[Intel-gfx] [PATCH 0/2] extend page_track for external usage

[Daniel Vetter]

On Fri, Nov 4, 2016 at 12:22 PM, Paolo Bonzini <pbonzini at redhat.com> wrote:
>
>
> On 04/11/2016 11:29, Zhenyu Wang wrote:
>> On 2016.11.04 11:02:05 +0100, Paolo Bonzini wrote:
>>>>> On 25/10/2016 09:50, Jike Song wrote:
>>>>>> page_track is currently used only by KVM internally to protect guest
>>>>>> page tables. To be used by external user like KVMGT, the callbacks
>>>>>> should carry more information other than KVM/VCPU, and the symbols
>>>>>> should be exported.
>>>>>>
>>>>>> Jike Song (2):
>>>>>>   kvm/page_track: call notifiers with kvm_page_track_notifier_node
>>>>>>   kvm/page_track: export symbols for external usage
>>>>>>
>>>>>>  arch/x86/include/asm/kvm_page_track.h | 7 +++++--
>>>>>>  arch/x86/kvm/mmu.c                    | 6 ++++--
>>>>>>  arch/x86/kvm/page_track.c             | 8 ++++++--
>>>>>>  3 files changed, 15 insertions(+), 6 deletions(-)
>>>>>
>>>>> Thanks, applied to kvm/queue.
>>>>>
>>>>> Paolo
>>>>>
>>>>
>>>> Hi Paolo,
>>>>
>>>> I just synced kvm repo and it seemd there was a forced-update on
>>>> kvm/queue, and the commits are dropped?
>>>
>>> I temporarily reverted everything after the APICv patches.  They'll be
>>> back soon (and in kvm/next).
>>
>> Paolo, for this case, do you think it's feasible we pick them through
>> drm/i915 merge path? As currently initial KVMGT patch sets require these
>> exported symbols, that's why I ask how we should handle this dependency.
>
> Then it's actually a good thing that I dropped from kvm/queue!  You can
> certainly include these patches, but please do that through a topic branch.
>
> I've prepared a branch for you
> (git://git.kernel.org/pub/scm/virt/kvm/kvm.git branch for-kvmgt).  Once
> Linus processes my outstanding pull request, the branch will only
> include the three page-tracking patches.  Please pull that topic branch
> into your own branch, and ensure you have a merge commit when you send
> the pull request to Daniel.  The merge commit ensures that the workflow
> was correct; use --no-ff if necessary.
>
> You can do the same for Jike's patches for the KVM-VFIO device, when
> Alex reviews them, and I suppose you'll need a topic branch for mdev
> too?  I didn't know that KVMGT was planned for 4.10.  In the future,
> let's synchronize ahead so that we can prepare topic branches for you.

Ok, back from the useless wifi at plumbers, I can mail again. Zhenyu
confirmed on irc that the initial code pile only needs this. For the
cross-maintainer topic tree I prefer a formal pull request with stable
tag. Please also cc: intel-gfx on that, since I plan to merge that one
directly into i915.

Thanks, Daniel
-- 
Daniel Vetter
Software Engineer, Intel Corporation
+41 (0) 79 365 57 48 - http://blog.ffwll.ch