When LLD links against an import library (for the regular, short import
libraries), it doesn't actually link in the header/trailer object files
at all, but synthesizes new corresponding data structures into the right
sections.
If the whole of such an import library is forced to be linked, e.g. with
the -wholearchive: option, we actually end up linking in those
header/trailer objects. The header objects contain a construct which LLD
fails to handle; previously we'd error out with the error ".idata$4
should not refer to special section 0".
Within the import library header object, in the import directory we have
relocations towards the IAT (.idata$4 and .idata$5), but the header
object itself doesn't contain any data for those sections.
In the case of GNU generated import libraries, the header objects
contain zero length sections .idata$4 and .idata$5, with relocations
against them. However in the case of LLVM generated import libraries,
the sections .idata$4 and .idata$5 are not included in the list of
sections. The symbol table does contain section symbols for these
sections, but without any actual associated section. This can probably
be seen as a declaration of an empty section.
If the header/trailer objects of a short import library are linked
forcibly and we also reference other functions in the library, we end up
with two import directory entries for this DLL, one that gets
synthesized by LLD, and one from the actual header object file. This is
inelegant, but should be acceptable.
While it would seem unusual to link import libraries with the
-wholearchive: option, this can happen in certain scenarios.
Rust builds libraries that contain relevant import libraries bundled
along with compiled Rust code as regular object files, all within one
single archive. Such an archive can then end up linked with the
-wholarchive: option, if build systems decide to use such an option for
including static libraries.
This should fix https://github.com/msys2/MINGW-packages/issues/21017.
This works for the header/trailer object files in import libraries
generated by LLVM; import libraries generated by MSVC are vaguely
different. ecb5ea6a266d5cc4e05252f6db4c73613b73cc3b did an attempt at
fixing the issue for MSVC generated libraries, but it's not entirely
correct, and isn't enough for making things work for that case.
This refactor prepares for further ARM64X hybrid support, where these
helpers will need to work with either the native or EC symbol table
based on context.
Move `LinkerDriver::addUndefined` to` SymbolTable` to allow its use with
both symbol tables on ARM64X and rename it to `addGCRoot` to clarify its
distinct role compared to the existing `SymbolTable::addUndefined`.
Command-line `-include` arguments now apply to the EC symbol table, with
`mainSymtab` introduced in `linkerMain`. There will be more similar
cases. For `.drectve` sections, the corresponding symbol table is used
based on the context.
This change ensures the load config in the hybrid image view is handled
correctly. It introduces a new Arm64XRelocVal class to abstract
relocation values, allowing them to be relative to a symbol. This class
will also be useful for managing ARM64X relocation offsets in the
future.
This change ensures that base relocations are sorted in the output,
aligning with MSVC linker behavior. While input files typically provide
sorted relocations, this update guarantees correct sorting even if the
input relocations are unordered.
cg_profile in object is from CGProfilePass and it is often inaccurate.
While call-graph-ordering-file is provided by user. It is weird to
aggregate them together especially when call-graph-ordering-file is
accurate enough.
This already worked without /wholearchive; now it works with it too.
(Only for thin archives containing relative file names, matching the ELF
and Mach-O ports.)
The addFile implementation does not rely on the SymbolTable object. With
#119294, the symbol table for input files is determined during the
construction of the objects representing them. To clarify that
relationship, this change moves the implementation from the SymbolTable
class to the LinkerDriver class.
That change forgot to set `lazy` to false before calling `addFile()` in
`forceLazy()` which caused `addFile()` to parse the file we want to
force a load for to be added as a lazy object again instead of adding
the file to `ctx.objFileInstances`.
This is caught by a pretty simple test (included).
If foo.obj is eagerly loaded (due to a prior undef referencing one if
its symbols) and has more than one symbol, we used to assert:
SymbolTable::addLazyObject() for the first symbol would set `lazy` to
false and load all symbols from the file, but the outer
ObjFile::parseLazy() loop would continue to run and call addLazyObject()
for the second symbol, which would assert.
Instead, just stop adding lazy symbols if the file got loaded for real
while adding a symbol.
(The ELF port has a similar early exit in `ObjFile<ELFT>::parseLazy()`.)
`symtab.ctx.symtab` is just `symtab`. Looks like #119296 added
this using a global find-and-replace.
This was the only instance of `symtab.ctx.symtab` in lld/.
No behavior change.
Previously, we'd collect all input files in Driver::filePaths, and then
write filePaths after all other flags in
createResponseFile(). This meant that `-start-lib foo.obj -end-lib`
would be written as `-start-lib -end-lib foo.obj`, changing semantics.
Instead, remove Driver::filePaths, and handle things that fed into it
directly:
* OPT_INPUT is now handled in the same way as other flags, so that we
now get `-start-lib foo.obj -end-lib` in response.txt as desired. Add a
test for -start-lib / -end-lib and /reproduce:.
* OPT_wholearchive_file needs explicit handling now -- but before, this
was buggy as well: We'd put the flag without a rewritten path in
response.txt, but also the rewritten input file without wholearchive
semantics via filePaths. So this commit makes --whole-archive work with
/reproduce: too, and adds test coverage.
* /defaultlib:foo is now written as /defaultlib:foo into response.txt,
instead of writing the resolved path previously. While response.txt
looks slightly differently, both should have the same semantics, and
this should be mostly a no-op. (It does require updating a test.)
* /defaultlib: from .drectve sections are no longer recorded in
response.txt. This seems like a progression -- in the non-repro case
they come from .obj files, so they should come (only) from there in the
repro case too. This adds test coverage for this case.
Makes createResponseFile() look more like the versions in the ELF and
MachO ports too.
On hybrid ARM64X targets, ARM64 and ARM64EC input files operate in
separate namespaces and cannot reference each other. This change
introduces separate `SymbolTable` instances and associates each
`InputFile` with the appropriate table to reflect this behavior.
This change prepares for hybrid ARM64X support, which requires two
`SymbolTable` instances: one for native symbols and one for EC symbols.
In such cases, `config.machine` will remain ARM64X, while the
`SymbolTable` instances will store ARM64 and ARM64EC machine types.
This change prepares for the introduction of separate hybrid namespaces.
Hybrid images will require two `SymbolTable` instances, making it
necessary to associate `InputFile` objects with the relevant one.
Apologies for the large change, I looked for ways to break this up and
all of the ones I saw added real complexity. This change focuses on the
option's prefixed names and the array of prefixes. These are present in
every option and the dominant source of dynamic relocations for PIE or
PIC users of LLVM and Clang tooling. In some cases, 100s or 1000s of
them for the Clang driver which has a huge number of options.
This PR addresses this by building a string table and a prefixes table
that can be referenced with indices rather than pointers that require
dynamic relocations. This removes almost 7k dynmaic relocations from the
`clang` binary, roughly 8% of the remaining dynmaic relocations outside
of vtables. For busy-boxing use cases where many different option tables
are linked into the same binary, the savings add up a bit more.
The string table is a straightforward mechanism, but the prefixes
required some subtlety. They are encoded in a Pascal-string fashion with
a size followed by a sequence of offsets. This works relatively well for
the small realistic prefixes arrays in use.
Lots of code has to change in order to land this though: both all the
option library code has to be updated to use the string table and
prefixes table, and all the users of the options library have to be
updated to correctly instantiate the objects.
Some follow-up patches in the works to provide an abstraction for this
style of code, and to start using the same technique for some of the
other strings here now that the infrastructure is in place.
This modifies the machine field in the hybrid view to be AMD64, aligning
it with expectations from ARM64EC modules. While this provides initial
support, additional relocations will be necessary for full
functionality. Many of these cases depend on implementing separate
namespace support first.
Move clearing of the .reloc section from addBaserels to assignAddresses
to ensure it is always cleared, regardless of the relocatable
configuration. This change also clarifies the reasoning for adding the
dynamic relocations chunk in that location.
Similar to #112319 for ELF. While there is some initial boilerplate, it
can simplify some call sites that use Twine, especially when a printed
element uses `ctx` or toString.
