Hi ! >From a users point of view i would like to have some clarification on this discussion. Lets take a (now real world) example. Having /dev/dvb/adapter0/demux0 /dev/dvb/adapter0/dvr0 /dev/dvb/adapter0/frontend0 /dev/dvb/adapter0/frontend1 /dev/dvb/adapter0/net0 /dev/dvb/adapter1/demux0 /dev/dvb/adapter1/dvr0 /dev/dvb/adapter1/frontend0 /dev/dvb/adapter1/frontend1 /dev/dvb/adapter1/net0 for a C/T card, where one fe is C and the other is T , connector is only one per adapter. How is the logic to handle this ? Two big issues i don't properly understand at the moment. 1.) VDR does not know that frontend1 is related to demux0. since no application changes magically on its own, can someone provide some information of what an application is expected to do, to handle this properly. With this information it could be discussed at i.e. vdr mailing list. 2.) How does an application/user/whatever know what can be used ? I mean there is only C connected OR T - how can the application know what fe can be used (assumed point 1. has been fixed). How would the user know, which is which and how one should specify what is connected ? 3.) ca/caio device handling - is this something an application should implement ... and how. Please help me to understand these points as this is something which pops up regular in discussion with our (yavdr.org) users. For 1 and 2 the only proper solution i see would be 1 frontend instead of 2 with a possibility to specifiy the transport in one way or another. (which -> to be discussed) Regards Steffen On Sat, 16 Jul 2011 17:40:53 +0200 Oliver Endriss <o.endriss@xxxxxx> wrote: > On Saturday 16 July 2011 16:54:50 Mauro Carvalho Chehab wrote: > > Em 16-07-2011 11:16, Antti Palosaari escreveu: > > > On 07/16/2011 03:25 PM, Mauro Carvalho Chehab wrote: > > >> Em 15-07-2011 20:41, Antti Palosaari escreveu: > > >>> On 07/15/2011 08:01 PM, Andreas Oberritter wrote: > > >>>> On 15.07.2011 15:25, Mauro Carvalho Chehab wrote: > > >>>>> Em 15-07-2011 05:26, Ralph Metzler escreveu: > > >>>>>> At the same time I want to add delivery system properties to > > >>>>>> support everything in one frontend device. > > >>>>>> Adding a parameter to select C or T as default should help > > >>>>>> in most cases where the application does not support > > >>>>>> switching yet. > > >>>>> > > >>>>> If I understood well, creating a multi-delivery type of > > >>>>> frontend for devices like DRX-K makes sense for me. > > >>>>> > > >>>>> We need to take some care about how to add support for them, > > >>>>> to avoid breaking userspace, or to follow kernel deprecating > > >>>>> rules, by adding some legacy compatibility glue for a few > > >>>>> kernel versions. So, the sooner we add such support, the > > >>>>> better, as less drivers will need to support a "fallback" > > >>>>> mechanism. > > >>>>> > > >>>>> The current DVB version 5 API doesn't prevent some userspace > > >>>>> application to change the delivery system[1] for a given > > >>>>> frontend. This feature is actually used by DVB-T2 and DVB-S2 > > >>>>> drivers. This actually improved the DVB API multi-fe support, > > >>>>> by avoiding the need of create of a secondary frontend for > > >>>>> T2/S2. > > >>>>> > > >>>>> Userspace applications can detect that feature by using > > >>>>> FE_CAN_2G_MODULATION flag, but this mechanism doesn't allow > > >>>>> other types of changes like from/to DVB-T/DVB-C or from/to > > >>>>> DVB-T/ISDB-T. So, drivers that allow such type of delivery > > >>>>> system switch, using the same chip ended by needing to add > > >>>>> two frontends. > > >>>>> > > >>>>> Maybe we can add a generic FE_CAN_MULTI_DELIVERY flag to > > >>>>> fe_caps_t, and add a way to query the type of delivery > > >>>>> systems supported by a driver. > > >>>>> > > >>>>> [1] > > >>>>> http://linuxtv.org/downloads/v4l-dvb-apis/FE_GET_SET_PROPERTY.html#DTV-DELIVERY-SYSTEM > > >>>> > > >>>> I don't think it's necessary to add a new flag. It should be > > >>>> sufficient to add a property like > > >>>> "DTV_SUPPORTED_DELIVERY_SYSTEMS", which should be read-only > > >>>> and return an array of type fe_delivery_system_t. > > >>>> > > >>>> Querying this new property on present kernels hopefully fails > > >>>> with a non-zero return code. in which case FE_GET_INFO should > > >>>> be used to query the delivery system. > > >>>> > > >>>> In future kernels we can provide a default implementation, > > >>>> returning exactly one fe_delivery_system_t for unported > > >>>> drivers. Other drivers should be able to override this default > > >>>> implementation in their get_property callback. > > >>> > > >>> One thing I want to say is that consider about devices which > > >>> does have MFE using two different *physical* demods, not > > >>> integrated to same silicon. > > >>> > > >>> If you add such FE delsys switch mechanism it needs some more > > >>> glue to bind two physical FEs to one virtual FE. I see much > > >>> easier to keep all FEs as own - just register those under the > > >>> same adapter if FEs are shared. > > >> > > >> In this case, the driver should just create two frontends, as > > >> currently. > > >> > > >> There's a difference when there are two physical FE's and just > > >> one FE: with 2 FE's, the userspace application can just keep > > >> both opened at the same time. Some applications (like vdr) > > >> assumes that all multi-fe are like that. > > > > > > Does this mean demod is not sleeping (.init() called)? > > > > > >> When there's just a single FE, but the driver needs to "fork" it > > >> in two due to the API troubles, the driver needs to prevent the > > >> usage of both fe's, either at open or at the ioctl level. So, > > >> applications like vdr will only use the first frontend. > > > > > > Lets take example. There is shared MFE having DVB-S, DVB-T and > > > DVB-C. DVB-T and DVB-C are integrated to one chip whilst DVB-S > > > have own. > > > > > > Currently it will shown as: > > > > Let me name the approaches: > > > > Approach 1) > > > * adapter0 > > > ** frontend0 (DVB-S) > > > ** frontend1 (DVB-T) > > > ** frontend2 (DVB-C) > > > > Approach 2) > > > Your new "ideal" solution will be: > > > * adapter0 > > > ** frontend0 (DVB-S/T/C) > > > > Approach 3) > > > What really happens (mixed old and new): > > Why does this happen? > > > > * adapter0 > > > ** frontend0 (DVB-S) > > > ** frontend1 (DVB-T/C) > > > > What I've said before is that approach 3 is the "ideal" solution. > > No, sorry. > > > > It does not look very good to offer this kind of mixed solution, > > > since it is possible to offer only one solution for userspace, > > > new or old, but not mixing. > > > > Good point. > > > > There's an additional aspect to handle: if a driver that uses > > approach 1, a conversion to either approach 2 or 3 would break > > existing applications that can't handle with the new approach. > > > > There's a 4th posibility: always offering fe0 with MFE > > capabilities, and creating additional fe's for old applications > > that can't cope with the new mode. For example, on a device that > > supports DVB-S/DVB-S2/DVB-T/DVB-T2/DVB-C/ISDB-T, it will be shown > > as: > > > > Approach 4) fe0 is a frontend "superset" > > > > *adapter0 > > *frontend0 (DVB-S/DVB-S2/DVB-T/DVB-T2/DVB-C/ISDB-T) - aka: FE > > superset *frontend1 (DVB-S/DVB-S2) > > *frontend2 (DVB-T/DVB-T2) > > *frontend3 (DVB-C) > > *frontend4 (ISDB-T) > > > > fe0 will need some special logic to allow redirecting a FE call to > > the right fe, if there are more than one physical frontend bound > > into the FE API. > > > > I'm starting to think that (4) is the better approach, as it won't > > break legacy applications, and it will provide an easier way for > > new applications to control the frontend with just one frontend. > > Nack. Do not make it more complicated than neccessary! > Approach (2) is the way to go. > > I consider the current way as a clear abuse of the DVB API. > It is a bug, not a feature! > > Originally it was intended to support multiple data paths per adapter. > For example, A dual tuner DVB-S card should have been implemented as > one adapter: > > adapterX +--- demux0/frontend0/ca0/dvr0/net0 > | > +--- demux1/frontend1/ca1/dvr1/net1 > > (Both tuners can be used concurrently without limitations.) > > My proposal is: > If there is any kind of shared hardware, i.e. the application cannot > use both adapters independently, these hardware must be folded into a > single frontend. > > It is not so hard to implement, even for separate chips: > The driver just has to "switch" from one set of frontend ops to > another. > > Btw, which applications do really handle this fronten0/1 stuff > correctly? VDR definitely does not. Access to the frontend1 fails. > > CU > Oliver > -- 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
