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

From: Ian Campbell <ian.campbell@citrix.com>
To: George Dunlap <George.Dunlap@eu.citrix.com>
Cc: "Jürgen Groß" <jgross@suse.com>, "Wei Liu" <wei.liu2@citrix.com>,
	"Ian Jackson" <Ian.Jackson@eu.citrix.com>,
	"George Dunlap" <george.dunlap@citrix.com>,
	"xen-devel@lists.xen.org" <xen-devel@lists.xen.org>,
	"Jim Fehlig" <jfehlig@suse.com>, "Chunyan Liu" <cyliu@suse.com>
Subject: Re: [PATCH V7 6/7] xl: add usb-assignable-list command
Date: Wed, 7 Oct 2015 12:09:48 +0100	[thread overview]
Message-ID: <1444216188.5302.333.camel@citrix.com> (raw)
In-Reply-To: <CAFLBxZY7=o96+LgKoWTA1wURdqfO-EfBPXgzGn52atbReM6kwQ@mail.gmail.com>

On Wed, 2015-10-07 at 11:10 +0100, George Dunlap wrote:

> > So IMHO xl usb-assignable-list should behave like pci-assignable-list by
> > default.
> 
> I don't think that's really suitable.

Then I'm terribly confused because I thought that is what you were
initially advocating.

[...]

> For USB, there is no "assignable" stage -- "usb-attach" will take it
> all the way from being assigned to a driver to being assigned to the
> guest.  (You can think of this as pci-attach with "seize=1" always.)
> So making "usb-assignable-list" act like "pci-assignable-list" doesn't
> actually make any sense.

Thanks. Jeurgen has also explained this.

Do you agree that adding a dummy usbback driver just for the purposes of
adding this extra "assignable" state doesn't make sense?

> > Now, maybe it should also support some sort of --all or --full or --host
> > option which lists everything, ideally with some indication as to whether
> > they are attached to usbback or not and using syntax which can just be cut
> > -and-pasted into a cfg file (without at least one of those it's just a
> > pointless reimplementation of lsusb).
> > 
> > However I think --all/full/host is an optional extra.
> 
> Juergen suggested having "usb-list" have an --all option in the v3
> discussion.  If like me you're concerned about confusing people, then
> having --all and --host is probably the best option.
> 
> Thoughts?

If there is no assignable state in usb then I guess I don't really
understand what usb-list-assignable would even be for, so I don't really
understand why anyone is arguing what semantics it should have (my initial
reply was predicated on this state existing and it therefore being useful
to discuss how the command should behave).

Given that doing something with usb-list seems most plausible _if_ we need
some sort of thing like that at all.

What would "usb-list --all" add over and above using lsusb?

I take it that as things stand in patch #5:
    # xl usb-list <vm>
will list the usb devices attached to <vm> and that:
    # xl usb-list
will list the usb devices attached to every vm, is that
correct?

So the idea would then be to add some way of listing the devices not included in "xl usb-list", which are notionally attached to dom0, but via physical USB and not PV usb.

Ian.