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;
99 /* HITM shared clines stats */
100 struct c2c_stats hitm_stats;
105 const char *coalesce;
118 static const char *display_str[DISPLAY_MAX] = {
119 [DISPLAY_LCL] = "Local",
120 [DISPLAY_RMT] = "Remote",
121 [DISPLAY_TOT] = "Total",
124 static const struct option c2c_options[] = {
125 OPT_INCR('v', "verbose", &verbose, "be more verbose (show counter open errors, etc)"),
129 static struct perf_c2c c2c;
131 static void *c2c_he_zalloc(size_t size)
133 struct c2c_hist_entry *c2c_he;
135 c2c_he = zalloc(size + sizeof(*c2c_he));
139 c2c_he->cpuset = bitmap_alloc(c2c.cpus_cnt);
143 c2c_he->nodeset = bitmap_alloc(c2c.nodes_cnt);
144 if (!c2c_he->nodeset)
147 c2c_he->node_stats = zalloc(c2c.nodes_cnt * sizeof(*c2c_he->node_stats));
148 if (!c2c_he->node_stats)
151 init_stats(&c2c_he->cstats.lcl_hitm);
152 init_stats(&c2c_he->cstats.rmt_hitm);
153 init_stats(&c2c_he->cstats.load);
158 static void c2c_he_free(void *he)
160 struct c2c_hist_entry *c2c_he;
162 c2c_he = container_of(he, struct c2c_hist_entry, he);
164 hists__delete_entries(&c2c_he->hists->hists);
168 free(c2c_he->cpuset);
169 free(c2c_he->nodeset);
170 free(c2c_he->nodestr);
171 free(c2c_he->node_stats);
175 static struct hist_entry_ops c2c_entry_ops = {
176 .new = c2c_he_zalloc,
180 static int c2c_hists__init(struct c2c_hists *hists,
182 int nr_header_lines);
184 static struct c2c_hists*
185 he__get_c2c_hists(struct hist_entry *he,
189 struct c2c_hist_entry *c2c_he;
190 struct c2c_hists *hists;
193 c2c_he = container_of(he, struct c2c_hist_entry, he);
195 return c2c_he->hists;
197 hists = c2c_he->hists = zalloc(sizeof(*hists));
201 ret = c2c_hists__init(hists, sort, nr_header_lines);
210 static void c2c_he__set_cpu(struct c2c_hist_entry *c2c_he,
211 struct perf_sample *sample)
213 if (WARN_ONCE(sample->cpu == (unsigned int) -1,
214 "WARNING: no sample cpu value"))
217 set_bit(sample->cpu, c2c_he->cpuset);
220 static void c2c_he__set_node(struct c2c_hist_entry *c2c_he,
221 struct perf_sample *sample)
225 if (!sample->phys_addr) {
226 c2c_he->paddr_zero = true;
230 node = mem2node__node(&c2c.mem2node, sample->phys_addr);
231 if (WARN_ONCE(node < 0, "WARNING: failed to find node\n"))
234 set_bit(node, c2c_he->nodeset);
236 if (c2c_he->paddr != sample->phys_addr) {
238 c2c_he->paddr = sample->phys_addr;
242 static void compute_stats(struct c2c_hist_entry *c2c_he,
243 struct c2c_stats *stats,
246 struct compute_stats *cstats = &c2c_he->cstats;
249 update_stats(&cstats->rmt_hitm, weight);
250 else if (stats->lcl_hitm)
251 update_stats(&cstats->lcl_hitm, weight);
252 else if (stats->load)
253 update_stats(&cstats->load, weight);
256 static int process_sample_event(struct perf_tool *tool __maybe_unused,
257 union perf_event *event,
258 struct perf_sample *sample,
260 struct machine *machine)
262 struct c2c_hists *c2c_hists = &c2c.hists;
263 struct c2c_hist_entry *c2c_he;
264 struct c2c_stats stats = { .nr_entries = 0, };
265 struct hist_entry *he;
266 struct addr_location al;
267 struct mem_info *mi, *mi_dup;
270 if (machine__resolve(machine, &al, sample) < 0) {
271 pr_debug("problem processing %d event, skipping it.\n",
276 ret = sample__resolve_callchain(sample, &callchain_cursor, NULL,
277 evsel, &al, sysctl_perf_event_max_stack);
281 mi = sample__resolve_mem(sample, &al);
286 * The mi object is released in hists__add_entry_ops,
287 * if it gets sorted out into existing data, so we need
288 * to take the copy now.
290 mi_dup = mem_info__get(mi);
292 c2c_decode_stats(&stats, mi);
294 he = hists__add_entry_ops(&c2c_hists->hists, &c2c_entry_ops,
300 c2c_he = container_of(he, struct c2c_hist_entry, he);
301 c2c_add_stats(&c2c_he->stats, &stats);
302 c2c_add_stats(&c2c_hists->stats, &stats);
304 c2c_he__set_cpu(c2c_he, sample);
305 c2c_he__set_node(c2c_he, sample);
307 hists__inc_nr_samples(&c2c_hists->hists, he->filtered);
308 ret = hist_entry__append_callchain(he, sample);
312 * There's already been warning about missing
313 * sample's cpu value. Let's account all to
314 * node 0 in this case, without any further
317 * Doing node stats only for single callchain data.
319 int cpu = sample->cpu == (unsigned int) -1 ? 0 : sample->cpu;
320 int node = c2c.cpu2node[cpu];
324 c2c_hists = he__get_c2c_hists(he, c2c.cl_sort, 2);
328 he = hists__add_entry_ops(&c2c_hists->hists, &c2c_entry_ops,
334 c2c_he = container_of(he, struct c2c_hist_entry, he);
335 c2c_add_stats(&c2c_he->stats, &stats);
336 c2c_add_stats(&c2c_hists->stats, &stats);
337 c2c_add_stats(&c2c_he->node_stats[node], &stats);
339 compute_stats(c2c_he, &stats, sample->weight);
341 c2c_he__set_cpu(c2c_he, sample);
342 c2c_he__set_node(c2c_he, sample);
344 hists__inc_nr_samples(&c2c_hists->hists, he->filtered);
345 ret = hist_entry__append_callchain(he, sample);
349 addr_location__put(&al);
353 mem_info__put(mi_dup);
359 static struct perf_c2c c2c = {
361 .sample = process_sample_event,
362 .mmap = perf_event__process_mmap,
363 .mmap2 = perf_event__process_mmap2,
364 .comm = perf_event__process_comm,
365 .exit = perf_event__process_exit,
366 .fork = perf_event__process_fork,
367 .lost = perf_event__process_lost,
368 .ordered_events = true,
369 .ordering_requires_timestamps = true,
373 static const char * const c2c_usage[] = {
374 "perf c2c {record|report}",
378 static const char * const __usage_report[] = {
383 static const char * const *report_c2c_usage = __usage_report;
385 #define C2C_HEADER_MAX 2
391 } line[C2C_HEADER_MAX];
394 struct c2c_dimension {
395 struct c2c_header header;
398 struct sort_entry *se;
400 int64_t (*cmp)(struct perf_hpp_fmt *fmt,
401 struct hist_entry *, struct hist_entry *);
402 int (*entry)(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
403 struct hist_entry *he);
404 int (*color)(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
405 struct hist_entry *he);
409 struct perf_hpp_fmt fmt;
410 struct c2c_dimension *dim;
413 #define SYMBOL_WIDTH 30
415 static struct c2c_dimension dim_symbol;
416 static struct c2c_dimension dim_srcline;
418 static int symbol_width(struct hists *hists, struct sort_entry *se)
420 int width = hists__col_len(hists, se->se_width_idx);
422 if (!c2c.symbol_full)
423 width = MIN(width, SYMBOL_WIDTH);
428 static int c2c_width(struct perf_hpp_fmt *fmt,
429 struct perf_hpp *hpp __maybe_unused,
432 struct c2c_fmt *c2c_fmt;
433 struct c2c_dimension *dim;
435 c2c_fmt = container_of(fmt, struct c2c_fmt, fmt);
438 if (dim == &dim_symbol || dim == &dim_srcline)
439 return symbol_width(hists, dim->se);
441 return dim->se ? hists__col_len(hists, dim->se->se_width_idx) :
445 static int c2c_header(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
446 struct hists *hists, int line, int *span)
448 struct perf_hpp_list *hpp_list = hists->hpp_list;
449 struct c2c_fmt *c2c_fmt;
450 struct c2c_dimension *dim;
451 const char *text = NULL;
452 int width = c2c_width(fmt, hpp, hists);
454 c2c_fmt = container_of(fmt, struct c2c_fmt, fmt);
458 text = dim->header.line[line].text;
459 /* Use the last line from sort_entry if not defined. */
460 if (!text && (line == hpp_list->nr_header_lines - 1))
461 text = dim->se->se_header;
463 text = dim->header.line[line].text;
469 *span = dim->header.line[line].span;
476 return scnprintf(hpp->buf, hpp->size, "%*s", width, text);
479 #define HEX_STR(__s, __v) \
481 scnprintf(__s, sizeof(__s), "0x%" PRIx64, __v); \
486 dcacheline_cmp(struct perf_hpp_fmt *fmt __maybe_unused,
487 struct hist_entry *left, struct hist_entry *right)
489 return sort__dcacheline_cmp(left, right);
492 static int dcacheline_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
493 struct hist_entry *he)
496 int width = c2c_width(fmt, hpp, he->hists);
500 addr = cl_address(he->mem_info->daddr.addr);
502 return scnprintf(hpp->buf, hpp->size, "%*s", width, HEX_STR(buf, addr));
506 dcacheline_node_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
507 struct hist_entry *he)
509 struct c2c_hist_entry *c2c_he;
510 int width = c2c_width(fmt, hpp, he->hists);
512 c2c_he = container_of(he, struct c2c_hist_entry, he);
513 if (WARN_ON_ONCE(!c2c_he->nodestr))
516 return scnprintf(hpp->buf, hpp->size, "%*s", width, c2c_he->nodestr);
520 dcacheline_node_count(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
521 struct hist_entry *he)
523 struct c2c_hist_entry *c2c_he;
524 int width = c2c_width(fmt, hpp, he->hists);
526 c2c_he = container_of(he, struct c2c_hist_entry, he);
527 return scnprintf(hpp->buf, hpp->size, "%*lu", width, c2c_he->paddr_cnt);
530 static int offset_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
531 struct hist_entry *he)
534 int width = c2c_width(fmt, hpp, he->hists);
538 addr = cl_offset(he->mem_info->daddr.al_addr);
540 return scnprintf(hpp->buf, hpp->size, "%*s", width, HEX_STR(buf, addr));
544 offset_cmp(struct perf_hpp_fmt *fmt __maybe_unused,
545 struct hist_entry *left, struct hist_entry *right)
547 uint64_t l = 0, r = 0;
550 l = cl_offset(left->mem_info->daddr.addr);
552 r = cl_offset(right->mem_info->daddr.addr);
554 return (int64_t)(r - l);
558 iaddr_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
559 struct hist_entry *he)
562 int width = c2c_width(fmt, hpp, he->hists);
566 addr = he->mem_info->iaddr.addr;
568 return scnprintf(hpp->buf, hpp->size, "%*s", width, HEX_STR(buf, addr));
572 iaddr_cmp(struct perf_hpp_fmt *fmt __maybe_unused,
573 struct hist_entry *left, struct hist_entry *right)
575 return sort__iaddr_cmp(left, right);
579 tot_hitm_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
580 struct hist_entry *he)
582 struct c2c_hist_entry *c2c_he;
583 int width = c2c_width(fmt, hpp, he->hists);
584 unsigned int tot_hitm;
586 c2c_he = container_of(he, struct c2c_hist_entry, he);
587 tot_hitm = c2c_he->stats.lcl_hitm + c2c_he->stats.rmt_hitm;
589 return scnprintf(hpp->buf, hpp->size, "%*u", width, tot_hitm);
593 tot_hitm_cmp(struct perf_hpp_fmt *fmt __maybe_unused,
594 struct hist_entry *left, struct hist_entry *right)
596 struct c2c_hist_entry *c2c_left;
597 struct c2c_hist_entry *c2c_right;
598 uint64_t tot_hitm_left;
599 uint64_t tot_hitm_right;
601 c2c_left = container_of(left, struct c2c_hist_entry, he);
602 c2c_right = container_of(right, struct c2c_hist_entry, he);
604 tot_hitm_left = c2c_left->stats.lcl_hitm + c2c_left->stats.rmt_hitm;
605 tot_hitm_right = c2c_right->stats.lcl_hitm + c2c_right->stats.rmt_hitm;
607 return tot_hitm_left - tot_hitm_right;
610 #define STAT_FN_ENTRY(__f) \
612 __f ## _entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp, \
613 struct hist_entry *he) \
615 struct c2c_hist_entry *c2c_he; \
616 int width = c2c_width(fmt, hpp, he->hists); \
618 c2c_he = container_of(he, struct c2c_hist_entry, he); \
619 return scnprintf(hpp->buf, hpp->size, "%*u", width, \
620 c2c_he->stats.__f); \
623 #define STAT_FN_CMP(__f) \
625 __f ## _cmp(struct perf_hpp_fmt *fmt __maybe_unused, \
626 struct hist_entry *left, struct hist_entry *right) \
628 struct c2c_hist_entry *c2c_left, *c2c_right; \
630 c2c_left = container_of(left, struct c2c_hist_entry, he); \
631 c2c_right = container_of(right, struct c2c_hist_entry, he); \
632 return (uint64_t) c2c_left->stats.__f - \
633 (uint64_t) c2c_right->stats.__f; \
636 #define STAT_FN(__f) \
651 static uint64_t llc_miss(struct c2c_stats *stats)
655 llcmiss = stats->lcl_dram +
664 ld_llcmiss_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
665 struct hist_entry *he)
667 struct c2c_hist_entry *c2c_he;
668 int width = c2c_width(fmt, hpp, he->hists);
670 c2c_he = container_of(he, struct c2c_hist_entry, he);
672 return scnprintf(hpp->buf, hpp->size, "%*lu", width,
673 llc_miss(&c2c_he->stats));
677 ld_llcmiss_cmp(struct perf_hpp_fmt *fmt __maybe_unused,
678 struct hist_entry *left, struct hist_entry *right)
680 struct c2c_hist_entry *c2c_left;
681 struct c2c_hist_entry *c2c_right;
683 c2c_left = container_of(left, struct c2c_hist_entry, he);
684 c2c_right = container_of(right, struct c2c_hist_entry, he);
686 return (uint64_t) llc_miss(&c2c_left->stats) -
687 (uint64_t) llc_miss(&c2c_right->stats);
690 static uint64_t total_records(struct c2c_stats *stats)
692 uint64_t lclmiss, ldcnt, total;
694 lclmiss = stats->lcl_dram +
714 tot_recs_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
715 struct hist_entry *he)
717 struct c2c_hist_entry *c2c_he;
718 int width = c2c_width(fmt, hpp, he->hists);
721 c2c_he = container_of(he, struct c2c_hist_entry, he);
722 tot_recs = total_records(&c2c_he->stats);
724 return scnprintf(hpp->buf, hpp->size, "%*" PRIu64, width, tot_recs);
728 tot_recs_cmp(struct perf_hpp_fmt *fmt __maybe_unused,
729 struct hist_entry *left, struct hist_entry *right)
731 struct c2c_hist_entry *c2c_left;
732 struct c2c_hist_entry *c2c_right;
733 uint64_t tot_recs_left;
734 uint64_t tot_recs_right;
736 c2c_left = container_of(left, struct c2c_hist_entry, he);
737 c2c_right = container_of(right, struct c2c_hist_entry, he);
739 tot_recs_left = total_records(&c2c_left->stats);
740 tot_recs_right = total_records(&c2c_right->stats);
742 return tot_recs_left - tot_recs_right;
745 static uint64_t total_loads(struct c2c_stats *stats)
747 uint64_t lclmiss, ldcnt;
749 lclmiss = stats->lcl_dram +
765 tot_loads_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
766 struct hist_entry *he)
768 struct c2c_hist_entry *c2c_he;
769 int width = c2c_width(fmt, hpp, he->hists);
772 c2c_he = container_of(he, struct c2c_hist_entry, he);
773 tot_recs = total_loads(&c2c_he->stats);
775 return scnprintf(hpp->buf, hpp->size, "%*" PRIu64, width, tot_recs);
779 tot_loads_cmp(struct perf_hpp_fmt *fmt __maybe_unused,
780 struct hist_entry *left, struct hist_entry *right)
782 struct c2c_hist_entry *c2c_left;
783 struct c2c_hist_entry *c2c_right;
784 uint64_t tot_recs_left;
785 uint64_t tot_recs_right;
787 c2c_left = container_of(left, struct c2c_hist_entry, he);
788 c2c_right = container_of(right, struct c2c_hist_entry, he);
790 tot_recs_left = total_loads(&c2c_left->stats);
791 tot_recs_right = total_loads(&c2c_right->stats);
793 return tot_recs_left - tot_recs_right;
796 typedef double (get_percent_cb)(struct c2c_hist_entry *);
799 percent_color(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
800 struct hist_entry *he, get_percent_cb get_percent)
802 struct c2c_hist_entry *c2c_he;
803 int width = c2c_width(fmt, hpp, he->hists);
806 c2c_he = container_of(he, struct c2c_hist_entry, he);
807 per = get_percent(c2c_he);
809 #ifdef HAVE_SLANG_SUPPORT
811 return __hpp__slsmg_color_printf(hpp, "%*.2f%%", width - 1, per);
813 return hpp_color_scnprintf(hpp, "%*.2f%%", width - 1, per);
816 static double percent_hitm(struct c2c_hist_entry *c2c_he)
818 struct c2c_hists *hists;
819 struct c2c_stats *stats;
820 struct c2c_stats *total;
824 hists = container_of(c2c_he->he.hists, struct c2c_hists, hists);
825 stats = &c2c_he->stats;
826 total = &hists->stats;
828 switch (c2c.display) {
830 st = stats->rmt_hitm;
831 tot = total->rmt_hitm;
834 st = stats->lcl_hitm;
835 tot = total->lcl_hitm;
838 st = stats->tot_hitm;
839 tot = total->tot_hitm;
844 p = tot ? (double) st / tot : 0;
849 #define PERC_STR(__s, __v) \
851 scnprintf(__s, sizeof(__s), "%.2F%%", __v); \
856 percent_hitm_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
857 struct hist_entry *he)
859 struct c2c_hist_entry *c2c_he;
860 int width = c2c_width(fmt, hpp, he->hists);
864 c2c_he = container_of(he, struct c2c_hist_entry, he);
865 per = percent_hitm(c2c_he);
866 return scnprintf(hpp->buf, hpp->size, "%*s", width, PERC_STR(buf, per));
870 percent_hitm_color(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
871 struct hist_entry *he)
873 return percent_color(fmt, hpp, he, percent_hitm);
877 percent_hitm_cmp(struct perf_hpp_fmt *fmt __maybe_unused,
878 struct hist_entry *left, struct hist_entry *right)
880 struct c2c_hist_entry *c2c_left;
881 struct c2c_hist_entry *c2c_right;
885 c2c_left = container_of(left, struct c2c_hist_entry, he);
886 c2c_right = container_of(right, struct c2c_hist_entry, he);
888 per_left = percent_hitm(c2c_left);
889 per_right = percent_hitm(c2c_right);
891 return per_left - per_right;
894 static struct c2c_stats *he_stats(struct hist_entry *he)
896 struct c2c_hist_entry *c2c_he;
898 c2c_he = container_of(he, struct c2c_hist_entry, he);
899 return &c2c_he->stats;
902 static struct c2c_stats *total_stats(struct hist_entry *he)
904 struct c2c_hists *hists;
906 hists = container_of(he->hists, struct c2c_hists, hists);
907 return &hists->stats;
910 static double percent(int st, int tot)
912 return tot ? 100. * (double) st / (double) tot : 0;
915 #define PERCENT(__h, __f) percent(he_stats(__h)->__f, total_stats(__h)->__f)
917 #define PERCENT_FN(__f) \
918 static double percent_ ## __f(struct c2c_hist_entry *c2c_he) \
920 struct c2c_hists *hists; \
922 hists = container_of(c2c_he->he.hists, struct c2c_hists, hists); \
923 return percent(c2c_he->stats.__f, hists->stats.__f); \
929 PERCENT_FN(st_l1miss)
932 percent_rmt_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, rmt_hitm);
939 return scnprintf(hpp->buf, hpp->size, "%*s", width, PERC_STR(buf, per));
943 percent_rmt_hitm_color(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
944 struct hist_entry *he)
946 return percent_color(fmt, hpp, he, percent_rmt_hitm);
950 percent_rmt_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_lcl_hitm_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, lcl_hitm);
970 return scnprintf(hpp->buf, hpp->size, "%*s", width, PERC_STR(buf, per));
974 percent_lcl_hitm_color(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
975 struct hist_entry *he)
977 return percent_color(fmt, hpp, he, percent_lcl_hitm);
981 percent_lcl_hitm_cmp(struct perf_hpp_fmt *fmt __maybe_unused,
982 struct hist_entry *left, struct hist_entry *right)
987 per_left = PERCENT(left, lcl_hitm);
988 per_right = PERCENT(right, lcl_hitm);
990 return per_left - per_right;
994 percent_stores_l1hit_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_l1hit);
1001 return scnprintf(hpp->buf, hpp->size, "%*s", width, PERC_STR(buf, per));
1005 percent_stores_l1hit_color(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
1006 struct hist_entry *he)
1008 return percent_color(fmt, hpp, he, percent_st_l1hit);
1012 percent_stores_l1hit_cmp(struct perf_hpp_fmt *fmt __maybe_unused,
1013 struct hist_entry *left, struct hist_entry *right)
1018 per_left = PERCENT(left, st_l1hit);
1019 per_right = PERCENT(right, st_l1hit);
1021 return per_left - per_right;
1025 percent_stores_l1miss_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
1026 struct hist_entry *he)
1028 int width = c2c_width(fmt, hpp, he->hists);
1029 double per = PERCENT(he, st_l1miss);
1032 return scnprintf(hpp->buf, hpp->size, "%*s", width, PERC_STR(buf, per));
1036 percent_stores_l1miss_color(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
1037 struct hist_entry *he)
1039 return percent_color(fmt, hpp, he, percent_st_l1miss);
1043 percent_stores_l1miss_cmp(struct perf_hpp_fmt *fmt __maybe_unused,
1044 struct hist_entry *left, struct hist_entry *right)
1049 per_left = PERCENT(left, st_l1miss);
1050 per_right = PERCENT(right, st_l1miss);
1052 return per_left - per_right;
1059 pid_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
1060 struct hist_entry *he)
1062 int width = c2c_width(fmt, hpp, he->hists);
1064 return scnprintf(hpp->buf, hpp->size, "%*d", width, he->thread->pid_);
1068 pid_cmp(struct perf_hpp_fmt *fmt __maybe_unused,
1069 struct hist_entry *left, struct hist_entry *right)
1071 return left->thread->pid_ - right->thread->pid_;
1075 empty_cmp(struct perf_hpp_fmt *fmt __maybe_unused,
1076 struct hist_entry *left __maybe_unused,
1077 struct hist_entry *right __maybe_unused)
1083 node_entry(struct perf_hpp_fmt *fmt __maybe_unused, struct perf_hpp *hpp,
1084 struct hist_entry *he)
1086 struct c2c_hist_entry *c2c_he;
1091 c2c_he = container_of(he, struct c2c_hist_entry, he);
1093 for (node = 0; node < c2c.nodes_cnt; node++) {
1094 DECLARE_BITMAP(set, c2c.cpus_cnt);
1096 bitmap_zero(set, c2c.cpus_cnt);
1097 bitmap_and(set, c2c_he->cpuset, c2c.nodes[node], c2c.cpus_cnt);
1099 if (!bitmap_weight(set, c2c.cpus_cnt)) {
1100 if (c2c.node_info == 1) {
1101 ret = scnprintf(hpp->buf, hpp->size, "%21s", " ");
1102 advance_hpp(hpp, ret);
1108 ret = scnprintf(hpp->buf, hpp->size, " ");
1109 advance_hpp(hpp, ret);
1112 switch (c2c.node_info) {
1114 ret = scnprintf(hpp->buf, hpp->size, "%2d", node);
1115 advance_hpp(hpp, ret);
1119 int num = bitmap_weight(set, c2c.cpus_cnt);
1120 struct c2c_stats *stats = &c2c_he->node_stats[node];
1122 ret = scnprintf(hpp->buf, hpp->size, "%2d{%2d ", node, num);
1123 advance_hpp(hpp, ret);
1125 #define DISPLAY_HITM(__h) \
1126 if (c2c_he->stats.__h> 0) { \
1127 ret = scnprintf(hpp->buf, hpp->size, "%5.1f%% ", \
1128 percent(stats->__h, c2c_he->stats.__h));\
1130 ret = scnprintf(hpp->buf, hpp->size, "%6s ", "n/a"); \
1133 switch (c2c.display) {
1135 DISPLAY_HITM(rmt_hitm);
1138 DISPLAY_HITM(lcl_hitm);
1141 DISPLAY_HITM(tot_hitm);
1148 advance_hpp(hpp, ret);
1150 if (c2c_he->stats.store > 0) {
1151 ret = scnprintf(hpp->buf, hpp->size, "%5.1f%%}",
1152 percent(stats->store, c2c_he->stats.store));
1154 ret = scnprintf(hpp->buf, hpp->size, "%6s}", "n/a");
1157 advance_hpp(hpp, ret);
1161 ret = scnprintf(hpp->buf, hpp->size, "%2d{", node);
1162 advance_hpp(hpp, ret);
1164 ret = bitmap_scnprintf(set, c2c.cpus_cnt, hpp->buf, hpp->size);
1165 advance_hpp(hpp, ret);
1167 ret = scnprintf(hpp->buf, hpp->size, "}");
1168 advance_hpp(hpp, ret);
1181 mean_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
1182 struct hist_entry *he, double mean)
1184 int width = c2c_width(fmt, hpp, he->hists);
1187 scnprintf(buf, 10, "%6.0f", mean);
1188 return scnprintf(hpp->buf, hpp->size, "%*s", width, buf);
1191 #define MEAN_ENTRY(__func, __val) \
1193 __func(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp, struct hist_entry *he) \
1195 struct c2c_hist_entry *c2c_he; \
1196 c2c_he = container_of(he, struct c2c_hist_entry, he); \
1197 return mean_entry(fmt, hpp, he, avg_stats(&c2c_he->cstats.__val)); \
1200 MEAN_ENTRY(mean_rmt_entry, rmt_hitm);
1201 MEAN_ENTRY(mean_lcl_entry, lcl_hitm);
1202 MEAN_ENTRY(mean_load_entry, load);
1205 cpucnt_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
1206 struct hist_entry *he)
1208 struct c2c_hist_entry *c2c_he;
1209 int width = c2c_width(fmt, hpp, he->hists);
1212 c2c_he = container_of(he, struct c2c_hist_entry, he);
1214 scnprintf(buf, 10, "%d", bitmap_weight(c2c_he->cpuset, c2c.cpus_cnt));
1215 return scnprintf(hpp->buf, hpp->size, "%*s", width, buf);
1219 cl_idx_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
1220 struct hist_entry *he)
1222 struct c2c_hist_entry *c2c_he;
1223 int width = c2c_width(fmt, hpp, he->hists);
1226 c2c_he = container_of(he, struct c2c_hist_entry, he);
1228 scnprintf(buf, 10, "%u", c2c_he->cacheline_idx);
1229 return scnprintf(hpp->buf, hpp->size, "%*s", width, buf);
1233 cl_idx_empty_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
1234 struct hist_entry *he)
1236 int width = c2c_width(fmt, hpp, he->hists);
1238 return scnprintf(hpp->buf, hpp->size, "%*s", width, "");
1241 #define HEADER_LOW(__h) \
1248 #define HEADER_BOTH(__h0, __h1) \
1258 #define HEADER_SPAN(__h0, __h1, __s) \
1269 #define HEADER_SPAN_LOW(__h) \
1276 static struct c2c_dimension dim_dcacheline = {
1277 .header = HEADER_SPAN("--- Cacheline ----", "Address", 2),
1278 .name = "dcacheline",
1279 .cmp = dcacheline_cmp,
1280 .entry = dcacheline_entry,
1284 static struct c2c_dimension dim_dcacheline_node = {
1285 .header = HEADER_LOW("Node"),
1286 .name = "dcacheline_node",
1288 .entry = dcacheline_node_entry,
1292 static struct c2c_dimension dim_dcacheline_count = {
1293 .header = HEADER_LOW("PA cnt"),
1294 .name = "dcacheline_count",
1296 .entry = dcacheline_node_count,
1300 static struct c2c_header header_offset_tui = HEADER_SPAN("-----", "Off", 2);
1302 static struct c2c_dimension dim_offset = {
1303 .header = HEADER_SPAN("--- Data address -", "Offset", 2),
1306 .entry = offset_entry,
1310 static struct c2c_dimension dim_offset_node = {
1311 .header = HEADER_LOW("Node"),
1312 .name = "offset_node",
1314 .entry = dcacheline_node_entry,
1318 static struct c2c_dimension dim_iaddr = {
1319 .header = HEADER_LOW("Code address"),
1322 .entry = iaddr_entry,
1326 static struct c2c_dimension dim_tot_hitm = {
1327 .header = HEADER_SPAN("----- LLC Load Hitm -----", "Total", 2),
1329 .cmp = tot_hitm_cmp,
1330 .entry = tot_hitm_entry,
1334 static struct c2c_dimension dim_lcl_hitm = {
1335 .header = HEADER_SPAN_LOW("Lcl"),
1337 .cmp = lcl_hitm_cmp,
1338 .entry = lcl_hitm_entry,
1342 static struct c2c_dimension dim_rmt_hitm = {
1343 .header = HEADER_SPAN_LOW("Rmt"),
1345 .cmp = rmt_hitm_cmp,
1346 .entry = rmt_hitm_entry,
1350 static struct c2c_dimension dim_cl_rmt_hitm = {
1351 .header = HEADER_SPAN("----- HITM -----", "Rmt", 1),
1352 .name = "cl_rmt_hitm",
1353 .cmp = rmt_hitm_cmp,
1354 .entry = rmt_hitm_entry,
1358 static struct c2c_dimension dim_cl_lcl_hitm = {
1359 .header = HEADER_SPAN_LOW("Lcl"),
1360 .name = "cl_lcl_hitm",
1361 .cmp = lcl_hitm_cmp,
1362 .entry = lcl_hitm_entry,
1366 static struct c2c_dimension dim_stores = {
1367 .header = HEADER_SPAN("---- Store Reference ----", "Total", 2),
1370 .entry = store_entry,
1374 static struct c2c_dimension dim_stores_l1hit = {
1375 .header = HEADER_SPAN_LOW("L1Hit"),
1376 .name = "stores_l1hit",
1377 .cmp = st_l1hit_cmp,
1378 .entry = st_l1hit_entry,
1382 static struct c2c_dimension dim_stores_l1miss = {
1383 .header = HEADER_SPAN_LOW("L1Miss"),
1384 .name = "stores_l1miss",
1385 .cmp = st_l1miss_cmp,
1386 .entry = st_l1miss_entry,
1390 static struct c2c_dimension dim_cl_stores_l1hit = {
1391 .header = HEADER_SPAN("-- Store Refs --", "L1 Hit", 1),
1392 .name = "cl_stores_l1hit",
1393 .cmp = st_l1hit_cmp,
1394 .entry = st_l1hit_entry,
1398 static struct c2c_dimension dim_cl_stores_l1miss = {
1399 .header = HEADER_SPAN_LOW("L1 Miss"),
1400 .name = "cl_stores_l1miss",
1401 .cmp = st_l1miss_cmp,
1402 .entry = st_l1miss_entry,
1406 static struct c2c_dimension dim_ld_fbhit = {
1407 .header = HEADER_SPAN("----- Core Load Hit -----", "FB", 2),
1409 .cmp = ld_fbhit_cmp,
1410 .entry = ld_fbhit_entry,
1414 static struct c2c_dimension dim_ld_l1hit = {
1415 .header = HEADER_SPAN_LOW("L1"),
1417 .cmp = ld_l1hit_cmp,
1418 .entry = ld_l1hit_entry,
1422 static struct c2c_dimension dim_ld_l2hit = {
1423 .header = HEADER_SPAN_LOW("L2"),
1425 .cmp = ld_l2hit_cmp,
1426 .entry = ld_l2hit_entry,
1430 static struct c2c_dimension dim_ld_llchit = {
1431 .header = HEADER_SPAN("-- LLC Load Hit --", "Llc", 1),
1432 .name = "ld_lclhit",
1433 .cmp = ld_llchit_cmp,
1434 .entry = ld_llchit_entry,
1438 static struct c2c_dimension dim_ld_rmthit = {
1439 .header = HEADER_SPAN_LOW("Rmt"),
1440 .name = "ld_rmthit",
1442 .entry = rmt_hit_entry,
1446 static struct c2c_dimension dim_ld_llcmiss = {
1447 .header = HEADER_BOTH("LLC", "Ld Miss"),
1448 .name = "ld_llcmiss",
1449 .cmp = ld_llcmiss_cmp,
1450 .entry = ld_llcmiss_entry,
1454 static struct c2c_dimension dim_tot_recs = {
1455 .header = HEADER_BOTH("Total", "records"),
1457 .cmp = tot_recs_cmp,
1458 .entry = tot_recs_entry,
1462 static struct c2c_dimension dim_tot_loads = {
1463 .header = HEADER_BOTH("Total", "Loads"),
1464 .name = "tot_loads",
1465 .cmp = tot_loads_cmp,
1466 .entry = tot_loads_entry,
1470 static struct c2c_header percent_hitm_header[] = {
1471 [DISPLAY_LCL] = HEADER_BOTH("Lcl", "Hitm"),
1472 [DISPLAY_RMT] = HEADER_BOTH("Rmt", "Hitm"),
1473 [DISPLAY_TOT] = HEADER_BOTH("Tot", "Hitm"),
1476 static struct c2c_dimension dim_percent_hitm = {
1477 .name = "percent_hitm",
1478 .cmp = percent_hitm_cmp,
1479 .entry = percent_hitm_entry,
1480 .color = percent_hitm_color,
1484 static struct c2c_dimension dim_percent_rmt_hitm = {
1485 .header = HEADER_SPAN("----- HITM -----", "Rmt", 1),
1486 .name = "percent_rmt_hitm",
1487 .cmp = percent_rmt_hitm_cmp,
1488 .entry = percent_rmt_hitm_entry,
1489 .color = percent_rmt_hitm_color,
1493 static struct c2c_dimension dim_percent_lcl_hitm = {
1494 .header = HEADER_SPAN_LOW("Lcl"),
1495 .name = "percent_lcl_hitm",
1496 .cmp = percent_lcl_hitm_cmp,
1497 .entry = percent_lcl_hitm_entry,
1498 .color = percent_lcl_hitm_color,
1502 static struct c2c_dimension dim_percent_stores_l1hit = {
1503 .header = HEADER_SPAN("-- Store Refs --", "L1 Hit", 1),
1504 .name = "percent_stores_l1hit",
1505 .cmp = percent_stores_l1hit_cmp,
1506 .entry = percent_stores_l1hit_entry,
1507 .color = percent_stores_l1hit_color,
1511 static struct c2c_dimension dim_percent_stores_l1miss = {
1512 .header = HEADER_SPAN_LOW("L1 Miss"),
1513 .name = "percent_stores_l1miss",
1514 .cmp = percent_stores_l1miss_cmp,
1515 .entry = percent_stores_l1miss_entry,
1516 .color = percent_stores_l1miss_color,
1520 static struct c2c_dimension dim_dram_lcl = {
1521 .header = HEADER_SPAN("--- Load Dram ----", "Lcl", 1),
1523 .cmp = lcl_dram_cmp,
1524 .entry = lcl_dram_entry,
1528 static struct c2c_dimension dim_dram_rmt = {
1529 .header = HEADER_SPAN_LOW("Rmt"),
1531 .cmp = rmt_dram_cmp,
1532 .entry = rmt_dram_entry,
1536 static struct c2c_dimension dim_pid = {
1537 .header = HEADER_LOW("Pid"),
1544 static struct c2c_dimension dim_tid = {
1545 .header = HEADER_LOW("Tid"),
1550 static struct c2c_dimension dim_symbol = {
1555 static struct c2c_dimension dim_dso = {
1556 .header = HEADER_BOTH("Shared", "Object"),
1561 static struct c2c_header header_node[3] = {
1563 HEADER_LOW("Node{cpus %hitms %stores}"),
1564 HEADER_LOW("Node{cpu list}"),
1567 static struct c2c_dimension dim_node = {
1570 .entry = node_entry,
1574 static struct c2c_dimension dim_mean_rmt = {
1575 .header = HEADER_SPAN("---------- cycles ----------", "rmt hitm", 2),
1578 .entry = mean_rmt_entry,
1582 static struct c2c_dimension dim_mean_lcl = {
1583 .header = HEADER_SPAN_LOW("lcl hitm"),
1586 .entry = mean_lcl_entry,
1590 static struct c2c_dimension dim_mean_load = {
1591 .header = HEADER_SPAN_LOW("load"),
1592 .name = "mean_load",
1594 .entry = mean_load_entry,
1598 static struct c2c_dimension dim_cpucnt = {
1599 .header = HEADER_BOTH("cpu", "cnt"),
1602 .entry = cpucnt_entry,
1606 static struct c2c_dimension dim_srcline = {
1607 .name = "cl_srcline",
1608 .se = &sort_srcline,
1611 static struct c2c_dimension dim_dcacheline_idx = {
1612 .header = HEADER_LOW("Index"),
1615 .entry = cl_idx_entry,
1619 static struct c2c_dimension dim_dcacheline_num = {
1620 .header = HEADER_LOW("Num"),
1623 .entry = cl_idx_entry,
1627 static struct c2c_dimension dim_dcacheline_num_empty = {
1628 .header = HEADER_LOW("Num"),
1629 .name = "cl_num_empty",
1631 .entry = cl_idx_empty_entry,
1635 static struct c2c_dimension *dimensions[] = {
1637 &dim_dcacheline_node,
1638 &dim_dcacheline_count,
1650 &dim_cl_stores_l1hit,
1651 &dim_cl_stores_l1miss,
1661 &dim_percent_rmt_hitm,
1662 &dim_percent_lcl_hitm,
1663 &dim_percent_stores_l1hit,
1664 &dim_percent_stores_l1miss,
1677 &dim_dcacheline_idx,
1678 &dim_dcacheline_num,
1679 &dim_dcacheline_num_empty,
1683 static void fmt_free(struct perf_hpp_fmt *fmt)
1685 struct c2c_fmt *c2c_fmt;
1687 c2c_fmt = container_of(fmt, struct c2c_fmt, fmt);
1691 static bool fmt_equal(struct perf_hpp_fmt *a, struct perf_hpp_fmt *b)
1693 struct c2c_fmt *c2c_a = container_of(a, struct c2c_fmt, fmt);
1694 struct c2c_fmt *c2c_b = container_of(b, struct c2c_fmt, fmt);
1696 return c2c_a->dim == c2c_b->dim;
1699 static struct c2c_dimension *get_dimension(const char *name)
1703 for (i = 0; dimensions[i]; i++) {
1704 struct c2c_dimension *dim = dimensions[i];
1706 if (!strcmp(dim->name, name))
1713 static int c2c_se_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
1714 struct hist_entry *he)
1716 struct c2c_fmt *c2c_fmt = container_of(fmt, struct c2c_fmt, fmt);
1717 struct c2c_dimension *dim = c2c_fmt->dim;
1718 size_t len = fmt->user_len;
1721 len = hists__col_len(he->hists, dim->se->se_width_idx);
1723 if (dim == &dim_symbol || dim == &dim_srcline)
1724 len = symbol_width(he->hists, dim->se);
1727 return dim->se->se_snprintf(he, hpp->buf, hpp->size, len);
1730 static int64_t c2c_se_cmp(struct perf_hpp_fmt *fmt,
1731 struct hist_entry *a, struct hist_entry *b)
1733 struct c2c_fmt *c2c_fmt = container_of(fmt, struct c2c_fmt, fmt);
1734 struct c2c_dimension *dim = c2c_fmt->dim;
1736 return dim->se->se_cmp(a, b);
1739 static int64_t c2c_se_collapse(struct perf_hpp_fmt *fmt,
1740 struct hist_entry *a, struct hist_entry *b)
1742 struct c2c_fmt *c2c_fmt = container_of(fmt, struct c2c_fmt, fmt);
1743 struct c2c_dimension *dim = c2c_fmt->dim;
1744 int64_t (*collapse_fn)(struct hist_entry *, struct hist_entry *);
1746 collapse_fn = dim->se->se_collapse ?: dim->se->se_cmp;
1747 return collapse_fn(a, b);
1750 static struct c2c_fmt *get_format(const char *name)
1752 struct c2c_dimension *dim = get_dimension(name);
1753 struct c2c_fmt *c2c_fmt;
1754 struct perf_hpp_fmt *fmt;
1759 c2c_fmt = zalloc(sizeof(*c2c_fmt));
1765 fmt = &c2c_fmt->fmt;
1766 INIT_LIST_HEAD(&fmt->list);
1767 INIT_LIST_HEAD(&fmt->sort_list);
1769 fmt->cmp = dim->se ? c2c_se_cmp : dim->cmp;
1770 fmt->sort = dim->se ? c2c_se_cmp : dim->cmp;
1771 fmt->color = dim->se ? NULL : dim->color;
1772 fmt->entry = dim->se ? c2c_se_entry : dim->entry;
1773 fmt->header = c2c_header;
1774 fmt->width = c2c_width;
1775 fmt->collapse = dim->se ? c2c_se_collapse : dim->cmp;
1776 fmt->equal = fmt_equal;
1777 fmt->free = fmt_free;
1782 static int c2c_hists__init_output(struct perf_hpp_list *hpp_list, char *name)
1784 struct c2c_fmt *c2c_fmt = get_format(name);
1788 return output_field_add(hpp_list, name);
1791 perf_hpp_list__column_register(hpp_list, &c2c_fmt->fmt);
1795 static int c2c_hists__init_sort(struct perf_hpp_list *hpp_list, char *name)
1797 struct c2c_fmt *c2c_fmt = get_format(name);
1798 struct c2c_dimension *dim;
1802 return sort_dimension__add(hpp_list, name, NULL, 0);
1806 if (dim == &dim_dso)
1809 perf_hpp_list__register_sort_field(hpp_list, &c2c_fmt->fmt);
1813 #define PARSE_LIST(_list, _fn) \
1821 for (tok = strtok_r((char *)_list, ", ", &tmp); \
1822 tok; tok = strtok_r(NULL, ", ", &tmp)) { \
1823 ret = _fn(hpp_list, tok); \
1824 if (ret == -EINVAL) { \
1825 pr_err("Invalid --fields key: `%s'", tok); \
1827 } else if (ret == -ESRCH) { \
1828 pr_err("Unknown --fields key: `%s'", tok); \
1834 static int hpp_list__parse(struct perf_hpp_list *hpp_list,
1835 const char *output_,
1838 char *output = output_ ? strdup(output_) : NULL;
1839 char *sort = sort_ ? strdup(sort_) : NULL;
1842 PARSE_LIST(output, c2c_hists__init_output);
1843 PARSE_LIST(sort, c2c_hists__init_sort);
1845 /* copy sort keys to output fields */
1846 perf_hpp__setup_output_field(hpp_list);
1849 * We dont need other sorting keys other than those
1850 * we already specified. It also really slows down
1851 * the processing a lot with big number of output
1852 * fields, so switching this off for c2c.
1856 /* and then copy output fields to sort keys */
1857 perf_hpp__append_sort_keys(&hists->list);
1865 static int c2c_hists__init(struct c2c_hists *hists,
1867 int nr_header_lines)
1869 __hists__init(&hists->hists, &hists->list);
1872 * Initialize only with sort fields, we need to resort
1873 * later anyway, and that's where we add output fields
1876 perf_hpp_list__init(&hists->list);
1878 /* Overload number of header lines.*/
1879 hists->list.nr_header_lines = nr_header_lines;
1881 return hpp_list__parse(&hists->list, NULL, sort);
1884 static int c2c_hists__reinit(struct c2c_hists *c2c_hists,
1888 perf_hpp__reset_output_field(&c2c_hists->list);
1889 return hpp_list__parse(&c2c_hists->list, output, sort);
1892 #define DISPLAY_LINE_LIMIT 0.001
1894 static bool he__display(struct hist_entry *he, struct c2c_stats *stats)
1896 struct c2c_hist_entry *c2c_he;
1902 c2c_he = container_of(he, struct c2c_hist_entry, he);
1904 #define FILTER_HITM(__h) \
1906 ld_dist = ((double)c2c_he->stats.__h / stats->__h); \
1907 if (ld_dist < DISPLAY_LINE_LIMIT) \
1908 he->filtered = HIST_FILTER__C2C; \
1910 he->filtered = HIST_FILTER__C2C; \
1913 switch (c2c.display) {
1915 FILTER_HITM(lcl_hitm);
1918 FILTER_HITM(rmt_hitm);
1921 FILTER_HITM(tot_hitm);
1928 return he->filtered == 0;
1931 static inline int valid_hitm_or_store(struct hist_entry *he)
1933 struct c2c_hist_entry *c2c_he;
1936 c2c_he = container_of(he, struct c2c_hist_entry, he);
1937 has_hitm = c2c.display == DISPLAY_TOT ? c2c_he->stats.tot_hitm :
1938 c2c.display == DISPLAY_LCL ? c2c_he->stats.lcl_hitm :
1939 c2c_he->stats.rmt_hitm;
1940 return has_hitm || c2c_he->stats.store;
1943 static void set_node_width(struct c2c_hist_entry *c2c_he, int len)
1945 struct c2c_dimension *dim;
1947 dim = &c2c.hists == c2c_he->hists ?
1948 &dim_dcacheline_node : &dim_offset_node;
1950 if (len > dim->width)
1954 static int set_nodestr(struct c2c_hist_entry *c2c_he)
1959 if (c2c_he->nodestr)
1962 if (bitmap_weight(c2c_he->nodeset, c2c.nodes_cnt)) {
1963 len = bitmap_scnprintf(c2c_he->nodeset, c2c.nodes_cnt,
1966 len = scnprintf(buf, sizeof(buf), "N/A");
1969 set_node_width(c2c_he, len);
1970 c2c_he->nodestr = strdup(buf);
1971 return c2c_he->nodestr ? 0 : -ENOMEM;
1974 static void calc_width(struct c2c_hist_entry *c2c_he)
1976 struct c2c_hists *c2c_hists;
1978 c2c_hists = container_of(c2c_he->he.hists, struct c2c_hists, hists);
1979 hists__calc_col_len(&c2c_hists->hists, &c2c_he->he);
1980 set_nodestr(c2c_he);
1983 static int filter_cb(struct hist_entry *he, void *arg __maybe_unused)
1985 struct c2c_hist_entry *c2c_he;
1987 c2c_he = container_of(he, struct c2c_hist_entry, he);
1989 if (c2c.show_src && !he->srcline)
1990 he->srcline = hist_entry__srcline(he);
1994 if (!valid_hitm_or_store(he))
1995 he->filtered = HIST_FILTER__C2C;
2000 static int resort_cl_cb(struct hist_entry *he, void *arg __maybe_unused)
2002 struct c2c_hist_entry *c2c_he;
2003 struct c2c_hists *c2c_hists;
2004 bool display = he__display(he, &c2c.hitm_stats);
2006 c2c_he = container_of(he, struct c2c_hist_entry, he);
2007 c2c_hists = c2c_he->hists;
2009 if (display && c2c_hists) {
2010 static unsigned int idx;
2012 c2c_he->cacheline_idx = idx++;
2015 c2c_hists__reinit(c2c_hists, c2c.cl_output, c2c.cl_resort);
2017 hists__collapse_resort(&c2c_hists->hists, NULL);
2018 hists__output_resort_cb(&c2c_hists->hists, NULL, filter_cb);
2024 static void setup_nodes_header(void)
2026 dim_node.header = header_node[c2c.node_info];
2029 static int setup_nodes(struct perf_session *session)
2031 struct numa_node *n;
2032 unsigned long **nodes;
2036 if (c2c.node_info > 2)
2039 c2c.nodes_cnt = session->header.env.nr_numa_nodes;
2040 c2c.cpus_cnt = session->header.env.nr_cpus_avail;
2042 n = session->header.env.numa_nodes;
2046 nodes = zalloc(sizeof(unsigned long *) * c2c.nodes_cnt);
2052 cpu2node = zalloc(sizeof(int) * c2c.cpus_cnt);
2056 for (cpu = 0; cpu < c2c.cpus_cnt; cpu++)
2059 c2c.cpu2node = cpu2node;
2061 for (node = 0; node < c2c.nodes_cnt; node++) {
2062 struct perf_cpu_map *map = n[node].map;
2065 set = bitmap_alloc(c2c.cpus_cnt);
2071 /* empty node, skip */
2072 if (perf_cpu_map__empty(map))
2075 for (cpu = 0; cpu < map->nr; cpu++) {
2076 set_bit(map->map[cpu], set);
2078 if (WARN_ONCE(cpu2node[map->map[cpu]] != -1, "node/cpu topology bug"))
2081 cpu2node[map->map[cpu]] = node;
2085 setup_nodes_header();
2089 #define HAS_HITMS(__h) ((__h)->stats.lcl_hitm || (__h)->stats.rmt_hitm)
2091 static int resort_hitm_cb(struct hist_entry *he, void *arg __maybe_unused)
2093 struct c2c_hist_entry *c2c_he;
2094 c2c_he = container_of(he, struct c2c_hist_entry, he);
2096 if (HAS_HITMS(c2c_he)) {
2097 c2c.shared_clines++;
2098 c2c_add_stats(&c2c.hitm_stats, &c2c_he->stats);
2104 static int hists__iterate_cb(struct hists *hists, hists__resort_cb_t cb)
2106 struct rb_node *next = rb_first_cached(&hists->entries);
2110 struct hist_entry *he;
2112 he = rb_entry(next, struct hist_entry, rb_node);
2116 next = rb_next(&he->rb_node);
2122 static void print_c2c__display_stats(FILE *out)
2125 struct c2c_stats *stats = &c2c.hists.stats;
2127 llc_misses = stats->lcl_dram +
2132 fprintf(out, "=================================================\n");
2133 fprintf(out, " Trace Event Information \n");
2134 fprintf(out, "=================================================\n");
2135 fprintf(out, " Total records : %10d\n", stats->nr_entries);
2136 fprintf(out, " Locked Load/Store Operations : %10d\n", stats->locks);
2137 fprintf(out, " Load Operations : %10d\n", stats->load);
2138 fprintf(out, " Loads - uncacheable : %10d\n", stats->ld_uncache);
2139 fprintf(out, " Loads - IO : %10d\n", stats->ld_io);
2140 fprintf(out, " Loads - Miss : %10d\n", stats->ld_miss);
2141 fprintf(out, " Loads - no mapping : %10d\n", stats->ld_noadrs);
2142 fprintf(out, " Load Fill Buffer Hit : %10d\n", stats->ld_fbhit);
2143 fprintf(out, " Load L1D hit : %10d\n", stats->ld_l1hit);
2144 fprintf(out, " Load L2D hit : %10d\n", stats->ld_l2hit);
2145 fprintf(out, " Load LLC hit : %10d\n", stats->ld_llchit + stats->lcl_hitm);
2146 fprintf(out, " Load Local HITM : %10d\n", stats->lcl_hitm);
2147 fprintf(out, " Load Remote HITM : %10d\n", stats->rmt_hitm);
2148 fprintf(out, " Load Remote HIT : %10d\n", stats->rmt_hit);
2149 fprintf(out, " Load Local DRAM : %10d\n", stats->lcl_dram);
2150 fprintf(out, " Load Remote DRAM : %10d\n", stats->rmt_dram);
2151 fprintf(out, " Load MESI State Exclusive : %10d\n", stats->ld_excl);
2152 fprintf(out, " Load MESI State Shared : %10d\n", stats->ld_shared);
2153 fprintf(out, " Load LLC Misses : %10d\n", llc_misses);
2154 fprintf(out, " LLC Misses to Local DRAM : %10.1f%%\n", ((double)stats->lcl_dram/(double)llc_misses) * 100.);
2155 fprintf(out, " LLC Misses to Remote DRAM : %10.1f%%\n", ((double)stats->rmt_dram/(double)llc_misses) * 100.);
2156 fprintf(out, " LLC Misses to Remote cache (HIT) : %10.1f%%\n", ((double)stats->rmt_hit /(double)llc_misses) * 100.);
2157 fprintf(out, " LLC Misses to Remote cache (HITM) : %10.1f%%\n", ((double)stats->rmt_hitm/(double)llc_misses) * 100.);
2158 fprintf(out, " Store Operations : %10d\n", stats->store);
2159 fprintf(out, " Store - uncacheable : %10d\n", stats->st_uncache);
2160 fprintf(out, " Store - no mapping : %10d\n", stats->st_noadrs);
2161 fprintf(out, " Store L1D Hit : %10d\n", stats->st_l1hit);
2162 fprintf(out, " Store L1D Miss : %10d\n", stats->st_l1miss);
2163 fprintf(out, " No Page Map Rejects : %10d\n", stats->nomap);
2164 fprintf(out, " Unable to parse data source : %10d\n", stats->noparse);
2167 static void print_shared_cacheline_info(FILE *out)
2169 struct c2c_stats *stats = &c2c.hitm_stats;
2170 int hitm_cnt = stats->lcl_hitm + stats->rmt_hitm;
2172 fprintf(out, "=================================================\n");
2173 fprintf(out, " Global Shared Cache Line Event Information \n");
2174 fprintf(out, "=================================================\n");
2175 fprintf(out, " Total Shared Cache Lines : %10d\n", c2c.shared_clines);
2176 fprintf(out, " Load HITs on shared lines : %10d\n", stats->load);
2177 fprintf(out, " Fill Buffer Hits on shared lines : %10d\n", stats->ld_fbhit);
2178 fprintf(out, " L1D hits on shared lines : %10d\n", stats->ld_l1hit);
2179 fprintf(out, " L2D hits on shared lines : %10d\n", stats->ld_l2hit);
2180 fprintf(out, " LLC hits on shared lines : %10d\n", stats->ld_llchit + stats->lcl_hitm);
2181 fprintf(out, " Locked Access on shared lines : %10d\n", stats->locks);
2182 fprintf(out, " Store HITs on shared lines : %10d\n", stats->store);
2183 fprintf(out, " Store L1D hits on shared lines : %10d\n", stats->st_l1hit);
2184 fprintf(out, " Total Merged records : %10d\n", hitm_cnt + stats->store);
2187 static void print_cacheline(struct c2c_hists *c2c_hists,
2188 struct hist_entry *he_cl,
2189 struct perf_hpp_list *hpp_list,
2193 struct perf_hpp hpp = {
2200 hists__fprintf_headers(&c2c_hists->hists, out);
2206 fprintf(out, " -------------------------------------------------------------\n");
2207 __hist_entry__snprintf(he_cl, &hpp, hpp_list);
2208 fprintf(out, "%s\n", bf);
2209 fprintf(out, " -------------------------------------------------------------\n");
2211 hists__fprintf(&c2c_hists->hists, false, 0, 0, 0, out, false);
2214 static void print_pareto(FILE *out)
2216 struct perf_hpp_list hpp_list;
2220 perf_hpp_list__init(&hpp_list);
2221 ret = hpp_list__parse(&hpp_list,
2230 if (WARN_ONCE(ret, "failed to setup sort entries\n"))
2233 nd = rb_first_cached(&c2c.hists.hists.entries);
2235 for (; nd; nd = rb_next(nd)) {
2236 struct hist_entry *he = rb_entry(nd, struct hist_entry, rb_node);
2237 struct c2c_hist_entry *c2c_he;
2242 c2c_he = container_of(he, struct c2c_hist_entry, he);
2243 print_cacheline(c2c_he->hists, he, &hpp_list, out);
2247 static void print_c2c_info(FILE *out, struct perf_session *session)
2249 struct evlist *evlist = session->evlist;
2250 struct evsel *evsel;
2253 fprintf(out, "=================================================\n");
2254 fprintf(out, " c2c details \n");
2255 fprintf(out, "=================================================\n");
2257 evlist__for_each_entry(evlist, evsel) {
2258 fprintf(out, "%-36s: %s\n", first ? " Events" : "",
2259 perf_evsel__name(evsel));
2262 fprintf(out, " Cachelines sort on : %s HITMs\n",
2263 display_str[c2c.display]);
2264 fprintf(out, " Cacheline data grouping : %s\n", c2c.cl_sort);
2267 static void perf_c2c__hists_fprintf(FILE *out, struct perf_session *session)
2271 print_c2c__display_stats(out);
2273 print_shared_cacheline_info(out);
2275 print_c2c_info(out, session);
2281 fprintf(out, "=================================================\n");
2282 fprintf(out, " Shared Data Cache Line Table \n");
2283 fprintf(out, "=================================================\n");
2284 fprintf(out, "#\n");
2286 hists__fprintf(&c2c.hists.hists, true, 0, 0, 0, stdout, true);
2289 fprintf(out, "=================================================\n");
2290 fprintf(out, " Shared Cache Line Distribution Pareto \n");
2291 fprintf(out, "=================================================\n");
2292 fprintf(out, "#\n");
2297 #ifdef HAVE_SLANG_SUPPORT
2298 static void c2c_browser__update_nr_entries(struct hist_browser *hb)
2301 struct rb_node *nd = rb_first_cached(&hb->hists->entries);
2304 struct hist_entry *he = rb_entry(nd, struct hist_entry, rb_node);
2312 hb->nr_non_filtered_entries = nr_entries;
2315 struct c2c_cacheline_browser {
2316 struct hist_browser hb;
2317 struct hist_entry *he;
2321 perf_c2c_cacheline_browser__title(struct hist_browser *browser,
2322 char *bf, size_t size)
2324 struct c2c_cacheline_browser *cl_browser;
2325 struct hist_entry *he;
2328 cl_browser = container_of(browser, struct c2c_cacheline_browser, hb);
2329 he = cl_browser->he;
2332 addr = cl_address(he->mem_info->daddr.addr);
2334 scnprintf(bf, size, "Cacheline 0x%lx", addr);
2338 static struct c2c_cacheline_browser*
2339 c2c_cacheline_browser__new(struct hists *hists, struct hist_entry *he)
2341 struct c2c_cacheline_browser *browser;
2343 browser = zalloc(sizeof(*browser));
2345 hist_browser__init(&browser->hb, hists);
2346 browser->hb.c2c_filter = true;
2347 browser->hb.title = perf_c2c_cacheline_browser__title;
2354 static int perf_c2c__browse_cacheline(struct hist_entry *he)
2356 struct c2c_hist_entry *c2c_he;
2357 struct c2c_hists *c2c_hists;
2358 struct c2c_cacheline_browser *cl_browser;
2359 struct hist_browser *browser;
2361 static const char help[] =
2362 " ENTER Toggle callchains (if present) \n"
2363 " n Toggle Node details info \n"
2364 " s Toggle full length of symbol and source line columns \n"
2365 " q Return back to cacheline list \n";
2370 /* Display compact version first. */
2371 c2c.symbol_full = false;
2373 c2c_he = container_of(he, struct c2c_hist_entry, he);
2374 c2c_hists = c2c_he->hists;
2376 cl_browser = c2c_cacheline_browser__new(&c2c_hists->hists, he);
2377 if (cl_browser == NULL)
2380 browser = &cl_browser->hb;
2382 /* reset abort key so that it can get Ctrl-C as a key */
2384 SLang_init_tty(0, 0, 0);
2386 c2c_browser__update_nr_entries(browser);
2389 key = hist_browser__run(browser, "? - help", true, 0);
2393 c2c.symbol_full = !c2c.symbol_full;
2396 c2c.node_info = (c2c.node_info + 1) % 3;
2397 setup_nodes_header();
2402 ui_browser__help_window(&browser->b, help);
2414 static int perf_c2c_browser__title(struct hist_browser *browser,
2415 char *bf, size_t size)
2418 "Shared Data Cache Line Table "
2419 "(%lu entries, sorted on %s HITMs)",
2420 browser->nr_non_filtered_entries,
2421 display_str[c2c.display]);
2425 static struct hist_browser*
2426 perf_c2c_browser__new(struct hists *hists)
2428 struct hist_browser *browser = hist_browser__new(hists);
2431 browser->title = perf_c2c_browser__title;
2432 browser->c2c_filter = true;
2438 static int perf_c2c__hists_browse(struct hists *hists)
2440 struct hist_browser *browser;
2442 static const char help[] =
2443 " d Display cacheline details \n"
2444 " ENTER Toggle callchains (if present) \n"
2447 browser = perf_c2c_browser__new(hists);
2448 if (browser == NULL)
2451 /* reset abort key so that it can get Ctrl-C as a key */
2453 SLang_init_tty(0, 0, 0);
2455 c2c_browser__update_nr_entries(browser);
2458 key = hist_browser__run(browser, "? - help", true, 0);
2464 perf_c2c__browse_cacheline(browser->he_selection);
2467 ui_browser__help_window(&browser->b, help);
2475 hist_browser__delete(browser);
2479 static void perf_c2c_display(struct perf_session *session)
2481 if (use_browser == 0)
2482 perf_c2c__hists_fprintf(stdout, session);
2484 perf_c2c__hists_browse(&c2c.hists.hists);
2487 static void perf_c2c_display(struct perf_session *session)
2490 perf_c2c__hists_fprintf(stdout, session);
2492 #endif /* HAVE_SLANG_SUPPORT */
2494 static char *fill_line(const char *orig, int len)
2496 int i, j, olen = strlen(orig);
2499 buf = zalloc(len + 1);
2503 j = len / 2 - olen / 2;
2505 for (i = 0; i < j - 1; i++)
2510 strcpy(buf + i, orig);
2516 for (; i < len; i++)
2522 static int ui_quirks(void)
2524 const char *nodestr = "Data address";
2527 if (!c2c.use_stdio) {
2528 dim_offset.width = 5;
2529 dim_offset.header = header_offset_tui;
2533 dim_percent_hitm.header = percent_hitm_header[c2c.display];
2535 /* Fix the zero line for dcacheline column. */
2536 buf = fill_line("Cacheline", dim_dcacheline.width +
2537 dim_dcacheline_node.width +
2538 dim_dcacheline_count.width + 4);
2542 dim_dcacheline.header.line[0].text = buf;
2544 /* Fix the zero line for offset column. */
2545 buf = fill_line(nodestr, dim_offset.width +
2546 dim_offset_node.width +
2547 dim_dcacheline_count.width + 4);
2551 dim_offset.header.line[0].text = buf;
2556 #define CALLCHAIN_DEFAULT_OPT "graph,0.5,caller,function,percent"
2558 const char callchain_help[] = "Display call graph (stack chain/backtrace):\n\n"
2559 CALLCHAIN_REPORT_HELP
2560 "\n\t\t\t\tDefault: " CALLCHAIN_DEFAULT_OPT;
2563 parse_callchain_opt(const struct option *opt, const char *arg, int unset)
2565 struct callchain_param *callchain = opt->value;
2567 callchain->enabled = !unset;
2572 symbol_conf.use_callchain = false;
2573 callchain->mode = CHAIN_NONE;
2577 return parse_callchain_report_opt(arg);
2580 static int setup_callchain(struct evlist *evlist)
2582 u64 sample_type = perf_evlist__combined_sample_type(evlist);
2583 enum perf_call_graph_mode mode = CALLCHAIN_NONE;
2585 if ((sample_type & PERF_SAMPLE_REGS_USER) &&
2586 (sample_type & PERF_SAMPLE_STACK_USER)) {
2587 mode = CALLCHAIN_DWARF;
2588 dwarf_callchain_users = true;
2589 } else if (sample_type & PERF_SAMPLE_BRANCH_STACK)
2590 mode = CALLCHAIN_LBR;
2591 else if (sample_type & PERF_SAMPLE_CALLCHAIN)
2592 mode = CALLCHAIN_FP;
2594 if (!callchain_param.enabled &&
2595 callchain_param.mode != CHAIN_NONE &&
2596 mode != CALLCHAIN_NONE) {
2597 symbol_conf.use_callchain = true;
2598 if (callchain_register_param(&callchain_param) < 0) {
2599 ui__error("Can't register callchain params.\n");
2604 callchain_param.record_mode = mode;
2605 callchain_param.min_percent = 0;
2609 static int setup_display(const char *str)
2611 const char *display = str ?: "tot";
2613 if (!strcmp(display, "tot"))
2614 c2c.display = DISPLAY_TOT;
2615 else if (!strcmp(display, "rmt"))
2616 c2c.display = DISPLAY_RMT;
2617 else if (!strcmp(display, "lcl"))
2618 c2c.display = DISPLAY_LCL;
2620 pr_err("failed: unknown display type: %s\n", str);
2627 #define for_each_token(__tok, __buf, __sep, __tmp) \
2628 for (__tok = strtok_r(__buf, __sep, &__tmp); __tok; \
2629 __tok = strtok_r(NULL, __sep, &__tmp))
2631 static int build_cl_output(char *cl_sort, bool no_source)
2633 char *tok, *tmp, *buf = strdup(cl_sort);
2634 bool add_pid = false;
2635 bool add_tid = false;
2636 bool add_iaddr = false;
2637 bool add_sym = false;
2638 bool add_dso = false;
2639 bool add_src = false;
2645 for_each_token(tok, buf, ",", tmp) {
2646 if (!strcmp(tok, "tid")) {
2648 } else if (!strcmp(tok, "pid")) {
2650 } else if (!strcmp(tok, "iaddr")) {
2654 add_src = no_source ? false : true;
2655 } else if (!strcmp(tok, "dso")) {
2657 } else if (strcmp(tok, "offset")) {
2658 pr_err("unrecognized sort token: %s\n", tok);
2664 if (asprintf(&c2c.cl_output,
2665 "%s%s%s%s%s%s%s%s%s%s",
2666 c2c.use_stdio ? "cl_num_empty," : "",
2669 "percent_stores_l1hit,"
2670 "percent_stores_l1miss,"
2671 "offset,offset_node,dcacheline_count,",
2672 add_pid ? "pid," : "",
2673 add_tid ? "tid," : "",
2674 add_iaddr ? "iaddr," : "",
2680 add_sym ? "symbol," : "",
2681 add_dso ? "dso," : "",
2682 add_src ? "cl_srcline," : "",
2688 c2c.show_src = add_src;
2694 static int setup_coalesce(const char *coalesce, bool no_source)
2696 const char *c = coalesce ?: coalesce_default;
2698 if (asprintf(&c2c.cl_sort, "offset,%s", c) < 0)
2701 if (build_cl_output(c2c.cl_sort, no_source))
2704 if (asprintf(&c2c.cl_resort, "offset,%s",
2705 c2c.display == DISPLAY_TOT ?
2707 c2c.display == DISPLAY_RMT ?
2708 "rmt_hitm,lcl_hitm" :
2709 "lcl_hitm,rmt_hitm") < 0)
2712 pr_debug("coalesce sort fields: %s\n", c2c.cl_sort);
2713 pr_debug("coalesce resort fields: %s\n", c2c.cl_resort);
2714 pr_debug("coalesce output fields: %s\n", c2c.cl_output);
2718 static int perf_c2c__report(int argc, const char **argv)
2720 struct perf_session *session;
2721 struct ui_progress prog;
2722 struct perf_data data = {
2723 .mode = PERF_DATA_MODE_READ,
2725 char callchain_default_opt[] = CALLCHAIN_DEFAULT_OPT;
2726 const char *display = NULL;
2727 const char *coalesce = NULL;
2728 bool no_source = false;
2729 const struct option options[] = {
2730 OPT_STRING('k', "vmlinux", &symbol_conf.vmlinux_name,
2731 "file", "vmlinux pathname"),
2732 OPT_STRING('i', "input", &input_name, "file",
2733 "the input file to process"),
2734 OPT_INCR('N', "node-info", &c2c.node_info,
2735 "show extra node info in report (repeat for more info)"),
2736 #ifdef HAVE_SLANG_SUPPORT
2737 OPT_BOOLEAN(0, "stdio", &c2c.use_stdio, "Use the stdio interface"),
2739 OPT_BOOLEAN(0, "stats", &c2c.stats_only,
2740 "Display only statistic tables (implies --stdio)"),
2741 OPT_BOOLEAN(0, "full-symbols", &c2c.symbol_full,
2742 "Display full length of symbols"),
2743 OPT_BOOLEAN(0, "no-source", &no_source,
2744 "Do not display Source Line column"),
2745 OPT_BOOLEAN(0, "show-all", &c2c.show_all,
2746 "Show all captured HITM lines."),
2747 OPT_CALLBACK_DEFAULT('g', "call-graph", &callchain_param,
2748 "print_type,threshold[,print_limit],order,sort_key[,branch],value",
2749 callchain_help, &parse_callchain_opt,
2750 callchain_default_opt),
2751 OPT_STRING('d', "display", &display, "Switch HITM output type", "lcl,rmt"),
2752 OPT_STRING('c', "coalesce", &coalesce, "coalesce fields",
2753 "coalesce fields: pid,tid,iaddr,dso"),
2754 OPT_BOOLEAN('f', "force", &symbol_conf.force, "don't complain, do it"),
2755 OPT_PARENT(c2c_options),
2760 argc = parse_options(argc, argv, options, report_c2c_usage,
2761 PARSE_OPT_STOP_AT_NON_OPTION);
2763 usage_with_options(report_c2c_usage, options);
2766 c2c.use_stdio = true;
2768 if (!input_name || !strlen(input_name))
2769 input_name = "perf.data";
2771 data.path = input_name;
2772 data.force = symbol_conf.force;
2774 err = setup_display(display);
2778 err = setup_coalesce(coalesce, no_source);
2780 pr_debug("Failed to initialize hists\n");
2784 err = c2c_hists__init(&c2c.hists, "dcacheline", 2);
2786 pr_debug("Failed to initialize hists\n");
2790 session = perf_session__new(&data, 0, &c2c.tool);
2791 if (IS_ERR(session)) {
2792 err = PTR_ERR(session);
2793 pr_debug("Error creating perf session\n");
2797 err = setup_nodes(session);
2799 pr_err("Failed setup nodes\n");
2803 err = mem2node__init(&c2c.mem2node, &session->header.env);
2807 err = setup_callchain(session->evlist);
2811 if (symbol__init(&session->header.env) < 0)
2814 /* No pipe support at the moment. */
2815 if (perf_data__is_pipe(session->data)) {
2816 pr_debug("No pipe support at the moment.\n");
2825 setup_browser(false);
2827 err = perf_session__process_events(session);
2829 pr_err("failed to process sample\n");
2833 c2c_hists__reinit(&c2c.hists,
2840 "tot_hitm,lcl_hitm,rmt_hitm,"
2841 "stores,stores_l1hit,stores_l1miss,"
2842 "dram_lcl,dram_rmt,"
2845 "ld_fbhit,ld_l1hit,ld_l2hit,"
2846 "ld_lclhit,ld_rmthit",
2847 c2c.display == DISPLAY_TOT ? "tot_hitm" :
2848 c2c.display == DISPLAY_LCL ? "lcl_hitm" : "rmt_hitm"
2851 ui_progress__init(&prog, c2c.hists.hists.nr_entries, "Sorting...");
2853 hists__collapse_resort(&c2c.hists.hists, NULL);
2854 hists__output_resort_cb(&c2c.hists.hists, &prog, resort_hitm_cb);
2855 hists__iterate_cb(&c2c.hists.hists, resort_cl_cb);
2857 ui_progress__finish();
2860 pr_err("failed to setup UI\n");
2864 perf_c2c_display(session);
2867 mem2node__exit(&c2c.mem2node);
2869 perf_session__delete(session);
2874 static int parse_record_events(const struct option *opt,
2875 const char *str, int unset __maybe_unused)
2877 bool *event_set = (bool *) opt->value;
2880 return perf_mem_events__parse(str);
2884 static const char * const __usage_record[] = {
2885 "perf c2c record [<options>] [<command>]",
2886 "perf c2c record [<options>] -- <command> [<options>]",
2890 static const char * const *record_mem_usage = __usage_record;
2892 static int perf_c2c__record(int argc, const char **argv)
2894 int rec_argc, i = 0, j;
2895 const char **rec_argv;
2897 bool all_user = false, all_kernel = false;
2898 bool event_set = false;
2899 struct option options[] = {
2900 OPT_CALLBACK('e', "event", &event_set, "event",
2901 "event selector. Use 'perf mem record -e list' to list available events",
2902 parse_record_events),
2903 OPT_BOOLEAN('u', "all-user", &all_user, "collect only user level data"),
2904 OPT_BOOLEAN('k', "all-kernel", &all_kernel, "collect only kernel level data"),
2905 OPT_UINTEGER('l', "ldlat", &perf_mem_events__loads_ldlat, "setup mem-loads latency"),
2906 OPT_PARENT(c2c_options),
2910 if (perf_mem_events__init()) {
2911 pr_err("failed: memory events not supported\n");
2915 argc = parse_options(argc, argv, options, record_mem_usage,
2916 PARSE_OPT_KEEP_UNKNOWN);
2918 rec_argc = argc + 11; /* max number of arguments */
2919 rec_argv = calloc(rec_argc + 1, sizeof(char *));
2923 rec_argv[i++] = "record";
2926 perf_mem_events[PERF_MEM_EVENTS__LOAD].record = true;
2927 perf_mem_events[PERF_MEM_EVENTS__STORE].record = true;
2930 if (perf_mem_events[PERF_MEM_EVENTS__LOAD].record)
2931 rec_argv[i++] = "-W";
2933 rec_argv[i++] = "-d";
2934 rec_argv[i++] = "--phys-data";
2935 rec_argv[i++] = "--sample-cpu";
2937 for (j = 0; j < PERF_MEM_EVENTS__MAX; j++) {
2938 if (!perf_mem_events[j].record)
2941 if (!perf_mem_events[j].supported) {
2942 pr_err("failed: event '%s' not supported\n",
2943 perf_mem_events[j].name);
2948 rec_argv[i++] = "-e";
2949 rec_argv[i++] = perf_mem_events__name(j);
2953 rec_argv[i++] = "--all-user";
2956 rec_argv[i++] = "--all-kernel";
2958 for (j = 0; j < argc; j++, i++)
2959 rec_argv[i] = argv[j];
2962 pr_debug("calling: ");
2966 while (rec_argv[j]) {
2967 pr_debug("%s ", rec_argv[j]);
2973 ret = cmd_record(i, rec_argv);
2978 int cmd_c2c(int argc, const char **argv)
2980 argc = parse_options(argc, argv, c2c_options, c2c_usage,
2981 PARSE_OPT_STOP_AT_NON_OPTION);
2984 usage_with_options(c2c_usage, c2c_options);
2986 if (!strncmp(argv[0], "rec", 3)) {
2987 return perf_c2c__record(argc, argv);
2988 } else if (!strncmp(argv[0], "rep", 3)) {
2989 return perf_c2c__report(argc, argv);
2991 usage_with_options(c2c_usage, c2c_options);