Re: [PATCH v3 00/15] xen/arm: Bug fixes for the vGIC

[Julien Grall <julien.grall@linaro.org>]

On 19/02/15 17:48, Ian Campbell wrote:
> On Thu, 2015-02-19 at 17:34 +0000, Julien Grall wrote:
>> On 19/02/15 17:21, Ian Campbell wrote:
>>> On Mon, 2015-02-16 at 14:50 +0000, Julien Grall wrote:
>>>
>>> All applied, thanks.
>>>
>>>> Most of this patches should be backported to Xen 4.5/Xen 4.5 (see each patch).
>>>> Although, the one in GICv2 are not critical.
>>>
>>> Do you have a handy index of which ones do/don't need backporting to
>>> save me trawling through the mails?
>>
>> The list of GICv2 patches candidate for backporting are: #10, #11, #12, #13.
>>
>> For GICv3, see my comment below.
>>
>>> Given the lack of GICv3 hardware on the market today I'm in two minds
>>> about the backports to the gicv3 functionality, especially given the
>>> number of patches involved and the amount of stuff they change. In
>>> practical terms GICv2 support isn't much more than tech-preview in 4.5
>>> anyway.
>>
>> I guess you mean GICv3 for the last one?
> 
> Yes.
> 
>> I agree that there is no hardware available, but someone may want to use
>> the latest Xen release (currently 4.5) on their GICv3 internal board.
> 
> For new hardware people should be strongly advised to use the latest
> version of Xen, not least because that is where their platform patches
> need to be based and because that is where all the bleeding edge h/w
> enablement is happening.

You may want to use a release version in order to make a product. Xen
4.5 should be able to boot on any platform, supposing you wrote the
platform file. Anyway, it's a matter of taste for this one.

>> IHMO, GICv3 was not a tech-preview on Xen 4.5. At least we never clearly
>> say it was. If it's the case we should write-down on the features list.
> 
> Whether we said so or not it is clearly the case, in practical terms,
> that the support is not production quality in 4.5. The question is then
> whether we are willing to accept large numbers of backports trying to
> remedy the situation in a stable release.
> 
> I'm willing to try applying the backport once these changes are through
> staging, but at the first sign of git cherry-pick asking me to resolve a
> non-trivial conflict I'd be inclined to declare that a sign that this
> series is too big and/or intrusive to be considering for a backport.

AFAICT we don't have any push-gate for GICv3 code. It would be nice to
have one at some point.

For the conflict, we did some renaming on Xen 4.5. Which make the
backport more tricky.


Based on the discussion, what about waiting until someone complain about
GICv3 support on Xen 4.5?

Regards,

-- 
Julien Grall