[PATCH 15/17] kvm: arm64: Get rid of fake page table levels

[Suzuki K Poulose]

On 12/04/16 13:14, Christoffer Dall wrote:
> On Mon, Apr 11, 2016 at 03:33:45PM +0100, Suzuki K Poulose wrote:
>> On 08/04/16 16:05, Christoffer Dall wrote:
>>> On Mon, Apr 04, 2016 at 05:26:15PM +0100, Suzuki K Poulose wrote:
>>

>>>> -#if PGDIR_SHIFT > KVM_PHYS_SHIFT
>>>> -#define PTRS_PER_S2_PGD_SHIFT	0
>>>> -#else
>>>> -#define PTRS_PER_S2_PGD_SHIFT	(KVM_PHYS_SHIFT - PGDIR_SHIFT)
>>>> -#endif
>>>> -#define PTRS_PER_S2_PGD		(1 << PTRS_PER_S2_PGD_SHIFT)
>>>> +#define STAGE2_PGTABLE_LEVELS		ARM64_HW_PGTABLE_LEVELS(KVM_PHYS_SHIFT - 4)
>>>>
>>>>   /*
>>>> - * If we are concatenating first level stage-2 page tables, we would have less
>>>> - * than or equal to 16 pointers in the fake PGD, because that's what the
>>>> - * architecture allows.  In this case, (4 - CONFIG_PGTABLE_LEVELS)
>>>> - * represents the first level for the host, and we add 1 to go to the next
>>>> - * level (which uses contatenation) for the stage-2 tables.
>
> just noticed: s/contatenation/concatenation/

Thanks for catching that. Will fix it.

>>> which case this should be reworded to just state the assumptions and why
>>> this is a good assumption.
>>>
>>> (If my assumptions are wrong here, then there are also weird cases where
>>> the host does huge pages at the PMD level and we don't.  Not sure I can
>>> see the full ramifications of that.)
>>
>> I am sorry, I didn't get your point about the PMD level.
>>
>
> Right, I expressed that terribly, and I may have gotten myself confused
> when writing that.
>
> My concern is this: if the number of levels between the host and stage-2
> are different, and the host uses huge pmd mappings (either THP or huge
> tlb fs), then do we always do the right thing for stage-2 tables, even
> if we support the case with more levels in Stage-2 than on the host?

Yes. We are safe with PMD (level 2). In the worst case we will have a 2level
page table at host and say 4 at stage2. In either case, we have a PMD level
(which in host is folded to PGD) and both will use section mapping at level 2
for huge pmd. Also pmd_huge() doesn't care if the PMD is folded or not. So,
we are fine with that. I hope that answers your question.

>
> Thanks for trying to parse my crytptic and potentially nonsensical
> questions.

No no, they make absolute sense. I had gone through these questions myself initially
when I wrote the series, so its good to share them :-).

Btw, I have rebased my series to kvmarm and have addressed the comments. I will post
them after a round of testing.

Cheers
Suzuki