blob: a5805e8345672b20fd8474b99e40c78add4bca15 [file] [log] [blame]
- the frame state consists of the following:
- ip current instruction pointer
- sp current stack pointer value
- bsp current backing store pointer
- cfm current frame mask
these are derived from the next younger (more deeply nested) frame
as follows:
- ip == saved return-link (may be b0 or an alternate branch-reg)
- sp == if younger frame has a fixed-sized frame, sp + size-of-frame,
else saved sp
- cfm == saved ar.pfs
- bsp == if ar.bsp has been saved, saved ar.bsp, otherwise,
ar.bsp \ominus saved ar.pfs.pfm.sol
The unwind cursor should represent the machine state as it existed at
the address contained in register ip. This state consists of the
*current* frame state and the save locations in the next younger
An unwind script current takes the old save locations and updates them
for the next older frame. With the new setup, we need to update the
frame state first, without updating the other save locations. For this
to work, we need the following info:
- save location of return-link
- save location of ar.pfs
- save location of bsp (if it has been saved)
- size of stack frame (fixed case) or save location of sp
func: ...
... foo <-- call site
... <-- ip
initial state:
The unwind cursor represents the (preserved) machine state
as it existed at "ip".
Evaluating the unwind descriptors for "ip" yields the following
- frame size at call site (or previous sp)
- what registers where saved where by func before
the call site was reached
Note that there is some procedure info that needs to be obtained
for the new "ip" which is contained in the unwind descriptors.
Specifically, the following is needed:
- procedure's start address
- personality address
- pointer to language-specific data area
This info is stored in a separate proc_info structure and needs
to be obtained right after running the unwind script for func.