PCI: separate pci_setup_bridge to small functions
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / kernel / trace / trace_selftest.c
blob280fea470d67ea603f230d8bfc7868d5d48e62b2
1 /* Include in trace.c */
3 #include <linux/stringify.h>
4 #include <linux/kthread.h>
5 #include <linux/delay.h>
7 static inline int trace_valid_entry(struct trace_entry *entry)
9 switch (entry->type) {
10 case TRACE_FN:
11 case TRACE_CTX:
12 case TRACE_WAKE:
13 case TRACE_STACK:
14 case TRACE_PRINT:
15 case TRACE_SPECIAL:
16 case TRACE_BRANCH:
17 case TRACE_GRAPH_ENT:
18 case TRACE_GRAPH_RET:
19 case TRACE_HW_BRANCHES:
20 case TRACE_KSYM:
21 return 1;
23 return 0;
26 static int trace_test_buffer_cpu(struct trace_array *tr, int cpu)
28 struct ring_buffer_event *event;
29 struct trace_entry *entry;
30 unsigned int loops = 0;
32 while ((event = ring_buffer_consume(tr->buffer, cpu, NULL))) {
33 entry = ring_buffer_event_data(event);
36 * The ring buffer is a size of trace_buf_size, if
37 * we loop more than the size, there's something wrong
38 * with the ring buffer.
40 if (loops++ > trace_buf_size) {
41 printk(KERN_CONT ".. bad ring buffer ");
42 goto failed;
44 if (!trace_valid_entry(entry)) {
45 printk(KERN_CONT ".. invalid entry %d ",
46 entry->type);
47 goto failed;
50 return 0;
52 failed:
53 /* disable tracing */
54 tracing_disabled = 1;
55 printk(KERN_CONT ".. corrupted trace buffer .. ");
56 return -1;
60 * Test the trace buffer to see if all the elements
61 * are still sane.
63 static int trace_test_buffer(struct trace_array *tr, unsigned long *count)
65 unsigned long flags, cnt = 0;
66 int cpu, ret = 0;
68 /* Don't allow flipping of max traces now */
69 local_irq_save(flags);
70 arch_spin_lock(&ftrace_max_lock);
72 cnt = ring_buffer_entries(tr->buffer);
75 * The trace_test_buffer_cpu runs a while loop to consume all data.
76 * If the calling tracer is broken, and is constantly filling
77 * the buffer, this will run forever, and hard lock the box.
78 * We disable the ring buffer while we do this test to prevent
79 * a hard lock up.
81 tracing_off();
82 for_each_possible_cpu(cpu) {
83 ret = trace_test_buffer_cpu(tr, cpu);
84 if (ret)
85 break;
87 tracing_on();
88 arch_spin_unlock(&ftrace_max_lock);
89 local_irq_restore(flags);
91 if (count)
92 *count = cnt;
94 return ret;
97 static inline void warn_failed_init_tracer(struct tracer *trace, int init_ret)
99 printk(KERN_WARNING "Failed to init %s tracer, init returned %d\n",
100 trace->name, init_ret);
102 #ifdef CONFIG_FUNCTION_TRACER
104 #ifdef CONFIG_DYNAMIC_FTRACE
106 /* Test dynamic code modification and ftrace filters */
107 int trace_selftest_startup_dynamic_tracing(struct tracer *trace,
108 struct trace_array *tr,
109 int (*func)(void))
111 int save_ftrace_enabled = ftrace_enabled;
112 int save_tracer_enabled = tracer_enabled;
113 unsigned long count;
114 char *func_name;
115 int ret;
117 /* The ftrace test PASSED */
118 printk(KERN_CONT "PASSED\n");
119 pr_info("Testing dynamic ftrace: ");
121 /* enable tracing, and record the filter function */
122 ftrace_enabled = 1;
123 tracer_enabled = 1;
125 /* passed in by parameter to fool gcc from optimizing */
126 func();
129 * Some archs *cough*PowerPC*cough* add characters to the
130 * start of the function names. We simply put a '*' to
131 * accommodate them.
133 func_name = "*" __stringify(DYN_FTRACE_TEST_NAME);
135 /* filter only on our function */
136 ftrace_set_filter(func_name, strlen(func_name), 1);
138 /* enable tracing */
139 ret = tracer_init(trace, tr);
140 if (ret) {
141 warn_failed_init_tracer(trace, ret);
142 goto out;
145 /* Sleep for a 1/10 of a second */
146 msleep(100);
148 /* we should have nothing in the buffer */
149 ret = trace_test_buffer(tr, &count);
150 if (ret)
151 goto out;
153 if (count) {
154 ret = -1;
155 printk(KERN_CONT ".. filter did not filter .. ");
156 goto out;
159 /* call our function again */
160 func();
162 /* sleep again */
163 msleep(100);
165 /* stop the tracing. */
166 tracing_stop();
167 ftrace_enabled = 0;
169 /* check the trace buffer */
170 ret = trace_test_buffer(tr, &count);
171 trace->reset(tr);
172 tracing_start();
174 /* we should only have one item */
175 if (!ret && count != 1) {
176 printk(KERN_CONT ".. filter failed count=%ld ..", count);
177 ret = -1;
178 goto out;
181 out:
182 ftrace_enabled = save_ftrace_enabled;
183 tracer_enabled = save_tracer_enabled;
185 /* Enable tracing on all functions again */
186 ftrace_set_filter(NULL, 0, 1);
188 return ret;
190 #else
191 # define trace_selftest_startup_dynamic_tracing(trace, tr, func) ({ 0; })
192 #endif /* CONFIG_DYNAMIC_FTRACE */
195 * Simple verification test of ftrace function tracer.
196 * Enable ftrace, sleep 1/10 second, and then read the trace
197 * buffer to see if all is in order.
200 trace_selftest_startup_function(struct tracer *trace, struct trace_array *tr)
202 int save_ftrace_enabled = ftrace_enabled;
203 int save_tracer_enabled = tracer_enabled;
204 unsigned long count;
205 int ret;
207 /* make sure msleep has been recorded */
208 msleep(1);
210 /* start the tracing */
211 ftrace_enabled = 1;
212 tracer_enabled = 1;
214 ret = tracer_init(trace, tr);
215 if (ret) {
216 warn_failed_init_tracer(trace, ret);
217 goto out;
220 /* Sleep for a 1/10 of a second */
221 msleep(100);
222 /* stop the tracing. */
223 tracing_stop();
224 ftrace_enabled = 0;
226 /* check the trace buffer */
227 ret = trace_test_buffer(tr, &count);
228 trace->reset(tr);
229 tracing_start();
231 if (!ret && !count) {
232 printk(KERN_CONT ".. no entries found ..");
233 ret = -1;
234 goto out;
237 ret = trace_selftest_startup_dynamic_tracing(trace, tr,
238 DYN_FTRACE_TEST_NAME);
240 out:
241 ftrace_enabled = save_ftrace_enabled;
242 tracer_enabled = save_tracer_enabled;
244 /* kill ftrace totally if we failed */
245 if (ret)
246 ftrace_kill();
248 return ret;
250 #endif /* CONFIG_FUNCTION_TRACER */
253 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
255 /* Maximum number of functions to trace before diagnosing a hang */
256 #define GRAPH_MAX_FUNC_TEST 100000000
258 static void __ftrace_dump(bool disable_tracing);
259 static unsigned int graph_hang_thresh;
261 /* Wrap the real function entry probe to avoid possible hanging */
262 static int trace_graph_entry_watchdog(struct ftrace_graph_ent *trace)
264 /* This is harmlessly racy, we want to approximately detect a hang */
265 if (unlikely(++graph_hang_thresh > GRAPH_MAX_FUNC_TEST)) {
266 ftrace_graph_stop();
267 printk(KERN_WARNING "BUG: Function graph tracer hang!\n");
268 if (ftrace_dump_on_oops)
269 __ftrace_dump(false);
270 return 0;
273 return trace_graph_entry(trace);
277 * Pretty much the same than for the function tracer from which the selftest
278 * has been borrowed.
281 trace_selftest_startup_function_graph(struct tracer *trace,
282 struct trace_array *tr)
284 int ret;
285 unsigned long count;
288 * Simulate the init() callback but we attach a watchdog callback
289 * to detect and recover from possible hangs
291 tracing_reset_online_cpus(tr);
292 set_graph_array(tr);
293 ret = register_ftrace_graph(&trace_graph_return,
294 &trace_graph_entry_watchdog);
295 if (ret) {
296 warn_failed_init_tracer(trace, ret);
297 goto out;
299 tracing_start_cmdline_record();
301 /* Sleep for a 1/10 of a second */
302 msleep(100);
304 /* Have we just recovered from a hang? */
305 if (graph_hang_thresh > GRAPH_MAX_FUNC_TEST) {
306 tracing_selftest_disabled = true;
307 ret = -1;
308 goto out;
311 tracing_stop();
313 /* check the trace buffer */
314 ret = trace_test_buffer(tr, &count);
316 trace->reset(tr);
317 tracing_start();
319 if (!ret && !count) {
320 printk(KERN_CONT ".. no entries found ..");
321 ret = -1;
322 goto out;
325 /* Don't test dynamic tracing, the function tracer already did */
327 out:
328 /* Stop it if we failed */
329 if (ret)
330 ftrace_graph_stop();
332 return ret;
334 #endif /* CONFIG_FUNCTION_GRAPH_TRACER */
337 #ifdef CONFIG_IRQSOFF_TRACER
339 trace_selftest_startup_irqsoff(struct tracer *trace, struct trace_array *tr)
341 unsigned long save_max = tracing_max_latency;
342 unsigned long count;
343 int ret;
345 /* start the tracing */
346 ret = tracer_init(trace, tr);
347 if (ret) {
348 warn_failed_init_tracer(trace, ret);
349 return ret;
352 /* reset the max latency */
353 tracing_max_latency = 0;
354 /* disable interrupts for a bit */
355 local_irq_disable();
356 udelay(100);
357 local_irq_enable();
360 * Stop the tracer to avoid a warning subsequent
361 * to buffer flipping failure because tracing_stop()
362 * disables the tr and max buffers, making flipping impossible
363 * in case of parallels max irqs off latencies.
365 trace->stop(tr);
366 /* stop the tracing. */
367 tracing_stop();
368 /* check both trace buffers */
369 ret = trace_test_buffer(tr, NULL);
370 if (!ret)
371 ret = trace_test_buffer(&max_tr, &count);
372 trace->reset(tr);
373 tracing_start();
375 if (!ret && !count) {
376 printk(KERN_CONT ".. no entries found ..");
377 ret = -1;
380 tracing_max_latency = save_max;
382 return ret;
384 #endif /* CONFIG_IRQSOFF_TRACER */
386 #ifdef CONFIG_PREEMPT_TRACER
388 trace_selftest_startup_preemptoff(struct tracer *trace, struct trace_array *tr)
390 unsigned long save_max = tracing_max_latency;
391 unsigned long count;
392 int ret;
395 * Now that the big kernel lock is no longer preemptable,
396 * and this is called with the BKL held, it will always
397 * fail. If preemption is already disabled, simply
398 * pass the test. When the BKL is removed, or becomes
399 * preemptible again, we will once again test this,
400 * so keep it in.
402 if (preempt_count()) {
403 printk(KERN_CONT "can not test ... force ");
404 return 0;
407 /* start the tracing */
408 ret = tracer_init(trace, tr);
409 if (ret) {
410 warn_failed_init_tracer(trace, ret);
411 return ret;
414 /* reset the max latency */
415 tracing_max_latency = 0;
416 /* disable preemption for a bit */
417 preempt_disable();
418 udelay(100);
419 preempt_enable();
422 * Stop the tracer to avoid a warning subsequent
423 * to buffer flipping failure because tracing_stop()
424 * disables the tr and max buffers, making flipping impossible
425 * in case of parallels max preempt off latencies.
427 trace->stop(tr);
428 /* stop the tracing. */
429 tracing_stop();
430 /* check both trace buffers */
431 ret = trace_test_buffer(tr, NULL);
432 if (!ret)
433 ret = trace_test_buffer(&max_tr, &count);
434 trace->reset(tr);
435 tracing_start();
437 if (!ret && !count) {
438 printk(KERN_CONT ".. no entries found ..");
439 ret = -1;
442 tracing_max_latency = save_max;
444 return ret;
446 #endif /* CONFIG_PREEMPT_TRACER */
448 #if defined(CONFIG_IRQSOFF_TRACER) && defined(CONFIG_PREEMPT_TRACER)
450 trace_selftest_startup_preemptirqsoff(struct tracer *trace, struct trace_array *tr)
452 unsigned long save_max = tracing_max_latency;
453 unsigned long count;
454 int ret;
457 * Now that the big kernel lock is no longer preemptable,
458 * and this is called with the BKL held, it will always
459 * fail. If preemption is already disabled, simply
460 * pass the test. When the BKL is removed, or becomes
461 * preemptible again, we will once again test this,
462 * so keep it in.
464 if (preempt_count()) {
465 printk(KERN_CONT "can not test ... force ");
466 return 0;
469 /* start the tracing */
470 ret = tracer_init(trace, tr);
471 if (ret) {
472 warn_failed_init_tracer(trace, ret);
473 goto out_no_start;
476 /* reset the max latency */
477 tracing_max_latency = 0;
479 /* disable preemption and interrupts for a bit */
480 preempt_disable();
481 local_irq_disable();
482 udelay(100);
483 preempt_enable();
484 /* reverse the order of preempt vs irqs */
485 local_irq_enable();
488 * Stop the tracer to avoid a warning subsequent
489 * to buffer flipping failure because tracing_stop()
490 * disables the tr and max buffers, making flipping impossible
491 * in case of parallels max irqs/preempt off latencies.
493 trace->stop(tr);
494 /* stop the tracing. */
495 tracing_stop();
496 /* check both trace buffers */
497 ret = trace_test_buffer(tr, NULL);
498 if (ret)
499 goto out;
501 ret = trace_test_buffer(&max_tr, &count);
502 if (ret)
503 goto out;
505 if (!ret && !count) {
506 printk(KERN_CONT ".. no entries found ..");
507 ret = -1;
508 goto out;
511 /* do the test by disabling interrupts first this time */
512 tracing_max_latency = 0;
513 tracing_start();
514 trace->start(tr);
516 preempt_disable();
517 local_irq_disable();
518 udelay(100);
519 preempt_enable();
520 /* reverse the order of preempt vs irqs */
521 local_irq_enable();
523 trace->stop(tr);
524 /* stop the tracing. */
525 tracing_stop();
526 /* check both trace buffers */
527 ret = trace_test_buffer(tr, NULL);
528 if (ret)
529 goto out;
531 ret = trace_test_buffer(&max_tr, &count);
533 if (!ret && !count) {
534 printk(KERN_CONT ".. no entries found ..");
535 ret = -1;
536 goto out;
539 out:
540 tracing_start();
541 out_no_start:
542 trace->reset(tr);
543 tracing_max_latency = save_max;
545 return ret;
547 #endif /* CONFIG_IRQSOFF_TRACER && CONFIG_PREEMPT_TRACER */
549 #ifdef CONFIG_NOP_TRACER
551 trace_selftest_startup_nop(struct tracer *trace, struct trace_array *tr)
553 /* What could possibly go wrong? */
554 return 0;
556 #endif
558 #ifdef CONFIG_SCHED_TRACER
559 static int trace_wakeup_test_thread(void *data)
561 /* Make this a RT thread, doesn't need to be too high */
562 struct sched_param param = { .sched_priority = 5 };
563 struct completion *x = data;
565 sched_setscheduler(current, SCHED_FIFO, &param);
567 /* Make it know we have a new prio */
568 complete(x);
570 /* now go to sleep and let the test wake us up */
571 set_current_state(TASK_INTERRUPTIBLE);
572 schedule();
574 /* we are awake, now wait to disappear */
575 while (!kthread_should_stop()) {
577 * This is an RT task, do short sleeps to let
578 * others run.
580 msleep(100);
583 return 0;
587 trace_selftest_startup_wakeup(struct tracer *trace, struct trace_array *tr)
589 unsigned long save_max = tracing_max_latency;
590 struct task_struct *p;
591 struct completion isrt;
592 unsigned long count;
593 int ret;
595 init_completion(&isrt);
597 /* create a high prio thread */
598 p = kthread_run(trace_wakeup_test_thread, &isrt, "ftrace-test");
599 if (IS_ERR(p)) {
600 printk(KERN_CONT "Failed to create ftrace wakeup test thread ");
601 return -1;
604 /* make sure the thread is running at an RT prio */
605 wait_for_completion(&isrt);
607 /* start the tracing */
608 ret = tracer_init(trace, tr);
609 if (ret) {
610 warn_failed_init_tracer(trace, ret);
611 return ret;
614 /* reset the max latency */
615 tracing_max_latency = 0;
617 /* sleep to let the RT thread sleep too */
618 msleep(100);
621 * Yes this is slightly racy. It is possible that for some
622 * strange reason that the RT thread we created, did not
623 * call schedule for 100ms after doing the completion,
624 * and we do a wakeup on a task that already is awake.
625 * But that is extremely unlikely, and the worst thing that
626 * happens in such a case, is that we disable tracing.
627 * Honestly, if this race does happen something is horrible
628 * wrong with the system.
631 wake_up_process(p);
633 /* give a little time to let the thread wake up */
634 msleep(100);
636 /* stop the tracing. */
637 tracing_stop();
638 /* check both trace buffers */
639 ret = trace_test_buffer(tr, NULL);
640 if (!ret)
641 ret = trace_test_buffer(&max_tr, &count);
644 trace->reset(tr);
645 tracing_start();
647 tracing_max_latency = save_max;
649 /* kill the thread */
650 kthread_stop(p);
652 if (!ret && !count) {
653 printk(KERN_CONT ".. no entries found ..");
654 ret = -1;
657 return ret;
659 #endif /* CONFIG_SCHED_TRACER */
661 #ifdef CONFIG_CONTEXT_SWITCH_TRACER
663 trace_selftest_startup_sched_switch(struct tracer *trace, struct trace_array *tr)
665 unsigned long count;
666 int ret;
668 /* start the tracing */
669 ret = tracer_init(trace, tr);
670 if (ret) {
671 warn_failed_init_tracer(trace, ret);
672 return ret;
675 /* Sleep for a 1/10 of a second */
676 msleep(100);
677 /* stop the tracing. */
678 tracing_stop();
679 /* check the trace buffer */
680 ret = trace_test_buffer(tr, &count);
681 trace->reset(tr);
682 tracing_start();
684 if (!ret && !count) {
685 printk(KERN_CONT ".. no entries found ..");
686 ret = -1;
689 return ret;
691 #endif /* CONFIG_CONTEXT_SWITCH_TRACER */
693 #ifdef CONFIG_SYSPROF_TRACER
695 trace_selftest_startup_sysprof(struct tracer *trace, struct trace_array *tr)
697 unsigned long count;
698 int ret;
700 /* start the tracing */
701 ret = tracer_init(trace, tr);
702 if (ret) {
703 warn_failed_init_tracer(trace, ret);
704 return ret;
707 /* Sleep for a 1/10 of a second */
708 msleep(100);
709 /* stop the tracing. */
710 tracing_stop();
711 /* check the trace buffer */
712 ret = trace_test_buffer(tr, &count);
713 trace->reset(tr);
714 tracing_start();
716 if (!ret && !count) {
717 printk(KERN_CONT ".. no entries found ..");
718 ret = -1;
721 return ret;
723 #endif /* CONFIG_SYSPROF_TRACER */
725 #ifdef CONFIG_BRANCH_TRACER
727 trace_selftest_startup_branch(struct tracer *trace, struct trace_array *tr)
729 unsigned long count;
730 int ret;
732 /* start the tracing */
733 ret = tracer_init(trace, tr);
734 if (ret) {
735 warn_failed_init_tracer(trace, ret);
736 return ret;
739 /* Sleep for a 1/10 of a second */
740 msleep(100);
741 /* stop the tracing. */
742 tracing_stop();
743 /* check the trace buffer */
744 ret = trace_test_buffer(tr, &count);
745 trace->reset(tr);
746 tracing_start();
748 if (!ret && !count) {
749 printk(KERN_CONT ".. no entries found ..");
750 ret = -1;
753 return ret;
755 #endif /* CONFIG_BRANCH_TRACER */
757 #ifdef CONFIG_HW_BRANCH_TRACER
759 trace_selftest_startup_hw_branches(struct tracer *trace,
760 struct trace_array *tr)
762 struct trace_iterator *iter;
763 struct tracer tracer;
764 unsigned long count;
765 int ret;
767 if (!trace->open) {
768 printk(KERN_CONT "missing open function...");
769 return -1;
772 ret = tracer_init(trace, tr);
773 if (ret) {
774 warn_failed_init_tracer(trace, ret);
775 return ret;
779 * The hw-branch tracer needs to collect the trace from the various
780 * cpu trace buffers - before tracing is stopped.
782 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
783 if (!iter)
784 return -ENOMEM;
786 memcpy(&tracer, trace, sizeof(tracer));
788 iter->trace = &tracer;
789 iter->tr = tr;
790 iter->pos = -1;
791 mutex_init(&iter->mutex);
793 trace->open(iter);
795 mutex_destroy(&iter->mutex);
796 kfree(iter);
798 tracing_stop();
800 ret = trace_test_buffer(tr, &count);
801 trace->reset(tr);
802 tracing_start();
804 if (!ret && !count) {
805 printk(KERN_CONT "no entries found..");
806 ret = -1;
809 return ret;
811 #endif /* CONFIG_HW_BRANCH_TRACER */
813 #ifdef CONFIG_KSYM_TRACER
814 static int ksym_selftest_dummy;
817 trace_selftest_startup_ksym(struct tracer *trace, struct trace_array *tr)
819 unsigned long count;
820 int ret;
822 /* start the tracing */
823 ret = tracer_init(trace, tr);
824 if (ret) {
825 warn_failed_init_tracer(trace, ret);
826 return ret;
829 ksym_selftest_dummy = 0;
830 /* Register the read-write tracing request */
832 ret = process_new_ksym_entry("ksym_selftest_dummy",
833 HW_BREAKPOINT_R | HW_BREAKPOINT_W,
834 (unsigned long)(&ksym_selftest_dummy));
836 if (ret < 0) {
837 printk(KERN_CONT "ksym_trace read-write startup test failed\n");
838 goto ret_path;
840 /* Perform a read and a write operation over the dummy variable to
841 * trigger the tracer
843 if (ksym_selftest_dummy == 0)
844 ksym_selftest_dummy++;
846 /* stop the tracing. */
847 tracing_stop();
848 /* check the trace buffer */
849 ret = trace_test_buffer(tr, &count);
850 trace->reset(tr);
851 tracing_start();
853 /* read & write operations - one each is performed on the dummy variable
854 * triggering two entries in the trace buffer
856 if (!ret && count != 2) {
857 printk(KERN_CONT "Ksym tracer startup test failed");
858 ret = -1;
861 ret_path:
862 return ret;
864 #endif /* CONFIG_KSYM_TRACER */