Commit | Line | Data |
---|---|---|
f06c3810 IM |
1 | /* |
2 | * trace stack traces | |
3 | * | |
4 | * Copyright (C) 2007 Steven Rostedt <srostedt@redhat.com> | |
5 | * Copyright (C) 2008 Ingo Molnar <mingo@redhat.com> | |
56a08bdc | 6 | * Copyright (C) 2004, 2005, Soeren Sandmann |
f06c3810 | 7 | */ |
f06c3810 | 8 | #include <linux/kallsyms.h> |
0075fa80 IM |
9 | #include <linux/debugfs.h> |
10 | #include <linux/hrtimer.h> | |
f06c3810 | 11 | #include <linux/uaccess.h> |
f06c3810 | 12 | #include <linux/ftrace.h> |
0075fa80 | 13 | #include <linux/module.h> |
56a08bdc | 14 | #include <linux/irq.h> |
0075fa80 | 15 | #include <linux/fs.h> |
f06c3810 | 16 | |
cd2134b1 SSP |
17 | #include <asm/stacktrace.h> |
18 | ||
f06c3810 IM |
19 | #include "trace.h" |
20 | ||
56a08bdc | 21 | static struct trace_array *sysprof_trace; |
f06c3810 IM |
22 | static int __read_mostly tracer_enabled; |
23 | ||
56a08bdc | 24 | /* |
d618b3e6 | 25 | * 1 msec sample interval by default: |
56a08bdc | 26 | */ |
d618b3e6 | 27 | static unsigned long sample_period = 1000000; |
842af315 | 28 | static const unsigned int sample_max_depth = 512; |
0075fa80 | 29 | |
d618b3e6 | 30 | static DEFINE_MUTEX(sample_timer_lock); |
0075fa80 IM |
31 | /* |
32 | * Per CPU hrtimers that do the profiling: | |
33 | */ | |
34 | static DEFINE_PER_CPU(struct hrtimer, stack_trace_hrtimer); | |
35 | ||
56a08bdc IM |
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 | { | |
9f6b4e3f IM |
43 | int ret; |
44 | ||
56a08bdc IM |
45 | if (!access_ok(VERIFY_READ, fp, sizeof(*frame))) |
46 | return 0; | |
47 | ||
9f6b4e3f IM |
48 | ret = 1; |
49 | pagefault_disable(); | |
50 | if (__copy_from_user_inatomic(frame, fp, sizeof(*frame))) | |
51 | ret = 0; | |
52 | pagefault_enable(); | |
56a08bdc | 53 | |
9f6b4e3f | 54 | return ret; |
56a08bdc IM |
55 | } |
56 | ||
cd2134b1 SSP |
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 struct pt_regs * | |
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 *user_stack; | |
105 | char *stack; | |
106 | ||
107 | info.tr = tr; | |
108 | info.data = data; | |
109 | info.pos = 1; | |
110 | ||
111 | __trace_special(info.tr, info.data, 1, regs->ip, 0); | |
112 | ||
113 | stack = ((char *)regs + sizeof(struct pt_regs)); | |
114 | #ifdef CONFIG_FRAME_POINTER | |
115 | bp = regs->bp; | |
116 | #else | |
117 | bp = 0; | |
118 | #endif | |
119 | ||
120 | dump_trace(NULL, regs, (void *)stack, bp, &backtrace_ops, &info); | |
121 | ||
122 | /* Now trace the user stack */ | |
123 | user_stack = ((char *)current->thread.sp0 - sizeof(struct pt_regs)); | |
124 | ||
125 | return (struct pt_regs *)user_stack; | |
126 | } | |
127 | ||
56a08bdc IM |
128 | static void timer_notify(struct pt_regs *regs, int cpu) |
129 | { | |
56a08bdc IM |
130 | struct trace_array_cpu *data; |
131 | struct stack_frame frame; | |
132 | struct trace_array *tr; | |
9f6b4e3f | 133 | const void __user *fp; |
56a08bdc IM |
134 | int is_user; |
135 | int i; | |
136 | ||
137 | if (!regs) | |
138 | return; | |
139 | ||
140 | tr = sysprof_trace; | |
141 | data = tr->data[cpu]; | |
142 | is_user = user_mode(regs); | |
143 | ||
144 | if (!current || current->pid == 0) | |
145 | return; | |
146 | ||
147 | if (is_user && current->state != TASK_RUNNING) | |
148 | return; | |
149 | ||
cd2134b1 | 150 | __trace_special(tr, data, 0, 0, current->pid); |
56a08bdc | 151 | |
cd2134b1 SSP |
152 | if (!is_user) |
153 | regs = trace_kernel(regs, tr, data); | |
56a08bdc | 154 | |
9f6b4e3f | 155 | fp = (void __user *)regs->bp; |
56a08bdc | 156 | |
cd2134b1 SSP |
157 | __trace_special(tr, data, 2, regs->ip, 0); |
158 | ||
842af315 | 159 | for (i = 0; i < sample_max_depth; i++) { |
9f6b4e3f IM |
160 | frame.next_fp = 0; |
161 | frame.return_address = 0; | |
162 | if (!copy_stack_frame(fp, &frame)) | |
56a08bdc | 163 | break; |
9f6b4e3f | 164 | if ((unsigned long)fp < regs->sp) |
56a08bdc IM |
165 | break; |
166 | ||
cd2134b1 | 167 | __trace_special(tr, data, 2, frame.return_address, |
9f6b4e3f IM |
168 | (unsigned long)fp); |
169 | fp = frame.next_fp; | |
56a08bdc IM |
170 | } |
171 | ||
cd2134b1 | 172 | __trace_special(tr, data, 3, current->pid, i); |
56a08bdc | 173 | |
9f6b4e3f IM |
174 | /* |
175 | * Special trace entry if we overflow the max depth: | |
176 | */ | |
842af315 | 177 | if (i == sample_max_depth) |
9caee613 | 178 | __trace_special(tr, data, -1, -1, -1); |
56a08bdc IM |
179 | } |
180 | ||
0075fa80 IM |
181 | static enum hrtimer_restart stack_trace_timer_fn(struct hrtimer *hrtimer) |
182 | { | |
183 | /* trace here */ | |
56a08bdc | 184 | timer_notify(get_irq_regs(), smp_processor_id()); |
0075fa80 IM |
185 | |
186 | hrtimer_forward_now(hrtimer, ns_to_ktime(sample_period)); | |
187 | ||
188 | return HRTIMER_RESTART; | |
189 | } | |
190 | ||
191 | static void start_stack_timer(int cpu) | |
192 | { | |
193 | struct hrtimer *hrtimer = &per_cpu(stack_trace_hrtimer, cpu); | |
194 | ||
195 | hrtimer_init(hrtimer, CLOCK_MONOTONIC, HRTIMER_MODE_REL); | |
196 | hrtimer->function = stack_trace_timer_fn; | |
197 | hrtimer->cb_mode = HRTIMER_CB_IRQSAFE_NO_SOFTIRQ; | |
198 | ||
199 | hrtimer_start(hrtimer, ns_to_ktime(sample_period), HRTIMER_MODE_REL); | |
200 | } | |
201 | ||
202 | static void start_stack_timers(void) | |
203 | { | |
204 | cpumask_t saved_mask = current->cpus_allowed; | |
205 | int cpu; | |
206 | ||
207 | for_each_online_cpu(cpu) { | |
208 | set_cpus_allowed_ptr(current, &cpumask_of_cpu(cpu)); | |
209 | start_stack_timer(cpu); | |
0075fa80 IM |
210 | } |
211 | set_cpus_allowed_ptr(current, &saved_mask); | |
212 | } | |
213 | ||
214 | static void stop_stack_timer(int cpu) | |
215 | { | |
216 | struct hrtimer *hrtimer = &per_cpu(stack_trace_hrtimer, cpu); | |
217 | ||
218 | hrtimer_cancel(hrtimer); | |
0075fa80 IM |
219 | } |
220 | ||
221 | static void stop_stack_timers(void) | |
222 | { | |
223 | int cpu; | |
224 | ||
225 | for_each_online_cpu(cpu) | |
226 | stop_stack_timer(cpu); | |
227 | } | |
228 | ||
ada6b835 | 229 | static void stack_reset(struct trace_array *tr) |
f06c3810 IM |
230 | { |
231 | int cpu; | |
232 | ||
233 | tr->time_start = ftrace_now(tr->cpu); | |
234 | ||
235 | for_each_online_cpu(cpu) | |
236 | tracing_reset(tr->data[cpu]); | |
237 | } | |
238 | ||
ada6b835 | 239 | static void start_stack_trace(struct trace_array *tr) |
f06c3810 | 240 | { |
d618b3e6 | 241 | mutex_lock(&sample_timer_lock); |
f06c3810 | 242 | stack_reset(tr); |
0075fa80 | 243 | start_stack_timers(); |
f06c3810 | 244 | tracer_enabled = 1; |
d618b3e6 | 245 | mutex_unlock(&sample_timer_lock); |
f06c3810 IM |
246 | } |
247 | ||
ada6b835 | 248 | static void stop_stack_trace(struct trace_array *tr) |
f06c3810 | 249 | { |
d618b3e6 | 250 | mutex_lock(&sample_timer_lock); |
0075fa80 | 251 | stop_stack_timers(); |
f06c3810 | 252 | tracer_enabled = 0; |
d618b3e6 | 253 | mutex_unlock(&sample_timer_lock); |
f06c3810 IM |
254 | } |
255 | ||
ada6b835 | 256 | static void stack_trace_init(struct trace_array *tr) |
f06c3810 | 257 | { |
56a08bdc | 258 | sysprof_trace = tr; |
f06c3810 IM |
259 | |
260 | if (tr->ctrl) | |
261 | start_stack_trace(tr); | |
262 | } | |
263 | ||
ada6b835 | 264 | static void stack_trace_reset(struct trace_array *tr) |
f06c3810 IM |
265 | { |
266 | if (tr->ctrl) | |
267 | stop_stack_trace(tr); | |
268 | } | |
269 | ||
270 | static void stack_trace_ctrl_update(struct trace_array *tr) | |
271 | { | |
272 | /* When starting a new trace, reset the buffers */ | |
273 | if (tr->ctrl) | |
274 | start_stack_trace(tr); | |
275 | else | |
276 | stop_stack_trace(tr); | |
277 | } | |
278 | ||
279 | static struct tracer stack_trace __read_mostly = | |
280 | { | |
281 | .name = "sysprof", | |
282 | .init = stack_trace_init, | |
283 | .reset = stack_trace_reset, | |
284 | .ctrl_update = stack_trace_ctrl_update, | |
285 | #ifdef CONFIG_FTRACE_SELFTEST | |
a6dd24f8 | 286 | .selftest = trace_selftest_startup_sysprof, |
f06c3810 IM |
287 | #endif |
288 | }; | |
289 | ||
290 | __init static int init_stack_trace(void) | |
291 | { | |
292 | return register_tracer(&stack_trace); | |
293 | } | |
294 | device_initcall(init_stack_trace); | |
d618b3e6 IM |
295 | |
296 | #define MAX_LONG_DIGITS 22 | |
297 | ||
298 | static ssize_t | |
299 | sysprof_sample_read(struct file *filp, char __user *ubuf, | |
300 | size_t cnt, loff_t *ppos) | |
301 | { | |
302 | char buf[MAX_LONG_DIGITS]; | |
303 | int r; | |
304 | ||
305 | r = sprintf(buf, "%ld\n", nsecs_to_usecs(sample_period)); | |
306 | ||
307 | return simple_read_from_buffer(ubuf, cnt, ppos, buf, r); | |
308 | } | |
309 | ||
310 | static ssize_t | |
311 | sysprof_sample_write(struct file *filp, const char __user *ubuf, | |
312 | size_t cnt, loff_t *ppos) | |
313 | { | |
314 | char buf[MAX_LONG_DIGITS]; | |
315 | unsigned long val; | |
316 | ||
317 | if (cnt > MAX_LONG_DIGITS-1) | |
318 | cnt = MAX_LONG_DIGITS-1; | |
319 | ||
320 | if (copy_from_user(&buf, ubuf, cnt)) | |
321 | return -EFAULT; | |
322 | ||
323 | buf[cnt] = 0; | |
324 | ||
325 | val = simple_strtoul(buf, NULL, 10); | |
326 | /* | |
327 | * Enforce a minimum sample period of 100 usecs: | |
328 | */ | |
329 | if (val < 100) | |
330 | val = 100; | |
331 | ||
332 | mutex_lock(&sample_timer_lock); | |
333 | stop_stack_timers(); | |
334 | sample_period = val * 1000; | |
335 | start_stack_timers(); | |
336 | mutex_unlock(&sample_timer_lock); | |
337 | ||
338 | return cnt; | |
339 | } | |
340 | ||
341 | static struct file_operations sysprof_sample_fops = { | |
342 | .read = sysprof_sample_read, | |
343 | .write = sysprof_sample_write, | |
344 | }; | |
345 | ||
346 | void init_tracer_sysprof_debugfs(struct dentry *d_tracer) | |
347 | { | |
348 | struct dentry *entry; | |
349 | ||
350 | entry = debugfs_create_file("sysprof_sample_period", 0644, | |
351 | d_tracer, NULL, &sysprof_sample_fops); | |
352 | if (entry) | |
353 | return; | |
354 | pr_warning("Could not create debugfs 'dyn_ftrace_total_info' entry\n"); | |
355 | } |