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 /* HITM shared clines stats */
101 struct c2c_stats hitm_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 .ordered_events = true,
373 .ordering_requires_timestamps = true,
377 static const char * const c2c_usage[] = {
378 "perf c2c {record|report}",
382 static const char * const __usage_report[] = {
387 static const char * const *report_c2c_usage = __usage_report;
389 #define C2C_HEADER_MAX 2
395 } line[C2C_HEADER_MAX];
398 struct c2c_dimension {
399 struct c2c_header header;
402 struct sort_entry *se;
404 int64_t (*cmp)(struct perf_hpp_fmt *fmt,
405 struct hist_entry *, struct hist_entry *);
406 int (*entry)(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
407 struct hist_entry *he);
408 int (*color)(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
409 struct hist_entry *he);
413 struct perf_hpp_fmt fmt;
414 struct c2c_dimension *dim;
417 #define SYMBOL_WIDTH 30
419 static struct c2c_dimension dim_symbol;
420 static struct c2c_dimension dim_srcline;
422 static int symbol_width(struct hists *hists, struct sort_entry *se)
424 int width = hists__col_len(hists, se->se_width_idx);
426 if (!c2c.symbol_full)
427 width = MIN(width, SYMBOL_WIDTH);
432 static int c2c_width(struct perf_hpp_fmt *fmt,
433 struct perf_hpp *hpp __maybe_unused,
436 struct c2c_fmt *c2c_fmt;
437 struct c2c_dimension *dim;
439 c2c_fmt = container_of(fmt, struct c2c_fmt, fmt);
442 if (dim == &dim_symbol || dim == &dim_srcline)
443 return symbol_width(hists, dim->se);
445 return dim->se ? hists__col_len(hists, dim->se->se_width_idx) :
449 static int c2c_header(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
450 struct hists *hists, int line, int *span)
452 struct perf_hpp_list *hpp_list = hists->hpp_list;
453 struct c2c_fmt *c2c_fmt;
454 struct c2c_dimension *dim;
455 const char *text = NULL;
456 int width = c2c_width(fmt, hpp, hists);
458 c2c_fmt = container_of(fmt, struct c2c_fmt, fmt);
462 text = dim->header.line[line].text;
463 /* Use the last line from sort_entry if not defined. */
464 if (!text && (line == hpp_list->nr_header_lines - 1))
465 text = dim->se->se_header;
467 text = dim->header.line[line].text;
473 *span = dim->header.line[line].span;
480 return scnprintf(hpp->buf, hpp->size, "%*s", width, text);
483 #define HEX_STR(__s, __v) \
485 scnprintf(__s, sizeof(__s), "0x%" PRIx64, __v); \
490 dcacheline_cmp(struct perf_hpp_fmt *fmt __maybe_unused,
491 struct hist_entry *left, struct hist_entry *right)
493 return sort__dcacheline_cmp(left, right);
496 static int dcacheline_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
497 struct hist_entry *he)
500 int width = c2c_width(fmt, hpp, he->hists);
504 addr = cl_address(he->mem_info->daddr.addr);
506 return scnprintf(hpp->buf, hpp->size, "%*s", width, HEX_STR(buf, addr));
510 dcacheline_node_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
511 struct hist_entry *he)
513 struct c2c_hist_entry *c2c_he;
514 int width = c2c_width(fmt, hpp, he->hists);
516 c2c_he = container_of(he, struct c2c_hist_entry, he);
517 if (WARN_ON_ONCE(!c2c_he->nodestr))
520 return scnprintf(hpp->buf, hpp->size, "%*s", width, c2c_he->nodestr);
524 dcacheline_node_count(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
525 struct hist_entry *he)
527 struct c2c_hist_entry *c2c_he;
528 int width = c2c_width(fmt, hpp, he->hists);
530 c2c_he = container_of(he, struct c2c_hist_entry, he);
531 return scnprintf(hpp->buf, hpp->size, "%*lu", width, c2c_he->paddr_cnt);
534 static int offset_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
535 struct hist_entry *he)
538 int width = c2c_width(fmt, hpp, he->hists);
542 addr = cl_offset(he->mem_info->daddr.al_addr);
544 return scnprintf(hpp->buf, hpp->size, "%*s", width, HEX_STR(buf, addr));
548 offset_cmp(struct perf_hpp_fmt *fmt __maybe_unused,
549 struct hist_entry *left, struct hist_entry *right)
551 uint64_t l = 0, r = 0;
554 l = cl_offset(left->mem_info->daddr.addr);
556 r = cl_offset(right->mem_info->daddr.addr);
558 return (int64_t)(r - l);
562 iaddr_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
563 struct hist_entry *he)
566 int width = c2c_width(fmt, hpp, he->hists);
570 addr = he->mem_info->iaddr.addr;
572 return scnprintf(hpp->buf, hpp->size, "%*s", width, HEX_STR(buf, addr));
576 iaddr_cmp(struct perf_hpp_fmt *fmt __maybe_unused,
577 struct hist_entry *left, struct hist_entry *right)
579 return sort__iaddr_cmp(left, right);
583 tot_hitm_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
584 struct hist_entry *he)
586 struct c2c_hist_entry *c2c_he;
587 int width = c2c_width(fmt, hpp, he->hists);
588 unsigned int tot_hitm;
590 c2c_he = container_of(he, struct c2c_hist_entry, he);
591 tot_hitm = c2c_he->stats.lcl_hitm + c2c_he->stats.rmt_hitm;
593 return scnprintf(hpp->buf, hpp->size, "%*u", width, tot_hitm);
597 tot_hitm_cmp(struct perf_hpp_fmt *fmt __maybe_unused,
598 struct hist_entry *left, struct hist_entry *right)
600 struct c2c_hist_entry *c2c_left;
601 struct c2c_hist_entry *c2c_right;
602 uint64_t tot_hitm_left;
603 uint64_t tot_hitm_right;
605 c2c_left = container_of(left, struct c2c_hist_entry, he);
606 c2c_right = container_of(right, struct c2c_hist_entry, he);
608 tot_hitm_left = c2c_left->stats.lcl_hitm + c2c_left->stats.rmt_hitm;
609 tot_hitm_right = c2c_right->stats.lcl_hitm + c2c_right->stats.rmt_hitm;
611 return tot_hitm_left - tot_hitm_right;
614 #define STAT_FN_ENTRY(__f) \
616 __f ## _entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp, \
617 struct hist_entry *he) \
619 struct c2c_hist_entry *c2c_he; \
620 int width = c2c_width(fmt, hpp, he->hists); \
622 c2c_he = container_of(he, struct c2c_hist_entry, he); \
623 return scnprintf(hpp->buf, hpp->size, "%*u", width, \
624 c2c_he->stats.__f); \
627 #define STAT_FN_CMP(__f) \
629 __f ## _cmp(struct perf_hpp_fmt *fmt __maybe_unused, \
630 struct hist_entry *left, struct hist_entry *right) \
632 struct c2c_hist_entry *c2c_left, *c2c_right; \
634 c2c_left = container_of(left, struct c2c_hist_entry, he); \
635 c2c_right = container_of(right, struct c2c_hist_entry, he); \
636 return (uint64_t) c2c_left->stats.__f - \
637 (uint64_t) c2c_right->stats.__f; \
640 #define STAT_FN(__f) \
655 static uint64_t llc_miss(struct c2c_stats *stats)
659 llcmiss = stats->lcl_dram +
668 ld_llcmiss_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
669 struct hist_entry *he)
671 struct c2c_hist_entry *c2c_he;
672 int width = c2c_width(fmt, hpp, he->hists);
674 c2c_he = container_of(he, struct c2c_hist_entry, he);
676 return scnprintf(hpp->buf, hpp->size, "%*lu", width,
677 llc_miss(&c2c_he->stats));
681 ld_llcmiss_cmp(struct perf_hpp_fmt *fmt __maybe_unused,
682 struct hist_entry *left, struct hist_entry *right)
684 struct c2c_hist_entry *c2c_left;
685 struct c2c_hist_entry *c2c_right;
687 c2c_left = container_of(left, struct c2c_hist_entry, he);
688 c2c_right = container_of(right, struct c2c_hist_entry, he);
690 return (uint64_t) llc_miss(&c2c_left->stats) -
691 (uint64_t) llc_miss(&c2c_right->stats);
694 static uint64_t total_records(struct c2c_stats *stats)
696 uint64_t lclmiss, ldcnt, total;
698 lclmiss = stats->lcl_dram +
718 tot_recs_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
719 struct hist_entry *he)
721 struct c2c_hist_entry *c2c_he;
722 int width = c2c_width(fmt, hpp, he->hists);
725 c2c_he = container_of(he, struct c2c_hist_entry, he);
726 tot_recs = total_records(&c2c_he->stats);
728 return scnprintf(hpp->buf, hpp->size, "%*" PRIu64, width, tot_recs);
732 tot_recs_cmp(struct perf_hpp_fmt *fmt __maybe_unused,
733 struct hist_entry *left, struct hist_entry *right)
735 struct c2c_hist_entry *c2c_left;
736 struct c2c_hist_entry *c2c_right;
737 uint64_t tot_recs_left;
738 uint64_t tot_recs_right;
740 c2c_left = container_of(left, struct c2c_hist_entry, he);
741 c2c_right = container_of(right, struct c2c_hist_entry, he);
743 tot_recs_left = total_records(&c2c_left->stats);
744 tot_recs_right = total_records(&c2c_right->stats);
746 return tot_recs_left - tot_recs_right;
749 static uint64_t total_loads(struct c2c_stats *stats)
751 uint64_t lclmiss, ldcnt;
753 lclmiss = stats->lcl_dram +
769 tot_loads_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
770 struct hist_entry *he)
772 struct c2c_hist_entry *c2c_he;
773 int width = c2c_width(fmt, hpp, he->hists);
776 c2c_he = container_of(he, struct c2c_hist_entry, he);
777 tot_recs = total_loads(&c2c_he->stats);
779 return scnprintf(hpp->buf, hpp->size, "%*" PRIu64, width, tot_recs);
783 tot_loads_cmp(struct perf_hpp_fmt *fmt __maybe_unused,
784 struct hist_entry *left, struct hist_entry *right)
786 struct c2c_hist_entry *c2c_left;
787 struct c2c_hist_entry *c2c_right;
788 uint64_t tot_recs_left;
789 uint64_t tot_recs_right;
791 c2c_left = container_of(left, struct c2c_hist_entry, he);
792 c2c_right = container_of(right, struct c2c_hist_entry, he);
794 tot_recs_left = total_loads(&c2c_left->stats);
795 tot_recs_right = total_loads(&c2c_right->stats);
797 return tot_recs_left - tot_recs_right;
800 typedef double (get_percent_cb)(struct c2c_hist_entry *);
803 percent_color(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
804 struct hist_entry *he, get_percent_cb get_percent)
806 struct c2c_hist_entry *c2c_he;
807 int width = c2c_width(fmt, hpp, he->hists);
810 c2c_he = container_of(he, struct c2c_hist_entry, he);
811 per = get_percent(c2c_he);
813 #ifdef HAVE_SLANG_SUPPORT
815 return __hpp__slsmg_color_printf(hpp, "%*.2f%%", width - 1, per);
817 return hpp_color_scnprintf(hpp, "%*.2f%%", width - 1, per);
820 static double percent_hitm(struct c2c_hist_entry *c2c_he)
822 struct c2c_hists *hists;
823 struct c2c_stats *stats;
824 struct c2c_stats *total;
828 hists = container_of(c2c_he->he.hists, struct c2c_hists, hists);
829 stats = &c2c_he->stats;
830 total = &hists->stats;
832 switch (c2c.display) {
834 st = stats->rmt_hitm;
835 tot = total->rmt_hitm;
838 st = stats->lcl_hitm;
839 tot = total->lcl_hitm;
842 st = stats->tot_hitm;
843 tot = total->tot_hitm;
848 p = tot ? (double) st / tot : 0;
853 #define PERC_STR(__s, __v) \
855 scnprintf(__s, sizeof(__s), "%.2F%%", __v); \
860 percent_hitm_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
861 struct hist_entry *he)
863 struct c2c_hist_entry *c2c_he;
864 int width = c2c_width(fmt, hpp, he->hists);
868 c2c_he = container_of(he, struct c2c_hist_entry, he);
869 per = percent_hitm(c2c_he);
870 return scnprintf(hpp->buf, hpp->size, "%*s", width, PERC_STR(buf, per));
874 percent_hitm_color(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
875 struct hist_entry *he)
877 return percent_color(fmt, hpp, he, percent_hitm);
881 percent_hitm_cmp(struct perf_hpp_fmt *fmt __maybe_unused,
882 struct hist_entry *left, struct hist_entry *right)
884 struct c2c_hist_entry *c2c_left;
885 struct c2c_hist_entry *c2c_right;
889 c2c_left = container_of(left, struct c2c_hist_entry, he);
890 c2c_right = container_of(right, struct c2c_hist_entry, he);
892 per_left = percent_hitm(c2c_left);
893 per_right = percent_hitm(c2c_right);
895 return per_left - per_right;
898 static struct c2c_stats *he_stats(struct hist_entry *he)
900 struct c2c_hist_entry *c2c_he;
902 c2c_he = container_of(he, struct c2c_hist_entry, he);
903 return &c2c_he->stats;
906 static struct c2c_stats *total_stats(struct hist_entry *he)
908 struct c2c_hists *hists;
910 hists = container_of(he->hists, struct c2c_hists, hists);
911 return &hists->stats;
914 static double percent(int st, int tot)
916 return tot ? 100. * (double) st / (double) tot : 0;
919 #define PERCENT(__h, __f) percent(he_stats(__h)->__f, total_stats(__h)->__f)
921 #define PERCENT_FN(__f) \
922 static double percent_ ## __f(struct c2c_hist_entry *c2c_he) \
924 struct c2c_hists *hists; \
926 hists = container_of(c2c_he->he.hists, struct c2c_hists, hists); \
927 return percent(c2c_he->stats.__f, hists->stats.__f); \
933 PERCENT_FN(st_l1miss)
936 percent_rmt_hitm_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
937 struct hist_entry *he)
939 int width = c2c_width(fmt, hpp, he->hists);
940 double per = PERCENT(he, rmt_hitm);
943 return scnprintf(hpp->buf, hpp->size, "%*s", width, PERC_STR(buf, per));
947 percent_rmt_hitm_color(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
948 struct hist_entry *he)
950 return percent_color(fmt, hpp, he, percent_rmt_hitm);
954 percent_rmt_hitm_cmp(struct perf_hpp_fmt *fmt __maybe_unused,
955 struct hist_entry *left, struct hist_entry *right)
960 per_left = PERCENT(left, lcl_hitm);
961 per_right = PERCENT(right, lcl_hitm);
963 return per_left - per_right;
967 percent_lcl_hitm_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
968 struct hist_entry *he)
970 int width = c2c_width(fmt, hpp, he->hists);
971 double per = PERCENT(he, lcl_hitm);
974 return scnprintf(hpp->buf, hpp->size, "%*s", width, PERC_STR(buf, per));
978 percent_lcl_hitm_color(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
979 struct hist_entry *he)
981 return percent_color(fmt, hpp, he, percent_lcl_hitm);
985 percent_lcl_hitm_cmp(struct perf_hpp_fmt *fmt __maybe_unused,
986 struct hist_entry *left, struct hist_entry *right)
991 per_left = PERCENT(left, lcl_hitm);
992 per_right = PERCENT(right, lcl_hitm);
994 return per_left - per_right;
998 percent_stores_l1hit_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
999 struct hist_entry *he)
1001 int width = c2c_width(fmt, hpp, he->hists);
1002 double per = PERCENT(he, st_l1hit);
1005 return scnprintf(hpp->buf, hpp->size, "%*s", width, PERC_STR(buf, per));
1009 percent_stores_l1hit_color(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
1010 struct hist_entry *he)
1012 return percent_color(fmt, hpp, he, percent_st_l1hit);
1016 percent_stores_l1hit_cmp(struct perf_hpp_fmt *fmt __maybe_unused,
1017 struct hist_entry *left, struct hist_entry *right)
1022 per_left = PERCENT(left, st_l1hit);
1023 per_right = PERCENT(right, st_l1hit);
1025 return per_left - per_right;
1029 percent_stores_l1miss_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
1030 struct hist_entry *he)
1032 int width = c2c_width(fmt, hpp, he->hists);
1033 double per = PERCENT(he, st_l1miss);
1036 return scnprintf(hpp->buf, hpp->size, "%*s", width, PERC_STR(buf, per));
1040 percent_stores_l1miss_color(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
1041 struct hist_entry *he)
1043 return percent_color(fmt, hpp, he, percent_st_l1miss);
1047 percent_stores_l1miss_cmp(struct perf_hpp_fmt *fmt __maybe_unused,
1048 struct hist_entry *left, struct hist_entry *right)
1053 per_left = PERCENT(left, st_l1miss);
1054 per_right = PERCENT(right, st_l1miss);
1056 return per_left - per_right;
1063 pid_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
1064 struct hist_entry *he)
1066 int width = c2c_width(fmt, hpp, he->hists);
1068 return scnprintf(hpp->buf, hpp->size, "%*d", width, he->thread->pid_);
1072 pid_cmp(struct perf_hpp_fmt *fmt __maybe_unused,
1073 struct hist_entry *left, struct hist_entry *right)
1075 return left->thread->pid_ - right->thread->pid_;
1079 empty_cmp(struct perf_hpp_fmt *fmt __maybe_unused,
1080 struct hist_entry *left __maybe_unused,
1081 struct hist_entry *right __maybe_unused)
1087 node_entry(struct perf_hpp_fmt *fmt __maybe_unused, struct perf_hpp *hpp,
1088 struct hist_entry *he)
1090 struct c2c_hist_entry *c2c_he;
1095 c2c_he = container_of(he, struct c2c_hist_entry, he);
1097 for (node = 0; node < c2c.nodes_cnt; node++) {
1098 DECLARE_BITMAP(set, c2c.cpus_cnt);
1100 bitmap_zero(set, c2c.cpus_cnt);
1101 bitmap_and(set, c2c_he->cpuset, c2c.nodes[node], c2c.cpus_cnt);
1103 if (!bitmap_weight(set, c2c.cpus_cnt)) {
1104 if (c2c.node_info == 1) {
1105 ret = scnprintf(hpp->buf, hpp->size, "%21s", " ");
1106 advance_hpp(hpp, ret);
1112 ret = scnprintf(hpp->buf, hpp->size, " ");
1113 advance_hpp(hpp, ret);
1116 switch (c2c.node_info) {
1118 ret = scnprintf(hpp->buf, hpp->size, "%2d", node);
1119 advance_hpp(hpp, ret);
1123 int num = bitmap_weight(set, c2c.cpus_cnt);
1124 struct c2c_stats *stats = &c2c_he->node_stats[node];
1126 ret = scnprintf(hpp->buf, hpp->size, "%2d{%2d ", node, num);
1127 advance_hpp(hpp, ret);
1129 #define DISPLAY_HITM(__h) \
1130 if (c2c_he->stats.__h> 0) { \
1131 ret = scnprintf(hpp->buf, hpp->size, "%5.1f%% ", \
1132 percent(stats->__h, c2c_he->stats.__h));\
1134 ret = scnprintf(hpp->buf, hpp->size, "%6s ", "n/a"); \
1137 switch (c2c.display) {
1139 DISPLAY_HITM(rmt_hitm);
1142 DISPLAY_HITM(lcl_hitm);
1145 DISPLAY_HITM(tot_hitm);
1152 advance_hpp(hpp, ret);
1154 if (c2c_he->stats.store > 0) {
1155 ret = scnprintf(hpp->buf, hpp->size, "%5.1f%%}",
1156 percent(stats->store, c2c_he->stats.store));
1158 ret = scnprintf(hpp->buf, hpp->size, "%6s}", "n/a");
1161 advance_hpp(hpp, ret);
1165 ret = scnprintf(hpp->buf, hpp->size, "%2d{", node);
1166 advance_hpp(hpp, ret);
1168 ret = bitmap_scnprintf(set, c2c.cpus_cnt, hpp->buf, hpp->size);
1169 advance_hpp(hpp, ret);
1171 ret = scnprintf(hpp->buf, hpp->size, "}");
1172 advance_hpp(hpp, ret);
1185 mean_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
1186 struct hist_entry *he, double mean)
1188 int width = c2c_width(fmt, hpp, he->hists);
1191 scnprintf(buf, 10, "%6.0f", mean);
1192 return scnprintf(hpp->buf, hpp->size, "%*s", width, buf);
1195 #define MEAN_ENTRY(__func, __val) \
1197 __func(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp, struct hist_entry *he) \
1199 struct c2c_hist_entry *c2c_he; \
1200 c2c_he = container_of(he, struct c2c_hist_entry, he); \
1201 return mean_entry(fmt, hpp, he, avg_stats(&c2c_he->cstats.__val)); \
1204 MEAN_ENTRY(mean_rmt_entry, rmt_hitm);
1205 MEAN_ENTRY(mean_lcl_entry, lcl_hitm);
1206 MEAN_ENTRY(mean_load_entry, load);
1209 cpucnt_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
1210 struct hist_entry *he)
1212 struct c2c_hist_entry *c2c_he;
1213 int width = c2c_width(fmt, hpp, he->hists);
1216 c2c_he = container_of(he, struct c2c_hist_entry, he);
1218 scnprintf(buf, 10, "%d", bitmap_weight(c2c_he->cpuset, c2c.cpus_cnt));
1219 return scnprintf(hpp->buf, hpp->size, "%*s", width, buf);
1223 cl_idx_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
1224 struct hist_entry *he)
1226 struct c2c_hist_entry *c2c_he;
1227 int width = c2c_width(fmt, hpp, he->hists);
1230 c2c_he = container_of(he, struct c2c_hist_entry, he);
1232 scnprintf(buf, 10, "%u", c2c_he->cacheline_idx);
1233 return scnprintf(hpp->buf, hpp->size, "%*s", width, buf);
1237 cl_idx_empty_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
1238 struct hist_entry *he)
1240 int width = c2c_width(fmt, hpp, he->hists);
1242 return scnprintf(hpp->buf, hpp->size, "%*s", width, "");
1245 #define HEADER_LOW(__h) \
1252 #define HEADER_BOTH(__h0, __h1) \
1262 #define HEADER_SPAN(__h0, __h1, __s) \
1273 #define HEADER_SPAN_LOW(__h) \
1280 static struct c2c_dimension dim_dcacheline = {
1281 .header = HEADER_SPAN("--- Cacheline ----", "Address", 2),
1282 .name = "dcacheline",
1283 .cmp = dcacheline_cmp,
1284 .entry = dcacheline_entry,
1288 static struct c2c_dimension dim_dcacheline_node = {
1289 .header = HEADER_LOW("Node"),
1290 .name = "dcacheline_node",
1292 .entry = dcacheline_node_entry,
1296 static struct c2c_dimension dim_dcacheline_count = {
1297 .header = HEADER_LOW("PA cnt"),
1298 .name = "dcacheline_count",
1300 .entry = dcacheline_node_count,
1304 static struct c2c_header header_offset_tui = HEADER_SPAN("-----", "Off", 2);
1306 static struct c2c_dimension dim_offset = {
1307 .header = HEADER_SPAN("--- Data address -", "Offset", 2),
1310 .entry = offset_entry,
1314 static struct c2c_dimension dim_offset_node = {
1315 .header = HEADER_LOW("Node"),
1316 .name = "offset_node",
1318 .entry = dcacheline_node_entry,
1322 static struct c2c_dimension dim_iaddr = {
1323 .header = HEADER_LOW("Code address"),
1326 .entry = iaddr_entry,
1330 static struct c2c_dimension dim_tot_hitm = {
1331 .header = HEADER_SPAN("----- LLC Load Hitm -----", "Total", 2),
1333 .cmp = tot_hitm_cmp,
1334 .entry = tot_hitm_entry,
1338 static struct c2c_dimension dim_lcl_hitm = {
1339 .header = HEADER_SPAN_LOW("Lcl"),
1341 .cmp = lcl_hitm_cmp,
1342 .entry = lcl_hitm_entry,
1346 static struct c2c_dimension dim_rmt_hitm = {
1347 .header = HEADER_SPAN_LOW("Rmt"),
1349 .cmp = rmt_hitm_cmp,
1350 .entry = rmt_hitm_entry,
1354 static struct c2c_dimension dim_cl_rmt_hitm = {
1355 .header = HEADER_SPAN("----- HITM -----", "Rmt", 1),
1356 .name = "cl_rmt_hitm",
1357 .cmp = rmt_hitm_cmp,
1358 .entry = rmt_hitm_entry,
1362 static struct c2c_dimension dim_cl_lcl_hitm = {
1363 .header = HEADER_SPAN_LOW("Lcl"),
1364 .name = "cl_lcl_hitm",
1365 .cmp = lcl_hitm_cmp,
1366 .entry = lcl_hitm_entry,
1370 static struct c2c_dimension dim_stores = {
1371 .header = HEADER_SPAN("---- Store Reference ----", "Total", 2),
1374 .entry = store_entry,
1378 static struct c2c_dimension dim_stores_l1hit = {
1379 .header = HEADER_SPAN_LOW("L1Hit"),
1380 .name = "stores_l1hit",
1381 .cmp = st_l1hit_cmp,
1382 .entry = st_l1hit_entry,
1386 static struct c2c_dimension dim_stores_l1miss = {
1387 .header = HEADER_SPAN_LOW("L1Miss"),
1388 .name = "stores_l1miss",
1389 .cmp = st_l1miss_cmp,
1390 .entry = st_l1miss_entry,
1394 static struct c2c_dimension dim_cl_stores_l1hit = {
1395 .header = HEADER_SPAN("-- Store Refs --", "L1 Hit", 1),
1396 .name = "cl_stores_l1hit",
1397 .cmp = st_l1hit_cmp,
1398 .entry = st_l1hit_entry,
1402 static struct c2c_dimension dim_cl_stores_l1miss = {
1403 .header = HEADER_SPAN_LOW("L1 Miss"),
1404 .name = "cl_stores_l1miss",
1405 .cmp = st_l1miss_cmp,
1406 .entry = st_l1miss_entry,
1410 static struct c2c_dimension dim_ld_fbhit = {
1411 .header = HEADER_SPAN("----- Core Load Hit -----", "FB", 2),
1413 .cmp = ld_fbhit_cmp,
1414 .entry = ld_fbhit_entry,
1418 static struct c2c_dimension dim_ld_l1hit = {
1419 .header = HEADER_SPAN_LOW("L1"),
1421 .cmp = ld_l1hit_cmp,
1422 .entry = ld_l1hit_entry,
1426 static struct c2c_dimension dim_ld_l2hit = {
1427 .header = HEADER_SPAN_LOW("L2"),
1429 .cmp = ld_l2hit_cmp,
1430 .entry = ld_l2hit_entry,
1434 static struct c2c_dimension dim_ld_llchit = {
1435 .header = HEADER_SPAN("-- LLC Load Hit --", "Llc", 1),
1436 .name = "ld_lclhit",
1437 .cmp = ld_llchit_cmp,
1438 .entry = ld_llchit_entry,
1442 static struct c2c_dimension dim_ld_rmthit = {
1443 .header = HEADER_SPAN_LOW("Rmt"),
1444 .name = "ld_rmthit",
1446 .entry = rmt_hit_entry,
1450 static struct c2c_dimension dim_ld_llcmiss = {
1451 .header = HEADER_BOTH("LLC", "Ld Miss"),
1452 .name = "ld_llcmiss",
1453 .cmp = ld_llcmiss_cmp,
1454 .entry = ld_llcmiss_entry,
1458 static struct c2c_dimension dim_tot_recs = {
1459 .header = HEADER_BOTH("Total", "records"),
1461 .cmp = tot_recs_cmp,
1462 .entry = tot_recs_entry,
1466 static struct c2c_dimension dim_tot_loads = {
1467 .header = HEADER_BOTH("Total", "Loads"),
1468 .name = "tot_loads",
1469 .cmp = tot_loads_cmp,
1470 .entry = tot_loads_entry,
1474 static struct c2c_header percent_hitm_header[] = {
1475 [DISPLAY_LCL] = HEADER_BOTH("Lcl", "Hitm"),
1476 [DISPLAY_RMT] = HEADER_BOTH("Rmt", "Hitm"),
1477 [DISPLAY_TOT] = HEADER_BOTH("Tot", "Hitm"),
1480 static struct c2c_dimension dim_percent_hitm = {
1481 .name = "percent_hitm",
1482 .cmp = percent_hitm_cmp,
1483 .entry = percent_hitm_entry,
1484 .color = percent_hitm_color,
1488 static struct c2c_dimension dim_percent_rmt_hitm = {
1489 .header = HEADER_SPAN("----- HITM -----", "Rmt", 1),
1490 .name = "percent_rmt_hitm",
1491 .cmp = percent_rmt_hitm_cmp,
1492 .entry = percent_rmt_hitm_entry,
1493 .color = percent_rmt_hitm_color,
1497 static struct c2c_dimension dim_percent_lcl_hitm = {
1498 .header = HEADER_SPAN_LOW("Lcl"),
1499 .name = "percent_lcl_hitm",
1500 .cmp = percent_lcl_hitm_cmp,
1501 .entry = percent_lcl_hitm_entry,
1502 .color = percent_lcl_hitm_color,
1506 static struct c2c_dimension dim_percent_stores_l1hit = {
1507 .header = HEADER_SPAN("-- Store Refs --", "L1 Hit", 1),
1508 .name = "percent_stores_l1hit",
1509 .cmp = percent_stores_l1hit_cmp,
1510 .entry = percent_stores_l1hit_entry,
1511 .color = percent_stores_l1hit_color,
1515 static struct c2c_dimension dim_percent_stores_l1miss = {
1516 .header = HEADER_SPAN_LOW("L1 Miss"),
1517 .name = "percent_stores_l1miss",
1518 .cmp = percent_stores_l1miss_cmp,
1519 .entry = percent_stores_l1miss_entry,
1520 .color = percent_stores_l1miss_color,
1524 static struct c2c_dimension dim_dram_lcl = {
1525 .header = HEADER_SPAN("--- Load Dram ----", "Lcl", 1),
1527 .cmp = lcl_dram_cmp,
1528 .entry = lcl_dram_entry,
1532 static struct c2c_dimension dim_dram_rmt = {
1533 .header = HEADER_SPAN_LOW("Rmt"),
1535 .cmp = rmt_dram_cmp,
1536 .entry = rmt_dram_entry,
1540 static struct c2c_dimension dim_pid = {
1541 .header = HEADER_LOW("Pid"),
1548 static struct c2c_dimension dim_tid = {
1549 .header = HEADER_LOW("Tid"),
1554 static struct c2c_dimension dim_symbol = {
1559 static struct c2c_dimension dim_dso = {
1560 .header = HEADER_BOTH("Shared", "Object"),
1565 static struct c2c_header header_node[3] = {
1567 HEADER_LOW("Node{cpus %hitms %stores}"),
1568 HEADER_LOW("Node{cpu list}"),
1571 static struct c2c_dimension dim_node = {
1574 .entry = node_entry,
1578 static struct c2c_dimension dim_mean_rmt = {
1579 .header = HEADER_SPAN("---------- cycles ----------", "rmt hitm", 2),
1582 .entry = mean_rmt_entry,
1586 static struct c2c_dimension dim_mean_lcl = {
1587 .header = HEADER_SPAN_LOW("lcl hitm"),
1590 .entry = mean_lcl_entry,
1594 static struct c2c_dimension dim_mean_load = {
1595 .header = HEADER_SPAN_LOW("load"),
1596 .name = "mean_load",
1598 .entry = mean_load_entry,
1602 static struct c2c_dimension dim_cpucnt = {
1603 .header = HEADER_BOTH("cpu", "cnt"),
1606 .entry = cpucnt_entry,
1610 static struct c2c_dimension dim_srcline = {
1611 .name = "cl_srcline",
1612 .se = &sort_srcline,
1615 static struct c2c_dimension dim_dcacheline_idx = {
1616 .header = HEADER_LOW("Index"),
1619 .entry = cl_idx_entry,
1623 static struct c2c_dimension dim_dcacheline_num = {
1624 .header = HEADER_LOW("Num"),
1627 .entry = cl_idx_entry,
1631 static struct c2c_dimension dim_dcacheline_num_empty = {
1632 .header = HEADER_LOW("Num"),
1633 .name = "cl_num_empty",
1635 .entry = cl_idx_empty_entry,
1639 static struct c2c_dimension *dimensions[] = {
1641 &dim_dcacheline_node,
1642 &dim_dcacheline_count,
1654 &dim_cl_stores_l1hit,
1655 &dim_cl_stores_l1miss,
1665 &dim_percent_rmt_hitm,
1666 &dim_percent_lcl_hitm,
1667 &dim_percent_stores_l1hit,
1668 &dim_percent_stores_l1miss,
1681 &dim_dcacheline_idx,
1682 &dim_dcacheline_num,
1683 &dim_dcacheline_num_empty,
1687 static void fmt_free(struct perf_hpp_fmt *fmt)
1689 struct c2c_fmt *c2c_fmt;
1691 c2c_fmt = container_of(fmt, struct c2c_fmt, fmt);
1695 static bool fmt_equal(struct perf_hpp_fmt *a, struct perf_hpp_fmt *b)
1697 struct c2c_fmt *c2c_a = container_of(a, struct c2c_fmt, fmt);
1698 struct c2c_fmt *c2c_b = container_of(b, struct c2c_fmt, fmt);
1700 return c2c_a->dim == c2c_b->dim;
1703 static struct c2c_dimension *get_dimension(const char *name)
1707 for (i = 0; dimensions[i]; i++) {
1708 struct c2c_dimension *dim = dimensions[i];
1710 if (!strcmp(dim->name, name))
1717 static int c2c_se_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
1718 struct hist_entry *he)
1720 struct c2c_fmt *c2c_fmt = container_of(fmt, struct c2c_fmt, fmt);
1721 struct c2c_dimension *dim = c2c_fmt->dim;
1722 size_t len = fmt->user_len;
1725 len = hists__col_len(he->hists, dim->se->se_width_idx);
1727 if (dim == &dim_symbol || dim == &dim_srcline)
1728 len = symbol_width(he->hists, dim->se);
1731 return dim->se->se_snprintf(he, hpp->buf, hpp->size, len);
1734 static int64_t c2c_se_cmp(struct perf_hpp_fmt *fmt,
1735 struct hist_entry *a, struct hist_entry *b)
1737 struct c2c_fmt *c2c_fmt = container_of(fmt, struct c2c_fmt, fmt);
1738 struct c2c_dimension *dim = c2c_fmt->dim;
1740 return dim->se->se_cmp(a, b);
1743 static int64_t c2c_se_collapse(struct perf_hpp_fmt *fmt,
1744 struct hist_entry *a, struct hist_entry *b)
1746 struct c2c_fmt *c2c_fmt = container_of(fmt, struct c2c_fmt, fmt);
1747 struct c2c_dimension *dim = c2c_fmt->dim;
1748 int64_t (*collapse_fn)(struct hist_entry *, struct hist_entry *);
1750 collapse_fn = dim->se->se_collapse ?: dim->se->se_cmp;
1751 return collapse_fn(a, b);
1754 static struct c2c_fmt *get_format(const char *name)
1756 struct c2c_dimension *dim = get_dimension(name);
1757 struct c2c_fmt *c2c_fmt;
1758 struct perf_hpp_fmt *fmt;
1763 c2c_fmt = zalloc(sizeof(*c2c_fmt));
1769 fmt = &c2c_fmt->fmt;
1770 INIT_LIST_HEAD(&fmt->list);
1771 INIT_LIST_HEAD(&fmt->sort_list);
1773 fmt->cmp = dim->se ? c2c_se_cmp : dim->cmp;
1774 fmt->sort = dim->se ? c2c_se_cmp : dim->cmp;
1775 fmt->color = dim->se ? NULL : dim->color;
1776 fmt->entry = dim->se ? c2c_se_entry : dim->entry;
1777 fmt->header = c2c_header;
1778 fmt->width = c2c_width;
1779 fmt->collapse = dim->se ? c2c_se_collapse : dim->cmp;
1780 fmt->equal = fmt_equal;
1781 fmt->free = fmt_free;
1786 static int c2c_hists__init_output(struct perf_hpp_list *hpp_list, char *name)
1788 struct c2c_fmt *c2c_fmt = get_format(name);
1792 return output_field_add(hpp_list, name);
1795 perf_hpp_list__column_register(hpp_list, &c2c_fmt->fmt);
1799 static int c2c_hists__init_sort(struct perf_hpp_list *hpp_list, char *name)
1801 struct c2c_fmt *c2c_fmt = get_format(name);
1802 struct c2c_dimension *dim;
1806 return sort_dimension__add(hpp_list, name, NULL, 0);
1810 if (dim == &dim_dso)
1813 perf_hpp_list__register_sort_field(hpp_list, &c2c_fmt->fmt);
1817 #define PARSE_LIST(_list, _fn) \
1825 for (tok = strtok_r((char *)_list, ", ", &tmp); \
1826 tok; tok = strtok_r(NULL, ", ", &tmp)) { \
1827 ret = _fn(hpp_list, tok); \
1828 if (ret == -EINVAL) { \
1829 pr_err("Invalid --fields key: `%s'", tok); \
1831 } else if (ret == -ESRCH) { \
1832 pr_err("Unknown --fields key: `%s'", tok); \
1838 static int hpp_list__parse(struct perf_hpp_list *hpp_list,
1839 const char *output_,
1842 char *output = output_ ? strdup(output_) : NULL;
1843 char *sort = sort_ ? strdup(sort_) : NULL;
1846 PARSE_LIST(output, c2c_hists__init_output);
1847 PARSE_LIST(sort, c2c_hists__init_sort);
1849 /* copy sort keys to output fields */
1850 perf_hpp__setup_output_field(hpp_list);
1853 * We dont need other sorting keys other than those
1854 * we already specified. It also really slows down
1855 * the processing a lot with big number of output
1856 * fields, so switching this off for c2c.
1860 /* and then copy output fields to sort keys */
1861 perf_hpp__append_sort_keys(&hists->list);
1869 static int c2c_hists__init(struct c2c_hists *hists,
1871 int nr_header_lines)
1873 __hists__init(&hists->hists, &hists->list);
1876 * Initialize only with sort fields, we need to resort
1877 * later anyway, and that's where we add output fields
1880 perf_hpp_list__init(&hists->list);
1882 /* Overload number of header lines.*/
1883 hists->list.nr_header_lines = nr_header_lines;
1885 return hpp_list__parse(&hists->list, NULL, sort);
1888 static int c2c_hists__reinit(struct c2c_hists *c2c_hists,
1892 perf_hpp__reset_output_field(&c2c_hists->list);
1893 return hpp_list__parse(&c2c_hists->list, output, sort);
1896 #define DISPLAY_LINE_LIMIT 0.001
1898 static bool he__display(struct hist_entry *he, struct c2c_stats *stats)
1900 struct c2c_hist_entry *c2c_he;
1906 c2c_he = container_of(he, struct c2c_hist_entry, he);
1908 #define FILTER_HITM(__h) \
1910 ld_dist = ((double)c2c_he->stats.__h / stats->__h); \
1911 if (ld_dist < DISPLAY_LINE_LIMIT) \
1912 he->filtered = HIST_FILTER__C2C; \
1914 he->filtered = HIST_FILTER__C2C; \
1917 switch (c2c.display) {
1919 FILTER_HITM(lcl_hitm);
1922 FILTER_HITM(rmt_hitm);
1925 FILTER_HITM(tot_hitm);
1932 return he->filtered == 0;
1935 static inline int valid_hitm_or_store(struct hist_entry *he)
1937 struct c2c_hist_entry *c2c_he;
1940 c2c_he = container_of(he, struct c2c_hist_entry, he);
1941 has_hitm = c2c.display == DISPLAY_TOT ? c2c_he->stats.tot_hitm :
1942 c2c.display == DISPLAY_LCL ? c2c_he->stats.lcl_hitm :
1943 c2c_he->stats.rmt_hitm;
1944 return has_hitm || c2c_he->stats.store;
1947 static void set_node_width(struct c2c_hist_entry *c2c_he, int len)
1949 struct c2c_dimension *dim;
1951 dim = &c2c.hists == c2c_he->hists ?
1952 &dim_dcacheline_node : &dim_offset_node;
1954 if (len > dim->width)
1958 static int set_nodestr(struct c2c_hist_entry *c2c_he)
1963 if (c2c_he->nodestr)
1966 if (bitmap_weight(c2c_he->nodeset, c2c.nodes_cnt)) {
1967 len = bitmap_scnprintf(c2c_he->nodeset, c2c.nodes_cnt,
1970 len = scnprintf(buf, sizeof(buf), "N/A");
1973 set_node_width(c2c_he, len);
1974 c2c_he->nodestr = strdup(buf);
1975 return c2c_he->nodestr ? 0 : -ENOMEM;
1978 static void calc_width(struct c2c_hist_entry *c2c_he)
1980 struct c2c_hists *c2c_hists;
1982 c2c_hists = container_of(c2c_he->he.hists, struct c2c_hists, hists);
1983 hists__calc_col_len(&c2c_hists->hists, &c2c_he->he);
1984 set_nodestr(c2c_he);
1987 static int filter_cb(struct hist_entry *he, void *arg __maybe_unused)
1989 struct c2c_hist_entry *c2c_he;
1991 c2c_he = container_of(he, struct c2c_hist_entry, he);
1993 if (c2c.show_src && !he->srcline)
1994 he->srcline = hist_entry__srcline(he);
1998 if (!valid_hitm_or_store(he))
1999 he->filtered = HIST_FILTER__C2C;
2004 static int resort_cl_cb(struct hist_entry *he, void *arg __maybe_unused)
2006 struct c2c_hist_entry *c2c_he;
2007 struct c2c_hists *c2c_hists;
2008 bool display = he__display(he, &c2c.hitm_stats);
2010 c2c_he = container_of(he, struct c2c_hist_entry, he);
2011 c2c_hists = c2c_he->hists;
2013 if (display && c2c_hists) {
2014 static unsigned int idx;
2016 c2c_he->cacheline_idx = idx++;
2019 c2c_hists__reinit(c2c_hists, c2c.cl_output, c2c.cl_resort);
2021 hists__collapse_resort(&c2c_hists->hists, NULL);
2022 hists__output_resort_cb(&c2c_hists->hists, NULL, filter_cb);
2028 static void setup_nodes_header(void)
2030 dim_node.header = header_node[c2c.node_info];
2033 static int setup_nodes(struct perf_session *session)
2035 struct numa_node *n;
2036 unsigned long **nodes;
2040 if (c2c.node_info > 2)
2043 c2c.nodes_cnt = session->header.env.nr_numa_nodes;
2044 c2c.cpus_cnt = session->header.env.nr_cpus_avail;
2046 n = session->header.env.numa_nodes;
2050 nodes = zalloc(sizeof(unsigned long *) * c2c.nodes_cnt);
2056 cpu2node = zalloc(sizeof(int) * c2c.cpus_cnt);
2060 for (cpu = 0; cpu < c2c.cpus_cnt; cpu++)
2063 c2c.cpu2node = cpu2node;
2065 for (node = 0; node < c2c.nodes_cnt; node++) {
2066 struct perf_cpu_map *map = n[node].map;
2069 set = bitmap_alloc(c2c.cpus_cnt);
2075 /* empty node, skip */
2076 if (perf_cpu_map__empty(map))
2079 for (cpu = 0; cpu < map->nr; cpu++) {
2080 set_bit(map->map[cpu], set);
2082 if (WARN_ONCE(cpu2node[map->map[cpu]] != -1, "node/cpu topology bug"))
2085 cpu2node[map->map[cpu]] = node;
2089 setup_nodes_header();
2093 #define HAS_HITMS(__h) ((__h)->stats.lcl_hitm || (__h)->stats.rmt_hitm)
2095 static int resort_hitm_cb(struct hist_entry *he, void *arg __maybe_unused)
2097 struct c2c_hist_entry *c2c_he;
2098 c2c_he = container_of(he, struct c2c_hist_entry, he);
2100 if (HAS_HITMS(c2c_he)) {
2101 c2c.shared_clines++;
2102 c2c_add_stats(&c2c.hitm_stats, &c2c_he->stats);
2108 static int hists__iterate_cb(struct hists *hists, hists__resort_cb_t cb)
2110 struct rb_node *next = rb_first_cached(&hists->entries);
2114 struct hist_entry *he;
2116 he = rb_entry(next, struct hist_entry, rb_node);
2120 next = rb_next(&he->rb_node);
2126 static void print_c2c__display_stats(FILE *out)
2129 struct c2c_stats *stats = &c2c.hists.stats;
2131 llc_misses = stats->lcl_dram +
2136 fprintf(out, "=================================================\n");
2137 fprintf(out, " Trace Event Information \n");
2138 fprintf(out, "=================================================\n");
2139 fprintf(out, " Total records : %10d\n", stats->nr_entries);
2140 fprintf(out, " Locked Load/Store Operations : %10d\n", stats->locks);
2141 fprintf(out, " Load Operations : %10d\n", stats->load);
2142 fprintf(out, " Loads - uncacheable : %10d\n", stats->ld_uncache);
2143 fprintf(out, " Loads - IO : %10d\n", stats->ld_io);
2144 fprintf(out, " Loads - Miss : %10d\n", stats->ld_miss);
2145 fprintf(out, " Loads - no mapping : %10d\n", stats->ld_noadrs);
2146 fprintf(out, " Load Fill Buffer Hit : %10d\n", stats->ld_fbhit);
2147 fprintf(out, " Load L1D hit : %10d\n", stats->ld_l1hit);
2148 fprintf(out, " Load L2D hit : %10d\n", stats->ld_l2hit);
2149 fprintf(out, " Load LLC hit : %10d\n", stats->ld_llchit + stats->lcl_hitm);
2150 fprintf(out, " Load Local HITM : %10d\n", stats->lcl_hitm);
2151 fprintf(out, " Load Remote HITM : %10d\n", stats->rmt_hitm);
2152 fprintf(out, " Load Remote HIT : %10d\n", stats->rmt_hit);
2153 fprintf(out, " Load Local DRAM : %10d\n", stats->lcl_dram);
2154 fprintf(out, " Load Remote DRAM : %10d\n", stats->rmt_dram);
2155 fprintf(out, " Load MESI State Exclusive : %10d\n", stats->ld_excl);
2156 fprintf(out, " Load MESI State Shared : %10d\n", stats->ld_shared);
2157 fprintf(out, " Load LLC Misses : %10d\n", llc_misses);
2158 fprintf(out, " LLC Misses to Local DRAM : %10.1f%%\n", ((double)stats->lcl_dram/(double)llc_misses) * 100.);
2159 fprintf(out, " LLC Misses to Remote DRAM : %10.1f%%\n", ((double)stats->rmt_dram/(double)llc_misses) * 100.);
2160 fprintf(out, " LLC Misses to Remote cache (HIT) : %10.1f%%\n", ((double)stats->rmt_hit /(double)llc_misses) * 100.);
2161 fprintf(out, " LLC Misses to Remote cache (HITM) : %10.1f%%\n", ((double)stats->rmt_hitm/(double)llc_misses) * 100.);
2162 fprintf(out, " Store Operations : %10d\n", stats->store);
2163 fprintf(out, " Store - uncacheable : %10d\n", stats->st_uncache);
2164 fprintf(out, " Store - no mapping : %10d\n", stats->st_noadrs);
2165 fprintf(out, " Store L1D Hit : %10d\n", stats->st_l1hit);
2166 fprintf(out, " Store L1D Miss : %10d\n", stats->st_l1miss);
2167 fprintf(out, " No Page Map Rejects : %10d\n", stats->nomap);
2168 fprintf(out, " Unable to parse data source : %10d\n", stats->noparse);
2171 static void print_shared_cacheline_info(FILE *out)
2173 struct c2c_stats *stats = &c2c.hitm_stats;
2174 int hitm_cnt = stats->lcl_hitm + stats->rmt_hitm;
2176 fprintf(out, "=================================================\n");
2177 fprintf(out, " Global Shared Cache Line Event Information \n");
2178 fprintf(out, "=================================================\n");
2179 fprintf(out, " Total Shared Cache Lines : %10d\n", c2c.shared_clines);
2180 fprintf(out, " Load HITs on shared lines : %10d\n", stats->load);
2181 fprintf(out, " Fill Buffer Hits on shared lines : %10d\n", stats->ld_fbhit);
2182 fprintf(out, " L1D hits on shared lines : %10d\n", stats->ld_l1hit);
2183 fprintf(out, " L2D hits on shared lines : %10d\n", stats->ld_l2hit);
2184 fprintf(out, " LLC hits on shared lines : %10d\n", stats->ld_llchit + stats->lcl_hitm);
2185 fprintf(out, " Locked Access on shared lines : %10d\n", stats->locks);
2186 fprintf(out, " Store HITs on shared lines : %10d\n", stats->store);
2187 fprintf(out, " Store L1D hits on shared lines : %10d\n", stats->st_l1hit);
2188 fprintf(out, " Total Merged records : %10d\n", hitm_cnt + stats->store);
2191 static void print_cacheline(struct c2c_hists *c2c_hists,
2192 struct hist_entry *he_cl,
2193 struct perf_hpp_list *hpp_list,
2197 struct perf_hpp hpp = {
2204 hists__fprintf_headers(&c2c_hists->hists, out);
2210 fprintf(out, " -------------------------------------------------------------\n");
2211 __hist_entry__snprintf(he_cl, &hpp, hpp_list);
2212 fprintf(out, "%s\n", bf);
2213 fprintf(out, " -------------------------------------------------------------\n");
2215 hists__fprintf(&c2c_hists->hists, false, 0, 0, 0, out, false);
2218 static void print_pareto(FILE *out)
2220 struct perf_hpp_list hpp_list;
2224 perf_hpp_list__init(&hpp_list);
2225 ret = hpp_list__parse(&hpp_list,
2234 if (WARN_ONCE(ret, "failed to setup sort entries\n"))
2237 nd = rb_first_cached(&c2c.hists.hists.entries);
2239 for (; nd; nd = rb_next(nd)) {
2240 struct hist_entry *he = rb_entry(nd, struct hist_entry, rb_node);
2241 struct c2c_hist_entry *c2c_he;
2246 c2c_he = container_of(he, struct c2c_hist_entry, he);
2247 print_cacheline(c2c_he->hists, he, &hpp_list, out);
2251 static void print_c2c_info(FILE *out, struct perf_session *session)
2253 struct evlist *evlist = session->evlist;
2254 struct evsel *evsel;
2257 fprintf(out, "=================================================\n");
2258 fprintf(out, " c2c details \n");
2259 fprintf(out, "=================================================\n");
2261 evlist__for_each_entry(evlist, evsel) {
2262 fprintf(out, "%-36s: %s\n", first ? " Events" : "", evsel__name(evsel));
2265 fprintf(out, " Cachelines sort on : %s HITMs\n",
2266 display_str[c2c.display]);
2267 fprintf(out, " Cacheline data grouping : %s\n", c2c.cl_sort);
2270 static void perf_c2c__hists_fprintf(FILE *out, struct perf_session *session)
2274 print_c2c__display_stats(out);
2276 print_shared_cacheline_info(out);
2278 print_c2c_info(out, session);
2284 fprintf(out, "=================================================\n");
2285 fprintf(out, " Shared Data Cache Line Table \n");
2286 fprintf(out, "=================================================\n");
2287 fprintf(out, "#\n");
2289 hists__fprintf(&c2c.hists.hists, true, 0, 0, 0, stdout, true);
2292 fprintf(out, "=================================================\n");
2293 fprintf(out, " Shared Cache Line Distribution Pareto \n");
2294 fprintf(out, "=================================================\n");
2295 fprintf(out, "#\n");
2300 #ifdef HAVE_SLANG_SUPPORT
2301 static void c2c_browser__update_nr_entries(struct hist_browser *hb)
2304 struct rb_node *nd = rb_first_cached(&hb->hists->entries);
2307 struct hist_entry *he = rb_entry(nd, struct hist_entry, rb_node);
2315 hb->nr_non_filtered_entries = nr_entries;
2318 struct c2c_cacheline_browser {
2319 struct hist_browser hb;
2320 struct hist_entry *he;
2324 perf_c2c_cacheline_browser__title(struct hist_browser *browser,
2325 char *bf, size_t size)
2327 struct c2c_cacheline_browser *cl_browser;
2328 struct hist_entry *he;
2331 cl_browser = container_of(browser, struct c2c_cacheline_browser, hb);
2332 he = cl_browser->he;
2335 addr = cl_address(he->mem_info->daddr.addr);
2337 scnprintf(bf, size, "Cacheline 0x%lx", addr);
2341 static struct c2c_cacheline_browser*
2342 c2c_cacheline_browser__new(struct hists *hists, struct hist_entry *he)
2344 struct c2c_cacheline_browser *browser;
2346 browser = zalloc(sizeof(*browser));
2348 hist_browser__init(&browser->hb, hists);
2349 browser->hb.c2c_filter = true;
2350 browser->hb.title = perf_c2c_cacheline_browser__title;
2357 static int perf_c2c__browse_cacheline(struct hist_entry *he)
2359 struct c2c_hist_entry *c2c_he;
2360 struct c2c_hists *c2c_hists;
2361 struct c2c_cacheline_browser *cl_browser;
2362 struct hist_browser *browser;
2364 static const char help[] =
2365 " ENTER Toggle callchains (if present) \n"
2366 " n Toggle Node details info \n"
2367 " s Toggle full length of symbol and source line columns \n"
2368 " q Return back to cacheline list \n";
2373 /* Display compact version first. */
2374 c2c.symbol_full = false;
2376 c2c_he = container_of(he, struct c2c_hist_entry, he);
2377 c2c_hists = c2c_he->hists;
2379 cl_browser = c2c_cacheline_browser__new(&c2c_hists->hists, he);
2380 if (cl_browser == NULL)
2383 browser = &cl_browser->hb;
2385 /* reset abort key so that it can get Ctrl-C as a key */
2387 SLang_init_tty(0, 0, 0);
2389 c2c_browser__update_nr_entries(browser);
2392 key = hist_browser__run(browser, "? - help", true, 0);
2396 c2c.symbol_full = !c2c.symbol_full;
2399 c2c.node_info = (c2c.node_info + 1) % 3;
2400 setup_nodes_header();
2405 ui_browser__help_window(&browser->b, help);
2417 static int perf_c2c_browser__title(struct hist_browser *browser,
2418 char *bf, size_t size)
2421 "Shared Data Cache Line Table "
2422 "(%lu entries, sorted on %s HITMs)",
2423 browser->nr_non_filtered_entries,
2424 display_str[c2c.display]);
2428 static struct hist_browser*
2429 perf_c2c_browser__new(struct hists *hists)
2431 struct hist_browser *browser = hist_browser__new(hists);
2434 browser->title = perf_c2c_browser__title;
2435 browser->c2c_filter = true;
2441 static int perf_c2c__hists_browse(struct hists *hists)
2443 struct hist_browser *browser;
2445 static const char help[] =
2446 " d Display cacheline details \n"
2447 " ENTER Toggle callchains (if present) \n"
2450 browser = perf_c2c_browser__new(hists);
2451 if (browser == NULL)
2454 /* reset abort key so that it can get Ctrl-C as a key */
2456 SLang_init_tty(0, 0, 0);
2458 c2c_browser__update_nr_entries(browser);
2461 key = hist_browser__run(browser, "? - help", true, 0);
2467 perf_c2c__browse_cacheline(browser->he_selection);
2470 ui_browser__help_window(&browser->b, help);
2478 hist_browser__delete(browser);
2482 static void perf_c2c_display(struct perf_session *session)
2484 if (use_browser == 0)
2485 perf_c2c__hists_fprintf(stdout, session);
2487 perf_c2c__hists_browse(&c2c.hists.hists);
2490 static void perf_c2c_display(struct perf_session *session)
2493 perf_c2c__hists_fprintf(stdout, session);
2495 #endif /* HAVE_SLANG_SUPPORT */
2497 static char *fill_line(const char *orig, int len)
2499 int i, j, olen = strlen(orig);
2502 buf = zalloc(len + 1);
2506 j = len / 2 - olen / 2;
2508 for (i = 0; i < j - 1; i++)
2513 strcpy(buf + i, orig);
2519 for (; i < len; i++)
2525 static int ui_quirks(void)
2527 const char *nodestr = "Data address";
2530 if (!c2c.use_stdio) {
2531 dim_offset.width = 5;
2532 dim_offset.header = header_offset_tui;
2536 dim_percent_hitm.header = percent_hitm_header[c2c.display];
2538 /* Fix the zero line for dcacheline column. */
2539 buf = fill_line("Cacheline", dim_dcacheline.width +
2540 dim_dcacheline_node.width +
2541 dim_dcacheline_count.width + 4);
2545 dim_dcacheline.header.line[0].text = buf;
2547 /* Fix the zero line for offset column. */
2548 buf = fill_line(nodestr, dim_offset.width +
2549 dim_offset_node.width +
2550 dim_dcacheline_count.width + 4);
2554 dim_offset.header.line[0].text = buf;
2559 #define CALLCHAIN_DEFAULT_OPT "graph,0.5,caller,function,percent"
2561 const char callchain_help[] = "Display call graph (stack chain/backtrace):\n\n"
2562 CALLCHAIN_REPORT_HELP
2563 "\n\t\t\t\tDefault: " CALLCHAIN_DEFAULT_OPT;
2566 parse_callchain_opt(const struct option *opt, const char *arg, int unset)
2568 struct callchain_param *callchain = opt->value;
2570 callchain->enabled = !unset;
2575 symbol_conf.use_callchain = false;
2576 callchain->mode = CHAIN_NONE;
2580 return parse_callchain_report_opt(arg);
2583 static int setup_callchain(struct evlist *evlist)
2585 u64 sample_type = evlist__combined_sample_type(evlist);
2586 enum perf_call_graph_mode mode = CALLCHAIN_NONE;
2588 if ((sample_type & PERF_SAMPLE_REGS_USER) &&
2589 (sample_type & PERF_SAMPLE_STACK_USER)) {
2590 mode = CALLCHAIN_DWARF;
2591 dwarf_callchain_users = true;
2592 } else if (sample_type & PERF_SAMPLE_BRANCH_STACK)
2593 mode = CALLCHAIN_LBR;
2594 else if (sample_type & PERF_SAMPLE_CALLCHAIN)
2595 mode = CALLCHAIN_FP;
2597 if (!callchain_param.enabled &&
2598 callchain_param.mode != CHAIN_NONE &&
2599 mode != CALLCHAIN_NONE) {
2600 symbol_conf.use_callchain = true;
2601 if (callchain_register_param(&callchain_param) < 0) {
2602 ui__error("Can't register callchain params.\n");
2607 if (c2c.stitch_lbr && (mode != CALLCHAIN_LBR)) {
2608 ui__warning("Can't find LBR callchain. Switch off --stitch-lbr.\n"
2609 "Please apply --call-graph lbr when recording.\n");
2610 c2c.stitch_lbr = false;
2613 callchain_param.record_mode = mode;
2614 callchain_param.min_percent = 0;
2618 static int setup_display(const char *str)
2620 const char *display = str ?: "tot";
2622 if (!strcmp(display, "tot"))
2623 c2c.display = DISPLAY_TOT;
2624 else if (!strcmp(display, "rmt"))
2625 c2c.display = DISPLAY_RMT;
2626 else if (!strcmp(display, "lcl"))
2627 c2c.display = DISPLAY_LCL;
2629 pr_err("failed: unknown display type: %s\n", str);
2636 #define for_each_token(__tok, __buf, __sep, __tmp) \
2637 for (__tok = strtok_r(__buf, __sep, &__tmp); __tok; \
2638 __tok = strtok_r(NULL, __sep, &__tmp))
2640 static int build_cl_output(char *cl_sort, bool no_source)
2642 char *tok, *tmp, *buf = strdup(cl_sort);
2643 bool add_pid = false;
2644 bool add_tid = false;
2645 bool add_iaddr = false;
2646 bool add_sym = false;
2647 bool add_dso = false;
2648 bool add_src = false;
2654 for_each_token(tok, buf, ",", tmp) {
2655 if (!strcmp(tok, "tid")) {
2657 } else if (!strcmp(tok, "pid")) {
2659 } else if (!strcmp(tok, "iaddr")) {
2663 add_src = no_source ? false : true;
2664 } else if (!strcmp(tok, "dso")) {
2666 } else if (strcmp(tok, "offset")) {
2667 pr_err("unrecognized sort token: %s\n", tok);
2673 if (asprintf(&c2c.cl_output,
2674 "%s%s%s%s%s%s%s%s%s%s",
2675 c2c.use_stdio ? "cl_num_empty," : "",
2678 "percent_stores_l1hit,"
2679 "percent_stores_l1miss,"
2680 "offset,offset_node,dcacheline_count,",
2681 add_pid ? "pid," : "",
2682 add_tid ? "tid," : "",
2683 add_iaddr ? "iaddr," : "",
2689 add_sym ? "symbol," : "",
2690 add_dso ? "dso," : "",
2691 add_src ? "cl_srcline," : "",
2697 c2c.show_src = add_src;
2703 static int setup_coalesce(const char *coalesce, bool no_source)
2705 const char *c = coalesce ?: coalesce_default;
2707 if (asprintf(&c2c.cl_sort, "offset,%s", c) < 0)
2710 if (build_cl_output(c2c.cl_sort, no_source))
2713 if (asprintf(&c2c.cl_resort, "offset,%s",
2714 c2c.display == DISPLAY_TOT ?
2716 c2c.display == DISPLAY_RMT ?
2717 "rmt_hitm,lcl_hitm" :
2718 "lcl_hitm,rmt_hitm") < 0)
2721 pr_debug("coalesce sort fields: %s\n", c2c.cl_sort);
2722 pr_debug("coalesce resort fields: %s\n", c2c.cl_resort);
2723 pr_debug("coalesce output fields: %s\n", c2c.cl_output);
2727 static int perf_c2c__report(int argc, const char **argv)
2729 struct perf_session *session;
2730 struct ui_progress prog;
2731 struct perf_data data = {
2732 .mode = PERF_DATA_MODE_READ,
2734 char callchain_default_opt[] = CALLCHAIN_DEFAULT_OPT;
2735 const char *display = NULL;
2736 const char *coalesce = NULL;
2737 bool no_source = false;
2738 const struct option options[] = {
2739 OPT_STRING('k', "vmlinux", &symbol_conf.vmlinux_name,
2740 "file", "vmlinux pathname"),
2741 OPT_STRING('i', "input", &input_name, "file",
2742 "the input file to process"),
2743 OPT_INCR('N', "node-info", &c2c.node_info,
2744 "show extra node info in report (repeat for more info)"),
2745 #ifdef HAVE_SLANG_SUPPORT
2746 OPT_BOOLEAN(0, "stdio", &c2c.use_stdio, "Use the stdio interface"),
2748 OPT_BOOLEAN(0, "stats", &c2c.stats_only,
2749 "Display only statistic tables (implies --stdio)"),
2750 OPT_BOOLEAN(0, "full-symbols", &c2c.symbol_full,
2751 "Display full length of symbols"),
2752 OPT_BOOLEAN(0, "no-source", &no_source,
2753 "Do not display Source Line column"),
2754 OPT_BOOLEAN(0, "show-all", &c2c.show_all,
2755 "Show all captured HITM lines."),
2756 OPT_CALLBACK_DEFAULT('g', "call-graph", &callchain_param,
2757 "print_type,threshold[,print_limit],order,sort_key[,branch],value",
2758 callchain_help, &parse_callchain_opt,
2759 callchain_default_opt),
2760 OPT_STRING('d', "display", &display, "Switch HITM output type", "lcl,rmt"),
2761 OPT_STRING('c', "coalesce", &coalesce, "coalesce fields",
2762 "coalesce fields: pid,tid,iaddr,dso"),
2763 OPT_BOOLEAN('f', "force", &symbol_conf.force, "don't complain, do it"),
2764 OPT_BOOLEAN(0, "stitch-lbr", &c2c.stitch_lbr,
2765 "Enable LBR callgraph stitching approach"),
2766 OPT_PARENT(c2c_options),
2771 argc = parse_options(argc, argv, options, report_c2c_usage,
2772 PARSE_OPT_STOP_AT_NON_OPTION);
2774 usage_with_options(report_c2c_usage, options);
2777 c2c.use_stdio = true;
2779 if (!input_name || !strlen(input_name))
2780 input_name = "perf.data";
2782 data.path = input_name;
2783 data.force = symbol_conf.force;
2785 err = setup_display(display);
2789 err = setup_coalesce(coalesce, no_source);
2791 pr_debug("Failed to initialize hists\n");
2795 err = c2c_hists__init(&c2c.hists, "dcacheline", 2);
2797 pr_debug("Failed to initialize hists\n");
2801 session = perf_session__new(&data, 0, &c2c.tool);
2802 if (IS_ERR(session)) {
2803 err = PTR_ERR(session);
2804 pr_debug("Error creating perf session\n");
2808 err = setup_nodes(session);
2810 pr_err("Failed setup nodes\n");
2814 err = mem2node__init(&c2c.mem2node, &session->header.env);
2818 err = setup_callchain(session->evlist);
2822 if (symbol__init(&session->header.env) < 0)
2825 /* No pipe support at the moment. */
2826 if (perf_data__is_pipe(session->data)) {
2827 pr_debug("No pipe support at the moment.\n");
2836 setup_browser(false);
2838 err = perf_session__process_events(session);
2840 pr_err("failed to process sample\n");
2844 c2c_hists__reinit(&c2c.hists,
2851 "tot_hitm,lcl_hitm,rmt_hitm,"
2852 "stores,stores_l1hit,stores_l1miss,"
2853 "dram_lcl,dram_rmt,"
2856 "ld_fbhit,ld_l1hit,ld_l2hit,"
2857 "ld_lclhit,ld_rmthit",
2858 c2c.display == DISPLAY_TOT ? "tot_hitm" :
2859 c2c.display == DISPLAY_LCL ? "lcl_hitm" : "rmt_hitm"
2862 ui_progress__init(&prog, c2c.hists.hists.nr_entries, "Sorting...");
2864 hists__collapse_resort(&c2c.hists.hists, NULL);
2865 hists__output_resort_cb(&c2c.hists.hists, &prog, resort_hitm_cb);
2866 hists__iterate_cb(&c2c.hists.hists, resort_cl_cb);
2868 ui_progress__finish();
2871 pr_err("failed to setup UI\n");
2875 perf_c2c_display(session);
2878 mem2node__exit(&c2c.mem2node);
2880 perf_session__delete(session);
2885 static int parse_record_events(const struct option *opt,
2886 const char *str, int unset __maybe_unused)
2888 bool *event_set = (bool *) opt->value;
2890 if (!strcmp(str, "list")) {
2891 perf_mem_events__list();
2894 if (perf_mem_events__parse(str))
2902 static const char * const __usage_record[] = {
2903 "perf c2c record [<options>] [<command>]",
2904 "perf c2c record [<options>] -- <command> [<options>]",
2908 static const char * const *record_mem_usage = __usage_record;
2910 static int perf_c2c__record(int argc, const char **argv)
2912 int rec_argc, i = 0, j;
2913 const char **rec_argv;
2915 bool all_user = false, all_kernel = false;
2916 bool event_set = false;
2917 struct option options[] = {
2918 OPT_CALLBACK('e', "event", &event_set, "event",
2919 "event selector. Use 'perf mem record -e list' to list available events",
2920 parse_record_events),
2921 OPT_BOOLEAN('u', "all-user", &all_user, "collect only user level data"),
2922 OPT_BOOLEAN('k', "all-kernel", &all_kernel, "collect only kernel level data"),
2923 OPT_UINTEGER('l', "ldlat", &perf_mem_events__loads_ldlat, "setup mem-loads latency"),
2924 OPT_PARENT(c2c_options),
2928 if (perf_mem_events__init()) {
2929 pr_err("failed: memory events not supported\n");
2933 argc = parse_options(argc, argv, options, record_mem_usage,
2934 PARSE_OPT_KEEP_UNKNOWN);
2936 rec_argc = argc + 11; /* max number of arguments */
2937 rec_argv = calloc(rec_argc + 1, sizeof(char *));
2941 rec_argv[i++] = "record";
2944 perf_mem_events[PERF_MEM_EVENTS__LOAD].record = true;
2945 perf_mem_events[PERF_MEM_EVENTS__STORE].record = true;
2948 if (perf_mem_events[PERF_MEM_EVENTS__LOAD].record)
2949 rec_argv[i++] = "-W";
2951 rec_argv[i++] = "-d";
2952 rec_argv[i++] = "--phys-data";
2953 rec_argv[i++] = "--sample-cpu";
2955 for (j = 0; j < PERF_MEM_EVENTS__MAX; j++) {
2956 if (!perf_mem_events[j].record)
2959 if (!perf_mem_events[j].supported) {
2960 pr_err("failed: event '%s' not supported\n",
2961 perf_mem_events[j].name);
2966 rec_argv[i++] = "-e";
2967 rec_argv[i++] = perf_mem_events__name(j);
2971 rec_argv[i++] = "--all-user";
2974 rec_argv[i++] = "--all-kernel";
2976 for (j = 0; j < argc; j++, i++)
2977 rec_argv[i] = argv[j];
2980 pr_debug("calling: ");
2984 while (rec_argv[j]) {
2985 pr_debug("%s ", rec_argv[j]);
2991 ret = cmd_record(i, rec_argv);
2996 int cmd_c2c(int argc, const char **argv)
2998 argc = parse_options(argc, argv, c2c_options, c2c_usage,
2999 PARSE_OPT_STOP_AT_NON_OPTION);
3002 usage_with_options(c2c_usage, c2c_options);
3004 if (!strncmp(argv[0], "rec", 3)) {
3005 return perf_c2c__record(argc, argv);
3006 } else if (!strncmp(argv[0], "rep", 3)) {
3007 return perf_c2c__report(argc, argv);
3009 usage_with_options(c2c_usage, c2c_options);