From: Danny Smith Subject: Re: [PATCH] ASoC: sigmadsp: safeload should be used from 4 bytes Date: Wed, 22 Aug 2018 15:40:22 +0200 Message-ID: References: <20180822061131.13191-1-robertr@axis.com> Mime-Version: 1.0 Content-Type: text/plain; charset="us-ascii"; Format="flowed" Content-Transfer-Encoding: 7bit Return-path: Received: from bastet.se.axis.com (bastet.se.axis.com [195.60.68.11]) by alsa0.perex.cz (Postfix) with ESMTP id 6C555267663 for ; Wed, 22 Aug 2018 15:40:38 +0200 (CEST) In-Reply-To: Content-Language: en-US List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Errors-To: alsa-devel-bounces@alsa-project.org Sender: alsa-devel-bounces@alsa-project.org To: Lars-Peter Clausen , Robert Rosengren , Liam Girdwood , Mark Brown Cc: alsa-devel@alsa-project.org List-Id: alsa-devel@alsa-project.org On 08/22/2018 03:05 PM, Lars-Peter Clausen wrote: > On 08/22/2018 09:53 AM, Danny Smith wrote: >> >> On 08/22/2018 08:34 AM, Lars-Peter Clausen wrote: >>> On 08/22/2018 08:11 AM, Robert Rosengren wrote: >>>> From: Danny Smith >>>> >>>> Fixed range in safeload conditional to allow safeload to be used from >>>> 4 bytes. >>> Hi, >>> >>> Thanks for the patch. The reason why 4 bytes is excluded is that up to 4 >>> bytes can be updated atomically with a single register write. But I could >>> see that if the firmware reads the same parameter multiple times during the >>> same run you could get inconsistent results. >>> >>> Can you explain a bit more why you need this? >>> >>> If we want to allow safeload for 4 byte parameters the same reasoning >>> applies for parameters with less than 4 bytes as well and so the check >>> should be removed completely. >>> >>> - Lars >> Hi, >> >> Thanks for the feedback. It is correct that up to 4 bytes can be updated >> atomically but does that also guarantee that the data is safeloaded, i.e. >> updated when the parameter is not in use? Maybe updating up to 4 byte >> parameters without safeload does not cause audio glitches? > The datasheet says: > > """ > To update parameters in real time while avoiding pop and click > noises on the output, the ADAU1761 uses a software safeload > mechanism. The software safeload mechanism enables the > SigmaDSP core to load new parameters into RAM while guaranteeing > that the parameters are not in use. This prevents an > undesirable condition where an instruction could execute with > a mix of old and new parameters. > """ > > The way I understand the last sentence is that there is no issue when > loading only 4 bytes or less since there is no risk of mixing old and new data. > > I don't mind changing this, I was just curious if you had seen any issues > with a 4 byte parameter update. > > - Lars I have not seen any issues doing a 4byte parameter update without safeload but have not performed any extensive testing on the subject. The reason for the patch was that I noticed that sigmastudio uses safeload when doing 4byte writes but I cannot say if it is necessary or not. Regards, Danny