Re: [PATCH 09/10] perf: Remove more fastpath code

[Steven Rostedt <rostedt@goodmis.org>]

On Fri, 2010-05-21 at 11:02 +0200, Peter Zijlstra wrote:
> plain text document attachment (perf-opt-output-more.patch)
> Sanity checks cost instructions
> 
> Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
> ---
> -
> -	/*
> -	 * Check we didn't copy past our reservation window, taking the
> -	 * possible unsigned int wrap into account.
> -	 */
> -	if (WARN_ON_ONCE(((long)(handle->head - handle->offset)) < 0))
> -		return;

Are you sure you want to remove this? I mean, sanity checks are a small
cost (I have several in the ftrace ring buffer) and can catch bugs in
case something happens during development. Especially when the code is
under a lot of flux.

I even have a sanity check that Ingo asked me to add, which would detect
if a tracer (not the ring buffer, but the user of the ring buffer)
recursed on itself. I think that check detected one bug in the function
tracer in the begging, but hasn't caught anything since.

-- Steve

> -
>  	do {
> -		unsigned long size = min(handle->size, len);
> +		unsigned long size = min_t(unsigned long, handle->size, len);
>  
>  		memcpy(handle->addr, buf, size);
>  
> @@ -3054,15 +3045,12 @@ int perf_output_begin(struct perf_output
>  			goto fail;
>  	} while (local_cmpxchg(&data->head, offset, head) != offset);
>  
> -	handle->offset	= offset;
> -	handle->head	= head;
> -
>  	if (head - local_read(&data->wakeup) > data->watermark)
>  		local_add(data->watermark, &data->wakeup);
>  
> -	handle->page = handle->offset >> (PAGE_SHIFT + page_order(data));
> +	handle->page = offset >> (PAGE_SHIFT + page_order(data));
>  	handle->page &= data->nr_pages - 1;
> -	handle->size = handle->offset & ((PAGE_SIZE << page_order(data)) - 1);
> +	handle->size = offset & ((PAGE_SIZE << page_order(data)) - 1);
>  	handle->addr = data->data_pages[handle->page];
>  	handle->addr += handle->size;
>  	handle->size = (PAGE_SIZE << page_order(data)) - handle->size;
> 
>