[Vorbis] arecord/oggenc stops after 3 hours 45 minutes 47 seconds

[Monty]

> ecasound works pretty well, but now I'm getting some warnings about
> dropped samples.  I don't recall the exact message, and I'm at work
> now so I can't look at it.  The message had "alsa" in it somewhere,
> but I'm not sure whether it's from ecasound or the ALSA library.
> The machine is a dual Athlon 1900+ running a 2.6.8 kernel, with
> nothing else going on, so I'm surprised that there would be any
> dropped samples.  But this isn't an Ogg Vorbis problem, so I'll
> take it up elsewhere.

Most AMD chipsets have bugs having to do with write combining and the
PCI latency timer that allows PCI bus devices to lock down the bus as
long as they desire and ignore the latency timer (allowing the sound
card record/playback buffers to overflow/starve).  Usually PCI video
cards trip this although it's possible for a disk controller to do the
same.  The workaround is to disable write-combining for the offending
hardware device using /proc/mtrr.

Very likely this isn't actually the issue, but it's worth mentioning
as it's bitten me on every Athlon I've owned.  This is specific to the
PCI bus, an AGP bus is physically seperate on AMD northbridges.

Monty

(Strictly speaking the bug isn't a bug; it's a PCI spec loophole many
devices abuse because it can increase their bus performance 5-10% on
the synthetic benchmarks most hardware mags are so in love with.... of
course, this is at the expense of the rest of the system)