[linux-dvb] Re: Erroneous unsupported PLL synthesizer: foundit!!

[Holger Waechtler <holger@convergence.de>]

Michael Hunold wrote:
> Hello Holger, hello Andrew,
>
> > > Here you are.. this should also still report address errors correctly.
> > looks good,
>
> This change is not as good as it looks. It may work on dvb cards 
> perfectly, but it has negative side effects on my analog saa7146 based 
> cards.
>
> For example, the video muxes (tea6415c), audio muxes (tea6420) and the 
> tuners may be present on very different i2c addresses, depending what 
> the engineer has chosen. Because of the fact, that the mxb driver for 
> example cannot specify on which addresses it expects this devices to be, 
> it uses intesive probing.
>
> Your change now retransmits these probe i2c messages for 3 retries, even 
> if the first one has already shown that the device is not there (address 
> error). This greatly increases the time the driver needs for loading, 
> from about 1 seconds to 5 seconds.
errmm, usually i2c controllers can see a NACK after a few milliseconds, 
so might there be a general problem in the SAA i2c driver again?

In any case you have to retry when errors occur in the adress phase, i2c 
is a multi-master protocol and you have to retransmit your message in 
case of an arbitration error. See 
http://www.semiconductors.philips.com/acrobat/various/I2C_BUS_SPECIFICATION_1995.pdf 
for an explanation of this situation.


> > but I wonder if we shouldn't add a delay'n'retry on adress error for 
> > the current message only...
> > (imagine the stv0299 i2c gate case: 1st msg is opening the gate, then 
> > the second message has to wait a bit, but we don't want to resend the 
> > entire message...)
>
> The problem is the crappy i2c behaviour of these chipsets. I mean, i2c 
> is such a simple protocol and nevertheless some engineers "succeed" in 
> screwing it up. 8-/
(well, IMHO i2c is crappy by specification,)


> > Anyway, I applied your patch to CVS, further optimisations can follow 
> > later.
>
> I modified it the following way: currently, all analog saa7146 based 
> cards use i2c transactions with i2c irqs enabled, for better 
> performance. dvb cards don't use this, since it screws up gpio/debi irqs 
> and results in "gpio oops"/"debi oops" messages and ts lockup.
>
> So your new address error behaviour is only chosen for cards that don't 
> use i2c irqs (0!= SAA7146_USE_I2C_IRQ & dev->ext->flags). So dvb cards 
> get the new behaviour, analog cards have the old behaviour back. Ok?
hmmm, that's working but I'd like more a spec-conform approach.

btw: While reviewing this code I was not sure anymore if the retry loop 
should retry all messages or just the failed ones. What do you think?

Holger



--
Info:
To unsubscribe send a mail to listar@linuxtv.org with "unsubscribe linux-dvb" as subject.