Loading...
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * This is rewrite of original c2c tool introduced in here:
4 * http://lwn.net/Articles/588866/
5 *
6 * The original tool was changed to fit in current perf state.
7 *
8 * Original authors:
9 * Don Zickus <dzickus@redhat.com>
10 * Dick Fowles <fowles@inreach.com>
11 * Joe Mario <jmario@redhat.com>
12 */
13#include <errno.h>
14#include <inttypes.h>
15#include <linux/compiler.h>
16#include <linux/err.h>
17#include <linux/kernel.h>
18#include <linux/stringify.h>
19#include <linux/zalloc.h>
20#include <asm/bug.h>
21#include <sys/param.h>
22#include "debug.h"
23#include "builtin.h"
24#include <perf/cpumap.h>
25#include <subcmd/pager.h>
26#include <subcmd/parse-options.h>
27#include "map_symbol.h"
28#include "mem-events.h"
29#include "session.h"
30#include "hist.h"
31#include "sort.h"
32#include "tool.h"
33#include "cacheline.h"
34#include "data.h"
35#include "event.h"
36#include "evlist.h"
37#include "evsel.h"
38#include "ui/browsers/hists.h"
39#include "thread.h"
40#include "mem2node.h"
41#include "symbol.h"
42#include "ui/ui.h"
43#include "ui/progress.h"
44#include "../perf.h"
45
46struct c2c_hists {
47 struct hists hists;
48 struct perf_hpp_list list;
49 struct c2c_stats stats;
50};
51
52struct compute_stats {
53 struct stats lcl_hitm;
54 struct stats rmt_hitm;
55 struct stats load;
56};
57
58struct c2c_hist_entry {
59 struct c2c_hists *hists;
60 struct c2c_stats stats;
61 unsigned long *cpuset;
62 unsigned long *nodeset;
63 struct c2c_stats *node_stats;
64 unsigned int cacheline_idx;
65
66 struct compute_stats cstats;
67
68 unsigned long paddr;
69 unsigned long paddr_cnt;
70 bool paddr_zero;
71 char *nodestr;
72
73 /*
74 * must be at the end,
75 * because of its callchain dynamic entry
76 */
77 struct hist_entry he;
78};
79
80static char const *coalesce_default = "iaddr";
81
82struct perf_c2c {
83 struct perf_tool tool;
84 struct c2c_hists hists;
85 struct mem2node mem2node;
86
87 unsigned long **nodes;
88 int nodes_cnt;
89 int cpus_cnt;
90 int *cpu2node;
91 int node_info;
92
93 bool show_src;
94 bool show_all;
95 bool use_stdio;
96 bool stats_only;
97 bool symbol_full;
98
99 /* HITM shared clines stats */
100 struct c2c_stats hitm_stats;
101 int shared_clines;
102
103 int display;
104
105 const char *coalesce;
106 char *cl_sort;
107 char *cl_resort;
108 char *cl_output;
109};
110
111enum {
112 DISPLAY_LCL,
113 DISPLAY_RMT,
114 DISPLAY_TOT,
115 DISPLAY_MAX,
116};
117
118static const char *display_str[DISPLAY_MAX] = {
119 [DISPLAY_LCL] = "Local",
120 [DISPLAY_RMT] = "Remote",
121 [DISPLAY_TOT] = "Total",
122};
123
124static const struct option c2c_options[] = {
125 OPT_INCR('v', "verbose", &verbose, "be more verbose (show counter open errors, etc)"),
126 OPT_END()
127};
128
129static struct perf_c2c c2c;
130
131static void *c2c_he_zalloc(size_t size)
132{
133 struct c2c_hist_entry *c2c_he;
134
135 c2c_he = zalloc(size + sizeof(*c2c_he));
136 if (!c2c_he)
137 return NULL;
138
139 c2c_he->cpuset = bitmap_alloc(c2c.cpus_cnt);
140 if (!c2c_he->cpuset)
141 return NULL;
142
143 c2c_he->nodeset = bitmap_alloc(c2c.nodes_cnt);
144 if (!c2c_he->nodeset)
145 return NULL;
146
147 c2c_he->node_stats = zalloc(c2c.nodes_cnt * sizeof(*c2c_he->node_stats));
148 if (!c2c_he->node_stats)
149 return NULL;
150
151 init_stats(&c2c_he->cstats.lcl_hitm);
152 init_stats(&c2c_he->cstats.rmt_hitm);
153 init_stats(&c2c_he->cstats.load);
154
155 return &c2c_he->he;
156}
157
158static void c2c_he_free(void *he)
159{
160 struct c2c_hist_entry *c2c_he;
161
162 c2c_he = container_of(he, struct c2c_hist_entry, he);
163 if (c2c_he->hists) {
164 hists__delete_entries(&c2c_he->hists->hists);
165 free(c2c_he->hists);
166 }
167
168 free(c2c_he->cpuset);
169 free(c2c_he->nodeset);
170 free(c2c_he->nodestr);
171 free(c2c_he->node_stats);
172 free(c2c_he);
173}
174
175static struct hist_entry_ops c2c_entry_ops = {
176 .new = c2c_he_zalloc,
177 .free = c2c_he_free,
178};
179
180static int c2c_hists__init(struct c2c_hists *hists,
181 const char *sort,
182 int nr_header_lines);
183
184static struct c2c_hists*
185he__get_c2c_hists(struct hist_entry *he,
186 const char *sort,
187 int nr_header_lines)
188{
189 struct c2c_hist_entry *c2c_he;
190 struct c2c_hists *hists;
191 int ret;
192
193 c2c_he = container_of(he, struct c2c_hist_entry, he);
194 if (c2c_he->hists)
195 return c2c_he->hists;
196
197 hists = c2c_he->hists = zalloc(sizeof(*hists));
198 if (!hists)
199 return NULL;
200
201 ret = c2c_hists__init(hists, sort, nr_header_lines);
202 if (ret) {
203 free(hists);
204 return NULL;
205 }
206
207 return hists;
208}
209
210static void c2c_he__set_cpu(struct c2c_hist_entry *c2c_he,
211 struct perf_sample *sample)
212{
213 if (WARN_ONCE(sample->cpu == (unsigned int) -1,
214 "WARNING: no sample cpu value"))
215 return;
216
217 set_bit(sample->cpu, c2c_he->cpuset);
218}
219
220static void c2c_he__set_node(struct c2c_hist_entry *c2c_he,
221 struct perf_sample *sample)
222{
223 int node;
224
225 if (!sample->phys_addr) {
226 c2c_he->paddr_zero = true;
227 return;
228 }
229
230 node = mem2node__node(&c2c.mem2node, sample->phys_addr);
231 if (WARN_ONCE(node < 0, "WARNING: failed to find node\n"))
232 return;
233
234 set_bit(node, c2c_he->nodeset);
235
236 if (c2c_he->paddr != sample->phys_addr) {
237 c2c_he->paddr_cnt++;
238 c2c_he->paddr = sample->phys_addr;
239 }
240}
241
242static void compute_stats(struct c2c_hist_entry *c2c_he,
243 struct c2c_stats *stats,
244 u64 weight)
245{
246 struct compute_stats *cstats = &c2c_he->cstats;
247
248 if (stats->rmt_hitm)
249 update_stats(&cstats->rmt_hitm, weight);
250 else if (stats->lcl_hitm)
251 update_stats(&cstats->lcl_hitm, weight);
252 else if (stats->load)
253 update_stats(&cstats->load, weight);
254}
255
256static int process_sample_event(struct perf_tool *tool __maybe_unused,
257 union perf_event *event,
258 struct perf_sample *sample,
259 struct evsel *evsel,
260 struct machine *machine)
261{
262 struct c2c_hists *c2c_hists = &c2c.hists;
263 struct c2c_hist_entry *c2c_he;
264 struct c2c_stats stats = { .nr_entries = 0, };
265 struct hist_entry *he;
266 struct addr_location al;
267 struct mem_info *mi, *mi_dup;
268 int ret;
269
270 if (machine__resolve(machine, &al, sample) < 0) {
271 pr_debug("problem processing %d event, skipping it.\n",
272 event->header.type);
273 return -1;
274 }
275
276 ret = sample__resolve_callchain(sample, &callchain_cursor, NULL,
277 evsel, &al, sysctl_perf_event_max_stack);
278 if (ret)
279 goto out;
280
281 mi = sample__resolve_mem(sample, &al);
282 if (mi == NULL)
283 return -ENOMEM;
284
285 /*
286 * The mi object is released in hists__add_entry_ops,
287 * if it gets sorted out into existing data, so we need
288 * to take the copy now.
289 */
290 mi_dup = mem_info__get(mi);
291
292 c2c_decode_stats(&stats, mi);
293
294 he = hists__add_entry_ops(&c2c_hists->hists, &c2c_entry_ops,
295 &al, NULL, NULL, mi,
296 sample, true);
297 if (he == NULL)
298 goto free_mi;
299
300 c2c_he = container_of(he, struct c2c_hist_entry, he);
301 c2c_add_stats(&c2c_he->stats, &stats);
302 c2c_add_stats(&c2c_hists->stats, &stats);
303
304 c2c_he__set_cpu(c2c_he, sample);
305 c2c_he__set_node(c2c_he, sample);
306
307 hists__inc_nr_samples(&c2c_hists->hists, he->filtered);
308 ret = hist_entry__append_callchain(he, sample);
309
310 if (!ret) {
311 /*
312 * There's already been warning about missing
313 * sample's cpu value. Let's account all to
314 * node 0 in this case, without any further
315 * warning.
316 *
317 * Doing node stats only for single callchain data.
318 */
319 int cpu = sample->cpu == (unsigned int) -1 ? 0 : sample->cpu;
320 int node = c2c.cpu2node[cpu];
321
322 mi = mi_dup;
323
324 c2c_hists = he__get_c2c_hists(he, c2c.cl_sort, 2);
325 if (!c2c_hists)
326 goto free_mi;
327
328 he = hists__add_entry_ops(&c2c_hists->hists, &c2c_entry_ops,
329 &al, NULL, NULL, mi,
330 sample, true);
331 if (he == NULL)
332 goto free_mi;
333
334 c2c_he = container_of(he, struct c2c_hist_entry, he);
335 c2c_add_stats(&c2c_he->stats, &stats);
336 c2c_add_stats(&c2c_hists->stats, &stats);
337 c2c_add_stats(&c2c_he->node_stats[node], &stats);
338
339 compute_stats(c2c_he, &stats, sample->weight);
340
341 c2c_he__set_cpu(c2c_he, sample);
342 c2c_he__set_node(c2c_he, sample);
343
344 hists__inc_nr_samples(&c2c_hists->hists, he->filtered);
345 ret = hist_entry__append_callchain(he, sample);
346 }
347
348out:
349 addr_location__put(&al);
350 return ret;
351
352free_mi:
353 mem_info__put(mi_dup);
354 mem_info__put(mi);
355 ret = -ENOMEM;
356 goto out;
357}
358
359static struct perf_c2c c2c = {
360 .tool = {
361 .sample = process_sample_event,
362 .mmap = perf_event__process_mmap,
363 .mmap2 = perf_event__process_mmap2,
364 .comm = perf_event__process_comm,
365 .exit = perf_event__process_exit,
366 .fork = perf_event__process_fork,
367 .lost = perf_event__process_lost,
368 .ordered_events = true,
369 .ordering_requires_timestamps = true,
370 },
371};
372
373static const char * const c2c_usage[] = {
374 "perf c2c {record|report}",
375 NULL
376};
377
378static const char * const __usage_report[] = {
379 "perf c2c report",
380 NULL
381};
382
383static const char * const *report_c2c_usage = __usage_report;
384
385#define C2C_HEADER_MAX 2
386
387struct c2c_header {
388 struct {
389 const char *text;
390 int span;
391 } line[C2C_HEADER_MAX];
392};
393
394struct c2c_dimension {
395 struct c2c_header header;
396 const char *name;
397 int width;
398 struct sort_entry *se;
399
400 int64_t (*cmp)(struct perf_hpp_fmt *fmt,
401 struct hist_entry *, struct hist_entry *);
402 int (*entry)(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
403 struct hist_entry *he);
404 int (*color)(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
405 struct hist_entry *he);
406};
407
408struct c2c_fmt {
409 struct perf_hpp_fmt fmt;
410 struct c2c_dimension *dim;
411};
412
413#define SYMBOL_WIDTH 30
414
415static struct c2c_dimension dim_symbol;
416static struct c2c_dimension dim_srcline;
417
418static int symbol_width(struct hists *hists, struct sort_entry *se)
419{
420 int width = hists__col_len(hists, se->se_width_idx);
421
422 if (!c2c.symbol_full)
423 width = MIN(width, SYMBOL_WIDTH);
424
425 return width;
426}
427
428static int c2c_width(struct perf_hpp_fmt *fmt,
429 struct perf_hpp *hpp __maybe_unused,
430 struct hists *hists)
431{
432 struct c2c_fmt *c2c_fmt;
433 struct c2c_dimension *dim;
434
435 c2c_fmt = container_of(fmt, struct c2c_fmt, fmt);
436 dim = c2c_fmt->dim;
437
438 if (dim == &dim_symbol || dim == &dim_srcline)
439 return symbol_width(hists, dim->se);
440
441 return dim->se ? hists__col_len(hists, dim->se->se_width_idx) :
442 c2c_fmt->dim->width;
443}
444
445static int c2c_header(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
446 struct hists *hists, int line, int *span)
447{
448 struct perf_hpp_list *hpp_list = hists->hpp_list;
449 struct c2c_fmt *c2c_fmt;
450 struct c2c_dimension *dim;
451 const char *text = NULL;
452 int width = c2c_width(fmt, hpp, hists);
453
454 c2c_fmt = container_of(fmt, struct c2c_fmt, fmt);
455 dim = c2c_fmt->dim;
456
457 if (dim->se) {
458 text = dim->header.line[line].text;
459 /* Use the last line from sort_entry if not defined. */
460 if (!text && (line == hpp_list->nr_header_lines - 1))
461 text = dim->se->se_header;
462 } else {
463 text = dim->header.line[line].text;
464
465 if (*span) {
466 (*span)--;
467 return 0;
468 } else {
469 *span = dim->header.line[line].span;
470 }
471 }
472
473 if (text == NULL)
474 text = "";
475
476 return scnprintf(hpp->buf, hpp->size, "%*s", width, text);
477}
478
479#define HEX_STR(__s, __v) \
480({ \
481 scnprintf(__s, sizeof(__s), "0x%" PRIx64, __v); \
482 __s; \
483})
484
485static int64_t
486dcacheline_cmp(struct perf_hpp_fmt *fmt __maybe_unused,
487 struct hist_entry *left, struct hist_entry *right)
488{
489 return sort__dcacheline_cmp(left, right);
490}
491
492static int dcacheline_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
493 struct hist_entry *he)
494{
495 uint64_t addr = 0;
496 int width = c2c_width(fmt, hpp, he->hists);
497 char buf[20];
498
499 if (he->mem_info)
500 addr = cl_address(he->mem_info->daddr.addr);
501
502 return scnprintf(hpp->buf, hpp->size, "%*s", width, HEX_STR(buf, addr));
503}
504
505static int
506dcacheline_node_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
507 struct hist_entry *he)
508{
509 struct c2c_hist_entry *c2c_he;
510 int width = c2c_width(fmt, hpp, he->hists);
511
512 c2c_he = container_of(he, struct c2c_hist_entry, he);
513 if (WARN_ON_ONCE(!c2c_he->nodestr))
514 return 0;
515
516 return scnprintf(hpp->buf, hpp->size, "%*s", width, c2c_he->nodestr);
517}
518
519static int
520dcacheline_node_count(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
521 struct hist_entry *he)
522{
523 struct c2c_hist_entry *c2c_he;
524 int width = c2c_width(fmt, hpp, he->hists);
525
526 c2c_he = container_of(he, struct c2c_hist_entry, he);
527 return scnprintf(hpp->buf, hpp->size, "%*lu", width, c2c_he->paddr_cnt);
528}
529
530static int offset_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
531 struct hist_entry *he)
532{
533 uint64_t addr = 0;
534 int width = c2c_width(fmt, hpp, he->hists);
535 char buf[20];
536
537 if (he->mem_info)
538 addr = cl_offset(he->mem_info->daddr.al_addr);
539
540 return scnprintf(hpp->buf, hpp->size, "%*s", width, HEX_STR(buf, addr));
541}
542
543static int64_t
544offset_cmp(struct perf_hpp_fmt *fmt __maybe_unused,
545 struct hist_entry *left, struct hist_entry *right)
546{
547 uint64_t l = 0, r = 0;
548
549 if (left->mem_info)
550 l = cl_offset(left->mem_info->daddr.addr);
551 if (right->mem_info)
552 r = cl_offset(right->mem_info->daddr.addr);
553
554 return (int64_t)(r - l);
555}
556
557static int
558iaddr_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
559 struct hist_entry *he)
560{
561 uint64_t addr = 0;
562 int width = c2c_width(fmt, hpp, he->hists);
563 char buf[20];
564
565 if (he->mem_info)
566 addr = he->mem_info->iaddr.addr;
567
568 return scnprintf(hpp->buf, hpp->size, "%*s", width, HEX_STR(buf, addr));
569}
570
571static int64_t
572iaddr_cmp(struct perf_hpp_fmt *fmt __maybe_unused,
573 struct hist_entry *left, struct hist_entry *right)
574{
575 return sort__iaddr_cmp(left, right);
576}
577
578static int
579tot_hitm_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
580 struct hist_entry *he)
581{
582 struct c2c_hist_entry *c2c_he;
583 int width = c2c_width(fmt, hpp, he->hists);
584 unsigned int tot_hitm;
585
586 c2c_he = container_of(he, struct c2c_hist_entry, he);
587 tot_hitm = c2c_he->stats.lcl_hitm + c2c_he->stats.rmt_hitm;
588
589 return scnprintf(hpp->buf, hpp->size, "%*u", width, tot_hitm);
590}
591
592static int64_t
593tot_hitm_cmp(struct perf_hpp_fmt *fmt __maybe_unused,
594 struct hist_entry *left, struct hist_entry *right)
595{
596 struct c2c_hist_entry *c2c_left;
597 struct c2c_hist_entry *c2c_right;
598 unsigned int tot_hitm_left;
599 unsigned int tot_hitm_right;
600
601 c2c_left = container_of(left, struct c2c_hist_entry, he);
602 c2c_right = container_of(right, struct c2c_hist_entry, he);
603
604 tot_hitm_left = c2c_left->stats.lcl_hitm + c2c_left->stats.rmt_hitm;
605 tot_hitm_right = c2c_right->stats.lcl_hitm + c2c_right->stats.rmt_hitm;
606
607 return tot_hitm_left - tot_hitm_right;
608}
609
610#define STAT_FN_ENTRY(__f) \
611static int \
612__f ## _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 \
618 c2c_he = container_of(he, struct c2c_hist_entry, he); \
619 return scnprintf(hpp->buf, hpp->size, "%*u", width, \
620 c2c_he->stats.__f); \
621}
622
623#define STAT_FN_CMP(__f) \
624static int64_t \
625__f ## _cmp(struct perf_hpp_fmt *fmt __maybe_unused, \
626 struct hist_entry *left, struct hist_entry *right) \
627{ \
628 struct c2c_hist_entry *c2c_left, *c2c_right; \
629 \
630 c2c_left = container_of(left, struct c2c_hist_entry, he); \
631 c2c_right = container_of(right, struct c2c_hist_entry, he); \
632 return c2c_left->stats.__f - c2c_right->stats.__f; \
633}
634
635#define STAT_FN(__f) \
636 STAT_FN_ENTRY(__f) \
637 STAT_FN_CMP(__f)
638
639STAT_FN(rmt_hitm)
640STAT_FN(lcl_hitm)
641STAT_FN(store)
642STAT_FN(st_l1hit)
643STAT_FN(st_l1miss)
644STAT_FN(ld_fbhit)
645STAT_FN(ld_l1hit)
646STAT_FN(ld_l2hit)
647STAT_FN(ld_llchit)
648STAT_FN(rmt_hit)
649
650static uint64_t llc_miss(struct c2c_stats *stats)
651{
652 uint64_t llcmiss;
653
654 llcmiss = stats->lcl_dram +
655 stats->rmt_dram +
656 stats->rmt_hitm +
657 stats->rmt_hit;
658
659 return llcmiss;
660}
661
662static int
663ld_llcmiss_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
664 struct hist_entry *he)
665{
666 struct c2c_hist_entry *c2c_he;
667 int width = c2c_width(fmt, hpp, he->hists);
668
669 c2c_he = container_of(he, struct c2c_hist_entry, he);
670
671 return scnprintf(hpp->buf, hpp->size, "%*lu", width,
672 llc_miss(&c2c_he->stats));
673}
674
675static int64_t
676ld_llcmiss_cmp(struct perf_hpp_fmt *fmt __maybe_unused,
677 struct hist_entry *left, struct hist_entry *right)
678{
679 struct c2c_hist_entry *c2c_left;
680 struct c2c_hist_entry *c2c_right;
681
682 c2c_left = container_of(left, struct c2c_hist_entry, he);
683 c2c_right = container_of(right, struct c2c_hist_entry, he);
684
685 return llc_miss(&c2c_left->stats) - llc_miss(&c2c_right->stats);
686}
687
688static uint64_t total_records(struct c2c_stats *stats)
689{
690 uint64_t lclmiss, ldcnt, total;
691
692 lclmiss = stats->lcl_dram +
693 stats->rmt_dram +
694 stats->rmt_hitm +
695 stats->rmt_hit;
696
697 ldcnt = lclmiss +
698 stats->ld_fbhit +
699 stats->ld_l1hit +
700 stats->ld_l2hit +
701 stats->ld_llchit +
702 stats->lcl_hitm;
703
704 total = ldcnt +
705 stats->st_l1hit +
706 stats->st_l1miss;
707
708 return total;
709}
710
711static int
712tot_recs_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
713 struct hist_entry *he)
714{
715 struct c2c_hist_entry *c2c_he;
716 int width = c2c_width(fmt, hpp, he->hists);
717 uint64_t tot_recs;
718
719 c2c_he = container_of(he, struct c2c_hist_entry, he);
720 tot_recs = total_records(&c2c_he->stats);
721
722 return scnprintf(hpp->buf, hpp->size, "%*" PRIu64, width, tot_recs);
723}
724
725static int64_t
726tot_recs_cmp(struct perf_hpp_fmt *fmt __maybe_unused,
727 struct hist_entry *left, struct hist_entry *right)
728{
729 struct c2c_hist_entry *c2c_left;
730 struct c2c_hist_entry *c2c_right;
731 uint64_t tot_recs_left;
732 uint64_t tot_recs_right;
733
734 c2c_left = container_of(left, struct c2c_hist_entry, he);
735 c2c_right = container_of(right, struct c2c_hist_entry, he);
736
737 tot_recs_left = total_records(&c2c_left->stats);
738 tot_recs_right = total_records(&c2c_right->stats);
739
740 return tot_recs_left - tot_recs_right;
741}
742
743static uint64_t total_loads(struct c2c_stats *stats)
744{
745 uint64_t lclmiss, ldcnt;
746
747 lclmiss = stats->lcl_dram +
748 stats->rmt_dram +
749 stats->rmt_hitm +
750 stats->rmt_hit;
751
752 ldcnt = lclmiss +
753 stats->ld_fbhit +
754 stats->ld_l1hit +
755 stats->ld_l2hit +
756 stats->ld_llchit +
757 stats->lcl_hitm;
758
759 return ldcnt;
760}
761
762static int
763tot_loads_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
764 struct hist_entry *he)
765{
766 struct c2c_hist_entry *c2c_he;
767 int width = c2c_width(fmt, hpp, he->hists);
768 uint64_t tot_recs;
769
770 c2c_he = container_of(he, struct c2c_hist_entry, he);
771 tot_recs = total_loads(&c2c_he->stats);
772
773 return scnprintf(hpp->buf, hpp->size, "%*" PRIu64, width, tot_recs);
774}
775
776static int64_t
777tot_loads_cmp(struct perf_hpp_fmt *fmt __maybe_unused,
778 struct hist_entry *left, struct hist_entry *right)
779{
780 struct c2c_hist_entry *c2c_left;
781 struct c2c_hist_entry *c2c_right;
782 uint64_t tot_recs_left;
783 uint64_t tot_recs_right;
784
785 c2c_left = container_of(left, struct c2c_hist_entry, he);
786 c2c_right = container_of(right, struct c2c_hist_entry, he);
787
788 tot_recs_left = total_loads(&c2c_left->stats);
789 tot_recs_right = total_loads(&c2c_right->stats);
790
791 return tot_recs_left - tot_recs_right;
792}
793
794typedef double (get_percent_cb)(struct c2c_hist_entry *);
795
796static int
797percent_color(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
798 struct hist_entry *he, get_percent_cb get_percent)
799{
800 struct c2c_hist_entry *c2c_he;
801 int width = c2c_width(fmt, hpp, he->hists);
802 double per;
803
804 c2c_he = container_of(he, struct c2c_hist_entry, he);
805 per = get_percent(c2c_he);
806
807#ifdef HAVE_SLANG_SUPPORT
808 if (use_browser)
809 return __hpp__slsmg_color_printf(hpp, "%*.2f%%", width - 1, per);
810#endif
811 return hpp_color_scnprintf(hpp, "%*.2f%%", width - 1, per);
812}
813
814static double percent_hitm(struct c2c_hist_entry *c2c_he)
815{
816 struct c2c_hists *hists;
817 struct c2c_stats *stats;
818 struct c2c_stats *total;
819 int tot = 0, st = 0;
820 double p;
821
822 hists = container_of(c2c_he->he.hists, struct c2c_hists, hists);
823 stats = &c2c_he->stats;
824 total = &hists->stats;
825
826 switch (c2c.display) {
827 case DISPLAY_RMT:
828 st = stats->rmt_hitm;
829 tot = total->rmt_hitm;
830 break;
831 case DISPLAY_LCL:
832 st = stats->lcl_hitm;
833 tot = total->lcl_hitm;
834 break;
835 case DISPLAY_TOT:
836 st = stats->tot_hitm;
837 tot = total->tot_hitm;
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_hitm_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_hitm(c2c_he);
864 return scnprintf(hpp->buf, hpp->size, "%*s", width, PERC_STR(buf, per));
865}
866
867static int
868percent_hitm_color(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
869 struct hist_entry *he)
870{
871 return percent_color(fmt, hpp, he, percent_hitm);
872}
873
874static int64_t
875percent_hitm_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_hitm(c2c_left);
887 per_right = percent_hitm(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(int st, int 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(st_l1hit)
927PERCENT_FN(st_l1miss)
928
929static int
930percent_rmt_hitm_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
931 struct hist_entry *he)
932{
933 int width = c2c_width(fmt, hpp, he->hists);
934 double per = PERCENT(he, rmt_hitm);
935 char buf[10];
936
937 return scnprintf(hpp->buf, hpp->size, "%*s", width, PERC_STR(buf, per));
938}
939
940static int
941percent_rmt_hitm_color(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
942 struct hist_entry *he)
943{
944 return percent_color(fmt, hpp, he, percent_rmt_hitm);
945}
946
947static int64_t
948percent_rmt_hitm_cmp(struct perf_hpp_fmt *fmt __maybe_unused,
949 struct hist_entry *left, struct hist_entry *right)
950{
951 double per_left;
952 double per_right;
953
954 per_left = PERCENT(left, lcl_hitm);
955 per_right = PERCENT(right, lcl_hitm);
956
957 return per_left - per_right;
958}
959
960static int
961percent_lcl_hitm_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
962 struct hist_entry *he)
963{
964 int width = c2c_width(fmt, hpp, he->hists);
965 double per = PERCENT(he, lcl_hitm);
966 char buf[10];
967
968 return scnprintf(hpp->buf, hpp->size, "%*s", width, PERC_STR(buf, per));
969}
970
971static int
972percent_lcl_hitm_color(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
973 struct hist_entry *he)
974{
975 return percent_color(fmt, hpp, he, percent_lcl_hitm);
976}
977
978static int64_t
979percent_lcl_hitm_cmp(struct perf_hpp_fmt *fmt __maybe_unused,
980 struct hist_entry *left, struct hist_entry *right)
981{
982 double per_left;
983 double per_right;
984
985 per_left = PERCENT(left, lcl_hitm);
986 per_right = PERCENT(right, lcl_hitm);
987
988 return per_left - per_right;
989}
990
991static int
992percent_stores_l1hit_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
993 struct hist_entry *he)
994{
995 int width = c2c_width(fmt, hpp, he->hists);
996 double per = PERCENT(he, st_l1hit);
997 char buf[10];
998
999 return scnprintf(hpp->buf, hpp->size, "%*s", width, PERC_STR(buf, per));
1000}
1001
1002static int
1003percent_stores_l1hit_color(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
1004 struct hist_entry *he)
1005{
1006 return percent_color(fmt, hpp, he, percent_st_l1hit);
1007}
1008
1009static int64_t
1010percent_stores_l1hit_cmp(struct perf_hpp_fmt *fmt __maybe_unused,
1011 struct hist_entry *left, struct hist_entry *right)
1012{
1013 double per_left;
1014 double per_right;
1015
1016 per_left = PERCENT(left, st_l1hit);
1017 per_right = PERCENT(right, st_l1hit);
1018
1019 return per_left - per_right;
1020}
1021
1022static int
1023percent_stores_l1miss_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
1024 struct hist_entry *he)
1025{
1026 int width = c2c_width(fmt, hpp, he->hists);
1027 double per = PERCENT(he, st_l1miss);
1028 char buf[10];
1029
1030 return scnprintf(hpp->buf, hpp->size, "%*s", width, PERC_STR(buf, per));
1031}
1032
1033static int
1034percent_stores_l1miss_color(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
1035 struct hist_entry *he)
1036{
1037 return percent_color(fmt, hpp, he, percent_st_l1miss);
1038}
1039
1040static int64_t
1041percent_stores_l1miss_cmp(struct perf_hpp_fmt *fmt __maybe_unused,
1042 struct hist_entry *left, struct hist_entry *right)
1043{
1044 double per_left;
1045 double per_right;
1046
1047 per_left = PERCENT(left, st_l1miss);
1048 per_right = PERCENT(right, st_l1miss);
1049
1050 return per_left - per_right;
1051}
1052
1053STAT_FN(lcl_dram)
1054STAT_FN(rmt_dram)
1055
1056static int
1057pid_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
1062 return scnprintf(hpp->buf, hpp->size, "%*d", width, he->thread->pid_);
1063}
1064
1065static int64_t
1066pid_cmp(struct perf_hpp_fmt *fmt __maybe_unused,
1067 struct hist_entry *left, struct hist_entry *right)
1068{
1069 return left->thread->pid_ - right->thread->pid_;
1070}
1071
1072static int64_t
1073empty_cmp(struct perf_hpp_fmt *fmt __maybe_unused,
1074 struct hist_entry *left __maybe_unused,
1075 struct hist_entry *right __maybe_unused)
1076{
1077 return 0;
1078}
1079
1080static int
1081node_entry(struct perf_hpp_fmt *fmt __maybe_unused, struct perf_hpp *hpp,
1082 struct hist_entry *he)
1083{
1084 struct c2c_hist_entry *c2c_he;
1085 bool first = true;
1086 int node;
1087 int ret = 0;
1088
1089 c2c_he = container_of(he, struct c2c_hist_entry, he);
1090
1091 for (node = 0; node < c2c.nodes_cnt; node++) {
1092 DECLARE_BITMAP(set, c2c.cpus_cnt);
1093
1094 bitmap_zero(set, c2c.cpus_cnt);
1095 bitmap_and(set, c2c_he->cpuset, c2c.nodes[node], c2c.cpus_cnt);
1096
1097 if (!bitmap_weight(set, c2c.cpus_cnt)) {
1098 if (c2c.node_info == 1) {
1099 ret = scnprintf(hpp->buf, hpp->size, "%21s", " ");
1100 advance_hpp(hpp, ret);
1101 }
1102 continue;
1103 }
1104
1105 if (!first) {
1106 ret = scnprintf(hpp->buf, hpp->size, " ");
1107 advance_hpp(hpp, ret);
1108 }
1109
1110 switch (c2c.node_info) {
1111 case 0:
1112 ret = scnprintf(hpp->buf, hpp->size, "%2d", node);
1113 advance_hpp(hpp, ret);
1114 break;
1115 case 1:
1116 {
1117 int num = bitmap_weight(set, c2c.cpus_cnt);
1118 struct c2c_stats *stats = &c2c_he->node_stats[node];
1119
1120 ret = scnprintf(hpp->buf, hpp->size, "%2d{%2d ", node, num);
1121 advance_hpp(hpp, ret);
1122
1123 #define DISPLAY_HITM(__h) \
1124 if (c2c_he->stats.__h> 0) { \
1125 ret = scnprintf(hpp->buf, hpp->size, "%5.1f%% ", \
1126 percent(stats->__h, c2c_he->stats.__h));\
1127 } else { \
1128 ret = scnprintf(hpp->buf, hpp->size, "%6s ", "n/a"); \
1129 }
1130
1131 switch (c2c.display) {
1132 case DISPLAY_RMT:
1133 DISPLAY_HITM(rmt_hitm);
1134 break;
1135 case DISPLAY_LCL:
1136 DISPLAY_HITM(lcl_hitm);
1137 break;
1138 case DISPLAY_TOT:
1139 DISPLAY_HITM(tot_hitm);
1140 default:
1141 break;
1142 }
1143
1144 #undef DISPLAY_HITM
1145
1146 advance_hpp(hpp, ret);
1147
1148 if (c2c_he->stats.store > 0) {
1149 ret = scnprintf(hpp->buf, hpp->size, "%5.1f%%}",
1150 percent(stats->store, c2c_he->stats.store));
1151 } else {
1152 ret = scnprintf(hpp->buf, hpp->size, "%6s}", "n/a");
1153 }
1154
1155 advance_hpp(hpp, ret);
1156 break;
1157 }
1158 case 2:
1159 ret = scnprintf(hpp->buf, hpp->size, "%2d{", node);
1160 advance_hpp(hpp, ret);
1161
1162 ret = bitmap_scnprintf(set, c2c.cpus_cnt, hpp->buf, hpp->size);
1163 advance_hpp(hpp, ret);
1164
1165 ret = scnprintf(hpp->buf, hpp->size, "}");
1166 advance_hpp(hpp, ret);
1167 break;
1168 default:
1169 break;
1170 }
1171
1172 first = false;
1173 }
1174
1175 return 0;
1176}
1177
1178static int
1179mean_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
1180 struct hist_entry *he, double mean)
1181{
1182 int width = c2c_width(fmt, hpp, he->hists);
1183 char buf[10];
1184
1185 scnprintf(buf, 10, "%6.0f", mean);
1186 return scnprintf(hpp->buf, hpp->size, "%*s", width, buf);
1187}
1188
1189#define MEAN_ENTRY(__func, __val) \
1190static int \
1191__func(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp, struct hist_entry *he) \
1192{ \
1193 struct c2c_hist_entry *c2c_he; \
1194 c2c_he = container_of(he, struct c2c_hist_entry, he); \
1195 return mean_entry(fmt, hpp, he, avg_stats(&c2c_he->cstats.__val)); \
1196}
1197
1198MEAN_ENTRY(mean_rmt_entry, rmt_hitm);
1199MEAN_ENTRY(mean_lcl_entry, lcl_hitm);
1200MEAN_ENTRY(mean_load_entry, load);
1201
1202static int
1203cpucnt_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
1204 struct hist_entry *he)
1205{
1206 struct c2c_hist_entry *c2c_he;
1207 int width = c2c_width(fmt, hpp, he->hists);
1208 char buf[10];
1209
1210 c2c_he = container_of(he, struct c2c_hist_entry, he);
1211
1212 scnprintf(buf, 10, "%d", bitmap_weight(c2c_he->cpuset, c2c.cpus_cnt));
1213 return scnprintf(hpp->buf, hpp->size, "%*s", width, buf);
1214}
1215
1216static int
1217cl_idx_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
1218 struct hist_entry *he)
1219{
1220 struct c2c_hist_entry *c2c_he;
1221 int width = c2c_width(fmt, hpp, he->hists);
1222 char buf[10];
1223
1224 c2c_he = container_of(he, struct c2c_hist_entry, he);
1225
1226 scnprintf(buf, 10, "%u", c2c_he->cacheline_idx);
1227 return scnprintf(hpp->buf, hpp->size, "%*s", width, buf);
1228}
1229
1230static int
1231cl_idx_empty_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
1232 struct hist_entry *he)
1233{
1234 int width = c2c_width(fmt, hpp, he->hists);
1235
1236 return scnprintf(hpp->buf, hpp->size, "%*s", width, "");
1237}
1238
1239#define HEADER_LOW(__h) \
1240 { \
1241 .line[1] = { \
1242 .text = __h, \
1243 }, \
1244 }
1245
1246#define HEADER_BOTH(__h0, __h1) \
1247 { \
1248 .line[0] = { \
1249 .text = __h0, \
1250 }, \
1251 .line[1] = { \
1252 .text = __h1, \
1253 }, \
1254 }
1255
1256#define HEADER_SPAN(__h0, __h1, __s) \
1257 { \
1258 .line[0] = { \
1259 .text = __h0, \
1260 .span = __s, \
1261 }, \
1262 .line[1] = { \
1263 .text = __h1, \
1264 }, \
1265 }
1266
1267#define HEADER_SPAN_LOW(__h) \
1268 { \
1269 .line[1] = { \
1270 .text = __h, \
1271 }, \
1272 }
1273
1274static struct c2c_dimension dim_dcacheline = {
1275 .header = HEADER_SPAN("--- Cacheline ----", "Address", 2),
1276 .name = "dcacheline",
1277 .cmp = dcacheline_cmp,
1278 .entry = dcacheline_entry,
1279 .width = 18,
1280};
1281
1282static struct c2c_dimension dim_dcacheline_node = {
1283 .header = HEADER_LOW("Node"),
1284 .name = "dcacheline_node",
1285 .cmp = empty_cmp,
1286 .entry = dcacheline_node_entry,
1287 .width = 4,
1288};
1289
1290static struct c2c_dimension dim_dcacheline_count = {
1291 .header = HEADER_LOW("PA cnt"),
1292 .name = "dcacheline_count",
1293 .cmp = empty_cmp,
1294 .entry = dcacheline_node_count,
1295 .width = 6,
1296};
1297
1298static struct c2c_header header_offset_tui = HEADER_SPAN("-----", "Off", 2);
1299
1300static struct c2c_dimension dim_offset = {
1301 .header = HEADER_SPAN("--- Data address -", "Offset", 2),
1302 .name = "offset",
1303 .cmp = offset_cmp,
1304 .entry = offset_entry,
1305 .width = 18,
1306};
1307
1308static struct c2c_dimension dim_offset_node = {
1309 .header = HEADER_LOW("Node"),
1310 .name = "offset_node",
1311 .cmp = empty_cmp,
1312 .entry = dcacheline_node_entry,
1313 .width = 4,
1314};
1315
1316static struct c2c_dimension dim_iaddr = {
1317 .header = HEADER_LOW("Code address"),
1318 .name = "iaddr",
1319 .cmp = iaddr_cmp,
1320 .entry = iaddr_entry,
1321 .width = 18,
1322};
1323
1324static struct c2c_dimension dim_tot_hitm = {
1325 .header = HEADER_SPAN("----- LLC Load Hitm -----", "Total", 2),
1326 .name = "tot_hitm",
1327 .cmp = tot_hitm_cmp,
1328 .entry = tot_hitm_entry,
1329 .width = 7,
1330};
1331
1332static struct c2c_dimension dim_lcl_hitm = {
1333 .header = HEADER_SPAN_LOW("Lcl"),
1334 .name = "lcl_hitm",
1335 .cmp = lcl_hitm_cmp,
1336 .entry = lcl_hitm_entry,
1337 .width = 7,
1338};
1339
1340static struct c2c_dimension dim_rmt_hitm = {
1341 .header = HEADER_SPAN_LOW("Rmt"),
1342 .name = "rmt_hitm",
1343 .cmp = rmt_hitm_cmp,
1344 .entry = rmt_hitm_entry,
1345 .width = 7,
1346};
1347
1348static struct c2c_dimension dim_cl_rmt_hitm = {
1349 .header = HEADER_SPAN("----- HITM -----", "Rmt", 1),
1350 .name = "cl_rmt_hitm",
1351 .cmp = rmt_hitm_cmp,
1352 .entry = rmt_hitm_entry,
1353 .width = 7,
1354};
1355
1356static struct c2c_dimension dim_cl_lcl_hitm = {
1357 .header = HEADER_SPAN_LOW("Lcl"),
1358 .name = "cl_lcl_hitm",
1359 .cmp = lcl_hitm_cmp,
1360 .entry = lcl_hitm_entry,
1361 .width = 7,
1362};
1363
1364static struct c2c_dimension dim_stores = {
1365 .header = HEADER_SPAN("---- Store Reference ----", "Total", 2),
1366 .name = "stores",
1367 .cmp = store_cmp,
1368 .entry = store_entry,
1369 .width = 7,
1370};
1371
1372static struct c2c_dimension dim_stores_l1hit = {
1373 .header = HEADER_SPAN_LOW("L1Hit"),
1374 .name = "stores_l1hit",
1375 .cmp = st_l1hit_cmp,
1376 .entry = st_l1hit_entry,
1377 .width = 7,
1378};
1379
1380static struct c2c_dimension dim_stores_l1miss = {
1381 .header = HEADER_SPAN_LOW("L1Miss"),
1382 .name = "stores_l1miss",
1383 .cmp = st_l1miss_cmp,
1384 .entry = st_l1miss_entry,
1385 .width = 7,
1386};
1387
1388static struct c2c_dimension dim_cl_stores_l1hit = {
1389 .header = HEADER_SPAN("-- Store Refs --", "L1 Hit", 1),
1390 .name = "cl_stores_l1hit",
1391 .cmp = st_l1hit_cmp,
1392 .entry = st_l1hit_entry,
1393 .width = 7,
1394};
1395
1396static struct c2c_dimension dim_cl_stores_l1miss = {
1397 .header = HEADER_SPAN_LOW("L1 Miss"),
1398 .name = "cl_stores_l1miss",
1399 .cmp = st_l1miss_cmp,
1400 .entry = st_l1miss_entry,
1401 .width = 7,
1402};
1403
1404static struct c2c_dimension dim_ld_fbhit = {
1405 .header = HEADER_SPAN("----- Core Load Hit -----", "FB", 2),
1406 .name = "ld_fbhit",
1407 .cmp = ld_fbhit_cmp,
1408 .entry = ld_fbhit_entry,
1409 .width = 7,
1410};
1411
1412static struct c2c_dimension dim_ld_l1hit = {
1413 .header = HEADER_SPAN_LOW("L1"),
1414 .name = "ld_l1hit",
1415 .cmp = ld_l1hit_cmp,
1416 .entry = ld_l1hit_entry,
1417 .width = 7,
1418};
1419
1420static struct c2c_dimension dim_ld_l2hit = {
1421 .header = HEADER_SPAN_LOW("L2"),
1422 .name = "ld_l2hit",
1423 .cmp = ld_l2hit_cmp,
1424 .entry = ld_l2hit_entry,
1425 .width = 7,
1426};
1427
1428static struct c2c_dimension dim_ld_llchit = {
1429 .header = HEADER_SPAN("-- LLC Load Hit --", "Llc", 1),
1430 .name = "ld_lclhit",
1431 .cmp = ld_llchit_cmp,
1432 .entry = ld_llchit_entry,
1433 .width = 8,
1434};
1435
1436static struct c2c_dimension dim_ld_rmthit = {
1437 .header = HEADER_SPAN_LOW("Rmt"),
1438 .name = "ld_rmthit",
1439 .cmp = rmt_hit_cmp,
1440 .entry = rmt_hit_entry,
1441 .width = 8,
1442};
1443
1444static struct c2c_dimension dim_ld_llcmiss = {
1445 .header = HEADER_BOTH("LLC", "Ld Miss"),
1446 .name = "ld_llcmiss",
1447 .cmp = ld_llcmiss_cmp,
1448 .entry = ld_llcmiss_entry,
1449 .width = 7,
1450};
1451
1452static struct c2c_dimension dim_tot_recs = {
1453 .header = HEADER_BOTH("Total", "records"),
1454 .name = "tot_recs",
1455 .cmp = tot_recs_cmp,
1456 .entry = tot_recs_entry,
1457 .width = 7,
1458};
1459
1460static struct c2c_dimension dim_tot_loads = {
1461 .header = HEADER_BOTH("Total", "Loads"),
1462 .name = "tot_loads",
1463 .cmp = tot_loads_cmp,
1464 .entry = tot_loads_entry,
1465 .width = 7,
1466};
1467
1468static struct c2c_header percent_hitm_header[] = {
1469 [DISPLAY_LCL] = HEADER_BOTH("Lcl", "Hitm"),
1470 [DISPLAY_RMT] = HEADER_BOTH("Rmt", "Hitm"),
1471 [DISPLAY_TOT] = HEADER_BOTH("Tot", "Hitm"),
1472};
1473
1474static struct c2c_dimension dim_percent_hitm = {
1475 .name = "percent_hitm",
1476 .cmp = percent_hitm_cmp,
1477 .entry = percent_hitm_entry,
1478 .color = percent_hitm_color,
1479 .width = 7,
1480};
1481
1482static struct c2c_dimension dim_percent_rmt_hitm = {
1483 .header = HEADER_SPAN("----- HITM -----", "Rmt", 1),
1484 .name = "percent_rmt_hitm",
1485 .cmp = percent_rmt_hitm_cmp,
1486 .entry = percent_rmt_hitm_entry,
1487 .color = percent_rmt_hitm_color,
1488 .width = 7,
1489};
1490
1491static struct c2c_dimension dim_percent_lcl_hitm = {
1492 .header = HEADER_SPAN_LOW("Lcl"),
1493 .name = "percent_lcl_hitm",
1494 .cmp = percent_lcl_hitm_cmp,
1495 .entry = percent_lcl_hitm_entry,
1496 .color = percent_lcl_hitm_color,
1497 .width = 7,
1498};
1499
1500static struct c2c_dimension dim_percent_stores_l1hit = {
1501 .header = HEADER_SPAN("-- Store Refs --", "L1 Hit", 1),
1502 .name = "percent_stores_l1hit",
1503 .cmp = percent_stores_l1hit_cmp,
1504 .entry = percent_stores_l1hit_entry,
1505 .color = percent_stores_l1hit_color,
1506 .width = 7,
1507};
1508
1509static struct c2c_dimension dim_percent_stores_l1miss = {
1510 .header = HEADER_SPAN_LOW("L1 Miss"),
1511 .name = "percent_stores_l1miss",
1512 .cmp = percent_stores_l1miss_cmp,
1513 .entry = percent_stores_l1miss_entry,
1514 .color = percent_stores_l1miss_color,
1515 .width = 7,
1516};
1517
1518static struct c2c_dimension dim_dram_lcl = {
1519 .header = HEADER_SPAN("--- Load Dram ----", "Lcl", 1),
1520 .name = "dram_lcl",
1521 .cmp = lcl_dram_cmp,
1522 .entry = lcl_dram_entry,
1523 .width = 8,
1524};
1525
1526static struct c2c_dimension dim_dram_rmt = {
1527 .header = HEADER_SPAN_LOW("Rmt"),
1528 .name = "dram_rmt",
1529 .cmp = rmt_dram_cmp,
1530 .entry = rmt_dram_entry,
1531 .width = 8,
1532};
1533
1534static struct c2c_dimension dim_pid = {
1535 .header = HEADER_LOW("Pid"),
1536 .name = "pid",
1537 .cmp = pid_cmp,
1538 .entry = pid_entry,
1539 .width = 7,
1540};
1541
1542static struct c2c_dimension dim_tid = {
1543 .header = HEADER_LOW("Tid"),
1544 .name = "tid",
1545 .se = &sort_thread,
1546};
1547
1548static struct c2c_dimension dim_symbol = {
1549 .name = "symbol",
1550 .se = &sort_sym,
1551};
1552
1553static struct c2c_dimension dim_dso = {
1554 .header = HEADER_BOTH("Shared", "Object"),
1555 .name = "dso",
1556 .se = &sort_dso,
1557};
1558
1559static struct c2c_header header_node[3] = {
1560 HEADER_LOW("Node"),
1561 HEADER_LOW("Node{cpus %hitms %stores}"),
1562 HEADER_LOW("Node{cpu list}"),
1563};
1564
1565static struct c2c_dimension dim_node = {
1566 .name = "node",
1567 .cmp = empty_cmp,
1568 .entry = node_entry,
1569 .width = 4,
1570};
1571
1572static struct c2c_dimension dim_mean_rmt = {
1573 .header = HEADER_SPAN("---------- cycles ----------", "rmt hitm", 2),
1574 .name = "mean_rmt",
1575 .cmp = empty_cmp,
1576 .entry = mean_rmt_entry,
1577 .width = 8,
1578};
1579
1580static struct c2c_dimension dim_mean_lcl = {
1581 .header = HEADER_SPAN_LOW("lcl hitm"),
1582 .name = "mean_lcl",
1583 .cmp = empty_cmp,
1584 .entry = mean_lcl_entry,
1585 .width = 8,
1586};
1587
1588static struct c2c_dimension dim_mean_load = {
1589 .header = HEADER_SPAN_LOW("load"),
1590 .name = "mean_load",
1591 .cmp = empty_cmp,
1592 .entry = mean_load_entry,
1593 .width = 8,
1594};
1595
1596static struct c2c_dimension dim_cpucnt = {
1597 .header = HEADER_BOTH("cpu", "cnt"),
1598 .name = "cpucnt",
1599 .cmp = empty_cmp,
1600 .entry = cpucnt_entry,
1601 .width = 8,
1602};
1603
1604static struct c2c_dimension dim_srcline = {
1605 .name = "cl_srcline",
1606 .se = &sort_srcline,
1607};
1608
1609static struct c2c_dimension dim_dcacheline_idx = {
1610 .header = HEADER_LOW("Index"),
1611 .name = "cl_idx",
1612 .cmp = empty_cmp,
1613 .entry = cl_idx_entry,
1614 .width = 5,
1615};
1616
1617static struct c2c_dimension dim_dcacheline_num = {
1618 .header = HEADER_LOW("Num"),
1619 .name = "cl_num",
1620 .cmp = empty_cmp,
1621 .entry = cl_idx_entry,
1622 .width = 5,
1623};
1624
1625static struct c2c_dimension dim_dcacheline_num_empty = {
1626 .header = HEADER_LOW("Num"),
1627 .name = "cl_num_empty",
1628 .cmp = empty_cmp,
1629 .entry = cl_idx_empty_entry,
1630 .width = 5,
1631};
1632
1633static struct c2c_dimension *dimensions[] = {
1634 &dim_dcacheline,
1635 &dim_dcacheline_node,
1636 &dim_dcacheline_count,
1637 &dim_offset,
1638 &dim_offset_node,
1639 &dim_iaddr,
1640 &dim_tot_hitm,
1641 &dim_lcl_hitm,
1642 &dim_rmt_hitm,
1643 &dim_cl_lcl_hitm,
1644 &dim_cl_rmt_hitm,
1645 &dim_stores,
1646 &dim_stores_l1hit,
1647 &dim_stores_l1miss,
1648 &dim_cl_stores_l1hit,
1649 &dim_cl_stores_l1miss,
1650 &dim_ld_fbhit,
1651 &dim_ld_l1hit,
1652 &dim_ld_l2hit,
1653 &dim_ld_llchit,
1654 &dim_ld_rmthit,
1655 &dim_ld_llcmiss,
1656 &dim_tot_recs,
1657 &dim_tot_loads,
1658 &dim_percent_hitm,
1659 &dim_percent_rmt_hitm,
1660 &dim_percent_lcl_hitm,
1661 &dim_percent_stores_l1hit,
1662 &dim_percent_stores_l1miss,
1663 &dim_dram_lcl,
1664 &dim_dram_rmt,
1665 &dim_pid,
1666 &dim_tid,
1667 &dim_symbol,
1668 &dim_dso,
1669 &dim_node,
1670 &dim_mean_rmt,
1671 &dim_mean_lcl,
1672 &dim_mean_load,
1673 &dim_cpucnt,
1674 &dim_srcline,
1675 &dim_dcacheline_idx,
1676 &dim_dcacheline_num,
1677 &dim_dcacheline_num_empty,
1678 NULL,
1679};
1680
1681static void fmt_free(struct perf_hpp_fmt *fmt)
1682{
1683 struct c2c_fmt *c2c_fmt;
1684
1685 c2c_fmt = container_of(fmt, struct c2c_fmt, fmt);
1686 free(c2c_fmt);
1687}
1688
1689static bool fmt_equal(struct perf_hpp_fmt *a, struct perf_hpp_fmt *b)
1690{
1691 struct c2c_fmt *c2c_a = container_of(a, struct c2c_fmt, fmt);
1692 struct c2c_fmt *c2c_b = container_of(b, struct c2c_fmt, fmt);
1693
1694 return c2c_a->dim == c2c_b->dim;
1695}
1696
1697static struct c2c_dimension *get_dimension(const char *name)
1698{
1699 unsigned int i;
1700
1701 for (i = 0; dimensions[i]; i++) {
1702 struct c2c_dimension *dim = dimensions[i];
1703
1704 if (!strcmp(dim->name, name))
1705 return dim;
1706 };
1707
1708 return NULL;
1709}
1710
1711static int c2c_se_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
1712 struct hist_entry *he)
1713{
1714 struct c2c_fmt *c2c_fmt = container_of(fmt, struct c2c_fmt, fmt);
1715 struct c2c_dimension *dim = c2c_fmt->dim;
1716 size_t len = fmt->user_len;
1717
1718 if (!len) {
1719 len = hists__col_len(he->hists, dim->se->se_width_idx);
1720
1721 if (dim == &dim_symbol || dim == &dim_srcline)
1722 len = symbol_width(he->hists, dim->se);
1723 }
1724
1725 return dim->se->se_snprintf(he, hpp->buf, hpp->size, len);
1726}
1727
1728static int64_t c2c_se_cmp(struct perf_hpp_fmt *fmt,
1729 struct hist_entry *a, struct hist_entry *b)
1730{
1731 struct c2c_fmt *c2c_fmt = container_of(fmt, struct c2c_fmt, fmt);
1732 struct c2c_dimension *dim = c2c_fmt->dim;
1733
1734 return dim->se->se_cmp(a, b);
1735}
1736
1737static int64_t c2c_se_collapse(struct perf_hpp_fmt *fmt,
1738 struct hist_entry *a, struct hist_entry *b)
1739{
1740 struct c2c_fmt *c2c_fmt = container_of(fmt, struct c2c_fmt, fmt);
1741 struct c2c_dimension *dim = c2c_fmt->dim;
1742 int64_t (*collapse_fn)(struct hist_entry *, struct hist_entry *);
1743
1744 collapse_fn = dim->se->se_collapse ?: dim->se->se_cmp;
1745 return collapse_fn(a, b);
1746}
1747
1748static struct c2c_fmt *get_format(const char *name)
1749{
1750 struct c2c_dimension *dim = get_dimension(name);
1751 struct c2c_fmt *c2c_fmt;
1752 struct perf_hpp_fmt *fmt;
1753
1754 if (!dim)
1755 return NULL;
1756
1757 c2c_fmt = zalloc(sizeof(*c2c_fmt));
1758 if (!c2c_fmt)
1759 return NULL;
1760
1761 c2c_fmt->dim = dim;
1762
1763 fmt = &c2c_fmt->fmt;
1764 INIT_LIST_HEAD(&fmt->list);
1765 INIT_LIST_HEAD(&fmt->sort_list);
1766
1767 fmt->cmp = dim->se ? c2c_se_cmp : dim->cmp;
1768 fmt->sort = dim->se ? c2c_se_cmp : dim->cmp;
1769 fmt->color = dim->se ? NULL : dim->color;
1770 fmt->entry = dim->se ? c2c_se_entry : dim->entry;
1771 fmt->header = c2c_header;
1772 fmt->width = c2c_width;
1773 fmt->collapse = dim->se ? c2c_se_collapse : dim->cmp;
1774 fmt->equal = fmt_equal;
1775 fmt->free = fmt_free;
1776
1777 return c2c_fmt;
1778}
1779
1780static int c2c_hists__init_output(struct perf_hpp_list *hpp_list, char *name)
1781{
1782 struct c2c_fmt *c2c_fmt = get_format(name);
1783
1784 if (!c2c_fmt) {
1785 reset_dimensions();
1786 return output_field_add(hpp_list, name);
1787 }
1788
1789 perf_hpp_list__column_register(hpp_list, &c2c_fmt->fmt);
1790 return 0;
1791}
1792
1793static int c2c_hists__init_sort(struct perf_hpp_list *hpp_list, char *name)
1794{
1795 struct c2c_fmt *c2c_fmt = get_format(name);
1796 struct c2c_dimension *dim;
1797
1798 if (!c2c_fmt) {
1799 reset_dimensions();
1800 return sort_dimension__add(hpp_list, name, NULL, 0);
1801 }
1802
1803 dim = c2c_fmt->dim;
1804 if (dim == &dim_dso)
1805 hpp_list->dso = 1;
1806
1807 perf_hpp_list__register_sort_field(hpp_list, &c2c_fmt->fmt);
1808 return 0;
1809}
1810
1811#define PARSE_LIST(_list, _fn) \
1812 do { \
1813 char *tmp, *tok; \
1814 ret = 0; \
1815 \
1816 if (!_list) \
1817 break; \
1818 \
1819 for (tok = strtok_r((char *)_list, ", ", &tmp); \
1820 tok; tok = strtok_r(NULL, ", ", &tmp)) { \
1821 ret = _fn(hpp_list, tok); \
1822 if (ret == -EINVAL) { \
1823 pr_err("Invalid --fields key: `%s'", tok); \
1824 break; \
1825 } else if (ret == -ESRCH) { \
1826 pr_err("Unknown --fields key: `%s'", tok); \
1827 break; \
1828 } \
1829 } \
1830 } while (0)
1831
1832static int hpp_list__parse(struct perf_hpp_list *hpp_list,
1833 const char *output_,
1834 const char *sort_)
1835{
1836 char *output = output_ ? strdup(output_) : NULL;
1837 char *sort = sort_ ? strdup(sort_) : NULL;
1838 int ret;
1839
1840 PARSE_LIST(output, c2c_hists__init_output);
1841 PARSE_LIST(sort, c2c_hists__init_sort);
1842
1843 /* copy sort keys to output fields */
1844 perf_hpp__setup_output_field(hpp_list);
1845
1846 /*
1847 * We dont need other sorting keys other than those
1848 * we already specified. It also really slows down
1849 * the processing a lot with big number of output
1850 * fields, so switching this off for c2c.
1851 */
1852
1853#if 0
1854 /* and then copy output fields to sort keys */
1855 perf_hpp__append_sort_keys(&hists->list);
1856#endif
1857
1858 free(output);
1859 free(sort);
1860 return ret;
1861}
1862
1863static int c2c_hists__init(struct c2c_hists *hists,
1864 const char *sort,
1865 int nr_header_lines)
1866{
1867 __hists__init(&hists->hists, &hists->list);
1868
1869 /*
1870 * Initialize only with sort fields, we need to resort
1871 * later anyway, and that's where we add output fields
1872 * as well.
1873 */
1874 perf_hpp_list__init(&hists->list);
1875
1876 /* Overload number of header lines.*/
1877 hists->list.nr_header_lines = nr_header_lines;
1878
1879 return hpp_list__parse(&hists->list, NULL, sort);
1880}
1881
1882static int c2c_hists__reinit(struct c2c_hists *c2c_hists,
1883 const char *output,
1884 const char *sort)
1885{
1886 perf_hpp__reset_output_field(&c2c_hists->list);
1887 return hpp_list__parse(&c2c_hists->list, output, sort);
1888}
1889
1890#define DISPLAY_LINE_LIMIT 0.001
1891
1892static bool he__display(struct hist_entry *he, struct c2c_stats *stats)
1893{
1894 struct c2c_hist_entry *c2c_he;
1895 double ld_dist;
1896
1897 if (c2c.show_all)
1898 return true;
1899
1900 c2c_he = container_of(he, struct c2c_hist_entry, he);
1901
1902#define FILTER_HITM(__h) \
1903 if (stats->__h) { \
1904 ld_dist = ((double)c2c_he->stats.__h / stats->__h); \
1905 if (ld_dist < DISPLAY_LINE_LIMIT) \
1906 he->filtered = HIST_FILTER__C2C; \
1907 } else { \
1908 he->filtered = HIST_FILTER__C2C; \
1909 }
1910
1911 switch (c2c.display) {
1912 case DISPLAY_LCL:
1913 FILTER_HITM(lcl_hitm);
1914 break;
1915 case DISPLAY_RMT:
1916 FILTER_HITM(rmt_hitm);
1917 break;
1918 case DISPLAY_TOT:
1919 FILTER_HITM(tot_hitm);
1920 default:
1921 break;
1922 };
1923
1924#undef FILTER_HITM
1925
1926 return he->filtered == 0;
1927}
1928
1929static inline int valid_hitm_or_store(struct hist_entry *he)
1930{
1931 struct c2c_hist_entry *c2c_he;
1932 bool has_hitm;
1933
1934 c2c_he = container_of(he, struct c2c_hist_entry, he);
1935 has_hitm = c2c.display == DISPLAY_TOT ? c2c_he->stats.tot_hitm :
1936 c2c.display == DISPLAY_LCL ? c2c_he->stats.lcl_hitm :
1937 c2c_he->stats.rmt_hitm;
1938 return has_hitm || c2c_he->stats.store;
1939}
1940
1941static void set_node_width(struct c2c_hist_entry *c2c_he, int len)
1942{
1943 struct c2c_dimension *dim;
1944
1945 dim = &c2c.hists == c2c_he->hists ?
1946 &dim_dcacheline_node : &dim_offset_node;
1947
1948 if (len > dim->width)
1949 dim->width = len;
1950}
1951
1952static int set_nodestr(struct c2c_hist_entry *c2c_he)
1953{
1954 char buf[30];
1955 int len;
1956
1957 if (c2c_he->nodestr)
1958 return 0;
1959
1960 if (bitmap_weight(c2c_he->nodeset, c2c.nodes_cnt)) {
1961 len = bitmap_scnprintf(c2c_he->nodeset, c2c.nodes_cnt,
1962 buf, sizeof(buf));
1963 } else {
1964 len = scnprintf(buf, sizeof(buf), "N/A");
1965 }
1966
1967 set_node_width(c2c_he, len);
1968 c2c_he->nodestr = strdup(buf);
1969 return c2c_he->nodestr ? 0 : -ENOMEM;
1970}
1971
1972static void calc_width(struct c2c_hist_entry *c2c_he)
1973{
1974 struct c2c_hists *c2c_hists;
1975
1976 c2c_hists = container_of(c2c_he->he.hists, struct c2c_hists, hists);
1977 hists__calc_col_len(&c2c_hists->hists, &c2c_he->he);
1978 set_nodestr(c2c_he);
1979}
1980
1981static int filter_cb(struct hist_entry *he, void *arg __maybe_unused)
1982{
1983 struct c2c_hist_entry *c2c_he;
1984
1985 c2c_he = container_of(he, struct c2c_hist_entry, he);
1986
1987 if (c2c.show_src && !he->srcline)
1988 he->srcline = hist_entry__srcline(he);
1989
1990 calc_width(c2c_he);
1991
1992 if (!valid_hitm_or_store(he))
1993 he->filtered = HIST_FILTER__C2C;
1994
1995 return 0;
1996}
1997
1998static int resort_cl_cb(struct hist_entry *he, void *arg __maybe_unused)
1999{
2000 struct c2c_hist_entry *c2c_he;
2001 struct c2c_hists *c2c_hists;
2002 bool display = he__display(he, &c2c.hitm_stats);
2003
2004 c2c_he = container_of(he, struct c2c_hist_entry, he);
2005 c2c_hists = c2c_he->hists;
2006
2007 if (display && c2c_hists) {
2008 static unsigned int idx;
2009
2010 c2c_he->cacheline_idx = idx++;
2011 calc_width(c2c_he);
2012
2013 c2c_hists__reinit(c2c_hists, c2c.cl_output, c2c.cl_resort);
2014
2015 hists__collapse_resort(&c2c_hists->hists, NULL);
2016 hists__output_resort_cb(&c2c_hists->hists, NULL, filter_cb);
2017 }
2018
2019 return 0;
2020}
2021
2022static void setup_nodes_header(void)
2023{
2024 dim_node.header = header_node[c2c.node_info];
2025}
2026
2027static int setup_nodes(struct perf_session *session)
2028{
2029 struct numa_node *n;
2030 unsigned long **nodes;
2031 int node, cpu;
2032 int *cpu2node;
2033
2034 if (c2c.node_info > 2)
2035 c2c.node_info = 2;
2036
2037 c2c.nodes_cnt = session->header.env.nr_numa_nodes;
2038 c2c.cpus_cnt = session->header.env.nr_cpus_avail;
2039
2040 n = session->header.env.numa_nodes;
2041 if (!n)
2042 return -EINVAL;
2043
2044 nodes = zalloc(sizeof(unsigned long *) * c2c.nodes_cnt);
2045 if (!nodes)
2046 return -ENOMEM;
2047
2048 c2c.nodes = nodes;
2049
2050 cpu2node = zalloc(sizeof(int) * c2c.cpus_cnt);
2051 if (!cpu2node)
2052 return -ENOMEM;
2053
2054 for (cpu = 0; cpu < c2c.cpus_cnt; cpu++)
2055 cpu2node[cpu] = -1;
2056
2057 c2c.cpu2node = cpu2node;
2058
2059 for (node = 0; node < c2c.nodes_cnt; node++) {
2060 struct perf_cpu_map *map = n[node].map;
2061 unsigned long *set;
2062
2063 set = bitmap_alloc(c2c.cpus_cnt);
2064 if (!set)
2065 return -ENOMEM;
2066
2067 nodes[node] = set;
2068
2069 /* empty node, skip */
2070 if (perf_cpu_map__empty(map))
2071 continue;
2072
2073 for (cpu = 0; cpu < map->nr; cpu++) {
2074 set_bit(map->map[cpu], set);
2075
2076 if (WARN_ONCE(cpu2node[map->map[cpu]] != -1, "node/cpu topology bug"))
2077 return -EINVAL;
2078
2079 cpu2node[map->map[cpu]] = node;
2080 }
2081 }
2082
2083 setup_nodes_header();
2084 return 0;
2085}
2086
2087#define HAS_HITMS(__h) ((__h)->stats.lcl_hitm || (__h)->stats.rmt_hitm)
2088
2089static int resort_hitm_cb(struct hist_entry *he, void *arg __maybe_unused)
2090{
2091 struct c2c_hist_entry *c2c_he;
2092 c2c_he = container_of(he, struct c2c_hist_entry, he);
2093
2094 if (HAS_HITMS(c2c_he)) {
2095 c2c.shared_clines++;
2096 c2c_add_stats(&c2c.hitm_stats, &c2c_he->stats);
2097 }
2098
2099 return 0;
2100}
2101
2102static int hists__iterate_cb(struct hists *hists, hists__resort_cb_t cb)
2103{
2104 struct rb_node *next = rb_first_cached(&hists->entries);
2105 int ret = 0;
2106
2107 while (next) {
2108 struct hist_entry *he;
2109
2110 he = rb_entry(next, struct hist_entry, rb_node);
2111 ret = cb(he, NULL);
2112 if (ret)
2113 break;
2114 next = rb_next(&he->rb_node);
2115 }
2116
2117 return ret;
2118}
2119
2120static void print_c2c__display_stats(FILE *out)
2121{
2122 int llc_misses;
2123 struct c2c_stats *stats = &c2c.hists.stats;
2124
2125 llc_misses = stats->lcl_dram +
2126 stats->rmt_dram +
2127 stats->rmt_hit +
2128 stats->rmt_hitm;
2129
2130 fprintf(out, "=================================================\n");
2131 fprintf(out, " Trace Event Information \n");
2132 fprintf(out, "=================================================\n");
2133 fprintf(out, " Total records : %10d\n", stats->nr_entries);
2134 fprintf(out, " Locked Load/Store Operations : %10d\n", stats->locks);
2135 fprintf(out, " Load Operations : %10d\n", stats->load);
2136 fprintf(out, " Loads - uncacheable : %10d\n", stats->ld_uncache);
2137 fprintf(out, " Loads - IO : %10d\n", stats->ld_io);
2138 fprintf(out, " Loads - Miss : %10d\n", stats->ld_miss);
2139 fprintf(out, " Loads - no mapping : %10d\n", stats->ld_noadrs);
2140 fprintf(out, " Load Fill Buffer Hit : %10d\n", stats->ld_fbhit);
2141 fprintf(out, " Load L1D hit : %10d\n", stats->ld_l1hit);
2142 fprintf(out, " Load L2D hit : %10d\n", stats->ld_l2hit);
2143 fprintf(out, " Load LLC hit : %10d\n", stats->ld_llchit + stats->lcl_hitm);
2144 fprintf(out, " Load Local HITM : %10d\n", stats->lcl_hitm);
2145 fprintf(out, " Load Remote HITM : %10d\n", stats->rmt_hitm);
2146 fprintf(out, " Load Remote HIT : %10d\n", stats->rmt_hit);
2147 fprintf(out, " Load Local DRAM : %10d\n", stats->lcl_dram);
2148 fprintf(out, " Load Remote DRAM : %10d\n", stats->rmt_dram);
2149 fprintf(out, " Load MESI State Exclusive : %10d\n", stats->ld_excl);
2150 fprintf(out, " Load MESI State Shared : %10d\n", stats->ld_shared);
2151 fprintf(out, " Load LLC Misses : %10d\n", llc_misses);
2152 fprintf(out, " LLC Misses to Local DRAM : %10.1f%%\n", ((double)stats->lcl_dram/(double)llc_misses) * 100.);
2153 fprintf(out, " LLC Misses to Remote DRAM : %10.1f%%\n", ((double)stats->rmt_dram/(double)llc_misses) * 100.);
2154 fprintf(out, " LLC Misses to Remote cache (HIT) : %10.1f%%\n", ((double)stats->rmt_hit /(double)llc_misses) * 100.);
2155 fprintf(out, " LLC Misses to Remote cache (HITM) : %10.1f%%\n", ((double)stats->rmt_hitm/(double)llc_misses) * 100.);
2156 fprintf(out, " Store Operations : %10d\n", stats->store);
2157 fprintf(out, " Store - uncacheable : %10d\n", stats->st_uncache);
2158 fprintf(out, " Store - no mapping : %10d\n", stats->st_noadrs);
2159 fprintf(out, " Store L1D Hit : %10d\n", stats->st_l1hit);
2160 fprintf(out, " Store L1D Miss : %10d\n", stats->st_l1miss);
2161 fprintf(out, " No Page Map Rejects : %10d\n", stats->nomap);
2162 fprintf(out, " Unable to parse data source : %10d\n", stats->noparse);
2163}
2164
2165static void print_shared_cacheline_info(FILE *out)
2166{
2167 struct c2c_stats *stats = &c2c.hitm_stats;
2168 int hitm_cnt = stats->lcl_hitm + stats->rmt_hitm;
2169
2170 fprintf(out, "=================================================\n");
2171 fprintf(out, " Global Shared Cache Line Event Information \n");
2172 fprintf(out, "=================================================\n");
2173 fprintf(out, " Total Shared Cache Lines : %10d\n", c2c.shared_clines);
2174 fprintf(out, " Load HITs on shared lines : %10d\n", stats->load);
2175 fprintf(out, " Fill Buffer Hits on shared lines : %10d\n", stats->ld_fbhit);
2176 fprintf(out, " L1D hits on shared lines : %10d\n", stats->ld_l1hit);
2177 fprintf(out, " L2D hits on shared lines : %10d\n", stats->ld_l2hit);
2178 fprintf(out, " LLC hits on shared lines : %10d\n", stats->ld_llchit + stats->lcl_hitm);
2179 fprintf(out, " Locked Access on shared lines : %10d\n", stats->locks);
2180 fprintf(out, " Store HITs on shared lines : %10d\n", stats->store);
2181 fprintf(out, " Store L1D hits on shared lines : %10d\n", stats->st_l1hit);
2182 fprintf(out, " Total Merged records : %10d\n", hitm_cnt + stats->store);
2183}
2184
2185static void print_cacheline(struct c2c_hists *c2c_hists,
2186 struct hist_entry *he_cl,
2187 struct perf_hpp_list *hpp_list,
2188 FILE *out)
2189{
2190 char bf[1000];
2191 struct perf_hpp hpp = {
2192 .buf = bf,
2193 .size = 1000,
2194 };
2195 static bool once;
2196
2197 if (!once) {
2198 hists__fprintf_headers(&c2c_hists->hists, out);
2199 once = true;
2200 } else {
2201 fprintf(out, "\n");
2202 }
2203
2204 fprintf(out, " -------------------------------------------------------------\n");
2205 __hist_entry__snprintf(he_cl, &hpp, hpp_list);
2206 fprintf(out, "%s\n", bf);
2207 fprintf(out, " -------------------------------------------------------------\n");
2208
2209 hists__fprintf(&c2c_hists->hists, false, 0, 0, 0, out, false);
2210}
2211
2212static void print_pareto(FILE *out)
2213{
2214 struct perf_hpp_list hpp_list;
2215 struct rb_node *nd;
2216 int ret;
2217
2218 perf_hpp_list__init(&hpp_list);
2219 ret = hpp_list__parse(&hpp_list,
2220 "cl_num,"
2221 "cl_rmt_hitm,"
2222 "cl_lcl_hitm,"
2223 "cl_stores_l1hit,"
2224 "cl_stores_l1miss,"
2225 "dcacheline",
2226 NULL);
2227
2228 if (WARN_ONCE(ret, "failed to setup sort entries\n"))
2229 return;
2230
2231 nd = rb_first_cached(&c2c.hists.hists.entries);
2232
2233 for (; nd; nd = rb_next(nd)) {
2234 struct hist_entry *he = rb_entry(nd, struct hist_entry, rb_node);
2235 struct c2c_hist_entry *c2c_he;
2236
2237 if (he->filtered)
2238 continue;
2239
2240 c2c_he = container_of(he, struct c2c_hist_entry, he);
2241 print_cacheline(c2c_he->hists, he, &hpp_list, out);
2242 }
2243}
2244
2245static void print_c2c_info(FILE *out, struct perf_session *session)
2246{
2247 struct evlist *evlist = session->evlist;
2248 struct evsel *evsel;
2249 bool first = true;
2250
2251 fprintf(out, "=================================================\n");
2252 fprintf(out, " c2c details \n");
2253 fprintf(out, "=================================================\n");
2254
2255 evlist__for_each_entry(evlist, evsel) {
2256 fprintf(out, "%-36s: %s\n", first ? " Events" : "",
2257 perf_evsel__name(evsel));
2258 first = false;
2259 }
2260 fprintf(out, " Cachelines sort on : %s HITMs\n",
2261 display_str[c2c.display]);
2262 fprintf(out, " Cacheline data grouping : %s\n", c2c.cl_sort);
2263}
2264
2265static void perf_c2c__hists_fprintf(FILE *out, struct perf_session *session)
2266{
2267 setup_pager();
2268
2269 print_c2c__display_stats(out);
2270 fprintf(out, "\n");
2271 print_shared_cacheline_info(out);
2272 fprintf(out, "\n");
2273 print_c2c_info(out, session);
2274
2275 if (c2c.stats_only)
2276 return;
2277
2278 fprintf(out, "\n");
2279 fprintf(out, "=================================================\n");
2280 fprintf(out, " Shared Data Cache Line Table \n");
2281 fprintf(out, "=================================================\n");
2282 fprintf(out, "#\n");
2283
2284 hists__fprintf(&c2c.hists.hists, true, 0, 0, 0, stdout, true);
2285
2286 fprintf(out, "\n");
2287 fprintf(out, "=================================================\n");
2288 fprintf(out, " Shared Cache Line Distribution Pareto \n");
2289 fprintf(out, "=================================================\n");
2290 fprintf(out, "#\n");
2291
2292 print_pareto(out);
2293}
2294
2295#ifdef HAVE_SLANG_SUPPORT
2296static void c2c_browser__update_nr_entries(struct hist_browser *hb)
2297{
2298 u64 nr_entries = 0;
2299 struct rb_node *nd = rb_first_cached(&hb->hists->entries);
2300
2301 while (nd) {
2302 struct hist_entry *he = rb_entry(nd, struct hist_entry, rb_node);
2303
2304 if (!he->filtered)
2305 nr_entries++;
2306
2307 nd = rb_next(nd);
2308 }
2309
2310 hb->nr_non_filtered_entries = nr_entries;
2311}
2312
2313struct c2c_cacheline_browser {
2314 struct hist_browser hb;
2315 struct hist_entry *he;
2316};
2317
2318static int
2319perf_c2c_cacheline_browser__title(struct hist_browser *browser,
2320 char *bf, size_t size)
2321{
2322 struct c2c_cacheline_browser *cl_browser;
2323 struct hist_entry *he;
2324 uint64_t addr = 0;
2325
2326 cl_browser = container_of(browser, struct c2c_cacheline_browser, hb);
2327 he = cl_browser->he;
2328
2329 if (he->mem_info)
2330 addr = cl_address(he->mem_info->daddr.addr);
2331
2332 scnprintf(bf, size, "Cacheline 0x%lx", addr);
2333 return 0;
2334}
2335
2336static struct c2c_cacheline_browser*
2337c2c_cacheline_browser__new(struct hists *hists, struct hist_entry *he)
2338{
2339 struct c2c_cacheline_browser *browser;
2340
2341 browser = zalloc(sizeof(*browser));
2342 if (browser) {
2343 hist_browser__init(&browser->hb, hists);
2344 browser->hb.c2c_filter = true;
2345 browser->hb.title = perf_c2c_cacheline_browser__title;
2346 browser->he = he;
2347 }
2348
2349 return browser;
2350}
2351
2352static int perf_c2c__browse_cacheline(struct hist_entry *he)
2353{
2354 struct c2c_hist_entry *c2c_he;
2355 struct c2c_hists *c2c_hists;
2356 struct c2c_cacheline_browser *cl_browser;
2357 struct hist_browser *browser;
2358 int key = -1;
2359 static const char help[] =
2360 " ENTER Toggle callchains (if present) \n"
2361 " n Toggle Node details info \n"
2362 " s Toggle full length of symbol and source line columns \n"
2363 " q Return back to cacheline list \n";
2364
2365 if (!he)
2366 return 0;
2367
2368 /* Display compact version first. */
2369 c2c.symbol_full = false;
2370
2371 c2c_he = container_of(he, struct c2c_hist_entry, he);
2372 c2c_hists = c2c_he->hists;
2373
2374 cl_browser = c2c_cacheline_browser__new(&c2c_hists->hists, he);
2375 if (cl_browser == NULL)
2376 return -1;
2377
2378 browser = &cl_browser->hb;
2379
2380 /* reset abort key so that it can get Ctrl-C as a key */
2381 SLang_reset_tty();
2382 SLang_init_tty(0, 0, 0);
2383
2384 c2c_browser__update_nr_entries(browser);
2385
2386 while (1) {
2387 key = hist_browser__run(browser, "? - help", true);
2388
2389 switch (key) {
2390 case 's':
2391 c2c.symbol_full = !c2c.symbol_full;
2392 break;
2393 case 'n':
2394 c2c.node_info = (c2c.node_info + 1) % 3;
2395 setup_nodes_header();
2396 break;
2397 case 'q':
2398 goto out;
2399 case '?':
2400 ui_browser__help_window(&browser->b, help);
2401 break;
2402 default:
2403 break;
2404 }
2405 }
2406
2407out:
2408 free(cl_browser);
2409 return 0;
2410}
2411
2412static int perf_c2c_browser__title(struct hist_browser *browser,
2413 char *bf, size_t size)
2414{
2415 scnprintf(bf, size,
2416 "Shared Data Cache Line Table "
2417 "(%lu entries, sorted on %s HITMs)",
2418 browser->nr_non_filtered_entries,
2419 display_str[c2c.display]);
2420 return 0;
2421}
2422
2423static struct hist_browser*
2424perf_c2c_browser__new(struct hists *hists)
2425{
2426 struct hist_browser *browser = hist_browser__new(hists);
2427
2428 if (browser) {
2429 browser->title = perf_c2c_browser__title;
2430 browser->c2c_filter = true;
2431 }
2432
2433 return browser;
2434}
2435
2436static int perf_c2c__hists_browse(struct hists *hists)
2437{
2438 struct hist_browser *browser;
2439 int key = -1;
2440 static const char help[] =
2441 " d Display cacheline details \n"
2442 " ENTER Toggle callchains (if present) \n"
2443 " q Quit \n";
2444
2445 browser = perf_c2c_browser__new(hists);
2446 if (browser == NULL)
2447 return -1;
2448
2449 /* reset abort key so that it can get Ctrl-C as a key */
2450 SLang_reset_tty();
2451 SLang_init_tty(0, 0, 0);
2452
2453 c2c_browser__update_nr_entries(browser);
2454
2455 while (1) {
2456 key = hist_browser__run(browser, "? - help", true);
2457
2458 switch (key) {
2459 case 'q':
2460 goto out;
2461 case 'd':
2462 perf_c2c__browse_cacheline(browser->he_selection);
2463 break;
2464 case '?':
2465 ui_browser__help_window(&browser->b, help);
2466 break;
2467 default:
2468 break;
2469 }
2470 }
2471
2472out:
2473 hist_browser__delete(browser);
2474 return 0;
2475}
2476
2477static void perf_c2c_display(struct perf_session *session)
2478{
2479 if (use_browser == 0)
2480 perf_c2c__hists_fprintf(stdout, session);
2481 else
2482 perf_c2c__hists_browse(&c2c.hists.hists);
2483}
2484#else
2485static void perf_c2c_display(struct perf_session *session)
2486{
2487 use_browser = 0;
2488 perf_c2c__hists_fprintf(stdout, session);
2489}
2490#endif /* HAVE_SLANG_SUPPORT */
2491
2492static char *fill_line(const char *orig, int len)
2493{
2494 int i, j, olen = strlen(orig);
2495 char *buf;
2496
2497 buf = zalloc(len + 1);
2498 if (!buf)
2499 return NULL;
2500
2501 j = len / 2 - olen / 2;
2502
2503 for (i = 0; i < j - 1; i++)
2504 buf[i] = '-';
2505
2506 buf[i++] = ' ';
2507
2508 strcpy(buf + i, orig);
2509
2510 i += olen;
2511
2512 buf[i++] = ' ';
2513
2514 for (; i < len; i++)
2515 buf[i] = '-';
2516
2517 return buf;
2518}
2519
2520static int ui_quirks(void)
2521{
2522 const char *nodestr = "Data address";
2523 char *buf;
2524
2525 if (!c2c.use_stdio) {
2526 dim_offset.width = 5;
2527 dim_offset.header = header_offset_tui;
2528 nodestr = "CL";
2529 }
2530
2531 dim_percent_hitm.header = percent_hitm_header[c2c.display];
2532
2533 /* Fix the zero line for dcacheline column. */
2534 buf = fill_line("Cacheline", dim_dcacheline.width +
2535 dim_dcacheline_node.width +
2536 dim_dcacheline_count.width + 4);
2537 if (!buf)
2538 return -ENOMEM;
2539
2540 dim_dcacheline.header.line[0].text = buf;
2541
2542 /* Fix the zero line for offset column. */
2543 buf = fill_line(nodestr, dim_offset.width +
2544 dim_offset_node.width +
2545 dim_dcacheline_count.width + 4);
2546 if (!buf)
2547 return -ENOMEM;
2548
2549 dim_offset.header.line[0].text = buf;
2550
2551 return 0;
2552}
2553
2554#define CALLCHAIN_DEFAULT_OPT "graph,0.5,caller,function,percent"
2555
2556const char callchain_help[] = "Display call graph (stack chain/backtrace):\n\n"
2557 CALLCHAIN_REPORT_HELP
2558 "\n\t\t\t\tDefault: " CALLCHAIN_DEFAULT_OPT;
2559
2560static int
2561parse_callchain_opt(const struct option *opt, const char *arg, int unset)
2562{
2563 struct callchain_param *callchain = opt->value;
2564
2565 callchain->enabled = !unset;
2566 /*
2567 * --no-call-graph
2568 */
2569 if (unset) {
2570 symbol_conf.use_callchain = false;
2571 callchain->mode = CHAIN_NONE;
2572 return 0;
2573 }
2574
2575 return parse_callchain_report_opt(arg);
2576}
2577
2578static int setup_callchain(struct evlist *evlist)
2579{
2580 u64 sample_type = perf_evlist__combined_sample_type(evlist);
2581 enum perf_call_graph_mode mode = CALLCHAIN_NONE;
2582
2583 if ((sample_type & PERF_SAMPLE_REGS_USER) &&
2584 (sample_type & PERF_SAMPLE_STACK_USER)) {
2585 mode = CALLCHAIN_DWARF;
2586 dwarf_callchain_users = true;
2587 } else if (sample_type & PERF_SAMPLE_BRANCH_STACK)
2588 mode = CALLCHAIN_LBR;
2589 else if (sample_type & PERF_SAMPLE_CALLCHAIN)
2590 mode = CALLCHAIN_FP;
2591
2592 if (!callchain_param.enabled &&
2593 callchain_param.mode != CHAIN_NONE &&
2594 mode != CALLCHAIN_NONE) {
2595 symbol_conf.use_callchain = true;
2596 if (callchain_register_param(&callchain_param) < 0) {
2597 ui__error("Can't register callchain params.\n");
2598 return -EINVAL;
2599 }
2600 }
2601
2602 callchain_param.record_mode = mode;
2603 callchain_param.min_percent = 0;
2604 return 0;
2605}
2606
2607static int setup_display(const char *str)
2608{
2609 const char *display = str ?: "tot";
2610
2611 if (!strcmp(display, "tot"))
2612 c2c.display = DISPLAY_TOT;
2613 else if (!strcmp(display, "rmt"))
2614 c2c.display = DISPLAY_RMT;
2615 else if (!strcmp(display, "lcl"))
2616 c2c.display = DISPLAY_LCL;
2617 else {
2618 pr_err("failed: unknown display type: %s\n", str);
2619 return -1;
2620 }
2621
2622 return 0;
2623}
2624
2625#define for_each_token(__tok, __buf, __sep, __tmp) \
2626 for (__tok = strtok_r(__buf, __sep, &__tmp); __tok; \
2627 __tok = strtok_r(NULL, __sep, &__tmp))
2628
2629static int build_cl_output(char *cl_sort, bool no_source)
2630{
2631 char *tok, *tmp, *buf = strdup(cl_sort);
2632 bool add_pid = false;
2633 bool add_tid = false;
2634 bool add_iaddr = false;
2635 bool add_sym = false;
2636 bool add_dso = false;
2637 bool add_src = false;
2638 int ret = 0;
2639
2640 if (!buf)
2641 return -ENOMEM;
2642
2643 for_each_token(tok, buf, ",", tmp) {
2644 if (!strcmp(tok, "tid")) {
2645 add_tid = true;
2646 } else if (!strcmp(tok, "pid")) {
2647 add_pid = true;
2648 } else if (!strcmp(tok, "iaddr")) {
2649 add_iaddr = true;
2650 add_sym = true;
2651 add_dso = true;
2652 add_src = no_source ? false : true;
2653 } else if (!strcmp(tok, "dso")) {
2654 add_dso = true;
2655 } else if (strcmp(tok, "offset")) {
2656 pr_err("unrecognized sort token: %s\n", tok);
2657 ret = -EINVAL;
2658 goto err;
2659 }
2660 }
2661
2662 if (asprintf(&c2c.cl_output,
2663 "%s%s%s%s%s%s%s%s%s%s",
2664 c2c.use_stdio ? "cl_num_empty," : "",
2665 "percent_rmt_hitm,"
2666 "percent_lcl_hitm,"
2667 "percent_stores_l1hit,"
2668 "percent_stores_l1miss,"
2669 "offset,offset_node,dcacheline_count,",
2670 add_pid ? "pid," : "",
2671 add_tid ? "tid," : "",
2672 add_iaddr ? "iaddr," : "",
2673 "mean_rmt,"
2674 "mean_lcl,"
2675 "mean_load,"
2676 "tot_recs,"
2677 "cpucnt,",
2678 add_sym ? "symbol," : "",
2679 add_dso ? "dso," : "",
2680 add_src ? "cl_srcline," : "",
2681 "node") < 0) {
2682 ret = -ENOMEM;
2683 goto err;
2684 }
2685
2686 c2c.show_src = add_src;
2687err:
2688 free(buf);
2689 return ret;
2690}
2691
2692static int setup_coalesce(const char *coalesce, bool no_source)
2693{
2694 const char *c = coalesce ?: coalesce_default;
2695
2696 if (asprintf(&c2c.cl_sort, "offset,%s", c) < 0)
2697 return -ENOMEM;
2698
2699 if (build_cl_output(c2c.cl_sort, no_source))
2700 return -1;
2701
2702 if (asprintf(&c2c.cl_resort, "offset,%s",
2703 c2c.display == DISPLAY_TOT ?
2704 "tot_hitm" :
2705 c2c.display == DISPLAY_RMT ?
2706 "rmt_hitm,lcl_hitm" :
2707 "lcl_hitm,rmt_hitm") < 0)
2708 return -ENOMEM;
2709
2710 pr_debug("coalesce sort fields: %s\n", c2c.cl_sort);
2711 pr_debug("coalesce resort fields: %s\n", c2c.cl_resort);
2712 pr_debug("coalesce output fields: %s\n", c2c.cl_output);
2713 return 0;
2714}
2715
2716static int perf_c2c__report(int argc, const char **argv)
2717{
2718 struct perf_session *session;
2719 struct ui_progress prog;
2720 struct perf_data data = {
2721 .mode = PERF_DATA_MODE_READ,
2722 };
2723 char callchain_default_opt[] = CALLCHAIN_DEFAULT_OPT;
2724 const char *display = NULL;
2725 const char *coalesce = NULL;
2726 bool no_source = false;
2727 const struct option options[] = {
2728 OPT_STRING('k', "vmlinux", &symbol_conf.vmlinux_name,
2729 "file", "vmlinux pathname"),
2730 OPT_STRING('i', "input", &input_name, "file",
2731 "the input file to process"),
2732 OPT_INCR('N', "node-info", &c2c.node_info,
2733 "show extra node info in report (repeat for more info)"),
2734#ifdef HAVE_SLANG_SUPPORT
2735 OPT_BOOLEAN(0, "stdio", &c2c.use_stdio, "Use the stdio interface"),
2736#endif
2737 OPT_BOOLEAN(0, "stats", &c2c.stats_only,
2738 "Display only statistic tables (implies --stdio)"),
2739 OPT_BOOLEAN(0, "full-symbols", &c2c.symbol_full,
2740 "Display full length of symbols"),
2741 OPT_BOOLEAN(0, "no-source", &no_source,
2742 "Do not display Source Line column"),
2743 OPT_BOOLEAN(0, "show-all", &c2c.show_all,
2744 "Show all captured HITM lines."),
2745 OPT_CALLBACK_DEFAULT('g', "call-graph", &callchain_param,
2746 "print_type,threshold[,print_limit],order,sort_key[,branch],value",
2747 callchain_help, &parse_callchain_opt,
2748 callchain_default_opt),
2749 OPT_STRING('d', "display", &display, "Switch HITM output type", "lcl,rmt"),
2750 OPT_STRING('c', "coalesce", &coalesce, "coalesce fields",
2751 "coalesce fields: pid,tid,iaddr,dso"),
2752 OPT_BOOLEAN('f', "force", &symbol_conf.force, "don't complain, do it"),
2753 OPT_PARENT(c2c_options),
2754 OPT_END()
2755 };
2756 int err = 0;
2757
2758 argc = parse_options(argc, argv, options, report_c2c_usage,
2759 PARSE_OPT_STOP_AT_NON_OPTION);
2760 if (argc)
2761 usage_with_options(report_c2c_usage, options);
2762
2763 if (c2c.stats_only)
2764 c2c.use_stdio = true;
2765
2766 if (!input_name || !strlen(input_name))
2767 input_name = "perf.data";
2768
2769 data.path = input_name;
2770 data.force = symbol_conf.force;
2771
2772 err = setup_display(display);
2773 if (err)
2774 goto out;
2775
2776 err = setup_coalesce(coalesce, no_source);
2777 if (err) {
2778 pr_debug("Failed to initialize hists\n");
2779 goto out;
2780 }
2781
2782 err = c2c_hists__init(&c2c.hists, "dcacheline", 2);
2783 if (err) {
2784 pr_debug("Failed to initialize hists\n");
2785 goto out;
2786 }
2787
2788 session = perf_session__new(&data, 0, &c2c.tool);
2789 if (IS_ERR(session)) {
2790 err = PTR_ERR(session);
2791 pr_debug("Error creating perf session\n");
2792 goto out;
2793 }
2794
2795 err = setup_nodes(session);
2796 if (err) {
2797 pr_err("Failed setup nodes\n");
2798 goto out;
2799 }
2800
2801 err = mem2node__init(&c2c.mem2node, &session->header.env);
2802 if (err)
2803 goto out_session;
2804
2805 err = setup_callchain(session->evlist);
2806 if (err)
2807 goto out_mem2node;
2808
2809 if (symbol__init(&session->header.env) < 0)
2810 goto out_mem2node;
2811
2812 /* No pipe support at the moment. */
2813 if (perf_data__is_pipe(session->data)) {
2814 pr_debug("No pipe support at the moment.\n");
2815 goto out_mem2node;
2816 }
2817
2818 if (c2c.use_stdio)
2819 use_browser = 0;
2820 else
2821 use_browser = 1;
2822
2823 setup_browser(false);
2824
2825 err = perf_session__process_events(session);
2826 if (err) {
2827 pr_err("failed to process sample\n");
2828 goto out_mem2node;
2829 }
2830
2831 c2c_hists__reinit(&c2c.hists,
2832 "cl_idx,"
2833 "dcacheline,"
2834 "dcacheline_node,"
2835 "dcacheline_count,"
2836 "tot_recs,"
2837 "percent_hitm,"
2838 "tot_hitm,lcl_hitm,rmt_hitm,"
2839 "stores,stores_l1hit,stores_l1miss,"
2840 "dram_lcl,dram_rmt,"
2841 "ld_llcmiss,"
2842 "tot_loads,"
2843 "ld_fbhit,ld_l1hit,ld_l2hit,"
2844 "ld_lclhit,ld_rmthit",
2845 c2c.display == DISPLAY_TOT ? "tot_hitm" :
2846 c2c.display == DISPLAY_LCL ? "lcl_hitm" : "rmt_hitm"
2847 );
2848
2849 ui_progress__init(&prog, c2c.hists.hists.nr_entries, "Sorting...");
2850
2851 hists__collapse_resort(&c2c.hists.hists, NULL);
2852 hists__output_resort_cb(&c2c.hists.hists, &prog, resort_hitm_cb);
2853 hists__iterate_cb(&c2c.hists.hists, resort_cl_cb);
2854
2855 ui_progress__finish();
2856
2857 if (ui_quirks()) {
2858 pr_err("failed to setup UI\n");
2859 goto out_mem2node;
2860 }
2861
2862 perf_c2c_display(session);
2863
2864out_mem2node:
2865 mem2node__exit(&c2c.mem2node);
2866out_session:
2867 perf_session__delete(session);
2868out:
2869 return err;
2870}
2871
2872static int parse_record_events(const struct option *opt,
2873 const char *str, int unset __maybe_unused)
2874{
2875 bool *event_set = (bool *) opt->value;
2876
2877 *event_set = true;
2878 return perf_mem_events__parse(str);
2879}
2880
2881
2882static const char * const __usage_record[] = {
2883 "perf c2c record [<options>] [<command>]",
2884 "perf c2c record [<options>] -- <command> [<options>]",
2885 NULL
2886};
2887
2888static const char * const *record_mem_usage = __usage_record;
2889
2890static int perf_c2c__record(int argc, const char **argv)
2891{
2892 int rec_argc, i = 0, j;
2893 const char **rec_argv;
2894 int ret;
2895 bool all_user = false, all_kernel = false;
2896 bool event_set = false;
2897 struct option options[] = {
2898 OPT_CALLBACK('e', "event", &event_set, "event",
2899 "event selector. Use 'perf mem record -e list' to list available events",
2900 parse_record_events),
2901 OPT_BOOLEAN('u', "all-user", &all_user, "collect only user level data"),
2902 OPT_BOOLEAN('k', "all-kernel", &all_kernel, "collect only kernel level data"),
2903 OPT_UINTEGER('l', "ldlat", &perf_mem_events__loads_ldlat, "setup mem-loads latency"),
2904 OPT_PARENT(c2c_options),
2905 OPT_END()
2906 };
2907
2908 if (perf_mem_events__init()) {
2909 pr_err("failed: memory events not supported\n");
2910 return -1;
2911 }
2912
2913 argc = parse_options(argc, argv, options, record_mem_usage,
2914 PARSE_OPT_KEEP_UNKNOWN);
2915
2916 rec_argc = argc + 11; /* max number of arguments */
2917 rec_argv = calloc(rec_argc + 1, sizeof(char *));
2918 if (!rec_argv)
2919 return -1;
2920
2921 rec_argv[i++] = "record";
2922
2923 if (!event_set) {
2924 perf_mem_events[PERF_MEM_EVENTS__LOAD].record = true;
2925 perf_mem_events[PERF_MEM_EVENTS__STORE].record = true;
2926 }
2927
2928 if (perf_mem_events[PERF_MEM_EVENTS__LOAD].record)
2929 rec_argv[i++] = "-W";
2930
2931 rec_argv[i++] = "-d";
2932 rec_argv[i++] = "--phys-data";
2933 rec_argv[i++] = "--sample-cpu";
2934
2935 for (j = 0; j < PERF_MEM_EVENTS__MAX; j++) {
2936 if (!perf_mem_events[j].record)
2937 continue;
2938
2939 if (!perf_mem_events[j].supported) {
2940 pr_err("failed: event '%s' not supported\n",
2941 perf_mem_events[j].name);
2942 free(rec_argv);
2943 return -1;
2944 }
2945
2946 rec_argv[i++] = "-e";
2947 rec_argv[i++] = perf_mem_events__name(j);
2948 };
2949
2950 if (all_user)
2951 rec_argv[i++] = "--all-user";
2952
2953 if (all_kernel)
2954 rec_argv[i++] = "--all-kernel";
2955
2956 for (j = 0; j < argc; j++, i++)
2957 rec_argv[i] = argv[j];
2958
2959 if (verbose > 0) {
2960 pr_debug("calling: ");
2961
2962 j = 0;
2963
2964 while (rec_argv[j]) {
2965 pr_debug("%s ", rec_argv[j]);
2966 j++;
2967 }
2968 pr_debug("\n");
2969 }
2970
2971 ret = cmd_record(i, rec_argv);
2972 free(rec_argv);
2973 return ret;
2974}
2975
2976int cmd_c2c(int argc, const char **argv)
2977{
2978 argc = parse_options(argc, argv, c2c_options, c2c_usage,
2979 PARSE_OPT_STOP_AT_NON_OPTION);
2980
2981 if (!argc)
2982 usage_with_options(c2c_usage, c2c_options);
2983
2984 if (!strncmp(argv[0], "rec", 3)) {
2985 return perf_c2c__record(argc, argv);
2986 } else if (!strncmp(argv[0], "rep", 3)) {
2987 return perf_c2c__report(argc, argv);
2988 } else {
2989 usage_with_options(c2c_usage, c2c_options);
2990 }
2991
2992 return 0;
2993}
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * This is rewrite of original c2c tool introduced in here:
4 * http://lwn.net/Articles/588866/
5 *
6 * The original tool was changed to fit in current perf state.
7 *
8 * Original authors:
9 * Don Zickus <dzickus@redhat.com>
10 * Dick Fowles <fowles@inreach.com>
11 * Joe Mario <jmario@redhat.com>
12 */
13#include <errno.h>
14#include <inttypes.h>
15#include <linux/compiler.h>
16#include <linux/err.h>
17#include <linux/kernel.h>
18#include <linux/stringify.h>
19#include <linux/zalloc.h>
20#include <asm/bug.h>
21#include <sys/param.h>
22#include "debug.h"
23#include "builtin.h"
24#include <perf/cpumap.h>
25#include <subcmd/pager.h>
26#include <subcmd/parse-options.h>
27#include "map_symbol.h"
28#include "mem-events.h"
29#include "session.h"
30#include "hist.h"
31#include "sort.h"
32#include "tool.h"
33#include "cacheline.h"
34#include "data.h"
35#include "event.h"
36#include "evlist.h"
37#include "evsel.h"
38#include "ui/browsers/hists.h"
39#include "thread.h"
40#include "mem2node.h"
41#include "symbol.h"
42#include "ui/ui.h"
43#include "ui/progress.h"
44#include "../perf.h"
45
46struct c2c_hists {
47 struct hists hists;
48 struct perf_hpp_list list;
49 struct c2c_stats stats;
50};
51
52struct compute_stats {
53 struct stats lcl_hitm;
54 struct stats rmt_hitm;
55 struct stats load;
56};
57
58struct c2c_hist_entry {
59 struct c2c_hists *hists;
60 struct c2c_stats stats;
61 unsigned long *cpuset;
62 unsigned long *nodeset;
63 struct c2c_stats *node_stats;
64 unsigned int cacheline_idx;
65
66 struct compute_stats cstats;
67
68 unsigned long paddr;
69 unsigned long paddr_cnt;
70 bool paddr_zero;
71 char *nodestr;
72
73 /*
74 * must be at the end,
75 * because of its callchain dynamic entry
76 */
77 struct hist_entry he;
78};
79
80static char const *coalesce_default = "iaddr";
81
82struct perf_c2c {
83 struct perf_tool tool;
84 struct c2c_hists hists;
85 struct mem2node mem2node;
86
87 unsigned long **nodes;
88 int nodes_cnt;
89 int cpus_cnt;
90 int *cpu2node;
91 int node_info;
92
93 bool show_src;
94 bool show_all;
95 bool use_stdio;
96 bool stats_only;
97 bool symbol_full;
98 bool stitch_lbr;
99
100 /* HITM shared clines stats */
101 struct c2c_stats hitm_stats;
102 int shared_clines;
103
104 int display;
105
106 const char *coalesce;
107 char *cl_sort;
108 char *cl_resort;
109 char *cl_output;
110};
111
112enum {
113 DISPLAY_LCL,
114 DISPLAY_RMT,
115 DISPLAY_TOT,
116 DISPLAY_MAX,
117};
118
119static const char *display_str[DISPLAY_MAX] = {
120 [DISPLAY_LCL] = "Local",
121 [DISPLAY_RMT] = "Remote",
122 [DISPLAY_TOT] = "Total",
123};
124
125static const struct option c2c_options[] = {
126 OPT_INCR('v', "verbose", &verbose, "be more verbose (show counter open errors, etc)"),
127 OPT_END()
128};
129
130static struct perf_c2c c2c;
131
132static void *c2c_he_zalloc(size_t size)
133{
134 struct c2c_hist_entry *c2c_he;
135
136 c2c_he = zalloc(size + sizeof(*c2c_he));
137 if (!c2c_he)
138 return NULL;
139
140 c2c_he->cpuset = bitmap_alloc(c2c.cpus_cnt);
141 if (!c2c_he->cpuset)
142 return NULL;
143
144 c2c_he->nodeset = bitmap_alloc(c2c.nodes_cnt);
145 if (!c2c_he->nodeset)
146 return NULL;
147
148 c2c_he->node_stats = zalloc(c2c.nodes_cnt * sizeof(*c2c_he->node_stats));
149 if (!c2c_he->node_stats)
150 return NULL;
151
152 init_stats(&c2c_he->cstats.lcl_hitm);
153 init_stats(&c2c_he->cstats.rmt_hitm);
154 init_stats(&c2c_he->cstats.load);
155
156 return &c2c_he->he;
157}
158
159static void c2c_he_free(void *he)
160{
161 struct c2c_hist_entry *c2c_he;
162
163 c2c_he = container_of(he, struct c2c_hist_entry, he);
164 if (c2c_he->hists) {
165 hists__delete_entries(&c2c_he->hists->hists);
166 free(c2c_he->hists);
167 }
168
169 free(c2c_he->cpuset);
170 free(c2c_he->nodeset);
171 free(c2c_he->nodestr);
172 free(c2c_he->node_stats);
173 free(c2c_he);
174}
175
176static struct hist_entry_ops c2c_entry_ops = {
177 .new = c2c_he_zalloc,
178 .free = c2c_he_free,
179};
180
181static int c2c_hists__init(struct c2c_hists *hists,
182 const char *sort,
183 int nr_header_lines);
184
185static struct c2c_hists*
186he__get_c2c_hists(struct hist_entry *he,
187 const char *sort,
188 int nr_header_lines)
189{
190 struct c2c_hist_entry *c2c_he;
191 struct c2c_hists *hists;
192 int ret;
193
194 c2c_he = container_of(he, struct c2c_hist_entry, he);
195 if (c2c_he->hists)
196 return c2c_he->hists;
197
198 hists = c2c_he->hists = zalloc(sizeof(*hists));
199 if (!hists)
200 return NULL;
201
202 ret = c2c_hists__init(hists, sort, nr_header_lines);
203 if (ret) {
204 free(hists);
205 return NULL;
206 }
207
208 return hists;
209}
210
211static void c2c_he__set_cpu(struct c2c_hist_entry *c2c_he,
212 struct perf_sample *sample)
213{
214 if (WARN_ONCE(sample->cpu == (unsigned int) -1,
215 "WARNING: no sample cpu value"))
216 return;
217
218 set_bit(sample->cpu, c2c_he->cpuset);
219}
220
221static void c2c_he__set_node(struct c2c_hist_entry *c2c_he,
222 struct perf_sample *sample)
223{
224 int node;
225
226 if (!sample->phys_addr) {
227 c2c_he->paddr_zero = true;
228 return;
229 }
230
231 node = mem2node__node(&c2c.mem2node, sample->phys_addr);
232 if (WARN_ONCE(node < 0, "WARNING: failed to find node\n"))
233 return;
234
235 set_bit(node, c2c_he->nodeset);
236
237 if (c2c_he->paddr != sample->phys_addr) {
238 c2c_he->paddr_cnt++;
239 c2c_he->paddr = sample->phys_addr;
240 }
241}
242
243static void compute_stats(struct c2c_hist_entry *c2c_he,
244 struct c2c_stats *stats,
245 u64 weight)
246{
247 struct compute_stats *cstats = &c2c_he->cstats;
248
249 if (stats->rmt_hitm)
250 update_stats(&cstats->rmt_hitm, weight);
251 else if (stats->lcl_hitm)
252 update_stats(&cstats->lcl_hitm, weight);
253 else if (stats->load)
254 update_stats(&cstats->load, weight);
255}
256
257static int process_sample_event(struct perf_tool *tool __maybe_unused,
258 union perf_event *event,
259 struct perf_sample *sample,
260 struct evsel *evsel,
261 struct machine *machine)
262{
263 struct c2c_hists *c2c_hists = &c2c.hists;
264 struct c2c_hist_entry *c2c_he;
265 struct c2c_stats stats = { .nr_entries = 0, };
266 struct hist_entry *he;
267 struct addr_location al;
268 struct mem_info *mi, *mi_dup;
269 int ret;
270
271 if (machine__resolve(machine, &al, sample) < 0) {
272 pr_debug("problem processing %d event, skipping it.\n",
273 event->header.type);
274 return -1;
275 }
276
277 if (c2c.stitch_lbr)
278 al.thread->lbr_stitch_enable = true;
279
280 ret = sample__resolve_callchain(sample, &callchain_cursor, NULL,
281 evsel, &al, sysctl_perf_event_max_stack);
282 if (ret)
283 goto out;
284
285 mi = sample__resolve_mem(sample, &al);
286 if (mi == NULL)
287 return -ENOMEM;
288
289 /*
290 * The mi object is released in hists__add_entry_ops,
291 * if it gets sorted out into existing data, so we need
292 * to take the copy now.
293 */
294 mi_dup = mem_info__get(mi);
295
296 c2c_decode_stats(&stats, mi);
297
298 he = hists__add_entry_ops(&c2c_hists->hists, &c2c_entry_ops,
299 &al, NULL, NULL, mi,
300 sample, true);
301 if (he == NULL)
302 goto free_mi;
303
304 c2c_he = container_of(he, struct c2c_hist_entry, he);
305 c2c_add_stats(&c2c_he->stats, &stats);
306 c2c_add_stats(&c2c_hists->stats, &stats);
307
308 c2c_he__set_cpu(c2c_he, sample);
309 c2c_he__set_node(c2c_he, sample);
310
311 hists__inc_nr_samples(&c2c_hists->hists, he->filtered);
312 ret = hist_entry__append_callchain(he, sample);
313
314 if (!ret) {
315 /*
316 * There's already been warning about missing
317 * sample's cpu value. Let's account all to
318 * node 0 in this case, without any further
319 * warning.
320 *
321 * Doing node stats only for single callchain data.
322 */
323 int cpu = sample->cpu == (unsigned int) -1 ? 0 : sample->cpu;
324 int node = c2c.cpu2node[cpu];
325
326 mi = mi_dup;
327
328 c2c_hists = he__get_c2c_hists(he, c2c.cl_sort, 2);
329 if (!c2c_hists)
330 goto free_mi;
331
332 he = hists__add_entry_ops(&c2c_hists->hists, &c2c_entry_ops,
333 &al, NULL, NULL, mi,
334 sample, true);
335 if (he == NULL)
336 goto free_mi;
337
338 c2c_he = container_of(he, struct c2c_hist_entry, he);
339 c2c_add_stats(&c2c_he->stats, &stats);
340 c2c_add_stats(&c2c_hists->stats, &stats);
341 c2c_add_stats(&c2c_he->node_stats[node], &stats);
342
343 compute_stats(c2c_he, &stats, sample->weight);
344
345 c2c_he__set_cpu(c2c_he, sample);
346 c2c_he__set_node(c2c_he, sample);
347
348 hists__inc_nr_samples(&c2c_hists->hists, he->filtered);
349 ret = hist_entry__append_callchain(he, sample);
350 }
351
352out:
353 addr_location__put(&al);
354 return ret;
355
356free_mi:
357 mem_info__put(mi_dup);
358 mem_info__put(mi);
359 ret = -ENOMEM;
360 goto out;
361}
362
363static struct perf_c2c c2c = {
364 .tool = {
365 .sample = process_sample_event,
366 .mmap = perf_event__process_mmap,
367 .mmap2 = perf_event__process_mmap2,
368 .comm = perf_event__process_comm,
369 .exit = perf_event__process_exit,
370 .fork = perf_event__process_fork,
371 .lost = perf_event__process_lost,
372 .ordered_events = true,
373 .ordering_requires_timestamps = true,
374 },
375};
376
377static const char * const c2c_usage[] = {
378 "perf c2c {record|report}",
379 NULL
380};
381
382static const char * const __usage_report[] = {
383 "perf c2c report",
384 NULL
385};
386
387static const char * const *report_c2c_usage = __usage_report;
388
389#define C2C_HEADER_MAX 2
390
391struct c2c_header {
392 struct {
393 const char *text;
394 int span;
395 } line[C2C_HEADER_MAX];
396};
397
398struct c2c_dimension {
399 struct c2c_header header;
400 const char *name;
401 int width;
402 struct sort_entry *se;
403
404 int64_t (*cmp)(struct perf_hpp_fmt *fmt,
405 struct hist_entry *, struct hist_entry *);
406 int (*entry)(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
407 struct hist_entry *he);
408 int (*color)(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
409 struct hist_entry *he);
410};
411
412struct c2c_fmt {
413 struct perf_hpp_fmt fmt;
414 struct c2c_dimension *dim;
415};
416
417#define SYMBOL_WIDTH 30
418
419static struct c2c_dimension dim_symbol;
420static struct c2c_dimension dim_srcline;
421
422static int symbol_width(struct hists *hists, struct sort_entry *se)
423{
424 int width = hists__col_len(hists, se->se_width_idx);
425
426 if (!c2c.symbol_full)
427 width = MIN(width, SYMBOL_WIDTH);
428
429 return width;
430}
431
432static int c2c_width(struct perf_hpp_fmt *fmt,
433 struct perf_hpp *hpp __maybe_unused,
434 struct hists *hists)
435{
436 struct c2c_fmt *c2c_fmt;
437 struct c2c_dimension *dim;
438
439 c2c_fmt = container_of(fmt, struct c2c_fmt, fmt);
440 dim = c2c_fmt->dim;
441
442 if (dim == &dim_symbol || dim == &dim_srcline)
443 return symbol_width(hists, dim->se);
444
445 return dim->se ? hists__col_len(hists, dim->se->se_width_idx) :
446 c2c_fmt->dim->width;
447}
448
449static int c2c_header(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
450 struct hists *hists, int line, int *span)
451{
452 struct perf_hpp_list *hpp_list = hists->hpp_list;
453 struct c2c_fmt *c2c_fmt;
454 struct c2c_dimension *dim;
455 const char *text = NULL;
456 int width = c2c_width(fmt, hpp, hists);
457
458 c2c_fmt = container_of(fmt, struct c2c_fmt, fmt);
459 dim = c2c_fmt->dim;
460
461 if (dim->se) {
462 text = dim->header.line[line].text;
463 /* Use the last line from sort_entry if not defined. */
464 if (!text && (line == hpp_list->nr_header_lines - 1))
465 text = dim->se->se_header;
466 } else {
467 text = dim->header.line[line].text;
468
469 if (*span) {
470 (*span)--;
471 return 0;
472 } else {
473 *span = dim->header.line[line].span;
474 }
475 }
476
477 if (text == NULL)
478 text = "";
479
480 return scnprintf(hpp->buf, hpp->size, "%*s", width, text);
481}
482
483#define HEX_STR(__s, __v) \
484({ \
485 scnprintf(__s, sizeof(__s), "0x%" PRIx64, __v); \
486 __s; \
487})
488
489static int64_t
490dcacheline_cmp(struct perf_hpp_fmt *fmt __maybe_unused,
491 struct hist_entry *left, struct hist_entry *right)
492{
493 return sort__dcacheline_cmp(left, right);
494}
495
496static int dcacheline_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
497 struct hist_entry *he)
498{
499 uint64_t addr = 0;
500 int width = c2c_width(fmt, hpp, he->hists);
501 char buf[20];
502
503 if (he->mem_info)
504 addr = cl_address(he->mem_info->daddr.addr);
505
506 return scnprintf(hpp->buf, hpp->size, "%*s", width, HEX_STR(buf, addr));
507}
508
509static int
510dcacheline_node_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
511 struct hist_entry *he)
512{
513 struct c2c_hist_entry *c2c_he;
514 int width = c2c_width(fmt, hpp, he->hists);
515
516 c2c_he = container_of(he, struct c2c_hist_entry, he);
517 if (WARN_ON_ONCE(!c2c_he->nodestr))
518 return 0;
519
520 return scnprintf(hpp->buf, hpp->size, "%*s", width, c2c_he->nodestr);
521}
522
523static int
524dcacheline_node_count(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
525 struct hist_entry *he)
526{
527 struct c2c_hist_entry *c2c_he;
528 int width = c2c_width(fmt, hpp, he->hists);
529
530 c2c_he = container_of(he, struct c2c_hist_entry, he);
531 return scnprintf(hpp->buf, hpp->size, "%*lu", width, c2c_he->paddr_cnt);
532}
533
534static int offset_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
535 struct hist_entry *he)
536{
537 uint64_t addr = 0;
538 int width = c2c_width(fmt, hpp, he->hists);
539 char buf[20];
540
541 if (he->mem_info)
542 addr = cl_offset(he->mem_info->daddr.al_addr);
543
544 return scnprintf(hpp->buf, hpp->size, "%*s", width, HEX_STR(buf, addr));
545}
546
547static int64_t
548offset_cmp(struct perf_hpp_fmt *fmt __maybe_unused,
549 struct hist_entry *left, struct hist_entry *right)
550{
551 uint64_t l = 0, r = 0;
552
553 if (left->mem_info)
554 l = cl_offset(left->mem_info->daddr.addr);
555 if (right->mem_info)
556 r = cl_offset(right->mem_info->daddr.addr);
557
558 return (int64_t)(r - l);
559}
560
561static int
562iaddr_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 = he->mem_info->iaddr.addr;
571
572 return scnprintf(hpp->buf, hpp->size, "%*s", width, HEX_STR(buf, addr));
573}
574
575static int64_t
576iaddr_cmp(struct perf_hpp_fmt *fmt __maybe_unused,
577 struct hist_entry *left, struct hist_entry *right)
578{
579 return sort__iaddr_cmp(left, right);
580}
581
582static int
583tot_hitm_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
584 struct hist_entry *he)
585{
586 struct c2c_hist_entry *c2c_he;
587 int width = c2c_width(fmt, hpp, he->hists);
588 unsigned int tot_hitm;
589
590 c2c_he = container_of(he, struct c2c_hist_entry, he);
591 tot_hitm = c2c_he->stats.lcl_hitm + c2c_he->stats.rmt_hitm;
592
593 return scnprintf(hpp->buf, hpp->size, "%*u", width, tot_hitm);
594}
595
596static int64_t
597tot_hitm_cmp(struct perf_hpp_fmt *fmt __maybe_unused,
598 struct hist_entry *left, struct hist_entry *right)
599{
600 struct c2c_hist_entry *c2c_left;
601 struct c2c_hist_entry *c2c_right;
602 uint64_t tot_hitm_left;
603 uint64_t tot_hitm_right;
604
605 c2c_left = container_of(left, struct c2c_hist_entry, he);
606 c2c_right = container_of(right, struct c2c_hist_entry, he);
607
608 tot_hitm_left = c2c_left->stats.lcl_hitm + c2c_left->stats.rmt_hitm;
609 tot_hitm_right = c2c_right->stats.lcl_hitm + c2c_right->stats.rmt_hitm;
610
611 return tot_hitm_left - tot_hitm_right;
612}
613
614#define STAT_FN_ENTRY(__f) \
615static int \
616__f ## _entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp, \
617 struct hist_entry *he) \
618{ \
619 struct c2c_hist_entry *c2c_he; \
620 int width = c2c_width(fmt, hpp, he->hists); \
621 \
622 c2c_he = container_of(he, struct c2c_hist_entry, he); \
623 return scnprintf(hpp->buf, hpp->size, "%*u", width, \
624 c2c_he->stats.__f); \
625}
626
627#define STAT_FN_CMP(__f) \
628static int64_t \
629__f ## _cmp(struct perf_hpp_fmt *fmt __maybe_unused, \
630 struct hist_entry *left, struct hist_entry *right) \
631{ \
632 struct c2c_hist_entry *c2c_left, *c2c_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 return (uint64_t) c2c_left->stats.__f - \
637 (uint64_t) c2c_right->stats.__f; \
638}
639
640#define STAT_FN(__f) \
641 STAT_FN_ENTRY(__f) \
642 STAT_FN_CMP(__f)
643
644STAT_FN(rmt_hitm)
645STAT_FN(lcl_hitm)
646STAT_FN(store)
647STAT_FN(st_l1hit)
648STAT_FN(st_l1miss)
649STAT_FN(ld_fbhit)
650STAT_FN(ld_l1hit)
651STAT_FN(ld_l2hit)
652STAT_FN(ld_llchit)
653STAT_FN(rmt_hit)
654
655static uint64_t llc_miss(struct c2c_stats *stats)
656{
657 uint64_t llcmiss;
658
659 llcmiss = stats->lcl_dram +
660 stats->rmt_dram +
661 stats->rmt_hitm +
662 stats->rmt_hit;
663
664 return llcmiss;
665}
666
667static int
668ld_llcmiss_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
669 struct hist_entry *he)
670{
671 struct c2c_hist_entry *c2c_he;
672 int width = c2c_width(fmt, hpp, he->hists);
673
674 c2c_he = container_of(he, struct c2c_hist_entry, he);
675
676 return scnprintf(hpp->buf, hpp->size, "%*lu", width,
677 llc_miss(&c2c_he->stats));
678}
679
680static int64_t
681ld_llcmiss_cmp(struct perf_hpp_fmt *fmt __maybe_unused,
682 struct hist_entry *left, struct hist_entry *right)
683{
684 struct c2c_hist_entry *c2c_left;
685 struct c2c_hist_entry *c2c_right;
686
687 c2c_left = container_of(left, struct c2c_hist_entry, he);
688 c2c_right = container_of(right, struct c2c_hist_entry, he);
689
690 return (uint64_t) llc_miss(&c2c_left->stats) -
691 (uint64_t) llc_miss(&c2c_right->stats);
692}
693
694static uint64_t total_records(struct c2c_stats *stats)
695{
696 uint64_t lclmiss, ldcnt, total;
697
698 lclmiss = stats->lcl_dram +
699 stats->rmt_dram +
700 stats->rmt_hitm +
701 stats->rmt_hit;
702
703 ldcnt = lclmiss +
704 stats->ld_fbhit +
705 stats->ld_l1hit +
706 stats->ld_l2hit +
707 stats->ld_llchit +
708 stats->lcl_hitm;
709
710 total = ldcnt +
711 stats->st_l1hit +
712 stats->st_l1miss;
713
714 return total;
715}
716
717static int
718tot_recs_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
719 struct hist_entry *he)
720{
721 struct c2c_hist_entry *c2c_he;
722 int width = c2c_width(fmt, hpp, he->hists);
723 uint64_t tot_recs;
724
725 c2c_he = container_of(he, struct c2c_hist_entry, he);
726 tot_recs = total_records(&c2c_he->stats);
727
728 return scnprintf(hpp->buf, hpp->size, "%*" PRIu64, width, tot_recs);
729}
730
731static int64_t
732tot_recs_cmp(struct perf_hpp_fmt *fmt __maybe_unused,
733 struct hist_entry *left, struct hist_entry *right)
734{
735 struct c2c_hist_entry *c2c_left;
736 struct c2c_hist_entry *c2c_right;
737 uint64_t tot_recs_left;
738 uint64_t tot_recs_right;
739
740 c2c_left = container_of(left, struct c2c_hist_entry, he);
741 c2c_right = container_of(right, struct c2c_hist_entry, he);
742
743 tot_recs_left = total_records(&c2c_left->stats);
744 tot_recs_right = total_records(&c2c_right->stats);
745
746 return tot_recs_left - tot_recs_right;
747}
748
749static uint64_t total_loads(struct c2c_stats *stats)
750{
751 uint64_t lclmiss, ldcnt;
752
753 lclmiss = stats->lcl_dram +
754 stats->rmt_dram +
755 stats->rmt_hitm +
756 stats->rmt_hit;
757
758 ldcnt = lclmiss +
759 stats->ld_fbhit +
760 stats->ld_l1hit +
761 stats->ld_l2hit +
762 stats->ld_llchit +
763 stats->lcl_hitm;
764
765 return ldcnt;
766}
767
768static int
769tot_loads_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
770 struct hist_entry *he)
771{
772 struct c2c_hist_entry *c2c_he;
773 int width = c2c_width(fmt, hpp, he->hists);
774 uint64_t tot_recs;
775
776 c2c_he = container_of(he, struct c2c_hist_entry, he);
777 tot_recs = total_loads(&c2c_he->stats);
778
779 return scnprintf(hpp->buf, hpp->size, "%*" PRIu64, width, tot_recs);
780}
781
782static int64_t
783tot_loads_cmp(struct perf_hpp_fmt *fmt __maybe_unused,
784 struct hist_entry *left, struct hist_entry *right)
785{
786 struct c2c_hist_entry *c2c_left;
787 struct c2c_hist_entry *c2c_right;
788 uint64_t tot_recs_left;
789 uint64_t tot_recs_right;
790
791 c2c_left = container_of(left, struct c2c_hist_entry, he);
792 c2c_right = container_of(right, struct c2c_hist_entry, he);
793
794 tot_recs_left = total_loads(&c2c_left->stats);
795 tot_recs_right = total_loads(&c2c_right->stats);
796
797 return tot_recs_left - tot_recs_right;
798}
799
800typedef double (get_percent_cb)(struct c2c_hist_entry *);
801
802static int
803percent_color(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
804 struct hist_entry *he, get_percent_cb get_percent)
805{
806 struct c2c_hist_entry *c2c_he;
807 int width = c2c_width(fmt, hpp, he->hists);
808 double per;
809
810 c2c_he = container_of(he, struct c2c_hist_entry, he);
811 per = get_percent(c2c_he);
812
813#ifdef HAVE_SLANG_SUPPORT
814 if (use_browser)
815 return __hpp__slsmg_color_printf(hpp, "%*.2f%%", width - 1, per);
816#endif
817 return hpp_color_scnprintf(hpp, "%*.2f%%", width - 1, per);
818}
819
820static double percent_hitm(struct c2c_hist_entry *c2c_he)
821{
822 struct c2c_hists *hists;
823 struct c2c_stats *stats;
824 struct c2c_stats *total;
825 int tot = 0, st = 0;
826 double p;
827
828 hists = container_of(c2c_he->he.hists, struct c2c_hists, hists);
829 stats = &c2c_he->stats;
830 total = &hists->stats;
831
832 switch (c2c.display) {
833 case DISPLAY_RMT:
834 st = stats->rmt_hitm;
835 tot = total->rmt_hitm;
836 break;
837 case DISPLAY_LCL:
838 st = stats->lcl_hitm;
839 tot = total->lcl_hitm;
840 break;
841 case DISPLAY_TOT:
842 st = stats->tot_hitm;
843 tot = total->tot_hitm;
844 default:
845 break;
846 }
847
848 p = tot ? (double) st / tot : 0;
849
850 return 100 * p;
851}
852
853#define PERC_STR(__s, __v) \
854({ \
855 scnprintf(__s, sizeof(__s), "%.2F%%", __v); \
856 __s; \
857})
858
859static int
860percent_hitm_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
861 struct hist_entry *he)
862{
863 struct c2c_hist_entry *c2c_he;
864 int width = c2c_width(fmt, hpp, he->hists);
865 char buf[10];
866 double per;
867
868 c2c_he = container_of(he, struct c2c_hist_entry, he);
869 per = percent_hitm(c2c_he);
870 return scnprintf(hpp->buf, hpp->size, "%*s", width, PERC_STR(buf, per));
871}
872
873static int
874percent_hitm_color(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
875 struct hist_entry *he)
876{
877 return percent_color(fmt, hpp, he, percent_hitm);
878}
879
880static int64_t
881percent_hitm_cmp(struct perf_hpp_fmt *fmt __maybe_unused,
882 struct hist_entry *left, struct hist_entry *right)
883{
884 struct c2c_hist_entry *c2c_left;
885 struct c2c_hist_entry *c2c_right;
886 double per_left;
887 double per_right;
888
889 c2c_left = container_of(left, struct c2c_hist_entry, he);
890 c2c_right = container_of(right, struct c2c_hist_entry, he);
891
892 per_left = percent_hitm(c2c_left);
893 per_right = percent_hitm(c2c_right);
894
895 return per_left - per_right;
896}
897
898static struct c2c_stats *he_stats(struct hist_entry *he)
899{
900 struct c2c_hist_entry *c2c_he;
901
902 c2c_he = container_of(he, struct c2c_hist_entry, he);
903 return &c2c_he->stats;
904}
905
906static struct c2c_stats *total_stats(struct hist_entry *he)
907{
908 struct c2c_hists *hists;
909
910 hists = container_of(he->hists, struct c2c_hists, hists);
911 return &hists->stats;
912}
913
914static double percent(int st, int tot)
915{
916 return tot ? 100. * (double) st / (double) tot : 0;
917}
918
919#define PERCENT(__h, __f) percent(he_stats(__h)->__f, total_stats(__h)->__f)
920
921#define PERCENT_FN(__f) \
922static double percent_ ## __f(struct c2c_hist_entry *c2c_he) \
923{ \
924 struct c2c_hists *hists; \
925 \
926 hists = container_of(c2c_he->he.hists, struct c2c_hists, hists); \
927 return percent(c2c_he->stats.__f, hists->stats.__f); \
928}
929
930PERCENT_FN(rmt_hitm)
931PERCENT_FN(lcl_hitm)
932PERCENT_FN(st_l1hit)
933PERCENT_FN(st_l1miss)
934
935static int
936percent_rmt_hitm_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
937 struct hist_entry *he)
938{
939 int width = c2c_width(fmt, hpp, he->hists);
940 double per = PERCENT(he, rmt_hitm);
941 char buf[10];
942
943 return scnprintf(hpp->buf, hpp->size, "%*s", width, PERC_STR(buf, per));
944}
945
946static int
947percent_rmt_hitm_color(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
948 struct hist_entry *he)
949{
950 return percent_color(fmt, hpp, he, percent_rmt_hitm);
951}
952
953static int64_t
954percent_rmt_hitm_cmp(struct perf_hpp_fmt *fmt __maybe_unused,
955 struct hist_entry *left, struct hist_entry *right)
956{
957 double per_left;
958 double per_right;
959
960 per_left = PERCENT(left, lcl_hitm);
961 per_right = PERCENT(right, lcl_hitm);
962
963 return per_left - per_right;
964}
965
966static int
967percent_lcl_hitm_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
968 struct hist_entry *he)
969{
970 int width = c2c_width(fmt, hpp, he->hists);
971 double per = PERCENT(he, lcl_hitm);
972 char buf[10];
973
974 return scnprintf(hpp->buf, hpp->size, "%*s", width, PERC_STR(buf, per));
975}
976
977static int
978percent_lcl_hitm_color(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
979 struct hist_entry *he)
980{
981 return percent_color(fmt, hpp, he, percent_lcl_hitm);
982}
983
984static int64_t
985percent_lcl_hitm_cmp(struct perf_hpp_fmt *fmt __maybe_unused,
986 struct hist_entry *left, struct hist_entry *right)
987{
988 double per_left;
989 double per_right;
990
991 per_left = PERCENT(left, lcl_hitm);
992 per_right = PERCENT(right, lcl_hitm);
993
994 return per_left - per_right;
995}
996
997static int
998percent_stores_l1hit_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
999 struct hist_entry *he)
1000{
1001 int width = c2c_width(fmt, hpp, he->hists);
1002 double per = PERCENT(he, st_l1hit);
1003 char buf[10];
1004
1005 return scnprintf(hpp->buf, hpp->size, "%*s", width, PERC_STR(buf, per));
1006}
1007
1008static int
1009percent_stores_l1hit_color(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
1010 struct hist_entry *he)
1011{
1012 return percent_color(fmt, hpp, he, percent_st_l1hit);
1013}
1014
1015static int64_t
1016percent_stores_l1hit_cmp(struct perf_hpp_fmt *fmt __maybe_unused,
1017 struct hist_entry *left, struct hist_entry *right)
1018{
1019 double per_left;
1020 double per_right;
1021
1022 per_left = PERCENT(left, st_l1hit);
1023 per_right = PERCENT(right, st_l1hit);
1024
1025 return per_left - per_right;
1026}
1027
1028static int
1029percent_stores_l1miss_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
1030 struct hist_entry *he)
1031{
1032 int width = c2c_width(fmt, hpp, he->hists);
1033 double per = PERCENT(he, st_l1miss);
1034 char buf[10];
1035
1036 return scnprintf(hpp->buf, hpp->size, "%*s", width, PERC_STR(buf, per));
1037}
1038
1039static int
1040percent_stores_l1miss_color(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
1041 struct hist_entry *he)
1042{
1043 return percent_color(fmt, hpp, he, percent_st_l1miss);
1044}
1045
1046static int64_t
1047percent_stores_l1miss_cmp(struct perf_hpp_fmt *fmt __maybe_unused,
1048 struct hist_entry *left, struct hist_entry *right)
1049{
1050 double per_left;
1051 double per_right;
1052
1053 per_left = PERCENT(left, st_l1miss);
1054 per_right = PERCENT(right, st_l1miss);
1055
1056 return per_left - per_right;
1057}
1058
1059STAT_FN(lcl_dram)
1060STAT_FN(rmt_dram)
1061
1062static int
1063pid_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
1064 struct hist_entry *he)
1065{
1066 int width = c2c_width(fmt, hpp, he->hists);
1067
1068 return scnprintf(hpp->buf, hpp->size, "%*d", width, he->thread->pid_);
1069}
1070
1071static int64_t
1072pid_cmp(struct perf_hpp_fmt *fmt __maybe_unused,
1073 struct hist_entry *left, struct hist_entry *right)
1074{
1075 return left->thread->pid_ - right->thread->pid_;
1076}
1077
1078static int64_t
1079empty_cmp(struct perf_hpp_fmt *fmt __maybe_unused,
1080 struct hist_entry *left __maybe_unused,
1081 struct hist_entry *right __maybe_unused)
1082{
1083 return 0;
1084}
1085
1086static int
1087node_entry(struct perf_hpp_fmt *fmt __maybe_unused, struct perf_hpp *hpp,
1088 struct hist_entry *he)
1089{
1090 struct c2c_hist_entry *c2c_he;
1091 bool first = true;
1092 int node;
1093 int ret = 0;
1094
1095 c2c_he = container_of(he, struct c2c_hist_entry, he);
1096
1097 for (node = 0; node < c2c.nodes_cnt; node++) {
1098 DECLARE_BITMAP(set, c2c.cpus_cnt);
1099
1100 bitmap_zero(set, c2c.cpus_cnt);
1101 bitmap_and(set, c2c_he->cpuset, c2c.nodes[node], c2c.cpus_cnt);
1102
1103 if (!bitmap_weight(set, c2c.cpus_cnt)) {
1104 if (c2c.node_info == 1) {
1105 ret = scnprintf(hpp->buf, hpp->size, "%21s", " ");
1106 advance_hpp(hpp, ret);
1107 }
1108 continue;
1109 }
1110
1111 if (!first) {
1112 ret = scnprintf(hpp->buf, hpp->size, " ");
1113 advance_hpp(hpp, ret);
1114 }
1115
1116 switch (c2c.node_info) {
1117 case 0:
1118 ret = scnprintf(hpp->buf, hpp->size, "%2d", node);
1119 advance_hpp(hpp, ret);
1120 break;
1121 case 1:
1122 {
1123 int num = bitmap_weight(set, c2c.cpus_cnt);
1124 struct c2c_stats *stats = &c2c_he->node_stats[node];
1125
1126 ret = scnprintf(hpp->buf, hpp->size, "%2d{%2d ", node, num);
1127 advance_hpp(hpp, ret);
1128
1129 #define DISPLAY_HITM(__h) \
1130 if (c2c_he->stats.__h> 0) { \
1131 ret = scnprintf(hpp->buf, hpp->size, "%5.1f%% ", \
1132 percent(stats->__h, c2c_he->stats.__h));\
1133 } else { \
1134 ret = scnprintf(hpp->buf, hpp->size, "%6s ", "n/a"); \
1135 }
1136
1137 switch (c2c.display) {
1138 case DISPLAY_RMT:
1139 DISPLAY_HITM(rmt_hitm);
1140 break;
1141 case DISPLAY_LCL:
1142 DISPLAY_HITM(lcl_hitm);
1143 break;
1144 case DISPLAY_TOT:
1145 DISPLAY_HITM(tot_hitm);
1146 default:
1147 break;
1148 }
1149
1150 #undef DISPLAY_HITM
1151
1152 advance_hpp(hpp, ret);
1153
1154 if (c2c_he->stats.store > 0) {
1155 ret = scnprintf(hpp->buf, hpp->size, "%5.1f%%}",
1156 percent(stats->store, c2c_he->stats.store));
1157 } else {
1158 ret = scnprintf(hpp->buf, hpp->size, "%6s}", "n/a");
1159 }
1160
1161 advance_hpp(hpp, ret);
1162 break;
1163 }
1164 case 2:
1165 ret = scnprintf(hpp->buf, hpp->size, "%2d{", node);
1166 advance_hpp(hpp, ret);
1167
1168 ret = bitmap_scnprintf(set, c2c.cpus_cnt, hpp->buf, hpp->size);
1169 advance_hpp(hpp, ret);
1170
1171 ret = scnprintf(hpp->buf, hpp->size, "}");
1172 advance_hpp(hpp, ret);
1173 break;
1174 default:
1175 break;
1176 }
1177
1178 first = false;
1179 }
1180
1181 return 0;
1182}
1183
1184static int
1185mean_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
1186 struct hist_entry *he, double mean)
1187{
1188 int width = c2c_width(fmt, hpp, he->hists);
1189 char buf[10];
1190
1191 scnprintf(buf, 10, "%6.0f", mean);
1192 return scnprintf(hpp->buf, hpp->size, "%*s", width, buf);
1193}
1194
1195#define MEAN_ENTRY(__func, __val) \
1196static int \
1197__func(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp, struct hist_entry *he) \
1198{ \
1199 struct c2c_hist_entry *c2c_he; \
1200 c2c_he = container_of(he, struct c2c_hist_entry, he); \
1201 return mean_entry(fmt, hpp, he, avg_stats(&c2c_he->cstats.__val)); \
1202}
1203
1204MEAN_ENTRY(mean_rmt_entry, rmt_hitm);
1205MEAN_ENTRY(mean_lcl_entry, lcl_hitm);
1206MEAN_ENTRY(mean_load_entry, load);
1207
1208static int
1209cpucnt_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
1210 struct hist_entry *he)
1211{
1212 struct c2c_hist_entry *c2c_he;
1213 int width = c2c_width(fmt, hpp, he->hists);
1214 char buf[10];
1215
1216 c2c_he = container_of(he, struct c2c_hist_entry, he);
1217
1218 scnprintf(buf, 10, "%d", bitmap_weight(c2c_he->cpuset, c2c.cpus_cnt));
1219 return scnprintf(hpp->buf, hpp->size, "%*s", width, buf);
1220}
1221
1222static int
1223cl_idx_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
1224 struct hist_entry *he)
1225{
1226 struct c2c_hist_entry *c2c_he;
1227 int width = c2c_width(fmt, hpp, he->hists);
1228 char buf[10];
1229
1230 c2c_he = container_of(he, struct c2c_hist_entry, he);
1231
1232 scnprintf(buf, 10, "%u", c2c_he->cacheline_idx);
1233 return scnprintf(hpp->buf, hpp->size, "%*s", width, buf);
1234}
1235
1236static int
1237cl_idx_empty_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
1238 struct hist_entry *he)
1239{
1240 int width = c2c_width(fmt, hpp, he->hists);
1241
1242 return scnprintf(hpp->buf, hpp->size, "%*s", width, "");
1243}
1244
1245#define HEADER_LOW(__h) \
1246 { \
1247 .line[1] = { \
1248 .text = __h, \
1249 }, \
1250 }
1251
1252#define HEADER_BOTH(__h0, __h1) \
1253 { \
1254 .line[0] = { \
1255 .text = __h0, \
1256 }, \
1257 .line[1] = { \
1258 .text = __h1, \
1259 }, \
1260 }
1261
1262#define HEADER_SPAN(__h0, __h1, __s) \
1263 { \
1264 .line[0] = { \
1265 .text = __h0, \
1266 .span = __s, \
1267 }, \
1268 .line[1] = { \
1269 .text = __h1, \
1270 }, \
1271 }
1272
1273#define HEADER_SPAN_LOW(__h) \
1274 { \
1275 .line[1] = { \
1276 .text = __h, \
1277 }, \
1278 }
1279
1280static struct c2c_dimension dim_dcacheline = {
1281 .header = HEADER_SPAN("--- Cacheline ----", "Address", 2),
1282 .name = "dcacheline",
1283 .cmp = dcacheline_cmp,
1284 .entry = dcacheline_entry,
1285 .width = 18,
1286};
1287
1288static struct c2c_dimension dim_dcacheline_node = {
1289 .header = HEADER_LOW("Node"),
1290 .name = "dcacheline_node",
1291 .cmp = empty_cmp,
1292 .entry = dcacheline_node_entry,
1293 .width = 4,
1294};
1295
1296static struct c2c_dimension dim_dcacheline_count = {
1297 .header = HEADER_LOW("PA cnt"),
1298 .name = "dcacheline_count",
1299 .cmp = empty_cmp,
1300 .entry = dcacheline_node_count,
1301 .width = 6,
1302};
1303
1304static struct c2c_header header_offset_tui = HEADER_SPAN("-----", "Off", 2);
1305
1306static struct c2c_dimension dim_offset = {
1307 .header = HEADER_SPAN("--- Data address -", "Offset", 2),
1308 .name = "offset",
1309 .cmp = offset_cmp,
1310 .entry = offset_entry,
1311 .width = 18,
1312};
1313
1314static struct c2c_dimension dim_offset_node = {
1315 .header = HEADER_LOW("Node"),
1316 .name = "offset_node",
1317 .cmp = empty_cmp,
1318 .entry = dcacheline_node_entry,
1319 .width = 4,
1320};
1321
1322static struct c2c_dimension dim_iaddr = {
1323 .header = HEADER_LOW("Code address"),
1324 .name = "iaddr",
1325 .cmp = iaddr_cmp,
1326 .entry = iaddr_entry,
1327 .width = 18,
1328};
1329
1330static struct c2c_dimension dim_tot_hitm = {
1331 .header = HEADER_SPAN("----- LLC Load Hitm -----", "Total", 2),
1332 .name = "tot_hitm",
1333 .cmp = tot_hitm_cmp,
1334 .entry = tot_hitm_entry,
1335 .width = 7,
1336};
1337
1338static struct c2c_dimension dim_lcl_hitm = {
1339 .header = HEADER_SPAN_LOW("Lcl"),
1340 .name = "lcl_hitm",
1341 .cmp = lcl_hitm_cmp,
1342 .entry = lcl_hitm_entry,
1343 .width = 7,
1344};
1345
1346static struct c2c_dimension dim_rmt_hitm = {
1347 .header = HEADER_SPAN_LOW("Rmt"),
1348 .name = "rmt_hitm",
1349 .cmp = rmt_hitm_cmp,
1350 .entry = rmt_hitm_entry,
1351 .width = 7,
1352};
1353
1354static struct c2c_dimension dim_cl_rmt_hitm = {
1355 .header = HEADER_SPAN("----- HITM -----", "Rmt", 1),
1356 .name = "cl_rmt_hitm",
1357 .cmp = rmt_hitm_cmp,
1358 .entry = rmt_hitm_entry,
1359 .width = 7,
1360};
1361
1362static struct c2c_dimension dim_cl_lcl_hitm = {
1363 .header = HEADER_SPAN_LOW("Lcl"),
1364 .name = "cl_lcl_hitm",
1365 .cmp = lcl_hitm_cmp,
1366 .entry = lcl_hitm_entry,
1367 .width = 7,
1368};
1369
1370static struct c2c_dimension dim_stores = {
1371 .header = HEADER_SPAN("---- Store Reference ----", "Total", 2),
1372 .name = "stores",
1373 .cmp = store_cmp,
1374 .entry = store_entry,
1375 .width = 7,
1376};
1377
1378static struct c2c_dimension dim_stores_l1hit = {
1379 .header = HEADER_SPAN_LOW("L1Hit"),
1380 .name = "stores_l1hit",
1381 .cmp = st_l1hit_cmp,
1382 .entry = st_l1hit_entry,
1383 .width = 7,
1384};
1385
1386static struct c2c_dimension dim_stores_l1miss = {
1387 .header = HEADER_SPAN_LOW("L1Miss"),
1388 .name = "stores_l1miss",
1389 .cmp = st_l1miss_cmp,
1390 .entry = st_l1miss_entry,
1391 .width = 7,
1392};
1393
1394static struct c2c_dimension dim_cl_stores_l1hit = {
1395 .header = HEADER_SPAN("-- Store Refs --", "L1 Hit", 1),
1396 .name = "cl_stores_l1hit",
1397 .cmp = st_l1hit_cmp,
1398 .entry = st_l1hit_entry,
1399 .width = 7,
1400};
1401
1402static struct c2c_dimension dim_cl_stores_l1miss = {
1403 .header = HEADER_SPAN_LOW("L1 Miss"),
1404 .name = "cl_stores_l1miss",
1405 .cmp = st_l1miss_cmp,
1406 .entry = st_l1miss_entry,
1407 .width = 7,
1408};
1409
1410static struct c2c_dimension dim_ld_fbhit = {
1411 .header = HEADER_SPAN("----- Core Load Hit -----", "FB", 2),
1412 .name = "ld_fbhit",
1413 .cmp = ld_fbhit_cmp,
1414 .entry = ld_fbhit_entry,
1415 .width = 7,
1416};
1417
1418static struct c2c_dimension dim_ld_l1hit = {
1419 .header = HEADER_SPAN_LOW("L1"),
1420 .name = "ld_l1hit",
1421 .cmp = ld_l1hit_cmp,
1422 .entry = ld_l1hit_entry,
1423 .width = 7,
1424};
1425
1426static struct c2c_dimension dim_ld_l2hit = {
1427 .header = HEADER_SPAN_LOW("L2"),
1428 .name = "ld_l2hit",
1429 .cmp = ld_l2hit_cmp,
1430 .entry = ld_l2hit_entry,
1431 .width = 7,
1432};
1433
1434static struct c2c_dimension dim_ld_llchit = {
1435 .header = HEADER_SPAN("-- LLC Load Hit --", "Llc", 1),
1436 .name = "ld_lclhit",
1437 .cmp = ld_llchit_cmp,
1438 .entry = ld_llchit_entry,
1439 .width = 8,
1440};
1441
1442static struct c2c_dimension dim_ld_rmthit = {
1443 .header = HEADER_SPAN_LOW("Rmt"),
1444 .name = "ld_rmthit",
1445 .cmp = rmt_hit_cmp,
1446 .entry = rmt_hit_entry,
1447 .width = 8,
1448};
1449
1450static struct c2c_dimension dim_ld_llcmiss = {
1451 .header = HEADER_BOTH("LLC", "Ld Miss"),
1452 .name = "ld_llcmiss",
1453 .cmp = ld_llcmiss_cmp,
1454 .entry = ld_llcmiss_entry,
1455 .width = 7,
1456};
1457
1458static struct c2c_dimension dim_tot_recs = {
1459 .header = HEADER_BOTH("Total", "records"),
1460 .name = "tot_recs",
1461 .cmp = tot_recs_cmp,
1462 .entry = tot_recs_entry,
1463 .width = 7,
1464};
1465
1466static struct c2c_dimension dim_tot_loads = {
1467 .header = HEADER_BOTH("Total", "Loads"),
1468 .name = "tot_loads",
1469 .cmp = tot_loads_cmp,
1470 .entry = tot_loads_entry,
1471 .width = 7,
1472};
1473
1474static struct c2c_header percent_hitm_header[] = {
1475 [DISPLAY_LCL] = HEADER_BOTH("Lcl", "Hitm"),
1476 [DISPLAY_RMT] = HEADER_BOTH("Rmt", "Hitm"),
1477 [DISPLAY_TOT] = HEADER_BOTH("Tot", "Hitm"),
1478};
1479
1480static struct c2c_dimension dim_percent_hitm = {
1481 .name = "percent_hitm",
1482 .cmp = percent_hitm_cmp,
1483 .entry = percent_hitm_entry,
1484 .color = percent_hitm_color,
1485 .width = 7,
1486};
1487
1488static struct c2c_dimension dim_percent_rmt_hitm = {
1489 .header = HEADER_SPAN("----- HITM -----", "Rmt", 1),
1490 .name = "percent_rmt_hitm",
1491 .cmp = percent_rmt_hitm_cmp,
1492 .entry = percent_rmt_hitm_entry,
1493 .color = percent_rmt_hitm_color,
1494 .width = 7,
1495};
1496
1497static struct c2c_dimension dim_percent_lcl_hitm = {
1498 .header = HEADER_SPAN_LOW("Lcl"),
1499 .name = "percent_lcl_hitm",
1500 .cmp = percent_lcl_hitm_cmp,
1501 .entry = percent_lcl_hitm_entry,
1502 .color = percent_lcl_hitm_color,
1503 .width = 7,
1504};
1505
1506static struct c2c_dimension dim_percent_stores_l1hit = {
1507 .header = HEADER_SPAN("-- Store Refs --", "L1 Hit", 1),
1508 .name = "percent_stores_l1hit",
1509 .cmp = percent_stores_l1hit_cmp,
1510 .entry = percent_stores_l1hit_entry,
1511 .color = percent_stores_l1hit_color,
1512 .width = 7,
1513};
1514
1515static struct c2c_dimension dim_percent_stores_l1miss = {
1516 .header = HEADER_SPAN_LOW("L1 Miss"),
1517 .name = "percent_stores_l1miss",
1518 .cmp = percent_stores_l1miss_cmp,
1519 .entry = percent_stores_l1miss_entry,
1520 .color = percent_stores_l1miss_color,
1521 .width = 7,
1522};
1523
1524static struct c2c_dimension dim_dram_lcl = {
1525 .header = HEADER_SPAN("--- Load Dram ----", "Lcl", 1),
1526 .name = "dram_lcl",
1527 .cmp = lcl_dram_cmp,
1528 .entry = lcl_dram_entry,
1529 .width = 8,
1530};
1531
1532static struct c2c_dimension dim_dram_rmt = {
1533 .header = HEADER_SPAN_LOW("Rmt"),
1534 .name = "dram_rmt",
1535 .cmp = rmt_dram_cmp,
1536 .entry = rmt_dram_entry,
1537 .width = 8,
1538};
1539
1540static struct c2c_dimension dim_pid = {
1541 .header = HEADER_LOW("Pid"),
1542 .name = "pid",
1543 .cmp = pid_cmp,
1544 .entry = pid_entry,
1545 .width = 7,
1546};
1547
1548static struct c2c_dimension dim_tid = {
1549 .header = HEADER_LOW("Tid"),
1550 .name = "tid",
1551 .se = &sort_thread,
1552};
1553
1554static struct c2c_dimension dim_symbol = {
1555 .name = "symbol",
1556 .se = &sort_sym,
1557};
1558
1559static struct c2c_dimension dim_dso = {
1560 .header = HEADER_BOTH("Shared", "Object"),
1561 .name = "dso",
1562 .se = &sort_dso,
1563};
1564
1565static struct c2c_header header_node[3] = {
1566 HEADER_LOW("Node"),
1567 HEADER_LOW("Node{cpus %hitms %stores}"),
1568 HEADER_LOW("Node{cpu list}"),
1569};
1570
1571static struct c2c_dimension dim_node = {
1572 .name = "node",
1573 .cmp = empty_cmp,
1574 .entry = node_entry,
1575 .width = 4,
1576};
1577
1578static struct c2c_dimension dim_mean_rmt = {
1579 .header = HEADER_SPAN("---------- cycles ----------", "rmt hitm", 2),
1580 .name = "mean_rmt",
1581 .cmp = empty_cmp,
1582 .entry = mean_rmt_entry,
1583 .width = 8,
1584};
1585
1586static struct c2c_dimension dim_mean_lcl = {
1587 .header = HEADER_SPAN_LOW("lcl hitm"),
1588 .name = "mean_lcl",
1589 .cmp = empty_cmp,
1590 .entry = mean_lcl_entry,
1591 .width = 8,
1592};
1593
1594static struct c2c_dimension dim_mean_load = {
1595 .header = HEADER_SPAN_LOW("load"),
1596 .name = "mean_load",
1597 .cmp = empty_cmp,
1598 .entry = mean_load_entry,
1599 .width = 8,
1600};
1601
1602static struct c2c_dimension dim_cpucnt = {
1603 .header = HEADER_BOTH("cpu", "cnt"),
1604 .name = "cpucnt",
1605 .cmp = empty_cmp,
1606 .entry = cpucnt_entry,
1607 .width = 8,
1608};
1609
1610static struct c2c_dimension dim_srcline = {
1611 .name = "cl_srcline",
1612 .se = &sort_srcline,
1613};
1614
1615static struct c2c_dimension dim_dcacheline_idx = {
1616 .header = HEADER_LOW("Index"),
1617 .name = "cl_idx",
1618 .cmp = empty_cmp,
1619 .entry = cl_idx_entry,
1620 .width = 5,
1621};
1622
1623static struct c2c_dimension dim_dcacheline_num = {
1624 .header = HEADER_LOW("Num"),
1625 .name = "cl_num",
1626 .cmp = empty_cmp,
1627 .entry = cl_idx_entry,
1628 .width = 5,
1629};
1630
1631static struct c2c_dimension dim_dcacheline_num_empty = {
1632 .header = HEADER_LOW("Num"),
1633 .name = "cl_num_empty",
1634 .cmp = empty_cmp,
1635 .entry = cl_idx_empty_entry,
1636 .width = 5,
1637};
1638
1639static struct c2c_dimension *dimensions[] = {
1640 &dim_dcacheline,
1641 &dim_dcacheline_node,
1642 &dim_dcacheline_count,
1643 &dim_offset,
1644 &dim_offset_node,
1645 &dim_iaddr,
1646 &dim_tot_hitm,
1647 &dim_lcl_hitm,
1648 &dim_rmt_hitm,
1649 &dim_cl_lcl_hitm,
1650 &dim_cl_rmt_hitm,
1651 &dim_stores,
1652 &dim_stores_l1hit,
1653 &dim_stores_l1miss,
1654 &dim_cl_stores_l1hit,
1655 &dim_cl_stores_l1miss,
1656 &dim_ld_fbhit,
1657 &dim_ld_l1hit,
1658 &dim_ld_l2hit,
1659 &dim_ld_llchit,
1660 &dim_ld_rmthit,
1661 &dim_ld_llcmiss,
1662 &dim_tot_recs,
1663 &dim_tot_loads,
1664 &dim_percent_hitm,
1665 &dim_percent_rmt_hitm,
1666 &dim_percent_lcl_hitm,
1667 &dim_percent_stores_l1hit,
1668 &dim_percent_stores_l1miss,
1669 &dim_dram_lcl,
1670 &dim_dram_rmt,
1671 &dim_pid,
1672 &dim_tid,
1673 &dim_symbol,
1674 &dim_dso,
1675 &dim_node,
1676 &dim_mean_rmt,
1677 &dim_mean_lcl,
1678 &dim_mean_load,
1679 &dim_cpucnt,
1680 &dim_srcline,
1681 &dim_dcacheline_idx,
1682 &dim_dcacheline_num,
1683 &dim_dcacheline_num_empty,
1684 NULL,
1685};
1686
1687static void fmt_free(struct perf_hpp_fmt *fmt)
1688{
1689 struct c2c_fmt *c2c_fmt;
1690
1691 c2c_fmt = container_of(fmt, struct c2c_fmt, fmt);
1692 free(c2c_fmt);
1693}
1694
1695static bool fmt_equal(struct perf_hpp_fmt *a, struct perf_hpp_fmt *b)
1696{
1697 struct c2c_fmt *c2c_a = container_of(a, struct c2c_fmt, fmt);
1698 struct c2c_fmt *c2c_b = container_of(b, struct c2c_fmt, fmt);
1699
1700 return c2c_a->dim == c2c_b->dim;
1701}
1702
1703static struct c2c_dimension *get_dimension(const char *name)
1704{
1705 unsigned int i;
1706
1707 for (i = 0; dimensions[i]; i++) {
1708 struct c2c_dimension *dim = dimensions[i];
1709
1710 if (!strcmp(dim->name, name))
1711 return dim;
1712 }
1713
1714 return NULL;
1715}
1716
1717static int c2c_se_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
1718 struct hist_entry *he)
1719{
1720 struct c2c_fmt *c2c_fmt = container_of(fmt, struct c2c_fmt, fmt);
1721 struct c2c_dimension *dim = c2c_fmt->dim;
1722 size_t len = fmt->user_len;
1723
1724 if (!len) {
1725 len = hists__col_len(he->hists, dim->se->se_width_idx);
1726
1727 if (dim == &dim_symbol || dim == &dim_srcline)
1728 len = symbol_width(he->hists, dim->se);
1729 }
1730
1731 return dim->se->se_snprintf(he, hpp->buf, hpp->size, len);
1732}
1733
1734static int64_t c2c_se_cmp(struct perf_hpp_fmt *fmt,
1735 struct hist_entry *a, struct hist_entry *b)
1736{
1737 struct c2c_fmt *c2c_fmt = container_of(fmt, struct c2c_fmt, fmt);
1738 struct c2c_dimension *dim = c2c_fmt->dim;
1739
1740 return dim->se->se_cmp(a, b);
1741}
1742
1743static int64_t c2c_se_collapse(struct perf_hpp_fmt *fmt,
1744 struct hist_entry *a, struct hist_entry *b)
1745{
1746 struct c2c_fmt *c2c_fmt = container_of(fmt, struct c2c_fmt, fmt);
1747 struct c2c_dimension *dim = c2c_fmt->dim;
1748 int64_t (*collapse_fn)(struct hist_entry *, struct hist_entry *);
1749
1750 collapse_fn = dim->se->se_collapse ?: dim->se->se_cmp;
1751 return collapse_fn(a, b);
1752}
1753
1754static struct c2c_fmt *get_format(const char *name)
1755{
1756 struct c2c_dimension *dim = get_dimension(name);
1757 struct c2c_fmt *c2c_fmt;
1758 struct perf_hpp_fmt *fmt;
1759
1760 if (!dim)
1761 return NULL;
1762
1763 c2c_fmt = zalloc(sizeof(*c2c_fmt));
1764 if (!c2c_fmt)
1765 return NULL;
1766
1767 c2c_fmt->dim = dim;
1768
1769 fmt = &c2c_fmt->fmt;
1770 INIT_LIST_HEAD(&fmt->list);
1771 INIT_LIST_HEAD(&fmt->sort_list);
1772
1773 fmt->cmp = dim->se ? c2c_se_cmp : dim->cmp;
1774 fmt->sort = dim->se ? c2c_se_cmp : dim->cmp;
1775 fmt->color = dim->se ? NULL : dim->color;
1776 fmt->entry = dim->se ? c2c_se_entry : dim->entry;
1777 fmt->header = c2c_header;
1778 fmt->width = c2c_width;
1779 fmt->collapse = dim->se ? c2c_se_collapse : dim->cmp;
1780 fmt->equal = fmt_equal;
1781 fmt->free = fmt_free;
1782
1783 return c2c_fmt;
1784}
1785
1786static int c2c_hists__init_output(struct perf_hpp_list *hpp_list, char *name)
1787{
1788 struct c2c_fmt *c2c_fmt = get_format(name);
1789
1790 if (!c2c_fmt) {
1791 reset_dimensions();
1792 return output_field_add(hpp_list, name);
1793 }
1794
1795 perf_hpp_list__column_register(hpp_list, &c2c_fmt->fmt);
1796 return 0;
1797}
1798
1799static int c2c_hists__init_sort(struct perf_hpp_list *hpp_list, char *name)
1800{
1801 struct c2c_fmt *c2c_fmt = get_format(name);
1802 struct c2c_dimension *dim;
1803
1804 if (!c2c_fmt) {
1805 reset_dimensions();
1806 return sort_dimension__add(hpp_list, name, NULL, 0);
1807 }
1808
1809 dim = c2c_fmt->dim;
1810 if (dim == &dim_dso)
1811 hpp_list->dso = 1;
1812
1813 perf_hpp_list__register_sort_field(hpp_list, &c2c_fmt->fmt);
1814 return 0;
1815}
1816
1817#define PARSE_LIST(_list, _fn) \
1818 do { \
1819 char *tmp, *tok; \
1820 ret = 0; \
1821 \
1822 if (!_list) \
1823 break; \
1824 \
1825 for (tok = strtok_r((char *)_list, ", ", &tmp); \
1826 tok; tok = strtok_r(NULL, ", ", &tmp)) { \
1827 ret = _fn(hpp_list, tok); \
1828 if (ret == -EINVAL) { \
1829 pr_err("Invalid --fields key: `%s'", tok); \
1830 break; \
1831 } else if (ret == -ESRCH) { \
1832 pr_err("Unknown --fields key: `%s'", tok); \
1833 break; \
1834 } \
1835 } \
1836 } while (0)
1837
1838static int hpp_list__parse(struct perf_hpp_list *hpp_list,
1839 const char *output_,
1840 const char *sort_)
1841{
1842 char *output = output_ ? strdup(output_) : NULL;
1843 char *sort = sort_ ? strdup(sort_) : NULL;
1844 int ret;
1845
1846 PARSE_LIST(output, c2c_hists__init_output);
1847 PARSE_LIST(sort, c2c_hists__init_sort);
1848
1849 /* copy sort keys to output fields */
1850 perf_hpp__setup_output_field(hpp_list);
1851
1852 /*
1853 * We dont need other sorting keys other than those
1854 * we already specified. It also really slows down
1855 * the processing a lot with big number of output
1856 * fields, so switching this off for c2c.
1857 */
1858
1859#if 0
1860 /* and then copy output fields to sort keys */
1861 perf_hpp__append_sort_keys(&hists->list);
1862#endif
1863
1864 free(output);
1865 free(sort);
1866 return ret;
1867}
1868
1869static int c2c_hists__init(struct c2c_hists *hists,
1870 const char *sort,
1871 int nr_header_lines)
1872{
1873 __hists__init(&hists->hists, &hists->list);
1874
1875 /*
1876 * Initialize only with sort fields, we need to resort
1877 * later anyway, and that's where we add output fields
1878 * as well.
1879 */
1880 perf_hpp_list__init(&hists->list);
1881
1882 /* Overload number of header lines.*/
1883 hists->list.nr_header_lines = nr_header_lines;
1884
1885 return hpp_list__parse(&hists->list, NULL, sort);
1886}
1887
1888static int c2c_hists__reinit(struct c2c_hists *c2c_hists,
1889 const char *output,
1890 const char *sort)
1891{
1892 perf_hpp__reset_output_field(&c2c_hists->list);
1893 return hpp_list__parse(&c2c_hists->list, output, sort);
1894}
1895
1896#define DISPLAY_LINE_LIMIT 0.001
1897
1898static bool he__display(struct hist_entry *he, struct c2c_stats *stats)
1899{
1900 struct c2c_hist_entry *c2c_he;
1901 double ld_dist;
1902
1903 if (c2c.show_all)
1904 return true;
1905
1906 c2c_he = container_of(he, struct c2c_hist_entry, he);
1907
1908#define FILTER_HITM(__h) \
1909 if (stats->__h) { \
1910 ld_dist = ((double)c2c_he->stats.__h / stats->__h); \
1911 if (ld_dist < DISPLAY_LINE_LIMIT) \
1912 he->filtered = HIST_FILTER__C2C; \
1913 } else { \
1914 he->filtered = HIST_FILTER__C2C; \
1915 }
1916
1917 switch (c2c.display) {
1918 case DISPLAY_LCL:
1919 FILTER_HITM(lcl_hitm);
1920 break;
1921 case DISPLAY_RMT:
1922 FILTER_HITM(rmt_hitm);
1923 break;
1924 case DISPLAY_TOT:
1925 FILTER_HITM(tot_hitm);
1926 default:
1927 break;
1928 }
1929
1930#undef FILTER_HITM
1931
1932 return he->filtered == 0;
1933}
1934
1935static inline int valid_hitm_or_store(struct hist_entry *he)
1936{
1937 struct c2c_hist_entry *c2c_he;
1938 bool has_hitm;
1939
1940 c2c_he = container_of(he, struct c2c_hist_entry, he);
1941 has_hitm = c2c.display == DISPLAY_TOT ? c2c_he->stats.tot_hitm :
1942 c2c.display == DISPLAY_LCL ? c2c_he->stats.lcl_hitm :
1943 c2c_he->stats.rmt_hitm;
1944 return has_hitm || c2c_he->stats.store;
1945}
1946
1947static void set_node_width(struct c2c_hist_entry *c2c_he, int len)
1948{
1949 struct c2c_dimension *dim;
1950
1951 dim = &c2c.hists == c2c_he->hists ?
1952 &dim_dcacheline_node : &dim_offset_node;
1953
1954 if (len > dim->width)
1955 dim->width = len;
1956}
1957
1958static int set_nodestr(struct c2c_hist_entry *c2c_he)
1959{
1960 char buf[30];
1961 int len;
1962
1963 if (c2c_he->nodestr)
1964 return 0;
1965
1966 if (bitmap_weight(c2c_he->nodeset, c2c.nodes_cnt)) {
1967 len = bitmap_scnprintf(c2c_he->nodeset, c2c.nodes_cnt,
1968 buf, sizeof(buf));
1969 } else {
1970 len = scnprintf(buf, sizeof(buf), "N/A");
1971 }
1972
1973 set_node_width(c2c_he, len);
1974 c2c_he->nodestr = strdup(buf);
1975 return c2c_he->nodestr ? 0 : -ENOMEM;
1976}
1977
1978static void calc_width(struct c2c_hist_entry *c2c_he)
1979{
1980 struct c2c_hists *c2c_hists;
1981
1982 c2c_hists = container_of(c2c_he->he.hists, struct c2c_hists, hists);
1983 hists__calc_col_len(&c2c_hists->hists, &c2c_he->he);
1984 set_nodestr(c2c_he);
1985}
1986
1987static int filter_cb(struct hist_entry *he, void *arg __maybe_unused)
1988{
1989 struct c2c_hist_entry *c2c_he;
1990
1991 c2c_he = container_of(he, struct c2c_hist_entry, he);
1992
1993 if (c2c.show_src && !he->srcline)
1994 he->srcline = hist_entry__srcline(he);
1995
1996 calc_width(c2c_he);
1997
1998 if (!valid_hitm_or_store(he))
1999 he->filtered = HIST_FILTER__C2C;
2000
2001 return 0;
2002}
2003
2004static int resort_cl_cb(struct hist_entry *he, void *arg __maybe_unused)
2005{
2006 struct c2c_hist_entry *c2c_he;
2007 struct c2c_hists *c2c_hists;
2008 bool display = he__display(he, &c2c.hitm_stats);
2009
2010 c2c_he = container_of(he, struct c2c_hist_entry, he);
2011 c2c_hists = c2c_he->hists;
2012
2013 if (display && c2c_hists) {
2014 static unsigned int idx;
2015
2016 c2c_he->cacheline_idx = idx++;
2017 calc_width(c2c_he);
2018
2019 c2c_hists__reinit(c2c_hists, c2c.cl_output, c2c.cl_resort);
2020
2021 hists__collapse_resort(&c2c_hists->hists, NULL);
2022 hists__output_resort_cb(&c2c_hists->hists, NULL, filter_cb);
2023 }
2024
2025 return 0;
2026}
2027
2028static void setup_nodes_header(void)
2029{
2030 dim_node.header = header_node[c2c.node_info];
2031}
2032
2033static int setup_nodes(struct perf_session *session)
2034{
2035 struct numa_node *n;
2036 unsigned long **nodes;
2037 int node, cpu;
2038 int *cpu2node;
2039
2040 if (c2c.node_info > 2)
2041 c2c.node_info = 2;
2042
2043 c2c.nodes_cnt = session->header.env.nr_numa_nodes;
2044 c2c.cpus_cnt = session->header.env.nr_cpus_avail;
2045
2046 n = session->header.env.numa_nodes;
2047 if (!n)
2048 return -EINVAL;
2049
2050 nodes = zalloc(sizeof(unsigned long *) * c2c.nodes_cnt);
2051 if (!nodes)
2052 return -ENOMEM;
2053
2054 c2c.nodes = nodes;
2055
2056 cpu2node = zalloc(sizeof(int) * c2c.cpus_cnt);
2057 if (!cpu2node)
2058 return -ENOMEM;
2059
2060 for (cpu = 0; cpu < c2c.cpus_cnt; cpu++)
2061 cpu2node[cpu] = -1;
2062
2063 c2c.cpu2node = cpu2node;
2064
2065 for (node = 0; node < c2c.nodes_cnt; node++) {
2066 struct perf_cpu_map *map = n[node].map;
2067 unsigned long *set;
2068
2069 set = bitmap_alloc(c2c.cpus_cnt);
2070 if (!set)
2071 return -ENOMEM;
2072
2073 nodes[node] = set;
2074
2075 /* empty node, skip */
2076 if (perf_cpu_map__empty(map))
2077 continue;
2078
2079 for (cpu = 0; cpu < map->nr; cpu++) {
2080 set_bit(map->map[cpu], set);
2081
2082 if (WARN_ONCE(cpu2node[map->map[cpu]] != -1, "node/cpu topology bug"))
2083 return -EINVAL;
2084
2085 cpu2node[map->map[cpu]] = node;
2086 }
2087 }
2088
2089 setup_nodes_header();
2090 return 0;
2091}
2092
2093#define HAS_HITMS(__h) ((__h)->stats.lcl_hitm || (__h)->stats.rmt_hitm)
2094
2095static int resort_hitm_cb(struct hist_entry *he, void *arg __maybe_unused)
2096{
2097 struct c2c_hist_entry *c2c_he;
2098 c2c_he = container_of(he, struct c2c_hist_entry, he);
2099
2100 if (HAS_HITMS(c2c_he)) {
2101 c2c.shared_clines++;
2102 c2c_add_stats(&c2c.hitm_stats, &c2c_he->stats);
2103 }
2104
2105 return 0;
2106}
2107
2108static int hists__iterate_cb(struct hists *hists, hists__resort_cb_t cb)
2109{
2110 struct rb_node *next = rb_first_cached(&hists->entries);
2111 int ret = 0;
2112
2113 while (next) {
2114 struct hist_entry *he;
2115
2116 he = rb_entry(next, struct hist_entry, rb_node);
2117 ret = cb(he, NULL);
2118 if (ret)
2119 break;
2120 next = rb_next(&he->rb_node);
2121 }
2122
2123 return ret;
2124}
2125
2126static void print_c2c__display_stats(FILE *out)
2127{
2128 int llc_misses;
2129 struct c2c_stats *stats = &c2c.hists.stats;
2130
2131 llc_misses = stats->lcl_dram +
2132 stats->rmt_dram +
2133 stats->rmt_hit +
2134 stats->rmt_hitm;
2135
2136 fprintf(out, "=================================================\n");
2137 fprintf(out, " Trace Event Information \n");
2138 fprintf(out, "=================================================\n");
2139 fprintf(out, " Total records : %10d\n", stats->nr_entries);
2140 fprintf(out, " Locked Load/Store Operations : %10d\n", stats->locks);
2141 fprintf(out, " Load Operations : %10d\n", stats->load);
2142 fprintf(out, " Loads - uncacheable : %10d\n", stats->ld_uncache);
2143 fprintf(out, " Loads - IO : %10d\n", stats->ld_io);
2144 fprintf(out, " Loads - Miss : %10d\n", stats->ld_miss);
2145 fprintf(out, " Loads - no mapping : %10d\n", stats->ld_noadrs);
2146 fprintf(out, " Load Fill Buffer Hit : %10d\n", stats->ld_fbhit);
2147 fprintf(out, " Load L1D hit : %10d\n", stats->ld_l1hit);
2148 fprintf(out, " Load L2D hit : %10d\n", stats->ld_l2hit);
2149 fprintf(out, " Load LLC hit : %10d\n", stats->ld_llchit + stats->lcl_hitm);
2150 fprintf(out, " Load Local HITM : %10d\n", stats->lcl_hitm);
2151 fprintf(out, " Load Remote HITM : %10d\n", stats->rmt_hitm);
2152 fprintf(out, " Load Remote HIT : %10d\n", stats->rmt_hit);
2153 fprintf(out, " Load Local DRAM : %10d\n", stats->lcl_dram);
2154 fprintf(out, " Load Remote DRAM : %10d\n", stats->rmt_dram);
2155 fprintf(out, " Load MESI State Exclusive : %10d\n", stats->ld_excl);
2156 fprintf(out, " Load MESI State Shared : %10d\n", stats->ld_shared);
2157 fprintf(out, " Load LLC Misses : %10d\n", llc_misses);
2158 fprintf(out, " LLC Misses to Local DRAM : %10.1f%%\n", ((double)stats->lcl_dram/(double)llc_misses) * 100.);
2159 fprintf(out, " LLC Misses to Remote DRAM : %10.1f%%\n", ((double)stats->rmt_dram/(double)llc_misses) * 100.);
2160 fprintf(out, " LLC Misses to Remote cache (HIT) : %10.1f%%\n", ((double)stats->rmt_hit /(double)llc_misses) * 100.);
2161 fprintf(out, " LLC Misses to Remote cache (HITM) : %10.1f%%\n", ((double)stats->rmt_hitm/(double)llc_misses) * 100.);
2162 fprintf(out, " Store Operations : %10d\n", stats->store);
2163 fprintf(out, " Store - uncacheable : %10d\n", stats->st_uncache);
2164 fprintf(out, " Store - no mapping : %10d\n", stats->st_noadrs);
2165 fprintf(out, " Store L1D Hit : %10d\n", stats->st_l1hit);
2166 fprintf(out, " Store L1D Miss : %10d\n", stats->st_l1miss);
2167 fprintf(out, " No Page Map Rejects : %10d\n", stats->nomap);
2168 fprintf(out, " Unable to parse data source : %10d\n", stats->noparse);
2169}
2170
2171static void print_shared_cacheline_info(FILE *out)
2172{
2173 struct c2c_stats *stats = &c2c.hitm_stats;
2174 int hitm_cnt = stats->lcl_hitm + stats->rmt_hitm;
2175
2176 fprintf(out, "=================================================\n");
2177 fprintf(out, " Global Shared Cache Line Event Information \n");
2178 fprintf(out, "=================================================\n");
2179 fprintf(out, " Total Shared Cache Lines : %10d\n", c2c.shared_clines);
2180 fprintf(out, " Load HITs on shared lines : %10d\n", stats->load);
2181 fprintf(out, " Fill Buffer Hits on shared lines : %10d\n", stats->ld_fbhit);
2182 fprintf(out, " L1D hits on shared lines : %10d\n", stats->ld_l1hit);
2183 fprintf(out, " L2D hits on shared lines : %10d\n", stats->ld_l2hit);
2184 fprintf(out, " LLC hits on shared lines : %10d\n", stats->ld_llchit + stats->lcl_hitm);
2185 fprintf(out, " Locked Access on shared lines : %10d\n", stats->locks);
2186 fprintf(out, " Store HITs on shared lines : %10d\n", stats->store);
2187 fprintf(out, " Store L1D hits on shared lines : %10d\n", stats->st_l1hit);
2188 fprintf(out, " Total Merged records : %10d\n", hitm_cnt + stats->store);
2189}
2190
2191static void print_cacheline(struct c2c_hists *c2c_hists,
2192 struct hist_entry *he_cl,
2193 struct perf_hpp_list *hpp_list,
2194 FILE *out)
2195{
2196 char bf[1000];
2197 struct perf_hpp hpp = {
2198 .buf = bf,
2199 .size = 1000,
2200 };
2201 static bool once;
2202
2203 if (!once) {
2204 hists__fprintf_headers(&c2c_hists->hists, out);
2205 once = true;
2206 } else {
2207 fprintf(out, "\n");
2208 }
2209
2210 fprintf(out, " -------------------------------------------------------------\n");
2211 __hist_entry__snprintf(he_cl, &hpp, hpp_list);
2212 fprintf(out, "%s\n", bf);
2213 fprintf(out, " -------------------------------------------------------------\n");
2214
2215 hists__fprintf(&c2c_hists->hists, false, 0, 0, 0, out, false);
2216}
2217
2218static void print_pareto(FILE *out)
2219{
2220 struct perf_hpp_list hpp_list;
2221 struct rb_node *nd;
2222 int ret;
2223
2224 perf_hpp_list__init(&hpp_list);
2225 ret = hpp_list__parse(&hpp_list,
2226 "cl_num,"
2227 "cl_rmt_hitm,"
2228 "cl_lcl_hitm,"
2229 "cl_stores_l1hit,"
2230 "cl_stores_l1miss,"
2231 "dcacheline",
2232 NULL);
2233
2234 if (WARN_ONCE(ret, "failed to setup sort entries\n"))
2235 return;
2236
2237 nd = rb_first_cached(&c2c.hists.hists.entries);
2238
2239 for (; nd; nd = rb_next(nd)) {
2240 struct hist_entry *he = rb_entry(nd, struct hist_entry, rb_node);
2241 struct c2c_hist_entry *c2c_he;
2242
2243 if (he->filtered)
2244 continue;
2245
2246 c2c_he = container_of(he, struct c2c_hist_entry, he);
2247 print_cacheline(c2c_he->hists, he, &hpp_list, out);
2248 }
2249}
2250
2251static void print_c2c_info(FILE *out, struct perf_session *session)
2252{
2253 struct evlist *evlist = session->evlist;
2254 struct evsel *evsel;
2255 bool first = true;
2256
2257 fprintf(out, "=================================================\n");
2258 fprintf(out, " c2c details \n");
2259 fprintf(out, "=================================================\n");
2260
2261 evlist__for_each_entry(evlist, evsel) {
2262 fprintf(out, "%-36s: %s\n", first ? " Events" : "", evsel__name(evsel));
2263 first = false;
2264 }
2265 fprintf(out, " Cachelines sort on : %s HITMs\n",
2266 display_str[c2c.display]);
2267 fprintf(out, " Cacheline data grouping : %s\n", c2c.cl_sort);
2268}
2269
2270static void perf_c2c__hists_fprintf(FILE *out, struct perf_session *session)
2271{
2272 setup_pager();
2273
2274 print_c2c__display_stats(out);
2275 fprintf(out, "\n");
2276 print_shared_cacheline_info(out);
2277 fprintf(out, "\n");
2278 print_c2c_info(out, session);
2279
2280 if (c2c.stats_only)
2281 return;
2282
2283 fprintf(out, "\n");
2284 fprintf(out, "=================================================\n");
2285 fprintf(out, " Shared Data Cache Line Table \n");
2286 fprintf(out, "=================================================\n");
2287 fprintf(out, "#\n");
2288
2289 hists__fprintf(&c2c.hists.hists, true, 0, 0, 0, stdout, true);
2290
2291 fprintf(out, "\n");
2292 fprintf(out, "=================================================\n");
2293 fprintf(out, " Shared Cache Line Distribution Pareto \n");
2294 fprintf(out, "=================================================\n");
2295 fprintf(out, "#\n");
2296
2297 print_pareto(out);
2298}
2299
2300#ifdef HAVE_SLANG_SUPPORT
2301static void c2c_browser__update_nr_entries(struct hist_browser *hb)
2302{
2303 u64 nr_entries = 0;
2304 struct rb_node *nd = rb_first_cached(&hb->hists->entries);
2305
2306 while (nd) {
2307 struct hist_entry *he = rb_entry(nd, struct hist_entry, rb_node);
2308
2309 if (!he->filtered)
2310 nr_entries++;
2311
2312 nd = rb_next(nd);
2313 }
2314
2315 hb->nr_non_filtered_entries = nr_entries;
2316}
2317
2318struct c2c_cacheline_browser {
2319 struct hist_browser hb;
2320 struct hist_entry *he;
2321};
2322
2323static int
2324perf_c2c_cacheline_browser__title(struct hist_browser *browser,
2325 char *bf, size_t size)
2326{
2327 struct c2c_cacheline_browser *cl_browser;
2328 struct hist_entry *he;
2329 uint64_t addr = 0;
2330
2331 cl_browser = container_of(browser, struct c2c_cacheline_browser, hb);
2332 he = cl_browser->he;
2333
2334 if (he->mem_info)
2335 addr = cl_address(he->mem_info->daddr.addr);
2336
2337 scnprintf(bf, size, "Cacheline 0x%lx", addr);
2338 return 0;
2339}
2340
2341static struct c2c_cacheline_browser*
2342c2c_cacheline_browser__new(struct hists *hists, struct hist_entry *he)
2343{
2344 struct c2c_cacheline_browser *browser;
2345
2346 browser = zalloc(sizeof(*browser));
2347 if (browser) {
2348 hist_browser__init(&browser->hb, hists);
2349 browser->hb.c2c_filter = true;
2350 browser->hb.title = perf_c2c_cacheline_browser__title;
2351 browser->he = he;
2352 }
2353
2354 return browser;
2355}
2356
2357static int perf_c2c__browse_cacheline(struct hist_entry *he)
2358{
2359 struct c2c_hist_entry *c2c_he;
2360 struct c2c_hists *c2c_hists;
2361 struct c2c_cacheline_browser *cl_browser;
2362 struct hist_browser *browser;
2363 int key = -1;
2364 static const char help[] =
2365 " ENTER Toggle callchains (if present) \n"
2366 " n Toggle Node details info \n"
2367 " s Toggle full length of symbol and source line columns \n"
2368 " q Return back to cacheline list \n";
2369
2370 if (!he)
2371 return 0;
2372
2373 /* Display compact version first. */
2374 c2c.symbol_full = false;
2375
2376 c2c_he = container_of(he, struct c2c_hist_entry, he);
2377 c2c_hists = c2c_he->hists;
2378
2379 cl_browser = c2c_cacheline_browser__new(&c2c_hists->hists, he);
2380 if (cl_browser == NULL)
2381 return -1;
2382
2383 browser = &cl_browser->hb;
2384
2385 /* reset abort key so that it can get Ctrl-C as a key */
2386 SLang_reset_tty();
2387 SLang_init_tty(0, 0, 0);
2388
2389 c2c_browser__update_nr_entries(browser);
2390
2391 while (1) {
2392 key = hist_browser__run(browser, "? - help", true, 0);
2393
2394 switch (key) {
2395 case 's':
2396 c2c.symbol_full = !c2c.symbol_full;
2397 break;
2398 case 'n':
2399 c2c.node_info = (c2c.node_info + 1) % 3;
2400 setup_nodes_header();
2401 break;
2402 case 'q':
2403 goto out;
2404 case '?':
2405 ui_browser__help_window(&browser->b, help);
2406 break;
2407 default:
2408 break;
2409 }
2410 }
2411
2412out:
2413 free(cl_browser);
2414 return 0;
2415}
2416
2417static int perf_c2c_browser__title(struct hist_browser *browser,
2418 char *bf, size_t size)
2419{
2420 scnprintf(bf, size,
2421 "Shared Data Cache Line Table "
2422 "(%lu entries, sorted on %s HITMs)",
2423 browser->nr_non_filtered_entries,
2424 display_str[c2c.display]);
2425 return 0;
2426}
2427
2428static struct hist_browser*
2429perf_c2c_browser__new(struct hists *hists)
2430{
2431 struct hist_browser *browser = hist_browser__new(hists);
2432
2433 if (browser) {
2434 browser->title = perf_c2c_browser__title;
2435 browser->c2c_filter = true;
2436 }
2437
2438 return browser;
2439}
2440
2441static int perf_c2c__hists_browse(struct hists *hists)
2442{
2443 struct hist_browser *browser;
2444 int key = -1;
2445 static const char help[] =
2446 " d Display cacheline details \n"
2447 " ENTER Toggle callchains (if present) \n"
2448 " q Quit \n";
2449
2450 browser = perf_c2c_browser__new(hists);
2451 if (browser == NULL)
2452 return -1;
2453
2454 /* reset abort key so that it can get Ctrl-C as a key */
2455 SLang_reset_tty();
2456 SLang_init_tty(0, 0, 0);
2457
2458 c2c_browser__update_nr_entries(browser);
2459
2460 while (1) {
2461 key = hist_browser__run(browser, "? - help", true, 0);
2462
2463 switch (key) {
2464 case 'q':
2465 goto out;
2466 case 'd':
2467 perf_c2c__browse_cacheline(browser->he_selection);
2468 break;
2469 case '?':
2470 ui_browser__help_window(&browser->b, help);
2471 break;
2472 default:
2473 break;
2474 }
2475 }
2476
2477out:
2478 hist_browser__delete(browser);
2479 return 0;
2480}
2481
2482static void perf_c2c_display(struct perf_session *session)
2483{
2484 if (use_browser == 0)
2485 perf_c2c__hists_fprintf(stdout, session);
2486 else
2487 perf_c2c__hists_browse(&c2c.hists.hists);
2488}
2489#else
2490static void perf_c2c_display(struct perf_session *session)
2491{
2492 use_browser = 0;
2493 perf_c2c__hists_fprintf(stdout, session);
2494}
2495#endif /* HAVE_SLANG_SUPPORT */
2496
2497static char *fill_line(const char *orig, int len)
2498{
2499 int i, j, olen = strlen(orig);
2500 char *buf;
2501
2502 buf = zalloc(len + 1);
2503 if (!buf)
2504 return NULL;
2505
2506 j = len / 2 - olen / 2;
2507
2508 for (i = 0; i < j - 1; i++)
2509 buf[i] = '-';
2510
2511 buf[i++] = ' ';
2512
2513 strcpy(buf + i, orig);
2514
2515 i += olen;
2516
2517 buf[i++] = ' ';
2518
2519 for (; i < len; i++)
2520 buf[i] = '-';
2521
2522 return buf;
2523}
2524
2525static int ui_quirks(void)
2526{
2527 const char *nodestr = "Data address";
2528 char *buf;
2529
2530 if (!c2c.use_stdio) {
2531 dim_offset.width = 5;
2532 dim_offset.header = header_offset_tui;
2533 nodestr = "CL";
2534 }
2535
2536 dim_percent_hitm.header = percent_hitm_header[c2c.display];
2537
2538 /* Fix the zero line for dcacheline column. */
2539 buf = fill_line("Cacheline", dim_dcacheline.width +
2540 dim_dcacheline_node.width +
2541 dim_dcacheline_count.width + 4);
2542 if (!buf)
2543 return -ENOMEM;
2544
2545 dim_dcacheline.header.line[0].text = buf;
2546
2547 /* Fix the zero line for offset column. */
2548 buf = fill_line(nodestr, dim_offset.width +
2549 dim_offset_node.width +
2550 dim_dcacheline_count.width + 4);
2551 if (!buf)
2552 return -ENOMEM;
2553
2554 dim_offset.header.line[0].text = buf;
2555
2556 return 0;
2557}
2558
2559#define CALLCHAIN_DEFAULT_OPT "graph,0.5,caller,function,percent"
2560
2561const char callchain_help[] = "Display call graph (stack chain/backtrace):\n\n"
2562 CALLCHAIN_REPORT_HELP
2563 "\n\t\t\t\tDefault: " CALLCHAIN_DEFAULT_OPT;
2564
2565static int
2566parse_callchain_opt(const struct option *opt, const char *arg, int unset)
2567{
2568 struct callchain_param *callchain = opt->value;
2569
2570 callchain->enabled = !unset;
2571 /*
2572 * --no-call-graph
2573 */
2574 if (unset) {
2575 symbol_conf.use_callchain = false;
2576 callchain->mode = CHAIN_NONE;
2577 return 0;
2578 }
2579
2580 return parse_callchain_report_opt(arg);
2581}
2582
2583static int setup_callchain(struct evlist *evlist)
2584{
2585 u64 sample_type = evlist__combined_sample_type(evlist);
2586 enum perf_call_graph_mode mode = CALLCHAIN_NONE;
2587
2588 if ((sample_type & PERF_SAMPLE_REGS_USER) &&
2589 (sample_type & PERF_SAMPLE_STACK_USER)) {
2590 mode = CALLCHAIN_DWARF;
2591 dwarf_callchain_users = true;
2592 } else if (sample_type & PERF_SAMPLE_BRANCH_STACK)
2593 mode = CALLCHAIN_LBR;
2594 else if (sample_type & PERF_SAMPLE_CALLCHAIN)
2595 mode = CALLCHAIN_FP;
2596
2597 if (!callchain_param.enabled &&
2598 callchain_param.mode != CHAIN_NONE &&
2599 mode != CALLCHAIN_NONE) {
2600 symbol_conf.use_callchain = true;
2601 if (callchain_register_param(&callchain_param) < 0) {
2602 ui__error("Can't register callchain params.\n");
2603 return -EINVAL;
2604 }
2605 }
2606
2607 if (c2c.stitch_lbr && (mode != CALLCHAIN_LBR)) {
2608 ui__warning("Can't find LBR callchain. Switch off --stitch-lbr.\n"
2609 "Please apply --call-graph lbr when recording.\n");
2610 c2c.stitch_lbr = false;
2611 }
2612
2613 callchain_param.record_mode = mode;
2614 callchain_param.min_percent = 0;
2615 return 0;
2616}
2617
2618static int setup_display(const char *str)
2619{
2620 const char *display = str ?: "tot";
2621
2622 if (!strcmp(display, "tot"))
2623 c2c.display = DISPLAY_TOT;
2624 else if (!strcmp(display, "rmt"))
2625 c2c.display = DISPLAY_RMT;
2626 else if (!strcmp(display, "lcl"))
2627 c2c.display = DISPLAY_LCL;
2628 else {
2629 pr_err("failed: unknown display type: %s\n", str);
2630 return -1;
2631 }
2632
2633 return 0;
2634}
2635
2636#define for_each_token(__tok, __buf, __sep, __tmp) \
2637 for (__tok = strtok_r(__buf, __sep, &__tmp); __tok; \
2638 __tok = strtok_r(NULL, __sep, &__tmp))
2639
2640static int build_cl_output(char *cl_sort, bool no_source)
2641{
2642 char *tok, *tmp, *buf = strdup(cl_sort);
2643 bool add_pid = false;
2644 bool add_tid = false;
2645 bool add_iaddr = false;
2646 bool add_sym = false;
2647 bool add_dso = false;
2648 bool add_src = false;
2649 int ret = 0;
2650
2651 if (!buf)
2652 return -ENOMEM;
2653
2654 for_each_token(tok, buf, ",", tmp) {
2655 if (!strcmp(tok, "tid")) {
2656 add_tid = true;
2657 } else if (!strcmp(tok, "pid")) {
2658 add_pid = true;
2659 } else if (!strcmp(tok, "iaddr")) {
2660 add_iaddr = true;
2661 add_sym = true;
2662 add_dso = true;
2663 add_src = no_source ? false : true;
2664 } else if (!strcmp(tok, "dso")) {
2665 add_dso = true;
2666 } else if (strcmp(tok, "offset")) {
2667 pr_err("unrecognized sort token: %s\n", tok);
2668 ret = -EINVAL;
2669 goto err;
2670 }
2671 }
2672
2673 if (asprintf(&c2c.cl_output,
2674 "%s%s%s%s%s%s%s%s%s%s",
2675 c2c.use_stdio ? "cl_num_empty," : "",
2676 "percent_rmt_hitm,"
2677 "percent_lcl_hitm,"
2678 "percent_stores_l1hit,"
2679 "percent_stores_l1miss,"
2680 "offset,offset_node,dcacheline_count,",
2681 add_pid ? "pid," : "",
2682 add_tid ? "tid," : "",
2683 add_iaddr ? "iaddr," : "",
2684 "mean_rmt,"
2685 "mean_lcl,"
2686 "mean_load,"
2687 "tot_recs,"
2688 "cpucnt,",
2689 add_sym ? "symbol," : "",
2690 add_dso ? "dso," : "",
2691 add_src ? "cl_srcline," : "",
2692 "node") < 0) {
2693 ret = -ENOMEM;
2694 goto err;
2695 }
2696
2697 c2c.show_src = add_src;
2698err:
2699 free(buf);
2700 return ret;
2701}
2702
2703static int setup_coalesce(const char *coalesce, bool no_source)
2704{
2705 const char *c = coalesce ?: coalesce_default;
2706
2707 if (asprintf(&c2c.cl_sort, "offset,%s", c) < 0)
2708 return -ENOMEM;
2709
2710 if (build_cl_output(c2c.cl_sort, no_source))
2711 return -1;
2712
2713 if (asprintf(&c2c.cl_resort, "offset,%s",
2714 c2c.display == DISPLAY_TOT ?
2715 "tot_hitm" :
2716 c2c.display == DISPLAY_RMT ?
2717 "rmt_hitm,lcl_hitm" :
2718 "lcl_hitm,rmt_hitm") < 0)
2719 return -ENOMEM;
2720
2721 pr_debug("coalesce sort fields: %s\n", c2c.cl_sort);
2722 pr_debug("coalesce resort fields: %s\n", c2c.cl_resort);
2723 pr_debug("coalesce output fields: %s\n", c2c.cl_output);
2724 return 0;
2725}
2726
2727static int perf_c2c__report(int argc, const char **argv)
2728{
2729 struct perf_session *session;
2730 struct ui_progress prog;
2731 struct perf_data data = {
2732 .mode = PERF_DATA_MODE_READ,
2733 };
2734 char callchain_default_opt[] = CALLCHAIN_DEFAULT_OPT;
2735 const char *display = NULL;
2736 const char *coalesce = NULL;
2737 bool no_source = false;
2738 const struct option options[] = {
2739 OPT_STRING('k', "vmlinux", &symbol_conf.vmlinux_name,
2740 "file", "vmlinux pathname"),
2741 OPT_STRING('i', "input", &input_name, "file",
2742 "the input file to process"),
2743 OPT_INCR('N', "node-info", &c2c.node_info,
2744 "show extra node info in report (repeat for more info)"),
2745#ifdef HAVE_SLANG_SUPPORT
2746 OPT_BOOLEAN(0, "stdio", &c2c.use_stdio, "Use the stdio interface"),
2747#endif
2748 OPT_BOOLEAN(0, "stats", &c2c.stats_only,
2749 "Display only statistic tables (implies --stdio)"),
2750 OPT_BOOLEAN(0, "full-symbols", &c2c.symbol_full,
2751 "Display full length of symbols"),
2752 OPT_BOOLEAN(0, "no-source", &no_source,
2753 "Do not display Source Line column"),
2754 OPT_BOOLEAN(0, "show-all", &c2c.show_all,
2755 "Show all captured HITM lines."),
2756 OPT_CALLBACK_DEFAULT('g', "call-graph", &callchain_param,
2757 "print_type,threshold[,print_limit],order,sort_key[,branch],value",
2758 callchain_help, &parse_callchain_opt,
2759 callchain_default_opt),
2760 OPT_STRING('d', "display", &display, "Switch HITM output type", "lcl,rmt"),
2761 OPT_STRING('c', "coalesce", &coalesce, "coalesce fields",
2762 "coalesce fields: pid,tid,iaddr,dso"),
2763 OPT_BOOLEAN('f', "force", &symbol_conf.force, "don't complain, do it"),
2764 OPT_BOOLEAN(0, "stitch-lbr", &c2c.stitch_lbr,
2765 "Enable LBR callgraph stitching approach"),
2766 OPT_PARENT(c2c_options),
2767 OPT_END()
2768 };
2769 int err = 0;
2770
2771 argc = parse_options(argc, argv, options, report_c2c_usage,
2772 PARSE_OPT_STOP_AT_NON_OPTION);
2773 if (argc)
2774 usage_with_options(report_c2c_usage, options);
2775
2776 if (c2c.stats_only)
2777 c2c.use_stdio = true;
2778
2779 if (!input_name || !strlen(input_name))
2780 input_name = "perf.data";
2781
2782 data.path = input_name;
2783 data.force = symbol_conf.force;
2784
2785 err = setup_display(display);
2786 if (err)
2787 goto out;
2788
2789 err = setup_coalesce(coalesce, no_source);
2790 if (err) {
2791 pr_debug("Failed to initialize hists\n");
2792 goto out;
2793 }
2794
2795 err = c2c_hists__init(&c2c.hists, "dcacheline", 2);
2796 if (err) {
2797 pr_debug("Failed to initialize hists\n");
2798 goto out;
2799 }
2800
2801 session = perf_session__new(&data, 0, &c2c.tool);
2802 if (IS_ERR(session)) {
2803 err = PTR_ERR(session);
2804 pr_debug("Error creating perf session\n");
2805 goto out;
2806 }
2807
2808 err = setup_nodes(session);
2809 if (err) {
2810 pr_err("Failed setup nodes\n");
2811 goto out;
2812 }
2813
2814 err = mem2node__init(&c2c.mem2node, &session->header.env);
2815 if (err)
2816 goto out_session;
2817
2818 err = setup_callchain(session->evlist);
2819 if (err)
2820 goto out_mem2node;
2821
2822 if (symbol__init(&session->header.env) < 0)
2823 goto out_mem2node;
2824
2825 /* No pipe support at the moment. */
2826 if (perf_data__is_pipe(session->data)) {
2827 pr_debug("No pipe support at the moment.\n");
2828 goto out_mem2node;
2829 }
2830
2831 if (c2c.use_stdio)
2832 use_browser = 0;
2833 else
2834 use_browser = 1;
2835
2836 setup_browser(false);
2837
2838 err = perf_session__process_events(session);
2839 if (err) {
2840 pr_err("failed to process sample\n");
2841 goto out_mem2node;
2842 }
2843
2844 c2c_hists__reinit(&c2c.hists,
2845 "cl_idx,"
2846 "dcacheline,"
2847 "dcacheline_node,"
2848 "dcacheline_count,"
2849 "tot_recs,"
2850 "percent_hitm,"
2851 "tot_hitm,lcl_hitm,rmt_hitm,"
2852 "stores,stores_l1hit,stores_l1miss,"
2853 "dram_lcl,dram_rmt,"
2854 "ld_llcmiss,"
2855 "tot_loads,"
2856 "ld_fbhit,ld_l1hit,ld_l2hit,"
2857 "ld_lclhit,ld_rmthit",
2858 c2c.display == DISPLAY_TOT ? "tot_hitm" :
2859 c2c.display == DISPLAY_LCL ? "lcl_hitm" : "rmt_hitm"
2860 );
2861
2862 ui_progress__init(&prog, c2c.hists.hists.nr_entries, "Sorting...");
2863
2864 hists__collapse_resort(&c2c.hists.hists, NULL);
2865 hists__output_resort_cb(&c2c.hists.hists, &prog, resort_hitm_cb);
2866 hists__iterate_cb(&c2c.hists.hists, resort_cl_cb);
2867
2868 ui_progress__finish();
2869
2870 if (ui_quirks()) {
2871 pr_err("failed to setup UI\n");
2872 goto out_mem2node;
2873 }
2874
2875 perf_c2c_display(session);
2876
2877out_mem2node:
2878 mem2node__exit(&c2c.mem2node);
2879out_session:
2880 perf_session__delete(session);
2881out:
2882 return err;
2883}
2884
2885static int parse_record_events(const struct option *opt,
2886 const char *str, int unset __maybe_unused)
2887{
2888 bool *event_set = (bool *) opt->value;
2889
2890 if (!strcmp(str, "list")) {
2891 perf_mem_events__list();
2892 exit(0);
2893 }
2894 if (perf_mem_events__parse(str))
2895 exit(-1);
2896
2897 *event_set = true;
2898 return 0;
2899}
2900
2901
2902static const char * const __usage_record[] = {
2903 "perf c2c record [<options>] [<command>]",
2904 "perf c2c record [<options>] -- <command> [<options>]",
2905 NULL
2906};
2907
2908static const char * const *record_mem_usage = __usage_record;
2909
2910static int perf_c2c__record(int argc, const char **argv)
2911{
2912 int rec_argc, i = 0, j;
2913 const char **rec_argv;
2914 int ret;
2915 bool all_user = false, all_kernel = false;
2916 bool event_set = false;
2917 struct option options[] = {
2918 OPT_CALLBACK('e', "event", &event_set, "event",
2919 "event selector. Use 'perf mem record -e list' to list available events",
2920 parse_record_events),
2921 OPT_BOOLEAN('u', "all-user", &all_user, "collect only user level data"),
2922 OPT_BOOLEAN('k', "all-kernel", &all_kernel, "collect only kernel level data"),
2923 OPT_UINTEGER('l', "ldlat", &perf_mem_events__loads_ldlat, "setup mem-loads latency"),
2924 OPT_PARENT(c2c_options),
2925 OPT_END()
2926 };
2927
2928 if (perf_mem_events__init()) {
2929 pr_err("failed: memory events not supported\n");
2930 return -1;
2931 }
2932
2933 argc = parse_options(argc, argv, options, record_mem_usage,
2934 PARSE_OPT_KEEP_UNKNOWN);
2935
2936 rec_argc = argc + 11; /* max number of arguments */
2937 rec_argv = calloc(rec_argc + 1, sizeof(char *));
2938 if (!rec_argv)
2939 return -1;
2940
2941 rec_argv[i++] = "record";
2942
2943 if (!event_set) {
2944 perf_mem_events[PERF_MEM_EVENTS__LOAD].record = true;
2945 perf_mem_events[PERF_MEM_EVENTS__STORE].record = true;
2946 }
2947
2948 if (perf_mem_events[PERF_MEM_EVENTS__LOAD].record)
2949 rec_argv[i++] = "-W";
2950
2951 rec_argv[i++] = "-d";
2952 rec_argv[i++] = "--phys-data";
2953 rec_argv[i++] = "--sample-cpu";
2954
2955 for (j = 0; j < PERF_MEM_EVENTS__MAX; j++) {
2956 if (!perf_mem_events[j].record)
2957 continue;
2958
2959 if (!perf_mem_events[j].supported) {
2960 pr_err("failed: event '%s' not supported\n",
2961 perf_mem_events[j].name);
2962 free(rec_argv);
2963 return -1;
2964 }
2965
2966 rec_argv[i++] = "-e";
2967 rec_argv[i++] = perf_mem_events__name(j);
2968 }
2969
2970 if (all_user)
2971 rec_argv[i++] = "--all-user";
2972
2973 if (all_kernel)
2974 rec_argv[i++] = "--all-kernel";
2975
2976 for (j = 0; j < argc; j++, i++)
2977 rec_argv[i] = argv[j];
2978
2979 if (verbose > 0) {
2980 pr_debug("calling: ");
2981
2982 j = 0;
2983
2984 while (rec_argv[j]) {
2985 pr_debug("%s ", rec_argv[j]);
2986 j++;
2987 }
2988 pr_debug("\n");
2989 }
2990
2991 ret = cmd_record(i, rec_argv);
2992 free(rec_argv);
2993 return ret;
2994}
2995
2996int cmd_c2c(int argc, const char **argv)
2997{
2998 argc = parse_options(argc, argv, c2c_options, c2c_usage,
2999 PARSE_OPT_STOP_AT_NON_OPTION);
3000
3001 if (!argc)
3002 usage_with_options(c2c_usage, c2c_options);
3003
3004 if (!strncmp(argv[0], "rec", 3)) {
3005 return perf_c2c__record(argc, argv);
3006 } else if (!strncmp(argv[0], "rep", 3)) {
3007 return perf_c2c__report(argc, argv);
3008 } else {
3009 usage_with_options(c2c_usage, c2c_options);
3010 }
3011
3012 return 0;
3013}