[Qemu-devel] Some confusion about live migration of usb device

[Qemu-devel] Some confusion about live migration of usb device

[linzhecheng]

Hi, Gerd

I'm going through relevant codes about live migration of usb devices, it seems that we will not save/load USBpacket in any vmstate, so pending usb packets will be lost after live migration, is it a problem?

Re: [Qemu-devel] Some confusion about live migration of usb device

[gerd hoffmann]

On Thu, Sep 06, 2018 at 10:25:20AM +0000, linzhecheng wrote:
> Hi, Gerd
> 
> I'm going through relevant codes about live migration of usb devices,
> it seems that we will not save/load USBpacket in any vmstate, so
> pending usb packets will be lost after live migration, is it a
> problem?

With usb-host or usb-redir?

cheers,
  Gerd

Re: [Qemu-devel] Some confusion about live migration of usb device

[linzhecheng]

You had said that copying vmstate of usb-host is pointless, so just unpulg and plug it after migration is all right,
 but will other usb devices like usb-storage devices lose pending USBPackets then?  

> -----Original Message-----
> From: gerd hoffmann [mailto:kraxel@redhat.com]
> Sent: Thursday, September 06, 2018 8:04 PM
> To: linzhecheng <linzhecheng@huawei.com>
> Cc: CheneyLin <linzc@zju.edu.cn>; wangxin (U)
> <wangxinxin.wang@huawei.com>; qemu-devel@nongnu.org
> Subject: Re: Some confusion about live migration of usb device
> 
> On Thu, Sep 06, 2018 at 10:25:20AM +0000, linzhecheng wrote:
> > Hi, Gerd
> >
> > I'm going through relevant codes about live migration of usb devices,
> > it seems that we will not save/load USBpacket in any vmstate, so
> > pending usb packets will be lost after live migration, is it a
> > problem?
> 
> With usb-host or usb-redir?
> 
> cheers,
>   Gerd

Re: [Qemu-devel] Some confusion about live migration of usb device

[gerd hoffmann]

On Thu, Sep 06, 2018 at 12:10:08PM +0000, linzhecheng wrote:
> You had said that copying vmstate of usb-host is pointless, so just unpulg and plug it after migration is all right,
>  but will other usb devices like usb-storage devices lose pending USBPackets then?  

Ah, emulated usb devices.  The usb host adapters will re-submit any
unfinished usb transfers on the target machine.

cheers,
  Gerd

Re: [Qemu-devel] Some confusion about live migration of usb device

[CheneyLin]

> -----原始邮件-----
> 发件人: "gerd hoffmann" <kraxel@redhat.com>
> 发送时间: 2018-09-06 21:52:23 (星期四)
> 收件人: linzhecheng <linzhecheng@huawei.com>
> 抄送: "wangxin (U)" <wangxinxin.wang@huawei.com>, CheneyLin <linzc@zju.edu.cn>, "qemu-devel@nongnu.org" <qemu-devel@nongnu.org>
> 主题: Re: [Qemu-devel] Some confusion about live migration of usb device
> 
> On Thu, Sep 06, 2018 at 12:10:08PM +0000, linzhecheng wrote:
> > You had said that copying vmstate of usb-host is pointless, so just unpulg and plug it after migration is all right,
> >  but will other usb devices like usb-storage devices lose pending USBPackets then?  
> 
> Ah, emulated usb devices.  The usb host adapters will re-submit any
> unfinished usb transfers on the target machine.
You mean target vm usb controllers will re-submit them? 
And what about usb-redir, btw?
> 
> cheers,
>   Gerd
> 

Re: [Qemu-devel] Some confusion about live migration of usb device

[gerd hoffmann]

On Thu, Sep 06, 2018 at 10:23:45PM +0800, CheneyLin wrote:
> 
> 
> 
> > -----原始邮件-----
> > 发件人: "gerd hoffmann" <kraxel@redhat.com>
> > 发送时间: 2018-09-06 21:52:23 (星期四)
> > 收件人: linzhecheng <linzhecheng@huawei.com>
> > 抄送: "wangxin (U)" <wangxinxin.wang@huawei.com>, CheneyLin <linzc@zju.edu.cn>, "qemu-devel@nongnu.org" <qemu-devel@nongnu.org>
> > 主题: Re: [Qemu-devel] Some confusion about live migration of usb device
> > 
> > On Thu, Sep 06, 2018 at 12:10:08PM +0000, linzhecheng wrote:
> > > You had said that copying vmstate of usb-host is pointless, so just unpulg and plug it after migration is all right,
> > >  but will other usb devices like usb-storage devices lose pending USBPackets then?  
> > 
> > Ah, emulated usb devices.  The usb host adapters will re-submit any
> > unfinished usb transfers on the target machine.
> You mean target vm usb controllers will re-submit them? 

Yes.

> And what about usb-redir, btw?

Basically the same.  But some extra care is needed to make sure the
physical device will not see requests twice when the target starts
replaying unfinished requests.  All usb requests get a unique tag
(USBPacket->id) so the usbredir server (typically spice client) has a
chance to handle this correctly.

cheers,
  Gerd

Re: [Qemu-devel] Some confusion about live migration of usb device

[linzhecheng]

> -----Original Message-----
> From: gerd hoffmann [mailto:kraxel@redhat.com]
> Sent: Friday, September 07, 2018 2:23 PM
> To: CheneyLin <linzc@zju.edu.cn>
> Cc: linzhecheng <linzhecheng@huawei.com>; wangxin (U)
> <wangxinxin.wang@huawei.com>; qemu-devel@nongnu.org
> Subject: Re: [Qemu-devel] Some confusion about live migration of usb device
> 
> On Thu, Sep 06, 2018 at 10:23:45PM +0800, CheneyLin wrote:
> >
> >
> >
> > > -----原始邮件-----
> > > 发件人: "gerd hoffmann" <kraxel@redhat.com>
> > > 发送时间: 2018-09-06 21:52:23 (星期四)
> > > 收件人: linzhecheng <linzhecheng@huawei.com>
> > > 抄送: "wangxin (U)" <wangxinxin.wang@huawei.com>, CheneyLin
> > > <linzc@zju.edu.cn>, "qemu-devel@nongnu.org" <qemu-devel@nongnu.org>
> > > 主题: Re: [Qemu-devel] Some confusion about live migration of usb
> > > device
> > >
> > > On Thu, Sep 06, 2018 at 12:10:08PM +0000, linzhecheng wrote:
> > > > You had said that copying vmstate of usb-host is pointless, so
> > > > just unpulg and plug it after migration is all right,  but will other usb
> devices like usb-storage devices lose pending USBPackets then?
> > >
> > > Ah, emulated usb devices.  The usb host adapters will re-submit any
> > > unfinished usb transfers on the target machine.
> > You mean target vm usb controllers will re-submit them?
> 
> Yes.
> 
> > And what about usb-redir, btw?
> 
> Basically the same.  But some extra care is needed to make sure the physical
> device will not see requests twice when the target starts replaying unfinished
> requests.  All usb requests get a unique tag
> (USBPacket->id) so the usbredir server (typically spice client) has a chance to
> handle this correctly.
I got it! Thank you for patiently answering my questions.
> 
> cheers,
>   Gerd