LLDB requires that the inferior process be stopped before, and remain
stopped during, certain accesses to process state.
Previously this was achieved with a POSIX rwlock which had a write lock
taken for the duration that the process was running, and released when
the process was stopped. Any access to process state was performed with
a read lock held.
However, POSIX requires that pthread_rwlock_unlock() be called from the
same thread as pthread_rwlock_wrlock(), and lldb needs to stop and start
the process from different threads. Violating this constraint is
technically undefined behaviour, although as it happens Linux and Darwin
result in the unlock proceeding in this case. FreeBSD follows POSIX
more strictly, and the unlock would fail, resulting in a hang later upon
the next attempt to take the lock.
All read lock consumers use ReadTryLock() and handle failure to obtain
the lock (typically by logging an error "process is running"). Thus,
instead of using the lock state itself to track the running state, this
change adds an explicit m_running flag. ReadTryLock tests the flag, and
if the process is not running it returns with the read lock held.
WriteLock and WriteTryLock are renamed to SetRunning and TrySetRunning,
and (if successful) they set m_running with the lock held. This way,
read consumers can determine if the process is running and act
appropriately, and write consumers are still held off from starting the
process if read consumers are active.
Note that with this change there are still some curious access patterns,
such as calling WriteUnlock / SetStopped twice in a row, and there's no
protection from multiple threads trying to simultaneously start the
process. In practice this does not seem to be a problem, and was
exposing other undefined POSIX behaviour prior to this change.
llvm-svn: 187377
bother checking if a region is safe to use. In
cases where regions need to be synthesized rather
than properly allocated, the memory reads required
to determine whether the area is used are
- insufficient, because intermediate locations
could be in use, and
- unsafe, because on some platforms reading from
memory can trigger events.
All this only makes a difference on platforms
where memory allocation in the target is impossible.
Behavior on platforms where it is possible should
stay the same.
<rdar://problem/14023970>
llvm-svn: 185046
Made sure that temporary object created from HarmonizeThreadIdsForProfileData() doesn’t get passed around without creating an object first.
Reviewed by Greg
llvm-svn: 184769
325,000 breakpoints for running "breakpoint set --func-regex ." on lldb itself (after hitting a breakpoint at main so that LLDB.framework is loaded) used to take up to an hour to set, now we are down under a minute. With warm file caches, we are at 40 seconds, and that is with setting 325,000 breakpoint through the GDB remote API. Linux and the native debuggers might be faster. I haven't timed what how much is debug info parsing and how much is the protocol traffic to/from GDB remote.
That there were many performance issues. Most of them were due to storing breakpoints in the wrong data structures, or using the wrong iterators to traverse the lists, traversing the lists in inefficient ways, and not optimizing certain function name lookups/symbol merges correctly.
Debugging after that is also now very efficient. There were issues with replacing the breakpoint opcodes in memory that was read, and those routines were also fixed.
llvm-svn: 183820
regions that aren't actually allocated in the
process. This cache is used by the expression
parser if the underlying process doesn't support
memory allocation, to avoid needless repeated
searches for unused address ranges.
Also fixed a silly bug in IRMemoryMap where it
would continue searching even after it found a
valid region.
<rdar://problem/13866629>
llvm-svn: 182028
Avoid a deadlock when using the OperatingSystemPython code and typing "process interrupt". There was a possible lock inversion between the target API lock and the process' thread list lock due to code trying to discard the thread list. This was fixed by adding a boolean to Process::Halt() that indicates if the thread plans should be discarded and doing it in the private state thread when we process the stopped state.
llvm-svn: 181651
<rdar://problem/13594769>
Main changes in this patch include:
- cleanup plug-in interface and use ConstStrings for plug-in names
- Modfiied the BSD Archive plug-in to be able to pick out the correct .o file when .a files contain multiple .o files with the same name by using the timestamp
- Modified SymbolFileDWARFDebugMap to properly verify the timestamp on .o files it loads to ensure we don't load updated .o files and cause problems when debugging
The plug-in interface changes:
Modified the lldb_private::PluginInterface class that all plug-ins inherit from:
Changed:
virtual const char * GetPluginName() = 0;
To:
virtual ConstString GetPluginName() = 0;
Removed:
virtual const char * GetShortPluginName() = 0;
- Fixed up all plug-in to adhere to the new interface and to return lldb_private::ConstString values for the plug-in names.
- Fixed all plug-ins to return simple names with no prefixes. Some plug-ins had prefixes and most ones didn't, so now they all don't have prefixed names, just simple names like "linux", "gdb-remote", etc.
llvm-svn: 181631
This re-submission of this patch fixes a problem where the code sometimes caused a deadlock. The Process::SetPrivateState method was locking the Process::m_private_state variable and then later calling ThreadList::DidStop, which locks the ThreadList mutex. Other methods in ThreadList which were being called from other threads lock the ThreadList mutex and then call Process::GetPrivateState which locks the Process::m_private_state mutex. To avoid deadlocks, Process::SetPrivateState now locks the ThreadList mutex before locking the Process::m_private_state mutex.
llvm-svn: 181609
- Played with the current dual run lock implementation for a few days, noticed
no regressions, so enabling in trunk so we see if any problems are detected
by buildbots.
llvm-svn: 181446
value. This fixes problems, for instance, with the StepRange plans, where they know that
they explained the stop because they were at their "run to here" breakpoint, then deleted
that breakpoint, so when they got asked again, doh! I had done this for a couple of plans
in an ad hoc fashion, this just formalizes it.
Also add a "ResumeRequested" in Process so that the code in the completion handlers can
tell the ShouldStop logic they want to resume rather than just directly resuming. That allows
us to handle resuming in a more controlled fashion.
Also, SetPublicState can take a "restarted" flag, so that it doesn't drop the run lock when
the target was immediately restarted.
--This line, and those below , will be ignored--
M test/lang/objc/objc-dynamic-value/TestObjCDynamicValue.py
M include/lldb/Target/ThreadList.h
M include/lldb/Target/ThreadPlanStepOut.h
M include/lldb/Target/Thread.h
M include/lldb/Target/ThreadPlanBase.h
M include/lldb/Target/ThreadPlanStepThrough.h
M include/lldb/Target/ThreadPlanStepInstruction.h
M include/lldb/Target/ThreadPlanStepInRange.h
M include/lldb/Target/ThreadPlanStepOverBreakpoint.h
M include/lldb/Target/ThreadPlanStepUntil.h
M include/lldb/Target/StopInfo.h
M include/lldb/Target/Process.h
M include/lldb/Target/ThreadPlanRunToAddress.h
M include/lldb/Target/ThreadPlan.h
M include/lldb/Target/ThreadPlanCallFunction.h
M include/lldb/Target/ThreadPlanStepOverRange.h
M source/Plugins/LanguageRuntime/ObjC/AppleObjCRuntime/AppleThreadPlanStepThroughObjCTrampoline.h
M source/Plugins/LanguageRuntime/ObjC/AppleObjCRuntime/AppleThreadPlanStepThroughObjCTrampoline.cpp
M source/Target/StopInfo.cpp
M source/Target/Process.cpp
M source/Target/ThreadPlanRunToAddress.cpp
M source/Target/ThreadPlan.cpp
M source/Target/ThreadPlanCallFunction.cpp
M source/Target/ThreadPlanStepOverRange.cpp
M source/Target/ThreadList.cpp
M source/Target/ThreadPlanStepOut.cpp
M source/Target/Thread.cpp
M source/Target/ThreadPlanBase.cpp
M source/Target/ThreadPlanStepThrough.cpp
M source/Target/ThreadPlanStepInstruction.cpp
M source/Target/ThreadPlanStepInRange.cpp
M source/Target/ThreadPlanStepOverBreakpoint.cpp
M source/Target/ThreadPlanStepUntil.cpp
M lldb.xcodeproj/xcshareddata/xcschemes/Run Testsuite.xcscheme
llvm-svn: 181381
while we develop a better understanding of how to manage the thread lists in a platform-independant fashion.
Reviewed by: Daniel Malea
llvm-svn: 181323
This checkin aims to fix this. The process now has two thread lists: a real thread list for threads that are created by the lldb_private::Process subclass, and the user visible threads. The user visible threads are the same as the real threas when no OS plug-in in used. But when an OS plug-in is used, the user thread can be a combination of real and "memory" threads. Real threads can be placed inside of memory threads so that a thread appears to be different, but is still controlled by the actual real thread. When the thread list needs updating, the lldb_private::Process class will call the: lldb_private::Process::UpdateThreadList() function with the old real thread list, and the function is expected to fill in the new real thread list with the current state of the process. After this function, the process will check if there is an OS plug-in being used, and if so, it will give the old user thread list, the new real thread list and the OS plug-in will create the new user thread list from both of these lists. If there is no OS plug-in, the real thread list is the user thread list.
These changes keep the lldb_private::Process subclasses clean and no changes are required.
llvm-svn: 181091
If someone on Linux and/or FreeBSD can try to comment out the " #if defined(__APPLE__)" that surrounds access to "m_private_run_lock" and run the test suite, that would be nice. The new location where the locking/unlocking happens is bulletproof on MacOSX, and I want to verify that it is good on linux as well.
llvm-svn: 181061
specially, and make sure we stop. This shouldn't happen, but if it does, the user will probably want to
see it.
<rdar://problem/13273125>
llvm-svn: 180244
This prevents unbounded reads (i.e. reads of GetMaximumSizeOfStringSummary() bytes)
from causing test failures (i.e. due to ptrace EIO or EFAULT on Linux).
Note that ReadCStringFromMemory is marked as deprecated because the loop that calls
ReadMemory does not continue until the string has been completely read.
The expected behavior is to read until until max_bytes or a null terminator.
Note: As discussed on lldb-dev, further testing will be performed with ReadStringFromMemory
before further changes are made for users of ReadCStringFromMemory.
Thanks to Enrico, Matt and Andy for their review feedback.
llvm-svn: 179857
Made some fixes to the OperatingSystemPython class:
- If any thread dictionary contains any "core=N" key/value pairs then the threads obtained from the lldb_private::Process itself will be placed inside the ThreadMemory threads and will be used to get the information for a thread.
- Cleaned up all the places where a thread inside a thread was causing problems
llvm-svn: 179405
Fixed a case there the OperatingSystemPython would try to access and play with SBValue objects when the process' public run lock was taken. Prior to this fix, all attempts to run any SBValue functions would fail if run from the private state thread (like updating the thread list). Now we have two run locks, one for public (all threads except the private state thread) and one for private.
llvm-svn: 179329
LLDB is crashing when logging is enabled from lldb-perf-clang. This has to do with the global destructor chain as the process and its threads are being torn down.
All logging channels now make one and only one instance that is kept in a global pointer which is never freed. This guarantees that logging can correctly continue as the process tears itself down.
llvm-svn: 178191
and use it to keep from doing the OS Plugin UpdateThreadList while destroying, since
if that does anything that requires the API lock it may deadlock against whoever is
running the Process::Destroy.
<rdar://problem/13308627>
llvm-svn: 176375
to have it not named appropriately. Also in StopInfoMachException, we aren't testing for software or not software, just
whether the thing is a breakpoint we set. So don't use "software"...
llvm-svn: 175241
When launching in the shell, make sure if you specify a relative path, and if the current working directory has a space in it, that we don't hose the shell invocation.
Currently if we launch with a relative path, we prepend the current working directory to the PATH using:
PATH=`cwd`:$PATH a.out ...
We needed to add quotes around the value for PATH to make sure if any paths in PATH contained spaces, that we don't hose the shell command. Now we do a:
PATH="`cwd`:$PATH" a.out ...
llvm-svn: 175135
hitting auto-continue signals while running a thread plan would cause us to lose control of the debug
session.
<rdar://problem/12993641>
llvm-svn: 174793
Fix in loading mach files from memory when using DynamicLoaderMacOSXDYLD.
Removed the uuid mismatch warning that could be spit out and any time during debugging and removed the test case that was looking for that. Currently the "add-dsym" or "target symbols add" command will report an error when the UUID's don't match.
Be more careful when checking and resolving section + offset addresses to make sure none of the base addresses are invalid.
llvm-svn: 174222
Add the ability to give breakpoints a "kind" string, and have the StopInfoBreakpoint
print that in the brief description if set. Also print the kind - if set - in the breakpoint
listing.
Give kinds to a bunch of the internal breakpoints.
We were deleting the Mac OS X dynamic loader breakpoint as though the id we had stored away was
a breakpoint site ID, but in fact it was a breakpoint id, so we never actually deleted it. Fixed that.
llvm-svn: 173555
Major fixed to allow reading files that are over 4GB. The main problems were that the DataExtractor was using 32 bit offsets as a data cursor, and since we mmap all of our object files we could run into cases where if we had a very large core file that was over 4GB, we were running into the 4GB boundary.
So I defined a new "lldb::offset_t" which should be used for all file offsets.
After making this change, I enabled warnings for data loss and for enexpected implicit conversions temporarily and found a ton of things that I fixed.
Any functions that take an index internally, should use "size_t" for any indexes and also should return "size_t" for any sizes of collections.
llvm-svn: 173463
Added the ability for OS plug-ins to lazily populate the thread this. The python OS plug-in classes can now implement the following method:
class OperatingSystemPlugin:
def create_thread(self, tid, context):
# Return a dictionary for a new thread to create it on demand
This will add a new thread to the thread list if it doesn't already exist. The example code in lldb/examples/python/operating_system.py has been updated to show how this call us used.
Cleaned up the code in PythonDataObjects.cpp/h:
- renamed all classes that started with PythonData* to be Python*.
- renamed PythonArray to PythonList. Cleaned up the code to use inheritance where
- Centralized the code that does ref counting in the PythonObject class to a single function.
- Made the "bool PythonObject::Reset(PyObject *)" function be virtual so each subclass can correctly check to ensure a PyObject is of the right type before adopting the object.
- Cleaned up all APIs and added new constructors for the Python* classes to they can all construct form:
- PyObject *
- const PythonObject &
- const lldb::ScriptInterpreterObjectSP &
Cleaned up code in ScriptInterpreterPython:
- Made calling python functions safer by templatizing the production of value formats. Python specifies the value formats based on built in C types (long, long long, etc), and code often uses typedefs for uint32_t, uint64_t, etc when passing arguments down to python. We will now always produce correct value formats as the templatized code will "do the right thing" all the time.
- Fixed issues with the ScriptInterpreterPython::Locker where entering the session and leaving the session had a bunch of issues that could cause the "lldb" module globals lldb.debugger, lldb.target, lldb.process, lldb.thread, and lldb.frame to not be initialized.
llvm-svn: 172873
Added a unique integer identifier to processes. Some systems, like JTAG or other simulators, might always assign the same process ID (pid) to the processes that are being debugged. In order for scripts and the APIs to uniquely identify the processes, there needs to be another ID. Now the SBProcess class has:
uint32_t SBProcess::GetUniqueID();
This integer ID will help to truly uniquely identify a process and help with appropriate caching that can be associated with a SBProcess object.
llvm-svn: 172628
controlled by the --unwind-on-error flag, and --ignore-breakpoint which separately controls behavior when a called
function hits a breakpoint. For breakpoints, we don't unwind, we either stop, or ignore the breakpoint, which makes
more sense.
Also make both these behaviors globally settable through "settings set".
Also handle the case where a breakpoint command calls code that ends up re-hitting the breakpoint. We were recursing
and crashing. Now we just stop without calling the second command.
<rdar://problem/12986644>
<rdar://problem/9119325>
llvm-svn: 172503
Fixed an issue where the platform auto select code was changing the architecture and causing the wrong architecture to be assigned to the target.
llvm-svn: 172251
Should be that if any of the threads wants to stop, we should stop. The opposite was what was actually happening
<rdar://problem/12869725>
llvm-svn: 170153
equality can be strict or loose and we want code to
explicitly choose one or the other.
Also renamed the Compare function to IsEqualTo, to
avoid confusion.
<rdar://problem/12856749>
llvm-svn: 170152
- add new header lldb-python.h to be included before other system headers
- short term fix (eventually python dependencies must be cleaned up)
Patch by Matt Kopec!
llvm-svn: 169341
Cleaned up the option parsing code to always pass around the short options as integers. Previously we cast this down to "char" and lost some information. I recently added an assert that would detect duplicate short character options which was firing during the test suite.
This fix does the following:
- make sure all short options are treated as "int"
- make sure that short options can be non-printable values when a short option is not required or when an option group is mixed into many commands and a short option is not desired
- fix the help printing to "do the right thing" in all cases. Previously if there were duplicate short character options, it would just not emit help for the duplicates
- fix option parsing when there are duplicates to parse options correctly. Previously the option parsing, when done for an OptionGroup, would just start parsing options incorrectly by omitting table entries and it would end up setting the wrong option value
llvm-svn: 169189
Prevent async and sync calls to get profile data from stomping on each other.
At the same time, don't use '$' as end delimiter per chunk of profile data.
llvm-svn: 168948
This commit does three things:
(a) introduces a new notification model for adding/removing/changing modules to a ModuleList, and applies it to the Target's ModuleList, so that we make sure to always trigger the right set of actions
whenever modules come and go in a target. Certain spots in the code still need to "manually" notify the Target for several reasons, so this is a work in progress
(b) adds a new capability to the Platforms: locating a scripting resources associated to a module. A scripting resource is a Python file that can load commands, formatters, ... and any other action
of interest corresponding to the loading of a module. At the moment, this is only implemented on Mac OS X and only for files inside .dSYM bundles - the next step is going to be letting
the frameworks themselves hold their scripting resources. Implementors of platforms for other systems are free to implement "the right thing" for their own worlds
(c) hooking up items (a) and (b) so that targets auto-load the scripting resources as the corresponding modules get loaded in a target. This has a few caveats at the moment:
- the user needs to manually add the .py file to the dSYM (soon, it will also work in the framework itself)
- if two modules with the same name show up during the lifetime of an LLDB session, the second one won't be able to load its scripting resource, but will otherwise work just fine
llvm-svn: 167569
I tracked down a leak that could happen when detaching from a process where the lldb_private::Process objects would stay around forever. This was caused by a eStateDetached event that was queued up on the lldb_private::Process private state thread listener. Since process events contain shared pointers to the process, this is dangerous if they don't get consume or cleared as having the lldb_private::Process class contain a collection of things that have a shared pointer to yourself is obviously bad.
To fix this I modified the Process::Finalize() function to clear this list. The actual thing that was holding onto the ModuleSP and thus the static archive, was a stack frame. Since the process wasn't going away, it still had thread objects and they still had frames. I modified the Thread::Destroy() to clear the stack frames to ensure this further doesn't happen.
llvm-svn: 166964
There was a generic catch-all type for path arguments
called "eArgTypePath," and a specialized version
called "eArgTypeFilename." It turns out all the
cases where we used eArgTypePath we could have
used Filename or we explicitly meant a directory.
I changed Path to DirectoryName, made it use the
directory completer, and rationalized the uses of
Path.
<rdar://problem/12559915>
llvm-svn: 166533
Added a new setting that allows a python OS plug-in to detect threads and provide registers for memory threads. To enable this you set the setting:
settings set target.process.python-os-plugin-path lldb/examples/python/operating_system.py
Then run your program and see the extra threads.
llvm-svn: 166244
LLDB changes argv[0] when debugging a symlink. Now we have the notion of argv0 in the target settings:
target.arg0 (string) =
There is also the program argument that are separate from the first argument that have existed for a while:
target.run-args (arguments) =
When running "target create <exe>", we will place the untouched "<exe>" into target.arg0 to ensure when we run, we run with what the user typed. This has been added to the ProcessLaunchInfo and all other needed places so we always carry around the:
- resolved executable path
- argv0
- program args
Some systems may not support separating argv0 from the resolved executable path and the ProcessLaunchInfo needs to carry all of this information along so that each platform can make that decision.
llvm-svn: 166137
enabled after we'd found a few bugs that were caused by shadowed
local variables; the most important issue this turned up was
a common mistake of trying to obtain a mutex lock for the scope
of a code block by doing
Mutex::Locker(m_map_mutex);
This doesn't assign the lock object to a local variable; it is
a temporary that has its dtor called immediately. Instead,
Mutex::Locker locker(m_map_mutex);
does what is intended. For some reason -Wshadow happened to
highlight these as shadowed variables.
I also fixed a few obivous and easy shadowed variable issues
across the code base but there are a couple dozen more that
should be fixed when someone has a free minute.
<rdar://problem/12437585>
llvm-svn: 165269
loaded at a random offset).
To get the kernel's UUID and load address I need to send a kdp
packet so I had to implement the kernel relocation (and attempt to
find the kernel if none was provided to lldb already) in ProcessKDP
-- but this code really properly belongs in DynamicLoaderDarwinKernel.
I also had to add an optional Stream to ConnectRemote so
ProcessKDP::DoConnectRemote can print feedback about the remote kernel's
UUID, load address, and notify the user if we auto-loaded the kernel via
the UUID.
<rdar://problem/7714201>
llvm-svn: 164881
Fixed an issue where if we call "Process::Destroy()" and the process is running, if we try to stop it and get "exited" back as the stop reason, we will still deliver the exited event.
llvm-svn: 163591
on, basic inlined stepping works, including step-over of inlined functions. But for some as yet mysterious reason i386 debugging gets an
assert and dies immediately. So for now its off.
llvm-svn: 163044
Make breakpoint setting by file and line much more efficient by only looking for inlined breakpoint locations if we are setting a breakpoint in anything but a source implementation file. Implementing this complex for a many reasons. Turns out that parsing compile units lazily had some issues with respect to how we need to do things with DWARF in .o files. So the fixes in the checkin for this makes these changes:
- Add a new setting called "target.inline-breakpoint-strategy" which can be set to "never", "always", or "headers". "never" will never try and set any inlined breakpoints (fastest). "always" always looks for inlined breakpoint locations (slowest, but most accurate). "headers", which is the default setting, will only look for inlined breakpoint locations if the breakpoint is set in what are consudered to be header files, which is realy defined as "not in an implementation source file".
- modify the breakpoint setting by file and line to check the current "target.inline-breakpoint-strategy" setting and act accordingly
- Modify compile units to be able to get their language and other info lazily. This allows us to create compile units from the debug map and not have to fill all of the details in, and then lazily discover this information as we go on debuggging. This is needed to avoid parsing all .o files when setting breakpoints in implementation only files (no inlines). Otherwise we would need to parse the .o file, the object file (mach-o in our case) and the symbol file (DWARF in the object file) just to see what the compile unit was.
- modify the "SymbolFileDWARFDebugMap" to subclass lldb_private::Module so that the virtual "GetObjectFile()" and "GetSymbolVendor()" functions can be intercepted when the .o file contenst are later lazilly needed. Prior to this fix, when we first instantiated the "SymbolFileDWARFDebugMap" class, we would also make modules, object files and symbol files for every .o file in the debug map because we needed to fix up the sections in the .o files with information that is in the executable debug map. Now we lazily do this in the DebugMapModule::GetObjectFile()
Cleaned up header includes a bit as well.
llvm-svn: 162860
particularly in the SBThread & SBFrame interfaces. Instead of filling the whole context & then getting
the API mutex, we now get only the target, acquire the API mutex from it, then fill out the rest of the
context. This removes a race condition where you get a ThreadSP, then wait on the API mutex while another
command Destroy's the Thread you've just gotten.
Also fixed the ExecutionContextRef::Get*SP calls so they don't return invalid objects.
Also fixed the ExecutionContext::Has*Scope calls so they don't claim to have a scope if the object representing
that scope has been destroyed.
Also fixed a think-o in Thread::IsValid which was causing it to return the opposite of the desired value.
<rdar://problem/11995490>
llvm-svn: 162401
- no setting auto completion
- very manual and error prone way of getting/setting variables
- tons of code duplication
- useless instance names for processes, threads
Now settings can easily be defined like option values. The new settings makes use of the "OptionValue" classes so we can re-use the option value code that we use to set settings in command options. No more instances, just "does the right thing".
llvm-svn: 162366
Ed Maste