Re: [Qemu-devel] [PATCH qom-next 3/5] pc: move apic_mapped initialization into common apic init code

[Jan Kiszka]

On 2012-05-22 11:24, Andreas Färber wrote:
> Am 22.05.2012 14:42, schrieb Igor Mammedov:
>> On 05/22/2012 12:48 PM, Jan Kiszka wrote:
>>> On 2012-05-22 07:35, Igor Mammedov wrote:
>>>> -        apic_mapped = 1;
>>>> -    }
>>>> -
>>>> -    /* KVM does not support MSI yet. */
>>>> -    if (!kvm_irqchip_in_kernel()) {
>>>> -        msi_supported = true;
>>>> -    }
>>>> -
>>>> -    if (xen_msi_support()) {
>>>> -        msi_supported = true;
>>>> -    }
>>>> -
>>>> -    return dev;
>>>> -}
>>>> -
>>>
>>> You are loosing some xen bits here. But this will collide with latest
>>> kvm pull request
>>> (http://thread.gmane.org/gmane.comp.emulators.kvm.devel/91171) anyway.
>>> You may want to base on uq/master.
>>>
>>
>> This patchset is based on Andreas' qom-next tree. Probably I should wait
>> till above mentioned kvm pull is pulled in and it aprears in qom-next.
> 
> Jan, we currently have a chaos of concurrent, colliding QOM series.
> qom-next was intended to resolve this but so far it's a rebasing patch
> queue on top of master and not a repository with stable hashes so I
> can't do PULLs myself but I could cherry-pick related patches if needed.
> 
> Igor, if you put the code movement init -> initfn into its own patch
> I'll apply it to qom-next right away. Haven't looked at the series
> in-depth yet.
> 
> We're not quite there yet with qom-next due to series and counterseries
> and lack of input on realize/QBus. My current merge plan is as follows:
> 
> * Apply QOM CPUState series part 3 (cpu_state_reset) - aggressively done
> last night, prerequisite for part 4.
> 
> * Apply the last two remaining ARM cpu_reset followup cleanups - waiting
> for one ack by PMM.
> 
> * Post QOM CPUState series part 4 (CPU_COMMON) - still fiddling with
> bisectability, hope to post today. This will show areas of conflicts wrt
> apic and x86 and is quite invasive (qom-cpu branch on GitHub).
> 
> * Mix and match patches from Paolo's and Anthony's series for realizefn.
> Hope to post a short-term compromise soon, leaving properties aside for
> now. Apply it so that we finally have a realizefn.
> 
> * Post QOM CPUState series part 5 (CPUState conditionally as device).
> WIP (qom-cpu-dev branch on GitHub), needed for hotplug IIUC and this
> will enable integration with machine reset. Doesn't depend on part 4 so far.
> 
> * Apply PMM's ARM copro series - waiting for acks, still need to
> carefully review the final CPUID movements.
> * Post realizefn implementation on top - probably to be merged after
> PMM's holiday, i.e. to master not to qom-next.
> 
> * Align part 4 with Igor's series, possibly rebase on part 5. See how
> close to 1.2 we get and how the review of all open series goes.
> 
> Whatever progress we make on qom-next, the idea is to have qom-next
> merged into master *first*, since it's getting really large. Don't know
> what KVM PULL Jan is referring to - if it's for 1.1 then I'll rebase on
> it but otherwise I expect series to get rebased onto master w/qom-next
> before sending a PULL. That's why I asked target maintainers to queue
> the patches from my part 3 in their queues, to avoid merge conflicts
> once the 1.2 window opens.

In this case, conflicts may only be caused between
http://thread.gmane.org/gmane.comp.emulators.kvm.devel/91186 (pc: Enable
MSI support at APIC level) and this particular patch (that requires some
more work anyway).

Maybe you can test-merge the KVM pull in your current tree that is
supposed to go in first and ask Avi/Marcelo to the provide uq/master
baseline on top of yours in case of conflicts. I would provide an update
of my patches on top of that afterward.

Jan

-- 
Siemens AG, Corporate Technology, CT T DE IT 1
Corporate Competence Center Embedded Linux