| Age | Commit message (Collapse) | Author |
|---|
|
The XZ file format supports random access if the file was
compressed with the right flags (eg xz --block-size=4096).
Add support for it so that we only decompress the parts
of mini-debug-info that we need.
This reduces mini-debug-info memory usage by 20% without any
specific optimizations which take advantage of the laziness.
(with perfetto system-wide profiling running for a minute)
BM_symbol_find_single_from_sorted (best case lookup scenario)
is 4x faster and uses 5x less memory.
BM_symbol_not_present_from_sorted (worst case lookup scenario)
is 2x faster and uses 3x less memory.
Bug: 110133331
Test: libunwindstack_test
Change-Id: Id067b2d07807463f86ed5fd908b83d079c5a057e
|
|
The symbol name related reads and memory operations
take about half of the symbol name reading cost now.
This CL adds ref-counted read-only shared string cache,
which essentially eliminates all of the costs.
BM_symbol_find_single_many_times is >10% faster on ARM.
(which is 20% if we exclude the fixed ELF loading cost)
Real-world profiles seem even more encouraging.
The extra memory cost is negligible (by definition,
a small fraction of the decompressed mini-debug-info:
specifically, the subset of strtab for hit functions).
Furthermore, this effectively dedups strings when
consecutive unwinds hit similar set of functions.
(perfetto might keep several unwind results live)
Test: m libunwindstack_unit_test
Change-Id: I5cf600bb972fdb9d0f3a57ed0997bead2efa38f4
|
|
This will be used to speed up JIT lookups, by trivially
skipping JIT entries that don't need to be decompressed.
Test: m libunwindstack_unit_test
Change-Id: I6dabc33de245c0b319c19dbfe1f2129987b42f4b
|
|
Reduce verbosity caused by the dual bitness support.
(following pattern that I have found useful in ART)
The code semantics should be unchanged.
Test: m libunwindstack_test
Change-Id: Ib65f913d5a8f5217aeb148a2c96f1a85ccfa4096
|
|
Mark a CIE with a S in its augmentation string as signal frame.
This allows the code to properly handle signal frame data if none
of the signal frame pattern matchers work.
For a signal frame, DwarfSectionImpl<AddressType>::Eval needs to
continue the unwinding even if PC is zero. A zero PC means that the
program has crashed, and we should try to recover the real PC using the
return address on the stack or LR. This behavior is tested by
UnwindOffline.signal_{x86,x86_64}, which modify the libc.so files
so that the signal frame pattern matcher fails and the CIE/FDE
data is used instead.
Test: libunwindstack_test
Change-Id: I4655b070028fd984345311a5e743796f8c30ed36
|
|
The code was not properly getting the variable addresses and using
the offset and address fields of the .data section.
Fix all of that, and update the tests.
Bug: 145162678
Test: Unit tests pass.
Test: ./art/test/run-test --dex2oat-jobs 4 --host --prebuild --compact-dex-level fast --jit --no-relocate --runtime-option -Xcheck:jni 137-cfi
Test: ./art/test/testrunner/testrunner.py -t 137 --host
Change-Id: Ic61c4487334fd2273cda9c56eb1a3b525a03edb7
|
|
The original code assumed that the load bias in the program headers
would be exactly the same as in eh_frame/eh_frame_hdr/debug_frame.
This isn't guaranteed, so add a section bias for use when creating
a DwarfSection. In addtion, make the load bias and section bias
a signed value. There is no reason that this value needs to be positive,
so don't force it to be.
Add a new offline test that has a different load bias in eh_frame than
in the executable load.
Add additional unit tests to verify the load bias values are set properly.
Clean up the tests in ElfInterfaceTest, making all tests names follow the
same convention.
Bug: 141888859
Bug: 142094469
Test: New units and old unit tests pass on host and taimen.
Change-Id: Ib878123ab5545f0f315c749cfe0d27b012d873ee
|
|
This reverts commit 91dc724d5ba73d60cc0f7c2b01e559e899ff3a87.
Reason for revert: Breaks ART tests, reverting to investigate.
Change-Id: I1bb905407e87cbd4f832646651133a9caf6fcfc8
|
|
Update the entries only when the list is modified by the runtime.
Check that the list wasn't concurrently modified when being read.
Bug: 124287208
Test: libunwindstack_test
Test: art/test.py -b --host -r -t 137-cfi
Change-Id: I87ba70322053a01b3d5be1fdf6310e1dc21bb084
|
|
Changes:
- Change GetSoname to always returns a std::string.
- Added new unit tests for the soname printing.
- Modify the GetElf() function to save the same elf when we see rosegment
linkers that split the read-only and read-write across a map. This
avoids creating multiple elf objects for each map.
- Fixed a few offline unwind tests.
Bug: 29218999
Test: Unit tests pass.
Change-Id: Iad7c38b5c2957a8c5fd4ba94ebec335bafcad57d
|
|
Change the GetBuildID function to return a std::string.
Added benchmark to check how long it takes to get the build id from
a file versus an elf object.
Added a way to get an elf without passing in a valid process_memory and
added tests for this.
Test: New unit tests.
Change-Id: I3029019767e0181c758d611fe635bc1bf72d6e8e
|
|
This will be used by heapprofd to allow us to correlate build ids with
memory leaks in libraries and binaries.
Test: m
Test: host libunwindstack_test
Test: run unwind_info against my phone's libc.so (32/64) and compare
to readelf Build ID output.
Bug: 120186412
Change-Id: I3cefd6cce9a8733509bf35b7175eb0f967783477
|
|
Here is the allowable issues with an elf file that will not result in an error:
- The program headers/section headers offset points to unreadable memory.
- Allow missing program header and/or section headers.
- Allow a symbol table section header to point to invalid symbol table values.
There is no real reason to require the elf file be perfect. Everything in
the code has sane defaults, so any missing information won't cause any
problems.
This gets rid of the warning that occurs any time an elf is loaded
from memory. In memory elf files never contain all of the section headers,
and do not contain the symbol table data.
Update tests to test these new cases.
Test: Builds and unit tests all pass.
Change-Id: Iaefe2cd6b6c965a01ed425a112d6afae339f3b78
|
|
In almost all cases, it is faster to read the entire structure rather
than do multiple reads using ReadField. The only case where it would be
slower is if doing a remote unwind and ptrace is the only way to read. In
all other cases, it's a single system call. In the ptrace call, it will be
multiple calls. Given that it is unusual to be forced to use ptrace,
it's better to avoid it.
It also reduces the code complexity to do a single read, and avoids
issues where the code forgets to read the field it needs.
Test: Unit tests pass on host and target.
Change-Id: I7b3875b2c85d0d88115b1776e1be28521dc0b932
|
|
Before, I was using p_vaddr to get the offset into the elf file where
the exidx frame starts. I changed that to use p_offset since this already
has the load bias offset in it and some elf files do not set p_vaddr
properly.
Also, use p_filesz instead of p_memsz, since again, some elf files do
not set p_memsz to the same as p_filesz.
Bug: 110704153
Test: All libbacktrace/libunwindstack unit tests pass.
Test: Randomly unwind process on a walleye.
Test: Verified that this properly dumps and unwinds the shared
Test: library that sets p_vaddr and p_memsz differently.
Change-Id: Ic7b1e5d07439f4636fa02cd884a8727a5737372b
|
|
It turns out that for the dwarf information, if a FDE indicates it's pc
relative, then pc has to be incremented by the load bias. If not, then
it should not be incremented.
Previously, the code always subtracted load bias values from pcs, and assumed
that all fdes were incremented by load bias values. The new code actually
reads the fdes and adjusted the pcs in the fde and in the eh frame hdr so
that load bias values are already handled properly.
In addition, add dumping of arm exidx values in unwind_reg_info. This allowed
verifying that the debug frame in those elf files was being handled properly.
Added a new unit test that only has a debug frame that has a non-zero load
bias and has fde entries that do not have pc relative encoding.
Fix a couple of other small bugs.
Bug: 109824792
Test: All libbacktrace/libunwindstack unit tests pass.
Test: Ran ART 137-cfi test and 004-ThreadStress.
Test: Verify that displaying the fde start and end pc actually match the
Test: real data for fde that have pc relative set, and that don't.
Test: Verified that the unwind information for arm exidx matches the
Test: debug frame data.
Change-Id: I707555286b5cb05df9f25489e8c5ede753cfe0fb
|
|
The dynamic section contained an address, not an offset into the elf
file to indicate where the soname exists. Changed to use the strtab
entries in the section headers to map this address to the actual offset.
Refactor the soname test a bit to make it easier to verify the code.
Bug: 73499044
Test: Passes new unit tests.
Test: Ran unwind_info on the failing shared elf and verified the soinfo
Test: is correct.
Change-Id: I16ba148389bcb9aadd3566fb442dac27f89fe894
|
|
The backtrace offline code uses these error codes to diagnose errors.
In addtion, I've had cases where seeing these errors would help diagnose
failures.
This also allows us to add a few features to indicate why an unwind
terminated (such as max frames exceeded).
Bug: 65682279
Test: Updated unit tests pass.
Change-Id: If82b5092698e8a194016d670efff1320f9b44d50
|
|
Changes:
- New JitDebug class to handle all of the jit gdb interface.
- Add unit tests for all, along with new offline test using debug data.
- Add new Memory type called MemoryOfflineParts that has multiple
MemoryOffline objects to support the offline test.
- Update the tools to use the JitDebug object.
- Modify libbacktrace to use the JitDebug, but only looking in libart.so
and libartd.so.
- Change the Format32Bits to Is32Bit since it's more accurate and I use
it in a different context where original name didn't make sense.
- Add a new function to find global variables in an elf file
(GetGlobalVariable).
- Add a new function to determine if a pc is valid for this elf (IsValidPc).
Bug: 68396769
Test: Ran new unit tests. Added new offline test that uses jit debug data.
Test: Ran art test that generates jit data and verified a crash unwinds
Test: through the jit data.
Change-Id: I6e7ee2f5bab2242028a06feece156dff21c0a974
|
|
- Add a load_bias field in MapInfo so that it can be loaded offline,
and also so it can be cached.
- Add an Add function to the Maps class so that it's possible to manually
create a map.
- Remove the OfflineMaps class since I haven't found a reason for this to
exist.
- Add a pointer to the gnu debugdata compressed section in the interface
itself and modify the step path to try eh_frame, then debug_frame, then
gnu_debugdata. This way arm can add exidx as the last step behind
gnu_debugdata. Add an offline test to verify the order of unwind.
- Fix x86_64_ucontext_t since it was a different size on 32 bit and 64 bit
systems.
Test: Pass new unit tests.
Change-Id: I978b70d6c244bd307c62a29886d24c1a8cb2af23
|
|
Add a static GetLoadBias method to the Elf object that only reads just
enough to get the load bias.
Add a method to MapInfo that gets the load bias. First attempt to get
it if the elf object already exists. If no elf object was created, use
the new static method to get the load bias.
In BacktraceMap, add a custom iterator so that when code dereferences
a map element, that's when the load bias will be retrieved if it hasn't
already been set.
Bug: 69871050
Test: New unit tests, verify tombstones have non-zero load bias values for
Test: libraries with a non-zero load bias.
Change-Id: I125f4abc827589957fce2f0df24b0f25d037d732
|
|
Static executables only have a .eh_frame section and no .eh_frame_hdr
section. Add support for this by rearranging the class hierarchy and
creating a DwarfEhFrameWithHdr class and a DwarfEhFrame class to handle
the different cases.
Add new unit tests for DwarfEhFrame and for the new functionality.
Bug: 68820189
Test: Passes new unit tests, unwinds static executables.
Change-Id: I63d7cb8c52a686e96579a2266e18c0d06bbb6e63
|
|
- Move the load bias stored out of ElfInterface into Elf. For the compressed
sections, the load bias was not the same as the data from the uncompressed
section.
- Move the initialization of the compressed section into Init. It was too easy
to forget to call the init of the compressed section.
- Do not automatically add in load bias to the pc before calling ElfInterface
code. Do all of the pc manipulations in the Elf object.
- Change the interface GetFunctionName code to pass in the load_bias instead
of modifying the pc inside the code.
- Modify the Step function to pass in the elf offset, not add it to the pc.
It is necessary to have two different relative values when executing the
Step: a pc that is relative to the beginning of the elf for the reading data
the actual instructions when trying to determine if this is in a signal
frame, and a pc that is relative to the map for finding the appropriate
unwind information.
- Add a feature to Unwinder so that an unwind can be stopped if it ends up
in map that has a specified suffix. This is so that the ART unwinding
code doesn't require skipping the compressed section. Instead, stop at
if trying to unwind through a known suffix code that means the code is
in java code. This is important because the compressed section data is
not only used by the jave compiled code, so that will continue to work.
- Fix tests for restructuring, add new tests for new functionality.
Test: Ran art test 137-cfi using new unwinder as default.
Test: Ran new unit tests.
Change-Id: I42e658c64c5e14f698ba34944a3043afac967884
|
|
If a function crashes by jumping into unexecutable code, the old method
could not unwind through that. Add a fallback method to set the pc from
the default return address location.
In addition, add a new finished check for steps. This will provide a method
to indicate that this step is the last step. This prevents cases where
the fallback method might be triggered incorrectly.
Update the libbacktrace code to unwind using the new methodology.
Update the unwind tool to use the new unwind methodology.
Add a new option to crasher that calls through a null function.
Create a new object, Unwinder, that encapsulates the a basic unwind. For now,
libbacktrace will still use the custom code.
Added new unit tests to cover the new cases. Also add a test that
crashes calling a nullptr as a function, and then has call frames in
the signal stack.
Bug: 65842173
Test: Pass all unit tests, verify crasher dumps properly.
Change-Id: Ia18430ab107e9f7bdf0e14a9b74710b1280bd7f4
|
|
- Add a method to get the max size of an elf file by reading the
section header offset + size. This will properly map an elf
file embedded into an apk, instead of just mapping in what is done
by the dynamic linker. It does assume that the section headers are
at the end of the elf file.
- Add new tests for the above functionality.
- Update the unwind_symbols tool to take an address for finding a
function instead of dumping the entire symbol table.
Bug: 23762183
Test: Unit tests pass, unwind through the camera process and verify
Test: the GoogleCamera.apk shows some function names.
Change-Id: I00c021680fe1d43b60d652bf91bbf6667d9617be
|
|
- Add namespace unwindstack everywhere so that it's easier for other
code to use the library.
- Move some of the header files into include/unwindstack so that they
can be exposed.
- Modify the headers so that only a limited number need to be exposed.
- Update the tools to use the new headers.
- Add a GetLoadBias() call on the Elf object. This prevents the need
to get the interface object out of the Elf object.
- Move the GetRelPc() call out of the Reg class, to the Elf class. It's
not always the case that a Reg object will be around when you want to
get a relative pc. The tests for this moved to ElfTest.cpp.
Bug: 23762183
Test: Unit tests pass.
Change-Id: Iac609dac1dd90ed83d1a1e24ff2579c96c023bc3
|
