Re: [PATCH] x86/traps: Rework #PF[Rsvd] bit handling

[Andrew Cooper <andrew.cooper3@xxxxxxxxxx>]

On 19/05/2020 17:09, Jan Beulich wrote:
> On 19.05.2020 17:33, Andrew Cooper wrote:
>> On 19/05/2020 15:48, Jan Beulich wrote:
>>> On 19.05.2020 16:11, Andrew Cooper wrote:
>>>> Given that shadow frames are limited to 44 bits anyway (and not yet
>>>> levelled safely in the migration stream), my suggestion for fixing this
>>>> was just to use one extra nibble for the extra 4 bits and call it done.
>>> Would you remind(?) me of where this 44-bit restriction is coming
>>> from?
>> From paging_max_paddr_bits(),
>>
>> /* Shadowed superpages store GFNs in 32-bit page_info fields. */
> Ah, that's an abuse of the backlink field. After some looking
> around I first thought the up field could be used to store the
> GFN instead, as it's supposedly used for single-page shadows
> only. Then however I found
>
> static inline int sh_type_has_up_pointer(struct domain *d, unsigned int t)
> {
>     /* Multi-page shadows don't have up-pointers */
>     if ( t == SH_type_l1_32_shadow
>          || t == SH_type_fl1_32_shadow
>          || t == SH_type_l2_32_shadow )
>         return 0;
>     /* Pinnable shadows don't have up-pointers either */
>     return !sh_type_is_pinnable(d, t);
> }
>
> It's unclear to me in which way SH_type_l1_32_shadow and
> SH_type_l2_32_shadow are "multi-page" shadows; I'd rather have
> expected all three SH_type_fl1_*_shadow to be. Tim?

I suspect the comment is incomplete, and should include "4k shadows
don't have up-pointers".

>
> In any event there would be 12 bits to reclaim from the up
> pointer - it being a physical address, there'll not be more
> than 52 significant bits.

Right, but for L1TF safety, the address bits in the PTE must not be
cacheable.

Currently, on fully populated multi-socket servers, the MMIO fastpath
relies on the top 4G of address space not being cacheable, which is the
safest we can reasonably manage.  Extending this by a nibble takes us to
16G which is not meaningfully less safe.

~Andrew