]> Pileus Git - ~andy/linux/blobdiff - drivers/kvm/kvm_main.c
[PATCH] KVM: Recover after an arch module load failure
[~andy/linux] / drivers / kvm / kvm_main.c
index fd1bb870545c28adc3fa8c4bb7927046504e67e8..b54caf0ceeb1305a2129177234a745cf079897a5 100644 (file)
@@ -113,6 +113,11 @@ unsigned long segment_base(u16 selector)
 }
 EXPORT_SYMBOL_GPL(segment_base);
 
+static inline int valid_vcpu(int n)
+{
+       return likely(n >= 0 && n < KVM_MAX_VCPUS);
+}
+
 int kvm_read_guest(struct kvm_vcpu *vcpu,
                             gva_t addr,
                             unsigned long size,
@@ -240,7 +245,8 @@ static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
        if (!dont || free->phys_mem != dont->phys_mem)
                if (free->phys_mem) {
                        for (i = 0; i < free->npages; ++i)
-                               __free_page(free->phys_mem[i]);
+                               if (free->phys_mem[i])
+                                       __free_page(free->phys_mem[i]);
                        vfree(free->phys_mem);
                }
 
@@ -393,7 +399,7 @@ void set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
                return;
        }
 
-       if (kvm_arch_ops->is_long_mode(vcpu)) {
+       if (is_long_mode(vcpu)) {
                if (!(cr4 & CR4_PAE_MASK)) {
                        printk(KERN_DEBUG "set_cr4: #GP, clearing PAE while "
                               "in long mode\n");
@@ -420,7 +426,7 @@ EXPORT_SYMBOL_GPL(set_cr4);
 
 void set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
 {
-       if (kvm_arch_ops->is_long_mode(vcpu)) {
+       if (is_long_mode(vcpu)) {
                if ( cr3 & CR3_L_MODE_RESEVED_BITS) {
                        printk(KERN_DEBUG "set_cr3: #GP, reserved bits\n");
                        inject_gp(vcpu);
@@ -494,7 +500,7 @@ static int kvm_dev_ioctl_create_vcpu(struct kvm *kvm, int n)
        struct kvm_vcpu *vcpu;
 
        r = -EINVAL;
-       if (n < 0 || n >= KVM_MAX_VCPUS)
+       if (!valid_vcpu(n))
                goto out;
 
        vcpu = &kvm->vcpus[n];
@@ -516,12 +522,14 @@ static int kvm_dev_ioctl_create_vcpu(struct kvm *kvm, int n)
        if (r < 0)
                goto out_free_vcpus;
 
-       kvm_arch_ops->vcpu_load(vcpu);
+       r = kvm_mmu_create(vcpu);
+       if (r < 0)
+               goto out_free_vcpus;
 
-       r = kvm_arch_ops->vcpu_setup(vcpu);
+       kvm_arch_ops->vcpu_load(vcpu);
+       r = kvm_mmu_setup(vcpu);
        if (r >= 0)
-               r = kvm_mmu_init(vcpu);
-
+               r = kvm_arch_ops->vcpu_setup(vcpu);
        vcpu_put(vcpu);
 
        if (r < 0)
@@ -1098,6 +1106,51 @@ void realmode_set_cr(struct kvm_vcpu *vcpu, int cr, unsigned long val,
        }
 }
 
+int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
+{
+       u64 data;
+
+       switch (msr) {
+       case 0xc0010010: /* SYSCFG */
+       case 0xc0010015: /* HWCR */
+       case MSR_IA32_PLATFORM_ID:
+       case MSR_IA32_P5_MC_ADDR:
+       case MSR_IA32_P5_MC_TYPE:
+       case MSR_IA32_MC0_CTL:
+       case MSR_IA32_MCG_STATUS:
+       case MSR_IA32_MCG_CAP:
+       case MSR_IA32_MC0_MISC:
+       case MSR_IA32_MC0_MISC+4:
+       case MSR_IA32_MC0_MISC+8:
+       case MSR_IA32_MC0_MISC+12:
+       case MSR_IA32_MC0_MISC+16:
+       case MSR_IA32_UCODE_REV:
+       case MSR_IA32_PERF_STATUS:
+               /* MTRR registers */
+       case 0xfe:
+       case 0x200 ... 0x2ff:
+               data = 0;
+               break;
+       case 0xcd: /* fsb frequency */
+               data = 3;
+               break;
+       case MSR_IA32_APICBASE:
+               data = vcpu->apic_base;
+               break;
+#ifdef CONFIG_X86_64
+       case MSR_EFER:
+               data = vcpu->shadow_efer;
+               break;
+#endif
+       default:
+               printk(KERN_ERR "kvm: unhandled rdmsr: 0x%x\n", msr);
+               return 1;
+       }
+       *pdata = data;
+       return 0;
+}
+EXPORT_SYMBOL_GPL(kvm_get_msr_common);
+
 /*
  * Reads an msr value (of 'msr_index') into 'pdata'.
  * Returns 0 on success, non-0 otherwise.
@@ -1110,7 +1163,7 @@ static int get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
 
 #ifdef CONFIG_X86_64
 
-void set_efer(struct kvm_vcpu *vcpu, u64 efer)
+static void set_efer(struct kvm_vcpu *vcpu, u64 efer)
 {
        if (efer & EFER_RESERVED_BITS) {
                printk(KERN_DEBUG "set_efer: 0x%llx #GP, reserved bits\n",
@@ -1133,10 +1186,36 @@ void set_efer(struct kvm_vcpu *vcpu, u64 efer)
 
        vcpu->shadow_efer = efer;
 }
-EXPORT_SYMBOL_GPL(set_efer);
 
 #endif
 
+int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
+{
+       switch (msr) {
+#ifdef CONFIG_X86_64
+       case MSR_EFER:
+               set_efer(vcpu, data);
+               break;
+#endif
+       case MSR_IA32_MC0_STATUS:
+               printk(KERN_WARNING "%s: MSR_IA32_MC0_STATUS 0x%llx, nop\n",
+                      __FUNCTION__, data);
+               break;
+       case MSR_IA32_UCODE_REV:
+       case MSR_IA32_UCODE_WRITE:
+       case 0x200 ... 0x2ff: /* MTRRs */
+               break;
+       case MSR_IA32_APICBASE:
+               vcpu->apic_base = data;
+               break;
+       default:
+               printk(KERN_ERR "kvm: unhandled wrmsr: 0x%x\n", msr);
+               return 1;
+       }
+       return 0;
+}
+EXPORT_SYMBOL_GPL(kvm_set_msr_common);
+
 /*
  * Writes msr value into into the appropriate "register".
  * Returns 0 on success, non-0 otherwise.
@@ -1179,7 +1258,7 @@ static int kvm_dev_ioctl_run(struct kvm *kvm, struct kvm_run *kvm_run)
        struct kvm_vcpu *vcpu;
        int r;
 
-       if (kvm_run->vcpu < 0 || kvm_run->vcpu >= KVM_MAX_VCPUS)
+       if (!valid_vcpu(kvm_run->vcpu))
                return -EINVAL;
 
        vcpu = vcpu_load(kvm, kvm_run->vcpu);
@@ -1208,7 +1287,7 @@ static int kvm_dev_ioctl_get_regs(struct kvm *kvm, struct kvm_regs *regs)
 {
        struct kvm_vcpu *vcpu;
 
-       if (regs->vcpu < 0 || regs->vcpu >= KVM_MAX_VCPUS)
+       if (!valid_vcpu(regs->vcpu))
                return -EINVAL;
 
        vcpu = vcpu_load(kvm, regs->vcpu);
@@ -1254,7 +1333,7 @@ static int kvm_dev_ioctl_set_regs(struct kvm *kvm, struct kvm_regs *regs)
 {
        struct kvm_vcpu *vcpu;
 
-       if (regs->vcpu < 0 || regs->vcpu >= KVM_MAX_VCPUS)
+       if (!valid_vcpu(regs->vcpu))
                return -EINVAL;
 
        vcpu = vcpu_load(kvm, regs->vcpu);
@@ -1301,7 +1380,7 @@ static int kvm_dev_ioctl_get_sregs(struct kvm *kvm, struct kvm_sregs *sregs)
        struct kvm_vcpu *vcpu;
        struct descriptor_table dt;
 
-       if (sregs->vcpu < 0 || sregs->vcpu >= KVM_MAX_VCPUS)
+       if (!valid_vcpu(sregs->vcpu))
                return -EINVAL;
        vcpu = vcpu_load(kvm, sregs->vcpu);
        if (!vcpu)
@@ -1353,7 +1432,7 @@ static int kvm_dev_ioctl_set_sregs(struct kvm *kvm, struct kvm_sregs *sregs)
        int i;
        struct descriptor_table dt;
 
-       if (sregs->vcpu < 0 || sregs->vcpu >= KVM_MAX_VCPUS)
+       if (!valid_vcpu(sregs->vcpu))
                return -EINVAL;
        vcpu = vcpu_load(kvm, sregs->vcpu);
        if (!vcpu)
@@ -1412,6 +1491,9 @@ static int kvm_dev_ioctl_set_sregs(struct kvm *kvm, struct kvm_sregs *sregs)
 /*
  * List of msr numbers which we expose to userspace through KVM_GET_MSRS
  * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
+ *
+ * This list is modified at module load time to reflect the
+ * capabilities of the host cpu.
  */
 static u32 msrs_to_save[] = {
        MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
@@ -1422,6 +1504,22 @@ static u32 msrs_to_save[] = {
        MSR_IA32_TIME_STAMP_COUNTER,
 };
 
+static unsigned num_msrs_to_save;
+
+static __init void kvm_init_msr_list(void)
+{
+       u32 dummy[2];
+       unsigned i, j;
+
+       for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
+               if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
+                       continue;
+               if (j < i)
+                       msrs_to_save[j] = msrs_to_save[i];
+               j++;
+       }
+       num_msrs_to_save = j;
+}
 
 /*
  * Adapt set_msr() to msr_io()'s calling convention
@@ -1444,7 +1542,7 @@ static int __msr_io(struct kvm *kvm, struct kvm_msrs *msrs,
        struct kvm_vcpu *vcpu;
        int i;
 
-       if (msrs->vcpu < 0 || msrs->vcpu >= KVM_MAX_VCPUS)
+       if (!valid_vcpu(msrs->vcpu))
                return -EINVAL;
 
        vcpu = vcpu_load(kvm, msrs->vcpu);
@@ -1537,7 +1635,7 @@ static int kvm_dev_ioctl_interrupt(struct kvm *kvm, struct kvm_interrupt *irq)
 {
        struct kvm_vcpu *vcpu;
 
-       if (irq->vcpu < 0 || irq->vcpu >= KVM_MAX_VCPUS)
+       if (!valid_vcpu(irq->vcpu))
                return -EINVAL;
        if (irq->irq < 0 || irq->irq >= 256)
                return -EINVAL;
@@ -1559,7 +1657,7 @@ static int kvm_dev_ioctl_debug_guest(struct kvm *kvm,
        struct kvm_vcpu *vcpu;
        int r;
 
-       if (dbg->vcpu < 0 || dbg->vcpu >= KVM_MAX_VCPUS)
+       if (!valid_vcpu(dbg->vcpu))
                return -EINVAL;
        vcpu = vcpu_load(kvm, dbg->vcpu);
        if (!vcpu)
@@ -1579,6 +1677,9 @@ static long kvm_dev_ioctl(struct file *filp,
        int r = -EINVAL;
 
        switch (ioctl) {
+       case KVM_GET_API_VERSION:
+               r = KVM_API_VERSION;
+               break;
        case KVM_CREATE_VCPU: {
                r = kvm_dev_ioctl_create_vcpu(kvm, arg);
                if (r)
@@ -1730,15 +1831,15 @@ static long kvm_dev_ioctl(struct file *filp,
                if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
                        goto out;
                n = msr_list.nmsrs;
-               msr_list.nmsrs = ARRAY_SIZE(msrs_to_save);
+               msr_list.nmsrs = num_msrs_to_save;
                if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
                        goto out;
                r = -E2BIG;
-               if (n < ARRAY_SIZE(msrs_to_save))
+               if (n < num_msrs_to_save)
                        goto out;
                r = -EFAULT;
                if (copy_to_user(user_msr_list->indices, &msrs_to_save,
-                                sizeof msrs_to_save))
+                                num_msrs_to_save * sizeof(u32)))
                        goto out;
                r = 0;
        }
@@ -1838,17 +1939,22 @@ int kvm_init_arch(struct kvm_arch_ops *ops, struct module *module)
 {
        int r;
 
-       kvm_arch_ops = ops;
+       if (kvm_arch_ops) {
+               printk(KERN_ERR "kvm: already loaded the other module\n");
+               return -EEXIST;
+       }
 
-       if (!kvm_arch_ops->cpu_has_kvm_support()) {
+       if (!ops->cpu_has_kvm_support()) {
                printk(KERN_ERR "kvm: no hardware support\n");
                return -EOPNOTSUPP;
        }
-       if (kvm_arch_ops->disabled_by_bios()) {
+       if (ops->disabled_by_bios()) {
                printk(KERN_ERR "kvm: disabled by bios\n");
                return -EOPNOTSUPP;
        }
 
+       kvm_arch_ops = ops;
+
        r = kvm_arch_ops->hardware_setup();
        if (r < 0)
            return r;
@@ -1880,6 +1986,7 @@ void kvm_exit_arch(void)
        unregister_reboot_notifier(&kvm_reboot_notifier);
        on_each_cpu(kvm_arch_ops->hardware_disable, 0, 0, 1);
        kvm_arch_ops->hardware_unsetup();
+       kvm_arch_ops = NULL;
 }
 
 static __init int kvm_init(void)
@@ -1889,6 +1996,8 @@ static __init int kvm_init(void)
 
        kvm_init_debug();
 
+       kvm_init_msr_list();
+
        if ((bad_page = alloc_page(GFP_KERNEL)) == NULL) {
                r = -ENOMEM;
                goto out;