blob: 737bd9484822498d9db077640ed7acc0859babb9 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/**
2 * @file buffer_sync.c
3 *
4 * @remark Copyright 2002 OProfile authors
5 * @remark Read the file COPYING
6 *
7 * @author John Levon <levon@movementarian.org>
Barry Kasindorf345c2572008-07-22 21:08:54 +02008 * @author Barry Kasindorf
Linus Torvalds1da177e2005-04-16 15:20:36 -07009 *
10 * This is the core of the buffer management. Each
11 * CPU buffer is processed and entered into the
12 * global event buffer. Such processing is necessary
13 * in several circumstances, mentioned below.
14 *
15 * The processing does the job of converting the
16 * transitory EIP value into a persistent dentry/offset
17 * value that the profiler can record at its leisure.
18 *
19 * See fs/dcookies.c for a description of the dentry/offset
20 * objects.
21 */
22
23#include <linux/mm.h>
24#include <linux/workqueue.h>
25#include <linux/notifier.h>
26#include <linux/dcookies.h>
27#include <linux/profile.h>
28#include <linux/module.h>
29#include <linux/fs.h>
Bob Nelson14748552007-07-20 21:39:53 +020030#include <linux/oprofile.h>
Alexey Dobriyane8edc6e2007-05-21 01:22:52 +040031#include <linux/sched.h>
Bob Nelson14748552007-07-20 21:39:53 +020032
Linus Torvalds1da177e2005-04-16 15:20:36 -070033#include "oprofile_stats.h"
34#include "event_buffer.h"
35#include "cpu_buffer.h"
36#include "buffer_sync.h"
Robert Richter73185e02008-07-22 21:08:51 +020037
Linus Torvalds1da177e2005-04-16 15:20:36 -070038static LIST_HEAD(dying_tasks);
39static LIST_HEAD(dead_tasks);
40static cpumask_t marked_cpus = CPU_MASK_NONE;
41static DEFINE_SPINLOCK(task_mortuary);
42static void process_task_mortuary(void);
43
Linus Torvalds1da177e2005-04-16 15:20:36 -070044/* Take ownership of the task struct and place it on the
45 * list for processing. Only after two full buffer syncs
46 * does the task eventually get freed, because by then
47 * we are sure we will not reference it again.
Paul E. McKenney4369ef32006-01-08 01:01:35 -080048 * Can be invoked from softirq via RCU callback due to
49 * call_rcu() of the task struct, hence the _irqsave.
Linus Torvalds1da177e2005-04-16 15:20:36 -070050 */
Robert Richter73185e02008-07-22 21:08:51 +020051static int
52task_free_notify(struct notifier_block *self, unsigned long val, void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -070053{
Paul E. McKenney4369ef32006-01-08 01:01:35 -080054 unsigned long flags;
Robert Richter73185e02008-07-22 21:08:51 +020055 struct task_struct *task = data;
Paul E. McKenney4369ef32006-01-08 01:01:35 -080056 spin_lock_irqsave(&task_mortuary, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -070057 list_add(&task->tasks, &dying_tasks);
Paul E. McKenney4369ef32006-01-08 01:01:35 -080058 spin_unlock_irqrestore(&task_mortuary, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -070059 return NOTIFY_OK;
60}
61
62
63/* The task is on its way out. A sync of the buffer means we can catch
64 * any remaining samples for this task.
65 */
Robert Richter73185e02008-07-22 21:08:51 +020066static int
67task_exit_notify(struct notifier_block *self, unsigned long val, void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -070068{
69 /* To avoid latency problems, we only process the current CPU,
70 * hoping that most samples for the task are on this CPU
71 */
Ingo Molnar39c715b2005-06-21 17:14:34 -070072 sync_buffer(raw_smp_processor_id());
Robert Richter73185e02008-07-22 21:08:51 +020073 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -070074}
75
76
77/* The task is about to try a do_munmap(). We peek at what it's going to
78 * do, and if it's an executable region, process the samples first, so
79 * we don't lose any. This does not have to be exact, it's a QoI issue
80 * only.
81 */
Robert Richter73185e02008-07-22 21:08:51 +020082static int
83munmap_notify(struct notifier_block *self, unsigned long val, void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -070084{
85 unsigned long addr = (unsigned long)data;
Robert Richter73185e02008-07-22 21:08:51 +020086 struct mm_struct *mm = current->mm;
87 struct vm_area_struct *mpnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -070088
89 down_read(&mm->mmap_sem);
90
91 mpnt = find_vma(mm, addr);
92 if (mpnt && mpnt->vm_file && (mpnt->vm_flags & VM_EXEC)) {
93 up_read(&mm->mmap_sem);
94 /* To avoid latency problems, we only process the current CPU,
95 * hoping that most samples for the task are on this CPU
96 */
Ingo Molnar39c715b2005-06-21 17:14:34 -070097 sync_buffer(raw_smp_processor_id());
Linus Torvalds1da177e2005-04-16 15:20:36 -070098 return 0;
99 }
100
101 up_read(&mm->mmap_sem);
102 return 0;
103}
104
Robert Richter73185e02008-07-22 21:08:51 +0200105
Linus Torvalds1da177e2005-04-16 15:20:36 -0700106/* We need to be told about new modules so we don't attribute to a previously
107 * loaded module, or drop the samples on the floor.
108 */
Robert Richter73185e02008-07-22 21:08:51 +0200109static int
110module_load_notify(struct notifier_block *self, unsigned long val, void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700111{
112#ifdef CONFIG_MODULES
113 if (val != MODULE_STATE_COMING)
114 return 0;
115
116 /* FIXME: should we process all CPU buffers ? */
Markus Armbruster59cc1852006-06-25 05:47:33 -0700117 mutex_lock(&buffer_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700118 add_event_entry(ESCAPE_CODE);
119 add_event_entry(MODULE_LOADED_CODE);
Markus Armbruster59cc1852006-06-25 05:47:33 -0700120 mutex_unlock(&buffer_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700121#endif
122 return 0;
123}
124
Robert Richter73185e02008-07-22 21:08:51 +0200125
Linus Torvalds1da177e2005-04-16 15:20:36 -0700126static struct notifier_block task_free_nb = {
127 .notifier_call = task_free_notify,
128};
129
130static struct notifier_block task_exit_nb = {
131 .notifier_call = task_exit_notify,
132};
133
134static struct notifier_block munmap_nb = {
135 .notifier_call = munmap_notify,
136};
137
138static struct notifier_block module_load_nb = {
139 .notifier_call = module_load_notify,
140};
141
Robert Richter73185e02008-07-22 21:08:51 +0200142
Linus Torvalds1da177e2005-04-16 15:20:36 -0700143static void end_sync(void)
144{
145 end_cpu_work();
146 /* make sure we don't leak task structs */
147 process_task_mortuary();
148 process_task_mortuary();
149}
150
151
152int sync_start(void)
153{
154 int err;
155
156 start_cpu_work();
157
158 err = task_handoff_register(&task_free_nb);
159 if (err)
160 goto out1;
161 err = profile_event_register(PROFILE_TASK_EXIT, &task_exit_nb);
162 if (err)
163 goto out2;
164 err = profile_event_register(PROFILE_MUNMAP, &munmap_nb);
165 if (err)
166 goto out3;
167 err = register_module_notifier(&module_load_nb);
168 if (err)
169 goto out4;
170
171out:
172 return err;
173out4:
174 profile_event_unregister(PROFILE_MUNMAP, &munmap_nb);
175out3:
176 profile_event_unregister(PROFILE_TASK_EXIT, &task_exit_nb);
177out2:
178 task_handoff_unregister(&task_free_nb);
179out1:
180 end_sync();
181 goto out;
182}
183
184
185void sync_stop(void)
186{
187 unregister_module_notifier(&module_load_nb);
188 profile_event_unregister(PROFILE_MUNMAP, &munmap_nb);
189 profile_event_unregister(PROFILE_TASK_EXIT, &task_exit_nb);
190 task_handoff_unregister(&task_free_nb);
191 end_sync();
192}
193
Jan Blunck448678a2008-02-14 19:38:36 -0800194
Linus Torvalds1da177e2005-04-16 15:20:36 -0700195/* Optimisation. We can manage without taking the dcookie sem
196 * because we cannot reach this code without at least one
197 * dcookie user still being registered (namely, the reader
198 * of the event buffer). */
Jan Blunck448678a2008-02-14 19:38:36 -0800199static inline unsigned long fast_get_dcookie(struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700200{
201 unsigned long cookie;
Jan Blunck448678a2008-02-14 19:38:36 -0800202
203 if (path->dentry->d_cookie)
204 return (unsigned long)path->dentry;
205 get_dcookie(path, &cookie);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700206 return cookie;
207}
208
Jan Blunck448678a2008-02-14 19:38:36 -0800209
Linus Torvalds1da177e2005-04-16 15:20:36 -0700210/* Look up the dcookie for the task's first VM_EXECUTABLE mapping,
211 * which corresponds loosely to "application name". This is
212 * not strictly necessary but allows oprofile to associate
213 * shared-library samples with particular applications
214 */
Robert Richter73185e02008-07-22 21:08:51 +0200215static unsigned long get_exec_dcookie(struct mm_struct *mm)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700216{
John Levon0c0a4002005-06-23 22:02:47 -0700217 unsigned long cookie = NO_COOKIE;
Robert Richter73185e02008-07-22 21:08:51 +0200218 struct vm_area_struct *vma;
219
Linus Torvalds1da177e2005-04-16 15:20:36 -0700220 if (!mm)
221 goto out;
Robert Richter73185e02008-07-22 21:08:51 +0200222
Linus Torvalds1da177e2005-04-16 15:20:36 -0700223 for (vma = mm->mmap; vma; vma = vma->vm_next) {
224 if (!vma->vm_file)
225 continue;
226 if (!(vma->vm_flags & VM_EXECUTABLE))
227 continue;
Jan Blunck448678a2008-02-14 19:38:36 -0800228 cookie = fast_get_dcookie(&vma->vm_file->f_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700229 break;
230 }
231
232out:
233 return cookie;
234}
235
236
237/* Convert the EIP value of a sample into a persistent dentry/offset
238 * pair that can then be added to the global event buffer. We make
239 * sure to do this lookup before a mm->mmap modification happens so
240 * we don't lose track.
241 */
Robert Richter73185e02008-07-22 21:08:51 +0200242static unsigned long
243lookup_dcookie(struct mm_struct *mm, unsigned long addr, off_t *offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700244{
John Levon0c0a4002005-06-23 22:02:47 -0700245 unsigned long cookie = NO_COOKIE;
Robert Richter73185e02008-07-22 21:08:51 +0200246 struct vm_area_struct *vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700247
248 for (vma = find_vma(mm, addr); vma; vma = vma->vm_next) {
Robert Richter73185e02008-07-22 21:08:51 +0200249
Linus Torvalds1da177e2005-04-16 15:20:36 -0700250 if (addr < vma->vm_start || addr >= vma->vm_end)
251 continue;
252
John Levon0c0a4002005-06-23 22:02:47 -0700253 if (vma->vm_file) {
Jan Blunck448678a2008-02-14 19:38:36 -0800254 cookie = fast_get_dcookie(&vma->vm_file->f_path);
John Levon0c0a4002005-06-23 22:02:47 -0700255 *offset = (vma->vm_pgoff << PAGE_SHIFT) + addr -
256 vma->vm_start;
257 } else {
258 /* must be an anonymous map */
259 *offset = addr;
260 }
261
Linus Torvalds1da177e2005-04-16 15:20:36 -0700262 break;
263 }
264
John Levon0c0a4002005-06-23 22:02:47 -0700265 if (!vma)
266 cookie = INVALID_COOKIE;
267
Linus Torvalds1da177e2005-04-16 15:20:36 -0700268 return cookie;
269}
270
John Levon0c0a4002005-06-23 22:02:47 -0700271static unsigned long last_cookie = INVALID_COOKIE;
Robert Richter73185e02008-07-22 21:08:51 +0200272
Linus Torvalds1da177e2005-04-16 15:20:36 -0700273static void add_cpu_switch(int i)
274{
275 add_event_entry(ESCAPE_CODE);
276 add_event_entry(CPU_SWITCH_CODE);
277 add_event_entry(i);
John Levon0c0a4002005-06-23 22:02:47 -0700278 last_cookie = INVALID_COOKIE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700279}
280
281static void add_kernel_ctx_switch(unsigned int in_kernel)
282{
283 add_event_entry(ESCAPE_CODE);
284 if (in_kernel)
Robert Richter73185e02008-07-22 21:08:51 +0200285 add_event_entry(KERNEL_ENTER_SWITCH_CODE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700286 else
Robert Richter73185e02008-07-22 21:08:51 +0200287 add_event_entry(KERNEL_EXIT_SWITCH_CODE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700288}
Robert Richter73185e02008-07-22 21:08:51 +0200289
Linus Torvalds1da177e2005-04-16 15:20:36 -0700290static void
Robert Richter73185e02008-07-22 21:08:51 +0200291add_user_ctx_switch(struct task_struct const *task, unsigned long cookie)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700292{
293 add_event_entry(ESCAPE_CODE);
Robert Richter73185e02008-07-22 21:08:51 +0200294 add_event_entry(CTX_SWITCH_CODE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700295 add_event_entry(task->pid);
296 add_event_entry(cookie);
297 /* Another code for daemon back-compat */
298 add_event_entry(ESCAPE_CODE);
299 add_event_entry(CTX_TGID_CODE);
300 add_event_entry(task->tgid);
301}
302
Robert Richter73185e02008-07-22 21:08:51 +0200303
Linus Torvalds1da177e2005-04-16 15:20:36 -0700304static void add_cookie_switch(unsigned long cookie)
305{
306 add_event_entry(ESCAPE_CODE);
307 add_event_entry(COOKIE_SWITCH_CODE);
308 add_event_entry(cookie);
309}
310
Robert Richter73185e02008-07-22 21:08:51 +0200311
Linus Torvalds1da177e2005-04-16 15:20:36 -0700312static void add_trace_begin(void)
313{
314 add_event_entry(ESCAPE_CODE);
315 add_event_entry(TRACE_BEGIN_CODE);
316}
317
Robert Richter852402c2008-07-22 21:09:06 +0200318#ifdef CONFIG_OPROFILE_IBS
319
Barry Kasindorf345c2572008-07-22 21:08:54 +0200320#define IBS_FETCH_CODE_SIZE 2
321#define IBS_OP_CODE_SIZE 5
Barry Kasindorf345c2572008-07-22 21:08:54 +0200322
323/*
324 * Add IBS fetch and op entries to event buffer
325 */
Robert Richter6dad8282008-12-09 01:21:32 +0100326static void add_ibs_begin(int cpu, int code, struct mm_struct *mm)
Barry Kasindorf345c2572008-07-22 21:08:54 +0200327{
328 unsigned long rip;
329 int i, count;
330 unsigned long ibs_cookie = 0;
331 off_t offset;
Robert Richter6dad8282008-12-09 01:21:32 +0100332 struct op_sample *sample;
Barry Kasindorf345c2572008-07-22 21:08:54 +0200333
Robert Richter6dad8282008-12-09 01:21:32 +0100334 sample = cpu_buffer_read_entry(cpu);
335 if (!sample)
336 goto Error;
337 rip = sample->eip;
Barry Kasindorf345c2572008-07-22 21:08:54 +0200338
339#ifdef __LP64__
Robert Richter6dad8282008-12-09 01:21:32 +0100340 rip += sample->event << 32;
Barry Kasindorf345c2572008-07-22 21:08:54 +0200341#endif
342
343 if (mm) {
344 ibs_cookie = lookup_dcookie(mm, rip, &offset);
345
346 if (ibs_cookie == NO_COOKIE)
347 offset = rip;
348 if (ibs_cookie == INVALID_COOKIE) {
349 atomic_inc(&oprofile_stats.sample_lost_no_mapping);
350 offset = rip;
351 }
352 if (ibs_cookie != last_cookie) {
353 add_cookie_switch(ibs_cookie);
354 last_cookie = ibs_cookie;
355 }
356 } else
357 offset = rip;
358
359 add_event_entry(ESCAPE_CODE);
360 add_event_entry(code);
361 add_event_entry(offset); /* Offset from Dcookie */
362
363 /* we send the Dcookie offset, but send the raw Linear Add also*/
Robert Richter6dad8282008-12-09 01:21:32 +0100364 add_event_entry(sample->eip);
365 add_event_entry(sample->event);
Barry Kasindorf345c2572008-07-22 21:08:54 +0200366
367 if (code == IBS_FETCH_CODE)
368 count = IBS_FETCH_CODE_SIZE; /*IBS FETCH is 2 int64s*/
369 else
370 count = IBS_OP_CODE_SIZE; /*IBS OP is 5 int64s*/
371
372 for (i = 0; i < count; i++) {
Robert Richter6dad8282008-12-09 01:21:32 +0100373 sample = cpu_buffer_read_entry(cpu);
374 if (!sample)
375 goto Error;
376 add_event_entry(sample->eip);
377 add_event_entry(sample->event);
Barry Kasindorf345c2572008-07-22 21:08:54 +0200378 }
Robert Richter6dad8282008-12-09 01:21:32 +0100379
380 return;
381
382Error:
383 return;
Barry Kasindorf345c2572008-07-22 21:08:54 +0200384}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700385
Robert Richter852402c2008-07-22 21:09:06 +0200386#endif
387
Linus Torvalds1da177e2005-04-16 15:20:36 -0700388static void add_sample_entry(unsigned long offset, unsigned long event)
389{
390 add_event_entry(offset);
391 add_event_entry(event);
392}
393
394
Robert Richter73185e02008-07-22 21:08:51 +0200395static int add_us_sample(struct mm_struct *mm, struct op_sample *s)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700396{
397 unsigned long cookie;
398 off_t offset;
Robert Richter73185e02008-07-22 21:08:51 +0200399
400 cookie = lookup_dcookie(mm, s->eip, &offset);
401
John Levon0c0a4002005-06-23 22:02:47 -0700402 if (cookie == INVALID_COOKIE) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700403 atomic_inc(&oprofile_stats.sample_lost_no_mapping);
404 return 0;
405 }
406
407 if (cookie != last_cookie) {
408 add_cookie_switch(cookie);
409 last_cookie = cookie;
410 }
411
412 add_sample_entry(offset, s->event);
413
414 return 1;
415}
416
Robert Richter73185e02008-07-22 21:08:51 +0200417
Linus Torvalds1da177e2005-04-16 15:20:36 -0700418/* Add a sample to the global event buffer. If possible the
419 * sample is converted into a persistent dentry/offset pair
420 * for later lookup from userspace.
421 */
422static int
Robert Richter73185e02008-07-22 21:08:51 +0200423add_sample(struct mm_struct *mm, struct op_sample *s, int in_kernel)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700424{
425 if (in_kernel) {
426 add_sample_entry(s->eip, s->event);
427 return 1;
428 } else if (mm) {
429 return add_us_sample(mm, s);
430 } else {
431 atomic_inc(&oprofile_stats.sample_lost_no_mm);
432 }
433 return 0;
434}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700435
Robert Richter73185e02008-07-22 21:08:51 +0200436
437static void release_mm(struct mm_struct *mm)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700438{
439 if (!mm)
440 return;
441 up_read(&mm->mmap_sem);
442 mmput(mm);
443}
444
445
Robert Richter73185e02008-07-22 21:08:51 +0200446static struct mm_struct *take_tasks_mm(struct task_struct *task)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700447{
Robert Richter73185e02008-07-22 21:08:51 +0200448 struct mm_struct *mm = get_task_mm(task);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700449 if (mm)
450 down_read(&mm->mmap_sem);
451 return mm;
452}
453
454
455static inline int is_code(unsigned long val)
456{
457 return val == ESCAPE_CODE;
458}
Robert Richter73185e02008-07-22 21:08:51 +0200459
Linus Torvalds1da177e2005-04-16 15:20:36 -0700460
Linus Torvalds1da177e2005-04-16 15:20:36 -0700461/* Move tasks along towards death. Any tasks on dead_tasks
462 * will definitely have no remaining references in any
463 * CPU buffers at this point, because we use two lists,
464 * and to have reached the list, it must have gone through
465 * one full sync already.
466 */
467static void process_task_mortuary(void)
468{
Paul E. McKenney4369ef32006-01-08 01:01:35 -0800469 unsigned long flags;
470 LIST_HEAD(local_dead_tasks);
Robert Richter73185e02008-07-22 21:08:51 +0200471 struct task_struct *task;
472 struct task_struct *ttask;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700473
Paul E. McKenney4369ef32006-01-08 01:01:35 -0800474 spin_lock_irqsave(&task_mortuary, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700475
Paul E. McKenney4369ef32006-01-08 01:01:35 -0800476 list_splice_init(&dead_tasks, &local_dead_tasks);
477 list_splice_init(&dying_tasks, &dead_tasks);
478
479 spin_unlock_irqrestore(&task_mortuary, flags);
480
481 list_for_each_entry_safe(task, ttask, &local_dead_tasks, tasks) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700482 list_del(&task->tasks);
483 free_task(task);
484 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700485}
486
487
488static void mark_done(int cpu)
489{
490 int i;
491
492 cpu_set(cpu, marked_cpus);
493
494 for_each_online_cpu(i) {
495 if (!cpu_isset(i, marked_cpus))
496 return;
497 }
498
499 /* All CPUs have been processed at least once,
500 * we can process the mortuary once
501 */
502 process_task_mortuary();
503
504 cpus_clear(marked_cpus);
505}
506
507
508/* FIXME: this is not sufficient if we implement syscall barrier backtrace
509 * traversal, the code switch to sb_sample_start at first kernel enter/exit
510 * switch so we need a fifth state and some special handling in sync_buffer()
511 */
512typedef enum {
513 sb_bt_ignore = -2,
514 sb_buffer_start,
515 sb_bt_start,
516 sb_sample_start,
517} sync_buffer_state;
518
519/* Sync one of the CPU's buffers into the global event buffer.
520 * Here we need to go through each batch of samples punctuated
521 * by context switch notes, taking the task's mmap_sem and doing
522 * lookup in task->mm->mmap to convert EIP into dcookie/offset
523 * value.
524 */
525void sync_buffer(int cpu)
526{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700527 struct mm_struct *mm = NULL;
Robert Richterfd7826d2008-09-26 17:50:31 -0400528 struct mm_struct *oldmm;
Robert Richter73185e02008-07-22 21:08:51 +0200529 struct task_struct *new;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700530 unsigned long cookie = 0;
531 int in_kernel = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700532 sync_buffer_state state = sb_buffer_start;
Barry Kasindorf9b1f2612008-07-15 00:10:36 +0200533 unsigned int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700534 unsigned long available;
535
Markus Armbruster59cc1852006-06-25 05:47:33 -0700536 mutex_lock(&buffer_mutex);
Robert Richter73185e02008-07-22 21:08:51 +0200537
Linus Torvalds1da177e2005-04-16 15:20:36 -0700538 add_cpu_switch(cpu);
539
Robert Richterfbc9bf92008-12-04 16:27:00 +0100540 cpu_buffer_reset(cpu);
Robert Richter6dad8282008-12-09 01:21:32 +0100541 available = cpu_buffer_entries(cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700542
543 for (i = 0; i < available; ++i) {
Robert Richter6dad8282008-12-09 01:21:32 +0100544 struct op_sample *s = cpu_buffer_read_entry(cpu);
545 if (!s)
546 break;
Robert Richter73185e02008-07-22 21:08:51 +0200547
Linus Torvalds1da177e2005-04-16 15:20:36 -0700548 if (is_code(s->eip)) {
Robert Richterfd7826d2008-09-26 17:50:31 -0400549 switch (s->event) {
550 case 0:
551 case CPU_IS_KERNEL:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700552 /* kernel/userspace switch */
553 in_kernel = s->event;
554 if (state == sb_buffer_start)
555 state = sb_sample_start;
556 add_kernel_ctx_switch(s->event);
Robert Richterfd7826d2008-09-26 17:50:31 -0400557 break;
558 case CPU_TRACE_BEGIN:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700559 state = sb_bt_start;
560 add_trace_begin();
Robert Richterfd7826d2008-09-26 17:50:31 -0400561 break;
Robert Richter852402c2008-07-22 21:09:06 +0200562#ifdef CONFIG_OPROFILE_IBS
Robert Richterfd7826d2008-09-26 17:50:31 -0400563 case IBS_FETCH_BEGIN:
Barry Kasindorf345c2572008-07-22 21:08:54 +0200564 state = sb_bt_start;
Robert Richter6dad8282008-12-09 01:21:32 +0100565 add_ibs_begin(cpu, IBS_FETCH_CODE, mm);
Robert Richterfd7826d2008-09-26 17:50:31 -0400566 break;
567 case IBS_OP_BEGIN:
Barry Kasindorf345c2572008-07-22 21:08:54 +0200568 state = sb_bt_start;
Robert Richter6dad8282008-12-09 01:21:32 +0100569 add_ibs_begin(cpu, IBS_OP_CODE, mm);
Robert Richterfd7826d2008-09-26 17:50:31 -0400570 break;
Robert Richter852402c2008-07-22 21:09:06 +0200571#endif
Robert Richterfd7826d2008-09-26 17:50:31 -0400572 default:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700573 /* userspace context switch */
Robert Richterfd7826d2008-09-26 17:50:31 -0400574 oldmm = mm;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700575 new = (struct task_struct *)s->event;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700576 release_mm(oldmm);
577 mm = take_tasks_mm(new);
578 if (mm != oldmm)
579 cookie = get_exec_dcookie(mm);
580 add_user_ctx_switch(new, cookie);
Robert Richterfd7826d2008-09-26 17:50:31 -0400581 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700582 }
Robert Richter73185e02008-07-22 21:08:51 +0200583 } else if (state >= sb_bt_start &&
584 !add_sample(mm, s, in_kernel)) {
585 if (state == sb_bt_start) {
586 state = sb_bt_ignore;
587 atomic_inc(&oprofile_stats.bt_lost_no_mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700588 }
589 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700590 }
591 release_mm(mm);
592
593 mark_done(cpu);
594
Markus Armbruster59cc1852006-06-25 05:47:33 -0700595 mutex_unlock(&buffer_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700596}
Carl Lovea5598ca2008-10-14 23:37:01 +0000597
598/* The function can be used to add a buffer worth of data directly to
599 * the kernel buffer. The buffer is assumed to be a circular buffer.
600 * Take the entries from index start and end at index end, wrapping
601 * at max_entries.
602 */
603void oprofile_put_buff(unsigned long *buf, unsigned int start,
604 unsigned int stop, unsigned int max)
605{
606 int i;
607
608 i = start;
609
610 mutex_lock(&buffer_mutex);
611 while (i != stop) {
612 add_event_entry(buf[i++]);
613
614 if (i >= max)
615 i = 0;
616 }
617
618 mutex_unlock(&buffer_mutex);
619}
620