summaryrefslogtreecommitdiffstats
diff options
context:
space:
mode:
Diffstat
-rw-r--r--ChangeLog12
-rw-r--r--main.cxx8
-rw-r--r--session.h1
-rw-r--r--stap.1.in4
-rw-r--r--tapsets.cxx26
-rw-r--r--translate.cxx32
6 files changed, 81 insertions, 2 deletions
diff --git a/ChangeLog b/ChangeLog
index 39ff7105..c21d67ac 100644
--- a/ChangeLog
+++ b/ChangeLog
@@ -1,3 +1,15 @@
+2006-05-02 Will Cohen <wcohen@redhat.com>
+
+ PR 2228
+ * stap.1.in: Document "-t" option.
+ * main.cxx (main): Add "-t" option for collecting timing information.
+ * session.h (systemtap_session): Corresponding changes.
+ * tapsets.cxx (emit_probe_prologue, emit_probe_epilogue,
+ emit_probe_entries):
+ * translate.cxx: (emit_common_header, emit_module_init,
+ translate_pass): Add time collection of timing information.
+
+
2006-05-01 Frank Ch. Eigler <fche@elastic.org>
* translate.cxx (visit_print_format): Fix regression in
diff --git a/main.cxx b/main.cxx
index 8005a7ec..ab13a516 100644
--- a/main.cxx
+++ b/main.cxx
@@ -91,6 +91,7 @@ usage (systemtap_session& s)
<< " -c CMD start the probes, run CMD, and exit when it finishes"
<< endl
<< " -x PID sets target() to PID" << endl
+ << " -t benchmarking timing information generated" << endl
;
// -d: dump safety-related external references
@@ -124,6 +125,7 @@ main (int argc, char * const argv [])
s.kernel_release = string (buf.release);
s.architecture = string (buf.machine);
s.verbose = 0;
+ s.timing = 0;
s.guru_mode = false;
s.bulk_mode = false;
s.unoptimized = false;
@@ -149,7 +151,7 @@ main (int argc, char * const argv [])
while (true)
{
- int grc = getopt (argc, argv, "hVvp:I:e:o:R:r:m:kgc:x:D:bs:u");
+ int grc = getopt (argc, argv, "hVvtp:I:e:o:R:r:m:kgc:x:D:bs:u");
if (grc < 0)
break;
switch (grc)
@@ -162,6 +164,10 @@ main (int argc, char * const argv [])
s.verbose ++;
break;
+ case 't':
+ s.timing ++;
+ break;
+
case 'p':
s.last_pass = atoi (optarg);
if (s.last_pass < 1 || s.last_pass > 5)
diff --git a/session.h b/session.h
index c0416e1b..ef966d78 100644
--- a/session.h
+++ b/session.h
@@ -70,6 +70,7 @@ struct systemtap_session
int target_pid;
int last_pass;
unsigned verbose;
+ unsigned timing;
bool keep_tmpdir;
bool guru_mode;
bool bulk_mode;
diff --git a/stap.1.in b/stap.1.in
index 387fe640..31e38819 100644
--- a/stap.1.in
+++ b/stap.1.in
@@ -105,6 +105,10 @@ Unoptimized mode. Disable unused code elision during elaboration.
.BI \-b
Use relayfs-based bulk mode for kernel-to-user data transfer.
.TP
+.B \-t
+Collect timing information on the number of times probe executes
+and average amount of time spent in each probe.
+.TP
.BI \-s NUM
Use NUM megabyte buffers for kernel-to-user data transfer. On a
multiprocessor in bulk mode, this is a per-processor amount.
diff --git a/tapsets.cxx b/tapsets.cxx
index 50386480..278cc60c 100644
--- a/tapsets.cxx
+++ b/tapsets.cxx
@@ -94,7 +94,13 @@ derived_probe::emit_probe_prologue (translator_output* o,
{
o->newline() << "struct context* c;";
o->newline() << "unsigned long flags;";
+ o->newline() << "#ifdef STP_TIMING";
+ o->newline() << "cycles_t cycles_atstart;";
+ o->newline() << "#endif";
o->newline() << "local_irq_save (flags);";
+ o->newline() << "#ifdef STP_TIMING";
+ o->newline() << "cycles_atstart = get_cycles ();";
+ o->newline() << "#endif";
o->newline() << "c = per_cpu_ptr (contexts, smp_processor_id());";
o->newline() << "if (atomic_read (&session_state) != " << statereq << ")";
o->newline(1) << "goto probe_epilogue;";
@@ -139,7 +145,17 @@ derived_probe::emit_probe_epilogue (translator_output* o)
o->newline() << "atomic_dec (&c->busy);";
o->newline(-1) << "probe_epilogue:";
- o->newline(1) << "local_irq_restore (flags);";
+ o->newline(1) << "#ifdef STP_TIMING";
+ o->newline() << "{";
+ o->newline(1) << "cycles_t cycles_atend = get_cycles ();";
+ o->newline() << "int64_t cycles_elapsed = (cycles_atend > cycles_atstart)";
+ o->newline(1) << "? (int64_t) (cycles_atend - cycles_atstart)";
+ o->newline() << ": (int64_t) (~(cycles_t)0) - cycles_atstart + cycles_atend + 1;";
+ o->newline() << "_stp_stat_add(time_" << name << ",cycles_elapsed);";
+ o->indent(-1);
+ o->newline(-1) << "}";
+ o->newline() << "#endif";
+ o->newline() << "local_irq_restore (flags);";
}
@@ -198,6 +214,10 @@ be_derived_probe::emit_deregistrations (translator_output* o)
void
be_derived_probe::emit_probe_entries (translator_output* o)
{
+ o->newline() << "#ifdef STP_TIMING";
+ o->newline() << "static __cacheline_aligned Stat " << "time_" << name << ";";
+ o->newline() << "#endif";
+
for (unsigned i=0; i<locations.size(); i++)
{
probe_point *l = locations[i];
@@ -3056,6 +3076,10 @@ dwarf_derived_probe::emit_probe_entries (translator_output* o)
}
o->newline(-1) << "};";
+ o->newline();
+ o->newline() << "#ifdef STP_TIMING";
+ o->newline() << "static __cacheline_aligned Stat " << "time_" << name << ";";
+ o->newline() << "#endif";
// Construct a single entry function, and a struct kprobe pointing into
// the entry function. The entry function will call the probe function.
diff --git a/translate.cxx b/translate.cxx
index b4c54a8d..e9a297e7 100644
--- a/translate.cxx
+++ b/translate.cxx
@@ -895,6 +895,14 @@ c_unparser::emit_common_header ()
if (!session->stat_decls.empty())
o->newline() << "#include \"stat.c\"\n";
+
+ // XXX: Cannot tell if statistics are being used for the timing collection.
+ o->newline();
+ o->newline() << "#ifdef STP_TIMING";
+ o->newline() << "#include \"stat.c\"";
+ o->newline() << "#include \"arith.c\"";
+ o->newline() << "#endif";
+
}
@@ -1006,6 +1014,18 @@ c_unparser::emit_module_init ()
o->newline() << "noinline void unregister_probe_" << i << " (void) {";
o->indent(1);
session->probes[i]->emit_deregistrations (o);
+ o->newline() << "#ifdef STP_TIMING";
+ o->newline(1) << "{";
+ o->newline() << "struct stat_data *stats = _stp_stat_get (time_"
+ << session->probes[i]->name << ", 0);";
+ o->newline() << "int64_t avg = 0;";
+ o->newline() << "const char *error;";
+ o->newline() << "if (stats->count) avg = _stp_div64(&error, stats->sum, stats->count);";
+ o->newline() << "_stp_printf (\"time_" << session->probes[i]->name
+ << " %lld@%lld\\n\"," << "stats->count, avg);";
+ o->newline() << "_stp_print_flush();";
+ o->newline() << "#endif";
+ o->newline(-1) << "}";
o->newline(-1) << "}";
}
@@ -1040,6 +1060,15 @@ c_unparser::emit_module_init ()
o->newline() << "rwlock_init (& global_" << c_varname (v->name) << "_lock);";
}
+ // initialize each Stat used for timing information
+ o->newline() << "#ifdef STP_TIMING";
+ for (unsigned i=0; i<session->probes.size(); i++)
+ {
+ o->newline() << "time_" << session->probes[i]->name
+ << " = _stp_stat_init (HIST_NONE);";
+ }
+ o->newline() << "#endif";
+
for (unsigned i=0; i<session->probes.size(); i++)
{
o->newline() << "rc = register_probe_" << i << "();";
@@ -3802,6 +3831,9 @@ translate_pass (systemtap_session& s)
if (s.bulk_mode)
s.op->newline() << "#define STP_RELAYFS";
+ if (s.timing)
+ s.op->newline() << "#define STP_TIMING" << " " << s.timing ;
+
s.op->newline() << "#include \"runtime.h\"";
s.op->newline() << "#include \"current.c\"";
s.op->newline() << "#include \"stack.c\"";
generated by cgit (git 2.34.1) at 2025-08-03 00:10:53 +0000