Excise the MMU's one-off self-changing mapping logic and instead detect self-changing mappings in the primary "walk" flow, and rely on kvm_mmu_hugepage_adjust() to naturally handle "disallowed hugepage due to shadow page" conditions. When is_self_change_mapping() was first added, KVM did hugepage adjustments before the primary walk, and so didn't account for shadow pages that were allocated for the current page fault, i.e. effectively consumed a stale disallow_lpage. Now that KVM adjust after allocating new shadow pages, the one-off code is superfluous. Dropping the one-off code fixes an issue where KVM will force 4KiB pages for a 1GiB guest page even when using a 2MiB would be safe (1GiB overlaps a shadow page but 2MiB does not). v2: - Track the "write #PF to shadow page" using an EMULTYPE flag. - Split the main patch in two. v1: https://lore.kernel.org/all/20221213125538.81209-1-jiangshanlai@xxxxxxxxx Lai Jiangshan (2): KVM: x86/mmu: Detect write #PF to shadow pages during FNAME(fetch) walk KVM: x86/mmu: Remove FNAME(is_self_change_mapping) Sean Christopherson (1): KVM: x86/mmu: Use EMULTYPE flag to track write #PFs to shadow pages arch/x86/include/asm/kvm_host.h | 37 +++++++++++--------- arch/x86/kvm/mmu/mmu.c | 5 +-- arch/x86/kvm/mmu/mmu_internal.h | 12 ++++++- arch/x86/kvm/mmu/paging_tmpl.h | 61 ++++++--------------------------- arch/x86/kvm/x86.c | 15 ++------ 5 files changed, 46 insertions(+), 84 deletions(-) base-commit: 11b36fe7d4500c8ef73677c087f302fd713101c2 -- 2.39.1.519.gcb327c4b5f-goog
