linux kernel KVM

最新推荐文章于 2025-06-17 11:43:59 发布

星空探索

最新推荐文章于 2025-06-17 11:43:59 发布

阅读量820

点赞数

CC 4.0 BY-SA版权

分类专栏： Linux KVM

本文链接：https://blog.youkuaiyun.com/sunlei0625/article/details/57128311

Linux KVM 专栏收录该内容

12 篇文章

订阅专栏

本文介绍KVM虚拟机的创建过程及ioctl命令的处理机制，详细解析了KVM内核模块中用于创建虚拟机的数据结构及ioctl命令的具体实现。

摘要生成于 C知道，由 DeepSeek-R1 满血版支持，前往体验 >

kvm 的linux内核描述：

include/linux/kvm_host.h
struct kvm {
 spinlock_t mmu_lock;
 struct mutex slots_lock;
 struct mm_struct *mm; /*进程地址空间描述*/
 struct kvm_memslots *memslots[KVM_ADDRESS_SPACE_NUM];KVM使用的内存描述
 struct srcu_struct srcu;
 struct srcu_struct irq_srcu;
 struct kvm_vcpu *vcpus[KVM_MAX_VCPUS];支持的最大虚拟CPU个数
 /*
  * created_vcpus is protected by kvm->lock, and is incremented
  * at the beginning of KVM_CREATE_VCPU.  online_vcpus is only
  * incremented after storing the kvm_vcpu pointer in vcpus,
  * and is accessed atomically.
  */
 atomic_t online_vcpus;
 int created_vcpus;
 int last_boosted_vcpu;
 struct list_head vm_list;
 struct mutex lock;
 struct kvm_io_bus *buses[KVM_NR_BUSES];
#ifdef CONFIG_HAVE_KVM_EVENTFD
 struct {
  spinlock_t        lock;
  struct list_head  items;
  struct list_head  resampler_list;
  struct mutex      resampler_lock;
 } irqfds;
 struct list_head ioeventfds;
#endif
 struct kvm_vm_stat stat;
 struct kvm_arch arch;
 atomic_t users_count;
#ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
 struct kvm_coalesced_mmio_ring *coalesced_mmio_ring;
 spinlock_t ring_lock;
 struct list_head coalesced_zones;
#endif
 struct mutex irq_lock;
#ifdef CONFIG_HAVE_KVM_IRQCHIP
 struct kvm_irq_routing_table __rcu *irq_routing;
#endif
#ifdef CONFIG_HAVE_KVM_IRQFD
 struct hlist_head irq_ack_notifier_list;
#endif
#if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 struct mmu_notifier mmu_notifier;
 unsigned long mmu_notifier_seq;
 long mmu_notifier_count;
#endif
 long tlbs_dirty;
 struct list_head devices;
 struct dentry *debugfs_dentry;
 struct kvm_stat_data **debugfs_stat_data;
};
相关的结构：
static struct file_operations kvm_chardev_ops = {
 .unlocked_ioctl = kvm_dev_ioctl,
 .compat_ioctl   = kvm_dev_ioctl,
 .llseek  = noop_llseek,
};
static struct miscdevice kvm_dev = {
 KVM_MINOR,
 "kvm",
 &kvm_chardev_ops,
};
用户空间工具通过ioctl与内核空间的kvm模块通信，kvm_dev_ioctl主要功能是创建虚拟机数据结构。
static long kvm_dev_ioctl(struct file *filp,
     unsigned int ioctl, unsigned long arg)
{
 long r = -EINVAL;
 switch (ioctl) {
 case KVM_GET_API_VERSION:
  if (arg)
   goto out;
  r = KVM_API_VERSION;
  break;
 case KVM_CREATE_VM:
  r = kvm_dev_ioctl_create_vm(arg);
  break;
 case KVM_CHECK_EXTENSION:
  r = kvm_vm_ioctl_check_extension_generic(NULL, arg);
  break;
 case KVM_GET_VCPU_MMAP_SIZE:
  if (arg)
   goto out;
  r = PAGE_SIZE;     /* struct kvm_run */
#ifdef CONFIG_X86
  r += PAGE_SIZE;    /* pio data page */
#endif
#ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
  r += PAGE_SIZE;    /* coalesced mmio ring page */
#endif
  break;
 case KVM_TRACE_ENABLE:
 case KVM_TRACE_PAUSE:
 case KVM_TRACE_DISABLE:
  r = -EOPNOTSUPP;
  break;
 default:
  return kvm_arch_dev_ioctl(filp, ioctl, arg);
 }
out:
 return r;
}
上面函数需要注意三点，一是支持的case类别，二是创建虚拟机的cmd类别为KVM_CREATE_VM，
三是体系结构可以对kvm_dev_ioctl支持的case增强，即实现kvm_arch_dev_ioctl函数，以达到更强的控制。
显然，KVM框架层支持的case以在上面函数给出，主要有：
#define KVM_GET_API_VERSION       _IO(KVMIO,   0x00)
#define KVM_CREATE_VM             _IO(KVMIO,   0x01) /* returns a VM fd */
#define KVM_GET_MSR_INDEX_LIST    _IOWR(KVMIO, 0x02, struct kvm_msr_list)
/*
 * Check if a kvm extension is available.  Argument is extension number,
 * return is 1 (yes) or 0 (no, sorry).
 */
#define KVM_CHECK_EXTENSION       _IO(KVMIO,   0x03)
/*
 * Get size for mmap(vcpu_fd)
 */
#define KVM_GET_VCPU_MMAP_SIZE    _IO(KVMIO,   0x04) /* in bytes */
#define KVM_GET_SUPPORTED_CPUID   _IOWR(KVMIO, 0x05, struct kvm_cpuid2)
#define KVM_TRACE_ENABLE          __KVM_DEPRECATED_MAIN_W_0x06
#define KVM_TRACE_PAUSE           __KVM_DEPRECATED_MAIN_0x07
#define KVM_TRACE_DISABLE         __KVM_DEPRECATED_MAIN_0x08
#define KVM_GET_EMULATED_CPUID   _IOWR(KVMIO, 0x09, struct kvm_cpuid2)
创建虚拟机的函数为kvm_dev_ioctl_create_vm()。
对于X86体系架构来说，其实现了kvm_arch_dev_ioctl()函数：
arch/x86/kvm/x86.c

long kvm_arch_dev_ioctl(struct file *filp,
   unsigned int ioctl, unsigned long arg)
{
 void __user *argp = (void __user *)arg;
 long r;
 switch (ioctl) {
 case KVM_GET_MSR_INDEX_LIST: {
  struct kvm_msr_list __user *user_msr_list = argp;
  struct kvm_msr_list msr_list;
  unsigned n;
  r = -EFAULT;
  if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
   goto out;
  n = msr_list.nmsrs;
  msr_list.nmsrs = num_msrs_to_save + num_emulated_msrs;
  if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
   goto out;
  r = -E2BIG;
  if (n < msr_list.nmsrs)
   goto out;
  r = -EFAULT;
  if (copy_to_user(user_msr_list->indices, &msrs_to_save,
     num_msrs_to_save * sizeof(u32)))
   goto out;
  if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
     &emulated_msrs,
     num_emulated_msrs * sizeof(u32)))
   goto out;
  r = 0;
  break;
 }
 case KVM_GET_SUPPORTED_CPUID:
 case KVM_GET_EMULATED_CPUID: {
  struct kvm_cpuid2 __user *cpuid_arg = argp;
  struct kvm_cpuid2 cpuid;
  r = -EFAULT;
  if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
   goto out;
  r = kvm_dev_ioctl_get_cpuid(&cpuid, cpuid_arg->entries,
         ioctl);
  if (r)
   goto out;
  r = -EFAULT;
  if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
   goto out;
  r = 0;
  break;
 }
 case KVM_X86_GET_MCE_CAP_SUPPORTED: {
  r = -EFAULT;
  if (copy_to_user(argp, &kvm_mce_cap_supported,
     sizeof(kvm_mce_cap_supported)))
   goto out;
  r = 0;
  break;
 }
 default:
  r = -EINVAL;
 }
out:
 return r;
}