Re: xhci_hcd 0000:0c:00.0 dies with "Abort failed to stop command ring: -110" exactly 24s post-init, tunneled USB4 xHCI behind Goshen Ridge (ASUS ThunderboltEX 4) + CalDigit TS4

[Michal Pecio <michal.pecio@gmail.com>]

On Sat, 26 Sep 2026 09:59:01 -0400, Ben wrote:
> > Was USB functional at all? The patch would break it. Maybe it wasn't
> > ideal, please try this revised patch which only touches the suspect
> > host controller. And don't use xhci_hcd.quirks=0x800000 anymore.  
> 
> USB was not functional on the initial patch. I had to get the dumps
> using SSH. On the revised patch, it was functional.
> 
> > Please confirm that the string "beef cafe" appears in dmesg, this shows
> > that the patch is applied and it recognized the suspect controller.
> > Then check if there are any "AMD-Vi" errors logged, particularly at
> > addresses 0xbeef0000 and 0xcafe0000, and check this:  
> 
> Interestingly, I could not find 'beef' or 'cafe' in dmesg in the
> original patch. In the revised patch, each was present in one way or
> another.

That's odd, because the original patch log shows no IOMMU errors from
any controller, even though all of them were programmed to read bogus
pointers and some (presumably?) had devices attached.

I also see no usual "Adding to iommu group N" messages. Was that kernel
running without iommu=soft or similar options? Was AMD IOMMU support
enabled in .config?

Regards,
Michal