GCC 4.3.2, FreeRTOS and C++

raphaelpereira wrote on Friday, November 28, 2008:

Hello,

When using GCC 4.3.2 in a ARM7 (LPC2387) environment, if I have one process that creates and deletes a lot one class instance using new and delete, and that process waits on serial interrupt, sometimes the SP register gets zeroed and I have a data abort. The weird thing is that creating and removing the instance using static declaration (creating a new block for example), the problem doesn’t occur. It always occur only with new and delete.

If I use GCC 4.2 (tried with 4.2.2 and 4.2.4), the problem doesn’t occur.

What I want to know is if someone have any similar problem or if someone have success with GCC 4.3

Best Regards,

davedoors wrote on Friday, November 28, 2008:

I have still on GCC 4.2 and have not tried 4.3 but I suspect it might be something to do with the libraries more than the compiler. new and delete must be allocating memory from the heap so there is the potential for fragmentation. When they are static the heap will not be used.

raphaelpereira wrote on Friday, November 28, 2008:

But then it would be the new and delete implementation rather malloc, because I’m using the same newlib version and only changing GCC.

I also suspect that it might be a FreeRTOS context switch incompatibility with some GCC 4.3 implementation.

rtel wrote on Friday, November 28, 2008:

I’m just downloading version 4.3.2 of Yagarto but will not be able to try it until later this evening.

Regards.

raphaelpereira wrote on Friday, November 28, 2008:

Not a problem. I will keep GCC 4.2.4 until someone else can try it out.

rtel wrote on Friday, November 28, 2008:

So far I have found the SAM7X demo to work fine at -O3, but not at all at -O0.  Still working on it.

Regards.

rtel wrote on Friday, November 28, 2008:

Actually - my mistake - it is working fine with GCC V4.3.2 and with both -O0 and -O3 settings.  It was just my dodgy debug setup that was causing problems.  If I flash the board and run it stand alone (not in the debugger) then everything is ok.

I am using just C, not C++ though.

Regards.

raphaelpereira wrote on Friday, December 19, 2008:

The problem seems to be on "new" implementation, so, if using just C will not trigger the error. Could you please try C++ wih a lot of "new" and "delete"?

Regards,