LinuxMusicians

No i'm absolutely tired.

i think in xhci_urb_enqueue
... should do the trick. No extra USB_HOST_LOCK_IS_ACQUIRED flag needed.

You need both flags.

URB_FAST_COMPLETION must be set in all submitted ISO URBs, no matter if they are submitted from inside and outside of the completion handler.
USB_HOST_LOCK_IS_ACQUIRED must be set in ISO urbs submitted from inside the completion handler only.

I see a solution for using the completion handler in hardirq context. The reason for your hardlock is, that when resubmitting an urb, xhci_urb_enqueue tries to acquire the xhci_lock that is already held by the interrupt handler. So adding a flag USB_HOST_LOCK_IS_ACQUIRED to the urb's flag when an ur...

The right thing to do would be to have a separate kernel thread that handles the ISO completion callbacks and is scheduled instead of the tasklet_hi_schedule.

I make it short: Doing a spin_unlock/spin_lock while inside hardirq context is the best way to create a kernel panic.

Since they started to build USB interfaces RME completely ignores Linux. So buying any of their newer interfaces is a complete waste of money. TotalMix doesn't work at all. Only USB class compliant work. tvaz Having a computer where the problems shows up very quickly is actually a good thing for deb...

You can force loading of the driver on boot by putting a line into the file /etc/modules (including the module parameters). See "man modules". You can check what driver the device is using by "lsusb -t" Important is to look at inderfaces 0,1,2. If they are using snd-usb-audio then just execute the f...

I bet that it's a sound card. :) But I don't know how to find mapping between devices connected to xhci controller and their IRQs. Easy. All devices use the first allocated irq of an xhci host controller. Simply because the use of xhci secondary event rings and allocation of devices/enpoints to the...

One more try

Yes, there is hope. The reason why i did not get the scheduling bug was, because i compiled the kernel without preemption. Now that i compile it as low-latency kernel i also get the "scheduling while atomic" bug (but without a hardlock), so i can have a look whats happening. Btw, for the new Ultrali...

I will be on irc freenode, #linuxmusicians this evening from 20:00 CEST. Better and faster to sort out the problems there.

Then split the patchfile into separate files for each patch and test them incrementally.

The patchfile consists of 3 patches, each starting with the line containing the "diff".

The 3rd patch is the important one for the channel hopping, so start with that one.

@bluebell Try to boot with the option "nowatchdog".

So, the attached patch should fix the following: - startup problems with pulseaudio or jack in duplex mode - channel hopping/bit crushing due to x-runs or sync issues - channel hopping when changing buffersize on the fly with jack_bufsize Note: It can take up to about 10 seconds for sound to come ba...

The patches will only go into the kernel after i get verification from users of the Ultralite AVB that they solve the problem. I can only check with my 16A AVB which uses the same vendor/product ID as all the other AVB products. With the patches, everything runs smooth. Without the patches, channel ...