[Bug runtime/10272] backtraces fail with 32-on-64 executables

[mjw at redhat dot com]

------- Additional Comments From mjw at redhat dot com  2010-07-21 19:02 -------
The original unwinder was written for in-kernel unwinding, so it hard codes
assumptions about register wides.

The reg_info struct that defines the knowledge about offsets and widths of
registers is defined in runtime/unwind/unwind.h. This file includes an
architecture specific header file (only i386.h and x86_64.h) that defines the
actual UNW_REGISTER_INFO used. Make sure to check the EXTRA_INFO and PTREGS_INFO
defines in unwind.h that extract the information as compile time constants from
architecture struct pt_regs in the unwind_frame_info struct. The unwinder works
by initializing a architecture specific struct unwind_frame_info with a pt_regs
struct which the unwinder then adjusts to show the state for previous frame (you
can then feed that back into the unwinder to unwind further).

The runtime/unwinder.c has some "sanity checks" like:
  unsigned long value = 0;
  #ifdef CONFIG_64BIT
        BUILD_BUG_ON(sizeof(u64) != sizeof(value));
  #else
        BUILD_BUG_ON(sizeof(u32) != sizeof(value));
  #endif

And checks for registers widths like:
  if (reg_info[retAddrReg].width != sizeof(unsigned long)) goto err;

It als uses compile time constant sizeofs to initialize and copy stuff:
  memcpy(&state->cfa, &badCFA, sizeof(state->cfa));
  memset(state->regs, 0, sizeof(state->regs));
  tableSize = sizeof(unsigned long);
 
To update the actual registers it has code like:

#define FRAME_REG(r, t) (((t *)frame)[reg_info[r].offs])

#ifndef CONFIG_64BIT
# define CASES CASE(8); CASE(16); CASE(32)
#else
# define CASES CASE(8); CASE(16); CASE(32); CASE(64)
#endif

                case Register:
                        switch (reg_info[i].width) {
#define CASE(n) case sizeof(u##n): \
                                FRAME_REG(i, u##n) = state.regs[i].value; \
                                break
                                CASES;
#undef CASE


-- 


http://sourceware.org/bugzilla/show_bug.cgi?id=10272

------- You are receiving this mail because: -------
You are the assignee for the bug, or are watching the assignee.