Re: Usb passthrough with upstream qemu question

[George Dunlap <George.Dunlap@eu.citrix.com>]

On Fri, Oct 4, 2013 at 3:59 PM, Fabio Fantoni <fabio.fantoni@m2r.biz> wrote:
> Il 04/10/2013 16:25, George Dunlap ha scritto:
>
>> On 04/10/13 15:00, Fabio Fantoni wrote:
>>>
>>> Il 04/10/2013 15:30, George Dunlap ha scritto:
>>>>
>>>> On 04/10/13 14:16, Fabio Fantoni wrote:
>>>>>
>>>>> Il 04/10/2013 11:59, George Dunlap ha scritto:
>>>>>>
>>>>>> On 25/09/13 15:11, Pasi Kärkkäinen wrote:
>>>>>>>
>>>>>>> On Tue, Sep 24, 2013 at 10:59:18AM +0100, George Dunlap wrote:
>>>>>>>>
>>>>>>>> On 09/23/2013 03:46 PM, Fabio Fantoni wrote:
>>>>>>>>>
>>>>>>>>> I tested my latest patches about usb controllers (1-2-3) and usb
>>>>>>>>> redirection (from Spice client) support for upstream qemu and are
>>>>>>>>> working.
>>>>>>>>> I also tested usb passthrough (from dom0) but seems that with older
>>>>>>>>> -usbdevice paraters is not working.
>>>>>>>>> I see the qemu docs/qdev-device-use.txt and based on this I tried
>>>>>>>>> the
>>>>>>>>> new parameters, for example:
>>>>>>>>>
>>>>>>>>> device_model_args=["-device","usb-host,vendorid=0x058f,productid=0x6387"]
>>>>>>>>> With new parameters usb passthrough is working, also with new usb
>>>>>>>>> controller and also with usb redirection enabled.
>>>>>>>>> I did all tests using latest xen and qemu version (1.6).
>>>>>>>>> I think should be good to change upstream qemu parameters for usb
>>>>>>>>> passthrough on libxl with the new ones.
>>>>>>>>
>>>>>>>> Yes, I think it would be better.  The main difficulty is that the
>>>>>>>> libxl parameter is basically the old-style usbdevice parameter
>>>>>>>> passed straight through.  To support the current libxl parameter
>>>>>>>> with the new qemu would mean interpreting the usbdevice parameter
>>>>>>>> and translating it into the new-style parameter.
>>>>>>>>
>>>>>>>> My USB hot-plug series begins to introduce a more reasonable
>>>>>>>> interface for USB devices. Once that's in, we can think about
>>>>>>>> introducing a similar new interface for creating domains, that
>>>>>>>> should match the new way of specifying devices better. Then we
>>>>>>>> could just say, "If you're using spice, you have to use the new
>>>>>>>> interface."
>>>>>>>>
>>>>>>>> Unfortunately, it's looking like the USB hot-plug series may not
>>>>>>>> make it into 4.4 unless someone else can pick it up (or unless we
>>>>>>>> slip the feature freeze).
>>>>>>>>
>>>>>>> Hi,
>>>>>>>
>>>>>>> Is the latest version of the usb hotplug series posted to xen-devel,
>>>>>>> or are these not-yet-posted changes in your (private) tree?
>>>>>>>
>>>>>>> Just wondering if someone else wants to take a look at that.
>>>>>>
>>>>>>
>>>>>> The latest patches I posted were version 6; the 1/2 can be found here:
>>>>>>
>>>>>> http://marc.info/?l=xen-devel&m=136638728928743
>>>>>>
>>>>>> I can't remember if I went through and made any changes as a result of
>>>>>> comments or not; My latest local version can be found here:
>>>>>>
>>>>>> git://xenbits.xen.org/people/gdunlap/xen.git
>>>>>> out/hvm-usb-hotplug.v6-plus
>>>>>>
>>>>>>  -George
>>>>>
>>>>>
>>>>> Thanks for reply, I see your patches but I didn't found changes on "xl
>>>>> create" parts.
>>>>> Please could you keep an eye on my yesterday's mail about "xl create"
>>>>> part changes?
>>>>> http://lists.xen.org/archives/html/xen-devel/2013-10/msg00157.html
>>>>> For now I'm trying to do patch for basic spice support to pv.
>>>>> After that I'll try to do a patch for changes usb qemu parameters to
>>>>> make usb passthrough full working also with usb2,usb3 and usb redirection
>>>>> enabled ifno one else can do.
>>>>
>>>>
>>>> So at the moment, USB devices specified in the config file are added to
>>>> the guest on the qemu command line.  However, most devices aren't like that:
>>>> if you specify a PCI pass-through device, it will create the guest, and then
>>>> the domain creation will call the pci add functions, just as it would if you
>>>> were hot-plugging the device afterwards.
>>>>
>>>> So the idea would be that once the new USB hotplug interface was fully
>>>> functional, then you'd do the same thing: just specify the USB controller on
>>>> the qemu command line, and add in the other USB devices afterwards (i.e.,
>>>> after qemu had come up but before actually unpausing the guest).
>>>>
>>>>  -George
>>>
>>>
>>> If I'm not wrong on what you mean, usb controller definition should be
>>> right with my patch "usbversion" while actual usbpassthrough part should be
>>> replaced/remade in order to have devices passed by usbdevice and
>>> usbdevice_list added after qemu launch with hotplug function you had already
>>> did with your patches.
>>> Did you mean something else?
>>
>>
>> I thought the "usbpassthrough" stuff had to do with spice.
>>
>> In any case, what you describe it sounds pretty close, with one caveat:
>> Because usbdevice and usbdevice_list are basically just passing through
>> strings that qemu understands on its command line, I don't think we can
>> re-use them: I think we'd have to introduce another config option that would
>> have a new interface which matches the new libxl interface better.  We'd
>> have to keep the old ones around for backwards compatibility, but we can
>> deprecate them.
>>
>>  -George
>
>
> So to be sure, you mean to keep the old parameters usbdevice and
> usbdevice_list with only usb1 (making them deprecated), and introduce a new
> parameter storing the devices usbpassthrough list (from dom0) which will use
> the new usb1-2-3 controller (with usbversion) and your usbhotplug functions?
> Is it right?

Yes, except that I think ideally you'd design the new interface not
only for pass-through devices, but for emulated devices as well --
tablets, mice, keyboard, &c.  The patches I posted should be
expandable to include things like that.

 -George