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