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/err.h>
17 #include <linux/kernel.h>
18 #include <linux/stringify.h>
19 #include <linux/zalloc.h>
21 #include <sys/param.h>
24 #include <perf/cpumap.h>
25 #include <subcmd/pager.h>
26 #include <subcmd/parse-options.h>
27 #include "map_symbol.h"
28 #include "mem-events.h"
33 #include "cacheline.h"
38 #include "ui/browsers/hists.h"
43 #include "ui/progress.h"
48 struct perf_hpp_list list;
49 struct c2c_stats stats;
52 struct compute_stats {
53 struct stats lcl_hitm;
54 struct stats rmt_hitm;
58 struct c2c_hist_entry {
59 struct c2c_hists *hists;
60 struct c2c_stats stats;
61 unsigned long *cpuset;
62 unsigned long *nodeset;
63 struct c2c_stats *node_stats;
64 unsigned int cacheline_idx;
66 struct compute_stats cstats;
69 unsigned long paddr_cnt;
75 * because of its callchain dynamic entry
80 static char const *coalesce_default = "iaddr";
83 struct perf_tool tool;
84 struct c2c_hists hists;
85 struct mem2node mem2node;
87 unsigned long **nodes;
100 /* Shared cache line stats */
101 struct c2c_stats shared_clines_stats;
106 const char *coalesce;
119 static const char *display_str[DISPLAY_MAX] = {
120 [DISPLAY_LCL] = "Local",
121 [DISPLAY_RMT] = "Remote",
122 [DISPLAY_TOT] = "Total",
125 static const struct option c2c_options[] = {
126 OPT_INCR('v', "verbose", &verbose, "be more verbose (show counter open errors, etc)"),
130 static struct perf_c2c c2c;
132 static void *c2c_he_zalloc(size_t size)
134 struct c2c_hist_entry *c2c_he;
136 c2c_he = zalloc(size + sizeof(*c2c_he));
140 c2c_he->cpuset = bitmap_alloc(c2c.cpus_cnt);
144 c2c_he->nodeset = bitmap_alloc(c2c.nodes_cnt);
145 if (!c2c_he->nodeset)
148 c2c_he->node_stats = zalloc(c2c.nodes_cnt * sizeof(*c2c_he->node_stats));
149 if (!c2c_he->node_stats)
152 init_stats(&c2c_he->cstats.lcl_hitm);
153 init_stats(&c2c_he->cstats.rmt_hitm);
154 init_stats(&c2c_he->cstats.load);
159 static void c2c_he_free(void *he)
161 struct c2c_hist_entry *c2c_he;
163 c2c_he = container_of(he, struct c2c_hist_entry, he);
165 hists__delete_entries(&c2c_he->hists->hists);
169 free(c2c_he->cpuset);
170 free(c2c_he->nodeset);
171 free(c2c_he->nodestr);
172 free(c2c_he->node_stats);
176 static struct hist_entry_ops c2c_entry_ops = {
177 .new = c2c_he_zalloc,
181 static int c2c_hists__init(struct c2c_hists *hists,
183 int nr_header_lines);
185 static struct c2c_hists*
186 he__get_c2c_hists(struct hist_entry *he,
190 struct c2c_hist_entry *c2c_he;
191 struct c2c_hists *hists;
194 c2c_he = container_of(he, struct c2c_hist_entry, he);
196 return c2c_he->hists;
198 hists = c2c_he->hists = zalloc(sizeof(*hists));
202 ret = c2c_hists__init(hists, sort, nr_header_lines);
211 static void c2c_he__set_cpu(struct c2c_hist_entry *c2c_he,
212 struct perf_sample *sample)
214 if (WARN_ONCE(sample->cpu == (unsigned int) -1,
215 "WARNING: no sample cpu value"))
218 set_bit(sample->cpu, c2c_he->cpuset);
221 static void c2c_he__set_node(struct c2c_hist_entry *c2c_he,
222 struct perf_sample *sample)
226 if (!sample->phys_addr) {
227 c2c_he->paddr_zero = true;
231 node = mem2node__node(&c2c.mem2node, sample->phys_addr);
232 if (WARN_ONCE(node < 0, "WARNING: failed to find node\n"))
235 set_bit(node, c2c_he->nodeset);
237 if (c2c_he->paddr != sample->phys_addr) {
239 c2c_he->paddr = sample->phys_addr;
243 static void compute_stats(struct c2c_hist_entry *c2c_he,
244 struct c2c_stats *stats,
247 struct compute_stats *cstats = &c2c_he->cstats;
250 update_stats(&cstats->rmt_hitm, weight);
251 else if (stats->lcl_hitm)
252 update_stats(&cstats->lcl_hitm, weight);
253 else if (stats->load)
254 update_stats(&cstats->load, weight);
257 static int process_sample_event(struct perf_tool *tool __maybe_unused,
258 union perf_event *event,
259 struct perf_sample *sample,
261 struct machine *machine)
263 struct c2c_hists *c2c_hists = &c2c.hists;
264 struct c2c_hist_entry *c2c_he;
265 struct c2c_stats stats = { .nr_entries = 0, };
266 struct hist_entry *he;
267 struct addr_location al;
268 struct mem_info *mi, *mi_dup;
271 if (machine__resolve(machine, &al, sample) < 0) {
272 pr_debug("problem processing %d event, skipping it.\n",
278 al.thread->lbr_stitch_enable = true;
280 ret = sample__resolve_callchain(sample, &callchain_cursor, NULL,
281 evsel, &al, sysctl_perf_event_max_stack);
285 mi = sample__resolve_mem(sample, &al);
290 * The mi object is released in hists__add_entry_ops,
291 * if it gets sorted out into existing data, so we need
292 * to take the copy now.
294 mi_dup = mem_info__get(mi);
296 c2c_decode_stats(&stats, mi);
298 he = hists__add_entry_ops(&c2c_hists->hists, &c2c_entry_ops,
304 c2c_he = container_of(he, struct c2c_hist_entry, he);
305 c2c_add_stats(&c2c_he->stats, &stats);
306 c2c_add_stats(&c2c_hists->stats, &stats);
308 c2c_he__set_cpu(c2c_he, sample);
309 c2c_he__set_node(c2c_he, sample);
311 hists__inc_nr_samples(&c2c_hists->hists, he->filtered);
312 ret = hist_entry__append_callchain(he, sample);
316 * There's already been warning about missing
317 * sample's cpu value. Let's account all to
318 * node 0 in this case, without any further
321 * Doing node stats only for single callchain data.
323 int cpu = sample->cpu == (unsigned int) -1 ? 0 : sample->cpu;
324 int node = c2c.cpu2node[cpu];
328 c2c_hists = he__get_c2c_hists(he, c2c.cl_sort, 2);
332 he = hists__add_entry_ops(&c2c_hists->hists, &c2c_entry_ops,
338 c2c_he = container_of(he, struct c2c_hist_entry, he);
339 c2c_add_stats(&c2c_he->stats, &stats);
340 c2c_add_stats(&c2c_hists->stats, &stats);
341 c2c_add_stats(&c2c_he->node_stats[node], &stats);
343 compute_stats(c2c_he, &stats, sample->weight);
345 c2c_he__set_cpu(c2c_he, sample);
346 c2c_he__set_node(c2c_he, sample);
348 hists__inc_nr_samples(&c2c_hists->hists, he->filtered);
349 ret = hist_entry__append_callchain(he, sample);
353 addr_location__put(&al);
357 mem_info__put(mi_dup);
363 static struct perf_c2c c2c = {
365 .sample = process_sample_event,
366 .mmap = perf_event__process_mmap,
367 .mmap2 = perf_event__process_mmap2,
368 .comm = perf_event__process_comm,
369 .exit = perf_event__process_exit,
370 .fork = perf_event__process_fork,
371 .lost = perf_event__process_lost,
372 .attr = perf_event__process_attr,
373 .auxtrace_info = perf_event__process_auxtrace_info,
374 .auxtrace = perf_event__process_auxtrace,
375 .auxtrace_error = perf_event__process_auxtrace_error,
376 .ordered_events = true,
377 .ordering_requires_timestamps = true,
381 static const char * const c2c_usage[] = {
382 "perf c2c {record|report}",
386 static const char * const __usage_report[] = {
391 static const char * const *report_c2c_usage = __usage_report;
393 #define C2C_HEADER_MAX 2
399 } line[C2C_HEADER_MAX];
402 struct c2c_dimension {
403 struct c2c_header header;
406 struct sort_entry *se;
408 int64_t (*cmp)(struct perf_hpp_fmt *fmt,
409 struct hist_entry *, struct hist_entry *);
410 int (*entry)(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
411 struct hist_entry *he);
412 int (*color)(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
413 struct hist_entry *he);
417 struct perf_hpp_fmt fmt;
418 struct c2c_dimension *dim;
421 #define SYMBOL_WIDTH 30
423 static struct c2c_dimension dim_symbol;
424 static struct c2c_dimension dim_srcline;
426 static int symbol_width(struct hists *hists, struct sort_entry *se)
428 int width = hists__col_len(hists, se->se_width_idx);
430 if (!c2c.symbol_full)
431 width = MIN(width, SYMBOL_WIDTH);
436 static int c2c_width(struct perf_hpp_fmt *fmt,
437 struct perf_hpp *hpp __maybe_unused,
440 struct c2c_fmt *c2c_fmt;
441 struct c2c_dimension *dim;
443 c2c_fmt = container_of(fmt, struct c2c_fmt, fmt);
446 if (dim == &dim_symbol || dim == &dim_srcline)
447 return symbol_width(hists, dim->se);
449 return dim->se ? hists__col_len(hists, dim->se->se_width_idx) :
453 static int c2c_header(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
454 struct hists *hists, int line, int *span)
456 struct perf_hpp_list *hpp_list = hists->hpp_list;
457 struct c2c_fmt *c2c_fmt;
458 struct c2c_dimension *dim;
459 const char *text = NULL;
460 int width = c2c_width(fmt, hpp, hists);
462 c2c_fmt = container_of(fmt, struct c2c_fmt, fmt);
466 text = dim->header.line[line].text;
467 /* Use the last line from sort_entry if not defined. */
468 if (!text && (line == hpp_list->nr_header_lines - 1))
469 text = dim->se->se_header;
471 text = dim->header.line[line].text;
477 *span = dim->header.line[line].span;
484 return scnprintf(hpp->buf, hpp->size, "%*s", width, text);
487 #define HEX_STR(__s, __v) \
489 scnprintf(__s, sizeof(__s), "0x%" PRIx64, __v); \
494 dcacheline_cmp(struct perf_hpp_fmt *fmt __maybe_unused,
495 struct hist_entry *left, struct hist_entry *right)
497 return sort__dcacheline_cmp(left, right);
500 static int dcacheline_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
501 struct hist_entry *he)
504 int width = c2c_width(fmt, hpp, he->hists);
508 addr = cl_address(he->mem_info->daddr.addr);
510 return scnprintf(hpp->buf, hpp->size, "%*s", width, HEX_STR(buf, addr));
514 dcacheline_node_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
515 struct hist_entry *he)
517 struct c2c_hist_entry *c2c_he;
518 int width = c2c_width(fmt, hpp, he->hists);
520 c2c_he = container_of(he, struct c2c_hist_entry, he);
521 if (WARN_ON_ONCE(!c2c_he->nodestr))
524 return scnprintf(hpp->buf, hpp->size, "%*s", width, c2c_he->nodestr);
528 dcacheline_node_count(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
529 struct hist_entry *he)
531 struct c2c_hist_entry *c2c_he;
532 int width = c2c_width(fmt, hpp, he->hists);
534 c2c_he = container_of(he, struct c2c_hist_entry, he);
535 return scnprintf(hpp->buf, hpp->size, "%*lu", width, c2c_he->paddr_cnt);
538 static int offset_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
539 struct hist_entry *he)
542 int width = c2c_width(fmt, hpp, he->hists);
546 addr = cl_offset(he->mem_info->daddr.al_addr);
548 return scnprintf(hpp->buf, hpp->size, "%*s", width, HEX_STR(buf, addr));
552 offset_cmp(struct perf_hpp_fmt *fmt __maybe_unused,
553 struct hist_entry *left, struct hist_entry *right)
555 uint64_t l = 0, r = 0;
558 l = cl_offset(left->mem_info->daddr.addr);
560 r = cl_offset(right->mem_info->daddr.addr);
562 return (int64_t)(r - l);
566 iaddr_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
567 struct hist_entry *he)
570 int width = c2c_width(fmt, hpp, he->hists);
574 addr = he->mem_info->iaddr.addr;
576 return scnprintf(hpp->buf, hpp->size, "%*s", width, HEX_STR(buf, addr));
580 iaddr_cmp(struct perf_hpp_fmt *fmt __maybe_unused,
581 struct hist_entry *left, struct hist_entry *right)
583 return sort__iaddr_cmp(left, right);
587 tot_hitm_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
588 struct hist_entry *he)
590 struct c2c_hist_entry *c2c_he;
591 int width = c2c_width(fmt, hpp, he->hists);
592 unsigned int tot_hitm;
594 c2c_he = container_of(he, struct c2c_hist_entry, he);
595 tot_hitm = c2c_he->stats.lcl_hitm + c2c_he->stats.rmt_hitm;
597 return scnprintf(hpp->buf, hpp->size, "%*u", width, tot_hitm);
601 tot_hitm_cmp(struct perf_hpp_fmt *fmt __maybe_unused,
602 struct hist_entry *left, struct hist_entry *right)
604 struct c2c_hist_entry *c2c_left;
605 struct c2c_hist_entry *c2c_right;
606 uint64_t tot_hitm_left;
607 uint64_t tot_hitm_right;
609 c2c_left = container_of(left, struct c2c_hist_entry, he);
610 c2c_right = container_of(right, struct c2c_hist_entry, he);
612 tot_hitm_left = c2c_left->stats.lcl_hitm + c2c_left->stats.rmt_hitm;
613 tot_hitm_right = c2c_right->stats.lcl_hitm + c2c_right->stats.rmt_hitm;
615 return tot_hitm_left - tot_hitm_right;
618 #define STAT_FN_ENTRY(__f) \
620 __f ## _entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp, \
621 struct hist_entry *he) \
623 struct c2c_hist_entry *c2c_he; \
624 int width = c2c_width(fmt, hpp, he->hists); \
626 c2c_he = container_of(he, struct c2c_hist_entry, he); \
627 return scnprintf(hpp->buf, hpp->size, "%*u", width, \
628 c2c_he->stats.__f); \
631 #define STAT_FN_CMP(__f) \
633 __f ## _cmp(struct perf_hpp_fmt *fmt __maybe_unused, \
634 struct hist_entry *left, struct hist_entry *right) \
636 struct c2c_hist_entry *c2c_left, *c2c_right; \
638 c2c_left = container_of(left, struct c2c_hist_entry, he); \
639 c2c_right = container_of(right, struct c2c_hist_entry, he); \
640 return (uint64_t) c2c_left->stats.__f - \
641 (uint64_t) c2c_right->stats.__f; \
644 #define STAT_FN(__f) \
659 static uint64_t total_records(struct c2c_stats *stats)
661 uint64_t lclmiss, ldcnt, total;
663 lclmiss = stats->lcl_dram +
683 tot_recs_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
684 struct hist_entry *he)
686 struct c2c_hist_entry *c2c_he;
687 int width = c2c_width(fmt, hpp, he->hists);
690 c2c_he = container_of(he, struct c2c_hist_entry, he);
691 tot_recs = total_records(&c2c_he->stats);
693 return scnprintf(hpp->buf, hpp->size, "%*" PRIu64, width, tot_recs);
697 tot_recs_cmp(struct perf_hpp_fmt *fmt __maybe_unused,
698 struct hist_entry *left, struct hist_entry *right)
700 struct c2c_hist_entry *c2c_left;
701 struct c2c_hist_entry *c2c_right;
702 uint64_t tot_recs_left;
703 uint64_t tot_recs_right;
705 c2c_left = container_of(left, struct c2c_hist_entry, he);
706 c2c_right = container_of(right, struct c2c_hist_entry, he);
708 tot_recs_left = total_records(&c2c_left->stats);
709 tot_recs_right = total_records(&c2c_right->stats);
711 return tot_recs_left - tot_recs_right;
714 static uint64_t total_loads(struct c2c_stats *stats)
716 uint64_t lclmiss, ldcnt;
718 lclmiss = stats->lcl_dram +
734 tot_loads_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
735 struct hist_entry *he)
737 struct c2c_hist_entry *c2c_he;
738 int width = c2c_width(fmt, hpp, he->hists);
741 c2c_he = container_of(he, struct c2c_hist_entry, he);
742 tot_recs = total_loads(&c2c_he->stats);
744 return scnprintf(hpp->buf, hpp->size, "%*" PRIu64, width, tot_recs);
748 tot_loads_cmp(struct perf_hpp_fmt *fmt __maybe_unused,
749 struct hist_entry *left, struct hist_entry *right)
751 struct c2c_hist_entry *c2c_left;
752 struct c2c_hist_entry *c2c_right;
753 uint64_t tot_recs_left;
754 uint64_t tot_recs_right;
756 c2c_left = container_of(left, struct c2c_hist_entry, he);
757 c2c_right = container_of(right, struct c2c_hist_entry, he);
759 tot_recs_left = total_loads(&c2c_left->stats);
760 tot_recs_right = total_loads(&c2c_right->stats);
762 return tot_recs_left - tot_recs_right;
765 typedef double (get_percent_cb)(struct c2c_hist_entry *);
768 percent_color(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
769 struct hist_entry *he, get_percent_cb get_percent)
771 struct c2c_hist_entry *c2c_he;
772 int width = c2c_width(fmt, hpp, he->hists);
775 c2c_he = container_of(he, struct c2c_hist_entry, he);
776 per = get_percent(c2c_he);
778 #ifdef HAVE_SLANG_SUPPORT
780 return __hpp__slsmg_color_printf(hpp, "%*.2f%%", width - 1, per);
782 return hpp_color_scnprintf(hpp, "%*.2f%%", width - 1, per);
785 static double percent_hitm(struct c2c_hist_entry *c2c_he)
787 struct c2c_hists *hists;
788 struct c2c_stats *stats;
789 struct c2c_stats *total;
793 hists = container_of(c2c_he->he.hists, struct c2c_hists, hists);
794 stats = &c2c_he->stats;
795 total = &hists->stats;
797 switch (c2c.display) {
799 st = stats->rmt_hitm;
800 tot = total->rmt_hitm;
803 st = stats->lcl_hitm;
804 tot = total->lcl_hitm;
807 st = stats->tot_hitm;
808 tot = total->tot_hitm;
813 p = tot ? (double) st / tot : 0;
818 #define PERC_STR(__s, __v) \
820 scnprintf(__s, sizeof(__s), "%.2F%%", __v); \
825 percent_hitm_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
826 struct hist_entry *he)
828 struct c2c_hist_entry *c2c_he;
829 int width = c2c_width(fmt, hpp, he->hists);
833 c2c_he = container_of(he, struct c2c_hist_entry, he);
834 per = percent_hitm(c2c_he);
835 return scnprintf(hpp->buf, hpp->size, "%*s", width, PERC_STR(buf, per));
839 percent_hitm_color(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
840 struct hist_entry *he)
842 return percent_color(fmt, hpp, he, percent_hitm);
846 percent_hitm_cmp(struct perf_hpp_fmt *fmt __maybe_unused,
847 struct hist_entry *left, struct hist_entry *right)
849 struct c2c_hist_entry *c2c_left;
850 struct c2c_hist_entry *c2c_right;
854 c2c_left = container_of(left, struct c2c_hist_entry, he);
855 c2c_right = container_of(right, struct c2c_hist_entry, he);
857 per_left = percent_hitm(c2c_left);
858 per_right = percent_hitm(c2c_right);
860 return per_left - per_right;
863 static struct c2c_stats *he_stats(struct hist_entry *he)
865 struct c2c_hist_entry *c2c_he;
867 c2c_he = container_of(he, struct c2c_hist_entry, he);
868 return &c2c_he->stats;
871 static struct c2c_stats *total_stats(struct hist_entry *he)
873 struct c2c_hists *hists;
875 hists = container_of(he->hists, struct c2c_hists, hists);
876 return &hists->stats;
879 static double percent(u32 st, u32 tot)
881 return tot ? 100. * (double) st / (double) tot : 0;
884 #define PERCENT(__h, __f) percent(he_stats(__h)->__f, total_stats(__h)->__f)
886 #define PERCENT_FN(__f) \
887 static double percent_ ## __f(struct c2c_hist_entry *c2c_he) \
889 struct c2c_hists *hists; \
891 hists = container_of(c2c_he->he.hists, struct c2c_hists, hists); \
892 return percent(c2c_he->stats.__f, hists->stats.__f); \
898 PERCENT_FN(st_l1miss)
901 percent_rmt_hitm_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
902 struct hist_entry *he)
904 int width = c2c_width(fmt, hpp, he->hists);
905 double per = PERCENT(he, rmt_hitm);
908 return scnprintf(hpp->buf, hpp->size, "%*s", width, PERC_STR(buf, per));
912 percent_rmt_hitm_color(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
913 struct hist_entry *he)
915 return percent_color(fmt, hpp, he, percent_rmt_hitm);
919 percent_rmt_hitm_cmp(struct perf_hpp_fmt *fmt __maybe_unused,
920 struct hist_entry *left, struct hist_entry *right)
925 per_left = PERCENT(left, lcl_hitm);
926 per_right = PERCENT(right, lcl_hitm);
928 return per_left - per_right;
932 percent_lcl_hitm_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
933 struct hist_entry *he)
935 int width = c2c_width(fmt, hpp, he->hists);
936 double per = PERCENT(he, lcl_hitm);
939 return scnprintf(hpp->buf, hpp->size, "%*s", width, PERC_STR(buf, per));
943 percent_lcl_hitm_color(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
944 struct hist_entry *he)
946 return percent_color(fmt, hpp, he, percent_lcl_hitm);
950 percent_lcl_hitm_cmp(struct perf_hpp_fmt *fmt __maybe_unused,
951 struct hist_entry *left, struct hist_entry *right)
956 per_left = PERCENT(left, lcl_hitm);
957 per_right = PERCENT(right, lcl_hitm);
959 return per_left - per_right;
963 percent_stores_l1hit_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
964 struct hist_entry *he)
966 int width = c2c_width(fmt, hpp, he->hists);
967 double per = PERCENT(he, st_l1hit);
970 return scnprintf(hpp->buf, hpp->size, "%*s", width, PERC_STR(buf, per));
974 percent_stores_l1hit_color(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
975 struct hist_entry *he)
977 return percent_color(fmt, hpp, he, percent_st_l1hit);
981 percent_stores_l1hit_cmp(struct perf_hpp_fmt *fmt __maybe_unused,
982 struct hist_entry *left, struct hist_entry *right)
987 per_left = PERCENT(left, st_l1hit);
988 per_right = PERCENT(right, st_l1hit);
990 return per_left - per_right;
994 percent_stores_l1miss_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
995 struct hist_entry *he)
997 int width = c2c_width(fmt, hpp, he->hists);
998 double per = PERCENT(he, st_l1miss);
1001 return scnprintf(hpp->buf, hpp->size, "%*s", width, PERC_STR(buf, per));
1005 percent_stores_l1miss_color(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
1006 struct hist_entry *he)
1008 return percent_color(fmt, hpp, he, percent_st_l1miss);
1012 percent_stores_l1miss_cmp(struct perf_hpp_fmt *fmt __maybe_unused,
1013 struct hist_entry *left, struct hist_entry *right)
1018 per_left = PERCENT(left, st_l1miss);
1019 per_right = PERCENT(right, st_l1miss);
1021 return per_left - per_right;
1028 pid_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
1029 struct hist_entry *he)
1031 int width = c2c_width(fmt, hpp, he->hists);
1033 return scnprintf(hpp->buf, hpp->size, "%*d", width, he->thread->pid_);
1037 pid_cmp(struct perf_hpp_fmt *fmt __maybe_unused,
1038 struct hist_entry *left, struct hist_entry *right)
1040 return left->thread->pid_ - right->thread->pid_;
1044 empty_cmp(struct perf_hpp_fmt *fmt __maybe_unused,
1045 struct hist_entry *left __maybe_unused,
1046 struct hist_entry *right __maybe_unused)
1051 static int display_metrics(struct perf_hpp *hpp, u32 val, u32 sum)
1056 ret = scnprintf(hpp->buf, hpp->size, "%5.1f%% ",
1059 ret = scnprintf(hpp->buf, hpp->size, "%6s ", "n/a");
1065 node_entry(struct perf_hpp_fmt *fmt __maybe_unused, struct perf_hpp *hpp,
1066 struct hist_entry *he)
1068 struct c2c_hist_entry *c2c_he;
1073 c2c_he = container_of(he, struct c2c_hist_entry, he);
1075 for (node = 0; node < c2c.nodes_cnt; node++) {
1076 DECLARE_BITMAP(set, c2c.cpus_cnt);
1078 bitmap_zero(set, c2c.cpus_cnt);
1079 bitmap_and(set, c2c_he->cpuset, c2c.nodes[node], c2c.cpus_cnt);
1081 if (!bitmap_weight(set, c2c.cpus_cnt)) {
1082 if (c2c.node_info == 1) {
1083 ret = scnprintf(hpp->buf, hpp->size, "%21s", " ");
1084 advance_hpp(hpp, ret);
1090 ret = scnprintf(hpp->buf, hpp->size, " ");
1091 advance_hpp(hpp, ret);
1094 switch (c2c.node_info) {
1096 ret = scnprintf(hpp->buf, hpp->size, "%2d", node);
1097 advance_hpp(hpp, ret);
1101 int num = bitmap_weight(set, c2c.cpus_cnt);
1102 struct c2c_stats *stats = &c2c_he->node_stats[node];
1104 ret = scnprintf(hpp->buf, hpp->size, "%2d{%2d ", node, num);
1105 advance_hpp(hpp, ret);
1107 switch (c2c.display) {
1109 ret = display_metrics(hpp, stats->rmt_hitm,
1110 c2c_he->stats.rmt_hitm);
1113 ret = display_metrics(hpp, stats->lcl_hitm,
1114 c2c_he->stats.lcl_hitm);
1117 ret = display_metrics(hpp, stats->tot_hitm,
1118 c2c_he->stats.tot_hitm);
1124 advance_hpp(hpp, ret);
1126 if (c2c_he->stats.store > 0) {
1127 ret = scnprintf(hpp->buf, hpp->size, "%5.1f%%}",
1128 percent(stats->store, c2c_he->stats.store));
1130 ret = scnprintf(hpp->buf, hpp->size, "%6s}", "n/a");
1133 advance_hpp(hpp, ret);
1137 ret = scnprintf(hpp->buf, hpp->size, "%2d{", node);
1138 advance_hpp(hpp, ret);
1140 ret = bitmap_scnprintf(set, c2c.cpus_cnt, hpp->buf, hpp->size);
1141 advance_hpp(hpp, ret);
1143 ret = scnprintf(hpp->buf, hpp->size, "}");
1144 advance_hpp(hpp, ret);
1157 mean_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
1158 struct hist_entry *he, double mean)
1160 int width = c2c_width(fmt, hpp, he->hists);
1163 scnprintf(buf, 10, "%6.0f", mean);
1164 return scnprintf(hpp->buf, hpp->size, "%*s", width, buf);
1167 #define MEAN_ENTRY(__func, __val) \
1169 __func(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp, struct hist_entry *he) \
1171 struct c2c_hist_entry *c2c_he; \
1172 c2c_he = container_of(he, struct c2c_hist_entry, he); \
1173 return mean_entry(fmt, hpp, he, avg_stats(&c2c_he->cstats.__val)); \
1176 MEAN_ENTRY(mean_rmt_entry, rmt_hitm);
1177 MEAN_ENTRY(mean_lcl_entry, lcl_hitm);
1178 MEAN_ENTRY(mean_load_entry, load);
1181 cpucnt_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
1182 struct hist_entry *he)
1184 struct c2c_hist_entry *c2c_he;
1185 int width = c2c_width(fmt, hpp, he->hists);
1188 c2c_he = container_of(he, struct c2c_hist_entry, he);
1190 scnprintf(buf, 10, "%d", bitmap_weight(c2c_he->cpuset, c2c.cpus_cnt));
1191 return scnprintf(hpp->buf, hpp->size, "%*s", width, buf);
1195 cl_idx_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
1196 struct hist_entry *he)
1198 struct c2c_hist_entry *c2c_he;
1199 int width = c2c_width(fmt, hpp, he->hists);
1202 c2c_he = container_of(he, struct c2c_hist_entry, he);
1204 scnprintf(buf, 10, "%u", c2c_he->cacheline_idx);
1205 return scnprintf(hpp->buf, hpp->size, "%*s", width, buf);
1209 cl_idx_empty_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
1210 struct hist_entry *he)
1212 int width = c2c_width(fmt, hpp, he->hists);
1214 return scnprintf(hpp->buf, hpp->size, "%*s", width, "");
1217 #define HEADER_LOW(__h) \
1224 #define HEADER_BOTH(__h0, __h1) \
1234 #define HEADER_SPAN(__h0, __h1, __s) \
1245 #define HEADER_SPAN_LOW(__h) \
1252 static struct c2c_dimension dim_dcacheline = {
1253 .header = HEADER_SPAN("--- Cacheline ----", "Address", 2),
1254 .name = "dcacheline",
1255 .cmp = dcacheline_cmp,
1256 .entry = dcacheline_entry,
1260 static struct c2c_dimension dim_dcacheline_node = {
1261 .header = HEADER_LOW("Node"),
1262 .name = "dcacheline_node",
1264 .entry = dcacheline_node_entry,
1268 static struct c2c_dimension dim_dcacheline_count = {
1269 .header = HEADER_LOW("PA cnt"),
1270 .name = "dcacheline_count",
1272 .entry = dcacheline_node_count,
1276 static struct c2c_header header_offset_tui = HEADER_SPAN("-----", "Off", 2);
1278 static struct c2c_dimension dim_offset = {
1279 .header = HEADER_SPAN("--- Data address -", "Offset", 2),
1282 .entry = offset_entry,
1286 static struct c2c_dimension dim_offset_node = {
1287 .header = HEADER_LOW("Node"),
1288 .name = "offset_node",
1290 .entry = dcacheline_node_entry,
1294 static struct c2c_dimension dim_iaddr = {
1295 .header = HEADER_LOW("Code address"),
1298 .entry = iaddr_entry,
1302 static struct c2c_dimension dim_tot_hitm = {
1303 .header = HEADER_SPAN("------- Load Hitm -------", "Total", 2),
1305 .cmp = tot_hitm_cmp,
1306 .entry = tot_hitm_entry,
1310 static struct c2c_dimension dim_lcl_hitm = {
1311 .header = HEADER_SPAN_LOW("LclHitm"),
1313 .cmp = lcl_hitm_cmp,
1314 .entry = lcl_hitm_entry,
1318 static struct c2c_dimension dim_rmt_hitm = {
1319 .header = HEADER_SPAN_LOW("RmtHitm"),
1321 .cmp = rmt_hitm_cmp,
1322 .entry = rmt_hitm_entry,
1326 static struct c2c_dimension dim_cl_rmt_hitm = {
1327 .header = HEADER_SPAN("----- HITM -----", "Rmt", 1),
1328 .name = "cl_rmt_hitm",
1329 .cmp = rmt_hitm_cmp,
1330 .entry = rmt_hitm_entry,
1334 static struct c2c_dimension dim_cl_lcl_hitm = {
1335 .header = HEADER_SPAN_LOW("Lcl"),
1336 .name = "cl_lcl_hitm",
1337 .cmp = lcl_hitm_cmp,
1338 .entry = lcl_hitm_entry,
1342 static struct c2c_dimension dim_tot_stores = {
1343 .header = HEADER_BOTH("Total", "Stores"),
1344 .name = "tot_stores",
1346 .entry = store_entry,
1350 static struct c2c_dimension dim_stores_l1hit = {
1351 .header = HEADER_SPAN("---- Stores ----", "L1Hit", 1),
1352 .name = "stores_l1hit",
1353 .cmp = st_l1hit_cmp,
1354 .entry = st_l1hit_entry,
1358 static struct c2c_dimension dim_stores_l1miss = {
1359 .header = HEADER_SPAN_LOW("L1Miss"),
1360 .name = "stores_l1miss",
1361 .cmp = st_l1miss_cmp,
1362 .entry = st_l1miss_entry,
1366 static struct c2c_dimension dim_cl_stores_l1hit = {
1367 .header = HEADER_SPAN("-- Store Refs --", "L1 Hit", 1),
1368 .name = "cl_stores_l1hit",
1369 .cmp = st_l1hit_cmp,
1370 .entry = st_l1hit_entry,
1374 static struct c2c_dimension dim_cl_stores_l1miss = {
1375 .header = HEADER_SPAN_LOW("L1 Miss"),
1376 .name = "cl_stores_l1miss",
1377 .cmp = st_l1miss_cmp,
1378 .entry = st_l1miss_entry,
1382 static struct c2c_dimension dim_ld_fbhit = {
1383 .header = HEADER_SPAN("----- Core Load Hit -----", "FB", 2),
1385 .cmp = ld_fbhit_cmp,
1386 .entry = ld_fbhit_entry,
1390 static struct c2c_dimension dim_ld_l1hit = {
1391 .header = HEADER_SPAN_LOW("L1"),
1393 .cmp = ld_l1hit_cmp,
1394 .entry = ld_l1hit_entry,
1398 static struct c2c_dimension dim_ld_l2hit = {
1399 .header = HEADER_SPAN_LOW("L2"),
1401 .cmp = ld_l2hit_cmp,
1402 .entry = ld_l2hit_entry,
1406 static struct c2c_dimension dim_ld_llchit = {
1407 .header = HEADER_SPAN("- LLC Load Hit --", "LclHit", 1),
1408 .name = "ld_lclhit",
1409 .cmp = ld_llchit_cmp,
1410 .entry = ld_llchit_entry,
1414 static struct c2c_dimension dim_ld_rmthit = {
1415 .header = HEADER_SPAN("- RMT Load Hit --", "RmtHit", 1),
1416 .name = "ld_rmthit",
1418 .entry = rmt_hit_entry,
1422 static struct c2c_dimension dim_tot_recs = {
1423 .header = HEADER_BOTH("Total", "records"),
1425 .cmp = tot_recs_cmp,
1426 .entry = tot_recs_entry,
1430 static struct c2c_dimension dim_tot_loads = {
1431 .header = HEADER_BOTH("Total", "Loads"),
1432 .name = "tot_loads",
1433 .cmp = tot_loads_cmp,
1434 .entry = tot_loads_entry,
1438 static struct c2c_header percent_hitm_header[] = {
1439 [DISPLAY_LCL] = HEADER_BOTH("Lcl", "Hitm"),
1440 [DISPLAY_RMT] = HEADER_BOTH("Rmt", "Hitm"),
1441 [DISPLAY_TOT] = HEADER_BOTH("Tot", "Hitm"),
1444 static struct c2c_dimension dim_percent_hitm = {
1445 .name = "percent_hitm",
1446 .cmp = percent_hitm_cmp,
1447 .entry = percent_hitm_entry,
1448 .color = percent_hitm_color,
1452 static struct c2c_dimension dim_percent_rmt_hitm = {
1453 .header = HEADER_SPAN("----- HITM -----", "RmtHitm", 1),
1454 .name = "percent_rmt_hitm",
1455 .cmp = percent_rmt_hitm_cmp,
1456 .entry = percent_rmt_hitm_entry,
1457 .color = percent_rmt_hitm_color,
1461 static struct c2c_dimension dim_percent_lcl_hitm = {
1462 .header = HEADER_SPAN_LOW("LclHitm"),
1463 .name = "percent_lcl_hitm",
1464 .cmp = percent_lcl_hitm_cmp,
1465 .entry = percent_lcl_hitm_entry,
1466 .color = percent_lcl_hitm_color,
1470 static struct c2c_dimension dim_percent_stores_l1hit = {
1471 .header = HEADER_SPAN("-- Store Refs --", "L1 Hit", 1),
1472 .name = "percent_stores_l1hit",
1473 .cmp = percent_stores_l1hit_cmp,
1474 .entry = percent_stores_l1hit_entry,
1475 .color = percent_stores_l1hit_color,
1479 static struct c2c_dimension dim_percent_stores_l1miss = {
1480 .header = HEADER_SPAN_LOW("L1 Miss"),
1481 .name = "percent_stores_l1miss",
1482 .cmp = percent_stores_l1miss_cmp,
1483 .entry = percent_stores_l1miss_entry,
1484 .color = percent_stores_l1miss_color,
1488 static struct c2c_dimension dim_dram_lcl = {
1489 .header = HEADER_SPAN("--- Load Dram ----", "Lcl", 1),
1491 .cmp = lcl_dram_cmp,
1492 .entry = lcl_dram_entry,
1496 static struct c2c_dimension dim_dram_rmt = {
1497 .header = HEADER_SPAN_LOW("Rmt"),
1499 .cmp = rmt_dram_cmp,
1500 .entry = rmt_dram_entry,
1504 static struct c2c_dimension dim_pid = {
1505 .header = HEADER_LOW("Pid"),
1512 static struct c2c_dimension dim_tid = {
1513 .header = HEADER_LOW("Tid"),
1518 static struct c2c_dimension dim_symbol = {
1523 static struct c2c_dimension dim_dso = {
1524 .header = HEADER_BOTH("Shared", "Object"),
1529 static struct c2c_header header_node[3] = {
1531 HEADER_LOW("Node{cpus %hitms %stores}"),
1532 HEADER_LOW("Node{cpu list}"),
1535 static struct c2c_dimension dim_node = {
1538 .entry = node_entry,
1542 static struct c2c_dimension dim_mean_rmt = {
1543 .header = HEADER_SPAN("---------- cycles ----------", "rmt hitm", 2),
1546 .entry = mean_rmt_entry,
1550 static struct c2c_dimension dim_mean_lcl = {
1551 .header = HEADER_SPAN_LOW("lcl hitm"),
1554 .entry = mean_lcl_entry,
1558 static struct c2c_dimension dim_mean_load = {
1559 .header = HEADER_SPAN_LOW("load"),
1560 .name = "mean_load",
1562 .entry = mean_load_entry,
1566 static struct c2c_dimension dim_cpucnt = {
1567 .header = HEADER_BOTH("cpu", "cnt"),
1570 .entry = cpucnt_entry,
1574 static struct c2c_dimension dim_srcline = {
1575 .name = "cl_srcline",
1576 .se = &sort_srcline,
1579 static struct c2c_dimension dim_dcacheline_idx = {
1580 .header = HEADER_LOW("Index"),
1583 .entry = cl_idx_entry,
1587 static struct c2c_dimension dim_dcacheline_num = {
1588 .header = HEADER_LOW("Num"),
1591 .entry = cl_idx_entry,
1595 static struct c2c_dimension dim_dcacheline_num_empty = {
1596 .header = HEADER_LOW("Num"),
1597 .name = "cl_num_empty",
1599 .entry = cl_idx_empty_entry,
1603 static struct c2c_dimension *dimensions[] = {
1605 &dim_dcacheline_node,
1606 &dim_dcacheline_count,
1618 &dim_cl_stores_l1hit,
1619 &dim_cl_stores_l1miss,
1628 &dim_percent_rmt_hitm,
1629 &dim_percent_lcl_hitm,
1630 &dim_percent_stores_l1hit,
1631 &dim_percent_stores_l1miss,
1644 &dim_dcacheline_idx,
1645 &dim_dcacheline_num,
1646 &dim_dcacheline_num_empty,
1650 static void fmt_free(struct perf_hpp_fmt *fmt)
1652 struct c2c_fmt *c2c_fmt;
1654 c2c_fmt = container_of(fmt, struct c2c_fmt, fmt);
1658 static bool fmt_equal(struct perf_hpp_fmt *a, struct perf_hpp_fmt *b)
1660 struct c2c_fmt *c2c_a = container_of(a, struct c2c_fmt, fmt);
1661 struct c2c_fmt *c2c_b = container_of(b, struct c2c_fmt, fmt);
1663 return c2c_a->dim == c2c_b->dim;
1666 static struct c2c_dimension *get_dimension(const char *name)
1670 for (i = 0; dimensions[i]; i++) {
1671 struct c2c_dimension *dim = dimensions[i];
1673 if (!strcmp(dim->name, name))
1680 static int c2c_se_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
1681 struct hist_entry *he)
1683 struct c2c_fmt *c2c_fmt = container_of(fmt, struct c2c_fmt, fmt);
1684 struct c2c_dimension *dim = c2c_fmt->dim;
1685 size_t len = fmt->user_len;
1688 len = hists__col_len(he->hists, dim->se->se_width_idx);
1690 if (dim == &dim_symbol || dim == &dim_srcline)
1691 len = symbol_width(he->hists, dim->se);
1694 return dim->se->se_snprintf(he, hpp->buf, hpp->size, len);
1697 static int64_t c2c_se_cmp(struct perf_hpp_fmt *fmt,
1698 struct hist_entry *a, struct hist_entry *b)
1700 struct c2c_fmt *c2c_fmt = container_of(fmt, struct c2c_fmt, fmt);
1701 struct c2c_dimension *dim = c2c_fmt->dim;
1703 return dim->se->se_cmp(a, b);
1706 static int64_t c2c_se_collapse(struct perf_hpp_fmt *fmt,
1707 struct hist_entry *a, struct hist_entry *b)
1709 struct c2c_fmt *c2c_fmt = container_of(fmt, struct c2c_fmt, fmt);
1710 struct c2c_dimension *dim = c2c_fmt->dim;
1711 int64_t (*collapse_fn)(struct hist_entry *, struct hist_entry *);
1713 collapse_fn = dim->se->se_collapse ?: dim->se->se_cmp;
1714 return collapse_fn(a, b);
1717 static struct c2c_fmt *get_format(const char *name)
1719 struct c2c_dimension *dim = get_dimension(name);
1720 struct c2c_fmt *c2c_fmt;
1721 struct perf_hpp_fmt *fmt;
1726 c2c_fmt = zalloc(sizeof(*c2c_fmt));
1732 fmt = &c2c_fmt->fmt;
1733 INIT_LIST_HEAD(&fmt->list);
1734 INIT_LIST_HEAD(&fmt->sort_list);
1736 fmt->cmp = dim->se ? c2c_se_cmp : dim->cmp;
1737 fmt->sort = dim->se ? c2c_se_cmp : dim->cmp;
1738 fmt->color = dim->se ? NULL : dim->color;
1739 fmt->entry = dim->se ? c2c_se_entry : dim->entry;
1740 fmt->header = c2c_header;
1741 fmt->width = c2c_width;
1742 fmt->collapse = dim->se ? c2c_se_collapse : dim->cmp;
1743 fmt->equal = fmt_equal;
1744 fmt->free = fmt_free;
1749 static int c2c_hists__init_output(struct perf_hpp_list *hpp_list, char *name)
1751 struct c2c_fmt *c2c_fmt = get_format(name);
1755 return output_field_add(hpp_list, name);
1758 perf_hpp_list__column_register(hpp_list, &c2c_fmt->fmt);
1762 static int c2c_hists__init_sort(struct perf_hpp_list *hpp_list, char *name)
1764 struct c2c_fmt *c2c_fmt = get_format(name);
1765 struct c2c_dimension *dim;
1769 return sort_dimension__add(hpp_list, name, NULL, 0);
1773 if (dim == &dim_dso)
1776 perf_hpp_list__register_sort_field(hpp_list, &c2c_fmt->fmt);
1780 #define PARSE_LIST(_list, _fn) \
1788 for (tok = strtok_r((char *)_list, ", ", &tmp); \
1789 tok; tok = strtok_r(NULL, ", ", &tmp)) { \
1790 ret = _fn(hpp_list, tok); \
1791 if (ret == -EINVAL) { \
1792 pr_err("Invalid --fields key: `%s'", tok); \
1794 } else if (ret == -ESRCH) { \
1795 pr_err("Unknown --fields key: `%s'", tok); \
1801 static int hpp_list__parse(struct perf_hpp_list *hpp_list,
1802 const char *output_,
1805 char *output = output_ ? strdup(output_) : NULL;
1806 char *sort = sort_ ? strdup(sort_) : NULL;
1809 PARSE_LIST(output, c2c_hists__init_output);
1810 PARSE_LIST(sort, c2c_hists__init_sort);
1812 /* copy sort keys to output fields */
1813 perf_hpp__setup_output_field(hpp_list);
1816 * We dont need other sorting keys other than those
1817 * we already specified. It also really slows down
1818 * the processing a lot with big number of output
1819 * fields, so switching this off for c2c.
1823 /* and then copy output fields to sort keys */
1824 perf_hpp__append_sort_keys(&hists->list);
1832 static int c2c_hists__init(struct c2c_hists *hists,
1834 int nr_header_lines)
1836 __hists__init(&hists->hists, &hists->list);
1839 * Initialize only with sort fields, we need to resort
1840 * later anyway, and that's where we add output fields
1843 perf_hpp_list__init(&hists->list);
1845 /* Overload number of header lines.*/
1846 hists->list.nr_header_lines = nr_header_lines;
1848 return hpp_list__parse(&hists->list, NULL, sort);
1851 static int c2c_hists__reinit(struct c2c_hists *c2c_hists,
1855 perf_hpp__reset_output_field(&c2c_hists->list);
1856 return hpp_list__parse(&c2c_hists->list, output, sort);
1859 #define DISPLAY_LINE_LIMIT 0.001
1861 static u8 filter_display(u32 val, u32 sum)
1863 if (sum == 0 || ((double)val / sum) < DISPLAY_LINE_LIMIT)
1864 return HIST_FILTER__C2C;
1869 static bool he__display(struct hist_entry *he, struct c2c_stats *stats)
1871 struct c2c_hist_entry *c2c_he;
1876 c2c_he = container_of(he, struct c2c_hist_entry, he);
1878 switch (c2c.display) {
1880 he->filtered = filter_display(c2c_he->stats.lcl_hitm,
1884 he->filtered = filter_display(c2c_he->stats.rmt_hitm,
1888 he->filtered = filter_display(c2c_he->stats.tot_hitm,
1895 return he->filtered == 0;
1898 static inline bool is_valid_hist_entry(struct hist_entry *he)
1900 struct c2c_hist_entry *c2c_he;
1901 bool has_record = false;
1903 c2c_he = container_of(he, struct c2c_hist_entry, he);
1905 /* It's a valid entry if contains stores */
1906 if (c2c_he->stats.store)
1909 switch (c2c.display) {
1911 has_record = !!c2c_he->stats.lcl_hitm;
1914 has_record = !!c2c_he->stats.rmt_hitm;
1917 has_record = !!c2c_he->stats.tot_hitm;
1926 static void set_node_width(struct c2c_hist_entry *c2c_he, int len)
1928 struct c2c_dimension *dim;
1930 dim = &c2c.hists == c2c_he->hists ?
1931 &dim_dcacheline_node : &dim_offset_node;
1933 if (len > dim->width)
1937 static int set_nodestr(struct c2c_hist_entry *c2c_he)
1942 if (c2c_he->nodestr)
1945 if (bitmap_weight(c2c_he->nodeset, c2c.nodes_cnt)) {
1946 len = bitmap_scnprintf(c2c_he->nodeset, c2c.nodes_cnt,
1949 len = scnprintf(buf, sizeof(buf), "N/A");
1952 set_node_width(c2c_he, len);
1953 c2c_he->nodestr = strdup(buf);
1954 return c2c_he->nodestr ? 0 : -ENOMEM;
1957 static void calc_width(struct c2c_hist_entry *c2c_he)
1959 struct c2c_hists *c2c_hists;
1961 c2c_hists = container_of(c2c_he->he.hists, struct c2c_hists, hists);
1962 hists__calc_col_len(&c2c_hists->hists, &c2c_he->he);
1963 set_nodestr(c2c_he);
1966 static int filter_cb(struct hist_entry *he, void *arg __maybe_unused)
1968 struct c2c_hist_entry *c2c_he;
1970 c2c_he = container_of(he, struct c2c_hist_entry, he);
1972 if (c2c.show_src && !he->srcline)
1973 he->srcline = hist_entry__srcline(he);
1977 if (!is_valid_hist_entry(he))
1978 he->filtered = HIST_FILTER__C2C;
1983 static int resort_cl_cb(struct hist_entry *he, void *arg __maybe_unused)
1985 struct c2c_hist_entry *c2c_he;
1986 struct c2c_hists *c2c_hists;
1987 bool display = he__display(he, &c2c.shared_clines_stats);
1989 c2c_he = container_of(he, struct c2c_hist_entry, he);
1990 c2c_hists = c2c_he->hists;
1992 if (display && c2c_hists) {
1993 static unsigned int idx;
1995 c2c_he->cacheline_idx = idx++;
1998 c2c_hists__reinit(c2c_hists, c2c.cl_output, c2c.cl_resort);
2000 hists__collapse_resort(&c2c_hists->hists, NULL);
2001 hists__output_resort_cb(&c2c_hists->hists, NULL, filter_cb);
2007 static void setup_nodes_header(void)
2009 dim_node.header = header_node[c2c.node_info];
2012 static int setup_nodes(struct perf_session *session)
2014 struct numa_node *n;
2015 unsigned long **nodes;
2019 if (c2c.node_info > 2)
2022 c2c.nodes_cnt = session->header.env.nr_numa_nodes;
2023 c2c.cpus_cnt = session->header.env.nr_cpus_avail;
2025 n = session->header.env.numa_nodes;
2029 nodes = zalloc(sizeof(unsigned long *) * c2c.nodes_cnt);
2035 cpu2node = zalloc(sizeof(int) * c2c.cpus_cnt);
2039 for (cpu = 0; cpu < c2c.cpus_cnt; cpu++)
2042 c2c.cpu2node = cpu2node;
2044 for (node = 0; node < c2c.nodes_cnt; node++) {
2045 struct perf_cpu_map *map = n[node].map;
2048 set = bitmap_alloc(c2c.cpus_cnt);
2054 /* empty node, skip */
2055 if (perf_cpu_map__empty(map))
2058 for (cpu = 0; cpu < map->nr; cpu++) {
2059 set_bit(map->map[cpu], set);
2061 if (WARN_ONCE(cpu2node[map->map[cpu]] != -1, "node/cpu topology bug"))
2064 cpu2node[map->map[cpu]] = node;
2068 setup_nodes_header();
2072 #define HAS_HITMS(__h) ((__h)->stats.lcl_hitm || (__h)->stats.rmt_hitm)
2074 static int resort_shared_cl_cb(struct hist_entry *he, void *arg __maybe_unused)
2076 struct c2c_hist_entry *c2c_he;
2077 c2c_he = container_of(he, struct c2c_hist_entry, he);
2079 if (HAS_HITMS(c2c_he)) {
2080 c2c.shared_clines++;
2081 c2c_add_stats(&c2c.shared_clines_stats, &c2c_he->stats);
2087 static int hists__iterate_cb(struct hists *hists, hists__resort_cb_t cb)
2089 struct rb_node *next = rb_first_cached(&hists->entries);
2093 struct hist_entry *he;
2095 he = rb_entry(next, struct hist_entry, rb_node);
2099 next = rb_next(&he->rb_node);
2105 static void print_c2c__display_stats(FILE *out)
2108 struct c2c_stats *stats = &c2c.hists.stats;
2110 llc_misses = stats->lcl_dram +
2115 fprintf(out, "=================================================\n");
2116 fprintf(out, " Trace Event Information \n");
2117 fprintf(out, "=================================================\n");
2118 fprintf(out, " Total records : %10d\n", stats->nr_entries);
2119 fprintf(out, " Locked Load/Store Operations : %10d\n", stats->locks);
2120 fprintf(out, " Load Operations : %10d\n", stats->load);
2121 fprintf(out, " Loads - uncacheable : %10d\n", stats->ld_uncache);
2122 fprintf(out, " Loads - IO : %10d\n", stats->ld_io);
2123 fprintf(out, " Loads - Miss : %10d\n", stats->ld_miss);
2124 fprintf(out, " Loads - no mapping : %10d\n", stats->ld_noadrs);
2125 fprintf(out, " Load Fill Buffer Hit : %10d\n", stats->ld_fbhit);
2126 fprintf(out, " Load L1D hit : %10d\n", stats->ld_l1hit);
2127 fprintf(out, " Load L2D hit : %10d\n", stats->ld_l2hit);
2128 fprintf(out, " Load LLC hit : %10d\n", stats->ld_llchit + stats->lcl_hitm);
2129 fprintf(out, " Load Local HITM : %10d\n", stats->lcl_hitm);
2130 fprintf(out, " Load Remote HITM : %10d\n", stats->rmt_hitm);
2131 fprintf(out, " Load Remote HIT : %10d\n", stats->rmt_hit);
2132 fprintf(out, " Load Local DRAM : %10d\n", stats->lcl_dram);
2133 fprintf(out, " Load Remote DRAM : %10d\n", stats->rmt_dram);
2134 fprintf(out, " Load MESI State Exclusive : %10d\n", stats->ld_excl);
2135 fprintf(out, " Load MESI State Shared : %10d\n", stats->ld_shared);
2136 fprintf(out, " Load LLC Misses : %10d\n", llc_misses);
2137 fprintf(out, " Load access blocked by data : %10d\n", stats->blk_data);
2138 fprintf(out, " Load access blocked by address : %10d\n", stats->blk_addr);
2139 fprintf(out, " LLC Misses to Local DRAM : %10.1f%%\n", ((double)stats->lcl_dram/(double)llc_misses) * 100.);
2140 fprintf(out, " LLC Misses to Remote DRAM : %10.1f%%\n", ((double)stats->rmt_dram/(double)llc_misses) * 100.);
2141 fprintf(out, " LLC Misses to Remote cache (HIT) : %10.1f%%\n", ((double)stats->rmt_hit /(double)llc_misses) * 100.);
2142 fprintf(out, " LLC Misses to Remote cache (HITM) : %10.1f%%\n", ((double)stats->rmt_hitm/(double)llc_misses) * 100.);
2143 fprintf(out, " Store Operations : %10d\n", stats->store);
2144 fprintf(out, " Store - uncacheable : %10d\n", stats->st_uncache);
2145 fprintf(out, " Store - no mapping : %10d\n", stats->st_noadrs);
2146 fprintf(out, " Store L1D Hit : %10d\n", stats->st_l1hit);
2147 fprintf(out, " Store L1D Miss : %10d\n", stats->st_l1miss);
2148 fprintf(out, " No Page Map Rejects : %10d\n", stats->nomap);
2149 fprintf(out, " Unable to parse data source : %10d\n", stats->noparse);
2152 static void print_shared_cacheline_info(FILE *out)
2154 struct c2c_stats *stats = &c2c.shared_clines_stats;
2155 int hitm_cnt = stats->lcl_hitm + stats->rmt_hitm;
2157 fprintf(out, "=================================================\n");
2158 fprintf(out, " Global Shared Cache Line Event Information \n");
2159 fprintf(out, "=================================================\n");
2160 fprintf(out, " Total Shared Cache Lines : %10d\n", c2c.shared_clines);
2161 fprintf(out, " Load HITs on shared lines : %10d\n", stats->load);
2162 fprintf(out, " Fill Buffer Hits on shared lines : %10d\n", stats->ld_fbhit);
2163 fprintf(out, " L1D hits on shared lines : %10d\n", stats->ld_l1hit);
2164 fprintf(out, " L2D hits on shared lines : %10d\n", stats->ld_l2hit);
2165 fprintf(out, " LLC hits on shared lines : %10d\n", stats->ld_llchit + stats->lcl_hitm);
2166 fprintf(out, " Locked Access on shared lines : %10d\n", stats->locks);
2167 fprintf(out, " Blocked Access on shared lines : %10d\n", stats->blk_data + stats->blk_addr);
2168 fprintf(out, " Store HITs on shared lines : %10d\n", stats->store);
2169 fprintf(out, " Store L1D hits on shared lines : %10d\n", stats->st_l1hit);
2170 fprintf(out, " Total Merged records : %10d\n", hitm_cnt + stats->store);
2173 static void print_cacheline(struct c2c_hists *c2c_hists,
2174 struct hist_entry *he_cl,
2175 struct perf_hpp_list *hpp_list,
2179 struct perf_hpp hpp = {
2186 hists__fprintf_headers(&c2c_hists->hists, out);
2192 fprintf(out, " -------------------------------------------------------------\n");
2193 __hist_entry__snprintf(he_cl, &hpp, hpp_list);
2194 fprintf(out, "%s\n", bf);
2195 fprintf(out, " -------------------------------------------------------------\n");
2197 hists__fprintf(&c2c_hists->hists, false, 0, 0, 0, out, false);
2200 static void print_pareto(FILE *out)
2202 struct perf_hpp_list hpp_list;
2205 const char *cl_output;
2207 cl_output = "cl_num,"
2214 perf_hpp_list__init(&hpp_list);
2215 ret = hpp_list__parse(&hpp_list, cl_output, NULL);
2217 if (WARN_ONCE(ret, "failed to setup sort entries\n"))
2220 nd = rb_first_cached(&c2c.hists.hists.entries);
2222 for (; nd; nd = rb_next(nd)) {
2223 struct hist_entry *he = rb_entry(nd, struct hist_entry, rb_node);
2224 struct c2c_hist_entry *c2c_he;
2229 c2c_he = container_of(he, struct c2c_hist_entry, he);
2230 print_cacheline(c2c_he->hists, he, &hpp_list, out);
2234 static void print_c2c_info(FILE *out, struct perf_session *session)
2236 struct evlist *evlist = session->evlist;
2237 struct evsel *evsel;
2240 fprintf(out, "=================================================\n");
2241 fprintf(out, " c2c details \n");
2242 fprintf(out, "=================================================\n");
2244 evlist__for_each_entry(evlist, evsel) {
2245 fprintf(out, "%-36s: %s\n", first ? " Events" : "", evsel__name(evsel));
2248 fprintf(out, " Cachelines sort on : %s HITMs\n",
2249 display_str[c2c.display]);
2250 fprintf(out, " Cacheline data grouping : %s\n", c2c.cl_sort);
2253 static void perf_c2c__hists_fprintf(FILE *out, struct perf_session *session)
2257 print_c2c__display_stats(out);
2259 print_shared_cacheline_info(out);
2261 print_c2c_info(out, session);
2267 fprintf(out, "=================================================\n");
2268 fprintf(out, " Shared Data Cache Line Table \n");
2269 fprintf(out, "=================================================\n");
2270 fprintf(out, "#\n");
2272 hists__fprintf(&c2c.hists.hists, true, 0, 0, 0, stdout, true);
2275 fprintf(out, "=================================================\n");
2276 fprintf(out, " Shared Cache Line Distribution Pareto \n");
2277 fprintf(out, "=================================================\n");
2278 fprintf(out, "#\n");
2283 #ifdef HAVE_SLANG_SUPPORT
2284 static void c2c_browser__update_nr_entries(struct hist_browser *hb)
2287 struct rb_node *nd = rb_first_cached(&hb->hists->entries);
2290 struct hist_entry *he = rb_entry(nd, struct hist_entry, rb_node);
2298 hb->nr_non_filtered_entries = nr_entries;
2301 struct c2c_cacheline_browser {
2302 struct hist_browser hb;
2303 struct hist_entry *he;
2307 perf_c2c_cacheline_browser__title(struct hist_browser *browser,
2308 char *bf, size_t size)
2310 struct c2c_cacheline_browser *cl_browser;
2311 struct hist_entry *he;
2314 cl_browser = container_of(browser, struct c2c_cacheline_browser, hb);
2315 he = cl_browser->he;
2318 addr = cl_address(he->mem_info->daddr.addr);
2320 scnprintf(bf, size, "Cacheline 0x%lx", addr);
2324 static struct c2c_cacheline_browser*
2325 c2c_cacheline_browser__new(struct hists *hists, struct hist_entry *he)
2327 struct c2c_cacheline_browser *browser;
2329 browser = zalloc(sizeof(*browser));
2331 hist_browser__init(&browser->hb, hists);
2332 browser->hb.c2c_filter = true;
2333 browser->hb.title = perf_c2c_cacheline_browser__title;
2340 static int perf_c2c__browse_cacheline(struct hist_entry *he)
2342 struct c2c_hist_entry *c2c_he;
2343 struct c2c_hists *c2c_hists;
2344 struct c2c_cacheline_browser *cl_browser;
2345 struct hist_browser *browser;
2347 static const char help[] =
2348 " ENTER Toggle callchains (if present) \n"
2349 " n Toggle Node details info \n"
2350 " s Toggle full length of symbol and source line columns \n"
2351 " q Return back to cacheline list \n";
2356 /* Display compact version first. */
2357 c2c.symbol_full = false;
2359 c2c_he = container_of(he, struct c2c_hist_entry, he);
2360 c2c_hists = c2c_he->hists;
2362 cl_browser = c2c_cacheline_browser__new(&c2c_hists->hists, he);
2363 if (cl_browser == NULL)
2366 browser = &cl_browser->hb;
2368 /* reset abort key so that it can get Ctrl-C as a key */
2370 SLang_init_tty(0, 0, 0);
2372 c2c_browser__update_nr_entries(browser);
2375 key = hist_browser__run(browser, "? - help", true, 0);
2379 c2c.symbol_full = !c2c.symbol_full;
2382 c2c.node_info = (c2c.node_info + 1) % 3;
2383 setup_nodes_header();
2388 ui_browser__help_window(&browser->b, help);
2400 static int perf_c2c_browser__title(struct hist_browser *browser,
2401 char *bf, size_t size)
2404 "Shared Data Cache Line Table "
2405 "(%lu entries, sorted on %s HITMs)",
2406 browser->nr_non_filtered_entries,
2407 display_str[c2c.display]);
2411 static struct hist_browser*
2412 perf_c2c_browser__new(struct hists *hists)
2414 struct hist_browser *browser = hist_browser__new(hists);
2417 browser->title = perf_c2c_browser__title;
2418 browser->c2c_filter = true;
2424 static int perf_c2c__hists_browse(struct hists *hists)
2426 struct hist_browser *browser;
2428 static const char help[] =
2429 " d Display cacheline details \n"
2430 " ENTER Toggle callchains (if present) \n"
2433 browser = perf_c2c_browser__new(hists);
2434 if (browser == NULL)
2437 /* reset abort key so that it can get Ctrl-C as a key */
2439 SLang_init_tty(0, 0, 0);
2441 c2c_browser__update_nr_entries(browser);
2444 key = hist_browser__run(browser, "? - help", true, 0);
2450 perf_c2c__browse_cacheline(browser->he_selection);
2453 ui_browser__help_window(&browser->b, help);
2461 hist_browser__delete(browser);
2465 static void perf_c2c_display(struct perf_session *session)
2467 if (use_browser == 0)
2468 perf_c2c__hists_fprintf(stdout, session);
2470 perf_c2c__hists_browse(&c2c.hists.hists);
2473 static void perf_c2c_display(struct perf_session *session)
2476 perf_c2c__hists_fprintf(stdout, session);
2478 #endif /* HAVE_SLANG_SUPPORT */
2480 static char *fill_line(const char *orig, int len)
2482 int i, j, olen = strlen(orig);
2485 buf = zalloc(len + 1);
2489 j = len / 2 - olen / 2;
2491 for (i = 0; i < j - 1; i++)
2496 strcpy(buf + i, orig);
2502 for (; i < len; i++)
2508 static int ui_quirks(void)
2510 const char *nodestr = "Data address";
2513 if (!c2c.use_stdio) {
2514 dim_offset.width = 5;
2515 dim_offset.header = header_offset_tui;
2519 dim_percent_hitm.header = percent_hitm_header[c2c.display];
2521 /* Fix the zero line for dcacheline column. */
2522 buf = fill_line("Cacheline", dim_dcacheline.width +
2523 dim_dcacheline_node.width +
2524 dim_dcacheline_count.width + 4);
2528 dim_dcacheline.header.line[0].text = buf;
2530 /* Fix the zero line for offset column. */
2531 buf = fill_line(nodestr, dim_offset.width +
2532 dim_offset_node.width +
2533 dim_dcacheline_count.width + 4);
2537 dim_offset.header.line[0].text = buf;
2542 #define CALLCHAIN_DEFAULT_OPT "graph,0.5,caller,function,percent"
2544 const char callchain_help[] = "Display call graph (stack chain/backtrace):\n\n"
2545 CALLCHAIN_REPORT_HELP
2546 "\n\t\t\t\tDefault: " CALLCHAIN_DEFAULT_OPT;
2549 parse_callchain_opt(const struct option *opt, const char *arg, int unset)
2551 struct callchain_param *callchain = opt->value;
2553 callchain->enabled = !unset;
2558 symbol_conf.use_callchain = false;
2559 callchain->mode = CHAIN_NONE;
2563 return parse_callchain_report_opt(arg);
2566 static int setup_callchain(struct evlist *evlist)
2568 u64 sample_type = evlist__combined_sample_type(evlist);
2569 enum perf_call_graph_mode mode = CALLCHAIN_NONE;
2571 if ((sample_type & PERF_SAMPLE_REGS_USER) &&
2572 (sample_type & PERF_SAMPLE_STACK_USER)) {
2573 mode = CALLCHAIN_DWARF;
2574 dwarf_callchain_users = true;
2575 } else if (sample_type & PERF_SAMPLE_BRANCH_STACK)
2576 mode = CALLCHAIN_LBR;
2577 else if (sample_type & PERF_SAMPLE_CALLCHAIN)
2578 mode = CALLCHAIN_FP;
2580 if (!callchain_param.enabled &&
2581 callchain_param.mode != CHAIN_NONE &&
2582 mode != CALLCHAIN_NONE) {
2583 symbol_conf.use_callchain = true;
2584 if (callchain_register_param(&callchain_param) < 0) {
2585 ui__error("Can't register callchain params.\n");
2590 if (c2c.stitch_lbr && (mode != CALLCHAIN_LBR)) {
2591 ui__warning("Can't find LBR callchain. Switch off --stitch-lbr.\n"
2592 "Please apply --call-graph lbr when recording.\n");
2593 c2c.stitch_lbr = false;
2596 callchain_param.record_mode = mode;
2597 callchain_param.min_percent = 0;
2601 static int setup_display(const char *str)
2603 const char *display = str ?: "tot";
2605 if (!strcmp(display, "tot"))
2606 c2c.display = DISPLAY_TOT;
2607 else if (!strcmp(display, "rmt"))
2608 c2c.display = DISPLAY_RMT;
2609 else if (!strcmp(display, "lcl"))
2610 c2c.display = DISPLAY_LCL;
2612 pr_err("failed: unknown display type: %s\n", str);
2619 #define for_each_token(__tok, __buf, __sep, __tmp) \
2620 for (__tok = strtok_r(__buf, __sep, &__tmp); __tok; \
2621 __tok = strtok_r(NULL, __sep, &__tmp))
2623 static int build_cl_output(char *cl_sort, bool no_source)
2625 char *tok, *tmp, *buf = strdup(cl_sort);
2626 bool add_pid = false;
2627 bool add_tid = false;
2628 bool add_iaddr = false;
2629 bool add_sym = false;
2630 bool add_dso = false;
2631 bool add_src = false;
2637 for_each_token(tok, buf, ",", tmp) {
2638 if (!strcmp(tok, "tid")) {
2640 } else if (!strcmp(tok, "pid")) {
2642 } else if (!strcmp(tok, "iaddr")) {
2646 add_src = no_source ? false : true;
2647 } else if (!strcmp(tok, "dso")) {
2649 } else if (strcmp(tok, "offset")) {
2650 pr_err("unrecognized sort token: %s\n", tok);
2656 if (asprintf(&c2c.cl_output,
2657 "%s%s%s%s%s%s%s%s%s%s",
2658 c2c.use_stdio ? "cl_num_empty," : "",
2661 "percent_stores_l1hit,"
2662 "percent_stores_l1miss,"
2663 "offset,offset_node,dcacheline_count,",
2664 add_pid ? "pid," : "",
2665 add_tid ? "tid," : "",
2666 add_iaddr ? "iaddr," : "",
2672 add_sym ? "symbol," : "",
2673 add_dso ? "dso," : "",
2674 add_src ? "cl_srcline," : "",
2680 c2c.show_src = add_src;
2686 static int setup_coalesce(const char *coalesce, bool no_source)
2688 const char *c = coalesce ?: coalesce_default;
2690 if (asprintf(&c2c.cl_sort, "offset,%s", c) < 0)
2693 if (build_cl_output(c2c.cl_sort, no_source))
2696 if (asprintf(&c2c.cl_resort, "offset,%s",
2697 c2c.display == DISPLAY_TOT ?
2699 c2c.display == DISPLAY_RMT ?
2700 "rmt_hitm,lcl_hitm" :
2701 "lcl_hitm,rmt_hitm") < 0)
2704 pr_debug("coalesce sort fields: %s\n", c2c.cl_sort);
2705 pr_debug("coalesce resort fields: %s\n", c2c.cl_resort);
2706 pr_debug("coalesce output fields: %s\n", c2c.cl_output);
2710 static int perf_c2c__report(int argc, const char **argv)
2712 struct itrace_synth_opts itrace_synth_opts = {
2714 .mem = true, /* Only enable memory event */
2715 .default_no_sample = true,
2718 struct perf_session *session;
2719 struct ui_progress prog;
2720 struct perf_data data = {
2721 .mode = PERF_DATA_MODE_READ,
2723 char callchain_default_opt[] = CALLCHAIN_DEFAULT_OPT;
2724 const char *display = NULL;
2725 const char *coalesce = NULL;
2726 bool no_source = false;
2727 const struct option options[] = {
2728 OPT_STRING('k', "vmlinux", &symbol_conf.vmlinux_name,
2729 "file", "vmlinux pathname"),
2730 OPT_STRING('i', "input", &input_name, "file",
2731 "the input file to process"),
2732 OPT_INCR('N', "node-info", &c2c.node_info,
2733 "show extra node info in report (repeat for more info)"),
2734 #ifdef HAVE_SLANG_SUPPORT
2735 OPT_BOOLEAN(0, "stdio", &c2c.use_stdio, "Use the stdio interface"),
2737 OPT_BOOLEAN(0, "stats", &c2c.stats_only,
2738 "Display only statistic tables (implies --stdio)"),
2739 OPT_BOOLEAN(0, "full-symbols", &c2c.symbol_full,
2740 "Display full length of symbols"),
2741 OPT_BOOLEAN(0, "no-source", &no_source,
2742 "Do not display Source Line column"),
2743 OPT_BOOLEAN(0, "show-all", &c2c.show_all,
2744 "Show all captured HITM lines."),
2745 OPT_CALLBACK_DEFAULT('g', "call-graph", &callchain_param,
2746 "print_type,threshold[,print_limit],order,sort_key[,branch],value",
2747 callchain_help, &parse_callchain_opt,
2748 callchain_default_opt),
2749 OPT_STRING('d', "display", &display, "Switch HITM output type", "lcl,rmt"),
2750 OPT_STRING('c', "coalesce", &coalesce, "coalesce fields",
2751 "coalesce fields: pid,tid,iaddr,dso"),
2752 OPT_BOOLEAN('f', "force", &symbol_conf.force, "don't complain, do it"),
2753 OPT_BOOLEAN(0, "stitch-lbr", &c2c.stitch_lbr,
2754 "Enable LBR callgraph stitching approach"),
2755 OPT_PARENT(c2c_options),
2759 const char *output_str, *sort_str = NULL;
2761 argc = parse_options(argc, argv, options, report_c2c_usage,
2762 PARSE_OPT_STOP_AT_NON_OPTION);
2764 usage_with_options(report_c2c_usage, options);
2767 c2c.use_stdio = true;
2769 if (!input_name || !strlen(input_name))
2770 input_name = "perf.data";
2772 data.path = input_name;
2773 data.force = symbol_conf.force;
2775 err = setup_display(display);
2779 err = setup_coalesce(coalesce, no_source);
2781 pr_debug("Failed to initialize hists\n");
2785 err = c2c_hists__init(&c2c.hists, "dcacheline", 2);
2787 pr_debug("Failed to initialize hists\n");
2791 session = perf_session__new(&data, 0, &c2c.tool);
2792 if (IS_ERR(session)) {
2793 err = PTR_ERR(session);
2794 pr_debug("Error creating perf session\n");
2798 session->itrace_synth_opts = &itrace_synth_opts;
2800 err = setup_nodes(session);
2802 pr_err("Failed setup nodes\n");
2806 err = mem2node__init(&c2c.mem2node, &session->header.env);
2810 err = setup_callchain(session->evlist);
2814 if (symbol__init(&session->header.env) < 0)
2817 /* No pipe support at the moment. */
2818 if (perf_data__is_pipe(session->data)) {
2819 pr_debug("No pipe support at the moment.\n");
2828 setup_browser(false);
2830 err = perf_session__process_events(session);
2832 pr_err("failed to process sample\n");
2836 output_str = "cl_idx,"
2841 "tot_hitm,lcl_hitm,rmt_hitm,"
2845 "stores_l1hit,stores_l1miss,"
2846 "ld_fbhit,ld_l1hit,ld_l2hit,"
2847 "ld_lclhit,lcl_hitm,"
2848 "ld_rmthit,rmt_hitm,"
2849 "dram_lcl,dram_rmt";
2851 if (c2c.display == DISPLAY_TOT)
2852 sort_str = "tot_hitm";
2853 else if (c2c.display == DISPLAY_RMT)
2854 sort_str = "rmt_hitm";
2855 else if (c2c.display == DISPLAY_LCL)
2856 sort_str = "lcl_hitm";
2858 c2c_hists__reinit(&c2c.hists, output_str, sort_str);
2860 ui_progress__init(&prog, c2c.hists.hists.nr_entries, "Sorting...");
2862 hists__collapse_resort(&c2c.hists.hists, NULL);
2863 hists__output_resort_cb(&c2c.hists.hists, &prog, resort_shared_cl_cb);
2864 hists__iterate_cb(&c2c.hists.hists, resort_cl_cb);
2866 ui_progress__finish();
2869 pr_err("failed to setup UI\n");
2873 perf_c2c_display(session);
2876 mem2node__exit(&c2c.mem2node);
2878 perf_session__delete(session);
2883 static int parse_record_events(const struct option *opt,
2884 const char *str, int unset __maybe_unused)
2886 bool *event_set = (bool *) opt->value;
2888 if (!strcmp(str, "list")) {
2889 perf_mem_events__list();
2892 if (perf_mem_events__parse(str))
2900 static const char * const __usage_record[] = {
2901 "perf c2c record [<options>] [<command>]",
2902 "perf c2c record [<options>] -- <command> [<options>]",
2906 static const char * const *record_mem_usage = __usage_record;
2908 static int perf_c2c__record(int argc, const char **argv)
2910 int rec_argc, i = 0, j;
2911 const char **rec_argv;
2913 bool all_user = false, all_kernel = false;
2914 bool event_set = false;
2915 struct perf_mem_event *e;
2916 struct option options[] = {
2917 OPT_CALLBACK('e', "event", &event_set, "event",
2918 "event selector. Use 'perf c2c record -e list' to list available events",
2919 parse_record_events),
2920 OPT_BOOLEAN('u', "all-user", &all_user, "collect only user level data"),
2921 OPT_BOOLEAN('k', "all-kernel", &all_kernel, "collect only kernel level data"),
2922 OPT_UINTEGER('l', "ldlat", &perf_mem_events__loads_ldlat, "setup mem-loads latency"),
2923 OPT_PARENT(c2c_options),
2927 if (perf_mem_events__init()) {
2928 pr_err("failed: memory events not supported\n");
2932 argc = parse_options(argc, argv, options, record_mem_usage,
2933 PARSE_OPT_KEEP_UNKNOWN);
2935 rec_argc = argc + 11; /* max number of arguments */
2936 rec_argv = calloc(rec_argc + 1, sizeof(char *));
2940 rec_argv[i++] = "record";
2943 e = perf_mem_events__ptr(PERF_MEM_EVENTS__LOAD_STORE);
2945 * The load and store operations are required, use the event
2946 * PERF_MEM_EVENTS__LOAD_STORE if it is supported.
2951 e = perf_mem_events__ptr(PERF_MEM_EVENTS__LOAD);
2954 e = perf_mem_events__ptr(PERF_MEM_EVENTS__STORE);
2959 e = perf_mem_events__ptr(PERF_MEM_EVENTS__LOAD);
2961 rec_argv[i++] = "-W";
2963 rec_argv[i++] = "-d";
2964 rec_argv[i++] = "--phys-data";
2965 rec_argv[i++] = "--sample-cpu";
2967 for (j = 0; j < PERF_MEM_EVENTS__MAX; j++) {
2968 e = perf_mem_events__ptr(j);
2972 if (!e->supported) {
2973 pr_err("failed: event '%s' not supported\n",
2974 perf_mem_events__name(j));
2979 rec_argv[i++] = "-e";
2980 rec_argv[i++] = perf_mem_events__name(j);
2984 rec_argv[i++] = "--all-user";
2987 rec_argv[i++] = "--all-kernel";
2989 for (j = 0; j < argc; j++, i++)
2990 rec_argv[i] = argv[j];
2993 pr_debug("calling: ");
2997 while (rec_argv[j]) {
2998 pr_debug("%s ", rec_argv[j]);
3004 ret = cmd_record(i, rec_argv);
3009 int cmd_c2c(int argc, const char **argv)
3011 argc = parse_options(argc, argv, c2c_options, c2c_usage,
3012 PARSE_OPT_STOP_AT_NON_OPTION);
3015 usage_with_options(c2c_usage, c2c_options);
3017 if (!strncmp(argv[0], "rec", 3)) {
3018 return perf_c2c__record(argc, argv);
3019 } else if (!strncmp(argv[0], "rep", 3)) {
3020 return perf_c2c__report(argc, argv);
3022 usage_with_options(c2c_usage, c2c_options);