# Implement AES_set_[en|de]crypt_key. Key schedule setup is avoided
# for 128-bit keys, if hardware support is detected.
-$t1="%r0";
-$t2="%r1";
-$t3="%r2"; $inp="%r2";
-$out="%r3"; $mask="%r3"; $bits="%r3";
+# Januray 2009.
+#
+# Add support for hardware AES192/256 and reschedule instructions to
+# minimize/avoid Address Generation Interlock hazard and to favour
+# dual-issue z10 pipeline. This gave ~25% improvement on z10 and
+# almost 50% on z9. The gain is smaller on z10, because being dual-
+# issue z10 makes it improssible to eliminate the interlock condition:
+# critial path is not long enough. Yet it spends ~24 cycles per byte
+# processed with 128-bit key.
+#
+# 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 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"; $mask="%r0";
+$t1="%r1";
+$t2="%r2"; $inp="%r2";
+$t3="%r3"; $out="%r3"; $bits="%r3";
$key="%r4";
$i1="%r5";
$i2="%r6";
$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; }
.text
.type AES_Te,\@object
-.align 128
+.align 256
AES_Te:
___
&_data_word(
.long 0x01000000, 0x02000000, 0x04000000, 0x08000000
.long 0x10000000, 0x20000000, 0x40000000, 0x80000000
.long 0x1B000000, 0x36000000, 0, 0, 0, 0, 0, 0
+.align 256
.size AES_Te,.-AES_Te
# void AES_encrypt(const unsigned char *inp, unsigned char *out,
.globl AES_encrypt
.type AES_encrypt,\@function
AES_encrypt:
- stg $ra,112($sp)
- lghi %r0,10
- c %r0,240($key)
- jne .Lesoft
- lghi %r0,0 # query capability vector
- la %r1,16($sp)
- .long 0xb92e0042 # km %r4,%r2
- lg %r0,16($sp)
- tmhl %r0,`0x8000>>2`
- jz .Lesoft128
- lghi %r0,`0x00|0x12` # encrypt AES-128
+___
+$code.=<<___ if (!$softonly);
+ l %r0,240($key)
+ lhi %r1,16
+ clr %r0,%r1
+ jl .Lesoft
+
la %r1,0($key)
#la %r2,0($inp)
la %r4,0($out)
lghi %r3,16 # single block length
.long 0xb92e0042 # km %r4,%r2
- bcr 8,%r14 # return if done
- la $out,0(%r4) # restore arguments
- la $key,0(%r1)
-.Lesoft128:
- lghi %r0,0
- c %r0,236($key)
- je .Lesoft
- stmg $inp,$key,16($sp)
- la $inp,0($key)
- lghi $bits,128
- bras $ra,.Lekey_internal # postponed key schedule setup
- lmg $inp,$key,16($sp)
+ brc 1,.-4 # can this happen?
+ br %r14
+.align 64
.Lesoft:
- stmg %r3,%r13,24($sp)
-
- bras $tbl,1f
-1: aghi $tbl,AES_Te-.
+___
+$code.=<<___;
+ stm${g} %r3,$ra,3*$SIZE_T($sp)
llgf $s0,0($inp)
llgf $s1,4($inp)
llgf $s2,8($inp)
llgf $s3,12($inp)
- llill $mask,`0xff<<3`
+ 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 $ra
.size AES_encrypt,.-AES_encrypt
.type _s390x_AES_encrypt,\@function
.align 16
_s390x_AES_encrypt:
+ st${g} $ra,15*$SIZE_T($sp)
x $s0,0($key)
x $s1,4($key)
x $s2,8($key)
x $s3,12($key)
l $rounds,240($key)
+ llill $mask,`0xff<<3`
aghi $rounds,-1
-
+ j .Lenc_loop
+.align 16
.Lenc_loop:
- sllg $i1,$s0,`0+3`
- srlg $i2,$s0,`8-3`
- srlg $i3,$s0,`16-3`
+ sllg $t1,$s0,`0+3`
+ srlg $t2,$s0,`8-3`
+ srlg $t3,$s0,`16-3`
srl $s0,`24-3`
nr $s0,$mask
- ngr $i1,$mask
- nr $i2,$mask
- nr $i3,$mask
- l $s0,0($s0,$tbl) # Te0[s0>>24]
- l $t1,1($i1,$tbl) # Te3[s0>>0]
- l $t2,2($i2,$tbl) # Te2[s0>>8]
- l $t3,3($i3,$tbl) # Te1[s0>>16]
+ ngr $t1,$mask
+ nr $t2,$mask
+ nr $t3,$mask
srlg $i1,$s1,`16-3` # i0
sllg $i2,$s1,`0+3`
nr $s1,$mask
ngr $i2,$mask
nr $i3,$mask
+
+ l $s0,0($s0,$tbl) # Te0[s0>>24]
+ l $t1,1($t1,$tbl) # Te3[s0>>0]
+ l $t2,2($t2,$tbl) # Te2[s0>>8]
+ l $t3,3($t3,$tbl) # Te1[s0>>16]
+
x $s0,3($i1,$tbl) # Te1[s1>>16]
l $s1,0($s1,$tbl) # Te0[s1>>24]
x $t2,1($i2,$tbl) # Te3[s1>>0]
x $t3,2($i3,$tbl) # Te2[s1>>8]
- xr $s1,$t1
srlg $i1,$s2,`8-3` # i0
srlg $i2,$s2,`16-3` # i1
- sllg $i3,$s2,`0+3`
- srl $s2,`24-3`
nr $i1,$mask
nr $i2,$mask
+ sllg $i3,$s2,`0+3`
+ srl $s2,`24-3`
nr $s2,$mask
ngr $i3,$mask
+
+ xr $s1,$t1
+ srlg $ra,$s3,`8-3` # i1
+ sllg $t1,$s3,`0+3` # i0
+ nr $ra,$mask
+ la $key,16($key)
+ ngr $t1,$mask
+
x $s0,2($i1,$tbl) # Te2[s2>>8]
x $s1,3($i2,$tbl) # Te1[s2>>16]
l $s2,0($s2,$tbl) # Te0[s2>>24]
x $t3,1($i3,$tbl) # Te3[s2>>0]
- xr $s2,$t2
- sllg $i1,$s3,`0+3` # i0
- srlg $i2,$s3,`8-3` # i1
srlg $i3,$s3,`16-3` # i2
+ xr $s2,$t2
srl $s3,`24-3`
- ngr $i1,$mask
- nr $i2,$mask
nr $i3,$mask
nr $s3,$mask
- x $s0,1($i1,$tbl) # Te3[s3>>0]
- x $s1,2($i2,$tbl) # Te2[s3>>8]
- x $s2,3($i3,$tbl) # Te1[s3>>16]
- l $s3,0($s3,$tbl) # Te0[s3>>24]
- xr $s3,$t3
- la $key,16($key)
x $s0,0($key)
x $s1,4($key)
x $s2,8($key)
- x $s3,12($key)
+ x $t3,12($key)
+
+ x $s0,1($t1,$tbl) # Te3[s3>>0]
+ x $s1,2($ra,$tbl) # Te2[s3>>8]
+ x $s2,3($i3,$tbl) # Te1[s3>>16]
+ l $s3,0($s3,$tbl) # Te0[s3>>24]
+ xr $s3,$t3
brct $rounds,.Lenc_loop
+ .align 16
- sllg $i1,$s0,`0+3`
- srlg $i2,$s0,`8-3`
- srlg $i3,$s0,`16-3`
+ sllg $t1,$s0,`0+3`
+ srlg $t2,$s0,`8-3`
+ ngr $t1,$mask
+ srlg $t3,$s0,`16-3`
srl $s0,`24-3`
nr $s0,$mask
- ngr $i1,$mask
- nr $i2,$mask
- nr $i3,$mask
- llgc $s0,2($s0,$tbl) # Te4[s0>>24]
- llgc $t1,2($i1,$tbl) # Te4[s0>>0]
- llgc $t2,2($i2,$tbl) # Te4[s0>>8]
- llgc $t3,2($i3,$tbl) # Te4[s0>>16]
- sll $s0,24
- sll $t2,8
- sll $t3,16
+ nr $t2,$mask
+ nr $t3,$mask
srlg $i1,$s1,`16-3` # i0
sllg $i2,$s1,`0+3`
+ ngr $i2,$mask
srlg $i3,$s1,`8-3`
srl $s1,`24-3`
nr $i1,$mask
nr $s1,$mask
- ngr $i2,$mask
nr $i3,$mask
+
+ llgc $s0,2($s0,$tbl) # Te4[s0>>24]
+ llgc $t1,2($t1,$tbl) # Te4[s0>>0]
+ sll $s0,24
+ llgc $t2,2($t2,$tbl) # Te4[s0>>8]
+ llgc $t3,2($t3,$tbl) # Te4[s0>>16]
+ sll $t2,8
+ sll $t3,16
+
llgc $i1,2($i1,$tbl) # Te4[s1>>16]
llgc $s1,2($s1,$tbl) # Te4[s1>>24]
llgc $i2,2($i2,$tbl) # Te4[s1>>0]
srlg $i1,$s2,`8-3` # i0
srlg $i2,$s2,`16-3` # i1
- sllg $i3,$s2,`0+3`
- srl $s2,`24-3`
nr $i1,$mask
nr $i2,$mask
- nr $s2,$mask
+ sllg $i3,$s2,`0+3`
+ srl $s2,`24-3`
ngr $i3,$mask
+ nr $s2,$mask
+
+ sllg $t1,$s3,`0+3` # i0
+ srlg $ra,$s3,`8-3` # i1
+ ngr $t1,$mask
+
llgc $i1,2($i1,$tbl) # Te4[s2>>8]
llgc $i2,2($i2,$tbl) # Te4[s2>>16]
+ sll $i1,8
llgc $s2,2($s2,$tbl) # Te4[s2>>24]
llgc $i3,2($i3,$tbl) # Te4[s2>>0]
- sll $i1,8
sll $i2,16
+ nr $ra,$mask
sll $s2,24
or $s0,$i1
or $s1,$i2
or $s2,$t2
or $t3,$i3
- sllg $i1,$s3,`0+3` # i0
- srlg $i2,$s3,`8-3` # i1
srlg $i3,$s3,`16-3` # i2
srl $s3,`24-3`
- ngr $i1,$mask
- nr $i2,$mask
nr $i3,$mask
nr $s3,$mask
- llgc $i1,2($i1,$tbl) # Te4[s3>>0]
- llgc $i2,2($i2,$tbl) # Te4[s3>>8]
+
+ l $t0,16($key)
+ l $t2,20($key)
+
+ llgc $i1,2($t1,$tbl) # Te4[s3>>0]
+ llgc $i2,2($ra,$tbl) # Te4[s3>>8]
llgc $i3,2($i3,$tbl) # Te4[s3>>16]
llgc $s3,2($s3,$tbl) # Te4[s3>>24]
sll $i2,8
or $s2,$i3
or $s3,$t3
- x $s0,16($key)
- x $s1,20($key)
+ l${g} $ra,15*$SIZE_T($sp)
+ xr $s0,$t0
+ xr $s1,$t2
x $s2,24($key)
x $s3,28($key)
$code.=<<___;
.type AES_Td,\@object
-.align 128
+.align 256
AES_Td:
___
&_data_word(
.globl AES_decrypt
.type AES_decrypt,\@function
AES_decrypt:
- stg $ra,112($sp)
- lghi %r0,10
- c %r0,240($key)
- jne .Ldsoft
- lghi %r0,0 # query capability vector
- la %r1,16($sp)
- .long 0xb92e0042 # km %r4,%r2
- lg %r0,16($sp)
- tmhl %r0,`0x8000>>2`
- jz .Ldsoft128
- lghi %r0,`0x80|0x12` # decrypt AES-128
- la %r1,160($key)
+___
+$code.=<<___ if (!$softonly);
+ l %r0,240($key)
+ lhi %r1,16
+ clr %r0,%r1
+ jl .Ldsoft
+
+ la %r1,0($key)
#la %r2,0($inp)
la %r4,0($out)
lghi %r3,16 # single block length
.long 0xb92e0042 # km %r4,%r2
- bcr 8,%r14 # return if done
- la $out,0(%r4) # restore arguments
- lghi $key,-160
- la $key,0($key,%r1)
-.Ldsoft128:
- lghi %r0,0
- c %r0,236($key)
- je .Ldsoft
- stmg $inp,$key,16($sp)
- la $inp,160($key)
- lghi $bits,128
- bras $ra,.Ldkey_internal # postponed key schedule setup
- lmg $inp,$key,16($sp)
+ brc 1,.-4 # can this happen?
+ br %r14
+.align 64
.Ldsoft:
- stmg %r3,%r13,24($sp)
-
- bras $tbl,1f
-1: aghi $tbl,AES_Td-.
+___
+$code.=<<___;
+ stm${g} %r3,$ra,3*$SIZE_T($sp)
llgf $s0,0($inp)
llgf $s1,4($inp)
llgf $s2,8($inp)
llgf $s3,12($inp)
- llill $mask,`0xff<<3`
+ 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 $ra
.size AES_decrypt,.-AES_decrypt
.type _s390x_AES_decrypt,\@function
.align 16
_s390x_AES_decrypt:
+ st${g} $ra,15*$SIZE_T($sp)
x $s0,0($key)
x $s1,4($key)
x $s2,8($key)
x $s3,12($key)
l $rounds,240($key)
+ llill $mask,`0xff<<3`
aghi $rounds,-1
-
+ j .Ldec_loop
+.align 16
.Ldec_loop:
- srlg $i1,$s0,`16-3`
- srlg $i2,$s0,`8-3`
- sllg $i3,$s0,`0+3`
+ srlg $t1,$s0,`16-3`
+ srlg $t2,$s0,`8-3`
+ sllg $t3,$s0,`0+3`
srl $s0,`24-3`
nr $s0,$mask
- nr $i1,$mask
- nr $i2,$mask
- ngr $i3,$mask
- l $s0,0($s0,$tbl) # Td0[s0>>24]
- l $t1,3($i1,$tbl) # Td1[s0>>16]
- l $t2,2($i2,$tbl) # Td2[s0>>8]
- l $t3,1($i3,$tbl) # Td3[s0>>0]
+ nr $t1,$mask
+ nr $t2,$mask
+ ngr $t3,$mask
sllg $i1,$s1,`0+3` # i0
srlg $i2,$s1,`16-3`
nr $s1,$mask
nr $i2,$mask
nr $i3,$mask
+
+ l $s0,0($s0,$tbl) # Td0[s0>>24]
+ l $t1,3($t1,$tbl) # Td1[s0>>16]
+ l $t2,2($t2,$tbl) # Td2[s0>>8]
+ l $t3,1($t3,$tbl) # Td3[s0>>0]
+
x $s0,1($i1,$tbl) # Td3[s1>>0]
l $s1,0($s1,$tbl) # Td0[s1>>24]
x $t2,3($i2,$tbl) # Td1[s1>>16]
x $t3,2($i3,$tbl) # Td2[s1>>8]
- xr $s1,$t1
srlg $i1,$s2,`8-3` # i0
sllg $i2,$s2,`0+3` # i1
ngr $i2,$mask
nr $s2,$mask
nr $i3,$mask
+
+ xr $s1,$t1
+ srlg $ra,$s3,`8-3` # i1
+ srlg $t1,$s3,`16-3` # i0
+ nr $ra,$mask
+ la $key,16($key)
+ nr $t1,$mask
+
x $s0,2($i1,$tbl) # Td2[s2>>8]
x $s1,1($i2,$tbl) # Td3[s2>>0]
l $s2,0($s2,$tbl) # Td0[s2>>24]
x $t3,3($i3,$tbl) # Td1[s2>>16]
- xr $s2,$t2
- srlg $i1,$s3,`16-3` # i0
- srlg $i2,$s3,`8-3` # i1
sllg $i3,$s3,`0+3` # i2
srl $s3,`24-3`
- nr $i1,$mask
- nr $i2,$mask
ngr $i3,$mask
nr $s3,$mask
- x $s0,3($i1,$tbl) # Td1[s3>>16]
- x $s1,2($i2,$tbl) # Td2[s3>>8]
- x $s2,1($i3,$tbl) # Td3[s3>>0]
- l $s3,0($s3,$tbl) # Td0[s3>>24]
- xr $s3,$t3
- la $key,16($key)
+ xr $s2,$t2
x $s0,0($key)
x $s1,4($key)
x $s2,8($key)
- x $s3,12($key)
+ x $t3,12($key)
+
+ x $s0,3($t1,$tbl) # Td1[s3>>16]
+ x $s1,2($ra,$tbl) # Td2[s3>>8]
+ x $s2,1($i3,$tbl) # Td3[s3>>0]
+ l $s3,0($s3,$tbl) # Td0[s3>>24]
+ xr $s3,$t3
brct $rounds,.Ldec_loop
+ .align 16
l $t1,`2048+0`($tbl) # prefetch Td4
- l $t2,`2048+32`($tbl)
- l $t3,`2048+64`($tbl)
- l $i1,`2048+96`($tbl)
- l $i2,`2048+128`($tbl)
- l $i3,`2048+160`($tbl)
- l $t1,`2048+192`($tbl)
- l $t2,`2048+224`($tbl)
+ l $t2,`2048+64`($tbl)
+ l $t3,`2048+128`($tbl)
+ l $i1,`2048+192`($tbl)
llill $mask,0xff
srlg $i3,$s0,24 # i0
- srlg $i1,$s0,16
- srlg $i2,$s0,8
+ srlg $t1,$s0,16
+ srlg $t2,$s0,8
nr $s0,$mask # i3
- nr $i1,$mask
+ nr $t1,$mask
+
+ srlg $i1,$s1,24
+ nr $t2,$mask
+ srlg $i2,$s1,16
+ srlg $ra,$s1,8
+ nr $s1,$mask # i0
nr $i2,$mask
+ nr $ra,$mask
+
llgc $i3,2048($i3,$tbl) # Td4[s0>>24]
- llgc $t1,2048($i1,$tbl) # Td4[s0>>16]
- llgc $t2,2048($i2,$tbl) # Td4[s0>>8]
+ llgc $t1,2048($t1,$tbl) # Td4[s0>>16]
+ llgc $t2,2048($t2,$tbl) # Td4[s0>>8]
+ sll $t1,16
llgc $t3,2048($s0,$tbl) # Td4[s0>>0]
sllg $s0,$i3,24
- sll $t1,16
sll $t2,8
- srlg $i1,$s1,24
- srlg $i2,$s1,16
- srlg $i3,$s1,8
- nr $s1,$mask # i0
- nr $i2,$mask
- nr $i3,$mask
llgc $s1,2048($s1,$tbl) # Td4[s1>>0]
llgc $i1,2048($i1,$tbl) # Td4[s1>>24]
llgc $i2,2048($i2,$tbl) # Td4[s1>>16]
- llgc $i3,2048($i3,$tbl) # Td4[s1>>8]
sll $i1,24
+ llgc $i3,2048($ra,$tbl) # Td4[s1>>8]
sll $i2,16
sll $i3,8
or $s0,$s1
llgc $i3,2048($i3,$tbl) # Td4[s2>>16]
sll $i1,8
sll $i2,24
- sll $i3,16
or $s0,$i1
- or $s1,$t1
+ sll $i3,16
or $t2,$i2
or $t3,$i3
nr $s3,$mask # i2
nr $i1,$mask
nr $i2,$mask
+
+ l${g} $ra,15*$SIZE_T($sp)
+ or $s1,$t1
+ l $t0,16($key)
+ l $t1,20($key)
+
llgc $i1,2048($i1,$tbl) # Td4[s3>>16]
llgc $i2,2048($i2,$tbl) # Td4[s3>>8]
+ sll $i1,16
llgc $s2,2048($s3,$tbl) # Td4[s3>>0]
llgc $s3,2048($i3,$tbl) # Td4[s3>>24]
- sll $i1,16
sll $i2,8
sll $s3,24
or $s0,$i1
or $s2,$t2
or $s3,$t3
- x $s0,16($key)
- x $s1,20($key)
+ xr $s0,$t0
+ xr $s1,$t1
x $s2,24($key)
x $s3,28($key)
br $ra
.size _s390x_AES_decrypt,.-_s390x_AES_decrypt
+___
+$code.=<<___;
# void AES_set_encrypt_key(const unsigned char *in, int bits,
# AES_KEY *key) {
-.globl AES_set_encrypt_key
-.type AES_set_encrypt_key,\@function
+.globl private_AES_set_encrypt_key
+.type private_AES_set_encrypt_key,\@function
.align 16
-AES_set_encrypt_key:
- lghi $t1,0
- clgr $inp,$t1
+private_AES_set_encrypt_key:
+_s390x_AES_set_encrypt_key:
+ lghi $t0,0
+ cl${g}r $inp,$t0
je .Lminus1
- clgr $key,$t1
+ cl${g}r $key,$t0
je .Lminus1
- lghi $t1,128
- clr $bits,$t1
- je .Lproceed128
- lghi $t1,192
- clr $bits,$t1
- je .Lekey_internal
- lghi $t1,256
- clr $bits,$t1
- je .Lekey_internal
+ lghi $t0,128
+ clr $bits,$t0
+ je .Lproceed
+ lghi $t0,192
+ clr $bits,$t0
+ je .Lproceed
+ lghi $t0,256
+ clr $bits,$t0
+ je .Lproceed
lghi %r2,-2
br %r14
-.align 4
-.Lproceed128:
+.align 16
+.Lproceed:
+___
+$code.=<<___ if (!$softonly);
+ # convert bits to km code, [128,192,256]->[18,19,20]
+ lhi %r5,-128
+ lhi %r0,18
+ ar %r5,$bits
+ srl %r5,6
+ ar %r5,%r0
+
+ larl %r1,OPENSSL_s390xcap_P
+ lg %r0,0(%r1)
+ tmhl %r0,0x4000 # check for message-security assist
+ jz .Lekey_internal
+
lghi %r0,0 # query capability vector
la %r1,16($sp)
- .long 0xb92e0042 # km %r4,%r2
- lg %r0,16($sp)
- tmhl %r0,`0x8000>>2`
+ .long 0xb92f0042 # kmc %r4,%r2
+
+ llihh %r1,0x8000
+ srlg %r1,%r1,0(%r5)
+ ng %r1,16($sp)
jz .Lekey_internal
- l $t1,0($inp) # just copy 128 bits...
- l $t2,4($inp)
- l $bits,8($inp)
- l $inp,12($inp)
- st $t1,0($key)
- st $t2,4($key)
- st $bits,8($key)
- st $inp,12($key)
- lghi $t1,10
- st $t1,236($key) # ... postpone key setup
- st $t1,240($key)
+ lmg %r0,%r1,0($inp) # just copy 128 bits...
+ stmg %r0,%r1,0($key)
+ lhi %r0,192
+ cr $bits,%r0
+ jl 1f
+ lg %r1,16($inp)
+ stg %r1,16($key)
+ je 1f
+ lg %r1,24($inp)
+ stg %r1,24($key)
+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
-
+___
+$code.=<<___;
.align 16
.Lekey_internal:
- stmg %r6,%r13,48($sp) # all volatile regs, but $ra!
+ stm${g} %r4,%r13,4*$SIZE_T($sp) # all non-volatile regs and $key
- bras $tbl,1f
-1: aghi $tbl,AES_Te+2048-.
+ larl $tbl,AES_Te+2048
llgf $s0,0($inp)
llgf $s1,4($inp)
st $s1,4($key)
st $s2,8($key)
st $s3,12($key)
- lghi $t1,128
- cr $bits,$t1
+ lghi $t0,128
+ cr $bits,$t0
jne .Lnot128
llill $mask,0xff
lghi $t3,0 # i=0
lghi $rounds,10
- st $t3,236($key) # mark as set up
st $rounds,240($key)
-.align 8
-.L128_loop:
llgfr $t2,$s3 # temp=rk[3]
srlg $i1,$s3,8
srlg $i2,$s3,16
nr $t2,$mask
nr $i1,$mask
nr $i2,$mask
+
+.align 16
+.L128_loop:
la $t2,0($t2,$tbl)
la $i1,0($i1,$tbl)
la $i2,0($i2,$tbl)
xr $s1,$s0 # rk[5]=rk[1]^rk[4]
xr $s2,$s1 # rk[6]=rk[2]^rk[5]
xr $s3,$s2 # rk[7]=rk[3]^rk[6]
+
+ llgfr $t2,$s3 # temp=rk[3]
+ srlg $i1,$s3,8
+ srlg $i2,$s3,16
+ nr $t2,$mask
+ nr $i1,$mask
+ srlg $i3,$s3,24
+ nr $i2,$mask
+
st $s0,16($key)
st $s1,20($key)
st $s2,24($key)
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 4
+.align 16
.Lnot128:
- llgf $t1,16($inp)
- llgf $t2,20($inp)
- st $t1,16($key)
- st $t2,20($key)
- lghi $t1,192
- cr $bits,$t1
+ llgf $t0,16($inp)
+ llgf $t1,20($inp)
+ st $t0,16($key)
+ st $t1,20($key)
+ lghi $t0,192
+ cr $bits,$t0
jne .Lnot192
llill $mask,0xff
st $rounds,240($key)
lghi $rounds,8
-.align 8
-.L192_loop:
- srlg $i1,$t2,8
- srlg $i2,$t2,16
- srlg $i3,$t2,24
- nr $t2,$mask
+ srlg $i1,$t1,8
+ srlg $i2,$t1,16
+ srlg $i3,$t1,24
+ nr $t1,$mask
nr $i1,$mask
nr $i2,$mask
- la $t2,0($t2,$tbl)
+
+.align 16
+.L192_loop:
+ la $t1,0($t1,$tbl)
la $i1,0($i1,$tbl)
la $i2,0($i2,$tbl)
la $i3,0($i3,$tbl)
- icm $t2,2,0($t2) # Te4[rk[5]>>0]<<8
- icm $t2,4,0($i1) # Te4[rk[5]>>8]<<16
- icm $t2,8,0($i2) # Te4[rk[5]>>16]<<24
- icm $t2,1,0($i3) # Te4[rk[5]>>24]
- x $t2,256($t3,$tbl) # rcon[i]
- xr $s0,$t2 # rk[6]=rk[0]^...
+ icm $t1,2,0($t1) # Te4[rk[5]>>0]<<8
+ icm $t1,4,0($i1) # Te4[rk[5]>>8]<<16
+ icm $t1,8,0($i2) # Te4[rk[5]>>16]<<24
+ icm $t1,1,0($i3) # Te4[rk[5]>>24]
+ x $t1,256($t3,$tbl) # rcon[i]
+ xr $s0,$t1 # rk[6]=rk[0]^...
xr $s1,$s0 # rk[7]=rk[1]^rk[6]
xr $s2,$s1 # rk[8]=rk[2]^rk[7]
xr $s3,$s2 # rk[9]=rk[3]^rk[8]
+
st $s0,24($key)
st $s1,28($key)
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 4
+
+.align 16
.L192_continue:
- lgr $t2,$s3
- x $t2,16($key) # rk[10]=rk[4]^rk[9]
- st $t2,40($key)
- x $t2,20($key) # rk[11]=rk[5]^rk[10]
- st $t2,44($key)
+ lgr $t1,$s3
+ x $t1,16($key) # rk[10]=rk[4]^rk[9]
+ st $t1,40($key)
+ x $t1,20($key) # rk[11]=rk[5]^rk[10]
+ st $t1,44($key)
+
+ srlg $i1,$t1,8
+ srlg $i2,$t1,16
+ srlg $i3,$t1,24
+ nr $t1,$mask
+ nr $i1,$mask
+ nr $i2,$mask
+
la $key,24($key) # key+=6
la $t3,4($t3) # i++
j .L192_loop
-.align 4
+.align 16
.Lnot192:
- llgf $t1,24($inp)
- llgf $t2,28($inp)
- st $t1,24($key)
- st $t2,28($key)
+ llgf $t0,24($inp)
+ llgf $t1,28($inp)
+ st $t0,24($key)
+ st $t1,28($key)
llill $mask,0xff
lghi $t3,0 # i=0
lghi $rounds,14
st $rounds,240($key)
lghi $rounds,7
-.align 8
-.L256_loop:
- srlg $i1,$t2,8
- srlg $i2,$t2,16
- srlg $i3,$t2,24
- nr $t2,$mask
+ srlg $i1,$t1,8
+ srlg $i2,$t1,16
+ srlg $i3,$t1,24
+ nr $t1,$mask
nr $i1,$mask
nr $i2,$mask
- la $t2,0($t2,$tbl)
+
+.align 16
+.L256_loop:
+ la $t1,0($t1,$tbl)
la $i1,0($i1,$tbl)
la $i2,0($i2,$tbl)
la $i3,0($i3,$tbl)
- icm $t2,2,0($t2) # Te4[rk[7]>>0]<<8
- icm $t2,4,0($i1) # Te4[rk[7]>>8]<<16
- icm $t2,8,0($i2) # Te4[rk[7]>>16]<<24
- icm $t2,1,0($i3) # Te4[rk[7]>>24]
- x $t2,256($t3,$tbl) # rcon[i]
- xr $s0,$t2 # rk[8]=rk[0]^...
+ icm $t1,2,0($t1) # Te4[rk[7]>>0]<<8
+ icm $t1,4,0($i1) # Te4[rk[7]>>8]<<16
+ icm $t1,8,0($i2) # Te4[rk[7]>>16]<<24
+ icm $t1,1,0($i3) # Te4[rk[7]>>24]
+ x $t1,256($t3,$tbl) # rcon[i]
+ xr $s0,$t1 # rk[8]=rk[0]^...
xr $s1,$s0 # rk[9]=rk[1]^rk[8]
xr $s2,$s1 # rk[10]=rk[2]^rk[9]
xr $s3,$s2 # rk[11]=rk[3]^rk[10]
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 4
+
+.align 16
.L256_continue:
- lgr $t2,$s3 # temp=rk[11]
+ lgr $t1,$s3 # temp=rk[11]
srlg $i1,$s3,8
srlg $i2,$s3,16
srlg $i3,$s3,24
- nr $t2,$mask
+ nr $t1,$mask
nr $i1,$mask
nr $i2,$mask
- la $t2,0($t2,$tbl)
+ la $t1,0($t1,$tbl)
la $i1,0($i1,$tbl)
la $i2,0($i2,$tbl)
la $i3,0($i3,$tbl)
- icm $t2,1,0($t2) # Te4[rk[11]>>0]
- icm $t2,2,0($i1) # Te4[rk[11]>>8]<<8
- icm $t2,4,0($i2) # Te4[rk[11]>>16]<<16
- icm $t2,8,0($i3) # Te4[rk[11]>>24]<<24
- x $t2,16($key) # rk[12]=rk[4]^...
- st $t2,48($key)
- x $t2,20($key) # rk[13]=rk[5]^rk[12]
- st $t2,52($key)
- x $t2,24($key) # rk[14]=rk[6]^rk[13]
- st $t2,56($key)
- x $t2,28($key) # rk[15]=rk[7]^rk[14]
- st $t2,60($key)
+ llgc $t1,0($t1) # Te4[rk[11]>>0]
+ icm $t1,2,0($i1) # Te4[rk[11]>>8]<<8
+ icm $t1,4,0($i2) # Te4[rk[11]>>16]<<16
+ icm $t1,8,0($i3) # Te4[rk[11]>>24]<<24
+ x $t1,16($key) # rk[12]=rk[4]^...
+ st $t1,48($key)
+ x $t1,20($key) # rk[13]=rk[5]^rk[12]
+ st $t1,52($key)
+ x $t1,24($key) # rk[14]=rk[6]^rk[13]
+ st $t1,56($key)
+ x $t1,28($key) # rk[15]=rk[7]^rk[14]
+ st $t1,60($key)
+
+ srlg $i1,$t1,8
+ srlg $i2,$t1,16
+ srlg $i3,$t1,24
+ nr $t1,$mask
+ nr $i1,$mask
+ nr $i2,$mask
la $key,32($key) # key+=8
la $t3,4($t3) # i++
j .L256_loop
-.align 4
+
.Lminus1:
lghi %r2,-1
- br %r14
-.size AES_set_encrypt_key,.-AES_set_encrypt_key
+ br $ra
+.size private_AES_set_encrypt_key,.-private_AES_set_encrypt_key
# void AES_set_decrypt_key(const unsigned char *in, int bits,
# AES_KEY *key) {
-.globl AES_set_decrypt_key
-.type AES_set_decrypt_key,\@function
+.globl private_AES_set_decrypt_key
+.type private_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 [other] volatile registers!
- bras $ra,AES_set_encrypt_key
- lg $key,32($sp)
- lg $ra,112($sp)
+private_AES_set_decrypt_key:
+ #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
-
- lghi $t1,10
- c $t1,240($key)
- jne .Lgo
- lghi $t1,0
- c $t1,236($key)
- je .Lgo
-
- l $t1,0($key) # just copy 128 bits otherwise
- l $t2,4($key)
- l $t3,8($key)
- l $bits,12($key)
- st $t1,160($key)
- st $t2,164($key)
- st $t3,168($key)
- st $bits,172($key)
- lghi %r2,0
+___
+$code.=<<___ if (!$softonly);
+ #l $t0,240($key)
+ lhi $t1,16
+ cr $t0,$t1
+ jl .Lgo
+ oill $t0,0x80 # set "decrypt" bit
+ st $t0,240($key)
br $ra
-
+___
+$code.=<<___;
.align 16
-.Ldkey_internal:
- stg $key,32($sp)
- stg $ra,40($sp)
- bras $ra,.Lekey_internal
- lg $key,32($sp)
- lg $ra,40($sp)
-
-.Lgo: llgf $rounds,240($key)
- lghi $i1,0
+.Lgo: lgr $rounds,$t0 #llgf $rounds,240($key)
+ la $i1,0($key)
sllg $i2,$rounds,4
+ la $i2,0($i2,$key)
srl $rounds,1
+ lghi $t1,-16
-.align 8
-.Linv: l $s0,0($i1,$key)
- l $s1,4($i1,$key)
- l $s2,8($i1,$key)
- l $s3,12($i1,$key)
- l $t1,0($i2,$key)
- l $t2,4($i2,$key)
- l $t3,8($i2,$key)
- l $i3,12($i2,$key)
- st $s0,0($i2,$key)
- st $s1,4($i2,$key)
- st $s2,8($i2,$key)
- st $s3,12($i2,$key)
- st $t1,0($i1,$key)
- st $t2,4($i1,$key)
- st $t3,8($i1,$key)
- st $i3,12($i1,$key)
- aghi $i1,16
- aghi $i2,-16
+.align 16
+.Linv: lmg $s0,$s1,0($i1)
+ lmg $s2,$s3,0($i2)
+ stmg $s0,$s1,0($i2)
+ stmg $s2,$s3,0($i1)
+ la $i1,16($i1)
+ la $i2,0($t1,$i2)
brct $rounds,.Linv
___
$mask80=$i1;
aghi $rounds,-1
sll $rounds,2 # (rounds-1)*4
llilh $mask80,0x8080
- oill $mask80,0x8080
llilh $mask1b,0x1b1b
- oill $mask1b,0x1b1b
llilh $maskfe,0xfefe
+ oill $mask80,0x8080
+ oill $mask1b,0x1b1b
oill $maskfe,0xfefe
-.align 8
+.align 16
.Lmix: l $s0,16($key) # tp1
lr $s1,$s0
ngr $s1,$mask80
xr $s1,$s0 # tp2^tp1
xr $s2,$s0 # tp4^tp1
rll $s0,$s0,24 # = ROTATE(tp1,8)
+ xr $s2,$s3 # ^=tp8
xr $s0,$s1 # ^=tp2^tp1
- xr $s0,$s2 # ^=tp4^tp1
- xr $s0,$s3 # ^= tp8[^(tp4^tp1)^(tp2^tp1)=tp4^tp2]
xr $s1,$s3 # tp2^tp1^tp8
+ xr $s0,$s2 # ^=tp4^tp1^tp8
rll $s1,$s1,8
- xr $s0,$s1 # ^= ROTATE(tp8^tp2^tp1,24)
- xr $s2,$s3 # tp4^tp1^tp8
rll $s2,$s2,16
- xr $s0,$s2 # ^= ROTATE(tp8^tp4^tp1,16)
+ xr $s0,$s1 # ^= ROTATE(tp8^tp2^tp1,24)
rll $s3,$s3,24
+ xr $s0,$s2 # ^= ROTATE(tp8^tp4^tp1,16)
xr $s0,$s3 # ^= ROTATE(tp8,8)
st $s0,16($key)
la $key,4($key)
brct $rounds,.Lmix
- lmg %r6,%r13,48($sp)# this 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
+.size private_AES_set_decrypt_key,.-private_AES_set_decrypt_key
+___
+
+########################################################################
+# void AES_cbc_encrypt(const unsigned char *in, unsigned char *out,
+# size_t length, const AES_KEY *key,
+# unsigned char *ivec, const int enc)
+{
+my $inp="%r2";
+my $out="%r4"; # length and out are swapped
+my $len="%r3";
+my $key="%r5";
+my $ivp="%r6";
+
+$code.=<<___;
+.globl AES_cbc_encrypt
+.type AES_cbc_encrypt,\@function
+.align 16
+AES_cbc_encrypt:
+ xgr %r3,%r4 # flip %r3 and %r4, out and len
+ xgr %r4,%r3
+ xgr %r3,%r4
+___
+$code.=<<___ if (!$softonly);
+ lhi %r0,16
+ cl %r0,240($key)
+ jh .Lcbc_software
+
+ lg %r0,0($ivp) # copy ivec
+ lg %r1,8($ivp)
+ stmg %r0,%r1,16($sp)
+ lmg %r0,%r1,0($key) # copy key, cover 256 bit
+ stmg %r0,%r1,32($sp)
+ lmg %r0,%r1,16($key)
+ stmg %r0,%r1,48($sp)
+ l %r0,240($key) # load kmc code
+ lghi $key,15 # res=len%16, len-=res;
+ ngr $key,$len
+ sl${g}r $len,$key
+ la %r1,16($sp) # parameter block - ivec || key
+ jz .Lkmc_truncated
+ .long 0xb92f0042 # kmc %r4,%r2
+ brc 1,.-4 # pay attention to "partial completion"
+ ltr $key,$key
+ jnz .Lkmc_truncated
+.Lkmc_done:
+ lmg %r0,%r1,16($sp) # copy ivec to caller
+ stg %r0,0($ivp)
+ stg %r1,8($ivp)
+ br $ra
+.align 16
+.Lkmc_truncated:
+ ahi $key,-1 # it's the way it's encoded in mvc
+ tmll %r0,0x80
+ jnz .Lkmc_truncated_dec
+ lghi %r1,0
+ stg %r1,16*$SIZE_T($sp)
+ stg %r1,16*$SIZE_T+8($sp)
+ bras %r1,1f
+ mvc 16*$SIZE_T(1,$sp),0($inp)
+1: ex $key,0(%r1)
+ la %r1,16($sp) # restore parameter block
+ la $inp,16*$SIZE_T($sp)
+ lghi $len,16
+ .long 0xb92f0042 # kmc %r4,%r2
+ j .Lkmc_done
+.align 16
+.Lkmc_truncated_dec:
+ st${g} $out,4*$SIZE_T($sp)
+ la $out,16*$SIZE_T($sp)
+ lghi $len,16
+ .long 0xb92f0042 # kmc %r4,%r2
+ l${g} $out,4*$SIZE_T($sp)
+ bras %r1,2f
+ mvc 0(1,$out),16*$SIZE_T($sp)
+2: ex $key,0(%r1)
+ j .Lkmc_done
+.align 16
+.Lcbc_software:
+___
+$code.=<<___;
+ stm${g} $key,$ra,5*$SIZE_T($sp)
+ lhi %r0,0
+ cl %r0,`$stdframe+$SIZE_T-4`($sp)
+ je .Lcbc_decrypt
+
+ larl $tbl,AES_Te
+
+ llgf $s0,0($ivp)
+ llgf $s1,4($ivp)
+ llgf $s2,8($ivp)
+ llgf $s3,12($ivp)
+
+ lghi $t0,16
+ sl${g}r $len,$t0
+ brc 4,.Lcbc_enc_tail # if borrow
+.Lcbc_enc_loop:
+ stm${g} $inp,$out,2*$SIZE_T($sp)
+ x $s0,0($inp)
+ x $s1,4($inp)
+ x $s2,8($inp)
+ x $s3,12($inp)
+ lgr %r4,$key
+
+ bras $ra,_s390x_AES_encrypt
+
+ lm${g} $inp,$key,2*$SIZE_T($sp)
+ st $s0,0($out)
+ st $s1,4($out)
+ st $s2,8($out)
+ st $s3,12($out)
+
+ la $inp,16($inp)
+ la $out,16($out)
+ lghi $t0,16
+ lt${g}r $len,$len
+ jz .Lcbc_enc_done
+ sl${g}r $len,$t0
+ brc 4,.Lcbc_enc_tail # if borrow
+ j .Lcbc_enc_loop
+.align 16
+.Lcbc_enc_done:
+ l${g} $ivp,6*$SIZE_T($sp)
+ st $s0,0($ivp)
+ st $s1,4($ivp)
+ st $s2,8($ivp)
+ st $s3,12($ivp)
+
+ lm${g} %r7,$ra,7*$SIZE_T($sp)
+ br $ra
+
+.align 16
+.Lcbc_enc_tail:
+ aghi $len,15
+ lghi $t0,0
+ stg $t0,16*$SIZE_T($sp)
+ stg $t0,16*$SIZE_T+8($sp)
+ bras $t1,3f
+ mvc 16*$SIZE_T(1,$sp),0($inp)
+3: ex $len,0($t1)
+ lghi $len,0
+ la $inp,16*$SIZE_T($sp)
+ j .Lcbc_enc_loop
+
+.align 16
+.Lcbc_decrypt:
+ larl $tbl,AES_Td
+
+ lg $t0,0($ivp)
+ lg $t1,8($ivp)
+ stmg $t0,$t1,16*$SIZE_T($sp)
+
+.Lcbc_dec_loop:
+ stm${g} $inp,$out,2*$SIZE_T($sp)
+ llgf $s0,0($inp)
+ llgf $s1,4($inp)
+ llgf $s2,8($inp)
+ llgf $s3,12($inp)
+ lgr %r4,$key
+
+ bras $ra,_s390x_AES_decrypt
+
+ lm${g} $inp,$key,2*$SIZE_T($sp)
+ sllg $s0,$s0,32
+ sllg $s2,$s2,32
+ lr $s0,$s1
+ lr $s2,$s3
+
+ lg $t0,0($inp)
+ lg $t1,8($inp)
+ xg $s0,16*$SIZE_T($sp)
+ xg $s2,16*$SIZE_T+8($sp)
+ lghi $s1,16
+ sl${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 $t0,$t1,16*$SIZE_T($sp)
+
+ la $inp,16($inp)
+ la $out,16($out)
+ j .Lcbc_dec_loop
+
+.Lcbc_dec_done:
+ stg $s0,0($out)
+ stg $s2,8($out)
+.Lcbc_dec_exit:
+ lm${g} %r6,$ra,6*$SIZE_T($sp)
+ stmg $t0,$t1,0($ivp)
+
+ br $ra
+
+.align 16
+.Lcbc_dec_tail:
+ aghi $len,15
+ stg $s0,16*$SIZE_T($sp)
+ stg $s2,16*$SIZE_T+8($sp)
+ bras $s1,4f
+ mvc 0(1,$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 $s1,%r0
+ lghi %r0,0 # query capability vector
+ la %r1,2*$SIZE_T($sp)
+ .long 0xb92e0042 # km %r4,%r2
+ llihh %r1,0x8000
+ srlg %r1,%r1,32($s1) # check for 32+function code
+ ng %r1,2*$SIZE_T($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(%r1) # 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
+ srlg $i2,$s0,63 # carry bit from lower half
+ sllg $s0,$s0,1
+ sllg $s1,$s1,1
+ xgr $s0,$i1
+ ogr $s1,$i2
+.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
+ srlg $i2,$s0,63 # carry bit from lower half
+ sllg $s0,$s0,1
+ sllg $s1,$s1,1
+ xgr $s0,$i1
+ ogr $s1,$i2
+
+ 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
+
+ stm${g} %r6,$s3,6*$SIZE_T($sp)
+ st${g} $ra,14*$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:
+ l${g} $ra,14*$SIZE_T($sp)
+ st${g} $sp,$tweak($sp) # wipe tweak
+ st${g} $sp,$tweak($sp)
+ lm${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
+
+ xgr $s0,$s0 # clear upper half
+ xgr $s1,$s1
+ lrv $s0,$stdframe+4($sp) # load secno
+ lrv $s1,$stdframe+0($sp)
+ xgr $s2,$s2
+ xgr $s3,$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
+ srlg %r0,$s1,63 # carry bit from lower half
+ sllg $s1,$s1,1
+ sllg $s3,$s3,1
+ xgr $s1,%r1
+ ogr $s3,%r0
+ 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
+ srlg %r0,$s1,63 # carry bit from lower half
+ sllg $s1,$s1,1
+ sllg $s3,$s3,1
+ xgr $s1,%r1
+ ogr $s3,%r0
+ 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 $ra
+.size AES_xts_encrypt,.-AES_xts_encrypt
+___
+# void AES_xts_decrypt(const char *inp,char *out,size_t len,
+# const AES_KEY *key1, const AES_KEY *key2,u64 secno);
+#
+$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
+
+ stm${g} %r6,$s3,6*$SIZE_T($sp)
+ st${g} $ra,14*$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
+ srlg $i2,$s0,63 # carry bit from lower half
+ sllg $s0,$s0,1
+ sllg $s1,$s1,1
+ xgr $s0,$i1
+ ogr $s1,$i2
+ 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:
+ l${g} $ra,14*$SIZE_T($sp)
+ st${g} $sp,$tweak($sp) # wipe tweak
+ st${g} $sp,$tweak($sp)
+ lm${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
+
+ xgr $s0,$s0 # clear upper half
+ xgr $s1,$s1
+ lrv $s0,$stdframe+4($sp) # load secno
+ lrv $s1,$stdframe+0($sp)
+ xgr $s2,$s2
+ xgr $s3,$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
+ srlg %r0,$s1,63 # carry bit from lower half
+ sllg $s1,$s1,1
+ sllg $s3,$s3,1
+ xgr $s1,%r1
+ ogr $s3,%r0
+ 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
+ srlg %r0,$s1,63 # carry bit from lower half
+ sllg $s1,$s1,1
+ sllg $s3,$s3,1
+ xgr $s1,%r1
+ ogr $s3,%r0
+ 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
+ srlg %r0,$s1,63 # carry bit from lower half
+ sllg $s1,$s1,1
+ sllg $s3,$s3,1
+ xgr $s1,%r1
+ ogr $s3,%r0
+ 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