Re: [libunwind] backtrace on EM64T

[Arun Sharma (अरुण)]

Yeshurun, Meir wrote:
> Hi,
>
> I failed to grep dwarf_cie_t.
>
> Anyway, I thought the cache should be ip -> dwarf_reg_state. Does
> that make sense?

I think you're right.

These routines are fetching the unwind information:

  if ((ret = fetch_proc_info (c, c->ip, 1)) < 0)
    return ret;

  if ((ret = create_state_record_for (c, &sr, c->ip)) < 0)
    return ret;

And this one is actually doing the unwind:

  apply_reg_state (c, &sr.rs_current)

So if we create a ip -> sr.rs_current cache, we don't have to execute 
the fetch_proc_info() and create_state_record_for() everytime.

I noticed that ia64 has state records as well, but prefers to build a 
"script" from those state records for faster execution.

Because the x86-64 unwind information is relatively less complicated, we 
can skip the scripts and just execute the sr.rs_current directly.

        -Arun