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, 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 "../../include/linux/libcfs/libcfs.h"
49 /* XXX move things up to the top, comment */
50 union cfs_trace_data_union (*cfs_trace_data[TCD_MAX_TYPES])[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 atomic_t cfs_tage_allocated = 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(struct list_head *list)
66 return list_entry(list, struct cfs_trace_page, linkage);
69 static struct cfs_trace_page *cfs_tage_alloc(gfp_t gfp)
72 struct cfs_trace_page *tage;
74 /* My caller is trying to free memory */
75 if (!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 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 atomic_dec(&cfs_tage_allocated);
108 static void cfs_tage_to_tail(struct cfs_trace_page *tage,
109 struct list_head *queue)
111 __LASSERT(tage != NULL);
112 __LASSERT(queue != NULL);
114 list_move_tail(&tage->linkage, queue);
117 int cfs_trace_refill_stock(struct cfs_trace_cpu_data *tcd, gfp_t gfp,
118 struct list_head *stock)
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 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(!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 list_del_init(&tage->linkage);
157 tage = cfs_tage_alloc(GFP_ATOMIC);
158 if (unlikely(tage == NULL)) {
159 if ((!memory_pressure_get() ||
160 in_interrupt()) && printk_ratelimit())
162 "cannot allocate a tage (%ld)\n",
169 tage->cpu = smp_processor_id();
170 tage->type = tcd->tcd_type;
171 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 wake_up(&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(KERN_WARNING "debug daemon buffer overflowed; discarding 10%% of pages (%d of %ld)\n",
200 pgcount + 1, tcd->tcd_cur_pages);
202 INIT_LIST_HEAD(&pc.pc_pages);
203 spin_lock_init(&pc.pc_lock);
205 list_for_each_entry_safe(tage, tmp, &tcd->tcd_pages, linkage) {
209 list_move_tail(&tage->linkage, &pc.pc_pages);
210 tcd->tcd_cur_pages--;
212 put_pages_on_tcd_daemon_list(&pc, tcd);
215 /* return a page that has 'len' bytes left at the end */
216 static struct cfs_trace_page *cfs_trace_get_tage(struct cfs_trace_cpu_data *tcd,
219 struct cfs_trace_page *tage;
222 * XXX nikita: do NOT call portals_debug_msg() (CDEBUG/ENTRY/EXIT)
223 * from here: this will lead to infinite recursion.
226 if (len > PAGE_CACHE_SIZE) {
228 "cowardly refusing to write %lu bytes in a page\n", len);
232 tage = cfs_trace_get_tage_try(tcd, len);
237 if (tcd->tcd_cur_pages > 0) {
238 tage = cfs_tage_from_list(tcd->tcd_pages.next);
240 cfs_tage_to_tail(tage, &tcd->tcd_pages);
245 int libcfs_debug_msg(struct libcfs_debug_msg_data *msgdata,
246 const char *format, ...)
251 va_start(args, format);
252 rc = libcfs_debug_vmsg2(msgdata, format, args, NULL);
257 EXPORT_SYMBOL(libcfs_debug_msg);
259 int libcfs_debug_vmsg2(struct libcfs_debug_msg_data *msgdata,
260 const char *format1, va_list args,
261 const char *format2, ...)
263 struct cfs_trace_cpu_data *tcd = NULL;
264 struct ptldebug_header header = {0};
265 struct cfs_trace_page *tage;
266 /* string_buf is used only if tcd != NULL, and is always set then */
267 char *string_buf = NULL;
270 int needed = 85; /* average message length */
276 int mask = msgdata->msg_mask;
277 const char *file = kbasename(msgdata->msg_file);
278 struct cfs_debug_limit_state *cdls = msgdata->msg_cdls;
280 tcd = cfs_trace_get_tcd();
282 /* cfs_trace_get_tcd() grabs a lock, which disables preemption and
283 * pins us to a particular CPU. This avoids an smp_processor_id()
284 * warning on Linux when debugging is enabled. */
285 cfs_set_ptldebug_header(&header, msgdata, CDEBUG_STACK());
287 if (tcd == NULL) /* arch may not log in IRQ context */
290 if (tcd->tcd_cur_pages == 0)
291 header.ph_flags |= PH_FLAG_FIRST_RECORD;
293 if (tcd->tcd_shutting_down) {
294 cfs_trace_put_tcd(tcd);
299 depth = __current_nesting_level();
300 known_size = strlen(file) + 1 + depth;
302 known_size += strlen(msgdata->msg_fn) + 1;
304 if (libcfs_debug_binary)
305 known_size += sizeof(header);
308 * '2' used because vsnprintf return real size required for output
309 * _without_ terminating NULL.
310 * if needed is to small for this format.
312 for (i = 0; i < 2; i++) {
313 tage = cfs_trace_get_tage(tcd, needed + known_size + 1);
315 if (needed + known_size > PAGE_CACHE_SIZE)
318 cfs_trace_put_tcd(tcd);
323 string_buf = (char *)page_address(tage->page) +
324 tage->used + known_size;
326 max_nob = PAGE_CACHE_SIZE - tage->used - known_size;
328 printk(KERN_EMERG "negative max_nob: %d\n",
331 cfs_trace_put_tcd(tcd);
339 needed = vsnprintf(string_buf, max_nob, format1, ap);
344 remain = max_nob - needed;
348 va_start(ap, format2);
349 needed += vsnprintf(string_buf + needed, remain,
354 if (needed < max_nob) /* well. printing ok.. */
358 if (*(string_buf+needed-1) != '\n')
359 printk(KERN_INFO "format at %s:%d:%s doesn't end in newline\n",
360 file, msgdata->msg_line, msgdata->msg_fn);
362 header.ph_len = known_size + needed;
363 debug_buf = (char *)page_address(tage->page) + tage->used;
365 if (libcfs_debug_binary) {
366 memcpy(debug_buf, &header, sizeof(header));
367 tage->used += sizeof(header);
368 debug_buf += sizeof(header);
371 /* indent message according to the nesting level */
372 while (depth-- > 0) {
373 *(debug_buf++) = '.';
377 strcpy(debug_buf, file);
378 tage->used += strlen(file) + 1;
379 debug_buf += strlen(file) + 1;
381 if (msgdata->msg_fn) {
382 strcpy(debug_buf, msgdata->msg_fn);
383 tage->used += strlen(msgdata->msg_fn) + 1;
384 debug_buf += strlen(msgdata->msg_fn) + 1;
387 __LASSERT(debug_buf == string_buf);
389 tage->used += needed;
390 __LASSERT (tage->used <= PAGE_CACHE_SIZE);
393 if ((mask & libcfs_printk) == 0) {
394 /* no console output requested */
396 cfs_trace_put_tcd(tcd);
401 if (libcfs_console_ratelimit &&
402 cdls->cdls_next != 0 && /* not first time ever */
403 !cfs_time_after(cfs_time_current(), cdls->cdls_next)) {
404 /* skipping a console message */
407 cfs_trace_put_tcd(tcd);
411 if (cfs_time_after(cfs_time_current(), cdls->cdls_next +
412 libcfs_console_max_delay
413 + cfs_time_seconds(10))) {
414 /* last timeout was a long time ago */
415 cdls->cdls_delay /= libcfs_console_backoff * 4;
417 cdls->cdls_delay *= libcfs_console_backoff;
420 if (cdls->cdls_delay < libcfs_console_min_delay)
421 cdls->cdls_delay = libcfs_console_min_delay;
422 else if (cdls->cdls_delay > libcfs_console_max_delay)
423 cdls->cdls_delay = libcfs_console_max_delay;
425 /* ensure cdls_next is never zero after it's been seen */
426 cdls->cdls_next = (cfs_time_current() + cdls->cdls_delay) | 1;
430 cfs_print_to_console(&header, mask, string_buf, needed, file,
432 cfs_trace_put_tcd(tcd);
434 string_buf = cfs_trace_get_console_buffer();
437 if (format1 != NULL) {
439 needed = vsnprintf(string_buf,
440 CFS_TRACE_CONSOLE_BUFFER_SIZE,
444 if (format2 != NULL) {
445 remain = CFS_TRACE_CONSOLE_BUFFER_SIZE - needed;
447 va_start(ap, format2);
448 needed += vsnprintf(string_buf+needed, remain,
453 cfs_print_to_console(&header, mask,
454 string_buf, needed, file, msgdata->msg_fn);
456 cfs_trace_put_console_buffer(string_buf);
459 if (cdls != NULL && cdls->cdls_count != 0) {
460 string_buf = cfs_trace_get_console_buffer();
462 needed = snprintf(string_buf, CFS_TRACE_CONSOLE_BUFFER_SIZE,
463 "Skipped %d previous similar message%s\n",
465 (cdls->cdls_count > 1) ? "s" : "");
467 cfs_print_to_console(&header, mask,
468 string_buf, needed, file, msgdata->msg_fn);
470 cfs_trace_put_console_buffer(string_buf);
471 cdls->cdls_count = 0;
476 EXPORT_SYMBOL(libcfs_debug_vmsg2);
479 cfs_trace_assertion_failed(const char *str,
480 struct libcfs_debug_msg_data *msgdata)
482 struct ptldebug_header hdr;
484 libcfs_panic_in_progress = 1;
485 libcfs_catastrophe = 1;
488 cfs_set_ptldebug_header(&hdr, msgdata, CDEBUG_STACK());
490 cfs_print_to_console(&hdr, D_EMERG, str, strlen(str),
491 msgdata->msg_file, msgdata->msg_fn);
493 panic("Lustre debug assertion failure\n");
499 panic_collect_pages(struct page_collection *pc)
501 /* Do the collect_pages job on a single CPU: assumes that all other
502 * CPUs have been stopped during a panic. If this isn't true for some
503 * arch, this will have to be implemented separately in each arch. */
506 struct cfs_trace_cpu_data *tcd;
508 INIT_LIST_HEAD(&pc->pc_pages);
510 cfs_tcd_for_each(tcd, i, j) {
511 list_splice_init(&tcd->tcd_pages, &pc->pc_pages);
512 tcd->tcd_cur_pages = 0;
514 if (pc->pc_want_daemon_pages) {
515 list_splice_init(&tcd->tcd_daemon_pages,
517 tcd->tcd_cur_daemon_pages = 0;
522 static void collect_pages_on_all_cpus(struct page_collection *pc)
524 struct cfs_trace_cpu_data *tcd;
527 spin_lock(&pc->pc_lock);
528 for_each_possible_cpu(cpu) {
529 cfs_tcd_for_each_type_lock(tcd, i, cpu) {
530 list_splice_init(&tcd->tcd_pages, &pc->pc_pages);
531 tcd->tcd_cur_pages = 0;
532 if (pc->pc_want_daemon_pages) {
533 list_splice_init(&tcd->tcd_daemon_pages,
535 tcd->tcd_cur_daemon_pages = 0;
539 spin_unlock(&pc->pc_lock);
542 static void collect_pages(struct page_collection *pc)
544 INIT_LIST_HEAD(&pc->pc_pages);
546 if (libcfs_panic_in_progress)
547 panic_collect_pages(pc);
549 collect_pages_on_all_cpus(pc);
552 static void put_pages_back_on_all_cpus(struct page_collection *pc)
554 struct cfs_trace_cpu_data *tcd;
555 struct list_head *cur_head;
556 struct cfs_trace_page *tage;
557 struct cfs_trace_page *tmp;
560 spin_lock(&pc->pc_lock);
561 for_each_possible_cpu(cpu) {
562 cfs_tcd_for_each_type_lock(tcd, i, cpu) {
563 cur_head = tcd->tcd_pages.next;
565 list_for_each_entry_safe(tage, tmp, &pc->pc_pages,
568 __LASSERT_TAGE_INVARIANT(tage);
570 if (tage->cpu != cpu || tage->type != i)
573 cfs_tage_to_tail(tage, cur_head);
574 tcd->tcd_cur_pages++;
578 spin_unlock(&pc->pc_lock);
581 static void put_pages_back(struct page_collection *pc)
583 if (!libcfs_panic_in_progress)
584 put_pages_back_on_all_cpus(pc);
587 /* Add pages to a per-cpu debug daemon ringbuffer. This buffer makes sure that
588 * we have a good amount of data at all times for dumping during an LBUG, even
589 * if we have been steadily writing (and otherwise discarding) pages via the
591 static void put_pages_on_tcd_daemon_list(struct page_collection *pc,
592 struct cfs_trace_cpu_data *tcd)
594 struct cfs_trace_page *tage;
595 struct cfs_trace_page *tmp;
597 spin_lock(&pc->pc_lock);
598 list_for_each_entry_safe(tage, tmp, &pc->pc_pages, linkage) {
600 __LASSERT_TAGE_INVARIANT(tage);
602 if (tage->cpu != tcd->tcd_cpu || tage->type != tcd->tcd_type)
605 cfs_tage_to_tail(tage, &tcd->tcd_daemon_pages);
606 tcd->tcd_cur_daemon_pages++;
608 if (tcd->tcd_cur_daemon_pages > tcd->tcd_max_pages) {
609 struct cfs_trace_page *victim;
611 __LASSERT(!list_empty(&tcd->tcd_daemon_pages));
612 victim = cfs_tage_from_list(tcd->tcd_daemon_pages.next);
614 __LASSERT_TAGE_INVARIANT(victim);
616 list_del(&victim->linkage);
617 cfs_tage_free(victim);
618 tcd->tcd_cur_daemon_pages--;
621 spin_unlock(&pc->pc_lock);
624 static void put_pages_on_daemon_list(struct page_collection *pc)
626 struct cfs_trace_cpu_data *tcd;
629 for_each_possible_cpu(cpu) {
630 cfs_tcd_for_each_type_lock(tcd, i, cpu)
631 put_pages_on_tcd_daemon_list(pc, tcd);
635 void cfs_trace_debug_print(void)
637 struct page_collection pc;
638 struct cfs_trace_page *tage;
639 struct cfs_trace_page *tmp;
641 spin_lock_init(&pc.pc_lock);
643 pc.pc_want_daemon_pages = 1;
645 list_for_each_entry_safe(tage, tmp, &pc.pc_pages, linkage) {
649 __LASSERT_TAGE_INVARIANT(tage);
652 p = page_address(page);
653 while (p < ((char *)page_address(page) + tage->used)) {
654 struct ptldebug_header *hdr;
659 p += strlen(file) + 1;
662 len = hdr->ph_len - (int)(p - (char *)hdr);
664 cfs_print_to_console(hdr, D_EMERG, p, len, file, fn);
669 list_del(&tage->linkage);
674 int cfs_tracefile_dump_all_pages(char *filename)
676 struct page_collection pc;
678 struct cfs_trace_page *tage;
679 struct cfs_trace_page *tmp;
685 cfs_tracefile_write_lock();
687 filp = filp_open(filename, O_CREAT|O_EXCL|O_WRONLY|O_LARGEFILE, 0600);
691 printk(KERN_ERR "LustreError: can't open %s for dump: rc %d\n",
696 spin_lock_init(&pc.pc_lock);
697 pc.pc_want_daemon_pages = 1;
699 if (list_empty(&pc.pc_pages)) {
704 /* ok, for now, just write the pages. in the future we'll be building
705 * iobufs with the pages and calling generic_direct_IO */
707 list_for_each_entry_safe(tage, tmp, &pc.pc_pages, linkage) {
709 __LASSERT_TAGE_INVARIANT(tage);
711 buf = kmap(tage->page);
712 rc = vfs_write(filp, (__force const char __user *)buf,
713 tage->used, &filp->f_pos);
716 if (rc != (int)tage->used) {
717 printk(KERN_WARNING "wanted to write %u but wrote %d\n",
720 __LASSERT(list_empty(&pc.pc_pages));
723 list_del(&tage->linkage);
727 rc = vfs_fsync(filp, 1);
729 printk(KERN_ERR "sync returns %d\n", rc);
731 filp_close(filp, NULL);
733 cfs_tracefile_write_unlock();
737 void cfs_trace_flush_pages(void)
739 struct page_collection pc;
740 struct cfs_trace_page *tage;
741 struct cfs_trace_page *tmp;
743 spin_lock_init(&pc.pc_lock);
745 pc.pc_want_daemon_pages = 1;
747 list_for_each_entry_safe(tage, tmp, &pc.pc_pages, linkage) {
749 __LASSERT_TAGE_INVARIANT(tage);
751 list_del(&tage->linkage);
756 int cfs_trace_copyin_string(char *knl_buffer, int knl_buffer_nob,
757 const char __user *usr_buffer, int usr_buffer_nob)
761 if (usr_buffer_nob > knl_buffer_nob)
764 if (copy_from_user((void *)knl_buffer,
765 usr_buffer, usr_buffer_nob))
768 nob = strnlen(knl_buffer, usr_buffer_nob);
769 while (nob-- >= 0) /* strip trailing whitespace */
770 if (!isspace(knl_buffer[nob]))
773 if (nob < 0) /* empty string */
776 if (nob == knl_buffer_nob) /* no space to terminate */
779 knl_buffer[nob + 1] = 0; /* terminate */
782 EXPORT_SYMBOL(cfs_trace_copyin_string);
784 int cfs_trace_copyout_string(char __user *usr_buffer, int usr_buffer_nob,
785 const char *knl_buffer, char *append)
787 /* NB if 'append' != NULL, it's a single character to append to the
788 * copied out string - usually "\n", for /proc entries and "" (i.e. a
789 * terminating zero byte) for sysctl entries */
790 int nob = strlen(knl_buffer);
792 if (nob > usr_buffer_nob)
793 nob = usr_buffer_nob;
795 if (copy_to_user(usr_buffer, knl_buffer, nob))
798 if (append != NULL && nob < usr_buffer_nob) {
799 if (copy_to_user(usr_buffer + nob, append, 1))
807 EXPORT_SYMBOL(cfs_trace_copyout_string);
809 int cfs_trace_allocate_string_buffer(char **str, int nob)
811 if (nob > 2 * PAGE_CACHE_SIZE) /* string must be "sensible" */
814 *str = kmalloc(nob, GFP_IOFS | __GFP_ZERO);
821 void cfs_trace_free_string_buffer(char *str, int nob)
826 int cfs_trace_dump_debug_buffer_usrstr(void __user *usr_str, int usr_str_nob)
831 rc = cfs_trace_allocate_string_buffer(&str, usr_str_nob + 1);
835 rc = cfs_trace_copyin_string(str, usr_str_nob + 1,
836 usr_str, usr_str_nob);
844 rc = cfs_tracefile_dump_all_pages(str);
846 cfs_trace_free_string_buffer(str, usr_str_nob + 1);
850 int cfs_trace_daemon_command(char *str)
854 cfs_tracefile_write_lock();
856 if (strcmp(str, "stop") == 0) {
857 cfs_tracefile_write_unlock();
858 cfs_trace_stop_thread();
859 cfs_tracefile_write_lock();
860 memset(cfs_tracefile, 0, sizeof(cfs_tracefile));
862 } else if (strncmp(str, "size=", 5) == 0) {
863 cfs_tracefile_size = simple_strtoul(str + 5, NULL, 0);
864 if (cfs_tracefile_size < 10 || cfs_tracefile_size > 20480)
865 cfs_tracefile_size = CFS_TRACEFILE_SIZE;
867 cfs_tracefile_size <<= 20;
869 } else if (strlen(str) >= sizeof(cfs_tracefile)) {
871 } else if (str[0] != '/') {
874 strcpy(cfs_tracefile, str);
877 "Lustre: debug daemon will attempt to start writing to %s (%lukB max)\n",
879 (long)(cfs_tracefile_size >> 10));
881 cfs_trace_start_thread();
884 cfs_tracefile_write_unlock();
888 int cfs_trace_daemon_command_usrstr(void __user *usr_str, int usr_str_nob)
893 rc = cfs_trace_allocate_string_buffer(&str, usr_str_nob + 1);
897 rc = cfs_trace_copyin_string(str, usr_str_nob + 1,
898 usr_str, usr_str_nob);
900 rc = cfs_trace_daemon_command(str);
902 cfs_trace_free_string_buffer(str, usr_str_nob + 1);
906 int cfs_trace_set_debug_mb(int mb)
911 int limit = cfs_trace_max_debug_mb();
912 struct cfs_trace_cpu_data *tcd;
914 if (mb < num_possible_cpus()) {
916 "Lustre: %d MB is too small for debug buffer size, setting it to %d MB.\n",
917 mb, num_possible_cpus());
918 mb = num_possible_cpus();
923 "Lustre: %d MB is too large for debug buffer size, setting it to %d MB.\n",
928 mb /= num_possible_cpus();
929 pages = mb << (20 - PAGE_CACHE_SHIFT);
931 cfs_tracefile_write_lock();
933 cfs_tcd_for_each(tcd, i, j)
934 tcd->tcd_max_pages = (pages * tcd->tcd_pages_factor) / 100;
936 cfs_tracefile_write_unlock();
941 int cfs_trace_set_debug_mb_usrstr(void __user *usr_str, int usr_str_nob)
946 rc = cfs_trace_copyin_string(str, sizeof(str), usr_str, usr_str_nob);
950 return cfs_trace_set_debug_mb(simple_strtoul(str, NULL, 0));
953 int cfs_trace_get_debug_mb(void)
957 struct cfs_trace_cpu_data *tcd;
960 cfs_tracefile_read_lock();
962 cfs_tcd_for_each(tcd, i, j)
963 total_pages += tcd->tcd_max_pages;
965 cfs_tracefile_read_unlock();
967 return (total_pages >> (20 - PAGE_CACHE_SHIFT)) + 1;
970 static int tracefiled(void *arg)
972 struct page_collection pc;
973 struct tracefiled_ctl *tctl = arg;
974 struct cfs_trace_page *tage;
975 struct cfs_trace_page *tmp;
983 /* we're started late enough that we pick up init's fs context */
984 /* this is so broken in uml? what on earth is going on? */
986 spin_lock_init(&pc.pc_lock);
987 complete(&tctl->tctl_start);
992 pc.pc_want_daemon_pages = 0;
994 if (list_empty(&pc.pc_pages))
998 cfs_tracefile_read_lock();
999 if (cfs_tracefile[0] != 0) {
1000 filp = filp_open(cfs_tracefile,
1001 O_CREAT | O_RDWR | O_LARGEFILE,
1006 printk(KERN_WARNING "couldn't open %s: %d\n",
1010 cfs_tracefile_read_unlock();
1012 put_pages_on_daemon_list(&pc);
1013 __LASSERT(list_empty(&pc.pc_pages));
1019 list_for_each_entry_safe(tage, tmp, &pc.pc_pages,
1021 static loff_t f_pos;
1023 __LASSERT_TAGE_INVARIANT(tage);
1025 if (f_pos >= (off_t)cfs_tracefile_size)
1027 else if (f_pos > i_size_read(filp->f_dentry->d_inode))
1028 f_pos = i_size_read(filp->f_dentry->d_inode);
1030 buf = kmap(tage->page);
1031 rc = vfs_write(filp, (__force const char __user *)buf,
1032 tage->used, &f_pos);
1035 if (rc != (int)tage->used) {
1036 printk(KERN_WARNING "wanted to write %u but wrote %d\n",
1038 put_pages_back(&pc);
1039 __LASSERT(list_empty(&pc.pc_pages));
1044 filp_close(filp, NULL);
1045 put_pages_on_daemon_list(&pc);
1046 if (!list_empty(&pc.pc_pages)) {
1049 printk(KERN_ALERT "Lustre: trace pages aren't empty\n");
1050 printk(KERN_ERR "total cpus(%d): ",
1051 num_possible_cpus());
1052 for (i = 0; i < num_possible_cpus(); i++)
1054 printk(KERN_ERR "%d(on) ", i);
1056 printk(KERN_ERR "%d(off) ", i);
1057 printk(KERN_ERR "\n");
1060 list_for_each_entry_safe(tage, tmp, &pc.pc_pages,
1062 printk(KERN_ERR "page %d belongs to cpu %d\n",
1064 printk(KERN_ERR "There are %d pages unwritten\n",
1067 __LASSERT(list_empty(&pc.pc_pages));
1069 if (atomic_read(&tctl->tctl_shutdown)) {
1070 if (last_loop == 0) {
1077 init_waitqueue_entry(&__wait, current);
1078 add_wait_queue(&tctl->tctl_waitq, &__wait);
1079 set_current_state(TASK_INTERRUPTIBLE);
1080 schedule_timeout(cfs_time_seconds(1));
1081 remove_wait_queue(&tctl->tctl_waitq, &__wait);
1083 complete(&tctl->tctl_stop);
1087 int cfs_trace_start_thread(void)
1089 struct tracefiled_ctl *tctl = &trace_tctl;
1092 mutex_lock(&cfs_trace_thread_mutex);
1096 init_completion(&tctl->tctl_start);
1097 init_completion(&tctl->tctl_stop);
1098 init_waitqueue_head(&tctl->tctl_waitq);
1099 atomic_set(&tctl->tctl_shutdown, 0);
1101 if (IS_ERR(kthread_run(tracefiled, tctl, "ktracefiled"))) {
1106 wait_for_completion(&tctl->tctl_start);
1109 mutex_unlock(&cfs_trace_thread_mutex);
1113 void cfs_trace_stop_thread(void)
1115 struct tracefiled_ctl *tctl = &trace_tctl;
1117 mutex_lock(&cfs_trace_thread_mutex);
1118 if (thread_running) {
1120 "Lustre: shutting down debug daemon thread...\n");
1121 atomic_set(&tctl->tctl_shutdown, 1);
1122 wait_for_completion(&tctl->tctl_stop);
1125 mutex_unlock(&cfs_trace_thread_mutex);
1128 int cfs_tracefile_init(int max_pages)
1130 struct cfs_trace_cpu_data *tcd;
1136 rc = cfs_tracefile_init_arch();
1140 cfs_tcd_for_each(tcd, i, j) {
1141 /* tcd_pages_factor is initialized int tracefile_init_arch. */
1142 factor = tcd->tcd_pages_factor;
1143 INIT_LIST_HEAD(&tcd->tcd_pages);
1144 INIT_LIST_HEAD(&tcd->tcd_stock_pages);
1145 INIT_LIST_HEAD(&tcd->tcd_daemon_pages);
1146 tcd->tcd_cur_pages = 0;
1147 tcd->tcd_cur_stock_pages = 0;
1148 tcd->tcd_cur_daemon_pages = 0;
1149 tcd->tcd_max_pages = (max_pages * factor) / 100;
1150 LASSERT(tcd->tcd_max_pages > 0);
1151 tcd->tcd_shutting_down = 0;
1157 static void trace_cleanup_on_all_cpus(void)
1159 struct cfs_trace_cpu_data *tcd;
1160 struct cfs_trace_page *tage;
1161 struct cfs_trace_page *tmp;
1164 for_each_possible_cpu(cpu) {
1165 cfs_tcd_for_each_type_lock(tcd, i, cpu) {
1166 tcd->tcd_shutting_down = 1;
1168 list_for_each_entry_safe(tage, tmp, &tcd->tcd_pages,
1170 __LASSERT_TAGE_INVARIANT(tage);
1172 list_del(&tage->linkage);
1173 cfs_tage_free(tage);
1176 tcd->tcd_cur_pages = 0;
1181 static void cfs_trace_cleanup(void)
1183 struct page_collection pc;
1185 INIT_LIST_HEAD(&pc.pc_pages);
1186 spin_lock_init(&pc.pc_lock);
1188 trace_cleanup_on_all_cpus();
1190 cfs_tracefile_fini_arch();
1193 void cfs_tracefile_exit(void)
1195 cfs_trace_stop_thread();
1196 cfs_trace_cleanup();