-#!/usr/bin/env perl
+#! /usr/bin/env perl
+# Copyright 2004-2016 The OpenSSL Project Authors. All Rights Reserved.
+#
+# Licensed under the OpenSSL license (the "License"). You may not use
+# this file except in compliance with the License. You can obtain a copy
+# in the file LICENSE in the source distribution or at
+# https://www.openssl.org/source/license.html
$0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
push(@INC, "${dir}perlasm", "perlasm");
require "x86asm.pl";
+$output = pop;
+open OUT,">$output";
+*STDOUT=*OUT;
+
&asm_init($ARGV[0],"x86cpuid");
for (@ARGV) { $sse2=1 if (/-DOPENSSL_IA32_SSE2/); }
&xor ("eax","eax");
&bt ("ecx",21);
&jnc (&label("nocpuid"));
+ &mov ("esi",&wparam(0));
+ &mov (&DWP(8,"esi"),"eax"); # clear 3rd word
&cpuid ();
&mov ("edi","eax"); # max value for standard query level
&jmp (&label("generic"));
&set_label("intel");
+ &cmp ("edi",7);
+ &jb (&label("cacheinfo"));
+
+ &mov ("esi",&wparam(0));
+ &mov ("eax",7);
+ &xor ("ecx","ecx");
+ &cpuid ();
+ &mov (&DWP(8,"esi"),"ebx");
+
+&set_label("cacheinfo");
&cmp ("edi",4);
&mov ("edi",-1);
&jb (&label("nocacheinfo"));
&and ("esi",0xfeffffff); # clear FXSR
&set_label("clear_avx");
&and ("ebp",0xefffe7ff); # clear AVX, FMA and AMD XOP bits
+ &mov ("edi",&wparam(0));
+ &and (&DWP(8,"edi"),0xffffffdf); # clear AVX2
&set_label("done");
&mov ("eax","esi");
&mov ("edx","ebp");
&function_begin_B("OPENSSL_far_spin");
&pushf ();
- &pop ("eax")
+ &pop ("eax");
&bt ("eax",9);
&jnc (&label("nospin")); # interrupts are disabled
&ret ();
&function_end_B("OPENSSL_cleanse");
-&function_begin_B("OPENSSL_ia32_rdrand");
+&function_begin_B("CRYPTO_memcmp");
+ &push ("esi");
+ &push ("edi");
+ &mov ("esi",&wparam(0));
+ &mov ("edi",&wparam(1));
+ &mov ("ecx",&wparam(2));
+ &xor ("eax","eax");
+ &xor ("edx","edx");
+ &cmp ("ecx",0);
+ &je (&label("no_data"));
+&set_label("loop");
+ &mov ("dl",&BP(0,"esi"));
+ &lea ("esi",&DWP(1,"esi"));
+ &xor ("dl",&BP(0,"edi"));
+ &lea ("edi",&DWP(1,"edi"));
+ &or ("al","dl");
+ &dec ("ecx");
+ &jnz (&label("loop"));
+ &neg ("eax");
+ &shr ("eax",31);
+&set_label("no_data");
+ &pop ("edi");
+ &pop ("esi");
+ &ret ();
+&function_end_B("CRYPTO_memcmp");
+{
+my $lasttick = "esi";
+my $lastdiff = "ebx";
+my $out = "edi";
+my $cnt = "ecx";
+my $max = "ebp";
+
+&function_begin("OPENSSL_instrument_bus");
+ &mov ("eax",0);
+ if ($sse2) {
+ &picmeup("edx","OPENSSL_ia32cap_P");
+ &bt (&DWP(0,"edx"),4);
+ &jnc (&label("nogo")); # no TSC
+ &bt (&DWP(0,"edx"),19);
+ &jnc (&label("nogo")); # no CLFLUSH
+
+ &mov ($out,&wparam(0)); # load arguments
+ &mov ($cnt,&wparam(1));
+
+ # collect 1st tick
+ &rdtsc ();
+ &mov ($lasttick,"eax"); # lasttick = tick
+ &mov ($lastdiff,0); # lastdiff = 0
+ &clflush(&DWP(0,$out));
+ &data_byte(0xf0); # lock
+ &add (&DWP(0,$out),$lastdiff);
+ &jmp (&label("loop"));
+
+&set_label("loop",16);
+ &rdtsc ();
+ &mov ("edx","eax"); # put aside tick (yes, I neglect edx)
+ &sub ("eax",$lasttick); # diff
+ &mov ($lasttick,"edx"); # lasttick = tick
+ &mov ($lastdiff,"eax"); # lastdiff = diff
+ &clflush(&DWP(0,$out));
+ &data_byte(0xf0); # lock
+ &add (&DWP(0,$out),"eax"); # accumulate diff
+ &lea ($out,&DWP(4,$out)); # ++$out
+ &sub ($cnt,1); # --$cnt
+ &jnz (&label("loop"));
+
+ &mov ("eax",&wparam(1));
+&set_label("nogo");
+ }
+&function_end("OPENSSL_instrument_bus");
+
+&function_begin("OPENSSL_instrument_bus2");
+ &mov ("eax",0);
+ if ($sse2) {
+ &picmeup("edx","OPENSSL_ia32cap_P");
+ &bt (&DWP(0,"edx"),4);
+ &jnc (&label("nogo")); # no TSC
+ &bt (&DWP(0,"edx"),19);
+ &jnc (&label("nogo")); # no CLFLUSH
+
+ &mov ($out,&wparam(0)); # load arguments
+ &mov ($cnt,&wparam(1));
+ &mov ($max,&wparam(2));
+
+ &rdtsc (); # collect 1st tick
+ &mov ($lasttick,"eax"); # lasttick = tick
+ &mov ($lastdiff,0); # lastdiff = 0
+
+ &clflush(&DWP(0,$out));
+ &data_byte(0xf0); # lock
+ &add (&DWP(0,$out),$lastdiff);
+
+ &rdtsc (); # collect 1st diff
+ &mov ("edx","eax"); # put aside tick (yes, I neglect edx)
+ &sub ("eax",$lasttick); # diff
+ &mov ($lasttick,"edx"); # lasttick = tick
+ &mov ($lastdiff,"eax"); # lastdiff = diff
+ &jmp (&label("loop2"));
+
+&set_label("loop2",16);
+ &clflush(&DWP(0,$out));
+ &data_byte(0xf0); # lock
+ &add (&DWP(0,$out),"eax"); # accumulate diff
+
+ &sub ($max,1);
+ &jz (&label("done2"));
+
+ &rdtsc ();
+ &mov ("edx","eax"); # put aside tick (yes, I neglect edx)
+ &sub ("eax",$lasttick); # diff
+ &mov ($lasttick,"edx"); # lasttick = tick
+ &cmp ("eax",$lastdiff);
+ &mov ($lastdiff,"eax"); # lastdiff = diff
+ &mov ("edx",0);
+ &setne ("dl");
+ &sub ($cnt,"edx"); # conditional --$cnt
+ &lea ($out,&DWP(0,$out,"edx",4)); # conditional ++$out
+ &jnz (&label("loop2"));
+
+&set_label("done2");
+ &mov ("eax",&wparam(1));
+ &sub ("eax",$cnt);
+&set_label("nogo");
+ }
+&function_end("OPENSSL_instrument_bus2");
+}
+
+sub gen_random {
+my $rdop = shift;
+&function_begin_B("OPENSSL_ia32_${rdop}");
&mov ("ecx",8);
&set_label("loop");
- &rdrand ("eax");
+ &${rdop}("eax");
&jc (&label("break"));
&loop (&label("loop"));
&set_label("break");
&cmp ("eax",0);
&cmove ("eax","ecx");
&ret ();
-&function_end_B("OPENSSL_ia32_rdrand");
+&function_end_B("OPENSSL_ia32_${rdop}");
+
+&function_begin_B("OPENSSL_ia32_${rdop}_bytes");
+ &push ("edi");
+ &push ("ebx");
+ &xor ("eax","eax"); # return value
+ &mov ("edi",&wparam(0));
+ &mov ("ebx",&wparam(1));
+
+ &cmp ("ebx",0);
+ &je (&label("done"));
+
+ &mov ("ecx",8);
+&set_label("loop");
+ &${rdop}("edx");
+ &jc (&label("break"));
+ &loop (&label("loop"));
+ &jmp (&label("done"));
+
+&set_label("break",16);
+ &cmp ("ebx",4);
+ &jb (&label("tail"));
+ &mov (&DWP(0,"edi"),"edx");
+ &lea ("edi",&DWP(4,"edi"));
+ &add ("eax",4);
+ &sub ("ebx",4);
+ &jz (&label("done"));
+ &mov ("ecx",8);
+ &jmp (&label("loop"));
+
+&set_label("tail",16);
+ &mov (&BP(0,"edi"),"dl");
+ &lea ("edi",&DWP(1,"edi"));
+ &inc ("eax");
+ &shr ("edx",8);
+ &dec ("ebx");
+ &jnz (&label("tail"));
+
+&set_label("done");
+ &pop ("ebx");
+ &pop ("edi");
+ &ret ();
+&function_end_B("OPENSSL_ia32_${rdop}_bytes");
+}
+&gen_random("rdrand");
+&gen_random("rdseed");
&initseg("OPENSSL_cpuid_setup");
+&hidden("OPENSSL_cpuid_setup");
+&hidden("OPENSSL_ia32cap_P");
+
&asm_finish();
+
+close STDOUT;
projects / cassiopeia.git / blobdiff