Forgot to add disclaimer... These are just my opinion based on the discussions I had on the mailing list in the past on similar topics... Murali Karicheri Software Design Engineer Texas Instruments Inc. Germantown, MD 20874 email: m-karicheri2@xxxxxx >-----Original Message----- >From: linux-media-owner@xxxxxxxxxxxxxxx [mailto:linux-media- >owner@xxxxxxxxxxxxxxx] On Behalf Of Karicheri, Muralidharan >Sent: Monday, September 28, 2009 11:25 AM >To: Yu, Jinlu; Mauro Carvalho Chehab >Cc: linux-media@xxxxxxxxxxxxxxx >Subject: RE: [PATCH 0/5] V4L2 patches for Intel Moorestown Camera Imaging > >Hi, > >First of all, based on Media controller proposal, you would be better >off by developing ISP also as a sub device. This way you will be able >to make connection as below... > >sensor -> isp -> video node > >Please check out the bus parameter settings RFC from Hans. Some of these >parameters are set using board/platform specific configuration and there >is a new API proposed to do this. Following are candidate for this. > >>bus_width; width of the buss connecting sensor and ISP. >>hpol; horizontal sync polarity >>vpol; vertical sync polarity >>edge; sampling edge > > >>field_sel; field selection, even or odd. >I am assuming this will be reported by the ISP based on what it receives >from sensor for interlaced scan. If not, probably you need to add this to >sub dev interface as a private ops if sub device can report it. > >>ycseq; YCbCr sequence, YCbCr or YCrCb or CbYCrY or CrYCbY > >This is also being discussed currently through another related RFC from SoC >camera side. Assuming sensor outputs YCbCr and ISP is capable of changing >the sequence (which is the case in DM355/DM6446/DM365 VPFE), then these are >enumerated at the video node for ENUM_FMT as different frame formats. > >>conv422; subsampling type, co-sited 4:4:4 or non-cosited 4:4:4 or color >>interpolation >Again assuming these are at the ISP side, you need to use driver specific >IOCTLs to do this. Converting ISP to sub device is the way to go based on >the Media controller proposal that allows you to have these ioctls setting >it in the sub device directly. > >>bpat; bayer sampling sequence, RGRG GBGB or GRGR BGBG or ... > >This is something to be discussed in the bus parameter RFC. Camera can >output different patterns. So ISP needs to know what pattern is output >by the sensor and bridge device probably needs to read it from the sensor >and set it in the ISP. data format setting is also discussed in the above >RFC or in a separate RFC. Please check it out. > > >Murali Karicheri >Software Design Engineer >Texas Instruments Inc. >Germantown, MD 20874 >email: m-karicheri2@xxxxxx > >>-----Original Message----- >>From: linux-media-owner@xxxxxxxxxxxxxxx [mailto:linux-media- >>owner@xxxxxxxxxxxxxxx] On Behalf Of Yu, Jinlu >>Sent: Monday, September 28, 2009 5:26 AM >>To: Mauro Carvalho Chehab >>Cc: linux-media@xxxxxxxxxxxxxxx >>Subject: RE: [PATCH 0/5] V4L2 patches for Intel Moorestown Camera Imaging >> >>We have a solution to this which is to make them as sensor's private data, >>and let ISP access them by share the same structure definition. >> >>But I don't think it's a good idea, because it is so driver-specific and >>can not be used by others. >> >>Best Regards >>Jinlu Yu >>UMG UPSG PRC >>INET: 8758 1603 >>TEL: 86 10 8217 1603 >>FAX: 86 10 8286 1400 >>-----Original Message----- >>From: linux-media-owner@xxxxxxxxxxxxxxx [mailto:linux-media- >>owner@xxxxxxxxxxxxxxx] On Behalf Of Yu, Jinlu >>Sent: 2009年9月27日 22:31 >>To: Mauro Carvalho Chehab >>Cc: linux-media@xxxxxxxxxxxxxxx >>Subject: RE: [PATCH 0/5] V4L2 patches for Intel Moorestown Camera Imaging >> >>Hi, Mauro >> >>Thank you for your suggestion on this. >> >>Now I have another problem. The ISP needs the following parameters of the >>sensor to set the acquisition interface, but I can not find a suitable >>subdev ioctls to get them from sensor driver. >> >>bus_width; width of the buss connecting sensor and ISP. >>field_sel; field selection, even or odd. >>ycseq; YCbCr sequence, YCbCr or YCrCb or CbYCrY or CrYCbY >>conv422; subsampling type, co-sited 4:4:4 or non-cosited 4:4:4 or color >>interpolation >>bpat; bayer sampling sequence, RGRG GBGB or GRGR BGBG or ... >>hpol; horizontal sync polarity >>vpol; vertical sync polarity >>edge; sampling edge >> >>Best Regards >>Jinlu Yu >>UMG UPSG PRC >>INET: 8758 1603 >>TEL: 86 10 8217 1603 >>FAX: 86 10 8286 1400 >>-----Original Message----- >>From: Mauro Carvalho Chehab [mailto:mchehab@xxxxxxxxxxxxx] >>Sent: 2009年9月24日 19:45 >>To: Yu, Jinlu >>Cc: linux-media@xxxxxxxxxxxxxxx >>Subject: Re: [PATCH 0/5] V4L2 patches for Intel Moorestown Camera Imaging >> >>Em Thu, 24 Sep 2009 19:21:40 +0800 >>"Yu, Jinlu" <jinlu.yu@xxxxxxxxx> escreveu: >> >>> Hi, Hans/Guennadi >>> >>> I am modifying these drivers to comply with v4l2 framework. I have >>finished replacing our buffer managing code with utility function from >>videobuf-core.c and videobuf-dma-contig.c. Now I am working on the subdev. >>One thing I am sure is that each sensor should be registered as a >>v4l2_subdev and ISP (Image Signal Processor) is registered as a >v4l2_device >>acting as the bridge device. >>> >>> But we have two ways to deal with the relationship of sensor and ISP, >and >>we don't know which one is better. Could you help me on this? >>> >>> No.1. Register the ISP as a video_device (/dev/video0) and treat each of >>the sensor (SOC and RAW) as an input of the ISP. If I want to change the >>sensor, use the VIDIOC_S_INPUT to change input from sensor A to sensor B. >>But I have a concern about this ioctl. Since I didn't find any code >related >>HW pipeline status checking and HW register setting in the implement of >>this ioctl (e.g. vino_s_input in /drivers/media/video/vino.c). So don't I >>have to stream-off the HW pipeline and change the HW register setting for >>the new input? Or is it application's responsibility to stream-off the >>pipeline and renegotiate the parameters for the new input? >>> >>> No.2. Combine the SOC sensor together with the ISP as Channel One and >>register it as /dev/video0, and combine the RAW sensor together with the >>ISP as Channel Two and register it as /dev/video1. Surely, only one >channel >>works at a certain time due to HW restriction. When I want to change the >>sensor (e.g. from SOC sensor to RAW sensor), just close /dev/video0 and >>open /dev/video1. >> >>The better seems to be No. 1. As you need to re-negotiate parameters for >>switching from one sensor to another, if some app tries to change from one >>input to another while streaming, you should just return -EBUSY, if it is >>not >>possible to switch (for example, if the selected format/resolution/frame >>rate >>is incompatible). >> >> >> >>Cheers, >>Mauro >>N嫥叉靣笡y氊b瞂千v豝?藓{.n?壏{睓鏱j)韰骅w*jg?秹殠娸/侁鋤罐枈?娹櫒璀?? >>摺玜囤 瓽珴閔?鎗:+v墾妛鑶佶 >>N�����r��y���b�X��ǧv�^�){.n�+����{���bj)���w* >>jg��� �����ݢj/���z�ޖ��2�ޙ���&�)ߡ�a���� �G���h��j:+v���w�٥ >N�����r��y���b�X��ǧv�^�){.n�+����{���bj)���w* >jg��� �����ݢj/���z�ޖ��2�ޙ���&�)ߡ�a���� �G���h��j:+v���w�٥ ��.n��������+%������w��{.n�����{��g����^n�r������&��z�ޗ�zf���h���~����������_��+v���)ߣ�m
