X-Git-Url: https://git.openssl.org/gitweb/?p=openssl.git;a=blobdiff_plain;f=crypto%2Fperlasm%2Fx86_64-xlate.pl;h=75092e058038dc62dd092445b2232a3e255d2197;hp=33ffc418a1e2dd19a237768273bfd9728f765fd3;hb=4f469342692d0b2804f466f82d6ad8fea604a647;hpb=1cfd258ed61721ef667ea8a6ca46b57f3765007e diff --git a/crypto/perlasm/x86_64-xlate.pl b/crypto/perlasm/x86_64-xlate.pl index 33ffc418a1..75092e0580 100755 --- a/crypto/perlasm/x86_64-xlate.pl +++ b/crypto/perlasm/x86_64-xlate.pl @@ -30,20 +30,55 @@ # stack frame allocation. If volatile storage is actually required # that is. If not, just leave the stack as is. # 3. Functions tagged with ".type name,@function" get crafted with -# unified Windows prologue and epilogue automatically. If you want +# unified Win64 prologue and epilogue automatically. If you want # to take care of ABI differences yourself, tag functions as -# ".type name,@abi-omnipotent." -# 4. As minor optimization you can specify number of input arguments -# as ".type name,@function,N." Keep in mind that if N is larger -# than 6, then you *have to* write "abi-omnipotent" code, because -# it can't be resolved with unified prologue. -# 5. Name local labels as .L*. -# 6. Don't use repret, it's generated automatically. +# ".type name,@abi-omnipotent" instead. +# 4. To optimize the Win64 prologue you can specify number of input +# arguments as ".type name,@function,N." Keep in mind that if N is +# larger than 6, then you *have to* write "abi-omnipotent" code, +# because >6 cases can't be addressed with unified prologue. +# 5. Name local labels as .L*, do *not* use dynamic labels such as 1: +# (sorry about latter). +# 6. Don't use [or hand-code with .byte] "rep ret." "ret" mnemonic is +# required to identify the spots, where to inject Win64 epilogue! +# But on the pros, it's then prefixed with rep automatically:-) +# 7. Due to MASM limitations [and certain general counter-intuitivity +# of ip-relative addressing] generation of position-independent +# code is assisted by synthetic directive, .picmeup, which puts +# address of the *next* instruction into target register. +# +# Example 1: +# .picmeup %rax +# lea .Label-.(%rax),%rax +# Example 2: +# .picmeup %rcx +# .Lpic_point: +# ... +# lea .Label-.Lpic_point(%rcx),%rbp 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 $masm=$masmref; +my $PTR=" PTR"; + +my $nasm=0; + +if ($win64) +{ if ($ENV{ASM} =~ m/nasm/) + { $nasm = 1; $PTR=""; } + elsif (`ml64 2>&1` =~ m/Version ([0-9]+)\.([0-9]+)(\.([0-9]+))?/) + { $masm = $1 + $2*2**-16 + $4*2**-32; } +} my $current_segment; my $current_function; @@ -54,16 +89,18 @@ my $current_function; 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/) { + $self->{sz} = "" + } elsif ($self->{op} =~ /([a-z]{3,})([qlwb])$/) { $self->{op} = $1; $self->{sz} = $2; } @@ -78,27 +115,32 @@ my $current_function; } 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} =~ /^j/ && $nasm) { + $self->{op} .= " NEAR"; } $self->{op}; } } + sub mnemonic { shift->{op}; } } { package const; # pick up constants, which start with $ sub re { @@ -115,7 +157,17 @@ my $current_function; } sub out { my $self = shift; - sprintf $masm?"%s":"\$%s",$self->{value}; + + if (!$win64) { + # Solaris /usr/ccs/bin/as can't handle multiplications + # in $self->{value} + $self->{value} =~ s/(?{value} =~ s/([0-9]+\s*[\*\/\%]\s*[0-9]+)/eval($1)/eg; + sprintf "\$%s",$self->{value}; + } else { + $self->{value} =~ s/0x([0-9a-f]+)/0$1h/ig; + sprintf "%s",$self->{value}; + } } } { package ea; # pick up effective addresses: expr(%reg,%reg,scale) @@ -131,7 +183,6 @@ my $current_function; $ret = $self; $line = substr($line,@+[0]); $line =~ s/^\s+//; - $self->{label} =~ s/\.L/\$L/g; $self->{base} =~ s/^%//; $self->{index} =~ s/^%// if (defined($self->{index})); } @@ -142,26 +193,43 @@ my $current_function; my $self = shift; my $sz = shift; - if (!$masm) { + # 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/(?{label} =~ s/([0-9]+\s*[\*\/\%]\s*[0-9]+)/eval($1)/eg; + if (defined($self->{index})) { - sprintf "%s(%%%s,%%%s,%d)", $self->{label},$self->{base}, + sprintf "%s(%%%s,%%%s,%d)", + $self->{label},$self->{base}, $self->{index},$self->{scale}; - } - else { + } else { sprintf "%s(%%%s)", $self->{label},$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/\./\$/g; + $self->{label} =~ s/0x([0-9a-f]+)/0$1h/ig; + $self->{label} = "($self->{label})" if ($self->{label} =~ /[\*\+\-\/]/); if (defined($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}, $self->{base}; - } - else { - sprintf "%s PTR %s[%s]",$szmap{$sz}, - $self->{label},$self->{base}; + } elsif ($self->{base} eq "rip") { + sprintf "%s[%s]",$szmap{$sz},$self->{label}; + } else { + sprintf "%s[%s%s]",$szmap{$sz}, + $self->{label}?"$self->{label}+":"", + $self->{base}; } } } @@ -198,7 +266,7 @@ my $current_function; } sub out { my $self = shift; - sprintf $masm?"%s":"%%%s",$self->{value}; + sprintf $win64?"%s":"%%%s",$self->{value}; } } { package label; # pick up labels, which end with : @@ -212,32 +280,32 @@ my $current_function; $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}:") { $self->{value}; } elsif ($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}".($nasm?":":"\tPROC")."\n". + " mov QWORD${PTR}[8+rsp],rdi\t;WIN64 prologue\n". + " mov QWORD${PTR}[16+rsp],rsi\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"); } } } @@ -252,7 +320,7 @@ my $current_function; $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; } @@ -267,13 +335,30 @@ my $current_function; local *line = shift; undef $ret; my $dir; + my %opcode = # lea 2f-1f(%rip),%dst; 1: nop; 2: + ( "%rax"=>0x01058d48, "%rcx"=>0x010d8d48, + "%rdx"=>0x01158d48, "%rbx"=>0x011d8d48, + "%rsp"=>0x01258d48, "%rbp"=>0x012d8d48, + "%rsi"=>0x01358d48, "%rdi"=>0x013d8d48, + "%r8" =>0x01058d4c, "%r9" =>0x010d8d4c, + "%r10"=>0x01158d4c, "%r11"=>0x011d8d4c, + "%r12"=>0x01258d4c, "%r13"=>0x012d8d4c, + "%r14"=>0x01358d4c, "%r15"=>0x013d8d4c ); if ($line =~ /^\s*(\.\w+)/) { - if (!$masm) { + if (!$win64) { $self->{value} = $1; $line =~ s/\@abi\-omnipotent/\@function/; $line =~ s/\@function.*/\@function/; - $self->{value} = $line; + if ($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; + } $line = ""; return $self; } @@ -283,31 +368,42 @@ my $current_function; undef $self->{value}; $line = substr($line,@+[0]); $line =~ s/^\s+//; SWITCH: for ($dir) { - /\.(text|data)/ + /\.(text)/ && 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"; + if ($nasm) { + $v ="section .$1 code align=64\n"; + $v.="default rel\n"; + $v.="%define PUBLIC global"; + } else { + $v="$current_segment\tENDS\n" if ($current_segment); + $current_segment = "_$1\$"; + $current_segment =~ tr/[a-z]/[A-Z]/; + $v.="$current_segment\tSEGMENT "; + $v.=$masm>=$masmref ? "ALIGN(64)" : "PAGE"; + $v.=" 'CODE'"; + } $self->{value} = $v; last; }; + /\.extern/ && do { $self->{value} = "EXTERN\t".$line; + $self->{value} .= ":BYTE" if (!$nasm); + last; + }; /\.globl/ && do { $self->{value} = "PUBLIC\t".$line; last; }; /\.type/ && do { ($sym,$type,$narg) = split(',',$line); - if ($type eq "\@function") - { undef $current_function; + if ($type eq "\@function") { + undef $current_function; $current_function->{name} = $sym; $current_function->{abi} = "svr4"; $current_function->{narg} = $narg; - } - elsif ($type eq "\@abi-omnipotent") - { undef $current_function; + } elsif ($type eq "\@abi-omnipotent") { + undef $current_function; $current_function->{name} = $sym; } last; }; - /\.size/ && do { if (defined($current_function)) - { $self->{value}="$current_function->{name}\tENDP"; + /\.size/ && do { if (defined($current_function)) { + $self->{value}="$current_function->{name}\tENDP" if(!$nasm); undef $current_function; } last; @@ -317,11 +413,31 @@ my $current_function; && do { my @arr = split(',',$line); my $sz = substr($1,0,1); my $last = pop(@arr); + my $conv = sub { my $var=shift; + if ($var=~s/0x([0-9a-f]+)/0$1h/i) { $var; } + else { sprintf"0%Xh",$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 { $self->{value} = sprintf"\tDD\t0%Xh,090000000h",$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; } + } + $self->{value}.="DB\t" + .join(",",@str) if (@str); + } last; }; } @@ -340,8 +456,9 @@ while($line=<>) { chomp($line); - $line =~ s/\[#!].*$//; # get rid of comments... - $line =~ s/^\s+//; # ... and skip white spaces + $line =~ s|[#!].*$||; # get rid of asm-style comments... + $line =~ s|/\*.*\*/||; # ... and C-style comments... + $line =~ s|^\s+||; # ... and skip white spaces in beginning undef $label; undef $opcode; @@ -373,16 +490,15 @@ while($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 { + } else { + undef $sz if ($nasm && $opcode->mnemonic() eq "lea"); printf "\t%s\t%s,%s", $opcode->out(), $dst->out($sz),$src->out($sz); } - } - elsif (defined($src)) { + } elsif (defined($src)) { printf "\t%s\t%s",$opcode->out(),$src->out($sz); } else { printf "\t%s",$opcode->out(); @@ -392,7 +508,7 @@ while($line=<>) { print $line,"\n"; } -print "\n$current_segment\tENDS\nEND\n" if ($masm); +print "\n$current_segment\tENDS\nEND\n" if ($current_segment); close STDOUT; @@ -433,7 +549,10 @@ close STDOUT; # 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, @@ -457,3 +576,120 @@ close STDOUT; # movq 16(%rsp),%rsi # endif # ret +# +################################################# +# 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: +# +# 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 +# +# 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; +# if (context->RipRax; +# else +# { 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 +# +# UNWIND_INFO structure for .xdata segment would be +# DB 9,0,0,0 +# DD imagerel handler +# denoting exception handler for a function with zero-length prologue, +# no stack frame or frame register. +# +# P.S. Attentive reader can notice that effectively no exceptions are +# expected in "gear" prologue and epilogue [discussed in "ABI +# cross-reference" above]. No, there are not. This is because if +# memory area used by them was subject to segmentation violation, +# then exception would be raised upon call to our function and be +# accounted to caller and unwound from its frame, which is not a +# problem. +# +# (*) 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).