[mythtv-users] Firewire troubles after upgrade. Is test-mepg2 mightier than MythTV?

Jarod Wilson jarod at wilsonet.com
Wed Jan 27 04:02:58 UTC 2010


On Tue, Jan 26, 2010 at 10:42 PM, Deyan <mythtv at bektchiev.net> wrote:
> On Tue, Jan 26, 2010 at 2:11 PM, Jarod Wilson <jarod at wilsonet.com> wrote:
>>
>> On Tue, Jan 26, 2010 at 4:44 PM, Eric Sharkey <eric at lisaneric.org> wrote:
>> > On Tue, Jan 26, 2010 at 4:24 PM, Jarod Wilson <jarod at wilsonet.com>
>> > wrote:
>> >>> I installed an old PCI firewire controller (VIA Technologies, Inc.
>> >>> VT6306 Fire II) and everything just worked.
>> >>
>> >> Is that Via chip reporting as OHCI 1.0 or 1.1? Trying to prove out my
>> >> theory that OHCI 1.1's dual-buffer isochronous receive is generally
>> >> boned...
>> >
>> > The Via chip is reporting as OHCI 1.0 and the TI is reporting as OHCI
>> > 1.1.
>>
>> Would be very interesting then to see if you're previously non-working
>> TI chip starts working in 2.6.33 (or in an earlier kernel respun with
>> a one-line change that'll keep ohci 1.1 cards from using
>> dual-buffer)...
>
> Is there some way to know what versions of the TI chipsets (and how do you
> even find the version) have the issue?

Chipset errata documentation, trial and error. Your lspci output shows
you have a TSB43AB23. From what I recall, that one has a quirk, where
if any of its iso receive buffers got mapped above the 4GB boundary
(which happens even on a box with only 4GB of RAM due to bios
remapping), and it was running in dual-buffer mode, it would stall
out. It was either that chip or the TSB43AB22/A that had that quirk,
maybe both...

> I have a TI chip on a Gigabyte motherboard and since upgrading to Fedora 11
> 64-bit Kernel (2.6.30.10-105.fc11.x86_64) it has been rock solid. Initially
> I had numerous issues with it but I thought it was due to the amount of
> memory installed and weird behavior that I was seeing then. I upgraded from
> 2GB to 4GB and the firewire card would not receive anything from the
> cablebox at all. Doing memory tests also showed problems during some
> iterations and it didn't matter what the DIMMs were or in which banks they
> were placed (4GB in 1GB and 2GB DIMMs in any combination would cause the
> issue but 2GB rotating the DIMMs in the four banks was always passing).
>
> In the end I thought it was a bad motherboard and decided to live with only
> 2GB RAM which was anyway sufficient, I just can't run a Windows VM on the
> box.
>
> Now after 7 months Firewire recordings have never been more reliable since
> the 2+ years that I've had it before.

This one *might* be okay with 4GB with a later kernel, not certain.
I'd have to go look at the driver code in that specific kernel... I
thought it was a while ago when we quirked that chipset to only use
packet-per-buffer, but maybe it wasn't that long ago, and/or it was
the AB22/A, but there were unconfirmed suspicions it might affect the
AB23 as well. A good test would be 4GB with a kernel built to only
enable packet-per-buffer on that chip.

> Below is the data from lspci and dmesg related to the card.
...
> 05:07.0 FireWire (IEEE 1394): Texas Instruments TSB43AB23 IEEE-1394a-2000


-- 
Jarod Wilson
jarod at wilsonet.com


More information about the mythtv-users mailing list