Merge branch 'sched/core' into cpus4096
[linux-2.6] / kernel / trace / trace_sysprof.c
1 /*
2  * trace stack traces
3  *
4  * Copyright (C) 2004-2008, Soeren Sandmann
5  * Copyright (C) 2007 Steven Rostedt <srostedt@redhat.com>
6  * Copyright (C) 2008 Ingo Molnar <mingo@redhat.com>
7  */
8 #include <linux/kallsyms.h>
9 #include <linux/debugfs.h>
10 #include <linux/hrtimer.h>
11 #include <linux/uaccess.h>
12 #include <linux/ftrace.h>
13 #include <linux/module.h>
14 #include <linux/irq.h>
15 #include <linux/fs.h>
16
17 #include <asm/stacktrace.h>
18
19 #include "trace.h"
20
21 static struct trace_array       *sysprof_trace;
22 static int __read_mostly        tracer_enabled;
23
24 /*
25  * 1 msec sample interval by default:
26  */
27 static unsigned long sample_period = 1000000;
28 static const unsigned int sample_max_depth = 512;
29
30 static DEFINE_MUTEX(sample_timer_lock);
31 /*
32  * Per CPU hrtimers that do the profiling:
33  */
34 static DEFINE_PER_CPU(struct hrtimer, stack_trace_hrtimer);
35
36 struct stack_frame {
37         const void __user       *next_fp;
38         unsigned long           return_address;
39 };
40
41 static int copy_stack_frame(const void __user *fp, struct stack_frame *frame)
42 {
43         int ret;
44
45         if (!access_ok(VERIFY_READ, fp, sizeof(*frame)))
46                 return 0;
47
48         ret = 1;
49         pagefault_disable();
50         if (__copy_from_user_inatomic(frame, fp, sizeof(*frame)))
51                 ret = 0;
52         pagefault_enable();
53
54         return ret;
55 }
56
57 struct backtrace_info {
58         struct trace_array_cpu  *data;
59         struct trace_array      *tr;
60         int                     pos;
61 };
62
63 static void
64 backtrace_warning_symbol(void *data, char *msg, unsigned long symbol)
65 {
66         /* Ignore warnings */
67 }
68
69 static void backtrace_warning(void *data, char *msg)
70 {
71         /* Ignore warnings */
72 }
73
74 static int backtrace_stack(void *data, char *name)
75 {
76         /* Don't bother with IRQ stacks for now */
77         return -1;
78 }
79
80 static void backtrace_address(void *data, unsigned long addr, int reliable)
81 {
82         struct backtrace_info *info = data;
83
84         if (info->pos < sample_max_depth && reliable) {
85                 __trace_special(info->tr, info->data, 1, addr, 0);
86
87                 info->pos++;
88         }
89 }
90
91 const static struct stacktrace_ops backtrace_ops = {
92         .warning                = backtrace_warning,
93         .warning_symbol         = backtrace_warning_symbol,
94         .stack                  = backtrace_stack,
95         .address                = backtrace_address,
96 };
97
98 static int
99 trace_kernel(struct pt_regs *regs, struct trace_array *tr,
100              struct trace_array_cpu *data)
101 {
102         struct backtrace_info info;
103         unsigned long bp;
104         char *stack;
105
106         info.tr = tr;
107         info.data = data;
108         info.pos = 1;
109
110         __trace_special(info.tr, info.data, 1, regs->ip, 0);
111
112         stack = ((char *)regs + sizeof(struct pt_regs));
113 #ifdef CONFIG_FRAME_POINTER
114         bp = regs->bp;
115 #else
116         bp = 0;
117 #endif
118
119         dump_trace(NULL, regs, (void *)stack, bp, &backtrace_ops, &info);
120
121         return info.pos;
122 }
123
124 static void timer_notify(struct pt_regs *regs, int cpu)
125 {
126         struct trace_array_cpu *data;
127         struct stack_frame frame;
128         struct trace_array *tr;
129         const void __user *fp;
130         int is_user;
131         int i;
132
133         if (!regs)
134                 return;
135
136         tr = sysprof_trace;
137         data = tr->data[cpu];
138         is_user = user_mode(regs);
139
140         if (!current || current->pid == 0)
141                 return;
142
143         if (is_user && current->state != TASK_RUNNING)
144                 return;
145
146         __trace_special(tr, data, 0, 0, current->pid);
147
148         if (!is_user)
149                 i = trace_kernel(regs, tr, data);
150         else
151                 i = 0;
152
153         /*
154          * Trace user stack if we are not a kernel thread
155          */
156         if (current->mm && i < sample_max_depth) {
157                 regs = (struct pt_regs *)current->thread.sp0 - 1;
158
159                 fp = (void __user *)regs->bp;
160
161                 __trace_special(tr, data, 2, regs->ip, 0);
162
163                 while (i < sample_max_depth) {
164                         frame.next_fp = NULL;
165                         frame.return_address = 0;
166                         if (!copy_stack_frame(fp, &frame))
167                                 break;
168                         if ((unsigned long)fp < regs->sp)
169                                 break;
170
171                         __trace_special(tr, data, 2, frame.return_address,
172                                         (unsigned long)fp);
173                         fp = frame.next_fp;
174
175                         i++;
176                 }
177
178         }
179
180         /*
181          * Special trace entry if we overflow the max depth:
182          */
183         if (i == sample_max_depth)
184                 __trace_special(tr, data, -1, -1, -1);
185
186         __trace_special(tr, data, 3, current->pid, i);
187 }
188
189 static enum hrtimer_restart stack_trace_timer_fn(struct hrtimer *hrtimer)
190 {
191         /* trace here */
192         timer_notify(get_irq_regs(), smp_processor_id());
193
194         hrtimer_forward_now(hrtimer, ns_to_ktime(sample_period));
195
196         return HRTIMER_RESTART;
197 }
198
199 static void start_stack_timer(int cpu)
200 {
201         struct hrtimer *hrtimer = &per_cpu(stack_trace_hrtimer, cpu);
202
203         hrtimer_init(hrtimer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
204         hrtimer->function = stack_trace_timer_fn;
205         hrtimer->cb_mode = HRTIMER_CB_IRQSAFE_PERCPU;
206
207         hrtimer_start(hrtimer, ns_to_ktime(sample_period), HRTIMER_MODE_REL);
208 }
209
210 static void start_stack_timers(void)
211 {
212         cpumask_t saved_mask = current->cpus_allowed;
213         int cpu;
214
215         for_each_online_cpu(cpu) {
216                 set_cpus_allowed_ptr(current, &cpumask_of_cpu(cpu));
217                 start_stack_timer(cpu);
218         }
219         set_cpus_allowed_ptr(current, &saved_mask);
220 }
221
222 static void stop_stack_timer(int cpu)
223 {
224         struct hrtimer *hrtimer = &per_cpu(stack_trace_hrtimer, cpu);
225
226         hrtimer_cancel(hrtimer);
227 }
228
229 static void stop_stack_timers(void)
230 {
231         int cpu;
232
233         for_each_online_cpu(cpu)
234                 stop_stack_timer(cpu);
235 }
236
237 static void stack_reset(struct trace_array *tr)
238 {
239         int cpu;
240
241         tr->time_start = ftrace_now(tr->cpu);
242
243         for_each_online_cpu(cpu)
244                 tracing_reset(tr, cpu);
245 }
246
247 static void start_stack_trace(struct trace_array *tr)
248 {
249         mutex_lock(&sample_timer_lock);
250         stack_reset(tr);
251         start_stack_timers();
252         tracer_enabled = 1;
253         mutex_unlock(&sample_timer_lock);
254 }
255
256 static void stop_stack_trace(struct trace_array *tr)
257 {
258         mutex_lock(&sample_timer_lock);
259         stop_stack_timers();
260         tracer_enabled = 0;
261         mutex_unlock(&sample_timer_lock);
262 }
263
264 static int stack_trace_init(struct trace_array *tr)
265 {
266         sysprof_trace = tr;
267
268         start_stack_trace(tr);
269         return 0;
270 }
271
272 static void stack_trace_reset(struct trace_array *tr)
273 {
274         stop_stack_trace(tr);
275 }
276
277 static struct tracer stack_trace __read_mostly =
278 {
279         .name           = "sysprof",
280         .init           = stack_trace_init,
281         .reset          = stack_trace_reset,
282 #ifdef CONFIG_FTRACE_SELFTEST
283         .selftest    = trace_selftest_startup_sysprof,
284 #endif
285 };
286
287 __init static int init_stack_trace(void)
288 {
289         return register_tracer(&stack_trace);
290 }
291 device_initcall(init_stack_trace);
292
293 #define MAX_LONG_DIGITS 22
294
295 static ssize_t
296 sysprof_sample_read(struct file *filp, char __user *ubuf,
297                     size_t cnt, loff_t *ppos)
298 {
299         char buf[MAX_LONG_DIGITS];
300         int r;
301
302         r = sprintf(buf, "%ld\n", nsecs_to_usecs(sample_period));
303
304         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
305 }
306
307 static ssize_t
308 sysprof_sample_write(struct file *filp, const char __user *ubuf,
309                      size_t cnt, loff_t *ppos)
310 {
311         char buf[MAX_LONG_DIGITS];
312         unsigned long val;
313
314         if (cnt > MAX_LONG_DIGITS-1)
315                 cnt = MAX_LONG_DIGITS-1;
316
317         if (copy_from_user(&buf, ubuf, cnt))
318                 return -EFAULT;
319
320         buf[cnt] = 0;
321
322         val = simple_strtoul(buf, NULL, 10);
323         /*
324          * Enforce a minimum sample period of 100 usecs:
325          */
326         if (val < 100)
327                 val = 100;
328
329         mutex_lock(&sample_timer_lock);
330         stop_stack_timers();
331         sample_period = val * 1000;
332         start_stack_timers();
333         mutex_unlock(&sample_timer_lock);
334
335         return cnt;
336 }
337
338 static struct file_operations sysprof_sample_fops = {
339         .read           = sysprof_sample_read,
340         .write          = sysprof_sample_write,
341 };
342
343 void init_tracer_sysprof_debugfs(struct dentry *d_tracer)
344 {
345         struct dentry *entry;
346
347         entry = debugfs_create_file("sysprof_sample_period", 0644,
348                         d_tracer, NULL, &sysprof_sample_fops);
349         if (entry)
350                 return;
351         pr_warning("Could not create debugfs 'dyn_ftrace_total_info' entry\n");
352 }