ftrace: build fix
[deliverable/linux.git] / kernel / trace / trace.c
index 9175ce91b8f6923bbfaf7e883bcfc0fc27b81fd6..71b25b79b3decce0ad1ccfe171a93b5ce2e07c51 100644 (file)
@@ -32,6 +32,8 @@
 unsigned long __read_mostly    tracing_max_latency = (cycle_t)ULONG_MAX;
 unsigned long __read_mostly    tracing_thresh;
 
+static int tracing_disabled = 1;
+
 static long notrace
 ns2usecs(cycle_t nsec)
 {
@@ -40,7 +42,11 @@ ns2usecs(cycle_t nsec)
        return nsec;
 }
 
-static atomic_t                        tracer_counter;
+notrace cycle_t ftrace_now(int cpu)
+{
+       return cpu_clock(cpu);
+}
+
 static struct trace_array      global_trace;
 
 static DEFINE_PER_CPU(struct trace_array_cpu, global_trace_cpu);
@@ -49,7 +55,7 @@ static struct trace_array     max_tr;
 
 static DEFINE_PER_CPU(struct trace_array_cpu, max_data);
 
-static int                     tracer_enabled;
+static int                     tracer_enabled = 1;
 static unsigned long           trace_nr_entries = 16384UL;
 
 static struct tracer           *trace_types __read_mostly;
@@ -69,11 +75,17 @@ static int __init set_nr_entries(char *str)
 }
 __setup("trace_entries=", set_nr_entries);
 
+unsigned long nsecs_to_usecs(unsigned long nsecs)
+{
+       return nsecs / 1000;
+}
+
 enum trace_type {
        __TRACE_FIRST_TYPE = 0,
 
        TRACE_FN,
        TRACE_CTX,
+       TRACE_SPECIAL,
 
        __TRACE_LAST_TYPE
 };
@@ -90,6 +102,8 @@ enum trace_iterator_flags {
        TRACE_ITER_SYM_OFFSET           = 0x02,
        TRACE_ITER_SYM_ADDR             = 0x04,
        TRACE_ITER_VERBOSE              = 0x08,
+       TRACE_ITER_RAW                  = 0x10,
+       TRACE_ITER_BIN                  = 0x20,
 };
 
 #define TRACE_ITER_SYM_MASK \
@@ -101,6 +115,8 @@ static const char *trace_options[] = {
        "sym-offset",
        "sym-addr",
        "verbose",
+       "raw",
+       "bin",
        NULL
 };
 
@@ -113,7 +129,7 @@ static DEFINE_SPINLOCK(ftrace_max_lock);
  * 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];
@@ -135,12 +151,135 @@ __update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
        tracing_record_cmdline(current);
 }
 
+void check_pages(struct trace_array_cpu *data)
+{
+       struct page *page, *tmp;
+
+       BUG_ON(data->trace_pages.next->prev != &data->trace_pages);
+       BUG_ON(data->trace_pages.prev->next != &data->trace_pages);
+
+       list_for_each_entry_safe(page, tmp, &data->trace_pages, lru) {
+               BUG_ON(page->lru.next->prev != &page->lru);
+               BUG_ON(page->lru.prev->next != &page->lru);
+       }
+}
+
+void *head_page(struct trace_array_cpu *data)
+{
+       struct page *page;
+
+       check_pages(data);
+       if (list_empty(&data->trace_pages))
+               return NULL;
+
+       page = list_entry(data->trace_pages.next, struct page, lru);
+       BUG_ON(&page->lru == &data->trace_pages);
+
+       return page_address(page);
+}
+
+static notrace int
+trace_seq_printf(struct trace_seq *s, const char *fmt, ...)
+{
+       int len = (PAGE_SIZE - 1) - s->len;
+       va_list ap;
+       int ret;
+
+       if (!len)
+               return 0;
+
+       va_start(ap, fmt);
+       ret = vsnprintf(s->buffer + s->len, len, fmt, ap);
+       va_end(ap);
+
+       /* If we can't write it all, don't bother writing anything */
+       if (ret > len)
+               return 0;
+
+       s->len += ret;
+
+       return len;
+}
+
+static notrace int
+trace_seq_puts(struct trace_seq *s, const char *str)
+{
+       int len = strlen(str);
+
+       if (len > ((PAGE_SIZE - 1) - s->len))
+               return 0;
+
+       memcpy(s->buffer + s->len, str, len);
+       s->len += len;
+
+       return len;
+}
+
+static notrace int
+trace_seq_putc(struct trace_seq *s, unsigned char c)
+{
+       if (s->len >= (PAGE_SIZE - 1))
+               return 0;
+
+       s->buffer[s->len++] = c;
+
+       return 1;
+}
+
+static notrace int
+trace_seq_putmem(struct trace_seq *s, void *mem, size_t len)
+{
+       if (len > ((PAGE_SIZE - 1) - s->len))
+               return 0;
+
+       memcpy(s->buffer + s->len, mem, len);
+       s->len += len;
+
+       return len;
+}
+
+static notrace void
+trace_seq_reset(struct trace_seq *s)
+{
+       s->len = 0;
+}
+
+static notrace 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_reset(s);
+}
+
+notrace static void
+flip_trace(struct trace_array_cpu *tr1, struct trace_array_cpu *tr2)
+{
+       struct list_head flip_pages;
+
+       INIT_LIST_HEAD(&flip_pages);
+
+       memcpy(&tr1->trace_head_idx, &tr2->trace_head_idx,
+               sizeof(struct trace_array_cpu) -
+               offsetof(struct trace_array_cpu, trace_head_idx));
+
+       check_pages(tr1);
+       check_pages(tr2);
+       list_splice_init(&tr1->trace_pages, &flip_pages);
+       list_splice_init(&tr2->trace_pages, &tr1->trace_pages);
+       list_splice_init(&flip_pages, &tr2->trace_pages);
+       BUG_ON(!list_empty(&flip_pages));
+       check_pages(tr1);
+       check_pages(tr2);
+}
+
 notrace void
 update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
 {
        struct trace_array_cpu *data;
-       void *save_trace;
-       struct list_head save_pages;
        int i;
 
        WARN_ON_ONCE(!irqs_disabled());
@@ -148,11 +287,8 @@ update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
        /* clear out all the previous traces */
        for_each_possible_cpu(i) {
                data = tr->data[i];
-               save_trace = max_tr.data[i]->trace;
-               save_pages = max_tr.data[i]->trace_pages;
-               memcpy(max_tr.data[i], data, sizeof(*data));
-               data->trace = save_trace;
-               data->trace_pages = save_pages;
+               flip_trace(max_tr.data[i], data);
+               tracing_reset(data);
        }
 
        __update_max_tr(tr, tsk, cpu);
@@ -169,8 +305,6 @@ notrace void
 update_max_tr_single(struct trace_array *tr, struct task_struct *tsk, int cpu)
 {
        struct trace_array_cpu *data = tr->data[cpu];
-       void *save_trace;
-       struct list_head save_pages;
        int i;
 
        WARN_ON_ONCE(!irqs_disabled());
@@ -178,11 +312,8 @@ update_max_tr_single(struct trace_array *tr, struct task_struct *tsk, int cpu)
        for_each_possible_cpu(i)
                tracing_reset(max_tr.data[i]);
 
-       save_trace = max_tr.data[cpu]->trace;
-       save_pages = max_tr.data[cpu]->trace_pages;
-       memcpy(max_tr.data[cpu], data, sizeof(*data));
-       data->trace = save_trace;
-       data->trace_pages = save_pages;
+       flip_trace(max_tr.data[cpu], data);
+       tracing_reset(data);
 
        __update_max_tr(tr, tsk, cpu);
        spin_unlock(&ftrace_max_lock);
@@ -210,11 +341,55 @@ int register_tracer(struct tracer *type)
                }
        }
 
+#ifdef CONFIG_FTRACE_STARTUP_TEST
+       if (type->selftest) {
+               struct tracer *saved_tracer = current_trace;
+               struct trace_array_cpu *data;
+               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
+                * tracer to be this tracer. The tracer can then run some
+                * internal tracing to verify that everything is in order.
+                * If we fail, we do not register this tracer.
+                */
+               for_each_possible_cpu(i) {
+                       data = tr->data[i];
+                       if (!head_page(data))
+                               continue;
+                       tracing_reset(data);
+               }
+               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_possible_cpu(i) {
+                       data = tr->data[i];
+                       if (!head_page(data))
+                               continue;
+                       tracing_reset(data);
+               }
+               printk(KERN_CONT "PASSED\n");
+       }
+#endif
+
        type->next = trace_types;
        trace_types = type;
        len = strlen(type->name);
        if (len > max_tracer_type_len)
                max_tracer_type_len = len;
+
  out:
        mutex_unlock(&trace_types_lock);
 
@@ -249,52 +424,12 @@ void unregister_tracer(struct tracer *type)
        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 = data->trace;
-       data->trace_current_idx = 0;
-}
-
-#ifdef CONFIG_FTRACE
-static void notrace
-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;
-
-       if (unlikely(!tracer_enabled))
-               return;
-
-       local_irq_save(flags);
-       cpu = raw_smp_processor_id();
-       data = tr->data[cpu];
-       disabled = atomic_inc_return(&data->disabled);
-
-       if (likely(disabled == 1))
-               ftrace(tr, data, ip, parent_ip, flags);
-
-       atomic_dec(&data->disabled);
-       local_irq_restore(flags);
-}
-
-static struct ftrace_ops trace_ops __read_mostly =
-{
-       .func = function_trace_call,
-};
-#endif
-
-notrace void tracing_start_function_trace(void)
-{
-       register_ftrace_function(&trace_ops);
-}
-
-notrace void tracing_stop_function_trace(void)
-{
-       unregister_ftrace_function(&trace_ops);
+       data->trace_head = data->trace_tail = head_page(data);
+       data->trace_head_idx = 0;
+       data->trace_tail_idx = 0;
 }
 
 #define SAVED_CMDLINES 128
@@ -314,7 +449,7 @@ static void trace_init_cmdlines(void)
 
 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;
@@ -378,50 +513,80 @@ notrace void tracing_record_cmdline(struct task_struct *tsk)
        trace_save_cmdline(tsk);
 }
 
+static inline notrace struct list_head *
+trace_next_list(struct trace_array_cpu *data, struct list_head *next)
+{
+       /*
+        * Roundrobin - but skip the head (which is not a real page):
+        */
+       next = next->next;
+       if (unlikely(next == &data->trace_pages))
+               next = next->next;
+       BUG_ON(next == &data->trace_pages);
+
+       return next;
+}
+
+static inline notrace void *
+trace_next_page(struct trace_array_cpu *data, void *addr)
+{
+       struct list_head *next;
+       struct page *page;
+
+       page = virt_to_page(addr);
+
+       next = trace_next_list(data, &page->lru);
+       page = list_entry(next, struct page, lru);
+
+       return page_address(page);
+}
+
 static inline notrace struct trace_entry *
-tracing_get_trace_entry(struct trace_array *tr,
-                       struct trace_array_cpu *data)
+tracing_get_trace_entry(struct trace_array *tr, struct trace_array_cpu *data)
 {
        unsigned long idx, idx_next;
        struct trace_entry *entry;
-       struct page *page;
-       struct list_head *next;
 
        data->trace_idx++;
-       idx = data->trace_current_idx;
+       idx = data->trace_head_idx;
        idx_next = idx + 1;
 
-       entry = data->trace_current + idx * TRACE_ENTRY_SIZE;
+       BUG_ON(idx * TRACE_ENTRY_SIZE >= PAGE_SIZE);
+
+       entry = data->trace_head + idx * TRACE_ENTRY_SIZE;
 
        if (unlikely(idx_next >= ENTRIES_PER_PAGE)) {
-               page = virt_to_page(data->trace_current);
-               if (unlikely(&page->lru == data->trace_pages.prev))
-                       next = data->trace_pages.next;
-               else
-                       next = page->lru.next;
-               page = list_entry(next, struct page, lru);
-               data->trace_current = page_address(page);
+               data->trace_head = trace_next_page(data, data->trace_head);
                idx_next = 0;
        }
 
-       data->trace_current_idx = idx_next;
+       if (data->trace_head == data->trace_tail &&
+           idx_next == data->trace_tail_idx) {
+               /* overrun */
+               data->trace_tail_idx++;
+               if (data->trace_tail_idx >= ENTRIES_PER_PAGE) {
+                       data->trace_tail =
+                               trace_next_page(data, data->trace_tail);
+                       data->trace_tail_idx = 0;
+               }
+       }
+
+       data->trace_head_idx = idx_next;
 
        return entry;
 }
 
 static inline notrace void
-tracing_generic_entry_update(struct trace_entry *entry,
-                            unsigned long flags)
+tracing_generic_entry_update(struct trace_entry *entry, unsigned long flags)
 {
        struct task_struct *tsk = current;
        unsigned long pc;
 
        pc = preempt_count();
 
-       entry->idx      = atomic_inc_return(&tracer_counter);
-       entry->preempt_count = pc & 0xff;
-       entry->pid       = tsk->pid;
-       entry->t         = now(raw_smp_processor_id());
+       entry->preempt_count    = pc & 0xff;
+       entry->pid              = tsk->pid;
+       entry->t                = ftrace_now(raw_smp_processor_id());
        entry->flags = (irqs_disabled_flags(flags) ? TRACE_FLAG_IRQS_OFF : 0) |
                ((pc & HARDIRQ_MASK) ? TRACE_FLAG_HARDIRQ : 0) |
                ((pc & SOFTIRQ_MASK) ? TRACE_FLAG_SOFTIRQ : 0) |
@@ -429,17 +594,44 @@ tracing_generic_entry_update(struct trace_entry *entry,
 }
 
 notrace void
-ftrace(struct trace_array *tr, struct trace_array_cpu *data,
-                      unsigned long ip, unsigned long parent_ip,
-                      unsigned long flags)
+__ftrace(struct trace_array *tr, struct trace_array_cpu *data,
+        unsigned long ip, unsigned long parent_ip, unsigned long flags)
 {
        struct trace_entry *entry;
+       unsigned long irq_flags;
 
-       entry = tracing_get_trace_entry(tr, data);
+       spin_lock_irqsave(&data->lock, irq_flags);
+       entry                   = tracing_get_trace_entry(tr, data);
        tracing_generic_entry_update(entry, flags);
-       entry->type         = TRACE_FN;
-       entry->fn.ip        = ip;
-       entry->fn.parent_ip = parent_ip;
+       entry->type             = TRACE_FN;
+       entry->fn.ip            = ip;
+       entry->fn.parent_ip     = parent_ip;
+       spin_unlock_irqrestore(&data->lock, irq_flags);
+}
+
+notrace void
+ftrace(struct trace_array *tr, struct trace_array_cpu *data,
+       unsigned long ip, unsigned long parent_ip, unsigned long flags)
+{
+       if (likely(!atomic_read(&data->disabled)))
+               __ftrace(tr, data, ip, parent_ip, flags);
+}
+
+notrace void
+trace_special(struct trace_array *tr, struct trace_array_cpu *data,
+             unsigned long arg1, unsigned long arg2, unsigned long arg3)
+{
+       struct trace_entry *entry;
+       unsigned long irq_flags;
+
+       spin_lock_irqsave(&data->lock, irq_flags);
+       entry                   = tracing_get_trace_entry(tr, data);
+       tracing_generic_entry_update(entry, 0);
+       entry->type             = TRACE_SPECIAL;
+       entry->special.arg1     = arg1;
+       entry->special.arg2     = arg2;
+       entry->special.arg3     = arg3;
+       spin_unlock_irqrestore(&data->lock, irq_flags);
 }
 
 notrace void
@@ -449,8 +641,10 @@ tracing_sched_switch_trace(struct trace_array *tr,
                           unsigned long flags)
 {
        struct trace_entry *entry;
+       unsigned long irq_flags;
 
-       entry = tracing_get_trace_entry(tr, data);
+       spin_lock_irqsave(&data->lock, irq_flags);
+       entry                   = tracing_get_trace_entry(tr, data);
        tracing_generic_entry_update(entry, flags);
        entry->type             = TRACE_CTX;
        entry->ctx.prev_pid     = prev->pid;
@@ -458,8 +652,50 @@ tracing_sched_switch_trace(struct trace_array *tr,
        entry->ctx.prev_state   = prev->state;
        entry->ctx.next_pid     = next->pid;
        entry->ctx.next_prio    = next->prio;
+       spin_unlock_irqrestore(&data->lock, irq_flags);
+}
+
+#ifdef CONFIG_FTRACE
+static notrace 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;
+
+       if (unlikely(!tracer_enabled))
+               return;
+
+       local_irq_save(flags);
+       cpu = raw_smp_processor_id();
+       data = tr->data[cpu];
+       disabled = atomic_inc_return(&data->disabled);
+
+       if (likely(disabled == 1))
+               __ftrace(tr, data, ip, parent_ip, flags);
+
+       atomic_dec(&data->disabled);
+       local_irq_restore(flags);
 }
 
+static struct ftrace_ops trace_ops __read_mostly =
+{
+       .func = function_trace_call,
+};
+
+notrace void tracing_start_function_trace(void)
+{
+       register_ftrace_function(&trace_ops);
+}
+
+notrace void tracing_stop_function_trace(void)
+{
+       unregister_ftrace_function(&trace_ops);
+}
+#endif
+
 enum trace_file_type {
        TRACE_FILE_LAT_FMT      = 1,
 };
@@ -472,34 +708,29 @@ trace_entry_idx(struct trace_array *tr, struct trace_array_cpu *data,
        struct trace_entry *array;
 
        if (iter->next_idx[cpu] >= tr->entries ||
-           iter->next_idx[cpu] >= data->trace_idx)
+           iter->next_idx[cpu] >= data->trace_idx ||
+           (data->trace_head == data->trace_tail &&
+            data->trace_head_idx == data->trace_tail_idx))
                return NULL;
 
        if (!iter->next_page[cpu]) {
-               /*
-                * Initialize. If the count of elements in
-                * this buffer is greater than the max entries
-                * we had an underrun. Which means we looped around.
-                * We can simply use the current pointer as our
-                * starting point.
-                */
-               if (data->trace_idx >= tr->entries) {
-                       page = virt_to_page(data->trace_current);
-                       iter->next_page[cpu] = &page->lru;
-                       iter->next_page_idx[cpu] = data->trace_current_idx;
-               } else {
-                       iter->next_page[cpu] = data->trace_pages.next;
-                       iter->next_page_idx[cpu] = 0;
-               }
+               /* Initialize the iterator for this cpu trace buffer */
+               WARN_ON(!data->trace_tail);
+               page = virt_to_page(data->trace_tail);
+               iter->next_page[cpu] = &page->lru;
+               iter->next_page_idx[cpu] = data->trace_tail_idx;
        }
 
        page = list_entry(iter->next_page[cpu], struct page, lru);
+       BUG_ON(&data->trace_pages == &page->lru);
+
        array = page_address(page);
 
+       WARN_ON(iter->next_page_idx[cpu] >= ENTRIES_PER_PAGE);
        return &array[iter->next_page_idx[cpu]];
 }
 
-static struct notrace trace_entry *
+static struct trace_entry * notrace
 find_next_entry(struct trace_iterator *iter, int *ent_cpu)
 {
        struct trace_array *tr = iter->tr;
@@ -508,11 +739,13 @@ find_next_entry(struct trace_iterator *iter, int *ent_cpu)
        int cpu;
 
        for_each_possible_cpu(cpu) {
-               if (!tr->data[cpu]->trace)
+               if (!head_page(tr->data[cpu]))
                        continue;
                ent = trace_entry_idx(tr, tr->data[cpu], iter, cpu);
-               if (ent &&
-                   (!next || (long)(next->idx - ent->idx) > 0)) {
+               /*
+                * Pick the entry with the smallest timestamp:
+                */
+               if (ent && (!next || ent->t < next->t)) {
                        next = ent;
                        next_cpu = cpu;
                }
@@ -524,41 +757,62 @@ find_next_entry(struct trace_iterator *iter, int *ent_cpu)
        return next;
 }
 
-static void *find_next_entry_inc(struct trace_iterator *iter)
+static notrace void trace_iterator_increment(struct trace_iterator *iter)
+{
+       iter->idx++;
+       iter->next_idx[iter->cpu]++;
+       iter->next_page_idx[iter->cpu]++;
+
+       if (iter->next_page_idx[iter->cpu] >= ENTRIES_PER_PAGE) {
+               struct trace_array_cpu *data = iter->tr->data[iter->cpu];
+
+               iter->next_page_idx[iter->cpu] = 0;
+               iter->next_page[iter->cpu] =
+                       trace_next_list(data, iter->next_page[iter->cpu]);
+       }
+}
+
+static notrace void trace_consume(struct trace_iterator *iter)
+{
+       struct trace_array_cpu *data = iter->tr->data[iter->cpu];
+
+       data->trace_tail_idx++;
+       if (data->trace_tail_idx >= ENTRIES_PER_PAGE) {
+               data->trace_tail = trace_next_page(data, data->trace_tail);
+               data->trace_tail_idx = 0;
+       }
+
+       /* Check if we empty it, then reset the index */
+       if (data->trace_head == data->trace_tail &&
+           data->trace_head_idx == data->trace_tail_idx)
+               data->trace_idx = 0;
+}
+
+static notrace void *find_next_entry_inc(struct trace_iterator *iter)
 {
        struct trace_entry *next;
        int next_cpu = -1;
 
        next = find_next_entry(iter, &next_cpu);
 
-       if (next) {
-               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];
-
-                       iter->next_page_idx[next_cpu] = 0;
-                       iter->next_page[next_cpu] =
-                               iter->next_page[next_cpu]->next;
-                       if (iter->next_page[next_cpu] == &data->trace_pages)
-                               iter->next_page[next_cpu] =
-                                       data->trace_pages.next;
-               }
-       }
+       iter->prev_ent = iter->ent;
+       iter->prev_cpu = iter->cpu;
+
        iter->ent = next;
        iter->cpu = next_cpu;
 
+       if (next)
+               trace_iterator_increment(iter);
+
        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)++;
 
@@ -604,6 +858,8 @@ static void *s_start(struct seq_file *m, loff_t *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;
@@ -634,27 +890,30 @@ static void s_stop(struct seq_file *m, void *p)
        mutex_unlock(&trace_types_lock);
 }
 
-static void
-seq_print_sym_short(struct seq_file *m, const char *fmt, unsigned long address)
+static int
+seq_print_sym_short(struct trace_seq *s, const char *fmt, unsigned long address)
 {
 #ifdef CONFIG_KALLSYMS
        char str[KSYM_SYMBOL_LEN];
 
        kallsyms_lookup(address, NULL, NULL, NULL, str);
 
-       seq_printf(m, fmt, str);
+       return trace_seq_printf(s, fmt, str);
 #endif
+       return 1;
 }
 
-static void
-seq_print_sym_offset(struct seq_file *m, const char *fmt, unsigned long address)
+static int
+seq_print_sym_offset(struct trace_seq *s, const char *fmt,
+                    unsigned long address)
 {
 #ifdef CONFIG_KALLSYMS
        char str[KSYM_SYMBOL_LEN];
 
        sprint_symbol(str, address);
-       seq_printf(m, fmt, str);
+       return trace_seq_printf(s, fmt, str);
 #endif
+       return 1;
 }
 
 #ifndef CONFIG_64BIT
@@ -663,24 +922,28 @@ seq_print_sym_offset(struct seq_file *m, const char *fmt, unsigned long address)
 # define IP_FMT "%016lx"
 #endif
 
-static void notrace
-seq_print_ip_sym(struct seq_file *m, unsigned long ip, unsigned long sym_flags)
+static notrace int
+seq_print_ip_sym(struct trace_seq *s, unsigned long ip, unsigned long sym_flags)
 {
-       if (!ip) {
-               seq_printf(m, "0");
-               return;
-       }
+       int ret;
+
+       if (!ip)
+               return trace_seq_printf(s, "0");
 
        if (sym_flags & TRACE_ITER_SYM_OFFSET)
-               seq_print_sym_offset(m, "%s", ip);
+               ret = seq_print_sym_offset(s, "%s", ip);
        else
-               seq_print_sym_short(m, "%s", ip);
+               ret = seq_print_sym_short(s, "%s", ip);
+
+       if (!ret)
+               return 0;
 
        if (sym_flags & TRACE_ITER_SYM_ADDR)
-               seq_printf(m, " <" IP_FMT ">", ip);
+               ret = trace_seq_printf(s, " <" IP_FMT ">", ip);
+       return ret;
 }
 
-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");
@@ -693,14 +956,14 @@ static void notrace print_lat_help_header(struct seq_file *m)
        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);
@@ -716,7 +979,7 @@ print_trace_header(struct seq_file *m, struct trace_iterator *iter)
                name = type->name;
 
        for_each_possible_cpu(cpu) {
-               if (tr->data[cpu]->trace) {
+               if (head_page(tr->data[cpu])) {
                        total += tr->data[cpu]->trace_idx;
                        if (tr->data[cpu]->trace_idx > tr->entries)
                                entries += tr->entries;
@@ -731,7 +994,7 @@ print_trace_header(struct seq_file *m, struct trace_iterator *iter)
                 "---------------------------------\n");
        seq_printf(m, " latency: %lu us, #%lu/%lu, CPU#%d |"
                   " (M:%s VP:%d, KP:%d, SP:%d HP:%d",
-                  data->saved_latency,
+                  nsecs_to_usecs(data->saved_latency),
                   entries,
                   total,
                   tr->cpu,
@@ -760,76 +1023,73 @@ print_trace_header(struct seq_file *m, struct trace_iterator *iter)
 
        if (data->critical_start) {
                seq_puts(m, " => started at: ");
-               seq_print_ip_sym(m, data->critical_start, sym_flags);
+               seq_print_ip_sym(&iter->seq, data->critical_start, sym_flags);
+               trace_print_seq(m, &iter->seq);
                seq_puts(m, "\n => ended at:   ");
-               seq_print_ip_sym(m, data->critical_end, sym_flags);
+               seq_print_ip_sym(&iter->seq, data->critical_end, sym_flags);
+               trace_print_seq(m, &iter->seq);
                seq_puts(m, "\n");
        }
 
        seq_puts(m, "\n");
 }
 
-unsigned long nsecs_to_usecs(unsigned long nsecs)
-{
-       return nsecs / 1000;
-}
-
-static void notrace
-lat_print_generic(struct seq_file *m, struct trace_entry *entry, int cpu)
+static notrace void
+lat_print_generic(struct trace_seq *s, struct trace_entry *entry, int cpu)
 {
        int hardirq, softirq;
        char *comm;
 
        comm = trace_find_cmdline(entry->pid);
 
-       seq_printf(m, "%8.8s-%-5d ", comm, entry->pid);
-       seq_printf(m, "%d", cpu);
-       seq_printf(m, "%c%c",
-                  (entry->flags & TRACE_FLAG_IRQS_OFF) ? 'd' : '.',
-                  ((entry->flags & TRACE_FLAG_NEED_RESCHED) ? 'N' : '.'));
+       trace_seq_printf(s, "%8.8s-%-5d ", comm, entry->pid);
+       trace_seq_printf(s, "%d", cpu);
+       trace_seq_printf(s, "%c%c",
+                       (entry->flags & TRACE_FLAG_IRQS_OFF) ? 'd' : '.',
+                       ((entry->flags & TRACE_FLAG_NEED_RESCHED) ? 'N' : '.'));
 
        hardirq = entry->flags & TRACE_FLAG_HARDIRQ;
        softirq = entry->flags & TRACE_FLAG_SOFTIRQ;
        if (hardirq && softirq)
-               seq_putc(m, 'H');
+               trace_seq_putc(s, 'H');
        else {
                if (hardirq)
-                       seq_putc(m, 'h');
+                       trace_seq_putc(s, 'h');
                else {
                        if (softirq)
-                               seq_putc(m, 's');
+                               trace_seq_putc(s, 's');
                        else
-                               seq_putc(m, '.');
+                               trace_seq_putc(s, '.');
                }
        }
 
        if (entry->preempt_count)
-               seq_printf(m, "%x", entry->preempt_count);
+               trace_seq_printf(s, "%x", entry->preempt_count);
        else
-               seq_puts(m, ".");
+               trace_seq_puts(s, ".");
 }
 
 unsigned long preempt_mark_thresh = 100;
 
-static void notrace
-lat_print_timestamp(struct seq_file *m, unsigned long long abs_usecs,
+static notrace void
+lat_print_timestamp(struct trace_seq *s, unsigned long long abs_usecs,
                    unsigned long rel_usecs)
 {
-       seq_printf(m, " %4lldus", abs_usecs);
+       trace_seq_printf(s, " %4lldus", abs_usecs);
        if (rel_usecs > preempt_mark_thresh)
-               seq_puts(m, "!: ");
+               trace_seq_puts(s, "!: ");
        else if (rel_usecs > 1)
-               seq_puts(m, "+: ");
+               trace_seq_puts(s, "+: ");
        else
-               seq_puts(m, " : ");
+               trace_seq_puts(s, " : ");
 }
 
 static const char state_to_char[] = TASK_STATE_TO_CHAR_STR;
 
-static void notrace
-print_lat_fmt(struct seq_file *m, struct trace_iterator *iter,
-             unsigned int trace_idx, int cpu)
+static notrace int
+print_lat_fmt(struct trace_iterator *iter, unsigned int trace_idx, int cpu)
 {
+       struct trace_seq *s = &iter->seq;
        unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
        struct trace_entry *next_entry = find_next_entry(iter, NULL);
        unsigned long verbose = (trace_flags & TRACE_ITER_VERBOSE);
@@ -846,50 +1106,63 @@ print_lat_fmt(struct seq_file *m, struct trace_iterator *iter,
 
        if (verbose) {
                comm = trace_find_cmdline(entry->pid);
-               seq_printf(m, "%16s %5d %d %d %08x %08x [%08lx]"
-                          " %ld.%03ldms (+%ld.%03ldms): ",
-                          comm,
-                          entry->pid, cpu, entry->flags,
-                          entry->preempt_count, trace_idx,
-                          ns2usecs(entry->t),
-                          abs_usecs/1000,
-                          abs_usecs % 1000, rel_usecs/1000, rel_usecs % 1000);
+               trace_seq_printf(s, "%16s %5d %d %d %08x %08x [%08lx]"
+                                " %ld.%03ldms (+%ld.%03ldms): ",
+                                comm,
+                                entry->pid, cpu, entry->flags,
+                                entry->preempt_count, trace_idx,
+                                ns2usecs(entry->t),
+                                abs_usecs/1000,
+                                abs_usecs % 1000, rel_usecs/1000,
+                                rel_usecs % 1000);
        } else {
-               lat_print_generic(m, entry, cpu);
-               lat_print_timestamp(m, abs_usecs, rel_usecs);
+               lat_print_generic(s, entry, cpu);
+               lat_print_timestamp(s, abs_usecs, rel_usecs);
        }
        switch (entry->type) {
        case TRACE_FN:
-               seq_print_ip_sym(m, entry->fn.ip, sym_flags);
-               seq_puts(m, " (");
-               seq_print_ip_sym(m, entry->fn.parent_ip, sym_flags);
-               seq_puts(m, ")\n");
+               seq_print_ip_sym(s, entry->fn.ip, sym_flags);
+               trace_seq_puts(s, " (");
+               seq_print_ip_sym(s, entry->fn.parent_ip, sym_flags);
+               trace_seq_puts(s, ")\n");
                break;
        case TRACE_CTX:
                S = entry->ctx.prev_state < sizeof(state_to_char) ?
                        state_to_char[entry->ctx.prev_state] : 'X';
                comm = trace_find_cmdline(entry->ctx.next_pid);
-               seq_printf(m, " %d:%d:%c --> %d:%d %s\n",
-                          entry->ctx.prev_pid,
-                          entry->ctx.prev_prio,
-                          S,
-                          entry->ctx.next_pid,
-                          entry->ctx.next_prio,
-                          comm);
+               trace_seq_printf(s, " %d:%d:%c --> %d:%d %s\n",
+                                entry->ctx.prev_pid,
+                                entry->ctx.prev_prio,
+                                S,
+                                entry->ctx.next_pid,
+                                entry->ctx.next_prio,
+                                comm);
+               break;
+       case TRACE_SPECIAL:
+               trace_seq_printf(s, " %lx %lx %lx\n",
+                                entry->special.arg1,
+                                entry->special.arg2,
+                                entry->special.arg3);
                break;
+       default:
+               trace_seq_printf(s, "Unknown type %d\n", entry->type);
        }
+       return 1;
 }
 
-static void notrace
-print_trace_fmt(struct seq_file *m, struct trace_iterator *iter)
+static notrace int print_trace_fmt(struct trace_iterator *iter)
 {
+       struct trace_seq *s = &iter->seq;
        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;
+       int ret;
+
+       entry = iter->ent;
 
        comm = trace_find_cmdline(iter->ent->pid);
 
@@ -897,31 +1170,140 @@ print_trace_fmt(struct seq_file *m, struct trace_iterator *iter)
        usec_rem = do_div(t, 1000000ULL);
        secs = (unsigned long)t;
 
-       seq_printf(m, "%16s-%-5d ", comm, entry->pid);
-       seq_printf(m, "[%02d] ", iter->cpu);
-       seq_printf(m, "%5lu.%06lu: ", secs, usec_rem);
+       ret = trace_seq_printf(s, "%16s-%-5d ", comm, entry->pid);
+       if (!ret)
+               return 0;
+       ret = trace_seq_printf(s, "[%02d] ", iter->cpu);
+       if (!ret)
+               return 0;
+       ret = trace_seq_printf(s, "%5lu.%06lu: ", secs, usec_rem);
+       if (!ret)
+               return 0;
 
        switch (entry->type) {
        case TRACE_FN:
-               seq_print_ip_sym(m, entry->fn.ip, sym_flags);
+               ret = seq_print_ip_sym(s, entry->fn.ip, sym_flags);
+               if (!ret)
+                       return 0;
                if ((sym_flags & TRACE_ITER_PRINT_PARENT) &&
                                                entry->fn.parent_ip) {
-                       seq_printf(m, " <-");
-                       seq_print_ip_sym(m, entry->fn.parent_ip, sym_flags);
+                       ret = trace_seq_printf(s, " <-");
+                       if (!ret)
+                               return 0;
+                       ret = seq_print_ip_sym(s, entry->fn.parent_ip,
+                                              sym_flags);
+                       if (!ret)
+                               return 0;
                }
+               ret = trace_seq_printf(s, "\n");
+               if (!ret)
+                       return 0;
                break;
        case TRACE_CTX:
                S = entry->ctx.prev_state < sizeof(state_to_char) ?
                        state_to_char[entry->ctx.prev_state] : 'X';
-               seq_printf(m, " %d:%d:%c ==> %d:%d\n",
-                          entry->ctx.prev_pid,
-                          entry->ctx.prev_prio,
-                          S,
-                          entry->ctx.next_pid,
-                          entry->ctx.next_prio);
+               ret = trace_seq_printf(s, " %d:%d:%c ==> %d:%d\n",
+                                      entry->ctx.prev_pid,
+                                      entry->ctx.prev_prio,
+                                      S,
+                                      entry->ctx.next_pid,
+                                      entry->ctx.next_prio);
+               if (!ret)
+                       return 0;
+               break;
+       case TRACE_SPECIAL:
+               ret = trace_seq_printf(s, " %lx %lx %lx\n",
+                                entry->special.arg1,
+                                entry->special.arg2,
+                                entry->special.arg3);
+               if (!ret)
+                       return 0;
+               break;
+       }
+       return 1;
+}
+
+static notrace int print_raw_fmt(struct trace_iterator *iter)
+{
+       struct trace_seq *s = &iter->seq;
+       struct trace_entry *entry;
+       int ret;
+       int S;
+
+       entry = iter->ent;
+
+       ret = trace_seq_printf(s, "%d %d %llu ",
+               entry->pid, iter->cpu, entry->t);
+       if (!ret)
+               return 0;
+
+       switch (entry->type) {
+       case TRACE_FN:
+               ret = trace_seq_printf(s, "%x %x\n",
+                                       entry->fn.ip, entry->fn.parent_ip);
+               if (!ret)
+                       return 0;
+               break;
+       case TRACE_CTX:
+               S = entry->ctx.prev_state < sizeof(state_to_char) ?
+                       state_to_char[entry->ctx.prev_state] : 'X';
+               ret = trace_seq_printf(s, "%d %d %c %d %d\n",
+                                      entry->ctx.prev_pid,
+                                      entry->ctx.prev_prio,
+                                      S,
+                                      entry->ctx.next_pid,
+                                      entry->ctx.next_prio);
+               if (!ret)
+                       return 0;
+               break;
+       case TRACE_SPECIAL:
+               ret = trace_seq_printf(s, " %lx %lx %lx\n",
+                                entry->special.arg1,
+                                entry->special.arg2,
+                                entry->special.arg3);
+               if (!ret)
+                       return 0;
                break;
        }
-       seq_printf(m, "\n");
+       return 1;
+}
+
+#define SEQ_PUT_FIELD_RET(s, x)                                \
+do {                                                   \
+       if (!trace_seq_putmem(s, &(x), sizeof(x)))      \
+               return 0;                               \
+} while (0)
+
+static notrace int print_bin_fmt(struct trace_iterator *iter)
+{
+       struct trace_seq *s = &iter->seq;
+       struct trace_entry *entry;
+
+       entry = iter->ent;
+
+       SEQ_PUT_FIELD_RET(s, entry->pid);
+       SEQ_PUT_FIELD_RET(s, entry->cpu);
+       SEQ_PUT_FIELD_RET(s, entry->t);
+
+       switch (entry->type) {
+       case TRACE_FN:
+               SEQ_PUT_FIELD_RET(s, entry->fn.ip);
+               SEQ_PUT_FIELD_RET(s, entry->fn.parent_ip);
+               break;
+       case TRACE_CTX:
+               SEQ_PUT_FIELD_RET(s, entry->ctx.prev_pid);
+               SEQ_PUT_FIELD_RET(s, entry->ctx.prev_prio);
+               SEQ_PUT_FIELD_RET(s, entry->ctx.prev_state);
+               SEQ_PUT_FIELD_RET(s, entry->ctx.next_pid);
+               SEQ_PUT_FIELD_RET(s, entry->ctx.next_prio);
+               break;
+       case TRACE_SPECIAL:
+               SEQ_PUT_FIELD_RET(s, entry->special.arg1);
+               SEQ_PUT_FIELD_RET(s, entry->special.arg2);
+               SEQ_PUT_FIELD_RET(s, entry->special.arg3);
+               break;
+       }
+       return 1;
 }
 
 static int trace_empty(struct trace_iterator *iter)
@@ -932,13 +1314,28 @@ static int trace_empty(struct trace_iterator *iter)
        for_each_possible_cpu(cpu) {
                data = iter->tr->data[cpu];
 
-               if (data->trace &&
-                   data->trace_idx)
+               if (head_page(data) && data->trace_idx &&
+                   (data->trace_tail != data->trace_head ||
+                    data->trace_tail_idx != data->trace_head_idx))
                        return 0;
        }
        return 1;
 }
 
+static int print_trace_line(struct trace_iterator *iter)
+{
+       if (trace_flags & TRACE_ITER_BIN)
+               return print_bin_fmt(iter);
+
+       if (trace_flags & TRACE_ITER_RAW)
+               return print_raw_fmt(iter);
+
+       if (iter->iter_flags & TRACE_FILE_LAT_FMT)
+               return print_lat_fmt(iter, iter->idx, iter->cpu);
+
+       return print_trace_fmt(iter);
+}
+
 static int s_show(struct seq_file *m, void *v)
 {
        struct trace_iterator *iter = v;
@@ -960,20 +1357,18 @@ static int s_show(struct seq_file *m, void *v)
                                print_func_help_header(m);
                }
        } else {
-               if (iter->iter_flags & TRACE_FILE_LAT_FMT)
-                       print_lat_fmt(m, iter, iter->idx, iter->cpu);
-               else
-                       print_trace_fmt(m, iter);
+               print_trace_line(iter);
+               trace_print_seq(m, &iter->seq);
        }
 
        return 0;
 }
 
 static struct seq_operations tracer_seq_ops = {
-       .start = s_start,
-       .next = s_next,
-       .stop = s_stop,
-       .show = s_show,
+       .start          = s_start,
+       .next           = s_next,
+       .stop           = s_stop,
+       .show           = s_show,
 };
 
 static struct trace_iterator notrace *
@@ -981,6 +1376,11 @@ __tracing_open(struct inode *inode, struct file *file, int *ret)
 {
        struct trace_iterator *iter;
 
+       if (tracing_disabled) {
+               *ret = -ENODEV;
+               return NULL;
+       }
+
        iter = kzalloc(sizeof(*iter), GFP_KERNEL);
        if (!iter) {
                *ret = -ENOMEM;
@@ -1019,6 +1419,9 @@ __tracing_open(struct inode *inode, struct file *file, int *ret)
 
 int tracing_open_generic(struct inode *inode, struct file *filp)
 {
+       if (tracing_disabled)
+               return -ENODEV;
+
        filp->private_data = inode->i_private;
        return 0;
 }
@@ -1065,7 +1468,7 @@ static int tracing_lt_open(struct inode *inode, struct file *file)
 }
 
 
-static void notrace *
+static notrace void *
 t_next(struct seq_file *m, void *v, loff_t *pos)
 {
        struct tracer *t = m->private;
@@ -1114,16 +1517,19 @@ static int t_show(struct seq_file *m, void *v)
 }
 
 static struct seq_operations show_traces_seq_ops = {
-       .start = t_start,
-       .next = t_next,
-       .stop = t_stop,
-       .show = t_show,
+       .start          = t_start,
+       .next           = t_next,
+       .stop           = t_stop,
+       .show           = t_show,
 };
 
 static int show_traces_open(struct inode *inode, struct file *file)
 {
        int ret;
 
+       if (tracing_disabled)
+               return -ENODEV;
+
        ret = seq_open(file, &show_traces_seq_ops);
        if (!ret) {
                struct seq_file *m = file->private_data;
@@ -1134,17 +1540,17 @@ static int show_traces_open(struct inode *inode, struct file *file)
 }
 
 static struct file_operations tracing_fops = {
-       .open = tracing_open,
-       .read = seq_read,
-       .llseek = seq_lseek,
-       .release = tracing_release,
+       .open           = tracing_open,
+       .read           = seq_read,
+       .llseek         = seq_lseek,
+       .release        = tracing_release,
 };
 
 static struct file_operations tracing_lt_fops = {
-       .open = tracing_lt_open,
-       .read = seq_read,
-       .llseek = seq_lseek,
-       .release = tracing_release,
+       .open           = tracing_lt_open,
+       .read           = seq_read,
+       .llseek         = seq_lseek,
+       .release        = tracing_release,
 };
 
 static struct file_operations show_traces_fops = {
@@ -1236,6 +1642,39 @@ static struct file_operations tracing_iter_fops = {
        .write = tracing_iter_ctrl_write,
 };
 
+static const char readme_msg[] =
+       "tracing mini-HOWTO:\n\n"
+       "# 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"
+       "# cat /debug/tracing/current_tracer\n"
+       "none\n"
+       "# echo sched_switch > /debug/tracing/current_tracer\n"
+       "# cat /debug/tracing/current_tracer\n"
+       "sched_switch\n"
+       "# cat /debug/tracing/iter_ctrl\n"
+       "noprint-parent nosym-offset nosym-addr noverbose\n"
+       "# echo print-parent > /debug/tracing/iter_ctrl\n"
+       "# echo 1 > /debug/tracing/tracing_enabled\n"
+       "# cat /debug/tracing/trace > /tmp/trace.txt\n"
+       "echo 0 > /debug/tracing/tracing_enabled\n"
+;
+
+static ssize_t
+tracing_readme_read(struct file *filp, char __user *ubuf,
+                      size_t cnt, loff_t *ppos)
+{
+       return simple_read_from_buffer(ubuf, cnt, ppos,
+                                       readme_msg, strlen(readme_msg));
+}
+
+static struct file_operations tracing_readme_fops = {
+       .open = tracing_open_generic,
+       .read = tracing_readme_read,
+};
+
+
 static ssize_t
 tracing_ctrl_read(struct file *filp, char __user *ubuf,
                  size_t cnt, loff_t *ppos)
@@ -1245,8 +1684,7 @@ tracing_ctrl_read(struct file *filp, char __user *ubuf,
        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
@@ -1302,8 +1740,7 @@ tracing_set_trace_read(struct file *filp, char __user *ubuf,
                r = sprintf(buf, "\n");
        mutex_unlock(&trace_types_lock);
 
-       return simple_read_from_buffer(ubuf, cnt, ppos,
-                                      buf, r);
+       return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
 }
 
 static ssize_t
@@ -1362,8 +1799,7 @@ tracing_max_lat_read(struct file *filp, char __user *ubuf,
                     *ptr == (unsigned long)-1 ? -1 : nsecs_to_usecs(*ptr));
        if (r > 64)
                r = 64;
-       return simple_read_from_buffer(ubuf, cnt, ppos,
-                                      buf, r);
+       return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
 }
 
 static ssize_t
@@ -1389,22 +1825,220 @@ tracing_max_lat_write(struct file *filp, const char __user *ubuf,
        return cnt;
 }
 
+static atomic_t tracing_reader;
+
+static int tracing_open_pipe(struct inode *inode, struct file *filp)
+{
+       struct trace_iterator *iter;
+
+       if (tracing_disabled)
+               return -ENODEV;
+
+       /* We only allow for reader of the pipe */
+       if (atomic_inc_return(&tracing_reader) != 1) {
+               atomic_dec(&tracing_reader);
+               return -EBUSY;
+       }
+
+       /* create a buffer to store the information to pass to userspace */
+       iter = kzalloc(sizeof(*iter), GFP_KERNEL);
+       if (!iter)
+               return -ENOMEM;
+
+       iter->tr = &global_trace;
+
+       filp->private_data = iter;
+
+       return 0;
+}
+
+static int tracing_release_pipe(struct inode *inode, struct file *file)
+{
+       struct trace_iterator *iter = file->private_data;
+
+       kfree(iter);
+       atomic_dec(&tracing_reader);
+
+       return 0;
+}
+
+/*
+ * Consumer reader.
+ */
+static ssize_t
+tracing_read_pipe(struct file *filp, char __user *ubuf,
+                 size_t cnt, loff_t *ppos)
+{
+       struct trace_iterator *iter = filp->private_data;
+       struct trace_array_cpu *data;
+       static cpumask_t mask;
+       static int start;
+       unsigned long flags;
+       int ftrace_save;
+       int read = 0;
+       int cpu;
+       int len;
+       int ret;
+
+       /* return any leftover data */
+       if (iter->seq.len > start) {
+               len = iter->seq.len - start;
+               if (cnt > len)
+                       cnt = len;
+               ret = copy_to_user(ubuf, iter->seq.buffer + start, cnt);
+               if (ret)
+                       cnt = -EFAULT;
+
+               start += len;
+
+               return cnt;
+       }
+
+       trace_seq_reset(&iter->seq);
+       start = 0;
+
+       while (trace_empty(iter)) {
+               /*
+                * This is a make-shift waitqueue. The reason we don't use
+                * an actual wait queue is because:
+                *  1) we only ever have one waiter
+                *  2) the tracing, traces all functions, we don't want
+                *     the overhead of calling wake_up and friends
+                *     (and tracing them too)
+                *     Anyway, this is really very primitive wakeup.
+                */
+               set_current_state(TASK_INTERRUPTIBLE);
+               iter->tr->waiter = current;
+
+               /* sleep for one second, and try again. */
+               schedule_timeout(HZ);
+
+               iter->tr->waiter = NULL;
+
+               if (signal_pending(current))
+                       return -EINTR;
+
+               /*
+                * We block until we read something and tracing is disabled.
+                * We still block if tracing is disabled, but we have never
+                * read anything. This allows a user to cat this file, and
+                * then enable tracing. But after we have read something,
+                * we give an EOF when tracing is again disabled.
+                *
+                * iter->pos will be 0 if we haven't read anything.
+                */
+               if (!tracer_enabled && iter->pos)
+                       break;
+
+               continue;
+       }
+
+       /* stop when tracing is finished */
+       if (trace_empty(iter))
+               return 0;
+
+       if (cnt >= PAGE_SIZE)
+               cnt = PAGE_SIZE - 1;
+
+       memset(iter, 0, sizeof(*iter));
+       iter->tr = &global_trace;
+       iter->pos = -1;
+
+       /*
+        * We need to stop all tracing on all CPUS to read the
+        * the next buffer. This is a bit expensive, but is
+        * not done often. We fill all what we can read,
+        * and then release the locks again.
+        */
+
+       cpus_clear(mask);
+       local_irq_save(flags);
+       ftrace_save = ftrace_enabled;
+       ftrace_enabled = 0;
+       smp_wmb();
+       for_each_possible_cpu(cpu) {
+               data = iter->tr->data[cpu];
+
+               if (!head_page(data) || !data->trace_idx)
+                       continue;
+
+               atomic_inc(&data->disabled);
+               cpu_set(cpu, mask);
+       }
+
+       for_each_cpu_mask(cpu, mask) {
+               data = iter->tr->data[cpu];
+               spin_lock(&data->lock);
+       }
+
+       while (find_next_entry_inc(iter) != NULL) {
+               int len = iter->seq.len;
+
+               ret = print_trace_line(iter);
+               if (!ret) {
+                       /* don't print partial lines */
+                       iter->seq.len = len;
+                       break;
+               }
+
+               trace_consume(iter);
+
+               if (iter->seq.len >= cnt)
+                       break;
+       }
+
+       for_each_cpu_mask(cpu, mask) {
+               data = iter->tr->data[cpu];
+               spin_unlock(&data->lock);
+       }
+
+       for_each_cpu_mask(cpu, mask) {
+               data = iter->tr->data[cpu];
+               atomic_dec(&data->disabled);
+       }
+       ftrace_enabled = ftrace_save;
+       local_irq_restore(flags);
+
+       /* Now copy what we have to the user */
+       read = iter->seq.len;
+       if (read > cnt)
+               read = cnt;
+
+       ret = copy_to_user(ubuf, iter->seq.buffer, read);
+
+       if (read < iter->seq.len)
+               start = read;
+       else
+               trace_seq_reset(&iter->seq);
+
+       if (ret)
+               read = -EFAULT;
+
+       return read;
+}
+
 static struct file_operations tracing_max_lat_fops = {
-       .open = tracing_open_generic,
-       .read = tracing_max_lat_read,
-       .write = tracing_max_lat_write,
+       .open           = tracing_open_generic,
+       .read           = tracing_max_lat_read,
+       .write          = tracing_max_lat_write,
 };
 
 static struct file_operations tracing_ctrl_fops = {
-       .open = tracing_open_generic,
-       .read = tracing_ctrl_read,
-       .write = tracing_ctrl_write,
+       .open           = tracing_open_generic,
+       .read           = tracing_ctrl_read,
+       .write          = tracing_ctrl_write,
 };
 
 static struct file_operations set_tracer_fops = {
-       .open = tracing_open_generic,
-       .read = tracing_set_trace_read,
-       .write = tracing_set_trace_write,
+       .open           = tracing_open_generic,
+       .read           = tracing_set_trace_read,
+       .write          = tracing_set_trace_write,
+};
+
+static struct file_operations tracing_pipe_fops = {
+       .open           = tracing_open_pipe,
+       .read           = tracing_read_pipe,
+       .release        = tracing_release_pipe,
 };
 
 #ifdef CONFIG_DYNAMIC_FTRACE
@@ -1418,13 +2052,13 @@ tracing_read_long(struct file *filp, char __user *ubuf,
        int r;
 
        r = sprintf(buf, "%ld\n", *p);
-       return simple_read_from_buffer(ubuf, cnt, ppos,
-                                      buf, r);
+
+       return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
 }
 
 static struct file_operations tracing_read_long_fops = {
-       .open = tracing_open_generic,
-       .read = tracing_read_long,
+       .open           = tracing_open_generic,
+       .read           = tracing_read_long,
 };
 #endif
 
@@ -1448,6 +2082,11 @@ struct dentry *tracing_init_dentry(void)
        return d_tracer;
 }
 
+#ifdef CONFIG_FTRACE_SELFTEST
+/* Let selftest have access to static functions in this file */
+#include "trace_selftest.c"
+#endif
+
 static __init void tracer_init_debugfs(void)
 {
        struct dentry *d_tracer;
@@ -1497,6 +2136,16 @@ static __init void tracer_init_debugfs(void)
        if (!entry)
                pr_warning("Could not create debugfs "
                           "'tracing_threash' 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", 0644, d_tracer,
+                                   NULL, &tracing_pipe_fops);
+       if (!entry)
+               pr_warning("Could not create debugfs "
+                          "'tracing_threash' entry\n");
 
 #ifdef CONFIG_DYNAMIC_FTRACE
        entry = debugfs_create_file("dyn_ftrace_total_info", 0444, d_tracer,
@@ -1511,15 +2160,15 @@ static __init void tracer_init_debugfs(void)
 /* dummy trace to disable tracing */
 static struct tracer no_tracer __read_mostly =
 {
-       .name = "none",
+       .name           = "none",
 };
 
 static int trace_alloc_page(void)
 {
        struct trace_array_cpu *data;
-       void *array;
        struct page *page, *tmp;
        LIST_HEAD(pages);
+       void *array;
        int i;
 
        /* first allocate a page for each CPU */
@@ -1550,15 +2199,19 @@ static int trace_alloc_page(void)
        /* Now that we successfully allocate a page per CPU, add them */
        for_each_possible_cpu(i) {
                data = global_trace.data[i];
+               spin_lock_init(&data->lock);
+               lockdep_set_class(&data->lock, &data->lock_key);
                page = list_entry(pages.next, struct page, lru);
-               list_del(&page->lru);
+               list_del_init(&page->lru);
                list_add_tail(&page->lru, &data->trace_pages);
                ClearPageLRU(page);
 
 #ifdef CONFIG_TRACER_MAX_TRACE
                data = max_tr.data[i];
+               spin_lock_init(&data->lock);
+               lockdep_set_class(&data->lock, &data->lock_key);
                page = list_entry(pages.next, struct page, lru);
-               list_del(&page->lru);
+               list_del_init(&page->lru);
                list_add_tail(&page->lru, &data->trace_pages);
                SetPageLRU(page);
 #endif
@@ -1569,7 +2222,7 @@ static int trace_alloc_page(void)
 
  free_pages:
        list_for_each_entry_safe(page, tmp, &pages, lru) {
-               list_del(&page->lru);
+               list_del_init(&page->lru);
                __free_page(page);
        }
        return -ENOMEM;
@@ -1581,8 +2234,11 @@ __init static int tracer_alloc_buffers(void)
        void *array;
        struct page *page;
        int pages = 0;
+       int ret = -ENOMEM;
        int i;
 
+       global_trace.ctrl = tracer_enabled;
+
        /* Allocate the first page for all buffers */
        for_each_possible_cpu(i) {
                data = global_trace.data[i] = &per_cpu(global_trace_cpu, i);
@@ -1594,7 +2250,6 @@ __init static int tracer_alloc_buffers(void)
                               "for trace buffer!\n");
                        goto free_buffers;
                }
-               data->trace = array;
 
                /* set the array to the list */
                INIT_LIST_HEAD(&data->trace_pages);
@@ -1611,7 +2266,6 @@ __init static int tracer_alloc_buffers(void)
                               "for trace buffer!\n");
                        goto free_buffers;
                }
-               max_tr.data[i]->trace = array;
 
                INIT_LIST_HEAD(&max_tr.data[i]->trace_pages);
                page = virt_to_page(array);
@@ -1625,7 +2279,6 @@ __init static int tracer_alloc_buffers(void)
         * round up a bit.
         */
        global_trace.entries = ENTRIES_PER_PAGE;
-       max_tr.entries = global_trace.entries;
        pages++;
 
        while (global_trace.entries < trace_nr_entries) {
@@ -1633,6 +2286,7 @@ __init static int tracer_alloc_buffers(void)
                        break;
                pages++;
        }
+       max_tr.entries = global_trace.entries;
 
        pr_info("tracer: %d pages allocated for %ld",
                pages, trace_nr_entries);
@@ -1646,6 +2300,9 @@ __init static int tracer_alloc_buffers(void)
        register_tracer(&no_tracer);
        current_trace = &no_tracer;
 
+       /* All seems OK, enable tracing */
+       tracing_disabled = 0;
+
        return 0;
 
  free_buffers:
@@ -1653,28 +2310,25 @@ __init static int tracer_alloc_buffers(void)
                struct page *page, *tmp;
                struct trace_array_cpu *data = global_trace.data[i];
 
-               if (data && data->trace) {
+               if (data) {
                        list_for_each_entry_safe(page, tmp,
                                                 &data->trace_pages, lru) {
-                               list_del(&page->lru);
+                               list_del_init(&page->lru);
                                __free_page(page);
                        }
-                       data->trace = NULL;
                }
 
 #ifdef CONFIG_TRACER_MAX_TRACE
                data = max_tr.data[i];
-               if (data && data->trace) {
+               if (data) {
                        list_for_each_entry_safe(page, tmp,
                                                 &data->trace_pages, lru) {
-                               list_del(&page->lru);
+                               list_del_init(&page->lru);
                                __free_page(page);
                        }
-                       data->trace = NULL;
                }
 #endif
        }
-       return -ENOMEM;
+       return ret;
 }
-
-device_initcall(tracer_alloc_buffers);
+fs_initcall(tracer_alloc_buffers);
This page took 0.046582 seconds and 5 git commands to generate.