1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
6 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
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.
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).
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
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
29 * Copyright 2008 Sun Microsystems, Inc. All rights reserved
30 * Use is subject to license terms.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
36 * libcfs/libcfs/tracefile.c
38 * Author: Zach Brown <zab@clusterfs.com>
39 * Author: Phil Schwan <phil@clusterfs.com>
43 #define DEBUG_SUBSYSTEM S_LNET
44 #define LUSTRE_TRACEFILE_PRIVATE
45 #include "tracefile.h"
47 #include <libcfs/libcfs.h>
49 /* XXX move things up to the top, comment */
50 union trace_data_union (*trace_data[TCD_MAX_TYPES])[NR_CPUS] __cacheline_aligned;
52 char tracefile[TRACEFILE_NAME_SIZE];
53 long long tracefile_size = TRACEFILE_SIZE;
54 static struct tracefiled_ctl trace_tctl;
55 struct semaphore trace_thread_sem;
56 static int thread_running = 0;
58 atomic_t tage_allocated = ATOMIC_INIT(0);
60 static void put_pages_on_tcd_daemon_list(struct page_collection *pc,
61 struct trace_cpu_data *tcd);
63 static inline struct trace_page *tage_from_list(struct list_head *list)
65 return list_entry(list, struct trace_page, linkage);
68 static struct trace_page *tage_alloc(int gfp)
71 struct trace_page *tage;
74 * Don't spam console with allocation failures: they will be reported
75 * by upper layer anyway.
77 gfp |= CFS_ALLOC_NOWARN;
78 page = cfs_alloc_page(gfp);
82 tage = cfs_alloc(sizeof(*tage), gfp);
89 atomic_inc(&tage_allocated);
93 static void tage_free(struct trace_page *tage)
95 __LASSERT(tage != NULL);
96 __LASSERT(tage->page != NULL);
98 cfs_free_page(tage->page);
100 atomic_dec(&tage_allocated);
103 static void tage_to_tail(struct trace_page *tage, struct list_head *queue)
105 __LASSERT(tage != NULL);
106 __LASSERT(queue != NULL);
108 list_move_tail(&tage->linkage, queue);
111 int trace_refill_stock(struct trace_cpu_data *tcd, int gfp,
112 struct list_head *stock)
117 * XXX nikita: do NOT call portals_debug_msg() (CDEBUG/ENTRY/EXIT)
118 * from here: this will lead to infinite recursion.
121 for (i = 0; i + tcd->tcd_cur_stock_pages < TCD_STOCK_PAGES ; ++ i) {
122 struct trace_page *tage;
124 tage = tage_alloc(gfp);
127 list_add_tail(&tage->linkage, stock);
132 /* return a page that has 'len' bytes left at the end */
133 static struct trace_page *trace_get_tage_try(struct trace_cpu_data *tcd,
136 struct trace_page *tage;
138 if (tcd->tcd_cur_pages > 0) {
139 __LASSERT(!list_empty(&tcd->tcd_pages));
140 tage = tage_from_list(tcd->tcd_pages.prev);
141 if (tage->used + len <= CFS_PAGE_SIZE)
145 if (tcd->tcd_cur_pages < tcd->tcd_max_pages) {
146 if (tcd->tcd_cur_stock_pages > 0) {
147 tage = tage_from_list(tcd->tcd_stock_pages.prev);
148 -- tcd->tcd_cur_stock_pages;
149 list_del_init(&tage->linkage);
151 tage = tage_alloc(CFS_ALLOC_ATOMIC);
153 if (printk_ratelimit())
155 "cannot allocate a tage (%ld)\n",
162 tage->cpu = smp_processor_id();
163 tage->type = tcd->tcd_type;
164 list_add_tail(&tage->linkage, &tcd->tcd_pages);
165 tcd->tcd_cur_pages++;
167 if (tcd->tcd_cur_pages > 8 && thread_running) {
168 struct tracefiled_ctl *tctl = &trace_tctl;
170 * wake up tracefiled to process some pages.
172 cfs_waitq_signal(&tctl->tctl_waitq);
179 static void tcd_shrink(struct trace_cpu_data *tcd)
181 int pgcount = tcd->tcd_cur_pages / 10;
182 struct page_collection pc;
183 struct trace_page *tage;
184 struct trace_page *tmp;
187 * XXX nikita: do NOT call portals_debug_msg() (CDEBUG/ENTRY/EXIT)
188 * from here: this will lead to infinite recursion.
191 if (printk_ratelimit())
192 printk(KERN_WARNING "debug daemon buffer overflowed; "
193 "discarding 10%% of pages (%d of %ld)\n",
194 pgcount + 1, tcd->tcd_cur_pages);
196 CFS_INIT_LIST_HEAD(&pc.pc_pages);
197 spin_lock_init(&pc.pc_lock);
199 cfs_list_for_each_entry_safe_typed(tage, tmp, &tcd->tcd_pages,
200 struct trace_page, linkage) {
204 list_move_tail(&tage->linkage, &pc.pc_pages);
205 tcd->tcd_cur_pages--;
207 put_pages_on_tcd_daemon_list(&pc, tcd);
210 /* return a page that has 'len' bytes left at the end */
211 static struct trace_page *trace_get_tage(struct trace_cpu_data *tcd,
214 struct trace_page *tage;
217 * XXX nikita: do NOT call portals_debug_msg() (CDEBUG/ENTRY/EXIT)
218 * from here: this will lead to infinite recursion.
221 if (len > CFS_PAGE_SIZE) {
223 "cowardly refusing to write %lu bytes in a page\n", len);
227 tage = trace_get_tage_try(tcd, len);
232 if (tcd->tcd_cur_pages > 0) {
233 tage = tage_from_list(tcd->tcd_pages.next);
235 tage_to_tail(tage, &tcd->tcd_pages);
240 int libcfs_debug_vmsg2(cfs_debug_limit_state_t *cdls, int subsys, int mask,
241 const char *file, const char *fn, const int line,
242 const char *format1, va_list args,
243 const char *format2, ...)
245 struct trace_cpu_data *tcd = NULL;
246 struct ptldebug_header header;
247 struct trace_page *tage;
248 /* string_buf is used only if tcd != NULL, and is always set then */
249 char *string_buf = NULL;
252 int needed = 85; /* average message length */
259 if (strchr(file, '/'))
260 file = strrchr(file, '/') + 1;
263 set_ptldebug_header(&header, subsys, mask, line, CDEBUG_STACK());
265 tcd = trace_get_tcd();
266 if (tcd == NULL) /* arch may not log in IRQ context */
269 if (tcd->tcd_shutting_down) {
275 depth = __current_nesting_level();
276 known_size = strlen(file) + 1 + depth;
278 known_size += strlen(fn) + 1;
280 if (libcfs_debug_binary)
281 known_size += sizeof(header);
284 * '2' used because vsnprintf return real size required for output
285 * _without_ terminating NULL.
286 * if needed is to small for this format.
288 for (i = 0; i < 2; i++) {
289 tage = trace_get_tage(tcd, needed + known_size + 1);
291 if (needed + known_size > CFS_PAGE_SIZE)
299 string_buf = (char *)cfs_page_address(tage->page) +
300 tage->used + known_size;
302 max_nob = CFS_PAGE_SIZE - tage->used - known_size;
304 printk(KERN_EMERG "negative max_nob: %i\n", max_nob);
314 needed = vsnprintf(string_buf, max_nob, format1, ap);
319 remain = max_nob - needed;
323 va_start(ap, format2);
324 needed += vsnprintf(string_buf + needed, remain,
329 if (needed < max_nob) /* well. printing ok.. */
333 if (*(string_buf+needed-1) != '\n')
334 printk(KERN_INFO "format at %s:%d:%s doesn't end in newline\n",
337 header.ph_len = known_size + needed;
338 debug_buf = (char *)cfs_page_address(tage->page) + tage->used;
340 if (libcfs_debug_binary) {
341 memcpy(debug_buf, &header, sizeof(header));
342 tage->used += sizeof(header);
343 debug_buf += sizeof(header);
346 /* indent message according to the nesting level */
347 while (depth-- > 0) {
348 *(debug_buf++) = '.';
352 strcpy(debug_buf, file);
353 tage->used += strlen(file) + 1;
354 debug_buf += strlen(file) + 1;
357 strcpy(debug_buf, fn);
358 tage->used += strlen(fn) + 1;
359 debug_buf += strlen(fn) + 1;
362 __LASSERT(debug_buf == string_buf);
364 tage->used += needed;
365 __LASSERT (tage->used <= CFS_PAGE_SIZE);
368 if ((mask & libcfs_printk) == 0) {
369 /* no console output requested */
376 if (libcfs_console_ratelimit &&
377 cdls->cdls_next != 0 && /* not first time ever */
378 !cfs_time_after(cfs_time_current(), cdls->cdls_next)) {
379 /* skipping a console message */
386 if (cfs_time_after(cfs_time_current(), cdls->cdls_next +
387 libcfs_console_max_delay
388 + cfs_time_seconds(10))) {
389 /* last timeout was a long time ago */
390 cdls->cdls_delay /= libcfs_console_backoff * 4;
392 cdls->cdls_delay *= libcfs_console_backoff;
394 if (cdls->cdls_delay < libcfs_console_min_delay)
395 cdls->cdls_delay = libcfs_console_min_delay;
396 else if (cdls->cdls_delay > libcfs_console_max_delay)
397 cdls->cdls_delay = libcfs_console_max_delay;
400 /* ensure cdls_next is never zero after it's been seen */
401 cdls->cdls_next = (cfs_time_current() + cdls->cdls_delay) | 1;
405 print_to_console(&header, mask, string_buf, needed, file, fn);
408 string_buf = trace_get_console_buffer();
411 if (format1 != NULL) {
413 needed = vsnprintf(string_buf, TRACE_CONSOLE_BUFFER_SIZE, format1, ap);
416 if (format2 != NULL) {
417 remain = TRACE_CONSOLE_BUFFER_SIZE - needed;
419 va_start(ap, format2);
420 needed += vsnprintf(string_buf+needed, remain, format2, ap);
424 print_to_console(&header, mask,
425 string_buf, needed, file, fn);
427 trace_put_console_buffer(string_buf);
430 if (cdls != NULL && cdls->cdls_count != 0) {
431 string_buf = trace_get_console_buffer();
433 needed = snprintf(string_buf, TRACE_CONSOLE_BUFFER_SIZE,
434 "Skipped %d previous similar message%s\n",
435 cdls->cdls_count, (cdls->cdls_count > 1) ? "s" : "");
437 print_to_console(&header, mask,
438 string_buf, needed, file, fn);
440 trace_put_console_buffer(string_buf);
441 cdls->cdls_count = 0;
446 EXPORT_SYMBOL(libcfs_debug_vmsg2);
449 libcfs_assertion_failed(const char *expr, const char *file,
450 const char *func, const int line)
452 libcfs_debug_msg(NULL, 0, D_EMERG, file, func, line,
453 "ASSERTION(%s) failed\n", expr);
454 /* cfs_enter_debugger(); */
455 lbug_with_loc(file, func, line);
457 EXPORT_SYMBOL(libcfs_assertion_failed);
460 trace_assertion_failed(const char *str,
461 const char *fn, const char *file, int line)
463 struct ptldebug_header hdr;
465 libcfs_panic_in_progress = 1;
466 libcfs_catastrophe = 1;
469 set_ptldebug_header(&hdr, DEBUG_SUBSYSTEM, D_EMERG, line,
472 print_to_console(&hdr, D_EMERG, str, strlen(str), file, fn);
474 LIBCFS_PANIC("Lustre debug assertion failure\n");
480 panic_collect_pages(struct page_collection *pc)
482 /* Do the collect_pages job on a single CPU: assumes that all other
483 * CPUs have been stopped during a panic. If this isn't true for some
484 * arch, this will have to be implemented separately in each arch. */
487 struct trace_cpu_data *tcd;
489 CFS_INIT_LIST_HEAD(&pc->pc_pages);
491 tcd_for_each(tcd, i, j) {
492 list_splice_init(&tcd->tcd_pages, &pc->pc_pages);
493 tcd->tcd_cur_pages = 0;
495 if (pc->pc_want_daemon_pages) {
496 list_splice_init(&tcd->tcd_daemon_pages, &pc->pc_pages);
497 tcd->tcd_cur_daemon_pages = 0;
502 static void collect_pages_on_all_cpus(struct page_collection *pc)
504 struct trace_cpu_data *tcd;
507 spin_lock(&pc->pc_lock);
508 for_each_possible_cpu(cpu) {
509 tcd_for_each_type_lock(tcd, i, cpu) {
510 list_splice_init(&tcd->tcd_pages, &pc->pc_pages);
511 tcd->tcd_cur_pages = 0;
512 if (pc->pc_want_daemon_pages) {
513 list_splice_init(&tcd->tcd_daemon_pages,
515 tcd->tcd_cur_daemon_pages = 0;
519 spin_unlock(&pc->pc_lock);
522 static void collect_pages(struct page_collection *pc)
524 CFS_INIT_LIST_HEAD(&pc->pc_pages);
526 if (libcfs_panic_in_progress)
527 panic_collect_pages(pc);
529 collect_pages_on_all_cpus(pc);
532 static void put_pages_back_on_all_cpus(struct page_collection *pc)
534 struct trace_cpu_data *tcd;
535 struct list_head *cur_head;
536 struct trace_page *tage;
537 struct trace_page *tmp;
540 spin_lock(&pc->pc_lock);
541 for_each_possible_cpu(cpu) {
542 tcd_for_each_type_lock(tcd, i, cpu) {
543 cur_head = tcd->tcd_pages.next;
545 cfs_list_for_each_entry_safe_typed(tage, tmp,
550 __LASSERT_TAGE_INVARIANT(tage);
552 if (tage->cpu != cpu || tage->type != i)
555 tage_to_tail(tage, cur_head);
556 tcd->tcd_cur_pages++;
560 spin_unlock(&pc->pc_lock);
563 static void put_pages_back(struct page_collection *pc)
565 if (!libcfs_panic_in_progress)
566 put_pages_back_on_all_cpus(pc);
569 /* Add pages to a per-cpu debug daemon ringbuffer. This buffer makes sure that
570 * we have a good amount of data at all times for dumping during an LBUG, even
571 * if we have been steadily writing (and otherwise discarding) pages via the
573 static void put_pages_on_tcd_daemon_list(struct page_collection *pc,
574 struct trace_cpu_data *tcd)
576 struct trace_page *tage;
577 struct trace_page *tmp;
579 spin_lock(&pc->pc_lock);
580 cfs_list_for_each_entry_safe_typed(tage, tmp, &pc->pc_pages,
581 struct trace_page, linkage) {
583 __LASSERT_TAGE_INVARIANT(tage);
585 if (tage->cpu != tcd->tcd_cpu || tage->type != tcd->tcd_type)
588 tage_to_tail(tage, &tcd->tcd_daemon_pages);
589 tcd->tcd_cur_daemon_pages++;
591 if (tcd->tcd_cur_daemon_pages > tcd->tcd_max_pages) {
592 struct trace_page *victim;
594 __LASSERT(!list_empty(&tcd->tcd_daemon_pages));
595 victim = tage_from_list(tcd->tcd_daemon_pages.next);
597 __LASSERT_TAGE_INVARIANT(victim);
599 list_del(&victim->linkage);
601 tcd->tcd_cur_daemon_pages--;
604 spin_unlock(&pc->pc_lock);
607 static void put_pages_on_daemon_list(struct page_collection *pc)
609 struct trace_cpu_data *tcd;
612 for_each_possible_cpu(cpu) {
613 tcd_for_each_type_lock(tcd, i, cpu)
614 put_pages_on_tcd_daemon_list(pc, tcd);
618 void trace_debug_print(void)
620 struct page_collection pc;
621 struct trace_page *tage;
622 struct trace_page *tmp;
624 spin_lock_init(&pc.pc_lock);
626 pc.pc_want_daemon_pages = 1;
628 cfs_list_for_each_entry_safe_typed(tage, tmp, &pc.pc_pages,
629 struct trace_page, linkage) {
633 __LASSERT_TAGE_INVARIANT(tage);
636 p = cfs_page_address(page);
637 while (p < ((char *)cfs_page_address(page) + tage->used)) {
638 struct ptldebug_header *hdr;
643 p += strlen(file) + 1;
646 len = hdr->ph_len - (int)(p - (char *)hdr);
648 print_to_console(hdr, D_EMERG, p, len, file, fn);
653 list_del(&tage->linkage);
658 int tracefile_dump_all_pages(char *filename)
660 struct page_collection pc;
662 struct trace_page *tage;
663 struct trace_page *tmp;
668 tracefile_write_lock();
670 filp = cfs_filp_open(filename,
671 O_CREAT|O_EXCL|O_WRONLY|O_LARGEFILE, 0600, &rc);
674 printk(KERN_ERR "LustreError: can't open %s for dump: rc %d\n",
679 spin_lock_init(&pc.pc_lock);
680 pc.pc_want_daemon_pages = 1;
682 if (list_empty(&pc.pc_pages)) {
687 /* ok, for now, just write the pages. in the future we'll be building
688 * iobufs with the pages and calling generic_direct_IO */
690 cfs_list_for_each_entry_safe_typed(tage, tmp, &pc.pc_pages,
691 struct trace_page, linkage) {
693 __LASSERT_TAGE_INVARIANT(tage);
695 rc = cfs_filp_write(filp, cfs_page_address(tage->page),
696 tage->used, cfs_filp_poff(filp));
697 if (rc != (int)tage->used) {
698 printk(KERN_WARNING "wanted to write %u but wrote "
699 "%d\n", tage->used, rc);
701 __LASSERT(list_empty(&pc.pc_pages));
704 list_del(&tage->linkage);
708 rc = cfs_filp_fsync(filp);
710 printk(KERN_ERR "sync returns %d\n", rc);
712 cfs_filp_close(filp);
714 tracefile_write_unlock();
718 void trace_flush_pages(void)
720 struct page_collection pc;
721 struct trace_page *tage;
722 struct trace_page *tmp;
724 spin_lock_init(&pc.pc_lock);
726 pc.pc_want_daemon_pages = 1;
728 cfs_list_for_each_entry_safe_typed(tage, tmp, &pc.pc_pages,
729 struct trace_page, linkage) {
731 __LASSERT_TAGE_INVARIANT(tage);
733 list_del(&tage->linkage);
738 int trace_copyin_string(char *knl_buffer, int knl_buffer_nob,
739 const char *usr_buffer, int usr_buffer_nob)
743 if (usr_buffer_nob > knl_buffer_nob)
746 if (copy_from_user((void *)knl_buffer,
747 (void *)usr_buffer, usr_buffer_nob))
750 nob = strnlen(knl_buffer, usr_buffer_nob);
751 while (nob-- >= 0) /* strip trailing whitespace */
752 if (!isspace(knl_buffer[nob]))
755 if (nob < 0) /* empty string */
758 if (nob == knl_buffer_nob) /* no space to terminate */
761 knl_buffer[nob + 1] = 0; /* terminate */
765 int trace_copyout_string(char *usr_buffer, int usr_buffer_nob,
766 const char *knl_buffer, char *append)
768 /* NB if 'append' != NULL, it's a single character to append to the
769 * copied out string - usually "\n", for /proc entries and "" (i.e. a
770 * terminating zero byte) for sysctl entries */
771 int nob = strlen(knl_buffer);
773 if (nob > usr_buffer_nob)
774 nob = usr_buffer_nob;
776 if (copy_to_user(usr_buffer, knl_buffer, nob))
779 if (append != NULL && nob < usr_buffer_nob) {
780 if (copy_to_user(usr_buffer + nob, append, 1))
789 int trace_allocate_string_buffer(char **str, int nob)
791 if (nob > 2 * CFS_PAGE_SIZE) /* string must be "sensible" */
794 *str = cfs_alloc(nob, CFS_ALLOC_STD | CFS_ALLOC_ZERO);
801 void trace_free_string_buffer(char *str, int nob)
806 int trace_dump_debug_buffer_usrstr(void *usr_str, int usr_str_nob)
811 rc = trace_allocate_string_buffer(&str, usr_str_nob + 1);
815 rc = trace_copyin_string(str, usr_str_nob + 1,
816 usr_str, usr_str_nob);
820 #if !defined(__WINNT__)
826 rc = tracefile_dump_all_pages(str);
828 trace_free_string_buffer(str, usr_str_nob + 1);
832 int trace_daemon_command(char *str)
836 tracefile_write_lock();
838 if (strcmp(str, "stop") == 0) {
839 tracefile_write_unlock();
841 tracefile_write_lock();
842 memset(tracefile, 0, sizeof(tracefile));
844 } else if (strncmp(str, "size=", 5) == 0) {
845 tracefile_size = simple_strtoul(str + 5, NULL, 0);
846 if (tracefile_size < 10 || tracefile_size > 20480)
847 tracefile_size = TRACEFILE_SIZE;
849 tracefile_size <<= 20;
851 } else if (strlen(str) >= sizeof(tracefile)) {
854 } else if (str[0] != '/') {
858 strcpy(tracefile, str);
860 printk(KERN_INFO "Lustre: debug daemon will attempt to start writing "
861 "to %s (%lukB max)\n", tracefile,
862 (long)(tracefile_size >> 10));
864 trace_start_thread();
867 tracefile_write_unlock();
871 int trace_daemon_command_usrstr(void *usr_str, int usr_str_nob)
876 rc = trace_allocate_string_buffer(&str, usr_str_nob + 1);
880 rc = trace_copyin_string(str, usr_str_nob + 1,
881 usr_str, usr_str_nob);
883 rc = trace_daemon_command(str);
885 trace_free_string_buffer(str, usr_str_nob + 1);
889 int trace_set_debug_mb(int mb)
894 int limit = trace_max_debug_mb();
895 struct trace_cpu_data *tcd;
897 if (mb < num_possible_cpus())
901 printk(KERN_ERR "Lustre: Refusing to set debug buffer size to "
902 "%dMB - limit is %d\n", mb, limit);
906 mb /= num_possible_cpus();
907 pages = mb << (20 - CFS_PAGE_SHIFT);
909 tracefile_write_lock();
911 tcd_for_each(tcd, i, j)
912 tcd->tcd_max_pages = (pages * tcd->tcd_pages_factor) / 100;
914 tracefile_write_unlock();
919 int trace_set_debug_mb_usrstr(void *usr_str, int usr_str_nob)
924 rc = trace_copyin_string(str, sizeof(str), usr_str, usr_str_nob);
928 return trace_set_debug_mb(simple_strtoul(str, NULL, 0));
931 int trace_get_debug_mb(void)
935 struct trace_cpu_data *tcd;
938 tracefile_read_lock();
940 tcd_for_each(tcd, i, j)
941 total_pages += tcd->tcd_max_pages;
943 tracefile_read_unlock();
945 return (total_pages >> (20 - CFS_PAGE_SHIFT)) + 1;
948 static int tracefiled(void *arg)
950 struct page_collection pc;
951 struct tracefiled_ctl *tctl = arg;
952 struct trace_page *tage;
953 struct trace_page *tmp;
954 struct ptldebug_header *hdr;
961 /* we're started late enough that we pick up init's fs context */
962 /* this is so broken in uml? what on earth is going on? */
963 cfs_daemonize("ktracefiled");
965 spin_lock_init(&pc.pc_lock);
966 complete(&tctl->tctl_start);
969 cfs_waitlink_t __wait;
971 pc.pc_want_daemon_pages = 0;
973 if (list_empty(&pc.pc_pages))
977 tracefile_read_lock();
978 if (tracefile[0] != 0) {
979 filp = cfs_filp_open(tracefile,
980 O_CREAT | O_RDWR | O_LARGEFILE,
983 printk(KERN_WARNING "couldn't open %s: %d\n",
986 tracefile_read_unlock();
988 put_pages_on_daemon_list(&pc);
989 __LASSERT(list_empty(&pc.pc_pages));
995 /* mark the first header, so we can sort in chunks */
996 tage = tage_from_list(pc.pc_pages.next);
997 __LASSERT_TAGE_INVARIANT(tage);
999 hdr = cfs_page_address(tage->page);
1000 hdr->ph_flags |= PH_FLAG_FIRST_RECORD;
1002 cfs_list_for_each_entry_safe_typed(tage, tmp, &pc.pc_pages,
1003 struct trace_page, linkage) {
1004 static loff_t f_pos;
1006 __LASSERT_TAGE_INVARIANT(tage);
1008 if (f_pos >= (off_t)tracefile_size)
1010 else if (f_pos > (off_t)cfs_filp_size(filp))
1011 f_pos = cfs_filp_size(filp);
1013 rc = cfs_filp_write(filp, cfs_page_address(tage->page),
1014 tage->used, &f_pos);
1015 if (rc != (int)tage->used) {
1016 printk(KERN_WARNING "wanted to write %u but "
1017 "wrote %d\n", tage->used, rc);
1018 put_pages_back(&pc);
1019 __LASSERT(list_empty(&pc.pc_pages));
1024 cfs_filp_close(filp);
1025 put_pages_on_daemon_list(&pc);
1026 if (!list_empty(&pc.pc_pages)) {
1029 printk(KERN_ALERT "Lustre: trace pages aren't empty\n");
1030 printk(KERN_ERR "total cpus(%d): ", num_possible_cpus());
1031 for (i = 0; i < num_possible_cpus(); i++)
1033 printk(KERN_ERR "%d(on) ", i);
1035 printk(KERN_ERR "%d(off) ", i);
1036 printk(KERN_ERR "\n");
1039 list_for_each_entry_safe(tage, tmp, &pc.pc_pages,
1041 printk(KERN_ERR "page %d belongs to cpu %d\n",
1043 printk(KERN_ERR "There are %d pages unwritten\n", i);
1045 __LASSERT(list_empty(&pc.pc_pages));
1047 if (atomic_read(&tctl->tctl_shutdown)) {
1048 if (last_loop == 0) {
1055 cfs_waitlink_init(&__wait);
1056 cfs_waitq_add(&tctl->tctl_waitq, &__wait);
1057 set_current_state(TASK_INTERRUPTIBLE);
1058 cfs_waitq_timedwait(&__wait, CFS_TASK_INTERRUPTIBLE,
1059 cfs_time_seconds(1));
1060 cfs_waitq_del(&tctl->tctl_waitq, &__wait);
1062 complete(&tctl->tctl_stop);
1066 int trace_start_thread(void)
1068 struct tracefiled_ctl *tctl = &trace_tctl;
1071 mutex_down(&trace_thread_sem);
1075 init_completion(&tctl->tctl_start);
1076 init_completion(&tctl->tctl_stop);
1077 cfs_waitq_init(&tctl->tctl_waitq);
1078 atomic_set(&tctl->tctl_shutdown, 0);
1080 if (cfs_kernel_thread(tracefiled, tctl, 0) < 0) {
1085 wait_for_completion(&tctl->tctl_start);
1088 mutex_up(&trace_thread_sem);
1092 void trace_stop_thread(void)
1094 struct tracefiled_ctl *tctl = &trace_tctl;
1096 mutex_down(&trace_thread_sem);
1097 if (thread_running) {
1098 printk(KERN_INFO "Lustre: shutting down debug daemon thread...\n");
1099 atomic_set(&tctl->tctl_shutdown, 1);
1100 wait_for_completion(&tctl->tctl_stop);
1103 mutex_up(&trace_thread_sem);
1106 int tracefile_init(int max_pages)
1108 struct trace_cpu_data *tcd;
1114 rc = tracefile_init_arch();
1118 tcd_for_each(tcd, i, j) {
1119 /* tcd_pages_factor is initialized int tracefile_init_arch. */
1120 factor = tcd->tcd_pages_factor;
1121 CFS_INIT_LIST_HEAD(&tcd->tcd_pages);
1122 CFS_INIT_LIST_HEAD(&tcd->tcd_stock_pages);
1123 CFS_INIT_LIST_HEAD(&tcd->tcd_daemon_pages);
1124 tcd->tcd_cur_pages = 0;
1125 tcd->tcd_cur_stock_pages = 0;
1126 tcd->tcd_cur_daemon_pages = 0;
1127 tcd->tcd_max_pages = (max_pages * factor) / 100;
1128 LASSERT(tcd->tcd_max_pages > 0);
1129 tcd->tcd_shutting_down = 0;
1135 static void trace_cleanup_on_all_cpus(void)
1137 struct trace_cpu_data *tcd;
1138 struct trace_page *tage;
1139 struct trace_page *tmp;
1142 for_each_possible_cpu(cpu) {
1143 tcd_for_each_type_lock(tcd, i, cpu) {
1144 tcd->tcd_shutting_down = 1;
1146 cfs_list_for_each_entry_safe_typed(tage, tmp,
1150 __LASSERT_TAGE_INVARIANT(tage);
1152 list_del(&tage->linkage);
1156 tcd->tcd_cur_pages = 0;
1161 static void trace_cleanup(void)
1163 struct page_collection pc;
1165 CFS_INIT_LIST_HEAD(&pc.pc_pages);
1166 spin_lock_init(&pc.pc_lock);
1168 trace_cleanup_on_all_cpus();
1170 tracefile_fini_arch();
1173 void tracefile_exit(void)
1175 trace_stop_thread();