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 static const struct stacktrace_ops backtrace_ops = {
92 .warning = backtrace_warning,
93 .warning_symbol = backtrace_warning_symbol,
94 .stack = backtrace_stack,
95 .address = backtrace_address,
96 .walk_stack = print_context_stack,
100 trace_kernel(struct pt_regs *regs, struct trace_array *tr,
101 struct trace_array_cpu *data)
103 struct backtrace_info info;
111 __trace_special(info.tr, info.data, 1, regs->ip, 0);
113 stack = ((char *)regs + sizeof(struct pt_regs));
114 #ifdef CONFIG_FRAME_POINTER
120 dump_trace(NULL, regs, (void *)stack, bp, &backtrace_ops, &info);
125 static void timer_notify(struct pt_regs *regs, int cpu)
127 struct trace_array_cpu *data;
128 struct stack_frame frame;
129 struct trace_array *tr;
130 const void __user *fp;
138 data = tr->data[cpu];
139 is_user = user_mode(regs);
141 if (!current || current->pid == 0)
144 if (is_user && current->state != TASK_RUNNING)
147 __trace_special(tr, data, 0, 0, current->pid);
150 i = trace_kernel(regs, tr, data);
155 * Trace user stack if we are not a kernel thread
157 if (current->mm && i < sample_max_depth) {
158 regs = (struct pt_regs *)current->thread.sp0 - 1;
160 fp = (void __user *)regs->bp;
162 __trace_special(tr, data, 2, regs->ip, 0);
164 while (i < sample_max_depth) {
165 frame.next_fp = NULL;
166 frame.return_address = 0;
167 if (!copy_stack_frame(fp, &frame))
169 if ((unsigned long)fp < regs->sp)
172 __trace_special(tr, data, 2, frame.return_address,
182 * Special trace entry if we overflow the max depth:
184 if (i == sample_max_depth)
185 __trace_special(tr, data, -1, -1, -1);
187 __trace_special(tr, data, 3, current->pid, i);
190 static enum hrtimer_restart stack_trace_timer_fn(struct hrtimer *hrtimer)
193 timer_notify(get_irq_regs(), smp_processor_id());
195 hrtimer_forward_now(hrtimer, ns_to_ktime(sample_period));
197 return HRTIMER_RESTART;
200 static void start_stack_timer(void *unused)
202 struct hrtimer *hrtimer = &__get_cpu_var(stack_trace_hrtimer);
204 hrtimer_init(hrtimer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
205 hrtimer->function = stack_trace_timer_fn;
207 hrtimer_start(hrtimer, ns_to_ktime(sample_period),
208 HRTIMER_MODE_REL_PINNED);
211 static void start_stack_timers(void)
213 on_each_cpu(start_stack_timer, NULL, 1);
216 static void stop_stack_timer(int cpu)
218 struct hrtimer *hrtimer = &per_cpu(stack_trace_hrtimer, cpu);
220 hrtimer_cancel(hrtimer);
223 static void stop_stack_timers(void)
227 for_each_online_cpu(cpu)
228 stop_stack_timer(cpu);
231 static void stop_stack_trace(struct trace_array *tr)
233 mutex_lock(&sample_timer_lock);
236 mutex_unlock(&sample_timer_lock);
239 static int stack_trace_init(struct trace_array *tr)
243 tracing_start_cmdline_record();
245 mutex_lock(&sample_timer_lock);
246 start_stack_timers();
248 mutex_unlock(&sample_timer_lock);
252 static void stack_trace_reset(struct trace_array *tr)
254 tracing_stop_cmdline_record();
255 stop_stack_trace(tr);
258 static struct tracer stack_trace __read_mostly =
261 .init = stack_trace_init,
262 .reset = stack_trace_reset,
263 #ifdef CONFIG_FTRACE_SELFTEST
264 .selftest = trace_selftest_startup_sysprof,
268 __init static int init_stack_trace(void)
270 return register_tracer(&stack_trace);
272 device_initcall(init_stack_trace);
274 #define MAX_LONG_DIGITS 22
277 sysprof_sample_read(struct file *filp, char __user *ubuf,
278 size_t cnt, loff_t *ppos)
280 char buf[MAX_LONG_DIGITS];
283 r = sprintf(buf, "%ld\n", nsecs_to_usecs(sample_period));
285 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
289 sysprof_sample_write(struct file *filp, const char __user *ubuf,
290 size_t cnt, loff_t *ppos)
292 char buf[MAX_LONG_DIGITS];
295 if (cnt > MAX_LONG_DIGITS-1)
296 cnt = MAX_LONG_DIGITS-1;
298 if (copy_from_user(&buf, ubuf, cnt))
303 val = simple_strtoul(buf, NULL, 10);
305 * Enforce a minimum sample period of 100 usecs:
310 mutex_lock(&sample_timer_lock);
312 sample_period = val * 1000;
313 start_stack_timers();
314 mutex_unlock(&sample_timer_lock);
319 static const struct file_operations sysprof_sample_fops = {
320 .read = sysprof_sample_read,
321 .write = sysprof_sample_write,
324 void init_tracer_sysprof_debugfs(struct dentry *d_tracer)
327 trace_create_file("sysprof_sample_period", 0644,
328 d_tracer, NULL, &sysprof_sample_fops);