Near Last Call Before Branching 4.11
dufault at hda.com
Thu Jul 16 18:59:27 UTC 2015
> On Jul 16, 2015, at 13:55 , Joel Sherrill <Joel.Sherrill at oarcorp.com> wrote:
> I am in the middle of a build sweep now with the tms570 patches
> and my bsp.h changes. Once this finishes, I plan to push those
> changes and cut the branch.
> Speak now.
I have a two concerns but they’ll have to be fixed after the branch. This is Phytec MPC5554.
- My flash code had a dereference of the NULL pointer (expected) and it triggered the GCC -O2 optimization that replaced it with a TRAP instruction. When I disassembled the application binary there were 39 calls to a TRAP instruction total, many in RTEMS code. I recompiled with -fno-delete-null-pointer-checks (both RTEMS and my application) and there were 2. I haven’t had time to go back and analyze this.
- I was having NFS crashes regularly. I removed a pthread_mutex_trylock() in an ISR (I assumed that was safe) while trying to make that ISR faster and the NFS issues disappeared. I haven’t had time to go back to that either.
On a plus note the -ffunction-sections -fdata-sections size optimization has worked great and I didn’t need to make any changes other than rebuilding with that.
I’m on a project that’s late and that’s why I haven’t kept up on this.
HD Associates, Inc. Software and System Engineering
More information about the devel