ring-buffer: add reader lock
authorSteven Rostedt <srostedt@redhat.com>
Tue, 11 Nov 2008 17:47:44 +0000 (18:47 +0100)
committerIngo Molnar <mingo@elte.hu>
Tue, 11 Nov 2008 17:47:44 +0000 (18:47 +0100)
Impact: serialize reader accesses to individual CPU ring buffers

The code in the ring buffer expects only one reader at a time, but currently
it puts that requirement on the caller. This is not strong enough, and this
patch adds a "reader_lock" that serializes the access to the reader API
of the ring buffer.

Signed-off-by: Steven Rostedt <srostedt@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
kernel/trace/ring_buffer.c

index a6b8f9d7ac96c95902c6f189de2cf47e829ca126..17c2ccebb567ead0f2b90a1856af9d105346fcef 100644 (file)
@@ -154,6 +154,7 @@ static inline int test_time_stamp(u64 delta)
 struct ring_buffer_per_cpu {
        int                             cpu;
        struct ring_buffer              *buffer;
+       spinlock_t                      reader_lock; /* serialize readers */
        raw_spinlock_t                  lock;
        struct lock_class_key           lock_key;
        struct list_head                pages;
@@ -321,6 +322,7 @@ rb_allocate_cpu_buffer(struct ring_buffer *buffer, int cpu)
 
        cpu_buffer->cpu = cpu;
        cpu_buffer->buffer = buffer;
+       spin_lock_init(&cpu_buffer->reader_lock);
        cpu_buffer->lock = (raw_spinlock_t)__RAW_SPIN_LOCK_UNLOCKED;
        INIT_LIST_HEAD(&cpu_buffer->pages);
 
@@ -1476,6 +1478,9 @@ unsigned long ring_buffer_overruns(struct ring_buffer *buffer)
 void ring_buffer_iter_reset(struct ring_buffer_iter *iter)
 {
        struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer;
+       unsigned long flags;
+
+       spin_lock_irqsave(&cpu_buffer->reader_lock, flags);
 
        /* Iterator usage is expected to have record disabled */
        if (list_empty(&cpu_buffer->reader_page->list)) {
@@ -1489,6 +1494,8 @@ void ring_buffer_iter_reset(struct ring_buffer_iter *iter)
                iter->read_stamp = cpu_buffer->read_stamp;
        else
                iter->read_stamp = iter->head_page->time_stamp;
+
+       spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags);
 }
 
 /**
@@ -1707,17 +1714,8 @@ static void rb_advance_iter(struct ring_buffer_iter *iter)
                rb_advance_iter(iter);
 }
 
-/**
- * ring_buffer_peek - peek at the next event to be read
- * @buffer: The ring buffer to read
- * @cpu: The cpu to peak at
- * @ts: The timestamp counter of this event.
- *
- * This will return the event that will be read next, but does
- * not consume the data.
- */
-struct ring_buffer_event *
-ring_buffer_peek(struct ring_buffer *buffer, int cpu, u64 *ts)
+static struct ring_buffer_event *
+rb_buffer_peek(struct ring_buffer *buffer, int cpu, u64 *ts)
 {
        struct ring_buffer_per_cpu *cpu_buffer;
        struct ring_buffer_event *event;
@@ -1779,16 +1777,8 @@ ring_buffer_peek(struct ring_buffer *buffer, int cpu, u64 *ts)
        return NULL;
 }
 
-/**
- * ring_buffer_iter_peek - peek at the next event to be read
- * @iter: The ring buffer iterator
- * @ts: The timestamp counter of this event.
- *
- * This will return the event that will be read next, but does
- * not increment the iterator.
- */
-struct ring_buffer_event *
-ring_buffer_iter_peek(struct ring_buffer_iter *iter, u64 *ts)
+static struct ring_buffer_event *
+rb_iter_peek(struct ring_buffer_iter *iter, u64 *ts)
 {
        struct ring_buffer *buffer;
        struct ring_buffer_per_cpu *cpu_buffer;
@@ -1849,6 +1839,51 @@ ring_buffer_iter_peek(struct ring_buffer_iter *iter, u64 *ts)
        return NULL;
 }
 
+/**
+ * ring_buffer_peek - peek at the next event to be read
+ * @buffer: The ring buffer to read
+ * @cpu: The cpu to peak at
+ * @ts: The timestamp counter of this event.
+ *
+ * This will return the event that will be read next, but does
+ * not consume the data.
+ */
+struct ring_buffer_event *
+ring_buffer_peek(struct ring_buffer *buffer, int cpu, u64 *ts)
+{
+       struct ring_buffer_per_cpu *cpu_buffer = buffer->buffers[cpu];
+       struct ring_buffer_event *event;
+       unsigned long flags;
+
+       spin_lock_irqsave(&cpu_buffer->reader_lock, flags);
+       event = rb_buffer_peek(buffer, cpu, ts);
+       spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags);
+
+       return event;
+}
+
+/**
+ * ring_buffer_iter_peek - peek at the next event to be read
+ * @iter: The ring buffer iterator
+ * @ts: The timestamp counter of this event.
+ *
+ * This will return the event that will be read next, but does
+ * not increment the iterator.
+ */
+struct ring_buffer_event *
+ring_buffer_iter_peek(struct ring_buffer_iter *iter, u64 *ts)
+{
+       struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer;
+       struct ring_buffer_event *event;
+       unsigned long flags;
+
+       spin_lock_irqsave(&cpu_buffer->reader_lock, flags);
+       event = rb_iter_peek(iter, ts);
+       spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags);
+
+       return event;
+}
+
 /**
  * ring_buffer_consume - return an event and consume it
  * @buffer: The ring buffer to get the next event from
@@ -1860,19 +1895,24 @@ ring_buffer_iter_peek(struct ring_buffer_iter *iter, u64 *ts)
 struct ring_buffer_event *
 ring_buffer_consume(struct ring_buffer *buffer, int cpu, u64 *ts)
 {
-       struct ring_buffer_per_cpu *cpu_buffer;
+       struct ring_buffer_per_cpu *cpu_buffer = buffer->buffers[cpu];
        struct ring_buffer_event *event;
+       unsigned long flags;
 
        if (!cpu_isset(cpu, buffer->cpumask))
                return NULL;
 
-       event = ring_buffer_peek(buffer, cpu, ts);
+       spin_lock_irqsave(&cpu_buffer->reader_lock, flags);
+
+       event = rb_buffer_peek(buffer, cpu, ts);
        if (!event)
-               return NULL;
+               goto out;
 
-       cpu_buffer = buffer->buffers[cpu];
        rb_advance_reader(cpu_buffer);
 
+ out:
+       spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags);
+
        return event;
 }
 
@@ -1909,11 +1949,11 @@ ring_buffer_read_start(struct ring_buffer *buffer, int cpu)
        atomic_inc(&cpu_buffer->record_disabled);
        synchronize_sched();
 
-       local_irq_save(flags);
+       spin_lock_irqsave(&cpu_buffer->reader_lock, flags);
        __raw_spin_lock(&cpu_buffer->lock);
        ring_buffer_iter_reset(iter);
        __raw_spin_unlock(&cpu_buffer->lock);
-       local_irq_restore(flags);
+       spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags);
 
        return iter;
 }
@@ -1945,12 +1985,17 @@ struct ring_buffer_event *
 ring_buffer_read(struct ring_buffer_iter *iter, u64 *ts)
 {
        struct ring_buffer_event *event;
+       struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer;
+       unsigned long flags;
 
-       event = ring_buffer_iter_peek(iter, ts);
+       spin_lock_irqsave(&cpu_buffer->reader_lock, flags);
+       event = rb_iter_peek(iter, ts);
        if (!event)
-               return NULL;
+               goto out;
 
        rb_advance_iter(iter);
+ out:
+       spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags);
 
        return event;
 }
@@ -1999,13 +2044,15 @@ void ring_buffer_reset_cpu(struct ring_buffer *buffer, int cpu)
        if (!cpu_isset(cpu, buffer->cpumask))
                return;
 
-       local_irq_save(flags);
+       spin_lock_irqsave(&cpu_buffer->reader_lock, flags);
+
        __raw_spin_lock(&cpu_buffer->lock);
 
        rb_reset_cpu(cpu_buffer);
 
        __raw_spin_unlock(&cpu_buffer->lock);
-       local_irq_restore(flags);
+
+       spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags);
 }
 
 /**