Merge branch 'master' of git://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux-2.6
[linux-2.6.git] / kernel / trace / trace.c
index d2e7547..c8760ec 100644 (file)
@@ -30,6 +30,7 @@
 #include <linux/gfp.h>
 #include <linux/fs.h>
 #include <linux/kprobes.h>
+#include <linux/seq_file.h>
 #include <linux/writeback.h>
 
 #include <linux/stacktrace.h>
 unsigned long __read_mostly    tracing_max_latency = (cycle_t)ULONG_MAX;
 unsigned long __read_mostly    tracing_thresh;
 
+/*
+ * We need to change this state when a selftest is running.
+ * A selftest will lurk into the ring-buffer to count the
+ * entries inserted during the selftest although some concurrent
+ * insertions into the ring-buffer such as ftrace_printk could occurred
+ * at the same time, giving false positive or negative results.
+ */
+static bool __read_mostly tracing_selftest_running;
+
+/* For tracers that don't implement custom flags */
+static struct tracer_opt dummy_tracer_opt[] = {
+       { }
+};
+
+static struct tracer_flags dummy_tracer_flags = {
+       .val = 0,
+       .opts = dummy_tracer_opt
+};
+
+static int dummy_set_flag(u32 old_flags, u32 bit, int set)
+{
+       return 0;
+}
+
+/*
+ * Kill all tracing for good (never come back).
+ * It is initialized to 1 but will turn to zero if the initialization
+ * of the tracer is successful. But that is the only place that sets
+ * this back to zero.
+ */
+int tracing_disabled = 1;
+
 static DEFINE_PER_CPU(local_t, ftrace_cpu_disabled);
 
 static inline void ftrace_disable_cpu(void)
@@ -62,7 +95,36 @@ static cpumask_t __read_mostly               tracing_buffer_mask;
 #define for_each_tracing_cpu(cpu)      \
        for_each_cpu_mask(cpu, tracing_buffer_mask)
 
-static int tracing_disabled = 1;
+/*
+ * ftrace_dump_on_oops - variable to dump ftrace buffer on oops
+ *
+ * If there is an oops (or kernel panic) and the ftrace_dump_on_oops
+ * is set, then ftrace_dump is called. This will output the contents
+ * of the ftrace buffers to the console.  This is very useful for
+ * capturing traces that lead to crashes and outputing it to a
+ * serial console.
+ *
+ * It is default off, but you can enable it with either specifying
+ * "ftrace_dump_on_oops" in the kernel command line, or setting
+ * /proc/sys/kernel/ftrace_dump_on_oops to true.
+ */
+int ftrace_dump_on_oops;
+
+static int tracing_set_tracer(char *buf);
+
+static int __init set_ftrace(char *str)
+{
+       tracing_set_tracer(str);
+       return 1;
+}
+__setup("ftrace", set_ftrace);
+
+static int __init set_ftrace_dump_on_oops(char *str)
+{
+       ftrace_dump_on_oops = 1;
+       return 1;
+}
+__setup("ftrace_dump_on_oops", set_ftrace_dump_on_oops);
 
 long
 ns2usecs(cycle_t nsec)
@@ -112,6 +174,19 @@ static DEFINE_PER_CPU(struct trace_array_cpu, max_data);
 /* tracer_enabled is used to toggle activation of a tracer */
 static int                     tracer_enabled = 1;
 
+/**
+ * tracing_is_enabled - return tracer_enabled status
+ *
+ * This function is used by other tracers to know the status
+ * of the tracer_enabled flag.  Tracers may use this function
+ * to know if it should enable their features when starting
+ * up. See irqsoff tracer for an example (start_irqsoff_tracer).
+ */
+int tracing_is_enabled(void)
+{
+       return tracer_enabled;
+}
+
 /* function tracing enabled */
 int                            ftrace_function_enabled;
 
@@ -153,8 +228,9 @@ static DEFINE_MUTEX(trace_types_lock);
 /* trace_wait is a waitqueue for tasks blocked on trace_poll */
 static DECLARE_WAIT_QUEUE_HEAD(trace_wait);
 
-/* trace_flags holds iter_ctrl options */
-unsigned long trace_flags = TRACE_ITER_PRINT_PARENT;
+/* trace_flags holds trace_options default values */
+unsigned long trace_flags = TRACE_ITER_PRINT_PARENT | TRACE_ITER_PRINTK |
+       TRACE_ITER_ANNOTATE;
 
 /**
  * trace_wake_up - wake up tasks waiting for trace input
@@ -193,13 +269,6 @@ unsigned long nsecs_to_usecs(unsigned long nsecs)
        return nsecs / 1000;
 }
 
-/*
- * TRACE_ITER_SYM_MASK masks the options in trace_flags that
- * control the output of kernel symbols.
- */
-#define TRACE_ITER_SYM_MASK \
-       (TRACE_ITER_PRINT_PARENT|TRACE_ITER_SYM_OFFSET|TRACE_ITER_SYM_ADDR)
-
 /* These must match the bit postions in trace_iterator_flags */
 static const char *trace_options[] = {
        "print-parent",
@@ -213,6 +282,12 @@ static const char *trace_options[] = {
        "stacktrace",
        "sched-tree",
        "ftrace_printk",
+       "ftrace_preempt",
+       "branch",
+       "annotate",
+       "userstacktrace",
+       "sym-userobj",
+       "printk-msg-only",
        NULL
 };
 
@@ -246,7 +321,7 @@ __update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
 
        memcpy(data->comm, tsk->comm, TASK_COMM_LEN);
        data->pid = tsk->pid;
-       data->uid = tsk->uid;
+       data->uid = task_uid(tsk);
        data->nice = tsk->static_prio - 20 - MAX_RT_PRIO;
        data->policy = tsk->policy;
        data->rt_priority = tsk->rt_priority;
@@ -359,6 +434,28 @@ trace_seq_putmem_hex(struct trace_seq *s, void *mem, size_t len)
        return trace_seq_putmem(s, hex, j);
 }
 
+static int
+trace_seq_path(struct trace_seq *s, struct path *path)
+{
+       unsigned char *p;
+
+       if (s->len >= (PAGE_SIZE - 1))
+               return 0;
+       p = d_path(path, s->buffer + s->len, PAGE_SIZE - s->len);
+       if (!IS_ERR(p)) {
+               p = mangle_path(s->buffer + s->len, p, "\n");
+               if (p) {
+                       s->len = p - s->buffer;
+                       return 1;
+               }
+       } else {
+               s->buffer[s->len++] = '?';
+               return 1;
+       }
+
+       return 0;
+}
+
 static void
 trace_seq_reset(struct trace_seq *s)
 {
@@ -470,7 +567,17 @@ int register_tracer(struct tracer *type)
                return -1;
        }
 
+       /*
+        * When this gets called we hold the BKL which means that
+        * preemption is disabled. Various trace selftests however
+        * need to disable and enable preemption for successful tests.
+        * So we drop the BKL here and grab it after the tests again.
+        */
+       unlock_kernel();
        mutex_lock(&trace_types_lock);
+
+       tracing_selftest_running = true;
+
        for (t = trace_types; t; t = t->next) {
                if (strcmp(type->name, t->name) == 0) {
                        /* already found */
@@ -481,12 +588,20 @@ int register_tracer(struct tracer *type)
                }
        }
 
+       if (!type->set_flag)
+               type->set_flag = &dummy_set_flag;
+       if (!type->flags)
+               type->flags = &dummy_tracer_flags;
+       else
+               if (!type->flags->opts)
+                       type->flags->opts = dummy_tracer_opt;
+
 #ifdef CONFIG_FTRACE_STARTUP_TEST
        if (type->selftest) {
                struct tracer *saved_tracer = current_trace;
                struct trace_array *tr = &global_trace;
-               int saved_ctrl = tr->ctrl;
                int i;
+
                /*
                 * Run a selftest on this tracer.
                 * Here we reset the trace buffer, and set the current
@@ -494,25 +609,23 @@ int register_tracer(struct tracer *type)
                 * internal tracing to verify that everything is in order.
                 * If we fail, we do not register this tracer.
                 */
-               for_each_tracing_cpu(i) {
+               for_each_tracing_cpu(i)
                        tracing_reset(tr, i);
-               }
+
                current_trace = type;
-               tr->ctrl = 0;
                /* the test is responsible for initializing and enabling */
                pr_info("Testing tracer %s: ", type->name);
                ret = type->selftest(type, tr);
                /* the test is responsible for resetting too */
                current_trace = saved_tracer;
-               tr->ctrl = saved_ctrl;
                if (ret) {
                        printk(KERN_CONT "FAILED!\n");
                        goto out;
                }
                /* Only reset on passing, to avoid touching corrupted buffers */
-               for_each_tracing_cpu(i) {
+               for_each_tracing_cpu(i)
                        tracing_reset(tr, i);
-               }
+
                printk(KERN_CONT "PASSED\n");
        }
 #endif
@@ -524,7 +637,9 @@ int register_tracer(struct tracer *type)
                max_tracer_type_len = len;
 
  out:
+       tracing_selftest_running = false;
        mutex_unlock(&trace_types_lock);
+       lock_kernel();
 
        return ret;
 }
@@ -564,6 +679,16 @@ void tracing_reset(struct trace_array *tr, int cpu)
        ftrace_enable_cpu();
 }
 
+void tracing_reset_online_cpus(struct trace_array *tr)
+{
+       int cpu;
+
+       tr->time_start = ftrace_now(tr->cpu);
+
+       for_each_online_cpu(cpu)
+               tracing_reset(tr, cpu);
+}
+
 #define SAVED_CMDLINES 128
 static unsigned map_pid_to_cmdline[PID_MAX_DEFAULT+1];
 static unsigned map_cmdline_to_pid[SAVED_CMDLINES];
@@ -581,6 +706,91 @@ static void trace_init_cmdlines(void)
        cmdline_idx = 0;
 }
 
+static int trace_stop_count;
+static DEFINE_SPINLOCK(tracing_start_lock);
+
+/**
+ * ftrace_off_permanent - disable all ftrace code permanently
+ *
+ * This should only be called when a serious anomally has
+ * been detected.  This will turn off the function tracing,
+ * ring buffers, and other tracing utilites. It takes no
+ * locks and can be called from any context.
+ */
+void ftrace_off_permanent(void)
+{
+       tracing_disabled = 1;
+       ftrace_stop();
+       tracing_off_permanent();
+}
+
+/**
+ * tracing_start - quick start of the tracer
+ *
+ * If tracing is enabled but was stopped by tracing_stop,
+ * this will start the tracer back up.
+ */
+void tracing_start(void)
+{
+       struct ring_buffer *buffer;
+       unsigned long flags;
+
+       if (tracing_disabled)
+               return;
+
+       spin_lock_irqsave(&tracing_start_lock, flags);
+       if (--trace_stop_count)
+               goto out;
+
+       if (trace_stop_count < 0) {
+               /* Someone screwed up their debugging */
+               WARN_ON_ONCE(1);
+               trace_stop_count = 0;
+               goto out;
+       }
+
+
+       buffer = global_trace.buffer;
+       if (buffer)
+               ring_buffer_record_enable(buffer);
+
+       buffer = max_tr.buffer;
+       if (buffer)
+               ring_buffer_record_enable(buffer);
+
+       ftrace_start();
+ out:
+       spin_unlock_irqrestore(&tracing_start_lock, flags);
+}
+
+/**
+ * tracing_stop - quick stop of the tracer
+ *
+ * Light weight way to stop tracing. Use in conjunction with
+ * tracing_start.
+ */
+void tracing_stop(void)
+{
+       struct ring_buffer *buffer;
+       unsigned long flags;
+
+       ftrace_stop();
+       spin_lock_irqsave(&tracing_start_lock, flags);
+       if (trace_stop_count++)
+               goto out;
+
+       buffer = global_trace.buffer;
+       if (buffer)
+               ring_buffer_record_disable(buffer);
+
+       buffer = max_tr.buffer;
+       if (buffer)
+               ring_buffer_record_disable(buffer);
+
+ out:
+       spin_unlock_irqrestore(&tracing_start_lock, flags);
+}
+
 void trace_stop_cmdline_recording(void);
 
 static void trace_save_cmdline(struct task_struct *tsk)
@@ -618,7 +828,7 @@ static void trace_save_cmdline(struct task_struct *tsk)
        spin_unlock(&trace_cmdline_lock);
 }
 
-static char *trace_find_cmdline(int pid)
+char *trace_find_cmdline(int pid)
 {
        char *cmdline = "<...>";
        unsigned map;
@@ -655,6 +865,7 @@ tracing_generic_entry_update(struct trace_entry *entry, unsigned long flags,
 
        entry->preempt_count            = pc & 0xff;
        entry->pid                      = (tsk) ? tsk->pid : 0;
+       entry->tgid                     = (tsk) ? tsk->tgid : 0;
        entry->flags =
 #ifdef CONFIG_TRACE_IRQFLAGS_SUPPORT
                (irqs_disabled_flags(flags) ? TRACE_FLAG_IRQS_OFF : 0) |
@@ -691,6 +902,56 @@ trace_function(struct trace_array *tr, struct trace_array_cpu *data,
        ring_buffer_unlock_commit(tr->buffer, event, irq_flags);
 }
 
+#ifdef CONFIG_FUNCTION_GRAPH_TRACER
+static void __trace_graph_entry(struct trace_array *tr,
+                               struct trace_array_cpu *data,
+                               struct ftrace_graph_ent *trace,
+                               unsigned long flags,
+                               int pc)
+{
+       struct ring_buffer_event *event;
+       struct ftrace_graph_ent_entry *entry;
+       unsigned long irq_flags;
+
+       if (unlikely(local_read(&__get_cpu_var(ftrace_cpu_disabled))))
+               return;
+
+       event = ring_buffer_lock_reserve(global_trace.buffer, sizeof(*entry),
+                                        &irq_flags);
+       if (!event)
+               return;
+       entry   = ring_buffer_event_data(event);
+       tracing_generic_entry_update(&entry->ent, flags, pc);
+       entry->ent.type                 = TRACE_GRAPH_ENT;
+       entry->graph_ent                        = *trace;
+       ring_buffer_unlock_commit(global_trace.buffer, event, irq_flags);
+}
+
+static void __trace_graph_return(struct trace_array *tr,
+                               struct trace_array_cpu *data,
+                               struct ftrace_graph_ret *trace,
+                               unsigned long flags,
+                               int pc)
+{
+       struct ring_buffer_event *event;
+       struct ftrace_graph_ret_entry *entry;
+       unsigned long irq_flags;
+
+       if (unlikely(local_read(&__get_cpu_var(ftrace_cpu_disabled))))
+               return;
+
+       event = ring_buffer_lock_reserve(global_trace.buffer, sizeof(*entry),
+                                        &irq_flags);
+       if (!event)
+               return;
+       entry   = ring_buffer_event_data(event);
+       tracing_generic_entry_update(&entry->ent, flags, pc);
+       entry->ent.type                 = TRACE_GRAPH_RET;
+       entry->ret                              = *trace;
+       ring_buffer_unlock_commit(global_trace.buffer, event, irq_flags);
+}
+#endif
+
 void
 ftrace(struct trace_array *tr, struct trace_array_cpu *data,
        unsigned long ip, unsigned long parent_ip, unsigned long flags,
@@ -742,6 +1003,46 @@ void __trace_stack(struct trace_array *tr,
        ftrace_trace_stack(tr, data, flags, skip, preempt_count());
 }
 
+static void ftrace_trace_userstack(struct trace_array *tr,
+                  struct trace_array_cpu *data,
+                  unsigned long flags, int pc)
+{
+#ifdef CONFIG_STACKTRACE
+       struct ring_buffer_event *event;
+       struct userstack_entry *entry;
+       struct stack_trace trace;
+       unsigned long irq_flags;
+
+       if (!(trace_flags & TRACE_ITER_USERSTACKTRACE))
+               return;
+
+       event = ring_buffer_lock_reserve(tr->buffer, sizeof(*entry),
+                                        &irq_flags);
+       if (!event)
+               return;
+       entry   = ring_buffer_event_data(event);
+       tracing_generic_entry_update(&entry->ent, flags, pc);
+       entry->ent.type         = TRACE_USER_STACK;
+
+       memset(&entry->caller, 0, sizeof(entry->caller));
+
+       trace.nr_entries        = 0;
+       trace.max_entries       = FTRACE_STACK_ENTRIES;
+       trace.skip              = 0;
+       trace.entries           = entry->caller;
+
+       save_stack_trace_user(&trace);
+       ring_buffer_unlock_commit(tr->buffer, event, irq_flags);
+#endif
+}
+
+void __trace_userstack(struct trace_array *tr,
+                  struct trace_array_cpu *data,
+                  unsigned long flags)
+{
+       ftrace_trace_userstack(tr, data, flags, preempt_count());
+}
+
 static void
 ftrace_trace_special(void *__tr, void *__data,
                     unsigned long arg1, unsigned long arg2, unsigned long arg3,
@@ -765,6 +1066,7 @@ ftrace_trace_special(void *__tr, void *__data,
        entry->arg3                     = arg3;
        ring_buffer_unlock_commit(tr->buffer, event, irq_flags);
        ftrace_trace_stack(tr, data, irq_flags, 4, pc);
+       ftrace_trace_userstack(tr, data, irq_flags, pc);
 
        trace_wake_up();
 }
@@ -803,6 +1105,7 @@ tracing_sched_switch_trace(struct trace_array *tr,
        entry->next_cpu = task_cpu(next);
        ring_buffer_unlock_commit(tr->buffer, event, irq_flags);
        ftrace_trace_stack(tr, data, flags, 5, pc);
+       ftrace_trace_userstack(tr, data, flags, pc);
 }
 
 void
@@ -832,6 +1135,7 @@ tracing_sched_wakeup_trace(struct trace_array *tr,
        entry->next_cpu                 = task_cpu(wakee);
        ring_buffer_unlock_commit(tr->buffer, event, irq_flags);
        ftrace_trace_stack(tr, data, flags, 6, pc);
+       ftrace_trace_userstack(tr, data, flags, pc);
 
        trace_wake_up();
 }
@@ -841,26 +1145,28 @@ ftrace_special(unsigned long arg1, unsigned long arg2, unsigned long arg3)
 {
        struct trace_array *tr = &global_trace;
        struct trace_array_cpu *data;
+       unsigned long flags;
        int cpu;
        int pc;
 
-       if (tracing_disabled || !tr->ctrl)
+       if (tracing_disabled)
                return;
 
        pc = preempt_count();
-       preempt_disable_notrace();
+       local_irq_save(flags);
        cpu = raw_smp_processor_id();
        data = tr->data[cpu];
 
-       if (likely(!atomic_read(&data->disabled)))
+       if (likely(atomic_inc_return(&data->disabled) == 1))
                ftrace_trace_special(tr, data, arg1, arg2, arg3, pc);
 
-       preempt_enable_notrace();
+       atomic_dec(&data->disabled);
+       local_irq_restore(flags);
 }
 
 #ifdef CONFIG_FUNCTION_TRACER
 static void
-function_trace_call(unsigned long ip, unsigned long parent_ip)
+function_trace_call_preempt_only(unsigned long ip, unsigned long parent_ip)
 {
        struct trace_array *tr = &global_trace;
        struct trace_array_cpu *data;
@@ -873,8 +1179,7 @@ function_trace_call(unsigned long ip, unsigned long parent_ip)
                return;
 
        pc = preempt_count();
-       resched = need_resched();
-       preempt_disable_notrace();
+       resched = ftrace_preempt_disable();
        local_save_flags(flags);
        cpu = raw_smp_processor_id();
        data = tr->data[cpu];
@@ -884,12 +1189,97 @@ function_trace_call(unsigned long ip, unsigned long parent_ip)
                trace_function(tr, data, ip, parent_ip, flags, pc);
 
        atomic_dec(&data->disabled);
-       if (resched)
-               preempt_enable_no_resched_notrace();
-       else
-               preempt_enable_notrace();
+       ftrace_preempt_enable(resched);
 }
 
+static void
+function_trace_call(unsigned long ip, unsigned long parent_ip)
+{
+       struct trace_array *tr = &global_trace;
+       struct trace_array_cpu *data;
+       unsigned long flags;
+       long disabled;
+       int cpu;
+       int pc;
+
+       if (unlikely(!ftrace_function_enabled))
+               return;
+
+       /*
+        * Need to use raw, since this must be called before the
+        * recursive protection is performed.
+        */
+       local_irq_save(flags);
+       cpu = raw_smp_processor_id();
+       data = tr->data[cpu];
+       disabled = atomic_inc_return(&data->disabled);
+
+       if (likely(disabled == 1)) {
+               pc = preempt_count();
+               trace_function(tr, data, ip, parent_ip, flags, pc);
+       }
+
+       atomic_dec(&data->disabled);
+       local_irq_restore(flags);
+}
+
+#ifdef CONFIG_FUNCTION_GRAPH_TRACER
+int trace_graph_entry(struct ftrace_graph_ent *trace)
+{
+       struct trace_array *tr = &global_trace;
+       struct trace_array_cpu *data;
+       unsigned long flags;
+       long disabled;
+       int cpu;
+       int pc;
+
+       if (!ftrace_trace_task(current))
+               return 0;
+
+       if (!ftrace_graph_addr(trace->func))
+               return 0;
+
+       local_irq_save(flags);
+       cpu = raw_smp_processor_id();
+       data = tr->data[cpu];
+       disabled = atomic_inc_return(&data->disabled);
+       if (likely(disabled == 1)) {
+               pc = preempt_count();
+               __trace_graph_entry(tr, data, trace, flags, pc);
+       }
+       /* 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;
+}
+
+void trace_graph_return(struct ftrace_graph_ret *trace)
+{
+       struct trace_array *tr = &global_trace;
+       struct trace_array_cpu *data;
+       unsigned long flags;
+       long disabled;
+       int cpu;
+       int pc;
+
+       local_irq_save(flags);
+       cpu = raw_smp_processor_id();
+       data = tr->data[cpu];
+       disabled = atomic_inc_return(&data->disabled);
+       if (likely(disabled == 1)) {
+               pc = preempt_count();
+               __trace_graph_return(tr, data, trace, flags, pc);
+       }
+       if (!trace->depth)
+               clear_tsk_trace_graph(current);
+       atomic_dec(&data->disabled);
+       local_irq_restore(flags);
+}
+#endif /* CONFIG_FUNCTION_GRAPH_TRACER */
+
 static struct ftrace_ops trace_ops __read_mostly =
 {
        .func = function_trace_call,
@@ -898,9 +1288,14 @@ static struct ftrace_ops trace_ops __read_mostly =
 void tracing_start_function_trace(void)
 {
        ftrace_function_enabled = 0;
+
+       if (trace_flags & TRACE_ITER_PREEMPTONLY)
+               trace_ops.func = function_trace_call_preempt_only;
+       else
+               trace_ops.func = function_trace_call;
+
        register_ftrace_function(&trace_ops);
-       if (tracer_enabled)
-               ftrace_function_enabled = 1;
+       ftrace_function_enabled = 1;
 }
 
 void tracing_stop_function_trace(void)
@@ -912,6 +1307,7 @@ void tracing_stop_function_trace(void)
 
 enum trace_file_type {
        TRACE_FILE_LAT_FMT      = 1,
+       TRACE_FILE_ANNOTATE     = 2,
 };
 
 static void trace_iterator_increment(struct trace_iterator *iter, int cpu)
@@ -1047,10 +1443,6 @@ static void *s_start(struct seq_file *m, loff_t *pos)
 
        atomic_inc(&trace_record_cmdline_disabled);
 
-       /* let the tracer grab locks here if needed */
-       if (current_trace->start)
-               current_trace->start(iter);
-
        if (*pos != iter->pos) {
                iter->ent = NULL;
                iter->cpu = 0;
@@ -1077,14 +1469,7 @@ static void *s_start(struct seq_file *m, loff_t *pos)
 
 static void s_stop(struct seq_file *m, void *p)
 {
-       struct trace_iterator *iter = m->private;
-
        atomic_dec(&trace_record_cmdline_disabled);
-
-       /* let the tracer release locks here if needed */
-       if (current_trace && current_trace == iter->trace && iter->trace->stop)
-               iter->trace->stop(iter);
-
        mutex_unlock(&trace_types_lock);
 }
 
@@ -1143,7 +1528,7 @@ seq_print_sym_offset(struct trace_seq *s, const char *fmt,
 # define IP_FMT "%016lx"
 #endif
 
-static int
+int
 seq_print_ip_sym(struct trace_seq *s, unsigned long ip, unsigned long sym_flags)
 {
        int ret;
@@ -1164,6 +1549,78 @@ seq_print_ip_sym(struct trace_seq *s, unsigned long ip, unsigned long sym_flags)
        return ret;
 }
 
+static inline int seq_print_user_ip(struct trace_seq *s, struct mm_struct *mm,
+                                   unsigned long ip, unsigned long sym_flags)
+{
+       struct file *file = NULL;
+       unsigned long vmstart = 0;
+       int ret = 1;
+
+       if (mm) {
+               const struct vm_area_struct *vma;
+
+               down_read(&mm->mmap_sem);
+               vma = find_vma(mm, ip);
+               if (vma) {
+                       file = vma->vm_file;
+                       vmstart = vma->vm_start;
+               }
+               if (file) {
+                       ret = trace_seq_path(s, &file->f_path);
+                       if (ret)
+                               ret = trace_seq_printf(s, "[+0x%lx]", ip - vmstart);
+               }
+               up_read(&mm->mmap_sem);
+       }
+       if (ret && ((sym_flags & TRACE_ITER_SYM_ADDR) || !file))
+               ret = trace_seq_printf(s, " <" IP_FMT ">", ip);
+       return ret;
+}
+
+static int
+seq_print_userip_objs(const struct userstack_entry *entry, struct trace_seq *s,
+                     unsigned long sym_flags)
+{
+       struct mm_struct *mm = NULL;
+       int ret = 1;
+       unsigned int i;
+
+       if (trace_flags & TRACE_ITER_SYM_USEROBJ) {
+               struct task_struct *task;
+               /*
+                * we do the lookup on the thread group leader,
+                * since individual threads might have already quit!
+                */
+               rcu_read_lock();
+               task = find_task_by_vpid(entry->ent.tgid);
+               if (task)
+                       mm = get_task_mm(task);
+               rcu_read_unlock();
+       }
+
+       for (i = 0; i < FTRACE_STACK_ENTRIES; i++) {
+               unsigned long ip = entry->caller[i];
+
+               if (ip == ULONG_MAX || !ret)
+                       break;
+               if (i && ret)
+                       ret = trace_seq_puts(s, " <- ");
+               if (!ip) {
+                       if (ret)
+                               ret = trace_seq_puts(s, "??");
+                       continue;
+               }
+               if (!ret)
+                       break;
+               if (ret)
+                       ret = seq_print_user_ip(s, mm, ip, sym_flags);
+       }
+
+       if (mm)
+               mmput(mm);
+       return ret;
+}
+
 static void print_lat_help_header(struct seq_file *m)
 {
        seq_puts(m, "#                  _------=> CPU#            \n");
@@ -1301,6 +1758,13 @@ lat_print_timestamp(struct trace_seq *s, u64 abs_usecs,
 
 static const char state_to_char[] = TASK_STATE_TO_CHAR_STR;
 
+static int task_state_char(unsigned long state)
+{
+       int bit = state ? __ffs(state) + 1 : 0;
+
+       return bit < sizeof(state_to_char) - 1 ? state_to_char[bit] : '?';
+}
+
 /*
  * The message is supposed to contain an ending newline.
  * If the printing stops prematurely, try to add a newline of our own.
@@ -1338,6 +1802,23 @@ void trace_seq_print_cont(struct trace_seq *s, struct trace_iterator *iter)
                trace_seq_putc(s, '\n');
 }
 
+static void test_cpu_buff_start(struct trace_iterator *iter)
+{
+       struct trace_seq *s = &iter->seq;
+
+       if (!(trace_flags & TRACE_ITER_ANNOTATE))
+               return;
+
+       if (!(iter->iter_flags & TRACE_FILE_ANNOTATE))
+               return;
+
+       if (cpu_isset(iter->cpu, iter->started))
+               return;
+
+       cpu_set(iter->cpu, iter->started);
+       trace_seq_printf(s, "##### CPU %u buffer started ####\n", iter->cpu);
+}
+
 static enum print_line_t
 print_lat_fmt(struct trace_iterator *iter, unsigned int trace_idx, int cpu)
 {
@@ -1352,11 +1833,12 @@ print_lat_fmt(struct trace_iterator *iter, unsigned int trace_idx, int cpu)
        char *comm;
        int S, T;
        int i;
-       unsigned state;
 
        if (entry->type == TRACE_CONT)
                return TRACE_TYPE_HANDLED;
 
+       test_cpu_buff_start(iter);
+
        next_entry = find_next_entry(iter, NULL, &next_ts);
        if (!next_entry)
                next_ts = iter->ts;
@@ -1396,12 +1878,8 @@ print_lat_fmt(struct trace_iterator *iter, unsigned int trace_idx, int cpu)
 
                trace_assign_type(field, entry);
 
-               T = field->next_state < sizeof(state_to_char) ?
-                       state_to_char[field->next_state] : 'X';
-
-               state = field->prev_state ?
-                       __ffs(field->prev_state) + 1 : 0;
-               S = state < sizeof(state_to_char) - 1 ? state_to_char[state] : 'X';
+               T = task_state_char(field->next_state);
+               S = task_state_char(field->prev_state);
                comm = trace_find_cmdline(field->next_pid);
                trace_seq_printf(s, " %5d:%3d:%c %s [%03d] %5d:%3d:%c %s\n",
                                 field->prev_pid,
@@ -1448,6 +1926,27 @@ print_lat_fmt(struct trace_iterator *iter, unsigned int trace_idx, int cpu)
                        trace_seq_print_cont(s, iter);
                break;
        }
+       case TRACE_BRANCH: {
+               struct trace_branch *field;
+
+               trace_assign_type(field, entry);
+
+               trace_seq_printf(s, "[%s] %s:%s:%d\n",
+                                field->correct ? "  ok  " : " MISS ",
+                                field->func,
+                                field->file,
+                                field->line);
+               break;
+       }
+       case TRACE_USER_STACK: {
+               struct userstack_entry *field;
+
+               trace_assign_type(field, entry);
+
+               seq_print_userip_objs(field, s, sym_flags);
+               trace_seq_putc(s, '\n');
+               break;
+       }
        default:
                trace_seq_printf(s, "Unknown type %d\n", entry->type);
        }
@@ -1472,6 +1971,8 @@ static enum print_line_t print_trace_fmt(struct trace_iterator *iter)
        if (entry->type == TRACE_CONT)
                return TRACE_TYPE_HANDLED;
 
+       test_cpu_buff_start(iter);
+
        comm = trace_find_cmdline(iter->ent->pid);
 
        t = ns2usecs(iter->ts);
@@ -1519,10 +2020,8 @@ static enum print_line_t print_trace_fmt(struct trace_iterator *iter)
 
                trace_assign_type(field, entry);
 
-               S = field->prev_state < sizeof(state_to_char) ?
-                       state_to_char[field->prev_state] : 'X';
-               T = field->next_state < sizeof(state_to_char) ?
-                       state_to_char[field->next_state] : 'X';
+               T = task_state_char(field->next_state);
+               S = task_state_char(field->prev_state);
                ret = trace_seq_printf(s, " %5d:%3d:%c %s [%03d] %5d:%3d:%c\n",
                                       field->prev_pid,
                                       field->prev_prio,
@@ -1581,6 +2080,37 @@ static enum print_line_t print_trace_fmt(struct trace_iterator *iter)
                        trace_seq_print_cont(s, iter);
                break;
        }
+       case TRACE_GRAPH_RET: {
+               return print_graph_function(iter);
+       }
+       case TRACE_GRAPH_ENT: {
+               return print_graph_function(iter);
+       }
+       case TRACE_BRANCH: {
+               struct trace_branch *field;
+
+               trace_assign_type(field, entry);
+
+               trace_seq_printf(s, "[%s] %s:%s:%d\n",
+                                field->correct ? "  ok  " : " MISS ",
+                                field->func,
+                                field->file,
+                                field->line);
+               break;
+       }
+       case TRACE_USER_STACK: {
+               struct userstack_entry *field;
+
+               trace_assign_type(field, entry);
+
+               ret = seq_print_userip_objs(field, s, sym_flags);
+               if (!ret)
+                       return TRACE_TYPE_PARTIAL_LINE;
+               ret = trace_seq_putc(s, '\n');
+               if (!ret)
+                       return TRACE_TYPE_PARTIAL_LINE;
+               break;
+       }
        }
        return TRACE_TYPE_HANDLED;
 }
@@ -1621,12 +2151,9 @@ static enum print_line_t print_raw_fmt(struct trace_iterator *iter)
 
                trace_assign_type(field, entry);
 
-               S = field->prev_state < sizeof(state_to_char) ?
-                       state_to_char[field->prev_state] : 'X';
-               T = field->next_state < sizeof(state_to_char) ?
-                       state_to_char[field->next_state] : 'X';
-               if (entry->type == TRACE_WAKE)
-                       S = '+';
+               T = task_state_char(field->next_state);
+               S = entry->type == TRACE_WAKE ? '+' :
+                       task_state_char(field->prev_state);
                ret = trace_seq_printf(s, "%d %d %c %d %d %d %c\n",
                                       field->prev_pid,
                                       field->prev_prio,
@@ -1640,6 +2167,7 @@ static enum print_line_t print_raw_fmt(struct trace_iterator *iter)
                break;
        }
        case TRACE_SPECIAL:
+       case TRACE_USER_STACK:
        case TRACE_STACK: {
                struct special_entry *field;
 
@@ -1712,12 +2240,9 @@ static enum print_line_t print_hex_fmt(struct trace_iterator *iter)
 
                trace_assign_type(field, entry);
 
-               S = field->prev_state < sizeof(state_to_char) ?
-                       state_to_char[field->prev_state] : 'X';
-               T = field->next_state < sizeof(state_to_char) ?
-                       state_to_char[field->next_state] : 'X';
-               if (entry->type == TRACE_WAKE)
-                       S = '+';
+               T = task_state_char(field->next_state);
+               S = entry->type == TRACE_WAKE ? '+' :
+                       task_state_char(field->prev_state);
                SEQ_PUT_HEX_FIELD_RET(s, field->prev_pid);
                SEQ_PUT_HEX_FIELD_RET(s, field->prev_prio);
                SEQ_PUT_HEX_FIELD_RET(s, S);
@@ -1728,6 +2253,7 @@ static enum print_line_t print_hex_fmt(struct trace_iterator *iter)
                break;
        }
        case TRACE_SPECIAL:
+       case TRACE_USER_STACK:
        case TRACE_STACK: {
                struct special_entry *field;
 
@@ -1744,6 +2270,25 @@ static enum print_line_t print_hex_fmt(struct trace_iterator *iter)
        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, field->buf);
+       if (!ret)
+               return TRACE_TYPE_PARTIAL_LINE;
+
+       if (entry->flags & TRACE_FLAG_CONT)
+               trace_seq_print_cont(s, iter);
+
+       return TRACE_TYPE_HANDLED;
+}
+
 static enum print_line_t print_bin_fmt(struct trace_iterator *iter)
 {
        struct trace_seq *s = &iter->seq;
@@ -1782,6 +2327,7 @@ static enum print_line_t print_bin_fmt(struct trace_iterator *iter)
                break;
        }
        case TRACE_SPECIAL:
+       case TRACE_USER_STACK:
        case TRACE_STACK: {
                struct special_entry *field;
 
@@ -1823,6 +2369,11 @@ static enum print_line_t print_trace_line(struct trace_iterator *iter)
                        return ret;
        }
 
+       if (iter->ent->type == TRACE_PRINT &&
+                       trace_flags & TRACE_ITER_PRINTK &&
+                       trace_flags & TRACE_ITER_PRINTK_MSGONLY)
+               return print_printk_msg_only(iter);
+
        if (trace_flags & TRACE_ITER_BIN)
                return print_bin_fmt(iter);
 
@@ -1847,7 +2398,9 @@ static int s_show(struct seq_file *m, void *v)
                        seq_printf(m, "# tracer: %s\n", iter->trace->name);
                        seq_puts(m, "#\n");
                }
-               if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
+               if (iter->trace && iter->trace->print_header)
+                       iter->trace->print_header(m);
+               else if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
                        /* print nothing if the buffers are empty */
                        if (trace_empty(iter))
                                return 0;
@@ -1899,6 +2452,15 @@ __tracing_open(struct inode *inode, struct file *file, int *ret)
        iter->trace = current_trace;
        iter->pos = -1;
 
+       /* Notify the tracer early; before we stop tracing. */
+       if (iter->trace && iter->trace->open)
+               iter->trace->open(iter);
+
+       /* Annotate start of buffers if we had overruns */
+       if (ring_buffer_overruns(iter->tr->buffer))
+               iter->iter_flags |= TRACE_FILE_ANNOTATE;
+
+
        for_each_tracing_cpu(cpu) {
 
                iter->buffer_iter[cpu] =
@@ -1917,13 +2479,7 @@ __tracing_open(struct inode *inode, struct file *file, int *ret)
        m->private = iter;
 
        /* stop the trace while dumping */
-       if (iter->tr->ctrl) {
-               tracer_enabled = 0;
-               ftrace_function_enabled = 0;
-       }
-
-       if (iter->trace && iter->trace->open)
-                       iter->trace->open(iter);
+       tracing_stop();
 
        mutex_unlock(&trace_types_lock);
 
@@ -1966,14 +2522,7 @@ int tracing_release(struct inode *inode, struct file *file)
                iter->trace->close(iter);
 
        /* reenable tracing if it was previously enabled */
-       if (iter->tr->ctrl) {
-               tracer_enabled = 1;
-               /*
-                * It is safe to enable function tracing even if it
-                * isn't used
-                */
-               ftrace_function_enabled = 1;
-       }
+       tracing_start();
        mutex_unlock(&trace_types_lock);
 
        seq_release(inode, file);
@@ -2151,7 +2700,7 @@ tracing_cpumask_write(struct file *filp, const char __user *ubuf,
        if (err)
                goto err_unlock;
 
-       raw_local_irq_disable();
+       local_irq_disable();
        __raw_spin_lock(&ftrace_max_lock);
        for_each_tracing_cpu(cpu) {
                /*
@@ -2168,7 +2717,7 @@ tracing_cpumask_write(struct file *filp, const char __user *ubuf,
                }
        }
        __raw_spin_unlock(&ftrace_max_lock);
-       raw_local_irq_enable();
+       local_irq_enable();
 
        tracing_cpumask = tracing_cpumask_new;
 
@@ -2189,13 +2738,16 @@ static struct file_operations tracing_cpumask_fops = {
 };
 
 static ssize_t
-tracing_iter_ctrl_read(struct file *filp, char __user *ubuf,
+tracing_trace_options_read(struct file *filp, char __user *ubuf,
                       size_t cnt, loff_t *ppos)
 {
+       int i;
        char *buf;
        int r = 0;
        int len = 0;
-       int i;
+       u32 tracer_flags = current_trace->flags->val;
+       struct tracer_opt *trace_opts = current_trace->flags->opts;
+
 
        /* calulate max size */
        for (i = 0; trace_options[i]; i++) {
@@ -2203,6 +2755,15 @@ tracing_iter_ctrl_read(struct file *filp, char __user *ubuf,
                len += 3; /* "no" and space */
        }
 
+       /*
+        * Increase the size with names of options specific
+        * of the current tracer.
+        */
+       for (i = 0; trace_opts[i].name; i++) {
+               len += strlen(trace_opts[i].name);
+               len += 3; /* "no" and space */
+       }
+
        /* +2 for \n and \0 */
        buf = kmalloc(len + 2, GFP_KERNEL);
        if (!buf)
@@ -2215,6 +2776,15 @@ tracing_iter_ctrl_read(struct file *filp, char __user *ubuf,
                        r += sprintf(buf + r, "no%s ", trace_options[i]);
        }
 
+       for (i = 0; trace_opts[i].name; i++) {
+               if (tracer_flags & trace_opts[i].bit)
+                       r += sprintf(buf + r, "%s ",
+                               trace_opts[i].name);
+               else
+                       r += sprintf(buf + r, "no%s ",
+                               trace_opts[i].name);
+       }
+
        r += sprintf(buf + r, "\n");
        WARN_ON(r >= len + 2);
 
@@ -2225,13 +2795,48 @@ tracing_iter_ctrl_read(struct file *filp, char __user *ubuf,
        return r;
 }
 
+/* Try to assign a tracer specific option */
+static int set_tracer_option(struct tracer *trace, char *cmp, int neg)
+{
+       struct tracer_flags *trace_flags = trace->flags;
+       struct tracer_opt *opts = NULL;
+       int ret = 0, i = 0;
+       int len;
+
+       for (i = 0; trace_flags->opts[i].name; i++) {
+               opts = &trace_flags->opts[i];
+               len = strlen(opts->name);
+
+               if (strncmp(cmp, opts->name, len) == 0) {
+                       ret = trace->set_flag(trace_flags->val,
+                               opts->bit, !neg);
+                       break;
+               }
+       }
+       /* Not found */
+       if (!trace_flags->opts[i].name)
+               return -EINVAL;
+
+       /* Refused to handle */
+       if (ret)
+               return ret;
+
+       if (neg)
+               trace_flags->val &= ~opts->bit;
+       else
+               trace_flags->val |= opts->bit;
+
+       return 0;
+}
+
 static ssize_t
-tracing_iter_ctrl_write(struct file *filp, const char __user *ubuf,
+tracing_trace_options_write(struct file *filp, const char __user *ubuf,
                        size_t cnt, loff_t *ppos)
 {
        char buf[64];
        char *cmp = buf;
        int neg = 0;
+       int ret;
        int i;
 
        if (cnt >= sizeof(buf))
@@ -2258,11 +2863,13 @@ tracing_iter_ctrl_write(struct file *filp, const char __user *ubuf,
                        break;
                }
        }
-       /*
-        * If no option could be set, return an error:
-        */
-       if (!trace_options[i])
-               return -EINVAL;
+
+       /* If no option could be set, test the specific tracer options */
+       if (!trace_options[i]) {
+               ret = set_tracer_option(current_trace, cmp, neg);
+               if (ret)
+                       return ret;
+       }
 
        filp->f_pos += cnt;
 
@@ -2271,8 +2878,8 @@ tracing_iter_ctrl_write(struct file *filp, const char __user *ubuf,
 
 static struct file_operations tracing_iter_fops = {
        .open           = tracing_open_generic,
-       .read           = tracing_iter_ctrl_read,
-       .write          = tracing_iter_ctrl_write,
+       .read           = tracing_trace_options_read,
+       .write          = tracing_trace_options_write,
 };
 
 static const char readme_msg[] =
@@ -2286,9 +2893,9 @@ static const char readme_msg[] =
        "# echo sched_switch > /debug/tracing/current_tracer\n"
        "# cat /debug/tracing/current_tracer\n"
        "sched_switch\n"
-       "# cat /debug/tracing/iter_ctrl\n"
+       "# cat /debug/tracing/trace_options\n"
        "noprint-parent nosym-offset nosym-addr noverbose\n"
-       "# echo print-parent > /debug/tracing/iter_ctrl\n"
+       "# echo print-parent > /debug/tracing/trace_options\n"
        "# echo 1 > /debug/tracing/tracing_enabled\n"
        "# cat /debug/tracing/trace > /tmp/trace.txt\n"
        "echo 0 > /debug/tracing/tracing_enabled\n"
@@ -2311,11 +2918,10 @@ static ssize_t
 tracing_ctrl_read(struct file *filp, char __user *ubuf,
                  size_t cnt, loff_t *ppos)
 {
-       struct trace_array *tr = filp->private_data;
        char buf[64];
        int r;
 
-       r = sprintf(buf, "%ld\n", tr->ctrl);
+       r = sprintf(buf, "%u\n", tracer_enabled);
        return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
 }
 
@@ -2343,16 +2949,18 @@ tracing_ctrl_write(struct file *filp, const char __user *ubuf,
        val = !!val;
 
        mutex_lock(&trace_types_lock);
-       if (tr->ctrl ^ val) {
-               if (val)
+       if (tracer_enabled ^ val) {
+               if (val) {
                        tracer_enabled = 1;
-               else
+                       if (current_trace->start)
+                               current_trace->start(tr);
+                       tracing_start();
+               } else {
                        tracer_enabled = 0;
-
-               tr->ctrl = val;
-
-               if (current_trace && current_trace->ctrl_update)
-                       current_trace->ctrl_update(tr);
+                       tracing_stop();
+                       if (current_trace->stop)
+                               current_trace->stop(tr);
+               }
        }
        mutex_unlock(&trace_types_lock);
 
@@ -2378,29 +2986,11 @@ tracing_set_trace_read(struct file *filp, char __user *ubuf,
        return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
 }
 
-static ssize_t
-tracing_set_trace_write(struct file *filp, const char __user *ubuf,
-                       size_t cnt, loff_t *ppos)
+static int tracing_set_tracer(char *buf)
 {
        struct trace_array *tr = &global_trace;
        struct tracer *t;
-       char buf[max_tracer_type_len+1];
-       int i;
-       size_t ret;
-
-       ret = cnt;
-
-       if (cnt > max_tracer_type_len)
-               cnt = max_tracer_type_len;
-
-       if (copy_from_user(&buf, ubuf, cnt))
-               return -EFAULT;
-
-       buf[cnt] = 0;
-
-       /* strip ending whitespace. */
-       for (i = cnt - 1; i > 0 && isspace(buf[i]); i--)
-               buf[i] = 0;
+       int ret = 0;
 
        mutex_lock(&trace_types_lock);
        for (t = trace_types; t; t = t->next) {
@@ -2414,18 +3004,52 @@ tracing_set_trace_write(struct file *filp, const char __user *ubuf,
        if (t == current_trace)
                goto out;
 
+       trace_branch_disable();
        if (current_trace && current_trace->reset)
                current_trace->reset(tr);
 
        current_trace = t;
-       if (t->init)
-               t->init(tr);
+       if (t->init) {
+               ret = t->init(tr);
+               if (ret)
+                       goto out;
+       }
 
+       trace_branch_enable(tr);
  out:
        mutex_unlock(&trace_types_lock);
 
-       if (ret > 0)
-               filp->f_pos += ret;
+       return ret;
+}
+
+static ssize_t
+tracing_set_trace_write(struct file *filp, const char __user *ubuf,
+                       size_t cnt, loff_t *ppos)
+{
+       char buf[max_tracer_type_len+1];
+       int i;
+       size_t ret;
+       int err;
+
+       ret = cnt;
+
+       if (cnt > max_tracer_type_len)
+               cnt = max_tracer_type_len;
+
+       if (copy_from_user(&buf, ubuf, cnt))
+               return -EFAULT;
+
+       buf[cnt] = 0;
+
+       /* strip ending whitespace. */
+       for (i = cnt - 1; i > 0 && isspace(buf[i]); i--)
+               buf[i] = 0;
+
+       err = tracing_set_tracer(buf);
+       if (err)
+               return err;
+
+       filp->f_pos += ret;
 
        return ret;
 }
@@ -2492,6 +3116,10 @@ static int tracing_open_pipe(struct inode *inode, struct file *filp)
                return -ENOMEM;
 
        mutex_lock(&trace_types_lock);
+
+       /* trace pipe does not show start of buffer */
+       cpus_setall(iter->started);
+
        iter->tr = &global_trace;
        iter->trace = current_trace;
        filp->private_data = iter;
@@ -2667,7 +3295,7 @@ tracing_entries_read(struct file *filp, char __user *ubuf,
        char buf[64];
        int r;
 
-       r = sprintf(buf, "%lu\n", tr->entries);
+       r = sprintf(buf, "%lu\n", tr->entries >> 10);
        return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
 }
 
@@ -2678,7 +3306,6 @@ tracing_entries_write(struct file *filp, const char __user *ubuf,
        unsigned long val;
        char buf[64];
        int ret, cpu;
-       struct trace_array *tr = filp->private_data;
 
        if (cnt >= sizeof(buf))
                return -EINVAL;
@@ -2698,12 +3325,7 @@ tracing_entries_write(struct file *filp, const char __user *ubuf,
 
        mutex_lock(&trace_types_lock);
 
-       if (tr->ctrl) {
-               cnt = -EBUSY;
-               pr_info("ftrace: please disable tracing"
-                       " before modifying buffer size\n");
-               goto out;
-       }
+       tracing_stop();
 
        /* disable all cpu buffers */
        for_each_tracing_cpu(cpu) {
@@ -2713,6 +3335,9 @@ tracing_entries_write(struct file *filp, const char __user *ubuf,
                        atomic_inc(&max_tr.data[cpu]->disabled);
        }
 
+       /* value is in KB */
+       val <<= 10;
+
        if (val != global_trace.entries) {
                ret = ring_buffer_resize(global_trace.buffer, val);
                if (ret < 0) {
@@ -2751,6 +3376,7 @@ tracing_entries_write(struct file *filp, const char __user *ubuf,
                        atomic_dec(&max_tr.data[cpu]->disabled);
        }
 
+       tracing_start();
        max_tr.entries = global_trace.entries;
        mutex_unlock(&trace_types_lock);
 
@@ -2762,7 +3388,7 @@ static int mark_printk(const char *fmt, ...)
        int ret;
        va_list args;
        va_start(args, fmt);
-       ret = trace_vprintk(0, fmt, args);
+       ret = trace_vprintk(0, -1, fmt, args);
        va_end(args);
        return ret;
 }
@@ -2773,9 +3399,8 @@ tracing_mark_write(struct file *filp, const char __user *ubuf,
 {
        char *buf;
        char *end;
-       struct trace_array *tr = &global_trace;
 
-       if (!tr->ctrl || tracing_disabled)
+       if (tracing_disabled)
                return -EINVAL;
 
        if (cnt > TRACE_BUF_SIZE)
@@ -2841,22 +3466,38 @@ static struct file_operations tracing_mark_fops = {
 
 #ifdef CONFIG_DYNAMIC_FTRACE
 
+int __weak ftrace_arch_read_dyn_info(char *buf, int size)
+{
+       return 0;
+}
+
 static ssize_t
-tracing_read_long(struct file *filp, char __user *ubuf,
+tracing_read_dyn_info(struct file *filp, char __user *ubuf,
                  size_t cnt, loff_t *ppos)
 {
+       static char ftrace_dyn_info_buffer[1024];
+       static DEFINE_MUTEX(dyn_info_mutex);
        unsigned long *p = filp->private_data;
-       char buf[64];
+       char *buf = ftrace_dyn_info_buffer;
+       int size = ARRAY_SIZE(ftrace_dyn_info_buffer);
        int r;
 
-       r = sprintf(buf, "%ld\n", *p);
+       mutex_lock(&dyn_info_mutex);
+       r = sprintf(buf, "%ld ", *p);
 
-       return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
+       r += ftrace_arch_read_dyn_info(buf+r, (size-1)-r);
+       buf[r++] = '\n';
+
+       r = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
+
+       mutex_unlock(&dyn_info_mutex);
+
+       return r;
 }
 
-static struct file_operations tracing_read_long_fops = {
+static struct file_operations tracing_dyn_info_fops = {
        .open           = tracing_open_generic,
-       .read           = tracing_read_long,
+       .read           = tracing_read_dyn_info,
 };
 #endif
 
@@ -2897,10 +3538,10 @@ static __init int tracer_init_debugfs(void)
        if (!entry)
                pr_warning("Could not create debugfs 'tracing_enabled' entry\n");
 
-       entry = debugfs_create_file("iter_ctrl", 0644, d_tracer,
+       entry = debugfs_create_file("trace_options", 0644, d_tracer,
                                    NULL, &tracing_iter_fops);
        if (!entry)
-               pr_warning("Could not create debugfs 'iter_ctrl' entry\n");
+               pr_warning("Could not create debugfs 'trace_options' entry\n");
 
        entry = debugfs_create_file("tracing_cpumask", 0644, d_tracer,
                                    NULL, &tracing_cpumask_fops);
@@ -2950,11 +3591,11 @@ static __init int tracer_init_debugfs(void)
                pr_warning("Could not create debugfs "
                           "'trace_pipe' entry\n");
 
-       entry = debugfs_create_file("trace_entries", 0644, d_tracer,
+       entry = debugfs_create_file("buffer_size_kb", 0644, d_tracer,
                                    &global_trace, &tracing_entries_fops);
        if (!entry)
                pr_warning("Could not create debugfs "
-                          "'trace_entries' entry\n");
+                          "'buffer_size_kb' entry\n");
 
        entry = debugfs_create_file("trace_marker", 0220, d_tracer,
                                    NULL, &tracing_mark_fops);
@@ -2965,7 +3606,7 @@ static __init int tracer_init_debugfs(void)
 #ifdef CONFIG_DYNAMIC_FTRACE
        entry = debugfs_create_file("dyn_ftrace_total_info", 0444, d_tracer,
                                    &ftrace_update_tot_cnt,
-                                   &tracing_read_long_fops);
+                                   &tracing_dyn_info_fops);
        if (!entry)
                pr_warning("Could not create debugfs "
                           "'dyn_ftrace_total_info' entry\n");
@@ -2976,7 +3617,7 @@ static __init int tracer_init_debugfs(void)
        return 0;
 }
 
-int trace_vprintk(unsigned long ip, const char *fmt, va_list args)
+int trace_vprintk(unsigned long ip, int depth, const char *fmt, va_list args)
 {
        static DEFINE_SPINLOCK(trace_buf_lock);
        static char trace_buf[TRACE_BUF_SIZE];
@@ -2984,11 +3625,11 @@ int trace_vprintk(unsigned long ip, const char *fmt, va_list args)
        struct ring_buffer_event *event;
        struct trace_array *tr = &global_trace;
        struct trace_array_cpu *data;
-       struct print_entry *entry;
-       unsigned long flags, irq_flags;
        int cpu, len = 0, size, pc;
+       struct print_entry *entry;
+       unsigned long irq_flags;
 
-       if (!tr->ctrl || tracing_disabled)
+       if (tracing_disabled || tracing_selftest_running)
                return 0;
 
        pc = preempt_count();
@@ -2999,7 +3640,8 @@ int trace_vprintk(unsigned long ip, const char *fmt, va_list args)
        if (unlikely(atomic_read(&data->disabled)))
                goto out;
 
-       spin_lock_irqsave(&trace_buf_lock, flags);
+       pause_graph_tracing();
+       spin_lock_irqsave(&trace_buf_lock, irq_flags);
        len = vsnprintf(trace_buf, TRACE_BUF_SIZE, fmt, args);
 
        len = min(len, TRACE_BUF_SIZE-1);
@@ -3010,17 +3652,18 @@ int trace_vprintk(unsigned long ip, const char *fmt, va_list args)
        if (!event)
                goto out_unlock;
        entry = ring_buffer_event_data(event);
-       tracing_generic_entry_update(&entry->ent, flags, pc);
+       tracing_generic_entry_update(&entry->ent, irq_flags, pc);
        entry->ent.type                 = TRACE_PRINT;
        entry->ip                       = ip;
+       entry->depth                    = depth;
 
        memcpy(&entry->buf, trace_buf, len);
        entry->buf[len] = 0;
        ring_buffer_unlock_commit(tr->buffer, event, irq_flags);
 
  out_unlock:
-       spin_unlock_irqrestore(&trace_buf_lock, flags);
-
+       spin_unlock_irqrestore(&trace_buf_lock, irq_flags);
+       unpause_graph_tracing();
  out:
        preempt_enable_notrace();
 
@@ -3037,7 +3680,7 @@ int __ftrace_printk(unsigned long ip, const char *fmt, ...)
                return 0;
 
        va_start(ap, fmt);
-       ret = trace_vprintk(ip, fmt, ap);
+       ret = trace_vprintk(ip, task_curr_ret_stack(current), fmt, ap);
        va_end(ap);
        return ret;
 }
@@ -3046,7 +3689,8 @@ EXPORT_SYMBOL_GPL(__ftrace_printk);
 static int trace_panic_handler(struct notifier_block *this,
                               unsigned long event, void *unused)
 {
-       ftrace_dump();
+       if (ftrace_dump_on_oops)
+               ftrace_dump();
        return NOTIFY_OK;
 }
 
@@ -3062,7 +3706,8 @@ static int trace_die_handler(struct notifier_block *self,
 {
        switch (val) {
        case DIE_OOPS:
-               ftrace_dump();
+               if (ftrace_dump_on_oops)
+                       ftrace_dump();
                break;
        default:
                break;
@@ -3103,7 +3748,6 @@ trace_printk_seq(struct trace_seq *s)
        trace_seq_reset(s);
 }
 
-
 void ftrace_dump(void)
 {
        static DEFINE_SPINLOCK(ftrace_dump_lock);
@@ -3128,6 +3772,9 @@ void ftrace_dump(void)
                atomic_inc(&global_trace.data[cpu]->disabled);
        }
 
+       /* don't look at user memory in panic mode */
+       trace_flags &= ~TRACE_ITER_SYM_USEROBJ;
+
        printk(KERN_TRACE "Dumping ftrace buffer:\n");
 
        iter.tr = &global_trace;
@@ -3221,7 +3868,6 @@ __init static int tracer_alloc_buffers(void)
 #endif
 
        /* All seems OK, enable tracing */
-       global_trace.ctrl = tracer_enabled;
        tracing_disabled = 0;
 
        atomic_notifier_chain_register(&panic_notifier_list,