KVM Simple Example

Background

Starting from ARM-v8.1, Virtualization Host Extension (VHE) feature supports running unmodified OS in EL2. mrs x0, ESR_EL1can be redirected to access ESR_EL2 with the help of system register redirection.

Host APP and Guest APP are now running in EL0, Guest Kernel executes in EL1, while Host Kernel & KVM live in EL2, EL1 is circumvented at most times.

Note

In this blog, KVM is integrated into Linux kernel directly, making module_init execute as device_initcall during kernel initialization.

Linux version 4.19 is chosen in the following parts.

Initialization

Initialization of KVM starts from virt/kvm/arm/arm.c:arm_init, invoking the only kvm_init inside the function：

kvm_init
-> kvm_arch_init
    -> in_hyp_mode = is_kernel_in_hyp_mode(); // return true, HCR_EL2.E2H == 1 --> whole kernel in EL2N
    -> init_common_resources
        -> kvm_set_ipa_limit
    -> init_subsystems
        -> _kvm_arch_hardware_enable    // enable hardware to access EL2
        -> kvm_vgic_hyp_init            // ??? debugger error
        -> kvm_timer_hyp_init           // ??? debugger error
        -> kvm_perf_init                // ??? debugger error
        -> kvm_coproc_table_init        // ??? debugger error
        -> _kvm_arch_hardware_disable
-> kvm_irqfd_init
-> register_reboot_notifier
-> kmem_cache_create_usercopy
-> kvm_async_pf_init
    -> KMEM_CACHE
-> misc_register // register /dev/kvm as a chardev, unlocked_ioctl = kvm_dev_ioctl
-> kvm_vfio_ops_init

---

KVM Simple Example

1. KVM_CREATE_VM (use kvmfd, return vmfd)

kvm_dev_ioctl_create_vm
-> kvm_create_vm
    -> kvm_arch_alloc_vm        // has_vhe ? vzalloc : kzalloc
    -> kvm_eventfd_init
    -> kvm_arch_init_vm         // initialize stage2 mm & vgic
        -> kvm_arm_setup_stage2     // VTCR_EL2 related, 2-level pgtable (not split host PMD huge pages)
        -> kvm_alloc_stage2_pgd
        -> create_hyp_mappings      // kernel already in hyp mode, do nothing
        -> kvm_vgic_early_init      // initialize static VGIC VCPU data structures
    -> kvm_init_mmu_notifier    // maybe for swap???
-> kvm_coalesced_mmio_init
-> get_unused_fd_flag           // alloc vmfd
-> fd_install

2. KVM_SET_USER_MEMORY_REGION (use vmfd, setup guest code)

copy_from_user
kvm_vm_ioctl_set_memory_region
    -> kvm_set_memory_region    // allocate memory, map at the given GPA

3. KVM_CREATE_VCPU (use vmfd, return vcpufd)

kvm_vm_ioctl_create_vcpu
-> kvm_arch_vcpu_create
    -> kmem_cache_zalloc
        -> kvm_vcpu_init
        -> kvm_arch_vcpu_init
            -> kvm_timer_vcpu_init
            -> kvm_vgic_vcpu_init
    -> create_hyp_mappings  // kernel already in hyp mode, do nothing
-> kvm_arch_vcpu_setup      // arm64 do nothing
-> create_vcpu_fd.          // kvm_vcpu_fops: map (struct kvm_run) to userspace
-> kvm_arch_vcpu_postcreate // arm64 do nothing

4. KVM_ARM_VCPU_INIT (use vcpufd)

copy_from_user
kvm_arch_vcpu_ioctl_vcpu_init
-> kvm_vcpu_set_target
-> vcpu_reset_hcr // hcr_el2 |= HCR_E2H, etc.

5. KVM_SET_ONE_REG (use vcpufd)

copy_from_user
kvm_arm_set_reg

6. KVM_RUN (use vcpufd)

kvm_arch_vcpu_ioctl_run
-> kvm_vcpu_first_run_init
-> vcpu_load
    -> kvm_arch_vcpu_load
        -> kvm_arm_set_running_vcpu  // this_cpu_write
        -> kvm_vgic_load
        -> kvm_vcpu_load_sysregs
    -> put_cpu // preempt_enable
-> while loop (ret > 0)
    -> cond_resched + update_vmid + check_vcpu_requests
    -> preempt_disable + local_irq_disable ...
    -> guest_enter_irqoff + kvm_arm_vhe_guest_enter
    -> kvm_vcpu_run_vhe (return exception_index)
        -> __activate_vm            // vtcr_el2, vttbr_el2
        -> __activate_traps         // cpacr_el1, vbar_el1
        -> sysreg_restore_guest_state_vhe
        -> while loop (fixup_guest_exit)
            -> __guest_enter                // arch/arm64/kvm/hyp/entry.S
        -> sysreg_save_guest_state_vhe
        -> __deactivate_traps
        -> sysreg_restore_host_state_vhe
    -> kvm_arm_vhe_guest_exit
    -> local_irq_enable
    -> guest_exit
    -> handle_exit_early + preempt_enable
    -> handle_exit
        -> handle_trap_exceptions   // if exception_index is ARM_EXCEPTION_TRAP
            -> arm_exit_handlers[hsr_ec]
-> vcpu_put
    -> preempt_disable
    -> kvm_arch_vcpu_put
        -> kvm_vcpu_put_sysregs
        -> kvm_vgic_put
        -> kvm_arm_set_running_vcpu
    -> preempt_enable