1 // SPDX-License-Identifier: GPL-2.0
3 * This is rewrite of original c2c tool introduced in here:
4 * http://lwn.net/Articles/588866/
6 * The original tool was changed to fit in current perf state.
9 * Don Zickus <dzickus@redhat.com>
10 * Dick Fowles <fowles@inreach.com>
11 * Joe Mario <jmario@redhat.com>
15 #include <linux/compiler.h>
16 #include <linux/kernel.h>
17 #include <linux/stringify.h>
19 #include <sys/param.h>
23 #include <subcmd/parse-options.h>
24 #include "mem-events.h"
35 #include "ui/browsers/hists.h"
41 struct perf_hpp_list list;
42 struct c2c_stats stats;
45 struct compute_stats {
46 struct stats lcl_hitm;
47 struct stats rmt_hitm;
51 struct c2c_hist_entry {
52 struct c2c_hists *hists;
53 struct c2c_stats stats;
54 unsigned long *cpuset;
55 struct c2c_stats *node_stats;
56 unsigned int cacheline_idx;
58 struct compute_stats cstats;
62 * because of its callchain dynamic entry
67 static char const *coalesce_default = "pid,iaddr";
70 struct perf_tool tool;
71 struct c2c_hists hists;
73 unsigned long **nodes;
85 /* HITM shared clines stats */
86 struct c2c_stats hitm_stats;
104 static const char *display_str[DISPLAY_MAX] = {
105 [DISPLAY_LCL] = "Local",
106 [DISPLAY_RMT] = "Remote",
107 [DISPLAY_TOT] = "Total",
110 static const struct option c2c_options[] = {
111 OPT_INCR('v', "verbose", &verbose, "be more verbose (show counter open errors, etc)"),
115 static struct perf_c2c c2c;
117 static void *c2c_he_zalloc(size_t size)
119 struct c2c_hist_entry *c2c_he;
121 c2c_he = zalloc(size + sizeof(*c2c_he));
125 c2c_he->cpuset = bitmap_alloc(c2c.cpus_cnt);
129 c2c_he->node_stats = zalloc(c2c.nodes_cnt * sizeof(*c2c_he->node_stats));
130 if (!c2c_he->node_stats)
133 init_stats(&c2c_he->cstats.lcl_hitm);
134 init_stats(&c2c_he->cstats.rmt_hitm);
135 init_stats(&c2c_he->cstats.load);
140 static void c2c_he_free(void *he)
142 struct c2c_hist_entry *c2c_he;
144 c2c_he = container_of(he, struct c2c_hist_entry, he);
146 hists__delete_entries(&c2c_he->hists->hists);
150 free(c2c_he->cpuset);
151 free(c2c_he->node_stats);
155 static struct hist_entry_ops c2c_entry_ops = {
156 .new = c2c_he_zalloc,
160 static int c2c_hists__init(struct c2c_hists *hists,
162 int nr_header_lines);
164 static struct c2c_hists*
165 he__get_c2c_hists(struct hist_entry *he,
169 struct c2c_hist_entry *c2c_he;
170 struct c2c_hists *hists;
173 c2c_he = container_of(he, struct c2c_hist_entry, he);
175 return c2c_he->hists;
177 hists = c2c_he->hists = zalloc(sizeof(*hists));
181 ret = c2c_hists__init(hists, sort, nr_header_lines);
190 static void c2c_he__set_cpu(struct c2c_hist_entry *c2c_he,
191 struct perf_sample *sample)
193 if (WARN_ONCE(sample->cpu == (unsigned int) -1,
194 "WARNING: no sample cpu value"))
197 set_bit(sample->cpu, c2c_he->cpuset);
200 static void compute_stats(struct c2c_hist_entry *c2c_he,
201 struct c2c_stats *stats,
204 struct compute_stats *cstats = &c2c_he->cstats;
207 update_stats(&cstats->rmt_hitm, weight);
208 else if (stats->lcl_hitm)
209 update_stats(&cstats->lcl_hitm, weight);
210 else if (stats->load)
211 update_stats(&cstats->load, weight);
214 static int process_sample_event(struct perf_tool *tool __maybe_unused,
215 union perf_event *event,
216 struct perf_sample *sample,
217 struct perf_evsel *evsel,
218 struct machine *machine)
220 struct c2c_hists *c2c_hists = &c2c.hists;
221 struct c2c_hist_entry *c2c_he;
222 struct c2c_stats stats = { .nr_entries = 0, };
223 struct hist_entry *he;
224 struct addr_location al;
225 struct mem_info *mi, *mi_dup;
228 if (machine__resolve(machine, &al, sample) < 0) {
229 pr_debug("problem processing %d event, skipping it.\n",
234 ret = sample__resolve_callchain(sample, &callchain_cursor, NULL,
235 evsel, &al, sysctl_perf_event_max_stack);
239 mi = sample__resolve_mem(sample, &al);
243 mi_dup = memdup(mi, sizeof(*mi));
247 c2c_decode_stats(&stats, mi);
249 he = hists__add_entry_ops(&c2c_hists->hists, &c2c_entry_ops,
255 c2c_he = container_of(he, struct c2c_hist_entry, he);
256 c2c_add_stats(&c2c_he->stats, &stats);
257 c2c_add_stats(&c2c_hists->stats, &stats);
259 c2c_he__set_cpu(c2c_he, sample);
261 hists__inc_nr_samples(&c2c_hists->hists, he->filtered);
262 ret = hist_entry__append_callchain(he, sample);
266 * There's already been warning about missing
267 * sample's cpu value. Let's account all to
268 * node 0 in this case, without any further
271 * Doing node stats only for single callchain data.
273 int cpu = sample->cpu == (unsigned int) -1 ? 0 : sample->cpu;
274 int node = c2c.cpu2node[cpu];
278 mi_dup = memdup(mi, sizeof(*mi));
282 c2c_hists = he__get_c2c_hists(he, c2c.cl_sort, 2);
286 he = hists__add_entry_ops(&c2c_hists->hists, &c2c_entry_ops,
292 c2c_he = container_of(he, struct c2c_hist_entry, he);
293 c2c_add_stats(&c2c_he->stats, &stats);
294 c2c_add_stats(&c2c_hists->stats, &stats);
295 c2c_add_stats(&c2c_he->node_stats[node], &stats);
297 compute_stats(c2c_he, &stats, sample->weight);
299 c2c_he__set_cpu(c2c_he, sample);
301 hists__inc_nr_samples(&c2c_hists->hists, he->filtered);
302 ret = hist_entry__append_callchain(he, sample);
306 addr_location__put(&al);
317 static struct perf_c2c c2c = {
319 .sample = process_sample_event,
320 .mmap = perf_event__process_mmap,
321 .mmap2 = perf_event__process_mmap2,
322 .comm = perf_event__process_comm,
323 .exit = perf_event__process_exit,
324 .fork = perf_event__process_fork,
325 .lost = perf_event__process_lost,
326 .ordered_events = true,
327 .ordering_requires_timestamps = true,
331 static const char * const c2c_usage[] = {
332 "perf c2c {record|report}",
336 static const char * const __usage_report[] = {
341 static const char * const *report_c2c_usage = __usage_report;
343 #define C2C_HEADER_MAX 2
349 } line[C2C_HEADER_MAX];
352 struct c2c_dimension {
353 struct c2c_header header;
356 struct sort_entry *se;
358 int64_t (*cmp)(struct perf_hpp_fmt *fmt,
359 struct hist_entry *, struct hist_entry *);
360 int (*entry)(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
361 struct hist_entry *he);
362 int (*color)(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
363 struct hist_entry *he);
367 struct perf_hpp_fmt fmt;
368 struct c2c_dimension *dim;
371 #define SYMBOL_WIDTH 30
373 static struct c2c_dimension dim_symbol;
374 static struct c2c_dimension dim_srcline;
376 static int symbol_width(struct hists *hists, struct sort_entry *se)
378 int width = hists__col_len(hists, se->se_width_idx);
380 if (!c2c.symbol_full)
381 width = MIN(width, SYMBOL_WIDTH);
386 static int c2c_width(struct perf_hpp_fmt *fmt,
387 struct perf_hpp *hpp __maybe_unused,
390 struct c2c_fmt *c2c_fmt;
391 struct c2c_dimension *dim;
393 c2c_fmt = container_of(fmt, struct c2c_fmt, fmt);
396 if (dim == &dim_symbol || dim == &dim_srcline)
397 return symbol_width(hists, dim->se);
399 return dim->se ? hists__col_len(hists, dim->se->se_width_idx) :
403 static int c2c_header(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
404 struct hists *hists, int line, int *span)
406 struct perf_hpp_list *hpp_list = hists->hpp_list;
407 struct c2c_fmt *c2c_fmt;
408 struct c2c_dimension *dim;
409 const char *text = NULL;
410 int width = c2c_width(fmt, hpp, hists);
412 c2c_fmt = container_of(fmt, struct c2c_fmt, fmt);
416 text = dim->header.line[line].text;
417 /* Use the last line from sort_entry if not defined. */
418 if (!text && (line == hpp_list->nr_header_lines - 1))
419 text = dim->se->se_header;
421 text = dim->header.line[line].text;
427 *span = dim->header.line[line].span;
434 return scnprintf(hpp->buf, hpp->size, "%*s", width, text);
437 #define HEX_STR(__s, __v) \
439 scnprintf(__s, sizeof(__s), "0x%" PRIx64, __v); \
444 dcacheline_cmp(struct perf_hpp_fmt *fmt __maybe_unused,
445 struct hist_entry *left, struct hist_entry *right)
447 return sort__dcacheline_cmp(left, right);
450 static int dcacheline_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
451 struct hist_entry *he)
454 int width = c2c_width(fmt, hpp, he->hists);
458 addr = cl_address(he->mem_info->daddr.addr);
460 return scnprintf(hpp->buf, hpp->size, "%*s", width, HEX_STR(buf, addr));
463 static int offset_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
464 struct hist_entry *he)
467 int width = c2c_width(fmt, hpp, he->hists);
471 addr = cl_offset(he->mem_info->daddr.al_addr);
473 return scnprintf(hpp->buf, hpp->size, "%*s", width, HEX_STR(buf, addr));
477 offset_cmp(struct perf_hpp_fmt *fmt __maybe_unused,
478 struct hist_entry *left, struct hist_entry *right)
480 uint64_t l = 0, r = 0;
483 l = cl_offset(left->mem_info->daddr.addr);
485 r = cl_offset(right->mem_info->daddr.addr);
487 return (int64_t)(r - l);
491 iaddr_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
492 struct hist_entry *he)
495 int width = c2c_width(fmt, hpp, he->hists);
499 addr = he->mem_info->iaddr.addr;
501 return scnprintf(hpp->buf, hpp->size, "%*s", width, HEX_STR(buf, addr));
505 iaddr_cmp(struct perf_hpp_fmt *fmt __maybe_unused,
506 struct hist_entry *left, struct hist_entry *right)
508 return sort__iaddr_cmp(left, right);
512 tot_hitm_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
513 struct hist_entry *he)
515 struct c2c_hist_entry *c2c_he;
516 int width = c2c_width(fmt, hpp, he->hists);
517 unsigned int tot_hitm;
519 c2c_he = container_of(he, struct c2c_hist_entry, he);
520 tot_hitm = c2c_he->stats.lcl_hitm + c2c_he->stats.rmt_hitm;
522 return scnprintf(hpp->buf, hpp->size, "%*u", width, tot_hitm);
526 tot_hitm_cmp(struct perf_hpp_fmt *fmt __maybe_unused,
527 struct hist_entry *left, struct hist_entry *right)
529 struct c2c_hist_entry *c2c_left;
530 struct c2c_hist_entry *c2c_right;
531 unsigned int tot_hitm_left;
532 unsigned int tot_hitm_right;
534 c2c_left = container_of(left, struct c2c_hist_entry, he);
535 c2c_right = container_of(right, struct c2c_hist_entry, he);
537 tot_hitm_left = c2c_left->stats.lcl_hitm + c2c_left->stats.rmt_hitm;
538 tot_hitm_right = c2c_right->stats.lcl_hitm + c2c_right->stats.rmt_hitm;
540 return tot_hitm_left - tot_hitm_right;
543 #define STAT_FN_ENTRY(__f) \
545 __f ## _entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp, \
546 struct hist_entry *he) \
548 struct c2c_hist_entry *c2c_he; \
549 int width = c2c_width(fmt, hpp, he->hists); \
551 c2c_he = container_of(he, struct c2c_hist_entry, he); \
552 return scnprintf(hpp->buf, hpp->size, "%*u", width, \
553 c2c_he->stats.__f); \
556 #define STAT_FN_CMP(__f) \
558 __f ## _cmp(struct perf_hpp_fmt *fmt __maybe_unused, \
559 struct hist_entry *left, struct hist_entry *right) \
561 struct c2c_hist_entry *c2c_left, *c2c_right; \
563 c2c_left = container_of(left, struct c2c_hist_entry, he); \
564 c2c_right = container_of(right, struct c2c_hist_entry, he); \
565 return c2c_left->stats.__f - c2c_right->stats.__f; \
568 #define STAT_FN(__f) \
583 static uint64_t llc_miss(struct c2c_stats *stats)
587 llcmiss = stats->lcl_dram +
596 ld_llcmiss_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
597 struct hist_entry *he)
599 struct c2c_hist_entry *c2c_he;
600 int width = c2c_width(fmt, hpp, he->hists);
602 c2c_he = container_of(he, struct c2c_hist_entry, he);
604 return scnprintf(hpp->buf, hpp->size, "%*lu", width,
605 llc_miss(&c2c_he->stats));
609 ld_llcmiss_cmp(struct perf_hpp_fmt *fmt __maybe_unused,
610 struct hist_entry *left, struct hist_entry *right)
612 struct c2c_hist_entry *c2c_left;
613 struct c2c_hist_entry *c2c_right;
615 c2c_left = container_of(left, struct c2c_hist_entry, he);
616 c2c_right = container_of(right, struct c2c_hist_entry, he);
618 return llc_miss(&c2c_left->stats) - llc_miss(&c2c_right->stats);
621 static uint64_t total_records(struct c2c_stats *stats)
623 uint64_t lclmiss, ldcnt, total;
625 lclmiss = stats->lcl_dram +
645 tot_recs_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
646 struct hist_entry *he)
648 struct c2c_hist_entry *c2c_he;
649 int width = c2c_width(fmt, hpp, he->hists);
652 c2c_he = container_of(he, struct c2c_hist_entry, he);
653 tot_recs = total_records(&c2c_he->stats);
655 return scnprintf(hpp->buf, hpp->size, "%*" PRIu64, width, tot_recs);
659 tot_recs_cmp(struct perf_hpp_fmt *fmt __maybe_unused,
660 struct hist_entry *left, struct hist_entry *right)
662 struct c2c_hist_entry *c2c_left;
663 struct c2c_hist_entry *c2c_right;
664 uint64_t tot_recs_left;
665 uint64_t tot_recs_right;
667 c2c_left = container_of(left, struct c2c_hist_entry, he);
668 c2c_right = container_of(right, struct c2c_hist_entry, he);
670 tot_recs_left = total_records(&c2c_left->stats);
671 tot_recs_right = total_records(&c2c_right->stats);
673 return tot_recs_left - tot_recs_right;
676 static uint64_t total_loads(struct c2c_stats *stats)
678 uint64_t lclmiss, ldcnt;
680 lclmiss = stats->lcl_dram +
696 tot_loads_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
697 struct hist_entry *he)
699 struct c2c_hist_entry *c2c_he;
700 int width = c2c_width(fmt, hpp, he->hists);
703 c2c_he = container_of(he, struct c2c_hist_entry, he);
704 tot_recs = total_loads(&c2c_he->stats);
706 return scnprintf(hpp->buf, hpp->size, "%*" PRIu64, width, tot_recs);
710 tot_loads_cmp(struct perf_hpp_fmt *fmt __maybe_unused,
711 struct hist_entry *left, struct hist_entry *right)
713 struct c2c_hist_entry *c2c_left;
714 struct c2c_hist_entry *c2c_right;
715 uint64_t tot_recs_left;
716 uint64_t tot_recs_right;
718 c2c_left = container_of(left, struct c2c_hist_entry, he);
719 c2c_right = container_of(right, struct c2c_hist_entry, he);
721 tot_recs_left = total_loads(&c2c_left->stats);
722 tot_recs_right = total_loads(&c2c_right->stats);
724 return tot_recs_left - tot_recs_right;
727 typedef double (get_percent_cb)(struct c2c_hist_entry *);
730 percent_color(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
731 struct hist_entry *he, get_percent_cb get_percent)
733 struct c2c_hist_entry *c2c_he;
734 int width = c2c_width(fmt, hpp, he->hists);
737 c2c_he = container_of(he, struct c2c_hist_entry, he);
738 per = get_percent(c2c_he);
740 #ifdef HAVE_SLANG_SUPPORT
742 return __hpp__slsmg_color_printf(hpp, "%*.2f%%", width - 1, per);
744 return hpp_color_scnprintf(hpp, "%*.2f%%", width - 1, per);
747 static double percent_hitm(struct c2c_hist_entry *c2c_he)
749 struct c2c_hists *hists;
750 struct c2c_stats *stats;
751 struct c2c_stats *total;
755 hists = container_of(c2c_he->he.hists, struct c2c_hists, hists);
756 stats = &c2c_he->stats;
757 total = &hists->stats;
759 switch (c2c.display) {
761 st = stats->rmt_hitm;
762 tot = total->rmt_hitm;
765 st = stats->lcl_hitm;
766 tot = total->lcl_hitm;
769 st = stats->tot_hitm;
770 tot = total->tot_hitm;
775 p = tot ? (double) st / tot : 0;
780 #define PERC_STR(__s, __v) \
782 scnprintf(__s, sizeof(__s), "%.2F%%", __v); \
787 percent_hitm_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
788 struct hist_entry *he)
790 struct c2c_hist_entry *c2c_he;
791 int width = c2c_width(fmt, hpp, he->hists);
795 c2c_he = container_of(he, struct c2c_hist_entry, he);
796 per = percent_hitm(c2c_he);
797 return scnprintf(hpp->buf, hpp->size, "%*s", width, PERC_STR(buf, per));
801 percent_hitm_color(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
802 struct hist_entry *he)
804 return percent_color(fmt, hpp, he, percent_hitm);
808 percent_hitm_cmp(struct perf_hpp_fmt *fmt __maybe_unused,
809 struct hist_entry *left, struct hist_entry *right)
811 struct c2c_hist_entry *c2c_left;
812 struct c2c_hist_entry *c2c_right;
816 c2c_left = container_of(left, struct c2c_hist_entry, he);
817 c2c_right = container_of(right, struct c2c_hist_entry, he);
819 per_left = percent_hitm(c2c_left);
820 per_right = percent_hitm(c2c_right);
822 return per_left - per_right;
825 static struct c2c_stats *he_stats(struct hist_entry *he)
827 struct c2c_hist_entry *c2c_he;
829 c2c_he = container_of(he, struct c2c_hist_entry, he);
830 return &c2c_he->stats;
833 static struct c2c_stats *total_stats(struct hist_entry *he)
835 struct c2c_hists *hists;
837 hists = container_of(he->hists, struct c2c_hists, hists);
838 return &hists->stats;
841 static double percent(int st, int tot)
843 return tot ? 100. * (double) st / (double) tot : 0;
846 #define PERCENT(__h, __f) percent(he_stats(__h)->__f, total_stats(__h)->__f)
848 #define PERCENT_FN(__f) \
849 static double percent_ ## __f(struct c2c_hist_entry *c2c_he) \
851 struct c2c_hists *hists; \
853 hists = container_of(c2c_he->he.hists, struct c2c_hists, hists); \
854 return percent(c2c_he->stats.__f, hists->stats.__f); \
860 PERCENT_FN(st_l1miss)
863 percent_rmt_hitm_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
864 struct hist_entry *he)
866 int width = c2c_width(fmt, hpp, he->hists);
867 double per = PERCENT(he, rmt_hitm);
870 return scnprintf(hpp->buf, hpp->size, "%*s", width, PERC_STR(buf, per));
874 percent_rmt_hitm_color(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
875 struct hist_entry *he)
877 return percent_color(fmt, hpp, he, percent_rmt_hitm);
881 percent_rmt_hitm_cmp(struct perf_hpp_fmt *fmt __maybe_unused,
882 struct hist_entry *left, struct hist_entry *right)
887 per_left = PERCENT(left, lcl_hitm);
888 per_right = PERCENT(right, lcl_hitm);
890 return per_left - per_right;
894 percent_lcl_hitm_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
895 struct hist_entry *he)
897 int width = c2c_width(fmt, hpp, he->hists);
898 double per = PERCENT(he, lcl_hitm);
901 return scnprintf(hpp->buf, hpp->size, "%*s", width, PERC_STR(buf, per));
905 percent_lcl_hitm_color(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
906 struct hist_entry *he)
908 return percent_color(fmt, hpp, he, percent_lcl_hitm);
912 percent_lcl_hitm_cmp(struct perf_hpp_fmt *fmt __maybe_unused,
913 struct hist_entry *left, struct hist_entry *right)
918 per_left = PERCENT(left, lcl_hitm);
919 per_right = PERCENT(right, lcl_hitm);
921 return per_left - per_right;
925 percent_stores_l1hit_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
926 struct hist_entry *he)
928 int width = c2c_width(fmt, hpp, he->hists);
929 double per = PERCENT(he, st_l1hit);
932 return scnprintf(hpp->buf, hpp->size, "%*s", width, PERC_STR(buf, per));
936 percent_stores_l1hit_color(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
937 struct hist_entry *he)
939 return percent_color(fmt, hpp, he, percent_st_l1hit);
943 percent_stores_l1hit_cmp(struct perf_hpp_fmt *fmt __maybe_unused,
944 struct hist_entry *left, struct hist_entry *right)
949 per_left = PERCENT(left, st_l1hit);
950 per_right = PERCENT(right, st_l1hit);
952 return per_left - per_right;
956 percent_stores_l1miss_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
957 struct hist_entry *he)
959 int width = c2c_width(fmt, hpp, he->hists);
960 double per = PERCENT(he, st_l1miss);
963 return scnprintf(hpp->buf, hpp->size, "%*s", width, PERC_STR(buf, per));
967 percent_stores_l1miss_color(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
968 struct hist_entry *he)
970 return percent_color(fmt, hpp, he, percent_st_l1miss);
974 percent_stores_l1miss_cmp(struct perf_hpp_fmt *fmt __maybe_unused,
975 struct hist_entry *left, struct hist_entry *right)
980 per_left = PERCENT(left, st_l1miss);
981 per_right = PERCENT(right, st_l1miss);
983 return per_left - per_right;
990 pid_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
991 struct hist_entry *he)
993 int width = c2c_width(fmt, hpp, he->hists);
995 return scnprintf(hpp->buf, hpp->size, "%*d", width, he->thread->pid_);
999 pid_cmp(struct perf_hpp_fmt *fmt __maybe_unused,
1000 struct hist_entry *left, struct hist_entry *right)
1002 return left->thread->pid_ - right->thread->pid_;
1006 empty_cmp(struct perf_hpp_fmt *fmt __maybe_unused,
1007 struct hist_entry *left __maybe_unused,
1008 struct hist_entry *right __maybe_unused)
1014 node_entry(struct perf_hpp_fmt *fmt __maybe_unused, struct perf_hpp *hpp,
1015 struct hist_entry *he)
1017 struct c2c_hist_entry *c2c_he;
1022 c2c_he = container_of(he, struct c2c_hist_entry, he);
1024 for (node = 0; node < c2c.nodes_cnt; node++) {
1025 DECLARE_BITMAP(set, c2c.cpus_cnt);
1027 bitmap_zero(set, c2c.cpus_cnt);
1028 bitmap_and(set, c2c_he->cpuset, c2c.nodes[node], c2c.cpus_cnt);
1030 if (!bitmap_weight(set, c2c.cpus_cnt)) {
1031 if (c2c.node_info == 1) {
1032 ret = scnprintf(hpp->buf, hpp->size, "%21s", " ");
1033 advance_hpp(hpp, ret);
1039 ret = scnprintf(hpp->buf, hpp->size, " ");
1040 advance_hpp(hpp, ret);
1043 switch (c2c.node_info) {
1045 ret = scnprintf(hpp->buf, hpp->size, "%2d", node);
1046 advance_hpp(hpp, ret);
1050 int num = bitmap_weight(c2c_he->cpuset, c2c.cpus_cnt);
1051 struct c2c_stats *stats = &c2c_he->node_stats[node];
1053 ret = scnprintf(hpp->buf, hpp->size, "%2d{%2d ", node, num);
1054 advance_hpp(hpp, ret);
1056 #define DISPLAY_HITM(__h) \
1057 if (c2c_he->stats.__h> 0) { \
1058 ret = scnprintf(hpp->buf, hpp->size, "%5.1f%% ", \
1059 percent(stats->__h, c2c_he->stats.__h));\
1061 ret = scnprintf(hpp->buf, hpp->size, "%6s ", "n/a"); \
1064 switch (c2c.display) {
1066 DISPLAY_HITM(rmt_hitm);
1069 DISPLAY_HITM(lcl_hitm);
1072 DISPLAY_HITM(tot_hitm);
1079 advance_hpp(hpp, ret);
1081 if (c2c_he->stats.store > 0) {
1082 ret = scnprintf(hpp->buf, hpp->size, "%5.1f%%}",
1083 percent(stats->store, c2c_he->stats.store));
1085 ret = scnprintf(hpp->buf, hpp->size, "%6s}", "n/a");
1088 advance_hpp(hpp, ret);
1092 ret = scnprintf(hpp->buf, hpp->size, "%2d{", node);
1093 advance_hpp(hpp, ret);
1095 ret = bitmap_scnprintf(set, c2c.cpus_cnt, hpp->buf, hpp->size);
1096 advance_hpp(hpp, ret);
1098 ret = scnprintf(hpp->buf, hpp->size, "}");
1099 advance_hpp(hpp, ret);
1112 mean_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
1113 struct hist_entry *he, double mean)
1115 int width = c2c_width(fmt, hpp, he->hists);
1118 scnprintf(buf, 10, "%6.0f", mean);
1119 return scnprintf(hpp->buf, hpp->size, "%*s", width, buf);
1122 #define MEAN_ENTRY(__func, __val) \
1124 __func(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp, struct hist_entry *he) \
1126 struct c2c_hist_entry *c2c_he; \
1127 c2c_he = container_of(he, struct c2c_hist_entry, he); \
1128 return mean_entry(fmt, hpp, he, avg_stats(&c2c_he->cstats.__val)); \
1131 MEAN_ENTRY(mean_rmt_entry, rmt_hitm);
1132 MEAN_ENTRY(mean_lcl_entry, lcl_hitm);
1133 MEAN_ENTRY(mean_load_entry, load);
1136 cpucnt_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
1137 struct hist_entry *he)
1139 struct c2c_hist_entry *c2c_he;
1140 int width = c2c_width(fmt, hpp, he->hists);
1143 c2c_he = container_of(he, struct c2c_hist_entry, he);
1145 scnprintf(buf, 10, "%d", bitmap_weight(c2c_he->cpuset, c2c.cpus_cnt));
1146 return scnprintf(hpp->buf, hpp->size, "%*s", width, buf);
1150 cl_idx_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
1151 struct hist_entry *he)
1153 struct c2c_hist_entry *c2c_he;
1154 int width = c2c_width(fmt, hpp, he->hists);
1157 c2c_he = container_of(he, struct c2c_hist_entry, he);
1159 scnprintf(buf, 10, "%u", c2c_he->cacheline_idx);
1160 return scnprintf(hpp->buf, hpp->size, "%*s", width, buf);
1164 cl_idx_empty_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
1165 struct hist_entry *he)
1167 int width = c2c_width(fmt, hpp, he->hists);
1169 return scnprintf(hpp->buf, hpp->size, "%*s", width, "");
1172 #define HEADER_LOW(__h) \
1179 #define HEADER_BOTH(__h0, __h1) \
1189 #define HEADER_SPAN(__h0, __h1, __s) \
1200 #define HEADER_SPAN_LOW(__h) \
1207 static struct c2c_dimension dim_dcacheline = {
1208 .header = HEADER_LOW("Cacheline"),
1209 .name = "dcacheline",
1210 .cmp = dcacheline_cmp,
1211 .entry = dcacheline_entry,
1215 static struct c2c_header header_offset_tui = HEADER_LOW("Off");
1217 static struct c2c_dimension dim_offset = {
1218 .header = HEADER_BOTH("Data address", "Offset"),
1221 .entry = offset_entry,
1225 static struct c2c_dimension dim_iaddr = {
1226 .header = HEADER_LOW("Code address"),
1229 .entry = iaddr_entry,
1233 static struct c2c_dimension dim_tot_hitm = {
1234 .header = HEADER_SPAN("----- LLC Load Hitm -----", "Total", 2),
1236 .cmp = tot_hitm_cmp,
1237 .entry = tot_hitm_entry,
1241 static struct c2c_dimension dim_lcl_hitm = {
1242 .header = HEADER_SPAN_LOW("Lcl"),
1244 .cmp = lcl_hitm_cmp,
1245 .entry = lcl_hitm_entry,
1249 static struct c2c_dimension dim_rmt_hitm = {
1250 .header = HEADER_SPAN_LOW("Rmt"),
1252 .cmp = rmt_hitm_cmp,
1253 .entry = rmt_hitm_entry,
1257 static struct c2c_dimension dim_cl_rmt_hitm = {
1258 .header = HEADER_SPAN("----- HITM -----", "Rmt", 1),
1259 .name = "cl_rmt_hitm",
1260 .cmp = rmt_hitm_cmp,
1261 .entry = rmt_hitm_entry,
1265 static struct c2c_dimension dim_cl_lcl_hitm = {
1266 .header = HEADER_SPAN_LOW("Lcl"),
1267 .name = "cl_lcl_hitm",
1268 .cmp = lcl_hitm_cmp,
1269 .entry = lcl_hitm_entry,
1273 static struct c2c_dimension dim_stores = {
1274 .header = HEADER_SPAN("---- Store Reference ----", "Total", 2),
1277 .entry = store_entry,
1281 static struct c2c_dimension dim_stores_l1hit = {
1282 .header = HEADER_SPAN_LOW("L1Hit"),
1283 .name = "stores_l1hit",
1284 .cmp = st_l1hit_cmp,
1285 .entry = st_l1hit_entry,
1289 static struct c2c_dimension dim_stores_l1miss = {
1290 .header = HEADER_SPAN_LOW("L1Miss"),
1291 .name = "stores_l1miss",
1292 .cmp = st_l1miss_cmp,
1293 .entry = st_l1miss_entry,
1297 static struct c2c_dimension dim_cl_stores_l1hit = {
1298 .header = HEADER_SPAN("-- Store Refs --", "L1 Hit", 1),
1299 .name = "cl_stores_l1hit",
1300 .cmp = st_l1hit_cmp,
1301 .entry = st_l1hit_entry,
1305 static struct c2c_dimension dim_cl_stores_l1miss = {
1306 .header = HEADER_SPAN_LOW("L1 Miss"),
1307 .name = "cl_stores_l1miss",
1308 .cmp = st_l1miss_cmp,
1309 .entry = st_l1miss_entry,
1313 static struct c2c_dimension dim_ld_fbhit = {
1314 .header = HEADER_SPAN("----- Core Load Hit -----", "FB", 2),
1316 .cmp = ld_fbhit_cmp,
1317 .entry = ld_fbhit_entry,
1321 static struct c2c_dimension dim_ld_l1hit = {
1322 .header = HEADER_SPAN_LOW("L1"),
1324 .cmp = ld_l1hit_cmp,
1325 .entry = ld_l1hit_entry,
1329 static struct c2c_dimension dim_ld_l2hit = {
1330 .header = HEADER_SPAN_LOW("L2"),
1332 .cmp = ld_l2hit_cmp,
1333 .entry = ld_l2hit_entry,
1337 static struct c2c_dimension dim_ld_llchit = {
1338 .header = HEADER_SPAN("-- LLC Load Hit --", "Llc", 1),
1339 .name = "ld_lclhit",
1340 .cmp = ld_llchit_cmp,
1341 .entry = ld_llchit_entry,
1345 static struct c2c_dimension dim_ld_rmthit = {
1346 .header = HEADER_SPAN_LOW("Rmt"),
1347 .name = "ld_rmthit",
1349 .entry = rmt_hit_entry,
1353 static struct c2c_dimension dim_ld_llcmiss = {
1354 .header = HEADER_BOTH("LLC", "Ld Miss"),
1355 .name = "ld_llcmiss",
1356 .cmp = ld_llcmiss_cmp,
1357 .entry = ld_llcmiss_entry,
1361 static struct c2c_dimension dim_tot_recs = {
1362 .header = HEADER_BOTH("Total", "records"),
1364 .cmp = tot_recs_cmp,
1365 .entry = tot_recs_entry,
1369 static struct c2c_dimension dim_tot_loads = {
1370 .header = HEADER_BOTH("Total", "Loads"),
1371 .name = "tot_loads",
1372 .cmp = tot_loads_cmp,
1373 .entry = tot_loads_entry,
1377 static struct c2c_header percent_hitm_header[] = {
1378 [DISPLAY_LCL] = HEADER_BOTH("Lcl", "Hitm"),
1379 [DISPLAY_RMT] = HEADER_BOTH("Rmt", "Hitm"),
1380 [DISPLAY_TOT] = HEADER_BOTH("Tot", "Hitm"),
1383 static struct c2c_dimension dim_percent_hitm = {
1384 .name = "percent_hitm",
1385 .cmp = percent_hitm_cmp,
1386 .entry = percent_hitm_entry,
1387 .color = percent_hitm_color,
1391 static struct c2c_dimension dim_percent_rmt_hitm = {
1392 .header = HEADER_SPAN("----- HITM -----", "Rmt", 1),
1393 .name = "percent_rmt_hitm",
1394 .cmp = percent_rmt_hitm_cmp,
1395 .entry = percent_rmt_hitm_entry,
1396 .color = percent_rmt_hitm_color,
1400 static struct c2c_dimension dim_percent_lcl_hitm = {
1401 .header = HEADER_SPAN_LOW("Lcl"),
1402 .name = "percent_lcl_hitm",
1403 .cmp = percent_lcl_hitm_cmp,
1404 .entry = percent_lcl_hitm_entry,
1405 .color = percent_lcl_hitm_color,
1409 static struct c2c_dimension dim_percent_stores_l1hit = {
1410 .header = HEADER_SPAN("-- Store Refs --", "L1 Hit", 1),
1411 .name = "percent_stores_l1hit",
1412 .cmp = percent_stores_l1hit_cmp,
1413 .entry = percent_stores_l1hit_entry,
1414 .color = percent_stores_l1hit_color,
1418 static struct c2c_dimension dim_percent_stores_l1miss = {
1419 .header = HEADER_SPAN_LOW("L1 Miss"),
1420 .name = "percent_stores_l1miss",
1421 .cmp = percent_stores_l1miss_cmp,
1422 .entry = percent_stores_l1miss_entry,
1423 .color = percent_stores_l1miss_color,
1427 static struct c2c_dimension dim_dram_lcl = {
1428 .header = HEADER_SPAN("--- Load Dram ----", "Lcl", 1),
1430 .cmp = lcl_dram_cmp,
1431 .entry = lcl_dram_entry,
1435 static struct c2c_dimension dim_dram_rmt = {
1436 .header = HEADER_SPAN_LOW("Rmt"),
1438 .cmp = rmt_dram_cmp,
1439 .entry = rmt_dram_entry,
1443 static struct c2c_dimension dim_pid = {
1444 .header = HEADER_LOW("Pid"),
1451 static struct c2c_dimension dim_tid = {
1452 .header = HEADER_LOW("Tid"),
1457 static struct c2c_dimension dim_symbol = {
1462 static struct c2c_dimension dim_dso = {
1463 .header = HEADER_BOTH("Shared", "Object"),
1468 static struct c2c_header header_node[3] = {
1470 HEADER_LOW("Node{cpus %hitms %stores}"),
1471 HEADER_LOW("Node{cpu list}"),
1474 static struct c2c_dimension dim_node = {
1477 .entry = node_entry,
1481 static struct c2c_dimension dim_mean_rmt = {
1482 .header = HEADER_SPAN("---------- cycles ----------", "rmt hitm", 2),
1485 .entry = mean_rmt_entry,
1489 static struct c2c_dimension dim_mean_lcl = {
1490 .header = HEADER_SPAN_LOW("lcl hitm"),
1493 .entry = mean_lcl_entry,
1497 static struct c2c_dimension dim_mean_load = {
1498 .header = HEADER_SPAN_LOW("load"),
1499 .name = "mean_load",
1501 .entry = mean_load_entry,
1505 static struct c2c_dimension dim_cpucnt = {
1506 .header = HEADER_BOTH("cpu", "cnt"),
1509 .entry = cpucnt_entry,
1513 static struct c2c_dimension dim_srcline = {
1514 .name = "cl_srcline",
1515 .se = &sort_srcline,
1518 static struct c2c_dimension dim_dcacheline_idx = {
1519 .header = HEADER_LOW("Index"),
1522 .entry = cl_idx_entry,
1526 static struct c2c_dimension dim_dcacheline_num = {
1527 .header = HEADER_LOW("Num"),
1530 .entry = cl_idx_entry,
1534 static struct c2c_dimension dim_dcacheline_num_empty = {
1535 .header = HEADER_LOW("Num"),
1536 .name = "cl_num_empty",
1538 .entry = cl_idx_empty_entry,
1542 static struct c2c_dimension *dimensions[] = {
1554 &dim_cl_stores_l1hit,
1555 &dim_cl_stores_l1miss,
1565 &dim_percent_rmt_hitm,
1566 &dim_percent_lcl_hitm,
1567 &dim_percent_stores_l1hit,
1568 &dim_percent_stores_l1miss,
1581 &dim_dcacheline_idx,
1582 &dim_dcacheline_num,
1583 &dim_dcacheline_num_empty,
1587 static void fmt_free(struct perf_hpp_fmt *fmt)
1589 struct c2c_fmt *c2c_fmt;
1591 c2c_fmt = container_of(fmt, struct c2c_fmt, fmt);
1595 static bool fmt_equal(struct perf_hpp_fmt *a, struct perf_hpp_fmt *b)
1597 struct c2c_fmt *c2c_a = container_of(a, struct c2c_fmt, fmt);
1598 struct c2c_fmt *c2c_b = container_of(b, struct c2c_fmt, fmt);
1600 return c2c_a->dim == c2c_b->dim;
1603 static struct c2c_dimension *get_dimension(const char *name)
1607 for (i = 0; dimensions[i]; i++) {
1608 struct c2c_dimension *dim = dimensions[i];
1610 if (!strcmp(dim->name, name))
1617 static int c2c_se_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
1618 struct hist_entry *he)
1620 struct c2c_fmt *c2c_fmt = container_of(fmt, struct c2c_fmt, fmt);
1621 struct c2c_dimension *dim = c2c_fmt->dim;
1622 size_t len = fmt->user_len;
1625 len = hists__col_len(he->hists, dim->se->se_width_idx);
1627 if (dim == &dim_symbol || dim == &dim_srcline)
1628 len = symbol_width(he->hists, dim->se);
1631 return dim->se->se_snprintf(he, hpp->buf, hpp->size, len);
1634 static int64_t c2c_se_cmp(struct perf_hpp_fmt *fmt,
1635 struct hist_entry *a, struct hist_entry *b)
1637 struct c2c_fmt *c2c_fmt = container_of(fmt, struct c2c_fmt, fmt);
1638 struct c2c_dimension *dim = c2c_fmt->dim;
1640 return dim->se->se_cmp(a, b);
1643 static int64_t c2c_se_collapse(struct perf_hpp_fmt *fmt,
1644 struct hist_entry *a, struct hist_entry *b)
1646 struct c2c_fmt *c2c_fmt = container_of(fmt, struct c2c_fmt, fmt);
1647 struct c2c_dimension *dim = c2c_fmt->dim;
1648 int64_t (*collapse_fn)(struct hist_entry *, struct hist_entry *);
1650 collapse_fn = dim->se->se_collapse ?: dim->se->se_cmp;
1651 return collapse_fn(a, b);
1654 static struct c2c_fmt *get_format(const char *name)
1656 struct c2c_dimension *dim = get_dimension(name);
1657 struct c2c_fmt *c2c_fmt;
1658 struct perf_hpp_fmt *fmt;
1663 c2c_fmt = zalloc(sizeof(*c2c_fmt));
1669 fmt = &c2c_fmt->fmt;
1670 INIT_LIST_HEAD(&fmt->list);
1671 INIT_LIST_HEAD(&fmt->sort_list);
1673 fmt->cmp = dim->se ? c2c_se_cmp : dim->cmp;
1674 fmt->sort = dim->se ? c2c_se_cmp : dim->cmp;
1675 fmt->color = dim->se ? NULL : dim->color;
1676 fmt->entry = dim->se ? c2c_se_entry : dim->entry;
1677 fmt->header = c2c_header;
1678 fmt->width = c2c_width;
1679 fmt->collapse = dim->se ? c2c_se_collapse : dim->cmp;
1680 fmt->equal = fmt_equal;
1681 fmt->free = fmt_free;
1686 static int c2c_hists__init_output(struct perf_hpp_list *hpp_list, char *name)
1688 struct c2c_fmt *c2c_fmt = get_format(name);
1692 return output_field_add(hpp_list, name);
1695 perf_hpp_list__column_register(hpp_list, &c2c_fmt->fmt);
1699 static int c2c_hists__init_sort(struct perf_hpp_list *hpp_list, char *name)
1701 struct c2c_fmt *c2c_fmt = get_format(name);
1702 struct c2c_dimension *dim;
1706 return sort_dimension__add(hpp_list, name, NULL, 0);
1710 if (dim == &dim_dso)
1713 perf_hpp_list__register_sort_field(hpp_list, &c2c_fmt->fmt);
1717 #define PARSE_LIST(_list, _fn) \
1725 for (tok = strtok_r((char *)_list, ", ", &tmp); \
1726 tok; tok = strtok_r(NULL, ", ", &tmp)) { \
1727 ret = _fn(hpp_list, tok); \
1728 if (ret == -EINVAL) { \
1729 pr_err("Invalid --fields key: `%s'", tok); \
1731 } else if (ret == -ESRCH) { \
1732 pr_err("Unknown --fields key: `%s'", tok); \
1738 static int hpp_list__parse(struct perf_hpp_list *hpp_list,
1739 const char *output_,
1742 char *output = output_ ? strdup(output_) : NULL;
1743 char *sort = sort_ ? strdup(sort_) : NULL;
1746 PARSE_LIST(output, c2c_hists__init_output);
1747 PARSE_LIST(sort, c2c_hists__init_sort);
1749 /* copy sort keys to output fields */
1750 perf_hpp__setup_output_field(hpp_list);
1753 * We dont need other sorting keys other than those
1754 * we already specified. It also really slows down
1755 * the processing a lot with big number of output
1756 * fields, so switching this off for c2c.
1760 /* and then copy output fields to sort keys */
1761 perf_hpp__append_sort_keys(&hists->list);
1769 static int c2c_hists__init(struct c2c_hists *hists,
1771 int nr_header_lines)
1773 __hists__init(&hists->hists, &hists->list);
1776 * Initialize only with sort fields, we need to resort
1777 * later anyway, and that's where we add output fields
1780 perf_hpp_list__init(&hists->list);
1782 /* Overload number of header lines.*/
1783 hists->list.nr_header_lines = nr_header_lines;
1785 return hpp_list__parse(&hists->list, NULL, sort);
1788 static int c2c_hists__reinit(struct c2c_hists *c2c_hists,
1792 perf_hpp__reset_output_field(&c2c_hists->list);
1793 return hpp_list__parse(&c2c_hists->list, output, sort);
1796 #define DISPLAY_LINE_LIMIT 0.0005
1798 static bool he__display(struct hist_entry *he, struct c2c_stats *stats)
1800 struct c2c_hist_entry *c2c_he;
1806 c2c_he = container_of(he, struct c2c_hist_entry, he);
1808 #define FILTER_HITM(__h) \
1810 ld_dist = ((double)c2c_he->stats.__h / stats->__h); \
1811 if (ld_dist < DISPLAY_LINE_LIMIT) \
1812 he->filtered = HIST_FILTER__C2C; \
1814 he->filtered = HIST_FILTER__C2C; \
1817 switch (c2c.display) {
1819 FILTER_HITM(lcl_hitm);
1822 FILTER_HITM(rmt_hitm);
1825 FILTER_HITM(tot_hitm);
1832 return he->filtered == 0;
1835 static inline int valid_hitm_or_store(struct hist_entry *he)
1837 struct c2c_hist_entry *c2c_he;
1840 c2c_he = container_of(he, struct c2c_hist_entry, he);
1841 has_hitm = c2c.display == DISPLAY_TOT ? c2c_he->stats.tot_hitm :
1842 c2c.display == DISPLAY_LCL ? c2c_he->stats.lcl_hitm :
1843 c2c_he->stats.rmt_hitm;
1844 return has_hitm || c2c_he->stats.store;
1847 static void calc_width(struct hist_entry *he)
1849 struct c2c_hists *c2c_hists;
1851 c2c_hists = container_of(he->hists, struct c2c_hists, hists);
1852 hists__calc_col_len(&c2c_hists->hists, he);
1855 static int filter_cb(struct hist_entry *he)
1857 if (c2c.show_src && !he->srcline)
1858 he->srcline = hist_entry__get_srcline(he);
1862 if (!valid_hitm_or_store(he))
1863 he->filtered = HIST_FILTER__C2C;
1868 static int resort_cl_cb(struct hist_entry *he)
1870 struct c2c_hist_entry *c2c_he;
1871 struct c2c_hists *c2c_hists;
1872 bool display = he__display(he, &c2c.hitm_stats);
1874 c2c_he = container_of(he, struct c2c_hist_entry, he);
1875 c2c_hists = c2c_he->hists;
1879 if (display && c2c_hists) {
1880 static unsigned int idx;
1882 c2c_he->cacheline_idx = idx++;
1884 c2c_hists__reinit(c2c_hists, c2c.cl_output, c2c.cl_resort);
1886 hists__collapse_resort(&c2c_hists->hists, NULL);
1887 hists__output_resort_cb(&c2c_hists->hists, NULL, filter_cb);
1893 static void setup_nodes_header(void)
1895 dim_node.header = header_node[c2c.node_info];
1898 static int setup_nodes(struct perf_session *session)
1900 struct numa_node *n;
1901 unsigned long **nodes;
1905 if (c2c.node_info > 2)
1908 c2c.nodes_cnt = session->header.env.nr_numa_nodes;
1909 c2c.cpus_cnt = session->header.env.nr_cpus_online;
1911 n = session->header.env.numa_nodes;
1915 nodes = zalloc(sizeof(unsigned long *) * c2c.nodes_cnt);
1921 cpu2node = zalloc(sizeof(int) * c2c.cpus_cnt);
1925 for (cpu = 0; cpu < c2c.cpus_cnt; cpu++)
1928 c2c.cpu2node = cpu2node;
1930 for (node = 0; node < c2c.nodes_cnt; node++) {
1931 struct cpu_map *map = n[node].map;
1934 set = bitmap_alloc(c2c.cpus_cnt);
1938 for (cpu = 0; cpu < map->nr; cpu++) {
1939 set_bit(map->map[cpu], set);
1941 if (WARN_ONCE(cpu2node[map->map[cpu]] != -1, "node/cpu topology bug"))
1944 cpu2node[map->map[cpu]] = node;
1950 setup_nodes_header();
1954 #define HAS_HITMS(__h) ((__h)->stats.lcl_hitm || (__h)->stats.rmt_hitm)
1956 static int resort_hitm_cb(struct hist_entry *he)
1958 struct c2c_hist_entry *c2c_he;
1959 c2c_he = container_of(he, struct c2c_hist_entry, he);
1961 if (HAS_HITMS(c2c_he)) {
1962 c2c.shared_clines++;
1963 c2c_add_stats(&c2c.hitm_stats, &c2c_he->stats);
1969 static int hists__iterate_cb(struct hists *hists, hists__resort_cb_t cb)
1971 struct rb_node *next = rb_first(&hists->entries);
1975 struct hist_entry *he;
1977 he = rb_entry(next, struct hist_entry, rb_node);
1981 next = rb_next(&he->rb_node);
1987 static void print_c2c__display_stats(FILE *out)
1990 struct c2c_stats *stats = &c2c.hists.stats;
1992 llc_misses = stats->lcl_dram +
1997 fprintf(out, "=================================================\n");
1998 fprintf(out, " Trace Event Information \n");
1999 fprintf(out, "=================================================\n");
2000 fprintf(out, " Total records : %10d\n", stats->nr_entries);
2001 fprintf(out, " Locked Load/Store Operations : %10d\n", stats->locks);
2002 fprintf(out, " Load Operations : %10d\n", stats->load);
2003 fprintf(out, " Loads - uncacheable : %10d\n", stats->ld_uncache);
2004 fprintf(out, " Loads - IO : %10d\n", stats->ld_io);
2005 fprintf(out, " Loads - Miss : %10d\n", stats->ld_miss);
2006 fprintf(out, " Loads - no mapping : %10d\n", stats->ld_noadrs);
2007 fprintf(out, " Load Fill Buffer Hit : %10d\n", stats->ld_fbhit);
2008 fprintf(out, " Load L1D hit : %10d\n", stats->ld_l1hit);
2009 fprintf(out, " Load L2D hit : %10d\n", stats->ld_l2hit);
2010 fprintf(out, " Load LLC hit : %10d\n", stats->ld_llchit + stats->lcl_hitm);
2011 fprintf(out, " Load Local HITM : %10d\n", stats->lcl_hitm);
2012 fprintf(out, " Load Remote HITM : %10d\n", stats->rmt_hitm);
2013 fprintf(out, " Load Remote HIT : %10d\n", stats->rmt_hit);
2014 fprintf(out, " Load Local DRAM : %10d\n", stats->lcl_dram);
2015 fprintf(out, " Load Remote DRAM : %10d\n", stats->rmt_dram);
2016 fprintf(out, " Load MESI State Exclusive : %10d\n", stats->ld_excl);
2017 fprintf(out, " Load MESI State Shared : %10d\n", stats->ld_shared);
2018 fprintf(out, " Load LLC Misses : %10d\n", llc_misses);
2019 fprintf(out, " LLC Misses to Local DRAM : %10.1f%%\n", ((double)stats->lcl_dram/(double)llc_misses) * 100.);
2020 fprintf(out, " LLC Misses to Remote DRAM : %10.1f%%\n", ((double)stats->rmt_dram/(double)llc_misses) * 100.);
2021 fprintf(out, " LLC Misses to Remote cache (HIT) : %10.1f%%\n", ((double)stats->rmt_hit /(double)llc_misses) * 100.);
2022 fprintf(out, " LLC Misses to Remote cache (HITM) : %10.1f%%\n", ((double)stats->rmt_hitm/(double)llc_misses) * 100.);
2023 fprintf(out, " Store Operations : %10d\n", stats->store);
2024 fprintf(out, " Store - uncacheable : %10d\n", stats->st_uncache);
2025 fprintf(out, " Store - no mapping : %10d\n", stats->st_noadrs);
2026 fprintf(out, " Store L1D Hit : %10d\n", stats->st_l1hit);
2027 fprintf(out, " Store L1D Miss : %10d\n", stats->st_l1miss);
2028 fprintf(out, " No Page Map Rejects : %10d\n", stats->nomap);
2029 fprintf(out, " Unable to parse data source : %10d\n", stats->noparse);
2032 static void print_shared_cacheline_info(FILE *out)
2034 struct c2c_stats *stats = &c2c.hitm_stats;
2035 int hitm_cnt = stats->lcl_hitm + stats->rmt_hitm;
2037 fprintf(out, "=================================================\n");
2038 fprintf(out, " Global Shared Cache Line Event Information \n");
2039 fprintf(out, "=================================================\n");
2040 fprintf(out, " Total Shared Cache Lines : %10d\n", c2c.shared_clines);
2041 fprintf(out, " Load HITs on shared lines : %10d\n", stats->load);
2042 fprintf(out, " Fill Buffer Hits on shared lines : %10d\n", stats->ld_fbhit);
2043 fprintf(out, " L1D hits on shared lines : %10d\n", stats->ld_l1hit);
2044 fprintf(out, " L2D hits on shared lines : %10d\n", stats->ld_l2hit);
2045 fprintf(out, " LLC hits on shared lines : %10d\n", stats->ld_llchit + stats->lcl_hitm);
2046 fprintf(out, " Locked Access on shared lines : %10d\n", stats->locks);
2047 fprintf(out, " Store HITs on shared lines : %10d\n", stats->store);
2048 fprintf(out, " Store L1D hits on shared lines : %10d\n", stats->st_l1hit);
2049 fprintf(out, " Total Merged records : %10d\n", hitm_cnt + stats->store);
2052 static void print_cacheline(struct c2c_hists *c2c_hists,
2053 struct hist_entry *he_cl,
2054 struct perf_hpp_list *hpp_list,
2058 struct perf_hpp hpp = {
2065 hists__fprintf_headers(&c2c_hists->hists, out);
2071 fprintf(out, " -------------------------------------------------------------\n");
2072 __hist_entry__snprintf(he_cl, &hpp, hpp_list);
2073 fprintf(out, "%s\n", bf);
2074 fprintf(out, " -------------------------------------------------------------\n");
2076 hists__fprintf(&c2c_hists->hists, false, 0, 0, 0, out, true);
2079 static void print_pareto(FILE *out)
2081 struct perf_hpp_list hpp_list;
2085 perf_hpp_list__init(&hpp_list);
2086 ret = hpp_list__parse(&hpp_list,
2095 if (WARN_ONCE(ret, "failed to setup sort entries\n"))
2098 nd = rb_first(&c2c.hists.hists.entries);
2100 for (; nd; nd = rb_next(nd)) {
2101 struct hist_entry *he = rb_entry(nd, struct hist_entry, rb_node);
2102 struct c2c_hist_entry *c2c_he;
2107 c2c_he = container_of(he, struct c2c_hist_entry, he);
2108 print_cacheline(c2c_he->hists, he, &hpp_list, out);
2112 static void print_c2c_info(FILE *out, struct perf_session *session)
2114 struct perf_evlist *evlist = session->evlist;
2115 struct perf_evsel *evsel;
2118 fprintf(out, "=================================================\n");
2119 fprintf(out, " c2c details \n");
2120 fprintf(out, "=================================================\n");
2122 evlist__for_each_entry(evlist, evsel) {
2123 fprintf(out, "%-36s: %s\n", first ? " Events" : "",
2124 perf_evsel__name(evsel));
2127 fprintf(out, " Cachelines sort on : %s HITMs\n",
2128 display_str[c2c.display]);
2129 fprintf(out, " Cacheline data grouping : %s\n", c2c.cl_sort);
2132 static void perf_c2c__hists_fprintf(FILE *out, struct perf_session *session)
2136 print_c2c__display_stats(out);
2138 print_shared_cacheline_info(out);
2140 print_c2c_info(out, session);
2146 fprintf(out, "=================================================\n");
2147 fprintf(out, " Shared Data Cache Line Table \n");
2148 fprintf(out, "=================================================\n");
2149 fprintf(out, "#\n");
2151 hists__fprintf(&c2c.hists.hists, true, 0, 0, 0, stdout, false);
2154 fprintf(out, "=================================================\n");
2155 fprintf(out, " Shared Cache Line Distribution Pareto \n");
2156 fprintf(out, "=================================================\n");
2157 fprintf(out, "#\n");
2162 #ifdef HAVE_SLANG_SUPPORT
2163 static void c2c_browser__update_nr_entries(struct hist_browser *hb)
2166 struct rb_node *nd = rb_first(&hb->hists->entries);
2169 struct hist_entry *he = rb_entry(nd, struct hist_entry, rb_node);
2177 hb->nr_non_filtered_entries = nr_entries;
2180 struct c2c_cacheline_browser {
2181 struct hist_browser hb;
2182 struct hist_entry *he;
2186 perf_c2c_cacheline_browser__title(struct hist_browser *browser,
2187 char *bf, size_t size)
2189 struct c2c_cacheline_browser *cl_browser;
2190 struct hist_entry *he;
2193 cl_browser = container_of(browser, struct c2c_cacheline_browser, hb);
2194 he = cl_browser->he;
2197 addr = cl_address(he->mem_info->daddr.addr);
2199 scnprintf(bf, size, "Cacheline 0x%lx", addr);
2203 static struct c2c_cacheline_browser*
2204 c2c_cacheline_browser__new(struct hists *hists, struct hist_entry *he)
2206 struct c2c_cacheline_browser *browser;
2208 browser = zalloc(sizeof(*browser));
2210 hist_browser__init(&browser->hb, hists);
2211 browser->hb.c2c_filter = true;
2212 browser->hb.title = perf_c2c_cacheline_browser__title;
2219 static int perf_c2c__browse_cacheline(struct hist_entry *he)
2221 struct c2c_hist_entry *c2c_he;
2222 struct c2c_hists *c2c_hists;
2223 struct c2c_cacheline_browser *cl_browser;
2224 struct hist_browser *browser;
2227 " ENTER Togle callchains (if present) \n"
2228 " n Togle Node details info \n"
2229 " s Togle full lenght of symbol and source line columns \n"
2230 " q Return back to cacheline list \n";
2232 /* Display compact version first. */
2233 c2c.symbol_full = false;
2235 c2c_he = container_of(he, struct c2c_hist_entry, he);
2236 c2c_hists = c2c_he->hists;
2238 cl_browser = c2c_cacheline_browser__new(&c2c_hists->hists, he);
2239 if (cl_browser == NULL)
2242 browser = &cl_browser->hb;
2244 /* reset abort key so that it can get Ctrl-C as a key */
2246 SLang_init_tty(0, 0, 0);
2248 c2c_browser__update_nr_entries(browser);
2251 key = hist_browser__run(browser, "? - help");
2255 c2c.symbol_full = !c2c.symbol_full;
2258 c2c.node_info = (c2c.node_info + 1) % 3;
2259 setup_nodes_header();
2264 ui_browser__help_window(&browser->b, help);
2276 static int perf_c2c_browser__title(struct hist_browser *browser,
2277 char *bf, size_t size)
2280 "Shared Data Cache Line Table "
2281 "(%lu entries, sorted on %s HITMs)",
2282 browser->nr_non_filtered_entries,
2283 display_str[c2c.display]);
2287 static struct hist_browser*
2288 perf_c2c_browser__new(struct hists *hists)
2290 struct hist_browser *browser = hist_browser__new(hists);
2293 browser->title = perf_c2c_browser__title;
2294 browser->c2c_filter = true;
2300 static int perf_c2c__hists_browse(struct hists *hists)
2302 struct hist_browser *browser;
2305 " d Display cacheline details \n"
2306 " ENTER Togle callchains (if present) \n"
2309 browser = perf_c2c_browser__new(hists);
2310 if (browser == NULL)
2313 /* reset abort key so that it can get Ctrl-C as a key */
2315 SLang_init_tty(0, 0, 0);
2317 c2c_browser__update_nr_entries(browser);
2320 key = hist_browser__run(browser, "? - help");
2326 perf_c2c__browse_cacheline(browser->he_selection);
2329 ui_browser__help_window(&browser->b, help);
2337 hist_browser__delete(browser);
2341 static void perf_c2c_display(struct perf_session *session)
2343 if (use_browser == 0)
2344 perf_c2c__hists_fprintf(stdout, session);
2346 perf_c2c__hists_browse(&c2c.hists.hists);
2349 static void perf_c2c_display(struct perf_session *session)
2352 perf_c2c__hists_fprintf(stdout, session);
2354 #endif /* HAVE_SLANG_SUPPORT */
2356 static void ui_quirks(void)
2358 if (!c2c.use_stdio) {
2359 dim_offset.width = 5;
2360 dim_offset.header = header_offset_tui;
2363 dim_percent_hitm.header = percent_hitm_header[c2c.display];
2366 #define CALLCHAIN_DEFAULT_OPT "graph,0.5,caller,function,percent"
2368 const char callchain_help[] = "Display call graph (stack chain/backtrace):\n\n"
2369 CALLCHAIN_REPORT_HELP
2370 "\n\t\t\t\tDefault: " CALLCHAIN_DEFAULT_OPT;
2373 parse_callchain_opt(const struct option *opt, const char *arg, int unset)
2375 struct callchain_param *callchain = opt->value;
2377 callchain->enabled = !unset;
2382 symbol_conf.use_callchain = false;
2383 callchain->mode = CHAIN_NONE;
2387 return parse_callchain_report_opt(arg);
2390 static int setup_callchain(struct perf_evlist *evlist)
2392 u64 sample_type = perf_evlist__combined_sample_type(evlist);
2393 enum perf_call_graph_mode mode = CALLCHAIN_NONE;
2395 if ((sample_type & PERF_SAMPLE_REGS_USER) &&
2396 (sample_type & PERF_SAMPLE_STACK_USER))
2397 mode = CALLCHAIN_DWARF;
2398 else if (sample_type & PERF_SAMPLE_BRANCH_STACK)
2399 mode = CALLCHAIN_LBR;
2400 else if (sample_type & PERF_SAMPLE_CALLCHAIN)
2401 mode = CALLCHAIN_FP;
2403 if (!callchain_param.enabled &&
2404 callchain_param.mode != CHAIN_NONE &&
2405 mode != CALLCHAIN_NONE) {
2406 symbol_conf.use_callchain = true;
2407 if (callchain_register_param(&callchain_param) < 0) {
2408 ui__error("Can't register callchain params.\n");
2413 callchain_param.record_mode = mode;
2414 callchain_param.min_percent = 0;
2418 static int setup_display(const char *str)
2420 const char *display = str ?: "tot";
2422 if (!strcmp(display, "tot"))
2423 c2c.display = DISPLAY_TOT;
2424 else if (!strcmp(display, "rmt"))
2425 c2c.display = DISPLAY_RMT;
2426 else if (!strcmp(display, "lcl"))
2427 c2c.display = DISPLAY_LCL;
2429 pr_err("failed: unknown display type: %s\n", str);
2436 #define for_each_token(__tok, __buf, __sep, __tmp) \
2437 for (__tok = strtok_r(__buf, __sep, &__tmp); __tok; \
2438 __tok = strtok_r(NULL, __sep, &__tmp))
2440 static int build_cl_output(char *cl_sort, bool no_source)
2442 char *tok, *tmp, *buf = strdup(cl_sort);
2443 bool add_pid = false;
2444 bool add_tid = false;
2445 bool add_iaddr = false;
2446 bool add_sym = false;
2447 bool add_dso = false;
2448 bool add_src = false;
2453 for_each_token(tok, buf, ",", tmp) {
2454 if (!strcmp(tok, "tid")) {
2456 } else if (!strcmp(tok, "pid")) {
2458 } else if (!strcmp(tok, "iaddr")) {
2462 add_src = no_source ? false : true;
2463 } else if (!strcmp(tok, "dso")) {
2465 } else if (strcmp(tok, "offset")) {
2466 pr_err("unrecognized sort token: %s\n", tok);
2471 if (asprintf(&c2c.cl_output,
2472 "%s%s%s%s%s%s%s%s%s%s",
2473 c2c.use_stdio ? "cl_num_empty," : "",
2476 "percent_stores_l1hit,"
2477 "percent_stores_l1miss,"
2479 add_pid ? "pid," : "",
2480 add_tid ? "tid," : "",
2481 add_iaddr ? "iaddr," : "",
2487 add_sym ? "symbol," : "",
2488 add_dso ? "dso," : "",
2489 add_src ? "cl_srcline," : "",
2493 c2c.show_src = add_src;
2499 static int setup_coalesce(const char *coalesce, bool no_source)
2501 const char *c = coalesce ?: coalesce_default;
2503 if (asprintf(&c2c.cl_sort, "offset,%s", c) < 0)
2506 if (build_cl_output(c2c.cl_sort, no_source))
2509 if (asprintf(&c2c.cl_resort, "offset,%s",
2510 c2c.display == DISPLAY_TOT ?
2512 c2c.display == DISPLAY_RMT ?
2513 "rmt_hitm,lcl_hitm" :
2514 "lcl_hitm,rmt_hitm") < 0)
2517 pr_debug("coalesce sort fields: %s\n", c2c.cl_sort);
2518 pr_debug("coalesce resort fields: %s\n", c2c.cl_resort);
2519 pr_debug("coalesce output fields: %s\n", c2c.cl_output);
2523 static int perf_c2c__report(int argc, const char **argv)
2525 struct perf_session *session;
2526 struct ui_progress prog;
2527 struct perf_data_file file = {
2528 .mode = PERF_DATA_MODE_READ,
2530 char callchain_default_opt[] = CALLCHAIN_DEFAULT_OPT;
2531 const char *display = NULL;
2532 const char *coalesce = NULL;
2533 bool no_source = false;
2534 const struct option options[] = {
2535 OPT_STRING('k', "vmlinux", &symbol_conf.vmlinux_name,
2536 "file", "vmlinux pathname"),
2537 OPT_STRING('i', "input", &input_name, "file",
2538 "the input file to process"),
2539 OPT_INCR('N', "node-info", &c2c.node_info,
2540 "show extra node info in report (repeat for more info)"),
2541 #ifdef HAVE_SLANG_SUPPORT
2542 OPT_BOOLEAN(0, "stdio", &c2c.use_stdio, "Use the stdio interface"),
2544 OPT_BOOLEAN(0, "stats", &c2c.stats_only,
2545 "Display only statistic tables (implies --stdio)"),
2546 OPT_BOOLEAN(0, "full-symbols", &c2c.symbol_full,
2547 "Display full length of symbols"),
2548 OPT_BOOLEAN(0, "no-source", &no_source,
2549 "Do not display Source Line column"),
2550 OPT_BOOLEAN(0, "show-all", &c2c.show_all,
2551 "Show all captured HITM lines."),
2552 OPT_CALLBACK_DEFAULT('g', "call-graph", &callchain_param,
2553 "print_type,threshold[,print_limit],order,sort_key[,branch],value",
2554 callchain_help, &parse_callchain_opt,
2555 callchain_default_opt),
2556 OPT_STRING('d', "display", &display, "Switch HITM output type", "lcl,rmt"),
2557 OPT_STRING('c', "coalesce", &coalesce, "coalesce fields",
2558 "coalesce fields: pid,tid,iaddr,dso"),
2559 OPT_BOOLEAN('f', "force", &symbol_conf.force, "don't complain, do it"),
2560 OPT_PARENT(c2c_options),
2565 argc = parse_options(argc, argv, options, report_c2c_usage,
2566 PARSE_OPT_STOP_AT_NON_OPTION);
2568 usage_with_options(report_c2c_usage, options);
2571 c2c.use_stdio = true;
2573 if (!input_name || !strlen(input_name))
2574 input_name = "perf.data";
2576 file.path = input_name;
2577 file.force = symbol_conf.force;
2579 err = setup_display(display);
2583 err = setup_coalesce(coalesce, no_source);
2585 pr_debug("Failed to initialize hists\n");
2589 err = c2c_hists__init(&c2c.hists, "dcacheline", 2);
2591 pr_debug("Failed to initialize hists\n");
2595 session = perf_session__new(&file, 0, &c2c.tool);
2596 if (session == NULL) {
2597 pr_debug("No memory for session\n");
2601 err = setup_nodes(session);
2603 pr_err("Failed setup nodes\n");
2607 err = setup_callchain(session->evlist);
2611 if (symbol__init(&session->header.env) < 0)
2614 /* No pipe support at the moment. */
2615 if (perf_data_file__is_pipe(session->file)) {
2616 pr_debug("No pipe support at the moment.\n");
2625 setup_browser(false);
2627 err = perf_session__process_events(session);
2629 pr_err("failed to process sample\n");
2633 c2c_hists__reinit(&c2c.hists,
2638 "tot_hitm,lcl_hitm,rmt_hitm,"
2639 "stores,stores_l1hit,stores_l1miss,"
2640 "dram_lcl,dram_rmt,"
2643 "ld_fbhit,ld_l1hit,ld_l2hit,"
2644 "ld_lclhit,ld_rmthit",
2645 c2c.display == DISPLAY_TOT ? "tot_hitm" :
2646 c2c.display == DISPLAY_LCL ? "lcl_hitm" : "rmt_hitm"
2649 ui_progress__init(&prog, c2c.hists.hists.nr_entries, "Sorting...");
2651 hists__collapse_resort(&c2c.hists.hists, NULL);
2652 hists__output_resort_cb(&c2c.hists.hists, &prog, resort_hitm_cb);
2653 hists__iterate_cb(&c2c.hists.hists, resort_cl_cb);
2655 ui_progress__finish();
2659 perf_c2c_display(session);
2662 perf_session__delete(session);
2667 static int parse_record_events(const struct option *opt,
2668 const char *str, int unset __maybe_unused)
2670 bool *event_set = (bool *) opt->value;
2673 return perf_mem_events__parse(str);
2677 static const char * const __usage_record[] = {
2678 "perf c2c record [<options>] [<command>]",
2679 "perf c2c record [<options>] -- <command> [<options>]",
2683 static const char * const *record_mem_usage = __usage_record;
2685 static int perf_c2c__record(int argc, const char **argv)
2687 int rec_argc, i = 0, j;
2688 const char **rec_argv;
2690 bool all_user = false, all_kernel = false;
2691 bool event_set = false;
2692 struct option options[] = {
2693 OPT_CALLBACK('e', "event", &event_set, "event",
2694 "event selector. Use 'perf mem record -e list' to list available events",
2695 parse_record_events),
2696 OPT_BOOLEAN('u', "all-user", &all_user, "collect only user level data"),
2697 OPT_BOOLEAN('k', "all-kernel", &all_kernel, "collect only kernel level data"),
2698 OPT_UINTEGER('l', "ldlat", &perf_mem_events__loads_ldlat, "setup mem-loads latency"),
2699 OPT_PARENT(c2c_options),
2703 if (perf_mem_events__init()) {
2704 pr_err("failed: memory events not supported\n");
2708 argc = parse_options(argc, argv, options, record_mem_usage,
2709 PARSE_OPT_KEEP_UNKNOWN);
2711 rec_argc = argc + 10; /* max number of arguments */
2712 rec_argv = calloc(rec_argc + 1, sizeof(char *));
2716 rec_argv[i++] = "record";
2719 perf_mem_events[PERF_MEM_EVENTS__LOAD].record = true;
2720 perf_mem_events[PERF_MEM_EVENTS__STORE].record = true;
2723 if (perf_mem_events[PERF_MEM_EVENTS__LOAD].record)
2724 rec_argv[i++] = "-W";
2726 rec_argv[i++] = "-d";
2727 rec_argv[i++] = "--sample-cpu";
2729 for (j = 0; j < PERF_MEM_EVENTS__MAX; j++) {
2730 if (!perf_mem_events[j].record)
2733 if (!perf_mem_events[j].supported) {
2734 pr_err("failed: event '%s' not supported\n",
2735 perf_mem_events[j].name);
2739 rec_argv[i++] = "-e";
2740 rec_argv[i++] = perf_mem_events__name(j);
2744 rec_argv[i++] = "--all-user";
2747 rec_argv[i++] = "--all-kernel";
2749 for (j = 0; j < argc; j++, i++)
2750 rec_argv[i] = argv[j];
2753 pr_debug("calling: ");
2757 while (rec_argv[j]) {
2758 pr_debug("%s ", rec_argv[j]);
2764 ret = cmd_record(i, rec_argv);
2769 int cmd_c2c(int argc, const char **argv)
2771 argc = parse_options(argc, argv, c2c_options, c2c_usage,
2772 PARSE_OPT_STOP_AT_NON_OPTION);
2775 usage_with_options(c2c_usage, c2c_options);
2777 if (!strncmp(argv[0], "rec", 3)) {
2778 return perf_c2c__record(argc, argv);
2779 } else if (!strncmp(argv[0], "rep", 3)) {
2780 return perf_c2c__report(argc, argv);
2782 usage_with_options(c2c_usage, c2c_options);