X-Git-Url: https://git.librecmc.org/?a=blobdiff_plain;f=crypto%2Fx86_64cpuid.pl;h=6cb152148b5b6eae2d61da23e18dd442e3edfff5;hb=6e290a25c2cbdc26119c0866c20d9292f9e64dd8;hp=7b76522bd881f8eac8a53939a07941d19f957185;hpb=4bb90087d745c26401e09a3bd10137d7b05e9ea3;p=oweals%2Fopenssl.git

diff --git a/crypto/x86_64cpuid.pl b/crypto/x86_64cpuid.pl
index 7b76522bd8..6cb152148b 100644
--- a/crypto/x86_64cpuid.pl
+++ b/crypto/x86_64cpuid.pl
@@ -1,4 +1,11 @@
-#!/usr/bin/env perl
+#! /usr/bin/env perl
+# Copyright 2005-2016 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
+
 
 $flavour = shift;
 $output  = shift;
@@ -7,7 +14,12 @@ if ($flavour =~ /\./) { $output = $flavour; undef $flavour; }
 $win64=0; $win64=1 if ($flavour =~ /[nm]asm|mingw64/ || $output =~ /\.asm$/);
 
 $0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
-open STDOUT,"| $^X ${dir}perlasm/x86_64-xlate.pl $flavour $output";
+( $xlate="${dir}x86_64-xlate.pl" and -f $xlate ) or
+( $xlate="${dir}perlasm/x86_64-xlate.pl" and -f $xlate) or
+die "can't locate x86_64-xlate.pl";
+
+open OUT,"| \"$^X\" \"$xlate\" $flavour \"$output\"";
+*STDOUT=*OUT;
 
 ($arg1,$arg2,$arg3,$arg4)=$win64?("%rcx","%rdx","%r8", "%r9") :	# Win64 order
 				 ("%rdi","%rsi","%rdx","%rcx");	# Unix order
@@ -19,7 +31,7 @@ print<<___;
 	call	OPENSSL_cpuid_setup
 
 .hidden	OPENSSL_ia32cap_P
-.comm	OPENSSL_ia32cap_P,8
+.comm	OPENSSL_ia32cap_P,16,4
 
 .text
 
@@ -48,12 +60,13 @@ OPENSSL_rdtsc:
 .size	OPENSSL_rdtsc,.-OPENSSL_rdtsc
 
 .globl	OPENSSL_ia32_cpuid
-.type	OPENSSL_ia32_cpuid,\@abi-omnipotent
+.type	OPENSSL_ia32_cpuid,\@function,1
 .align	16
 OPENSSL_ia32_cpuid:
 	mov	%rbx,%r8		# save %rbx
 
 	xor	%eax,%eax
+	mov	%eax,8(%rdi)		# clear 3rd word
 	cpuid
 	mov	%eax,%r11d		# max value for standard query level
 
@@ -121,6 +134,14 @@ OPENSSL_ia32_cpuid:
 	shr	\$14,%r10d
 	and	\$0xfff,%r10d		# number of cores -1 per L1D
 
+	cmp	\$7,%r11d
+	jb	.Lnocacheinfo
+
+	mov	\$7,%eax
+	xor	%ecx,%ecx
+	cpuid
+	mov	%ebx,8(%rdi)
+
 .Lnocacheinfo:
 	mov	\$1,%eax
 	cpuid
@@ -160,6 +181,7 @@ OPENSSL_ia32_cpuid:
 .Lclear_avx:
 	mov	\$0xefffe7ff,%eax	# ~(1<<28|1<<12|1<<11)
 	and	%eax,%r9d		# clear AVX, FMA and AMD XOP bits
+	andl	\$0xffffffdf,8(%rdi)	# cleax AVX2, ~(1<<5)
 .Ldone:
 	shl	\$32,%r9
 	mov	%r10d,%eax
@@ -202,6 +224,28 @@ OPENSSL_cleanse:
 	jne	.Little
 	ret
 .size	OPENSSL_cleanse,.-OPENSSL_cleanse
+
+.globl  CRYPTO_memcmp
+.type   CRYPTO_memcmp,\@abi-omnipotent
+.align  16
+CRYPTO_memcmp:
+	xor	%rax,%rax
+	xor	%r10,%r10
+	cmp	\$0,$arg3
+	je	.Lno_data
+.Loop_cmp:
+	mov	($arg1),%r10b
+	lea	1($arg1),$arg1
+	xor	($arg2),%r10b
+	lea	1($arg2),$arg2
+	or	%r10b,%al
+	dec	$arg3
+	jnz	.Loop_cmp
+	neg	%rax
+	shr	\$63,%rax
+.Lno_data:
+	ret
+.size	CRYPTO_memcmp,.-CRYPTO_memcmp
 ___
 
 print<<___ if (!$win64);
@@ -349,4 +393,67 @@ OPENSSL_instrument_bus2:
 ___
 }
 
+sub gen_random {
+my $rdop = shift;
+print<<___;
+.globl	OPENSSL_ia32_${rdop}
+.type	OPENSSL_ia32_${rdop},\@abi-omnipotent
+.align	16
+OPENSSL_ia32_${rdop}:
+	mov	\$8,%ecx
+.Loop_${rdop}:
+	${rdop}	%rax
+	jc	.Lbreak_${rdop}
+	loop	.Loop_${rdop}
+.Lbreak_${rdop}:
+	cmp	\$0,%rax
+	cmove	%rcx,%rax
+	ret
+.size	OPENSSL_ia32_${rdop},.-OPENSSL_ia32_${rdop}
+
+.globl	OPENSSL_ia32_${rdop}_bytes
+.type	OPENSSL_ia32_${rdop}_bytes,\@abi-omnipotent
+.align	16
+OPENSSL_ia32_${rdop}_bytes:
+	xor	%rax, %rax	# return value
+	cmp	\$0,$arg2
+	je	.Ldone_${rdop}_bytes
+
+	mov	\$8,%r11
+.Loop_${rdop}_bytes:
+	${rdop}	%r10
+	jc	.Lbreak_${rdop}_bytes
+	dec	%r11
+	jnz	.Loop_${rdop}_bytes
+	jmp	.Ldone_${rdop}_bytes
+
+.align	16
+.Lbreak_${rdop}_bytes:
+	cmp	\$8,$arg2
+	jb	.Ltail_${rdop}_bytes
+	mov	%r10,($arg1)
+	lea	8($arg1),$arg1
+	add	\$8,%rax
+	sub	\$8,$arg2
+	jz	.Ldone_${rdop}_bytes
+	mov	\$8,%r11
+	jmp	.Loop_${rdop}_bytes
+
+.align	16
+.Ltail_${rdop}_bytes:
+	mov	%r10b,($arg1)
+	lea	1($arg1),$arg1
+	inc	%rax
+	shr	\$8,%r8
+	dec	$arg2
+	jnz	.Ltail_${rdop}_bytes
+
+.Ldone_${rdop}_bytes:
+	ret
+.size	OPENSSL_ia32_${rdop}_bytes,.-OPENSSL_ia32_${rdop}_bytes
+___
+}
+gen_random("rdrand");
+gen_random("rdseed");
+
 close STDOUT;	# flush