University project - recommendations?

[Chris Johns]

On 1/12/2014 12:17 am, Dominik Taborsky wrote:
> On Sun, 30 Nov 2014 01:27:06 +0100, Chris Johns <chrisj at rtems.org> wrote:
>
>> On 29/11/2014 11:14 pm, Dominik Taborsky wrote:
>>> On Sat, 29 Nov 2014 04:44:53 +0100, Gedare Bloom <gedare at rtems.org>
>>> wrote:
>>>
>>>> On Fri, Nov 28, 2014 at 4:17 PM, Dominik Taborsky <bremby at seznam.cz>
>>>> wrote:
>>>>> I have browsed through the wiki/trac and I found these 3:
>>>>> 1) CFI-standard flash device interface,
>>>>> 2) CEXP integration,
>>>>> 3) TCP stack rewrite.
>>>>>
>>>> I don't know that CEXP integration is that interesting anymore. One of
>>>> the key features of CEXP is dynamic loading, which is not supported
>>>> through the RTEMS linker and loader projects (RTL). You might ask
>>>> Chris Johns if there are projects available for RTL.
>>>
>>> Chris is subscribed to this ML so he can reply now. Or should I address
>>> him directly?
>>>
>>
>> Here is fine and welcome to RTEMS.
>>
>> The RTL is in the main tree now under libld. There are some extra
>> tasks to complete, one relate to ARM veneers and adding veneer support
>> in general that is outstanding but I am not sure there is enough work
>> left in them for you. The veneer is an interesting task and has some
>> challengers.
>
> OK, I digress.
>
>>
>> One area that has plenty of work that needs doing plus ways to
>> innovate is trace. Jennifer is adding trace support to the capture
>> engine [1] for SMP and I have written a trace linker [2] that has
>> support to trace calls using printk. We need to hook all this together
>> to make a workable solution for users and we are currently looking at
>> the Common Trace Format (CTF) [3] to do this. Generating CTF lets us
>> integrate with Babletrace and the Linux Trace Toolkit Viewer (LTTV).
>>
>> The trace support we are adding attempts to solve some tough user
>> requirements from the space community such as instrumenting and
>> tracing software at the object level rather than instrumenting at the
>> source level. This means you take software that is compiled and
>> production ready and link in trace support. The support needs to be
>> efficient because it is a target side software only trace.
>> Transporting data off target needs to be generic and flexible because
>> we have so many different target types with differing hardware.
>>
>> The work involves adding suitable support to the capture engine to
>> extract the trace data and convert it to the CTF format. This may be
>> on target or off target depending on the load it places on the target.
>> We have lots of pieces of code in place but the code is new and green
>> so you would need to work across all parts. We need support for
>> generating trigger and trace control maps and this may be linked to
>> the CTF format and Babeltrace. There will be work at real-time capture
>> level, the trace linking parts on the host, and CTF integrations.
>>
>> Joel and I met three of the EfficiOS people at GSoC this year and
>> discussed this work so we would work with them.
>>
>
> So you're working on the capture engine itself,

I am not but Jennifer is. I am not sure what remaining work she has 
planned. I have been doing little bits with the trace linker and that 
code needs work to integrate with the capture engine. For example 
trigger maps plus other controls.

> while my job would be to compile the trace captured into a CTF,

If you are lucky this might be the case. How we integrate to that code 
base has not been looked at. Using this format lets us use the nice 
front end tools that exist. Any changes will require us working with the 
upstream project.

> possibly transmitting it over to some other machine and decompiling
> it again?

Yes these tasks exist. The transmission task is complex because of the 
possible overhead it can create. As Joel said we need to create trigger 
maps and merge them into the system some how plus there are possible 
stability issues where the trace overloads the target to be considered.

The capture engine has 2 parts, the capture part and a command line 
interface via the RTEMS shell. This work relates to just the capture 
engine code and not the CLI.

>
> To make sure, this means implementing some kind of state automata on
> both ends, right?
>

I am not sure this process is that exact.

Chris

>
>> Chris
>>
>> [1] https://devel.rtems.org/browser/rtems/cpukit/libmisc/capture
>> [2] https://devel.rtems.org/browser/rtems-tools/linkers/rtems-tld.cpp
>> [3] http://www.efficios.com/ctf