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 (c) 2008, 2010, Oracle and/or its affiliates. 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 cfs_trace_data_union (*cfs_trace_data[TCD_MAX_TYPES])[CFS_NR_CPUS] __cacheline_aligned;
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;
58 cfs_atomic_t cfs_tage_allocated = CFS_ATOMIC_INIT(0);
60 static void put_pages_on_tcd_daemon_list(struct page_collection *pc,
61 struct cfs_trace_cpu_data *tcd);
63 static inline struct cfs_trace_page *
64 cfs_tage_from_list(cfs_list_t *list)
66 return cfs_list_entry(list, struct cfs_trace_page, linkage);
69 static struct cfs_trace_page *cfs_tage_alloc(int gfp)
72 struct cfs_trace_page *tage;
74 /* My caller is trying to free memory */
75 if (!cfs_in_interrupt() && cfs_memory_pressure_get())
79 * Don't spam console with allocation failures: they will be reported
80 * by upper layer anyway.
82 gfp |= CFS_ALLOC_NOWARN;
83 page = cfs_alloc_page(gfp);
87 tage = cfs_alloc(sizeof(*tage), gfp);
94 cfs_atomic_inc(&cfs_tage_allocated);
98 static void cfs_tage_free(struct cfs_trace_page *tage)
100 __LASSERT(tage != NULL);
101 __LASSERT(tage->page != NULL);
103 cfs_free_page(tage->page);
105 cfs_atomic_dec(&cfs_tage_allocated);
108 static void cfs_tage_to_tail(struct cfs_trace_page *tage,
111 __LASSERT(tage != NULL);
112 __LASSERT(queue != NULL);
114 cfs_list_move_tail(&tage->linkage, queue);
117 int cfs_trace_refill_stock(struct cfs_trace_cpu_data *tcd, int gfp,
123 * XXX nikita: do NOT call portals_debug_msg() (CDEBUG/ENTRY/EXIT)
124 * from here: this will lead to infinite recursion.
127 for (i = 0; i + tcd->tcd_cur_stock_pages < TCD_STOCK_PAGES ; ++ i) {
128 struct cfs_trace_page *tage;
130 tage = cfs_tage_alloc(gfp);
133 cfs_list_add_tail(&tage->linkage, stock);
138 /* return a page that has 'len' bytes left at the end */
139 static struct cfs_trace_page *
140 cfs_trace_get_tage_try(struct cfs_trace_cpu_data *tcd, unsigned long len)
142 struct cfs_trace_page *tage;
144 if (tcd->tcd_cur_pages > 0) {
145 __LASSERT(!cfs_list_empty(&tcd->tcd_pages));
146 tage = cfs_tage_from_list(tcd->tcd_pages.prev);
147 if (tage->used + len <= CFS_PAGE_SIZE)
151 if (tcd->tcd_cur_pages < tcd->tcd_max_pages) {
152 if (tcd->tcd_cur_stock_pages > 0) {
153 tage = cfs_tage_from_list(tcd->tcd_stock_pages.prev);
154 -- tcd->tcd_cur_stock_pages;
155 cfs_list_del_init(&tage->linkage);
157 tage = cfs_tage_alloc(CFS_ALLOC_ATOMIC);
159 if (printk_ratelimit())
160 printk(CFS_KERN_WARNING
161 "cannot allocate a tage (%ld)\n",
168 tage->cpu = cfs_smp_processor_id();
169 tage->type = tcd->tcd_type;
170 cfs_list_add_tail(&tage->linkage, &tcd->tcd_pages);
171 tcd->tcd_cur_pages++;
173 if (tcd->tcd_cur_pages > 8 && thread_running) {
174 struct tracefiled_ctl *tctl = &trace_tctl;
176 * wake up tracefiled to process some pages.
178 cfs_waitq_signal(&tctl->tctl_waitq);
185 static void cfs_tcd_shrink(struct cfs_trace_cpu_data *tcd)
187 int pgcount = tcd->tcd_cur_pages / 10;
188 struct page_collection pc;
189 struct cfs_trace_page *tage;
190 struct cfs_trace_page *tmp;
193 * XXX nikita: do NOT call portals_debug_msg() (CDEBUG/ENTRY/EXIT)
194 * from here: this will lead to infinite recursion.
197 if (printk_ratelimit())
198 printk(CFS_KERN_WARNING "debug daemon buffer overflowed; "
199 "discarding 10%% of pages (%d of %ld)\n",
200 pgcount + 1, tcd->tcd_cur_pages);
202 CFS_INIT_LIST_HEAD(&pc.pc_pages);
203 cfs_spin_lock_init(&pc.pc_lock);
205 cfs_list_for_each_entry_safe_typed(tage, tmp, &tcd->tcd_pages,
206 struct cfs_trace_page, linkage) {
210 cfs_list_move_tail(&tage->linkage, &pc.pc_pages);
211 tcd->tcd_cur_pages--;
213 put_pages_on_tcd_daemon_list(&pc, tcd);
216 /* return a page that has 'len' bytes left at the end */
217 static struct cfs_trace_page *cfs_trace_get_tage(struct cfs_trace_cpu_data *tcd,
220 struct cfs_trace_page *tage;
223 * XXX nikita: do NOT call portals_debug_msg() (CDEBUG/ENTRY/EXIT)
224 * from here: this will lead to infinite recursion.
227 if (len > CFS_PAGE_SIZE) {
229 "cowardly refusing to write %lu bytes in a page\n", len);
233 tage = cfs_trace_get_tage_try(tcd, len);
238 if (tcd->tcd_cur_pages > 0) {
239 tage = cfs_tage_from_list(tcd->tcd_pages.next);
241 cfs_tage_to_tail(tage, &tcd->tcd_pages);
246 int libcfs_debug_msg(struct libcfs_debug_msg_data *msgdata,
247 const char *format, ...)
252 va_start(args, format);
253 rc = libcfs_debug_vmsg2(msgdata, format, args, NULL);
258 EXPORT_SYMBOL(libcfs_debug_msg);
260 int libcfs_debug_vmsg2(struct libcfs_debug_msg_data *msgdata,
261 const char *format1, va_list args,
262 const char *format2, ...)
264 struct cfs_trace_cpu_data *tcd = NULL;
265 struct ptldebug_header header = {0};
266 struct cfs_trace_page *tage;
267 /* string_buf is used only if tcd != NULL, and is always set then */
268 char *string_buf = NULL;
271 int needed = 85; /* average message length */
277 int mask = msgdata->msg_mask;
278 char *file = (char *)msgdata->msg_file;
279 cfs_debug_limit_state_t *cdls = msgdata->msg_cdls;
281 if (strchr(file, '/'))
282 file = strrchr(file, '/') + 1;
284 tcd = cfs_trace_get_tcd();
286 /* cfs_trace_get_tcd() grabs a lock, which disables preemption and
287 * pins us to a particular CPU. This avoids an smp_processor_id()
288 * warning on Linux when debugging is enabled. */
289 cfs_set_ptldebug_header(&header, msgdata, CDEBUG_STACK());
291 if (tcd == NULL) /* arch may not log in IRQ context */
294 if (tcd->tcd_cur_pages == 0)
295 header.ph_flags |= PH_FLAG_FIRST_RECORD;
297 if (tcd->tcd_shutting_down) {
298 cfs_trace_put_tcd(tcd);
303 depth = __current_nesting_level();
304 known_size = strlen(file) + 1 + depth;
306 known_size += strlen(msgdata->msg_fn) + 1;
308 if (libcfs_debug_binary)
309 known_size += sizeof(header);
312 * '2' used because vsnprintf return real size required for output
313 * _without_ terminating NULL.
314 * if needed is to small for this format.
316 for (i = 0; i < 2; i++) {
317 tage = cfs_trace_get_tage(tcd, needed + known_size + 1);
319 if (needed + known_size > CFS_PAGE_SIZE)
322 cfs_trace_put_tcd(tcd);
327 string_buf = (char *)cfs_page_address(tage->page) +
328 tage->used + known_size;
330 max_nob = CFS_PAGE_SIZE - tage->used - known_size;
332 printk(CFS_KERN_EMERG "negative max_nob: %d\n",
335 cfs_trace_put_tcd(tcd);
343 needed = vsnprintf(string_buf, max_nob, format1, ap);
348 remain = max_nob - needed;
352 va_start(ap, format2);
353 needed += vsnprintf(string_buf + needed, remain,
358 if (needed < max_nob) /* well. printing ok.. */
362 if (*(string_buf+needed-1) != '\n')
363 printk(CFS_KERN_INFO "format at %s:%d:%s doesn't end in "
364 "newline\n", file, msgdata->msg_line, msgdata->msg_fn);
366 header.ph_len = known_size + needed;
367 debug_buf = (char *)cfs_page_address(tage->page) + tage->used;
369 if (libcfs_debug_binary) {
370 memcpy(debug_buf, &header, sizeof(header));
371 tage->used += sizeof(header);
372 debug_buf += sizeof(header);
375 /* indent message according to the nesting level */
376 while (depth-- > 0) {
377 *(debug_buf++) = '.';
381 strcpy(debug_buf, file);
382 tage->used += strlen(file) + 1;
383 debug_buf += strlen(file) + 1;
385 if (msgdata->msg_fn) {
386 strcpy(debug_buf, msgdata->msg_fn);
387 tage->used += strlen(msgdata->msg_fn) + 1;
388 debug_buf += strlen(msgdata->msg_fn) + 1;
391 __LASSERT(debug_buf == string_buf);
393 tage->used += needed;
394 __LASSERT (tage->used <= CFS_PAGE_SIZE);
397 if ((mask & libcfs_printk) == 0) {
398 /* no console output requested */
400 cfs_trace_put_tcd(tcd);
405 if (libcfs_console_ratelimit &&
406 cdls->cdls_next != 0 && /* not first time ever */
407 !cfs_time_after(cfs_time_current(), cdls->cdls_next)) {
408 /* skipping a console message */
411 cfs_trace_put_tcd(tcd);
415 if (cfs_time_after(cfs_time_current(), cdls->cdls_next +
416 libcfs_console_max_delay
417 + cfs_time_seconds(10))) {
418 /* last timeout was a long time ago */
419 cdls->cdls_delay /= libcfs_console_backoff * 4;
421 cdls->cdls_delay *= libcfs_console_backoff;
423 if (cdls->cdls_delay < libcfs_console_min_delay)
424 cdls->cdls_delay = libcfs_console_min_delay;
425 else if (cdls->cdls_delay > libcfs_console_max_delay)
426 cdls->cdls_delay = libcfs_console_max_delay;
429 /* ensure cdls_next is never zero after it's been seen */
430 cdls->cdls_next = (cfs_time_current() + cdls->cdls_delay) | 1;
434 cfs_print_to_console(&header, mask, string_buf, needed, file,
436 cfs_trace_put_tcd(tcd);
438 string_buf = cfs_trace_get_console_buffer();
441 if (format1 != NULL) {
443 needed = vsnprintf(string_buf,
444 CFS_TRACE_CONSOLE_BUFFER_SIZE,
448 if (format2 != NULL) {
449 remain = CFS_TRACE_CONSOLE_BUFFER_SIZE - needed;
451 va_start(ap, format2);
452 needed += vsnprintf(string_buf+needed, remain,
457 cfs_print_to_console(&header, mask,
458 string_buf, needed, file, msgdata->msg_fn);
460 cfs_trace_put_console_buffer(string_buf);
463 if (cdls != NULL && cdls->cdls_count != 0) {
464 string_buf = cfs_trace_get_console_buffer();
466 needed = snprintf(string_buf, CFS_TRACE_CONSOLE_BUFFER_SIZE,
467 "Skipped %d previous similar message%s\n",
469 (cdls->cdls_count > 1) ? "s" : "");
471 cfs_print_to_console(&header, mask,
472 string_buf, needed, file, msgdata->msg_fn);
474 cfs_trace_put_console_buffer(string_buf);
475 cdls->cdls_count = 0;
480 EXPORT_SYMBOL(libcfs_debug_vmsg2);
483 libcfs_assertion_failed(const char *expr, struct libcfs_debug_msg_data *msgdata)
485 libcfs_debug_msg(msgdata, "ASSERTION(%s) failed\n", expr);
486 /* cfs_enter_debugger(); */
487 lbug_with_loc(msgdata);
489 EXPORT_SYMBOL(libcfs_assertion_failed);
492 cfs_trace_assertion_failed(const char *str,
493 struct libcfs_debug_msg_data *msgdata)
495 struct ptldebug_header hdr;
497 libcfs_panic_in_progress = 1;
498 libcfs_catastrophe = 1;
501 cfs_set_ptldebug_header(&hdr, msgdata, CDEBUG_STACK());
503 cfs_print_to_console(&hdr, D_EMERG, str, strlen(str),
504 msgdata->msg_file, msgdata->msg_fn);
506 LIBCFS_PANIC("Lustre debug assertion failure\n");
512 panic_collect_pages(struct page_collection *pc)
514 /* Do the collect_pages job on a single CPU: assumes that all other
515 * CPUs have been stopped during a panic. If this isn't true for some
516 * arch, this will have to be implemented separately in each arch. */
519 struct cfs_trace_cpu_data *tcd;
521 CFS_INIT_LIST_HEAD(&pc->pc_pages);
523 cfs_tcd_for_each(tcd, i, j) {
524 cfs_list_splice_init(&tcd->tcd_pages, &pc->pc_pages);
525 tcd->tcd_cur_pages = 0;
527 if (pc->pc_want_daemon_pages) {
528 cfs_list_splice_init(&tcd->tcd_daemon_pages,
530 tcd->tcd_cur_daemon_pages = 0;
535 static void collect_pages_on_all_cpus(struct page_collection *pc)
537 struct cfs_trace_cpu_data *tcd;
540 cfs_spin_lock(&pc->pc_lock);
541 cfs_for_each_possible_cpu(cpu) {
542 cfs_tcd_for_each_type_lock(tcd, i, cpu) {
543 cfs_list_splice_init(&tcd->tcd_pages, &pc->pc_pages);
544 tcd->tcd_cur_pages = 0;
545 if (pc->pc_want_daemon_pages) {
546 cfs_list_splice_init(&tcd->tcd_daemon_pages,
548 tcd->tcd_cur_daemon_pages = 0;
552 cfs_spin_unlock(&pc->pc_lock);
555 static void collect_pages(struct page_collection *pc)
557 CFS_INIT_LIST_HEAD(&pc->pc_pages);
559 if (libcfs_panic_in_progress)
560 panic_collect_pages(pc);
562 collect_pages_on_all_cpus(pc);
565 static void put_pages_back_on_all_cpus(struct page_collection *pc)
567 struct cfs_trace_cpu_data *tcd;
568 cfs_list_t *cur_head;
569 struct cfs_trace_page *tage;
570 struct cfs_trace_page *tmp;
573 cfs_spin_lock(&pc->pc_lock);
574 cfs_for_each_possible_cpu(cpu) {
575 cfs_tcd_for_each_type_lock(tcd, i, cpu) {
576 cur_head = tcd->tcd_pages.next;
578 cfs_list_for_each_entry_safe_typed(tage, tmp,
580 struct cfs_trace_page,
583 __LASSERT_TAGE_INVARIANT(tage);
585 if (tage->cpu != cpu || tage->type != i)
588 cfs_tage_to_tail(tage, cur_head);
589 tcd->tcd_cur_pages++;
593 cfs_spin_unlock(&pc->pc_lock);
596 static void put_pages_back(struct page_collection *pc)
598 if (!libcfs_panic_in_progress)
599 put_pages_back_on_all_cpus(pc);
602 /* Add pages to a per-cpu debug daemon ringbuffer. This buffer makes sure that
603 * we have a good amount of data at all times for dumping during an LBUG, even
604 * if we have been steadily writing (and otherwise discarding) pages via the
606 static void put_pages_on_tcd_daemon_list(struct page_collection *pc,
607 struct cfs_trace_cpu_data *tcd)
609 struct cfs_trace_page *tage;
610 struct cfs_trace_page *tmp;
612 cfs_spin_lock(&pc->pc_lock);
613 cfs_list_for_each_entry_safe_typed(tage, tmp, &pc->pc_pages,
614 struct cfs_trace_page, linkage) {
616 __LASSERT_TAGE_INVARIANT(tage);
618 if (tage->cpu != tcd->tcd_cpu || tage->type != tcd->tcd_type)
621 cfs_tage_to_tail(tage, &tcd->tcd_daemon_pages);
622 tcd->tcd_cur_daemon_pages++;
624 if (tcd->tcd_cur_daemon_pages > tcd->tcd_max_pages) {
625 struct cfs_trace_page *victim;
627 __LASSERT(!cfs_list_empty(&tcd->tcd_daemon_pages));
628 victim = cfs_tage_from_list(tcd->tcd_daemon_pages.next);
630 __LASSERT_TAGE_INVARIANT(victim);
632 cfs_list_del(&victim->linkage);
633 cfs_tage_free(victim);
634 tcd->tcd_cur_daemon_pages--;
637 cfs_spin_unlock(&pc->pc_lock);
640 static void put_pages_on_daemon_list(struct page_collection *pc)
642 struct cfs_trace_cpu_data *tcd;
645 cfs_for_each_possible_cpu(cpu) {
646 cfs_tcd_for_each_type_lock(tcd, i, cpu)
647 put_pages_on_tcd_daemon_list(pc, tcd);
651 void cfs_trace_debug_print(void)
653 struct page_collection pc;
654 struct cfs_trace_page *tage;
655 struct cfs_trace_page *tmp;
657 cfs_spin_lock_init(&pc.pc_lock);
659 pc.pc_want_daemon_pages = 1;
661 cfs_list_for_each_entry_safe_typed(tage, tmp, &pc.pc_pages,
662 struct cfs_trace_page, linkage) {
666 __LASSERT_TAGE_INVARIANT(tage);
669 p = cfs_page_address(page);
670 while (p < ((char *)cfs_page_address(page) + tage->used)) {
671 struct ptldebug_header *hdr;
676 p += strlen(file) + 1;
679 len = hdr->ph_len - (int)(p - (char *)hdr);
681 cfs_print_to_console(hdr, D_EMERG, p, len, file, fn);
686 cfs_list_del(&tage->linkage);
691 int cfs_tracefile_dump_all_pages(char *filename)
693 struct page_collection pc;
695 struct cfs_trace_page *tage;
696 struct cfs_trace_page *tmp;
701 cfs_tracefile_write_lock();
703 filp = cfs_filp_open(filename,
704 O_CREAT|O_EXCL|O_WRONLY|O_LARGEFILE, 0600, &rc);
708 "LustreError: can't open %s for dump: rc %d\n",
713 cfs_spin_lock_init(&pc.pc_lock);
714 pc.pc_want_daemon_pages = 1;
716 if (cfs_list_empty(&pc.pc_pages)) {
721 /* ok, for now, just write the pages. in the future we'll be building
722 * iobufs with the pages and calling generic_direct_IO */
724 cfs_list_for_each_entry_safe_typed(tage, tmp, &pc.pc_pages,
725 struct cfs_trace_page, linkage) {
727 __LASSERT_TAGE_INVARIANT(tage);
729 rc = cfs_filp_write(filp, cfs_page_address(tage->page),
730 tage->used, cfs_filp_poff(filp));
731 if (rc != (int)tage->used) {
732 printk(CFS_KERN_WARNING "wanted to write %u but wrote "
733 "%d\n", tage->used, rc);
735 __LASSERT(cfs_list_empty(&pc.pc_pages));
738 cfs_list_del(&tage->linkage);
742 rc = cfs_filp_fsync(filp);
744 printk(CFS_KERN_ERR "sync returns %d\n", rc);
746 cfs_filp_close(filp);
748 cfs_tracefile_write_unlock();
752 void cfs_trace_flush_pages(void)
754 struct page_collection pc;
755 struct cfs_trace_page *tage;
756 struct cfs_trace_page *tmp;
758 cfs_spin_lock_init(&pc.pc_lock);
760 pc.pc_want_daemon_pages = 1;
762 cfs_list_for_each_entry_safe_typed(tage, tmp, &pc.pc_pages,
763 struct cfs_trace_page, linkage) {
765 __LASSERT_TAGE_INVARIANT(tage);
767 cfs_list_del(&tage->linkage);
772 int cfs_trace_copyin_string(char *knl_buffer, int knl_buffer_nob,
773 const char *usr_buffer, int usr_buffer_nob)
777 if (usr_buffer_nob > knl_buffer_nob)
780 if (cfs_copy_from_user((void *)knl_buffer,
781 (void *)usr_buffer, usr_buffer_nob))
784 nob = strnlen(knl_buffer, usr_buffer_nob);
785 while (nob-- >= 0) /* strip trailing whitespace */
786 if (!isspace(knl_buffer[nob]))
789 if (nob < 0) /* empty string */
792 if (nob == knl_buffer_nob) /* no space to terminate */
795 knl_buffer[nob + 1] = 0; /* terminate */
799 int cfs_trace_copyout_string(char *usr_buffer, int usr_buffer_nob,
800 const char *knl_buffer, char *append)
802 /* NB if 'append' != NULL, it's a single character to append to the
803 * copied out string - usually "\n", for /proc entries and "" (i.e. a
804 * terminating zero byte) for sysctl entries */
805 int nob = strlen(knl_buffer);
807 if (nob > usr_buffer_nob)
808 nob = usr_buffer_nob;
810 if (cfs_copy_to_user(usr_buffer, knl_buffer, nob))
813 if (append != NULL && nob < usr_buffer_nob) {
814 if (cfs_copy_to_user(usr_buffer + nob, append, 1))
822 EXPORT_SYMBOL(cfs_trace_copyout_string);
824 int cfs_trace_allocate_string_buffer(char **str, int nob)
826 if (nob > 2 * CFS_PAGE_SIZE) /* string must be "sensible" */
829 *str = cfs_alloc(nob, CFS_ALLOC_STD | CFS_ALLOC_ZERO);
836 void cfs_trace_free_string_buffer(char *str, int nob)
841 int cfs_trace_dump_debug_buffer_usrstr(void *usr_str, int usr_str_nob)
846 rc = cfs_trace_allocate_string_buffer(&str, usr_str_nob + 1);
850 rc = cfs_trace_copyin_string(str, usr_str_nob + 1,
851 usr_str, usr_str_nob);
855 #if !defined(__WINNT__)
861 rc = cfs_tracefile_dump_all_pages(str);
863 cfs_trace_free_string_buffer(str, usr_str_nob + 1);
867 int cfs_trace_daemon_command(char *str)
871 cfs_tracefile_write_lock();
873 if (strcmp(str, "stop") == 0) {
874 cfs_tracefile_write_unlock();
875 cfs_trace_stop_thread();
876 cfs_tracefile_write_lock();
877 memset(cfs_tracefile, 0, sizeof(cfs_tracefile));
879 } else if (strncmp(str, "size=", 5) == 0) {
880 cfs_tracefile_size = simple_strtoul(str + 5, NULL, 0);
881 if (cfs_tracefile_size < 10 || cfs_tracefile_size > 20480)
882 cfs_tracefile_size = CFS_TRACEFILE_SIZE;
884 cfs_tracefile_size <<= 20;
886 } else if (strlen(str) >= sizeof(cfs_tracefile)) {
889 } else if (str[0] != '/') {
893 strcpy(cfs_tracefile, str);
896 "Lustre: debug daemon will attempt to start writing "
897 "to %s (%lukB max)\n", cfs_tracefile,
898 (long)(cfs_tracefile_size >> 10));
900 cfs_trace_start_thread();
903 cfs_tracefile_write_unlock();
907 int cfs_trace_daemon_command_usrstr(void *usr_str, int usr_str_nob)
912 rc = cfs_trace_allocate_string_buffer(&str, usr_str_nob + 1);
916 rc = cfs_trace_copyin_string(str, usr_str_nob + 1,
917 usr_str, usr_str_nob);
919 rc = cfs_trace_daemon_command(str);
921 cfs_trace_free_string_buffer(str, usr_str_nob + 1);
925 int cfs_trace_set_debug_mb(int mb)
930 int limit = cfs_trace_max_debug_mb();
931 struct cfs_trace_cpu_data *tcd;
933 if (mb < cfs_num_possible_cpus()) {
934 printk(KERN_ERR "Cannot set debug_mb to %d, the value should be >= %d\n",
935 mb, cfs_num_possible_cpus());
940 printk(CFS_KERN_ERR "Lustre: Refusing to set debug buffer size "
941 "to %dMB - limit is %d\n", mb, limit);
945 mb /= cfs_num_possible_cpus();
946 pages = mb << (20 - CFS_PAGE_SHIFT);
948 cfs_tracefile_write_lock();
950 cfs_tcd_for_each(tcd, i, j)
951 tcd->tcd_max_pages = (pages * tcd->tcd_pages_factor) / 100;
953 cfs_tracefile_write_unlock();
958 int cfs_trace_set_debug_mb_usrstr(void *usr_str, int usr_str_nob)
963 rc = cfs_trace_copyin_string(str, sizeof(str), usr_str, usr_str_nob);
967 return cfs_trace_set_debug_mb(simple_strtoul(str, NULL, 0));
970 int cfs_trace_get_debug_mb(void)
974 struct cfs_trace_cpu_data *tcd;
977 cfs_tracefile_read_lock();
979 cfs_tcd_for_each(tcd, i, j)
980 total_pages += tcd->tcd_max_pages;
982 cfs_tracefile_read_unlock();
984 return (total_pages >> (20 - CFS_PAGE_SHIFT)) + 1;
987 static int tracefiled(void *arg)
989 struct page_collection pc;
990 struct tracefiled_ctl *tctl = arg;
991 struct cfs_trace_page *tage;
992 struct cfs_trace_page *tmp;
999 /* we're started late enough that we pick up init's fs context */
1000 /* this is so broken in uml? what on earth is going on? */
1001 cfs_daemonize("ktracefiled");
1003 cfs_spin_lock_init(&pc.pc_lock);
1004 cfs_complete(&tctl->tctl_start);
1007 cfs_waitlink_t __wait;
1009 pc.pc_want_daemon_pages = 0;
1011 if (cfs_list_empty(&pc.pc_pages))
1015 cfs_tracefile_read_lock();
1016 if (cfs_tracefile[0] != 0) {
1017 filp = cfs_filp_open(cfs_tracefile,
1018 O_CREAT | O_RDWR | O_LARGEFILE,
1021 printk(CFS_KERN_WARNING "couldn't open %s: "
1022 "%d\n", cfs_tracefile, rc);
1024 cfs_tracefile_read_unlock();
1026 put_pages_on_daemon_list(&pc);
1027 __LASSERT(cfs_list_empty(&pc.pc_pages));
1033 cfs_list_for_each_entry_safe_typed(tage, tmp, &pc.pc_pages,
1034 struct cfs_trace_page,
1036 static loff_t f_pos;
1038 __LASSERT_TAGE_INVARIANT(tage);
1040 if (f_pos >= (off_t)cfs_tracefile_size)
1042 else if (f_pos > (off_t)cfs_filp_size(filp))
1043 f_pos = cfs_filp_size(filp);
1045 rc = cfs_filp_write(filp, cfs_page_address(tage->page),
1046 tage->used, &f_pos);
1047 if (rc != (int)tage->used) {
1048 printk(CFS_KERN_WARNING "wanted to write %u "
1049 "but wrote %d\n", tage->used, rc);
1050 put_pages_back(&pc);
1051 __LASSERT(cfs_list_empty(&pc.pc_pages));
1056 cfs_filp_close(filp);
1057 put_pages_on_daemon_list(&pc);
1058 if (!cfs_list_empty(&pc.pc_pages)) {
1061 printk(CFS_KERN_ALERT "Lustre: trace pages aren't "
1063 printk(CFS_KERN_ERR "total cpus(%d): ",
1064 cfs_num_possible_cpus());
1065 for (i = 0; i < cfs_num_possible_cpus(); i++)
1066 if (cfs_cpu_online(i))
1067 printk(CFS_KERN_ERR "%d(on) ", i);
1069 printk(CFS_KERN_ERR "%d(off) ", i);
1070 printk(CFS_KERN_ERR "\n");
1073 cfs_list_for_each_entry_safe(tage, tmp, &pc.pc_pages,
1075 printk(CFS_KERN_ERR "page %d belongs to cpu "
1076 "%d\n", ++i, tage->cpu);
1077 printk(CFS_KERN_ERR "There are %d pages unwritten\n",
1080 __LASSERT(cfs_list_empty(&pc.pc_pages));
1082 if (cfs_atomic_read(&tctl->tctl_shutdown)) {
1083 if (last_loop == 0) {
1090 cfs_waitlink_init(&__wait);
1091 cfs_waitq_add(&tctl->tctl_waitq, &__wait);
1092 cfs_set_current_state(CFS_TASK_INTERRUPTIBLE);
1093 cfs_waitq_timedwait(&__wait, CFS_TASK_INTERRUPTIBLE,
1094 cfs_time_seconds(1));
1095 cfs_waitq_del(&tctl->tctl_waitq, &__wait);
1097 cfs_complete(&tctl->tctl_stop);
1101 int cfs_trace_start_thread(void)
1103 struct tracefiled_ctl *tctl = &trace_tctl;
1106 cfs_mutex_down(&cfs_trace_thread_sem);
1110 cfs_init_completion(&tctl->tctl_start);
1111 cfs_init_completion(&tctl->tctl_stop);
1112 cfs_waitq_init(&tctl->tctl_waitq);
1113 cfs_atomic_set(&tctl->tctl_shutdown, 0);
1115 if (cfs_create_thread(tracefiled, tctl, 0) < 0) {
1120 cfs_wait_for_completion(&tctl->tctl_start);
1123 cfs_mutex_up(&cfs_trace_thread_sem);
1127 void cfs_trace_stop_thread(void)
1129 struct tracefiled_ctl *tctl = &trace_tctl;
1131 cfs_mutex_down(&cfs_trace_thread_sem);
1132 if (thread_running) {
1133 printk(CFS_KERN_INFO
1134 "Lustre: shutting down debug daemon thread...\n");
1135 cfs_atomic_set(&tctl->tctl_shutdown, 1);
1136 cfs_wait_for_completion(&tctl->tctl_stop);
1139 cfs_mutex_up(&cfs_trace_thread_sem);
1142 int cfs_tracefile_init(int max_pages)
1144 struct cfs_trace_cpu_data *tcd;
1150 rc = cfs_tracefile_init_arch();
1154 cfs_tcd_for_each(tcd, i, j) {
1155 /* tcd_pages_factor is initialized int tracefile_init_arch. */
1156 factor = tcd->tcd_pages_factor;
1157 CFS_INIT_LIST_HEAD(&tcd->tcd_pages);
1158 CFS_INIT_LIST_HEAD(&tcd->tcd_stock_pages);
1159 CFS_INIT_LIST_HEAD(&tcd->tcd_daemon_pages);
1160 tcd->tcd_cur_pages = 0;
1161 tcd->tcd_cur_stock_pages = 0;
1162 tcd->tcd_cur_daemon_pages = 0;
1163 tcd->tcd_max_pages = (max_pages * factor) / 100;
1164 LASSERT(tcd->tcd_max_pages > 0);
1165 tcd->tcd_shutting_down = 0;
1171 static void trace_cleanup_on_all_cpus(void)
1173 struct cfs_trace_cpu_data *tcd;
1174 struct cfs_trace_page *tage;
1175 struct cfs_trace_page *tmp;
1178 cfs_for_each_possible_cpu(cpu) {
1179 cfs_tcd_for_each_type_lock(tcd, i, cpu) {
1180 tcd->tcd_shutting_down = 1;
1182 cfs_list_for_each_entry_safe_typed(tage, tmp,
1184 struct cfs_trace_page,
1186 __LASSERT_TAGE_INVARIANT(tage);
1188 cfs_list_del(&tage->linkage);
1189 cfs_tage_free(tage);
1192 tcd->tcd_cur_pages = 0;
1197 static void cfs_trace_cleanup(void)
1199 struct page_collection pc;
1201 CFS_INIT_LIST_HEAD(&pc.pc_pages);
1202 cfs_spin_lock_init(&pc.pc_lock);
1204 trace_cleanup_on_all_cpus();
1206 cfs_tracefile_fini_arch();
1209 void cfs_tracefile_exit(void)
1211 cfs_trace_stop_thread();
1212 cfs_trace_cleanup();