diff options
Diffstat (limited to 'meta/recipes-kernel/oprofile/oprofile/0001-Allow-ppc64-events-to-be-specified-with-or-without-_.patch')
-rw-r--r-- | meta/recipes-kernel/oprofile/oprofile/0001-Allow-ppc64-events-to-be-specified-with-or-without-_.patch | 206 |
1 files changed, 206 insertions, 0 deletions
diff --git a/meta/recipes-kernel/oprofile/oprofile/0001-Allow-ppc64-events-to-be-specified-with-or-without-_.patch b/meta/recipes-kernel/oprofile/oprofile/0001-Allow-ppc64-events-to-be-specified-with-or-without-_.patch new file mode 100644 index 0000000000..5eb8b8ff4f --- /dev/null +++ b/meta/recipes-kernel/oprofile/oprofile/0001-Allow-ppc64-events-to-be-specified-with-or-without-_.patch @@ -0,0 +1,206 @@ +From 36028035555297695f52e856f21920012fd64f79 Mon Sep 17 00:00:00 2001 +From: Maynard Johnson <maynardj@us.ibm.com> +Date: Fri, 11 Jan 2013 13:29:57 -0600 +Subject: [PATCH] Allow ppc64 events to be specified with or without _GRP<n> + suffix + +All events for IBM PowerPC server processors (except CYCLES) have +a _GRP<n> suffix. This is because the legacy opcontrol profiler +can only profile events in the same group (i.e., having the same +_GRP<n> suffix). But operf has no such restriction because it +can multiplex events; thus, so we should allow the user to pass +event names without the _GRP<n> suffix. + +Signed-off-by: Maynard Johnson <maynardj@us.ibm.com> +--- + doc/operf.1.in | 6 +++ + doc/oprofile.xml | 12 +++++- + pe_profiling/operf.cpp | 107 ++++++++++++++++++++++++++++++++++++++++++++++++ + utils/ophelp.c | 4 ++ + 4 files changed, 127 insertions(+), 2 deletions(-) + +diff --git a/doc/operf.1.in b/doc/operf.1.in +index b109324..03027ca 100644 +--- a/doc/operf.1.in ++++ b/doc/operf.1.in +@@ -110,6 +110,12 @@ be specified using the symbolic name. If no unit mask is specified, 0x0 will be + used as the default. + .P + .RS ++On IBM PowerPC systems, events may be specified with or without the ++.I _GRP<n> ++suffix. If no group number suffix is given, one will be automatically ++assigned; thus, OProfile post-processing tools will always show real event ++names that include the group number suffix. ++.P + When no event specification is given, the default event for the running + processor type will be used for profiling. + Use +diff --git a/doc/oprofile.xml b/doc/oprofile.xml +index 0ae2b0b..0f74726 100644 +--- a/doc/oprofile.xml ++++ b/doc/oprofile.xml +@@ -1106,10 +1106,18 @@ shown by the output of <command>ophelp</command>. Unit masks with "extra:" para + specified using the symbolic name. + </para> + <note><para> +-When using legacy mode <command>opcontrol</command> on PowerPC platforms, all events specified must be in the same group; ++When using legacy mode <command>opcontrol</command> on IBM PowerPC platforms, all events specified must be in the same group; + i.e., the group number appended to the event name (e.g. <constant><<emphasis>some-event-name</emphasis>>_GRP9 + </constant>) must be the same. +-</para></note> ++</para> ++<para> ++When profiling with <command>operf</command> on IBM PowerPC platforms, the above restriction ++regarding the same group number does not apply, and events may be ++specified with or without the group number suffix. If no group number suffix is given, one will be automatically ++assigned; thus, OProfile post-processing tools will always show real event ++names that include the group number suffix. ++</para> ++</note> + <para> + If OProfile is using timer-interrupt mode, there is no event configuration possible. + </para> +diff --git a/pe_profiling/operf.cpp b/pe_profiling/operf.cpp +index 4416b29..a776e71 100644 +--- a/pe_profiling/operf.cpp ++++ b/pe_profiling/operf.cpp +@@ -1146,6 +1146,108 @@ static void _get_event_code(operf_event_t * event) + event->evt_code = config; + } + ++#if (defined(__powerpc__) || defined(__powerpc64__)) ++/* All ppc64 events (except CYCLES) have a _GRP<n> suffix. This is ++ * because the legacy opcontrol profiler can only profile events in ++ * the same group (i.e., having the same _GRP<n> suffix). But operf ++ * can multiplex events, so we should allow the user to pass event ++ * names without the _GRP<n> suffix. ++ * ++ * If event name is not CYCLES or does not have a _GRP<n> suffix, ++ * we'll call ophelp and scan the list of events, searching for one ++ * that matches up to the _GRP<n> suffix. If we don't find a match, ++ * then we'll exit with the expected error message for invalid event name. ++ */ ++static string _handle_powerpc_event_spec(string event_spec) ++{ ++ FILE * fp; ++ char line[MAX_INPUT]; ++ size_t grp_pos; ++ string evt, retval, err_msg; ++ size_t evt_name_len; ++ bool first_non_cyc_evt_found = false; ++ bool event_found = false; ++ char event_name[OP_MAX_EVT_NAME_LEN], event_spec_str[OP_MAX_EVT_NAME_LEN + 20], * count_str; ++ string cmd = OP_BINDIR; ++ cmd += "/ophelp"; ++ ++ strncpy(event_spec_str, event_spec.c_str(), event_spec.length() + 1); ++ ++ strncpy(event_name, strtok(event_spec_str, ":"), OP_MAX_EVT_NAME_LEN); ++ count_str = strtok(NULL, ":"); ++ if (!count_str) { ++ err_msg = "Invalid count for event "; ++ goto out; ++ } ++ ++ if (!strcmp("CYCLES", event_name)) { ++ event_found = true; ++ goto out; ++ } ++ ++ evt = event_name; ++ // Need to make sure the event name truly has a _GRP<n> suffix. ++ grp_pos = evt.rfind("_GRP"); ++ if ((grp_pos != string::npos) && ((evt = evt.substr(grp_pos, string::npos))).length() > 4) { ++ unsigned long value; ++ char * end; ++ value = strtoul(evt.substr(4, string::npos).c_str(), &end, 0); ++ if (end && (*end == '\0')) { ++ // Valid group number found after _GRP, so we can skip to the end. ++ event_found = true; ++ goto out; ++ } ++ } ++ ++ // If we get here, it implies the user passed a non-CYCLES event without a GRP suffix. ++ // Lets try to find a valid suffix for it. ++ fp = popen(cmd.c_str(), "r"); ++ if (fp == NULL) { ++ cerr << "Unable to execute ophelp to get info for event " ++ << event_spec << endl; ++ exit(EXIT_FAILURE); ++ } ++ evt_name_len = strlen(event_name); ++ err_msg = "Cannot find event "; ++ while (fgets(line, MAX_INPUT, fp)) { ++ if (!first_non_cyc_evt_found) { ++ if (!strncmp(line, "PM_", 3)) ++ first_non_cyc_evt_found = true; ++ else ++ continue; ++ } ++ if (line[0] == ' ' || line[0] == '\t') ++ continue; ++ if (!strncmp(line, event_name, evt_name_len)) { ++ // Found a potential match. Check if it's a perfect match. ++ string save_event_name = event_name; ++ size_t full_evt_len = index(line, ':') - line; ++ memset(event_name, '\0', OP_MAX_EVT_NAME_LEN); ++ strncpy(event_name, line, full_evt_len); ++ string candidate = event_name; ++ if (candidate.rfind("_GRP") == evt_name_len) { ++ event_found = true; ++ break; ++ } else { ++ memset(event_name, '\0', OP_MAX_EVT_NAME_LEN); ++ strncpy(event_name, save_event_name.c_str(), evt_name_len); ++ } ++ } ++ } ++ pclose(fp); ++ ++out: ++ if (!event_found) { ++ cerr << err_msg << event_name << endl; ++ cerr << "Error retrieving info for event " ++ << event_spec << endl; ++ exit(EXIT_FAILURE); ++ } ++ retval = event_name; ++ return retval + ":" + count_str; ++} ++#endif ++ + static void _process_events_list(void) + { + string cmd = OP_BINDIR; +@@ -1154,6 +1256,11 @@ static void _process_events_list(void) + FILE * fp; + string full_cmd = cmd; + string event_spec = operf_options::evts[i]; ++ ++#if (defined(__powerpc__) || defined(__powerpc64__)) ++ event_spec = _handle_powerpc_event_spec(event_spec); ++#endif ++ + if (operf_options::callgraph) { + full_cmd += " --callgraph=1 "; + } +diff --git a/utils/ophelp.c b/utils/ophelp.c +index 53a5dde..63895c8 100644 +--- a/utils/ophelp.c ++++ b/utils/ophelp.c +@@ -652,6 +652,10 @@ int main(int argc, char const * argv[]) + case CPU_PPC64_POWER7: + case CPU_PPC64_IBM_COMPAT_V1: + event_doc = ++ "When using operf, events may be specified without a '_GRP<n>' suffix.\n" ++ "If _GRP<n> (i.e., group number) is not specified, one will be automatically\n" ++ "selected for use by the profiler. OProfile post-processing tools will\n" ++ "always show real event names that include the group number suffix.\n\n" + "Documentation for IBM POWER7 can be obtained at:\n" + "http://www.power.org/events/Power7/\n" + "No public performance monitoring doc available for older processors.\n"; +-- +1.7.9.7 + |