-#!/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
-push(@INC,"perlasm");
+$0 =~ m/(.*[\/\\])[^\/\\]+$/; $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/); }
&pushf ();
&pop ("eax");
&xor ("ecx","eax");
+ &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
+
+ &xor ("eax","eax");
+ &cmp ("ebx",0x756e6547); # "Genu"
+ &setne (&LB("eax"));
+ &mov ("ebp","eax");
+ &cmp ("edx",0x49656e69); # "ineI"
+ &setne (&LB("eax"));
+ &or ("ebp","eax");
+ &cmp ("ecx",0x6c65746e); # "ntel"
+ &setne (&LB("eax"));
+ &or ("ebp","eax"); # 0 indicates Intel CPU
+ &jz (&label("intel"));
+
+ &cmp ("ebx",0x68747541); # "Auth"
+ &setne (&LB("eax"));
+ &mov ("esi","eax");
+ &cmp ("edx",0x69746E65); # "enti"
+ &setne (&LB("eax"));
+ &or ("esi","eax");
+ &cmp ("ecx",0x444D4163); # "cAMD"
+ &setne (&LB("eax"));
+ &or ("esi","eax"); # 0 indicates AMD CPU
+ &jnz (&label("intel"));
+
+ # AMD specific
+ &mov ("eax",0x80000000);
+ &cpuid ();
+ &cmp ("eax",0x80000001);
+ &jb (&label("intel"));
+ &mov ("esi","eax");
+ &mov ("eax",0x80000001);
+ &cpuid ();
+ &or ("ebp","ecx");
+ &and ("ebp",1<<11|1); # isolate XOP bit
+ &cmp ("esi",0x80000008);
+ &jb (&label("intel"));
+
+ &mov ("eax",0x80000008);
+ &cpuid ();
+ &movz ("esi",&LB("ecx")); # number of cores - 1
+ &inc ("esi"); # number of cores
+
&mov ("eax",1);
+ &xor ("ecx","ecx");
+ &cpuid ();
+ &bt ("edx",28);
+ &jnc (&label("generic"));
+ &shr ("ebx",16);
+ &and ("ebx",0xff);
+ &cmp ("ebx","esi");
+ &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"));
+
+ &mov ("eax",4);
+ &mov ("ecx",0); # query L1D
+ &cpuid ();
+ &mov ("edi","eax");
+ &shr ("edi",14);
+ &and ("edi",0xfff); # number of cores -1 per L1D
+
+&set_label("nocacheinfo");
+ &mov ("eax",1);
+ &xor ("ecx","ecx");
+ &cpuid ();
+ &and ("edx",0xbfefffff); # force reserved bits #20, #30 to 0
+ &cmp ("ebp",0);
+ &jne (&label("notintel"));
+ &or ("edx",1<<30); # set reserved bit#30 on Intel CPUs
+ &and (&HB("eax"),15); # familiy ID
+ &cmp (&HB("eax"),15); # P4?
+ &jne (&label("notintel"));
+ &or ("edx",1<<20); # set reserved bit#20 to engage RC4_CHAR
+&set_label("notintel");
+ &bt ("edx",28); # test hyper-threading bit
+ &jnc (&label("generic"));
+ &and ("edx",0xefffffff);
+ &cmp ("edi",0);
+ &je (&label("generic"));
+
+ &or ("edx",0x10000000);
+ &shr ("ebx",16);
+ &cmp (&LB("ebx"),1);
+ &ja (&label("generic"));
+ &and ("edx",0xefffffff); # clear hyper-threading bit if not
+
+&set_label("generic");
+ &and ("ebp",1<<11); # isolate AMD XOP flag
+ &and ("ecx",0xfffff7ff); # force 11th bit to 0
+ &mov ("esi","edx");
+ &or ("ebp","ecx"); # merge AMD XOP flag
+
+ &bt ("ecx",27); # check OSXSAVE bit
+ &jnc (&label("clear_avx"));
+ &xor ("ecx","ecx");
+ &data_byte(0x0f,0x01,0xd0); # xgetbv
+ &and ("eax",6);
+ &cmp ("eax",6);
+ &je (&label("done"));
+ &cmp ("eax",2);
+ &je (&label("clear_avx"));
+&set_label("clear_xmm");
+ &and ("ebp",0xfdfffffd); # clear AESNI and PCLMULQDQ bits
+ &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");
- &mov ("eax","edx");
- &mov ("edx","ecx");
&function_end("OPENSSL_ia32_cpuid");
&external_label("OPENSSL_ia32cap_P");
&jnz (&label("nohalt")); # not enough privileges
&pushf ();
- &pop ("eax")
+ &pop ("eax");
&bt ("eax",9);
&jnc (&label("nohalt")); # interrupts are disabled
&function_begin_B("OPENSSL_far_spin");
&pushf ();
- &pop ("eax")
+ &pop ("eax");
&bt ("eax",9);
&jnc (&label("nospin")); # interrupts are disabled
&bt (&DWP(0,"ecx"),1);
&jnc (&label("no_x87"));
if ($sse2) {
- &bt (&DWP(0,"ecx"),26);
- &jnc (&label("no_sse2"));
+ &and ("ecx",1<<26|1<<24); # check SSE2 and FXSR bits
+ &cmp ("ecx",1<<26|1<<24);
+ &jne (&label("no_sse2"));
&pxor ("xmm0","xmm0");
&pxor ("xmm1","xmm1");
&pxor ("xmm2","xmm2");
&ret ();
&function_end_B("OPENSSL_atomic_add");
+# This function can become handy under Win32 in situations when
+# we don't know which calling convention, __stdcall or __cdecl(*),
+# indirect callee is using. In C it can be deployed as
+#
+#ifdef OPENSSL_CPUID_OBJ
+# type OPENSSL_indirect_call(void *f,...);
+# ...
+# OPENSSL_indirect_call(func,[up to $max arguments]);
+#endif
+#
+# (*) it's designed to work even for __fastcall if number of
+# arguments is 1 or 2!
+&function_begin_B("OPENSSL_indirect_call");
+ {
+ my ($max,$i)=(7,); # $max has to be chosen as 4*n-1
+ # in order to preserve eventual
+ # stack alignment
+ &push ("ebp");
+ &mov ("ebp","esp");
+ &sub ("esp",$max*4);
+ &mov ("ecx",&DWP(12,"ebp"));
+ &mov (&DWP(0,"esp"),"ecx");
+ &mov ("edx",&DWP(16,"ebp"));
+ &mov (&DWP(4,"esp"),"edx");
+ for($i=2;$i<$max;$i++)
+ {
+ # Some copies will be redundant/bogus...
+ &mov ("eax",&DWP(12+$i*4,"ebp"));
+ &mov (&DWP(0+$i*4,"esp"),"eax");
+ }
+ &call_ptr (&DWP(8,"ebp"));# make the call...
+ &mov ("esp","ebp"); # ... and just restore the stack pointer
+ # without paying attention to what we called,
+ # (__cdecl *func) or (__stdcall *one).
+ &pop ("ebp");
+ &ret ();
+ }
+&function_end_B("OPENSSL_indirect_call");
+
+&function_begin_B("OPENSSL_cleanse");
+ &mov ("edx",&wparam(0));
+ &mov ("ecx",&wparam(1));
+ &xor ("eax","eax");
+ &cmp ("ecx",7);
+ &jae (&label("lot"));
+ &cmp ("ecx",0);
+ &je (&label("ret"));
+&set_label("little");
+ &mov (&BP(0,"edx"),"al");
+ &sub ("ecx",1);
+ &lea ("edx",&DWP(1,"edx"));
+ &jnz (&label("little"));
+&set_label("ret");
+ &ret ();
+
+&set_label("lot",16);
+ &test ("edx",3);
+ &jz (&label("aligned"));
+ &mov (&BP(0,"edx"),"al");
+ &lea ("ecx",&DWP(-1,"ecx"));
+ &lea ("edx",&DWP(1,"edx"));
+ &jmp (&label("lot"));
+&set_label("aligned");
+ &mov (&DWP(0,"edx"),"eax");
+ &lea ("ecx",&DWP(-4,"ecx"));
+ &test ("ecx",-4);
+ &lea ("edx",&DWP(4,"edx"));
+ &jnz (&label("aligned"));
+ &cmp ("ecx",0);
+ &jne (&label("little"));
+ &ret ();
+&function_end_B("OPENSSL_cleanse");
+
+{
+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");
+}
+
+&function_begin_B("OPENSSL_ia32_rdrand");
+ &mov ("ecx",8);
+&set_label("loop");
+ &rdrand ("eax");
+ &jc (&label("break"));
+ &loop (&label("loop"));
+&set_label("break");
+ &cmp ("eax",0);
+ &cmove ("eax","ecx");
+ &ret ();
+&function_end_B("OPENSSL_ia32_rdrand");
+
+&function_begin_B("OPENSSL_ia32_rdseed");
+ &mov ("ecx",8);
+&set_label("loop");
+ &rdseed ("eax");
+ &jc (&label("break"));
+ &loop (&label("loop"));
+&set_label("break");
+ &cmp ("eax",0);
+ &cmove ("eax","ecx");
+ &ret ();
+&function_end_B("OPENSSL_ia32_rdseed");
+
&initseg("OPENSSL_cpuid_setup");
+&hidden("OPENSSL_cpuid_setup");
+&hidden("OPENSSL_ia32cap_P");
+
&asm_finish();
+
+close STDOUT;
projects / openssl.git / blobdiff