KVM: x86/mmu: Embed direct bits into gpa for KVM_PRE_FAULT_MEMORY
authorPaolo Bonzini <pbonzini@redhat.com>
Wed, 11 Jun 2025 00:10:18 +0000 (20:10 -0400)
committerPaolo Bonzini <pbonzini@redhat.com>
Thu, 12 Jun 2025 04:43:39 +0000 (00:43 -0400)
Bug[*] reported for TDX case when enabling KVM_PRE_FAULT_MEMORY in QEMU.

It turns out that @gpa passed to kvm_mmu_do_page_fault() doesn't have
shared bit set when the memory attribute of it is shared, and it leads
to wrong root in tdp_mmu_get_root_for_fault().

Fix it by embedding the direct bits in the gpa that is passed to
kvm_tdp_map_page(), when the memory of the gpa is not private.

[*] https://lore.kernel.org/qemu-devel/4a757796-11c2-47f1-ae0d-335626e818fd@intel.com/

Reported-by: Xiaoyao Li <xiaoyao.li@intel.com>
Closes: https://lore.kernel.org/qemu-devel/4a757796-11c2-47f1-ae0d-335626e818fd@intel.com/
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Xiaoyao Li <xiaoyao.li@intel.com>
Message-ID: <20250611001018.2179964-1-xiaoyao.li@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
arch/x86/kvm/mmu/mmu.c

index cbc84c6abc2e3f8b1f18ed8d1c9629fd7f827466..a4040578b5370cec275978453caece53b1a39d3a 100644 (file)
@@ -4896,6 +4896,7 @@ long kvm_arch_vcpu_pre_fault_memory(struct kvm_vcpu *vcpu,
 {
        u64 error_code = PFERR_GUEST_FINAL_MASK;
        u8 level = PG_LEVEL_4K;
+       u64 direct_bits;
        u64 end;
        int r;
 
@@ -4910,15 +4911,18 @@ long kvm_arch_vcpu_pre_fault_memory(struct kvm_vcpu *vcpu,
        if (r)
                return r;
 
+       direct_bits = 0;
        if (kvm_arch_has_private_mem(vcpu->kvm) &&
            kvm_mem_is_private(vcpu->kvm, gpa_to_gfn(range->gpa)))
                error_code |= PFERR_PRIVATE_ACCESS;
+       else
+               direct_bits = gfn_to_gpa(kvm_gfn_direct_bits(vcpu->kvm));
 
        /*
         * Shadow paging uses GVA for kvm page fault, so restrict to
         * two-dimensional paging.
         */
-       r = kvm_tdp_map_page(vcpu, range->gpa, error_code, &level);
+       r = kvm_tdp_map_page(vcpu, range->gpa | direct_bits, error_code, &level);
        if (r < 0)
                return r;