On Thu, 2022-11-03 at 05:40 +0300, Kirill A. Shutemov wrote:
> On Thu, Nov 03, 2022 at 09:04:23AM +0800, Robert Hoo wrote:
> > I also notice that skip_tlb_flush is set when pcid_enabled && (CR3
> > & X86_CR3_PCID_NOFLUSH). Under this condition, do you think (0,0)
> > -->
> > (1,0) need to flip it back to false?
>
> Yes, I think we should. We know it is a safe choice.
If so, then judging the (0,0) --> (1,0) case in the else{} branch is
inevitable, isn't it?
Or totally remove the skip_tlb_flush logic in this function, but this
would break existing logic. You won't like it.
>
> It also would be nice to get LAM documentation updated on the
> expected
> behaviour. It is not clear from current documentation if enabling LAM
> causes flush. We can only guess that it should at least for some
> scenarios.
>
> Phantom TLB entires that resurface after LAM gets disable would be
> fun to
> debug.
>
Agree, and echo your conservativeness.
