Inside Macintosh: Mac OS Runtime Architectures /

Chapter 2 - Indirect Addressing in the CFM-Based Architecture / PowerPC Implementation

Glue Code for Named Indirect Calls

If the routine call is referenced indirectly by name, the linker generates glue code in the calling fragment and directs the compiler-generated branch to this code. If the linker finds that a call is local (that is, not cross-fragment), it replaces the GPR2 restore instruction with a NOP instruction.

The glue code takes the following steps to switch the direct data areas and execute the actual called routine:

1. Loads the pointer for the transition vector into GPR12.
2. Saves the current value of GPR2.
3. Loads the next 4 bytes of the transition vector into GPR2 (effectively switching the direct data area).
4. Jumps to the start of the actual routine.

Upon return, control passes directly to the caller (not the glue code) which restores the saved value of GPR2.

Listing 2-1 shows some sample glue code.

Listing 2-1 Glue code for a cross-fragment call

   bl moo_glue             ; call the cross-fragment glue
   lwz   R2, R2_save_offs(SP); restore the caller's base pointer

   ...

moo_glue:
   lwz   R12, tvect_of_moo(R2); get pointer to moo's transition
                           ;  vector
   stw   R2, R2_save_offs(SP); save the caller's base pointer
   lwz   R0, 0(R12)        ; get moo's entry point
   lwz   R2, 4(R12)        ; load moo's base pointer
   mtctr R0                ; move entry point to Count Register
   bctr                    ; and jump to moo

Note

The linker generates custom glue for each routine since the glue code contains the direct data area offset of the routine's transition vector.