[LAU] USB audio interface and a buggy USB controller

[Jeremy Jongepier]

On 09/15/2010 09:59 AM, Jeremy Jongepier wrote:
> On 09/14/2010 06:26 PM, Monty Montgomery wrote:
>> On Tue, Sep 14, 2010 at 8:29 AM, Jeremy Jongepier <jeremy at autostatic.com> wrote:
>>> On 09/14/2010 02:23 PM, Clemens Ladisch wrote:
>>>> Jeremy Jongepier wrote:
>>>>> cannot submit datapipe for urb 0, error -28: not enough bandwidth
>>>>
>>>> Is CONFIG_USB_EHCI_TT_NEWSCHED enabled in your kernel's configuration?
>>>>
>>>> Can you try with or without a hub?
>>>>
>>>
>>> I can't try with a hub, I don't have any.
>>
>> The bandwidth has been fragmented by other devices.  USB-1 audio
>> devices require large _contiguous_ blocks of bandwidth schedule, and a
>> mouse or whatever that has carved its tiny little block out of the
>> middle of the bandwidth schedule makes all the bandwidth on either
>> side useless to an isochronous device.  Simply unplugging it (or
>> whatever caused the scheduling problem) is not enough; the schedule
>> remains fragmented, as the linux EHCI driver does not know how to
>> reconsolidate fragmented bandwidth chunks.  You have to rmmod EHCI or
>> reboot to get a fresh start.
>>
>> It's possible you're on a hub and don't know it.  Most machines with
>> multiple USB ports are actually using a single controller and a built
>> in hub.  Hubs make things *far* worse, as the hub is a dumb puppet
>> controlled entirely by the host controller, and Linux's hub scheduling
>> algorithm (even the 'new improved' one mentioned above) is rather
>> poor.  I wrote a new one years ago that approached theoretical maximum
>> efficiency and could rebalance/reconsolidate the schedule, but I
>> couldn't keep up with repatching it during the complete free-for-all
>> that has been kernel 2.6 (I'd literally spent months of full-time work
>> *just* keeping up with the churn) and it was dropped from -mm.
>>
>> Monty
> 
> Hello Monty,
> 
> I've tried with all the ports on the machine but alas, forgot to run
> lsusb so I don't know how many USB buses there are available. I use a
> real-time kernel with rtirq though and will prioritize USB to see if
> that helps.
> 
> Best,
> 
> Jeremy

I've checked with lsusb. Only two buses, and even if I try the least
crowded bus and use rtirq to prioritize that bus my UA-25 refuses to
work in duplex mode. So it might be that I'm behind a hub after all and
it's probably got nothing to do with a buggy USB chipset in my case.
But I already had a suspicion that USB soundcards wouldn't work on this
machine so I've equipped it with a decent FireWire PCIe card (the one on
the motherboard is crap also) and a FireWire soundcard.

Best,

Jeremy