Re: [Xen-devel] [PATCH RFC V2 3/5] libxl: add pvusb API

[George Dunlap <george.dunlap@xxxxxxxxxxxxx>]

On 03/09/2015 09:39 AM, Ian Campbell wrote:
> On Fri, 2015-03-06 at 16:50 +0000, George Dunlap wrote:
>>> +libxl_device_usb *
>>> +libxl_device_usb_assignable_list(libxl_ctx *ctx, int *num)
>>
>> I'm a bit ambivalent about this one.  For people using xl, "lsusb"
>> should be just fine.  Is anyone building a toolstack on top of libxl
>> directly going to need this functionality?  Would libvirt use this
>> functionality, for instance, or would it use libusb?  Or just leave
>> that to the caller?
> 
> Do USB devices not need to be bound to some sort of placeholder driver
> (a la xen-pciback) or at least unbound from their actual driver before
> they can be assigned to a guest?
> 
> If so then I would expect a function such as this to list the subset of
> devices which have been correctly configured for USB passthrough, which
> would be a subset of the output of lsusb I think.

Are you saying you think this is what the function *should* do, or are
you saying that you think it should be renamed because that's what the
name would lead you to expect it did?

This code, following xend's lead, seems to do the plugging stuff
automatically; similar to the "seize=1" flag I added for PCI devices.  I
think this is probably the best interface here; the reason it's not
enabled by default in the PCI case is that it's so dangerous (you might
yank out your own hard disk by accident).

Given this "auto-seize" is what xend did, and what qemu does (and what
VirtualBox does, and what presumably every other virtualization-related
software does), I think adding the "make assignable" step would be
unnecessary and confusing.

 -George

_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxx
http://lists.xen.org/xen-devel