diff options
Diffstat (limited to 'kernel/trace/trace_event_perf.c')
-rw-r--r-- | kernel/trace/trace_event_perf.c | 208 |
1 files changed, 166 insertions, 42 deletions
diff --git a/kernel/trace/trace_event_perf.c b/kernel/trace/trace_event_perf.c index 19a359d5e6d5..fee3752ae8f6 100644 --- a/kernel/trace/trace_event_perf.c +++ b/kernel/trace/trace_event_perf.c | |||
@@ -24,6 +24,11 @@ static int total_ref_count; | |||
24 | static int perf_trace_event_perm(struct ftrace_event_call *tp_event, | 24 | static int perf_trace_event_perm(struct ftrace_event_call *tp_event, |
25 | struct perf_event *p_event) | 25 | struct perf_event *p_event) |
26 | { | 26 | { |
27 | /* The ftrace function trace is allowed only for root. */ | ||
28 | if (ftrace_event_is_function(tp_event) && | ||
29 | perf_paranoid_kernel() && !capable(CAP_SYS_ADMIN)) | ||
30 | return -EPERM; | ||
31 | |||
27 | /* No tracing, just counting, so no obvious leak */ | 32 | /* No tracing, just counting, so no obvious leak */ |
28 | if (!(p_event->attr.sample_type & PERF_SAMPLE_RAW)) | 33 | if (!(p_event->attr.sample_type & PERF_SAMPLE_RAW)) |
29 | return 0; | 34 | return 0; |
@@ -44,23 +49,17 @@ static int perf_trace_event_perm(struct ftrace_event_call *tp_event, | |||
44 | return 0; | 49 | return 0; |
45 | } | 50 | } |
46 | 51 | ||
47 | static int perf_trace_event_init(struct ftrace_event_call *tp_event, | 52 | static int perf_trace_event_reg(struct ftrace_event_call *tp_event, |
48 | struct perf_event *p_event) | 53 | struct perf_event *p_event) |
49 | { | 54 | { |
50 | struct hlist_head __percpu *list; | 55 | struct hlist_head __percpu *list; |
51 | int ret; | 56 | int ret = -ENOMEM; |
52 | int cpu; | 57 | int cpu; |
53 | 58 | ||
54 | ret = perf_trace_event_perm(tp_event, p_event); | ||
55 | if (ret) | ||
56 | return ret; | ||
57 | |||
58 | p_event->tp_event = tp_event; | 59 | p_event->tp_event = tp_event; |
59 | if (tp_event->perf_refcount++ > 0) | 60 | if (tp_event->perf_refcount++ > 0) |
60 | return 0; | 61 | return 0; |
61 | 62 | ||
62 | ret = -ENOMEM; | ||
63 | |||
64 | list = alloc_percpu(struct hlist_head); | 63 | list = alloc_percpu(struct hlist_head); |
65 | if (!list) | 64 | if (!list) |
66 | goto fail; | 65 | goto fail; |
@@ -83,7 +82,7 @@ static int perf_trace_event_init(struct ftrace_event_call *tp_event, | |||
83 | } | 82 | } |
84 | } | 83 | } |
85 | 84 | ||
86 | ret = tp_event->class->reg(tp_event, TRACE_REG_PERF_REGISTER); | 85 | ret = tp_event->class->reg(tp_event, TRACE_REG_PERF_REGISTER, NULL); |
87 | if (ret) | 86 | if (ret) |
88 | goto fail; | 87 | goto fail; |
89 | 88 | ||
@@ -108,6 +107,69 @@ fail: | |||
108 | return ret; | 107 | return ret; |
109 | } | 108 | } |
110 | 109 | ||
110 | static void perf_trace_event_unreg(struct perf_event *p_event) | ||
111 | { | ||
112 | struct ftrace_event_call *tp_event = p_event->tp_event; | ||
113 | int i; | ||
114 | |||
115 | if (--tp_event->perf_refcount > 0) | ||
116 | goto out; | ||
117 | |||
118 | tp_event->class->reg(tp_event, TRACE_REG_PERF_UNREGISTER, NULL); | ||
119 | |||
120 | /* | ||
121 | * Ensure our callback won't be called anymore. The buffers | ||
122 | * will be freed after that. | ||
123 | */ | ||
124 | tracepoint_synchronize_unregister(); | ||
125 | |||
126 | free_percpu(tp_event->perf_events); | ||
127 | tp_event->perf_events = NULL; | ||
128 | |||
129 | if (!--total_ref_count) { | ||
130 | for (i = 0; i < PERF_NR_CONTEXTS; i++) { | ||
131 | free_percpu(perf_trace_buf[i]); | ||
132 | perf_trace_buf[i] = NULL; | ||
133 | } | ||
134 | } | ||
135 | out: | ||
136 | module_put(tp_event->mod); | ||
137 | } | ||
138 | |||
139 | static int perf_trace_event_open(struct perf_event *p_event) | ||
140 | { | ||
141 | struct ftrace_event_call *tp_event = p_event->tp_event; | ||
142 | return tp_event->class->reg(tp_event, TRACE_REG_PERF_OPEN, p_event); | ||
143 | } | ||
144 | |||
145 | static void perf_trace_event_close(struct perf_event *p_event) | ||
146 | { | ||
147 | struct ftrace_event_call *tp_event = p_event->tp_event; | ||
148 | tp_event->class->reg(tp_event, TRACE_REG_PERF_CLOSE, p_event); | ||
149 | } | ||
150 | |||
151 | static int perf_trace_event_init(struct ftrace_event_call *tp_event, | ||
152 | struct perf_event *p_event) | ||
153 | { | ||
154 | int ret; | ||
155 | |||
156 | ret = perf_trace_event_perm(tp_event, p_event); | ||
157 | if (ret) | ||
158 | return ret; | ||
159 | |||
160 | ret = perf_trace_event_reg(tp_event, p_event); | ||
161 | if (ret) | ||
162 | return ret; | ||
163 | |||
164 | ret = perf_trace_event_open(p_event); | ||
165 | if (ret) { | ||
166 | perf_trace_event_unreg(p_event); | ||
167 | return ret; | ||
168 | } | ||
169 | |||
170 | return 0; | ||
171 | } | ||
172 | |||
111 | int perf_trace_init(struct perf_event *p_event) | 173 | int perf_trace_init(struct perf_event *p_event) |
112 | { | 174 | { |
113 | struct ftrace_event_call *tp_event; | 175 | struct ftrace_event_call *tp_event; |
@@ -130,6 +192,14 @@ int perf_trace_init(struct perf_event *p_event) | |||
130 | return ret; | 192 | return ret; |
131 | } | 193 | } |
132 | 194 | ||
195 | void perf_trace_destroy(struct perf_event *p_event) | ||
196 | { | ||
197 | mutex_lock(&event_mutex); | ||
198 | perf_trace_event_close(p_event); | ||
199 | perf_trace_event_unreg(p_event); | ||
200 | mutex_unlock(&event_mutex); | ||
201 | } | ||
202 | |||
133 | int perf_trace_add(struct perf_event *p_event, int flags) | 203 | int perf_trace_add(struct perf_event *p_event, int flags) |
134 | { | 204 | { |
135 | struct ftrace_event_call *tp_event = p_event->tp_event; | 205 | struct ftrace_event_call *tp_event = p_event->tp_event; |
@@ -146,43 +216,14 @@ int perf_trace_add(struct perf_event *p_event, int flags) | |||
146 | list = this_cpu_ptr(pcpu_list); | 216 | list = this_cpu_ptr(pcpu_list); |
147 | hlist_add_head_rcu(&p_event->hlist_entry, list); | 217 | hlist_add_head_rcu(&p_event->hlist_entry, list); |
148 | 218 | ||
149 | return 0; | 219 | return tp_event->class->reg(tp_event, TRACE_REG_PERF_ADD, p_event); |
150 | } | 220 | } |
151 | 221 | ||
152 | void perf_trace_del(struct perf_event *p_event, int flags) | 222 | void perf_trace_del(struct perf_event *p_event, int flags) |
153 | { | 223 | { |
154 | hlist_del_rcu(&p_event->hlist_entry); | ||
155 | } | ||
156 | |||
157 | void perf_trace_destroy(struct perf_event *p_event) | ||
158 | { | ||
159 | struct ftrace_event_call *tp_event = p_event->tp_event; | 224 | struct ftrace_event_call *tp_event = p_event->tp_event; |
160 | int i; | 225 | hlist_del_rcu(&p_event->hlist_entry); |
161 | 226 | tp_event->class->reg(tp_event, TRACE_REG_PERF_DEL, p_event); | |
162 | mutex_lock(&event_mutex); | ||
163 | if (--tp_event->perf_refcount > 0) | ||
164 | goto out; | ||
165 | |||
166 | tp_event->class->reg(tp_event, TRACE_REG_PERF_UNREGISTER); | ||
167 | |||
168 | /* | ||
169 | * Ensure our callback won't be called anymore. The buffers | ||
170 | * will be freed after that. | ||
171 | */ | ||
172 | tracepoint_synchronize_unregister(); | ||
173 | |||
174 | free_percpu(tp_event->perf_events); | ||
175 | tp_event->perf_events = NULL; | ||
176 | |||
177 | if (!--total_ref_count) { | ||
178 | for (i = 0; i < PERF_NR_CONTEXTS; i++) { | ||
179 | free_percpu(perf_trace_buf[i]); | ||
180 | perf_trace_buf[i] = NULL; | ||
181 | } | ||
182 | } | ||
183 | out: | ||
184 | module_put(tp_event->mod); | ||
185 | mutex_unlock(&event_mutex); | ||
186 | } | 227 | } |
187 | 228 | ||
188 | __kprobes void *perf_trace_buf_prepare(int size, unsigned short type, | 229 | __kprobes void *perf_trace_buf_prepare(int size, unsigned short type, |
@@ -214,3 +255,86 @@ __kprobes void *perf_trace_buf_prepare(int size, unsigned short type, | |||
214 | return raw_data; | 255 | return raw_data; |
215 | } | 256 | } |
216 | EXPORT_SYMBOL_GPL(perf_trace_buf_prepare); | 257 | EXPORT_SYMBOL_GPL(perf_trace_buf_prepare); |
258 | |||
259 | #ifdef CONFIG_FUNCTION_TRACER | ||
260 | static void | ||
261 | perf_ftrace_function_call(unsigned long ip, unsigned long parent_ip) | ||
262 | { | ||
263 | struct ftrace_entry *entry; | ||
264 | struct hlist_head *head; | ||
265 | struct pt_regs regs; | ||
266 | int rctx; | ||
267 | |||
268 | #define ENTRY_SIZE (ALIGN(sizeof(struct ftrace_entry) + sizeof(u32), \ | ||
269 | sizeof(u64)) - sizeof(u32)) | ||
270 | |||
271 | BUILD_BUG_ON(ENTRY_SIZE > PERF_MAX_TRACE_SIZE); | ||
272 | |||
273 | perf_fetch_caller_regs(®s); | ||
274 | |||
275 | entry = perf_trace_buf_prepare(ENTRY_SIZE, TRACE_FN, NULL, &rctx); | ||
276 | if (!entry) | ||
277 | return; | ||
278 | |||
279 | entry->ip = ip; | ||
280 | entry->parent_ip = parent_ip; | ||
281 | |||
282 | head = this_cpu_ptr(event_function.perf_events); | ||
283 | perf_trace_buf_submit(entry, ENTRY_SIZE, rctx, 0, | ||
284 | 1, ®s, head); | ||
285 | |||
286 | #undef ENTRY_SIZE | ||
287 | } | ||
288 | |||
289 | static int perf_ftrace_function_register(struct perf_event *event) | ||
290 | { | ||
291 | struct ftrace_ops *ops = &event->ftrace_ops; | ||
292 | |||
293 | ops->flags |= FTRACE_OPS_FL_CONTROL; | ||
294 | ops->func = perf_ftrace_function_call; | ||
295 | return register_ftrace_function(ops); | ||
296 | } | ||
297 | |||
298 | static int perf_ftrace_function_unregister(struct perf_event *event) | ||
299 | { | ||
300 | struct ftrace_ops *ops = &event->ftrace_ops; | ||
301 | int ret = unregister_ftrace_function(ops); | ||
302 | ftrace_free_filter(ops); | ||
303 | return ret; | ||
304 | } | ||
305 | |||
306 | static void perf_ftrace_function_enable(struct perf_event *event) | ||
307 | { | ||
308 | ftrace_function_local_enable(&event->ftrace_ops); | ||
309 | } | ||
310 | |||
311 | static void perf_ftrace_function_disable(struct perf_event *event) | ||
312 | { | ||
313 | ftrace_function_local_disable(&event->ftrace_ops); | ||
314 | } | ||
315 | |||
316 | int perf_ftrace_event_register(struct ftrace_event_call *call, | ||
317 | enum trace_reg type, void *data) | ||
318 | { | ||
319 | switch (type) { | ||
320 | case TRACE_REG_REGISTER: | ||
321 | case TRACE_REG_UNREGISTER: | ||
322 | break; | ||
323 | case TRACE_REG_PERF_REGISTER: | ||
324 | case TRACE_REG_PERF_UNREGISTER: | ||
325 | return 0; | ||
326 | case TRACE_REG_PERF_OPEN: | ||
327 | return perf_ftrace_function_register(data); | ||
328 | case TRACE_REG_PERF_CLOSE: | ||
329 | return perf_ftrace_function_unregister(data); | ||
330 | case TRACE_REG_PERF_ADD: | ||
331 | perf_ftrace_function_enable(data); | ||
332 | return 0; | ||
333 | case TRACE_REG_PERF_DEL: | ||
334 | perf_ftrace_function_disable(data); | ||
335 | return 0; | ||
336 | } | ||
337 | |||
338 | return -EINVAL; | ||
339 | } | ||
340 | #endif /* CONFIG_FUNCTION_TRACER */ | ||