[PATCH] ringbuffer: synchronize for resizing

[PATCH] ringbuffer: synchronize for resizing

[Lai Jiangshan]

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>
---
diff --git a/kernel/trace/ring_buffer.c b/kernel/trace/ring_buffer.c
index 63446f1..db223fe 100644
--- a/kernel/trace/ring_buffer.c
+++ b/kernel/trace/ring_buffer.c
@@ -1196,6 +1196,7 @@ 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++) {
@@ -1210,6 +1211,7 @@ 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);
 

Re: [PATCH] ringbuffer: synchronize for resizing

[Steven Rostedt]

On Tue, 2009-11-03 at 19:42 +0800, Lai Jiangshan wrote:
> 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 ...)

Thanks! I'll push this out for inclusion right away.

-- Steve

> 
> Reported-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
> Signed-off-by: Lai Jiangshan <laijs@cn.fujitsu.com>
> ---
> diff --git a/kernel/trace/ring_buffer.c b/kernel/trace/ring_buffer.c
> index 63446f1..db223fe 100644
> --- a/kernel/trace/ring_buffer.c
> +++ b/kernel/trace/ring_buffer.c
> @@ -1196,6 +1196,7 @@ 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++) {
> @@ -1210,6 +1211,7 @@ 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);
>  
> 
> 
> 

[tip:tracing/urgent] ring-buffer: Synchronize resizing buffer with reader lock

[tip-bot for Lai Jiangshan]

Commit-ID:  f7112949f6a4cd6883d66c882d568c2197321de6
Gitweb:     http://git.kernel.org/tip/f7112949f6a4cd6883d66c882d568c2197321de6
Author:     Lai Jiangshan <laijs@cn.fujitsu.com>
AuthorDate: Tue, 3 Nov 2009 19:42:45 +0800
Committer:  Steven Rostedt <rostedt@goodmis.org>
CommitDate: Wed, 4 Nov 2009 00:04:20 -0500

ring-buffer: Synchronize resizing buffer with reader lock

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 |    2 ++
 1 files changed, 2 insertions(+), 0 deletions(-)

diff --git a/kernel/trace/ring_buffer.c b/kernel/trace/ring_buffer.c
index 3ffa502..5dd017f 100644
--- a/kernel/trace/ring_buffer.c
+++ b/kernel/trace/ring_buffer.c
@@ -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);