RTOS/66AK2G12: NDK POSIX pthread RTS SYS/BIOS STDOUT - Stuck before main()

Bernhard,

There is a known issue with use of BIOS with C++ code using GCC 7.3.1 that was reported here that may be impacting your setup as well:

https://e2e.ti.com/support/processors/f/791/p/816940/3032988#3032988

Currently, the work around is to use GCC 7.2.1. You can download the version of the tool chain from:

https://github.com/gnu-mcu-eclipse/arm-none-eabi-gcc/releases/tag/v7.2.1-1.1  

Please unzip this package and then discover the compiler within CCS.

Regards,

Rahul

Dear Rahul,

if you had looked at the test case I provided, you would know that I am working on the DSP core and therefore do not use a GCC compiler. Instead I use the C6000 CGT 8.3.4 compiler. In fact I started with 8.2.6 and moved to 8.3.4 to see whether the compiler is the cause.

Why should SYS/BIOS and C++ don't work together? We already have a working setup on C6657 and also on the 66AK2G12.

On C6657 we use NDK 2.26 that don't uses pthreads. We have no problems there.

Only on 66AK2G12 after adding the latest NDK 3.60 with pthreads usage i have the problems.

So it is likely that the issue is related to pthreads, SYS/BIOS, RTS.

Looking forward to any further feedback.

Bernhard

Rahul,

I have some further thoughts.

There is also a problem if I dont use std::cout for printing. The XDC print and the C print are also not working. So the issue is not only related to C++. Am I wrong?

What about the --multithread compiler option? If I dont use it I have no problems. In the final application I need it because we have several tasks running.

Is there maybe a misbehaviour regarding --multithread and pthreads?

Bernhard

Dear Rahul,

I did some further investigations and found out that this is definitively related to POSIX.

If I remove the NDK dependency from the project. In the product configuration and in the CFG file. I have no problems.

If I enable POSIX support of SYS/BIOS by adding the line:

var Ti_Posix = xdc.useModule('ti.posix.tirtos.Settings');

Then the problem is back.

The "problem" is that I am not able to print anything on the CCS console. None of the available print fuctions (System_printf(), printf(), std::cout) is working anymore.

I really need some help to figure out what is happening.

Bernhard

Bernhard,

The only change in NDK and POSIX in the latest SDKs happened with the ProcSDK 5.01 release where NDK 2.x to NDK 3.x migration occurred:

http://software-dl.ti.com/targetcontent/ndk/3_40_01_01/exports/ndk_3_40_01_01/docs/ndk/NDK_2_to_3_Migration_Guide.html

I have looped in the TI RTOS team to comment on any changes in Posix CIO support.  As a work around can you use UART_printf so you can see debug logs on the UART ?

Regards,

Rahul

Dear Rahul,

thank you for your feedback!

Yes, in our C6657 project we use NDK2.x version with no underlying POSIX. For our new project on the 66AK2G12 I wanted to use the latest NDK to have the new layer for easy possibility to use transport encryptoin.

Unfortunately on our own PCB we dont have UART available. So the UART_printf() does not help here.

I am looking forward to the comments of you colleagues about POSIX and CIO support. Thank your for triggering them!

Bernhard

Dear Rahul,

is there some feedback availabel from the RTOS team?

Bernhard

Hi Bernhard,

I just tried out your setup and it appears to work. Big caveats..I'm using a different device and compiler since I don't have a C6000 board handy right now. I want to make sure I'm testing the right thing though, so can you confirm what I'm testing. If so, I'll try to mimic your setup (e.g. CCS version, C6000 board, etc.)

Setup

CortexM4F device 

TI Arm compiler

SYS/BIOS 6.76.02.02 

CCS 9.1

Example

bigtime example with the following changes

1. Add posix into the .cfg (and remove all the semaphore, clock and task creation) (attached at the bottom)

2. replace contents of the main source file with your main.cpp code (minus all the osal stuff) (attached at the bottom).

Test Cases

1. Don't have --multithread setting: all three prints (System_printf, printf, and std::cout) send output to the CCS console

2. Have --multithread setting: all three prints (System_printf, printf, and std::cout) send output to the CCS console

/cfs-file/__key/communityserver-discussions-components-files/791/bigtime.cpp

/cfs-file/__key/communityserver-discussions-components-files/791/bigtime.cfg

Todd

Dear Todd,

thank you for having a look at this!

Unfortunately I still can reproduce with your test case with a C6000 compiler 8.3.4 when --multithread is active.

Please have a look at the call stack where the iostream init aborts before main() is reached:

Thanks for confirming my example has the same issue. I'm getting an engineer here to replicate your setup. It may take a day or two though.

Hi Bernhard,

Could you try adding the following lines to your .cfg file:

tls = xdc.useModule('ti.sysbios.rts.ti.ThreadLocalStorage');
tls.enableTLSSupport = true;

And were you getting any warnings when building your project without these lines in your .cfg file?

Jas

Dear Jaskaran,

I modified the loading of the POSIX module from

var Settings = xdc.useModule('ti.sysbios.posix.Settings');

to 

var Settings = xdc.useModule('ti.posix.tirtos.Settings');

according to a warning when processing the CFG file.

Yes, you are right. There are some linker warnings. I have not recoginzed. Sorry for this.

Attached find the build.log (without TLS).

With the TLS statements in the CFG file the warnings are gone. But still the same issue:

Please find the latest CFG and CPP file and the build log attached.

bseiz_test.zip

Dear Todd,

I was on vacation last week. Therefore the late reply.

Thank you for the feedback!

What are the drawbacks of this? I think it is mandatory to use the --multithread option in our project since we are using multiple tasks. 

Is there some time schedule about a potential fix?

Kind Regards,

Bernhard

Dear Todd,

sorry for the late reply I was on vacation.

What is the --multithread option effect on the RTS library? 

Is it really not necessary to activate it on a multithread environment?

What about if I want to start to exchange data between ARM and DSP core via IPC, OpenMP?

Bernhard