From mboxrd@z Thu Jan 1 00:00:00 1970 From: George Dunlap Subject: Re: RFC: Proposed libxl USB hot-plug interface Date: Wed, 3 Apr 2013 17:04:21 +0100 Message-ID: <515C5305.3080603@eu.citrix.com> References: <515C46A4.8060506@eu.citrix.com> <20828.20190.263789.332076@mariner.uk.xensource.com> Mime-Version: 1.0 Content-Type: text/plain; charset="us-ascii"; Format="flowed" Content-Transfer-Encoding: 7bit Return-path: In-Reply-To: <20828.20190.263789.332076@mariner.uk.xensource.com> List-Unsubscribe: , List-Post: List-Help: List-Subscribe: , Sender: xen-devel-bounces@lists.xen.org Errors-To: xen-devel-bounces@lists.xen.org To: Ian Jackson Cc: Marek Marczykowski , Roger Pau Monne , Ian Campbell , "xen-devel@lists.xen.org" List-Id: xen-devel@lists.xenproject.org On 03/04/13 16:46, Ian Jackson wrote: > George Dunlap writes ("RFC: Proposed libxl USB hot-plug interface"): >> OK, based on the feedback, what about an interface like the following? > This looks plausible. > >> The idea would be that long-term, AUTO would always do PV for PV VMs, >> and would somehow decide whether to do HVM or PV for HVM domains. > Right. > >> The "type" and the "union" fields are designed to allow the interface to >> be extended to include other types of devices, such as adding emulated >> tablets, mice, keyboards, usb sticks, &c. > Right. > >> "list" should return all available USB devices including the "handle" >> that can be used to remove a device. >> struct libxl_device_usb { >> uint16_t protocol; /* AUTO, PV, HVM */ >> uint16_t type; /* HOST; later can be emulated devices like tablet, >> disk, &c */ >> uint32_t backend_domain_id; /* For PVUSB */ > I don't see why the backend domain is necessarily invalid for hvm > domains. In the stub-dm, you're going to have both an emulated > interface (at the guest-dm interface) and a PV one (at the dm-backend > interface). (Not that this will necessarily be implemented right > away!) Hmm -- I hadn't thought of using qemu-usb in a stubdom. But in any case, I think the comment is still valid -- the backend_domain_id is being used for the PVUSB part of the "connection". (And if an HVM domain has a usbfront drivers, it can of course use this field as well.) > >> uint64_t handle; /* OUT: Unique (per domain) handle that must be >> used to remove a device */ > I'm not sure mixing in and out parameters in the single struct is a > good idea. Well it's particularly interesting I think to have it here for the "usb-list" function (in which case the entire structure is written by the library). We could I suppose have usb-add just put it in the return value instead of modifying the structure in-place. >> int >> union { >> struct { >> int hostbus, hostaddr, vendorid, productid; >> } host; > This needs some more documentation. Why is it a union ? What are its > other options, etc. ? This was just meant to be a brief description to talk about -- definitely not the final version. :-) Initially this would be a union because I had envisioned something like this: union { struct { blah blah } host; struct { blah blah } wacom-tablet; struct { blah blah blah } disk; }; Then it's the same functions to add/remove/list an emulated tablet, or an emulated USB disk. > I guess you can set the values to -1 to mean "wildcard" ? or LIBXL_USB_DEVICE_HOST_ANY, which is #defined to -1. :-) -George