Re: [PATCH RFC 01/91] [media] dvb-core: allow demods to specify the supported delivery systems supported standards.

[Mauro Carvalho Chehab <mchehab@infradead.org>]

On 27-12-2011 18:37, Andreas Oberritter wrote:
> On 27.12.2011 18:06, Mauro Carvalho Chehab wrote:
>> On 27-12-2011 12:33, Andreas Oberritter wrote:
>>> On 27.12.2011 14:28, Mauro Carvalho Chehab wrote:
>>>> On 27-12-2011 10:11, Andreas Oberritter wrote:
>>>>> On 27.12.2011 02:07, Mauro Carvalho Chehab wrote:
>>>>>> DVB-S and DVB-T, as those were the standards supported by DVBv3.
>>>>>
>>>>> The description seems to be incomplete.
>>>>>
>>>>>> New standards like DSS, ISDB and CTTB don't fit on any of the
>>>>>> above types.
>>>>>>
>>>>>> while there's a way for the drivers to explicitly change whatever
>>>>>> default DELSYS were filled inside the core, still a fake value is
>>>>>> needed there, and a "compat" code to allow DVBv3 applications to
>>>>>> work with those delivery systems is needed. This is good for a
>>>>>> short term solution, while applications aren't using DVBv5 directly.
>>>>>>
>>>>>> However, at long term, this is bad, as the compat code runs even
>>>>>> if the application is using DVBv5. Also, the compat code is not
>>>>>> perfect, and only works when the frontend is capable of auto-detecting
>>>>>> the parameters that aren't visible by the faked delivery systems.
>>>>>>
>>>>>> So, let the frontend fill the supported delivery systems at the
>>>>>> device properties directly, and, in the future, let the core to use
>>>>>> the delsys to fill the reported info::type based on the delsys.
>>>>>>
>>>>>> Signed-off-by: Mauro Carvalho Chehab <mchehab@redhat.com>
>>>>>> ---
>>>>>>  drivers/media/dvb/dvb-core/dvb_frontend.c |   13 +++++++++++++
>>>>>>  drivers/media/dvb/dvb-core/dvb_frontend.h |    8 ++++++++
>>>>>>  2 files changed, 21 insertions(+), 0 deletions(-)
>>>>>>
>>>>>> diff --git a/drivers/media/dvb/dvb-core/dvb_frontend.c b/drivers/media/dvb/dvb-core/dvb_frontend.c
>>>>>> index 8dedff4..f17c411 100644
>>>>>> --- a/drivers/media/dvb/dvb-core/dvb_frontend.c
>>>>>> +++ b/drivers/media/dvb/dvb-core/dvb_frontend.c
>>>>>> @@ -1252,6 +1252,19 @@ static void dtv_set_default_delivery_caps(const struct dvb_frontend *fe, struct
>>>>>>  	const struct dvb_frontend_info *info = &fe->ops.info;
>>>>>>  	u32 ncaps = 0;
>>>>>>  
>>>>>> +	/*
>>>>>> +	 * If the frontend explicitly sets a list, use it, instead of
>>>>>> +	 * filling based on the info->type
>>>>>> +	 */
>>>>>> +	if (fe->ops.delsys[ncaps]) {
>>>>>> +		while (fe->ops.delsys[ncaps] && ncaps < MAX_DELSYS) {
>>>>>> +			p->u.buffer.data[ncaps] = fe->ops.delsys[ncaps];
>>>>>> +			ncaps++;
>>>>>> +		}
>>>>>> +		p->u.buffer.len = ncaps;
>>>>>> +		return;
>>>>>> +	}
>>>>>> +
>>>>>
>>>>> I don't understand what this is trying to solve. This is already handled
>>>>> by the get_property driver callback.
>>>>>
>>>>> dtv_set_default_delivery_caps() only sets some defaults for drivers not
>>>>> implementing get_property yet.
>>>>
>>>> dtv_set_default_delivery_caps() does the wrong thing for delivery systems
>>>> like ISDB-T, ISDB-S, DSS, DMB-TH, as it fills data with a fake value that
>>>> is there at fe->ops.info.type. 
>>>>
>>>> The fake values there should be used only for DVBv3 legacy calls emulation
>>>> on those delivery systems that are not fully compatible with a DVBv3 call.
>>>
>>> That's right. Still, there's no need to introduce fe->ops.delsys,
>>> because the drivers in question could just implement get_property
>>> instead. At least that's what we discussed and AFAIR agreed upon when
>>> Manu recently submitted his patches regarding enumeration of delivery
>>> systems.
>>
>> Manu's patches were applied (well, except for two patches related to af9013
>> driver that are/were under discussion between Manu and Antti).
>>
>> Manu's approach is good, as it provided a way to enumerate the 
>> standards without much changes, offering a way for userspace to
>> query the delivery system, at the expense of serializing a driver 
>> call for each property. 
>>
>> Yet, it doesn't allow the DVB core to detect the supported
>> delivery systems on a sane way [1].
>>
>> The addition of fe->ops.delsys is going one step further, as it will
>> allow, at the long term, the removal of info.type.
> 
> You can't remove info.type, as it would break userspace. What you can do
> is deprecate its use inside drivers.

Yes, that's what I'm meaning: remove it from the drivers, and let 
dvb_frontend fill it for DVBv3 calls, according with the delivery
system.

>> There are two reasons why we need to get rid of info.type:
>>
>> 1) dvb_frontend core can be changed to use fe->ops.delsys
>>    internally, instead of info.type, in order to fix some
>>    bugs inside it, where it does the wrong assumption, because
>>    the frontend is lying about the delivery system;
>>
>> 2) There is no sane way to fill fe->ops.info.type for Multi delivery
>>    system frontends, like DRX-K, that supports both DVB-T and DVB-C.
>>    The type can be filled with either FE_QAM or FE_OFDM, not with both.
>>    So, choosing either type will be plain wrong, and may cause bad
>>    side effects inside dvb_frontend.
>>
>> [1] Ok, it would be possible to do a hack internally, for it to call
>> get_property() but it is a way more sane to just read fe->ops.delsys.
> 
> I don't see why using the already defined interface to query the
> possible delivery systems would be a hack.

It is a hack, as instead of just doing a single table lookup, the current
approach is to have a code filling information based on fe->ops.info.type,
with can be wrong, and then jump to the driver's code to get the right
information, and, for each DTV properties, on every call, call get_property()
to do nothing for all calls.

> I also don't see what the enumeration has to do with the deprecation of
> fe->ops.info.type. If you want to replace fe->ops.info.type, then what
> you need to know is the currently selected delivery system, not the
> supported delivery systems. The currently selected system is available
> from the property cache.

Yes, most of the code should be using the current delivery system.

Yet, there are some places at dvb_frontend.c that uses fe->ops.info.type,
like for instance, at dtv_property_cache_init():
 */
static void dtv_property_cache_init(struct dvb_frontend *fe,
                                    struct dtv_frontend_properties *c)
{
        switch (fe->ops.info.type) {
...

The initial selected system is based on fe->ops.info.type. The 
best thing to do, instead, would be to use fe->ops.delsys[0].
Without properly setting the initial current delivery system, a DVBv3 
call won't work.

So, in order to remove fe->ops.info.type filling from the drivers, the
DVB core should have an initialization code like:

	c->delivery_system = fe->ops.delsys[0];
	switch (fe->ops.delsys[0]) {
	case SYS_DVBT:
	case SYS_ISDBT:
	case SYS_DMBTH:
		fe->ops.info.type = FE_OFDM;
		break;
...
	}

Regards,
Mauro