crypto/x86[_64]cpuid.pl: add OPENSSL_ia32_rd[rand|seed]_bytes.

[openssl.git] / crypto / x86cpuid.pl

diff --git a/crypto/x86cpuid.pl b/crypto/x86cpuid.pl
index a1285b9d002af77bd44337d7d9a0d2049d2318b5..c45b18356bb12814d028cd92c3a94292a2b23229 100644 (file)
--- a/crypto/x86cpuid.pl
+++ b/crypto/x86cpuid.pl
@@ -1,9 +1,19 @@
-#!/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/); }
@@ -22,6 +32,8 @@ 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
 
@@ -67,6 +79,7 @@ for (@ARGV) { $sse2=1 if (/-DOPENSSL_IA32_SSE2/); }
        &inc    ("esi");                # number of cores
 
        &mov    ("eax",1);
+       &xor    ("ecx","ecx");
        &cpuid  ();
        &bt     ("edx",28);
        &jnc    (&label("generic"));
@@ -78,6 +91,16 @@ for (@ARGV) { $sse2=1 if (/-DOPENSSL_IA32_SSE2/); }
        &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"));
@@ -91,6 +114,7 @@ for (@ARGV) { $sse2=1 if (/-DOPENSSL_IA32_SSE2/); }
 
 &set_label("nocacheinfo");
        &mov    ("eax",1);
+       &xor    ("ecx","ecx");
        &cpuid  ();
        &and    ("edx",0xbfefffff);     # force reserved bits #20, #30 to 0
        &cmp    ("ebp",0);
@@ -133,6 +157,8 @@ for (@ARGV) { $sse2=1 if (/-DOPENSSL_IA32_SSE2/); }
        &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");
@@ -165,7 +191,7 @@ for (@ARGV) { $sse2=1 if (/-DOPENSSL_IA32_SSE2/); }
        &jnz    (&label("nohalt"));     # not enough privileges
 
        &pushf  ();
-       &pop    ("eax")
+       &pop    ("eax");
        &bt     ("eax",9);
        &jnc    (&label("nohalt"));     # interrupts are disabled
 
@@ -196,7 +222,7 @@ for (@ARGV) { $sse2=1 if (/-DOPENSSL_IA32_SSE2/); }
 
 &function_begin_B("OPENSSL_far_spin");
        &pushf  ();
-       &pop    ("eax")
+       &pop    ("eax");
        &bt     ("eax",9);
        &jnc    (&label("nospin"));     # interrupts are disabled
 
@@ -280,7 +306,7 @@ for (@ARGV) { $sse2=1 if (/-DOPENSSL_IA32_SSE2/); }
 #      arguments is 1 or 2!
 &function_begin_B("OPENSSL_indirect_call");
        {
-       my $i,$max=7;           # $max has to be chosen as 4*n-1
+       my ($max,$i)=(7,);      # $max has to be chosen as 4*n-1
                                # in order to preserve eventual
                                # stack alignment
        &push   ("ebp");
@@ -339,6 +365,31 @@ for (@ARGV) { $sse2=1 if (/-DOPENSSL_IA32_SSE2/); }
        &ret    ();
 &function_end_B("OPENSSL_cleanse");
 
+&function_begin_B("CRYPTO_memcmp");
+       &push   ("esi");
+       &push   ("edi");
+       &mov    ("esi",&wparam(0));
+       &mov    ("edi",&wparam(1));
+       &mov    ("ecx",&wparam(2));
+       &xor    ("eax","eax");
+       &xor    ("edx","edx");
+       &cmp    ("ecx",0);
+       &je     (&label("no_data"));
+&set_label("loop");
+       &mov    ("dl",&BP(0,"esi"));
+       &lea    ("esi",&DWP(1,"esi"));
+       &xor    ("dl",&BP(0,"edi"));
+       &lea    ("edi",&DWP(1,"edi"));
+       &or     ("al","dl");
+       &dec    ("ecx");
+       &jnz    (&label("loop"));
+       &neg    ("eax");
+       &shr    ("eax",31);
+&set_label("no_data");
+       &pop    ("edi");
+       &pop    ("esi");
+       &ret    ();
+&function_end_B("CRYPTO_memcmp");
 {
 my $lasttick = "esi";
 my $lastdiff = "ebx";
@@ -441,18 +492,70 @@ my $max = "ebp";
 &function_end("OPENSSL_instrument_bus2");
 }
 
-&function_begin_B("OPENSSL_ia32_rdrand");
+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;