kernel/trace/trace_event_perf.c

   1 /*
   2  * trace event based perf event profiling/tracing
   3  *
   4  * Copyright (C) 2009 Red Hat Inc, Peter Zijlstra <pzijlstr@redhat.com>
   5  * Copyright (C) 2009-2010 Frederic Weisbecker <fweisbec@gmail.com>
   6  */
   7
   8 #include <linux/module.h>
   9 #include <linux/kprobes.h>
  10 #include "trace.h"
  11
  12 static char *perf_trace_buf[4];
  13
  14 /*
  15  * Force it to be aligned to unsigned long to avoid misaligned accesses
  16  * suprises
  17  */
  18 typedef typeof(unsigned long [PERF_MAX_TRACE_SIZE / sizeof(unsigned long)])
  19         perf_trace_t;
  20
  21 /* Count the events in use (per event id, not per instance) */
  22 static int      total_ref_count;
  23
  24 static int perf_trace_event_init(struct ftrace_event_call *tp_event,
  25                                  struct perf_event *p_event)
  26 {
  27         struct hlist_head *list;
  28         int ret = -ENOMEM;
  29         int cpu;
  30
  31         p_event->tp_event = tp_event;
  32         if (tp_event->perf_refcount++ > 0)
  33                 return 0;
  34
  35         list = alloc_percpu(struct hlist_head);
  36         if (!list)
  37                 goto fail;
  38
  39         for_each_possible_cpu(cpu)
  40                 INIT_HLIST_HEAD(per_cpu_ptr(list, cpu));
  41
  42         tp_event->perf_events = list;
  43
  44         if (!total_ref_count) {
  45                 char *buf;
  46                 int i;
  47
  48                 for (i = 0; i < 4; i++) {
  49                         buf = (char *)alloc_percpu(perf_trace_t);
  50                         if (!buf)
  51                                 goto fail;
  52
  53                         perf_trace_buf[i] = buf;
  54                 }
  55         }
  56
  57         if (tp_event->class->reg)
  58                 ret = tp_event->class->reg(tp_event, TRACE_REG_PERF_REGISTER);
  59         else
  60                 ret = tracepoint_probe_register(tp_event->name,
  61                                                 tp_event->class->perf_probe,
  62                                                 tp_event);
  63
  64         if (ret)
  65                 goto fail;
  66
  67         total_ref_count++;
  68         return 0;
  69
  70 fail:
  71         if (!total_ref_count) {
  72                 int i;
  73
  74                 for (i = 0; i < 4; i++) {
  75                         free_percpu(perf_trace_buf[i]);
  76                         perf_trace_buf[i] = NULL;
  77                 }
  78         }
  79
  80         if (!--tp_event->perf_refcount) {
  81                 free_percpu(tp_event->perf_events);
  82                 tp_event->perf_events = NULL;
  83         }
  84
  85         return ret;
  86 }
  87
  88 int perf_trace_init(struct perf_event *p_event)
  89 {
  90         struct ftrace_event_call *tp_event;
  91         int event_id = p_event->attr.config;
  92         int ret = -EINVAL;
  93
  94         mutex_lock(&event_mutex);
  95         list_for_each_entry(tp_event, &ftrace_events, list) {
  96                 if (tp_event->event.type == event_id &&
  97                     tp_event->class &&
  98                     (tp_event->class->perf_probe ||
  99                      tp_event->class->reg) &&
 100                     try_module_get(tp_event->mod)) {
 101                         ret = perf_trace_event_init(tp_event, p_event);
 102                         break;
 103                 }
 104         }
 105         mutex_unlock(&event_mutex);
 106
 107         return ret;
 108 }
 109
 110 int perf_trace_enable(struct perf_event *p_event)
 111 {
 112         struct ftrace_event_call *tp_event = p_event->tp_event;
 113         struct hlist_head *list;
 114
 115         list = tp_event->perf_events;
 116         if (WARN_ON_ONCE(!list))
 117                 return -EINVAL;
 118
 119         list = this_cpu_ptr(list);
 120         hlist_add_head_rcu(&p_event->hlist_entry, list);
 121
 122         return 0;
 123 }
 124
 125 void perf_trace_disable(struct perf_event *p_event)
 126 {
 127         hlist_del_rcu(&p_event->hlist_entry);
 128 }
 129
 130 void perf_trace_destroy(struct perf_event *p_event)
 131 {
 132         struct ftrace_event_call *tp_event = p_event->tp_event;
 133         int i;
 134
 135         mutex_lock(&event_mutex);
 136         if (--tp_event->perf_refcount > 0)
 137                 goto out;
 138
 139         if (tp_event->class->reg)
 140                 tp_event->class->reg(tp_event, TRACE_REG_PERF_UNREGISTER);
 141         else
 142                 tracepoint_probe_unregister(tp_event->name,
 143                                             tp_event->class->perf_probe,
 144                                             tp_event);
 145
 146         /*
 147          * Ensure our callback won't be called anymore. See
 148          * tracepoint_probe_unregister() and __DO_TRACE().
 149          */
 150         synchronize_sched();
 151
 152         free_percpu(tp_event->perf_events);
 153         tp_event->perf_events = NULL;
 154
 155         if (!--total_ref_count) {
 156                 for (i = 0; i < 4; i++) {
 157                         free_percpu(perf_trace_buf[i]);
 158                         perf_trace_buf[i] = NULL;
 159                 }
 160         }
 161 out:
 162         mutex_unlock(&event_mutex);
 163 }
 164
 165 __kprobes void *perf_trace_buf_prepare(int size, unsigned short type,
 166                                        struct pt_regs *regs, int *rctxp)
 167 {
 168         struct trace_entry *entry;
 169         unsigned long flags;
 170         char *raw_data;
 171         int pc;
 172
 173         BUILD_BUG_ON(PERF_MAX_TRACE_SIZE % sizeof(unsigned long));
 174
 175         pc = preempt_count();
 176
 177         *rctxp = perf_swevent_get_recursion_context();
 178         if (*rctxp < 0)
 179                 return NULL;
 180
 181         raw_data = this_cpu_ptr(perf_trace_buf[*rctxp]);
 182
 183         /* zero the dead bytes from align to not leak stack to user */
 184         memset(&raw_data[size - sizeof(u64)], 0, sizeof(u64));
 185
 186         entry = (struct trace_entry *)raw_data;
 187         local_save_flags(flags);
 188         tracing_generic_entry_update(entry, flags, pc);
 189         entry->type = type;
 190
 191         return raw_data;
 192 }
 193 EXPORT_SYMBOL_GPL(perf_trace_buf_prepare);