]>
Commit | Line | Data |
---|---|---|
d7e09d03 PT |
1 | /* |
2 | * GPL HEADER START | |
3 | * | |
4 | * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. | |
5 | * | |
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. | |
9 | * | |
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). | |
15 | * | |
16 | * You should have received a copy of the GNU General Public License | |
17 | * version 2 along with this program; If not, see | |
6a5b99a4 | 18 | * http://www.gnu.org/licenses/gpl-2.0.html |
d7e09d03 | 19 | * |
d7e09d03 PT |
20 | * GPL HEADER END |
21 | */ | |
22 | /* | |
23 | * Copyright (c) 2008, 2010, Oracle and/or its affiliates. All rights reserved. | |
24 | * Use is subject to license terms. | |
25 | * | |
26 | * Copyright (c) 2012, Intel Corporation. | |
27 | */ | |
28 | /* | |
29 | * This file is part of Lustre, http://www.lustre.org/ | |
30 | * Lustre is a trademark of Sun Microsystems, Inc. | |
31 | * | |
32 | * libcfs/libcfs/tracefile.c | |
33 | * | |
34 | * Author: Zach Brown <zab@clusterfs.com> | |
35 | * Author: Phil Schwan <phil@clusterfs.com> | |
36 | */ | |
37 | ||
d7e09d03 PT |
38 | #define DEBUG_SUBSYSTEM S_LNET |
39 | #define LUSTRE_TRACEFILE_PRIVATE | |
217d31c4 | 40 | #define pr_fmt(fmt) "Lustre: " fmt |
d7e09d03 PT |
41 | #include "tracefile.h" |
42 | ||
9fdaf8c0 | 43 | #include "../../include/linux/libcfs/libcfs.h" |
d7e09d03 PT |
44 | |
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; | |
47 | ||
48 | char cfs_tracefile[TRACEFILE_NAME_SIZE]; | |
49 | long long cfs_tracefile_size = CFS_TRACEFILE_SIZE; | |
50 | static struct tracefiled_ctl trace_tctl; | |
2f4246f7 | 51 | static DEFINE_MUTEX(cfs_trace_thread_mutex); |
225f597c | 52 | static int thread_running; |
d7e09d03 | 53 | |
2a74b9bd | 54 | static atomic_t cfs_tage_allocated = ATOMIC_INIT(0); |
d7e09d03 | 55 | |
69c7c854 OD |
56 | struct page_collection { |
57 | struct list_head pc_pages; | |
58 | /* | |
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. | |
62 | */ | |
715475ae | 63 | int pc_want_daemon_pages; |
69c7c854 OD |
64 | }; |
65 | ||
66 | struct tracefiled_ctl { | |
67 | struct completion tctl_start; | |
68 | struct completion tctl_stop; | |
715475ae | 69 | wait_queue_head_t tctl_waitq; |
69c7c854 OD |
70 | pid_t tctl_pid; |
71 | atomic_t tctl_shutdown; | |
72 | }; | |
73 | ||
74 | /* | |
75 | * small data-structure for each page owned by tracefiled. | |
76 | */ | |
77 | struct cfs_trace_page { | |
78 | /* | |
79 | * page itself | |
80 | */ | |
715475ae | 81 | struct page *page; |
69c7c854 OD |
82 | /* |
83 | * linkage into one of the lists in trace_data_union or | |
84 | * page_collection | |
85 | */ | |
715475ae | 86 | struct list_head linkage; |
69c7c854 OD |
87 | /* |
88 | * number of bytes used within this page | |
89 | */ | |
715475ae | 90 | unsigned int used; |
69c7c854 OD |
91 | /* |
92 | * cpu that owns this page | |
93 | */ | |
715475ae | 94 | unsigned short cpu; |
69c7c854 OD |
95 | /* |
96 | * type(context) of this page | |
97 | */ | |
715475ae | 98 | unsigned short type; |
69c7c854 OD |
99 | }; |
100 | ||
d7e09d03 PT |
101 | static void put_pages_on_tcd_daemon_list(struct page_collection *pc, |
102 | struct cfs_trace_cpu_data *tcd); | |
103 | ||
104 | static inline struct cfs_trace_page * | |
105 | cfs_tage_from_list(struct list_head *list) | |
106 | { | |
107 | return list_entry(list, struct cfs_trace_page, linkage); | |
108 | } | |
109 | ||
cad7aa13 | 110 | static struct cfs_trace_page *cfs_tage_alloc(gfp_t gfp) |
d7e09d03 | 111 | { |
715475ae | 112 | struct page *page; |
d7e09d03 PT |
113 | struct cfs_trace_page *tage; |
114 | ||
115 | /* My caller is trying to free memory */ | |
116 | if (!in_interrupt() && memory_pressure_get()) | |
117 | return NULL; | |
118 | ||
119 | /* | |
120 | * Don't spam console with allocation failures: they will be reported | |
121 | * by upper layer anyway. | |
122 | */ | |
123 | gfp |= __GFP_NOWARN; | |
124 | page = alloc_page(gfp); | |
15d9f520 | 125 | if (!page) |
d7e09d03 PT |
126 | return NULL; |
127 | ||
128 | tage = kmalloc(sizeof(*tage), gfp); | |
15d9f520 | 129 | if (!tage) { |
d7e09d03 PT |
130 | __free_page(page); |
131 | return NULL; | |
132 | } | |
133 | ||
134 | tage->page = page; | |
135 | atomic_inc(&cfs_tage_allocated); | |
136 | return tage; | |
137 | } | |
138 | ||
139 | static void cfs_tage_free(struct cfs_trace_page *tage) | |
140 | { | |
d7e09d03 PT |
141 | __free_page(tage->page); |
142 | kfree(tage); | |
143 | atomic_dec(&cfs_tage_allocated); | |
144 | } | |
145 | ||
146 | static void cfs_tage_to_tail(struct cfs_trace_page *tage, | |
147 | struct list_head *queue) | |
148 | { | |
d7e09d03 PT |
149 | list_move_tail(&tage->linkage, queue); |
150 | } | |
151 | ||
cad7aa13 | 152 | int cfs_trace_refill_stock(struct cfs_trace_cpu_data *tcd, gfp_t gfp, |
d7e09d03 PT |
153 | struct list_head *stock) |
154 | { | |
155 | int i; | |
156 | ||
157 | /* | |
158 | * XXX nikita: do NOT call portals_debug_msg() (CDEBUG/ENTRY/EXIT) | |
159 | * from here: this will lead to infinite recursion. | |
160 | */ | |
161 | ||
b2952d62 | 162 | for (i = 0; i + tcd->tcd_cur_stock_pages < TCD_STOCK_PAGES ; ++i) { |
d7e09d03 PT |
163 | struct cfs_trace_page *tage; |
164 | ||
165 | tage = cfs_tage_alloc(gfp); | |
15d9f520 | 166 | if (!tage) |
d7e09d03 PT |
167 | break; |
168 | list_add_tail(&tage->linkage, stock); | |
169 | } | |
170 | return i; | |
171 | } | |
172 | ||
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) | |
176 | { | |
177 | struct cfs_trace_page *tage; | |
178 | ||
179 | if (tcd->tcd_cur_pages > 0) { | |
180 | __LASSERT(!list_empty(&tcd->tcd_pages)); | |
181 | tage = cfs_tage_from_list(tcd->tcd_pages.prev); | |
09cbfeaf | 182 | if (tage->used + len <= PAGE_SIZE) |
d7e09d03 PT |
183 | return tage; |
184 | } | |
185 | ||
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); | |
191 | } else { | |
192 | tage = cfs_tage_alloc(GFP_ATOMIC); | |
15d9f520 | 193 | if (unlikely(!tage)) { |
93e7ea8c KM |
194 | if (!memory_pressure_get() || in_interrupt()) |
195 | pr_warn_ratelimited("cannot allocate a tage (%ld)\n", | |
196 | tcd->tcd_cur_pages); | |
d7e09d03 PT |
197 | return NULL; |
198 | } | |
199 | } | |
200 | ||
201 | tage->used = 0; | |
202 | tage->cpu = smp_processor_id(); | |
203 | tage->type = tcd->tcd_type; | |
204 | list_add_tail(&tage->linkage, &tcd->tcd_pages); | |
205 | tcd->tcd_cur_pages++; | |
206 | ||
207 | if (tcd->tcd_cur_pages > 8 && thread_running) { | |
208 | struct tracefiled_ctl *tctl = &trace_tctl; | |
209 | /* | |
210 | * wake up tracefiled to process some pages. | |
211 | */ | |
212 | wake_up(&tctl->tctl_waitq); | |
213 | } | |
214 | return tage; | |
215 | } | |
216 | return NULL; | |
217 | } | |
218 | ||
219 | static void cfs_tcd_shrink(struct cfs_trace_cpu_data *tcd) | |
220 | { | |
221 | int pgcount = tcd->tcd_cur_pages / 10; | |
222 | struct page_collection pc; | |
223 | struct cfs_trace_page *tage; | |
224 | struct cfs_trace_page *tmp; | |
225 | ||
226 | /* | |
227 | * XXX nikita: do NOT call portals_debug_msg() (CDEBUG/ENTRY/EXIT) | |
228 | * from here: this will lead to infinite recursion. | |
229 | */ | |
230 | ||
93e7ea8c KM |
231 | pr_warn_ratelimited("debug daemon buffer overflowed; discarding 10%% of pages (%d of %ld)\n", |
232 | pgcount + 1, tcd->tcd_cur_pages); | |
d7e09d03 PT |
233 | |
234 | INIT_LIST_HEAD(&pc.pc_pages); | |
d7e09d03 PT |
235 | |
236 | list_for_each_entry_safe(tage, tmp, &tcd->tcd_pages, linkage) { | |
6fc68ea5 | 237 | if (!pgcount--) |
d7e09d03 PT |
238 | break; |
239 | ||
240 | list_move_tail(&tage->linkage, &pc.pc_pages); | |
241 | tcd->tcd_cur_pages--; | |
242 | } | |
243 | put_pages_on_tcd_daemon_list(&pc, tcd); | |
244 | } | |
245 | ||
246 | /* return a page that has 'len' bytes left at the end */ | |
247 | static struct cfs_trace_page *cfs_trace_get_tage(struct cfs_trace_cpu_data *tcd, | |
248 | unsigned long len) | |
249 | { | |
250 | struct cfs_trace_page *tage; | |
251 | ||
252 | /* | |
253 | * XXX nikita: do NOT call portals_debug_msg() (CDEBUG/ENTRY/EXIT) | |
254 | * from here: this will lead to infinite recursion. | |
255 | */ | |
256 | ||
09cbfeaf | 257 | if (len > PAGE_SIZE) { |
324588ec | 258 | pr_err("cowardly refusing to write %lu bytes in a page\n", len); |
d7e09d03 PT |
259 | return NULL; |
260 | } | |
261 | ||
262 | tage = cfs_trace_get_tage_try(tcd, len); | |
15d9f520 | 263 | if (tage) |
d7e09d03 PT |
264 | return tage; |
265 | if (thread_running) | |
266 | cfs_tcd_shrink(tcd); | |
267 | if (tcd->tcd_cur_pages > 0) { | |
268 | tage = cfs_tage_from_list(tcd->tcd_pages.next); | |
269 | tage->used = 0; | |
270 | cfs_tage_to_tail(tage, &tcd->tcd_pages); | |
271 | } | |
272 | return tage; | |
273 | } | |
274 | ||
275 | int libcfs_debug_msg(struct libcfs_debug_msg_data *msgdata, | |
276 | const char *format, ...) | |
277 | { | |
278 | va_list args; | |
715475ae | 279 | int rc; |
d7e09d03 PT |
280 | |
281 | va_start(args, format); | |
282 | rc = libcfs_debug_vmsg2(msgdata, format, args, NULL); | |
283 | va_end(args); | |
284 | ||
285 | return rc; | |
286 | } | |
287 | EXPORT_SYMBOL(libcfs_debug_msg); | |
288 | ||
289 | int libcfs_debug_vmsg2(struct libcfs_debug_msg_data *msgdata, | |
290 | const char *format1, va_list args, | |
291 | const char *format2, ...) | |
292 | { | |
293 | struct cfs_trace_cpu_data *tcd = NULL; | |
715475ae JS |
294 | struct ptldebug_header header = { 0 }; |
295 | struct cfs_trace_page *tage; | |
d7e09d03 | 296 | /* string_buf is used only if tcd != NULL, and is always set then */ |
715475ae JS |
297 | char *string_buf = NULL; |
298 | char *debug_buf; | |
299 | int known_size; | |
300 | int needed = 85; /* average message length */ | |
301 | int max_nob; | |
302 | va_list ap; | |
303 | int depth; | |
304 | int i; | |
305 | int remain; | |
306 | int mask = msgdata->msg_mask; | |
307 | const char *file = kbasename(msgdata->msg_file); | |
308 | struct cfs_debug_limit_state *cdls = msgdata->msg_cdls; | |
d7e09d03 | 309 | |
d7e09d03 PT |
310 | tcd = cfs_trace_get_tcd(); |
311 | ||
312 | /* cfs_trace_get_tcd() grabs a lock, which disables preemption and | |
313 | * pins us to a particular CPU. This avoids an smp_processor_id() | |
a3fbcb3c OD |
314 | * warning on Linux when debugging is enabled. |
315 | */ | |
d7e09d03 PT |
316 | cfs_set_ptldebug_header(&header, msgdata, CDEBUG_STACK()); |
317 | ||
15d9f520 | 318 | if (!tcd) /* arch may not log in IRQ context */ |
d7e09d03 PT |
319 | goto console; |
320 | ||
6fc68ea5 | 321 | if (!tcd->tcd_cur_pages) |
d7e09d03 PT |
322 | header.ph_flags |= PH_FLAG_FIRST_RECORD; |
323 | ||
324 | if (tcd->tcd_shutting_down) { | |
325 | cfs_trace_put_tcd(tcd); | |
326 | tcd = NULL; | |
327 | goto console; | |
328 | } | |
329 | ||
330 | depth = __current_nesting_level(); | |
331 | known_size = strlen(file) + 1 + depth; | |
332 | if (msgdata->msg_fn) | |
333 | known_size += strlen(msgdata->msg_fn) + 1; | |
334 | ||
335 | if (libcfs_debug_binary) | |
336 | known_size += sizeof(header); | |
337 | ||
a3fbcb3c | 338 | /* |
d7e09d03 PT |
339 | * '2' used because vsnprintf return real size required for output |
340 | * _without_ terminating NULL. | |
341 | * if needed is to small for this format. | |
342 | */ | |
343 | for (i = 0; i < 2; i++) { | |
344 | tage = cfs_trace_get_tage(tcd, needed + known_size + 1); | |
15d9f520 | 345 | if (!tage) { |
09cbfeaf | 346 | if (needed + known_size > PAGE_SIZE) |
d7e09d03 PT |
347 | mask |= D_ERROR; |
348 | ||
349 | cfs_trace_put_tcd(tcd); | |
350 | tcd = NULL; | |
351 | goto console; | |
352 | } | |
353 | ||
354 | string_buf = (char *)page_address(tage->page) + | |
355 | tage->used + known_size; | |
356 | ||
09cbfeaf | 357 | max_nob = PAGE_SIZE - tage->used - known_size; |
d7e09d03 | 358 | if (max_nob <= 0) { |
f7b24d53 | 359 | pr_emerg("negative max_nob: %d\n", max_nob); |
d7e09d03 PT |
360 | mask |= D_ERROR; |
361 | cfs_trace_put_tcd(tcd); | |
362 | tcd = NULL; | |
363 | goto console; | |
364 | } | |
365 | ||
366 | needed = 0; | |
367 | if (format1) { | |
368 | va_copy(ap, args); | |
369 | needed = vsnprintf(string_buf, max_nob, format1, ap); | |
370 | va_end(ap); | |
371 | } | |
372 | ||
373 | if (format2) { | |
374 | remain = max_nob - needed; | |
375 | if (remain < 0) | |
376 | remain = 0; | |
377 | ||
378 | va_start(ap, format2); | |
379 | needed += vsnprintf(string_buf + needed, remain, | |
380 | format2, ap); | |
381 | va_end(ap); | |
382 | } | |
383 | ||
384 | if (needed < max_nob) /* well. printing ok.. */ | |
385 | break; | |
386 | } | |
387 | ||
60b156e3 | 388 | if (*(string_buf + needed - 1) != '\n') |
f7b24d53 NAMV |
389 | pr_info("format at %s:%d:%s doesn't end in newline\n", file, |
390 | msgdata->msg_line, msgdata->msg_fn); | |
d7e09d03 PT |
391 | |
392 | header.ph_len = known_size + needed; | |
393 | debug_buf = (char *)page_address(tage->page) + tage->used; | |
394 | ||
395 | if (libcfs_debug_binary) { | |
396 | memcpy(debug_buf, &header, sizeof(header)); | |
397 | tage->used += sizeof(header); | |
398 | debug_buf += sizeof(header); | |
399 | } | |
400 | ||
401 | /* indent message according to the nesting level */ | |
402 | while (depth-- > 0) { | |
403 | *(debug_buf++) = '.'; | |
b2952d62 | 404 | ++tage->used; |
d7e09d03 PT |
405 | } |
406 | ||
407 | strcpy(debug_buf, file); | |
408 | tage->used += strlen(file) + 1; | |
409 | debug_buf += strlen(file) + 1; | |
410 | ||
411 | if (msgdata->msg_fn) { | |
412 | strcpy(debug_buf, msgdata->msg_fn); | |
413 | tage->used += strlen(msgdata->msg_fn) + 1; | |
414 | debug_buf += strlen(msgdata->msg_fn) + 1; | |
415 | } | |
416 | ||
417 | __LASSERT(debug_buf == string_buf); | |
418 | ||
419 | tage->used += needed; | |
09cbfeaf | 420 | __LASSERT(tage->used <= PAGE_SIZE); |
d7e09d03 PT |
421 | |
422 | console: | |
6fc68ea5 | 423 | if (!(mask & libcfs_printk)) { |
d7e09d03 | 424 | /* no console output requested */ |
15d9f520 | 425 | if (tcd) |
d7e09d03 PT |
426 | cfs_trace_put_tcd(tcd); |
427 | return 1; | |
428 | } | |
429 | ||
15d9f520 | 430 | if (cdls) { |
d7e09d03 | 431 | if (libcfs_console_ratelimit && |
d95531fc | 432 | cdls->cdls_next && /* not first time ever */ |
d7e09d03 PT |
433 | !cfs_time_after(cfs_time_current(), cdls->cdls_next)) { |
434 | /* skipping a console message */ | |
435 | cdls->cdls_count++; | |
15d9f520 | 436 | if (tcd) |
d7e09d03 PT |
437 | cfs_trace_put_tcd(tcd); |
438 | return 1; | |
439 | } | |
440 | ||
417ad156 OD |
441 | if (cfs_time_after(cfs_time_current(), |
442 | cdls->cdls_next + libcfs_console_max_delay + | |
443 | cfs_time_seconds(10))) { | |
d7e09d03 PT |
444 | /* last timeout was a long time ago */ |
445 | cdls->cdls_delay /= libcfs_console_backoff * 4; | |
446 | } else { | |
447 | cdls->cdls_delay *= libcfs_console_backoff; | |
d7e09d03 PT |
448 | } |
449 | ||
b7d0254c RH |
450 | if (cdls->cdls_delay < libcfs_console_min_delay) |
451 | cdls->cdls_delay = libcfs_console_min_delay; | |
452 | else if (cdls->cdls_delay > libcfs_console_max_delay) | |
453 | cdls->cdls_delay = libcfs_console_max_delay; | |
454 | ||
d7e09d03 PT |
455 | /* ensure cdls_next is never zero after it's been seen */ |
456 | cdls->cdls_next = (cfs_time_current() + cdls->cdls_delay) | 1; | |
457 | } | |
458 | ||
15d9f520 | 459 | if (tcd) { |
d7e09d03 PT |
460 | cfs_print_to_console(&header, mask, string_buf, needed, file, |
461 | msgdata->msg_fn); | |
462 | cfs_trace_put_tcd(tcd); | |
463 | } else { | |
464 | string_buf = cfs_trace_get_console_buffer(); | |
465 | ||
466 | needed = 0; | |
15d9f520 | 467 | if (format1) { |
d7e09d03 PT |
468 | va_copy(ap, args); |
469 | needed = vsnprintf(string_buf, | |
470 | CFS_TRACE_CONSOLE_BUFFER_SIZE, | |
471 | format1, ap); | |
472 | va_end(ap); | |
473 | } | |
15d9f520 | 474 | if (format2) { |
d7e09d03 PT |
475 | remain = CFS_TRACE_CONSOLE_BUFFER_SIZE - needed; |
476 | if (remain > 0) { | |
477 | va_start(ap, format2); | |
60b156e3 | 478 | needed += vsnprintf(string_buf + needed, remain, |
d7e09d03 PT |
479 | format2, ap); |
480 | va_end(ap); | |
481 | } | |
482 | } | |
483 | cfs_print_to_console(&header, mask, | |
484 | string_buf, needed, file, msgdata->msg_fn); | |
485 | ||
87af1d2e | 486 | put_cpu(); |
d7e09d03 PT |
487 | } |
488 | ||
d95531fc | 489 | if (cdls && cdls->cdls_count) { |
d7e09d03 PT |
490 | string_buf = cfs_trace_get_console_buffer(); |
491 | ||
492 | needed = snprintf(string_buf, CFS_TRACE_CONSOLE_BUFFER_SIZE, | |
493 | "Skipped %d previous similar message%s\n", | |
494 | cdls->cdls_count, | |
495 | (cdls->cdls_count > 1) ? "s" : ""); | |
496 | ||
497 | cfs_print_to_console(&header, mask, | |
498 | string_buf, needed, file, msgdata->msg_fn); | |
499 | ||
87af1d2e | 500 | put_cpu(); |
d7e09d03 PT |
501 | cdls->cdls_count = 0; |
502 | } | |
503 | ||
504 | return 0; | |
505 | } | |
506 | EXPORT_SYMBOL(libcfs_debug_vmsg2); | |
507 | ||
508 | void | |
509 | cfs_trace_assertion_failed(const char *str, | |
510 | struct libcfs_debug_msg_data *msgdata) | |
511 | { | |
512 | struct ptldebug_header hdr; | |
513 | ||
514 | libcfs_panic_in_progress = 1; | |
515 | libcfs_catastrophe = 1; | |
516 | mb(); | |
517 | ||
518 | cfs_set_ptldebug_header(&hdr, msgdata, CDEBUG_STACK()); | |
519 | ||
520 | cfs_print_to_console(&hdr, D_EMERG, str, strlen(str), | |
521 | msgdata->msg_file, msgdata->msg_fn); | |
522 | ||
523 | panic("Lustre debug assertion failure\n"); | |
524 | ||
525 | /* not reached */ | |
526 | } | |
527 | ||
528 | static void | |
529 | panic_collect_pages(struct page_collection *pc) | |
530 | { | |
531 | /* Do the collect_pages job on a single CPU: assumes that all other | |
532 | * CPUs have been stopped during a panic. If this isn't true for some | |
a3fbcb3c OD |
533 | * arch, this will have to be implemented separately in each arch. |
534 | */ | |
d7e09d03 | 535 | struct cfs_trace_cpu_data *tcd; |
715475ae JS |
536 | int i; |
537 | int j; | |
d7e09d03 PT |
538 | |
539 | INIT_LIST_HEAD(&pc->pc_pages); | |
540 | ||
541 | cfs_tcd_for_each(tcd, i, j) { | |
542 | list_splice_init(&tcd->tcd_pages, &pc->pc_pages); | |
543 | tcd->tcd_cur_pages = 0; | |
544 | ||
545 | if (pc->pc_want_daemon_pages) { | |
ae0b4833 | 546 | list_splice_init(&tcd->tcd_daemon_pages, &pc->pc_pages); |
d7e09d03 PT |
547 | tcd->tcd_cur_daemon_pages = 0; |
548 | } | |
549 | } | |
550 | } | |
551 | ||
552 | static void collect_pages_on_all_cpus(struct page_collection *pc) | |
553 | { | |
554 | struct cfs_trace_cpu_data *tcd; | |
555 | int i, cpu; | |
556 | ||
7ac24eda | 557 | for_each_possible_cpu(cpu) { |
d7e09d03 PT |
558 | cfs_tcd_for_each_type_lock(tcd, i, cpu) { |
559 | list_splice_init(&tcd->tcd_pages, &pc->pc_pages); | |
560 | tcd->tcd_cur_pages = 0; | |
561 | if (pc->pc_want_daemon_pages) { | |
562 | list_splice_init(&tcd->tcd_daemon_pages, | |
ae0b4833 | 563 | &pc->pc_pages); |
d7e09d03 PT |
564 | tcd->tcd_cur_daemon_pages = 0; |
565 | } | |
566 | } | |
567 | } | |
d7e09d03 PT |
568 | } |
569 | ||
570 | static void collect_pages(struct page_collection *pc) | |
571 | { | |
572 | INIT_LIST_HEAD(&pc->pc_pages); | |
573 | ||
574 | if (libcfs_panic_in_progress) | |
575 | panic_collect_pages(pc); | |
576 | else | |
577 | collect_pages_on_all_cpus(pc); | |
578 | } | |
579 | ||
580 | static void put_pages_back_on_all_cpus(struct page_collection *pc) | |
581 | { | |
582 | struct cfs_trace_cpu_data *tcd; | |
583 | struct list_head *cur_head; | |
584 | struct cfs_trace_page *tage; | |
585 | struct cfs_trace_page *tmp; | |
586 | int i, cpu; | |
587 | ||
7ac24eda | 588 | for_each_possible_cpu(cpu) { |
d7e09d03 PT |
589 | cfs_tcd_for_each_type_lock(tcd, i, cpu) { |
590 | cur_head = tcd->tcd_pages.next; | |
591 | ||
592 | list_for_each_entry_safe(tage, tmp, &pc->pc_pages, | |
593 | linkage) { | |
d7e09d03 PT |
594 | __LASSERT_TAGE_INVARIANT(tage); |
595 | ||
596 | if (tage->cpu != cpu || tage->type != i) | |
597 | continue; | |
598 | ||
599 | cfs_tage_to_tail(tage, cur_head); | |
600 | tcd->tcd_cur_pages++; | |
601 | } | |
602 | } | |
603 | } | |
d7e09d03 PT |
604 | } |
605 | ||
606 | static void put_pages_back(struct page_collection *pc) | |
607 | { | |
608 | if (!libcfs_panic_in_progress) | |
609 | put_pages_back_on_all_cpus(pc); | |
610 | } | |
611 | ||
612 | /* Add pages to a per-cpu debug daemon ringbuffer. This buffer makes sure that | |
613 | * we have a good amount of data at all times for dumping during an LBUG, even | |
614 | * if we have been steadily writing (and otherwise discarding) pages via the | |
a3fbcb3c OD |
615 | * debug daemon. |
616 | */ | |
d7e09d03 PT |
617 | static void put_pages_on_tcd_daemon_list(struct page_collection *pc, |
618 | struct cfs_trace_cpu_data *tcd) | |
619 | { | |
620 | struct cfs_trace_page *tage; | |
621 | struct cfs_trace_page *tmp; | |
622 | ||
d7e09d03 | 623 | list_for_each_entry_safe(tage, tmp, &pc->pc_pages, linkage) { |
d7e09d03 PT |
624 | __LASSERT_TAGE_INVARIANT(tage); |
625 | ||
626 | if (tage->cpu != tcd->tcd_cpu || tage->type != tcd->tcd_type) | |
627 | continue; | |
628 | ||
629 | cfs_tage_to_tail(tage, &tcd->tcd_daemon_pages); | |
630 | tcd->tcd_cur_daemon_pages++; | |
631 | ||
632 | if (tcd->tcd_cur_daemon_pages > tcd->tcd_max_pages) { | |
633 | struct cfs_trace_page *victim; | |
634 | ||
635 | __LASSERT(!list_empty(&tcd->tcd_daemon_pages)); | |
636 | victim = cfs_tage_from_list(tcd->tcd_daemon_pages.next); | |
637 | ||
638 | __LASSERT_TAGE_INVARIANT(victim); | |
639 | ||
640 | list_del(&victim->linkage); | |
641 | cfs_tage_free(victim); | |
642 | tcd->tcd_cur_daemon_pages--; | |
643 | } | |
644 | } | |
d7e09d03 PT |
645 | } |
646 | ||
647 | static void put_pages_on_daemon_list(struct page_collection *pc) | |
648 | { | |
649 | struct cfs_trace_cpu_data *tcd; | |
650 | int i, cpu; | |
651 | ||
7ac24eda | 652 | for_each_possible_cpu(cpu) { |
d7e09d03 PT |
653 | cfs_tcd_for_each_type_lock(tcd, i, cpu) |
654 | put_pages_on_tcd_daemon_list(pc, tcd); | |
655 | } | |
656 | } | |
657 | ||
658 | void cfs_trace_debug_print(void) | |
659 | { | |
660 | struct page_collection pc; | |
661 | struct cfs_trace_page *tage; | |
662 | struct cfs_trace_page *tmp; | |
663 | ||
d7e09d03 PT |
664 | pc.pc_want_daemon_pages = 1; |
665 | collect_pages(&pc); | |
666 | list_for_each_entry_safe(tage, tmp, &pc.pc_pages, linkage) { | |
667 | char *p, *file, *fn; | |
668 | struct page *page; | |
669 | ||
670 | __LASSERT_TAGE_INVARIANT(tage); | |
671 | ||
672 | page = tage->page; | |
673 | p = page_address(page); | |
674 | while (p < ((char *)page_address(page) + tage->used)) { | |
675 | struct ptldebug_header *hdr; | |
676 | int len; | |
50ffcb7e | 677 | |
d7e09d03 PT |
678 | hdr = (void *)p; |
679 | p += sizeof(*hdr); | |
680 | file = p; | |
681 | p += strlen(file) + 1; | |
682 | fn = p; | |
683 | p += strlen(fn) + 1; | |
684 | len = hdr->ph_len - (int)(p - (char *)hdr); | |
685 | ||
686 | cfs_print_to_console(hdr, D_EMERG, p, len, file, fn); | |
687 | ||
688 | p += len; | |
689 | } | |
690 | ||
691 | list_del(&tage->linkage); | |
692 | cfs_tage_free(tage); | |
693 | } | |
694 | } | |
695 | ||
696 | int cfs_tracefile_dump_all_pages(char *filename) | |
697 | { | |
715475ae JS |
698 | struct page_collection pc; |
699 | struct file *filp; | |
700 | struct cfs_trace_page *tage; | |
701 | struct cfs_trace_page *tmp; | |
702 | char *buf; | |
e3e30e10 | 703 | mm_segment_t __oldfs; |
d7e09d03 PT |
704 | int rc; |
705 | ||
d7e09d03 PT |
706 | cfs_tracefile_write_lock(); |
707 | ||
60b156e3 OD |
708 | filp = filp_open(filename, O_CREAT | O_EXCL | O_WRONLY | O_LARGEFILE, |
709 | 0600); | |
d7e09d03 PT |
710 | if (IS_ERR(filp)) { |
711 | rc = PTR_ERR(filp); | |
712 | filp = NULL; | |
324588ec | 713 | pr_err("LustreError: can't open %s for dump: rc %d\n", |
ae0b4833 | 714 | filename, rc); |
d7e09d03 PT |
715 | goto out; |
716 | } | |
717 | ||
d7e09d03 PT |
718 | pc.pc_want_daemon_pages = 1; |
719 | collect_pages(&pc); | |
720 | if (list_empty(&pc.pc_pages)) { | |
721 | rc = 0; | |
722 | goto close; | |
723 | } | |
e3e30e10 JS |
724 | __oldfs = get_fs(); |
725 | set_fs(get_ds()); | |
d7e09d03 PT |
726 | |
727 | /* ok, for now, just write the pages. in the future we'll be building | |
a3fbcb3c OD |
728 | * iobufs with the pages and calling generic_direct_IO |
729 | */ | |
d7e09d03 | 730 | list_for_each_entry_safe(tage, tmp, &pc.pc_pages, linkage) { |
d7e09d03 PT |
731 | __LASSERT_TAGE_INVARIANT(tage); |
732 | ||
7ebcb731 PT |
733 | buf = kmap(tage->page); |
734 | rc = vfs_write(filp, (__force const char __user *)buf, | |
735 | tage->used, &filp->f_pos); | |
736 | kunmap(tage->page); | |
737 | ||
d7e09d03 | 738 | if (rc != (int)tage->used) { |
f7b24d53 NAMV |
739 | pr_warn("wanted to write %u but wrote %d\n", tage->used, |
740 | rc); | |
d7e09d03 PT |
741 | put_pages_back(&pc); |
742 | __LASSERT(list_empty(&pc.pc_pages)); | |
743 | break; | |
744 | } | |
745 | list_del(&tage->linkage); | |
746 | cfs_tage_free(tage); | |
747 | } | |
e3e30e10 | 748 | set_fs(__oldfs); |
a630b22d | 749 | rc = vfs_fsync(filp, 1); |
d7e09d03 | 750 | if (rc) |
324588ec | 751 | pr_err("sync returns %d\n", rc); |
d7e09d03 PT |
752 | close: |
753 | filp_close(filp, NULL); | |
754 | out: | |
755 | cfs_tracefile_write_unlock(); | |
756 | return rc; | |
757 | } | |
758 | ||
759 | void cfs_trace_flush_pages(void) | |
760 | { | |
761 | struct page_collection pc; | |
762 | struct cfs_trace_page *tage; | |
763 | struct cfs_trace_page *tmp; | |
764 | ||
d7e09d03 PT |
765 | pc.pc_want_daemon_pages = 1; |
766 | collect_pages(&pc); | |
767 | list_for_each_entry_safe(tage, tmp, &pc.pc_pages, linkage) { | |
d7e09d03 PT |
768 | __LASSERT_TAGE_INVARIANT(tage); |
769 | ||
770 | list_del(&tage->linkage); | |
771 | cfs_tage_free(tage); | |
772 | } | |
773 | } | |
774 | ||
775 | int cfs_trace_copyin_string(char *knl_buffer, int knl_buffer_nob, | |
5a08f067 | 776 | const char __user *usr_buffer, int usr_buffer_nob) |
d7e09d03 | 777 | { |
715475ae | 778 | int nob; |
d7e09d03 PT |
779 | |
780 | if (usr_buffer_nob > knl_buffer_nob) | |
781 | return -EOVERFLOW; | |
782 | ||
783 | if (copy_from_user((void *)knl_buffer, | |
5a08f067 | 784 | usr_buffer, usr_buffer_nob)) |
d7e09d03 PT |
785 | return -EFAULT; |
786 | ||
787 | nob = strnlen(knl_buffer, usr_buffer_nob); | |
788 | while (nob-- >= 0) /* strip trailing whitespace */ | |
789 | if (!isspace(knl_buffer[nob])) | |
790 | break; | |
791 | ||
792 | if (nob < 0) /* empty string */ | |
793 | return -EINVAL; | |
794 | ||
795 | if (nob == knl_buffer_nob) /* no space to terminate */ | |
796 | return -EOVERFLOW; | |
797 | ||
798 | knl_buffer[nob + 1] = 0; /* terminate */ | |
799 | return 0; | |
800 | } | |
801 | EXPORT_SYMBOL(cfs_trace_copyin_string); | |
802 | ||
5a08f067 | 803 | int cfs_trace_copyout_string(char __user *usr_buffer, int usr_buffer_nob, |
d7e09d03 PT |
804 | const char *knl_buffer, char *append) |
805 | { | |
3092242c OD |
806 | /* |
807 | * NB if 'append' != NULL, it's a single character to append to the | |
808 | * copied out string - usually "\n" or "" (i.e. a terminating zero byte) | |
809 | */ | |
715475ae | 810 | int nob = strlen(knl_buffer); |
d7e09d03 PT |
811 | |
812 | if (nob > usr_buffer_nob) | |
813 | nob = usr_buffer_nob; | |
814 | ||
815 | if (copy_to_user(usr_buffer, knl_buffer, nob)) | |
816 | return -EFAULT; | |
817 | ||
15d9f520 | 818 | if (append && nob < usr_buffer_nob) { |
d7e09d03 PT |
819 | if (copy_to_user(usr_buffer + nob, append, 1)) |
820 | return -EFAULT; | |
821 | ||
822 | nob++; | |
823 | } | |
824 | ||
825 | return nob; | |
826 | } | |
827 | EXPORT_SYMBOL(cfs_trace_copyout_string); | |
828 | ||
829 | int cfs_trace_allocate_string_buffer(char **str, int nob) | |
830 | { | |
09cbfeaf | 831 | if (nob > 2 * PAGE_SIZE) /* string must be "sensible" */ |
d7e09d03 PT |
832 | return -EINVAL; |
833 | ||
40113370 | 834 | *str = kmalloc(nob, GFP_KERNEL | __GFP_ZERO); |
15d9f520 | 835 | if (!*str) |
d7e09d03 PT |
836 | return -ENOMEM; |
837 | ||
838 | return 0; | |
839 | } | |
840 | ||
5a08f067 | 841 | int cfs_trace_dump_debug_buffer_usrstr(void __user *usr_str, int usr_str_nob) |
d7e09d03 | 842 | { |
715475ae JS |
843 | char *str; |
844 | int rc; | |
d7e09d03 PT |
845 | |
846 | rc = cfs_trace_allocate_string_buffer(&str, usr_str_nob + 1); | |
d95531fc | 847 | if (rc) |
d7e09d03 PT |
848 | return rc; |
849 | ||
850 | rc = cfs_trace_copyin_string(str, usr_str_nob + 1, | |
851 | usr_str, usr_str_nob); | |
d95531fc | 852 | if (rc) |
d7e09d03 PT |
853 | goto out; |
854 | ||
855 | if (str[0] != '/') { | |
856 | rc = -EINVAL; | |
857 | goto out; | |
858 | } | |
859 | rc = cfs_tracefile_dump_all_pages(str); | |
860 | out: | |
7cbf673d | 861 | kfree(str); |
d7e09d03 PT |
862 | return rc; |
863 | } | |
864 | ||
865 | int cfs_trace_daemon_command(char *str) | |
866 | { | |
715475ae | 867 | int rc = 0; |
d7e09d03 PT |
868 | |
869 | cfs_tracefile_write_lock(); | |
870 | ||
6fc68ea5 | 871 | if (!strcmp(str, "stop")) { |
d7e09d03 PT |
872 | cfs_tracefile_write_unlock(); |
873 | cfs_trace_stop_thread(); | |
874 | cfs_tracefile_write_lock(); | |
875 | memset(cfs_tracefile, 0, sizeof(cfs_tracefile)); | |
876 | ||
6fc68ea5 | 877 | } else if (!strncmp(str, "size=", 5)) { |
4d8a00ef OD |
878 | unsigned long tmp; |
879 | ||
880 | rc = kstrtoul(str + 5, 10, &tmp); | |
881 | if (!rc) { | |
882 | if (tmp < 10 || tmp > 20480) | |
883 | cfs_tracefile_size = CFS_TRACEFILE_SIZE; | |
884 | else | |
885 | cfs_tracefile_size = tmp << 20; | |
886 | } | |
d7e09d03 PT |
887 | } else if (strlen(str) >= sizeof(cfs_tracefile)) { |
888 | rc = -ENAMETOOLONG; | |
889 | } else if (str[0] != '/') { | |
890 | rc = -EINVAL; | |
891 | } else { | |
892 | strcpy(cfs_tracefile, str); | |
893 | ||
217d31c4 | 894 | pr_info("debug daemon will attempt to start writing to %s (%lukB max)\n", |
f7b24d53 NAMV |
895 | cfs_tracefile, |
896 | (long)(cfs_tracefile_size >> 10)); | |
d7e09d03 PT |
897 | |
898 | cfs_trace_start_thread(); | |
899 | } | |
900 | ||
901 | cfs_tracefile_write_unlock(); | |
902 | return rc; | |
903 | } | |
904 | ||
5a08f067 | 905 | int cfs_trace_daemon_command_usrstr(void __user *usr_str, int usr_str_nob) |
d7e09d03 PT |
906 | { |
907 | char *str; | |
715475ae | 908 | int rc; |
d7e09d03 PT |
909 | |
910 | rc = cfs_trace_allocate_string_buffer(&str, usr_str_nob + 1); | |
d95531fc | 911 | if (rc) |
d7e09d03 PT |
912 | return rc; |
913 | ||
914 | rc = cfs_trace_copyin_string(str, usr_str_nob + 1, | |
ae0b4833 | 915 | usr_str, usr_str_nob); |
6fc68ea5 | 916 | if (!rc) |
d7e09d03 PT |
917 | rc = cfs_trace_daemon_command(str); |
918 | ||
7cbf673d | 919 | kfree(str); |
d7e09d03 PT |
920 | return rc; |
921 | } | |
922 | ||
923 | int cfs_trace_set_debug_mb(int mb) | |
924 | { | |
925 | int i; | |
926 | int j; | |
927 | int pages; | |
928 | int limit = cfs_trace_max_debug_mb(); | |
929 | struct cfs_trace_cpu_data *tcd; | |
930 | ||
931 | if (mb < num_possible_cpus()) { | |
217d31c4 | 932 | pr_warn("%d MB is too small for debug buffer size, setting it to %d MB.\n", |
f7b24d53 | 933 | mb, num_possible_cpus()); |
d7e09d03 PT |
934 | mb = num_possible_cpus(); |
935 | } | |
936 | ||
937 | if (mb > limit) { | |
217d31c4 | 938 | pr_warn("%d MB is too large for debug buffer size, setting it to %d MB.\n", |
f7b24d53 | 939 | mb, limit); |
d7e09d03 PT |
940 | mb = limit; |
941 | } | |
942 | ||
943 | mb /= num_possible_cpus(); | |
09cbfeaf | 944 | pages = mb << (20 - PAGE_SHIFT); |
d7e09d03 PT |
945 | |
946 | cfs_tracefile_write_lock(); | |
947 | ||
948 | cfs_tcd_for_each(tcd, i, j) | |
949 | tcd->tcd_max_pages = (pages * tcd->tcd_pages_factor) / 100; | |
950 | ||
951 | cfs_tracefile_write_unlock(); | |
952 | ||
953 | return 0; | |
954 | } | |
955 | ||
d7e09d03 PT |
956 | int cfs_trace_get_debug_mb(void) |
957 | { | |
958 | int i; | |
959 | int j; | |
960 | struct cfs_trace_cpu_data *tcd; | |
961 | int total_pages = 0; | |
962 | ||
963 | cfs_tracefile_read_lock(); | |
964 | ||
965 | cfs_tcd_for_each(tcd, i, j) | |
966 | total_pages += tcd->tcd_max_pages; | |
967 | ||
968 | cfs_tracefile_read_unlock(); | |
969 | ||
09cbfeaf | 970 | return (total_pages >> (20 - PAGE_SHIFT)) + 1; |
d7e09d03 PT |
971 | } |
972 | ||
973 | static int tracefiled(void *arg) | |
974 | { | |
975 | struct page_collection pc; | |
976 | struct tracefiled_ctl *tctl = arg; | |
977 | struct cfs_trace_page *tage; | |
978 | struct cfs_trace_page *tmp; | |
e3e30e10 | 979 | mm_segment_t __oldfs; |
d7e09d03 | 980 | struct file *filp; |
7ebcb731 | 981 | char *buf; |
d7e09d03 PT |
982 | int last_loop = 0; |
983 | int rc; | |
984 | ||
d7e09d03 PT |
985 | /* we're started late enough that we pick up init's fs context */ |
986 | /* this is so broken in uml? what on earth is going on? */ | |
987 | ||
d7e09d03 PT |
988 | complete(&tctl->tctl_start); |
989 | ||
990 | while (1) { | |
991 | wait_queue_t __wait; | |
992 | ||
993 | pc.pc_want_daemon_pages = 0; | |
994 | collect_pages(&pc); | |
995 | if (list_empty(&pc.pc_pages)) | |
996 | goto end_loop; | |
997 | ||
998 | filp = NULL; | |
999 | cfs_tracefile_read_lock(); | |
d95531fc | 1000 | if (cfs_tracefile[0]) { |
d7e09d03 PT |
1001 | filp = filp_open(cfs_tracefile, |
1002 | O_CREAT | O_RDWR | O_LARGEFILE, | |
1003 | 0600); | |
1004 | if (IS_ERR(filp)) { | |
1005 | rc = PTR_ERR(filp); | |
1006 | filp = NULL; | |
f7b24d53 NAMV |
1007 | pr_warn("couldn't open %s: %d\n", cfs_tracefile, |
1008 | rc); | |
d7e09d03 PT |
1009 | } |
1010 | } | |
1011 | cfs_tracefile_read_unlock(); | |
15d9f520 | 1012 | if (!filp) { |
d7e09d03 PT |
1013 | put_pages_on_daemon_list(&pc); |
1014 | __LASSERT(list_empty(&pc.pc_pages)); | |
1015 | goto end_loop; | |
1016 | } | |
e3e30e10 JS |
1017 | __oldfs = get_fs(); |
1018 | set_fs(get_ds()); | |
d7e09d03 | 1019 | |
ae0b4833 | 1020 | list_for_each_entry_safe(tage, tmp, &pc.pc_pages, linkage) { |
d7e09d03 PT |
1021 | static loff_t f_pos; |
1022 | ||
1023 | __LASSERT_TAGE_INVARIANT(tage); | |
1024 | ||
1025 | if (f_pos >= (off_t)cfs_tracefile_size) | |
1026 | f_pos = 0; | |
2a8a3597 AV |
1027 | else if (f_pos > i_size_read(file_inode(filp))) |
1028 | f_pos = i_size_read(file_inode(filp)); | |
d7e09d03 | 1029 | |
7ebcb731 PT |
1030 | buf = kmap(tage->page); |
1031 | rc = vfs_write(filp, (__force const char __user *)buf, | |
1032 | tage->used, &f_pos); | |
1033 | kunmap(tage->page); | |
1034 | ||
d7e09d03 | 1035 | if (rc != (int)tage->used) { |
f7b24d53 NAMV |
1036 | pr_warn("wanted to write %u but wrote %d\n", |
1037 | tage->used, rc); | |
d7e09d03 PT |
1038 | put_pages_back(&pc); |
1039 | __LASSERT(list_empty(&pc.pc_pages)); | |
2661e6c4 | 1040 | break; |
d7e09d03 PT |
1041 | } |
1042 | } | |
e3e30e10 | 1043 | set_fs(__oldfs); |
d7e09d03 PT |
1044 | |
1045 | filp_close(filp, NULL); | |
1046 | put_pages_on_daemon_list(&pc); | |
1047 | if (!list_empty(&pc.pc_pages)) { | |
1048 | int i; | |
1049 | ||
217d31c4 | 1050 | pr_alert("trace pages aren't empty\n"); |
ae0b4833 | 1051 | pr_err("total cpus(%d): ", num_possible_cpus()); |
d7e09d03 PT |
1052 | for (i = 0; i < num_possible_cpus(); i++) |
1053 | if (cpu_online(i)) | |
324588ec | 1054 | pr_cont("%d(on) ", i); |
d7e09d03 | 1055 | else |
324588ec TJ |
1056 | pr_cont("%d(off) ", i); |
1057 | pr_cont("\n"); | |
d7e09d03 PT |
1058 | |
1059 | i = 0; | |
1060 | list_for_each_entry_safe(tage, tmp, &pc.pc_pages, | |
ae0b4833 | 1061 | linkage) |
324588ec | 1062 | pr_err("page %d belongs to cpu %d\n", |
ae0b4833 | 1063 | ++i, tage->cpu); |
324588ec | 1064 | pr_err("There are %d pages unwritten\n", i); |
d7e09d03 PT |
1065 | } |
1066 | __LASSERT(list_empty(&pc.pc_pages)); | |
1067 | end_loop: | |
1068 | if (atomic_read(&tctl->tctl_shutdown)) { | |
6fc68ea5 | 1069 | if (!last_loop) { |
d7e09d03 PT |
1070 | last_loop = 1; |
1071 | continue; | |
1072 | } else { | |
1073 | break; | |
1074 | } | |
1075 | } | |
9e795d35 | 1076 | init_waitqueue_entry(&__wait, current); |
d7e09d03 PT |
1077 | add_wait_queue(&tctl->tctl_waitq, &__wait); |
1078 | set_current_state(TASK_INTERRUPTIBLE); | |
b7efb98d | 1079 | schedule_timeout(cfs_time_seconds(1)); |
d7e09d03 PT |
1080 | remove_wait_queue(&tctl->tctl_waitq, &__wait); |
1081 | } | |
1082 | complete(&tctl->tctl_stop); | |
1083 | return 0; | |
1084 | } | |
1085 | ||
1086 | int cfs_trace_start_thread(void) | |
1087 | { | |
1088 | struct tracefiled_ctl *tctl = &trace_tctl; | |
060c2820 | 1089 | struct task_struct *task; |
d7e09d03 PT |
1090 | int rc = 0; |
1091 | ||
1092 | mutex_lock(&cfs_trace_thread_mutex); | |
1093 | if (thread_running) | |
1094 | goto out; | |
1095 | ||
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); | |
1100 | ||
060c2820 JH |
1101 | task = kthread_run(tracefiled, tctl, "ktracefiled"); |
1102 | if (IS_ERR(task)) { | |
1103 | rc = PTR_ERR(task); | |
d7e09d03 PT |
1104 | goto out; |
1105 | } | |
1106 | ||
1107 | wait_for_completion(&tctl->tctl_start); | |
1108 | thread_running = 1; | |
1109 | out: | |
1110 | mutex_unlock(&cfs_trace_thread_mutex); | |
1111 | return rc; | |
1112 | } | |
1113 | ||
1114 | void cfs_trace_stop_thread(void) | |
1115 | { | |
1116 | struct tracefiled_ctl *tctl = &trace_tctl; | |
1117 | ||
1118 | mutex_lock(&cfs_trace_thread_mutex); | |
1119 | if (thread_running) { | |
217d31c4 | 1120 | pr_info("shutting down debug daemon thread...\n"); |
d7e09d03 PT |
1121 | atomic_set(&tctl->tctl_shutdown, 1); |
1122 | wait_for_completion(&tctl->tctl_stop); | |
1123 | thread_running = 0; | |
1124 | } | |
1125 | mutex_unlock(&cfs_trace_thread_mutex); | |
1126 | } | |
1127 | ||
1128 | int cfs_tracefile_init(int max_pages) | |
1129 | { | |
1130 | struct cfs_trace_cpu_data *tcd; | |
715475ae JS |
1131 | int i; |
1132 | int j; | |
1133 | int rc; | |
1134 | int factor; | |
d7e09d03 PT |
1135 | |
1136 | rc = cfs_tracefile_init_arch(); | |
d95531fc | 1137 | if (rc) |
d7e09d03 PT |
1138 | return rc; |
1139 | ||
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; | |
1152 | } | |
1153 | ||
1154 | return 0; | |
1155 | } | |
1156 | ||
1157 | static void trace_cleanup_on_all_cpus(void) | |
1158 | { | |
1159 | struct cfs_trace_cpu_data *tcd; | |
1160 | struct cfs_trace_page *tage; | |
1161 | struct cfs_trace_page *tmp; | |
1162 | int i, cpu; | |
1163 | ||
7ac24eda | 1164 | for_each_possible_cpu(cpu) { |
d7e09d03 PT |
1165 | cfs_tcd_for_each_type_lock(tcd, i, cpu) { |
1166 | tcd->tcd_shutting_down = 1; | |
1167 | ||
1168 | list_for_each_entry_safe(tage, tmp, &tcd->tcd_pages, | |
ae0b4833 | 1169 | linkage) { |
d7e09d03 PT |
1170 | __LASSERT_TAGE_INVARIANT(tage); |
1171 | ||
1172 | list_del(&tage->linkage); | |
1173 | cfs_tage_free(tage); | |
1174 | } | |
1175 | ||
1176 | tcd->tcd_cur_pages = 0; | |
1177 | } | |
1178 | } | |
1179 | } | |
1180 | ||
1181 | static void cfs_trace_cleanup(void) | |
1182 | { | |
1183 | struct page_collection pc; | |
1184 | ||
1185 | INIT_LIST_HEAD(&pc.pc_pages); | |
d7e09d03 PT |
1186 | |
1187 | trace_cleanup_on_all_cpus(); | |
1188 | ||
1189 | cfs_tracefile_fini_arch(); | |
1190 | } | |
1191 | ||
1192 | void cfs_tracefile_exit(void) | |
1193 | { | |
1194 | cfs_trace_stop_thread(); | |
1195 | cfs_trace_cleanup(); | |
1196 | } |