tracing/stat: remember to free root node
[safe/jmp/linux-2.6] / kernel / trace / trace.c
index 2f994ca..a3a8a87 100644 (file)
@@ -30,6 +30,7 @@
 #include <linux/percpu.h>
 #include <linux/splice.h>
 #include <linux/kdebug.h>
+#include <linux/string.h>
 #include <linux/ctype.h>
 #include <linux/init.h>
 #include <linux/poll.h>
@@ -147,8 +148,7 @@ static int __init set_ftrace_dump_on_oops(char *str)
 }
 __setup("ftrace_dump_on_oops", set_ftrace_dump_on_oops);
 
-long
-ns2usecs(cycle_t nsec)
+unsigned long long ns2usecs(cycle_t nsec)
 {
        nsec += 500;
        do_div(nsec, 1000);
@@ -171,6 +171,13 @@ static struct trace_array  global_trace;
 
 static DEFINE_PER_CPU(struct trace_array_cpu, global_trace_cpu);
 
+int filter_current_check_discard(struct ftrace_event_call *call, void *rec,
+                                struct ring_buffer_event *event)
+{
+       return filter_check_discard(call, rec, global_trace.buffer, event);
+}
+EXPORT_SYMBOL_GPL(filter_current_check_discard);
+
 cycle_t ftrace_now(int cpu)
 {
        u64 ts;
@@ -255,7 +262,8 @@ static DECLARE_WAIT_QUEUE_HEAD(trace_wait);
 
 /* trace_flags holds trace_options default values */
 unsigned long trace_flags = TRACE_ITER_PRINT_PARENT | TRACE_ITER_PRINTK |
-       TRACE_ITER_ANNOTATE | TRACE_ITER_CONTEXT_INFO;
+       TRACE_ITER_ANNOTATE | TRACE_ITER_CONTEXT_INFO | TRACE_ITER_SLEEP_TIME |
+       TRACE_ITER_GRAPH_TIME;
 
 /**
  * trace_wake_up - wake up tasks waiting for trace input
@@ -316,6 +324,8 @@ static const char *trace_options[] = {
        "context-info",
        "latency-format",
        "global-clock",
+       "sleep-time",
+       "graph-time",
        NULL
 };
 
@@ -382,7 +392,7 @@ ssize_t trace_seq_to_user(struct trace_seq *s, char __user *ubuf, size_t cnt)
        return cnt;
 }
 
-ssize_t trace_seq_to_buffer(struct trace_seq *s, void *buf, size_t cnt)
+static ssize_t trace_seq_to_buffer(struct trace_seq *s, void *buf, size_t cnt)
 {
        int len;
        void *ret;
@@ -401,17 +411,6 @@ ssize_t trace_seq_to_buffer(struct trace_seq *s, void *buf, size_t cnt)
        return cnt;
 }
 
-static void
-trace_print_seq(struct seq_file *m, struct trace_seq *s)
-{
-       int len = s->len >= PAGE_SIZE ? PAGE_SIZE - 1 : s->len;
-
-       s->buffer[len] = 0;
-       seq_puts(m, s->buffer);
-
-       trace_seq_init(s);
-}
-
 /**
  * update_max_tr - snapshot all trace buffers from global_trace to max_tr
  * @tr: tracer
@@ -640,7 +639,18 @@ void tracing_reset_online_cpus(struct trace_array *tr)
                tracing_reset(tr, cpu);
 }
 
+void tracing_reset_current(int cpu)
+{
+       tracing_reset(&global_trace, cpu);
+}
+
+void tracing_reset_current_online_cpus(void)
+{
+       tracing_reset_online_cpus(&global_trace);
+}
+
 #define SAVED_CMDLINES 128
+#define NO_CMDLINE_MAP UINT_MAX
 static unsigned map_pid_to_cmdline[PID_MAX_DEFAULT+1];
 static unsigned map_cmdline_to_pid[SAVED_CMDLINES];
 static char saved_cmdlines[SAVED_CMDLINES][TASK_COMM_LEN];
@@ -652,8 +662,8 @@ static atomic_t trace_record_cmdline_disabled __read_mostly;
 
 static void trace_init_cmdlines(void)
 {
-       memset(&map_pid_to_cmdline, -1, sizeof(map_pid_to_cmdline));
-       memset(&map_cmdline_to_pid, -1, sizeof(map_cmdline_to_pid));
+       memset(&map_pid_to_cmdline, NO_CMDLINE_MAP, sizeof(map_pid_to_cmdline));
+       memset(&map_cmdline_to_pid, NO_CMDLINE_MAP, sizeof(map_cmdline_to_pid));
        cmdline_idx = 0;
 }
 
@@ -745,8 +755,7 @@ void trace_stop_cmdline_recording(void);
 
 static void trace_save_cmdline(struct task_struct *tsk)
 {
-       unsigned map;
-       unsigned idx;
+       unsigned pid, idx;
 
        if (!tsk->pid || unlikely(tsk->pid > PID_MAX_DEFAULT))
                return;
@@ -761,13 +770,20 @@ static void trace_save_cmdline(struct task_struct *tsk)
                return;
 
        idx = map_pid_to_cmdline[tsk->pid];
-       if (idx >= SAVED_CMDLINES) {
+       if (idx == NO_CMDLINE_MAP) {
                idx = (cmdline_idx + 1) % SAVED_CMDLINES;
 
-               map = map_cmdline_to_pid[idx];
-               if (map <= PID_MAX_DEFAULT)
-                       map_pid_to_cmdline[map] = (unsigned)-1;
+               /*
+                * Check whether the cmdline buffer at idx has a pid
+                * mapped. We are going to overwrite that entry so we
+                * need to clear the map_pid_to_cmdline. Otherwise we
+                * would read the new comm for the old pid.
+                */
+               pid = map_cmdline_to_pid[idx];
+               if (pid != NO_CMDLINE_MAP)
+                       map_pid_to_cmdline[pid] = NO_CMDLINE_MAP;
 
+               map_cmdline_to_pid[idx] = tsk->pid;
                map_pid_to_cmdline[tsk->pid] = idx;
 
                cmdline_idx = idx;
@@ -792,20 +808,22 @@ void trace_find_cmdline(int pid, char comm[])
                return;
        }
 
+       preempt_disable();
        __raw_spin_lock(&trace_cmdline_lock);
        map = map_pid_to_cmdline[pid];
-       if (map >= SAVED_CMDLINES)
-               goto out;
-
-       strcpy(comm, saved_cmdlines[map]);
+       if (map != NO_CMDLINE_MAP)
+               strcpy(comm, saved_cmdlines[map]);
+       else
+               strcpy(comm, "<...>");
 
- out:
        __raw_spin_unlock(&trace_cmdline_lock);
+       preempt_enable();
 }
 
 void tracing_record_cmdline(struct task_struct *tsk)
 {
-       if (atomic_read(&trace_record_cmdline_disabled) || !tracing_is_on())
+       if (atomic_read(&trace_record_cmdline_disabled) || !tracer_enabled ||
+           !tracing_is_on())
                return;
 
        trace_save_cmdline(tsk);
@@ -832,7 +850,7 @@ tracing_generic_entry_update(struct trace_entry *entry, unsigned long flags,
 }
 
 struct ring_buffer_event *trace_buffer_lock_reserve(struct trace_array *tr,
-                                                   unsigned char type,
+                                                   int type,
                                                    unsigned long len,
                                                    unsigned long flags, int pc)
 {
@@ -853,36 +871,62 @@ static void ftrace_trace_stack(struct trace_array *tr,
 static void ftrace_trace_userstack(struct trace_array *tr,
                                   unsigned long flags, int pc);
 
-void trace_buffer_unlock_commit(struct trace_array *tr,
-                               struct ring_buffer_event *event,
-                               unsigned long flags, int pc)
+static inline void __trace_buffer_unlock_commit(struct trace_array *tr,
+                                       struct ring_buffer_event *event,
+                                       unsigned long flags, int pc,
+                                       int wake)
 {
        ring_buffer_unlock_commit(tr->buffer, event);
 
        ftrace_trace_stack(tr, flags, 6, pc);
        ftrace_trace_userstack(tr, flags, pc);
-       trace_wake_up();
+
+       if (wake)
+               trace_wake_up();
+}
+
+void trace_buffer_unlock_commit(struct trace_array *tr,
+                                       struct ring_buffer_event *event,
+                                       unsigned long flags, int pc)
+{
+       __trace_buffer_unlock_commit(tr, event, flags, pc, 1);
 }
 
 struct ring_buffer_event *
-trace_current_buffer_lock_reserve(unsigned char type, unsigned long len,
+trace_current_buffer_lock_reserve(int type, unsigned long len,
                                  unsigned long flags, int pc)
 {
        return trace_buffer_lock_reserve(&global_trace,
                                         type, len, flags, pc);
 }
+EXPORT_SYMBOL_GPL(trace_current_buffer_lock_reserve);
 
 void trace_current_buffer_unlock_commit(struct ring_buffer_event *event,
                                        unsigned long flags, int pc)
 {
-       return trace_buffer_unlock_commit(&global_trace, event, flags, pc);
+       __trace_buffer_unlock_commit(&global_trace, event, flags, pc, 1);
+}
+EXPORT_SYMBOL_GPL(trace_current_buffer_unlock_commit);
+
+void trace_nowake_buffer_unlock_commit(struct ring_buffer_event *event,
+                                       unsigned long flags, int pc)
+{
+       __trace_buffer_unlock_commit(&global_trace, event, flags, pc, 0);
+}
+EXPORT_SYMBOL_GPL(trace_nowake_buffer_unlock_commit);
+
+void trace_current_buffer_discard_commit(struct ring_buffer_event *event)
+{
+       ring_buffer_discard_commit(global_trace.buffer, event);
 }
+EXPORT_SYMBOL_GPL(trace_current_buffer_discard_commit);
 
 void
 trace_function(struct trace_array *tr,
               unsigned long ip, unsigned long parent_ip, unsigned long flags,
               int pc)
 {
+       struct ftrace_event_call *call = &event_function;
        struct ring_buffer_event *event;
        struct ftrace_entry *entry;
 
@@ -897,28 +941,34 @@ trace_function(struct trace_array *tr,
        entry   = ring_buffer_event_data(event);
        entry->ip                       = ip;
        entry->parent_ip                = parent_ip;
-       ring_buffer_unlock_commit(tr->buffer, event);
+
+       if (!filter_check_discard(call, entry, tr->buffer, event))
+               ring_buffer_unlock_commit(tr->buffer, event);
 }
 
 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
-static void __trace_graph_entry(struct trace_array *tr,
+static int __trace_graph_entry(struct trace_array *tr,
                                struct ftrace_graph_ent *trace,
                                unsigned long flags,
                                int pc)
 {
+       struct ftrace_event_call *call = &event_funcgraph_entry;
        struct ring_buffer_event *event;
        struct ftrace_graph_ent_entry *entry;
 
        if (unlikely(local_read(&__get_cpu_var(ftrace_cpu_disabled))))
-               return;
+               return 0;
 
        event = trace_buffer_lock_reserve(&global_trace, TRACE_GRAPH_ENT,
                                          sizeof(*entry), flags, pc);
        if (!event)
-               return;
+               return 0;
        entry   = ring_buffer_event_data(event);
        entry->graph_ent                        = *trace;
-       ring_buffer_unlock_commit(global_trace.buffer, event);
+       if (!filter_current_check_discard(call, entry, event))
+               ring_buffer_unlock_commit(global_trace.buffer, event);
+
+       return 1;
 }
 
 static void __trace_graph_return(struct trace_array *tr,
@@ -926,6 +976,7 @@ static void __trace_graph_return(struct trace_array *tr,
                                unsigned long flags,
                                int pc)
 {
+       struct ftrace_event_call *call = &event_funcgraph_exit;
        struct ring_buffer_event *event;
        struct ftrace_graph_ret_entry *entry;
 
@@ -938,7 +989,8 @@ static void __trace_graph_return(struct trace_array *tr,
                return;
        entry   = ring_buffer_event_data(event);
        entry->ret                              = *trace;
-       ring_buffer_unlock_commit(global_trace.buffer, event);
+       if (!filter_current_check_discard(call, entry, event))
+               ring_buffer_unlock_commit(global_trace.buffer, event);
 }
 #endif
 
@@ -956,6 +1008,7 @@ static void __ftrace_trace_stack(struct trace_array *tr,
                                 int skip, int pc)
 {
 #ifdef CONFIG_STACKTRACE
+       struct ftrace_event_call *call = &event_kernel_stack;
        struct ring_buffer_event *event;
        struct stack_entry *entry;
        struct stack_trace trace;
@@ -973,7 +1026,8 @@ static void __ftrace_trace_stack(struct trace_array *tr,
        trace.entries           = entry->caller;
 
        save_stack_trace(&trace);
-       ring_buffer_unlock_commit(tr->buffer, event);
+       if (!filter_check_discard(call, entry, tr->buffer, event))
+               ring_buffer_unlock_commit(tr->buffer, event);
 #endif
 }
 
@@ -998,6 +1052,7 @@ static void ftrace_trace_userstack(struct trace_array *tr,
                                   unsigned long flags, int pc)
 {
 #ifdef CONFIG_STACKTRACE
+       struct ftrace_event_call *call = &event_user_stack;
        struct ring_buffer_event *event;
        struct userstack_entry *entry;
        struct stack_trace trace;
@@ -1019,7 +1074,8 @@ static void ftrace_trace_userstack(struct trace_array *tr,
        trace.entries           = entry->caller;
 
        save_stack_trace_user(&trace);
-       ring_buffer_unlock_commit(tr->buffer, event);
+       if (!filter_check_discard(call, entry, tr->buffer, event))
+               ring_buffer_unlock_commit(tr->buffer, event);
 #endif
 }
 
@@ -1063,6 +1119,7 @@ tracing_sched_switch_trace(struct trace_array *tr,
                           struct task_struct *next,
                           unsigned long flags, int pc)
 {
+       struct ftrace_event_call *call = &event_context_switch;
        struct ring_buffer_event *event;
        struct ctx_switch_entry *entry;
 
@@ -1078,7 +1135,9 @@ tracing_sched_switch_trace(struct trace_array *tr,
        entry->next_prio                = next->prio;
        entry->next_state               = next->state;
        entry->next_cpu = task_cpu(next);
-       trace_buffer_unlock_commit(tr, event, flags, pc);
+
+       if (!filter_check_discard(call, entry, tr->buffer, event))
+               trace_buffer_unlock_commit(tr, event, flags, pc);
 }
 
 void
@@ -1087,6 +1146,7 @@ tracing_sched_wakeup_trace(struct trace_array *tr,
                           struct task_struct *curr,
                           unsigned long flags, int pc)
 {
+       struct ftrace_event_call *call = &event_wakeup;
        struct ring_buffer_event *event;
        struct ctx_switch_entry *entry;
 
@@ -1103,7 +1163,8 @@ tracing_sched_wakeup_trace(struct trace_array *tr,
        entry->next_state               = wakee->state;
        entry->next_cpu                 = task_cpu(wakee);
 
-       ring_buffer_unlock_commit(tr->buffer, event);
+       if (!filter_check_discard(call, entry, tr->buffer, event))
+               ring_buffer_unlock_commit(tr->buffer, event);
        ftrace_trace_stack(tr, flags, 6, pc);
        ftrace_trace_userstack(tr, flags, pc);
 }
@@ -1139,6 +1200,7 @@ int trace_graph_entry(struct ftrace_graph_ent *trace)
        struct trace_array_cpu *data;
        unsigned long flags;
        long disabled;
+       int ret;
        int cpu;
        int pc;
 
@@ -1154,15 +1216,18 @@ int trace_graph_entry(struct ftrace_graph_ent *trace)
        disabled = atomic_inc_return(&data->disabled);
        if (likely(disabled == 1)) {
                pc = preempt_count();
-               __trace_graph_entry(tr, trace, flags, pc);
+               ret = __trace_graph_entry(tr, trace, flags, pc);
+       } else {
+               ret = 0;
        }
        /* Only do the atomic if it is not already set */
        if (!test_tsk_trace_graph(current))
                set_tsk_trace_graph(current);
+
        atomic_dec(&data->disabled);
        local_irq_restore(flags);
 
-       return 1;
+       return ret;
 }
 
 void trace_graph_return(struct ftrace_graph_ret *trace)
@@ -1194,17 +1259,19 @@ void trace_graph_return(struct ftrace_graph_ret *trace)
  * trace_vbprintk - write binary msg to tracing buffer
  *
  */
-int trace_vbprintk(unsigned long ip, int depth, const char *fmt, va_list args)
+int trace_vbprintk(unsigned long ip, const char *fmt, va_list args)
 {
        static raw_spinlock_t trace_buf_lock =
                (raw_spinlock_t)__RAW_SPIN_LOCK_UNLOCKED;
        static u32 trace_buf[TRACE_BUF_SIZE];
 
+       struct ftrace_event_call *call = &event_bprint;
        struct ring_buffer_event *event;
        struct trace_array *tr = &global_trace;
        struct trace_array_cpu *data;
        struct bprint_entry *entry;
        unsigned long flags;
+       int disable;
        int resched;
        int cpu, len = 0, size, pc;
 
@@ -1219,7 +1286,8 @@ int trace_vbprintk(unsigned long ip, int depth, const char *fmt, va_list args)
        cpu = raw_smp_processor_id();
        data = tr->data[cpu];
 
-       if (unlikely(atomic_read(&data->disabled)))
+       disable = atomic_inc_return(&data->disabled);
+       if (unlikely(disable != 1))
                goto out;
 
        /* Lockdep uses trace_printk for lock tracing */
@@ -1236,17 +1304,18 @@ int trace_vbprintk(unsigned long ip, int depth, const char *fmt, va_list args)
                goto out_unlock;
        entry = ring_buffer_event_data(event);
        entry->ip                       = ip;
-       entry->depth                    = depth;
        entry->fmt                      = fmt;
 
        memcpy(entry->buf, trace_buf, sizeof(u32) * len);
-       ring_buffer_unlock_commit(tr->buffer, event);
+       if (!filter_check_discard(call, entry, tr->buffer, event))
+               ring_buffer_unlock_commit(tr->buffer, event);
 
 out_unlock:
        __raw_spin_unlock(&trace_buf_lock);
        local_irq_restore(flags);
 
 out:
+       atomic_dec_return(&data->disabled);
        ftrace_preempt_enable(resched);
        unpause_graph_tracing();
 
@@ -1254,17 +1323,19 @@ out:
 }
 EXPORT_SYMBOL_GPL(trace_vbprintk);
 
-int trace_vprintk(unsigned long ip, int depth, const char *fmt, va_list args)
+int trace_vprintk(unsigned long ip, const char *fmt, va_list args)
 {
        static raw_spinlock_t trace_buf_lock = __RAW_SPIN_LOCK_UNLOCKED;
        static char trace_buf[TRACE_BUF_SIZE];
 
+       struct ftrace_event_call *call = &event_print;
        struct ring_buffer_event *event;
        struct trace_array *tr = &global_trace;
        struct trace_array_cpu *data;
        int cpu, len = 0, size, pc;
        struct print_entry *entry;
        unsigned long irq_flags;
+       int disable;
 
        if (tracing_disabled || tracing_selftest_running)
                return 0;
@@ -1274,7 +1345,8 @@ int trace_vprintk(unsigned long ip, int depth, const char *fmt, va_list args)
        cpu = raw_smp_processor_id();
        data = tr->data[cpu];
 
-       if (unlikely(atomic_read(&data->disabled)))
+       disable = atomic_inc_return(&data->disabled);
+       if (unlikely(disable != 1))
                goto out;
 
        pause_graph_tracing();
@@ -1291,17 +1363,18 @@ int trace_vprintk(unsigned long ip, int depth, const char *fmt, va_list args)
                goto out_unlock;
        entry = ring_buffer_event_data(event);
        entry->ip                       = ip;
-       entry->depth                    = depth;
 
        memcpy(&entry->buf, trace_buf, len);
        entry->buf[len] = 0;
-       ring_buffer_unlock_commit(tr->buffer, event);
+       if (!filter_check_discard(call, entry, tr->buffer, event))
+               ring_buffer_unlock_commit(tr->buffer, event);
 
  out_unlock:
        __raw_spin_unlock(&trace_buf_lock);
        raw_local_irq_restore(irq_flags);
        unpause_graph_tracing();
  out:
+       atomic_dec_return(&data->disabled);
        preempt_enable_notrace();
 
        return len;
@@ -1498,12 +1571,14 @@ static void *s_start(struct seq_file *m, loff_t *pos)
                p = s_next(m, p, &l);
        }
 
+       trace_event_read_lock();
        return p;
 }
 
 static void s_stop(struct seq_file *m, void *p)
 {
        atomic_dec(&trace_record_cmdline_disabled);
+       trace_event_read_unlock();
 }
 
 static void print_lat_help_header(struct seq_file *m)
@@ -1604,7 +1679,11 @@ static void test_cpu_buff_start(struct trace_iterator *iter)
                return;
 
        cpumask_set_cpu(iter->cpu, iter->started);
-       trace_seq_printf(s, "##### CPU %u buffer started ####\n", iter->cpu);
+
+       /* Don't print started cpu buffer for the first entry of the trace */
+       if (iter->idx > 1)
+               trace_seq_printf(s, "##### CPU %u buffer started ####\n",
+                               iter->cpu);
 }
 
 static enum print_line_t print_trace_fmt(struct trace_iterator *iter)
@@ -1694,38 +1773,6 @@ static enum print_line_t print_hex_fmt(struct trace_iterator *iter)
        return TRACE_TYPE_HANDLED;
 }
 
-static enum print_line_t print_bprintk_msg_only(struct trace_iterator *iter)
-{
-       struct trace_seq *s = &iter->seq;
-       struct trace_entry *entry = iter->ent;
-       struct bprint_entry *field;
-       int ret;
-
-       trace_assign_type(field, entry);
-
-       ret = trace_seq_bprintf(s, field->fmt, field->buf);
-       if (!ret)
-               return TRACE_TYPE_PARTIAL_LINE;
-
-       return TRACE_TYPE_HANDLED;
-}
-
-static enum print_line_t print_printk_msg_only(struct trace_iterator *iter)
-{
-       struct trace_seq *s = &iter->seq;
-       struct trace_entry *entry = iter->ent;
-       struct print_entry *field;
-       int ret;
-
-       trace_assign_type(field, entry);
-
-       ret = trace_seq_printf(s, "%s", field->buf);
-       if (!ret)
-               return TRACE_TYPE_PARTIAL_LINE;
-
-       return TRACE_TYPE_HANDLED;
-}
-
 static enum print_line_t print_bin_fmt(struct trace_iterator *iter)
 {
        struct trace_seq *s = &iter->seq;
@@ -1774,6 +1821,7 @@ static int trace_empty(struct trace_iterator *iter)
        return 1;
 }
 
+/*  Called with trace_event_read_lock() held. */
 static enum print_line_t print_trace_line(struct trace_iterator *iter)
 {
        enum print_line_t ret;
@@ -1787,12 +1835,12 @@ static enum print_line_t print_trace_line(struct trace_iterator *iter)
        if (iter->ent->type == TRACE_BPRINT &&
                        trace_flags & TRACE_ITER_PRINTK &&
                        trace_flags & TRACE_ITER_PRINTK_MSGONLY)
-               return print_bprintk_msg_only(iter);
+               return trace_print_bprintk_msg_only(iter);
 
        if (iter->ent->type == TRACE_PRINT &&
                        trace_flags & TRACE_ITER_PRINTK &&
                        trace_flags & TRACE_ITER_PRINTK_MSGONLY)
-               return print_printk_msg_only(iter);
+               return trace_print_printk_msg_only(iter);
 
        if (trace_flags & TRACE_ITER_BIN)
                return print_bin_fmt(iter);
@@ -1871,6 +1919,11 @@ __tracing_open(struct inode *inode, struct file *file)
        if (current_trace)
                *iter->trace = *current_trace;
 
+       if (!alloc_cpumask_var(&iter->started, GFP_KERNEL))
+               goto fail;
+
+       cpumask_clear(iter->started);
+
        if (current_trace && current_trace->print_max)
                iter->tr = &max_tr;
        else
@@ -1921,6 +1974,7 @@ __tracing_open(struct inode *inode, struct file *file)
                if (iter->buffer_iter[cpu])
                        ring_buffer_read_finish(iter->buffer_iter[cpu]);
        }
+       free_cpumask_var(iter->started);
  fail:
        mutex_unlock(&trace_types_lock);
        kfree(iter->trace);
@@ -1941,9 +1995,14 @@ int tracing_open_generic(struct inode *inode, struct file *filp)
 static int tracing_release(struct inode *inode, struct file *file)
 {
        struct seq_file *m = (struct seq_file *)file->private_data;
-       struct trace_iterator *iter = m->private;
+       struct trace_iterator *iter;
        int cpu;
 
+       if (!(file->f_mode & FMODE_READ))
+               return 0;
+
+       iter = m->private;
+
        mutex_lock(&trace_types_lock);
        for_each_tracing_cpu(cpu) {
                if (iter->buffer_iter[cpu])
@@ -1959,6 +2018,7 @@ static int tracing_release(struct inode *inode, struct file *file)
 
        seq_release(inode, file);
        mutex_destroy(&iter->mutex);
+       free_cpumask_var(iter->started);
        kfree(iter->trace);
        kfree(iter);
        return 0;
@@ -1969,12 +2029,24 @@ static int tracing_open(struct inode *inode, struct file *file)
        struct trace_iterator *iter;
        int ret = 0;
 
-       iter = __tracing_open(inode, file);
-       if (IS_ERR(iter))
-               ret = PTR_ERR(iter);
-       else if (trace_flags & TRACE_ITER_LATENCY_FMT)
-               iter->iter_flags |= TRACE_FILE_LAT_FMT;
+       /* If this file was open for write, then erase contents */
+       if ((file->f_mode & FMODE_WRITE) &&
+           !(file->f_flags & O_APPEND)) {
+               long cpu = (long) inode->i_private;
+
+               if (cpu == TRACE_PIPE_ALL_CPU)
+                       tracing_reset_online_cpus(&global_trace);
+               else
+                       tracing_reset(&global_trace, cpu);
+       }
 
+       if (file->f_mode & FMODE_READ) {
+               iter = __tracing_open(inode, file);
+               if (IS_ERR(iter))
+                       ret = PTR_ERR(iter);
+               else if (trace_flags & TRACE_ITER_LATENCY_FMT)
+                       iter->iter_flags |= TRACE_FILE_LAT_FMT;
+       }
        return ret;
 }
 
@@ -2049,9 +2121,17 @@ static int show_traces_open(struct inode *inode, struct file *file)
        return ret;
 }
 
+static ssize_t
+tracing_write_stub(struct file *filp, const char __user *ubuf,
+                  size_t count, loff_t *ppos)
+{
+       return count;
+}
+
 static const struct file_operations tracing_fops = {
        .open           = tracing_open,
        .read           = seq_read,
+       .write          = tracing_write_stub,
        .llseek         = seq_lseek,
        .release        = tracing_release,
 };
@@ -2337,9 +2417,9 @@ static const char readme_msg[] =
        "# mkdir /debug\n"
        "# mount -t debugfs nodev /debug\n\n"
        "# cat /debug/tracing/available_tracers\n"
-       "wakeup preemptirqsoff preemptoff irqsoff ftrace sched_switch none\n\n"
+       "wakeup preemptirqsoff preemptoff irqsoff function sched_switch nop\n\n"
        "# cat /debug/tracing/current_tracer\n"
-       "none\n"
+       "nop\n"
        "# echo sched_switch > /debug/tracing/current_tracer\n"
        "# cat /debug/tracing/current_tracer\n"
        "sched_switch\n"
@@ -2365,6 +2445,56 @@ static const struct file_operations tracing_readme_fops = {
 };
 
 static ssize_t
+tracing_saved_cmdlines_read(struct file *file, char __user *ubuf,
+                               size_t cnt, loff_t *ppos)
+{
+       char *buf_comm;
+       char *file_buf;
+       char *buf;
+       int len = 0;
+       int pid;
+       int i;
+
+       file_buf = kmalloc(SAVED_CMDLINES*(16+TASK_COMM_LEN), GFP_KERNEL);
+       if (!file_buf)
+               return -ENOMEM;
+
+       buf_comm = kmalloc(TASK_COMM_LEN, GFP_KERNEL);
+       if (!buf_comm) {
+               kfree(file_buf);
+               return -ENOMEM;
+       }
+
+       buf = file_buf;
+
+       for (i = 0; i < SAVED_CMDLINES; i++) {
+               int r;
+
+               pid = map_cmdline_to_pid[i];
+               if (pid == -1 || pid == NO_CMDLINE_MAP)
+                       continue;
+
+               trace_find_cmdline(pid, buf_comm);
+               r = sprintf(buf, "%d %s\n", pid, buf_comm);
+               buf += r;
+               len += r;
+       }
+
+       len = simple_read_from_buffer(ubuf, cnt, ppos,
+                                     file_buf, len);
+
+       kfree(file_buf);
+       kfree(buf_comm);
+
+       return len;
+}
+
+static const struct file_operations tracing_saved_cmdlines_fops = {
+    .open       = tracing_open_generic,
+    .read       = tracing_saved_cmdlines_read,
+};
+
+static ssize_t
 tracing_ctrl_read(struct file *filp, char __user *ubuf,
                  size_t cnt, loff_t *ppos)
 {
@@ -2531,7 +2661,7 @@ static int tracing_set_tracer(const char *buf)
        if (!ring_buffer_expanded) {
                ret = tracing_resize_ring_buffer(trace_buf_size);
                if (ret < 0)
-                       return ret;
+                       goto out;
                ret = 0;
        }
 
@@ -2883,6 +3013,7 @@ waitagain:
               offsetof(struct trace_iterator, seq));
        iter->pos = -1;
 
+       trace_event_read_lock();
        while (find_next_entry_inc(iter) != NULL) {
                enum print_line_t ret;
                int len = iter->seq.len;
@@ -2899,6 +3030,7 @@ waitagain:
                if (iter->seq.len >= cnt)
                        break;
        }
+       trace_event_read_unlock();
 
        /* Now copy what we have to the user */
        sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
@@ -3021,6 +3153,8 @@ static ssize_t tracing_splice_read_pipe(struct file *filp,
                goto out_err;
        }
 
+       trace_event_read_lock();
+
        /* Fill as many pages as possible. */
        for (i = 0, rem = len; i < PIPE_BUFFERS && rem; i++) {
                pages[i] = alloc_page(GFP_KERNEL);
@@ -3043,6 +3177,7 @@ static ssize_t tracing_splice_read_pipe(struct file *filp,
                trace_seq_init(&iter->seq);
        }
 
+       trace_event_read_unlock();
        mutex_unlock(&iter->mutex);
 
        spd.nr_pages = i;
@@ -3147,7 +3282,7 @@ static int mark_printk(const char *fmt, ...)
        int ret;
        va_list args;
        va_start(args, fmt);
-       ret = trace_vprintk(0, -1, fmt, args);
+       ret = trace_vprintk(0, fmt, args);
        va_end(args);
        return ret;
 }
@@ -3245,19 +3380,13 @@ static int tracing_buffers_open(struct inode *inode, struct file *filp)
 
        info->tr        = &global_trace;
        info->cpu       = cpu;
-       info->spare     = ring_buffer_alloc_read_page(info->tr->buffer);
+       info->spare     = NULL;
        /* Force reading ring buffer for first read */
        info->read      = (unsigned int)-1;
-       if (!info->spare)
-               goto out;
 
        filp->private_data = info;
 
-       return 0;
-
- out:
-       kfree(info);
-       return -ENOMEM;
+       return nonseekable_open(inode, filp);
 }
 
 static ssize_t
@@ -3272,6 +3401,11 @@ tracing_buffers_read(struct file *filp, char __user *ubuf,
        if (!count)
                return 0;
 
+       if (!info->spare)
+               info->spare = ring_buffer_alloc_read_page(info->tr->buffer);
+       if (!info->spare)
+               return -ENOMEM;
+
        /* Do we have previous read data to read? */
        if (info->read < PAGE_SIZE)
                goto read;
@@ -3310,7 +3444,8 @@ static int tracing_buffers_release(struct inode *inode, struct file *file)
 {
        struct ftrace_buffer_info *info = file->private_data;
 
-       ring_buffer_free_read_page(info->tr->buffer, info->spare);
+       if (info->spare)
+               ring_buffer_free_read_page(info->tr->buffer, info->spare);
        kfree(info);
 
        return 0;
@@ -3393,17 +3528,24 @@ tracing_buffers_splice_read(struct file *file, loff_t *ppos,
                .spd_release    = buffer_spd_release,
        };
        struct buffer_ref *ref;
-       int size, i;
+       int entries, size, i;
        size_t ret;
 
-       /*
-        * We can't seek on a buffer input
-        */
-       if (unlikely(*ppos))
-               return -ESPIPE;
+       if (*ppos & (PAGE_SIZE - 1)) {
+               WARN_ONCE(1, "Ftrace: previous read must page-align\n");
+               return -EINVAL;
+       }
 
+       if (len & (PAGE_SIZE - 1)) {
+               WARN_ONCE(1, "Ftrace: splice_read should page-align\n");
+               if (len < PAGE_SIZE)
+                       return -EINVAL;
+               len &= PAGE_MASK;
+       }
+
+       entries = ring_buffer_entries_cpu(info->tr->buffer, info->cpu);
 
-       for (i = 0; i < PIPE_BUFFERS && len; i++, len -= size) {
+       for (i = 0; i < PIPE_BUFFERS && len && entries; i++, len -= PAGE_SIZE) {
                struct page *page;
                int r;
 
@@ -3411,6 +3553,7 @@ tracing_buffers_splice_read(struct file *file, loff_t *ppos,
                if (!ref)
                        break;
 
+               ref->ref = 1;
                ref->buffer = info->tr->buffer;
                ref->page = ring_buffer_alloc_read_page(ref->buffer);
                if (!ref->page) {
@@ -3419,7 +3562,7 @@ tracing_buffers_splice_read(struct file *file, loff_t *ppos,
                }
 
                r = ring_buffer_read_page(ref->buffer, &ref->page,
-                                         len, info->cpu, 0);
+                                         len, info->cpu, 1);
                if (r < 0) {
                        ring_buffer_free_read_page(ref->buffer,
                                                   ref->page);
@@ -3442,6 +3585,9 @@ tracing_buffers_splice_read(struct file *file, loff_t *ppos,
                spd.partial[i].offset = 0;
                spd.partial[i].private = (unsigned long)ref;
                spd.nr_pages++;
+               *ppos += PAGE_SIZE;
+
+               entries = ring_buffer_entries_cpu(info->tr->buffer, info->cpu);
        }
 
        spd.nr_pages = i;
@@ -3469,6 +3615,45 @@ static const struct file_operations tracing_buffers_fops = {
        .llseek         = no_llseek,
 };
 
+static ssize_t
+tracing_stats_read(struct file *filp, char __user *ubuf,
+                  size_t count, loff_t *ppos)
+{
+       unsigned long cpu = (unsigned long)filp->private_data;
+       struct trace_array *tr = &global_trace;
+       struct trace_seq *s;
+       unsigned long cnt;
+
+       s = kmalloc(sizeof(*s), GFP_ATOMIC);
+       if (!s)
+               return ENOMEM;
+
+       trace_seq_init(s);
+
+       cnt = ring_buffer_entries_cpu(tr->buffer, cpu);
+       trace_seq_printf(s, "entries: %ld\n", cnt);
+
+       cnt = ring_buffer_overrun_cpu(tr->buffer, cpu);
+       trace_seq_printf(s, "overrun: %ld\n", cnt);
+
+       cnt = ring_buffer_commit_overrun_cpu(tr->buffer, cpu);
+       trace_seq_printf(s, "commit overrun: %ld\n", cnt);
+
+       cnt = ring_buffer_nmi_dropped_cpu(tr->buffer, cpu);
+       trace_seq_printf(s, "nmi dropped: %ld\n", cnt);
+
+       count = simple_read_from_buffer(ubuf, count, ppos, s->buffer, s->len);
+
+       kfree(s);
+
+       return count;
+}
+
+static const struct file_operations tracing_stats_fops = {
+       .open           = tracing_open_generic,
+       .read           = tracing_stats_read,
+};
+
 #ifdef CONFIG_DYNAMIC_FTRACE
 
 int __weak ftrace_arch_read_dyn_info(char *buf, int size)
@@ -3515,6 +3700,9 @@ struct dentry *tracing_init_dentry(void)
        if (d_tracer)
                return d_tracer;
 
+       if (!debugfs_initialized())
+               return NULL;
+
        d_tracer = debugfs_create_dir("tracing", NULL);
 
        if (!d_tracer && !once) {
@@ -3555,7 +3743,7 @@ struct dentry *tracing_dentry_percpu(void)
 static void tracing_init_debugfs_percpu(long cpu)
 {
        struct dentry *d_percpu = tracing_dentry_percpu();
-       struct dentry *entry, *d_cpu;
+       struct dentry *d_cpu;
        /* strlen(cpu) + MAX(log10(cpu)) + '\0' */
        char cpu_dir[7];
 
@@ -3570,21 +3758,18 @@ static void tracing_init_debugfs_percpu(long cpu)
        }
 
        /* per cpu trace_pipe */
-       entry = debugfs_create_file("trace_pipe", 0444, d_cpu,
-                               (void *) cpu, &tracing_pipe_fops);
-       if (!entry)
-               pr_warning("Could not create debugfs 'trace_pipe' entry\n");
+       trace_create_file("trace_pipe", 0444, d_cpu,
+                       (void *) cpu, &tracing_pipe_fops);
 
        /* per cpu trace */
-       entry = debugfs_create_file("trace", 0444, d_cpu,
-                               (void *) cpu, &tracing_fops);
-       if (!entry)
-               pr_warning("Could not create debugfs 'trace' entry\n");
+       trace_create_file("trace", 0644, d_cpu,
+                       (void *) cpu, &tracing_fops);
+
+       trace_create_file("trace_pipe_raw", 0444, d_cpu,
+                       (void *) cpu, &tracing_buffers_fops);
 
-       entry = debugfs_create_file("trace_pipe_raw", 0444, d_cpu,
-                                   (void *) cpu, &tracing_buffers_fops);
-       if (!entry)
-               pr_warning("Could not create debugfs 'trace_pipe_raw' entry\n");
+       trace_create_file("stats", 0444, d_cpu,
+                       (void *) cpu, &tracing_stats_fops);
 }
 
 #ifdef CONFIG_FTRACE_SELFTEST
@@ -3740,6 +3925,22 @@ static const struct file_operations trace_options_core_fops = {
        .write = trace_options_core_write,
 };
 
+struct dentry *trace_create_file(const char *name,
+                                mode_t mode,
+                                struct dentry *parent,
+                                void *data,
+                                const struct file_operations *fops)
+{
+       struct dentry *ret;
+
+       ret = debugfs_create_file(name, mode, parent, data, fops);
+       if (!ret)
+               pr_warning("Could not create debugfs '%s' entry\n", name);
+
+       return ret;
+}
+
+
 static struct dentry *trace_options_init_dentry(void)
 {
        struct dentry *d_tracer;
@@ -3767,7 +3968,6 @@ create_trace_option_file(struct trace_option_dentry *topt,
                         struct tracer_opt *opt)
 {
        struct dentry *t_options;
-       struct dentry *entry;
 
        t_options = trace_options_init_dentry();
        if (!t_options)
@@ -3776,11 +3976,9 @@ create_trace_option_file(struct trace_option_dentry *topt,
        topt->flags = flags;
        topt->opt = opt;
 
-       entry = debugfs_create_file(opt->name, 0644, t_options, topt,
+       topt->entry = trace_create_file(opt->name, 0644, t_options, topt,
                                    &trace_options_fops);
 
-       topt->entry = entry;
-
 }
 
 static struct trace_option_dentry *
@@ -3835,123 +4033,84 @@ static struct dentry *
 create_trace_option_core_file(const char *option, long index)
 {
        struct dentry *t_options;
-       struct dentry *entry;
 
        t_options = trace_options_init_dentry();
        if (!t_options)
                return NULL;
 
-       entry = debugfs_create_file(option, 0644, t_options, (void *)index,
+       return trace_create_file(option, 0644, t_options, (void *)index,
                                    &trace_options_core_fops);
-
-       return entry;
 }
 
 static __init void create_trace_options_dir(void)
 {
        struct dentry *t_options;
-       struct dentry *entry;
        int i;
 
        t_options = trace_options_init_dentry();
        if (!t_options)
                return;
 
-       for (i = 0; trace_options[i]; i++) {
-               entry = create_trace_option_core_file(trace_options[i], i);
-               if (!entry)
-                       pr_warning("Could not create debugfs %s entry\n",
-                                  trace_options[i]);
-       }
+       for (i = 0; trace_options[i]; i++)
+               create_trace_option_core_file(trace_options[i], i);
 }
 
 static __init int tracer_init_debugfs(void)
 {
        struct dentry *d_tracer;
-       struct dentry *entry;
        int cpu;
 
        d_tracer = tracing_init_dentry();
 
-       entry = debugfs_create_file("tracing_enabled", 0644, d_tracer,
-                                   &global_trace, &tracing_ctrl_fops);
-       if (!entry)
-               pr_warning("Could not create debugfs 'tracing_enabled' entry\n");
+       trace_create_file("tracing_enabled", 0644, d_tracer,
+                       &global_trace, &tracing_ctrl_fops);
 
-       entry = debugfs_create_file("trace_options", 0644, d_tracer,
-                                   NULL, &tracing_iter_fops);
-       if (!entry)
-               pr_warning("Could not create debugfs 'trace_options' entry\n");
+       trace_create_file("trace_options", 0644, d_tracer,
+                       NULL, &tracing_iter_fops);
 
-       create_trace_options_dir();
+       trace_create_file("tracing_cpumask", 0644, d_tracer,
+                       NULL, &tracing_cpumask_fops);
+
+       trace_create_file("trace", 0644, d_tracer,
+                       (void *) TRACE_PIPE_ALL_CPU, &tracing_fops);
+
+       trace_create_file("available_tracers", 0444, d_tracer,
+                       &global_trace, &show_traces_fops);
+
+       trace_create_file("current_tracer", 0644, d_tracer,
+                       &global_trace, &set_tracer_fops);
+
+       trace_create_file("tracing_max_latency", 0644, d_tracer,
+                       &tracing_max_latency, &tracing_max_lat_fops);
 
-       entry = debugfs_create_file("tracing_cpumask", 0644, d_tracer,
-                                   NULL, &tracing_cpumask_fops);
-       if (!entry)
-               pr_warning("Could not create debugfs 'tracing_cpumask' entry\n");
-
-       entry = debugfs_create_file("trace", 0444, d_tracer,
-                                (void *) TRACE_PIPE_ALL_CPU, &tracing_fops);
-       if (!entry)
-               pr_warning("Could not create debugfs 'trace' entry\n");
-
-       entry = debugfs_create_file("available_tracers", 0444, d_tracer,
-                                   &global_trace, &show_traces_fops);
-       if (!entry)
-               pr_warning("Could not create debugfs 'available_tracers' entry\n");
-
-       entry = debugfs_create_file("current_tracer", 0444, d_tracer,
-                                   &global_trace, &set_tracer_fops);
-       if (!entry)
-               pr_warning("Could not create debugfs 'current_tracer' entry\n");
-
-       entry = debugfs_create_file("tracing_max_latency", 0644, d_tracer,
-                                   &tracing_max_latency,
-                                   &tracing_max_lat_fops);
-       if (!entry)
-               pr_warning("Could not create debugfs "
-                          "'tracing_max_latency' entry\n");
-
-       entry = debugfs_create_file("tracing_thresh", 0644, d_tracer,
-                                   &tracing_thresh, &tracing_max_lat_fops);
-       if (!entry)
-               pr_warning("Could not create debugfs "
-                          "'tracing_thresh' entry\n");
-       entry = debugfs_create_file("README", 0644, d_tracer,
-                                   NULL, &tracing_readme_fops);
-       if (!entry)
-               pr_warning("Could not create debugfs 'README' entry\n");
-
-       entry = debugfs_create_file("trace_pipe", 0444, d_tracer,
+       trace_create_file("tracing_thresh", 0644, d_tracer,
+                       &tracing_thresh, &tracing_max_lat_fops);
+
+       trace_create_file("README", 0444, d_tracer,
+                       NULL, &tracing_readme_fops);
+
+       trace_create_file("trace_pipe", 0444, d_tracer,
                        (void *) TRACE_PIPE_ALL_CPU, &tracing_pipe_fops);
-       if (!entry)
-               pr_warning("Could not create debugfs "
-                          "'trace_pipe' entry\n");
-
-       entry = debugfs_create_file("buffer_size_kb", 0644, d_tracer,
-                                   &global_trace, &tracing_entries_fops);
-       if (!entry)
-               pr_warning("Could not create debugfs "
-                          "'buffer_size_kb' entry\n");
-
-       entry = debugfs_create_file("trace_marker", 0220, d_tracer,
-                                   NULL, &tracing_mark_fops);
-       if (!entry)
-               pr_warning("Could not create debugfs "
-                          "'trace_marker' entry\n");
+
+       trace_create_file("buffer_size_kb", 0644, d_tracer,
+                       &global_trace, &tracing_entries_fops);
+
+       trace_create_file("trace_marker", 0220, d_tracer,
+                       NULL, &tracing_mark_fops);
+
+       trace_create_file("saved_cmdlines", 0444, d_tracer,
+                       NULL, &tracing_saved_cmdlines_fops);
 
 #ifdef CONFIG_DYNAMIC_FTRACE
-       entry = debugfs_create_file("dyn_ftrace_total_info", 0444, d_tracer,
-                                   &ftrace_update_tot_cnt,
-                                   &tracing_dyn_info_fops);
-       if (!entry)
-               pr_warning("Could not create debugfs "
-                          "'dyn_ftrace_total_info' entry\n");
+       trace_create_file("dyn_ftrace_total_info", 0444, d_tracer,
+                       &ftrace_update_tot_cnt, &tracing_dyn_info_fops);
 #endif
 #ifdef CONFIG_SYSPROF_TRACER
        init_tracer_sysprof_debugfs(d_tracer);
 #endif
 
+       create_trace_options_dir();
+
        for_each_tracing_cpu(cpu)
                tracing_init_debugfs_percpu(cpu);
 
@@ -4020,30 +4179,36 @@ trace_printk_seq(struct trace_seq *s)
        trace_seq_init(s);
 }
 
-void ftrace_dump(void)
+static void __ftrace_dump(bool disable_tracing)
 {
-       static DEFINE_SPINLOCK(ftrace_dump_lock);
+       static raw_spinlock_t ftrace_dump_lock =
+               (raw_spinlock_t)__RAW_SPIN_LOCK_UNLOCKED;
        /* use static because iter can be a bit big for the stack */
        static struct trace_iterator iter;
+       unsigned int old_userobj;
        static int dump_ran;
        unsigned long flags;
        int cnt = 0, cpu;
 
        /* only one dump */
-       spin_lock_irqsave(&ftrace_dump_lock, flags);
+       local_irq_save(flags);
+       __raw_spin_lock(&ftrace_dump_lock);
        if (dump_ran)
                goto out;
 
        dump_ran = 1;
 
-       /* No turning back! */
        tracing_off();
-       ftrace_kill();
+
+       if (disable_tracing)
+               ftrace_kill();
 
        for_each_tracing_cpu(cpu) {
                atomic_inc(&global_trace.data[cpu]->disabled);
        }
 
+       old_userobj = trace_flags & TRACE_ITER_SYM_USEROBJ;
+
        /* don't look at user memory in panic mode */
        trace_flags &= ~TRACE_ITER_SYM_USEROBJ;
 
@@ -4088,8 +4253,25 @@ void ftrace_dump(void)
        else
                printk(KERN_TRACE "---------------------------------\n");
 
+       /* Re-enable tracing if requested */
+       if (!disable_tracing) {
+               trace_flags |= old_userobj;
+
+               for_each_tracing_cpu(cpu) {
+                       atomic_dec(&global_trace.data[cpu]->disabled);
+               }
+               tracing_on();
+       }
+
  out:
-       spin_unlock_irqrestore(&ftrace_dump_lock, flags);
+       __raw_spin_unlock(&ftrace_dump_lock);
+       local_irq_restore(flags);
+}
+
+/* By default: disable tracing after the dump */
+void ftrace_dump(void)
+{
+       __ftrace_dump(true);
 }
 
 __init static int tracer_alloc_buffers(void)
@@ -4162,7 +4344,8 @@ __init static int tracer_alloc_buffers(void)
                                       &trace_panic_notifier);
 
        register_die_notifier(&trace_die_notifier);
-       ret = 0;
+
+       return 0;
 
 out_free_cpumask:
        free_cpumask_var(tracing_reader_cpumask);