[new PV device]

[new PV device]

[Vincenzo Maffione]

Hi,
   I would like to design a new paravirtual device for Xen PV guests
(under Linux),
for research purposes.
This device should be a "simplification" of netfront/netback in that
the interface
with the domU/dom0 O.S. is not the normal TCP/IP stack, but something way
simpler (however, this is not so important w.r.t. my question).

Looking at the netback code
     linux:drivers/net/xen-netback/xenbus.c
I've run into some "hotplug scripts" support code.
The netfront code, on the other hand, has no hotplug-related code.

I don't really understand what is the purpose of this hotplug support
(Is it something
related to the vif-* scripts?), but I'm wondering whether I can get
rid of this code for my
"simplified" device.

It would be enough for me to be able to do the following:
  1) manually load my backend driver kernel module in Dom0
  2) manually load my frontend driver kernel module in DomU
  3) being able to use my device (think as I have userspace access to
the device through a character device).
I don't want/need any configuration scripts in Dom0 nor DomU (no sw
bridges or routing
tables to configure).  Just plain simple Dom0 <==> DomU buffer exchange.

Is it possibile to achieve this without any "hotplug code"? Or do I need some
other "software pieces" apart from the frontend and backend drivers?


Thanks in advance.

Best regards,
  Vincenzo Maffione

Re: [new PV device]

[Wei Liu]

On Mon, Apr 14, 2014 at 05:45:00PM +0200, Vincenzo Maffione wrote:
> Hi,
>    I would like to design a new paravirtual device for Xen PV guests
> (under Linux),
> for research purposes.
> This device should be a "simplification" of netfront/netback in that
> the interface
> with the domU/dom0 O.S. is not the normal TCP/IP stack, but something way
> simpler (however, this is not so important w.r.t. my question).
> 
> Looking at the netback code
>      linux:drivers/net/xen-netback/xenbus.c
> I've run into some "hotplug scripts" support code.
> The netfront code, on the other hand, has no hotplug-related code.
> 
> I don't really understand what is the purpose of this hotplug support
> (Is it something
> related to the vif-* scripts?), but I'm wondering whether I can get

Yes, you're right.

> rid of this code for my
> "simplified" device.
> 

Of course you can, if you don't need user space event handling or don't
mind doing some tedious configurations by hand.

> It would be enough for me to be able to do the following:
>   1) manually load my backend driver kernel module in Dom0
>   2) manually load my frontend driver kernel module in DomU
>   3) being able to use my device (think as I have userspace access to
> the device through a character device).
> I don't want/need any configuration scripts in Dom0 nor DomU (no sw
> bridges or routing
> tables to configure).  Just plain simple Dom0 <==> DomU buffer exchange.
> 
> Is it possibile to achieve this without any "hotplug code"? Or do I need some

Yes, that's possible.

You can have a look at other Xen PV drivers as well. Probably Xen PV
framebuffer can be a good starting point.

Xen wiki also has some general information on PV driver.

Wei.

Re: [new PV device]

[Vincenzo Maffione]

2014-04-14 23:46 GMT+02:00 Wei Liu <wei.liu2@citrix.com>:
> On Mon, Apr 14, 2014 at 05:45:00PM +0200, Vincenzo Maffione wrote:
>> Hi,
>>    I would like to design a new paravirtual device for Xen PV guests
>> (under Linux),
>> for research purposes.
>> This device should be a "simplification" of netfront/netback in that
>> the interface
>> with the domU/dom0 O.S. is not the normal TCP/IP stack, but something way
>> simpler (however, this is not so important w.r.t. my question).
>>
>> Looking at the netback code
>>      linux:drivers/net/xen-netback/xenbus.c
>> I've run into some "hotplug scripts" support code.
>> The netfront code, on the other hand, has no hotplug-related code.
>>
>> I don't really understand what is the purpose of this hotplug support
>> (Is it something
>> related to the vif-* scripts?), but I'm wondering whether I can get
>
> Yes, you're right.
>
>> rid of this code for my
>> "simplified" device.
>>
>
> Of course you can, if you don't need user space event handling or don't
> mind doing some tedious configurations by hand.
>
>> It would be enough for me to be able to do the following:
>>   1) manually load my backend driver kernel module in Dom0
>>   2) manually load my frontend driver kernel module in DomU
>>   3) being able to use my device (think as I have userspace access to
>> the device through a character device).
>> I don't want/need any configuration scripts in Dom0 nor DomU (no sw
>> bridges or routing
>> tables to configure).  Just plain simple Dom0 <==> DomU buffer exchange.
>>
>> Is it possibile to achieve this without any "hotplug code"? Or do I need some
>
> Yes, that's possible.
>
> You can have a look at other Xen PV drivers as well. Probably Xen PV
> framebuffer can be a good starting point.
>
> Xen wiki also has some general information on PV driver.
>
> Wei.

Thank you.
So Xen is really a minimal hypervisor, if adding a new device doesn't
require modifications in the hypervisor itself (but only some new
kernel modules for the domains). Great!

I'm not able to find the framebuffer backend in the linux kernel. Is
it somewhere else?

-- 
Vincenzo Maffione

Re: [new PV device]

[Wei Liu]

[...]
> 
> Thank you.
> So Xen is really a minimal hypervisor, if adding a new device doesn't
> require modifications in the hypervisor itself (but only some new
> kernel modules for the domains). Great!
> 
> I'm not able to find the framebuffer backend in the linux kernel. Is
> it somewhere else?
> 

video/xen-fbfront.c

but the backend is in QEMU, so this might not fit your need, sorry.

In any case you can go to drivers/ and do "find | grep xen" to get a
list of Xen drivers.

Wei.

> -- 
> Vincenzo Maffione

Re: [new PV device]

[Konrad Rzeszutek Wilk]

[-- Attachment #1: Type: text/plain, Size: 822 bytes --]

On Tue, Apr 15, 2014 at 11:20:47AM +0100, Wei Liu wrote:
> [...]
> > 
> > Thank you.
> > So Xen is really a minimal hypervisor, if adding a new device doesn't
> > require modifications in the hypervisor itself (but only some new
> > kernel modules for the domains). Great!
> > 
> > I'm not able to find the framebuffer backend in the linux kernel. Is
> > it somewhere else?
> > 
> 
> video/xen-fbfront.c
> 
> but the backend is in QEMU, so this might not fit your need, sorry.

There was an standalone frame buffer at some point.

See attached.
> 
> In any case you can go to drivers/ and do "find | grep xen" to get a
> list of Xen drivers.
> 
> Wei.
> 
> > -- 
> > Vincenzo Maffione
> 
> _______________________________________________
> Xen-devel mailing list
> Xen-devel@lists.xen.org
> http://lists.xen.org/xen-devel

[-- Attachment #2: xenfb.tgz --]
[-- Type: application/gzip, Size: 11776 bytes --]

[-- Attachment #3: Type: text/plain, Size: 126 bytes --]

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

Re: [new PV device]

[Roger Pau Monné]

On 14/04/14 17:45, Vincenzo Maffione wrote:
> Hi,
>    I would like to design a new paravirtual device for Xen PV guests
> (under Linux),
> for research purposes.
> This device should be a "simplification" of netfront/netback in that
> the interface
> with the domU/dom0 O.S. is not the normal TCP/IP stack, but something way
> simpler (however, this is not so important w.r.t. my question).
> 
> Looking at the netback code
>      linux:drivers/net/xen-netback/xenbus.c
> I've run into some "hotplug scripts" support code.
> The netfront code, on the other hand, has no hotplug-related code.
> 
> I don't really understand what is the purpose of this hotplug support
> (Is it something
> related to the vif-* scripts?), but I'm wondering whether I can get
> rid of this code for my
> "simplified" device.

This hooks were used by xend, which relies on udev for launching hotplug
scripts. libxl launches scripts directly, so if you are adding a new
device, and don't need xend compatibility, you can get rid of them AFAIK.

Roger.