[vdr] Re: consecutive poll timeouts

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

"C.Y.M." wrote:
> 
> After performing a few recordings yesterday, I noticed that there were
> several hours of ring buffer overflow errors in my syslog. I am using a
> Nexus-S DVB card. The only thing that has changed is updating the following
> file:
> 
> CVSROOT:        /cvs/linuxtv
> Module name:    dvb-kernel
> Changes by:     kenneth 20041001 05:28:43
> 
> Modified files:
>         linux/drivers/media/dvb/ttpci: av7110_ca.c
> 
> Log message:
> - Fix dvb-ttpci ca write poll.

I doubt that this fix has anything to do with it (although I'm not
yet using the latest CVS driver version). However, to be sure you
might want to try the same scenario without that change.

> Here are a few snips of my log:
> 
> Oct  1 04:00:56 poseidon vdr[8340]: buffer usage: 90% (tid=753672)
> Oct  1 04:00:56 poseidon vdr[8340]: buffer usage: 100% (tid=753672)
> Oct  1 04:00:56 poseidon vdr[8340]: buffer usage: 90% (tid=753672)
> Oct  1 04:00:56 poseidon vdr[8340]: buffer usage: 100% (tid=753672)
> Oct  1 04:00:57 poseidon vdr[8340]: ERROR: 2279 ring buffer overflows
> (428341 bytes dropped)
> Oct  1 04:00:57 poseidon vdr[8340]: buffer usage: 90% (tid=753672)
> Oct  1 04:00:57 poseidon vdr[8340]: buffer usage: 100% (tid=753672)
> Oct  1 04:00:57 poseidon vdr[8340]: buffer usage: 90% (tid=753672)
> 
> <repeat>
> 
> Oct  1 04:08:20 poseidon vdr[8340]: buffer usage: 90% (tid=753672)
> Oct  1 04:08:20 poseidon vdr[8340]: buffer usage: 100% (tid=753672)
> Oct  1 04:08:20 poseidon vdr[8340]: buffer usage: 90% (tid=753672)
> Oct  1 04:08:21 poseidon vdr[8347]: clearing device because of consecutive
> poll timeouts
> Oct  1 04:08:21 poseidon vdr[8340]: buffer usage: 0% (tid=753672)
> 
> <repeat>
> 
> Oct  1 05:59:59 poseidon vdr[8340]: buffer usage: 90% (tid=753672)
> Oct  1 05:59:59 poseidon vdr[8340]: buffer usage: 100% (tid=753672)
> Oct  1 06:00:00 poseidon vdr[8340]: ERROR: 5638 ring buffer overflows
> (1059833 bytes dropped)
> Oct  1 06:00:00 poseidon vdr[10744]: recording thread ended (pid=10744,
> tid=868363)
> Oct  1 06:00:00 poseidon vdr[1911]: buffer stats: 141188 (1%) used
> 
> Is it possible that this driver update could have caused this?  I would also
> like to mention that even though I had all of these buffer errors, the
> recording was still good.

The "clearing device because of consecutive poll timeouts" would indicate
that there was a "Transfer Mode" going on. That would explain why the recording
was fine. Apparently it was the Transfer Mode's buffer that ran full.

One of the things in my current work of optimizing buffer usage in VDR
was to clear the ring buffer in such a case, because an "almost full"
ring buffer in Transfer Mode tends to cause buffer overflows every now and then.

Klaus