The Joys of Backtrace

Just spent a good long while getting a fairly simple backtrace to work. Now there is something crying out for a HowTo. I have PC, function name (if available), and function address. I can’t figure out how to extract the saved registers, as it appears to just push what is necessary (so without finding the STMFD instruction and taking it apart, can you even tell which registers were saved…if any?
I wonder, also, if there is a stored register dump from the point of a fault happening? I am testing my code by intentionally raising a type 5 (SIGSEGV).
I presume there is, it has just been a long time since I’ve seen a C style backtrace. Usually the ones I see these days are from NetSurf. ;-)

Along the way I noticed some Korean outfit has grabbed a copy of my assembler tutorial (I guess they don’t understand what off-line means?). http://www.pldworld.com/altera/html/excalibur/asm-site/
That’s fine. The Koreans can help themselves to my programming documents, I’ll help myself to some movies, then everybody is happy. Oh, wait, copyright only works for the big guys, right? <sigh>

I did notice that CC 5.70 can be easily trashed by referring to PC in an inline __asm block.
Something like the following:

   __asm
   {
      CMP   a1, a1
      TEQ   pc, pc
      MOVEQ a1, #32
      MOVNE a1, #26
      MOV   cpuworld, a1
   }

The result?

“c.main”, line 79: Warning: PSR corrupted but possibly reused later. This code may not work correctly
“c.main”, line 82: Fatal internal error: outinstrS(e13f000f)

Internal inconsistency: either resource shortage or compiler fault. If you
cannot alter your program to avoid this failure, please contact your supplier
AMU: * exit (1) *

So, back to writing proper assembler. ;-)

Along the way I noticed some Korean outfit has grabbed a copy of my assembler tutorial

They left in the copyright references. You could look on it as free hosting.

After looking at how CLib does it, yeah. You have the frame creation instruction loaded, so it is just a small matter of pulling it apart to work out which registers were actually saved, to then retrieve them. A bit of work with Theo’s docs, *Memory, and the ARM ARM and I got it sussed.

 Backtrace:
  PC = &FC17B1DC : Function <unknown> (at &FC17B180)
  PC = &0000BC48 : Function IO_ParseCode (at &0000ADB8)
       R0 = &0001C990  R1 = &00015C90
  PC = &0000ACA4 : Function IO_SendString (at &0000AB8C)
       R0 = &00022D8C
  PC = &0000AB30 : Function IO_SendObject (at &0000AAF0)
       R0 = &000086D8
  PC = &0000847C : Function main (at &00008098)
       R0 = &00000003  R1 = &0001CFDC

It would be nice to have some crash-site registers, but I’m not sure if these are recorded. Removing all my signal handling gave me an old style CLib backtrace which didn’t list at-crash registers, so I am guessing maybe CLib doesn’t do this?

Oh my. It’s quarter past one (already?). Time to quit for tonight!