BIOS heapsize

Hi all

I am currently developing software for a board with a C6678 and 2GB DDR3 memory, using SYS/BIOS 6.34.04.22.

I'm using my own platform-file, where I have defined a memory segement "DDR3" of length 0x5c000000 at 0x84000000.

In my .cfg-file, I have created a heapMem instance of size 0x30000000 in section ".data_heap_sec" in that segment, and I have also set that instance to be the BIOS heap my stating: Memory.defaultHeapInstance = Program.gloabl.data_heap in the cfg-file.

Compiling and linking is fine, but looking in the map-file, it seems that my section ".data_heap_sec" is located at 0x84000000 (ok) but only of size 0x10000000 (not ok), and that the .text section follows directly after at 0x94000000.

However, if I in the code use the HeapMem_getStats, I get that the size is 0x30000000 and that I have almost that size free. Testting to allocate memory reveals that I can allocate 0x10000000 two times, and the second memory chunk is well within the next section (.text) and if I try to write to my allocated memory I the program crashes.

So, to me it seems to be a conflict between the cfg-file and the map-file, and between the map-file and what I get using the code.

Is there any size limitation I have missed regarding BIOS heap?

/OH

over 11 years ago

0 ToddMullanix over 11 years ago

TI__Guru* 96960 points

Hi Oh,

It is not a SYS/BIOS limitation. It is a issue with the CodeGen Tools. Please refer to http://e2e.ti.com/support/development_tools/compiler/f/343/p/253348/886334.aspx#886334 for more details.

Todd

0 OH over 11 years ago in reply to ToddMullanix

Intellectual 260 points

Thanks for the reply.

So even if I never dynamically allocates any data bigger than 256 MB, the limitation will hit me because the heap itself is a big chunk of data allocated by the defective tools?

And is the descrepancy between the map-file that shows 0x10000000 and the BIOS function HeapMem_getStats that shows 0x30000000 also due to the same bug?

/OH

0 ToddMullanix over 11 years ago in reply to OH

TI__Guru* 96960 points

When a heap is statically created (in the .cfg), a char array is defined. The length of that array is the size of the heap. The codegen issue is for any large array (I believe it is for 512MB and over, not 256MB...it looks like the top 3 bits are being ignored). So the buffer in the heap is the wrong size. This is reflected in mapfile. I expect bad things are happening during runtime allocation since the buffer is smaller than the heap implementation thinks it is. I'll confirm this later today.

Todd

0 ToddMullanix over 11 years ago in reply to ToddMullanix

TI__Guru* 96960 points

Yep. Bad things could happen. SYS/BIOS thinks the heap is the requested size. So corruption can occur if something is placed into the memory right after the "too-small" heap.

Todd

Processors

Processors forum

BIOS heapsize