[linux-dvb] Re: A/V dropouts in dvb-kernel driver

[Klaus Schmidinger <Klaus.Schmidinger@cadsoft.de>]

Klaus Schmidinger wrote:
> 
> I've tracked down the problem somewhat more and found that
> in ves1x93.c, in case 'FE_READ_STATUS', the 'sync' value
> sometimes has somewhat strange values. Mostly, when it's
> wrong, one of the lower two bits is 0 (sometimes even both),
> so this would indicate that there _is_ VITERBI _and_ SYNC,
> but no SIGNAL or CARRIER (which clearly can't be, because without
> signal or carrier there can't be anything else).
> 
> As a workaround (since the actual problem apparently is inside the
> VES1893 chip) I have added the following code, which re-reads the
> 'sync' value in case it doesn't make sense:
> 
> --- linux/drivers/media/dvb/frontends/ves1x93.c  2004-08-25 21:27:08.000000000 +0200
> +++ linux/drivers/media/dvb/frontends/ves1x93.c  2004-09-11 15:03:09.000000000 +0200
> @@ -452,6 +452,11 @@
>         {
>                 fe_status_t *status = arg;
>                 u8 sync = ves1x93_readreg (i2c, 0x0e);
> +               int maxtry = 10; // just for safety - let's not get stuck here
> +               while ((sync & 0x03) != 0x03 && (sync & 0x0c) && maxtry--) {
> +                       msleep(10);
> +                       sync = ves1x93_readreg (i2c, 0x0e);
> +                }
> 
>                 *status = 0;
> 
> 
> With this modification the dropouts are gone on my system.
> Maybe somebody more familiar with the driver internals could comment
> on this. Does it make sense?
> Since it is in a hardware specific source file, it shouldn't have
> any side effects on other hardware.
> 
> Maybe it should be even enhanced so that any 'sync' value that doesn't
> have its '1' bits contiguously set starting at the lowest bit will be
> considered faulty.

Ok, I've tried that now, but it didn't work as expected.
So the above code still appears to be a good fix.

Klaus