ring-buffer: Synchronize resizing buffer with reader lock
authorLai Jiangshan <laijs@cn.fujitsu.com>
Tue, 3 Nov 2009 11:42:45 +0000 (19:42 +0800)
committerSteven Rostedt <rostedt@goodmis.org>
Wed, 4 Nov 2009 05:04:20 +0000 (00:04 -0500)
We got a sudden panic when we reduced the size of the
ringbuffer.

We can reproduce the panic by the following steps:

echo 1 > events/sched/enable
cat trace_pipe > /dev/null &

while ((1))
do
echo 12000 > buffer_size_kb
echo 512 > buffer_size_kb
done

(not more than 5 seconds, panic ...)

Reported-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Signed-off-by: Lai Jiangshan <laijs@cn.fujitsu.com>
LKML-Reference: <4AF01735.9060409@cn.fujitsu.com>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
kernel/trace/ring_buffer.c

index 3ffa502fb2433352cb986a6e6821cf203c63652d..5dd017fea6f58bf8cbb3824770c99ba6f83fa4f2 100644 (file)
@@ -1193,6 +1193,7 @@ rb_remove_pages(struct ring_buffer_per_cpu *cpu_buffer, unsigned nr_pages)
        atomic_inc(&cpu_buffer->record_disabled);
        synchronize_sched();
 
+       spin_lock_irq(&cpu_buffer->reader_lock);
        rb_head_page_deactivate(cpu_buffer);
 
        for (i = 0; i < nr_pages; i++) {
@@ -1207,6 +1208,7 @@ rb_remove_pages(struct ring_buffer_per_cpu *cpu_buffer, unsigned nr_pages)
                return;
 
        rb_reset_cpu(cpu_buffer);
+       spin_unlock_irq(&cpu_buffer->reader_lock);
 
        rb_check_pages(cpu_buffer);