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>
42 #define DEBUG_SUBSYSTEM S_LNET
43 #define LUSTRE_TRACEFILE_PRIVATE
44 #include "tracefile.h"
46 #include "../../include/linux/libcfs/libcfs.h"
48 /* XXX move things up to the top, comment */
49 union cfs_trace_data_union (*cfs_trace_data[TCD_MAX_TYPES])[NR_CPUS] __cacheline_aligned;
51 char cfs_tracefile[TRACEFILE_NAME_SIZE];
52 long long cfs_tracefile_size = CFS_TRACEFILE_SIZE;
53 static struct tracefiled_ctl trace_tctl;
54 static DEFINE_MUTEX(cfs_trace_thread_mutex);
55 static int thread_running;
57 static atomic_t cfs_tage_allocated = ATOMIC_INIT(0);
59 static void put_pages_on_tcd_daemon_list(struct page_collection *pc,
60 struct cfs_trace_cpu_data *tcd);
62 static inline struct cfs_trace_page *
63 cfs_tage_from_list(struct list_head *list)
65 return list_entry(list, struct cfs_trace_page, linkage);
68 static struct cfs_trace_page *cfs_tage_alloc(gfp_t gfp)
71 struct cfs_trace_page *tage;
73 /* My caller is trying to free memory */
74 if (!in_interrupt() && memory_pressure_get())
78 * Don't spam console with allocation failures: they will be reported
79 * by upper layer anyway.
82 page = alloc_page(gfp);
86 tage = kmalloc(sizeof(*tage), gfp);
93 atomic_inc(&cfs_tage_allocated);
97 static void cfs_tage_free(struct cfs_trace_page *tage)
99 __free_page(tage->page);
101 atomic_dec(&cfs_tage_allocated);
104 static void cfs_tage_to_tail(struct cfs_trace_page *tage,
105 struct list_head *queue)
107 list_move_tail(&tage->linkage, queue);
110 int cfs_trace_refill_stock(struct cfs_trace_cpu_data *tcd, gfp_t gfp,
111 struct list_head *stock)
116 * XXX nikita: do NOT call portals_debug_msg() (CDEBUG/ENTRY/EXIT)
117 * from here: this will lead to infinite recursion.
120 for (i = 0; i + tcd->tcd_cur_stock_pages < TCD_STOCK_PAGES ; ++i) {
121 struct cfs_trace_page *tage;
123 tage = cfs_tage_alloc(gfp);
126 list_add_tail(&tage->linkage, stock);
131 /* return a page that has 'len' bytes left at the end */
132 static struct cfs_trace_page *
133 cfs_trace_get_tage_try(struct cfs_trace_cpu_data *tcd, unsigned long len)
135 struct cfs_trace_page *tage;
137 if (tcd->tcd_cur_pages > 0) {
138 __LASSERT(!list_empty(&tcd->tcd_pages));
139 tage = cfs_tage_from_list(tcd->tcd_pages.prev);
140 if (tage->used + len <= PAGE_CACHE_SIZE)
144 if (tcd->tcd_cur_pages < tcd->tcd_max_pages) {
145 if (tcd->tcd_cur_stock_pages > 0) {
146 tage = cfs_tage_from_list(tcd->tcd_stock_pages.prev);
147 --tcd->tcd_cur_stock_pages;
148 list_del_init(&tage->linkage);
150 tage = cfs_tage_alloc(GFP_ATOMIC);
151 if (unlikely(!tage)) {
152 if ((!memory_pressure_get() ||
153 in_interrupt()) && 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 wake_up(&tctl->tctl_waitq);
179 static void cfs_tcd_shrink(struct cfs_trace_cpu_data *tcd)
181 int pgcount = tcd->tcd_cur_pages / 10;
182 struct page_collection pc;
183 struct cfs_trace_page *tage;
184 struct cfs_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; discarding 10%% of pages (%d of %ld)\n",
193 pgcount + 1, tcd->tcd_cur_pages);
195 INIT_LIST_HEAD(&pc.pc_pages);
197 list_for_each_entry_safe(tage, tmp, &tcd->tcd_pages, linkage) {
201 list_move_tail(&tage->linkage, &pc.pc_pages);
202 tcd->tcd_cur_pages--;
204 put_pages_on_tcd_daemon_list(&pc, tcd);
207 /* return a page that has 'len' bytes left at the end */
208 static struct cfs_trace_page *cfs_trace_get_tage(struct cfs_trace_cpu_data *tcd,
211 struct cfs_trace_page *tage;
214 * XXX nikita: do NOT call portals_debug_msg() (CDEBUG/ENTRY/EXIT)
215 * from here: this will lead to infinite recursion.
218 if (len > PAGE_CACHE_SIZE) {
219 pr_err("cowardly refusing to write %lu bytes in a page\n", len);
223 tage = cfs_trace_get_tage_try(tcd, len);
228 if (tcd->tcd_cur_pages > 0) {
229 tage = cfs_tage_from_list(tcd->tcd_pages.next);
231 cfs_tage_to_tail(tage, &tcd->tcd_pages);
236 int libcfs_debug_msg(struct libcfs_debug_msg_data *msgdata,
237 const char *format, ...)
242 va_start(args, format);
243 rc = libcfs_debug_vmsg2(msgdata, format, args, NULL);
248 EXPORT_SYMBOL(libcfs_debug_msg);
250 int libcfs_debug_vmsg2(struct libcfs_debug_msg_data *msgdata,
251 const char *format1, va_list args,
252 const char *format2, ...)
254 struct cfs_trace_cpu_data *tcd = NULL;
255 struct ptldebug_header header = {0};
256 struct cfs_trace_page *tage;
257 /* string_buf is used only if tcd != NULL, and is always set then */
258 char *string_buf = NULL;
261 int needed = 85; /* average message length */
267 int mask = msgdata->msg_mask;
268 const char *file = kbasename(msgdata->msg_file);
269 struct cfs_debug_limit_state *cdls = msgdata->msg_cdls;
271 tcd = cfs_trace_get_tcd();
273 /* cfs_trace_get_tcd() grabs a lock, which disables preemption and
274 * pins us to a particular CPU. This avoids an smp_processor_id()
275 * warning on Linux when debugging is enabled. */
276 cfs_set_ptldebug_header(&header, msgdata, CDEBUG_STACK());
278 if (!tcd) /* arch may not log in IRQ context */
281 if (tcd->tcd_cur_pages == 0)
282 header.ph_flags |= PH_FLAG_FIRST_RECORD;
284 if (tcd->tcd_shutting_down) {
285 cfs_trace_put_tcd(tcd);
290 depth = __current_nesting_level();
291 known_size = strlen(file) + 1 + depth;
293 known_size += strlen(msgdata->msg_fn) + 1;
295 if (libcfs_debug_binary)
296 known_size += sizeof(header);
299 * '2' used because vsnprintf return real size required for output
300 * _without_ terminating NULL.
301 * if needed is to small for this format.
303 for (i = 0; i < 2; i++) {
304 tage = cfs_trace_get_tage(tcd, needed + known_size + 1);
306 if (needed + known_size > PAGE_CACHE_SIZE)
309 cfs_trace_put_tcd(tcd);
314 string_buf = (char *)page_address(tage->page) +
315 tage->used + known_size;
317 max_nob = PAGE_CACHE_SIZE - tage->used - known_size;
319 printk(KERN_EMERG "negative max_nob: %d\n",
322 cfs_trace_put_tcd(tcd);
330 needed = vsnprintf(string_buf, max_nob, format1, ap);
335 remain = max_nob - needed;
339 va_start(ap, format2);
340 needed += vsnprintf(string_buf + needed, remain,
345 if (needed < max_nob) /* well. printing ok.. */
349 if (*(string_buf + needed - 1) != '\n')
350 printk(KERN_INFO "format at %s:%d:%s doesn't end in newline\n",
351 file, msgdata->msg_line, msgdata->msg_fn);
353 header.ph_len = known_size + needed;
354 debug_buf = (char *)page_address(tage->page) + tage->used;
356 if (libcfs_debug_binary) {
357 memcpy(debug_buf, &header, sizeof(header));
358 tage->used += sizeof(header);
359 debug_buf += sizeof(header);
362 /* indent message according to the nesting level */
363 while (depth-- > 0) {
364 *(debug_buf++) = '.';
368 strcpy(debug_buf, file);
369 tage->used += strlen(file) + 1;
370 debug_buf += strlen(file) + 1;
372 if (msgdata->msg_fn) {
373 strcpy(debug_buf, msgdata->msg_fn);
374 tage->used += strlen(msgdata->msg_fn) + 1;
375 debug_buf += strlen(msgdata->msg_fn) + 1;
378 __LASSERT(debug_buf == string_buf);
380 tage->used += needed;
381 __LASSERT(tage->used <= PAGE_CACHE_SIZE);
384 if ((mask & libcfs_printk) == 0) {
385 /* no console output requested */
387 cfs_trace_put_tcd(tcd);
392 if (libcfs_console_ratelimit &&
393 cdls->cdls_next != 0 && /* not first time ever */
394 !cfs_time_after(cfs_time_current(), cdls->cdls_next)) {
395 /* skipping a console message */
398 cfs_trace_put_tcd(tcd);
402 if (cfs_time_after(cfs_time_current(), cdls->cdls_next +
403 libcfs_console_max_delay
404 + cfs_time_seconds(10))) {
405 /* last timeout was a long time ago */
406 cdls->cdls_delay /= libcfs_console_backoff * 4;
408 cdls->cdls_delay *= libcfs_console_backoff;
411 if (cdls->cdls_delay < libcfs_console_min_delay)
412 cdls->cdls_delay = libcfs_console_min_delay;
413 else if (cdls->cdls_delay > libcfs_console_max_delay)
414 cdls->cdls_delay = libcfs_console_max_delay;
416 /* ensure cdls_next is never zero after it's been seen */
417 cdls->cdls_next = (cfs_time_current() + cdls->cdls_delay) | 1;
421 cfs_print_to_console(&header, mask, string_buf, needed, file,
423 cfs_trace_put_tcd(tcd);
425 string_buf = cfs_trace_get_console_buffer();
430 needed = vsnprintf(string_buf,
431 CFS_TRACE_CONSOLE_BUFFER_SIZE,
436 remain = CFS_TRACE_CONSOLE_BUFFER_SIZE - needed;
438 va_start(ap, format2);
439 needed += vsnprintf(string_buf + needed, remain,
444 cfs_print_to_console(&header, mask,
445 string_buf, needed, file, msgdata->msg_fn);
450 if (cdls && cdls->cdls_count != 0) {
451 string_buf = cfs_trace_get_console_buffer();
453 needed = snprintf(string_buf, CFS_TRACE_CONSOLE_BUFFER_SIZE,
454 "Skipped %d previous similar message%s\n",
456 (cdls->cdls_count > 1) ? "s" : "");
458 cfs_print_to_console(&header, mask,
459 string_buf, needed, file, msgdata->msg_fn);
462 cdls->cdls_count = 0;
467 EXPORT_SYMBOL(libcfs_debug_vmsg2);
470 cfs_trace_assertion_failed(const char *str,
471 struct libcfs_debug_msg_data *msgdata)
473 struct ptldebug_header hdr;
475 libcfs_panic_in_progress = 1;
476 libcfs_catastrophe = 1;
479 cfs_set_ptldebug_header(&hdr, msgdata, CDEBUG_STACK());
481 cfs_print_to_console(&hdr, D_EMERG, str, strlen(str),
482 msgdata->msg_file, msgdata->msg_fn);
484 panic("Lustre debug assertion failure\n");
490 panic_collect_pages(struct page_collection *pc)
492 /* Do the collect_pages job on a single CPU: assumes that all other
493 * CPUs have been stopped during a panic. If this isn't true for some
494 * arch, this will have to be implemented separately in each arch. */
497 struct cfs_trace_cpu_data *tcd;
499 INIT_LIST_HEAD(&pc->pc_pages);
501 cfs_tcd_for_each(tcd, i, j) {
502 list_splice_init(&tcd->tcd_pages, &pc->pc_pages);
503 tcd->tcd_cur_pages = 0;
505 if (pc->pc_want_daemon_pages) {
506 list_splice_init(&tcd->tcd_daemon_pages, &pc->pc_pages);
507 tcd->tcd_cur_daemon_pages = 0;
512 static void collect_pages_on_all_cpus(struct page_collection *pc)
514 struct cfs_trace_cpu_data *tcd;
517 for_each_possible_cpu(cpu) {
518 cfs_tcd_for_each_type_lock(tcd, i, cpu) {
519 list_splice_init(&tcd->tcd_pages, &pc->pc_pages);
520 tcd->tcd_cur_pages = 0;
521 if (pc->pc_want_daemon_pages) {
522 list_splice_init(&tcd->tcd_daemon_pages,
524 tcd->tcd_cur_daemon_pages = 0;
530 static void collect_pages(struct page_collection *pc)
532 INIT_LIST_HEAD(&pc->pc_pages);
534 if (libcfs_panic_in_progress)
535 panic_collect_pages(pc);
537 collect_pages_on_all_cpus(pc);
540 static void put_pages_back_on_all_cpus(struct page_collection *pc)
542 struct cfs_trace_cpu_data *tcd;
543 struct list_head *cur_head;
544 struct cfs_trace_page *tage;
545 struct cfs_trace_page *tmp;
548 for_each_possible_cpu(cpu) {
549 cfs_tcd_for_each_type_lock(tcd, i, cpu) {
550 cur_head = tcd->tcd_pages.next;
552 list_for_each_entry_safe(tage, tmp, &pc->pc_pages,
555 __LASSERT_TAGE_INVARIANT(tage);
557 if (tage->cpu != cpu || tage->type != i)
560 cfs_tage_to_tail(tage, cur_head);
561 tcd->tcd_cur_pages++;
567 static void put_pages_back(struct page_collection *pc)
569 if (!libcfs_panic_in_progress)
570 put_pages_back_on_all_cpus(pc);
573 /* Add pages to a per-cpu debug daemon ringbuffer. This buffer makes sure that
574 * we have a good amount of data at all times for dumping during an LBUG, even
575 * if we have been steadily writing (and otherwise discarding) pages via the
577 static void put_pages_on_tcd_daemon_list(struct page_collection *pc,
578 struct cfs_trace_cpu_data *tcd)
580 struct cfs_trace_page *tage;
581 struct cfs_trace_page *tmp;
583 list_for_each_entry_safe(tage, tmp, &pc->pc_pages, linkage) {
585 __LASSERT_TAGE_INVARIANT(tage);
587 if (tage->cpu != tcd->tcd_cpu || tage->type != tcd->tcd_type)
590 cfs_tage_to_tail(tage, &tcd->tcd_daemon_pages);
591 tcd->tcd_cur_daemon_pages++;
593 if (tcd->tcd_cur_daemon_pages > tcd->tcd_max_pages) {
594 struct cfs_trace_page *victim;
596 __LASSERT(!list_empty(&tcd->tcd_daemon_pages));
597 victim = cfs_tage_from_list(tcd->tcd_daemon_pages.next);
599 __LASSERT_TAGE_INVARIANT(victim);
601 list_del(&victim->linkage);
602 cfs_tage_free(victim);
603 tcd->tcd_cur_daemon_pages--;
608 static void put_pages_on_daemon_list(struct page_collection *pc)
610 struct cfs_trace_cpu_data *tcd;
613 for_each_possible_cpu(cpu) {
614 cfs_tcd_for_each_type_lock(tcd, i, cpu)
615 put_pages_on_tcd_daemon_list(pc, tcd);
619 void cfs_trace_debug_print(void)
621 struct page_collection pc;
622 struct cfs_trace_page *tage;
623 struct cfs_trace_page *tmp;
625 pc.pc_want_daemon_pages = 1;
627 list_for_each_entry_safe(tage, tmp, &pc.pc_pages, linkage) {
631 __LASSERT_TAGE_INVARIANT(tage);
634 p = page_address(page);
635 while (p < ((char *)page_address(page) + tage->used)) {
636 struct ptldebug_header *hdr;
642 p += strlen(file) + 1;
645 len = hdr->ph_len - (int)(p - (char *)hdr);
647 cfs_print_to_console(hdr, D_EMERG, p, len, file, fn);
652 list_del(&tage->linkage);
657 int cfs_tracefile_dump_all_pages(char *filename)
659 struct page_collection pc;
661 struct cfs_trace_page *tage;
662 struct cfs_trace_page *tmp;
668 cfs_tracefile_write_lock();
670 filp = filp_open(filename, O_CREAT | O_EXCL | O_WRONLY | O_LARGEFILE,
675 pr_err("LustreError: can't open %s for dump: rc %d\n",
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 list_for_each_entry_safe(tage, tmp, &pc.pc_pages, linkage) {
692 __LASSERT_TAGE_INVARIANT(tage);
694 buf = kmap(tage->page);
695 rc = vfs_write(filp, (__force const char __user *)buf,
696 tage->used, &filp->f_pos);
699 if (rc != (int)tage->used) {
700 printk(KERN_WARNING "wanted to write %u but wrote %d\n",
703 __LASSERT(list_empty(&pc.pc_pages));
706 list_del(&tage->linkage);
710 rc = vfs_fsync(filp, 1);
712 pr_err("sync returns %d\n", rc);
714 filp_close(filp, NULL);
716 cfs_tracefile_write_unlock();
720 void cfs_trace_flush_pages(void)
722 struct page_collection pc;
723 struct cfs_trace_page *tage;
724 struct cfs_trace_page *tmp;
726 pc.pc_want_daemon_pages = 1;
728 list_for_each_entry_safe(tage, tmp, &pc.pc_pages, linkage) {
730 __LASSERT_TAGE_INVARIANT(tage);
732 list_del(&tage->linkage);
737 int cfs_trace_copyin_string(char *knl_buffer, int knl_buffer_nob,
738 const char __user *usr_buffer, int usr_buffer_nob)
742 if (usr_buffer_nob > knl_buffer_nob)
745 if (copy_from_user((void *)knl_buffer,
746 usr_buffer, usr_buffer_nob))
749 nob = strnlen(knl_buffer, usr_buffer_nob);
750 while (nob-- >= 0) /* strip trailing whitespace */
751 if (!isspace(knl_buffer[nob]))
754 if (nob < 0) /* empty string */
757 if (nob == knl_buffer_nob) /* no space to terminate */
760 knl_buffer[nob + 1] = 0; /* terminate */
763 EXPORT_SYMBOL(cfs_trace_copyin_string);
765 int cfs_trace_copyout_string(char __user *usr_buffer, int usr_buffer_nob,
766 const char *knl_buffer, char *append)
769 * NB if 'append' != NULL, it's a single character to append to the
770 * copied out string - usually "\n" or "" (i.e. a terminating zero byte)
772 int nob = strlen(knl_buffer);
774 if (nob > usr_buffer_nob)
775 nob = usr_buffer_nob;
777 if (copy_to_user(usr_buffer, knl_buffer, nob))
780 if (append && nob < usr_buffer_nob) {
781 if (copy_to_user(usr_buffer + nob, append, 1))
789 EXPORT_SYMBOL(cfs_trace_copyout_string);
791 int cfs_trace_allocate_string_buffer(char **str, int nob)
793 if (nob > 2 * PAGE_CACHE_SIZE) /* string must be "sensible" */
796 *str = kmalloc(nob, GFP_KERNEL | __GFP_ZERO);
803 int cfs_trace_dump_debug_buffer_usrstr(void __user *usr_str, int usr_str_nob)
808 rc = cfs_trace_allocate_string_buffer(&str, usr_str_nob + 1);
812 rc = cfs_trace_copyin_string(str, usr_str_nob + 1,
813 usr_str, usr_str_nob);
821 rc = cfs_tracefile_dump_all_pages(str);
827 int cfs_trace_daemon_command(char *str)
831 cfs_tracefile_write_lock();
833 if (strcmp(str, "stop") == 0) {
834 cfs_tracefile_write_unlock();
835 cfs_trace_stop_thread();
836 cfs_tracefile_write_lock();
837 memset(cfs_tracefile, 0, sizeof(cfs_tracefile));
839 } else if (strncmp(str, "size=", 5) == 0) {
840 cfs_tracefile_size = simple_strtoul(str + 5, NULL, 0);
841 if (cfs_tracefile_size < 10 || cfs_tracefile_size > 20480)
842 cfs_tracefile_size = CFS_TRACEFILE_SIZE;
844 cfs_tracefile_size <<= 20;
846 } else if (strlen(str) >= sizeof(cfs_tracefile)) {
848 } else if (str[0] != '/') {
851 strcpy(cfs_tracefile, str);
854 "Lustre: debug daemon will attempt to start writing to %s (%lukB max)\n",
856 (long)(cfs_tracefile_size >> 10));
858 cfs_trace_start_thread();
861 cfs_tracefile_write_unlock();
865 int cfs_trace_daemon_command_usrstr(void __user *usr_str, int usr_str_nob)
870 rc = cfs_trace_allocate_string_buffer(&str, usr_str_nob + 1);
874 rc = cfs_trace_copyin_string(str, usr_str_nob + 1,
875 usr_str, usr_str_nob);
877 rc = cfs_trace_daemon_command(str);
883 int cfs_trace_set_debug_mb(int mb)
888 int limit = cfs_trace_max_debug_mb();
889 struct cfs_trace_cpu_data *tcd;
891 if (mb < num_possible_cpus()) {
893 "Lustre: %d MB is too small for debug buffer size, setting it to %d MB.\n",
894 mb, num_possible_cpus());
895 mb = num_possible_cpus();
900 "Lustre: %d MB is too large for debug buffer size, setting it to %d MB.\n",
905 mb /= num_possible_cpus();
906 pages = mb << (20 - PAGE_CACHE_SHIFT);
908 cfs_tracefile_write_lock();
910 cfs_tcd_for_each(tcd, i, j)
911 tcd->tcd_max_pages = (pages * tcd->tcd_pages_factor) / 100;
913 cfs_tracefile_write_unlock();
918 int cfs_trace_get_debug_mb(void)
922 struct cfs_trace_cpu_data *tcd;
925 cfs_tracefile_read_lock();
927 cfs_tcd_for_each(tcd, i, j)
928 total_pages += tcd->tcd_max_pages;
930 cfs_tracefile_read_unlock();
932 return (total_pages >> (20 - PAGE_CACHE_SHIFT)) + 1;
935 static int tracefiled(void *arg)
937 struct page_collection pc;
938 struct tracefiled_ctl *tctl = arg;
939 struct cfs_trace_page *tage;
940 struct cfs_trace_page *tmp;
948 /* we're started late enough that we pick up init's fs context */
949 /* this is so broken in uml? what on earth is going on? */
951 complete(&tctl->tctl_start);
956 pc.pc_want_daemon_pages = 0;
958 if (list_empty(&pc.pc_pages))
962 cfs_tracefile_read_lock();
963 if (cfs_tracefile[0] != 0) {
964 filp = filp_open(cfs_tracefile,
965 O_CREAT | O_RDWR | O_LARGEFILE,
970 printk(KERN_WARNING "couldn't open %s: %d\n",
974 cfs_tracefile_read_unlock();
976 put_pages_on_daemon_list(&pc);
977 __LASSERT(list_empty(&pc.pc_pages));
983 list_for_each_entry_safe(tage, tmp, &pc.pc_pages, linkage) {
986 __LASSERT_TAGE_INVARIANT(tage);
988 if (f_pos >= (off_t)cfs_tracefile_size)
990 else if (f_pos > i_size_read(file_inode(filp)))
991 f_pos = i_size_read(file_inode(filp));
993 buf = kmap(tage->page);
994 rc = vfs_write(filp, (__force const char __user *)buf,
998 if (rc != (int)tage->used) {
999 printk(KERN_WARNING "wanted to write %u but wrote %d\n",
1001 put_pages_back(&pc);
1002 __LASSERT(list_empty(&pc.pc_pages));
1008 filp_close(filp, NULL);
1009 put_pages_on_daemon_list(&pc);
1010 if (!list_empty(&pc.pc_pages)) {
1013 printk(KERN_ALERT "Lustre: trace pages aren't empty\n");
1014 pr_err("total cpus(%d): ", num_possible_cpus());
1015 for (i = 0; i < num_possible_cpus(); i++)
1017 pr_cont("%d(on) ", i);
1019 pr_cont("%d(off) ", i);
1023 list_for_each_entry_safe(tage, tmp, &pc.pc_pages,
1025 pr_err("page %d belongs to cpu %d\n",
1027 pr_err("There are %d pages unwritten\n", i);
1029 __LASSERT(list_empty(&pc.pc_pages));
1031 if (atomic_read(&tctl->tctl_shutdown)) {
1032 if (last_loop == 0) {
1039 init_waitqueue_entry(&__wait, current);
1040 add_wait_queue(&tctl->tctl_waitq, &__wait);
1041 set_current_state(TASK_INTERRUPTIBLE);
1042 schedule_timeout(cfs_time_seconds(1));
1043 remove_wait_queue(&tctl->tctl_waitq, &__wait);
1045 complete(&tctl->tctl_stop);
1049 int cfs_trace_start_thread(void)
1051 struct tracefiled_ctl *tctl = &trace_tctl;
1052 struct task_struct *task;
1055 mutex_lock(&cfs_trace_thread_mutex);
1059 init_completion(&tctl->tctl_start);
1060 init_completion(&tctl->tctl_stop);
1061 init_waitqueue_head(&tctl->tctl_waitq);
1062 atomic_set(&tctl->tctl_shutdown, 0);
1064 task = kthread_run(tracefiled, tctl, "ktracefiled");
1070 wait_for_completion(&tctl->tctl_start);
1073 mutex_unlock(&cfs_trace_thread_mutex);
1077 void cfs_trace_stop_thread(void)
1079 struct tracefiled_ctl *tctl = &trace_tctl;
1081 mutex_lock(&cfs_trace_thread_mutex);
1082 if (thread_running) {
1084 "Lustre: shutting down debug daemon thread...\n");
1085 atomic_set(&tctl->tctl_shutdown, 1);
1086 wait_for_completion(&tctl->tctl_stop);
1089 mutex_unlock(&cfs_trace_thread_mutex);
1092 int cfs_tracefile_init(int max_pages)
1094 struct cfs_trace_cpu_data *tcd;
1100 rc = cfs_tracefile_init_arch();
1104 cfs_tcd_for_each(tcd, i, j) {
1105 /* tcd_pages_factor is initialized int tracefile_init_arch. */
1106 factor = tcd->tcd_pages_factor;
1107 INIT_LIST_HEAD(&tcd->tcd_pages);
1108 INIT_LIST_HEAD(&tcd->tcd_stock_pages);
1109 INIT_LIST_HEAD(&tcd->tcd_daemon_pages);
1110 tcd->tcd_cur_pages = 0;
1111 tcd->tcd_cur_stock_pages = 0;
1112 tcd->tcd_cur_daemon_pages = 0;
1113 tcd->tcd_max_pages = (max_pages * factor) / 100;
1114 LASSERT(tcd->tcd_max_pages > 0);
1115 tcd->tcd_shutting_down = 0;
1121 static void trace_cleanup_on_all_cpus(void)
1123 struct cfs_trace_cpu_data *tcd;
1124 struct cfs_trace_page *tage;
1125 struct cfs_trace_page *tmp;
1128 for_each_possible_cpu(cpu) {
1129 cfs_tcd_for_each_type_lock(tcd, i, cpu) {
1130 tcd->tcd_shutting_down = 1;
1132 list_for_each_entry_safe(tage, tmp, &tcd->tcd_pages,
1134 __LASSERT_TAGE_INVARIANT(tage);
1136 list_del(&tage->linkage);
1137 cfs_tage_free(tage);
1140 tcd->tcd_cur_pages = 0;
1145 static void cfs_trace_cleanup(void)
1147 struct page_collection pc;
1149 INIT_LIST_HEAD(&pc.pc_pages);
1151 trace_cleanup_on_all_cpus();
1153 cfs_tracefile_fini_arch();
1156 void cfs_tracefile_exit(void)
1158 cfs_trace_stop_thread();
1159 cfs_trace_cleanup();