[Qemu-devel] Bug (Regression?) in hw/usb/hcd-uhci.c causes failure of ICH9 host controller and attached Xbox 360 Wireless Receiver

[Qemu-devel] Bug (Regression?) in hw/usb/hcd-uhci.c causes failure of ICH9 host controller and attached Xbox 360 Wireless Receiver

[Joe Clifford]

Hi folks,

I'm not a coder by any stretch of the imagination so I don't have a 
patch unfortunately however, this commit:

https://github.com/qemu/qemu/commit/5f77e06baa84323e5bbc96c2c7f4fe627078b210

causes the failure of the ICH9 Universal Host Controller - 2934 and the 
attached "Microsoft Common Controller For Windows Class"->"Xbox 360 
Wireless Receiver for Windows" which is a pass-through usb device 
attached to a Q35 Windows 7 x64 VM running on a Debian Jessie host.

Reverting this commit on current master (as of today) fixes the problem.

I spent some time today bisecting the git master code to find this.

If you require any more information for debugging, please just let me know.

Regards,

Joe

Re: [Qemu-devel] Bug (Regression?) in hw/usb/hcd-uhci.c causes failure of ICH9 host controller and attached Xbox 360 Wireless Receiver

[Gerd Hoffmann]

On Mi, 2016-04-20 at 22:13 +0100, Joe Clifford wrote:
> Hi folks,
> 
> I'm not a coder by any stretch of the imagination so I don't have a 
> patch unfortunately however, this commit:
> 
> https://github.com/qemu/qemu/commit/5f77e06baa84323e5bbc96c2c7f4fe627078b210
> 
> causes the failure of the ICH9 Universal Host Controller - 2934 and the 
> attached "Microsoft Common Controller For Windows Class"->"Xbox 360 
> Wireless Receiver for Windows" which is a pass-through usb device 
> attached to a Q35 Windows 7 x64 VM running on a Debian Jessie host.
> 
> Reverting this commit on current master (as of today) fixes the problem.
> 
> I spent some time today bisecting the git master code to find this.
> 
> If you require any more information for debugging, please just let me know.

Hmm, quick attempt to reproduce locally failed (slightly different
device though, a microsoft nano receiver for a mouse).  It's also not
obvious how that commit breaks usb devices.

Does the device show up in device manager?
If so:  It probably has a yellow exclamation mark?
Any error description?

Can you send the "lsusb -v" output of the device?

cheers,
  Gerd

Re: [Qemu-devel] Bug (Regression?) in hw/usb/hcd-uhci.c causes failure of ICH9 host controller and attached Xbox 360 Wireless Receiver

[Joe Clifford]

On 21/04/16 07:44, Gerd Hoffmann wrote:
> On Mi, 2016-04-20 at 22:13 +0100, Joe Clifford wrote:
>> Hi folks,
>>
>> I'm not a coder by any stretch of the imagination so I don't have a
>> patch unfortunately however, this commit:
>>
>> https://github.com/qemu/qemu/commit/5f77e06baa84323e5bbc96c2c7f4fe627078b210
>>
>> causes the failure of the ICH9 Universal Host Controller - 2934 and the
>> attached "Microsoft Common Controller For Windows Class"->"Xbox 360
>> Wireless Receiver for Windows" which is a pass-through usb device
>> attached to a Q35 Windows 7 x64 VM running on a Debian Jessie host.
>>
>> Reverting this commit on current master (as of today) fixes the problem.
>>
>> I spent some time today bisecting the git master code to find this.
>>
>> If you require any more information for debugging, please just let me know.
> Hmm, quick attempt to reproduce locally failed (slightly different
> device though, a microsoft nano receiver for a mouse).  It's also not
> obvious how that commit breaks usb devices.
>
> Does the device show up in device manager?
> If so:  It probably has a yellow exclamation mark?
> Any error description?
>
> Can you send the "lsusb -v" output of the device?
>
> cheers,
>    Gerd
>
Hi Gerd,

Apologies, I should have provided more information yesterday but it was 
a tired, end of the day email. Hopefully the following provides a better 
understanding. Please excuse my assumption that the problem lies with 
the commit 5f77e0. I'm guessing that the problem could actually be a 
result of a bug in my motherboard's UEFI firmware or in the OVMF 
firmware or some other Qemu code that compounds the issue.

I wasn't sure if it was appropriate to attach files to mailing list 
emails so I've uploaded them to my github account.

lspci and lsub output: 
https://gist.github.com/7hunderbug/d4c79cc7dfb928a4d7e8bd4fb9500c5a
Windows device manager screen-shots: 
https://github.com/7hunderbug/Qemu-Win7-screenshots

Overview:

Host machine:
- Debian Jessie x86_64, Intel Core i5-4570S, Z87 chipset, Nvidia GTX
670, standard 3.16.0-4-amd64 kernel (backported 4.3.0-0.bpo.1-amd64
kernel also tried)
- Host booted with UEFI firmware.
- Unmodded EVGA NVidia card with UEFI compatible ROM.
- Host OS uses Intel CPU HD4600 IGD for display.
- VGA pass-through to Windows 7 VM of NVidia card using "intel_iommu=on"
kernel command line and vfio module with IOMMU groups: works as expected.
- PCI pass-through of host audio device
- USB pass-through of three input devices, all connected to same USB 2.0
hub (mouse, keyboard, XBox wireless receiver)

Guest VM:
- Windows 7 x86_64 Q35 machine
- OVMF-pure-efi UEFI firmware (extracted from
https://www.kraxel.org/repos/jenkins/ - tried 3 different releases
including edk2.git-ovmf-x64-0-20160420.b1702.g6b17c11.noarch.rpm and
edk2.git-ovmf-x64-0-20160308.b1583.g231ad7d.noarch.rpm)
- NVidia VGA card pass-through from host
- PCI pass-through of host audio device
- USB pass-through of three input devices from host

Qemu command line:
./sources/qemu/x86_64-softmmu/qemu-system-x86_64 -machine q35 \
-serial none -parallel none -enable-kvm -name Windows \
-cpu host,kvm=off,acpi -smp sockets=1,cores=4,threads=1 -m 8192 \
-drive if=pflash,format=raw,readonly,file=vms/OVMF.fd \
-rtc base=localtime \
-readconfig ./sources/qemu/docs/q35-chipset.cfg \
-device ich9-ahci,id=ahci \
-drive if=virtio,id=drive0,file=vms/w7p64-g1.qcow2,format=qcow2 \
-drive if=virtio,id=drive1,file=vms/w7pro64-pf.qcow2,format=qcow2 \
-drive if=virtio,id=drive2,file=/dev/mapper/winf,format=raw,cache=none \
-device vfio-pci,host=01:00.0,multifunction=on,x-vga=on \
-device vfio-pci,host=01:00.1,multifunction=on \
-device vfio-pci,host=00:1b.0,multifunction=on \
-netdev type=tap,id=net0,ifname=tap1,script=no \
-device virtio-net-pci,netdev=net0,mac=52:54:xx:xx:xx:xx \
-device usb-host,bus=ich9-ehci-1.0,hostbus=3,hostport=2.1 \
-device usb-host,bus=ich9-ehci-2.0,hostbus=3,hostport=2.2 \
-device usb-host,bus=ich9-ehci-2.0,hostbus=3,hostport=2.4

The symptoms on booting the Windows 7 VM are:

*Using Qemu compiled from master:
- When booting with the device attached, sometimes the mouse and 
keyboard in the guest are unresponsive for a minute or so. Checking the 
device manager when they become responsive again shows the yellow 
warning triangle next to the first ICH9 UHCI device.
- When booting without the device attached, then running the device 
manager to watch what happens, inserting the device results in it 
appearing as normal for ~30-50 seconds before the device manager view 
refreshes and the yellow warning triangle appears next to the IC9 UHCI 
device and the Xbox USB device disappears. (See screen-shot 
w7-devman1.png and w7-devman2.png). Sometimes this results in 
unresponsive keyboard and mouse in the guest and other times not.

*Using Qemu compiled from master with commit5f77e0 reverted:
- Booting with the device attached or unattached makes no difference, in 
both scenarios the device works as expected. (See screen-shot 
w7-devman3.png)

This is reproducible for me every time, with my current Windows 7 image 
and with a freshly built image.

After researching other user's problems, I suspected the issue might be 
related to all the pass-through USB devices being manufactured by 
Microsoft however, the same problem occurs with a Logitech keyboard and 
mouse.

Regards,

Joe

Re: [Qemu-devel] Bug (Regression?) in hw/usb/hcd-uhci.c causes failure of ICH9 host controller and attached Xbox 360 Wireless Receiver

[Gerd Hoffmann]

> - When booting without the device attached, then running the device 
> manager to watch what happens, inserting the device results in it 
> appearing as normal for ~30-50 seconds before the device manager view 
> refreshes and the yellow warning triangle appears next to the IC9 UHCI 
> device and the Xbox USB device disappears. (See screen-shot 
> w7-devman1.png and w7-devman2.png). Sometimes this results in 
> unresponsive keyboard and mouse in the guest and other times not.

Ok, can you try
https://www.kraxel.org/cgit/qemu/commit/?h=work/usb-uhci-usb-host-debug&id=f8a731045bbe46a315622a1cc76afc6deb14534e
?

Probably the message gets printed when the yellow warning appears in
device manager.

thanks,
  Gerd

Re: [Qemu-devel] Bug (Regression?) in hw/usb/hcd-uhci.c causes failure of ICH9 host controller and attached Xbox 360 Wireless Receiver

[Joe Clifford]

On 21/04/16 15:17, Gerd Hoffmann wrote:
>> - When booting without the device attached, then running the device
>> manager to watch what happens, inserting the device results in it
>> appearing as normal for ~30-50 seconds before the device manager view
>> refreshes and the yellow warning triangle appears next to the IC9 UHCI
>> device and the Xbox USB device disappears. (See screen-shot
>> w7-devman1.png and w7-devman2.png). Sometimes this results in
>> unresponsive keyboard and mouse in the guest and other times not.
> Ok, can you try
> https://www.kraxel.org/cgit/qemu/commit/?h=work/usb-uhci-usb-host-debug&id=f8a731045bbe46a315622a1cc76afc6deb14534e
> ?
>
> Probably the message gets printed when the yellow warning appears in
> device manager.
>
> thanks,
>    Gerd
>
Starting the guest VM with the UHCI debug patch and the device unplugged 
results in these two lines just after it starts:

uhci_handle_td: invalid pid token 0x0, raising HCPERR
uhci_handle_td: invalid pid token 0x0, raising HCPERR

and then the same line repeated 18 times once the Windows desktop is 
reached and the device is plugged in.

When the yellow warning appears in the device manager, nothing is 
printed to stderr.

Shutdown of the guest results in 7 more identical lines.

Starting the guest with the device already plugged in is not much 
different; 20 lines of 'uhci_handle_td: invalid pid token 0x0, raising 
HCPERR' until desktop is reached, nothing when yellow exclamation 
appears, 7 more lines on shutdown.

There are various warnings printed in the host syslog that occur in line 
with the guest starting, the device being plugged in and guest shutdown 
- you can see the output here:

https://gist.github.com/7hunderbug/a1d271b0e8cab34800e046be06414ab9

Thanks,

Joe

Re: [Qemu-devel] Bug (Regression?) in hw/usb/hcd-uhci.c causes failure of ICH9 host controller and attached Xbox 360 Wireless Receiver

[Joe Clifford]

On 21/04/16 16:41, Joe Clifford wrote:
> On 21/04/16 15:17, Gerd Hoffmann wrote:
>>> - When booting without the device attached, then running the device
>>> manager to watch what happens, inserting the device results in it
>>> appearing as normal for ~30-50 seconds before the device manager view
>>> refreshes and the yellow warning triangle appears next to the IC9 UHCI
>>> device and the Xbox USB device disappears. (See screen-shot
>>> w7-devman1.png and w7-devman2.png). Sometimes this results in
>>> unresponsive keyboard and mouse in the guest and other times not.
>> Ok, can you try
>> https://www.kraxel.org/cgit/qemu/commit/?h=work/usb-uhci-usb-host-debug&id=f8a731045bbe46a315622a1cc76afc6deb14534e 
>>
>> ?
>>
>> Probably the message gets printed when the yellow warning appears in
>> device manager.
>>
>> thanks,
>>    Gerd
>>
> Starting the guest VM with the UHCI debug patch and the device 
> unplugged results in these two lines just after it starts:
>
> uhci_handle_td: invalid pid token 0x0, raising HCPERR
> uhci_handle_td: invalid pid token 0x0, raising HCPERR
>
> and then the same line repeated 18 times once the Windows desktop is 
> reached and the device is plugged in.
>
> When the yellow warning appears in the device manager, nothing is 
> printed to stderr.
>
> Shutdown of the guest results in 7 more identical lines.
>
> Starting the guest with the device already plugged in is not much 
> different; 20 lines of 'uhci_handle_td: invalid pid token 0x0, raising 
> HCPERR' until desktop is reached, nothing when yellow exclamation 
> appears, 7 more lines on shutdown.
>
> There are various warnings printed in the host syslog that occur in 
> line with the guest starting, the device being plugged in and guest 
> shutdown - you can see the output here:
>
> https://gist.github.com/7hunderbug/a1d271b0e8cab34800e046be06414ab9
>
> Thanks,
>
> Joe
>
I should add that for some unknown reason, my host machine sometimes 
changes the host bus enumeration that has the USB 2.0 hub hanging off 
it, that has the pass-through devices attached to it. It varies from Bus 
03.Port 2 to Bus 01.Port 2. Usually it is Host Bus 03.Port2 but I've no 
idea why it changes from boot to boot, possibly something to do with 
udev.....or a buggy UEFI....

My initial report showed Bus 03.Port 2, the recent debug output shows 
Bus 01.Port 2.

Re: [Qemu-devel] Bug (Regression?) in hw/usb/hcd-uhci.c causes failure of ICH9 host controller and attached Xbox 360 Wireless Receiver

[Gerd Hoffmann]

  Hi,

> Starting the guest VM with the UHCI debug patch and the device unplugged 
> results in these two lines just after it starts:
> 
> uhci_handle_td: invalid pid token 0x0, raising HCPERR
> uhci_handle_td: invalid pid token 0x0, raising HCPERR
> 
> and then the same line repeated 18 times once the Windows desktop is 
> reached and the device is plugged in.

Think I've nailed it.  Can you fetch the latest version of the git
branch (fyi: it's rebased) and try again?

thanks,
  Gerd

Re: [Qemu-devel] Bug (Regression?) in hw/usb/hcd-uhci.c causes failure of ICH9 host controller and attached Xbox 360 Wireless Receiver

[Gerd Hoffmann]

  Hi,

> I should add that for some unknown reason, my host machine sometimes 
> changes the host bus enumeration that has the USB 2.0 hub hanging off 
> it, that has the pass-through devices attached to it.

Probably because linux probes the pci devices in parallel, so whoever
finished first gets bus number one.

cheers,
  Gerd

Re: [Qemu-devel] Bug (Regression?) in hw/usb/hcd-uhci.c causes failure of ICH9 host controller and attached Xbox 360 Wireless Receiver

[joe]

On 2016-04-22 11:04, Gerd Hoffmann wrote:
> Hi,
> 
>> Starting the guest VM with the UHCI debug patch and the device 
>> unplugged
>> results in these two lines just after it starts:
>> 
>> uhci_handle_td: invalid pid token 0x0, raising HCPERR
>> uhci_handle_td: invalid pid token 0x0, raising HCPERR
>> 
>> and then the same line repeated 18 times once the Windows desktop is
>> reached and the device is plugged in.
> 
> Think I've nailed it.  Can you fetch the latest version of the git
> branch (fyi: it's rebased) and try again?
> 
> thanks,
>   Gerd

Yep, you nailed it!

I've applied 
https://www.kraxel.org/cgit/qemu/commit/?h=work/usb-uhci-usb-host-debug&id=c69e5efac15db4340b880c80f27ebe536acd1aeb 
to my local copy of current qemu git master and the problem no longer 
exists.

The stderr debugging output is also beautifully silent :)

Awesome work, many thanks.

Joe