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.gnu.org/licenses/gpl-2.0.html
23 * Copyright (c) 2008, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2012, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
30 * Lustre is a trademark of Sun Microsystems, Inc.
32 * libcfs/libcfs/tracefile.c
34 * Author: Zach Brown <zab@clusterfs.com>
35 * Author: Phil Schwan <phil@clusterfs.com>
38 #define DEBUG_SUBSYSTEM S_LNET
39 #define LUSTRE_TRACEFILE_PRIVATE
40 #define pr_fmt(fmt) "Lustre: " fmt
41 #include "tracefile.h"
43 #include "../../include/linux/libcfs/libcfs.h"
45 /* XXX move things up to the top, comment */
46 union cfs_trace_data_union (*cfs_trace_data[TCD_MAX_TYPES])[NR_CPUS] __cacheline_aligned;
48 char cfs_tracefile[TRACEFILE_NAME_SIZE];
49 long long cfs_tracefile_size = CFS_TRACEFILE_SIZE;
50 static struct tracefiled_ctl trace_tctl;
51 static DEFINE_MUTEX(cfs_trace_thread_mutex);
52 static int thread_running;
54 static atomic_t cfs_tage_allocated = ATOMIC_INIT(0);
56 struct page_collection {
57 struct list_head pc_pages;
59 * if this flag is set, collect_pages() will spill both
60 * ->tcd_daemon_pages and ->tcd_pages to the ->pc_pages. Otherwise,
61 * only ->tcd_pages are spilled.
63 int pc_want_daemon_pages;
66 struct tracefiled_ctl {
67 struct completion tctl_start;
68 struct completion tctl_stop;
69 wait_queue_head_t tctl_waitq;
71 atomic_t tctl_shutdown;
75 * small data-structure for each page owned by tracefiled.
77 struct cfs_trace_page {
83 * linkage into one of the lists in trace_data_union or
86 struct list_head linkage;
88 * number of bytes used within this page
92 * cpu that owns this page
96 * type(context) of this page
101 static void put_pages_on_tcd_daemon_list(struct page_collection *pc,
102 struct cfs_trace_cpu_data *tcd);
104 static inline struct cfs_trace_page *
105 cfs_tage_from_list(struct list_head *list)
107 return list_entry(list, struct cfs_trace_page, linkage);
110 static struct cfs_trace_page *cfs_tage_alloc(gfp_t gfp)
113 struct cfs_trace_page *tage;
115 /* My caller is trying to free memory */
116 if (!in_interrupt() && memory_pressure_get())
120 * Don't spam console with allocation failures: they will be reported
121 * by upper layer anyway.
124 page = alloc_page(gfp);
128 tage = kmalloc(sizeof(*tage), gfp);
135 atomic_inc(&cfs_tage_allocated);
139 static void cfs_tage_free(struct cfs_trace_page *tage)
141 __free_page(tage->page);
143 atomic_dec(&cfs_tage_allocated);
146 static void cfs_tage_to_tail(struct cfs_trace_page *tage,
147 struct list_head *queue)
149 list_move_tail(&tage->linkage, queue);
152 int cfs_trace_refill_stock(struct cfs_trace_cpu_data *tcd, gfp_t gfp,
153 struct list_head *stock)
158 * XXX nikita: do NOT call portals_debug_msg() (CDEBUG/ENTRY/EXIT)
159 * from here: this will lead to infinite recursion.
162 for (i = 0; i + tcd->tcd_cur_stock_pages < TCD_STOCK_PAGES ; ++i) {
163 struct cfs_trace_page *tage;
165 tage = cfs_tage_alloc(gfp);
168 list_add_tail(&tage->linkage, stock);
173 /* return a page that has 'len' bytes left at the end */
174 static struct cfs_trace_page *
175 cfs_trace_get_tage_try(struct cfs_trace_cpu_data *tcd, unsigned long len)
177 struct cfs_trace_page *tage;
179 if (tcd->tcd_cur_pages > 0) {
180 __LASSERT(!list_empty(&tcd->tcd_pages));
181 tage = cfs_tage_from_list(tcd->tcd_pages.prev);
182 if (tage->used + len <= PAGE_SIZE)
186 if (tcd->tcd_cur_pages < tcd->tcd_max_pages) {
187 if (tcd->tcd_cur_stock_pages > 0) {
188 tage = cfs_tage_from_list(tcd->tcd_stock_pages.prev);
189 --tcd->tcd_cur_stock_pages;
190 list_del_init(&tage->linkage);
192 tage = cfs_tage_alloc(GFP_ATOMIC);
193 if (unlikely(!tage)) {
194 if ((!memory_pressure_get() ||
195 in_interrupt()) && printk_ratelimit())
196 pr_warn("cannot allocate a tage (%ld)\n",
203 tage->cpu = smp_processor_id();
204 tage->type = tcd->tcd_type;
205 list_add_tail(&tage->linkage, &tcd->tcd_pages);
206 tcd->tcd_cur_pages++;
208 if (tcd->tcd_cur_pages > 8 && thread_running) {
209 struct tracefiled_ctl *tctl = &trace_tctl;
211 * wake up tracefiled to process some pages.
213 wake_up(&tctl->tctl_waitq);
220 static void cfs_tcd_shrink(struct cfs_trace_cpu_data *tcd)
222 int pgcount = tcd->tcd_cur_pages / 10;
223 struct page_collection pc;
224 struct cfs_trace_page *tage;
225 struct cfs_trace_page *tmp;
228 * XXX nikita: do NOT call portals_debug_msg() (CDEBUG/ENTRY/EXIT)
229 * from here: this will lead to infinite recursion.
232 if (printk_ratelimit())
233 pr_warn("debug daemon buffer overflowed; discarding 10%% of pages (%d of %ld)\n",
234 pgcount + 1, tcd->tcd_cur_pages);
236 INIT_LIST_HEAD(&pc.pc_pages);
238 list_for_each_entry_safe(tage, tmp, &tcd->tcd_pages, linkage) {
242 list_move_tail(&tage->linkage, &pc.pc_pages);
243 tcd->tcd_cur_pages--;
245 put_pages_on_tcd_daemon_list(&pc, tcd);
248 /* return a page that has 'len' bytes left at the end */
249 static struct cfs_trace_page *cfs_trace_get_tage(struct cfs_trace_cpu_data *tcd,
252 struct cfs_trace_page *tage;
255 * XXX nikita: do NOT call portals_debug_msg() (CDEBUG/ENTRY/EXIT)
256 * from here: this will lead to infinite recursion.
259 if (len > PAGE_SIZE) {
260 pr_err("cowardly refusing to write %lu bytes in a page\n", len);
264 tage = cfs_trace_get_tage_try(tcd, len);
269 if (tcd->tcd_cur_pages > 0) {
270 tage = cfs_tage_from_list(tcd->tcd_pages.next);
272 cfs_tage_to_tail(tage, &tcd->tcd_pages);
277 int libcfs_debug_msg(struct libcfs_debug_msg_data *msgdata,
278 const char *format, ...)
283 va_start(args, format);
284 rc = libcfs_debug_vmsg2(msgdata, format, args, NULL);
289 EXPORT_SYMBOL(libcfs_debug_msg);
291 int libcfs_debug_vmsg2(struct libcfs_debug_msg_data *msgdata,
292 const char *format1, va_list args,
293 const char *format2, ...)
295 struct cfs_trace_cpu_data *tcd = NULL;
296 struct ptldebug_header header = { 0 };
297 struct cfs_trace_page *tage;
298 /* string_buf is used only if tcd != NULL, and is always set then */
299 char *string_buf = NULL;
302 int needed = 85; /* average message length */
308 int mask = msgdata->msg_mask;
309 const char *file = kbasename(msgdata->msg_file);
310 struct cfs_debug_limit_state *cdls = msgdata->msg_cdls;
312 tcd = cfs_trace_get_tcd();
314 /* cfs_trace_get_tcd() grabs a lock, which disables preemption and
315 * pins us to a particular CPU. This avoids an smp_processor_id()
316 * warning on Linux when debugging is enabled.
318 cfs_set_ptldebug_header(&header, msgdata, CDEBUG_STACK());
320 if (!tcd) /* arch may not log in IRQ context */
323 if (!tcd->tcd_cur_pages)
324 header.ph_flags |= PH_FLAG_FIRST_RECORD;
326 if (tcd->tcd_shutting_down) {
327 cfs_trace_put_tcd(tcd);
332 depth = __current_nesting_level();
333 known_size = strlen(file) + 1 + depth;
335 known_size += strlen(msgdata->msg_fn) + 1;
337 if (libcfs_debug_binary)
338 known_size += sizeof(header);
341 * '2' used because vsnprintf return real size required for output
342 * _without_ terminating NULL.
343 * if needed is to small for this format.
345 for (i = 0; i < 2; i++) {
346 tage = cfs_trace_get_tage(tcd, needed + known_size + 1);
348 if (needed + known_size > PAGE_SIZE)
351 cfs_trace_put_tcd(tcd);
356 string_buf = (char *)page_address(tage->page) +
357 tage->used + known_size;
359 max_nob = PAGE_SIZE - tage->used - known_size;
361 pr_emerg("negative max_nob: %d\n", max_nob);
363 cfs_trace_put_tcd(tcd);
371 needed = vsnprintf(string_buf, max_nob, format1, ap);
376 remain = max_nob - needed;
380 va_start(ap, format2);
381 needed += vsnprintf(string_buf + needed, remain,
386 if (needed < max_nob) /* well. printing ok.. */
390 if (*(string_buf + needed - 1) != '\n')
391 pr_info("format at %s:%d:%s doesn't end in newline\n", file,
392 msgdata->msg_line, msgdata->msg_fn);
394 header.ph_len = known_size + needed;
395 debug_buf = (char *)page_address(tage->page) + tage->used;
397 if (libcfs_debug_binary) {
398 memcpy(debug_buf, &header, sizeof(header));
399 tage->used += sizeof(header);
400 debug_buf += sizeof(header);
403 /* indent message according to the nesting level */
404 while (depth-- > 0) {
405 *(debug_buf++) = '.';
409 strcpy(debug_buf, file);
410 tage->used += strlen(file) + 1;
411 debug_buf += strlen(file) + 1;
413 if (msgdata->msg_fn) {
414 strcpy(debug_buf, msgdata->msg_fn);
415 tage->used += strlen(msgdata->msg_fn) + 1;
416 debug_buf += strlen(msgdata->msg_fn) + 1;
419 __LASSERT(debug_buf == string_buf);
421 tage->used += needed;
422 __LASSERT(tage->used <= PAGE_SIZE);
425 if (!(mask & libcfs_printk)) {
426 /* no console output requested */
428 cfs_trace_put_tcd(tcd);
433 if (libcfs_console_ratelimit &&
434 cdls->cdls_next && /* not first time ever */
435 !cfs_time_after(cfs_time_current(), cdls->cdls_next)) {
436 /* skipping a console message */
439 cfs_trace_put_tcd(tcd);
443 if (cfs_time_after(cfs_time_current(),
444 cdls->cdls_next + libcfs_console_max_delay +
445 cfs_time_seconds(10))) {
446 /* last timeout was a long time ago */
447 cdls->cdls_delay /= libcfs_console_backoff * 4;
449 cdls->cdls_delay *= libcfs_console_backoff;
452 if (cdls->cdls_delay < libcfs_console_min_delay)
453 cdls->cdls_delay = libcfs_console_min_delay;
454 else if (cdls->cdls_delay > libcfs_console_max_delay)
455 cdls->cdls_delay = libcfs_console_max_delay;
457 /* ensure cdls_next is never zero after it's been seen */
458 cdls->cdls_next = (cfs_time_current() + cdls->cdls_delay) | 1;
462 cfs_print_to_console(&header, mask, string_buf, needed, file,
464 cfs_trace_put_tcd(tcd);
466 string_buf = cfs_trace_get_console_buffer();
471 needed = vsnprintf(string_buf,
472 CFS_TRACE_CONSOLE_BUFFER_SIZE,
477 remain = CFS_TRACE_CONSOLE_BUFFER_SIZE - needed;
479 va_start(ap, format2);
480 needed += vsnprintf(string_buf + needed, remain,
485 cfs_print_to_console(&header, mask,
486 string_buf, needed, file, msgdata->msg_fn);
491 if (cdls && cdls->cdls_count) {
492 string_buf = cfs_trace_get_console_buffer();
494 needed = snprintf(string_buf, CFS_TRACE_CONSOLE_BUFFER_SIZE,
495 "Skipped %d previous similar message%s\n",
497 (cdls->cdls_count > 1) ? "s" : "");
499 cfs_print_to_console(&header, mask,
500 string_buf, needed, file, msgdata->msg_fn);
503 cdls->cdls_count = 0;
508 EXPORT_SYMBOL(libcfs_debug_vmsg2);
511 cfs_trace_assertion_failed(const char *str,
512 struct libcfs_debug_msg_data *msgdata)
514 struct ptldebug_header hdr;
516 libcfs_panic_in_progress = 1;
517 libcfs_catastrophe = 1;
520 cfs_set_ptldebug_header(&hdr, msgdata, CDEBUG_STACK());
522 cfs_print_to_console(&hdr, D_EMERG, str, strlen(str),
523 msgdata->msg_file, msgdata->msg_fn);
525 panic("Lustre debug assertion failure\n");
531 panic_collect_pages(struct page_collection *pc)
533 /* Do the collect_pages job on a single CPU: assumes that all other
534 * CPUs have been stopped during a panic. If this isn't true for some
535 * arch, this will have to be implemented separately in each arch.
537 struct cfs_trace_cpu_data *tcd;
541 INIT_LIST_HEAD(&pc->pc_pages);
543 cfs_tcd_for_each(tcd, i, j) {
544 list_splice_init(&tcd->tcd_pages, &pc->pc_pages);
545 tcd->tcd_cur_pages = 0;
547 if (pc->pc_want_daemon_pages) {
548 list_splice_init(&tcd->tcd_daemon_pages, &pc->pc_pages);
549 tcd->tcd_cur_daemon_pages = 0;
554 static void collect_pages_on_all_cpus(struct page_collection *pc)
556 struct cfs_trace_cpu_data *tcd;
559 for_each_possible_cpu(cpu) {
560 cfs_tcd_for_each_type_lock(tcd, i, cpu) {
561 list_splice_init(&tcd->tcd_pages, &pc->pc_pages);
562 tcd->tcd_cur_pages = 0;
563 if (pc->pc_want_daemon_pages) {
564 list_splice_init(&tcd->tcd_daemon_pages,
566 tcd->tcd_cur_daemon_pages = 0;
572 static void collect_pages(struct page_collection *pc)
574 INIT_LIST_HEAD(&pc->pc_pages);
576 if (libcfs_panic_in_progress)
577 panic_collect_pages(pc);
579 collect_pages_on_all_cpus(pc);
582 static void put_pages_back_on_all_cpus(struct page_collection *pc)
584 struct cfs_trace_cpu_data *tcd;
585 struct list_head *cur_head;
586 struct cfs_trace_page *tage;
587 struct cfs_trace_page *tmp;
590 for_each_possible_cpu(cpu) {
591 cfs_tcd_for_each_type_lock(tcd, i, cpu) {
592 cur_head = tcd->tcd_pages.next;
594 list_for_each_entry_safe(tage, tmp, &pc->pc_pages,
596 __LASSERT_TAGE_INVARIANT(tage);
598 if (tage->cpu != cpu || tage->type != i)
601 cfs_tage_to_tail(tage, cur_head);
602 tcd->tcd_cur_pages++;
608 static void put_pages_back(struct page_collection *pc)
610 if (!libcfs_panic_in_progress)
611 put_pages_back_on_all_cpus(pc);
614 /* Add pages to a per-cpu debug daemon ringbuffer. This buffer makes sure that
615 * we have a good amount of data at all times for dumping during an LBUG, even
616 * if we have been steadily writing (and otherwise discarding) pages via the
619 static void put_pages_on_tcd_daemon_list(struct page_collection *pc,
620 struct cfs_trace_cpu_data *tcd)
622 struct cfs_trace_page *tage;
623 struct cfs_trace_page *tmp;
625 list_for_each_entry_safe(tage, tmp, &pc->pc_pages, linkage) {
626 __LASSERT_TAGE_INVARIANT(tage);
628 if (tage->cpu != tcd->tcd_cpu || tage->type != tcd->tcd_type)
631 cfs_tage_to_tail(tage, &tcd->tcd_daemon_pages);
632 tcd->tcd_cur_daemon_pages++;
634 if (tcd->tcd_cur_daemon_pages > tcd->tcd_max_pages) {
635 struct cfs_trace_page *victim;
637 __LASSERT(!list_empty(&tcd->tcd_daemon_pages));
638 victim = cfs_tage_from_list(tcd->tcd_daemon_pages.next);
640 __LASSERT_TAGE_INVARIANT(victim);
642 list_del(&victim->linkage);
643 cfs_tage_free(victim);
644 tcd->tcd_cur_daemon_pages--;
649 static void put_pages_on_daemon_list(struct page_collection *pc)
651 struct cfs_trace_cpu_data *tcd;
654 for_each_possible_cpu(cpu) {
655 cfs_tcd_for_each_type_lock(tcd, i, cpu)
656 put_pages_on_tcd_daemon_list(pc, tcd);
660 void cfs_trace_debug_print(void)
662 struct page_collection pc;
663 struct cfs_trace_page *tage;
664 struct cfs_trace_page *tmp;
666 pc.pc_want_daemon_pages = 1;
668 list_for_each_entry_safe(tage, tmp, &pc.pc_pages, linkage) {
672 __LASSERT_TAGE_INVARIANT(tage);
675 p = page_address(page);
676 while (p < ((char *)page_address(page) + tage->used)) {
677 struct ptldebug_header *hdr;
683 p += strlen(file) + 1;
686 len = hdr->ph_len - (int)(p - (char *)hdr);
688 cfs_print_to_console(hdr, D_EMERG, p, len, file, fn);
693 list_del(&tage->linkage);
698 int cfs_tracefile_dump_all_pages(char *filename)
700 struct page_collection pc;
702 struct cfs_trace_page *tage;
703 struct cfs_trace_page *tmp;
705 mm_segment_t __oldfs;
708 cfs_tracefile_write_lock();
710 filp = filp_open(filename, O_CREAT | O_EXCL | O_WRONLY | O_LARGEFILE,
715 pr_err("LustreError: can't open %s for dump: rc %d\n",
720 pc.pc_want_daemon_pages = 1;
722 if (list_empty(&pc.pc_pages)) {
729 /* ok, for now, just write the pages. in the future we'll be building
730 * iobufs with the pages and calling generic_direct_IO
732 list_for_each_entry_safe(tage, tmp, &pc.pc_pages, linkage) {
733 __LASSERT_TAGE_INVARIANT(tage);
735 buf = kmap(tage->page);
736 rc = vfs_write(filp, (__force const char __user *)buf,
737 tage->used, &filp->f_pos);
740 if (rc != (int)tage->used) {
741 pr_warn("wanted to write %u but wrote %d\n", tage->used,
744 __LASSERT(list_empty(&pc.pc_pages));
747 list_del(&tage->linkage);
751 rc = vfs_fsync(filp, 1);
753 pr_err("sync returns %d\n", rc);
755 filp_close(filp, NULL);
757 cfs_tracefile_write_unlock();
761 void cfs_trace_flush_pages(void)
763 struct page_collection pc;
764 struct cfs_trace_page *tage;
765 struct cfs_trace_page *tmp;
767 pc.pc_want_daemon_pages = 1;
769 list_for_each_entry_safe(tage, tmp, &pc.pc_pages, linkage) {
770 __LASSERT_TAGE_INVARIANT(tage);
772 list_del(&tage->linkage);
777 int cfs_trace_copyin_string(char *knl_buffer, int knl_buffer_nob,
778 const char __user *usr_buffer, int usr_buffer_nob)
782 if (usr_buffer_nob > knl_buffer_nob)
785 if (copy_from_user((void *)knl_buffer,
786 usr_buffer, usr_buffer_nob))
789 nob = strnlen(knl_buffer, usr_buffer_nob);
790 while (nob-- >= 0) /* strip trailing whitespace */
791 if (!isspace(knl_buffer[nob]))
794 if (nob < 0) /* empty string */
797 if (nob == knl_buffer_nob) /* no space to terminate */
800 knl_buffer[nob + 1] = 0; /* terminate */
803 EXPORT_SYMBOL(cfs_trace_copyin_string);
805 int cfs_trace_copyout_string(char __user *usr_buffer, int usr_buffer_nob,
806 const char *knl_buffer, char *append)
809 * NB if 'append' != NULL, it's a single character to append to the
810 * copied out string - usually "\n" or "" (i.e. a terminating zero byte)
812 int nob = strlen(knl_buffer);
814 if (nob > usr_buffer_nob)
815 nob = usr_buffer_nob;
817 if (copy_to_user(usr_buffer, knl_buffer, nob))
820 if (append && nob < usr_buffer_nob) {
821 if (copy_to_user(usr_buffer + nob, append, 1))
829 EXPORT_SYMBOL(cfs_trace_copyout_string);
831 int cfs_trace_allocate_string_buffer(char **str, int nob)
833 if (nob > 2 * PAGE_SIZE) /* string must be "sensible" */
836 *str = kmalloc(nob, GFP_KERNEL | __GFP_ZERO);
843 int cfs_trace_dump_debug_buffer_usrstr(void __user *usr_str, int usr_str_nob)
848 rc = cfs_trace_allocate_string_buffer(&str, usr_str_nob + 1);
852 rc = cfs_trace_copyin_string(str, usr_str_nob + 1,
853 usr_str, usr_str_nob);
861 rc = cfs_tracefile_dump_all_pages(str);
867 int cfs_trace_daemon_command(char *str)
871 cfs_tracefile_write_lock();
873 if (!strcmp(str, "stop")) {
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)) {
882 rc = kstrtoul(str + 5, 10, &tmp);
884 if (tmp < 10 || tmp > 20480)
885 cfs_tracefile_size = CFS_TRACEFILE_SIZE;
887 cfs_tracefile_size = tmp << 20;
889 } else if (strlen(str) >= sizeof(cfs_tracefile)) {
891 } else if (str[0] != '/') {
894 strcpy(cfs_tracefile, str);
896 pr_info("debug daemon will attempt to start writing to %s (%lukB max)\n",
898 (long)(cfs_tracefile_size >> 10));
900 cfs_trace_start_thread();
903 cfs_tracefile_write_unlock();
907 int cfs_trace_daemon_command_usrstr(void __user *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);
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 < num_possible_cpus()) {
934 pr_warn("%d MB is too small for debug buffer size, setting it to %d MB.\n",
935 mb, num_possible_cpus());
936 mb = num_possible_cpus();
940 pr_warn("%d MB is too large for debug buffer size, setting it to %d MB.\n",
945 mb /= num_possible_cpus();
946 pages = mb << (20 - 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_get_debug_mb(void)
962 struct cfs_trace_cpu_data *tcd;
965 cfs_tracefile_read_lock();
967 cfs_tcd_for_each(tcd, i, j)
968 total_pages += tcd->tcd_max_pages;
970 cfs_tracefile_read_unlock();
972 return (total_pages >> (20 - PAGE_SHIFT)) + 1;
975 static int tracefiled(void *arg)
977 struct page_collection pc;
978 struct tracefiled_ctl *tctl = arg;
979 struct cfs_trace_page *tage;
980 struct cfs_trace_page *tmp;
981 mm_segment_t __oldfs;
987 /* we're started late enough that we pick up init's fs context */
988 /* this is so broken in uml? what on earth is going on? */
990 complete(&tctl->tctl_start);
993 wait_queue_entry_t __wait;
995 pc.pc_want_daemon_pages = 0;
997 if (list_empty(&pc.pc_pages))
1001 cfs_tracefile_read_lock();
1002 if (cfs_tracefile[0]) {
1003 filp = filp_open(cfs_tracefile,
1004 O_CREAT | O_RDWR | O_LARGEFILE,
1009 pr_warn("couldn't open %s: %d\n", cfs_tracefile,
1013 cfs_tracefile_read_unlock();
1015 put_pages_on_daemon_list(&pc);
1016 __LASSERT(list_empty(&pc.pc_pages));
1022 list_for_each_entry_safe(tage, tmp, &pc.pc_pages, linkage) {
1023 static loff_t f_pos;
1025 __LASSERT_TAGE_INVARIANT(tage);
1027 if (f_pos >= (off_t)cfs_tracefile_size)
1029 else if (f_pos > i_size_read(file_inode(filp)))
1030 f_pos = i_size_read(file_inode(filp));
1032 buf = kmap(tage->page);
1033 rc = vfs_write(filp, (__force const char __user *)buf,
1034 tage->used, &f_pos);
1037 if (rc != (int)tage->used) {
1038 pr_warn("wanted to write %u but wrote %d\n",
1040 put_pages_back(&pc);
1041 __LASSERT(list_empty(&pc.pc_pages));
1047 filp_close(filp, NULL);
1048 put_pages_on_daemon_list(&pc);
1049 if (!list_empty(&pc.pc_pages)) {
1052 pr_alert("trace pages aren't empty\n");
1053 pr_err("total cpus(%d): ", num_possible_cpus());
1054 for (i = 0; i < num_possible_cpus(); i++)
1056 pr_cont("%d(on) ", i);
1058 pr_cont("%d(off) ", i);
1062 list_for_each_entry_safe(tage, tmp, &pc.pc_pages,
1064 pr_err("page %d belongs to cpu %d\n",
1066 pr_err("There are %d pages unwritten\n", i);
1068 __LASSERT(list_empty(&pc.pc_pages));
1070 if (atomic_read(&tctl->tctl_shutdown)) {
1078 init_waitqueue_entry(&__wait, current);
1079 add_wait_queue(&tctl->tctl_waitq, &__wait);
1080 set_current_state(TASK_INTERRUPTIBLE);
1081 schedule_timeout(cfs_time_seconds(1));
1082 remove_wait_queue(&tctl->tctl_waitq, &__wait);
1084 complete(&tctl->tctl_stop);
1088 int cfs_trace_start_thread(void)
1090 struct tracefiled_ctl *tctl = &trace_tctl;
1091 struct task_struct *task;
1094 mutex_lock(&cfs_trace_thread_mutex);
1098 init_completion(&tctl->tctl_start);
1099 init_completion(&tctl->tctl_stop);
1100 init_waitqueue_head(&tctl->tctl_waitq);
1101 atomic_set(&tctl->tctl_shutdown, 0);
1103 task = kthread_run(tracefiled, tctl, "ktracefiled");
1109 wait_for_completion(&tctl->tctl_start);
1112 mutex_unlock(&cfs_trace_thread_mutex);
1116 void cfs_trace_stop_thread(void)
1118 struct tracefiled_ctl *tctl = &trace_tctl;
1120 mutex_lock(&cfs_trace_thread_mutex);
1121 if (thread_running) {
1122 pr_info("shutting down debug daemon thread...\n");
1123 atomic_set(&tctl->tctl_shutdown, 1);
1124 wait_for_completion(&tctl->tctl_stop);
1127 mutex_unlock(&cfs_trace_thread_mutex);
1130 int cfs_tracefile_init(int max_pages)
1132 struct cfs_trace_cpu_data *tcd;
1138 rc = cfs_tracefile_init_arch();
1142 cfs_tcd_for_each(tcd, i, j) {
1143 /* tcd_pages_factor is initialized int tracefile_init_arch. */
1144 factor = tcd->tcd_pages_factor;
1145 INIT_LIST_HEAD(&tcd->tcd_pages);
1146 INIT_LIST_HEAD(&tcd->tcd_stock_pages);
1147 INIT_LIST_HEAD(&tcd->tcd_daemon_pages);
1148 tcd->tcd_cur_pages = 0;
1149 tcd->tcd_cur_stock_pages = 0;
1150 tcd->tcd_cur_daemon_pages = 0;
1151 tcd->tcd_max_pages = (max_pages * factor) / 100;
1152 LASSERT(tcd->tcd_max_pages > 0);
1153 tcd->tcd_shutting_down = 0;
1159 static void trace_cleanup_on_all_cpus(void)
1161 struct cfs_trace_cpu_data *tcd;
1162 struct cfs_trace_page *tage;
1163 struct cfs_trace_page *tmp;
1166 for_each_possible_cpu(cpu) {
1167 cfs_tcd_for_each_type_lock(tcd, i, cpu) {
1168 tcd->tcd_shutting_down = 1;
1170 list_for_each_entry_safe(tage, tmp, &tcd->tcd_pages,
1172 __LASSERT_TAGE_INVARIANT(tage);
1174 list_del(&tage->linkage);
1175 cfs_tage_free(tage);
1178 tcd->tcd_cur_pages = 0;
1183 static void cfs_trace_cleanup(void)
1185 struct page_collection pc;
1187 INIT_LIST_HEAD(&pc.pc_pages);
1189 trace_cleanup_on_all_cpus();
1191 cfs_tracefile_fini_arch();
1194 void cfs_tracefile_exit(void)
1196 cfs_trace_stop_thread();
1197 cfs_trace_cleanup();