Summary:
Valid section or chunk alignments are powers of 2 in the range [1,
8192]. These can be stored more canonically in log2 form to free up some
bits in Chunk. Combined with D61696, SectionChunk gets 8 bytes smaller.
Reviewers: ruiu, aganea
Subscribers: llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D61698
llvm-svn: 361206
Summary:
Prior to this change, every implementation of writeTo would add
OutputSectionOff to the output section buffer start before writing data.
Instead, do this math in the caller, so that it can be written once
instead of many times.
The output section offset is always equivalent to the difference between
the chunk RVA and the output section RVA, so we can replace the one
remaining usage of OutputSectionOff with that subtraction.
This doesn't change the size of SectionChunk because of alignment
requirements, but I will rearrange the fields in a follow-up change to
accomplish that.
Reviewers: ruiu, aganea
Subscribers: llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D61696
llvm-svn: 360376
SectionChunk is one of the most frequently allocated data structures in
LLD, since there are about four per function when optimizations and
debug info are enabled (.text, .pdata, .xdata, .debug$S).
A PE COFF file cannot be larger than 2GB, so there is an inherent limit
on the length of the section name and the number of relocations.
Decompose the ArrayRef and StringRef into pointer and size, and put them
back together in the accessors for section name and relocation list.
I plan to gather complete performance numbers later by padding
SectionChunk with dead data and measuring performance after all the size
optimizations are done.
llvm-svn: 359923
Summary:
Reorder the fields in both to use padding more efficiently, and add more
comments on the purpose of the fields.
Replace `std::vector<SectionChunk*> AssociativeChildren` with a
singly-linked list. This avoids the separate vector allocation to list
associative children, and shrinks the 3 pointers used for the typically
empty vector down to 1.
In the end, this reduces the sum of heap allocations used to link
browser_tests.exe with NO PDB by 13.10%, going from 2,248,728 KB to
1,954,071 KB of heap. These numbers exclude memory mapped files, which
are of course a significant factor in LLD's memory usage.
Reviewers: ruiu, mstorsjo, aganea
Subscribers: jdoerfert, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D59797
llvm-svn: 357535
Summary:
This avoids allocating O(#relocs) of intermediate data for each section
when range extension thunks aren't needed for that section. This also
removes a std::vector from SectionChunk, which further reduces its size.
Instead, this change adds the range extension thunk symbols to the
object files that contain sections that need extension thunks. By adding
them to the symbol table of the parent object, that means they now have
a symbol table index. Then we can then modify the original relocation,
after copying it to read-write memory, to use the new symbol table
index.
This makes linking browser_tests.exe with no PDB 10.46% faster, moving
it from 11.364s to 10.288s averaged over five runs.
Reviewers: mstorsjo, ruiu
Subscribers: aganea, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D59902
llvm-svn: 357200
On ARM64, this is normally necessary only after a module exceeds
128 MB in size (while the limit for thumb is 16 MB). For conditional
branches, the range limit is only 1 MB though (the same as for thumb),
and for the tbz instruction, the range is only 32 KB, which allows for
a test much smaller than the full 128 MB.
This fixes PR40467.
Differential Revision: https://reviews.llvm.org/D57575
llvm-svn: 352929
I need the comdat selection for PR40094. To keep the patch for that smaller,
I'm adding it here, and as a first application I'm using it to reject
associative comdats referring to earlier associative comdats. Depends on
D56929; together with that all associative comdats referring to other
associative comdats are now rejected.
Differential Revision: https://reviews.llvm.org/D56931
llvm-svn: 352254
to reflect the new license.
We understand that people may be surprised that we're moving the header
entirely to discuss the new license. We checked this carefully with the
Foundation's lawyer and we believe this is the correct approach.
Essentially, all code in the project is now made available by the LLVM
project under our new license, so you will see that the license headers
include that license only. Some of our contributors have contributed
code under our old license, and accordingly, we have retained a copy of
our old license notice in the top-level files in each project and
repository.
llvm-svn: 351636
Changes a few things I noticed while reading this code.
- fix a few typos in comments
- remove two `auto` uses where the type wasn't clear to me
- add comment saying that two sequential checks for `if (SparseChunks[SectionNumber] == PendingComdat)` are intentional
- name two parameters
No behavior change.
Differential Revision: https://reviews.llvm.org/D56677
llvm-svn: 351101
When these are accessed with load/store instructions on ARM64,
it becomes strictly necessary to have them properly aligned.
This fixes PR39228.
Differential Revision: https://reviews.llvm.org/D53128
llvm-svn: 344264
(patch by Benoit Rousseau)
This patch fixes a bug where the global variable initializers were sometimes not invoked in the correct order when it involved a C++ template instantiation.
Differential Revision: https://reviews.llvm.org/D52749
llvm-svn: 343847
This is a feature that MS link.exe lacks; it currently errors out on
such relocations, just like lld did before.
This allows linking clang.exe for ARM - practically, any image over
16 MB will likely run into the issue.
Differential Revision: https://reviews.llvm.org/D52156
llvm-svn: 342962
MinGW uses these kind of list terminator symbols for traversing
the constructor/destructor lists. These list terminators are
actual pointers entries in the lists, with the values 0 and
(uintptr_t)-1 (instead of just symbols pointing to the start/end
of the list).
(This mechanism exists in both the mingw-w64 crt startup code and
in libgcc; normally the mingw-w64 one is used, but a DLL build of
libgcc uses the libgcc one. Therefore it's not trivial to change
the mechanism without lots of cross-project synchronization and
potentially invalidating some combinations of old/new versions
of them.)
When mingw-w64 has been used with lld so far, the CRT startup object
files have so far provided these symbols, ending up with different,
incompatible builds of the CRT startup object files depending on
whether binutils or lld are going to be used.
In order to avoid the need of different configuration of the CRT startup
object files depending on what linker to be used, provide these symbols
in lld instead. (Mingw-w64 checks at build time whether the linker
provides these symbols or not.) This unifies this particular detail
between the two linkers.
This does disallow the use of the very latest lld with older versions
of mingw-w64 (the configure check for the list was added recently;
earlier it simply checked whether the CRT was built with gcc or clang),
and requires rebuilding the mingw-w64 CRT. But the number of users of
lld+mingw still is low enough that such a change should be tolerable,
and unifies this aspect of the toolchains, easing interoperability
between the toolchains for the future.
The actual test for this feature is added in ctors_dtors_priority.s,
but a number of other tests that checked absolute output addresses
are updated.
Differential Revision: https://reviews.llvm.org/D52053
llvm-svn: 342294
After fixing up the runtime pseudo relocation, the .refptr.<var>
will be a plain pointer with the same value as the IAT entry itself.
To save a little binary size and reduce the number of runtime pseudo
relocations, redirect references to the IAT entry (via the __imp_<var>
symbol) itself and discard the .refptr.<var> chunk (as long as the
same section chunk doesn't contain anything else than the single
pointer).
As there are now cases for both setting the Live variable to true
and false externally, remove the accessors and setters and just make
the variable public instead.
Differential Revision: https://reviews.llvm.org/D51456
llvm-svn: 341175
Normally, in order to reference exported data symbols from a different
DLL, the declarations need to have the dllimport attribute, in order to
use the __imp_<var> symbol (which contains an address to the actual
variable) instead of the variable itself directly. This isn't an issue
in the same way for functions, since any reference to the function without
the dllimport attribute will end up as a reference to a thunk which loads
the actual target function from the import address table (IAT).
GNU ld, in MinGW environments, supports automatically importing data
symbols from DLLs, even if the references didn't have the appropriate
dllimport attribute. Since the PE/COFF format doesn't support the kind
of relocations that this would require, the MinGW's CRT startup code
has an custom framework of their own for manually fixing the missing
relocations once module is loaded and the target addresses in the IAT
are known.
For this to work, the linker (originall in GNU ld) creates a list of
remaining references needing fixup, which the runtime processes on
startup before handing over control to user code.
While this feature is rather controversial, it's one of the main features
allowing unix style libraries to be used on windows without any extra
porting effort.
Some sort of automatic fixing of data imports is also necessary for the
itanium C++ ABI on windows (as clang implements it right now) for importing
vtable pointers in certain cases, see D43184 for some discussion on that.
The runtime pseudo relocation handler supports 8/16/32/64 bit addresses,
either PC relative references (like IMAGE_REL_*_REL32*) or absolute
references (IMAGE_REL_AMD64_ADDR32, IMAGE_REL_AMD64_ADDR32,
IMAGE_REL_I386_DIR32). On linking, the relocation is handled as a
relocation against the corresponding IAT slot. For the absolute references,
a normal base relocation is created, to update the embedded address
in case the image is loaded at a different address.
The list of runtime pseudo relocations contains the RVA of the
imported symbol (the IAT slot), the RVA of the location the relocation
should be applied to, and a size of the memory location. When the
relocations are fixed at runtime, the difference between the actual
IAT slot value and the IAT slot address is added to the reference,
doing the right thing for both absolute and relative references.
With this patch alone, things work fine for i386 binaries, and mostly
for x86_64 binaries, with feature parity with GNU ld. Despite this,
there are a few gotchas:
- References to data from within code works fine on both x86 architectures,
since their relocations consist of plain 32 or 64 bit absolute/relative
references. On ARM and AArch64, references to data doesn't consist of
a plain 32 or 64 bit embedded address or offset in the code. On ARMNT,
it's usually a MOVW+MOVT instruction pair represented by a
IMAGE_REL_ARM_MOV32T relocation, each instruction containing 16 bit of
the target address), on AArch64, it's usually an ADRP+ADD/LDR/STR
instruction pair with an even more complex encoding, storing a PC
relative address (with a range of +/- 4 GB). This could theoretically
be remedied by extending the runtime pseudo relocation handler with new
relocation types, to support these instruction encodings. This isn't an
issue for GCC/GNU ld since they don't support windows on ARMNT/AArch64.
- For x86_64, if references in code are encoded as 32 bit PC relative
offsets, the runtime relocation will fail if the target turns out to be
out of range for a 32 bit offset.
- Fixing up the relocations at runtime requires making sections writable
if necessary, with the VirtualProtect function. In Windows Store/UWP apps,
this function is forbidden.
These limitations are addressed by a few later patches in lld and
llvm.
Differential Revision: https://reviews.llvm.org/D50917
llvm-svn: 340726
In an upcoming change I will need to make a distinction between section
type (code, data, bss) and permissions. The term that I use for both
of these things is "output characteristics".
Differential Revision: https://reviews.llvm.org/D45799
llvm-svn: 330361
Summary:
This change does three things:
- Try to find the file and line number of an undefined symbol
reference by reading codeview debug info.
- Try to find the name of the function or global variable with the
undefined symbol reference by searching the object file's symbol
table.
- Prints the information in the same style as the ELF linker.
Differential Revision: https://reviews.llvm.org/D45467
llvm-svn: 330235
In COFF, duplicate string literals are merged by placing them in a
comdat whose leader symbol name contains a specific prefix followed
by the hash and partial contents of the string literal. This gives
us an easy way to identify sections containing string literals in
the linker: check for leader symbol names with the given prefix.
Any sections that are identified in this way as containing string
literals may be tail merged. We do so using the StringTableBuilder
class, which is also used to tail merge string literals in the ELF
linker. Tail merging is enabled only if ICF is enabled, as this
provides a signal as to whether the user cares about binary size.
Differential Revision: https://reviews.llvm.org/D44504
llvm-svn: 327668
Summary:
This patch adds some initial support for Windows control flow guard. At
the end of the day, the linker needs to synthesize a table of RVAs very
similar to the structured exception handler table (/safeseh).
Both /safeseh and /guard:cf take sections of symbol table indices
(.sxdata and .gfids$y) and turn them into RVA tables referenced by the
load config struct in the CRT through special symbols.
Reviewers: ruiu, amccarth
Subscribers: llvm-commits
Differential Revision: https://reviews.llvm.org/D42592
llvm-svn: 324306
If /debug was not specified, readSection will return a null
pointer for debug sections. If the debug section is associative with
another section, we need to make sure that the section returned from
readSection is not a null pointer before adding it as an associative
section.
Differential Revision: https://reviews.llvm.org/D40533
llvm-svn: 319133
With this change, instead of creating a SectionChunk for each section
in the object file, we only create them when we encounter a prevailing
comdat section.
Also change how symbol resolution occurs between comdat symbols. Now
only the comdat leader participates in comdat resolution, and not any
other external associated symbols. This is more in line with how COFF
semantics are defined, and should allow for a more straightforward
implementation of non-ANY comdat types.
On my machine, this change reduces our runtime linking a release
build of chrome_child.dll with /nopdb from 5.65s to 4.54s (median of
50 runs).
Differential Revision: https://reviews.llvm.org/D40238
llvm-svn: 319090
Now that we have only SymbolBody as the symbol class. So, "SymbolBody"
is a bit strange name now. This is a mechanical change generated by
perl -i -pe s/SymbolBody/Symbol/g $(git grep -l SymbolBody lld/ELF lld/COFF)
nd clang-format-diff.
Differential Revision: https://reviews.llvm.org/D39459
llvm-svn: 317370
New lld's files are spread under lib subdirectory, and it isn't easy
to find which files are actually maintained. This patch moves maintained
files to Common subdirectory.
Differential Revision: https://reviews.llvm.org/D37645
llvm-svn: 314719
These are emitted for comm symbols in object files, when targeting
a GNU environment.
Alternatively, just ignore them since we already align CommonChunk
to the natural size of the content (up to 32 bytes). That would only
trade away the possibility to overalign small symbols, which doesn't
sound like something that might not need to be handled?
Differential Revision: https://reviews.llvm.org/D36304
llvm-svn: 310871
In order to get dbghelp to load our pdb, we have to fill in the
PointerToRawData field as well as the AddressOfRawData field. One is the
file offset and the other is the RVA.
llvm-svn: 309900
DWARF debug sections can also contain relocations against symbols in
discared segments. LLD should accept such relocations.
Differential Revision: https://reviews.llvm.org/D35526
llvm-svn: 308315
Summary:
This would have caught the invalid object file I used in my test case in
r307726. The OOB was only caught by ASan later, which is slow and
doesn't work on some platforms. LLD should do some basic input
validation itself. This check isn't perfect, so relocations can reach
OOB by up to seven bytes, but it's better than what we had and probably
cheap.
Reviewers: ruiu
Subscribers: llvm-commits
Differential Revision: https://reviews.llvm.org/D35371
llvm-svn: 307948
This is enough to link a working hello world executable, with
a call to an imported function, a string constant passed to
the imported function, and loads from a global variable.
Differential Revision: https://reviews.llvm.org/D34964
llvm-svn: 307629
Summary:
In order to do this without switching on the symbol kind multiple times,
I created Defined::getChunkAndOffset and use that instead of
SymbolBody::getRVA in the inner relocation loop.
Now we get the symbol's chunk before switching over relocation types, so
we can test if it has been discarded outside the inner relocation type
switch. This also simplifies application of section relative
relocations. Previously we would switch on symbol kind to compute the
RVA, then the relocation type, and then the symbol kind again to get the
output section so we could subtract that from the symbol RVA. Now we
*always* have an OutputSection, so applying SECREL and SECTION
relocations isn't as much of a special case.
I'm still not quite happy with the cleanliness of this code. I'm not
sure what offsets and bases we should be using during the relocation
processing loop: VA, RVA, or OutputSectionOffset.
Reviewers: ruiu, pcc
Reviewed By: ruiu
Subscribers: majnemer, inglorion, llvm-commits, aprantl
Differential Revision: https://reviews.llvm.org/D34650
llvm-svn: 306566
Summary:
Previously we didn't add debug info chunks to the SparseChunks array, so
they didn't participate in section GC. Now we do.
Reviewers: ruiu
Subscribers: aprantl, llvm-commits
Differential Revision: https://reviews.llvm.org/D34356
llvm-svn: 305811
Summary:
This is a first step towards getting line info to show up in VS and
windbg. So far, only llvm-pdbutil can parse the PDBs that we produce.
cvdump doesn't like something about our file checksum tables. I'll have
to dig into that next.
This patch adds a new DebugSubsectionRecordBuilder which takes bytes
directly from some other producer, such as a linker, and sticks it into
the PDB. Line tables only need to be relocated. No data needs to be
rewritten.
File checksums and string tables, on the other hand, need to be re-done.
Reviewers: zturner, ruiu
Subscribers: llvm-commits, hiraditya
Differential Revision: https://reviews.llvm.org/D34257
llvm-svn: 305713
Summary:
Adds a "Discarded" bool to SectionChunk to indicate if the section was
discarded by COMDAT deduplication. The Writer still just checks
`isLive()`.
Fixes PR33446
Reviewers: ruiu
Subscribers: llvm-commits
Differential Revision: https://reviews.llvm.org/D34288
llvm-svn: 305582
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
LLD used to take 11.73 seconds to link Clang. Now it is 6.94 seconds.
MSVC link takes 83.02 seconds. Note that ICF is enabled by default on
Windows, so a low latency ICF is more important than in ELF.
llvm-svn: 288487
Reid Kleckner