X-Git-Url: https://git.openssl.org/gitweb/?a=blobdiff_plain;f=crypto%2Fx86cpuid.pl;h=176e8e38c8d6f21a260c91abb594e4c7dcbb7c61;hb=6577e00892fc7b2ce02223e1eff77330e05ea660;hp=e8eaef758262343b541942476157b644eac2e7d8;hpb=bdf40fd2511d61439d169146760a121bcb17f194;p=openssl.git

diff --git a/crypto/x86cpuid.pl b/crypto/x86cpuid.pl
index e8eaef7582..176e8e38c8 100644
--- 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/); }
@@ -19,9 +29,11 @@ for (@ARGV) { $sse2=1 if (/-DOPENSSL_IA32_SSE2/); }
 	&pushf	();
 	&pop	("eax");
 	&xor	("ecx","eax");
-	&bt	("ecx",21);
-	&jnc	(&label("generic"));
 	&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"));
@@ -76,8 +89,18 @@ for (@ARGV) { $sse2=1 if (/-DOPENSSL_IA32_SSE2/); }
 	&ja	(&label("generic"));
 	&and	("edx",0xefffffff);	# clear hyper-threading bit
 	&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);
@@ -133,9 +157,12 @@ 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");
+&set_label("nocpuid");
 &function_end("OPENSSL_ia32_cpuid");
 
 &external_label("OPENSSL_ia32cap_P");
@@ -164,7 +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
 
@@ -195,7 +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
 
@@ -279,7 +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");
@@ -338,6 +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";
@@ -440,18 +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;