Re: [PATCH] kvm: Take into account the unaligned section size when preparing bitmap

[Zenghui Yu]

On 2020/12/10 10:53, zhukeqian wrote:
> On 2020/12/10 10:08, Peter Xu wrote:
> > Keqian,
> >
> > On Thu, Dec 10, 2020 at 09:46:06AM +0800, zhukeqian wrote:
> > > Hi,
> > >
> > > I see that if start or size is not PAGE aligned, it also clears areas
> > > which beyond caller's expectation, so do we also need to consider this?

Ah, I really don't anticipate this to happen but ... The question is
what's the expectation of the caller if a non-PAGE aligned @start is
given. Should we clear dirty state for the page which covers the
unaligned @start? Or just skip it?

> > Could you elaborate?
> >
> > If start_delta != 0, kvm_log_clear_one_slot() should already go the slow path.
> >
> > Thanks,
>
> Hi Peter,
>
> start_delta /= psize;
>
> If start is not PAGE aligned, then start_delta is not PAGE aligned.
> so I think the above code will implicitly extend our start to be PAGE aligned.
>
> I suggest that we should shrink the start and (start + size) to be PAGE aligned
> at beginning of this function.

I don't object to do so (though haven't checked the code carefully).
I also don't think there is much we can do given that the caller
doesn't care about which *exact pages* to clear.


Thanks,
Zenghui