[LAU] UA-101

[Clemens Ladisch]

Colin Fletcher wrote:
> ua-1000-ehci-oldsched.log is the result with CONFIG_USB_EHCI_TT_NEWSCHED
> not set: the URBs still complete out-of-order.
> 
> Out of curiousity, because I have a dual-core CPU, I also tried booting 
> the CONFIG_USB_EHCI_TT_NEWSCHED=y kernel with maxcpus=0 
> (ua-1000-ehci-newsched-maxcpus-0.log) and maxcpus=1 
> (ua-1000-ehci-newsched-maxcpus-1.log), to see what would happen.
> 
> Curiously, maxcpus=0 shows the URBs completing in order, except that 
> submitting URB 20 is delayed until after 0-19 have completed. maxcpus=1 
> has the URBs out-of-order. I don't know what to conclude from this; 
> maybe it makes some sense to you?

With maxcpus=0, several URBs are completed out of order after timestamp
203.58, and they should not complete at that time but ten milliseconds
later.

As far as I can see, none of these changes make much of a difference.
(The TT scheduler should affect only low or full speed transfers.)

> Is there anything else you'd like me to try?

Please change line 376 to

	printk(KERN_DEBUG "completed URB %d, status: %d, bytes: %u\n", err, urb->status, urb->iso_frame_desc[0].actual_length);

and show the logs for both maxcpus=1 and multiple cores.

Can you try this on any other computer?

> Or anything else you might need to know about my system?

Is this really an unpatched kernel.org kernel?
What USB controller do you have (see lspci)?

> Is there a more appropriate list for this, or are we still on-topic for LAU?

I'm not sure if this is a problem with the EHCI driver or your USB
controller, but both would be appropriate for the linux-usb list.


Regards,
Clemens