#!/usr/bin/env perl
-# Ascetic x86_64 AT&T to MASM assembler translator by <appro>.
+# Ascetic x86_64 AT&T to MASM/NASM assembler translator by <appro>.
#
# Why AT&T to MASM and not vice versa? Several reasons. Because AT&T
# format is way easier to parse. Because it's simpler to "gear" from
# Currently recognized limitations:
#
# - can't use multiple ops per line;
-# - indirect calls and jumps are not supported;
#
# Dual-ABI styling rules.
#
-# 1. Adhere to Unix register and stack layout [see the end for
-# explanation].
+# 1. Adhere to Unix register and stack layout [see cross-reference
+# ABI "card" at the end for explanation].
# 2. Forget about "red zone," stick to more traditional blended
# stack frame allocation. If volatile storage is actually required
# that is. If not, just leave the stack as is.
# .Lpic_point:
# ...
# lea .Label-.Lpic_point(%rcx),%rbp
-
+#
+# 8. In order to provide for structured exception handling unified
+# Win64 prologue copies %rsp value to %rax. For further details
+# see SEH paragraph at the end.
+\f
my $output = shift;
-open STDOUT,">$output" || die "can't open $output: $!";
-my $masm=1 if ($output =~ /\.asm/);
+{ my ($stddev,$stdino,@junk)=stat(STDOUT);
+ my ($outdev,$outino,@junk)=stat($output);
+
+ open STDOUT,">$output" || die "can't open $output: $!"
+ if ($stddev!=$outdev || $stdino!=$outino);
+}
+
+my $win64=1 if ($output =~ /\.asm$/);
+
+my $masmref=8 + 50727*2**-32; # 8.00.50727 shipped with VS2005
+my $masm=0;
+my $PTR=" PTR";
+
+my $nasmref=2.03;
+my $nasm=0;
+
+if ($win64)
+{ if ($ENV{ASM} =~ m/nasm/ && `nasm -v` =~ m/version ([0-9]+)\.([0-9]+)/i)
+ { $nasm = $1 + $2*0.01; $PTR=""; }
+ elsif (`ml64 2>&1` =~ m/Version ([0-9]+)\.([0-9]+)(\.([0-9]+))?/)
+ { $masm = $1 + $2*2**-16 + $4*2**-32; }
+ die "no assembler found on %PATH" if (!($nasm || $masm));
+}
my $current_segment;
my $current_function;
+my %globals;
{ package opcode; # pick up opcodes
sub re {
local *line = shift;
undef $ret;
- if ($line =~ /^([a-z]+)/i) {
+ if ($line =~ /^([a-z][a-z0-9]*)/i) {
$self->{op} = $1;
$ret = $self;
$line = substr($line,@+[0]); $line =~ s/^\s+//;
undef $self->{sz};
- if ($self->{op} =~ /(movz)b.*/) { # movz is pain...
+ if ($self->{op} =~ /^(movz)b.*/) { # movz is pain...
$self->{op} = $1;
$self->{sz} = "b";
- } elsif ($self->{op} =~ /([a-z]{3,})([qlwb])/) {
+ } elsif ($self->{op} =~ /call|jmp/) {
+ $self->{sz} = ""
+ } elsif ($self->{op} =~ /([a-z]{3,})([qlwb])$/) {
$self->{op} = $1;
$self->{sz} = $2;
}
}
sub out {
my $self = shift;
- if (!$masm) {
- if ($self->{op} eq "movz") { # movz in pain...
+ if (!$win64) {
+ if ($self->{op} eq "movz") { # movz is pain...
sprintf "%s%s%s",$self->{op},$self->{sz},shift;
+ } elsif ($self->{op} =~ /^set/) {
+ "$self->{op}";
} elsif ($self->{op} eq "ret") {
".byte 0xf3,0xc3";
} else {
"$self->{op}$self->{sz}";
}
} else {
- $self->{op} =~ s/movz/movzx/;
+ $self->{op} =~ s/^movz/movzx/;
if ($self->{op} eq "ret") {
$self->{op} = "";
if ($current_function->{abi} eq "svr4") {
- $self->{op} = "mov rdi,QWORD PTR 8[rsp]\t;WIN64 epilogue\n\t".
- "mov rsi,QWORD PTR 16[rsp]\n\t";
+ $self->{op} = "mov rdi,QWORD${PTR}[8+rsp]\t;WIN64 epilogue\n\t".
+ "mov rsi,QWORD${PTR}[16+rsp]\n\t";
}
$self->{op} .= "DB\t0F3h,0C3h\t\t;repret";
}
+ elsif ($self->{op} =~ /^(pop|push)f/) {
+ $self->{op} .= $self->{sz};
+ }
$self->{op};
}
}
+ sub mnemonic { shift->{op}; }
}
{ package const; # pick up constants, which start with $
sub re {
sub out {
my $self = shift;
- if (!$masm) {
+ if (!$win64) {
+ # Solaris /usr/ccs/bin/as can't handle multiplications
+ # in $self->{value}
+ $self->{value} =~ s/(?<![0-9a-f])(0[x0-9a-f]+)/oct($1)/egi;
+ $self->{value} =~ s/([0-9]+\s*[\*\/\%]\s*[0-9]+)/eval($1)/eg;
sprintf "\$%s",$self->{value};
} else {
$self->{value} =~ s/0x([0-9a-f]+)/0$1h/ig;
local *line = shift;
undef $ret;
- if ($line =~ /^([^\(,]*)\(([%\w,]+)\)/) {
- $self->{label} = $1;
- ($self->{base},$self->{index},$self->{scale})=split(/,/,$2);
+ # optional * ---vvv--- appears in indirect jmp/call
+ if ($line =~ /^(\*?)([^\(,]*)\(([%\w,]+)\)/) {
+ $self->{asterisk} = $1;
+ $self->{label} = $2;
+ ($self->{base},$self->{index},$self->{scale})=split(/,/,$3);
$self->{scale} = 1 if (!defined($self->{scale}));
$ret = $self;
$line = substr($line,@+[0]); $line =~ s/^\s+//;
my $self = shift;
my $sz = shift;
- if (!$masm) {
- # elder GNU assembler insists on 64-bit EAs:-(
- # on pros side, this results in more compact code:-)
- $self->{index} =~ s/^[er](.?[0-9xp])[d]?$/r\1/;
- $self->{base} =~ s/^[er](.?[0-9xp])[d]?$/r\1/;
+ # Silently convert all EAs to 64-bit. This is required for
+ # elder GNU assembler and results in more compact code,
+ # *but* most importantly AES module depends on this feature!
+ $self->{index} =~ s/^[er](.?[0-9xpi])[d]?$/r\1/;
+ $self->{base} =~ s/^[er](.?[0-9xpi])[d]?$/r\1/;
+
+ if (!$win64) {
+ # Solaris /usr/ccs/bin/as can't handle multiplications
+ # in $self->{label}
+ $self->{label} =~ s/(?<![0-9a-f])(0[x0-9a-f]+)/oct($1)/egi;
+ $self->{label} =~ s/([0-9]+\s*[\*\/\%]\s*[0-9]+)/eval($1)/eg;
if (defined($self->{index})) {
- sprintf "%s(%%%s,%%%s,%d)",
+ sprintf "%s%s(%%%s,%%%s,%d)",$self->{asterisk},
$self->{label},$self->{base},
$self->{index},$self->{scale};
} else {
- sprintf "%s(%%%s)", $self->{label},$self->{base};
+ sprintf "%s%s(%%%s)", $self->{asterisk},$self->{label},$self->{base};
}
} else {
- %szmap = ( b=>"BYTE", w=>"WORD", l=>"DWORD", q=>"QWORD" );
+ %szmap = ( b=>"BYTE$PTR", w=>"WORD$PTR", l=>"DWORD$PTR", q=>"QWORD$PTR" );
+ $self->{label} =~ s/\.L/\$L\$/g;
$self->{label} =~ s/\./\$/g;
$self->{label} =~ s/0x([0-9a-f]+)/0$1h/ig;
$self->{label} = "($self->{label})" if ($self->{label} =~ /[\*\+\-\/]/);
+ $sz="q" if ($self->{asterisk});
if (defined($self->{index})) {
- sprintf "%s PTR %s[%s*%d+%s]",$szmap{$sz},
- $self->{label},
+ sprintf "%s[%s%s*%d+%s]",$szmap{$sz},
+ $self->{label}?"$self->{label}+":"",
$self->{index},$self->{scale},
$self->{base};
+ } elsif ($self->{base} eq "rip") {
+ sprintf "%s[%s]",$szmap{$sz},$self->{label};
} else {
- sprintf "%s PTR %s[%s]",$szmap{$sz},
- $self->{label},$self->{base};
+ sprintf "%s[%s%s]",$szmap{$sz},
+ $self->{label}?"$self->{label}+":"",
+ $self->{base};
}
}
}
local *line = shift;
undef $ret;
- if ($line =~ /^%(\w+)/) {
+ # optional * ---vvv--- appears in indirect jmp/call
+ if ($line =~ /^(\*?)%(\w+)/) {
bless $self,$class;
- $self->{value} = $1;
+ $self->{asterisk} = $1;
+ $self->{value} = $2;
$ret = $self;
$line = substr($line,@+[0]); $line =~ s/^\s+//;
}
}
sub out {
my $self = shift;
- sprintf $masm?"%s":"%%%s",$self->{value};
+ if (!$win64) { sprintf "%s%%%s",$self->{asterisk},$self->{value}; }
+ else { $self->{value}; }
}
}
{ package label; # pick up labels, which end with :
$ret = $self;
$line = substr($line,@+[0]); $line =~ s/^\s+//;
- $self->{value} =~ s/\.L/\$L/ if ($masm);
+ $self->{value} =~ s/\.L/\$L\$/ if ($win64);
}
$ret;
}
sub out {
my $self = shift;
- if (!$masm) {
+ if (!$win64) {
$self->{value};
} elsif ($self->{value} ne "$current_function->{name}:") {
+ $self->{value} .= ":" if ($masm && $ret!~m/^\$/);
$self->{value};
} elsif ($current_function->{abi} eq "svr4") {
- my $func = "$current_function->{name} PROC\n".
- " mov QWORD PTR 8[rsp],rdi\t;WIN64 prologue\n".
- " mov QWORD PTR 16[rsp],rsi\n";
+ my $func = "$current_function->{name}" .
+ ($nasm ? ":" : "\tPROC $current_function->{scope}") .
+ "\n";
+ $func .= " mov QWORD${PTR}[8+rsp],rdi\t;WIN64 prologue\n";
+ $func .= " mov QWORD${PTR}[16+rsp],rsi\n";
+ $func .= " mov rax,rsp\n";
+ $func .= "\$L\$SEH_begin_$current_function->{name}:";
+ $func .= ":" if ($masm);
+ $func .= "\n";
my $narg = $current_function->{narg};
$narg=6 if (!defined($narg));
$func .= " mov rdi,rcx\n" if ($narg>0);
$func .= " mov rsi,rdx\n" if ($narg>1);
$func .= " mov rdx,r8\n" if ($narg>2);
$func .= " mov rcx,r9\n" if ($narg>3);
- $func .= " mov r8,QWORD PTR 40[rsp]\n" if ($narg>4);
- $func .= " mov r9,QWORD PTR 48[rsp]\n" if ($narg>5);
+ $func .= " mov r8,QWORD${PTR}[40+rsp]\n" if ($narg>4);
+ $func .= " mov r9,QWORD${PTR}[48+rsp]\n" if ($narg>5);
$func .= "\n";
} else {
- "$current_function->{name} PROC";
+ "$current_function->{name}".
+ ($nasm ? ":" : "\tPROC $current_function->{scope}");
}
}
}
$ret = $self;
$line = substr($line,@+[0]); $line =~ s/^\s+//;
- $self->{value} =~ s/\.L/\$L/g if ($masm);
+ $self->{value} =~ s/\.L/\$L\$/g if ($win64);
}
$ret;
}
sub out {
my $self = shift;
- $self->{value};
+ if ($nasm && opcode->mnemonic()=~m/^j/) {
+ "NEAR ".$self->{value};
+ } else {
+ $self->{value};
+ }
}
}
{ package directive; # pick up directives, which start with .
"%r14"=>0x01358d4c, "%r15"=>0x013d8d4c );
if ($line =~ /^\s*(\.\w+)/) {
- if (!$masm) {
+ if (!$win64) {
$self->{value} = $1;
$line =~ s/\@abi\-omnipotent/\@function/;
$line =~ s/\@function.*/\@function/;
if ($line =~ /\.picmeup\s+(%r[\w]+)/i) {
$self->{value} = sprintf "\t.long\t0x%x,0x90000000",$opcode{$1};
+ } elsif ($line =~ /\.asciz\s+"(.*)"$/) {
+ $self->{value} = ".byte\t".join(",",unpack("C*",$1),0);
+ } elsif ($line =~ /\.extern/) {
+ $self->{value} = ""; # swallow extern
} else {
$self->{value} = $line;
}
undef $self->{value};
$line = substr($line,@+[0]); $line =~ s/^\s+//;
SWITCH: for ($dir) {
- /\.(text|data)/
- && do { my $v=undef;
- $v="$current_segment\tENDS\n" if ($current_segment);
- $current_segment = "_$1";
- $current_segment =~ tr/[a-z]/[A-Z]/;
- $v.="$current_segment\tSEGMENT PARA";
+ /\.text/ && do { my $v=undef;
+ if ($nasm) {
+ $v="section .text code align=64\n";
+ } else {
+ $v="$current_segment\tENDS\n" if ($current_segment);
+ $current_segment = ".text\$";
+ $v.="$current_segment\tSEGMENT ";
+ $v.=$masm>=$masmref ? "ALIGN(64)" : "PAGE";
+ $v.=" 'CODE'";
+ }
+ $self->{value} = $v;
+ last;
+ };
+ /\.data/ && do { my $v=undef;
+ if ($nasm) {
+ $v="section .data data align=8\n";
+ } else {
+ $v="$current_segment\tENDS\n" if ($current_segment);
+ $current_segment = "_DATA";
+ $v.="$current_segment\tSEGMENT";
+ }
$self->{value} = $v;
last;
};
- /\.globl/ && do { $self->{value} = "PUBLIC\t".$line; last; };
+ /\.section/ && do { my $v=undef;
+ if ($nasm) {
+ $v="section $line";
+ if ($line=~/\.([px])data/) {
+ $v.=" rdata align=";
+ $v.=$1 eq "p"? 4 : 8;
+ }
+ } else {
+ $v="$current_segment\tENDS\n" if ($current_segment);
+ $v.="$line\tSEGMENT";
+ if ($line=~/\.([px])data/) {
+ $v.=" READONLY";
+ $v.=" ALIGN(".($1 eq "p" ? 4 : 8).")" if ($masm>=$masmref);
+ }
+ }
+ $current_segment = $line;
+ $self->{value} = $v;
+ last;
+ };
+ /\.extern/ && do { $self->{value} = "EXTERN\t".$line;
+ $self->{value} .= ":NEAR" if ($masm);
+ last;
+ };
+ /\.globl|.global/
+ && do { $self->{value} = $masm?"PUBLIC":"global";
+ $self->{value} .= "\t".$line;
+ $globals{$line} = $line;
+ last;
+ };
/\.type/ && do { ($sym,$type,$narg) = split(',',$line);
if ($type eq "\@function") {
undef $current_function;
$current_function->{name} = $sym;
$current_function->{abi} = "svr4";
$current_function->{narg} = $narg;
+ $current_function->{scope} = defined($globals{$sym})?"PUBLIC":"PRIVATE";
} elsif ($type eq "\@abi-omnipotent") {
undef $current_function;
$current_function->{name} = $sym;
+ $current_function->{scope} = defined($globals{$sym})?"PUBLIC":"PRIVATE";
}
last;
};
/\.size/ && do { if (defined($current_function)) {
- $self->{value}="$current_function->{name}\tENDP";
+ undef $self->{value};
+ if ($current_function->{abi} eq "svr4") {
+ $self->{value}="\$L\$SEH_end_$current_function->{name}:";
+ $self->{value}.=":\n" if($masm);
+ }
+ $self->{value}.="$current_function->{name}\tENDP" if($masm);
undef $current_function;
}
last;
};
/\.align/ && do { $self->{value} = "ALIGN\t".$line; last; };
/\.(byte|value|long|quad)/
- && do { my @arr = split(',',$line);
- my $sz = substr($1,0,1);
+ && do { my $sz = substr($1,0,1);
+ my @arr = split(',',$line);
my $last = pop(@arr);
+ my $conv = sub { my $var=shift;
+ $var=~s/0x([0-9a-f]+)/0$1h/ig;
+ $var=~s/\.L/\$L\$/g;
+ if ($current_segment=~/.[px]data/)
+ { $var=~s/([_a-z\$\@][_a-z0-9\$\@]*)/$nasm?"$1 wrt ..imagebase":"imagerel $1"/egi; }
+ $var;
+ };
$sz =~ tr/bvlq/BWDQ/;
$self->{value} = "\tD$sz\t";
- for (@arr) { $self->{value} .= sprintf"0%Xh,",oct; }
- $self->{value} .= sprintf"0%Xh",oct($last);
+ for (@arr) { $self->{value} .= &$conv($_).","; }
+ $self->{value} .= &$conv($last);
last;
};
- /\.picmeup/ && do { $self->{value} = sprintf"\tDD\t 0%Xh,090000000h",$opcode{$line};
+ /\.picmeup/ && do { $self->{value} = sprintf"\tDD\t0%Xh,090000000h",$opcode{$line};
+ last;
+ };
+ /\.asciz/ && do { if ($line =~ /^"(.*)"$/) {
+ my @str=unpack("C*",$1);
+ push @str,0;
+ while ($#str>15) {
+ $self->{value}.="DB\t"
+ .join(",",@str[0..15])."\n";
+ foreach (0..15) { shift @str; }
+ }
+ $self->{value}.="DB\t"
+ .join(",",@str) if (@str);
+ }
last;
};
}
}
}
+if ($nasm) {
+ print <<___;
+default rel
+___
+} elsif ($masm) {
+ print <<___;
+OPTION DOTNAME
+___
+}
while($line=<>) {
chomp($line);
$sz=opcode->size();
if (defined($dst)) {
- if (!$masm) {
+ if (!$win64) {
printf "\t%s\t%s,%s", $opcode->out($dst->size()),
$src->out($sz),$dst->out($sz);
} else {
+ undef $sz if ($nasm && $opcode->mnemonic() eq "lea");
printf "\t%s\t%s,%s", $opcode->out(),
$dst->out($sz),$src->out($sz);
}
print $line,"\n";
}
-print "\n$current_segment\tENDS\nEND\n" if ($masm);
+print "\n$current_segment\tENDS\nEND\n" if ($current_segment && $masm);
close STDOUT;
-#################################################
+\f#################################################
# Cross-reference x86_64 ABI "card"
#
# Unix Win64
# arguments passed to callee, *but* not less than 4! This means that
# upon function entry point 5th argument resides at 40(%rsp), as well
# as that 32 bytes from 8(%rsp) can always be used as temporal
-# storage [without allocating a frame].
+# storage [without allocating a frame]. One can actually argue that
+# one can assume a "red zone" above stack pointer under Win64 as well.
+# Point is that at apparently no occasion Windows kernel would alter
+# the area above user stack pointer in true asynchronous manner...
#
# All the above means that if assembler programmer adheres to Unix
# register and stack layout, but disregards the "red zone" existense,
# movq 16(%rsp),%rsi
# endif
# ret
+#
+\f#################################################
+# Win64 SEH, Structured Exception Handling.
+#
+# Unlike on Unix systems(*) lack of Win64 stack unwinding information
+# has undesired side-effect at run-time: if an exception is raised in
+# assembler subroutine such as those in question (basically we're
+# referring to segmentation violations caused by malformed input
+# parameters), the application is briskly terminated without invoking
+# any exception handlers, most notably without generating memory dump
+# or any user notification whatsoever. This poses a problem. It's
+# possible to address it by registering custom language-specific
+# handler that would restore processor context to the state at
+# subroutine entry point and return "exception is not handled, keep
+# unwinding" code. Writing such handler can be a challenge... But it's
+# doable, though requires certain coding convention. Consider following
+# snippet:
+#
+# .type function,@function
+# function:
+# movq %rsp,%rax # copy rsp to volatile register
+# pushq %r15 # save non-volatile registers
+# pushq %rbx
+# pushq %rbp
+# movq %rsp,%r11
+# subq %rdi,%r11 # prepare [variable] stack frame
+# andq $-64,%r11
+# movq %rax,0(%r11) # check for exceptions
+# movq %r11,%rsp # allocate [variable] stack frame
+# movq %rax,0(%rsp) # save original rsp value
+# magic_point:
+# ...
+# movq 0(%rsp),%rcx # pull original rsp value
+# movq -24(%rcx),%rbp # restore non-volatile registers
+# movq -16(%rcx),%rbx
+# movq -8(%rcx),%r15
+# movq %rcx,%rsp # restore original rsp
+# ret
+# .size function,.-function
+#
+# The key is that up to magic_point copy of original rsp value remains
+# in chosen volatile register and no non-volatile register, except for
+# rsp, is modified. While past magic_point rsp remains constant till
+# the very end of the function. In this case custom language-specific
+# exception handler would look like this:
+#
+# EXCEPTION_DISPOSITION handler (EXCEPTION_RECORD *rec,ULONG64 frame,
+# CONTEXT *context,DISPATCHER_CONTEXT *disp)
+# { ULONG64 *rsp = (ULONG64 *)context->Rax;
+# if (context->Rip >= magic_point)
+# { rsp = ((ULONG64 **)context->Rsp)[0];
+# context->Rbp = rsp[-3];
+# context->Rbx = rsp[-2];
+# context->R15 = rsp[-1];
+# }
+# context->Rsp = (ULONG64)rsp;
+# context->Rdi = rsp[1];
+# context->Rsi = rsp[2];
+#
+# memcpy (disp->ContextRecord,context,sizeof(CONTEXT));
+# RtlVirtualUnwind(UNW_FLAG_NHANDLER,disp->ImageBase,
+# dips->ControlPc,disp->FunctionEntry,disp->ContextRecord,
+# &disp->HandlerData,&disp->EstablisherFrame,NULL);
+# return ExceptionContinueSearch;
+# }
+#
+# It's appropriate to implement this handler in assembler, directly in
+# function's module. In order to do that one has to know members'
+# offsets in CONTEXT and DISPATCHER_CONTEXT structures and some constant
+# values. Here they are:
+#
+# CONTEXT.Rax 120
+# CONTEXT.Rcx 128
+# CONTEXT.Rdx 136
+# CONTEXT.Rbx 144
+# CONTEXT.Rsp 152
+# CONTEXT.Rbp 160
+# CONTEXT.Rsi 168
+# CONTEXT.Rdi 176
+# CONTEXT.R8 184
+# CONTEXT.R9 192
+# CONTEXT.R10 200
+# CONTEXT.R11 208
+# CONTEXT.R12 216
+# CONTEXT.R13 224
+# CONTEXT.R14 232
+# CONTEXT.R15 240
+# CONTEXT.Rip 248
+# sizeof(CONTEXT) 1232
+# DISPATCHER_CONTEXT.ControlPc 0
+# DISPATCHER_CONTEXT.ImageBase 8
+# DISPATCHER_CONTEXT.FunctionEntry 16
+# DISPATCHER_CONTEXT.EstablisherFrame 24
+# DISPATCHER_CONTEXT.TargetIp 32
+# DISPATCHER_CONTEXT.ContextRecord 40
+# DISPATCHER_CONTEXT.LanguageHandler 48
+# DISPATCHER_CONTEXT.HandlerData 56
+# UNW_FLAG_NHANDLER 0
+# ExceptionContinueSearch 1
+#
+# In order to tie the handler to the function one has to compose
+# couple of structures: one for .xdata segment and one for .pdata.
+#
+# UNWIND_INFO structure for .xdata segment would be
+#
+# function_unwind_info:
+# .byte 9,0,0,0
+# .long handler
+#
+# This structure designates exception handler for a function with
+# zero-length prologue, no stack frame or frame register.
+#
+# To facilitate composing of .pdata structures, auto-generated "gear"
+# prologue copies rsp value to rax and denotes next instruction with
+# $L$SEH_begin_{function_name} label. This essentially defines the SEH
+# styling rule mentioned in the beginning. Position of this label is
+# chosen in such manner that possible exceptions raised in the "gear"
+# prologue would be accounted to caller and unwound from latter's frame.
+# End of function is marked with respective $L$SEH_end_{function_name}
+# label. To summarize, .pdata segment would contain
+#
+# .long $L$SEH_begin_function
+# .long $L$SEH_end_function
+# .long function_unwind_info
+#
+# Reference to functon_unwind_info from .xdata segment is the anchor.
+# In case you wonder why references are 32-bit .longs and not 64-bit
+# .quads. References put into these two segments are required to be
+# *relative* to the base address of the current binary module, a.k.a.
+# image base. No Win64 module, be it .exe or .dll, can be larger than
+# 2GB and thus such relative references can be and are accommodated in
+# 32 bits.
+#
+# Having reviewed the example function code, one can argue that "movq
+# %rsp,%rax" above is redundant. It is not! Keep in mind that on Unix
+# rax would contain an undefined value. If this "offends" you, use
+# another register and refrain from modifying rax till magic_point is
+# reached, i.e. as if it was a non-volatile register. If more registers
+# are required prior [variable] frame setup is completed, note that
+# nobody says that you can have only one "magic point." You can
+# "liberate" non-volatile registers by denoting last stack off-load
+# instruction and reflecting it in finer grade unwind logic in handler.
+# After all, isn't it why it's called *language-specific* handler...
+#
+# Attentive reader can notice that exceptions would be mishandled in
+# auto-generated "gear" epilogue. Well, exception effectively can't
+# occur there, because if memory area used by it was subject to
+# segmentation violation, then it would be raised upon call to the
+# function (and as already mentioned be accounted to caller, which is
+# not a problem). If you're still not comfortable, then define tail
+# "magic point" just prior ret instruction and have handler treat it...
+#
+# (*) Note that we're talking about run-time, not debug-time. Lack of
+# unwind information makes debugging hard on both Windows and
+# Unix. "Unlike" referes to the fact that on Unix signal handler
+# will always be invoked, core dumped and appropriate exit code
+# returned to parent (for user notification).