[fpc-devel] CFI

[Joost van der Sluis]

Hi all,

I'm working on FpDebug to add support for the Call Frame Information 
(CFI) which is part of the Dwarf specification. (Version 2 onwards)

This is a snippet of the information that the compiler typically 
generates: (x86_64-linux)

fde:
<    0><0x00401090:0x004010c1><JoJo><cie offset 0x00000000::cie index 
  0><fde offset 0x00000014 length: 0x00000024><eh offset none>
         0x00401090: <off cfa=08(r7) > <off r16=-8(cfa) >
         0x00401091: <off cfa=16(r7) > <off r3=-16(cfa) > <off 
r16=-8(cfa) >
         0x004010c0: <off cfa=08(r7) > <off r16=-8(cfa) >

This basically means that when the instruction-pointer is at 0x00401090, 
the CFA (frame) can be obtained by taking register 7 (rsp) and add 8 to it.
The return address (called r16 here) if obtained by taking the CFA, 
substract 8 and then read the value at that memory location.

So far so good. But then I need the next frame. So use the return 
address, substract one (To obtain the call address). And start over... 
but, the value of r7 is not available anymore.

But it seems like it that GDB has no problem with it.

This is what the Dwarf-5 specs say: (We/FPC always create CFI version 1 
information, which corresponds with Dwarf-2, but the description in 
Dwarf-5 is easier to understand.)

"The default rule for all columns before interpretation of the initial 
instructions is the undefined rule. However, an ABI authoring body or a 
compilation system authoring body may specify an alternate default value 
for any or all columns."

'Undefined' means that the register is not available. So in principle we 
can not unwind the stack more the one level using the CFI provided by 
FreePascal. Unless.... there is some ABI rule that overrules the value 
of the r7 register.

Does anyone know if this is the case. And what are those ABI rules 
precisely, or should we add a column to the CFI information to describe r7?

Regards,

Joost.