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"
46 #include "pmu-hybrid.h"
51 struct perf_hpp_list list;
52 struct c2c_stats stats;
55 struct compute_stats {
56 struct stats lcl_hitm;
57 struct stats rmt_hitm;
58 struct stats lcl_peer;
59 struct stats rmt_peer;
63 struct c2c_hist_entry {
64 struct c2c_hists *hists;
65 struct c2c_stats stats;
66 unsigned long *cpuset;
67 unsigned long *nodeset;
68 struct c2c_stats *node_stats;
69 unsigned int cacheline_idx;
71 struct compute_stats cstats;
74 unsigned long paddr_cnt;
80 * because of its callchain dynamic entry
85 static char const *coalesce_default = "iaddr";
88 struct perf_tool tool;
89 struct c2c_hists hists;
90 struct mem2node mem2node;
92 unsigned long **nodes;
105 /* Shared cache line stats */
106 struct c2c_stats shared_clines_stats;
111 const char *coalesce;
125 static const char *display_str[DISPLAY_MAX] = {
126 [DISPLAY_LCL_HITM] = "Local HITMs",
127 [DISPLAY_RMT_HITM] = "Remote HITMs",
128 [DISPLAY_TOT_HITM] = "Total HITMs",
129 [DISPLAY_SNP_PEER] = "Peer Snoop",
132 static const struct option c2c_options[] = {
133 OPT_INCR('v', "verbose", &verbose, "be more verbose (show counter open errors, etc)"),
137 static struct perf_c2c c2c;
139 static void *c2c_he_zalloc(size_t size)
141 struct c2c_hist_entry *c2c_he;
143 c2c_he = zalloc(size + sizeof(*c2c_he));
147 c2c_he->cpuset = bitmap_zalloc(c2c.cpus_cnt);
151 c2c_he->nodeset = bitmap_zalloc(c2c.nodes_cnt);
152 if (!c2c_he->nodeset)
155 c2c_he->node_stats = zalloc(c2c.nodes_cnt * sizeof(*c2c_he->node_stats));
156 if (!c2c_he->node_stats)
159 init_stats(&c2c_he->cstats.lcl_hitm);
160 init_stats(&c2c_he->cstats.rmt_hitm);
161 init_stats(&c2c_he->cstats.lcl_peer);
162 init_stats(&c2c_he->cstats.rmt_peer);
163 init_stats(&c2c_he->cstats.load);
168 free(c2c_he->nodeset);
169 free(c2c_he->cpuset);
174 static void c2c_he_free(void *he)
176 struct c2c_hist_entry *c2c_he;
178 c2c_he = container_of(he, struct c2c_hist_entry, he);
180 hists__delete_entries(&c2c_he->hists->hists);
184 free(c2c_he->cpuset);
185 free(c2c_he->nodeset);
186 free(c2c_he->nodestr);
187 free(c2c_he->node_stats);
191 static struct hist_entry_ops c2c_entry_ops = {
192 .new = c2c_he_zalloc,
196 static int c2c_hists__init(struct c2c_hists *hists,
198 int nr_header_lines);
200 static struct c2c_hists*
201 he__get_c2c_hists(struct hist_entry *he,
205 struct c2c_hist_entry *c2c_he;
206 struct c2c_hists *hists;
209 c2c_he = container_of(he, struct c2c_hist_entry, he);
211 return c2c_he->hists;
213 hists = c2c_he->hists = zalloc(sizeof(*hists));
217 ret = c2c_hists__init(hists, sort, nr_header_lines);
226 static void c2c_he__set_cpu(struct c2c_hist_entry *c2c_he,
227 struct perf_sample *sample)
229 if (WARN_ONCE(sample->cpu == (unsigned int) -1,
230 "WARNING: no sample cpu value"))
233 __set_bit(sample->cpu, c2c_he->cpuset);
236 static void c2c_he__set_node(struct c2c_hist_entry *c2c_he,
237 struct perf_sample *sample)
241 if (!sample->phys_addr) {
242 c2c_he->paddr_zero = true;
246 node = mem2node__node(&c2c.mem2node, sample->phys_addr);
247 if (WARN_ONCE(node < 0, "WARNING: failed to find node\n"))
250 __set_bit(node, c2c_he->nodeset);
252 if (c2c_he->paddr != sample->phys_addr) {
254 c2c_he->paddr = sample->phys_addr;
258 static void compute_stats(struct c2c_hist_entry *c2c_he,
259 struct c2c_stats *stats,
262 struct compute_stats *cstats = &c2c_he->cstats;
265 update_stats(&cstats->rmt_hitm, weight);
266 else if (stats->lcl_hitm)
267 update_stats(&cstats->lcl_hitm, weight);
268 else if (stats->rmt_peer)
269 update_stats(&cstats->rmt_peer, weight);
270 else if (stats->lcl_peer)
271 update_stats(&cstats->lcl_peer, weight);
272 else if (stats->load)
273 update_stats(&cstats->load, weight);
276 static int process_sample_event(struct perf_tool *tool __maybe_unused,
277 union perf_event *event,
278 struct perf_sample *sample,
280 struct machine *machine)
282 struct c2c_hists *c2c_hists = &c2c.hists;
283 struct c2c_hist_entry *c2c_he;
284 struct c2c_stats stats = { .nr_entries = 0, };
285 struct hist_entry *he;
286 struct addr_location al;
287 struct mem_info *mi, *mi_dup;
290 if (machine__resolve(machine, &al, sample) < 0) {
291 pr_debug("problem processing %d event, skipping it.\n",
297 al.thread->lbr_stitch_enable = true;
299 ret = sample__resolve_callchain(sample, &callchain_cursor, NULL,
300 evsel, &al, sysctl_perf_event_max_stack);
304 mi = sample__resolve_mem(sample, &al);
309 * The mi object is released in hists__add_entry_ops,
310 * if it gets sorted out into existing data, so we need
311 * to take the copy now.
313 mi_dup = mem_info__get(mi);
315 c2c_decode_stats(&stats, mi);
317 he = hists__add_entry_ops(&c2c_hists->hists, &c2c_entry_ops,
323 c2c_he = container_of(he, struct c2c_hist_entry, he);
324 c2c_add_stats(&c2c_he->stats, &stats);
325 c2c_add_stats(&c2c_hists->stats, &stats);
327 c2c_he__set_cpu(c2c_he, sample);
328 c2c_he__set_node(c2c_he, sample);
330 hists__inc_nr_samples(&c2c_hists->hists, he->filtered);
331 ret = hist_entry__append_callchain(he, sample);
335 * There's already been warning about missing
336 * sample's cpu value. Let's account all to
337 * node 0 in this case, without any further
340 * Doing node stats only for single callchain data.
342 int cpu = sample->cpu == (unsigned int) -1 ? 0 : sample->cpu;
343 int node = c2c.cpu2node[cpu];
347 c2c_hists = he__get_c2c_hists(he, c2c.cl_sort, 2);
351 he = hists__add_entry_ops(&c2c_hists->hists, &c2c_entry_ops,
357 c2c_he = container_of(he, struct c2c_hist_entry, he);
358 c2c_add_stats(&c2c_he->stats, &stats);
359 c2c_add_stats(&c2c_hists->stats, &stats);
360 c2c_add_stats(&c2c_he->node_stats[node], &stats);
362 compute_stats(c2c_he, &stats, sample->weight);
364 c2c_he__set_cpu(c2c_he, sample);
365 c2c_he__set_node(c2c_he, sample);
367 hists__inc_nr_samples(&c2c_hists->hists, he->filtered);
368 ret = hist_entry__append_callchain(he, sample);
372 addr_location__put(&al);
376 mem_info__put(mi_dup);
382 static struct perf_c2c c2c = {
384 .sample = process_sample_event,
385 .mmap = perf_event__process_mmap,
386 .mmap2 = perf_event__process_mmap2,
387 .comm = perf_event__process_comm,
388 .exit = perf_event__process_exit,
389 .fork = perf_event__process_fork,
390 .lost = perf_event__process_lost,
391 .attr = perf_event__process_attr,
392 .auxtrace_info = perf_event__process_auxtrace_info,
393 .auxtrace = perf_event__process_auxtrace,
394 .auxtrace_error = perf_event__process_auxtrace_error,
395 .ordered_events = true,
396 .ordering_requires_timestamps = true,
400 static const char * const c2c_usage[] = {
401 "perf c2c {record|report}",
405 static const char * const __usage_report[] = {
410 static const char * const *report_c2c_usage = __usage_report;
412 #define C2C_HEADER_MAX 2
418 } line[C2C_HEADER_MAX];
421 struct c2c_dimension {
422 struct c2c_header header;
425 struct sort_entry *se;
427 int64_t (*cmp)(struct perf_hpp_fmt *fmt,
428 struct hist_entry *, struct hist_entry *);
429 int (*entry)(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
430 struct hist_entry *he);
431 int (*color)(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
432 struct hist_entry *he);
436 struct perf_hpp_fmt fmt;
437 struct c2c_dimension *dim;
440 #define SYMBOL_WIDTH 30
442 static struct c2c_dimension dim_symbol;
443 static struct c2c_dimension dim_srcline;
445 static int symbol_width(struct hists *hists, struct sort_entry *se)
447 int width = hists__col_len(hists, se->se_width_idx);
449 if (!c2c.symbol_full)
450 width = MIN(width, SYMBOL_WIDTH);
455 static int c2c_width(struct perf_hpp_fmt *fmt,
456 struct perf_hpp *hpp __maybe_unused,
459 struct c2c_fmt *c2c_fmt;
460 struct c2c_dimension *dim;
462 c2c_fmt = container_of(fmt, struct c2c_fmt, fmt);
465 if (dim == &dim_symbol || dim == &dim_srcline)
466 return symbol_width(hists, dim->se);
468 return dim->se ? hists__col_len(hists, dim->se->se_width_idx) :
472 static int c2c_header(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
473 struct hists *hists, int line, int *span)
475 struct perf_hpp_list *hpp_list = hists->hpp_list;
476 struct c2c_fmt *c2c_fmt;
477 struct c2c_dimension *dim;
478 const char *text = NULL;
479 int width = c2c_width(fmt, hpp, hists);
481 c2c_fmt = container_of(fmt, struct c2c_fmt, fmt);
485 text = dim->header.line[line].text;
486 /* Use the last line from sort_entry if not defined. */
487 if (!text && (line == hpp_list->nr_header_lines - 1))
488 text = dim->se->se_header;
490 text = dim->header.line[line].text;
496 *span = dim->header.line[line].span;
503 return scnprintf(hpp->buf, hpp->size, "%*s", width, text);
506 #define HEX_STR(__s, __v) \
508 scnprintf(__s, sizeof(__s), "0x%" PRIx64, __v); \
513 dcacheline_cmp(struct perf_hpp_fmt *fmt __maybe_unused,
514 struct hist_entry *left, struct hist_entry *right)
516 return sort__dcacheline_cmp(left, right);
519 static int dcacheline_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
520 struct hist_entry *he)
523 int width = c2c_width(fmt, hpp, he->hists);
527 addr = cl_address(he->mem_info->daddr.addr);
529 return scnprintf(hpp->buf, hpp->size, "%*s", width, HEX_STR(buf, addr));
533 dcacheline_node_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
534 struct hist_entry *he)
536 struct c2c_hist_entry *c2c_he;
537 int width = c2c_width(fmt, hpp, he->hists);
539 c2c_he = container_of(he, struct c2c_hist_entry, he);
540 if (WARN_ON_ONCE(!c2c_he->nodestr))
543 return scnprintf(hpp->buf, hpp->size, "%*s", width, c2c_he->nodestr);
547 dcacheline_node_count(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
548 struct hist_entry *he)
550 struct c2c_hist_entry *c2c_he;
551 int width = c2c_width(fmt, hpp, he->hists);
553 c2c_he = container_of(he, struct c2c_hist_entry, he);
554 return scnprintf(hpp->buf, hpp->size, "%*lu", width, c2c_he->paddr_cnt);
557 static int offset_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
558 struct hist_entry *he)
561 int width = c2c_width(fmt, hpp, he->hists);
565 addr = cl_offset(he->mem_info->daddr.al_addr);
567 return scnprintf(hpp->buf, hpp->size, "%*s", width, HEX_STR(buf, addr));
571 offset_cmp(struct perf_hpp_fmt *fmt __maybe_unused,
572 struct hist_entry *left, struct hist_entry *right)
574 uint64_t l = 0, r = 0;
577 l = cl_offset(left->mem_info->daddr.addr);
579 r = cl_offset(right->mem_info->daddr.addr);
581 return (int64_t)(r - l);
585 iaddr_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
586 struct hist_entry *he)
589 int width = c2c_width(fmt, hpp, he->hists);
593 addr = he->mem_info->iaddr.addr;
595 return scnprintf(hpp->buf, hpp->size, "%*s", width, HEX_STR(buf, addr));
599 iaddr_cmp(struct perf_hpp_fmt *fmt __maybe_unused,
600 struct hist_entry *left, struct hist_entry *right)
602 return sort__iaddr_cmp(left, right);
606 tot_hitm_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
607 struct hist_entry *he)
609 struct c2c_hist_entry *c2c_he;
610 int width = c2c_width(fmt, hpp, he->hists);
611 unsigned int tot_hitm;
613 c2c_he = container_of(he, struct c2c_hist_entry, he);
614 tot_hitm = c2c_he->stats.lcl_hitm + c2c_he->stats.rmt_hitm;
616 return scnprintf(hpp->buf, hpp->size, "%*u", width, tot_hitm);
620 tot_hitm_cmp(struct perf_hpp_fmt *fmt __maybe_unused,
621 struct hist_entry *left, struct hist_entry *right)
623 struct c2c_hist_entry *c2c_left;
624 struct c2c_hist_entry *c2c_right;
625 uint64_t tot_hitm_left;
626 uint64_t tot_hitm_right;
628 c2c_left = container_of(left, struct c2c_hist_entry, he);
629 c2c_right = container_of(right, struct c2c_hist_entry, he);
631 tot_hitm_left = c2c_left->stats.lcl_hitm + c2c_left->stats.rmt_hitm;
632 tot_hitm_right = c2c_right->stats.lcl_hitm + c2c_right->stats.rmt_hitm;
634 return tot_hitm_left - tot_hitm_right;
637 #define STAT_FN_ENTRY(__f) \
639 __f ## _entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp, \
640 struct hist_entry *he) \
642 struct c2c_hist_entry *c2c_he; \
643 int width = c2c_width(fmt, hpp, he->hists); \
645 c2c_he = container_of(he, struct c2c_hist_entry, he); \
646 return scnprintf(hpp->buf, hpp->size, "%*u", width, \
647 c2c_he->stats.__f); \
650 #define STAT_FN_CMP(__f) \
652 __f ## _cmp(struct perf_hpp_fmt *fmt __maybe_unused, \
653 struct hist_entry *left, struct hist_entry *right) \
655 struct c2c_hist_entry *c2c_left, *c2c_right; \
657 c2c_left = container_of(left, struct c2c_hist_entry, he); \
658 c2c_right = container_of(right, struct c2c_hist_entry, he); \
659 return (uint64_t) c2c_left->stats.__f - \
660 (uint64_t) c2c_right->stats.__f; \
663 #define STAT_FN(__f) \
682 static uint64_t get_load_llc_misses(struct c2c_stats *stats)
684 return stats->lcl_dram +
690 static uint64_t get_load_cache_hits(struct c2c_stats *stats)
692 return stats->ld_fbhit +
699 static uint64_t get_stores(struct c2c_stats *stats)
701 return stats->st_l1hit +
706 static uint64_t total_records(struct c2c_stats *stats)
708 return get_load_llc_misses(stats) +
709 get_load_cache_hits(stats) +
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 return get_load_llc_misses(stats) +
748 get_load_cache_hits(stats);
752 tot_loads_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
753 struct hist_entry *he)
755 struct c2c_hist_entry *c2c_he;
756 int width = c2c_width(fmt, hpp, he->hists);
759 c2c_he = container_of(he, struct c2c_hist_entry, he);
760 tot_recs = total_loads(&c2c_he->stats);
762 return scnprintf(hpp->buf, hpp->size, "%*" PRIu64, width, tot_recs);
766 tot_loads_cmp(struct perf_hpp_fmt *fmt __maybe_unused,
767 struct hist_entry *left, struct hist_entry *right)
769 struct c2c_hist_entry *c2c_left;
770 struct c2c_hist_entry *c2c_right;
771 uint64_t tot_recs_left;
772 uint64_t tot_recs_right;
774 c2c_left = container_of(left, struct c2c_hist_entry, he);
775 c2c_right = container_of(right, struct c2c_hist_entry, he);
777 tot_recs_left = total_loads(&c2c_left->stats);
778 tot_recs_right = total_loads(&c2c_right->stats);
780 return tot_recs_left - tot_recs_right;
783 typedef double (get_percent_cb)(struct c2c_hist_entry *);
786 percent_color(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
787 struct hist_entry *he, get_percent_cb get_percent)
789 struct c2c_hist_entry *c2c_he;
790 int width = c2c_width(fmt, hpp, he->hists);
793 c2c_he = container_of(he, struct c2c_hist_entry, he);
794 per = get_percent(c2c_he);
796 #ifdef HAVE_SLANG_SUPPORT
798 return __hpp__slsmg_color_printf(hpp, "%*.2f%%", width - 1, per);
800 return hpp_color_scnprintf(hpp, "%*.2f%%", width - 1, per);
803 static double percent_costly_snoop(struct c2c_hist_entry *c2c_he)
805 struct c2c_hists *hists;
806 struct c2c_stats *stats;
807 struct c2c_stats *total;
811 hists = container_of(c2c_he->he.hists, struct c2c_hists, hists);
812 stats = &c2c_he->stats;
813 total = &hists->stats;
815 switch (c2c.display) {
816 case DISPLAY_RMT_HITM:
817 st = stats->rmt_hitm;
818 tot = total->rmt_hitm;
820 case DISPLAY_LCL_HITM:
821 st = stats->lcl_hitm;
822 tot = total->lcl_hitm;
824 case DISPLAY_TOT_HITM:
825 st = stats->tot_hitm;
826 tot = total->tot_hitm;
828 case DISPLAY_SNP_PEER:
829 st = stats->tot_peer;
830 tot = total->tot_peer;
836 p = tot ? (double) st / tot : 0;
841 #define PERC_STR(__s, __v) \
843 scnprintf(__s, sizeof(__s), "%.2F%%", __v); \
848 percent_costly_snoop_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
849 struct hist_entry *he)
851 struct c2c_hist_entry *c2c_he;
852 int width = c2c_width(fmt, hpp, he->hists);
856 c2c_he = container_of(he, struct c2c_hist_entry, he);
857 per = percent_costly_snoop(c2c_he);
858 return scnprintf(hpp->buf, hpp->size, "%*s", width, PERC_STR(buf, per));
862 percent_costly_snoop_color(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
863 struct hist_entry *he)
865 return percent_color(fmt, hpp, he, percent_costly_snoop);
869 percent_costly_snoop_cmp(struct perf_hpp_fmt *fmt __maybe_unused,
870 struct hist_entry *left, struct hist_entry *right)
872 struct c2c_hist_entry *c2c_left;
873 struct c2c_hist_entry *c2c_right;
877 c2c_left = container_of(left, struct c2c_hist_entry, he);
878 c2c_right = container_of(right, struct c2c_hist_entry, he);
880 per_left = percent_costly_snoop(c2c_left);
881 per_right = percent_costly_snoop(c2c_right);
883 return per_left - per_right;
886 static struct c2c_stats *he_stats(struct hist_entry *he)
888 struct c2c_hist_entry *c2c_he;
890 c2c_he = container_of(he, struct c2c_hist_entry, he);
891 return &c2c_he->stats;
894 static struct c2c_stats *total_stats(struct hist_entry *he)
896 struct c2c_hists *hists;
898 hists = container_of(he->hists, struct c2c_hists, hists);
899 return &hists->stats;
902 static double percent(u32 st, u32 tot)
904 return tot ? 100. * (double) st / (double) tot : 0;
907 #define PERCENT(__h, __f) percent(he_stats(__h)->__f, total_stats(__h)->__f)
909 #define PERCENT_FN(__f) \
910 static double percent_ ## __f(struct c2c_hist_entry *c2c_he) \
912 struct c2c_hists *hists; \
914 hists = container_of(c2c_he->he.hists, struct c2c_hists, hists); \
915 return percent(c2c_he->stats.__f, hists->stats.__f); \
923 PERCENT_FN(st_l1miss)
927 percent_rmt_hitm_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
928 struct hist_entry *he)
930 int width = c2c_width(fmt, hpp, he->hists);
931 double per = PERCENT(he, rmt_hitm);
934 return scnprintf(hpp->buf, hpp->size, "%*s", width, PERC_STR(buf, per));
938 percent_rmt_hitm_color(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
939 struct hist_entry *he)
941 return percent_color(fmt, hpp, he, percent_rmt_hitm);
945 percent_rmt_hitm_cmp(struct perf_hpp_fmt *fmt __maybe_unused,
946 struct hist_entry *left, struct hist_entry *right)
951 per_left = PERCENT(left, rmt_hitm);
952 per_right = PERCENT(right, rmt_hitm);
954 return per_left - per_right;
958 percent_lcl_hitm_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
959 struct hist_entry *he)
961 int width = c2c_width(fmt, hpp, he->hists);
962 double per = PERCENT(he, lcl_hitm);
965 return scnprintf(hpp->buf, hpp->size, "%*s", width, PERC_STR(buf, per));
969 percent_lcl_hitm_color(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
970 struct hist_entry *he)
972 return percent_color(fmt, hpp, he, percent_lcl_hitm);
976 percent_lcl_hitm_cmp(struct perf_hpp_fmt *fmt __maybe_unused,
977 struct hist_entry *left, struct hist_entry *right)
982 per_left = PERCENT(left, lcl_hitm);
983 per_right = PERCENT(right, lcl_hitm);
985 return per_left - per_right;
989 percent_lcl_peer_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
990 struct hist_entry *he)
992 int width = c2c_width(fmt, hpp, he->hists);
993 double per = PERCENT(he, lcl_peer);
996 return scnprintf(hpp->buf, hpp->size, "%*s", width, PERC_STR(buf, per));
1000 percent_lcl_peer_color(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
1001 struct hist_entry *he)
1003 return percent_color(fmt, hpp, he, percent_lcl_peer);
1007 percent_lcl_peer_cmp(struct perf_hpp_fmt *fmt __maybe_unused,
1008 struct hist_entry *left, struct hist_entry *right)
1013 per_left = PERCENT(left, lcl_peer);
1014 per_right = PERCENT(right, lcl_peer);
1016 return per_left - per_right;
1020 percent_rmt_peer_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
1021 struct hist_entry *he)
1023 int width = c2c_width(fmt, hpp, he->hists);
1024 double per = PERCENT(he, rmt_peer);
1027 return scnprintf(hpp->buf, hpp->size, "%*s", width, PERC_STR(buf, per));
1031 percent_rmt_peer_color(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
1032 struct hist_entry *he)
1034 return percent_color(fmt, hpp, he, percent_rmt_peer);
1038 percent_rmt_peer_cmp(struct perf_hpp_fmt *fmt __maybe_unused,
1039 struct hist_entry *left, struct hist_entry *right)
1044 per_left = PERCENT(left, rmt_peer);
1045 per_right = PERCENT(right, rmt_peer);
1047 return per_left - per_right;
1051 percent_stores_l1hit_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
1052 struct hist_entry *he)
1054 int width = c2c_width(fmt, hpp, he->hists);
1055 double per = PERCENT(he, st_l1hit);
1058 return scnprintf(hpp->buf, hpp->size, "%*s", width, PERC_STR(buf, per));
1062 percent_stores_l1hit_color(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
1063 struct hist_entry *he)
1065 return percent_color(fmt, hpp, he, percent_st_l1hit);
1069 percent_stores_l1hit_cmp(struct perf_hpp_fmt *fmt __maybe_unused,
1070 struct hist_entry *left, struct hist_entry *right)
1075 per_left = PERCENT(left, st_l1hit);
1076 per_right = PERCENT(right, st_l1hit);
1078 return per_left - per_right;
1082 percent_stores_l1miss_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
1083 struct hist_entry *he)
1085 int width = c2c_width(fmt, hpp, he->hists);
1086 double per = PERCENT(he, st_l1miss);
1089 return scnprintf(hpp->buf, hpp->size, "%*s", width, PERC_STR(buf, per));
1093 percent_stores_l1miss_color(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
1094 struct hist_entry *he)
1096 return percent_color(fmt, hpp, he, percent_st_l1miss);
1100 percent_stores_l1miss_cmp(struct perf_hpp_fmt *fmt __maybe_unused,
1101 struct hist_entry *left, struct hist_entry *right)
1106 per_left = PERCENT(left, st_l1miss);
1107 per_right = PERCENT(right, st_l1miss);
1109 return per_left - per_right;
1113 percent_stores_na_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
1114 struct hist_entry *he)
1116 int width = c2c_width(fmt, hpp, he->hists);
1117 double per = PERCENT(he, st_na);
1120 return scnprintf(hpp->buf, hpp->size, "%*s", width, PERC_STR(buf, per));
1124 percent_stores_na_color(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
1125 struct hist_entry *he)
1127 return percent_color(fmt, hpp, he, percent_st_na);
1131 percent_stores_na_cmp(struct perf_hpp_fmt *fmt __maybe_unused,
1132 struct hist_entry *left, struct hist_entry *right)
1137 per_left = PERCENT(left, st_na);
1138 per_right = PERCENT(right, st_na);
1140 return per_left - per_right;
1147 pid_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
1148 struct hist_entry *he)
1150 int width = c2c_width(fmt, hpp, he->hists);
1152 return scnprintf(hpp->buf, hpp->size, "%*d", width, he->thread->pid_);
1156 pid_cmp(struct perf_hpp_fmt *fmt __maybe_unused,
1157 struct hist_entry *left, struct hist_entry *right)
1159 return left->thread->pid_ - right->thread->pid_;
1163 empty_cmp(struct perf_hpp_fmt *fmt __maybe_unused,
1164 struct hist_entry *left __maybe_unused,
1165 struct hist_entry *right __maybe_unused)
1170 static int display_metrics(struct perf_hpp *hpp, u32 val, u32 sum)
1175 ret = scnprintf(hpp->buf, hpp->size, "%5.1f%% ",
1178 ret = scnprintf(hpp->buf, hpp->size, "%6s ", "n/a");
1184 node_entry(struct perf_hpp_fmt *fmt __maybe_unused, struct perf_hpp *hpp,
1185 struct hist_entry *he)
1187 struct c2c_hist_entry *c2c_he;
1192 c2c_he = container_of(he, struct c2c_hist_entry, he);
1194 for (node = 0; node < c2c.nodes_cnt; node++) {
1195 DECLARE_BITMAP(set, c2c.cpus_cnt);
1197 bitmap_zero(set, c2c.cpus_cnt);
1198 bitmap_and(set, c2c_he->cpuset, c2c.nodes[node], c2c.cpus_cnt);
1200 if (bitmap_empty(set, c2c.cpus_cnt)) {
1201 if (c2c.node_info == 1) {
1202 ret = scnprintf(hpp->buf, hpp->size, "%21s", " ");
1203 advance_hpp(hpp, ret);
1209 ret = scnprintf(hpp->buf, hpp->size, " ");
1210 advance_hpp(hpp, ret);
1213 switch (c2c.node_info) {
1215 ret = scnprintf(hpp->buf, hpp->size, "%2d", node);
1216 advance_hpp(hpp, ret);
1220 int num = bitmap_weight(set, c2c.cpus_cnt);
1221 struct c2c_stats *stats = &c2c_he->node_stats[node];
1223 ret = scnprintf(hpp->buf, hpp->size, "%2d{%2d ", node, num);
1224 advance_hpp(hpp, ret);
1226 switch (c2c.display) {
1227 case DISPLAY_RMT_HITM:
1228 ret = display_metrics(hpp, stats->rmt_hitm,
1229 c2c_he->stats.rmt_hitm);
1231 case DISPLAY_LCL_HITM:
1232 ret = display_metrics(hpp, stats->lcl_hitm,
1233 c2c_he->stats.lcl_hitm);
1235 case DISPLAY_TOT_HITM:
1236 ret = display_metrics(hpp, stats->tot_hitm,
1237 c2c_he->stats.tot_hitm);
1239 case DISPLAY_SNP_PEER:
1240 ret = display_metrics(hpp, stats->tot_peer,
1241 c2c_he->stats.tot_peer);
1247 advance_hpp(hpp, ret);
1249 if (c2c_he->stats.store > 0) {
1250 ret = scnprintf(hpp->buf, hpp->size, "%5.1f%%}",
1251 percent(stats->store, c2c_he->stats.store));
1253 ret = scnprintf(hpp->buf, hpp->size, "%6s}", "n/a");
1256 advance_hpp(hpp, ret);
1260 ret = scnprintf(hpp->buf, hpp->size, "%2d{", node);
1261 advance_hpp(hpp, ret);
1263 ret = bitmap_scnprintf(set, c2c.cpus_cnt, hpp->buf, hpp->size);
1264 advance_hpp(hpp, ret);
1266 ret = scnprintf(hpp->buf, hpp->size, "}");
1267 advance_hpp(hpp, ret);
1280 mean_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
1281 struct hist_entry *he, double mean)
1283 int width = c2c_width(fmt, hpp, he->hists);
1286 scnprintf(buf, 10, "%6.0f", mean);
1287 return scnprintf(hpp->buf, hpp->size, "%*s", width, buf);
1290 #define MEAN_ENTRY(__func, __val) \
1292 __func(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp, struct hist_entry *he) \
1294 struct c2c_hist_entry *c2c_he; \
1295 c2c_he = container_of(he, struct c2c_hist_entry, he); \
1296 return mean_entry(fmt, hpp, he, avg_stats(&c2c_he->cstats.__val)); \
1299 MEAN_ENTRY(mean_rmt_entry, rmt_hitm);
1300 MEAN_ENTRY(mean_lcl_entry, lcl_hitm);
1301 MEAN_ENTRY(mean_load_entry, load);
1302 MEAN_ENTRY(mean_rmt_peer_entry, rmt_peer);
1303 MEAN_ENTRY(mean_lcl_peer_entry, lcl_peer);
1306 cpucnt_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
1307 struct hist_entry *he)
1309 struct c2c_hist_entry *c2c_he;
1310 int width = c2c_width(fmt, hpp, he->hists);
1313 c2c_he = container_of(he, struct c2c_hist_entry, he);
1315 scnprintf(buf, 10, "%d", bitmap_weight(c2c_he->cpuset, c2c.cpus_cnt));
1316 return scnprintf(hpp->buf, hpp->size, "%*s", width, buf);
1320 cl_idx_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
1321 struct hist_entry *he)
1323 struct c2c_hist_entry *c2c_he;
1324 int width = c2c_width(fmt, hpp, he->hists);
1327 c2c_he = container_of(he, struct c2c_hist_entry, he);
1329 scnprintf(buf, 10, "%u", c2c_he->cacheline_idx);
1330 return scnprintf(hpp->buf, hpp->size, "%*s", width, buf);
1334 cl_idx_empty_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
1335 struct hist_entry *he)
1337 int width = c2c_width(fmt, hpp, he->hists);
1339 return scnprintf(hpp->buf, hpp->size, "%*s", width, "");
1342 #define HEADER_LOW(__h) \
1349 #define HEADER_BOTH(__h0, __h1) \
1359 #define HEADER_SPAN(__h0, __h1, __s) \
1370 #define HEADER_SPAN_LOW(__h) \
1377 static struct c2c_dimension dim_dcacheline = {
1378 .header = HEADER_SPAN("--- Cacheline ----", "Address", 2),
1379 .name = "dcacheline",
1380 .cmp = dcacheline_cmp,
1381 .entry = dcacheline_entry,
1385 static struct c2c_dimension dim_dcacheline_node = {
1386 .header = HEADER_LOW("Node"),
1387 .name = "dcacheline_node",
1389 .entry = dcacheline_node_entry,
1393 static struct c2c_dimension dim_dcacheline_count = {
1394 .header = HEADER_LOW("PA cnt"),
1395 .name = "dcacheline_count",
1397 .entry = dcacheline_node_count,
1401 static struct c2c_header header_offset_tui = HEADER_SPAN("-----", "Off", 2);
1403 static struct c2c_dimension dim_offset = {
1404 .header = HEADER_SPAN("--- Data address -", "Offset", 2),
1407 .entry = offset_entry,
1411 static struct c2c_dimension dim_offset_node = {
1412 .header = HEADER_LOW("Node"),
1413 .name = "offset_node",
1415 .entry = dcacheline_node_entry,
1419 static struct c2c_dimension dim_iaddr = {
1420 .header = HEADER_LOW("Code address"),
1423 .entry = iaddr_entry,
1427 static struct c2c_dimension dim_tot_hitm = {
1428 .header = HEADER_SPAN("------- Load Hitm -------", "Total", 2),
1430 .cmp = tot_hitm_cmp,
1431 .entry = tot_hitm_entry,
1435 static struct c2c_dimension dim_lcl_hitm = {
1436 .header = HEADER_SPAN_LOW("LclHitm"),
1438 .cmp = lcl_hitm_cmp,
1439 .entry = lcl_hitm_entry,
1443 static struct c2c_dimension dim_rmt_hitm = {
1444 .header = HEADER_SPAN_LOW("RmtHitm"),
1446 .cmp = rmt_hitm_cmp,
1447 .entry = rmt_hitm_entry,
1451 static struct c2c_dimension dim_tot_peer = {
1452 .header = HEADER_SPAN("------- Load Peer -------", "Total", 2),
1454 .cmp = tot_peer_cmp,
1455 .entry = tot_peer_entry,
1459 static struct c2c_dimension dim_lcl_peer = {
1460 .header = HEADER_SPAN_LOW("Local"),
1462 .cmp = lcl_peer_cmp,
1463 .entry = lcl_peer_entry,
1467 static struct c2c_dimension dim_rmt_peer = {
1468 .header = HEADER_SPAN_LOW("Remote"),
1470 .cmp = rmt_peer_cmp,
1471 .entry = rmt_peer_entry,
1475 static struct c2c_dimension dim_cl_rmt_hitm = {
1476 .header = HEADER_SPAN("----- HITM -----", "Rmt", 1),
1477 .name = "cl_rmt_hitm",
1478 .cmp = rmt_hitm_cmp,
1479 .entry = rmt_hitm_entry,
1483 static struct c2c_dimension dim_cl_lcl_hitm = {
1484 .header = HEADER_SPAN_LOW("Lcl"),
1485 .name = "cl_lcl_hitm",
1486 .cmp = lcl_hitm_cmp,
1487 .entry = lcl_hitm_entry,
1491 static struct c2c_dimension dim_cl_rmt_peer = {
1492 .header = HEADER_SPAN("----- Peer -----", "Rmt", 1),
1493 .name = "cl_rmt_peer",
1494 .cmp = rmt_peer_cmp,
1495 .entry = rmt_peer_entry,
1499 static struct c2c_dimension dim_cl_lcl_peer = {
1500 .header = HEADER_SPAN_LOW("Lcl"),
1501 .name = "cl_lcl_peer",
1502 .cmp = lcl_peer_cmp,
1503 .entry = lcl_peer_entry,
1507 static struct c2c_dimension dim_tot_stores = {
1508 .header = HEADER_BOTH("Total", "Stores"),
1509 .name = "tot_stores",
1511 .entry = store_entry,
1515 static struct c2c_dimension dim_stores_l1hit = {
1516 .header = HEADER_SPAN("--------- Stores --------", "L1Hit", 2),
1517 .name = "stores_l1hit",
1518 .cmp = st_l1hit_cmp,
1519 .entry = st_l1hit_entry,
1523 static struct c2c_dimension dim_stores_l1miss = {
1524 .header = HEADER_SPAN_LOW("L1Miss"),
1525 .name = "stores_l1miss",
1526 .cmp = st_l1miss_cmp,
1527 .entry = st_l1miss_entry,
1531 static struct c2c_dimension dim_stores_na = {
1532 .header = HEADER_SPAN_LOW("N/A"),
1533 .name = "stores_na",
1535 .entry = st_na_entry,
1539 static struct c2c_dimension dim_cl_stores_l1hit = {
1540 .header = HEADER_SPAN("------- Store Refs ------", "L1 Hit", 2),
1541 .name = "cl_stores_l1hit",
1542 .cmp = st_l1hit_cmp,
1543 .entry = st_l1hit_entry,
1547 static struct c2c_dimension dim_cl_stores_l1miss = {
1548 .header = HEADER_SPAN_LOW("L1 Miss"),
1549 .name = "cl_stores_l1miss",
1550 .cmp = st_l1miss_cmp,
1551 .entry = st_l1miss_entry,
1555 static struct c2c_dimension dim_cl_stores_na = {
1556 .header = HEADER_SPAN_LOW("N/A"),
1557 .name = "cl_stores_na",
1559 .entry = st_na_entry,
1563 static struct c2c_dimension dim_ld_fbhit = {
1564 .header = HEADER_SPAN("----- Core Load Hit -----", "FB", 2),
1566 .cmp = ld_fbhit_cmp,
1567 .entry = ld_fbhit_entry,
1571 static struct c2c_dimension dim_ld_l1hit = {
1572 .header = HEADER_SPAN_LOW("L1"),
1574 .cmp = ld_l1hit_cmp,
1575 .entry = ld_l1hit_entry,
1579 static struct c2c_dimension dim_ld_l2hit = {
1580 .header = HEADER_SPAN_LOW("L2"),
1582 .cmp = ld_l2hit_cmp,
1583 .entry = ld_l2hit_entry,
1587 static struct c2c_dimension dim_ld_llchit = {
1588 .header = HEADER_SPAN("- LLC Load Hit --", "LclHit", 1),
1589 .name = "ld_lclhit",
1590 .cmp = ld_llchit_cmp,
1591 .entry = ld_llchit_entry,
1595 static struct c2c_dimension dim_ld_rmthit = {
1596 .header = HEADER_SPAN("- RMT Load Hit --", "RmtHit", 1),
1597 .name = "ld_rmthit",
1599 .entry = rmt_hit_entry,
1603 static struct c2c_dimension dim_tot_recs = {
1604 .header = HEADER_BOTH("Total", "records"),
1606 .cmp = tot_recs_cmp,
1607 .entry = tot_recs_entry,
1611 static struct c2c_dimension dim_tot_loads = {
1612 .header = HEADER_BOTH("Total", "Loads"),
1613 .name = "tot_loads",
1614 .cmp = tot_loads_cmp,
1615 .entry = tot_loads_entry,
1619 static struct c2c_header percent_costly_snoop_header[] = {
1620 [DISPLAY_LCL_HITM] = HEADER_BOTH("Lcl", "Hitm"),
1621 [DISPLAY_RMT_HITM] = HEADER_BOTH("Rmt", "Hitm"),
1622 [DISPLAY_TOT_HITM] = HEADER_BOTH("Tot", "Hitm"),
1623 [DISPLAY_SNP_PEER] = HEADER_BOTH("Peer", "Snoop"),
1626 static struct c2c_dimension dim_percent_costly_snoop = {
1627 .name = "percent_costly_snoop",
1628 .cmp = percent_costly_snoop_cmp,
1629 .entry = percent_costly_snoop_entry,
1630 .color = percent_costly_snoop_color,
1634 static struct c2c_dimension dim_percent_rmt_hitm = {
1635 .header = HEADER_SPAN("----- HITM -----", "RmtHitm", 1),
1636 .name = "percent_rmt_hitm",
1637 .cmp = percent_rmt_hitm_cmp,
1638 .entry = percent_rmt_hitm_entry,
1639 .color = percent_rmt_hitm_color,
1643 static struct c2c_dimension dim_percent_lcl_hitm = {
1644 .header = HEADER_SPAN_LOW("LclHitm"),
1645 .name = "percent_lcl_hitm",
1646 .cmp = percent_lcl_hitm_cmp,
1647 .entry = percent_lcl_hitm_entry,
1648 .color = percent_lcl_hitm_color,
1652 static struct c2c_dimension dim_percent_rmt_peer = {
1653 .header = HEADER_SPAN("-- Peer Snoop --", "Rmt", 1),
1654 .name = "percent_rmt_peer",
1655 .cmp = percent_rmt_peer_cmp,
1656 .entry = percent_rmt_peer_entry,
1657 .color = percent_rmt_peer_color,
1661 static struct c2c_dimension dim_percent_lcl_peer = {
1662 .header = HEADER_SPAN_LOW("Lcl"),
1663 .name = "percent_lcl_peer",
1664 .cmp = percent_lcl_peer_cmp,
1665 .entry = percent_lcl_peer_entry,
1666 .color = percent_lcl_peer_color,
1670 static struct c2c_dimension dim_percent_stores_l1hit = {
1671 .header = HEADER_SPAN("------- Store Refs ------", "L1 Hit", 2),
1672 .name = "percent_stores_l1hit",
1673 .cmp = percent_stores_l1hit_cmp,
1674 .entry = percent_stores_l1hit_entry,
1675 .color = percent_stores_l1hit_color,
1679 static struct c2c_dimension dim_percent_stores_l1miss = {
1680 .header = HEADER_SPAN_LOW("L1 Miss"),
1681 .name = "percent_stores_l1miss",
1682 .cmp = percent_stores_l1miss_cmp,
1683 .entry = percent_stores_l1miss_entry,
1684 .color = percent_stores_l1miss_color,
1688 static struct c2c_dimension dim_percent_stores_na = {
1689 .header = HEADER_SPAN_LOW("N/A"),
1690 .name = "percent_stores_na",
1691 .cmp = percent_stores_na_cmp,
1692 .entry = percent_stores_na_entry,
1693 .color = percent_stores_na_color,
1697 static struct c2c_dimension dim_dram_lcl = {
1698 .header = HEADER_SPAN("--- Load Dram ----", "Lcl", 1),
1700 .cmp = lcl_dram_cmp,
1701 .entry = lcl_dram_entry,
1705 static struct c2c_dimension dim_dram_rmt = {
1706 .header = HEADER_SPAN_LOW("Rmt"),
1708 .cmp = rmt_dram_cmp,
1709 .entry = rmt_dram_entry,
1713 static struct c2c_dimension dim_pid = {
1714 .header = HEADER_LOW("Pid"),
1721 static struct c2c_dimension dim_tid = {
1722 .header = HEADER_LOW("Tid"),
1727 static struct c2c_dimension dim_symbol = {
1732 static struct c2c_dimension dim_dso = {
1733 .header = HEADER_BOTH("Shared", "Object"),
1738 static struct c2c_dimension dim_node = {
1741 .entry = node_entry,
1745 static struct c2c_dimension dim_mean_rmt = {
1746 .header = HEADER_SPAN("---------- cycles ----------", "rmt hitm", 2),
1749 .entry = mean_rmt_entry,
1753 static struct c2c_dimension dim_mean_lcl = {
1754 .header = HEADER_SPAN_LOW("lcl hitm"),
1757 .entry = mean_lcl_entry,
1761 static struct c2c_dimension dim_mean_load = {
1762 .header = HEADER_SPAN_LOW("load"),
1763 .name = "mean_load",
1765 .entry = mean_load_entry,
1769 static struct c2c_dimension dim_mean_rmt_peer = {
1770 .header = HEADER_SPAN("---------- cycles ----------", "rmt peer", 2),
1771 .name = "mean_rmt_peer",
1773 .entry = mean_rmt_peer_entry,
1777 static struct c2c_dimension dim_mean_lcl_peer = {
1778 .header = HEADER_SPAN_LOW("lcl peer"),
1779 .name = "mean_lcl_peer",
1781 .entry = mean_lcl_peer_entry,
1785 static struct c2c_dimension dim_cpucnt = {
1786 .header = HEADER_BOTH("cpu", "cnt"),
1789 .entry = cpucnt_entry,
1793 static struct c2c_dimension dim_srcline = {
1794 .name = "cl_srcline",
1795 .se = &sort_srcline,
1798 static struct c2c_dimension dim_dcacheline_idx = {
1799 .header = HEADER_LOW("Index"),
1802 .entry = cl_idx_entry,
1806 static struct c2c_dimension dim_dcacheline_num = {
1807 .header = HEADER_LOW("Num"),
1810 .entry = cl_idx_entry,
1814 static struct c2c_dimension dim_dcacheline_num_empty = {
1815 .header = HEADER_LOW("Num"),
1816 .name = "cl_num_empty",
1818 .entry = cl_idx_empty_entry,
1822 static struct c2c_dimension *dimensions[] = {
1824 &dim_dcacheline_node,
1825 &dim_dcacheline_count,
1843 &dim_cl_stores_l1hit,
1844 &dim_cl_stores_l1miss,
1853 &dim_percent_costly_snoop,
1854 &dim_percent_rmt_hitm,
1855 &dim_percent_lcl_hitm,
1856 &dim_percent_rmt_peer,
1857 &dim_percent_lcl_peer,
1858 &dim_percent_stores_l1hit,
1859 &dim_percent_stores_l1miss,
1860 &dim_percent_stores_na,
1875 &dim_dcacheline_idx,
1876 &dim_dcacheline_num,
1877 &dim_dcacheline_num_empty,
1881 static void fmt_free(struct perf_hpp_fmt *fmt)
1883 struct c2c_fmt *c2c_fmt;
1885 c2c_fmt = container_of(fmt, struct c2c_fmt, fmt);
1889 static bool fmt_equal(struct perf_hpp_fmt *a, struct perf_hpp_fmt *b)
1891 struct c2c_fmt *c2c_a = container_of(a, struct c2c_fmt, fmt);
1892 struct c2c_fmt *c2c_b = container_of(b, struct c2c_fmt, fmt);
1894 return c2c_a->dim == c2c_b->dim;
1897 static struct c2c_dimension *get_dimension(const char *name)
1901 for (i = 0; dimensions[i]; i++) {
1902 struct c2c_dimension *dim = dimensions[i];
1904 if (!strcmp(dim->name, name))
1911 static int c2c_se_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
1912 struct hist_entry *he)
1914 struct c2c_fmt *c2c_fmt = container_of(fmt, struct c2c_fmt, fmt);
1915 struct c2c_dimension *dim = c2c_fmt->dim;
1916 size_t len = fmt->user_len;
1919 len = hists__col_len(he->hists, dim->se->se_width_idx);
1921 if (dim == &dim_symbol || dim == &dim_srcline)
1922 len = symbol_width(he->hists, dim->se);
1925 return dim->se->se_snprintf(he, hpp->buf, hpp->size, len);
1928 static int64_t c2c_se_cmp(struct perf_hpp_fmt *fmt,
1929 struct hist_entry *a, struct hist_entry *b)
1931 struct c2c_fmt *c2c_fmt = container_of(fmt, struct c2c_fmt, fmt);
1932 struct c2c_dimension *dim = c2c_fmt->dim;
1934 return dim->se->se_cmp(a, b);
1937 static int64_t c2c_se_collapse(struct perf_hpp_fmt *fmt,
1938 struct hist_entry *a, struct hist_entry *b)
1940 struct c2c_fmt *c2c_fmt = container_of(fmt, struct c2c_fmt, fmt);
1941 struct c2c_dimension *dim = c2c_fmt->dim;
1942 int64_t (*collapse_fn)(struct hist_entry *, struct hist_entry *);
1944 collapse_fn = dim->se->se_collapse ?: dim->se->se_cmp;
1945 return collapse_fn(a, b);
1948 static struct c2c_fmt *get_format(const char *name)
1950 struct c2c_dimension *dim = get_dimension(name);
1951 struct c2c_fmt *c2c_fmt;
1952 struct perf_hpp_fmt *fmt;
1957 c2c_fmt = zalloc(sizeof(*c2c_fmt));
1963 fmt = &c2c_fmt->fmt;
1964 INIT_LIST_HEAD(&fmt->list);
1965 INIT_LIST_HEAD(&fmt->sort_list);
1967 fmt->cmp = dim->se ? c2c_se_cmp : dim->cmp;
1968 fmt->sort = dim->se ? c2c_se_cmp : dim->cmp;
1969 fmt->color = dim->se ? NULL : dim->color;
1970 fmt->entry = dim->se ? c2c_se_entry : dim->entry;
1971 fmt->header = c2c_header;
1972 fmt->width = c2c_width;
1973 fmt->collapse = dim->se ? c2c_se_collapse : dim->cmp;
1974 fmt->equal = fmt_equal;
1975 fmt->free = fmt_free;
1980 static int c2c_hists__init_output(struct perf_hpp_list *hpp_list, char *name)
1982 struct c2c_fmt *c2c_fmt = get_format(name);
1986 return output_field_add(hpp_list, name);
1989 perf_hpp_list__column_register(hpp_list, &c2c_fmt->fmt);
1993 static int c2c_hists__init_sort(struct perf_hpp_list *hpp_list, char *name)
1995 struct c2c_fmt *c2c_fmt = get_format(name);
1996 struct c2c_dimension *dim;
2000 return sort_dimension__add(hpp_list, name, NULL, 0);
2004 if (dim == &dim_dso)
2007 perf_hpp_list__register_sort_field(hpp_list, &c2c_fmt->fmt);
2011 #define PARSE_LIST(_list, _fn) \
2019 for (tok = strtok_r((char *)_list, ", ", &tmp); \
2020 tok; tok = strtok_r(NULL, ", ", &tmp)) { \
2021 ret = _fn(hpp_list, tok); \
2022 if (ret == -EINVAL) { \
2023 pr_err("Invalid --fields key: `%s'", tok); \
2025 } else if (ret == -ESRCH) { \
2026 pr_err("Unknown --fields key: `%s'", tok); \
2032 static int hpp_list__parse(struct perf_hpp_list *hpp_list,
2033 const char *output_,
2036 char *output = output_ ? strdup(output_) : NULL;
2037 char *sort = sort_ ? strdup(sort_) : NULL;
2040 PARSE_LIST(output, c2c_hists__init_output);
2041 PARSE_LIST(sort, c2c_hists__init_sort);
2043 /* copy sort keys to output fields */
2044 perf_hpp__setup_output_field(hpp_list);
2047 * We dont need other sorting keys other than those
2048 * we already specified. It also really slows down
2049 * the processing a lot with big number of output
2050 * fields, so switching this off for c2c.
2054 /* and then copy output fields to sort keys */
2055 perf_hpp__append_sort_keys(&hists->list);
2063 static int c2c_hists__init(struct c2c_hists *hists,
2065 int nr_header_lines)
2067 __hists__init(&hists->hists, &hists->list);
2070 * Initialize only with sort fields, we need to resort
2071 * later anyway, and that's where we add output fields
2074 perf_hpp_list__init(&hists->list);
2076 /* Overload number of header lines.*/
2077 hists->list.nr_header_lines = nr_header_lines;
2079 return hpp_list__parse(&hists->list, NULL, sort);
2082 static int c2c_hists__reinit(struct c2c_hists *c2c_hists,
2086 perf_hpp__reset_output_field(&c2c_hists->list);
2087 return hpp_list__parse(&c2c_hists->list, output, sort);
2090 #define DISPLAY_LINE_LIMIT 0.001
2092 static u8 filter_display(u32 val, u32 sum)
2094 if (sum == 0 || ((double)val / sum) < DISPLAY_LINE_LIMIT)
2095 return HIST_FILTER__C2C;
2100 static bool he__display(struct hist_entry *he, struct c2c_stats *stats)
2102 struct c2c_hist_entry *c2c_he;
2107 c2c_he = container_of(he, struct c2c_hist_entry, he);
2109 switch (c2c.display) {
2110 case DISPLAY_LCL_HITM:
2111 he->filtered = filter_display(c2c_he->stats.lcl_hitm,
2114 case DISPLAY_RMT_HITM:
2115 he->filtered = filter_display(c2c_he->stats.rmt_hitm,
2118 case DISPLAY_TOT_HITM:
2119 he->filtered = filter_display(c2c_he->stats.tot_hitm,
2122 case DISPLAY_SNP_PEER:
2123 he->filtered = filter_display(c2c_he->stats.tot_peer,
2130 return he->filtered == 0;
2133 static inline bool is_valid_hist_entry(struct hist_entry *he)
2135 struct c2c_hist_entry *c2c_he;
2136 bool has_record = false;
2138 c2c_he = container_of(he, struct c2c_hist_entry, he);
2140 /* It's a valid entry if contains stores */
2141 if (c2c_he->stats.store)
2144 switch (c2c.display) {
2145 case DISPLAY_LCL_HITM:
2146 has_record = !!c2c_he->stats.lcl_hitm;
2148 case DISPLAY_RMT_HITM:
2149 has_record = !!c2c_he->stats.rmt_hitm;
2151 case DISPLAY_TOT_HITM:
2152 has_record = !!c2c_he->stats.tot_hitm;
2154 case DISPLAY_SNP_PEER:
2155 has_record = !!c2c_he->stats.tot_peer;
2163 static void set_node_width(struct c2c_hist_entry *c2c_he, int len)
2165 struct c2c_dimension *dim;
2167 dim = &c2c.hists == c2c_he->hists ?
2168 &dim_dcacheline_node : &dim_offset_node;
2170 if (len > dim->width)
2174 static int set_nodestr(struct c2c_hist_entry *c2c_he)
2179 if (c2c_he->nodestr)
2182 if (!bitmap_empty(c2c_he->nodeset, c2c.nodes_cnt)) {
2183 len = bitmap_scnprintf(c2c_he->nodeset, c2c.nodes_cnt,
2186 len = scnprintf(buf, sizeof(buf), "N/A");
2189 set_node_width(c2c_he, len);
2190 c2c_he->nodestr = strdup(buf);
2191 return c2c_he->nodestr ? 0 : -ENOMEM;
2194 static void calc_width(struct c2c_hist_entry *c2c_he)
2196 struct c2c_hists *c2c_hists;
2198 c2c_hists = container_of(c2c_he->he.hists, struct c2c_hists, hists);
2199 hists__calc_col_len(&c2c_hists->hists, &c2c_he->he);
2200 set_nodestr(c2c_he);
2203 static int filter_cb(struct hist_entry *he, void *arg __maybe_unused)
2205 struct c2c_hist_entry *c2c_he;
2207 c2c_he = container_of(he, struct c2c_hist_entry, he);
2209 if (c2c.show_src && !he->srcline)
2210 he->srcline = hist_entry__srcline(he);
2214 if (!is_valid_hist_entry(he))
2215 he->filtered = HIST_FILTER__C2C;
2220 static int resort_cl_cb(struct hist_entry *he, void *arg __maybe_unused)
2222 struct c2c_hist_entry *c2c_he;
2223 struct c2c_hists *c2c_hists;
2224 bool display = he__display(he, &c2c.shared_clines_stats);
2226 c2c_he = container_of(he, struct c2c_hist_entry, he);
2227 c2c_hists = c2c_he->hists;
2229 if (display && c2c_hists) {
2230 static unsigned int idx;
2232 c2c_he->cacheline_idx = idx++;
2235 c2c_hists__reinit(c2c_hists, c2c.cl_output, c2c.cl_resort);
2237 hists__collapse_resort(&c2c_hists->hists, NULL);
2238 hists__output_resort_cb(&c2c_hists->hists, NULL, filter_cb);
2244 static struct c2c_header header_node_0 = HEADER_LOW("Node");
2245 static struct c2c_header header_node_1_hitms_stores =
2246 HEADER_LOW("Node{cpus %hitms %stores}");
2247 static struct c2c_header header_node_1_peers_stores =
2248 HEADER_LOW("Node{cpus %peers %stores}");
2249 static struct c2c_header header_node_2 = HEADER_LOW("Node{cpu list}");
2251 static void setup_nodes_header(void)
2253 switch (c2c.node_info) {
2255 dim_node.header = header_node_0;
2258 if (c2c.display == DISPLAY_SNP_PEER)
2259 dim_node.header = header_node_1_peers_stores;
2261 dim_node.header = header_node_1_hitms_stores;
2264 dim_node.header = header_node_2;
2273 static int setup_nodes(struct perf_session *session)
2275 struct numa_node *n;
2276 unsigned long **nodes;
2278 struct perf_cpu cpu;
2281 if (c2c.node_info > 2)
2284 c2c.nodes_cnt = session->header.env.nr_numa_nodes;
2285 c2c.cpus_cnt = session->header.env.nr_cpus_avail;
2287 n = session->header.env.numa_nodes;
2291 nodes = zalloc(sizeof(unsigned long *) * c2c.nodes_cnt);
2297 cpu2node = zalloc(sizeof(int) * c2c.cpus_cnt);
2301 for (idx = 0; idx < c2c.cpus_cnt; idx++)
2304 c2c.cpu2node = cpu2node;
2306 for (node = 0; node < c2c.nodes_cnt; node++) {
2307 struct perf_cpu_map *map = n[node].map;
2310 set = bitmap_zalloc(c2c.cpus_cnt);
2316 /* empty node, skip */
2317 if (perf_cpu_map__empty(map))
2320 perf_cpu_map__for_each_cpu(cpu, idx, map) {
2321 __set_bit(cpu.cpu, set);
2323 if (WARN_ONCE(cpu2node[cpu.cpu] != -1, "node/cpu topology bug"))
2326 cpu2node[cpu.cpu] = node;
2330 setup_nodes_header();
2334 #define HAS_HITMS(__h) ((__h)->stats.lcl_hitm || (__h)->stats.rmt_hitm)
2335 #define HAS_PEER(__h) ((__h)->stats.lcl_peer || (__h)->stats.rmt_peer)
2337 static int resort_shared_cl_cb(struct hist_entry *he, void *arg __maybe_unused)
2339 struct c2c_hist_entry *c2c_he;
2340 c2c_he = container_of(he, struct c2c_hist_entry, he);
2342 if (HAS_HITMS(c2c_he) || HAS_PEER(c2c_he)) {
2343 c2c.shared_clines++;
2344 c2c_add_stats(&c2c.shared_clines_stats, &c2c_he->stats);
2350 static int hists__iterate_cb(struct hists *hists, hists__resort_cb_t cb)
2352 struct rb_node *next = rb_first_cached(&hists->entries);
2356 struct hist_entry *he;
2358 he = rb_entry(next, struct hist_entry, rb_node);
2362 next = rb_next(&he->rb_node);
2368 static void print_c2c__display_stats(FILE *out)
2371 struct c2c_stats *stats = &c2c.hists.stats;
2373 llc_misses = get_load_llc_misses(stats);
2375 fprintf(out, "=================================================\n");
2376 fprintf(out, " Trace Event Information \n");
2377 fprintf(out, "=================================================\n");
2378 fprintf(out, " Total records : %10d\n", stats->nr_entries);
2379 fprintf(out, " Locked Load/Store Operations : %10d\n", stats->locks);
2380 fprintf(out, " Load Operations : %10d\n", stats->load);
2381 fprintf(out, " Loads - uncacheable : %10d\n", stats->ld_uncache);
2382 fprintf(out, " Loads - IO : %10d\n", stats->ld_io);
2383 fprintf(out, " Loads - Miss : %10d\n", stats->ld_miss);
2384 fprintf(out, " Loads - no mapping : %10d\n", stats->ld_noadrs);
2385 fprintf(out, " Load Fill Buffer Hit : %10d\n", stats->ld_fbhit);
2386 fprintf(out, " Load L1D hit : %10d\n", stats->ld_l1hit);
2387 fprintf(out, " Load L2D hit : %10d\n", stats->ld_l2hit);
2388 fprintf(out, " Load LLC hit : %10d\n", stats->ld_llchit + stats->lcl_hitm);
2389 fprintf(out, " Load Local HITM : %10d\n", stats->lcl_hitm);
2390 fprintf(out, " Load Remote HITM : %10d\n", stats->rmt_hitm);
2391 fprintf(out, " Load Remote HIT : %10d\n", stats->rmt_hit);
2392 fprintf(out, " Load Local DRAM : %10d\n", stats->lcl_dram);
2393 fprintf(out, " Load Remote DRAM : %10d\n", stats->rmt_dram);
2394 fprintf(out, " Load MESI State Exclusive : %10d\n", stats->ld_excl);
2395 fprintf(out, " Load MESI State Shared : %10d\n", stats->ld_shared);
2396 fprintf(out, " Load LLC Misses : %10d\n", llc_misses);
2397 fprintf(out, " Load access blocked by data : %10d\n", stats->blk_data);
2398 fprintf(out, " Load access blocked by address : %10d\n", stats->blk_addr);
2399 fprintf(out, " Load HIT Local Peer : %10d\n", stats->lcl_peer);
2400 fprintf(out, " Load HIT Remote Peer : %10d\n", stats->rmt_peer);
2401 fprintf(out, " LLC Misses to Local DRAM : %10.1f%%\n", ((double)stats->lcl_dram/(double)llc_misses) * 100.);
2402 fprintf(out, " LLC Misses to Remote DRAM : %10.1f%%\n", ((double)stats->rmt_dram/(double)llc_misses) * 100.);
2403 fprintf(out, " LLC Misses to Remote cache (HIT) : %10.1f%%\n", ((double)stats->rmt_hit /(double)llc_misses) * 100.);
2404 fprintf(out, " LLC Misses to Remote cache (HITM) : %10.1f%%\n", ((double)stats->rmt_hitm/(double)llc_misses) * 100.);
2405 fprintf(out, " Store Operations : %10d\n", stats->store);
2406 fprintf(out, " Store - uncacheable : %10d\n", stats->st_uncache);
2407 fprintf(out, " Store - no mapping : %10d\n", stats->st_noadrs);
2408 fprintf(out, " Store L1D Hit : %10d\n", stats->st_l1hit);
2409 fprintf(out, " Store L1D Miss : %10d\n", stats->st_l1miss);
2410 fprintf(out, " Store No available memory level : %10d\n", stats->st_na);
2411 fprintf(out, " No Page Map Rejects : %10d\n", stats->nomap);
2412 fprintf(out, " Unable to parse data source : %10d\n", stats->noparse);
2415 static void print_shared_cacheline_info(FILE *out)
2417 struct c2c_stats *stats = &c2c.shared_clines_stats;
2418 int hitm_cnt = stats->lcl_hitm + stats->rmt_hitm;
2420 fprintf(out, "=================================================\n");
2421 fprintf(out, " Global Shared Cache Line Event Information \n");
2422 fprintf(out, "=================================================\n");
2423 fprintf(out, " Total Shared Cache Lines : %10d\n", c2c.shared_clines);
2424 fprintf(out, " Load HITs on shared lines : %10d\n", stats->load);
2425 fprintf(out, " Fill Buffer Hits on shared lines : %10d\n", stats->ld_fbhit);
2426 fprintf(out, " L1D hits on shared lines : %10d\n", stats->ld_l1hit);
2427 fprintf(out, " L2D hits on shared lines : %10d\n", stats->ld_l2hit);
2428 fprintf(out, " LLC hits on shared lines : %10d\n", stats->ld_llchit + stats->lcl_hitm);
2429 fprintf(out, " Load hits on peer cache or nodes : %10d\n", stats->lcl_peer + stats->rmt_peer);
2430 fprintf(out, " Locked Access on shared lines : %10d\n", stats->locks);
2431 fprintf(out, " Blocked Access on shared lines : %10d\n", stats->blk_data + stats->blk_addr);
2432 fprintf(out, " Store HITs on shared lines : %10d\n", stats->store);
2433 fprintf(out, " Store L1D hits on shared lines : %10d\n", stats->st_l1hit);
2434 fprintf(out, " Store No available memory level : %10d\n", stats->st_na);
2435 fprintf(out, " Total Merged records : %10d\n", hitm_cnt + stats->store);
2438 static void print_cacheline(struct c2c_hists *c2c_hists,
2439 struct hist_entry *he_cl,
2440 struct perf_hpp_list *hpp_list,
2444 struct perf_hpp hpp = {
2451 hists__fprintf_headers(&c2c_hists->hists, out);
2457 fprintf(out, " ----------------------------------------------------------------------\n");
2458 __hist_entry__snprintf(he_cl, &hpp, hpp_list);
2459 fprintf(out, "%s\n", bf);
2460 fprintf(out, " ----------------------------------------------------------------------\n");
2462 hists__fprintf(&c2c_hists->hists, false, 0, 0, 0, out, false);
2465 static void print_pareto(FILE *out)
2467 struct perf_hpp_list hpp_list;
2470 const char *cl_output;
2472 if (c2c.display != DISPLAY_SNP_PEER)
2473 cl_output = "cl_num,"
2481 cl_output = "cl_num,"
2489 perf_hpp_list__init(&hpp_list);
2490 ret = hpp_list__parse(&hpp_list, cl_output, NULL);
2492 if (WARN_ONCE(ret, "failed to setup sort entries\n"))
2495 nd = rb_first_cached(&c2c.hists.hists.entries);
2497 for (; nd; nd = rb_next(nd)) {
2498 struct hist_entry *he = rb_entry(nd, struct hist_entry, rb_node);
2499 struct c2c_hist_entry *c2c_he;
2504 c2c_he = container_of(he, struct c2c_hist_entry, he);
2505 print_cacheline(c2c_he->hists, he, &hpp_list, out);
2509 static void print_c2c_info(FILE *out, struct perf_session *session)
2511 struct evlist *evlist = session->evlist;
2512 struct evsel *evsel;
2515 fprintf(out, "=================================================\n");
2516 fprintf(out, " c2c details \n");
2517 fprintf(out, "=================================================\n");
2519 evlist__for_each_entry(evlist, evsel) {
2520 fprintf(out, "%-36s: %s\n", first ? " Events" : "", evsel__name(evsel));
2523 fprintf(out, " Cachelines sort on : %s\n",
2524 display_str[c2c.display]);
2525 fprintf(out, " Cacheline data grouping : %s\n", c2c.cl_sort);
2528 static void perf_c2c__hists_fprintf(FILE *out, struct perf_session *session)
2532 print_c2c__display_stats(out);
2534 print_shared_cacheline_info(out);
2536 print_c2c_info(out, session);
2542 fprintf(out, "=================================================\n");
2543 fprintf(out, " Shared Data Cache Line Table \n");
2544 fprintf(out, "=================================================\n");
2545 fprintf(out, "#\n");
2547 hists__fprintf(&c2c.hists.hists, true, 0, 0, 0, stdout, true);
2550 fprintf(out, "=================================================\n");
2551 fprintf(out, " Shared Cache Line Distribution Pareto \n");
2552 fprintf(out, "=================================================\n");
2553 fprintf(out, "#\n");
2558 #ifdef HAVE_SLANG_SUPPORT
2559 static void c2c_browser__update_nr_entries(struct hist_browser *hb)
2562 struct rb_node *nd = rb_first_cached(&hb->hists->entries);
2565 struct hist_entry *he = rb_entry(nd, struct hist_entry, rb_node);
2573 hb->nr_non_filtered_entries = nr_entries;
2576 struct c2c_cacheline_browser {
2577 struct hist_browser hb;
2578 struct hist_entry *he;
2582 perf_c2c_cacheline_browser__title(struct hist_browser *browser,
2583 char *bf, size_t size)
2585 struct c2c_cacheline_browser *cl_browser;
2586 struct hist_entry *he;
2589 cl_browser = container_of(browser, struct c2c_cacheline_browser, hb);
2590 he = cl_browser->he;
2593 addr = cl_address(he->mem_info->daddr.addr);
2595 scnprintf(bf, size, "Cacheline 0x%lx", addr);
2599 static struct c2c_cacheline_browser*
2600 c2c_cacheline_browser__new(struct hists *hists, struct hist_entry *he)
2602 struct c2c_cacheline_browser *browser;
2604 browser = zalloc(sizeof(*browser));
2606 hist_browser__init(&browser->hb, hists);
2607 browser->hb.c2c_filter = true;
2608 browser->hb.title = perf_c2c_cacheline_browser__title;
2615 static int perf_c2c__browse_cacheline(struct hist_entry *he)
2617 struct c2c_hist_entry *c2c_he;
2618 struct c2c_hists *c2c_hists;
2619 struct c2c_cacheline_browser *cl_browser;
2620 struct hist_browser *browser;
2622 static const char help[] =
2623 " ENTER Toggle callchains (if present) \n"
2624 " n Toggle Node details info \n"
2625 " s Toggle full length of symbol and source line columns \n"
2626 " q Return back to cacheline list \n";
2631 /* Display compact version first. */
2632 c2c.symbol_full = false;
2634 c2c_he = container_of(he, struct c2c_hist_entry, he);
2635 c2c_hists = c2c_he->hists;
2637 cl_browser = c2c_cacheline_browser__new(&c2c_hists->hists, he);
2638 if (cl_browser == NULL)
2641 browser = &cl_browser->hb;
2643 /* reset abort key so that it can get Ctrl-C as a key */
2645 SLang_init_tty(0, 0, 0);
2647 c2c_browser__update_nr_entries(browser);
2650 key = hist_browser__run(browser, "? - help", true, 0);
2654 c2c.symbol_full = !c2c.symbol_full;
2657 c2c.node_info = (c2c.node_info + 1) % 3;
2658 setup_nodes_header();
2663 ui_browser__help_window(&browser->b, help);
2675 static int perf_c2c_browser__title(struct hist_browser *browser,
2676 char *bf, size_t size)
2679 "Shared Data Cache Line Table "
2680 "(%lu entries, sorted on %s)",
2681 browser->nr_non_filtered_entries,
2682 display_str[c2c.display]);
2686 static struct hist_browser*
2687 perf_c2c_browser__new(struct hists *hists)
2689 struct hist_browser *browser = hist_browser__new(hists);
2692 browser->title = perf_c2c_browser__title;
2693 browser->c2c_filter = true;
2699 static int perf_c2c__hists_browse(struct hists *hists)
2701 struct hist_browser *browser;
2703 static const char help[] =
2704 " d Display cacheline details \n"
2705 " ENTER Toggle callchains (if present) \n"
2708 browser = perf_c2c_browser__new(hists);
2709 if (browser == NULL)
2712 /* reset abort key so that it can get Ctrl-C as a key */
2714 SLang_init_tty(0, 0, 0);
2716 c2c_browser__update_nr_entries(browser);
2719 key = hist_browser__run(browser, "? - help", true, 0);
2725 perf_c2c__browse_cacheline(browser->he_selection);
2728 ui_browser__help_window(&browser->b, help);
2736 hist_browser__delete(browser);
2740 static void perf_c2c_display(struct perf_session *session)
2742 if (use_browser == 0)
2743 perf_c2c__hists_fprintf(stdout, session);
2745 perf_c2c__hists_browse(&c2c.hists.hists);
2748 static void perf_c2c_display(struct perf_session *session)
2751 perf_c2c__hists_fprintf(stdout, session);
2753 #endif /* HAVE_SLANG_SUPPORT */
2755 static char *fill_line(const char *orig, int len)
2757 int i, j, olen = strlen(orig);
2760 buf = zalloc(len + 1);
2764 j = len / 2 - olen / 2;
2766 for (i = 0; i < j - 1; i++)
2771 strcpy(buf + i, orig);
2777 for (; i < len; i++)
2783 static int ui_quirks(void)
2785 const char *nodestr = "Data address";
2788 if (!c2c.use_stdio) {
2789 dim_offset.width = 5;
2790 dim_offset.header = header_offset_tui;
2794 dim_percent_costly_snoop.header = percent_costly_snoop_header[c2c.display];
2796 /* Fix the zero line for dcacheline column. */
2797 buf = fill_line("Cacheline", dim_dcacheline.width +
2798 dim_dcacheline_node.width +
2799 dim_dcacheline_count.width + 4);
2803 dim_dcacheline.header.line[0].text = buf;
2805 /* Fix the zero line for offset column. */
2806 buf = fill_line(nodestr, dim_offset.width +
2807 dim_offset_node.width +
2808 dim_dcacheline_count.width + 4);
2812 dim_offset.header.line[0].text = buf;
2817 #define CALLCHAIN_DEFAULT_OPT "graph,0.5,caller,function,percent"
2819 const char callchain_help[] = "Display call graph (stack chain/backtrace):\n\n"
2820 CALLCHAIN_REPORT_HELP
2821 "\n\t\t\t\tDefault: " CALLCHAIN_DEFAULT_OPT;
2824 parse_callchain_opt(const struct option *opt, const char *arg, int unset)
2826 struct callchain_param *callchain = opt->value;
2828 callchain->enabled = !unset;
2833 symbol_conf.use_callchain = false;
2834 callchain->mode = CHAIN_NONE;
2838 return parse_callchain_report_opt(arg);
2841 static int setup_callchain(struct evlist *evlist)
2843 u64 sample_type = evlist__combined_sample_type(evlist);
2844 enum perf_call_graph_mode mode = CALLCHAIN_NONE;
2846 if ((sample_type & PERF_SAMPLE_REGS_USER) &&
2847 (sample_type & PERF_SAMPLE_STACK_USER)) {
2848 mode = CALLCHAIN_DWARF;
2849 dwarf_callchain_users = true;
2850 } else if (sample_type & PERF_SAMPLE_BRANCH_STACK)
2851 mode = CALLCHAIN_LBR;
2852 else if (sample_type & PERF_SAMPLE_CALLCHAIN)
2853 mode = CALLCHAIN_FP;
2855 if (!callchain_param.enabled &&
2856 callchain_param.mode != CHAIN_NONE &&
2857 mode != CALLCHAIN_NONE) {
2858 symbol_conf.use_callchain = true;
2859 if (callchain_register_param(&callchain_param) < 0) {
2860 ui__error("Can't register callchain params.\n");
2865 if (c2c.stitch_lbr && (mode != CALLCHAIN_LBR)) {
2866 ui__warning("Can't find LBR callchain. Switch off --stitch-lbr.\n"
2867 "Please apply --call-graph lbr when recording.\n");
2868 c2c.stitch_lbr = false;
2871 callchain_param.record_mode = mode;
2872 callchain_param.min_percent = 0;
2876 static int setup_display(const char *str)
2878 const char *display = str;
2880 if (!strcmp(display, "tot"))
2881 c2c.display = DISPLAY_TOT_HITM;
2882 else if (!strcmp(display, "rmt"))
2883 c2c.display = DISPLAY_RMT_HITM;
2884 else if (!strcmp(display, "lcl"))
2885 c2c.display = DISPLAY_LCL_HITM;
2886 else if (!strcmp(display, "peer"))
2887 c2c.display = DISPLAY_SNP_PEER;
2889 pr_err("failed: unknown display type: %s\n", str);
2896 #define for_each_token(__tok, __buf, __sep, __tmp) \
2897 for (__tok = strtok_r(__buf, __sep, &__tmp); __tok; \
2898 __tok = strtok_r(NULL, __sep, &__tmp))
2900 static int build_cl_output(char *cl_sort, bool no_source)
2902 char *tok, *tmp, *buf = strdup(cl_sort);
2903 bool add_pid = false;
2904 bool add_tid = false;
2905 bool add_iaddr = false;
2906 bool add_sym = false;
2907 bool add_dso = false;
2908 bool add_src = false;
2914 for_each_token(tok, buf, ",", tmp) {
2915 if (!strcmp(tok, "tid")) {
2917 } else if (!strcmp(tok, "pid")) {
2919 } else if (!strcmp(tok, "iaddr")) {
2923 add_src = no_source ? false : true;
2924 } else if (!strcmp(tok, "dso")) {
2926 } else if (strcmp(tok, "offset")) {
2927 pr_err("unrecognized sort token: %s\n", tok);
2933 if (asprintf(&c2c.cl_output,
2934 "%s%s%s%s%s%s%s%s%s%s%s%s",
2935 c2c.use_stdio ? "cl_num_empty," : "",
2936 c2c.display == DISPLAY_SNP_PEER ? "percent_rmt_peer,"
2937 "percent_lcl_peer," :
2939 "percent_lcl_hitm,",
2940 "percent_stores_l1hit,"
2941 "percent_stores_l1miss,"
2942 "percent_stores_na,"
2943 "offset,offset_node,dcacheline_count,",
2944 add_pid ? "pid," : "",
2945 add_tid ? "tid," : "",
2946 add_iaddr ? "iaddr," : "",
2947 c2c.display == DISPLAY_SNP_PEER ? "mean_rmt_peer,"
2954 add_sym ? "symbol," : "",
2955 add_dso ? "dso," : "",
2956 add_src ? "cl_srcline," : "",
2962 c2c.show_src = add_src;
2968 static int setup_coalesce(const char *coalesce, bool no_source)
2970 const char *c = coalesce ?: coalesce_default;
2971 const char *sort_str = NULL;
2973 if (asprintf(&c2c.cl_sort, "offset,%s", c) < 0)
2976 if (build_cl_output(c2c.cl_sort, no_source))
2979 if (c2c.display == DISPLAY_TOT_HITM)
2980 sort_str = "tot_hitm";
2981 else if (c2c.display == DISPLAY_RMT_HITM)
2982 sort_str = "rmt_hitm,lcl_hitm";
2983 else if (c2c.display == DISPLAY_LCL_HITM)
2984 sort_str = "lcl_hitm,rmt_hitm";
2985 else if (c2c.display == DISPLAY_SNP_PEER)
2986 sort_str = "tot_peer";
2988 if (asprintf(&c2c.cl_resort, "offset,%s", sort_str) < 0)
2991 pr_debug("coalesce sort fields: %s\n", c2c.cl_sort);
2992 pr_debug("coalesce resort fields: %s\n", c2c.cl_resort);
2993 pr_debug("coalesce output fields: %s\n", c2c.cl_output);
2997 static int perf_c2c__report(int argc, const char **argv)
2999 struct itrace_synth_opts itrace_synth_opts = {
3001 .mem = true, /* Only enable memory event */
3002 .default_no_sample = true,
3005 struct perf_session *session;
3006 struct ui_progress prog;
3007 struct perf_data data = {
3008 .mode = PERF_DATA_MODE_READ,
3010 char callchain_default_opt[] = CALLCHAIN_DEFAULT_OPT;
3011 const char *display = NULL;
3012 const char *coalesce = NULL;
3013 bool no_source = false;
3014 const struct option options[] = {
3015 OPT_STRING('k', "vmlinux", &symbol_conf.vmlinux_name,
3016 "file", "vmlinux pathname"),
3017 OPT_STRING('i', "input", &input_name, "file",
3018 "the input file to process"),
3019 OPT_INCR('N', "node-info", &c2c.node_info,
3020 "show extra node info in report (repeat for more info)"),
3021 OPT_BOOLEAN(0, "stdio", &c2c.use_stdio, "Use the stdio interface"),
3022 OPT_BOOLEAN(0, "stats", &c2c.stats_only,
3023 "Display only statistic tables (implies --stdio)"),
3024 OPT_BOOLEAN(0, "full-symbols", &c2c.symbol_full,
3025 "Display full length of symbols"),
3026 OPT_BOOLEAN(0, "no-source", &no_source,
3027 "Do not display Source Line column"),
3028 OPT_BOOLEAN(0, "show-all", &c2c.show_all,
3029 "Show all captured HITM lines."),
3030 OPT_CALLBACK_DEFAULT('g', "call-graph", &callchain_param,
3031 "print_type,threshold[,print_limit],order,sort_key[,branch],value",
3032 callchain_help, &parse_callchain_opt,
3033 callchain_default_opt),
3034 OPT_STRING('d', "display", &display, "Switch HITM output type", "tot,lcl,rmt,peer"),
3035 OPT_STRING('c', "coalesce", &coalesce, "coalesce fields",
3036 "coalesce fields: pid,tid,iaddr,dso"),
3037 OPT_BOOLEAN('f', "force", &symbol_conf.force, "don't complain, do it"),
3038 OPT_BOOLEAN(0, "stitch-lbr", &c2c.stitch_lbr,
3039 "Enable LBR callgraph stitching approach"),
3040 OPT_PARENT(c2c_options),
3044 const char *output_str, *sort_str = NULL;
3046 argc = parse_options(argc, argv, options, report_c2c_usage,
3047 PARSE_OPT_STOP_AT_NON_OPTION);
3049 usage_with_options(report_c2c_usage, options);
3051 #ifndef HAVE_SLANG_SUPPORT
3052 c2c.use_stdio = true;
3056 c2c.use_stdio = true;
3058 err = symbol__validate_sym_arguments();
3062 if (!input_name || !strlen(input_name))
3063 input_name = "perf.data";
3065 data.path = input_name;
3066 data.force = symbol_conf.force;
3068 session = perf_session__new(&data, &c2c.tool);
3069 if (IS_ERR(session)) {
3070 err = PTR_ERR(session);
3071 pr_debug("Error creating perf session\n");
3076 * Use the 'tot' as default display type if user doesn't specify it;
3077 * since Arm64 platform doesn't support HITMs flag, use 'peer' as the
3078 * default display type.
3081 if (!strcmp(perf_env__arch(&session->header.env), "arm64"))
3087 err = setup_display(display);
3091 err = setup_coalesce(coalesce, no_source);
3093 pr_debug("Failed to initialize hists\n");
3097 err = c2c_hists__init(&c2c.hists, "dcacheline", 2);
3099 pr_debug("Failed to initialize hists\n");
3103 session->itrace_synth_opts = &itrace_synth_opts;
3105 err = setup_nodes(session);
3107 pr_err("Failed setup nodes\n");
3111 err = mem2node__init(&c2c.mem2node, &session->header.env);
3115 err = setup_callchain(session->evlist);
3119 if (symbol__init(&session->header.env) < 0)
3122 /* No pipe support at the moment. */
3123 if (perf_data__is_pipe(session->data)) {
3124 pr_debug("No pipe support at the moment.\n");
3133 setup_browser(false);
3135 err = perf_session__process_events(session);
3137 pr_err("failed to process sample\n");
3141 if (c2c.display != DISPLAY_SNP_PEER)
3142 output_str = "cl_idx,"
3146 "percent_costly_snoop,"
3147 "tot_hitm,lcl_hitm,rmt_hitm,"
3151 "stores_l1hit,stores_l1miss,stores_na,"
3152 "ld_fbhit,ld_l1hit,ld_l2hit,"
3153 "ld_lclhit,lcl_hitm,"
3154 "ld_rmthit,rmt_hitm,"
3155 "dram_lcl,dram_rmt";
3157 output_str = "cl_idx,"
3161 "percent_costly_snoop,"
3162 "tot_peer,lcl_peer,rmt_peer,"
3166 "stores_l1hit,stores_l1miss,stores_na,"
3167 "ld_fbhit,ld_l1hit,ld_l2hit,"
3168 "ld_lclhit,lcl_hitm,"
3169 "ld_rmthit,rmt_hitm,"
3170 "dram_lcl,dram_rmt";
3172 if (c2c.display == DISPLAY_TOT_HITM)
3173 sort_str = "tot_hitm";
3174 else if (c2c.display == DISPLAY_RMT_HITM)
3175 sort_str = "rmt_hitm";
3176 else if (c2c.display == DISPLAY_LCL_HITM)
3177 sort_str = "lcl_hitm";
3178 else if (c2c.display == DISPLAY_SNP_PEER)
3179 sort_str = "tot_peer";
3181 c2c_hists__reinit(&c2c.hists, output_str, sort_str);
3183 ui_progress__init(&prog, c2c.hists.hists.nr_entries, "Sorting...");
3185 hists__collapse_resort(&c2c.hists.hists, NULL);
3186 hists__output_resort_cb(&c2c.hists.hists, &prog, resort_shared_cl_cb);
3187 hists__iterate_cb(&c2c.hists.hists, resort_cl_cb);
3189 ui_progress__finish();
3192 pr_err("failed to setup UI\n");
3196 perf_c2c_display(session);
3199 mem2node__exit(&c2c.mem2node);
3201 perf_session__delete(session);
3206 static int parse_record_events(const struct option *opt,
3207 const char *str, int unset __maybe_unused)
3209 bool *event_set = (bool *) opt->value;
3211 if (!strcmp(str, "list")) {
3212 perf_mem_events__list();
3215 if (perf_mem_events__parse(str))
3223 static const char * const __usage_record[] = {
3224 "perf c2c record [<options>] [<command>]",
3225 "perf c2c record [<options>] -- <command> [<options>]",
3229 static const char * const *record_mem_usage = __usage_record;
3231 static int perf_c2c__record(int argc, const char **argv)
3233 int rec_argc, i = 0, j, rec_tmp_nr = 0;
3234 const char **rec_argv;
3237 bool all_user = false, all_kernel = false;
3238 bool event_set = false;
3239 struct perf_mem_event *e;
3240 struct option options[] = {
3241 OPT_CALLBACK('e', "event", &event_set, "event",
3242 "event selector. Use 'perf c2c record -e list' to list available events",
3243 parse_record_events),
3244 OPT_BOOLEAN('u', "all-user", &all_user, "collect only user level data"),
3245 OPT_BOOLEAN('k', "all-kernel", &all_kernel, "collect only kernel level data"),
3246 OPT_UINTEGER('l', "ldlat", &perf_mem_events__loads_ldlat, "setup mem-loads latency"),
3247 OPT_PARENT(c2c_options),
3251 if (perf_mem_events__init()) {
3252 pr_err("failed: memory events not supported\n");
3256 argc = parse_options(argc, argv, options, record_mem_usage,
3257 PARSE_OPT_KEEP_UNKNOWN);
3259 if (!perf_pmu__has_hybrid())
3260 rec_argc = argc + 11; /* max number of arguments */
3262 rec_argc = argc + 11 * perf_pmu__hybrid_pmu_num();
3264 rec_argv = calloc(rec_argc + 1, sizeof(char *));
3268 rec_tmp = calloc(rec_argc + 1, sizeof(char *));
3274 rec_argv[i++] = "record";
3277 e = perf_mem_events__ptr(PERF_MEM_EVENTS__LOAD_STORE);
3279 * The load and store operations are required, use the event
3280 * PERF_MEM_EVENTS__LOAD_STORE if it is supported.
3284 rec_argv[i++] = "-W";
3286 e = perf_mem_events__ptr(PERF_MEM_EVENTS__LOAD);
3289 e = perf_mem_events__ptr(PERF_MEM_EVENTS__STORE);
3294 e = perf_mem_events__ptr(PERF_MEM_EVENTS__LOAD);
3296 rec_argv[i++] = "-W";
3298 rec_argv[i++] = "-d";
3299 rec_argv[i++] = "--phys-data";
3300 rec_argv[i++] = "--sample-cpu";
3302 ret = perf_mem_events__record_args(rec_argv, &i, rec_tmp, &rec_tmp_nr);
3307 rec_argv[i++] = "--all-user";
3310 rec_argv[i++] = "--all-kernel";
3312 for (j = 0; j < argc; j++, i++)
3313 rec_argv[i] = argv[j];
3316 pr_debug("calling: ");
3320 while (rec_argv[j]) {
3321 pr_debug("%s ", rec_argv[j]);
3327 ret = cmd_record(i, rec_argv);
3329 for (i = 0; i < rec_tmp_nr; i++)
3337 int cmd_c2c(int argc, const char **argv)
3339 argc = parse_options(argc, argv, c2c_options, c2c_usage,
3340 PARSE_OPT_STOP_AT_NON_OPTION);
3343 usage_with_options(c2c_usage, c2c_options);
3345 if (strlen(argv[0]) > 2 && strstarts("record", argv[0])) {
3346 return perf_c2c__record(argc, argv);
3347 } else if (strlen(argv[0]) > 2 && strstarts("report", argv[0])) {
3348 return perf_c2c__report(argc, argv);
3350 usage_with_options(c2c_usage, c2c_options);