4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
27 * Copyright (c) 2008, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2012, 2013, Intel Corporation.
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 struct mutex cfs_trace_thread_mutex;
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() && memory_pressure_get())
79 * Don't spam console with allocation failures: they will be reported
80 * by upper layer anyway.
83 page = alloc_page(gfp);
87 tage = kmalloc(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 __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 <= PAGE_CACHE_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(GFP_ATOMIC);
158 if (unlikely(tage == NULL)) {
159 if ((!memory_pressure_get() ||
160 cfs_in_interrupt()) && printk_ratelimit())
161 printk(CFS_KERN_WARNING
162 "cannot allocate a tage (%ld)\n",
169 tage->cpu = smp_processor_id();
170 tage->type = tcd->tcd_type;
171 cfs_list_add_tail(&tage->linkage, &tcd->tcd_pages);
172 tcd->tcd_cur_pages++;
174 if (tcd->tcd_cur_pages > 8 && thread_running) {
175 struct tracefiled_ctl *tctl = &trace_tctl;
177 * wake up tracefiled to process some pages.
179 cfs_waitq_signal(&tctl->tctl_waitq);
186 static void cfs_tcd_shrink(struct cfs_trace_cpu_data *tcd)
188 int pgcount = tcd->tcd_cur_pages / 10;
189 struct page_collection pc;
190 struct cfs_trace_page *tage;
191 struct cfs_trace_page *tmp;
194 * XXX nikita: do NOT call portals_debug_msg() (CDEBUG/ENTRY/EXIT)
195 * from here: this will lead to infinite recursion.
198 if (printk_ratelimit())
199 printk(CFS_KERN_WARNING "debug daemon buffer overflowed; "
200 "discarding 10%% of pages (%d of %ld)\n",
201 pgcount + 1, tcd->tcd_cur_pages);
203 CFS_INIT_LIST_HEAD(&pc.pc_pages);
204 spin_lock_init(&pc.pc_lock);
206 cfs_list_for_each_entry_safe_typed(tage, tmp, &tcd->tcd_pages,
207 struct cfs_trace_page, linkage) {
211 cfs_list_move_tail(&tage->linkage, &pc.pc_pages);
212 tcd->tcd_cur_pages--;
214 put_pages_on_tcd_daemon_list(&pc, tcd);
217 /* return a page that has 'len' bytes left at the end */
218 static struct cfs_trace_page *cfs_trace_get_tage(struct cfs_trace_cpu_data *tcd,
221 struct cfs_trace_page *tage;
224 * XXX nikita: do NOT call portals_debug_msg() (CDEBUG/ENTRY/EXIT)
225 * from here: this will lead to infinite recursion.
228 if (len > PAGE_CACHE_SIZE) {
230 "cowardly refusing to write %lu bytes in a page\n", len);
234 tage = cfs_trace_get_tage_try(tcd, len);
239 if (tcd->tcd_cur_pages > 0) {
240 tage = cfs_tage_from_list(tcd->tcd_pages.next);
242 cfs_tage_to_tail(tage, &tcd->tcd_pages);
247 int libcfs_debug_msg(struct libcfs_debug_msg_data *msgdata,
248 const char *format, ...)
253 va_start(args, format);
254 rc = libcfs_debug_vmsg2(msgdata, format, args, NULL);
259 EXPORT_SYMBOL(libcfs_debug_msg);
261 int libcfs_debug_vmsg2(struct libcfs_debug_msg_data *msgdata,
262 const char *format1, va_list args,
263 const char *format2, ...)
265 struct cfs_trace_cpu_data *tcd = NULL;
266 struct ptldebug_header header = {0};
267 struct cfs_trace_page *tage;
268 /* string_buf is used only if tcd != NULL, and is always set then */
269 char *string_buf = NULL;
272 int needed = 85; /* average message length */
278 int mask = msgdata->msg_mask;
279 char *file = (char *)msgdata->msg_file;
280 cfs_debug_limit_state_t *cdls = msgdata->msg_cdls;
282 if (strchr(file, '/'))
283 file = strrchr(file, '/') + 1;
285 tcd = cfs_trace_get_tcd();
287 /* cfs_trace_get_tcd() grabs a lock, which disables preemption and
288 * pins us to a particular CPU. This avoids an smp_processor_id()
289 * warning on Linux when debugging is enabled. */
290 cfs_set_ptldebug_header(&header, msgdata, CDEBUG_STACK());
292 if (tcd == NULL) /* arch may not log in IRQ context */
295 if (tcd->tcd_cur_pages == 0)
296 header.ph_flags |= PH_FLAG_FIRST_RECORD;
298 if (tcd->tcd_shutting_down) {
299 cfs_trace_put_tcd(tcd);
304 depth = __current_nesting_level();
305 known_size = strlen(file) + 1 + depth;
307 known_size += strlen(msgdata->msg_fn) + 1;
309 if (libcfs_debug_binary)
310 known_size += sizeof(header);
313 * '2' used because vsnprintf return real size required for output
314 * _without_ terminating NULL.
315 * if needed is to small for this format.
317 for (i = 0; i < 2; i++) {
318 tage = cfs_trace_get_tage(tcd, needed + known_size + 1);
320 if (needed + known_size > PAGE_CACHE_SIZE)
323 cfs_trace_put_tcd(tcd);
328 string_buf = (char *)page_address(tage->page) +
329 tage->used + known_size;
331 max_nob = PAGE_CACHE_SIZE - tage->used - known_size;
333 printk(CFS_KERN_EMERG "negative max_nob: %d\n",
336 cfs_trace_put_tcd(tcd);
344 needed = vsnprintf(string_buf, max_nob, format1, ap);
349 remain = max_nob - needed;
353 va_start(ap, format2);
354 needed += vsnprintf(string_buf + needed, remain,
359 if (needed < max_nob) /* well. printing ok.. */
363 if (*(string_buf+needed-1) != '\n')
364 printk(CFS_KERN_INFO "format at %s:%d:%s doesn't end in "
365 "newline\n", file, msgdata->msg_line, msgdata->msg_fn);
367 header.ph_len = known_size + needed;
368 debug_buf = (char *)page_address(tage->page) + tage->used;
370 if (libcfs_debug_binary) {
371 memcpy(debug_buf, &header, sizeof(header));
372 tage->used += sizeof(header);
373 debug_buf += sizeof(header);
376 /* indent message according to the nesting level */
377 while (depth-- > 0) {
378 *(debug_buf++) = '.';
382 strcpy(debug_buf, file);
383 tage->used += strlen(file) + 1;
384 debug_buf += strlen(file) + 1;
386 if (msgdata->msg_fn) {
387 strcpy(debug_buf, msgdata->msg_fn);
388 tage->used += strlen(msgdata->msg_fn) + 1;
389 debug_buf += strlen(msgdata->msg_fn) + 1;
392 __LASSERT(debug_buf == string_buf);
394 tage->used += needed;
395 __LASSERT(tage->used <= PAGE_CACHE_SIZE);
398 if ((mask & libcfs_printk) == 0) {
399 /* no console output requested */
401 cfs_trace_put_tcd(tcd);
406 if (libcfs_console_ratelimit &&
407 cdls->cdls_next != 0 && /* not first time ever */
408 !cfs_time_after(cfs_time_current(), cdls->cdls_next)) {
409 /* skipping a console message */
412 cfs_trace_put_tcd(tcd);
416 if (cfs_time_after(cfs_time_current(), cdls->cdls_next +
417 libcfs_console_max_delay
418 + cfs_time_seconds(10))) {
419 /* last timeout was a long time ago */
420 cdls->cdls_delay /= libcfs_console_backoff * 4;
422 cdls->cdls_delay *= libcfs_console_backoff;
424 if (cdls->cdls_delay < libcfs_console_min_delay)
425 cdls->cdls_delay = libcfs_console_min_delay;
426 else if (cdls->cdls_delay > libcfs_console_max_delay)
427 cdls->cdls_delay = libcfs_console_max_delay;
430 /* ensure cdls_next is never zero after it's been seen */
431 cdls->cdls_next = (cfs_time_current() + cdls->cdls_delay) | 1;
435 cfs_print_to_console(&header, mask, string_buf, needed, file,
437 cfs_trace_put_tcd(tcd);
439 string_buf = cfs_trace_get_console_buffer();
442 if (format1 != NULL) {
444 needed = vsnprintf(string_buf,
445 CFS_TRACE_CONSOLE_BUFFER_SIZE,
449 if (format2 != NULL) {
450 remain = CFS_TRACE_CONSOLE_BUFFER_SIZE - needed;
452 va_start(ap, format2);
453 needed += vsnprintf(string_buf+needed, remain,
458 cfs_print_to_console(&header, mask,
459 string_buf, needed, file, msgdata->msg_fn);
461 cfs_trace_put_console_buffer(string_buf);
464 if (cdls != NULL && cdls->cdls_count != 0) {
465 string_buf = cfs_trace_get_console_buffer();
467 needed = snprintf(string_buf, CFS_TRACE_CONSOLE_BUFFER_SIZE,
468 "Skipped %d previous similar message%s\n",
470 (cdls->cdls_count > 1) ? "s" : "");
472 cfs_print_to_console(&header, mask,
473 string_buf, needed, file, msgdata->msg_fn);
475 cfs_trace_put_console_buffer(string_buf);
476 cdls->cdls_count = 0;
481 EXPORT_SYMBOL(libcfs_debug_vmsg2);
484 cfs_trace_assertion_failed(const char *str,
485 struct libcfs_debug_msg_data *msgdata)
487 struct ptldebug_header hdr;
489 libcfs_panic_in_progress = 1;
490 libcfs_catastrophe = 1;
493 cfs_set_ptldebug_header(&hdr, msgdata, CDEBUG_STACK());
495 cfs_print_to_console(&hdr, D_EMERG, str, strlen(str),
496 msgdata->msg_file, msgdata->msg_fn);
498 panic("Lustre debug assertion failure\n");
504 panic_collect_pages(struct page_collection *pc)
506 /* Do the collect_pages job on a single CPU: assumes that all other
507 * CPUs have been stopped during a panic. If this isn't true for some
508 * arch, this will have to be implemented separately in each arch. */
511 struct cfs_trace_cpu_data *tcd;
513 CFS_INIT_LIST_HEAD(&pc->pc_pages);
515 cfs_tcd_for_each(tcd, i, j) {
516 cfs_list_splice_init(&tcd->tcd_pages, &pc->pc_pages);
517 tcd->tcd_cur_pages = 0;
519 if (pc->pc_want_daemon_pages) {
520 cfs_list_splice_init(&tcd->tcd_daemon_pages,
522 tcd->tcd_cur_daemon_pages = 0;
527 static void collect_pages_on_all_cpus(struct page_collection *pc)
529 struct cfs_trace_cpu_data *tcd;
532 spin_lock(&pc->pc_lock);
533 cfs_for_each_possible_cpu(cpu) {
534 cfs_tcd_for_each_type_lock(tcd, i, cpu) {
535 cfs_list_splice_init(&tcd->tcd_pages, &pc->pc_pages);
536 tcd->tcd_cur_pages = 0;
537 if (pc->pc_want_daemon_pages) {
538 cfs_list_splice_init(&tcd->tcd_daemon_pages,
540 tcd->tcd_cur_daemon_pages = 0;
544 spin_unlock(&pc->pc_lock);
547 static void collect_pages(struct page_collection *pc)
549 CFS_INIT_LIST_HEAD(&pc->pc_pages);
551 if (libcfs_panic_in_progress)
552 panic_collect_pages(pc);
554 collect_pages_on_all_cpus(pc);
557 static void put_pages_back_on_all_cpus(struct page_collection *pc)
559 struct cfs_trace_cpu_data *tcd;
560 cfs_list_t *cur_head;
561 struct cfs_trace_page *tage;
562 struct cfs_trace_page *tmp;
565 spin_lock(&pc->pc_lock);
566 cfs_for_each_possible_cpu(cpu) {
567 cfs_tcd_for_each_type_lock(tcd, i, cpu) {
568 cur_head = tcd->tcd_pages.next;
570 cfs_list_for_each_entry_safe_typed(tage, tmp,
572 struct cfs_trace_page,
575 __LASSERT_TAGE_INVARIANT(tage);
577 if (tage->cpu != cpu || tage->type != i)
580 cfs_tage_to_tail(tage, cur_head);
581 tcd->tcd_cur_pages++;
585 spin_unlock(&pc->pc_lock);
588 static void put_pages_back(struct page_collection *pc)
590 if (!libcfs_panic_in_progress)
591 put_pages_back_on_all_cpus(pc);
594 /* Add pages to a per-cpu debug daemon ringbuffer. This buffer makes sure that
595 * we have a good amount of data at all times for dumping during an LBUG, even
596 * if we have been steadily writing (and otherwise discarding) pages via the
598 static void put_pages_on_tcd_daemon_list(struct page_collection *pc,
599 struct cfs_trace_cpu_data *tcd)
601 struct cfs_trace_page *tage;
602 struct cfs_trace_page *tmp;
604 spin_lock(&pc->pc_lock);
605 cfs_list_for_each_entry_safe_typed(tage, tmp, &pc->pc_pages,
606 struct cfs_trace_page, linkage) {
608 __LASSERT_TAGE_INVARIANT(tage);
610 if (tage->cpu != tcd->tcd_cpu || tage->type != tcd->tcd_type)
613 cfs_tage_to_tail(tage, &tcd->tcd_daemon_pages);
614 tcd->tcd_cur_daemon_pages++;
616 if (tcd->tcd_cur_daemon_pages > tcd->tcd_max_pages) {
617 struct cfs_trace_page *victim;
619 __LASSERT(!cfs_list_empty(&tcd->tcd_daemon_pages));
620 victim = cfs_tage_from_list(tcd->tcd_daemon_pages.next);
622 __LASSERT_TAGE_INVARIANT(victim);
624 cfs_list_del(&victim->linkage);
625 cfs_tage_free(victim);
626 tcd->tcd_cur_daemon_pages--;
629 spin_unlock(&pc->pc_lock);
632 static void put_pages_on_daemon_list(struct page_collection *pc)
634 struct cfs_trace_cpu_data *tcd;
637 cfs_for_each_possible_cpu(cpu) {
638 cfs_tcd_for_each_type_lock(tcd, i, cpu)
639 put_pages_on_tcd_daemon_list(pc, tcd);
643 void cfs_trace_debug_print(void)
645 struct page_collection pc;
646 struct cfs_trace_page *tage;
647 struct cfs_trace_page *tmp;
649 spin_lock_init(&pc.pc_lock);
651 pc.pc_want_daemon_pages = 1;
653 cfs_list_for_each_entry_safe_typed(tage, tmp, &pc.pc_pages,
654 struct cfs_trace_page, linkage) {
658 __LASSERT_TAGE_INVARIANT(tage);
661 p = page_address(page);
662 while (p < ((char *)page_address(page) + tage->used)) {
663 struct ptldebug_header *hdr;
668 p += strlen(file) + 1;
671 len = hdr->ph_len - (int)(p - (char *)hdr);
673 cfs_print_to_console(hdr, D_EMERG, p, len, file, fn);
678 cfs_list_del(&tage->linkage);
683 int cfs_tracefile_dump_all_pages(char *filename)
685 struct page_collection pc;
687 struct cfs_trace_page *tage;
688 struct cfs_trace_page *tmp;
693 cfs_tracefile_write_lock();
695 filp = filp_open(filename, O_CREAT|O_EXCL|O_WRONLY|O_LARGEFILE, 0600);
699 printk(KERN_ERR "LustreError: can't open %s for dump: rc %d\n",
704 spin_lock_init(&pc.pc_lock);
705 pc.pc_want_daemon_pages = 1;
707 if (cfs_list_empty(&pc.pc_pages)) {
712 /* ok, for now, just write the pages. in the future we'll be building
713 * iobufs with the pages and calling generic_direct_IO */
715 cfs_list_for_each_entry_safe_typed(tage, tmp, &pc.pc_pages,
716 struct cfs_trace_page, linkage) {
718 __LASSERT_TAGE_INVARIANT(tage);
720 rc = filp_write(filp, page_address(tage->page),
721 tage->used, filp_poff(filp));
722 if (rc != (int)tage->used) {
723 printk(CFS_KERN_WARNING "wanted to write %u but wrote "
724 "%d\n", tage->used, rc);
726 __LASSERT(cfs_list_empty(&pc.pc_pages));
729 cfs_list_del(&tage->linkage);
733 rc = filp_fsync(filp);
735 printk(CFS_KERN_ERR "sync returns %d\n", rc);
737 filp_close(filp, NULL);
739 cfs_tracefile_write_unlock();
743 void cfs_trace_flush_pages(void)
745 struct page_collection pc;
746 struct cfs_trace_page *tage;
747 struct cfs_trace_page *tmp;
749 spin_lock_init(&pc.pc_lock);
751 pc.pc_want_daemon_pages = 1;
753 cfs_list_for_each_entry_safe_typed(tage, tmp, &pc.pc_pages,
754 struct cfs_trace_page, linkage) {
756 __LASSERT_TAGE_INVARIANT(tage);
758 cfs_list_del(&tage->linkage);
763 int cfs_trace_copyin_string(char *knl_buffer, int knl_buffer_nob,
764 const char *usr_buffer, int usr_buffer_nob)
768 if (usr_buffer_nob > knl_buffer_nob)
771 if (copy_from_user((void *)knl_buffer,
772 (void *)usr_buffer, usr_buffer_nob))
775 nob = strnlen(knl_buffer, usr_buffer_nob);
776 while (nob-- >= 0) /* strip trailing whitespace */
777 if (!isspace(knl_buffer[nob]))
780 if (nob < 0) /* empty string */
783 if (nob == knl_buffer_nob) /* no space to terminate */
786 knl_buffer[nob + 1] = 0; /* terminate */
789 EXPORT_SYMBOL(cfs_trace_copyin_string);
791 int cfs_trace_copyout_string(char *usr_buffer, int usr_buffer_nob,
792 const char *knl_buffer, char *append)
794 /* NB if 'append' != NULL, it's a single character to append to the
795 * copied out string - usually "\n", for /proc entries and "" (i.e. a
796 * terminating zero byte) for sysctl entries */
797 int nob = strlen(knl_buffer);
799 if (nob > usr_buffer_nob)
800 nob = usr_buffer_nob;
802 if (copy_to_user(usr_buffer, knl_buffer, nob))
805 if (append != NULL && nob < usr_buffer_nob) {
806 if (copy_to_user(usr_buffer + nob, append, 1))
814 EXPORT_SYMBOL(cfs_trace_copyout_string);
816 int cfs_trace_allocate_string_buffer(char **str, int nob)
818 if (nob > 2 * PAGE_CACHE_SIZE) /* string must be "sensible" */
821 *str = kmalloc(nob, GFP_IOFS | __GFP_ZERO);
828 void cfs_trace_free_string_buffer(char *str, int nob)
833 int cfs_trace_dump_debug_buffer_usrstr(void *usr_str, int usr_str_nob)
838 rc = cfs_trace_allocate_string_buffer(&str, usr_str_nob + 1);
842 rc = cfs_trace_copyin_string(str, usr_str_nob + 1,
843 usr_str, usr_str_nob);
847 #if !defined(__WINNT__)
853 rc = cfs_tracefile_dump_all_pages(str);
855 cfs_trace_free_string_buffer(str, usr_str_nob + 1);
859 int cfs_trace_daemon_command(char *str)
863 cfs_tracefile_write_lock();
865 if (strcmp(str, "stop") == 0) {
866 cfs_tracefile_write_unlock();
867 cfs_trace_stop_thread();
868 cfs_tracefile_write_lock();
869 memset(cfs_tracefile, 0, sizeof(cfs_tracefile));
871 } else if (strncmp(str, "size=", 5) == 0) {
872 cfs_tracefile_size = simple_strtoul(str + 5, NULL, 0);
873 if (cfs_tracefile_size < 10 || cfs_tracefile_size > 20480)
874 cfs_tracefile_size = CFS_TRACEFILE_SIZE;
876 cfs_tracefile_size <<= 20;
878 } else if (strlen(str) >= sizeof(cfs_tracefile)) {
881 } else if (str[0] != '/') {
885 strcpy(cfs_tracefile, str);
888 "Lustre: debug daemon will attempt to start writing "
889 "to %s (%lukB max)\n", cfs_tracefile,
890 (long)(cfs_tracefile_size >> 10));
892 cfs_trace_start_thread();
895 cfs_tracefile_write_unlock();
899 int cfs_trace_daemon_command_usrstr(void *usr_str, int usr_str_nob)
904 rc = cfs_trace_allocate_string_buffer(&str, usr_str_nob + 1);
908 rc = cfs_trace_copyin_string(str, usr_str_nob + 1,
909 usr_str, usr_str_nob);
911 rc = cfs_trace_daemon_command(str);
913 cfs_trace_free_string_buffer(str, usr_str_nob + 1);
917 int cfs_trace_set_debug_mb(int mb)
922 int limit = cfs_trace_max_debug_mb();
923 struct cfs_trace_cpu_data *tcd;
925 if (mb < num_possible_cpus()) {
926 printk(CFS_KERN_WARNING
927 "Lustre: %d MB is too small for debug buffer size, "
928 "setting it to %d MB.\n", mb, num_possible_cpus());
929 mb = num_possible_cpus();
933 printk(CFS_KERN_WARNING
934 "Lustre: %d MB is too large for debug buffer size, "
935 "setting it to %d MB.\n", mb, limit);
939 mb /= num_possible_cpus();
940 pages = mb << (20 - PAGE_CACHE_SHIFT);
942 cfs_tracefile_write_lock();
944 cfs_tcd_for_each(tcd, i, j)
945 tcd->tcd_max_pages = (pages * tcd->tcd_pages_factor) / 100;
947 cfs_tracefile_write_unlock();
952 int cfs_trace_set_debug_mb_usrstr(void *usr_str, int usr_str_nob)
957 rc = cfs_trace_copyin_string(str, sizeof(str), usr_str, usr_str_nob);
961 return cfs_trace_set_debug_mb(simple_strtoul(str, NULL, 0));
964 int cfs_trace_get_debug_mb(void)
968 struct cfs_trace_cpu_data *tcd;
971 cfs_tracefile_read_lock();
973 cfs_tcd_for_each(tcd, i, j)
974 total_pages += tcd->tcd_max_pages;
976 cfs_tracefile_read_unlock();
978 return (total_pages >> (20 - PAGE_CACHE_SHIFT)) + 1;
981 static int tracefiled(void *arg)
983 struct page_collection pc;
984 struct tracefiled_ctl *tctl = arg;
985 struct cfs_trace_page *tage;
986 struct cfs_trace_page *tmp;
993 /* we're started late enough that we pick up init's fs context */
994 /* this is so broken in uml? what on earth is going on? */
996 spin_lock_init(&pc.pc_lock);
997 complete(&tctl->tctl_start);
1000 cfs_waitlink_t __wait;
1002 pc.pc_want_daemon_pages = 0;
1004 if (cfs_list_empty(&pc.pc_pages))
1008 cfs_tracefile_read_lock();
1009 if (cfs_tracefile[0] != 0) {
1010 filp = filp_open(cfs_tracefile,
1011 O_CREAT | O_RDWR | O_LARGEFILE,
1016 printk(CFS_KERN_WARNING "couldn't open %s: "
1017 "%d\n", cfs_tracefile, rc);
1020 cfs_tracefile_read_unlock();
1022 put_pages_on_daemon_list(&pc);
1023 __LASSERT(cfs_list_empty(&pc.pc_pages));
1029 cfs_list_for_each_entry_safe_typed(tage, tmp, &pc.pc_pages,
1030 struct cfs_trace_page,
1032 static loff_t f_pos;
1034 __LASSERT_TAGE_INVARIANT(tage);
1036 if (f_pos >= (off_t)cfs_tracefile_size)
1038 else if (f_pos > (off_t)filp_size(filp))
1039 f_pos = filp_size(filp);
1041 rc = filp_write(filp, page_address(tage->page),
1042 tage->used, &f_pos);
1043 if (rc != (int)tage->used) {
1044 printk(CFS_KERN_WARNING "wanted to write %u "
1045 "but wrote %d\n", tage->used, rc);
1046 put_pages_back(&pc);
1047 __LASSERT(cfs_list_empty(&pc.pc_pages));
1052 filp_close(filp, NULL);
1053 put_pages_on_daemon_list(&pc);
1054 if (!cfs_list_empty(&pc.pc_pages)) {
1057 printk(CFS_KERN_ALERT "Lustre: trace pages aren't "
1059 printk(CFS_KERN_ERR "total cpus(%d): ",
1060 num_possible_cpus());
1061 for (i = 0; i < num_possible_cpus(); i++)
1063 printk(CFS_KERN_ERR "%d(on) ", i);
1065 printk(CFS_KERN_ERR "%d(off) ", i);
1066 printk(CFS_KERN_ERR "\n");
1069 cfs_list_for_each_entry_safe(tage, tmp, &pc.pc_pages,
1071 printk(CFS_KERN_ERR "page %d belongs to cpu "
1072 "%d\n", ++i, tage->cpu);
1073 printk(CFS_KERN_ERR "There are %d pages unwritten\n",
1076 __LASSERT(cfs_list_empty(&pc.pc_pages));
1078 if (cfs_atomic_read(&tctl->tctl_shutdown)) {
1079 if (last_loop == 0) {
1086 cfs_waitlink_init(&__wait);
1087 cfs_waitq_add(&tctl->tctl_waitq, &__wait);
1088 cfs_set_current_state(CFS_TASK_INTERRUPTIBLE);
1089 cfs_waitq_timedwait(&__wait, CFS_TASK_INTERRUPTIBLE,
1090 cfs_time_seconds(1));
1091 cfs_waitq_del(&tctl->tctl_waitq, &__wait);
1093 complete(&tctl->tctl_stop);
1097 int cfs_trace_start_thread(void)
1099 struct tracefiled_ctl *tctl = &trace_tctl;
1102 mutex_lock(&cfs_trace_thread_mutex);
1106 init_completion(&tctl->tctl_start);
1107 init_completion(&tctl->tctl_stop);
1108 cfs_waitq_init(&tctl->tctl_waitq);
1109 cfs_atomic_set(&tctl->tctl_shutdown, 0);
1111 if (IS_ERR(kthread_run(tracefiled, tctl, "ktracefiled"))) {
1116 wait_for_completion(&tctl->tctl_start);
1119 mutex_unlock(&cfs_trace_thread_mutex);
1123 void cfs_trace_stop_thread(void)
1125 struct tracefiled_ctl *tctl = &trace_tctl;
1127 mutex_lock(&cfs_trace_thread_mutex);
1128 if (thread_running) {
1129 printk(CFS_KERN_INFO
1130 "Lustre: shutting down debug daemon thread...\n");
1131 cfs_atomic_set(&tctl->tctl_shutdown, 1);
1132 wait_for_completion(&tctl->tctl_stop);
1135 mutex_unlock(&cfs_trace_thread_mutex);
1138 int cfs_tracefile_init(int max_pages)
1140 struct cfs_trace_cpu_data *tcd;
1146 rc = cfs_tracefile_init_arch();
1150 cfs_tcd_for_each(tcd, i, j) {
1151 /* tcd_pages_factor is initialized int tracefile_init_arch. */
1152 factor = tcd->tcd_pages_factor;
1153 CFS_INIT_LIST_HEAD(&tcd->tcd_pages);
1154 CFS_INIT_LIST_HEAD(&tcd->tcd_stock_pages);
1155 CFS_INIT_LIST_HEAD(&tcd->tcd_daemon_pages);
1156 tcd->tcd_cur_pages = 0;
1157 tcd->tcd_cur_stock_pages = 0;
1158 tcd->tcd_cur_daemon_pages = 0;
1159 tcd->tcd_max_pages = (max_pages * factor) / 100;
1160 LASSERT(tcd->tcd_max_pages > 0);
1161 tcd->tcd_shutting_down = 0;
1167 static void trace_cleanup_on_all_cpus(void)
1169 struct cfs_trace_cpu_data *tcd;
1170 struct cfs_trace_page *tage;
1171 struct cfs_trace_page *tmp;
1174 cfs_for_each_possible_cpu(cpu) {
1175 cfs_tcd_for_each_type_lock(tcd, i, cpu) {
1176 tcd->tcd_shutting_down = 1;
1178 cfs_list_for_each_entry_safe_typed(tage, tmp,
1180 struct cfs_trace_page,
1182 __LASSERT_TAGE_INVARIANT(tage);
1184 cfs_list_del(&tage->linkage);
1185 cfs_tage_free(tage);
1188 tcd->tcd_cur_pages = 0;
1193 static void cfs_trace_cleanup(void)
1195 struct page_collection pc;
1197 CFS_INIT_LIST_HEAD(&pc.pc_pages);
1198 spin_lock_init(&pc.pc_lock);
1200 trace_cleanup_on_all_cpus();
1202 cfs_tracefile_fini_arch();
1205 void cfs_tracefile_exit(void)
1207 cfs_trace_stop_thread();
1208 cfs_trace_cleanup();