New function to create the process memory object. This allows for
a future where different remote process memory objects could be created
depending on the way remote memory can be created. Even different local
memory objects that access memory without doing any checks.
It also allows MemoryRange objects to share one single process memory object
and could help if the process memory object caches data.
Small changes to MapInfo::CreateMemory to when some errors are detected.
- Always check if the map is a device map, instead of only if the name
is not empty.
- Check if a memory map is readable before creating the memory from process
memory.
Bug: 23762183
Test: Ran unit tests, unwound on device using the new code.
Change-Id: I12a93c2dc19639689a528ec41c67bfac74d431b3
- Change all load_base to load_bias in the test files.
- Add the test files to the backtrace_test.
- Add a function to get the path to the test libraries.
- Change aarch64 to arm64 for offline test code.
- Modify the offline tests so that they can be easily updated when
unwinding on any arch for any other arch is possible.
- Add tests of CreateNew for remote debugging.
Test: Ran unit tests on host and angler.
Change-Id: Id6c5afe73aeb2ac22463dd81f061799fcb1c178b
There are no actual users of the machine type output parameter to
Regs::RemoteGet. The concrete implementations of Regs know what machine
type they represent anyway, so provide an accessor to query.
Test: treehugger
Test: libunwindstack tests on 32/64-bit host, hikey960
Change-Id: Ia25910531d36c41b2b6919f154cfa914aae63117
- Rewrite the Maps::Parse to use open, and a buffer on the stack.
- Rewrite the line parser away from sscanf. The current way sscanf
is used does not catch many malformed lines. In addition, this
new version improves performance by 50% over sscanf on sailfish.
- Add a lot of unit tests for the parser to make sure there are
no problems. In addition, add a special line that was not rejected
with the previous version of the code.
- Add new accessor to get the map at a particular index.
- Add a backtrace benchmark for map creation for both new and old.
This cl results in ~5% speedup and makes the new unwinder map creation
about the same for 64 bit. It's still a bit slower, but not by much.
On 32 bit, we are still about 5% slower than the old creation method,
though.
Bug: 23762183
Test: libunwindstack unit tests pass. Ran the new benchmarks.
Change-Id: Id4431e539f400984e6fad62153fdf4152d518322
As a VNDK-SP module, Android.bp must have 'vndk' tag as well as
'vendor_available: true'.
The 'vndk' tag for VNDK-SP formated as follows:
vndk: {
enabled: true,
support_system_process: true,
},
VNDK-SP modules will be installed both in system/lib(64) as normal
and in system/lib(64)/vndk-sp as a vendor variant.
Bug: 63866913
Test: build and boot with BOARD_VNDK_VERSION=current
Merged-In: I51fe0859f63ad58b7b91909e7d7d4206443228cd
Change-Id: I51fe0859f63ad58b7b91909e7d7d4206443228cd
(cherry picked from commit aeb68e86e4)
Also add a comment to the GetElf function to indicate that it never returns
nullptr.
Also needed to add this library to the a million and one places that the vndk
has hard-coded this data.
Bug: 23762183
Test: Built, nothing uses the new code.
Test: However, I did run backtrace_test using this code, and all tests pass.
Change-Id: Ib270665dcb7a7607075e36d88be76dbde6e2faa8
(cherry picked from commit dc4104b720)
Add NOLINT comment to work around clang-tidy
error in checking macro arguments used in
type expressions.
Bug: 28705665
Test: make with WITH_TIDY=1 WITH_TIDY_CHECKS=-*,misc-macro-* \
WITH_TIDY_FLAGS=-header-filter=system/core/.*
Change-Id: I7619978c1804e151a11a8b0477e80076bcf21cab
Also add a comment to the GetElf function to indicate that it never returns
nullptr.
Bug: 23762183
Test: Built, nothing uses the new code.
Test: However, I did run backtrace_test using this code, and all tests pass.
Change-Id: I252b9c2497e2d3d94347dd6e506170bf50cbfe16
- Change the field name load_base to load_bias (which is what it really is).
- Add a rel_pc field so that callers do not need to compute it themselves.
- Remove the BacktraceMap::GetRelativePc() since nobody should need to
compute this themselves.
Bug: 23762183
Test: Compiles and unit tests pass (debuggerd, libbacktrace).
Change-Id: I2cb579767120adf08c407a58f3c487ee3f2b45fc
Test: Boot on bullhead.
Test: Ran the libbacktrace tests on bullhead.
Test: Added a temporary log message in the signal handler, and ran the
Test: backtrace tests.
Change-Id: I0a6888c9f311af2c8cc7fbb4929315911bd2bb3c
This is used in CallStack.h.
Test: m -j libutils
Test: links
Test: trying to compile with BOARD_VNDK_VERSION := current
Bug: 33241851
Change-Id: I076c9551d356824d6e69b82349b890369bc7eea3
Merged-In: If7258e7b38000a72fd520f725030fb0038f90167
By setting vendor_available, the following may become true:
* a prebuilt library from this release may be used at runtime by
in a later releasse (by vendor code compiled against this release).
so this library shouldn't depend on runtime state that may change
in the future.
* this library may be loaded twice into a single process (potentially
an old version and a newer version). The symbols will be isolated
using linker namespaces, but this may break assumptions about 1
library in 1 process (your singletons will run twice).
Background:
This means that these modules may be built and installed twice --
once for the system partition and once for the vendor partition. The
system version will build just like today, and will be used by the
framework components on /system. The vendor version will build
against a reduced set of exports and libraries -- similar to, but
separate from, the NDK. This means that all your dependencies must
also mark vendor_available.
At runtime, /system binaries will load libraries from /system/lib*,
while /vendor binaries will load libraries from /vendor/lib*. There
are some exceptions in both directions -- bionic(libc,etc) and liblog
are always loaded from /system. And SP-HALs (OpenGL, etc) may load
/vendor code into /system processes, but the dependencies of those
libraries will load from /vendor until it reaches a library that's
always on /system. In the SP-HAL case, if both framework and vendor
libraries depend on a library of the same name, both versions will be
loaded, but they will be isolated from each other.
It's possible to compile differently -- reducing your source files,
exporting different include directories, etc. For details see:
https://android-review.googlesource.com/368372
None of this is enabled unless the device opts into the system/vendor
split with BOARD_VNDK_VERSION := current.
Bug: 33241851
Test: build and flash internal marlin
Test: m -j libbacktrace
Test: build with BOARD_VNDK_VERSION := current
(cherry picked from commit 4c0e956c76)
Merged-In: Idab4880e011416ebc40b225205c30fb5ed8661db
Change-Id: Idab4880e011416ebc40b225205c30fb5ed8661db
Also make backtrace_testlib.c C++.
Bug: 34468756
Test: Ran new unit test on host x86, host x86_64, angler (32 bit, 64 bit),
Test: fugu.
Change-Id: Ia810f596c2df56cd56e8ab17e6c19be0c48d737f
It's possible that a device map has memory controlled by a single entry
device driver. Thus, you can deadlock if a process is touching that
device memory and we try to unwind it and also touch that device memory.
Simply skip any attempts to step through, or get function names from
device memory maps.
Bug: 36130325
Test: Ran new unit tests, ran bionic unit tests, ran art ThreadStress.
Change-Id: Ibc62d7ec8106c619ee08968f05e04aea55d7cbfa
Some extra initialization needs to occur before calling the local/remote
GetFunctioneName.
Also, cleanup the test code a bit:
- Make all local functions static.
- Create a common function to create and finish remote processes.
- Remove unused function.
- Fix the formatting a bit by making it match clang format.
Test: Ran the unit tests.
Change-Id: I998bed1378d582df59fdf9263df6f208db5d795a
Make the default the 4 tab space version.
Add a link to the 2 space version for libbacktrace.
The only other difference between the 4 space and 2 space tab version
is that the 2 space tab version allows short functions on a single line.
This is for things like constructors, short destructors, or accessor
functions. I believe this is closer to the Google C++ style guide.
Bug: 36046320
Test: NA
Change-Id: Ida08db18902eeb101f42869dd5590182529d54ef
There was a problem where errno could be set in a signal handler
and cause bugs if other system calls were interrupted.
There isn't strong evidence this is causing any issues, but add this
proactively.
Bug: 31448909
Test: Ran the backtrace unit tests, backtraces few random processes,
Test: forced the ANR path for some zygote based processes.
Test: Ran the art ThreadStress test a few times.
Change-Id: I5abc9e330a5e86cea7015e2362f66044c6bc37d0
Make libbacktraceoffline thread-safe by protecting the global variable with
a lock. So it can be used in a multi-thread environment, like by
simpleperf inplace sampling.
Bug: http://b/30974760
Test: no
Change-Id: I4699bf15dfa69ac75faeb4e79a73fb3af0f08dfc
1. In /system/lib/libart.so, the ranges of functions covered by .eh_frame
and .ARM.exidx overlap with each other. Currently, we don't check .eh_frame
if the address is in the ranges of functions covered by .ARM.exidx. It leads
to a wrong unwinding result. So change the logic to first try .eh_frame and
.debug_frame, if it fails, then try .ARM.exidx.
2. Add test for Item 1.
Bug: http://b/35251937
Test: run backtrace_test.
Change-Id: Ic412705617c300bc69b4ece178cc2e1bb3eb765b
The libbacktrace common flags don't need the flags from
LLVM_HOST_BUILD_MK, they are added directly to libbacktrace_offline.
The -fno-omit-frame-pointer is unnecessary, as we no longer import
flags only used to build LLVM into everything that depends on LLVM.
Removing -Wno-extern-c-compat doesn't cause any warnings.
Test: mma -j
Change-Id: Ia01380e8572c3d84268843291a54ec8c932a327f
See build/soong/README.md for more information.
Reapplies I61c49fb62fdeafb66db25b8f97621239b8fd6e75 after fixing
an unused parameter warning on mips builds.
Test: mma -j
Test: backtrace_test
Change-Id: I6dc353a274cafcdb05ee866b41c3c2731612941c
libbacktrace_offline was inheriting -Wno-unused-parameter from llvm's
makefiles. Fix an unused parameter warning when compiling for an
unsupported architecture, and return false since *value wasn't
written.
Test: lunch aosp_mips-eng && mma -j
Change-Id: I7df990f01e1025ca17ec3717107df56d2349f3cb
1. Read .ARM.exidx and .ARM.extab, provides these data to libunwind
through AccessMem.
2. In FindProcInfo, search .ARM.exidx for the idx entry of function
containing ip.
3. Add structures to store .ARM.exidx info in DebugFrameInfo, and
clean up the structure.
4. Add tests to test each unwind information (.eh_frame, .debug_frame,
.gnu_debugdata, .ARM.exidx) separately.
Bug: http://b/30974693
Test: run backtrace_test.
Test: run simpleperf runtest.py.
Change-Id: I9377a54379158bd35860efcf0e011423be7350d4
Bug: http://b/28866258
Remove dependency of backtrace_test on libLLVM. Also clean up
specification for libbacktrace_offline module. Since it is a static
library now, dependencies are only needed to automatically include
necessary headers.
Test: 32-bit and 64-bit backtrace_test passes on host and on Angler.
mma in system/extras/simpleperf, cts/tests/tests/simpleperf
m test-art-host
Change-Id: I76b0db14d067fa7b140a99edd237370ac04fe873
This reverts commit 279843e139.
Bug: 32806783
Change-Id: I61df26d979a8139a571ff5a6442fb5499ca20547
Test: Verified that the stack and read data works with the revert.
Bug: http://b/31320715
Test: No regression in 32-bit and 64-bit backtrace_test on host and
angler.
Change-Id: Ic5c5511e89ca7707ed9c13d5ba0539e09310dfe3
Host builds were compiling without -Wl,--no-undefined. Add missing
host ldlibs so that -Wl,--no-undefined can be added again.
Test: m -j host
Bug: 32305815
Change-Id: I0ceabc3f4ec74e2f737afaac58810c5d53d6ccb6
Point to log/log.h where necessary, define LOG_TAG where necessary.
Accept that private/android_logger.h is suitable replacement for
log/logger.h and android/log.h.
Correct liblog/README
Effectively a cleanup and controlled select revert of
'system/core: drop or replace log/logger.h' and
'system/core: Replace log/log.h with android/log.h'.
Test: compile
Bug: 30465923
Change-Id: Ic2ad157bad6f5efe2c6af293a73bb753300b17a2
Modify BacktracePtrace::Read to use the process_vm_readv syscall
instead of doing multiple ptrace calls.
Test: Ran the backtrace unit tests on host and on x86_64 emulator.
Change-Id: I24c2787f6713d8eb44312f3aa751d1f6d9d34719
Should use android/log.h instead of log/log.h as a good example
to all others. Adjust header order to comply with Android Coding
standards.
Test: Compile
Bug: 26552300
Bug: 31289077
Change-Id: I33a8fb4e754d2dc4754d335660c450e0a67190fc
libziparchive-host is no longer necessary, and mixing libziparchive-host
and libziparchive can cause ODR violations.
Change-Id: I27b2cc9974000ee7bf1d1ac15fd7b069862d9b01
* Use const reference type for parameters to avoid unnecessary copy.
* Suppress warning of not using faster overloaded string find function.
Bug: 30407689
Bug: 30411878
Change-Id: I6cfdbbd50cf5e8f3db6e5263076d3a17a9a791ee
Test: build with WITH_TIDY=1
Merged-In: Ie79dbe21899867bc62031f8618bb1322b8071525
If the underlying local map changes, it's possible for multiple
threads to try and modify the map data associated with the UnwindLocalMap
object. Add a lock when generating the local map to avoid this problem.
In addition, add a read lock whenever any caller gets the maps iterator.
Updated all iterator callers to make this lock.
Bug: 29387050
Change-Id: Ie34822c3d8fd3bdb3dd126aeeb399969c36508c1
LOCAL_LDLIBS should only be used with -l flags for NDK and some host
libraries. Other uses should use LOCAL_SHARED_LIBRARIES.
LOCAL_LDFLAGS should not contain -l entries, they should be in
LOCAL_LDLIBS instead.
LOCAL_CFLAGS should not contain -I entries, those should use
LOCAL_C_INCLUDES instead.
Change-Id: Ic901aa14651572ac682166becd4f0cd86baecc34
As device may not have libbacktrace_offline shared library, simpleperf
no longer rely on it. So there is no reason to keep libbacktrace_offline
shared library.
Bug: 28152982
Change-Id: If8e31b1434646dbbfed60264eb25f034bc5d6946
Remove the logging of an error if a thread disappears before the unwind
can begin. This can happen, so allow the caller to determine if this
is really a problem worth logging.
Bug: 27449879
(cherry picked from commit 206a3b9798)
Change-Id: If9e7cfeb6eb7b122679a734c1a9eacee8354ef18
Remove the logging of an error if a thread disappears before the unwind
can begin. This can happen, so allow the caller to determine if this
is really a problem worth logging.
Bug: 27449879
Change-Id: Ie81718d53fb0e519fa0a7db9fd5f314b72bfa431
Sometimes, the new pss is significantly less than the stable pss. Do
not fail if this case occurs.
Change-Id: Ic7da87275a821818e375d05cac10e84c62710c5e
When creating an UnwindMapLocal fails in the Build() function call,
the destructor for UnwindMap is called. Unfortunately, the map_cursor_
member variable has not been initialized, so the call to destroy it
winds up operating on garbage data.
Part of this is a result of a bad class hierarchy, so this refactors
the classes slightly, and properly initializes the map_cursor_ member
variable in the base class.
Bug: 26931578
Change-Id: I885596bf65e4ef63559cee2c56cd41576d5ecc1b
BacktraceOffline is the only code that uses libLLVM. Create
a new shared library, backtrace_offline.so, that references this
so that including libbacktrace.so doesn't wind up pulling in libLLVM.
Change-Id: I90c502d833f6c464c3810dffde52a0c6da241cce
The check_unreadble_elf_* tests only create a readable map. I changed
libunwind to disallow unwinding through non-executable maps, so make
the maps executable too.
Bug: 26589772
Change-Id: Idbe6426ac4d3243779b819c8f1aebdcdacb0669f
Before, an anonymous map wound up printing the pc as relative.
Unfortunately, this meant that it was impossible to tell the actual
pc. The new code prints the map name as <anonymous:map_start> and
still prints the pc as relative.
In addition, add the start of the map for map names that begin with a
'[' character.
Bug: 25844836
Change-Id: Ie0b6149dde258fe13f0e5a3e5739d85374512f4b
This change makes sure -fno-omit-frame-pointer flag is not
overrided by flags set in $(LLVM_HOST_BUILD_MK).
Change-Id: I53037b0eab6bd1272cec0f5c9caa058ad1a55419
(This might want to be unified with the very similar code in bionic, but
that's a job for another day.)
Change-Id: I5dade498f781305fa9909ee01f1f4bebc537c169
This reverts commit ab5e583327.
And statically link llvm libraries on host to remove dependency
on 32-bit llvm shared library which is not included in the prebuilt.
Bug: 22229391
Change-Id: I8210687655ee1809fd820ab2a6ca5dfaf3f9096d
Using llvm makes libbacktrace stops building for 32 bit on linux host x86_64.
This reverts commit c9c6b6ee77.
Change-Id: Ieb807db881d7bb3132cd09e25fd08c28766ca0a8
If the signal handler doesn't fire in the given time when trying to unwind
a thread, put on a logging handler. This prevents crashes if the signal
does eventually fire.
Bug: 23783762
(cherry picked from commit d7226f9a14)
Change-Id: Ib990a06733cc93717752ab4998f4ae26afd7e249
If the signal handler doesn't fire in the given time when trying to unwind
a thread, put on a logging handler. This prevents crashes if the signal
does eventually fire.
Bug: 23783762
Change-Id: Ib7abb36b71d079a7043117697c41b535319586fd
The dlopen of a shared library in an apk results in large map offsets.
Unfortunately, the current way that the frame data is printed, it's
impossible to tell what the relative pc is relative to. With the
addition of the offset, it's possible to figure out what the relative
pc actually references.
Bug: 23348999
(cherry picked from commit e0ab23223a)
Change-Id: I950f92c1cb29ee05eed777f47453efa03318cf3e
The dlopen of a shared library in an apk results in large map offsets.
Unfortunately, the current way that the frame data is printed, it's
impossible to tell what the relative pc is relative to. With the
addition of the offset, it's possible to figure out what the relative
pc actually references.
Bug: 23348999
Change-Id: Ia51b669ea3f810158cfd0d71d9ae89bf9a3170d3
Upstream clang has a new diagnostic, '-Wmismatched-new-delete' to
generate a warning (that becomes an error with -Werror) when delete is
used to deallocate pointers created with 'new[]'. This patch fixes
'delete's that trigger this warning/error.
Change-Id: I7551e856b4b13bfae4cb78b5664d8652322683cf
This error message has no meaning and makes it look like an unwinding
failure occurred. These messages are showing up more often now
that a lot more of debuggerd is using the ReadWord and Read calls.
If a higher level function wants to indicate there is an error, then
it can add a more meaningful error.
Bug: 21818730
(cherry picked from commit 35c8136d96)
Change-Id: Ie6e20fbc18702fde935f2cf8cbf1b297fda43282
This error message has no meaning and makes it look like an unwinding
failure occurred. These messages are showing up more often now
that a lot more of debuggerd is using the ReadWord and Read calls.
If a higher level function wants to indicate there is an error, then
it can add a more meaningful error.
Bug: 21818730
Change-Id: I83aca9cf241aee99e7425059b1b6e0d229c08a5e
If the signal to dump a thread is never delivered, then it's
possible for a deadlock. The signal handler is responsible for
unlocking and deleting the ThreadEntry created for the pid/tid
combination. This means if the signal is lost, the ThreadEntry
gets stuck locked and never deleted. If a second attempt to get
a backtrace of this thread occurs, there is a deadlock.
Also, decrease the timeout from 10 seconds to 5 seconds. The original
10 seconds was because the unwind was actually done in the signal
handler. Now the signal handler does nothing but copy the ucontext
structure and let the caller do the unwind.
Bug: 21086132
(cherry picked from commit 2d09171758)
Change-Id: I414c500eb08983a5017caf3fce4f499465575a9d
If the signal to dump a thread is never delivered, then it's
possible for a deadlock. The signal handler is responsible for
unlocking and deleting the ThreadEntry created for the pid/tid
combination. This means if the signal is lost, the ThreadEntry
gets stuck locked and never deleted. If a second attempt to get
a backtrace of this thread occurs, there is a deadlock.
Also, decrease the timeout from 10 seconds to 5 seconds. The original
10 seconds was because the unwind was actually done in the signal
handler. Now the signal handler does nothing but copy the ucontext
structure and let the caller do the unwind.
Bug: 21086132
Change-Id: Idc735dbf6147ec879d35bd4f034c5d227e26a98d
Under some conditions, /proc/<pid>/maps might return nothing. If we
try and unwind in this case, we'll crash. Check this case and fail
the unwind.
Add checks that no other functions try and use map_ without
checking for nullptr.
Add logging when an unwind fails so it's clear what happened.
Bug: 21162746
Change-Id: I56ce51dda0cfc9db20475a441f118108196aa07c
(cherry picked from commit 30c942cf10)
Under some conditions, /proc/<pid>/maps might return nothing. If we
try and unwind in this case, we'll crash. Check this case and fail
the unwind.
Add checks that no other functions try and use map_ without
checking for nullptr.
Add logging when an unwind fails so it's clear what happened.
Bug: 21162746
Change-Id: I56ce51dda0cfc9db20475a441f118108196aa07c
Also simplifies the Darwin implementation of gettid, because
apparently libbacktrace had a better way of doing it.
Bug: 19517541
(cherry picked from commit 23f750b068)
Change-Id: I2f888e8ed7a2f5719973786cbcbb399a81867ee9
The BacktracePtrace::Read function crashes if the number of bytes to
read is less than the number of bytes needed to align the read to
a word_t boundary.
Fix this and add a test for this case.
Change-Id: I50808849ece44928f65dba1d25309e3885c829a2
(cherry picked from commit 944f417ccb)
The BacktracePtrace::Read function crashes if the number of bytes to
read is less than the number of bytes needed to align the read to
a word_t boundary.
Fix this and add a test for this case.
Change-Id: I50808849ece44928f65dba1d25309e3885c829a2
The new linker relocation packing support uses non-zero load bases,
so we better handle them properly.
Also print out the load base for a map if it's non-zero.
Bug: 20687795
(cherry picked from commit 329ed7dae4)
Change-Id: Ibc37d8c8bb032820dca4e7531184349ba6d402d2
The new linker relocation packing support uses non-zero load bases,
so we better handle them properly.
Also print out the load base for a map if it's non-zero.
Bug: 20687795
Change-Id: Iec2d1db2051e7b4a278c1dfa57d745128a7f2974
Also simplifies the Darwin implementation of gettid, because
apparently libbacktrace had a better way of doing it.
Change-Id: I9e14f9176b167405a5aaa5b8a0db0327c7126271
When doing a local unwind, do not include the frames that come
from either libunwind or libbacktrace.
Bug: 11518609
Change-Id: I0ec8d823aebbfa0903e61b16b7e5663f3fd65e78
The object hierarchy was confusing and convoluted. This removes
a lot of unnecessary code, and consolidates the BacktraceCurrent
and BacktraceThread code into BacktraceCurrent.
Change-Id: I01c8407d493712a48169df49dd3ff46db4a7c3ae
The latest clang (r230699) does not allow SP/PC to be declared in inline asm
lists. This is a problem for libraries that are attempting to save all
register state.
Change-Id: I1ce80530f8be85fd74e949d0c65b60080e0aaf2a
Included tests for this new feature.
Changed the NULLs to nullptr in backtrace_test.
Changed UniquePtr to std::unique_ptr in backtrace_test.
Change-Id: I92375465b8f8ba84589834cc162db5915bf1be81
The backtrace structure used to include a pointer to a backtrace_map_t
that represented the map data for a particular pc. This introduced a
race condition where the pointer could be discarded, but the backtrace
structure still contained a pointer to garbage memory. Now all of the map
information is right in the structure.
Bug: 19028453
Change-Id: If7088a73f3c6bf1f3bc8cdd2bb4b62e7cab831c0
Switch to the better supported pthread_cond to handle the Wait/Wake
functions.
Also, increase the number of simultaneous threads in the thread tests.
Bug: 18381207
(cherry picked from commit db44538387)
Change-Id: Id326a7a7b92cb61573def3f761597c40f3ef2f4b
