Summary:
NPL used to be peppered with casts of the NativeThreadProtocol objects into NativeThreadLinux. I
move these closer to the source where we obtain these objects. This way, the rest of the code can
assume we are working with the correct type of objects.
Reviewers: ovyalov, tberghammer
Subscribers: lldb-commits
Differential Revision: http://reviews.llvm.org/D12187
llvm-svn: 245681
Summary:
There was a bug in NativeProcessLinux, where doing an instruction-level single-step over the
thread-creation syscall resulted in loss of control over the inferior. This happened because
after the inferior entered the thread-creation maintenance stop, we unconditionally performed a
PTRACE_CONT, even though the original intention was to do a PTRACE_SINGLESTEP. This is fixed by
storing the original state of the thread before the stop (stepping or running) and then
performing the appropriate action when resuming.
I also get rid of the callback in the ThreadContext structure, which stored the lambda used to
resume the thread, but which was not used consistently.
A test verifying the correctness of the new behavior is included.
Reviewers: ovyalov, tberghammer
Subscribers: lldb-commits
Differential Revision: http://reviews.llvm.org/D12104
llvm-svn: 245545
Summary:
This commit integrates MainLoop into NativeProcessLinux. By registering a SIGCHLD handler with
the llgs main loop, we can get rid of the special monitor thread in NPL, which saves as a lot of
thread ping-pong when responding to client requests (e.g. qThreadInfo processing time has been
reduced by about 40%). It also makes the code simpler, IMHO.
Reviewers: ovyalov, clayborg, tberghammer, chaoren
Subscribers: lldb-commits
Differential Revision: http://reviews.llvm.org/D11150
This is a resubmission of r242305 after it was reverted due to bad interactions with the stdio
thread.
llvm-svn: 242783
Summary:
This commit integrates MainLoop into NativeProcessLinux. By registering a SIGCHLD handler with
the llgs main loop, we can get rid of the special monitor thread in NPL, which saves as a lot of
thread ping-pong when responding to client requests (e.g. qThreadInfo processing time has been
reduced by about 40%). It also makes the code simpler, IMHO.
Reviewers: ovyalov, clayborg, tberghammer, chaoren
Subscribers: lldb-commits
Differential Revision: http://reviews.llvm.org/D11150
llvm-svn: 242305
Summary:
This commit avoids the Platform instance when spawning or attaching to a process in lldb-server.
Instead, I have the server call a (static) method of NativeProcessProtocol directly. The reason
for this is that I believe that NativeProcessProtocol should be decoupled from the Platform
(after all, it always knows which platform it is running on, unlike the rest of lldb).
Additionally, the kind of platform actions a NativeProcessProtocol instance is likely to differ
greatly from the platform actions of the lldb client, so I think the separation makes sense.
After this, the only dependency NativeProcessLinux has on PlatformLinux is the ResolveExecutable
method, which needs additional refactoring.
This is a resubmit of r241672, after it was reverted due to build failueres on non-linux
platforms.
Reviewers: ovyalov, clayborg
Subscribers: lldb-commits
Differential Revision: http://reviews.llvm.org/D10996
llvm-svn: 241796
platform-specific symbols that are not implemented on OS X.
The build error that caused this is
Undefined symbols for architecture x86_64:
"lldb_private::NativeProcessProtocol::Attach(unsigned long long, lldb_private::NativeProcessProtocol::NativeDelegate&, std::__1::shared_ptr<lldb_private::NativeProcessProtocol>&)", referenced from:
lldb_private::process_gdb_remote::GDBRemoteCommunicationServerLLGS::AttachToProcess(unsigned long long) in liblldb-core.a(GDBRemoteCommunicationServerLLGS.o)
"lldb_private::NativeProcessProtocol::Launch(lldb_private::ProcessLaunchInfo&, lldb_private::NativeProcessProtocol::NativeDelegate&, std::__1::shared_ptr<lldb_private::NativeProcessProtocol>&)", referenced from:
lldb_private::process_gdb_remote::GDBRemoteCommunicationServerLLGS::LaunchProcess() in liblldb-core.a(GDBRemoteCommunicationServerLLGS.o)
ld: symbol(s) not found for architecture x86_64
clang: error: linker command failed with exit code 1 (use -v to see invocation)
llvm-svn: 241688
Summary:
This commit avoids the Platform instance when spawning or attaching to a process in lldb-server.
Instead, I have the server call a (static) method of NativeProcessProtocol directly. The reason
for this is that I believe that NativeProcessProtocol should be decoupled from the Platform
(after all, it always knows which platform it is running on, unlike the rest of lldb).
Additionally, the kind of platform actions a NativeProcessProtocol instance is likely to differ
greatly from the platform actions of the lldb client, so I think the separation makes sense.
After this, the only dependency NativeProcessLinux has on PlatformLinux is the ResolveExecutable
method, which needs additional refactoring.
Reviewers: ovyalov, clayborg
Subscribers: lldb-commits
Differential Revision: http://reviews.llvm.org/D10996
llvm-svn: 241672
Summary:
This changes PtraceWrapper to return an Error, while the actual result is in an pointer parameter
(instead of the other way around). Also made a couple of PtraceWrapper arguments default to zero.
This arrangement makes a lot of the code much simpler.
Test Plan: Tests pass on linux. It compiles on android arm64/mips64.
Reviewers: chaoren, mohit.bhakkad
Subscribers: tberghammer, aemerson, lldb-commits
Differential Revision: http://reviews.llvm.org/D10808
llvm-svn: 241079
Summary:
This removes a lot of boilerplate, which was needed to execute monitor operations. Previously one
needed do declare a separate class for each operation which would manually capture all needed
arguments, which was very verbose. In addition to less code, I believe this also makes the code
more readable, since now the implementation of the operation can be physically closer to the code
that invokes it.
Test Plan: Code compiles on x86, arm and mips, tests pass on x86 linux.
Reviewers: tberghammer, chaoren
Subscribers: aemerson, lldb-commits
Differential Revision: http://reviews.llvm.org/D10694
llvm-svn: 240772
Summary:
This should solve the issue of sending denormalized paths over gdb-remote
if we stick to GetPath(false) in GDBRemoteCommunicationClient, and let the
server handle any denormalization.
Reviewers: ovyalov, zturner, vharron, clayborg
Reviewed By: clayborg
Subscribers: tberghammer, emaste, lldb-commits
Differential Revision: http://reviews.llvm.org/D9728
llvm-svn: 238604
This change reorganize the register read/write code inside lldb-server on Linux
with moving the architecture independent code into a new class called
NativeRegisterContextLinux and all of the architecture dependent code into the
appropriate NativeRegisterContextLinux_* class. As part of it the compilation of
the architecture specific register contexts are only compiled on the specific
architecture because they can't be used in other cases.
The purpose of this change is to remove a lot of duplicated code from the different
register contexts and to remove the architecture dependent codes from the global
NativeProcessLinux class.
Differential revision: http://reviews.llvm.org/D9935
llvm-svn: 238196
Summary:
There was an issue in NPL, where we attempted removal of temporary breakpoints (used to implement
software single stepping), while some threads of the process were running. This is a problem
since we currently always use the main thread's ID in the removal ptrace call. Therefore, if the
main thread was still running, the ptrace call would fail, and the software breakpoint would
remain, causing all kinds of problems. This change removes the breakpoints after all threads have
stopped. This fixes TestExitDuringStep on Android arm and can also potentially help in other
situations, as previously the breakpoint would not get removed if the thread stopped for another
reason.
Test Plan: TestExitDuringStep passes, other tests remain unchanged.
Reviewers: tberghammer
Subscribers: tberghammer, aemerson, lldb-commits
Differential Revision: http://reviews.llvm.org/D9792
llvm-svn: 237448
Summary:
NPL::Resume attempted to handle eStateStopped as a resume action. However:
- GDBRemoteCommunicationServerLLGS (the only user of NPL) never sets this action
- it could set this action in response to a vCont:t packet, but LLDB never produces this packet
- gdb-remote protocol documentation says vCont:t packet is used only in non-stop mode, but LLDB
does not support non-stop mode
- even if LLDB supported non-stop mode, this implementation of eStateStopped does something
different from what the spec says it should (according to spec, it should stop the specified
thread, but this seems to want to stop all threads).
Given the facts above, I believe we should remove this unused and untested code, as it probably
doesn't even work and removing it makes the rest of the code noticably simpler.
Reviewers: ovyalov, chaoren
Subscribers: lldb-commits
Differential Revision: http://reviews.llvm.org/D9657
llvm-svn: 237103
Summary:
Since the former-TSC events are now processed synchronously, there is no need for to protect them
with a separate mutex - all the actions are now guarded by the big m_threads_mutex.
With the mutex gone, the following functions, no longer have any purpose and were removed:
NotifyThreadCreate: replaced by direct calls to ThreadWasCreated
NotifyThreadStop: replaced by direct calls to ThreadDidStop
NotifyThreadDeath: folded into StopTrackingThread
ResetForExec: inlined as it consisted of a single line of code
RequestThreadResume(AsNeeded): replaced by direct calls to ResumeThread
StopThreads: removed, as it was never called
Test Plan: tests continue to pass
Reviewers: ovyalov, chaoren
Subscribers: lldb-commits
Differential Revision: http://reviews.llvm.org/D9603
llvm-svn: 237101
Summary:
Now that all thread events are processed synchronously, there is no need to have separate records
of whether a thread is running. This changes the (ever-dwindling) remains of the TSC to use
NativeThreadLinux as the authoritative source of the state of threads. The rest of the
ThreadContext we need has been moved to a member of NTL.
Test Plan: ninja check-lldb continues to pass
Reviewers: chaoren, ovyalov
Subscribers: lldb-commits
Differential Revision: http://reviews.llvm.org/D9562
llvm-svn: 236983
Summary:
The stop callback is a remnant of the ThreadStateCoordinator. We don't need it now that TSC is
gone, as we know exactly which function to call when threads stop. This also removes some
stop-related functions, which were just forwarding calls to one another.
Test Plan: ninja check-lldb continues to pass
Reviewers: chaoren, ovyalov
Subscribers: lldb-commits
Differential Revision: http://reviews.llvm.org/D9531
llvm-svn: 236814
Summary:
These are remnants of the thread state coordinator, which are now unnecessary. I have basically
inlined the callbacks. No functional change.
Test Plan: Tests continue to pass.
Reviewers: chaoren, vharron
Subscribers: lldb-commits
Differential Revision: http://reviews.llvm.org/D9343
llvm-svn: 236707
Summary:
The lambda was always calling SetState(eStateStopped) with small variations, so I have inlined
the code. Given that we don't have the TSC anymore, I believe we don't need to be so generic.
The only major change here is the way we choose a stop reason thread when we're interrupting a
program on client request. Previously, we were setting a null stop reason for all threads and
then fixing up the reason for one victim thread in the lambda. Now, I make sure the stop reason
is set for the victim thread correctly in the first place.
I also take the opportunity to rename CallAfter* functions into something more appropriate.
Test Plan: All tests continue to pass.
Reviewers: chaoren, vharron
Subscribers: lldb-commits
Differential Revision: http://reviews.llvm.org/D9321
llvm-svn: 236595
Summary:
Since all TSC operations are now executed synchronously, TSC has become a little more than a
messenger between different parts of NativeProcessLinux. Therefore, the reason for its existance
has disappeared.
This commit moves the contents of the TSC into the NPL class. This will enable us to remove all
the boilerplate code in NPL (as it stands now, this is most of the class), which I plan to do in
subsequent commits.
Unfortunately, this also means we will lose the unit tests for the TSC. However, since the size
of the TSC has diminished, the unit tests were not testing much at this point anyway, so it's not
a big loss.
No functional change.
Test Plan: All tests continue to pass.
Reviewers: vharron, chaoren
Subscribers: lldb-commits
Differential Revision: http://reviews.llvm.org/D9296
llvm-svn: 236587
Summary:
This change removes the thread state coordinator thread by making all the operations it was
performing synchronous. In order to prevent deadlock, NativeProcessLinux must now always call
m_monitor->DoOperation with the m_threads_mutex released. This is needed because HandleWait
callbacks lock the mutex (which means the monitor thread will block waiting on whoever holds the
lock). If the other thread now requests a monitor operation, it will wait for the monitor thread
do process it, creating a deadlock.
To preserve this invariant I have introduced two new Monitor commands: "begin operation block"
and "end operation block". They begin command blocks the monitor from processing waitpid
events until the corresponding end command, thereby assuring the monitor does not attempt to
acquire the mutex.
Test Plan: Run the test suite locally, verify no tests fail.
Reviewers: vharron, chaoren
Subscribers: lldb-commits
Differential Revision: http://reviews.llvm.org/D9227
llvm-svn: 236501
Summary:
NativeProcessProtocol uses ReadMemory internally for setting/checking
breakpoints but also for generic memory reads (Handle_m), this change adds a
ReadMemoryWithoutTrap for that purpose. Also fixes a bunch of misuses of addr_t
as size/length.
Test Plan: `disassemble` no longer shows the trap code.
Reviewers: jingham, vharron, clayborg
Reviewed By: clayborg
Subscribers: lldb-commits
Differential Revision: http://reviews.llvm.org/D9330
llvm-svn: 236132
Summary:
Without the synchronisation between the two thread creation events the following case could
happen:
- threads A and B are running. A hits a breakpoint. We note that we want to stop B.
- before we could stop it, B creates a new thread C, we get the stop notification for B, but we
don't record C's existence yet.
- we resume B
- before we get the C notification, B stops again (e.g. hits a breakpoint, gets our SIGSTOP,
etc.)
- we see all known threads have stopped, and we notify LLDB
- C notification comes, we note it's existence and resume it
=> we have an inconsistent state (LLDB thinks we've stopped, but C is running)
I resolve this by doing a blocking wait for for the C notification when we get the creation
notification on the parent (B) thread. This way the two events are synchronised, but we don't
need to introduce the intermediate "launching" state which would complicate handling of thread
states as all code would need to be aware of the third possible state.
Test Plan:
This is an obscure corner case, which I had not observed in practise, so I have no
test for it. I have tested that this commit does not regress in existing tests though.
Reviewers: chaoren, vharron
Subscribers: lldb-commits
Differential Revision: http://reviews.llvm.org/D9217
llvm-svn: 235969
The following situation occured if we were stopping a process (due to breakpoint, watchpoint, ...
hit) while a new thread was being created.
- process has two threads: A and B.
- thread A hits a breakpoint: we send a STOP signal to thread B and register a callback with
ThreadStateCoordinator to send a stop notification after the thread stops.
- thread B stops, but not due to the SIGSTOP, but on a thread creation event (of a new thread C).
We are unaware of our desire to stop, so we queue ThreadStopped and RequestResume operations
with TSC, so the thread can continue running.
- TSC receives the ThreadStopped event, sees that all threads are stopped and fires the delayed
stop notification.
- immediately after that TSC gets the RequestResume operation, so it resumes the thread.
At this point the state is inconsistent because LLDB thinks the process is stopped and will start
issuing commands to it, but one of the threads is in fact running. Things eventually break.
I address this problem by omitting the two TSC events altogether and Resuming the thread B
directly. This way the short stop is invisible to the TSC and the delayed notification will not
fire. We will fire the notification when we actually process the SIGSTOP on thread B.
When we get the initial SIGSTOP for thread C, we also resume the thread and send a
ThreadWasCreated message (is_stopped = false) to the TSC. This way, the TSC can stop the thread
on its own and handle the stop event later. This way the state of the new thread is correctly
handled as well (thanks Chaoren for the idea).
This patch also removes the synchronisation between the thread creation notifications on threads
B and C. The need for this synchronisation is unclear (the comments seem to hint that the new
thread is "fully created" only after we process both events, but I have noticed no regressions in
treating it as "created" even after just processing the initial C event), but it is a source for
many kinds of obscure races, since it introduces a new thread state "Launching" and the rest of
the code does not handle this state at all (what happens if we get a resume request from LLDB
while this thread is launching? what happens if we get a stop request? etc.).
This fixes the "spurious $O packet" problem in TestPrintStackTraces.py. However, the test remains
disabled on i386 due to the VDSO issue.
Test Plan:
TestPrintStackTraces works on x86_64. No regressions in the rest of the test suite.
Reviewers: vharron, chaoren
Subscribers: lldb-commits
Differential Revision: http://reviews.llvm.org/D9145
llvm-svn: 235579
On linux-arm we use software single stepping where setting the new
breakpoint is only possible while the process is in stopped state.
This CL moves the setup code for single stepping form the SigneStep
operation into the Resum method to avoid an error when the process
already started when we want to step one of the thread.
Differential revision: http://reviews.llvm.org/D9108
llvm-svn: 235494
Summary:
This commit moves the functionality of the operation thread into the new monitor thread. This is
required to avoid a kernel race between the two threads and I believe it actually makes the code
cleaner.
Test Plan: Ran the test suite a couple of times, no regressions.
Reviewers: ovyalov, tberghammer, vharron
Subscribers: tberghammer, lldb-commits
Differential Revision: http://reviews.llvm.org/D9080
llvm-svn: 235304
Summary:
This is the first phase of the merging of Monitor and Operation threads in NativeProcessLinux
(which is necessary since the two threads race inside Linux kernel). Here, I reimplement the
Monitor thread do use non-blocking waitpid calls, which enables later addition of code from the
operation thread.
Test Plan: Ran the test suite a couple of times, no regressions detected.
Reviewers: vharron, ovyalov, tberghammer
Subscribers: lldb-commits
Differential Revision: http://reviews.llvm.org/D9048
llvm-svn: 235193
Linux arm don't support hardware stepping (neither mismatch
breakpoints). This patch implement signle stepping with doing a software
emulation of the next instruction and then setting a temporary
breakpoint at the address where the thread will stop next.
Differential revision: http://reviews.llvm.org/D8976
llvm-svn: 234987
Previously the remote module sepcification was fetched only from the
remote platform. With this CL if we have a remote process then we ask it
if it have any information from a given module. It is required because
on android the dynamic linker only reports the name of the SO file and
the platform can't always find it without a full path (the process can
do it based on /proc/<pid>/maps).
Differential revision: http://reviews.llvm.org/D8547
llvm-svn: 233061
This CL change the logic used to terminate the monitor thread of
NativeProcessLinux to use a signal instead of pthread_cancel as
pthread_cancel is not supported on android.
Differential revision: http://reviews.llvm.org/D8205
llvm-svn: 232155
This was previously initialized by ProcessGDBRemote::Initialize but lldb-server does not contain ProcessGDBRemote anymore so this needs to be initialized directly.
Differential Revision: http://reviews.llvm.org/D8186
llvm-svn: 231966
Previously it was initialized by ProcessLinux but lldb-server don't
contain ProcessLinux anymore so it have to be initialized by
NativeProcessLinux also.
Differential revision: http://reviews.llvm.org/D8080
llvm-svn: 231482
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
* 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
With this change, both local-process llgs and remote-target llgs stdout/stderr
handling from inferior work correctly.
Several log lines have been added around PTY and stdout/stderr redirection
logic on the lldb client side.
Regarding remote llgs execution, see the following:
With these changes, remote llgs with $O now works properly:
$ lldb
(lldb) platform select remote-linux
(lldb) target create ~/some/inferior/exe
(lldb) gdb-remote {some-target}:{port}
(lldb) run
The sequence above will correctly redirect stdout/stderr over gdb-remote $O,
as is needed for remote debugging. That sequence assumes there is a lldb-gdbserver
exe running on the target with {some-host}:{port}.
You can replace the gdb-remote command with a '(lldb) platform connect
connect://{target-ip}:{target-port}'. If you do this and have a
lldb-platform running on the remote end, it will go ahead and launch
llgs for lldb for each target instance that is run/attached.
For local debugging with llgs, the following sequence also works, and
uses local PTYs instead to avoid $O and extra gdb-remote messages:
$ lldb
(lldb) settings set platform.plugin.linux.use-llgs true
(lldb) target create ~/some/inferior/exe
(lldb) run
The above will run the inferior using llgs on the local host, and
will use PTYs rather than $O redirection.
This change also removes the logging that happened after the fork but
before the exec when llgs is launching a new inferior process. Some
aspect of the file handling during that portion of code would not do
the right thing with log handling. We might want to go back later
and have that communicate over a pipe from the child to parent to pass
along any messages that previously were logged in that section of code.
llvm-svn: 219578
* Sends a SIGSTOP to the process.
* Fixes busted SIGSTOP handling. Now builds a list of non-stopped
that we wait for the PTRACE group-stop for. When the final must-stop
tid gets its group stop, we propagate the process state change.
Only the signal receiving the notification of the pending SIGSTOP
is marked with the SIGSTOP signal. All the rest, if they weren't
already stopped, are marked as stopped with signal 0.
* Fixes a few broken tests.
* Marks the Linux test I added earlier as expect-pass (no longer XFAIL).
Implements fix for http://llvm.org/bugs/show_bug.cgi?id=20908.
llvm-svn: 217647
This patch moves creates a thread abstraction that represents a
thread running inside the LLDB process. This is a replacement for
otherwise using lldb::thread_t, and provides a platform agnostic
interface to managing these threads.
Differential Revision: http://reviews.llvm.org/D5198
Reviewed by: Jim Ingham
llvm-svn: 217460
This change brings in lldb-gdbserver (llgs) specifically for Linux x86_64.
(More architectures coming soon).
Not every debugserver option is covered yet. Currently
the lldb-gdbserver command line can start unattached,
start attached to a pid (process-name attach not supported yet),
or accept lldb attaching and launching a process or connecting
by process id.
The history of this large change can be found here:
https://github.com/tfiala/lldb/tree/dev-tfiala-native-protocol-linux-x86_64
Until mid/late April, I was not sharing the work and continued
to rebase it off of head (developed via id tfiala@google.com). I switched over to
user todd.fiala@gmail.com in the middle, and once I went to github, I did
merges rather than rebasing so I could share with others.
llvm-svn: 212069
Pavel Labath