Re: [PATCH V7 6/7] xl: add usb-assignable-list command

[Juergen Gross <jgross@suse.com>]

On 10/07/2015 05:03 PM, George Dunlap wrote:
> On 07/10/15 15:47, Juergen Gross wrote:
>> On 10/07/2015 04:35 PM, George Dunlap wrote:
>>> On Wed, Oct 7, 2015 at 3:26 PM, Juergen Gross <jgross@suse.com> wrote:
>>>> On 10/07/2015 04:05 PM, Ian Campbell wrote:
>>>>>
>>>>> On Wed, 2015-10-07 at 15:54 +0200, Juergen Gross wrote:
>>>>>
>>>>>> Hmm, technically all unassigned USB-devices are usable from Dom0.
>>>>>> So why
>>>>>> not list them there.
>>>>>
>>>>>
>>>>> I think you'd at least want to distinguish USB devices available to
>>>>> dom0
>>>>> as
>>>>> via a PCI host controller from those which are available via pvusb.
>>>>
>>>>
>>>> Yeah, the non-pvusb devices should be listed as "assignable" if they
>>>> are handled by a driver domain.
>>>>
>>>>> This also doesn't quite work when driver domains are in the picture.
>>>>
>>>>
>>>> --host does?
>>>>
>>>> TBH: driver domains seem to be a reason to not support the --host
>>>> option. So the options are --all or no special option. I'm still
>>>> fine with both.
>>>
>>> I think the idea is that --host doesn't pretend to.  But yes, if we
>>> made it clear which devices were *assigned* to dom0 from another VM,
>>> and which were *available from* dom0 to be assigned, then actually
>>> that could be extended at some point in the future to include devices
>>> available *from* driver domains.
>>>
>>> Alternately, perhaps we should just have
>>> # xl usb-available [domain]
>>
>> Would work for me.
>>
>> Wait, wasn't there a similar command suggested? Something like
>> "xl usb-assignable-list"? ;-)
>
> Indeed there was!  A very useful command with a very sensible name...
> unfortunately one with a very *similar* name to another command but with
> different functionality.  And this similarity even confused one of the
> tools maintainers, so would almost certainly confuse our users. :-)
>
>>> Which will tell you the usb devices available to be assigned from that
>>> domain; or all available usb devices from all driver domains.  And of
>>> course for the time being it only works on domain 0, but future
>>> functionality could implement it for driver domains as well.
>>>
>>>> BTW: Do Xen concepts allow to assign a USB-device via pvUSB from a
>>>> driver domain to Dom0?
>>>
>>> Sure, why not?
>>
>> With driver domains the picture is becoming more complicated.
>>
>> The main question now is: What is the real purpose of "xl usb-list"?
>> Should it list the devices which are assigned to a domain, or should it
>> list the devices which are usable by a domain (probably attributed
>> whether the device is assignable to other domains)?
>
> For *all* of our other -list commands relating to guest devices --
> pci-list, disk-list, network-list, vtpm-list, channel-list, tmem-list --
> the list command shows the virtual devices *actually assigned to the
> VM*.   Based on that, the core information that any user will *expect*
> from "usb-list" will be "what usb devices are assigned to this VM".
>
> I'm not opposed to extending usb-list to include extra useful
> information; but it should be made clear to the user which information
> is the "core expected" information (i.e., what usb devices are assigned
> to this VM) and what is "extra" information (i.e., what usb devices are
> available to be assigned *from* this VM).
>
> So I think we have 3 options that people don't find objectionable:
>
> 1. Add --all and --host options to usb-list. (Or perhaps, --available.)
>
> 2. Have usb-list also list "available" devices from driver domains,
> clearly marked distinct from devices *assigned to* that driver domain
> from another domain, and to begin with only working for dom0.
>
> 3. A new command, "usb-available" to list available devices from driver
> domains, to begin with only working for dom0.
>
> I think I'd go for #3, but mostly I think we need to decide *something*
> so we can move forward.

+1 for both. :-)

Juergen