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>
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>
17 #include <asm/stacktrace.h>
21 static struct trace_array *sysprof_trace;
22 static int __read_mostly tracer_enabled;
25 * 1 msec sample interval by default:
27 static unsigned long sample_period = 1000000;
28 static const unsigned int sample_max_depth = 512;
30 static DEFINE_MUTEX(sample_timer_lock);
32 * Per CPU hrtimers that do the profiling:
34 static DEFINE_PER_CPU(struct hrtimer, stack_trace_hrtimer);
37 const void __user *next_fp;
38 unsigned long return_address;
41 static int copy_stack_frame(const void __user *fp, struct stack_frame *frame)
45 if (!access_ok(VERIFY_READ, fp, sizeof(*frame)))
50 if (__copy_from_user_inatomic(frame, fp, sizeof(*frame)))
57 struct backtrace_info {
58 struct trace_array_cpu *data;
59 struct trace_array *tr;
64 backtrace_warning_symbol(void *data, char *msg, unsigned long symbol)
69 static void backtrace_warning(void *data, char *msg)
74 static int backtrace_stack(void *data, char *name)
76 /* Don't bother with IRQ stacks for now */
80 static void backtrace_address(void *data, unsigned long addr, int reliable)
82 struct backtrace_info *info = data;
84 if (info->pos < sample_max_depth && reliable) {
85 __trace_special(info->tr, info->data, 1, addr, 0);
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,
99 trace_kernel(struct pt_regs *regs, struct trace_array *tr,
100 struct trace_array_cpu *data)
102 struct backtrace_info info;
110 __trace_special(info.tr, info.data, 1, regs->ip, 0);
112 stack = ((char *)regs + sizeof(struct pt_regs));
113 #ifdef CONFIG_FRAME_POINTER
119 dump_trace(NULL, regs, (void *)stack, bp, &backtrace_ops, &info);
124 static void timer_notify(struct pt_regs *regs, int cpu)
126 struct trace_array_cpu *data;
127 struct stack_frame frame;
128 struct trace_array *tr;
129 const void __user *fp;
137 data = tr->data[cpu];
138 is_user = user_mode(regs);
140 if (!current || current->pid == 0)
143 if (is_user && current->state != TASK_RUNNING)
146 __trace_special(tr, data, 0, 0, current->pid);
149 i = trace_kernel(regs, tr, data);
154 * Trace user stack if we are not a kernel thread
156 if (current->mm && i < sample_max_depth) {
157 regs = (struct pt_regs *)current->thread.sp0 - 1;
159 fp = (void __user *)regs->bp;
161 __trace_special(tr, data, 2, regs->ip, 0);
163 while (i < sample_max_depth) {
164 frame.next_fp = NULL;
165 frame.return_address = 0;
166 if (!copy_stack_frame(fp, &frame))
168 if ((unsigned long)fp < regs->sp)
171 __trace_special(tr, data, 2, frame.return_address,
181 * Special trace entry if we overflow the max depth:
183 if (i == sample_max_depth)
184 __trace_special(tr, data, -1, -1, -1);
186 __trace_special(tr, data, 3, current->pid, i);
189 static enum hrtimer_restart stack_trace_timer_fn(struct hrtimer *hrtimer)
192 timer_notify(get_irq_regs(), smp_processor_id());
194 hrtimer_forward_now(hrtimer, ns_to_ktime(sample_period));
196 return HRTIMER_RESTART;
199 static void start_stack_timer(int cpu)
201 struct hrtimer *hrtimer = &per_cpu(stack_trace_hrtimer, cpu);
203 hrtimer_init(hrtimer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
204 hrtimer->function = stack_trace_timer_fn;
205 hrtimer->cb_mode = HRTIMER_CB_IRQSAFE_PERCPU;
207 hrtimer_start(hrtimer, ns_to_ktime(sample_period), HRTIMER_MODE_REL);
210 static void start_stack_timers(void)
212 cpumask_t saved_mask = current->cpus_allowed;
215 for_each_online_cpu(cpu) {
216 set_cpus_allowed_ptr(current, &cpumask_of_cpu(cpu));
217 start_stack_timer(cpu);
219 set_cpus_allowed_ptr(current, &saved_mask);
222 static void stop_stack_timer(int cpu)
224 struct hrtimer *hrtimer = &per_cpu(stack_trace_hrtimer, cpu);
226 hrtimer_cancel(hrtimer);
229 static void stop_stack_timers(void)
233 for_each_online_cpu(cpu)
234 stop_stack_timer(cpu);
237 static void stack_reset(struct trace_array *tr)
241 tr->time_start = ftrace_now(tr->cpu);
243 for_each_online_cpu(cpu)
244 tracing_reset(tr, cpu);
247 static void start_stack_trace(struct trace_array *tr)
249 mutex_lock(&sample_timer_lock);
251 start_stack_timers();
253 mutex_unlock(&sample_timer_lock);
256 static void stop_stack_trace(struct trace_array *tr)
258 mutex_lock(&sample_timer_lock);
261 mutex_unlock(&sample_timer_lock);
264 static void stack_trace_init(struct trace_array *tr)
269 start_stack_trace(tr);
272 static void stack_trace_reset(struct trace_array *tr)
275 stop_stack_trace(tr);
278 static void stack_trace_ctrl_update(struct trace_array *tr)
280 /* When starting a new trace, reset the buffers */
282 start_stack_trace(tr);
284 stop_stack_trace(tr);
287 static struct tracer stack_trace __read_mostly =
290 .init = stack_trace_init,
291 .reset = stack_trace_reset,
292 .ctrl_update = stack_trace_ctrl_update,
293 #ifdef CONFIG_FTRACE_SELFTEST
294 .selftest = trace_selftest_startup_sysprof,
298 __init static int init_stack_trace(void)
300 return register_tracer(&stack_trace);
302 device_initcall(init_stack_trace);
304 #define MAX_LONG_DIGITS 22
307 sysprof_sample_read(struct file *filp, char __user *ubuf,
308 size_t cnt, loff_t *ppos)
310 char buf[MAX_LONG_DIGITS];
313 r = sprintf(buf, "%ld\n", nsecs_to_usecs(sample_period));
315 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
319 sysprof_sample_write(struct file *filp, const char __user *ubuf,
320 size_t cnt, loff_t *ppos)
322 char buf[MAX_LONG_DIGITS];
325 if (cnt > MAX_LONG_DIGITS-1)
326 cnt = MAX_LONG_DIGITS-1;
328 if (copy_from_user(&buf, ubuf, cnt))
333 val = simple_strtoul(buf, NULL, 10);
335 * Enforce a minimum sample period of 100 usecs:
340 mutex_lock(&sample_timer_lock);
342 sample_period = val * 1000;
343 start_stack_timers();
344 mutex_unlock(&sample_timer_lock);
349 static struct file_operations sysprof_sample_fops = {
350 .read = sysprof_sample_read,
351 .write = sysprof_sample_write,
354 void init_tracer_sysprof_debugfs(struct dentry *d_tracer)
356 struct dentry *entry;
358 entry = debugfs_create_file("sysprof_sample_period", 0644,
359 d_tracer, NULL, &sysprof_sample_fops);
362 pr_warning("Could not create debugfs 'dyn_ftrace_total_info' entry\n");