Startup error

I did some more debugging and I think there is an error with the HeapMem implementation:

At line 307 of HeapMem.c (bios_6_34_02_18) it says:

Error_raise(eb, HeapMem_E_memory, (IArg)obj, (IArg)reqSize);

Ii can see, that eb equals 0, so in that Error_raise function, bios will halt and abort.

In other implementations (like HeapBuf) I can see that the Error_raise-call is encapsulated in an if-clause, checking whether an error is actually to be raised, or not.

Is it correct, that in HeapMem, there is no if-statement around the Error_raise call? Maybe a bug?

Regarding my error, I'm not yet sure if that's what causes my trouble, I will investigate further. But I wanted to check with you (especially the TI guys), whether the above stated may be a bug.

Hi Kai,

The code is correct. If you pass in a NULL Error_Block and an "error" happens, the program will terminate. The original reasoning was that it helps developers find error conditions quickly. Are you calling Memory_alloc with a NULL Error_Block?  To pass in an Error_Block, it must be initialized first

Error_Block eb;
Error_init(&eb);
buffer = Memory_alloc(heap, size, align, &eb);

Note there are two ways to check to see if there is a memory allocation failure:

1. if buffer == NULL (standard way)

2. Error_check(&eb). You can refer to http://rtsc.eclipse.org/docs-tip/Using_xdc.runtime_Errors/Example_5 or xdc.runtime.Error documenation for more details.

Todd

Hi Todd,

thanks your answer. 

I think I was not completely clear. I am not calling the Error_raise function, the HeapMem-Implementation of the SYS/BIOS is doing that. 

And when I backtrace through the call stack, I see that the HeapMem_alloc function itself is called with the eb parameter set to 0.

At the moment I cannot tell you exactly, which function is the first one calling some function with eb=0, because I replaced the used heap by HeapBuf. That implementation checks first whether there is actually an error existing, and then calls (if necessary) the Error_raise function (of course with eb=0, but that is alright, because we actually have an error).

So all in all:

At the end of the HeapMem_alloc function it reads:

  Gate_leaveModule(key);

  Error_raise(eb, HeapMem_E_memory, (IArg)obj, (IArg)reqSize);

  return (NULL);

}

So you see, the implementation does not check, whether there is any error to raise at all.

In contrast the HeapBuf implementation (HeapBuf_alloc, at the beginning):

if (size > obj->blockSize) {
  Error_raise(eb, HeapBuf_E_size, (IArg)obj, (IArg)size);
  return (NULL);
}

So, it checks if there is an error, and depending on that, it raises the error.

I hope I was a little more clear.

The way I see it, there is definitely a bug in the HeapMem implementation. I just downloaded the most current release of SYS/BIOS, and it's the same in that release.

Hi Todd,

yes you're absolutely right. Thanky you! I wonder how that return statement slipped by me (must be the hours of working with that code and thus not being able to see the forest for the trees).

Thanks!