LLGS debugging is outputting different thread stop reasons than
local linux debugging. The stop reasons are reasonable so I've left
left them alone. Might update them to match darwin in the future.
llvm-svn: 227920
NativeProcessLinux::MonitorSignal was automatically resuming threads
that stopped due to a signal. This is inconsistent with the
behavior of lldb and gdb. This change removes the automatic resume.
Fixes
TestSendSignal.py
TestSignalsAPI.py
if PLATFORM_LINUX_FORCE_LLGS_LOCAL is in the environment vars.
llvm-svn: 227918
Note this code path should not happen - it implies a bug in another part of
the code. For the thread to receive the stop signal as it is handled, the
and for it to already have a stop reason, it implies the kernel was able to
tell the thread that it stopped while it was stopped. More likely this
seems to indicate a bug where an actual thread start was not getting correctly
logged. If it does get hit, we'll want to understand the sequence to figure
out if it is truly legitimate or if it implies another bug.
llvm-svn: 227916
It looks like Shawn's fix addresses what the initial hijacking was trying
to accomplish per conversations with Greg and Jim. The hijacking was
causing several tests to hang (#61, #62, #63, #64, #67, possibly more).
These tests now just fail rather than hang with this modification.
llvm-svn: 227914
* When the thread state coordinator is told to skip sending a stop request
for a running thread that is ignored (e.g. the thread that steps in a
step operation is technically running and should not have a stop sent
to it, since it will stop of its own accord per the kernel step operation),
ensure the deferred signal notification logic still waits for the
skipped thread. (i.e. we want to defer the notification until the
stepping thread is indeed stopped, we just don't want to send it a tgkill).
* Add ThreadStateCoordinator::RequestResumeAsNeeded(). This variant of the
RequestResume() method does not call the error function when the thread
is already running. Instead, it just logs that the thread is already
running and skips the resume operation. This is useful for the case of
vCont;c handling, where we tell all threads that they should be running.
At the place we're calling, all we know is "we want this thread running if
it isn't already," and that's exactly what this command does.
* Formatting change (minor) in NativeThreadLinux logging.
llvm-svn: 227913
See https://github.com/tfiala/lldb/issues/75. Not fixed yet but
continuing to push this further.
Fixes:
* Resume() now skips doing deferred notifications if we're doing a
vCont;{c,C}. In this case, we're trying to start something up,
not defer a stop notification. The default thread action stop
mode pickup was triggering a stop because it had at least one
stop, which was wrong in the case of a continue. (Bug introduced
by previous change.)
* Added a variant to ThreadStateCoordinator to specify a set of
thread ids to be skipped when triggering stop notifications to
non-stopped threads on a deferred signal call. For the case of
a stepping thread, it is actually told to step (and is running)
for a brief moment, but the thread state coordinator would think
it needed to send the stepping thread a stop, which id doesn't
need to do. This facility allows me to get around that cleanly.
With this change, behavior is now reduced to something I think is
essentially a different bug:
* Doing a step into libc code from my code crashes llgs.
* Doing a next out of a function in my own code crashes llgs.
llvm-svn: 227912
Tracked down while working on https://github.com/tfiala/lldb/issues/75.
This is not a complete fix for that issue, but moves us farther along.
Fixes:
* When a thread step is requested via vCont:{s,S}, Resume() now marks
the stepping thread as (1) currently stepping and (2) does trigger
the deferred signal for the stepped thread. This fixes a bug where
we were actually triggering a deferred stop cycle here for the non-stepping
thread since the single step thread was not part of the Resume()
deferred signal mechanism. The stepping thread is also marked in
the thread state coordinator as running (via a resume callback).
* When we get the SIGTRAP signal for the step completion, we don't
do a deferred signal call - that happened during the vCont:{s,S}
processing in Resume() already. Now we just need to mark that
the stepping thread is now stopped. If this is the last thread
in the set that needs to stop, it will trigger the process/delegate
stop call that will notify lldb. Otherwise, that'll happen when
the final thead we're waiting for stops.
Misc:
* Fixed up thread stop logging to use a leading 0 (0x%PRIx32) so
we don't get log lines like 0x5 for 0x05 SIGTRAP.
llvm-svn: 227911
* Fixed bug in run loop where run loop return enum was being treated
erroneously like an int, causing the TSC event loop to terminate
prematurely.
* Added an explicit scope in NativeProcessLinux::Resume() for the
threads lock lifetime. (This was likely unnecessary but is
more explicit.)
* Fixed a bug in ThreadStateCoordinator where resume execution was
not updating the internal state about the thread assumed to be
running now. I'll add a test and upstream this in a moment.
* Added a verbose logging mechanism to event processing within
ThreadStateCoordinator. It is currently enabled when the
'log enable lldb thread' is true upon inferior launch/attach.
llvm-svn: 227909
When it runs, lldb-mi creates a log file. There is an option --noLog
to disable the creation but it does not work and you end up with log
file. This happens as log is enabled in a few places in the code
although it has been disabled by this option.
This commit tried to fix the problem in the following way.
Log file generation is disabled by default.
You can enable it by giving --log option. It is the only way to enable
it. Rest of the call have been removed.
So the code basically remove the calls that enable the log
unconditionally and changes the option --noLog to --log.
llvm-svn: 227810
Currently it was marked as mandatory for many commands. Although it
is recommended that fronends generate this argument, it is not
mandatory. GDB seems to treat it as optional too.
llvm-svn: 227798
create its own threads with 8MB additional maximum stack size.
Extra room is needed for the bookkeeping needed for this
instrumentation.
llvm-svn: 227421
for executable binaries on the local filesystem so the user doesn't
need to provide the path to the correct binary manually.
Also have lldb search for kexts/the kernel in the current working
directory in addition to all the usual places.
<rdar://problem/18126501>
llvm-svn: 227419
./dotest.py -A x86_64 -C clang -v -t -f TestImageListMultiArchitecture.test_image_list_shows_multiple_architectures
The problem was that if the platform wasn't compatible with the current file in the "target create" command, it wasn't finding a platform that was like it used to.
Also, the currently selected platform was being used upload the file _before_ the target was created which was incorrect as "target create a.out" might switch platforms if its architecture doesn't match, so I moved the uploading to happen after the target was created so we use the right platform (the one in the target, not the selected one).
llvm-svn: 227380
The change was made so we could re-use a platform if one was already created instead of creating a new one, but it would fail in the above case. To fix this, if we have a selected platform, we verify that the platform matches the current platform before we try to re-use it. We do this by asking the OptionGroupPlatform if the platform matches. If so, it returns true and we don't create a new platform, else we do.
llvm-svn: 227288
And since enough of these are doing the right thing, add a test case to verify we are doing the right thing with freeze drying ObjC object types
Fixes rdar://18092770
llvm-svn: 227282
Namely, this commit provides an actual implementation of how to retrieve the byte size in a sane way for an ObjC class, by scanning ivar offsets and byte sizes, figuring out the farthest-from-base ivar, and adding its byte size to that
Still NFC
llvm-svn: 227277
This is necessary because the byte size of an ObjC class type is not reliably statically knowable (e.g. because superclasses sit deep in frameworks that we have no debug info for)
The lack of reliable size info is a problem when trying to freeze-dry an ObjC instance (not the pointer, the pointee)
This commit lays the foundation for having language runtimes help in figuring out byte sizes, and having ClangASTType ask for runtime help
No feature change as no runtime actually implements the logic, and nowhere is an ExecutionContext passed in yet
llvm-svn: 227274
This test tests the equivalent of:
breakpoint set --name count
breakpoint set --selector count
breakpoint set --name isEqual:
breakpoint set --selector isEqual:
breakpoint set --name "-[MyClass(MyCategory) myCategoryFunction]"
breakpoint set --name "-[MyClass myCategoryFunction]"
breakpoint set --name "[MyClass myCategoryFunction]"
llvm-svn: 227271
This patch adds lldbmi_testcase.MiTestCaseBase class and removes
a lot of repitition in the lldb-mi test cases. Also cleans import
list and code-style.
Presented for review in http://reviews.llvm.org/D7162.
Patch from Ilia K <ki.stfu@gmail.com>.
llvm-svn: 227218
it does call, and implementing it so that we once again look up external symbols in the JIT.
Also juked the error reporting from the JIT a little bit.
This resolves:
http://llvm.org/bugs/show_bug.cgi?id=22314
llvm-svn: 227217
This was causing problems on clang 602 branches that use MemoryObjects to as the container for opcode bytes instead of a plain array of bytes. So we were asking for 3 bytes to be disassembled at address 0xffffffffffffffff which would cause an unsigned overflow and cause the MemoryObject to refuse to read anymore bytes.
llvm-svn: 227153
Without this overload, attempts to edit the value of a variable with synthetic children enabled would change the value inside the synthetic ValueObject, but not propagate the changes to the underlying storage, hence resulting in no write for any meaningful purpose
Comes with a test case, and fixes rdar://19586311
llvm-svn: 227120
This patch fixes test_launch_in_terminal test which doesn't work
in OS X since the moment as it was added in r225284. The test fails
because Target::Launch returns the following error: "the darwin-debug
executable doesn't exist at
<output_dir>/lib/python2.7/site-packages/lldb/darwin-debug'".
Patch by Ilia K
Differential Revision: http://reviews.llvm.org/D7102
llvm-svn: 227096
Ilia K <ki.stfu@gmail.com> sent this patch. He noted that it was possible for
child to be terminated after isAlive call and before we send it quit command.
In this case, sending command to it causes OSError. This patch just adds this
exception to the except list.
llvm-svn: 227086
This patch fixes TestRegisters on Linux with LLGS
Introduce GetUserRegisterCount on RegisterInfoInterface to distinguish
lldb internal registers (e.g.: DR0-DR7) during register counting.
Update GDBRemoteCommunicationServer to skip lldb internal registers on
read/write register and on discover register.
Submitted for Tamas Berghammer
llvm-svn: 226959
Other platforms may benefit from something similar if issues arise. The
libedit library doesn't explicitly initialize the curses termcap library,
which it gets away with until TERM is set to VT100 where it stumbles over
an implementation assumption that may not exist on other platforms.
<rdar://problem/17581929>
llvm-svn: 226891
i386/x86_64 functions. The stack size was being multiplied by the
pointer size incorrectly. The register permutation placeholders
(UNWIND_X86_REG_NONE) were decrementing the stack offset of the
saved registers when it should not have been.
<rdar://problem/19570035>
llvm-svn: 226889
paths we get from dladdr to have "//" in it internally, and while that is
formally correct it is just asking for somebody to misparse it...
llvm-svn: 226886
On Windows we copy python27(_d).dll to the bin directory. We do
this by looking at the PYTHON_LIBRARY specified by the user, which
is something like C:\foo\python27_d.lib, and replacing ".lib" with
".dll". But ".lib" as a regex will also match "flib", etc. So
make this a literal . instead of a wildcard .
llvm-svn: 226858
This matches the behavior of the default constructor, so is
technically more correct.
Patch by Chaoren Lin
Differential Revision: http://reviews.llvm.org/D7113
llvm-svn: 226851
When you create a target, it tries to look for the platform's list
of supported architectures for a match. The match it finds can
contain specific triples, like i386-pc-windows-msvc. Later, we
overwrite this value with the most generic triple that can apply
to any platform with COFF support, causing some of the fields of
the triple to get overwritten.
This patch changes the behavior to only merge in values from the COFF
triple if the fields of the matching triple were unknown/unspecified
to begin with.
This fixes load address resolution on Windows, since it enables the
DynamicLoaderWindows to be used instead of DynamicLoaderStatic.
Reviewed by: Greg Clayton
Differential Revision: http://reviews.llvm.org/D7120
llvm-svn: 226849
buffer and to add a nul terminator byte, was incorrectly resizing
its buffer so the nul terminator was not included.
Problem found by clang ASAN instrumentation when using an
expression prefix file which was read via this mechanism.
<rdar://problem/19556459>
llvm-svn: 226753
inferring the function signature. This works well where the ABI doesn't
distinguish between variadic and fixed argument lists, but on arm64 the
calling conventions differ. The default assumption works for fixed argument
lists, but variadic functions require explicit prototypes to be called.
By far the most common case where this is an issue is when attempting to use
printf(). This change augments the default expression prefix to include a
working variadic prototype for the function.
<rdar://problem/19024779>
llvm-svn: 226744
The reason why one mechanism is so much faster than the other (as in, ~10 seconds vs. ~1 minute) is quite baffling to me at the moment, but these are not process handling tests, so do what's faster
llvm-svn: 226730
Make sure the selected platform is always used
Make sure that the host uses the connect://hostname to connect to both
the lldb-platform and the lldb-gdbserver rather than what the platform
reports as the hostname of the lldb-gdbserver
Make sure that lldb-platform uses the IP address on it's connection
back to the host instead of the hostname that the host sends to it
when launching lldb-gdbserver with the remote host information
Tested on OSX and Linux
llvm-svn: 226712
Since REG_ENHANCED is available on MacOSX, this allow the use of \d (digits) \b (word boundaries) and much more without affecting other systems.
<rdar://problem/12082562>
llvm-svn: 226704
This reverts commit r226679. For some reason it was
not generating the same behavior as manually specifying
the include dir, library path, and exe path, and it was
causing the test suite to fail to run.
llvm-svn: 226683
CMake FindPythonLibs will look for multiple versions of Python
including both debug and release, and build up a list such as
(debug <debugpath> optimized <optimizedpath>). This confuses
the logic we have in CMake to copy the correct python dll to
the output directory so that it need not be in your system's PATH.
To alleviate this, we manually split this list and extract out
the debug and release versions of the python library, and copy
only the correct one to the output directory.
llvm-svn: 226679
Most of the time, we can use context information just fine to choose a language (i.e. the language of the frame that the root object was defined in, if any); but in some cases, synthetic children may be fabricated as root frame-less entities, and then we wouldn't know any better
This patch allows (internal) synthetic child providers to set a display language on the children they generate, should they so choose
llvm-svn: 226634
introduced subtle bugs in two places in
RegisterContextLLDB::GetFullUnwindPlanForFrame where
it specifically wanted to get an eh_frame unwind plan
and was using "Get CallSite UnwindPlan" as synonymous
with that. But now we have two different types of
unwind plan that can be returned in that case, and
compact unwind won't behaves as needed.
<rdar://problem/19528559>
llvm-svn: 226631
We were referring to hardcoded paths /bin/ls and /bin/cat. For
the purposes of this test, the actual value it's set to doesn't
matter, and it might as well be a non-existent path. All that
matters is that the before and after values have to match, and
that trailing whitespace is stripped. On Windows FileSpec
(correctly) converts /bin/ls to D:\bin\ls though, so the before
and after values won't match. So this patch just correctly builds
up a valid path in a platform-agnostic manner, and verifies that
it matches before and after the set.
llvm-svn: 226625
At some point we will need to either provide a pexpect equivalent
on Windows, or provide some other method of doing out-of-process
tests.
Even with a pexpect replacement, it may be worth re-evaluating
some of these tests to see if they would be better served as
in-process tests. The larger issue of coming up with a pexpect
replacement on Windows is tracked in http://llvm.org/pr22274.
llvm-svn: 226614
lldb::pid_t
Host::LaunchApplication (const FileSpec &app_file_spec);
This had use of a function FSPathMakeRef(const UInt8*, FSRef *, ...) that was deprecated in 10.8.
Removing this fucntion since it wasn't used and was causing warnings.
llvm-svn: 226608
This fixes file paths on Windows, as you can now write, for example,
file d:\foo\bar.txt, but does not break the case that this tokenization
logic was originally designed for, which is to allow escaping of things
like quotes and double quotes, so that all of the escapable characters
can appear in the same string together.
Reviewed by: Jim Ingham, Greg Clayton
Differential Revision: http://reviews.llvm.org/D7018
llvm-svn: 226587
so it doesn't assume that the SmallVector<char> will have
nul terminator. It did not in at least one case.
Caught by ASAN instrumentation.
llvm-svn: 226544
Include paths were switched to be user include paths, if this breaks the linux build we will need to fix the Makefiles/cmake stuff.
<rdar://problem/19198581>
llvm-svn: 226530
CommandInterpreter's execution context AFTER the process had started running
and before it initially stopped. Also fixed one test case that was implicitly
using this (and an abuse of the async mode) to accidentally succeed.
<rdar://problem/16814726>
llvm-svn: 226528
This function returns a URI of the resource that the connection is connected to. This is especially important for connections established by accepting a connection from a remote host.
Also added implementations for ConnectionMachPort, ConnectionSharedMemory,
Also fixed up some documentation in Connection::Write
Renamed ConnectionFileDescriptorPosix::SocketListen to ConnectionFileDescriptorPosix::SocketListenAndAccept
Fixed a log message in Socket.cpp
Differential Review: http://reviews.llvm.org/D7026
llvm-svn: 226362
the register state when debugging AArch32 programs (armv7
programs running on an armv8 processor). Most notably,
there is no "fpscr" register in the register context -
there is an fpsr and an fpcr.
Also fix a bug where the floating point values could not
be written in armv7 processes.
<rdar://problem/18977767>
llvm-svn: 226244
`ninja lldb` used to always run "echo -n", which on OS X results in literally
echoing "-n" to the screen. Just remove the command from add_custom_target,
then it only adds an alias and `ninja lldb` now reports "no work to do".
Other than that, no intended behavior change.
llvm-svn: 226233
This hooks up the changes necessary to set the trap flag on the
CPU and properly manage the process and thread's resume state
and private state so that the ThreadPlan does its thing.
Stepping still doesn't work as of this change, because there are
some issues with stack frames where it doesn't update the thread's
frame list correctly when it breaks inside of a function, but
I will try to fix that separately.
llvm-svn: 226221
On Windows, opening with "w" opens it as text instead of binary.
This causes translation of newline characters, so that "\n" turns
into "\r\n", which in turn leads to git detecting that the file
has changed and wanting to commit it.
llvm-svn: 226220
The refactor was motivated by some comments that Greg made
http://reviews.llvm.org/D6918
and also to break a dependency cascade that caused functions linking
in string->int conversion functions to pull in most of lldb
llvm-svn: 226199
moved into the #else branch of the #if/#elif/#endif, so it wasn't getting done in the #if case anymore.
Keep the code to add the demangled name outside of the #if, and then just free the demangled_name
and set it back to NULL in the Windows case.
<rdar://problem/19479499>
llvm-svn: 226088
This is done by adding a "Variable *" to SymbolContext and allowing SymbolFile::ResolveSymbolContext() so if an address is resolved into a symbol context, we can include the global or static variable for that address.
This means you can now find global variables that are merged globals when doing a "image lookup --verbose --address 0x1230000". Previously we would resolve a symbol and show "_MergedGlobals123 + 1234". But now we can show the global variable name.
The eSymbolContextEverything purposely does not include the new eSymbolContextVariable in its lookup since stack frame code does many lookups and we don't want it triggering the global variable lookups.
<rdar://problem/18945678>
llvm-svn: 226084
the hit count is not updated.
Also, keep the hit count for the breakpoint in the breakpoint. We were
using just the sum of the location's hit counts, but that was wrong since if a shared library is
unloaded, and the location goes away, the breakpoint hit count should not suddenly drop
by the number of hits there were on that location.
llvm-svn: 226074
The default help display now shows the alias collection by default, and hides commands whose named begin with an underscore. Help is primarily useful to those unfamiliar with LLDB and should aim to answer typical questions while still being able to provide more esoteric answers when required. To that latter end an argument to include the hidden commands in help has been added, and instead of having a help flag to show aliases there is now one to hide them. This final change might be controversial as it repurposes the -a shorthand as the opposite of its original meaning.
The previous implementation of OutputFormattedHelpText was easily confused by embedded newlines. The new algorithm correctly breaks on the FIRST newline or LAST space/tab before the target column count rather than treating all whitespace interchangeably.
Command interpreters now have the ability to specify help prologue text and a command prefix string. Neither are used in the current LLDB sources but are required to support REPL-like extensions where LLDB commands must be prefixed and additional help text is required to explain how to access traditional debugging commands.
<rdar://problem/17751929>
<rdar://problem/16953815>
<rdar://problem/16953841>
<rdar://problem/16930173>
<rdar://problem/16879028>
llvm-svn: 226068
While there is quite a bit of potential for mishaps due to tagged pointers, and after quite some internal discussion, this seems a saner behavior given how "po" stands for "print OBJECT". The argument being that we should make at least some sensible attempt to print the thing the user passed as-if it was an object
Fixes rdar://19423124
llvm-svn: 226062
These fix various issues with path handling and disable a few tests
which use features of LLVM which are not yet supported on Windows.
llvm-svn: 226042
BUILD_SHARED_LIBS=TRUE currently isn't working for Linux x86_64
This patch fixes the link errors and also some runtime errors
Test Plan:
CC=clang CXX=clang++ cmake -GNinja -DBUILD_SHARED_LIBS=TRUE -DCMAKE_LINKER=ld.gold -DCMAKE_BUILD_TYPE=Debug ../../llvm
ninja
ninja check-lldb
llvm-svn: 226039
If stdout is not a terminal Python executes
rl_variable_bind ("enable-meta-key", "off");
This produces a warning with FreeBSD's libedit because the
enable-meta-key variable is unknown. Not an issue on Apple because
cpython commit f0ab6f9f0603 added a #ifndef __APPLE__ around the call.
See http://bugs.python.org/issue19884 for more information.
For now we just discard the warning output to get the tests working
again on FreeBSD.
llvm-svn: 226037
We now verify that the debugger's input file is a valid terminal file descriptor before allowing the "gui" command to try to run.
Xcode would crash if you typed "gui" at the command line prior to this fix.
<rdar://problem/18775851>
llvm-svn: 226027
saved/restored across a mid-function epilogue. We ignore
repeated push/pops of a register so once we saw one 'pop %rbp',
we'd ignore it the second time we saw it.
<rdar://problem/19417410>
llvm-svn: 225853
This is currently controlled by a setting:
(lldb) settings set target.process.python-os-plugin-path <path>
Or clearing it with:
(lldb) settings clear target.process.python-os-plugin-path
The process will now reload the OperatingSystem plug-in.
This was implemented by:
- adding the ability to set a notify callback for when an option value is changed
- added the ability for the process plug-in to load the operating system plug-in on the fly
- fixed bugs in the Process::GetStatus() so all threads are displayed if their thread IDs are larger than 32 bits
- adding a callback in ProcessProperties to tell when the "python-os-plugin-path" is changed by the user
- fixing a crasher in ProcessMachCore that happens when updating the thread list when the OS plugin is reloaded
llvm-svn: 225831
it will do the right thing on x86 routines with a mid-function
epilogue sequence (where the unwind rules need to be reinstalled
after the epilogue has completed).
<rdar://problem/19417410>
llvm-svn: 225773
which will verify if the eh_frame instructions include details about
the prologue or not. Both clang and gcc include prologue instructions
but there's no requirement for them to do so -- and I'm sure we'll
have to interoperate with a compiler that doesn't generate prologue
info at some point.
I don't have any compilers that omit the prologue instructions so the
testing was of the "makre sure augmented unwind info is still created".
With an eh_frame without prologue, this code should reject the
augmentation scheme altogether and we should fall back to using assembly
instruction profiling.
llvm-svn: 225771
step through the complete function looking for any epilogue
instructions. If we find an epilogue sequence, re-instate
the correct unwind instructions if there is more code past
that epilogue -- this will correctly handle an x86 function
with multiple epilogues in it.
NB there is still a bug with the "eh_frame augmented"
UnwindPlans and mid-function epilogues. Looking at that next.
<rdar://problem/18863406>
llvm-svn: 225770
Chaoren Lin