1 files changed, 126 insertions, 10 deletions

diff --git a/src/lib/libcrypto/bn/asm/x86_64-mont.pl b/src/lib/libcrypto/bn/asm/x86_64-mont.pl
index c43b69592a..3b7a6f243f 100755
--- a/src/lib/libcrypto/bn/asm/x86_64-mont.pl
+++ b/src/lib/libcrypto/bn/asm/x86_64-mont.pl
@@ -15,14 +15,18 @@
 # respectful 50%. It remains to be seen if loop unrolling and
 # dedicated squaring routine can provide further improvement...
 
-$output=shift;
+$flavour = shift;
+$output  = shift;
+if ($flavour =~ /\./) { $output = $flavour; undef $flavour; }
+
+$win64=0; $win64=1 if ($flavour =~ /[nm]asm|mingw64/ || $output =~ /\.asm$/);
 
 $0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
 ( $xlate="${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 STDOUT,"| $^X $xlate $output";
+open STDOUT,"| $^X $xlate $flavour $output";
 
 # int bn_mul_mont(
 $rp="%rdi";     # BN_ULONG *rp,
@@ -55,13 +59,14 @@ bn_mul_mont:
         push    %r15
 
         mov     ${num}d,${num}d
-        lea     2($num),%rax
-        mov     %rsp,%rbp
-        neg     %rax
-        lea     (%rsp,%rax,8),%rsp      # tp=alloca(8*(num+2))
+        lea     2($num),%r10
+        mov     %rsp,%r11
+        neg     %r10
+        lea     (%rsp,%r10,8),%rsp      # tp=alloca(8*(num+2))
         and     \$-1024,%rsp            # minimize TLB usage
 
-        mov     %rbp,8(%rsp,$num,8)     # tp[num+1]=%rsp
+        mov     %r11,8(%rsp,$num,8)     # tp[num+1]=%rsp
+.Lprologue:
         mov     %rdx,$bp                # $bp reassigned, remember?
 
         mov     ($n0),$n0               # pull n0[0] value
@@ -197,18 +202,129 @@ bn_mul_mont:
         dec     $j
         jge     .Lcopy
 
-        mov     8(%rsp,$num,8),%rsp     # restore %rsp
+        mov     8(%rsp,$num,8),%rsi     # restore %rsp
         mov     \$1,%rax
+        mov     (%rsi),%r15
+        mov     8(%rsi),%r14
+        mov     16(%rsi),%r13
+        mov     24(%rsi),%r12
+        mov     32(%rsi),%rbp
+        mov     40(%rsi),%rbx
+        lea     48(%rsi),%rsp
+.Lepilogue:
+        ret
+.size   bn_mul_mont,.-bn_mul_mont
+.asciz  "Montgomery Multiplication for x86_64, CRYPTOGAMS by <appro\@openssl.org>"
+.align  16
+___
+
+# EXCEPTION_DISPOSITION handler (EXCEPTION_RECORD *rec,ULONG64 frame,
+#               CONTEXT *context,DISPATCHER_CONTEXT *disp)
+if ($win64) {
+$rec="%rcx";
+$frame="%rdx";
+$context="%r8";
+$disp="%r9";
+
+$code.=<<___;
+.extern __imp_RtlVirtualUnwind
+.type   se_handler,\@abi-omnipotent
+.align  16
+se_handler:
+        push    %rsi
+        push    %rdi
+        push    %rbx
+        push    %rbp
+        push    %r12
+        push    %r13
+        push    %r14
+        push    %r15
+        pushfq
+        sub     \$64,%rsp
+
+        mov     120($context),%rax      # pull context->Rax
+        mov     248($context),%rbx      # pull context->Rip
+
+        lea     .Lprologue(%rip),%r10
+        cmp     %r10,%rbx               # context->Rip<.Lprologue
+        jb      .Lin_prologue
+
+        mov     152($context),%rax      # pull context->Rsp
+
+        lea     .Lepilogue(%rip),%r10
+        cmp     %r10,%rbx               # context->Rip>=.Lepilogue
+        jae     .Lin_prologue
+
+        mov     192($context),%r10      # pull $num
+        mov     8(%rax,%r10,8),%rax     # pull saved stack pointer
+        lea     48(%rax),%rax
+
+        mov     -8(%rax),%rbx
+        mov     -16(%rax),%rbp
+        mov     -24(%rax),%r12
+        mov     -32(%rax),%r13
+        mov     -40(%rax),%r14
+        mov     -48(%rax),%r15
+        mov     %rbx,144($context)      # restore context->Rbx
+        mov     %rbp,160($context)      # restore context->Rbp
+        mov     %r12,216($context)      # restore context->R12
+        mov     %r13,224($context)      # restore context->R13
+        mov     %r14,232($context)      # restore context->R14
+        mov     %r15,240($context)      # restore context->R15
+
+.Lin_prologue:
+        mov     8(%rax),%rdi
+        mov     16(%rax),%rsi
+        mov     %rax,152($context)      # restore context->Rsp
+        mov     %rsi,168($context)      # restore context->Rsi
+        mov     %rdi,176($context)      # restore context->Rdi
+
+        mov     40($disp),%rdi          # disp->ContextRecord
+        mov     $context,%rsi           # context
+        mov     \$154,%ecx              # sizeof(CONTEXT)
+        .long   0xa548f3fc              # cld; rep movsq
+
+        mov     $disp,%rsi
+        xor     %rcx,%rcx               # arg1, UNW_FLAG_NHANDLER
+        mov     8(%rsi),%rdx            # arg2, disp->ImageBase
+        mov     0(%rsi),%r8             # arg3, disp->ControlPc
+        mov     16(%rsi),%r9            # arg4, disp->FunctionEntry
+        mov     40(%rsi),%r10           # disp->ContextRecord
+        lea     56(%rsi),%r11           # &disp->HandlerData
+        lea     24(%rsi),%r12           # &disp->EstablisherFrame
+        mov     %r10,32(%rsp)           # arg5
+        mov     %r11,40(%rsp)           # arg6
+        mov     %r12,48(%rsp)           # arg7
+        mov     %rcx,56(%rsp)           # arg8, (NULL)
+        call    *__imp_RtlVirtualUnwind(%rip)
+
+        mov     \$1,%eax                # ExceptionContinueSearch
+        add     \$64,%rsp
+        popfq
         pop     %r15
         pop     %r14
         pop     %r13
         pop     %r12
         pop     %rbp
         pop     %rbx
+        pop     %rdi
+        pop     %rsi
         ret
-.size   bn_mul_mont,.-bn_mul_mont
-.asciz  "Montgomery Multiplication for x86_64, CRYPTOGAMS by <appro\@openssl.org>"
+.size   se_handler,.-se_handler
+
+.section        .pdata
+.align  4
+        .rva    .LSEH_begin_bn_mul_mont
+        .rva    .LSEH_end_bn_mul_mont
+        .rva    .LSEH_info_bn_mul_mont
+
+.section        .xdata
+.align  8
+.LSEH_info_bn_mul_mont:
+        .byte   9,0,0,0
+        .rva    se_handler
 ___
+}
 
 print $code;
 close STDOUT;