[PATCH 0/9] perf: consolidate all the open counters loops

[PATCH 0/9] perf: consolidate all the open counters loops

[David Ahern]

ACME was a litle slow today (ACME Component Mgmt Env that is) so managed
to add perf-stat to the list and do a decent amount of testing. This
consolidates all of the open counters loops into a single common one.

David Ahern (9):
  perf python: add ui stubs file
  perf top: make use of perf_record_opts
  perf evlist: introduce open counters method
  perf top: use the new perf_evlist__open_counters method
  perf record: use the new perf_evlist__open_counters method
  perf stat: move user options to perf_record_opts
  perf evlist: add stat unique code to open_counters method
  perf stat: move to perf_evlist__open_counters
  perf evsel: remove perf_evsel__open_per_cpu

 tools/perf/builtin-record.c        |  109 +---------------
 tools/perf/builtin-stat.c          |  240 ++++++++++++++++--------------------
 tools/perf/builtin-top.c           |  142 ++++++---------------
 tools/perf/util/evlist.c           |  139 ++++++++++++++++++++-
 tools/perf/util/evlist.h           |    4 +
 tools/perf/util/evsel.c            |    6 -
 tools/perf/util/evsel.h            |    2 -
 tools/perf/util/python-ext-sources |    1 +
 tools/perf/util/top.c              |   20 +--
 tools/perf/util/top.h              |    9 +-
 10 files changed, 303 insertions(+), 369 deletions(-)

-- 
1.7.10.1

[PATCH 1/9] perf python: add ui stubs file

[David Ahern]

stderr based implementations of ui_xxxx functions for the python
library. Needed for patch 3 - consolidating open counters method.

Signed-off-by: David Ahern <dsahern@gmail.com>
Cc: Arnaldo Carvalho de Melo <acme@ghostprotocols.net>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Peter Zijlstra <peterz@infradead.org>
---
 tools/perf/util/python-ext-sources |    1 +
 1 file changed, 1 insertion(+)

diff --git a/tools/perf/util/python-ext-sources b/tools/perf/util/python-ext-sources
index 2133628..8a45370 100644
--- a/tools/perf/util/python-ext-sources
+++ b/tools/perf/util/python-ext-sources
@@ -19,3 +19,4 @@ util/debugfs.c
 util/rblist.c
 util/strlist.c
 ../../lib/rbtree.c
+util/ui_stubs.c
-- 
1.7.10.1

[PATCH 2/9] perf top: make use of perf_record_opts

[David Ahern]

Changes top code to use the perf_record_opts struct. Stepping stone to
consolidating the open counters code.

Signed-off-by: David Ahern <dsahern@gmail.com>
Cc: Arnaldo Carvalho de Melo <acme@ghostprotocols.net>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Peter Zijlstra <peterz@infradead.org>
---
 tools/perf/builtin-top.c |   84 ++++++++++++++++++++++++----------------------
 tools/perf/util/top.c    |   20 +++++------
 tools/perf/util/top.h    |    9 +----
 3 files changed, 54 insertions(+), 59 deletions(-)

diff --git a/tools/perf/builtin-top.c b/tools/perf/builtin-top.c
index fb9da71..33c3825 100644
--- a/tools/perf/builtin-top.c
+++ b/tools/perf/builtin-top.c
@@ -591,7 +591,7 @@ static void *display_thread_tui(void *arg)
 	 * via --uid.
 	 */
 	list_for_each_entry(pos, &top->evlist->entries, node)
-		pos->hists.uid_filter_str = top->target.uid_str;
+		pos->hists.uid_filter_str = top->opts.target.uid_str;
 
 	perf_evlist__tui_browse_hists(top->evlist, help,
 				      perf_top__sort_new_samples,
@@ -891,7 +891,7 @@ static void perf_top__start_counters(struct perf_top *top)
 	struct perf_evsel *counter;
 	struct perf_evlist *evlist = top->evlist;
 
-	if (top->group)
+	if (top->opts.group)
 		perf_evlist__set_leader(evlist);
 
 	list_for_each_entry(counter, &evlist->entries, node) {
@@ -899,10 +899,10 @@ static void perf_top__start_counters(struct perf_top *top)
 
 		attr->sample_type = PERF_SAMPLE_IP | PERF_SAMPLE_TID;
 
-		if (top->freq) {
+		if (top->opts.freq) {
 			attr->sample_type |= PERF_SAMPLE_PERIOD;
 			attr->freq	  = 1;
-			attr->sample_freq = top->freq;
+			attr->sample_freq = top->opts.freq;
 		}
 
 		if (evlist->nr_entries > 1) {
@@ -910,7 +910,7 @@ static void perf_top__start_counters(struct perf_top *top)
 			attr->read_format |= PERF_FORMAT_ID;
 		}
 
-		if (perf_target__has_cpu(&top->target))
+		if (perf_target__has_cpu(&top->opts.target))
 			attr->sample_type |= PERF_SAMPLE_CPU;
 
 		if (symbol_conf.use_callchain)
@@ -918,12 +918,12 @@ static void perf_top__start_counters(struct perf_top *top)
 
 		attr->mmap = 1;
 		attr->comm = 1;
-		attr->inherit = top->inherit;
+		attr->inherit = !top->opts.no_inherit;
 fallback_missing_features:
-		if (top->exclude_guest_missing)
+		if (top->opts.exclude_guest_missing)
 			attr->exclude_guest = attr->exclude_host = 0;
 retry_sample_id:
-		attr->sample_id_all = top->sample_id_all_missing ? 0 : 1;
+		attr->sample_id_all = top->opts.sample_id_all_missing ? 0 : 1;
 try_again:
 		if (perf_evsel__open(counter, top->evlist->cpus,
 				     top->evlist->threads) < 0) {
@@ -933,17 +933,17 @@ try_again:
 				ui__error_paranoid();
 				goto out_err;
 			} else if (err == EINVAL) {
-				if (!top->exclude_guest_missing &&
+				if (!top->opts.exclude_guest_missing &&
 				    (attr->exclude_guest || attr->exclude_host)) {
 					pr_debug("Old kernel, cannot exclude "
 						 "guest or host samples.\n");
-					top->exclude_guest_missing = true;
+					top->opts.exclude_guest_missing = true;
 					goto fallback_missing_features;
-				} else if (!top->sample_id_all_missing) {
+				} else if (!top->opts.sample_id_all_missing) {
 					/*
 					 * Old kernel, no attr->sample_id_type_all field
 					 */
-					top->sample_id_all_missing = true;
+					top->opts.sample_id_all_missing = true;
 					goto retry_sample_id;
 				}
 			}
@@ -988,7 +988,7 @@ try_again:
 		}
 	}
 
-	if (perf_evlist__mmap(evlist, top->mmap_pages, false) < 0) {
+	if (perf_evlist__mmap(evlist, top->opts.mmap_pages, false) < 0) {
 		ui__error("Failed to mmap with %d (%s)\n",
 			    errno, strerror(errno));
 		goto out_err;
@@ -1034,7 +1034,7 @@ static int __cmd_top(struct perf_top *top)
 	if (ret)
 		goto out_delete;
 
-	if (perf_target__has_task(&top->target))
+	if (perf_target__has_task(&top->opts.target))
 		perf_event__synthesize_thread_map(&top->tool, top->evlist->threads,
 						  perf_event__process,
 						  &top->session->host_machine);
@@ -1168,11 +1168,13 @@ int cmd_top(int argc, const char **argv, const char *prefix __maybe_unused)
 	struct perf_top top = {
 		.count_filter	     = 5,
 		.delay_secs	     = 2,
-		.freq		     = 4000, /* 4 KHz */
-		.mmap_pages	     = 128,
 		.sym_pcnt_filter     = 5,
-		.target		     = {
-			.uses_mmap   = true,
+		.opts = {
+			.freq		     = 4000, /* 4 KHz */
+			.mmap_pages	     = 128,
+			.target		     = {
+				.uses_mmap   = true,
+			},
 		},
 	};
 	char callchain_default_opt[] = "fractal,0.5,callee";
@@ -1180,21 +1182,21 @@ int cmd_top(int argc, const char **argv, const char *prefix __maybe_unused)
 	OPT_CALLBACK('e', "event", &top.evlist, "event",
 		     "event selector. use 'perf list' to list available events",
 		     parse_events_option),
-	OPT_INTEGER('c', "count", &top.default_interval,
-		    "event period to sample"),
-	OPT_STRING('p', "pid", &top.target.pid, "pid",
+	OPT_U64('c', "count", &top.opts.default_interval,
+		"event period to sample"),
+	OPT_STRING('p', "pid", &top.opts.target.pid, "pid",
 		    "profile events on existing process id"),
-	OPT_STRING('t', "tid", &top.target.tid, "tid",
+	OPT_STRING('t', "tid", &top.opts.target.tid, "tid",
 		    "profile events on existing thread id"),
-	OPT_BOOLEAN('a', "all-cpus", &top.target.system_wide,
+	OPT_BOOLEAN('a', "all-cpus", &top.opts.target.system_wide,
 			    "system-wide collection from all CPUs"),
-	OPT_STRING('C', "cpu", &top.target.cpu_list, "cpu",
+	OPT_STRING('C', "cpu", &top.opts.target.cpu_list, "cpu",
 		    "list of cpus to monitor"),
 	OPT_STRING('k', "vmlinux", &symbol_conf.vmlinux_name,
 		   "file", "vmlinux pathname"),
 	OPT_BOOLEAN('K', "hide_kernel_symbols", &top.hide_kernel_symbols,
 		    "hide kernel symbols"),
-	OPT_UINTEGER('m', "mmap-pages", &top.mmap_pages, "number of mmap data pages"),
+	OPT_UINTEGER('m', "mmap-pages", &top.opts.mmap_pages, "number of mmap data pages"),
 	OPT_INTEGER('r', "realtime", &top.realtime_prio,
 		    "collect data with this RT SCHED_FIFO priority"),
 	OPT_INTEGER('d', "delay", &top.delay_secs,
@@ -1203,15 +1205,15 @@ int cmd_top(int argc, const char **argv, const char *prefix __maybe_unused)
 			    "dump the symbol table used for profiling"),
 	OPT_INTEGER('f', "count-filter", &top.count_filter,
 		    "only display functions with more events than this"),
-	OPT_BOOLEAN('g', "group", &top.group,
+	OPT_BOOLEAN('g', "group", &top.opts.group,
 			    "put the counters into a counter group"),
-	OPT_BOOLEAN('i', "inherit", &top.inherit,
+	OPT_BOOLEAN('i', "inherit", &top.opts.no_inherit,
 		    "child tasks inherit counters"),
 	OPT_STRING(0, "sym-annotate", &top.sym_filter, "symbol name",
 		    "symbol to annotate"),
 	OPT_BOOLEAN('z', "zero", &top.zero,
 		    "zero history across updates"),
-	OPT_INTEGER('F', "freq", &top.freq,
+	OPT_UINTEGER('F', "freq", &top.opts.freq,
 		    "profile at this frequency"),
 	OPT_INTEGER('E', "entries", &top.print_entries,
 		    "display this many functions"),
@@ -1243,7 +1245,7 @@ int cmd_top(int argc, const char **argv, const char *prefix __maybe_unused)
 		    "Display raw encoding of assembly instructions (default)"),
 	OPT_STRING('M', "disassembler-style", &disassembler_style, "disassembler style",
 		   "Specify disassembler style (e.g. -M intel for intel syntax)"),
-	OPT_STRING('u', "uid", &top.target.uid_str, "user", "user to profile"),
+	OPT_STRING('u', "uid", &top.opts.target.uid_str, "user", "user to profile"),
 	OPT_END()
 	};
 	const char * const top_usage[] = {
@@ -1273,27 +1275,27 @@ int cmd_top(int argc, const char **argv, const char *prefix __maybe_unused)
 
 	setup_browser(false);
 
-	status = perf_target__validate(&top.target);
+	status = perf_target__validate(&top.opts.target);
 	if (status) {
-		perf_target__strerror(&top.target, status, errbuf, BUFSIZ);
+		perf_target__strerror(&top.opts.target, status, errbuf, BUFSIZ);
 		ui__warning("%s", errbuf);
 	}
 
-	status = perf_target__parse_uid(&top.target);
+	status = perf_target__parse_uid(&top.opts.target);
 	if (status) {
 		int saved_errno = errno;
 
-		perf_target__strerror(&top.target, status, errbuf, BUFSIZ);
+		perf_target__strerror(&top.opts.target, status, errbuf, BUFSIZ);
 		ui__error("%s", errbuf);
 
 		status = -saved_errno;
 		goto out_delete_evlist;
 	}
 
-	if (perf_target__none(&top.target))
-		top.target.system_wide = true;
+	if (perf_target__none(&top.opts.target))
+		top.opts.target.system_wide = true;
 
-	if (perf_evlist__create_maps(top.evlist, &top.target) < 0)
+	if (perf_evlist__create_maps(top.evlist, &top.opts.target) < 0)
 		usage_with_options(top_usage, options);
 
 	if (!top.evlist->nr_entries &&
@@ -1310,10 +1312,10 @@ int cmd_top(int argc, const char **argv, const char *prefix __maybe_unused)
 	/*
 	 * User specified count overrides default frequency.
 	 */
-	if (top.default_interval)
-		top.freq = 0;
-	else if (top.freq) {
-		top.default_interval = top.freq;
+	if (top.opts.default_interval)
+		top.opts.freq = 0;
+	else if (top.opts.freq) {
+		top.opts.default_interval = top.opts.freq;
 	} else {
 		ui__error("frequency and count are zero, aborting\n");
 		exit(EXIT_FAILURE);
@@ -1324,7 +1326,7 @@ int cmd_top(int argc, const char **argv, const char *prefix __maybe_unused)
 		 * Fill in the ones not specifically initialized via -c:
 		 */
 		if (!pos->attr.sample_period)
-			pos->attr.sample_period = top.default_interval;
+			pos->attr.sample_period = top.opts.default_interval;
 	}
 
 	top.sym_evsel = perf_evlist__first(top.evlist);
diff --git a/tools/perf/util/top.c b/tools/perf/util/top.c
index 884dde9..deea8fd 100644
--- a/tools/perf/util/top.c
+++ b/tools/perf/util/top.c
@@ -61,31 +61,31 @@ size_t perf_top__header_snprintf(struct perf_top *top, char *bf, size_t size)
 		struct perf_evsel *first = perf_evlist__first(top->evlist);
 		ret += SNPRINTF(bf + ret, size - ret, "%" PRIu64 "%s ",
 				(uint64_t)first->attr.sample_period,
-				top->freq ? "Hz" : "");
+				top->opts.freq ? "Hz" : "");
 	}
 
 	ret += SNPRINTF(bf + ret, size - ret, "%s", perf_evsel__name(top->sym_evsel));
 
 	ret += SNPRINTF(bf + ret, size - ret, "], ");
 
-	if (top->target.pid)
+	if (top->opts.target.pid)
 		ret += SNPRINTF(bf + ret, size - ret, " (target_pid: %s",
-				top->target.pid);
-	else if (top->target.tid)
+				top->opts.target.pid);
+	else if (top->opts.target.tid)
 		ret += SNPRINTF(bf + ret, size - ret, " (target_tid: %s",
-				top->target.tid);
-	else if (top->target.uid_str != NULL)
+				top->opts.target.tid);
+	else if (top->opts.target.uid_str != NULL)
 		ret += SNPRINTF(bf + ret, size - ret, " (uid: %s",
-				top->target.uid_str);
+				top->opts.target.uid_str);
 	else
 		ret += SNPRINTF(bf + ret, size - ret, " (all");
 
-	if (top->target.cpu_list)
+	if (top->opts.target.cpu_list)
 		ret += SNPRINTF(bf + ret, size - ret, ", CPU%s: %s)",
 				top->evlist->cpus->nr > 1 ? "s" : "",
-				top->target.cpu_list);
+				top->opts.target.cpu_list);
 	else {
-		if (top->target.tid)
+		if (top->opts.target.tid)
 			ret += SNPRINTF(bf + ret, size - ret, ")");
 		else
 			ret += SNPRINTF(bf + ret, size - ret, ", %d CPU%s)",
diff --git a/tools/perf/util/top.h b/tools/perf/util/top.h
index 86ff1b1..9728740 100644
--- a/tools/perf/util/top.h
+++ b/tools/perf/util/top.h
@@ -14,7 +14,7 @@ struct perf_session;
 struct perf_top {
 	struct perf_tool   tool;
 	struct perf_evlist *evlist;
-	struct perf_target target;
+	struct perf_record_opts opts;
 	/*
 	 * Symbols will be added here in perf_event__process_sample and will
 	 * get out after decayed.
@@ -24,24 +24,17 @@ struct perf_top {
 	u64		   exact_samples;
 	u64		   guest_us_samples, guest_kernel_samples;
 	int		   print_entries, count_filter, delay_secs;
-	int		   freq;
 	bool		   hide_kernel_symbols, hide_user_symbols, zero;
 	bool		   use_tui, use_stdio;
 	bool		   sort_has_symbols;
 	bool		   dont_use_callchains;
 	bool		   kptr_restrict_warned;
 	bool		   vmlinux_warned;
-	bool		   inherit;
-	bool		   group;
-	bool		   sample_id_all_missing;
-	bool		   exclude_guest_missing;
 	bool		   dump_symtab;
 	struct hist_entry  *sym_filter_entry;
 	struct perf_evsel  *sym_evsel;
 	struct perf_session *session;
 	struct winsize	   winsize;
-	unsigned int	   mmap_pages;
-	int		   default_interval;
 	int		   realtime_prio;
 	int		   sym_pcnt_filter;
 	const char	   *sym_filter;
-- 
1.7.10.1

[PATCH 3/9] perf evlist: introduce open counters method

[David Ahern]

Superset of the open counters code in perf-top and perf-record -
combining retry handling and error handling. Should be functionally
equivalent.

Signed-off-by: David Ahern <dsahern@gmail.com>
Cc: Arnaldo Carvalho de Melo <acme@ghostprotocols.net>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Peter Zijlstra <peterz@infradead.org>
---
 tools/perf/util/evlist.c |  127 +++++++++++++++++++++++++++++++++++++++++++++-
 tools/perf/util/evlist.h |    3 ++
 2 files changed, 129 insertions(+), 1 deletion(-)

diff --git a/tools/perf/util/evlist.c b/tools/perf/util/evlist.c
index a41dc4a..bce2f58 100644
--- a/tools/perf/util/evlist.c
+++ b/tools/perf/util/evlist.c
@@ -15,7 +15,7 @@
 #include "evlist.h"
 #include "evsel.h"
 #include <unistd.h>
-
+#include "debug.h"
 #include "parse-events.h"
 
 #include <sys/mman.h>
@@ -838,3 +838,128 @@ size_t perf_evlist__fprintf(struct perf_evlist *evlist, FILE *fp)
 
 	return printed + fprintf(fp, "\n");;
 }
+
+int perf_evlist__open_counters(struct perf_evlist *evlist,
+			       struct perf_record_opts *opts)
+{
+	struct perf_evsel *pos;
+	int rc = 0;
+
+	list_for_each_entry(pos, &evlist->entries, node) {
+		struct perf_event_attr *attr = &pos->attr;
+
+		/*
+		 * Carried over from perf-record:
+		 * Check if parse_single_tracepoint_event has already asked for
+		 * PERF_SAMPLE_TIME.
+		 *
+		 * XXX this is kludgy but short term fix for problems introduced by
+		 * eac23d1c that broke 'perf script' by having different sample_types
+		 * when using multiple tracepoint events when we use a perf binary
+		 * that tries to use sample_id_all on an older kernel.
+		 *
+		 * We need to move counter creation to perf_session, support
+		 * different sample_types, etc.
+		 */
+		bool time_needed = attr->sample_type & PERF_SAMPLE_TIME;
+
+fallback_missing_features:
+		if (opts->exclude_guest_missing)
+			attr->exclude_guest = attr->exclude_host = 0;
+retry_sample_id:
+		attr->sample_id_all = opts->sample_id_all_missing ? 0 : 1;
+try_again:
+		if (perf_evsel__open(pos, evlist->cpus, evlist->threads) < 0) {
+			int err = errno;
+
+			if (err == EPERM || err == EACCES) {
+				ui__error_paranoid();
+				rc = -err;
+				goto out;
+			} else if (err ==  ENODEV && opts->target.cpu_list) {
+				pr_err("No such device - did you specify"
+				       " an out-of-range profile CPU?\n");
+				rc = -err;
+				goto out;
+			} else if (err == EINVAL) {
+				if (!opts->exclude_guest_missing &&
+				    (attr->exclude_guest || attr->exclude_host)) {
+					pr_debug("Old kernel, cannot exclude "
+						 "guest or host samples.\n");
+					opts->exclude_guest_missing = true;
+					goto fallback_missing_features;
+				} else if (!opts->sample_id_all_missing) {
+					/*
+					 * Old kernel, no attr->sample_id_type_all field
+					 */
+					opts->sample_id_all_missing = true;
+					if (!opts->sample_time &&
+					    !opts->raw_samples &&
+					    !time_needed)
+						attr->sample_type &= ~PERF_SAMPLE_TIME;
+					goto retry_sample_id;
+				}
+			}
+
+			/*
+			 * If it's cycles then fall back to hrtimer
+			 * based cpu-clock-tick sw counter, which
+			 * is always available even if no PMU support:
+			 *
+			 * PPC returns ENXIO until 2.6.37 (behavior changed
+			 * with commit b0a873e).
+			 */
+			if ((err == ENOENT || err == ENXIO) &&
+			    (attr->type == PERF_TYPE_HARDWARE) &&
+			    (attr->config == PERF_COUNT_HW_CPU_CYCLES)) {
+
+				if (verbose)
+					ui__warning("Cycles event not supported,\n"
+						    "trying to fall back to cpu-clock-ticks\n");
+
+				attr->type = PERF_TYPE_SOFTWARE;
+				attr->config = PERF_COUNT_SW_CPU_CLOCK;
+				if (pos->name) {
+					free(pos->name);
+					pos->name = NULL;
+				}
+				goto try_again;
+			}
+
+			if (err == ENOENT) {
+				ui__error("The %s event is not supported.\n",
+					  perf_evsel__name(pos));
+				rc = -err;
+				goto out;
+			} else if (err == EMFILE) {
+				ui__error("Too many events are opened.\n"
+					  "Try again after reducing the number of events\n");
+				rc = -err;
+				goto out;
+			}
+
+			ui__error("sys_perf_event_open() syscall returned with "
+				  "%d (%s) for event %s.  /bin/dmesg may provide"
+				  "additional information.\n",
+				  err, strerror(err), perf_evsel__name(pos));
+
+#if defined(__i386__) || defined(__x86_64__)
+			if ((attr->type == PERF_TYPE_HARDWARE) &&
+			    (err == EOPNOTSUPP)) {
+				pr_err("No hardware sampling interrupt available."
+				       " No APIC? If so then you can boot the kernel"
+				       " with the \"lapic\" boot parameter to"
+				       " force-enable it.\n");
+				rc = -err;
+				goto out;
+			}
+#endif
+
+			pr_err("No CONFIG_PERF_EVENTS=y kernel support configured?\n");
+			rc = -err;
+			goto out;
+		}
+	}
+out:
+	return rc;
+}
diff --git a/tools/perf/util/evlist.h b/tools/perf/util/evlist.h
index 56003f7..270e546 100644
--- a/tools/perf/util/evlist.h
+++ b/tools/perf/util/evlist.h
@@ -135,4 +135,7 @@ static inline struct perf_evsel *perf_evlist__last(struct perf_evlist *evlist)
 }
 
 size_t perf_evlist__fprintf(struct perf_evlist *evlist, FILE *fp);
+
+int perf_evlist__open_counters(struct perf_evlist *evlist,
+			       struct perf_record_opts *opts);
 #endif /* __PERF_EVLIST_H */
-- 
1.7.10.1

[PATCH 4/9] perf top: use the new perf_evlist__open_counters method

[David Ahern]

Remove open counters code with all the retry and error handling in
favor of the new perf_evlist__open_counters method which is based
on the top code.

Signed-off-by: David Ahern <dsahern@gmail.com>
Cc: Arnaldo Carvalho de Melo <acme@ghostprotocols.net>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Peter Zijlstra <peterz@infradead.org>
---
 tools/perf/builtin-top.c |   70 ++--------------------------------------------
 1 file changed, 3 insertions(+), 67 deletions(-)

diff --git a/tools/perf/builtin-top.c b/tools/perf/builtin-top.c
index 33c3825..2ffc32e 100644
--- a/tools/perf/builtin-top.c
+++ b/tools/perf/builtin-top.c
@@ -919,75 +919,11 @@ static void perf_top__start_counters(struct perf_top *top)
 		attr->mmap = 1;
 		attr->comm = 1;
 		attr->inherit = !top->opts.no_inherit;
-fallback_missing_features:
-		if (top->opts.exclude_guest_missing)
-			attr->exclude_guest = attr->exclude_host = 0;
-retry_sample_id:
-		attr->sample_id_all = top->opts.sample_id_all_missing ? 0 : 1;
-try_again:
-		if (perf_evsel__open(counter, top->evlist->cpus,
-				     top->evlist->threads) < 0) {
-			int err = errno;
-
-			if (err == EPERM || err == EACCES) {
-				ui__error_paranoid();
-				goto out_err;
-			} else if (err == EINVAL) {
-				if (!top->opts.exclude_guest_missing &&
-				    (attr->exclude_guest || attr->exclude_host)) {
-					pr_debug("Old kernel, cannot exclude "
-						 "guest or host samples.\n");
-					top->opts.exclude_guest_missing = true;
-					goto fallback_missing_features;
-				} else if (!top->opts.sample_id_all_missing) {
-					/*
-					 * Old kernel, no attr->sample_id_type_all field
-					 */
-					top->opts.sample_id_all_missing = true;
-					goto retry_sample_id;
-				}
-			}
-			/*
-			 * If it's cycles then fall back to hrtimer
-			 * based cpu-clock-tick sw counter, which
-			 * is always available even if no PMU support:
-			 */
-			if ((err == ENOENT || err == ENXIO) &&
-			    (attr->type == PERF_TYPE_HARDWARE) &&
-			    (attr->config == PERF_COUNT_HW_CPU_CYCLES)) {
-
-				if (verbose)
-					ui__warning("Cycles event not supported,\n"
-						    "trying to fall back to cpu-clock-ticks\n");
-
-				attr->type = PERF_TYPE_SOFTWARE;
-				attr->config = PERF_COUNT_SW_CPU_CLOCK;
-				if (counter->name) {
-					free(counter->name);
-					counter->name = NULL;
-				}
-				goto try_again;
-			}
-
-			if (err == ENOENT) {
-				ui__error("The %s event is not supported.\n",
-					  perf_evsel__name(counter));
-				goto out_err;
-			} else if (err == EMFILE) {
-				ui__error("Too many events are opened.\n"
-					    "Try again after reducing the number of events\n");
-				goto out_err;
-			}
-
-			ui__error("The sys_perf_event_open() syscall "
-				    "returned with %d (%s).  /bin/dmesg "
-				    "may provide additional information.\n"
-				    "No CONFIG_PERF_EVENTS=y kernel support "
-				    "configured?\n", err, strerror(err));
-			goto out_err;
-		}
 	}
 
+	if (perf_evlist__open_counters(evlist, &top->opts) != 0)
+		goto out_err;
+
 	if (perf_evlist__mmap(evlist, top->opts.mmap_pages, false) < 0) {
 		ui__error("Failed to mmap with %d (%s)\n",
 			    errno, strerror(errno));
-- 
1.7.10.1

[PATCH 5/9] perf record: use the new perf_evlist__open_counters method

[David Ahern]

Remove open counters code with all the retry and error handling in
favor of the new perf_evlist__open_counters method which is based
on the existing code.

Signed-off-by: David Ahern <dsahern@gmail.com>
Cc: Arnaldo Carvalho de Melo <acme@ghostprotocols.net>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Peter Zijlstra <peterz@infradead.org>
---
 tools/perf/builtin-record.c |  109 +------------------------------------------
 1 file changed, 2 insertions(+), 107 deletions(-)

diff --git a/tools/perf/builtin-record.c b/tools/perf/builtin-record.c
index 73b5d7f..b9dcc01 100644
--- a/tools/perf/builtin-record.c
+++ b/tools/perf/builtin-record.c
@@ -224,7 +224,6 @@ static bool perf_evlist__equal(struct perf_evlist *evlist,
 
 static int perf_record__open(struct perf_record *rec)
 {
-	struct perf_evsel *pos;
 	struct perf_evlist *evlist = rec->evlist;
 	struct perf_session *session = rec->session;
 	struct perf_record_opts *opts = &rec->opts;
@@ -235,113 +234,9 @@ static int perf_record__open(struct perf_record *rec)
 	if (opts->group)
 		perf_evlist__set_leader(evlist);
 
-	list_for_each_entry(pos, &evlist->entries, node) {
-		struct perf_event_attr *attr = &pos->attr;
-		/*
-		 * Check if parse_single_tracepoint_event has already asked for
-		 * PERF_SAMPLE_TIME.
-		 *
-		 * XXX this is kludgy but short term fix for problems introduced by
-		 * eac23d1c that broke 'perf script' by having different sample_types
-		 * when using multiple tracepoint events when we use a perf binary
-		 * that tries to use sample_id_all on an older kernel.
-		 *
-		 * We need to move counter creation to perf_session, support
-		 * different sample_types, etc.
-		 */
-		bool time_needed = attr->sample_type & PERF_SAMPLE_TIME;
-
-fallback_missing_features:
-		if (opts->exclude_guest_missing)
-			attr->exclude_guest = attr->exclude_host = 0;
-retry_sample_id:
-		attr->sample_id_all = opts->sample_id_all_missing ? 0 : 1;
-try_again:
-		if (perf_evsel__open(pos, evlist->cpus, evlist->threads) < 0) {
-			int err = errno;
-
-			if (err == EPERM || err == EACCES) {
-				ui__error_paranoid();
-				rc = -err;
-				goto out;
-			} else if (err ==  ENODEV && opts->target.cpu_list) {
-				pr_err("No such device - did you specify"
-				       " an out-of-range profile CPU?\n");
-				rc = -err;
-				goto out;
-			} else if (err == EINVAL) {
-				if (!opts->exclude_guest_missing &&
-				    (attr->exclude_guest || attr->exclude_host)) {
-					pr_debug("Old kernel, cannot exclude "
-						 "guest or host samples.\n");
-					opts->exclude_guest_missing = true;
-					goto fallback_missing_features;
-				} else if (!opts->sample_id_all_missing) {
-					/*
-					 * Old kernel, no attr->sample_id_type_all field
-					 */
-					opts->sample_id_all_missing = true;
-					if (!opts->sample_time && !opts->raw_samples && !time_needed)
-						attr->sample_type &= ~PERF_SAMPLE_TIME;
-
-					goto retry_sample_id;
-				}
-			}
-
-			/*
-			 * If it's cycles then fall back to hrtimer
-			 * based cpu-clock-tick sw counter, which
-			 * is always available even if no PMU support.
-			 *
-			 * PPC returns ENXIO until 2.6.37 (behavior changed
-			 * with commit b0a873e).
-			 */
-			if ((err == ENOENT || err == ENXIO)
-					&& attr->type == PERF_TYPE_HARDWARE
-					&& attr->config == PERF_COUNT_HW_CPU_CYCLES) {
-
-				if (verbose)
-					ui__warning("The cycles event is not supported, "
-						    "trying to fall back to cpu-clock-ticks\n");
-				attr->type = PERF_TYPE_SOFTWARE;
-				attr->config = PERF_COUNT_SW_CPU_CLOCK;
-				if (pos->name) {
-					free(pos->name);
-					pos->name = NULL;
-				}
-				goto try_again;
-			}
-
-			if (err == ENOENT) {
-				ui__error("The %s event is not supported.\n",
-					  perf_evsel__name(pos));
-				rc = -err;
-				goto out;
-			}
-
-			printf("\n");
-			error("sys_perf_event_open() syscall returned with %d "
-			      "(%s) for event %s. /bin/dmesg may provide "
-			      "additional information.\n",
-			      err, strerror(err), perf_evsel__name(pos));
-
-#if defined(__i386__) || defined(__x86_64__)
-			if (attr->type == PERF_TYPE_HARDWARE &&
-			    err == EOPNOTSUPP) {
-				pr_err("No hardware sampling interrupt available."
-				       " No APIC? If so then you can boot the kernel"
-				       " with the \"lapic\" boot parameter to"
-				       " force-enable it.\n");
-				rc = -err;
-				goto out;
-			}
-#endif
-
-			pr_err("No CONFIG_PERF_EVENTS=y kernel support configured?\n");
-			rc = -err;
+	rc = perf_evlist__open_counters(evlist, opts);
+	if (rc != 0)
 			goto out;
-		}
-	}
 
 	if (perf_evlist__apply_filters(evlist)) {
 		error("failed to set filter with %d (%s)\n", errno,
-- 
1.7.10.1

[PATCH 6/9] perf stat: move user options to perf_record_opts

[David Ahern]

This is required for perf-stat to use perf_evlist__open_counters.
And move opts to a stack variable.

Signed-off-by: David Ahern <dsahern@gmail.com>
Cc: Arnaldo Carvalho de Melo <acme@ghostprotocols.net>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Peter Zijlstra <peterz@infradead.org>
---
 tools/perf/builtin-stat.c |  161 +++++++++++++++++++++++++++------------------
 1 file changed, 97 insertions(+), 64 deletions(-)

diff --git a/tools/perf/builtin-stat.c b/tools/perf/builtin-stat.c
index 93b9011..9727d217 100644
--- a/tools/perf/builtin-stat.c
+++ b/tools/perf/builtin-stat.c
@@ -66,12 +66,8 @@
 
 static struct perf_evlist	*evsel_list;
 
-static struct perf_target	target = {
-	.uid	= UINT_MAX,
-};
 
 static int			run_count			=  1;
-static bool			no_inherit			= false;
 static bool			scale				=  true;
 static bool			no_aggr				= false;
 static pid_t			child_pid			= -1;
@@ -81,7 +77,6 @@ static bool			big_num				=  true;
 static int			big_num_opt			=  -1;
 static const char		*csv_sep			= NULL;
 static bool			csv_output			= false;
-static bool			group				= false;
 static FILE			*output				= NULL;
 
 static volatile int done = 0;
@@ -102,14 +97,16 @@ static void perf_evsel__free_stat_priv(struct perf_evsel *evsel)
 	evsel->priv = NULL;
 }
 
-static inline struct cpu_map *perf_evsel__cpus(struct perf_evsel *evsel)
+static inline struct cpu_map *perf_evsel__cpus(struct perf_evsel *evsel,
+					       struct perf_target *target)
 {
-	return (evsel->cpus && !target.cpu_list) ? evsel->cpus : evsel_list->cpus;
+	return (evsel->cpus && !target->cpu_list) ? evsel->cpus : evsel_list->cpus;
 }
 
-static inline int perf_evsel__nr_cpus(struct perf_evsel *evsel)
+static inline int perf_evsel__nr_cpus(struct perf_evsel *evsel,
+				      struct perf_target *target)
 {
-	return perf_evsel__cpus(evsel)->nr;
+	return perf_evsel__cpus(evsel, target)->nr;
 }
 
 static struct stats runtime_nsecs_stats[MAX_NR_CPUS];
@@ -126,8 +123,10 @@ static struct stats runtime_dtlb_cache_stats[MAX_NR_CPUS];
 static struct stats walltime_nsecs_stats;
 
 static int create_perf_stat_counter(struct perf_evsel *evsel,
-				    struct perf_evsel *first)
+				    struct perf_evsel *first,
+				    struct perf_record_opts *opts)
 {
+	struct perf_target *target = &opts->target;
 	struct perf_event_attr *attr = &evsel->attr;
 	bool exclude_guest_missing = false;
 	int ret;
@@ -136,20 +135,22 @@ static int create_perf_stat_counter(struct perf_evsel *evsel,
 		attr->read_format = PERF_FORMAT_TOTAL_TIME_ENABLED |
 				    PERF_FORMAT_TOTAL_TIME_RUNNING;
 
-	attr->inherit = !no_inherit;
+	attr->inherit = !opts->no_inherit;
 
 retry:
 	if (exclude_guest_missing)
 		evsel->attr.exclude_guest = evsel->attr.exclude_host = 0;
 
-	if (perf_target__has_cpu(&target)) {
-		ret = perf_evsel__open_per_cpu(evsel, perf_evsel__cpus(evsel));
+	if (perf_target__has_cpu(target)) {
+		ret = perf_evsel__open_per_cpu(evsel,
+					       perf_evsel__cpus(evsel, target));
 		if (ret)
 			goto check_ret;
 		return 0;
 	}
 
-	if (!perf_target__has_task(&target) && (!group || evsel == first)) {
+	if (!perf_target__has_task(target) &&
+	    (!opts->group || evsel == first)) {
 		attr->disabled = 1;
 		attr->enable_on_exec = 1;
 	}
@@ -218,13 +219,15 @@ static void update_shadow_stats(struct perf_evsel *counter, u64 *count)
  * Read out the results of a single counter:
  * aggregate counts across CPUs in system-wide mode
  */
-static int read_counter_aggr(struct perf_evsel *counter)
+static int read_counter_aggr(struct perf_evsel *counter,
+			     struct perf_record_opts *opts)
 {
+	struct perf_target *target = &opts->target;
 	struct perf_stat *ps = counter->priv;
 	u64 *count = counter->counts->aggr.values;
 	int i;
 
-	if (__perf_evsel__read(counter, perf_evsel__nr_cpus(counter),
+	if (__perf_evsel__read(counter, perf_evsel__nr_cpus(counter, target),
 			       evsel_list->threads->nr, scale) < 0)
 		return -1;
 
@@ -248,12 +251,14 @@ static int read_counter_aggr(struct perf_evsel *counter)
  * Read out the results of a single counter:
  * do not aggregate counts across CPUs in system-wide mode
  */
-static int read_counter(struct perf_evsel *counter)
+static int read_counter(struct perf_evsel *counter,
+			struct perf_record_opts *opts)
 {
+	struct cpu_map *cmap = perf_evsel__cpus(counter, &opts->target);
 	u64 *count;
 	int cpu;
 
-	for (cpu = 0; cpu < perf_evsel__nr_cpus(counter); cpu++) {
+	for (cpu = 0; cpu < cmap->nr; cpu++) {
 		if (__perf_evsel__read_on_cpu(counter, cpu, 0, scale) < 0)
 			return -1;
 
@@ -265,10 +270,13 @@ static int read_counter(struct perf_evsel *counter)
 	return 0;
 }
 
-static int run_perf_stat(int argc __maybe_unused, const char **argv)
+static int run_perf_stat(int argc __maybe_unused,
+			 const char **argv,
+			 struct perf_record_opts *opts)
 {
 	unsigned long long t0, t1;
 	struct perf_evsel *counter, *first;
+	struct cpu_map *cmap;
 	int status = 0;
 	int child_ready_pipe[2], go_pipe[2];
 	const bool forks = (argc > 0);
@@ -312,7 +320,7 @@ static int run_perf_stat(int argc __maybe_unused, const char **argv)
 			exit(-1);
 		}
 
-		if (perf_target__none(&target))
+		if (perf_target__none(&opts->target))
 			evsel_list->threads->map[0] = child_pid;
 
 		/*
@@ -325,13 +333,13 @@ static int run_perf_stat(int argc __maybe_unused, const char **argv)
 		close(child_ready_pipe[0]);
 	}
 
-	if (group)
+	if (opts->group)
 		perf_evlist__set_leader(evsel_list);
 
 	first = perf_evlist__first(evsel_list);
 
 	list_for_each_entry(counter, &evsel_list->entries, node) {
-		if (create_perf_stat_counter(counter, first) < 0) {
+		if (create_perf_stat_counter(counter, first, opts) < 0) {
 			/*
 			 * PPC returns ENXIO for HW counters until 2.6.37
 			 * (behavior changed with commit b0a873e).
@@ -350,7 +358,7 @@ static int run_perf_stat(int argc __maybe_unused, const char **argv)
 				error("You may not have permission to collect %sstats.\n"
 				      "\t Consider tweaking"
 				      " /proc/sys/kernel/perf_event_paranoid or running as root.",
-				      target.system_wide ? "system-wide " : "");
+				      opts->target.system_wide ? "system-wide " : "");
 			} else {
 				error("open_counter returned with %d (%s). "
 				      "/bin/dmesg may provide additional information.\n",
@@ -391,13 +399,15 @@ static int run_perf_stat(int argc __maybe_unused, const char **argv)
 
 	if (no_aggr) {
 		list_for_each_entry(counter, &evsel_list->entries, node) {
-			read_counter(counter);
-			perf_evsel__close_fd(counter, perf_evsel__nr_cpus(counter), 1);
+			cmap = perf_evsel__cpus(counter, &opts->target);
+			read_counter(counter, opts);
+			perf_evsel__close_fd(counter, cmap->nr, 1);
 		}
 	} else {
 		list_for_each_entry(counter, &evsel_list->entries, node) {
-			read_counter_aggr(counter);
-			perf_evsel__close_fd(counter, perf_evsel__nr_cpus(counter),
+			cmap = perf_evsel__cpus(counter, &opts->target);
+			read_counter_aggr(counter, opts);
+			perf_evsel__close_fd(counter, cmap->nr,
 					     evsel_list->threads->nr);
 		}
 	}
@@ -426,16 +436,20 @@ static void print_noise(struct perf_evsel *evsel, double avg)
 	print_noise_pct(stddev_stats(&ps->res_stats[0]), avg);
 }
 
-static void nsec_printout(int cpu, struct perf_evsel *evsel, double avg)
+static void nsec_printout(int cpu, struct perf_evsel *evsel,
+			  double avg, struct perf_record_opts *opts)
 {
 	double msecs = avg / 1e6;
 	char cpustr[16] = { '\0', };
 	const char *fmt = csv_output ? "%s%.6f%s%s" : "%s%18.6f%s%-25s";
+	struct cpu_map *cmap;
 
-	if (no_aggr)
+	if (no_aggr) {
+		cmap = perf_evsel__cpus(evsel, &opts->target);
 		sprintf(cpustr, "CPU%*d%s",
 			csv_output ? 0 : -4,
-			perf_evsel__cpus(evsel)->map[cpu], csv_sep);
+			cmap->map[cpu], csv_sep);
+	}
 
 	fprintf(output, fmt, cpustr, msecs, csv_sep, perf_evsel__name(evsel));
 
@@ -631,11 +645,13 @@ static void print_ll_cache_misses(int cpu,
 	fprintf(output, " of all LL-cache hits   ");
 }
 
-static void abs_printout(int cpu, struct perf_evsel *evsel, double avg)
+static void abs_printout(int cpu, struct perf_evsel *evsel,
+			  double avg, struct perf_record_opts *opts)
 {
 	double total, ratio = 0.0;
 	char cpustr[16] = { '\0', };
 	const char *fmt;
+	struct cpu_map *cmap;
 
 	if (csv_output)
 		fmt = "%s%.0f%s%s";
@@ -644,11 +660,12 @@ static void abs_printout(int cpu, struct perf_evsel *evsel, double avg)
 	else
 		fmt = "%s%18.0f%s%-25s";
 
-	if (no_aggr)
+	if (no_aggr) {
+		cmap = perf_evsel__cpus(evsel, &opts->target);
 		sprintf(cpustr, "CPU%*d%s",
 			csv_output ? 0 : -4,
-			perf_evsel__cpus(evsel)->map[cpu], csv_sep);
-	else
+			cmap->map[cpu], csv_sep);
+	} else
 		cpu = 0;
 
 	fprintf(output, fmt, cpustr, avg, csv_sep, perf_evsel__name(evsel));
@@ -755,7 +772,8 @@ static void abs_printout(int cpu, struct perf_evsel *evsel, double avg)
  * Print out the results of a single counter:
  * aggregated counts in system-wide mode
  */
-static void print_counter_aggr(struct perf_evsel *counter)
+static void print_counter_aggr(struct perf_evsel *counter,
+			       struct perf_record_opts *opts)
 {
 	struct perf_stat *ps = counter->priv;
 	double avg = avg_stats(&ps->res_stats[0]);
@@ -777,9 +795,9 @@ static void print_counter_aggr(struct perf_evsel *counter)
 	}
 
 	if (nsec_counter(counter))
-		nsec_printout(-1, counter, avg);
+		nsec_printout(-1, counter, avg, opts);
 	else
-		abs_printout(-1, counter, avg);
+		abs_printout(-1, counter, avg, opts);
 
 	print_noise(counter, avg);
 
@@ -803,19 +821,21 @@ static void print_counter_aggr(struct perf_evsel *counter)
  * Print out the results of a single counter:
  * does not use aggregated count in system-wide
  */
-static void print_counter(struct perf_evsel *counter)
+static void print_counter(struct perf_evsel *counter,
+			  struct perf_record_opts *opts)
 {
+	struct cpu_map *cmap = perf_evsel__cpus(counter, &opts->target);
 	u64 ena, run, val;
 	int cpu;
 
-	for (cpu = 0; cpu < perf_evsel__nr_cpus(counter); cpu++) {
+	for (cpu = 0; cpu < cmap->nr; cpu++) {
 		val = counter->counts->cpu[cpu].val;
 		ena = counter->counts->cpu[cpu].ena;
 		run = counter->counts->cpu[cpu].run;
 		if (run == 0 || ena == 0) {
 			fprintf(output, "CPU%*d%s%*s%s%*s",
 				csv_output ? 0 : -4,
-				perf_evsel__cpus(counter)->map[cpu], csv_sep,
+				cmap->map[cpu], csv_sep,
 				csv_output ? 0 : 18,
 				counter->supported ? CNTR_NOT_COUNTED : CNTR_NOT_SUPPORTED,
 				csv_sep,
@@ -831,9 +851,9 @@ static void print_counter(struct perf_evsel *counter)
 		}
 
 		if (nsec_counter(counter))
-			nsec_printout(cpu, counter, val);
+			nsec_printout(cpu, counter, val, opts);
 		else
-			abs_printout(cpu, counter, val);
+			abs_printout(cpu, counter, val, opts);
 
 		if (!csv_output) {
 			print_noise(counter, 1.0);
@@ -846,9 +866,11 @@ static void print_counter(struct perf_evsel *counter)
 	}
 }
 
-static void print_stat(int argc, const char **argv)
+static void print_stat(int argc, const char **argv,
+		       struct perf_record_opts *opts)
 {
 	struct perf_evsel *counter;
+	struct perf_target *target = &opts->target;
 	int i;
 
 	fflush(stdout);
@@ -856,14 +878,14 @@ static void print_stat(int argc, const char **argv)
 	if (!csv_output) {
 		fprintf(output, "\n");
 		fprintf(output, " Performance counter stats for ");
-		if (!perf_target__has_task(&target)) {
+		if (!perf_target__has_task(target)) {
 			fprintf(output, "\'%s", argv[0]);
 			for (i = 1; i < argc; i++)
 				fprintf(output, " %s", argv[i]);
-		} else if (target.pid)
-			fprintf(output, "process id \'%s", target.pid);
+		} else if (target->pid)
+			fprintf(output, "process id \'%s", target->pid);
 		else
-			fprintf(output, "thread id \'%s", target.tid);
+			fprintf(output, "thread id \'%s", target->tid);
 
 		fprintf(output, "\'");
 		if (run_count > 1)
@@ -873,10 +895,10 @@ static void print_stat(int argc, const char **argv)
 
 	if (no_aggr) {
 		list_for_each_entry(counter, &evsel_list->entries, node)
-			print_counter(counter);
+			print_counter(counter, opts);
 	} else {
 		list_for_each_entry(counter, &evsel_list->entries, node)
-			print_counter_aggr(counter);
+			print_counter_aggr(counter, opts);
 	}
 
 	if (!csv_output) {
@@ -1073,21 +1095,31 @@ int cmd_stat(int argc, const char **argv, const char *prefix __maybe_unused)
 	     sync_run = false;
 	int output_fd = 0;
 	const char *output_name	= NULL;
+
+	struct perf_record_opts opts = {
+		.target = {
+			.uid	= UINT_MAX,
+		},
+		.no_inherit	= false,
+		.group		= false,
+	};
+	struct perf_target *target = &opts.target;
+
 	const struct option options[] = {
 	OPT_CALLBACK('e', "event", &evsel_list, "event",
 		     "event selector. use 'perf list' to list available events",
 		     parse_events_option),
 	OPT_CALLBACK(0, "filter", &evsel_list, "filter",
 		     "event filter", parse_filter),
-	OPT_BOOLEAN('i', "no-inherit", &no_inherit,
+	OPT_BOOLEAN('i', "no-inherit", &opts.no_inherit,
 		    "child tasks do not inherit counters"),
-	OPT_STRING('p', "pid", &target.pid, "pid",
+	OPT_STRING('p', "pid", &target->pid, "pid",
 		   "stat events on existing process id"),
-	OPT_STRING('t', "tid", &target.tid, "tid",
+	OPT_STRING('t', "tid", &target->tid, "tid",
 		   "stat events on existing thread id"),
-	OPT_BOOLEAN('a', "all-cpus", &target.system_wide,
+	OPT_BOOLEAN('a', "all-cpus", &target->system_wide,
 		    "system-wide collection from all CPUs"),
-	OPT_BOOLEAN('g', "group", &group,
+	OPT_BOOLEAN('g', "group", &opts.group,
 		    "put the counters into a counter group"),
 	OPT_BOOLEAN('c', "scale", &scale, "scale/normalize counters"),
 	OPT_INCR('v', "verbose", &verbose,
@@ -1103,7 +1135,7 @@ int cmd_stat(int argc, const char **argv, const char *prefix __maybe_unused)
 	OPT_CALLBACK_NOOPT('B', "big-num", NULL, NULL, 
 			   "print large numbers with thousands\' separators",
 			   stat__set_big_num),
-	OPT_STRING('C', "cpu", &target.cpu_list, "cpu",
+	OPT_STRING('C', "cpu", &target->cpu_list, "cpu",
 		    "list of cpus to monitor in system-wide"),
 	OPT_BOOLEAN('A', "no-aggr", &no_aggr, "disable CPU count aggregation"),
 	OPT_STRING('x', "field-separator", &csv_sep, "separator",
@@ -1187,13 +1219,13 @@ int cmd_stat(int argc, const char **argv, const char *prefix __maybe_unused)
 	} else if (big_num_opt == 0) /* User passed --no-big-num */
 		big_num = false;
 
-	if (!argc && !perf_target__has_task(&target))
+	if (!argc && !perf_target__has_task(target))
 		usage_with_options(stat_usage, options);
 	if (run_count <= 0)
 		usage_with_options(stat_usage, options);
 
 	/* no_aggr, cgroup are for system-wide only */
-	if ((no_aggr || nr_cgroups) && !perf_target__has_cpu(&target)) {
+	if ((no_aggr || nr_cgroups) && !perf_target__has_cpu(target)) {
 		fprintf(stderr, "both cgroup and no-aggregation "
 			"modes only available in system-wide mode\n");
 
@@ -1203,12 +1235,12 @@ int cmd_stat(int argc, const char **argv, const char *prefix __maybe_unused)
 	if (add_default_attributes())
 		goto out;
 
-	perf_target__validate(&target);
+	perf_target__validate(target);
 
-	if (perf_evlist__create_maps(evsel_list, &target) < 0) {
-		if (perf_target__has_task(&target))
+	if (perf_evlist__create_maps(evsel_list, target) < 0) {
+		if (perf_target__has_task(target))
 			pr_err("Problems finding threads of monitor\n");
-		if (perf_target__has_cpu(&target))
+		if (perf_target__has_cpu(target))
 			perror("failed to parse CPUs map");
 
 		usage_with_options(stat_usage, options);
@@ -1216,8 +1248,9 @@ int cmd_stat(int argc, const char **argv, const char *prefix __maybe_unused)
 	}
 
 	list_for_each_entry(pos, &evsel_list->entries, node) {
+		struct cpu_map *cmap = perf_evsel__cpus(pos, target);
 		if (perf_evsel__alloc_stat_priv(pos) < 0 ||
-		    perf_evsel__alloc_counts(pos, perf_evsel__nr_cpus(pos)) < 0)
+		    perf_evsel__alloc_counts(pos, cmap->nr) < 0)
 			goto out_free_fd;
 	}
 
@@ -1241,11 +1274,11 @@ int cmd_stat(int argc, const char **argv, const char *prefix __maybe_unused)
 		if (sync_run)
 			sync();
 
-		status = run_perf_stat(argc, argv);
+		status = run_perf_stat(argc, argv, &opts);
 	}
 
 	if (status != -1)
-		print_stat(argc, argv);
+		print_stat(argc, argv, &opts);
 out_free_fd:
 	list_for_each_entry(pos, &evsel_list->entries, node)
 		perf_evsel__free_stat_priv(pos);
-- 
1.7.10.1

[PATCH 7/9] perf evlist: add stat unique code to open_counters method

[David Ahern]

Mainly the addition is an argument to keep going for some open
failures.

Signed-off-by: David Ahern <dsahern@gmail.com>
Cc: Arnaldo Carvalho de Melo <acme@ghostprotocols.net>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Peter Zijlstra <peterz@infradead.org>
---
 tools/perf/builtin-record.c |    2 +-
 tools/perf/builtin-top.c    |    2 +-
 tools/perf/util/evlist.c    |   16 ++++++++++++++--
 tools/perf/util/evlist.h    |    3 ++-
 4 files changed, 18 insertions(+), 5 deletions(-)

diff --git a/tools/perf/builtin-record.c b/tools/perf/builtin-record.c
index b9dcc01..663ccc8 100644
--- a/tools/perf/builtin-record.c
+++ b/tools/perf/builtin-record.c
@@ -234,7 +234,7 @@ static int perf_record__open(struct perf_record *rec)
 	if (opts->group)
 		perf_evlist__set_leader(evlist);
 
-	rc = perf_evlist__open_counters(evlist, opts);
+	rc = perf_evlist__open_counters(evlist, opts, false);
 	if (rc != 0)
 			goto out;
 
diff --git a/tools/perf/builtin-top.c b/tools/perf/builtin-top.c
index 2ffc32e..2c3b3c7 100644
--- a/tools/perf/builtin-top.c
+++ b/tools/perf/builtin-top.c
@@ -921,7 +921,7 @@ static void perf_top__start_counters(struct perf_top *top)
 		attr->inherit = !top->opts.no_inherit;
 	}
 
-	if (perf_evlist__open_counters(evlist, &top->opts) != 0)
+	if (perf_evlist__open_counters(evlist, &top->opts, false) != 0)
 		goto out_err;
 
 	if (perf_evlist__mmap(evlist, top->opts.mmap_pages, false) < 0) {
diff --git a/tools/perf/util/evlist.c b/tools/perf/util/evlist.c
index bce2f58..fa0daac 100644
--- a/tools/perf/util/evlist.c
+++ b/tools/perf/util/evlist.c
@@ -840,7 +840,8 @@ size_t perf_evlist__fprintf(struct perf_evlist *evlist, FILE *fp)
 }
 
 int perf_evlist__open_counters(struct perf_evlist *evlist,
-			       struct perf_record_opts *opts)
+			       struct perf_record_opts *opts,
+			       bool continue_on_fail)
 {
 	struct perf_evsel *pos;
 	int rc = 0;
@@ -872,6 +873,16 @@ try_again:
 		if (perf_evsel__open(pos, evlist->cpus, evlist->threads) < 0) {
 			int err = errno;
 
+			if (continue_on_fail &&
+			    (err == EINVAL || err == ENOSYS || err == ENXIO ||
+			     err == ENOENT || err == EOPNOTSUPP)) {
+				if (verbose)
+					ui__warning("%s event is not supported by the kernel.\n",
+						    perf_evsel__name(pos));
+				pos->supported = false;
+				continue;
+			}
+
 			if (err == EPERM || err == EACCES) {
 				ui__error_paranoid();
 				rc = -err;
@@ -958,7 +969,8 @@ try_again:
 			pr_err("No CONFIG_PERF_EVENTS=y kernel support configured?\n");
 			rc = -err;
 			goto out;
-		}
+		} else
+			pos->supported = true;
 	}
 out:
 	return rc;
diff --git a/tools/perf/util/evlist.h b/tools/perf/util/evlist.h
index 270e546..0747b6f 100644
--- a/tools/perf/util/evlist.h
+++ b/tools/perf/util/evlist.h
@@ -137,5 +137,6 @@ static inline struct perf_evsel *perf_evlist__last(struct perf_evlist *evlist)
 size_t perf_evlist__fprintf(struct perf_evlist *evlist, FILE *fp);
 
 int perf_evlist__open_counters(struct perf_evlist *evlist,
-			       struct perf_record_opts *opts);
+			       struct perf_record_opts *opts,
+			       bool continue_on_fail);
 #endif /* __PERF_EVLIST_H */
-- 
1.7.10.1

[PATCH 8/9] perf stat: move to perf_evlist__open_counters

[David Ahern]

Removes a lot of duplicated code moving to the common
open method.

Signed-off-by: David Ahern <dsahern@gmail.com>
Cc: Arnaldo Carvalho de Melo <acme@ghostprotocols.net>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Peter Zijlstra <peterz@infradead.org>
---
 tools/perf/builtin-stat.c |  103 ++++++++++-----------------------------------
 1 file changed, 22 insertions(+), 81 deletions(-)

diff --git a/tools/perf/builtin-stat.c b/tools/perf/builtin-stat.c
index 9727d217..affbada 100644
--- a/tools/perf/builtin-stat.c
+++ b/tools/perf/builtin-stat.c
@@ -122,55 +122,6 @@ static struct stats runtime_itlb_cache_stats[MAX_NR_CPUS];
 static struct stats runtime_dtlb_cache_stats[MAX_NR_CPUS];
 static struct stats walltime_nsecs_stats;
 
-static int create_perf_stat_counter(struct perf_evsel *evsel,
-				    struct perf_evsel *first,
-				    struct perf_record_opts *opts)
-{
-	struct perf_target *target = &opts->target;
-	struct perf_event_attr *attr = &evsel->attr;
-	bool exclude_guest_missing = false;
-	int ret;
-
-	if (scale)
-		attr->read_format = PERF_FORMAT_TOTAL_TIME_ENABLED |
-				    PERF_FORMAT_TOTAL_TIME_RUNNING;
-
-	attr->inherit = !opts->no_inherit;
-
-retry:
-	if (exclude_guest_missing)
-		evsel->attr.exclude_guest = evsel->attr.exclude_host = 0;
-
-	if (perf_target__has_cpu(target)) {
-		ret = perf_evsel__open_per_cpu(evsel,
-					       perf_evsel__cpus(evsel, target));
-		if (ret)
-			goto check_ret;
-		return 0;
-	}
-
-	if (!perf_target__has_task(target) &&
-	    (!opts->group || evsel == first)) {
-		attr->disabled = 1;
-		attr->enable_on_exec = 1;
-	}
-
-	ret = perf_evsel__open_per_thread(evsel, evsel_list->threads);
-	if (!ret)
-		return 0;
-	/* fall through */
-check_ret:
-	if (ret && errno == EINVAL) {
-		if (!exclude_guest_missing &&
-		    (evsel->attr.exclude_guest || evsel->attr.exclude_host)) {
-			pr_debug("Old kernel, cannot exclude "
-				 "guest or host samples.\n");
-			exclude_guest_missing = true;
-			goto retry;
-		}
-	}
-	return ret;
-}
 
 /*
  * Does the counter have nsecs as a unit?
@@ -277,6 +228,7 @@ static int run_perf_stat(int argc __maybe_unused,
 	unsigned long long t0, t1;
 	struct perf_evsel *counter, *first;
 	struct cpu_map *cmap;
+	struct perf_target *target = &opts->target;
 	int status = 0;
 	int child_ready_pipe[2], go_pipe[2];
 	const bool forks = (argc > 0);
@@ -320,7 +272,7 @@ static int run_perf_stat(int argc __maybe_unused,
 			exit(-1);
 		}
 
-		if (perf_target__none(&opts->target))
+		if (perf_target__none(target))
 			evsel_list->threads->map[0] = child_pid;
 
 		/*
@@ -339,38 +291,27 @@ static int run_perf_stat(int argc __maybe_unused,
 	first = perf_evlist__first(evsel_list);
 
 	list_for_each_entry(counter, &evsel_list->entries, node) {
-		if (create_perf_stat_counter(counter, first, opts) < 0) {
-			/*
-			 * PPC returns ENXIO for HW counters until 2.6.37
-			 * (behavior changed with commit b0a873e).
-			 */
-			if (errno == EINVAL || errno == ENOSYS ||
-			    errno == ENOENT || errno == EOPNOTSUPP ||
-			    errno == ENXIO) {
-				if (verbose)
-					ui__warning("%s event is not supported by the kernel.\n",
-						    perf_evsel__name(counter));
-				counter->supported = false;
-				continue;
-			}
-
-			if (errno == EPERM || errno == EACCES) {
-				error("You may not have permission to collect %sstats.\n"
-				      "\t Consider tweaking"
-				      " /proc/sys/kernel/perf_event_paranoid or running as root.",
-				      opts->target.system_wide ? "system-wide " : "");
-			} else {
-				error("open_counter returned with %d (%s). "
-				      "/bin/dmesg may provide additional information.\n",
-				       errno, strerror(errno));
-			}
-			if (child_pid != -1)
-				kill(child_pid, SIGTERM);
-
-			pr_err("Not all events could be opened.\n");
-			return -1;
+		struct perf_event_attr *attr = &counter->attr;
+
+		if (scale)
+			attr->read_format = PERF_FORMAT_TOTAL_TIME_ENABLED |
+					    PERF_FORMAT_TOTAL_TIME_RUNNING;
+
+		attr->inherit = !opts->no_inherit;
+
+		if (perf_target__none(target) &&
+		    (!opts->group || counter == first)) {
+			attr->disabled = 1;
+			attr->enable_on_exec = 1;
 		}
-		counter->supported = true;
+	}
+
+	if (perf_evlist__open_counters(evsel_list, opts, true) != 0) {
+		if (child_pid != -1)
+			kill(child_pid, SIGTERM);
+
+		pr_err("Not all events could be opened.\n");
+		return -1;
 	}
 
 	if (perf_evlist__apply_filters(evsel_list)) {
-- 
1.7.10.1

[PATCH 9/9] perf evsel: remove perf_evsel__open_per_cpu

[David Ahern]

No longer needed with perf-stat converted to perf_evlist__open_counters.

Signed-off-by: David Ahern <dsahern@gmail.com>
Cc: Arnaldo Carvalho de Melo <acme@ghostprotocols.net>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Peter Zijlstra <peterz@infradead.org>
---
 tools/perf/util/evsel.c |    6 ------
 tools/perf/util/evsel.h |    2 --
 2 files changed, 8 deletions(-)

diff --git a/tools/perf/util/evsel.c b/tools/perf/util/evsel.c
index ffdd94e..ab3d1c8 100644
--- a/tools/perf/util/evsel.c
+++ b/tools/perf/util/evsel.c
@@ -774,12 +774,6 @@ int perf_evsel__open(struct perf_evsel *evsel, struct cpu_map *cpus,
 	return __perf_evsel__open(evsel, cpus, threads);
 }
 
-int perf_evsel__open_per_cpu(struct perf_evsel *evsel,
-			     struct cpu_map *cpus)
-{
-	return __perf_evsel__open(evsel, cpus, &empty_thread_map.map);
-}
-
 int perf_evsel__open_per_thread(struct perf_evsel *evsel,
 				struct thread_map *threads)
 {
diff --git a/tools/perf/util/evsel.h b/tools/perf/util/evsel.h
index 3ead0d5..bf32de4 100644
--- a/tools/perf/util/evsel.h
+++ b/tools/perf/util/evsel.h
@@ -121,8 +121,6 @@ void perf_evsel__close_fd(struct perf_evsel *evsel, int ncpus, int nthreads);
 int perf_evsel__set_filter(struct perf_evsel *evsel, int ncpus, int nthreads,
 			   const char *filter);
 
-int perf_evsel__open_per_cpu(struct perf_evsel *evsel,
-			     struct cpu_map *cpus);
 int perf_evsel__open_per_thread(struct perf_evsel *evsel,
 				struct thread_map *threads);
 int perf_evsel__open(struct perf_evsel *evsel, struct cpu_map *cpus,
-- 
1.7.10.1

[PATCH 3/9] perf evlist: introduce open counters method

[David Ahern]

Superset of the open counters code in perf-top and perf-record -
combining retry handling and error handling. Should be functionally
equivalent.

Signed-off-by: David Ahern <dsahern@gmail.com>
Cc: Arnaldo Carvalho de Melo <acme@ghostprotocols.net>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Peter Zijlstra <peterz@infradead.org>
---
 tools/perf/util/evlist.c |  131 +++++++++++++++++++++++++++++++++++++++++++++-
 tools/perf/util/evlist.h |    3 ++
 2 files changed, 133 insertions(+), 1 deletion(-)

diff --git a/tools/perf/util/evlist.c b/tools/perf/util/evlist.c
index a41dc4a..b24ebc1 100644
--- a/tools/perf/util/evlist.c
+++ b/tools/perf/util/evlist.c
@@ -15,7 +15,7 @@
 #include "evlist.h"
 #include "evsel.h"
 #include <unistd.h>
-
+#include "debug.h"
 #include "parse-events.h"
 
 #include <sys/mman.h>
@@ -838,3 +838,132 @@ size_t perf_evlist__fprintf(struct perf_evlist *evlist, FILE *fp)
 
 	return printed + fprintf(fp, "\n");;
 }
+
+int perf_evlist__open_counters(struct perf_evlist *evlist,
+			       struct perf_record_opts *opts)
+{
+	struct perf_evsel *pos;
+	int rc = 0;
+
+	list_for_each_entry(pos, &evlist->entries, node) {
+		struct perf_event_attr *attr = &pos->attr;
+
+		/*
+		 * Carried over from perf-record:
+		 * Check if parse_single_tracepoint_event has already asked for
+		 * PERF_SAMPLE_TIME.
+		 *
+		 * XXX this is kludgy but short term fix for problems introduced by
+		 * eac23d1c that broke 'perf script' by having different sample_types
+		 * when using multiple tracepoint events when we use a perf binary
+		 * that tries to use sample_id_all on an older kernel.
+		 *
+		 * We need to move counter creation to perf_session, support
+		 * different sample_types, etc.
+		 */
+		bool time_needed = attr->sample_type & PERF_SAMPLE_TIME;
+
+fallback_missing_features:
+		if (opts->exclude_guest_missing)
+			attr->exclude_guest = attr->exclude_host = 0;
+retry_sample_id:
+		attr->sample_id_all = opts->sample_id_all_missing ? 0 : 1;
+try_again:
+		if (perf_evsel__open(pos, evlist->cpus, evlist->threads) < 0) {
+			int err = errno;
+
+			if (err == EPERM || err == EACCES) {
+				ui__error_paranoid();
+				rc = -err;
+				goto out;
+			} else if (err ==  ENODEV && opts->target.cpu_list) {
+				pr_err("No such device - did you specify"
+				       " an out-of-range profile CPU?\n");
+				rc = -err;
+				goto out;
+			} else if (err == EINVAL) {
+				if (!opts->exclude_guest_missing &&
+				    (attr->exclude_guest || attr->exclude_host)) {
+					pr_debug("Old kernel, cannot exclude "
+						 "guest or host samples.\n");
+					opts->exclude_guest_missing = true;
+					goto fallback_missing_features;
+				} else if (!opts->sample_id_all_missing) {
+					/*
+					 * Old kernel, no attr->sample_id_type_all field
+					 */
+					opts->sample_id_all_missing = true;
+					if (!opts->sample_time &&
+					    !opts->raw_samples &&
+					    !time_needed)
+						attr->sample_type &= ~PERF_SAMPLE_TIME;
+					goto retry_sample_id;
+				}
+			}
+
+			/*
+			 * If it's cycles then fall back to hrtimer
+			 * based cpu-clock-tick sw counter, which
+			 * is always available even if no PMU support:
+			 *
+			 * PPC returns ENXIO until 2.6.37 (behavior changed
+			 * with commit b0a873e).
+			 */
+			if ((err == ENOENT || err == ENXIO) &&
+			    (attr->type == PERF_TYPE_HARDWARE) &&
+			    (attr->config == PERF_COUNT_HW_CPU_CYCLES)) {
+
+				if (verbose)
+					ui__warning("Cycles event not supported,\n"
+						    "trying to fall back to cpu-clock-ticks\n");
+
+				attr->type = PERF_TYPE_SOFTWARE;
+				attr->config = PERF_COUNT_SW_CPU_CLOCK;
+				if (pos->name) {
+					free(pos->name);
+					pos->name = NULL;
+				}
+				goto try_again;
+			}
+
+			if (err == ENOENT) {
+				ui__error("The %s event is not supported.\n",
+					  perf_evsel__name(pos));
+				rc = -err;
+				goto out;
+			} else if (err == EMFILE) {
+				ui__error("Too many events are opened.\n"
+					  "Try again after reducing the number of events\n");
+				rc = -err;
+				goto out;
+			} else if ((err == EOPNOTSUPP) && (attr->precise_ip)) {
+				ui__error("\'precise\' request may not be supported. "
+					  "Try removing 'p' modifier\n");
+				goto out;
+			}
+
+			ui__error("sys_perf_event_open() syscall returned with "
+				  "%d (%s) for event %s.  /bin/dmesg may provide"
+				  "additional information.\n",
+				  err, strerror(err), perf_evsel__name(pos));
+
+#if defined(__i386__) || defined(__x86_64__)
+			if ((attr->type == PERF_TYPE_HARDWARE) &&
+			    (err == EOPNOTSUPP)) {
+				pr_err("No hardware sampling interrupt available."
+				       " No APIC? If so then you can boot the kernel"
+				       " with the \"lapic\" boot parameter to"
+				       " force-enable it.\n");
+				rc = -err;
+				goto out;
+			}
+#endif
+
+			pr_err("No CONFIG_PERF_EVENTS=y kernel support configured?\n");
+			rc = -err;
+			goto out;
+		}
+	}
+out:
+	return rc;
+}
diff --git a/tools/perf/util/evlist.h b/tools/perf/util/evlist.h
index 56003f7..270e546 100644
--- a/tools/perf/util/evlist.h
+++ b/tools/perf/util/evlist.h
@@ -135,4 +135,7 @@ static inline struct perf_evsel *perf_evlist__last(struct perf_evlist *evlist)
 }
 
 size_t perf_evlist__fprintf(struct perf_evlist *evlist, FILE *fp);
+
+int perf_evlist__open_counters(struct perf_evlist *evlist,
+			       struct perf_record_opts *opts);
 #endif /* __PERF_EVLIST_H */
-- 
1.7.10.1