[Spice-devel] [Qemu-devel] seamless migration with spice

[Hans de Goede]

Hi,

On 03/15/2012 03:22 PM, Alon Levy wrote:
> On Thu, Mar 15, 2012 at 01:36:23PM +0100, Hans de Goede wrote:
>> Hi,
>>
>> On 03/15/2012 01:11 PM, Yonit Halperin wrote:
>>> On 03/13/2012 09:40 AM, Gerd Hoffmann wrote:
>>>> Hi,
>>>>
>>>>> It is not easy when you have 2 components, and it is much less easy when
>>>>> you have 3 or 4 components. So why make it more complicated if you can
>>>>> avoid it. Especially since there is no functional reason for making the
>>>>> qemu/client capabilities/versions dependent on the server internal data.
>>>>
>>>> qemu has ways to handle compatibility in the vmstate format. We can use
>>>> those capabilities. That of course requires exposing the structs to be
>>>> saved to qemu and adds some complexity to the qemu<->  spice interface.
>>>>
>>>> What session state is needed by the target?
>>>> What of this can be negotiated between client and target host without
>>>> bothering the source?
>>>> What needs be transfered from source to target, either directly or via
>>>> client?
>>>>
>>>>>> If this is a hard requirement then using the vmstate channel isn't going
>>>>>> to work. The vmstate is a one-way channel, no way to negotiate anything
>>>>>> between source and target.
>>>>>>
>>>>> We can do this via the client.
>>>>
>>>> Then you can send the actual state via client too.
>>>> Out-of-band negotiation for the blob send via vmstate scares me.
>>>>
>>>> Can we please start with a look at which state we actually have to send
>>>> over?
>>> Ok, I can take the display and sound channels.
>>> Alon, can you take the smartcard?
> Smartcard is actually using spicevmc state in qemu as well - the qemu
> part is the same as the usb part, uses spicevmc chardev. And there is
> already vmstate for the ccid-card-passthru device. In flight messages
> can be dealt with the same way Hans suggested for spicevmc channel as
> well.
>
> The channel itself has a buffer it needs to save unless we go the same
> route Hans suggested and flush it. IIRC there are messages for doing
> that in the protocol already - server asks client to get ready for
> migration, client notifies server when it is ready. Server can pass all
> messages it has to write to client before sending it PREPARE_TO_MIGRATE,
> then it can push anything it gets from the client to qemu (hmm, that
> could take an undefined amount of time if spicevmc/usb ccid are not
> processing it fast enough), while not polling qemu for new stuff, and
> the client shouldn't be sending anything new anyway after the
> PREPARE_TO_MIGRATE.

There seems to be more state in spice/server/smartcard.c then just
buffers, ie if a reader is attached or not. Then again that seems
to reflect whether a client is connected to the channel or not,
so you're probably right.

Hmm, interesting you allow for multiple smartcard chardevs in theory,
but then in smartcard_char_device_add_to_readers you unconditionally
call smartcard_init which does:

SmartCardChannel *g_smartcard_channel;

static void smartcard_init(void)
{
     ChannelCbs channel_cbs = { NULL, };
     ClientCbs client_cbs = { NULL, };

     ASSERT(!g_smartcard_channel);

<snip>

     g_smartcard_channel = (SmartCardChannel*)red_channel_create(sizeof(...

So in essence you allow only 1 smartcard chardev, and registering a
second will trigger the ASSERT (if enabled), or override
g_smartcard_channel ...

Not really migration related just something I noticed.

Regards,

Hans