USB disk on CS5536 unstable

[Anselm Strauss]

On 10/02/10 16:39, Hans Petter Selasky wrote:
> On Saturday 02 October 2010 14:44:07 Anselm Strauss wrote:
>> On 09/30/10 21:38, Hans Petter Selasky wrote:
>>> On Thursday 30 September 2010 21:10:59 Anselm Strauss wrote:
>>>> Maybe sending it to just the USB list was too specific ...
>>>>
>>>> On 09/30/10 00:08, Anselm Strauss wrote:
>>>>> Hi
>>>>>
>>>>> I have an ALIX board that has an AMD Geode and the CS5536 companion
>>>>> chip with integrated USB on it. When I connect a USB disk I have
>>>>> observed various problems. For example when I run fsck_ufs on a 250 GB
>>>>> partition the process gets stuck in biord state and fsck reports
>>>>> unreadable sectors. When I do a dd over the whole disk and direct it
>>>>> to /dev/null it suddenly returns with no error, but having read only a
>>>>> small fraction of the disk. I tried it with two different disks and
>>>>> two different ALIX boards. I'm pretty sure the disks are okay since I
>>>>> tried them on other hardware.
>>>>>
>>>>> As far as I know there was some trouble with the chip regarding
>>>>> timeouts. Under load after some time the USB just stops responding. I
>>>>> have tried 8.0 and 8.1. Is there any known problem? How can I track
>>>>> this down?
>>>>>
>>>>> Anselm
>>>
>>> If you compile the kernel with USB_DEBUG, then there are some sysctls
>>> under hw.usb.ehci which you can tweak. Needs to be set before boot.
>>>
>>> --HPS
>>
>> Did not know that there were configurable bug workarounds in sysctl.
>> When I set hw.usb.ehci.lostintrbug=1 in /boot/loader.conf the problems
>> seem gone.
>>
>> Without this setting I got the following kernel message when dd did abort:
>>
>> ehci_timeout: xfer=0xc29cd3c8
>>
>>
>> Thanks,
>> Anselm
> 
> Maybe you can report the PCI vendor ID and product so that we can add this 
> quirk.
> 
> --HPS

Not sure what a "PCI vendor ID" is and how to determine it. It's a ALIX
2d2 from http://pcengines.ch/alix.

Anselm