[Qemu-devel] Should new USB devices such as usb-ccid support legacy -usbdevice?

[Qemu-devel] Should new USB devices such as usb-ccid support legacy -usbdevice?

[Markus Armbruster]

When Gerd qdevified USB, he kept legacy -usbdevice working (commit
0958b4cc...).  What about new USB devices?  Should they get a legacy
syntax, too?

The only existing new device is usb-ccid, and it got one in commit
36707144.

Re: [Qemu-devel] Should new USB devices such as usb-ccid support legacy -usbdevice?

[Anthony Liguori]

On 05/13/2011 11:36 AM, Markus Armbruster wrote:
> When Gerd qdevified USB, he kept legacy -usbdevice working (commit
> 0958b4cc...).  What about new USB devices?  Should they get a legacy
> syntax, too?
>
> The only existing new device is usb-ccid, and it got one in commit
> 36707144.

What keeps -usbdevice from being a light wrapper to -device such that no 
future code is needed for this?

I know some of the names are different but presumably we could use a map 
for existing ones and pass through names for newer ones.

Regards,

Anthony Liguori

Re: [Qemu-devel] Should new USB devices such as usb-ccid support legacy -usbdevice?

[Gerd Hoffmann]

On 05/13/11 18:47, Anthony Liguori wrote:
> On 05/13/2011 11:36 AM, Markus Armbruster wrote:
>> When Gerd qdevified USB, he kept legacy -usbdevice working (commit
>> 0958b4cc...). What about new USB devices? Should they get a legacy
>> syntax, too?

Any reason to do that?  We already have a number of devices which can be 
setup via -device only, and that is perfectly fine IMHO.  Also I don't 
want extend the legacy syntax when adding new properties for new 
features to usb devices, which will add more cases where you have to use 
-device to use new features.

>> The only existing new device is usb-ccid, and it got one in commit
>> 36707144.
>
> What keeps -usbdevice from being a light wrapper to -device such that no
> future code is needed for this?

Different command line syntax.  Each device (which needs/accepts 
parameters) has a init function to handle the -usbdevice command line, 
which basically does:

   * parse old syntax
   * usb_create()
   * qdev_set_prop_*() calls
   * qdev_init

> I know some of the names are different but presumably we could use a map
> for existing ones and pass through names for newer ones.

Just the device name is easy, there is a field in USBDeviceInfo for that 
already, and in fact "-usbdevice tablet" acts like "-device usb-tablet" 
thanks to that entry ;)

cheers,
   Gerd

Re: [Qemu-devel] Should new USB devices such as usb-ccid support legacy -usbdevice?

[Markus Armbruster]

Anthony Liguori <anthony@codemonkey.ws> writes:

> On 05/13/2011 11:36 AM, Markus Armbruster wrote:
>> When Gerd qdevified USB, he kept legacy -usbdevice working (commit
>> 0958b4cc...).  What about new USB devices?  Should they get a legacy
>> syntax, too?
>>
>> The only existing new device is usb-ccid, and it got one in commit
>> 36707144.
>
> What keeps -usbdevice from being a light wrapper to -device such that
> no future code is needed for this?

What would that buy us?  Inhowfar is -usbdevice any lighter than
-device?

> I know some of the names are different but presumably we could use a
> map for existing ones and pass through names for newer ones.

Here's what an USB device needs to do for -usbdevice:

* For -usbdevice NAME (no parameters), set USBDeviceInfo member
  usbdevice_name to NAME.

* For -usbdevice NAME:PARAMS, additionally set USBDeviceInfo member
  usbdevice_init() to a function that parses PARAMS and creates the
  device.

  Legacy PARAMs use various ad hoc syntax, and that's why we have a
  callback here.

  For new devices, we could require regular syntax, and then a common
  callback would do.  The common NAME=VALUE,... syntax is the obvious
  choice, where the NAMEs are property names.

  But then -usbdevice is *exactly* like -device, except you the "usb"
  goes in a different place: "-usbdevice ccid" vs. "-device usb-ccid".
  And except that -usbdevice is crippled for some devices.  For
  instance, usb-ccid doesn't support parameters with -usbdevice, even
  though it has a qdev property.

Re: [Qemu-devel] Should new USB devices such as usb-ccid support legacy -usbdevice?

[Kevin Wolf]

Am 16.05.2011 10:33, schrieb Markus Armbruster:
> Anthony Liguori <anthony@codemonkey.ws> writes:
> 
>> On 05/13/2011 11:36 AM, Markus Armbruster wrote:
>>> When Gerd qdevified USB, he kept legacy -usbdevice working (commit
>>> 0958b4cc...).  What about new USB devices?  Should they get a legacy
>>> syntax, too?
>>>
>>> The only existing new device is usb-ccid, and it got one in commit
>>> 36707144.
>>
>> What keeps -usbdevice from being a light wrapper to -device such that
>> no future code is needed for this?
> 
> What would that buy us?  Inhowfar is -usbdevice any lighter than
> -device?

One difference is that it automagically gives you an USB host
controller, whereas with -device you need to specify one explicitly.

Kevin