Re: [PATCH] x86/pv: Flush TLB in response to paging structure changes

[Andrew Cooper <andrew.cooper3@xxxxxxxxxx>]

On 20/10/2020 18:10, Andrew Cooper wrote:
> On 20/10/2020 17:20, Andrew Cooper wrote:
>> On 20/10/2020 16:48, Jan Beulich wrote:
>>> On 20.10.2020 17:24, Andrew Cooper wrote:
>>>> With MMU_UPDATE, a PV guest can make changes to higher level pagetables.  
>>>> This
>>>> is from Xen's point of view (as the update only affects guest mappings), 
>>>> and
>>>> the guest is required to flush suitably after making updates.
>>>>
>>>> However, Xen's use of linear pagetables (UPDATE_VA_MAPPING, GNTTABOP_map,
>>>> writeable pagetables, etc.) is an implementation detail outside of the
>>>> API/ABI.
>>>>
>>>> Changes in the paging structure require invalidations in the linear 
>>>> pagetable
>>>> range for subsequent accesses into the linear pagetables to access 
>>>> non-stale
>>>> mappings.  Xen must provide suitable flushing to prevent intermixed guest
>>>> actions from accidentally accessing/modifying the wrong pagetable.
>>>>
>>>> For all L2 and higher modifications, flush the full TLB.  (This could in
>>>> principle be an order 39 flush starting at LINEAR_PT_VIRT_START, but no 
>>>> such
>>>> mechanism exists in practice.)
>>>>
>>>> As this combines with sync_guest for XPTI L4 "shadowing", replace the
>>>> sync_guest boolean with flush_flags and accumulate flags.  The sync_guest 
>>>> case
>>>> now always needs to flush, there is no point trying to exclude the current 
>>>> CPU
>>>> from the flush mask.  Use pt_owner->dirty_cpumask directly.
>>> Why is there no point? There's no need for the FLUSH_ROOT_PGTBL
>>> part of the flushing on the local CPU. The draft you had sent
>>> earlier looked better in this regard.
>> This was the area which broke.  It is to do with subtle difference in
>> the scope of L4 updates.
>>
>> ROOT_PGTBL needs to resync current (if in use), and be broadcasted if
>> other references to the pages are found.
>>
>> The TLB flush needs to be broadcast to the whole domain dirty mask, as
>> we can't (easily) know if the update was part of the current structure.
> Actually - we can know whether an L4 update needs flushing locally or
> not, in exactly the same way as the sync logic currently works.
>
> However, unlike the opencoded get_cpu_info()->root_pgt_changed = true,
> we can't just flush locally for free.
>
> This is quite awkward to express.

And not safe.  Flushes may accumulate from multiple levels in a batch,
and pt_owner may not be equal to current.

I stand by the version submitted as the security fix.

~Andrew