libcfs/libcfs/tracefile.c

   1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
   2  * vim:expandtab:shiftwidth=8:tabstop=8:
   3  *
   4  * GPL HEADER START
   5  *
   6  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   7  *
   8  * This program is free software; you can redistribute it and/or modify
   9  * it under the terms of the GNU General Public License version 2 only,
  10  * as published by the Free Software Foundation.
  11  *
  12  * This program is distributed in the hope that it will be useful, but
  13  * WITHOUT ANY WARRANTY; without even the implied warranty of
  14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  15  * General Public License version 2 for more details (a copy is included
  16  * in the LICENSE file that accompanied this code).
  17  *
  18  * You should have received a copy of the GNU General Public License
  19  * version 2 along with this program; If not, see
  20  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  21  *
  22  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  23  * CA 95054 USA or visit www.sun.com if you need additional information or
  24  * have any questions.
  25  *
  26  * GPL HEADER END
  27  */
  28 /*
  29  * Copyright  2008 Sun Microsystems, Inc. All rights reserved
  30  * Use is subject to license terms.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  *
  36  * libcfs/libcfs/tracefile.c
  37  *
  38  * Author: Zach Brown <zab@clusterfs.com>
  39  * Author: Phil Schwan <phil@clusterfs.com>
  40  */
  41
  42
  43 #define DEBUG_SUBSYSTEM S_LNET
  44 #define LUSTRE_TRACEFILE_PRIVATE
  45 #include "tracefile.h"
  46
  47 #include <libcfs/libcfs.h>
  48
  49 /* XXX move things up to the top, comment */
  50 union cfs_trace_data_union (*cfs_trace_data[TCD_MAX_TYPES])[CFS_NR_CPUS] __cacheline_aligned;
  51
  52 char cfs_tracefile[TRACEFILE_NAME_SIZE];
  53 long long cfs_tracefile_size = CFS_TRACEFILE_SIZE;
  54 static struct tracefiled_ctl trace_tctl;
  55 cfs_semaphore_t cfs_trace_thread_sem;
  56 static int thread_running = 0;
  57
  58 cfs_atomic_t cfs_tage_allocated = CFS_ATOMIC_INIT(0);
  59
  60 static void put_pages_on_tcd_daemon_list(struct page_collection *pc,
  61                                          struct cfs_trace_cpu_data *tcd);
  62
  63 static inline struct cfs_trace_page *
  64 cfs_tage_from_list(cfs_list_t *list)
  65 {
  66         return cfs_list_entry(list, struct cfs_trace_page, linkage);
  67 }
  68
  69 static struct cfs_trace_page *cfs_tage_alloc(int gfp)
  70 {
  71         cfs_page_t            *page;
  72         struct cfs_trace_page *tage;
  73
  74         /*
  75          * Don't spam console with allocation failures: they will be reported
  76          * by upper layer anyway.
  77          */
  78         gfp |= CFS_ALLOC_NOWARN;
  79         page = cfs_alloc_page(gfp);
  80         if (page == NULL)
  81                 return NULL;
  82
  83         tage = cfs_alloc(sizeof(*tage), gfp);
  84         if (tage == NULL) {
  85                 cfs_free_page(page);
  86                 return NULL;
  87         }
  88
  89         tage->page = page;
  90         cfs_atomic_inc(&cfs_tage_allocated);
  91         return tage;
  92 }
  93
  94 static void cfs_tage_free(struct cfs_trace_page *tage)
  95 {
  96         __LASSERT(tage != NULL);
  97         __LASSERT(tage->page != NULL);
  98
  99         cfs_free_page(tage->page);
 100         cfs_free(tage);
 101         cfs_atomic_dec(&cfs_tage_allocated);
 102 }
 103
 104 static void cfs_tage_to_tail(struct cfs_trace_page *tage,
 105                              cfs_list_t *queue)
 106 {
 107         __LASSERT(tage != NULL);
 108         __LASSERT(queue != NULL);
 109
 110         cfs_list_move_tail(&tage->linkage, queue);
 111 }
 112
 113 int cfs_trace_refill_stock(struct cfs_trace_cpu_data *tcd, int gfp,
 114                            cfs_list_t *stock)
 115 {
 116         int i;
 117
 118         /*
 119          * XXX nikita: do NOT call portals_debug_msg() (CDEBUG/ENTRY/EXIT)
 120          * from here: this will lead to infinite recursion.
 121          */
 122
 123         for (i = 0; i + tcd->tcd_cur_stock_pages < TCD_STOCK_PAGES ; ++ i) {
 124                 struct cfs_trace_page *tage;
 125
 126                 tage = cfs_tage_alloc(gfp);
 127                 if (tage == NULL)
 128                         break;
 129                 cfs_list_add_tail(&tage->linkage, stock);
 130         }
 131         return i;
 132 }
 133
 134 /* return a page that has 'len' bytes left at the end */
 135 static struct cfs_trace_page *
 136 cfs_trace_get_tage_try(struct cfs_trace_cpu_data *tcd, unsigned long len)
 137 {
 138         struct cfs_trace_page *tage;
 139
 140         if (tcd->tcd_cur_pages > 0) {
 141                 __LASSERT(!cfs_list_empty(&tcd->tcd_pages));
 142                 tage = cfs_tage_from_list(tcd->tcd_pages.prev);
 143                 if (tage->used + len <= CFS_PAGE_SIZE)
 144                         return tage;
 145         }
 146
 147         if (tcd->tcd_cur_pages < tcd->tcd_max_pages) {
 148                 if (tcd->tcd_cur_stock_pages > 0) {
 149                         tage = cfs_tage_from_list(tcd->tcd_stock_pages.prev);
 150                         -- tcd->tcd_cur_stock_pages;
 151                         cfs_list_del_init(&tage->linkage);
 152                 } else {
 153                         tage = cfs_tage_alloc(CFS_ALLOC_ATOMIC);
 154                         if (tage == NULL) {
 155                                 if (printk_ratelimit())
 156                                         printk(CFS_KERN_WARNING
 157                                                "cannot allocate a tage (%ld)\n",
 158                                        tcd->tcd_cur_pages);
 159                                 return NULL;
 160                         }
 161                 }
 162
 163                 tage->used = 0;
 164                 tage->cpu = cfs_smp_processor_id();
 165                 tage->type = tcd->tcd_type;
 166                 cfs_list_add_tail(&tage->linkage, &tcd->tcd_pages);
 167                 tcd->tcd_cur_pages++;
 168
 169                 if (tcd->tcd_cur_pages > 8 && thread_running) {
 170                         struct tracefiled_ctl *tctl = &trace_tctl;
 171                         /*
 172                          * wake up tracefiled to process some pages.
 173                          */
 174                         cfs_waitq_signal(&tctl->tctl_waitq);
 175                 }
 176                 return tage;
 177         }
 178         return NULL;
 179 }
 180
 181 static void cfs_tcd_shrink(struct cfs_trace_cpu_data *tcd)
 182 {
 183         int pgcount = tcd->tcd_cur_pages / 10;
 184         struct page_collection pc;
 185         struct cfs_trace_page *tage;
 186         struct cfs_trace_page *tmp;
 187
 188         /*
 189          * XXX nikita: do NOT call portals_debug_msg() (CDEBUG/ENTRY/EXIT)
 190          * from here: this will lead to infinite recursion.
 191          */
 192
 193         if (printk_ratelimit())
 194                 printk(CFS_KERN_WARNING "debug daemon buffer overflowed; "
 195                        "discarding 10%% of pages (%d of %ld)\n",
 196                        pgcount + 1, tcd->tcd_cur_pages);
 197
 198         CFS_INIT_LIST_HEAD(&pc.pc_pages);
 199         cfs_spin_lock_init(&pc.pc_lock);
 200
 201         cfs_list_for_each_entry_safe_typed(tage, tmp, &tcd->tcd_pages,
 202                                            struct cfs_trace_page, linkage) {
 203                 if (pgcount-- == 0)
 204                         break;
 205
 206                 cfs_list_move_tail(&tage->linkage, &pc.pc_pages);
 207                 tcd->tcd_cur_pages--;
 208         }
 209         put_pages_on_tcd_daemon_list(&pc, tcd);
 210 }
 211
 212 /* return a page that has 'len' bytes left at the end */
 213 static struct cfs_trace_page *cfs_trace_get_tage(struct cfs_trace_cpu_data *tcd,
 214                                                  unsigned long len)
 215 {
 216         struct cfs_trace_page *tage;
 217
 218         /*
 219          * XXX nikita: do NOT call portals_debug_msg() (CDEBUG/ENTRY/EXIT)
 220          * from here: this will lead to infinite recursion.
 221          */
 222
 223         if (len > CFS_PAGE_SIZE) {
 224                 printk(CFS_KERN_ERR
 225                        "cowardly refusing to write %lu bytes in a page\n", len);
 226                 return NULL;
 227         }
 228
 229         tage = cfs_trace_get_tage_try(tcd, len);
 230         if (tage != NULL)
 231                 return tage;
 232         if (thread_running)
 233                 cfs_tcd_shrink(tcd);
 234         if (tcd->tcd_cur_pages > 0) {
 235                 tage = cfs_tage_from_list(tcd->tcd_pages.next);
 236                 tage->used = 0;
 237                 cfs_tage_to_tail(tage, &tcd->tcd_pages);
 238         }
 239         return tage;
 240 }
 241
 242 int libcfs_debug_vmsg2(cfs_debug_limit_state_t *cdls, int subsys, int mask,
 243                        const char *file, const char *fn, const int line,
 244                        const char *format1, va_list args,
 245                        const char *format2, ...)
 246 {
 247         struct cfs_trace_cpu_data *tcd = NULL;
 248         struct ptldebug_header     header = {0};
 249         struct cfs_trace_page     *tage;
 250         /* string_buf is used only if tcd != NULL, and is always set then */
 251         char                      *string_buf = NULL;
 252         char                      *debug_buf;
 253         int                        known_size;
 254         int                        needed = 85; /* average message length */
 255         int                        max_nob;
 256         va_list                    ap;
 257         int                        depth;
 258         int                        i;
 259         int                        remain;
 260
 261         if (strchr(file, '/'))
 262                 file = strrchr(file, '/') + 1;
 263
 264
 265         cfs_set_ptldebug_header(&header, subsys, mask, line, CDEBUG_STACK());
 266
 267         tcd = cfs_trace_get_tcd();
 268         if (tcd == NULL)                /* arch may not log in IRQ context */
 269                 goto console;
 270
 271         if (tcd->tcd_cur_pages == 0)
 272                 header.ph_flags |= PH_FLAG_FIRST_RECORD;
 273
 274         if (tcd->tcd_shutting_down) {
 275                 cfs_trace_put_tcd(tcd);
 276                 tcd = NULL;
 277                 goto console;
 278         }
 279
 280         depth = __current_nesting_level();
 281         known_size = strlen(file) + 1 + depth;
 282         if (fn)
 283                 known_size += strlen(fn) + 1;
 284
 285         if (libcfs_debug_binary)
 286                 known_size += sizeof(header);
 287
 288         /*/
 289          * '2' used because vsnprintf return real size required for output
 290          * _without_ terminating NULL.
 291          * if needed is to small for this format.
 292          */
 293         for (i = 0; i < 2; i++) {
 294                 tage = cfs_trace_get_tage(tcd, needed + known_size + 1);
 295                 if (tage == NULL) {
 296                         if (needed + known_size > CFS_PAGE_SIZE)
 297                                 mask |= D_ERROR;
 298
 299                         cfs_trace_put_tcd(tcd);
 300                         tcd = NULL;
 301                         goto console;
 302                 }
 303
 304                 string_buf = (char *)cfs_page_address(tage->page) +
 305                                         tage->used + known_size;
 306
 307                 max_nob = CFS_PAGE_SIZE - tage->used - known_size;
 308                 if (max_nob <= 0) {
 309                         printk(CFS_KERN_EMERG "negative max_nob: %i\n",
 310                                max_nob);
 311                         mask |= D_ERROR;
 312                         cfs_trace_put_tcd(tcd);
 313                         tcd = NULL;
 314                         goto console;
 315                 }
 316
 317                 needed = 0;
 318                 if (format1) {
 319                         va_copy(ap, args);
 320                         needed = vsnprintf(string_buf, max_nob, format1, ap);
 321                         va_end(ap);
 322                 }
 323
 324                 if (format2) {
 325                         remain = max_nob - needed;
 326                         if (remain < 0)
 327                                 remain = 0;
 328
 329                         va_start(ap, format2);
 330                         needed += vsnprintf(string_buf + needed, remain,
 331                                             format2, ap);
 332                         va_end(ap);
 333                 }
 334
 335                 if (needed < max_nob) /* well. printing ok.. */
 336                         break;
 337         }
 338
 339         if (*(string_buf+needed-1) != '\n')
 340                 printk(CFS_KERN_INFO "format at %s:%d:%s doesn't end in newline\n",
 341                        file, line, fn);
 342
 343         header.ph_len = known_size + needed;
 344         debug_buf = (char *)cfs_page_address(tage->page) + tage->used;
 345
 346         if (libcfs_debug_binary) {
 347                 memcpy(debug_buf, &header, sizeof(header));
 348                 tage->used += sizeof(header);
 349                 debug_buf += sizeof(header);
 350         }
 351
 352         /* indent message according to the nesting level */
 353         while (depth-- > 0) {
 354                 *(debug_buf++) = '.';
 355                 ++ tage->used;
 356         }
 357
 358         strcpy(debug_buf, file);
 359         tage->used += strlen(file) + 1;
 360         debug_buf += strlen(file) + 1;
 361
 362         if (fn) {
 363                 strcpy(debug_buf, fn);
 364                 tage->used += strlen(fn) + 1;
 365                 debug_buf += strlen(fn) + 1;
 366         }
 367
 368         __LASSERT(debug_buf == string_buf);
 369
 370         tage->used += needed;
 371         __LASSERT (tage->used <= CFS_PAGE_SIZE);
 372
 373 console:
 374         if ((mask & libcfs_printk) == 0) {
 375                 /* no console output requested */
 376                 if (tcd != NULL)
 377                         cfs_trace_put_tcd(tcd);
 378                 return 1;
 379         }
 380
 381         if (cdls != NULL) {
 382                 if (libcfs_console_ratelimit &&
 383                     cdls->cdls_next != 0 &&     /* not first time ever */
 384                     !cfs_time_after(cfs_time_current(), cdls->cdls_next)) {
 385                         /* skipping a console message */
 386                         cdls->cdls_count++;
 387                         if (tcd != NULL)
 388                                 cfs_trace_put_tcd(tcd);
 389                         return 1;
 390                 }
 391
 392                 if (cfs_time_after(cfs_time_current(), cdls->cdls_next +
 393                                                        libcfs_console_max_delay
 394                                                        + cfs_time_seconds(10))) {
 395                         /* last timeout was a long time ago */
 396                         cdls->cdls_delay /= libcfs_console_backoff * 4;
 397                 } else {
 398                         cdls->cdls_delay *= libcfs_console_backoff;
 399
 400                         if (cdls->cdls_delay < libcfs_console_min_delay)
 401                                 cdls->cdls_delay = libcfs_console_min_delay;
 402                         else if (cdls->cdls_delay > libcfs_console_max_delay)
 403                                 cdls->cdls_delay = libcfs_console_max_delay;
 404                 }
 405
 406                 /* ensure cdls_next is never zero after it's been seen */
 407                 cdls->cdls_next = (cfs_time_current() + cdls->cdls_delay) | 1;
 408         }
 409
 410         if (tcd != NULL) {
 411                 cfs_print_to_console(&header, mask, string_buf, needed, file,
 412                                      fn);
 413                 cfs_trace_put_tcd(tcd);
 414         } else {
 415                 string_buf = cfs_trace_get_console_buffer();
 416
 417                 needed = 0;
 418                 if (format1 != NULL) {
 419                         va_copy(ap, args);
 420                         needed = vsnprintf(string_buf,
 421                                            CFS_TRACE_CONSOLE_BUFFER_SIZE,
 422                                            format1, ap);
 423                         va_end(ap);
 424                 }
 425                 if (format2 != NULL) {
 426                         remain = CFS_TRACE_CONSOLE_BUFFER_SIZE - needed;
 427                         if (remain > 0) {
 428                                 va_start(ap, format2);
 429                                 needed += vsnprintf(string_buf+needed, remain, format2, ap);
 430                                 va_end(ap);
 431                         }
 432                 }
 433                 cfs_print_to_console(&header, mask,
 434                                      string_buf, needed, file, fn);
 435
 436                 cfs_trace_put_console_buffer(string_buf);
 437         }
 438
 439         if (cdls != NULL && cdls->cdls_count != 0) {
 440                 string_buf = cfs_trace_get_console_buffer();
 441
 442                 needed = snprintf(string_buf, CFS_TRACE_CONSOLE_BUFFER_SIZE,
 443                          "Skipped %d previous similar message%s\n",
 444                          cdls->cdls_count, (cdls->cdls_count > 1) ? "s" : "");
 445
 446                 cfs_print_to_console(&header, mask,
 447                                  string_buf, needed, file, fn);
 448
 449                 cfs_trace_put_console_buffer(string_buf);
 450                 cdls->cdls_count = 0;
 451         }
 452
 453         return 0;
 454 }
 455 EXPORT_SYMBOL(libcfs_debug_vmsg2);
 456
 457 void
 458 libcfs_assertion_failed(const char *expr, const char *file,
 459                         const char *func, const int line)
 460 {
 461         libcfs_debug_msg(NULL, 0, D_EMERG, file, func, line,
 462                          "ASSERTION(%s) failed\n", expr);
 463         /* cfs_enter_debugger(); */
 464         lbug_with_loc(file, func, line);
 465 }
 466 EXPORT_SYMBOL(libcfs_assertion_failed);
 467
 468 void
 469 cfs_trace_assertion_failed(const char *str,
 470                            const char *fn, const char *file, int line)
 471 {
 472         struct ptldebug_header hdr;
 473
 474         libcfs_panic_in_progress = 1;
 475         libcfs_catastrophe = 1;
 476         cfs_mb();
 477
 478         cfs_set_ptldebug_header(&hdr, DEBUG_SUBSYSTEM, D_EMERG, line,
 479                                 CDEBUG_STACK());
 480
 481         cfs_print_to_console(&hdr, D_EMERG, str, strlen(str), file, fn);
 482
 483         LIBCFS_PANIC("Lustre debug assertion failure\n");
 484
 485         /* not reached */
 486 }
 487
 488 static void
 489 panic_collect_pages(struct page_collection *pc)
 490 {
 491         /* Do the collect_pages job on a single CPU: assumes that all other
 492          * CPUs have been stopped during a panic.  If this isn't true for some
 493          * arch, this will have to be implemented separately in each arch.  */
 494         int                        i;
 495         int                        j;
 496         struct cfs_trace_cpu_data *tcd;
 497
 498         CFS_INIT_LIST_HEAD(&pc->pc_pages);
 499
 500         cfs_tcd_for_each(tcd, i, j) {
 501                 cfs_list_splice_init(&tcd->tcd_pages, &pc->pc_pages);
 502                 tcd->tcd_cur_pages = 0;
 503
 504                 if (pc->pc_want_daemon_pages) {
 505                         cfs_list_splice_init(&tcd->tcd_daemon_pages,
 506                                              &pc->pc_pages);
 507                         tcd->tcd_cur_daemon_pages = 0;
 508                 }
 509         }
 510 }
 511
 512 static void collect_pages_on_all_cpus(struct page_collection *pc)
 513 {
 514         struct cfs_trace_cpu_data *tcd;
 515         int i, cpu;
 516
 517         cfs_spin_lock(&pc->pc_lock);
 518         cfs_for_each_possible_cpu(cpu) {
 519                 cfs_tcd_for_each_type_lock(tcd, i, cpu) {
 520                         cfs_list_splice_init(&tcd->tcd_pages, &pc->pc_pages);
 521                         tcd->tcd_cur_pages = 0;
 522                         if (pc->pc_want_daemon_pages) {
 523                                 cfs_list_splice_init(&tcd->tcd_daemon_pages,
 524                                                      &pc->pc_pages);
 525                                 tcd->tcd_cur_daemon_pages = 0;
 526                         }
 527                 }
 528         }
 529         cfs_spin_unlock(&pc->pc_lock);
 530 }
 531
 532 static void collect_pages(struct page_collection *pc)
 533 {
 534         CFS_INIT_LIST_HEAD(&pc->pc_pages);
 535
 536         if (libcfs_panic_in_progress)
 537                 panic_collect_pages(pc);
 538         else
 539                 collect_pages_on_all_cpus(pc);
 540 }
 541
 542 static void put_pages_back_on_all_cpus(struct page_collection *pc)
 543 {
 544         struct cfs_trace_cpu_data *tcd;
 545         cfs_list_t *cur_head;
 546         struct cfs_trace_page *tage;
 547         struct cfs_trace_page *tmp;
 548         int i, cpu;
 549
 550         cfs_spin_lock(&pc->pc_lock);
 551         cfs_for_each_possible_cpu(cpu) {
 552                 cfs_tcd_for_each_type_lock(tcd, i, cpu) {
 553                         cur_head = tcd->tcd_pages.next;
 554
 555                         cfs_list_for_each_entry_safe_typed(tage, tmp,
 556                                                            &pc->pc_pages,
 557                                                            struct cfs_trace_page,
 558                                                            linkage) {
 559
 560                                 __LASSERT_TAGE_INVARIANT(tage);
 561
 562                                 if (tage->cpu != cpu || tage->type != i)
 563                                         continue;
 564
 565                                 cfs_tage_to_tail(tage, cur_head);
 566                                 tcd->tcd_cur_pages++;
 567                         }
 568                 }
 569         }
 570         cfs_spin_unlock(&pc->pc_lock);
 571 }
 572
 573 static void put_pages_back(struct page_collection *pc)
 574 {
 575         if (!libcfs_panic_in_progress)
 576                 put_pages_back_on_all_cpus(pc);
 577 }
 578
 579 /* Add pages to a per-cpu debug daemon ringbuffer.  This buffer makes sure that
 580  * we have a good amount of data at all times for dumping during an LBUG, even
 581  * if we have been steadily writing (and otherwise discarding) pages via the
 582  * debug daemon. */
 583 static void put_pages_on_tcd_daemon_list(struct page_collection *pc,
 584                                          struct cfs_trace_cpu_data *tcd)
 585 {
 586         struct cfs_trace_page *tage;
 587         struct cfs_trace_page *tmp;
 588
 589         cfs_spin_lock(&pc->pc_lock);
 590         cfs_list_for_each_entry_safe_typed(tage, tmp, &pc->pc_pages,
 591                                            struct cfs_trace_page, linkage) {
 592
 593                 __LASSERT_TAGE_INVARIANT(tage);
 594
 595                 if (tage->cpu != tcd->tcd_cpu || tage->type != tcd->tcd_type)
 596                         continue;
 597
 598                 cfs_tage_to_tail(tage, &tcd->tcd_daemon_pages);
 599                 tcd->tcd_cur_daemon_pages++;
 600
 601                 if (tcd->tcd_cur_daemon_pages > tcd->tcd_max_pages) {
 602                         struct cfs_trace_page *victim;
 603
 604                         __LASSERT(!cfs_list_empty(&tcd->tcd_daemon_pages));
 605                         victim = cfs_tage_from_list(tcd->tcd_daemon_pages.next);
 606
 607                         __LASSERT_TAGE_INVARIANT(victim);
 608
 609                         cfs_list_del(&victim->linkage);
 610                         cfs_tage_free(victim);
 611                         tcd->tcd_cur_daemon_pages--;
 612                 }
 613         }
 614         cfs_spin_unlock(&pc->pc_lock);
 615 }
 616
 617 static void put_pages_on_daemon_list(struct page_collection *pc)
 618 {
 619         struct cfs_trace_cpu_data *tcd;
 620         int i, cpu;
 621
 622         cfs_for_each_possible_cpu(cpu) {
 623                 cfs_tcd_for_each_type_lock(tcd, i, cpu)
 624                         put_pages_on_tcd_daemon_list(pc, tcd);
 625         }
 626 }
 627
 628 void cfs_trace_debug_print(void)
 629 {
 630         struct page_collection pc;
 631         struct cfs_trace_page *tage;
 632         struct cfs_trace_page *tmp;
 633
 634         cfs_spin_lock_init(&pc.pc_lock);
 635
 636         pc.pc_want_daemon_pages = 1;
 637         collect_pages(&pc);
 638         cfs_list_for_each_entry_safe_typed(tage, tmp, &pc.pc_pages,
 639                                            struct cfs_trace_page, linkage) {
 640                 char *p, *file, *fn;
 641                 cfs_page_t *page;
 642
 643                 __LASSERT_TAGE_INVARIANT(tage);
 644
 645                 page = tage->page;
 646                 p = cfs_page_address(page);
 647                 while (p < ((char *)cfs_page_address(page) + tage->used)) {
 648                         struct ptldebug_header *hdr;
 649                         int len;
 650                         hdr = (void *)p;
 651                         p += sizeof(*hdr);
 652                         file = p;
 653                         p += strlen(file) + 1;
 654                         fn = p;
 655                         p += strlen(fn) + 1;
 656                         len = hdr->ph_len - (int)(p - (char *)hdr);
 657
 658                         cfs_print_to_console(hdr, D_EMERG, p, len, file, fn);
 659
 660                         p += len;
 661                 }
 662
 663                 cfs_list_del(&tage->linkage);
 664                 cfs_tage_free(tage);
 665         }
 666 }
 667
 668 int cfs_tracefile_dump_all_pages(char *filename)
 669 {
 670         struct page_collection pc;
 671         cfs_file_t *filp;
 672         struct cfs_trace_page *tage;
 673         struct cfs_trace_page *tmp;
 674         int rc;
 675
 676         CFS_DECL_MMSPACE;
 677
 678         cfs_tracefile_write_lock();
 679
 680         filp = cfs_filp_open(filename,
 681                              O_CREAT|O_EXCL|O_WRONLY|O_LARGEFILE, 0600, &rc);
 682         if (!filp) {
 683                 if (rc != -EEXIST)
 684                         printk(CFS_KERN_ERR
 685                                "LustreError: can't open %s for dump: rc %d\n",
 686                                filename, rc);
 687                 goto out;
 688         }
 689
 690         cfs_spin_lock_init(&pc.pc_lock);
 691         pc.pc_want_daemon_pages = 1;
 692         collect_pages(&pc);
 693         if (cfs_list_empty(&pc.pc_pages)) {
 694                 rc = 0;
 695                 goto close;
 696         }
 697
 698         /* ok, for now, just write the pages.  in the future we'll be building
 699          * iobufs with the pages and calling generic_direct_IO */
 700         CFS_MMSPACE_OPEN;
 701         cfs_list_for_each_entry_safe_typed(tage, tmp, &pc.pc_pages,
 702                                            struct cfs_trace_page, linkage) {
 703
 704                 __LASSERT_TAGE_INVARIANT(tage);
 705
 706                 rc = cfs_filp_write(filp, cfs_page_address(tage->page),
 707                                     tage->used, cfs_filp_poff(filp));
 708                 if (rc != (int)tage->used) {
 709                         printk(CFS_KERN_WARNING "wanted to write %u but wrote "
 710                                "%d\n", tage->used, rc);
 711                         put_pages_back(&pc);
 712                         __LASSERT(cfs_list_empty(&pc.pc_pages));
 713                         break;
 714                 }
 715                 cfs_list_del(&tage->linkage);
 716                 cfs_tage_free(tage);
 717         }
 718         CFS_MMSPACE_CLOSE;
 719         rc = cfs_filp_fsync(filp);
 720         if (rc)
 721                 printk(CFS_KERN_ERR "sync returns %d\n", rc);
 722  close:
 723         cfs_filp_close(filp);
 724  out:
 725         cfs_tracefile_write_unlock();
 726         return rc;
 727 }
 728
 729 void cfs_trace_flush_pages(void)
 730 {
 731         struct page_collection pc;
 732         struct cfs_trace_page *tage;
 733         struct cfs_trace_page *tmp;
 734
 735         cfs_spin_lock_init(&pc.pc_lock);
 736
 737         pc.pc_want_daemon_pages = 1;
 738         collect_pages(&pc);
 739         cfs_list_for_each_entry_safe_typed(tage, tmp, &pc.pc_pages,
 740                                            struct cfs_trace_page, linkage) {
 741
 742                 __LASSERT_TAGE_INVARIANT(tage);
 743
 744                 cfs_list_del(&tage->linkage);
 745                 cfs_tage_free(tage);
 746         }
 747 }
 748
 749 int cfs_trace_copyin_string(char *knl_buffer, int knl_buffer_nob,
 750                             const char *usr_buffer, int usr_buffer_nob)
 751 {
 752         int    nob;
 753
 754         if (usr_buffer_nob > knl_buffer_nob)
 755                 return -EOVERFLOW;
 756
 757         if (cfs_copy_from_user((void *)knl_buffer,
 758                            (void *)usr_buffer, usr_buffer_nob))
 759                 return -EFAULT;
 760
 761         nob = strnlen(knl_buffer, usr_buffer_nob);
 762         while (nob-- >= 0)                      /* strip trailing whitespace */
 763                 if (!isspace(knl_buffer[nob]))
 764                         break;
 765
 766         if (nob < 0)                            /* empty string */
 767                 return -EINVAL;
 768
 769         if (nob == knl_buffer_nob)              /* no space to terminate */
 770                 return -EOVERFLOW;
 771
 772         knl_buffer[nob + 1] = 0;                /* terminate */
 773         return 0;
 774 }
 775
 776 int cfs_trace_copyout_string(char *usr_buffer, int usr_buffer_nob,
 777                              const char *knl_buffer, char *append)
 778 {
 779         /* NB if 'append' != NULL, it's a single character to append to the
 780          * copied out string - usually "\n", for /proc entries and "" (i.e. a
 781          * terminating zero byte) for sysctl entries */
 782         int   nob = strlen(knl_buffer);
 783
 784         if (nob > usr_buffer_nob)
 785                 nob = usr_buffer_nob;
 786
 787         if (cfs_copy_to_user(usr_buffer, knl_buffer, nob))
 788                 return -EFAULT;
 789
 790         if (append != NULL && nob < usr_buffer_nob) {
 791                 if (cfs_copy_to_user(usr_buffer + nob, append, 1))
 792                         return -EFAULT;
 793
 794                 nob++;
 795         }
 796
 797         return nob;
 798 }
 799 EXPORT_SYMBOL(cfs_trace_copyout_string);
 800
 801 int cfs_trace_allocate_string_buffer(char **str, int nob)
 802 {
 803         if (nob > 2 * CFS_PAGE_SIZE)            /* string must be "sensible" */
 804                 return -EINVAL;
 805
 806         *str = cfs_alloc(nob, CFS_ALLOC_STD | CFS_ALLOC_ZERO);
 807         if (*str == NULL)
 808                 return -ENOMEM;
 809
 810         return 0;
 811 }
 812
 813 void cfs_trace_free_string_buffer(char *str, int nob)
 814 {
 815         cfs_free(str);
 816 }
 817
 818 int cfs_trace_dump_debug_buffer_usrstr(void *usr_str, int usr_str_nob)
 819 {
 820         char         *str;
 821         int           rc;
 822
 823         rc = cfs_trace_allocate_string_buffer(&str, usr_str_nob + 1);
 824         if (rc != 0)
 825                 return rc;
 826
 827         rc = cfs_trace_copyin_string(str, usr_str_nob + 1,
 828                                      usr_str, usr_str_nob);
 829         if (rc != 0)
 830                 goto out;
 831
 832 #if !defined(__WINNT__)
 833         if (str[0] != '/') {
 834                 rc = -EINVAL;
 835                 goto out;
 836         }
 837 #endif
 838         rc = cfs_tracefile_dump_all_pages(str);
 839 out:
 840         cfs_trace_free_string_buffer(str, usr_str_nob + 1);
 841         return rc;
 842 }
 843
 844 int cfs_trace_daemon_command(char *str)
 845 {
 846         int       rc = 0;
 847
 848         cfs_tracefile_write_lock();
 849
 850         if (strcmp(str, "stop") == 0) {
 851                 cfs_tracefile_write_unlock();
 852                 cfs_trace_stop_thread();
 853                 cfs_tracefile_write_lock();
 854                 memset(cfs_tracefile, 0, sizeof(cfs_tracefile));
 855
 856         } else if (strncmp(str, "size=", 5) == 0) {
 857                 cfs_tracefile_size = simple_strtoul(str + 5, NULL, 0);
 858                 if (cfs_tracefile_size < 10 || cfs_tracefile_size > 20480)
 859                         cfs_tracefile_size = CFS_TRACEFILE_SIZE;
 860                 else
 861                         cfs_tracefile_size <<= 20;
 862
 863         } else if (strlen(str) >= sizeof(cfs_tracefile)) {
 864                 rc = -ENAMETOOLONG;
 865 #ifndef __WINNT__
 866         } else if (str[0] != '/') {
 867                 rc = -EINVAL;
 868 #endif
 869         } else {
 870                 strcpy(cfs_tracefile, str);
 871
 872                 printk(CFS_KERN_INFO
 873                        "Lustre: debug daemon will attempt to start writing "
 874                        "to %s (%lukB max)\n", cfs_tracefile,
 875                        (long)(cfs_tracefile_size >> 10));
 876
 877                 cfs_trace_start_thread();
 878         }
 879
 880         cfs_tracefile_write_unlock();
 881         return rc;
 882 }
 883
 884 int cfs_trace_daemon_command_usrstr(void *usr_str, int usr_str_nob)
 885 {
 886         char *str;
 887         int   rc;
 888
 889         rc = cfs_trace_allocate_string_buffer(&str, usr_str_nob + 1);
 890         if (rc != 0)
 891                 return rc;
 892
 893         rc = cfs_trace_copyin_string(str, usr_str_nob + 1,
 894                                  usr_str, usr_str_nob);
 895         if (rc == 0)
 896                 rc = cfs_trace_daemon_command(str);
 897
 898         cfs_trace_free_string_buffer(str, usr_str_nob + 1);
 899         return rc;
 900 }
 901
 902 int cfs_trace_set_debug_mb(int mb)
 903 {
 904         int i;
 905         int j;
 906         int pages;
 907         int limit = cfs_trace_max_debug_mb();
 908         struct cfs_trace_cpu_data *tcd;
 909
 910         if (mb < cfs_num_possible_cpus()) {
 911                 printk(KERN_ERR "Cannot set debug_mb to %d, the value should be >= %d\n",
 912                        mb, num_possible_cpus());
 913                 return -EINVAL;
 914         }
 915
 916         if (mb > limit) {
 917                 printk(CFS_KERN_ERR "Lustre: Refusing to set debug buffer size "
 918                        "to %dMB - limit is %d\n", mb, limit);
 919                 return -EINVAL;
 920         }
 921
 922         mb /= cfs_num_possible_cpus();
 923         pages = mb << (20 - CFS_PAGE_SHIFT);
 924
 925         cfs_tracefile_write_lock();
 926
 927         cfs_tcd_for_each(tcd, i, j)
 928                 tcd->tcd_max_pages = (pages * tcd->tcd_pages_factor) / 100;
 929
 930         cfs_tracefile_write_unlock();
 931
 932         return 0;
 933 }
 934
 935 int cfs_trace_set_debug_mb_usrstr(void *usr_str, int usr_str_nob)
 936 {
 937         char     str[32];
 938         int      rc;
 939
 940         rc = cfs_trace_copyin_string(str, sizeof(str), usr_str, usr_str_nob);
 941         if (rc < 0)
 942                 return rc;
 943
 944         return cfs_trace_set_debug_mb(simple_strtoul(str, NULL, 0));
 945 }
 946
 947 int cfs_trace_get_debug_mb(void)
 948 {
 949         int i;
 950         int j;
 951         struct cfs_trace_cpu_data *tcd;
 952         int total_pages = 0;
 953
 954         cfs_tracefile_read_lock();
 955
 956         cfs_tcd_for_each(tcd, i, j)
 957                 total_pages += tcd->tcd_max_pages;
 958
 959         cfs_tracefile_read_unlock();
 960
 961         return (total_pages >> (20 - CFS_PAGE_SHIFT)) + 1;
 962 }
 963
 964 static int tracefiled(void *arg)
 965 {
 966         struct page_collection pc;
 967         struct tracefiled_ctl *tctl = arg;
 968         struct cfs_trace_page *tage;
 969         struct cfs_trace_page *tmp;
 970         cfs_file_t *filp;
 971         int last_loop = 0;
 972         int rc;
 973
 974         CFS_DECL_MMSPACE;
 975
 976         /* we're started late enough that we pick up init's fs context */
 977         /* this is so broken in uml?  what on earth is going on? */
 978         cfs_daemonize("ktracefiled");
 979
 980         cfs_spin_lock_init(&pc.pc_lock);
 981         cfs_complete(&tctl->tctl_start);
 982
 983         while (1) {
 984                 cfs_waitlink_t __wait;
 985
 986                 pc.pc_want_daemon_pages = 0;
 987                 collect_pages(&pc);
 988                 if (cfs_list_empty(&pc.pc_pages))
 989                         goto end_loop;
 990
 991                 filp = NULL;
 992                 cfs_tracefile_read_lock();
 993                 if (cfs_tracefile[0] != 0) {
 994                         filp = cfs_filp_open(cfs_tracefile,
 995                                              O_CREAT | O_RDWR | O_LARGEFILE,
 996                                              0600, &rc);
 997                         if (!(filp))
 998                                 printk(CFS_KERN_WARNING "couldn't open %s: "
 999                                        "%d\n", cfs_tracefile, rc);
1000                 }
1001                 cfs_tracefile_read_unlock();
1002                 if (filp == NULL) {
1003                         put_pages_on_daemon_list(&pc);
1004                         __LASSERT(cfs_list_empty(&pc.pc_pages));
1005                         goto end_loop;
1006                 }
1007
1008                 CFS_MMSPACE_OPEN;
1009
1010                 cfs_list_for_each_entry_safe_typed(tage, tmp, &pc.pc_pages,
1011                                                    struct cfs_trace_page,
1012                                                    linkage) {
1013                         static loff_t f_pos;
1014
1015                         __LASSERT_TAGE_INVARIANT(tage);
1016
1017                         if (f_pos >= (off_t)cfs_tracefile_size)
1018                                 f_pos = 0;
1019                         else if (f_pos > (off_t)cfs_filp_size(filp))
1020                                 f_pos = cfs_filp_size(filp);
1021
1022                         rc = cfs_filp_write(filp, cfs_page_address(tage->page),
1023                                             tage->used, &f_pos);
1024                         if (rc != (int)tage->used) {
1025                                 printk(CFS_KERN_WARNING "wanted to write %u "
1026                                        "but wrote %d\n", tage->used, rc);
1027                                 put_pages_back(&pc);
1028                                 __LASSERT(cfs_list_empty(&pc.pc_pages));
1029                         }
1030                 }
1031                 CFS_MMSPACE_CLOSE;
1032
1033                 cfs_filp_close(filp);
1034                 put_pages_on_daemon_list(&pc);
1035                 if (!cfs_list_empty(&pc.pc_pages)) {
1036                         int i;
1037
1038                         printk(CFS_KERN_ALERT "Lustre: trace pages aren't "
1039                                " empty\n");
1040                         printk(CFS_KERN_ERR "total cpus(%d): ",
1041                                cfs_num_possible_cpus());
1042                         for (i = 0; i < cfs_num_possible_cpus(); i++)
1043                                 if (cfs_cpu_online(i))
1044                                         printk(CFS_KERN_ERR "%d(on) ", i);
1045                                 else
1046                                         printk(CFS_KERN_ERR "%d(off) ", i);
1047                         printk(CFS_KERN_ERR "\n");
1048
1049                         i = 0;
1050                         cfs_list_for_each_entry_safe(tage, tmp, &pc.pc_pages,
1051                                                      linkage)
1052                                 printk(CFS_KERN_ERR "page %d belongs to cpu "
1053                                        "%d\n", ++i, tage->cpu);
1054                         printk(CFS_KERN_ERR "There are %d pages unwritten\n",
1055                                i);
1056                 }
1057                 __LASSERT(cfs_list_empty(&pc.pc_pages));
1058 end_loop:
1059                 if (cfs_atomic_read(&tctl->tctl_shutdown)) {
1060                         if (last_loop == 0) {
1061                                 last_loop = 1;
1062                                 continue;
1063                         } else {
1064                                 break;
1065                         }
1066                 }
1067                 cfs_waitlink_init(&__wait);
1068                 cfs_waitq_add(&tctl->tctl_waitq, &__wait);
1069                 cfs_set_current_state(CFS_TASK_INTERRUPTIBLE);
1070                 cfs_waitq_timedwait(&__wait, CFS_TASK_INTERRUPTIBLE,
1071                                     cfs_time_seconds(1));
1072                 cfs_waitq_del(&tctl->tctl_waitq, &__wait);
1073         }
1074         cfs_complete(&tctl->tctl_stop);
1075         return 0;
1076 }
1077
1078 int cfs_trace_start_thread(void)
1079 {
1080         struct tracefiled_ctl *tctl = &trace_tctl;
1081         int rc = 0;
1082
1083         cfs_mutex_down(&cfs_trace_thread_sem);
1084         if (thread_running)
1085                 goto out;
1086
1087         cfs_init_completion(&tctl->tctl_start);
1088         cfs_init_completion(&tctl->tctl_stop);
1089         cfs_waitq_init(&tctl->tctl_waitq);
1090         cfs_atomic_set(&tctl->tctl_shutdown, 0);
1091
1092         if (cfs_kernel_thread(tracefiled, tctl, 0) < 0) {
1093                 rc = -ECHILD;
1094                 goto out;
1095         }
1096
1097         cfs_wait_for_completion(&tctl->tctl_start);
1098         thread_running = 1;
1099 out:
1100         cfs_mutex_up(&cfs_trace_thread_sem);
1101         return rc;
1102 }
1103
1104 void cfs_trace_stop_thread(void)
1105 {
1106         struct tracefiled_ctl *tctl = &trace_tctl;
1107
1108         cfs_mutex_down(&cfs_trace_thread_sem);
1109         if (thread_running) {
1110                 printk(CFS_KERN_INFO
1111                        "Lustre: shutting down debug daemon thread...\n");
1112                 cfs_atomic_set(&tctl->tctl_shutdown, 1);
1113                 cfs_wait_for_completion(&tctl->tctl_stop);
1114                 thread_running = 0;
1115         }
1116         cfs_mutex_up(&cfs_trace_thread_sem);
1117 }
1118
1119 int cfs_tracefile_init(int max_pages)
1120 {
1121         struct cfs_trace_cpu_data *tcd;
1122         int                    i;
1123         int                    j;
1124         int                    rc;
1125         int                    factor;
1126
1127         rc = cfs_tracefile_init_arch();
1128         if (rc != 0)
1129                 return rc;
1130
1131         cfs_tcd_for_each(tcd, i, j) {
1132                 /* tcd_pages_factor is initialized int tracefile_init_arch. */
1133                 factor = tcd->tcd_pages_factor;
1134                 CFS_INIT_LIST_HEAD(&tcd->tcd_pages);
1135                 CFS_INIT_LIST_HEAD(&tcd->tcd_stock_pages);
1136                 CFS_INIT_LIST_HEAD(&tcd->tcd_daemon_pages);
1137                 tcd->tcd_cur_pages = 0;
1138                 tcd->tcd_cur_stock_pages = 0;
1139                 tcd->tcd_cur_daemon_pages = 0;
1140                 tcd->tcd_max_pages = (max_pages * factor) / 100;
1141                 LASSERT(tcd->tcd_max_pages > 0);
1142                 tcd->tcd_shutting_down = 0;
1143         }
1144
1145         return 0;
1146 }
1147
1148 static void trace_cleanup_on_all_cpus(void)
1149 {
1150         struct cfs_trace_cpu_data *tcd;
1151         struct cfs_trace_page *tage;
1152         struct cfs_trace_page *tmp;
1153         int i, cpu;
1154
1155         cfs_for_each_possible_cpu(cpu) {
1156                 cfs_tcd_for_each_type_lock(tcd, i, cpu) {
1157                         tcd->tcd_shutting_down = 1;
1158
1159                         cfs_list_for_each_entry_safe_typed(tage, tmp,
1160                                                            &tcd->tcd_pages,
1161                                                            struct cfs_trace_page,
1162                                                            linkage) {
1163                                 __LASSERT_TAGE_INVARIANT(tage);
1164
1165                                 cfs_list_del(&tage->linkage);
1166                                 cfs_tage_free(tage);
1167                         }
1168
1169                         tcd->tcd_cur_pages = 0;
1170                 }
1171         }
1172 }
1173
1174 static void cfs_trace_cleanup(void)
1175 {
1176         struct page_collection pc;
1177
1178         CFS_INIT_LIST_HEAD(&pc.pc_pages);
1179         cfs_spin_lock_init(&pc.pc_lock);
1180
1181         trace_cleanup_on_all_cpus();
1182
1183         cfs_tracefile_fini_arch();
1184 }
1185
1186 void cfs_tracefile_exit(void)
1187 {
1188         cfs_trace_stop_thread();
1189         cfs_trace_cleanup();
1190 }