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