Re: Xen 4.0.0 first release candidate / graphics passthru

[Weidong Han <weidong.han@intel.com>]

Pasi Kärkkäinen wrote:
> On Thu, Jan 21, 2010 at 05:52:15PM +0800, Weidong Han wrote:
>   
>> Jean Guyader wrote:
>>     
>>> 2010/1/21 Pasi Kärkkäinen <pasik@iki.fi>:
>>>   
>>>       
>>>> On Thu, Jan 21, 2010 at 09:09:35AM +0000, Jean Guyader wrote:
>>>>     
>>>>         
>>>>> 2010/1/21 Pasi Kärkkäinen <pasik@iki.fi>:
>>>>>       
>>>>>           
>>>>>> On Tue, Jan 12, 2010 at 10:05:12AM +0000, Keir Fraser wrote:
>>>>>>         
>>>>>>             
>>>>>>> On 12/01/2010 09:32, "Pasi Kärkkäinen" <pasik@iki.fi> wrote:
>>>>>>>
>>>>>>>           
>>>>>>>               
>>>>>>>>>> Does 4.0 include gpu passthru features? If not are there any updated
>>>>>>>>>> patches to try gpu passthru with 4.x?
>>>>>>>>>>                 
>>>>>>>>>>                     
>>>>>>>>> There is support (c/s 20144).
>>>>>>>>>
>>>>>>>>>               
>>>>>>>>>                   
>>>>>>>> Btw is it possible now to specify a file to load the VGA bios from?
>>>>>>>> earlier it was required to hardcode it into some header file or so.
>>>>>>>>             
>>>>>>>>                 
>>>>>>> I think it is hardcoded that qemu will take a copy of the host system's BIOS
>>>>>>> (i.e., for the primary host VGA). Other options would be easy to implement
>>>>>>> I'm sure, if someone's interested.
>>>>>>>
>>>>>>>           
>>>>>>>               
>>>>>> I was just looking at same older mails, and it seems for example Teo En Ming
>>>>>> and others who got Xen VGA passthrough working had to copy the vgabios-pt.bin
>>>>>> to tools/firmware/vgabios/ before building Xen.
>>>>>>
>>>>>> That's not very practical..
>>>>>>
>>>>>>         
>>>>>>             
>>>>> Right now the code copy the bios from 0xC0000. So if you want to pass through
>>>>> a secondary display adapter with the vga bios you are not copying the right one,
>>>>> that why this hack was for.
>>>>> Surely the code could be modified to add an option to copy the bios of
>>>>> the right card.
>>>>>
>>>>>       
>>>>>           
>>>> Ok, thanks for the info.
>>>>
>>>> Do I remember correctly that there was a problem with copying the BIOS from some cards,
>>>> or on some setups? And that's why the patch to load it from a file was made?
>>>>
>>>>     
>>>>         
>>> Yes, I think that was because of that, and I'm not sure about the
>>> intel IGD but I think that was causing some issue.
>>> The code could have a cleanup and use sysfs to get the rombios, if I
>>> works with the intel graphic card I think we should do that.
>>>
>>> Jean
>>>   
>>>       
>> vbios copied from 0xC0000 doesn't work for some graphics cards, e.g.  
>> Nvidia GS 8400, due to the vbios code is modified after initialization.  
>> So it needs to load it from file. yes, it's not practical, Intel IGD  
>> cards don't have this issue in our tests.  One possible good solution is  
>> to load vbios from expansion ROM of graphics cards, but we found  
>> expansion ROM address are invalid for some graphics cards, even using  
>> Linux upstream.  It needs to look into Linux code.
>>
>>     
>
> So at this point should be add two things:
>
> 1) option in guest cfgfile telling from which card to copy the vbios from
>   
I think adding an option in guest config file to indicate primary or 
secondary gfx assignment can cover this.
> 2) option to load the vbios from a file specified in guest cfgfile
>   
I don't prefer this. It involves much manual work for users. It's better 
to find a general solution which is also easy to use.

Regards,
Weidong

> How does that sound like? 
>
> -- Pasi
>
>