Re: [PATCH v5 05/14] KVM: X86: Implement ring-based dirty memory tracking

[Michael S. Tsirkin]

On Tue, Mar 10, 2020 at 10:19:01AM -0400, Peter Xu wrote:
> On Tue, Mar 10, 2020 at 10:11:30AM -0400, Michael S. Tsirkin wrote:
> > On Tue, Mar 10, 2020 at 10:09:21AM -0400, Peter Xu wrote:
> > > On Tue, Mar 10, 2020 at 02:31:55AM -0400, Michael S. Tsirkin wrote:
> > > > On Mon, Mar 09, 2020 at 05:35:54PM -0400, Peter Xu wrote:
> > > > > I'll probably also
> > > > > move KVM_DIRTY_LOG_PAGE_OFFSET==0 definition to uapi/linux/kvm.h.
> > > > 
> > > > 
> > > > IMHO KVM_DIRTY_LOG_PAGE_OFFSET is kind of pointless anyway - 
> > > > we won't be able to move data around just by changing the
> > > > uapi value since userspace isn't
> > > > recompiled when kernel changes ...
> > > 
> > > Yes I think we can even drop this KVM_DIRTY_LOG_PAGE_OFFSET==0
> > > definition.  IMHO it's only a matter of whether we would like to
> > > directly reference this value in the common code (e.g., for kernel
> > > virt/kvm_main.c) or we want quite a few of this instead:
> > > 
> > > #ifdef KVM_DIRTY_LOG_PAGE_OFFSET
> > > ..
> > > #endif
> > 
> > Hmm do other arches define it to a different value?
> > Maybe I'm confused.
> > If they do then it makes sense.
> 
> Yes they can. So far with this series only x86 will define it to
> nonzero (64). But logically other archs can define it to different
> values.


Oh ok then. somehow I thought it's 0 for all arches.
Sorry about the noise, pls ignore this comment.

> 
> We can reference this to existing offsets that we've defined already
> for different archs, like KVM_COALESCED_MMIO_PAGE_OFFSET:
> 
>   - For ppc, it's defined as 1 (arch/powerpc/include/uapi/asm/kvm.h)
>   - For x86, it's defined as 2 (arch/x86/include/uapi/asm/kvm.h)
>   - ...
> 
> Thanks,
> > 
> > > I slightly prefer to not use lots of "#ifdef"s so I chose to make sure
> > > it's defined.  However I've no strong opinion on this either. So I'm
> > > open to change that if anyone insists with some reasons.
> > > 
> > > Thanks,
> > > 
> > > -- 
> > > Peter Xu
> > 
> 
> -- 
> Peter Xu