diff options
author | Josh Stone <jistone@redhat.com> | 2009-05-07 14:49:08 -0700 |
---|---|---|
committer | Josh Stone <jistone@redhat.com> | 2009-05-07 14:52:49 -0700 |
commit | 01c2eefe76e5a0eb8dd5a9504b522cdcb11f2665 (patch) | |
tree | 4bad9b6ff586602718774470231243af8ed267ef | |
parent | bae55db94d702c373cf66006d6736fb6adb69f23 (diff) | |
download | systemtap-steved-01c2eefe76e5a0eb8dd5a9504b522cdcb11f2665.tar.gz systemtap-steved-01c2eefe76e5a0eb8dd5a9504b522cdcb11f2665.tar.xz systemtap-steved-01c2eefe76e5a0eb8dd5a9504b522cdcb11f2665.zip |
Separate the permon tapset
-rw-r--r-- | Makefile.am | 3 | ||||
-rw-r--r-- | Makefile.in | 19 | ||||
-rw-r--r-- | tapset-perfmon.cxx | 463 | ||||
-rw-r--r-- | tapsets.cxx | 437 | ||||
-rw-r--r-- | tapsets.h | 1 |
5 files changed, 484 insertions, 439 deletions
diff --git a/Makefile.am b/Makefile.am index 08775a74..dc8f060d 100644 --- a/Makefile.am +++ b/Makefile.am @@ -38,7 +38,8 @@ stap_SOURCES = main.cxx \ parse.cxx staptree.cxx elaborate.cxx translate.cxx \ tapsets.cxx buildrun.cxx loc2c.c hash.cxx mdfour.c \ cache.cxx util.cxx coveragedb.cxx dwarf_wrappers.cxx \ - tapset-been.cxx tapset-procfs.cxx tapset-timers.cxx + tapset-been.cxx tapset-procfs.cxx tapset-timers.cxx \ + tapset-perfmon.cxx stap_LDADD = @stap_LIBS@ @sqlite3_LIBS@ BUILT_SOURCES = diff --git a/Makefile.in b/Makefile.in index 7d12e709..eded2636 100644 --- a/Makefile.in +++ b/Makefile.in @@ -123,7 +123,7 @@ am_stap_OBJECTS = stap-main.$(OBJEXT) stap-parse.$(OBJEXT) \ stap-util.$(OBJEXT) stap-coveragedb.$(OBJEXT) \ stap-dwarf_wrappers.$(OBJEXT) stap-tapset-been.$(OBJEXT) \ stap-tapset-procfs.$(OBJEXT) stap-tapset-timers.$(OBJEXT) \ - $(am__objects_1) + stap-tapset-perfmon.$(OBJEXT) $(am__objects_1) stap_OBJECTS = $(am_stap_OBJECTS) stap_LINK = $(CXXLD) $(stap_CXXFLAGS) $(CXXFLAGS) $(stap_LDFLAGS) \ $(LDFLAGS) -o $@ @@ -330,7 +330,7 @@ stap_SOURCES = main.cxx parse.cxx staptree.cxx elaborate.cxx \ translate.cxx tapsets.cxx buildrun.cxx loc2c.c hash.cxx \ mdfour.c cache.cxx util.cxx coveragedb.cxx dwarf_wrappers.cxx \ tapset-been.cxx tapset-procfs.cxx tapset-timers.cxx \ - $(am__append_4) + tapset-perfmon.cxx $(am__append_4) stap_LDADD = @stap_LIBS@ @sqlite3_LIBS@ $(am__append_6) # Arrange for git_version.h to be regenerated at every "make". @@ -605,6 +605,7 @@ distclean-compile: @AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/stap-parse.Po@am__quote@ @AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/stap-staptree.Po@am__quote@ @AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/stap-tapset-been.Po@am__quote@ +@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/stap-tapset-perfmon.Po@am__quote@ @AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/stap-tapset-procfs.Po@am__quote@ @AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/stap-tapset-timers.Po@am__quote@ @AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/stap-tapsets.Po@am__quote@ @@ -1129,6 +1130,20 @@ stap-tapset-timers.obj: tapset-timers.cxx @AMDEP_TRUE@@am__fastdepCXX_FALSE@ DEPDIR=$(DEPDIR) $(CXXDEPMODE) $(depcomp) @AMDEPBACKSLASH@ @am__fastdepCXX_FALSE@ $(CXX) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(stap_CPPFLAGS) $(CPPFLAGS) $(stap_CXXFLAGS) $(CXXFLAGS) -c -o stap-tapset-timers.obj `if test -f 'tapset-timers.cxx'; then $(CYGPATH_W) 'tapset-timers.cxx'; else $(CYGPATH_W) '$(srcdir)/tapset-timers.cxx'; fi` +stap-tapset-perfmon.o: tapset-perfmon.cxx +@am__fastdepCXX_TRUE@ $(CXX) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(stap_CPPFLAGS) $(CPPFLAGS) $(stap_CXXFLAGS) $(CXXFLAGS) -MT stap-tapset-perfmon.o -MD -MP -MF $(DEPDIR)/stap-tapset-perfmon.Tpo -c -o stap-tapset-perfmon.o `test -f 'tapset-perfmon.cxx' || echo '$(srcdir)/'`tapset-perfmon.cxx +@am__fastdepCXX_TRUE@ mv -f $(DEPDIR)/stap-tapset-perfmon.Tpo $(DEPDIR)/stap-tapset-perfmon.Po +@AMDEP_TRUE@@am__fastdepCXX_FALSE@ source='tapset-perfmon.cxx' object='stap-tapset-perfmon.o' libtool=no @AMDEPBACKSLASH@ +@AMDEP_TRUE@@am__fastdepCXX_FALSE@ DEPDIR=$(DEPDIR) $(CXXDEPMODE) $(depcomp) @AMDEPBACKSLASH@ +@am__fastdepCXX_FALSE@ $(CXX) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(stap_CPPFLAGS) $(CPPFLAGS) $(stap_CXXFLAGS) $(CXXFLAGS) -c -o stap-tapset-perfmon.o `test -f 'tapset-perfmon.cxx' || echo '$(srcdir)/'`tapset-perfmon.cxx + +stap-tapset-perfmon.obj: tapset-perfmon.cxx +@am__fastdepCXX_TRUE@ $(CXX) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(stap_CPPFLAGS) $(CPPFLAGS) $(stap_CXXFLAGS) $(CXXFLAGS) -MT stap-tapset-perfmon.obj -MD -MP -MF $(DEPDIR)/stap-tapset-perfmon.Tpo -c -o stap-tapset-perfmon.obj `if test -f 'tapset-perfmon.cxx'; then $(CYGPATH_W) 'tapset-perfmon.cxx'; else $(CYGPATH_W) '$(srcdir)/tapset-perfmon.cxx'; fi` +@am__fastdepCXX_TRUE@ mv -f $(DEPDIR)/stap-tapset-perfmon.Tpo $(DEPDIR)/stap-tapset-perfmon.Po +@AMDEP_TRUE@@am__fastdepCXX_FALSE@ source='tapset-perfmon.cxx' object='stap-tapset-perfmon.obj' libtool=no @AMDEPBACKSLASH@ +@AMDEP_TRUE@@am__fastdepCXX_FALSE@ DEPDIR=$(DEPDIR) $(CXXDEPMODE) $(depcomp) @AMDEPBACKSLASH@ +@am__fastdepCXX_FALSE@ $(CXX) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(stap_CPPFLAGS) $(CPPFLAGS) $(stap_CXXFLAGS) $(CXXFLAGS) -c -o stap-tapset-perfmon.obj `if test -f 'tapset-perfmon.cxx'; then $(CYGPATH_W) 'tapset-perfmon.cxx'; else $(CYGPATH_W) '$(srcdir)/tapset-perfmon.cxx'; fi` + stap-modsign.o: modsign.cxx @am__fastdepCXX_TRUE@ $(CXX) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(stap_CPPFLAGS) $(CPPFLAGS) $(stap_CXXFLAGS) $(CXXFLAGS) -MT stap-modsign.o -MD -MP -MF $(DEPDIR)/stap-modsign.Tpo -c -o stap-modsign.o `test -f 'modsign.cxx' || echo '$(srcdir)/'`modsign.cxx @am__fastdepCXX_TRUE@ mv -f $(DEPDIR)/stap-modsign.Tpo $(DEPDIR)/stap-modsign.Po diff --git a/tapset-perfmon.cxx b/tapset-perfmon.cxx new file mode 100644 index 00000000..e3f30ece --- /dev/null +++ b/tapset-perfmon.cxx @@ -0,0 +1,463 @@ +// tapset for HW performance monitoring +// Copyright (C) 2005-2009 Red Hat Inc. +// Copyright (C) 2005-2007 Intel Corporation. +// Copyright (C) 2008 James.Bottomley@HansenPartnership.com +// +// This file is part of systemtap, and is free software. You can +// redistribute it and/or modify it under the terms of the GNU General +// Public License (GPL); either version 2, or (at your option) any +// later version. + +#include "session.h" +#include "tapsets.h" +#include "util.h" + +#include <string> + +#ifdef PERFMON +#include <perfmon/pfmlib.h> +#include <perfmon/perfmon.h> +#endif + + +using namespace std; +using namespace __gnu_cxx; + + + +// ------------------------------------------------------------------------ +// perfmon derived probes +// ------------------------------------------------------------------------ +// This is a new interface to the perfmon hw. +// + + +struct perfmon_var_expanding_visitor: public var_expanding_visitor +{ + systemtap_session & sess; + unsigned counter_number; + perfmon_var_expanding_visitor(systemtap_session & s, unsigned c): + sess(s), counter_number(c) {} + void visit_target_symbol (target_symbol* e); +}; + + +void +perfmon_var_expanding_visitor::visit_target_symbol (target_symbol *e) +{ + assert(e->base_name.size() > 0 && e->base_name[0] == '$'); + + // Synthesize a function. + functiondecl *fdecl = new functiondecl; + fdecl->tok = e->tok; + embeddedcode *ec = new embeddedcode; + ec->tok = e->tok; + bool lvalue = is_active_lvalue(e); + + if (lvalue ) + throw semantic_error("writes to $counter not permitted"); + + string fname = string("_perfmon_tvar_get") + + "_" + e->base_name.substr(1) + + "_" + lex_cast<string>(counter_number); + + if (e->base_name != "$counter") + throw semantic_error ("target variables not available to perfmon probes"); + + if (e->components.size() > 0) + { + switch (e->components[0].first) + { + case target_symbol::comp_literal_array_index: + throw semantic_error("perfmon probe '$counter' variable may not be used as array", + e->tok); + break; + case target_symbol::comp_struct_member: + throw semantic_error("perfmon probe '$counter' variable may not be used as a structure", + e->tok); + break; + default: + throw semantic_error ("invalid use of perfmon probe '$counter' variable", + e->tok); + break; + } + } + + ec->code = "THIS->__retvalue = _pfm_pmd_x[" + + lex_cast<string>(counter_number) + "].reg_num;"; + ec->code += "/* pure */"; + fdecl->name = fname; + fdecl->body = ec; + fdecl->type = pe_long; + sess.functions[fdecl->name]=fdecl; + + // Synthesize a functioncall. + functioncall* n = new functioncall; + n->tok = e->tok; + n->function = fname; + n->referent = 0; // NB: must not resolve yet, to ensure inclusion in session + + provide (n); +} + + +enum perfmon_mode +{ + perfmon_count, + perfmon_sample +}; + + +struct perfmon_derived_probe: public derived_probe +{ +protected: + static unsigned probes_allocated; + +public: + systemtap_session & sess; + string event; + perfmon_mode mode; + + perfmon_derived_probe (probe* p, probe_point* l, systemtap_session &s, + string e, perfmon_mode m); + virtual void join_group (systemtap_session& s); +}; + + +struct perfmon_derived_probe_group: public generic_dpg<perfmon_derived_probe> +{ +public: + void emit_module_decls (systemtap_session&) {} + void emit_module_init (systemtap_session&) {} + void emit_module_exit (systemtap_session&) {} +}; + + +struct perfmon_builder: public derived_probe_builder +{ + perfmon_builder() {} + virtual void build(systemtap_session & sess, + probe * base, + probe_point * location, + literal_map_t const & parameters, + vector<derived_probe *> & finished_results) + { + string event; + if (!get_param (parameters, "counter", event)) + throw semantic_error("perfmon requires an event"); + + sess.perfmon++; + + // XXX: need to revise when doing sampling + finished_results.push_back(new perfmon_derived_probe(base, location, + sess, event, + perfmon_count)); + } +}; + + +unsigned perfmon_derived_probe::probes_allocated; + +perfmon_derived_probe::perfmon_derived_probe (probe* p, probe_point* l, + systemtap_session &s, + string e, perfmon_mode m) + : derived_probe (p, l), sess(s), event(e), mode(m) +{ + ++probes_allocated; + + // Now expand the local variables in the probe body + perfmon_var_expanding_visitor v (sess, probes_allocated-1); + this->body = v.require (this->body); + + if (sess.verbose > 1) + clog << "perfmon-based probe" << endl; +} + + +void +perfmon_derived_probe::join_group (systemtap_session& s) +{ + throw semantic_error ("incomplete", this->tok); + + if (! s.perfmon_derived_probes) + s.perfmon_derived_probes = new perfmon_derived_probe_group (); + s.perfmon_derived_probes->enroll (this); +} + + +#if 0 +void +perfmon_derived_probe::emit_registrations_start (translator_output* o, + unsigned index) +{ + for (unsigned i=0; i<locations.size(); i++) + o->newline() << "enter_" << name << "_" << i << " ();"; +} + + +void +perfmon_derived_probe::emit_registrations_end (translator_output * o, + unsigned index) +{ +} + + +void +perfmon_derived_probe::emit_deregistrations (translator_output * o) +{ +} + + +void +perfmon_derived_probe::emit_probe_entries (translator_output * o) +{ + o->newline() << "#ifdef STP_TIMING"; + // NB: This variable may be multiply (but identically) defined. + o->newline() << "static __cacheline_aligned Stat " << "time_" << basest()->name << ";"; + o->newline() << "#endif"; + + for (unsigned i=0; i<locations.size(); i++) + { + probe_point *l = locations[i]; + o->newline() << "/* location " << i << ": " << *l << " */"; + o->newline() << "static void enter_" << name << "_" << i << " (void) {"; + + o->indent(1); + o->newline() << "const char* probe_point = " + << lex_cast_qstring(*l) << ";"; + + o->newline() << "static struct pfarg_ctx _pfm_context;"; + o->newline() << "static void *_pfm_desc;"; + o->newline() << "static struct pfarg_pmc *_pfm_pmc_x;"; + o->newline() << "static int _pfm_num_pmc_x;"; + o->newline() << "static struct pfarg_pmd *_pfm_pmd_x;"; + o->newline() << "static int _pfm_num_pmd_x;"; + + emit_probe_prologue (o, + (mode == perfmon_count ? + "STAP_SESSION_STARTING" : + "STAP_SESSION_RUNNING"), + "probe_point"); + + // NB: locals are initialized by probe function itself + o->newline() << name << " (c);"; + + emit_probe_epilogue (o); + + o->newline(-1) << "}\n"; + } +} +#endif + + +#if 0 +void no_pfm_event_error (string s) +{ + string msg(string("Cannot find event:" + s)); + throw semantic_error(msg); +} + + +void no_pfm_mask_error (string s) +{ + string msg(string("Cannot find mask:" + s)); + throw semantic_error(msg); +} + + +void +split(const string& s, vector<string>& v, const string & separator) +{ + string::size_type last_pos = s.find_first_not_of(separator, 0); + string::size_type pos = s.find_first_of(separator, last_pos); + + while (string::npos != pos || string::npos != last_pos) { + v.push_back(s.substr(last_pos, pos - last_pos)); + last_pos = s.find_first_not_of(separator, pos); + pos = s.find_first_of(separator, last_pos); + } +} + + +void +perfmon_derived_probe_group::emit_probes (translator_output* op, unparser* up) +{ + for (unsigned i=0; i < probes.size(); i++) + { + op->newline (); + up->emit_probe (probes[i]); + } +} + + +void +perfmon_derived_probe_group::emit_module_init (translator_output* o) +{ + int ret; + pfmlib_input_param_t inp; + pfmlib_output_param_t outp; + pfarg_pmd_t pd[PFMLIB_MAX_PMDS]; + pfarg_pmc_t pc[PFMLIB_MAX_PMCS]; + pfarg_ctx_t ctx; + pfarg_load_t load_args; + pfmlib_options_t pfmlib_options; + unsigned int max_counters; + + if ( probes.size() == 0) + return; + ret = pfm_initialize(); + if (ret != PFMLIB_SUCCESS) + throw semantic_error("Unable to generate performance monitoring events (no libpfm)"); + + pfm_get_num_counters(&max_counters); + + memset(&pfmlib_options, 0, sizeof(pfmlib_options)); + pfmlib_options.pfm_debug = 0; /* set to 1 for debug */ + pfmlib_options.pfm_verbose = 0; /* set to 1 for debug */ + pfm_set_options(&pfmlib_options); + + memset(pd, 0, sizeof(pd)); + memset(pc, 0, sizeof(pc)); + memset(&ctx, 0, sizeof(ctx)); + memset(&load_args, 0, sizeof(load_args)); + + /* + * prepare parameters to library. + */ + memset(&inp,0, sizeof(inp)); + memset(&outp,0, sizeof(outp)); + + /* figure out the events */ + for (unsigned i=0; i<probes.size(); ++i) + { + if (probes[i]->event == "cycles") { + if (pfm_get_cycle_event( &inp.pfp_events[i].event) != PFMLIB_SUCCESS) + no_pfm_event_error(probes[i]->event); + } else if (probes[i]->event == "instructions") { + if (pfm_get_inst_retired_event( &inp.pfp_events[i].event) != + PFMLIB_SUCCESS) + no_pfm_event_error(probes[i]->event); + } else { + unsigned int event_id = 0; + unsigned int mask_id = 0; + vector<string> event_spec; + split(probes[i]->event, event_spec, ":"); + int num = event_spec.size(); + int masks = num - 1; + + if (num == 0) + throw semantic_error("No events found"); + + /* setup event */ + if (pfm_find_event(event_spec[0].c_str(), &event_id) != PFMLIB_SUCCESS) + no_pfm_event_error(event_spec[0]); + inp.pfp_events[i].event = event_id; + + /* set up masks */ + if (masks > PFMLIB_MAX_MASKS_PER_EVENT) + throw semantic_error("Too many unit masks specified"); + + for (int j=0; j < masks; j++) { + if (pfm_find_event_mask(event_id, event_spec[j+1].c_str(), + &mask_id) != PFMLIB_SUCCESS) + no_pfm_mask_error(string(event_spec[j+1])); + inp.pfp_events[i].unit_masks[j] = mask_id; + } + inp.pfp_events[i].num_masks = masks; + } + } + + /* number of counters in use */ + inp.pfp_event_count = probes.size(); + + // XXX: no elimination of duplicated counters + if (inp.pfp_event_count>max_counters) + throw semantic_error("Too many performance monitoring events."); + + /* count events both in kernel and user-space */ + inp.pfp_dfl_plm = PFM_PLM0 | PFM_PLM3; + + /* XXX: some cases a perfmon register might be used of watch dog + this code doesn't handle that case */ + + /* figure out the pmcs for the events */ + if ((ret=pfm_dispatch_events(&inp, NULL, &outp, NULL)) != PFMLIB_SUCCESS) + throw semantic_error("Cannot configure events"); + + for (unsigned i=0; i < outp.pfp_pmc_count; i++) { + pc[i].reg_num = outp.pfp_pmcs[i].reg_num; + pc[i].reg_value = outp.pfp_pmcs[i].reg_value; + } + + /* + * There could be more pmc settings than pmd. + * Figure out the actual pmds to use. + */ + for (unsigned i=0, j=0; i < inp.pfp_event_count; i++) { + pd[i].reg_num = outp.pfp_pmcs[j].reg_pmd_num; + for(; j < outp.pfp_pmc_count; j++) + if (outp.pfp_pmcs[j].reg_evt_idx != i) break; + } + + // Output the be probes create function + o->newline() << "static int register_perfmon_probes (void) {"; + o->newline(1) << "int rc = 0;"; + + o->newline() << "/* data for perfmon */"; + o->newline() << "static int _pfm_num_pmc = " << outp.pfp_pmc_count << ";"; + o->newline() << "static struct pfarg_pmc _pfm_pmc[" << outp.pfp_pmc_count + << "] = {"; + /* output the needed bits for pmc here */ + for (unsigned i=0; i < outp.pfp_pmc_count; i++) { + o->newline() << "{.reg_num=" << pc[i].reg_num << ", " + << ".reg_value=" << lex_cast_hex<string>(pc[i].reg_value) + << "},"; + } + + o->newline() << "};"; + o->newline() << "static int _pfm_num_pmd = " << inp.pfp_event_count << ";"; + o->newline() << "static struct pfarg_pmd _pfm_pmd[" << inp.pfp_event_count + << "] = {"; + /* output the needed bits for pmd here */ + for (unsigned i=0; i < inp.pfp_event_count; i++) { + o->newline() << "{.reg_num=" << pd[i].reg_num << ", " + << ".reg_value=" << pd[i].reg_value << "},"; + } + o->newline() << "};"; + o->newline(); + + o->newline() << "_pfm_pmc_x=_pfm_pmc;"; + o->newline() << "_pfm_num_pmc_x=_pfm_num_pmc;"; + o->newline() << "_pfm_pmd_x=_pfm_pmd;"; + o->newline() << "_pfm_num_pmd_x=_pfm_num_pmd;"; + + // call all the function bodies associated with perfcounters + for (unsigned i=0; i < probes.size (); i++) + probes[i]->emit_registrations_start (o,i); + + /* generate call to turn on instrumentation */ + o->newline() << "_pfm_context.ctx_flags |= PFM_FL_SYSTEM_WIDE;"; + o->newline() << "rc = rc || _stp_perfmon_setup(&_pfm_desc, &_pfm_context,"; + o->newline(1) << "_pfm_pmc, _pfm_num_pmc,"; + o->newline() << "_pfm_pmd, _pfm_num_pmd);"; + o->newline(-1); + + o->newline() << "return rc;"; + o->newline(-1) << "}\n"; + + // Output the be probes destroy function + o->newline() << "static void unregister_perfmon_probes (void) {"; + o->newline(1) << "_stp_perfmon_shutdown(_pfm_desc);"; + o->newline(-1) << "}\n"; +} +#endif + + +void +register_tapset_perfmon(systemtap_session& s) +{ + s.pattern_root->bind("perfmon")->bind_str("counter") + ->bind(new perfmon_builder()); +} + +/* vim: set sw=2 ts=8 cino=>4,n-2,{2,^-2,t0,(0,u0,w1,M1 : */ diff --git a/tapsets.cxx b/tapsets.cxx index a943163e..b4b58292 100644 --- a/tapsets.cxx +++ b/tapsets.cxx @@ -58,11 +58,6 @@ extern "C" { } -#ifdef PERFMON -#include <perfmon/pfmlib.h> -#include <perfmon/perfmon.h> -#endif - using namespace std; using namespace __gnu_cxx; @@ -86,15 +81,6 @@ common_probe_entryfn_prologue (translator_output* o, string statestr, o->newline() << "cycles_t cycles_atstart = get_cycles ();"; o->newline() << "#endif"; -#if 0 /* XXX: PERFMON */ - o->newline() << "static struct pfarg_ctx _pfm_context;"; - o->newline() << "static void *_pfm_desc;"; - o->newline() << "static struct pfarg_pmc *_pfm_pmc_x;"; - o->newline() << "static int _pfm_num_pmc_x;"; - o->newline() << "static struct pfarg_pmd *_pfm_pmd_x;"; - o->newline() << "static int _pfm_num_pmd_x;"; -#endif - o->newline() << "#if INTERRUPTIBLE"; o->newline() << "preempt_disable ();"; o->newline() << "#else"; @@ -10047,426 +10033,6 @@ tracepoint_builder::build(systemtap_session& s, // ------------------------------------------------------------------------ -// perfmon derived probes -// ------------------------------------------------------------------------ -// This is a new interface to the perfmon hw. -// - - -struct perfmon_var_expanding_visitor: public var_expanding_visitor -{ - systemtap_session & sess; - unsigned counter_number; - perfmon_var_expanding_visitor(systemtap_session & s, unsigned c): - sess(s), counter_number(c) {} - void visit_target_symbol (target_symbol* e); -}; - - -void -perfmon_var_expanding_visitor::visit_target_symbol (target_symbol *e) -{ - assert(e->base_name.size() > 0 && e->base_name[0] == '$'); - - // Synthesize a function. - functiondecl *fdecl = new functiondecl; - fdecl->tok = e->tok; - embeddedcode *ec = new embeddedcode; - ec->tok = e->tok; - bool lvalue = is_active_lvalue(e); - - if (lvalue ) - throw semantic_error("writes to $counter not permitted"); - - string fname = string("_perfmon_tvar_get") - + "_" + e->base_name.substr(1) - + "_" + lex_cast<string>(counter_number); - - if (e->base_name != "$counter") - throw semantic_error ("target variables not available to perfmon probes"); - - if (e->components.size() > 0) - { - switch (e->components[0].first) - { - case target_symbol::comp_literal_array_index: - throw semantic_error("perfmon probe '$counter' variable may not be used as array", - e->tok); - break; - case target_symbol::comp_struct_member: - throw semantic_error("perfmon probe '$counter' variable may not be used as a structure", - e->tok); - break; - default: - throw semantic_error ("invalid use of perfmon probe '$counter' variable", - e->tok); - break; - } - } - - ec->code = "THIS->__retvalue = _pfm_pmd_x[" + - lex_cast<string>(counter_number) + "].reg_num;"; - ec->code += "/* pure */"; - fdecl->name = fname; - fdecl->body = ec; - fdecl->type = pe_long; - sess.functions[fdecl->name]=fdecl; - - // Synthesize a functioncall. - functioncall* n = new functioncall; - n->tok = e->tok; - n->function = fname; - n->referent = 0; // NB: must not resolve yet, to ensure inclusion in session - - provide (n); -} - - -enum perfmon_mode -{ - perfmon_count, - perfmon_sample -}; - - -struct perfmon_derived_probe: public derived_probe -{ -protected: - static unsigned probes_allocated; - -public: - systemtap_session & sess; - string event; - perfmon_mode mode; - - perfmon_derived_probe (probe* p, probe_point* l, systemtap_session &s, - string e, perfmon_mode m); - virtual void join_group (systemtap_session& s); -}; - - -struct perfmon_derived_probe_group: public generic_dpg<perfmon_derived_probe> -{ -public: - void emit_module_decls (systemtap_session&) {} - void emit_module_init (systemtap_session&) {} - void emit_module_exit (systemtap_session&) {} -}; - - -struct perfmon_builder: public derived_probe_builder -{ - perfmon_builder() {} - virtual void build(systemtap_session & sess, - probe * base, - probe_point * location, - literal_map_t const & parameters, - vector<derived_probe *> & finished_results) - { - string event; - if (!get_param (parameters, "counter", event)) - throw semantic_error("perfmon requires an event"); - - sess.perfmon++; - - // XXX: need to revise when doing sampling - finished_results.push_back(new perfmon_derived_probe(base, location, - sess, event, - perfmon_count)); - } -}; - - -unsigned perfmon_derived_probe::probes_allocated; - -perfmon_derived_probe::perfmon_derived_probe (probe* p, probe_point* l, - systemtap_session &s, - string e, perfmon_mode m) - : derived_probe (p, l), sess(s), event(e), mode(m) -{ - ++probes_allocated; - - // Now expand the local variables in the probe body - perfmon_var_expanding_visitor v (sess, probes_allocated-1); - this->body = v.require (this->body); - - if (sess.verbose > 1) - clog << "perfmon-based probe" << endl; -} - - -void -perfmon_derived_probe::join_group (systemtap_session& s) -{ - throw semantic_error ("incomplete", this->tok); - - if (! s.perfmon_derived_probes) - s.perfmon_derived_probes = new perfmon_derived_probe_group (); - s.perfmon_derived_probes->enroll (this); -} - - -#if 0 -void -perfmon_derived_probe::emit_registrations_start (translator_output* o, - unsigned index) -{ - for (unsigned i=0; i<locations.size(); i++) - o->newline() << "enter_" << name << "_" << i << " ();"; -} - - -void -perfmon_derived_probe::emit_registrations_end (translator_output * o, - unsigned index) -{ -} - - -void -perfmon_derived_probe::emit_deregistrations (translator_output * o) -{ -} - - -void -perfmon_derived_probe::emit_probe_entries (translator_output * o) -{ - o->newline() << "#ifdef STP_TIMING"; - // NB: This variable may be multiply (but identically) defined. - o->newline() << "static __cacheline_aligned Stat " << "time_" << basest()->name << ";"; - o->newline() << "#endif"; - - for (unsigned i=0; i<locations.size(); i++) - { - probe_point *l = locations[i]; - o->newline() << "/* location " << i << ": " << *l << " */"; - o->newline() << "static void enter_" << name << "_" << i << " (void) {"; - - o->indent(1); - o->newline() << "const char* probe_point = " - << lex_cast_qstring(*l) << ";"; - emit_probe_prologue (o, - (mode == perfmon_count ? - "STAP_SESSION_STARTING" : - "STAP_SESSION_RUNNING"), - "probe_point"); - - // NB: locals are initialized by probe function itself - o->newline() << name << " (c);"; - - emit_probe_epilogue (o); - - o->newline(-1) << "}\n"; - } -} -#endif - - -#if 0 -void no_pfm_event_error (string s) -{ - string msg(string("Cannot find event:" + s)); - throw semantic_error(msg); -} - - -void no_pfm_mask_error (string s) -{ - string msg(string("Cannot find mask:" + s)); - throw semantic_error(msg); -} - - -void -split(const string& s, vector<string>& v, const string & separator) -{ - string::size_type last_pos = s.find_first_not_of(separator, 0); - string::size_type pos = s.find_first_of(separator, last_pos); - - while (string::npos != pos || string::npos != last_pos) { - v.push_back(s.substr(last_pos, pos - last_pos)); - last_pos = s.find_first_not_of(separator, pos); - pos = s.find_first_of(separator, last_pos); - } -} - - -void -perfmon_derived_probe_group::emit_probes (translator_output* op, unparser* up) -{ - for (unsigned i=0; i < probes.size(); i++) - { - op->newline (); - up->emit_probe (probes[i]); - } -} - - -void -perfmon_derived_probe_group::emit_module_init (translator_output* o) -{ - int ret; - pfmlib_input_param_t inp; - pfmlib_output_param_t outp; - pfarg_pmd_t pd[PFMLIB_MAX_PMDS]; - pfarg_pmc_t pc[PFMLIB_MAX_PMCS]; - pfarg_ctx_t ctx; - pfarg_load_t load_args; - pfmlib_options_t pfmlib_options; - unsigned int max_counters; - - if ( probes.size() == 0) - return; - ret = pfm_initialize(); - if (ret != PFMLIB_SUCCESS) - throw semantic_error("Unable to generate performance monitoring events (no libpfm)"); - - pfm_get_num_counters(&max_counters); - - memset(&pfmlib_options, 0, sizeof(pfmlib_options)); - pfmlib_options.pfm_debug = 0; /* set to 1 for debug */ - pfmlib_options.pfm_verbose = 0; /* set to 1 for debug */ - pfm_set_options(&pfmlib_options); - - memset(pd, 0, sizeof(pd)); - memset(pc, 0, sizeof(pc)); - memset(&ctx, 0, sizeof(ctx)); - memset(&load_args, 0, sizeof(load_args)); - - /* - * prepare parameters to library. - */ - memset(&inp,0, sizeof(inp)); - memset(&outp,0, sizeof(outp)); - - /* figure out the events */ - for (unsigned i=0; i<probes.size(); ++i) - { - if (probes[i]->event == "cycles") { - if (pfm_get_cycle_event( &inp.pfp_events[i].event) != PFMLIB_SUCCESS) - no_pfm_event_error(probes[i]->event); - } else if (probes[i]->event == "instructions") { - if (pfm_get_inst_retired_event( &inp.pfp_events[i].event) != - PFMLIB_SUCCESS) - no_pfm_event_error(probes[i]->event); - } else { - unsigned int event_id = 0; - unsigned int mask_id = 0; - vector<string> event_spec; - split(probes[i]->event, event_spec, ":"); - int num = event_spec.size(); - int masks = num - 1; - - if (num == 0) - throw semantic_error("No events found"); - - /* setup event */ - if (pfm_find_event(event_spec[0].c_str(), &event_id) != PFMLIB_SUCCESS) - no_pfm_event_error(event_spec[0]); - inp.pfp_events[i].event = event_id; - - /* set up masks */ - if (masks > PFMLIB_MAX_MASKS_PER_EVENT) - throw semantic_error("Too many unit masks specified"); - - for (int j=0; j < masks; j++) { - if (pfm_find_event_mask(event_id, event_spec[j+1].c_str(), - &mask_id) != PFMLIB_SUCCESS) - no_pfm_mask_error(string(event_spec[j+1])); - inp.pfp_events[i].unit_masks[j] = mask_id; - } - inp.pfp_events[i].num_masks = masks; - } - } - - /* number of counters in use */ - inp.pfp_event_count = probes.size(); - - // XXX: no elimination of duplicated counters - if (inp.pfp_event_count>max_counters) - throw semantic_error("Too many performance monitoring events."); - - /* count events both in kernel and user-space */ - inp.pfp_dfl_plm = PFM_PLM0 | PFM_PLM3; - - /* XXX: some cases a perfmon register might be used of watch dog - this code doesn't handle that case */ - - /* figure out the pmcs for the events */ - if ((ret=pfm_dispatch_events(&inp, NULL, &outp, NULL)) != PFMLIB_SUCCESS) - throw semantic_error("Cannot configure events"); - - for (unsigned i=0; i < outp.pfp_pmc_count; i++) { - pc[i].reg_num = outp.pfp_pmcs[i].reg_num; - pc[i].reg_value = outp.pfp_pmcs[i].reg_value; - } - - /* - * There could be more pmc settings than pmd. - * Figure out the actual pmds to use. - */ - for (unsigned i=0, j=0; i < inp.pfp_event_count; i++) { - pd[i].reg_num = outp.pfp_pmcs[j].reg_pmd_num; - for(; j < outp.pfp_pmc_count; j++) - if (outp.pfp_pmcs[j].reg_evt_idx != i) break; - } - - // Output the be probes create function - o->newline() << "static int register_perfmon_probes (void) {"; - o->newline(1) << "int rc = 0;"; - - o->newline() << "/* data for perfmon */"; - o->newline() << "static int _pfm_num_pmc = " << outp.pfp_pmc_count << ";"; - o->newline() << "static struct pfarg_pmc _pfm_pmc[" << outp.pfp_pmc_count - << "] = {"; - /* output the needed bits for pmc here */ - for (unsigned i=0; i < outp.pfp_pmc_count; i++) { - o->newline() << "{.reg_num=" << pc[i].reg_num << ", " - << ".reg_value=" << lex_cast_hex<string>(pc[i].reg_value) - << "},"; - } - - o->newline() << "};"; - o->newline() << "static int _pfm_num_pmd = " << inp.pfp_event_count << ";"; - o->newline() << "static struct pfarg_pmd _pfm_pmd[" << inp.pfp_event_count - << "] = {"; - /* output the needed bits for pmd here */ - for (unsigned i=0; i < inp.pfp_event_count; i++) { - o->newline() << "{.reg_num=" << pd[i].reg_num << ", " - << ".reg_value=" << pd[i].reg_value << "},"; - } - o->newline() << "};"; - o->newline(); - - o->newline() << "_pfm_pmc_x=_pfm_pmc;"; - o->newline() << "_pfm_num_pmc_x=_pfm_num_pmc;"; - o->newline() << "_pfm_pmd_x=_pfm_pmd;"; - o->newline() << "_pfm_num_pmd_x=_pfm_num_pmd;"; - - // call all the function bodies associated with perfcounters - for (unsigned i=0; i < probes.size (); i++) - probes[i]->emit_registrations_start (o,i); - - /* generate call to turn on instrumentation */ - o->newline() << "_pfm_context.ctx_flags |= PFM_FL_SYSTEM_WIDE;"; - o->newline() << "rc = rc || _stp_perfmon_setup(&_pfm_desc, &_pfm_context,"; - o->newline(1) << "_pfm_pmc, _pfm_num_pmc,"; - o->newline() << "_pfm_pmd, _pfm_num_pmd);"; - o->newline(-1); - - o->newline() << "return rc;"; - o->newline(-1) << "}\n"; - - // Output the be probes destroy function - o->newline() << "static void unregister_perfmon_probes (void) {"; - o->newline(1) << "_stp_perfmon_shutdown(_pfm_desc);"; - o->newline(-1) << "}\n"; -} -#endif - - -// ------------------------------------------------------------------------ // Standard tapset registry. // ------------------------------------------------------------------------ @@ -10474,11 +10040,10 @@ void register_standard_tapsets(systemtap_session & s) { register_tapset_been(s); + register_tapset_perfmon(s); register_tapset_procfs(s); register_tapset_timers(s); - s.pattern_root->bind("perfmon")->bind_str("counter") - ->bind(new perfmon_builder()); // dwarf-based kprobe/uprobe parts dwarf_derived_probe::register_patterns(s); @@ -21,6 +21,7 @@ void common_probe_entryfn_prologue (translator_output* o, std::string statestr, void common_probe_entryfn_epilogue (translator_output* o, bool overload_processing = true); void register_tapset_been(systemtap_session& sess); +void register_tapset_perfmon(systemtap_session& sess); void register_tapset_procfs(systemtap_session& sess); void register_tapset_timers(systemtap_session& sess); |