kgdb: gdb "monitor" -> kdb passthrough
[safe/jmp/linux-2.6] / kernel / trace / trace_events.c
index 83cc2c0..c697c70 100644 (file)
 #include <linux/uaccess.h>
 #include <linux/module.h>
 #include <linux/ctype.h>
+#include <linux/slab.h>
 #include <linux/delay.h>
 
 #include <asm/setup.h>
 
 #include "trace_output.h"
 
+#undef TRACE_SYSTEM
 #define TRACE_SYSTEM "TRACE_SYSTEM"
 
 DEFINE_MUTEX(event_mutex);
@@ -59,10 +61,8 @@ int trace_define_field(struct ftrace_event_call *call, const char *type,
        return 0;
 
 err:
-       if (field) {
+       if (field)
                kfree(field->name);
-               kfree(field->type);
-       }
        kfree(field);
 
        return -ENOMEM;
@@ -77,7 +77,7 @@ EXPORT_SYMBOL_GPL(trace_define_field);
        if (ret)                                                        \
                return ret;
 
-int trace_define_common_fields(struct ftrace_event_call *call)
+static int trace_define_common_fields(struct ftrace_event_call *call)
 {
        int ret;
        struct trace_entry ent;
@@ -86,11 +86,10 @@ int trace_define_common_fields(struct ftrace_event_call *call)
        __common_field(unsigned char, flags);
        __common_field(unsigned char, preempt_count);
        __common_field(int, pid);
-       __common_field(int, tgid);
+       __common_field(int, lock_depth);
 
        return ret;
 }
-EXPORT_SYMBOL_GPL(trace_define_common_fields);
 
 void trace_destroy_fields(struct ftrace_event_call *call)
 {
@@ -104,9 +103,25 @@ void trace_destroy_fields(struct ftrace_event_call *call)
        }
 }
 
-static void ftrace_event_enable_disable(struct ftrace_event_call *call,
+int trace_event_raw_init(struct ftrace_event_call *call)
+{
+       int id;
+
+       id = register_ftrace_event(call->event);
+       if (!id)
+               return -ENODEV;
+       call->id = id;
+       INIT_LIST_HEAD(&call->fields);
+
+       return 0;
+}
+EXPORT_SYMBOL_GPL(trace_event_raw_init);
+
+static int ftrace_event_enable_disable(struct ftrace_event_call *call,
                                        int enable)
 {
+       int ret = 0;
+
        switch (enable) {
        case 0:
                if (call->enabled) {
@@ -117,12 +132,20 @@ static void ftrace_event_enable_disable(struct ftrace_event_call *call,
                break;
        case 1:
                if (!call->enabled) {
-                       call->enabled = 1;
                        tracing_start_cmdline_record();
-                       call->regfunc(call);
+                       ret = call->regfunc(call);
+                       if (ret) {
+                               tracing_stop_cmdline_record();
+                               pr_info("event trace: Could not enable event "
+                                       "%s\n", call->name);
+                               break;
+                       }
+                       call->enabled = 1;
                }
                break;
        }
+
+       return ret;
 }
 
 static void ftrace_clear_events(void)
@@ -226,73 +249,38 @@ static ssize_t
 ftrace_event_write(struct file *file, const char __user *ubuf,
                   size_t cnt, loff_t *ppos)
 {
-       size_t read = 0;
-       int i, set = 1;
-       ssize_t ret;
-       char *buf;
-       char ch;
+       struct trace_parser parser;
+       ssize_t read, ret;
 
-       if (!cnt || cnt < 0)
+       if (!cnt)
                return 0;
 
        ret = tracing_update_buffers();
        if (ret < 0)
                return ret;
 
-       ret = get_user(ch, ubuf++);
-       if (ret)
-               return ret;
-       read++;
-       cnt--;
-
-       /* skip white space */
-       while (cnt && isspace(ch)) {
-               ret = get_user(ch, ubuf++);
-               if (ret)
-                       return ret;
-               read++;
-               cnt--;
-       }
-
-       /* Only white space found? */
-       if (isspace(ch)) {
-               file->f_pos += read;
-               ret = read;
-               return ret;
-       }
-
-       buf = kmalloc(EVENT_BUF_SIZE+1, GFP_KERNEL);
-       if (!buf)
+       if (trace_parser_get_init(&parser, EVENT_BUF_SIZE + 1))
                return -ENOMEM;
 
-       if (cnt > EVENT_BUF_SIZE)
-               cnt = EVENT_BUF_SIZE;
+       read = trace_get_user(&parser, ubuf, cnt, ppos);
 
-       i = 0;
-       while (cnt && !isspace(ch)) {
-               if (!i && ch == '!')
+       if (read >= 0 && trace_parser_loaded((&parser))) {
+               int set = 1;
+
+               if (*parser.buffer == '!')
                        set = 0;
-               else
-                       buf[i++] = ch;
 
-               ret = get_user(ch, ubuf++);
+               parser.buffer[parser.idx] = 0;
+
+               ret = ftrace_set_clr_event(parser.buffer + !set, set);
                if (ret)
-                       goto out_free;
-               read++;
-               cnt--;
+                       goto out_put;
        }
-       buf[i] = 0;
-
-       file->f_pos += read;
-
-       ret = ftrace_set_clr_event(buf, set);
-       if (ret)
-               goto out_free;
 
        ret = read;
 
- out_free:
-       kfree(buf);
+ out_put:
+       trace_parser_put(&parser);
 
        return ret;
 }
@@ -300,42 +288,32 @@ ftrace_event_write(struct file *file, const char __user *ubuf,
 static void *
 t_next(struct seq_file *m, void *v, loff_t *pos)
 {
-       struct list_head *list = m->private;
-       struct ftrace_event_call *call;
+       struct ftrace_event_call *call = v;
 
        (*pos)++;
 
-       for (;;) {
-               if (list == &ftrace_events)
-                       return NULL;
-
-               call = list_entry(list, struct ftrace_event_call, list);
-
+       list_for_each_entry_continue(call, &ftrace_events, list) {
                /*
                 * The ftrace subsystem is for showing formats only.
                 * They can not be enabled or disabled via the event files.
                 */
                if (call->regfunc)
-                       break;
-
-               list = list->next;
+                       return call;
        }
 
-       m->private = list->next;
-
-       return call;
+       return NULL;
 }
 
 static void *t_start(struct seq_file *m, loff_t *pos)
 {
-       struct ftrace_event_call *call = NULL;
+       struct ftrace_event_call *call;
        loff_t l;
 
        mutex_lock(&event_mutex);
 
-       m->private = ftrace_events.next;
+       call = list_entry(&ftrace_events, struct ftrace_event_call, list);
        for (l = 0; l <= *pos; ) {
-               call = t_next(m, NULL, &l);
+               call = t_next(m, call, &l);
                if (!call)
                        break;
        }
@@ -345,37 +323,28 @@ static void *t_start(struct seq_file *m, loff_t *pos)
 static void *
 s_next(struct seq_file *m, void *v, loff_t *pos)
 {
-       struct list_head *list = m->private;
-       struct ftrace_event_call *call;
+       struct ftrace_event_call *call = v;
 
        (*pos)++;
 
- retry:
-       if (list == &ftrace_events)
-               return NULL;
-
-       call = list_entry(list, struct ftrace_event_call, list);
-
-       if (!call->enabled) {
-               list = list->next;
-               goto retry;
+       list_for_each_entry_continue(call, &ftrace_events, list) {
+               if (call->enabled)
+                       return call;
        }
 
-       m->private = list->next;
-
-       return call;
+       return NULL;
 }
 
 static void *s_start(struct seq_file *m, loff_t *pos)
 {
-       struct ftrace_event_call *call = NULL;
+       struct ftrace_event_call *call;
        loff_t l;
 
        mutex_lock(&event_mutex);
 
-       m->private = ftrace_events.next;
+       call = list_entry(&ftrace_events, struct ftrace_event_call, list);
        for (l = 0; l <= *pos; ) {
-               call = s_next(m, NULL, &l);
+               call = s_next(m, call, &l);
                if (!call)
                        break;
        }
@@ -455,7 +424,7 @@ event_enable_write(struct file *filp, const char __user *ubuf, size_t cnt,
        case 0:
        case 1:
                mutex_lock(&event_mutex);
-               ftrace_event_enable_disable(call, val);
+               ret = ftrace_event_enable_disable(call, val);
                mutex_unlock(&event_mutex);
                break;
 
@@ -465,7 +434,7 @@ event_enable_write(struct file *filp, const char __user *ubuf, size_t cnt,
 
        *ppos += cnt;
 
-       return cnt;
+       return ret ? ret : cnt;
 }
 
 static ssize_t
@@ -550,41 +519,16 @@ out:
        return ret;
 }
 
-extern char *__bad_type_size(void);
-
-#undef FIELD
-#define FIELD(type, name)                                              \
-       sizeof(type) != sizeof(field.name) ? __bad_type_size() :        \
-       #type, "common_" #name, offsetof(typeof(field), name),          \
-               sizeof(field.name)
-
-static int trace_write_header(struct trace_seq *s)
-{
-       struct trace_entry field;
-
-       /* struct trace_entry */
-       return trace_seq_printf(s,
-                               "\tfield:%s %s;\toffset:%zu;\tsize:%zu;\n"
-                               "\tfield:%s %s;\toffset:%zu;\tsize:%zu;\n"
-                               "\tfield:%s %s;\toffset:%zu;\tsize:%zu;\n"
-                               "\tfield:%s %s;\toffset:%zu;\tsize:%zu;\n"
-                               "\tfield:%s %s;\toffset:%zu;\tsize:%zu;\n"
-                               "\n",
-                               FIELD(unsigned short, type),
-                               FIELD(unsigned char, flags),
-                               FIELD(unsigned char, preempt_count),
-                               FIELD(int, pid),
-                               FIELD(int, tgid));
-}
-
 static ssize_t
 event_format_read(struct file *filp, char __user *ubuf, size_t cnt,
                  loff_t *ppos)
 {
        struct ftrace_event_call *call = filp->private_data;
+       struct ftrace_event_field *field;
        struct trace_seq *s;
+       int common_field_count = 5;
        char *buf;
-       int r;
+       int r = 0;
 
        if (*ppos)
                return 0;
@@ -595,14 +539,48 @@ event_format_read(struct file *filp, char __user *ubuf, size_t cnt,
 
        trace_seq_init(s);
 
-       /* If any of the first writes fail, so will the show_format. */
-
        trace_seq_printf(s, "name: %s\n", call->name);
        trace_seq_printf(s, "ID: %d\n", call->id);
        trace_seq_printf(s, "format:\n");
-       trace_write_header(s);
 
-       r = call->show_format(call, s);
+       list_for_each_entry_reverse(field, &call->fields, link) {
+               /*
+                * Smartly shows the array type(except dynamic array).
+                * Normal:
+                *      field:TYPE VAR
+                * If TYPE := TYPE[LEN], it is shown:
+                *      field:TYPE VAR[LEN]
+                */
+               const char *array_descriptor = strchr(field->type, '[');
+
+               if (!strncmp(field->type, "__data_loc", 10))
+                       array_descriptor = NULL;
+
+               if (!array_descriptor) {
+                       r = trace_seq_printf(s, "\tfield:%s %s;\toffset:%u;"
+                                       "\tsize:%u;\tsigned:%d;\n",
+                                       field->type, field->name, field->offset,
+                                       field->size, !!field->is_signed);
+               } else {
+                       r = trace_seq_printf(s, "\tfield:%.*s %s%s;\toffset:%u;"
+                                       "\tsize:%u;\tsigned:%d;\n",
+                                       (int)(array_descriptor - field->type),
+                                       field->type, field->name,
+                                       array_descriptor, field->offset,
+                                       field->size, !!field->is_signed);
+               }
+
+               if (--common_field_count == 0)
+                       r = trace_seq_printf(s, "\n");
+
+               if (!r)
+                       break;
+       }
+
+       if (r)
+               r = trace_seq_printf(s, "\nprint fmt: %s\n",
+                               call->print_fmt);
+
        if (!r) {
                /*
                 * ug!  The format output is bigger than a PAGE!!
@@ -927,9 +905,9 @@ event_subsystem_dir(const char *name, struct dentry *d_events)
                           "'%s/filter' entry\n", name);
        }
 
-       entry = trace_create_file("enable", 0644, system->entry,
-                                 (void *)system->name,
-                                 &ftrace_system_enable_fops);
+       trace_create_file("enable", 0644, system->entry,
+                         (void *)system->name,
+                         &ftrace_system_enable_fops);
 
        return system->entry;
 }
@@ -941,7 +919,6 @@ event_create_dir(struct ftrace_event_call *call, struct dentry *d_events,
                 const struct file_operations *filter,
                 const struct file_operations *format)
 {
-       struct dentry *entry;
        int ret;
 
        /*
@@ -959,30 +936,28 @@ event_create_dir(struct ftrace_event_call *call, struct dentry *d_events,
        }
 
        if (call->regfunc)
-               entry = trace_create_file("enable", 0644, call->dir, call,
-                                         enable);
+               trace_create_file("enable", 0644, call->dir, call,
+                                 enable);
 
-       if (call->id && call->profile_enable)
-               entry = trace_create_file("id", 0444, call->dir, call,
-                                         id);
+       if (call->id && call->perf_event_enable)
+               trace_create_file("id", 0444, call->dir, call,
+                                 id);
 
        if (call->define_fields) {
-               ret = call->define_fields(call);
+               ret = trace_define_common_fields(call);
+               if (!ret)
+                       ret = call->define_fields(call);
                if (ret < 0) {
                        pr_warning("Could not initialize trace point"
                                   " events/%s\n", call->name);
                        return ret;
                }
-               entry = trace_create_file("filter", 0644, call->dir, call,
-                                         filter);
+               trace_create_file("filter", 0644, call->dir, call,
+                                 filter);
        }
 
-       /* A trace may not want to export its format */
-       if (!call->show_format)
-               return 0;
-
-       entry = trace_create_file("format", 0444, call->dir, call,
-                                 format);
+       trace_create_file("format", 0444, call->dir, call,
+                         format);
 
        return 0;
 }
@@ -1009,12 +984,12 @@ static int __trace_add_event_call(struct ftrace_event_call *call)
        if (!d_events)
                return -ENOENT;
 
-       list_add(&call->list, &ftrace_events);
        ret = event_create_dir(call, d_events, &ftrace_event_id_fops,
                                &ftrace_enable_fops, &ftrace_event_filter_fops,
                                &ftrace_event_format_fops);
-       if (ret < 0)
-               list_del(&call->list);
+       if (!ret)
+               list_add(&call->list, &ftrace_events);
+
        return ret;
 }
 
@@ -1054,6 +1029,9 @@ static void remove_subsystem_dir(const char *name)
        }
 }
 
+/*
+ * Must be called under locking both of event_mutex and trace_event_mutex.
+ */
 static void __trace_remove_event_call(struct ftrace_event_call *call)
 {
        ftrace_event_enable_disable(call, 0);
@@ -1070,7 +1048,9 @@ static void __trace_remove_event_call(struct ftrace_event_call *call)
 void trace_remove_event_call(struct ftrace_event_call *call)
 {
        mutex_lock(&event_mutex);
+       down_write(&trace_event_mutex);
        __trace_remove_event_call(call);
+       up_write(&trace_event_mutex);
        mutex_unlock(&event_mutex);
 }
 
@@ -1171,10 +1151,11 @@ static void trace_module_add_events(struct module *mod)
                                return;
                }
                call->mod = mod;
-               list_add(&call->list, &ftrace_events);
-               event_create_dir(call, d_events,
-                                &file_ops->id, &file_ops->enable,
-                                &file_ops->filter, &file_ops->format);
+               ret = event_create_dir(call, d_events,
+                                      &file_ops->id, &file_ops->enable,
+                                      &file_ops->filter, &file_ops->format);
+               if (!ret)
+                       list_add(&call->list, &ftrace_events);
        }
 }
 
@@ -1237,7 +1218,7 @@ static int trace_module_notify(struct notifier_block *self,
 }
 #endif /* CONFIG_MODULES */
 
-struct notifier_block trace_module_nb = {
+static struct notifier_block trace_module_nb = {
        .notifier_call = trace_module_notify,
        .priority = 0,
 };
@@ -1314,10 +1295,12 @@ static __init int event_trace_init(void)
                                continue;
                        }
                }
-               list_add(&call->list, &ftrace_events);
-               event_create_dir(call, d_events, &ftrace_event_id_fops,
-                                &ftrace_enable_fops, &ftrace_event_filter_fops,
-                                &ftrace_event_format_fops);
+               ret = event_create_dir(call, d_events, &ftrace_event_id_fops,
+                                      &ftrace_enable_fops,
+                                      &ftrace_event_filter_fops,
+                                      &ftrace_event_format_fops);
+               if (!ret)
+                       list_add(&call->list, &ftrace_events);
        }
 
        while (true) {
@@ -1409,6 +1392,18 @@ static __init void event_trace_self_tests(void)
                if (!call->regfunc)
                        continue;
 
+/*
+ * Testing syscall events here is pretty useless, but
+ * we still do it if configured. But this is time consuming.
+ * What we really need is a user thread to perform the
+ * syscalls as we test.
+ */
+#ifndef CONFIG_EVENT_TRACE_TEST_SYSCALLS
+               if (call->system &&
+                   strcmp(call->system, "syscalls") == 0)
+                       continue;
+#endif
+
                pr_info("Testing event %s: ", call->name);
 
                /*
@@ -1482,7 +1477,7 @@ static __init void event_trace_self_tests(void)
 
 #ifdef CONFIG_FUNCTION_TRACER
 
-static DEFINE_PER_CPU(atomic_t, test_event_disable);
+static DEFINE_PER_CPU(atomic_t, ftrace_test_event_disable);
 
 static void
 function_test_events_call(unsigned long ip, unsigned long parent_ip)
@@ -1499,7 +1494,7 @@ function_test_events_call(unsigned long ip, unsigned long parent_ip)
        pc = preempt_count();
        resched = ftrace_preempt_disable();
        cpu = raw_smp_processor_id();
-       disabled = atomic_inc_return(&per_cpu(test_event_disable, cpu));
+       disabled = atomic_inc_return(&per_cpu(ftrace_test_event_disable, cpu));
 
        if (disabled != 1)
                goto out;
@@ -1518,7 +1513,7 @@ function_test_events_call(unsigned long ip, unsigned long parent_ip)
        trace_nowake_buffer_unlock_commit(buffer, event, flags, pc);
 
  out:
-       atomic_dec(&per_cpu(test_event_disable, cpu));
+       atomic_dec(&per_cpu(ftrace_test_event_disable, cpu));
        ftrace_preempt_enable(resched);
 }