1 /*- 2 * Copyright (c) 2009, Fabien Thomas 3 * All rights reserved. 4 * 5 * Redistribution and use in source and binary forms, with or without 6 * modification, are permitted provided that the following conditions 7 * are met: 8 * 1. Redistributions of source code must retain the above copyright 9 * notice, this list of conditions and the following disclaimer. 10 * 2. Redistributions in binary form must reproduce the above copyright 11 * notice, this list of conditions and the following disclaimer in the 12 * documentation and/or other materials provided with the distribution. 13 * 14 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 15 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 16 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 17 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 18 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 19 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 20 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 21 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 22 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 23 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 24 * SUCH DAMAGE. 25 */ 26 27 /* 28 * Process hwpmc(4) samples as calltree. 29 * 30 * Output file format compatible with Kcachegrind (kdesdk). 31 * Handle top mode with a sorted tree display. 32 */ 33 34 #include <sys/cdefs.h> 35 __FBSDID("$FreeBSD$"); 36 37 #include <sys/param.h> 38 #include <sys/endian.h> 39 #include <sys/queue.h> 40 41 #include <assert.h> 42 #include <curses.h> 43 #include <ctype.h> 44 #include <err.h> 45 #include <errno.h> 46 #include <fcntl.h> 47 #include <pmc.h> 48 #include <pmclog.h> 49 #include <sysexits.h> 50 #include <stdint.h> 51 #include <stdio.h> 52 #include <stdlib.h> 53 #include <string.h> 54 #include <unistd.h> 55 #include <sysexits.h> 56 57 #include "pmcstat.h" 58 #include "pmcstat_log.h" 59 #include "pmcstat_top.h" 60 #include "pmcpl_calltree.h" 61 62 #define PMCPL_CT_GROWSIZE 4 63 64 static pmcstat_interned_string pmcpl_ct_prevfn; 65 66 static int pmcstat_skiplink = 0; 67 68 struct pmcpl_ct_node; 69 70 /* Get the sample value for PMC a. */ 71 #define PMCPL_CT_SAMPLE(a, b) \ 72 ((a) < (b)->npmcs ? (b)->sb[a] : 0) 73 74 /* Get the sample value in percent related to rsamples. */ 75 #define PMCPL_CT_SAMPLEP(a, b) \ 76 (PMCPL_CT_SAMPLE(a, b) * 100.0 / rsamples->sb[a]) 77 78 struct pmcpl_ct_sample { 79 int npmcs; /* Max pmc index available. */ 80 unsigned *sb; /* Sample buffer for 0..npmcs. */ 81 }; 82 83 struct pmcpl_ct_arc { 84 struct pmcpl_ct_sample pcta_samples; 85 struct pmcpl_ct_sample pcta_callid; 86 unsigned pcta_call; 87 struct pmcpl_ct_node *pcta_child; 88 }; 89 90 struct pmcpl_ct_instr { 91 uintfptr_t pctf_func; 92 struct pmcpl_ct_sample pctf_samples; 93 }; 94 95 /* 96 * Each calltree node is tracked by a pmcpl_ct_node struct. 97 */ 98 struct pmcpl_ct_node { 99 #define PMCPL_PCT_TAG 0x00000001 /* Loop detection. */ 100 uint32_t pct_flags; 101 struct pmcstat_image *pct_image; 102 uintfptr_t pct_func; 103 struct pmcpl_ct_sample pct_samples; 104 105 int pct_narc; 106 int pct_arc_c; 107 struct pmcpl_ct_arc *pct_arc; 108 109 /* TODO: optimize for large number of items. */ 110 int pct_ninstr; 111 int pct_instr_c; 112 struct pmcpl_ct_instr *pct_instr; 113 }; 114 115 struct pmcpl_ct_node_hash { 116 struct pmcpl_ct_node *pch_ctnode; 117 LIST_ENTRY(pmcpl_ct_node_hash) pch_next; 118 }; 119 120 struct pmcpl_ct_sample pmcpl_ct_callid; 121 122 #define PMCPL_CT_MAXCOL PMC_CALLCHAIN_DEPTH_MAX 123 #define PMCPL_CT_MAXLINE 256 124 struct pmcpl_ct_node *pmcpl_ct_topscreen[PMCPL_CT_MAXCOL][PMCPL_CT_MAXLINE]; 125 126 /* 127 * All nodes indexed by function/image name are placed in a hash table. 128 */ 129 static LIST_HEAD(,pmcpl_ct_node_hash) pmcpl_ct_node_hash[PMCSTAT_NHASH]; 130 131 /* 132 * Root node for the graph. 133 */ 134 static struct pmcpl_ct_node *pmcpl_ct_root; 135 136 /* 137 * Prototypes 138 */ 139 140 /* 141 * Initialize a samples. 142 */ 143 144 static void 145 pmcpl_ct_samples_init(struct pmcpl_ct_sample *samples) 146 { 147 148 samples->npmcs = 0; 149 samples->sb = NULL; 150 } 151 152 /* 153 * Free a samples. 154 */ 155 156 static void 157 pmcpl_ct_samples_free(struct pmcpl_ct_sample *samples) 158 { 159 160 samples->npmcs = 0; 161 free(samples->sb); 162 samples->sb = NULL; 163 } 164 165 /* 166 * Grow a sample block to store pmcstat_npmcs PMCs. 167 */ 168 169 static void 170 pmcpl_ct_samples_grow(struct pmcpl_ct_sample *samples) 171 { 172 int npmcs; 173 174 /* Enough storage. */ 175 if (pmcstat_npmcs <= samples->npmcs) 176 return; 177 178 npmcs = samples->npmcs + 179 max(pmcstat_npmcs - samples->npmcs, PMCPL_CT_GROWSIZE); 180 samples->sb = realloc(samples->sb, npmcs * sizeof(unsigned)); 181 if (samples->sb == NULL) 182 errx(EX_SOFTWARE, "ERROR: out of memory"); 183 bzero((char *)samples->sb + samples->npmcs * sizeof(unsigned), 184 (npmcs - samples->npmcs) * sizeof(unsigned)); 185 samples->npmcs = npmcs; 186 } 187 188 /* 189 * Compute the sum of all root arcs. 190 */ 191 192 static void 193 pmcpl_ct_samples_root(struct pmcpl_ct_sample *samples) 194 { 195 int i, pmcin; 196 197 pmcpl_ct_samples_init(samples); 198 pmcpl_ct_samples_grow(samples); 199 200 for (i = 0; i < pmcpl_ct_root->pct_narc; i++) 201 for (pmcin = 0; pmcin < pmcstat_npmcs; pmcin++) 202 samples->sb[pmcin] += PMCPL_CT_SAMPLE(pmcin, 203 &pmcpl_ct_root->pct_arc[i].pcta_samples); 204 } 205 206 /* 207 * Grow the arc table. 208 */ 209 210 static void 211 pmcpl_ct_arc_grow(int cursize, int *maxsize, struct pmcpl_ct_arc **items) 212 { 213 int nmaxsize; 214 215 if (cursize < *maxsize) 216 return; 217 218 nmaxsize = *maxsize + max(cursize + 1 - *maxsize, PMCPL_CT_GROWSIZE); 219 *items = realloc(*items, nmaxsize * sizeof(struct pmcpl_ct_arc)); 220 if (*items == NULL) 221 errx(EX_SOFTWARE, "ERROR: out of memory"); 222 bzero((char *)*items + *maxsize * sizeof(struct pmcpl_ct_arc), 223 (nmaxsize - *maxsize) * sizeof(struct pmcpl_ct_arc)); 224 *maxsize = nmaxsize; 225 } 226 227 /* 228 * Compare two arc by samples value. 229 */ 230 static int 231 pmcpl_ct_arc_compare(void *thunk, const void *a, const void *b) 232 { 233 const struct pmcpl_ct_arc *ct1, *ct2; 234 int pmcin = *(int *)thunk; 235 236 ct1 = (const struct pmcpl_ct_arc *) a; 237 ct2 = (const struct pmcpl_ct_arc *) b; 238 239 /* Sort in reverse order */ 240 if (PMCPL_CT_SAMPLE(pmcin, &ct1->pcta_samples) < 241 PMCPL_CT_SAMPLE(pmcin, &ct2->pcta_samples)) 242 return (1); 243 if (PMCPL_CT_SAMPLE(pmcin, &ct1->pcta_samples) > 244 PMCPL_CT_SAMPLE(pmcin, &ct2->pcta_samples)) 245 return (-1); 246 return (0); 247 } 248 249 /* 250 * Grow the instr table. 251 */ 252 253 static void 254 pmcpl_ct_instr_grow(int cursize, int *maxsize, struct pmcpl_ct_instr **items) 255 { 256 int nmaxsize; 257 258 if (cursize < *maxsize) 259 return; 260 261 nmaxsize = *maxsize + max(cursize + 1 - *maxsize, PMCPL_CT_GROWSIZE); 262 *items = realloc(*items, nmaxsize * sizeof(struct pmcpl_ct_instr)); 263 if (*items == NULL) 264 errx(EX_SOFTWARE, "ERROR: out of memory"); 265 bzero((char *)*items + *maxsize * sizeof(struct pmcpl_ct_instr), 266 (nmaxsize - *maxsize) * sizeof(struct pmcpl_ct_instr)); 267 *maxsize = nmaxsize; 268 } 269 270 /* 271 * Add a new instruction sample to given node. 272 */ 273 274 static void 275 pmcpl_ct_instr_add(struct pmcpl_ct_node *ct, int pmcin, uintfptr_t pc) 276 { 277 int i; 278 struct pmcpl_ct_instr *in; 279 280 for (i = 0; i<ct->pct_ninstr; i++) { 281 if (ct->pct_instr[i].pctf_func == pc) { 282 in = &ct->pct_instr[i]; 283 pmcpl_ct_samples_grow(&in->pctf_samples); 284 in->pctf_samples.sb[pmcin]++; 285 return; 286 } 287 } 288 289 pmcpl_ct_instr_grow(ct->pct_ninstr, &ct->pct_instr_c, &ct->pct_instr); 290 in = &ct->pct_instr[ct->pct_ninstr]; 291 in->pctf_func = pc; 292 pmcpl_ct_samples_init(&in->pctf_samples); 293 pmcpl_ct_samples_grow(&in->pctf_samples); 294 in->pctf_samples.sb[pmcin] = 1; 295 ct->pct_ninstr++; 296 } 297 298 /* 299 * Allocate a new node. 300 */ 301 302 static struct pmcpl_ct_node * 303 pmcpl_ct_node_allocate(struct pmcstat_image *image, uintfptr_t pc) 304 { 305 struct pmcpl_ct_node *ct; 306 307 if ((ct = malloc(sizeof(*ct))) == NULL) 308 err(EX_OSERR, "ERROR: Cannot allocate callgraph node"); 309 310 ct->pct_flags = 0; 311 ct->pct_image = image; 312 ct->pct_func = pc; 313 314 pmcpl_ct_samples_init(&ct->pct_samples); 315 316 ct->pct_narc = 0; 317 ct->pct_arc_c = 0; 318 ct->pct_arc = NULL; 319 320 ct->pct_ninstr = 0; 321 ct->pct_instr_c = 0; 322 ct->pct_instr = NULL; 323 324 return (ct); 325 } 326 327 /* 328 * Free a node. 329 */ 330 331 static void 332 pmcpl_ct_node_free(struct pmcpl_ct_node *ct) 333 { 334 int i; 335 336 for (i = 0; i < ct->pct_narc; i++) { 337 pmcpl_ct_samples_free(&ct->pct_arc[i].pcta_samples); 338 pmcpl_ct_samples_free(&ct->pct_arc[i].pcta_callid); 339 } 340 341 pmcpl_ct_samples_free(&ct->pct_samples); 342 free(ct->pct_arc); 343 free(ct->pct_instr); 344 free(ct); 345 } 346 347 /* 348 * Clear the graph tag on each node. 349 */ 350 static void 351 pmcpl_ct_node_cleartag(void) 352 { 353 int i; 354 struct pmcpl_ct_node_hash *pch; 355 356 for (i = 0; i < PMCSTAT_NHASH; i++) 357 LIST_FOREACH(pch, &pmcpl_ct_node_hash[i], pch_next) 358 pch->pch_ctnode->pct_flags &= ~PMCPL_PCT_TAG; 359 360 pmcpl_ct_root->pct_flags &= ~PMCPL_PCT_TAG; 361 } 362 363 /* 364 * Print the callchain line by line with maximum cost at top. 365 */ 366 367 static int 368 pmcpl_ct_node_dumptop(int pmcin, struct pmcpl_ct_node *ct, 369 struct pmcpl_ct_sample *rsamples, int x, int *y, int maxy) 370 { 371 int i; 372 373 if (ct->pct_flags & PMCPL_PCT_TAG) 374 return 0; 375 376 ct->pct_flags |= PMCPL_PCT_TAG; 377 378 if (x >= PMCPL_CT_MAXCOL) { 379 pmcpl_ct_topscreen[x][*y] = NULL; 380 return 1; 381 } 382 pmcpl_ct_topscreen[x][*y] = ct; 383 384 /* 385 * This is a terminal node 386 */ 387 if (ct->pct_narc == 0) { 388 pmcpl_ct_topscreen[x+1][*y] = NULL; 389 if (*y >= PMCPL_CT_MAXLINE || 390 *y >= maxy) 391 return 1; 392 *y = *y + 1; 393 for (i=0; i < x; i++) 394 pmcpl_ct_topscreen[i][*y] = 395 pmcpl_ct_topscreen[i][*y - 1]; 396 return 0; 397 } 398 399 /* 400 * Quicksort the arcs. 401 */ 402 qsort_r(ct->pct_arc, ct->pct_narc, sizeof(struct pmcpl_ct_arc), 403 &pmcin, pmcpl_ct_arc_compare); 404 405 for (i = 0; i < ct->pct_narc; i++) { 406 /* Skip this arc if there is no sample at all. */ 407 if (PMCPL_CT_SAMPLE(pmcin, 408 &ct->pct_arc[i].pcta_samples) == 0) 409 continue; 410 if (PMCPL_CT_SAMPLEP(pmcin, 411 &ct->pct_arc[i].pcta_samples) > pmcstat_threshold) { 412 if (pmcpl_ct_node_dumptop(pmcin, 413 ct->pct_arc[i].pcta_child, 414 rsamples, x+1, y, maxy)) 415 return 1; 416 } 417 } 418 419 return 0; 420 } 421 422 /* 423 * Format and display given PMC index. 424 */ 425 426 static void 427 pmcpl_ct_node_printtop(struct pmcpl_ct_sample *rsamples, int pmcin, int maxy) 428 { 429 int v_attrs, ns_len, vs_len, is_len, width, indentwidth, x, y; 430 float v; 431 char ns[30], vs[10], is[20]; 432 struct pmcpl_ct_node *ct; 433 struct pmcstat_symbol *sym; 434 const char *space = " "; 435 436 for (y = 0; y < maxy; y++) { 437 /* Output image. */ 438 ct = pmcpl_ct_topscreen[0][y]; 439 snprintf(is, sizeof(is), "%-10.10s", 440 pmcstat_string_unintern(ct->pct_image->pi_name)); 441 PMCSTAT_PRINTW("%s ", is); 442 width = indentwidth = 11; 443 444 for (x = 0; pmcpl_ct_topscreen[x][y] !=NULL; x++) { 445 446 ct = pmcpl_ct_topscreen[x][y]; 447 448 ns[0] = '\0'; ns_len = 0; 449 vs[0] = '\0'; vs_len = 0; 450 is[0] = '\0'; is_len = 0; 451 452 /* Format value. */ 453 v = PMCPL_CT_SAMPLEP(pmcin, &ct->pct_samples); 454 if (v > pmcstat_threshold) 455 vs_len = snprintf(vs, sizeof(vs), "(%.1f%%)", v); 456 v_attrs = PMCSTAT_ATTRPERCENT(v); 457 458 if (pmcstat_skiplink && v <= pmcstat_threshold) { 459 PMCSTAT_PRINTW(". "); 460 width += 2; 461 continue; 462 } 463 sym = pmcstat_symbol_search(ct->pct_image, ct->pct_func); 464 if (sym != NULL) { 465 ns_len = snprintf(ns, sizeof(ns), "%s", 466 pmcstat_string_unintern(sym->ps_name)); 467 } else 468 ns_len = snprintf(ns, sizeof(ns), "%p", 469 (void *)ct->pct_func); 470 471 /* Format image. */ 472 if (x > 0 && pmcpl_ct_topscreen[x-1][y]->pct_image != ct->pct_image) 473 is_len = snprintf(is, sizeof(is), "@%s", 474 pmcstat_string_unintern(ct->pct_image->pi_name)); 475 476 /* Check for line wrap. */ 477 width += ns_len + is_len + vs_len + 1; 478 if (width >= pmcstat_displaywidth) { 479 maxy--; 480 if (y >= maxy) 481 break; 482 PMCSTAT_PRINTW("\n%*s", indentwidth, space); 483 width = indentwidth + ns_len + is_len + vs_len; 484 } 485 486 PMCSTAT_ATTRON(v_attrs); 487 PMCSTAT_PRINTW("%s%s%s ", ns, is, vs); 488 PMCSTAT_ATTROFF(v_attrs); 489 } 490 PMCSTAT_PRINTW("\n"); 491 } 492 } 493 494 /* 495 * Output top mode snapshot. 496 */ 497 498 void 499 pmcpl_ct_topdisplay(void) 500 { 501 int i, x, y, pmcin; 502 struct pmcpl_ct_sample rsamples; 503 504 pmcpl_ct_samples_root(&rsamples); 505 506 PMCSTAT_PRINTW("%-10.10s %s\n", "IMAGE", "CALLTREE"); 507 508 for (pmcin = 0; pmcin < pmcstat_npmcs; pmcin++) { 509 /* Filter PMCs. */ 510 if (pmcstat_pmcinfilter != pmcin) 511 continue; 512 513 pmcpl_ct_node_cleartag(); 514 515 /* Quicksort the arcs. */ 516 qsort_r(pmcpl_ct_root->pct_arc, 517 pmcpl_ct_root->pct_narc, 518 sizeof(struct pmcpl_ct_arc), 519 &pmcin, pmcpl_ct_arc_compare); 520 521 x = y = 0; 522 for (i = 0; i < pmcpl_ct_root->pct_narc; i++) { 523 /* Skip this arc if there is no sample at all. */ 524 if (PMCPL_CT_SAMPLE(pmcin, 525 &pmcpl_ct_root->pct_arc[i].pcta_samples) == 0) 526 continue; 527 if (pmcpl_ct_node_dumptop(pmcin, 528 pmcpl_ct_root->pct_arc[i].pcta_child, 529 &rsamples, x, &y, pmcstat_displayheight - 2)) { 530 break; 531 } 532 } 533 534 pmcpl_ct_node_printtop(&rsamples, pmcin, y); 535 } 536 pmcpl_ct_samples_free(&rsamples); 537 } 538 539 /* 540 * Handle top mode keypress. 541 */ 542 543 int 544 pmcpl_ct_topkeypress(int c, WINDOW *w) 545 { 546 547 switch (c) { 548 case 'f': 549 pmcstat_skiplink = !pmcstat_skiplink; 550 wprintw(w, "skip empty link %s", pmcstat_skiplink ? "on" : "off"); 551 break; 552 } 553 554 return 0; 555 } 556 557 /* 558 * Look for a callgraph node associated with pmc `pmcid' in the global 559 * hash table that corresponds to the given `pc' value in the process map 560 * `ppm'. 561 */ 562 563 static struct pmcpl_ct_node * 564 pmcpl_ct_node_hash_lookup_pc(struct pmcpl_ct_node *parent, 565 struct pmcstat_pcmap *ppm, uintfptr_t pc, int pmcin) 566 { 567 struct pmcstat_symbol *sym; 568 struct pmcstat_image *image; 569 struct pmcpl_ct_node *ct; 570 struct pmcpl_ct_node_hash *h; 571 struct pmcpl_ct_arc *arc; 572 uintfptr_t loadaddress; 573 int i; 574 unsigned int hash; 575 576 assert(parent != NULL); 577 578 image = ppm->ppm_image; 579 580 loadaddress = ppm->ppm_lowpc + image->pi_vaddr - image->pi_start; 581 pc -= loadaddress; /* Convert to an offset in the image. */ 582 583 /* 584 * Try determine the function at this offset. If we can't 585 * find a function round leave the `pc' value alone. 586 */ 587 if ((sym = pmcstat_symbol_search(image, pc)) != NULL) 588 pc = sym->ps_start; 589 590 for (hash = i = 0; i < (int)sizeof(uintfptr_t); i++) 591 hash += (pc >> i) & 0xFF; 592 593 hash &= PMCSTAT_HASH_MASK; 594 595 ct = NULL; 596 LIST_FOREACH(h, &pmcpl_ct_node_hash[hash], pch_next) { 597 ct = h->pch_ctnode; 598 599 assert(ct != NULL); 600 601 if (ct->pct_image == image && ct->pct_func == pc) { 602 /* 603 * Find related arc in parent node and 604 * increment the sample count. 605 */ 606 for (i = 0; i < parent->pct_narc; i++) { 607 if (parent->pct_arc[i].pcta_child == ct) { 608 arc = &parent->pct_arc[i]; 609 pmcpl_ct_samples_grow(&arc->pcta_samples); 610 arc->pcta_samples.sb[pmcin]++; 611 /* Estimate call count. */ 612 pmcpl_ct_samples_grow(&arc->pcta_callid); 613 if (pmcpl_ct_callid.sb[pmcin] - 614 arc->pcta_callid.sb[pmcin] > 1) 615 arc->pcta_call++; 616 arc->pcta_callid.sb[pmcin] = 617 pmcpl_ct_callid.sb[pmcin]; 618 return (ct); 619 } 620 } 621 622 /* 623 * No arc found for us, add ourself to the parent. 624 */ 625 pmcpl_ct_arc_grow(parent->pct_narc, 626 &parent->pct_arc_c, &parent->pct_arc); 627 arc = &parent->pct_arc[parent->pct_narc]; 628 pmcpl_ct_samples_grow(&arc->pcta_samples); 629 arc->pcta_samples.sb[pmcin] = 1; 630 arc->pcta_call = 1; 631 pmcpl_ct_samples_grow(&arc->pcta_callid); 632 arc->pcta_callid.sb[pmcin] = pmcpl_ct_callid.sb[pmcin]; 633 arc->pcta_child = ct; 634 parent->pct_narc++; 635 return (ct); 636 } 637 } 638 639 /* 640 * We haven't seen this (pmcid, pc) tuple yet, so allocate a 641 * new callgraph node and a new hash table entry for it. 642 */ 643 ct = pmcpl_ct_node_allocate(image, pc); 644 if ((h = malloc(sizeof(*h))) == NULL) 645 err(EX_OSERR, "ERROR: Could not allocate callgraph node"); 646 647 h->pch_ctnode = ct; 648 LIST_INSERT_HEAD(&pmcpl_ct_node_hash[hash], h, pch_next); 649 650 pmcpl_ct_arc_grow(parent->pct_narc, 651 &parent->pct_arc_c, &parent->pct_arc); 652 arc = &parent->pct_arc[parent->pct_narc]; 653 pmcpl_ct_samples_grow(&arc->pcta_samples); 654 arc->pcta_samples.sb[pmcin] = 1; 655 arc->pcta_call = 1; 656 pmcpl_ct_samples_grow(&arc->pcta_callid); 657 arc->pcta_callid.sb[pmcin] = pmcpl_ct_callid.sb[pmcin]; 658 arc->pcta_child = ct; 659 parent->pct_narc++; 660 return (ct); 661 } 662 663 /* 664 * Record a callchain. 665 */ 666 667 void 668 pmcpl_ct_process(struct pmcstat_process *pp, struct pmcstat_pmcrecord *pmcr, 669 uint32_t nsamples, uintfptr_t *cc, int usermode, uint32_t cpu) 670 { 671 int n, pmcin; 672 struct pmcstat_pcmap *ppm[PMC_CALLCHAIN_DEPTH_MAX]; 673 struct pmcstat_process *km; 674 struct pmcpl_ct_node *parent, *child; 675 676 (void) cpu; 677 678 assert(nsamples>0 && nsamples<=PMC_CALLCHAIN_DEPTH_MAX); 679 680 /* Get the PMC index. */ 681 pmcin = pmcr->pr_pmcin; 682 683 /* 684 * Validate mapping for the callchain. 685 * Go from bottom to first invalid entry. 686 */ 687 km = pmcstat_kernproc; 688 for (n = 0; n < (int)nsamples; n++) { 689 ppm[n] = pmcstat_process_find_map(usermode ? 690 pp : km, cc[n]); 691 if (ppm[n] == NULL) { 692 /* Detect full frame capture (kernel + user). */ 693 if (!usermode) { 694 ppm[n] = pmcstat_process_find_map(pp, cc[n]); 695 if (ppm[n] != NULL) 696 km = pp; 697 } 698 } 699 if (ppm[n] == NULL) 700 break; 701 } 702 if (n-- == 0) { 703 pmcstat_stats.ps_callchain_dubious_frames++; 704 pmcr->pr_dubious_frames++; 705 return; 706 } 707 708 /* Increase the call generation counter. */ 709 pmcpl_ct_samples_grow(&pmcpl_ct_callid); 710 pmcpl_ct_callid.sb[pmcin]++; 711 712 /* 713 * Iterate remaining addresses. 714 */ 715 for (parent = pmcpl_ct_root, child = NULL; n >= 0; n--) { 716 child = pmcpl_ct_node_hash_lookup_pc(parent, ppm[n], cc[n], 717 pmcin); 718 if (child == NULL) { 719 pmcstat_stats.ps_callchain_dubious_frames++; 720 continue; 721 } 722 parent = child; 723 } 724 725 /* 726 * Increment the sample count for this PMC. 727 */ 728 if (child != NULL) { 729 pmcpl_ct_samples_grow(&child->pct_samples); 730 child->pct_samples.sb[pmcin]++; 731 732 /* Update per instruction sample if required. */ 733 if (args.pa_ctdumpinstr) 734 pmcpl_ct_instr_add(child, pmcin, cc[0] - 735 (ppm[0]->ppm_lowpc + ppm[0]->ppm_image->pi_vaddr - 736 ppm[0]->ppm_image->pi_start)); 737 } 738 } 739 740 /* 741 * Print node self cost. 742 */ 743 744 static void 745 pmcpl_ct_node_printself(struct pmcpl_ct_node *ct) 746 { 747 int i, j, line; 748 uintptr_t addr; 749 struct pmcstat_symbol *sym; 750 char sourcefile[PATH_MAX]; 751 char funcname[PATH_MAX]; 752 753 /* 754 * Object binary. 755 */ 756 #ifdef PMCPL_CT_OPTIMIZEFN 757 if (pmcpl_ct_prevfn != ct->pct_image->pi_fullpath) { 758 #endif 759 pmcpl_ct_prevfn = ct->pct_image->pi_fullpath; 760 fprintf(args.pa_graphfile, "ob=%s\n", 761 pmcstat_string_unintern(pmcpl_ct_prevfn)); 762 #ifdef PMCPL_CT_OPTIMIZEFN 763 } 764 #endif 765 766 /* 767 * Function name. 768 */ 769 if (pmcstat_image_addr2line(ct->pct_image, ct->pct_func, 770 sourcefile, sizeof(sourcefile), &line, 771 funcname, sizeof(funcname))) { 772 fprintf(args.pa_graphfile, "fn=%s\n", 773 funcname); 774 } else { 775 sym = pmcstat_symbol_search(ct->pct_image, ct->pct_func); 776 if (sym != NULL) 777 fprintf(args.pa_graphfile, "fn=%s\n", 778 pmcstat_string_unintern(sym->ps_name)); 779 else 780 fprintf(args.pa_graphfile, "fn=%p\n", 781 (void *)(ct->pct_image->pi_vaddr + ct->pct_func)); 782 } 783 784 /* 785 * Self cost. 786 */ 787 if (ct->pct_ninstr > 0) { 788 for (i = 0; i < ct->pct_ninstr; i++) { 789 addr = ct->pct_image->pi_vaddr + 790 ct->pct_instr[i].pctf_func; 791 line = 0; 792 if (pmcstat_image_addr2line(ct->pct_image, addr, 793 sourcefile, sizeof(sourcefile), &line, 794 funcname, sizeof(funcname))) 795 fprintf(args.pa_graphfile, "fl=%s\n", sourcefile); 796 fprintf(args.pa_graphfile, "%p %u", (void *)addr, line); 797 for (j = 0; j<pmcstat_npmcs; j++) 798 fprintf(args.pa_graphfile, " %u", 799 PMCPL_CT_SAMPLE(j, 800 &ct->pct_instr[i].pctf_samples)); 801 fprintf(args.pa_graphfile, "\n"); 802 } 803 } else { 804 addr = ct->pct_image->pi_vaddr + ct->pct_func; 805 line = 0; 806 if (pmcstat_image_addr2line(ct->pct_image, addr, 807 sourcefile, sizeof(sourcefile), &line, 808 funcname, sizeof(funcname))) 809 fprintf(args.pa_graphfile, "fl=%s\n", sourcefile); 810 fprintf(args.pa_graphfile, "* *"); 811 for (i = 0; i<pmcstat_npmcs ; i++) 812 fprintf(args.pa_graphfile, " %u", 813 PMCPL_CT_SAMPLE(i, &ct->pct_samples)); 814 fprintf(args.pa_graphfile, "\n"); 815 } 816 } 817 818 /* 819 * Print node child cost. 820 */ 821 822 static void 823 pmcpl_ct_node_printchild(struct pmcpl_ct_node *ct) 824 { 825 int i, j, line; 826 uintptr_t addr; 827 struct pmcstat_symbol *sym; 828 struct pmcpl_ct_node *child; 829 char sourcefile[PATH_MAX]; 830 char funcname[PATH_MAX]; 831 832 /* 833 * Child cost. 834 * TODO: attach child cost to the real position in the funtion. 835 * TODO: cfn=<fn> / call <ncall> addr(<fn>) / addr(call <fn>) <arccost> 836 */ 837 for (i=0 ; i<ct->pct_narc; i++) { 838 child = ct->pct_arc[i].pcta_child; 839 840 /* Object binary. */ 841 #ifdef PMCPL_CT_OPTIMIZEFN 842 if (pmcpl_ct_prevfn != child->pct_image->pi_fullpath) { 843 #endif 844 pmcpl_ct_prevfn = child->pct_image->pi_fullpath; 845 fprintf(args.pa_graphfile, "cob=%s\n", 846 pmcstat_string_unintern(pmcpl_ct_prevfn)); 847 #if PMCPL_CT_OPTIMIZEFN 848 } 849 #endif 850 /* Child function name. */ 851 addr = child->pct_image->pi_vaddr + child->pct_func; 852 /* Child function source file. */ 853 if (pmcstat_image_addr2line(child->pct_image, addr, 854 sourcefile, sizeof(sourcefile), &line, 855 funcname, sizeof(funcname))) { 856 fprintf(args.pa_graphfile, "cfn=%s\n", funcname); 857 fprintf(args.pa_graphfile, "cfl=%s\n", sourcefile); 858 } else { 859 sym = pmcstat_symbol_search(child->pct_image, 860 child->pct_func); 861 if (sym != NULL) 862 fprintf(args.pa_graphfile, "cfn=%s\n", 863 pmcstat_string_unintern(sym->ps_name)); 864 else 865 fprintf(args.pa_graphfile, "cfn=%p\n", (void *)addr); 866 } 867 868 /* Child function address, line and call count. */ 869 fprintf(args.pa_graphfile, "calls=%u %p %u\n", 870 ct->pct_arc[i].pcta_call, (void *)addr, line); 871 872 if (ct->pct_image != NULL) { 873 /* Call address, line, sample. */ 874 addr = ct->pct_image->pi_vaddr + ct->pct_func; 875 line = 0; 876 pmcstat_image_addr2line(ct->pct_image, addr, sourcefile, 877 sizeof(sourcefile), &line, 878 funcname, sizeof(funcname)); 879 fprintf(args.pa_graphfile, "%p %u", (void *)addr, line); 880 } 881 else 882 fprintf(args.pa_graphfile, "* *"); 883 for (j = 0; j<pmcstat_npmcs; j++) 884 fprintf(args.pa_graphfile, " %u", 885 PMCPL_CT_SAMPLE(j, &ct->pct_arc[i].pcta_samples)); 886 fprintf(args.pa_graphfile, "\n"); 887 } 888 } 889 890 /* 891 * Clean the PMC name for Kcachegrind formula 892 */ 893 894 static void 895 pmcpl_ct_fixup_pmcname(char *s) 896 { 897 char *p; 898 899 for (p = s; *p; p++) 900 if (!isalnum(*p)) 901 *p = '_'; 902 } 903 904 /* 905 * Print a calltree (KCachegrind) for all PMCs. 906 */ 907 908 static void 909 pmcpl_ct_print(void) 910 { 911 int n, i; 912 struct pmcpl_ct_node_hash *pch; 913 struct pmcpl_ct_sample rsamples; 914 char name[40]; 915 916 pmcpl_ct_samples_root(&rsamples); 917 pmcpl_ct_prevfn = NULL; 918 919 fprintf(args.pa_graphfile, 920 "version: 1\n" 921 "creator: pmcstat\n" 922 "positions: instr line\n" 923 "events:"); 924 for (i=0; i<pmcstat_npmcs; i++) { 925 snprintf(name, sizeof(name), "%s_%d", 926 pmcstat_pmcindex_to_name(i), i); 927 pmcpl_ct_fixup_pmcname(name); 928 fprintf(args.pa_graphfile, " %s", name); 929 } 930 fprintf(args.pa_graphfile, "\nsummary:"); 931 for (i=0; i<pmcstat_npmcs ; i++) 932 fprintf(args.pa_graphfile, " %u", 933 PMCPL_CT_SAMPLE(i, &rsamples)); 934 fprintf(args.pa_graphfile, "\n\n"); 935 936 /* 937 * Fake root node 938 */ 939 fprintf(args.pa_graphfile, "ob=FreeBSD\n"); 940 fprintf(args.pa_graphfile, "fn=ROOT\n"); 941 fprintf(args.pa_graphfile, "* *"); 942 for (i = 0; i<pmcstat_npmcs ; i++) 943 fprintf(args.pa_graphfile, " 0"); 944 fprintf(args.pa_graphfile, "\n"); 945 pmcpl_ct_node_printchild(pmcpl_ct_root); 946 947 for (n = 0; n < PMCSTAT_NHASH; n++) 948 LIST_FOREACH(pch, &pmcpl_ct_node_hash[n], pch_next) { 949 pmcpl_ct_node_printself(pch->pch_ctnode); 950 pmcpl_ct_node_printchild(pch->pch_ctnode); 951 } 952 953 pmcpl_ct_samples_free(&rsamples); 954 } 955 956 int 957 pmcpl_ct_configure(char *opt) 958 { 959 960 if (strncmp(opt, "skiplink=", 9) == 0) { 961 pmcstat_skiplink = atoi(opt+9); 962 } else 963 return (0); 964 965 return (1); 966 } 967 968 int 969 pmcpl_ct_init(void) 970 { 971 int i; 972 973 pmcpl_ct_prevfn = NULL; 974 pmcpl_ct_root = pmcpl_ct_node_allocate(NULL, 0); 975 976 for (i = 0; i < PMCSTAT_NHASH; i++) 977 LIST_INIT(&pmcpl_ct_node_hash[i]); 978 979 pmcpl_ct_samples_init(&pmcpl_ct_callid); 980 981 return (0); 982 } 983 984 void 985 pmcpl_ct_shutdown(FILE *mf) 986 { 987 int i; 988 struct pmcpl_ct_node_hash *pch, *pchtmp; 989 990 (void) mf; 991 992 if (args.pa_flags & FLAG_DO_CALLGRAPHS) 993 pmcpl_ct_print(); 994 995 /* 996 * Free memory. 997 */ 998 999 for (i = 0; i < PMCSTAT_NHASH; i++) { 1000 LIST_FOREACH_SAFE(pch, &pmcpl_ct_node_hash[i], pch_next, 1001 pchtmp) { 1002 pmcpl_ct_node_free(pch->pch_ctnode); 1003 free(pch); 1004 } 1005 } 1006 1007 pmcpl_ct_node_free(pmcpl_ct_root); 1008 pmcpl_ct_root = NULL; 1009 1010 pmcpl_ct_samples_free(&pmcpl_ct_callid); 1011 } 1012 1013