svm: Introduce AMD IOMMU avic_ga_log_notifier
authorSuravee Suthikulpanit <suravee.suthikulpanit@amd.com>
Tue, 23 Aug 2016 18:52:42 +0000 (13:52 -0500)
committerPaolo Bonzini <pbonzini@redhat.com>
Thu, 8 Sep 2016 11:11:56 +0000 (13:11 +0200)
This patch introduces avic_ga_log_notifier, which will be called
by IOMMU driver whenever it handles the Guest vAPIC (GA) log entry.

Reviewed-by: Radim Krčmář <rkrcmar@redhat.com>
Signed-off-by: Suravee Suthikulpanit <suravee.suthikulpanit@amd.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
arch/x86/include/asm/kvm_host.h
arch/x86/kvm/svm.c

index d36176bb1b6d0c247c9bdb2652f1cb44f5ccab99..4c738c206be3830b66d387b951e4feac861985ed 100644 (file)
@@ -785,6 +785,7 @@ struct kvm_arch {
        u32 ldr_mode;
        struct page *avic_logical_id_table_page;
        struct page *avic_physical_id_table_page;
+       struct hlist_node hnode;
 
        bool x2apic_format;
        bool x2apic_broadcast_quirk_disabled;
index 27e37d8044b3793d23f4aa8afe6625dfcdd6a65a..0a9e43b2bd2604c585e2add24c4c6aeb9ba17c81 100644 (file)
@@ -34,6 +34,8 @@
 #include <linux/sched.h>
 #include <linux/trace_events.h>
 #include <linux/slab.h>
+#include <linux/amd-iommu.h>
+#include <linux/hashtable.h>
 
 #include <asm/apic.h>
 #include <asm/perf_event.h>
@@ -945,6 +947,55 @@ static void svm_disable_lbrv(struct vcpu_svm *svm)
        set_msr_interception(msrpm, MSR_IA32_LASTINTTOIP, 0, 0);
 }
 
+/* Note:
+ * This hash table is used to map VM_ID to a struct kvm_arch,
+ * when handling AMD IOMMU GALOG notification to schedule in
+ * a particular vCPU.
+ */
+#define SVM_VM_DATA_HASH_BITS  8
+DECLARE_HASHTABLE(svm_vm_data_hash, SVM_VM_DATA_HASH_BITS);
+static spinlock_t svm_vm_data_hash_lock;
+
+/* Note:
+ * This function is called from IOMMU driver to notify
+ * SVM to schedule in a particular vCPU of a particular VM.
+ */
+static int avic_ga_log_notifier(u32 ga_tag)
+{
+       unsigned long flags;
+       struct kvm_arch *ka = NULL;
+       struct kvm_vcpu *vcpu = NULL;
+       u32 vm_id = AVIC_GATAG_TO_VMID(ga_tag);
+       u32 vcpu_id = AVIC_GATAG_TO_VCPUID(ga_tag);
+
+       pr_debug("SVM: %s: vm_id=%#x, vcpu_id=%#x\n", __func__, vm_id, vcpu_id);
+
+       spin_lock_irqsave(&svm_vm_data_hash_lock, flags);
+       hash_for_each_possible(svm_vm_data_hash, ka, hnode, vm_id) {
+               struct kvm *kvm = container_of(ka, struct kvm, arch);
+               struct kvm_arch *vm_data = &kvm->arch;
+
+               if (vm_data->avic_vm_id != vm_id)
+                       continue;
+               vcpu = kvm_get_vcpu_by_id(kvm, vcpu_id);
+               break;
+       }
+       spin_unlock_irqrestore(&svm_vm_data_hash_lock, flags);
+
+       if (!vcpu)
+               return 0;
+
+       /* Note:
+        * At this point, the IOMMU should have already set the pending
+        * bit in the vAPIC backing page. So, we just need to schedule
+        * in the vcpu.
+        */
+       if (vcpu->mode == OUTSIDE_GUEST_MODE)
+               kvm_vcpu_wake_up(vcpu);
+
+       return 0;
+}
+
 static __init int svm_hardware_setup(void)
 {
        int cpu;
@@ -1003,10 +1054,15 @@ static __init int svm_hardware_setup(void)
        if (avic) {
                if (!npt_enabled ||
                    !boot_cpu_has(X86_FEATURE_AVIC) ||
-                   !IS_ENABLED(CONFIG_X86_LOCAL_APIC))
+                   !IS_ENABLED(CONFIG_X86_LOCAL_APIC)) {
                        avic = false;
-               else
+               } else {
                        pr_info("AVIC enabled\n");
+
+                       hash_init(svm_vm_data_hash);
+                       spin_lock_init(&svm_vm_data_hash_lock);
+                       amd_iommu_register_ga_log_notifier(&avic_ga_log_notifier);
+               }
        }
 
        return 0;
@@ -1327,6 +1383,7 @@ static inline int avic_free_vm_id(int id)
 
 static void avic_vm_destroy(struct kvm *kvm)
 {
+       unsigned long flags;
        struct kvm_arch *vm_data = &kvm->arch;
 
        avic_free_vm_id(vm_data->avic_vm_id);
@@ -1335,10 +1392,15 @@ static void avic_vm_destroy(struct kvm *kvm)
                __free_page(vm_data->avic_logical_id_table_page);
        if (vm_data->avic_physical_id_table_page)
                __free_page(vm_data->avic_physical_id_table_page);
+
+       spin_lock_irqsave(&svm_vm_data_hash_lock, flags);
+       hash_del(&vm_data->hnode);
+       spin_unlock_irqrestore(&svm_vm_data_hash_lock, flags);
 }
 
 static int avic_vm_init(struct kvm *kvm)
 {
+       unsigned long flags;
        int err = -ENOMEM;
        struct kvm_arch *vm_data = &kvm->arch;
        struct page *p_page;
@@ -1367,6 +1429,10 @@ static int avic_vm_init(struct kvm *kvm)
        vm_data->avic_logical_id_table_page = l_page;
        clear_page(page_address(l_page));
 
+       spin_lock_irqsave(&svm_vm_data_hash_lock, flags);
+       hash_add(svm_vm_data_hash, &vm_data->hnode, vm_data->avic_vm_id);
+       spin_unlock_irqrestore(&svm_vm_data_hash_lock, flags);
+
        return 0;
 
 free_avic: