2 # Copyright 2005-2016 The OpenSSL Project Authors. All Rights Reserved.
4 # Licensed under the OpenSSL license (the "License"). You may not use
5 # this file except in compliance with the License. You can obtain a copy
6 # in the file LICENSE in the source distribution or at
7 # https://www.openssl.org/source/license.html
12 if ($flavour =~ /\./) { $output = $flavour; undef $flavour; }
14 $win64=0; $win64=1 if ($flavour =~ /[nm]asm|mingw64/ || $output =~ /\.asm$/);
16 $0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
17 ( $xlate="${dir}x86_64-xlate.pl" and -f $xlate ) or
18 ( $xlate="${dir}perlasm/x86_64-xlate.pl" and -f $xlate) or
19 die "can't locate x86_64-xlate.pl";
21 open OUT,"| \"$^X\" \"$xlate\" $flavour \"$output\"";
24 ($arg1,$arg2,$arg3,$arg4)=$win64?("%rcx","%rdx","%r8", "%r9") : # Win64 order
25 ("%rdi","%rsi","%rdx","%rcx"); # Unix order
28 .extern OPENSSL_cpuid_setup
29 .hidden OPENSSL_cpuid_setup
31 call OPENSSL_cpuid_setup
33 .hidden OPENSSL_ia32cap_P
34 .comm OPENSSL_ia32cap_P,16,4
38 .globl OPENSSL_atomic_add
39 .type OPENSSL_atomic_add,\@abi-omnipotent
43 .Lspin: leaq ($arg2,%rax),%r8
48 .byte 0x48,0x98 # cltq/cdqe
50 .size OPENSSL_atomic_add,.-OPENSSL_atomic_add
53 .type OPENSSL_rdtsc,\@abi-omnipotent
60 .size OPENSSL_rdtsc,.-OPENSSL_rdtsc
62 .globl OPENSSL_ia32_cpuid
63 .type OPENSSL_ia32_cpuid,\@function,1
67 mov %rbx,%r8 # save %rbx
68 .cfi_register %rbx,%r8
71 mov %eax,8(%rdi) # clear 3rd word
73 mov %eax,%r11d # max value for standard query level
86 cmp \$0x756e6547,%ebx # "Genu"
89 cmp \$0x49656e69,%edx # "ineI"
92 cmp \$0x6c65746e,%ecx # "ntel"
94 or %eax,%r9d # 0 indicates Intel CPU
97 cmp \$0x68747541,%ebx # "Auth"
100 cmp \$0x69746E65,%edx # "enti"
103 cmp \$0x444D4163,%ecx # "cAMD"
105 or %eax,%r10d # 0 indicates AMD CPU
109 mov \$0x80000000,%eax
111 cmp \$0x80000001,%eax
114 mov \$0x80000001,%eax
117 and \$0x00000801,%r9d # isolate AMD XOP bit, 1<<11
119 cmp \$0x80000008,%r10d
122 mov \$0x80000008,%eax
124 movzb %cl,%r10 # number of cores - 1
125 inc %r10 # number of cores
129 bt \$28,%edx # test hyper-threading bit
131 shr \$16,%ebx # number of logical processors
134 and \$0xefffffff,%edx # ~(1<<28)
143 mov \$0,%ecx # query L1D
147 and \$0xfff,%r10d # number of cores -1 per L1D
152 and \$0xbfefffff,%edx # force reserved bits to 0
155 or \$0x40000000,%edx # set reserved bit#30 on Intel CPUs
157 cmp \$15,%ah # examine Family ID
159 or \$0x00100000,%edx # set reserved bit#20 to engage RC4_CHAR
161 bt \$28,%edx # test hyper-threading bit
163 and \$0xefffffff,%edx # ~(1<<28)
167 or \$0x10000000,%edx # 1<<28
169 cmp \$1,%bl # see if cache is shared
171 and \$0xefffffff,%edx # ~(1<<28)
173 and \$0x00000800,%r9d # isolate AMD XOP flag
174 and \$0xfffff7ff,%ecx
175 or %ecx,%r9d # merge AMD XOP flag
177 mov %edx,%r10d # %r9d:%r10d is copy of %ecx:%edx
178 bt \$27,%r9d # check OSXSAVE bit
181 .byte 0x0f,0x01,0xd0 # xgetbv
182 and \$0xe6,%eax # isolate XMM, YMM and ZMM state support
185 andl \$0xfffeffff,8(%rdi) # clear AVX512F, ~(1<<16)
186 # note that we don't touch other AVX512
187 # extensions, because they can be used
188 # with YMM (without opmasking though)
189 and \$6,%eax # isolate XMM and YMM state support
193 mov \$0xefffe7ff,%eax # ~(1<<28|1<<12|1<<11)
194 and %eax,%r9d # clear AVX, FMA and AMD XOP bits
195 mov \$0x3fdeffdf,%eax # ~(1<<31|1<<30|1<<21|1<<16|1<<5)
196 and %eax,8(%rdi) # cleax AVX2 and AVX512* bits
200 mov %r8,%rbx # restore %rbx
205 .size OPENSSL_ia32_cpuid,.-OPENSSL_ia32_cpuid
207 .globl OPENSSL_cleanse
208 .type OPENSSL_cleanse,\@abi-omnipotent
240 .size OPENSSL_cleanse,.-OPENSSL_cleanse
243 .type CRYPTO_memcmp,\@abi-omnipotent
262 .size CRYPTO_memcmp,.-CRYPTO_memcmp
265 print<<___ if (!$win64);
266 .globl OPENSSL_wipe_cpu
267 .type OPENSSL_wipe_cpu,\@abi-omnipotent
296 .size OPENSSL_wipe_cpu,.-OPENSSL_wipe_cpu
298 print<<___ if ($win64);
299 .globl OPENSSL_wipe_cpu
300 .type OPENSSL_wipe_cpu,\@abi-omnipotent
317 .size OPENSSL_wipe_cpu,.-OPENSSL_wipe_cpu
325 my $redzone=win64?8:-8;
328 .globl OPENSSL_instrument_bus
329 .type OPENSSL_instrument_bus,\@abi-omnipotent
331 OPENSSL_instrument_bus:
332 mov $arg1,$out # tribute to Win64
336 rdtsc # collect 1st tick
337 mov %eax,$lasttick # lasttick = tick
338 mov \$0,$lastdiff # lastdiff = 0
358 .size OPENSSL_instrument_bus,.-OPENSSL_instrument_bus
360 .globl OPENSSL_instrument_bus2
361 .type OPENSSL_instrument_bus2,\@abi-omnipotent
363 OPENSSL_instrument_bus2:
364 mov $arg1,$out # tribute to Win64
367 mov $cnt,$redzone(%rsp)
369 rdtsc # collect 1st tick
370 mov %eax,$lasttick # lasttick = tick
371 mov \$0,$lastdiff # lastdiff = 0
377 rdtsc # collect 1st diff
379 sub $lasttick,%eax # diff
380 mov %edx,$lasttick # lasttick = tick
381 mov %eax,$lastdiff # lastdiff = diff
385 add %eax,($out) # accumulate diff
392 sub $lasttick,%eax # diff
393 mov %edx,$lasttick # lasttick = tick
395 mov %eax,$lastdiff # lastdiff = diff
398 sub %rdx,$cnt # conditional --$cnt
399 lea ($out,%rdx,4),$out # conditional ++$out
403 mov $redzone(%rsp),%rax
406 .size OPENSSL_instrument_bus2,.-OPENSSL_instrument_bus2
413 .globl OPENSSL_ia32_${rdop}
414 .type OPENSSL_ia32_${rdop},\@abi-omnipotent
416 OPENSSL_ia32_${rdop}:
426 .size OPENSSL_ia32_${rdop},.-OPENSSL_ia32_${rdop}
428 .globl OPENSSL_ia32_${rdop}_bytes
429 .type OPENSSL_ia32_${rdop}_bytes,\@abi-omnipotent
431 OPENSSL_ia32_${rdop}_bytes:
432 xor %rax, %rax # return value
434 je .Ldone_${rdop}_bytes
439 jc .Lbreak_${rdop}_bytes
441 jnz .Loop_${rdop}_bytes
442 jmp .Ldone_${rdop}_bytes
445 .Lbreak_${rdop}_bytes:
447 jb .Ltail_${rdop}_bytes
452 jz .Ldone_${rdop}_bytes
454 jmp .Loop_${rdop}_bytes
457 .Ltail_${rdop}_bytes:
463 jnz .Ltail_${rdop}_bytes
465 .Ldone_${rdop}_bytes:
467 .size OPENSSL_ia32_${rdop}_bytes,.-OPENSSL_ia32_${rdop}_bytes
470 gen_random("rdrand");
471 gen_random("rdseed");
473 close STDOUT; # flush