On Thu, Dec 12, 2019 at 12:28:27AM +0100, Paolo Bonzini wrote: > On 11/12/19 20:13, Sean Christopherson wrote: > > Assuming we haven't missed something, the easiest fix would be to reduce > > the MMIO generation by one bit and use bits 62:54 for the MMIO generation. > > Yes, and I mistakenly thought it would be done just by adjusting > PT64_SECOND_AVAIL_BITS_SHIFT. > > I will test and send formally something like this: > > diff --git a/arch/x86/kvm/mmu/mmu.c b/arch/x86/kvm/mmu/mmu.c > index 6f92b40d798c..aa2d86f42b9a 100644 > --- a/arch/x86/kvm/mmu/mmu.c > +++ b/arch/x86/kvm/mmu/mmu.c > @@ -405,11 +405,13 @@ static inline bool is_access_track_spte(u64 spte) > } > > /* > - * Due to limited space in PTEs, the MMIO generation is a 19 bit subset of > + * Due to limited space in PTEs, the MMIO generation is a 18 bit subset of > * the memslots generation and is derived as follows: > * > * Bits 0-8 of the MMIO generation are propagated to spte bits 3-11 > - * Bits 9-18 of the MMIO generation are propagated to spte bits 52-61 > + * Bits 9-17 of the MMIO generation are propagated to spte bits 54-62 > * > + * We don't use bit 63 to avoid conflicting with the SVE bit in EPT PTEs. > + * > * The KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS flag is intentionally not included in > * the MMIO generation number, as doing so would require stealing a bit from > @@ -418,15 +418,16 @@ static inline bool is_access_track_spte(u64 spte) > * requires a full MMU zap). The flag is instead explicitly queried when > * checking for MMIO spte cache hits. > */ > -#define MMIO_SPTE_GEN_MASK GENMASK_ULL(18, 0) > +#define MMIO_SPTE_GEN_MASK GENMASK_ULL(17, 0) > > #define MMIO_SPTE_GEN_LOW_START 3 > #define MMIO_SPTE_GEN_LOW_END 11 > #define MMIO_SPTE_GEN_LOW_MASK GENMASK_ULL(MMIO_SPTE_GEN_LOW_END, \ > MMIO_SPTE_GEN_LOW_START) > > -#define MMIO_SPTE_GEN_HIGH_START 52 > -#define MMIO_SPTE_GEN_HIGH_END 61 > +/* Leave room for SPTE_SPECIAL_MASK. */ > +#define MMIO_SPTE_GEN_HIGH_START PT64_SECOND_AVAIL_BITS_SHIFT I'd rather have GEN_HIGH_START be an explicit bit number and then add a BUILD_BUG_ON(GEN_HIGH_START < PT64_SECOND_AVAIL_BITS_SHIFT) to ensure the MMIO gen doesn't overlap other stuff. That way we get a build error if someone changes PT64_SECOND_AVAIL_BITS_SHIFT, otherwise the MMIO gen will end up who knows where and probably overwrite NX or EPT.SUPPRESS_VE. > +#define MMIO_SPTE_GEN_HIGH_END 62 > #define MMIO_SPTE_GEN_HIGH_MASK GENMASK_ULL(MMIO_SPTE_GEN_HIGH_END, \ > MMIO_SPTE_GEN_HIGH_START) > static u64 generation_mmio_spte_mask(u64 gen) > > > Paolo >
