* structure. (this way the maximum trace is permanently saved,
  * for later retrieval via /debugfs/tracing/latency_trace)
  */
-static void notrace
+static notrace void
 __update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
 {
        struct trace_array_cpu *data = tr->data[cpu];
        mutex_unlock(&trace_types_lock);
 }
 
-void notrace tracing_reset(struct trace_array_cpu *data)
+notrace void tracing_reset(struct trace_array_cpu *data)
 {
        data->trace_idx = 0;
        data->trace_current = head_page(data);
        data->trace_current_idx = 0;
+       data->time_offset = 0;
 }
 
 #ifdef CONFIG_FTRACE
-static void notrace
+static notrace void
 function_trace_call(unsigned long ip, unsigned long parent_ip)
 {
        struct trace_array *tr = &global_trace;
 
 notrace void trace_stop_cmdline_recording(void);
 
-static void notrace trace_save_cmdline(struct task_struct *tsk)
+static notrace void trace_save_cmdline(struct task_struct *tsk)
 {
        unsigned map;
        unsigned idx;
                iter->idx++;
                iter->next_idx[next_cpu]++;
                iter->next_page_idx[next_cpu]++;
+
                if (iter->next_page_idx[next_cpu] >= ENTRIES_PER_PAGE) {
                        struct trace_array_cpu *data = iter->tr->data[next_cpu];
 
                                        data->trace_pages.next;
                }
        }
+       iter->prev_ent = iter->ent;
+       iter->prev_cpu = iter->cpu;
+
        iter->ent = next;
        iter->cpu = next_cpu;
 
        return next ? iter : NULL;
 }
 
-static void notrace *
-s_next(struct seq_file *m, void *v, loff_t *pos)
+static notrace void *s_next(struct seq_file *m, void *v, loff_t *pos)
 {
        struct trace_iterator *iter = m->private;
-       void *ent;
        void *last_ent = iter->ent;
        int i = (int)*pos;
+       void *ent;
 
        (*pos)++;
 
                iter->ent = NULL;
                iter->cpu = 0;
                iter->idx = -1;
+               iter->prev_ent = NULL;
+               iter->prev_cpu = -1;
 
                for_each_possible_cpu(i) {
                        iter->next_idx[i] = 0;
 # define IP_FMT "%016lx"
 #endif
 
-static void notrace
+static notrace void
 seq_print_ip_sym(struct seq_file *m, unsigned long ip, unsigned long sym_flags)
 {
        if (!ip) {
                seq_printf(m, " <" IP_FMT ">", ip);
 }
 
-static void notrace print_lat_help_header(struct seq_file *m)
+static notrace void print_lat_help_header(struct seq_file *m)
 {
        seq_puts(m, "#                _------=> CPU#            \n");
        seq_puts(m, "#               / _-----=> irqs-off        \n");
        seq_puts(m, "#     \\   /    |||||   \\   |   /           \n");
 }
 
-static void notrace print_func_help_header(struct seq_file *m)
+static notrace void print_func_help_header(struct seq_file *m)
 {
        seq_puts(m, "#           TASK-PID   CPU#    TIMESTAMP  FUNCTION\n");
        seq_puts(m, "#              | |      |          |         |\n");
 }
 
 
-static void notrace
+static notrace void
 print_trace_header(struct seq_file *m, struct trace_iterator *iter)
 {
        unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
        seq_puts(m, "\n");
 }
 
-static void notrace
+static notrace void
 lat_print_generic(struct seq_file *m, struct trace_entry *entry, int cpu)
 {
        int hardirq, softirq;
 
 unsigned long preempt_mark_thresh = 100;
 
-static void notrace
+static notrace void
 lat_print_timestamp(struct seq_file *m, unsigned long long abs_usecs,
                    unsigned long rel_usecs)
 {
 
 static const char state_to_char[] = TASK_STATE_TO_CHAR_STR;
 
-static void notrace
+static notrace void
 print_lat_fmt(struct seq_file *m, struct trace_iterator *iter,
              unsigned int trace_idx, int cpu)
 {
        }
 }
 
-static void notrace
+static notrace void sync_time_offset(struct trace_iterator *iter)
+{
+       struct trace_array_cpu *prev_array, *array;
+       struct trace_entry *prev_entry, *entry;
+       cycle_t prev_t, t;
+
+       entry = iter->ent;
+       prev_entry = iter->prev_ent;
+       if (!prev_entry)
+               return;
+
+       prev_array = iter->tr->data[iter->prev_cpu];
+       array = iter->tr->data[iter->cpu];
+
+       prev_t = prev_entry->t + prev_array->time_offset;
+       t = entry->t + array->time_offset;
+
+       /*
+        * If time goes backwards we increase the offset of
+        * the current array, to not have observable time warps.
+        * This will quickly synchronize the time offsets of
+        * multiple CPUs:
+        */
+       if (t < prev_t)
+               array->time_offset += prev_t - t;
+}
+
+static notrace void
 print_trace_fmt(struct seq_file *m, struct trace_iterator *iter)
 {
        unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
-       struct trace_entry *entry = iter->ent;
+       struct trace_entry *entry;
        unsigned long usec_rem;
        unsigned long long t;
        unsigned long secs;
        char *comm;
        int S;
 
+       sync_time_offset(iter);
+       entry = iter->ent;
+
        comm = trace_find_cmdline(iter->ent->pid);
 
-       t = ns2usecs(entry->t);
+       t = ns2usecs(entry->t + iter->tr->data[iter->cpu]->time_offset);
        usec_rem = do_div(t, 1000000ULL);
        secs = (unsigned long)t;
 
 }
 
 
-static void notrace *
+static notrace void *
 t_next(struct seq_file *m, void *v, loff_t *pos)
 {
        struct tracer *t = m->private;
        int r;
 
        r = sprintf(buf, "%ld\n", tr->ctrl);
-       return simple_read_from_buffer(ubuf, cnt, ppos,
-                                      buf, r);
+       return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
 }
 
 static ssize_t