Bug#596528: Debuging large transfer under usb3 device
Hi,
In February, Sarah Sharp wrote:
> On Mon, Feb 27, 2012 at 08:18:13PM +0100, guillaume.jaouen@free.fr wrote:
>> Hello Jonathan,
>>
>> I followed your advices and try to track the usb 3.0 bug occuring
>> during large transfer on usb 3.0 device with kernel 3.2.0.1.
>>
>> You'll find in attachements as asked :
Sorry for the slow reply, and thanks to Sarah for the cc.
Guillaume, I never received your message [1]. Would you mind
resending it to me privately (jrnieder@gmail.com)?
>> - full "dmesg" output from booting
>> - "lsusb -vvv" output
>
> Can you also run `sudo lspci -vvv` for me, so I know which host
> controller you have?
Accordingly to [2] it's a Buffalo USB 3.0 Express Card. Looks like
it wasn't plugged in when the report was filed so the vendor and
product id and other PCI registers aren't in the report.
[...]
> BTW, what are these errors in your dmesg about? If some userspace
> program is sending bad SCSI commands to the device, then it's no wonder
> the device was stalling the commands:
>
>> [ 2.265783] md: raid10 personality registered for level 10
>> [ 2.284315] mdadm: sending ioctl 800c0910 to a partition!
>> [ 2.284377] mdadm: sending ioctl 800c0910 to a partition!
>> [ 2.284436] mdadm: sending ioctl 1261 to a partition!
>> [ 2.284487] mdadm: sending ioctl 1261 to a partition!
>> [ 2.284965] mdadm: sending ioctl 1261 to a partition!
>> [ 2.285016] mdadm: sending ioctl 1261 to a partition!
>> [ 2.285379] mdadm: sending ioctl 1261 to a partition!
>> [ 2.285431] mdadm: sending ioctl 1261 to a partition!
>> [ 2.285809] mdadm: sending ioctl 1261 to a partition!
>> [ 2.285860] mdadm: sending ioctl 1261 to a partition!
These look like instances of <http://bugs.debian.org/656899>, which
should be unrelated.
[...]
> I can't be sure that your host controller is the thing that's broken
> unless you rebuild your kernel with CONFIG_USB_DEBUGGING and
> CONFIG_USB_XHCI_HCD_DEBUGGING turned on, and capture the full dmesg
> starting just before that transfer error. You'll really want to be
> running 3.3, since that cleaned up a lot of the xHCI driver debugging,
> and the log file will be much smaller.
Did anything come of these questions? Some instructions for building
a custom kernel on Debian are at [3], for what it's worth.
> It's possible, although not likely, that we're over writing the link TRB
> on the command ring and causing the host controller to step off into
> lala land and access bad memory. My other theory is that your express
> card is just broken and can't handle the throughput. Or perhaps it's an
> early prototype that made it out into the market without having good
> transfer error support.
Thanks for your help.
Sincerely,
Jonathan
[1] message-id 1631973011.35885137.1330370293927.JavaMail.root@zimbra47-e8.priv.proxad.net
[2] http://bugs.debian.org/596528
[3] http://kernel-handbook.alioth.debian.org/ch-common-tasks.html#s-kernel-org-package
Reply to: