Em 01-03-2011 11:38, Hans Verkuil escreveu: > Hi Mauro, > > On Tuesday, March 01, 2011 13:28:35 Mauro Carvalho Chehab wrote: >> Hi Martin, >> >> Em 01-03-2011 06:59, Martin Bugge (marbugge) escreveu: >>> Author: Martin Bugge <marbugge@xxxxxxxxx> >>> Date: Tue, 1 March 2010 >>> ====================== >>> >>> This is a proposal for adding a Consumer Electronic Control (CEC) API to > V4L2. >>> This document describes the changes and new ioctls needed. >>> >>> Version 1.0 (This is first version) >>> >>> Background >>> ========== >>> CEC is a protocol that provides high-level control functions between > various audiovisual products. >>> It is an optional supplement to the High-Definition Multimedia Interface > Specification (HDMI). >>> Physical layer is a one-wire bidirectional serial bus that uses the > industry-standard AV.link protocol. >>> >>> In short: CEC uses pin 13 on the HDMI connector to transmit and receive > small data-packets >>> (maximum 16 bytes including a 1 byte header) at low data rates > (~400 bits/s). >>> >>> A CEC device may have any of 15 logical addresses (0 - 14). >>> (address 15 is broadcast and some addresses are reserved) >>> >>> >>> References >>> ========== >>> [1] High-Definition Multimedia Interface Specification version 1.3a, >>> Supplement 1 Consumer Electronic Control (CEC). >>> http://www.hdmi.org/manufacturer/specification.aspx >>> >>> [2] > http://www.hdmi.org/pdf/whitepaper/DesigningCECintoYourNextHDMIProduct.pdf >>> >>> >>> Proposed solution >>> ================= >>> >>> Two new ioctls: >>> VIDIOC_CEC_CAP (read) >>> VIDIOC_CEC_CMD (read/write) >> >> How this proposal will interact with RC core? The way I see it, HDMI-CEC is > just a way to get/send >> Remote Controller data, and should be interacting with the proper Kernel > subsystems, e. g., >> with Remote Controller and input/event subsystems. > > I knew you were going to mention this :-) > > Actually, while CEC does support IR commands, this is only a very small part > of the standard. Routing IR commands to the IR core is possible to do, > although it is not in this initial version. Should this be needed, then a flag > can be created that tells V4L to route IR commands to the IR core. > > This should be optional, though, because if you are a repeater you do not want > to pass such IR commands to the IR core, instead you want to retransmit them > to a CEC output. > >> >> I don't think we need two ioctls for that, as RC capabilities are already > exported via >> sysfs, and we have two interfaces already for receiving events (input/event > and lirc). >> For sending, lirc interface might be used, but it is currently focused only > on sending >> raw pulse/space sequences. So, we'll need to add some capability there for > IR/CEC TX. >> I had a few discussions about that with Jarod, but we didn't write yet an > interface for it. > > Again, CEC != IR. All you need is a simple API to be able to send and receive > CEC packets and a libcec that you can use to do the topology discovery and > send/receive the commands. You don't want nor need that in the kernel. > > The only place where routing things to the IR core is useful is when someone > points a remote at a TV (for example), which then passes it over CEC to your > device which is not a repeater but can actually handle the remote command. > > This is a future extension, though. There are two separate things when dealing with CEC: the low-level kernel implementation of a bus for connecting with CEC devices, and userspace APIs for using its features. If you were needing it only internally inside the kernel, there's no need for new ioctl's. So, your proposal seems to add a raw interface for it, and do all the work in userspace. An alternative approach, that it is the way most Kernel API's do is to write/use higher userspace APIs, abstracting the hardware internals. V4L, DVB and RC, input/event, vfs, tty, etc are good examples of how we do APIs in Linux. We should only go to a raw API if the high-level ones won't work. Also, a raw-level implementation of CEC may/will interfere on higher level interfaces. For example, assuming that we have both raw and RC interfaces using HDMI-CIC, a raw access on one process during a RC reception or transmit could interfere on another process using the high-level interface for RC (as a raw access to a block device may actually corrupt data). So, raw interfaces are evil, and generally require CAP_SYS_ADMIN. So, I think we should first discuss what are the needs, and then discuss how to implement them. Cheers, Mauro. -- To unsubscribe from this list: send the line "unsubscribe linux-media" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
