Phase 2 takes too much time to complete

[Roberto Natella <rnatella at gmail dot com>]

Dear SystemTap developers,
I noticed that compile time for a non-cached module greatly increased
since SystemTap snapshot 20070721.
On my system (RHEL4 with 2.6.25 kernel; dual Intel Xeon(TM) CPU
2.80GHz with HT; 5 GB RAM), phase 2
takes about half an hour to complete. This problem also occurs in
SystemTap 0.8 and in most recent snapshots.
Using strace, I found that stap spends most of time for generating the
header "stap-symbols.h" in function
dump_unwindsyms() (in translate.cxx). Here it is the "guilty" code:

  // Add unwind data to be included if it exists for this module.
  size_t len = 0;
  void *unwind = get_unwind_data (m, &len);
  if (unwind != NULL)
    {
      c->output << "#ifdef STP_USE_DWARF_UNWINDER" << endl;
      c->output << "static uint8_t _stp_module_" << stpmod_idx
        << "_unwind_data[] = " << endl;
      c->output << "  {";
      for (size_t i = 0; i < len; i++)
    {
      int h = ((uint8_t *)unwind)[i];
      c->output << "0x" << hex << h << dec << ",";
      if ((i + 1) % 16 == 0)
        c->output << endl << "   ";
    }
      c->output << "};" << endl;
      c->output << "#endif /* STP_USE_DWARF_UNWINDER */" << endl;
    }

and

  for (unsigned secidx = 0; secidx < seclist.size(); secidx++)
    {
      c->output << "struct _stp_symbol "
                << "_stp_module_" << stpmod_idx<< "_symbols_" <<
secidx << "[] = {" << endl;

      // We write out a *sorted* symbol table, so the runtime doesn't
have to sort them later.
      for (addrmap_t::iterator it = addrmap[secidx].begin(); it !=
addrmap[secidx].end(); it++)
        {
          if (it->first < extra_offset)
            continue; // skip symbols that occur before our chosen base address

          c->output << "  { 0x" << hex << it->first-extra_offset << dec
                    << ", " << lex_cast_qstring (it->second) << " }," << endl;
        }
      c->output << "};" << endl;
    }

As far as I know, the "stap-symbols.h" file contains the list of
kernel symbols (preliminarly ordered by name)
in order to speed-up symbol resolution at run-time. Because the symbol
list remains the same if the kernel
is unchanged, I think that stap-symbols.h should be cached in order to
speed up phase 2. You can find a
very simple patch I made on SystemTap 0.8 which caches stap-symbols.h
into ~/.systemtap
(it only takes into account the modification time of
/boot/vmlinux-VERSION, so it can be greatly improved).
Please consider the addition of caching mechanisms for stap-symbols.h
in future releases of SystemTap.
Thank you for your support
Roberto Natella

Attachment: cache_unwind_symbols.patch
Description: Binary data