This adds the capability for a task to read a perf counter.

usage syntax:
perf.type(0).config(0).counter("a")
...
probe process("name")...
{...
  x = @perf("name")
}

translation data structures
tapsets.cxx
  dwarf_var_expanding_visitor
   std::vector<derived_probe*> perf_counter_refs;
    A vector of derived probes corresponding to @perf ops used by this 
probe.
session.h
  std::map<std::string,derived_probe*> perf_counters;
   A mapping from the counter name to the associated probe

runtime data structures
  stap_inode_uprobe_consumers
   perf_counters is a list of indices into stap_perf_probes
   corresponding to the perf counters that a process probe uses.  e.g.
   {0, 2}  AND
   static struct stap_perf_probe stap_perf_probes [3] = {...}
uprobes-inode.c::stapiu_target_reg
  calls _stp_perf_read_init to setup a counter corresponding to each
   @perf using the index to stap_perf_probes
  _stp_perf_read is passed the same index to read the counter

EXAMPLE
for the script:
global perm_begin
global perm_end
global tower_begin
global tower_end
probe perf.type(0).config(0).counter("a") { }
probe perf.type(0).config(0).counter("b") { }
probe perf.type(0).config(1).counter("c") { }
probe process ("/work/scox/stap/perf/bench.x").statement("*@bench.c:1131")
{
  perm_begin <<< @perf("a")
  perm_begin <<< @perf("c")
}
probe process ("/work/scox/stap/perf/bench.x").statement("tower@bench.c")
{}
probe process ("/work/scox/stap/perf/bench.x").statement("*@bench.c:1131")
{
  perm_end <<< @perf("a")
  perm_begin <<< @perf("c")
}
probe process ("/work/scox/stap/perf/bench.x").statement("*@bench.c:1135")
{
  tower_begin <<< @perf("b")
}
probe process ("/work/scox/stap/perf/bench.x").statement("*@bench.c:1136")
{
  tower_end <<< @perf("b")
}


long perf_counters_0[] = {0, 2};
long perf_counters_1[] = {};
long perf_counters_2[] = {0, 2};
long perf_counters_3[] = {1};
long perf_counters_4[] = {1};
static struct stapiu_consumer stap_inode_uprobe_consumers[] = {
   { .target=&stap_inode_uprobe_targets[0], .offset=(loff_t)0x31f1ULL, 
.perf_counters_dim=ARRAY_SIZE(perf_counters_0), 
.perf_counters=&perf_counters_0, .probe=(&stap_probes[3]), },
   { .target=&stap_inode_uprobe_targets[0], .offset=(loff_t)0xa7aULL, 
.perf_counters_dim=ARRAY_SIZE(perf_counters_1), 
.perf_counters=&perf_counters_1, .probe=(&stap_probes[4]), },

...
static struct stap_perf_probe stap_perf_probes [3] = {
   {
     .attr={ .type=0ULL, .config=0ULL, { .sample_period=0ULL }},
     .callback=enter_perf_probe_0,
     .probe=(&stap_probes[0]),
     .per_thread=1,
   },
   {
     .attr={ .type=0ULL, .config=0ULL, { .sample_period=0ULL }},
     .callback=enter_perf_probe_1,
     .probe=(&stap_probes[1]),
     .per_thread=1,
   },
   {
     .attr={ .type=0ULL, .config=1ULL, { .sample_period=0ULL }},
     .callback=enter_perf_probe_2,
     .probe=(&stap_probes[2]),
     .per_thread=1,
   },
};
... and a read is done with:
l->l___perf_read_b = (((int64_t) 
(_stp_perf_read(smp_processor_id(),1)))); # where 1 is the index into 
stap_perf_probes corresponding to counter("b")