KVM/nVMX: Use kvm_vcpu_map when mapping the posted interrupt descriptor table
authorKarimAllah Ahmed <karahmed@amazon.de>
Thu, 31 Jan 2019 20:24:38 +0000 (21:24 +0100)
committerPaolo Bonzini <pbonzini@redhat.com>
Tue, 30 Apr 2019 19:45:21 +0000 (21:45 +0200)
Use kvm_vcpu_map when mapping the posted interrupt descriptor table since
using kvm_vcpu_gpa_to_page() and kmap() will only work for guest memory
that has a "struct page".

One additional semantic change is that the virtual host mapping lifecycle
has changed a bit. It now has the same lifetime of the pinning of the
interrupt descriptor table page on the host side.

Signed-off-by: KarimAllah Ahmed <karahmed@amazon.de>
Reviewed-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
arch/x86/kvm/vmx/nested.c
arch/x86/kvm/vmx/vmx.h

index a00c4ded272d74293c04fd0ac61dd13864d05064..69c505fcaad7ebe83503bc930bd3281386e42ce5 100644 (file)
@@ -230,12 +230,8 @@ static void free_nested(struct kvm_vcpu *vcpu)
                vmx->nested.apic_access_page = NULL;
        }
        kvm_vcpu_unmap(vcpu, &vmx->nested.virtual_apic_map, true);
-       if (vmx->nested.pi_desc_page) {
-               kunmap(vmx->nested.pi_desc_page);
-               kvm_release_page_dirty(vmx->nested.pi_desc_page);
-               vmx->nested.pi_desc_page = NULL;
-               vmx->nested.pi_desc = NULL;
-       }
+       kvm_vcpu_unmap(vcpu, &vmx->nested.pi_desc_map, true);
+       vmx->nested.pi_desc = NULL;
 
        kvm_mmu_free_roots(vcpu, &vcpu->arch.guest_mmu, KVM_MMU_ROOTS_ALL);
 
@@ -2891,26 +2887,15 @@ static void nested_get_vmcs12_pages(struct kvm_vcpu *vcpu)
        }
 
        if (nested_cpu_has_posted_intr(vmcs12)) {
-               if (vmx->nested.pi_desc_page) { /* shouldn't happen */
-                       kunmap(vmx->nested.pi_desc_page);
-                       kvm_release_page_dirty(vmx->nested.pi_desc_page);
-                       vmx->nested.pi_desc_page = NULL;
-                       vmx->nested.pi_desc = NULL;
-                       vmcs_write64(POSTED_INTR_DESC_ADDR, -1ull);
+               map = &vmx->nested.pi_desc_map;
+
+               if (!kvm_vcpu_map(vcpu, gpa_to_gfn(vmcs12->posted_intr_desc_addr), map)) {
+                       vmx->nested.pi_desc =
+                               (struct pi_desc *)(((void *)map->hva) +
+                               offset_in_page(vmcs12->posted_intr_desc_addr));
+                       vmcs_write64(POSTED_INTR_DESC_ADDR,
+                                    pfn_to_hpa(map->pfn) + offset_in_page(vmcs12->posted_intr_desc_addr));
                }
-               page = kvm_vcpu_gpa_to_page(vcpu, vmcs12->posted_intr_desc_addr);
-               if (is_error_page(page))
-                       return;
-               vmx->nested.pi_desc_page = page;
-               vmx->nested.pi_desc = kmap(vmx->nested.pi_desc_page);
-               vmx->nested.pi_desc =
-                       (struct pi_desc *)((void *)vmx->nested.pi_desc +
-                       (unsigned long)(vmcs12->posted_intr_desc_addr &
-                       (PAGE_SIZE - 1)));
-               vmcs_write64(POSTED_INTR_DESC_ADDR,
-                       page_to_phys(vmx->nested.pi_desc_page) +
-                       (unsigned long)(vmcs12->posted_intr_desc_addr &
-                       (PAGE_SIZE - 1)));
        }
        if (nested_vmx_prepare_msr_bitmap(vcpu, vmcs12))
                vmcs_set_bits(CPU_BASED_VM_EXEC_CONTROL,
@@ -3952,12 +3937,8 @@ void nested_vmx_vmexit(struct kvm_vcpu *vcpu, u32 exit_reason,
                vmx->nested.apic_access_page = NULL;
        }
        kvm_vcpu_unmap(vcpu, &vmx->nested.virtual_apic_map, true);
-       if (vmx->nested.pi_desc_page) {
-               kunmap(vmx->nested.pi_desc_page);
-               kvm_release_page_dirty(vmx->nested.pi_desc_page);
-               vmx->nested.pi_desc_page = NULL;
-               vmx->nested.pi_desc = NULL;
-       }
+       kvm_vcpu_unmap(vcpu, &vmx->nested.pi_desc_map, true);
+       vmx->nested.pi_desc = NULL;
 
        /*
         * We are now running in L2, mmu_notifier will force to reload the
index b03b18cf9b6bc8a4ffbf748fc63c65a22c48fa50..c0ff305d59f7819d2294f939c2266a243bde44cc 100644 (file)
@@ -143,7 +143,7 @@ struct nested_vmx {
         */
        struct page *apic_access_page;
        struct kvm_host_map virtual_apic_map;
-       struct page *pi_desc_page;
+       struct kvm_host_map pi_desc_map;
 
        struct kvm_host_map msr_bitmap_map;