Re: [PATCH] libxl: fix cirrus vga video memory setting with upstream qemu

[Fabio Fantoni <fabio.fantoni@m2r.biz>]

Il 09/05/2014 16:24, Fabio Fantoni ha scritto:
> Il 09/05/2014 16:03, Stefano Stabellini ha scritto:
>> On Fri, 9 May 2014, Ian Campbell wrote:
>>> On Fri, 2014-05-09 at 10:01 +0200, Fabio Fantoni wrote:
>>>> Il 08/05/2014 17:19, Ian Campbell ha scritto:
>>>>> On Thu, 2014-05-08 at 17:03 +0200, Fabio Fantoni wrote:
>>>>>> Il 08/05/2014 13:33, Ian Campbell ha scritto:
>>>>>>> On Thu, 2014-05-08 at 12:41 +0200, Fabio Fantoni wrote:
>>>>>>>
>>>>>>>> About qemu version when vgamem_mb property is added is qemu 1.3, I
>>>>>>>> already did detailed reply about it in stdvga patch:
>>>>>>>> http://lists.xen.org/archives/html/xen-devel/2014-05/msg00259.html
>>>>>>>> Or you mean something other?
>>>>>>> "in the commit message" is the bit you keep missing.
>>>>>> I not found 1399030886.32736.63.camel@kazak.uk.xensource.com with 
>>>>>> google
>>>>>> or thunderbird on mail archive, I not understand what post you 
>>>>>> refer :(
>>>>> mid.gmane.org is a good way to find these things. e.g.:
>>>>> http://mid.gmane.org/<1399030886.32736.63.camel@kazak.uk.xensource.com> 
>>>>>
>>>>>
>>>>> Note that the <> which I quoted *are* part of the message id.
>>>>>
>>>>> The questions there were:
>>>>>           Is this change correct for all versions of mainline qemu 
>>>>> which
>>>>>           people might be using with Xen?
>>>> vgamem_mb property was added in qemu 1.3 (same for stdvga patch),
>>> How was the video RAM size controller in qemu prior to that? Or was it
>>> not possible?
>>>
>>>>   xen
>>>> 4.4 from source use qemu 1.6, and from distributions package is newer
>>>> all case I saw, then FWIK there should be no problems. Xen 4.4 if I
>>>> remember good change other important qemu value for hvm domUs that
>>>> require qemu>=1.6. (1.6.1 because 1.6.0 have critical regression 
>>>> for all
>>>> hvm domUs)
>>> Stefano/Anthony -- can you confirm that we already depend on qemu >=
>>> 1.6.x (or otherwise that it is OK to depend on qemu >= 1.3)?
>>  From upstream QEMU POV anything older than 1.7 is "ancient" and
>> unmaintained.
>>
>>  From Xen POV the upstream QEMU version that we released with Xen 4.2 
>> was
>> based on QEMU v1.0.1, the one we released with Xen 4.3 was based on
>> v1.3.0.
>>
>> Given that upstream QEMU with Xen 4.2 was a tech preview and that if I
>> am not mistaken we don't maintain Xen 4.2 anymore, it should be OK to
>> depend on QEMU >= 1.3.
>
> If I remember good the machine/pv qemu parameters part changed by 
> Anthony in xen 4.4 require at least qemu 1.6.
> In that case xen >=4.4 need already qemu>=1.6.1 (1.6.1 because 1.6.0 
> have critical regression for all
> hvm domUs)

Found:
http://xenbits.xen.org/gitweb/?p=xen.git;a=commitdiff;h=2bc047635b51abd41c917aa2b813211ee0de2c38
http://git.qemu.org/?p=qemu.git;a=commit;h=a97d6fe6fbb97630d77253d20bdce78f76d01850
 From a fast look seems that require qemu>=1.6 only for the case of 
xen_platform_pci=0

I remember also another critical regression in qemu 1.5 and solved in 
qemu 1.6 that cause qemu crash in some cases hvm domUs with high ram 
(even without pci passthrough).

These are probably a good reasons to recommend qemu>=1.6.1 with xen>=4.4

>
>>
>>
>>> An ack from you guys on this change would be appreciated.
>> I think the change is OK but I wonder if we should write down somewhere,
>> maybe in the 4.5 release notes or on the wiki, that we depend on QEMU >=
>> 1.3.
>>
>>
>>> It's also been proposed that we backport this to stable branches, how
>>> far back would it be safe/acceptable for us to go with that?
>>> [...]
>> Anything newer than 4.2 should be OK.
>