From mboxrd@z Thu Jan 1 00:00:00 1970 Return-path: Received: from mx1.redhat.com ([209.132.183.28]:23868 "EHLO mx1.redhat.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1751398Ab2G1Kld (ORCPT ); Sat, 28 Jul 2012 06:41:33 -0400 Message-ID: <5013C30B.7030600@redhat.com> Date: Sat, 28 Jul 2012 12:46:35 +0200 From: Hans de Goede MIME-Version: 1.0 To: abel@uni-bielefeld.de CC: "Ivan T. Ivanov" , Sergio Aguirre , linux-media@vger.kernel.org, Laurent Pinchart , Atsuo Kuwahara , Stefan Herbrechtsmeier Subject: Re: Advice on extending libv4l for media controller support References: <1336662597.15542.15.camel@iivanov-desktop> <5011AD6A.9040609@uni-bielefeld.de> In-Reply-To: <5011AD6A.9040609@uni-bielefeld.de> Content-Type: text/plain; charset=UTF-8; format=flowed Content-Transfer-Encoding: 7bit Sender: linux-media-owner@vger.kernel.org List-ID: Hi, On 07/26/2012 10:49 PM, Robert Abel wrote: > Hi, > > Sorry to be late to the party... I wanted to follow up on this discussion, but forgot and haven't read anything about it since... > > On 10.05.2012 17:09, Ivan T. Ivanov wrote: >> On Wed, May 9, 2012 at 7:08 PM, Sergio Aguirre >> wrote: >>> I want to create some sort of plugin with specific media >>> controller configurations, >>> to avoid userspace to worry about component names and specific >>> usecases (use sensor resizer, or SoC ISP resizer, etc.). >> Probably following links can help you. They have been tested >> with the OMAP3 ISP. >> >> Regards, >> iivanov >> >> [1]http://www.spinics.net/lists/linux-media/msg31901.html >> [2] >> http://comments.gmane.org/gmane.linux.drivers.video-input-infrastructure/32704 > > I recently extended Yordan Kamenov's libv4l-mcplugin to support multiple trees per device with extended configurations (-stolen from- inspired by media-ctl) not tied to specific device nodes (but to device names instead). > > I uploaded the patches here (16kB). Basically, I used Yordan's patches as a base and worked from there to fix up his source code and Makefile for cross-compiling using OpenEmbedded/Yocto. > > There are a ton of minor issues with this, starting with the fact that I did not put proper copyright notices in any of these files. Please advise if this poses a problem. > Only integral frame size support and no support for native read() calls. There's a dummy read() function, because for some reason this is required in libv4l2 0.9.0-test though it's not mentioned anywhere. As the original plug-in by Yordan, there is currently no cleaning-up of the internal data structures. > > I used this in conjunction with the Gumstix CASPA FS (MT9V032) camera using some of Laurent's patches and some custom patches which add ENUM_FMT support to the driver. > > Basically, upon opening a given device, all trees are configured once to load the respective end-point's formats for emulation of setting and getting formats. Then regular format negotiation by the user application takes place. As discussed higher up in this thread, since the initial libv4l-mcplugin was done for the omap3, we've had several meetings on the topic of libv4l and media-controller using devices and we came to the following conclusions: 1) The existing mediactl lib would be extended with a libmediactlvideo lib, which would be able to control media-ctrl video chains, ie it can: -give a list of possibly supported formats / sizes / framerates -setup the chain to deliver a requested format Since the optimal setup will be hardware specific the idea was to give this libs per soc plugins, and a generic plugin for simple socs / as fallback. 2) A cmdline utility to set up a chain using libmediactlvideo, so that things can be tested using raw devices, ie without libv4l2 coming into play, just like apps like v4l2-ctl allow low level control mostly for testing purposes 3) There would then be a libv4l2 plugin much like the above linked omap3 plugin, but then generic for any mediactl using video devices, which would use libmediactlvideo to do the work of setting up the chain (and which will fail to init when the to be opened device is not part of a mediactl controlled chain). And AFAIK some work was done in this direction. Sakari? Laurent? Eitherway it is about time someone started working on this, and I would greatly prefer the above plan to be implemented. Once we have this in place, then we can do a new v4l-utils release which officially supports the plugin API (which currently only lives in master, not in any releases). Regards, Hans