On 27.09.2021 22:25, Tim Deegan wrote: > At 13:31 +0200 on 24 Sep (1632490304), Jan Beulich wrote: >> I'm afraid you're still my best guess to hopefully get an insight >> on issues like this one. > > I'm now very rusty on all this but I'll do my best! I suspect I'll > just be following you through the code.
Thanks much! >> While doing IOMMU superpage work I was, just in the background, >> considering in how far the superpage re-coalescing to be used there >> couldn't be re-used for P2M / EPT / NPT. Which got me to think about >> shadow mode's using of p2m-pt.c: That's purely software use of the >> tables in that case, isn't it? In which case hardware support for >> superpages shouldn't matter at all. > > ISTR at the time we used the same table for p2m and NPT. > If that's gone away, then yes, we could have superpages > in the p2m without caring about hardware support. No, that code is still used two ways, but it can't be used for the same domain in both of these ways. IOW I'm wondering whether the check for 2M pages to be usable shouldn't be "!hap || hap_2mb", as opposed to the 1G check continuing to be "hap && hap_1gb". Of course once I make that change, I may end up learning what "potential errors" that other commit was talking about ... As to the further parts of your reply, I guess I'll try to transform this (largely supporting my observations) and the above into one or more patches then. Jan