sysctl: add proc_do_large_bitmap
[safe/jmp/linux-2.6] / kernel / trace / trace_kprobe.c
index f6821f1..1251e36 100644 (file)
@@ -1,5 +1,5 @@
 /*
- * kprobe based kernel tracer
+ * Kprobes-based tracing events
  *
  * Created by Masami Hiramatsu <mhiramat@redhat.com>
  *
@@ -28,7 +28,7 @@
 #include <linux/string.h>
 #include <linux/ctype.h>
 #include <linux/ptrace.h>
-#include <linux/perf_counter.h>
+#include <linux/perf_event.h>
 
 #include "trace.h"
 #include "trace_output.h"
 #define MAX_EVENT_NAME_LEN 64
 #define KPROBE_EVENT_SYSTEM "kprobes"
 
-/* currently, trace_kprobe only supports X86. */
+/* Reserved field names */
+#define FIELD_STRING_IP "__probe_ip"
+#define FIELD_STRING_NARGS "__probe_nargs"
+#define FIELD_STRING_RETIP "__probe_ret_ip"
+#define FIELD_STRING_FUNC "__probe_func"
+
+const char *reserved_field_names[] = {
+       "common_type",
+       "common_flags",
+       "common_preempt_count",
+       "common_pid",
+       "common_tgid",
+       "common_lock_depth",
+       FIELD_STRING_IP,
+       FIELD_STRING_NARGS,
+       FIELD_STRING_RETIP,
+       FIELD_STRING_FUNC,
+};
 
 struct fetch_func {
        unsigned long (*func)(struct pt_regs *, void *);
@@ -74,22 +91,12 @@ static __kprobes unsigned long fetch_memory(struct pt_regs *regs, void *addr)
        return retval;
 }
 
-static __kprobes unsigned long fetch_argument(struct pt_regs *regs, void *num)
-{
-       return regs_get_argument_nth(regs, (unsigned int)((unsigned long)num));
-}
-
 static __kprobes unsigned long fetch_retvalue(struct pt_regs *regs,
                                              void *dummy)
 {
        return regs_return_value(regs);
 }
 
-static __kprobes unsigned long fetch_ip(struct pt_regs *regs, void *dummy)
-{
-       return instruction_pointer(regs);
-}
-
 static __kprobes unsigned long fetch_stack_address(struct pt_regs *regs,
                                                   void *dummy)
 {
@@ -177,7 +184,7 @@ static __kprobes void free_indirect_fetch_data(struct indirect_fetch_data *data)
 }
 
 /**
- * Kprobe tracer core functions
+ * Kprobe event core functions
  */
 
 struct probe_arg {
@@ -219,25 +226,25 @@ static int probe_arg_string(char *buf, size_t n, struct fetch_func *ff)
 {
        int ret = -EINVAL;
 
-       if (ff->func == fetch_argument)
-               ret = snprintf(buf, n, "a%lu", (unsigned long)ff->data);
-       else if (ff->func == fetch_register) {
+       if (ff->func == fetch_register) {
                const char *name;
                name = regs_query_register_name((unsigned int)((long)ff->data));
                ret = snprintf(buf, n, "%%%s", name);
        } else if (ff->func == fetch_stack)
-               ret = snprintf(buf, n, "s%lu", (unsigned long)ff->data);
+               ret = snprintf(buf, n, "$stack%lu", (unsigned long)ff->data);
        else if (ff->func == fetch_memory)
                ret = snprintf(buf, n, "@0x%p", ff->data);
        else if (ff->func == fetch_symbol) {
                struct symbol_cache *sc = ff->data;
-               ret = snprintf(buf, n, "@%s%+ld", sc->symbol, sc->offset);
+               if (sc->offset)
+                       ret = snprintf(buf, n, "@%s%+ld", sc->symbol,
+                                       sc->offset);
+               else
+                       ret = snprintf(buf, n, "@%s", sc->symbol);
        } else if (ff->func == fetch_retvalue)
-               ret = snprintf(buf, n, "rv");
-       else if (ff->func == fetch_ip)
-               ret = snprintf(buf, n, "ra");
+               ret = snprintf(buf, n, "$retval");
        else if (ff->func == fetch_stack_address)
-               ret = snprintf(buf, n, "sa");
+               ret = snprintf(buf, n, "$stack");
        else if (ff->func == fetch_indirect) {
                struct indirect_fetch_data *id = ff->data;
                size_t l = 0;
@@ -268,6 +275,18 @@ static int kprobe_dispatcher(struct kprobe *kp, struct pt_regs *regs);
 static int kretprobe_dispatcher(struct kretprobe_instance *ri,
                                struct pt_regs *regs);
 
+/* Check the name is good for event/group */
+static int check_event_name(const char *name)
+{
+       if (!isalpha(*name) && *name != '_')
+               return 0;
+       while (*++name != '\0') {
+               if (!isalpha(*name) && !isdigit(*name) && *name != '_')
+                       return 0;
+       }
+       return 1;
+}
+
 /*
  * Allocate new trace_probe and initialize it (including kprobes).
  */
@@ -279,10 +298,11 @@ static struct trace_probe *alloc_trace_probe(const char *group,
                                             int nargs, int is_return)
 {
        struct trace_probe *tp;
+       int ret = -ENOMEM;
 
        tp = kzalloc(SIZEOF_TRACE_PROBE(nargs), GFP_KERNEL);
        if (!tp)
-               return ERR_PTR(-ENOMEM);
+               return ERR_PTR(ret);
 
        if (symbol) {
                tp->symbol = kstrdup(symbol, GFP_KERNEL);
@@ -298,14 +318,20 @@ static struct trace_probe *alloc_trace_probe(const char *group,
        else
                tp->rp.kp.pre_handler = kprobe_dispatcher;
 
-       if (!event)
+       if (!event || !check_event_name(event)) {
+               ret = -EINVAL;
                goto error;
+       }
+
        tp->call.name = kstrdup(event, GFP_KERNEL);
        if (!tp->call.name)
                goto error;
 
-       if (!group)
+       if (!group || !check_event_name(group)) {
+               ret = -EINVAL;
                goto error;
+       }
+
        tp->call.system = kstrdup(group, GFP_KERNEL);
        if (!tp->call.system)
                goto error;
@@ -316,7 +342,7 @@ error:
        kfree(tp->call.name);
        kfree(tp->symbol);
        kfree(tp);
-       return ERR_PTR(-ENOMEM);
+       return ERR_PTR(ret);
 }
 
 static void free_probe_arg(struct probe_arg *arg)
@@ -341,12 +367,14 @@ static void free_trace_probe(struct trace_probe *tp)
        kfree(tp);
 }
 
-static struct trace_probe *find_probe_event(const char *event)
+static struct trace_probe *find_probe_event(const char *event,
+                                           const char *group)
 {
        struct trace_probe *tp;
 
        list_for_each_entry(tp, &probe_list, list)
-               if (!strcmp(tp->call.name, event))
+               if (strcmp(tp->call.name, event) == 0 &&
+                   strcmp(tp->call.system, group) == 0)
                        return tp;
        return NULL;
 }
@@ -371,7 +399,7 @@ static int register_trace_probe(struct trace_probe *tp)
        mutex_lock(&probe_lock);
 
        /* register as an event */
-       old_tp = find_probe_event(tp->call.name);
+       old_tp = find_probe_event(tp->call.name, tp->call.system);
        if (old_tp) {
                /* delete old event */
                unregister_trace_probe(old_tp);
@@ -429,53 +457,54 @@ static int split_symbol_offset(char *symbol, unsigned long *offset)
 #define PARAM_MAX_ARGS 16
 #define PARAM_MAX_STACK (THREAD_SIZE / sizeof(unsigned long))
 
-static int parse_probe_arg(char *arg, struct fetch_func *ff, int is_return)
+static int parse_probe_vars(char *arg, struct fetch_func *ff, int is_return)
 {
        int ret = 0;
        unsigned long param;
-       long offset;
-       char *tmp;
 
-       switch (arg[0]) {
-       case 'a':       /* argument */
-               ret = strict_strtoul(arg + 1, 10, &param);
-               if (ret || param > PARAM_MAX_ARGS)
-                       ret = -EINVAL;
-               else {
-                       ff->func = fetch_argument;
-                       ff->data = (void *)param;
-               }
-               break;
-       case 'r':       /* retval or retaddr */
-               if (is_return && arg[1] == 'v') {
+       if (strcmp(arg, "retval") == 0) {
+               if (is_return) {
                        ff->func = fetch_retvalue;
                        ff->data = NULL;
-               } else if (is_return && arg[1] == 'a') {
-                       ff->func = fetch_ip;
-                       ff->data = NULL;
                } else
                        ret = -EINVAL;
-               break;
-       case '%':       /* named register */
-               ret = regs_query_register_offset(arg + 1);
-               if (ret >= 0) {
-                       ff->func = fetch_register;
-                       ff->data = (void *)(unsigned long)ret;
-                       ret = 0;
-               }
-               break;
-       case 's':       /* stack */
-               if (arg[1] == 'a') {
+       } else if (strncmp(arg, "stack", 5) == 0) {
+               if (arg[5] == '\0') {
                        ff->func = fetch_stack_address;
                        ff->data = NULL;
-               } else {
-                       ret = strict_strtoul(arg + 1, 10, &param);
+               } else if (isdigit(arg[5])) {
+                       ret = strict_strtoul(arg + 5, 10, &param);
                        if (ret || param > PARAM_MAX_STACK)
                                ret = -EINVAL;
                        else {
                                ff->func = fetch_stack;
                                ff->data = (void *)param;
                        }
+               } else
+                       ret = -EINVAL;
+       } else
+               ret = -EINVAL;
+       return ret;
+}
+
+/* Recursive argument parser */
+static int __parse_probe_arg(char *arg, struct fetch_func *ff, int is_return)
+{
+       int ret = 0;
+       unsigned long param;
+       long offset;
+       char *tmp;
+
+       switch (arg[0]) {
+       case '$':
+               ret = parse_probe_vars(arg + 1, ff, is_return);
+               break;
+       case '%':       /* named register */
+               ret = regs_query_register_offset(arg + 1);
+               if (ret >= 0) {
+                       ff->func = fetch_register;
+                       ff->data = (void *)(unsigned long)ret;
+                       ret = 0;
                }
                break;
        case '@':       /* memory or symbol */
@@ -489,8 +518,7 @@ static int parse_probe_arg(char *arg, struct fetch_func *ff, int is_return)
                        ret = split_symbol_offset(arg + 1, &offset);
                        if (ret)
                                break;
-                       ff->data = alloc_symbol_cache(arg + 1,
-                                                             offset);
+                       ff->data = alloc_symbol_cache(arg + 1, offset);
                        if (ff->data)
                                ff->func = fetch_symbol;
                        else
@@ -520,7 +548,7 @@ static int parse_probe_arg(char *arg, struct fetch_func *ff, int is_return)
                        if (!id)
                                return -ENOMEM;
                        id->offset = offset;
-                       ret = parse_probe_arg(arg, &id->orig, is_return);
+                       ret = __parse_probe_arg(arg, &id->orig, is_return);
                        if (ret)
                                kfree(id);
                        else {
@@ -537,6 +565,30 @@ static int parse_probe_arg(char *arg, struct fetch_func *ff, int is_return)
        return ret;
 }
 
+/* String length checking wrapper */
+static int parse_probe_arg(char *arg, struct fetch_func *ff, int is_return)
+{
+       if (strlen(arg) > MAX_ARGSTR_LEN) {
+               pr_info("Argument is too long.: %s\n",  arg);
+               return -ENOSPC;
+       }
+       return __parse_probe_arg(arg, ff, is_return);
+}
+
+/* Return 1 if name is reserved or already used by another argument */
+static int conflict_field_name(const char *name,
+                              struct probe_arg *args, int narg)
+{
+       int i;
+       for (i = 0; i < ARRAY_SIZE(reserved_field_names); i++)
+               if (strcmp(reserved_field_names[i], name) == 0)
+                       return 1;
+       for (i = 0; i < narg; i++)
+               if (strcmp(args[i].name, name) == 0)
+                       return 1;
+       return 0;
+}
+
 static int create_trace_probe(int argc, char **argv)
 {
        /*
@@ -544,11 +596,9 @@ static int create_trace_probe(int argc, char **argv)
         *  - Add kprobe: p[:[GRP/]EVENT] KSYM[+OFFS]|KADDR [FETCHARGS]
         *  - Add kretprobe: r[:[GRP/]EVENT] KSYM[+0] [FETCHARGS]
         * Fetch args:
-        *  aN  : fetch Nth of function argument. (N:0-)
-        *  rv  : fetch return value
-        *  ra  : fetch return address
-        *  sa  : fetch stack address
-        *  sN  : fetch Nth of stack (N:0-)
+        *  $retval     : fetch return value
+        *  $stack      : fetch stack address
+        *  $stackN     : fetch Nth of stack (N:0-)
         *  @ADDR       : fetch memory at ADDR (ADDR should be in kernel)
         *  @SYM[+|-offs] : fetch memory at SYM +|- offs (SYM is a data symbol)
         *  %REG        : fetch register REG
@@ -559,21 +609,24 @@ static int create_trace_probe(int argc, char **argv)
         */
        struct trace_probe *tp;
        int i, ret = 0;
-       int is_return = 0;
+       int is_return = 0, is_delete = 0;
        char *symbol = NULL, *event = NULL, *arg = NULL, *group = NULL;
        unsigned long offset = 0;
        void *addr = NULL;
        char buf[MAX_EVENT_NAME_LEN];
 
-       if (argc < 2)
-               return -EINVAL;
-
+       /* argc must be >= 1 */
        if (argv[0][0] == 'p')
                is_return = 0;
        else if (argv[0][0] == 'r')
                is_return = 1;
-       else
+       else if (argv[0][0] == '-')
+               is_delete = 1;
+       else {
+               pr_info("Probe definition must be started with 'p', 'r' or"
+                       " '-'.\n");
                return -EINVAL;
+       }
 
        if (argv[0][1] == ':') {
                event = &argv[0][2];
@@ -582,52 +635,83 @@ static int create_trace_probe(int argc, char **argv)
                        event = strchr(group, '/') + 1;
                        event[-1] = '\0';
                        if (strlen(group) == 0) {
-                               pr_info("Group name is not specifiled\n");
+                               pr_info("Group name is not specified\n");
                                return -EINVAL;
                        }
                }
                if (strlen(event) == 0) {
-                       pr_info("Event name is not specifiled\n");
+                       pr_info("Event name is not specified\n");
+                       return -EINVAL;
+               }
+       }
+       if (!group)
+               group = KPROBE_EVENT_SYSTEM;
+
+       if (is_delete) {
+               if (!event) {
+                       pr_info("Delete command needs an event name.\n");
                        return -EINVAL;
                }
+               tp = find_probe_event(event, group);
+               if (!tp) {
+                       pr_info("Event %s/%s doesn't exist.\n", group, event);
+                       return -ENOENT;
+               }
+               /* delete an event */
+               unregister_trace_probe(tp);
+               free_trace_probe(tp);
+               return 0;
        }
 
+       if (argc < 2) {
+               pr_info("Probe point is not specified.\n");
+               return -EINVAL;
+       }
        if (isdigit(argv[1][0])) {
-               if (is_return)
+               if (is_return) {
+                       pr_info("Return probe point must be a symbol.\n");
                        return -EINVAL;
+               }
                /* an address specified */
-               ret = strict_strtoul(&argv[0][2], 0, (unsigned long *)&addr);
-               if (ret)
+               ret = strict_strtoul(&argv[1][0], 0, (unsigned long *)&addr);
+               if (ret) {
+                       pr_info("Failed to parse address.\n");
                        return ret;
+               }
        } else {
                /* a symbol specified */
                symbol = argv[1];
                /* TODO: support .init module functions */
                ret = split_symbol_offset(symbol, &offset);
-               if (ret)
+               if (ret) {
+                       pr_info("Failed to parse symbol.\n");
                        return ret;
-               if (offset && is_return)
+               }
+               if (offset && is_return) {
+                       pr_info("Return probe must be used without offset.\n");
                        return -EINVAL;
+               }
        }
        argc -= 2; argv += 2;
 
        /* setup a probe */
-       if (!group)
-               group = KPROBE_EVENT_SYSTEM;
        if (!event) {
                /* Make a new event name */
                if (symbol)
-                       snprintf(buf, MAX_EVENT_NAME_LEN, "%c@%s%+ld",
+                       snprintf(buf, MAX_EVENT_NAME_LEN, "%c_%s_%ld",
                                 is_return ? 'r' : 'p', symbol, offset);
                else
-                       snprintf(buf, MAX_EVENT_NAME_LEN, "%c@0x%p",
+                       snprintf(buf, MAX_EVENT_NAME_LEN, "%c_0x%p",
                                 is_return ? 'r' : 'p', addr);
                event = buf;
        }
        tp = alloc_trace_probe(group, event, addr, symbol, offset, argc,
                               is_return);
-       if (IS_ERR(tp))
+       if (IS_ERR(tp)) {
+               pr_info("Failed to allocate trace_probe.(%d)\n",
+                       (int)PTR_ERR(tp));
                return PTR_ERR(tp);
+       }
 
        /* parse arguments */
        ret = 0;
@@ -638,19 +722,32 @@ static int create_trace_probe(int argc, char **argv)
                        *arg++ = '\0';
                else
                        arg = argv[i];
-               tp->args[i].name = kstrdup(argv[i], GFP_KERNEL);
 
-               /* Parse fetch argument */
-               if (strlen(arg) > MAX_ARGSTR_LEN) {
-                       pr_info("Argument%d(%s) is too long.\n", i, arg);
-                       ret = -ENOSPC;
+               if (conflict_field_name(argv[i], tp->args, i)) {
+                       pr_info("Argument%d name '%s' conflicts with "
+                               "another field.\n", i, argv[i]);
+                       ret = -EINVAL;
+                       goto error;
+               }
+
+               tp->args[i].name = kstrdup(argv[i], GFP_KERNEL);
+               if (!tp->args[i].name) {
+                       pr_info("Failed to allocate argument%d name '%s'.\n",
+                               i, argv[i]);
+                       ret = -ENOMEM;
                        goto error;
                }
+
+               /* Parse fetch argument */
                ret = parse_probe_arg(arg, &tp->args[i].fetch, is_return);
-               if (ret)
+               if (ret) {
+                       pr_info("Parse error at argument%d. (%d)\n", i, ret);
+                       kfree(tp->args[i].name);
                        goto error;
+               }
+
+               tp->nr_args++;
        }
-       tp->nr_args = i;
 
        ret = register_trace_probe(tp);
        if (ret)
@@ -701,12 +798,14 @@ static int probes_seq_show(struct seq_file *m, void *v)
        char buf[MAX_ARGSTR_LEN + 1];
 
        seq_printf(m, "%c", probe_is_return(tp) ? 'r' : 'p');
-       seq_printf(m, ":%s", tp->call.name);
+       seq_printf(m, ":%s/%s", tp->call.system, tp->call.name);
 
-       if (tp->symbol)
+       if (!tp->symbol)
+               seq_printf(m, " 0x%p", tp->rp.kp.addr);
+       else if (tp->rp.kp.offset)
                seq_printf(m, " %s+%u", probe_symbol(tp), tp->rp.kp.offset);
        else
-               seq_printf(m, " 0x%p", tp->rp.kp.addr);
+               seq_printf(m, " %s", probe_symbol(tp));
 
        for (i = 0; i < tp->nr_args; i++) {
                ret = probe_arg_string(buf, MAX_ARGSTR_LEN, &tp->args[i].fetch);
@@ -843,7 +942,7 @@ static const struct file_operations kprobe_profile_ops = {
 };
 
 /* Kprobe handler */
-static __kprobes int kprobe_trace_func(struct kprobe *kp, struct pt_regs *regs)
+static __kprobes void kprobe_trace_func(struct kprobe *kp, struct pt_regs *regs)
 {
        struct trace_probe *tp = container_of(kp, struct trace_probe, rp.kp);
        struct kprobe_trace_entry *entry;
@@ -863,7 +962,7 @@ static __kprobes int kprobe_trace_func(struct kprobe *kp, struct pt_regs *regs)
        event = trace_current_buffer_lock_reserve(&buffer, call->id, size,
                                                  irq_flags, pc);
        if (!event)
-               return 0;
+               return;
 
        entry = ring_buffer_event_data(event);
        entry->nargs = tp->nr_args;
@@ -873,11 +972,10 @@ static __kprobes int kprobe_trace_func(struct kprobe *kp, struct pt_regs *regs)
 
        if (!filter_current_check_discard(buffer, call, entry, event))
                trace_nowake_buffer_unlock_commit(buffer, event, irq_flags, pc);
-       return 0;
 }
 
 /* Kretprobe handler */
-static __kprobes int kretprobe_trace_func(struct kretprobe_instance *ri,
+static __kprobes void kretprobe_trace_func(struct kretprobe_instance *ri,
                                          struct pt_regs *regs)
 {
        struct trace_probe *tp = container_of(ri->rp, struct trace_probe, rp);
@@ -896,7 +994,7 @@ static __kprobes int kretprobe_trace_func(struct kretprobe_instance *ri,
        event = trace_current_buffer_lock_reserve(&buffer, call->id, size,
                                                  irq_flags, pc);
        if (!event)
-               return 0;
+               return;
 
        entry = ring_buffer_event_data(event);
        entry->nargs = tp->nr_args;
@@ -907,8 +1005,6 @@ static __kprobes int kretprobe_trace_func(struct kretprobe_instance *ri,
 
        if (!filter_current_check_discard(buffer, call, entry, event))
                trace_nowake_buffer_unlock_commit(buffer, event, irq_flags, pc);
-
-       return 0;
 }
 
 /* Event entry printers */
@@ -1036,12 +1132,8 @@ static int kprobe_event_define_fields(struct ftrace_event_call *event_call)
        struct kprobe_trace_entry field;
        struct trace_probe *tp = (struct trace_probe *)event_call->data;
 
-       ret = trace_define_common_fields(event_call);
-       if (!ret)
-               return ret;
-
-       DEFINE_FIELD(unsigned long, ip, "ip", 0);
-       DEFINE_FIELD(int, nargs, "nargs", 1);
+       DEFINE_FIELD(unsigned long, ip, FIELD_STRING_IP, 0);
+       DEFINE_FIELD(int, nargs, FIELD_STRING_NARGS, 1);
        /* Set argument names as fields */
        for (i = 0; i < tp->nr_args; i++)
                DEFINE_FIELD(unsigned long, args[i], tp->args[i].name, 0);
@@ -1054,199 +1146,161 @@ static int kretprobe_event_define_fields(struct ftrace_event_call *event_call)
        struct kretprobe_trace_entry field;
        struct trace_probe *tp = (struct trace_probe *)event_call->data;
 
-       ret = trace_define_common_fields(event_call);
-       if (!ret)
-               return ret;
-
-       DEFINE_FIELD(unsigned long, func, "func", 0);
-       DEFINE_FIELD(unsigned long, ret_ip, "ret_ip", 0);
-       DEFINE_FIELD(int, nargs, "nargs", 1);
+       DEFINE_FIELD(unsigned long, func, FIELD_STRING_FUNC, 0);
+       DEFINE_FIELD(unsigned long, ret_ip, FIELD_STRING_RETIP, 0);
+       DEFINE_FIELD(int, nargs, FIELD_STRING_NARGS, 1);
        /* Set argument names as fields */
        for (i = 0; i < tp->nr_args; i++)
                DEFINE_FIELD(unsigned long, args[i], tp->args[i].name, 0);
        return 0;
 }
 
-static int __probe_event_show_format(struct trace_seq *s,
-                                    struct trace_probe *tp, const char *fmt,
-                                    const char *arg)
+static int __set_print_fmt(struct trace_probe *tp, char *buf, int len)
 {
        int i;
+       int pos = 0;
 
-       /* Show format */
-       if (!trace_seq_printf(s, "\nprint fmt: \"%s", fmt))
-               return 0;
-
-       for (i = 0; i < tp->nr_args; i++)
-               if (!trace_seq_printf(s, " %s=%%lx", tp->args[i].name))
-                       return 0;
+       const char *fmt, *arg;
 
-       if (!trace_seq_printf(s, "\", %s", arg))
-               return 0;
+       if (!probe_is_return(tp)) {
+               fmt = "(%lx)";
+               arg = "REC->" FIELD_STRING_IP;
+       } else {
+               fmt = "(%lx <- %lx)";
+               arg = "REC->" FIELD_STRING_FUNC ", REC->" FIELD_STRING_RETIP;
+       }
 
-       for (i = 0; i < tp->nr_args; i++)
-               if (!trace_seq_printf(s, ", REC->%s", tp->args[i].name))
-                       return 0;
+       /* When len=0, we just calculate the needed length */
+#define LEN_OR_ZERO (len ? len - pos : 0)
 
-       return trace_seq_puts(s, "\n");
-}
+       pos += snprintf(buf + pos, LEN_OR_ZERO, "\"%s", fmt);
 
-#undef SHOW_FIELD
-#define SHOW_FIELD(type, item, name)                                   \
-       do {                                                            \
-               ret = trace_seq_printf(s, "\tfield: " #type " %s;\t"    \
-                               "offset:%u;\tsize:%u;\n", name,         \
-                               (unsigned int)offsetof(typeof(field), item),\
-                               (unsigned int)sizeof(type));            \
-               if (!ret)                                               \
-                       return 0;                                       \
-       } while (0)
+       for (i = 0; i < tp->nr_args; i++) {
+               pos += snprintf(buf + pos, LEN_OR_ZERO, " %s=%%lx",
+                               tp->args[i].name);
+       }
 
-static int kprobe_event_show_format(struct ftrace_event_call *call,
-                                   struct trace_seq *s)
-{
-       struct kprobe_trace_entry field __attribute__((unused));
-       int ret, i;
-       struct trace_probe *tp = (struct trace_probe *)call->data;
+       pos += snprintf(buf + pos, LEN_OR_ZERO, "\", %s", arg);
 
-       SHOW_FIELD(unsigned long, ip, "ip");
-       SHOW_FIELD(int, nargs, "nargs");
+       for (i = 0; i < tp->nr_args; i++) {
+               pos += snprintf(buf + pos, LEN_OR_ZERO, ", REC->%s",
+                               tp->args[i].name);
+       }
 
-       /* Show fields */
-       for (i = 0; i < tp->nr_args; i++)
-               SHOW_FIELD(unsigned long, args[i], tp->args[i].name);
-       trace_seq_puts(s, "\n");
+#undef LEN_OR_ZERO
 
-       return __probe_event_show_format(s, tp, "(%lx)", "REC->ip");
+       /* return the length of print_fmt */
+       return pos;
 }
 
-static int kretprobe_event_show_format(struct ftrace_event_call *call,
-                                      struct trace_seq *s)
+static int set_print_fmt(struct trace_probe *tp)
 {
-       struct kretprobe_trace_entry field __attribute__((unused));
-       int ret, i;
-       struct trace_probe *tp = (struct trace_probe *)call->data;
+       int len;
+       char *print_fmt;
 
-       SHOW_FIELD(unsigned long, func, "func");
-       SHOW_FIELD(unsigned long, ret_ip, "ret_ip");
-       SHOW_FIELD(int, nargs, "nargs");
+       /* First: called with 0 length to calculate the needed length */
+       len = __set_print_fmt(tp, NULL, 0);
+       print_fmt = kmalloc(len + 1, GFP_KERNEL);
+       if (!print_fmt)
+               return -ENOMEM;
 
-       /* Show fields */
-       for (i = 0; i < tp->nr_args; i++)
-               SHOW_FIELD(unsigned long, args[i], tp->args[i].name);
-       trace_seq_puts(s, "\n");
+       /* Second: actually write the @print_fmt */
+       __set_print_fmt(tp, print_fmt, len + 1);
+       tp->call.print_fmt = print_fmt;
 
-       return __probe_event_show_format(s, tp, "(%lx <- %lx)",
-                                         "REC->func, REC->ret_ip");
+       return 0;
 }
 
-#ifdef CONFIG_EVENT_PROFILE
+#ifdef CONFIG_PERF_EVENTS
 
 /* Kprobe profile handler */
-static __kprobes int kprobe_profile_func(struct kprobe *kp,
+static __kprobes void kprobe_perf_func(struct kprobe *kp,
                                         struct pt_regs *regs)
 {
        struct trace_probe *tp = container_of(kp, struct trace_probe, rp.kp);
        struct ftrace_event_call *call = &tp->call;
        struct kprobe_trace_entry *entry;
-       int size, __size, i, pc;
+       int size, __size, i;
        unsigned long irq_flags;
-
-       local_save_flags(irq_flags);
-       pc = preempt_count();
+       int rctx;
 
        __size = SIZEOF_KPROBE_TRACE_ENTRY(tp->nr_args);
        size = ALIGN(__size + sizeof(u32), sizeof(u64));
        size -= sizeof(u32);
+       if (WARN_ONCE(size > PERF_MAX_TRACE_SIZE,
+                    "profile buffer not large enough"))
+               return;
 
-       do {
-               char raw_data[size];
-               struct trace_entry *ent;
-               /*
-                * Zero dead bytes from alignment to avoid stack leak
-                * to userspace
-                */
-               *(u64 *)(&raw_data[size - sizeof(u64)]) = 0ULL;
-               entry = (struct kprobe_trace_entry *)raw_data;
-               ent = &entry->ent;
-
-               tracing_generic_entry_update(ent, irq_flags, pc);
-               ent->type = call->id;
-               entry->nargs = tp->nr_args;
-               entry->ip = (unsigned long)kp->addr;
-               for (i = 0; i < tp->nr_args; i++)
-                       entry->args[i] = call_fetch(&tp->args[i].fetch, regs);
-               perf_tpcounter_event(call->id, entry->ip, 1, entry, size);
-       } while (0);
-       return 0;
+       entry = perf_trace_buf_prepare(size, call->id, &rctx, &irq_flags);
+       if (!entry)
+               return;
+
+       entry->nargs = tp->nr_args;
+       entry->ip = (unsigned long)kp->addr;
+       for (i = 0; i < tp->nr_args; i++)
+               entry->args[i] = call_fetch(&tp->args[i].fetch, regs);
+
+       perf_trace_buf_submit(entry, size, rctx, entry->ip, 1, irq_flags, regs);
 }
 
 /* Kretprobe profile handler */
-static __kprobes int kretprobe_profile_func(struct kretprobe_instance *ri,
+static __kprobes void kretprobe_perf_func(struct kretprobe_instance *ri,
                                            struct pt_regs *regs)
 {
        struct trace_probe *tp = container_of(ri->rp, struct trace_probe, rp);
        struct ftrace_event_call *call = &tp->call;
        struct kretprobe_trace_entry *entry;
-       int size, __size, i, pc;
+       int size, __size, i;
        unsigned long irq_flags;
-
-       local_save_flags(irq_flags);
-       pc = preempt_count();
+       int rctx;
 
        __size = SIZEOF_KRETPROBE_TRACE_ENTRY(tp->nr_args);
        size = ALIGN(__size + sizeof(u32), sizeof(u64));
        size -= sizeof(u32);
+       if (WARN_ONCE(size > PERF_MAX_TRACE_SIZE,
+                    "profile buffer not large enough"))
+               return;
 
-       do {
-               char raw_data[size];
-               struct trace_entry *ent;
-
-               *(u64 *)(&raw_data[size - sizeof(u64)]) = 0ULL;
-               entry = (struct kretprobe_trace_entry *)raw_data;
-               ent = &entry->ent;
-
-               tracing_generic_entry_update(ent, irq_flags, pc);
-               ent->type = call->id;
-               entry->nargs = tp->nr_args;
-               entry->func = (unsigned long)tp->rp.kp.addr;
-               entry->ret_ip = (unsigned long)ri->ret_addr;
-               for (i = 0; i < tp->nr_args; i++)
-                       entry->args[i] = call_fetch(&tp->args[i].fetch, regs);
-               perf_tpcounter_event(call->id, entry->ret_ip, 1, entry, size);
-       } while (0);
-       return 0;
+       entry = perf_trace_buf_prepare(size, call->id, &rctx, &irq_flags);
+       if (!entry)
+               return;
+
+       entry->nargs = tp->nr_args;
+       entry->func = (unsigned long)tp->rp.kp.addr;
+       entry->ret_ip = (unsigned long)ri->ret_addr;
+       for (i = 0; i < tp->nr_args; i++)
+               entry->args[i] = call_fetch(&tp->args[i].fetch, regs);
+
+       perf_trace_buf_submit(entry, size, rctx, entry->ret_ip, 1,
+                              irq_flags, regs);
 }
 
-static int probe_profile_enable(struct ftrace_event_call *call)
+static int probe_perf_enable(struct ftrace_event_call *call)
 {
        struct trace_probe *tp = (struct trace_probe *)call->data;
 
-       if (atomic_inc_return(&call->profile_count))
-               return 0;
-
        tp->flags |= TP_FLAG_PROFILE;
+
        if (probe_is_return(tp))
                return enable_kretprobe(&tp->rp);
        else
                return enable_kprobe(&tp->rp.kp);
 }
 
-static void probe_profile_disable(struct ftrace_event_call *call)
+static void probe_perf_disable(struct ftrace_event_call *call)
 {
        struct trace_probe *tp = (struct trace_probe *)call->data;
 
-       if (atomic_add_negative(-1, &call->profile_count))
-               tp->flags &= ~TP_FLAG_PROFILE;
+       tp->flags &= ~TP_FLAG_PROFILE;
 
-       if (!(tp->flags & (TP_FLAG_TRACE | TP_FLAG_PROFILE))) {
+       if (!(tp->flags & TP_FLAG_TRACE)) {
                if (probe_is_return(tp))
                        disable_kretprobe(&tp->rp);
                else
                        disable_kprobe(&tp->rp.kp);
        }
 }
-#endif /* CONFIG_EVENT_PROFILE */
+#endif /* CONFIG_PERF_EVENTS */
 
 
 static __kprobes
@@ -1256,10 +1310,10 @@ int kprobe_dispatcher(struct kprobe *kp, struct pt_regs *regs)
 
        if (tp->flags & TP_FLAG_TRACE)
                kprobe_trace_func(kp, regs);
-#ifdef CONFIG_EVENT_PROFILE
+#ifdef CONFIG_PERF_EVENTS
        if (tp->flags & TP_FLAG_PROFILE)
-               kprobe_profile_func(kp, regs);
-#endif /* CONFIG_EVENT_PROFILE */
+               kprobe_perf_func(kp, regs);
+#endif
        return 0;       /* We don't tweek kernel, so just return 0 */
 }
 
@@ -1270,10 +1324,10 @@ int kretprobe_dispatcher(struct kretprobe_instance *ri, struct pt_regs *regs)
 
        if (tp->flags & TP_FLAG_TRACE)
                kretprobe_trace_func(ri, regs);
-#ifdef CONFIG_EVENT_PROFILE
+#ifdef CONFIG_PERF_EVENTS
        if (tp->flags & TP_FLAG_PROFILE)
-               kretprobe_profile_func(ri, regs);
-#endif /* CONFIG_EVENT_PROFILE */
+               kretprobe_perf_func(ri, regs);
+#endif
        return 0;       /* We don't tweek kernel, so just return 0 */
 }
 
@@ -1286,31 +1340,33 @@ static int register_probe_event(struct trace_probe *tp)
        if (probe_is_return(tp)) {
                tp->event.trace = print_kretprobe_event;
                call->raw_init = probe_event_raw_init;
-               call->show_format = kretprobe_event_show_format;
                call->define_fields = kretprobe_event_define_fields;
        } else {
                tp->event.trace = print_kprobe_event;
                call->raw_init = probe_event_raw_init;
-               call->show_format = kprobe_event_show_format;
                call->define_fields = kprobe_event_define_fields;
        }
+       if (set_print_fmt(tp) < 0)
+               return -ENOMEM;
        call->event = &tp->event;
        call->id = register_ftrace_event(&tp->event);
-       if (!call->id)
+       if (!call->id) {
+               kfree(call->print_fmt);
                return -ENODEV;
+       }
        call->enabled = 0;
        call->regfunc = probe_event_enable;
        call->unregfunc = probe_event_disable;
 
-#ifdef CONFIG_EVENT_PROFILE
-       atomic_set(&call->profile_count, -1);
-       call->profile_enable = probe_profile_enable;
-       call->profile_disable = probe_profile_disable;
+#ifdef CONFIG_PERF_EVENTS
+       call->perf_event_enable = probe_perf_enable;
+       call->perf_event_disable = probe_perf_disable;
 #endif
        call->data = tp;
        ret = trace_add_event_call(call);
        if (ret) {
                pr_info("Failed to register kprobe event: %s\n", call->name);
+               kfree(call->print_fmt);
                unregister_ftrace_event(&tp->event);
        }
        return ret;
@@ -1320,6 +1376,7 @@ static void unregister_probe_event(struct trace_probe *tp)
 {
        /* tp->event is unregistered in trace_remove_event_call() */
        trace_remove_event_call(&tp->call);
+       kfree(tp->call.print_fmt);
 }
 
 /* Make a debugfs interface for controling probe points */
@@ -1362,28 +1419,67 @@ static int kprobe_trace_selftest_target(int a1, int a2, int a3,
 
 static __init int kprobe_trace_self_tests_init(void)
 {
-       int ret;
+       int ret, warn = 0;
        int (*target)(int, int, int, int, int, int);
+       struct trace_probe *tp;
 
        target = kprobe_trace_selftest_target;
 
        pr_info("Testing kprobe tracing: ");
 
        ret = command_trace_probe("p:testprobe kprobe_trace_selftest_target "
-                                 "a1 a2 a3 a4 a5 a6");
-       if (WARN_ON_ONCE(ret))
-               pr_warning("error enabling function entry\n");
+                                 "$stack $stack0 +0($stack)");
+       if (WARN_ON_ONCE(ret)) {
+               pr_warning("error on probing function entry.\n");
+               warn++;
+       } else {
+               /* Enable trace point */
+               tp = find_probe_event("testprobe", KPROBE_EVENT_SYSTEM);
+               if (WARN_ON_ONCE(tp == NULL)) {
+                       pr_warning("error on getting new probe.\n");
+                       warn++;
+               } else
+                       probe_event_enable(&tp->call);
+       }
 
        ret = command_trace_probe("r:testprobe2 kprobe_trace_selftest_target "
-                                 "ra rv");
-       if (WARN_ON_ONCE(ret))
-               pr_warning("error enabling function return\n");
+                                 "$retval");
+       if (WARN_ON_ONCE(ret)) {
+               pr_warning("error on probing function return.\n");
+               warn++;
+       } else {
+               /* Enable trace point */
+               tp = find_probe_event("testprobe2", KPROBE_EVENT_SYSTEM);
+               if (WARN_ON_ONCE(tp == NULL)) {
+                       pr_warning("error on getting new probe.\n");
+                       warn++;
+               } else
+                       probe_event_enable(&tp->call);
+       }
+
+       if (warn)
+               goto end;
 
        ret = target(1, 2, 3, 4, 5, 6);
 
-       cleanup_all_probes();
+       ret = command_trace_probe("-:testprobe");
+       if (WARN_ON_ONCE(ret)) {
+               pr_warning("error on deleting a probe.\n");
+               warn++;
+       }
+
+       ret = command_trace_probe("-:testprobe2");
+       if (WARN_ON_ONCE(ret)) {
+               pr_warning("error on deleting a probe.\n");
+               warn++;
+       }
 
-       pr_cont("OK\n");
+end:
+       cleanup_all_probes();
+       if (warn)
+               pr_cont("NG: Some tests are failed. Please check them.\n");
+       else
+               pr_cont("OK\n");
        return 0;
 }