3 @c Copyright 1991, 1992, 1993, 1994, 1995, 1996, 1997, 1998, 1999, 2000,
4 @c 2001, 2002, 2003, 2004, 2005, 2006, 2007, 2008, 2009, 2010, 2011, 2012
5 @c Free Software Foundation, Inc.
8 @include configdoc.texi
9 @c (configdoc.texi is generated by the Makefile)
15 @macro gcctabopt{body}
21 @c Configure for the generation of man pages
44 @dircategory Software development
46 * Ld: (ld). The GNU linker.
51 This file documents the @sc{gnu} linker LD
52 @ifset VERSION_PACKAGE
53 @value{VERSION_PACKAGE}
55 version @value{VERSION}.
57 Copyright @copyright{} 1991, 1992, 1993, 1994, 1995, 1996, 1997, 1998, 1999, 2000,
58 2001, 2002, 2003, 2004, 2005, 2006, 2007, 2008, 2009 Free Software Foundation, Inc.
60 Permission is granted to copy, distribute and/or modify this document
61 under the terms of the GNU Free Documentation License, Version 1.3
62 or any later version published by the Free Software Foundation;
63 with no Invariant Sections, with no Front-Cover Texts, and with no
64 Back-Cover Texts. A copy of the license is included in the
65 section entitled ``GNU Free Documentation License''.
69 @setchapternewpage odd
70 @settitle The GNU linker
75 @ifset VERSION_PACKAGE
76 @subtitle @value{VERSION_PACKAGE}
78 @subtitle Version @value{VERSION}
79 @author Steve Chamberlain
80 @author Ian Lance Taylor
85 \hfill Red Hat Inc\par
86 \hfill nickc\@credhat.com, doc\@redhat.com\par
87 \hfill {\it The GNU linker}\par
88 \hfill Edited by Jeffrey Osier (jeffrey\@cygnus.com)\par
90 \global\parindent=0pt % Steve likes it this way.
93 @vskip 0pt plus 1filll
94 @c man begin COPYRIGHT
95 Copyright @copyright{} 1991, 1992, 1993, 1994, 1995, 1996, 1997, 1998,
96 1999, 2000, 2001, 2002, 2003, 2004, 2005, 2006, 2007, 2008, 2009 Free
97 Software Foundation, Inc.
99 Permission is granted to copy, distribute and/or modify this document
100 under the terms of the GNU Free Documentation License, Version 1.3
101 or any later version published by the Free Software Foundation;
102 with no Invariant Sections, with no Front-Cover Texts, and with no
103 Back-Cover Texts. A copy of the license is included in the
104 section entitled ``GNU Free Documentation License''.
110 @c FIXME: Talk about importance of *order* of args, cmds to linker!
115 This file documents the @sc{gnu} linker ld
116 @ifset VERSION_PACKAGE
117 @value{VERSION_PACKAGE}
119 version @value{VERSION}.
121 This document is distributed under the terms of the GNU Free
122 Documentation License version 1.3. A copy of the license is included
123 in the section entitled ``GNU Free Documentation License''.
126 * Overview:: Overview
127 * Invocation:: Invocation
128 * Scripts:: Linker Scripts
130 * Machine Dependent:: Machine Dependent Features
134 * H8/300:: ld and the H8/300
137 * Renesas:: ld and other Renesas micros
140 * i960:: ld and the Intel 960 family
143 * ARM:: ld and the ARM family
146 * HPPA ELF32:: ld and HPPA 32-bit ELF
149 * M68HC11/68HC12:: ld and the Motorola 68HC11 and 68HC12 families
152 * M68K:: ld and Motorola 68K family
155 * PowerPC ELF32:: ld and PowerPC 32-bit ELF Support
158 * PowerPC64 ELF64:: ld and PowerPC64 64-bit ELF Support
161 * SPU ELF:: ld and SPU ELF Support
164 * TI COFF:: ld and the TI COFF
167 * Win32:: ld and WIN32 (cygwin/mingw)
170 * Xtensa:: ld and Xtensa Processors
173 @ifclear SingleFormat
176 @c Following blank line required for remaining bug in makeinfo conds/menus
178 * Reporting Bugs:: Reporting Bugs
179 * MRI:: MRI Compatible Script Files
180 * GNU Free Documentation License:: GNU Free Documentation License
181 * LD Index:: LD Index
188 @cindex @sc{gnu} linker
189 @cindex what is this?
192 @c man begin SYNOPSIS
193 ld [@b{options}] @var{objfile} @dots{}
197 ar(1), nm(1), objcopy(1), objdump(1), readelf(1) and
198 the Info entries for @file{binutils} and
203 @c man begin DESCRIPTION
205 @command{ld} combines a number of object and archive files, relocates
206 their data and ties up symbol references. Usually the last step in
207 compiling a program is to run @command{ld}.
209 @command{ld} accepts Linker Command Language files written in
210 a superset of AT&T's Link Editor Command Language syntax,
211 to provide explicit and total control over the linking process.
215 This man page does not describe the command language; see the
216 @command{ld} entry in @code{info} for full details on the command
217 language and on other aspects of the GNU linker.
220 @ifclear SingleFormat
221 This version of @command{ld} uses the general purpose BFD libraries
222 to operate on object files. This allows @command{ld} to read, combine, and
223 write object files in many different formats---for example, COFF or
224 @code{a.out}. Different formats may be linked together to produce any
225 available kind of object file. @xref{BFD}, for more information.
228 Aside from its flexibility, the @sc{gnu} linker is more helpful than other
229 linkers in providing diagnostic information. Many linkers abandon
230 execution immediately upon encountering an error; whenever possible,
231 @command{ld} continues executing, allowing you to identify other errors
232 (or, in some cases, to get an output file in spite of the error).
239 @c man begin DESCRIPTION
241 The @sc{gnu} linker @command{ld} is meant to cover a broad range of situations,
242 and to be as compatible as possible with other linkers. As a result,
243 you have many choices to control its behavior.
249 * Options:: Command Line Options
250 * Environment:: Environment Variables
254 @section Command Line Options
262 The linker supports a plethora of command-line options, but in actual
263 practice few of them are used in any particular context.
264 @cindex standard Unix system
265 For instance, a frequent use of @command{ld} is to link standard Unix
266 object files on a standard, supported Unix system. On such a system, to
267 link a file @code{hello.o}:
270 ld -o @var{output} /lib/crt0.o hello.o -lc
273 This tells @command{ld} to produce a file called @var{output} as the
274 result of linking the file @code{/lib/crt0.o} with @code{hello.o} and
275 the library @code{libc.a}, which will come from the standard search
276 directories. (See the discussion of the @samp{-l} option below.)
278 Some of the command-line options to @command{ld} may be specified at any
279 point in the command line. However, options which refer to files, such
280 as @samp{-l} or @samp{-T}, cause the file to be read at the point at
281 which the option appears in the command line, relative to the object
282 files and other file options. Repeating non-file options with a
283 different argument will either have no further effect, or override prior
284 occurrences (those further to the left on the command line) of that
285 option. Options which may be meaningfully specified more than once are
286 noted in the descriptions below.
289 Non-option arguments are object files or archives which are to be linked
290 together. They may follow, precede, or be mixed in with command-line
291 options, except that an object file argument may not be placed between
292 an option and its argument.
294 Usually the linker is invoked with at least one object file, but you can
295 specify other forms of binary input files using @samp{-l}, @samp{-R},
296 and the script command language. If @emph{no} binary input files at all
297 are specified, the linker does not produce any output, and issues the
298 message @samp{No input files}.
300 If the linker cannot recognize the format of an object file, it will
301 assume that it is a linker script. A script specified in this way
302 augments the main linker script used for the link (either the default
303 linker script or the one specified by using @samp{-T}). This feature
304 permits the linker to link against a file which appears to be an object
305 or an archive, but actually merely defines some symbol values, or uses
306 @code{INPUT} or @code{GROUP} to load other objects. Specifying a
307 script in this way merely augments the main linker script, with the
308 extra commands placed after the main script; use the @samp{-T} option
309 to replace the default linker script entirely, but note the effect of
310 the @code{INSERT} command. @xref{Scripts}.
312 For options whose names are a single letter,
313 option arguments must either follow the option letter without intervening
314 whitespace, or be given as separate arguments immediately following the
315 option that requires them.
317 For options whose names are multiple letters, either one dash or two can
318 precede the option name; for example, @samp{-trace-symbol} and
319 @samp{--trace-symbol} are equivalent. Note---there is one exception to
320 this rule. Multiple letter options that start with a lower case 'o' can
321 only be preceded by two dashes. This is to reduce confusion with the
322 @samp{-o} option. So for example @samp{-omagic} sets the output file
323 name to @samp{magic} whereas @samp{--omagic} sets the NMAGIC flag on the
326 Arguments to multiple-letter options must either be separated from the
327 option name by an equals sign, or be given as separate arguments
328 immediately following the option that requires them. For example,
329 @samp{--trace-symbol foo} and @samp{--trace-symbol=foo} are equivalent.
330 Unique abbreviations of the names of multiple-letter options are
333 Note---if the linker is being invoked indirectly, via a compiler driver
334 (e.g. @samp{gcc}) then all the linker command line options should be
335 prefixed by @samp{-Wl,} (or whatever is appropriate for the particular
336 compiler driver) like this:
339 gcc -Wl,--start-group foo.o bar.o -Wl,--end-group
342 This is important, because otherwise the compiler driver program may
343 silently drop the linker options, resulting in a bad link. Confusion
344 may also arise when passing options that require values through a
345 driver, as the use of a space between option and argument acts as
346 a separator, and causes the driver to pass only the option to the linker
347 and the argument to the compiler. In this case, it is simplest to use
348 the joined forms of both single- and multiple-letter options, such as:
351 gcc foo.o bar.o -Wl,-eENTRY -Wl,-Map=a.map
354 Here is a table of the generic command line switches accepted by the GNU
358 @include at-file.texi
360 @kindex -a @var{keyword}
361 @item -a @var{keyword}
362 This option is supported for HP/UX compatibility. The @var{keyword}
363 argument must be one of the strings @samp{archive}, @samp{shared}, or
364 @samp{default}. @samp{-aarchive} is functionally equivalent to
365 @samp{-Bstatic}, and the other two keywords are functionally equivalent
366 to @samp{-Bdynamic}. This option may be used any number of times.
368 @kindex --audit @var{AUDITLIB}
369 @item --audit @var{AUDITLIB}
370 Adds @var{AUDITLIB} to the @code{DT_AUDIT} entry of the dynamic section.
371 @var{AUDITLIB} is not checked for existence, nor will it use the DT_SONAME
372 specified in the library. If specified multiple times @code{DT_AUDIT}
373 will contain a colon separated list of audit interfaces to use. If the linker
374 finds an object with an audit entry while searching for shared libraries,
375 it will add a corresponding @code{DT_DEPAUDIT} entry in the output file.
376 This option is only meaningful on ELF platforms supporting the rtld-audit
380 @cindex architectures
381 @kindex -A @var{arch}
382 @item -A @var{architecture}
383 @kindex --architecture=@var{arch}
384 @itemx --architecture=@var{architecture}
385 In the current release of @command{ld}, this option is useful only for the
386 Intel 960 family of architectures. In that @command{ld} configuration, the
387 @var{architecture} argument identifies the particular architecture in
388 the 960 family, enabling some safeguards and modifying the
389 archive-library search path. @xref{i960,,@command{ld} and the Intel 960
390 family}, for details.
392 Future releases of @command{ld} may support similar functionality for
393 other architecture families.
396 @ifclear SingleFormat
397 @cindex binary input format
398 @kindex -b @var{format}
399 @kindex --format=@var{format}
402 @item -b @var{input-format}
403 @itemx --format=@var{input-format}
404 @command{ld} may be configured to support more than one kind of object
405 file. If your @command{ld} is configured this way, you can use the
406 @samp{-b} option to specify the binary format for input object files
407 that follow this option on the command line. Even when @command{ld} is
408 configured to support alternative object formats, you don't usually need
409 to specify this, as @command{ld} should be configured to expect as a
410 default input format the most usual format on each machine.
411 @var{input-format} is a text string, the name of a particular format
412 supported by the BFD libraries. (You can list the available binary
413 formats with @samp{objdump -i}.)
416 You may want to use this option if you are linking files with an unusual
417 binary format. You can also use @samp{-b} to switch formats explicitly (when
418 linking object files of different formats), by including
419 @samp{-b @var{input-format}} before each group of object files in a
422 The default format is taken from the environment variable
427 You can also define the input format from a script, using the command
430 see @ref{Format Commands}.
434 @kindex -c @var{MRI-cmdfile}
435 @kindex --mri-script=@var{MRI-cmdfile}
436 @cindex compatibility, MRI
437 @item -c @var{MRI-commandfile}
438 @itemx --mri-script=@var{MRI-commandfile}
439 For compatibility with linkers produced by MRI, @command{ld} accepts script
440 files written in an alternate, restricted command language, described in
442 @ref{MRI,,MRI Compatible Script Files}.
445 the MRI Compatible Script Files section of GNU ld documentation.
447 Introduce MRI script files with
448 the option @samp{-c}; use the @samp{-T} option to run linker
449 scripts written in the general-purpose @command{ld} scripting language.
450 If @var{MRI-cmdfile} does not exist, @command{ld} looks for it in the directories
451 specified by any @samp{-L} options.
453 @cindex common allocation
460 These three options are equivalent; multiple forms are supported for
461 compatibility with other linkers. They assign space to common symbols
462 even if a relocatable output file is specified (with @samp{-r}). The
463 script command @code{FORCE_COMMON_ALLOCATION} has the same effect.
464 @xref{Miscellaneous Commands}.
466 @kindex --depaudit @var{AUDITLIB}
467 @kindex -P @var{AUDITLIB}
468 @item --depaudit @var{AUDITLIB}
469 @itemx -P @var{AUDITLIB}
470 Adds @var{AUDITLIB} to the @code{DT_DEPAUDIT} entry of the dynamic section.
471 @var{AUDITLIB} is not checked for existence, nor will it use the DT_SONAME
472 specified in the library. If specified multiple times @code{DT_DEPAUDIT}
473 will contain a colon separated list of audit interfaces to use. This
474 option is only meaningful on ELF platforms supporting the rtld-audit interface.
475 The -P option is provided for Solaris compatibility.
477 @cindex entry point, from command line
478 @kindex -e @var{entry}
479 @kindex --entry=@var{entry}
481 @itemx --entry=@var{entry}
482 Use @var{entry} as the explicit symbol for beginning execution of your
483 program, rather than the default entry point. If there is no symbol
484 named @var{entry}, the linker will try to parse @var{entry} as a number,
485 and use that as the entry address (the number will be interpreted in
486 base 10; you may use a leading @samp{0x} for base 16, or a leading
487 @samp{0} for base 8). @xref{Entry Point}, for a discussion of defaults
488 and other ways of specifying the entry point.
490 @kindex --exclude-libs
491 @item --exclude-libs @var{lib},@var{lib},...
492 Specifies a list of archive libraries from which symbols should not be automatically
493 exported. The library names may be delimited by commas or colons. Specifying
494 @code{--exclude-libs ALL} excludes symbols in all archive libraries from
495 automatic export. This option is available only for the i386 PE targeted
496 port of the linker and for ELF targeted ports. For i386 PE, symbols
497 explicitly listed in a .def file are still exported, regardless of this
498 option. For ELF targeted ports, symbols affected by this option will
499 be treated as hidden.
501 @kindex --exclude-modules-for-implib
502 @item --exclude-modules-for-implib @var{module},@var{module},...
503 Specifies a list of object files or archive members, from which symbols
504 should not be automatically exported, but which should be copied wholesale
505 into the import library being generated during the link. The module names
506 may be delimited by commas or colons, and must match exactly the filenames
507 used by @command{ld} to open the files; for archive members, this is simply
508 the member name, but for object files the name listed must include and
509 match precisely any path used to specify the input file on the linker's
510 command-line. This option is available only for the i386 PE targeted port
511 of the linker. Symbols explicitly listed in a .def file are still exported,
512 regardless of this option.
514 @cindex dynamic symbol table
516 @kindex --export-dynamic
517 @kindex --no-export-dynamic
519 @itemx --export-dynamic
520 @itemx --no-export-dynamic
521 When creating a dynamically linked executable, using the @option{-E}
522 option or the @option{--export-dynamic} option causes the linker to add
523 all symbols to the dynamic symbol table. The dynamic symbol table is the
524 set of symbols which are visible from dynamic objects at run time.
526 If you do not use either of these options (or use the
527 @option{--no-export-dynamic} option to restore the default behavior), the
528 dynamic symbol table will normally contain only those symbols which are
529 referenced by some dynamic object mentioned in the link.
531 If you use @code{dlopen} to load a dynamic object which needs to refer
532 back to the symbols defined by the program, rather than some other
533 dynamic object, then you will probably need to use this option when
534 linking the program itself.
536 You can also use the dynamic list to control what symbols should
537 be added to the dynamic symbol table if the output format supports it.
538 See the description of @samp{--dynamic-list}.
540 Note that this option is specific to ELF targeted ports. PE targets
541 support a similar function to export all symbols from a DLL or EXE; see
542 the description of @samp{--export-all-symbols} below.
544 @ifclear SingleFormat
545 @cindex big-endian objects
549 Link big-endian objects. This affects the default output format.
551 @cindex little-endian objects
554 Link little-endian objects. This affects the default output format.
557 @kindex -f @var{name}
558 @kindex --auxiliary=@var{name}
560 @itemx --auxiliary=@var{name}
561 When creating an ELF shared object, set the internal DT_AUXILIARY field
562 to the specified name. This tells the dynamic linker that the symbol
563 table of the shared object should be used as an auxiliary filter on the
564 symbol table of the shared object @var{name}.
566 If you later link a program against this filter object, then, when you
567 run the program, the dynamic linker will see the DT_AUXILIARY field. If
568 the dynamic linker resolves any symbols from the filter object, it will
569 first check whether there is a definition in the shared object
570 @var{name}. If there is one, it will be used instead of the definition
571 in the filter object. The shared object @var{name} need not exist.
572 Thus the shared object @var{name} may be used to provide an alternative
573 implementation of certain functions, perhaps for debugging or for
574 machine specific performance.
576 This option may be specified more than once. The DT_AUXILIARY entries
577 will be created in the order in which they appear on the command line.
579 @kindex -F @var{name}
580 @kindex --filter=@var{name}
582 @itemx --filter=@var{name}
583 When creating an ELF shared object, set the internal DT_FILTER field to
584 the specified name. This tells the dynamic linker that the symbol table
585 of the shared object which is being created should be used as a filter
586 on the symbol table of the shared object @var{name}.
588 If you later link a program against this filter object, then, when you
589 run the program, the dynamic linker will see the DT_FILTER field. The
590 dynamic linker will resolve symbols according to the symbol table of the
591 filter object as usual, but it will actually link to the definitions
592 found in the shared object @var{name}. Thus the filter object can be
593 used to select a subset of the symbols provided by the object
596 Some older linkers used the @option{-F} option throughout a compilation
597 toolchain for specifying object-file format for both input and output
599 @ifclear SingleFormat
600 The @sc{gnu} linker uses other mechanisms for this purpose: the
601 @option{-b}, @option{--format}, @option{--oformat} options, the
602 @code{TARGET} command in linker scripts, and the @code{GNUTARGET}
603 environment variable.
605 The @sc{gnu} linker will ignore the @option{-F} option when not
606 creating an ELF shared object.
608 @cindex finalization function
609 @kindex -fini=@var{name}
610 @item -fini=@var{name}
611 When creating an ELF executable or shared object, call NAME when the
612 executable or shared object is unloaded, by setting DT_FINI to the
613 address of the function. By default, the linker uses @code{_fini} as
614 the function to call.
618 Ignored. Provided for compatibility with other tools.
620 @kindex -G @var{value}
621 @kindex --gpsize=@var{value}
624 @itemx --gpsize=@var{value}
625 Set the maximum size of objects to be optimized using the GP register to
626 @var{size}. This is only meaningful for object file formats such as
627 MIPS ECOFF which supports putting large and small objects into different
628 sections. This is ignored for other object file formats.
630 @cindex runtime library name
631 @kindex -h @var{name}
632 @kindex -soname=@var{name}
634 @itemx -soname=@var{name}
635 When creating an ELF shared object, set the internal DT_SONAME field to
636 the specified name. When an executable is linked with a shared object
637 which has a DT_SONAME field, then when the executable is run the dynamic
638 linker will attempt to load the shared object specified by the DT_SONAME
639 field rather than the using the file name given to the linker.
642 @cindex incremental link
644 Perform an incremental link (same as option @samp{-r}).
646 @cindex initialization function
647 @kindex -init=@var{name}
648 @item -init=@var{name}
649 When creating an ELF executable or shared object, call NAME when the
650 executable or shared object is loaded, by setting DT_INIT to the address
651 of the function. By default, the linker uses @code{_init} as the
654 @cindex archive files, from cmd line
655 @kindex -l @var{namespec}
656 @kindex --library=@var{namespec}
657 @item -l @var{namespec}
658 @itemx --library=@var{namespec}
659 Add the archive or object file specified by @var{namespec} to the
660 list of files to link. This option may be used any number of times.
661 If @var{namespec} is of the form @file{:@var{filename}}, @command{ld}
662 will search the library path for a file called @var{filename}, otherwise it
663 will search the library path for a file called @file{lib@var{namespec}.a}.
665 On systems which support shared libraries, @command{ld} may also search for
666 files other than @file{lib@var{namespec}.a}. Specifically, on ELF
667 and SunOS systems, @command{ld} will search a directory for a library
668 called @file{lib@var{namespec}.so} before searching for one called
669 @file{lib@var{namespec}.a}. (By convention, a @code{.so} extension
670 indicates a shared library.) Note that this behavior does not apply
671 to @file{:@var{filename}}, which always specifies a file called
674 The linker will search an archive only once, at the location where it is
675 specified on the command line. If the archive defines a symbol which
676 was undefined in some object which appeared before the archive on the
677 command line, the linker will include the appropriate file(s) from the
678 archive. However, an undefined symbol in an object appearing later on
679 the command line will not cause the linker to search the archive again.
681 See the @option{-(} option for a way to force the linker to search
682 archives multiple times.
684 You may list the same archive multiple times on the command line.
687 This type of archive searching is standard for Unix linkers. However,
688 if you are using @command{ld} on AIX, note that it is different from the
689 behaviour of the AIX linker.
692 @cindex search directory, from cmd line
694 @kindex --library-path=@var{dir}
695 @item -L @var{searchdir}
696 @itemx --library-path=@var{searchdir}
697 Add path @var{searchdir} to the list of paths that @command{ld} will search
698 for archive libraries and @command{ld} control scripts. You may use this
699 option any number of times. The directories are searched in the order
700 in which they are specified on the command line. Directories specified
701 on the command line are searched before the default directories. All
702 @option{-L} options apply to all @option{-l} options, regardless of the
703 order in which the options appear. @option{-L} options do not affect
704 how @command{ld} searches for a linker script unless @option{-T}
707 If @var{searchdir} begins with @code{=}, then the @code{=} will be replaced
708 by the @dfn{sysroot prefix}, a path specified when the linker is configured.
711 The default set of paths searched (without being specified with
712 @samp{-L}) depends on which emulation mode @command{ld} is using, and in
713 some cases also on how it was configured. @xref{Environment}.
716 The paths can also be specified in a link script with the
717 @code{SEARCH_DIR} command. Directories specified this way are searched
718 at the point in which the linker script appears in the command line.
721 @kindex -m @var{emulation}
722 @item -m @var{emulation}
723 Emulate the @var{emulation} linker. You can list the available
724 emulations with the @samp{--verbose} or @samp{-V} options.
726 If the @samp{-m} option is not used, the emulation is taken from the
727 @code{LDEMULATION} environment variable, if that is defined.
729 Otherwise, the default emulation depends upon how the linker was
737 Print a link map to the standard output. A link map provides
738 information about the link, including the following:
742 Where object files are mapped into memory.
744 How common symbols are allocated.
746 All archive members included in the link, with a mention of the symbol
747 which caused the archive member to be brought in.
749 The values assigned to symbols.
751 Note - symbols whose values are computed by an expression which
752 involves a reference to a previous value of the same symbol may not
753 have correct result displayed in the link map. This is because the
754 linker discards intermediate results and only retains the final value
755 of an expression. Under such circumstances the linker will display
756 the final value enclosed by square brackets. Thus for example a
757 linker script containing:
765 will produce the following output in the link map if the @option{-M}
770 [0x0000000c] foo = (foo * 0x4)
771 [0x0000000c] foo = (foo + 0x8)
774 See @ref{Expressions} for more information about expressions in linker
779 @cindex read-only text
784 Turn off page alignment of sections, and disable linking against shared
785 libraries. If the output format supports Unix style magic numbers,
786 mark the output as @code{NMAGIC}.
790 @cindex read/write from cmd line
794 Set the text and data sections to be readable and writable. Also, do
795 not page-align the data segment, and disable linking against shared
796 libraries. If the output format supports Unix style magic numbers,
797 mark the output as @code{OMAGIC}. Note: Although a writable text section
798 is allowed for PE-COFF targets, it does not conform to the format
799 specification published by Microsoft.
804 This option negates most of the effects of the @option{-N} option. It
805 sets the text section to be read-only, and forces the data segment to
806 be page-aligned. Note - this option does not enable linking against
807 shared libraries. Use @option{-Bdynamic} for this.
809 @kindex -o @var{output}
810 @kindex --output=@var{output}
811 @cindex naming the output file
812 @item -o @var{output}
813 @itemx --output=@var{output}
814 Use @var{output} as the name for the program produced by @command{ld}; if this
815 option is not specified, the name @file{a.out} is used by default. The
816 script command @code{OUTPUT} can also specify the output file name.
818 @kindex -O @var{level}
819 @cindex generating optimized output
821 If @var{level} is a numeric values greater than zero @command{ld} optimizes
822 the output. This might take significantly longer and therefore probably
823 should only be enabled for the final binary. At the moment this
824 option only affects ELF shared library generation. Future releases of
825 the linker may make more use of this option. Also currently there is
826 no difference in the linker's behaviour for different non-zero values
827 of this option. Again this may change with future releases.
830 @kindex --emit-relocs
831 @cindex retain relocations in final executable
834 Leave relocation sections and contents in fully linked executables.
835 Post link analysis and optimization tools may need this information in
836 order to perform correct modifications of executables. This results
837 in larger executables.
839 This option is currently only supported on ELF platforms.
841 @kindex --force-dynamic
842 @cindex forcing the creation of dynamic sections
843 @item --force-dynamic
844 Force the output file to have dynamic sections. This option is specific
848 @cindex relocatable output
850 @kindex --relocatable
853 Generate relocatable output---i.e., generate an output file that can in
854 turn serve as input to @command{ld}. This is often called @dfn{partial
855 linking}. As a side effect, in environments that support standard Unix
856 magic numbers, this option also sets the output file's magic number to
858 @c ; see @option{-N}.
859 If this option is not specified, an absolute file is produced. When
860 linking C++ programs, this option @emph{will not} resolve references to
861 constructors; to do that, use @samp{-Ur}.
863 When an input file does not have the same format as the output file,
864 partial linking is only supported if that input file does not contain any
865 relocations. Different output formats can have further restrictions; for
866 example some @code{a.out}-based formats do not support partial linking
867 with input files in other formats at all.
869 This option does the same thing as @samp{-i}.
871 @kindex -R @var{file}
872 @kindex --just-symbols=@var{file}
873 @cindex symbol-only input
874 @item -R @var{filename}
875 @itemx --just-symbols=@var{filename}
876 Read symbol names and their addresses from @var{filename}, but do not
877 relocate it or include it in the output. This allows your output file
878 to refer symbolically to absolute locations of memory defined in other
879 programs. You may use this option more than once.
881 For compatibility with other ELF linkers, if the @option{-R} option is
882 followed by a directory name, rather than a file name, it is treated as
883 the @option{-rpath} option.
887 @cindex strip all symbols
890 Omit all symbol information from the output file.
893 @kindex --strip-debug
894 @cindex strip debugger symbols
897 Omit debugger symbol information (but not all symbols) from the output file.
901 @cindex input files, displaying
904 Print the names of the input files as @command{ld} processes them.
906 @kindex -T @var{script}
907 @kindex --script=@var{script}
909 @item -T @var{scriptfile}
910 @itemx --script=@var{scriptfile}
911 Use @var{scriptfile} as the linker script. This script replaces
912 @command{ld}'s default linker script (rather than adding to it), so
913 @var{commandfile} must specify everything necessary to describe the
914 output file. @xref{Scripts}. If @var{scriptfile} does not exist in
915 the current directory, @code{ld} looks for it in the directories
916 specified by any preceding @samp{-L} options. Multiple @samp{-T}
919 @kindex -dT @var{script}
920 @kindex --default-script=@var{script}
922 @item -dT @var{scriptfile}
923 @itemx --default-script=@var{scriptfile}
924 Use @var{scriptfile} as the default linker script. @xref{Scripts}.
926 This option is similar to the @option{--script} option except that
927 processing of the script is delayed until after the rest of the
928 command line has been processed. This allows options placed after the
929 @option{--default-script} option on the command line to affect the
930 behaviour of the linker script, which can be important when the linker
931 command line cannot be directly controlled by the user. (eg because
932 the command line is being constructed by another tool, such as
935 @kindex -u @var{symbol}
936 @kindex --undefined=@var{symbol}
937 @cindex undefined symbol
938 @item -u @var{symbol}
939 @itemx --undefined=@var{symbol}
940 Force @var{symbol} to be entered in the output file as an undefined
941 symbol. Doing this may, for example, trigger linking of additional
942 modules from standard libraries. @samp{-u} may be repeated with
943 different option arguments to enter additional undefined symbols. This
944 option is equivalent to the @code{EXTERN} linker script command.
949 For anything other than C++ programs, this option is equivalent to
950 @samp{-r}: it generates relocatable output---i.e., an output file that can in
951 turn serve as input to @command{ld}. When linking C++ programs, @samp{-Ur}
952 @emph{does} resolve references to constructors, unlike @samp{-r}.
953 It does not work to use @samp{-Ur} on files that were themselves linked
954 with @samp{-Ur}; once the constructor table has been built, it cannot
955 be added to. Use @samp{-Ur} only for the last partial link, and
956 @samp{-r} for the others.
958 @kindex --unique[=@var{SECTION}]
959 @item --unique[=@var{SECTION}]
960 Creates a separate output section for every input section matching
961 @var{SECTION}, or if the optional wildcard @var{SECTION} argument is
962 missing, for every orphan input section. An orphan section is one not
963 specifically mentioned in a linker script. You may use this option
964 multiple times on the command line; It prevents the normal merging of
965 input sections with the same name, overriding output section assignments
975 Display the version number for @command{ld}. The @option{-V} option also
976 lists the supported emulations.
979 @kindex --discard-all
980 @cindex deleting local symbols
983 Delete all local symbols.
986 @kindex --discard-locals
987 @cindex local symbols, deleting
989 @itemx --discard-locals
990 Delete all temporary local symbols. (These symbols start with
991 system-specific local label prefixes, typically @samp{.L} for ELF systems
992 or @samp{L} for traditional a.out systems.)
994 @kindex -y @var{symbol}
995 @kindex --trace-symbol=@var{symbol}
996 @cindex symbol tracing
997 @item -y @var{symbol}
998 @itemx --trace-symbol=@var{symbol}
999 Print the name of each linked file in which @var{symbol} appears. This
1000 option may be given any number of times. On many systems it is necessary
1001 to prepend an underscore.
1003 This option is useful when you have an undefined symbol in your link but
1004 don't know where the reference is coming from.
1006 @kindex -Y @var{path}
1008 Add @var{path} to the default library search path. This option exists
1009 for Solaris compatibility.
1011 @kindex -z @var{keyword}
1012 @item -z @var{keyword}
1013 The recognized keywords are:
1017 Combines multiple reloc sections and sorts them to make dynamic symbol
1018 lookup caching possible.
1021 Disallows undefined symbols in object files. Undefined symbols in
1022 shared libraries are still allowed.
1025 Marks the object as requiring executable stack.
1028 This option is only meaningful when building a shared object. It makes
1029 the symbols defined by this shared object available for symbol resolution
1030 of subsequently loaded libraries.
1033 This option is only meaningful when building a shared object.
1034 It marks the object so that its runtime initialization will occur
1035 before the runtime initialization of any other objects brought into
1036 the process at the same time. Similarly the runtime finalization of
1037 the object will occur after the runtime finalization of any other
1041 Marks the object that its symbol table interposes before all symbols
1042 but the primary executable.
1045 When generating an executable or shared library, mark it to tell the
1046 dynamic linker to defer function call resolution to the point when
1047 the function is called (lazy binding), rather than at load time.
1048 Lazy binding is the default.
1051 Marks the object that its filters be processed immediately at
1055 Allows multiple definitions.
1058 Disables multiple reloc sections combining.
1061 Disables production of copy relocs.
1064 Marks the object that the search for dependencies of this object will
1065 ignore any default library search paths.
1068 Marks the object shouldn't be unloaded at runtime.
1071 Marks the object not available to @code{dlopen}.
1074 Marks the object can not be dumped by @code{dldump}.
1077 Marks the object as not requiring executable stack.
1080 Don't create an ELF @code{PT_GNU_RELRO} segment header in the object.
1083 When generating an executable or shared library, mark it to tell the
1084 dynamic linker to resolve all symbols when the program is started, or
1085 when the shared library is linked to using dlopen, instead of
1086 deferring function call resolution to the point when the function is
1090 Marks the object may contain $ORIGIN.
1093 Create an ELF @code{PT_GNU_RELRO} segment header in the object.
1095 @item max-page-size=@var{value}
1096 Set the emulation maximum page size to @var{value}.
1098 @item common-page-size=@var{value}
1099 Set the emulation common page size to @var{value}.
1101 @item stack-size=@var{value}
1102 Specify a stack size for in an ELF @code{PT_GNU_STACK} segment.
1103 Specifying zero will override any default non-zero sized
1104 @code{PT_GNU_STACK} segment creation.
1108 Other keywords are ignored for Solaris compatibility.
1111 @cindex groups of archives
1112 @item -( @var{archives} -)
1113 @itemx --start-group @var{archives} --end-group
1114 The @var{archives} should be a list of archive files. They may be
1115 either explicit file names, or @samp{-l} options.
1117 The specified archives are searched repeatedly until no new undefined
1118 references are created. Normally, an archive is searched only once in
1119 the order that it is specified on the command line. If a symbol in that
1120 archive is needed to resolve an undefined symbol referred to by an
1121 object in an archive that appears later on the command line, the linker
1122 would not be able to resolve that reference. By grouping the archives,
1123 they all be searched repeatedly until all possible references are
1126 Using this option has a significant performance cost. It is best to use
1127 it only when there are unavoidable circular references between two or
1130 @kindex --accept-unknown-input-arch
1131 @kindex --no-accept-unknown-input-arch
1132 @item --accept-unknown-input-arch
1133 @itemx --no-accept-unknown-input-arch
1134 Tells the linker to accept input files whose architecture cannot be
1135 recognised. The assumption is that the user knows what they are doing
1136 and deliberately wants to link in these unknown input files. This was
1137 the default behaviour of the linker, before release 2.14. The default
1138 behaviour from release 2.14 onwards is to reject such input files, and
1139 so the @samp{--accept-unknown-input-arch} option has been added to
1140 restore the old behaviour.
1143 @kindex --no-as-needed
1145 @itemx --no-as-needed
1146 This option affects ELF DT_NEEDED tags for dynamic libraries mentioned
1147 on the command line after the @option{--as-needed} option. Normally
1148 the linker will add a DT_NEEDED tag for each dynamic library mentioned
1149 on the command line, regardless of whether the library is actually
1150 needed or not. @option{--as-needed} causes a DT_NEEDED tag to only be
1151 emitted for a library that satisfies an undefined symbol reference
1152 from a regular object file or, if the library is not found in the
1153 DT_NEEDED lists of other libraries linked up to that point, an
1154 undefined symbol reference from another dynamic library.
1155 @option{--no-as-needed} restores the default behaviour.
1157 @kindex --add-needed
1158 @kindex --no-add-needed
1160 @itemx --no-add-needed
1161 These two options have been deprecated because of the similarity of
1162 their names to the @option{--as-needed} and @option{--no-as-needed}
1163 options. They have been replaced by @option{--copy-dt-needed-entries}
1164 and @option{--no-copy-dt-needed-entries}.
1166 @kindex -assert @var{keyword}
1167 @item -assert @var{keyword}
1168 This option is ignored for SunOS compatibility.
1172 @kindex -call_shared
1176 Link against dynamic libraries. This is only meaningful on platforms
1177 for which shared libraries are supported. This option is normally the
1178 default on such platforms. The different variants of this option are
1179 for compatibility with various systems. You may use this option
1180 multiple times on the command line: it affects library searching for
1181 @option{-l} options which follow it.
1185 Set the @code{DF_1_GROUP} flag in the @code{DT_FLAGS_1} entry in the dynamic
1186 section. This causes the runtime linker to handle lookups in this
1187 object and its dependencies to be performed only inside the group.
1188 @option{--unresolved-symbols=report-all} is implied. This option is
1189 only meaningful on ELF platforms which support shared libraries.
1199 Do not link against shared libraries. This is only meaningful on
1200 platforms for which shared libraries are supported. The different
1201 variants of this option are for compatibility with various systems. You
1202 may use this option multiple times on the command line: it affects
1203 library searching for @option{-l} options which follow it. This
1204 option also implies @option{--unresolved-symbols=report-all}. This
1205 option can be used with @option{-shared}. Doing so means that a
1206 shared library is being created but that all of the library's external
1207 references must be resolved by pulling in entries from static
1212 When creating a shared library, bind references to global symbols to the
1213 definition within the shared library, if any. Normally, it is possible
1214 for a program linked against a shared library to override the definition
1215 within the shared library. This option is only meaningful on ELF
1216 platforms which support shared libraries.
1218 @kindex -Bsymbolic-functions
1219 @item -Bsymbolic-functions
1220 When creating a shared library, bind references to global function
1221 symbols to the definition within the shared library, if any.
1222 This option is only meaningful on ELF platforms which support shared
1225 @kindex --dynamic-list=@var{dynamic-list-file}
1226 @item --dynamic-list=@var{dynamic-list-file}
1227 Specify the name of a dynamic list file to the linker. This is
1228 typically used when creating shared libraries to specify a list of
1229 global symbols whose references shouldn't be bound to the definition
1230 within the shared library, or creating dynamically linked executables
1231 to specify a list of symbols which should be added to the symbol table
1232 in the executable. This option is only meaningful on ELF platforms
1233 which support shared libraries.
1235 The format of the dynamic list is the same as the version node without
1236 scope and node name. See @ref{VERSION} for more information.
1238 @kindex --dynamic-list-data
1239 @item --dynamic-list-data
1240 Include all global data symbols to the dynamic list.
1242 @kindex --dynamic-list-cpp-new
1243 @item --dynamic-list-cpp-new
1244 Provide the builtin dynamic list for C++ operator new and delete. It
1245 is mainly useful for building shared libstdc++.
1247 @kindex --dynamic-list-cpp-typeinfo
1248 @item --dynamic-list-cpp-typeinfo
1249 Provide the builtin dynamic list for C++ runtime type identification.
1251 @kindex --check-sections
1252 @kindex --no-check-sections
1253 @item --check-sections
1254 @itemx --no-check-sections
1255 Asks the linker @emph{not} to check section addresses after they have
1256 been assigned to see if there are any overlaps. Normally the linker will
1257 perform this check, and if it finds any overlaps it will produce
1258 suitable error messages. The linker does know about, and does make
1259 allowances for sections in overlays. The default behaviour can be
1260 restored by using the command line switch @option{--check-sections}.
1261 Section overlap is not usually checked for relocatable links. You can
1262 force checking in that case by using the @option{--check-sections}
1265 @kindex --copy-dt-needed-entries
1266 @kindex --no-copy-dt-needed-entries
1267 @item --copy-dt-needed-entries
1268 @itemx --no-copy-dt-needed-entries
1269 This option affects the treatment of dynamic libraries referred to
1270 by DT_NEEDED tags @emph{inside} ELF dynamic libraries mentioned on the
1271 command line. Normally the linker won't add a DT_NEEDED tag to the
1272 output binary for each library mentioned in a DT_NEEDED tag in an
1273 input dynamic library. With @option{--copy-dt-needed-entries}
1274 specified on the command line however any dynamic libraries that
1275 follow it will have their DT_NEEDED entries added. The default
1276 behaviour can be restored with @option{--no-copy-dt-needed-entries}.
1278 This option also has an effect on the resolution of symbols in dynamic
1279 libraries. With @option{--copy-dt-needed-entries} dynamic libraries
1280 mentioned on the command line will be recursively searched, following
1281 their DT_NEEDED tags to other libraries, in order to resolve symbols
1282 required by the output binary. With the default setting however
1283 the searching of dynamic libraries that follow it will stop with the
1284 dynamic library itself. No DT_NEEDED links will be traversed to resolve
1287 @cindex cross reference table
1290 Output a cross reference table. If a linker map file is being
1291 generated, the cross reference table is printed to the map file.
1292 Otherwise, it is printed on the standard output.
1294 The format of the table is intentionally simple, so that it may be
1295 easily processed by a script if necessary. The symbols are printed out,
1296 sorted by name. For each symbol, a list of file names is given. If the
1297 symbol is defined, the first file listed is the location of the
1298 definition. The remaining files contain references to the symbol.
1300 @cindex common allocation
1301 @kindex --no-define-common
1302 @item --no-define-common
1303 This option inhibits the assignment of addresses to common symbols.
1304 The script command @code{INHIBIT_COMMON_ALLOCATION} has the same effect.
1305 @xref{Miscellaneous Commands}.
1307 The @samp{--no-define-common} option allows decoupling
1308 the decision to assign addresses to Common symbols from the choice
1309 of the output file type; otherwise a non-Relocatable output type
1310 forces assigning addresses to Common symbols.
1311 Using @samp{--no-define-common} allows Common symbols that are referenced
1312 from a shared library to be assigned addresses only in the main program.
1313 This eliminates the unused duplicate space in the shared library,
1314 and also prevents any possible confusion over resolving to the wrong
1315 duplicate when there are many dynamic modules with specialized search
1316 paths for runtime symbol resolution.
1318 @cindex symbols, from command line
1319 @kindex --defsym=@var{symbol}=@var{exp}
1320 @item --defsym=@var{symbol}=@var{expression}
1321 Create a global symbol in the output file, containing the absolute
1322 address given by @var{expression}. You may use this option as many
1323 times as necessary to define multiple symbols in the command line. A
1324 limited form of arithmetic is supported for the @var{expression} in this
1325 context: you may give a hexadecimal constant or the name of an existing
1326 symbol, or use @code{+} and @code{-} to add or subtract hexadecimal
1327 constants or symbols. If you need more elaborate expressions, consider
1328 using the linker command language from a script (@pxref{Assignments,,
1329 Assignment: Symbol Definitions}). @emph{Note:} there should be no white
1330 space between @var{symbol}, the equals sign (``@key{=}''), and
1333 @cindex demangling, from command line
1334 @kindex --demangle[=@var{style}]
1335 @kindex --no-demangle
1336 @item --demangle[=@var{style}]
1337 @itemx --no-demangle
1338 These options control whether to demangle symbol names in error messages
1339 and other output. When the linker is told to demangle, it tries to
1340 present symbol names in a readable fashion: it strips leading
1341 underscores if they are used by the object file format, and converts C++
1342 mangled symbol names into user readable names. Different compilers have
1343 different mangling styles. The optional demangling style argument can be used
1344 to choose an appropriate demangling style for your compiler. The linker will
1345 demangle by default unless the environment variable @samp{COLLECT_NO_DEMANGLE}
1346 is set. These options may be used to override the default.
1348 @cindex dynamic linker, from command line
1349 @kindex -I@var{file}
1350 @kindex --dynamic-linker=@var{file}
1352 @itemx --dynamic-linker=@var{file}
1353 Set the name of the dynamic linker. This is only meaningful when
1354 generating dynamically linked ELF executables. The default dynamic
1355 linker is normally correct; don't use this unless you know what you are
1358 @kindex --fatal-warnings
1359 @kindex --no-fatal-warnings
1360 @item --fatal-warnings
1361 @itemx --no-fatal-warnings
1362 Treat all warnings as errors. The default behaviour can be restored
1363 with the option @option{--no-fatal-warnings}.
1365 @kindex --force-exe-suffix
1366 @item --force-exe-suffix
1367 Make sure that an output file has a .exe suffix.
1369 If a successfully built fully linked output file does not have a
1370 @code{.exe} or @code{.dll} suffix, this option forces the linker to copy
1371 the output file to one of the same name with a @code{.exe} suffix. This
1372 option is useful when using unmodified Unix makefiles on a Microsoft
1373 Windows host, since some versions of Windows won't run an image unless
1374 it ends in a @code{.exe} suffix.
1376 @kindex --gc-sections
1377 @kindex --no-gc-sections
1378 @cindex garbage collection
1380 @itemx --no-gc-sections
1381 Enable garbage collection of unused input sections. It is ignored on
1382 targets that do not support this option. The default behaviour (of not
1383 performing this garbage collection) can be restored by specifying
1384 @samp{--no-gc-sections} on the command line.
1386 @samp{--gc-sections} decides which input sections are used by
1387 examining symbols and relocations. The section containing the entry
1388 symbol and all sections containing symbols undefined on the
1389 command-line will be kept, as will sections containing symbols
1390 referenced by dynamic objects. Note that when building shared
1391 libraries, the linker must assume that any visible symbol is
1392 referenced. Once this initial set of sections has been determined,
1393 the linker recursively marks as used any section referenced by their
1394 relocations. See @samp{--entry} and @samp{--undefined}.
1396 This option can be set when doing a partial link (enabled with option
1397 @samp{-r}). In this case the root of symbols kept must be explicitly
1398 specified either by an @samp{--entry} or @samp{--undefined} option or by
1399 a @code{ENTRY} command in the linker script.
1401 @kindex --print-gc-sections
1402 @kindex --no-print-gc-sections
1403 @cindex garbage collection
1404 @item --print-gc-sections
1405 @itemx --no-print-gc-sections
1406 List all sections removed by garbage collection. The listing is
1407 printed on stderr. This option is only effective if garbage
1408 collection has been enabled via the @samp{--gc-sections}) option. The
1409 default behaviour (of not listing the sections that are removed) can
1410 be restored by specifying @samp{--no-print-gc-sections} on the command
1413 @kindex --print-output-format
1414 @cindex output format
1415 @item --print-output-format
1416 Print the name of the default output format (perhaps influenced by
1417 other command-line options). This is the string that would appear
1418 in an @code{OUTPUT_FORMAT} linker script command (@pxref{File Commands}).
1424 Print a summary of the command-line options on the standard output and exit.
1426 @kindex --target-help
1428 Print a summary of all target specific options on the standard output and exit.
1430 @kindex -Map=@var{mapfile}
1431 @item -Map=@var{mapfile}
1432 Print a link map to the file @var{mapfile}. See the description of the
1433 @option{-M} option, above.
1435 @cindex memory usage
1436 @kindex --no-keep-memory
1437 @item --no-keep-memory
1438 @command{ld} normally optimizes for speed over memory usage by caching the
1439 symbol tables of input files in memory. This option tells @command{ld} to
1440 instead optimize for memory usage, by rereading the symbol tables as
1441 necessary. This may be required if @command{ld} runs out of memory space
1442 while linking a large executable.
1444 @kindex --no-undefined
1446 @item --no-undefined
1448 Report unresolved symbol references from regular object files. This
1449 is done even if the linker is creating a non-symbolic shared library.
1450 The switch @option{--[no-]allow-shlib-undefined} controls the
1451 behaviour for reporting unresolved references found in shared
1452 libraries being linked in.
1454 @kindex --allow-multiple-definition
1456 @item --allow-multiple-definition
1458 Normally when a symbol is defined multiple times, the linker will
1459 report a fatal error. These options allow multiple definitions and the
1460 first definition will be used.
1462 @kindex --allow-shlib-undefined
1463 @kindex --no-allow-shlib-undefined
1464 @item --allow-shlib-undefined
1465 @itemx --no-allow-shlib-undefined
1466 Allows or disallows undefined symbols in shared libraries.
1467 This switch is similar to @option{--no-undefined} except that it
1468 determines the behaviour when the undefined symbols are in a
1469 shared library rather than a regular object file. It does not affect
1470 how undefined symbols in regular object files are handled.
1472 The default behaviour is to report errors for any undefined symbols
1473 referenced in shared libraries if the linker is being used to create
1474 an executable, but to allow them if the linker is being used to create
1477 The reasons for allowing undefined symbol references in shared
1478 libraries specified at link time are that:
1482 A shared library specified at link time may not be the same as the one
1483 that is available at load time, so the symbol might actually be
1484 resolvable at load time.
1486 There are some operating systems, eg BeOS and HPPA, where undefined
1487 symbols in shared libraries are normal.
1489 The BeOS kernel for example patches shared libraries at load time to
1490 select whichever function is most appropriate for the current
1491 architecture. This is used, for example, to dynamically select an
1492 appropriate memset function.
1495 @kindex --no-undefined-version
1496 @item --no-undefined-version
1497 Normally when a symbol has an undefined version, the linker will ignore
1498 it. This option disallows symbols with undefined version and a fatal error
1499 will be issued instead.
1501 @kindex --default-symver
1502 @item --default-symver
1503 Create and use a default symbol version (the soname) for unversioned
1506 @kindex --default-imported-symver
1507 @item --default-imported-symver
1508 Create and use a default symbol version (the soname) for unversioned
1511 @kindex --no-warn-mismatch
1512 @item --no-warn-mismatch
1513 Normally @command{ld} will give an error if you try to link together input
1514 files that are mismatched for some reason, perhaps because they have
1515 been compiled for different processors or for different endiannesses.
1516 This option tells @command{ld} that it should silently permit such possible
1517 errors. This option should only be used with care, in cases when you
1518 have taken some special action that ensures that the linker errors are
1521 @kindex --no-warn-search-mismatch
1522 @item --no-warn-search-mismatch
1523 Normally @command{ld} will give a warning if it finds an incompatible
1524 library during a library search. This option silences the warning.
1526 @kindex --no-whole-archive
1527 @item --no-whole-archive
1528 Turn off the effect of the @option{--whole-archive} option for subsequent
1531 @cindex output file after errors
1532 @kindex --noinhibit-exec
1533 @item --noinhibit-exec
1534 Retain the executable output file whenever it is still usable.
1535 Normally, the linker will not produce an output file if it encounters
1536 errors during the link process; it exits without writing an output file
1537 when it issues any error whatsoever.
1541 Only search library directories explicitly specified on the
1542 command line. Library directories specified in linker scripts
1543 (including linker scripts specified on the command line) are ignored.
1545 @ifclear SingleFormat
1546 @kindex --oformat=@var{output-format}
1547 @item --oformat=@var{output-format}
1548 @command{ld} may be configured to support more than one kind of object
1549 file. If your @command{ld} is configured this way, you can use the
1550 @samp{--oformat} option to specify the binary format for the output
1551 object file. Even when @command{ld} is configured to support alternative
1552 object formats, you don't usually need to specify this, as @command{ld}
1553 should be configured to produce as a default output format the most
1554 usual format on each machine. @var{output-format} is a text string, the
1555 name of a particular format supported by the BFD libraries. (You can
1556 list the available binary formats with @samp{objdump -i}.) The script
1557 command @code{OUTPUT_FORMAT} can also specify the output format, but
1558 this option overrides it. @xref{BFD}.
1562 @kindex --pic-executable
1564 @itemx --pic-executable
1565 @cindex position independent executables
1566 Create a position independent executable. This is currently only supported on
1567 ELF platforms. Position independent executables are similar to shared
1568 libraries in that they are relocated by the dynamic linker to the virtual
1569 address the OS chooses for them (which can vary between invocations). Like
1570 normal dynamically linked executables they can be executed and symbols
1571 defined in the executable cannot be overridden by shared libraries.
1575 This option is ignored for Linux compatibility.
1579 This option is ignored for SVR4 compatibility.
1582 @cindex synthesizing linker
1583 @cindex relaxing addressing modes
1587 An option with machine dependent effects.
1589 This option is only supported on a few targets.
1592 @xref{H8/300,,@command{ld} and the H8/300}.
1595 @xref{i960,, @command{ld} and the Intel 960 family}.
1598 @xref{Xtensa,, @command{ld} and Xtensa Processors}.
1601 @xref{M68HC11/68HC12,,@command{ld} and the 68HC11 and 68HC12}.
1604 @xref{PowerPC ELF32,,@command{ld} and PowerPC 32-bit ELF Support}.
1607 On some platforms the @samp{--relax} option performs target specific,
1608 global optimizations that become possible when the linker resolves
1609 addressing in the program, such as relaxing address modes,
1610 synthesizing new instructions, selecting shorter version of current
1611 instructions, and combinig constant values.
1613 On some platforms these link time global optimizations may make symbolic
1614 debugging of the resulting executable impossible.
1616 This is known to be the case for the Matsushita MN10200 and MN10300
1617 family of processors.
1621 On platforms where this is not supported, @samp{--relax} is accepted,
1625 On platforms where @samp{--relax} is accepted the option
1626 @samp{--no-relax} can be used to disable the feature.
1628 @cindex retaining specified symbols
1629 @cindex stripping all but some symbols
1630 @cindex symbols, retaining selectively
1631 @kindex --retain-symbols-file=@var{filename}
1632 @item --retain-symbols-file=@var{filename}
1633 Retain @emph{only} the symbols listed in the file @var{filename},
1634 discarding all others. @var{filename} is simply a flat file, with one
1635 symbol name per line. This option is especially useful in environments
1639 where a large global symbol table is accumulated gradually, to conserve
1642 @samp{--retain-symbols-file} does @emph{not} discard undefined symbols,
1643 or symbols needed for relocations.
1645 You may only specify @samp{--retain-symbols-file} once in the command
1646 line. It overrides @samp{-s} and @samp{-S}.
1649 @item -rpath=@var{dir}
1650 @cindex runtime library search path
1651 @kindex -rpath=@var{dir}
1652 Add a directory to the runtime library search path. This is used when
1653 linking an ELF executable with shared objects. All @option{-rpath}
1654 arguments are concatenated and passed to the runtime linker, which uses
1655 them to locate shared objects at runtime. The @option{-rpath} option is
1656 also used when locating shared objects which are needed by shared
1657 objects explicitly included in the link; see the description of the
1658 @option{-rpath-link} option. If @option{-rpath} is not used when linking an
1659 ELF executable, the contents of the environment variable
1660 @code{LD_RUN_PATH} will be used if it is defined.
1662 The @option{-rpath} option may also be used on SunOS. By default, on
1663 SunOS, the linker will form a runtime search patch out of all the
1664 @option{-L} options it is given. If a @option{-rpath} option is used, the
1665 runtime search path will be formed exclusively using the @option{-rpath}
1666 options, ignoring the @option{-L} options. This can be useful when using
1667 gcc, which adds many @option{-L} options which may be on NFS mounted
1670 For compatibility with other ELF linkers, if the @option{-R} option is
1671 followed by a directory name, rather than a file name, it is treated as
1672 the @option{-rpath} option.
1676 @cindex link-time runtime library search path
1677 @kindex -rpath-link=@var{dir}
1678 @item -rpath-link=@var{dir}
1679 When using ELF or SunOS, one shared library may require another. This
1680 happens when an @code{ld -shared} link includes a shared library as one
1683 When the linker encounters such a dependency when doing a non-shared,
1684 non-relocatable link, it will automatically try to locate the required
1685 shared library and include it in the link, if it is not included
1686 explicitly. In such a case, the @option{-rpath-link} option
1687 specifies the first set of directories to search. The
1688 @option{-rpath-link} option may specify a sequence of directory names
1689 either by specifying a list of names separated by colons, or by
1690 appearing multiple times.
1692 This option should be used with caution as it overrides the search path
1693 that may have been hard compiled into a shared library. In such a case it
1694 is possible to use unintentionally a different search path than the
1695 runtime linker would do.
1697 The linker uses the following search paths to locate required shared
1701 Any directories specified by @option{-rpath-link} options.
1703 Any directories specified by @option{-rpath} options. The difference
1704 between @option{-rpath} and @option{-rpath-link} is that directories
1705 specified by @option{-rpath} options are included in the executable and
1706 used at runtime, whereas the @option{-rpath-link} option is only effective
1707 at link time. Searching @option{-rpath} in this way is only supported
1708 by native linkers and cross linkers which have been configured with
1709 the @option{--with-sysroot} option.
1711 On an ELF system, for native linkers, if the @option{-rpath} and
1712 @option{-rpath-link} options were not used, search the contents of the
1713 environment variable @code{LD_RUN_PATH}.
1715 On SunOS, if the @option{-rpath} option was not used, search any
1716 directories specified using @option{-L} options.
1718 For a native linker, the search the contents of the environment
1719 variable @code{LD_LIBRARY_PATH}.
1721 For a native ELF linker, the directories in @code{DT_RUNPATH} or
1722 @code{DT_RPATH} of a shared library are searched for shared
1723 libraries needed by it. The @code{DT_RPATH} entries are ignored if
1724 @code{DT_RUNPATH} entries exist.
1726 The default directories, normally @file{/lib} and @file{/usr/lib}.
1728 For a native linker on an ELF system, if the file @file{/etc/ld.so.conf}
1729 exists, the list of directories found in that file.
1732 If the required shared library is not found, the linker will issue a
1733 warning and continue with the link.
1740 @cindex shared libraries
1741 Create a shared library. This is currently only supported on ELF, XCOFF
1742 and SunOS platforms. On SunOS, the linker will automatically create a
1743 shared library if the @option{-e} option is not used and there are
1744 undefined symbols in the link.
1746 @kindex --sort-common
1748 @itemx --sort-common=ascending
1749 @itemx --sort-common=descending
1750 This option tells @command{ld} to sort the common symbols by alignment in
1751 ascending or descending order when it places them in the appropriate output
1752 sections. The symbol alignments considered are sixteen-byte or larger,
1753 eight-byte, four-byte, two-byte, and one-byte. This is to prevent gaps
1754 between symbols due to alignment constraints. If no sorting order is
1755 specified, then descending order is assumed.
1757 @kindex --sort-section=name
1758 @item --sort-section=name
1759 This option will apply @code{SORT_BY_NAME} to all wildcard section
1760 patterns in the linker script.
1762 @kindex --sort-section=alignment
1763 @item --sort-section=alignment
1764 This option will apply @code{SORT_BY_ALIGNMENT} to all wildcard section
1765 patterns in the linker script.
1767 @kindex --split-by-file
1768 @item --split-by-file[=@var{size}]
1769 Similar to @option{--split-by-reloc} but creates a new output section for
1770 each input file when @var{size} is reached. @var{size} defaults to a
1771 size of 1 if not given.
1773 @kindex --split-by-reloc
1774 @item --split-by-reloc[=@var{count}]
1775 Tries to creates extra sections in the output file so that no single
1776 output section in the file contains more than @var{count} relocations.
1777 This is useful when generating huge relocatable files for downloading into
1778 certain real time kernels with the COFF object file format; since COFF
1779 cannot represent more than 65535 relocations in a single section. Note
1780 that this will fail to work with object file formats which do not
1781 support arbitrary sections. The linker will not split up individual
1782 input sections for redistribution, so if a single input section contains
1783 more than @var{count} relocations one output section will contain that
1784 many relocations. @var{count} defaults to a value of 32768.
1788 Compute and display statistics about the operation of the linker, such
1789 as execution time and memory usage.
1791 @kindex --sysroot=@var{directory}
1792 @item --sysroot=@var{directory}
1793 Use @var{directory} as the location of the sysroot, overriding the
1794 configure-time default. This option is only supported by linkers
1795 that were configured using @option{--with-sysroot}.
1797 @kindex --traditional-format
1798 @cindex traditional format
1799 @item --traditional-format
1800 For some targets, the output of @command{ld} is different in some ways from
1801 the output of some existing linker. This switch requests @command{ld} to
1802 use the traditional format instead.
1805 For example, on SunOS, @command{ld} combines duplicate entries in the
1806 symbol string table. This can reduce the size of an output file with
1807 full debugging information by over 30 percent. Unfortunately, the SunOS
1808 @code{dbx} program can not read the resulting program (@code{gdb} has no
1809 trouble). The @samp{--traditional-format} switch tells @command{ld} to not
1810 combine duplicate entries.
1812 @kindex --section-start=@var{sectionname}=@var{org}
1813 @item --section-start=@var{sectionname}=@var{org}
1814 Locate a section in the output file at the absolute
1815 address given by @var{org}. You may use this option as many
1816 times as necessary to locate multiple sections in the command
1818 @var{org} must be a single hexadecimal integer;
1819 for compatibility with other linkers, you may omit the leading
1820 @samp{0x} usually associated with hexadecimal values. @emph{Note:} there
1821 should be no white space between @var{sectionname}, the equals
1822 sign (``@key{=}''), and @var{org}.
1824 @kindex -Tbss=@var{org}
1825 @kindex -Tdata=@var{org}
1826 @kindex -Ttext=@var{org}
1827 @cindex segment origins, cmd line
1828 @item -Tbss=@var{org}
1829 @itemx -Tdata=@var{org}
1830 @itemx -Ttext=@var{org}
1831 Same as @option{--section-start}, with @code{.bss}, @code{.data} or
1832 @code{.text} as the @var{sectionname}.
1834 @kindex -Ttext-segment=@var{org}
1835 @item -Ttext-segment=@var{org}
1836 @cindex text segment origin, cmd line
1837 When creating an ELF executable or shared object, it will set the address
1838 of the first byte of the text segment.
1840 @kindex --unresolved-symbols
1841 @item --unresolved-symbols=@var{method}
1842 Determine how to handle unresolved symbols. There are four possible
1843 values for @samp{method}:
1847 Do not report any unresolved symbols.
1850 Report all unresolved symbols. This is the default.
1852 @item ignore-in-object-files
1853 Report unresolved symbols that are contained in shared libraries, but
1854 ignore them if they come from regular object files.
1856 @item ignore-in-shared-libs
1857 Report unresolved symbols that come from regular object files, but
1858 ignore them if they come from shared libraries. This can be useful
1859 when creating a dynamic binary and it is known that all the shared
1860 libraries that it should be referencing are included on the linker's
1864 The behaviour for shared libraries on their own can also be controlled
1865 by the @option{--[no-]allow-shlib-undefined} option.
1867 Normally the linker will generate an error message for each reported
1868 unresolved symbol but the option @option{--warn-unresolved-symbols}
1869 can change this to a warning.
1871 @kindex --verbose[=@var{NUMBER}]
1872 @cindex verbose[=@var{NUMBER}]
1874 @itemx --verbose[=@var{NUMBER}]
1875 Display the version number for @command{ld} and list the linker emulations
1876 supported. Display which input files can and cannot be opened. Display
1877 the linker script being used by the linker. If the optional @var{NUMBER}
1878 argument > 1, plugin symbol status will also be displayed.
1880 @kindex --version-script=@var{version-scriptfile}
1881 @cindex version script, symbol versions
1882 @item --version-script=@var{version-scriptfile}
1883 Specify the name of a version script to the linker. This is typically
1884 used when creating shared libraries to specify additional information
1885 about the version hierarchy for the library being created. This option
1886 is only fully supported on ELF platforms which support shared libraries;
1887 see @ref{VERSION}. It is partially supported on PE platforms, which can
1888 use version scripts to filter symbol visibility in auto-export mode: any
1889 symbols marked @samp{local} in the version script will not be exported.
1892 @kindex --warn-common
1893 @cindex warnings, on combining symbols
1894 @cindex combining symbols, warnings on
1896 Warn when a common symbol is combined with another common symbol or with
1897 a symbol definition. Unix linkers allow this somewhat sloppy practise,
1898 but linkers on some other operating systems do not. This option allows
1899 you to find potential problems from combining global symbols.
1900 Unfortunately, some C libraries use this practise, so you may get some
1901 warnings about symbols in the libraries as well as in your programs.
1903 There are three kinds of global symbols, illustrated here by C examples:
1907 A definition, which goes in the initialized data section of the output
1911 An undefined reference, which does not allocate space.
1912 There must be either a definition or a common symbol for the
1916 A common symbol. If there are only (one or more) common symbols for a
1917 variable, it goes in the uninitialized data area of the output file.
1918 The linker merges multiple common symbols for the same variable into a
1919 single symbol. If they are of different sizes, it picks the largest
1920 size. The linker turns a common symbol into a declaration, if there is
1921 a definition of the same variable.
1924 The @samp{--warn-common} option can produce five kinds of warnings.
1925 Each warning consists of a pair of lines: the first describes the symbol
1926 just encountered, and the second describes the previous symbol
1927 encountered with the same name. One or both of the two symbols will be
1932 Turning a common symbol into a reference, because there is already a
1933 definition for the symbol.
1935 @var{file}(@var{section}): warning: common of `@var{symbol}'
1936 overridden by definition
1937 @var{file}(@var{section}): warning: defined here
1941 Turning a common symbol into a reference, because a later definition for
1942 the symbol is encountered. This is the same as the previous case,
1943 except that the symbols are encountered in a different order.
1945 @var{file}(@var{section}): warning: definition of `@var{symbol}'
1947 @var{file}(@var{section}): warning: common is here
1951 Merging a common symbol with a previous same-sized common symbol.
1953 @var{file}(@var{section}): warning: multiple common
1955 @var{file}(@var{section}): warning: previous common is here
1959 Merging a common symbol with a previous larger common symbol.
1961 @var{file}(@var{section}): warning: common of `@var{symbol}'
1962 overridden by larger common
1963 @var{file}(@var{section}): warning: larger common is here
1967 Merging a common symbol with a previous smaller common symbol. This is
1968 the same as the previous case, except that the symbols are
1969 encountered in a different order.
1971 @var{file}(@var{section}): warning: common of `@var{symbol}'
1972 overriding smaller common
1973 @var{file}(@var{section}): warning: smaller common is here
1977 @kindex --warn-constructors
1978 @item --warn-constructors
1979 Warn if any global constructors are used. This is only useful for a few
1980 object file formats. For formats like COFF or ELF, the linker can not
1981 detect the use of global constructors.
1983 @kindex --warn-multiple-gp
1984 @item --warn-multiple-gp
1985 Warn if multiple global pointer values are required in the output file.
1986 This is only meaningful for certain processors, such as the Alpha.
1987 Specifically, some processors put large-valued constants in a special
1988 section. A special register (the global pointer) points into the middle
1989 of this section, so that constants can be loaded efficiently via a
1990 base-register relative addressing mode. Since the offset in
1991 base-register relative mode is fixed and relatively small (e.g., 16
1992 bits), this limits the maximum size of the constant pool. Thus, in
1993 large programs, it is often necessary to use multiple global pointer
1994 values in order to be able to address all possible constants. This
1995 option causes a warning to be issued whenever this case occurs.
1998 @cindex warnings, on undefined symbols
1999 @cindex undefined symbols, warnings on
2001 Only warn once for each undefined symbol, rather than once per module
2004 @kindex --warn-section-align
2005 @cindex warnings, on section alignment
2006 @cindex section alignment, warnings on
2007 @item --warn-section-align
2008 Warn if the address of an output section is changed because of
2009 alignment. Typically, the alignment will be set by an input section.
2010 The address will only be changed if it not explicitly specified; that
2011 is, if the @code{SECTIONS} command does not specify a start address for
2012 the section (@pxref{SECTIONS}).
2014 @kindex --warn-shared-textrel
2015 @item --warn-shared-textrel
2016 Warn if the linker adds a DT_TEXTREL to a shared object.
2018 @kindex --warn-alternate-em
2019 @item --warn-alternate-em
2020 Warn if an object has alternate ELF machine code.
2022 @kindex --warn-unresolved-symbols
2023 @item --warn-unresolved-symbols
2024 If the linker is going to report an unresolved symbol (see the option
2025 @option{--unresolved-symbols}) it will normally generate an error.
2026 This option makes it generate a warning instead.
2028 @kindex --error-unresolved-symbols
2029 @item --error-unresolved-symbols
2030 This restores the linker's default behaviour of generating errors when
2031 it is reporting unresolved symbols.
2033 @kindex --whole-archive
2034 @cindex including an entire archive
2035 @item --whole-archive
2036 For each archive mentioned on the command line after the
2037 @option{--whole-archive} option, include every object file in the archive
2038 in the link, rather than searching the archive for the required object
2039 files. This is normally used to turn an archive file into a shared
2040 library, forcing every object to be included in the resulting shared
2041 library. This option may be used more than once.
2043 Two notes when using this option from gcc: First, gcc doesn't know
2044 about this option, so you have to use @option{-Wl,-whole-archive}.
2045 Second, don't forget to use @option{-Wl,-no-whole-archive} after your
2046 list of archives, because gcc will add its own list of archives to
2047 your link and you may not want this flag to affect those as well.
2049 @kindex --wrap=@var{symbol}
2050 @item --wrap=@var{symbol}
2051 Use a wrapper function for @var{symbol}. Any undefined reference to
2052 @var{symbol} will be resolved to @code{__wrap_@var{symbol}}. Any
2053 undefined reference to @code{__real_@var{symbol}} will be resolved to
2056 This can be used to provide a wrapper for a system function. The
2057 wrapper function should be called @code{__wrap_@var{symbol}}. If it
2058 wishes to call the system function, it should call
2059 @code{__real_@var{symbol}}.
2061 Here is a trivial example:
2065 __wrap_malloc (size_t c)
2067 printf ("malloc called with %zu\n", c);
2068 return __real_malloc (c);
2072 If you link other code with this file using @option{--wrap malloc}, then
2073 all calls to @code{malloc} will call the function @code{__wrap_malloc}
2074 instead. The call to @code{__real_malloc} in @code{__wrap_malloc} will
2075 call the real @code{malloc} function.
2077 You may wish to provide a @code{__real_malloc} function as well, so that
2078 links without the @option{--wrap} option will succeed. If you do this,
2079 you should not put the definition of @code{__real_malloc} in the same
2080 file as @code{__wrap_malloc}; if you do, the assembler may resolve the
2081 call before the linker has a chance to wrap it to @code{malloc}.
2083 @kindex --eh-frame-hdr
2084 @item --eh-frame-hdr
2085 Request creation of @code{.eh_frame_hdr} section and ELF
2086 @code{PT_GNU_EH_FRAME} segment header.
2088 @kindex --ld-generated-unwind-info
2089 @item --no-ld-generated-unwind-info
2090 Request creation of @code{.eh_frame} unwind info for linker
2091 generated code sections like PLT. This option is on by default
2092 if linker generated unwind info is supported.
2094 @kindex --enable-new-dtags
2095 @kindex --disable-new-dtags
2096 @item --enable-new-dtags
2097 @itemx --disable-new-dtags
2098 This linker can create the new dynamic tags in ELF. But the older ELF
2099 systems may not understand them. If you specify
2100 @option{--enable-new-dtags}, the dynamic tags will be created as needed.
2101 If you specify @option{--disable-new-dtags}, no new dynamic tags will be
2102 created. By default, the new dynamic tags are not created. Note that
2103 those options are only available for ELF systems.
2105 @kindex --hash-size=@var{number}
2106 @item --hash-size=@var{number}
2107 Set the default size of the linker's hash tables to a prime number
2108 close to @var{number}. Increasing this value can reduce the length of
2109 time it takes the linker to perform its tasks, at the expense of
2110 increasing the linker's memory requirements. Similarly reducing this
2111 value can reduce the memory requirements at the expense of speed.
2113 @kindex --hash-style=@var{style}
2114 @item --hash-style=@var{style}
2115 Set the type of linker's hash table(s). @var{style} can be either
2116 @code{sysv} for classic ELF @code{.hash} section, @code{gnu} for
2117 new style GNU @code{.gnu.hash} section or @code{both} for both
2118 the classic ELF @code{.hash} and new style GNU @code{.gnu.hash}
2119 hash tables. The default is @code{sysv}.
2121 @kindex --reduce-memory-overheads
2122 @item --reduce-memory-overheads
2123 This option reduces memory requirements at ld runtime, at the expense of
2124 linking speed. This was introduced to select the old O(n^2) algorithm
2125 for link map file generation, rather than the new O(n) algorithm which uses
2126 about 40% more memory for symbol storage.
2128 Another effect of the switch is to set the default hash table size to
2129 1021, which again saves memory at the cost of lengthening the linker's
2130 run time. This is not done however if the @option{--hash-size} switch
2133 The @option{--reduce-memory-overheads} switch may be also be used to
2134 enable other tradeoffs in future versions of the linker.
2137 @kindex --build-id=@var{style}
2139 @itemx --build-id=@var{style}
2140 Request creation of @code{.note.gnu.build-id} ELF note section.
2141 The contents of the note are unique bits identifying this linked
2142 file. @var{style} can be @code{uuid} to use 128 random bits,
2143 @code{sha1} to use a 160-bit @sc{SHA1} hash on the normative
2144 parts of the output contents, @code{md5} to use a 128-bit
2145 @sc{MD5} hash on the normative parts of the output contents, or
2146 @code{0x@var{hexstring}} to use a chosen bit string specified as
2147 an even number of hexadecimal digits (@code{-} and @code{:}
2148 characters between digit pairs are ignored). If @var{style} is
2149 omitted, @code{sha1} is used.
2151 The @code{md5} and @code{sha1} styles produces an identifier
2152 that is always the same in an identical output file, but will be
2153 unique among all nonidentical output files. It is not intended
2154 to be compared as a checksum for the file's contents. A linked
2155 file may be changed later by other tools, but the build ID bit
2156 string identifying the original linked file does not change.
2158 Passing @code{none} for @var{style} disables the setting from any
2159 @code{--build-id} options earlier on the command line.
2164 @subsection Options Specific to i386 PE Targets
2166 @c man begin OPTIONS
2168 The i386 PE linker supports the @option{-shared} option, which causes
2169 the output to be a dynamically linked library (DLL) instead of a
2170 normal executable. You should name the output @code{*.dll} when you
2171 use this option. In addition, the linker fully supports the standard
2172 @code{*.def} files, which may be specified on the linker command line
2173 like an object file (in fact, it should precede archives it exports
2174 symbols from, to ensure that they get linked in, just like a normal
2177 In addition to the options common to all targets, the i386 PE linker
2178 support additional command line options that are specific to the i386
2179 PE target. Options that take values may be separated from their
2180 values by either a space or an equals sign.
2184 @kindex --add-stdcall-alias
2185 @item --add-stdcall-alias
2186 If given, symbols with a stdcall suffix (@@@var{nn}) will be exported
2187 as-is and also with the suffix stripped.
2188 [This option is specific to the i386 PE targeted port of the linker]
2191 @item --base-file @var{file}
2192 Use @var{file} as the name of a file in which to save the base
2193 addresses of all the relocations needed for generating DLLs with
2195 [This is an i386 PE specific option]
2199 Create a DLL instead of a regular executable. You may also use
2200 @option{-shared} or specify a @code{LIBRARY} in a given @code{.def}
2202 [This option is specific to the i386 PE targeted port of the linker]
2204 @kindex --enable-long-section-names
2205 @kindex --disable-long-section-names
2206 @item --enable-long-section-names
2207 @itemx --disable-long-section-names
2208 The PE variants of the Coff object format add an extension that permits
2209 the use of section names longer than eight characters, the normal limit
2210 for Coff. By default, these names are only allowed in object files, as
2211 fully-linked executable images do not carry the Coff string table required
2212 to support the longer names. As a GNU extension, it is possible to
2213 allow their use in executable images as well, or to (probably pointlessly!)
2214 disallow it in object files, by using these two options. Executable images
2215 generated with these long section names are slightly non-standard, carrying
2216 as they do a string table, and may generate confusing output when examined
2217 with non-GNU PE-aware tools, such as file viewers and dumpers. However,
2218 GDB relies on the use of PE long section names to find Dwarf-2 debug
2219 information sections in an executable image at runtime, and so if neither
2220 option is specified on the command-line, @command{ld} will enable long
2221 section names, overriding the default and technically correct behaviour,
2222 when it finds the presence of debug information while linking an executable
2223 image and not stripping symbols.
2224 [This option is valid for all PE targeted ports of the linker]
2226 @kindex --enable-stdcall-fixup
2227 @kindex --disable-stdcall-fixup
2228 @item --enable-stdcall-fixup
2229 @itemx --disable-stdcall-fixup
2230 If the link finds a symbol that it cannot resolve, it will attempt to
2231 do ``fuzzy linking'' by looking for another defined symbol that differs
2232 only in the format of the symbol name (cdecl vs stdcall) and will
2233 resolve that symbol by linking to the match. For example, the
2234 undefined symbol @code{_foo} might be linked to the function
2235 @code{_foo@@12}, or the undefined symbol @code{_bar@@16} might be linked
2236 to the function @code{_bar}. When the linker does this, it prints a
2237 warning, since it normally should have failed to link, but sometimes
2238 import libraries generated from third-party dlls may need this feature
2239 to be usable. If you specify @option{--enable-stdcall-fixup}, this
2240 feature is fully enabled and warnings are not printed. If you specify
2241 @option{--disable-stdcall-fixup}, this feature is disabled and such
2242 mismatches are considered to be errors.
2243 [This option is specific to the i386 PE targeted port of the linker]
2245 @kindex --leading-underscore
2246 @kindex --no-leading-underscore
2247 @item --leading-underscore
2248 @itemx --no-leading-underscore
2249 For most targets default symbol-prefix is an underscore and is defined
2250 in target's description. By this option it is possible to
2251 disable/enable the default underscore symbol-prefix.
2253 @cindex DLLs, creating
2254 @kindex --export-all-symbols
2255 @item --export-all-symbols
2256 If given, all global symbols in the objects used to build a DLL will
2257 be exported by the DLL. Note that this is the default if there
2258 otherwise wouldn't be any exported symbols. When symbols are
2259 explicitly exported via DEF files or implicitly exported via function
2260 attributes, the default is to not export anything else unless this
2261 option is given. Note that the symbols @code{DllMain@@12},
2262 @code{DllEntryPoint@@0}, @code{DllMainCRTStartup@@12}, and
2263 @code{impure_ptr} will not be automatically
2264 exported. Also, symbols imported from other DLLs will not be
2265 re-exported, nor will symbols specifying the DLL's internal layout
2266 such as those beginning with @code{_head_} or ending with
2267 @code{_iname}. In addition, no symbols from @code{libgcc},
2268 @code{libstd++}, @code{libmingw32}, or @code{crtX.o} will be exported.
2269 Symbols whose names begin with @code{__rtti_} or @code{__builtin_} will
2270 not be exported, to help with C++ DLLs. Finally, there is an
2271 extensive list of cygwin-private symbols that are not exported
2272 (obviously, this applies on when building DLLs for cygwin targets).
2273 These cygwin-excludes are: @code{_cygwin_dll_entry@@12},
2274 @code{_cygwin_crt0_common@@8}, @code{_cygwin_noncygwin_dll_entry@@12},
2275 @code{_fmode}, @code{_impure_ptr}, @code{cygwin_attach_dll},
2276 @code{cygwin_premain0}, @code{cygwin_premain1}, @code{cygwin_premain2},
2277 @code{cygwin_premain3}, and @code{environ}.
2278 [This option is specific to the i386 PE targeted port of the linker]
2280 @kindex --exclude-symbols
2281 @item --exclude-symbols @var{symbol},@var{symbol},...
2282 Specifies a list of symbols which should not be automatically
2283 exported. The symbol names may be delimited by commas or colons.
2284 [This option is specific to the i386 PE targeted port of the linker]
2286 @kindex --exclude-all-symbols
2287 @item --exclude-all-symbols
2288 Specifies no symbols should be automatically exported.
2289 [This option is specific to the i386 PE targeted port of the linker]
2291 @kindex --file-alignment
2292 @item --file-alignment
2293 Specify the file alignment. Sections in the file will always begin at
2294 file offsets which are multiples of this number. This defaults to
2296 [This option is specific to the i386 PE targeted port of the linker]
2300 @item --heap @var{reserve}
2301 @itemx --heap @var{reserve},@var{commit}
2302 Specify the number of bytes of memory to reserve (and optionally commit)
2303 to be used as heap for this program. The default is 1Mb reserved, 4K
2305 [This option is specific to the i386 PE targeted port of the linker]
2308 @kindex --image-base
2309 @item --image-base @var{value}
2310 Use @var{value} as the base address of your program or dll. This is
2311 the lowest memory location that will be used when your program or dll
2312 is loaded. To reduce the need to relocate and improve performance of
2313 your dlls, each should have a unique base address and not overlap any
2314 other dlls. The default is 0x400000 for executables, and 0x10000000
2316 [This option is specific to the i386 PE targeted port of the linker]
2320 If given, the stdcall suffixes (@@@var{nn}) will be stripped from
2321 symbols before they are exported.
2322 [This option is specific to the i386 PE targeted port of the linker]
2324 @kindex --large-address-aware
2325 @item --large-address-aware
2326 If given, the appropriate bit in the ``Characteristics'' field of the COFF
2327 header is set to indicate that this executable supports virtual addresses
2328 greater than 2 gigabytes. This should be used in conjunction with the /3GB
2329 or /USERVA=@var{value} megabytes switch in the ``[operating systems]''
2330 section of the BOOT.INI. Otherwise, this bit has no effect.
2331 [This option is specific to PE targeted ports of the linker]
2333 @kindex --major-image-version
2334 @item --major-image-version @var{value}
2335 Sets the major number of the ``image version''. Defaults to 1.
2336 [This option is specific to the i386 PE targeted port of the linker]
2338 @kindex --major-os-version
2339 @item --major-os-version @var{value}
2340 Sets the major number of the ``os version''. Defaults to 4.
2341 [This option is specific to the i386 PE targeted port of the linker]
2343 @kindex --major-subsystem-version
2344 @item --major-subsystem-version @var{value}
2345 Sets the major number of the ``subsystem version''. Defaults to 4.
2346 [This option is specific to the i386 PE targeted port of the linker]
2348 @kindex --minor-image-version
2349 @item --minor-image-version @var{value}
2350 Sets the minor number of the ``image version''. Defaults to 0.
2351 [This option is specific to the i386 PE targeted port of the linker]
2353 @kindex --minor-os-version
2354 @item --minor-os-version @var{value}
2355 Sets the minor number of the ``os version''. Defaults to 0.
2356 [This option is specific to the i386 PE targeted port of the linker]
2358 @kindex --minor-subsystem-version
2359 @item --minor-subsystem-version @var{value}
2360 Sets the minor number of the ``subsystem version''. Defaults to 0.
2361 [This option is specific to the i386 PE targeted port of the linker]
2363 @cindex DEF files, creating
2364 @cindex DLLs, creating
2365 @kindex --output-def
2366 @item --output-def @var{file}
2367 The linker will create the file @var{file} which will contain a DEF
2368 file corresponding to the DLL the linker is generating. This DEF file
2369 (which should be called @code{*.def}) may be used to create an import
2370 library with @code{dlltool} or may be used as a reference to
2371 automatically or implicitly exported symbols.
2372 [This option is specific to the i386 PE targeted port of the linker]
2374 @cindex DLLs, creating
2375 @kindex --out-implib
2376 @item --out-implib @var{file}
2377 The linker will create the file @var{file} which will contain an
2378 import lib corresponding to the DLL the linker is generating. This
2379 import lib (which should be called @code{*.dll.a} or @code{*.a}
2380 may be used to link clients against the generated DLL; this behaviour
2381 makes it possible to skip a separate @code{dlltool} import library
2383 [This option is specific to the i386 PE targeted port of the linker]
2385 @kindex --enable-auto-image-base
2386 @item --enable-auto-image-base
2387 Automatically choose the image base for DLLs, unless one is specified
2388 using the @code{--image-base} argument. By using a hash generated
2389 from the dllname to create unique image bases for each DLL, in-memory
2390 collisions and relocations which can delay program execution are
2392 [This option is specific to the i386 PE targeted port of the linker]
2394 @kindex --disable-auto-image-base
2395 @item --disable-auto-image-base
2396 Do not automatically generate a unique image base. If there is no
2397 user-specified image base (@code{--image-base}) then use the platform
2399 [This option is specific to the i386 PE targeted port of the linker]
2401 @cindex DLLs, linking to
2402 @kindex --dll-search-prefix
2403 @item --dll-search-prefix @var{string}
2404 When linking dynamically to a dll without an import library,
2405 search for @code{<string><basename>.dll} in preference to
2406 @code{lib<basename>.dll}. This behaviour allows easy distinction
2407 between DLLs built for the various "subplatforms": native, cygwin,
2408 uwin, pw, etc. For instance, cygwin DLLs typically use
2409 @code{--dll-search-prefix=cyg}.
2410 [This option is specific to the i386 PE targeted port of the linker]
2412 @kindex --enable-auto-import
2413 @item --enable-auto-import
2414 Do sophisticated linking of @code{_symbol} to @code{__imp__symbol} for
2415 DATA imports from DLLs, and create the necessary thunking symbols when
2416 building the import libraries with those DATA exports. Note: Use of the
2417 'auto-import' extension will cause the text section of the image file
2418 to be made writable. This does not conform to the PE-COFF format
2419 specification published by Microsoft.
2421 Note - use of the 'auto-import' extension will also cause read only
2422 data which would normally be placed into the .rdata section to be
2423 placed into the .data section instead. This is in order to work
2424 around a problem with consts that is described here:
2425 http://www.cygwin.com/ml/cygwin/2004-09/msg01101.html
2427 Using 'auto-import' generally will 'just work' -- but sometimes you may
2430 "variable '<var>' can't be auto-imported. Please read the
2431 documentation for ld's @code{--enable-auto-import} for details."
2433 This message occurs when some (sub)expression accesses an address
2434 ultimately given by the sum of two constants (Win32 import tables only
2435 allow one). Instances where this may occur include accesses to member
2436 fields of struct variables imported from a DLL, as well as using a
2437 constant index into an array variable imported from a DLL. Any
2438 multiword variable (arrays, structs, long long, etc) may trigger
2439 this error condition. However, regardless of the exact data type
2440 of the offending exported variable, ld will always detect it, issue
2441 the warning, and exit.
2443 There are several ways to address this difficulty, regardless of the
2444 data type of the exported variable:
2446 One way is to use --enable-runtime-pseudo-reloc switch. This leaves the task
2447 of adjusting references in your client code for runtime environment, so
2448 this method works only when runtime environment supports this feature.
2450 A second solution is to force one of the 'constants' to be a variable --
2451 that is, unknown and un-optimizable at compile time. For arrays,
2452 there are two possibilities: a) make the indexee (the array's address)
2453 a variable, or b) make the 'constant' index a variable. Thus:
2456 extern type extern_array[];
2458 @{ volatile type *t=extern_array; t[1] @}
2464 extern type extern_array[];
2466 @{ volatile int t=1; extern_array[t] @}
2469 For structs (and most other multiword data types) the only option
2470 is to make the struct itself (or the long long, or the ...) variable:
2473 extern struct s extern_struct;
2474 extern_struct.field -->
2475 @{ volatile struct s *t=&extern_struct; t->field @}
2481 extern long long extern_ll;
2483 @{ volatile long long * local_ll=&extern_ll; *local_ll @}
2486 A third method of dealing with this difficulty is to abandon
2487 'auto-import' for the offending symbol and mark it with
2488 @code{__declspec(dllimport)}. However, in practise that
2489 requires using compile-time #defines to indicate whether you are
2490 building a DLL, building client code that will link to the DLL, or
2491 merely building/linking to a static library. In making the choice
2492 between the various methods of resolving the 'direct address with
2493 constant offset' problem, you should consider typical real-world usage:
2501 void main(int argc, char **argv)@{
2502 printf("%d\n",arr[1]);
2512 void main(int argc, char **argv)@{
2513 /* This workaround is for win32 and cygwin; do not "optimize" */
2514 volatile int *parr = arr;
2515 printf("%d\n",parr[1]);
2522 /* Note: auto-export is assumed (no __declspec(dllexport)) */
2523 #if (defined(_WIN32) || defined(__CYGWIN__)) && \
2524 !(defined(FOO_BUILD_DLL) || defined(FOO_STATIC))
2525 #define FOO_IMPORT __declspec(dllimport)
2529 extern FOO_IMPORT int arr[];
2532 void main(int argc, char **argv)@{
2533 printf("%d\n",arr[1]);
2537 A fourth way to avoid this problem is to re-code your
2538 library to use a functional interface rather than a data interface
2539 for the offending variables (e.g. set_foo() and get_foo() accessor
2541 [This option is specific to the i386 PE targeted port of the linker]
2543 @kindex --disable-auto-import
2544 @item --disable-auto-import
2545 Do not attempt to do sophisticated linking of @code{_symbol} to
2546 @code{__imp__symbol} for DATA imports from DLLs.
2547 [This option is specific to the i386 PE targeted port of the linker]
2549 @kindex --enable-runtime-pseudo-reloc
2550 @item --enable-runtime-pseudo-reloc
2551 If your code contains expressions described in --enable-auto-import section,
2552 that is, DATA imports from DLL with non-zero offset, this switch will create
2553 a vector of 'runtime pseudo relocations' which can be used by runtime
2554 environment to adjust references to such data in your client code.
2555 [This option is specific to the i386 PE targeted port of the linker]
2557 @kindex --disable-runtime-pseudo-reloc
2558 @item --disable-runtime-pseudo-reloc
2559 Do not create pseudo relocations for non-zero offset DATA imports from
2560 DLLs. This is the default.
2561 [This option is specific to the i386 PE targeted port of the linker]
2563 @kindex --enable-extra-pe-debug
2564 @item --enable-extra-pe-debug
2565 Show additional debug info related to auto-import symbol thunking.
2566 [This option is specific to the i386 PE targeted port of the linker]
2568 @kindex --section-alignment
2569 @item --section-alignment
2570 Sets the section alignment. Sections in memory will always begin at
2571 addresses which are a multiple of this number. Defaults to 0x1000.
2572 [This option is specific to the i386 PE targeted port of the linker]
2576 @item --stack @var{reserve}
2577 @itemx --stack @var{reserve},@var{commit}
2578 Specify the number of bytes of memory to reserve (and optionally commit)
2579 to be used as stack for this program. The default is 2Mb reserved, 4K
2581 [This option is specific to the i386 PE targeted port of the linker]
2584 @item --subsystem @var{which}
2585 @itemx --subsystem @var{which}:@var{major}
2586 @itemx --subsystem @var{which}:@var{major}.@var{minor}
2587 Specifies the subsystem under which your program will execute. The
2588 legal values for @var{which} are @code{native}, @code{windows},
2589 @code{console}, @code{posix}, and @code{xbox}. You may optionally set
2590 the subsystem version also. Numeric values are also accepted for
2592 [This option is specific to the i386 PE targeted port of the linker]
2594 The following options set flags in the @code{DllCharacteristics} field
2595 of the PE file header:
2596 [These options are specific to PE targeted ports of the linker]
2598 @kindex --dynamicbase
2600 The image base address may be relocated using address space layout
2601 randomization (ASLR). This feature was introduced with MS Windows
2602 Vista for i386 PE targets.
2604 @kindex --forceinteg
2606 Code integrity checks are enforced.
2610 The image is compatible with the Data Execution Prevention.
2611 This feature was introduced with MS Windows XP SP2 for i386 PE targets.
2613 @kindex --no-isolation
2614 @item --no-isolation
2615 Although the image understands isolation, do not isolate the image.
2619 The image does not use SEH. No SE handler may be called from
2624 Do not bind this image.
2628 The driver uses the MS Windows Driver Model.
2632 The image is Terminal Server aware.
2639 @subsection Options specific to C6X uClinux targets
2641 @c man begin OPTIONS
2643 The C6X uClinux target uses a binary format called DSBT to support shared
2644 libraries. Each shared library in the system needs to have a unique index;
2645 all executables use an index of 0.
2650 @item --dsbt-size @var{size}
2651 This option sets the number of entires in the DSBT of the current executable
2652 or shared library to @var{size}. The default is to create a table with 64
2655 @kindex --dsbt-index
2656 @item --dsbt-index @var{index}
2657 This option sets the DSBT index of the current executable or shared library
2658 to @var{index}. The default is 0, which is appropriate for generating
2659 executables. If a shared library is generated with a DSBT index of 0, the
2660 @code{R_C6000_DSBT_INDEX} relocs are copied into the output file.
2662 @kindex --no-merge-exidx-entries
2663 The @samp{--no-merge-exidx-entries} switch disables the merging of adjacent
2664 exidx entries in frame unwind info.
2672 @subsection Options specific to Motorola 68HC11 and 68HC12 targets
2674 @c man begin OPTIONS
2676 The 68HC11 and 68HC12 linkers support specific options to control the
2677 memory bank switching mapping and trampoline code generation.
2681 @kindex --no-trampoline
2682 @item --no-trampoline
2683 This option disables the generation of trampoline. By default a trampoline
2684 is generated for each far function which is called using a @code{jsr}
2685 instruction (this happens when a pointer to a far function is taken).
2687 @kindex --bank-window
2688 @item --bank-window @var{name}
2689 This option indicates to the linker the name of the memory region in
2690 the @samp{MEMORY} specification that describes the memory bank window.
2691 The definition of such region is then used by the linker to compute
2692 paging and addresses within the memory window.
2700 @subsection Options specific to Motorola 68K target
2702 @c man begin OPTIONS
2704 The following options are supported to control handling of GOT generation
2705 when linking for 68K targets.
2710 @item --got=@var{type}
2711 This option tells the linker which GOT generation scheme to use.
2712 @var{type} should be one of @samp{single}, @samp{negative},
2713 @samp{multigot} or @samp{target}. For more information refer to the
2714 Info entry for @file{ld}.
2723 @section Environment Variables
2725 @c man begin ENVIRONMENT
2727 You can change the behaviour of @command{ld} with the environment variables
2728 @ifclear SingleFormat
2731 @code{LDEMULATION} and @code{COLLECT_NO_DEMANGLE}.
2733 @ifclear SingleFormat
2735 @cindex default input format
2736 @code{GNUTARGET} determines the input-file object format if you don't
2737 use @samp{-b} (or its synonym @samp{--format}). Its value should be one
2738 of the BFD names for an input format (@pxref{BFD}). If there is no
2739 @code{GNUTARGET} in the environment, @command{ld} uses the natural format
2740 of the target. If @code{GNUTARGET} is set to @code{default} then BFD
2741 attempts to discover the input format by examining binary input files;
2742 this method often succeeds, but there are potential ambiguities, since
2743 there is no method of ensuring that the magic number used to specify
2744 object-file formats is unique. However, the configuration procedure for
2745 BFD on each system places the conventional format for that system first
2746 in the search-list, so ambiguities are resolved in favor of convention.
2750 @cindex default emulation
2751 @cindex emulation, default
2752 @code{LDEMULATION} determines the default emulation if you don't use the
2753 @samp{-m} option. The emulation can affect various aspects of linker
2754 behaviour, particularly the default linker script. You can list the
2755 available emulations with the @samp{--verbose} or @samp{-V} options. If
2756 the @samp{-m} option is not used, and the @code{LDEMULATION} environment
2757 variable is not defined, the default emulation depends upon how the
2758 linker was configured.
2760 @kindex COLLECT_NO_DEMANGLE
2761 @cindex demangling, default
2762 Normally, the linker will default to demangling symbols. However, if
2763 @code{COLLECT_NO_DEMANGLE} is set in the environment, then it will
2764 default to not demangling symbols. This environment variable is used in
2765 a similar fashion by the @code{gcc} linker wrapper program. The default
2766 may be overridden by the @samp{--demangle} and @samp{--no-demangle}
2773 @chapter Linker Scripts
2776 @cindex linker scripts
2777 @cindex command files
2778 Every link is controlled by a @dfn{linker script}. This script is
2779 written in the linker command language.
2781 The main purpose of the linker script is to describe how the sections in
2782 the input files should be mapped into the output file, and to control
2783 the memory layout of the output file. Most linker scripts do nothing
2784 more than this. However, when necessary, the linker script can also
2785 direct the linker to perform many other operations, using the commands
2788 The linker always uses a linker script. If you do not supply one
2789 yourself, the linker will use a default script that is compiled into the
2790 linker executable. You can use the @samp{--verbose} command line option
2791 to display the default linker script. Certain command line options,
2792 such as @samp{-r} or @samp{-N}, will affect the default linker script.
2794 You may supply your own linker script by using the @samp{-T} command
2795 line option. When you do this, your linker script will replace the
2796 default linker script.
2798 You may also use linker scripts implicitly by naming them as input files
2799 to the linker, as though they were files to be linked. @xref{Implicit
2803 * Basic Script Concepts:: Basic Linker Script Concepts
2804 * Script Format:: Linker Script Format
2805 * Simple Example:: Simple Linker Script Example
2806 * Simple Commands:: Simple Linker Script Commands
2807 * Assignments:: Assigning Values to Symbols
2808 * SECTIONS:: SECTIONS Command
2809 * MEMORY:: MEMORY Command
2810 * PHDRS:: PHDRS Command
2811 * VERSION:: VERSION Command
2812 * Expressions:: Expressions in Linker Scripts
2813 * Implicit Linker Scripts:: Implicit Linker Scripts
2816 @node Basic Script Concepts
2817 @section Basic Linker Script Concepts
2818 @cindex linker script concepts
2819 We need to define some basic concepts and vocabulary in order to
2820 describe the linker script language.
2822 The linker combines input files into a single output file. The output
2823 file and each input file are in a special data format known as an
2824 @dfn{object file format}. Each file is called an @dfn{object file}.
2825 The output file is often called an @dfn{executable}, but for our
2826 purposes we will also call it an object file. Each object file has,
2827 among other things, a list of @dfn{sections}. We sometimes refer to a
2828 section in an input file as an @dfn{input section}; similarly, a section
2829 in the output file is an @dfn{output section}.
2831 Each section in an object file has a name and a size. Most sections
2832 also have an associated block of data, known as the @dfn{section
2833 contents}. A section may be marked as @dfn{loadable}, which mean that
2834 the contents should be loaded into memory when the output file is run.
2835 A section with no contents may be @dfn{allocatable}, which means that an
2836 area in memory should be set aside, but nothing in particular should be
2837 loaded there (in some cases this memory must be zeroed out). A section
2838 which is neither loadable nor allocatable typically contains some sort
2839 of debugging information.
2841 Every loadable or allocatable output section has two addresses. The
2842 first is the @dfn{VMA}, or virtual memory address. This is the address
2843 the section will have when the output file is run. The second is the
2844 @dfn{LMA}, or load memory address. This is the address at which the
2845 section will be loaded. In most cases the two addresses will be the
2846 same. An example of when they might be different is when a data section
2847 is loaded into ROM, and then copied into RAM when the program starts up
2848 (this technique is often used to initialize global variables in a ROM
2849 based system). In this case the ROM address would be the LMA, and the
2850 RAM address would be the VMA.
2852 You can see the sections in an object file by using the @code{objdump}
2853 program with the @samp{-h} option.
2855 Every object file also has a list of @dfn{symbols}, known as the
2856 @dfn{symbol table}. A symbol may be defined or undefined. Each symbol
2857 has a name, and each defined symbol has an address, among other
2858 information. If you compile a C or C++ program into an object file, you
2859 will get a defined symbol for every defined function and global or
2860 static variable. Every undefined function or global variable which is
2861 referenced in the input file will become an undefined symbol.
2863 You can see the symbols in an object file by using the @code{nm}
2864 program, or by using the @code{objdump} program with the @samp{-t}
2868 @section Linker Script Format
2869 @cindex linker script format
2870 Linker scripts are text files.
2872 You write a linker script as a series of commands. Each command is
2873 either a keyword, possibly followed by arguments, or an assignment to a
2874 symbol. You may separate commands using semicolons. Whitespace is
2877 Strings such as file or format names can normally be entered directly.
2878 If the file name contains a character such as a comma which would
2879 otherwise serve to separate file names, you may put the file name in
2880 double quotes. There is no way to use a double quote character in a
2883 You may include comments in linker scripts just as in C, delimited by
2884 @samp{/*} and @samp{*/}. As in C, comments are syntactically equivalent
2887 @node Simple Example
2888 @section Simple Linker Script Example
2889 @cindex linker script example
2890 @cindex example of linker script
2891 Many linker scripts are fairly simple.
2893 The simplest possible linker script has just one command:
2894 @samp{SECTIONS}. You use the @samp{SECTIONS} command to describe the
2895 memory layout of the output file.
2897 The @samp{SECTIONS} command is a powerful command. Here we will
2898 describe a simple use of it. Let's assume your program consists only of
2899 code, initialized data, and uninitialized data. These will be in the
2900 @samp{.text}, @samp{.data}, and @samp{.bss} sections, respectively.
2901 Let's assume further that these are the only sections which appear in
2904 For this example, let's say that the code should be loaded at address
2905 0x10000, and that the data should start at address 0x8000000. Here is a
2906 linker script which will do that:
2911 .text : @{ *(.text) @}
2913 .data : @{ *(.data) @}
2914 .bss : @{ *(.bss) @}
2918 You write the @samp{SECTIONS} command as the keyword @samp{SECTIONS},
2919 followed by a series of symbol assignments and output section
2920 descriptions enclosed in curly braces.
2922 The first line inside the @samp{SECTIONS} command of the above example
2923 sets the value of the special symbol @samp{.}, which is the location
2924 counter. If you do not specify the address of an output section in some
2925 other way (other ways are described later), the address is set from the
2926 current value of the location counter. The location counter is then
2927 incremented by the size of the output section. At the start of the
2928 @samp{SECTIONS} command, the location counter has the value @samp{0}.
2930 The second line defines an output section, @samp{.text}. The colon is
2931 required syntax which may be ignored for now. Within the curly braces
2932 after the output section name, you list the names of the input sections
2933 which should be placed into this output section. The @samp{*} is a
2934 wildcard which matches any file name. The expression @samp{*(.text)}
2935 means all @samp{.text} input sections in all input files.
2937 Since the location counter is @samp{0x10000} when the output section
2938 @samp{.text} is defined, the linker will set the address of the
2939 @samp{.text} section in the output file to be @samp{0x10000}.
2941 The remaining lines define the @samp{.data} and @samp{.bss} sections in
2942 the output file. The linker will place the @samp{.data} output section
2943 at address @samp{0x8000000}. After the linker places the @samp{.data}
2944 output section, the value of the location counter will be
2945 @samp{0x8000000} plus the size of the @samp{.data} output section. The
2946 effect is that the linker will place the @samp{.bss} output section
2947 immediately after the @samp{.data} output section in memory.
2949 The linker will ensure that each output section has the required
2950 alignment, by increasing the location counter if necessary. In this
2951 example, the specified addresses for the @samp{.text} and @samp{.data}
2952 sections will probably satisfy any alignment constraints, but the linker
2953 may have to create a small gap between the @samp{.data} and @samp{.bss}
2956 That's it! That's a simple and complete linker script.
2958 @node Simple Commands
2959 @section Simple Linker Script Commands
2960 @cindex linker script simple commands
2961 In this section we describe the simple linker script commands.
2964 * Entry Point:: Setting the entry point
2965 * File Commands:: Commands dealing with files
2966 @ifclear SingleFormat
2967 * Format Commands:: Commands dealing with object file formats
2970 * REGION_ALIAS:: Assign alias names to memory regions
2971 * Miscellaneous Commands:: Other linker script commands
2975 @subsection Setting the Entry Point
2976 @kindex ENTRY(@var{symbol})
2977 @cindex start of execution
2978 @cindex first instruction
2980 The first instruction to execute in a program is called the @dfn{entry
2981 point}. You can use the @code{ENTRY} linker script command to set the
2982 entry point. The argument is a symbol name:
2987 There are several ways to set the entry point. The linker will set the
2988 entry point by trying each of the following methods in order, and
2989 stopping when one of them succeeds:
2992 the @samp{-e} @var{entry} command-line option;
2994 the @code{ENTRY(@var{symbol})} command in a linker script;
2996 the value of a target specific symbol, if it is defined; For many
2997 targets this is @code{start}, but PE and BeOS based systems for example
2998 check a list of possible entry symbols, matching the first one found.
3000 the address of the first byte of the @samp{.text} section, if present;
3002 The address @code{0}.
3006 @subsection Commands Dealing with Files
3007 @cindex linker script file commands
3008 Several linker script commands deal with files.
3011 @item INCLUDE @var{filename}
3012 @kindex INCLUDE @var{filename}
3013 @cindex including a linker script
3014 Include the linker script @var{filename} at this point. The file will
3015 be searched for in the current directory, and in any directory specified
3016 with the @option{-L} option. You can nest calls to @code{INCLUDE} up to
3019 You can place @code{INCLUDE} directives at the top level, in @code{MEMORY} or
3020 @code{SECTIONS} commands, or in output section descriptions.
3022 @item INPUT(@var{file}, @var{file}, @dots{})
3023 @itemx INPUT(@var{file} @var{file} @dots{})
3024 @kindex INPUT(@var{files})
3025 @cindex input files in linker scripts
3026 @cindex input object files in linker scripts
3027 @cindex linker script input object files
3028 The @code{INPUT} command directs the linker to include the named files
3029 in the link, as though they were named on the command line.
3031 For example, if you always want to include @file{subr.o} any time you do
3032 a link, but you can't be bothered to put it on every link command line,
3033 then you can put @samp{INPUT (subr.o)} in your linker script.
3035 In fact, if you like, you can list all of your input files in the linker
3036 script, and then invoke the linker with nothing but a @samp{-T} option.
3038 In case a @dfn{sysroot prefix} is configured, and the filename starts
3039 with the @samp{/} character, and the script being processed was
3040 located inside the @dfn{sysroot prefix}, the filename will be looked
3041 for in the @dfn{sysroot prefix}. Otherwise, the linker will try to
3042 open the file in the current directory. If it is not found, the
3043 linker will search through the archive library search path. See the
3044 description of @samp{-L} in @ref{Options,,Command Line Options}.
3046 If you use @samp{INPUT (-l@var{file})}, @command{ld} will transform the
3047 name to @code{lib@var{file}.a}, as with the command line argument
3050 When you use the @code{INPUT} command in an implicit linker script, the
3051 files will be included in the link at the point at which the linker
3052 script file is included. This can affect archive searching.
3054 @item GROUP(@var{file}, @var{file}, @dots{})
3055 @itemx GROUP(@var{file} @var{file} @dots{})
3056 @kindex GROUP(@var{files})
3057 @cindex grouping input files
3058 The @code{GROUP} command is like @code{INPUT}, except that the named
3059 files should all be archives, and they are searched repeatedly until no
3060 new undefined references are created. See the description of @samp{-(}
3061 in @ref{Options,,Command Line Options}.
3063 @item AS_NEEDED(@var{file}, @var{file}, @dots{})
3064 @itemx AS_NEEDED(@var{file} @var{file} @dots{})
3065 @kindex AS_NEEDED(@var{files})
3066 This construct can appear only inside of the @code{INPUT} or @code{GROUP}
3067 commands, among other filenames. The files listed will be handled
3068 as if they appear directly in the @code{INPUT} or @code{GROUP} commands,
3069 with the exception of ELF shared libraries, that will be added only
3070 when they are actually needed. This construct essentially enables
3071 @option{--as-needed} option for all the files listed inside of it
3072 and restores previous @option{--as-needed} resp. @option{--no-as-needed}
3075 @item OUTPUT(@var{filename})
3076 @kindex OUTPUT(@var{filename})
3077 @cindex output file name in linker script
3078 The @code{OUTPUT} command names the output file. Using
3079 @code{OUTPUT(@var{filename})} in the linker script is exactly like using
3080 @samp{-o @var{filename}} on the command line (@pxref{Options,,Command
3081 Line Options}). If both are used, the command line option takes
3084 You can use the @code{OUTPUT} command to define a default name for the
3085 output file other than the usual default of @file{a.out}.
3087 @item SEARCH_DIR(@var{path})
3088 @kindex SEARCH_DIR(@var{path})
3089 @cindex library search path in linker script
3090 @cindex archive search path in linker script
3091 @cindex search path in linker script
3092 The @code{SEARCH_DIR} command adds @var{path} to the list of paths where
3093 @command{ld} looks for archive libraries. Using
3094 @code{SEARCH_DIR(@var{path})} is exactly like using @samp{-L @var{path}}
3095 on the command line (@pxref{Options,,Command Line Options}). If both
3096 are used, then the linker will search both paths. Paths specified using
3097 the command line option are searched first.
3099 @item STARTUP(@var{filename})
3100 @kindex STARTUP(@var{filename})
3101 @cindex first input file
3102 The @code{STARTUP} command is just like the @code{INPUT} command, except
3103 that @var{filename} will become the first input file to be linked, as
3104 though it were specified first on the command line. This may be useful
3105 when using a system in which the entry point is always the start of the
3109 @ifclear SingleFormat
3110 @node Format Commands
3111 @subsection Commands Dealing with Object File Formats
3112 A couple of linker script commands deal with object file formats.
3115 @item OUTPUT_FORMAT(@var{bfdname})
3116 @itemx OUTPUT_FORMAT(@var{default}, @var{big}, @var{little})
3117 @kindex OUTPUT_FORMAT(@var{bfdname})
3118 @cindex output file format in linker script
3119 The @code{OUTPUT_FORMAT} command names the BFD format to use for the
3120 output file (@pxref{BFD}). Using @code{OUTPUT_FORMAT(@var{bfdname})} is
3121 exactly like using @samp{--oformat @var{bfdname}} on the command line
3122 (@pxref{Options,,Command Line Options}). If both are used, the command
3123 line option takes precedence.
3125 You can use @code{OUTPUT_FORMAT} with three arguments to use different
3126 formats based on the @samp{-EB} and @samp{-EL} command line options.
3127 This permits the linker script to set the output format based on the
3130 If neither @samp{-EB} nor @samp{-EL} are used, then the output format
3131 will be the first argument, @var{default}. If @samp{-EB} is used, the
3132 output format will be the second argument, @var{big}. If @samp{-EL} is
3133 used, the output format will be the third argument, @var{little}.
3135 For example, the default linker script for the MIPS ELF target uses this
3138 OUTPUT_FORMAT(elf32-bigmips, elf32-bigmips, elf32-littlemips)
3140 This says that the default format for the output file is
3141 @samp{elf32-bigmips}, but if the user uses the @samp{-EL} command line
3142 option, the output file will be created in the @samp{elf32-littlemips}
3145 @item TARGET(@var{bfdname})
3146 @kindex TARGET(@var{bfdname})
3147 @cindex input file format in linker script
3148 The @code{TARGET} command names the BFD format to use when reading input
3149 files. It affects subsequent @code{INPUT} and @code{GROUP} commands.
3150 This command is like using @samp{-b @var{bfdname}} on the command line
3151 (@pxref{Options,,Command Line Options}). If the @code{TARGET} command
3152 is used but @code{OUTPUT_FORMAT} is not, then the last @code{TARGET}
3153 command is also used to set the format for the output file. @xref{BFD}.
3158 @subsection Assign alias names to memory regions
3159 @kindex REGION_ALIAS(@var{alias}, @var{region})
3160 @cindex region alias
3161 @cindex region names
3163 Alias names can be added to existing memory regions created with the
3164 @ref{MEMORY} command. Each name corresponds to at most one memory region.
3167 REGION_ALIAS(@var{alias}, @var{region})
3170 The @code{REGION_ALIAS} function creates an alias name @var{alias} for the
3171 memory region @var{region}. This allows a flexible mapping of output sections
3172 to memory regions. An example follows.
3174 Suppose we have an application for embedded systems which come with various
3175 memory storage devices. All have a general purpose, volatile memory @code{RAM}
3176 that allows code execution or data storage. Some may have a read-only,
3177 non-volatile memory @code{ROM} that allows code execution and read-only data
3178 access. The last variant is a read-only, non-volatile memory @code{ROM2} with
3179 read-only data access and no code execution capability. We have four output
3184 @code{.text} program code;
3186 @code{.rodata} read-only data;
3188 @code{.data} read-write initialized data;
3190 @code{.bss} read-write zero initialized data.
3193 The goal is to provide a linker command file that contains a system independent
3194 part defining the output sections and a system dependent part mapping the
3195 output sections to the memory regions available on the system. Our embedded
3196 systems come with three different memory setups @code{A}, @code{B} and
3198 @multitable @columnfractions .25 .25 .25 .25
3199 @item Section @tab Variant A @tab Variant B @tab Variant C
3200 @item .text @tab RAM @tab ROM @tab ROM
3201 @item .rodata @tab RAM @tab ROM @tab ROM2
3202 @item .data @tab RAM @tab RAM/ROM @tab RAM/ROM2
3203 @item .bss @tab RAM @tab RAM @tab RAM
3205 The notation @code{RAM/ROM} or @code{RAM/ROM2} means that this section is
3206 loaded into region @code{ROM} or @code{ROM2} respectively. Please note that
3207 the load address of the @code{.data} section starts in all three variants at
3208 the end of the @code{.rodata} section.
3210 The base linker script that deals with the output sections follows. It
3211 includes the system dependent @code{linkcmds.memory} file that describes the
3214 INCLUDE linkcmds.memory
3227 .data : AT (rodata_end)
3232 data_size = SIZEOF(.data);
3233 data_load_start = LOADADDR(.data);
3241 Now we need three different @code{linkcmds.memory} files to define memory
3242 regions and alias names. The content of @code{linkcmds.memory} for the three
3243 variants @code{A}, @code{B} and @code{C}:
3246 Here everything goes into the @code{RAM}.
3250 RAM : ORIGIN = 0, LENGTH = 4M
3253 REGION_ALIAS("REGION_TEXT", RAM);
3254 REGION_ALIAS("REGION_RODATA", RAM);
3255 REGION_ALIAS("REGION_DATA", RAM);
3256 REGION_ALIAS("REGION_BSS", RAM);
3259 Program code and read-only data go into the @code{ROM}. Read-write data goes
3260 into the @code{RAM}. An image of the initialized data is loaded into the
3261 @code{ROM} and will be copied during system start into the @code{RAM}.
3265 ROM : ORIGIN = 0, LENGTH = 3M
3266 RAM : ORIGIN = 0x10000000, LENGTH = 1M
3269 REGION_ALIAS("REGION_TEXT", ROM);
3270 REGION_ALIAS("REGION_RODATA", ROM);
3271 REGION_ALIAS("REGION_DATA", RAM);
3272 REGION_ALIAS("REGION_BSS", RAM);
3275 Program code goes into the @code{ROM}. Read-only data goes into the
3276 @code{ROM2}. Read-write data goes into the @code{RAM}. An image of the
3277 initialized data is loaded into the @code{ROM2} and will be copied during
3278 system start into the @code{RAM}.
3282 ROM : ORIGIN = 0, LENGTH = 2M
3283 ROM2 : ORIGIN = 0x10000000, LENGTH = 1M
3284 RAM : ORIGIN = 0x20000000, LENGTH = 1M
3287 REGION_ALIAS("REGION_TEXT", ROM);
3288 REGION_ALIAS("REGION_RODATA", ROM2);
3289 REGION_ALIAS("REGION_DATA", RAM);
3290 REGION_ALIAS("REGION_BSS", RAM);
3294 It is possible to write a common system initialization routine to copy the
3295 @code{.data} section from @code{ROM} or @code{ROM2} into the @code{RAM} if
3300 extern char data_start [];
3301 extern char data_size [];
3302 extern char data_load_start [];
3304 void copy_data(void)
3306 if (data_start != data_load_start)
3308 memcpy(data_start, data_load_start, (size_t) data_size);
3313 @node Miscellaneous Commands
3314 @subsection Other Linker Script Commands
3315 There are a few other linker scripts commands.
3318 @item ASSERT(@var{exp}, @var{message})
3320 @cindex assertion in linker script
3321 Ensure that @var{exp} is non-zero. If it is zero, then exit the linker
3322 with an error code, and print @var{message}.
3324 @item EXTERN(@var{symbol} @var{symbol} @dots{})
3326 @cindex undefined symbol in linker script
3327 Force @var{symbol} to be entered in the output file as an undefined
3328 symbol. Doing this may, for example, trigger linking of additional
3329 modules from standard libraries. You may list several @var{symbol}s for
3330 each @code{EXTERN}, and you may use @code{EXTERN} multiple times. This
3331 command has the same effect as the @samp{-u} command-line option.
3333 @item FORCE_COMMON_ALLOCATION
3334 @kindex FORCE_COMMON_ALLOCATION
3335 @cindex common allocation in linker script
3336 This command has the same effect as the @samp{-d} command-line option:
3337 to make @command{ld} assign space to common symbols even if a relocatable
3338 output file is specified (@samp{-r}).
3340 @item INHIBIT_COMMON_ALLOCATION
3341 @kindex INHIBIT_COMMON_ALLOCATION
3342 @cindex common allocation in linker script
3343 This command has the same effect as the @samp{--no-define-common}
3344 command-line option: to make @code{ld} omit the assignment of addresses
3345 to common symbols even for a non-relocatable output file.
3347 @item INSERT [ AFTER | BEFORE ] @var{output_section}
3349 @cindex insert user script into default script
3350 This command is typically used in a script specified by @samp{-T} to
3351 augment the default @code{SECTIONS} with, for example, overlays. It
3352 inserts all prior linker script statements after (or before)
3353 @var{output_section}, and also causes @samp{-T} to not override the
3354 default linker script. The exact insertion point is as for orphan
3355 sections. @xref{Location Counter}. The insertion happens after the
3356 linker has mapped input sections to output sections. Prior to the
3357 insertion, since @samp{-T} scripts are parsed before the default
3358 linker script, statements in the @samp{-T} script occur before the
3359 default linker script statements in the internal linker representation
3360 of the script. In particular, input section assignments will be made
3361 to @samp{-T} output sections before those in the default script. Here
3362 is an example of how a @samp{-T} script using @code{INSERT} might look:
3369 .ov1 @{ ov1*(.text) @}
3370 .ov2 @{ ov2*(.text) @}
3376 @item NOCROSSREFS(@var{section} @var{section} @dots{})
3377 @kindex NOCROSSREFS(@var{sections})
3378 @cindex cross references
3379 This command may be used to tell @command{ld} to issue an error about any
3380 references among certain output sections.
3382 In certain types of programs, particularly on embedded systems when
3383 using overlays, when one section is loaded into memory, another section
3384 will not be. Any direct references between the two sections would be
3385 errors. For example, it would be an error if code in one section called
3386 a function defined in the other section.
3388 The @code{NOCROSSREFS} command takes a list of output section names. If
3389 @command{ld} detects any cross references between the sections, it reports
3390 an error and returns a non-zero exit status. Note that the
3391 @code{NOCROSSREFS} command uses output section names, not input section
3394 @ifclear SingleFormat
3395 @item OUTPUT_ARCH(@var{bfdarch})
3396 @kindex OUTPUT_ARCH(@var{bfdarch})
3397 @cindex machine architecture
3398 @cindex architecture
3399 Specify a particular output machine architecture. The argument is one
3400 of the names used by the BFD library (@pxref{BFD}). You can see the
3401 architecture of an object file by using the @code{objdump} program with
3402 the @samp{-f} option.
3405 @item LD_FEATURE(@var{string})
3406 @kindex LD_FEATURE(@var{string})
3407 This command may be used to modify @command{ld} behavior. If
3408 @var{string} is @code{"SANE_EXPR"} then absolute symbols and numbers
3409 in a script are simply treated as numbers everywhere.
3410 @xref{Expression Section}.
3414 @section Assigning Values to Symbols
3415 @cindex assignment in scripts
3416 @cindex symbol definition, scripts
3417 @cindex variables, defining
3418 You may assign a value to a symbol in a linker script. This will define
3419 the symbol and place it into the symbol table with a global scope.
3422 * Simple Assignments:: Simple Assignments
3425 * PROVIDE_HIDDEN:: PROVIDE_HIDDEN
3426 * Source Code Reference:: How to use a linker script defined symbol in source code
3429 @node Simple Assignments
3430 @subsection Simple Assignments
3432 You may assign to a symbol using any of the C assignment operators:
3435 @item @var{symbol} = @var{expression} ;
3436 @itemx @var{symbol} += @var{expression} ;
3437 @itemx @var{symbol} -= @var{expression} ;
3438 @itemx @var{symbol} *= @var{expression} ;
3439 @itemx @var{symbol} /= @var{expression} ;
3440 @itemx @var{symbol} <<= @var{expression} ;
3441 @itemx @var{symbol} >>= @var{expression} ;
3442 @itemx @var{symbol} &= @var{expression} ;
3443 @itemx @var{symbol} |= @var{expression} ;
3446 The first case will define @var{symbol} to the value of
3447 @var{expression}. In the other cases, @var{symbol} must already be
3448 defined, and the value will be adjusted accordingly.
3450 The special symbol name @samp{.} indicates the location counter. You
3451 may only use this within a @code{SECTIONS} command. @xref{Location Counter}.
3453 The semicolon after @var{expression} is required.
3455 Expressions are defined below; see @ref{Expressions}.
3457 You may write symbol assignments as commands in their own right, or as
3458 statements within a @code{SECTIONS} command, or as part of an output
3459 section description in a @code{SECTIONS} command.
3461 The section of the symbol will be set from the section of the
3462 expression; for more information, see @ref{Expression Section}.
3464 Here is an example showing the three different places that symbol
3465 assignments may be used:
3476 _bdata = (. + 3) & ~ 3;
3477 .data : @{ *(.data) @}
3481 In this example, the symbol @samp{floating_point} will be defined as
3482 zero. The symbol @samp{_etext} will be defined as the address following
3483 the last @samp{.text} input section. The symbol @samp{_bdata} will be
3484 defined as the address following the @samp{.text} output section aligned
3485 upward to a 4 byte boundary.
3490 For ELF targeted ports, define a symbol that will be hidden and won't be
3491 exported. The syntax is @code{HIDDEN(@var{symbol} = @var{expression})}.
3493 Here is the example from @ref{Simple Assignments}, rewritten to use
3497 HIDDEN(floating_point = 0);
3505 HIDDEN(_bdata = (. + 3) & ~ 3);
3506 .data : @{ *(.data) @}
3510 In this case none of the three symbols will be visible outside this module.
3515 In some cases, it is desirable for a linker script to define a symbol
3516 only if it is referenced and is not defined by any object included in
3517 the link. For example, traditional linkers defined the symbol
3518 @samp{etext}. However, ANSI C requires that the user be able to use
3519 @samp{etext} as a function name without encountering an error. The
3520 @code{PROVIDE} keyword may be used to define a symbol, such as
3521 @samp{etext}, only if it is referenced but not defined. The syntax is
3522 @code{PROVIDE(@var{symbol} = @var{expression})}.
3524 Here is an example of using @code{PROVIDE} to define @samp{etext}:
3537 In this example, if the program defines @samp{_etext} (with a leading
3538 underscore), the linker will give a multiple definition error. If, on
3539 the other hand, the program defines @samp{etext} (with no leading
3540 underscore), the linker will silently use the definition in the program.
3541 If the program references @samp{etext} but does not define it, the
3542 linker will use the definition in the linker script.
3544 @node PROVIDE_HIDDEN
3545 @subsection PROVIDE_HIDDEN
3546 @cindex PROVIDE_HIDDEN
3547 Similar to @code{PROVIDE}. For ELF targeted ports, the symbol will be
3548 hidden and won't be exported.
3550 @node Source Code Reference
3551 @subsection Source Code Reference
3553 Accessing a linker script defined variable from source code is not
3554 intuitive. In particular a linker script symbol is not equivalent to
3555 a variable declaration in a high level language, it is instead a
3556 symbol that does not have a value.
3558 Before going further, it is important to note that compilers often
3559 transform names in the source code into different names when they are
3560 stored in the symbol table. For example, Fortran compilers commonly
3561 prepend or append an underscore, and C++ performs extensive @samp{name
3562 mangling}. Therefore there might be a discrepancy between the name
3563 of a variable as it is used in source code and the name of the same
3564 variable as it is defined in a linker script. For example in C a
3565 linker script variable might be referred to as:
3571 But in the linker script it might be defined as:
3577 In the remaining examples however it is assumed that no name
3578 transformation has taken place.
3580 When a symbol is declared in a high level language such as C, two
3581 things happen. The first is that the compiler reserves enough space
3582 in the program's memory to hold the @emph{value} of the symbol. The
3583 second is that the compiler creates an entry in the program's symbol
3584 table which holds the symbol's @emph{address}. ie the symbol table
3585 contains the address of the block of memory holding the symbol's
3586 value. So for example the following C declaration, at file scope:
3592 creates a entry called @samp{foo} in the symbol table. This entry
3593 holds the address of an @samp{int} sized block of memory where the
3594 number 1000 is initially stored.
3596 When a program references a symbol the compiler generates code that
3597 first accesses the symbol table to find the address of the symbol's
3598 memory block and then code to read the value from that memory block.
3605 looks up the symbol @samp{foo} in the symbol table, gets the address
3606 associated with this symbol and then writes the value 1 into that
3613 looks up the symbol @samp{foo} in the symbol table, gets it address
3614 and then copies this address into the block of memory associated with
3615 the variable @samp{a}.
3617 Linker scripts symbol declarations, by contrast, create an entry in
3618 the symbol table but do not assign any memory to them. Thus they are
3619 an address without a value. So for example the linker script definition:
3625 creates an entry in the symbol table called @samp{foo} which holds
3626 the address of memory location 1000, but nothing special is stored at
3627 address 1000. This means that you cannot access the @emph{value} of a
3628 linker script defined symbol - it has no value - all you can do is
3629 access the @emph{address} of a linker script defined symbol.
3631 Hence when you are using a linker script defined symbol in source code
3632 you should always take the address of the symbol, and never attempt to
3633 use its value. For example suppose you want to copy the contents of a
3634 section of memory called .ROM into a section called .FLASH and the
3635 linker script contains these declarations:
3639 start_of_ROM = .ROM;
3640 end_of_ROM = .ROM + sizeof (.ROM) - 1;
3641 start_of_FLASH = .FLASH;
3645 Then the C source code to perform the copy would be:
3649 extern char start_of_ROM, end_of_ROM, start_of_FLASH;
3651 memcpy (& start_of_FLASH, & start_of_ROM, & end_of_ROM - & start_of_ROM);
3655 Note the use of the @samp{&} operators. These are correct.
3658 @section SECTIONS Command
3660 The @code{SECTIONS} command tells the linker how to map input sections
3661 into output sections, and how to place the output sections in memory.
3663 The format of the @code{SECTIONS} command is:
3667 @var{sections-command}
3668 @var{sections-command}
3673 Each @var{sections-command} may of be one of the following:
3677 an @code{ENTRY} command (@pxref{Entry Point,,Entry command})
3679 a symbol assignment (@pxref{Assignments})
3681 an output section description
3683 an overlay description
3686 The @code{ENTRY} command and symbol assignments are permitted inside the
3687 @code{SECTIONS} command for convenience in using the location counter in
3688 those commands. This can also make the linker script easier to
3689 understand because you can use those commands at meaningful points in
3690 the layout of the output file.
3692 Output section descriptions and overlay descriptions are described
3695 If you do not use a @code{SECTIONS} command in your linker script, the
3696 linker will place each input section into an identically named output
3697 section in the order that the sections are first encountered in the
3698 input files. If all input sections are present in the first file, for
3699 example, the order of sections in the output file will match the order
3700 in the first input file. The first section will be at address zero.
3703 * Output Section Description:: Output section description
3704 * Output Section Name:: Output section name
3705 * Output Section Address:: Output section address
3706 * Input Section:: Input section description
3707 * Output Section Data:: Output section data
3708 * Output Section Keywords:: Output section keywords
3709 * Output Section Discarding:: Output section discarding
3710 * Output Section Attributes:: Output section attributes
3711 * Overlay Description:: Overlay description
3714 @node Output Section Description
3715 @subsection Output Section Description
3716 The full description of an output section looks like this:
3719 @var{section} [@var{address}] [(@var{type})] :
3721 [ALIGN(@var{section_align})]
3722 [SUBALIGN(@var{subsection_align})]
3725 @var{output-section-command}
3726 @var{output-section-command}
3728 @} [>@var{region}] [AT>@var{lma_region}] [:@var{phdr} :@var{phdr} @dots{}] [=@var{fillexp}]
3732 Most output sections do not use most of the optional section attributes.
3734 The whitespace around @var{section} is required, so that the section
3735 name is unambiguous. The colon and the curly braces are also required.
3736 The line breaks and other white space are optional.
3738 Each @var{output-section-command} may be one of the following:
3742 a symbol assignment (@pxref{Assignments})
3744 an input section description (@pxref{Input Section})
3746 data values to include directly (@pxref{Output Section Data})
3748 a special output section keyword (@pxref{Output Section Keywords})
3751 @node Output Section Name
3752 @subsection Output Section Name
3753 @cindex name, section
3754 @cindex section name
3755 The name of the output section is @var{section}. @var{section} must
3756 meet the constraints of your output format. In formats which only
3757 support a limited number of sections, such as @code{a.out}, the name
3758 must be one of the names supported by the format (@code{a.out}, for
3759 example, allows only @samp{.text}, @samp{.data} or @samp{.bss}). If the
3760 output format supports any number of sections, but with numbers and not
3761 names (as is the case for Oasys), the name should be supplied as a
3762 quoted numeric string. A section name may consist of any sequence of
3763 characters, but a name which contains any unusual characters such as
3764 commas must be quoted.
3766 The output section name @samp{/DISCARD/} is special; @ref{Output Section
3769 @node Output Section Address
3770 @subsection Output Section Address
3771 @cindex address, section
3772 @cindex section address
3773 The @var{address} is an expression for the VMA (the virtual memory
3774 address) of the output section. This address is optional, but if it
3775 is provided then the output address will be set exactly as specified.
3777 If the output address is not specified then one will be chosen for the
3778 section, based on the heuristic below. This address will be adjusted
3779 to fit the alignment requirement of the output section. The
3780 alignment requirement is the strictest alignment of any input section
3781 contained within the output section.
3783 The output section address heuristic is as follows:
3787 If an output memory @var{region} is set for the section then it
3788 is added to this region and its address will be the next free address
3792 If the MEMORY command has been used to create a list of memory
3793 regions then the first region which has attributes compatible with the
3794 section is selected to contain it. The section's output address will
3795 be the next free address in that region; @ref{MEMORY}.
3798 If no memory regions were specified, or none match the section then
3799 the output address will be based on the current value of the location
3807 .text . : @{ *(.text) @}
3814 .text : @{ *(.text) @}
3818 are subtly different. The first will set the address of the
3819 @samp{.text} output section to the current value of the location
3820 counter. The second will set it to the current value of the location
3821 counter aligned to the strictest alignment of any of the @samp{.text}
3824 The @var{address} may be an arbitrary expression; @ref{Expressions}.
3825 For example, if you want to align the section on a 0x10 byte boundary,
3826 so that the lowest four bits of the section address are zero, you could
3827 do something like this:
3829 .text ALIGN(0x10) : @{ *(.text) @}
3832 This works because @code{ALIGN} returns the current location counter
3833 aligned upward to the specified value.
3835 Specifying @var{address} for a section will change the value of the
3836 location counter, provided that the section is non-empty. (Empty
3837 sections are ignored).
3840 @subsection Input Section Description
3841 @cindex input sections
3842 @cindex mapping input sections to output sections
3843 The most common output section command is an input section description.
3845 The input section description is the most basic linker script operation.
3846 You use output sections to tell the linker how to lay out your program
3847 in memory. You use input section descriptions to tell the linker how to
3848 map the input files into your memory layout.
3851 * Input Section Basics:: Input section basics
3852 * Input Section Wildcards:: Input section wildcard patterns
3853 * Input Section Common:: Input section for common symbols
3854 * Input Section Keep:: Input section and garbage collection
3855 * Input Section Example:: Input section example
3858 @node Input Section Basics
3859 @subsubsection Input Section Basics
3860 @cindex input section basics
3861 An input section description consists of a file name optionally followed
3862 by a list of section names in parentheses.
3864 The file name and the section name may be wildcard patterns, which we
3865 describe further below (@pxref{Input Section Wildcards}).
3867 The most common input section description is to include all input
3868 sections with a particular name in the output section. For example, to
3869 include all input @samp{.text} sections, you would write:
3874 Here the @samp{*} is a wildcard which matches any file name. To exclude a list
3875 of files from matching the file name wildcard, EXCLUDE_FILE may be used to
3876 match all files except the ones specified in the EXCLUDE_FILE list. For
3879 *(EXCLUDE_FILE (*crtend.o *otherfile.o) .ctors)
3881 will cause all .ctors sections from all files except @file{crtend.o} and
3882 @file{otherfile.o} to be included.
3884 There are two ways to include more than one section:
3890 The difference between these is the order in which the @samp{.text} and
3891 @samp{.rdata} input sections will appear in the output section. In the
3892 first example, they will be intermingled, appearing in the same order as
3893 they are found in the linker input. In the second example, all
3894 @samp{.text} input sections will appear first, followed by all
3895 @samp{.rdata} input sections.
3897 You can specify a file name to include sections from a particular file.
3898 You would do this if one or more of your files contain special data that
3899 needs to be at a particular location in memory. For example:
3904 To refine the sections that are included based on the section flags
3905 of an input section, INPUT_SECTION_FLAGS may be used.
3907 Here is a simple example for using Section header flags for ELF sections:
3912 .text : @{ INPUT_SECTION_FLAGS (SHF_MERGE & SHF_STRINGS) *(.text) @}
3913 .text2 : @{ INPUT_SECTION_FLAGS (!SHF_WRITE) *(.text) @}
3918 In this example, the output section @samp{.text} will be comprised of any
3919 input section matching the name *(.text) whose section header flags
3920 @code{SHF_MERGE} and @code{SHF_STRINGS} are set. The output section
3921 @samp{.text2} will be comprised of any input section matching the name *(.text)
3922 whose section header flag @code{SHF_WRITE} is clear.
3924 You can also specify files within archives by writing a pattern
3925 matching the archive, a colon, then the pattern matching the file,
3926 with no whitespace around the colon.
3930 matches file within archive
3932 matches the whole archive
3934 matches file but not one in an archive
3937 Either one or both of @samp{archive} and @samp{file} can contain shell
3938 wildcards. On DOS based file systems, the linker will assume that a
3939 single letter followed by a colon is a drive specifier, so
3940 @samp{c:myfile.o} is a simple file specification, not @samp{myfile.o}
3941 within an archive called @samp{c}. @samp{archive:file} filespecs may
3942 also be used within an @code{EXCLUDE_FILE} list, but may not appear in
3943 other linker script contexts. For instance, you cannot extract a file
3944 from an archive by using @samp{archive:file} in an @code{INPUT}
3947 If you use a file name without a list of sections, then all sections in
3948 the input file will be included in the output section. This is not
3949 commonly done, but it may by useful on occasion. For example:
3954 When you use a file name which is not an @samp{archive:file} specifier
3955 and does not contain any wild card
3956 characters, the linker will first see if you also specified the file
3957 name on the linker command line or in an @code{INPUT} command. If you
3958 did not, the linker will attempt to open the file as an input file, as
3959 though it appeared on the command line. Note that this differs from an
3960 @code{INPUT} command, because the linker will not search for the file in
3961 the archive search path.
3963 @node Input Section Wildcards
3964 @subsubsection Input Section Wildcard Patterns
3965 @cindex input section wildcards
3966 @cindex wildcard file name patterns
3967 @cindex file name wildcard patterns
3968 @cindex section name wildcard patterns
3969 In an input section description, either the file name or the section
3970 name or both may be wildcard patterns.
3972 The file name of @samp{*} seen in many examples is a simple wildcard
3973 pattern for the file name.
3975 The wildcard patterns are like those used by the Unix shell.
3979 matches any number of characters
3981 matches any single character
3983 matches a single instance of any of the @var{chars}; the @samp{-}
3984 character may be used to specify a range of characters, as in
3985 @samp{[a-z]} to match any lower case letter
3987 quotes the following character
3990 When a file name is matched with a wildcard, the wildcard characters
3991 will not match a @samp{/} character (used to separate directory names on
3992 Unix). A pattern consisting of a single @samp{*} character is an
3993 exception; it will always match any file name, whether it contains a
3994 @samp{/} or not. In a section name, the wildcard characters will match
3995 a @samp{/} character.
3997 File name wildcard patterns only match files which are explicitly
3998 specified on the command line or in an @code{INPUT} command. The linker
3999 does not search directories to expand wildcards.
4001 If a file name matches more than one wildcard pattern, or if a file name
4002 appears explicitly and is also matched by a wildcard pattern, the linker
4003 will use the first match in the linker script. For example, this
4004 sequence of input section descriptions is probably in error, because the
4005 @file{data.o} rule will not be used:
4007 .data : @{ *(.data) @}
4008 .data1 : @{ data.o(.data) @}
4011 @cindex SORT_BY_NAME
4012 Normally, the linker will place files and sections matched by wildcards
4013 in the order in which they are seen during the link. You can change
4014 this by using the @code{SORT_BY_NAME} keyword, which appears before a wildcard
4015 pattern in parentheses (e.g., @code{SORT_BY_NAME(.text*)}). When the
4016 @code{SORT_BY_NAME} keyword is used, the linker will sort the files or sections
4017 into ascending order by name before placing them in the output file.
4019 @cindex SORT_BY_ALIGNMENT
4020 @code{SORT_BY_ALIGNMENT} is very similar to @code{SORT_BY_NAME}. The
4021 difference is @code{SORT_BY_ALIGNMENT} will sort sections into
4022 ascending order by alignment before placing them in the output file.
4024 @cindex SORT_BY_INIT_PRIORITY
4025 @code{SORT_BY_INIT_PRIORITY} is very similar to @code{SORT_BY_NAME}. The
4026 difference is @code{SORT_BY_INIT_PRIORITY} will sort sections into
4027 ascending order by numerical value of the GCC init_priority attribute
4028 encoded in the section name before placing them in the output file.
4031 @code{SORT} is an alias for @code{SORT_BY_NAME}.
4033 When there are nested section sorting commands in linker script, there
4034 can be at most 1 level of nesting for section sorting commands.
4038 @code{SORT_BY_NAME} (@code{SORT_BY_ALIGNMENT} (wildcard section pattern)).
4039 It will sort the input sections by name first, then by alignment if 2
4040 sections have the same name.
4042 @code{SORT_BY_ALIGNMENT} (@code{SORT_BY_NAME} (wildcard section pattern)).
4043 It will sort the input sections by alignment first, then by name if 2
4044 sections have the same alignment.
4046 @code{SORT_BY_NAME} (@code{SORT_BY_NAME} (wildcard section pattern)) is
4047 treated the same as @code{SORT_BY_NAME} (wildcard section pattern).
4049 @code{SORT_BY_ALIGNMENT} (@code{SORT_BY_ALIGNMENT} (wildcard section pattern))
4050 is treated the same as @code{SORT_BY_ALIGNMENT} (wildcard section pattern).
4052 All other nested section sorting commands are invalid.
4055 When both command line section sorting option and linker script
4056 section sorting command are used, section sorting command always
4057 takes precedence over the command line option.
4059 If the section sorting command in linker script isn't nested, the
4060 command line option will make the section sorting command to be
4061 treated as nested sorting command.
4065 @code{SORT_BY_NAME} (wildcard section pattern ) with
4066 @option{--sort-sections alignment} is equivalent to
4067 @code{SORT_BY_NAME} (@code{SORT_BY_ALIGNMENT} (wildcard section pattern)).
4069 @code{SORT_BY_ALIGNMENT} (wildcard section pattern) with
4070 @option{--sort-section name} is equivalent to
4071 @code{SORT_BY_ALIGNMENT} (@code{SORT_BY_NAME} (wildcard section pattern)).
4074 If the section sorting command in linker script is nested, the
4075 command line option will be ignored.
4078 @code{SORT_NONE} disables section sorting by ignoring the command line
4079 section sorting option.
4081 If you ever get confused about where input sections are going, use the
4082 @samp{-M} linker option to generate a map file. The map file shows
4083 precisely how input sections are mapped to output sections.
4085 This example shows how wildcard patterns might be used to partition
4086 files. This linker script directs the linker to place all @samp{.text}
4087 sections in @samp{.text} and all @samp{.bss} sections in @samp{.bss}.
4088 The linker will place the @samp{.data} section from all files beginning
4089 with an upper case character in @samp{.DATA}; for all other files, the
4090 linker will place the @samp{.data} section in @samp{.data}.
4094 .text : @{ *(.text) @}
4095 .DATA : @{ [A-Z]*(.data) @}
4096 .data : @{ *(.data) @}
4097 .bss : @{ *(.bss) @}
4102 @node Input Section Common
4103 @subsubsection Input Section for Common Symbols
4104 @cindex common symbol placement
4105 @cindex uninitialized data placement
4106 A special notation is needed for common symbols, because in many object
4107 file formats common symbols do not have a particular input section. The
4108 linker treats common symbols as though they are in an input section
4109 named @samp{COMMON}.
4111 You may use file names with the @samp{COMMON} section just as with any
4112 other input sections. You can use this to place common symbols from a
4113 particular input file in one section while common symbols from other
4114 input files are placed in another section.
4116 In most cases, common symbols in input files will be placed in the
4117 @samp{.bss} section in the output file. For example:
4119 .bss @{ *(.bss) *(COMMON) @}
4122 @cindex scommon section
4123 @cindex small common symbols
4124 Some object file formats have more than one type of common symbol. For
4125 example, the MIPS ELF object file format distinguishes standard common
4126 symbols and small common symbols. In this case, the linker will use a
4127 different special section name for other types of common symbols. In
4128 the case of MIPS ELF, the linker uses @samp{COMMON} for standard common
4129 symbols and @samp{.scommon} for small common symbols. This permits you
4130 to map the different types of common symbols into memory at different
4134 You will sometimes see @samp{[COMMON]} in old linker scripts. This
4135 notation is now considered obsolete. It is equivalent to
4138 @node Input Section Keep
4139 @subsubsection Input Section and Garbage Collection
4141 @cindex garbage collection
4142 When link-time garbage collection is in use (@samp{--gc-sections}),
4143 it is often useful to mark sections that should not be eliminated.
4144 This is accomplished by surrounding an input section's wildcard entry
4145 with @code{KEEP()}, as in @code{KEEP(*(.init))} or
4146 @code{KEEP(SORT_BY_NAME(*)(.ctors))}.
4148 @node Input Section Example
4149 @subsubsection Input Section Example
4150 The following example is a complete linker script. It tells the linker
4151 to read all of the sections from file @file{all.o} and place them at the
4152 start of output section @samp{outputa} which starts at location
4153 @samp{0x10000}. All of section @samp{.input1} from file @file{foo.o}
4154 follows immediately, in the same output section. All of section
4155 @samp{.input2} from @file{foo.o} goes into output section
4156 @samp{outputb}, followed by section @samp{.input1} from @file{foo1.o}.
4157 All of the remaining @samp{.input1} and @samp{.input2} sections from any
4158 files are written to output section @samp{outputc}.
4186 @node Output Section Data
4187 @subsection Output Section Data
4189 @cindex section data
4190 @cindex output section data
4191 @kindex BYTE(@var{expression})
4192 @kindex SHORT(@var{expression})
4193 @kindex LONG(@var{expression})
4194 @kindex QUAD(@var{expression})
4195 @kindex SQUAD(@var{expression})
4196 You can include explicit bytes of data in an output section by using
4197 @code{BYTE}, @code{SHORT}, @code{LONG}, @code{QUAD}, or @code{SQUAD} as
4198 an output section command. Each keyword is followed by an expression in
4199 parentheses providing the value to store (@pxref{Expressions}). The
4200 value of the expression is stored at the current value of the location
4203 The @code{BYTE}, @code{SHORT}, @code{LONG}, and @code{QUAD} commands
4204 store one, two, four, and eight bytes (respectively). After storing the
4205 bytes, the location counter is incremented by the number of bytes
4208 For example, this will store the byte 1 followed by the four byte value
4209 of the symbol @samp{addr}:
4215 When using a 64 bit host or target, @code{QUAD} and @code{SQUAD} are the
4216 same; they both store an 8 byte, or 64 bit, value. When both host and
4217 target are 32 bits, an expression is computed as 32 bits. In this case
4218 @code{QUAD} stores a 32 bit value zero extended to 64 bits, and
4219 @code{SQUAD} stores a 32 bit value sign extended to 64 bits.
4221 If the object file format of the output file has an explicit endianness,
4222 which is the normal case, the value will be stored in that endianness.
4223 When the object file format does not have an explicit endianness, as is
4224 true of, for example, S-records, the value will be stored in the
4225 endianness of the first input object file.
4227 Note---these commands only work inside a section description and not
4228 between them, so the following will produce an error from the linker:
4230 SECTIONS @{@ .text : @{@ *(.text) @}@ LONG(1) .data : @{@ *(.data) @}@ @}@
4232 whereas this will work:
4234 SECTIONS @{@ .text : @{@ *(.text) ; LONG(1) @}@ .data : @{@ *(.data) @}@ @}@
4237 @kindex FILL(@var{expression})
4238 @cindex holes, filling
4239 @cindex unspecified memory
4240 You may use the @code{FILL} command to set the fill pattern for the
4241 current section. It is followed by an expression in parentheses. Any
4242 otherwise unspecified regions of memory within the section (for example,
4243 gaps left due to the required alignment of input sections) are filled
4244 with the value of the expression, repeated as
4245 necessary. A @code{FILL} statement covers memory locations after the
4246 point at which it occurs in the section definition; by including more
4247 than one @code{FILL} statement, you can have different fill patterns in
4248 different parts of an output section.
4250 This example shows how to fill unspecified regions of memory with the
4256 The @code{FILL} command is similar to the @samp{=@var{fillexp}} output
4257 section attribute, but it only affects the
4258 part of the section following the @code{FILL} command, rather than the
4259 entire section. If both are used, the @code{FILL} command takes
4260 precedence. @xref{Output Section Fill}, for details on the fill
4263 @node Output Section Keywords
4264 @subsection Output Section Keywords
4265 There are a couple of keywords which can appear as output section
4269 @kindex CREATE_OBJECT_SYMBOLS
4270 @cindex input filename symbols
4271 @cindex filename symbols
4272 @item CREATE_OBJECT_SYMBOLS
4273 The command tells the linker to create a symbol for each input file.
4274 The name of each symbol will be the name of the corresponding input
4275 file. The section of each symbol will be the output section in which
4276 the @code{CREATE_OBJECT_SYMBOLS} command appears.
4278 This is conventional for the a.out object file format. It is not
4279 normally used for any other object file format.
4281 @kindex CONSTRUCTORS
4282 @cindex C++ constructors, arranging in link
4283 @cindex constructors, arranging in link
4285 When linking using the a.out object file format, the linker uses an
4286 unusual set construct to support C++ global constructors and
4287 destructors. When linking object file formats which do not support
4288 arbitrary sections, such as ECOFF and XCOFF, the linker will
4289 automatically recognize C++ global constructors and destructors by name.
4290 For these object file formats, the @code{CONSTRUCTORS} command tells the
4291 linker to place constructor information in the output section where the
4292 @code{CONSTRUCTORS} command appears. The @code{CONSTRUCTORS} command is
4293 ignored for other object file formats.
4295 The symbol @w{@code{__CTOR_LIST__}} marks the start of the global
4296 constructors, and the symbol @w{@code{__CTOR_END__}} marks the end.
4297 Similarly, @w{@code{__DTOR_LIST__}} and @w{@code{__DTOR_END__}} mark
4298 the start and end of the global destructors. The
4299 first word in the list is the number of entries, followed by the address
4300 of each constructor or destructor, followed by a zero word. The
4301 compiler must arrange to actually run the code. For these object file
4302 formats @sc{gnu} C++ normally calls constructors from a subroutine
4303 @code{__main}; a call to @code{__main} is automatically inserted into
4304 the startup code for @code{main}. @sc{gnu} C++ normally runs
4305 destructors either by using @code{atexit}, or directly from the function
4308 For object file formats such as @code{COFF} or @code{ELF} which support
4309 arbitrary section names, @sc{gnu} C++ will normally arrange to put the
4310 addresses of global constructors and destructors into the @code{.ctors}
4311 and @code{.dtors} sections. Placing the following sequence into your
4312 linker script will build the sort of table which the @sc{gnu} C++
4313 runtime code expects to see.
4317 LONG((__CTOR_END__ - __CTOR_LIST__) / 4 - 2)
4322 LONG((__DTOR_END__ - __DTOR_LIST__) / 4 - 2)
4328 If you are using the @sc{gnu} C++ support for initialization priority,
4329 which provides some control over the order in which global constructors
4330 are run, you must sort the constructors at link time to ensure that they
4331 are executed in the correct order. When using the @code{CONSTRUCTORS}
4332 command, use @samp{SORT_BY_NAME(CONSTRUCTORS)} instead. When using the
4333 @code{.ctors} and @code{.dtors} sections, use @samp{*(SORT_BY_NAME(.ctors))} and
4334 @samp{*(SORT_BY_NAME(.dtors))} instead of just @samp{*(.ctors)} and
4337 Normally the compiler and linker will handle these issues automatically,
4338 and you will not need to concern yourself with them. However, you may
4339 need to consider this if you are using C++ and writing your own linker
4344 @node Output Section Discarding
4345 @subsection Output Section Discarding
4346 @cindex discarding sections
4347 @cindex sections, discarding
4348 @cindex removing sections
4349 The linker will not create output sections with no contents. This is
4350 for convenience when referring to input sections that may or may not
4351 be present in any of the input files. For example:
4353 .foo : @{ *(.foo) @}
4356 will only create a @samp{.foo} section in the output file if there is a
4357 @samp{.foo} section in at least one input file, and if the input
4358 sections are not all empty. Other link script directives that allocate
4359 space in an output section will also create the output section.
4361 The linker will ignore address assignments (@pxref{Output Section Address})
4362 on discarded output sections, except when the linker script defines
4363 symbols in the output section. In that case the linker will obey
4364 the address assignments, possibly advancing dot even though the
4365 section is discarded.
4368 The special output section name @samp{/DISCARD/} may be used to discard
4369 input sections. Any input sections which are assigned to an output
4370 section named @samp{/DISCARD/} are not included in the output file.
4372 @node Output Section Attributes
4373 @subsection Output Section Attributes
4374 @cindex output section attributes
4375 We showed above that the full description of an output section looked
4380 @var{section} [@var{address}] [(@var{type})] :
4382 [ALIGN(@var{section_align})]
4383 [SUBALIGN(@var{subsection_align})]
4386 @var{output-section-command}
4387 @var{output-section-command}
4389 @} [>@var{region}] [AT>@var{lma_region}] [:@var{phdr} :@var{phdr} @dots{}] [=@var{fillexp}]
4393 We've already described @var{section}, @var{address}, and
4394 @var{output-section-command}. In this section we will describe the
4395 remaining section attributes.
4398 * Output Section Type:: Output section type
4399 * Output Section LMA:: Output section LMA
4400 * Forced Output Alignment:: Forced Output Alignment
4401 * Forced Input Alignment:: Forced Input Alignment
4402 * Output Section Constraint:: Output section constraint
4403 * Output Section Region:: Output section region
4404 * Output Section Phdr:: Output section phdr
4405 * Output Section Fill:: Output section fill
4408 @node Output Section Type
4409 @subsubsection Output Section Type
4410 Each output section may have a type. The type is a keyword in
4411 parentheses. The following types are defined:
4415 The section should be marked as not loadable, so that it will not be
4416 loaded into memory when the program is run.
4421 These type names are supported for backward compatibility, and are
4422 rarely used. They all have the same effect: the section should be
4423 marked as not allocatable, so that no memory is allocated for the
4424 section when the program is run.
4428 @cindex prevent unnecessary loading
4429 @cindex loading, preventing
4430 The linker normally sets the attributes of an output section based on
4431 the input sections which map into it. You can override this by using
4432 the section type. For example, in the script sample below, the
4433 @samp{ROM} section is addressed at memory location @samp{0} and does not
4434 need to be loaded when the program is run.
4438 ROM 0 (NOLOAD) : @{ @dots{} @}
4444 @node Output Section LMA
4445 @subsubsection Output Section LMA
4446 @kindex AT>@var{lma_region}
4447 @kindex AT(@var{lma})
4448 @cindex load address
4449 @cindex section load address
4450 Every section has a virtual address (VMA) and a load address (LMA); see
4451 @ref{Basic Script Concepts}. The virtual address is specified by the
4452 @pxref{Output Section Address} described earlier. The load address is
4453 specified by the @code{AT} or @code{AT>} keywords. Specifying a load
4454 address is optional.
4456 The @code{AT} keyword takes an expression as an argument. This
4457 specifies the exact load address of the section. The @code{AT>} keyword
4458 takes the name of a memory region as an argument. @xref{MEMORY}. The
4459 load address of the section is set to the next free address in the
4460 region, aligned to the section's alignment requirements.
4462 If neither @code{AT} nor @code{AT>} is specified for an allocatable
4463 section, the linker will use the following heuristic to determine the
4468 If the section has a specific VMA address, then this is used as
4469 the LMA address as well.
4472 If the section is not allocatable then its LMA is set to its VMA.
4475 Otherwise if a memory region can be found that is compatible
4476 with the current section, and this region contains at least one
4477 section, then the LMA is set so the difference between the
4478 VMA and LMA is the same as the difference between the VMA and LMA of
4479 the last section in the located region.
4482 If no memory regions have been declared then a default region
4483 that covers the entire address space is used in the previous step.
4486 If no suitable region could be found, or there was no previous
4487 section then the LMA is set equal to the VMA.
4490 @cindex ROM initialized data
4491 @cindex initialized data in ROM
4492 This feature is designed to make it easy to build a ROM image. For
4493 example, the following linker script creates three output sections: one
4494 called @samp{.text}, which starts at @code{0x1000}, one called
4495 @samp{.mdata}, which is loaded at the end of the @samp{.text} section
4496 even though its VMA is @code{0x2000}, and one called @samp{.bss} to hold
4497 uninitialized data at address @code{0x3000}. The symbol @code{_data} is
4498 defined with the value @code{0x2000}, which shows that the location
4499 counter holds the VMA value, not the LMA value.
4505 .text 0x1000 : @{ *(.text) _etext = . ; @}
4507 AT ( ADDR (.text) + SIZEOF (.text) )
4508 @{ _data = . ; *(.data); _edata = . ; @}
4510 @{ _bstart = . ; *(.bss) *(COMMON) ; _bend = . ;@}
4515 The run-time initialization code for use with a program generated with
4516 this linker script would include something like the following, to copy
4517 the initialized data from the ROM image to its runtime address. Notice
4518 how this code takes advantage of the symbols defined by the linker
4523 extern char _etext, _data, _edata, _bstart, _bend;
4524 char *src = &_etext;
4527 /* ROM has data at end of text; copy it. */
4528 while (dst < &_edata)
4532 for (dst = &_bstart; dst< &_bend; dst++)
4537 @node Forced Output Alignment
4538 @subsubsection Forced Output Alignment
4539 @kindex ALIGN(@var{section_align})
4540 @cindex forcing output section alignment
4541 @cindex output section alignment
4542 You can increase an output section's alignment by using ALIGN.
4544 @node Forced Input Alignment
4545 @subsubsection Forced Input Alignment
4546 @kindex SUBALIGN(@var{subsection_align})
4547 @cindex forcing input section alignment
4548 @cindex input section alignment
4549 You can force input section alignment within an output section by using
4550 SUBALIGN. The value specified overrides any alignment given by input
4551 sections, whether larger or smaller.
4553 @node Output Section Constraint
4554 @subsubsection Output Section Constraint
4557 @cindex constraints on output sections
4558 You can specify that an output section should only be created if all
4559 of its input sections are read-only or all of its input sections are
4560 read-write by using the keyword @code{ONLY_IF_RO} and
4561 @code{ONLY_IF_RW} respectively.
4563 @node Output Section Region
4564 @subsubsection Output Section Region
4565 @kindex >@var{region}
4566 @cindex section, assigning to memory region
4567 @cindex memory regions and sections
4568 You can assign a section to a previously defined region of memory by
4569 using @samp{>@var{region}}. @xref{MEMORY}.
4571 Here is a simple example:
4574 MEMORY @{ rom : ORIGIN = 0x1000, LENGTH = 0x1000 @}
4575 SECTIONS @{ ROM : @{ *(.text) @} >rom @}
4579 @node Output Section Phdr
4580 @subsubsection Output Section Phdr
4582 @cindex section, assigning to program header
4583 @cindex program headers and sections
4584 You can assign a section to a previously defined program segment by
4585 using @samp{:@var{phdr}}. @xref{PHDRS}. If a section is assigned to
4586 one or more segments, then all subsequent allocated sections will be
4587 assigned to those segments as well, unless they use an explicitly
4588 @code{:@var{phdr}} modifier. You can use @code{:NONE} to tell the
4589 linker to not put the section in any segment at all.
4591 Here is a simple example:
4594 PHDRS @{ text PT_LOAD ; @}
4595 SECTIONS @{ .text : @{ *(.text) @} :text @}
4599 @node Output Section Fill
4600 @subsubsection Output Section Fill
4601 @kindex =@var{fillexp}
4602 @cindex section fill pattern
4603 @cindex fill pattern, entire section
4604 You can set the fill pattern for an entire section by using
4605 @samp{=@var{fillexp}}. @var{fillexp} is an expression
4606 (@pxref{Expressions}). Any otherwise unspecified regions of memory
4607 within the output section (for example, gaps left due to the required
4608 alignment of input sections) will be filled with the value, repeated as
4609 necessary. If the fill expression is a simple hex number, ie. a string
4610 of hex digit starting with @samp{0x} and without a trailing @samp{k} or @samp{M}, then
4611 an arbitrarily long sequence of hex digits can be used to specify the
4612 fill pattern; Leading zeros become part of the pattern too. For all
4613 other cases, including extra parentheses or a unary @code{+}, the fill
4614 pattern is the four least significant bytes of the value of the
4615 expression. In all cases, the number is big-endian.
4617 You can also change the fill value with a @code{FILL} command in the
4618 output section commands; (@pxref{Output Section Data}).
4620 Here is a simple example:
4623 SECTIONS @{ .text : @{ *(.text) @} =0x90909090 @}
4627 @node Overlay Description
4628 @subsection Overlay Description
4631 An overlay description provides an easy way to describe sections which
4632 are to be loaded as part of a single memory image but are to be run at
4633 the same memory address. At run time, some sort of overlay manager will
4634 copy the overlaid sections in and out of the runtime memory address as
4635 required, perhaps by simply manipulating addressing bits. This approach
4636 can be useful, for example, when a certain region of memory is faster
4639 Overlays are described using the @code{OVERLAY} command. The
4640 @code{OVERLAY} command is used within a @code{SECTIONS} command, like an
4641 output section description. The full syntax of the @code{OVERLAY}
4642 command is as follows:
4645 OVERLAY [@var{start}] : [NOCROSSREFS] [AT ( @var{ldaddr} )]
4649 @var{output-section-command}
4650 @var{output-section-command}
4652 @} [:@var{phdr}@dots{}] [=@var{fill}]
4655 @var{output-section-command}
4656 @var{output-section-command}
4658 @} [:@var{phdr}@dots{}] [=@var{fill}]
4660 @} [>@var{region}] [:@var{phdr}@dots{}] [=@var{fill}]
4664 Everything is optional except @code{OVERLAY} (a keyword), and each
4665 section must have a name (@var{secname1} and @var{secname2} above). The
4666 section definitions within the @code{OVERLAY} construct are identical to
4667 those within the general @code{SECTIONS} contruct (@pxref{SECTIONS}),
4668 except that no addresses and no memory regions may be defined for
4669 sections within an @code{OVERLAY}.
4671 The sections are all defined with the same starting address. The load
4672 addresses of the sections are arranged such that they are consecutive in
4673 memory starting at the load address used for the @code{OVERLAY} as a
4674 whole (as with normal section definitions, the load address is optional,
4675 and defaults to the start address; the start address is also optional,
4676 and defaults to the current value of the location counter).
4678 If the @code{NOCROSSREFS} keyword is used, and there any references
4679 among the sections, the linker will report an error. Since the sections
4680 all run at the same address, it normally does not make sense for one
4681 section to refer directly to another. @xref{Miscellaneous Commands,
4684 For each section within the @code{OVERLAY}, the linker automatically
4685 provides two symbols. The symbol @code{__load_start_@var{secname}} is
4686 defined as the starting load address of the section. The symbol
4687 @code{__load_stop_@var{secname}} is defined as the final load address of
4688 the section. Any characters within @var{secname} which are not legal
4689 within C identifiers are removed. C (or assembler) code may use these
4690 symbols to move the overlaid sections around as necessary.
4692 At the end of the overlay, the value of the location counter is set to
4693 the start address of the overlay plus the size of the largest section.
4695 Here is an example. Remember that this would appear inside a
4696 @code{SECTIONS} construct.
4699 OVERLAY 0x1000 : AT (0x4000)
4701 .text0 @{ o1/*.o(.text) @}
4702 .text1 @{ o2/*.o(.text) @}
4707 This will define both @samp{.text0} and @samp{.text1} to start at
4708 address 0x1000. @samp{.text0} will be loaded at address 0x4000, and
4709 @samp{.text1} will be loaded immediately after @samp{.text0}. The
4710 following symbols will be defined if referenced: @code{__load_start_text0},
4711 @code{__load_stop_text0}, @code{__load_start_text1},
4712 @code{__load_stop_text1}.
4714 C code to copy overlay @code{.text1} into the overlay area might look
4719 extern char __load_start_text1, __load_stop_text1;
4720 memcpy ((char *) 0x1000, &__load_start_text1,
4721 &__load_stop_text1 - &__load_start_text1);
4725 Note that the @code{OVERLAY} command is just syntactic sugar, since
4726 everything it does can be done using the more basic commands. The above
4727 example could have been written identically as follows.
4731 .text0 0x1000 : AT (0x4000) @{ o1/*.o(.text) @}
4732 PROVIDE (__load_start_text0 = LOADADDR (.text0));
4733 PROVIDE (__load_stop_text0 = LOADADDR (.text0) + SIZEOF (.text0));
4734 .text1 0x1000 : AT (0x4000 + SIZEOF (.text0)) @{ o2/*.o(.text) @}
4735 PROVIDE (__load_start_text1 = LOADADDR (.text1));
4736 PROVIDE (__load_stop_text1 = LOADADDR (.text1) + SIZEOF (.text1));
4737 . = 0x1000 + MAX (SIZEOF (.text0), SIZEOF (.text1));
4742 @section MEMORY Command
4744 @cindex memory regions
4745 @cindex regions of memory
4746 @cindex allocating memory
4747 @cindex discontinuous memory
4748 The linker's default configuration permits allocation of all available
4749 memory. You can override this by using the @code{MEMORY} command.
4751 The @code{MEMORY} command describes the location and size of blocks of
4752 memory in the target. You can use it to describe which memory regions
4753 may be used by the linker, and which memory regions it must avoid. You
4754 can then assign sections to particular memory regions. The linker will
4755 set section addresses based on the memory regions, and will warn about
4756 regions that become too full. The linker will not shuffle sections
4757 around to fit into the available regions.
4759 A linker script may contain at most one use of the @code{MEMORY}
4760 command. However, you can define as many blocks of memory within it as
4761 you wish. The syntax is:
4766 @var{name} [(@var{attr})] : ORIGIN = @var{origin}, LENGTH = @var{len}
4772 The @var{name} is a name used in the linker script to refer to the
4773 region. The region name has no meaning outside of the linker script.
4774 Region names are stored in a separate name space, and will not conflict
4775 with symbol names, file names, or section names. Each memory region
4776 must have a distinct name within the @code{MEMORY} command. However you can
4777 add later alias names to existing memory regions with the @ref{REGION_ALIAS}
4780 @cindex memory region attributes
4781 The @var{attr} string is an optional list of attributes that specify
4782 whether to use a particular memory region for an input section which is
4783 not explicitly mapped in the linker script. As described in
4784 @ref{SECTIONS}, if you do not specify an output section for some input
4785 section, the linker will create an output section with the same name as
4786 the input section. If you define region attributes, the linker will use
4787 them to select the memory region for the output section that it creates.
4789 The @var{attr} string must consist only of the following characters:
4804 Invert the sense of any of the attributes that follow
4807 If a unmapped section matches any of the listed attributes other than
4808 @samp{!}, it will be placed in the memory region. The @samp{!}
4809 attribute reverses this test, so that an unmapped section will be placed
4810 in the memory region only if it does not match any of the listed
4816 The @var{origin} is an numerical expression for the start address of
4817 the memory region. The expression must evaluate to a constant and it
4818 cannot involve any symbols. The keyword @code{ORIGIN} may be
4819 abbreviated to @code{org} or @code{o} (but not, for example,
4825 The @var{len} is an expression for the size in bytes of the memory
4826 region. As with the @var{origin} expression, the expression must
4827 be numerical only and must evaluate to a constant. The keyword
4828 @code{LENGTH} may be abbreviated to @code{len} or @code{l}.
4830 In the following example, we specify that there are two memory regions
4831 available for allocation: one starting at @samp{0} for 256 kilobytes,
4832 and the other starting at @samp{0x40000000} for four megabytes. The
4833 linker will place into the @samp{rom} memory region every section which
4834 is not explicitly mapped into a memory region, and is either read-only
4835 or executable. The linker will place other sections which are not
4836 explicitly mapped into a memory region into the @samp{ram} memory
4843 rom (rx) : ORIGIN = 0, LENGTH = 256K
4844 ram (!rx) : org = 0x40000000, l = 4M
4849 Once you define a memory region, you can direct the linker to place
4850 specific output sections into that memory region by using the
4851 @samp{>@var{region}} output section attribute. For example, if you have
4852 a memory region named @samp{mem}, you would use @samp{>mem} in the
4853 output section definition. @xref{Output Section Region}. If no address
4854 was specified for the output section, the linker will set the address to
4855 the next available address within the memory region. If the combined
4856 output sections directed to a memory region are too large for the
4857 region, the linker will issue an error message.
4859 It is possible to access the origin and length of a memory in an
4860 expression via the @code{ORIGIN(@var{memory})} and
4861 @code{LENGTH(@var{memory})} functions:
4865 _fstack = ORIGIN(ram) + LENGTH(ram) - 4;
4870 @section PHDRS Command
4872 @cindex program headers
4873 @cindex ELF program headers
4874 @cindex program segments
4875 @cindex segments, ELF
4876 The ELF object file format uses @dfn{program headers}, also knows as
4877 @dfn{segments}. The program headers describe how the program should be
4878 loaded into memory. You can print them out by using the @code{objdump}
4879 program with the @samp{-p} option.
4881 When you run an ELF program on a native ELF system, the system loader
4882 reads the program headers in order to figure out how to load the
4883 program. This will only work if the program headers are set correctly.
4884 This manual does not describe the details of how the system loader
4885 interprets program headers; for more information, see the ELF ABI.
4887 The linker will create reasonable program headers by default. However,
4888 in some cases, you may need to specify the program headers more
4889 precisely. You may use the @code{PHDRS} command for this purpose. When
4890 the linker sees the @code{PHDRS} command in the linker script, it will
4891 not create any program headers other than the ones specified.
4893 The linker only pays attention to the @code{PHDRS} command when
4894 generating an ELF output file. In other cases, the linker will simply
4895 ignore @code{PHDRS}.
4897 This is the syntax of the @code{PHDRS} command. The words @code{PHDRS},
4898 @code{FILEHDR}, @code{AT}, and @code{FLAGS} are keywords.
4904 @var{name} @var{type} [ FILEHDR ] [ PHDRS ] [ AT ( @var{address} ) ]
4905 [ FLAGS ( @var{flags} ) ] ;
4910 The @var{name} is used only for reference in the @code{SECTIONS} command
4911 of the linker script. It is not put into the output file. Program
4912 header names are stored in a separate name space, and will not conflict
4913 with symbol names, file names, or section names. Each program header
4914 must have a distinct name. The headers are processed in order and it
4915 is usual for them to map to sections in ascending load address order.
4917 Certain program header types describe segments of memory which the
4918 system loader will load from the file. In the linker script, you
4919 specify the contents of these segments by placing allocatable output
4920 sections in the segments. You use the @samp{:@var{phdr}} output section
4921 attribute to place a section in a particular segment. @xref{Output
4924 It is normal to put certain sections in more than one segment. This
4925 merely implies that one segment of memory contains another. You may
4926 repeat @samp{:@var{phdr}}, using it once for each segment which should
4927 contain the section.
4929 If you place a section in one or more segments using @samp{:@var{phdr}},
4930 then the linker will place all subsequent allocatable sections which do
4931 not specify @samp{:@var{phdr}} in the same segments. This is for
4932 convenience, since generally a whole set of contiguous sections will be
4933 placed in a single segment. You can use @code{:NONE} to override the
4934 default segment and tell the linker to not put the section in any
4939 You may use the @code{FILEHDR} and @code{PHDRS} keywords after
4940 the program header type to further describe the contents of the segment.
4941 The @code{FILEHDR} keyword means that the segment should include the ELF
4942 file header. The @code{PHDRS} keyword means that the segment should
4943 include the ELF program headers themselves. If applied to a loadable
4944 segment (@code{PT_LOAD}), all prior loadable segments must have one of
4947 The @var{type} may be one of the following. The numbers indicate the
4948 value of the keyword.
4951 @item @code{PT_NULL} (0)
4952 Indicates an unused program header.
4954 @item @code{PT_LOAD} (1)
4955 Indicates that this program header describes a segment to be loaded from
4958 @item @code{PT_DYNAMIC} (2)
4959 Indicates a segment where dynamic linking information can be found.
4961 @item @code{PT_INTERP} (3)
4962 Indicates a segment where the name of the program interpreter may be
4965 @item @code{PT_NOTE} (4)
4966 Indicates a segment holding note information.
4968 @item @code{PT_SHLIB} (5)
4969 A reserved program header type, defined but not specified by the ELF
4972 @item @code{PT_PHDR} (6)
4973 Indicates a segment where the program headers may be found.
4975 @item @var{expression}
4976 An expression giving the numeric type of the program header. This may
4977 be used for types not defined above.
4980 You can specify that a segment should be loaded at a particular address
4981 in memory by using an @code{AT} expression. This is identical to the
4982 @code{AT} command used as an output section attribute (@pxref{Output
4983 Section LMA}). The @code{AT} command for a program header overrides the
4984 output section attribute.
4986 The linker will normally set the segment flags based on the sections
4987 which comprise the segment. You may use the @code{FLAGS} keyword to
4988 explicitly specify the segment flags. The value of @var{flags} must be
4989 an integer. It is used to set the @code{p_flags} field of the program
4992 Here is an example of @code{PHDRS}. This shows a typical set of program
4993 headers used on a native ELF system.
4999 headers PT_PHDR PHDRS ;
5001 text PT_LOAD FILEHDR PHDRS ;
5003 dynamic PT_DYNAMIC ;
5009 .interp : @{ *(.interp) @} :text :interp
5010 .text : @{ *(.text) @} :text
5011 .rodata : @{ *(.rodata) @} /* defaults to :text */
5013 . = . + 0x1000; /* move to a new page in memory */
5014 .data : @{ *(.data) @} :data
5015 .dynamic : @{ *(.dynamic) @} :data :dynamic
5022 @section VERSION Command
5023 @kindex VERSION @{script text@}
5024 @cindex symbol versions
5025 @cindex version script
5026 @cindex versions of symbols
5027 The linker supports symbol versions when using ELF. Symbol versions are
5028 only useful when using shared libraries. The dynamic linker can use
5029 symbol versions to select a specific version of a function when it runs
5030 a program that may have been linked against an earlier version of the
5033 You can include a version script directly in the main linker script, or
5034 you can supply the version script as an implicit linker script. You can
5035 also use the @samp{--version-script} linker option.
5037 The syntax of the @code{VERSION} command is simply
5039 VERSION @{ version-script-commands @}
5042 The format of the version script commands is identical to that used by
5043 Sun's linker in Solaris 2.5. The version script defines a tree of
5044 version nodes. You specify the node names and interdependencies in the
5045 version script. You can specify which symbols are bound to which
5046 version nodes, and you can reduce a specified set of symbols to local
5047 scope so that they are not globally visible outside of the shared
5050 The easiest way to demonstrate the version script language is with a few
5076 This example version script defines three version nodes. The first
5077 version node defined is @samp{VERS_1.1}; it has no other dependencies.
5078 The script binds the symbol @samp{foo1} to @samp{VERS_1.1}. It reduces
5079 a number of symbols to local scope so that they are not visible outside
5080 of the shared library; this is done using wildcard patterns, so that any
5081 symbol whose name begins with @samp{old}, @samp{original}, or @samp{new}
5082 is matched. The wildcard patterns available are the same as those used
5083 in the shell when matching filenames (also known as ``globbing'').
5084 However, if you specify the symbol name inside double quotes, then the
5085 name is treated as literal, rather than as a glob pattern.
5087 Next, the version script defines node @samp{VERS_1.2}. This node
5088 depends upon @samp{VERS_1.1}. The script binds the symbol @samp{foo2}
5089 to the version node @samp{VERS_1.2}.
5091 Finally, the version script defines node @samp{VERS_2.0}. This node
5092 depends upon @samp{VERS_1.2}. The scripts binds the symbols @samp{bar1}
5093 and @samp{bar2} are bound to the version node @samp{VERS_2.0}.
5095 When the linker finds a symbol defined in a library which is not
5096 specifically bound to a version node, it will effectively bind it to an
5097 unspecified base version of the library. You can bind all otherwise
5098 unspecified symbols to a given version node by using @samp{global: *;}
5099 somewhere in the version script. Note that it's slightly crazy to use
5100 wildcards in a global spec except on the last version node. Global
5101 wildcards elsewhere run the risk of accidentally adding symbols to the
5102 set exported for an old version. That's wrong since older versions
5103 ought to have a fixed set of symbols.
5105 The names of the version nodes have no specific meaning other than what
5106 they might suggest to the person reading them. The @samp{2.0} version
5107 could just as well have appeared in between @samp{1.1} and @samp{1.2}.
5108 However, this would be a confusing way to write a version script.
5110 Node name can be omitted, provided it is the only version node
5111 in the version script. Such version script doesn't assign any versions to
5112 symbols, only selects which symbols will be globally visible out and which
5116 @{ global: foo; bar; local: *; @};
5119 When you link an application against a shared library that has versioned
5120 symbols, the application itself knows which version of each symbol it
5121 requires, and it also knows which version nodes it needs from each
5122 shared library it is linked against. Thus at runtime, the dynamic
5123 loader can make a quick check to make sure that the libraries you have
5124 linked against do in fact supply all of the version nodes that the
5125 application will need to resolve all of the dynamic symbols. In this
5126 way it is possible for the dynamic linker to know with certainty that
5127 all external symbols that it needs will be resolvable without having to
5128 search for each symbol reference.
5130 The symbol versioning is in effect a much more sophisticated way of
5131 doing minor version checking that SunOS does. The fundamental problem
5132 that is being addressed here is that typically references to external
5133 functions are bound on an as-needed basis, and are not all bound when
5134 the application starts up. If a shared library is out of date, a
5135 required interface may be missing; when the application tries to use
5136 that interface, it may suddenly and unexpectedly fail. With symbol
5137 versioning, the user will get a warning when they start their program if
5138 the libraries being used with the application are too old.
5140 There are several GNU extensions to Sun's versioning approach. The
5141 first of these is the ability to bind a symbol to a version node in the
5142 source file where the symbol is defined instead of in the versioning
5143 script. This was done mainly to reduce the burden on the library
5144 maintainer. You can do this by putting something like:
5146 __asm__(".symver original_foo,foo@@VERS_1.1");
5149 in the C source file. This renames the function @samp{original_foo} to
5150 be an alias for @samp{foo} bound to the version node @samp{VERS_1.1}.
5151 The @samp{local:} directive can be used to prevent the symbol
5152 @samp{original_foo} from being exported. A @samp{.symver} directive
5153 takes precedence over a version script.
5155 The second GNU extension is to allow multiple versions of the same
5156 function to appear in a given shared library. In this way you can make
5157 an incompatible change to an interface without increasing the major
5158 version number of the shared library, while still allowing applications
5159 linked against the old interface to continue to function.
5161 To do this, you must use multiple @samp{.symver} directives in the
5162 source file. Here is an example:
5165 __asm__(".symver original_foo,foo@@");
5166 __asm__(".symver old_foo,foo@@VERS_1.1");
5167 __asm__(".symver old_foo1,foo@@VERS_1.2");
5168 __asm__(".symver new_foo,foo@@@@VERS_2.0");
5171 In this example, @samp{foo@@} represents the symbol @samp{foo} bound to the
5172 unspecified base version of the symbol. The source file that contains this
5173 example would define 4 C functions: @samp{original_foo}, @samp{old_foo},
5174 @samp{old_foo1}, and @samp{new_foo}.
5176 When you have multiple definitions of a given symbol, there needs to be
5177 some way to specify a default version to which external references to
5178 this symbol will be bound. You can do this with the
5179 @samp{foo@@@@VERS_2.0} type of @samp{.symver} directive. You can only
5180 declare one version of a symbol as the default in this manner; otherwise
5181 you would effectively have multiple definitions of the same symbol.
5183 If you wish to bind a reference to a specific version of the symbol
5184 within the shared library, you can use the aliases of convenience
5185 (i.e., @samp{old_foo}), or you can use the @samp{.symver} directive to
5186 specifically bind to an external version of the function in question.
5188 You can also specify the language in the version script:
5191 VERSION extern "lang" @{ version-script-commands @}
5194 The supported @samp{lang}s are @samp{C}, @samp{C++}, and @samp{Java}.
5195 The linker will iterate over the list of symbols at the link time and
5196 demangle them according to @samp{lang} before matching them to the
5197 patterns specified in @samp{version-script-commands}. The default
5198 @samp{lang} is @samp{C}.
5200 Demangled names may contains spaces and other special characters. As
5201 described above, you can use a glob pattern to match demangled names,
5202 or you can use a double-quoted string to match the string exactly. In
5203 the latter case, be aware that minor differences (such as differing
5204 whitespace) between the version script and the demangler output will
5205 cause a mismatch. As the exact string generated by the demangler
5206 might change in the future, even if the mangled name does not, you
5207 should check that all of your version directives are behaving as you
5208 expect when you upgrade.
5211 @section Expressions in Linker Scripts
5214 The syntax for expressions in the linker script language is identical to
5215 that of C expressions. All expressions are evaluated as integers. All
5216 expressions are evaluated in the same size, which is 32 bits if both the
5217 host and target are 32 bits, and is otherwise 64 bits.
5219 You can use and set symbol values in expressions.
5221 The linker defines several special purpose builtin functions for use in
5225 * Constants:: Constants
5226 * Symbolic Constants:: Symbolic constants
5227 * Symbols:: Symbol Names
5228 * Orphan Sections:: Orphan Sections
5229 * Location Counter:: The Location Counter
5230 * Operators:: Operators
5231 * Evaluation:: Evaluation
5232 * Expression Section:: The Section of an Expression
5233 * Builtin Functions:: Builtin Functions
5237 @subsection Constants
5238 @cindex integer notation
5239 @cindex constants in linker scripts
5240 All constants are integers.
5242 As in C, the linker considers an integer beginning with @samp{0} to be
5243 octal, and an integer beginning with @samp{0x} or @samp{0X} to be
5244 hexadecimal. Alternatively the linker accepts suffixes of @samp{h} or
5245 @samp{H} for hexadeciaml, @samp{o} or @samp{O} for octal, @samp{b} or
5246 @samp{B} for binary and @samp{d} or @samp{D} for decimal. Any integer
5247 value without a prefix or a suffix is considered to be decimal.
5249 @cindex scaled integers
5250 @cindex K and M integer suffixes
5251 @cindex M and K integer suffixes
5252 @cindex suffixes for integers
5253 @cindex integer suffixes
5254 In addition, you can use the suffixes @code{K} and @code{M} to scale a
5258 @c END TEXI2ROFF-KILL
5259 @code{1024} or @code{1024*1024}
5263 ${\rm 1024}$ or ${\rm 1024}^2$
5265 @c END TEXI2ROFF-KILL
5266 respectively. For example, the following
5267 all refer to the same quantity:
5276 Note - the @code{K} and @code{M} suffixes cannot be used in
5277 conjunction with the base suffixes mentioned above.
5279 @node Symbolic Constants
5280 @subsection Symbolic Constants
5281 @cindex symbolic constants
5283 It is possible to refer to target specific constants via the use of
5284 the @code{CONSTANT(@var{name})} operator, where @var{name} is one of:
5289 The target's maximum page size.
5291 @item COMMONPAGESIZE
5292 @kindex COMMONPAGESIZE
5293 The target's default page size.
5299 .text ALIGN (CONSTANT (MAXPAGESIZE)) : @{ *(.text) @}
5302 will create a text section aligned to the largest page boundary
5303 supported by the target.
5306 @subsection Symbol Names
5307 @cindex symbol names
5309 @cindex quoted symbol names
5311 Unless quoted, symbol names start with a letter, underscore, or period
5312 and may include letters, digits, underscores, periods, and hyphens.
5313 Unquoted symbol names must not conflict with any keywords. You can
5314 specify a symbol which contains odd characters or has the same name as a
5315 keyword by surrounding the symbol name in double quotes:
5318 "with a space" = "also with a space" + 10;
5321 Since symbols can contain many non-alphabetic characters, it is safest
5322 to delimit symbols with spaces. For example, @samp{A-B} is one symbol,
5323 whereas @samp{A - B} is an expression involving subtraction.
5325 @node Orphan Sections
5326 @subsection Orphan Sections
5328 Orphan sections are sections present in the input files which
5329 are not explicitly placed into the output file by the linker
5330 script. The linker will still copy these sections into the
5331 output file, but it has to guess as to where they should be
5332 placed. The linker uses a simple heuristic to do this. It
5333 attempts to place orphan sections after non-orphan sections of the
5334 same attribute, such as code vs data, loadable vs non-loadable, etc.
5335 If there is not enough room to do this then it places
5336 at the end of the file.
5338 For ELF targets, the attribute of the section includes section type as
5339 well as section flag.
5341 If an orphaned section's name is representable as a C identifier then
5342 the linker will automatically @pxref{PROVIDE} two symbols:
5343 __start_SECNAME and __stop_SECNAME, where SECNAME is the name of the
5344 section. These indicate the start address and end address of the
5345 orphaned section respectively. Note: most section names are not
5346 representable as C identifiers because they contain a @samp{.}
5349 @node Location Counter
5350 @subsection The Location Counter
5353 @cindex location counter
5354 @cindex current output location
5355 The special linker variable @dfn{dot} @samp{.} always contains the
5356 current output location counter. Since the @code{.} always refers to a
5357 location in an output section, it may only appear in an expression
5358 within a @code{SECTIONS} command. The @code{.} symbol may appear
5359 anywhere that an ordinary symbol is allowed in an expression.
5362 Assigning a value to @code{.} will cause the location counter to be
5363 moved. This may be used to create holes in the output section. The
5364 location counter may not be moved backwards inside an output section,
5365 and may not be moved backwards outside of an output section if so
5366 doing creates areas with overlapping LMAs.
5382 In the previous example, the @samp{.text} section from @file{file1} is
5383 located at the beginning of the output section @samp{output}. It is
5384 followed by a 1000 byte gap. Then the @samp{.text} section from
5385 @file{file2} appears, also with a 1000 byte gap following before the
5386 @samp{.text} section from @file{file3}. The notation @samp{= 0x12345678}
5387 specifies what data to write in the gaps (@pxref{Output Section Fill}).
5389 @cindex dot inside sections
5390 Note: @code{.} actually refers to the byte offset from the start of the
5391 current containing object. Normally this is the @code{SECTIONS}
5392 statement, whose start address is 0, hence @code{.} can be used as an
5393 absolute address. If @code{.} is used inside a section description
5394 however, it refers to the byte offset from the start of that section,
5395 not an absolute address. Thus in a script like this:
5413 The @samp{.text} section will be assigned a starting address of 0x100
5414 and a size of exactly 0x200 bytes, even if there is not enough data in
5415 the @samp{.text} input sections to fill this area. (If there is too
5416 much data, an error will be produced because this would be an attempt to
5417 move @code{.} backwards). The @samp{.data} section will start at 0x500
5418 and it will have an extra 0x600 bytes worth of space after the end of
5419 the values from the @samp{.data} input sections and before the end of
5420 the @samp{.data} output section itself.
5422 @cindex dot outside sections
5423 Setting symbols to the value of the location counter outside of an
5424 output section statement can result in unexpected values if the linker
5425 needs to place orphan sections. For example, given the following:
5431 .text: @{ *(.text) @}
5435 .data: @{ *(.data) @}
5440 If the linker needs to place some input section, e.g. @code{.rodata},
5441 not mentioned in the script, it might choose to place that section
5442 between @code{.text} and @code{.data}. You might think the linker
5443 should place @code{.rodata} on the blank line in the above script, but
5444 blank lines are of no particular significance to the linker. As well,
5445 the linker doesn't associate the above symbol names with their
5446 sections. Instead, it assumes that all assignments or other
5447 statements belong to the previous output section, except for the
5448 special case of an assignment to @code{.}. I.e., the linker will
5449 place the orphan @code{.rodata} section as if the script was written
5456 .text: @{ *(.text) @}
5460 .rodata: @{ *(.rodata) @}
5461 .data: @{ *(.data) @}
5466 This may or may not be the script author's intention for the value of
5467 @code{start_of_data}. One way to influence the orphan section
5468 placement is to assign the location counter to itself, as the linker
5469 assumes that an assignment to @code{.} is setting the start address of
5470 a following output section and thus should be grouped with that
5471 section. So you could write:
5477 .text: @{ *(.text) @}
5482 .data: @{ *(.data) @}
5487 Now, the orphan @code{.rodata} section will be placed between
5488 @code{end_of_text} and @code{start_of_data}.
5492 @subsection Operators
5493 @cindex operators for arithmetic
5494 @cindex arithmetic operators
5495 @cindex precedence in expressions
5496 The linker recognizes the standard C set of arithmetic operators, with
5497 the standard bindings and precedence levels:
5500 @c END TEXI2ROFF-KILL
5502 precedence associativity Operators Notes
5508 5 left == != > < <= >=
5514 11 right &= += -= *= /= (2)
5518 (1) Prefix operators
5519 (2) @xref{Assignments}.
5523 \vskip \baselineskip
5524 %"lispnarrowing" is the extra indent used generally for smallexample
5525 \hskip\lispnarrowing\vbox{\offinterlineskip
5528 {\vrule#&\strut\hfil\ #\ \hfil&\vrule#&\strut\hfil\ #\ \hfil&\vrule#&\strut\hfil\ {\tt #}\ \hfil&\vrule#\cr
5529 height2pt&\omit&&\omit&&\omit&\cr
5530 &Precedence&& Associativity &&{\rm Operators}&\cr
5531 height2pt&\omit&&\omit&&\omit&\cr
5533 height2pt&\omit&&\omit&&\omit&\cr
5535 % '176 is tilde, '~' in tt font
5536 &1&&left&&\qquad- \char'176\ !\qquad\dag&\cr
5537 &2&&left&&* / \%&\cr
5540 &5&&left&&== != > < <= >=&\cr
5543 &8&&left&&{\&\&}&\cr
5546 &11&&right&&\qquad\&= += -= *= /=\qquad\ddag&\cr
5548 height2pt&\omit&&\omit&&\omit&\cr}
5553 @obeylines@parskip=0pt@parindent=0pt
5554 @dag@quad Prefix operators.
5555 @ddag@quad @xref{Assignments}.
5558 @c END TEXI2ROFF-KILL
5561 @subsection Evaluation
5562 @cindex lazy evaluation
5563 @cindex expression evaluation order
5564 The linker evaluates expressions lazily. It only computes the value of
5565 an expression when absolutely necessary.
5567 The linker needs some information, such as the value of the start
5568 address of the first section, and the origins and lengths of memory
5569 regions, in order to do any linking at all. These values are computed
5570 as soon as possible when the linker reads in the linker script.
5572 However, other values (such as symbol values) are not known or needed
5573 until after storage allocation. Such values are evaluated later, when
5574 other information (such as the sizes of output sections) is available
5575 for use in the symbol assignment expression.
5577 The sizes of sections cannot be known until after allocation, so
5578 assignments dependent upon these are not performed until after
5581 Some expressions, such as those depending upon the location counter
5582 @samp{.}, must be evaluated during section allocation.
5584 If the result of an expression is required, but the value is not
5585 available, then an error results. For example, a script like the
5591 .text 9+this_isnt_constant :
5597 will cause the error message @samp{non constant expression for initial
5600 @node Expression Section
5601 @subsection The Section of an Expression
5602 @cindex expression sections
5603 @cindex absolute expressions
5604 @cindex relative expressions
5605 @cindex absolute and relocatable symbols
5606 @cindex relocatable and absolute symbols
5607 @cindex symbols, relocatable and absolute
5608 Addresses and symbols may be section relative, or absolute. A section
5609 relative symbol is relocatable. If you request relocatable output
5610 using the @samp{-r} option, a further link operation may change the
5611 value of a section relative symbol. On the other hand, an absolute
5612 symbol will retain the same value throughout any further link
5615 Some terms in linker expressions are addresses. This is true of
5616 section relative symbols and for builtin functions that return an
5617 address, such as @code{ADDR}, @code{LOADADDR}, @code{ORIGIN} and
5618 @code{SEGMENT_START}. Other terms are simply numbers, or are builtin
5619 functions that return a non-address value, such as @code{LENGTH}.
5620 One complication is that unless you set @code{LD_FEATURE ("SANE_EXPR")}
5621 (@pxref{Miscellaneous Commands}), numbers and absolute symbols are treated
5622 differently depending on their location, for compatibility with older
5623 versions of @code{ld}. Expressions appearing outside an output
5624 section definition treat all numbers as absolute addresses.
5625 Expressions appearing inside an output section definition treat
5626 absolute symbols as numbers. If @code{LD_FEATURE ("SANE_EXPR")} is
5627 given, then absolute symbols and numbers are simply treated as numbers
5630 In the following simple example,
5637 __executable_start = 0x100;
5641 __data_start = 0x10;
5649 both @code{.} and @code{__executable_start} are set to the absolute
5650 address 0x100 in the first two assignments, then both @code{.} and
5651 @code{__data_start} are set to 0x10 relative to the @code{.data}
5652 section in the second two assignments.
5654 For expressions involving numbers, relative addresses and absolute
5655 addresses, ld follows these rules to evaluate terms:
5659 Unary operations on an absolute address or number, and binary
5660 operations on two absolute addresses or two numbers, or between one
5661 absolute address and a number, apply the operator to the value(s).
5663 Unary operations on a relative address, and binary operations on two
5664 relative addresses in the same section or between one relative address
5665 and a number, apply the operator to the offset part of the address(es).
5667 Other binary operations, that is, between two relative addresses not
5668 in the same section, or between a relative address and an absolute
5669 address, first convert any non-absolute term to an absolute address
5670 before applying the operator.
5673 The result section of each sub-expression is as follows:
5677 An operation involving only numbers results in a number.
5679 The result of comparisons, @samp{&&} and @samp{||} is also a number.
5681 The result of other binary arithmetic and logical operations on two
5682 relative addresses in the same section or two absolute addresess
5683 (after above conversions) is also a number.
5685 The result of other operations on relative addresses or one
5686 relative address and a number, is a relative address in the same
5687 section as the relative operand(s).
5689 The result of other operations on absolute addresses (after above
5690 conversions) is an absolute address.
5693 You can use the builtin function @code{ABSOLUTE} to force an expression
5694 to be absolute when it would otherwise be relative. For example, to
5695 create an absolute symbol set to the address of the end of the output
5696 section @samp{.data}:
5700 .data : @{ *(.data) _edata = ABSOLUTE(.); @}
5704 If @samp{ABSOLUTE} were not used, @samp{_edata} would be relative to the
5705 @samp{.data} section.
5707 Using @code{LOADADDR} also forces an expression absolute, since this
5708 particular builtin function returns an absolute address.
5710 @node Builtin Functions
5711 @subsection Builtin Functions
5712 @cindex functions in expressions
5713 The linker script language includes a number of builtin functions for
5714 use in linker script expressions.
5717 @item ABSOLUTE(@var{exp})
5718 @kindex ABSOLUTE(@var{exp})
5719 @cindex expression, absolute
5720 Return the absolute (non-relocatable, as opposed to non-negative) value
5721 of the expression @var{exp}. Primarily useful to assign an absolute
5722 value to a symbol within a section definition, where symbol values are
5723 normally section relative. @xref{Expression Section}.
5725 @item ADDR(@var{section})
5726 @kindex ADDR(@var{section})
5727 @cindex section address in expression
5728 Return the address (VMA) of the named @var{section}. Your
5729 script must previously have defined the location of that section. In
5730 the following example, @code{start_of_output_1}, @code{symbol_1} and
5731 @code{symbol_2} are assigned equivalent values, except that
5732 @code{symbol_1} will be relative to the @code{.output1} section while
5733 the other two will be absolute:
5739 start_of_output_1 = ABSOLUTE(.);
5744 symbol_1 = ADDR(.output1);
5745 symbol_2 = start_of_output_1;
5751 @item ALIGN(@var{align})
5752 @itemx ALIGN(@var{exp},@var{align})
5753 @kindex ALIGN(@var{align})
5754 @kindex ALIGN(@var{exp},@var{align})
5755 @cindex round up location counter
5756 @cindex align location counter
5757 @cindex round up expression
5758 @cindex align expression
5759 Return the location counter (@code{.}) or arbitrary expression aligned
5760 to the next @var{align} boundary. The single operand @code{ALIGN}
5761 doesn't change the value of the location counter---it just does
5762 arithmetic on it. The two operand @code{ALIGN} allows an arbitrary
5763 expression to be aligned upwards (@code{ALIGN(@var{align})} is
5764 equivalent to @code{ALIGN(., @var{align})}).
5766 Here is an example which aligns the output @code{.data} section to the
5767 next @code{0x2000} byte boundary after the preceding section and sets a
5768 variable within the section to the next @code{0x8000} boundary after the
5773 .data ALIGN(0x2000): @{
5775 variable = ALIGN(0x8000);
5781 The first use of @code{ALIGN} in this example specifies the location of
5782 a section because it is used as the optional @var{address} attribute of
5783 a section definition (@pxref{Output Section Address}). The second use
5784 of @code{ALIGN} is used to defines the value of a symbol.
5786 The builtin function @code{NEXT} is closely related to @code{ALIGN}.
5788 @item ALIGNOF(@var{section})
5789 @kindex ALIGNOF(@var{section})
5790 @cindex section alignment
5791 Return the alignment in bytes of the named @var{section}, if that section has
5792 been allocated. If the section has not been allocated when this is
5793 evaluated, the linker will report an error. In the following example,
5794 the alignment of the @code{.output} section is stored as the first
5795 value in that section.
5800 LONG (ALIGNOF (.output))
5807 @item BLOCK(@var{exp})
5808 @kindex BLOCK(@var{exp})
5809 This is a synonym for @code{ALIGN}, for compatibility with older linker
5810 scripts. It is most often seen when setting the address of an output
5813 @item DATA_SEGMENT_ALIGN(@var{maxpagesize}, @var{commonpagesize})
5814 @kindex DATA_SEGMENT_ALIGN(@var{maxpagesize}, @var{commonpagesize})
5815 This is equivalent to either
5817 (ALIGN(@var{maxpagesize}) + (. & (@var{maxpagesize} - 1)))
5821 (ALIGN(@var{maxpagesize}) + (. & (@var{maxpagesize} - @var{commonpagesize})))
5824 depending on whether the latter uses fewer @var{commonpagesize} sized pages
5825 for the data segment (area between the result of this expression and
5826 @code{DATA_SEGMENT_END}) than the former or not.
5827 If the latter form is used, it means @var{commonpagesize} bytes of runtime
5828 memory will be saved at the expense of up to @var{commonpagesize} wasted
5829 bytes in the on-disk file.
5831 This expression can only be used directly in @code{SECTIONS} commands, not in
5832 any output section descriptions and only once in the linker script.
5833 @var{commonpagesize} should be less or equal to @var{maxpagesize} and should
5834 be the system page size the object wants to be optimized for (while still
5835 working on system page sizes up to @var{maxpagesize}).
5840 . = DATA_SEGMENT_ALIGN(0x10000, 0x2000);
5843 @item DATA_SEGMENT_END(@var{exp})
5844 @kindex DATA_SEGMENT_END(@var{exp})
5845 This defines the end of data segment for @code{DATA_SEGMENT_ALIGN}
5846 evaluation purposes.
5849 . = DATA_SEGMENT_END(.);
5852 @item DATA_SEGMENT_RELRO_END(@var{offset}, @var{exp})
5853 @kindex DATA_SEGMENT_RELRO_END(@var{offset}, @var{exp})
5854 This defines the end of the @code{PT_GNU_RELRO} segment when
5855 @samp{-z relro} option is used. Second argument is returned.
5856 When @samp{-z relro} option is not present, @code{DATA_SEGMENT_RELRO_END}
5857 does nothing, otherwise @code{DATA_SEGMENT_ALIGN} is padded so that
5858 @var{exp} + @var{offset} is aligned to the most commonly used page
5859 boundary for particular target. If present in the linker script,
5860 it must always come in between @code{DATA_SEGMENT_ALIGN} and
5861 @code{DATA_SEGMENT_END}.
5864 . = DATA_SEGMENT_RELRO_END(24, .);
5867 @item DEFINED(@var{symbol})
5868 @kindex DEFINED(@var{symbol})
5869 @cindex symbol defaults
5870 Return 1 if @var{symbol} is in the linker global symbol table and is
5871 defined before the statement using DEFINED in the script, otherwise
5872 return 0. You can use this function to provide
5873 default values for symbols. For example, the following script fragment
5874 shows how to set a global symbol @samp{begin} to the first location in
5875 the @samp{.text} section---but if a symbol called @samp{begin} already
5876 existed, its value is preserved:
5882 begin = DEFINED(begin) ? begin : . ;
5890 @item LENGTH(@var{memory})
5891 @kindex LENGTH(@var{memory})
5892 Return the length of the memory region named @var{memory}.
5894 @item LOADADDR(@var{section})
5895 @kindex LOADADDR(@var{section})
5896 @cindex section load address in expression
5897 Return the absolute LMA of the named @var{section}. (@pxref{Output
5901 @item MAX(@var{exp1}, @var{exp2})
5902 Returns the maximum of @var{exp1} and @var{exp2}.
5905 @item MIN(@var{exp1}, @var{exp2})
5906 Returns the minimum of @var{exp1} and @var{exp2}.
5908 @item NEXT(@var{exp})
5909 @kindex NEXT(@var{exp})
5910 @cindex unallocated address, next
5911 Return the next unallocated address that is a multiple of @var{exp}.
5912 This function is closely related to @code{ALIGN(@var{exp})}; unless you
5913 use the @code{MEMORY} command to define discontinuous memory for the
5914 output file, the two functions are equivalent.
5916 @item ORIGIN(@var{memory})
5917 @kindex ORIGIN(@var{memory})
5918 Return the origin of the memory region named @var{memory}.
5920 @item SEGMENT_START(@var{segment}, @var{default})
5921 @kindex SEGMENT_START(@var{segment}, @var{default})
5922 Return the base address of the named @var{segment}. If an explicit
5923 value has been given for this segment (with a command-line @samp{-T}
5924 option) that value will be returned; otherwise the value will be
5925 @var{default}. At present, the @samp{-T} command-line option can only
5926 be used to set the base address for the ``text'', ``data'', and
5927 ``bss'' sections, but you can use @code{SEGMENT_START} with any segment
5930 @item SIZEOF(@var{section})
5931 @kindex SIZEOF(@var{section})
5932 @cindex section size
5933 Return the size in bytes of the named @var{section}, if that section has
5934 been allocated. If the section has not been allocated when this is
5935 evaluated, the linker will report an error. In the following example,
5936 @code{symbol_1} and @code{symbol_2} are assigned identical values:
5945 symbol_1 = .end - .start ;
5946 symbol_2 = SIZEOF(.output);
5951 @item SIZEOF_HEADERS
5952 @itemx sizeof_headers
5953 @kindex SIZEOF_HEADERS
5955 Return the size in bytes of the output file's headers. This is
5956 information which appears at the start of the output file. You can use
5957 this number when setting the start address of the first section, if you
5958 choose, to facilitate paging.
5960 @cindex not enough room for program headers
5961 @cindex program headers, not enough room
5962 When producing an ELF output file, if the linker script uses the
5963 @code{SIZEOF_HEADERS} builtin function, the linker must compute the
5964 number of program headers before it has determined all the section
5965 addresses and sizes. If the linker later discovers that it needs
5966 additional program headers, it will report an error @samp{not enough
5967 room for program headers}. To avoid this error, you must avoid using
5968 the @code{SIZEOF_HEADERS} function, or you must rework your linker
5969 script to avoid forcing the linker to use additional program headers, or
5970 you must define the program headers yourself using the @code{PHDRS}
5971 command (@pxref{PHDRS}).
5974 @node Implicit Linker Scripts
5975 @section Implicit Linker Scripts
5976 @cindex implicit linker scripts
5977 If you specify a linker input file which the linker can not recognize as
5978 an object file or an archive file, it will try to read the file as a
5979 linker script. If the file can not be parsed as a linker script, the
5980 linker will report an error.
5982 An implicit linker script will not replace the default linker script.
5984 Typically an implicit linker script would contain only symbol
5985 assignments, or the @code{INPUT}, @code{GROUP}, or @code{VERSION}
5988 Any input files read because of an implicit linker script will be read
5989 at the position in the command line where the implicit linker script was
5990 read. This can affect archive searching.
5993 @node Machine Dependent
5994 @chapter Machine Dependent Features
5996 @cindex machine dependencies
5997 @command{ld} has additional features on some platforms; the following
5998 sections describe them. Machines where @command{ld} has no additional
5999 functionality are not listed.
6003 * H8/300:: @command{ld} and the H8/300
6006 * i960:: @command{ld} and the Intel 960 family
6009 * ARM:: @command{ld} and the ARM family
6012 * HPPA ELF32:: @command{ld} and HPPA 32-bit ELF
6015 * M68K:: @command{ld} and the Motorola 68K family
6018 * MMIX:: @command{ld} and MMIX
6021 * MSP430:: @command{ld} and MSP430
6024 * M68HC11/68HC12:: @code{ld} and the Motorola 68HC11 and 68HC12 families
6027 * PowerPC ELF32:: @command{ld} and PowerPC 32-bit ELF Support
6030 * PowerPC64 ELF64:: @command{ld} and PowerPC64 64-bit ELF Support
6033 * SPU ELF:: @command{ld} and SPU ELF Support
6036 * TI COFF:: @command{ld} and TI COFF
6039 * WIN32:: @command{ld} and WIN32 (cygwin/mingw)
6042 * Xtensa:: @command{ld} and Xtensa Processors
6053 @section @command{ld} and the H8/300
6055 @cindex H8/300 support
6056 For the H8/300, @command{ld} can perform these global optimizations when
6057 you specify the @samp{--relax} command-line option.
6060 @cindex relaxing on H8/300
6061 @item relaxing address modes
6062 @command{ld} finds all @code{jsr} and @code{jmp} instructions whose
6063 targets are within eight bits, and turns them into eight-bit
6064 program-counter relative @code{bsr} and @code{bra} instructions,
6067 @cindex synthesizing on H8/300
6068 @item synthesizing instructions
6069 @c FIXME: specifically mov.b, or any mov instructions really?
6070 @command{ld} finds all @code{mov.b} instructions which use the
6071 sixteen-bit absolute address form, but refer to the top
6072 page of memory, and changes them to use the eight-bit address form.
6073 (That is: the linker turns @samp{mov.b @code{@@}@var{aa}:16} into
6074 @samp{mov.b @code{@@}@var{aa}:8} whenever the address @var{aa} is in the
6075 top page of memory).
6077 @item bit manipulation instructions
6078 @command{ld} finds all bit manipulation instructions like @code{band, bclr,
6079 biand, bild, bior, bist, bixor, bld, bnot, bor, bset, bst, btst, bxor}
6080 which use 32 bit and 16 bit absolute address form, but refer to the top
6081 page of memory, and changes them to use the 8 bit address form.
6082 (That is: the linker turns @samp{bset #xx:3,@code{@@}@var{aa}:32} into
6083 @samp{bset #xx:3,@code{@@}@var{aa}:8} whenever the address @var{aa} is in
6084 the top page of memory).
6086 @item system control instructions
6087 @command{ld} finds all @code{ldc.w, stc.w} instructions which use the
6088 32 bit absolute address form, but refer to the top page of memory, and
6089 changes them to use 16 bit address form.
6090 (That is: the linker turns @samp{ldc.w @code{@@}@var{aa}:32,ccr} into
6091 @samp{ldc.w @code{@@}@var{aa}:16,ccr} whenever the address @var{aa} is in
6092 the top page of memory).
6102 @c This stuff is pointless to say unless you're especially concerned
6103 @c with Renesas chips; don't enable it for generic case, please.
6105 @chapter @command{ld} and Other Renesas Chips
6107 @command{ld} also supports the Renesas (formerly Hitachi) H8/300H,
6108 H8/500, and SH chips. No special features, commands, or command-line
6109 options are required for these chips.
6119 @section @command{ld} and the Intel 960 Family
6121 @cindex i960 support
6123 You can use the @samp{-A@var{architecture}} command line option to
6124 specify one of the two-letter names identifying members of the 960
6125 family; the option specifies the desired output target, and warns of any
6126 incompatible instructions in the input files. It also modifies the
6127 linker's search strategy for archive libraries, to support the use of
6128 libraries specific to each particular architecture, by including in the
6129 search loop names suffixed with the string identifying the architecture.
6131 For example, if your @command{ld} command line included @w{@samp{-ACA}} as
6132 well as @w{@samp{-ltry}}, the linker would look (in its built-in search
6133 paths, and in any paths you specify with @samp{-L}) for a library with
6146 The first two possibilities would be considered in any event; the last
6147 two are due to the use of @w{@samp{-ACA}}.
6149 You can meaningfully use @samp{-A} more than once on a command line, since
6150 the 960 architecture family allows combination of target architectures; each
6151 use will add another pair of name variants to search for when @w{@samp{-l}}
6152 specifies a library.
6154 @cindex @option{--relax} on i960
6155 @cindex relaxing on i960
6156 @command{ld} supports the @samp{--relax} option for the i960 family. If
6157 you specify @samp{--relax}, @command{ld} finds all @code{balx} and
6158 @code{calx} instructions whose targets are within 24 bits, and turns
6159 them into 24-bit program-counter relative @code{bal} and @code{cal}
6160 instructions, respectively. @command{ld} also turns @code{cal}
6161 instructions into @code{bal} instructions when it determines that the
6162 target subroutine is a leaf routine (that is, the target subroutine does
6163 not itself call any subroutines).
6165 @cindex Cortex-A8 erratum workaround
6166 @kindex --fix-cortex-a8
6167 @kindex --no-fix-cortex-a8
6168 The @samp{--fix-cortex-a8} switch enables a link-time workaround for an erratum in certain Cortex-A8 processors. The workaround is enabled by default if you are targeting the ARM v7-A architecture profile. It can be enabled otherwise by specifying @samp{--fix-cortex-a8}, or disabled unconditionally by specifying @samp{--no-fix-cortex-a8}.
6170 The erratum only affects Thumb-2 code. Please contact ARM for further details.
6172 @kindex --merge-exidx-entries
6173 @kindex --no-merge-exidx-entries
6174 The @samp{--no-merge-exidx-entries} switch disables the merging of adjacent exidx entries in debuginfo.
6191 @node M68HC11/68HC12
6192 @section @command{ld} and the Motorola 68HC11 and 68HC12 families
6194 @cindex M68HC11 and 68HC12 support
6196 @subsection Linker Relaxation
6198 For the Motorola 68HC11, @command{ld} can perform these global
6199 optimizations when you specify the @samp{--relax} command-line option.
6202 @cindex relaxing on M68HC11
6203 @item relaxing address modes
6204 @command{ld} finds all @code{jsr} and @code{jmp} instructions whose
6205 targets are within eight bits, and turns them into eight-bit
6206 program-counter relative @code{bsr} and @code{bra} instructions,
6209 @command{ld} also looks at all 16-bit extended addressing modes and
6210 transforms them in a direct addressing mode when the address is in
6211 page 0 (between 0 and 0x0ff).
6213 @item relaxing gcc instruction group
6214 When @command{gcc} is called with @option{-mrelax}, it can emit group
6215 of instructions that the linker can optimize to use a 68HC11 direct
6216 addressing mode. These instructions consists of @code{bclr} or
6217 @code{bset} instructions.
6221 @subsection Trampoline Generation
6223 @cindex trampoline generation on M68HC11
6224 @cindex trampoline generation on M68HC12
6225 For 68HC11 and 68HC12, @command{ld} can generate trampoline code to
6226 call a far function using a normal @code{jsr} instruction. The linker
6227 will also change the relocation to some far function to use the
6228 trampoline address instead of the function address. This is typically the
6229 case when a pointer to a function is taken. The pointer will in fact
6230 point to the function trampoline.
6238 @section @command{ld} and the ARM family
6240 @cindex ARM interworking support
6241 @kindex --support-old-code
6242 For the ARM, @command{ld} will generate code stubs to allow functions calls
6243 between ARM and Thumb code. These stubs only work with code that has
6244 been compiled and assembled with the @samp{-mthumb-interwork} command
6245 line option. If it is necessary to link with old ARM object files or
6246 libraries, which have not been compiled with the -mthumb-interwork
6247 option then the @samp{--support-old-code} command line switch should be
6248 given to the linker. This will make it generate larger stub functions
6249 which will work with non-interworking aware ARM code. Note, however,
6250 the linker does not support generating stubs for function calls to
6251 non-interworking aware Thumb code.
6253 @cindex thumb entry point
6254 @cindex entry point, thumb
6255 @kindex --thumb-entry=@var{entry}
6256 The @samp{--thumb-entry} switch is a duplicate of the generic
6257 @samp{--entry} switch, in that it sets the program's starting address.
6258 But it also sets the bottom bit of the address, so that it can be
6259 branched to using a BX instruction, and the program will start
6260 executing in Thumb mode straight away.
6262 @cindex PE import table prefixing
6263 @kindex --use-nul-prefixed-import-tables
6264 The @samp{--use-nul-prefixed-import-tables} switch is specifying, that
6265 the import tables idata4 and idata5 have to be generated with a zero
6266 elememt prefix for import libraries. This is the old style to generate
6267 import tables. By default this option is turned off.
6271 The @samp{--be8} switch instructs @command{ld} to generate BE8 format
6272 executables. This option is only valid when linking big-endian objects.
6273 The resulting image will contain big-endian data and little-endian code.
6276 @kindex --target1-rel
6277 @kindex --target1-abs
6278 The @samp{R_ARM_TARGET1} relocation is typically used for entries in the
6279 @samp{.init_array} section. It is interpreted as either @samp{R_ARM_REL32}
6280 or @samp{R_ARM_ABS32}, depending on the target. The @samp{--target1-rel}
6281 and @samp{--target1-abs} switches override the default.
6284 @kindex --target2=@var{type}
6285 The @samp{--target2=type} switch overrides the default definition of the
6286 @samp{R_ARM_TARGET2} relocation. Valid values for @samp{type}, their
6287 meanings, and target defaults are as follows:
6290 @samp{R_ARM_REL32} (arm*-*-elf, arm*-*-eabi)
6292 @samp{R_ARM_ABS32} (arm*-*-symbianelf)
6294 @samp{R_ARM_GOT_PREL} (arm*-*-linux, arm*-*-*bsd)
6299 The @samp{R_ARM_V4BX} relocation (defined by the ARM AAELF
6300 specification) enables objects compiled for the ARMv4 architecture to be
6301 interworking-safe when linked with other objects compiled for ARMv4t, but
6302 also allows pure ARMv4 binaries to be built from the same ARMv4 objects.
6304 In the latter case, the switch @option{--fix-v4bx} must be passed to the
6305 linker, which causes v4t @code{BX rM} instructions to be rewritten as
6306 @code{MOV PC,rM}, since v4 processors do not have a @code{BX} instruction.
6308 In the former case, the switch should not be used, and @samp{R_ARM_V4BX}
6309 relocations are ignored.
6311 @cindex FIX_V4BX_INTERWORKING
6312 @kindex --fix-v4bx-interworking
6313 Replace @code{BX rM} instructions identified by @samp{R_ARM_V4BX}
6314 relocations with a branch to the following veneer:
6322 This allows generation of libraries/applications that work on ARMv4 cores
6323 and are still interworking safe. Note that the above veneer clobbers the
6324 condition flags, so may cause incorrect progrm behavior in rare cases.
6328 The @samp{--use-blx} switch enables the linker to use ARM/Thumb
6329 BLX instructions (available on ARMv5t and above) in various
6330 situations. Currently it is used to perform calls via the PLT from Thumb
6331 code using BLX rather than using BX and a mode-switching stub before
6332 each PLT entry. This should lead to such calls executing slightly faster.
6334 This option is enabled implicitly for SymbianOS, so there is no need to
6335 specify it if you are using that target.
6337 @cindex VFP11_DENORM_FIX
6338 @kindex --vfp11-denorm-fix
6339 The @samp{--vfp11-denorm-fix} switch enables a link-time workaround for a
6340 bug in certain VFP11 coprocessor hardware, which sometimes allows
6341 instructions with denorm operands (which must be handled by support code)
6342 to have those operands overwritten by subsequent instructions before
6343 the support code can read the intended values.
6345 The bug may be avoided in scalar mode if you allow at least one
6346 intervening instruction between a VFP11 instruction which uses a register
6347 and another instruction which writes to the same register, or at least two
6348 intervening instructions if vector mode is in use. The bug only affects
6349 full-compliance floating-point mode: you do not need this workaround if
6350 you are using "runfast" mode. Please contact ARM for further details.
6352 If you know you are using buggy VFP11 hardware, you can
6353 enable this workaround by specifying the linker option
6354 @samp{--vfp-denorm-fix=scalar} if you are using the VFP11 scalar
6355 mode only, or @samp{--vfp-denorm-fix=vector} if you are using
6356 vector mode (the latter also works for scalar code). The default is
6357 @samp{--vfp-denorm-fix=none}.
6359 If the workaround is enabled, instructions are scanned for
6360 potentially-troublesome sequences, and a veneer is created for each
6361 such sequence which may trigger the erratum. The veneer consists of the
6362 first instruction of the sequence and a branch back to the subsequent
6363 instruction. The original instruction is then replaced with a branch to
6364 the veneer. The extra cycles required to call and return from the veneer
6365 are sufficient to avoid the erratum in both the scalar and vector cases.
6367 @cindex ARM1176 erratum workaround
6368 @kindex --fix-arm1176
6369 @kindex --no-fix-arm1176
6370 The @samp{--fix-arm1176} switch enables a link-time workaround for an erratum
6371 in certain ARM1176 processors. The workaround is enabled by default if you
6372 are targetting ARM v6 (excluding ARM v6T2) or earlier. It can be disabled
6373 unconditionally by specifying @samp{--no-fix-arm1176}.
6375 Further information is available in the ``ARM1176JZ-S and ARM1176JZF-S
6376 Programmer Advice Notice'' available on the ARM documentaion website at:
6377 http://infocenter.arm.com/.
6379 @cindex NO_ENUM_SIZE_WARNING
6380 @kindex --no-enum-size-warning
6381 The @option{--no-enum-size-warning} switch prevents the linker from
6382 warning when linking object files that specify incompatible EABI
6383 enumeration size attributes. For example, with this switch enabled,
6384 linking of an object file using 32-bit enumeration values with another
6385 using enumeration values fitted into the smallest possible space will
6388 @cindex NO_WCHAR_SIZE_WARNING
6389 @kindex --no-wchar-size-warning
6390 The @option{--no-wchar-size-warning} switch prevents the linker from
6391 warning when linking object files that specify incompatible EABI
6392 @code{wchar_t} size attributes. For example, with this switch enabled,
6393 linking of an object file using 32-bit @code{wchar_t} values with another
6394 using 16-bit @code{wchar_t} values will not be diagnosed.
6397 @kindex --pic-veneer
6398 The @samp{--pic-veneer} switch makes the linker use PIC sequences for
6399 ARM/Thumb interworking veneers, even if the rest of the binary
6400 is not PIC. This avoids problems on uClinux targets where
6401 @samp{--emit-relocs} is used to generate relocatable binaries.
6403 @cindex STUB_GROUP_SIZE
6404 @kindex --stub-group-size=@var{N}
6405 The linker will automatically generate and insert small sequences of
6406 code into a linked ARM ELF executable whenever an attempt is made to
6407 perform a function call to a symbol that is too far away. The
6408 placement of these sequences of instructions - called stubs - is
6409 controlled by the command line option @option{--stub-group-size=N}.
6410 The placement is important because a poor choice can create a need for
6411 duplicate stubs, increasing the code sizw. The linker will try to
6412 group stubs together in order to reduce interruptions to the flow of
6413 code, but it needs guidance as to how big these groups should be and
6414 where they should be placed.
6416 The value of @samp{N}, the parameter to the
6417 @option{--stub-group-size=} option controls where the stub groups are
6418 placed. If it is negative then all stubs are placed after the first
6419 branch that needs them. If it is positive then the stubs can be
6420 placed either before or after the branches that need them. If the
6421 value of @samp{N} is 1 (either +1 or -1) then the linker will choose
6422 exactly where to place groups of stubs, using its built in heuristics.
6423 A value of @samp{N} greater than 1 (or smaller than -1) tells the
6424 linker that a single group of stubs can service at most @samp{N} bytes
6425 from the input sections.
6427 The default, if @option{--stub-group-size=} is not specified, is
6430 Farcalls stubs insertion is fully supported for the ARM-EABI target
6431 only, because it relies on object files properties not present
6445 @section @command{ld} and HPPA 32-bit ELF Support
6446 @cindex HPPA multiple sub-space stubs
6447 @kindex --multi-subspace
6448 When generating a shared library, @command{ld} will by default generate
6449 import stubs suitable for use with a single sub-space application.
6450 The @samp{--multi-subspace} switch causes @command{ld} to generate export
6451 stubs, and different (larger) import stubs suitable for use with
6452 multiple sub-spaces.
6454 @cindex HPPA stub grouping
6455 @kindex --stub-group-size=@var{N}
6456 Long branch stubs and import/export stubs are placed by @command{ld} in
6457 stub sections located between groups of input sections.
6458 @samp{--stub-group-size} specifies the maximum size of a group of input
6459 sections handled by one stub section. Since branch offsets are signed,
6460 a stub section may serve two groups of input sections, one group before
6461 the stub section, and one group after it. However, when using
6462 conditional branches that require stubs, it may be better (for branch
6463 prediction) that stub sections only serve one group of input sections.
6464 A negative value for @samp{N} chooses this scheme, ensuring that
6465 branches to stubs always use a negative offset. Two special values of
6466 @samp{N} are recognized, @samp{1} and @samp{-1}. These both instruct
6467 @command{ld} to automatically size input section groups for the branch types
6468 detected, with the same behaviour regarding stub placement as other
6469 positive or negative values of @samp{N} respectively.
6471 Note that @samp{--stub-group-size} does not split input sections. A
6472 single input section larger than the group size specified will of course
6473 create a larger group (of one section). If input sections are too
6474 large, it may not be possible for a branch to reach its stub.
6487 @section @command{ld} and the Motorola 68K family
6489 @cindex Motorola 68K GOT generation
6490 @kindex --got=@var{type}
6491 The @samp{--got=@var{type}} option lets you choose the GOT generation scheme.
6492 The choices are @samp{single}, @samp{negative}, @samp{multigot} and
6493 @samp{target}. When @samp{target} is selected the linker chooses
6494 the default GOT generation scheme for the current target.
6495 @samp{single} tells the linker to generate a single GOT with
6496 entries only at non-negative offsets.
6497 @samp{negative} instructs the linker to generate a single GOT with
6498 entries at both negative and positive offsets. Not all environments
6500 @samp{multigot} allows the linker to generate several GOTs in the
6501 output file. All GOT references from a single input object
6502 file access the same GOT, but references from different input object
6503 files might access different GOTs. Not all environments support such GOTs.
6516 @section @code{ld} and MMIX
6517 For MMIX, there is a choice of generating @code{ELF} object files or
6518 @code{mmo} object files when linking. The simulator @code{mmix}
6519 understands the @code{mmo} format. The binutils @code{objcopy} utility
6520 can translate between the two formats.
6522 There is one special section, the @samp{.MMIX.reg_contents} section.
6523 Contents in this section is assumed to correspond to that of global
6524 registers, and symbols referring to it are translated to special symbols,
6525 equal to registers. In a final link, the start address of the
6526 @samp{.MMIX.reg_contents} section corresponds to the first allocated
6527 global register multiplied by 8. Register @code{$255} is not included in
6528 this section; it is always set to the program entry, which is at the
6529 symbol @code{Main} for @code{mmo} files.
6531 Global symbols with the prefix @code{__.MMIX.start.}, for example
6532 @code{__.MMIX.start..text} and @code{__.MMIX.start..data} are special.
6533 The default linker script uses these to set the default start address
6536 Initial and trailing multiples of zero-valued 32-bit words in a section,
6537 are left out from an mmo file.
6550 @section @code{ld} and MSP430
6551 For the MSP430 it is possible to select the MPU architecture. The flag @samp{-m [mpu type]}
6552 will select an appropriate linker script for selected MPU type. (To get a list of known MPUs
6553 just pass @samp{-m help} option to the linker).
6555 @cindex MSP430 extra sections
6556 The linker will recognize some extra sections which are MSP430 specific:
6559 @item @samp{.vectors}
6560 Defines a portion of ROM where interrupt vectors located.
6562 @item @samp{.bootloader}
6563 Defines the bootloader portion of the ROM (if applicable). Any code
6564 in this section will be uploaded to the MPU.
6566 @item @samp{.infomem}
6567 Defines an information memory section (if applicable). Any code in
6568 this section will be uploaded to the MPU.
6570 @item @samp{.infomemnobits}
6571 This is the same as the @samp{.infomem} section except that any code
6572 in this section will not be uploaded to the MPU.
6574 @item @samp{.noinit}
6575 Denotes a portion of RAM located above @samp{.bss} section.
6577 The last two sections are used by gcc.
6591 @section @command{ld} and PowerPC 32-bit ELF Support
6592 @cindex PowerPC long branches
6593 @kindex --relax on PowerPC
6594 Branches on PowerPC processors are limited to a signed 26-bit
6595 displacement, which may result in @command{ld} giving
6596 @samp{relocation truncated to fit} errors with very large programs.
6597 @samp{--relax} enables the generation of trampolines that can access
6598 the entire 32-bit address space. These trampolines are inserted at
6599 section boundaries, so may not themselves be reachable if an input
6600 section exceeds 33M in size. You may combine @samp{-r} and
6601 @samp{--relax} to add trampolines in a partial link. In that case
6602 both branches to undefined symbols and inter-section branches are also
6603 considered potentially out of range, and trampolines inserted.
6605 @cindex PowerPC ELF32 options
6610 Current PowerPC GCC accepts a @samp{-msecure-plt} option that
6611 generates code capable of using a newer PLT and GOT layout that has
6612 the security advantage of no executable section ever needing to be
6613 writable and no writable section ever being executable. PowerPC
6614 @command{ld} will generate this layout, including stubs to access the
6615 PLT, if all input files (including startup and static libraries) were
6616 compiled with @samp{-msecure-plt}. @samp{--bss-plt} forces the old
6617 BSS PLT (and GOT layout) which can give slightly better performance.
6619 @kindex --secure-plt
6621 @command{ld} will use the new PLT and GOT layout if it is linking new
6622 @samp{-fpic} or @samp{-fPIC} code, but does not do so automatically
6623 when linking non-PIC code. This option requests the new PLT and GOT
6624 layout. A warning will be given if some object file requires the old
6630 The new secure PLT and GOT are placed differently relative to other
6631 sections compared to older BSS PLT and GOT placement. The location of
6632 @code{.plt} must change because the new secure PLT is an initialized
6633 section while the old PLT is uninitialized. The reason for the
6634 @code{.got} change is more subtle: The new placement allows
6635 @code{.got} to be read-only in applications linked with
6636 @samp{-z relro -z now}. However, this placement means that
6637 @code{.sdata} cannot always be used in shared libraries, because the
6638 PowerPC ABI accesses @code{.sdata} in shared libraries from the GOT
6639 pointer. @samp{--sdata-got} forces the old GOT placement. PowerPC
6640 GCC doesn't use @code{.sdata} in shared libraries, so this option is
6641 really only useful for other compilers that may do so.
6643 @cindex PowerPC stub symbols
6644 @kindex --emit-stub-syms
6645 @item --emit-stub-syms
6646 This option causes @command{ld} to label linker stubs with a local
6647 symbol that encodes the stub type and destination.
6649 @cindex PowerPC TLS optimization
6650 @kindex --no-tls-optimize
6651 @item --no-tls-optimize
6652 PowerPC @command{ld} normally performs some optimization of code
6653 sequences used to access Thread-Local Storage. Use this option to
6654 disable the optimization.
6667 @node PowerPC64 ELF64
6668 @section @command{ld} and PowerPC64 64-bit ELF Support
6670 @cindex PowerPC64 ELF64 options
6672 @cindex PowerPC64 stub grouping
6673 @kindex --stub-group-size
6674 @item --stub-group-size
6675 Long branch stubs, PLT call stubs and TOC adjusting stubs are placed
6676 by @command{ld} in stub sections located between groups of input sections.
6677 @samp{--stub-group-size} specifies the maximum size of a group of input
6678 sections handled by one stub section. Since branch offsets are signed,
6679 a stub section may serve two groups of input sections, one group before
6680 the stub section, and one group after it. However, when using
6681 conditional branches that require stubs, it may be better (for branch
6682 prediction) that stub sections only serve one group of input sections.
6683 A negative value for @samp{N} chooses this scheme, ensuring that
6684 branches to stubs always use a negative offset. Two special values of
6685 @samp{N} are recognized, @samp{1} and @samp{-1}. These both instruct
6686 @command{ld} to automatically size input section groups for the branch types
6687 detected, with the same behaviour regarding stub placement as other
6688 positive or negative values of @samp{N} respectively.
6690 Note that @samp{--stub-group-size} does not split input sections. A
6691 single input section larger than the group size specified will of course
6692 create a larger group (of one section). If input sections are too
6693 large, it may not be possible for a branch to reach its stub.
6695 @cindex PowerPC64 stub symbols
6696 @kindex --emit-stub-syms
6697 @item --emit-stub-syms
6698 This option causes @command{ld} to label linker stubs with a local
6699 symbol that encodes the stub type and destination.
6701 @cindex PowerPC64 dot symbols
6703 @kindex --no-dotsyms
6704 @item --dotsyms, --no-dotsyms
6705 These two options control how @command{ld} interprets version patterns
6706 in a version script. Older PowerPC64 compilers emitted both a
6707 function descriptor symbol with the same name as the function, and a
6708 code entry symbol with the name prefixed by a dot (@samp{.}). To
6709 properly version a function @samp{foo}, the version script thus needs
6710 to control both @samp{foo} and @samp{.foo}. The option
6711 @samp{--dotsyms}, on by default, automatically adds the required
6712 dot-prefixed patterns. Use @samp{--no-dotsyms} to disable this
6715 @cindex PowerPC64 TLS optimization
6716 @kindex --no-tls-optimize
6717 @item --no-tls-optimize
6718 PowerPC64 @command{ld} normally performs some optimization of code
6719 sequences used to access Thread-Local Storage. Use this option to
6720 disable the optimization.
6722 @cindex PowerPC64 OPD optimization
6723 @kindex --no-opd-optimize
6724 @item --no-opd-optimize
6725 PowerPC64 @command{ld} normally removes @code{.opd} section entries
6726 corresponding to deleted link-once functions, or functions removed by
6727 the action of @samp{--gc-sections} or linker script @code{/DISCARD/}.
6728 Use this option to disable @code{.opd} optimization.
6730 @cindex PowerPC64 OPD spacing
6731 @kindex --non-overlapping-opd
6732 @item --non-overlapping-opd
6733 Some PowerPC64 compilers have an option to generate compressed
6734 @code{.opd} entries spaced 16 bytes apart, overlapping the third word,
6735 the static chain pointer (unused in C) with the first word of the next
6736 entry. This option expands such entries to the full 24 bytes.
6738 @cindex PowerPC64 TOC optimization
6739 @kindex --no-toc-optimize
6740 @item --no-toc-optimize
6741 PowerPC64 @command{ld} normally removes unused @code{.toc} section
6742 entries. Such entries are detected by examining relocations that
6743 reference the TOC in code sections. A reloc in a deleted code section
6744 marks a TOC word as unneeded, while a reloc in a kept code section
6745 marks a TOC word as needed. Since the TOC may reference itself, TOC
6746 relocs are also examined. TOC words marked as both needed and
6747 unneeded will of course be kept. TOC words without any referencing
6748 reloc are assumed to be part of a multi-word entry, and are kept or
6749 discarded as per the nearest marked preceding word. This works
6750 reliably for compiler generated code, but may be incorrect if assembly
6751 code is used to insert TOC entries. Use this option to disable the
6754 @cindex PowerPC64 multi-TOC
6755 @kindex --no-multi-toc
6756 @item --no-multi-toc
6757 If given any toc option besides @code{-mcmodel=medium} or
6758 @code{-mcmodel=large}, PowerPC64 GCC generates code for a TOC model
6760 entries are accessed with a 16-bit offset from r2. This limits the
6761 total TOC size to 64K. PowerPC64 @command{ld} extends this limit by
6762 grouping code sections such that each group uses less than 64K for its
6763 TOC entries, then inserts r2 adjusting stubs between inter-group
6764 calls. @command{ld} does not split apart input sections, so cannot
6765 help if a single input file has a @code{.toc} section that exceeds
6766 64K, most likely from linking multiple files with @command{ld -r}.
6767 Use this option to turn off this feature.
6769 @cindex PowerPC64 TOC sorting
6770 @kindex --no-toc-sort
6772 By default, @command{ld} sorts TOC sections so that those whose file
6773 happens to have a section called @code{.init} or @code{.fini} are
6774 placed first, followed by TOC sections referenced by code generated
6775 with PowerPC64 gcc's @code{-mcmodel=small}, and lastly TOC sections
6776 referenced only by code generated with PowerPC64 gcc's
6777 @code{-mcmodel=medium} or @code{-mcmodel=large} options. Doing this
6778 results in better TOC grouping for multi-TOC. Use this option to turn
6781 @cindex PowerPC64 PLT stub alignment
6783 @kindex --no-plt-align
6785 @itemx --no-plt-align
6786 Use these options to control whether individual PLT call stubs are
6787 aligned to a 32-byte boundary, or to the specified power of two
6788 boundary when using @code{--plt-align=}. By default PLT call stubs
6791 @cindex PowerPC64 PLT call stub static chain
6792 @kindex --plt-static-chain
6793 @kindex --no-plt-static-chain
6794 @item --plt-static-chain
6795 @itemx --no-plt-static-chain
6796 Use these options to control whether PLT call stubs load the static
6797 chain pointer (r11). @code{ld} defaults to not loading the static
6798 chain since there is never any need to do so on a PLT call.
6800 @cindex PowerPC64 PLT call stub thread safety
6801 @kindex --plt-thread-safe
6802 @kindex --no-plt-thread-safe
6803 @item --plt-thread-safe
6804 @itemx --no-thread-safe
6805 With power7's weakly ordered memory model, it is possible when using
6806 lazy binding for ld.so to update a plt entry in one thread and have
6807 another thread see the individual plt entry words update in the wrong
6808 order, despite ld.so carefully writing in the correct order and using
6809 memory write barriers. To avoid this we need some sort of read
6810 barrier in the call stub, or use LD_BIND_NOW=1. By default, @code{ld}
6811 looks for calls to commonly used functions that create threads, and if
6812 seen, adds the necessary barriers. Use these options to change the
6827 @section @command{ld} and SPU ELF Support
6829 @cindex SPU ELF options
6835 This option marks an executable as a PIC plugin module.
6837 @cindex SPU overlays
6838 @kindex --no-overlays
6840 Normally, @command{ld} recognizes calls to functions within overlay
6841 regions, and redirects such calls to an overlay manager via a stub.
6842 @command{ld} also provides a built-in overlay manager. This option
6843 turns off all this special overlay handling.
6845 @cindex SPU overlay stub symbols
6846 @kindex --emit-stub-syms
6847 @item --emit-stub-syms
6848 This option causes @command{ld} to label overlay stubs with a local
6849 symbol that encodes the stub type and destination.
6851 @cindex SPU extra overlay stubs
6852 @kindex --extra-overlay-stubs
6853 @item --extra-overlay-stubs
6854 This option causes @command{ld} to add overlay call stubs on all
6855 function calls out of overlay regions. Normally stubs are not added
6856 on calls to non-overlay regions.
6858 @cindex SPU local store size
6859 @kindex --local-store=lo:hi
6860 @item --local-store=lo:hi
6861 @command{ld} usually checks that a final executable for SPU fits in
6862 the address range 0 to 256k. This option may be used to change the
6863 range. Disable the check entirely with @option{--local-store=0:0}.
6866 @kindex --stack-analysis
6867 @item --stack-analysis
6868 SPU local store space is limited. Over-allocation of stack space
6869 unnecessarily limits space available for code and data, while
6870 under-allocation results in runtime failures. If given this option,
6871 @command{ld} will provide an estimate of maximum stack usage.
6872 @command{ld} does this by examining symbols in code sections to
6873 determine the extents of functions, and looking at function prologues
6874 for stack adjusting instructions. A call-graph is created by looking
6875 for relocations on branch instructions. The graph is then searched
6876 for the maximum stack usage path. Note that this analysis does not
6877 find calls made via function pointers, and does not handle recursion
6878 and other cycles in the call graph. Stack usage may be
6879 under-estimated if your code makes such calls. Also, stack usage for
6880 dynamic allocation, e.g. alloca, will not be detected. If a link map
6881 is requested, detailed information about each function's stack usage
6882 and calls will be given.
6885 @kindex --emit-stack-syms
6886 @item --emit-stack-syms
6887 This option, if given along with @option{--stack-analysis} will result
6888 in @command{ld} emitting stack sizing symbols for each function.
6889 These take the form @code{__stack_<function_name>} for global
6890 functions, and @code{__stack_<number>_<function_name>} for static
6891 functions. @code{<number>} is the section id in hex. The value of
6892 such symbols is the stack requirement for the corresponding function.
6893 The symbol size will be zero, type @code{STT_NOTYPE}, binding
6894 @code{STB_LOCAL}, and section @code{SHN_ABS}.
6908 @section @command{ld}'s Support for Various TI COFF Versions
6909 @cindex TI COFF versions
6910 @kindex --format=@var{version}
6911 The @samp{--format} switch allows selection of one of the various
6912 TI COFF versions. The latest of this writing is 2; versions 0 and 1 are
6913 also supported. The TI COFF versions also vary in header byte-order
6914 format; @command{ld} will read any version or byte order, but the output
6915 header format depends on the default specified by the specific target.
6928 @section @command{ld} and WIN32 (cygwin/mingw)
6930 This section describes some of the win32 specific @command{ld} issues.
6931 See @ref{Options,,Command Line Options} for detailed description of the
6932 command line options mentioned here.
6935 @cindex import libraries
6936 @item import libraries
6937 The standard Windows linker creates and uses so-called import
6938 libraries, which contains information for linking to dll's. They are
6939 regular static archives and are handled as any other static
6940 archive. The cygwin and mingw ports of @command{ld} have specific
6941 support for creating such libraries provided with the
6942 @samp{--out-implib} command line option.
6944 @item exporting DLL symbols
6945 @cindex exporting DLL symbols
6946 The cygwin/mingw @command{ld} has several ways to export symbols for dll's.
6949 @item using auto-export functionality
6950 @cindex using auto-export functionality
6951 By default @command{ld} exports symbols with the auto-export functionality,
6952 which is controlled by the following command line options:
6955 @item --export-all-symbols [This is the default]
6956 @item --exclude-symbols
6957 @item --exclude-libs
6958 @item --exclude-modules-for-implib
6959 @item --version-script
6962 When auto-export is in operation, @command{ld} will export all the non-local
6963 (global and common) symbols it finds in a DLL, with the exception of a few
6964 symbols known to belong to the system's runtime and libraries. As it will
6965 often not be desirable to export all of a DLL's symbols, which may include
6966 private functions that are not part of any public interface, the command-line
6967 options listed above may be used to filter symbols out from the list for
6968 exporting. The @samp{--output-def} option can be used in order to see the
6969 final list of exported symbols with all exclusions taken into effect.
6971 If @samp{--export-all-symbols} is not given explicitly on the
6972 command line, then the default auto-export behavior will be @emph{disabled}
6973 if either of the following are true:
6976 @item A DEF file is used.
6977 @item Any symbol in any object file was marked with the __declspec(dllexport) attribute.
6980 @item using a DEF file
6981 @cindex using a DEF file
6982 Another way of exporting symbols is using a DEF file. A DEF file is
6983 an ASCII file containing definitions of symbols which should be
6984 exported when a dll is created. Usually it is named @samp{<dll
6985 name>.def} and is added as any other object file to the linker's
6986 command line. The file's name must end in @samp{.def} or @samp{.DEF}.
6989 gcc -o <output> <objectfiles> <dll name>.def
6992 Using a DEF file turns off the normal auto-export behavior, unless the
6993 @samp{--export-all-symbols} option is also used.
6995 Here is an example of a DEF file for a shared library called @samp{xyz.dll}:
6998 LIBRARY "xyz.dll" BASE=0x20000000
7004 another_foo = abc.dll.afoo
7010 This example defines a DLL with a non-default base address and seven
7011 symbols in the export table. The third exported symbol @code{_bar} is an
7012 alias for the second. The fourth symbol, @code{another_foo} is resolved
7013 by "forwarding" to another module and treating it as an alias for
7014 @code{afoo} exported from the DLL @samp{abc.dll}. The final symbol
7015 @code{var1} is declared to be a data object. The @samp{doo} symbol in
7016 export library is an alias of @samp{foo}, which gets the string name
7017 in export table @samp{foo2}. The @samp{eoo} symbol is an data export
7018 symbol, which gets in export table the name @samp{var1}.
7020 The optional @code{LIBRARY <name>} command indicates the @emph{internal}
7021 name of the output DLL. If @samp{<name>} does not include a suffix,
7022 the default library suffix, @samp{.DLL} is appended.
7024 When the .DEF file is used to build an application, rather than a
7025 library, the @code{NAME <name>} command should be used instead of
7026 @code{LIBRARY}. If @samp{<name>} does not include a suffix, the default
7027 executable suffix, @samp{.EXE} is appended.
7029 With either @code{LIBRARY <name>} or @code{NAME <name>} the optional
7030 specification @code{BASE = <number>} may be used to specify a
7031 non-default base address for the image.
7033 If neither @code{LIBRARY <name>} nor @code{NAME <name>} is specified,
7034 or they specify an empty string, the internal name is the same as the
7035 filename specified on the command line.
7037 The complete specification of an export symbol is:
7041 ( ( ( <name1> [ = <name2> ] )
7042 | ( <name1> = <module-name> . <external-name>))
7043 [ @@ <integer> ] [NONAME] [DATA] [CONSTANT] [PRIVATE] [== <name3>] ) *
7046 Declares @samp{<name1>} as an exported symbol from the DLL, or declares
7047 @samp{<name1>} as an exported alias for @samp{<name2>}; or declares
7048 @samp{<name1>} as a "forward" alias for the symbol
7049 @samp{<external-name>} in the DLL @samp{<module-name>}.
7050 Optionally, the symbol may be exported by the specified ordinal
7051 @samp{<integer>} alias. The optional @samp{<name3>} is the to be used
7052 string in import/export table for the symbol.
7054 The optional keywords that follow the declaration indicate:
7056 @code{NONAME}: Do not put the symbol name in the DLL's export table. It
7057 will still be exported by its ordinal alias (either the value specified
7058 by the .def specification or, otherwise, the value assigned by the
7059 linker). The symbol name, however, does remain visible in the import
7060 library (if any), unless @code{PRIVATE} is also specified.
7062 @code{DATA}: The symbol is a variable or object, rather than a function.
7063 The import lib will export only an indirect reference to @code{foo} as
7064 the symbol @code{_imp__foo} (ie, @code{foo} must be resolved as
7067 @code{CONSTANT}: Like @code{DATA}, but put the undecorated @code{foo} as
7068 well as @code{_imp__foo} into the import library. Both refer to the
7069 read-only import address table's pointer to the variable, not to the
7070 variable itself. This can be dangerous. If the user code fails to add
7071 the @code{dllimport} attribute and also fails to explicitly add the
7072 extra indirection that the use of the attribute enforces, the
7073 application will behave unexpectedly.
7075 @code{PRIVATE}: Put the symbol in the DLL's export table, but do not put
7076 it into the static import library used to resolve imports at link time. The
7077 symbol can still be imported using the @code{LoadLibrary/GetProcAddress}
7078 API at runtime or by by using the GNU ld extension of linking directly to
7079 the DLL without an import library.
7081 See ld/deffilep.y in the binutils sources for the full specification of
7082 other DEF file statements
7084 @cindex creating a DEF file
7085 While linking a shared dll, @command{ld} is able to create a DEF file
7086 with the @samp{--output-def <file>} command line option.
7088 @item Using decorations
7089 @cindex Using decorations
7090 Another way of marking symbols for export is to modify the source code
7091 itself, so that when building the DLL each symbol to be exported is
7095 __declspec(dllexport) int a_variable
7096 __declspec(dllexport) void a_function(int with_args)
7099 All such symbols will be exported from the DLL. If, however,
7100 any of the object files in the DLL contain symbols decorated in
7101 this way, then the normal auto-export behavior is disabled, unless
7102 the @samp{--export-all-symbols} option is also used.
7104 Note that object files that wish to access these symbols must @emph{not}
7105 decorate them with dllexport. Instead, they should use dllimport,
7109 __declspec(dllimport) int a_variable
7110 __declspec(dllimport) void a_function(int with_args)
7113 This complicates the structure of library header files, because
7114 when included by the library itself the header must declare the
7115 variables and functions as dllexport, but when included by client
7116 code the header must declare them as dllimport. There are a number
7117 of idioms that are typically used to do this; often client code can
7118 omit the __declspec() declaration completely. See
7119 @samp{--enable-auto-import} and @samp{automatic data imports} for more
7123 @cindex automatic data imports
7124 @item automatic data imports
7125 The standard Windows dll format supports data imports from dlls only
7126 by adding special decorations (dllimport/dllexport), which let the
7127 compiler produce specific assembler instructions to deal with this
7128 issue. This increases the effort necessary to port existing Un*x
7129 code to these platforms, especially for large
7130 c++ libraries and applications. The auto-import feature, which was
7131 initially provided by Paul Sokolovsky, allows one to omit the
7132 decorations to achieve a behavior that conforms to that on POSIX/Un*x
7133 platforms. This feature is enabled with the @samp{--enable-auto-import}
7134 command-line option, although it is enabled by default on cygwin/mingw.
7135 The @samp{--enable-auto-import} option itself now serves mainly to
7136 suppress any warnings that are ordinarily emitted when linked objects
7137 trigger the feature's use.
7139 auto-import of variables does not always work flawlessly without
7140 additional assistance. Sometimes, you will see this message
7142 "variable '<var>' can't be auto-imported. Please read the
7143 documentation for ld's @code{--enable-auto-import} for details."
7145 The @samp{--enable-auto-import} documentation explains why this error
7146 occurs, and several methods that can be used to overcome this difficulty.
7147 One of these methods is the @emph{runtime pseudo-relocs} feature, described
7150 @cindex runtime pseudo-relocation
7151 For complex variables imported from DLLs (such as structs or classes),
7152 object files typically contain a base address for the variable and an
7153 offset (@emph{addend}) within the variable--to specify a particular
7154 field or public member, for instance. Unfortunately, the runtime loader used
7155 in win32 environments is incapable of fixing these references at runtime
7156 without the additional information supplied by dllimport/dllexport decorations.
7157 The standard auto-import feature described above is unable to resolve these
7160 The @samp{--enable-runtime-pseudo-relocs} switch allows these references to
7161 be resolved without error, while leaving the task of adjusting the references
7162 themselves (with their non-zero addends) to specialized code provided by the
7163 runtime environment. Recent versions of the cygwin and mingw environments and
7164 compilers provide this runtime support; older versions do not. However, the
7165 support is only necessary on the developer's platform; the compiled result will
7166 run without error on an older system.
7168 @samp{--enable-runtime-pseudo-relocs} is not the default; it must be explicitly
7171 @cindex direct linking to a dll
7172 @item direct linking to a dll
7173 The cygwin/mingw ports of @command{ld} support the direct linking,
7174 including data symbols, to a dll without the usage of any import
7175 libraries. This is much faster and uses much less memory than does the
7176 traditional import library method, especially when linking large
7177 libraries or applications. When @command{ld} creates an import lib, each
7178 function or variable exported from the dll is stored in its own bfd, even
7179 though a single bfd could contain many exports. The overhead involved in
7180 storing, loading, and processing so many bfd's is quite large, and explains the
7181 tremendous time, memory, and storage needed to link against particularly
7182 large or complex libraries when using import libs.
7184 Linking directly to a dll uses no extra command-line switches other than
7185 @samp{-L} and @samp{-l}, because @command{ld} already searches for a number
7186 of names to match each library. All that is needed from the developer's
7187 perspective is an understanding of this search, in order to force ld to
7188 select the dll instead of an import library.
7191 For instance, when ld is called with the argument @samp{-lxxx} it will attempt
7192 to find, in the first directory of its search path,
7204 before moving on to the next directory in the search path.
7206 (*) Actually, this is not @samp{cygxxx.dll} but in fact is @samp{<prefix>xxx.dll},
7207 where @samp{<prefix>} is set by the @command{ld} option
7208 @samp{--dll-search-prefix=<prefix>}. In the case of cygwin, the standard gcc spec
7209 file includes @samp{--dll-search-prefix=cyg}, so in effect we actually search for
7212 Other win32-based unix environments, such as mingw or pw32, may use other
7213 @samp{<prefix>}es, although at present only cygwin makes use of this feature. It
7214 was originally intended to help avoid name conflicts among dll's built for the
7215 various win32/un*x environments, so that (for example) two versions of a zlib dll
7216 could coexist on the same machine.
7218 The generic cygwin/mingw path layout uses a @samp{bin} directory for
7219 applications and dll's and a @samp{lib} directory for the import
7220 libraries (using cygwin nomenclature):
7226 libxxx.dll.a (in case of dll's)
7227 libxxx.a (in case of static archive)
7230 Linking directly to a dll without using the import library can be
7233 1. Use the dll directly by adding the @samp{bin} path to the link line
7235 gcc -Wl,-verbose -o a.exe -L../bin/ -lxxx
7238 However, as the dll's often have version numbers appended to their names
7239 (@samp{cygncurses-5.dll}) this will often fail, unless one specifies
7240 @samp{-L../bin -lncurses-5} to include the version. Import libs are generally
7241 not versioned, and do not have this difficulty.
7243 2. Create a symbolic link from the dll to a file in the @samp{lib}
7244 directory according to the above mentioned search pattern. This
7245 should be used to avoid unwanted changes in the tools needed for
7249 ln -s bin/cygxxx.dll lib/[cyg|lib|]xxx.dll[.a]
7252 Then you can link without any make environment changes.
7255 gcc -Wl,-verbose -o a.exe -L../lib/ -lxxx
7258 This technique also avoids the version number problems, because the following is
7265 libxxx.dll.a -> ../bin/cygxxx-5.dll
7268 Linking directly to a dll without using an import lib will work
7269 even when auto-import features are exercised, and even when
7270 @samp{--enable-runtime-pseudo-relocs} is used.
7272 Given the improvements in speed and memory usage, one might justifiably
7273 wonder why import libraries are used at all. There are three reasons:
7275 1. Until recently, the link-directly-to-dll functionality did @emph{not}
7276 work with auto-imported data.
7278 2. Sometimes it is necessary to include pure static objects within the
7279 import library (which otherwise contains only bfd's for indirection
7280 symbols that point to the exports of a dll). Again, the import lib
7281 for the cygwin kernel makes use of this ability, and it is not
7282 possible to do this without an import lib.
7284 3. Symbol aliases can only be resolved using an import lib. This is
7285 critical when linking against OS-supplied dll's (eg, the win32 API)
7286 in which symbols are usually exported as undecorated aliases of their
7287 stdcall-decorated assembly names.
7289 So, import libs are not going away. But the ability to replace
7290 true import libs with a simple symbolic link to (or a copy of)
7291 a dll, in many cases, is a useful addition to the suite of tools
7292 binutils makes available to the win32 developer. Given the
7293 massive improvements in memory requirements during linking, storage
7294 requirements, and linking speed, we expect that many developers
7295 will soon begin to use this feature whenever possible.
7297 @item symbol aliasing
7299 @item adding additional names
7300 Sometimes, it is useful to export symbols with additional names.
7301 A symbol @samp{foo} will be exported as @samp{foo}, but it can also be
7302 exported as @samp{_foo} by using special directives in the DEF file
7303 when creating the dll. This will affect also the optional created
7304 import library. Consider the following DEF file:
7307 LIBRARY "xyz.dll" BASE=0x61000000
7314 The line @samp{_foo = foo} maps the symbol @samp{foo} to @samp{_foo}.
7316 Another method for creating a symbol alias is to create it in the
7317 source code using the "weak" attribute:
7320 void foo () @{ /* Do something. */; @}
7321 void _foo () __attribute__ ((weak, alias ("foo")));
7324 See the gcc manual for more information about attributes and weak
7327 @item renaming symbols
7328 Sometimes it is useful to rename exports. For instance, the cygwin
7329 kernel does this regularly. A symbol @samp{_foo} can be exported as
7330 @samp{foo} but not as @samp{_foo} by using special directives in the
7331 DEF file. (This will also affect the import library, if it is
7332 created). In the following example:
7335 LIBRARY "xyz.dll" BASE=0x61000000
7341 The line @samp{_foo = foo} maps the exported symbol @samp{foo} to
7345 Note: using a DEF file disables the default auto-export behavior,
7346 unless the @samp{--export-all-symbols} command line option is used.
7347 If, however, you are trying to rename symbols, then you should list
7348 @emph{all} desired exports in the DEF file, including the symbols
7349 that are not being renamed, and do @emph{not} use the
7350 @samp{--export-all-symbols} option. If you list only the
7351 renamed symbols in the DEF file, and use @samp{--export-all-symbols}
7352 to handle the other symbols, then the both the new names @emph{and}
7353 the original names for the renamed symbols will be exported.
7354 In effect, you'd be aliasing those symbols, not renaming them,
7355 which is probably not what you wanted.
7357 @cindex weak externals
7358 @item weak externals
7359 The Windows object format, PE, specifies a form of weak symbols called
7360 weak externals. When a weak symbol is linked and the symbol is not
7361 defined, the weak symbol becomes an alias for some other symbol. There
7362 are three variants of weak externals:
7364 @item Definition is searched for in objects and libraries, historically
7365 called lazy externals.
7366 @item Definition is searched for only in other objects, not in libraries.
7367 This form is not presently implemented.
7368 @item No search; the symbol is an alias. This form is not presently
7371 As a GNU extension, weak symbols that do not specify an alternate symbol
7372 are supported. If the symbol is undefined when linking, the symbol
7373 uses a default value.
7375 @cindex aligned common symbols
7376 @item aligned common symbols
7377 As a GNU extension to the PE file format, it is possible to specify the
7378 desired alignment for a common symbol. This information is conveyed from
7379 the assembler or compiler to the linker by means of GNU-specific commands
7380 carried in the object file's @samp{.drectve} section, which are recognized
7381 by @command{ld} and respected when laying out the common symbols. Native
7382 tools will be able to process object files employing this GNU extension,
7383 but will fail to respect the alignment instructions, and may issue noisy
7384 warnings about unknown linker directives.
7398 @section @code{ld} and Xtensa Processors
7400 @cindex Xtensa processors
7401 The default @command{ld} behavior for Xtensa processors is to interpret
7402 @code{SECTIONS} commands so that lists of explicitly named sections in a
7403 specification with a wildcard file will be interleaved when necessary to
7404 keep literal pools within the range of PC-relative load offsets. For
7405 example, with the command:
7417 @command{ld} may interleave some of the @code{.literal}
7418 and @code{.text} sections from different object files to ensure that the
7419 literal pools are within the range of PC-relative load offsets. A valid
7420 interleaving might place the @code{.literal} sections from an initial
7421 group of files followed by the @code{.text} sections of that group of
7422 files. Then, the @code{.literal} sections from the rest of the files
7423 and the @code{.text} sections from the rest of the files would follow.
7425 @cindex @option{--relax} on Xtensa
7426 @cindex relaxing on Xtensa
7427 Relaxation is enabled by default for the Xtensa version of @command{ld} and
7428 provides two important link-time optimizations. The first optimization
7429 is to combine identical literal values to reduce code size. A redundant
7430 literal will be removed and all the @code{L32R} instructions that use it
7431 will be changed to reference an identical literal, as long as the
7432 location of the replacement literal is within the offset range of all
7433 the @code{L32R} instructions. The second optimization is to remove
7434 unnecessary overhead from assembler-generated ``longcall'' sequences of
7435 @code{L32R}/@code{CALLX@var{n}} when the target functions are within
7436 range of direct @code{CALL@var{n}} instructions.
7438 For each of these cases where an indirect call sequence can be optimized
7439 to a direct call, the linker will change the @code{CALLX@var{n}}
7440 instruction to a @code{CALL@var{n}} instruction, remove the @code{L32R}
7441 instruction, and remove the literal referenced by the @code{L32R}
7442 instruction if it is not used for anything else. Removing the
7443 @code{L32R} instruction always reduces code size but can potentially
7444 hurt performance by changing the alignment of subsequent branch targets.
7445 By default, the linker will always preserve alignments, either by
7446 switching some instructions between 24-bit encodings and the equivalent
7447 density instructions or by inserting a no-op in place of the @code{L32R}
7448 instruction that was removed. If code size is more important than
7449 performance, the @option{--size-opt} option can be used to prevent the
7450 linker from widening density instructions or inserting no-ops, except in
7451 a few cases where no-ops are required for correctness.
7453 The following Xtensa-specific command-line options can be used to
7456 @cindex Xtensa options
7459 When optimizing indirect calls to direct calls, optimize for code size
7460 more than performance. With this option, the linker will not insert
7461 no-ops or widen density instructions to preserve branch target
7462 alignment. There may still be some cases where no-ops are required to
7463 preserve the correctness of the code.
7471 @ifclear SingleFormat
7476 @cindex object file management
7477 @cindex object formats available
7479 The linker accesses object and archive files using the BFD libraries.
7480 These libraries allow the linker to use the same routines to operate on
7481 object files whatever the object file format. A different object file
7482 format can be supported simply by creating a new BFD back end and adding
7483 it to the library. To conserve runtime memory, however, the linker and
7484 associated tools are usually configured to support only a subset of the
7485 object file formats available. You can use @code{objdump -i}
7486 (@pxref{objdump,,objdump,binutils.info,The GNU Binary Utilities}) to
7487 list all the formats available for your configuration.
7489 @cindex BFD requirements
7490 @cindex requirements for BFD
7491 As with most implementations, BFD is a compromise between
7492 several conflicting requirements. The major factor influencing
7493 BFD design was efficiency: any time used converting between
7494 formats is time which would not have been spent had BFD not
7495 been involved. This is partly offset by abstraction payback; since
7496 BFD simplifies applications and back ends, more time and care
7497 may be spent optimizing algorithms for a greater speed.
7499 One minor artifact of the BFD solution which you should bear in
7500 mind is the potential for information loss. There are two places where
7501 useful information can be lost using the BFD mechanism: during
7502 conversion and during output. @xref{BFD information loss}.
7505 * BFD outline:: How it works: an outline of BFD
7509 @section How It Works: An Outline of BFD
7510 @cindex opening object files
7511 @include bfdsumm.texi
7514 @node Reporting Bugs
7515 @chapter Reporting Bugs
7516 @cindex bugs in @command{ld}
7517 @cindex reporting bugs in @command{ld}
7519 Your bug reports play an essential role in making @command{ld} reliable.
7521 Reporting a bug may help you by bringing a solution to your problem, or
7522 it may not. But in any case the principal function of a bug report is
7523 to help the entire community by making the next version of @command{ld}
7524 work better. Bug reports are your contribution to the maintenance of
7527 In order for a bug report to serve its purpose, you must include the
7528 information that enables us to fix the bug.
7531 * Bug Criteria:: Have you found a bug?
7532 * Bug Reporting:: How to report bugs
7536 @section Have You Found a Bug?
7537 @cindex bug criteria
7539 If you are not sure whether you have found a bug, here are some guidelines:
7542 @cindex fatal signal
7543 @cindex linker crash
7544 @cindex crash of linker
7546 If the linker gets a fatal signal, for any input whatever, that is a
7547 @command{ld} bug. Reliable linkers never crash.
7549 @cindex error on valid input
7551 If @command{ld} produces an error message for valid input, that is a bug.
7553 @cindex invalid input
7555 If @command{ld} does not produce an error message for invalid input, that
7556 may be a bug. In the general case, the linker can not verify that
7557 object files are correct.
7560 If you are an experienced user of linkers, your suggestions for
7561 improvement of @command{ld} are welcome in any case.
7565 @section How to Report Bugs
7567 @cindex @command{ld} bugs, reporting
7569 A number of companies and individuals offer support for @sc{gnu}
7570 products. If you obtained @command{ld} from a support organization, we
7571 recommend you contact that organization first.
7573 You can find contact information for many support companies and
7574 individuals in the file @file{etc/SERVICE} in the @sc{gnu} Emacs
7578 Otherwise, send bug reports for @command{ld} to
7582 The fundamental principle of reporting bugs usefully is this:
7583 @strong{report all the facts}. If you are not sure whether to state a
7584 fact or leave it out, state it!
7586 Often people omit facts because they think they know what causes the
7587 problem and assume that some details do not matter. Thus, you might
7588 assume that the name of a symbol you use in an example does not
7589 matter. Well, probably it does not, but one cannot be sure. Perhaps
7590 the bug is a stray memory reference which happens to fetch from the
7591 location where that name is stored in memory; perhaps, if the name
7592 were different, the contents of that location would fool the linker
7593 into doing the right thing despite the bug. Play it safe and give a
7594 specific, complete example. That is the easiest thing for you to do,
7595 and the most helpful.
7597 Keep in mind that the purpose of a bug report is to enable us to fix
7598 the bug if it is new to us. Therefore, always write your bug reports
7599 on the assumption that the bug has not been reported previously.
7601 Sometimes people give a few sketchy facts and ask, ``Does this ring a
7602 bell?'' This cannot help us fix a bug, so it is basically useless. We
7603 respond by asking for enough details to enable us to investigate.
7604 You might as well expedite matters by sending them to begin with.
7606 To enable us to fix the bug, you should include all these things:
7610 The version of @command{ld}. @command{ld} announces it if you start it with
7611 the @samp{--version} argument.
7613 Without this, we will not know whether there is any point in looking for
7614 the bug in the current version of @command{ld}.
7617 Any patches you may have applied to the @command{ld} source, including any
7618 patches made to the @code{BFD} library.
7621 The type of machine you are using, and the operating system name and
7625 What compiler (and its version) was used to compile @command{ld}---e.g.
7629 The command arguments you gave the linker to link your example and
7630 observe the bug. To guarantee you will not omit something important,
7631 list them all. A copy of the Makefile (or the output from make) is
7634 If we were to try to guess the arguments, we would probably guess wrong
7635 and then we might not encounter the bug.
7638 A complete input file, or set of input files, that will reproduce the
7639 bug. It is generally most helpful to send the actual object files
7640 provided that they are reasonably small. Say no more than 10K. For
7641 bigger files you can either make them available by FTP or HTTP or else
7642 state that you are willing to send the object file(s) to whomever
7643 requests them. (Note - your email will be going to a mailing list, so
7644 we do not want to clog it up with large attachments). But small
7645 attachments are best.
7647 If the source files were assembled using @code{gas} or compiled using
7648 @code{gcc}, then it may be OK to send the source files rather than the
7649 object files. In this case, be sure to say exactly what version of
7650 @code{gas} or @code{gcc} was used to produce the object files. Also say
7651 how @code{gas} or @code{gcc} were configured.
7654 A description of what behavior you observe that you believe is
7655 incorrect. For example, ``It gets a fatal signal.''
7657 Of course, if the bug is that @command{ld} gets a fatal signal, then we
7658 will certainly notice it. But if the bug is incorrect output, we might
7659 not notice unless it is glaringly wrong. You might as well not give us
7660 a chance to make a mistake.
7662 Even if the problem you experience is a fatal signal, you should still
7663 say so explicitly. Suppose something strange is going on, such as, your
7664 copy of @command{ld} is out of sync, or you have encountered a bug in the
7665 C library on your system. (This has happened!) Your copy might crash
7666 and ours would not. If you told us to expect a crash, then when ours
7667 fails to crash, we would know that the bug was not happening for us. If
7668 you had not told us to expect a crash, then we would not be able to draw
7669 any conclusion from our observations.
7672 If you wish to suggest changes to the @command{ld} source, send us context
7673 diffs, as generated by @code{diff} with the @samp{-u}, @samp{-c}, or
7674 @samp{-p} option. Always send diffs from the old file to the new file.
7675 If you even discuss something in the @command{ld} source, refer to it by
7676 context, not by line number.
7678 The line numbers in our development sources will not match those in your
7679 sources. Your line numbers would convey no useful information to us.
7682 Here are some things that are not necessary:
7686 A description of the envelope of the bug.
7688 Often people who encounter a bug spend a lot of time investigating
7689 which changes to the input file will make the bug go away and which
7690 changes will not affect it.
7692 This is often time consuming and not very useful, because the way we
7693 will find the bug is by running a single example under the debugger
7694 with breakpoints, not by pure deduction from a series of examples.
7695 We recommend that you save your time for something else.
7697 Of course, if you can find a simpler example to report @emph{instead}
7698 of the original one, that is a convenience for us. Errors in the
7699 output will be easier to spot, running under the debugger will take
7700 less time, and so on.
7702 However, simplification is not vital; if you do not want to do this,
7703 report the bug anyway and send us the entire test case you used.
7706 A patch for the bug.
7708 A patch for the bug does help us if it is a good one. But do not omit
7709 the necessary information, such as the test case, on the assumption that
7710 a patch is all we need. We might see problems with your patch and decide
7711 to fix the problem another way, or we might not understand it at all.
7713 Sometimes with a program as complicated as @command{ld} it is very hard to
7714 construct an example that will make the program follow a certain path
7715 through the code. If you do not send us the example, we will not be
7716 able to construct one, so we will not be able to verify that the bug is
7719 And if we cannot understand what bug you are trying to fix, or why your
7720 patch should be an improvement, we will not install it. A test case will
7721 help us to understand.
7724 A guess about what the bug is or what it depends on.
7726 Such guesses are usually wrong. Even we cannot guess right about such
7727 things without first using the debugger to find the facts.
7731 @appendix MRI Compatible Script Files
7732 @cindex MRI compatibility
7733 To aid users making the transition to @sc{gnu} @command{ld} from the MRI
7734 linker, @command{ld} can use MRI compatible linker scripts as an
7735 alternative to the more general-purpose linker scripting language
7736 described in @ref{Scripts}. MRI compatible linker scripts have a much
7737 simpler command set than the scripting language otherwise used with
7738 @command{ld}. @sc{gnu} @command{ld} supports the most commonly used MRI
7739 linker commands; these commands are described here.
7741 In general, MRI scripts aren't of much use with the @code{a.out} object
7742 file format, since it only has three sections and MRI scripts lack some
7743 features to make use of them.
7745 You can specify a file containing an MRI-compatible script using the
7746 @samp{-c} command-line option.
7748 Each command in an MRI-compatible script occupies its own line; each
7749 command line starts with the keyword that identifies the command (though
7750 blank lines are also allowed for punctuation). If a line of an
7751 MRI-compatible script begins with an unrecognized keyword, @command{ld}
7752 issues a warning message, but continues processing the script.
7754 Lines beginning with @samp{*} are comments.
7756 You can write these commands using all upper-case letters, or all
7757 lower case; for example, @samp{chip} is the same as @samp{CHIP}.
7758 The following list shows only the upper-case form of each command.
7761 @cindex @code{ABSOLUTE} (MRI)
7762 @item ABSOLUTE @var{secname}
7763 @itemx ABSOLUTE @var{secname}, @var{secname}, @dots{} @var{secname}
7764 Normally, @command{ld} includes in the output file all sections from all
7765 the input files. However, in an MRI-compatible script, you can use the
7766 @code{ABSOLUTE} command to restrict the sections that will be present in
7767 your output program. If the @code{ABSOLUTE} command is used at all in a
7768 script, then only the sections named explicitly in @code{ABSOLUTE}
7769 commands will appear in the linker output. You can still use other
7770 input sections (whatever you select on the command line, or using
7771 @code{LOAD}) to resolve addresses in the output file.
7773 @cindex @code{ALIAS} (MRI)
7774 @item ALIAS @var{out-secname}, @var{in-secname}
7775 Use this command to place the data from input section @var{in-secname}
7776 in a section called @var{out-secname} in the linker output file.
7778 @var{in-secname} may be an integer.
7780 @cindex @code{ALIGN} (MRI)
7781 @item ALIGN @var{secname} = @var{expression}
7782 Align the section called @var{secname} to @var{expression}. The
7783 @var{expression} should be a power of two.
7785 @cindex @code{BASE} (MRI)
7786 @item BASE @var{expression}
7787 Use the value of @var{expression} as the lowest address (other than
7788 absolute addresses) in the output file.
7790 @cindex @code{CHIP} (MRI)
7791 @item CHIP @var{expression}
7792 @itemx CHIP @var{expression}, @var{expression}
7793 This command does nothing; it is accepted only for compatibility.
7795 @cindex @code{END} (MRI)
7797 This command does nothing whatever; it's only accepted for compatibility.
7799 @cindex @code{FORMAT} (MRI)
7800 @item FORMAT @var{output-format}
7801 Similar to the @code{OUTPUT_FORMAT} command in the more general linker
7802 language, but restricted to one of these output formats:
7806 S-records, if @var{output-format} is @samp{S}
7809 IEEE, if @var{output-format} is @samp{IEEE}
7812 COFF (the @samp{coff-m68k} variant in BFD), if @var{output-format} is
7816 @cindex @code{LIST} (MRI)
7817 @item LIST @var{anything}@dots{}
7818 Print (to the standard output file) a link map, as produced by the
7819 @command{ld} command-line option @samp{-M}.
7821 The keyword @code{LIST} may be followed by anything on the
7822 same line, with no change in its effect.
7824 @cindex @code{LOAD} (MRI)
7825 @item LOAD @var{filename}
7826 @itemx LOAD @var{filename}, @var{filename}, @dots{} @var{filename}
7827 Include one or more object file @var{filename} in the link; this has the
7828 same effect as specifying @var{filename} directly on the @command{ld}
7831 @cindex @code{NAME} (MRI)
7832 @item NAME @var{output-name}
7833 @var{output-name} is the name for the program produced by @command{ld}; the
7834 MRI-compatible command @code{NAME} is equivalent to the command-line
7835 option @samp{-o} or the general script language command @code{OUTPUT}.
7837 @cindex @code{ORDER} (MRI)
7838 @item ORDER @var{secname}, @var{secname}, @dots{} @var{secname}
7839 @itemx ORDER @var{secname} @var{secname} @var{secname}
7840 Normally, @command{ld} orders the sections in its output file in the
7841 order in which they first appear in the input files. In an MRI-compatible
7842 script, you can override this ordering with the @code{ORDER} command. The
7843 sections you list with @code{ORDER} will appear first in your output
7844 file, in the order specified.
7846 @cindex @code{PUBLIC} (MRI)
7847 @item PUBLIC @var{name}=@var{expression}
7848 @itemx PUBLIC @var{name},@var{expression}
7849 @itemx PUBLIC @var{name} @var{expression}
7850 Supply a value (@var{expression}) for external symbol
7851 @var{name} used in the linker input files.
7853 @cindex @code{SECT} (MRI)
7854 @item SECT @var{secname}, @var{expression}
7855 @itemx SECT @var{secname}=@var{expression}
7856 @itemx SECT @var{secname} @var{expression}
7857 You can use any of these three forms of the @code{SECT} command to
7858 specify the start address (@var{expression}) for section @var{secname}.
7859 If you have more than one @code{SECT} statement for the same
7860 @var{secname}, only the @emph{first} sets the start address.
7863 @node GNU Free Documentation License
7864 @appendix GNU Free Documentation License
7868 @unnumbered LD Index
7873 % I think something like @colophon should be in texinfo. In the
7875 \long\def\colophon{\hbox to0pt{}\vfill
7876 \centerline{The body of this manual is set in}
7877 \centerline{\fontname\tenrm,}
7878 \centerline{with headings in {\bf\fontname\tenbf}}
7879 \centerline{and examples in {\tt\fontname\tentt}.}
7880 \centerline{{\it\fontname\tenit\/} and}
7881 \centerline{{\sl\fontname\tensl\/}}
7882 \centerline{are used for emphasis.}\vfill}
7884 % Blame: doc@cygnus.com, 28mar91.