[Openchrome-users] Fwd: opencrome in Debian 5 on Epia M.

[Wander Winkelhorst]

On Fri, Mar 27, 2009 at 1:21 PM, John Robinson
<john.robinson at anonymous.org.uk> wrote:
> On 23/03/2009 18:21, Francesco Bochicchio wrote:
>> 2009/3/22 Xavier Bachelot <xavier at bachelot.org <mailto:xavier at bachelot.org>>
> [...]
>> ? ? I think it was never really fixed and the support is disabled by
>> ? ? default. Grep for CONFIG_X86_LONGHAUL in your kernel config.
>>
>> I can add that this option is one of the differences between the
>> 'faulty' kernel, which has the option enabled as module, and the 'good'
>> one, which has it disabled.
>> However, I'm not sure about this diagnosis: my box is permanently on
>> and running a P2P program ?which usually uses ?between 15% and 40% of
>> CPU time. When I play a movie, I don't stop that program, so at least
>> 15% of CPU is always used. If the failure was due to an attempt to
>> downclock the CPU, it should occur sometime also without playing a
>> movie, e.g. when the P2P program run out of sources. It never occurred
>> despite an uptime of days with the P2P program always running.
>
> Yes, but almost any time you do anything interactive, like start playing
> a DVD, the CPU speed gets put back up to the maximum, and the crashes
> only seem to occur on speed changes when there's heavy I/O: your P2P
> program's I/O won't be much more than your upstream connection speed x2
> (from disc, to network), while playing video involves much more data
> coming from disc then shuffling a screen's worth of data around 50-60
> times per second.

Speed doesn't actually have anything to do with these crashes; It is
because the DMA engine get confused by the switching CPU speed in
combination with Busmaster DMA.  The acceleration code uses BM-DMA
(and so does the IDE and some capture cards) network doesn't use
BM-DMA.