2 # Copyright 2005-2018 The OpenSSL Project Authors. All Rights Reserved.
4 # Licensed under the Apache License 2.0 (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
10 # $output is the last argument if it looks like a file (it has an extension)
11 # $flavour is the first argument if it doesn't look like a file
12 $output = $#ARGV >= 0 && $ARGV[$#ARGV] =~ m|\.\w+$| ? pop : undef;
13 $flavour = $#ARGV >= 0 && $ARGV[0] !~ m|\.| ? shift : undef;
15 $win64=0; $win64=1 if ($flavour =~ /[nm]asm|mingw64/ || $output =~ /\.asm$/);
17 $0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
18 ( $xlate="${dir}x86_64-xlate.pl" and -f $xlate ) or
19 ( $xlate="${dir}perlasm/x86_64-xlate.pl" and -f $xlate) or
20 die "can't locate x86_64-xlate.pl";
22 open OUT,"| \"$^X\" \"$xlate\" $flavour \"$output\""
23 or die "can't call $xlate: $!";
26 ($arg1,$arg2,$arg3,$arg4)=$win64?("%rcx","%rdx","%r8", "%r9") : # Win64 order
27 ("%rdi","%rsi","%rdx","%rcx"); # Unix order
30 .extern OPENSSL_cpuid_setup
31 .hidden OPENSSL_cpuid_setup
33 call OPENSSL_cpuid_setup
35 .hidden OPENSSL_ia32cap_P
36 .comm OPENSSL_ia32cap_P,16,4
40 .globl OPENSSL_atomic_add
41 .type OPENSSL_atomic_add,\@abi-omnipotent
45 .Lspin: leaq ($arg2,%rax),%r8
50 .byte 0x48,0x98 # cltq/cdqe
52 .size OPENSSL_atomic_add,.-OPENSSL_atomic_add
55 .type OPENSSL_rdtsc,\@abi-omnipotent
62 .size OPENSSL_rdtsc,.-OPENSSL_rdtsc
64 .globl OPENSSL_ia32_cpuid
65 .type OPENSSL_ia32_cpuid,\@function,1
69 mov %rbx,%r8 # save %rbx
70 .cfi_register %rbx,%r8
73 mov %rax,8(%rdi) # clear extended feature flags
75 mov %eax,%r11d # max value for standard query level
78 cmp \$0x756e6547,%ebx # "Genu"
81 cmp \$0x49656e69,%edx # "ineI"
84 cmp \$0x6c65746e,%ecx # "ntel"
86 or %eax,%r9d # 0 indicates Intel CPU
89 cmp \$0x68747541,%ebx # "Auth"
92 cmp \$0x69746E65,%edx # "enti"
95 cmp \$0x444D4163,%ecx # "cAMD"
97 or %eax,%r10d # 0 indicates AMD CPU
101 mov \$0x80000000,%eax
103 cmp \$0x80000001,%eax
106 mov \$0x80000001,%eax
109 and \$0x00000801,%r9d # isolate AMD XOP bit, 1<<11
111 cmp \$0x80000008,%r10d
114 mov \$0x80000008,%eax
116 movzb %cl,%r10 # number of cores - 1
117 inc %r10 # number of cores
121 bt \$28,%edx # test hyper-threading bit
123 shr \$16,%ebx # number of logical processors
126 and \$0xefffffff,%edx # ~(1<<28)
135 mov \$0,%ecx # query L1D
139 and \$0xfff,%r10d # number of cores -1 per L1D
144 movd %eax,%xmm0 # put aside processor id
145 and \$0xbfefffff,%edx # force reserved bits to 0
148 or \$0x40000000,%edx # set reserved bit#30 on Intel CPUs
150 cmp \$15,%ah # examine Family ID
152 or \$0x00100000,%edx # set reserved bit#20 to engage RC4_CHAR
156 and \$0x0fff0ff0,%eax
157 cmp \$0x00050670,%eax # Knights Landing
159 cmp \$0x00080650,%eax # Knights Mill (according to sde)
162 and \$0xfbffffff,%ecx # clear XSAVE flag to mimic Silvermont
165 bt \$28,%edx # test hyper-threading bit
167 and \$0xefffffff,%edx # ~(1<<28)
171 or \$0x10000000,%edx # 1<<28
173 cmp \$1,%bl # see if cache is shared
175 and \$0xefffffff,%edx # ~(1<<28)
177 and \$0x00000800,%r9d # isolate AMD XOP flag
178 and \$0xfffff7ff,%ecx
179 or %ecx,%r9d # merge AMD XOP flag
181 mov %edx,%r10d # %r9d:%r10d is copy of %ecx:%edx
184 jb .Lno_extended_info
188 bt \$26,%r9d # check XSAVE bit, cleared on Knights
190 and \$0xfff7ffff,%ebx # clear ADCX/ADOX flag
192 movd %xmm0,%eax # restore processor id
193 and \$0x0fff0ff0,%eax
194 cmp \$0x00050650,%eax # Skylake-X
196 and \$0xfffeffff,%ebx # ~(1<<16)
197 # suppress AVX512F flag on Skylake-X
199 mov %ebx,8(%rdi) # save extended feature flags
203 bt \$27,%r9d # check OSXSAVE bit
206 .byte 0x0f,0x01,0xd0 # xgetbv
207 and \$0xe6,%eax # isolate XMM, YMM and ZMM state support
210 andl \$0x3fdeffff,8(%rdi) # ~(1<<31|1<<30|1<<21|1<<16)
211 # clear AVX512F+BW+VL+FIMA, all of
212 # them are EVEX-encoded, which requires
213 # ZMM state support even if one uses
214 # only XMM and YMM :-(
215 and \$6,%eax # isolate XMM and YMM state support
219 mov \$0xefffe7ff,%eax # ~(1<<28|1<<12|1<<11)
220 and %eax,%r9d # clear AVX, FMA and AMD XOP bits
221 mov \$0x3fdeffdf,%eax # ~(1<<31|1<<30|1<<21|1<<16|1<<5)
222 and %eax,8(%rdi) # clear AVX2 and AVX512* bits
226 mov %r8,%rbx # restore %rbx
231 .size OPENSSL_ia32_cpuid,.-OPENSSL_ia32_cpuid
233 .globl OPENSSL_cleanse
234 .type OPENSSL_cleanse,\@abi-omnipotent
266 .size OPENSSL_cleanse,.-OPENSSL_cleanse
269 .type CRYPTO_memcmp,\@abi-omnipotent
300 .size CRYPTO_memcmp,.-CRYPTO_memcmp
303 print<<___ if (!$win64);
304 .globl OPENSSL_wipe_cpu
305 .type OPENSSL_wipe_cpu,\@abi-omnipotent
334 .size OPENSSL_wipe_cpu,.-OPENSSL_wipe_cpu
336 print<<___ if ($win64);
337 .globl OPENSSL_wipe_cpu
338 .type OPENSSL_wipe_cpu,\@abi-omnipotent
355 .size OPENSSL_wipe_cpu,.-OPENSSL_wipe_cpu
363 my $redzone=win64?8:-8;
366 .globl OPENSSL_instrument_bus
367 .type OPENSSL_instrument_bus,\@abi-omnipotent
369 OPENSSL_instrument_bus:
370 mov $arg1,$out # tribute to Win64
374 rdtsc # collect 1st tick
375 mov %eax,$lasttick # lasttick = tick
376 mov \$0,$lastdiff # lastdiff = 0
396 .size OPENSSL_instrument_bus,.-OPENSSL_instrument_bus
398 .globl OPENSSL_instrument_bus2
399 .type OPENSSL_instrument_bus2,\@abi-omnipotent
401 OPENSSL_instrument_bus2:
402 mov $arg1,$out # tribute to Win64
405 mov $cnt,$redzone(%rsp)
407 rdtsc # collect 1st tick
408 mov %eax,$lasttick # lasttick = tick
409 mov \$0,$lastdiff # lastdiff = 0
415 rdtsc # collect 1st diff
417 sub $lasttick,%eax # diff
418 mov %edx,$lasttick # lasttick = tick
419 mov %eax,$lastdiff # lastdiff = diff
423 add %eax,($out) # accumulate diff
430 sub $lasttick,%eax # diff
431 mov %edx,$lasttick # lasttick = tick
433 mov %eax,$lastdiff # lastdiff = diff
436 sub %rdx,$cnt # conditional --$cnt
437 lea ($out,%rdx,4),$out # conditional ++$out
441 mov $redzone(%rsp),%rax
444 .size OPENSSL_instrument_bus2,.-OPENSSL_instrument_bus2
451 .globl OPENSSL_ia32_${rdop}_bytes
452 .type OPENSSL_ia32_${rdop}_bytes,\@abi-omnipotent
454 OPENSSL_ia32_${rdop}_bytes:
455 xor %rax, %rax # return value
457 je .Ldone_${rdop}_bytes
462 jc .Lbreak_${rdop}_bytes
464 jnz .Loop_${rdop}_bytes
465 jmp .Ldone_${rdop}_bytes
468 .Lbreak_${rdop}_bytes:
470 jb .Ltail_${rdop}_bytes
475 jz .Ldone_${rdop}_bytes
477 jmp .Loop_${rdop}_bytes
480 .Ltail_${rdop}_bytes:
486 jnz .Ltail_${rdop}_bytes
488 .Ldone_${rdop}_bytes:
489 xor %r10,%r10 # Clear sensitive data from register
491 .size OPENSSL_ia32_${rdop}_bytes,.-OPENSSL_ia32_${rdop}_bytes
494 gen_random("rdrand");
495 gen_random("rdseed");
497 close STDOUT; # flush