perf_event: Provide vmalloc() based mmap() backing
[safe/jmp/linux-2.6] / kernel / trace / trace_sysprof.c
index 6c139bc..f669396 100644 (file)
 /*
  * trace stack traces
  *
+ * Copyright (C) 2004-2008, Soeren Sandmann
  * Copyright (C) 2007 Steven Rostedt <srostedt@redhat.com>
  * Copyright (C) 2008 Ingo Molnar <mingo@redhat.com>
- *
  */
-#include <linux/module.h>
-#include <linux/fs.h>
-#include <linux/debugfs.h>
 #include <linux/kallsyms.h>
+#include <linux/debugfs.h>
+#include <linux/hrtimer.h>
 #include <linux/uaccess.h>
-#include <linux/marker.h>
 #include <linux/ftrace.h>
+#include <linux/module.h>
+#include <linux/irq.h>
+#include <linux/fs.h>
+
+#include <asm/stacktrace.h>
 
 #include "trace.h"
 
-static struct trace_array      *ctx_trace;
+static struct trace_array      *sysprof_trace;
 static int __read_mostly       tracer_enabled;
 
-static notrace void stack_reset(struct trace_array *tr)
+/*
+ * 1 msec sample interval by default:
+ */
+static unsigned long sample_period = 1000000;
+static const unsigned int sample_max_depth = 512;
+
+static DEFINE_MUTEX(sample_timer_lock);
+/*
+ * Per CPU hrtimers that do the profiling:
+ */
+static DEFINE_PER_CPU(struct hrtimer, stack_trace_hrtimer);
+
+struct stack_frame {
+       const void __user       *next_fp;
+       unsigned long           return_address;
+};
+
+static int copy_stack_frame(const void __user *fp, struct stack_frame *frame)
 {
-       int cpu;
+       int ret;
 
-       tr->time_start = ftrace_now(tr->cpu);
+       if (!access_ok(VERIFY_READ, fp, sizeof(*frame)))
+               return 0;
 
-       for_each_online_cpu(cpu)
-               tracing_reset(tr->data[cpu]);
+       ret = 1;
+       pagefault_disable();
+       if (__copy_from_user_inatomic(frame, fp, sizeof(*frame)))
+               ret = 0;
+       pagefault_enable();
+
+       return ret;
 }
 
-static notrace void start_stack_trace(struct trace_array *tr)
+struct backtrace_info {
+       struct trace_array_cpu  *data;
+       struct trace_array      *tr;
+       int                     pos;
+};
+
+static void
+backtrace_warning_symbol(void *data, char *msg, unsigned long symbol)
 {
-       stack_reset(tr);
-       tracer_enabled = 1;
+       /* Ignore warnings */
 }
 
-static notrace void stop_stack_trace(struct trace_array *tr)
+static void backtrace_warning(void *data, char *msg)
 {
-       tracer_enabled = 0;
+       /* Ignore warnings */
+}
+
+static int backtrace_stack(void *data, char *name)
+{
+       /* Don't bother with IRQ stacks for now */
+       return -1;
 }
 
-static notrace void stack_trace_init(struct trace_array *tr)
+static void backtrace_address(void *data, unsigned long addr, int reliable)
 {
-       ctx_trace = tr;
+       struct backtrace_info *info = data;
+
+       if (info->pos < sample_max_depth && reliable) {
+               __trace_special(info->tr, info->data, 1, addr, 0);
 
-       if (tr->ctrl)
-               start_stack_trace(tr);
+               info->pos++;
+       }
 }
 
-static notrace void stack_trace_reset(struct trace_array *tr)
+static const struct stacktrace_ops backtrace_ops = {
+       .warning                = backtrace_warning,
+       .warning_symbol         = backtrace_warning_symbol,
+       .stack                  = backtrace_stack,
+       .address                = backtrace_address,
+};
+
+static int
+trace_kernel(struct pt_regs *regs, struct trace_array *tr,
+            struct trace_array_cpu *data)
 {
-       if (tr->ctrl)
-               stop_stack_trace(tr);
+       struct backtrace_info info;
+       unsigned long bp;
+       char *stack;
+
+       info.tr = tr;
+       info.data = data;
+       info.pos = 1;
+
+       __trace_special(info.tr, info.data, 1, regs->ip, 0);
+
+       stack = ((char *)regs + sizeof(struct pt_regs));
+#ifdef CONFIG_FRAME_POINTER
+       bp = regs->bp;
+#else
+       bp = 0;
+#endif
+
+       dump_trace(NULL, regs, (void *)stack, bp, &backtrace_ops, &info);
+
+       return info.pos;
 }
 
-static void stack_trace_ctrl_update(struct trace_array *tr)
+static void timer_notify(struct pt_regs *regs, int cpu)
 {
-       /* When starting a new trace, reset the buffers */
-       if (tr->ctrl)
-               start_stack_trace(tr);
+       struct trace_array_cpu *data;
+       struct stack_frame frame;
+       struct trace_array *tr;
+       const void __user *fp;
+       int is_user;
+       int i;
+
+       if (!regs)
+               return;
+
+       tr = sysprof_trace;
+       data = tr->data[cpu];
+       is_user = user_mode(regs);
+
+       if (!current || current->pid == 0)
+               return;
+
+       if (is_user && current->state != TASK_RUNNING)
+               return;
+
+       __trace_special(tr, data, 0, 0, current->pid);
+
+       if (!is_user)
+               i = trace_kernel(regs, tr, data);
        else
-               stop_stack_trace(tr);
+               i = 0;
+
+       /*
+        * Trace user stack if we are not a kernel thread
+        */
+       if (current->mm && i < sample_max_depth) {
+               regs = (struct pt_regs *)current->thread.sp0 - 1;
+
+               fp = (void __user *)regs->bp;
+
+               __trace_special(tr, data, 2, regs->ip, 0);
+
+               while (i < sample_max_depth) {
+                       frame.next_fp = NULL;
+                       frame.return_address = 0;
+                       if (!copy_stack_frame(fp, &frame))
+                               break;
+                       if ((unsigned long)fp < regs->sp)
+                               break;
+
+                       __trace_special(tr, data, 2, frame.return_address,
+                                       (unsigned long)fp);
+                       fp = frame.next_fp;
+
+                       i++;
+               }
+
+       }
+
+       /*
+        * Special trace entry if we overflow the max depth:
+        */
+       if (i == sample_max_depth)
+               __trace_special(tr, data, -1, -1, -1);
+
+       __trace_special(tr, data, 3, current->pid, i);
+}
+
+static enum hrtimer_restart stack_trace_timer_fn(struct hrtimer *hrtimer)
+{
+       /* trace here */
+       timer_notify(get_irq_regs(), smp_processor_id());
+
+       hrtimer_forward_now(hrtimer, ns_to_ktime(sample_period));
+
+       return HRTIMER_RESTART;
+}
+
+static void start_stack_timer(void *unused)
+{
+       struct hrtimer *hrtimer = &__get_cpu_var(stack_trace_hrtimer);
+
+       hrtimer_init(hrtimer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
+       hrtimer->function = stack_trace_timer_fn;
+
+       hrtimer_start(hrtimer, ns_to_ktime(sample_period),
+                     HRTIMER_MODE_REL_PINNED);
+}
+
+static void start_stack_timers(void)
+{
+       on_each_cpu(start_stack_timer, NULL, 1);
+}
+
+static void stop_stack_timer(int cpu)
+{
+       struct hrtimer *hrtimer = &per_cpu(stack_trace_hrtimer, cpu);
+
+       hrtimer_cancel(hrtimer);
+}
+
+static void stop_stack_timers(void)
+{
+       int cpu;
+
+       for_each_online_cpu(cpu)
+               stop_stack_timer(cpu);
+}
+
+static void stop_stack_trace(struct trace_array *tr)
+{
+       mutex_lock(&sample_timer_lock);
+       stop_stack_timers();
+       tracer_enabled = 0;
+       mutex_unlock(&sample_timer_lock);
+}
+
+static int stack_trace_init(struct trace_array *tr)
+{
+       sysprof_trace = tr;
+
+       tracing_start_cmdline_record();
+
+       mutex_lock(&sample_timer_lock);
+       start_stack_timers();
+       tracer_enabled = 1;
+       mutex_unlock(&sample_timer_lock);
+       return 0;
+}
+
+static void stack_trace_reset(struct trace_array *tr)
+{
+       tracing_stop_cmdline_record();
+       stop_stack_trace(tr);
 }
 
 static struct tracer stack_trace __read_mostly =
@@ -67,9 +259,8 @@ static struct tracer stack_trace __read_mostly =
        .name           = "sysprof",
        .init           = stack_trace_init,
        .reset          = stack_trace_reset,
-       .ctrl_update    = stack_trace_ctrl_update,
 #ifdef CONFIG_FTRACE_SELFTEST
-       .selftest    = trace_selftest_startup_stack,
+       .selftest    = trace_selftest_startup_sysprof,
 #endif
 };
 
@@ -78,3 +269,60 @@ __init static int init_stack_trace(void)
        return register_tracer(&stack_trace);
 }
 device_initcall(init_stack_trace);
+
+#define MAX_LONG_DIGITS 22
+
+static ssize_t
+sysprof_sample_read(struct file *filp, char __user *ubuf,
+                   size_t cnt, loff_t *ppos)
+{
+       char buf[MAX_LONG_DIGITS];
+       int r;
+
+       r = sprintf(buf, "%ld\n", nsecs_to_usecs(sample_period));
+
+       return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
+}
+
+static ssize_t
+sysprof_sample_write(struct file *filp, const char __user *ubuf,
+                    size_t cnt, loff_t *ppos)
+{
+       char buf[MAX_LONG_DIGITS];
+       unsigned long val;
+
+       if (cnt > MAX_LONG_DIGITS-1)
+               cnt = MAX_LONG_DIGITS-1;
+
+       if (copy_from_user(&buf, ubuf, cnt))
+               return -EFAULT;
+
+       buf[cnt] = 0;
+
+       val = simple_strtoul(buf, NULL, 10);
+       /*
+        * Enforce a minimum sample period of 100 usecs:
+        */
+       if (val < 100)
+               val = 100;
+
+       mutex_lock(&sample_timer_lock);
+       stop_stack_timers();
+       sample_period = val * 1000;
+       start_stack_timers();
+       mutex_unlock(&sample_timer_lock);
+
+       return cnt;
+}
+
+static const struct file_operations sysprof_sample_fops = {
+       .read           = sysprof_sample_read,
+       .write          = sysprof_sample_write,
+};
+
+void init_tracer_sysprof_debugfs(struct dentry *d_tracer)
+{
+
+       trace_create_file("sysprof_sample_period", 0644,
+                       d_tracer, NULL, &sysprof_sample_fops);
+}