[PATCH] ring-buffer: mark racy accesses on work->wait_index

[PATCH] ring-buffer: mark racy accesses on work->wait_index

[linke li]

Mark data races to work->wait_index as benign using READ_ONCE and WRITE_ONCE. These accesses are expected to be racy.

Signed-off-by: linke li <lilinke99@qq.com>
---
 kernel/trace/ring_buffer.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/kernel/trace/ring_buffer.c b/kernel/trace/ring_buffer.c
index 0699027b4f4c..a47e9e9750cc 100644
--- a/kernel/trace/ring_buffer.c
+++ b/kernel/trace/ring_buffer.c
@@ -798,7 +798,7 @@ void ring_buffer_wake_waiters(struct trace_buffer *buffer, int cpu)
 		rbwork = &cpu_buffer->irq_work;
 	}
 
-	rbwork->wait_index++;
+	WRITE_ONCE(rbwork->wait_index, READ_ONCE(rbwork->wait_index) + 1);
 	/* make sure the waiters see the new index */
 	smp_wmb();
 
@@ -906,7 +906,7 @@ int ring_buffer_wait(struct trace_buffer *buffer, int cpu, int full)
 
 		/* Make sure to see the new wait index */
 		smp_rmb();
-		if (wait_index != work->wait_index)
+		if (wait_index != READ_ONCE(work->wait_index))
 			break;
 	}
 
-- 
2.39.3 (Apple Git-145)

Re: [PATCH] ring-buffer: mark racy accesses on work->wait_index

[Steven Rostedt]

On Wed,  6 Mar 2024 10:55:34 +0800
linke li <lilinke99@qq.com> wrote:

> Mark data races to work->wait_index as benign using READ_ONCE and WRITE_ONCE. These accesses are expected to be racy.

Are we now to the point that every single access of a variable (long size
or less) needs a READ_ONCE/WRITE_ONCE even with all the necessary smp_r/wmb()s?


> 
> Signed-off-by: linke li <lilinke99@qq.com>
> ---
>  kernel/trace/ring_buffer.c | 4 ++--
>  1 file changed, 2 insertions(+), 2 deletions(-)
> 
> diff --git a/kernel/trace/ring_buffer.c b/kernel/trace/ring_buffer.c
> index 0699027b4f4c..a47e9e9750cc 100644
> --- a/kernel/trace/ring_buffer.c
> +++ b/kernel/trace/ring_buffer.c
> @@ -798,7 +798,7 @@ void ring_buffer_wake_waiters(struct trace_buffer *buffer, int cpu)
>  		rbwork = &cpu_buffer->irq_work;
>  	}
>  
> -	rbwork->wait_index++;
> +	WRITE_ONCE(rbwork->wait_index, READ_ONCE(rbwork->wait_index) + 1);

I mean the above is really ugly. If this is the new thing to do, we need
better macros.

If anything, just convert it to an atomic_t.

-- Steve


>  	/* make sure the waiters see the new index */
>  	smp_wmb();
>  
> @@ -906,7 +906,7 @@ int ring_buffer_wait(struct trace_buffer *buffer, int cpu, int full)
>  
>  		/* Make sure to see the new wait index */
>  		smp_rmb();
> -		if (wait_index != work->wait_index)
> +		if (wait_index != READ_ONCE(work->wait_index))
>  			break;
>  	}
>