We would previously accept `--threads=4`, but this option just turns on
threading and does not specify a number of threads.
I ran into this by accident because I was passing `--threads=<n>` but
the number didn't seem to affect anything.
llvm-svn: 270963
MIPS .reginfo and .MIPS.options sections are consumed by the linker, and
the linker produces a single output section. But it is possible that
input files contain section symbol points to the corresponding input
section. In case of generation a relocatable output we need to write
such symbols to the output file.
Fixes bug 27878.
Differential Revision: http://reviews.llvm.org/D20688
llvm-svn: 270910
D15779 introduced basic approach to support new relaxations.
This patch implements relaxations for jmp and call instructions,
described in System V Application Binary Interface AMD64 Architecture Processor
Supplement Draft Version 0.99.8 (https://github.com/hjl-tools/x86-psABI/wiki/x86-64-psABI-r249.pdf,
B.2 "B.2 Optimize GOTPCRELX Relocations")
Differential revision: http://reviews.llvm.org/D20622
llvm-svn: 270721
This patch makes SectionPiece class 8 bytes smaller on platforms
on which pointer size is 8 bytes. Sean suggested in a post commit
review for r270340 that this could make a differentce, and it
actually is. Time to link clang (with debug info) improved from
6.725 seconds to 6.589 seconds or by about 2%.
Differential Revision: http://reviews.llvm.org/D20613
llvm-svn: 270717
Add another possible error that may be reported for the same case. The
original reproduction case that prompted r270706 produced the error
"corrupted CIE" instead of "corrupted or unsupported CIE information".
The specific error depends on arbitrary data later in the file so
check that neither is emitted in case the input is ever changed.
Document the process used to create the input .o and rename the test
file to .s, as requested by Rafael.
llvm-svn: 270709
"A zero length string indicates that no augmentation data is present."
The FreeBSD/mips toolchain (GCC 4.2.1) generates .debug_frame sections
containing CIE records that have an empty augmentation string.
Differential Revision: http://reviews.llvm.org/D19928
llvm-svn: 270706
System V Application Binary Interface AMD64 Architecture Processor Supplement Draft Version 0.99.8
(https://github.com/hjl-tools/x86-psABI/wiki/x86-64-psABI-r249.pdf, B.2 "B.2 Optimize GOTPCRELX Relocations")
introduces possible relaxations for R_X86_64_GOTPCRELX and R_X86_64_REX_GOTPCRELX.
That patch implements the next relaxation:
mov foo@GOTPCREL(%rip), %reg => lea foo(%rip), %reg
and also opens door for implementing all other ones.
Implementation was suggested by Rafael Ávila de Espíndola with few additions and testcases by myself.
Differential revision: http://reviews.llvm.org/D15779
llvm-svn: 270705
scanReloc and the functions on which scanReloc depends is in total
more than 600 lines of code. Since scanReloc does not depend on Writer,
it is better to move it into a separate file.
Differential Revision: http://reviews.llvm.org/D20554
llvm-svn: 270606
This patch addresses a post-commit review for r270325. r270325
introduced getReloc function that searches a relocation for a
given range. It always started searching from beginning of relocation
vector, so it was slower than before. Previously, we used to use
the fact that the relocations are sorted. This patch restore it.
llvm-svn: 270572
Previously, we created a .bss section when needed. We had a function
ensureBss() for that purpose. Turned out that was error-prone
because it was easy to forget to call that function before accessing
the .bss section.
This patch always make the BSS section. The section is added to the
output when it's not empty.
llvm-svn: 270527
Copy relocations are relocations to copy data from DSOs to
executable's .bss segment at runtime. It doesn't make sense to
create such relocations for zero-sized symbols.
GNU linkers don't agree with each other. ld rejects such
relocation/symbol pair. gold don't reject that but do not create
copy relocations as well. I took the former approach because
I don't think the latter is what user wants.
llvm-svn: 270525
Previously, mergeable section's constructors did more than just
setting member variables; it split section contents into small
pieces. It is not always computationally cheap task because if
the section is a mergeable string section, it needs to scan the
entire section to split them by NUL characters.
If a section would be thrown away by GC, that cost ended up
being a waste of time. It is going to be larger problem if the
section is compressed -- the whole time to uncompress it and
split it up is going to be a waste.
Luckily, we can defer section splitting after GC. We just have
to remember which offsets are in use during GC and apply that later.
This patch implements it.
Differential Revision: http://reviews.llvm.org/D20516
llvm-svn: 270455
.eh_frame_hdr assumes that there is only one .eh_frame and
ensures it by assertions. This patch makes .eh_frame a real
singleton object to simplify.
llvm-svn: 270445
Previously, EhFrameHdr section computed addresses to which FDEs are
applied to. This is not an ideal design because EhFrameHdr does not
know much about FDEs unless EhFrame passes the information to EhFrameHdr.
It is what we did.
This patch simplifies the code by making EhFrame to compute the
values and pass the cooked information to EhFrameHdr. EhFrameHdr no
longer have to know about the details of FDEs such as FDE encodings.
llvm-svn: 270393
This patch refactors EHOutputSection using SectionPiece struct.
EHRegion class was removed since we can now directly use SectionPiece.
An incomplete support of large CIE/FDE record (> 2^32 bytes) was removed
because it silently created broken executable. There are several places
in the existing code that "size" field is always 4 bytes and at offset 4
in the record, which is not true for 64-bit size records. We will have to
support that in future, but it is better to error out instead of creating
malformed eh_frame sections.
llvm-svn: 270382
This patch adds Size member to SectionPiece so that getRangeAndSize
can just return a SectionPiece instead of a std::pair<SectionPiece *, uint_t>.
Also renamed the function.
llvm-svn: 270346
We were using std::pair to represents pieces of splittable section
contents. It hurt readability because "first" and "second" are not
meaningful. This patch give them names.
One more thing is that piecewise liveness information is stored to
the second element of the pair as a special value of output section
offset. It was confusing, so I defiend a new bit, "Live", in the
new struct.
llvm-svn: 270340
This fixes a potential bug when cross linking very large executables
on LLP64 machines such as Windows. On such platform, uintX_t is 64 bits
while unsigned is 32 bits.
llvm-svn: 270327
Most functions take destination buffers as the first arguments
just like memcpy, so this order is easier to read.
Also simplified the function.
llvm-svn: 270324
This makes it explicit that each R_RELAX_TLS_* is equivalent to some
other expression.
With this I think we are at a sweet spot for how much is done in
Target.cpp. I did experiment with moving *all* the value math out of it.
It has the advantage that we know the final value in target independent
code, but it gets quite verbose.
llvm-svn: 270277
This adds direct support for computing offsets from the thread pointer
for both variants. Of the architectures we support, variant 1 is used
only by aarch64 (but that doesn't seem to be documented anywhere.)
llvm-svn: 270243
the linker script. The cycle in the ELF/LinkerScript.cpp:assignAddresses()
routine will be used to go through all the sections and set all the
addresses correctly.
Add new test to check this case.
llvm-svn: 270090
Lazy binding is quite important for use case like a shared build of
llvm. Also, if someone wants to disable it, it is better done in the
compiler (disable plt generation).
The only reason to keep it is to make it easier to add a new
architecture. But it doesn't really help much as it is possible to start
with non lazy relocation and plt code but still let the generic part
create a dedicated .got.plt and .rela.plt.
llvm-svn: 269982
New names reflect purpose of corresponding GOT entries better.
Both expression types related to entries allocated in the 'local'
part of MIPS GOT. R_MIPS_GOT_LOCAL_PAGE is for entries contain 'page'
addresses. R_MIPS_GOT_LOCAL is for entries contain 'full' address.
llvm-svn: 269597
If you specify the option in the form of --build-id=0x<hexstring>,
that hexstring is set as a build ID. We observed that the feature
is actually in use in some builds, so we want this feature.
llvm-svn: 269495
The Elf_Rela has an explicit addend. It doesn't need the addend to be
written to the section being relocated.
Since relative relocations are very common in the output, this is a
noticeable speedup. The results I got were
chromium
master 4.778149487
patch 4.761120792 0.996436131802
chromium fast
master 1.896253636
patch 1.840990582 0.970856718241
the gold plugin
master 0.399337811
patch 0.392279276 0.982324401032
clang
master 0.666873675
patch 0.665895708 0.998533504865
llvm-as
master 0.037101095
patch 0.037123149 1.00059442989
the gold plugin fsds
master 0.422473396
patch 0.414192879 0.980399909016
clang fsds
master 0.747302008
patch 0.744843964 0.996710775599
llvm-as fsds
master 0.033146245
patch 0.033064531 0.997534743377
scylla
master 4.08857525
patch 4.082245184 0.998451767275
llvm-svn: 269417
Just do not allow to link shared library if there are
undefined symbols.
This fixes PR27447
Differential revision: http://reviews.llvm.org/D20169
llvm-svn: 269183
win32 was my case.
Before that change test failed with next error for me:
23> ******************** TEST 'lld :: ELF/mips-64-got.s' FAILED ********************
....
23> Command 3 Stderr:
23> relocation R_MIPS_GOT_PAGE out of range
llvm-svn: 269166
This is the option which sorts relocs to optimize dynamic linker performance.
-z combelocs is the default in gold, also it ignores -z nocombreloc,
this patch do the same.
Patch sorts relocations by symbols only and do not create any
DT_REL[A]COUNT entries. That is different with what gold/bfd do.
More information about option is here:
http://www.airs.com/blog/archives/186http://people.redhat.com/jakub/prelink.pdf, p.2
Differential revision: http://reviews.llvm.org/D19528
llvm-svn: 269066
MIPS N64 ABI packs multiple relocations into the single relocation
record. In general, all up to three relocations can have arbitrary types.
In fact, Clang and GCC uses only a few combinations. For now, we support
two of them. That is allow to pass at least all LLVM test suite cases.
<any relocation> / R_MIPS_SUB / R_MIPS_HI16 | R_MIPS_LO16
<any relocation> / R_MIPS_64 / R_MIPS_NONE
The first relocation is a 'real' relocation which is calculated using
the corresponding symbol's value. The second and the third relocations
used to modify result of the first one: extend it to 64-bit, extract
high or low part etc. For details, see part 2.9 'Relocation' at
https://dmz-portal.mips.com/mw/images/8/82/007-4658-001.pdf
llvm-svn: 268876
In case of MIPS ABI relocation has R_GOTREL expression's type iif the
relocation type is either R_MIPS_GPREL16 or R_MIPS_GPREL32. So it is
enough to check expression's type only.
llvm-svn: 268741
We were creating the copy relocations just fine, but then thinking that
the .bss position could be preempted and creating a dynamic relocation
to it, which would crash at runtime since that memory is read only.
llvm-svn: 268668
This allows the combined LTO object to provide a definition with the same
name as a symbol that was internalized without causing a duplicate symbol
error. This normally happens during parallel codegen which externalizes
originally-internal symbols, for example.
In order to make this work, I needed to relax the undefined symbol error to
only report an error for symbols that are used in regular objects.
Differential Revision: http://reviews.llvm.org/D19954
llvm-svn: 268649
We were previously using an output offset of -1 for both GC'd and tail
merged pieces. We need to distinguish these two cases in order to filter
GC'd symbols from the symbol table -- we were previously asserting when we
asked for the VA of a symbol pointing into a dead piece, which would end
up asking the tail merging string table for an offset even though we hadn't
initialized it properly.
This patch fixes the bug by using an offset of -1 to exclusively mean GC'd
pieces, using 0 for tail merges, and distinguishing the tail merge case from
an offset of 0 by asking the output section whether it is tail merge.
Differential Revision: http://reviews.llvm.org/D19953
llvm-svn: 268604
It is insanely hard to write a test that works both on Windows and Unix.
I tried to workaround it with cpio's minor options, but the behaviors of
the options were myterious. It just doesn't worth to spend time on it.
And probably minor options could break buildbots that doesn't have the
GNU version of cpio command.
In this patch, I simply added a separate test file that runs only on
Windows.
llvm-svn: 268596
MIPS N64 ABI packs multiple relocations into the single relocation
record. Particularly it requires to represent dynamic relative
relocation as a combination of R_MIPS_REL32 and R_MIPS_64 relocations.
llvm-svn: 268565
We were already checking for non relative relocations.
If we ever decide to add support for rw text segments this means we will
have a single spot to add the flag.
llvm-svn: 268558
Also improves the error message. Previously it would just print out
the cause (e.g. "permission denied"). Now it prints out something like
"--reproduce: failed to open foo.cpio: permission denied".
llvm-svn: 268551
Currently we don't check when creating relative relocations if the
section is read only or not. I am about to fix that, so first update the
patches that depend on the current behavior.
llvm-svn: 268542
These relocations introduced by MIPS N64 ABI. R_MIPS_GOT_DISP references
GOT entry with full symbol's address, R_MIPS_GOT_PAGE creates GOT entry
with address of memory page which includes symbol's address,
R_MIPS_GOT_OFST used together with R_MIPS_GOT_PAGE. This relocation
calculates offset from beginning of memory page to the symbol address.
llvm-svn: 268525
As requested by Rafael Espindola in his post-commit comments on r268036. This
makes the previous behaviour the default while still allowing verification of
IAS.
llvm-svn: 268496
This is both simpler and safer. If we crash at any point, there is a
valid cpio file to reproduce the crash.
Thanks to Rui for the suggestion.
llvm-svn: 268495
MIPS N64 ABI introduces .MIPS.options section which specifies miscellaneous
options to be applied to an object/shared/executable file. LLVM as well as
modern versions of GNU tools read and write the only type of the options -
ODK_REGINFO. It is exact copy of .reginfo section used by O32 ABI.
llvm-svn: 268485
Both bfd and gold have this. It allows disabling build-id when it is the
default with by adding -Wl,--build-id=none no the clang command line.
llvm-svn: 268435
Introduce a special symbol type to indicate that we have not yet seen a type
for the symbol, so we should not report TLS mismatches for that symbol.
Differential Revision: http://reviews.llvm.org/D19836
llvm-svn: 268411
We want --reproduce to
* not rewrite scripts and thin archives
* work with absolute paths
Given that, it pretty much has to create a full directory tree. On windows that
is problematic because of the very short maximum path limit. On most cases
users can still work around it with "--repro c:\r", but that is annoying and
not viable for automated testing.
We then need to produce some form of archive with the files. The first option
that comes to mind is .a files since we already have code for writing them.
There are a few problems with them
The format has a dedicated string table, so we cannot start writing it until
all members are known.
Regular implementations don't support creating directories. We could make
llvm-ar support that, but that is probably not a good idea.
The next natural option would be tar. The problem is that to support long path
names (which is how this started) it needs a "pax extended header" making this
an annoying format to write.
The next option I looked at seems a natural fit: cpio files.
They are available on pretty much every unix, support directories and long path
names and are really easy to write. The only slightly annoying part is a
terminator, but at least gnu cpio only prints a warning if it is missing, which
is handy for crashes. This patch still makes an effort to always create it.
llvm-svn: 268404
The test is now unexpectedly passing on
llvm-clang-lld-x86_64-scei-ps4-ubuntu-fast which is treated as an error.
For now, disable Windows testing of the feature.
Rafael is working on generating an archive, which will hopefully allow
us to turn this test back on.
Unfortunately, we don't have a way to temporarily XFAIL this test just
on llvm-clang-lld-x86_64-scei-ps4-windows10pro-fast.
llvm-svn: 268351
Weak undefined symbols resolve to the image base. This is a little strange,
but it allows us to link function calls to such symbols. Normally such a
call will be guarded with a comparison, which will load a zero from the GOT.
There's one example of such a function call in crti.o in Linux's CRT.
As part of this change, I also needed to make the synthetic start and end
symbols image base relative in the case where their sections were empty,
so that PC-relative references to those symbols would continue to work.
Differential Revision: http://reviews.llvm.org/D19844
llvm-svn: 268350
`REQUIRES: shell` is not appropriate because that would mean that there
are no windows bots testing this, and that is precisely where it needs
the most testing.
Rafael or Rui are working on generating an archive directly, which
should avoid this issue.
We can try to move the bot to a shorter build directory path.
llvm-svn: 268345
This change simplifies the BuildId classes by removing a few member
functions and variables from them. It should also make it easy to
parallelize hash computation in future because now each BuildId object
see all inputs rather than one at a time.
llvm-svn: 268333
This patch increases the size of Undefined by the size of a pointer,
but it wouldn't actually increase the size of memory that LLD uses
because we are not allocating the exact size but the size of the
largest SymbolBody.
llvm-svn: 268310
With this it is possible to use chroot/fakechroot to have a completely
reproducible link even when thin archives or linker scripts have
absolute paths.
llvm-svn: 268231
Patch implements one of suggestions from Rafael Ávila de Espíndola,
to fix segfault after section that contains personality being
garbage collected.
Suggestion was just to keep alive all non executable sections
referenced by .eh_frame.
This fixes PR27529.
Differential revision: http://reviews.llvm.org/D19656
llvm-svn: 268228
This patch implements a new design for the symbol table that stores
SymbolBodies within a memory region of the Symbol object. Symbols are mutated
by constructing SymbolBodies in place over existing SymbolBodies, rather
than by mutating pointers. As mentioned in the initial proposal [1], this
memory layout helps reduce the cache miss rate by improving memory locality.
Performance numbers:
old(s) new(s)
Without debug info:
chrome 7.178 6.432 (-11.5%)
LLVMgold.so 0.505 0.502 (-0.5%)
clang 0.954 0.827 (-15.4%)
llvm-as 0.052 0.045 (-15.5%)
With debug info:
scylla 5.695 5.613 (-1.5%)
clang 14.396 14.143 (-1.8%)
Performance counter results show that the fewer required indirections is
indeed the cause of the improved performance. For example, when linking
chrome, stalled cycles decreases from 14,556,444,002 to 12,959,238,310, and
instructions per cycle increases from 0.78 to 0.83. We are also executing
many fewer instructions (15,516,401,933 down to 15,002,434,310), probably
because we spend less time allocating SymbolBodies.
The new mechanism by which symbols are added to the symbol table is by calling
add* functions on the SymbolTable.
In this patch, I handle local symbols by storing them inside "unparented"
SymbolBodies. This is suboptimal, but if we do want to try to avoid allocating
these SymbolBodies, we can probably do that separately.
I also removed a few members from the SymbolBody class that were only being
used to pass information from the input file to the symbol table.
This patch implements the new design for the ELF linker only. I intend to
prepare a similar patch for the COFF linker.
[1] http://lists.llvm.org/pipermail/llvm-dev/2016-April/098832.html
Differential Revision: http://reviews.llvm.org/D19752
llvm-svn: 268178
The aim of this patch is to make it easy to re-run the command without
updating paths in the command line. Here is a use case.
Assume that Alice is having an issue with lld and is reporting the issue
to developer Bob. Alice's current directly is /home/alice/work and her
command line is "ld.lld -o foo foo.o ../bar.o". She adds "--reproduce repro"
to the command line and re-run. Then the following text will be produced as
response.txt (notice that the paths are rewritten so that they are
relative to /home/alice/work/repro.)
-o home/alice/work/foo home/alice/work/foo.o home/alice/bar.o
The command also produces the following files by copying inputs.
/home/alice/repro/home/alice/work/foo.o
/home/alice/repro/home/alice/bar.o
Alice zips the directory and send it to Bob. Bob get an archive from Alice
and extract it to his home directory as /home/bob/repro. Now his directory
have the following files.
/home/bob/repro/response.txt
/home/bob/repro/home/alice/work/foo.o
/home/bob/repro/home/alice/bar.o
Bob then re-run the command with these files by the following commands.
cd /home/bob/repro
ld.lld @response.txt
This command will run the linker with the same command line options and
the same input files as Alice's, so it is very likely that Bob will see
the same issue as Alice saw.
Differential Revision: http://reviews.llvm.org/D19737
llvm-svn: 268169
These would just crash at runtime.
If we ever decide to support rw text segments this should make it easier
to implement as there is now a single point where we notice the problem.
I have tested this with a freebsd buildworld. It found a non pic
assembly file being linked into a .so,. With that fixed, buildworld
finished.
llvm-svn: 268149
We currently don't do a good job of diagnosing inputs that would require
dynamic relocations to be applied to read only segments.
I am about to improve lld in that area, but unfortunately we developed
tests that depend on the current behavior.
To make clear what is actually changing, this first patch just updates
tests to not depend on the current behavior. In most cases this just
means using a rw section instead of a ro one, but that unfortunately
changes many addresses.
llvm-svn: 268145
This patch redefines the default optimization level as 1 and adds
new level 0. In the command line, it is -O0. The flag disables
costly but optional features so that the linker produces semantically
correct but larger output quickly. Currently it only disables
section merging.
This flag is not intended to be used for final production linking.
It is intended to be used in compile-link-test cycle.
Time to link clang with debug info is about 2x faster with the flag.
Head:
13.24 seconds
Output size: 1227189664 bytes
With this patch:
7.41 seconds
Output size: 2490281784 bytes
Differential Revision: http://reviews.llvm.org/D19705
llvm-svn: 268056
Using multiple context used to be a really big memory saving because we
could free memory from each file while the linker proceeded with the
symbol resolution. We are getting lazier about reading data from the
bitcode, so I was curious if this was still a good tradeoff.
One thing that is a bit annoying is that we still have to copy the
symbol names. The problem is that the names are stored in the Module and
get freed when we move the module bits during linking.
Long term I think the solution is to add a symbol table to the bitcode.
That way IRObject file will not need to use a Module or a Context and we
can drop it while still keeping a StringRef to the names.
This patch is still be an interesting medium term improvement.
When linking llvm-as without debug info this patch is a small speedup:
master: 29.861877513 seconds
patch: 29.814533787 seconds
With debug info the numbers are
master: 34.765181469 seconds
patch: 34.563351584 seconds
The peak memory usage when linking llvm-as with debug info was
master: 599.10MB
patch: 600.13MB
llvm-svn: 267921
Relocations against sections with no SHF_ALLOC bit are R_ABS relocations.
Currently we are creating Relocations vector for them, but that is wasteful.
This patch is to skip vector construction and to directly apply relocations
in place.
This patch seems to be pretty effective for large executables with debug info.
r266158 (Rafael's patch to change the way how we apply relocations) caused a
temporary performance degradation for such executables, but this patch makes
it even faster than before.
Time to link clang with debug info (output size is 1070 MB):
before r266158: 15.312 seconds (0%)
r266158: 17.301 seconds (+13.0%)
Head: 16.484 seconds (+7.7%)
w/patch: 13.166 seconds (-14.0%)
Differential Revision: http://reviews.llvm.org/D19645
llvm-svn: 267917
There seems to be no reason to keep st_size of undefined symbols.
This patch removes the member for it. This patch will change outputs
in cases that undefined symbols are copied to output, but I think
this is unimportant.
Differential Revision: http://reviews.llvm.org/D19574
llvm-svn: 267826
Sean Silva