Re: [RFC Part2 PATCH 04/30] x86/mm: split the physmap when adding the page in RMP table

[Brijesh Singh]

On 4/19/21 7:32 AM, Borislav Petkov wrote:
> On Wed, Mar 24, 2021 at 12:04:10PM -0500, Brijesh Singh wrote:
>> A write from the hypervisor goes through the RMP checks. When the
>> hypervisor writes to pages, hardware checks to ensures that the assigned
>> bit in the RMP is zero (i.e page is shared). If the page table entry that
>> gives the sPA indicates that the target page size is a large page, then
>> all RMP entries for the 4KB constituting pages of the target must have the
>> assigned bit 0.
> Hmm, so this is important: I read this such that we can have a 2M
> page table entry but the RMP table can contain 4K entries for the
> corresponding 512 4K pages. Is that correct?

Yes that is correct.


>
> If so, then there's a certain discrepancy here and I'd expect that if
> the page gets split/collapsed, depending on the result, the RMP table
> should be updated too, so that it remains in sync.

Yes that is correct. For write access to succeed we need both the x86
and RMP page tables in sync.

>
> For example:
>
> * mm decides to group all 512 4K entries into a 2M entry, RMP table gets
> updated in the end to reflect that

To my understanding, we don't group 512 4K entries into a 2M for the
kernel address range. We do this for the userspace address through
khugepage daemon. If page tables get out of sync then it will cause an
RMP violation, the Patch #7 adds support to split the pages on demand.


>
> * mm decides to split a page, RMP table gets updated too, for the same
> reason.
>
> In this way, RMP table will be always in sync with the pagetables.
>
> I know, I probably am missing something but that makes most sense to
> me instead of noticing the discrepancy and getting to work then, when
> handling the RMP violation.
>
> Or?
>
> Thx.
>