From mboxrd@z Thu Jan 1 00:00:00 1970 Return-path: Received: from perceval.ideasonboard.com ([95.142.166.194]:35654 "EHLO perceval.ideasonboard.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1754731AbaBUNSO (ORCPT ); Fri, 21 Feb 2014 08:18:14 -0500 From: Laurent Pinchart To: Sakari Ailus Cc: Hans Verkuil , linux-media@vger.kernel.org, k.debski@samsung.com Subject: Re: [PATCH v5.1 3/7] v4l: Add timestamp source flags, mask and document them Date: Fri, 21 Feb 2014 14:19:26 +0100 Message-ID: <3197008.QzfCa0jNuY@avalon> In-Reply-To: <20140221130448.GC15635@valkosipuli.retiisi.org.uk> References: <20140217232931.GW15635@valkosipuli.retiisi.org.uk> <1627498.vqEXNhvDry@avalon> <20140221130448.GC15635@valkosipuli.retiisi.org.uk> MIME-Version: 1.0 Content-Transfer-Encoding: 7Bit Content-Type: text/plain; charset="us-ascii" Sender: linux-media-owner@vger.kernel.org List-ID: Hi Sakari, On Friday 21 February 2014 15:04:48 Sakari Ailus wrote: > Hi Laurent, > > On Fri, Feb 21, 2014 at 12:58:58PM +0100, Laurent Pinchart wrote: > ... > > > > It's possible to calculate it (decrementing the readout time + exposure > > > time from the end of frame timestamp) and that's what the devices > > > supposedly do. The pre-frame exposure time isn't available to the host, > > > so the end of frame timestamp cannot be calculated by the host from the > > > camera generated timestamp. > > > > > > However the link to the host is USB which has a lot more latency than > > > almost anything else which makes even hardware generated timestamps a > > > little imprecise. > > > > Why so ? There will be a jitter in frame arrival, but the hardware > > timestamp should be accurate (at least if properly generated by the > > camera firmware). > > Yes, the hardware timestamp should be accurate on its own, but as there's > delay and jitter converting that into something that's relevant on the host > adds some uncertainty. AFAIR the accuracy of the camera generated timestamp > was still much better than that of the driver generated one, right? Yes it is, as the conversion mechanism is stats-based and takes jitter and delays into account. -- Regards, Laurent Pinchart