minwin-binutils-gdb/ld/ldint.texinfo

\input texinfo
@setfilename ldint.info
@c Copyright (C) 1992-2015 Free Software Foundation, Inc.

@ifnottex
@dircategory Software development
@direntry
* Ld-Internals: (ldint).	The GNU linker internals.
@end direntry
@end ifnottex

@copying
This file documents the internals of the GNU linker ld.

Copyright @copyright{} 1992-2015 Free Software Foundation, Inc.
Contributed by Cygnus Support.

Permission is granted to copy, distribute and/or modify this document
under the terms of the GNU Free Documentation License, Version 1.3 or
any later version published by the Free Software Foundation; with the
Invariant Sections being ``GNU General Public License'' and ``Funding
Free Software'', the Front-Cover texts being (a) (see below), and with
the Back-Cover Texts being (b) (see below).  A copy of the license is
included in the section entitled ``GNU Free Documentation License''.

(a) The FSF's Front-Cover Text is:

     A GNU Manual

(b) The FSF's Back-Cover Text is:

     You have freedom to copy and modify this GNU Manual, like GNU
     software.  Copies published by the Free Software Foundation raise
     funds for GNU development.
@end copying

@iftex
@finalout
@setchapternewpage off
@settitle GNU Linker Internals
@titlepage
@title{A guide to the internals of the GNU linker}
@author Per Bothner, Steve Chamberlain, Ian Lance Taylor, DJ Delorie
@author Cygnus Support
@page

@tex
\def\$#1${{#1}}  % Kluge: collect RCS revision info without $...$
\xdef\manvers{2.10.91}  % For use in headers, footers too
{\parskip=0pt
\hfill Cygnus Support\par
\hfill \manvers\par
\hfill \TeX{}info \texinfoversion\par
}
@end tex

@vskip 0pt plus 1filll
Copyright @copyright{} 1992-2015 Free Software Foundation, Inc.

      Permission is granted to copy, distribute and/or modify this document
      under the terms of the GNU Free Documentation License, Version 1.3
      or any later version published by the Free Software Foundation;
      with no Invariant Sections, with no Front-Cover Texts, and with no
      Back-Cover Texts.  A copy of the license is included in the
      section entitled "GNU Free Documentation License".

@end titlepage
@end iftex

@node Top
@top

This file documents the internals of the GNU linker @code{ld}.  It is a
collection of miscellaneous information with little form at this point.
Mostly, it is a repository into which you can put information about
GNU @code{ld} as you discover it (or as you design changes to @code{ld}).

This document is distributed under the terms of the GNU Free
Documentation License.  A copy of the license is included in the
section entitled "GNU Free Documentation License".

@menu
* README::			The README File
* Emulations::			How linker emulations are generated
* Emulation Walkthrough::	A Walkthrough of a Typical Emulation
* Architecture Specific::	Some Architecture Specific Notes
* GNU Free Documentation License::  GNU Free Documentation License
@end menu

@node README
@chapter The @file{README} File

Check the @file{README} file; it often has useful information that does not
appear anywhere else in the directory.

@node Emulations
@chapter How linker emulations are generated

Each linker target has an @dfn{emulation}.  The emulation includes the
default linker script, and certain emulations also modify certain types
of linker behaviour.

Emulations are created during the build process by the shell script
@file{genscripts.sh}.

The @file{genscripts.sh} script starts by reading a file in the
@file{emulparams} directory.  This is a shell script which sets various
shell variables used by @file{genscripts.sh} and the other shell scripts
it invokes.

The @file{genscripts.sh} script will invoke a shell script in the
@file{scripttempl} directory in order to create default linker scripts
written in the linker command language.  The @file{scripttempl} script
will be invoked 5 (or, in some cases, 6) times, with different
assignments to shell variables, to create different default scripts.
The choice of script is made based on the command line options.

After creating the scripts, @file{genscripts.sh} will invoke yet another
shell script, this time in the @file{emultempl} directory.  That shell
script will create the emulation source file, which contains C code.
This C code permits the linker emulation to override various linker
behaviours.  Most targets use the generic emulation code, which is in
@file{emultempl/generic.em}.

To summarize, @file{genscripts.sh} reads three shell scripts: an
emulation parameters script in the @file{emulparams} directory, a linker
script generation script in the @file{scripttempl} directory, and an
emulation source file generation script in the @file{emultempl}
directory.

For example, the Sun 4 linker sets up variables in
@file{emulparams/sun4.sh}, creates linker scripts using
@file{scripttempl/aout.sc}, and creates the emulation code using
@file{emultempl/sunos.em}.

Note that the linker can support several emulations simultaneously,
depending upon how it is configured.  An emulation can be selected with
the @code{-m} option.  The @code{-V} option will list all supported
emulations.

@menu
* emulation parameters::        @file{emulparams} scripts
* linker scripts::              @file{scripttempl} scripts
* linker emulations::           @file{emultempl} scripts
@end menu

@node emulation parameters
@section @file{emulparams} scripts

Each target selects a particular file in the @file{emulparams} directory
by setting the shell variable @code{targ_emul} in @file{configure.tgt}.
This shell variable is used by the @file{configure} script to control
building an emulation source file.

Certain conventions are enforced.  Suppose the @code{targ_emul} variable
is set to @var{emul} in @file{configure.tgt}.  The name of the emulation
shell script will be @file{emulparams/@var{emul}.sh}.  The
@file{Makefile} must have a target named @file{e@var{emul}.c}; this
target must depend upon @file{emulparams/@var{emul}.sh}, as well as the
appropriate scripts in the @file{scripttempl} and @file{emultempl}
directories.  The @file{Makefile} target must invoke @code{GENSCRIPTS}
with two arguments: @var{emul}, and the value of the make variable
@code{tdir_@var{emul}}.  The value of the latter variable will be set by
the @file{configure} script, and is used to set the default target
directory to search.

By convention, the @file{emulparams/@var{emul}.sh} shell script should
only set shell variables.  It may set shell variables which are to be
interpreted by the @file{scripttempl} and the @file{emultempl} scripts.
Certain shell variables are interpreted directly by the
@file{genscripts.sh} script.

Here is a list of shell variables interpreted by @file{genscripts.sh},
as well as some conventional shell variables interpreted by the
@file{scripttempl} and @file{emultempl} scripts.

@table @code
@item SCRIPT_NAME
This is the name of the @file{scripttempl} script to use.  If
@code{SCRIPT_NAME} is set to @var{script}, @file{genscripts.sh} will use
the script @file{scripttempl/@var{script}.sc}.

@item TEMPLATE_NAME
This is the name of the @file{emultempl} script to use.  If
@code{TEMPLATE_NAME} is set to @var{template}, @file{genscripts.sh} will
use the script @file{emultempl/@var{template}.em}.  If this variable is
not set, the default value is @samp{generic}.

@item GENERATE_SHLIB_SCRIPT
If this is set to a nonempty string, @file{genscripts.sh} will invoke
the @file{scripttempl} script an extra time to create a shared library
script.  @ref{linker scripts}.

@item OUTPUT_FORMAT
This is normally set to indicate the BFD output format use (e.g.,
@samp{"a.out-sunos-big"}.  The @file{scripttempl} script will normally
use it in an @code{OUTPUT_FORMAT} expression in the linker script.

@item ARCH
This is normally set to indicate the architecture to use (e.g.,
@samp{sparc}).  The @file{scripttempl} script will normally use it in an
@code{OUTPUT_ARCH} expression in the linker script.

@item ENTRY
Some @file{scripttempl} scripts use this to set the entry address, in an
@code{ENTRY} expression in the linker script.

@item TEXT_START_ADDR
Some @file{scripttempl} scripts use this to set the start address of the
@samp{.text} section.

@item SEGMENT_SIZE
The @file{genscripts.sh} script uses this to set the default value of
@code{DATA_ALIGNMENT} when running the @file{scripttempl} script.

@item TARGET_PAGE_SIZE
If @code{SEGMENT_SIZE} is not defined, the @file{genscripts.sh} script
uses this to define it.

@item ALIGNMENT
Some @file{scripttempl} scripts set this to a number to pass to
@code{ALIGN} to set the required alignment for the @code{end} symbol.
@end table

@node linker scripts
@section @file{scripttempl} scripts

Each linker target uses a @file{scripttempl} script to generate the
default linker scripts.  The name of the @file{scripttempl} script is
set by the @code{SCRIPT_NAME} variable in the @file{emulparams} script.
If @code{SCRIPT_NAME} is set to @var{script}, @code{genscripts.sh} will
invoke @file{scripttempl/@var{script}.sc}.

The @file{genscripts.sh} script will invoke the @file{scripttempl}
script 5 to 9 times.  Each time it will set the shell variable
@code{LD_FLAG} to a different value.  When the linker is run, the
options used will direct it to select a particular script.  (Script
selection is controlled by the @code{get_script} emulation entry point;
this describes the conventional behaviour).

The @file{scripttempl} script should just write a linker script, written
in the linker command language, to standard output.  If the emulation
name--the name of the @file{emulparams} file without the @file{.sc}
extension--is @var{emul}, then the output will be directed to
@file{ldscripts/@var{emul}.@var{extension}} in the build directory,
where @var{extension} changes each time the @file{scripttempl} script is
invoked.

Here is the list of values assigned to @code{LD_FLAG}.

@table @code
@item (empty)
The script generated is used by default (when none of the following
cases apply).  The output has an extension of @file{.x}.
@item n
The script generated is used when the linker is invoked with the
@code{-n} option.  The output has an extension of @file{.xn}.
@item N
The script generated is used when the linker is invoked with the
@code{-N} option.  The output has an extension of @file{.xbn}.
@item r
The script generated is used when the linker is invoked with the
@code{-r} option.  The output has an extension of @file{.xr}.
@item u
The script generated is used when the linker is invoked with the
@code{-Ur} option.  The output has an extension of @file{.xu}.
@item shared
The @file{scripttempl} script is only invoked with @code{LD_FLAG} set to
this value if @code{GENERATE_SHLIB_SCRIPT} is defined in the
@file{emulparams} file.  The @file{emultempl} script must arrange to use
this script at the appropriate time, normally when the linker is invoked
with the @code{-shared} option.  The output has an extension of
@file{.xs}.
@item c
The @file{scripttempl} script is only invoked with @code{LD_FLAG} set to
this value if @code{GENERATE_COMBRELOC_SCRIPT} is defined in the
@file{emulparams} file or if @code{SCRIPT_NAME} is @code{elf}. The
@file{emultempl} script must arrange to use this script at the appropriate
time, normally when the linker is invoked with the @code{-z combreloc}
option.  The output has an extension of
@file{.xc}.
@item cshared
The @file{scripttempl} script is only invoked with @code{LD_FLAG} set to
this value if @code{GENERATE_COMBRELOC_SCRIPT} is defined in the
@file{emulparams} file or if @code{SCRIPT_NAME} is @code{elf} and
@code{GENERATE_SHLIB_SCRIPT} is defined in the @file{emulparams} file.
The @file{emultempl} script must arrange to use this script at the
appropriate time, normally when the linker is invoked with the @code{-shared
-z combreloc} option.  The output has an extension of @file{.xsc}.
@item auto_import
The @file{scripttempl} script is only invoked with @code{LD_FLAG} set to
this value if @code{GENERATE_AUTO_IMPORT_SCRIPT} is defined in the
@file{emulparams} file.  The @file{emultempl} script must arrange to
use this script at the appropriate time, normally when the linker is
invoked with the @code{--enable-auto-import} option.  The output has
an extension of @file{.xa}.
@end table

Besides the shell variables set by the @file{emulparams} script, and the
@code{LD_FLAG} variable, the @file{genscripts.sh} script will set
certain variables for each run of the @file{scripttempl} script.

@table @code
@item RELOCATING
This will be set to a non-empty string when the linker is doing a final
relocation (e.g., all scripts other than @code{-r} and @code{-Ur}).

@item CONSTRUCTING
This will be set to a non-empty string when the linker is building
global constructor and destructor tables (e.g., all scripts other than
@code{-r}).

@item DATA_ALIGNMENT
This will be set to an @code{ALIGN} expression when the output should be
page aligned, or to @samp{.} when generating the @code{-N} script.

@item CREATE_SHLIB
This will be set to a non-empty string when generating a @code{-shared}
script.

@item COMBRELOC
This will be set to a non-empty string when generating @code{-z combreloc}
scripts to a temporary file name which can be used during script generation.
@end table

The conventional way to write a @file{scripttempl} script is to first
set a few shell variables, and then write out a linker script using
@code{cat} with a here document.  The linker script will use variable
substitutions, based on the above variables and those set in the
@file{emulparams} script, to control its behaviour.

When there are parts of the @file{scripttempl} script which should only
be run when doing a final relocation, they should be enclosed within a
variable substitution based on @code{RELOCATING}.  For example, on many
targets special symbols such as @code{_end} should be defined when doing
a final link.  Naturally, those symbols should not be defined when doing
a relocatable link using @code{-r}.  The @file{scripttempl} script
could use a construct like this to define those symbols:
@smallexample
  $@{RELOCATING+ _end = .;@}
@end smallexample
This will do the symbol assignment only if the @code{RELOCATING}
variable is defined.

The basic job of the linker script is to put the sections in the correct
order, and at the correct memory addresses.  For some targets, the
linker script may have to do some other operations.

For example, on most MIPS platforms, the linker is responsible for
defining the special symbol @code{_gp}, used to initialize the
@code{$gp} register.  It must be set to the start of the small data
section plus @code{0x8000}.  Naturally, it should only be defined when
doing a final relocation.  This will typically be done like this:
@smallexample
  $@{RELOCATING+ _gp = ALIGN(16) + 0x8000;@}
@end smallexample
This line would appear just before the sections which compose the small
data section (@samp{.sdata}, @samp{.sbss}).  All those sections would be
contiguous in memory.

Many COFF systems build constructor tables in the linker script.  The
compiler will arrange to output the address of each global constructor
in a @samp{.ctor} section, and the address of each global destructor in
a @samp{.dtor} section (this is done by defining
@code{ASM_OUTPUT_CONSTRUCTOR} and @code{ASM_OUTPUT_DESTRUCTOR} in the
@code{gcc} configuration files).  The @code{gcc} runtime support
routines expect the constructor table to be named @code{__CTOR_LIST__}.
They expect it to be a list of words, with the first word being the
count of the number of entries.  There should be a trailing zero word.
(Actually, the count may be -1 if the trailing word is present, and the
trailing word may be omitted if the count is correct, but, as the
@code{gcc} behaviour has changed slightly over the years, it is safest
to provide both).  Here is a typical way that might be handled in a
@file{scripttempl} file.
@smallexample
    $@{CONSTRUCTING+ __CTOR_LIST__ = .;@}
    $@{CONSTRUCTING+ LONG((__CTOR_END__ - __CTOR_LIST__) / 4 - 2)@}
    $@{CONSTRUCTING+ *(.ctors)@}
    $@{CONSTRUCTING+ LONG(0)@}
    $@{CONSTRUCTING+ __CTOR_END__ = .;@}
    $@{CONSTRUCTING+ __DTOR_LIST__ = .;@}
    $@{CONSTRUCTING+ LONG((__DTOR_END__ - __DTOR_LIST__) / 4 - 2)@}
    $@{CONSTRUCTING+ *(.dtors)@}
    $@{CONSTRUCTING+ LONG(0)@}
    $@{CONSTRUCTING+ __DTOR_END__ = .;@}
@end smallexample
The use of @code{CONSTRUCTING} ensures that these linker script commands
will only appear when the linker is supposed to be building the
constructor and destructor tables.  This example is written for a target
which uses 4 byte pointers.

Embedded systems often need to set a stack address.  This is normally
best done by using the @code{PROVIDE} construct with a default stack
address.  This permits the user to easily override the stack address
using the @code{--defsym} option.  Here is an example:
@smallexample
  $@{RELOCATING+ PROVIDE (__stack = 0x80000000);@}
@end smallexample
The value of the symbol @code{__stack} would then be used in the startup
code to initialize the stack pointer.

@node linker emulations
@section @file{emultempl} scripts

Each linker target uses an @file{emultempl} script to generate the
emulation code.  The name of the @file{emultempl} script is set by the
@code{TEMPLATE_NAME} variable in the @file{emulparams} script.  If the
@code{TEMPLATE_NAME} variable is not set, the default is
@samp{generic}.  If the value of @code{TEMPLATE_NAME} is @var{template},
@file{genscripts.sh} will use @file{emultempl/@var{template}.em}.

Most targets use the generic @file{emultempl} script,
@file{emultempl/generic.em}.  A different @file{emultempl} script is
only needed if the linker must support unusual actions, such as linking
against shared libraries.

The @file{emultempl} script is normally written as a simple invocation
of @code{cat} with a here document.  The document will use a few
variable substitutions.  Typically each function names uses a
substitution involving @code{EMULATION_NAME}, for ease of debugging when
the linker supports multiple emulations.

Every function and variable in the emitted file should be static.  The
only globally visible object must be named
@code{ld_@var{EMULATION_NAME}_emulation}, where @var{EMULATION_NAME} is
the name of the emulation set in @file{configure.tgt} (this is also the
name of the @file{emulparams} file without the @file{.sh} extension).
The @file{genscripts.sh} script will set the shell variable
@code{EMULATION_NAME} before invoking the @file{emultempl} script.

The @code{ld_@var{EMULATION_NAME}_emulation} variable must be a
@code{struct ld_emulation_xfer_struct}, as defined in @file{ldemul.h}.
It defines a set of function pointers which are invoked by the linker,
as well as strings for the emulation name (normally set from the shell
variable @code{EMULATION_NAME} and the default BFD target name (normally
set from the shell variable @code{OUTPUT_FORMAT} which is normally set
by the @file{emulparams} file).

The @file{genscripts.sh} script will set the shell variable
@code{COMPILE_IN} when it invokes the @file{emultempl} script for the
default emulation.  In this case, the @file{emultempl} script should
include the linker scripts directly, and return them from the
@code{get_scripts} entry point.  When the emulation is not the default,
the @code{get_scripts} entry point should just return a file name.  See
@file{emultempl/generic.em} for an example of how this is done.

At some point, the linker emulation entry points should be documented.

@node Emulation Walkthrough
@chapter A Walkthrough of a Typical Emulation

This chapter is to help people who are new to the way emulations
interact with the linker, or who are suddenly thrust into the position
of having to work with existing emulations.  It will discuss the files
you need to be aware of.  It will tell you when the given "hooks" in
the emulation will be called.  It will, hopefully, give you enough
information about when and how things happen that you'll be able to
get by.  As always, the source is the definitive reference to this.

The starting point for the linker is in @file{ldmain.c} where
@code{main} is defined.  The bulk of the code that's emulation
specific will initially be in @code{emultempl/@var{emulation}.em} but
will end up in @code{e@var{emulation}.c} when the build is done.
Most of the work to select and interface with emulations is in
@code{ldemul.h} and @code{ldemul.c}.  Specifically, @code{ldemul.h}
defines the @code{ld_emulation_xfer_struct} structure your emulation
exports.

Your emulation file exports a symbol
@code{ld_@var{EMULATION_NAME}_emulation}.  If your emulation is
selected (it usually is, since usually there's only one),
@code{ldemul.c} sets the variable @var{ld_emulation} to point to it.
@code{ldemul.c} also defines a number of API functions that interface
to your emulation, like @code{ldemul_after_parse} which simply calls
your @code{ld_@var{EMULATION}_emulation.after_parse} function.  For
the rest of this section, the functions will be mentioned, but you
should assume the indirect reference to your emulation also.

We will also skip or gloss over parts of the link process that don't
relate to emulations, like setting up internationalization.

After initialization, @code{main} selects an emulation by pre-scanning
the command line arguments.  It calls @code{ldemul_choose_target} to
choose a target.  If you set @code{choose_target} to
@code{ldemul_default_target}, it picks your @code{target_name} by
default.

@code{main} calls @code{ldemul_before_parse}, then @code{parse_args}.
@code{parse_args} calls @code{ldemul_parse_args} for each arg, which
must update the @code{getopt} globals if it recognizes the argument.
If the emulation doesn't recognize it, then parse_args checks to see
if it recognizes it.

Now that the emulation has had access to all its command-line options,
@code{main} calls @code{ldemul_set_symbols}.  This can be used for any
initialization that may be affected by options.  It is also supposed
to set up any variables needed by the emulation script.

@code{main} now calls @code{ldemul_get_script} to get the emulation
script to use (based on arguments, no doubt, @pxref{Emulations}) and
runs it.  While parsing, @code{ldgram.y} may call @code{ldemul_hll} or
@code{ldemul_syslib} to handle the @code{HLL} or @code{SYSLIB}
commands.  It may call @code{ldemul_unrecognized_file} if you asked
the linker to link a file it doesn't recognize.  It will call
@code{ldemul_recognized_file} for each file it does recognize, in case
the emulation wants to handle some files specially.  All the while,
it's loading the files (possibly calling
@code{ldemul_open_dynamic_archive}) and symbols and stuff.  After it's
done reading the script, @code{main} calls @code{ldemul_after_parse}.
Use the after-parse hook to set up anything that depends on stuff the
script might have set up, like the entry point.

@code{main} next calls @code{lang_process} in @code{ldlang.c}.  This
appears to be the main core of the linking itself, as far as emulation
hooks are concerned(*).  It first opens the output file's BFD, calling
@code{ldemul_set_output_arch}, and calls
@code{ldemul_create_output_section_statements} in case you need to use
other means to find or create object files (i.e. shared libraries
found on a path, or fake stub objects).  Despite the name, nobody
creates output sections here.

(*) In most cases, the BFD library does the bulk of the actual
linking, handling symbol tables, symbol resolution, relocations, and
building the final output file.  See the BFD reference for all the
details.  Your emulation is usually concerned more with managing
things at the file and section level, like "put this here, add this
section", etc.

Next, the objects to be linked are opened and BFDs created for them,
and @code{ldemul_after_open} is called.  At this point, you have all
the objects and symbols loaded, but none of the data has been placed
yet.

Next comes the Big Linking Thingy (except for the parts BFD does).
All input sections are mapped to output sections according to the
script.  If a section doesn't get mapped by default,
@code{ldemul_place_orphan} will get called to figure out where it goes.
Next it figures out the offsets for each section, calling
@code{ldemul_before_allocation} before and
@code{ldemul_after_allocation} after deciding where each input section
ends up in the output sections.

The last part of @code{lang_process} is to figure out all the symbols'
values.  After assigning final values to the symbols,
@code{ldemul_finish} is called, and after that, any undefined symbols
are turned into fatal errors.

OK, back to @code{main}, which calls @code{ldwrite} in
@file{ldwrite.c}.  @code{ldwrite} calls BFD's final_link, which does
all the relocation fixups and writes the output bfd to disk, and we're
done.

In summary,

@itemize @bullet

@item @code{main()} in @file{ldmain.c}
@item @file{emultempl/@var{EMULATION}.em} has your code
@item @code{ldemul_choose_target} (defaults to your @code{target_name})
@item @code{ldemul_before_parse}
@item Parse argv, calls @code{ldemul_parse_args} for each
@item @code{ldemul_set_symbols}
@item @code{ldemul_get_script}
@item parse script

@itemize @bullet
@item may call @code{ldemul_hll} or @code{ldemul_syslib}
@item may call @code{ldemul_open_dynamic_archive}
@end itemize

@item @code{ldemul_after_parse}
@item @code{lang_process()} in @file{ldlang.c}

@itemize @bullet
@item create @code{output_bfd}
@item @code{ldemul_set_output_arch}
@item @code{ldemul_create_output_section_statements}
@item read objects, create input bfds - all symbols exist, but have no values
@item may call @code{ldemul_unrecognized_file}
@item will call @code{ldemul_recognized_file}
@item @code{ldemul_after_open}
@item map input sections to output sections
@item may call @code{ldemul_place_orphan} for remaining sections
@item @code{ldemul_before_allocation}
@item gives input sections offsets into output sections, places output sections
@item @code{ldemul_after_allocation} - section addresses valid
@item assigns values to symbols
@item @code{ldemul_finish} - symbol values valid
@end itemize

@item output bfd is written to disk

@end itemize

@node Architecture Specific
@chapter Some Architecture Specific Notes

This is the place for notes on the behavior of @code{ld} on
specific platforms.  Currently, only Intel x86 is documented (and
of that, only the auto-import behavior for DLLs).

@menu
* ix86::                        Intel x86
@end menu

@node ix86
@section Intel x86

@table @emph
@code{ld} can create DLLs that operate with various runtimes available
on a common x86 operating system.  These runtimes include native (using
the mingw "platform"), cygwin, and pw.

@item auto-import from DLLs
@enumerate
@item
With this feature on, DLL clients can import variables from DLL
without any concern from their side (for example, without any source
code modifications).  Auto-import can be enabled using the
@code{--enable-auto-import} flag, or disabled via the
@code{--disable-auto-import} flag.  Auto-import is disabled by default.

@item
This is done completely in bounds of the PE specification (to be fair,
there's a minor violation of the spec at one point, but in practice
auto-import works on all known variants of that common x86 operating
system)  So, the resulting DLL can be used with any other PE
compiler/linker.

@item
Auto-import is fully compatible with standard import method, in which
variables are decorated using attribute modifiers. Libraries of either
type may be mixed together.

@item
Overhead (space): 8 bytes per imported symbol, plus 20 for each
reference to it; Overhead (load time): negligible; Overhead
(virtual/physical memory): should be less than effect of DLL
relocation.
@end enumerate

Motivation

The obvious and only way to get rid of dllimport insanity is
to make client access variable directly in the DLL, bypassing
the extra dereference imposed by ordinary DLL runtime linking.
I.e., whenever client contains something like

@code{mov dll_var,%eax,}

address of dll_var in the command should be relocated to point
into loaded DLL. The aim is to make OS loader do so, and than
make ld help with that.  Import section of PE made following
way: there's a vector of structures each describing imports
from particular DLL. Each such structure points to two other
parallel vectors: one holding imported names, and one which
will hold address of corresponding imported name. So, the
solution is de-vectorize these structures, making import
locations be sparse and pointing directly into code.

Implementation

For each reference of data symbol to be imported from DLL (to
set of which belong symbols with name <sym>, if __imp_<sym> is
found in implib), the import fixup entry is generated. That
entry is of type IMAGE_IMPORT_DESCRIPTOR and stored in .idata$3
subsection. Each fixup entry contains pointer to symbol's address
within .text section (marked with __fuN_<sym> symbol, where N is
integer), pointer to DLL name (so, DLL name is referenced by
multiple entries), and pointer to symbol name thunk. Symbol name
thunk is singleton vector (__nm_th_<symbol>) pointing to
IMAGE_IMPORT_BY_NAME structure (__nm_<symbol>) directly containing
imported name. Here comes that "om the edge" problem mentioned above:
PE specification rambles that name vector (OriginalFirstThunk) should
run in parallel with addresses vector (FirstThunk), i.e. that they
should have same number of elements and terminated with zero. We violate
this, since FirstThunk points directly into machine code. But in
practice, OS loader implemented the sane way: it goes thru
OriginalFirstThunk and puts addresses to FirstThunk, not something
else. It once again should be noted that dll and symbol name
structures are reused across fixup entries and should be there
anyway to support standard import stuff, so sustained overhead is
20 bytes per reference. Other question is whether having several
IMAGE_IMPORT_DESCRIPTORS for the same DLL is possible. Answer is yes,
it is done even by native compiler/linker (libth32's functions are in
fact resident in windows9x kernel32.dll, so if you use it, you have
two IMAGE_IMPORT_DESCRIPTORS for kernel32.dll). Yet other question is
whether referencing the same PE structures several times is valid.
The answer is why not, prohibiting that (detecting violation) would
require more work on behalf of loader than not doing it.

@end table

@node GNU Free Documentation License
@chapter GNU Free Documentation License

@include fdl.texi

@contents
@bye