Large part of this change is required due to
https://code.google.com/p/android/issues/detail?id=61799
dlsym() crashes when symbol resolution fails, which means
we have to limit the interceptor list instead of relying on
runtime detection.
There are minor differencies in system headers, too.
llvm-svn: 212273
With this change all values passed through blacklisted functions
become fully initialized. Previous behavior was to initialize all
loads in blacklisted functions, but apply normal shadow propagation
logic for all other operation.
This makes blacklist applicable in a wider range of situations.
It also makes code for blacklisted functions a lot shorter, which
works as yet another workaround for PR17409.
llvm-svn: 212268
With this change all values passed through blacklisted functions
become fully initialized. Previous behavior was to initialize all
loads in blacklisted functions, but apply normal shadow propagation
logic for all other operation.
This makes blacklist applicable in a wider range of situations.
It also makes code for blacklisted functions a lot shorter, which
works as yet another workaround for PR17409.
llvm-svn: 212265
See https://code.google.com/p/address-sanitizer/issues/detail?id=299 for the
original feature request.
Introduce llvm.asan.globals metadata, which Clang (or any other frontend)
may use to report extra information about global variables to ASan
instrumentation pass in the backend. This metadata replaces
llvm.asan.dynamically_initialized_globals that was used to detect init-order
bugs. llvm.asan.globals contains the following data for each global:
1) source location (file/line/column info);
2) whether it is dynamically initialized;
3) whether it is blacklisted (shouldn't be instrumented).
Source location data is then emitted in the binary and can be picked up
by ASan runtime in case it needs to print error report involving some global.
For example:
0x... is located 4 bytes to the right of global variable 'C::array' defined in '/path/to/file:17:8' (0x...) of size 40
These source locations are printed even if the binary doesn't have any
debug info.
This is an ABI-breaking change. ASan initialization is renamed to
__asan_init_v4(). Pre-built libraries compiled with older Clang will not work
with the fresh runtime.
llvm-svn: 212188
The former used to crash with a null deref if it was given a not owned pointer,
while the latter returned 0. Now they both return 0. This is still not the best possible
behavior: it is better to print an error report with a stack trace, pointing
to the error in user code, as we do in ASan.
llvm-svn: 212112
Origin history should only be recorded for uninitialized values, because it is
meaningless otherwise. This change moves __msan_chain_origin to the runtime
library side and makes it conditional on the corresponding shadow value.
Previous code was correct, but _very_ inefficient.
llvm-svn: 211700
Summary: The patch supports both the clang cross-compiler and native compiler
Patch by Kumar Sukhani <Kumar.Sukhani@imgtec.com>
Test Plan:
Kumar had the following asan test results when compiled on a MIPS board:
Expected Passes : 96
Expected Failures : 2
Unsupported Tests : 84
Unexpected Passes : 4
Unexpected Failures: 19
The list of unexpected failures can be found in the review.
Reviewers: kcc, petarj, dsanders
Reviewed By: kcc
Subscribers: farazs, kcc, llvm-commits
Differential Revision: http://reviews.llvm.org/D4208
llvm-svn: 211587
Storage Class 3 is static storage. These symbols need to be marked as external
(storage class 2) so that they can be referenced. Note that this external is
not the same as ELF "external" visibility, which is indicated by DLL Storage
Class (i.e. __declspec(dllexport) or __declspec(dllimport)).
llvm-svn: 211428
Add a test for the frexp() interceptor.
Annotate the interceptors that may potentially corrupt stack IDs of freed buffers with FIXME comments.
llvm-svn: 211153
Multiplication by an integer with a number of trailing zero bits leaves
the same number of lower bits of the result initialized to zero.
This change makes MSan take this into account in the case of multiplication by
a compile-time constant.
We don't handle the general, non-constant, case because
(a) it's not going to be cheap (computation-wise);
(b) multiplication by a partially uninitialized value in user code is
a bad idea anyway.
Constant case must be handled because it appears from LLVM optimization of a
completely valid user code, as the test case in compiler-rt demonstrates.
llvm-svn: 211092
When possible, use Thumb or Thumb-2 over ARM instructions. This is particularly
important for pure-Thumb environments (e.g. Windows on ARM). Although, it is
possible to conditionalise this for that target specifically, this is available
on most newer ARM CPUs, and the code remains compatible with older CPUs with no
adverse effects. It therefore feels better to always prefer Thumb when
possible.
llvm-svn: 211032
Init-order and use-after-return modes can currently be enabled
by runtime flags. use-after-scope mode is not really working at the
moment.
The only problem I see is that users won't be able to disable extra
instrumentation for init-order and use-after-scope by a top-level Clang flag.
But this instrumentation was implicitly enabled for quite a while and
we didn't hear from users hurt by it.
llvm-svn: 210924
Clang's lit cfg already detects the currently selected SDK via
"xcrun --show-sdk-path". The same thing should be done for compiler-rt tests,
to make them work on recent OS X versions. Instead of duplicating the detection
code, this patch extracts the detection function into a lit.util method.
Patch by Kuba Brecka (kuba.brecka@gmail.com),
reviewed at http://reviews.llvm.org/D4072
llvm-svn: 210534
Instructions from __nodebug__ functions don't have file:line
information even when inlined into no-nodebug functions. As a result,
intrinsics (SSE and other) from <*intrin.h> clang headers _never_
have file:line information.
With this change, an instruction without !dbg metadata gets one from
the call instruction when inlined.
Fixes PR19001.
llvm-svn: 210459
Use caller pc of __sanitizer_cov_module_init to figure out
when 2 sequential calls are from the same module; skip
.sancov.map file update in this case.
llvm-svn: 210267
This sound like a good idea in general.
Also, without this on Android we get add_lit_testsuite() with empty DEPENDS
list, and it does not work well.
llvm-svn: 210257
Reset coverage data on fork().
For memory-mapped mode (coverage_direct=1) this helps avoid loss of data
(before this change two processes would write to the same file simultaneously).
For normal mode, this reduces coverage dump size, because PCs from the parent
process are no longer inherited by the child.
llvm-svn: 210180
No longer need to set ANDROID if COMPILER_RT_TEST_TARGET_TRIPLE is
arm-linux-androideabi.
No need to set ANDROID_COMMON_FLAGS. These flags are already in
CMAKE_CXX_FLAGS which are used in try_compile().
llvm-svn: 210053
Place constants into .rdata if targeting ELF or COFF/PE. This should be
functionally identical, however, the data would be placed into a different
section. This is purely a cleanup change.
llvm-svn: 209986
Make the whitespace a bit more uniform in the various assembly routines. This
also makes the assembly files a bit more uniform on the ARM side by explicitly
stating that it is using the unified syntax and that the contents of the code is
in the text section (or segment). No functional change.
llvm-svn: 209985
Add a script that is used to deflake inherently flaky tsan tests.
It is invoked from lit tests as:
%deflake %run %t
The script runs the target program up to 10 times,
until it produces a tsan warning.
llvm-svn: 209898
The optimization is two-fold:
First, the algorithm now uses SSE instructions to
handle all 4 shadow slots at once. This makes processing
faster.
Second, if shadow contains the same access, we do not
store the event into trace. This increases effective
trace size, that is, tsan can remember up to 10x more
previous memory accesses.
Perofrmance impact:
Before:
[ OK ] DISABLED_BENCH.Mop8Read (2461 ms)
[ OK ] DISABLED_BENCH.Mop8Write (1836 ms)
After:
[ OK ] DISABLED_BENCH.Mop8Read (1204 ms)
[ OK ] DISABLED_BENCH.Mop8Write (976 ms)
But this measures only fast-path.
On large real applications the speedup is ~20%.
Trace size impact:
On app1:
Memory accesses : 1163265870
Including same : 791312905 (68%)
on app2:
Memory accesses : 166875345
Including same : 150449689 (90%)
90% of filtered events means that trace size is effectively 10x larger.
llvm-svn: 209897
You can expect the sanitizers to be built under any of the following conditions:
1) CMAKE_C_COMPILER is GCC built to cross-compile to ARM
2) CMAKE_C_COMPILER is Clang built to cross-compile to ARM (ARM is default target)
3) CMAKE_C_COMPILER is Clang and CMAKE_C_FLAGS contains -target and --sysroot
Differential Revision: http://reviews.llvm.org/D3794
llvm-svn: 209835
The new storage (MetaMap) is based on direct shadow (instead of a hashmap + per-block lists).
This solves a number of problems:
- eliminates quadratic behaviour in SyncTab::GetAndLock (https://code.google.com/p/thread-sanitizer/issues/detail?id=26)
- eliminates contention in SyncTab
- eliminates contention in internal allocator during allocation of sync objects
- removes a bunch of ad-hoc code in java interface
- reduces java shadow from 2x to 1/2x
- allows to memorize heap block meta info for Java and Go
- allows to cleanup sync object meta info for Go
- which in turn enabled deadlock detector for Go
llvm-svn: 209810
The refactoring makes suppressions more flexible
and allow to suppress based on arbitrary number of stacks.
In particular it fixes:
https://code.google.com/p/thread-sanitizer/issues/detail?id=64
"Make it possible to suppress deadlock reports by any stack (not just first)"
llvm-svn: 209757
This way does not require a __sanitizer_cov_dump() call. That's
important on Android, where apps can be killed at arbitrary time.
We write raw PCs to disk instead of module offsets; we also write
memory layout to a separate file. This increases dump size by the
factor of 2 on 64-bit systems.
llvm-svn: 209653
Call it "libclang_rt.builtins-<arch>.a" to be consistent
with sanitizers/profile libraries naming. Modify Makefile
and CMake build systems and Clang driver accordingly.
Fixes PR19822.
llvm-svn: 209473
Generalize StackDepot and create a new specialized instance of it to
efficiently (i.e. without duplicating stack trace data) store the
origin history tree.
This reduces memory usage for chained origins roughly by an order of
magnitude.
Most importantly, this new design allows us to put two limits on
stored history data (exposed in MSAN_OPTIONS) that help avoid
exponential growth in used memory on certain workloads.
See comments in lib/msan/msan_origin.h for more details.
llvm-svn: 209284
This change also enables asm instrumentation in asan tests that was
accidentally disabled yearlier, and adds a sanity test for that.
Patch by Yuri Gorshenin.
llvm-svn: 209282
For Linux/x86-64, pointers passed to internal_syscall should be casted
to uptr first. Otherwise, they won't be properly extended to 64-bit for
x32.
Patch by H.J. Lu
llvm-svn: 209278
The patch adds better target_triple and target_arch defaults for lit tests,
which allows us to XFAIL tests based on architecture.
Was:
target_triple = LLVM_DEFAULT_TARGET_TRIPLE
target_arch = HOST_ARCH
Now:
target_triple = COMPILER_RT_TEST_TARGET_TRIPLE
, otherwise LLVM_DEFAULT_TARGET_TRIPLE
target_arch = first item in COMPILER_RT_TEST_TARGET_TRIPLE
Differential Revision: http://reviews.llvm.org/D3855
llvm-svn: 209256
Summary:
Sandboxed code may now pass additional arguments to
__sanitizer_sandbox_on_notify() to force all coverage data to be dumped to a
single file (the default is one file per module). The user may supply a file or
socket to write to. The latter option can be used to broker out the file writing
functionality. If -1 is passed, we pre-open a file.
llvm-svn: 209121
The instrprofile-write-file-only test was failing on the builtbots. The runtime
library initialization is explicitly being discarded to avoid the atexit hook.
However, this would also prevent the filename from being initialized. Thus,
when the write file was invoked, the filename would not be setup, and the test
would fail as the profiling data would never be written out. Explicitly
initialize the filename to ensure that the data is written out when requested.
This should hopefully finally get the build bots all green again.
llvm-svn: 209099
dlopen expects that the flags contains RTLD_LAZY or RTLD_NOW. RTLD_LAZY will
resolve the symbol on first use, and is generally preferred due to efficiency.
Add this to fix the test failure on the build bots.
llvm-svn: 209098
Add logging to report any failures that may occur on calls to libdl. Without
this, it is difficult to identify the actual problem if the test fails.
llvm-svn: 209097
Add an explicit link against libdl. libdl may not be indirectly pulled on some
Linux hosts. Explicitly link against it. This should hopefully improve the
state of the build bots.
llvm-svn: 209096
Extend the function definition macros further to support COFF object emission.
The function definition in COFF includes the type and storage class in the
symbol definition context. This is needed to make the assembly routines
possible to be built for COFF environments (i.e. Windows).
llvm-svn: 209095
Rename the HIDDEN_DIRECTIVE macro to HIDDEN and give it a parameter providing
the name of the symbol to be given hidden visibility. This makes the macros
more amenable to COFF.
llvm-svn: 209094
Shared objects are hard. After this commit, we do the right thing when
profiling two separate shared objects that have been dlopen'd with
`RTLD_LOCAL`, when the main executable is *not* being profiled.
This mainly simplifies the writer logic.
- At initialization, determine the output filename and truncate the
file. Depending on whether shared objects can see each other, this
may happen multiple times.
- At exit, each executable writes its own profile in append mode.
<rdar://problem/16918688>
llvm-svn: 209053
These tests were XPASS-ing on Linux bots creating Mach-O, which makes
sense, since the real difference is the object format.
I'm hoping a short-term fix to get these tests passing on ELF is to
create two copies of the runtime -- one built with -fPIC, and one
without. A follow-up patch will change clang's driver to pick between
them depending on whether `-shared` is specified.
llvm-svn: 208947
According to the buildbots, the new features for shared objects don't
work on ELF since it requires an -fPIC when building the profile
library. XFAIL these tests for now.
It's possible we'll have to build two versions of the profile library on
Linux (one with -fPIC and one without), but it's not clear to me exactly
how to resolve this.
llvm-svn: 208946
Change the API of the instrumented profiling library to work with shared
objects.
- Most things are now declared hidden, so that each executable gets
its own copy.
- Initialization hooks up a linked list of writers.
- The raw format with shared objects that are profiled consists of a
concatenated series of profiles. llvm-profdata knows how to deal
with that since r208938.
<rdar://problem/16918688>
llvm-svn: 208940
for sanitizers to pass the C++ compilation and exe linking flags through
from the host CMake configuration. We pass the target flags afterward,
allowing them to trump flags as needed. This is particularly important
when the flags direct Clang, even the just-built-Clang, toward the
standard library, linker, and other tools to use.
llvm-svn: 208896
User-visible instances of xdr_ops always seem to be allocated statically, and
don't need unpoisoning. Also, it's size differs between platforms.
llvm-svn: 208851
Evgeniy Stepanov