Break-before-make (BBM) can be expensive, as transitioning via an invalid mapping (i.e. the "break" step) requires the completion of TLB invalidation and can also cause other agents to fault concurrently on the invalid mapping. Since BBM is not required when changing only the software bits of a PTE, avoid the sequence in this case and just update the PTE directly. Signed-off-by: Will Deacon <will@xxxxxxxxxx> --- arch/arm64/kvm/hyp/pgtable.c | 11 ++++++++++- 1 file changed, 10 insertions(+), 1 deletion(-) diff --git a/arch/arm64/kvm/hyp/pgtable.c b/arch/arm64/kvm/hyp/pgtable.c index 2069e6833831..756bbb15c1f3 100644 --- a/arch/arm64/kvm/hyp/pgtable.c +++ b/arch/arm64/kvm/hyp/pgtable.c @@ -744,6 +744,13 @@ static int stage2_map_walker_try_leaf(u64 addr, u64 end, u32 level, if (!stage2_pte_needs_update(old, new)) return -EAGAIN; + /* + * If we're only changing software bits, then we don't need to + * do anything else/ + */ + if (!((old ^ new) & ~KVM_PTE_LEAF_ATTR_HI_SW)) + goto out_set_pte; + stage2_put_pte(ptep, data->mmu, addr, level, mm_ops); } @@ -754,9 +761,11 @@ static int stage2_map_walker_try_leaf(u64 addr, u64 end, u32 level, if (mm_ops->icache_inval_pou && stage2_pte_executable(new)) mm_ops->icache_inval_pou(kvm_pte_follow(new, mm_ops), granule); - smp_store_release(ptep, new); if (stage2_pte_is_counted(new)) mm_ops->get_page(ptep); + +out_set_pte: + smp_store_release(ptep, new); if (kvm_phys_is_valid(phys)) data->phys += granule; return 0; -- 2.36.1.124.g0e6072fb45-goog _______________________________________________ kvmarm mailing list kvmarm@xxxxxxxxxxxxxxxxxxxxx https://lists.cs.columbia.edu/mailman/listinfo/kvmarm