Re: [PATCH] pcf50633: revise locking for ADC

[Samuel Ortiz <sameo@linux.intel.com>]

Hi Paul,

On Tue, Jul 28, 2009 at 12:58:48AM +0400, Paul Fertser wrote:
> Current implementation is prone to races, this patch attempts to remove all
> but one (in pcf50633_adc_sync_read).
> 
> The idea is that we need to guard the queue access only on inserting and
> removing items. If we insert and there're no more items in the queue it
> means that the last irq already happened and we need to trigger ADC
> manually. If not, then the next conversion will be triggered by the irq
> handler upon completion of the previous.


> @@ -136,9 +133,13 @@ int pcf50633_adc_sync_read(struct pcf50633 *pcf, int mux, int avg)
>  	req->callback_param = req;
>  
>  	init_completion(&req->completion);
> -	adc_enqueue_request(pcf, req);
> +	err = adc_enqueue_request(pcf, req);
> +	if (err)
> +		return err;
> +
>  	wait_for_completion(&req->completion);
>  
> +	/* FIXME by this time req might be already freed */
In fact, this is problematic.
Shouldn't the request be freed by the callback (in the async request), or by
sync_read() ?

Your patch looks fine though, I applied it to my for-next branch.

Cheers,
Samuel.

-- 
Intel Open Source Technology Centre
http://oss.intel.com/