Tool for using Intel(R) Processor Trace hardware feature

[abhishek.aggarwal]

Summary

1. Provide single library for all Intel specific hardware features instead of individual libraries for each feature
2. Added Intel(R) Processor Trace hardware feature in this single library. Details about the tool implementing this feature is as follows:
   
   Tool developed on top of LLDB to provide its users the execution trace of the debugged inferiors. Tool's API are exposed as C++ object oriented interface in a shared library. API are designed especially to be easily integrable with IDEs providing LLDB as an application debugger. Entire API is also available as Python functions through a script bridging interface allowing development of python modules.
   
   This patch also provides a CLI wrapper to use the Tool through LLDB's command line. Highlights of the Tool and the wrapper are given below:
   
   ****** Intel(R) Processor Trace Tool: ******
   • • Provides execution trace of the debugged application
     • Uses Intel(R) Processor Trace hardware feature (already implemented inside LLDB) for this purpose
   • Collects trace packets generated by this feature from LLDB, decodes and post-processes them
   • Constructs the execution trace of the application
   • Presents execution trace as a list of assembly instructions
     • Provides 4 APIs (exposed as C++ object oriented interface)
   • start trace with configuration options for a thread/process,
   • stop trace for a thread/process,
   • get the execution flow (assembly instructions) for a thread,
   • get trace specific information for a thread
     • Easily integrable into IDEs providing LLDB as application debugger
     • Entire API available as Python functions through script bridging interface
   • Allows developing python apps on top of Tool
     • README_TOOL.txt provides more details about the Tool, its dependencies, building steps and API usage
     • Tool ready to use through LLDB's command line
   • CLI wrapper has been developed on top of the Tool for this purpose
     
     ******* CLI wrapper: cli-wrapper-pt.cpp *******
     • Provides 4 commands (syntax similar to LLDB's CLI commands):
   • processor-trace start
   • processor-trace stop
   • processor-trace show-trace-options
   • processor-trace show-instr-log
     • README_CLI.txt provides more details about commands and their options

Signed-off-by: Abhishek Aggarwal <[e-mail]>

[clayborg]

So we don't want clients of SBStructuredData to have to text scrape and write manual decoders. See my inlined comments on the minimal modifications we need to make to SBStructureData first. This then removes a ton of code from your patch. The SBStructureData modifications should probably be done in separate patch as well. internally lldb_private::StructuredData has everything (dicts, array, strings, integers, bools, nulls) inherit from lldb_private::StructuredData::Object. Then SBStructuredData owns a StructuredDataImpl which has a lldb_private::StructuredData::Object shared pointer, so a lldb::SBStructureData can be any object (dicts, array, strings, integers, bools, nulls). So we should expose the ability to get data from SBStructuredData. See my inlined SBStructureData additions I proposed. We really don't want anyone doing manual JSON text scraping.

tools/CMakeLists.txt

8

Can we default this to enabled?

tools/intel-pt/Decoder.cpp

18 ↗

(On Diff #98434)

This kind of reach in to internal plug-in sources shouldn't happen as it violates the boundaries. Remove and see comment below.

27 ↗

(On Diff #98434)

We really shouldn't be interpreting integer error codes here. The string in the "sberror" should be what you use. Modify the code that creates these errors in the first place to also put a string values you have here into the lldb_private::Error to begin with and remove this function.

177 ↗

(On Diff #98434)

Why do we care which debugger this belongs to?

200 ↗

(On Diff #98434)

We meed to add API to SBStructuredData to expose some of the stuff from lldb_private::StructuredData so you don't end up text scraping here. Just do what you need for now but I would suggest at the very least:

class SBStructuredData {
  enum Type {
    eTypeArray,
    eTypeDictionary,
    eTypeString,
    eTypeInteger,
    eTypeBoolean
  }
  // Get the type of data in this structured data.
  Type GetType();
  // Get a dictionary for a key value given a key from the top level of this structured data if type is eTypeDictionary
  SBStructuredData GetValueForKey(const char *key);
  // Get the value of this object as a string if type is eTypeString
  bool GetStringValue(char *s, size_t max_len);
  // Get an integer of this object as an integer if type is eTypeInteger
  bool GetIntegerValue(uint64_t &value);
  ...
};

See cleanup code below if we do this.

211–212 ↗

(On Diff #98434)

if (!sbstructdata.GetValueForKey("trace-tech").IsValid())

218–219 ↗

(On Diff #98434)

if (!sbstructdata.GetValueForKey("intel-pt").IsValid())

538–547 ↗

(On Diff #98434)

No text scraping. Please use SBStructureData methods that we will need to add.

602–603 ↗

(On Diff #98434)

Remove this function, add real API to SBStructuredData

1046–1047 ↗

(On Diff #98434)

Remove this function?

tools/intel-pt/Decoder.h

86 ↗

(On Diff #98434)

Should this be named better? IntelTraceOptions? Also does it need to inherit from lldb::SBTraceOptions? Would it make more sense to have this be a "has a" where SBTraceOptions is a member variable?

90 ↗

(On Diff #98434)

Does this need to be virtual? Remember that lldb::SBTraceOptions ins't virtual since it is in the public API.

131–134 ↗

(On Diff #98434)

Do you actually need the debugger here? It might be nice to avoid this and just get the debugger from the process:

auto debugger = process.GetTarget().GetDebugger();

177 ↗

(On Diff #98434)

You probably don't need this function as you shouldn't have to store the SBDebugger anywhere in this class as you can always get it from the process:

auto debugger = process.GetTarget().GetDebugger();

Why do we care which debugger this belongs to?

[krytarowski]

Can we please use the generic process plugin code, not the specific Linux one? If the generic code is insufficient - we should adjust it and alter instances of it (Linux and BSD).

[clayborg]

In D33035#751264, @krytarowski wrote:

Can we please use the generic process plugin code, not the specific Linux one? If the generic code is insufficient - we should adjust it and alter instances of it (Linux and BSD).

There should be no need to ever #include something from a plug-in in code the gets loaded externally. We should rely on the error strings themselves, not be comparing them to some value. SBError has an error type (generic, posix, mach-kernel, expression results, windows). It we are going to do anything we can make up a new set of them, but the code in this patch is relying on the values, when it should just be using the string value. If the string value wasn't set correctly, fix that is the code that creates the errors.

[krytarowski]

In D33035#751280, @clayborg wrote:

In D33035#751264, @krytarowski wrote:

Can we please use the generic process plugin code, not the specific Linux one? If the generic code is insufficient - we should adjust it and alter instances of it (Linux and BSD).

There should be no need to ever #include something from a plug-in in code the gets loaded externally. We should rely on the error strings themselves, not be comparing them to some value. SBError has an error type (generic, posix, mach-kernel, expression results, windows). It we are going to do anything we can make up a new set of them, but the code in this patch is relying on the values, when it should just be using the string value. If the string value wasn't set correctly, fix that is the code that creates the errors.

Thank you for explanation. This is even better and ideal.

[labath]

I don't really like that we are adding a public shared library for every tiny intel feature. Could we at least merge this "plugin" with the existing "intel-mpx plugin" to create one "intel support" library?

Also, adding an external dependency probably deserves a discussion on lldb-dev.

tools/CMakeLists.txt

8

We probably can't, as this code depends on a third party library. In any case, this option should go to LLDBConfig.cmake

tools/intel-pt/CMakeLists.txt

42 ↗

(On Diff #98434)

any reason you're not using add_lldb_library here?

53 ↗

(On Diff #98434)

All of this needs to go away. I think you only needed it because you are plucking NativeProcessLinux internals, so fixing that should fix this too.

[abhishek.aggarwal]

tools/CMakeLists.txt

8

The tool has a dependency on a decoder library. People who are not interested in building this tool will have to explicitly set this flag OFF to avoid build failures caused by absence of this decoder library on their system. I wanted to avoid that. But if you think enabling it by default is a better idea then I will change it.

tools/intel-pt/Decoder.cpp

18 ↗

(On Diff #98434)

I am removing this include from here.

27 ↗

(On Diff #98434)

Removing this function.

177 ↗

(On Diff #98434)

Please see my reply to your comment in Decoder.h file.

200 ↗

(On Diff #98434)

I am on it.

538–547 ↗

(On Diff #98434)

Working on it.

602–603 ↗

(On Diff #98434)

Working on it.

1046–1047 ↗

(On Diff #98434)

Please see reply to your comment in Decoder.h file.

tools/intel-pt/Decoder.h

86 ↗

(On Diff #98434)

1. I am changing its name to "TraceOptions".
2. Inheriting it from SBTraceOptions makes life easier in the sense of reusing the APIs without re-defining them here. This is a backend class. Exposing any new API in PTInfo class (in case SBTraceOptions class undergo any change/addition in future) will require change only in PTDecoder.h file and corresponding implementation file. Plus, concept wise both "SBTraceOptions" and "Info" represent trace options. These were the motives behind using inheritance. Am I missing some important thing here? Please let me know.

90 ↗

(On Diff #98434)

You are right. I will remove it.

131–134 ↗

(On Diff #98434)

Please see my reply to your next comment below in this file.

177 ↗

(On Diff #98434)

As per my understanding, the Unique IDs of SBProcess instances will be unique within 1 SBDebugger instance but will not be unique across two different SBDebugger instances. If I am wrong then I don't need to store Debugger ID here. Otherwise, for each SBProcess instance provided by user through public APIs of the Tool, I will have to check which particular SBDebugger instance it belongs to. Please correct me if I am wrong.

[abhishek.aggarwal]

In D33035#754640, @labath wrote:

I don't really like that we are adding a public shared library for every tiny intel feature. Could we at least merge this "plugin" with the existing "intel-mpx plugin" to create one "intel support" library?

Also, adding an external dependency probably deserves a discussion on lldb-dev.

Hi Paval ... Before starting the development of this whole feature, we had submitted the full proposal to lldb dev list 1.5 years ago. During the discussions, it was proposed to keep the external dependency outside LLDB (i.e. not to be bundled in liblldb shared library). The External dependency required for this tool is not and will never be a part of lldb repository. Users who are interested to use this tool, will download this external dependency separately.

tools/intel-pt/CMakeLists.txt

42 ↗

(On Diff #98434)

No. I will try with that.

[labath]

In D33035#767029, @abhishek.aggarwal wrote:

In D33035#754640, @labath wrote:

I don't really like that we are adding a public shared library for every tiny intel feature. Could we at least merge this "plugin" with the existing "intel-mpx plugin" to create one "intel support" library?

Also, adding an external dependency probably deserves a discussion on lldb-dev.

Hi Paval ... Before starting the development of this whole feature, we had submitted the full proposal to lldb dev list 1.5 years ago. During the discussions, it was proposed to keep the external dependency outside LLDB (i.e. not to be bundled in liblldb shared library). The External dependency required for this tool is not and will never be a part of lldb repository. Users who are interested to use this tool, will download this external dependency separately.

Yes I remember that. But, as you say, that was 1.5 years ago, and we haven't heard anything since. Honestly, I had assumed you abandoned that work until you reappeared last month. :)
So I think it's worth updating that thread, as new things may have come up since then (for example, the decision whether to merge the intel stuff into a single library).

[abhishek.aggarwal]

Single feature library development for all intel specific features

• Merged the developed tool for Intel(R) Processor Trace hardware feature provided in this patch and already existing tool for Intel(R) Memory Protection Extensions support

[abhishek.aggarwal]

tools/intel-pt/CMakeLists.txt

53 ↗

(On Diff #98434)

Fixing this.

tools/intel-pt/Decoder.cpp

27 ↗

(On Diff #98434)

Currently, the codes are generated by lldb server and remote packets don't support sending error strings from lldb server to lldb client.
Now, I can think of 2 ways:

1. modify remote packets to send error strings as well (in addition to error codes).
2. or decode error codes and generate error strings at lldb client side

Which one do you prefer? @labath prefers 1st one (in discussions of https://reviews.llvm.org/D33674). If you also prefers 1st one then I (or @ravitheja) can work on modifying packets accordingly and submit a separate patch for that.

My idea is to keep error codes and its decoding logic here for now. Once we submit patch of modified packets and that patch gets approved, I can remove this decoding function from here all together. Please let me know what you prefer.

[labath]

tools/intel-features/CMakeLists.txt

51

"bin" sounds wrong here. Shouldn't this go ti lib${LLVM_LIBDIR_SUFFIX}?
To properly handle DLL targets (i don't know whether ipt support windows) you'd need something like (see function add_lldb_library):

install(TARGETS ${name}
  COMPONENT ${name}
  RUNTIME DESTINATION bin
  LIBRARY DESTINATION lib${LLVM_LIBDIR_SUFFIX}
  ARCHIVE DESTINATION lib${LLVM_LIBDIR_SUFFIX})

Although it may be than you can just call that function yourself.

tools/intel-features/intel-mpx/CMakeLists.txt

6

Normally we build a single .a file for each source folder, which are then linked into other targets as appropriate, and I don't see a reason to deviate from this here.

Same goes for the ipt subfolder.

tools/intel-features/scripts/python-typemaps.txt

2

Could we just use standard lldb typemaps? I don't see anything ipt specific here...

tools/intel-pt/Decoder.cpp

27 ↗

(On Diff #98434)

How about we just avoid interpreting the error codes for now and print a generic "error 47" message instead. This avoids the awkward state, where we have two enums that need to be kept in sync, which is a really bad idea.

I think having more generic error code will be very useful - there are a lot of packets that would benefit from more helpful error messages.

[abhishek.aggarwal]

Hi Pavel .. My comments are inlined. Please let me know if you have more concerns. I am submitting the patch with all the proposed changes.

tools/intel-features/CMakeLists.txt

51

Fixing it.

tools/intel-features/intel-mpx/CMakeLists.txt

6

Agreed. I am submitting the changes.

tools/intel-features/scripts/python-typemaps.txt

2

Unfortunately, we can't because that file uses lldb_private::PythonString, lldb_private::PythonList etc and for that I will have to link to liblldbPluginScripInterpreterPython.

tools/intel-pt/Decoder.cpp

27 ↗

(On Diff #98434)

I am removing this function as @ravitheja is working on enabling lldb remote packets to send error strings directly. So, we don't even need "error 47" message.

[abhishek.aggarwal]

Changes after feedback

• Created static libs for each hardware feature to combine it to generate single shared library
• Removed error codes decoding logic and directly using error strings

[labath]

Hi Abhishek,

Thank you for incorporating the changes. I still see some room for improvement in the cmake files. I realize that this is something most people are not familiar with, and is not exactly glamorous work, but it's still part of our codebase and we should make sure it follows best practices.

tools/intel-features/CMakeLists.txt

17

Could we avoid building the shared library altogether if both features are OFF?

tools/intel-features/cli-wrapper.cpp

28

You will need some ifdef magic to make sure these still compile when the feature is off.

tools/intel-features/intel-mpx/CMakeLists.txt

10

What you want here is to define an INTERFACE dependency on the MPX library instead.
vanilla cmake way would be target_link_libraries(lldbIntelMPX INTERFACE LLVMSupport). However, we should use the llvm function instead, as that also handles other llvm-specific magic (for example, this code will break if someone does a LLVM_LINK_LLVM_DYLIB build).

So, I am asking for the third time:
Have you tried using add_lldb_library instead?

The correct invocation should be add_lldb_library(foo.cpp LINK_LIBS Support) and the rest of this file can just go away.

tools/intel-features/scripts/lldb-intel-features.swig

10

There appear to be no typedefs here.

15

You are already defining this as a part of the swig invocation in cmake.

tools/intel-features/scripts/python-typemaps.txt

2

Ok, let's leave this as-is then. We could try factoring out common part of those typemaps, but I'm not sure that's such a good idea at the moment.

[abhishek.aggarwal]

Hi Pavel .. I have made the changes you suggested. My apologies for misinterpreting your previous comments but during written communications, it is sometimes difficult to interpret everything correctly. I have tried following LLDB's coding conventions and guidelines. Please let me know if I still missed things that you would have liked to see in this patch. Thanks for your patience :)

tools/intel-features/CMakeLists.txt

17

yes. Adding the check.

tools/intel-features/cli-wrapper.cpp

28

Fixed it.

tools/intel-features/intel-mpx/CMakeLists.txt

10

I am extremely sorry Pavel but I understood it now what you were trying to say in previous comments. Sorry about misinterpreting your comments before. I have used add_lldb_library function now. Please see them in the next patch set.

tools/intel-features/scripts/lldb-intel-features.swig

10

Forgot to remove this. Doing it.

15

removing it.

[abhishek.aggarwal]

Cmake files related changes

• Using lldb's cmake functions insted of vanilla ones for cmake files

[abhishek.aggarwal]

Fixed one minor thing in CMakeFile

[labath]

Thank you for making the changes. I concede that I am not always clear in my communications, and I apologize if I was being rude.

I am happy with the cmake stuff, but I noticed a new issue now. Exceptions are banned in llvm, so we need to figure out how to remove them from this code.

tools/intel-features/intel-mpx/CMakeLists.txt

10

Looks much better. Thanks.

tools/intel-features/intel-pt/Decoder.cpp

411

We can't have exceptions in llvm code. You will have to achieve this differently. Your trick with manually adding -fexceptions will not work anyway if the rest of the code is compiled without exceptions. Although I'm not really sure why you need to protect this vector append in particular, as we don't do this for any other vector elsewhere.

[abhishek.aggarwal]

Thanks for your review Pavel. My comments are inlined. Let me know your opinion :)

tools/intel-features/intel-pt/Decoder.cpp

411

I kept the exception handling around stl containers only to catch bad_alloc exception because if this exception occurs, I didn't want the code to just exit but provide user with whatever amount of instruction log is available in the vector. That much amount of instruction log might still be helpful to the user. What is your opinion on that?

Plus, If rest of the code is not being compiled with -fexceptions but just this file, will it not solve the purpose? Let me know what you think about it. I can make changes accordingly then.

[labath]

tools/intel-features/intel-pt/Decoder.cpp

411

I don't think there's any negotiating on this point (not with me anyway, you'd need to take this much higher up). But here's what I think anyway:

• the exception catch will be generally useless as most (all?) current OSs will overcommit virtual memory (and then just kill you if they really run out of memory and swap space)
• even if you disable overcommit, chances are you will hit an OOM in one of the zillion other places which allocate memory (all of which are unchecked) instead of here. So this single catch will not make a difference.

[abhishek.aggarwal]

tools/intel-features/intel-pt/Decoder.cpp

411

Got it. Then I remove exception handling code from here. Submit the patch again. Thanks for elaborating.

[abhishek.aggarwal]

Removed exception handling code

[abhishek.aggarwal]

Hi Pavel .. I could remove all exception handling code from source files. However, I am still wondering how to disable exception handling while providing python functions for C++ APIs of the features. Can you suggest something here? The whole problem is occurring because of the use of vector<> as an argument to one of the C++ APIs. To provide a python interface for these APIs, I am forced to include std_vector.i (please see tools/intel-features/intel-pt/interface/PTDecoder.i file) which is introducing exception handling code.

[labath]

In D33035#798857, @abhishek.aggarwal wrote:

Hi Pavel .. I could remove all exception handling code from source files. However, I am still wondering how to disable exception handling while providing python functions for C++ APIs of the features. Can you suggest something here? The whole problem is occurring because of the use of vector<> as an argument to one of the C++ APIs. To provide a python interface for these APIs, I am forced to include std_vector.i (please see tools/intel-features/intel-pt/interface/PTDecoder.i file) which is introducing exception handling code.

Hm... that's a bit of a drag. I guess the SB API never ran into this problem because it always provides it's own vector-like classes (SBModuleList, SBFileSpecList, etc.). I guess the most canonical way would be to follow that example and have your own class for a list of instructions. However, that is a bit annoying, so I do see a case for making code generated by swig as an exception to the rule.

[abhishek.aggarwal]

In D33035#798885, @labath wrote:

In D33035#798857, @abhishek.aggarwal wrote:

Hi Pavel .. I could remove all exception handling code from source files. However, I am still wondering how to disable exception handling while providing python functions for C++ APIs of the features. Can you suggest something here? The whole problem is occurring because of the use of vector<> as an argument to one of the C++ APIs. To provide a python interface for these APIs, I am forced to include std_vector.i (please see tools/intel-features/intel-pt/interface/PTDecoder.i file) which is introducing exception handling code.

Hm... that's a bit of a drag. I guess the SB API never ran into this problem because it always provides it's own vector-like classes (SBModuleList, SBFileSpecList, etc.). I guess the most canonical way would be to follow that example and have your own class for a list of instructions. However, that is a bit annoying, so I do see a case for making code generated by swig as an exception to the rule.

Hmm .. Is the decision about enabling exception handling only for swig generated code lie in LLDB's community's hands or we will have to ask LLVM community for this too? I can add another class to resolve this problem but I am not sure whether it is the right way to go.

[labath]

In D33035#798925, @abhishek.aggarwal wrote:

In D33035#798885, @labath wrote:

Hm... that's a bit of a drag. I guess the SB API never ran into this problem because it always provides it's own vector-like classes (SBModuleList, SBFileSpecList, etc.). I guess the most canonical way would be to follow that example and have your own class for a list of instructions. However, that is a bit annoying, so I do see a case for making code generated by swig as an exception to the rule.

Hmm .. Is the decision about enabling exception handling only for swig generated code lie in LLDB's community's hands or we will have to ask LLVM community for this too? I can add another class to resolve this problem but I am not sure whether it is the right way to go.

I think we can start with an email to lldb-dev and see how it goes. Nobody on the llvm side uses swig.

I'm not sure what would be a better solution either... I don't like either of them :)

[clayborg]

So std::vector shouldn't be used in a public API. You should make a class that wraps your objects. LLDB's public API has lldb::SBInstruction and lldb::SBInstructionList as examples. std::vector on other systems compiles differently for debug and release builds and things will crash. If you need this via swig for internal kind of stuff, then use a typemap where you map std::vector<T> to a list() with T instances in it in python.

I am a bit confused here as well. Are we exporting a plug-in specific python bindings for the Intel PT data? It seems like it would be nice to wrap this API into the SBTrace or other lldb interface? Am I not understanding this correctly?

[abhishek.aggarwal]

In D33035#799058, @clayborg wrote:

So std::vector shouldn't be used in a public API. You should make a class that wraps your objects. LLDB's public API has lldb::SBInstruction and lldb::SBInstructionList as examples. std::vector on other systems compiles differently for debug and release builds and things will crash.

Hi Greg .. Does it go for the tools built on top of LLDB (e.g. in my case where feature library is not a part of liblldb shared lib but built on top of it)? If yes, then I will proceed to remove std::vector from C++ public API of the tool and create a new class for it.

If you need this via swig for internal kind of stuff, then use a typemap where you map std::vector<T> to a list() with T instances in it in python.

I want to provide a python interface for the tool's C++ public API as well so that the API can be used in python modules as well. Therefore, I think typemapping to list() will not solve the problem. Am I right?

I am a bit confused here as well. Are we exporting a plug-in specific python bindings for the Intel PT data? It seems like it would be nice to wrap this API into the SBTrace or other lldb interface? Am I not understanding this correctly?

Probably, I didn't understand this comment of yours. We are not exporting python binding for Intel PT data. It is a python binding for Tool's C++ API and this Tool is built on top of LLDB. Did I answer your question? Please let me know if I misunderstood your comment.

[clayborg]

In D33035#799404, @abhishek.aggarwal wrote:

In D33035#799058, @clayborg wrote:

So std::vector shouldn't be used in a public API. You should make a class that wraps your objects. LLDB's public API has lldb::SBInstruction and lldb::SBInstructionList as examples. std::vector on other systems compiles differently for debug and release builds and things will crash.

Hi Greg .. Does it go for the tools built on top of LLDB (e.g. in my case where feature library is not a part of liblldb shared lib but built on top of it)? If yes, then I will proceed to remove std::vector from C++ public API of the tool and create a new class for it.

It really comes down to the issue of how stable you want this API to be. If this API is a "you are expected to rebuild any tools you make against this API every time we release a new version" then you can really do anything you want. If you want a stable API to build against, you could choose to follow the LLDB public API rules: no inheritance, fixed number of member variables that never change (which usually means a std::shared_ptr<T>, std::unique_ptr<T> or a T*), and no virtual functions. This effectively renders the C++ API into a C API, but it does force you to have your public API classes have an internal implementation (T) and then the class that you export for your API be the public facing API that everyone uses externally. The nice thing about this is that swig has a really easy time creating all of the script bindings for you if you do it this way. So it really depends on what your future use case of this API is.

If you need this via swig for internal kind of stuff, then use a typemap where you map std::vector<T> to a list() with T instances in it in python.

I want to provide a python interface for the tool's C++ public API as well so that the API can be used in python modules as well. Therefore, I think typemapping to list() will not solve the problem. Am I right?

I believe it would fix your issue for you. You write the glue code that can translate any argument (std::vector<ptdecoder::PTInstruction>) into something more python like like a list of PTInstruction python objects. We do this in a lot of places with LLDB.

I am a bit confused here as well. Are we exporting a plug-in specific python bindings for the Intel PT data? It seems like it would be nice to wrap this API into the SBTrace or other lldb interface? Am I not understanding this correctly?

Probably, I didn't understand this comment of yours. We are not exporting python binding for Intel PT data. It is a python binding for Tool's C++ API and this Tool is built on top of LLDB. Did I answer your question? Please let me know if I misunderstood your comment.

Now I understand. I didn't realize this was a stand alone tool. The main thing to figure out is how stable you want this C++ API that you are exporting to be. If that isn't an issue, feel free to use anything you want in that API. If stability is an issue you want to solve and you want to vend a C++ API, you might think about adopting LLDB's public API rules to help ensure stability.

[abhishek.aggarwal]

In D33035#799656, @clayborg wrote:

In D33035#799404, @abhishek.aggarwal wrote:

In D33035#799058, @clayborg wrote:

So std::vector shouldn't be used in a public API. You should make a class that wraps your objects. LLDB's public API has lldb::SBInstruction and lldb::SBInstructionList as examples. std::vector on other systems compiles differently for debug and release builds and things will crash.

Hi Greg .. Does it go for the tools built on top of LLDB (e.g. in my case where feature library is not a part of liblldb shared lib but built on top of it)? If yes, then I will proceed to remove std::vector from C++ public API of the tool and create a new class for it.

It really comes down to the issue of how stable you want this API to be. If this API is a "you are expected to rebuild any tools you make against this API every time we release a new version" then you can really do anything you want. If you want a stable API to build against, you could choose to follow the LLDB public API rules: no inheritance, fixed number of member variables that never change (which usually means a std::shared_ptr<T>, std::unique_ptr<T> or a T*), and no virtual functions. This effectively renders the C++ API into a C API, but it does force you to have your public API classes have an internal implementation (T) and then the class that you export for your API be the public facing API that everyone uses externally. The nice thing about this is that swig has a really easy time creating all of the script bindings for you if you do it this way. So it really depends on what your future use case of this API is.

If you need this via swig for internal kind of stuff, then use a typemap where you map std::vector<T> to a list() with T instances in it in python.

I want to provide a python interface for the tool's C++ public API as well so that the API can be used in python modules as well. Therefore, I think typemapping to list() will not solve the problem. Am I right?

I believe it would fix your issue for you. You write the glue code that can translate any argument (std::vector<ptdecoder::PTInstruction>) into something more python like like a list of PTInstruction python objects. We do this in a lot of places with LLDB.

I am a bit confused here as well. Are we exporting a plug-in specific python bindings for the Intel PT data? It seems like it would be nice to wrap this API into the SBTrace or other lldb interface? Am I not understanding this correctly?

Probably, I didn't understand this comment of yours. We are not exporting python binding for Intel PT data. It is a python binding for Tool's C++ API and this Tool is built on top of LLDB. Did I answer your question? Please let me know if I misunderstood your comment.

Now I understand. I didn't realize this was a stand alone tool. The main thing to figure out is how stable you want this C++ API that you are exporting to be. If that isn't an issue, feel free to use anything you want in that API. If stability is an issue you want to solve and you want to vend a C++ API, you might think about adopting LLDB's public API rules to help ensure stability.

Hi Greg .. Thanks a lot for your feedback and detailed explanation. I am choosing to follow what LLDB SB API are following. Removing std::vector<> from interface. Submitting a new patch.

[abhishek.aggarwal]

Removed std::vector<> from public APIs

[clayborg]

Much better on the API with Swig. Just a few inlined questions around GetRawBytes.

tools/intel-features/intel-pt/PTDecoder.h

55–56

In the python version of this function, this should return a python string that contains the bytes or return None of there is nothing. Is that how it currently works? One alternate way of doing this is doing what we do with lldb::SBProcess:

size_t ReadMemory(addr_t addr, void *buf, size_t size, lldb::SBError &error);

In python, we use a type map to detect the args "void *buf, size_t size", and we turn this into:

def ReadMemory(self, addr, error):
    return # Python string with bytes from read memory

So one option would be to change GetRawBytes() into:

size_t GetRawBytes(void *buf, size_t size) const;

The return value is now many bytes were filled in. If "buf" is NULL or "size" is zero, you can return the length that you would need as the return value.

And have the python version be:

def GetRawBytes(self):
    return # bytes in a python string or None

59

This might not be needed if GetRawBytes is changed to:

size_t GetRawBytes(void *buf, size_t size) const;

[abhishek.aggarwal]

In D33035#803801, @clayborg wrote:

Much better on the API with Swig. Just a few inlined questions around GetRawBytes.

Hi Greg .. Thanks for feedback. My comments are inlined.

tools/intel-features/intel-pt/PTDecoder.h

55–56

Hi Greg.. As per my understanding, if we change ReadMemory() function as you proposed then both typemaps:
%typemap(in) (void *buf, size_t size)
%typemap(argout) (void *buf, size_t size)

will have to be defined. Am i right?

If it is so then I am afraid that I can't reuse these maps from lldb as the 2nd typemap refers to lldb_private::PythonBytes which is an internal library of lldb. Please let me know if I am wrong.

[clayborg]

I am not sure how sensitive typemaps are to the names of the arguments? Maybe we can just have you use different names and then the two typemaps won't collide? You can also write some manual code to do the remapping without typemaps.

Regardless of whether you are going to change the API or not, you will need to write something to make it work in python.

const uint8_t * ptdecoder_private:: Instruction::GetRawBytes() const;

What does swig do right now when the above code is converted to python? It needs to make a python string from this and the GetRawBytesSize()...

[abhishek.aggarwal]

In D33035#816337, @clayborg wrote:

I am not sure how sensitive typemaps are to the names of the arguments? Maybe we can just have you use different names and then the two typemaps won't collide? You can also write some manual code to do the remapping without typemaps.

Regardless of whether you are going to change the API or not, you will need to write something to make it work in python.

const uint8_t * ptdecoder_private:: Instruction::GetRawBytes() const;

What does swig do right now when the above code is converted to python? It needs to make a python string from this and the GetRawBytesSize()...

Hi Greg .. You are right. I checked the swig converted code for this API and it was not using GetRawBytesSize(). I have changed the C++ API signature as you had suggested and wrote a new typemap to handle it. It is mostly similar to the one defined in lldb typemaps but doesn't use internal lldb implementation of PythonBytes. Let me know if something is not right. I am submitting new patch set. Thanks a lot for your feedback :)

In D33035#816337, @clayborg wrote:

I am not sure how sensitive typemaps are to the names of the arguments? Maybe we can just have you use different names and then the two typemaps won't collide? You can also write some manual code to do the remapping without typemaps.

Regardless of whether you are going to change the API or not, you will need to write something to make it work in python.

const uint8_t * ptdecoder_private:: Instruction::GetRawBytes() const;

What does swig do right now when the above code is converted to python? It needs to make a python string from this and the GetRawBytesSize()...

[abhishek.aggarwal]

Changes to make a C++ API work in python modules

• Added a new python typemap to handle it

[abhishek.aggarwal]

Hi Greg .. Can you please review the changes? Please let me know if something needs to be changed. Thanks :)

[clayborg]

I was out on vacation, sorry for the delay. Looks good.

[abhishek.aggarwal]

In D33035#833876, @clayborg wrote:

I was out on vacation, sorry for the delay. Looks good.

Thanks a lot Greg for your feedback. I am merging the patch now.

tools/CMakeLists.txt

 if (CMAKE_SYSTEM_NAME MATCHES "Darwin")
   add_subdirectory(darwin-debug)
   add_subdirectory(debugserver)
 endif()
 add_subdirectory(argdumper)
 add_subdirectory(driver)
 add_subdirectory(lldb-mi)
 if (LLDB_CAN_USE_LLDB_SERVER)
   add_subdirectory(lldb-server)
 endif()
-add_subdirectory(intel-mpx)
+add_subdirectory(intel-features)

tools/intel-features/CMakeLists.txt

+include(${LLDB_PROJECT_ROOT}/cmake/LLDBDependencies.cmake)
+
+# Flags to control each individual feature
+option(LLDB_BUILD_INTEL_MPX "Enable Building of Intel(R) Memory Protection Extensions" ON)
+option(LLDB_BUILD_INTEL_PT "Enable Building of Intel(R) Processor Trace Tool" OFF)
+
+# Return if all features are OFF
+if (NOT LLDB_BUILD_INTEL_MPX AND NOT LLDB_BUILD_INTEL_PT)
+  return()
+endif()
+
+LIST (APPEND FEATURE_LIBS "")
+
+# Add feature specific subdirectories based on flags
+if (LLDB_BUILD_INTEL_MPX AND CMAKE_SYSTEM_NAME MATCHES "Linux")
+  add_subdirectory(intel-mpx)
+  LIST (APPEND FEATURE_LIBS ${FEATURE_LIBS} lldbIntelMPX)
+  SET (CLI_WRAPPER_PREPROCESSORS "${CLI_WRAPPER_PREPROCESSORS} -DBUILD_INTEL_MPX")
+endif()
+
+if (LLDB_BUILD_INTEL_PT)
+  add_subdirectory(intel-pt)
+  LIST (APPEND FEATURE_LIBS ${FEATURE_LIBS} lldbIntelPT)
+  SET (CLI_WRAPPER_PREPROCESSORS "${CLI_WRAPPER_PREPROCESSORS} -DBUILD_INTEL_PT")
+endif()
+
+# Add python wrapper if python not disabled
+if (NOT LLDB_DISABLE_PYTHON AND LLDB_BUILD_INTEL_PT)
+  set(LLDB_INTEL_FEATURES_PYTHON_WRAP
+     ${LLDB_BINARY_DIR}/tools/intel-features/scripts/IntelFeaturesPythonWrap.cpp)
+  set_source_files_properties(${LLDB_INTEL_FEATURES_PYTHON_WRAP}
+                              PROPERTIES GENERATED 1
+                              COMPILE_FLAGS " -fexceptions -frtti")
+
+  if (CLANG_CL)
+    set_source_files_properties(${LLDB_INTEL_FEATURES_PYTHON_WRAP}
+                                 PROPERTIES COMPILE_FLAGS -Wno-unused-function)
+  endif()
+
+  if (LLVM_COMPILER_IS_GCC_COMPATIBLE AND
+      NOT "${CMAKE_SYSTEM_NAME}" MATCHES "Darwin")
+    set_property(SOURCE ${LLDB_INTEL_FEATURES_PYTHON_WRAP}
+                 APPEND_STRING PROPERTY COMPILE_FLAGS
+                 " -Wno-sequence-point -Wno-cast-qual")
+  endif ()
+  add_subdirectory(scripts)
+endif()
+
+set_source_files_properties(cli-wrapper.cpp PROPERTIES
+                            COMPILE_FLAGS ${CLI_WRAPPER_PREPROCESSORS})
+
+add_lldb_library(lldbIntelFeatures SHARED
+  cli-wrapper.cpp
+  ${LLDB_INTEL_FEATURES_PYTHON_WRAP}
+
+  LINK_LIBS
+    ${FEATURE_LIBS}
+  )
+
+# Add link dependencies for python wrapper
+if (NOT LLDB_DISABLE_PYTHON AND LLDB_BUILD_INTEL_PT)
+  add_dependencies(lldbIntelFeatures intel-features-swig_wrapper)
+  target_link_libraries(lldbIntelFeatures PRIVATE ${LLDB_SYSTEM_LIBS})
+endif()
+
+install(TARGETS lldbIntelFeatures
+  LIBRARY DESTINATION lib${LLVM_LIBDIR_SUFFIX})

tools/intel-features/README.txt

+****************************************************************************
+*                            README                                        *
+*                                                                          *
+* This file provides all the information regarding new CLI commands that   *
+* enable using various hardware features of Intel(R) architecture based    *
+* processors from LLDB's CLI.                                              *
+****************************************************************************
+
+
+============
+Introduction
+============
+A shared library has been developed to use various hardware features of
+Intel(R) architecture based processors through LLDB's command line. The library
+currently comprises of hardware features namely Intel(R) Processor Trace and
+Intel(R) Memory Protection Extensions.
+
+
+============
+Details
+============
+A C++ based cli wrapper (cli-wrapper.cpp) has been developed here that
+agglomerates all cli commands for various hardware features. This wrapper is
+build to generate a shared library (lldbIntelFeatures) to provide all these
+commands.
+
+For each hardware feature, separate cli commands have been developed that are
+provided by wrappers (cli-wrapper-pt.cpp and cli-wrapper-mpxtable.cpp) residing
+in feature specific folders ("intel-pt" and "intel-mpx" respectively).
+
+For details regarding cli commands of each feature, please refer to these
+feature specific wrappers.
+
+
+
+============
+How to Build
+============
+The shared library (lldbIntelFeatures) has a cmake based build and can be built
+while building LLDB with cmake. "cli-wrapper.cpp" file is compiled along with all
+the feature specific source files (residing in feature specific folders).
+
+Furthermore, flexibility is provided to the user to include/exclude a particular
+feature while building lldbIntelFeatures library. This is done by flags described
+below:
+
+  - LLDB_BUILD_INTEL_PT - The flag enables building of Intel(R) Processor Trace
+    feature (inside intel-pt folder). This flag defaults to "OFF" meaning the
+    feature is excluded while building lldbIntelFeatures library. Set it to "ON"
+    in order to include it.
+
+  - LLDB_BUILD_INTEL_MPX - Enables building Intel(R) Memory Protection Extensions
+    feature (inside intel-mpx folder). This flag defaults to "ON" meaning
+    the feature is excluded while building lldbIntelFeatures library.
+
+Please refer to README files in feature specific folders to know about additional
+flags that need to be set in order to build that feature successfully.
+
+
+============
+How to Use
+============
+All CLI commands provided by this shared library can be used through the LLDB's
+CLI by executing "plugin load <shared_lib_name>" on LLDB CLI. shared_lib_name here
+is lldbIntelFeatures
+
+
+
+============
+Description
+============
+Please refer to README_CLI file of each feature to know about details of CLI
+commands.

tools/intel-features/cli-wrapper.cpp

+//===-- cli-wrapper.cpp -----------------------------------------*- C++ -*-===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+// CLI Wrapper for hardware features of Intel(R) architecture based processors
+// to enable them to be used through LLDB's CLI. For details, please refer to
+// cli wrappers of each individual feature, residing in their respective
+// folders.
+//
+// Compile this into a shared lib and load by placing at appropriate locations
+// on disk or by using "plugin load" command at the LLDB command line.
+//
+//===----------------------------------------------------------------------===//
+
+#include "intel-mpx/cli-wrapper-mpxtable.h"
+#include "intel-pt/cli-wrapper-pt.h"
+#include "lldb/API/SBDebugger.h"
+
+namespace lldb {
+bool PluginInitialize(lldb::SBDebugger debugger);
+}
+
+bool lldb::PluginInitialize(lldb::SBDebugger debugger) {
+
+#ifdef BUILD_INTEL_PT
+  PTPluginInitialize(debugger);
+#endif
+
+#ifdef BUILD_INTEL_MPX
+  MPXPluginInitialize(debugger);
+#endif
+
+  return true;
+}

tools/intel-features/intel-mpx/CMakeLists.txt

+add_lldb_library(lldbIntelMPX
+  cli-wrapper-mpxtable.cpp
+
+  LINK_LIBS
+    liblldb
+
+  LINK_COMPONENTS
+    Support
+  )

tools/intel-features/intel-mpx/cli-wrapper-mpxtable.h

+//===-- cli-wrapper-mpxtable.h----------------------------------*- C++ -*-===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+
+#include "lldb/API/SBDebugger.h"
+
+bool MPXPluginInitialize(lldb::SBDebugger& debugger);

tools/intel-features/intel-mpx/cli-wrapper-mpxtable.cpp

-//===-- IntelMPXTablePlugin.cpp----------------------------------*- C++ -*-===//
+//===-- cli-wrapper-mpxtable.cpp----------------------------------*- C++ -*-===//
 //
 //                     The LLVM Compiler Infrastructure
 //
 // This file is distributed under the University of Illinois Open Source
 // License. See LICENSE.TXT for details.
 //
 //===----------------------------------------------------------------------===//
 
 // C++ includes
 #include <cerrno>
 #include <string>
 
 // Project includes
+#include "cli-wrapper-mpxtable.h"
 #include "lldb/API/SBCommandInterpreter.h"
 #include "lldb/API/SBCommandReturnObject.h"
 #include "lldb/API/SBMemoryRegionInfo.h"
 #include "lldb/API/SBProcess.h"
 #include "lldb/API/SBTarget.h"
 #include "lldb/API/SBThread.h"
 
 #include "llvm/ADT/Triple.h"
 
-namespace lldb {
-bool PluginInitialize(lldb::SBDebugger debugger);
-}
-
 static bool GetPtr(char *cptr, uint64_t &ptr, lldb::SBFrame &frame,
                    lldb::SBCommandReturnObject &result) {
   if (!cptr) {
     result.SetError("Bad argument.");
     result.SetStatus(lldb::eReturnStatusFailed);
     return false;
   }
 
if (!frame.IsValid()) {
     result.SetError("No valid process, thread or frame.");
     result.SetStatus(lldb::eReturnStatusFailed);
     return false;
   }
 
   lldb::SBValue bndcfgu_val = frame.FindRegister("bndcfgu");
   if (!bndcfgu_val.IsValid()) {
     result.SetError(
-        "Cannot access register BNDCFGU. Does the target support MPX?");
+        "Cannot access register BNDCFGU. Does the target support "
+        "Intel(R) Memory Protection Extensions (Intel(R) MPX)?");
     result.SetStatus(lldb::eReturnStatusFailed);
     return false;
   }
 
   lldb::SBData bndcfgu_data = bndcfgu_val.GetData();
   bndcfgu = bndcfgu_data.GetUnsignedInt64(error, 0);
   if (!error.Success()) {
     result.SetError(error, "Invalid read of register BNDCFGU.");
if (command) {
                         error);
     }
 
     result.SetError("Too few arguments. See help.");
     return false;
   }
 };
 
-bool lldb::PluginInitialize(lldb::SBDebugger debugger) {
+bool MPXPluginInitialize(lldb::SBDebugger& debugger) {
   lldb::SBCommandInterpreter interpreter = debugger.GetCommandInterpreter();
   lldb::SBCommand mpxTable = interpreter.AddMultiwordCommand(
-      "mpx-table", "A utility to access the MPX table entries.");
+      "mpx-table", "A utility to access the Intel(R) MPX table entries.");
 
-  const char *mpx_show_help = "Show the MPX table entry of a pointer.\n"
-                              "mpx-table show <pointer>";
+  const char *mpx_show_help = "Show the Intel(R) MPX table entry of a pointer."
+                              "\nmpx-table show <pointer>";
   mpxTable.AddCommand("show", new MPXTableShow(), mpx_show_help);
 
   const char *mpx_set_help =
-      "Set the MPX table entry of a pointer.\n"
+      "Set the Intel(R) MPX table entry of a pointer.\n"
       "mpx-table set <pointer> <lower bound> <upper bound>";
   mpxTable.AddCommand("set", new MPXTableSet(), mpx_set_help);
 
   return true;
 }

tools/intel-features/intel-mpx/test/Makefile

 LEVEL = ../../../../make
 
 CXX_SOURCES := main.cpp
 
 CFLAGS_EXTRAS += -mmpx -fcheck-pointer-bounds -lmpxwrappers -lmpx -fuse-ld=bfd
 
 include $(LEVEL)/Makefile.rules

tools/intel-features/intel-mpx/test/README.txt

 In order to run this test, create the following directory:
 
     packages/Python/lldbsuite/test/functionalities/plugins/commands/mpxtablecmd
 
-and copy into it the contents of this direcotry.
+and copy into it the contents of this directory.
 

tools/intel-features/intel-mpx/test/TestMPXTable.py

class TestMPXTable(TestBase):
     mydir = TestBase.compute_mydir(__file__)
 
     def setUp(self):
         TestBase.setUp(self)
 
     @skipIf(compiler="clang")
     @skipIf(oslist=no_match(['linux']))
     @skipIf(archs=no_match(['i386', 'x86_64']))
-    @skipIf(compiler="gcc", compiler_version=["<", "5"]) #GCC version >= 5 supports Intel(R) MPX.
+    @skipIf(compiler="gcc", compiler_version=["<", "5"]) #GCC version >= 5 supports
+    #Intel(R) Memory Protection Extensions (Intel(R) MPX).
     def test_show_command(self):
         """Test 'mpx-table show' command"""
         self.build()
 
         lldb_exec_dir = os.environ["LLDB_IMPLIB_DIR"]
         lldb_lib_dir = os.path.join(lldb_exec_dir, os.pardir, "lib")
-        plugin_file = os.path.join(lldb_lib_dir, "liblldb-intel-mpxtable.so")
+        plugin_file = os.path.join(lldb_lib_dir, "liblldbIntelFeatures.so")
         if not os.path.isfile(plugin_file):
-            self.skipTest("Intel(R) mpx-table plugin missing.")
+            self.skipTest("features plugin missing.")
         plugin_command = " "
         seq = ("plugin", "load", plugin_file)
         plugin_command = plugin_command.join(seq)
         self.runCmd(plugin_command)
 
         exe = os.path.join(os.getcwd(), "a.out")
         self.runCmd("file " + exe, CURRENT_EXECUTABLE_SET)
 
def test_show_command(self):
                      error = False)
 
     def test_set_command(self):
         """Test 'mpx-table set' command"""
         self.build()
 
         lldb_exec_dir = os.environ["LLDB_IMPLIB_DIR"]
         lldb_lib_dir = os.path.join(lldb_exec_dir, os.pardir, "lib")
-        plugin_file = os.path.join(lldb_lib_dir, "liblldb-intel-mpxtable.so")
+        plugin_file = os.path.join(lldb_lib_dir, "liblldbIntelFeatures.so")
         if not os.path.isfile(plugin_file):
-            self.skipTest("Intel(R) mpx-table plugin missing.")
+            self.skipTest("features plugin missing.")
         plugin_command = " "
         seq = ("plugin", "load", plugin_file)
         plugin_command = plugin_command.join(seq)
         self.runCmd(plugin_command)
 
         exe = os.path.join(os.getcwd(), "a.out")
         self.runCmd("file " + exe, CURRENT_EXECUTABLE_SET)
 

tools/intel-features/intel-mpx/test/main.cpp

#endif
   tmp = (int*)0x2; // Break 2.
 
   return; // Break 3.
 }
 
 int
 main(int argc, char const *argv[])
 {
-  // This call returns 0 only if the CPU and the kernel support Intel(R) MPX.
+  // This call returns 0 only if the CPU and the kernel support
+  // Intel(R) Memory Protection Extensions (Intel(R) MPX).
   if (prctl(PR_MPX_ENABLE_MANAGEMENT, 0, 0, 0, 0) != 0)
         return -1;
 
   int*  a = (int *) calloc(size, sizeof(int));
 #if defined  __GNUC__ && !defined __INTEL_COMPILER
   __builtin___bnd_store_ptr_bounds ((void**)&a, a);
 #endif
   func(a); // Break 1.
 
   free(a); // Break 4.
 
   return 0;
 }

tools/intel-features/intel-pt/CMakeLists.txt

+if (NOT LIBIPT_INCLUDE_PATH)
+  message (FATAL_ERROR "libipt include path not provided")
+endif()
+
+if (NOT EXISTS "${LIBIPT_INCLUDE_PATH}")
+  message (FATAL_ERROR "invalid libipt include path provided")
+endif()
+include_directories(${LIBIPT_INCLUDE_PATH})
+
+if (NOT LIBIPT_LIBRARY_PATH)
+  find_library(LIBIPT_LIBRARY ipt)
+else()
+  if (NOT EXISTS "${LIBIPT_LIBRARY_PATH}")
+    message (FATAL_ERROR "invalid libipt library path provided")
+  endif()
+  find_library(LIBIPT_LIBRARY ipt PATHS ${LIBIPT_LIBRARY_PATH})
+endif()
+
+if (NOT LIBIPT_LIBRARY)
+  message (FATAL_ERROR "libipt library not found")
+endif()
+
+set_source_files_properties(PTDecoder.cpp PROPERTIES COMPILE_FLAGS -fexceptions)
+set_source_files_properties(Decoder.cpp PROPERTIES COMPILE_FLAGS -fexceptions)
+
+add_lldb_library(lldbIntelPT
+  PTDecoder.cpp
+  Decoder.cpp
+  cli-wrapper-pt.cpp
+
+  LINK_LIBS
+    ${LIBIPT_LIBRARY}
+    liblldb
+  )

tools/intel-features/intel-pt/Decoder.h

+//===-- Decoder.h -----------------------------------------------*- C++ -*-===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+
+#ifndef Decoder_h_
+#define Decoder_h_
+
+// C/C++ Includes
+#include <map>
+#include <mutex>
+#include <string>
+#include <vector>
+
+// Project includes, Other libraries and framework includes
+#include "lldb/API/SBDebugger.h"
+#include "lldb/API/SBError.h"
+#include "lldb/API/SBProcess.h"
+#include "lldb/API/SBStream.h"
+#include "lldb/API/SBStructuredData.h"
+#include "lldb/API/SBTarget.h"
+#include "lldb/API/SBTrace.h"
+#include "lldb/API/SBTraceOptions.h"
+#include "lldb/lldb-enumerations.h"
+#include "lldb/lldb-types.h"
+
+#include "intel-pt.h"
+
+namespace ptdecoder_private {
+//----------------------------------------------------------------------
+/// @class Instruction
+/// @brief Represents an assembly instruction containing raw
+///     instruction bytes, instruction address along with information
+///     regarding execution flow context and Intel(R) Processor Trace
+///     context.
+//----------------------------------------------------------------------
+class Instruction {
+public:
+  Instruction() : ip(0), data(), error(), iclass(ptic_error), speculative(0) {}
+
+  Instruction(const Instruction &insn) = default;
+
+  Instruction(const struct pt_insn &insn)
+      : ip(insn.ip), data(), error(insn.size == 0 ? "invalid instruction" : ""),
+        iclass(insn.iclass), speculative(insn.speculative) {
+    if (insn.size != 0)
+      data.assign(insn.raw, insn.raw + insn.size);
+  }
+
+  Instruction(const char *err)
+      : ip(0), data(), error(err ? err : "unknown error"), iclass(ptic_error),
+        speculative(0) {}
+
+  ~Instruction() {}
+
+  uint64_t GetInsnAddress() const { return ip; }
+
+  const uint8_t *GetRawBytes() const { return data.data(); }
+
+  uint8_t GetRawBytesSize() const { return data.size(); }
+
+  const std::string &GetError() const { return error; }
+
+  bool GetSpeculative() const { return speculative; }
+
+private:
+  uint64_t ip;               // instruction address in inferior's memory image
+  std::vector<uint8_t> data; // raw bytes
+  std::string error;         // Error string if instruction is invalid
+  enum pt_insn_class iclass; // classification of the instruction
+  // A collection of flags giving additional information about instruction
+  uint32_t speculative : 1; // Instruction was executed speculatively or not
+};
+
+//----------------------------------------------------------------------
+/// @class TraceOptions
+/// @brief Provides Intel(R) Processor Trace specific configuration options and
+///     other information obtained by decoding and post-processing the trace
+///     data. Currently, this information comprises of the total number of
+///     assembly instructions executed for an inferior.
+//----------------------------------------------------------------------
+class TraceOptions : public lldb::SBTraceOptions {
+public:
+  TraceOptions() : lldb::SBTraceOptions(), m_insn_log_size(0) {}
+
+  ~TraceOptions() {}
+
+  //------------------------------------------------------------------
+  /// Get total number of assembly instructions obtained after decoding the
+  /// complete Intel(R) Processor Trace data obtained from LLDB.
+  ///
+  /// @return
+  ///     Total number of instructions.
+  //------------------------------------------------------------------
+  uint32_t getInstructionLogSize() const { return m_insn_log_size; }
+
+  //------------------------------------------------------------------
+  /// Set total number of assembly instructions.
+  ///
+  /// @param[in] size
+  ///     Value to be set.
+  //------------------------------------------------------------------
+  void setInstructionLogSize(uint32_t size) { m_insn_log_size = size; }
+
+private:
+  uint32_t m_insn_log_size;
+};
+
+//----------------------------------------------------------------------
+/// @class Decoder
+/// @brief This class makes use of Intel(R) Processor Trace hardware feature
+///     (implememted inside LLDB) to gather trace data for an inferior (being
+///     debugged with LLDB) to provide meaningful information out of it.
+///
+///     Currently the meaningful information comprises of the execution flow
+///     of the inferior (in terms of assembly instructions executed). The class
+///     enables user to:
+///     - start the trace with configuration options for a thread/process,
+///     - stop the trace for a thread/process,
+///     - get the execution flow (assembly instructions) for a thread and
+///     - get trace specific information for a thread
+//----------------------------------------------------------------------
+class Decoder {
+public:
+  typedef std::vector<Instruction> Instructions;
+
+  Decoder(lldb::SBDebugger &sbdebugger)
+      : m_mapProcessUID_mapThreadID_TraceInfo_mutex(),
+        m_mapProcessUID_mapThreadID_TraceInfo(),
+        m_debugger_user_id(sbdebugger.GetID()) {}
+
+  ~Decoder() {}
+
+  void StartProcessorTrace(lldb::SBProcess &sbprocess,
+                           lldb::SBTraceOptions &sbtraceoptions,
+                           lldb::SBError &sberror);
+
+  void StopProcessorTrace(lldb::SBProcess &sbprocess, lldb::SBError &sberror,
+                          lldb::tid_t tid = LLDB_INVALID_THREAD_ID);
+
+  void GetInstructionLogAtOffset(
+      lldb::SBProcess &sbprocess, lldb::tid_t tid, uint32_t offset,
+      uint32_t count, std::vector<std::shared_ptr<Instruction>> &result_list,
+      lldb::SBError &sberror);
+
+  void GetProcessorTraceInfo(lldb::SBProcess &sbprocess, lldb::tid_t tid,
+                             TraceOptions &traceinfo, lldb::SBError &sberror);
+
+private:
+  class ThreadTraceInfo;
+  typedef std::vector<uint8_t> Buffer;
+
+  // internal class to manage inferior's read-execute section information
+  class ReadExecuteSectionInfo {
+  public:
+    uint64_t load_address;
+    uint64_t file_offset;
+    uint64_t size;
+    std::string image_path;
+
+    ReadExecuteSectionInfo(const uint64_t addr, const uint64_t offset,
+                           const uint64_t sz, const std::string &path)
+        : load_address(addr), file_offset(offset), size(sz), image_path(path) {}
+
+    ReadExecuteSectionInfo(const ReadExecuteSectionInfo &rxsection) = default;
+  };
+
+  typedef struct pt_cpu CPUInfo;
+  typedef std::vector<ReadExecuteSectionInfo> ReadExecuteSectionInfos;
+
+  // Check whether the provided SBProcess belongs to the same SBDebugger with
+  // which Decoder class instance was constructed.
+  void CheckDebuggerID(lldb::SBProcess &sbprocess, lldb::SBError &sberror);
+
+  // Function to remove entries of finished processes/threads in the class
+  void RemoveDeadProcessesAndThreads(lldb::SBProcess &sbprocess);
+
+  // Parse cpu information from trace configuration received from LLDB
+  void ParseCPUInfo(CPUInfo &pt_cpu, lldb::SBStructuredData &s,
+                    lldb::SBError &sberror);
+
+  ///------------------------------------------------------------------------
+  /// Function performs following tasks for a given process and thread:
+  ///  - Checks if the given thread is registered in the class or not. If not
+  ///  then tries to register it if trace was ever started on the entire
+  ///  process. Else returns error.
+  ///  - fetches trace and other necessary information from LLDB (using
+  ///  ReadTraceDataAndImageInfo()) and decodes the trace (using
+  ///  DecodeProcessorTrace())
+  ///------------------------------------------------------------------------
+  void FetchAndDecode(lldb::SBProcess &sbprocess, lldb::tid_t tid,
+                      lldb::SBError &sberror,
+                      ThreadTraceInfo **threadTraceInfo);
+
+  // Helper function of FetchAndDecode() to get raw trace data and memory image
+  // info of inferior from LLDB
+  void ReadTraceDataAndImageInfo(lldb::SBProcess &sbprocess, lldb::tid_t tid,
+                                 lldb::SBError &sberror,
+                                 ThreadTraceInfo &threadTraceInfo);
+
+  // Helper function of FetchAndDecode() to initialize raw trace decoder and
+  // start trace decoding
+  void DecodeProcessorTrace(lldb::SBProcess &sbprocess, lldb::tid_t tid,
+                            lldb::SBError &sberror,
+                            ThreadTraceInfo &threadTraceInfo);
+
+  // Helper function of ReadTraceDataAndImageInfo() function for gathering
+  // inferior's memory image info along with all dynamic libraries linked with
+  // it
+  void GetTargetModulesInfo(lldb::SBTarget &sbtarget,
+                            ReadExecuteSectionInfos &readExecuteSectionInfos,
+                            lldb::SBError &sberror);
+
+  ///------------------------------------------------------------------------
+  /// Helper functions of DecodeProcessorTrace() function for:
+  ///  - initializing raw trace decoder (provided by Intel(R) Processor Trace
+  ///    Decoding library)
+  ///  - start trace decoding
+  ///------------------------------------------------------------------------
+  void InitializePTInstDecoder(
+      struct pt_insn_decoder **decoder, struct pt_config *config,
+      const CPUInfo &pt_cpu, Buffer &pt_buffer,
+      const ReadExecuteSectionInfos &readExecuteSectionInfos,
+      lldb::SBError &sberror) const;
+  void DecodeTrace(struct pt_insn_decoder *decoder,
+                   Instructions &instruction_list, lldb::SBError &sberror);
+
+  // Function to diagnose and indicate errors during raw trace decoding
+  void Diagnose(struct pt_insn_decoder *decoder, int errcode,
+                lldb::SBError &sberror, const struct pt_insn *insn = nullptr);
+
+  class ThreadTraceInfo {
+  public:
+    ThreadTraceInfo()
+        : m_pt_buffer(), m_readExecuteSectionInfos(), m_thread_stop_id(0),
+          m_trace(), m_pt_cpu(), m_instruction_log() {}
+
+    ThreadTraceInfo(const ThreadTraceInfo &trace_info) = default;
+
+    ~ThreadTraceInfo() {}
+
+    Buffer &GetPTBuffer() { return m_pt_buffer; }
+
+    void AllocatePTBuffer(uint64_t size) { m_pt_buffer.assign(size, 0); }
+
+    ReadExecuteSectionInfos &GetReadExecuteSectionInfos() {
+      return m_readExecuteSectionInfos;
+    }
+
+    CPUInfo &GetCPUInfo() { return m_pt_cpu; }
+
+    Instructions &GetInstructionLog() { return m_instruction_log; }
+
+    uint32_t GetStopID() const { return m_thread_stop_id; }
+
+    void SetStopID(uint32_t stop_id) { m_thread_stop_id = stop_id; }
+
+    lldb::SBTrace &GetUniqueTraceInstance() { return m_trace; }
+
+    void SetUniqueTraceInstance(lldb::SBTrace &trace) { m_trace = trace; }
+
+    friend class Decoder;
+
+  private:
+    Buffer m_pt_buffer; // raw trace buffer
+    ReadExecuteSectionInfos
+        m_readExecuteSectionInfos; // inferior's memory image info
+    uint32_t m_thread_stop_id;     // stop id for thread
+    lldb::SBTrace m_trace; // unique tracing instance of a thread/process
+    CPUInfo m_pt_cpu; // cpu info of the target on which inferior is running
+    Instructions m_instruction_log; // complete instruction log
+  };
+
+  typedef std::map<lldb::user_id_t, ThreadTraceInfo> MapThreadID_TraceInfo;
+  typedef std::map<uint32_t, MapThreadID_TraceInfo>
+      MapProcessUID_MapThreadID_TraceInfo;
+
+  std::mutex m_mapProcessUID_mapThreadID_TraceInfo_mutex;
+  MapProcessUID_MapThreadID_TraceInfo
+      m_mapProcessUID_mapThreadID_TraceInfo; // to store trace information for
+                                             // each process and its associated
+                                             // threads
+  lldb::user_id_t m_debugger_user_id; // SBDebugger instance which is associated
+                                      // to this Decoder instance
+};
+
+} // namespace ptdecoder_private
+#endif // Decoder_h_

tools/intel-features/intel-pt/Decoder.cpp

+//===-- Decoder.cpp ---------------------------------------------*- C++ -*-===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+
+// Project includes
+#include "Decoder.h"
+
+// C/C++ Includes
+#include <cinttypes>
+#include <cstring>
+#include <stdexcept>
+
+// Other libraries and framework includes
+#include "lldb/API/SBModule.h"
+#include "lldb/API/SBProcess.h"
+#include "lldb/API/SBThread.h"
+
+using namespace ptdecoder_private;
+
+// This function removes entries of all the processes/threads which were once
+// registered in the class but are not alive anymore because they died or
+// finished executing.
+void Decoder::RemoveDeadProcessesAndThreads(lldb::SBProcess &sbprocess) {
+  lldb::SBTarget sbtarget = sbprocess.GetTarget();
+  lldb::SBDebugger sbdebugger = sbtarget.GetDebugger();
+  uint32_t num_targets = sbdebugger.GetNumTargets();
+
+  auto itr_process = m_mapProcessUID_mapThreadID_TraceInfo.begin();
+  while (itr_process != m_mapProcessUID_mapThreadID_TraceInfo.end()) {
+    bool process_found = false;
+    lldb::SBTarget target;
+    lldb::SBProcess process;
+    for (uint32_t i = 0; i < num_targets; i++) {
+      target = sbdebugger.GetTargetAtIndex(i);
+      process = target.GetProcess();
+      if (process.GetUniqueID() == itr_process->first) {
+        process_found = true;
+        break;
+      }
+    }
+
+    // Remove the process's entry if it was not found in SBDebugger
+    if (!process_found) {
+      itr_process = m_mapProcessUID_mapThreadID_TraceInfo.erase(itr_process);
+      continue;
+    }
+
+    // If the state of the process is exited or detached then remove process's
+    // entry. If not then remove entry for all those registered threads of this
+    // process that are not alive anymore.
+    lldb::StateType state = process.GetState();
+    if ((state == lldb::StateType::eStateDetached) ||
+        (state == lldb::StateType::eStateExited))
+      itr_process = m_mapProcessUID_mapThreadID_TraceInfo.erase(itr_process);
+    else {
+      auto itr_thread = itr_process->second.begin();
+      while (itr_thread != itr_process->second.end()) {
+        if (itr_thread->first == LLDB_INVALID_THREAD_ID) {
+          ++itr_thread;
+          continue;
+        }
+
+        lldb::SBThread thread = process.GetThreadByID(itr_thread->first);
+        if (!thread.IsValid())
+          itr_thread = itr_process->second.erase(itr_thread);
+        else
+          ++itr_thread;
+      }
+      ++itr_process;
+    }
+  }
+}
+
+void Decoder::StartProcessorTrace(lldb::SBProcess &sbprocess,
+                                  lldb::SBTraceOptions &sbtraceoptions,
+                                  lldb::SBError &sberror) {
+  sberror.Clear();
+  CheckDebuggerID(sbprocess, sberror);
+  if (!sberror.Success())
+    return;
+
+  std::lock_guard<std::mutex> guard(
+      m_mapProcessUID_mapThreadID_TraceInfo_mutex);
+  RemoveDeadProcessesAndThreads(sbprocess);
+
+  if (sbtraceoptions.getType() != lldb::TraceType::eTraceTypeProcessorTrace) {
+    sberror.SetErrorStringWithFormat("SBTraceOptions::TraceType not set to "
+                                     "eTraceTypeProcessorTrace; ProcessID = "
+                                     "%" PRIu64,
+                                     sbprocess.GetProcessID());
+    return;
+  }
+  lldb::SBStructuredData sbstructdata = sbtraceoptions.getTraceParams(sberror);
+  if (!sberror.Success())
+    return;
+
+  const char *trace_tech_key = "trace-tech";
+  std::string trace_tech_value("intel-pt");
+  lldb::SBStructuredData value = sbstructdata.GetValueForKey(trace_tech_key);
+  if (!value.IsValid()) {
+    sberror.SetErrorStringWithFormat(
+        "key \"%s\" not set in custom trace parameters", trace_tech_key);
+    return;
+  }
+
+  char string_value[9];
+  size_t bytes_written = value.GetStringValue(
+      string_value, sizeof(string_value) / sizeof(*string_value));
+  if (!bytes_written ||
+      (bytes_written > (sizeof(string_value) / sizeof(*string_value)))) {
+    sberror.SetErrorStringWithFormat(
+        "key \"%s\" not set in custom trace parameters", trace_tech_key);
+    return;
+  }
+
+  std::size_t pos =
+      trace_tech_value.find((const char *)string_value, 0, bytes_written);
+  if ((pos == std::string::npos)) {
+    sberror.SetErrorStringWithFormat(
+        "key \"%s\" not set to \"%s\" in custom trace parameters",
+        trace_tech_key, trace_tech_value.c_str());
+    return;
+  }
+
+  // Start Tracing
+  lldb::SBError error;
+  uint32_t unique_id = sbprocess.GetUniqueID();
+  lldb::tid_t tid = sbtraceoptions.getThreadID();
+  lldb::SBTrace trace = sbprocess.StartTrace(sbtraceoptions, error);
+  if (!error.Success()) {
+    if (tid == LLDB_INVALID_THREAD_ID)
+      sberror.SetErrorStringWithFormat("%s; ProcessID = %" PRIu64,
+                                       error.GetCString(),
+                                       sbprocess.GetProcessID());
+    else
+      sberror.SetErrorStringWithFormat(
+          "%s; thread_id = %" PRIu64 ", ProcessID = %" PRIu64,
+          error.GetCString(), tid, sbprocess.GetProcessID());
+    return;
+  }
+
+  MapThreadID_TraceInfo &mapThreadID_TraceInfo =
+      m_mapProcessUID_mapThreadID_TraceInfo[unique_id];
+  ThreadTraceInfo &trace_info = mapThreadID_TraceInfo[tid];
+  trace_info.SetUniqueTraceInstance(trace);
+  trace_info.SetStopID(sbprocess.GetStopID());
+}
+
+void Decoder::StopProcessorTrace(lldb::SBProcess &sbprocess,
+                                 lldb::SBError &sberror, lldb::tid_t tid) {
+  sberror.Clear();
+  CheckDebuggerID(sbprocess, sberror);
+  if (!sberror.Success()) {
+    return;
+  }
+
+  std::lock_guard<std::mutex> guard(
+      m_mapProcessUID_mapThreadID_TraceInfo_mutex);
+  RemoveDeadProcessesAndThreads(sbprocess);
+
+  uint32_t unique_id = sbprocess.GetUniqueID();
+  auto itr_process = m_mapProcessUID_mapThreadID_TraceInfo.find(unique_id);
+  if (itr_process == m_mapProcessUID_mapThreadID_TraceInfo.end()) {
+    sberror.SetErrorStringWithFormat(
+        "tracing not active for this process; ProcessID = %" PRIu64,
+        sbprocess.GetProcessID());
+    return;
+  }
+
+  lldb::SBError error;
+  if (tid == LLDB_INVALID_THREAD_ID) {
+    // This implies to stop tracing on the whole process
+    lldb::user_id_t id_to_be_ignored = LLDB_INVALID_UID;
+    auto itr_thread = itr_process->second.begin();
+    while (itr_thread != itr_process->second.end()) {
+      // In the case when user started trace on the entire process and then
+      // registered newly spawned threads of this process in the class later,
+      // these newly spawned threads will have same trace id. If we stopped
+      // trace on the entire process then tracing stops automatically for these
+      // newly spawned registered threads. Stopping trace on them again will
+      // return error and therefore we need to skip stopping trace on them
+      // again.
+      lldb::SBTrace &trace = itr_thread->second.GetUniqueTraceInstance();
+      lldb::user_id_t lldb_pt_user_id = trace.GetTraceUID();
+      if (lldb_pt_user_id != id_to_be_ignored) {
+        trace.StopTrace(error, itr_thread->first);
+        if (!error.Success()) {
+          std::string error_string(error.GetCString());
+          if ((error_string.find("tracing not active for this process") ==
+               std::string::npos) &&
+              (error_string.find("tracing not active for this thread") ==
+               std::string::npos)) {
+            sberror.SetErrorStringWithFormat(
+                "%s; thread id=%" PRIu64 ", ProcessID = %" PRIu64,
+                error_string.c_str(), itr_thread->first,
+                sbprocess.GetProcessID());
+            return;
+          }
+        }
+
+        if (itr_thread->first == LLDB_INVALID_THREAD_ID)
+          id_to_be_ignored = lldb_pt_user_id;
+      }
+      itr_thread = itr_process->second.erase(itr_thread);
+    }
+    m_mapProcessUID_mapThreadID_TraceInfo.erase(itr_process);
+  } else {
+    // This implies to stop tracing on a single thread.
+    // if 'tid' is registered in the class then get the trace id and stop trace
+    // on it. If it is not then check if tracing was ever started on the entire
+    // process (because there is a possibility that trace is still running for
+    // 'tid' but it was not registered in the class because user had started
+    // trace on the whole process and 'tid' spawned later). In that case, get
+    // the trace id of the process trace instance and stop trace on this thread.
+    // If tracing was never started on the entire process then return error
+    // because there is no way tracing is active on 'tid'.
+    MapThreadID_TraceInfo &mapThreadID_TraceInfo = itr_process->second;
+    lldb::SBTrace trace;
+    auto itr = mapThreadID_TraceInfo.find(tid);
+    if (itr != mapThreadID_TraceInfo.end()) {
+      trace = itr->second.GetUniqueTraceInstance();
+    } else {
+      auto itr = mapThreadID_TraceInfo.find(LLDB_INVALID_THREAD_ID);
+      if (itr != mapThreadID_TraceInfo.end()) {
+        trace = itr->second.GetUniqueTraceInstance();
+      } else {
+        sberror.SetErrorStringWithFormat(
+            "tracing not active for this thread; thread id=%" PRIu64
+            ", ProcessID = %" PRIu64,
+            tid, sbprocess.GetProcessID());
+        return;
+      }
+    }
+
+    // Stop Tracing
+    trace.StopTrace(error, tid);
+    if (!error.Success()) {
+      std::string error_string(error.GetCString());
+      sberror.SetErrorStringWithFormat(
+          "%s; thread id=%" PRIu64 ", ProcessID = %" PRIu64,
+          error_string.c_str(), tid, sbprocess.GetProcessID());
+      if (error_string.find("tracing not active") == std::string::npos)
+        return;
+    }
+    // Delete the entry of 'tid' from this class (if any)
+    mapThreadID_TraceInfo.erase(tid);
+  }
+}
+
+void Decoder::ReadTraceDataAndImageInfo(lldb::SBProcess &sbprocess,
+                                        lldb::tid_t tid, lldb::SBError &sberror,
+                                        ThreadTraceInfo &threadTraceInfo) {
+  // Allocate trace data buffer and parse cpu info for 'tid' if it is registered
+  // for the first time in class
+  lldb::SBTrace &trace = threadTraceInfo.GetUniqueTraceInstance();
+  Buffer &pt_buffer = threadTraceInfo.GetPTBuffer();
+  lldb::SBError error;
+  if (pt_buffer.size() == 0) {
+    lldb::SBTraceOptions traceoptions;
+    traceoptions.setThreadID(tid);
+    trace.GetTraceConfig(traceoptions, error);
+    if (!error.Success()) {
+      sberror.SetErrorStringWithFormat("%s; ProcessID = %" PRIu64,
+                                       error.GetCString(),
+                                       sbprocess.GetProcessID());
+      return;
+    }
+    if (traceoptions.getType() != lldb::TraceType::eTraceTypeProcessorTrace) {
+      sberror.SetErrorStringWithFormat("invalid TraceType received from LLDB "
+                                       "for this thread; thread id=%" PRIu64
+                                       ", ProcessID = %" PRIu64,
+                                       tid, sbprocess.GetProcessID());
+      return;
+    }
+
+    threadTraceInfo.AllocatePTBuffer(traceoptions.getTraceBufferSize());
+    lldb::SBStructuredData sbstructdata = traceoptions.getTraceParams(sberror);
+    if (!sberror.Success())
+      return;
+    CPUInfo &pt_cpu = threadTraceInfo.GetCPUInfo();
+    ParseCPUInfo(pt_cpu, sbstructdata, sberror);
+    if (!sberror.Success())
+      return;
+  }
+
+  // Call LLDB API to get raw trace data for this thread
+  size_t bytes_written = trace.GetTraceData(error, (void *)pt_buffer.data(),
+                                            pt_buffer.size(), 0, tid);
+  if (!error.Success()) {
+    sberror.SetErrorStringWithFormat(
+        "%s; thread_id = %" PRIu64 ",  ProcessID = %" PRIu64,
+        error.GetCString(), tid, sbprocess.GetProcessID());
+    return;
+  }
+  std::fill(pt_buffer.begin() + bytes_written, pt_buffer.end(), 0);
+
+  // Get information of all the modules of the inferior
+  lldb::SBTarget sbtarget = sbprocess.GetTarget();
+  ReadExecuteSectionInfos &readExecuteSectionInfos =
+      threadTraceInfo.GetReadExecuteSectionInfos();
+  GetTargetModulesInfo(sbtarget, readExecuteSectionInfos, sberror);
+  if (!sberror.Success())
+    return;
+}
+
+void Decoder::DecodeProcessorTrace(lldb::SBProcess &sbprocess, lldb::tid_t tid,
+                                   lldb::SBError &sberror,
+                                   ThreadTraceInfo &threadTraceInfo) {
+  // Initialize instruction decoder
+  struct pt_insn_decoder *decoder = nullptr;
+  struct pt_config config;
+  Buffer &pt_buffer = threadTraceInfo.GetPTBuffer();
+  CPUInfo &pt_cpu = threadTraceInfo.GetCPUInfo();
+  ReadExecuteSectionInfos &readExecuteSectionInfos =
+      threadTraceInfo.GetReadExecuteSectionInfos();
+
+  InitializePTInstDecoder(&decoder, &config, pt_cpu, pt_buffer,
+                          readExecuteSectionInfos, sberror);
+  if (!sberror.Success())
+    return;
+
+  // Start raw trace decoding
+  Instructions &instruction_list = threadTraceInfo.GetInstructionLog();
+  instruction_list.clear();
+  DecodeTrace(decoder, instruction_list, sberror);
+}
+
+// Raw trace decoding requires information of Read & Execute sections of each
+// module of the inferior. This function updates internal state of the class to
+// store this information.
+void Decoder::GetTargetModulesInfo(
+    lldb::SBTarget &sbtarget, ReadExecuteSectionInfos &readExecuteSectionInfos,
+    lldb::SBError &sberror) {
+  if (!sbtarget.IsValid()) {
+    sberror.SetErrorStringWithFormat("Can't get target's modules info from "
+                                     "LLDB; process has an invalid target");
+    return;
+  }
+
+  lldb::SBFileSpec target_file_spec = sbtarget.GetExecutable();
+  if (!target_file_spec.IsValid()) {
+    sberror.SetErrorStringWithFormat("Target has an invalid file spec");
+    return;
+  }
+
+  uint32_t num_modules = sbtarget.GetNumModules();
+  readExecuteSectionInfos.clear();
+
+  // Store information of all RX sections of each module of inferior
+  for (uint32_t i = 0; i < num_modules; i++) {
+    lldb::SBModule module = sbtarget.GetModuleAtIndex(i);
+    if (!module.IsValid()) {
+      sberror.SetErrorStringWithFormat(
+          "Can't get module info [ %" PRIu32
+          " ] of target \"%s\" from LLDB, invalid module",
+          i, target_file_spec.GetFilename());
+      return;
+    }
+
+    lldb::SBFileSpec module_file_spec = module.GetPlatformFileSpec();
+    if (!module_file_spec.IsValid()) {
+      sberror.SetErrorStringWithFormat(
+          "Can't get module info [ %" PRIu32
+          " ] of target \"%s\" from LLDB, invalid file spec",
+          i, target_file_spec.GetFilename());
+      return;
+    }
+
+    const char *image(module_file_spec.GetFilename());
+    lldb::SBError error;
+    char image_complete_path[1024];
+    uint32_t path_length = module_file_spec.GetPath(
+        image_complete_path, sizeof(image_complete_path));
+    size_t num_sections = module.GetNumSections();
+
+    // Store information of only RX sections
+    for (size_t idx = 0; idx < num_sections; idx++) {
+      lldb::SBSection section = module.GetSectionAtIndex(idx);
+      uint32_t section_permission = section.GetPermissions();
+      if ((section_permission & lldb::Permissions::ePermissionsReadable) &&
+          (section_permission & lldb::Permissions::ePermissionsExecutable)) {
+        lldb::SBData section_data = section.GetSectionData();
+        if (!section_data.IsValid()) {
+          sberror.SetErrorStringWithFormat(
+              "Can't get module info [ %" PRIu32 " ]   \"%s\"  of target "
+              "\"%s\" from LLDB, invalid "
+              "data in \"%s\" section",
+              i, image, target_file_spec.GetFilename(), section.GetName());
+          return;
+        }
+
+        // In case section has no data, skip it.
+        if (section_data.GetByteSize() == 0)
+          continue;
+
+        if (!path_length) {
+          sberror.SetErrorStringWithFormat(
+              "Can't get module info [ %" PRIu32 " ]   \"%s\"  of target "
+              "\"%s\" from LLDB, module "
+              "has an invalid path length",
+              i, image, target_file_spec.GetFilename());
+          return;
+        }
+
+        std::string image_path(image_complete_path, path_length);
+        try {
+          readExecuteSectionInfos.emplace_back(
+              section.GetLoadAddress(sbtarget), section.GetFileOffset(),
+              section_data.GetByteSize(), image_path);
+        } catch (const std::bad_alloc &le) {
+          continue;
+        }
+      }
+    }
+  }
+}
+
+// Raw trace decoding requires information of the target cpu on which inferior
+// is running. This function gets the Trace Configuration from LLDB, parses it
+// for cpu model, family, stepping and vendor id info and updates the internal
+// state of the class to store this information.
+void Decoder::ParseCPUInfo(CPUInfo &pt_cpu, lldb::SBStructuredData &s,
+                           lldb::SBError &sberror) {
+  lldb::SBStructuredData custom_trace_params = s.GetValueForKey("intel-pt");
+  if (!custom_trace_params.IsValid()) {
+    sberror.SetErrorStringWithFormat("lldb couldn't provide cpuinfo");
+    return;
+  }
+
+  uint64_t family = 0, model = 0, stepping = 0;
+  char vendor[32];
+  const char *key_family = "cpu_family";
+  const char *key_model = "cpu_model";
+  const char *key_stepping = "cpu_stepping";
+  const char *key_vendor = "cpu_vendor";
+
+  // parse family
+  lldb::SBStructuredData struct_family =
+      custom_trace_params.GetValueForKey(key_family);
+  if (!struct_family.IsValid()) {
+    sberror.SetErrorStringWithFormat(
+        "%s info missing in custom trace parameters", key_family);
+    return;
+  }
+  family = struct_family.GetIntegerValue(0x10000);
+  if (family > UINT16_MAX) {
+    sberror.SetErrorStringWithFormat(
+        "invalid CPU family value extracted from custom trace parameters");
+    return;
+  }
+  pt_cpu.family = (uint16_t)family;
+
+  // parse model
+  lldb::SBStructuredData struct_model =
+      custom_trace_params.GetValueForKey(key_model);
+  if (!struct_model.IsValid()) {
+    sberror.SetErrorStringWithFormat(
+        "%s info missing in custom trace parameters; family=%" PRIu16,
+        key_model, pt_cpu.family);
+    return;
+  }
+  model = struct_model.GetIntegerValue(0x100);
+  if (model > UINT8_MAX) {
+    sberror.SetErrorStringWithFormat("invalid CPU model value extracted from "
+                                     "custom trace parameters; family=%" PRIu16,
+                                     pt_cpu.family);
+    return;
+  }
+  pt_cpu.model = (uint8_t)model;
+
+  // parse stepping
+  lldb::SBStructuredData struct_stepping =
+      custom_trace_params.GetValueForKey(key_stepping);
+  if (!struct_stepping.IsValid()) {
+    sberror.SetErrorStringWithFormat(
+        "%s info missing in custom trace parameters; family=%" PRIu16
+        ", model=%" PRIu8,
+        key_stepping, pt_cpu.family, pt_cpu.model);
+    return;
+  }
+  stepping = struct_stepping.GetIntegerValue(0x100);
+  if (stepping > UINT8_MAX) {
+    sberror.SetErrorStringWithFormat("invalid CPU stepping value extracted "
+                                     "from custom trace parameters; "
+                                     "family=%" PRIu16 ", model=%" PRIu8,
+                                     pt_cpu.family, pt_cpu.model);
+    return;
+  }
+  pt_cpu.stepping = (uint8_t)stepping;
+
+  // parse vendor info
+  pt_cpu.vendor = pcv_unknown;
+  lldb::SBStructuredData struct_vendor =
+      custom_trace_params.GetValueForKey(key_vendor);
+  if (!struct_vendor.IsValid()) {
+    sberror.SetErrorStringWithFormat(
+        "%s info missing in custom trace parameters; family=%" PRIu16
+        ", model=%" PRIu8 ", stepping=%" PRIu8,
+        key_vendor, pt_cpu.family, pt_cpu.model, pt_cpu.stepping);
+    return;
+  }
+  auto length = struct_vendor.GetStringValue(vendor, sizeof(vendor));
+  if (length && strstr(vendor, "GenuineIntel"))
+    pt_cpu.vendor = pcv_intel;
+}
+
+// Initialize trace decoder with pt_config structure and populate its image
+// structure with inferior's memory image information. pt_config structure is
+// initialized with trace buffer and cpu info of the inferior before storing it
+// in trace decoder.
+void Decoder::InitializePTInstDecoder(
+    struct pt_insn_decoder **decoder, struct pt_config *config,
+    const CPUInfo &pt_cpu, Buffer &pt_buffer,
+    const ReadExecuteSectionInfos &readExecuteSectionInfos,
+    lldb::SBError &sberror) const {
+  if (!decoder || !config) {
+    sberror.SetErrorStringWithFormat("internal error");
+    return;
+  }
+
+  // Load cpu info of inferior's target in pt_config struct
+  pt_config_init(config);
+  config->cpu = pt_cpu;
+  int errcode = pt_cpu_errata(&(config->errata), &(config->cpu));
+  if (errcode < 0) {
+    sberror.SetErrorStringWithFormat("processor trace decoding library: "
+                                     "pt_cpu_errata() failed with error: "
+                                     "\"%s\"",
+                                     pt_errstr(pt_errcode(errcode)));
+    return;
+  }
+
+  // Load trace buffer's starting and end address in pt_config struct
+  config->begin = pt_buffer.data();
+  config->end = pt_buffer.data() + pt_buffer.size();
+
+  // Fill trace decoder with pt_config struct
+  *decoder = pt_insn_alloc_decoder(config);
+  if (*decoder == nullptr) {
+    sberror.SetErrorStringWithFormat("processor trace decoding library:  "
+                                     "pt_insn_alloc_decoder() returned null "
+                                     "pointer");
+    return;
+  }
+
+  // Fill trace decoder's image with inferior's memory image information
+  struct pt_image *image = pt_insn_get_image(*decoder);
+  if (!image) {
+    sberror.SetErrorStringWithFormat("processor trace decoding library:  "
+                                     "pt_insn_get_image() returned null "
+                                     "pointer");
+    pt_insn_free_decoder(*decoder);
+    return;
+  }
+
+  for (auto &itr : readExecuteSectionInfos) {
+    errcode = pt_image_add_file(image, itr.image_path.c_str(), itr.file_offset,
+                                itr.size, nullptr, itr.load_address);
+    if (errcode < 0) {
+      sberror.SetErrorStringWithFormat("processor trace decoding library:  "
+                                       "pt_image_add_file() failed with error: "
+                                       "\"%s\"",
+                                       pt_errstr(pt_errcode(errcode)));
+      pt_insn_free_decoder(*decoder);
+      return;
+    }
+  }
+}
+
+// Start actual decoding of raw trace
+void Decoder::DecodeTrace(struct pt_insn_decoder *decoder,
+                          Instructions &instruction_list,
+                          lldb::SBError &sberror) {
+  uint64_t decoder_offset = 0;
+
+  while (1) {
+    struct pt_insn insn;
+
+    // Try to sync the decoder. If it fails then get the decoder_offset and try
+    // to sync again. If the new_decoder_offset is same as decoder_offset then
+    // we will not succeed in syncing for any number of pt_insn_sync_forward()
+    // operations. Return in that case. Else keep resyncing until either end of
+    // trace stream is reached or pt_insn_sync_forward() passes.
+    int errcode = pt_insn_sync_forward(decoder);
+    if (errcode < 0) {
+      if (errcode == -pte_eos)
+        return;
+
+      int errcode_off = pt_insn_get_offset(decoder, &decoder_offset);
+      if (errcode_off < 0) {
+        sberror.SetErrorStringWithFormat(
+            "processor trace decoding library: \"%s\"",
+            pt_errstr(pt_errcode(errcode)));
+        try {
+          instruction_list.emplace_back(sberror.GetCString());
+        } catch (const std::bad_alloc &le) {
+        }
+        return;
+      }
+
+      sberror.SetErrorStringWithFormat(
+          "processor trace decoding library: \"%s\"  [decoder_offset] => "
+          "[0x%" PRIu64 "]",
+          pt_errstr(pt_errcode(errcode)), decoder_offset);
+      try {
+        instruction_list.emplace_back(sberror.GetCString());
+      } catch (const std::bad_alloc &le) {
+        return;
+      }
+      while (1) {
+        errcode = pt_insn_sync_forward(decoder);
+        if (errcode >= 0)
+          break;
+
+        if (errcode == -pte_eos)
+          return;
+
+        uint64_t new_decoder_offset = 0;
+        errcode_off = pt_insn_get_offset(decoder, &new_decoder_offset);
+        if (errcode_off < 0) {
+          sberror.SetErrorStringWithFormat(
+              "processor trace decoding library: \"%s\"",
+              pt_errstr(pt_errcode(errcode)));
+          try {
+            instruction_list.emplace_back(sberror.GetCString());
+          } catch (const std::bad_alloc &le) {
+          }
+          return;
+        } else if (new_decoder_offset <= decoder_offset) {
+          // We tried resyncing the decoder and decoder didn't make any
+          // progress because the offset didn't change. We will not make any
+          // progress further. Hence, returning in this situation.
+          return;
+        }
+        sberror.SetErrorStringWithFormat(
+            "processor trace decoding library: \"%s\"  [decoder_offset] => "
+            "[0x%" PRIu64 "]",
+            pt_errstr(pt_errcode(errcode)), new_decoder_offset);
+        try {
+          instruction_list.emplace_back(sberror.GetCString());
+        } catch (const std::bad_alloc &le) {
+        }
+        decoder_offset = new_decoder_offset;
+      }
+    }
+
+    while (1) {
+      errcode = pt_insn_next(decoder, &insn, sizeof(insn));
+      if (errcode < 0) {
+        if (insn.iclass == ptic_error)
+          break;
+
+        try {
+          instruction_list.emplace_back(insn);
+        } catch (const std::bad_alloc &le) {
+          return;
+        }
+
+        if (errcode == -pte_eos)
+          return;
+
+        Diagnose(decoder, errcode, sberror, &insn);
+        try {
+          instruction_list.emplace_back(sberror.GetCString());
+        } catch (const std::bad_alloc &le) {
+          return;
+        }
+        break;
+      }
+      try {
+        instruction_list.emplace_back(insn);
+      } catch (const std::bad_alloc &le) {
+        return;
+      }
+      if (errcode & pts_eos)
+        return;
+    }
+  }
+}
+
+// Function to diagnose and indicate errors during raw trace decoding
+void Decoder::Diagnose(struct pt_insn_decoder *decoder, int decode_error,
+                       lldb::SBError &sberror, const struct pt_insn *insn) {
+  int errcode;
+  uint64_t offset;
+
+  errcode = pt_insn_get_offset(decoder, &offset);
+  if (insn) {
+    if (errcode < 0)
+      sberror.SetErrorStringWithFormat(
+          "processor trace decoding library: \"%s\"  [decoder_offset, "
+          "last_successful_decoded_ip] => [?, 0x%" PRIu64 "]",
+          pt_errstr(pt_errcode(decode_error)), insn->ip);
+    else
+      sberror.SetErrorStringWithFormat(
+          "processor trace decoding library: \"%s\"  [decoder_offset, "
+          "last_successful_decoded_ip] => [0x%" PRIu64 ", 0x%" PRIu64 "]",
+          pt_errstr(pt_errcode(decode_error)), offset, insn->ip);
+  } else {
+    if (errcode < 0)
+      sberror.SetErrorStringWithFormat(
+          "processor trace decoding library: \"%s\"",
+          pt_errstr(pt_errcode(decode_error)));
+    else
+      sberror.SetErrorStringWithFormat(
+          "processor trace decoding library: \"%s\"  [decoder_offset] => "
+          "[0x%" PRIu64 "]",
+          pt_errstr(pt_errcode(decode_error)), offset);
+  }
+}
+
+void Decoder::GetInstructionLogAtOffset(
+    lldb::SBProcess &sbprocess, lldb::tid_t tid, uint32_t offset,
+    uint32_t count, std::vector<std::shared_ptr<Instruction>> &result_list,
+    lldb::SBError &sberror) {
+  sberror.Clear();
+  CheckDebuggerID(sbprocess, sberror);
+  if (!sberror.Success()) {
+    return;
+  }
+
+  std::lock_guard<std::mutex> guard(
+      m_mapProcessUID_mapThreadID_TraceInfo_mutex);
+  RemoveDeadProcessesAndThreads(sbprocess);
+
+  ThreadTraceInfo *threadTraceInfo = nullptr;
+  FetchAndDecode(sbprocess, tid, sberror, &threadTraceInfo);
+  if (!sberror.Success()) {
+    return;
+  }
+  if (threadTraceInfo == nullptr) {
+    sberror.SetErrorStringWithFormat("internal error");
+    return;
+  }
+
+  // Return instruction log by populating 'result_list'
+  Instructions &insn_list = threadTraceInfo->GetInstructionLog();
+  uint64_t sum = (uint64_t)offset + 1;
+  if (((insn_list.size() <= offset) && (count <= sum) &&
+       ((sum - count) >= insn_list.size())) ||
+      (count < 1)) {
+    sberror.SetErrorStringWithFormat(
+        "Instruction Log not available for offset=%" PRIu32
+        " and count=%" PRIu32 ", ProcessID = %" PRIu64,
+        offset, count, sbprocess.GetProcessID());
+    return;
+  }
+
+  Instructions::iterator itr_first =
+      (insn_list.size() <= offset) ? insn_list.begin()
+                                   : insn_list.begin() + insn_list.size() - sum;
+  Instructions::iterator itr_last =
+      (count <= sum) ? insn_list.begin() + insn_list.size() - (sum - count)
+                     : insn_list.end();
+  Instructions::iterator itr = itr_first;
+  while (itr != itr_last) {
+    try {
+      result_list.push_back(
+          std::shared_ptr<Instruction>(new Instruction(*itr)));
+    } catch (const std::bad_alloc &le) {
+      return;
+    }
+    ++itr;
+  }
+}
+
+void Decoder::GetProcessorTraceInfo(lldb::SBProcess &sbprocess, lldb::tid_t tid,
+                                    TraceOptions &options,
+                                    lldb::SBError &sberror) {
+  sberror.Clear();
+  CheckDebuggerID(sbprocess, sberror);
+  if (!sberror.Success()) {
+    return;
+  }
+
+  std::lock_guard<std::mutex> guard(
+      m_mapProcessUID_mapThreadID_TraceInfo_mutex);
+  RemoveDeadProcessesAndThreads(sbprocess);
+
+  ThreadTraceInfo *threadTraceInfo = nullptr;
+  FetchAndDecode(sbprocess, tid, sberror, &threadTraceInfo);
+  if (!sberror.Success()) {
+    return;
+  }
+  if (threadTraceInfo == nullptr) {
+    sberror.SetErrorStringWithFormat("internal error");
+    return;
+  }
+
+  // Get SBTraceOptions from LLDB for 'tid', populate 'traceoptions' with it
+  lldb::SBTrace &trace = threadTraceInfo->GetUniqueTraceInstance();
+  lldb::SBTraceOptions traceoptions;
+  lldb::SBError error;
+  traceoptions.setThreadID(tid);
+  trace.GetTraceConfig(traceoptions, error);
+  if (!error.Success()) {
+    std::string error_string(error.GetCString());
+    if (error_string.find("tracing not active") != std::string::npos) {
+      uint32_t unique_id = sbprocess.GetUniqueID();
+      auto itr_process = m_mapProcessUID_mapThreadID_TraceInfo.find(unique_id);
+      if (itr_process == m_mapProcessUID_mapThreadID_TraceInfo.end())
+        return;
+      itr_process->second.erase(tid);
+    }
+    sberror.SetErrorStringWithFormat("%s; ProcessID = %" PRIu64,
+                                     error_string.c_str(),
+                                     sbprocess.GetProcessID());
+    return;
+  }
+  if (traceoptions.getType() != lldb::TraceType::eTraceTypeProcessorTrace) {
+    sberror.SetErrorStringWithFormat("invalid TraceType received from LLDB "
+                                     "for this thread; thread id=%" PRIu64
+                                     ", ProcessID = %" PRIu64,
+                                     tid, sbprocess.GetProcessID());
+    return;
+  }
+  options.setType(traceoptions.getType());
+  options.setTraceBufferSize(traceoptions.getTraceBufferSize());
+  options.setMetaDataBufferSize(traceoptions.getMetaDataBufferSize());
+  lldb::SBStructuredData sbstructdata = traceoptions.getTraceParams(sberror);
+  if (!sberror.Success())
+    return;
+  options.setTraceParams(sbstructdata);
+  options.setInstructionLogSize(threadTraceInfo->GetInstructionLog().size());
+}
+
+void Decoder::FetchAndDecode(lldb::SBProcess &sbprocess, lldb::tid_t tid,
+                             lldb::SBError &sberror,
+                             ThreadTraceInfo **threadTraceInfo) {
+  // Return with error if 'sbprocess' is not registered in the class
+  uint32_t unique_id = sbprocess.GetUniqueID();
+  auto itr_process = m_mapProcessUID_mapThreadID_TraceInfo.find(unique_id);
+  if (itr_process == m_mapProcessUID_mapThreadID_TraceInfo.end()) {
+    sberror.SetErrorStringWithFormat(
+        "tracing not active for this process; ProcessID = %" PRIu64,
+        sbprocess.GetProcessID());
+    return;
+  }
+
+  if (tid == LLDB_INVALID_THREAD_ID) {
+    sberror.SetErrorStringWithFormat(
+        "invalid thread id provided; thread_id = %" PRIu64
+        ", ProcessID = %" PRIu64,
+        tid, sbprocess.GetProcessID());
+    return;
+  }
+
+  // Check whether 'tid' thread is registered in the class. If it is then in
+  // case StopID didn't change then return without doing anything (no need to
+  // read and decode trace data then). Otherwise, save new StopID and proceed
+  // with reading and decoding trace.
+  if (threadTraceInfo == nullptr) {
+    sberror.SetErrorStringWithFormat("internal error");
+    return;
+  }
+
+  MapThreadID_TraceInfo &mapThreadID_TraceInfo = itr_process->second;
+  auto itr_thread = mapThreadID_TraceInfo.find(tid);
+  if (itr_thread != mapThreadID_TraceInfo.end()) {
+    if (itr_thread->second.GetStopID() == sbprocess.GetStopID()) {
+      *threadTraceInfo = &(itr_thread->second);
+      return;
+    }
+    itr_thread->second.SetStopID(sbprocess.GetStopID());
+  } else {
+    // Implies 'tid' is not registered in the class. If tracing was never
+    // started on the entire process then return an error. Else try to register
+    // this thread and proceed with reading and decoding trace.
+    lldb::SBError error;
+    itr_thread = mapThreadID_TraceInfo.find(LLDB_INVALID_THREAD_ID);
+    if (itr_thread == mapThreadID_TraceInfo.end()) {
+      sberror.SetErrorStringWithFormat(
+          "tracing not active for this thread; ProcessID = %" PRIu64,
+          sbprocess.GetProcessID());
+      return;
+    }
+
+    lldb::SBTrace &trace = itr_thread->second.GetUniqueTraceInstance();
+    ThreadTraceInfo &trace_info = mapThreadID_TraceInfo[tid];
+    trace_info.SetUniqueTraceInstance(trace);
+    trace_info.SetStopID(sbprocess.GetStopID());
+    itr_thread = mapThreadID_TraceInfo.find(tid);
+  }
+
+  // Get raw trace data and inferior image from LLDB for the registered thread
+  ReadTraceDataAndImageInfo(sbprocess, tid, sberror, itr_thread->second);
+  if (!sberror.Success()) {
+    std::string error_string(sberror.GetCString());
+    if (error_string.find("tracing not active") != std::string::npos)
+      mapThreadID_TraceInfo.erase(itr_thread);
+    return;
+  }
+  // Decode raw trace data
+  DecodeProcessorTrace(sbprocess, tid, sberror, itr_thread->second);
+  if (!sberror.Success()) {
+    return;
+  }
+  *threadTraceInfo = &(itr_thread->second);
+}
+
+// This function checks whether the provided SBProcess instance belongs to same
+// SBDebugger with which this tool instance is associated.
+void Decoder::CheckDebuggerID(lldb::SBProcess &sbprocess,
+                              lldb::SBError &sberror) {
+  if (!sbprocess.IsValid()) {
+    sberror.SetErrorStringWithFormat("invalid process instance");
+    return;
+  }
+
+  lldb::SBTarget sbtarget = sbprocess.GetTarget();
+  if (!sbtarget.IsValid()) {
+    sberror.SetErrorStringWithFormat(
+        "process contains an invalid target; ProcessID = %" PRIu64,
+        sbprocess.GetProcessID());
+    return;
+  }
+
+  lldb::SBDebugger sbdebugger = sbtarget.GetDebugger();
+  if (!sbdebugger.IsValid()) {
+    sberror.SetErrorStringWithFormat("process's target contains an invalid "
+                                     "debugger instance; ProcessID = %" PRIu64,
+                                     sbprocess.GetProcessID());
+    return;
+  }
+
+  if (sbdebugger.GetID() != m_debugger_user_id) {
+    sberror.SetErrorStringWithFormat(
+        "process belongs to a different SBDebugger instance than the one for "
+        "which the tool is instantiated; ProcessID = %" PRIu64,
+        sbprocess.GetProcessID());
+    return;
+  }
+}

tools/intel-features/intel-pt/PTDecoder.h

+//===-- PTDecoder.h ---------------------------------------------*- C++ -*-===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+
+#ifndef PTDecoder_h_
+#define PTDecoder_h_
+
+// C/C++ Includes
+#include <vector>
+
+// Project includes, Other libraries and framework includes
+#include "lldb/API/SBDebugger.h"
+#include "lldb/API/SBError.h"
+#include "lldb/API/SBProcess.h"
+#include "lldb/API/SBStructuredData.h"
+#include "lldb/API/SBTraceOptions.h"
+#include "lldb/lldb-enumerations.h"
+#include "lldb/lldb-types.h"
+
+namespace ptdecoder_private {
+class Instruction;
+class TraceOptions;
+class Decoder;
+} // namespace ptdecoder_private
+
+namespace ptdecoder {
+//----------------------------------------------------------------------
+/// @class PTInstruction
+/// @brief Represents an assembly instruction containing raw
+///     instruction bytes, instruction address along with information
+///     regarding execution flow context and Intel(R) Processor Trace
+///     context.
+//----------------------------------------------------------------------
+class PTInstruction {
+public:
+  PTInstruction();
+
+  PTInstruction(const PTInstruction &insn);
+
+  PTInstruction(const std::shared_ptr<ptdecoder_private::Instruction> &ptr);
+
+  ~PTInstruction();
+
+  // Get instruction address in inferior's memory image
+  uint64_t GetInsnAddress() const;
+
+  // Get raw bytes of the instruction
+  const uint8_t *GetRawBytes() const;
+
+  // Get size of raw bytes of the instruction
+  uint8_t GetRawBytesSize() const;
+
+  // Get error string if it represents an invalid instruction. For a valid
+  // instruction, an empty string is returned
+  std::string GetError() const;
+
+  // Instruction was executed speculatively or not
+  bool GetSpeculative() const;
+
+private:
+  std::shared_ptr<ptdecoder_private::Instruction> m_opaque_sp;
+};
+
+//----------------------------------------------------------------------
+/// @class PTTraceOptions
+/// @brief Provides configuration options like trace type, trace buffer size,
+///     meta data buffer size along with other Intel(R) Processor Trace
+///     specific options.
+//----------------------------------------------------------------------
+class PTTraceOptions {
+public:
+  PTTraceOptions();
+
+  PTTraceOptions(const PTTraceOptions &options);
+
+  ~PTTraceOptions();
+
+  lldb::TraceType GetType() const;
+
+  uint64_t GetTraceBufferSize() const;
+
+  uint64_t GetMetaDataBufferSize() const;
+
+  //------------------------------------------------------------------
+  /// Get Intel(R) Processor Trace specific configuration options (apart from
+  /// trace buffer size, meta data buffer size and TraceType) formatted as json
+  /// text i.e. {"Name":Value,"Name":Value} pairs, where "Value" is a 64-bit
+  /// unsigned integer in hex format. For "Name", please refer to
+  /// SBProcess::StartTrace API description for setting SBTraceOptions.
+  ///
+  /// @return
+  ///     A string formatted as json text {"Name":Value,"Name":Value}
+  //------------------------------------------------------------------
+  lldb::SBStructuredData GetTraceParams(lldb::SBError &error);
+
+private:
+  friend class PTDecoder;
+
+  void SetSP(const std::shared_ptr<ptdecoder_private::TraceOptions> &ptr);
+
+  std::shared_ptr<ptdecoder_private::TraceOptions> m_opaque_sp;
+};
+
+//----------------------------------------------------------------------
+/// @class PTDecoder
+/// @brief This class makes use of Intel(R) Processor Trace hardware feature
+///     (implememted inside LLDB) to gather trace data for an inferior (being
+///     debugged with LLDB) to provide meaningful information out of it.
+///
+///     Currently the meaningful information comprises of the execution flow
+///     of the inferior (in terms of assembly instructions executed). The class
+///     enables user to:
+///     - start the trace with configuration options for a thread/process,
+///     - stop the trace for a thread/process,
+///     - get the execution flow (assembly instructions) for a thread and
+///     - get trace specific information for a thread
+//----------------------------------------------------------------------
+class PTDecoder {
+public:
+  PTDecoder(lldb::SBDebugger &sbdebugger);
+
+  PTDecoder(const PTDecoder &ptdecoder);
+
+  ~PTDecoder();
+
+  //------------------------------------------------------------------
+  /// Start Intel(R) Processor Trace on a thread or complete process with
+  /// Intel(R) Processor Trace specific configuration options
+  ///
+  /// @param[in] sbprocess
+  ///     A valid process on which this operation will be performed. An error is
+  ///     returned in case of an invalid process.
+  ///
+  /// @param[in] sbtraceoptions
+  ///     Contains thread id information and configuration options:
+  ///
+  ///     For tracing a single thread, provide a valid thread id. If sbprocess
+  ///     doesn't contain this thread id, error will be returned. For tracing
+  ///     complete process, set it to lldb::LLDB_INVALID_THREAD_ID
+  ///     Configuration options comprises of:
+  ///     a) trace buffer size, meta data buffer size, TraceType and
+  ///     b) All other possible Intel(R) Processor Trace specific configuration
+  ///     options (hereafter collectively referred as CUSTOM_OPTIONS), formatted
+  ///     as json text i.e. {"Name":Value,"Name":Value,..} inside
+  ///     sbtraceoptions, where "Value" should be a 64-bit unsigned integer in
+  ///     hex format. For information regarding what all configuration options
+  ///     are currently supported by LLDB and detailed information about
+  ///     CUSTOM_OPTIONS usage, please refer to SBProcess::StartTrace() API
+  ///     description. To know about all possible configuration options of
+  ///     Intel(R) Processor Trace, please refer to Intel(R) 64 and IA-32
+  ///     Architectures Software Developer's Manual.
+  ///
+  ///     TraceType should be set to lldb::TraceType::eTraceTypeProcessorTrace,
+  ///     else error is returned. To find out any other requirement to start
+  ///     tracing successfully, please refer to SBProcess::StartTrace() API
+  ///     description. LLDB's current implementation of Intel(R) Processor Trace
+  ///     feature may round off invalid values for configuration options.
+  ///     Therefore, the configuration options with which the trace was actually
+  ///     started, might be different to the ones with which trace was asked to
+  ///     be started by user. The actual used configuration options can be
+  ///     obtained from GetProcessorTraceInfo() API.
+  ///
+  /// @param[out] sberror
+  ///     An error with the failure reason if API fails. Else success.
+  //------------------------------------------------------------------
+  void StartProcessorTrace(lldb::SBProcess &sbprocess,
+                           lldb::SBTraceOptions &sbtraceoptions,
+                           lldb::SBError &sberror);
+
+  //------------------------------------------------------------------
+  /// Stop Intel(R) Processor Trace on a thread or complete process.
+  ///
+  /// @param[in] sbprocess
+  ///     A valid process on which this operation will be performed. An error is
+  ///     returned in case of an invalid process.
+  ///
+  /// @param[in] tid
+  ///     Case 1: To stop tracing a single thread, provide a valid thread id. If
+  ///     sbprocess doesn't contain the thread tid, error will be returned.
+  ///     Case 2: To stop tracing complete process, use
+  ///     lldb::LLDB_INVALID_THREAD_ID.
+  ///
+  /// @param[out] sberror
+  ///     An error with the failure reason if API fails. Else success.
+  //------------------------------------------------------------------
+  void StopProcessorTrace(lldb::SBProcess &sbprocess, lldb::SBError &sberror,
+                          lldb::tid_t tid = LLDB_INVALID_THREAD_ID);
+
+  //------------------------------------------------------------------
+  /// Get instruction log containing the execution flow for a thread of a
+  /// process in terms of assembly instructions executed.
+  ///
+  /// @param[in] sbprocess
+  ///     A valid process on which this operation will be performed. An error is
+  ///     returned in case of an invalid process.
+  ///
+  /// @param[in] tid
+  ///     A valid thread id of the thread for which instruction log is desired.
+  ///     If sbprocess doesn't contain the thread tid, error will be returned.
+  ///
+  /// @param[in] count
+  ///     The number of instructions requested by the user to be returned from
+  ///     the complete instruction log. Complete instruction log refers to all
+  ///     the assembly instructions obtained after decoding the complete raw
+  ///     trace data obtained from LLDB. The length of the complete instruction
+  ///     log is dependent on the trace buffer size with which processor tracing
+  ///     was started for this thread.
+  ///     The number of instructions actually returned are dependent on 'count'
+  ///     and 'offset' parameters of this API.
+  ///
+  /// @param[in] offset
+  ///     The offset in the complete instruction log from where 'count' number
+  ///     of instructions are requested by the user. offset is counted from the
+  ///     end of of this complete instruction log (which means the last executed
+  ///     instruction is at offset 0 (zero)).
+  ///
+  /// @param[out] result_list
+  ///     Depending upon 'count' and 'offset' values, instructions will be
+  ///     appended to the end of this list.
+  ///
+  /// @param[out] sberror
+  ///     An error with the failure reason if API fails. Else success.
+  //------------------------------------------------------------------
+  void GetInstructionLogAtOffset(lldb::SBProcess &sbprocess, lldb::tid_t tid,
+                                 uint32_t offset, uint32_t count,
+                                 std::vector<PTInstruction> &result_list,
+                                 lldb::SBError &sberror);
+
+  //------------------------------------------------------------------
+  /// Get Intel(R) Processor Trace specific information for a thread of a
+  /// process. The information contains the actual configuration options with
+  /// which the trace was started for this thread.
+  ///
+  /// @param[in] sbprocess
+  ///     A valid process on which this operation will be performed. An error is
+  ///     returned in case of an invalid process.
+  ///
+  /// @param[in] tid
+  ///     A valid thread id of the thread for which the trace specific
+  ///     information is required. If sbprocess doesn't contain the thread tid,
+  ///     an error will be returned.
+  ///
+  /// @param[out] options
+  ///     Contains actual configuration options (they may be different to the
+  ///     ones with which tracing was asked to be started for this thread during
+  ///     StartProcessorTrace() API call).
+  ///
+  /// @param[out] sberror
+  ///     An error with the failure reason if API fails. Else success.
+  //------------------------------------------------------------------
+  void GetProcessorTraceInfo(lldb::SBProcess &sbprocess, lldb::tid_t tid,
+                             PTTraceOptions &options, lldb::SBError &sberror);
+
+private:
+  std::shared_ptr<ptdecoder_private::Decoder> m_opaque_sp;
+};
+
+} // namespace ptdecoder
+#endif // PTDecoder_h_

tools/intel-features/intel-pt/PTDecoder.cpp

+//===-- PTDecoder.cpp -------------------------------------------*- C++ -*-===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+
+#include <stdexcept>
+// Project includes
+#include "Decoder.h"
+#include "PTDecoder.h"
+
+using namespace ptdecoder;
+using namespace ptdecoder_private;
+
+// PTInstruction class member functions definitions
+PTInstruction::PTInstruction() : m_opaque_sp() {}
+
+PTInstruction::PTInstruction(const PTInstruction &insn)
+    : m_opaque_sp(insn.m_opaque_sp) {}
+
+PTInstruction::PTInstruction(
+    const std::shared_ptr<ptdecoder_private::Instruction> &ptr)
+    : m_opaque_sp(ptr) {}
+
+PTInstruction::~PTInstruction() {}
+
+uint64_t PTInstruction::GetInsnAddress() const {
+  return (m_opaque_sp ? m_opaque_sp->GetInsnAddress() : 0);
+}
+
+const uint8_t *PTInstruction::GetRawBytes() const {
+  return (m_opaque_sp ? m_opaque_sp->GetRawBytes() : nullptr);
+}
+
+uint8_t PTInstruction::GetRawBytesSize() const {
+  return (m_opaque_sp ? m_opaque_sp->GetRawBytesSize() : 0);
+}
+
+std::string PTInstruction::GetError() const {
+  return (m_opaque_sp ? m_opaque_sp->GetError() : "null pointer");
+}
+
+bool PTInstruction::GetSpeculative() const {
+  return (m_opaque_sp ? m_opaque_sp->GetSpeculative() : 0);
+}
+
+// PTTraceOptions class member functions definitions
+PTTraceOptions::PTTraceOptions() : m_opaque_sp() {}
+
+PTTraceOptions::PTTraceOptions(const PTTraceOptions &options)
+    : m_opaque_sp(options.m_opaque_sp) {}
+
+PTTraceOptions::~PTTraceOptions() {}
+
+lldb::TraceType PTTraceOptions::GetType() const {
+  return (m_opaque_sp ? m_opaque_sp->getType()
+                      : lldb::TraceType::eTraceTypeNone);
+}
+
+uint64_t PTTraceOptions::GetTraceBufferSize() const {
+  return (m_opaque_sp ? m_opaque_sp->getTraceBufferSize() : 0);
+}
+
+uint64_t PTTraceOptions::GetMetaDataBufferSize() const {
+  return (m_opaque_sp ? m_opaque_sp->getMetaDataBufferSize() : 0);
+}
+
+lldb::SBStructuredData PTTraceOptions::GetTraceParams(lldb::SBError &error) {
+  if (!m_opaque_sp)
+    error.SetErrorString("null pointer");
+  return (m_opaque_sp ? m_opaque_sp->getTraceParams(error)
+                      : lldb::SBStructuredData());
+}
+
+void PTTraceOptions::SetSP(
+    const std::shared_ptr<ptdecoder_private::TraceOptions> &ptr) {
+  m_opaque_sp = ptr;
+}
+
+// PTDecoder class member functions definitions
+PTDecoder::PTDecoder(lldb::SBDebugger &sbdebugger)
+    : m_opaque_sp(new ptdecoder_private::Decoder(sbdebugger)) {}
+
+PTDecoder::PTDecoder(const PTDecoder &ptdecoder)
+    : m_opaque_sp(ptdecoder.m_opaque_sp) {}
+
+PTDecoder::~PTDecoder() {}
+
+void PTDecoder::StartProcessorTrace(lldb::SBProcess &sbprocess,
+                                    lldb::SBTraceOptions &sbtraceoptions,
+                                    lldb::SBError &sberror) {
+  if (m_opaque_sp == nullptr) {
+    sberror.SetErrorStringWithFormat("invalid PTDecoder instance");
+    return;
+  }
+
+  m_opaque_sp->StartProcessorTrace(sbprocess, sbtraceoptions, sberror);
+}
+
+void PTDecoder::StopProcessorTrace(lldb::SBProcess &sbprocess,
+                                   lldb::SBError &sberror, lldb::tid_t tid) {
+  if (m_opaque_sp == nullptr) {
+    sberror.SetErrorStringWithFormat("invalid PTDecoder instance");
+    return;
+  }
+
+  m_opaque_sp->StopProcessorTrace(sbprocess, sberror, tid);
+}
+
+void PTDecoder::GetInstructionLogAtOffset(
+    lldb::SBProcess &sbprocess, lldb::tid_t tid, uint32_t offset,
+    uint32_t count, std::vector<PTInstruction> &result_list,
+    lldb::SBError &sberror) {
+  if (m_opaque_sp == nullptr) {
+    sberror.SetErrorStringWithFormat("invalid PTDecoder instance");
+    return;
+  }
+
+  std::vector<std::shared_ptr<ptdecoder_private::Instruction>> vec_insn_ptr;
+  m_opaque_sp->GetInstructionLogAtOffset(sbprocess, tid, offset, count,
+                                         vec_insn_ptr, sberror);
+  if (!sberror.Success())
+    return;
+
+  for (auto &i : vec_insn_ptr) {
+    try {
+      result_list.emplace_back(
+          std::shared_ptr<ptdecoder_private::Instruction>(i));
+    } catch (const std::bad_alloc &le) {
+      return;
+    }
+  }
+}
+
+void PTDecoder::GetProcessorTraceInfo(lldb::SBProcess &sbprocess,
+                                      lldb::tid_t tid, PTTraceOptions &options,
+                                      lldb::SBError &sberror) {
+  if (m_opaque_sp == nullptr) {
+    sberror.SetErrorStringWithFormat("invalid PTDecoder instance");
+    return;
+  }
+
+  std::shared_ptr<ptdecoder_private::TraceOptions> trace_options_ptr(
+      new TraceOptions());
+  m_opaque_sp->GetProcessorTraceInfo(sbprocess, tid, *trace_options_ptr,
+                                     sberror);
+  if (!sberror.Success())
+    return;
+
+  options.SetSP(trace_options_ptr);
+}

tools/intel-features/intel-pt/README_CLI.txt

+****************************************************************************
+*                            README                                        *
+*                                                                          *
+* This file provides all the information regarding 4 new CLI commands that *
+* enable using Intel(R) Processor Trace Tool from LLDB's CLI.              *
+****************************************************************************
+
+
+============
+Introduction
+============
+A C++ based cli wrapper has been developed to use Intel(R) Processor Trace Tool
+through LLDB's command line. This also provides an idea to all developers on how
+to integrate the Tool into various IDEs providing LLDB as a debugger.
+
+
+
+============
+How to Build
+============
+The wrapper cli-wrapper-pt.cpp needs to be compiled and linked with the shared
+library of the Intel(R) Processor Trace Tool in order to be used through LLDB's
+CLI. The procedure to build shared library of the Intel(R) Processor Trace Tool
+is given in README_TOOL.txt file.
+
+
+
+============
+How to Use
+============
+All these commands are available via shared library (lldbIntelFeatures)
+obtained after building intel-features folder from top. Please refer to
+cli-wrapper.cpp and README files of "intel-features" folder for this purpose.
+
+
+
+============
+Description
+============
+4 CLI commands have been designed keeping the LLDB's existing CLI command syntax
+in mind.
+
+   1) processor-trace start [-b <buffer-size>] [<thread-index>]
+
+      Start Intel(R) Processor Trace on a specific thread or on the whole process
+
+      Syntax: processor-trace start  <cmd-options>
+
+      cmd-options Usage:
+        processor-trace start [-b <buffer-size>] [<thread-index>]
+
+          -b <buffer-size>
+             size of the trace buffer to store the trace data. If not specified
+             then a default value (=4KB) will be taken
+
+          <thread-index>
+             thread index of the thread. If no threads are specified, currently
+             selected thread is taken. Use the thread-index 'all' to start
+             tracing the whole process
+
+
+
+   2) processor-trace stop [<thread-index>]
+
+      Stop Intel(R) Processor Trace on a specific thread or on the whole process
+
+      Syntax: processor-trace stop  <cmd-options>
+
+      cmd-options Usage:
+      processor-trace stop [<thread-index>]
+
+          <thread-index>
+             thread index of the thread. If no threads are specified, currently
+             selected thread is taken. Use the thread-index 'all' to stop
+             tracing the whole process
+
+
+
+   3) processor-trace show-trace-options [<thread-index>]
+
+      Display all the information regarding Intel(R) Processor Trace for a specific
+      thread or for the whole process. The information contains trace buffer
+      size and configuration options of Intel(R) Processor Trace.
+
+      Syntax: processor-trace show-trace-options <cmd-options>
+
+      cmd-options Usage:
+        processor-trace show-trace-options [<thread-index>]
+
+          <thread-index>
+             thread index of the thread. If no threads are specified, currently
+             selected thread is taken. Use the thread-index 'all' to display
+             information for all threads of the process
+
+
+
+   4) processor-trace show-instr-log [-o <offset>] [-c <count>] [<thread-index>]
+
+      Display a log of assembly instructions executed for a specific thread or
+      for the whole process. The length of the log to be displayed and the
+      offset in the whole instruction log from where the log needs to be
+      displayed can also be provided. The offset is counted from the end of this
+      whole instruction log which means the last executed instruction is at
+      offset 0 (zero).
+
+      Syntax: processor-trace show-instr-log  <cmd-options>
+
+      cmd-options Usage:
+        processor-trace show-instr-log [-o <offset>] [-c <count>] [<thread-index>]
+
+          -c <count>
+             number of instructions to be displayed. If not specified then a
+             default value (=10) will be taken
+
+          -o <offset>
+             offset in the whole instruction log from where the log will be
+             displayed. If not specified then default value is calculated as
+             offset = count -1
+
+          <thread-index>
+             thread index of the thread. If no threads are specified, currently
+             selected thread is taken. Use the thread-index 'all' to show
+             instruction log for all the threads of the process

tools/intel-features/intel-pt/README_TOOL.txt

+*******************************************************************************
+*                            README                                           *
+*                                                                             *
+* This file provides all the information regarding Intel(R) Processor Trace   *
+* Tool. It consists explanation about how Tool internally works, its hardware *
+* and software dependencies, build procedure and usage of the API.            *
+*******************************************************************************
+
+
+
+============
+Introduction
+============
+The Intel(R) Processor Trace Tool is developed on top of LLDB and provides its
+its users execution trace of the debugged applications. Tool makes use of
+Intel(R) Processor Trace hardware feature implementation inside LLDB for this
+purpose. This hardware feature generates a set of trace packets that
+encapsulates program flow information. These trace packets along with the binary
+of the application can be decoded with the help of a software decoder to
+construct the execution trace of the application.
+
+More information about Intel(R) Processor Trace feature can be obtained from
+website: https://software.intel.com/en-us/blogs/2013/09/18/processor-tracing
+
+
+
+
+=========
+Details
+=========
+The functionality of the Tool consists three parts:
+
+1. Raw Trace Collection from LLDB
+      With the help of API of this Tool (given below), Intel(R) Processor Trace
+      can be started on the application being debugged with LLDB. The generated
+      trace of the application is gathered inside LLDB and is collected by the
+      Tool from LLDB through LLDB's public API.
+
+2. Raw Trace Decoding
+      For decoding the raw trace data, the Tool makes use of "libipt", an
+      Intel(R) Processor Trace Decoder Library. The library needs binary of
+      the application and information about the cpu on which the application is
+      running in order to decode the raw trace. The Tool gathers this
+      information from LLDB public API and provide it to "libipt". More
+      information about "libipt" can be found at:
+      https://software.intel.com/en-us/blogs/2013/09/18/processor-tracing and
+      https://github.com/01org/processor-trace
+
+3. Decoded Trace Post-processing
+      The decoded trace is post-processed to reconstruct the execution flow of
+      the application. The execution flow contains the list of assembly
+      instructions (called instruction log hereafter).
+
+
+
+
+=============
+Dependencies
+=============
+The Tool has following hardware and software dependencies:
+
+  - Hardware dependency: The Tool makes use of this hardware feature to capture
+    raw trace of an application from LLDB. This hardware feature may not be
+    present in all processors. The hardware feature is supported on Broadwell
+    and other succeeding CPUs such as Skylake etc. In order for Tool to provide
+    something meaningful, the target machine on which the application is running
+    should have this feature.
+
+  - Software dependency: The Tool has an indirect dependency on the Operating
+    System level software support for Intel(R) Processor Trace on the target
+    machine where the application is running and being debugged by LLDB. This
+    support is required to enable raw trace generation on the target machine.
+    Currently, the Tool works for applications running on Linux OS as till now
+    the Operating System level support for the feature is present only in Linux
+    (more specifically starting from the 4.1 kernel). In Linux, this feature is
+    implemented in perf_events subsystem and is usable through perf_event_open
+    system call. In the User space level, the Tool has a direct dependency on
+    "libipt" to decode the captured raw trace. This library might be
+    pre-installed on host systems. If not then the library can be built from
+    its sources (available at): https://github.com/01org/processor-trace
+
+
+
+
+============
+How to Build
+============
+The Tool has a cmake based build and can be built by specifying some extra flags
+while building LLDB with cmake. The following cmake flags need to be provided to
+build the Tool:
+
+  - LIBIPT_INCLUDE_PATH - The flag specifies the directory where the header
+    file of "libipt" resides. If the library is not pre-installed on the host
+    system and is built directly from "libipt" project sources then this file
+    may either come as a part of the sources itself or will be generated in
+    build folder while building library.
+
+  - LIBIPT_LIBRARY_PATH - The flag points to the location of "libipt" shared
+    library.
+
+The Tool currently works successfully with following versions of this library:
+  - v1.4, v1.5, v1.6
+
+
+
+============
+How to Use
+============
+The Tool's API are exposed as a C++ object oriented interface (file PTDecoder.h)
+in a shared library. The main class that implements the whole functionality is
+PTDecoder. This class makes use of 2 other classes, one to return instruction
+log (class PTInstruction) and the other to return trace specific information
+(class PTTraceOptions). The users can use these API to develop their own products. All
+API are also available as python functions through a script bridging interface,
+allowing them to be used directly from python either interactively or to build
+python apps.
+
+Currently, cli wrapper has been developed on top of the Tool to use it through
+LLDB's command line. Please refer to README_CLI.txt file for command line usage.
+
+
+A brief introduction about the classes and their API are given below.
+
+  class PTDecoder
+  ===============
+    This class makes use of Intel(R) Processor Trace hardware feature
+    (implemented inside LLDB) to gather trace data for an inferior (being
+    debugged with LLDB) to provide meaningful information out of it. Currently
+    the meaningful information comprises of the execution flow of the inferior
+    (in terms of assembly instructions executed). The class enables user to:
+
+    - start the trace with configuration options for a thread/process,
+    - stop the trace for a thread/process,
+    - get the execution flow (assembly instructions) for a thread and
+    - get trace specific information for a thread
+
+    Corresponding API are explained below:
+    a) void StartProcessorTrace(lldb::SBProcess &sbprocess,
+                                lldb::SBTraceOptions &sbtraceoptions,
+                                lldb::SBError &sberror)
+       ------------------------------------------------------------------------
+           This API allows the user to start trace on a particular thread or on
+           the whole process with Intel(R) Processor Trace specific
+           configuration options.
+
+           @param[in] sbprocess      : A valid process on which this operation
+               will be performed. An error is returned in case of an invalid
+               process.
+
+           @param[out] sberror       : An error with the failure reason if API
+               fails. Else success.
+
+           @param[in] sbtraceoptions : Contains thread id information and
+               configuration options:
+               For tracing a single thread, provide a valid thread id. If
+               sbprocess doesn't contain this thread id, error will be returned.
+               For tracing complete process, set to lldb::LLDB_INVALID_THREAD_ID
+               Configuration options comprises of:
+                - trace buffer size, meta data buffer size, TraceType and
+                - All other possible Intel(R) Processor Trace specific
+                  configuration options (hereafter collectively referred as
+                  CUSTOM_OPTIONS)
+
+                Trace buffer, meant to store the trace data read from target
+                machine, inside LLDB is configured as a cyclic buffer. Hence,
+                depending upon the trace buffer size provided here, buffer
+                overwrites may happen while LLDB writes trace data into it.
+                CUSTOM_OPTIONS are formatted as json text i.e. {"Name":Value,
+                "Name":Value,...} inside sbtraceoptions, where "Value" should be
+                a 64-bit unsigned integer in hex format. For information
+                regarding what all configuration options are currently supported
+                by LLDB and detailed information about CUSTOM_OPTIONS usage,
+                please refer to SBProcess::StartTrace() API description. An
+                overview of some of the various CUSTOM_OPTIONS are briefly given
+                below. Please refer to "Intel(R) 64 and IA-32 Architectures
+                Software Developer's Manual" for more details about them.
+                  - CYCEn       Enable/Disable Cycle Count Packet (CYC) Packet
+                  - OS          Packet generation enabled/disabled if
+                                Current Privilege Level (CPL)=0
+                  - User        Packet generation enabled/disabled if CPL>0
+                  - CR3Filter   Enable/Disable CR3 Filtering
+                  - MTCEn       Enable/disable MTC packets
+                  - TSCEn       Enable/disable TSC packets
+                  - DisRETC     Enable/disable RET Compression
+                  - BranchEn    Enable/disable COFI-based packets
+                  - MTCFreq     Defines MTC Packet Frequency
+                  - CycThresh   CYC Packet threshold
+                  - PSBFreq     Frequency of PSB Packets
+
+                TraceType should be set to
+                lldb::TraceType::eTraceTypeProcessorTrace, else error is
+                returned. To find out any other requirement to start tracing
+                successfully, refer to SBProcess::StartTrace() API description.
+                LLDB's current implementation of Intel(R) Processor Trace
+                feature may round off invalid values for configuration options.
+                Therefore, the configuration options with which the trace was
+                actually started, might be different to the ones with which
+                trace was asked to be started by user. The actual used
+                configuration options can be obtained from
+                GetProcessorTraceInfo() API.
+
+
+
+    b) void StopProcessorTrace(lldb::SBProcess &sbprocess,
+                               lldb::SBError &sberror,
+                               lldb::tid_t tid = LLDB_INVALID_THREAD_ID)
+       ------------------------------------------------------------------------
+           This API allows the user to Stop trace on a particular thread or on
+           the whole process.
+
+           @param[in] sbprocess : A valid process on which this operation will
+               be performed. An error is returned in case of an invalid process.
+
+           @param[in] tid       : To stop tracing a single thread, provide a
+               valid thread id. If sbprocess doesn't contain the thread tid,
+               error will be returned. To stop tracing complete process, use
+               lldb::LLDB_INVALID_THREAD_ID
+
+           @param[out] sberror  : An error with the failure reason if API fails.
+               Else success
+
+
+
+    c) void GetInstructionLogAtOffset(lldb::SBProcess &sbprocess, lldb::tid_t tid,
+                                      uint32_t offset, uint32_t count,
+                                      std::vector<PTInstruction> &result_list,
+                                      lldb::SBError &sberror)
+       ------------------------------------------------------------------------
+           This API provides instruction log that contains the execution flow
+           for a thread of a process in terms of assembly instruction executed.
+           The API works on only 1 thread at a time. To gather this information
+           for whole process, this API needs to be called for each thread.
+
+           @param[in] sbprocess    : A valid process on which this operation
+               will be performed. An error is returned in case of an invalid
+               process.
+
+           @param[in] tid          : A valid thread id of the thread for which
+               instruction log is desired. If sbprocess doesn't contain the
+               thread tid, error will be returned.
+
+           @param[in] count        : Number of instructions requested by the
+               user to be returned from the complete instruction log. Complete
+               instruction log refers to all the assembly instructions obtained
+               after decoding the complete raw trace data obtained from LLDB.
+               The length of the complete instruction log is dependent on the
+               trace buffer size with which processor tracing was started for
+               this thread.
+               The number of instructions actually returned are dependent on
+               'count' and 'offset' parameters of this API.
+
+           @param[in] offset       : The offset in the complete instruction log
+               from where 'count' number of instructions are requested by the
+               user. offset is counted from the end of of this complete
+               instruction log (which means the last executed instruction
+               is at offset 0 (zero)).
+
+           @param[out] result_list : Depending upon 'count' and 'offset' values,
+               instructions will be appended to the end of this list.
+
+           @param[out] sberror     : An error with the failure reason if API
+               fails. Else success
+
+
+
+    d) void GetProcessorTraceInfo(lldb::SBProcess &sbprocess, lldb::tid_t tid,
+                                  PTTraceOptions &options, lldb::SBError &sberror)
+       ------------------------------------------------------------------------
+           This API provides Intel(R) Processor Trace specific information for
+           a thread of a process. The API works on only 1 thread at a time. To
+           gather this information for whole process, this API needs to be
+           called for each thread. The information contains the actual
+           configuration options with which the trace was started for this
+           thread.
+
+           @param[in] sbprocess  : The valid process on which this operation
+               will be performed. An error is returned in case of an invalid
+               process.
+
+           @param[in] tid        : A valid thread id of the thread for which the
+               trace specific information is required. If sbprocess doesn't
+               contain the thread tid, an error will be returned.
+
+           @param[out] options   : Contains actual configuration options (they
+               may be different to the ones with which tracing was asked to be
+               started for this thread during StartProcessorTrace() API call).
+
+           @param[out] sberror   : An error with the failure reason if API
+               fails. Else success
+
+
+  class PTInstruction
+  =================
+      This class represents an assembly instruction containing raw instruction
+      bytes, instruction address along with execution flow context and
+      Intel(R) Processor Trace context. For more details, please refer to
+      PTDecoder.h file.
+
+
+  class PTTraceOptions
+  ============
+      This class provides Intel(R) Processor Trace specific configuration
+      options like trace type, trace buffer size, meta data buffer size along
+      with other trace specific options. For more details, please refer to
+      PTDecoder.h file.

tools/intel-features/intel-pt/interface/PTDecoder.i

+%include "stdint.i"
+%include "std_string.i"
+%include "std_vector.i"
+
+%include "lldb/lldb-defines.h"
+%include "lldb/lldb-enumerations.h"
+%include "lldb/lldb-forward.h"
+%include "lldb/lldb-types.h"
+
+%include "lldb/API/SBDefines.h"
+
+%include "../PTDecoder.h"
+%template(PTInstructions) std::vector<ptdecoder::PTInstruction>;

tools/intel-features/scripts/CMakeLists.txt

+file(GLOB_RECURSE SWIG_SOURCES *.swig)
+
+set(FLAGS
+  -c++
+  -shadow
+  -python
+  -D__STDC_LIMIT_MACROS
+  -D__STDC_CONSTANT_MACROS
+  )
+
+set(INCLUDES
+  -I${LLDB_SOURCE_DIR}/include
+  -I${LLDB_SOURCE_DIR}/tools/intel-features/intel-pt
+  )
+
+set(OUTPUT_PYTHON_WRAPPER
+  ${CMAKE_CURRENT_BINARY_DIR}/IntelFeaturesPythonWrap.cpp
+  )
+
+set(OUTPUT_PYTHON_SCRIPT_DIR
+  ${CMAKE_CURRENT_BINARY_DIR}
+  )
+
+find_package(SWIG REQUIRED)
+add_custom_command(
+  OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/IntelFeaturesPythonWrap.cpp
+  OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/lldbIntelFeatures.py
+  DEPENDS ${SWIG_SOURCES}
+  COMMAND ${SWIG_EXECUTABLE} ${FLAGS} ${INCLUDES} -o ${OUTPUT_PYTHON_WRAPPER} -outdir ${OUTPUT_PYTHON_SCRIPT_DIR} ${SWIG_SOURCES}
+  COMMENT "Generating python wrapper for features library")
+
+set_source_files_properties(${CMAKE_CURRENT_BINARY_DIR}/IntelFeaturesPythonWrap.cpp PROPERTIES GENERATED 1)
+set_source_files_properties(${CMAKE_CURRENT_BINARY_DIR}/lldbIntelFeatures.py PROPERTIES GENERATED 1)
+
+add_custom_target(intel-features-swig_wrapper ALL
+  DEPENDS ${CMAKE_CURRENT_BINARY_DIR}/IntelFeaturesPythonWrap.cpp
+  )

tools/intel-features/scripts/lldb-intel-features.swig

+%module lldbIntelFeatures
+
+%{
+#include "lldb/lldb-public.h"
+#include "intel-pt/PTDecoder.h"
+using namespace ptdecoder;
+%}
+
+/* Undefine GCC keyword to make Swig happy when processing glibc's stdint.h */
+#define __extension__
+
+/* Combined python typemap for all features */
+%include "python-typemaps.txt"
+
+/* Feature specific python interface files*/
+%include "../intel-pt/interface/PTDecoder.i"

tools/intel-features/scripts/python-typemaps.txt

+/* Typemap definitions to allow SWIG to properly handle some data types */
+
+// typemap for a char buffer
+%typemap(in) (char *dst, size_t dst_len) {
+   if (!PyInt_Check($input)) {
+       PyErr_SetString(PyExc_ValueError, "Expecting an integer");
+       return NULL;
+   }
+   $2 = PyInt_AsLong($input);
+   if ($2 <= 0) {
+       PyErr_SetString(PyExc_ValueError, "Positive integer expected");
+       return NULL;
+   }
+   $1 = (char *) malloc($2);
+}

tools/intel-mpx/CMakeLists.txt

-if (NOT CMAKE_SYSTEM_NAME MATCHES "Linux")
-  return ()
-endif ()
-
-include(${LLDB_PROJECT_ROOT}/cmake/LLDBDependencies.cmake)
-
-add_library(lldb-intel-mpxtable SHARED
-  IntelMPXTablePlugin.cpp
-  )
-
-target_link_libraries(lldb-intel-mpxtable
-                      PUBLIC liblldb LLVMSupport)
-
-install(TARGETS lldb-intel-mpxtable
-  LIBRARY DESTINATION bin)