When the -no-keep-memory option is given, BFD linker tries to save
memory in their own way. Since our internal architecture is completely
different from that linker, that option doesn't make sense to us.
llvm-svn: 301772
For an option -foo-bar-baz, we have Config->FooBarBaz. Since -rpath is
-rpath and not -r-path, it should be Config->Rpath instead Config->RPath.
llvm-svn: 301759
This version uses a set to speed up the synchronize method.
Original message:
Remove LinkerScript::flush.
This patch replaces flush with a last ditch attempt at synchronizing
the section list with the linker script "AST".
The synchronization is a bit of a hack and should in time be avoided
by creating the AST earlier so that modifications can be made directly
to it instead of modifying the section list and synchronizing it back.
This is the main step for fixing
https://bugs.llvm.org/show_bug.cgi?id=32816. With this in place I
think the only missing thing would be to have processCommands assign
section indexes as dummy offsets so that the sort in
OutputSection::finalize works.
With this LinkerScript::assignAddresses becomes much simpler, which
should help with the thunk work.
llvm-svn: 301745
Since the output format has been simplified, the class to print
out a map file doesn't seem to be needed anymore. We can replace
it with a few non-member functions.
llvm-svn: 301715
This patch is to ignore .debug_gnu_pub{names,types} sections if the
-gdb-index option was given.
Differential Revision: https://reviews.llvm.org/D32662
llvm-svn: 301710
This reverts commit r301678 since that change significantly slowed
down the linker. Before this patch, LLD could link clang in 8 seconds,
but with this patch it took 40 seconds.
llvm-svn: 301709
This change was motivated by output from lld-link.exe and link.exe
getting intermixed. There's already a flush() call in message(), so
there's precedence.
llvm-svn: 301693
Previously, we printed out input sections and input files in
separate columns as shown below.
Address Size Align Out In File Symbol
0000000000201000 0000000000000015 4 .text
0000000000201000 000000000000000e 4 .text
0000000000201000 000000000000000e 4 foo.o
0000000000201000 0000000000000000 0 _start
0000000000201005 0000000000000000 0 f(int)
000000000020100e 0000000000000000 0 local
0000000000201010 0000000000000002 4 bar.o
0000000000201010 0000000000000000 0 foo
0000000000201011 0000000000000000 0 bar
This format doesn't make much sense because for each input section,
there's always exactly one input file. This patch changes the format
to this.
Address Size Align Out In Symbol
0000000000201000 0000000000000015 4 .text
0000000000201000 000000000000000e 4 foo.o:(.text)
0000000000201000 0000000000000000 0 _start
0000000000201005 0000000000000000 0 f(int)
000000000020100e 0000000000000000 0 local
0000000000201010 0000000000000002 4 bar.o:(.text)
0000000000201010 0000000000000000 0 foo
0000000000201011 0000000000000000 0 bar
Differential Revision: https://reviews.llvm.org/D32657
llvm-svn: 301683
This patch replaces flush with a last ditch attempt at synchronizing
the section list with the linker script "AST".
The synchronization is a bit of a hack and should in time be avoided
by creating the AST earlier so that modifications can be made directly
to it instead of modifying the section list and synchronizing it back.
This is the main step for fixing
https://bugs.llvm.org/show_bug.cgi?id=32816. With this in place I
think the only missing thing would be to have processCommands assign
section indexes as dummy offsets so that the sort in
OutputSection::finalize works.
With this LinkerScript::assignAddresses becomes much simpler, which
should help with the thunk work.
llvm-svn: 301678
We found that some part of code for the -Map option takes O(m*n)
where m is the number of input sections in some file and n is
the number of symbols in the same file. If you do LTO, we usually
have only a few object files as inputs for the -Map option
feature, so this performance characteristic was worse than I
expected.
This patch rewrites the -Map option feature to speed it up.
I eliminated the O(m*n) bottleneck and also used multi-threading.
As a result, clang link time with the -Map option improved from
18.7 seconds to 11.2 seconds. Without -Map, it takes 7.7 seconds,
so the -Map option is now about 3x faster than before for this
test case (from 11.0 seconds to 3.5 seconds.) The generated output
file size was 223 MiB, and the file contains 1.2M lines.
Differential Revision: https://reviews.llvm.org/D32631
llvm-svn: 301659
This patch is to reduce amount of template uses. The new code is less
exciting and boring than before, but I think it is easier to read.
Differential Revision: https://reviews.llvm.org/D32467
llvm-svn: 301488
We were already pretty close, the one exception was when a name was
reused in another SECTIONS directive:
SECTIONS {
.text : { *(.text) }
.data : { *(.data) }
}
SECTIONS {
.data : { *(other) }
}
In this case we would create a single .data and magically output
"other" while looking at the first OutputSectionCommand.
We now create two .data sections. This matches what gold does. If we
really want to create a single one, we should change the parser so that
the above is parsed as if the user had written
SECTIONS {
.text : { *(.text) }
.data : { *(.data) *(other)}
}
That is, there should be only one OutputSectionCommand for .data and
it would have two InputSectionDescriptions.
By itself this patch makes the code a bit more complicated, but is an
important step in allowing assignAddresses to operate just on the
linker script.
llvm-svn: 301484
isFileWritable() checked if a file is writable and then emitted
an error message if it is not writable. So it did more than the
name says. This patch moves error() calls to Driver.
llvm-svn: 301422
This code was not used because of
handleARMTlsRelocation and handleMipsTlsRelocation methods that are called
for these platforms instead of regular TLS code.
Differential revision: https://reviews.llvm.org/D32355
llvm-svn: 301414
gnu ld description of option is:
--defsym=symbol=expression
Create a global symbol in the output file, containing the absolute address given
by expression. You may use this option as many times as necessary to define multiple
symbols in the command line. A limited form of arithmetic is supported for the
expression in this context: you may give a hexadecimal constant or the name of an
existing symbol, or use "+" and "-" to add or subtract hexadecimal constants or
symbols. If you need more elaborate expressions, consider using the linker command
language from a script. Note: there should be no white space between symbol,
the equals sign ("="), and expression.
In compare with D32082, this patch does not support math expressions and absolute
symbols. It implemented via code similar to --wrap. That covers 1 of 3 possible
--defsym cases.
Differential revision: https://reviews.llvm.org/D32171
llvm-svn: 301391
Previously we were not printing out the flags of the incompatible
section which made it difficult to determine what the problem was.
The error message format has been change to the following:
error: incompatible section flags for .bar
>>> /foo/bar/incompatible-section-flags.s.tmp.o:(.bar): 0x403
>>> output section .bar: 0x3
Patch by Alexander Richardson.
Differential Revision: https://reviews.llvm.org/D32484
llvm-svn: 301319
addIgnored defines a given symbol even if there is no existing
symbol with the same name. So, even if libc provides __tls_get_addr,
we should still be able to call addIgnored.
Differential Revision: https://reviews.llvm.org/D32053
llvm-svn: 301290
BSD's __progname symbol is defined in crt1.o and linked against main
executables. The libc expects that main executables export __progname
symbol via .dynsym sections. In order to handle this case, we scan
undefined symbols in DSOs and exported them by setting Sym->ExportDynamic
to true.
But it turned out that setting that variable is not enough to make sure
that symbols are exported in all use cases. If a -dynamic-list option is
given, all symbols not explicitly mentioned in a version script are
hidden by default. That hides __progname symbol. This patch fixes the issue.
Fixes https://bugs.llvm.org/show_bug.cgi?id=32703
llvm-svn: 301282
We can just use the existing SoName member variable. It now initially
contains what was in DefaultSoName and is modified if the .so has an
actual soname.
llvm-svn: 301259
This relocation type has an implicit addend. Account for it when
processing the relocation. Add an offset to an existing test to ensure
it gets processed correctly.
Fixes PR32634.
Differential Revision: https://reviews.llvm.org/D32336
llvm-svn: 301207
This change fabricates linker script commands for the case where there is
no linker script SECTIONS to control address assignment. This permits us
to have a single Script->assignAddresses() function.
There is a small change in user-visible-behavior with respect to the
handling of .tbss SHT_NOBITS, SHF_TLS as the Script->assignAddresses()
requires setDot() to be called with monotically increasing addresses.
The tls-offset.s test has been updated so that the script and non-script
results match.
This change should make the non-script behavior of lld closer to an
equivalent linker script.
Differential Revision: https://reviews.llvm.org/D31888
llvm-svn: 300687
createHeader didn't use data members of Elf_Chdr type and write
directly to a given buffer. That is not a good practice because
the function had a knowledge of the struct layout.
llvm-svn: 300674
Patch implements --compress-debug-sections=zlib.
In compare with D20211 (a year old patch, abandoned), it implementation
uses streaming and fully reimplemented, does not support zlib-gnu for
simplification.
This is PR32308.
Differential revision: https://reviews.llvm.org/D31941
llvm-svn: 300444
Imagine next script:
SECTIONS { BYTE(0x11); }
Section content written to disk will be 0x11. Previous LLD behavior was to make this
section SHT_NOBITS. What is not correct because section has content.
ld.bfd makes such sections SHT_PROGBITS, this patch do the same.
This fixes PR32537
Differential revision: https://reviews.llvm.org/D32016
llvm-svn: 300317
This fixes an assertion `Align != 0u && "Align can't be 0."'
in llvm::alignTo() when a linker script references a globally
defined variable in an ALIGN() context.
Patch by Alexander Richardson !
Differential revision: https://reviews.llvm.org/D31984
llvm-svn: 300315
Start using it in LLD to avoid needing to read bitcode again just to get the
target triple, and in llvm-lto2 to avoid printing symbol table information
that is inappropriate for the target.
Differential Revision: https://reviews.llvm.org/D32038
llvm-svn: 300300
We generally want to use uint64_t instead of uintX_t if the 64-bit
type works for both 32-bit and 64-bit because it is simpler than
the variable-size type.
llvm-svn: 300293
"read" is used as a prefix for functions that read tokens from input
streams. This function doesn't really read anything, but just parses
a given string as an integer, so rename.
llvm-svn: 300281
Replace addModuleReloc with AddTlsReloc so that we can use it for both the
module relocation and the offset relocation.
Differential Revision: https://reviews.llvm.org/D31751
llvm-svn: 300192
ARM Exception Index Table sections .ARM.exidx have an implicit dependency
on code sections via SHF_LINK_ORDER. When code sections are folded by ICF
we must mark the unique .ARM.exidx table that describes it as not live
to prevent an illegal entry in the exception table.
Note that we do not try and follow the relocations from the .ARM.exidx
section to the .ARM.extab sections to mark these as not live. Leaving
these sections is not a correctness problem. In theory these could be
removed via an application of garbage collection.
Fixes https://bugs.llvm.org/show_bug.cgi?id=32614
Differential Revision: https://reviews.llvm.org/D31932
llvm-svn: 300182
RELRO is a feature to make segments read-only after dynamic relocations
are applied. It is different from read-only segments because RELRO is
initially writable. And of course RELRO is different from writable
segments.
RELRO is not a very well known feature. We have a series of checks to
make a decision whether a section should be in a RELRO segment or not,
but we didn't describe why. This patch adds comments to explain how
that decision is made.
llvm-svn: 300176
Previously, we allowed only integers in this context. Now you can
write expressions there. LLD is now able to handle the following
linker, for example.
MEMORY { rom (rx) : ORIGIN = (1024 * 1024) }
llvm-svn: 300131
Fixes PR32572.
When
(a) a library has no soname
and (b) library is given on the command line with path (and not through -L/-l flags)
DT_NEEDED entry for such library keeps the path as given.
This behavior is consistent with gold and bfd, and is used in compiler-rt test suite.
This is a second attempt after r300007 got reverted. This time relro-omagic test is
changed in a way to avoid hardcoding the path to the test directory in the objdump'd
binary.
llvm-svn: 300011
Fixes PR32572.
When
(a) a library has no soname
and (b) library is given on the command line with path (and not through -L/-l flags)
DT_NEEDED entry for such library keeps the path as given.
This behavior is consistent with gold and bfd, and is used in compiler-rt test suite.
llvm-svn: 300007
Previously we silently produced broken output for R_386_GOT32X/R_386_GOT32
relocations if they were used to compute the address of the symbol’s global
offset table entry without base register when position-independent code is disabled.
Situation happened because of recent ABI changes. Released ABI mentions that
R_386_GOT32X can be calculated in a two different ways (so we did not follow ABI here
before this patch), but draft ABI also mentions R_386_GOT32 relocation here.
We should use the same calculations for both relocations.
Problem is that we always calculated them as G + A - GOT (offset from end of GOT),
but for case when PIC is disabled, according to i386 ABI calculation should be G + A,
what should produce just an address in GOT finally.
ABI: https://github.com/hjl-tools/x86-psABI/wiki/intel386-psABI-draft.pdf (p36, p60).
llvm-svn: 299812
Both functions always use the same GOT sections In<ELFT>::Got and
In<ELFT>::MipsGot respectively, so we do not need to pass them as an
argument.
llvm-svn: 299773
When the target of the TlsOffsetRel is non-preemptible we can write the
offset directly into the GOT without needing a dynamic relocation. This
is optional for dynamically linked executables but is required for static
linking.
This change adds the relocation to the GOT entry and a test case for
non-0 offsets so that if we miss out the offset the test won't spuriously
pass by virtue of the default value being 0.
Differential Revision: https://reviews.llvm.org/D31749
llvm-svn: 299751
The handleNoRelaxTlsRelocation handled both ARM and Mips as at a
high-level the actions of what to do when encountering a local dynamic or
global dynamic TLS relocation are the same. However due to Mips using a
custom GOT the differences of the implementation are enough that the
function became difficult to understand.
This change replaces handleNotRelaxTlsRelocation into
handleARMTlsRelocation() and handleMipsTlsRelocation() so that the ARM and
Mips specific code is isolated.
Differential Revision: https://reviews.llvm.org/D31748
llvm-svn: 299750
Executable sections should not be padded with zero by default. On some
architectures, 0x00 is the start of a valid instruction sequence, so can confuse
disassembly between InputSections (and indeed the start of the next InputSection
in some situations). Further, in the case of misjumps into padding, padding may
start to be executed silently.
On x86, the "0xcc" byte represents the int3 trap instruction. It is a single
byte long so can serve well as padding. This change switches x86 (and x86_64) to
use this value for padding in executable sections, if no linker script directive
overrides it. It also puts the behaviour into place making it easy to change the
behaviour of other targets when desired. I do not know the relevant instruction
sequences for trap instructions on other targets however, so somebody should add
this separately.
Because the old behaviour simply wrote padding in the whole section before
overwriting most of it, this change also modifies the padding algorithm to write
padding only where needed. This in turn has caused a small behaviour change with
regards to what values are written via Fill commands in linker scripts, bringing
it into line with ld.bfd. The fill value is now written starting from the end of
the previous block, which means that it always starts from the first byte of the
fill, whereas the old behaviour meant that the padding sometimes started mid-way
through the fill value. See the test changes for more details.
Reviewed by: ruiu
Differential Revision: https://reviews.llvm.org/D30886
Bugzilla: http://bugs.llvm.org/show_bug.cgi?id=32227
llvm-svn: 299635
scanRelocs() does a lot of things. It fills InputSection's Relocations vector,
making a decision whether a TLS relocation should be relaxed or not,
and making a decision whether a GOT/PLT slot needs to be created or not.
They don't actually have to be done in a single loop. I want to separate
them so that some of them can be run concurently. As a first step, this
patch moves PLT/GOT slot assignment to beginning of the loop, so that
they just fall through to the next statements. This should make it clear
that that code doesn't affect other parts of the loop.
llvm-svn: 299615
Relocations are abstracted as platform-independent R_TLS_* relocations,
so we don't need to check platform-specific ones to see if a relocation
is TLS GD.
llvm-svn: 299614
Symbols referenced by linker scripts are not necessarily be undefined,
so the previous name didn't convey the meaining of the variable.
llvm-svn: 299573
Previously, the code we set to our .plt entries expected that .got
and .got.plt are consecutive in the virtual address space.
Since %ebx points to the last entry of .got for position-independent
code, it assumed that .got is accessible with small negative
displacements and .got.plt are accessible with small positive
displacements.
That assumption was simply wrong. We don't impose any restrictions on
relative layout of .got and .got.plt. As a result, the control is
transferred to a bogus address from .plt at runtime, which resulted in
segfaults.
This patch removes that wrong assumption. We still assume that .got.plt
has a fixed relative address to .got, but we no longer assume that they
are consecutive in memory.
With this change, a "hello world" program compiled with -fPIC works.
Fixes https://bugs.llvm.org/show_bug.cgi?id=31332.
Differential Revision: https://reviews.llvm.org/D31682
llvm-svn: 299553
For range extension thunks we will need to repeatedly call createThunks()
until no more thunks are created. We will need to retain the state of
Thunks that we have created so far to avoid recreating them on later
passes. This change does not change the functionality of createThunks().
Differential Revision: https://reviews.llvm.org/D31654
llvm-svn: 299530
GNU linkers define __bss_start symbol.
Patch teaches LLD to do that. This is PR32051.
Below is part of standart ld.bfd script:
.data1 : { *(.data1) }
_edata = .; PROVIDE (edata = .);
. = .;
__bss_start = .;
.bss :
{
Currently LLD can emit up to 3 .bss* sections as one of testcase shows.
Implementation inserts this symbol before first .bss* output section.
Differential revision: https://reviews.llvm.org/D30419
llvm-svn: 299528
Rui Ueyama