Loading...
1/*
2 * This is rewrite of original c2c tool introduced in here:
3 * http://lwn.net/Articles/588866/
4 *
5 * The original tool was changed to fit in current perf state.
6 *
7 * Original authors:
8 * Don Zickus <dzickus@redhat.com>
9 * Dick Fowles <fowles@inreach.com>
10 * Joe Mario <jmario@redhat.com>
11 */
12#include <linux/compiler.h>
13#include <linux/kernel.h>
14#include <linux/stringify.h>
15#include <asm/bug.h>
16#include "util.h"
17#include "debug.h"
18#include "builtin.h"
19#include <subcmd/parse-options.h>
20#include "mem-events.h"
21#include "session.h"
22#include "hist.h"
23#include "sort.h"
24#include "tool.h"
25#include "data.h"
26#include "sort.h"
27#include "evlist.h"
28#include "evsel.h"
29#include <asm/bug.h>
30#include "ui/browsers/hists.h"
31#include "evlist.h"
32
33struct c2c_hists {
34 struct hists hists;
35 struct perf_hpp_list list;
36 struct c2c_stats stats;
37};
38
39struct compute_stats {
40 struct stats lcl_hitm;
41 struct stats rmt_hitm;
42 struct stats load;
43};
44
45struct c2c_hist_entry {
46 struct c2c_hists *hists;
47 struct c2c_stats stats;
48 unsigned long *cpuset;
49 struct c2c_stats *node_stats;
50 unsigned int cacheline_idx;
51
52 struct compute_stats cstats;
53
54 /*
55 * must be at the end,
56 * because of its callchain dynamic entry
57 */
58 struct hist_entry he;
59};
60
61static char const *coalesce_default = "pid,tid,iaddr";
62
63struct perf_c2c {
64 struct perf_tool tool;
65 struct c2c_hists hists;
66
67 unsigned long **nodes;
68 int nodes_cnt;
69 int cpus_cnt;
70 int *cpu2node;
71 int node_info;
72
73 bool show_src;
74 bool show_all;
75 bool use_stdio;
76 bool stats_only;
77 bool symbol_full;
78
79 /* HITM shared clines stats */
80 struct c2c_stats hitm_stats;
81 int shared_clines;
82
83 int display;
84
85 const char *coalesce;
86 char *cl_sort;
87 char *cl_resort;
88 char *cl_output;
89};
90
91enum {
92 DISPLAY_LCL,
93 DISPLAY_RMT,
94 DISPLAY_TOT,
95 DISPLAY_MAX,
96};
97
98static const char *display_str[DISPLAY_MAX] = {
99 [DISPLAY_LCL] = "Local",
100 [DISPLAY_RMT] = "Remote",
101 [DISPLAY_TOT] = "Total",
102};
103
104static const struct option c2c_options[] = {
105 OPT_INCR('v', "verbose", &verbose, "be more verbose (show counter open errors, etc)"),
106 OPT_END()
107};
108
109static struct perf_c2c c2c;
110
111static void *c2c_he_zalloc(size_t size)
112{
113 struct c2c_hist_entry *c2c_he;
114
115 c2c_he = zalloc(size + sizeof(*c2c_he));
116 if (!c2c_he)
117 return NULL;
118
119 c2c_he->cpuset = bitmap_alloc(c2c.cpus_cnt);
120 if (!c2c_he->cpuset)
121 return NULL;
122
123 c2c_he->node_stats = zalloc(c2c.nodes_cnt * sizeof(*c2c_he->node_stats));
124 if (!c2c_he->node_stats)
125 return NULL;
126
127 init_stats(&c2c_he->cstats.lcl_hitm);
128 init_stats(&c2c_he->cstats.rmt_hitm);
129 init_stats(&c2c_he->cstats.load);
130
131 return &c2c_he->he;
132}
133
134static void c2c_he_free(void *he)
135{
136 struct c2c_hist_entry *c2c_he;
137
138 c2c_he = container_of(he, struct c2c_hist_entry, he);
139 if (c2c_he->hists) {
140 hists__delete_entries(&c2c_he->hists->hists);
141 free(c2c_he->hists);
142 }
143
144 free(c2c_he->cpuset);
145 free(c2c_he->node_stats);
146 free(c2c_he);
147}
148
149static struct hist_entry_ops c2c_entry_ops = {
150 .new = c2c_he_zalloc,
151 .free = c2c_he_free,
152};
153
154static int c2c_hists__init(struct c2c_hists *hists,
155 const char *sort,
156 int nr_header_lines);
157
158static struct c2c_hists*
159he__get_c2c_hists(struct hist_entry *he,
160 const char *sort,
161 int nr_header_lines)
162{
163 struct c2c_hist_entry *c2c_he;
164 struct c2c_hists *hists;
165 int ret;
166
167 c2c_he = container_of(he, struct c2c_hist_entry, he);
168 if (c2c_he->hists)
169 return c2c_he->hists;
170
171 hists = c2c_he->hists = zalloc(sizeof(*hists));
172 if (!hists)
173 return NULL;
174
175 ret = c2c_hists__init(hists, sort, nr_header_lines);
176 if (ret) {
177 free(hists);
178 return NULL;
179 }
180
181 return hists;
182}
183
184static void c2c_he__set_cpu(struct c2c_hist_entry *c2c_he,
185 struct perf_sample *sample)
186{
187 if (WARN_ONCE(sample->cpu == (unsigned int) -1,
188 "WARNING: no sample cpu value"))
189 return;
190
191 set_bit(sample->cpu, c2c_he->cpuset);
192}
193
194static void compute_stats(struct c2c_hist_entry *c2c_he,
195 struct c2c_stats *stats,
196 u64 weight)
197{
198 struct compute_stats *cstats = &c2c_he->cstats;
199
200 if (stats->rmt_hitm)
201 update_stats(&cstats->rmt_hitm, weight);
202 else if (stats->lcl_hitm)
203 update_stats(&cstats->lcl_hitm, weight);
204 else if (stats->load)
205 update_stats(&cstats->load, weight);
206}
207
208static int process_sample_event(struct perf_tool *tool __maybe_unused,
209 union perf_event *event,
210 struct perf_sample *sample,
211 struct perf_evsel *evsel,
212 struct machine *machine)
213{
214 struct c2c_hists *c2c_hists = &c2c.hists;
215 struct c2c_hist_entry *c2c_he;
216 struct c2c_stats stats = { .nr_entries = 0, };
217 struct hist_entry *he;
218 struct addr_location al;
219 struct mem_info *mi, *mi_dup;
220 int ret;
221
222 if (machine__resolve(machine, &al, sample) < 0) {
223 pr_debug("problem processing %d event, skipping it.\n",
224 event->header.type);
225 return -1;
226 }
227
228 ret = sample__resolve_callchain(sample, &callchain_cursor, NULL,
229 evsel, &al, sysctl_perf_event_max_stack);
230 if (ret)
231 goto out;
232
233 mi = sample__resolve_mem(sample, &al);
234 if (mi == NULL)
235 return -ENOMEM;
236
237 mi_dup = memdup(mi, sizeof(*mi));
238 if (!mi_dup)
239 goto free_mi;
240
241 c2c_decode_stats(&stats, mi);
242
243 he = hists__add_entry_ops(&c2c_hists->hists, &c2c_entry_ops,
244 &al, NULL, NULL, mi,
245 sample, true);
246 if (he == NULL)
247 goto free_mi_dup;
248
249 c2c_he = container_of(he, struct c2c_hist_entry, he);
250 c2c_add_stats(&c2c_he->stats, &stats);
251 c2c_add_stats(&c2c_hists->stats, &stats);
252
253 c2c_he__set_cpu(c2c_he, sample);
254
255 hists__inc_nr_samples(&c2c_hists->hists, he->filtered);
256 ret = hist_entry__append_callchain(he, sample);
257
258 if (!ret) {
259 /*
260 * There's already been warning about missing
261 * sample's cpu value. Let's account all to
262 * node 0 in this case, without any further
263 * warning.
264 *
265 * Doing node stats only for single callchain data.
266 */
267 int cpu = sample->cpu == (unsigned int) -1 ? 0 : sample->cpu;
268 int node = c2c.cpu2node[cpu];
269
270 mi = mi_dup;
271
272 mi_dup = memdup(mi, sizeof(*mi));
273 if (!mi_dup)
274 goto free_mi;
275
276 c2c_hists = he__get_c2c_hists(he, c2c.cl_sort, 2);
277 if (!c2c_hists)
278 goto free_mi_dup;
279
280 he = hists__add_entry_ops(&c2c_hists->hists, &c2c_entry_ops,
281 &al, NULL, NULL, mi,
282 sample, true);
283 if (he == NULL)
284 goto free_mi_dup;
285
286 c2c_he = container_of(he, struct c2c_hist_entry, he);
287 c2c_add_stats(&c2c_he->stats, &stats);
288 c2c_add_stats(&c2c_hists->stats, &stats);
289 c2c_add_stats(&c2c_he->node_stats[node], &stats);
290
291 compute_stats(c2c_he, &stats, sample->weight);
292
293 c2c_he__set_cpu(c2c_he, sample);
294
295 hists__inc_nr_samples(&c2c_hists->hists, he->filtered);
296 ret = hist_entry__append_callchain(he, sample);
297 }
298
299out:
300 addr_location__put(&al);
301 return ret;
302
303free_mi_dup:
304 free(mi_dup);
305free_mi:
306 free(mi);
307 ret = -ENOMEM;
308 goto out;
309}
310
311static struct perf_c2c c2c = {
312 .tool = {
313 .sample = process_sample_event,
314 .mmap = perf_event__process_mmap,
315 .mmap2 = perf_event__process_mmap2,
316 .comm = perf_event__process_comm,
317 .exit = perf_event__process_exit,
318 .fork = perf_event__process_fork,
319 .lost = perf_event__process_lost,
320 .ordered_events = true,
321 .ordering_requires_timestamps = true,
322 },
323};
324
325static const char * const c2c_usage[] = {
326 "perf c2c {record|report}",
327 NULL
328};
329
330static const char * const __usage_report[] = {
331 "perf c2c report",
332 NULL
333};
334
335static const char * const *report_c2c_usage = __usage_report;
336
337#define C2C_HEADER_MAX 2
338
339struct c2c_header {
340 struct {
341 const char *text;
342 int span;
343 } line[C2C_HEADER_MAX];
344};
345
346struct c2c_dimension {
347 struct c2c_header header;
348 const char *name;
349 int width;
350 struct sort_entry *se;
351
352 int64_t (*cmp)(struct perf_hpp_fmt *fmt,
353 struct hist_entry *, struct hist_entry *);
354 int (*entry)(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
355 struct hist_entry *he);
356 int (*color)(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
357 struct hist_entry *he);
358};
359
360struct c2c_fmt {
361 struct perf_hpp_fmt fmt;
362 struct c2c_dimension *dim;
363};
364
365#define SYMBOL_WIDTH 30
366
367static struct c2c_dimension dim_symbol;
368static struct c2c_dimension dim_srcline;
369
370static int symbol_width(struct hists *hists, struct sort_entry *se)
371{
372 int width = hists__col_len(hists, se->se_width_idx);
373
374 if (!c2c.symbol_full)
375 width = MIN(width, SYMBOL_WIDTH);
376
377 return width;
378}
379
380static int c2c_width(struct perf_hpp_fmt *fmt,
381 struct perf_hpp *hpp __maybe_unused,
382 struct hists *hists)
383{
384 struct c2c_fmt *c2c_fmt;
385 struct c2c_dimension *dim;
386
387 c2c_fmt = container_of(fmt, struct c2c_fmt, fmt);
388 dim = c2c_fmt->dim;
389
390 if (dim == &dim_symbol || dim == &dim_srcline)
391 return symbol_width(hists, dim->se);
392
393 return dim->se ? hists__col_len(hists, dim->se->se_width_idx) :
394 c2c_fmt->dim->width;
395}
396
397static int c2c_header(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
398 struct hists *hists, int line, int *span)
399{
400 struct perf_hpp_list *hpp_list = hists->hpp_list;
401 struct c2c_fmt *c2c_fmt;
402 struct c2c_dimension *dim;
403 const char *text = NULL;
404 int width = c2c_width(fmt, hpp, hists);
405
406 c2c_fmt = container_of(fmt, struct c2c_fmt, fmt);
407 dim = c2c_fmt->dim;
408
409 if (dim->se) {
410 text = dim->header.line[line].text;
411 /* Use the last line from sort_entry if not defined. */
412 if (!text && (line == hpp_list->nr_header_lines - 1))
413 text = dim->se->se_header;
414 } else {
415 text = dim->header.line[line].text;
416
417 if (*span) {
418 (*span)--;
419 return 0;
420 } else {
421 *span = dim->header.line[line].span;
422 }
423 }
424
425 if (text == NULL)
426 text = "";
427
428 return scnprintf(hpp->buf, hpp->size, "%*s", width, text);
429}
430
431#define HEX_STR(__s, __v) \
432({ \
433 scnprintf(__s, sizeof(__s), "0x%" PRIx64, __v); \
434 __s; \
435})
436
437static int64_t
438dcacheline_cmp(struct perf_hpp_fmt *fmt __maybe_unused,
439 struct hist_entry *left, struct hist_entry *right)
440{
441 return sort__dcacheline_cmp(left, right);
442}
443
444static int dcacheline_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
445 struct hist_entry *he)
446{
447 uint64_t addr = 0;
448 int width = c2c_width(fmt, hpp, he->hists);
449 char buf[20];
450
451 if (he->mem_info)
452 addr = cl_address(he->mem_info->daddr.addr);
453
454 return scnprintf(hpp->buf, hpp->size, "%*s", width, HEX_STR(buf, addr));
455}
456
457static int offset_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
458 struct hist_entry *he)
459{
460 uint64_t addr = 0;
461 int width = c2c_width(fmt, hpp, he->hists);
462 char buf[20];
463
464 if (he->mem_info)
465 addr = cl_offset(he->mem_info->daddr.al_addr);
466
467 return scnprintf(hpp->buf, hpp->size, "%*s", width, HEX_STR(buf, addr));
468}
469
470static int64_t
471offset_cmp(struct perf_hpp_fmt *fmt __maybe_unused,
472 struct hist_entry *left, struct hist_entry *right)
473{
474 uint64_t l = 0, r = 0;
475
476 if (left->mem_info)
477 l = cl_offset(left->mem_info->daddr.addr);
478 if (right->mem_info)
479 r = cl_offset(right->mem_info->daddr.addr);
480
481 return (int64_t)(r - l);
482}
483
484static int
485iaddr_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
486 struct hist_entry *he)
487{
488 uint64_t addr = 0;
489 int width = c2c_width(fmt, hpp, he->hists);
490 char buf[20];
491
492 if (he->mem_info)
493 addr = he->mem_info->iaddr.addr;
494
495 return scnprintf(hpp->buf, hpp->size, "%*s", width, HEX_STR(buf, addr));
496}
497
498static int64_t
499iaddr_cmp(struct perf_hpp_fmt *fmt __maybe_unused,
500 struct hist_entry *left, struct hist_entry *right)
501{
502 return sort__iaddr_cmp(left, right);
503}
504
505static int
506tot_hitm_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
507 struct hist_entry *he)
508{
509 struct c2c_hist_entry *c2c_he;
510 int width = c2c_width(fmt, hpp, he->hists);
511 unsigned int tot_hitm;
512
513 c2c_he = container_of(he, struct c2c_hist_entry, he);
514 tot_hitm = c2c_he->stats.lcl_hitm + c2c_he->stats.rmt_hitm;
515
516 return scnprintf(hpp->buf, hpp->size, "%*u", width, tot_hitm);
517}
518
519static int64_t
520tot_hitm_cmp(struct perf_hpp_fmt *fmt __maybe_unused,
521 struct hist_entry *left, struct hist_entry *right)
522{
523 struct c2c_hist_entry *c2c_left;
524 struct c2c_hist_entry *c2c_right;
525 unsigned int tot_hitm_left;
526 unsigned int tot_hitm_right;
527
528 c2c_left = container_of(left, struct c2c_hist_entry, he);
529 c2c_right = container_of(right, struct c2c_hist_entry, he);
530
531 tot_hitm_left = c2c_left->stats.lcl_hitm + c2c_left->stats.rmt_hitm;
532 tot_hitm_right = c2c_right->stats.lcl_hitm + c2c_right->stats.rmt_hitm;
533
534 return tot_hitm_left - tot_hitm_right;
535}
536
537#define STAT_FN_ENTRY(__f) \
538static int \
539__f ## _entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp, \
540 struct hist_entry *he) \
541{ \
542 struct c2c_hist_entry *c2c_he; \
543 int width = c2c_width(fmt, hpp, he->hists); \
544 \
545 c2c_he = container_of(he, struct c2c_hist_entry, he); \
546 return scnprintf(hpp->buf, hpp->size, "%*u", width, \
547 c2c_he->stats.__f); \
548}
549
550#define STAT_FN_CMP(__f) \
551static int64_t \
552__f ## _cmp(struct perf_hpp_fmt *fmt __maybe_unused, \
553 struct hist_entry *left, struct hist_entry *right) \
554{ \
555 struct c2c_hist_entry *c2c_left, *c2c_right; \
556 \
557 c2c_left = container_of(left, struct c2c_hist_entry, he); \
558 c2c_right = container_of(right, struct c2c_hist_entry, he); \
559 return c2c_left->stats.__f - c2c_right->stats.__f; \
560}
561
562#define STAT_FN(__f) \
563 STAT_FN_ENTRY(__f) \
564 STAT_FN_CMP(__f)
565
566STAT_FN(rmt_hitm)
567STAT_FN(lcl_hitm)
568STAT_FN(store)
569STAT_FN(st_l1hit)
570STAT_FN(st_l1miss)
571STAT_FN(ld_fbhit)
572STAT_FN(ld_l1hit)
573STAT_FN(ld_l2hit)
574STAT_FN(ld_llchit)
575STAT_FN(rmt_hit)
576
577static uint64_t llc_miss(struct c2c_stats *stats)
578{
579 uint64_t llcmiss;
580
581 llcmiss = stats->lcl_dram +
582 stats->rmt_dram +
583 stats->rmt_hitm +
584 stats->rmt_hit;
585
586 return llcmiss;
587}
588
589static int
590ld_llcmiss_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
591 struct hist_entry *he)
592{
593 struct c2c_hist_entry *c2c_he;
594 int width = c2c_width(fmt, hpp, he->hists);
595
596 c2c_he = container_of(he, struct c2c_hist_entry, he);
597
598 return scnprintf(hpp->buf, hpp->size, "%*lu", width,
599 llc_miss(&c2c_he->stats));
600}
601
602static int64_t
603ld_llcmiss_cmp(struct perf_hpp_fmt *fmt __maybe_unused,
604 struct hist_entry *left, struct hist_entry *right)
605{
606 struct c2c_hist_entry *c2c_left;
607 struct c2c_hist_entry *c2c_right;
608
609 c2c_left = container_of(left, struct c2c_hist_entry, he);
610 c2c_right = container_of(right, struct c2c_hist_entry, he);
611
612 return llc_miss(&c2c_left->stats) - llc_miss(&c2c_right->stats);
613}
614
615static uint64_t total_records(struct c2c_stats *stats)
616{
617 uint64_t lclmiss, ldcnt, total;
618
619 lclmiss = stats->lcl_dram +
620 stats->rmt_dram +
621 stats->rmt_hitm +
622 stats->rmt_hit;
623
624 ldcnt = lclmiss +
625 stats->ld_fbhit +
626 stats->ld_l1hit +
627 stats->ld_l2hit +
628 stats->ld_llchit +
629 stats->lcl_hitm;
630
631 total = ldcnt +
632 stats->st_l1hit +
633 stats->st_l1miss;
634
635 return total;
636}
637
638static int
639tot_recs_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
640 struct hist_entry *he)
641{
642 struct c2c_hist_entry *c2c_he;
643 int width = c2c_width(fmt, hpp, he->hists);
644 uint64_t tot_recs;
645
646 c2c_he = container_of(he, struct c2c_hist_entry, he);
647 tot_recs = total_records(&c2c_he->stats);
648
649 return scnprintf(hpp->buf, hpp->size, "%*" PRIu64, width, tot_recs);
650}
651
652static int64_t
653tot_recs_cmp(struct perf_hpp_fmt *fmt __maybe_unused,
654 struct hist_entry *left, struct hist_entry *right)
655{
656 struct c2c_hist_entry *c2c_left;
657 struct c2c_hist_entry *c2c_right;
658 uint64_t tot_recs_left;
659 uint64_t tot_recs_right;
660
661 c2c_left = container_of(left, struct c2c_hist_entry, he);
662 c2c_right = container_of(right, struct c2c_hist_entry, he);
663
664 tot_recs_left = total_records(&c2c_left->stats);
665 tot_recs_right = total_records(&c2c_right->stats);
666
667 return tot_recs_left - tot_recs_right;
668}
669
670static uint64_t total_loads(struct c2c_stats *stats)
671{
672 uint64_t lclmiss, ldcnt;
673
674 lclmiss = stats->lcl_dram +
675 stats->rmt_dram +
676 stats->rmt_hitm +
677 stats->rmt_hit;
678
679 ldcnt = lclmiss +
680 stats->ld_fbhit +
681 stats->ld_l1hit +
682 stats->ld_l2hit +
683 stats->ld_llchit +
684 stats->lcl_hitm;
685
686 return ldcnt;
687}
688
689static int
690tot_loads_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
691 struct hist_entry *he)
692{
693 struct c2c_hist_entry *c2c_he;
694 int width = c2c_width(fmt, hpp, he->hists);
695 uint64_t tot_recs;
696
697 c2c_he = container_of(he, struct c2c_hist_entry, he);
698 tot_recs = total_loads(&c2c_he->stats);
699
700 return scnprintf(hpp->buf, hpp->size, "%*" PRIu64, width, tot_recs);
701}
702
703static int64_t
704tot_loads_cmp(struct perf_hpp_fmt *fmt __maybe_unused,
705 struct hist_entry *left, struct hist_entry *right)
706{
707 struct c2c_hist_entry *c2c_left;
708 struct c2c_hist_entry *c2c_right;
709 uint64_t tot_recs_left;
710 uint64_t tot_recs_right;
711
712 c2c_left = container_of(left, struct c2c_hist_entry, he);
713 c2c_right = container_of(right, struct c2c_hist_entry, he);
714
715 tot_recs_left = total_loads(&c2c_left->stats);
716 tot_recs_right = total_loads(&c2c_right->stats);
717
718 return tot_recs_left - tot_recs_right;
719}
720
721typedef double (get_percent_cb)(struct c2c_hist_entry *);
722
723static int
724percent_color(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
725 struct hist_entry *he, get_percent_cb get_percent)
726{
727 struct c2c_hist_entry *c2c_he;
728 int width = c2c_width(fmt, hpp, he->hists);
729 double per;
730
731 c2c_he = container_of(he, struct c2c_hist_entry, he);
732 per = get_percent(c2c_he);
733
734#ifdef HAVE_SLANG_SUPPORT
735 if (use_browser)
736 return __hpp__slsmg_color_printf(hpp, "%*.2f%%", width - 1, per);
737#endif
738 return hpp_color_scnprintf(hpp, "%*.2f%%", width - 1, per);
739}
740
741static double percent_hitm(struct c2c_hist_entry *c2c_he)
742{
743 struct c2c_hists *hists;
744 struct c2c_stats *stats;
745 struct c2c_stats *total;
746 int tot = 0, st = 0;
747 double p;
748
749 hists = container_of(c2c_he->he.hists, struct c2c_hists, hists);
750 stats = &c2c_he->stats;
751 total = &hists->stats;
752
753 switch (c2c.display) {
754 case DISPLAY_RMT:
755 st = stats->rmt_hitm;
756 tot = total->rmt_hitm;
757 break;
758 case DISPLAY_LCL:
759 st = stats->lcl_hitm;
760 tot = total->lcl_hitm;
761 break;
762 case DISPLAY_TOT:
763 st = stats->tot_hitm;
764 tot = total->tot_hitm;
765 default:
766 break;
767 }
768
769 p = tot ? (double) st / tot : 0;
770
771 return 100 * p;
772}
773
774#define PERC_STR(__s, __v) \
775({ \
776 scnprintf(__s, sizeof(__s), "%.2F%%", __v); \
777 __s; \
778})
779
780static int
781percent_hitm_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
782 struct hist_entry *he)
783{
784 struct c2c_hist_entry *c2c_he;
785 int width = c2c_width(fmt, hpp, he->hists);
786 char buf[10];
787 double per;
788
789 c2c_he = container_of(he, struct c2c_hist_entry, he);
790 per = percent_hitm(c2c_he);
791 return scnprintf(hpp->buf, hpp->size, "%*s", width, PERC_STR(buf, per));
792}
793
794static int
795percent_hitm_color(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
796 struct hist_entry *he)
797{
798 return percent_color(fmt, hpp, he, percent_hitm);
799}
800
801static int64_t
802percent_hitm_cmp(struct perf_hpp_fmt *fmt __maybe_unused,
803 struct hist_entry *left, struct hist_entry *right)
804{
805 struct c2c_hist_entry *c2c_left;
806 struct c2c_hist_entry *c2c_right;
807 double per_left;
808 double per_right;
809
810 c2c_left = container_of(left, struct c2c_hist_entry, he);
811 c2c_right = container_of(right, struct c2c_hist_entry, he);
812
813 per_left = percent_hitm(c2c_left);
814 per_right = percent_hitm(c2c_right);
815
816 return per_left - per_right;
817}
818
819static struct c2c_stats *he_stats(struct hist_entry *he)
820{
821 struct c2c_hist_entry *c2c_he;
822
823 c2c_he = container_of(he, struct c2c_hist_entry, he);
824 return &c2c_he->stats;
825}
826
827static struct c2c_stats *total_stats(struct hist_entry *he)
828{
829 struct c2c_hists *hists;
830
831 hists = container_of(he->hists, struct c2c_hists, hists);
832 return &hists->stats;
833}
834
835static double percent(int st, int tot)
836{
837 return tot ? 100. * (double) st / (double) tot : 0;
838}
839
840#define PERCENT(__h, __f) percent(he_stats(__h)->__f, total_stats(__h)->__f)
841
842#define PERCENT_FN(__f) \
843static double percent_ ## __f(struct c2c_hist_entry *c2c_he) \
844{ \
845 struct c2c_hists *hists; \
846 \
847 hists = container_of(c2c_he->he.hists, struct c2c_hists, hists); \
848 return percent(c2c_he->stats.__f, hists->stats.__f); \
849}
850
851PERCENT_FN(rmt_hitm)
852PERCENT_FN(lcl_hitm)
853PERCENT_FN(st_l1hit)
854PERCENT_FN(st_l1miss)
855
856static int
857percent_rmt_hitm_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
858 struct hist_entry *he)
859{
860 int width = c2c_width(fmt, hpp, he->hists);
861 double per = PERCENT(he, rmt_hitm);
862 char buf[10];
863
864 return scnprintf(hpp->buf, hpp->size, "%*s", width, PERC_STR(buf, per));
865}
866
867static int
868percent_rmt_hitm_color(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
869 struct hist_entry *he)
870{
871 return percent_color(fmt, hpp, he, percent_rmt_hitm);
872}
873
874static int64_t
875percent_rmt_hitm_cmp(struct perf_hpp_fmt *fmt __maybe_unused,
876 struct hist_entry *left, struct hist_entry *right)
877{
878 double per_left;
879 double per_right;
880
881 per_left = PERCENT(left, lcl_hitm);
882 per_right = PERCENT(right, lcl_hitm);
883
884 return per_left - per_right;
885}
886
887static int
888percent_lcl_hitm_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
889 struct hist_entry *he)
890{
891 int width = c2c_width(fmt, hpp, he->hists);
892 double per = PERCENT(he, lcl_hitm);
893 char buf[10];
894
895 return scnprintf(hpp->buf, hpp->size, "%*s", width, PERC_STR(buf, per));
896}
897
898static int
899percent_lcl_hitm_color(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
900 struct hist_entry *he)
901{
902 return percent_color(fmt, hpp, he, percent_lcl_hitm);
903}
904
905static int64_t
906percent_lcl_hitm_cmp(struct perf_hpp_fmt *fmt __maybe_unused,
907 struct hist_entry *left, struct hist_entry *right)
908{
909 double per_left;
910 double per_right;
911
912 per_left = PERCENT(left, lcl_hitm);
913 per_right = PERCENT(right, lcl_hitm);
914
915 return per_left - per_right;
916}
917
918static int
919percent_stores_l1hit_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
920 struct hist_entry *he)
921{
922 int width = c2c_width(fmt, hpp, he->hists);
923 double per = PERCENT(he, st_l1hit);
924 char buf[10];
925
926 return scnprintf(hpp->buf, hpp->size, "%*s", width, PERC_STR(buf, per));
927}
928
929static int
930percent_stores_l1hit_color(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
931 struct hist_entry *he)
932{
933 return percent_color(fmt, hpp, he, percent_st_l1hit);
934}
935
936static int64_t
937percent_stores_l1hit_cmp(struct perf_hpp_fmt *fmt __maybe_unused,
938 struct hist_entry *left, struct hist_entry *right)
939{
940 double per_left;
941 double per_right;
942
943 per_left = PERCENT(left, st_l1hit);
944 per_right = PERCENT(right, st_l1hit);
945
946 return per_left - per_right;
947}
948
949static int
950percent_stores_l1miss_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
951 struct hist_entry *he)
952{
953 int width = c2c_width(fmt, hpp, he->hists);
954 double per = PERCENT(he, st_l1miss);
955 char buf[10];
956
957 return scnprintf(hpp->buf, hpp->size, "%*s", width, PERC_STR(buf, per));
958}
959
960static int
961percent_stores_l1miss_color(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
962 struct hist_entry *he)
963{
964 return percent_color(fmt, hpp, he, percent_st_l1miss);
965}
966
967static int64_t
968percent_stores_l1miss_cmp(struct perf_hpp_fmt *fmt __maybe_unused,
969 struct hist_entry *left, struct hist_entry *right)
970{
971 double per_left;
972 double per_right;
973
974 per_left = PERCENT(left, st_l1miss);
975 per_right = PERCENT(right, st_l1miss);
976
977 return per_left - per_right;
978}
979
980STAT_FN(lcl_dram)
981STAT_FN(rmt_dram)
982
983static int
984pid_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
985 struct hist_entry *he)
986{
987 int width = c2c_width(fmt, hpp, he->hists);
988
989 return scnprintf(hpp->buf, hpp->size, "%*d", width, he->thread->pid_);
990}
991
992static int64_t
993pid_cmp(struct perf_hpp_fmt *fmt __maybe_unused,
994 struct hist_entry *left, struct hist_entry *right)
995{
996 return left->thread->pid_ - right->thread->pid_;
997}
998
999static int64_t
1000empty_cmp(struct perf_hpp_fmt *fmt __maybe_unused,
1001 struct hist_entry *left __maybe_unused,
1002 struct hist_entry *right __maybe_unused)
1003{
1004 return 0;
1005}
1006
1007static int
1008node_entry(struct perf_hpp_fmt *fmt __maybe_unused, struct perf_hpp *hpp,
1009 struct hist_entry *he)
1010{
1011 struct c2c_hist_entry *c2c_he;
1012 bool first = true;
1013 int node;
1014 int ret = 0;
1015
1016 c2c_he = container_of(he, struct c2c_hist_entry, he);
1017
1018 for (node = 0; node < c2c.nodes_cnt; node++) {
1019 DECLARE_BITMAP(set, c2c.cpus_cnt);
1020
1021 bitmap_zero(set, c2c.cpus_cnt);
1022 bitmap_and(set, c2c_he->cpuset, c2c.nodes[node], c2c.cpus_cnt);
1023
1024 if (!bitmap_weight(set, c2c.cpus_cnt)) {
1025 if (c2c.node_info == 1) {
1026 ret = scnprintf(hpp->buf, hpp->size, "%21s", " ");
1027 advance_hpp(hpp, ret);
1028 }
1029 continue;
1030 }
1031
1032 if (!first) {
1033 ret = scnprintf(hpp->buf, hpp->size, " ");
1034 advance_hpp(hpp, ret);
1035 }
1036
1037 switch (c2c.node_info) {
1038 case 0:
1039 ret = scnprintf(hpp->buf, hpp->size, "%2d", node);
1040 advance_hpp(hpp, ret);
1041 break;
1042 case 1:
1043 {
1044 int num = bitmap_weight(c2c_he->cpuset, c2c.cpus_cnt);
1045 struct c2c_stats *stats = &c2c_he->node_stats[node];
1046
1047 ret = scnprintf(hpp->buf, hpp->size, "%2d{%2d ", node, num);
1048 advance_hpp(hpp, ret);
1049
1050 #define DISPLAY_HITM(__h) \
1051 if (c2c_he->stats.__h> 0) { \
1052 ret = scnprintf(hpp->buf, hpp->size, "%5.1f%% ", \
1053 percent(stats->__h, c2c_he->stats.__h));\
1054 } else { \
1055 ret = scnprintf(hpp->buf, hpp->size, "%6s ", "n/a"); \
1056 }
1057
1058 switch (c2c.display) {
1059 case DISPLAY_RMT:
1060 DISPLAY_HITM(rmt_hitm);
1061 break;
1062 case DISPLAY_LCL:
1063 DISPLAY_HITM(lcl_hitm);
1064 break;
1065 case DISPLAY_TOT:
1066 DISPLAY_HITM(tot_hitm);
1067 default:
1068 break;
1069 }
1070
1071 #undef DISPLAY_HITM
1072
1073 advance_hpp(hpp, ret);
1074
1075 if (c2c_he->stats.store > 0) {
1076 ret = scnprintf(hpp->buf, hpp->size, "%5.1f%%}",
1077 percent(stats->store, c2c_he->stats.store));
1078 } else {
1079 ret = scnprintf(hpp->buf, hpp->size, "%6s}", "n/a");
1080 }
1081
1082 advance_hpp(hpp, ret);
1083 break;
1084 }
1085 case 2:
1086 ret = scnprintf(hpp->buf, hpp->size, "%2d{", node);
1087 advance_hpp(hpp, ret);
1088
1089 ret = bitmap_scnprintf(set, c2c.cpus_cnt, hpp->buf, hpp->size);
1090 advance_hpp(hpp, ret);
1091
1092 ret = scnprintf(hpp->buf, hpp->size, "}");
1093 advance_hpp(hpp, ret);
1094 break;
1095 default:
1096 break;
1097 }
1098
1099 first = false;
1100 }
1101
1102 return 0;
1103}
1104
1105static int
1106mean_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
1107 struct hist_entry *he, double mean)
1108{
1109 int width = c2c_width(fmt, hpp, he->hists);
1110 char buf[10];
1111
1112 scnprintf(buf, 10, "%6.0f", mean);
1113 return scnprintf(hpp->buf, hpp->size, "%*s", width, buf);
1114}
1115
1116#define MEAN_ENTRY(__func, __val) \
1117static int \
1118__func(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp, struct hist_entry *he) \
1119{ \
1120 struct c2c_hist_entry *c2c_he; \
1121 c2c_he = container_of(he, struct c2c_hist_entry, he); \
1122 return mean_entry(fmt, hpp, he, avg_stats(&c2c_he->cstats.__val)); \
1123}
1124
1125MEAN_ENTRY(mean_rmt_entry, rmt_hitm);
1126MEAN_ENTRY(mean_lcl_entry, lcl_hitm);
1127MEAN_ENTRY(mean_load_entry, load);
1128
1129static int
1130cpucnt_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
1131 struct hist_entry *he)
1132{
1133 struct c2c_hist_entry *c2c_he;
1134 int width = c2c_width(fmt, hpp, he->hists);
1135 char buf[10];
1136
1137 c2c_he = container_of(he, struct c2c_hist_entry, he);
1138
1139 scnprintf(buf, 10, "%d", bitmap_weight(c2c_he->cpuset, c2c.cpus_cnt));
1140 return scnprintf(hpp->buf, hpp->size, "%*s", width, buf);
1141}
1142
1143static int
1144cl_idx_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
1145 struct hist_entry *he)
1146{
1147 struct c2c_hist_entry *c2c_he;
1148 int width = c2c_width(fmt, hpp, he->hists);
1149 char buf[10];
1150
1151 c2c_he = container_of(he, struct c2c_hist_entry, he);
1152
1153 scnprintf(buf, 10, "%u", c2c_he->cacheline_idx);
1154 return scnprintf(hpp->buf, hpp->size, "%*s", width, buf);
1155}
1156
1157static int
1158cl_idx_empty_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
1159 struct hist_entry *he)
1160{
1161 int width = c2c_width(fmt, hpp, he->hists);
1162
1163 return scnprintf(hpp->buf, hpp->size, "%*s", width, "");
1164}
1165
1166#define HEADER_LOW(__h) \
1167 { \
1168 .line[1] = { \
1169 .text = __h, \
1170 }, \
1171 }
1172
1173#define HEADER_BOTH(__h0, __h1) \
1174 { \
1175 .line[0] = { \
1176 .text = __h0, \
1177 }, \
1178 .line[1] = { \
1179 .text = __h1, \
1180 }, \
1181 }
1182
1183#define HEADER_SPAN(__h0, __h1, __s) \
1184 { \
1185 .line[0] = { \
1186 .text = __h0, \
1187 .span = __s, \
1188 }, \
1189 .line[1] = { \
1190 .text = __h1, \
1191 }, \
1192 }
1193
1194#define HEADER_SPAN_LOW(__h) \
1195 { \
1196 .line[1] = { \
1197 .text = __h, \
1198 }, \
1199 }
1200
1201static struct c2c_dimension dim_dcacheline = {
1202 .header = HEADER_LOW("Cacheline"),
1203 .name = "dcacheline",
1204 .cmp = dcacheline_cmp,
1205 .entry = dcacheline_entry,
1206 .width = 18,
1207};
1208
1209static struct c2c_header header_offset_tui = HEADER_LOW("Off");
1210
1211static struct c2c_dimension dim_offset = {
1212 .header = HEADER_BOTH("Data address", "Offset"),
1213 .name = "offset",
1214 .cmp = offset_cmp,
1215 .entry = offset_entry,
1216 .width = 18,
1217};
1218
1219static struct c2c_dimension dim_iaddr = {
1220 .header = HEADER_LOW("Code address"),
1221 .name = "iaddr",
1222 .cmp = iaddr_cmp,
1223 .entry = iaddr_entry,
1224 .width = 18,
1225};
1226
1227static struct c2c_dimension dim_tot_hitm = {
1228 .header = HEADER_SPAN("----- LLC Load Hitm -----", "Total", 2),
1229 .name = "tot_hitm",
1230 .cmp = tot_hitm_cmp,
1231 .entry = tot_hitm_entry,
1232 .width = 7,
1233};
1234
1235static struct c2c_dimension dim_lcl_hitm = {
1236 .header = HEADER_SPAN_LOW("Lcl"),
1237 .name = "lcl_hitm",
1238 .cmp = lcl_hitm_cmp,
1239 .entry = lcl_hitm_entry,
1240 .width = 7,
1241};
1242
1243static struct c2c_dimension dim_rmt_hitm = {
1244 .header = HEADER_SPAN_LOW("Rmt"),
1245 .name = "rmt_hitm",
1246 .cmp = rmt_hitm_cmp,
1247 .entry = rmt_hitm_entry,
1248 .width = 7,
1249};
1250
1251static struct c2c_dimension dim_cl_rmt_hitm = {
1252 .header = HEADER_SPAN("----- HITM -----", "Rmt", 1),
1253 .name = "cl_rmt_hitm",
1254 .cmp = rmt_hitm_cmp,
1255 .entry = rmt_hitm_entry,
1256 .width = 7,
1257};
1258
1259static struct c2c_dimension dim_cl_lcl_hitm = {
1260 .header = HEADER_SPAN_LOW("Lcl"),
1261 .name = "cl_lcl_hitm",
1262 .cmp = lcl_hitm_cmp,
1263 .entry = lcl_hitm_entry,
1264 .width = 7,
1265};
1266
1267static struct c2c_dimension dim_stores = {
1268 .header = HEADER_SPAN("---- Store Reference ----", "Total", 2),
1269 .name = "stores",
1270 .cmp = store_cmp,
1271 .entry = store_entry,
1272 .width = 7,
1273};
1274
1275static struct c2c_dimension dim_stores_l1hit = {
1276 .header = HEADER_SPAN_LOW("L1Hit"),
1277 .name = "stores_l1hit",
1278 .cmp = st_l1hit_cmp,
1279 .entry = st_l1hit_entry,
1280 .width = 7,
1281};
1282
1283static struct c2c_dimension dim_stores_l1miss = {
1284 .header = HEADER_SPAN_LOW("L1Miss"),
1285 .name = "stores_l1miss",
1286 .cmp = st_l1miss_cmp,
1287 .entry = st_l1miss_entry,
1288 .width = 7,
1289};
1290
1291static struct c2c_dimension dim_cl_stores_l1hit = {
1292 .header = HEADER_SPAN("-- Store Refs --", "L1 Hit", 1),
1293 .name = "cl_stores_l1hit",
1294 .cmp = st_l1hit_cmp,
1295 .entry = st_l1hit_entry,
1296 .width = 7,
1297};
1298
1299static struct c2c_dimension dim_cl_stores_l1miss = {
1300 .header = HEADER_SPAN_LOW("L1 Miss"),
1301 .name = "cl_stores_l1miss",
1302 .cmp = st_l1miss_cmp,
1303 .entry = st_l1miss_entry,
1304 .width = 7,
1305};
1306
1307static struct c2c_dimension dim_ld_fbhit = {
1308 .header = HEADER_SPAN("----- Core Load Hit -----", "FB", 2),
1309 .name = "ld_fbhit",
1310 .cmp = ld_fbhit_cmp,
1311 .entry = ld_fbhit_entry,
1312 .width = 7,
1313};
1314
1315static struct c2c_dimension dim_ld_l1hit = {
1316 .header = HEADER_SPAN_LOW("L1"),
1317 .name = "ld_l1hit",
1318 .cmp = ld_l1hit_cmp,
1319 .entry = ld_l1hit_entry,
1320 .width = 7,
1321};
1322
1323static struct c2c_dimension dim_ld_l2hit = {
1324 .header = HEADER_SPAN_LOW("L2"),
1325 .name = "ld_l2hit",
1326 .cmp = ld_l2hit_cmp,
1327 .entry = ld_l2hit_entry,
1328 .width = 7,
1329};
1330
1331static struct c2c_dimension dim_ld_llchit = {
1332 .header = HEADER_SPAN("-- LLC Load Hit --", "Llc", 1),
1333 .name = "ld_lclhit",
1334 .cmp = ld_llchit_cmp,
1335 .entry = ld_llchit_entry,
1336 .width = 8,
1337};
1338
1339static struct c2c_dimension dim_ld_rmthit = {
1340 .header = HEADER_SPAN_LOW("Rmt"),
1341 .name = "ld_rmthit",
1342 .cmp = rmt_hit_cmp,
1343 .entry = rmt_hit_entry,
1344 .width = 8,
1345};
1346
1347static struct c2c_dimension dim_ld_llcmiss = {
1348 .header = HEADER_BOTH("LLC", "Ld Miss"),
1349 .name = "ld_llcmiss",
1350 .cmp = ld_llcmiss_cmp,
1351 .entry = ld_llcmiss_entry,
1352 .width = 7,
1353};
1354
1355static struct c2c_dimension dim_tot_recs = {
1356 .header = HEADER_BOTH("Total", "records"),
1357 .name = "tot_recs",
1358 .cmp = tot_recs_cmp,
1359 .entry = tot_recs_entry,
1360 .width = 7,
1361};
1362
1363static struct c2c_dimension dim_tot_loads = {
1364 .header = HEADER_BOTH("Total", "Loads"),
1365 .name = "tot_loads",
1366 .cmp = tot_loads_cmp,
1367 .entry = tot_loads_entry,
1368 .width = 7,
1369};
1370
1371static struct c2c_header percent_hitm_header[] = {
1372 [DISPLAY_LCL] = HEADER_BOTH("Lcl", "Hitm"),
1373 [DISPLAY_RMT] = HEADER_BOTH("Rmt", "Hitm"),
1374 [DISPLAY_TOT] = HEADER_BOTH("Tot", "Hitm"),
1375};
1376
1377static struct c2c_dimension dim_percent_hitm = {
1378 .name = "percent_hitm",
1379 .cmp = percent_hitm_cmp,
1380 .entry = percent_hitm_entry,
1381 .color = percent_hitm_color,
1382 .width = 7,
1383};
1384
1385static struct c2c_dimension dim_percent_rmt_hitm = {
1386 .header = HEADER_SPAN("----- HITM -----", "Rmt", 1),
1387 .name = "percent_rmt_hitm",
1388 .cmp = percent_rmt_hitm_cmp,
1389 .entry = percent_rmt_hitm_entry,
1390 .color = percent_rmt_hitm_color,
1391 .width = 7,
1392};
1393
1394static struct c2c_dimension dim_percent_lcl_hitm = {
1395 .header = HEADER_SPAN_LOW("Lcl"),
1396 .name = "percent_lcl_hitm",
1397 .cmp = percent_lcl_hitm_cmp,
1398 .entry = percent_lcl_hitm_entry,
1399 .color = percent_lcl_hitm_color,
1400 .width = 7,
1401};
1402
1403static struct c2c_dimension dim_percent_stores_l1hit = {
1404 .header = HEADER_SPAN("-- Store Refs --", "L1 Hit", 1),
1405 .name = "percent_stores_l1hit",
1406 .cmp = percent_stores_l1hit_cmp,
1407 .entry = percent_stores_l1hit_entry,
1408 .color = percent_stores_l1hit_color,
1409 .width = 7,
1410};
1411
1412static struct c2c_dimension dim_percent_stores_l1miss = {
1413 .header = HEADER_SPAN_LOW("L1 Miss"),
1414 .name = "percent_stores_l1miss",
1415 .cmp = percent_stores_l1miss_cmp,
1416 .entry = percent_stores_l1miss_entry,
1417 .color = percent_stores_l1miss_color,
1418 .width = 7,
1419};
1420
1421static struct c2c_dimension dim_dram_lcl = {
1422 .header = HEADER_SPAN("--- Load Dram ----", "Lcl", 1),
1423 .name = "dram_lcl",
1424 .cmp = lcl_dram_cmp,
1425 .entry = lcl_dram_entry,
1426 .width = 8,
1427};
1428
1429static struct c2c_dimension dim_dram_rmt = {
1430 .header = HEADER_SPAN_LOW("Rmt"),
1431 .name = "dram_rmt",
1432 .cmp = rmt_dram_cmp,
1433 .entry = rmt_dram_entry,
1434 .width = 8,
1435};
1436
1437static struct c2c_dimension dim_pid = {
1438 .header = HEADER_LOW("Pid"),
1439 .name = "pid",
1440 .cmp = pid_cmp,
1441 .entry = pid_entry,
1442 .width = 7,
1443};
1444
1445static struct c2c_dimension dim_tid = {
1446 .header = HEADER_LOW("Tid"),
1447 .name = "tid",
1448 .se = &sort_thread,
1449};
1450
1451static struct c2c_dimension dim_symbol = {
1452 .name = "symbol",
1453 .se = &sort_sym,
1454};
1455
1456static struct c2c_dimension dim_dso = {
1457 .header = HEADER_BOTH("Shared", "Object"),
1458 .name = "dso",
1459 .se = &sort_dso,
1460};
1461
1462static struct c2c_header header_node[3] = {
1463 HEADER_LOW("Node"),
1464 HEADER_LOW("Node{cpus %hitms %stores}"),
1465 HEADER_LOW("Node{cpu list}"),
1466};
1467
1468static struct c2c_dimension dim_node = {
1469 .name = "node",
1470 .cmp = empty_cmp,
1471 .entry = node_entry,
1472 .width = 4,
1473};
1474
1475static struct c2c_dimension dim_mean_rmt = {
1476 .header = HEADER_SPAN("---------- cycles ----------", "rmt hitm", 2),
1477 .name = "mean_rmt",
1478 .cmp = empty_cmp,
1479 .entry = mean_rmt_entry,
1480 .width = 8,
1481};
1482
1483static struct c2c_dimension dim_mean_lcl = {
1484 .header = HEADER_SPAN_LOW("lcl hitm"),
1485 .name = "mean_lcl",
1486 .cmp = empty_cmp,
1487 .entry = mean_lcl_entry,
1488 .width = 8,
1489};
1490
1491static struct c2c_dimension dim_mean_load = {
1492 .header = HEADER_SPAN_LOW("load"),
1493 .name = "mean_load",
1494 .cmp = empty_cmp,
1495 .entry = mean_load_entry,
1496 .width = 8,
1497};
1498
1499static struct c2c_dimension dim_cpucnt = {
1500 .header = HEADER_BOTH("cpu", "cnt"),
1501 .name = "cpucnt",
1502 .cmp = empty_cmp,
1503 .entry = cpucnt_entry,
1504 .width = 8,
1505};
1506
1507static struct c2c_dimension dim_srcline = {
1508 .name = "cl_srcline",
1509 .se = &sort_srcline,
1510};
1511
1512static struct c2c_dimension dim_dcacheline_idx = {
1513 .header = HEADER_LOW("Index"),
1514 .name = "cl_idx",
1515 .cmp = empty_cmp,
1516 .entry = cl_idx_entry,
1517 .width = 5,
1518};
1519
1520static struct c2c_dimension dim_dcacheline_num = {
1521 .header = HEADER_LOW("Num"),
1522 .name = "cl_num",
1523 .cmp = empty_cmp,
1524 .entry = cl_idx_entry,
1525 .width = 5,
1526};
1527
1528static struct c2c_dimension dim_dcacheline_num_empty = {
1529 .header = HEADER_LOW("Num"),
1530 .name = "cl_num_empty",
1531 .cmp = empty_cmp,
1532 .entry = cl_idx_empty_entry,
1533 .width = 5,
1534};
1535
1536static struct c2c_dimension *dimensions[] = {
1537 &dim_dcacheline,
1538 &dim_offset,
1539 &dim_iaddr,
1540 &dim_tot_hitm,
1541 &dim_lcl_hitm,
1542 &dim_rmt_hitm,
1543 &dim_cl_lcl_hitm,
1544 &dim_cl_rmt_hitm,
1545 &dim_stores,
1546 &dim_stores_l1hit,
1547 &dim_stores_l1miss,
1548 &dim_cl_stores_l1hit,
1549 &dim_cl_stores_l1miss,
1550 &dim_ld_fbhit,
1551 &dim_ld_l1hit,
1552 &dim_ld_l2hit,
1553 &dim_ld_llchit,
1554 &dim_ld_rmthit,
1555 &dim_ld_llcmiss,
1556 &dim_tot_recs,
1557 &dim_tot_loads,
1558 &dim_percent_hitm,
1559 &dim_percent_rmt_hitm,
1560 &dim_percent_lcl_hitm,
1561 &dim_percent_stores_l1hit,
1562 &dim_percent_stores_l1miss,
1563 &dim_dram_lcl,
1564 &dim_dram_rmt,
1565 &dim_pid,
1566 &dim_tid,
1567 &dim_symbol,
1568 &dim_dso,
1569 &dim_node,
1570 &dim_mean_rmt,
1571 &dim_mean_lcl,
1572 &dim_mean_load,
1573 &dim_cpucnt,
1574 &dim_srcline,
1575 &dim_dcacheline_idx,
1576 &dim_dcacheline_num,
1577 &dim_dcacheline_num_empty,
1578 NULL,
1579};
1580
1581static void fmt_free(struct perf_hpp_fmt *fmt)
1582{
1583 struct c2c_fmt *c2c_fmt;
1584
1585 c2c_fmt = container_of(fmt, struct c2c_fmt, fmt);
1586 free(c2c_fmt);
1587}
1588
1589static bool fmt_equal(struct perf_hpp_fmt *a, struct perf_hpp_fmt *b)
1590{
1591 struct c2c_fmt *c2c_a = container_of(a, struct c2c_fmt, fmt);
1592 struct c2c_fmt *c2c_b = container_of(b, struct c2c_fmt, fmt);
1593
1594 return c2c_a->dim == c2c_b->dim;
1595}
1596
1597static struct c2c_dimension *get_dimension(const char *name)
1598{
1599 unsigned int i;
1600
1601 for (i = 0; dimensions[i]; i++) {
1602 struct c2c_dimension *dim = dimensions[i];
1603
1604 if (!strcmp(dim->name, name))
1605 return dim;
1606 };
1607
1608 return NULL;
1609}
1610
1611static int c2c_se_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
1612 struct hist_entry *he)
1613{
1614 struct c2c_fmt *c2c_fmt = container_of(fmt, struct c2c_fmt, fmt);
1615 struct c2c_dimension *dim = c2c_fmt->dim;
1616 size_t len = fmt->user_len;
1617
1618 if (!len) {
1619 len = hists__col_len(he->hists, dim->se->se_width_idx);
1620
1621 if (dim == &dim_symbol || dim == &dim_srcline)
1622 len = symbol_width(he->hists, dim->se);
1623 }
1624
1625 return dim->se->se_snprintf(he, hpp->buf, hpp->size, len);
1626}
1627
1628static int64_t c2c_se_cmp(struct perf_hpp_fmt *fmt,
1629 struct hist_entry *a, struct hist_entry *b)
1630{
1631 struct c2c_fmt *c2c_fmt = container_of(fmt, struct c2c_fmt, fmt);
1632 struct c2c_dimension *dim = c2c_fmt->dim;
1633
1634 return dim->se->se_cmp(a, b);
1635}
1636
1637static int64_t c2c_se_collapse(struct perf_hpp_fmt *fmt,
1638 struct hist_entry *a, struct hist_entry *b)
1639{
1640 struct c2c_fmt *c2c_fmt = container_of(fmt, struct c2c_fmt, fmt);
1641 struct c2c_dimension *dim = c2c_fmt->dim;
1642 int64_t (*collapse_fn)(struct hist_entry *, struct hist_entry *);
1643
1644 collapse_fn = dim->se->se_collapse ?: dim->se->se_cmp;
1645 return collapse_fn(a, b);
1646}
1647
1648static struct c2c_fmt *get_format(const char *name)
1649{
1650 struct c2c_dimension *dim = get_dimension(name);
1651 struct c2c_fmt *c2c_fmt;
1652 struct perf_hpp_fmt *fmt;
1653
1654 if (!dim)
1655 return NULL;
1656
1657 c2c_fmt = zalloc(sizeof(*c2c_fmt));
1658 if (!c2c_fmt)
1659 return NULL;
1660
1661 c2c_fmt->dim = dim;
1662
1663 fmt = &c2c_fmt->fmt;
1664 INIT_LIST_HEAD(&fmt->list);
1665 INIT_LIST_HEAD(&fmt->sort_list);
1666
1667 fmt->cmp = dim->se ? c2c_se_cmp : dim->cmp;
1668 fmt->sort = dim->se ? c2c_se_cmp : dim->cmp;
1669 fmt->color = dim->se ? NULL : dim->color;
1670 fmt->entry = dim->se ? c2c_se_entry : dim->entry;
1671 fmt->header = c2c_header;
1672 fmt->width = c2c_width;
1673 fmt->collapse = dim->se ? c2c_se_collapse : dim->cmp;
1674 fmt->equal = fmt_equal;
1675 fmt->free = fmt_free;
1676
1677 return c2c_fmt;
1678}
1679
1680static int c2c_hists__init_output(struct perf_hpp_list *hpp_list, char *name)
1681{
1682 struct c2c_fmt *c2c_fmt = get_format(name);
1683
1684 if (!c2c_fmt) {
1685 reset_dimensions();
1686 return output_field_add(hpp_list, name);
1687 }
1688
1689 perf_hpp_list__column_register(hpp_list, &c2c_fmt->fmt);
1690 return 0;
1691}
1692
1693static int c2c_hists__init_sort(struct perf_hpp_list *hpp_list, char *name)
1694{
1695 struct c2c_fmt *c2c_fmt = get_format(name);
1696 struct c2c_dimension *dim;
1697
1698 if (!c2c_fmt) {
1699 reset_dimensions();
1700 return sort_dimension__add(hpp_list, name, NULL, 0);
1701 }
1702
1703 dim = c2c_fmt->dim;
1704 if (dim == &dim_dso)
1705 hpp_list->dso = 1;
1706
1707 perf_hpp_list__register_sort_field(hpp_list, &c2c_fmt->fmt);
1708 return 0;
1709}
1710
1711#define PARSE_LIST(_list, _fn) \
1712 do { \
1713 char *tmp, *tok; \
1714 ret = 0; \
1715 \
1716 if (!_list) \
1717 break; \
1718 \
1719 for (tok = strtok_r((char *)_list, ", ", &tmp); \
1720 tok; tok = strtok_r(NULL, ", ", &tmp)) { \
1721 ret = _fn(hpp_list, tok); \
1722 if (ret == -EINVAL) { \
1723 error("Invalid --fields key: `%s'", tok); \
1724 break; \
1725 } else if (ret == -ESRCH) { \
1726 error("Unknown --fields key: `%s'", tok); \
1727 break; \
1728 } \
1729 } \
1730 } while (0)
1731
1732static int hpp_list__parse(struct perf_hpp_list *hpp_list,
1733 const char *output_,
1734 const char *sort_)
1735{
1736 char *output = output_ ? strdup(output_) : NULL;
1737 char *sort = sort_ ? strdup(sort_) : NULL;
1738 int ret;
1739
1740 PARSE_LIST(output, c2c_hists__init_output);
1741 PARSE_LIST(sort, c2c_hists__init_sort);
1742
1743 /* copy sort keys to output fields */
1744 perf_hpp__setup_output_field(hpp_list);
1745
1746 /*
1747 * We dont need other sorting keys other than those
1748 * we already specified. It also really slows down
1749 * the processing a lot with big number of output
1750 * fields, so switching this off for c2c.
1751 */
1752
1753#if 0
1754 /* and then copy output fields to sort keys */
1755 perf_hpp__append_sort_keys(&hists->list);
1756#endif
1757
1758 free(output);
1759 free(sort);
1760 return ret;
1761}
1762
1763static int c2c_hists__init(struct c2c_hists *hists,
1764 const char *sort,
1765 int nr_header_lines)
1766{
1767 __hists__init(&hists->hists, &hists->list);
1768
1769 /*
1770 * Initialize only with sort fields, we need to resort
1771 * later anyway, and that's where we add output fields
1772 * as well.
1773 */
1774 perf_hpp_list__init(&hists->list);
1775
1776 /* Overload number of header lines.*/
1777 hists->list.nr_header_lines = nr_header_lines;
1778
1779 return hpp_list__parse(&hists->list, NULL, sort);
1780}
1781
1782static int c2c_hists__reinit(struct c2c_hists *c2c_hists,
1783 const char *output,
1784 const char *sort)
1785{
1786 perf_hpp__reset_output_field(&c2c_hists->list);
1787 return hpp_list__parse(&c2c_hists->list, output, sort);
1788}
1789
1790#define DISPLAY_LINE_LIMIT 0.0005
1791
1792static bool he__display(struct hist_entry *he, struct c2c_stats *stats)
1793{
1794 struct c2c_hist_entry *c2c_he;
1795 double ld_dist;
1796
1797 if (c2c.show_all)
1798 return true;
1799
1800 c2c_he = container_of(he, struct c2c_hist_entry, he);
1801
1802#define FILTER_HITM(__h) \
1803 if (stats->__h) { \
1804 ld_dist = ((double)c2c_he->stats.__h / stats->__h); \
1805 if (ld_dist < DISPLAY_LINE_LIMIT) \
1806 he->filtered = HIST_FILTER__C2C; \
1807 } else { \
1808 he->filtered = HIST_FILTER__C2C; \
1809 }
1810
1811 switch (c2c.display) {
1812 case DISPLAY_LCL:
1813 FILTER_HITM(lcl_hitm);
1814 break;
1815 case DISPLAY_RMT:
1816 FILTER_HITM(rmt_hitm);
1817 break;
1818 case DISPLAY_TOT:
1819 FILTER_HITM(tot_hitm);
1820 default:
1821 break;
1822 };
1823
1824#undef FILTER_HITM
1825
1826 return he->filtered == 0;
1827}
1828
1829static inline int valid_hitm_or_store(struct hist_entry *he)
1830{
1831 struct c2c_hist_entry *c2c_he;
1832 bool has_hitm;
1833
1834 c2c_he = container_of(he, struct c2c_hist_entry, he);
1835 has_hitm = c2c.display == DISPLAY_TOT ? c2c_he->stats.tot_hitm :
1836 c2c.display == DISPLAY_LCL ? c2c_he->stats.lcl_hitm :
1837 c2c_he->stats.rmt_hitm;
1838 return has_hitm || c2c_he->stats.store;
1839}
1840
1841static void calc_width(struct hist_entry *he)
1842{
1843 struct c2c_hists *c2c_hists;
1844
1845 c2c_hists = container_of(he->hists, struct c2c_hists, hists);
1846 hists__calc_col_len(&c2c_hists->hists, he);
1847}
1848
1849static int filter_cb(struct hist_entry *he)
1850{
1851 if (c2c.show_src && !he->srcline)
1852 he->srcline = hist_entry__get_srcline(he);
1853
1854 calc_width(he);
1855
1856 if (!valid_hitm_or_store(he))
1857 he->filtered = HIST_FILTER__C2C;
1858
1859 return 0;
1860}
1861
1862static int resort_cl_cb(struct hist_entry *he)
1863{
1864 struct c2c_hist_entry *c2c_he;
1865 struct c2c_hists *c2c_hists;
1866 bool display = he__display(he, &c2c.hitm_stats);
1867
1868 c2c_he = container_of(he, struct c2c_hist_entry, he);
1869 c2c_hists = c2c_he->hists;
1870
1871 calc_width(he);
1872
1873 if (display && c2c_hists) {
1874 static unsigned int idx;
1875
1876 c2c_he->cacheline_idx = idx++;
1877
1878 c2c_hists__reinit(c2c_hists, c2c.cl_output, c2c.cl_resort);
1879
1880 hists__collapse_resort(&c2c_hists->hists, NULL);
1881 hists__output_resort_cb(&c2c_hists->hists, NULL, filter_cb);
1882 }
1883
1884 return 0;
1885}
1886
1887static void setup_nodes_header(void)
1888{
1889 dim_node.header = header_node[c2c.node_info];
1890}
1891
1892static int setup_nodes(struct perf_session *session)
1893{
1894 struct numa_node *n;
1895 unsigned long **nodes;
1896 int node, cpu;
1897 int *cpu2node;
1898
1899 if (c2c.node_info > 2)
1900 c2c.node_info = 2;
1901
1902 c2c.nodes_cnt = session->header.env.nr_numa_nodes;
1903 c2c.cpus_cnt = session->header.env.nr_cpus_online;
1904
1905 n = session->header.env.numa_nodes;
1906 if (!n)
1907 return -EINVAL;
1908
1909 nodes = zalloc(sizeof(unsigned long *) * c2c.nodes_cnt);
1910 if (!nodes)
1911 return -ENOMEM;
1912
1913 c2c.nodes = nodes;
1914
1915 cpu2node = zalloc(sizeof(int) * c2c.cpus_cnt);
1916 if (!cpu2node)
1917 return -ENOMEM;
1918
1919 for (cpu = 0; cpu < c2c.cpus_cnt; cpu++)
1920 cpu2node[cpu] = -1;
1921
1922 c2c.cpu2node = cpu2node;
1923
1924 for (node = 0; node < c2c.nodes_cnt; node++) {
1925 struct cpu_map *map = n[node].map;
1926 unsigned long *set;
1927
1928 set = bitmap_alloc(c2c.cpus_cnt);
1929 if (!set)
1930 return -ENOMEM;
1931
1932 for (cpu = 0; cpu < map->nr; cpu++) {
1933 set_bit(map->map[cpu], set);
1934
1935 if (WARN_ONCE(cpu2node[map->map[cpu]] != -1, "node/cpu topology bug"))
1936 return -EINVAL;
1937
1938 cpu2node[map->map[cpu]] = node;
1939 }
1940
1941 nodes[node] = set;
1942 }
1943
1944 setup_nodes_header();
1945 return 0;
1946}
1947
1948#define HAS_HITMS(__h) ((__h)->stats.lcl_hitm || (__h)->stats.rmt_hitm)
1949
1950static int resort_hitm_cb(struct hist_entry *he)
1951{
1952 struct c2c_hist_entry *c2c_he;
1953 c2c_he = container_of(he, struct c2c_hist_entry, he);
1954
1955 if (HAS_HITMS(c2c_he)) {
1956 c2c.shared_clines++;
1957 c2c_add_stats(&c2c.hitm_stats, &c2c_he->stats);
1958 }
1959
1960 return 0;
1961}
1962
1963static int hists__iterate_cb(struct hists *hists, hists__resort_cb_t cb)
1964{
1965 struct rb_node *next = rb_first(&hists->entries);
1966 int ret = 0;
1967
1968 while (next) {
1969 struct hist_entry *he;
1970
1971 he = rb_entry(next, struct hist_entry, rb_node);
1972 ret = cb(he);
1973 if (ret)
1974 break;
1975 next = rb_next(&he->rb_node);
1976 }
1977
1978 return ret;
1979}
1980
1981static void print_c2c__display_stats(FILE *out)
1982{
1983 int llc_misses;
1984 struct c2c_stats *stats = &c2c.hists.stats;
1985
1986 llc_misses = stats->lcl_dram +
1987 stats->rmt_dram +
1988 stats->rmt_hit +
1989 stats->rmt_hitm;
1990
1991 fprintf(out, "=================================================\n");
1992 fprintf(out, " Trace Event Information \n");
1993 fprintf(out, "=================================================\n");
1994 fprintf(out, " Total records : %10d\n", stats->nr_entries);
1995 fprintf(out, " Locked Load/Store Operations : %10d\n", stats->locks);
1996 fprintf(out, " Load Operations : %10d\n", stats->load);
1997 fprintf(out, " Loads - uncacheable : %10d\n", stats->ld_uncache);
1998 fprintf(out, " Loads - IO : %10d\n", stats->ld_io);
1999 fprintf(out, " Loads - Miss : %10d\n", stats->ld_miss);
2000 fprintf(out, " Loads - no mapping : %10d\n", stats->ld_noadrs);
2001 fprintf(out, " Load Fill Buffer Hit : %10d\n", stats->ld_fbhit);
2002 fprintf(out, " Load L1D hit : %10d\n", stats->ld_l1hit);
2003 fprintf(out, " Load L2D hit : %10d\n", stats->ld_l2hit);
2004 fprintf(out, " Load LLC hit : %10d\n", stats->ld_llchit + stats->lcl_hitm);
2005 fprintf(out, " Load Local HITM : %10d\n", stats->lcl_hitm);
2006 fprintf(out, " Load Remote HITM : %10d\n", stats->rmt_hitm);
2007 fprintf(out, " Load Remote HIT : %10d\n", stats->rmt_hit);
2008 fprintf(out, " Load Local DRAM : %10d\n", stats->lcl_dram);
2009 fprintf(out, " Load Remote DRAM : %10d\n", stats->rmt_dram);
2010 fprintf(out, " Load MESI State Exclusive : %10d\n", stats->ld_excl);
2011 fprintf(out, " Load MESI State Shared : %10d\n", stats->ld_shared);
2012 fprintf(out, " Load LLC Misses : %10d\n", llc_misses);
2013 fprintf(out, " LLC Misses to Local DRAM : %10.1f%%\n", ((double)stats->lcl_dram/(double)llc_misses) * 100.);
2014 fprintf(out, " LLC Misses to Remote DRAM : %10.1f%%\n", ((double)stats->rmt_dram/(double)llc_misses) * 100.);
2015 fprintf(out, " LLC Misses to Remote cache (HIT) : %10.1f%%\n", ((double)stats->rmt_hit /(double)llc_misses) * 100.);
2016 fprintf(out, " LLC Misses to Remote cache (HITM) : %10.1f%%\n", ((double)stats->rmt_hitm/(double)llc_misses) * 100.);
2017 fprintf(out, " Store Operations : %10d\n", stats->store);
2018 fprintf(out, " Store - uncacheable : %10d\n", stats->st_uncache);
2019 fprintf(out, " Store - no mapping : %10d\n", stats->st_noadrs);
2020 fprintf(out, " Store L1D Hit : %10d\n", stats->st_l1hit);
2021 fprintf(out, " Store L1D Miss : %10d\n", stats->st_l1miss);
2022 fprintf(out, " No Page Map Rejects : %10d\n", stats->nomap);
2023 fprintf(out, " Unable to parse data source : %10d\n", stats->noparse);
2024}
2025
2026static void print_shared_cacheline_info(FILE *out)
2027{
2028 struct c2c_stats *stats = &c2c.hitm_stats;
2029 int hitm_cnt = stats->lcl_hitm + stats->rmt_hitm;
2030
2031 fprintf(out, "=================================================\n");
2032 fprintf(out, " Global Shared Cache Line Event Information \n");
2033 fprintf(out, "=================================================\n");
2034 fprintf(out, " Total Shared Cache Lines : %10d\n", c2c.shared_clines);
2035 fprintf(out, " Load HITs on shared lines : %10d\n", stats->load);
2036 fprintf(out, " Fill Buffer Hits on shared lines : %10d\n", stats->ld_fbhit);
2037 fprintf(out, " L1D hits on shared lines : %10d\n", stats->ld_l1hit);
2038 fprintf(out, " L2D hits on shared lines : %10d\n", stats->ld_l2hit);
2039 fprintf(out, " LLC hits on shared lines : %10d\n", stats->ld_llchit + stats->lcl_hitm);
2040 fprintf(out, " Locked Access on shared lines : %10d\n", stats->locks);
2041 fprintf(out, " Store HITs on shared lines : %10d\n", stats->store);
2042 fprintf(out, " Store L1D hits on shared lines : %10d\n", stats->st_l1hit);
2043 fprintf(out, " Total Merged records : %10d\n", hitm_cnt + stats->store);
2044}
2045
2046static void print_cacheline(struct c2c_hists *c2c_hists,
2047 struct hist_entry *he_cl,
2048 struct perf_hpp_list *hpp_list,
2049 FILE *out)
2050{
2051 char bf[1000];
2052 struct perf_hpp hpp = {
2053 .buf = bf,
2054 .size = 1000,
2055 };
2056 static bool once;
2057
2058 if (!once) {
2059 hists__fprintf_headers(&c2c_hists->hists, out);
2060 once = true;
2061 } else {
2062 fprintf(out, "\n");
2063 }
2064
2065 fprintf(out, " -------------------------------------------------------------\n");
2066 __hist_entry__snprintf(he_cl, &hpp, hpp_list);
2067 fprintf(out, "%s\n", bf);
2068 fprintf(out, " -------------------------------------------------------------\n");
2069
2070 hists__fprintf(&c2c_hists->hists, false, 0, 0, 0, out, true);
2071}
2072
2073static void print_pareto(FILE *out)
2074{
2075 struct perf_hpp_list hpp_list;
2076 struct rb_node *nd;
2077 int ret;
2078
2079 perf_hpp_list__init(&hpp_list);
2080 ret = hpp_list__parse(&hpp_list,
2081 "cl_num,"
2082 "cl_rmt_hitm,"
2083 "cl_lcl_hitm,"
2084 "cl_stores_l1hit,"
2085 "cl_stores_l1miss,"
2086 "dcacheline",
2087 NULL);
2088
2089 if (WARN_ONCE(ret, "failed to setup sort entries\n"))
2090 return;
2091
2092 nd = rb_first(&c2c.hists.hists.entries);
2093
2094 for (; nd; nd = rb_next(nd)) {
2095 struct hist_entry *he = rb_entry(nd, struct hist_entry, rb_node);
2096 struct c2c_hist_entry *c2c_he;
2097
2098 if (he->filtered)
2099 continue;
2100
2101 c2c_he = container_of(he, struct c2c_hist_entry, he);
2102 print_cacheline(c2c_he->hists, he, &hpp_list, out);
2103 }
2104}
2105
2106static void print_c2c_info(FILE *out, struct perf_session *session)
2107{
2108 struct perf_evlist *evlist = session->evlist;
2109 struct perf_evsel *evsel;
2110 bool first = true;
2111
2112 fprintf(out, "=================================================\n");
2113 fprintf(out, " c2c details \n");
2114 fprintf(out, "=================================================\n");
2115
2116 evlist__for_each_entry(evlist, evsel) {
2117 fprintf(out, "%-36s: %s\n", first ? " Events" : "",
2118 perf_evsel__name(evsel));
2119 first = false;
2120 }
2121 fprintf(out, " Cachelines sort on : %s HITMs\n",
2122 display_str[c2c.display]);
2123 fprintf(out, " Cacheline data grouping : %s\n", c2c.cl_sort);
2124}
2125
2126static void perf_c2c__hists_fprintf(FILE *out, struct perf_session *session)
2127{
2128 setup_pager();
2129
2130 print_c2c__display_stats(out);
2131 fprintf(out, "\n");
2132 print_shared_cacheline_info(out);
2133 fprintf(out, "\n");
2134 print_c2c_info(out, session);
2135
2136 if (c2c.stats_only)
2137 return;
2138
2139 fprintf(out, "\n");
2140 fprintf(out, "=================================================\n");
2141 fprintf(out, " Shared Data Cache Line Table \n");
2142 fprintf(out, "=================================================\n");
2143 fprintf(out, "#\n");
2144
2145 hists__fprintf(&c2c.hists.hists, true, 0, 0, 0, stdout, false);
2146
2147 fprintf(out, "\n");
2148 fprintf(out, "=================================================\n");
2149 fprintf(out, " Shared Cache Line Distribution Pareto \n");
2150 fprintf(out, "=================================================\n");
2151 fprintf(out, "#\n");
2152
2153 print_pareto(out);
2154}
2155
2156#ifdef HAVE_SLANG_SUPPORT
2157static void c2c_browser__update_nr_entries(struct hist_browser *hb)
2158{
2159 u64 nr_entries = 0;
2160 struct rb_node *nd = rb_first(&hb->hists->entries);
2161
2162 while (nd) {
2163 struct hist_entry *he = rb_entry(nd, struct hist_entry, rb_node);
2164
2165 if (!he->filtered)
2166 nr_entries++;
2167
2168 nd = rb_next(nd);
2169 }
2170
2171 hb->nr_non_filtered_entries = nr_entries;
2172}
2173
2174struct c2c_cacheline_browser {
2175 struct hist_browser hb;
2176 struct hist_entry *he;
2177};
2178
2179static int
2180perf_c2c_cacheline_browser__title(struct hist_browser *browser,
2181 char *bf, size_t size)
2182{
2183 struct c2c_cacheline_browser *cl_browser;
2184 struct hist_entry *he;
2185 uint64_t addr = 0;
2186
2187 cl_browser = container_of(browser, struct c2c_cacheline_browser, hb);
2188 he = cl_browser->he;
2189
2190 if (he->mem_info)
2191 addr = cl_address(he->mem_info->daddr.addr);
2192
2193 scnprintf(bf, size, "Cacheline 0x%lx", addr);
2194 return 0;
2195}
2196
2197static struct c2c_cacheline_browser*
2198c2c_cacheline_browser__new(struct hists *hists, struct hist_entry *he)
2199{
2200 struct c2c_cacheline_browser *browser;
2201
2202 browser = zalloc(sizeof(*browser));
2203 if (browser) {
2204 hist_browser__init(&browser->hb, hists);
2205 browser->hb.c2c_filter = true;
2206 browser->hb.title = perf_c2c_cacheline_browser__title;
2207 browser->he = he;
2208 }
2209
2210 return browser;
2211}
2212
2213static int perf_c2c__browse_cacheline(struct hist_entry *he)
2214{
2215 struct c2c_hist_entry *c2c_he;
2216 struct c2c_hists *c2c_hists;
2217 struct c2c_cacheline_browser *cl_browser;
2218 struct hist_browser *browser;
2219 int key = -1;
2220 const char help[] =
2221 " ENTER Togle callchains (if present) \n"
2222 " n Togle Node details info \n"
2223 " s Togle full lenght of symbol and source line columns \n"
2224 " q Return back to cacheline list \n";
2225
2226 /* Display compact version first. */
2227 c2c.symbol_full = false;
2228
2229 c2c_he = container_of(he, struct c2c_hist_entry, he);
2230 c2c_hists = c2c_he->hists;
2231
2232 cl_browser = c2c_cacheline_browser__new(&c2c_hists->hists, he);
2233 if (cl_browser == NULL)
2234 return -1;
2235
2236 browser = &cl_browser->hb;
2237
2238 /* reset abort key so that it can get Ctrl-C as a key */
2239 SLang_reset_tty();
2240 SLang_init_tty(0, 0, 0);
2241
2242 c2c_browser__update_nr_entries(browser);
2243
2244 while (1) {
2245 key = hist_browser__run(browser, "? - help");
2246
2247 switch (key) {
2248 case 's':
2249 c2c.symbol_full = !c2c.symbol_full;
2250 break;
2251 case 'n':
2252 c2c.node_info = (c2c.node_info + 1) % 3;
2253 setup_nodes_header();
2254 break;
2255 case 'q':
2256 goto out;
2257 case '?':
2258 ui_browser__help_window(&browser->b, help);
2259 break;
2260 default:
2261 break;
2262 }
2263 }
2264
2265out:
2266 free(cl_browser);
2267 return 0;
2268}
2269
2270static int perf_c2c_browser__title(struct hist_browser *browser,
2271 char *bf, size_t size)
2272{
2273 scnprintf(bf, size,
2274 "Shared Data Cache Line Table "
2275 "(%lu entries, sorted on %s HITMs)",
2276 browser->nr_non_filtered_entries,
2277 display_str[c2c.display]);
2278 return 0;
2279}
2280
2281static struct hist_browser*
2282perf_c2c_browser__new(struct hists *hists)
2283{
2284 struct hist_browser *browser = hist_browser__new(hists);
2285
2286 if (browser) {
2287 browser->title = perf_c2c_browser__title;
2288 browser->c2c_filter = true;
2289 }
2290
2291 return browser;
2292}
2293
2294static int perf_c2c__hists_browse(struct hists *hists)
2295{
2296 struct hist_browser *browser;
2297 int key = -1;
2298 const char help[] =
2299 " d Display cacheline details \n"
2300 " ENTER Togle callchains (if present) \n"
2301 " q Quit \n";
2302
2303 browser = perf_c2c_browser__new(hists);
2304 if (browser == NULL)
2305 return -1;
2306
2307 /* reset abort key so that it can get Ctrl-C as a key */
2308 SLang_reset_tty();
2309 SLang_init_tty(0, 0, 0);
2310
2311 c2c_browser__update_nr_entries(browser);
2312
2313 while (1) {
2314 key = hist_browser__run(browser, "? - help");
2315
2316 switch (key) {
2317 case 'q':
2318 goto out;
2319 case 'd':
2320 perf_c2c__browse_cacheline(browser->he_selection);
2321 break;
2322 case '?':
2323 ui_browser__help_window(&browser->b, help);
2324 break;
2325 default:
2326 break;
2327 }
2328 }
2329
2330out:
2331 hist_browser__delete(browser);
2332 return 0;
2333}
2334
2335static void perf_c2c_display(struct perf_session *session)
2336{
2337 if (c2c.use_stdio)
2338 perf_c2c__hists_fprintf(stdout, session);
2339 else
2340 perf_c2c__hists_browse(&c2c.hists.hists);
2341}
2342#else
2343static void perf_c2c_display(struct perf_session *session)
2344{
2345 use_browser = 0;
2346 perf_c2c__hists_fprintf(stdout, session);
2347}
2348#endif /* HAVE_SLANG_SUPPORT */
2349
2350static void ui_quirks(void)
2351{
2352 if (!c2c.use_stdio) {
2353 dim_offset.width = 5;
2354 dim_offset.header = header_offset_tui;
2355 }
2356
2357 dim_percent_hitm.header = percent_hitm_header[c2c.display];
2358}
2359
2360#define CALLCHAIN_DEFAULT_OPT "graph,0.5,caller,function,percent"
2361
2362const char callchain_help[] = "Display call graph (stack chain/backtrace):\n\n"
2363 CALLCHAIN_REPORT_HELP
2364 "\n\t\t\t\tDefault: " CALLCHAIN_DEFAULT_OPT;
2365
2366static int
2367parse_callchain_opt(const struct option *opt, const char *arg, int unset)
2368{
2369 struct callchain_param *callchain = opt->value;
2370
2371 callchain->enabled = !unset;
2372 /*
2373 * --no-call-graph
2374 */
2375 if (unset) {
2376 symbol_conf.use_callchain = false;
2377 callchain->mode = CHAIN_NONE;
2378 return 0;
2379 }
2380
2381 return parse_callchain_report_opt(arg);
2382}
2383
2384static int setup_callchain(struct perf_evlist *evlist)
2385{
2386 u64 sample_type = perf_evlist__combined_sample_type(evlist);
2387 enum perf_call_graph_mode mode = CALLCHAIN_NONE;
2388
2389 if ((sample_type & PERF_SAMPLE_REGS_USER) &&
2390 (sample_type & PERF_SAMPLE_STACK_USER))
2391 mode = CALLCHAIN_DWARF;
2392 else if (sample_type & PERF_SAMPLE_BRANCH_STACK)
2393 mode = CALLCHAIN_LBR;
2394 else if (sample_type & PERF_SAMPLE_CALLCHAIN)
2395 mode = CALLCHAIN_FP;
2396
2397 if (!callchain_param.enabled &&
2398 callchain_param.mode != CHAIN_NONE &&
2399 mode != CALLCHAIN_NONE) {
2400 symbol_conf.use_callchain = true;
2401 if (callchain_register_param(&callchain_param) < 0) {
2402 ui__error("Can't register callchain params.\n");
2403 return -EINVAL;
2404 }
2405 }
2406
2407 callchain_param.record_mode = mode;
2408 callchain_param.min_percent = 0;
2409 return 0;
2410}
2411
2412static int setup_display(const char *str)
2413{
2414 const char *display = str ?: "tot";
2415
2416 if (!strcmp(display, "tot"))
2417 c2c.display = DISPLAY_TOT;
2418 else if (!strcmp(display, "rmt"))
2419 c2c.display = DISPLAY_RMT;
2420 else if (!strcmp(display, "lcl"))
2421 c2c.display = DISPLAY_LCL;
2422 else {
2423 pr_err("failed: unknown display type: %s\n", str);
2424 return -1;
2425 }
2426
2427 return 0;
2428}
2429
2430#define for_each_token(__tok, __buf, __sep, __tmp) \
2431 for (__tok = strtok_r(__buf, __sep, &__tmp); __tok; \
2432 __tok = strtok_r(NULL, __sep, &__tmp))
2433
2434static int build_cl_output(char *cl_sort, bool no_source)
2435{
2436 char *tok, *tmp, *buf = strdup(cl_sort);
2437 bool add_pid = false;
2438 bool add_tid = false;
2439 bool add_iaddr = false;
2440 bool add_sym = false;
2441 bool add_dso = false;
2442 bool add_src = false;
2443
2444 if (!buf)
2445 return -ENOMEM;
2446
2447 for_each_token(tok, buf, ",", tmp) {
2448 if (!strcmp(tok, "tid")) {
2449 add_tid = true;
2450 } else if (!strcmp(tok, "pid")) {
2451 add_pid = true;
2452 } else if (!strcmp(tok, "iaddr")) {
2453 add_iaddr = true;
2454 add_sym = true;
2455 add_dso = true;
2456 add_src = no_source ? false : true;
2457 } else if (!strcmp(tok, "dso")) {
2458 add_dso = true;
2459 } else if (strcmp(tok, "offset")) {
2460 pr_err("unrecognized sort token: %s\n", tok);
2461 return -EINVAL;
2462 }
2463 }
2464
2465 if (asprintf(&c2c.cl_output,
2466 "%s%s%s%s%s%s%s%s%s%s",
2467 c2c.use_stdio ? "cl_num_empty," : "",
2468 "percent_rmt_hitm,"
2469 "percent_lcl_hitm,"
2470 "percent_stores_l1hit,"
2471 "percent_stores_l1miss,"
2472 "offset,",
2473 add_pid ? "pid," : "",
2474 add_tid ? "tid," : "",
2475 add_iaddr ? "iaddr," : "",
2476 "mean_rmt,"
2477 "mean_lcl,"
2478 "mean_load,"
2479 "cpucnt,",
2480 add_sym ? "symbol," : "",
2481 add_dso ? "dso," : "",
2482 add_src ? "cl_srcline," : "",
2483 "node") < 0)
2484 return -ENOMEM;
2485
2486 c2c.show_src = add_src;
2487
2488 free(buf);
2489 return 0;
2490}
2491
2492static int setup_coalesce(const char *coalesce, bool no_source)
2493{
2494 const char *c = coalesce ?: coalesce_default;
2495
2496 if (asprintf(&c2c.cl_sort, "offset,%s", c) < 0)
2497 return -ENOMEM;
2498
2499 if (build_cl_output(c2c.cl_sort, no_source))
2500 return -1;
2501
2502 if (asprintf(&c2c.cl_resort, "offset,%s",
2503 c2c.display == DISPLAY_TOT ?
2504 "tot_hitm" :
2505 c2c.display == DISPLAY_RMT ?
2506 "rmt_hitm,lcl_hitm" :
2507 "lcl_hitm,rmt_hitm") < 0)
2508 return -ENOMEM;
2509
2510 pr_debug("coalesce sort fields: %s\n", c2c.cl_sort);
2511 pr_debug("coalesce resort fields: %s\n", c2c.cl_resort);
2512 pr_debug("coalesce output fields: %s\n", c2c.cl_output);
2513 return 0;
2514}
2515
2516static int perf_c2c__report(int argc, const char **argv)
2517{
2518 struct perf_session *session;
2519 struct ui_progress prog;
2520 struct perf_data_file file = {
2521 .mode = PERF_DATA_MODE_READ,
2522 };
2523 char callchain_default_opt[] = CALLCHAIN_DEFAULT_OPT;
2524 const char *display = NULL;
2525 const char *coalesce = NULL;
2526 bool no_source = false;
2527 const struct option options[] = {
2528 OPT_STRING('k', "vmlinux", &symbol_conf.vmlinux_name,
2529 "file", "vmlinux pathname"),
2530 OPT_STRING('i', "input", &input_name, "file",
2531 "the input file to process"),
2532 OPT_INCR('N', "node-info", &c2c.node_info,
2533 "show extra node info in report (repeat for more info)"),
2534#ifdef HAVE_SLANG_SUPPORT
2535 OPT_BOOLEAN(0, "stdio", &c2c.use_stdio, "Use the stdio interface"),
2536#endif
2537 OPT_BOOLEAN(0, "stats", &c2c.stats_only,
2538 "Use the stdio interface"),
2539 OPT_BOOLEAN(0, "full-symbols", &c2c.symbol_full,
2540 "Display full length of symbols"),
2541 OPT_BOOLEAN(0, "no-source", &no_source,
2542 "Do not display Source Line column"),
2543 OPT_BOOLEAN(0, "show-all", &c2c.show_all,
2544 "Show all captured HITM lines."),
2545 OPT_CALLBACK_DEFAULT('g', "call-graph", &callchain_param,
2546 "print_type,threshold[,print_limit],order,sort_key[,branch],value",
2547 callchain_help, &parse_callchain_opt,
2548 callchain_default_opt),
2549 OPT_STRING('d', "display", &display, "Switch HITM output type", "lcl,rmt"),
2550 OPT_STRING('c', "coalesce", &coalesce, "coalesce fields",
2551 "coalesce fields: pid,tid,iaddr,dso"),
2552 OPT_BOOLEAN('f', "force", &symbol_conf.force, "don't complain, do it"),
2553 OPT_PARENT(c2c_options),
2554 OPT_END()
2555 };
2556 int err = 0;
2557
2558 argc = parse_options(argc, argv, options, report_c2c_usage,
2559 PARSE_OPT_STOP_AT_NON_OPTION);
2560 if (argc)
2561 usage_with_options(report_c2c_usage, options);
2562
2563 if (c2c.stats_only)
2564 c2c.use_stdio = true;
2565
2566 if (!input_name || !strlen(input_name))
2567 input_name = "perf.data";
2568
2569 file.path = input_name;
2570 file.force = symbol_conf.force;
2571
2572 err = setup_display(display);
2573 if (err)
2574 goto out;
2575
2576 err = setup_coalesce(coalesce, no_source);
2577 if (err) {
2578 pr_debug("Failed to initialize hists\n");
2579 goto out;
2580 }
2581
2582 err = c2c_hists__init(&c2c.hists, "dcacheline", 2);
2583 if (err) {
2584 pr_debug("Failed to initialize hists\n");
2585 goto out;
2586 }
2587
2588 session = perf_session__new(&file, 0, &c2c.tool);
2589 if (session == NULL) {
2590 pr_debug("No memory for session\n");
2591 goto out;
2592 }
2593
2594 err = setup_nodes(session);
2595 if (err) {
2596 pr_err("Failed setup nodes\n");
2597 goto out;
2598 }
2599
2600 err = setup_callchain(session->evlist);
2601 if (err)
2602 goto out_session;
2603
2604 if (symbol__init(&session->header.env) < 0)
2605 goto out_session;
2606
2607 /* No pipe support at the moment. */
2608 if (perf_data_file__is_pipe(session->file)) {
2609 pr_debug("No pipe support at the moment.\n");
2610 goto out_session;
2611 }
2612
2613 if (c2c.use_stdio)
2614 use_browser = 0;
2615 else
2616 use_browser = 1;
2617
2618 setup_browser(false);
2619
2620 err = perf_session__process_events(session);
2621 if (err) {
2622 pr_err("failed to process sample\n");
2623 goto out_session;
2624 }
2625
2626 c2c_hists__reinit(&c2c.hists,
2627 "cl_idx,"
2628 "dcacheline,"
2629 "tot_recs,"
2630 "percent_hitm,"
2631 "tot_hitm,lcl_hitm,rmt_hitm,"
2632 "stores,stores_l1hit,stores_l1miss,"
2633 "dram_lcl,dram_rmt,"
2634 "ld_llcmiss,"
2635 "tot_loads,"
2636 "ld_fbhit,ld_l1hit,ld_l2hit,"
2637 "ld_lclhit,ld_rmthit",
2638 c2c.display == DISPLAY_TOT ? "tot_hitm" :
2639 c2c.display == DISPLAY_LCL ? "lcl_hitm" : "rmt_hitm"
2640 );
2641
2642 ui_progress__init(&prog, c2c.hists.hists.nr_entries, "Sorting...");
2643
2644 hists__collapse_resort(&c2c.hists.hists, NULL);
2645 hists__output_resort_cb(&c2c.hists.hists, &prog, resort_hitm_cb);
2646 hists__iterate_cb(&c2c.hists.hists, resort_cl_cb);
2647
2648 ui_progress__finish();
2649
2650 ui_quirks();
2651
2652 perf_c2c_display(session);
2653
2654out_session:
2655 perf_session__delete(session);
2656out:
2657 return err;
2658}
2659
2660static int parse_record_events(const struct option *opt,
2661 const char *str, int unset __maybe_unused)
2662{
2663 bool *event_set = (bool *) opt->value;
2664
2665 *event_set = true;
2666 return perf_mem_events__parse(str);
2667}
2668
2669
2670static const char * const __usage_record[] = {
2671 "perf c2c record [<options>] [<command>]",
2672 "perf c2c record [<options>] -- <command> [<options>]",
2673 NULL
2674};
2675
2676static const char * const *record_mem_usage = __usage_record;
2677
2678static int perf_c2c__record(int argc, const char **argv)
2679{
2680 int rec_argc, i = 0, j;
2681 const char **rec_argv;
2682 int ret;
2683 bool all_user = false, all_kernel = false;
2684 bool event_set = false;
2685 struct option options[] = {
2686 OPT_CALLBACK('e', "event", &event_set, "event",
2687 "event selector. Use 'perf mem record -e list' to list available events",
2688 parse_record_events),
2689 OPT_BOOLEAN('u', "all-user", &all_user, "collect only user level data"),
2690 OPT_BOOLEAN('k', "all-kernel", &all_kernel, "collect only kernel level data"),
2691 OPT_UINTEGER('l', "ldlat", &perf_mem_events__loads_ldlat, "setup mem-loads latency"),
2692 OPT_PARENT(c2c_options),
2693 OPT_END()
2694 };
2695
2696 if (perf_mem_events__init()) {
2697 pr_err("failed: memory events not supported\n");
2698 return -1;
2699 }
2700
2701 argc = parse_options(argc, argv, options, record_mem_usage,
2702 PARSE_OPT_KEEP_UNKNOWN);
2703
2704 rec_argc = argc + 10; /* max number of arguments */
2705 rec_argv = calloc(rec_argc + 1, sizeof(char *));
2706 if (!rec_argv)
2707 return -1;
2708
2709 rec_argv[i++] = "record";
2710
2711 if (!event_set) {
2712 perf_mem_events[PERF_MEM_EVENTS__LOAD].record = true;
2713 perf_mem_events[PERF_MEM_EVENTS__STORE].record = true;
2714 }
2715
2716 if (perf_mem_events[PERF_MEM_EVENTS__LOAD].record)
2717 rec_argv[i++] = "-W";
2718
2719 rec_argv[i++] = "-d";
2720 rec_argv[i++] = "--sample-cpu";
2721
2722 for (j = 0; j < PERF_MEM_EVENTS__MAX; j++) {
2723 if (!perf_mem_events[j].record)
2724 continue;
2725
2726 if (!perf_mem_events[j].supported) {
2727 pr_err("failed: event '%s' not supported\n",
2728 perf_mem_events[j].name);
2729 return -1;
2730 }
2731
2732 rec_argv[i++] = "-e";
2733 rec_argv[i++] = perf_mem_events__name(j);
2734 };
2735
2736 if (all_user)
2737 rec_argv[i++] = "--all-user";
2738
2739 if (all_kernel)
2740 rec_argv[i++] = "--all-kernel";
2741
2742 for (j = 0; j < argc; j++, i++)
2743 rec_argv[i] = argv[j];
2744
2745 if (verbose > 0) {
2746 pr_debug("calling: ");
2747
2748 j = 0;
2749
2750 while (rec_argv[j]) {
2751 pr_debug("%s ", rec_argv[j]);
2752 j++;
2753 }
2754 pr_debug("\n");
2755 }
2756
2757 ret = cmd_record(i, rec_argv, NULL);
2758 free(rec_argv);
2759 return ret;
2760}
2761
2762int cmd_c2c(int argc, const char **argv, const char *prefix __maybe_unused)
2763{
2764 argc = parse_options(argc, argv, c2c_options, c2c_usage,
2765 PARSE_OPT_STOP_AT_NON_OPTION);
2766
2767 if (!argc)
2768 usage_with_options(c2c_usage, c2c_options);
2769
2770 if (!strncmp(argv[0], "rec", 3)) {
2771 return perf_c2c__record(argc, argv);
2772 } else if (!strncmp(argv[0], "rep", 3)) {
2773 return perf_c2c__report(argc, argv);
2774 } else {
2775 usage_with_options(c2c_usage, c2c_options);
2776 }
2777
2778 return 0;
2779}
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * This is rewrite of original c2c tool introduced in here:
4 * http://lwn.net/Articles/588866/
5 *
6 * The original tool was changed to fit in current perf state.
7 *
8 * Original authors:
9 * Don Zickus <dzickus@redhat.com>
10 * Dick Fowles <fowles@inreach.com>
11 * Joe Mario <jmario@redhat.com>
12 */
13#include <errno.h>
14#include <inttypes.h>
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>
20#include <asm/bug.h>
21#include <sys/param.h>
22#include "debug.h"
23#include "builtin.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"
29#include "session.h"
30#include "hist.h"
31#include "sort.h"
32#include "tool.h"
33#include "cacheline.h"
34#include "data.h"
35#include "event.h"
36#include "evlist.h"
37#include "evsel.h"
38#include "ui/browsers/hists.h"
39#include "thread.h"
40#include "mem2node.h"
41#include "symbol.h"
42#include "ui/ui.h"
43#include "ui/progress.h"
44#include "pmus.h"
45#include "string2.h"
46#include "util/util.h"
47
48struct c2c_hists {
49 struct hists hists;
50 struct perf_hpp_list list;
51 struct c2c_stats stats;
52};
53
54struct compute_stats {
55 struct stats lcl_hitm;
56 struct stats rmt_hitm;
57 struct stats lcl_peer;
58 struct stats rmt_peer;
59 struct stats load;
60};
61
62struct c2c_hist_entry {
63 struct c2c_hists *hists;
64 struct c2c_stats stats;
65 unsigned long *cpuset;
66 unsigned long *nodeset;
67 struct c2c_stats *node_stats;
68 unsigned int cacheline_idx;
69
70 struct compute_stats cstats;
71
72 unsigned long paddr;
73 unsigned long paddr_cnt;
74 bool paddr_zero;
75 char *nodestr;
76
77 /*
78 * must be at the end,
79 * because of its callchain dynamic entry
80 */
81 struct hist_entry he;
82};
83
84static char const *coalesce_default = "iaddr";
85
86struct perf_c2c {
87 struct perf_tool tool;
88 struct c2c_hists hists;
89 struct mem2node mem2node;
90
91 unsigned long **nodes;
92 int nodes_cnt;
93 int cpus_cnt;
94 int *cpu2node;
95 int node_info;
96
97 bool show_src;
98 bool show_all;
99 bool use_stdio;
100 bool stats_only;
101 bool symbol_full;
102 bool stitch_lbr;
103
104 /* Shared cache line stats */
105 struct c2c_stats shared_clines_stats;
106 int shared_clines;
107
108 int display;
109
110 const char *coalesce;
111 char *cl_sort;
112 char *cl_resort;
113 char *cl_output;
114};
115
116enum {
117 DISPLAY_LCL_HITM,
118 DISPLAY_RMT_HITM,
119 DISPLAY_TOT_HITM,
120 DISPLAY_SNP_PEER,
121 DISPLAY_MAX,
122};
123
124static const char *display_str[DISPLAY_MAX] = {
125 [DISPLAY_LCL_HITM] = "Local HITMs",
126 [DISPLAY_RMT_HITM] = "Remote HITMs",
127 [DISPLAY_TOT_HITM] = "Total HITMs",
128 [DISPLAY_SNP_PEER] = "Peer Snoop",
129};
130
131static const struct option c2c_options[] = {
132 OPT_INCR('v', "verbose", &verbose, "be more verbose (show counter open errors, etc)"),
133 OPT_END()
134};
135
136static struct perf_c2c c2c;
137
138static void *c2c_he_zalloc(size_t size)
139{
140 struct c2c_hist_entry *c2c_he;
141
142 c2c_he = zalloc(size + sizeof(*c2c_he));
143 if (!c2c_he)
144 return NULL;
145
146 c2c_he->cpuset = bitmap_zalloc(c2c.cpus_cnt);
147 if (!c2c_he->cpuset)
148 goto out_free;
149
150 c2c_he->nodeset = bitmap_zalloc(c2c.nodes_cnt);
151 if (!c2c_he->nodeset)
152 goto out_free;
153
154 c2c_he->node_stats = zalloc(c2c.nodes_cnt * sizeof(*c2c_he->node_stats));
155 if (!c2c_he->node_stats)
156 goto out_free;
157
158 init_stats(&c2c_he->cstats.lcl_hitm);
159 init_stats(&c2c_he->cstats.rmt_hitm);
160 init_stats(&c2c_he->cstats.lcl_peer);
161 init_stats(&c2c_he->cstats.rmt_peer);
162 init_stats(&c2c_he->cstats.load);
163
164 return &c2c_he->he;
165
166out_free:
167 zfree(&c2c_he->nodeset);
168 zfree(&c2c_he->cpuset);
169 free(c2c_he);
170 return NULL;
171}
172
173static void c2c_he_free(void *he)
174{
175 struct c2c_hist_entry *c2c_he;
176
177 c2c_he = container_of(he, struct c2c_hist_entry, he);
178 if (c2c_he->hists) {
179 hists__delete_entries(&c2c_he->hists->hists);
180 zfree(&c2c_he->hists);
181 }
182
183 zfree(&c2c_he->cpuset);
184 zfree(&c2c_he->nodeset);
185 zfree(&c2c_he->nodestr);
186 zfree(&c2c_he->node_stats);
187 free(c2c_he);
188}
189
190static struct hist_entry_ops c2c_entry_ops = {
191 .new = c2c_he_zalloc,
192 .free = c2c_he_free,
193};
194
195static int c2c_hists__init(struct c2c_hists *hists,
196 const char *sort,
197 int nr_header_lines);
198
199static struct c2c_hists*
200he__get_c2c_hists(struct hist_entry *he,
201 const char *sort,
202 int nr_header_lines)
203{
204 struct c2c_hist_entry *c2c_he;
205 struct c2c_hists *hists;
206 int ret;
207
208 c2c_he = container_of(he, struct c2c_hist_entry, he);
209 if (c2c_he->hists)
210 return c2c_he->hists;
211
212 hists = c2c_he->hists = zalloc(sizeof(*hists));
213 if (!hists)
214 return NULL;
215
216 ret = c2c_hists__init(hists, sort, nr_header_lines);
217 if (ret) {
218 free(hists);
219 return NULL;
220 }
221
222 return hists;
223}
224
225static void c2c_he__set_cpu(struct c2c_hist_entry *c2c_he,
226 struct perf_sample *sample)
227{
228 if (WARN_ONCE(sample->cpu == (unsigned int) -1,
229 "WARNING: no sample cpu value"))
230 return;
231
232 __set_bit(sample->cpu, c2c_he->cpuset);
233}
234
235static void c2c_he__set_node(struct c2c_hist_entry *c2c_he,
236 struct perf_sample *sample)
237{
238 int node;
239
240 if (!sample->phys_addr) {
241 c2c_he->paddr_zero = true;
242 return;
243 }
244
245 node = mem2node__node(&c2c.mem2node, sample->phys_addr);
246 if (WARN_ONCE(node < 0, "WARNING: failed to find node\n"))
247 return;
248
249 __set_bit(node, c2c_he->nodeset);
250
251 if (c2c_he->paddr != sample->phys_addr) {
252 c2c_he->paddr_cnt++;
253 c2c_he->paddr = sample->phys_addr;
254 }
255}
256
257static void compute_stats(struct c2c_hist_entry *c2c_he,
258 struct c2c_stats *stats,
259 u64 weight)
260{
261 struct compute_stats *cstats = &c2c_he->cstats;
262
263 if (stats->rmt_hitm)
264 update_stats(&cstats->rmt_hitm, weight);
265 else if (stats->lcl_hitm)
266 update_stats(&cstats->lcl_hitm, weight);
267 else if (stats->rmt_peer)
268 update_stats(&cstats->rmt_peer, weight);
269 else if (stats->lcl_peer)
270 update_stats(&cstats->lcl_peer, weight);
271 else if (stats->load)
272 update_stats(&cstats->load, weight);
273}
274
275static int process_sample_event(struct perf_tool *tool __maybe_unused,
276 union perf_event *event,
277 struct perf_sample *sample,
278 struct evsel *evsel,
279 struct machine *machine)
280{
281 struct c2c_hists *c2c_hists = &c2c.hists;
282 struct c2c_hist_entry *c2c_he;
283 struct c2c_stats stats = { .nr_entries = 0, };
284 struct hist_entry *he;
285 struct addr_location al;
286 struct mem_info *mi, *mi_dup;
287 struct callchain_cursor *cursor;
288 int ret;
289
290 addr_location__init(&al);
291 if (machine__resolve(machine, &al, sample) < 0) {
292 pr_debug("problem processing %d event, skipping it.\n",
293 event->header.type);
294 ret = -1;
295 goto out;
296 }
297
298 if (c2c.stitch_lbr)
299 thread__set_lbr_stitch_enable(al.thread, true);
300
301 cursor = get_tls_callchain_cursor();
302 ret = sample__resolve_callchain(sample, cursor, NULL,
303 evsel, &al, sysctl_perf_event_max_stack);
304 if (ret)
305 goto out;
306
307 mi = sample__resolve_mem(sample, &al);
308 if (mi == NULL) {
309 ret = -ENOMEM;
310 goto out;
311 }
312
313 /*
314 * The mi object is released in hists__add_entry_ops,
315 * if it gets sorted out into existing data, so we need
316 * to take the copy now.
317 */
318 mi_dup = mem_info__get(mi);
319
320 c2c_decode_stats(&stats, mi);
321
322 he = hists__add_entry_ops(&c2c_hists->hists, &c2c_entry_ops,
323 &al, NULL, NULL, mi, NULL,
324 sample, true);
325 if (he == NULL)
326 goto free_mi;
327
328 c2c_he = container_of(he, struct c2c_hist_entry, he);
329 c2c_add_stats(&c2c_he->stats, &stats);
330 c2c_add_stats(&c2c_hists->stats, &stats);
331
332 c2c_he__set_cpu(c2c_he, sample);
333 c2c_he__set_node(c2c_he, sample);
334
335 hists__inc_nr_samples(&c2c_hists->hists, he->filtered);
336 ret = hist_entry__append_callchain(he, sample);
337
338 if (!ret) {
339 /*
340 * There's already been warning about missing
341 * sample's cpu value. Let's account all to
342 * node 0 in this case, without any further
343 * warning.
344 *
345 * Doing node stats only for single callchain data.
346 */
347 int cpu = sample->cpu == (unsigned int) -1 ? 0 : sample->cpu;
348 int node = c2c.cpu2node[cpu];
349
350 mi = mi_dup;
351
352 c2c_hists = he__get_c2c_hists(he, c2c.cl_sort, 2);
353 if (!c2c_hists)
354 goto free_mi;
355
356 he = hists__add_entry_ops(&c2c_hists->hists, &c2c_entry_ops,
357 &al, NULL, NULL, mi, NULL,
358 sample, true);
359 if (he == NULL)
360 goto free_mi;
361
362 c2c_he = container_of(he, struct c2c_hist_entry, he);
363 c2c_add_stats(&c2c_he->stats, &stats);
364 c2c_add_stats(&c2c_hists->stats, &stats);
365 c2c_add_stats(&c2c_he->node_stats[node], &stats);
366
367 compute_stats(c2c_he, &stats, sample->weight);
368
369 c2c_he__set_cpu(c2c_he, sample);
370 c2c_he__set_node(c2c_he, sample);
371
372 hists__inc_nr_samples(&c2c_hists->hists, he->filtered);
373 ret = hist_entry__append_callchain(he, sample);
374 }
375
376out:
377 addr_location__exit(&al);
378 return ret;
379
380free_mi:
381 mem_info__put(mi_dup);
382 mem_info__put(mi);
383 ret = -ENOMEM;
384 goto out;
385}
386
387static struct perf_c2c c2c = {
388 .tool = {
389 .sample = process_sample_event,
390 .mmap = perf_event__process_mmap,
391 .mmap2 = perf_event__process_mmap2,
392 .comm = perf_event__process_comm,
393 .exit = perf_event__process_exit,
394 .fork = perf_event__process_fork,
395 .lost = perf_event__process_lost,
396 .attr = perf_event__process_attr,
397 .auxtrace_info = perf_event__process_auxtrace_info,
398 .auxtrace = perf_event__process_auxtrace,
399 .auxtrace_error = perf_event__process_auxtrace_error,
400 .ordered_events = true,
401 .ordering_requires_timestamps = true,
402 },
403};
404
405static const char * const c2c_usage[] = {
406 "perf c2c {record|report}",
407 NULL
408};
409
410static const char * const __usage_report[] = {
411 "perf c2c report",
412 NULL
413};
414
415static const char * const *report_c2c_usage = __usage_report;
416
417#define C2C_HEADER_MAX 2
418
419struct c2c_header {
420 struct {
421 const char *text;
422 int span;
423 } line[C2C_HEADER_MAX];
424};
425
426struct c2c_dimension {
427 struct c2c_header header;
428 const char *name;
429 int width;
430 struct sort_entry *se;
431
432 int64_t (*cmp)(struct perf_hpp_fmt *fmt,
433 struct hist_entry *, struct hist_entry *);
434 int (*entry)(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
435 struct hist_entry *he);
436 int (*color)(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
437 struct hist_entry *he);
438};
439
440struct c2c_fmt {
441 struct perf_hpp_fmt fmt;
442 struct c2c_dimension *dim;
443};
444
445#define SYMBOL_WIDTH 30
446
447static struct c2c_dimension dim_symbol;
448static struct c2c_dimension dim_srcline;
449
450static int symbol_width(struct hists *hists, struct sort_entry *se)
451{
452 int width = hists__col_len(hists, se->se_width_idx);
453
454 if (!c2c.symbol_full)
455 width = MIN(width, SYMBOL_WIDTH);
456
457 return width;
458}
459
460static int c2c_width(struct perf_hpp_fmt *fmt,
461 struct perf_hpp *hpp __maybe_unused,
462 struct hists *hists)
463{
464 struct c2c_fmt *c2c_fmt;
465 struct c2c_dimension *dim;
466
467 c2c_fmt = container_of(fmt, struct c2c_fmt, fmt);
468 dim = c2c_fmt->dim;
469
470 if (dim == &dim_symbol || dim == &dim_srcline)
471 return symbol_width(hists, dim->se);
472
473 return dim->se ? hists__col_len(hists, dim->se->se_width_idx) :
474 c2c_fmt->dim->width;
475}
476
477static int c2c_header(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
478 struct hists *hists, int line, int *span)
479{
480 struct perf_hpp_list *hpp_list = hists->hpp_list;
481 struct c2c_fmt *c2c_fmt;
482 struct c2c_dimension *dim;
483 const char *text = NULL;
484 int width = c2c_width(fmt, hpp, hists);
485
486 c2c_fmt = container_of(fmt, struct c2c_fmt, fmt);
487 dim = c2c_fmt->dim;
488
489 if (dim->se) {
490 text = dim->header.line[line].text;
491 /* Use the last line from sort_entry if not defined. */
492 if (!text && (line == hpp_list->nr_header_lines - 1))
493 text = dim->se->se_header;
494 } else {
495 text = dim->header.line[line].text;
496
497 if (*span) {
498 (*span)--;
499 return 0;
500 } else {
501 *span = dim->header.line[line].span;
502 }
503 }
504
505 if (text == NULL)
506 text = "";
507
508 return scnprintf(hpp->buf, hpp->size, "%*s", width, text);
509}
510
511#define HEX_STR(__s, __v) \
512({ \
513 scnprintf(__s, sizeof(__s), "0x%" PRIx64, __v); \
514 __s; \
515})
516
517static int64_t
518dcacheline_cmp(struct perf_hpp_fmt *fmt __maybe_unused,
519 struct hist_entry *left, struct hist_entry *right)
520{
521 return sort__dcacheline_cmp(left, right);
522}
523
524static int dcacheline_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
525 struct hist_entry *he)
526{
527 uint64_t addr = 0;
528 int width = c2c_width(fmt, hpp, he->hists);
529 char buf[20];
530
531 if (he->mem_info)
532 addr = cl_address(he->mem_info->daddr.addr, chk_double_cl);
533
534 return scnprintf(hpp->buf, hpp->size, "%*s", width, HEX_STR(buf, addr));
535}
536
537static int
538dcacheline_node_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
539 struct hist_entry *he)
540{
541 struct c2c_hist_entry *c2c_he;
542 int width = c2c_width(fmt, hpp, he->hists);
543
544 c2c_he = container_of(he, struct c2c_hist_entry, he);
545 if (WARN_ON_ONCE(!c2c_he->nodestr))
546 return 0;
547
548 return scnprintf(hpp->buf, hpp->size, "%*s", width, c2c_he->nodestr);
549}
550
551static int
552dcacheline_node_count(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
553 struct hist_entry *he)
554{
555 struct c2c_hist_entry *c2c_he;
556 int width = c2c_width(fmt, hpp, he->hists);
557
558 c2c_he = container_of(he, struct c2c_hist_entry, he);
559 return scnprintf(hpp->buf, hpp->size, "%*lu", width, c2c_he->paddr_cnt);
560}
561
562static int offset_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
563 struct hist_entry *he)
564{
565 uint64_t addr = 0;
566 int width = c2c_width(fmt, hpp, he->hists);
567 char buf[20];
568
569 if (he->mem_info)
570 addr = cl_offset(he->mem_info->daddr.al_addr, chk_double_cl);
571
572 return scnprintf(hpp->buf, hpp->size, "%*s", width, HEX_STR(buf, addr));
573}
574
575static int64_t
576offset_cmp(struct perf_hpp_fmt *fmt __maybe_unused,
577 struct hist_entry *left, struct hist_entry *right)
578{
579 uint64_t l = 0, r = 0;
580
581 if (left->mem_info)
582 l = cl_offset(left->mem_info->daddr.addr, chk_double_cl);
583
584 if (right->mem_info)
585 r = cl_offset(right->mem_info->daddr.addr, chk_double_cl);
586
587 return (int64_t)(r - l);
588}
589
590static int
591iaddr_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
592 struct hist_entry *he)
593{
594 uint64_t addr = 0;
595 int width = c2c_width(fmt, hpp, he->hists);
596 char buf[20];
597
598 if (he->mem_info)
599 addr = he->mem_info->iaddr.addr;
600
601 return scnprintf(hpp->buf, hpp->size, "%*s", width, HEX_STR(buf, addr));
602}
603
604static int64_t
605iaddr_cmp(struct perf_hpp_fmt *fmt __maybe_unused,
606 struct hist_entry *left, struct hist_entry *right)
607{
608 return sort__iaddr_cmp(left, right);
609}
610
611static int
612tot_hitm_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
613 struct hist_entry *he)
614{
615 struct c2c_hist_entry *c2c_he;
616 int width = c2c_width(fmt, hpp, he->hists);
617 unsigned int tot_hitm;
618
619 c2c_he = container_of(he, struct c2c_hist_entry, he);
620 tot_hitm = c2c_he->stats.lcl_hitm + c2c_he->stats.rmt_hitm;
621
622 return scnprintf(hpp->buf, hpp->size, "%*u", width, tot_hitm);
623}
624
625static int64_t
626tot_hitm_cmp(struct perf_hpp_fmt *fmt __maybe_unused,
627 struct hist_entry *left, struct hist_entry *right)
628{
629 struct c2c_hist_entry *c2c_left;
630 struct c2c_hist_entry *c2c_right;
631 uint64_t tot_hitm_left;
632 uint64_t tot_hitm_right;
633
634 c2c_left = container_of(left, struct c2c_hist_entry, he);
635 c2c_right = container_of(right, struct c2c_hist_entry, he);
636
637 tot_hitm_left = c2c_left->stats.lcl_hitm + c2c_left->stats.rmt_hitm;
638 tot_hitm_right = c2c_right->stats.lcl_hitm + c2c_right->stats.rmt_hitm;
639
640 return tot_hitm_left - tot_hitm_right;
641}
642
643#define STAT_FN_ENTRY(__f) \
644static int \
645__f ## _entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp, \
646 struct hist_entry *he) \
647{ \
648 struct c2c_hist_entry *c2c_he; \
649 int width = c2c_width(fmt, hpp, he->hists); \
650 \
651 c2c_he = container_of(he, struct c2c_hist_entry, he); \
652 return scnprintf(hpp->buf, hpp->size, "%*u", width, \
653 c2c_he->stats.__f); \
654}
655
656#define STAT_FN_CMP(__f) \
657static int64_t \
658__f ## _cmp(struct perf_hpp_fmt *fmt __maybe_unused, \
659 struct hist_entry *left, struct hist_entry *right) \
660{ \
661 struct c2c_hist_entry *c2c_left, *c2c_right; \
662 \
663 c2c_left = container_of(left, struct c2c_hist_entry, he); \
664 c2c_right = container_of(right, struct c2c_hist_entry, he); \
665 return (uint64_t) c2c_left->stats.__f - \
666 (uint64_t) c2c_right->stats.__f; \
667}
668
669#define STAT_FN(__f) \
670 STAT_FN_ENTRY(__f) \
671 STAT_FN_CMP(__f)
672
673STAT_FN(rmt_hitm)
674STAT_FN(lcl_hitm)
675STAT_FN(rmt_peer)
676STAT_FN(lcl_peer)
677STAT_FN(tot_peer)
678STAT_FN(store)
679STAT_FN(st_l1hit)
680STAT_FN(st_l1miss)
681STAT_FN(st_na)
682STAT_FN(ld_fbhit)
683STAT_FN(ld_l1hit)
684STAT_FN(ld_l2hit)
685STAT_FN(ld_llchit)
686STAT_FN(rmt_hit)
687
688static uint64_t get_load_llc_misses(struct c2c_stats *stats)
689{
690 return stats->lcl_dram +
691 stats->rmt_dram +
692 stats->rmt_hitm +
693 stats->rmt_hit;
694}
695
696static uint64_t get_load_cache_hits(struct c2c_stats *stats)
697{
698 return stats->ld_fbhit +
699 stats->ld_l1hit +
700 stats->ld_l2hit +
701 stats->ld_llchit +
702 stats->lcl_hitm;
703}
704
705static uint64_t get_stores(struct c2c_stats *stats)
706{
707 return stats->st_l1hit +
708 stats->st_l1miss +
709 stats->st_na;
710}
711
712static uint64_t total_records(struct c2c_stats *stats)
713{
714 return get_load_llc_misses(stats) +
715 get_load_cache_hits(stats) +
716 get_stores(stats);
717}
718
719static int
720tot_recs_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
721 struct hist_entry *he)
722{
723 struct c2c_hist_entry *c2c_he;
724 int width = c2c_width(fmt, hpp, he->hists);
725 uint64_t tot_recs;
726
727 c2c_he = container_of(he, struct c2c_hist_entry, he);
728 tot_recs = total_records(&c2c_he->stats);
729
730 return scnprintf(hpp->buf, hpp->size, "%*" PRIu64, width, tot_recs);
731}
732
733static int64_t
734tot_recs_cmp(struct perf_hpp_fmt *fmt __maybe_unused,
735 struct hist_entry *left, struct hist_entry *right)
736{
737 struct c2c_hist_entry *c2c_left;
738 struct c2c_hist_entry *c2c_right;
739 uint64_t tot_recs_left;
740 uint64_t tot_recs_right;
741
742 c2c_left = container_of(left, struct c2c_hist_entry, he);
743 c2c_right = container_of(right, struct c2c_hist_entry, he);
744
745 tot_recs_left = total_records(&c2c_left->stats);
746 tot_recs_right = total_records(&c2c_right->stats);
747
748 return tot_recs_left - tot_recs_right;
749}
750
751static uint64_t total_loads(struct c2c_stats *stats)
752{
753 return get_load_llc_misses(stats) +
754 get_load_cache_hits(stats);
755}
756
757static int
758tot_loads_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
759 struct hist_entry *he)
760{
761 struct c2c_hist_entry *c2c_he;
762 int width = c2c_width(fmt, hpp, he->hists);
763 uint64_t tot_recs;
764
765 c2c_he = container_of(he, struct c2c_hist_entry, he);
766 tot_recs = total_loads(&c2c_he->stats);
767
768 return scnprintf(hpp->buf, hpp->size, "%*" PRIu64, width, tot_recs);
769}
770
771static int64_t
772tot_loads_cmp(struct perf_hpp_fmt *fmt __maybe_unused,
773 struct hist_entry *left, struct hist_entry *right)
774{
775 struct c2c_hist_entry *c2c_left;
776 struct c2c_hist_entry *c2c_right;
777 uint64_t tot_recs_left;
778 uint64_t tot_recs_right;
779
780 c2c_left = container_of(left, struct c2c_hist_entry, he);
781 c2c_right = container_of(right, struct c2c_hist_entry, he);
782
783 tot_recs_left = total_loads(&c2c_left->stats);
784 tot_recs_right = total_loads(&c2c_right->stats);
785
786 return tot_recs_left - tot_recs_right;
787}
788
789typedef double (get_percent_cb)(struct c2c_hist_entry *);
790
791static int
792percent_color(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
793 struct hist_entry *he, get_percent_cb get_percent)
794{
795 struct c2c_hist_entry *c2c_he;
796 int width = c2c_width(fmt, hpp, he->hists);
797 double per;
798
799 c2c_he = container_of(he, struct c2c_hist_entry, he);
800 per = get_percent(c2c_he);
801
802#ifdef HAVE_SLANG_SUPPORT
803 if (use_browser)
804 return __hpp__slsmg_color_printf(hpp, "%*.2f%%", width - 1, per);
805#endif
806 return hpp_color_scnprintf(hpp, "%*.2f%%", width - 1, per);
807}
808
809static double percent_costly_snoop(struct c2c_hist_entry *c2c_he)
810{
811 struct c2c_hists *hists;
812 struct c2c_stats *stats;
813 struct c2c_stats *total;
814 int tot = 0, st = 0;
815 double p;
816
817 hists = container_of(c2c_he->he.hists, struct c2c_hists, hists);
818 stats = &c2c_he->stats;
819 total = &hists->stats;
820
821 switch (c2c.display) {
822 case DISPLAY_RMT_HITM:
823 st = stats->rmt_hitm;
824 tot = total->rmt_hitm;
825 break;
826 case DISPLAY_LCL_HITM:
827 st = stats->lcl_hitm;
828 tot = total->lcl_hitm;
829 break;
830 case DISPLAY_TOT_HITM:
831 st = stats->tot_hitm;
832 tot = total->tot_hitm;
833 break;
834 case DISPLAY_SNP_PEER:
835 st = stats->tot_peer;
836 tot = total->tot_peer;
837 break;
838 default:
839 break;
840 }
841
842 p = tot ? (double) st / tot : 0;
843
844 return 100 * p;
845}
846
847#define PERC_STR(__s, __v) \
848({ \
849 scnprintf(__s, sizeof(__s), "%.2F%%", __v); \
850 __s; \
851})
852
853static int
854percent_costly_snoop_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
855 struct hist_entry *he)
856{
857 struct c2c_hist_entry *c2c_he;
858 int width = c2c_width(fmt, hpp, he->hists);
859 char buf[10];
860 double per;
861
862 c2c_he = container_of(he, struct c2c_hist_entry, he);
863 per = percent_costly_snoop(c2c_he);
864 return scnprintf(hpp->buf, hpp->size, "%*s", width, PERC_STR(buf, per));
865}
866
867static int
868percent_costly_snoop_color(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
869 struct hist_entry *he)
870{
871 return percent_color(fmt, hpp, he, percent_costly_snoop);
872}
873
874static int64_t
875percent_costly_snoop_cmp(struct perf_hpp_fmt *fmt __maybe_unused,
876 struct hist_entry *left, struct hist_entry *right)
877{
878 struct c2c_hist_entry *c2c_left;
879 struct c2c_hist_entry *c2c_right;
880 double per_left;
881 double per_right;
882
883 c2c_left = container_of(left, struct c2c_hist_entry, he);
884 c2c_right = container_of(right, struct c2c_hist_entry, he);
885
886 per_left = percent_costly_snoop(c2c_left);
887 per_right = percent_costly_snoop(c2c_right);
888
889 return per_left - per_right;
890}
891
892static struct c2c_stats *he_stats(struct hist_entry *he)
893{
894 struct c2c_hist_entry *c2c_he;
895
896 c2c_he = container_of(he, struct c2c_hist_entry, he);
897 return &c2c_he->stats;
898}
899
900static struct c2c_stats *total_stats(struct hist_entry *he)
901{
902 struct c2c_hists *hists;
903
904 hists = container_of(he->hists, struct c2c_hists, hists);
905 return &hists->stats;
906}
907
908static double percent(u32 st, u32 tot)
909{
910 return tot ? 100. * (double) st / (double) tot : 0;
911}
912
913#define PERCENT(__h, __f) percent(he_stats(__h)->__f, total_stats(__h)->__f)
914
915#define PERCENT_FN(__f) \
916static double percent_ ## __f(struct c2c_hist_entry *c2c_he) \
917{ \
918 struct c2c_hists *hists; \
919 \
920 hists = container_of(c2c_he->he.hists, struct c2c_hists, hists); \
921 return percent(c2c_he->stats.__f, hists->stats.__f); \
922}
923
924PERCENT_FN(rmt_hitm)
925PERCENT_FN(lcl_hitm)
926PERCENT_FN(rmt_peer)
927PERCENT_FN(lcl_peer)
928PERCENT_FN(st_l1hit)
929PERCENT_FN(st_l1miss)
930PERCENT_FN(st_na)
931
932static int
933percent_rmt_hitm_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
934 struct hist_entry *he)
935{
936 int width = c2c_width(fmt, hpp, he->hists);
937 double per = PERCENT(he, rmt_hitm);
938 char buf[10];
939
940 return scnprintf(hpp->buf, hpp->size, "%*s", width, PERC_STR(buf, per));
941}
942
943static int
944percent_rmt_hitm_color(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
945 struct hist_entry *he)
946{
947 return percent_color(fmt, hpp, he, percent_rmt_hitm);
948}
949
950static int64_t
951percent_rmt_hitm_cmp(struct perf_hpp_fmt *fmt __maybe_unused,
952 struct hist_entry *left, struct hist_entry *right)
953{
954 double per_left;
955 double per_right;
956
957 per_left = PERCENT(left, rmt_hitm);
958 per_right = PERCENT(right, rmt_hitm);
959
960 return per_left - per_right;
961}
962
963static int
964percent_lcl_hitm_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
965 struct hist_entry *he)
966{
967 int width = c2c_width(fmt, hpp, he->hists);
968 double per = PERCENT(he, lcl_hitm);
969 char buf[10];
970
971 return scnprintf(hpp->buf, hpp->size, "%*s", width, PERC_STR(buf, per));
972}
973
974static int
975percent_lcl_hitm_color(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
976 struct hist_entry *he)
977{
978 return percent_color(fmt, hpp, he, percent_lcl_hitm);
979}
980
981static int64_t
982percent_lcl_hitm_cmp(struct perf_hpp_fmt *fmt __maybe_unused,
983 struct hist_entry *left, struct hist_entry *right)
984{
985 double per_left;
986 double per_right;
987
988 per_left = PERCENT(left, lcl_hitm);
989 per_right = PERCENT(right, lcl_hitm);
990
991 return per_left - per_right;
992}
993
994static int
995percent_lcl_peer_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
996 struct hist_entry *he)
997{
998 int width = c2c_width(fmt, hpp, he->hists);
999 double per = PERCENT(he, lcl_peer);
1000 char buf[10];
1001
1002 return scnprintf(hpp->buf, hpp->size, "%*s", width, PERC_STR(buf, per));
1003}
1004
1005static int
1006percent_lcl_peer_color(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
1007 struct hist_entry *he)
1008{
1009 return percent_color(fmt, hpp, he, percent_lcl_peer);
1010}
1011
1012static int64_t
1013percent_lcl_peer_cmp(struct perf_hpp_fmt *fmt __maybe_unused,
1014 struct hist_entry *left, struct hist_entry *right)
1015{
1016 double per_left;
1017 double per_right;
1018
1019 per_left = PERCENT(left, lcl_peer);
1020 per_right = PERCENT(right, lcl_peer);
1021
1022 return per_left - per_right;
1023}
1024
1025static int
1026percent_rmt_peer_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
1027 struct hist_entry *he)
1028{
1029 int width = c2c_width(fmt, hpp, he->hists);
1030 double per = PERCENT(he, rmt_peer);
1031 char buf[10];
1032
1033 return scnprintf(hpp->buf, hpp->size, "%*s", width, PERC_STR(buf, per));
1034}
1035
1036static int
1037percent_rmt_peer_color(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
1038 struct hist_entry *he)
1039{
1040 return percent_color(fmt, hpp, he, percent_rmt_peer);
1041}
1042
1043static int64_t
1044percent_rmt_peer_cmp(struct perf_hpp_fmt *fmt __maybe_unused,
1045 struct hist_entry *left, struct hist_entry *right)
1046{
1047 double per_left;
1048 double per_right;
1049
1050 per_left = PERCENT(left, rmt_peer);
1051 per_right = PERCENT(right, rmt_peer);
1052
1053 return per_left - per_right;
1054}
1055
1056static int
1057percent_stores_l1hit_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
1058 struct hist_entry *he)
1059{
1060 int width = c2c_width(fmt, hpp, he->hists);
1061 double per = PERCENT(he, st_l1hit);
1062 char buf[10];
1063
1064 return scnprintf(hpp->buf, hpp->size, "%*s", width, PERC_STR(buf, per));
1065}
1066
1067static int
1068percent_stores_l1hit_color(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
1069 struct hist_entry *he)
1070{
1071 return percent_color(fmt, hpp, he, percent_st_l1hit);
1072}
1073
1074static int64_t
1075percent_stores_l1hit_cmp(struct perf_hpp_fmt *fmt __maybe_unused,
1076 struct hist_entry *left, struct hist_entry *right)
1077{
1078 double per_left;
1079 double per_right;
1080
1081 per_left = PERCENT(left, st_l1hit);
1082 per_right = PERCENT(right, st_l1hit);
1083
1084 return per_left - per_right;
1085}
1086
1087static int
1088percent_stores_l1miss_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
1089 struct hist_entry *he)
1090{
1091 int width = c2c_width(fmt, hpp, he->hists);
1092 double per = PERCENT(he, st_l1miss);
1093 char buf[10];
1094
1095 return scnprintf(hpp->buf, hpp->size, "%*s", width, PERC_STR(buf, per));
1096}
1097
1098static int
1099percent_stores_l1miss_color(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
1100 struct hist_entry *he)
1101{
1102 return percent_color(fmt, hpp, he, percent_st_l1miss);
1103}
1104
1105static int64_t
1106percent_stores_l1miss_cmp(struct perf_hpp_fmt *fmt __maybe_unused,
1107 struct hist_entry *left, struct hist_entry *right)
1108{
1109 double per_left;
1110 double per_right;
1111
1112 per_left = PERCENT(left, st_l1miss);
1113 per_right = PERCENT(right, st_l1miss);
1114
1115 return per_left - per_right;
1116}
1117
1118static int
1119percent_stores_na_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
1120 struct hist_entry *he)
1121{
1122 int width = c2c_width(fmt, hpp, he->hists);
1123 double per = PERCENT(he, st_na);
1124 char buf[10];
1125
1126 return scnprintf(hpp->buf, hpp->size, "%*s", width, PERC_STR(buf, per));
1127}
1128
1129static int
1130percent_stores_na_color(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
1131 struct hist_entry *he)
1132{
1133 return percent_color(fmt, hpp, he, percent_st_na);
1134}
1135
1136static int64_t
1137percent_stores_na_cmp(struct perf_hpp_fmt *fmt __maybe_unused,
1138 struct hist_entry *left, struct hist_entry *right)
1139{
1140 double per_left;
1141 double per_right;
1142
1143 per_left = PERCENT(left, st_na);
1144 per_right = PERCENT(right, st_na);
1145
1146 return per_left - per_right;
1147}
1148
1149STAT_FN(lcl_dram)
1150STAT_FN(rmt_dram)
1151
1152static int
1153pid_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
1154 struct hist_entry *he)
1155{
1156 int width = c2c_width(fmt, hpp, he->hists);
1157
1158 return scnprintf(hpp->buf, hpp->size, "%*d", width, thread__pid(he->thread));
1159}
1160
1161static int64_t
1162pid_cmp(struct perf_hpp_fmt *fmt __maybe_unused,
1163 struct hist_entry *left, struct hist_entry *right)
1164{
1165 return thread__pid(left->thread) - thread__pid(right->thread);
1166}
1167
1168static int64_t
1169empty_cmp(struct perf_hpp_fmt *fmt __maybe_unused,
1170 struct hist_entry *left __maybe_unused,
1171 struct hist_entry *right __maybe_unused)
1172{
1173 return 0;
1174}
1175
1176static int display_metrics(struct perf_hpp *hpp, u32 val, u32 sum)
1177{
1178 int ret;
1179
1180 if (sum != 0)
1181 ret = scnprintf(hpp->buf, hpp->size, "%5.1f%% ",
1182 percent(val, sum));
1183 else
1184 ret = scnprintf(hpp->buf, hpp->size, "%6s ", "n/a");
1185
1186 return ret;
1187}
1188
1189static int
1190node_entry(struct perf_hpp_fmt *fmt __maybe_unused, struct perf_hpp *hpp,
1191 struct hist_entry *he)
1192{
1193 struct c2c_hist_entry *c2c_he;
1194 bool first = true;
1195 int node;
1196 int ret = 0;
1197
1198 c2c_he = container_of(he, struct c2c_hist_entry, he);
1199
1200 for (node = 0; node < c2c.nodes_cnt; node++) {
1201 DECLARE_BITMAP(set, c2c.cpus_cnt);
1202
1203 bitmap_zero(set, c2c.cpus_cnt);
1204 bitmap_and(set, c2c_he->cpuset, c2c.nodes[node], c2c.cpus_cnt);
1205
1206 if (bitmap_empty(set, c2c.cpus_cnt)) {
1207 if (c2c.node_info == 1) {
1208 ret = scnprintf(hpp->buf, hpp->size, "%21s", " ");
1209 advance_hpp(hpp, ret);
1210 }
1211 continue;
1212 }
1213
1214 if (!first) {
1215 ret = scnprintf(hpp->buf, hpp->size, " ");
1216 advance_hpp(hpp, ret);
1217 }
1218
1219 switch (c2c.node_info) {
1220 case 0:
1221 ret = scnprintf(hpp->buf, hpp->size, "%2d", node);
1222 advance_hpp(hpp, ret);
1223 break;
1224 case 1:
1225 {
1226 int num = bitmap_weight(set, c2c.cpus_cnt);
1227 struct c2c_stats *stats = &c2c_he->node_stats[node];
1228
1229 ret = scnprintf(hpp->buf, hpp->size, "%2d{%2d ", node, num);
1230 advance_hpp(hpp, ret);
1231
1232 switch (c2c.display) {
1233 case DISPLAY_RMT_HITM:
1234 ret = display_metrics(hpp, stats->rmt_hitm,
1235 c2c_he->stats.rmt_hitm);
1236 break;
1237 case DISPLAY_LCL_HITM:
1238 ret = display_metrics(hpp, stats->lcl_hitm,
1239 c2c_he->stats.lcl_hitm);
1240 break;
1241 case DISPLAY_TOT_HITM:
1242 ret = display_metrics(hpp, stats->tot_hitm,
1243 c2c_he->stats.tot_hitm);
1244 break;
1245 case DISPLAY_SNP_PEER:
1246 ret = display_metrics(hpp, stats->tot_peer,
1247 c2c_he->stats.tot_peer);
1248 break;
1249 default:
1250 break;
1251 }
1252
1253 advance_hpp(hpp, ret);
1254
1255 if (c2c_he->stats.store > 0) {
1256 ret = scnprintf(hpp->buf, hpp->size, "%5.1f%%}",
1257 percent(stats->store, c2c_he->stats.store));
1258 } else {
1259 ret = scnprintf(hpp->buf, hpp->size, "%6s}", "n/a");
1260 }
1261
1262 advance_hpp(hpp, ret);
1263 break;
1264 }
1265 case 2:
1266 ret = scnprintf(hpp->buf, hpp->size, "%2d{", node);
1267 advance_hpp(hpp, ret);
1268
1269 ret = bitmap_scnprintf(set, c2c.cpus_cnt, hpp->buf, hpp->size);
1270 advance_hpp(hpp, ret);
1271
1272 ret = scnprintf(hpp->buf, hpp->size, "}");
1273 advance_hpp(hpp, ret);
1274 break;
1275 default:
1276 break;
1277 }
1278
1279 first = false;
1280 }
1281
1282 return 0;
1283}
1284
1285static int
1286mean_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
1287 struct hist_entry *he, double mean)
1288{
1289 int width = c2c_width(fmt, hpp, he->hists);
1290 char buf[10];
1291
1292 scnprintf(buf, 10, "%6.0f", mean);
1293 return scnprintf(hpp->buf, hpp->size, "%*s", width, buf);
1294}
1295
1296#define MEAN_ENTRY(__func, __val) \
1297static int \
1298__func(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp, struct hist_entry *he) \
1299{ \
1300 struct c2c_hist_entry *c2c_he; \
1301 c2c_he = container_of(he, struct c2c_hist_entry, he); \
1302 return mean_entry(fmt, hpp, he, avg_stats(&c2c_he->cstats.__val)); \
1303}
1304
1305MEAN_ENTRY(mean_rmt_entry, rmt_hitm);
1306MEAN_ENTRY(mean_lcl_entry, lcl_hitm);
1307MEAN_ENTRY(mean_load_entry, load);
1308MEAN_ENTRY(mean_rmt_peer_entry, rmt_peer);
1309MEAN_ENTRY(mean_lcl_peer_entry, lcl_peer);
1310
1311static int
1312cpucnt_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
1313 struct hist_entry *he)
1314{
1315 struct c2c_hist_entry *c2c_he;
1316 int width = c2c_width(fmt, hpp, he->hists);
1317 char buf[10];
1318
1319 c2c_he = container_of(he, struct c2c_hist_entry, he);
1320
1321 scnprintf(buf, 10, "%d", bitmap_weight(c2c_he->cpuset, c2c.cpus_cnt));
1322 return scnprintf(hpp->buf, hpp->size, "%*s", width, buf);
1323}
1324
1325static int
1326cl_idx_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
1327 struct hist_entry *he)
1328{
1329 struct c2c_hist_entry *c2c_he;
1330 int width = c2c_width(fmt, hpp, he->hists);
1331 char buf[10];
1332
1333 c2c_he = container_of(he, struct c2c_hist_entry, he);
1334
1335 scnprintf(buf, 10, "%u", c2c_he->cacheline_idx);
1336 return scnprintf(hpp->buf, hpp->size, "%*s", width, buf);
1337}
1338
1339static int
1340cl_idx_empty_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
1341 struct hist_entry *he)
1342{
1343 int width = c2c_width(fmt, hpp, he->hists);
1344
1345 return scnprintf(hpp->buf, hpp->size, "%*s", width, "");
1346}
1347
1348#define HEADER_LOW(__h) \
1349 { \
1350 .line[1] = { \
1351 .text = __h, \
1352 }, \
1353 }
1354
1355#define HEADER_BOTH(__h0, __h1) \
1356 { \
1357 .line[0] = { \
1358 .text = __h0, \
1359 }, \
1360 .line[1] = { \
1361 .text = __h1, \
1362 }, \
1363 }
1364
1365#define HEADER_SPAN(__h0, __h1, __s) \
1366 { \
1367 .line[0] = { \
1368 .text = __h0, \
1369 .span = __s, \
1370 }, \
1371 .line[1] = { \
1372 .text = __h1, \
1373 }, \
1374 }
1375
1376#define HEADER_SPAN_LOW(__h) \
1377 { \
1378 .line[1] = { \
1379 .text = __h, \
1380 }, \
1381 }
1382
1383static struct c2c_dimension dim_dcacheline = {
1384 .header = HEADER_SPAN("--- Cacheline ----", "Address", 2),
1385 .name = "dcacheline",
1386 .cmp = dcacheline_cmp,
1387 .entry = dcacheline_entry,
1388 .width = 18,
1389};
1390
1391static struct c2c_dimension dim_dcacheline_node = {
1392 .header = HEADER_LOW("Node"),
1393 .name = "dcacheline_node",
1394 .cmp = empty_cmp,
1395 .entry = dcacheline_node_entry,
1396 .width = 4,
1397};
1398
1399static struct c2c_dimension dim_dcacheline_count = {
1400 .header = HEADER_LOW("PA cnt"),
1401 .name = "dcacheline_count",
1402 .cmp = empty_cmp,
1403 .entry = dcacheline_node_count,
1404 .width = 6,
1405};
1406
1407static struct c2c_header header_offset_tui = HEADER_SPAN("-----", "Off", 2);
1408
1409static struct c2c_dimension dim_offset = {
1410 .header = HEADER_SPAN("--- Data address -", "Offset", 2),
1411 .name = "offset",
1412 .cmp = offset_cmp,
1413 .entry = offset_entry,
1414 .width = 18,
1415};
1416
1417static struct c2c_dimension dim_offset_node = {
1418 .header = HEADER_LOW("Node"),
1419 .name = "offset_node",
1420 .cmp = empty_cmp,
1421 .entry = dcacheline_node_entry,
1422 .width = 4,
1423};
1424
1425static struct c2c_dimension dim_iaddr = {
1426 .header = HEADER_LOW("Code address"),
1427 .name = "iaddr",
1428 .cmp = iaddr_cmp,
1429 .entry = iaddr_entry,
1430 .width = 18,
1431};
1432
1433static struct c2c_dimension dim_tot_hitm = {
1434 .header = HEADER_SPAN("------- Load Hitm -------", "Total", 2),
1435 .name = "tot_hitm",
1436 .cmp = tot_hitm_cmp,
1437 .entry = tot_hitm_entry,
1438 .width = 7,
1439};
1440
1441static struct c2c_dimension dim_lcl_hitm = {
1442 .header = HEADER_SPAN_LOW("LclHitm"),
1443 .name = "lcl_hitm",
1444 .cmp = lcl_hitm_cmp,
1445 .entry = lcl_hitm_entry,
1446 .width = 7,
1447};
1448
1449static struct c2c_dimension dim_rmt_hitm = {
1450 .header = HEADER_SPAN_LOW("RmtHitm"),
1451 .name = "rmt_hitm",
1452 .cmp = rmt_hitm_cmp,
1453 .entry = rmt_hitm_entry,
1454 .width = 7,
1455};
1456
1457static struct c2c_dimension dim_tot_peer = {
1458 .header = HEADER_SPAN("------- Load Peer -------", "Total", 2),
1459 .name = "tot_peer",
1460 .cmp = tot_peer_cmp,
1461 .entry = tot_peer_entry,
1462 .width = 7,
1463};
1464
1465static struct c2c_dimension dim_lcl_peer = {
1466 .header = HEADER_SPAN_LOW("Local"),
1467 .name = "lcl_peer",
1468 .cmp = lcl_peer_cmp,
1469 .entry = lcl_peer_entry,
1470 .width = 7,
1471};
1472
1473static struct c2c_dimension dim_rmt_peer = {
1474 .header = HEADER_SPAN_LOW("Remote"),
1475 .name = "rmt_peer",
1476 .cmp = rmt_peer_cmp,
1477 .entry = rmt_peer_entry,
1478 .width = 7,
1479};
1480
1481static struct c2c_dimension dim_cl_rmt_hitm = {
1482 .header = HEADER_SPAN("----- HITM -----", "Rmt", 1),
1483 .name = "cl_rmt_hitm",
1484 .cmp = rmt_hitm_cmp,
1485 .entry = rmt_hitm_entry,
1486 .width = 7,
1487};
1488
1489static struct c2c_dimension dim_cl_lcl_hitm = {
1490 .header = HEADER_SPAN_LOW("Lcl"),
1491 .name = "cl_lcl_hitm",
1492 .cmp = lcl_hitm_cmp,
1493 .entry = lcl_hitm_entry,
1494 .width = 7,
1495};
1496
1497static struct c2c_dimension dim_cl_rmt_peer = {
1498 .header = HEADER_SPAN("----- Peer -----", "Rmt", 1),
1499 .name = "cl_rmt_peer",
1500 .cmp = rmt_peer_cmp,
1501 .entry = rmt_peer_entry,
1502 .width = 7,
1503};
1504
1505static struct c2c_dimension dim_cl_lcl_peer = {
1506 .header = HEADER_SPAN_LOW("Lcl"),
1507 .name = "cl_lcl_peer",
1508 .cmp = lcl_peer_cmp,
1509 .entry = lcl_peer_entry,
1510 .width = 7,
1511};
1512
1513static struct c2c_dimension dim_tot_stores = {
1514 .header = HEADER_BOTH("Total", "Stores"),
1515 .name = "tot_stores",
1516 .cmp = store_cmp,
1517 .entry = store_entry,
1518 .width = 7,
1519};
1520
1521static struct c2c_dimension dim_stores_l1hit = {
1522 .header = HEADER_SPAN("--------- Stores --------", "L1Hit", 2),
1523 .name = "stores_l1hit",
1524 .cmp = st_l1hit_cmp,
1525 .entry = st_l1hit_entry,
1526 .width = 7,
1527};
1528
1529static struct c2c_dimension dim_stores_l1miss = {
1530 .header = HEADER_SPAN_LOW("L1Miss"),
1531 .name = "stores_l1miss",
1532 .cmp = st_l1miss_cmp,
1533 .entry = st_l1miss_entry,
1534 .width = 7,
1535};
1536
1537static struct c2c_dimension dim_stores_na = {
1538 .header = HEADER_SPAN_LOW("N/A"),
1539 .name = "stores_na",
1540 .cmp = st_na_cmp,
1541 .entry = st_na_entry,
1542 .width = 7,
1543};
1544
1545static struct c2c_dimension dim_cl_stores_l1hit = {
1546 .header = HEADER_SPAN("------- Store Refs ------", "L1 Hit", 2),
1547 .name = "cl_stores_l1hit",
1548 .cmp = st_l1hit_cmp,
1549 .entry = st_l1hit_entry,
1550 .width = 7,
1551};
1552
1553static struct c2c_dimension dim_cl_stores_l1miss = {
1554 .header = HEADER_SPAN_LOW("L1 Miss"),
1555 .name = "cl_stores_l1miss",
1556 .cmp = st_l1miss_cmp,
1557 .entry = st_l1miss_entry,
1558 .width = 7,
1559};
1560
1561static struct c2c_dimension dim_cl_stores_na = {
1562 .header = HEADER_SPAN_LOW("N/A"),
1563 .name = "cl_stores_na",
1564 .cmp = st_na_cmp,
1565 .entry = st_na_entry,
1566 .width = 7,
1567};
1568
1569static struct c2c_dimension dim_ld_fbhit = {
1570 .header = HEADER_SPAN("----- Core Load Hit -----", "FB", 2),
1571 .name = "ld_fbhit",
1572 .cmp = ld_fbhit_cmp,
1573 .entry = ld_fbhit_entry,
1574 .width = 7,
1575};
1576
1577static struct c2c_dimension dim_ld_l1hit = {
1578 .header = HEADER_SPAN_LOW("L1"),
1579 .name = "ld_l1hit",
1580 .cmp = ld_l1hit_cmp,
1581 .entry = ld_l1hit_entry,
1582 .width = 7,
1583};
1584
1585static struct c2c_dimension dim_ld_l2hit = {
1586 .header = HEADER_SPAN_LOW("L2"),
1587 .name = "ld_l2hit",
1588 .cmp = ld_l2hit_cmp,
1589 .entry = ld_l2hit_entry,
1590 .width = 7,
1591};
1592
1593static struct c2c_dimension dim_ld_llchit = {
1594 .header = HEADER_SPAN("- LLC Load Hit --", "LclHit", 1),
1595 .name = "ld_lclhit",
1596 .cmp = ld_llchit_cmp,
1597 .entry = ld_llchit_entry,
1598 .width = 8,
1599};
1600
1601static struct c2c_dimension dim_ld_rmthit = {
1602 .header = HEADER_SPAN("- RMT Load Hit --", "RmtHit", 1),
1603 .name = "ld_rmthit",
1604 .cmp = rmt_hit_cmp,
1605 .entry = rmt_hit_entry,
1606 .width = 8,
1607};
1608
1609static struct c2c_dimension dim_tot_recs = {
1610 .header = HEADER_BOTH("Total", "records"),
1611 .name = "tot_recs",
1612 .cmp = tot_recs_cmp,
1613 .entry = tot_recs_entry,
1614 .width = 7,
1615};
1616
1617static struct c2c_dimension dim_tot_loads = {
1618 .header = HEADER_BOTH("Total", "Loads"),
1619 .name = "tot_loads",
1620 .cmp = tot_loads_cmp,
1621 .entry = tot_loads_entry,
1622 .width = 7,
1623};
1624
1625static struct c2c_header percent_costly_snoop_header[] = {
1626 [DISPLAY_LCL_HITM] = HEADER_BOTH("Lcl", "Hitm"),
1627 [DISPLAY_RMT_HITM] = HEADER_BOTH("Rmt", "Hitm"),
1628 [DISPLAY_TOT_HITM] = HEADER_BOTH("Tot", "Hitm"),
1629 [DISPLAY_SNP_PEER] = HEADER_BOTH("Peer", "Snoop"),
1630};
1631
1632static struct c2c_dimension dim_percent_costly_snoop = {
1633 .name = "percent_costly_snoop",
1634 .cmp = percent_costly_snoop_cmp,
1635 .entry = percent_costly_snoop_entry,
1636 .color = percent_costly_snoop_color,
1637 .width = 7,
1638};
1639
1640static struct c2c_dimension dim_percent_rmt_hitm = {
1641 .header = HEADER_SPAN("----- HITM -----", "RmtHitm", 1),
1642 .name = "percent_rmt_hitm",
1643 .cmp = percent_rmt_hitm_cmp,
1644 .entry = percent_rmt_hitm_entry,
1645 .color = percent_rmt_hitm_color,
1646 .width = 7,
1647};
1648
1649static struct c2c_dimension dim_percent_lcl_hitm = {
1650 .header = HEADER_SPAN_LOW("LclHitm"),
1651 .name = "percent_lcl_hitm",
1652 .cmp = percent_lcl_hitm_cmp,
1653 .entry = percent_lcl_hitm_entry,
1654 .color = percent_lcl_hitm_color,
1655 .width = 7,
1656};
1657
1658static struct c2c_dimension dim_percent_rmt_peer = {
1659 .header = HEADER_SPAN("-- Peer Snoop --", "Rmt", 1),
1660 .name = "percent_rmt_peer",
1661 .cmp = percent_rmt_peer_cmp,
1662 .entry = percent_rmt_peer_entry,
1663 .color = percent_rmt_peer_color,
1664 .width = 7,
1665};
1666
1667static struct c2c_dimension dim_percent_lcl_peer = {
1668 .header = HEADER_SPAN_LOW("Lcl"),
1669 .name = "percent_lcl_peer",
1670 .cmp = percent_lcl_peer_cmp,
1671 .entry = percent_lcl_peer_entry,
1672 .color = percent_lcl_peer_color,
1673 .width = 7,
1674};
1675
1676static struct c2c_dimension dim_percent_stores_l1hit = {
1677 .header = HEADER_SPAN("------- Store Refs ------", "L1 Hit", 2),
1678 .name = "percent_stores_l1hit",
1679 .cmp = percent_stores_l1hit_cmp,
1680 .entry = percent_stores_l1hit_entry,
1681 .color = percent_stores_l1hit_color,
1682 .width = 7,
1683};
1684
1685static struct c2c_dimension dim_percent_stores_l1miss = {
1686 .header = HEADER_SPAN_LOW("L1 Miss"),
1687 .name = "percent_stores_l1miss",
1688 .cmp = percent_stores_l1miss_cmp,
1689 .entry = percent_stores_l1miss_entry,
1690 .color = percent_stores_l1miss_color,
1691 .width = 7,
1692};
1693
1694static struct c2c_dimension dim_percent_stores_na = {
1695 .header = HEADER_SPAN_LOW("N/A"),
1696 .name = "percent_stores_na",
1697 .cmp = percent_stores_na_cmp,
1698 .entry = percent_stores_na_entry,
1699 .color = percent_stores_na_color,
1700 .width = 7,
1701};
1702
1703static struct c2c_dimension dim_dram_lcl = {
1704 .header = HEADER_SPAN("--- Load Dram ----", "Lcl", 1),
1705 .name = "dram_lcl",
1706 .cmp = lcl_dram_cmp,
1707 .entry = lcl_dram_entry,
1708 .width = 8,
1709};
1710
1711static struct c2c_dimension dim_dram_rmt = {
1712 .header = HEADER_SPAN_LOW("Rmt"),
1713 .name = "dram_rmt",
1714 .cmp = rmt_dram_cmp,
1715 .entry = rmt_dram_entry,
1716 .width = 8,
1717};
1718
1719static struct c2c_dimension dim_pid = {
1720 .header = HEADER_LOW("Pid"),
1721 .name = "pid",
1722 .cmp = pid_cmp,
1723 .entry = pid_entry,
1724 .width = 7,
1725};
1726
1727static struct c2c_dimension dim_tid = {
1728 .header = HEADER_LOW("Tid"),
1729 .name = "tid",
1730 .se = &sort_thread,
1731};
1732
1733static struct c2c_dimension dim_symbol = {
1734 .name = "symbol",
1735 .se = &sort_sym,
1736};
1737
1738static struct c2c_dimension dim_dso = {
1739 .header = HEADER_BOTH("Shared", "Object"),
1740 .name = "dso",
1741 .se = &sort_dso,
1742};
1743
1744static struct c2c_dimension dim_node = {
1745 .name = "node",
1746 .cmp = empty_cmp,
1747 .entry = node_entry,
1748 .width = 4,
1749};
1750
1751static struct c2c_dimension dim_mean_rmt = {
1752 .header = HEADER_SPAN("---------- cycles ----------", "rmt hitm", 2),
1753 .name = "mean_rmt",
1754 .cmp = empty_cmp,
1755 .entry = mean_rmt_entry,
1756 .width = 8,
1757};
1758
1759static struct c2c_dimension dim_mean_lcl = {
1760 .header = HEADER_SPAN_LOW("lcl hitm"),
1761 .name = "mean_lcl",
1762 .cmp = empty_cmp,
1763 .entry = mean_lcl_entry,
1764 .width = 8,
1765};
1766
1767static struct c2c_dimension dim_mean_load = {
1768 .header = HEADER_SPAN_LOW("load"),
1769 .name = "mean_load",
1770 .cmp = empty_cmp,
1771 .entry = mean_load_entry,
1772 .width = 8,
1773};
1774
1775static struct c2c_dimension dim_mean_rmt_peer = {
1776 .header = HEADER_SPAN("---------- cycles ----------", "rmt peer", 2),
1777 .name = "mean_rmt_peer",
1778 .cmp = empty_cmp,
1779 .entry = mean_rmt_peer_entry,
1780 .width = 8,
1781};
1782
1783static struct c2c_dimension dim_mean_lcl_peer = {
1784 .header = HEADER_SPAN_LOW("lcl peer"),
1785 .name = "mean_lcl_peer",
1786 .cmp = empty_cmp,
1787 .entry = mean_lcl_peer_entry,
1788 .width = 8,
1789};
1790
1791static struct c2c_dimension dim_cpucnt = {
1792 .header = HEADER_BOTH("cpu", "cnt"),
1793 .name = "cpucnt",
1794 .cmp = empty_cmp,
1795 .entry = cpucnt_entry,
1796 .width = 8,
1797};
1798
1799static struct c2c_dimension dim_srcline = {
1800 .name = "cl_srcline",
1801 .se = &sort_srcline,
1802};
1803
1804static struct c2c_dimension dim_dcacheline_idx = {
1805 .header = HEADER_LOW("Index"),
1806 .name = "cl_idx",
1807 .cmp = empty_cmp,
1808 .entry = cl_idx_entry,
1809 .width = 5,
1810};
1811
1812static struct c2c_dimension dim_dcacheline_num = {
1813 .header = HEADER_LOW("Num"),
1814 .name = "cl_num",
1815 .cmp = empty_cmp,
1816 .entry = cl_idx_entry,
1817 .width = 5,
1818};
1819
1820static struct c2c_dimension dim_dcacheline_num_empty = {
1821 .header = HEADER_LOW("Num"),
1822 .name = "cl_num_empty",
1823 .cmp = empty_cmp,
1824 .entry = cl_idx_empty_entry,
1825 .width = 5,
1826};
1827
1828static struct c2c_dimension *dimensions[] = {
1829 &dim_dcacheline,
1830 &dim_dcacheline_node,
1831 &dim_dcacheline_count,
1832 &dim_offset,
1833 &dim_offset_node,
1834 &dim_iaddr,
1835 &dim_tot_hitm,
1836 &dim_lcl_hitm,
1837 &dim_rmt_hitm,
1838 &dim_tot_peer,
1839 &dim_lcl_peer,
1840 &dim_rmt_peer,
1841 &dim_cl_lcl_hitm,
1842 &dim_cl_rmt_hitm,
1843 &dim_cl_lcl_peer,
1844 &dim_cl_rmt_peer,
1845 &dim_tot_stores,
1846 &dim_stores_l1hit,
1847 &dim_stores_l1miss,
1848 &dim_stores_na,
1849 &dim_cl_stores_l1hit,
1850 &dim_cl_stores_l1miss,
1851 &dim_cl_stores_na,
1852 &dim_ld_fbhit,
1853 &dim_ld_l1hit,
1854 &dim_ld_l2hit,
1855 &dim_ld_llchit,
1856 &dim_ld_rmthit,
1857 &dim_tot_recs,
1858 &dim_tot_loads,
1859 &dim_percent_costly_snoop,
1860 &dim_percent_rmt_hitm,
1861 &dim_percent_lcl_hitm,
1862 &dim_percent_rmt_peer,
1863 &dim_percent_lcl_peer,
1864 &dim_percent_stores_l1hit,
1865 &dim_percent_stores_l1miss,
1866 &dim_percent_stores_na,
1867 &dim_dram_lcl,
1868 &dim_dram_rmt,
1869 &dim_pid,
1870 &dim_tid,
1871 &dim_symbol,
1872 &dim_dso,
1873 &dim_node,
1874 &dim_mean_rmt,
1875 &dim_mean_lcl,
1876 &dim_mean_rmt_peer,
1877 &dim_mean_lcl_peer,
1878 &dim_mean_load,
1879 &dim_cpucnt,
1880 &dim_srcline,
1881 &dim_dcacheline_idx,
1882 &dim_dcacheline_num,
1883 &dim_dcacheline_num_empty,
1884 NULL,
1885};
1886
1887static void fmt_free(struct perf_hpp_fmt *fmt)
1888{
1889 struct c2c_fmt *c2c_fmt;
1890
1891 c2c_fmt = container_of(fmt, struct c2c_fmt, fmt);
1892 free(c2c_fmt);
1893}
1894
1895static bool fmt_equal(struct perf_hpp_fmt *a, struct perf_hpp_fmt *b)
1896{
1897 struct c2c_fmt *c2c_a = container_of(a, struct c2c_fmt, fmt);
1898 struct c2c_fmt *c2c_b = container_of(b, struct c2c_fmt, fmt);
1899
1900 return c2c_a->dim == c2c_b->dim;
1901}
1902
1903static struct c2c_dimension *get_dimension(const char *name)
1904{
1905 unsigned int i;
1906
1907 for (i = 0; dimensions[i]; i++) {
1908 struct c2c_dimension *dim = dimensions[i];
1909
1910 if (!strcmp(dim->name, name))
1911 return dim;
1912 }
1913
1914 return NULL;
1915}
1916
1917static int c2c_se_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
1918 struct hist_entry *he)
1919{
1920 struct c2c_fmt *c2c_fmt = container_of(fmt, struct c2c_fmt, fmt);
1921 struct c2c_dimension *dim = c2c_fmt->dim;
1922 size_t len = fmt->user_len;
1923
1924 if (!len) {
1925 len = hists__col_len(he->hists, dim->se->se_width_idx);
1926
1927 if (dim == &dim_symbol || dim == &dim_srcline)
1928 len = symbol_width(he->hists, dim->se);
1929 }
1930
1931 return dim->se->se_snprintf(he, hpp->buf, hpp->size, len);
1932}
1933
1934static int64_t c2c_se_cmp(struct perf_hpp_fmt *fmt,
1935 struct hist_entry *a, struct hist_entry *b)
1936{
1937 struct c2c_fmt *c2c_fmt = container_of(fmt, struct c2c_fmt, fmt);
1938 struct c2c_dimension *dim = c2c_fmt->dim;
1939
1940 return dim->se->se_cmp(a, b);
1941}
1942
1943static int64_t c2c_se_collapse(struct perf_hpp_fmt *fmt,
1944 struct hist_entry *a, struct hist_entry *b)
1945{
1946 struct c2c_fmt *c2c_fmt = container_of(fmt, struct c2c_fmt, fmt);
1947 struct c2c_dimension *dim = c2c_fmt->dim;
1948 int64_t (*collapse_fn)(struct hist_entry *, struct hist_entry *);
1949
1950 collapse_fn = dim->se->se_collapse ?: dim->se->se_cmp;
1951 return collapse_fn(a, b);
1952}
1953
1954static struct c2c_fmt *get_format(const char *name)
1955{
1956 struct c2c_dimension *dim = get_dimension(name);
1957 struct c2c_fmt *c2c_fmt;
1958 struct perf_hpp_fmt *fmt;
1959
1960 if (!dim)
1961 return NULL;
1962
1963 c2c_fmt = zalloc(sizeof(*c2c_fmt));
1964 if (!c2c_fmt)
1965 return NULL;
1966
1967 c2c_fmt->dim = dim;
1968
1969 fmt = &c2c_fmt->fmt;
1970 INIT_LIST_HEAD(&fmt->list);
1971 INIT_LIST_HEAD(&fmt->sort_list);
1972
1973 fmt->cmp = dim->se ? c2c_se_cmp : dim->cmp;
1974 fmt->sort = dim->se ? c2c_se_cmp : dim->cmp;
1975 fmt->color = dim->se ? NULL : dim->color;
1976 fmt->entry = dim->se ? c2c_se_entry : dim->entry;
1977 fmt->header = c2c_header;
1978 fmt->width = c2c_width;
1979 fmt->collapse = dim->se ? c2c_se_collapse : dim->cmp;
1980 fmt->equal = fmt_equal;
1981 fmt->free = fmt_free;
1982
1983 return c2c_fmt;
1984}
1985
1986static int c2c_hists__init_output(struct perf_hpp_list *hpp_list, char *name)
1987{
1988 struct c2c_fmt *c2c_fmt = get_format(name);
1989
1990 if (!c2c_fmt) {
1991 reset_dimensions();
1992 return output_field_add(hpp_list, name);
1993 }
1994
1995 perf_hpp_list__column_register(hpp_list, &c2c_fmt->fmt);
1996 return 0;
1997}
1998
1999static int c2c_hists__init_sort(struct perf_hpp_list *hpp_list, char *name)
2000{
2001 struct c2c_fmt *c2c_fmt = get_format(name);
2002 struct c2c_dimension *dim;
2003
2004 if (!c2c_fmt) {
2005 reset_dimensions();
2006 return sort_dimension__add(hpp_list, name, NULL, 0);
2007 }
2008
2009 dim = c2c_fmt->dim;
2010 if (dim == &dim_dso)
2011 hpp_list->dso = 1;
2012
2013 perf_hpp_list__register_sort_field(hpp_list, &c2c_fmt->fmt);
2014 return 0;
2015}
2016
2017#define PARSE_LIST(_list, _fn) \
2018 do { \
2019 char *tmp, *tok; \
2020 ret = 0; \
2021 \
2022 if (!_list) \
2023 break; \
2024 \
2025 for (tok = strtok_r((char *)_list, ", ", &tmp); \
2026 tok; tok = strtok_r(NULL, ", ", &tmp)) { \
2027 ret = _fn(hpp_list, tok); \
2028 if (ret == -EINVAL) { \
2029 pr_err("Invalid --fields key: `%s'", tok); \
2030 break; \
2031 } else if (ret == -ESRCH) { \
2032 pr_err("Unknown --fields key: `%s'", tok); \
2033 break; \
2034 } \
2035 } \
2036 } while (0)
2037
2038static int hpp_list__parse(struct perf_hpp_list *hpp_list,
2039 const char *output_,
2040 const char *sort_)
2041{
2042 char *output = output_ ? strdup(output_) : NULL;
2043 char *sort = sort_ ? strdup(sort_) : NULL;
2044 int ret;
2045
2046 PARSE_LIST(output, c2c_hists__init_output);
2047 PARSE_LIST(sort, c2c_hists__init_sort);
2048
2049 /* copy sort keys to output fields */
2050 perf_hpp__setup_output_field(hpp_list);
2051
2052 /*
2053 * We dont need other sorting keys other than those
2054 * we already specified. It also really slows down
2055 * the processing a lot with big number of output
2056 * fields, so switching this off for c2c.
2057 */
2058
2059#if 0
2060 /* and then copy output fields to sort keys */
2061 perf_hpp__append_sort_keys(&hists->list);
2062#endif
2063
2064 free(output);
2065 free(sort);
2066 return ret;
2067}
2068
2069static int c2c_hists__init(struct c2c_hists *hists,
2070 const char *sort,
2071 int nr_header_lines)
2072{
2073 __hists__init(&hists->hists, &hists->list);
2074
2075 /*
2076 * Initialize only with sort fields, we need to resort
2077 * later anyway, and that's where we add output fields
2078 * as well.
2079 */
2080 perf_hpp_list__init(&hists->list);
2081
2082 /* Overload number of header lines.*/
2083 hists->list.nr_header_lines = nr_header_lines;
2084
2085 return hpp_list__parse(&hists->list, NULL, sort);
2086}
2087
2088static int c2c_hists__reinit(struct c2c_hists *c2c_hists,
2089 const char *output,
2090 const char *sort)
2091{
2092 perf_hpp__reset_output_field(&c2c_hists->list);
2093 return hpp_list__parse(&c2c_hists->list, output, sort);
2094}
2095
2096#define DISPLAY_LINE_LIMIT 0.001
2097
2098static u8 filter_display(u32 val, u32 sum)
2099{
2100 if (sum == 0 || ((double)val / sum) < DISPLAY_LINE_LIMIT)
2101 return HIST_FILTER__C2C;
2102
2103 return 0;
2104}
2105
2106static bool he__display(struct hist_entry *he, struct c2c_stats *stats)
2107{
2108 struct c2c_hist_entry *c2c_he;
2109
2110 if (c2c.show_all)
2111 return true;
2112
2113 c2c_he = container_of(he, struct c2c_hist_entry, he);
2114
2115 switch (c2c.display) {
2116 case DISPLAY_LCL_HITM:
2117 he->filtered = filter_display(c2c_he->stats.lcl_hitm,
2118 stats->lcl_hitm);
2119 break;
2120 case DISPLAY_RMT_HITM:
2121 he->filtered = filter_display(c2c_he->stats.rmt_hitm,
2122 stats->rmt_hitm);
2123 break;
2124 case DISPLAY_TOT_HITM:
2125 he->filtered = filter_display(c2c_he->stats.tot_hitm,
2126 stats->tot_hitm);
2127 break;
2128 case DISPLAY_SNP_PEER:
2129 he->filtered = filter_display(c2c_he->stats.tot_peer,
2130 stats->tot_peer);
2131 break;
2132 default:
2133 break;
2134 }
2135
2136 return he->filtered == 0;
2137}
2138
2139static inline bool is_valid_hist_entry(struct hist_entry *he)
2140{
2141 struct c2c_hist_entry *c2c_he;
2142 bool has_record = false;
2143
2144 c2c_he = container_of(he, struct c2c_hist_entry, he);
2145
2146 /* It's a valid entry if contains stores */
2147 if (c2c_he->stats.store)
2148 return true;
2149
2150 switch (c2c.display) {
2151 case DISPLAY_LCL_HITM:
2152 has_record = !!c2c_he->stats.lcl_hitm;
2153 break;
2154 case DISPLAY_RMT_HITM:
2155 has_record = !!c2c_he->stats.rmt_hitm;
2156 break;
2157 case DISPLAY_TOT_HITM:
2158 has_record = !!c2c_he->stats.tot_hitm;
2159 break;
2160 case DISPLAY_SNP_PEER:
2161 has_record = !!c2c_he->stats.tot_peer;
2162 default:
2163 break;
2164 }
2165
2166 return has_record;
2167}
2168
2169static void set_node_width(struct c2c_hist_entry *c2c_he, int len)
2170{
2171 struct c2c_dimension *dim;
2172
2173 dim = &c2c.hists == c2c_he->hists ?
2174 &dim_dcacheline_node : &dim_offset_node;
2175
2176 if (len > dim->width)
2177 dim->width = len;
2178}
2179
2180static int set_nodestr(struct c2c_hist_entry *c2c_he)
2181{
2182 char buf[30];
2183 int len;
2184
2185 if (c2c_he->nodestr)
2186 return 0;
2187
2188 if (!bitmap_empty(c2c_he->nodeset, c2c.nodes_cnt)) {
2189 len = bitmap_scnprintf(c2c_he->nodeset, c2c.nodes_cnt,
2190 buf, sizeof(buf));
2191 } else {
2192 len = scnprintf(buf, sizeof(buf), "N/A");
2193 }
2194
2195 set_node_width(c2c_he, len);
2196 c2c_he->nodestr = strdup(buf);
2197 return c2c_he->nodestr ? 0 : -ENOMEM;
2198}
2199
2200static void calc_width(struct c2c_hist_entry *c2c_he)
2201{
2202 struct c2c_hists *c2c_hists;
2203
2204 c2c_hists = container_of(c2c_he->he.hists, struct c2c_hists, hists);
2205 hists__calc_col_len(&c2c_hists->hists, &c2c_he->he);
2206 set_nodestr(c2c_he);
2207}
2208
2209static int filter_cb(struct hist_entry *he, void *arg __maybe_unused)
2210{
2211 struct c2c_hist_entry *c2c_he;
2212
2213 c2c_he = container_of(he, struct c2c_hist_entry, he);
2214
2215 if (c2c.show_src && !he->srcline)
2216 he->srcline = hist_entry__srcline(he);
2217
2218 calc_width(c2c_he);
2219
2220 if (!is_valid_hist_entry(he))
2221 he->filtered = HIST_FILTER__C2C;
2222
2223 return 0;
2224}
2225
2226static int resort_cl_cb(struct hist_entry *he, void *arg __maybe_unused)
2227{
2228 struct c2c_hist_entry *c2c_he;
2229 struct c2c_hists *c2c_hists;
2230 bool display = he__display(he, &c2c.shared_clines_stats);
2231
2232 c2c_he = container_of(he, struct c2c_hist_entry, he);
2233 c2c_hists = c2c_he->hists;
2234
2235 if (display && c2c_hists) {
2236 static unsigned int idx;
2237
2238 c2c_he->cacheline_idx = idx++;
2239 calc_width(c2c_he);
2240
2241 c2c_hists__reinit(c2c_hists, c2c.cl_output, c2c.cl_resort);
2242
2243 hists__collapse_resort(&c2c_hists->hists, NULL);
2244 hists__output_resort_cb(&c2c_hists->hists, NULL, filter_cb);
2245 }
2246
2247 return 0;
2248}
2249
2250static struct c2c_header header_node_0 = HEADER_LOW("Node");
2251static struct c2c_header header_node_1_hitms_stores =
2252 HEADER_LOW("Node{cpus %hitms %stores}");
2253static struct c2c_header header_node_1_peers_stores =
2254 HEADER_LOW("Node{cpus %peers %stores}");
2255static struct c2c_header header_node_2 = HEADER_LOW("Node{cpu list}");
2256
2257static void setup_nodes_header(void)
2258{
2259 switch (c2c.node_info) {
2260 case 0:
2261 dim_node.header = header_node_0;
2262 break;
2263 case 1:
2264 if (c2c.display == DISPLAY_SNP_PEER)
2265 dim_node.header = header_node_1_peers_stores;
2266 else
2267 dim_node.header = header_node_1_hitms_stores;
2268 break;
2269 case 2:
2270 dim_node.header = header_node_2;
2271 break;
2272 default:
2273 break;
2274 }
2275
2276 return;
2277}
2278
2279static int setup_nodes(struct perf_session *session)
2280{
2281 struct numa_node *n;
2282 unsigned long **nodes;
2283 int node, idx;
2284 struct perf_cpu cpu;
2285 int *cpu2node;
2286
2287 if (c2c.node_info > 2)
2288 c2c.node_info = 2;
2289
2290 c2c.nodes_cnt = session->header.env.nr_numa_nodes;
2291 c2c.cpus_cnt = session->header.env.nr_cpus_avail;
2292
2293 n = session->header.env.numa_nodes;
2294 if (!n)
2295 return -EINVAL;
2296
2297 nodes = zalloc(sizeof(unsigned long *) * c2c.nodes_cnt);
2298 if (!nodes)
2299 return -ENOMEM;
2300
2301 c2c.nodes = nodes;
2302
2303 cpu2node = zalloc(sizeof(int) * c2c.cpus_cnt);
2304 if (!cpu2node)
2305 return -ENOMEM;
2306
2307 for (idx = 0; idx < c2c.cpus_cnt; idx++)
2308 cpu2node[idx] = -1;
2309
2310 c2c.cpu2node = cpu2node;
2311
2312 for (node = 0; node < c2c.nodes_cnt; node++) {
2313 struct perf_cpu_map *map = n[node].map;
2314 unsigned long *set;
2315
2316 set = bitmap_zalloc(c2c.cpus_cnt);
2317 if (!set)
2318 return -ENOMEM;
2319
2320 nodes[node] = set;
2321
2322 /* empty node, skip */
2323 if (perf_cpu_map__has_any_cpu_or_is_empty(map))
2324 continue;
2325
2326 perf_cpu_map__for_each_cpu(cpu, idx, map) {
2327 __set_bit(cpu.cpu, set);
2328
2329 if (WARN_ONCE(cpu2node[cpu.cpu] != -1, "node/cpu topology bug"))
2330 return -EINVAL;
2331
2332 cpu2node[cpu.cpu] = node;
2333 }
2334 }
2335
2336 setup_nodes_header();
2337 return 0;
2338}
2339
2340#define HAS_HITMS(__h) ((__h)->stats.lcl_hitm || (__h)->stats.rmt_hitm)
2341#define HAS_PEER(__h) ((__h)->stats.lcl_peer || (__h)->stats.rmt_peer)
2342
2343static int resort_shared_cl_cb(struct hist_entry *he, void *arg __maybe_unused)
2344{
2345 struct c2c_hist_entry *c2c_he;
2346 c2c_he = container_of(he, struct c2c_hist_entry, he);
2347
2348 if (HAS_HITMS(c2c_he) || HAS_PEER(c2c_he)) {
2349 c2c.shared_clines++;
2350 c2c_add_stats(&c2c.shared_clines_stats, &c2c_he->stats);
2351 }
2352
2353 return 0;
2354}
2355
2356static int hists__iterate_cb(struct hists *hists, hists__resort_cb_t cb)
2357{
2358 struct rb_node *next = rb_first_cached(&hists->entries);
2359 int ret = 0;
2360
2361 while (next) {
2362 struct hist_entry *he;
2363
2364 he = rb_entry(next, struct hist_entry, rb_node);
2365 ret = cb(he, NULL);
2366 if (ret)
2367 break;
2368 next = rb_next(&he->rb_node);
2369 }
2370
2371 return ret;
2372}
2373
2374static void print_c2c__display_stats(FILE *out)
2375{
2376 int llc_misses;
2377 struct c2c_stats *stats = &c2c.hists.stats;
2378
2379 llc_misses = get_load_llc_misses(stats);
2380
2381 fprintf(out, "=================================================\n");
2382 fprintf(out, " Trace Event Information \n");
2383 fprintf(out, "=================================================\n");
2384 fprintf(out, " Total records : %10d\n", stats->nr_entries);
2385 fprintf(out, " Locked Load/Store Operations : %10d\n", stats->locks);
2386 fprintf(out, " Load Operations : %10d\n", stats->load);
2387 fprintf(out, " Loads - uncacheable : %10d\n", stats->ld_uncache);
2388 fprintf(out, " Loads - IO : %10d\n", stats->ld_io);
2389 fprintf(out, " Loads - Miss : %10d\n", stats->ld_miss);
2390 fprintf(out, " Loads - no mapping : %10d\n", stats->ld_noadrs);
2391 fprintf(out, " Load Fill Buffer Hit : %10d\n", stats->ld_fbhit);
2392 fprintf(out, " Load L1D hit : %10d\n", stats->ld_l1hit);
2393 fprintf(out, " Load L2D hit : %10d\n", stats->ld_l2hit);
2394 fprintf(out, " Load LLC hit : %10d\n", stats->ld_llchit + stats->lcl_hitm);
2395 fprintf(out, " Load Local HITM : %10d\n", stats->lcl_hitm);
2396 fprintf(out, " Load Remote HITM : %10d\n", stats->rmt_hitm);
2397 fprintf(out, " Load Remote HIT : %10d\n", stats->rmt_hit);
2398 fprintf(out, " Load Local DRAM : %10d\n", stats->lcl_dram);
2399 fprintf(out, " Load Remote DRAM : %10d\n", stats->rmt_dram);
2400 fprintf(out, " Load MESI State Exclusive : %10d\n", stats->ld_excl);
2401 fprintf(out, " Load MESI State Shared : %10d\n", stats->ld_shared);
2402 fprintf(out, " Load LLC Misses : %10d\n", llc_misses);
2403 fprintf(out, " Load access blocked by data : %10d\n", stats->blk_data);
2404 fprintf(out, " Load access blocked by address : %10d\n", stats->blk_addr);
2405 fprintf(out, " Load HIT Local Peer : %10d\n", stats->lcl_peer);
2406 fprintf(out, " Load HIT Remote Peer : %10d\n", stats->rmt_peer);
2407 fprintf(out, " LLC Misses to Local DRAM : %10.1f%%\n", ((double)stats->lcl_dram/(double)llc_misses) * 100.);
2408 fprintf(out, " LLC Misses to Remote DRAM : %10.1f%%\n", ((double)stats->rmt_dram/(double)llc_misses) * 100.);
2409 fprintf(out, " LLC Misses to Remote cache (HIT) : %10.1f%%\n", ((double)stats->rmt_hit /(double)llc_misses) * 100.);
2410 fprintf(out, " LLC Misses to Remote cache (HITM) : %10.1f%%\n", ((double)stats->rmt_hitm/(double)llc_misses) * 100.);
2411 fprintf(out, " Store Operations : %10d\n", stats->store);
2412 fprintf(out, " Store - uncacheable : %10d\n", stats->st_uncache);
2413 fprintf(out, " Store - no mapping : %10d\n", stats->st_noadrs);
2414 fprintf(out, " Store L1D Hit : %10d\n", stats->st_l1hit);
2415 fprintf(out, " Store L1D Miss : %10d\n", stats->st_l1miss);
2416 fprintf(out, " Store No available memory level : %10d\n", stats->st_na);
2417 fprintf(out, " No Page Map Rejects : %10d\n", stats->nomap);
2418 fprintf(out, " Unable to parse data source : %10d\n", stats->noparse);
2419}
2420
2421static void print_shared_cacheline_info(FILE *out)
2422{
2423 struct c2c_stats *stats = &c2c.shared_clines_stats;
2424 int hitm_cnt = stats->lcl_hitm + stats->rmt_hitm;
2425
2426 fprintf(out, "=================================================\n");
2427 fprintf(out, " Global Shared Cache Line Event Information \n");
2428 fprintf(out, "=================================================\n");
2429 fprintf(out, " Total Shared Cache Lines : %10d\n", c2c.shared_clines);
2430 fprintf(out, " Load HITs on shared lines : %10d\n", stats->load);
2431 fprintf(out, " Fill Buffer Hits on shared lines : %10d\n", stats->ld_fbhit);
2432 fprintf(out, " L1D hits on shared lines : %10d\n", stats->ld_l1hit);
2433 fprintf(out, " L2D hits on shared lines : %10d\n", stats->ld_l2hit);
2434 fprintf(out, " LLC hits on shared lines : %10d\n", stats->ld_llchit + stats->lcl_hitm);
2435 fprintf(out, " Load hits on peer cache or nodes : %10d\n", stats->lcl_peer + stats->rmt_peer);
2436 fprintf(out, " Locked Access on shared lines : %10d\n", stats->locks);
2437 fprintf(out, " Blocked Access on shared lines : %10d\n", stats->blk_data + stats->blk_addr);
2438 fprintf(out, " Store HITs on shared lines : %10d\n", stats->store);
2439 fprintf(out, " Store L1D hits on shared lines : %10d\n", stats->st_l1hit);
2440 fprintf(out, " Store No available memory level : %10d\n", stats->st_na);
2441 fprintf(out, " Total Merged records : %10d\n", hitm_cnt + stats->store);
2442}
2443
2444static void print_cacheline(struct c2c_hists *c2c_hists,
2445 struct hist_entry *he_cl,
2446 struct perf_hpp_list *hpp_list,
2447 FILE *out)
2448{
2449 char bf[1000];
2450 struct perf_hpp hpp = {
2451 .buf = bf,
2452 .size = 1000,
2453 };
2454 static bool once;
2455
2456 if (!once) {
2457 hists__fprintf_headers(&c2c_hists->hists, out);
2458 once = true;
2459 } else {
2460 fprintf(out, "\n");
2461 }
2462
2463 fprintf(out, " ----------------------------------------------------------------------\n");
2464 __hist_entry__snprintf(he_cl, &hpp, hpp_list);
2465 fprintf(out, "%s\n", bf);
2466 fprintf(out, " ----------------------------------------------------------------------\n");
2467
2468 hists__fprintf(&c2c_hists->hists, false, 0, 0, 0, out, false);
2469}
2470
2471static void print_pareto(FILE *out)
2472{
2473 struct perf_hpp_list hpp_list;
2474 struct rb_node *nd;
2475 int ret;
2476 const char *cl_output;
2477
2478 if (c2c.display != DISPLAY_SNP_PEER)
2479 cl_output = "cl_num,"
2480 "cl_rmt_hitm,"
2481 "cl_lcl_hitm,"
2482 "cl_stores_l1hit,"
2483 "cl_stores_l1miss,"
2484 "cl_stores_na,"
2485 "dcacheline";
2486 else
2487 cl_output = "cl_num,"
2488 "cl_rmt_peer,"
2489 "cl_lcl_peer,"
2490 "cl_stores_l1hit,"
2491 "cl_stores_l1miss,"
2492 "cl_stores_na,"
2493 "dcacheline";
2494
2495 perf_hpp_list__init(&hpp_list);
2496 ret = hpp_list__parse(&hpp_list, cl_output, NULL);
2497
2498 if (WARN_ONCE(ret, "failed to setup sort entries\n"))
2499 return;
2500
2501 nd = rb_first_cached(&c2c.hists.hists.entries);
2502
2503 for (; nd; nd = rb_next(nd)) {
2504 struct hist_entry *he = rb_entry(nd, struct hist_entry, rb_node);
2505 struct c2c_hist_entry *c2c_he;
2506
2507 if (he->filtered)
2508 continue;
2509
2510 c2c_he = container_of(he, struct c2c_hist_entry, he);
2511 print_cacheline(c2c_he->hists, he, &hpp_list, out);
2512 }
2513}
2514
2515static void print_c2c_info(FILE *out, struct perf_session *session)
2516{
2517 struct evlist *evlist = session->evlist;
2518 struct evsel *evsel;
2519 bool first = true;
2520
2521 fprintf(out, "=================================================\n");
2522 fprintf(out, " c2c details \n");
2523 fprintf(out, "=================================================\n");
2524
2525 evlist__for_each_entry(evlist, evsel) {
2526 fprintf(out, "%-36s: %s\n", first ? " Events" : "", evsel__name(evsel));
2527 first = false;
2528 }
2529 fprintf(out, " Cachelines sort on : %s\n",
2530 display_str[c2c.display]);
2531 fprintf(out, " Cacheline data grouping : %s\n", c2c.cl_sort);
2532}
2533
2534static void perf_c2c__hists_fprintf(FILE *out, struct perf_session *session)
2535{
2536 setup_pager();
2537
2538 print_c2c__display_stats(out);
2539 fprintf(out, "\n");
2540 print_shared_cacheline_info(out);
2541 fprintf(out, "\n");
2542 print_c2c_info(out, session);
2543
2544 if (c2c.stats_only)
2545 return;
2546
2547 fprintf(out, "\n");
2548 fprintf(out, "=================================================\n");
2549 fprintf(out, " Shared Data Cache Line Table \n");
2550 fprintf(out, "=================================================\n");
2551 fprintf(out, "#\n");
2552
2553 hists__fprintf(&c2c.hists.hists, true, 0, 0, 0, stdout, true);
2554
2555 fprintf(out, "\n");
2556 fprintf(out, "=================================================\n");
2557 fprintf(out, " Shared Cache Line Distribution Pareto \n");
2558 fprintf(out, "=================================================\n");
2559 fprintf(out, "#\n");
2560
2561 print_pareto(out);
2562}
2563
2564#ifdef HAVE_SLANG_SUPPORT
2565static void c2c_browser__update_nr_entries(struct hist_browser *hb)
2566{
2567 u64 nr_entries = 0;
2568 struct rb_node *nd = rb_first_cached(&hb->hists->entries);
2569
2570 while (nd) {
2571 struct hist_entry *he = rb_entry(nd, struct hist_entry, rb_node);
2572
2573 if (!he->filtered)
2574 nr_entries++;
2575
2576 nd = rb_next(nd);
2577 }
2578
2579 hb->nr_non_filtered_entries = nr_entries;
2580}
2581
2582struct c2c_cacheline_browser {
2583 struct hist_browser hb;
2584 struct hist_entry *he;
2585};
2586
2587static int
2588perf_c2c_cacheline_browser__title(struct hist_browser *browser,
2589 char *bf, size_t size)
2590{
2591 struct c2c_cacheline_browser *cl_browser;
2592 struct hist_entry *he;
2593 uint64_t addr = 0;
2594
2595 cl_browser = container_of(browser, struct c2c_cacheline_browser, hb);
2596 he = cl_browser->he;
2597
2598 if (he->mem_info)
2599 addr = cl_address(he->mem_info->daddr.addr, chk_double_cl);
2600
2601 scnprintf(bf, size, "Cacheline 0x%lx", addr);
2602 return 0;
2603}
2604
2605static struct c2c_cacheline_browser*
2606c2c_cacheline_browser__new(struct hists *hists, struct hist_entry *he)
2607{
2608 struct c2c_cacheline_browser *browser;
2609
2610 browser = zalloc(sizeof(*browser));
2611 if (browser) {
2612 hist_browser__init(&browser->hb, hists);
2613 browser->hb.c2c_filter = true;
2614 browser->hb.title = perf_c2c_cacheline_browser__title;
2615 browser->he = he;
2616 }
2617
2618 return browser;
2619}
2620
2621static int perf_c2c__browse_cacheline(struct hist_entry *he)
2622{
2623 struct c2c_hist_entry *c2c_he;
2624 struct c2c_hists *c2c_hists;
2625 struct c2c_cacheline_browser *cl_browser;
2626 struct hist_browser *browser;
2627 int key = -1;
2628 static const char help[] =
2629 " ENTER Toggle callchains (if present) \n"
2630 " n Toggle Node details info \n"
2631 " s Toggle full length of symbol and source line columns \n"
2632 " q Return back to cacheline list \n";
2633
2634 if (!he)
2635 return 0;
2636
2637 /* Display compact version first. */
2638 c2c.symbol_full = false;
2639
2640 c2c_he = container_of(he, struct c2c_hist_entry, he);
2641 c2c_hists = c2c_he->hists;
2642
2643 cl_browser = c2c_cacheline_browser__new(&c2c_hists->hists, he);
2644 if (cl_browser == NULL)
2645 return -1;
2646
2647 browser = &cl_browser->hb;
2648
2649 /* reset abort key so that it can get Ctrl-C as a key */
2650 SLang_reset_tty();
2651 SLang_init_tty(0, 0, 0);
2652
2653 c2c_browser__update_nr_entries(browser);
2654
2655 while (1) {
2656 key = hist_browser__run(browser, "? - help", true, 0);
2657
2658 switch (key) {
2659 case 's':
2660 c2c.symbol_full = !c2c.symbol_full;
2661 break;
2662 case 'n':
2663 c2c.node_info = (c2c.node_info + 1) % 3;
2664 setup_nodes_header();
2665 break;
2666 case 'q':
2667 goto out;
2668 case '?':
2669 ui_browser__help_window(&browser->b, help);
2670 break;
2671 default:
2672 break;
2673 }
2674 }
2675
2676out:
2677 free(cl_browser);
2678 return 0;
2679}
2680
2681static int perf_c2c_browser__title(struct hist_browser *browser,
2682 char *bf, size_t size)
2683{
2684 scnprintf(bf, size,
2685 "Shared Data Cache Line Table "
2686 "(%lu entries, sorted on %s)",
2687 browser->nr_non_filtered_entries,
2688 display_str[c2c.display]);
2689 return 0;
2690}
2691
2692static struct hist_browser*
2693perf_c2c_browser__new(struct hists *hists)
2694{
2695 struct hist_browser *browser = hist_browser__new(hists);
2696
2697 if (browser) {
2698 browser->title = perf_c2c_browser__title;
2699 browser->c2c_filter = true;
2700 }
2701
2702 return browser;
2703}
2704
2705static int perf_c2c__hists_browse(struct hists *hists)
2706{
2707 struct hist_browser *browser;
2708 int key = -1;
2709 static const char help[] =
2710 " d Display cacheline details \n"
2711 " ENTER Toggle callchains (if present) \n"
2712 " q Quit \n";
2713
2714 browser = perf_c2c_browser__new(hists);
2715 if (browser == NULL)
2716 return -1;
2717
2718 /* reset abort key so that it can get Ctrl-C as a key */
2719 SLang_reset_tty();
2720 SLang_init_tty(0, 0, 0);
2721
2722 c2c_browser__update_nr_entries(browser);
2723
2724 while (1) {
2725 key = hist_browser__run(browser, "? - help", true, 0);
2726
2727 switch (key) {
2728 case 'q':
2729 goto out;
2730 case 'd':
2731 perf_c2c__browse_cacheline(browser->he_selection);
2732 break;
2733 case '?':
2734 ui_browser__help_window(&browser->b, help);
2735 break;
2736 default:
2737 break;
2738 }
2739 }
2740
2741out:
2742 hist_browser__delete(browser);
2743 return 0;
2744}
2745
2746static void perf_c2c_display(struct perf_session *session)
2747{
2748 if (use_browser == 0)
2749 perf_c2c__hists_fprintf(stdout, session);
2750 else
2751 perf_c2c__hists_browse(&c2c.hists.hists);
2752}
2753#else
2754static void perf_c2c_display(struct perf_session *session)
2755{
2756 use_browser = 0;
2757 perf_c2c__hists_fprintf(stdout, session);
2758}
2759#endif /* HAVE_SLANG_SUPPORT */
2760
2761static char *fill_line(const char *orig, int len)
2762{
2763 int i, j, olen = strlen(orig);
2764 char *buf;
2765
2766 buf = zalloc(len + 1);
2767 if (!buf)
2768 return NULL;
2769
2770 j = len / 2 - olen / 2;
2771
2772 for (i = 0; i < j - 1; i++)
2773 buf[i] = '-';
2774
2775 buf[i++] = ' ';
2776
2777 strcpy(buf + i, orig);
2778
2779 i += olen;
2780
2781 buf[i++] = ' ';
2782
2783 for (; i < len; i++)
2784 buf[i] = '-';
2785
2786 return buf;
2787}
2788
2789static int ui_quirks(void)
2790{
2791 const char *nodestr = "Data address";
2792 char *buf;
2793
2794 if (!c2c.use_stdio) {
2795 dim_offset.width = 5;
2796 dim_offset.header = header_offset_tui;
2797 nodestr = chk_double_cl ? "Double-CL" : "CL";
2798 }
2799
2800 dim_percent_costly_snoop.header = percent_costly_snoop_header[c2c.display];
2801
2802 /* Fix the zero line for dcacheline column. */
2803 buf = fill_line(chk_double_cl ? "Double-Cacheline" : "Cacheline",
2804 dim_dcacheline.width +
2805 dim_dcacheline_node.width +
2806 dim_dcacheline_count.width + 4);
2807 if (!buf)
2808 return -ENOMEM;
2809
2810 dim_dcacheline.header.line[0].text = buf;
2811
2812 /* Fix the zero line for offset column. */
2813 buf = fill_line(nodestr, dim_offset.width +
2814 dim_offset_node.width +
2815 dim_dcacheline_count.width + 4);
2816 if (!buf)
2817 return -ENOMEM;
2818
2819 dim_offset.header.line[0].text = buf;
2820
2821 return 0;
2822}
2823
2824#define CALLCHAIN_DEFAULT_OPT "graph,0.5,caller,function,percent"
2825
2826const char callchain_help[] = "Display call graph (stack chain/backtrace):\n\n"
2827 CALLCHAIN_REPORT_HELP
2828 "\n\t\t\t\tDefault: " CALLCHAIN_DEFAULT_OPT;
2829
2830static int
2831parse_callchain_opt(const struct option *opt, const char *arg, int unset)
2832{
2833 struct callchain_param *callchain = opt->value;
2834
2835 callchain->enabled = !unset;
2836 /*
2837 * --no-call-graph
2838 */
2839 if (unset) {
2840 symbol_conf.use_callchain = false;
2841 callchain->mode = CHAIN_NONE;
2842 return 0;
2843 }
2844
2845 return parse_callchain_report_opt(arg);
2846}
2847
2848static int setup_callchain(struct evlist *evlist)
2849{
2850 u64 sample_type = evlist__combined_sample_type(evlist);
2851 enum perf_call_graph_mode mode = CALLCHAIN_NONE;
2852
2853 if ((sample_type & PERF_SAMPLE_REGS_USER) &&
2854 (sample_type & PERF_SAMPLE_STACK_USER)) {
2855 mode = CALLCHAIN_DWARF;
2856 dwarf_callchain_users = true;
2857 } else if (sample_type & PERF_SAMPLE_BRANCH_STACK)
2858 mode = CALLCHAIN_LBR;
2859 else if (sample_type & PERF_SAMPLE_CALLCHAIN)
2860 mode = CALLCHAIN_FP;
2861
2862 if (!callchain_param.enabled &&
2863 callchain_param.mode != CHAIN_NONE &&
2864 mode != CALLCHAIN_NONE) {
2865 symbol_conf.use_callchain = true;
2866 if (callchain_register_param(&callchain_param) < 0) {
2867 ui__error("Can't register callchain params.\n");
2868 return -EINVAL;
2869 }
2870 }
2871
2872 if (c2c.stitch_lbr && (mode != CALLCHAIN_LBR)) {
2873 ui__warning("Can't find LBR callchain. Switch off --stitch-lbr.\n"
2874 "Please apply --call-graph lbr when recording.\n");
2875 c2c.stitch_lbr = false;
2876 }
2877
2878 callchain_param.record_mode = mode;
2879 callchain_param.min_percent = 0;
2880 return 0;
2881}
2882
2883static int setup_display(const char *str)
2884{
2885 const char *display = str;
2886
2887 if (!strcmp(display, "tot"))
2888 c2c.display = DISPLAY_TOT_HITM;
2889 else if (!strcmp(display, "rmt"))
2890 c2c.display = DISPLAY_RMT_HITM;
2891 else if (!strcmp(display, "lcl"))
2892 c2c.display = DISPLAY_LCL_HITM;
2893 else if (!strcmp(display, "peer"))
2894 c2c.display = DISPLAY_SNP_PEER;
2895 else {
2896 pr_err("failed: unknown display type: %s\n", str);
2897 return -1;
2898 }
2899
2900 return 0;
2901}
2902
2903#define for_each_token(__tok, __buf, __sep, __tmp) \
2904 for (__tok = strtok_r(__buf, __sep, &__tmp); __tok; \
2905 __tok = strtok_r(NULL, __sep, &__tmp))
2906
2907static int build_cl_output(char *cl_sort, bool no_source)
2908{
2909 char *tok, *tmp, *buf = strdup(cl_sort);
2910 bool add_pid = false;
2911 bool add_tid = false;
2912 bool add_iaddr = false;
2913 bool add_sym = false;
2914 bool add_dso = false;
2915 bool add_src = false;
2916 int ret = 0;
2917
2918 if (!buf)
2919 return -ENOMEM;
2920
2921 for_each_token(tok, buf, ",", tmp) {
2922 if (!strcmp(tok, "tid")) {
2923 add_tid = true;
2924 } else if (!strcmp(tok, "pid")) {
2925 add_pid = true;
2926 } else if (!strcmp(tok, "iaddr")) {
2927 add_iaddr = true;
2928 add_sym = true;
2929 add_dso = true;
2930 add_src = no_source ? false : true;
2931 } else if (!strcmp(tok, "dso")) {
2932 add_dso = true;
2933 } else if (strcmp(tok, "offset")) {
2934 pr_err("unrecognized sort token: %s\n", tok);
2935 ret = -EINVAL;
2936 goto err;
2937 }
2938 }
2939
2940 if (asprintf(&c2c.cl_output,
2941 "%s%s%s%s%s%s%s%s%s%s%s%s",
2942 c2c.use_stdio ? "cl_num_empty," : "",
2943 c2c.display == DISPLAY_SNP_PEER ? "percent_rmt_peer,"
2944 "percent_lcl_peer," :
2945 "percent_rmt_hitm,"
2946 "percent_lcl_hitm,",
2947 "percent_stores_l1hit,"
2948 "percent_stores_l1miss,"
2949 "percent_stores_na,"
2950 "offset,offset_node,dcacheline_count,",
2951 add_pid ? "pid," : "",
2952 add_tid ? "tid," : "",
2953 add_iaddr ? "iaddr," : "",
2954 c2c.display == DISPLAY_SNP_PEER ? "mean_rmt_peer,"
2955 "mean_lcl_peer," :
2956 "mean_rmt,"
2957 "mean_lcl,",
2958 "mean_load,"
2959 "tot_recs,"
2960 "cpucnt,",
2961 add_sym ? "symbol," : "",
2962 add_dso ? "dso," : "",
2963 add_src ? "cl_srcline," : "",
2964 "node") < 0) {
2965 ret = -ENOMEM;
2966 goto err;
2967 }
2968
2969 c2c.show_src = add_src;
2970err:
2971 free(buf);
2972 return ret;
2973}
2974
2975static int setup_coalesce(const char *coalesce, bool no_source)
2976{
2977 const char *c = coalesce ?: coalesce_default;
2978 const char *sort_str = NULL;
2979
2980 if (asprintf(&c2c.cl_sort, "offset,%s", c) < 0)
2981 return -ENOMEM;
2982
2983 if (build_cl_output(c2c.cl_sort, no_source))
2984 return -1;
2985
2986 if (c2c.display == DISPLAY_TOT_HITM)
2987 sort_str = "tot_hitm";
2988 else if (c2c.display == DISPLAY_RMT_HITM)
2989 sort_str = "rmt_hitm,lcl_hitm";
2990 else if (c2c.display == DISPLAY_LCL_HITM)
2991 sort_str = "lcl_hitm,rmt_hitm";
2992 else if (c2c.display == DISPLAY_SNP_PEER)
2993 sort_str = "tot_peer";
2994
2995 if (asprintf(&c2c.cl_resort, "offset,%s", sort_str) < 0)
2996 return -ENOMEM;
2997
2998 pr_debug("coalesce sort fields: %s\n", c2c.cl_sort);
2999 pr_debug("coalesce resort fields: %s\n", c2c.cl_resort);
3000 pr_debug("coalesce output fields: %s\n", c2c.cl_output);
3001 return 0;
3002}
3003
3004static int perf_c2c__report(int argc, const char **argv)
3005{
3006 struct itrace_synth_opts itrace_synth_opts = {
3007 .set = true,
3008 .mem = true, /* Only enable memory event */
3009 .default_no_sample = true,
3010 };
3011
3012 struct perf_session *session;
3013 struct ui_progress prog;
3014 struct perf_data data = {
3015 .mode = PERF_DATA_MODE_READ,
3016 };
3017 char callchain_default_opt[] = CALLCHAIN_DEFAULT_OPT;
3018 const char *display = NULL;
3019 const char *coalesce = NULL;
3020 bool no_source = false;
3021 const struct option options[] = {
3022 OPT_STRING('k', "vmlinux", &symbol_conf.vmlinux_name,
3023 "file", "vmlinux pathname"),
3024 OPT_STRING('i', "input", &input_name, "file",
3025 "the input file to process"),
3026 OPT_INCR('N', "node-info", &c2c.node_info,
3027 "show extra node info in report (repeat for more info)"),
3028 OPT_BOOLEAN(0, "stdio", &c2c.use_stdio, "Use the stdio interface"),
3029 OPT_BOOLEAN(0, "stats", &c2c.stats_only,
3030 "Display only statistic tables (implies --stdio)"),
3031 OPT_BOOLEAN(0, "full-symbols", &c2c.symbol_full,
3032 "Display full length of symbols"),
3033 OPT_BOOLEAN(0, "no-source", &no_source,
3034 "Do not display Source Line column"),
3035 OPT_BOOLEAN(0, "show-all", &c2c.show_all,
3036 "Show all captured HITM lines."),
3037 OPT_CALLBACK_DEFAULT('g', "call-graph", &callchain_param,
3038 "print_type,threshold[,print_limit],order,sort_key[,branch],value",
3039 callchain_help, &parse_callchain_opt,
3040 callchain_default_opt),
3041 OPT_STRING('d', "display", &display, "Switch HITM output type", "tot,lcl,rmt,peer"),
3042 OPT_STRING('c', "coalesce", &coalesce, "coalesce fields",
3043 "coalesce fields: pid,tid,iaddr,dso"),
3044 OPT_BOOLEAN('f', "force", &symbol_conf.force, "don't complain, do it"),
3045 OPT_BOOLEAN(0, "stitch-lbr", &c2c.stitch_lbr,
3046 "Enable LBR callgraph stitching approach"),
3047 OPT_BOOLEAN(0, "double-cl", &chk_double_cl, "Detect adjacent cacheline false sharing"),
3048 OPT_PARENT(c2c_options),
3049 OPT_END()
3050 };
3051 int err = 0;
3052 const char *output_str, *sort_str = NULL;
3053
3054 argc = parse_options(argc, argv, options, report_c2c_usage,
3055 PARSE_OPT_STOP_AT_NON_OPTION);
3056 if (argc)
3057 usage_with_options(report_c2c_usage, options);
3058
3059#ifndef HAVE_SLANG_SUPPORT
3060 c2c.use_stdio = true;
3061#endif
3062
3063 if (c2c.stats_only)
3064 c2c.use_stdio = true;
3065
3066 err = symbol__validate_sym_arguments();
3067 if (err)
3068 goto out;
3069
3070 if (!input_name || !strlen(input_name))
3071 input_name = "perf.data";
3072
3073 data.path = input_name;
3074 data.force = symbol_conf.force;
3075
3076 session = perf_session__new(&data, &c2c.tool);
3077 if (IS_ERR(session)) {
3078 err = PTR_ERR(session);
3079 pr_debug("Error creating perf session\n");
3080 goto out;
3081 }
3082
3083 /*
3084 * Use the 'tot' as default display type if user doesn't specify it;
3085 * since Arm64 platform doesn't support HITMs flag, use 'peer' as the
3086 * default display type.
3087 */
3088 if (!display) {
3089 if (!strcmp(perf_env__arch(&session->header.env), "arm64"))
3090 display = "peer";
3091 else
3092 display = "tot";
3093 }
3094
3095 err = setup_display(display);
3096 if (err)
3097 goto out_session;
3098
3099 err = setup_coalesce(coalesce, no_source);
3100 if (err) {
3101 pr_debug("Failed to initialize hists\n");
3102 goto out_session;
3103 }
3104
3105 err = c2c_hists__init(&c2c.hists, "dcacheline", 2);
3106 if (err) {
3107 pr_debug("Failed to initialize hists\n");
3108 goto out_session;
3109 }
3110
3111 session->itrace_synth_opts = &itrace_synth_opts;
3112
3113 err = setup_nodes(session);
3114 if (err) {
3115 pr_err("Failed setup nodes\n");
3116 goto out_session;
3117 }
3118
3119 err = mem2node__init(&c2c.mem2node, &session->header.env);
3120 if (err)
3121 goto out_session;
3122
3123 err = setup_callchain(session->evlist);
3124 if (err)
3125 goto out_mem2node;
3126
3127 if (symbol__init(&session->header.env) < 0)
3128 goto out_mem2node;
3129
3130 /* No pipe support at the moment. */
3131 if (perf_data__is_pipe(session->data)) {
3132 pr_debug("No pipe support at the moment.\n");
3133 goto out_mem2node;
3134 }
3135
3136 if (c2c.use_stdio)
3137 use_browser = 0;
3138 else
3139 use_browser = 1;
3140
3141 setup_browser(false);
3142
3143 err = perf_session__process_events(session);
3144 if (err) {
3145 pr_err("failed to process sample\n");
3146 goto out_mem2node;
3147 }
3148
3149 if (c2c.display != DISPLAY_SNP_PEER)
3150 output_str = "cl_idx,"
3151 "dcacheline,"
3152 "dcacheline_node,"
3153 "dcacheline_count,"
3154 "percent_costly_snoop,"
3155 "tot_hitm,lcl_hitm,rmt_hitm,"
3156 "tot_recs,"
3157 "tot_loads,"
3158 "tot_stores,"
3159 "stores_l1hit,stores_l1miss,stores_na,"
3160 "ld_fbhit,ld_l1hit,ld_l2hit,"
3161 "ld_lclhit,lcl_hitm,"
3162 "ld_rmthit,rmt_hitm,"
3163 "dram_lcl,dram_rmt";
3164 else
3165 output_str = "cl_idx,"
3166 "dcacheline,"
3167 "dcacheline_node,"
3168 "dcacheline_count,"
3169 "percent_costly_snoop,"
3170 "tot_peer,lcl_peer,rmt_peer,"
3171 "tot_recs,"
3172 "tot_loads,"
3173 "tot_stores,"
3174 "stores_l1hit,stores_l1miss,stores_na,"
3175 "ld_fbhit,ld_l1hit,ld_l2hit,"
3176 "ld_lclhit,lcl_hitm,"
3177 "ld_rmthit,rmt_hitm,"
3178 "dram_lcl,dram_rmt";
3179
3180 if (c2c.display == DISPLAY_TOT_HITM)
3181 sort_str = "tot_hitm";
3182 else if (c2c.display == DISPLAY_RMT_HITM)
3183 sort_str = "rmt_hitm";
3184 else if (c2c.display == DISPLAY_LCL_HITM)
3185 sort_str = "lcl_hitm";
3186 else if (c2c.display == DISPLAY_SNP_PEER)
3187 sort_str = "tot_peer";
3188
3189 c2c_hists__reinit(&c2c.hists, output_str, sort_str);
3190
3191 ui_progress__init(&prog, c2c.hists.hists.nr_entries, "Sorting...");
3192
3193 hists__collapse_resort(&c2c.hists.hists, NULL);
3194 hists__output_resort_cb(&c2c.hists.hists, &prog, resort_shared_cl_cb);
3195 hists__iterate_cb(&c2c.hists.hists, resort_cl_cb);
3196
3197 ui_progress__finish();
3198
3199 if (ui_quirks()) {
3200 pr_err("failed to setup UI\n");
3201 goto out_mem2node;
3202 }
3203
3204 perf_c2c_display(session);
3205
3206out_mem2node:
3207 mem2node__exit(&c2c.mem2node);
3208out_session:
3209 perf_session__delete(session);
3210out:
3211 return err;
3212}
3213
3214static int parse_record_events(const struct option *opt,
3215 const char *str, int unset __maybe_unused)
3216{
3217 bool *event_set = (bool *) opt->value;
3218 struct perf_pmu *pmu;
3219
3220 pmu = perf_mem_events_find_pmu();
3221 if (!pmu) {
3222 pr_err("failed: there is no PMU that supports perf c2c\n");
3223 exit(-1);
3224 }
3225
3226 if (!strcmp(str, "list")) {
3227 perf_pmu__mem_events_list(pmu);
3228 exit(0);
3229 }
3230 if (perf_pmu__mem_events_parse(pmu, str))
3231 exit(-1);
3232
3233 *event_set = true;
3234 return 0;
3235}
3236
3237
3238static const char * const __usage_record[] = {
3239 "perf c2c record [<options>] [<command>]",
3240 "perf c2c record [<options>] -- <command> [<options>]",
3241 NULL
3242};
3243
3244static const char * const *record_mem_usage = __usage_record;
3245
3246static int perf_c2c__record(int argc, const char **argv)
3247{
3248 int rec_argc, i = 0, j;
3249 const char **rec_argv;
3250 int ret;
3251 bool all_user = false, all_kernel = false;
3252 bool event_set = false;
3253 struct perf_mem_event *e;
3254 struct perf_pmu *pmu;
3255 struct option options[] = {
3256 OPT_CALLBACK('e', "event", &event_set, "event",
3257 "event selector. Use 'perf c2c record -e list' to list available events",
3258 parse_record_events),
3259 OPT_BOOLEAN('u', "all-user", &all_user, "collect only user level data"),
3260 OPT_BOOLEAN('k', "all-kernel", &all_kernel, "collect only kernel level data"),
3261 OPT_UINTEGER('l', "ldlat", &perf_mem_events__loads_ldlat, "setup mem-loads latency"),
3262 OPT_PARENT(c2c_options),
3263 OPT_END()
3264 };
3265
3266 pmu = perf_mem_events_find_pmu();
3267 if (!pmu) {
3268 pr_err("failed: no PMU supports the memory events\n");
3269 return -1;
3270 }
3271
3272 if (perf_pmu__mem_events_init(pmu)) {
3273 pr_err("failed: memory events not supported\n");
3274 return -1;
3275 }
3276
3277 argc = parse_options(argc, argv, options, record_mem_usage,
3278 PARSE_OPT_KEEP_UNKNOWN);
3279
3280 /* Max number of arguments multiplied by number of PMUs that can support them. */
3281 rec_argc = argc + 11 * (perf_pmu__mem_events_num_mem_pmus(pmu) + 1);
3282
3283 rec_argv = calloc(rec_argc + 1, sizeof(char *));
3284 if (!rec_argv)
3285 return -1;
3286
3287 rec_argv[i++] = "record";
3288
3289 if (!event_set) {
3290 e = perf_pmu__mem_events_ptr(pmu, PERF_MEM_EVENTS__LOAD_STORE);
3291 /*
3292 * The load and store operations are required, use the event
3293 * PERF_MEM_EVENTS__LOAD_STORE if it is supported.
3294 */
3295 if (e->tag) {
3296 e->record = true;
3297 rec_argv[i++] = "-W";
3298 } else {
3299 e = perf_pmu__mem_events_ptr(pmu, PERF_MEM_EVENTS__LOAD);
3300 e->record = true;
3301
3302 e = perf_pmu__mem_events_ptr(pmu, PERF_MEM_EVENTS__STORE);
3303 e->record = true;
3304 }
3305 }
3306
3307 e = perf_pmu__mem_events_ptr(pmu, PERF_MEM_EVENTS__LOAD);
3308 if (e->record)
3309 rec_argv[i++] = "-W";
3310
3311 rec_argv[i++] = "-d";
3312 rec_argv[i++] = "--phys-data";
3313 rec_argv[i++] = "--sample-cpu";
3314
3315 ret = perf_mem_events__record_args(rec_argv, &i);
3316 if (ret)
3317 goto out;
3318
3319 if (all_user)
3320 rec_argv[i++] = "--all-user";
3321
3322 if (all_kernel)
3323 rec_argv[i++] = "--all-kernel";
3324
3325 for (j = 0; j < argc; j++, i++)
3326 rec_argv[i] = argv[j];
3327
3328 if (verbose > 0) {
3329 pr_debug("calling: ");
3330
3331 j = 0;
3332
3333 while (rec_argv[j]) {
3334 pr_debug("%s ", rec_argv[j]);
3335 j++;
3336 }
3337 pr_debug("\n");
3338 }
3339
3340 ret = cmd_record(i, rec_argv);
3341out:
3342 free(rec_argv);
3343 return ret;
3344}
3345
3346int cmd_c2c(int argc, const char **argv)
3347{
3348 argc = parse_options(argc, argv, c2c_options, c2c_usage,
3349 PARSE_OPT_STOP_AT_NON_OPTION);
3350
3351 if (!argc)
3352 usage_with_options(c2c_usage, c2c_options);
3353
3354 if (strlen(argv[0]) > 2 && strstarts("record", argv[0])) {
3355 return perf_c2c__record(argc, argv);
3356 } else if (strlen(argv[0]) > 2 && strstarts("report", argv[0])) {
3357 return perf_c2c__report(argc, argv);
3358 } else {
3359 usage_with_options(c2c_usage, c2c_options);
3360 }
3361
3362 return 0;
3363}