These references are used to implement MachO/x64-64 subtractor relocations
where the minuend is being fixed up, rather than the subtrahend. The 64-bit
version was not previously supported, the 32-bit version was partially
implemented but contained bugs not caught by existing test cases. This
patch fixes both functionality and test coverage.
llvm-svn: 273759
This reverts commit r264945.
The commit only removed an unreachable in a method with a covered switch, but
GCC is likely to warn on this, and the coding standards recommend just leaving
in the unreachable.
llvm-svn: 264983
make_dynamic_error_code was used to create a std::error_code with
a std::string message. Now that we are migrating to llvm::Error,
there are no calls to these make_dynamic_error_code methods.
There is one single call to make_dynamic_error_code remaining, the one
inside GenericError::convertToErrorCode(). That method is only called
from File::doParse() which should be a temporary situation. We need
to work out how to deal with File::parse() caching the error result from
doParse(). Caching errors isn't supported in the new scheme, and probably
isn't needed here, but we need to work that out.
Once thats done, dynamic error and all utilities around it can be deleted.
llvm-svn: 264982
These methods weren't really throwing errors. The only error used
was that a file could not be found, which isn't really an error at all
as we are searching paths and libraries for a file. All of the callers
also ignored errors and just used the returned path if one was available.
Changing to return Optional<StringRef> as that actually reflects what
we are trying to do here: optionally find a given path.
llvm-svn: 264979
These methods were responsible for some of the few remaining calls
to llvm::errorCodeToError. Converting them makes us have more Error's
in the api and fewer error_code's.
llvm-svn: 264974
Pretty mechanical change here. Just replacing all the std::error_code() with
llvm::Error() and make_dynamic_error_code with make_error<GenericError>
llvm-svn: 264917
Adds a GenericError class to lld/Core which can carry a string. This is
analygous to the dynamic_error we currently use in lld/Core.
Use this GenericError instead of make_dynamic_error_code. Also, provide
an implemention of GenericError::convertToErrorCode which for now converts
it in to the dynamic_error_code we used to have. This will go away once
all the APIs are converted.
llvm-svn: 264910
searchArchivesToOverrideTentativeDefinitions and
searchSharedLibrariesToOverrideTentativeDefinitions are always false.
For the dead flags, we have a fairly large amount of code which is
never be executed.
http://reviews.llvm.org/D17791
llvm-svn: 264653
The stack-size.yaml test had an empty atom content array. This is
legal, but asking a BumpPtrAllocator for 0 sized data may not be
legal. Instead just avoid requesting any data when we can just return
an empty ArrayRef instead.
llvm-svn: 264234
Its possible for file to have no entry atom which means that there
is no atom to check for being a thumb function. Instead just skip
the thumb check and set the entry address to 0, which matches the
current behaviour of getting a default initialised int from a map.
llvm-svn: 264233
On a 32-bit output, we may write LC_MAIN (which contains a uint64_t) to
an unaligned address. This changes it to use a memcpy instead which is UB safe.
llvm-svn: 264232
We were casting a potentially unaligned pointer to uint32_t and
dereferencing. As the pointer ultimately comes from the object file,
there's no way to guarantee alignment, so use the little32_t read instead.
Also, little32_t knows about endianness, so in theory this may have broken on
big endian machines.
llvm-svn: 264231
The .o path always makes sure to store a power of 2 value in the
Section alignment. However, the YAML code didn't verify this.
Added verification and updated all the tests which had a 3 but meant
to have 2^3.
llvm-svn: 264228
The size of a section can be zero, even when it contains atoms, so
long as all of the atoms are also size 0. In this case we were
allocating space for a 0 sized buffer.
Changed this to only allocate when we need the space, but also cleaned
up all the code to use MutableArrayRef instead of uint8_t* so its much much
safer as we get bounds checking on all of our section creation logic.
llvm-svn: 264204
On a 32-bit output, we may write LC_SOURCE_VERSION (which contains a uint64_t) to
an unaligned address. This changes it to use a memcpy instead which is UB safe.
llvm-svn: 264202
We were already copying this data to a temporary for endian swaps. Now
we just always copy it, but still only do the endian swaps when needed.
llvm-svn: 264172
Turns out that checking only x86 for empty atoms to fix UBSan then
requires the same code in the other targets too. Better to just
check this in the main run loop instead of in each target.
Should be NFC, other than fixing UBSan failures.
llvm-svn: 264116
This was caught by the UBSan bot. When the atom has no size, we would
issue a memcpy with size0 and a nullptr for the source.
Also, this code should never have references inside an empty atom so
add an assert for that while we're here.
llvm-svn: 264115
OwningAtomPtr does not have OwningAtomPtr(OwningAtomPtr&) or the equivalent
operator= as we only want to use rvalue references in it.
SortKey didn't like this on MSVC as it was synthesizing SortKey(SortKey&) and
trying to use the OwningAtomPtr(OwningAtomPtr&) method which was private an
unimplemented.
Now we explicitly have the methods on SortKey so hopefully the bot will be
happier.
llvm-svn: 264077
This is a re-commit of r264022 with a fix for MSVC. The issue there was
that the code was running DefinedAtom::~Atom() for some value and instead
needed to cast to Atom before running ~Atom. Original commit message follows.
Currently each File contains an BumpPtrAllocator in which Atom's are
allocated. Some Atom's contain data structures like std::vector which
leak as we don't run ~Atom when they are BumpPtrAllocate'd.
Now each File actually owns its Atom's using an OwningAtomPtr. This
is analygous to std::unique_ptr and may be replaced by it if possible.
An Atom can therefore only be owned by a single File, so the Resolver now
moves them from one File to another. The MachOLinkingContext owns the File's
and so clears all the Atom's in ~MachOLinkingContext, then delete's all the
File's. This makes sure all Atom's have been destructed before any of the
BumpPtrAllocator's in which they run have gone away.
Should hopefully fix the remaining leaks. Will keep an eye on the bots to
make sure.
llvm-svn: 264067
This reverts commit r264022.
This breaks the Window's bots which don't like that i'm calling ~Atom when
the this pointer is a sublcass of Atom.
Reverting for now until I try find a better fix. I tried using std::unique_ptr with
a custom deleter as a quick fix, but it didn't work well in the YAML parser.
llvm-svn: 264023
Currently each File contains an BumpPtrAllocator in which Atom's are
allocated. Some Atom's contain data structures like std::vector which
leak as we don't run ~Atom when they are BumpPtrAllocate'd.
Now each File actually owns its Atom's using an OwningAtomPtr. This
is analygous to std::unique_ptr and may be replaced by it if possible.
An Atom can therefore only be owned by a single File, so the Resolver now
moves them from one File to another. The MachOLinkingContext owns the File's
and so clears all the Atom's in ~MachOLinkingContext, then delete's all the
File's. This makes sure all Atom's have been destructed before any of the
BumpPtrAllocator's in which they run have gone away.
Should hopefully fix the remaining leaks. Will keep an eye on the bots to
make sure.
llvm-svn: 264022
In trying to fix the leaks in the MachO lld codebase, we need to have
a better model for file and atom ownership. Having the context own
everything seems like the simplest model, so change all the passes to
allocate File's on the context instead of owning files as a member.
llvm-svn: 264004
The YAML traits new's when not passed an allocator to parse data.
For atom types, this is a leak as we don't destruct atoms. For
the File here, we do actually destruct File's so that single case of
not using an allocator will be fine.
Should fix a bunch more leaks.
llvm-svn: 263680
In lld we allocate atoms on an allocator and so don't run their
destructors. This means we also shouldn't allocate memory inside
them without that also being on an allocator.
Reviewed by Lang Hames and Rafael Espindola.
llvm-svn: 263677
The current code for processCIE and processFDE returns out if it sees
any references. The problem with this is that some references could be
explicit in the binary, while others are implicit as they can be
inferred from the content of the EHFrame itself.
This change walks the references we have against the references we
need, and verifies that all explicit references are in the correct place,
and generates any missing implicit ones.
Reviewed by Lang Hames and Nick Kledzik.
Differential Revision: http://reviews.llvm.org/D15439
llvm-svn: 263590
Now that DarwinLdDriver is the only derived class of Driver.
This patch merges them and actually removed the class because
they can now just be non-member functions. This change simplifies
a common header, Driver.h.
http://reviews.llvm.org/D17788
llvm-svn: 262502
CoreDriver implements a driver for a hypothetical platform.
It is intended to be used in unittests. However, it is actually
redundant because the features are tested using the real driver
for the real platforms. So we can remove this.
http://reviews.llvm.org/D17698
llvm-svn: 262421
UniversalDriver was used as a dispatcher to each platform-specific driver.
It had its own Options.td file. It was not just too much to parse only a
few options (we only want to parse -core, -flavor or argv[0]),
but also interpreted arguments too early. For example, if you invoke lld as
"lld -flavor gnu ... -help", then you'd get the UniversalDriver's help
message instead of GnuDriver's. This patch eliminates the use of Options
from the dispatcher.
http://reviews.llvm.org/D17686
llvm-svn: 262190
We have a vector of all of the compact unwind entries anyway, and
its live as long as we need it to be. So instead of copying from that
vector to another, just take references to the range of the original vector
we need for each compact unwind page.
llvm-svn: 260437
The non lazy atoms generated in the stubs pass use an image cache to
hold all of the pointers. On arm archs, this is the __got section,
but on x86 archs it should be __nl_symbol_ptr.
rdar://problem/24572729
llvm-svn: 260271
ld64 aligns most of the stub's to 2 byte alignment, expect for
the stub helper common atoms which are 4 byte aligned.
This adds a new field to StubInfo which tracks this alignment
and ensures that this is the alignment we get in the final image.
rdar://problem/24570220
llvm-svn: 260248
Also added the defaults for whether to generate this load command, which
the cmdline options are able to override.
There was also a difference to ld64 which is fixed here in that ld64 will
generate an empty data in code command if requested.
rdar://problem/24472630
llvm-svn: 260191
This load command generates data in the LINKEDIT section which
is a list of ULEB128 delta's to all of the functions in the __text section.
It is then 0 terminated and pointer aligned to pad.
ld64 exposes the -function-starts and no-function-starts cmdline options
to override behaviour from the defaults based on file types.
rdar://problem/24472630
llvm-svn: 260188
This was a bug in our handling of these symbols compared to ld64.
Turns out that ld64 always marks these symbols as being not dead stripped.
llvm-svn: 260185
The atom content type enum is used as a tie breaker to sort atoms.
In that case, we want MachHeader to be before typeCode as it really will
be before the code in the final executable.
Test case to follow in the next commit or two.
llvm-svn: 260184
The initial segment protection was also being used to set the maximum
segment protection level. Instead, the maximum should be set according
to the architecture we are linking. For example on Mac OS it should be
RWX on most pages, but on iOS is often on R_X.
rdar://problem/24515136
llvm-svn: 259966
We currently tag on a "__LINKEDIT" when we are emitting the segments.
However, an upcoming patch aims to set the initprot and maxprot segment members
to their correct values, and in order to share code, its better to create this
segment for real and handle it in buildFileOffsets the same way ld64 does.
The commit for segment protections will add a test for this all being correct so
no test here until that code is committed.
llvm-svn: 259960
On Mac OS 10.5 and later, with X86_64 and outputting a dynamic executable,
ld64 set the CPU_SUBTYPE_LIB64 mask on the cpusubtype in the mach_header.
This adds the same functionality to lld.
rdar://problem/24507177
llvm-svn: 259826
This is of the form A.B.C.D.E and to match ld64's behaviour, is
always output to files, even when the version is 0.
rdar://problem/24472630
llvm-svn: 259746
ld64 sets both S_ATTR_PURE_INSTRUCTIONS and S_ATTR_SOME_INSTRUCTIONS
on __TEXT, __text. We only had the S_ATTR_PURE_INSTRUCTIONS attribute.
rdar://problem/24495801
llvm-svn: 259744
In the case where we are emitting to an object file, the platform is
possibly unknown, and the source object files contained load commands
for version min, we can take the maximum of those min versions and
emit in in the output object file.
This test also tests r259739.
llvm-svn: 259742
Defaulting to unknown matches ld64, but it also makes sure that all
of our code can handle not knowing the platform. For example, a later
commit will add support for version min load commands with an unknown
platform, which is a feature supported by ld64.
No test case here. The next commit will have one with the version min
code that needed this patch.
llvm-svn: 259739
This option is emitted in the min_version load commands.
Note, there's currently a difference in behaviour compared to ld64 in
that we emit a warning if we generate a min_version load command and
didn't give an sdk_version. We need to decide what the correct behaviour
is here as its possible we want to emit an error and force clients to
provide the option.
llvm-svn: 259729
If the command line contains something like -macosx_version_min and we
don't explicitly disable generation with -no_version_load_command then
we generate the LC_VERSION_MIN command in the output file.
There's a couple of FIXME's in here. These will be handled soon with
more tests but I didn't want to grow this patch any more than it already was.
rdar://problem/24472630
llvm-svn: 259718
In r259574 I fixed some of the issues with the mach header symbols
and DSO handles.
This is the next issue whereby the __mh_execute_header has to not
be dead stripped, and (to match ld64) should be dynamically referenced.
The test here should also have been added in r259574 to make sure that
we emit this symbol. But checking that it is not only emitted but also
has the correct reference type is fine.
llvm-svn: 259589
The magic file which contained these symbols inherited from archive
which meant that the resolver didn't add the required atoms as archive
members only get added when referenced. Instead we now inherit from
SimpleFile which always links in the atoms needed.
The second issue was in the handling of these symbols when we emit
the MachO. The mach header symbol needs to be in the atom list as
it gets an offset (0), and being in the atom list makes sure it is
emitted to the symbol table. DSO handles are not emitted to the
symbol table.
rdar://problem/24450654
llvm-svn: 259574
When we do debug printing of atoms, they expect their parent file
to have an ordinal. Right now we trigger a bunch of asserts.
This doesn't actually manifest in differences in any real tests, but
even if the only thing it does is improve debuggability then its still
useful.
llvm-svn: 259450
The ObjCPass is going to need to create pointer sized relocations in
the ObjC sections. This method will be used to give us a target independent
way of getting the correct kind for the refererence.
llvm-svn: 259441
An upcoming patch will use this to create lists of ObjC methods.
Adding it now to reduce the amount of code in that patch.
Test cases will follow in the other patch too.
llvm-svn: 259440
This adds the no_objc_category_merging cmdline option which will
be used in an upcoming commit to disable the category optimizer.
It is on by default in ld64 so we match that here.
Test case will come soon with the patch to make use of this option.
llvm-svn: 259439
Most of the other methods to access Reference's were on DefinedAtom so
this just keeps them all together.
This will be used in a future patch in ObjCPass which needs to add
new references. The method is virtual because we may add references to
different data structures depending on whether we parsed a macho file or
yaml.
llvm-svn: 259436
When we parse a MachoFile, we set a number of members from the parsed
file, for example, subsectionsViaSymbols.
However, a number of passes, such as ObjCPass, create local copies of
MachoFile and don't get the benefit of setting flags and other fields in
the parser. Instead we can just give a more sensible default as the parser
will definitely get the correct value from the file anyway.
llvm-svn: 259426
Now that MachoFile has classof(), we can use dyn_cast instead which
is actually the only safe way to handle this.
Turns out this actually manifests as a bug as we were incorrectly
casting instances which weren't MachoFile in to a MachoFile.
Unfortunately, there's no reliable way of checking for this as it
requires that the file we are looking for has a 0 at exactly the byte
we need for the load of subsectionsViaSymbols.
llvm-svn: 259413
__DATA, __objc_catlist contains a list of pointers to categories.
We want to atomize it so that the ObjC pass can later optimize and remove
categories. That will be a later patch.
llvm-svn: 259386
In many situations, we don't want to exit at the first error even in the
process model. For example, it is better to report all undefined symbols
rather than reporting the first one that the linker picked up randomly.
In order to handle such errors, we don't need to wrap everything with
ErrorOr (thanks for David Blaikie for pointing this out!) Instead, we
can set a flag to record the fact that we found an error and keep it
going until it reaches a reasonable checkpoint.
This idea should be applicable to other places. For example, we can
ignore broken relocations and check for errors after visiting all relocs.
In this patch, I rename error to fatal, and introduce another version of
error which doesn't call exit. That function instead sets HasError to true.
Once HasError becomes true, it stays true, so that we know that there
was an error if it is true.
I think introducing a non-noreturn error reporting function is by itself
a good idea, and it looks to me that this also provides a gradual path
towards lld-as-a-library (or at least embed-lld-to-your-program) without
sacrificing code readability with lots of ErrorOr's.
http://reviews.llvm.org/D16641
llvm-svn: 259069
The TrieNode/TrieEdge data structures here are allocated in a bumpptrallocator.
Unfortunately, TrieNode contained a std::list<TrieEdge> and as the allocator doesn't
call the TrieNode destructor, we ended up leaking the memory allocated by the std::list
itself.
Instead we can use an intrusive list as then we save the extra allocations anyway.
llvm-svn: 258725
This option matches the behaviour of ld64, that is it prevents globals
from being dead stripped in executables and dylibs.
Reviewed by Lang Hames
Differential Revision: http://reviews.llvm.org/D16026
llvm-svn: 258554
This pass currently emits an objc image info section if one is required.
This section contains the aggregated version and flags for all of the input
files.
llvm-svn: 258197
Like arch, os, etc, when we know we are going to use a file, we check
that the file has compatible objc constraints to the context, throw
appropriate errors where that is not the case, and hopefully set the
objc constraints on the context for use later.
Added 2 tests to ensure that we don't have incompatibilities between
host and simulator code as both will get x86 based architectures.
llvm-svn: 258173
Image info flags describe the objc constraint which is GC/retain/release/etc.
These need to be parsed and stored in the file so that we can do error checking.
That will come in a later commit.
llvm-svn: 258160
When generating a relocatable file, its only valid to set this flag if
all of the inputs also had the flag. Otherwise we may atomize incorrectly
when we link the relocatable file again.
Reviewed by Lang Hames.
Differential Revision: http://reviews.llvm.org/D16018
llvm-svn: 257976
The image info struct contains flags for what kind of GC/retain/release is required.
Give an error if we parse GC flags as these are unsupported.
llvm-svn: 257974
This patch makes use of the handleLoadedFile hook added in r257814.
That method is used to check the arch and the OS of the files we are linking
against the arch and OS on the context.
The first test to use this ensures that we do not try to combine i386 Mac OS code
with i386 simulator code.
llvm-svn: 257837
This is to enable isa<> support for any files which need it.
It will be used in an upcoming patch to differentiate MachOFile from other implicitly generated files.
Reviewed by Lang Hames.
Differential Revision: http://reviews.llvm.org/D16103
llvm-svn: 257830
This is called from the resolver on each file we decide we actually want to use.
Future commits will make use of this to extract useful information from the files and do
error checking against the context. For example, ensure that files are the same arch as
each other.
Reviewed by Lang Hames.
Differential Revision: http://reviews.llvm.org/D16093
llvm-svn: 257814
Summary: This is no longer needed now that the new ELF implementation supports AMDGPU.
Reviewers: ruiu, rafael
Subscribers: llvm-commits
Differential Revision: http://reviews.llvm.org/D15954
llvm-svn: 257390
The __eh_frame section contains relocations which can always be implicitly generated.
This patch tracks whether sections have only implicitly relocations and skips emitting them to the object file if that is the case.
The test case here ensures that this is the case for __eh_frame sections.
Reviewed by Lang Hames.
http://reviews.llvm.org/D15594
llvm-svn: 257099
In a UI such as XCode, it can group the headers for a library with that library.
This is done in the CMakeLists.txt for the library itself by setting the path(s)
as ADDITIONAL_HEADER_DIRS.
LLVM already does this for all of its libraries, so just adding this to lld to
make things easier. Should be NFC.
llvm-svn: 257002
In a UI such as XCode, LLVM source files are in 'libraries' while clang
files are in 'clang libraries'.
This change moves the lld source to 'lld libraries' to make code browsing easier.
It should be NFC as the build itself is still the same, just the structure in a
UI differs.
llvm-svn: 257001
The fixup content we encode here should be the offset from the
fixup location back to the last nonlocal label. We were only encoding
the address of the fixup, and not taking in to account the base address
of the atom we are in.
Updated the test case here to have a text section which will come before
the data section where the relocation lives. .data being at offset 0 had
previously been hiding this bug.
llvm-svn: 256974
The final section order in relocatable files was just a side effect
of the atom sorter. This meant that sections like __data were before
__text because __data has RW permissions and __text RX and RW was less
than RX in our enum.
Final linked images had an actual section/segment sorter. There was no
reason for the difference, so simplify a bunch of code and just use the
same sorted for everything.
Reviewed by Lang Hames.
http://reviews.llvm.org/D15868
llvm-svn: 256786
The encoded value should be an offset from the fixup location, which
means that it should take in to account the fixup offset in its section.
We weren't subtracting the base address of the atom, which meant that when
we parsed the file again for a round trip, we had 2x the atom address in our
target address.
I've also improved comments for these to try and describe what is going on.
There's no test case right now, as the bug is only exhibited when __data is at
a non-zero address in a -r link. A commit will soon sort the sections differently
and move __data to after __text. Then these relocations in
test/mach-o/parse-data-relocs-x86_64.yaml will test for this bug.
llvm-svn: 256779
negDelta32 is only ever implicitly generated as the FDE->CIE reference.
We therefore don't emit a relocation for it in the object file in -r mode.
The value we write in to the FDE location therefore needs to point to the
final target address of the CIE, and not the inAtomAddress as it was currently
doing.
llvm-svn: 255835
We used to parse the LLVM options in Driver::link. However, that is
after parse() where we load files. By moving the LLVM option handling
earlier, we can add DEBUG() to code such as MachONormalizedFileToAtoms.cpp
and have it enabled correctly by '-mllvm --debug'.
llvm-svn: 255819
We had some DEBUG prints these passes, but add more so that its clear where we are dumping
things, and what state we are in when we do so.
I'll be adding more and more DEBUG printing to try make it easier to observe whats going on
without having to attach a debugger.
llvm-svn: 255805
The delta64 relocation is represented as the pair ARM64_RELOC_SUBTRACTOR and ARM64_RELOC_UNSIGNED.
Those should always have the same offset, so this adds a check and tests to ensure this is the case.
Also updated the error printing in this case to shows both relocs when erroring on pair.
llvm-svn: 255274
table.
The first entry in the MachO symbol table is always the empty string: make sure
we reserve space for it, or we will overflow the symbol table by one byte.
No test case - this manifests as an occasional memory error. In the near future
I hope to set up a bot building and runnnig LLD with sanitizers - that should
catch future instances of this issue.
llvm-svn: 255178
The gcc_except_tab was generating these references to point to the typeinfo in the data section.
gcc_except_tab also had the DW_EH_PE_indirect flag set which means that at runtime we are going
to dereference this entry as if it is in the GOT.
Reviewed by Nick Kledzik in http://reviews.llvm.org/D15360.
llvm-svn: 255085
MachODefinedCustomSectionAtom.
The section names for these atoms are initialized from temporaries (e.g.
segName + "/" + sectName), so we can't use StringRef here.
llvm-svn: 251610
GNU linkers accept both variants and at least for MIPS target gcc passes
joined variant of the '-m' option.
Differential Revision: http://reviews.llvm.org/D14133
llvm-svn: 251497
This is a basic initial implementation of the -flat_namespace and
-undefined options for LLD-darwin. It ignores several subtlties,
but the result is close enough that we can now link LLVM (but not
clang) on Darwin and pass all regression tests.
llvm-svn: 248732
Now it is possible to have mips-linux-gnu-ld executable and link MIPS 64-bit
little-endian binaries providing -melf64ltsmip command line argument.
llvm-svn: 246335
This is a basic implementation that allows lld to emit binaries
consumable by the HSA runtime.
Differential Revision: http://reviews.llvm.org/D11267
llvm-svn: 246155
Add PT_PHDR segment depending on its availability in linker script's
PHDRS command, fallback if no linker script is given.
Handle FILEHDR, PHDRS and FLAGS attributes of program header.
Differential Revision: http://reviews.llvm.org/D11589
llvm-svn: 244743
This has a few advantages
* Less C++ code (about 300 lines less).
* Less machine code (about 14 KB of text on a linux x86_64 build).
* It is more debugger friendly. Just set a breakpoint on the exit function and
you get the complete lld stack trace of when the error was found.
* It is a more robust API. The errors are handled early and we don't get a
std::error_code hot potato being passed around.
* In most cases the error function in a better position to print diagnostics
(it has more context).
llvm-svn: 244215
Lang Hames