Re: [REGRESSION] [REPRO] USB-A devices not working on boot after recent USB merge

[Mathias Nyman <mathias.nyman@linux.intel.com>]

On 10/9/25 14:14, Michal Pecio wrote:
> On Thu, 9 Oct 2025 00:25:55 +0200, Arisa Snowbell wrote:
>> This is what I get when I use good kernel:
>>
>> kernel: xhci_hcd 0000:7a:00.0: xHCI Host Controller
>> kernel: xhci_hcd 0000:7a:00.0: new USB bus registered, assigned bus number 9
>> kernel: xhci_hcd 0000:7a:00.0: USB3 root hub has no ports
>> kernel: xhci_hcd 0000:7a:00.0: hcc params 0x0110ffc5 hci version 0x120
>> quirks 0x0000000200000010
>> kernel: xhci_hcd 0000:7a:00.0: xHCI Host Controller
>> kernel: xhci_hcd 0000:7a:00.0: new USB bus registered, assigned bus number 10
>> kernel: xhci_hcd 0000:7a:00.0: Host supports USB 3.0 SuperSpeed
>> kernel: usb usb9: New USB device found, idVendor=1d6b, idProduct=0002,
>> bcdDevice= 6.17
>> kernel: usb usb9: New USB device strings: Mfr=3, Product=2, SerialNumber=1
>> kernel: usb usb9: Product: xHCI Host Controller
>> kernel: usb usb9: Manufacturer: Linux
>> 6.17.0-1-mainline-12298-gf5bd2142c274 xhci-hcd
>> kernel: usb usb9: SerialNumber: 0000:7a:00.0
>> kernel: hub 9-0:1.0: USB hub found
>> kernel: hub 9-0:1.0: 1 port detected
>> kernel: usb usb10: We don't know the algorithms for LPM for this host,
>> disabling LPM.
>> kernel: usb usb10: New USB device found, idVendor=1d6b,
>> idProduct=0003, bcdDevice= 6.17
>> kernel: usb usb10: New USB device strings: Mfr=3, Product=2, SerialNumber=1
>> kernel: usb usb10: Product: xHCI Host Controller
>> kernel: usb usb10: Manufacturer: Linux
>> 6.17.0-1-mainline-12298-gf5bd2142c274 xhci-hcd
>> kernel: usb usb10: SerialNumber: 0000:7a:00.0
>> kernel: hub 10-0:1.0: USB hub found
>> kernel: hub 10-0:1.0: config failed, hub doesn't have any ports! (err -19)
>>
>> where the 2.0 USB's work, mice is powered on, with the bad kernel the
>> mice doesn't even power the LED's on.
>> In the bad kernel its missing the New USB and all.
> 
> Okay, thanks. I now see what's going on.
> 
> I have successfully reproduced it on a normal controller by patching
> the driver to simply ignore any USB3 ports. With this patch, no root
> hubs are registered at all until I revert the "bad" commit.
> 
> In my case it's an idiotic little problem:
> 
> xhci_pci_common_probe()
> {
> 	usb_hcd_pci_probe() {
> 		// allocate xhci
> 		xhci_run(xhci);
> 	}
> 
> 	xhci->allow_single_roothub = 1;
> }
> 
> The thing is that xhci_run() needs allow_single_roothub to already
> be set when it executes, but we can't do it before xhci is allocated.
> And some non-PCI drivers don't want it to be set.
> 

Nice catch,

When that flag isn't set yet in xhci_run() it will lead to
HCD_FLAG_DEFER_RH_REGISTER flag being set, prevent hcd driver from registering
the USB 2.0 roothub before usb 3.0 hcd is added. (which never happens)

Maybe the xhci->allow_single_roothub = 1; flag could be set in xhci_pci_setup()
when called for primary hcd

-Mathias