add custom vendor commands

Hello all,

Background:
Intel CPUs that support MPX have a limited number of bound registers. For any program that has more objects than fit into these registers, the bounds must be kept elsewhere. For this purpose, Bounds Tables (BT) are stored in application memory: for each pointer there is a bound table entry with lower bound, upper bound, check pointer value.

It would be convenient for the user to be able to access the BT and possibly also manipulate it, through commands like: 'set/show mpx-bounds <pointer_name> <lower-bound-value> <upper-bound-value>'.

Is there a way to add customized commands for vendor features, such as this one?
I have seen that CommandObjectCommands has a class to add commands interpreted by command interpreter scripts and an interface for command aliases, so I wonder if it would make sense to add something else for vendor commands. For example, I could create a new class CommandObjectVendor that allows vendors to create their own specific commands. In my case, for the MPX bound table, the commands could be:
  
  intel show mpx-bounds <pointer_name> <lower-bound-value> <upper-bound-value>
  intel set mpx-bounds <pointer_name> <lower-bound-value> <upper-bound-value>

Thanks in advance for your help,

- V.

Intel Deutschland GmbH
Registered Address: Am Campeon 10-12, 85579 Neubiberg, Germany
Tel: +49 89 99 8853-0, www.intel.de
Managing Directors: Christin Eisenschmid, Christian Lamprechter
Chairperson of the Supervisory Board: Nicole Lau
Registered Office: Munich
Commercial Register: Amtsgericht Muenchen HRB 186928

Hello all,

Background:
Intel CPUs that support MPX have a limited number of bound registers. For any program that has more objects than fit into these registers, the bounds must be kept elsewhere. For this purpose, Bounds Tables (BT) are stored in application memory: for each pointer there is a bound table entry with lower bound, upper bound, check pointer value.

It would be convenient for the user to be able to access the BT and possibly also manipulate it, through commands like: 'set/show mpx-bounds <pointer_name> <lower-bound-value> <upper-bound-value>'.

Is there a way to add customized commands for vendor features, such as this one?
I have seen that CommandObjectCommands has a class to add commands interpreted by command interpreter scripts and an interface for command aliases, so I wonder if it would make sense to add something else for vendor commands. For example, I could create a new class CommandObjectVendor that allows vendors to create their own specific commands. In my case, for the MPX bound table, the commands could be:
  
  intel show mpx-bounds <pointer_name> <lower-bound-value> <upper-bound-value>
  intel set mpx-bounds <pointer_name> <lower-bound-value> <upper-bound-value>

Thanks in advance for your help,

You can install new python commands that can do the job to work out the details.

http://lldb.llvm.org/python-reference.html

See the section named "CREATE A NEW LLDB COMMAND USING A PYTHON FUNCTION".

You can basically install a new "intel" command and parse all of the options "show mpx-bounds ..." or "set mpx-bounds ..." from within this new command. You can use the public LLDB API to look through the process and do things. Let me know if you need any help with this. There is also a API to add custom commands from C++ plug-ins. Enrico Granata did this work and can answer more questions on that. Either way, both commands (python and C++ plug-ins) have access to the public API for LLDB, so the code you will write will use the same API. So I would stick with python for now to get things working.

Some questions for you:
- How do plan on detecting that you have an intel processor?
- Do you need access to any process registers? If so, are these registers thread specific? Are these registers available currently on linux and MacOS?
- How do you locate the BT? (or do you even need to?). Is there symbol? Can you extract all values in the bounds table once you locate it?

Let me know if you have any questions,

Greg Clayton

Hi Greg,

Thanks a lot for your reply, please find below my answers.

You can install new python commands that can do the job to work out the
details.

http://lldb.llvm.org/python-reference.html

See the section named "CREATE A NEW LLDB COMMAND USING A PYTHON
FUNCTION".

You can basically install a new "intel" command and parse all of the options
"show mpx-bounds ..." or "set mpx-bounds ..." from within this new command.
You can use the public LLDB API to look through the process and do things. Let
me know if you need any help with this. There is also a API to add custom
commands from C++ plug-ins. Enrico Granata did this work and can answer
more questions on that. Either way, both commands (python and C++ plug-ins)
have access to the public API for LLDB, so the code you will write will use the
same API. So I would stick with python for now to get things working.

I think I would rather start right away with the C++ approach. If I understand correctly, it means to create a specific Intel plugin under the directory source/Plugins and use the LLDB API to add the custom command.

Some questions for you:
- How do plan on detecting that you have an intel processor?

I thought of checking if the target architecture is either llvm::Triple::x86 or llvm::Triple::x86_64. Do you think there is a better approach to this?

- Do you need access to any process registers? If so, are these registers thread
specific? Are these registers available currently on linux and MacOS?

I only need to access the MPX configuration register, BNDCFGU. As far as I know, MacOS doesn't have MPX support, so I also didn't implement the MPX support for MacOS in LLDB, which means it is only available in Linux at the moment.

- How do you locate the BT? (or do you even need to?). Is there symbol? Can
you extract all values in the bounds table once you locate it?

The bound tables are allocated contiguously in the process memory, starting from the Bound Directory address which is stored in the BNDCFGU register. Therefore, in the plugin I only need to access this register, do the appropriate calculations and then access the process memory to get the requested bound table entry.

- Val

Intel Deutschland GmbH
Registered Address: Am Campeon 10-12, 85579 Neubiberg, Germany
Tel: +49 89 99 8853-0, www.intel.de
Managing Directors: Christin Eisenschmid, Christian Lamprechter
Chairperson of the Supervisory Board: Nicole Lau
Registered Office: Munich
Commercial Register: Amtsgericht Muenchen HRB 186928

Hi Greg,

Thanks a lot for your reply, please find below my answers.

You can install new python commands that can do the job to work out the
details.

http://lldb.llvm.org/python-reference.html

See the section named "CREATE A NEW LLDB COMMAND USING A PYTHON
FUNCTION".

You can basically install a new "intel" command and parse all of the options
"show mpx-bounds ..." or "set mpx-bounds ..." from within this new command.
You can use the public LLDB API to look through the process and do things. Let
me know if you need any help with this. There is also a API to add custom
commands from C++ plug-ins. Enrico Granata did this work and can answer
more questions on that. Either way, both commands (python and C++ plug-ins)
have access to the public API for LLDB, so the code you will write will use the
same API. So I would stick with python for now to get things working.

I think I would rather start right away with the C++ approach. If I understand correctly, it means to create a specific Intel plugin under the directory source/Plugins and use the LLDB API to add the custom command.

That is one way to do this, but any change you make will require a recompile of LLDB.

The only external plug-ins we currently have with LLDB are for adding commands. See the following code in our test suite where $(trunk) is the root of your LLDB checkout:

$(trunk)/packages/Python/lldbsuite/test/functionalities/plugins/commands

This show you how to compile a C++ plug-in that will be loaded by LLDB if it is placed in the right plug-in location. So this would allow you to distribute a plug-in that can work with existing released LLDB in case you care to do so.

Some questions for you:
- How do plan on detecting that you have an intel processor?

I thought of checking if the target architecture is either llvm::Triple::x86 or llvm::Triple::x86_64. Do you think there is a better approach to this?

I guess just checking for the "BNDCFGU" register is what you will need to do?

- Do you need access to any process registers? If so, are these registers thread
specific? Are these registers available currently on linux and MacOS?

I only need to access the MPX configuration register, BNDCFGU. As far as I know, MacOS doesn't have MPX support, so I also didn't implement the MPX support for MacOS in LLDB, which means it is only available in Linux at the moment.

- How do you locate the BT? (or do you even need to?). Is there symbol? Can
you extract all values in the bounds table once you locate it?

The bound tables are allocated contiguously in the process memory, starting from the Bound Directory address which is stored in the BNDCFGU register. Therefore, in the plugin I only need to access this register, do the appropriate calculations and then access the process memory to get the requested bound table entry.

Great. This sounds easy to do. Just decide if you prefer a built in command or to create an external command shared library plug-in and you are all set to go!

Greg

>
> Hi Greg,
>
> Thanks a lot for your reply, please find below my answers.
>
>> You can install new python commands that can do the job to work out
>> the details.
>>
>> http://lldb.llvm.org/python-reference.html
>>
>> See the section named "CREATE A NEW LLDB COMMAND USING A PYTHON
>> FUNCTION".
>>
>> You can basically install a new "intel" command and parse all of the
>> options "show mpx-bounds ..." or "set mpx-bounds ..." from within this new
command.
>> You can use the public LLDB API to look through the process and do
>> things. Let me know if you need any help with this. There is also a
>> API to add custom commands from C++ plug-ins. Enrico Granata did this
>> work and can answer more questions on that. Either way, both commands
>> (python and C++ plug-ins) have access to the public API for LLDB, so
>> the code you will write will use the same API. So I would stick with python for
now to get things working.
>>
>
> I think I would rather start right away with the C++ approach. If I understand
correctly, it means to create a specific Intel plugin under the directory
source/Plugins and use the LLDB API to add the custom command.

That is one way to do this, but any change you make will require a recompile of
LLDB.

The only external plug-ins we currently have with LLDB are for adding
commands. See the following code in our test suite where $(trunk) is the root of
your LLDB checkout:

$(trunk)/packages/Python/lldbsuite/test/functionalities/plugins/commands

This show you how to compile a C++ plug-in that will be loaded by LLDB if it is
placed in the right plug-in location. So this would allow you to distribute a plug-
in that can work with existing released LLDB in case you care to do so.

Thanks for mentioning this, I will keep it in mind. However I think I prefer to implement a builtin command.

>
>> Some questions for you:
>> - How do plan on detecting that you have an intel processor?
>
> I thought of checking if the target architecture is either llvm::Triple::x86 or
llvm::Triple::x86_64. Do you think there is a better approach to this?

I guess just checking for the "BNDCFGU" register is what you will need to do?

Yeah, I can probably skip the arch checking step.

>
>> - Do you need access to any process registers? If so, are these
>> registers thread specific? Are these registers available currently on linux and
MacOS?
>
> I only need to access the MPX configuration register, BNDCFGU. As far as I
know, MacOS doesn't have MPX support, so I also didn't implement the MPX
support for MacOS in LLDB, which means it is only available in Linux at the
moment.
>
>> - How do you locate the BT? (or do you even need to?). Is there
>> symbol? Can you extract all values in the bounds table once you locate it?
>>
>
> The bound tables are allocated contiguously in the process memory, starting
from the Bound Directory address which is stored in the BNDCFGU register.
Therefore, in the plugin I only need to access this register, do the appropriate
calculations and then access the process memory to get the requested bound
table entry.

Great. This sounds easy to do. Just decide if you prefer a built in command or to
create an external command shared library plug-in and you are all set to go!

Seems so, thanks!

- Val

Intel Deutschland GmbH
Registered Address: Am Campeon 10-12, 85579 Neubiberg, Germany
Tel: +49 89 99 8853-0, www.intel.de
Managing Directors: Christin Eisenschmid, Christian Lamprechter
Chairperson of the Supervisory Board: Nicole Lau
Registered Office: Munich
Commercial Register: Amtsgericht Muenchen HRB 186928