3 # ====================================================================
4 # Written by Andy Polyakov <appro@openssl.org> for the OpenSSL
5 # project. The module is, however, dual licensed under OpenSSL and
6 # CRYPTOGAMS licenses depending on where you obtain it. For further
7 # details see http://www.openssl.org/~appro/cryptogams/.
8 # ====================================================================
12 # Companion to x86_64-mont.pl that optimizes cache-timing attack
13 # countermeasures. The subroutines are produced by replacing bp[i]
14 # references in their x86_64-mont.pl counterparts with cache-neutral
15 # references to powers table computed in BN_mod_exp_mont_consttime.
16 # In addition subroutine that scatters elements of the powers table
17 # is implemented, so that scatter-/gathering can be tuned without
18 # bn_exp.c modifications.
22 if ($flavour =~ /\./) { $output = $flavour; undef $flavour; }
24 $win64=0; $win64=1 if ($flavour =~ /[nm]asm|mingw64/ || $output =~ /\.asm$/);
26 $0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
27 ( $xlate="${dir}x86_64-xlate.pl" and -f $xlate ) or
28 ( $xlate="${dir}../../perlasm/x86_64-xlate.pl" and -f $xlate) or
29 die "can't locate x86_64-xlate.pl";
31 open OUT,"| \"$^X\" $xlate $flavour $output";
34 # int bn_mul_mont_gather5(
35 $rp="%rdi"; # BN_ULONG *rp,
36 $ap="%rsi"; # const BN_ULONG *ap,
37 $bp="%rdx"; # const BN_ULONG *bp,
38 $np="%rcx"; # const BN_ULONG *np,
39 $n0="%r8"; # const BN_ULONG *n0,
40 $num="%r9"; # int num,
41 # int idx); # 0 to 2^5-1, "index" in $bp holding
42 # pre-computed powers of a', interlaced
43 # in such manner that b[0] is $bp[idx],
44 # b[1] is [2^5+idx], etc.
56 .globl bn_mul_mont_gather5
57 .type bn_mul_mont_gather5,\@function,6
69 movd `($win64?56:8)`(%rsp),%xmm5 # load 7th argument
82 lea -264(%rsp,%r11,8),%rsp # tp=alloca(8*(num+2)+256+8)
83 and \$-1024,%rsp # minimize TLB usage
85 mov %rax,8(%rsp,$num,8) # tp[num+1]=%rsp
87 lea 128($bp),%r12 # reassign $bp (+size optimization)
90 $STRIDE=2**5*8; # 5 is "window size"
91 $N=$STRIDE/4; # should match cache line size
93 movdqa 0(%r10),%xmm0 # 00000001000000010000000000000000
94 movdqa 16(%r10),%xmm1 # 00000002000000020000000200000002
95 lea 24-112(%rsp,$num,8),%r10# place the mask after tp[num+3] (+ICache optimization)
98 pshufd \$0,%xmm5,%xmm5 # broadcast index
102 ########################################################################
103 # calculate mask by comparing 0..31 to index and save result to stack
107 pcmpeqd %xmm5,%xmm0 # compare to 1,0
111 for($k=0;$k<$STRIDE/16-4;$k+=4) {
114 pcmpeqd %xmm5,%xmm1 # compare to 3,2
115 movdqa %xmm0,`16*($k+0)+112`(%r10)
119 pcmpeqd %xmm5,%xmm2 # compare to 5,4
120 movdqa %xmm1,`16*($k+1)+112`(%r10)
124 pcmpeqd %xmm5,%xmm3 # compare to 7,6
125 movdqa %xmm2,`16*($k+2)+112`(%r10)
130 movdqa %xmm3,`16*($k+3)+112`(%r10)
134 $code.=<<___; # last iteration can be optimized
137 movdqa %xmm0,`16*($k+0)+112`(%r10)
142 movdqa %xmm1,`16*($k+1)+112`(%r10)
145 movdqa %xmm2,`16*($k+2)+112`(%r10)
146 pand `16*($k+0)-128`($bp),%xmm0 # while it's still in register
148 pand `16*($k+1)-128`($bp),%xmm1
149 pand `16*($k+2)-128`($bp),%xmm2
150 movdqa %xmm3,`16*($k+3)+112`(%r10)
151 pand `16*($k+3)-128`($bp),%xmm3
155 for($k=0;$k<$STRIDE/16-4;$k+=4) {
157 movdqa `16*($k+0)-128`($bp),%xmm4
158 movdqa `16*($k+1)-128`($bp),%xmm5
159 movdqa `16*($k+2)-128`($bp),%xmm2
160 pand `16*($k+0)+112`(%r10),%xmm4
161 movdqa `16*($k+3)-128`($bp),%xmm3
162 pand `16*($k+1)+112`(%r10),%xmm5
164 pand `16*($k+2)+112`(%r10),%xmm2
166 pand `16*($k+3)+112`(%r10),%xmm3
173 pshufd \$0x4e,%xmm0,%xmm1
176 movq %xmm0,$m0 # m0=bp[0]
178 mov ($n0),$n0 # pull n0[0] value
185 mulq $m0 # ap[0]*bp[0]
189 imulq $lo0,$m1 # "tp[0]"*n0
193 add %rax,$lo0 # discarded
206 add $hi0,$hi1 # np[j]*m1+ap[j]*bp[0]
209 mov $hi1,-16(%rsp,$j,8) # tp[j-1]
213 mulq $m0 # ap[j]*bp[0]
225 mov ($ap),%rax # ap[0]
227 add $hi0,$hi1 # np[j]*m1+ap[j]*bp[0]
229 mov $hi1,-16(%rsp,$j,8) # tp[j-1]
236 mov $hi1,-8(%rsp,$num,8)
237 mov %rdx,(%rsp,$num,8) # store upmost overflow bit
243 lea 24+128(%rsp,$num,8),%rdx # where 256-byte mask is (+size optimization)
248 for($k=0;$k<$STRIDE/16;$k+=4) {
250 movdqa `16*($k+0)-128`($bp),%xmm0
251 movdqa `16*($k+1)-128`($bp),%xmm1
252 movdqa `16*($k+2)-128`($bp),%xmm2
253 movdqa `16*($k+3)-128`($bp),%xmm3
254 pand `16*($k+0)-128`(%rdx),%xmm0
255 pand `16*($k+1)-128`(%rdx),%xmm1
257 pand `16*($k+2)-128`(%rdx),%xmm2
259 pand `16*($k+3)-128`(%rdx),%xmm3
266 pshufd \$0x4e,%xmm4,%xmm0
269 movq %xmm0,$m0 # m0=bp[i]
275 mulq $m0 # ap[0]*bp[i]
276 add %rax,$lo0 # ap[0]*bp[i]+tp[0]
280 imulq $lo0,$m1 # tp[0]*n0
284 add %rax,$lo0 # discarded
287 mov 8(%rsp),$lo0 # tp[1]
298 add $lo0,$hi1 # np[j]*m1+ap[j]*bp[i]+tp[j]
301 mov $hi1,-16(%rsp,$j,8) # tp[j-1]
305 mulq $m0 # ap[j]*bp[i]
309 add $hi0,$lo0 # ap[j]*bp[i]+tp[j]
319 mov ($ap),%rax # ap[0]
321 add $lo0,$hi1 # np[j]*m1+ap[j]*bp[i]+tp[j]
324 mov $hi1,-16(%rsp,$j,8) # tp[j-1]
330 add $lo0,$hi1 # pull upmost overflow bit
332 mov $hi1,-8(%rsp,$num,8)
333 mov %rdx,(%rsp,$num,8) # store upmost overflow bit
339 xor $i,$i # i=0 and clear CF!
340 mov (%rsp),%rax # tp[0]
341 lea (%rsp),$ap # borrow ap for tp
345 .Lsub: sbb ($np,$i,8),%rax
346 mov %rax,($rp,$i,8) # rp[i]=tp[i]-np[i]
347 mov 8($ap,$i,8),%rax # tp[i+1]
349 dec $j # doesnn't affect CF!
352 sbb \$0,%rax # handle upmost overflow bit
359 or $np,$ap # ap=borrow?tp:rp
361 .Lcopy: # copy or in-place refresh
363 mov $i,(%rsp,$i,8) # zap temporary vector
364 mov %rax,($rp,$i,8) # rp[i]=tp[i]
369 mov 8(%rsp,$num,8),%rsi # restore %rsp
381 .size bn_mul_mont_gather5,.-bn_mul_mont_gather5
384 my @A=("%r10","%r11");
385 my @N=("%r13","%rdi");
387 .type bn_mul4x_mont_gather5,\@function,6
389 bn_mul4x_mont_gather5:
392 movd `($win64?56:8)`(%rsp),%xmm5 # load 7th argument
405 lea -256(%rsp,%r11,8),%rsp # tp=alloca(8*(num+4)+256)
406 and \$-1024,%rsp # minimize TLB usage
408 mov %rax,8(%rsp,$num,8) # tp[num+1]=%rsp
410 mov $rp,16(%rsp,$num,8) # tp[num+2]=$rp
411 lea 128(%rdx),%r12 # reassign $bp (+size optimization)
414 $STRIDE=2**5*8; # 5 is "window size"
415 $N=$STRIDE/4; # should match cache line size
417 movdqa 0(%r10),%xmm0 # 00000001000000010000000000000000
418 movdqa 16(%r10),%xmm1 # 00000002000000020000000200000002
419 lea 32-112(%rsp,$num,8),%r10# place the mask after tp[num+4] (+ICache optimization)
421 pshufd \$0,%xmm5,%xmm5 # broadcast index
426 ########################################################################
427 # calculate mask by comparing 0..31 to index and save result to stack
431 pcmpeqd %xmm5,%xmm0 # compare to 1,0
435 for($k=0;$k<$STRIDE/16-4;$k+=4) {
438 pcmpeqd %xmm5,%xmm1 # compare to 3,2
439 movdqa %xmm0,`16*($k+0)+112`(%r10)
443 pcmpeqd %xmm5,%xmm2 # compare to 5,4
444 movdqa %xmm1,`16*($k+1)+112`(%r10)
448 pcmpeqd %xmm5,%xmm3 # compare to 7,6
449 movdqa %xmm2,`16*($k+2)+112`(%r10)
454 movdqa %xmm3,`16*($k+3)+112`(%r10)
458 $code.=<<___; # last iteration can be optimized
461 movdqa %xmm0,`16*($k+0)+112`(%r10)
466 movdqa %xmm1,`16*($k+1)+112`(%r10)
469 movdqa %xmm2,`16*($k+2)+112`(%r10)
470 pand `16*($k+0)-128`($bp),%xmm0 # while it's still in register
472 pand `16*($k+1)-128`($bp),%xmm1
473 pand `16*($k+2)-128`($bp),%xmm2
474 movdqa %xmm3,`16*($k+3)+112`(%r10)
475 pand `16*($k+3)-128`($bp),%xmm3
479 for($k=0;$k<$STRIDE/16-4;$k+=4) {
481 movdqa `16*($k+0)-128`($bp),%xmm4
482 movdqa `16*($k+1)-128`($bp),%xmm5
483 movdqa `16*($k+2)-128`($bp),%xmm2
484 pand `16*($k+0)+112`(%r10),%xmm4
485 movdqa `16*($k+3)-128`($bp),%xmm3
486 pand `16*($k+1)+112`(%r10),%xmm5
488 pand `16*($k+2)+112`(%r10),%xmm2
490 pand `16*($k+3)+112`(%r10),%xmm3
497 pshufd \$0x4e,%xmm0,%xmm1
500 movq %xmm0,$m0 # m0=bp[0]
502 mov ($n0),$n0 # pull n0[0] value
509 mulq $m0 # ap[0]*bp[0]
513 imulq $A[0],$m1 # "tp[0]"*n0
517 add %rax,$A[0] # discarded
540 mulq $m0 # ap[j]*bp[0]
542 mov -16($np,$j,8),%rax
548 mov -8($ap,$j,8),%rax
550 add $A[0],$N[0] # np[j]*m1+ap[j]*bp[0]
552 mov $N[0],-24(%rsp,$j,8) # tp[j-1]
555 mulq $m0 # ap[j]*bp[0]
557 mov -8($np,$j,8),%rax
565 add $A[1],$N[1] # np[j]*m1+ap[j]*bp[0]
567 mov $N[1],-16(%rsp,$j,8) # tp[j-1]
570 mulq $m0 # ap[j]*bp[0]
580 add $A[0],$N[0] # np[j]*m1+ap[j]*bp[0]
582 mov $N[0],-8(%rsp,$j,8) # tp[j-1]
585 mulq $m0 # ap[j]*bp[0]
594 mov -16($ap,$j,8),%rax
596 add $A[1],$N[1] # np[j]*m1+ap[j]*bp[0]
598 mov $N[1],-32(%rsp,$j,8) # tp[j-1]
603 mulq $m0 # ap[j]*bp[0]
605 mov -16($np,$j,8),%rax
611 mov -8($ap,$j,8),%rax
613 add $A[0],$N[0] # np[j]*m1+ap[j]*bp[0]
615 mov $N[0],-24(%rsp,$j,8) # tp[j-1]
618 mulq $m0 # ap[j]*bp[0]
620 mov -8($np,$j,8),%rax
626 mov ($ap),%rax # ap[0]
628 add $A[1],$N[1] # np[j]*m1+ap[j]*bp[0]
630 mov $N[1],-16(%rsp,$j,8) # tp[j-1]
636 mov $N[0],-8(%rsp,$j,8)
637 mov $N[1],(%rsp,$j,8) # store upmost overflow bit
642 lea 32+128(%rsp,$num,8),%rdx # where 256-byte mask is (+size optimization)
646 for($k=0;$k<$STRIDE/16;$k+=4) {
648 movdqa `16*($k+0)-128`($bp),%xmm0
649 movdqa `16*($k+1)-128`($bp),%xmm1
650 movdqa `16*($k+2)-128`($bp),%xmm2
651 movdqa `16*($k+3)-128`($bp),%xmm3
652 pand `16*($k+0)-128`(%rdx),%xmm0
653 pand `16*($k+1)-128`(%rdx),%xmm1
655 pand `16*($k+2)-128`(%rdx),%xmm2
657 pand `16*($k+3)-128`(%rdx),%xmm3
664 pshufd \$0x4e,%xmm4,%xmm0
667 movq %xmm0,$m0 # m0=bp[i]
673 mulq $m0 # ap[0]*bp[i]
674 add %rax,$A[0] # ap[0]*bp[i]+tp[0]
678 imulq $A[0],$m1 # tp[0]*n0
682 add %rax,$A[0] # "$N[0]", discarded
687 mulq $m0 # ap[j]*bp[i]
691 add 8(%rsp),$A[1] # +tp[1]
699 add $A[1],$N[1] # np[j]*m1+ap[j]*bp[i]+tp[j]
706 mulq $m0 # ap[j]*bp[i]
708 mov -16($np,$j,8),%rax
710 add -16(%rsp,$j,8),$A[0] # ap[j]*bp[i]+tp[j]
716 mov -8($ap,$j,8),%rax
720 mov $N[1],-32(%rsp,$j,8) # tp[j-1]
723 mulq $m0 # ap[j]*bp[i]
725 mov -8($np,$j,8),%rax
727 add -8(%rsp,$j,8),$A[1]
737 mov $N[0],-24(%rsp,$j,8) # tp[j-1]
740 mulq $m0 # ap[j]*bp[i]
744 add (%rsp,$j,8),$A[0] # ap[j]*bp[i]+tp[j]
754 mov $N[1],-16(%rsp,$j,8) # tp[j-1]
757 mulq $m0 # ap[j]*bp[i]
761 add 8(%rsp,$j,8),$A[1]
768 mov -16($ap,$j,8),%rax
772 mov $N[0],-40(%rsp,$j,8) # tp[j-1]
777 mulq $m0 # ap[j]*bp[i]
779 mov -16($np,$j,8),%rax
781 add -16(%rsp,$j,8),$A[0] # ap[j]*bp[i]+tp[j]
787 mov -8($ap,$j,8),%rax
791 mov $N[1],-32(%rsp,$j,8) # tp[j-1]
794 mulq $m0 # ap[j]*bp[i]
796 mov -8($np,$j,8),%rax
798 add -8(%rsp,$j,8),$A[1]
805 mov ($ap),%rax # ap[0]
809 mov $N[0],-24(%rsp,$j,8) # tp[j-1]
812 mov $N[1],-16(%rsp,$j,8) # tp[j-1]
817 add (%rsp,$num,8),$N[0] # pull upmost overflow bit
819 mov $N[0],-8(%rsp,$j,8)
820 mov $N[1],(%rsp,$j,8) # store upmost overflow bit
826 my @ri=("%rax","%rdx",$m0,$m1);
828 mov 16(%rsp,$num,8),$rp # restore $rp
829 mov 0(%rsp),@ri[0] # tp[0]
831 mov 8(%rsp),@ri[1] # tp[1]
832 shr \$2,$num # num/=4
833 lea (%rsp),$ap # borrow ap for tp
834 xor $i,$i # i=0 and clear CF!
837 mov 16($ap),@ri[2] # tp[2]
838 mov 24($ap),@ri[3] # tp[3]
840 lea -1($num),$j # j=num/4-1
844 mov @ri[0],0($rp,$i,8) # rp[i]=tp[i]-np[i]
845 mov @ri[1],8($rp,$i,8) # rp[i]=tp[i]-np[i]
846 sbb 16($np,$i,8),@ri[2]
847 mov 32($ap,$i,8),@ri[0] # tp[i+1]
848 mov 40($ap,$i,8),@ri[1]
849 sbb 24($np,$i,8),@ri[3]
850 mov @ri[2],16($rp,$i,8) # rp[i]=tp[i]-np[i]
851 mov @ri[3],24($rp,$i,8) # rp[i]=tp[i]-np[i]
852 sbb 32($np,$i,8),@ri[0]
853 mov 48($ap,$i,8),@ri[2]
854 mov 56($ap,$i,8),@ri[3]
855 sbb 40($np,$i,8),@ri[1]
857 dec $j # doesnn't affect CF!
860 mov @ri[0],0($rp,$i,8) # rp[i]=tp[i]-np[i]
861 mov 32($ap,$i,8),@ri[0] # load overflow bit
862 sbb 16($np,$i,8),@ri[2]
863 mov @ri[1],8($rp,$i,8) # rp[i]=tp[i]-np[i]
864 sbb 24($np,$i,8),@ri[3]
865 mov @ri[2],16($rp,$i,8) # rp[i]=tp[i]-np[i]
867 sbb \$0,@ri[0] # handle upmost overflow bit
868 mov @ri[3],24($rp,$i,8) # rp[i]=tp[i]-np[i]
875 or $np,$ap # ap=borrow?tp:rp
882 .Lcopy4x: # copy or in-place refresh
883 movdqu 16($ap,$i),%xmm2
884 movdqu 32($ap,$i),%xmm1
885 movdqa %xmm0,16(%rsp,$i)
886 movdqu %xmm2,16($rp,$i)
887 movdqa %xmm0,32(%rsp,$i)
888 movdqu %xmm1,32($rp,$i)
894 movdqu 16($ap,$i),%xmm2
895 movdqa %xmm0,16(%rsp,$i)
896 movdqu %xmm2,16($rp,$i)
900 mov 8(%rsp,$num,8),%rsi # restore %rsp
912 .size bn_mul4x_mont_gather5,.-bn_mul4x_mont_gather5
917 my ($inp,$num,$tbl,$idx)=$win64?("%rcx","%rdx","%r8", "%r9d") : # Win64 order
918 ("%rdi","%rsi","%rdx","%ecx"); # Unix order
925 .type bn_scatter5,\@abi-omnipotent
929 jz .Lscatter_epilogue
930 lea ($tbl,$idx,8),$tbl
940 .size bn_scatter5,.-bn_scatter5
943 .type bn_gather5,\@abi-omnipotent
946 .LSEH_begin_bn_gather5: # Win64 thing, but harmless in other cases
947 # I can't trust assembler to use specific encoding:-(
948 .byte 0x4c,0x8d,0x14,0x24 # lea (%rsp),%r10
949 .byte 0x48,0x81,0xec,0x08,0x01,0x00,0x00 # sub $0x108,%rsp
951 and \$-16,%rsp # shouldn't be formally required
954 movdqa 0(%rax),%xmm0 # 00000001000000010000000000000000
955 movdqa 16(%rax),%xmm1 # 00000002000000020000000200000002
956 lea 128($tbl),%r11 # size optimization
957 lea 128(%rsp),%rax # size optimization
959 pshufd \$0,%xmm5,%xmm5 # broadcast $idx
963 ########################################################################
964 # calculate mask by comparing 0..31 to $idx and save result to stack
966 for($i=0;$i<$STRIDE/16;$i+=4) {
969 pcmpeqd %xmm5,%xmm0 # compare to 1,0
971 $code.=<<___ if ($i);
972 movdqa %xmm3,`16*($i-1)-128`(%rax)
978 pcmpeqd %xmm5,%xmm1 # compare to 3,2
979 movdqa %xmm0,`16*($i+0)-128`(%rax)
983 pcmpeqd %xmm5,%xmm2 # compare to 5,4
984 movdqa %xmm1,`16*($i+1)-128`(%rax)
988 pcmpeqd %xmm5,%xmm3 # compare to 7,6
989 movdqa %xmm2,`16*($i+2)-128`(%rax)
994 movdqa %xmm3,`16*($i-1)-128`(%rax)
1002 for($i=0;$i<$STRIDE/16;$i+=4) {
1004 movdqa `16*($i+0)-128`(%r11),%xmm0
1005 movdqa `16*($i+1)-128`(%r11),%xmm1
1006 movdqa `16*($i+2)-128`(%r11),%xmm2
1007 pand `16*($i+0)-128`(%rax),%xmm0
1008 movdqa `16*($i+3)-128`(%r11),%xmm3
1009 pand `16*($i+1)-128`(%rax),%xmm1
1011 pand `16*($i+2)-128`(%rax),%xmm2
1013 pand `16*($i+3)-128`(%rax),%xmm3
1020 lea $STRIDE(%r11),%r11
1021 pshufd \$0x4e,%xmm4,%xmm0
1023 movq %xmm0,($out) # m0=bp[0]
1030 .LSEH_end_bn_gather5:
1031 .size bn_gather5,.-bn_gather5
1039 .asciz "Montgomery Multiplication with scatter/gather for x86_64, CRYPTOGAMS by <appro\@openssl.org>"
1042 # EXCEPTION_DISPOSITION handler (EXCEPTION_RECORD *rec,ULONG64 frame,
1043 # CONTEXT *context,DISPATCHER_CONTEXT *disp)
1051 .extern __imp_RtlVirtualUnwind
1052 .type mul_handler,\@abi-omnipotent
1066 mov 120($context),%rax # pull context->Rax
1067 mov 248($context),%rbx # pull context->Rip
1069 mov 8($disp),%rsi # disp->ImageBase
1070 mov 56($disp),%r11 # disp->HandlerData
1072 mov 0(%r11),%r10d # HandlerData[0]
1073 lea (%rsi,%r10),%r10 # end of prologue label
1074 cmp %r10,%rbx # context->Rip<end of prologue label
1075 jb .Lcommon_seh_tail
1079 mov 4(%r11),%r10d # HandlerData[1]
1080 lea (%rsi,%r10),%r10 # end of alloca label
1081 cmp %r10,%rbx # context->Rip<end of alloca label
1082 jb .Lcommon_seh_tail
1084 mov 152($context),%rax # pull context->Rsp
1086 mov 8(%r11),%r10d # HandlerData[2]
1087 lea (%rsi,%r10),%r10 # epilogue label
1088 cmp %r10,%rbx # context->Rip>=epilogue label
1089 jae .Lcommon_seh_tail
1091 mov 192($context),%r10 # pull $num
1092 mov 8(%rax,%r10,8),%rax # pull saved stack pointer
1102 mov %rbx,144($context) # restore context->Rbx
1103 mov %rbp,160($context) # restore context->Rbp
1104 mov %r12,216($context) # restore context->R12
1105 mov %r13,224($context) # restore context->R13
1106 mov %r14,232($context) # restore context->R14
1107 mov %r15,240($context) # restore context->R15
1112 mov %rax,152($context) # restore context->Rsp
1113 mov %rsi,168($context) # restore context->Rsi
1114 mov %rdi,176($context) # restore context->Rdi
1116 mov 40($disp),%rdi # disp->ContextRecord
1117 mov $context,%rsi # context
1118 mov \$154,%ecx # sizeof(CONTEXT)
1119 .long 0xa548f3fc # cld; rep movsq
1122 xor %rcx,%rcx # arg1, UNW_FLAG_NHANDLER
1123 mov 8(%rsi),%rdx # arg2, disp->ImageBase
1124 mov 0(%rsi),%r8 # arg3, disp->ControlPc
1125 mov 16(%rsi),%r9 # arg4, disp->FunctionEntry
1126 mov 40(%rsi),%r10 # disp->ContextRecord
1127 lea 56(%rsi),%r11 # &disp->HandlerData
1128 lea 24(%rsi),%r12 # &disp->EstablisherFrame
1129 mov %r10,32(%rsp) # arg5
1130 mov %r11,40(%rsp) # arg6
1131 mov %r12,48(%rsp) # arg7
1132 mov %rcx,56(%rsp) # arg8, (NULL)
1133 call *__imp_RtlVirtualUnwind(%rip)
1135 mov \$1,%eax # ExceptionContinueSearch
1147 .size mul_handler,.-mul_handler
1151 .rva .LSEH_begin_bn_mul_mont_gather5
1152 .rva .LSEH_end_bn_mul_mont_gather5
1153 .rva .LSEH_info_bn_mul_mont_gather5
1155 .rva .LSEH_begin_bn_mul4x_mont_gather5
1156 .rva .LSEH_end_bn_mul4x_mont_gather5
1157 .rva .LSEH_info_bn_mul4x_mont_gather5
1159 .rva .LSEH_begin_bn_gather5
1160 .rva .LSEH_end_bn_gather5
1161 .rva .LSEH_info_bn_gather5
1165 .LSEH_info_bn_mul_mont_gather5:
1168 .rva .Lmul_alloca,.Lmul_body,.Lmul_epilogue # HandlerData[]
1170 .LSEH_info_bn_mul4x_mont_gather5:
1173 .rva .Lmul4x_alloca,.Lmul4x_body,.Lmul4x_epilogue # HandlerData[]
1175 .LSEH_info_bn_gather5:
1176 .byte 0x01,0x0b,0x03,0x0a
1177 .byte 0x0b,0x01,0x21,0x00 # sub rsp,0x108
1178 .byte 0x04,0xa3,0x00,0x00 # lea r10,(rsp), set_frame r10
1183 $code =~ s/\`([^\`]*)\`/eval($1)/gem;