HeapMem_alloc failure terminates execution even when error block is used

Vesa Peltonen

Other Parts Discussed in Thread: SYSBIOS

Why the execution is terminated when allocation fails with the following code? If I've understood correctly, the execution should be returned to the caller in case that error block != NULL.

Error_Block errorBlock;
Error_init(&errorBlock);

pPacket = (Header *) HeapMem_alloc(heap, size , 1, &errorBlock);
if (Error_check(&errorBlock))
{
// this is not fatal for us. we can do here something to recover from the situation.
}

All other parameters are ok, i.e. allocation works fine when there is room in the heap.

over 12 years ago

0 Sasha Slijepcevic over 12 years ago

TI__Mastermind 37985 points

Vesa,
there is a configuration parameter 'policy' in Error that has to be set to 'Error.UNWIND' for Error.raise to return to the caller. http://rtsc.eclipse.org/cdoc-tip/xdc/runtime/Error.html#.Policy

What's the actual error message? You could be hitting an Assert, and for them there is no Error_Block and the app terminates. See this thread for that use case http://e2e.ti.com/support/embedded/tirtos/f/355/t/258134.aspx

0 Vesa Peltonen over 12 years ago in reply to Sasha Slijepcevic

Intellectual 525 points

Hi Sasha,

I've defined in app.cfg:

var Error = xdc.useModule('xdc.runtime.Error');
Error.policy = Error.UNWIND;

The termination print is as following:

ti.sysbios.heaps.HeapMem: line 307: out of memory: handle=0x866940, size=26327
xdc.runtime.Error.raise: terminating execution

0 Vesa Peltonen over 12 years ago in reply to Vesa Peltonen

Intellectual 525 points

I noticed that the "errorblock"-thing behaves as expected before BIOS_start() is called.

The problem I'm seeing happens inside a Task. Can anyone suggest why the behavior is different in these two cases?

0 Sasha Slijepcevic over 12 years ago in reply to Vesa Peltonen

TI__Mastermind 37985 points

I don't see anything obvious in Error source code that would explain the difference. Are you setting Error.raiseHook or Error.policyFxn in your CFG script?
Can you attach the generated C file, which you can find in the package/cfg subdirectory of your app?

0 Vesa Peltonen over 12 years ago in reply to Sasha Slijepcevic

Intellectual 525 points

I'm not sure if I'm allowed to do that... :) Anyway, I'm not setting any hooks or policy functions.

One interesting test just managed to do:

1. I implemented my own Error raise function and linked it in instead of the xdc one:

--symbol_map=xdc_runtime_Error_raiseX__F=myErrorRaise

2. The function doesn't anything but print the address of the error block:

Void myErrorRaise(Error_Block *eb, Types_ModuleId mod, CString file,
    Int line, Error_Id id, IArg arg1, IArg arg2)
{
	  System_printf("My Error raise eb: %p", eb);
}

3. And surprise, the address of eb is non-zero when called from main:

My Error raise eb: @00864490

And zero when called from a task:

My Error raise eb: @00000000

for exactly the same code.

I printed the address of eb before calling HeapMem_alloc in the task, and it is non-zero:

errorBlock: @00842e90

(The error block is in the stack of the task)

0 Sasha Slijepcevic over 12 years ago in reply to Vesa Peltonen

TI__Mastermind 37985 points

Are you working in CCS? Can you set a breakpoint in 'myErrorRaise' and post the call stack?

0 Vesa Peltonen over 12 years ago in reply to Sasha Slijepcevic

Intellectual 525 points

Thanks Sasha for your help. Again my stupidness, as most of the times :(

There was another task calling HeapMem_alloc with eb = NULL about the same time as to other one. Forgot to change that to call same interface.

The tricky thing here was that the allocated size was the same (related to same thing), hence the confusion.

No problems with TI code at all.

Processors

Processors forum

HeapMem_alloc failure terminates execution even when error block is used