On Fri, Apr 14, 2023, Jeremi Piotrowski wrote: > On 4/14/2023 1:12 AM, Sean Christopherson wrote: > > Preserve TDP MMU roots until they are explicitly invalidated by gifting > > the TDP MMU itself a reference to a root when it is allocated. Keeping a > > reference in the TDP MMU fixes a flaw where the TDP MMU exhibits terrible > > performance, and can potentially even soft-hang a vCPU, if a vCPU > > frequently unloads its roots, e.g. when KVM is emulating SMI+RSM. ... > Thank you, I just tested this and it works wonderfully! Is this still on time for 6.3? This is too risky for 6.3, but I am comfortable applying it for 6.4. > In case you need it: > > Tested-by: Jeremi Piotrowski <jpiotrowski@xxxxxxxxxxxxxxxxxxx> > > I'd also like to get this backported all the way back to 5.15 because the issue is > already present there. I tried it myself, but this was before async zap and i'm > doing something wrong with refcounts: For 5.15, I think our best bet is to just disable the TDP MMU by default. There have been a _lot_ of relevant changes since 5.15, I am skeptical that this patch can be backported to 5.15 without pulling in a big pile of changes from between 5.15 and 6.1 or so. I added you to a related thread[*] about TDP MMU backports for 5.15, let's continue the 5.15 discussion there. Thanks! [*] https://lore.kernel.org/all/ZDmEGM+CgYpvDLh6@xxxxxxxxxx