aboutsummaryrefslogtreecommitdiffstats
path: root/kernel/trace/trace_functions_graph.c
diff options
context:
space:
mode:
Diffstat (limited to 'kernel/trace/trace_functions_graph.c')
-rw-r--r--kernel/trace/trace_functions_graph.c267
1 files changed, 246 insertions, 21 deletions
diff --git a/kernel/trace/trace_functions_graph.c b/kernel/trace/trace_functions_graph.c
index d31d695174aa..894b50bca313 100644
--- a/kernel/trace/trace_functions_graph.c
+++ b/kernel/trace/trace_functions_graph.c
@@ -15,15 +15,24 @@
15 15
16#define TRACE_GRAPH_INDENT 2 16#define TRACE_GRAPH_INDENT 2
17 17
18/* Flag options */
18#define TRACE_GRAPH_PRINT_OVERRUN 0x1 19#define TRACE_GRAPH_PRINT_OVERRUN 0x1
20#define TRACE_GRAPH_PRINT_CPU 0x2
21#define TRACE_GRAPH_PRINT_OVERHEAD 0x4
22
19static struct tracer_opt trace_opts[] = { 23static struct tracer_opt trace_opts[] = {
20 /* Display overruns or not */ 24 /* Display overruns ? */
21 { TRACER_OPT(overrun, TRACE_GRAPH_PRINT_OVERRUN) }, 25 { TRACER_OPT(funcgraph-overrun, TRACE_GRAPH_PRINT_OVERRUN) },
26 /* Display CPU ? */
27 { TRACER_OPT(funcgraph-cpu, TRACE_GRAPH_PRINT_CPU) },
28 /* Display Overhead ? */
29 { TRACER_OPT(funcgraph-overhead, TRACE_GRAPH_PRINT_OVERHEAD) },
22 { } /* Empty entry */ 30 { } /* Empty entry */
23}; 31};
24 32
25static struct tracer_flags tracer_flags = { 33static struct tracer_flags tracer_flags = {
26 .val = 0, /* Don't display overruns by default */ 34 /* Don't display overruns by default */
35 .val = TRACE_GRAPH_PRINT_CPU | TRACE_GRAPH_PRINT_OVERHEAD,
27 .opts = trace_opts 36 .opts = trace_opts
28}; 37};
29 38
@@ -52,37 +61,208 @@ static void graph_trace_reset(struct trace_array *tr)
52 unregister_ftrace_graph(); 61 unregister_ftrace_graph();
53} 62}
54 63
64static inline int log10_cpu(int nb)
65{
66 if (nb / 100)
67 return 3;
68 if (nb / 10)
69 return 2;
70 return 1;
71}
72
73static enum print_line_t
74print_graph_cpu(struct trace_seq *s, int cpu)
75{
76 int i;
77 int ret;
78 int log10_this = log10_cpu(cpu);
79 int log10_all = log10_cpu(cpus_weight_nr(cpu_online_map));
80
81
82 /*
83 * Start with a space character - to make it stand out
84 * to the right a bit when trace output is pasted into
85 * email:
86 */
87 ret = trace_seq_printf(s, " ");
88
89 /*
90 * Tricky - we space the CPU field according to the max
91 * number of online CPUs. On a 2-cpu system it would take
92 * a maximum of 1 digit - on a 128 cpu system it would
93 * take up to 3 digits:
94 */
95 for (i = 0; i < log10_all - log10_this; i++) {
96 ret = trace_seq_printf(s, " ");
97 if (!ret)
98 return TRACE_TYPE_PARTIAL_LINE;
99 }
100 ret = trace_seq_printf(s, "%d) ", cpu);
101 if (!ret)
102 return TRACE_TYPE_PARTIAL_LINE;
103
104 return TRACE_TYPE_HANDLED;
105}
106
107
55/* If the pid changed since the last trace, output this event */ 108/* If the pid changed since the last trace, output this event */
56static int verif_pid(struct trace_seq *s, pid_t pid, int cpu) 109static int verif_pid(struct trace_seq *s, pid_t pid, int cpu)
57{ 110{
58 char *comm; 111 char *comm, *prev_comm;
112 pid_t prev_pid;
113 int ret;
59 114
60 if (last_pid[cpu] != -1 && last_pid[cpu] == pid) 115 if (last_pid[cpu] != -1 && last_pid[cpu] == pid)
61 return 1; 116 return 1;
62 117
118 prev_pid = last_pid[cpu];
63 last_pid[cpu] = pid; 119 last_pid[cpu] = pid;
120
64 comm = trace_find_cmdline(pid); 121 comm = trace_find_cmdline(pid);
122 prev_comm = trace_find_cmdline(prev_pid);
65 123
66 return trace_seq_printf(s, "\nCPU[%03d]" 124/*
67 " ------------8<---------- thread %s-%d" 125 * Context-switch trace line:
68 " ------------8<----------\n\n", 126
69 cpu, comm, pid); 127 ------------------------------------------
128 | 1) migration/0--1 => sshd-1755
129 ------------------------------------------
130
131 */
132 ret = trace_seq_printf(s,
133 " ------------------------------------------\n");
134 ret += trace_seq_printf(s, " | %d) %s-%d => %s-%d\n",
135 cpu, prev_comm, prev_pid, comm, pid);
136 ret += trace_seq_printf(s,
137 " ------------------------------------------\n\n");
138 return ret;
70} 139}
71 140
141static bool
142trace_branch_is_leaf(struct trace_iterator *iter,
143 struct ftrace_graph_ent_entry *curr)
144{
145 struct ring_buffer_iter *ring_iter;
146 struct ring_buffer_event *event;
147 struct ftrace_graph_ret_entry *next;
148
149 ring_iter = iter->buffer_iter[iter->cpu];
150
151 if (!ring_iter)
152 return false;
153
154 event = ring_buffer_iter_peek(ring_iter, NULL);
155
156 if (!event)
157 return false;
158
159 next = ring_buffer_event_data(event);
160
161 if (next->ent.type != TRACE_GRAPH_RET)
162 return false;
163
164 if (curr->ent.pid != next->ent.pid ||
165 curr->graph_ent.func != next->ret.func)
166 return false;
167
168 return true;
169}
170
171
172static inline int
173print_graph_duration(unsigned long long duration, struct trace_seq *s)
174{
175 unsigned long nsecs_rem = do_div(duration, 1000);
176 return trace_seq_printf(s, "%4llu.%3lu us | ", duration, nsecs_rem);
177}
178
179/* Signal a overhead of time execution to the output */
180static int
181print_graph_overhead(unsigned long long duration, struct trace_seq *s)
182{
183 /* Duration exceeded 100 msecs */
184 if (duration > 100000ULL)
185 return trace_seq_printf(s, "! ");
186
187 /* Duration exceeded 10 msecs */
188 if (duration > 10000ULL)
189 return trace_seq_printf(s, "+ ");
190
191 return trace_seq_printf(s, " ");
192}
193
194/* Case of a leaf function on its call entry */
72static enum print_line_t 195static enum print_line_t
73print_graph_entry(struct ftrace_graph_ent *call, struct trace_seq *s, 196print_graph_entry_leaf(struct trace_iterator *iter,
74 struct trace_entry *ent, int cpu) 197 struct ftrace_graph_ent_entry *entry, struct trace_seq *s)
75{ 198{
76 int i; 199 struct ftrace_graph_ret_entry *ret_entry;
200 struct ftrace_graph_ret *graph_ret;
201 struct ring_buffer_event *event;
202 struct ftrace_graph_ent *call;
203 unsigned long long duration;
77 int ret; 204 int ret;
205 int i;
78 206
79 if (!verif_pid(s, ent->pid, cpu)) 207 event = ring_buffer_read(iter->buffer_iter[iter->cpu], NULL);
208 ret_entry = ring_buffer_event_data(event);
209 graph_ret = &ret_entry->ret;
210 call = &entry->graph_ent;
211 duration = graph_ret->rettime - graph_ret->calltime;
212
213 /* Must not exceed 8 characters: 9999.999 us */
214 if (duration > 10000000ULL)
215 duration = 9999999ULL;
216
217 /* Overhead */
218 if (tracer_flags.val & TRACE_GRAPH_PRINT_OVERHEAD) {
219 ret = print_graph_overhead(duration, s);
220 if (!ret)
221 return TRACE_TYPE_PARTIAL_LINE;
222 }
223
224 /* Duration */
225 ret = print_graph_duration(duration, s);
226 if (!ret)
80 return TRACE_TYPE_PARTIAL_LINE; 227 return TRACE_TYPE_PARTIAL_LINE;
81 228
82 ret = trace_seq_printf(s, "CPU[%03d] ", cpu); 229 /* Function */
230 for (i = 0; i < call->depth * TRACE_GRAPH_INDENT; i++) {
231 ret = trace_seq_printf(s, " ");
232 if (!ret)
233 return TRACE_TYPE_PARTIAL_LINE;
234 }
235
236 ret = seq_print_ip_sym(s, call->func, 0);
237 if (!ret)
238 return TRACE_TYPE_PARTIAL_LINE;
239
240 ret = trace_seq_printf(s, "();\n");
83 if (!ret) 241 if (!ret)
84 return TRACE_TYPE_PARTIAL_LINE; 242 return TRACE_TYPE_PARTIAL_LINE;
85 243
244 return TRACE_TYPE_HANDLED;
245}
246
247static enum print_line_t
248print_graph_entry_nested(struct ftrace_graph_ent_entry *entry,
249 struct trace_seq *s)
250{
251 int i;
252 int ret;
253 struct ftrace_graph_ent *call = &entry->graph_ent;
254
255 /* No overhead */
256 if (tracer_flags.val & TRACE_GRAPH_PRINT_OVERHEAD) {
257 ret = trace_seq_printf(s, " ");
258 if (!ret)
259 return TRACE_TYPE_PARTIAL_LINE;
260 }
261
262 /* No time */
263 ret = trace_seq_printf(s, " | ");
264
265 /* Function */
86 for (i = 0; i < call->depth * TRACE_GRAPH_INDENT; i++) { 266 for (i = 0; i < call->depth * TRACE_GRAPH_INDENT; i++) {
87 ret = trace_seq_printf(s, " "); 267 ret = trace_seq_printf(s, " ");
88 if (!ret) 268 if (!ret)
@@ -96,37 +276,82 @@ print_graph_entry(struct ftrace_graph_ent *call, struct trace_seq *s,
96 ret = trace_seq_printf(s, "() {\n"); 276 ret = trace_seq_printf(s, "() {\n");
97 if (!ret) 277 if (!ret)
98 return TRACE_TYPE_PARTIAL_LINE; 278 return TRACE_TYPE_PARTIAL_LINE;
279
99 return TRACE_TYPE_HANDLED; 280 return TRACE_TYPE_HANDLED;
100} 281}
101 282
102static enum print_line_t 283static enum print_line_t
284print_graph_entry(struct ftrace_graph_ent_entry *field, struct trace_seq *s,
285 struct trace_iterator *iter, int cpu)
286{
287 int ret;
288 struct trace_entry *ent = iter->ent;
289
290 /* Pid */
291 if (!verif_pid(s, ent->pid, cpu))
292 return TRACE_TYPE_PARTIAL_LINE;
293
294 /* Cpu */
295 if (tracer_flags.val & TRACE_GRAPH_PRINT_CPU) {
296 ret = print_graph_cpu(s, cpu);
297 if (!ret)
298 return TRACE_TYPE_PARTIAL_LINE;
299 }
300
301 if (trace_branch_is_leaf(iter, field))
302 return print_graph_entry_leaf(iter, field, s);
303 else
304 return print_graph_entry_nested(field, s);
305
306}
307
308static enum print_line_t
103print_graph_return(struct ftrace_graph_ret *trace, struct trace_seq *s, 309print_graph_return(struct ftrace_graph_ret *trace, struct trace_seq *s,
104 struct trace_entry *ent, int cpu) 310 struct trace_entry *ent, int cpu)
105{ 311{
106 int i; 312 int i;
107 int ret; 313 int ret;
314 unsigned long long duration = trace->rettime - trace->calltime;
315
316 /* Must not exceed 8 characters: xxxx.yyy us */
317 if (duration > 10000000ULL)
318 duration = 9999999ULL;
108 319
320 /* Pid */
109 if (!verif_pid(s, ent->pid, cpu)) 321 if (!verif_pid(s, ent->pid, cpu))
110 return TRACE_TYPE_PARTIAL_LINE; 322 return TRACE_TYPE_PARTIAL_LINE;
111 323
112 ret = trace_seq_printf(s, "CPU[%03d] ", cpu); 324 /* Cpu */
325 if (tracer_flags.val & TRACE_GRAPH_PRINT_CPU) {
326 ret = print_graph_cpu(s, cpu);
327 if (!ret)
328 return TRACE_TYPE_PARTIAL_LINE;
329 }
330
331 /* Overhead */
332 if (tracer_flags.val & TRACE_GRAPH_PRINT_OVERHEAD) {
333 ret = print_graph_overhead(duration, s);
334 if (!ret)
335 return TRACE_TYPE_PARTIAL_LINE;
336 }
337
338 /* Duration */
339 ret = print_graph_duration(duration, s);
113 if (!ret) 340 if (!ret)
114 return TRACE_TYPE_PARTIAL_LINE; 341 return TRACE_TYPE_PARTIAL_LINE;
115 342
343 /* Closing brace */
116 for (i = 0; i < trace->depth * TRACE_GRAPH_INDENT; i++) { 344 for (i = 0; i < trace->depth * TRACE_GRAPH_INDENT; i++) {
117 ret = trace_seq_printf(s, " "); 345 ret = trace_seq_printf(s, " ");
118 if (!ret) 346 if (!ret)
119 return TRACE_TYPE_PARTIAL_LINE; 347 return TRACE_TYPE_PARTIAL_LINE;
120 } 348 }
121 349
122 ret = trace_seq_printf(s, "} "); 350 ret = trace_seq_printf(s, "}\n");
123 if (!ret)
124 return TRACE_TYPE_PARTIAL_LINE;
125
126 ret = trace_seq_printf(s, "%llu\n", trace->rettime - trace->calltime);
127 if (!ret) 351 if (!ret)
128 return TRACE_TYPE_PARTIAL_LINE; 352 return TRACE_TYPE_PARTIAL_LINE;
129 353
354 /* Overrun */
130 if (tracer_flags.val & TRACE_GRAPH_PRINT_OVERRUN) { 355 if (tracer_flags.val & TRACE_GRAPH_PRINT_OVERRUN) {
131 ret = trace_seq_printf(s, " (Overruns: %lu)\n", 356 ret = trace_seq_printf(s, " (Overruns: %lu)\n",
132 trace->overrun); 357 trace->overrun);
@@ -146,7 +371,7 @@ print_graph_function(struct trace_iterator *iter)
146 case TRACE_GRAPH_ENT: { 371 case TRACE_GRAPH_ENT: {
147 struct ftrace_graph_ent_entry *field; 372 struct ftrace_graph_ent_entry *field;
148 trace_assign_type(field, entry); 373 trace_assign_type(field, entry);
149 return print_graph_entry(&field->graph_ent, s, entry, 374 return print_graph_entry(field, s, iter,
150 iter->cpu); 375 iter->cpu);
151 } 376 }
152 case TRACE_GRAPH_RET: { 377 case TRACE_GRAPH_RET: {
@@ -160,7 +385,7 @@ print_graph_function(struct trace_iterator *iter)
160} 385}
161 386
162static struct tracer graph_trace __read_mostly = { 387static struct tracer graph_trace __read_mostly = {
163 .name = "function-graph", 388 .name = "function_graph",
164 .init = graph_trace_init, 389 .init = graph_trace_init,
165 .reset = graph_trace_reset, 390 .reset = graph_trace_reset,
166 .print_line = print_graph_function, 391 .print_line = print_graph_function,