#
# ====================================================================
# Written by Andy Polyakov <appro@fy.chalmers.se> for the OpenSSL
-# project. Rights for redistribution and usage in source and binary
-# forms are granted according to the OpenSSL license.
+# project. The module is, however, dual licensed under OpenSSL and
+# CRYPTOGAMS licenses depending on where you obtain it. For further
+# details see http://www.openssl.org/~appro/cryptogams/.
# ====================================================================
#
# Version 4.3.
# P4 56[60] 84[100] 23
# AMD K8 48[44] 70[79] 18
# PIII 41[50] 61[91] 24
+# Core 2 32[38] 45[70] 18.5
# Pentium 120 160 77
#
# Version 4.1 switches to compact S-box even in key schedule setup.
#
# Version 4.3 implements switch between compact and non-compact block
# functions in AES_cbc_encrypt depending on how much data was asked
-# to process in one stroke.
+# to be processed in one stroke.
#
+######################################################################
# Timing attacks are classified in two classes: synchronous when
-# attacker consciously initiates cryptographic operation and collect
+# attacker consciously initiates cryptographic operation and collects
# timing data of various character afterwards, and asynchronous when
# malicious code is executed on same CPU simultaneously with AES,
# instruments itself and performs statistical analysis of this data.
# timing. But note that *if* plain-text was concealed in such way that
# input to block function is distributed *uniformly*, then attack
# wouldn't apply. Now note that some encryption modes, most notably
-# CBC, do masks the plain-text in this exact way [secure cipher output
+# CBC, do mask the plain-text in this exact way [secure cipher output
# is distributed uniformly]. Yes, one still might find input that
# would reveal the information about given key, but if amount of
-# candidate inputs to be tried is larger than amount possible key
+# candidate inputs to be tried is larger than amount of possible key
# combinations then attack becomes infeasible. This is why revised
# AES_cbc_encrypt "dares" to switch to larger S-box when larger chunk
# of data is to be processed in one stroke. The current size limit of
# Current implementation accesses *all* cache-lines within ~50 cycles
# window, which is actually *less* than RDTSC latency on Intel P4!
-push(@INC,"perlasm","../../perlasm");
+$0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
+push(@INC,"${dir}","${dir}../../perlasm");
require "x86asm.pl";
&asm_init($ARGV[0],"aes-586.pl",$x86only = $ARGV[$#ARGV] eq "386");
+&static_label("AES_Te");
+&static_label("AES_Td");
$s0="eax";
$s1="ebx";
&mov ($acc,$s[$i]);
&and ($acc,0x80808080);
&mov ($tmp,$acc);
- &mov ($r2,$s[$i]);
&shr ($tmp,7);
- &and ($r2,0x7f7f7f7f);
+ &lea ($r2,&DWP(0,$s[$i],$s[$i]));
&sub ($acc,$tmp);
- &lea ($r2,&DWP(0,$r2,$r2));
+ &and ($r2,0xfefefefe);
&and ($acc,0x1b1b1b1b);
&mov ($tmp,$s[$i]);
&xor ($acc,$r2); # r2
&xor ($s[$i],$tmp);
}
-&public_label("AES_Te");
&function_begin_B("_x86_AES_encrypt_compact");
# note that caller is expected to allocate stack frame for me!
&mov ($__key,$key); # save key
}
if (!$x86only) {
-&public_label("AES_Te");
&function_begin_B("_sse_AES_encrypt_compact");
&pxor ("mm0",&QWP(0,$key)); # 7, 6, 5, 4, 3, 2, 1, 0
&pxor ("mm4",&QWP(8,$key)); # 15,14,13,12,11,10, 9, 8
if ($i==3) { &mov ($s[3],$acc); }
}
-&public_label("AES_Te");
&function_begin_B("_x86_AES_encrypt");
if ($vertical_spin) {
# I need high parts of volatile registers to be accessible...
&function_end_B("_x86_AES_encrypt");
# void AES_encrypt (const void *inp,void *out,const AES_KEY *key);
-&public_label("AES_Te");
&function_begin("AES_encrypt");
&mov ($acc,&wparam(0)); # load inp
&mov ($key,&wparam(2)); # load key
&mov ($acc,$s[$i]);
&and ($acc,0x80808080);
&mov ($tmp,$acc);
- &mov ($tp2,$s[$i]);
&shr ($tmp,7);
- &and ($tp2,0x7f7f7f7f);
+ &lea ($tp2,&DWP(0,$s[$i],$s[$i]));
&sub ($acc,$tmp);
- &add ($tp2,$tp2);
+ &and ($tp2,0xfefefefe);
&and ($acc,0x1b1b1b1b);
&xor ($acc,$tp2);
&mov ($tp2,$acc);
&and ($acc,0x80808080);
&mov ($tmp,$acc);
- &mov ($tp4,$tp2);
- &xor ($tp2,$s[$i]); # tp2^tp1
&shr ($tmp,7);
- &and ($tp4,0x7f7f7f7f);
+ &lea ($tp4,&DWP(0,$tp2,$tp2));
&sub ($acc,$tmp);
- &add ($tp4,$tp4);
+ &and ($tp4,0xfefefefe);
&and ($acc,0x1b1b1b1b);
+ &xor ($tp2,$s[$i]); # tp2^tp1
&xor ($acc,$tp4);
&mov ($tp4,$acc);
&and ($acc,0x80808080);
&mov ($tmp,$acc);
- &mov ($tp8,$tp4);
- &xor ($tp4,$s[$i]); # tp4^tp1
&shr ($tmp,7);
- &and ($tp8,0x7f7f7f7f);
+ &lea ($tp8,&DWP(0,$tp4,$tp4));
&sub ($acc,$tmp);
- &add ($tp8,$tp8);
+ &and ($tp8,0xfefefefe);
&and ($acc,0x1b1b1b1b);
+ &xor ($tp4,$s[$i]); # tp4^tp1
&rotl ($s[$i],8); # = ROTATE(tp1,8)
&xor ($tp8,$acc);
&xor ($s[$i],$tp2);
&xor ($tp2,$tp8);
- &xor ($s[$i],$tp4);
&rotl ($tp2,24);
+ &xor ($s[$i],$tp4);
&xor ($tp4,$tp8);
- &xor ($s[$i],$tp8); # ^= tp8^(tp4^tp1)^(tp2^tp1)
&rotl ($tp4,16);
- &xor ($s[$i],$tp2); # ^= ROTATE(tp8^tp2^tp1,24)
+ &xor ($s[$i],$tp8); # ^= tp8^(tp4^tp1)^(tp2^tp1)
&rotl ($tp8,8);
+ &xor ($s[$i],$tp2); # ^= ROTATE(tp8^tp2^tp1,24)
&xor ($s[$i],$tp4); # ^= ROTATE(tp8^tp4^tp1,16)
+ &mov ($s[0],$__s0) if($i==2); #prefetch $s0
+ &mov ($s[1],$__s1) if($i==3); #prefetch $s1
+ &mov ($s[2],$__s2) if($i==1);
&xor ($s[$i],$tp8); # ^= ROTATE(tp8,8)
- &mov ($s[0],$__s0) if($i==2); #prefetch $s0
- &mov ($s[1],$__s1) if($i==3); #prefetch $s1
- &mov ($s[2],$__s2) if($i==1);
&mov ($s[3],$__s3) if($i==1);
&mov (&DWP(4+4*$i,"esp"),$s[$i]) if($i>=2);
}
-&public_label("AES_Td");
&function_begin_B("_x86_AES_decrypt_compact");
# note that caller is expected to allocate stack frame for me!
&mov ($__key,$key); # save key
}
if (!$x86only) {
-&public_label("AES_Td");
&function_begin_B("_sse_AES_decrypt_compact");
&pxor ("mm0",&QWP(0,$key)); # 7, 6, 5, 4, 3, 2, 1, 0
&pxor ("mm4",&QWP(8,$key)); # 15,14,13,12,11,10, 9, 8
&lea ($td,&DWP(-2048,$td)); }
}
-&public_label("AES_Td");
&function_begin_B("_x86_AES_decrypt");
# note that caller is expected to allocate stack frame for me!
&mov ($__key,$key); # save key
&function_end_B("_x86_AES_decrypt");
# void AES_decrypt (const void *inp,void *out,const AES_KEY *key);
-&public_label("AES_Td");
&function_begin("AES_decrypt");
&mov ($acc,&wparam(0)); # load inp
&mov ($key,&wparam(2)); # load key
my $aes_key=&DWP(76,"esp"); # copy of aes_key
my $mark=&DWP(76+240,"esp"); # copy of aes_key->rounds
-&public_label("AES_Te");
-&public_label("AES_Td");
&function_begin("AES_cbc_encrypt");
&mov ($s2 eq "ecx"? $s2 : "",&wparam(2)); # load len
&cmp ($s2,0);
#--------------------------- SLOW ENCRYPT ---------------------------#
&cmp ($s2,16);
+ &mov ($s3,$s1);
&jb (&label("slow_enc_tail"));
if (!$x86only) {
&pushf (); # kludge, never executed
&set_label("slow_enc_tail",16);
- &emms ();
+ &emms () if (!$x86only);
&mov ($key eq "edi"? $key:"",$s3); # load out to edi
&mov ($s1,16);
&sub ($s1,$s2);
&align (4);
&data_word(0xAAF3F689); # rep stosb # zero tail
- &lea ($key,&DWP(-16,$s3)); # restore ivp
+ &mov ($key,$_ivp); # restore ivp
&mov ($acc,$s3); # output as input
&mov ($s0,&DWP(0,$key));
&mov ($s1,&DWP(4,$key));
&mov ($acc,$_inp); # load inp
&lea ($acc,&DWP(16,$acc)); # advance inp
&mov ($_inp,$acc); # save inp
- &mov ($_len,$s2); # save len
&jnz (&label("slow_dec_loop_x86"));
&mov ("esp",$_esp);
&popf ();
&xor ("eax",&DWP(1024-128,$tbl,"ecx",4)); # rcon
}
-# int AES_set_encrypt_key(const unsigned char *userKey, const int bits,
-# AES_KEY *key)
-&public_label("AES_Te");
-&function_begin("AES_set_encrypt_key");
- &mov ("esi",&wparam(0)); # user supplied key
- &mov ("edi",&wparam(2)); # private key schedule
+&function_begin("_x86_AES_set_encrypt_key");
+ &mov ("esi",&wparam(1)); # user supplied key
+ &mov ("edi",&wparam(3)); # private key schedule
&test ("esi",-1);
&jz (&label("badpointer"));
&mov ("ecx",&DWP(192-128,$tbl));
&mov ("edx",&DWP(224-128,$tbl));
- &mov ("ecx",&wparam(1)); # number of bits in key
+ &mov ("ecx",&wparam(2)); # number of bits in key
&cmp ("ecx",128);
&je (&label("10rounds"));
&cmp ("ecx",192);
&set_label("badpointer");
&mov ("eax",-1);
&set_label("exit");
-&function_end("AES_set_encrypt_key");
+&function_end("_x86_AES_set_encrypt_key");
+
+# int AES_set_encrypt_key(const unsigned char *userKey, const int bits,
+# AES_KEY *key)
+&function_begin_B("AES_set_encrypt_key");
+ &call ("_x86_AES_set_encrypt_key");
+ &ret ();
+&function_end_B("AES_set_encrypt_key");
sub deckey()
{ my ($i,$key,$tp1,$tp2,$tp4,$tp8) = @_;
&mov ($acc,$tp1);
&and ($acc,0x80808080);
&mov ($tmp,$acc);
- &mov ($tp2,$tp1);
&shr ($tmp,7);
- &and ($tp2,0x7f7f7f7f);
+ &lea ($tp2,&DWP(0,$tp1,$tp1));
&sub ($acc,$tmp);
- &add ($tp2,$tp2);
+ &and ($tp2,0xfefefefe);
&and ($acc,0x1b1b1b1b);
&xor ($acc,$tp2);
&mov ($tp2,$acc);
&and ($acc,0x80808080);
&mov ($tmp,$acc);
- &mov ($tp4,$tp2);
- &xor ($tp2,$tp1); # tp2^tp1
&shr ($tmp,7);
- &and ($tp4,0x7f7f7f7f);
+ &lea ($tp4,&DWP(0,$tp2,$tp2));
&sub ($acc,$tmp);
- &add ($tp4,$tp4);
+ &and ($tp4,0xfefefefe);
&and ($acc,0x1b1b1b1b);
+ &xor ($tp2,$tp1); # tp2^tp1
&xor ($acc,$tp4);
&mov ($tp4,$acc);
&and ($acc,0x80808080);
&mov ($tmp,$acc);
- &mov ($tp8,$tp4);
- &xor ($tp4,$tp1); # tp4^tp1
&shr ($tmp,7);
- &and ($tp8,0x7f7f7f7f);
+ &lea ($tp8,&DWP(0,$tp4,$tp4));
+ &xor ($tp4,$tp1); # tp4^tp1
&sub ($acc,$tmp);
- &add ($tp8,$tp8);
+ &and ($tp8,0xfefefefe);
&and ($acc,0x1b1b1b1b);
&rotl ($tp1,8); # = ROTATE(tp1,8)
&xor ($tp8,$acc);
# int AES_set_decrypt_key(const unsigned char *userKey, const int bits,
# AES_KEY *key)
-&public_label("AES_Td");
-&public_label("AES_Te");
&function_begin_B("AES_set_decrypt_key");
- &mov ("eax",&wparam(0));
- &mov ("ecx",&wparam(1));
- &mov ("edx",&wparam(2));
- &sub ("esp",12);
- &mov (&DWP(0,"esp"),"eax");
- &mov (&DWP(4,"esp"),"ecx");
- &mov (&DWP(8,"esp"),"edx");
- &call ("AES_set_encrypt_key");
- &add ("esp",12);
+ &call ("_x86_AES_set_encrypt_key");
&cmp ("eax",0);
&je (&label("proceed"));
&ret ();
&xor ("eax","eax"); # return success
&function_end("AES_set_decrypt_key");
+&asciz("AES for x86, CRYPTOGAMS by <appro\@openssl.org>");
&asm_finish();