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