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
46 struct c2c_hists {
47 struct hists hists;
48 struct perf_hpp_list list;
49 struct c2c_stats stats;
50 };
51
52 struct compute_stats {
53 struct stats lcl_hitm;
54 struct stats rmt_hitm;
55 struct stats load;
56 };
57
58 struct 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
80 static char const *coalesce_default = "iaddr";
81
82 struct 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
111 enum {
112 DISPLAY_LCL,
113 DISPLAY_RMT,
114 DISPLAY_TOT,
115 DISPLAY_MAX,
116 };
117
118 static const char *display_str[DISPLAY_MAX] = {
119 [DISPLAY_LCL] = "Local",
120 [DISPLAY_RMT] = "Remote",
121 [DISPLAY_TOT] = "Total",
122 };
123
124 static const struct option c2c_options[] = {
125 OPT_INCR('v', "verbose", &verbose, "be more verbose (show counter open errors, etc)"),
126 OPT_END()
127 };
128
129 static struct perf_c2c c2c;
130
c2c_he_zalloc(size_t size)131 static 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
c2c_he_free(void * he)158 static 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
175 static struct hist_entry_ops c2c_entry_ops = {
176 .new = c2c_he_zalloc,
177 .free = c2c_he_free,
178 };
179
180 static int c2c_hists__init(struct c2c_hists *hists,
181 const char *sort,
182 int nr_header_lines);
183
184 static struct c2c_hists*
he__get_c2c_hists(struct hist_entry * he,const char * sort,int nr_header_lines)185 he__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
c2c_he__set_cpu(struct c2c_hist_entry * c2c_he,struct perf_sample * sample)210 static 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
c2c_he__set_node(struct c2c_hist_entry * c2c_he,struct perf_sample * sample)220 static 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
compute_stats(struct c2c_hist_entry * c2c_he,struct c2c_stats * stats,u64 weight)242 static 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
process_sample_event(struct perf_tool * tool __maybe_unused,union perf_event * event,struct perf_sample * sample,struct evsel * evsel,struct machine * machine)256 static 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
348 out:
349 addr_location__put(&al);
350 return ret;
351
352 free_mi:
353 mem_info__put(mi_dup);
354 mem_info__put(mi);
355 ret = -ENOMEM;
356 goto out;
357 }
358
359 static 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
373 static const char * const c2c_usage[] = {
374 "perf c2c {record|report}",
375 NULL
376 };
377
378 static const char * const __usage_report[] = {
379 "perf c2c report",
380 NULL
381 };
382
383 static const char * const *report_c2c_usage = __usage_report;
384
385 #define C2C_HEADER_MAX 2
386
387 struct c2c_header {
388 struct {
389 const char *text;
390 int span;
391 } line[C2C_HEADER_MAX];
392 };
393
394 struct 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
408 struct c2c_fmt {
409 struct perf_hpp_fmt fmt;
410 struct c2c_dimension *dim;
411 };
412
413 #define SYMBOL_WIDTH 30
414
415 static struct c2c_dimension dim_symbol;
416 static struct c2c_dimension dim_srcline;
417
symbol_width(struct hists * hists,struct sort_entry * se)418 static 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
c2c_width(struct perf_hpp_fmt * fmt,struct perf_hpp * hpp __maybe_unused,struct hists * hists)428 static 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
c2c_header(struct perf_hpp_fmt * fmt,struct perf_hpp * hpp,struct hists * hists,int line,int * span)445 static 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
485 static int64_t
dcacheline_cmp(struct perf_hpp_fmt * fmt __maybe_unused,struct hist_entry * left,struct hist_entry * right)486 dcacheline_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
dcacheline_entry(struct perf_hpp_fmt * fmt,struct perf_hpp * hpp,struct hist_entry * he)492 static 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
505 static int
dcacheline_node_entry(struct perf_hpp_fmt * fmt,struct perf_hpp * hpp,struct hist_entry * he)506 dcacheline_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
519 static int
dcacheline_node_count(struct perf_hpp_fmt * fmt,struct perf_hpp * hpp,struct hist_entry * he)520 dcacheline_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
offset_entry(struct perf_hpp_fmt * fmt,struct perf_hpp * hpp,struct hist_entry * he)530 static 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
543 static int64_t
offset_cmp(struct perf_hpp_fmt * fmt __maybe_unused,struct hist_entry * left,struct hist_entry * right)544 offset_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
557 static int
iaddr_entry(struct perf_hpp_fmt * fmt,struct perf_hpp * hpp,struct hist_entry * he)558 iaddr_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
571 static int64_t
iaddr_cmp(struct perf_hpp_fmt * fmt __maybe_unused,struct hist_entry * left,struct hist_entry * right)572 iaddr_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
578 static int
tot_hitm_entry(struct perf_hpp_fmt * fmt,struct perf_hpp * hpp,struct hist_entry * he)579 tot_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
592 static int64_t
tot_hitm_cmp(struct perf_hpp_fmt * fmt __maybe_unused,struct hist_entry * left,struct hist_entry * right)593 tot_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) \
611 static 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) \
624 static 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
639 STAT_FN(rmt_hitm)
STAT_FN(lcl_hitm)640 STAT_FN(lcl_hitm)
641 STAT_FN(store)
642 STAT_FN(st_l1hit)
643 STAT_FN(st_l1miss)
644 STAT_FN(ld_fbhit)
645 STAT_FN(ld_l1hit)
646 STAT_FN(ld_l2hit)
647 STAT_FN(ld_llchit)
648 STAT_FN(rmt_hit)
649
650 static 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
662 static int
ld_llcmiss_entry(struct perf_hpp_fmt * fmt,struct perf_hpp * hpp,struct hist_entry * he)663 ld_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
675 static int64_t
ld_llcmiss_cmp(struct perf_hpp_fmt * fmt __maybe_unused,struct hist_entry * left,struct hist_entry * right)676 ld_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
total_records(struct c2c_stats * stats)688 static 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
711 static int
tot_recs_entry(struct perf_hpp_fmt * fmt,struct perf_hpp * hpp,struct hist_entry * he)712 tot_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
725 static int64_t
tot_recs_cmp(struct perf_hpp_fmt * fmt __maybe_unused,struct hist_entry * left,struct hist_entry * right)726 tot_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
total_loads(struct c2c_stats * stats)743 static 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
762 static int
tot_loads_entry(struct perf_hpp_fmt * fmt,struct perf_hpp * hpp,struct hist_entry * he)763 tot_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
776 static int64_t
tot_loads_cmp(struct perf_hpp_fmt * fmt __maybe_unused,struct hist_entry * left,struct hist_entry * right)777 tot_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
794 typedef double (get_percent_cb)(struct c2c_hist_entry *);
795
796 static int
percent_color(struct perf_hpp_fmt * fmt,struct perf_hpp * hpp,struct hist_entry * he,get_percent_cb get_percent)797 percent_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
percent_hitm(struct c2c_hist_entry * c2c_he)814 static 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
853 static int
percent_hitm_entry(struct perf_hpp_fmt * fmt,struct perf_hpp * hpp,struct hist_entry * he)854 percent_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
867 static int
percent_hitm_color(struct perf_hpp_fmt * fmt,struct perf_hpp * hpp,struct hist_entry * he)868 percent_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
874 static int64_t
percent_hitm_cmp(struct perf_hpp_fmt * fmt __maybe_unused,struct hist_entry * left,struct hist_entry * right)875 percent_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
he_stats(struct hist_entry * he)892 static 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
total_stats(struct hist_entry * he)900 static 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
percent(int st,int tot)908 static 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) \
916 static 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
924 PERCENT_FN(rmt_hitm)
PERCENT_FN(lcl_hitm)925 PERCENT_FN(lcl_hitm)
926 PERCENT_FN(st_l1hit)
927 PERCENT_FN(st_l1miss)
928
929 static int
930 percent_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
940 static int
percent_rmt_hitm_color(struct perf_hpp_fmt * fmt,struct perf_hpp * hpp,struct hist_entry * he)941 percent_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
947 static int64_t
percent_rmt_hitm_cmp(struct perf_hpp_fmt * fmt __maybe_unused,struct hist_entry * left,struct hist_entry * right)948 percent_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
960 static int
percent_lcl_hitm_entry(struct perf_hpp_fmt * fmt,struct perf_hpp * hpp,struct hist_entry * he)961 percent_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
971 static int
percent_lcl_hitm_color(struct perf_hpp_fmt * fmt,struct perf_hpp * hpp,struct hist_entry * he)972 percent_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
978 static int64_t
percent_lcl_hitm_cmp(struct perf_hpp_fmt * fmt __maybe_unused,struct hist_entry * left,struct hist_entry * right)979 percent_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
991 static int
percent_stores_l1hit_entry(struct perf_hpp_fmt * fmt,struct perf_hpp * hpp,struct hist_entry * he)992 percent_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
1002 static int
percent_stores_l1hit_color(struct perf_hpp_fmt * fmt,struct perf_hpp * hpp,struct hist_entry * he)1003 percent_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
1009 static int64_t
percent_stores_l1hit_cmp(struct perf_hpp_fmt * fmt __maybe_unused,struct hist_entry * left,struct hist_entry * right)1010 percent_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
1022 static int
percent_stores_l1miss_entry(struct perf_hpp_fmt * fmt,struct perf_hpp * hpp,struct hist_entry * he)1023 percent_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
1033 static int
percent_stores_l1miss_color(struct perf_hpp_fmt * fmt,struct perf_hpp * hpp,struct hist_entry * he)1034 percent_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
1040 static int64_t
percent_stores_l1miss_cmp(struct perf_hpp_fmt * fmt __maybe_unused,struct hist_entry * left,struct hist_entry * right)1041 percent_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
1053 STAT_FN(lcl_dram)
STAT_FN(rmt_dram)1054 STAT_FN(rmt_dram)
1055
1056 static int
1057 pid_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
1065 static int64_t
pid_cmp(struct perf_hpp_fmt * fmt __maybe_unused,struct hist_entry * left,struct hist_entry * right)1066 pid_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
1072 static int64_t
empty_cmp(struct perf_hpp_fmt * fmt __maybe_unused,struct hist_entry * left __maybe_unused,struct hist_entry * right __maybe_unused)1073 empty_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
1080 static int
node_entry(struct perf_hpp_fmt * fmt __maybe_unused,struct perf_hpp * hpp,struct hist_entry * he)1081 node_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
1178 static int
mean_entry(struct perf_hpp_fmt * fmt,struct perf_hpp * hpp,struct hist_entry * he,double mean)1179 mean_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) \
1190 static 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
1198 MEAN_ENTRY(mean_rmt_entry, rmt_hitm);
1199 MEAN_ENTRY(mean_lcl_entry, lcl_hitm);
1200 MEAN_ENTRY(mean_load_entry, load);
1201
1202 static int
cpucnt_entry(struct perf_hpp_fmt * fmt,struct perf_hpp * hpp,struct hist_entry * he)1203 cpucnt_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
1216 static int
cl_idx_entry(struct perf_hpp_fmt * fmt,struct perf_hpp * hpp,struct hist_entry * he)1217 cl_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
1230 static int
cl_idx_empty_entry(struct perf_hpp_fmt * fmt,struct perf_hpp * hpp,struct hist_entry * he)1231 cl_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
1274 static 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
1282 static 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
1290 static 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
1298 static struct c2c_header header_offset_tui = HEADER_SPAN("-----", "Off", 2);
1299
1300 static 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
1308 static 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
1316 static 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
1324 static 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
1332 static 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
1340 static 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
1348 static 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
1356 static 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
1364 static 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
1372 static 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
1380 static 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
1388 static 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
1396 static 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
1404 static 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
1412 static 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
1420 static 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
1428 static 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
1436 static 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
1444 static 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
1452 static 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
1460 static 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
1468 static 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
1474 static 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
1482 static 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
1491 static 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
1500 static 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
1509 static 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
1518 static 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
1526 static 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
1534 static 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
1542 static struct c2c_dimension dim_tid = {
1543 .header = HEADER_LOW("Tid"),
1544 .name = "tid",
1545 .se = &sort_thread,
1546 };
1547
1548 static struct c2c_dimension dim_symbol = {
1549 .name = "symbol",
1550 .se = &sort_sym,
1551 };
1552
1553 static struct c2c_dimension dim_dso = {
1554 .header = HEADER_BOTH("Shared", "Object"),
1555 .name = "dso",
1556 .se = &sort_dso,
1557 };
1558
1559 static 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
1565 static struct c2c_dimension dim_node = {
1566 .name = "node",
1567 .cmp = empty_cmp,
1568 .entry = node_entry,
1569 .width = 4,
1570 };
1571
1572 static 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
1580 static 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
1588 static 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
1596 static 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
1604 static struct c2c_dimension dim_srcline = {
1605 .name = "cl_srcline",
1606 .se = &sort_srcline,
1607 };
1608
1609 static 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
1617 static 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
1625 static 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
1633 static 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
fmt_free(struct perf_hpp_fmt * fmt)1681 static 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
fmt_equal(struct perf_hpp_fmt * a,struct perf_hpp_fmt * b)1689 static 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
get_dimension(const char * name)1697 static 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
c2c_se_entry(struct perf_hpp_fmt * fmt,struct perf_hpp * hpp,struct hist_entry * he)1711 static 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
c2c_se_cmp(struct perf_hpp_fmt * fmt,struct hist_entry * a,struct hist_entry * b)1728 static 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
c2c_se_collapse(struct perf_hpp_fmt * fmt,struct hist_entry * a,struct hist_entry * b)1737 static 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
get_format(const char * name)1748 static 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
c2c_hists__init_output(struct perf_hpp_list * hpp_list,char * name)1780 static 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
c2c_hists__init_sort(struct perf_hpp_list * hpp_list,char * name)1793 static 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
hpp_list__parse(struct perf_hpp_list * hpp_list,const char * output_,const char * sort_)1832 static 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
c2c_hists__init(struct c2c_hists * hists,const char * sort,int nr_header_lines)1863 static 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
c2c_hists__reinit(struct c2c_hists * c2c_hists,const char * output,const char * sort)1882 static 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
he__display(struct hist_entry * he,struct c2c_stats * stats)1892 static 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
valid_hitm_or_store(struct hist_entry * he)1929 static 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
set_node_width(struct c2c_hist_entry * c2c_he,int len)1941 static 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
set_nodestr(struct c2c_hist_entry * c2c_he)1952 static 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
calc_width(struct c2c_hist_entry * c2c_he)1972 static 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
filter_cb(struct hist_entry * he,void * arg __maybe_unused)1981 static 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
resort_cl_cb(struct hist_entry * he,void * arg __maybe_unused)1998 static 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
setup_nodes_header(void)2022 static void setup_nodes_header(void)
2023 {
2024 dim_node.header = header_node[c2c.node_info];
2025 }
2026
setup_nodes(struct perf_session * session)2027 static 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
resort_hitm_cb(struct hist_entry * he,void * arg __maybe_unused)2089 static 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
hists__iterate_cb(struct hists * hists,hists__resort_cb_t cb)2102 static 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
print_c2c__display_stats(FILE * out)2120 static 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
print_shared_cacheline_info(FILE * out)2165 static 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
print_cacheline(struct c2c_hists * c2c_hists,struct hist_entry * he_cl,struct perf_hpp_list * hpp_list,FILE * out)2185 static 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
print_pareto(FILE * out)2212 static 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
print_c2c_info(FILE * out,struct perf_session * session)2245 static 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
perf_c2c__hists_fprintf(FILE * out,struct perf_session * session)2265 static 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
c2c_browser__update_nr_entries(struct hist_browser * hb)2296 static 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
2313 struct c2c_cacheline_browser {
2314 struct hist_browser hb;
2315 struct hist_entry *he;
2316 };
2317
2318 static int
perf_c2c_cacheline_browser__title(struct hist_browser * browser,char * bf,size_t size)2319 perf_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
2336 static struct c2c_cacheline_browser*
c2c_cacheline_browser__new(struct hists * hists,struct hist_entry * he)2337 c2c_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
perf_c2c__browse_cacheline(struct hist_entry * he)2352 static 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
2407 out:
2408 free(cl_browser);
2409 return 0;
2410 }
2411
perf_c2c_browser__title(struct hist_browser * browser,char * bf,size_t size)2412 static 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
2423 static struct hist_browser*
perf_c2c_browser__new(struct hists * hists)2424 perf_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
perf_c2c__hists_browse(struct hists * hists)2436 static 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
2472 out:
2473 hist_browser__delete(browser);
2474 return 0;
2475 }
2476
perf_c2c_display(struct perf_session * session)2477 static 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
perf_c2c_display(struct perf_session * session)2485 static 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
fill_line(const char * orig,int len)2492 static 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
ui_quirks(void)2520 static 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
2556 const 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
2560 static int
parse_callchain_opt(const struct option * opt,const char * arg,int unset)2561 parse_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
setup_callchain(struct evlist * evlist)2578 static 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
setup_display(const char * str)2607 static 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
build_cl_output(char * cl_sort,bool no_source)2629 static 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;
2687 err:
2688 free(buf);
2689 return ret;
2690 }
2691
setup_coalesce(const char * coalesce,bool no_source)2692 static 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
perf_c2c__report(int argc,const char ** argv)2716 static 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
2864 out_mem2node:
2865 mem2node__exit(&c2c.mem2node);
2866 out_session:
2867 perf_session__delete(session);
2868 out:
2869 return err;
2870 }
2871
parse_record_events(const struct option * opt,const char * str,int unset __maybe_unused)2872 static 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
2882 static const char * const __usage_record[] = {
2883 "perf c2c record [<options>] [<command>]",
2884 "perf c2c record [<options>] -- <command> [<options>]",
2885 NULL
2886 };
2887
2888 static const char * const *record_mem_usage = __usage_record;
2889
perf_c2c__record(int argc,const char ** argv)2890 static 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
cmd_c2c(int argc,const char ** argv)2976 int 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 }
2994