I added the ability for a process plug-in to implement custom commands. All the lldb_private::Process plug-in has to do is override:
virtual CommandObject *
GetPluginCommandObject();
This object returned should be a multi-word command that vends LLDB commands. There is a sample implementation in ProcessGDBRemote that is hollowed out. It is intended to be used for sending a custom packet, though the body of the command execute function has yet to be implemented!
llvm-svn: 165861
Then make the Thread a Broadcaster, and get it to broadcast when the selected frame is changed (but only from the Command Line) and when Thread::ReturnFromFrame
changes the stack.
Made the Driver use this notification to print the new thread status rather than doing it in the command.
Fixed a few places where people were setting their broadcaster class by hand rather than using the static broadcaster class call.
<rdar://problem/12383087>
llvm-svn: 165640
if we have a kernel binary, set the target's architecture to match.
Include the target's architecture in the ModuleSpec when we're searching for the
kext binaries on the local system -- otherwise we won't get a specific slice of
a fat file picked out for us and we won't use the returned Module correctly.
Remove the redundant attempt to find a file on the local filesystem from this method.
In ProcessGDBRemote::CheckForKernel(), if we have a kernel binary in memory, mark
the canJIT as false. There is no jitting code in kernel debug sessions.
llvm-svn: 165357
remove the duplicates of this code in ProcessGDBRemote and ProcessKDP.
These two Process plugins will hardcode their DynamicLoader name to be
the DynamicLoaderDarwinKernel so the correct DynamicLoader is picked,
and return the kernel load address as the ImageInfosAddress.
<rdar://problem/12417038>
llvm-svn: 165080
When attaching to a remote system that does not look like a typical vendor system, and no
executable binary was specified to lldb, check a couple of fixed locations where kernels
running in ASLR mode (slid in memory to a random address) store their load addr when booted
in debug mode, and relocate the symbols or load the kernel wholesale from the host computer
if we can find it.
<rdar://problem/7714201>
llvm-svn: 164888
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
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
Added code the initialize the register context in the OperatingSystemPython plug-in with the new PythonData classes, and added a test OperatingSystemPython module in lldb/examples/python/operating_system.py that we can use for testing.
llvm-svn: 162530
Previously we put a WatchpointSentry object within StopInfo.cpp to disable-and-then-enable the watchpoint itself
while we are performing the actions associated with the triggered watchpoint, which can cause the user-initiated
watchpoint disabling action to be negated.
Add a test case to verify that a watchpoint can be disabled during the callbacks.
llvm-svn: 162483
Convert from calling Halt in the lldb Driver.cpp's input reader's sigint handler to sending this AsyncInterrupt so it can be handled in the
event loop.
If you are attaching and get an async interrupt, abort the attach attempt.
Also remember to destroy the process if get interrupted while attaching.
Getting this to work also required handing the eBroadcastBitInterrupt in a few more places in Process WaitForEvent & friends.
<rdar://problem/10792425>
llvm-svn: 160903
calling functions. This is necessary on Mac OS X, since bad things can happen if you set
the registers of a thread that's sitting in a kernel trap.
<rdar://problem/11145013>
llvm-svn: 160756
Fixed an issue where GDB servers that don't support the thread suffix could get registers states incorrectly due to an incorrect assumption that the current register thread (set using the "Hg%x" packet) will always be cached between runs. Now we clear the cached register thred when the process is resumed.
llvm-svn: 159603
than being given the pthread_mutex_t from the Mutex and locks that. That allows us to
track ownership of the Mutex better.
Used this to switch the LLDB_CONFIGURATION_DEBUG enabled assert when we can't get the
gdb-remote sequence mutex to assert when the thread that had the mutex releases it. This
is generally more useful information than saying just who failed to get it (since the
code that had it locked often had released it by the time the assert fired.)
llvm-svn: 158240
m_interrupt_sent into account. Also don't reset m_interrupt_sent in SendInterrupt but do so in SendPacketAndWaitForResponse
when we know we've handled the interrupt.
Fix a code path through ProcessGDBRemote::DoDestroy where we were tearing down the debug session but
not setting the exit status.
llvm-svn: 158043
Fixed a case where multiple threads can be asking to send a packet to the GDB server and one of three things will happen:
1 - everything works
2 - one thread will fail to send the packet due to not being able to get the sequence mutex
3 - one thread will try and interrupt the other packet sending and fail and not send the packet
Now the flow is a bit different. Prior to this fix we did:
if (try_get_sequence_mutex()) {
send_packet()
return success;
} else {
if (async_ok) {
interrupt()
send_packet()
resume()
return success;
}
}
return fail
The issue is that the call to "try_get_sequence_mutex()" could fail if another thread was sending a packet and could cause us to just not send the packet and an error would be returned.
What we really want is to try and get the sequence mutex, and if this succeeds, send the packet. Else check if we are running and if we are, do what we used to do. The big difference is when we aren't running, we wait for the sequence mutex so we don't drop packets. Pseudo code is:
if (try_get_sequence_mutex()) {
// Safe to send the packet right away
send_packet()
return success;
} else {
if (running) {
// We are running, interrupt and send async packet if ok to do so,
// else it is ok to fail
if (async_ok) {
interrupt()
send_packet()
resume()
return success;
}
}
else {
// Not running, wait for the sequence mutex so we don't drop packets
get_sequence_mutex()
send_packet()
return success;
}
}
return fail
llvm-svn: 157751
Sending async packets can deadlock a program on darwin. We currently allow breakpoint packets and memory read/write packets (for software breakpoints) to be sent while a program is running. In the GDB remote plug-in, we will interrupt the run, send the async packet and resume (currently with the continue packet that caused the program to resume). If the GDB server supports the "vCont" packet, we might have initially continued with each thread stating it should continue. If new threads show up while we are stopped, which happend when running GCD, we can end up with new threads that we aren't mentioning in the continue list. So we start with a thread list of 1,2,3 and continue:
continue thread 1, continue thread 2, continue thread 3
Now we interrupt and set a breakpoint and we actually have threads 1,2,3,4 now when we are about to resume, yet we send:
continue thread 1, continue thread 2, continue thread 3
Any thread that isn't mentioned is currently going to stay suspended. This causes the deadlock.
llvm-svn: 157439
Add default Process::GetWatchpointSupportInfo() impl which returns an error of "not supported".
Add "qWatchpointSupportInfo" packet to the gdb communication layer to support this, and modify TestWatchpointCommands.py to test it.
llvm-svn: 157345
the value_regs field, which is useful for future expansion purposes. As of now, we have:
calculated_offset_of_eax = offset_of_rax + (offset_of_eax_from_the_descriptor which is 0)
llvm-svn: 157275
Add convenience registers eax, ebx, ecx, edx, edi, esi, ebp, esp to the 'register read' command for x86_64.
Add a GDBRemoteRegisterContext::Addx86_64ConvenienceRegisters() method called from ProcessGDBRemote::BuildDynamicRegisterInfo().
Servicing of eax, for example, is accomplished by delegating to rax with an adjusted offset into the register context.
llvm-svn: 157230
that dynamically discovers remote register context information.
o GDBRemoteRegisterContext.h:
Change the prototype of HardcodeARMRegisters() to take a boolean flag, which now becomes
void
HardcodeARMRegisters(bool from_scratch);
o GDBRemoteRegisterContext.cpp:
HardcodeARMRegisters() now checks the from_scratch flag and decides whether to add composite registers to the already
existing primordial registers based on a table called g_composites which describes the composite registers.
o ProcessGDBRemote.cpp:
Modify the logic of ProcessGDBRemote::BuildDynamicRegisterInfo() to call m_register_info.HardcodeARMRegisters()
with the newly introduced 'bool from_scrach' flag.
rdar://problem/10652076
llvm-svn: 156773
Switch over to the "*-apple-macosx" for desktop and "*-apple-ios" for iOS triples.
Also make the selection process for auto selecting platforms based off of an arch much better.
llvm-svn: 156354
No one was using it and Locker(pthread_mutex_t *) immediately asserts for
pthread_mutex_t's that don't come from a Mutex anyway. Rather than try to make
that work, we should maintain the Mutex abstraction and not pass around the
platform implementation...
Make Mutex::Locker::Lock take a Mutex & or a Mutex *, and remove the constructor
taking a pthread_mutex_t *. You no longer need to call Mutex::GetMutex to pass
your mutex to a Locker (you can't in fact, since I made it private.)
llvm-svn: 156221
us of its architecture, use that to set the Target's arch if it
doesn't already have one set.
In Process::CompleteAttach(), if the Target has a valid arch make
sure that the Platform we pick up is compatible with that arch; if
not, find a Platform that is compatible. Don't let the the default
platform override the Target's arch.
<rdar://problem/11185420>
llvm-svn: 156116
Enable logging the packet history when registers fail to read due to not getting the sequence mutex if "--verbose" is enabled on the log channel for the "gdb-remote" log category.
This will help us track down some issues.
llvm-svn: 154704
The less locks there are, the better. I removed the thread ID mutex and now just shared the m_thread_list's mutex to make sure we don't deadlock due to lock inversion.
llvm-svn: 154652
for packet confirmation.
Also added a bit more logging.
Also, unlock the writer end of the run lock in Process.cpp on our way out of the private state
thread so that the Process can shut down cleanly.
<rdar://problem/11228538>
llvm-svn: 154601
Cleaned up the Mutex::Locker and the ReadWriteLock classes a bit.
Also cleaned up the GDBRemoteCommunication class to not have so many packet functions. Used the "NoLock" versions of send/receive packet functions when possible for a bit of performance.
llvm-svn: 154458
QListThreadsInStopReply
This GDB remote query command can enable added a "threads" key/value pair to all stop reply packets so that we always get a list of all threads in each stop reply packet. It increases performance if enabled (the reply to the "QListThreadsInStopReply" is "OK") by saving us from sending to command/reply pairs (the "qfThreadInfo" and "qsThreadInfo" packets), and also helps us keep the current process state up to date.
llvm-svn: 154380
The next step is to have our stop reply packets send the thread list in the actual stop reply packet to avoid a 2 packet overhead of sending the qfThreadInfo + response and qfThreadInfo + response.
llvm-svn: 154376
The current ProcessGDBRemote function that updates the threads could end up with an empty list if any other thread had the sequence mutex. We now don't clear the thread list when we can't access it, and we also have changed how lldb_private::Process handles the return code from the:
virtual bool
Process::UpdateThreadList (lldb_private::ThreadList &old_thread_list,
lldb_private::ThreadList &new_thread_list) = 0;
A bool is now returned to indicate if the list was actually updated or not and the lldb_private::Process class will only update the stop ID of the validity of the thread list if "true" is returned.
The ProcessGDBRemote also got an extra assertion that will hopefully assert when running debug builds so we can find the source of this issue.
llvm-svn: 154365
spin up a temporary "private state thread" that will respond to events from the lower level process plugins. This check-in should work to do
that, but it is still buggy. However, if you don't call functions on the private state thread, these changes make no difference.
This patch also moves the code in the AppleObjCRuntime step-through-trampoline handler that might call functions (in the case where the debug
server doesn't support the memory allocate/deallocate packet) out to a safe place to do that call.
llvm-svn: 154230
<rdar://problem/11051056>
Found a race condition when sending async packets in the ProcessGDBRemote.
A little background: GDB remote clients can only send one packet at a time. You must send a packet and wait for a response. So when we continue, we obviously can't hold up the calling thread waiting for the process to stop again, so we have an async thread in the ProcessGDBRemote whose only job is to run packets that control the inferior process. When you send a continue packet, the only packet you can send is an interrupt packet (which consists of sending a CTRL+C (or a '\x03' byte)). This then stops the inferior and we can send the async packet, and then resume the target. There was a race condition that often happened during stepping where we are doing a source level single step which consists of many instruction steps and a few runs here and there when we step into a function. So the flow looks like:
inst single step
inst single step
inst single step
inst single step
inst single step
step BP and run
inst single step
inst single step
inst single step
Now if we got an async packet while the program is running we get something like:
send --> continue
send --> interrupt
recv <-- interrupt stop reply packet
send --> async packet
recv <-- async response
send --> continue again and wait for actual stop
Problems arise when this was happening when single stepping a thread where we would get:
send --> step thread 123
send --> interrupt
send --> stop reply for thread 123 (from the step)
Now we _might_ have an extra stop reply packet from the "interrupt" which we weren't checking for and we could end up with:
send --> async packet (like memory read!)
recv <-- async response (which is the interrupt stop reply packet)
Now we have the read memroy reply sitting in our buffer and waiting to be used as the reply for the next packet...
To further complicate things, the single step should have exited the async thread since the run control is finished, but now it will continue if it was interrupted.
The fixes I checked in to two major things:
- watch for the extra stop reply if we need to
- make sure we exit from the async thread run loop when the previous run control (like the instruction level single step) is finished.
Needless to say this makes very fast stepping in Xcode much more reliable.
llvm-svn: 153629
This fix really needed to happen as a previous fix I had submitted for
calculating symbol sizes made many symbols appear to have zero size since
the function that was calculating the symbol size was calling another function
that would cause the calculation to happen again. This resulted in some symbols
having zero size when they shouldn't. This could then cause infinite stack
traces and many other side affects.
llvm-svn: 152244
Fixed STDERR to not be opened as readable. Also cleaned up some of the code that implemented the file actions as some of the code was using the wrong variables, they now use the right ones (in for stdin, out for stdout, err for stderr).
llvm-svn: 152102
Incremental check in to calculate the offsets of registers correctly. Registers can be primordial or composite,
for example, r0-r12 are primordial, s0-s31 are primordial, while q0 is composite consisting of (s0, s1, s2, s3).
Modify q0-q8 to be composed of the primordial s0-s31 registers.
llvm-svn: 151757
Initial step -- infrastructure change -- to fix the bug. Change the RegisterInfo data structure
to contain two additional fields (uint32_t *value_rges and uint32_t *invalidate_regs) to facilitate
architectures which have register mapping.
Update all existing RegsiterInfo arrays to have two extra NULL's (the additional fields) in each row,
GDBRemoteRegisterContext.cpp is modified to add d0-d15 and q0-q15 register info entries which take
advantage of the value_regs field to specify the containment relationship:
d0 -> (s0, s1)
...
d15 -> (s30, s31)
q0 -> (d0, d1)
...
q15 -> (d30, d31)
llvm-svn: 151686
more of the local path, platform path, associated symbol file, UUID, arch,
object name and object offset. This allows many of the calls that were
GetSharedModule to reduce the number of arguments that were used in a call
to these functions. It also allows a module to be created with a ModuleSpec
which allows many things to be specified prior to any accessors being called
on the Module class itself.
I was running into problems when adding support for "target symbol add"
where you can specify a stand alone debug info file after debugging has started
where I needed to specify the associated symbol file path and if I waited until
after construction, the wrong symbol file had already been located. By using
the ModuleSpec it allows us to construct a module with as little or as much
information as needed and not have to change the parameter list.
llvm-svn: 151476
objects for the backlink to the lldb_private::Process. The issues we were
running into before was someone was holding onto a shared pointer to a
lldb_private::Thread for too long, and the lldb_private::Process parent object
would get destroyed and the lldb_private::Thread had a "Process &m_process"
member which would just treat whatever memory that used to be a Process as a
valid Process. This was mostly happening for lldb_private::StackFrame objects
that had a member like "Thread &m_thread". So this completes the internal
strong/weak changes.
Documented the ExecutionContext and ExecutionContextRef classes so that our
LLDB developers can understand when and where to use ExecutionContext and
ExecutionContextRef objects.
llvm-svn: 151009
user space programs. The core file support is implemented by making a process
plug-in that will dress up the threads and stack frames by using the core file
memory.
Added many default implementations for the lldb_private::Process functions so
that plug-ins like the ProcessMachCore don't need to override many many
functions only to have to return an error.
Added new virtual functions to the ObjectFile class for extracting the frozen
thread states that might be stored in object files. The default implementations
return no thread information, but any platforms that support core files that
contain frozen thread states (like mach-o) can make a module using the core
file and then extract the information. The object files can enumerate the
threads and also provide the register state for each thread. Since each object
file knows how the thread registers are stored, they are responsible for
creating a suitable register context that can be used by the core file threads.
Changed the process CreateInstace callbacks to return a shared pointer and
to also take an "const FileSpec *core_file" parameter to allow for core file
support. This will also allow for lldb_private::Process subclasses to be made
that could load crash logs. This should be possible on darwin where the crash
logs contain all of the stack frames for all of the threads, yet the crash
logs only contain the registers for the crashed thrad. It should also allow
some variables to be viewed for the thread that crashed.
llvm-svn: 150154
due to RTTI worries since llvm and clang don't use RTTI, but I was able to
switch back with no issues as far as I can tell. Once the RTTI issue wasn't
an issue, we were looking for a way to properly track weak pointers to objects
to solve some of the threading issues we have been running into which naturally
led us back to std::tr1::weak_ptr. We also wanted the ability to make a shared
pointer from just a pointer, which is also easily solved using the
std::tr1::enable_shared_from_this class.
The main reason for this move back is so we can start properly having weak
references to objects. Currently a lldb_private::Thread class has a refrence
to its parent lldb_private::Process. This doesn't work well when we now hand
out a SBThread object that contains a shared pointer to a lldb_private::Thread
as this SBThread can be held onto by external clients and if they end up
using one of these objects we can easily crash.
So the next task is to start adopting std::tr1::weak_ptr where ever it makes
sense which we can do with lldb_private::Debugger, lldb_private::Target,
lldb_private::Process, lldb_private::Thread, lldb_private::StackFrame, and
many more objects now that they are no longer using intrusive ref counted
pointer objects (you can't do std::tr1::weak_ptr functionality with intrusive
pointers).
llvm-svn: 149207
if this is a mapped/executable region of memory. If it isn't, we've jumped
through a bad pointer and we know how to unwind the stack correctly based
on the ABI.
Previously I had 0x0 special cased but if you jumped to 0x2 on x86_64 one
frame would be skipped because the unwinder would try using the x86_64
ArchDefaultUnwindPlan which relied on the rbp.
Fixes <rdar://problem/10508291>
llvm-svn: 146477
will allow us to represent a process/thread ID using a pointer for the OS
plug-ins where they might want to represent the process or thread ID using
the address of the process or thread structure.
llvm-svn: 145644
1 - the DIE collections no longer have the NULL tags which saves up to 25%
of the memory on typical C++ code
2 - faster parsing by not having to run the SetDIERelations() function anymore
it is done when parsing the DWARF very efficiently.
llvm-svn: 144983
from a process and hooked it up to the new packet that was recently added
to our GDB remote executable named debugserver. Now Process has the following
new calls:
virtual Error
Process::GetMemoryRegionInfo (lldb::addr_t load_addr, MemoryRegionInfo &range_info);
virtual uint32_t
GetLoadAddressPermissions (lldb::addr_t load_addr);
Only the first one needs to be implemented by subclasses that can add this
support.
Cleaned up the way the new packet was implemented in debugserver to be more
useful as an API inside debugserver. Also found an error where finding a region
for an address actually will pick up the next region that follows the address
in the query so we also need ot make sure that the address we requested the
region for falls into the region that gets returned.
llvm-svn: 144976
turned out to be unitialized data in the ProcessLaunchInfo default constructor.
Turning on MallocScribble in the environment helped track this down.
When we launch and attach using the host layer, we now inform the process that
it shouldn't detach when by calling an accessor.
llvm-svn: 144882
After recent changes we weren't reaping child processes resulting in many
zombie processes.
This was fixed by adding more settings to the ProcessLaunchOptions class
that allow clients to specify a callback function and baton to be notified
when their process dies. If one is not supplied a default callback will be
used that "does the right thing".
Cleaned up a race condition in the ProcessGDBRemote class that would attempt
to monitor when debugserver died.
Added an extra boolean to the process monitor callbacks that indicate if a
process exited or not. If your process exited with a zero exit status and no
signal, both items could be zero.
Modified the process monitor functions to not require a callback function
in order to reap the child process.
llvm-svn: 144780
info for us to attach by pid, or by name and will also allow us to eventually
do a lot more powerful attaches. If you look at the options for the "platform
process list" command, there are many options which we should be able to
specify. This will allow us to do things like "attach to a process named 'tcsh'
that has a parent process ID of 123", or "attach to a process named 'x' which
has an effective user ID of 345".
I finished up the --shell implementation so that it can be used without the
--tty option in "process launch". The "--shell" option now can take an
optional argument which is the path to the shell to use (or a partial name
like "sh" which we will find using the current PATH environment variable).
Modified the Process::Attach to use the new ProcessAttachInfo as the sole
argument and centralized a lot of code that was in the "process attach"
Execute function so that everyone can take advantage of the powerful new
attach functionality.
llvm-svn: 144615
on internal only (public API hasn't changed) to simplify the paramter list
to the launch calls down into just one argument. Also all of the argument,
envronment and stdio things are now handled in a much more centralized fashion.
llvm-svn: 143656
Fixed an issue where async packets were incurring a delay even though they
were sent correctly. We now properly broadcast the private run state being
resumed correctly. Also fixed logging to reflect what is happening.
llvm-svn: 143154
lldb_private::Error objects the rules are:
- short strings that don't start with a capitol letter unless the name is a
class or anything else that is always capitolized
- no trailing newline character
- should be one line if possible
Implemented a first pass at adding "--gdb-format" support to anything that
accepts format with optional size/count.
llvm-svn: 142999
process IDs, and thread IDs, but was mainly needed for for the UserID's for
Types so that DWARF with debug map can work flawlessly. With DWARF in .o files
the type ID was the DIE offset in the DWARF for the .o file which is not
unique across all .o files, so now the SymbolFileDWARFDebugMap class will
make the .o file index part (the high 32 bits) of the unique type identifier
so it can uniquely identify the types.
llvm-svn: 142534
Jim Ingham