Summary: MSVC allows linker options to be specified in source code. One of these is the /INCLUDE directive, which specifies that a symbol must be added to the symbol table, even if it otherwise wouldn't be. Existing tests cover the case where the linker is given an object file with an /INCLUDE directive, but we also need to cover the case where /INCLUDE is specified in a bitcode file (as would happen when using LTO). This new test covers that case.
Reviewers: pcc, ruiu
Reviewed By: ruiu
Differential Revision: https://reviews.llvm.org/D29096
llvm-svn: 293107
This patch is to merge type info in multiple .debug$T sections.
One mystery that needs to be solved is that it is not clear how
the MSVC linker uses TPI and IPI streams. Both streams contain
type info, and it is not obvious what kind of record should go
which.
dumppdb command in microsoft-pdb repository prints out IPI stream
contents as "IDs" and TPI stream as "TYPES", but looks like the tool
don't really care about which stream type recrods were read from.
For now, in this patch, I emit all type records to TPI stream.
It might just work with other tools. If not, we need to investigate
it more.
llvm-svn: 291739
This broke the following two bots:
lld-x86_64-win7: the test failed because a diff command is not available
on that bot. That's a configuration error of the bot and will be fixed soon.
llvm-clang-lld-x86_64-scei-ps4-windows10pro-fast: "tar xf" failed on that
bot. I suspect that it is due to the maximum path limitation on Windows.
A build directory contains a buildbot name, so it's longer than usual on
that machine. On Windows, many filesystem operations fail if a path is
longer than 255 characters. I'll try to address that in another patch.
llvm-svn: 291527
I think generated tar files are more compatible with old tar commands
because of r291494 and r291340, so I want to enable this test on buildbots.
llvm-svn: 291526
This is how we use TarWriter in LLD. Now LLD does not append
a file extension, so you need to pass `--reproduce foo.tar`
instead of `--reproduce foo`.
Differential Revision: https://reviews.llvm.org/D28103
llvm-svn: 291210
The GUID should match between the RSDS and the PDB. This should repair
the build bots, though we should be ensuring that the GUIDs match.
Unfortunately, different build bots seem to be getting different GUIDs.
llvm-svn: 290981
The PDB GUID, Age, and version are tied together by the RSDS record in
the binary. Pass along the BuildId information into the createPDB to
allow us to tie the binary and the PDB together.
llvm-svn: 290975
Profiling revealed that the majority of lld's execution time on Windows was
spent opening and mapping input files. We can reduce this cost significantly
by performing these operations asynchronously.
This change introduces a queue for all operations on input file data. When
we discover that we need to load a file (for example, when we find a lazy
archive for an undefined symbol, or when we read a linker directive to
load a file from disk), the file operation is launched using a future and
the symbol resolution operation is enqueued. This implies another change
to symbol resolution semantics, but it seems to be harmless ("ninja All"
in Chromium still succeeds).
To measure the perf impact of this change I linked Chromium's chrome_child.dll
with both thin and fat archives.
Thin archives:
Before (median of 5 runs): 19.50s
After: 10.93s
Fat archives:
Before: 12.00s
After: 9.90s
On Linux I found that doing this asynchronously had a negative effect on
performance, probably because the cost of mapping a file is small enough that
it becomes outweighed by the cost of managing the futures. So on non-Windows
platforms I use the deferred execution strategy.
Differential Revision: https://reviews.llvm.org/D27768
llvm-svn: 289760
This patch replaces the symbol table's object and archive queues, as well as
the convergent loop in the linker driver, with a design more similar to the
ELF linker where symbol resolution directly causes input files to be added to
the link, including input files arising from linker directives. Effectively
this removes the last vestiges of the old parallel input file loader.
Differential Revision: https://reviews.llvm.org/D27660
llvm-svn: 289409
This ports the ELF linker's symbol table design, introduced in r268178,
to the COFF linker.
Differential Revision: http://reviews.llvm.org/D21166
llvm-svn: 289280
The former option bases the filename on the output name, e.g. if the
link output is a.exe, the map will be written to a.map. This matches the
behaviour of link.exe's /MAP option and is useful for creating a map
file of each executable when building a large project.
Differential Revision: https://reviews.llvm.org/D27595
llvm-svn: 289271
I don't think the data I add to a TPI stream in this patch is correct,
but at least it can be displayed using llvm-pdbdump. Until I add more
streams to a PDB file, I'm not able to know whether the data will be
accepted by MSVC tools or not.
llvm-svn: 289183
Previously, we had different way to stringize SymbolBody and InputFile
to construct error messages. This patch defines overloaded function
toString() so that we don't need to memorize all these different
function names.
With that change, it is now easy to include demangled names in error
messages. Now, if there is a symbol name conflict, we'll print out
both mangled and demangled names.
llvm-svn: 288992
Associative sections are sections that need to be linked if their associated
sections are linked. Associative sections are used to append auxiliary data
such as debug info.
Previously, we compared all associative sections when comparing two comdat
sections. Because usually assocative sections are not mergeable sections,
we missed a lot of mergeable sections. MSVC linker doesn't seem to check
the identity of associative sections.
This patch makes LLD to ignore associative sections when doing ICF.
llvm-svn: 288483
Previously, we discarded .debug$ sections. This patch adds them to
files so that PDB.cpp can access them.
This patch also adds a debug option, /dumppdb, to dump debug info
fed to createPDB so that we can verify that valid data has been passed.
llvm-svn: 287555
Object files compiled with cl.exe /GL contain intermediate code for LTO.
We can't (and don't want to) interpret such code, but we should print
out a user-friendly error message.
Differential Revision: https://reviews.llvm.org/D26647
llvm-svn: 286921
I don't really understand what is failing on lld-x86_64-darwin13 bot,
but this patch should at least reduces the number of moving parts.
llvm-svn: 286876
Following the lazy reference might bring in an object file that depends
on bitcode files that weren't part of the LTO step.
Differential Revision: https://reviews.llvm.org/D25461
llvm-svn: 283989
I don't really understand why we get a larger .rodata section only
on this bot. I guess it may be picking up a library which contains
a .rodata. I removed the specific values since their values are not
important for this test case.
llvm-svn: 283931
With this, "llvm-pdbdump yaml -ipi-stream" prints out an IPI stream.
Previously it crashed because it can't handle the case where IPI
stream doesn't exist.
llvm-svn: 283392
Handle this in the exact same way as IMAGE_REL_AMD64_SECREL
and IMAGE_REL_I386_SECREL.
Differential revision: https://reviews.llvm.org/D24608
llvm-svn: 282531
test/COFF/rsds.test checks only RSDS directory in a DLL and
didn't check the validity of the PDF file produced.
(Technically the produced PDB is not valid because it is really
a stub, but this test is still good to have.)
llvm-svn: 281678
Although the GUID seems to be stable across test runs now, it seems to be
unstable across hosts. Lets be a bit more lax about the reading of the RSDS
record.
llvm-svn: 281083
Change the way we calculate the build id to use MD5 to give reproducible build
ids. Previously we would generate random bytes for the build id GUID.
llvm-svn: 281079
The IMAGE_FILE_HEADER structure contains a (RVA, size) to an array of
COFF_DEBUG_DIRECTORY records. Each one of these records contains an RVA to a OMF
Debug Directory. These OMF debug directories are derived into newer types such
as PDB70, PDB20, etc. This constructs a PDB70 structure which will allow us to
associate a GUID with a build to actually tie debug information.
llvm-svn: 280012
Don't blindly OR in the new value, but clear the existing one, since it can be
nonzero. Read out the existing value before, and add into the desired offset.
(The add is done outside of the applyMOV, to handle potential overflow between
the two.)
Patch by Martin Storsjö!
llvm-svn: 277846
The opcode for the bl branches can initially be F000 F800, i.e.
the J1 and J2 bits are already set. Therefore mask these bits out
before or'ing in the new bits.
Patch by Martin Storsjö!
llvm-svn: 277836
This flag is implemented similarly to --reproduce in the ELF linker.
This patch implements /linkrepro by moving the cpio writer and associated
utility functions to lldCore, and using that implementation in both linkers.
One COFF-specific detail is that we store the object file from which the
resource files were created in our reproducer, rather than the resource
files themselves. This allows the reproducer to be used on non-Windows
systems for example.
Differential Revision: https://reviews.llvm.org/D22418
llvm-svn: 276719
Manifest file is a separate or embedded XML file having metadata
of an executable. As it is XML, it can contain various types of
information. Probably the most popular one is to request escalated
priviledges.
Usually the linker creates an XML file and embed that file into
an executable. However, there's a way to supply an XML file from
command line. /manifestniput is it.
Apparently it is over-designed here, but if you supply two or more
manifest files, then the linker needs to merge the files into a
single XML file. A good news is that we don't need to do that ourselves.
MT.exe command can do that, so we call the command from the linker
in this patch.
llvm-svn: 266704
Some COFF tests used INT_MIN for the alignment of the directive section.
This is invalid; replace the alignment with something more sensible: 1.
llvm-svn: 263723
This fixes a test which exposed an ASan issue.
We assumed that a symbol's section number had a corresponding section
without performing validation.
llvm-svn: 263558
The load configuration directory is a structure whose size varies as the
OS gains additional functionality. To account for this, the structure's
layout begins with a size field; this allows loaders to know which
fields are available.
However, LLD hard-coded the sizes (112 bytes for 64-bit and 64 for
32-bit). This means that we might not inform the loader of all the
pertinent fields or we might claim that there are more fields than are
actually present.
To correctly account for this, the size field must be loaded from the
_load_config_used symbol.
N.B. The COFF spec is either wrong or out of date, the load
configuration directory is not correctly documented in the
specification: it omits the size field.
llvm-svn: 263543
The TLS directory has a different layout depending on the bitness of the
machine the image will run on. LLD would always use the 64-bit TLS
directory for the data directory entry instead of an appropriately sized
TLS directory.
llvm-svn: 263539
This test is flaky for more than half a year or so on buildbots
and has been causing confusion. Remove it while I'm investing the
cause.
llvm-svn: 261709
DLL export tables usually contain dllexport'ed symbol RVAs so that
applications which use the DLLs can find symbols from the DLLs.
However, there's a minor feature to "forward" DLL symbols to other
DLLs.
If you set an RVA to a string whose form is "<dllname>.<symbolname>"
(e.g. "KERNEL32.ExitProcess") instead of symbol RVA to the export
table, the loader interprets that as a forwarder symbol, and resolve
that symbol from the specified DLL.
This patch implements that feature.
llvm-svn: 257243
If a section symbol is not external, that COMDAT section should never
be merge with other sections in other compilation unit. Previously,
we didn't take visibility into account.
Note that COMDAT sections with non-external visibility makes sense
because they can be removed by dead-stripping.
Fixes https://llvm.org/bugs/show_bug.cgi?id=25686
llvm-svn: 254578
There's actually a room to improve this patch. Instead of not merging
sections that have different alignements, we can choose the section that
has the largest alignment requirement among all sections that are otherwise
considered the same. Then all section alignments are satisfied, so we can
merge them.
I don't know if that improvement could make any difference for real-world
input, so I'll leave it alone. Would be interesting to revisit later.
llvm-svn: 248581
This is an LLD extension to MSVC link.exe command line. MSVC linker
does not write symbol tables for executables. We do unless no /debug
option is given.
There's a situation that we want to enable debug info but don't want
to emit the symbol table. One example is when we are comparing output
file size. With this patch, you can tell the linker to not create
a symbol table by just specifying /nosymtab.
llvm-svn: 248225
This patch fixes a regression introduced by r247964. Relocations that
are referring the same symbol should be considered equal, but they
were not if they were pointing to non-section chunks.
llvm-svn: 248132
Only live symbols are written to the symbol table. Because isLive()
returned false if dead-stripping was disabled entirely, only
non-COMDAT sections were written to the symbol table. This patch fixes
the issue.
llvm-svn: 247856
This is a patch to make LLD to be on par with MSVC in terms of ICF
effectiveness. MSVC produces a 27.14MB executable when linking LLD.
LLD previously produced a 27.61MB when self-linking. Now the size
is reduced to 27.11MB. Note that without ICF the size is 29.63MB.
In r247387, I implemented an algorithm that handles section graphs
as cyclic graphs and merge them using SCC. The algorithm did not
always work as intended as I demonstrated in r247721. The new
algortihm implemented in this patch is different from the previous
one. If you are interested the details, you want to read the file
comment of ICF.cpp.
llvm-svn: 247770
In this test, we have two functions, foo and bar. MSVC linker can
choose one and discard the other using ICF. LLD cannot. I add this
test as a TODO.
foo and bar are conceptually equivalent to the following:
void foo() { foo(); }
void bar() { foo(); }
foo and bar are effectively the same function. If foo and bar are
compiled to the same instructions, both their contents (foo and bar)
and relocation targets (foo) become the same, so from the ICF point
of view, they are reducible. But their graphs are not isomorphic!
LLD's ICF algorithm cannot handle this case yet.
llvm-svn: 247721
Previously, LLD's ICF couldn't merge cyclic graphs. That was unfortunate
because, in COFF, cyclic graphs are not exceptional at all. That is
pretty common.
In this patch, sections are grouped by Tarjan's strongly connected
component algorithm to get acyclic graphs. And then we try to merge
SCCs whose outdegree is zero, and remove them from the graph. This
makes other SCCs to have outdegree zero, so we can repeat the
process until all SCCs are removed. When comparing two SCCs, we handle
cycles properly.
This algorithm works better than previous one. Previously, self-linking
produced a 29.0MB executable. It now produces a 27.7MB. There's still some
gap compared to MSVC linker which produces a 27.1MB executable for the
same input. So the gap is narrowed, but still LLD is not on par with MSVC.
I'll investigate that later.
llvm-svn: 247387
I don't understand why the previous code is pretty flaky and
the new code is at least less flaky, but the original test
occasionally failed on the second run of lib.exe.
My guess was that lib.exe was failing because the output of
the echo command executed immediately before lib.exe was not
flushed to a file, but as far as I can say, the file
descriptor is properly closed in TestRunner.py, so this's
probably not correct. Other theory is that, on Windows, file
output is not guaranteed to be visible to other processes even
if a process flushes file descriptors, but I'd think that's
unlikely. So honestly I don't know the cause yet.
llvm-svn: 246621
This patch fixes a subtle incompatibility with MSVC linker.
MSVC linker preserves the original spelling of a DLL in the
import descriptor table. LLD previously converted all
characters to lowercase. Usually this difference is benign,
but if a program explicitly checks for DLL file names, the
program could fail.
llvm-svn: 246620
In r246424, I made a change that disables non-DLL to export
symbols. It turned out that the change was not correct. Both
DLLs and executables are able to export symbols (although the
latter is relatively rare). This change restores the feature.
llvm-svn: 246537
I have totally no idea why, but MSVC linker is sensitive about
file names of archive members. If we do not make import library
file names to the same as the DLL name, MSVC link *crashes*
when it is processing the library file. This patch is to set
the same name.
llvm-svn: 246535
The rules for dllexported symbols are overly complicated due to
x86 name decoration, fuzzy symbol resolution, and the fact that
one symbol can be resolved by so many different names. The rules
are probably intended to be "intuitive", so that users don't have
to understand the name mangling schemes, but it seems that it can
lead to unintended symbol exports.
To make it clear what I'm trying to do with this patch, let me
write how the export rules are subtle and complicated.
- x86 name decoration: If machine type is i386 and export name
is given by a command line option, like /export:foo, the
real symbol name the linker has to search for is _foo because
all symbols are decorated with "_" prefixes. This doesn't happen
on non-x86 machines. This automatic name decoration happens only
when the name is not C++ mangled.
However, the symbol name exported from DLLs are ones without "_"
on all platforms.
Moreover, if the option is given via .drectve section, no
symbol decoration is done (the reason being that the .drectve
section is created by a compiler and the compiler should always
know the exact name of the symbol, I guess).
- Fuzzy symbol resolution: In addition to x86 name decoration,
the linker has to look for cdecl or C++ mangled symbols
for a given /export. For example, it searches for not only
_foo but also _foo@<number> or ??foo@... for /export:foo.
Previous implementation didn't get it right. I'm trying to make
it as compatible with MSVC linker as possible with this patch
however the rules are. The new code looks a bit messy to me, but
I don't think it can be simpler due to the ad-hoc-ness of the rules.
llvm-svn: 246424
This is exposed via a new flag /opt:lldltojobs=N, where N is the number of
code generation threads.
Differential Revision: http://reviews.llvm.org/D12309
llvm-svn: 246342
ICF is a feature to merge sections not by name (which is the regular
COMDAT merging) but by contents. If two or more sections have the
identical contents and relocations, ICF merges them to save space.
Accessors or templated functions tend to have the same contents, and
ICF can hold them.
If we consider sections as vertices and relocations as edges, the
problem is to find as many isomorphic graphs as possile from input
graphs. MSVC linker is smart enough to identify isomorphic graphs
even if they contain circles (GNU gold cannot handle circles
according to http://research.google.com/pubs/pub36912.html, so this
is impressive).
Circular references are not uncommon in COFF object files.
One example is .pdata. .pdata sections contain exception handler info
for functions, so they naturally have relocations for the functions.
The functions in turn have references to the .pdata sections so that
the functions and their .pdata are linked together. As a result, they
form circles.
This is a test case for circular graphs. LLD is not able to handle
this test case yet. I'll add code soon.
llvm-svn: 245827
The old test files were just compiler outputs, so it was hard to
debug if something goes wrong. The new test file is carefully
hand-crafted to trigger ICF to avoid that.
llvm-svn: 245826
Previously, weak external symbols could reference only symbols that
appeared before them. Although that covers almost all use cases
of weak externals, there are object files out there which contains
weak externals that have forward references.
This patch supports such weak externals.
llvm-svn: 245258
There are some DLLs whose initializers depends on other DLLs'
initializers. The initialization order matters for them.
MSVC linker uses the order of the libraries from the command line.
LLD used ASCII-betical order. So they were incompatible.
This patch makes LLD compatible with MSVC.
llvm-svn: 245201
Sections must start at page boundaries in memory, but they
can be aligned to sector boundaries (512-bytes) on disk.
We aligned them to 4096-byte boundaries even on disk, so we
wasted disk space a bit.
llvm-svn: 244691
We were printing an error but exiting with 0.
Not sure how to test this. We could add a no-winlib feature,
but that is probably not worth it.
llvm-svn: 244109
Right now PE image section addresses are RVAs and symbol addresses are
VAs. We should probably fix this by changing section addresses to match
symbol addresses. Fixing this might take a few hours, so temporarily
disable the objdump part of this test.
llvm-svn: 243758
We want to convince the NT loader not to map these sections into memory.
A good first step is to move them to the end of the executable.
Differential Revision: http://reviews.llvm.org/D11655
llvm-svn: 243680
Windows ARM is the thumb ARM environment, and pointers to thumb code
needs to have its LSB set. When we apply relocations, we need to
adjust the LSB if it points to an executable section.
llvm-svn: 243560
SECREL should sets the 32-bit offset of the target from the beginning
of *target's* output section. Previously, the offset from the beginning
of source's output section was used instead.
SECTION means the target section's index, and not the source section's
index. This patch fixes that issue too.
llvm-svn: 243535
I don't fully understand the rationale behind the name mangling
scheme used for the DLL export table and the import library.
Why only leading "_" is dropped for the import library while
both "_" and "@" are dropped from DLL symbol table? But this seems
to be what MSVC linker does.
llvm-svn: 243490
Previously, we ignore /merge option if /debug is specified
because I thought that was MSVC linker did. This was wrong.
/merge shouldn't be ignored even in debug mode.
llvm-svn: 243375
An object file compatible with Safe SEH contains a .sxdata section.
The section contains a list of symbol table indices, each of which
is an exception handler function. A safe SEH-enabled executable
contains a list of exception handler RVAs. So, what the linker has
to do to support Safe SEH is basically to read the .sxdata section,
interpret the contents as a list of symbol indices, unique-fy and
sort their RVAs, and then emit that list to .rdata. This patch
implements that feature.
llvm-svn: 243182
__ImageBase is a special symbol whose value is the image base address.
Previously, we handled __ImageBase symbol as an absolute symbol.
Absolute symbols point to specific locations in memory and the locations
never change even if an image is base-relocated. That means that we
don't have base relocation entries for absolute symbols.
This is not a case for __ImageBase. If an image is base-relocated, its
base address changes, and __ImageBase needs to be shifted as well.
So we have to have base relocations for __ImageBase. That means that
__ImageBase is not really an absolute symbol but a different kind of
symbol.
In this patch, I introduced a new type of symbol -- DefinedRelative.
DefinedRelative is similar to DefinedAbsolute, but it has not a VA but RVA
and is a subject of base relocation. Currently only __ImageBase is of
the new symbol type.
llvm-svn: 243176
Load Configuration field points to a structure containing information
for SEH. That data strucutre is not created by the linker but provided
by an external file. What we have to do is just to set __load_config_used
address to the header.
llvm-svn: 242427
If a symbol is exported as /export:foo, and foo is resolved as a
mangled name (_foo@<number> or ?foo@@Y...), that mangled name should
be written to the export table. Previously, we wrote the original
name to the export table.
llvm-svn: 242342
Because thunks for dllimported symbols contain absolute addresses on x86,
they need to be relocated at load-time. This bug was a cause of crashes
in DLL initialization routines.
llvm-svn: 242259
Entry name selection rule is already complicated on x64, but it's more
complicated on x86 because of the underscore name mangling scheme.
If one of _main, _main@<number> (a C function) or ?main@@... (a C++ function)
is defined, entry name is _mainCRTStartup. If _wmain, _wmain@<number or
?wmain@@... is defined, entry name is _wmainCRTStartup. And so on.
llvm-svn: 242110
If /delayload option is given, we have to resolve __delayLoadHelper2
since the function is the dynamic loader to delay-load DLLs.
The function name is mangled in x86 as ___delayLoadHelper2@8.
llvm-svn: 242078
Symbol foo is mangled as _foo in C and ?foo@@... in C++ on x86.
findMangle has to remove prefix underscore before mangle a given name
as a C++ symbol.
llvm-svn: 241874
Symbol names are usually mangled by appending "_" prefix on x86.
But the mangled name is not used in DLL export table. The export
table contains unmangled names.
llvm-svn: 241872
With this patch, LLD is now able to self-link an .exe file for x86
that runs correctly, although I don't think some headers (particularly
SEH) are not correct. DLL support is coming soon.
llvm-svn: 241857
Previously, we infer machine type at the very end of linking after
all symbols are resolved. That's actually too late because machine
type affects how we mangle symbols (whether or not we need to
add "_").
For example, /entry:foo adds "_foo" to the symbol table if x86 but
"foo" if x64.
This patch moves the code to infer machine type, so that machine
type is inferred based on input files given via the command line
(but not based on .directives files).
llvm-svn: 241843
Symbols exported by DLLs are listed in import library files.
Exported names may be mangled by "Import Name Type" field as
described in PE/COFF spec 7.3. This patch implements that
mangling scheme.
llvm-svn: 241719
Providing a symbol table in the executable is quite useful when
debugging a fully-linked executable without having to reconstruct one
from DWARF.
Differential Revision: http://reviews.llvm.org/D11023
llvm-svn: 241689
Previously we were unnecessarily loading lazy symbols if they appeared in an
archive multiple times, as can happen with comdat symbols. This change fixes
the bug by only loading symbols from archives at load time if the original
symbol was undefined.
Differential Revision: http://reviews.llvm.org/D10980
llvm-svn: 241538
TLS table header field is supposed to have address and size of TLS table.
The linker doesn't have to understand what TLS table is. TLS table's name
is always "_tls_used", so if there's that symbol, the linker simply sets
that symbol's RVA to the header. The size of the TLS table is always 40 bytes.
llvm-svn: 241426
We were previously hitting assertion failures in the writer in cases where
a regular object file defined a weak external symbol that was defined by
a bitcode file. Because /export and /entry name mangling were implemented
using weak externals, the same problem affected mangled symbol names in
bitcode files.
The underlying cause of the problem was that weak external symbols were
being resolved before doing LTO, so the symbol table may have contained stale
references to bitcode symbols. The fix here is to defer weak external symbol
resolution until after LTO.
Also implement support for weak external symbols in bitcode files
by modelling them as replaceable DefinedBitcode symbols.
Differential Revision: http://reviews.llvm.org/D10940
llvm-svn: 241391
This worked before, but only by accident, and only with assertions disabled.
We ended up storing a DefinedRegular symbol in the WeakAlias field,
and never using it as an Undefined.
Differential Revision: http://reviews.llvm.org/D10934
llvm-svn: 241376
DLLs can export symbols only by ordinal, and DLLs are also able to be
delay-loaded. The combination of the two is valid. I didn't expect
that combination. This patch implements that feature.
With this patch, LLD is now able to link a working executable of Chrome
for 64-bit debug build. The browser seemed to be working fine. Chrome is
good for testing because of its variety and size. It contains various
open-source libraries written by various people. The largest file in
Chrome is chrome.dll whose size is 496MB. LLD can link it in 24 seconds.
MSVC linker takes 48 seconds. So it is exactly 2x faster. (I measured
that with debug info and ICF being turned off.)
With this achievement, I think I can say that the new COFF linker is
now mostly feature complete for x86-64 Windows. I believe there are
still many lingering bugs, though.
llvm-svn: 241318
Previously, __ImageBase symbol got a different value than the one
specified by /base:<number> because the symbol was created in the
SymbolTable's constructor. When the constructor is called,
no command line options are processed yet, so the symbol was
created always with the initial value. This caused wrong relocations
and thus caused mysterious crashes of some executables linked by LLD.
llvm-svn: 241313
Previously, pointers pointed by locally-imported symbols were broken.
It has only 4 bytes although the correct size is 8 byte. This patch
fixes that bug.
llvm-svn: 241295
On Windows, we have four different main functions, {w,}{main,WinMain}.
The linker has to choose a corresponding entry point function among
{w,}{main,WinMain}CRTStartup. These entry point functions are defined
in the standard library. The linker resolves one of them by looking at
which main function is defined and adding a corresponding undefined
symbol to the symbol table.
Object files containing entry point functions conflicts each other.
For example, we cannot resolve both mainCRTStartup and WinMainCRTStartup
because other symbols defined in the files conflict.
Previously, we inferred CRT function name at the very end of name
resolution. I found that that is sometimes too late. If the linker
already linked one of these four archive member objects, it's too late
to change the decision.
The right thing to do here is to infer entry point name after adding
all symbols from command line files and before adding any other files
(which are specified by directive sections). This patch does that.
llvm-svn: 241236
Previously, the order of adding symbols to the symbol table was simple.
We have a list of all input files. We read each file from beginning of
the list and add all symbols in it to the symbol table.
This patch changes that order. Now all archive files are added to the
symbol table first, and then all the other object files are added.
This shouldn't change the behavior in single-threading, and make room
to parallelize in multi-threading.
In the first step, only lazy symbols are added to the symbol table
because archives contain only Lazy symbols. Member object files
found to be necessary are queued. In the second step, defined and
undefined symbols are added from object files. Adding an undefined
symbol to the symbol table may cause more member files to be added
to the queue. We simply continue reading all object files until the
queue is empty.
Finally, new archive or object files may be added to the queues by
object files' directive sections (which contain new command line
options).
The above process is repeated until we get no new files.
Symbols defined both in object files and in archives can make results
undeterministic. If an archive is read before an object, a new member
file gets linked, while in the other way, no new file would be added.
That is the most popular cause of an undeterministic result or linking
failure as I observed. Separating phases of adding lazy symbols and
undefined symbols makes that deterministic. Adding symbols in each
phase should be parallelizable.
llvm-svn: 241107
Compilers recognize "main" function and don't mangle its name.
But if you use a different function as a user-defined entry name,
and if you didn't define that function with extern C, your entry
point function name is mangled. And the linker has to be able to
find that. This is relatively rare but can happen.
llvm-svn: 240953
The previous logic to find default entry name or subsystem does not
seem correct (i.e. was not compatible with MSVC linker). Previously,
default entry name was inferred from CRT functions and user-defined
entry functions. Subsystem was inferred from CRT functions.
Default entry name and subsystem are now inferred based on the
following table. Note that we no longer use CRT functions to infer
them.
Entry name Subsystem
main mainCRTStartup console
wmain wmainCRTStartup console
WinMain WinMainCRTStartup windows
wWinMain wWinMainCRTStartup windows
llvm-svn: 240922
Usually dllexported symbols are defined with 'extern "C"',
so identifying them is easy. We can just do hash table lookup
to look up exported symbols.
However, C++ non-member functions are also allowed to be exported,
and they can be specified with unmangled name. So, if /export:foo
is given, we need to look up not only "foo" but also its all
mangled names. In MSVC mangling scheme, that means that we need to
look up any symbol which starts with "?foo@@Y".
In this patch, we scan the entire symbol table to search for
a mangled symbol. The symbol table is a DenseMap, and that doesn't
support table lookup by string prefix. This is of course very
inefficient. But that should be probably OK because the user
should always add 'extern "C"' to dllexported symbols.
llvm-svn: 240919
This option is to ignore remaining undefined symbols and force
the linker to create an output file anyways.
The existing code assumes that there's no undefined symbol after
reportRemainingUndefines(). That assumption is legitimate.
I also don't want to mess up the existing code for this minor feature.
In order to keep it as is, remaining undefined symbols are replaced
with dummy defined symbols.
llvm-svn: 240913
When comparing two COMDAT sections, we need to take section values
and associative sections into account. This patch fixes that bug.
It fixes a crash bug of llvm-tblgen when linked with /opt:lldicf.
One thing I don't understand yet is that this logic seems to be
too strict. MSVC linker is able to create more compact executables
(which of course work correctly). With this ICF algorithm, LLD is
able to make executable smaller, but the outputs are larger than
MSVC's. There must be something I'm missing here.
llvm-svn: 240897
There were a few issues with the previous delay-import tables.
- "Attribute" field should have been 1 instead of 0.
(I don't know the meaning of this field, though.)
- LEA and CALL operands had wrong addresses.
- Address tables are in .didat (which is read-only).
They should have been in .data.
llvm-svn: 240837
This flag can be used to produce a map file, which is essentially a list
of objects linked into the final output file together with the RVAs of
their symbols. Because our format differs from MSVC's we expose it as a
separate flag.
Differential Revision: http://reviews.llvm.org/D10773
llvm-svn: 240812
We were resolving entry symbols and /include'd symbols after all other
symbols are resolved. But looks like it's too late. I found that it
causes some program to fail to link.
Let's say we have an object file A which defines symbols X and Y in an
archive. We also have another file B after A which defines X, Y and
_DLLMainCRTStartup in another archive. They conflict each other, so
either A or B can be linked.
If we have _DLLMainCRTStartup as an undefined symbol, file B is always
chosen. If not, there's a chance that A is chosen. If the linker
find it needs _DllMainCRTStartup after that, it's too late.
This patch adds undefined symbols to the symbol table as soon as
possible to fix the issue.
llvm-svn: 240757
Absolute symbols were always handled as external symbols, so if two
or more object files define the same absolute symbol, they would
conflict even if the symbol is private to each file.
This patch fixes that bug.
llvm-svn: 240756
ICF implemented in LLD is so experimental that we don't want to
enable that even if /opt:icf option is passed. I'll rename it back
once the feature is complete.
llvm-svn: 240721
The change I made in r240620 was not correct. If a symbol foo is
defined, and if you use __imp_foo, __imp_foo symbol is automatically
defined as a pointer (not just an alias) to foo.
Now that we need to create a chunk for automatically-created symbols.
I defined LocalImportChunk class for them.
llvm-svn: 240622
MSVC linker is able to link an object file created from the following code.
Note that __imp_hello is not defined anywhere.
void hello() { printf("Hello\n"); }
extern void (*__imp_hello)();
int main() { __imp_hello(); }
Function symbols exported from DLLs are automatically mangled by appending
__imp_ prefix, so they have two names (original one and with the prefix).
This "feature" seems to simulate that behavior even for non-DLL symbols.
This is in my opnion very odd feature. Even MSVC linker warns if you use this.
I'm adding that anyway for the sake of compatibiltiy.
llvm-svn: 240620
Identical COMDAT Folding (ICF) is an optimization to reduce binary
size by merging COMDAT sections that contain the same metadata,
actual data and relocations. MSVC link.exe and many other linkers
have this feature. LLD achieves on per with MSVC in terms produced
binary size with this patch.
This technique is pretty effective. For example, LLD's size is
reduced from 64MB to 54MB by enaling this optimization.
The algorithm implemented in this patch is extremely inefficient.
It puts all COMDAT sections into a set to identify duplicates.
Time to self-link with/without ICF are 3.3 and 320 seconds,
respectively. So this option roughly makes LLD 100x slower.
But it's okay as I wanted to achieve correctness first.
LLD is still able to link itself with this optimization.
I'm going to make it more efficient in followup patches.
Note that this optimization is *not* entirely safe. C/C++ require
different functions have different addresses. If your program
relies on that property, your program wouldn't work with ICF.
However, it's not going to be an issue on Windows because MSVC
link.exe turns ICF on by default. As long as your program works
with default settings (or not passing /opt:noicf), your program
would work with LLD too.
llvm-svn: 240519
Previously, we added files in directive sections to the symbol
table as we read the sections, so the link order was depth-first.
That's not compatible with MSVC link.exe nor the old LLD.
This patch is to queue files so that new files are added to the
end of the queue and processed last. Now addFile() doesn't parse
files nor resolve symbols. You need to call run() to process
queued files.
llvm-svn: 240483
DLLs are usually resolved at process startup, but you can
delay-load them by passing /delayload option to the linker.
If a /delayload is specified, the linker has to create data
which is similar to regular import table.
One notable difference is that the pointers in a delay-load
import table are originally pointing to thunks that resolves
themselves. Each thunk loads a DLL, resolve its name, and then
overwrites the pointer with the result so that subsequent
function calls directly call a desired function. The linker
has to emit thunks.
llvm-svn: 240250
.pdata section contains a list of triplets of function start address,
function end address and its unwind information. Linkers have to
sort section contents by function start address and set the section
address to the file header (so that runtime is able to find it and
do binary search.)
This change seems to resolve all but one remaining test failures in
check{,-clang,-lld} when building the entire stuff with clang-cl and
lld-link.
llvm-svn: 240231
This is a case that one mistake caused a very mysterious bug.
I made a mistake to calculate addresses of common symbols, so
each common symbol pointed not to the beginning of its location
but to the end of its location. (Ouch!)
Common symbols are aligned on 16 byte boundaries. If a common
symbol is small enough to fit between the end of its real
location and whatever comes next, this bug didn't cause any harm.
However, if a common symbol is larger than that, its memory
naturally overlapped with other symbols. That means some
uninitialized variables accidentally shared memory. Because
totally unrelated memory writes mutated other varaibles, it was
hard to debug.
It's surprising that LLD was able to link itself and all LLD
tests except gunit tests passed with this nasty bug.
With this fix, the new COFF linker is able to pass all tests
for LLVM, Clang and LLD if I use MSVC cl.exe as a compiler.
Only three tests are failing when used with clang-cl.
llvm-svn: 240216
In this linker model, adding an undefined symbol may trigger chain
reactions. It may trigger a Lazy symbol to read a new file.
A new file may contain a directive section, which may contain various
command line options.
Previously, we didn't handle chain reactions well. We visited /include'd
symbols only once, so newly-added /include symbols were ignored.
This patch fixes that bug.
Now, the symbol table is versioned; every time the symbol table is
updated, the version number is incremented. We repeat adding undefined
symbols until the version number does not change. It is guaranteed to
converge -- the number of undefined symbol in the system is finite,
and adding the same undefined symbol more than once is basically no-op.
llvm-svn: 240177
Alternatename option is in the form of /alternatename:<from>=<to>.
It's effect is to resolve <from> as <to> if <from> is still undefined
at end of name resolution.
If <from> is not undefined but completely a new symbol, alternatename
shouldn't do anything. Previously, it introduced a new undefined
symbol for <from>, which resulted in undefined symbol error.
llvm-svn: 240161
We don't want to insert a new symbol to the symbol table while reading
a .drectve section because it's going to be too complicated.
That we are reading a directive section means that we are currently
reading some object file. Adding a new undefined symbol to the symbol
table can trigger a library file to read a new file, so it would make
the call stack too deep.
In this patch, I add new symbol names to a list to resolve them later.
llvm-svn: 240076
Alternatename option is in the form of /alternatename:<from>=<to>.
It is an error if there are two options having the same <from> but
different <to>. It is *not* an error if both are the same.
llvm-svn: 240075
We skip unknown options in the command line with a warning message
being printed out, but we shouldn't do that for .drectve section.
The section is not visible to the user. We should handle unknown
options as an error.
llvm-svn: 240067
The linker has to create an XML file for each executable.
This patch supports that feature.
You can optionally embed an XML file to an executable as .rsrc
section. If you choose to do that (by passing /manifest:embed
option), the linker has to create a textual resource file
containing an XML file, compile that using rc.exe to a binary
resource file, conver that resource file to a COFF file using
cvtres.exe, and then link that COFF file. This patch implements
that feature too.
llvm-svn: 239978
On Windows, we have to create a .lib file for each .dll.
When linking against DLLs, the linker doesn't use the DLL files,
but instead read a list of dllexported symbols from corresponding
lib files.
A library file containing descriptors of a DLL is called an
import library file.
lib.exe has a feature to create an import library file from a
module-definition file. In this patch, we create a module-definition
file and pass that to lib.exe.
We eventually want to create an import library file by ourselves
to eliminate dependency to lib.exe. For now, we just use the MSVC
tool.
llvm-svn: 239937
Module-definition files (.def files) are yet another way to
specify parameters to the linker. You can write a list of dllexported
symbols in module-definition files instead of using /export command
line option. It also supports a few more directives.
The parser code is taken from lib/Driver/WinLinkModuleDef.cpp
with the following modifications.
- variable names are updated to comply with the LLVM coding style.
- Instead of returning parsing results as "directive" objects,
it updates Config object directly.
llvm-svn: 239929
DLL files are in the same format as executables but they have export tables.
The format of the export table is described in PE/COFF spec section 5.3.
A new class, EdataContents, takes care of creating chunks for export tables.
What we need to do is to parse command line flags for dllexports, and then
instantiate the class to create chunks. For the writer, export table chunks
are opaque data -- it just add chunks to .edata section.
llvm-svn: 239869
PE/COFF executables/DLLs usually contain data which is called
base relocations. Base relocations are a list of addresses that
need to be fixed by the loader if load-time relocation is needed.
Base relocations are in .reloc section.
We emit one base relocation entry for each IMAGE_REL_AMD64_ADDR64
relocation.
In order to save disk space, base relocations are grouped by page.
Each group is called a block. A block starts with a 32-bit page
address followed by 16-bit offsets in the page. That is more
efficient representation of addresses than just an array of 32-bit
addresses.
llvm-svn: 239710
Resource files are data files containing i18n messages, icon images, etc.
MSVC has a tool to convert a resource file to a regular COFF file so that
you can just link that file to embed resources to an executable.
However, you can directly pass resource files to the linker. If you do that,
the linker invokes the tool automatically. This patch implements that feature.
llvm-svn: 239704
In the case where either a bitcode file and a regular file or two bitcode
files export a common or comdat symbol with the same name, the linker needs
to pick one of them following COFF semantics. This patch implements a design
for resolving such symbols that pushes most of the work onto either LLD's
regular mechanism for resolving common or comdat symbols or the IR linker's
mechanism for doing the same.
We modify SymbolBody::compare to always prefer non-bitcode symbols, so that
during the initial phase of symbol resolution, the symbol table always contains
a regular symbol in any case where we need to choose between a regular and
a bitcode symbol. In SymbolTable::addCombinedLTOObject, we force export
any bitcode symbols that were initially pre-empted by a regular symbol,
and later use SymbolBody::compare to choose between the regular symbol in
the symbol table and the regular symbol from the combined LTO object file.
This design seems to be sound, so long as the resolution mechanism is defined
to be commutative and associative modulo arbitrary choices between symbols
(which seems to be the case for COFF).
Differential Revision: http://reviews.llvm.org/D10329
llvm-svn: 239563
The code generator may create references to runtime library symbols such as
__chkstk which were not visible via LTOModule. Handle these cases by loading
the object file from the library, but abort if we end up having loaded any
bitcode objects.
Because loading the object file may have introduced new undefined references,
call reportRemainingUndefines again to detect and report them.
Differential Revision: http://reviews.llvm.org/D10332
llvm-svn: 239386
The LLVM code generator can sometimes synthesize symbols, such as SSE
constants, that are not visible via the LTOModule interface. Allow such
symbols so long as they have definitions.
Differential Revision: http://reviews.llvm.org/D10331
llvm-svn: 239385
We forgot to check for auxiliary symbol's type. So we sometimes read
garbage as associative section definitions.
Associative sections are considered as not live themselves by the
garbage collector because they are live only when associaited sections
are live.
By reading more data (or garbage) as associative section definitions,
we treated more sections as non-GC-roots, that caused the linker to
discard too many sections by mistake. That caused another mysterious
bug (such as some global constructors don't run at all for some reason.)
llvm-svn: 239287
Not only entry point symbol but also symbols specified by /include
option must be preserved, as they will never be dead-stripped.
http://reviews.llvm.org/D10220
llvm-svn: 239005
Symbols exported by DLLs can be imported not by name but by
small number or ordinal. Usually, symbols have both ordinals
and names, and in that case ordinals are called "hints" and
used by the loader as hints.
However, symbols can have only ordinals. They are called
import-by-ordinal symbols. You need to manage ordinals by hand
so that they will never change if you choose to use the feature.
But it's supposed to make dynamic linking faster because
it needs no string comparison. Not sure if that claim still
stands in year 2015, though. Anyways, the feature exists,
and this patch implements that.
llvm-svn: 238780
Previously, this feature was implemented using a special type of
undefined symbol, in addition to an intricate way to make the resolver
read a virtual file containing that renaming symbols.
Now the feature is directly handled by the symbol table.
The symbol table has a function, rename(), to rename symbols, whose
definition is 4 lines long. Symbol renaming is naturally modeled using
Symbol and SymbolBody.
llvm-svn: 238696
It does not involve notions of virtual archives or virtual files,
nor store a list of undefined symbols somewhere else to consume them later.
We did that before. In this patch, undefined symbols are just added to
the symbol table, which now can be done in very few lines of code.
llvm-svn: 238681
`main` is not the only main function in Windows. You can choose one
from these four -- {w,}{WinMain,main}. There are four different entry
point functions for them, {w,}{WinMain,main}CRTStartup, respectively.
The linker needs to choose the right one depending on which `main`
function is defined.
llvm-svn: 238667
Section names were truncated to 8 bytes because the section table's
name field is 8 byte long. This patch creates the string table to
store long names.
llvm-svn: 238661
The new mechanism is less code, and fixes the case where all inputs
are archives.
Differential Revision: http://reviews.llvm.org/D10136
llvm-svn: 238618
Currently we set the field to zero, but as per the spec, we should
set numbers we read from import library files. The loader uses the
values as starting offsets for binary search when looking up imported
symbols from DLL.
llvm-svn: 238562
This is an initial patch for a section-based COFF linker.
The patch has 2300 lines of code including comments and blank lines.
Before diving into details, you want to start from reading README
because it should give you an overview of the design.
All important things are written in the README file, so I write
summary here.
- The linker is already able to self-link on Windows.
- It's significantly faster than the existing implementation.
The existing one takes 5 seconds to link LLD on my machine,
while the new one only takes 1.2 seconds, even though the new
one is not multi-threaded yet. (And a proof-of-concept multi-
threaded version was able to link it in 0.5 seconds.)
- It uses much less memory (250MB vs. 2GB virtual memory space
to self-host).
- IMHO the new code is much simpler and easier to read than
the existing PE/COFF port.
http://reviews.llvm.org/D10036
llvm-svn: 238458
Bob Haarman