[vfio-users] passthrough usb controler pci-e card ?

[thibaut noah]

After some trys i managed to run the vm after using the script for binding.
Problem is at each reboot obviously the card looses vfio-pci kernel, i try
adding the vendor and product id into modprob.d/local.conf (after ids for
gpus) but it didn't work, why it didn't work is beyond my range of
expertise, i assumed there would be no problem doing this.

2016-01-15 12:59 GMT+01:00 Quentin Deldycke <quentindeldycke at gmail.com>:

> Was the card used by a driver before the script?
>
> The bug possibly occurs because the driver it was on left the card in an
> incorrect state.
>
> Try to plug this device directly at boot to vfio-pci.
> On 15 Jan 2016 12:02 pm, "thibaut noah" <thibaut.noah at gmail.com> wrote:
>
>> Hum, the card is already linked to vfio-pci with the script you gave me,
>> did i miss something?
>>
>> 2016-01-15 11:55 GMT+01:00 Quentin Deldycke <quentindeldycke at gmail.com>:
>>
>>> If it is the case, try to boot with the card directly binded to
>>> vfio-pci. I think on aw blog there is an explenation on how to do it.
>>> (I don't use such, i manage other ways :P But to blacklist xhci is a bad
>>> idea for you i think.)
>>>
>>>
>>>
>>> --
>>> Deldycke Quentin
>>>
>>>
>>> On 15 January 2016 at 11:42, thibaut noah <thibaut.noah at gmail.com>
>>> wrote:
>>>
>>>> Hum, the card doesn't need linux support right? Since it will only be
>>>> use by the vm
>>>>
>>>> 2016-01-15 11:41 GMT+01:00 Quentin Deldycke <quentindeldycke at gmail.com>
>>>> :
>>>>
>>>>> Don't know if these things are affected by pci reset.
>>>>>
>>>>> I did not encounter such problems.
>>>>>
>>>>> --
>>>>> Deldycke Quentin
>>>>>
>>>>>
>>>>> On 15 January 2016 at 11:40, thibaut noah <thibaut.noah at gmail.com>
>>>>> wrote:
>>>>>
>>>>>> Seems that i cannot boot the vm with the card attached through the
>>>>>> xml file in libvirt, i stay on the loading screen (tianocore whathever).
>>>>>> Anybody knows this issue?
>>>>>>
>>>>>> 2016-01-15 10:59 GMT+01:00 thibaut noah <thibaut.noah at gmail.com>:
>>>>>>
>>>>>>> thanks for the help, the output after launching the script is
>>>>>>> totally weird :
>>>>>>> new_id invalid argument
>>>>>>> bind no such device
>>>>>>>
>>>>>>> But the kernel in use is vfio-pci, gonna try the vm asap, thanks for
>>>>>>> the help :)
>>>>>>>
>>>>>>> 2016-01-15 10:50 GMT+01:00 Quentin Deldycke <
>>>>>>> quentindeldycke at gmail.com>:
>>>>>>>
>>>>>>>> Ah no, sorry -_-'
>>>>>>>>
>>>>>>>> This is the vendor id / product id of the pci device.
>>>>>>>>
>>>>>>>> This part could be ommited, but not sure. I let it in the script so
>>>>>>>> better that it stay there =)
>>>>>>>>
>>>>>>>> --
>>>>>>>> Deldycke Quentin
>>>>>>>>
>>>>>>>>
>>>>>>>> On 15 January 2016 at 10:48, thibaut noah <thibaut.noah at gmail.com>
>>>>>>>> wrote:
>>>>>>>>
>>>>>>>>> Just so we're clear when you do echo "1b21 0612" | sudo tee -a
>>>>>>>>> /sys/bus/pci/drivers/vfio-pci/new_id
>>>>>>>>> you're creating a new id for the device in vfio-pci right? This
>>>>>>>>> number can be totally random?
>>>>>>>>>
>>>>>>>>> 2016-01-15 10:44 GMT+01:00 Quentin Deldycke <
>>>>>>>>> quentindeldycke at gmail.com>:
>>>>>>>>>
>>>>>>>>>> To do this dynamically, i use scripts which in sum do:
>>>>>>>>>>
>>>>>>>>>> Unbinding:
>>>>>>>>>> echo 0000:0a:00.0 | sudo tee
>>>>>>>>>> /sys/bus/pci/devices/0000:0a:00.0/driver/unbind
>>>>>>>>>> It means => on the driver actually used by 0000:0a:00.0, unbind
>>>>>>>>>> the device 0000:0a:00.0
>>>>>>>>>>
>>>>>>>>>> Binding to vfio:
>>>>>>>>>> echo "1b21 0612" | sudo tee -a
>>>>>>>>>> /sys/bus/pci/drivers/vfio-pci/new_id
>>>>>>>>>> echo "0000:0a:00:0" | sudo tee -a
>>>>>>>>>> /sys/bus/pci/drivers/vfio-pci/bind
>>>>>>>>>>
>>>>>>>>>>
>>>>>>>>>> I use them for:
>>>>>>>>>>  - nvme device
>>>>>>>>>>  - sata controller
>>>>>>>>>>
>>>>>>>>>> You could do same for GPU. But actually, it break dpm, so you
>>>>>>>>>> have shitty performance.
>>>>>>>>>> It is usefull for me to bind back to host as the nvme device
>>>>>>>>>> seems to be lost with vfio-pci and host sleep. (can be recovered with
>>>>>>>>>> loading nvme)
>>>>>>>>>>
>>>>>>>>>> --
>>>>>>>>>> Deldycke Quentin
>>>>>>>>>>
>>>>>>>>>>
>>>>>>>>>> On 15 January 2016 at 10:37, thibaut noah <thibaut.noah at gmail.com
>>>>>>>>>> > wrote:
>>>>>>>>>>
>>>>>>>>>>> I actually have no clue about how to unbind the device from xhci
>>>>>>>>>>> driver and bind it to vfio.
>>>>>>>>>>> It was explain in some tutorials for graphic cards but not for
>>>>>>>>>>> other devices so i'm totally lost.
>>>>>>>>>>>
>>>>>>>>>>>
>>>>>>>>>>> 2016-01-15 10:35 GMT+01:00 Quentin Deldycke <
>>>>>>>>>>> quentindeldycke at gmail.com>:
>>>>>>>>>>>
>>>>>>>>>>>> There is no difference.
>>>>>>>>>>>>
>>>>>>>>>>>> lspci to find your card, you bind it to vfio-pci and use
>>>>>>>>>>>> virt-manager to add the card.
>>>>>>>>>>>>
>>>>>>>>>>>> Should work out of the box. Juste take care that this device is
>>>>>>>>>>>> unbind from your xhci driver on host. then binded to vfio-pci.
>>>>>>>>>>>>
>>>>>>>>>>>> Take care of the iommu groups like the graphic card.
>>>>>>>>>>>>
>>>>>>>>>>>> --
>>>>>>>>>>>> Deldycke Quentin
>>>>>>>>>>>>
>>>>>>>>>>>>
>>>>>>>>>>>> On 15 January 2016 at 10:32, thibaut noah <
>>>>>>>>>>>> thibaut.noah at gmail.com> wrote:
>>>>>>>>>>>>
>>>>>>>>>>>>> Hello, i'm currently trying to passthrough an usb card
>>>>>>>>>>>>> controler to solve my usb issues (like audio glitches on headset and allow
>>>>>>>>>>>>> plug and play), problem is it seems that no one did tutorials for this and
>>>>>>>>>>>>> i'm a bit lost, only tutorial i found was to passthrough an internal usb
>>>>>>>>>>>>> controler.
>>>>>>>>>>>>> Anybody got some documentation to share?
>>>>>>>>>>>>>
>>>>>>>>>>>>> _______________________________________________
>>>>>>>>>>>>> vfio-users mailing list
>>>>>>>>>>>>> vfio-users at redhat.com
>>>>>>>>>>>>> https://www.redhat.com/mailman/listinfo/vfio-users
>>>>>>>>>>>>>
>>>>>>>>>>>>>
>>>>>>>>>>>>
>>>>>>>>>>>
>>>>>>>>>>
>>>>>>>>>
>>>>>>>>
>>>>>>>
>>>>>>
>>>>>
>>>>
>>>
>>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://listman.redhat.com/archives/vfio-users/attachments/20160115/8a82e99f/attachment.htm>