On 18.10.2021 13:19, Roger Pau Monné wrote:
> On Thu, May 20, 2021 at 03:34:28PM +0200, Jan Beulich wrote:
>> The SDM specifically allows for earlier writes to fully overlapping
>> ranges to be dropped. If a guest did so, hvmemul_phys_mmio_access()
>> would crash it if varying data was written to the same address. Detect
>> overlaps early, as doing so in hvmemul_{linear,phys}_mmio_access() would
>> be quite a bit more difficult. To maintain proper faulting behavior,
>> instead of dropping earlier write instances of fully overlapping slots
>> altogether, write the data of the final of these slots multiple times.
>
> Is it possible for a later (non duplicated slot) to cause a fault
> ending the instruction without reaching that final slot that contains
> the written data?
Yes, but that's not a problem: Only faults are required to be ordered,
and when a fault occurs guarantees are made only towards lower indices
(read: all lower index writes would have completed, while nothing can
be said about higher indices). All non-faulting writes can go out in
any order (unless there are [partial] overlaps, but afaict that case
still gets dealt with within spec by the proposed new logic).
Jan
