Quelle  builtin-report.c   Sprache: C

// SPDX-License-Identifier: GPL-2.0
/*
 * builtin-report.c
 *
 * Builtin report command: Analyze the perf.data input file,
 * look up and read DSOs and symbol information and display
 * a histogram of results, along various sorting keys.
 */
#include "builtin.h"

#include "util/config.h"

#include "util/annotate.h"
#include "util/color.h"
#include "util/dso.h"
#include <linux/list.h>
#include <linux/rbtree.h>
#include <linux/err.h>
#include <linux/zalloc.h>
#include "util/map.h"
#include "util/symbol.h"
#include "util/map_symbol.h"
#include "util/mem-events.h"
#include "util/branch.h"
#include "util/callchain.h"
#include "util/values.h"

#include "perf.h"
#include "util/debug.h"
#include "util/evlist.h"
#include "util/evsel.h"
#include "util/evswitch.h"
#include "util/header.h"
#include "util/mem-info.h"
#include "util/session.h"
#include "util/srcline.h"
#include "util/tool.h"

#include <subcmd/parse-options.h>
#include <subcmd/exec-cmd.h>
#include "util/parse-events.h"

#include "util/thread.h"
#include "util/sort.h"
#include "util/hist.h"
#include "util/data.h"
#include "arch/common.h"
#include "util/time-utils.h"
#include "util/auxtrace.h"
#include "util/units.h"
#include "util/util.h" // perf_tip()
#include "ui/ui.h"
#include "ui/progress.h"
#include "util/block-info.h"

#include <dlfcn.h>
#include <errno.h>
#include <inttypes.h>
#include <regex.h>
#include <linux/ctype.h>
#include <signal.h>
#include <linux/bitmap.h>
#include <linux/list_sort.h>
#include <linux/string.h>
#include <linux/stringify.h>
#include <linux/time64.h>
#include <sys/types.h>
#include <sys/stat.h>
#include <unistd.h>
#include <linux/mman.h>

#ifdef HAVE_LIBTRACEEVENT
#include <event-parse.h>
#endif

struct report {
 struct perf_tool tool;
 struct perf_session *session;
 struct evswitch  evswitch;
#ifdef HAVE_SLANG_SUPPORT
 bool   use_tui;
#endif
#ifdef HAVE_GTK2_SUPPORT
 bool   use_gtk;
#endif
 bool   use_stdio;
 bool   show_full_info;
 bool   show_threads;
 bool   inverted_callchain;
 bool   mem_mode;
 bool   stats_mode;
 bool   tasks_mode;
 bool   mmaps_mode;
 bool   header;
 bool   header_only;
 bool   nonany_branch_mode;
 bool   group_set;
 bool   stitch_lbr;
 bool   disable_order;
 bool   skip_empty;
 bool   data_type;
 int   max_stack;
 struct perf_read_values show_threads_values;
 const char  *pretty_printing_style;
 const char  *cpu_list;
 const char  *symbol_filter_str;
 const char  *time_str;
 struct perf_time_interval *ptime_range;
 int   range_size;
 int   range_num;
 float   min_percent;
 u64   nr_entries;
 u64   queue_size;
 u64   total_cycles;
 u64   total_samples;
 u64   singlethreaded_samples;
 int   socket_filter;
 DECLARE_BITMAP(cpu_bitmap, MAX_NR_CPUS);
 struct branch_type_stat brtype_stat;
 bool   symbol_ipc;
 bool   total_cycles_mode;
 struct block_report *block_reports;
 int   nr_block_reports;
};

static int report__config(const char *var, const char *value, void *cb)
{
 struct report *rep = cb;

 if (!strcmp(var, "report.group")) {
  symbol_conf.event_group = perf_config_bool(var, value);
  return 0;
 }
 if (!strcmp(var, "report.percent-limit")) {
  double pcnt = strtof(value, NULL);

  rep->min_percent = pcnt;
  callchain_param.min_percent = pcnt;
  return 0;
 }
 if (!strcmp(var, "report.children")) {
  symbol_conf.cumulate_callchain = perf_config_bool(var, value);
  return 0;
 }
 if (!strcmp(var, "report.queue-size"))
  return perf_config_u64(&rep->queue_size, var, value);

 if (!strcmp(var, "report.sort_order")) {
  default_sort_order = strdup(value);
  if (!default_sort_order) {
   pr_err("Not enough memory for report.sort_order\n");
   return -1;
  }
  return 0;
 }

 if (!strcmp(var, "report.skip-empty")) {
  rep->skip_empty = perf_config_bool(var, value);
  return 0;
 }

 pr_debug("%s variable unknown, ignoring...", var);
 return 0;
}

static int hist_iter__report_callback(struct hist_entry_iter *iter,
          struct addr_location *al, bool single,
          void *arg)
{
 int err = 0;
 struct report *rep = arg;
 struct hist_entry *he = iter->he;
 struct evsel *evsel = iter->evsel;
 struct perf_sample *sample = iter->sample;
 struct mem_info *mi;
 struct branch_info *bi;

 if (!ui__has_annotation() && !rep->symbol_ipc)
  return 0;

 if (sort__mode == SORT_MODE__BRANCH) {
  bi = he->branch_info;
  err = addr_map_symbol__inc_samples(&bi->from, sample, evsel);
  if (err)
   goto out;

  err = addr_map_symbol__inc_samples(&bi->to, sample, evsel);

 } else if (rep->mem_mode) {
  mi = he->mem_info;
  err = addr_map_symbol__inc_samples(mem_info__daddr(mi), sample, evsel);
  if (err)
   goto out;

  err = hist_entry__inc_addr_samples(he, sample, evsel, al->addr);

 } else if (symbol_conf.cumulate_callchain) {
  if (single)
   err = hist_entry__inc_addr_samples(he, sample, evsel, al->addr);
 } else {
  err = hist_entry__inc_addr_samples(he, sample, evsel, al->addr);
 }

out:
 return err;
}

static int hist_iter__branch_callback(struct hist_entry_iter *iter,
          struct addr_location *al __maybe_unused,
          bool single __maybe_unused,
          void *arg)
{
 struct hist_entry *he = iter->he;
 struct report *rep = arg;
 struct branch_info *bi = he->branch_info;
 struct perf_sample *sample = iter->sample;
 struct evsel *evsel = iter->evsel;
 int err;

 branch_type_count(&rep->brtype_stat, &bi->flags,
     bi->from.addr, bi->to.addr);

 if (!ui__has_annotation() && !rep->symbol_ipc)
  return 0;

 err = addr_map_symbol__inc_samples(&bi->from, sample, evsel);
 if (err)
  goto out;

 err = addr_map_symbol__inc_samples(&bi->to, sample, evsel);

out:
 return err;
}

static void setup_forced_leader(struct report *report,
    struct evlist *evlist)
{
 if (report->group_set)
  evlist__force_leader(evlist);
}

static int process_feature_event(struct perf_session *session,
     union perf_event *event)
{
 struct report *rep = container_of(session->tool, struct report, tool);

 if (event->feat.feat_id < HEADER_LAST_FEATURE)
  return perf_event__process_feature(session, event);

 if (event->feat.feat_id != HEADER_LAST_FEATURE) {
  pr_err("failed: wrong feature ID: %" PRI_lu64 "\n",
         event->feat.feat_id);
  return -1;
 } else if (rep->header_only) {
  session_done = 1;
 }

 /*
 * (feat_id = HEADER_LAST_FEATURE) is the end marker which
 * means all features are received, now we can force the
 * group if needed.
 */
 setup_forced_leader(rep, session->evlist);
 return 0;
}

static int process_sample_event(const struct perf_tool *tool,
    union perf_event *event,
    struct perf_sample *sample,
    struct evsel *evsel,
    struct machine *machine)
{
 struct report *rep = container_of(tool, struct report, tool);
 struct addr_location al;
 struct hist_entry_iter iter = {
  .evsel    = evsel,
  .sample   = sample,
  .hide_unresolved  = symbol_conf.hide_unresolved,
  .add_entry_cb   = hist_iter__report_callback,
 };
 int ret = 0;

 if (perf_time__ranges_skip_sample(rep->ptime_range, rep->range_num,
       sample->time)) {
  return 0;
 }

 if (evswitch__discard(&rep->evswitch, evsel))
  return 0;

 addr_location__init(&al);
 if (machine__resolve(machine, &al, sample) < 0) {
  pr_debug("problem processing %d event, skipping it.\n",
    event->header.type);
  ret = -1;
  goto out_put;
 }

 if (rep->stitch_lbr)
  thread__set_lbr_stitch_enable(al.thread, true);

 if (symbol_conf.hide_unresolved && al.sym == NULL)
  goto out_put;

 if (rep->cpu_list && !test_bit(sample->cpu, rep->cpu_bitmap))
  goto out_put;

 if (sort__mode == SORT_MODE__BRANCH) {
  /*
 * A non-synthesized event might not have a branch stack if
 * branch stacks have been synthesized (using itrace options).
 */
  if (!sample->branch_stack)
   goto out_put;

  iter.add_entry_cb = hist_iter__branch_callback;
  iter.ops = &hist_iter_branch;
 } else if (rep->mem_mode) {
  iter.ops = &hist_iter_mem;
 } else if (symbol_conf.cumulate_callchain) {
  iter.ops = &hist_iter_cumulative;
 } else {
  iter.ops = &hist_iter_normal;
 }

 if (al.map != NULL)
  dso__set_hit(map__dso(al.map));

 if (ui__has_annotation() || rep->symbol_ipc || rep->total_cycles_mode) {
  hist__account_cycles(sample->branch_stack, &al, sample,
         rep->nonany_branch_mode,
         &rep->total_cycles, evsel);
 }

 rep->total_samples++;
 if (al.parallelism == 1)
  rep->singlethreaded_samples++;

 ret = hist_entry_iter__add(&iter, &al, rep->max_stack, rep);
 if (ret < 0)
  pr_debug("problem adding hist entry, skipping event\n");
out_put:
 addr_location__exit(&al);
 return ret;
}

static int process_read_event(const struct perf_tool *tool,
         union perf_event *event,
         struct perf_sample *sample __maybe_unused,
         struct evsel *evsel,
         struct machine *machine __maybe_unused)
{
 struct report *rep = container_of(tool, struct report, tool);

 if (rep->show_threads) {
  int err = perf_read_values_add_value(&rep->show_threads_values,
        event->read.pid, event->read.tid,
        evsel,
        event->read.value);

  if (err)
   return err;
 }

 return 0;
}

/* For pipe mode, sample_type is not currently set */
static int report__setup_sample_type(struct report *rep)
{
 struct perf_session *session = rep->session;
 u64 sample_type = evlist__combined_sample_type(session->evlist);
 bool is_pipe = perf_data__is_pipe(session->data);
 struct evsel *evsel;

 if (session->itrace_synth_opts->callchain ||
     session->itrace_synth_opts->add_callchain ||
     (!is_pipe &&
      perf_header__has_feat(&session->header, HEADER_AUXTRACE) &&
      !session->itrace_synth_opts->set))
  sample_type |= PERF_SAMPLE_CALLCHAIN;

 if (session->itrace_synth_opts->last_branch ||
     session->itrace_synth_opts->add_last_branch)
  sample_type |= PERF_SAMPLE_BRANCH_STACK;

 if (!is_pipe && !(sample_type & PERF_SAMPLE_CALLCHAIN)) {
  if (perf_hpp_list.parent) {
   ui__error("Selected --sort parent, but no "
        "callchain data. Did you call "
        "'perf record' without -g?\n");
   return -EINVAL;
  }
  if (symbol_conf.use_callchain &&
   !symbol_conf.show_branchflag_count) {
   ui__error("Selected -g or --branch-history.\n"
      "But no callchain or branch data.\n"
      "Did you call 'perf record' without -g or -b?\n");
   return -1;
  }
 } else if (!callchain_param.enabled &&
     callchain_param.mode != CHAIN_NONE &&
     !symbol_conf.use_callchain) {
   symbol_conf.use_callchain = true;
   if (callchain_register_param(&callchain_param) < 0) {
    ui__error("Can't register callchain params.\n");
    return -EINVAL;
   }
 }

 if (symbol_conf.cumulate_callchain) {
  /* Silently ignore if callchain is missing */
  if (!(sample_type & PERF_SAMPLE_CALLCHAIN)) {
   symbol_conf.cumulate_callchain = false;
   perf_hpp__cancel_cumulate(session->evlist);
  }
 }

 if (sort__mode == SORT_MODE__BRANCH) {
  if (!is_pipe &&
      !(sample_type & PERF_SAMPLE_BRANCH_STACK)) {
   ui__error("Selected -b but no branch data. "
      "Did you call perf record without -b?\n");
   return -1;
  }
 }

 if (sort__mode == SORT_MODE__MEMORY) {
  /*
 * FIXUP: prior to kernel 5.18, Arm SPE missed to set
 * PERF_SAMPLE_DATA_SRC bit in sample type.  For backward
 * compatibility, set the bit if it's an old perf data file.
 */
  evlist__for_each_entry(session->evlist, evsel) {
   if (strstr(evsel__name(evsel), "arm_spe") &&
    !(sample_type & PERF_SAMPLE_DATA_SRC)) {
    evsel->core.attr.sample_type |= PERF_SAMPLE_DATA_SRC;
    sample_type |= PERF_SAMPLE_DATA_SRC;
   }
  }

  if (!is_pipe && !(sample_type & PERF_SAMPLE_DATA_SRC)) {
   ui__error("Selected --mem-mode but no mem data. "
      "Did you call perf record without -d?\n");
   return -1;
  }
 }

 callchain_param_setup(sample_type, perf_env__arch(perf_session__env(rep->session)));

 if (rep->stitch_lbr && (callchain_param.record_mode != CALLCHAIN_LBR)) {
  ui__warning("Can't find LBR callchain. Switch off --stitch-lbr.\n"
       "Please apply --call-graph lbr when recording.\n");
  rep->stitch_lbr = false;
 }

 /* ??? handle more cases than just ANY? */
 if (!(evlist__combined_branch_type(session->evlist) & PERF_SAMPLE_BRANCH_ANY))
  rep->nonany_branch_mode = true;

#if !defined(HAVE_LIBUNWIND_SUPPORT) && !defined(HAVE_LIBDW_SUPPORT)
 if (dwarf_callchain_users) {
  ui__warning("Please install libunwind or libdw "
       "development packages during the perf build.\n");
 }
#endif

 return 0;
}

static void sig_handler(int sig __maybe_unused)
{
 session_done = 1;
}

static size_t hists__fprintf_nr_sample_events(struct hists *hists, struct report *rep,
           const char *evname, FILE *fp)
{
 size_t ret;
 char unit;
 unsigned long nr_samples = hists->stats.nr_samples;
 u64 nr_events = hists->stats.total_period;
 struct evsel *evsel = hists_to_evsel(hists);
 char buf[512];
 size_t size = sizeof(buf);
 int socked_id = hists->socket_filter;

 if (quiet)
  return 0;

 if (symbol_conf.filter_relative) {
  nr_samples = hists->stats.nr_non_filtered_samples;
  nr_events = hists->stats.total_non_filtered_period;
 }

 if (evsel__is_group_event(evsel)) {
  struct evsel *pos;

  evsel__group_desc(evsel, buf, size);
  evname = buf;

  for_each_group_member(pos, evsel) {
   const struct hists *pos_hists = evsel__hists(pos);

   if (symbol_conf.filter_relative) {
    nr_samples += pos_hists->stats.nr_non_filtered_samples;
    nr_events += pos_hists->stats.total_non_filtered_period;
   } else {
    nr_samples += pos_hists->stats.nr_samples;
    nr_events += pos_hists->stats.total_period;
   }
  }
 }

 nr_samples = convert_unit(nr_samples, &unit);
 ret = fprintf(fp, "# Samples: %lu%c", nr_samples, unit);
 if (evname != NULL) {
  ret += fprintf(fp, " of event%s '%s'",
          evsel->core.nr_members > 1 ? "s" : "", evname);
 }

 if (rep->time_str)
  ret += fprintf(fp, " (time slices: %s)", rep->time_str);

 if (symbol_conf.show_ref_callgraph && evname && strstr(evname, "call-graph=no")) {
  ret += fprintf(fp, ", show reference callgraph");
 }

 if (rep->mem_mode) {
  ret += fprintf(fp, "\n# Total weight : %" PRIu64, nr_events);
  if (sort_order || !field_order) {
   ret += fprintf(fp, "\n# Sort order : %s",
           sort_order ? : default_mem_sort_order);
  }
 } else
  ret += fprintf(fp, "\n# Event count (approx.): %" PRIu64, nr_events);

 if (socked_id > -1)
  ret += fprintf(fp, "\n# Processor Socket: %d", socked_id);

 return ret + fprintf(fp, "\n#\n");
}

static int evlist__tui_block_hists_browse(struct evlist *evlist, struct report *rep)
{
 struct evsel *pos;
 int i = 0, ret;

 evlist__for_each_entry(evlist, pos) {
  ret = report__browse_block_hists(&rep->block_reports[i++].hist,
       rep->min_percent, pos,
       perf_session__env(rep->session));
  if (ret != 0)
   return ret;
 }

 return 0;
}

static int evlist__tty_browse_hists(struct evlist *evlist, struct report *rep, const char *help)
{
 struct evsel *pos;
 int i = 0;

 if (!quiet) {
  fprintf(stdout, "#\n# Total Lost Samples: %" PRIu64 "\n#\n",
   evlist->stats.total_lost_samples);
 }

 evlist__for_each_entry(evlist, pos) {
  struct hists *hists = evsel__hists(pos);
  const char *evname = evsel__name(pos);

  i++;
  if (symbol_conf.event_group && !evsel__is_group_leader(pos))
   continue;

  if (rep->skip_empty && !hists->stats.nr_samples)
   continue;

  hists__fprintf_nr_sample_events(hists, rep, evname, stdout);

  if (rep->total_cycles_mode) {
   char *buf;

   if (!annotation_br_cntr_abbr_list(&buf, pos, true)) {
    fprintf(stdout, "%s", buf);
    fprintf(stdout, "#\n");
    free(buf);
   }
   report__browse_block_hists(&rep->block_reports[i - 1].hist,
         rep->min_percent, pos, NULL);
   continue;
  }

  hists__fprintf(hists, !quiet, 0, 0, rep->min_percent, stdout,
          !(symbol_conf.use_callchain ||
            symbol_conf.show_branchflag_count));
  fprintf(stdout, "\n\n");
 }

 if (!quiet)
  fprintf(stdout, "#\n# (%s)\n#\n", help);

 if (rep->show_threads) {
  bool style = !strcmp(rep->pretty_printing_style, "raw");
  perf_read_values_display(stdout, &rep->show_threads_values,
      style);
  perf_read_values_destroy(&rep->show_threads_values);
 }

 if (sort__mode == SORT_MODE__BRANCH)
  branch_type_stat_display(stdout, &rep->brtype_stat);

 return 0;
}

static void report__warn_kptr_restrict(const struct report *rep)
{
 struct map *kernel_map = machine__kernel_map(&rep->session->machines.host);
 struct kmap *kernel_kmap = kernel_map ? map__kmap(kernel_map) : NULL;

 if (evlist__exclude_kernel(rep->session->evlist))
  return;

 if (kernel_map == NULL ||
     (dso__hit(map__dso(kernel_map)) &&
      (kernel_kmap->ref_reloc_sym == NULL ||
       kernel_kmap->ref_reloc_sym->addr == 0))) {
  const char *desc =
      "As no suitable kallsyms nor vmlinux was found, kernel samples\n"
      "can't be resolved.";

  if (kernel_map && map__has_symbols(kernel_map)) {
   desc = "If some relocation was applied (e.g. "
          "kexec) symbols may be misresolved.";
  }

  ui__warning(
"Kernel address maps (/proc/{kallsyms,modules}) were restricted.\n\n"
"Check /proc/sys/kernel/kptr_restrict before running 'perf record'.\n\n%s\n\n"
"Samples in kernel modules can't be resolved as well.\n\n",
  desc);
 }
}

static int report__gtk_browse_hists(struct report *rep, const char *help)
{
 int (*hist_browser)(struct evlist *evlist, const char *help,
       struct hist_browser_timer *timer, float min_pcnt);

 hist_browser = dlsym(perf_gtk_handle, "evlist__gtk_browse_hists");

 if (hist_browser == NULL) {
  ui__error("GTK browser not found!\n");
  return -1;
 }

 return hist_browser(rep->session->evlist, help, NULL, rep->min_percent);
}

static int report__browse_hists(struct report *rep)
{
 int ret;
 struct perf_session *session = rep->session;
 struct evlist *evlist = session->evlist;
 char *help = NULL, *path = NULL;

 path = system_path(TIPDIR);
 if (perf_tip(&help, path) || help == NULL) {
  /* fallback for people who don't install perf ;-) */
  free(path);
  path = system_path(DOCDIR);
  if (perf_tip(&help, path) || help == NULL)
   help = strdup("Cannot load tips.txt file, please install perf!");
 }
 free(path);

 switch (use_browser) {
 case 1:
  if (rep->total_cycles_mode) {
   ret = evlist__tui_block_hists_browse(evlist, rep);
   break;
  }

  ret = evlist__tui_browse_hists(evlist, help, NULL, rep->min_percent,
            perf_session__env(session), true);
  /*
 * Usually "ret" is the last pressed key, and we only
 * care if the key notifies us to switch data file.
 */
  if (ret != K_SWITCH_INPUT_DATA && ret != K_RELOAD)
   ret = 0;
  break;
 case 2:
  ret = report__gtk_browse_hists(rep, help);
  break;
 default:
  ret = evlist__tty_browse_hists(evlist, rep, help);
  break;
 }
 free(help);
 return ret;
}

static int report__collapse_hists(struct report *rep)
{
 struct perf_session *session = rep->session;
 struct evlist *evlist = session->evlist;
 struct ui_progress prog;
 struct evsel *pos;
 int ret = 0;

 /*
 * The pipe data needs to setup hierarchy hpp formats now, because it
 * cannot know about evsels in the data before reading the data.  The
 * normal file data saves the event (attribute) info in the header
 * section, but pipe does not have the luxury.
 */
 if (perf_data__is_pipe(session->data)) {
  if (perf_hpp__setup_hists_formats(&perf_hpp_list, evlist) < 0) {
   ui__error("Failed to setup hierarchy output formats\n");
   return -1;
  }
 }

 ui_progress__init(&prog, rep->nr_entries, "Merging related events...");

 evlist__for_each_entry(rep->session->evlist, pos) {
  struct hists *hists = evsel__hists(pos);

  if (pos->core.idx == 0)
   hists->symbol_filter_str = rep->symbol_filter_str;

  hists->socket_filter = rep->socket_filter;

  ret = hists__collapse_resort(hists, &prog);
  if (ret < 0)
   break;

  /* Non-group events are considered as leader */
  if (symbol_conf.event_group && !evsel__is_group_leader(pos)) {
   struct hists *leader_hists = evsel__hists(evsel__leader(pos));

   hists__match(leader_hists, hists);
   hists__link(leader_hists, hists);
  }
 }

 ui_progress__finish();
 return ret;
}

static int hists__resort_cb(struct hist_entry *he, void *arg)
{
 struct report *rep = arg;
 struct symbol *sym = he->ms.sym;

 if (rep->symbol_ipc && sym && !sym->annotate2) {
  struct evsel *evsel = hists_to_evsel(he->hists);

  symbol__annotate2(&he->ms, evsel, NULL);
 }

 return 0;
}

static void report__output_resort(struct report *rep)
{
 struct ui_progress prog;
 struct evsel *pos;

 ui_progress__init(&prog, rep->nr_entries, "Sorting events for output...");

 evlist__for_each_entry(rep->session->evlist, pos) {
  evsel__output_resort_cb(pos, &prog, hists__resort_cb, rep);
 }

 ui_progress__finish();
}

static int count_sample_event(const struct perf_tool *tool __maybe_unused,
         union perf_event *event __maybe_unused,
         struct perf_sample *sample __maybe_unused,
         struct evsel *evsel,
         struct machine *machine __maybe_unused)
{
 struct hists *hists = evsel__hists(evsel);

 hists__inc_nr_events(hists);
 return 0;
}

static int count_lost_samples_event(const struct perf_tool *tool,
        union perf_event *event,
        struct perf_sample *sample,
        struct machine *machine __maybe_unused)
{
 struct report *rep = container_of(tool, struct report, tool);
 struct evsel *evsel;

 evsel = evlist__id2evsel(rep->session->evlist, sample->id);
 if (evsel) {
  struct hists *hists = evsel__hists(evsel);
  u32 count = event->lost_samples.lost;

  if (event->header.misc & PERF_RECORD_MISC_LOST_SAMPLES_BPF)
   hists__inc_nr_dropped_samples(hists, count);
  else
   hists__inc_nr_lost_samples(hists, count);
 }
 return 0;
}

static int process_attr(const struct perf_tool *tool __maybe_unused,
   union perf_event *event,
   struct evlist **pevlist);

static void stats_setup(struct report *rep)
{
 perf_tool__init(&rep->tool, /*ordered_events=*/false);
 rep->tool.attr = process_attr;
 rep->tool.sample = count_sample_event;
 rep->tool.lost_samples = count_lost_samples_event;
 rep->tool.event_update = perf_event__process_event_update;
 rep->tool.no_warn = true;
}

static int stats_print(struct report *rep)
{
 struct perf_session *session = rep->session;

 perf_session__fprintf_nr_events(session, stdout);
 evlist__fprintf_nr_events(session->evlist, stdout);
 return 0;
}

static void tasks_setup(struct report *rep)
{
 perf_tool__init(&rep->tool, /*ordered_events=*/true);
 if (rep->mmaps_mode) {
  rep->tool.mmap = perf_event__process_mmap;
  rep->tool.mmap2 = perf_event__process_mmap2;
 }
 rep->tool.attr = process_attr;
 rep->tool.comm = perf_event__process_comm;
 rep->tool.exit = perf_event__process_exit;
 rep->tool.fork = perf_event__process_fork;
 rep->tool.no_warn = true;
}

struct maps__fprintf_task_args {
 int indent;
 FILE *fp;
 size_t printed;
};

static int maps__fprintf_task_cb(struct map *map, void *data)
{
 struct maps__fprintf_task_args *args = data;
 const struct dso *dso = map__dso(map);
 u32 prot = map__prot(map);
 const struct dso_id *dso_id = dso__id_const(dso);
 int ret;
 char buf[SBUILD_ID_SIZE];

 if (dso_id->mmap2_valid)
  snprintf(buf, sizeof(buf), "%" PRIu64, dso_id->ino);
 else
  build_id__snprintf(&dso_id->build_id, buf, sizeof(buf));

 ret = fprintf(args->fp,
  "%*s %" PRIx64 "-%" PRIx64 " %c%c%c%c %08" PRIx64 " %s %s\n",
  args->indent, "", map__start(map), map__end(map),
  prot & PROT_READ ? 'r' : '-',
  prot & PROT_WRITE ? 'w' : '-',
  prot & PROT_EXEC ? 'x' : '-',
  map__flags(map) ? 's' : 'p',
  map__pgoff(map),
  buf, dso__name(dso));

 if (ret < 0)
  return ret;

 args->printed += ret;
 return 0;
}

static size_t maps__fprintf_task(struct maps *maps, int indent, FILE *fp)
{
 struct maps__fprintf_task_args args = {
  .indent = indent,
  .fp = fp,
  .printed = 0,
 };

 maps__for_each_map(maps, maps__fprintf_task_cb, &args);

 return args.printed;
}

static int thread_level(struct machine *machine, const struct thread *thread)
{
 struct thread *parent_thread;
 int res;

 if (thread__tid(thread) <= 0)
  return 0;

 if (thread__ppid(thread) <= 0)
  return 1;

 parent_thread = machine__find_thread(machine, -1, thread__ppid(thread));
 if (!parent_thread) {
  pr_err("Missing parent thread of %d\n", thread__tid(thread));
  return 0;
 }
 res = 1 + thread_level(machine, parent_thread);
 thread__put(parent_thread);
 return res;
}

static void task__print_level(struct machine *machine, struct thread *thread, FILE *fp)
{
 int level = thread_level(machine, thread);
 int comm_indent = fprintf(fp, " %8d %8d %8d |%*s",
      thread__pid(thread), thread__tid(thread),
      thread__ppid(thread), level, "");

 fprintf(fp, "%s\n", thread__comm_str(thread));

 maps__fprintf_task(thread__maps(thread), comm_indent, fp);
}

/*
 * Sort two thread list nodes such that they form a tree. The first node is the
 * root of the tree, its children are ordered numerically after it. If a child
 * has children itself then they appear immediately after their parent. For
 * example, the 4 threads in the order they'd appear in the list:
 * - init with a TID 1 and a parent of 0
 * - systemd with a TID 3000 and a parent of init/1
 * - systemd child thread with TID 4000, the parent is 3000
 * - NetworkManager is a child of init with a TID of 3500.
 */
static int task_list_cmp(void *priv, const struct list_head *la, const struct list_head *lb)
{
 struct machine *machine = priv;
 struct thread_list *task_a = list_entry(la, struct thread_list, list);
 struct thread_list *task_b = list_entry(lb, struct thread_list, list);
 struct thread *a = task_a->thread;
 struct thread *b = task_b->thread;
 int level_a, level_b, res;

 /* Same thread? */
 if (thread__tid(a) == thread__tid(b))
  return 0;

 /* Compare a and b to root. */
 if (thread__tid(a) == 0)
  return -1;

 if (thread__tid(b) == 0)
  return 1;

 /* If parents match sort by tid. */
 if (thread__ppid(a) == thread__ppid(b))
  return thread__tid(a) < thread__tid(b) ? -1 : 1;

 /*
 * Find a and b such that if they are a child of each other a and b's
 * tid's match, otherwise a and b have a common parent and distinct
 * tid's to sort by. First make the depths of the threads match.
 */
 level_a = thread_level(machine, a);
 level_b = thread_level(machine, b);
 a = thread__get(a);
 b = thread__get(b);
 for (int i = level_a; i > level_b; i--) {
  struct thread *parent = machine__find_thread(machine, -1, thread__ppid(a));

  thread__put(a);
  if (!parent) {
   pr_err("Missing parent thread of %d\n", thread__tid(a));
   thread__put(b);
   return -1;
  }
  a = parent;
 }
 for (int i = level_b; i > level_a; i--) {
  struct thread *parent = machine__find_thread(machine, -1, thread__ppid(b));

  thread__put(b);
  if (!parent) {
   pr_err("Missing parent thread of %d\n", thread__tid(b));
   thread__put(a);
   return 1;
  }
  b = parent;
 }
 /* Search up to a common parent. */
 while (thread__ppid(a) != thread__ppid(b)) {
  struct thread *parent;

  parent = machine__find_thread(machine, -1, thread__ppid(a));
  thread__put(a);
  if (!parent)
   pr_err("Missing parent thread of %d\n", thread__tid(a));
  a = parent;
  parent = machine__find_thread(machine, -1, thread__ppid(b));
  thread__put(b);
  if (!parent)
   pr_err("Missing parent thread of %d\n", thread__tid(b));
  b = parent;
  if (!a || !b) {
   /* Handle missing parent (unexpected) with some sanity. */
   thread__put(a);
   thread__put(b);
   return !a && !b ? 0 : (!a ? -1 : 1);
  }
 }
 if (thread__tid(a) == thread__tid(b)) {
  /* a is a child of b or vice-versa, deeper levels appear later. */
  res = level_a < level_b ? -1 : (level_a > level_b ? 1 : 0);
 } else {
  /* Sort by tid now the parent is the same. */
  res = thread__tid(a) < thread__tid(b) ? -1 : 1;
 }
 thread__put(a);
 thread__put(b);
 return res;
}

static int tasks_print(struct report *rep, FILE *fp)
{
 struct machine *machine = &rep->session->machines.host;
 LIST_HEAD(tasks);
 int ret;

 ret = machine__thread_list(machine, &tasks);
 if (!ret) {
  struct thread_list *task;

  list_sort(machine, &tasks, task_list_cmp);

  fprintf(fp, "# %8s %8s %8s %s\n", "pid", "tid", "ppid", "comm");

  list_for_each_entry(task, &tasks, list)
   task__print_level(machine, task->thread, fp);
 }
 thread_list__delete(&tasks);
 return ret;
}

static int __cmd_report(struct report *rep)
{
 int ret;
 struct perf_session *session = rep->session;
 struct evsel *pos;
 struct perf_data *data = session->data;

 signal(SIGINT, sig_handler);

 if (rep->cpu_list) {
  ret = perf_session__cpu_bitmap(session, rep->cpu_list,
            rep->cpu_bitmap);
  if (ret) {
   ui__error("failed to set cpu bitmap\n");
   return ret;
  }
  session->itrace_synth_opts->cpu_bitmap = rep->cpu_bitmap;
 }

 if (rep->show_threads) {
  ret = perf_read_values_init(&rep->show_threads_values);
  if (ret)
   return ret;
 }

 ret = report__setup_sample_type(rep);
 if (ret) {
  /* report__setup_sample_type() already showed error message */
  return ret;
 }

 if (rep->stats_mode)
  stats_setup(rep);

 if (rep->tasks_mode)
  tasks_setup(rep);

 ret = perf_session__process_events(session);
 if (ret) {
  ui__error("failed to process sample\n");
  return ret;
 }

 /* Don't show Latency column for non-parallel profiles by default. */
 if (!symbol_conf.prefer_latency && rep->total_samples &&
  rep->singlethreaded_samples * 100 / rep->total_samples >= 99)
  perf_hpp__cancel_latency(session->evlist);

 evlist__check_mem_load_aux(session->evlist);

 if (rep->stats_mode)
  return stats_print(rep);

 if (rep->tasks_mode)
  return tasks_print(rep, stdout);

 report__warn_kptr_restrict(rep);

 evlist__for_each_entry(session->evlist, pos)
  rep->nr_entries += evsel__hists(pos)->nr_entries;

 if (use_browser == 0) {
  if (verbose > 3)
   perf_session__fprintf(session, stdout);

  if (verbose > 2)
   perf_session__fprintf_dsos(session, stdout);

  if (dump_trace) {
   stats_print(rep);
   return 0;
  }
 }

 ret = report__collapse_hists(rep);
 if (ret) {
  ui__error("failed to process hist entry\n");
  return ret;
 }

 if (session_done())
  return 0;

 /*
 * recalculate number of entries after collapsing since it
 * might be changed during the collapse phase.
 */
 rep->nr_entries = 0;
 evlist__for_each_entry(session->evlist, pos)
  rep->nr_entries += evsel__hists(pos)->nr_entries;

 if (rep->nr_entries == 0) {
  ui__error("The %s data has no samples!\n", data->path);
  return 0;
 }

 report__output_resort(rep);

 if (rep->total_cycles_mode) {
  int nr_hpps = 4;
  int block_hpps[PERF_HPP_REPORT__BLOCK_MAX_INDEX] = {
   PERF_HPP_REPORT__BLOCK_TOTAL_CYCLES_PCT,
   PERF_HPP_REPORT__BLOCK_LBR_CYCLES,
   PERF_HPP_REPORT__BLOCK_CYCLES_PCT,
   PERF_HPP_REPORT__BLOCK_AVG_CYCLES,
  };

  if (session->evlist->nr_br_cntr > 0)
   block_hpps[nr_hpps++] = PERF_HPP_REPORT__BLOCK_BRANCH_COUNTER;

  block_hpps[nr_hpps++] = PERF_HPP_REPORT__BLOCK_RANGE;
  block_hpps[nr_hpps++] = PERF_HPP_REPORT__BLOCK_DSO;

  rep->block_reports = block_info__create_report(session->evlist,
              rep->total_cycles,
              block_hpps, nr_hpps,
              &rep->nr_block_reports);
  if (!rep->block_reports)
   return -1;
 }

 return report__browse_hists(rep);
}

static int
report_parse_callchain_opt(const struct option *opt, const char *arg, int unset)
{
 struct callchain_param *callchain = opt->value;

 callchain->enabled = !unset;
 /*
 * --no-call-graph
 */
 if (unset) {
  symbol_conf.use_callchain = false;
  callchain->mode = CHAIN_NONE;
  return 0;
 }

 return parse_callchain_report_opt(arg);
}

static int
parse_time_quantum(const struct option *opt, const char *arg,
     int unset __maybe_unused)
{
 unsigned long *time_q = opt->value;
 char *end;

 *time_q = strtoul(arg, &end, 0);
 if (end == arg)
  goto parse_err;
 if (*time_q == 0) {
  pr_err("time quantum cannot be 0");
  return -1;
 }
 end = skip_spaces(end);
 if (*end == 0)
  return 0;
 if (!strcmp(end, "s")) {
  *time_q *= NSEC_PER_SEC;
  return 0;
 }
 if (!strcmp(end, "ms")) {
  *time_q *= NSEC_PER_MSEC;
  return 0;
 }
 if (!strcmp(end, "us")) {
  *time_q *= NSEC_PER_USEC;
  return 0;
 }
 if (!strcmp(end, "ns"))
  return 0;
parse_err:
 pr_err("Cannot parse time quantum `%s'\n", arg);
 return -1;
}

int
report_parse_ignore_callees_opt(const struct option *opt __maybe_unused,
    const char *arg, int unset __maybe_unused)
{
 if (arg) {
  int err = regcomp(&ignore_callees_regex, arg, REG_EXTENDED);
  if (err) {
   char buf[BUFSIZ];
   regerror(err, &ignore_callees_regex, buf, sizeof(buf));
   pr_err("Invalid --ignore-callees regex: %s\n%s", arg, buf);
   return -1;
  }
  have_ignore_callees = 1;
 }

 return 0;
}

static int
parse_branch_mode(const struct option *opt,
    const char *str __maybe_unused, int unset)
{
 int *branch_mode = opt->value;

 *branch_mode = !unset;
 return 0;
}

static int
parse_percent_limit(const struct option *opt, const char *str,
      int unset __maybe_unused)
{
 struct report *rep = opt->value;
 double pcnt = strtof(str, NULL);

 rep->min_percent = pcnt;
 callchain_param.min_percent = pcnt;
 return 0;
}

static int process_attr(const struct perf_tool *tool __maybe_unused,
   union perf_event *event,
   struct evlist **pevlist)
{
 struct perf_session *session;
 struct perf_env *env;
 u64 sample_type;
 int err;

 err = perf_event__process_attr(tool, event, pevlist);
 if (err)
  return err;

 /*
 * Check if we need to enable callchains based
 * on events sample_type.
 */
 sample_type = evlist__combined_sample_type(*pevlist);
 session = (*pevlist)->session;
 env = perf_session__env(session);
 callchain_param_setup(sample_type, perf_env__arch(env));
 return 0;
}

#define CALLCHAIN_BRANCH_SORT_ORDER \
 "srcline,symbol,dso,callchain_branch_predicted," \
 "callchain_branch_abort,callchain_branch_cycles"

int cmd_report(int argc, const char **argv)
{
 struct perf_session *session;
 struct itrace_synth_opts itrace_synth_opts = { .set = 0, };
 struct stat st;
 bool has_br_stack = false;
 int branch_mode = -1;
 int last_key = 0;
 bool branch_call_mode = false;
#define CALLCHAIN_DEFAULT_OPT  "graph,0.5,caller,function,percent"
 static const char report_callchain_help[] = "Display call graph (stack chain/backtrace):\n\n"
          CALLCHAIN_REPORT_HELP
          "\n\t\t\t\tDefault: " CALLCHAIN_DEFAULT_OPT;
 char callchain_default_opt[] = CALLCHAIN_DEFAULT_OPT;
 const char * const report_usage[] = {
  "perf report []",
  NULL
 };
 struct report report = {
  .max_stack   = PERF_MAX_STACK_DEPTH,
  .pretty_printing_style  = "normal",
  .socket_filter   = -1,
  .skip_empty   = true,
 };
 char *sort_order_help = sort_help("sort by key(s):", SORT_MODE__NORMAL);
 char *field_order_help = sort_help("output field(s):", SORT_MODE__NORMAL);
 const char *disassembler_style = NULL, *objdump_path = NULL, *addr2line_path = NULL;
 const struct option options[] = {
 OPT_STRING('i', "input", &input_name, "file",
      "input file name"),
 OPT_INCR('v', "verbose", &verbose,
      "be more verbose (show symbol address, etc)"),
 OPT_BOOLEAN('q', "quiet", &quiet, "Do not show any warnings or messages"),
 OPT_BOOLEAN('D', "dump-raw-trace", &dump_trace,
      "dump raw trace in ASCII"),
 OPT_BOOLEAN(0, "stats", &report.stats_mode, "Display event stats"),
 OPT_BOOLEAN(0, "tasks", &report.tasks_mode, "Display recorded tasks"),
 OPT_BOOLEAN(0, "mmaps", &report.mmaps_mode, "Display recorded tasks memory maps"),
 OPT_STRING('k', "vmlinux", &symbol_conf.vmlinux_name,
     "file", "vmlinux pathname"),
 OPT_BOOLEAN(0, "ignore-vmlinux", &symbol_conf.ignore_vmlinux,
                    "don't load vmlinux even if found"),
 OPT_STRING(0, "kallsyms", &symbol_conf.kallsyms_name,
     "file", "kallsyms pathname"),
 OPT_BOOLEAN('f', "force", &symbol_conf.force, "don't complain, do it"),
 OPT_BOOLEAN('m', "modules", &symbol_conf.use_modules,
      "load module symbols - WARNING: use only with -k and LIVE kernel"),
 OPT_BOOLEAN('n', "show-nr-samples", &symbol_conf.show_nr_samples,
      "Show a column with the number of samples"),
 OPT_BOOLEAN('T', "threads", &report.show_threads,
      "Show per-thread event counters"),
 OPT_STRING(0, "pretty", &report.pretty_printing_style, "key",
     "pretty printing style key: normal raw"),
#ifdef HAVE_SLANG_SUPPORT
 OPT_BOOLEAN(0, "tui", &report.use_tui, "Use the TUI interface"),
#endif
#ifdef HAVE_GTK2_SUPPORT
 OPT_BOOLEAN(0, "gtk", &report.use_gtk, "Use the GTK2 interface"),
#endif
 OPT_BOOLEAN(0, "stdio", &report.use_stdio,
      "Use the stdio interface"),
 OPT_BOOLEAN(0, "header", &report.header, "Show data header."),
 OPT_BOOLEAN(0, "header-only", &report.header_only,
      "Show only data header."),
 OPT_STRING('s', "sort", &sort_order, "key[,key2...]",
     sort_order_help),
 OPT_STRING('F', "fields", &field_order, "key[,keys...]",
     field_order_help),
 OPT_BOOLEAN(0, "show-cpu-utilization", &symbol_conf.show_cpu_utilization,
      "Show sample percentage for different cpu modes"),
 OPT_BOOLEAN_FLAG(0, "showcpuutilization", &symbol_conf.show_cpu_utilization,
      "Show sample percentage for different cpu modes", PARSE_OPT_HIDDEN),
 OPT_STRING('p', "parent", &parent_pattern, "regex",
     "regex filter to identify parent, see: '--sort parent'"),
 OPT_BOOLEAN('x', "exclude-other", &symbol_conf.exclude_other,
      "Only display entries with parent-match"),
 OPT_CALLBACK_DEFAULT('g', "call-graph", &callchain_param,
        "print_type,threshold[,print_limit],order,sort_key[,branch],value",
        report_callchain_help, &report_parse_callchain_opt,
        callchain_default_opt),
 OPT_BOOLEAN(0, "children", &symbol_conf.cumulate_callchain,
      "Accumulate callchains of children and show total overhead as well. "
      "Enabled by default, use --no-children to disable."),
 OPT_INTEGER(0, "max-stack", &report.max_stack,
      "Set the maximum stack depth when parsing the callchain, "
      "anything beyond the specified depth will be ignored. "
      "Default: kernel.perf_event_max_stack or " __stringify(PERF_MAX_STACK_DEPTH)),
 OPT_BOOLEAN('G', "inverted", &report.inverted_callchain,
      "alias for inverted call graph"),
 OPT_CALLBACK(0, "ignore-callees", NULL, "regex",
     "ignore callees of these functions in call graphs",
     report_parse_ignore_callees_opt),
 OPT_STRING('d', "dsos", &symbol_conf.dso_list_str, "dso[,dso...]",
     "only consider symbols in these dsos"),
 OPT_STRING('c', "comms", &symbol_conf.comm_list_str, "comm[,comm...]",
     "only consider symbols in these comms"),
 OPT_STRING(0, "pid", &symbol_conf.pid_list_str, "pid[,pid...]",
     "only consider symbols in these pids"),
 OPT_STRING(0, "tid", &symbol_conf.tid_list_str, "tid[,tid...]",
     "only consider symbols in these tids"),
 OPT_STRING('S', "symbols", &symbol_conf.sym_list_str, "symbol[,symbol...]",
     "only consider these symbols"),
 OPT_STRING(0, "symbol-filter", &report.symbol_filter_str, "filter",
     "only show symbols that (partially) match with this filter"),
 OPT_STRING('w', "column-widths", &symbol_conf.col_width_list_str,
     "width[,width...]",
     "don't try to adjust column width, use these fixed values"),
 OPT_STRING_NOEMPTY('t', "field-separator", &symbol_conf.field_sep, "separator",
     "separator for columns, no spaces will be added between "
     "columns '.' is reserved."),
 OPT_BOOLEAN('U', "hide-unresolved", &symbol_conf.hide_unresolved,
      "Only display entries resolved to a symbol"),
 OPT_CALLBACK(0, "symfs", NULL, "directory",
       "Look for files with symbols relative to this directory",
       symbol__config_symfs),
 OPT_STRING('C', "cpu", &report.cpu_list, "cpu",
     "list of cpus to profile"),
 OPT_STRING(0, "parallelism", &symbol_conf.parallelism_list_str, "parallelism",
     "only consider these parallelism levels (cpu set format)"),
 OPT_BOOLEAN('I', "show-info", &report.show_full_info,
      "Display extended information about perf.data file"),
 OPT_BOOLEAN(0, "source", &annotate_opts.annotate_src,
      "Interleave source code with assembly code (default)"),
 OPT_BOOLEAN(0, "asm-raw", &annotate_opts.show_asm_raw,
      "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(0, "prefix", &annotate_opts.prefix, "prefix",
      "Add prefix to source file path names in programs (with --prefix-strip)"),
 OPT_STRING(0, "prefix-strip", &annotate_opts.prefix_strip, "N",
      "Strip first N entries of source file path name in programs (with --prefix)"),
 OPT_BOOLEAN(0, "show-total-period", &symbol_conf.show_total_period,
      "Show a column with the sum of periods"),
 OPT_BOOLEAN_SET(0, "group", &symbol_conf.event_group, &report.group_set,
      "Show event group information together"),
 OPT_INTEGER(0, "group-sort-idx", &symbol_conf.group_sort_idx,
      "Sort the output by the event at the index n in group. "
      "If n is invalid, sort by the first event. "
      "WARNING: should be used on grouped events."),
 OPT_CALLBACK_NOOPT('b', "branch-stack", &branch_mode, "",
      "use branch records for per branch histogram filling",
      parse_branch_mode),
 OPT_BOOLEAN(0, "branch-history", &branch_call_mode,
      "add last branch records to call history"),
 OPT_STRING(0, "objdump", &objdump_path, "path",
     "objdump binary to use for disassembly and annotations"),
 OPT_STRING(0, "addr2line", &addr2line_path, "path",
     "addr2line binary to use for line numbers"),
 OPT_BOOLEAN(0, "demangle", &symbol_conf.demangle,
      "Symbol demangling. Enabled by default, use --no-demangle to disable."),
 OPT_BOOLEAN(0, "demangle-kernel", &symbol_conf.demangle_kernel,
      "Enable kernel symbol demangling"),
 OPT_BOOLEAN(0, "mem-mode", &report.mem_mode, "mem access profile"),
 OPT_INTEGER(0, "samples", &symbol_conf.res_sample,
      "Number of samples to save per histogram entry for individual browsing"),
 OPT_CALLBACK(0, "percent-limit", &report, "percent",
       "Don't show entries under that percent", parse_percent_limit),
 OPT_CALLBACK(0, "percentage", NULL, "relative|absolute",
       "how to display percentage of filtered entries", parse_filter_percentage),
 OPT_CALLBACK_OPTARG(0, "itrace", &itrace_synth_opts, NULL, "opts",
       "Instruction Tracing options\n" ITRACE_HELP,
       itrace_parse_synth_opts),
 OPT_BOOLEAN(0, "full-source-path", &srcline_full_filename,
   "Show full source file name path for source lines"),
 OPT_BOOLEAN(0, "show-ref-call-graph", &symbol_conf.show_ref_callgraph,
      "Show callgraph from reference event"),
 OPT_BOOLEAN(0, "stitch-lbr", &report.stitch_lbr,
      "Enable LBR callgraph stitching approach"),
 OPT_INTEGER(0, "socket-filter", &report.socket_filter,
      "only show processor socket that match with this filter"),
 OPT_BOOLEAN(0, "raw-trace", &symbol_conf.raw_trace,
      "Show raw trace event output (do not use print fmt or plugins)"),
 OPT_BOOLEAN('H', "hierarchy", &symbol_conf.report_hierarchy,
      "Show entries in a hierarchy"),
 OPT_CALLBACK_DEFAULT(0, "stdio-color", NULL, "mode",
        "'always' (default), 'never' or 'auto' only applicable to --stdio mode",
        stdio__config_color, "always"),
 OPT_STRING(0, "time", &report.time_str, "str",
     "Time span of interest (start,stop)"),
 OPT_BOOLEAN(0, "inline", &symbol_conf.inline_name,
      "Show inline function"),
 OPT_CALLBACK(0, "percent-type", &annotate_opts, "local-period",
       "Set percent type local/global-period/hits",
       annotate_parse_percent_type),
 OPT_BOOLEAN(0, "ns", &symbol_conf.nanosecs, "Show times in nanosecs"),
 OPT_CALLBACK(0, "time-quantum", &symbol_conf.time_quantum, "time (ms|us|ns|s)",
       "Set time quantum for time sort key (default 100ms)",
       parse_time_quantum),
 OPTS_EVSWITCH(&report.evswitch),
 OPT_BOOLEAN(0, "total-cycles", &report.total_cycles_mode,
      "Sort all blocks by 'Sampled Cycles%'"),
 OPT_BOOLEAN(0, "disable-order", &report.disable_order,
      "Disable raw trace ordering"),
 OPT_BOOLEAN(0, "skip-empty", &report.skip_empty,
      "Do not display empty (or dummy) events in the output"),
 OPT_BOOLEAN(0, "latency", &symbol_conf.prefer_latency,
      "Show latency-centric profile rather than the default\n"
      "\t\t\t CPU-consumption-centric profile\n"
      "\t\t\t (requires perf record --latency flag)."),
 OPT_END()
 };
 struct perf_data data = {
  .mode  = PERF_DATA_MODE_READ,
 };
 int ret = hists__init();
 char sort_tmp[128];
 bool ordered_events = true;

 if (ret < 0)
  goto exit;

 /*
 * tasks_mode require access to exited threads to list those that are in
 * the data file. Off-cpu events are synthesized after other events and
 * reference exited threads.
 */
 symbol_conf.keep_exited_threads = true;

 annotation_options__init();

 ret = perf_config(report__config, &report);
 if (ret)
  goto exit;

 argc = parse_options(argc, argv, options, report_usage, 0);
 if (argc) {
  /*
 * Special case: if there's an argument left then assume that
 * it's a symbol filter:
 */
  if (argc > 1)
   usage_with_options(report_usage, options);

  report.symbol_filter_str = argv[0];
 }

 if (disassembler_style) {
  annotate_opts.disassembler_style = strdup(disassembler_style);
  if (!annotate_opts.disassembler_style)
   return -ENOMEM;
 }
 if (objdump_path) {
  annotate_opts.objdump_path = strdup(objdump_path);
  if (!annotate_opts.objdump_path)
   return -ENOMEM;
 }
 if (addr2line_path) {
  symbol_conf.addr2line_path = strdup(addr2line_path);
  if (!symbol_conf.addr2line_path)
   return -ENOMEM;
 }

 if (annotate_check_args() < 0) {
  ret = -EINVAL;
  goto exit;
 }

 if (report.mmaps_mode)
  report.tasks_mode = true;

 if (dump_trace && report.disable_order)
  ordered_events = false;

 if (quiet)
  perf_quiet_option();

 ret = symbol__validate_sym_arguments();
 if (ret)
  goto exit;

 if (report.inverted_callchain)
  callchain_param.order = ORDER_CALLER;
 if (symbol_conf.cumulate_callchain && !callchain_param.order_set)
  callchain_param.order = ORDER_CALLER;

 if ((itrace_synth_opts.callchain || itrace_synth_opts.add_callchain) &&
     (int)itrace_synth_opts.callchain_sz > report.max_stack)
  report.max_stack = itrace_synth_opts.callchain_sz;

 if (!input_name || !strlen(input_name)) {
  if (!fstat(STDIN_FILENO, &st) && S_ISFIFO(st.st_mode))
   input_name = "-";
  else
   input_name = "perf.data";
 }

repeat:
 data.path  = input_name;
 data.force = symbol_conf.force;

 symbol_conf.skip_empty = report.skip_empty;

 perf_tool__init(&report.tool, ordered_events);
 report.tool.sample   = process_sample_event;
 report.tool.mmap   = perf_event__process_mmap;
 report.tool.mmap2   = perf_event__process_mmap2;
 report.tool.comm   = perf_event__process_comm;
 report.tool.namespaces   = perf_event__process_namespaces;
 report.tool.cgroup   = perf_event__process_cgroup;
 report.tool.exit   = perf_event__process_exit;
 report.tool.fork   = perf_event__process_fork;
 report.tool.context_switch  = perf_event__process_switch;
 report.tool.lost   = perf_event__process_lost;
 report.tool.read   = process_read_event;
 report.tool.attr   = process_attr;
#ifdef HAVE_LIBTRACEEVENT
 report.tool.tracing_data  = perf_event__process_tracing_data;
#endif
 report.tool.build_id   = perf_event__process_build_id;
 report.tool.id_index   = perf_event__process_id_index;
 report.tool.auxtrace_info  = perf_event__process_auxtrace_info;
 report.tool.auxtrace   = perf_event__process_auxtrace;
 report.tool.event_update  = perf_event__process_event_update;
 report.tool.feature   = process_feature_event;
 report.tool.ordering_requires_timestamps = true;

 session = perf_session__new(&data, &report.tool);
 if (IS_ERR(session)) {
  ret = PTR_ERR(session);
  goto exit;
 }

 ret = evswitch__init(&report.evswitch, session->evlist, stderr);
 if (ret)
  goto exit;

 if (zstd_init(&(session->zstd_data), 0) < 0)
  pr_warning("Decompression initialization failed. Reported data may be incomplete.\n");

 if (report.queue_size) {
  ordered_events__set_alloc_size(&session->ordered_events,
            report.queue_size);
 }

 session->itrace_synth_opts = &itrace_synth_opts;

 report.session = session;

 has_br_stack = perf_header__has_feat(&session->header,
          HEADER_BRANCH_STACK);
 if (evlist__combined_sample_type(session->evlist) & PERF_SAMPLE_STACK_USER)
  has_br_stack = false;

 setup_forced_leader(&report, session->evlist);

 if (symbol_conf.group_sort_idx && evlist__nr_groups(session->evlist) == 0) {
  parse_options_usage(NULL, options, "group-sort-idx", 0);
  ret = -EINVAL;
  goto error;
 }

 if (itrace_synth_opts.last_branch || itrace_synth_opts.add_last_branch)
  has_br_stack = true;

 if (has_br_stack && branch_call_mode)
  symbol_conf.show_branchflag_count = true;

 memset(&report.brtype_stat, 0, sizeof(struct branch_type_stat));

 /*
 * Branch mode is a tristate:
 * -1 means default, so decide based on the file having branch data.
 * 0/1 means the user chose a mode.
 */
 if (((branch_mode == -1 && has_br_stack) || branch_mode == 1) &&
     !branch_call_mode) {
  sort__mode = SORT_MODE__BRANCH;
  symbol_conf.cumulate_callchain = false;
 }
 if (branch_call_mode) {
  callchain_param.key = CCKEY_ADDRESS;
  callchain_param.branch_callstack = true;
  symbol_conf.use_callchain = true;
  callchain_register_param(&callchain_param);
  if (sort_order == NULL)
   sort_order = CALLCHAIN_BRANCH_SORT_ORDER;
 }

 if (report.mem_mode) {
  if (sort__mode == SORT_MODE__BRANCH) {
   pr_err("branch and mem mode incompatible\n");
   goto error;
  }
  sort__mode = SORT_MODE__MEMORY;
  symbol_conf.cumulate_callchain = false;
 }

 if (symbol_conf.report_hierarchy) {
  /*
 * The hist entries in hierarchy are added during the collpase
 * phase.  Let's enable it even if no sort keys require it.
 */
  perf_hpp_list.need_collapse = true;
 }

 if (report.use_stdio)
  use_browser = 0;
#ifdef HAVE_SLANG_SUPPORT
 else if (report.use_tui)
  use_browser = 1;
#endif
#ifdef HAVE_GTK2_SUPPORT
 else if (report.use_gtk)
  use_browser = 2;
#endif

 /* Force tty output for header output and per-thread stat. */
 if (report.header || report.header_only || report.show_threads)
  use_browser = 0;
 if (report.header || report.header_only)
  report.tool.show_feat_hdr = SHOW_FEAT_HEADER;
 if (report.show_full_info)
  report.tool.show_feat_hdr = SHOW_FEAT_HEADER_FULL_INFO;
 if (report.stats_mode || report.tasks_mode)
  use_browser = 0;
 if (report.stats_mode && report.tasks_mode) {
  pr_err("Error: --tasks and --mmaps can't be used together with --stats\n");
  goto error;
 }

 if (report.total_cycles_mode) {
  if (sort__mode != SORT_MODE__BRANCH)
   report.total_cycles_mode = false;
  else
   sort_order = NULL;
 }

 if (sort_order && strstr(sort_order, "type")) {
  report.data_type = true;
  annotate_opts.annotate_src = false;

  /* disable incompatible options */
  symbol_conf.cumulate_callchain = false;

#ifndef HAVE_LIBDW_SUPPORT
  pr_err("Error: Data type profiling is disabled due to missing DWARF support\n");
  goto error;
#endif
 }

 if (strcmp(input_name, "-") != 0)
  setup_browser(true);
 else
  use_browser = 0;

 if (report.data_type && use_browser == 1) {
  symbol_conf.annotate_data_member = true;
  symbol_conf.annotate_data_sample = true;
 }

 symbol_conf.enable_latency = true;
 if (report.disable_order || !perf_session__has_switch_events(session)) {
  if (symbol_conf.parallelism_list_str ||
   symbol_conf.prefer_latency ||
   (sort_order && (strstr(sort_order, "latency") ||
    strstr(sort_order, "parallelism"))) ||
   (field_order && (strstr(field_order, "latency") ||
    strstr(field_order, "parallelism")))) {
   if (report.disable_order)
    ui__error("Use of latency profile or parallelism is incompatible with --disable-order.\n");
   else
    ui__error("Use of latency profile or parallelism requires --latency flag during record.\n");
   return -1;
  }
  /*
 * If user did not ask for anything related to
 * latency/parallelism explicitly, just don't show it.
 */
  symbol_conf.enable_latency = false;
 }

 if (last_key != K_SWITCH_INPUT_DATA) {
  if (sort_order && strstr(sort_order, "ipc")) {
   parse_options_usage(report_usage, options, "s", 1);
   goto error;
  }

  if (sort_order && strstr(sort_order, "symbol")) {
   if (sort__mode == SORT_MODE__BRANCH) {
    snprintf(sort_tmp, sizeof(sort_tmp), "%s,%s",
      sort_order, "ipc_lbr");
    report.symbol_ipc = true;
   } else {
    snprintf(sort_tmp, sizeof(sort_tmp), "%s,%s",
      sort_order, "ipc_null");
   }

   sort_order = sort_tmp;
  }
 }

 if ((last_key != K_SWITCH_INPUT_DATA && last_key != K_RELOAD) &&
     (setup_sorting(session->evlist, perf_session__env(session)) < 0)) {
  if (sort_order)
   parse_options_usage(report_usage, options, "s", 1);
  if (field_order)
   parse_options_usage(sort_order ? NULL : report_usage,
         options, "F", 1);
  goto error;
 }

 if ((report.header || report.header_only) && !quiet) {
  perf_session__fprintf_info(session, stdout,
        report.show_full_info);
  if (report.header_only) {
   if (data.is_pipe) {
    /*
 * we need to process first few records
 * which contains PERF_RECORD_HEADER_FEATURE.
 */
    perf_session__process_events(session);
   }
   ret = 0;
   goto error;
  }
 } else if (use_browser == 0 && !quiet &&
     !report.stats_mode && !report.tasks_mode) {
  fputs("# To display the perf.data header info, please use --header/--header-only options.\n#\n",
        stdout);
 }

 /*
 * Only in the TUI browser we are doing integrated annotation,
 * so don't allocate extra space that won't be used in the stdio
 * implementation.
 */
 if (ui__has_annotation() || report.symbol_ipc || report.data_type ||
     report.total_cycles_mode) {
  ret = symbol__annotation_init();
  if (ret < 0)
   goto error;
  /*
   * For searching by name on the "Browse map details".
   * providing it only in verbose mode not to bloat too
   * much struct symbol.
   */
  if (verbose > 0) {
   /*
 * XXX: Need to provide a less kludgy way to ask for
 * more space per symbol, the u32 is for the index on
 * the ui browser.
 * See symbol__browser_index.
 */
   symbol_conf.priv_size += sizeof(u32);
  }
  annotation_config__init();
 }

 if (symbol__init(perf_session__env(session)) < 0)
  goto error;

 if (report.time_str) {
  ret = perf_time__parse_for_ranges(report.time_str, session,
        &report.ptime_range,
        &report.range_size,
        &report.range_num);
  if (ret < 0)
   goto error;

  itrace_synth_opts__set_time_range(&itrace_synth_opts,
        report.ptime_range,
        report.range_num);
 }

#ifdef HAVE_LIBTRACEEVENT
 if (session->tevent.pevent &&
     tep_set_function_resolver(session->tevent.pevent,
          machine__resolve_kernel_addr,
          &session->machines.host) < 0) {
  pr_err("%s: failed to set libtraceevent function resolver\n",
         __func__);
  return -1;
 }
#endif
 sort__setup_elide(stdout);

 ret = __cmd_report(&report);
 if (ret == K_SWITCH_INPUT_DATA || ret == K_RELOAD) {
  perf_session__delete(session);
  last_key = K_SWITCH_INPUT_DATA;
  /*
 * To support switching between data with and without callchains.
 * report__setup_sample_type() will update it properly.
 */
  symbol_conf.use_callchain = false;
  goto repeat;
 } else
  ret = 0;

 if (!use_browser && (verbose > 2 || debug_kmaps))
  perf_session__dump_kmaps(session);
error:
 if (report.ptime_range) {
  itrace_synth_opts__clear_time_range(&itrace_synth_opts);
  zfree(&report.ptime_range);
 }

 if (report.block_reports) {
  block_info__free_report(report.block_reports,
     report.nr_block_reports);
  report.block_reports = NULL;
 }

 zstd_fini(&(session->zstd_data));
 perf_session__delete(session);
exit:
 annotation_options__exit();
 free(sort_order_help);
 free(field_order_help);
 return ret;
}