,$i1]		@ Td4[s2>>8]
+	eor	$t3,$t3,$i3,lsl#8
 	ldrb	$i2,[$tbl,$i2]		@ Td4[s2>>0]
+	and	$i3,lr,$s2,lsr#16
+
 	ldrb	$s2,[$tbl,$s2,lsr#24]	@ Td4[s2>>24]
-	ldrb	$i3,[$tbl,$i3]		@ Td4[s2>>16]
 	eor	$s0,$s0,$i1,lsl#8
+	ldrb	$i3,[$tbl,$i3]		@ Td4[s2>>16]
 	eor	$s1,$i2,$s1,lsl#16
-	eor	$s2,$t2,$s2,lsl#16
-	eor	$t3,$t3,$i3,lsl#16
-
 	and	$i1,lr,$s3,lsr#16	@ i0
+	eor	$s2,$t2,$s2,lsl#16
 	and	$i2,lr,$s3,lsr#8	@ i1
-	and	$i3,lr,$s3		@ i2
 	ldrb	$i1,[$tbl,$i1]		@ Td4[s3>>16]
+	eor	$t3,$t3,$i3,lsl#16
 	ldrb	$i2,[$tbl,$i2]		@ Td4[s3>>8]
+	and	$i3,lr,$s3		@ i2
+
 	ldrb	$i3,[$tbl,$i3]		@ Td4[s3>>0]
 	ldrb	$s3,[$tbl,$s3,lsr#24]	@ Td4[s3>>24]
 	eor	$s0,$s0,$i1,lsl#16
+	ldr	$i1,[$key,#0]
 	eor	$s1,$s1,$i2,lsl#8
+	ldr	$t1,[$key,#4]
 	eor	$s2,$i3,$s2,lsl#8
+	ldr	$t2,[$key,#8]
 	eor	$s3,$t3,$s3,lsl#24
+	ldr	$t3,[$key,#12]
 
-	ldr	lr,[sp],#4		@ pop lr
-	ldr	$t1,[$key,#0]
-	ldr	$t2,[$key,#4]
-	ldr	$t3,[$key,#8]
-	ldr	$i1,[$key,#12]
-	eor	$s0,$s0,$t1
-	eor	$s1,$s1,$t2
-	eor	$s2,$s2,$t3
-	eor	$s3,$s3,$i1
+	eor	$s0,$s0,$i1
+	eor	$s1,$s1,$t1
+	eor	$s2,$s2,$t2
+	eor	$s3,$s3,$t3
 
 	sub	$tbl,$tbl,#1024
-	mov	pc,lr			@ return
+	ldr	pc,[sp],#4		@ pop and return
 .size	_armv4_AES_decrypt,.-_armv4_AES_decrypt
 .asciz	"AES for ARMv4, CRYPTOGAMS by <appro\@openssl.org>"
+.align	2
 ___
 
 $code =~ s/\bbx\s+lr\b/.word\t0xe12fff1e/gm;	# make it possible to compile with -march=armv4
 print $code;
+close STDOUT;	# enforce flush
diff --git a/lib/libcrypto/aes/asm/aes-s390x.pl b/lib/libcrypto/aes/asm/aes-s390x.pl
index 4b27afd92fc..71d5b55077f 100644
--- a/lib/libcrypto/aes/asm/aes-s390x.pl
+++ b/lib/libcrypto/aes/asm/aes-s390x.pl
@@ -44,12 +44,57 @@
 # Unlike previous version hardware support detection takes place only
 # at the moment of key schedule setup, which is denoted in key->rounds.
 # This is done, because deferred key setup can't be made MT-safe, not
-# for key lengthes longer than 128 bits.
+# for keys longer than 128 bits.
 #
 # Add AES_cbc_encrypt, which gives incredible performance improvement,
 # it was measured to be ~6.6x. It's less than previously mentioned 8x,
 # because software implementation was optimized.
 
+# May 2010.
+#
+# Add AES_ctr32_encrypt. If hardware-assisted, it provides up to 4.3x
+# performance improvement over "generic" counter mode routine relying
+# on single-block, also hardware-assisted, AES_encrypt. "Up to" refers
+# to the fact that exact throughput value depends on current stack
+# frame alignment within 4KB page. In worst case you get ~75% of the
+# maximum, but *on average* it would be as much as ~98%. Meaning that
+# worst case is unlike, it's like hitting ravine on plateau.
+
+# November 2010.
+#
+# Adapt for -m31 build. If kernel supports what's called "highgprs"
+# feature on Linux [see /proc/cpuinfo], it's possible to use 64-bit
+# instructions and achieve "64-bit" performance even in 31-bit legacy
+# application context. The feature is not specific to any particular
+# processor, as long as it's "z-CPU". Latter implies that the code
+# remains z/Architecture specific. On z990 it was measured to perform
+# 2x better than code generated by gcc 4.3.
+
+# December 2010.
+#
+# Add support for z196 "cipher message with counter" instruction.
+# Note however that it's disengaged, because it was measured to
+# perform ~12% worse than vanilla km-based code...
+
+# February 2011.
+#
+# Add AES_xts_[en|de]crypt. This includes support for z196 km-xts-aes
+# instructions, which deliver ~70% improvement at 8KB block size over
+# vanilla km-based code, 37% - at most like 512-bytes block size.
+
+$flavour = shift;
+
+if ($flavour =~ /3[12]/) {
+	$SIZE_T=4;
+	$g="";
+} else {
+	$SIZE_T=8;
+	$g="g";
+}
+
+while (($output=shift) && ($output!~/^\w[\w\-]*\.\w+$/)) {}
+open STDOUT,">$output";
+
 $softonly=0;	# allow hardware support
 
 $t0="%r0";	$mask="%r0";
@@ -69,6 +114,8 @@ $rounds="%r13";
 $ra="%r14";
 $sp="%r15";
 
+$stdframe=16*$SIZE_T+4*8;
+
 sub _data_word()
 { my $i;
     while(defined($i=shift)) { $code.=sprintf".long\t0x%08x,0x%08x\n",$i,$i; }
@@ -210,7 +257,7 @@ $code.=<<___ if (!$softonly);
 .Lesoft:
 ___
 $code.=<<___;
-	stmg	%r3,$ra,24($sp)
+	stm${g}	%r3,$ra,3*$SIZE_T($sp)
 
 	llgf	$s0,0($inp)
 	llgf	$s1,4($inp)
@@ -220,20 +267,20 @@ $code.=<<___;
 	larl	$tbl,AES_Te
 	bras	$ra,_s390x_AES_encrypt
 
-	lg	$out,24($sp)
+	l${g}	$out,3*$SIZE_T($sp)
 	st	$s0,0($out)
 	st	$s1,4($out)
 	st	$s2,8($out)
 	st	$s3,12($out)
 
-	lmg	%r6,$ra,48($sp)
+	lm${g}	%r6,$ra,6*$SIZE_T($sp)
 	br	$ra
 .size	AES_encrypt,.-AES_encrypt
 
 .type   _s390x_AES_encrypt,\@function
 .align	16
 _s390x_AES_encrypt:
-	stg	$ra,152($sp)
+	st${g}	$ra,15*$SIZE_T($sp)
 	x	$s0,0($key)
 	x	$s1,4($key)
 	x	$s2,8($key)
@@ -397,7 +444,7 @@ _s390x_AES_encrypt:
 	or	$s2,$i3
 	or	$s3,$t3
 
-	lg	$ra,152($sp)
+	l${g}	$ra,15*$SIZE_T($sp)
 	xr	$s0,$t0
 	xr	$s1,$t2
 	x	$s2,24($key)
@@ -536,7 +583,7 @@ $code.=<<___ if (!$softonly);
 .Ldsoft:
 ___
 $code.=<<___;
-	stmg	%r3,$ra,24($sp)
+	stm${g}	%r3,$ra,3*$SIZE_T($sp)
 
 	llgf	$s0,0($inp)
 	llgf	$s1,4($inp)
@@ -546,20 +593,20 @@ $code.=<<___;
 	larl	$tbl,AES_Td
 	bras	$ra,_s390x_AES_decrypt
 
-	lg	$out,24($sp)
+	l${g}	$out,3*$SIZE_T($sp)
 	st	$s0,0($out)
 	st	$s1,4($out)
 	st	$s2,8($out)
 	st	$s3,12($out)
 
-	lmg	%r6,$ra,48($sp)
+	lm${g}	%r6,$ra,6*$SIZE_T($sp)
 	br	$ra
 .size	AES_decrypt,.-AES_decrypt
 
 .type   _s390x_AES_decrypt,\@function
 .align	16
 _s390x_AES_decrypt:
-	stg	$ra,152($sp)
+	st${g}	$ra,15*$SIZE_T($sp)
 	x	$s0,0($key)
 	x	$s1,4($key)
 	x	$s2,8($key)
@@ -703,7 +750,7 @@ _s390x_AES_decrypt:
 	nr	$i1,$mask
 	nr	$i2,$mask
 
-	lg	$ra,152($sp)
+	l${g}	$ra,15*$SIZE_T($sp)
 	or	$s1,$t1
 	l	$t0,16($key)
 	l	$t1,20($key)
@@ -736,10 +783,11 @@ $code.=<<___;
 .type	AES_set_encrypt_key,\@function
 .align	16
 AES_set_encrypt_key:
+_s390x_AES_set_encrypt_key:
 	lghi	$t0,0
-	clgr	$inp,$t0
+	cl${g}r	$inp,$t0
 	je	.Lminus1
-	clgr	$key,$t0
+	cl${g}r	$key,$t0
 	je	.Lminus1
 
 	lghi	$t0,128
@@ -765,6 +813,11 @@ $code.=<<___ if (!$softonly);
 	srl	%r5,6
 	ar	%r5,%r0
 
+	larl	%r1,OPENSSL_s390xcap_P
+	lg	%r0,0(%r1)
+	tmhl	%r0,0x4000	# check for message-security assist
+	jz	.Lekey_internal
+
 	lghi	%r0,0		# query capability vector
 	la	%r1,16($sp)
 	.long	0xb92f0042	# kmc %r4,%r2
@@ -784,7 +837,8 @@ $code.=<<___ if (!$softonly);
 	je	1f
 	lg	%r1,24($inp)
 	stg	%r1,24($key)
-1:	st	$bits,236($key)	# save bits
+1:	st	$bits,236($key)	# save bits [for debugging purposes]
+	lgr	$t0,%r5
 	st	%r5,240($key)	# save km code
 	lghi	%r2,0
 	br	%r14
@@ -792,7 +846,7 @@ ___
 $code.=<<___;
 .align	16
 .Lekey_internal:
-	stmg	%r6,%r13,48($sp)	# all non-volatile regs
+	stm${g}	%r4,%r13,4*$SIZE_T($sp)	# all non-volatile regs and $key
 
 	larl	$tbl,AES_Te+2048
 
@@ -852,8 +906,9 @@ $code.=<<___;
 	la	$key,16($key)		# key+=4
 	la	$t3,4($t3)		# i++
 	brct	$rounds,.L128_loop
+	lghi	$t0,10
 	lghi	%r2,0
-	lmg	%r6,%r13,48($sp)
+	lm${g}	%r4,%r13,4*$SIZE_T($sp)
 	br	$ra
 
 .align	16
@@ -900,8 +955,9 @@ $code.=<<___;
 	st	$s2,32($key)
 	st	$s3,36($key)
 	brct	$rounds,.L192_continue
+	lghi	$t0,12
 	lghi	%r2,0
-	lmg	%r6,%r13,48($sp)
+	lm${g}	%r4,%r13,4*$SIZE_T($sp)
 	br	$ra
 
 .align	16
@@ -962,8 +1018,9 @@ $code.=<<___;
 	st	$s2,40($key)
 	st	$s3,44($key)
 	brct	$rounds,.L256_continue
+	lghi	$t0,14
 	lghi	%r2,0
-	lmg	%r6,%r13,48($sp)
+	lm${g}	%r4,%r13,4*$SIZE_T($sp)
 	br	$ra
 
 .align	16
@@ -1014,34 +1071,26 @@ $code.=<<___;
 .type	AES_set_decrypt_key,\@function
 .align	16
 AES_set_decrypt_key:
-	stg	$key,32($sp)		# I rely on AES_set_encrypt_key to
-	stg	$ra,112($sp)		# save non-volatile registers!
-	bras	$ra,AES_set_encrypt_key
-	lg	$key,32($sp)
-	lg	$ra,112($sp)
+	#st${g}	$key,4*$SIZE_T($sp)	# I rely on AES_set_encrypt_key to
+	st${g}	$ra,14*$SIZE_T($sp)	# save non-volatile registers and $key!
+	bras	$ra,_s390x_AES_set_encrypt_key
+	#l${g}	$key,4*$SIZE_T($sp)
+	l${g}	$ra,14*$SIZE_T($sp)
 	ltgr	%r2,%r2
 	bnzr	$ra
 ___
 $code.=<<___ if (!$softonly);
-	l	$t0,240($key)
+	#l	$t0,240($key)
 	lhi	$t1,16
 	cr	$t0,$t1
 	jl	.Lgo
 	oill	$t0,0x80	# set "decrypt" bit
 	st	$t0,240($key)
 	br	$ra
-
-.align	16
-.Ldkey_internal:
-	stg	$key,32($sp)
-	stg	$ra,40($sp)
-	bras	$ra,.Lekey_internal
-	lg	$key,32($sp)
-	lg	$ra,40($sp)
 ___
 $code.=<<___;
-
-.Lgo:	llgf	$rounds,240($key)
+.align	16
+.Lgo:	lgr	$rounds,$t0	#llgf	$rounds,240($key)
 	la	$i1,0($key)
 	sllg	$i2,$rounds,4
 	la	$i2,0($i2,$key)
@@ -1118,13 +1167,14 @@ $code.=<<___;
 	la	$key,4($key)
 	brct	$rounds,.Lmix
 
-	lmg	%r6,%r13,48($sp)# as was saved by AES_set_encrypt_key!
+	lm${g}	%r6,%r13,6*$SIZE_T($sp)# as was saved by AES_set_encrypt_key!
 	lghi	%r2,0
 	br	$ra
 .size	AES_set_decrypt_key,.-AES_set_decrypt_key
 ___
 
-#void AES_cbc_encrypt(const unsigned char *in, unsigned char *out,
+########################################################################
+# void AES_cbc_encrypt(const unsigned char *in, unsigned char *out,
 #                     size_t length, const AES_KEY *key,
 #                     unsigned char *ivec, const int enc)
 {
@@ -1158,7 +1208,7 @@ $code.=<<___ if (!$softonly);
 	l	%r0,240($key)	# load kmc code
 	lghi	$key,15		# res=len%16, len-=res;
 	ngr	$key,$len
-	slgr	$len,$key
+	sl${g}r	$len,$key
 	la	%r1,16($sp)	# parameter block - ivec || key
 	jz	.Lkmc_truncated
 	.long	0xb92f0042	# kmc %r4,%r2
@@ -1176,34 +1226,34 @@ $code.=<<___ if (!$softonly);
 	tmll	%r0,0x80
 	jnz	.Lkmc_truncated_dec
 	lghi	%r1,0
-	stg	%r1,128($sp)
-	stg	%r1,136($sp)
+	stg	%r1,16*$SIZE_T($sp)
+	stg	%r1,16*$SIZE_T+8($sp)
 	bras	%r1,1f
-	mvc	128(1,$sp),0($inp)
+	mvc	16*$SIZE_T(1,$sp),0($inp)
 1:	ex	$key,0(%r1)
 	la	%r1,16($sp)	# restore parameter block
-	la	$inp,128($sp)
+	la	$inp,16*$SIZE_T($sp)
 	lghi	$len,16
 	.long	0xb92f0042	# kmc %r4,%r2
 	j	.Lkmc_done
 .align	16
 .Lkmc_truncated_dec:
-	stg	$out,64($sp)
-	la	$out,128($sp)
+	st${g}	$out,4*$SIZE_T($sp)
+	la	$out,16*$SIZE_T($sp)
 	lghi	$len,16
 	.long	0xb92f0042	# kmc %r4,%r2
-	lg	$out,64($sp)
+	l${g}	$out,4*$SIZE_T($sp)
 	bras	%r1,2f
-	mvc	0(1,$out),128($sp)
+	mvc	0(1,$out),16*$SIZE_T($sp)
 2:	ex	$key,0(%r1)
 	j	.Lkmc_done
 .align	16
 .Lcbc_software:
 ___
 $code.=<<___;
-	stmg	$key,$ra,40($sp)
+	stm${g}	$key,$ra,5*$SIZE_T($sp)
 	lhi	%r0,0
-	cl	%r0,164($sp)
+	cl	%r0,`$stdframe+$SIZE_T-4`($sp)
 	je	.Lcbc_decrypt
 
 	larl	$tbl,AES_Te
@@ -1214,10 +1264,10 @@ $code.=<<___;
 	llgf	$s3,12($ivp)
 
 	lghi	$t0,16
-	slgr	$len,$t0
+	sl${g}r	$len,$t0
 	brc	4,.Lcbc_enc_tail	# if borrow
 .Lcbc_enc_loop:
-	stmg	$inp,$out,16($sp)
+	stm${g}	$inp,$out,2*$SIZE_T($sp)
 	x	$s0,0($inp)
 	x	$s1,4($inp)
 	x	$s2,8($inp)
@@ -1226,7 +1276,7 @@ $code.=<<___;
 
 	bras	$ra,_s390x_AES_encrypt
 
-	lmg	$inp,$key,16($sp)
+	lm${g}	$inp,$key,2*$SIZE_T($sp)
 	st	$s0,0($out)
 	st	$s1,4($out)
 	st	$s2,8($out)
@@ -1235,33 +1285,33 @@ $code.=<<___;
 	la	$inp,16($inp)
 	la	$out,16($out)
 	lghi	$t0,16
-	ltgr	$len,$len
+	lt${g}r	$len,$len
 	jz	.Lcbc_enc_done
-	slgr	$len,$t0
+	sl${g}r	$len,$t0
 	brc	4,.Lcbc_enc_tail	# if borrow
 	j	.Lcbc_enc_loop
 .align	16
 .Lcbc_enc_done:
-	lg	$ivp,48($sp)
+	l${g}	$ivp,6*$SIZE_T($sp)
 	st	$s0,0($ivp)
 	st	$s1,4($ivp)	
 	st	$s2,8($ivp)
 	st	$s3,12($ivp)
 
-	lmg	%r7,$ra,56($sp)
+	lm${g}	%r7,$ra,7*$SIZE_T($sp)
 	br	$ra
 
 .align	16
 .Lcbc_enc_tail:
 	aghi	$len,15
 	lghi	$t0,0
-	stg	$t0,128($sp)
-	stg	$t0,136($sp)
+	stg	$t0,16*$SIZE_T($sp)
+	stg	$t0,16*$SIZE_T+8($sp)
 	bras	$t1,3f
-	mvc	128(1,$sp),0($inp)
+	mvc	16*$SIZE_T(1,$sp),0($inp)
 3:	ex	$len,0($t1)
 	lghi	$len,0
-	la	$inp,128($sp)
+	la	$inp,16*$SIZE_T($sp)
 	j	.Lcbc_enc_loop
 
 .align	16
@@ -1270,10 +1320,10 @@ $code.=<<___;
 
 	lg	$t0,0($ivp)
 	lg	$t1,8($ivp)
-	stmg	$t0,$t1,128($sp)
+	stmg	$t0,$t1,16*$SIZE_T($sp)
 
 .Lcbc_dec_loop:
-	stmg	$inp,$out,16($sp)
+	stm${g}	$inp,$out,2*$SIZE_T($sp)
 	llgf	$s0,0($inp)
 	llgf	$s1,4($inp)
 	llgf	$s2,8($inp)
@@ -1282,7 +1332,7 @@ $code.=<<___;
 
 	bras	$ra,_s390x_AES_decrypt
 
-	lmg	$inp,$key,16($sp)
+	lm${g}	$inp,$key,2*$SIZE_T($sp)
 	sllg	$s0,$s0,32
 	sllg	$s2,$s2,32
 	lr	$s0,$s1
@@ -1290,15 +1340,15 @@ $code.=<<___;
 
 	lg	$t0,0($inp)
 	lg	$t1,8($inp)
-	xg	$s0,128($sp)
-	xg	$s2,136($sp)
+	xg	$s0,16*$SIZE_T($sp)
+	xg	$s2,16*$SIZE_T+8($sp)
 	lghi	$s1,16
-	slgr	$len,$s1
+	sl${g}r	$len,$s1
 	brc	4,.Lcbc_dec_tail	# if borrow
 	brc	2,.Lcbc_dec_done	# if zero
 	stg	$s0,0($out)
 	stg	$s2,8($out)
-	stmg	$t0,$t1,128($sp)
+	stmg	$t0,$t1,16*$SIZE_T($sp)
 
 	la	$inp,16($inp)
 	la	$out,16($out)
@@ -1308,7 +1358,7 @@ $code.=<<___;
 	stg	$s0,0($out)
 	stg	$s2,8($out)
 .Lcbc_dec_exit:
-	lmg	$ivp,$ra,48($sp)
+	lm${g}	%r6,$ra,6*$SIZE_T($sp)
 	stmg	$t0,$t1,0($ivp)
 
 	br	$ra
@@ -1316,18 +1366,872 @@ $code.=<<___;
 .align	16
 .Lcbc_dec_tail:
 	aghi	$len,15
-	stg	$s0,128($sp)
-	stg	$s2,136($sp)
+	stg	$s0,16*$SIZE_T($sp)
+	stg	$s2,16*$SIZE_T+8($sp)
 	bras	$s1,4f
-	mvc	0(1,$out),128($sp)
+	mvc	0(1,$out),16*$SIZE_T($sp)
 4:	ex	$len,0($s1)
 	j	.Lcbc_dec_exit
 .size	AES_cbc_encrypt,.-AES_cbc_encrypt
 ___
 }
+########################################################################
+# void AES_ctr32_encrypt(const unsigned char *in, unsigned char *out,
+#                     size_t blocks, const AES_KEY *key,
+#                     const unsigned char *ivec)
+{
+my $inp="%r2";
+my $out="%r4";	# blocks and out are swapped
+my $len="%r3";
+my $key="%r5";	my $iv0="%r5";
+my $ivp="%r6";
+my $fp ="%r7";
+
+$code.=<<___;
+.globl	AES_ctr32_encrypt
+.type	AES_ctr32_encrypt,\@function
+.align	16
+AES_ctr32_encrypt:
+	xgr	%r3,%r4		# flip %r3 and %r4, $out and $len
+	xgr	%r4,%r3
+	xgr	%r3,%r4
+	llgfr	$len,$len	# safe in ctr32 subroutine even in 64-bit case
+___
+$code.=<<___ if (!$softonly);
+	l	%r0,240($key)
+	lhi	%r1,16
+	clr	%r0,%r1
+	jl	.Lctr32_software
+
+	stm${g}	%r6,$s3,6*$SIZE_T($sp)
+
+	slgr	$out,$inp
+	la	%r1,0($key)	# %r1 is permanent copy of $key
+	lg	$iv0,0($ivp)	# load ivec
+	lg	$ivp,8($ivp)
+
+	# prepare and allocate stack frame at the top of 4K page
+	# with 1K reserved for eventual signal handling
+	lghi	$s0,-1024-256-16# guarantee at least 256-bytes buffer
+	lghi	$s1,-4096
+	algr	$s0,$sp
+	lgr	$fp,$sp
+	ngr	$s0,$s1		# align at page boundary
+	slgr	$fp,$s0		# total buffer size
+	lgr	$s2,$sp
+	lghi	$s1,1024+16	# sl[g]fi is extended-immediate facility
+	slgr	$fp,$s1		# deduct reservation to get usable buffer size
+	# buffer size is at lest 256 and at most 3072+256-16
+
+	la	$sp,1024($s0)	# alloca
+	srlg	$fp,$fp,4	# convert bytes to blocks, minimum 16
+	st${g}	$s2,0($sp)	# back-chain
+	st${g}	$fp,$SIZE_T($sp)
+
+	slgr	$len,$fp
+	brc	1,.Lctr32_hw_switch	# not zero, no borrow
+	algr	$fp,$len	# input is shorter than allocated buffer
+	lghi	$len,0
+	st${g}	$fp,$SIZE_T($sp)
+
+.Lctr32_hw_switch:
+___
+$code.=<<___ if (0);	######### kmctr code was measured to be ~12% slower
+	larl	$s0,OPENSSL_s390xcap_P
+	lg	$s0,8($s0)
+	tmhh	$s0,0x0004	# check for message_security-assist-4
+	jz	.Lctr32_km_loop
+
+	llgfr	$s0,%r0
+	lgr	$s1,%r1
+	lghi	%r0,0
+	la	%r1,16($sp)
+	.long	0xb92d2042	# kmctr %r4,%r2,%r2
+
+	llihh	%r0,0x8000	# check if kmctr supports the function code
+	srlg	%r0,%r0,0($s0)
+	ng	%r0,16($sp)
+	lgr	%r0,$s0
+	lgr	%r1,$s1
+	jz	.Lctr32_km_loop
+
+####### kmctr code
+	algr	$out,$inp	# restore $out
+	lgr	$s1,$len	# $s1 undertakes $len
+	j	.Lctr32_kmctr_loop
+.align	16
+.Lctr32_kmctr_loop:
+	la	$s2,16($sp)
+	lgr	$s3,$fp
+.Lctr32_kmctr_prepare:
+	stg	$iv0,0($s2)
+	stg	$ivp,8($s2)
+	la	$s2,16($s2)
+	ahi	$ivp,1		# 32-bit increment, preserves upper half
+	brct	$s3,.Lctr32_kmctr_prepare
+
+	#la	$inp,0($inp)	# inp
+	sllg	$len,$fp,4	# len
+	#la	$out,0($out)	# out
+	la	$s2,16($sp)	# iv
+	.long	0xb92da042	# kmctr $out,$s2,$inp
+	brc	1,.-4		# pay attention to "partial completion"
+
+	slgr	$s1,$fp
+	brc	1,.Lctr32_kmctr_loop	# not zero, no borrow
+	algr	$fp,$s1
+	lghi	$s1,0
+	brc	4+1,.Lctr32_kmctr_loop	# not zero
+
+	l${g}	$sp,0($sp)
+	lm${g}	%r6,$s3,6*$SIZE_T($sp)
+	br	$ra
+.align	16
+___
+$code.=<<___;
+.Lctr32_km_loop:
+	la	$s2,16($sp)
+	lgr	$s3,$fp
+.Lctr32_km_prepare:
+	stg	$iv0,0($s2)
+	stg	$ivp,8($s2)
+	la	$s2,16($s2)
+	ahi	$ivp,1		# 32-bit increment, preserves upper half
+	brct	$s3,.Lctr32_km_prepare
+
+	la	$s0,16($sp)	# inp
+	sllg	$s1,$fp,4	# len
+	la	$s2,16($sp)	# out
+	.long	0xb92e00a8	# km %r10,%r8
+	brc	1,.-4		# pay attention to "partial completion"
+
+	la	$s2,16($sp)
+	lgr	$s3,$fp
+	slgr	$s2,$inp
+.Lctr32_km_xor:
+	lg	$s0,0($inp)
+	lg	$s1,8($inp)
+	xg	$s0,0($s2,$inp)
+	xg	$s1,8($s2,$inp)
+	stg	$s0,0($out,$inp)
+	stg	$s1,8($out,$inp)
+	la	$inp,16($inp)
+	brct	$s3,.Lctr32_km_xor
+
+	slgr	$len,$fp
+	brc	1,.Lctr32_km_loop	# not zero, no borrow
+	algr	$fp,$len
+	lghi	$len,0
+	brc	4+1,.Lctr32_km_loop	# not zero
+
+	l${g}	$s0,0($sp)
+	l${g}	$s1,$SIZE_T($sp)
+	la	$s2,16($sp)
+.Lctr32_km_zap:
+	stg	$s0,0($s2)
+	stg	$s0,8($s2)
+	la	$s2,16($s2)
+	brct	$s1,.Lctr32_km_zap
+
+	la	$sp,0($s0)
+	lm${g}	%r6,$s3,6*$SIZE_T($sp)
+	br	$ra
+.align	16
+.Lctr32_software:
+___
+$code.=<<___;
+	stm${g}	$key,$ra,5*$SIZE_T($sp)
+	sl${g}r	$inp,$out
+	larl	$tbl,AES_Te
+	llgf	$t1,12($ivp)
+
+.Lctr32_loop:
+	stm${g}	$inp,$out,2*$SIZE_T($sp)
+	llgf	$s0,0($ivp)
+	llgf	$s1,4($ivp)
+	llgf	$s2,8($ivp)
+	lgr	$s3,$t1
+	st	$t1,16*$SIZE_T($sp)
+	lgr	%r4,$key
+
+	bras	$ra,_s390x_AES_encrypt
+
+	lm${g}	$inp,$ivp,2*$SIZE_T($sp)
+	llgf	$t1,16*$SIZE_T($sp)
+	x	$s0,0($inp,$out)
+	x	$s1,4($inp,$out)
+	x	$s2,8($inp,$out)
+	x	$s3,12($inp,$out)
+	stm	$s0,$s3,0($out)
+
+	la	$out,16($out)
+	ahi	$t1,1		# 32-bit increment
+	brct	$len,.Lctr32_loop
+
+	lm${g}	%r6,$ra,6*$SIZE_T($sp)
+	br	$ra
+.size	AES_ctr32_encrypt,.-AES_ctr32_encrypt
+___
+}
+
+########################################################################
+# void AES_xts_encrypt(const char *inp,char *out,size_t len,
+#	const AES_KEY *key1, const AES_KEY *key2,
+#	const unsigned char iv[16]);
+#
+{
+my $inp="%r2";
+my $out="%r4";	# len and out are swapped
+my $len="%r3";
+my $key1="%r5";	# $i1
+my $key2="%r6";	# $i2
+my $fp="%r7";	# $i3
+my $tweak=16*$SIZE_T+16;	# or $stdframe-16, bottom of the frame...
+
+$code.=<<___;
+.type	_s390x_xts_km,\@function
+.align	16
+_s390x_xts_km:
+___
+$code.=<<___ if(1);
+	llgfr	$s0,%r0			# put aside the function code
+	lghi	$s1,0x7f
+	nr	$s1,%r0
+	lghi	%r0,0			# query capability vector
+	la	%r1,$tweak-16($sp)
+	.long	0xb92e0042		# km %r4,%r2
+	llihh	%r1,0x8000
+	srlg	%r1,%r1,32($s1)		# check for 32+function code
+	ng	%r1,$tweak-16($sp)
+	lgr	%r0,$s0			# restore the function code
+	la	%r1,0($key1)		# restore $key1
+	jz	.Lxts_km_vanilla
+
+	lmg	$i2,$i3,$tweak($sp)	# put aside the tweak value
+	algr	$out,$inp
+
+	oill	%r0,32			# switch to xts function code
+	aghi	$s1,-18			#
+	sllg	$s1,$s1,3		# (function code - 18)*8, 0 or 16
+	la	%r1,$tweak-16($sp)
+	slgr	%r1,$s1			# parameter block position
+	lmg	$s0,$s3,0($key1)	# load 256 bits of key material,
+	stmg	$s0,$s3,0(%r1)		# and copy it to parameter block.
+					# yes, it contains junk and overlaps
+					# with the tweak in 128-bit case.
+					# it's done to avoid conditional
+					# branch.
+	stmg	$i2,$i3,$tweak($sp)	# "re-seat" the tweak value
+
+	.long	0xb92e0042		# km %r4,%r2
+	brc	1,.-4			# pay attention to "partial completion"
+
+	lrvg	$s0,$tweak+0($sp)	# load the last tweak
+	lrvg	$s1,$tweak+8($sp)
+	stmg	%r0,%r3,$tweak-32($sp)	# wipe copy of the key
+
+	nill	%r0,0xffdf		# switch back to original function code
+	la	%r1,0($key1)		# restore pointer to $key1
+	slgr	$out,$inp
+
+	llgc	$len,2*$SIZE_T-1($sp)
+	nill	$len,0x0f		# $len%=16
+	br	$ra
+	
+.align	16
+.Lxts_km_vanilla:
+___
+$code.=<<___;
+	# prepare and allocate stack frame at the top of 4K page
+	# with 1K reserved for eventual signal handling
+	lghi	$s0,-1024-256-16# guarantee at least 256-bytes buffer
+	lghi	$s1,-4096
+	algr	$s0,$sp
+	lgr	$fp,$sp
+	ngr	$s0,$s1		# align at page boundary
+	slgr	$fp,$s0		# total buffer size
+	lgr	$s2,$sp
+	lghi	$s1,1024+16	# sl[g]fi is extended-immediate facility
+	slgr	$fp,$s1		# deduct reservation to get usable buffer size
+	# buffer size is at lest 256 and at most 3072+256-16
+
+	la	$sp,1024($s0)	# alloca
+	nill	$fp,0xfff0	# round to 16*n
+	st${g}	$s2,0($sp)	# back-chain
+	nill	$len,0xfff0	# redundant
+	st${g}	$fp,$SIZE_T($sp)
+
+	slgr	$len,$fp
+	brc	1,.Lxts_km_go	# not zero, no borrow
+	algr	$fp,$len	# input is shorter than allocated buffer
+	lghi	$len,0
+	st${g}	$fp,$SIZE_T($sp)
+
+.Lxts_km_go:
+	lrvg	$s0,$tweak+0($s2)	# load the tweak value in little-endian
+	lrvg	$s1,$tweak+8($s2)
+
+	la	$s2,16($sp)		# vector of ascending tweak values
+	slgr	$s2,$inp
+	srlg	$s3,$fp,4
+	j	.Lxts_km_start
+
+.Lxts_km_loop:
+	la	$s2,16($sp)
+	slgr	$s2,$inp
+	srlg	$s3,$fp,4
+.Lxts_km_prepare:
+	lghi	$i1,0x87
+	srag	$i2,$s1,63		# broadcast upper bit
+	ngr	$i1,$i2			# rem
+	algr	$s0,$s0
+	alcgr	$s1,$s1
+	xgr	$s0,$i1
+.Lxts_km_start:
+	lrvgr	$i1,$s0			# flip byte order
+	lrvgr	$i2,$s1
+	stg	$i1,0($s2,$inp)
+	stg	$i2,8($s2,$inp)
+	xg	$i1,0($inp)
+	xg	$i2,8($inp)
+	stg	$i1,0($out,$inp)
+	stg	$i2,8($out,$inp)
+	la	$inp,16($inp)
+	brct	$s3,.Lxts_km_prepare
+
+	slgr	$inp,$fp		# rewind $inp
+	la	$s2,0($out,$inp)
+	lgr	$s3,$fp
+	.long	0xb92e00aa		# km $s2,$s2
+	brc	1,.-4			# pay attention to "partial completion"
+
+	la	$s2,16($sp)
+	slgr	$s2,$inp
+	srlg	$s3,$fp,4
+.Lxts_km_xor:
+	lg	$i1,0($out,$inp)
+	lg	$i2,8($out,$inp)
+	xg	$i1,0($s2,$inp)
+	xg	$i2,8($s2,$inp)
+	stg	$i1,0($out,$inp)
+	stg	$i2,8($out,$inp)
+	la	$inp,16($inp)
+	brct	$s3,.Lxts_km_xor
+
+	slgr	$len,$fp
+	brc	1,.Lxts_km_loop		# not zero, no borrow
+	algr	$fp,$len
+	lghi	$len,0
+	brc	4+1,.Lxts_km_loop	# not zero
+
+	l${g}	$i1,0($sp)		# back-chain
+	llgf	$fp,`2*$SIZE_T-4`($sp)	# bytes used
+	la	$i2,16($sp)
+	srlg	$fp,$fp,4
+.Lxts_km_zap:
+	stg	$i1,0($i2)
+	stg	$i1,8($i2)
+	la	$i2,16($i2)
+	brct	$fp,.Lxts_km_zap
+
+	la	$sp,0($i1)
+	llgc	$len,2*$SIZE_T-1($i1)
+	nill	$len,0x0f		# $len%=16
+	bzr	$ra
+
+	# generate one more tweak...
+	lghi	$i1,0x87
+	srag	$i2,$s1,63		# broadcast upper bit
+	ngr	$i1,$i2			# rem
+	algr	$s0,$s0
+	alcgr	$s1,$s1
+	xgr	$s0,$i1
+
+	ltr	$len,$len		# clear zero flag
+	br	$ra
+.size	_s390x_xts_km,.-_s390x_xts_km
+
+.globl	AES_xts_encrypt
+.type	AES_xts_encrypt,\@function
+.align	16
+AES_xts_encrypt:
+	xgr	%r3,%r4			# flip %r3 and %r4, $out and $len
+	xgr	%r4,%r3
+	xgr	%r3,%r4
+___
+$code.=<<___ if ($SIZE_T==4);
+	llgfr	$len,$len
+___
+$code.=<<___;
+	st${g}	$len,1*$SIZE_T($sp)	# save copy of $len
+	srag	$len,$len,4		# formally wrong, because it expands
+					# sign byte, but who can afford asking
+					# to process more than 2^63-1 bytes?
+					# I use it, because it sets condition
+					# code...
+	bcr	8,$ra			# abort if zero (i.e. less than 16)
+___
+$code.=<<___ if (!$softonly);
+	llgf	%r0,240($key2)
+	lhi	%r1,16
+	clr	%r0,%r1
+	jl	.Lxts_enc_software
+
+	st${g}	$ra,5*$SIZE_T($sp)
+	stm${g}	%r6,$s3,6*$SIZE_T($sp)
+
+	sllg	$len,$len,4		# $len&=~15
+	slgr	$out,$inp
+
+	# generate the tweak value
+	l${g}	$s3,$stdframe($sp)	# pointer to iv
+	la	$s2,$tweak($sp)
+	lmg	$s0,$s1,0($s3)
+	lghi	$s3,16
+	stmg	$s0,$s1,0($s2)
+	la	%r1,0($key2)		# $key2 is not needed anymore
+	.long	0xb92e00aa		# km $s2,$s2, generate the tweak
+	brc	1,.-4			# can this happen?
+
+	l	%r0,240($key1)
+	la	%r1,0($key1)		# $key1 is not needed anymore
+	bras	$ra,_s390x_xts_km
+	jz	.Lxts_enc_km_done
+
+	aghi	$inp,-16		# take one step back
+	la	$i3,0($out,$inp)	# put aside real $out
+.Lxts_enc_km_steal:
+	llgc	$i1,16($inp)
+	llgc	$i2,0($out,$inp)
+	stc	$i1,0($out,$inp)
+	stc	$i2,16($out,$inp)
+	la	$inp,1($inp)
+	brct	$len,.Lxts_enc_km_steal
+
+	la	$s2,0($i3)
+	lghi	$s3,16
+	lrvgr	$i1,$s0			# flip byte order
+	lrvgr	$i2,$s1
+	xg	$i1,0($s2)
+	xg	$i2,8($s2)
+	stg	$i1,0($s2)
+	stg	$i2,8($s2)
+	.long	0xb92e00aa		# km $s2,$s2
+	brc	1,.-4			# can this happen?
+	lrvgr	$i1,$s0			# flip byte order
+	lrvgr	$i2,$s1
+	xg	$i1,0($i3)
+	xg	$i2,8($i3)
+	stg	$i1,0($i3)
+	stg	$i2,8($i3)
+
+.Lxts_enc_km_done:
+	stg	$sp,$tweak+0($sp)	# wipe tweak
+	stg	$sp,$tweak+8($sp)
+	l${g}	$ra,5*$SIZE_T($sp)
+	lm${g}	%r6,$s3,6*$SIZE_T($sp)
+	br	$ra
+.align	16
+.Lxts_enc_software:
+___
+$code.=<<___;
+	stm${g}	%r6,$ra,6*$SIZE_T($sp)
+
+	slgr	$out,$inp
+
+	l${g}	$s3,$stdframe($sp)	# ivp
+	llgf	$s0,0($s3)		# load iv
+	llgf	$s1,4($s3)
+	llgf	$s2,8($s3)
+	llgf	$s3,12($s3)
+	stm${g}	%r2,%r5,2*$SIZE_T($sp)
+	la	$key,0($key2)
+	larl	$tbl,AES_Te
+	bras	$ra,_s390x_AES_encrypt	# generate the tweak
+	lm${g}	%r2,%r5,2*$SIZE_T($sp)
+	stm	$s0,$s3,$tweak($sp)	# save the tweak
+	j	.Lxts_enc_enter
+
+.align	16
+.Lxts_enc_loop:
+	lrvg	$s1,$tweak+0($sp)	# load the tweak in little-endian
+	lrvg	$s3,$tweak+8($sp)
+	lghi	%r1,0x87
+	srag	%r0,$s3,63		# broadcast upper bit
+	ngr	%r1,%r0			# rem
+	algr	$s1,$s1
+	alcgr	$s3,$s3
+	xgr	$s1,%r1
+	lrvgr	$s1,$s1			# flip byte order
+	lrvgr	$s3,$s3
+	srlg	$s0,$s1,32		# smash the tweak to 4x32-bits 
+	stg	$s1,$tweak+0($sp)	# save the tweak
+	llgfr	$s1,$s1
+	srlg	$s2,$s3,32
+	stg	$s3,$tweak+8($sp)
+	llgfr	$s3,$s3
+	la	$inp,16($inp)		# $inp+=16
+.Lxts_enc_enter:
+	x	$s0,0($inp)		# ^=*($inp)
+	x	$s1,4($inp)
+	x	$s2,8($inp)
+	x	$s3,12($inp)
+	stm${g}	%r2,%r3,2*$SIZE_T($sp)	# only two registers are changing
+	la	$key,0($key1)
+	bras	$ra,_s390x_AES_encrypt
+	lm${g}	%r2,%r5,2*$SIZE_T($sp)
+	x	$s0,$tweak+0($sp)	# ^=tweak
+	x	$s1,$tweak+4($sp)
+	x	$s2,$tweak+8($sp)
+	x	$s3,$tweak+12($sp)
+	st	$s0,0($out,$inp)
+	st	$s1,4($out,$inp)
+	st	$s2,8($out,$inp)
+	st	$s3,12($out,$inp)
+	brct${g}	$len,.Lxts_enc_loop
+
+	llgc	$len,`2*$SIZE_T-1`($sp)
+	nill	$len,0x0f		# $len%16
+	jz	.Lxts_enc_done
+
+	la	$i3,0($inp,$out)	# put aside real $out
+.Lxts_enc_steal:
+	llgc	%r0,16($inp)
+	llgc	%r1,0($out,$inp)
+	stc	%r0,0($out,$inp)
+	stc	%r1,16($out,$inp)
+	la	$inp,1($inp)
+	brct	$len,.Lxts_enc_steal
+	la	$out,0($i3)		# restore real $out
+
+	# generate last tweak...
+	lrvg	$s1,$tweak+0($sp)	# load the tweak in little-endian
+	lrvg	$s3,$tweak+8($sp)
+	lghi	%r1,0x87
+	srag	%r0,$s3,63		# broadcast upper bit
+	ngr	%r1,%r0			# rem
+	algr	$s1,$s1
+	alcgr	$s3,$s3
+	xgr	$s1,%r1
+	lrvgr	$s1,$s1			# flip byte order
+	lrvgr	$s3,$s3
+	srlg	$s0,$s1,32		# smash the tweak to 4x32-bits 
+	stg	$s1,$tweak+0($sp)	# save the tweak
+	llgfr	$s1,$s1
+	srlg	$s2,$s3,32
+	stg	$s3,$tweak+8($sp)
+	llgfr	$s3,$s3
+
+	x	$s0,0($out)		# ^=*(inp)|stolen cipther-text
+	x	$s1,4($out)
+	x	$s2,8($out)
+	x	$s3,12($out)
+	st${g}	$out,4*$SIZE_T($sp)
+	la	$key,0($key1)
+	bras	$ra,_s390x_AES_encrypt
+	l${g}	$out,4*$SIZE_T($sp)
+	x	$s0,`$tweak+0`($sp)	# ^=tweak
+	x	$s1,`$tweak+4`($sp)
+	x	$s2,`$tweak+8`($sp)
+	x	$s3,`$tweak+12`($sp)
+	st	$s0,0($out)
+	st	$s1,4($out)
+	st	$s2,8($out)
+	st	$s3,12($out)
+
+.Lxts_enc_done:
+	stg	$sp,$tweak+0($sp)	# wipe tweak
+	stg	$sp,$twesk+8($sp)
+	lm${g}	%r6,$ra,6*$SIZE_T($sp)
+	br	$ra
+.size	AES_xts_encrypt,.-AES_xts_encrypt
+___
+# void AES_xts_decrypt(const char *inp,char *out,size_t len,
+#	const AES_KEY *key1, const AES_KEY *key2,
+#	const unsigned char iv[16]);
+#
+$code.=<<___;
+.globl	AES_xts_decrypt
+.type	AES_xts_decrypt,\@function
+.align	16
+AES_xts_decrypt:
+	xgr	%r3,%r4			# flip %r3 and %r4, $out and $len
+	xgr	%r4,%r3
+	xgr	%r3,%r4
+___
+$code.=<<___ if ($SIZE_T==4);
+	llgfr	$len,$len
+___
+$code.=<<___;
+	st${g}	$len,1*$SIZE_T($sp)	# save copy of $len
+	aghi	$len,-16
+	bcr	4,$ra			# abort if less than zero. formally
+					# wrong, because $len is unsigned,
+					# but who can afford asking to
+					# process more than 2^63-1 bytes?
+	tmll	$len,0x0f
+	jnz	.Lxts_dec_proceed
+	aghi	$len,16
+.Lxts_dec_proceed:
+___
+$code.=<<___ if (!$softonly);
+	llgf	%r0,240($key2)
+	lhi	%r1,16
+	clr	%r0,%r1
+	jl	.Lxts_dec_software
+
+	st${g}	$ra,5*$SIZE_T($sp)
+	stm${g}	%r6,$s3,6*$SIZE_T($sp)
+
+	nill	$len,0xfff0		# $len&=~15
+	slgr	$out,$inp
+
+	# generate the tweak value
+	l${g}	$s3,$stdframe($sp)	# pointer to iv
+	la	$s2,$tweak($sp)
+	lmg	$s0,$s1,0($s3)
+	lghi	$s3,16
+	stmg	$s0,$s1,0($s2)
+	la	%r1,0($key2)		# $key2 is not needed past this point
+	.long	0xb92e00aa		# km $s2,$s2, generate the tweak
+	brc	1,.-4			# can this happen?
+
+	l	%r0,240($key1)
+	la	%r1,0($key1)		# $key1 is not needed anymore
+
+	ltgr	$len,$len
+	jz	.Lxts_dec_km_short
+	bras	$ra,_s390x_xts_km
+	jz	.Lxts_dec_km_done
+
+	lrvgr	$s2,$s0			# make copy in reverse byte order
+	lrvgr	$s3,$s1
+	j	.Lxts_dec_km_2ndtweak
+
+.Lxts_dec_km_short:
+	llgc	$len,`2*$SIZE_T-1`($sp)
+	nill	$len,0x0f		# $len%=16
+	lrvg	$s0,$tweak+0($sp)	# load the tweak
+	lrvg	$s1,$tweak+8($sp)
+	lrvgr	$s2,$s0			# make copy in reverse byte order
+	lrvgr	$s3,$s1
+
+.Lxts_dec_km_2ndtweak:
+	lghi	$i1,0x87
+	srag	$i2,$s1,63		# broadcast upper bit
+	ngr	$i1,$i2			# rem
+	algr	$s0,$s0
+	alcgr	$s1,$s1
+	xgr	$s0,$i1
+	lrvgr	$i1,$s0			# flip byte order
+	lrvgr	$i2,$s1
+
+	xg	$i1,0($inp)
+	xg	$i2,8($inp)
+	stg	$i1,0($out,$inp)
+	stg	$i2,8($out,$inp)
+	la	$i2,0($out,$inp)
+	lghi	$i3,16
+	.long	0xb92e0066		# km $i2,$i2
+	brc	1,.-4			# can this happen?
+	lrvgr	$i1,$s0
+	lrvgr	$i2,$s1
+	xg	$i1,0($out,$inp)
+	xg	$i2,8($out,$inp)
+	stg	$i1,0($out,$inp)
+	stg	$i2,8($out,$inp)
+
+	la	$i3,0($out,$inp)	# put aside real $out
+.Lxts_dec_km_steal:
+	llgc	$i1,16($inp)
+	llgc	$i2,0($out,$inp)
+	stc	$i1,0($out,$inp)
+	stc	$i2,16($out,$inp)
+	la	$inp,1($inp)
+	brct	$len,.Lxts_dec_km_steal
+
+	lgr	$s0,$s2
+	lgr	$s1,$s3
+	xg	$s0,0($i3)
+	xg	$s1,8($i3)
+	stg	$s0,0($i3)
+	stg	$s1,8($i3)
+	la	$s0,0($i3)
+	lghi	$s1,16
+	.long	0xb92e0088		# km $s0,$s0
+	brc	1,.-4			# can this happen?
+	xg	$s2,0($i3)
+	xg	$s3,8($i3)
+	stg	$s2,0($i3)
+	stg	$s3,8($i3)
+.Lxts_dec_km_done:
+	stg	$sp,$tweak+0($sp)	# wipe tweak
+	stg	$sp,$tweak+8($sp)
+	l${g}	$ra,5*$SIZE_T($sp)
+	lm${g}	%r6,$s3,6*$SIZE_T($sp)
+	br	$ra
+.align	16
+.Lxts_dec_software:
+___
+$code.=<<___;
+	stm${g}	%r6,$ra,6*$SIZE_T($sp)
+
+	srlg	$len,$len,4
+	slgr	$out,$inp
+
+	l${g}	$s3,$stdframe($sp)	# ivp
+	llgf	$s0,0($s3)		# load iv
+	llgf	$s1,4($s3)
+	llgf	$s2,8($s3)
+	llgf	$s3,12($s3)
+	stm${g}	%r2,%r5,2*$SIZE_T($sp)
+	la	$key,0($key2)
+	larl	$tbl,AES_Te
+	bras	$ra,_s390x_AES_encrypt	# generate the tweak
+	lm${g}	%r2,%r5,2*$SIZE_T($sp)
+	larl	$tbl,AES_Td
+	lt${g}r	$len,$len
+	stm	$s0,$s3,$tweak($sp)	# save the tweak
+	jz	.Lxts_dec_short
+	j	.Lxts_dec_enter
+
+.align	16
+.Lxts_dec_loop:
+	lrvg	$s1,$tweak+0($sp)	# load the tweak in little-endian
+	lrvg	$s3,$tweak+8($sp)
+	lghi	%r1,0x87
+	srag	%r0,$s3,63		# broadcast upper bit
+	ngr	%r1,%r0			# rem
+	algr	$s1,$s1
+	alcgr	$s3,$s3
+	xgr	$s1,%r1
+	lrvgr	$s1,$s1			# flip byte order
+	lrvgr	$s3,$s3
+	srlg	$s0,$s1,32		# smash the tweak to 4x32-bits 
+	stg	$s1,$tweak+0($sp)	# save the tweak
+	llgfr	$s1,$s1
+	srlg	$s2,$s3,32
+	stg	$s3,$tweak+8($sp)
+	llgfr	$s3,$s3
+.Lxts_dec_enter:
+	x	$s0,0($inp)		# tweak^=*(inp)
+	x	$s1,4($inp)
+	x	$s2,8($inp)
+	x	$s3,12($inp)
+	stm${g}	%r2,%r3,2*$SIZE_T($sp)	# only two registers are changing
+	la	$key,0($key1)
+	bras	$ra,_s390x_AES_decrypt
+	lm${g}	%r2,%r5,2*$SIZE_T($sp)
+	x	$s0,$tweak+0($sp)	# ^=tweak
+	x	$s1,$tweak+4($sp)
+	x	$s2,$tweak+8($sp)
+	x	$s3,$tweak+12($sp)
+	st	$s0,0($out,$inp)
+	st	$s1,4($out,$inp)
+	st	$s2,8($out,$inp)
+	st	$s3,12($out,$inp)
+	la	$inp,16($inp)
+	brct${g}	$len,.Lxts_dec_loop
+
+	llgc	$len,`2*$SIZE_T-1`($sp)
+	nill	$len,0x0f		# $len%16
+	jz	.Lxts_dec_done
+
+	# generate pair of tweaks...
+	lrvg	$s1,$tweak+0($sp)	# load the tweak in little-endian
+	lrvg	$s3,$tweak+8($sp)
+	lghi	%r1,0x87
+	srag	%r0,$s3,63		# broadcast upper bit
+	ngr	%r1,%r0			# rem
+	algr	$s1,$s1
+	alcgr	$s3,$s3
+	xgr	$s1,%r1
+	lrvgr	$i2,$s1			# flip byte order
+	lrvgr	$i3,$s3
+	stmg	$i2,$i3,$tweak($sp)	# save the 1st tweak
+	j	.Lxts_dec_2ndtweak
+
+.align	16
+.Lxts_dec_short:
+	llgc	$len,`2*$SIZE_T-1`($sp)
+	nill	$len,0x0f		# $len%16
+	lrvg	$s1,$tweak+0($sp)	# load the tweak in little-endian
+	lrvg	$s3,$tweak+8($sp)
+.Lxts_dec_2ndtweak:
+	lghi	%r1,0x87
+	srag	%r0,$s3,63		# broadcast upper bit
+	ngr	%r1,%r0			# rem
+	algr	$s1,$s1
+	alcgr	$s3,$s3
+	xgr	$s1,%r1
+	lrvgr	$s1,$s1			# flip byte order
+	lrvgr	$s3,$s3
+	srlg	$s0,$s1,32		# smash the tweak to 4x32-bits
+	stg	$s1,$tweak-16+0($sp)	# save the 2nd tweak
+	llgfr	$s1,$s1
+	srlg	$s2,$s3,32
+	stg	$s3,$tweak-16+8($sp)
+	llgfr	$s3,$s3
+
+	x	$s0,0($inp)		# tweak_the_2nd^=*(inp)
+	x	$s1,4($inp)
+	x	$s2,8($inp)
+	x	$s3,12($inp)
+	stm${g}	%r2,%r3,2*$SIZE_T($sp)
+	la	$key,0($key1)
+	bras	$ra,_s390x_AES_decrypt
+	lm${g}	%r2,%r5,2*$SIZE_T($sp)
+	x	$s0,$tweak-16+0($sp)	# ^=tweak_the_2nd
+	x	$s1,$tweak-16+4($sp)
+	x	$s2,$tweak-16+8($sp)
+	x	$s3,$tweak-16+12($sp)
+	st	$s0,0($out,$inp)
+	st	$s1,4($out,$inp)
+	st	$s2,8($out,$inp)
+	st	$s3,12($out,$inp)
+
+	la	$i3,0($out,$inp)	# put aside real $out
+.Lxts_dec_steal:
+	llgc	%r0,16($inp)
+	llgc	%r1,0($out,$inp)
+	stc	%r0,0($out,$inp)
+	stc	%r1,16($out,$inp)
+	la	$inp,1($inp)
+	brct	$len,.Lxts_dec_steal
+	la	$out,0($i3)		# restore real $out
+
+	lm	$s0,$s3,$tweak($sp)	# load the 1st tweak
+	x	$s0,0($out)		# tweak^=*(inp)|stolen cipher-text
+	x	$s1,4($out)
+	x	$s2,8($out)
+	x	$s3,12($out)
+	st${g}	$out,4*$SIZE_T($sp)
+	la	$key,0($key1)
+	bras	$ra,_s390x_AES_decrypt
+	l${g}	$out,4*$SIZE_T($sp)
+	x	$s0,$tweak+0($sp)	# ^=tweak
+	x	$s1,$tweak+4($sp)
+	x	$s2,$tweak+8($sp)
+	x	$s3,$tweak+12($sp)
+	st	$s0,0($out)
+	st	$s1,4($out)
+	st	$s2,8($out)
+	st	$s3,12($out)
+	stg	$sp,$tweak-16+0($sp)	# wipe 2nd tweak
+	stg	$sp,$tweak-16+8($sp)
+.Lxts_dec_done:
+	stg	$sp,$tweak+0($sp)	# wipe tweak
+	stg	$sp,$twesk+8($sp)
+	lm${g}	%r6,$ra,6*$SIZE_T($sp)
+	br	$ra
+.size	AES_xts_decrypt,.-AES_xts_decrypt
+___
+}
 $code.=<<___;
 .string	"AES for s390x, CRYPTOGAMS by <appro\@openssl.org>"
+.comm	OPENSSL_s390xcap_P,16,8
 ___
 
 $code =~ s/\`([^\`]*)\`/eval $1/gem;
 print $code;
+close STDOUT;	# force flush
diff --git a/lib/libcrypto/aes/asm/aes-x86_64.pl b/lib/libcrypto/aes/asm/aes-x86_64.pl
index 34cbb5d8442..f75e90ba87d 100755
--- a/lib/libcrypto/aes/asm/aes-x86_64.pl
+++ b/lib/libcrypto/aes/asm/aes-x86_64.pl
@@ -1284,13 +1284,13 @@ $code.=<<___;
 ___
 }
 
-# int private_AES_set_encrypt_key(const unsigned char *userKey, const int bits,
+# int AES_set_encrypt_key(const unsigned char *userKey, const int bits,
 #                        AES_KEY *key)
 $code.=<<___;
-.globl	private_AES_set_encrypt_key
-.type	private_AES_set_encrypt_key,\@function,3
+.globl	AES_set_encrypt_key
+.type	AES_set_encrypt_key,\@function,3
 .align	16
-private_AES_set_encrypt_key:
+AES_set_encrypt_key:
 	push	%rbx
 	push	%rbp
 	push	%r12			# redundant, but allows to share 
@@ -1311,7 +1311,7 @@ private_AES_set_encrypt_key:
 	add	\$56,%rsp
 .Lenc_key_epilogue:
 	ret
-.size	private_AES_set_encrypt_key,.-private_AES_set_encrypt_key
+.size	AES_set_encrypt_key,.-AES_set_encrypt_key
 
 .type	_x86_64_AES_set_encrypt_key,\@abi-omnipotent
 .align	16
@@ -1554,13 +1554,13 @@ $code.=<<___;
 ___
 }
 
-# int private_AES_set_decrypt_key(const unsigned char *userKey, const int bits,
+# int AES_set_decrypt_key(const unsigned char *userKey, const int bits,
 #                        AES_KEY *key)
 $code.=<<___;
-.globl	private_AES_set_decrypt_key
-.type	private_AES_set_decrypt_key,\@function,3
+.globl	AES_set_decrypt_key
+.type	AES_set_decrypt_key,\@function,3
 .align	16
-private_AES_set_decrypt_key:
+AES_set_decrypt_key:
 	push	%rbx
 	push	%rbp
 	push	%r12
@@ -1629,7 +1629,7 @@ $code.=<<___;
 	add	\$56,%rsp
 .Ldec_key_epilogue:
 	ret
-.size	private_AES_set_decrypt_key,.-private_AES_set_decrypt_key
+.size	AES_set_decrypt_key,.-AES_set_decrypt_key
 ___
 
 # void AES_cbc_encrypt (const void char *inp, unsigned char *out,
@@ -2776,13 +2776,13 @@ cbc_se_handler:
 	.rva	.LSEH_end_AES_decrypt
 	.rva	.LSEH_info_AES_decrypt
 
-	.rva	.LSEH_begin_private_AES_set_encrypt_key
-	.rva	.LSEH_end_private_AES_set_encrypt_key
-	.rva	.LSEH_info_private_AES_set_encrypt_key
+	.rva	.LSEH_begin_AES_set_encrypt_key
+	.rva	.LSEH_end_AES_set_encrypt_key
+	.rva	.LSEH_info_AES_set_encrypt_key
 
-	.rva	.LSEH_begin_private_AES_set_decrypt_key
-	.rva	.LSEH_end_private_AES_set_decrypt_key
-	.rva	.LSEH_info_private_AES_set_decrypt_key
+	.rva	.LSEH_begin_AES_set_decrypt_key
+	.rva	.LSEH_end_AES_set_decrypt_key
+	.rva	.LSEH_info_AES_set_decrypt_key
 
 	.rva	.LSEH_begin_AES_cbc_encrypt
 	.rva	.LSEH_end_AES_cbc_encrypt
@@ -2798,11 +2798,11 @@ cbc_se_handler:
 	.byte	9,0,0,0
 	.rva	block_se_handler
 	.rva	.Ldec_prologue,.Ldec_epilogue	# HandlerData[]
-.LSEH_info_private_AES_set_encrypt_key:
+.LSEH_info_AES_set_encrypt_key:
 	.byte	9,0,0,0
 	.rva	key_se_handler
 	.rva	.Lenc_key_prologue,.Lenc_key_epilogue	# HandlerData[]
-.LSEH_info_private_AES_set_decrypt_key:
+.LSEH_info_AES_set_decrypt_key:
 	.byte	9,0,0,0
 	.rva	key_se_handler
 	.rva	.Ldec_key_prologue,.Ldec_key_epilogue	# HandlerData[]