[linux-dvb] Re: dvb-kernel TS continuity errors

[Manu Abraham <manu@kromtek.com>]

On Sat September 25 2004 3:47 pm, Klaus Schmidinger wrote:
> I'm currently testing VDR under kernel 2.6 with the dvb-kernel
> driver (version from linuxtv.org CVS as of 2004-09-10).
>
> When I run 5 parallel recordings on the same transponder
> (RTL, RTL 2, Super RTL, VOX and RTL Shop) it works just fine on a full
> featured DVB card, while I get TS continuity errors when I
> do the same thing on a WinTV NOVA-S budget card.


		Many other people including myself had the same problem, in similar 
conditions when using a Twinhan card. It is not limited to the AV7110 cards. 
I was wondering why only a few people had these errors ?

		The problem, what i understood is in dvb_demux.c where sections are copied 
without checking whether that section is complete. patrick provided a 
patch ... to check the section.. it was even applied to the CVS, but later on 
removed since the patch was resulting in buffer overflows.

Manu

> What makes this so strange is that normally it is said that
> a _budget_ card can deliver the full TS, while a full featured
> card is limited in what it can deliver. In this scenario it would
> appear that the FF card works better than a budget card.
>
> Could it be that there is a bug in the budget card driver
> that causes problems with multiple recordings from the same
> transponder (which means high data transfer rate)?
>
> Note that with two recordings on the budget card (RTL and RTL 2)
> I don't get these problems. The more recordings I run on the same
> budget card, the more TS continuity errors I get. On the FF card
> with 5 recordings I don't get a single continuity error. Needless
> to say that the recordings where the continuity errors happened are
> broken (video and audio artefacts), while those made with the FF card
> are perfectly fine.
>
> Also, during all these tests there were no buffer overflows in VDR,
> and also none in the driver (at least none that were reported by the
> read() call).
>
> One observation is that with 5 recordings the FF card typically returns
> some 30KB in one read(), while the budget card delivers around 4KB per
> read(). So with the budget card there are apparently a lot more read()
> calls. Don't know whether this constitues a problem, though.
>
> Any ideas what could be causing this?
>
> Klaus