Re: [Libunwind-devel] [PATCH 0/1] Fast back-trace for x86_64 for only collecting the call stack

[Arun Sharma]

On Wed, May 26, 2010 at 6:25 AM, Lassi Tuura <address@hidden> wrote:

Hi Lassi,

>
> > This patch adds new function to perform a pure stack walk without
> > unwinding, functionally similar to backtrace() but accelerated by an
> > address attribute cache the caller maintains across calls.
>
> Any news on this patch? This is rather useful for us and I would like to
> see some accelerated lookup functionality included in libunwind, even if
> it isn't this patch.

I'm a bit behind on libunwind related emails/patches, but hope to look
at your patch in more detail in the next week or so.

I've reproduced the core of your proposal here - so others can chime in too.

+     unw_getcontext(&ctx);
+     memcpy(&saved, &ctx, sizeof(ctx));
+
+     unw_init_local(&cur, &ctx);
+     if (! cache || (ret = unw_tdep_trace(&cur, addrs, &depth, cache)) < 0)
+     {
+       depth = 0;
+       unw_init_local(&cur, &saved);
+       while (depth < 128)
+       {
+         unw_word_t ip;
+         unw_get_reg(&cur, UNW_REG_IP, &ip);
+         addresses[depth++] = (void *) ip;
+         if ((ret = unw_step(&cur)) <= 0)
+           break;
+       }
+     }

I generally like the idea of having a fast back trace with a fall back
to a correct, slower, but more general case. A couple of initial
comments:

* Why not use a call back based API instead of a backtrace() style API? For eg:

+       while (depth < 128)
+       {
+         unw_word_t ip;
+         unw_get_reg(&cur, UNW_REG_IP, &ip);
+         addresses[depth++] = (void *) ip;
+         if ((ret = unw_fast_step(&cur)) <= 0)
+           break;
+       }

* I couldn't tell what is the locking protocol on the frame cache. In
the multi-threaded case, do you expect to have a global frame cache or
a per-thread frame cache?

 -Arun