kernel/trace/trace_sysprof.c

   1 /*
   2  * trace stack traces
   3  *
   4  * Copyright (C) 2007 Steven Rostedt <srostedt@redhat.com>
   5  * Copyright (C) 2008 Ingo Molnar <mingo@redhat.com>
   6  * Copyright (C) 2004, 2005, Soeren Sandmann
   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 "trace.h"
  18
  19 static struct trace_array       *sysprof_trace;
  20 static int __read_mostly        tracer_enabled;
  21
  22 /*
  23  * 1 msec sample interval by default:
  24  */
  25 static unsigned long sample_period = 1000000;
  26 static const unsigned int sample_max_depth = 512;
  27
  28 static DEFINE_MUTEX(sample_timer_lock);
  29 /*
  30  * Per CPU hrtimers that do the profiling:
  31  */
  32 static DEFINE_PER_CPU(struct hrtimer, stack_trace_hrtimer);
  33
  34 struct stack_frame {
  35         const void __user       *next_fp;
  36         unsigned long           return_address;
  37 };
  38
  39 static int copy_stack_frame(const void __user *fp, struct stack_frame *frame)
  40 {
  41         int ret;
  42
  43         if (!access_ok(VERIFY_READ, fp, sizeof(*frame)))
  44                 return 0;
  45
  46         ret = 1;
  47         pagefault_disable();
  48         if (__copy_from_user_inatomic(frame, fp, sizeof(*frame)))
  49                 ret = 0;
  50         pagefault_enable();
  51
  52         return ret;
  53 }
  54
  55 static void timer_notify(struct pt_regs *regs, int cpu)
  56 {
  57         struct trace_array_cpu *data;
  58         struct stack_frame frame;
  59         struct trace_array *tr;
  60         const void __user *fp;
  61         int is_user;
  62         int i;
  63
  64         if (!regs)
  65                 return;
  66
  67         tr = sysprof_trace;
  68         data = tr->data[cpu];
  69         is_user = user_mode(regs);
  70
  71         if (!current || current->pid == 0)
  72                 return;
  73
  74         if (is_user && current->state != TASK_RUNNING)
  75                 return;
  76
  77         if (!is_user) {
  78                 /* kernel */
  79                 ftrace(tr, data, current->pid, 1, 0);
  80                 return;
  81
  82         }
  83
  84         trace_special(tr, data, 0, current->pid, regs->ip);
  85
  86         fp = (void __user *)regs->bp;
  87
  88         for (i = 0; i < sample_max_depth; i++) {
  89                 frame.next_fp = 0;
  90                 frame.return_address = 0;
  91                 if (!copy_stack_frame(fp, &frame))
  92                         break;
  93                 if ((unsigned long)fp < regs->sp)
  94                         break;
  95
  96                 trace_special(tr, data, 1, frame.return_address,
  97                               (unsigned long)fp);
  98                 fp = frame.next_fp;
  99         }
 100
 101         trace_special(tr, data, 2, current->pid, i);
 102
 103         /*
 104          * Special trace entry if we overflow the max depth:
 105          */
 106         if (i == sample_max_depth)
 107                 trace_special(tr, data, -1, -1, -1);
 108 }
 109
 110 static enum hrtimer_restart stack_trace_timer_fn(struct hrtimer *hrtimer)
 111 {
 112         /* trace here */
 113         timer_notify(get_irq_regs(), smp_processor_id());
 114
 115         hrtimer_forward_now(hrtimer, ns_to_ktime(sample_period));
 116
 117         return HRTIMER_RESTART;
 118 }
 119
 120 static void start_stack_timer(int cpu)
 121 {
 122         struct hrtimer *hrtimer = &per_cpu(stack_trace_hrtimer, cpu);
 123
 124         hrtimer_init(hrtimer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
 125         hrtimer->function = stack_trace_timer_fn;
 126         hrtimer->cb_mode = HRTIMER_CB_IRQSAFE_NO_SOFTIRQ;
 127
 128         hrtimer_start(hrtimer, ns_to_ktime(sample_period), HRTIMER_MODE_REL);
 129 }
 130
 131 static void start_stack_timers(void)
 132 {
 133         cpumask_t saved_mask = current->cpus_allowed;
 134         int cpu;
 135
 136         for_each_online_cpu(cpu) {
 137                 set_cpus_allowed_ptr(current, &cpumask_of_cpu(cpu));
 138                 start_stack_timer(cpu);
 139         }
 140         set_cpus_allowed_ptr(current, &saved_mask);
 141 }
 142
 143 static void stop_stack_timer(int cpu)
 144 {
 145         struct hrtimer *hrtimer = &per_cpu(stack_trace_hrtimer, cpu);
 146
 147         hrtimer_cancel(hrtimer);
 148 }
 149
 150 static void stop_stack_timers(void)
 151 {
 152         int cpu;
 153
 154         for_each_online_cpu(cpu)
 155                 stop_stack_timer(cpu);
 156 }
 157
 158 static void stack_reset(struct trace_array *tr)
 159 {
 160         int cpu;
 161
 162         tr->time_start = ftrace_now(tr->cpu);
 163
 164         for_each_online_cpu(cpu)
 165                 tracing_reset(tr->data[cpu]);
 166 }
 167
 168 static void start_stack_trace(struct trace_array *tr)
 169 {
 170         mutex_lock(&sample_timer_lock);
 171         stack_reset(tr);
 172         start_stack_timers();
 173         tracer_enabled = 1;
 174         mutex_unlock(&sample_timer_lock);
 175 }
 176
 177 static void stop_stack_trace(struct trace_array *tr)
 178 {
 179         mutex_lock(&sample_timer_lock);
 180         stop_stack_timers();
 181         tracer_enabled = 0;
 182         mutex_unlock(&sample_timer_lock);
 183 }
 184
 185 static void stack_trace_init(struct trace_array *tr)
 186 {
 187         sysprof_trace = tr;
 188
 189         if (tr->ctrl)
 190                 start_stack_trace(tr);
 191 }
 192
 193 static void stack_trace_reset(struct trace_array *tr)
 194 {
 195         if (tr->ctrl)
 196                 stop_stack_trace(tr);
 197 }
 198
 199 static void stack_trace_ctrl_update(struct trace_array *tr)
 200 {
 201         /* When starting a new trace, reset the buffers */
 202         if (tr->ctrl)
 203                 start_stack_trace(tr);
 204         else
 205                 stop_stack_trace(tr);
 206 }
 207
 208 static struct tracer stack_trace __read_mostly =
 209 {
 210         .name           = "sysprof",
 211         .init           = stack_trace_init,
 212         .reset          = stack_trace_reset,
 213         .ctrl_update    = stack_trace_ctrl_update,
 214 #ifdef CONFIG_FTRACE_SELFTEST
 215         .selftest    = trace_selftest_startup_sysprof,
 216 #endif
 217 };
 218
 219 __init static int init_stack_trace(void)
 220 {
 221         return register_tracer(&stack_trace);
 222 }
 223 device_initcall(init_stack_trace);
 224
 225 #define MAX_LONG_DIGITS 22
 226
 227 static ssize_t
 228 sysprof_sample_read(struct file *filp, char __user *ubuf,
 229                     size_t cnt, loff_t *ppos)
 230 {
 231         char buf[MAX_LONG_DIGITS];
 232         int r;
 233
 234         r = sprintf(buf, "%ld\n", nsecs_to_usecs(sample_period));
 235
 236         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
 237 }
 238
 239 static ssize_t
 240 sysprof_sample_write(struct file *filp, const char __user *ubuf,
 241                      size_t cnt, loff_t *ppos)
 242 {
 243         char buf[MAX_LONG_DIGITS];
 244         unsigned long val;
 245
 246         if (cnt > MAX_LONG_DIGITS-1)
 247                 cnt = MAX_LONG_DIGITS-1;
 248
 249         if (copy_from_user(&buf, ubuf, cnt))
 250                 return -EFAULT;
 251
 252         buf[cnt] = 0;
 253
 254         val = simple_strtoul(buf, NULL, 10);
 255         /*
 256          * Enforce a minimum sample period of 100 usecs:
 257          */
 258         if (val < 100)
 259                 val = 100;
 260
 261         mutex_lock(&sample_timer_lock);
 262         stop_stack_timers();
 263         sample_period = val * 1000;
 264         start_stack_timers();
 265         mutex_unlock(&sample_timer_lock);
 266
 267         return cnt;
 268 }
 269
 270 static struct file_operations sysprof_sample_fops = {
 271         .read           = sysprof_sample_read,
 272         .write          = sysprof_sample_write,
 273 };
 274
 275 void init_tracer_sysprof_debugfs(struct dentry *d_tracer)
 276 {
 277         struct dentry *entry;
 278
 279         entry = debugfs_create_file("sysprof_sample_period", 0644,
 280                         d_tracer, NULL, &sysprof_sample_fops);
 281         if (entry)
 282                 return;
 283         pr_warning("Could not create debugfs 'dyn_ftrace_total_info' entry\n");
 284 }