2 * This is rewrite of original c2c tool introduced in here:
3 * http://lwn.net/Articles/588866/
5 * The original tool was changed to fit in current perf state.
8 * Don Zickus <dzickus@redhat.com>
9 * Dick Fowles <fowles@inreach.com>
10 * Joe Mario <jmario@redhat.com>
14 #include <linux/compiler.h>
15 #include <linux/kernel.h>
16 #include <linux/stringify.h>
18 #include <sys/param.h>
22 #include <subcmd/parse-options.h>
23 #include "mem-events.h"
33 #include "ui/browsers/hists.h"
39 struct perf_hpp_list list;
40 struct c2c_stats stats;
43 struct compute_stats {
44 struct stats lcl_hitm;
45 struct stats rmt_hitm;
49 struct c2c_hist_entry {
50 struct c2c_hists *hists;
51 struct c2c_stats stats;
52 unsigned long *cpuset;
53 struct c2c_stats *node_stats;
54 unsigned int cacheline_idx;
56 struct compute_stats cstats;
60 * because of its callchain dynamic entry
65 static char const *coalesce_default = "pid,iaddr";
68 struct perf_tool tool;
69 struct c2c_hists hists;
71 unsigned long **nodes;
83 /* HITM shared clines stats */
84 struct c2c_stats hitm_stats;
102 static const char *display_str[DISPLAY_MAX] = {
103 [DISPLAY_LCL] = "Local",
104 [DISPLAY_RMT] = "Remote",
105 [DISPLAY_TOT] = "Total",
108 static const struct option c2c_options[] = {
109 OPT_INCR('v', "verbose", &verbose, "be more verbose (show counter open errors, etc)"),
113 static struct perf_c2c c2c;
115 static void *c2c_he_zalloc(size_t size)
117 struct c2c_hist_entry *c2c_he;
119 c2c_he = zalloc(size + sizeof(*c2c_he));
123 c2c_he->cpuset = bitmap_alloc(c2c.cpus_cnt);
127 c2c_he->node_stats = zalloc(c2c.nodes_cnt * sizeof(*c2c_he->node_stats));
128 if (!c2c_he->node_stats)
131 init_stats(&c2c_he->cstats.lcl_hitm);
132 init_stats(&c2c_he->cstats.rmt_hitm);
133 init_stats(&c2c_he->cstats.load);
138 static void c2c_he_free(void *he)
140 struct c2c_hist_entry *c2c_he;
142 c2c_he = container_of(he, struct c2c_hist_entry, he);
144 hists__delete_entries(&c2c_he->hists->hists);
148 free(c2c_he->cpuset);
149 free(c2c_he->node_stats);
153 static struct hist_entry_ops c2c_entry_ops = {
154 .new = c2c_he_zalloc,
158 static int c2c_hists__init(struct c2c_hists *hists,
160 int nr_header_lines);
162 static struct c2c_hists*
163 he__get_c2c_hists(struct hist_entry *he,
167 struct c2c_hist_entry *c2c_he;
168 struct c2c_hists *hists;
171 c2c_he = container_of(he, struct c2c_hist_entry, he);
173 return c2c_he->hists;
175 hists = c2c_he->hists = zalloc(sizeof(*hists));
179 ret = c2c_hists__init(hists, sort, nr_header_lines);
188 static void c2c_he__set_cpu(struct c2c_hist_entry *c2c_he,
189 struct perf_sample *sample)
191 if (WARN_ONCE(sample->cpu == (unsigned int) -1,
192 "WARNING: no sample cpu value"))
195 set_bit(sample->cpu, c2c_he->cpuset);
198 static void compute_stats(struct c2c_hist_entry *c2c_he,
199 struct c2c_stats *stats,
202 struct compute_stats *cstats = &c2c_he->cstats;
205 update_stats(&cstats->rmt_hitm, weight);
206 else if (stats->lcl_hitm)
207 update_stats(&cstats->lcl_hitm, weight);
208 else if (stats->load)
209 update_stats(&cstats->load, weight);
212 static int process_sample_event(struct perf_tool *tool __maybe_unused,
213 union perf_event *event,
214 struct perf_sample *sample,
215 struct perf_evsel *evsel,
216 struct machine *machine)
218 struct c2c_hists *c2c_hists = &c2c.hists;
219 struct c2c_hist_entry *c2c_he;
220 struct c2c_stats stats = { .nr_entries = 0, };
221 struct hist_entry *he;
222 struct addr_location al;
223 struct mem_info *mi, *mi_dup;
226 if (machine__resolve(machine, &al, sample) < 0) {
227 pr_debug("problem processing %d event, skipping it.\n",
232 ret = sample__resolve_callchain(sample, &callchain_cursor, NULL,
233 evsel, &al, sysctl_perf_event_max_stack);
237 mi = sample__resolve_mem(sample, &al);
241 mi_dup = memdup(mi, sizeof(*mi));
245 c2c_decode_stats(&stats, mi);
247 he = hists__add_entry_ops(&c2c_hists->hists, &c2c_entry_ops,
253 c2c_he = container_of(he, struct c2c_hist_entry, he);
254 c2c_add_stats(&c2c_he->stats, &stats);
255 c2c_add_stats(&c2c_hists->stats, &stats);
257 c2c_he__set_cpu(c2c_he, sample);
259 hists__inc_nr_samples(&c2c_hists->hists, he->filtered);
260 ret = hist_entry__append_callchain(he, sample);
264 * There's already been warning about missing
265 * sample's cpu value. Let's account all to
266 * node 0 in this case, without any further
269 * Doing node stats only for single callchain data.
271 int cpu = sample->cpu == (unsigned int) -1 ? 0 : sample->cpu;
272 int node = c2c.cpu2node[cpu];
276 mi_dup = memdup(mi, sizeof(*mi));
280 c2c_hists = he__get_c2c_hists(he, c2c.cl_sort, 2);
284 he = hists__add_entry_ops(&c2c_hists->hists, &c2c_entry_ops,
290 c2c_he = container_of(he, struct c2c_hist_entry, he);
291 c2c_add_stats(&c2c_he->stats, &stats);
292 c2c_add_stats(&c2c_hists->stats, &stats);
293 c2c_add_stats(&c2c_he->node_stats[node], &stats);
295 compute_stats(c2c_he, &stats, sample->weight);
297 c2c_he__set_cpu(c2c_he, sample);
299 hists__inc_nr_samples(&c2c_hists->hists, he->filtered);
300 ret = hist_entry__append_callchain(he, sample);
304 addr_location__put(&al);
315 static struct perf_c2c c2c = {
317 .sample = process_sample_event,
318 .mmap = perf_event__process_mmap,
319 .mmap2 = perf_event__process_mmap2,
320 .comm = perf_event__process_comm,
321 .exit = perf_event__process_exit,
322 .fork = perf_event__process_fork,
323 .lost = perf_event__process_lost,
324 .ordered_events = true,
325 .ordering_requires_timestamps = true,
329 static const char * const c2c_usage[] = {
330 "perf c2c {record|report}",
334 static const char * const __usage_report[] = {
339 static const char * const *report_c2c_usage = __usage_report;
341 #define C2C_HEADER_MAX 2
347 } line[C2C_HEADER_MAX];
350 struct c2c_dimension {
351 struct c2c_header header;
354 struct sort_entry *se;
356 int64_t (*cmp)(struct perf_hpp_fmt *fmt,
357 struct hist_entry *, struct hist_entry *);
358 int (*entry)(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
359 struct hist_entry *he);
360 int (*color)(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
361 struct hist_entry *he);
365 struct perf_hpp_fmt fmt;
366 struct c2c_dimension *dim;
369 #define SYMBOL_WIDTH 30
371 static struct c2c_dimension dim_symbol;
372 static struct c2c_dimension dim_srcline;
374 static int symbol_width(struct hists *hists, struct sort_entry *se)
376 int width = hists__col_len(hists, se->se_width_idx);
378 if (!c2c.symbol_full)
379 width = MIN(width, SYMBOL_WIDTH);
384 static int c2c_width(struct perf_hpp_fmt *fmt,
385 struct perf_hpp *hpp __maybe_unused,
388 struct c2c_fmt *c2c_fmt;
389 struct c2c_dimension *dim;
391 c2c_fmt = container_of(fmt, struct c2c_fmt, fmt);
394 if (dim == &dim_symbol || dim == &dim_srcline)
395 return symbol_width(hists, dim->se);
397 return dim->se ? hists__col_len(hists, dim->se->se_width_idx) :
401 static int c2c_header(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
402 struct hists *hists, int line, int *span)
404 struct perf_hpp_list *hpp_list = hists->hpp_list;
405 struct c2c_fmt *c2c_fmt;
406 struct c2c_dimension *dim;
407 const char *text = NULL;
408 int width = c2c_width(fmt, hpp, hists);
410 c2c_fmt = container_of(fmt, struct c2c_fmt, fmt);
414 text = dim->header.line[line].text;
415 /* Use the last line from sort_entry if not defined. */
416 if (!text && (line == hpp_list->nr_header_lines - 1))
417 text = dim->se->se_header;
419 text = dim->header.line[line].text;
425 *span = dim->header.line[line].span;
432 return scnprintf(hpp->buf, hpp->size, "%*s", width, text);
435 #define HEX_STR(__s, __v) \
437 scnprintf(__s, sizeof(__s), "0x%" PRIx64, __v); \
442 dcacheline_cmp(struct perf_hpp_fmt *fmt __maybe_unused,
443 struct hist_entry *left, struct hist_entry *right)
445 return sort__dcacheline_cmp(left, right);
448 static int dcacheline_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
449 struct hist_entry *he)
452 int width = c2c_width(fmt, hpp, he->hists);
456 addr = cl_address(he->mem_info->daddr.addr);
458 return scnprintf(hpp->buf, hpp->size, "%*s", width, HEX_STR(buf, addr));
461 static int offset_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
462 struct hist_entry *he)
465 int width = c2c_width(fmt, hpp, he->hists);
469 addr = cl_offset(he->mem_info->daddr.al_addr);
471 return scnprintf(hpp->buf, hpp->size, "%*s", width, HEX_STR(buf, addr));
475 offset_cmp(struct perf_hpp_fmt *fmt __maybe_unused,
476 struct hist_entry *left, struct hist_entry *right)
478 uint64_t l = 0, r = 0;
481 l = cl_offset(left->mem_info->daddr.addr);
483 r = cl_offset(right->mem_info->daddr.addr);
485 return (int64_t)(r - l);
489 iaddr_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
490 struct hist_entry *he)
493 int width = c2c_width(fmt, hpp, he->hists);
497 addr = he->mem_info->iaddr.addr;
499 return scnprintf(hpp->buf, hpp->size, "%*s", width, HEX_STR(buf, addr));
503 iaddr_cmp(struct perf_hpp_fmt *fmt __maybe_unused,
504 struct hist_entry *left, struct hist_entry *right)
506 return sort__iaddr_cmp(left, right);
510 tot_hitm_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
511 struct hist_entry *he)
513 struct c2c_hist_entry *c2c_he;
514 int width = c2c_width(fmt, hpp, he->hists);
515 unsigned int tot_hitm;
517 c2c_he = container_of(he, struct c2c_hist_entry, he);
518 tot_hitm = c2c_he->stats.lcl_hitm + c2c_he->stats.rmt_hitm;
520 return scnprintf(hpp->buf, hpp->size, "%*u", width, tot_hitm);
524 tot_hitm_cmp(struct perf_hpp_fmt *fmt __maybe_unused,
525 struct hist_entry *left, struct hist_entry *right)
527 struct c2c_hist_entry *c2c_left;
528 struct c2c_hist_entry *c2c_right;
529 unsigned int tot_hitm_left;
530 unsigned int tot_hitm_right;
532 c2c_left = container_of(left, struct c2c_hist_entry, he);
533 c2c_right = container_of(right, struct c2c_hist_entry, he);
535 tot_hitm_left = c2c_left->stats.lcl_hitm + c2c_left->stats.rmt_hitm;
536 tot_hitm_right = c2c_right->stats.lcl_hitm + c2c_right->stats.rmt_hitm;
538 return tot_hitm_left - tot_hitm_right;
541 #define STAT_FN_ENTRY(__f) \
543 __f ## _entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp, \
544 struct hist_entry *he) \
546 struct c2c_hist_entry *c2c_he; \
547 int width = c2c_width(fmt, hpp, he->hists); \
549 c2c_he = container_of(he, struct c2c_hist_entry, he); \
550 return scnprintf(hpp->buf, hpp->size, "%*u", width, \
551 c2c_he->stats.__f); \
554 #define STAT_FN_CMP(__f) \
556 __f ## _cmp(struct perf_hpp_fmt *fmt __maybe_unused, \
557 struct hist_entry *left, struct hist_entry *right) \
559 struct c2c_hist_entry *c2c_left, *c2c_right; \
561 c2c_left = container_of(left, struct c2c_hist_entry, he); \
562 c2c_right = container_of(right, struct c2c_hist_entry, he); \
563 return c2c_left->stats.__f - c2c_right->stats.__f; \
566 #define STAT_FN(__f) \
581 static uint64_t llc_miss(struct c2c_stats *stats)
585 llcmiss = stats->lcl_dram +
594 ld_llcmiss_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
595 struct hist_entry *he)
597 struct c2c_hist_entry *c2c_he;
598 int width = c2c_width(fmt, hpp, he->hists);
600 c2c_he = container_of(he, struct c2c_hist_entry, he);
602 return scnprintf(hpp->buf, hpp->size, "%*lu", width,
603 llc_miss(&c2c_he->stats));
607 ld_llcmiss_cmp(struct perf_hpp_fmt *fmt __maybe_unused,
608 struct hist_entry *left, struct hist_entry *right)
610 struct c2c_hist_entry *c2c_left;
611 struct c2c_hist_entry *c2c_right;
613 c2c_left = container_of(left, struct c2c_hist_entry, he);
614 c2c_right = container_of(right, struct c2c_hist_entry, he);
616 return llc_miss(&c2c_left->stats) - llc_miss(&c2c_right->stats);
619 static uint64_t total_records(struct c2c_stats *stats)
621 uint64_t lclmiss, ldcnt, total;
623 lclmiss = stats->lcl_dram +
643 tot_recs_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
644 struct hist_entry *he)
646 struct c2c_hist_entry *c2c_he;
647 int width = c2c_width(fmt, hpp, he->hists);
650 c2c_he = container_of(he, struct c2c_hist_entry, he);
651 tot_recs = total_records(&c2c_he->stats);
653 return scnprintf(hpp->buf, hpp->size, "%*" PRIu64, width, tot_recs);
657 tot_recs_cmp(struct perf_hpp_fmt *fmt __maybe_unused,
658 struct hist_entry *left, struct hist_entry *right)
660 struct c2c_hist_entry *c2c_left;
661 struct c2c_hist_entry *c2c_right;
662 uint64_t tot_recs_left;
663 uint64_t tot_recs_right;
665 c2c_left = container_of(left, struct c2c_hist_entry, he);
666 c2c_right = container_of(right, struct c2c_hist_entry, he);
668 tot_recs_left = total_records(&c2c_left->stats);
669 tot_recs_right = total_records(&c2c_right->stats);
671 return tot_recs_left - tot_recs_right;
674 static uint64_t total_loads(struct c2c_stats *stats)
676 uint64_t lclmiss, ldcnt;
678 lclmiss = stats->lcl_dram +
694 tot_loads_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
695 struct hist_entry *he)
697 struct c2c_hist_entry *c2c_he;
698 int width = c2c_width(fmt, hpp, he->hists);
701 c2c_he = container_of(he, struct c2c_hist_entry, he);
702 tot_recs = total_loads(&c2c_he->stats);
704 return scnprintf(hpp->buf, hpp->size, "%*" PRIu64, width, tot_recs);
708 tot_loads_cmp(struct perf_hpp_fmt *fmt __maybe_unused,
709 struct hist_entry *left, struct hist_entry *right)
711 struct c2c_hist_entry *c2c_left;
712 struct c2c_hist_entry *c2c_right;
713 uint64_t tot_recs_left;
714 uint64_t tot_recs_right;
716 c2c_left = container_of(left, struct c2c_hist_entry, he);
717 c2c_right = container_of(right, struct c2c_hist_entry, he);
719 tot_recs_left = total_loads(&c2c_left->stats);
720 tot_recs_right = total_loads(&c2c_right->stats);
722 return tot_recs_left - tot_recs_right;
725 typedef double (get_percent_cb)(struct c2c_hist_entry *);
728 percent_color(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
729 struct hist_entry *he, get_percent_cb get_percent)
731 struct c2c_hist_entry *c2c_he;
732 int width = c2c_width(fmt, hpp, he->hists);
735 c2c_he = container_of(he, struct c2c_hist_entry, he);
736 per = get_percent(c2c_he);
738 #ifdef HAVE_SLANG_SUPPORT
740 return __hpp__slsmg_color_printf(hpp, "%*.2f%%", width - 1, per);
742 return hpp_color_scnprintf(hpp, "%*.2f%%", width - 1, per);
745 static double percent_hitm(struct c2c_hist_entry *c2c_he)
747 struct c2c_hists *hists;
748 struct c2c_stats *stats;
749 struct c2c_stats *total;
753 hists = container_of(c2c_he->he.hists, struct c2c_hists, hists);
754 stats = &c2c_he->stats;
755 total = &hists->stats;
757 switch (c2c.display) {
759 st = stats->rmt_hitm;
760 tot = total->rmt_hitm;
763 st = stats->lcl_hitm;
764 tot = total->lcl_hitm;
767 st = stats->tot_hitm;
768 tot = total->tot_hitm;
773 p = tot ? (double) st / tot : 0;
778 #define PERC_STR(__s, __v) \
780 scnprintf(__s, sizeof(__s), "%.2F%%", __v); \
785 percent_hitm_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
786 struct hist_entry *he)
788 struct c2c_hist_entry *c2c_he;
789 int width = c2c_width(fmt, hpp, he->hists);
793 c2c_he = container_of(he, struct c2c_hist_entry, he);
794 per = percent_hitm(c2c_he);
795 return scnprintf(hpp->buf, hpp->size, "%*s", width, PERC_STR(buf, per));
799 percent_hitm_color(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
800 struct hist_entry *he)
802 return percent_color(fmt, hpp, he, percent_hitm);
806 percent_hitm_cmp(struct perf_hpp_fmt *fmt __maybe_unused,
807 struct hist_entry *left, struct hist_entry *right)
809 struct c2c_hist_entry *c2c_left;
810 struct c2c_hist_entry *c2c_right;
814 c2c_left = container_of(left, struct c2c_hist_entry, he);
815 c2c_right = container_of(right, struct c2c_hist_entry, he);
817 per_left = percent_hitm(c2c_left);
818 per_right = percent_hitm(c2c_right);
820 return per_left - per_right;
823 static struct c2c_stats *he_stats(struct hist_entry *he)
825 struct c2c_hist_entry *c2c_he;
827 c2c_he = container_of(he, struct c2c_hist_entry, he);
828 return &c2c_he->stats;
831 static struct c2c_stats *total_stats(struct hist_entry *he)
833 struct c2c_hists *hists;
835 hists = container_of(he->hists, struct c2c_hists, hists);
836 return &hists->stats;
839 static double percent(int st, int tot)
841 return tot ? 100. * (double) st / (double) tot : 0;
844 #define PERCENT(__h, __f) percent(he_stats(__h)->__f, total_stats(__h)->__f)
846 #define PERCENT_FN(__f) \
847 static double percent_ ## __f(struct c2c_hist_entry *c2c_he) \
849 struct c2c_hists *hists; \
851 hists = container_of(c2c_he->he.hists, struct c2c_hists, hists); \
852 return percent(c2c_he->stats.__f, hists->stats.__f); \
858 PERCENT_FN(st_l1miss)
861 percent_rmt_hitm_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
862 struct hist_entry *he)
864 int width = c2c_width(fmt, hpp, he->hists);
865 double per = PERCENT(he, rmt_hitm);
868 return scnprintf(hpp->buf, hpp->size, "%*s", width, PERC_STR(buf, per));
872 percent_rmt_hitm_color(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
873 struct hist_entry *he)
875 return percent_color(fmt, hpp, he, percent_rmt_hitm);
879 percent_rmt_hitm_cmp(struct perf_hpp_fmt *fmt __maybe_unused,
880 struct hist_entry *left, struct hist_entry *right)
885 per_left = PERCENT(left, lcl_hitm);
886 per_right = PERCENT(right, lcl_hitm);
888 return per_left - per_right;
892 percent_lcl_hitm_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
893 struct hist_entry *he)
895 int width = c2c_width(fmt, hpp, he->hists);
896 double per = PERCENT(he, lcl_hitm);
899 return scnprintf(hpp->buf, hpp->size, "%*s", width, PERC_STR(buf, per));
903 percent_lcl_hitm_color(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
904 struct hist_entry *he)
906 return percent_color(fmt, hpp, he, percent_lcl_hitm);
910 percent_lcl_hitm_cmp(struct perf_hpp_fmt *fmt __maybe_unused,
911 struct hist_entry *left, struct hist_entry *right)
916 per_left = PERCENT(left, lcl_hitm);
917 per_right = PERCENT(right, lcl_hitm);
919 return per_left - per_right;
923 percent_stores_l1hit_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
924 struct hist_entry *he)
926 int width = c2c_width(fmt, hpp, he->hists);
927 double per = PERCENT(he, st_l1hit);
930 return scnprintf(hpp->buf, hpp->size, "%*s", width, PERC_STR(buf, per));
934 percent_stores_l1hit_color(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
935 struct hist_entry *he)
937 return percent_color(fmt, hpp, he, percent_st_l1hit);
941 percent_stores_l1hit_cmp(struct perf_hpp_fmt *fmt __maybe_unused,
942 struct hist_entry *left, struct hist_entry *right)
947 per_left = PERCENT(left, st_l1hit);
948 per_right = PERCENT(right, st_l1hit);
950 return per_left - per_right;
954 percent_stores_l1miss_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
955 struct hist_entry *he)
957 int width = c2c_width(fmt, hpp, he->hists);
958 double per = PERCENT(he, st_l1miss);
961 return scnprintf(hpp->buf, hpp->size, "%*s", width, PERC_STR(buf, per));
965 percent_stores_l1miss_color(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
966 struct hist_entry *he)
968 return percent_color(fmt, hpp, he, percent_st_l1miss);
972 percent_stores_l1miss_cmp(struct perf_hpp_fmt *fmt __maybe_unused,
973 struct hist_entry *left, struct hist_entry *right)
978 per_left = PERCENT(left, st_l1miss);
979 per_right = PERCENT(right, st_l1miss);
981 return per_left - per_right;
988 pid_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
989 struct hist_entry *he)
991 int width = c2c_width(fmt, hpp, he->hists);
993 return scnprintf(hpp->buf, hpp->size, "%*d", width, he->thread->pid_);
997 pid_cmp(struct perf_hpp_fmt *fmt __maybe_unused,
998 struct hist_entry *left, struct hist_entry *right)
1000 return left->thread->pid_ - right->thread->pid_;
1004 empty_cmp(struct perf_hpp_fmt *fmt __maybe_unused,
1005 struct hist_entry *left __maybe_unused,
1006 struct hist_entry *right __maybe_unused)
1012 node_entry(struct perf_hpp_fmt *fmt __maybe_unused, struct perf_hpp *hpp,
1013 struct hist_entry *he)
1015 struct c2c_hist_entry *c2c_he;
1020 c2c_he = container_of(he, struct c2c_hist_entry, he);
1022 for (node = 0; node < c2c.nodes_cnt; node++) {
1023 DECLARE_BITMAP(set, c2c.cpus_cnt);
1025 bitmap_zero(set, c2c.cpus_cnt);
1026 bitmap_and(set, c2c_he->cpuset, c2c.nodes[node], c2c.cpus_cnt);
1028 if (!bitmap_weight(set, c2c.cpus_cnt)) {
1029 if (c2c.node_info == 1) {
1030 ret = scnprintf(hpp->buf, hpp->size, "%21s", " ");
1031 advance_hpp(hpp, ret);
1037 ret = scnprintf(hpp->buf, hpp->size, " ");
1038 advance_hpp(hpp, ret);
1041 switch (c2c.node_info) {
1043 ret = scnprintf(hpp->buf, hpp->size, "%2d", node);
1044 advance_hpp(hpp, ret);
1048 int num = bitmap_weight(c2c_he->cpuset, c2c.cpus_cnt);
1049 struct c2c_stats *stats = &c2c_he->node_stats[node];
1051 ret = scnprintf(hpp->buf, hpp->size, "%2d{%2d ", node, num);
1052 advance_hpp(hpp, ret);
1054 #define DISPLAY_HITM(__h) \
1055 if (c2c_he->stats.__h> 0) { \
1056 ret = scnprintf(hpp->buf, hpp->size, "%5.1f%% ", \
1057 percent(stats->__h, c2c_he->stats.__h));\
1059 ret = scnprintf(hpp->buf, hpp->size, "%6s ", "n/a"); \
1062 switch (c2c.display) {
1064 DISPLAY_HITM(rmt_hitm);
1067 DISPLAY_HITM(lcl_hitm);
1070 DISPLAY_HITM(tot_hitm);
1077 advance_hpp(hpp, ret);
1079 if (c2c_he->stats.store > 0) {
1080 ret = scnprintf(hpp->buf, hpp->size, "%5.1f%%}",
1081 percent(stats->store, c2c_he->stats.store));
1083 ret = scnprintf(hpp->buf, hpp->size, "%6s}", "n/a");
1086 advance_hpp(hpp, ret);
1090 ret = scnprintf(hpp->buf, hpp->size, "%2d{", node);
1091 advance_hpp(hpp, ret);
1093 ret = bitmap_scnprintf(set, c2c.cpus_cnt, hpp->buf, hpp->size);
1094 advance_hpp(hpp, ret);
1096 ret = scnprintf(hpp->buf, hpp->size, "}");
1097 advance_hpp(hpp, ret);
1110 mean_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
1111 struct hist_entry *he, double mean)
1113 int width = c2c_width(fmt, hpp, he->hists);
1116 scnprintf(buf, 10, "%6.0f", mean);
1117 return scnprintf(hpp->buf, hpp->size, "%*s", width, buf);
1120 #define MEAN_ENTRY(__func, __val) \
1122 __func(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp, struct hist_entry *he) \
1124 struct c2c_hist_entry *c2c_he; \
1125 c2c_he = container_of(he, struct c2c_hist_entry, he); \
1126 return mean_entry(fmt, hpp, he, avg_stats(&c2c_he->cstats.__val)); \
1129 MEAN_ENTRY(mean_rmt_entry, rmt_hitm);
1130 MEAN_ENTRY(mean_lcl_entry, lcl_hitm);
1131 MEAN_ENTRY(mean_load_entry, load);
1134 cpucnt_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
1135 struct hist_entry *he)
1137 struct c2c_hist_entry *c2c_he;
1138 int width = c2c_width(fmt, hpp, he->hists);
1141 c2c_he = container_of(he, struct c2c_hist_entry, he);
1143 scnprintf(buf, 10, "%d", bitmap_weight(c2c_he->cpuset, c2c.cpus_cnt));
1144 return scnprintf(hpp->buf, hpp->size, "%*s", width, buf);
1148 cl_idx_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
1149 struct hist_entry *he)
1151 struct c2c_hist_entry *c2c_he;
1152 int width = c2c_width(fmt, hpp, he->hists);
1155 c2c_he = container_of(he, struct c2c_hist_entry, he);
1157 scnprintf(buf, 10, "%u", c2c_he->cacheline_idx);
1158 return scnprintf(hpp->buf, hpp->size, "%*s", width, buf);
1162 cl_idx_empty_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
1163 struct hist_entry *he)
1165 int width = c2c_width(fmt, hpp, he->hists);
1167 return scnprintf(hpp->buf, hpp->size, "%*s", width, "");
1170 #define HEADER_LOW(__h) \
1177 #define HEADER_BOTH(__h0, __h1) \
1187 #define HEADER_SPAN(__h0, __h1, __s) \
1198 #define HEADER_SPAN_LOW(__h) \
1205 static struct c2c_dimension dim_dcacheline = {
1206 .header = HEADER_LOW("Cacheline"),
1207 .name = "dcacheline",
1208 .cmp = dcacheline_cmp,
1209 .entry = dcacheline_entry,
1213 static struct c2c_header header_offset_tui = HEADER_LOW("Off");
1215 static struct c2c_dimension dim_offset = {
1216 .header = HEADER_BOTH("Data address", "Offset"),
1219 .entry = offset_entry,
1223 static struct c2c_dimension dim_iaddr = {
1224 .header = HEADER_LOW("Code address"),
1227 .entry = iaddr_entry,
1231 static struct c2c_dimension dim_tot_hitm = {
1232 .header = HEADER_SPAN("----- LLC Load Hitm -----", "Total", 2),
1234 .cmp = tot_hitm_cmp,
1235 .entry = tot_hitm_entry,
1239 static struct c2c_dimension dim_lcl_hitm = {
1240 .header = HEADER_SPAN_LOW("Lcl"),
1242 .cmp = lcl_hitm_cmp,
1243 .entry = lcl_hitm_entry,
1247 static struct c2c_dimension dim_rmt_hitm = {
1248 .header = HEADER_SPAN_LOW("Rmt"),
1250 .cmp = rmt_hitm_cmp,
1251 .entry = rmt_hitm_entry,
1255 static struct c2c_dimension dim_cl_rmt_hitm = {
1256 .header = HEADER_SPAN("----- HITM -----", "Rmt", 1),
1257 .name = "cl_rmt_hitm",
1258 .cmp = rmt_hitm_cmp,
1259 .entry = rmt_hitm_entry,
1263 static struct c2c_dimension dim_cl_lcl_hitm = {
1264 .header = HEADER_SPAN_LOW("Lcl"),
1265 .name = "cl_lcl_hitm",
1266 .cmp = lcl_hitm_cmp,
1267 .entry = lcl_hitm_entry,
1271 static struct c2c_dimension dim_stores = {
1272 .header = HEADER_SPAN("---- Store Reference ----", "Total", 2),
1275 .entry = store_entry,
1279 static struct c2c_dimension dim_stores_l1hit = {
1280 .header = HEADER_SPAN_LOW("L1Hit"),
1281 .name = "stores_l1hit",
1282 .cmp = st_l1hit_cmp,
1283 .entry = st_l1hit_entry,
1287 static struct c2c_dimension dim_stores_l1miss = {
1288 .header = HEADER_SPAN_LOW("L1Miss"),
1289 .name = "stores_l1miss",
1290 .cmp = st_l1miss_cmp,
1291 .entry = st_l1miss_entry,
1295 static struct c2c_dimension dim_cl_stores_l1hit = {
1296 .header = HEADER_SPAN("-- Store Refs --", "L1 Hit", 1),
1297 .name = "cl_stores_l1hit",
1298 .cmp = st_l1hit_cmp,
1299 .entry = st_l1hit_entry,
1303 static struct c2c_dimension dim_cl_stores_l1miss = {
1304 .header = HEADER_SPAN_LOW("L1 Miss"),
1305 .name = "cl_stores_l1miss",
1306 .cmp = st_l1miss_cmp,
1307 .entry = st_l1miss_entry,
1311 static struct c2c_dimension dim_ld_fbhit = {
1312 .header = HEADER_SPAN("----- Core Load Hit -----", "FB", 2),
1314 .cmp = ld_fbhit_cmp,
1315 .entry = ld_fbhit_entry,
1319 static struct c2c_dimension dim_ld_l1hit = {
1320 .header = HEADER_SPAN_LOW("L1"),
1322 .cmp = ld_l1hit_cmp,
1323 .entry = ld_l1hit_entry,
1327 static struct c2c_dimension dim_ld_l2hit = {
1328 .header = HEADER_SPAN_LOW("L2"),
1330 .cmp = ld_l2hit_cmp,
1331 .entry = ld_l2hit_entry,
1335 static struct c2c_dimension dim_ld_llchit = {
1336 .header = HEADER_SPAN("-- LLC Load Hit --", "Llc", 1),
1337 .name = "ld_lclhit",
1338 .cmp = ld_llchit_cmp,
1339 .entry = ld_llchit_entry,
1343 static struct c2c_dimension dim_ld_rmthit = {
1344 .header = HEADER_SPAN_LOW("Rmt"),
1345 .name = "ld_rmthit",
1347 .entry = rmt_hit_entry,
1351 static struct c2c_dimension dim_ld_llcmiss = {
1352 .header = HEADER_BOTH("LLC", "Ld Miss"),
1353 .name = "ld_llcmiss",
1354 .cmp = ld_llcmiss_cmp,
1355 .entry = ld_llcmiss_entry,
1359 static struct c2c_dimension dim_tot_recs = {
1360 .header = HEADER_BOTH("Total", "records"),
1362 .cmp = tot_recs_cmp,
1363 .entry = tot_recs_entry,
1367 static struct c2c_dimension dim_tot_loads = {
1368 .header = HEADER_BOTH("Total", "Loads"),
1369 .name = "tot_loads",
1370 .cmp = tot_loads_cmp,
1371 .entry = tot_loads_entry,
1375 static struct c2c_header percent_hitm_header[] = {
1376 [DISPLAY_LCL] = HEADER_BOTH("Lcl", "Hitm"),
1377 [DISPLAY_RMT] = HEADER_BOTH("Rmt", "Hitm"),
1378 [DISPLAY_TOT] = HEADER_BOTH("Tot", "Hitm"),
1381 static struct c2c_dimension dim_percent_hitm = {
1382 .name = "percent_hitm",
1383 .cmp = percent_hitm_cmp,
1384 .entry = percent_hitm_entry,
1385 .color = percent_hitm_color,
1389 static struct c2c_dimension dim_percent_rmt_hitm = {
1390 .header = HEADER_SPAN("----- HITM -----", "Rmt", 1),
1391 .name = "percent_rmt_hitm",
1392 .cmp = percent_rmt_hitm_cmp,
1393 .entry = percent_rmt_hitm_entry,
1394 .color = percent_rmt_hitm_color,
1398 static struct c2c_dimension dim_percent_lcl_hitm = {
1399 .header = HEADER_SPAN_LOW("Lcl"),
1400 .name = "percent_lcl_hitm",
1401 .cmp = percent_lcl_hitm_cmp,
1402 .entry = percent_lcl_hitm_entry,
1403 .color = percent_lcl_hitm_color,
1407 static struct c2c_dimension dim_percent_stores_l1hit = {
1408 .header = HEADER_SPAN("-- Store Refs --", "L1 Hit", 1),
1409 .name = "percent_stores_l1hit",
1410 .cmp = percent_stores_l1hit_cmp,
1411 .entry = percent_stores_l1hit_entry,
1412 .color = percent_stores_l1hit_color,
1416 static struct c2c_dimension dim_percent_stores_l1miss = {
1417 .header = HEADER_SPAN_LOW("L1 Miss"),
1418 .name = "percent_stores_l1miss",
1419 .cmp = percent_stores_l1miss_cmp,
1420 .entry = percent_stores_l1miss_entry,
1421 .color = percent_stores_l1miss_color,
1425 static struct c2c_dimension dim_dram_lcl = {
1426 .header = HEADER_SPAN("--- Load Dram ----", "Lcl", 1),
1428 .cmp = lcl_dram_cmp,
1429 .entry = lcl_dram_entry,
1433 static struct c2c_dimension dim_dram_rmt = {
1434 .header = HEADER_SPAN_LOW("Rmt"),
1436 .cmp = rmt_dram_cmp,
1437 .entry = rmt_dram_entry,
1441 static struct c2c_dimension dim_pid = {
1442 .header = HEADER_LOW("Pid"),
1449 static struct c2c_dimension dim_tid = {
1450 .header = HEADER_LOW("Tid"),
1455 static struct c2c_dimension dim_symbol = {
1460 static struct c2c_dimension dim_dso = {
1461 .header = HEADER_BOTH("Shared", "Object"),
1466 static struct c2c_header header_node[3] = {
1468 HEADER_LOW("Node{cpus %hitms %stores}"),
1469 HEADER_LOW("Node{cpu list}"),
1472 static struct c2c_dimension dim_node = {
1475 .entry = node_entry,
1479 static struct c2c_dimension dim_mean_rmt = {
1480 .header = HEADER_SPAN("---------- cycles ----------", "rmt hitm", 2),
1483 .entry = mean_rmt_entry,
1487 static struct c2c_dimension dim_mean_lcl = {
1488 .header = HEADER_SPAN_LOW("lcl hitm"),
1491 .entry = mean_lcl_entry,
1495 static struct c2c_dimension dim_mean_load = {
1496 .header = HEADER_SPAN_LOW("load"),
1497 .name = "mean_load",
1499 .entry = mean_load_entry,
1503 static struct c2c_dimension dim_cpucnt = {
1504 .header = HEADER_BOTH("cpu", "cnt"),
1507 .entry = cpucnt_entry,
1511 static struct c2c_dimension dim_srcline = {
1512 .name = "cl_srcline",
1513 .se = &sort_srcline,
1516 static struct c2c_dimension dim_dcacheline_idx = {
1517 .header = HEADER_LOW("Index"),
1520 .entry = cl_idx_entry,
1524 static struct c2c_dimension dim_dcacheline_num = {
1525 .header = HEADER_LOW("Num"),
1528 .entry = cl_idx_entry,
1532 static struct c2c_dimension dim_dcacheline_num_empty = {
1533 .header = HEADER_LOW("Num"),
1534 .name = "cl_num_empty",
1536 .entry = cl_idx_empty_entry,
1540 static struct c2c_dimension *dimensions[] = {
1552 &dim_cl_stores_l1hit,
1553 &dim_cl_stores_l1miss,
1563 &dim_percent_rmt_hitm,
1564 &dim_percent_lcl_hitm,
1565 &dim_percent_stores_l1hit,
1566 &dim_percent_stores_l1miss,
1579 &dim_dcacheline_idx,
1580 &dim_dcacheline_num,
1581 &dim_dcacheline_num_empty,
1585 static void fmt_free(struct perf_hpp_fmt *fmt)
1587 struct c2c_fmt *c2c_fmt;
1589 c2c_fmt = container_of(fmt, struct c2c_fmt, fmt);
1593 static bool fmt_equal(struct perf_hpp_fmt *a, struct perf_hpp_fmt *b)
1595 struct c2c_fmt *c2c_a = container_of(a, struct c2c_fmt, fmt);
1596 struct c2c_fmt *c2c_b = container_of(b, struct c2c_fmt, fmt);
1598 return c2c_a->dim == c2c_b->dim;
1601 static struct c2c_dimension *get_dimension(const char *name)
1605 for (i = 0; dimensions[i]; i++) {
1606 struct c2c_dimension *dim = dimensions[i];
1608 if (!strcmp(dim->name, name))
1615 static int c2c_se_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
1616 struct hist_entry *he)
1618 struct c2c_fmt *c2c_fmt = container_of(fmt, struct c2c_fmt, fmt);
1619 struct c2c_dimension *dim = c2c_fmt->dim;
1620 size_t len = fmt->user_len;
1623 len = hists__col_len(he->hists, dim->se->se_width_idx);
1625 if (dim == &dim_symbol || dim == &dim_srcline)
1626 len = symbol_width(he->hists, dim->se);
1629 return dim->se->se_snprintf(he, hpp->buf, hpp->size, len);
1632 static int64_t c2c_se_cmp(struct perf_hpp_fmt *fmt,
1633 struct hist_entry *a, struct hist_entry *b)
1635 struct c2c_fmt *c2c_fmt = container_of(fmt, struct c2c_fmt, fmt);
1636 struct c2c_dimension *dim = c2c_fmt->dim;
1638 return dim->se->se_cmp(a, b);
1641 static int64_t c2c_se_collapse(struct perf_hpp_fmt *fmt,
1642 struct hist_entry *a, struct hist_entry *b)
1644 struct c2c_fmt *c2c_fmt = container_of(fmt, struct c2c_fmt, fmt);
1645 struct c2c_dimension *dim = c2c_fmt->dim;
1646 int64_t (*collapse_fn)(struct hist_entry *, struct hist_entry *);
1648 collapse_fn = dim->se->se_collapse ?: dim->se->se_cmp;
1649 return collapse_fn(a, b);
1652 static struct c2c_fmt *get_format(const char *name)
1654 struct c2c_dimension *dim = get_dimension(name);
1655 struct c2c_fmt *c2c_fmt;
1656 struct perf_hpp_fmt *fmt;
1661 c2c_fmt = zalloc(sizeof(*c2c_fmt));
1667 fmt = &c2c_fmt->fmt;
1668 INIT_LIST_HEAD(&fmt->list);
1669 INIT_LIST_HEAD(&fmt->sort_list);
1671 fmt->cmp = dim->se ? c2c_se_cmp : dim->cmp;
1672 fmt->sort = dim->se ? c2c_se_cmp : dim->cmp;
1673 fmt->color = dim->se ? NULL : dim->color;
1674 fmt->entry = dim->se ? c2c_se_entry : dim->entry;
1675 fmt->header = c2c_header;
1676 fmt->width = c2c_width;
1677 fmt->collapse = dim->se ? c2c_se_collapse : dim->cmp;
1678 fmt->equal = fmt_equal;
1679 fmt->free = fmt_free;
1684 static int c2c_hists__init_output(struct perf_hpp_list *hpp_list, char *name)
1686 struct c2c_fmt *c2c_fmt = get_format(name);
1690 return output_field_add(hpp_list, name);
1693 perf_hpp_list__column_register(hpp_list, &c2c_fmt->fmt);
1697 static int c2c_hists__init_sort(struct perf_hpp_list *hpp_list, char *name)
1699 struct c2c_fmt *c2c_fmt = get_format(name);
1700 struct c2c_dimension *dim;
1704 return sort_dimension__add(hpp_list, name, NULL, 0);
1708 if (dim == &dim_dso)
1711 perf_hpp_list__register_sort_field(hpp_list, &c2c_fmt->fmt);
1715 #define PARSE_LIST(_list, _fn) \
1723 for (tok = strtok_r((char *)_list, ", ", &tmp); \
1724 tok; tok = strtok_r(NULL, ", ", &tmp)) { \
1725 ret = _fn(hpp_list, tok); \
1726 if (ret == -EINVAL) { \
1727 error("Invalid --fields key: `%s'", tok); \
1729 } else if (ret == -ESRCH) { \
1730 error("Unknown --fields key: `%s'", tok); \
1736 static int hpp_list__parse(struct perf_hpp_list *hpp_list,
1737 const char *output_,
1740 char *output = output_ ? strdup(output_) : NULL;
1741 char *sort = sort_ ? strdup(sort_) : NULL;
1744 PARSE_LIST(output, c2c_hists__init_output);
1745 PARSE_LIST(sort, c2c_hists__init_sort);
1747 /* copy sort keys to output fields */
1748 perf_hpp__setup_output_field(hpp_list);
1751 * We dont need other sorting keys other than those
1752 * we already specified. It also really slows down
1753 * the processing a lot with big number of output
1754 * fields, so switching this off for c2c.
1758 /* and then copy output fields to sort keys */
1759 perf_hpp__append_sort_keys(&hists->list);
1767 static int c2c_hists__init(struct c2c_hists *hists,
1769 int nr_header_lines)
1771 __hists__init(&hists->hists, &hists->list);
1774 * Initialize only with sort fields, we need to resort
1775 * later anyway, and that's where we add output fields
1778 perf_hpp_list__init(&hists->list);
1780 /* Overload number of header lines.*/
1781 hists->list.nr_header_lines = nr_header_lines;
1783 return hpp_list__parse(&hists->list, NULL, sort);
1786 static int c2c_hists__reinit(struct c2c_hists *c2c_hists,
1790 perf_hpp__reset_output_field(&c2c_hists->list);
1791 return hpp_list__parse(&c2c_hists->list, output, sort);
1794 #define DISPLAY_LINE_LIMIT 0.0005
1796 static bool he__display(struct hist_entry *he, struct c2c_stats *stats)
1798 struct c2c_hist_entry *c2c_he;
1804 c2c_he = container_of(he, struct c2c_hist_entry, he);
1806 #define FILTER_HITM(__h) \
1808 ld_dist = ((double)c2c_he->stats.__h / stats->__h); \
1809 if (ld_dist < DISPLAY_LINE_LIMIT) \
1810 he->filtered = HIST_FILTER__C2C; \
1812 he->filtered = HIST_FILTER__C2C; \
1815 switch (c2c.display) {
1817 FILTER_HITM(lcl_hitm);
1820 FILTER_HITM(rmt_hitm);
1823 FILTER_HITM(tot_hitm);
1830 return he->filtered == 0;
1833 static inline int valid_hitm_or_store(struct hist_entry *he)
1835 struct c2c_hist_entry *c2c_he;
1838 c2c_he = container_of(he, struct c2c_hist_entry, he);
1839 has_hitm = c2c.display == DISPLAY_TOT ? c2c_he->stats.tot_hitm :
1840 c2c.display == DISPLAY_LCL ? c2c_he->stats.lcl_hitm :
1841 c2c_he->stats.rmt_hitm;
1842 return has_hitm || c2c_he->stats.store;
1845 static void calc_width(struct hist_entry *he)
1847 struct c2c_hists *c2c_hists;
1849 c2c_hists = container_of(he->hists, struct c2c_hists, hists);
1850 hists__calc_col_len(&c2c_hists->hists, he);
1853 static int filter_cb(struct hist_entry *he)
1855 if (c2c.show_src && !he->srcline)
1856 he->srcline = hist_entry__get_srcline(he);
1860 if (!valid_hitm_or_store(he))
1861 he->filtered = HIST_FILTER__C2C;
1866 static int resort_cl_cb(struct hist_entry *he)
1868 struct c2c_hist_entry *c2c_he;
1869 struct c2c_hists *c2c_hists;
1870 bool display = he__display(he, &c2c.hitm_stats);
1872 c2c_he = container_of(he, struct c2c_hist_entry, he);
1873 c2c_hists = c2c_he->hists;
1877 if (display && c2c_hists) {
1878 static unsigned int idx;
1880 c2c_he->cacheline_idx = idx++;
1882 c2c_hists__reinit(c2c_hists, c2c.cl_output, c2c.cl_resort);
1884 hists__collapse_resort(&c2c_hists->hists, NULL);
1885 hists__output_resort_cb(&c2c_hists->hists, NULL, filter_cb);
1891 static void setup_nodes_header(void)
1893 dim_node.header = header_node[c2c.node_info];
1896 static int setup_nodes(struct perf_session *session)
1898 struct numa_node *n;
1899 unsigned long **nodes;
1903 if (c2c.node_info > 2)
1906 c2c.nodes_cnt = session->header.env.nr_numa_nodes;
1907 c2c.cpus_cnt = session->header.env.nr_cpus_online;
1909 n = session->header.env.numa_nodes;
1913 nodes = zalloc(sizeof(unsigned long *) * c2c.nodes_cnt);
1919 cpu2node = zalloc(sizeof(int) * c2c.cpus_cnt);
1923 for (cpu = 0; cpu < c2c.cpus_cnt; cpu++)
1926 c2c.cpu2node = cpu2node;
1928 for (node = 0; node < c2c.nodes_cnt; node++) {
1929 struct cpu_map *map = n[node].map;
1932 set = bitmap_alloc(c2c.cpus_cnt);
1936 for (cpu = 0; cpu < map->nr; cpu++) {
1937 set_bit(map->map[cpu], set);
1939 if (WARN_ONCE(cpu2node[map->map[cpu]] != -1, "node/cpu topology bug"))
1942 cpu2node[map->map[cpu]] = node;
1948 setup_nodes_header();
1952 #define HAS_HITMS(__h) ((__h)->stats.lcl_hitm || (__h)->stats.rmt_hitm)
1954 static int resort_hitm_cb(struct hist_entry *he)
1956 struct c2c_hist_entry *c2c_he;
1957 c2c_he = container_of(he, struct c2c_hist_entry, he);
1959 if (HAS_HITMS(c2c_he)) {
1960 c2c.shared_clines++;
1961 c2c_add_stats(&c2c.hitm_stats, &c2c_he->stats);
1967 static int hists__iterate_cb(struct hists *hists, hists__resort_cb_t cb)
1969 struct rb_node *next = rb_first(&hists->entries);
1973 struct hist_entry *he;
1975 he = rb_entry(next, struct hist_entry, rb_node);
1979 next = rb_next(&he->rb_node);
1985 static void print_c2c__display_stats(FILE *out)
1988 struct c2c_stats *stats = &c2c.hists.stats;
1990 llc_misses = stats->lcl_dram +
1995 fprintf(out, "=================================================\n");
1996 fprintf(out, " Trace Event Information \n");
1997 fprintf(out, "=================================================\n");
1998 fprintf(out, " Total records : %10d\n", stats->nr_entries);
1999 fprintf(out, " Locked Load/Store Operations : %10d\n", stats->locks);
2000 fprintf(out, " Load Operations : %10d\n", stats->load);
2001 fprintf(out, " Loads - uncacheable : %10d\n", stats->ld_uncache);
2002 fprintf(out, " Loads - IO : %10d\n", stats->ld_io);
2003 fprintf(out, " Loads - Miss : %10d\n", stats->ld_miss);
2004 fprintf(out, " Loads - no mapping : %10d\n", stats->ld_noadrs);
2005 fprintf(out, " Load Fill Buffer Hit : %10d\n", stats->ld_fbhit);
2006 fprintf(out, " Load L1D hit : %10d\n", stats->ld_l1hit);
2007 fprintf(out, " Load L2D hit : %10d\n", stats->ld_l2hit);
2008 fprintf(out, " Load LLC hit : %10d\n", stats->ld_llchit + stats->lcl_hitm);
2009 fprintf(out, " Load Local HITM : %10d\n", stats->lcl_hitm);
2010 fprintf(out, " Load Remote HITM : %10d\n", stats->rmt_hitm);
2011 fprintf(out, " Load Remote HIT : %10d\n", stats->rmt_hit);
2012 fprintf(out, " Load Local DRAM : %10d\n", stats->lcl_dram);
2013 fprintf(out, " Load Remote DRAM : %10d\n", stats->rmt_dram);
2014 fprintf(out, " Load MESI State Exclusive : %10d\n", stats->ld_excl);
2015 fprintf(out, " Load MESI State Shared : %10d\n", stats->ld_shared);
2016 fprintf(out, " Load LLC Misses : %10d\n", llc_misses);
2017 fprintf(out, " LLC Misses to Local DRAM : %10.1f%%\n", ((double)stats->lcl_dram/(double)llc_misses) * 100.);
2018 fprintf(out, " LLC Misses to Remote DRAM : %10.1f%%\n", ((double)stats->rmt_dram/(double)llc_misses) * 100.);
2019 fprintf(out, " LLC Misses to Remote cache (HIT) : %10.1f%%\n", ((double)stats->rmt_hit /(double)llc_misses) * 100.);
2020 fprintf(out, " LLC Misses to Remote cache (HITM) : %10.1f%%\n", ((double)stats->rmt_hitm/(double)llc_misses) * 100.);
2021 fprintf(out, " Store Operations : %10d\n", stats->store);
2022 fprintf(out, " Store - uncacheable : %10d\n", stats->st_uncache);
2023 fprintf(out, " Store - no mapping : %10d\n", stats->st_noadrs);
2024 fprintf(out, " Store L1D Hit : %10d\n", stats->st_l1hit);
2025 fprintf(out, " Store L1D Miss : %10d\n", stats->st_l1miss);
2026 fprintf(out, " No Page Map Rejects : %10d\n", stats->nomap);
2027 fprintf(out, " Unable to parse data source : %10d\n", stats->noparse);
2030 static void print_shared_cacheline_info(FILE *out)
2032 struct c2c_stats *stats = &c2c.hitm_stats;
2033 int hitm_cnt = stats->lcl_hitm + stats->rmt_hitm;
2035 fprintf(out, "=================================================\n");
2036 fprintf(out, " Global Shared Cache Line Event Information \n");
2037 fprintf(out, "=================================================\n");
2038 fprintf(out, " Total Shared Cache Lines : %10d\n", c2c.shared_clines);
2039 fprintf(out, " Load HITs on shared lines : %10d\n", stats->load);
2040 fprintf(out, " Fill Buffer Hits on shared lines : %10d\n", stats->ld_fbhit);
2041 fprintf(out, " L1D hits on shared lines : %10d\n", stats->ld_l1hit);
2042 fprintf(out, " L2D hits on shared lines : %10d\n", stats->ld_l2hit);
2043 fprintf(out, " LLC hits on shared lines : %10d\n", stats->ld_llchit + stats->lcl_hitm);
2044 fprintf(out, " Locked Access on shared lines : %10d\n", stats->locks);
2045 fprintf(out, " Store HITs on shared lines : %10d\n", stats->store);
2046 fprintf(out, " Store L1D hits on shared lines : %10d\n", stats->st_l1hit);
2047 fprintf(out, " Total Merged records : %10d\n", hitm_cnt + stats->store);
2050 static void print_cacheline(struct c2c_hists *c2c_hists,
2051 struct hist_entry *he_cl,
2052 struct perf_hpp_list *hpp_list,
2056 struct perf_hpp hpp = {
2063 hists__fprintf_headers(&c2c_hists->hists, out);
2069 fprintf(out, " -------------------------------------------------------------\n");
2070 __hist_entry__snprintf(he_cl, &hpp, hpp_list);
2071 fprintf(out, "%s\n", bf);
2072 fprintf(out, " -------------------------------------------------------------\n");
2074 hists__fprintf(&c2c_hists->hists, false, 0, 0, 0, out, true);
2077 static void print_pareto(FILE *out)
2079 struct perf_hpp_list hpp_list;
2083 perf_hpp_list__init(&hpp_list);
2084 ret = hpp_list__parse(&hpp_list,
2093 if (WARN_ONCE(ret, "failed to setup sort entries\n"))
2096 nd = rb_first(&c2c.hists.hists.entries);
2098 for (; nd; nd = rb_next(nd)) {
2099 struct hist_entry *he = rb_entry(nd, struct hist_entry, rb_node);
2100 struct c2c_hist_entry *c2c_he;
2105 c2c_he = container_of(he, struct c2c_hist_entry, he);
2106 print_cacheline(c2c_he->hists, he, &hpp_list, out);
2110 static void print_c2c_info(FILE *out, struct perf_session *session)
2112 struct perf_evlist *evlist = session->evlist;
2113 struct perf_evsel *evsel;
2116 fprintf(out, "=================================================\n");
2117 fprintf(out, " c2c details \n");
2118 fprintf(out, "=================================================\n");
2120 evlist__for_each_entry(evlist, evsel) {
2121 fprintf(out, "%-36s: %s\n", first ? " Events" : "",
2122 perf_evsel__name(evsel));
2125 fprintf(out, " Cachelines sort on : %s HITMs\n",
2126 display_str[c2c.display]);
2127 fprintf(out, " Cacheline data grouping : %s\n", c2c.cl_sort);
2130 static void perf_c2c__hists_fprintf(FILE *out, struct perf_session *session)
2134 print_c2c__display_stats(out);
2136 print_shared_cacheline_info(out);
2138 print_c2c_info(out, session);
2144 fprintf(out, "=================================================\n");
2145 fprintf(out, " Shared Data Cache Line Table \n");
2146 fprintf(out, "=================================================\n");
2147 fprintf(out, "#\n");
2149 hists__fprintf(&c2c.hists.hists, true, 0, 0, 0, stdout, false);
2152 fprintf(out, "=================================================\n");
2153 fprintf(out, " Shared Cache Line Distribution Pareto \n");
2154 fprintf(out, "=================================================\n");
2155 fprintf(out, "#\n");
2160 #ifdef HAVE_SLANG_SUPPORT
2161 static void c2c_browser__update_nr_entries(struct hist_browser *hb)
2164 struct rb_node *nd = rb_first(&hb->hists->entries);
2167 struct hist_entry *he = rb_entry(nd, struct hist_entry, rb_node);
2175 hb->nr_non_filtered_entries = nr_entries;
2178 struct c2c_cacheline_browser {
2179 struct hist_browser hb;
2180 struct hist_entry *he;
2184 perf_c2c_cacheline_browser__title(struct hist_browser *browser,
2185 char *bf, size_t size)
2187 struct c2c_cacheline_browser *cl_browser;
2188 struct hist_entry *he;
2191 cl_browser = container_of(browser, struct c2c_cacheline_browser, hb);
2192 he = cl_browser->he;
2195 addr = cl_address(he->mem_info->daddr.addr);
2197 scnprintf(bf, size, "Cacheline 0x%lx", addr);
2201 static struct c2c_cacheline_browser*
2202 c2c_cacheline_browser__new(struct hists *hists, struct hist_entry *he)
2204 struct c2c_cacheline_browser *browser;
2206 browser = zalloc(sizeof(*browser));
2208 hist_browser__init(&browser->hb, hists);
2209 browser->hb.c2c_filter = true;
2210 browser->hb.title = perf_c2c_cacheline_browser__title;
2217 static int perf_c2c__browse_cacheline(struct hist_entry *he)
2219 struct c2c_hist_entry *c2c_he;
2220 struct c2c_hists *c2c_hists;
2221 struct c2c_cacheline_browser *cl_browser;
2222 struct hist_browser *browser;
2225 " ENTER Togle callchains (if present) \n"
2226 " n Togle Node details info \n"
2227 " s Togle full lenght of symbol and source line columns \n"
2228 " q Return back to cacheline list \n";
2230 /* Display compact version first. */
2231 c2c.symbol_full = false;
2233 c2c_he = container_of(he, struct c2c_hist_entry, he);
2234 c2c_hists = c2c_he->hists;
2236 cl_browser = c2c_cacheline_browser__new(&c2c_hists->hists, he);
2237 if (cl_browser == NULL)
2240 browser = &cl_browser->hb;
2242 /* reset abort key so that it can get Ctrl-C as a key */
2244 SLang_init_tty(0, 0, 0);
2246 c2c_browser__update_nr_entries(browser);
2249 key = hist_browser__run(browser, "? - help");
2253 c2c.symbol_full = !c2c.symbol_full;
2256 c2c.node_info = (c2c.node_info + 1) % 3;
2257 setup_nodes_header();
2262 ui_browser__help_window(&browser->b, help);
2274 static int perf_c2c_browser__title(struct hist_browser *browser,
2275 char *bf, size_t size)
2278 "Shared Data Cache Line Table "
2279 "(%lu entries, sorted on %s HITMs)",
2280 browser->nr_non_filtered_entries,
2281 display_str[c2c.display]);
2285 static struct hist_browser*
2286 perf_c2c_browser__new(struct hists *hists)
2288 struct hist_browser *browser = hist_browser__new(hists);
2291 browser->title = perf_c2c_browser__title;
2292 browser->c2c_filter = true;
2298 static int perf_c2c__hists_browse(struct hists *hists)
2300 struct hist_browser *browser;
2303 " d Display cacheline details \n"
2304 " ENTER Togle callchains (if present) \n"
2307 browser = perf_c2c_browser__new(hists);
2308 if (browser == NULL)
2311 /* reset abort key so that it can get Ctrl-C as a key */
2313 SLang_init_tty(0, 0, 0);
2315 c2c_browser__update_nr_entries(browser);
2318 key = hist_browser__run(browser, "? - help");
2324 perf_c2c__browse_cacheline(browser->he_selection);
2327 ui_browser__help_window(&browser->b, help);
2335 hist_browser__delete(browser);
2339 static void perf_c2c_display(struct perf_session *session)
2341 if (use_browser == 0)
2342 perf_c2c__hists_fprintf(stdout, session);
2344 perf_c2c__hists_browse(&c2c.hists.hists);
2347 static void perf_c2c_display(struct perf_session *session)
2350 perf_c2c__hists_fprintf(stdout, session);
2352 #endif /* HAVE_SLANG_SUPPORT */
2354 static void ui_quirks(void)
2356 if (!c2c.use_stdio) {
2357 dim_offset.width = 5;
2358 dim_offset.header = header_offset_tui;
2361 dim_percent_hitm.header = percent_hitm_header[c2c.display];
2364 #define CALLCHAIN_DEFAULT_OPT "graph,0.5,caller,function,percent"
2366 const char callchain_help[] = "Display call graph (stack chain/backtrace):\n\n"
2367 CALLCHAIN_REPORT_HELP
2368 "\n\t\t\t\tDefault: " CALLCHAIN_DEFAULT_OPT;
2371 parse_callchain_opt(const struct option *opt, const char *arg, int unset)
2373 struct callchain_param *callchain = opt->value;
2375 callchain->enabled = !unset;
2380 symbol_conf.use_callchain = false;
2381 callchain->mode = CHAIN_NONE;
2385 return parse_callchain_report_opt(arg);
2388 static int setup_callchain(struct perf_evlist *evlist)
2390 u64 sample_type = perf_evlist__combined_sample_type(evlist);
2391 enum perf_call_graph_mode mode = CALLCHAIN_NONE;
2393 if ((sample_type & PERF_SAMPLE_REGS_USER) &&
2394 (sample_type & PERF_SAMPLE_STACK_USER))
2395 mode = CALLCHAIN_DWARF;
2396 else if (sample_type & PERF_SAMPLE_BRANCH_STACK)
2397 mode = CALLCHAIN_LBR;
2398 else if (sample_type & PERF_SAMPLE_CALLCHAIN)
2399 mode = CALLCHAIN_FP;
2401 if (!callchain_param.enabled &&
2402 callchain_param.mode != CHAIN_NONE &&
2403 mode != CALLCHAIN_NONE) {
2404 symbol_conf.use_callchain = true;
2405 if (callchain_register_param(&callchain_param) < 0) {
2406 ui__error("Can't register callchain params.\n");
2411 callchain_param.record_mode = mode;
2412 callchain_param.min_percent = 0;
2416 static int setup_display(const char *str)
2418 const char *display = str ?: "tot";
2420 if (!strcmp(display, "tot"))
2421 c2c.display = DISPLAY_TOT;
2422 else if (!strcmp(display, "rmt"))
2423 c2c.display = DISPLAY_RMT;
2424 else if (!strcmp(display, "lcl"))
2425 c2c.display = DISPLAY_LCL;
2427 pr_err("failed: unknown display type: %s\n", str);
2434 #define for_each_token(__tok, __buf, __sep, __tmp) \
2435 for (__tok = strtok_r(__buf, __sep, &__tmp); __tok; \
2436 __tok = strtok_r(NULL, __sep, &__tmp))
2438 static int build_cl_output(char *cl_sort, bool no_source)
2440 char *tok, *tmp, *buf = strdup(cl_sort);
2441 bool add_pid = false;
2442 bool add_tid = false;
2443 bool add_iaddr = false;
2444 bool add_sym = false;
2445 bool add_dso = false;
2446 bool add_src = false;
2451 for_each_token(tok, buf, ",", tmp) {
2452 if (!strcmp(tok, "tid")) {
2454 } else if (!strcmp(tok, "pid")) {
2456 } else if (!strcmp(tok, "iaddr")) {
2460 add_src = no_source ? false : true;
2461 } else if (!strcmp(tok, "dso")) {
2463 } else if (strcmp(tok, "offset")) {
2464 pr_err("unrecognized sort token: %s\n", tok);
2469 if (asprintf(&c2c.cl_output,
2470 "%s%s%s%s%s%s%s%s%s%s",
2471 c2c.use_stdio ? "cl_num_empty," : "",
2474 "percent_stores_l1hit,"
2475 "percent_stores_l1miss,"
2477 add_pid ? "pid," : "",
2478 add_tid ? "tid," : "",
2479 add_iaddr ? "iaddr," : "",
2485 add_sym ? "symbol," : "",
2486 add_dso ? "dso," : "",
2487 add_src ? "cl_srcline," : "",
2491 c2c.show_src = add_src;
2497 static int setup_coalesce(const char *coalesce, bool no_source)
2499 const char *c = coalesce ?: coalesce_default;
2501 if (asprintf(&c2c.cl_sort, "offset,%s", c) < 0)
2504 if (build_cl_output(c2c.cl_sort, no_source))
2507 if (asprintf(&c2c.cl_resort, "offset,%s",
2508 c2c.display == DISPLAY_TOT ?
2510 c2c.display == DISPLAY_RMT ?
2511 "rmt_hitm,lcl_hitm" :
2512 "lcl_hitm,rmt_hitm") < 0)
2515 pr_debug("coalesce sort fields: %s\n", c2c.cl_sort);
2516 pr_debug("coalesce resort fields: %s\n", c2c.cl_resort);
2517 pr_debug("coalesce output fields: %s\n", c2c.cl_output);
2521 static int perf_c2c__report(int argc, const char **argv)
2523 struct perf_session *session;
2524 struct ui_progress prog;
2525 struct perf_data_file file = {
2526 .mode = PERF_DATA_MODE_READ,
2528 char callchain_default_opt[] = CALLCHAIN_DEFAULT_OPT;
2529 const char *display = NULL;
2530 const char *coalesce = NULL;
2531 bool no_source = false;
2532 const struct option options[] = {
2533 OPT_STRING('k', "vmlinux", &symbol_conf.vmlinux_name,
2534 "file", "vmlinux pathname"),
2535 OPT_STRING('i', "input", &input_name, "file",
2536 "the input file to process"),
2537 OPT_INCR('N', "node-info", &c2c.node_info,
2538 "show extra node info in report (repeat for more info)"),
2539 #ifdef HAVE_SLANG_SUPPORT
2540 OPT_BOOLEAN(0, "stdio", &c2c.use_stdio, "Use the stdio interface"),
2542 OPT_BOOLEAN(0, "stats", &c2c.stats_only,
2543 "Display only statistic tables (implies --stdio)"),
2544 OPT_BOOLEAN(0, "full-symbols", &c2c.symbol_full,
2545 "Display full length of symbols"),
2546 OPT_BOOLEAN(0, "no-source", &no_source,
2547 "Do not display Source Line column"),
2548 OPT_BOOLEAN(0, "show-all", &c2c.show_all,
2549 "Show all captured HITM lines."),
2550 OPT_CALLBACK_DEFAULT('g', "call-graph", &callchain_param,
2551 "print_type,threshold[,print_limit],order,sort_key[,branch],value",
2552 callchain_help, &parse_callchain_opt,
2553 callchain_default_opt),
2554 OPT_STRING('d', "display", &display, "Switch HITM output type", "lcl,rmt"),
2555 OPT_STRING('c', "coalesce", &coalesce, "coalesce fields",
2556 "coalesce fields: pid,tid,iaddr,dso"),
2557 OPT_BOOLEAN('f', "force", &symbol_conf.force, "don't complain, do it"),
2558 OPT_PARENT(c2c_options),
2563 argc = parse_options(argc, argv, options, report_c2c_usage,
2564 PARSE_OPT_STOP_AT_NON_OPTION);
2566 usage_with_options(report_c2c_usage, options);
2569 c2c.use_stdio = true;
2571 if (!input_name || !strlen(input_name))
2572 input_name = "perf.data";
2574 file.path = input_name;
2575 file.force = symbol_conf.force;
2577 err = setup_display(display);
2581 err = setup_coalesce(coalesce, no_source);
2583 pr_debug("Failed to initialize hists\n");
2587 err = c2c_hists__init(&c2c.hists, "dcacheline", 2);
2589 pr_debug("Failed to initialize hists\n");
2593 session = perf_session__new(&file, 0, &c2c.tool);
2594 if (session == NULL) {
2595 pr_debug("No memory for session\n");
2599 err = setup_nodes(session);
2601 pr_err("Failed setup nodes\n");
2605 err = setup_callchain(session->evlist);
2609 if (symbol__init(&session->header.env) < 0)
2612 /* No pipe support at the moment. */
2613 if (perf_data_file__is_pipe(session->file)) {
2614 pr_debug("No pipe support at the moment.\n");
2623 setup_browser(false);
2625 err = perf_session__process_events(session);
2627 pr_err("failed to process sample\n");
2631 c2c_hists__reinit(&c2c.hists,
2636 "tot_hitm,lcl_hitm,rmt_hitm,"
2637 "stores,stores_l1hit,stores_l1miss,"
2638 "dram_lcl,dram_rmt,"
2641 "ld_fbhit,ld_l1hit,ld_l2hit,"
2642 "ld_lclhit,ld_rmthit",
2643 c2c.display == DISPLAY_TOT ? "tot_hitm" :
2644 c2c.display == DISPLAY_LCL ? "lcl_hitm" : "rmt_hitm"
2647 ui_progress__init(&prog, c2c.hists.hists.nr_entries, "Sorting...");
2649 hists__collapse_resort(&c2c.hists.hists, NULL);
2650 hists__output_resort_cb(&c2c.hists.hists, &prog, resort_hitm_cb);
2651 hists__iterate_cb(&c2c.hists.hists, resort_cl_cb);
2653 ui_progress__finish();
2657 perf_c2c_display(session);
2660 perf_session__delete(session);
2665 static int parse_record_events(const struct option *opt,
2666 const char *str, int unset __maybe_unused)
2668 bool *event_set = (bool *) opt->value;
2671 return perf_mem_events__parse(str);
2675 static const char * const __usage_record[] = {
2676 "perf c2c record [<options>] [<command>]",
2677 "perf c2c record [<options>] -- <command> [<options>]",
2681 static const char * const *record_mem_usage = __usage_record;
2683 static int perf_c2c__record(int argc, const char **argv)
2685 int rec_argc, i = 0, j;
2686 const char **rec_argv;
2688 bool all_user = false, all_kernel = false;
2689 bool event_set = false;
2690 struct option options[] = {
2691 OPT_CALLBACK('e', "event", &event_set, "event",
2692 "event selector. Use 'perf mem record -e list' to list available events",
2693 parse_record_events),
2694 OPT_BOOLEAN('u', "all-user", &all_user, "collect only user level data"),
2695 OPT_BOOLEAN('k', "all-kernel", &all_kernel, "collect only kernel level data"),
2696 OPT_UINTEGER('l', "ldlat", &perf_mem_events__loads_ldlat, "setup mem-loads latency"),
2697 OPT_PARENT(c2c_options),
2701 if (perf_mem_events__init()) {
2702 pr_err("failed: memory events not supported\n");
2706 argc = parse_options(argc, argv, options, record_mem_usage,
2707 PARSE_OPT_KEEP_UNKNOWN);
2709 rec_argc = argc + 10; /* max number of arguments */
2710 rec_argv = calloc(rec_argc + 1, sizeof(char *));
2714 rec_argv[i++] = "record";
2717 perf_mem_events[PERF_MEM_EVENTS__LOAD].record = true;
2718 perf_mem_events[PERF_MEM_EVENTS__STORE].record = true;
2721 if (perf_mem_events[PERF_MEM_EVENTS__LOAD].record)
2722 rec_argv[i++] = "-W";
2724 rec_argv[i++] = "-d";
2725 rec_argv[i++] = "--sample-cpu";
2727 for (j = 0; j < PERF_MEM_EVENTS__MAX; j++) {
2728 if (!perf_mem_events[j].record)
2731 if (!perf_mem_events[j].supported) {
2732 pr_err("failed: event '%s' not supported\n",
2733 perf_mem_events[j].name);
2737 rec_argv[i++] = "-e";
2738 rec_argv[i++] = perf_mem_events__name(j);
2742 rec_argv[i++] = "--all-user";
2745 rec_argv[i++] = "--all-kernel";
2747 for (j = 0; j < argc; j++, i++)
2748 rec_argv[i] = argv[j];
2751 pr_debug("calling: ");
2755 while (rec_argv[j]) {
2756 pr_debug("%s ", rec_argv[j]);
2762 ret = cmd_record(i, rec_argv);
2767 int cmd_c2c(int argc, const char **argv)
2769 argc = parse_options(argc, argv, c2c_options, c2c_usage,
2770 PARSE_OPT_STOP_AT_NON_OPTION);
2773 usage_with_options(c2c_usage, c2c_options);
2775 if (!strncmp(argv[0], "rec", 3)) {
2776 return perf_c2c__record(argc, argv);
2777 } else if (!strncmp(argv[0], "rep", 3)) {
2778 return perf_c2c__report(argc, argv);
2780 usage_with_options(c2c_usage, c2c_options);