Posted on October 21, 2015 at 18:04Doing postmortem analysis with a debugger into a hot system can be a challenge. You'll have to practice and figure out what level of visibility you need to figure anything out. I can halt things with a JLink and poke around in memory, but I suspect you want a full-on symbolic debugger.
If you suspect a memory leak within your system, presumably dynamic allocation, then may be you want to instrument the malloc/free, and tag the allocations, so you can figure out where it's going wrong before it fails days later. Is it a problem with heap fragmentation?
Make sure you don't have any exit paths that fail to free memory. At a minimum count the malloc/free and watch for the number diverging. It will quickly show if you're allocating and not freeing, or that you're freeing the same allocation multiple times.
My approach would be to add a ''monitor'' application into the fault handler, so I could interact with the broken system via a serial port, and have functions to scan the heap chain, and dump memory and statistics, etc. At a minimum have the fault handler except a key input, and output a diagnostic log, and sit in a loop that does that.
Use some imagination.
