x86_64.rs

// Copyright 2020 The ChromiumOS Authors
// Use of this source code is governed by a BSD-style license that can be
// found in the LICENSE file.

use std::arch::x86_64::CpuidResult;
#[cfg(any(unix, feature = "haxm", feature = "whpx"))]
use std::arch::x86_64::__cpuid;
use std::arch::x86_64::_rdtsc;
use std::collections::BTreeMap;
use std::collections::HashSet;

use anyhow::Context;
use base::custom_serde::deserialize_seq_to_arr;
use base::custom_serde::serialize_arr;
use base::error;
use base::warn;
use base::Result;
use bit_field::*;
use downcast_rs::impl_downcast;
use libc::c_void;
use serde::Deserialize;
use serde::Serialize;
use snapshot::AnySnapshot;
use vm_memory::GuestAddress;

use crate::Hypervisor;
use crate::IrqRoute;
use crate::IrqSource;
use crate::IrqSourceChip;
use crate::Vcpu;
use crate::Vm;

const MSR_F15H_PERF_CTL0: u32 = 0xc0010200;
const MSR_F15H_PERF_CTL1: u32 = 0xc0010202;
const MSR_F15H_PERF_CTL2: u32 = 0xc0010204;
const MSR_F15H_PERF_CTL3: u32 = 0xc0010206;
const MSR_F15H_PERF_CTL4: u32 = 0xc0010208;
const MSR_F15H_PERF_CTL5: u32 = 0xc001020a;
const MSR_F15H_PERF_CTR0: u32 = 0xc0010201;
const MSR_F15H_PERF_CTR1: u32 = 0xc0010203;
const MSR_F15H_PERF_CTR2: u32 = 0xc0010205;
const MSR_F15H_PERF_CTR3: u32 = 0xc0010207;
const MSR_F15H_PERF_CTR4: u32 = 0xc0010209;
const MSR_F15H_PERF_CTR5: u32 = 0xc001020b;
const MSR_IA32_PERF_CAPABILITIES: u32 = 0x00000345;

/// A trait for managing cpuids for an x86_64 hypervisor and for checking its capabilities.
pub trait HypervisorX86_64: Hypervisor {
    /// Get the system supported CPUID values.
    fn get_supported_cpuid(&self) -> Result<CpuId>;

    /// Gets the list of supported MSRs.
    fn get_msr_index_list(&self) -> Result<Vec<u32>>;
}

/// A wrapper for using a VM on x86_64 and getting/setting its state.
pub trait VmX86_64: Vm {
    /// Gets the `HypervisorX86_64` that created this VM.
    fn get_hypervisor(&self) -> &dyn HypervisorX86_64;

    /// Create a Vcpu with the specified Vcpu ID.
    fn create_vcpu(&self, id: usize) -> Result<Box<dyn VcpuX86_64>>;

    /// Sets the address of the three-page region in the VM's address space.
    fn set_tss_addr(&self, addr: GuestAddress) -> Result<()>;

    /// Sets the address of a one-page region in the VM's address space.
    fn set_identity_map_addr(&self, addr: GuestAddress) -> Result<()>;

    /// Load pVM firmware for the VM, creating a memslot for it as needed.
    ///
    /// Only works on protected VMs (i.e. those with vm_type == KVM_X86_PKVM_PROTECTED_VM).
    fn load_protected_vm_firmware(&mut self, fw_addr: GuestAddress, fw_max_size: u64)
        -> Result<()>;
}

/// A wrapper around creating and using a VCPU on x86_64.
pub trait VcpuX86_64: Vcpu {
    /// Sets or clears the flag that requests the VCPU to exit when it becomes possible to inject
    /// interrupts into the guest.
    fn set_interrupt_window_requested(&self, requested: bool);

    /// Checks if we can inject an interrupt into the VCPU.
    fn ready_for_interrupt(&self) -> bool;

    /// Injects interrupt vector `irq` into the VCPU.
    ///
    /// This function should only be called when [`Self::ready_for_interrupt`] returns true.
    /// Otherwise the interrupt injection may fail or the next VCPU run may fail. However, if
    /// [`Self::interrupt`] returns [`Ok`], the implementation must guarantee that the interrupt
    /// isn't injected in an uninterruptible window (e.g. right after the mov ss instruction).
    ///
    /// The caller should avoid calling this function more than 1 time for one VMEXIT, because the
    /// hypervisor may behave differently: some hypervisors(e.g. WHPX, KVM) will only try to inject
    /// the last `irq` requested, while some other hypervisors(e.g. HAXM) may try to inject all
    /// `irq`s requested.
    fn interrupt(&self, irq: u8) -> Result<()>;

    /// Injects a non-maskable interrupt into the VCPU.
    fn inject_nmi(&self) -> Result<()>;

    /// Gets the VCPU general purpose registers.
    fn get_regs(&self) -> Result<Regs>;

    /// Sets the VCPU general purpose registers.
    fn set_regs(&self, regs: &Regs) -> Result<()>;

    /// Gets the VCPU special registers.
    fn get_sregs(&self) -> Result<Sregs>;

    /// Sets the VCPU special registers.
    fn set_sregs(&self, sregs: &Sregs) -> Result<()>;

    /// Gets the VCPU FPU registers.
    fn get_fpu(&self) -> Result<Fpu>;

    /// Sets the VCPU FPU registers.
    fn set_fpu(&self, fpu: &Fpu) -> Result<()>;

    /// Gets the VCPU debug registers.
    fn get_debugregs(&self) -> Result<DebugRegs>;

    /// Sets the VCPU debug registers.
    fn set_debugregs(&self, debugregs: &DebugRegs) -> Result<()>;

    /// Gets the VCPU extended control registers.
    fn get_xcrs(&self) -> Result<BTreeMap<u32, u64>>;

    /// Sets a VCPU extended control register.
    fn set_xcr(&self, xcr: u32, value: u64) -> Result<()>;

    /// Gets the VCPU x87 FPU, MMX, XMM, YMM and MXCSR registers.
    fn get_xsave(&self) -> Result<Xsave>;

    /// Sets the VCPU x87 FPU, MMX, XMM, YMM and MXCSR registers.
    fn set_xsave(&self, xsave: &Xsave) -> Result<()>;

    /// Gets hypervisor specific state for this VCPU that must be
    /// saved/restored for snapshotting.
    /// This state is fetched after VCPUs are frozen and interrupts are flushed.
    fn get_hypervisor_specific_state(&self) -> Result<AnySnapshot>;

    /// Sets hypervisor specific state for this VCPU. Only used for
    /// snapshotting.
    fn set_hypervisor_specific_state(&self, data: AnySnapshot) -> Result<()>;

    /// Gets a single model-specific register's value.
    fn get_msr(&self, msr_index: u32) -> Result<u64>;

    /// Gets the model-specific registers. Returns all the MSRs for the VCPU.
    fn get_all_msrs(&self) -> Result<BTreeMap<u32, u64>>;

    /// Sets a single model-specific register's value.
    fn set_msr(&self, msr_index: u32, value: u64) -> Result<()>;

    /// Sets up the data returned by the CPUID instruction.
    fn set_cpuid(&self, cpuid: &CpuId) -> Result<()>;

    /// Sets up debug registers and configure vcpu for handling guest debug events.
    fn set_guest_debug(&self, addrs: &[GuestAddress], enable_singlestep: bool) -> Result<()>;

    /// This function should be called after `Vcpu::run` returns `VcpuExit::Cpuid`, and `entry`
    /// should represent the result of emulating the CPUID instruction. The `handle_cpuid` function
    /// will then set the appropriate registers on the vcpu.
    fn handle_cpuid(&mut self, entry: &CpuIdEntry) -> Result<()>;

    /// Gets the guest->host TSC offset.
    ///
    /// The default implementation uses [`VcpuX86_64::get_msr()`] to read the guest TSC.
    fn get_tsc_offset(&self) -> Result<u64> {
        // SAFETY:
        // Safe because _rdtsc takes no arguments
        let host_before_tsc = unsafe { _rdtsc() };

        // get guest TSC value from our hypervisor
        let guest_tsc = self.get_msr(crate::MSR_IA32_TSC)?;

        // SAFETY:
        // Safe because _rdtsc takes no arguments
        let host_after_tsc = unsafe { _rdtsc() };

        // Average the before and after host tsc to get the best value
        let host_tsc = ((host_before_tsc as u128 + host_after_tsc as u128) / 2) as u64;

        Ok(guest_tsc.wrapping_sub(host_tsc))
    }

    /// Sets the guest->host TSC offset.
    ///
    /// The default implementation uses [`VcpuX86_64::set_tsc_value()`] to set the TSC value.
    ///
    /// It sets TSC_OFFSET (VMCS / CB field) by setting the TSC MSR to the current
    /// host TSC value plus the desired offset. We rely on the fact that hypervisors
    /// determine the value of TSC_OFFSET by computing TSC_OFFSET = `new_tsc_value - _rdtsc()` =
    /// `_rdtsc() + offset - _rdtsc()` ~= `offset`. Note that the ~= is important: this is an
    /// approximate operation, because the two _rdtsc() calls
    /// are separated by at least a few ticks.
    ///
    /// Note: TSC_OFFSET, host TSC, guest TSC, and TSC MSR are all different
    /// concepts.
    /// * When a guest executes rdtsc, the value (guest TSC) returned is host_tsc * TSC_MULTIPLIER +
    ///   TSC_OFFSET + TSC_ADJUST.
    /// * The TSC MSR is a special MSR that when written to by the host, will cause TSC_OFFSET to be
    ///   set accordingly by the hypervisor.
    /// * When the guest *writes* to TSC MSR, it actually changes the TSC_ADJUST MSR *for the
    ///   guest*. Generally this is only happens if the guest is trying to re-zero or synchronize
    ///   TSCs.
    fn set_tsc_offset(&self, offset: u64) -> Result<()> {
        // SAFETY: _rdtsc takes no arguments.
        let host_tsc = unsafe { _rdtsc() };
        self.set_tsc_value(host_tsc.wrapping_add(offset))
    }

    /// Sets the guest TSC exactly to the provided value.
    ///
    /// The default implementation sets the guest's TSC by writing the value to the MSR directly.
    ///
    /// See [`VcpuX86_64::set_tsc_offset()`] for an explanation of how this value is actually read
    /// by the guest after being set.
    fn set_tsc_value(&self, value: u64) -> Result<()> {
        self.set_msr(crate::MSR_IA32_TSC, value)
    }

    /// Some hypervisors require special handling to restore timekeeping when
    /// a snapshot is restored. They are provided with a host TSC reference
    /// moment, guaranteed to be the same across all Vcpus, and the Vcpu's TSC
    /// offset at the moment it was snapshotted.
    fn restore_timekeeping(&self, host_tsc_reference_moment: u64, tsc_offset: u64) -> Result<()>;

    /// Snapshot vCPU state
    fn snapshot(&self) -> anyhow::Result<VcpuSnapshot> {
        Ok(VcpuSnapshot {
            vcpu_id: self.id(),
            regs: self.get_regs()?,
            sregs: self.get_sregs()?,
            debug_regs: self.get_debugregs()?,
            xcrs: self.get_xcrs()?,
            msrs: self.get_all_msrs()?,
            xsave: self.get_xsave()?,
            hypervisor_data: self.get_hypervisor_specific_state()?,
            tsc_offset: self.get_tsc_offset()?,
        })
    }

    fn restore(
        &mut self,
        snapshot: &VcpuSnapshot,
        host_tsc_reference_moment: u64,
    ) -> anyhow::Result<()> {
        // List of MSRs that may fail to restore due to lack of support in the host kernel.
        // Some hosts are may be running older kernels which do not support all MSRs, but
        // get_all_msrs will still fetch the MSRs supported by the CPU. Trying to set those MSRs
        // will result in failures, so they will throw a warning instead.
        let msr_allowlist = HashSet::from([
            MSR_F15H_PERF_CTL0,
            MSR_F15H_PERF_CTL1,
            MSR_F15H_PERF_CTL2,
            MSR_F15H_PERF_CTL3,
            MSR_F15H_PERF_CTL4,
            MSR_F15H_PERF_CTL5,
            MSR_F15H_PERF_CTR0,
            MSR_F15H_PERF_CTR1,
            MSR_F15H_PERF_CTR2,
            MSR_F15H_PERF_CTR3,
            MSR_F15H_PERF_CTR4,
            MSR_F15H_PERF_CTR5,
            MSR_IA32_PERF_CAPABILITIES,
        ]);
        assert_eq!(snapshot.vcpu_id, self.id());
        self.set_regs(&snapshot.regs)?;
        self.set_sregs(&snapshot.sregs)?;
        self.set_debugregs(&snapshot.debug_regs)?;
        for (xcr_index, value) in &snapshot.xcrs {
            self.set_xcr(*xcr_index, *value)?;
        }

        for (msr_index, value) in snapshot.msrs.iter() {
            if self.get_msr(*msr_index) == Ok(*value) {
                continue; // no need to set MSR since the values are the same.
            }
            if let Err(e) = self.set_msr(*msr_index, *value) {
                if msr_allowlist.contains(msr_index) {
                    warn!(
                        "Failed to set MSR. MSR might not be supported in this kernel. Err: {}",
                        e
                    );
                } else {
                    return Err(e).context(
                        "Failed to set MSR. MSR might not be supported by the CPU or by the kernel,
                         and was not allow-listed.",
                    );
                }
            };
        }
        self.set_xsave(&snapshot.xsave)?;
        self.set_hypervisor_specific_state(snapshot.hypervisor_data.clone())?;
        self.restore_timekeeping(host_tsc_reference_moment, snapshot.tsc_offset)?;
        Ok(())
    }
}

/// x86 specific vCPU snapshot.
#[derive(Clone, Debug, Serialize, Deserialize)]
pub struct VcpuSnapshot {
    pub vcpu_id: usize,
    regs: Regs,
    sregs: Sregs,
    debug_regs: DebugRegs,
    xcrs: BTreeMap<u32, u64>,
    msrs: BTreeMap<u32, u64>,
    xsave: Xsave,
    hypervisor_data: AnySnapshot,
    tsc_offset: u64,
}

impl_downcast!(VcpuX86_64);

// TSC MSR
pub const MSR_IA32_TSC: u32 = 0x00000010;

/// Gets host cpu max physical address bits.
#[cfg(any(unix, feature = "haxm", feature = "whpx"))]
pub(crate) fn host_phys_addr_bits() -> u8 {
    // SAFETY: trivially safe
    let highest_ext_function = unsafe { __cpuid(0x80000000) };
    if highest_ext_function.eax >= 0x80000008 {
        // SAFETY: trivially safe
        let addr_size = unsafe { __cpuid(0x80000008) };
        // Low 8 bits of 0x80000008 leaf: host physical address size in bits.
        addr_size.eax as u8
    } else {
        36
    }
}

/// Initial state for x86_64 VCPUs.
#[derive(Clone, Default)]
pub struct VcpuInitX86_64 {
    /// General-purpose registers.
    pub regs: Regs,

    /// Special registers.
    pub sregs: Sregs,

    /// Floating-point registers.
    pub fpu: Fpu,

    /// Machine-specific registers.
    pub msrs: BTreeMap<u32, u64>,
}

/// Hold the CPU feature configurations that are needed to setup a vCPU.
#[derive(Clone, Debug, PartialEq, Eq)]
pub struct CpuConfigX86_64 {
    /// whether to force using a calibrated TSC leaf (0x15).
    pub force_calibrated_tsc_leaf: bool,

    /// whether enabling host cpu topology.
    pub host_cpu_topology: bool,

    /// whether expose HWP feature to the guest.
    pub enable_hwp: bool,

    /// Wheter diabling SMT (Simultaneous Multithreading).
    pub no_smt: bool,

    /// whether enabling ITMT scheduler
    pub itmt: bool,

    /// whether setting hybrid CPU type
    pub hybrid_type: Option<CpuHybridType>,
}

impl CpuConfigX86_64 {
    pub fn new(
        force_calibrated_tsc_leaf: bool,
        host_cpu_topology: bool,
        enable_hwp: bool,
        no_smt: bool,
        itmt: bool,
        hybrid_type: Option<CpuHybridType>,
    ) -> Self {
        CpuConfigX86_64 {
            force_calibrated_tsc_leaf,
            host_cpu_topology,
            enable_hwp,
            no_smt,
            itmt,
            hybrid_type,
        }
    }
}

/// A CpuId Entry contains supported feature information for the given processor.
/// This can be modified by the hypervisor to pass additional information to the guest kernel
/// about the hypervisor or vm. Information is returned in the eax, ebx, ecx and edx registers
/// by the cpu for a given function and index/subfunction (passed into the cpu via the eax and ecx
/// register respectively).
#[repr(C)]
#[derive(Clone, Copy, Debug, PartialEq, Eq)]
pub struct CpuIdEntry {
    pub function: u32,
    pub index: u32,
    // flags is needed for KVM.  We store it on CpuIdEntry to preserve the flags across
    // get_supported_cpuids() -> kvm_cpuid2 -> CpuId -> kvm_cpuid2 -> set_cpuid().
    pub flags: u32,
    pub cpuid: CpuidResult,
}

/// A container for the list of cpu id entries for the hypervisor and underlying cpu.
pub struct CpuId {
    pub cpu_id_entries: Vec<CpuIdEntry>,
}

impl CpuId {
    /// Constructs a new CpuId, with space allocated for `initial_capacity` CpuIdEntries.
    pub fn new(initial_capacity: usize) -> Self {
        CpuId {
            cpu_id_entries: Vec::with_capacity(initial_capacity),
        }
    }
}

#[bitfield]
#[derive(Clone, Copy, Debug, PartialEq, Eq)]
pub enum DestinationMode {
    Physical = 0,
    Logical = 1,
}

#[bitfield]
#[derive(Clone, Copy, Debug, PartialEq, Eq)]
pub enum TriggerMode {
    Edge = 0,
    Level = 1,
}

#[bitfield]
#[derive(Debug, Clone, Copy, PartialEq, Eq)]
pub enum DeliveryMode {
    Fixed = 0b000,
    Lowest = 0b001,
    SMI = 0b010,        // System management interrupt
    RemoteRead = 0b011, // This is no longer supported by intel.
    NMI = 0b100,        // Non maskable interrupt
    Init = 0b101,
    Startup = 0b110,
    External = 0b111,
}

// These MSI structures are for Intel's implementation of MSI.  The PCI spec defines most of MSI,
// but the Intel spec defines the format of messages for raising interrupts.  The PCI spec defines
// three u32s -- the address, address_high, and data -- but Intel only makes use of the address and
// data.  The Intel portion of the specification is in Volume 3 section 10.11.
#[bitfield]
#[derive(Clone, Copy, PartialEq, Eq)]
pub struct MsiAddressMessage {
    pub reserved: BitField2,
    #[bits = 1]
    pub destination_mode: DestinationMode,
    pub redirection_hint: BitField1,
    pub reserved_2: BitField8,
    pub destination_id: BitField8,
    // According to Intel's implementation of MSI, these bits must always be 0xfee.
    pub always_0xfee: BitField12,
}

#[bitfield]
#[derive(Clone, Copy, PartialEq, Eq)]
pub struct MsiDataMessage {
    pub vector: BitField8,
    #[bits = 3]
    pub delivery_mode: DeliveryMode,
    pub reserved: BitField3,
    #[bits = 1]
    pub level: Level,
    #[bits = 1]
    pub trigger: TriggerMode,
    pub reserved2: BitField16,
}

#[bitfield]
#[derive(Debug, Clone, Copy, PartialEq, Eq)]
pub enum DeliveryStatus {
    Idle = 0,
    Pending = 1,
}

/// The level of a level-triggered interrupt: asserted or deasserted.
#[bitfield]
#[derive(Debug, Clone, Copy, PartialEq, Eq)]
pub enum Level {
    Deassert = 0,
    Assert = 1,
}

/// Represents a IOAPIC redirection table entry.
#[bitfield]
#[derive(Clone, Copy, Default, PartialEq, Eq, Serialize, Deserialize)]
pub struct IoapicRedirectionTableEntry {
    vector: BitField8,
    #[bits = 3]
    delivery_mode: DeliveryMode,
    #[bits = 1]
    dest_mode: DestinationMode,
    #[bits = 1]
    delivery_status: DeliveryStatus,
    polarity: BitField1,
    remote_irr: bool,
    #[bits = 1]
    trigger_mode: TriggerMode,
    interrupt_mask: bool, // true iff interrupts are masked.
    reserved: BitField39,
    dest_id: BitField8,
}

/// Number of pins on the standard KVM/IOAPIC.
pub const NUM_IOAPIC_PINS: usize = 24;

/// Represents the state of the IOAPIC.
#[repr(C)]
#[derive(Clone, Copy, Debug, PartialEq, Eq, Serialize, Deserialize)]
pub struct IoapicState {
    /// base_address is the memory base address for this IOAPIC. It cannot be changed.
    pub base_address: u64,
    /// ioregsel register. Used for selecting which entry of the redirect table to read/write.
    pub ioregsel: u8,
    /// ioapicid register. Bits 24 - 27 contain the APIC ID for this device.
    pub ioapicid: u32,
    /// current_interrupt_level_bitmap represents a bitmap of the state of all of the irq lines
    pub current_interrupt_level_bitmap: u32,
    /// redirect_table contains the irq settings for each irq line
    #[serde(
        serialize_with = "serialize_arr",
        deserialize_with = "deserialize_seq_to_arr"
    )]
    pub redirect_table: [IoapicRedirectionTableEntry; NUM_IOAPIC_PINS],
}

impl Default for IoapicState {
    fn default() -> IoapicState {
        // SAFETY: trivially safe
        unsafe { std::mem::zeroed() }
    }
}

#[repr(C)]
#[derive(Debug, Clone, Copy, PartialEq, Eq)]
pub enum PicSelect {
    Primary = 0,
    Secondary = 1,
}

#[repr(C)]
#[derive(enumn::N, Debug, Clone, Copy, Default, PartialEq, Eq, Serialize, Deserialize)]
pub enum PicInitState {
    #[default]
    Icw1 = 0,
    Icw2 = 1,
    Icw3 = 2,
    Icw4 = 3,
}

/// Convenience implementation for converting from a u8
impl From<u8> for PicInitState {
    fn from(item: u8) -> Self {
        PicInitState::n(item).unwrap_or_else(|| {
            error!("Invalid PicInitState {}, setting to 0", item);
            PicInitState::Icw1
        })
    }
}

/// Represents the state of the PIC.
#[repr(C)]
#[derive(Clone, Copy, Default, Debug, PartialEq, Eq, Serialize, Deserialize)]
pub struct PicState {
    /// Edge detection.
    pub last_irr: u8,
    /// Interrupt Request Register.
    pub irr: u8,
    /// Interrupt Mask Register.
    pub imr: u8,
    /// Interrupt Service Register.
    pub isr: u8,
    /// Highest priority, for priority rotation.
    pub priority_add: u8,
    pub irq_base: u8,
    pub read_reg_select: bool,
    pub poll: bool,
    pub special_mask: bool,
    pub init_state: PicInitState,
    pub auto_eoi: bool,
    pub rotate_on_auto_eoi: bool,
    pub special_fully_nested_mode: bool,
    /// PIC takes either 3 or 4 bytes of initialization command word during
    /// initialization. use_4_byte_icw is true if 4 bytes of ICW are needed.
    pub use_4_byte_icw: bool,
    /// "Edge/Level Control Registers", for edge trigger selection.
    /// When a particular bit is set, the corresponding IRQ is in level-triggered mode. Otherwise
    /// it is in edge-triggered mode.
    pub elcr: u8,
    pub elcr_mask: u8,
}

/// The LapicState represents the state of an x86 CPU's Local APIC.
/// The Local APIC consists of 64 128-bit registers, but only the first 32-bits of each register
/// can be used, so this structure only stores the first 32-bits of each register.
#[repr(C)]
#[derive(Clone, Copy, Serialize, Deserialize)]
pub struct LapicState {
    #[serde(
        serialize_with = "serialize_arr",
        deserialize_with = "deserialize_seq_to_arr"
    )]
    pub regs: [LapicRegister; 64],
}

pub type LapicRegister = u32;

// rust arrays longer than 32 need custom implementations of Debug
impl std::fmt::Debug for LapicState {
    fn fmt(&self, formatter: &mut std::fmt::Formatter) -> std::fmt::Result {
        self.regs[..].fmt(formatter)
    }
}

// rust arrays longer than 32 need custom implementations of PartialEq
impl PartialEq for LapicState {
    fn eq(&self, other: &LapicState) -> bool {
        self.regs[..] == other.regs[..]
    }
}

// Lapic equality is reflexive, so we impl Eq
impl Eq for LapicState {}

/// The PitState represents the state of the PIT (aka the Programmable Interval Timer).
/// The state is simply the state of it's three channels.
#[repr(C)]
#[derive(Clone, Copy, Debug, PartialEq, Eq, Serialize, Deserialize)]
pub struct PitState {
    pub channels: [PitChannelState; 3],
    /// Hypervisor-specific flags for setting the pit state.
    pub flags: u32,
}

/// The PitRWMode enum represents the access mode of a PIT channel.
/// Reads and writes to the Pit happen over Port-mapped I/O, which happens one byte at a time,
/// but the count values and latch values are two bytes. So the access mode controls which of the
/// two bytes will be read when.
#[repr(C)]
#[derive(enumn::N, Clone, Copy, Debug, PartialEq, Eq, Serialize, Deserialize)]
pub enum PitRWMode {
    /// None mode means that no access mode has been set.
    None = 0,
    /// Least mode means all reads/writes will read/write the least significant byte.
    Least = 1,
    /// Most mode means all reads/writes will read/write the most significant byte.
    Most = 2,
    /// Both mode means first the least significant byte will be read/written, then the
    /// next read/write will read/write the most significant byte.
    Both = 3,
}

/// Convenience implementation for converting from a u8
impl From<u8> for PitRWMode {
    fn from(item: u8) -> Self {
        PitRWMode::n(item).unwrap_or_else(|| {
            error!("Invalid PitRWMode value {}, setting to 0", item);
            PitRWMode::None
        })
    }
}

/// The PitRWState enum represents the state of reading to or writing from a channel.
/// This is related to the PitRWMode, it mainly gives more detail about the state of the channel
/// with respect to PitRWMode::Both.
#[repr(C)]
#[derive(enumn::N, Clone, Copy, Debug, PartialEq, Eq, Serialize, Deserialize)]
pub enum PitRWState {
    /// None mode means that no access mode has been set.
    None = 0,
    /// LSB means that the channel is in PitRWMode::Least access mode.
    LSB = 1,
    /// MSB means that the channel is in PitRWMode::Most access mode.
    MSB = 2,
    /// Word0 means that the channel is in PitRWMode::Both mode, and the least sginificant byte
    /// has not been read/written yet.
    Word0 = 3,
    /// Word1 means that the channel is in PitRWMode::Both mode and the least significant byte
    /// has already been read/written, and the next byte to be read/written will be the most
    /// significant byte.
    Word1 = 4,
}

/// Convenience implementation for converting from a u8
impl From<u8> for PitRWState {
    fn from(item: u8) -> Self {
        PitRWState::n(item).unwrap_or_else(|| {
            error!("Invalid PitRWState value {}, setting to 0", item);
            PitRWState::None
        })
    }
}

/// The PitChannelState represents the state of one of the PIT's three counters.
#[repr(C)]
#[derive(Clone, Copy, Debug, PartialEq, Eq, Serialize, Deserialize)]
pub struct PitChannelState {
    /// The starting value for the counter.
    pub count: u32,
    /// Stores the channel count from the last time the count was latched.
    pub latched_count: u16,
    /// Indicates the PitRWState state of reading the latch value.
    pub count_latched: PitRWState,
    /// Indicates whether ReadBack status has been latched.
    pub status_latched: bool,
    /// Stores the channel status from the last time the status was latched. The status contains
    /// information about the access mode of this channel, but changing those bits in the status
    /// will not change the behavior of the pit.
    pub status: u8,
    /// Indicates the PitRWState state of reading the counter.
    pub read_state: PitRWState,
    /// Indicates the PitRWState state of writing the counter.
    pub write_state: PitRWState,
    /// Stores the value with which the counter was initialized. Counters are 16-
    /// bit values with an effective range of 1-65536 (65536 represented by 0).
    pub reload_value: u16,
    /// The command access mode of this channel.
    pub rw_mode: PitRWMode,
    /// The operation mode of this channel.
    pub mode: u8,
    /// Whether or not we are in bcd mode. Not supported by KVM or crosvm's PIT implementation.
    pub bcd: bool,
    /// Value of the gate input pin. This only applies to channel 2.
    pub gate: bool,
    /// Nanosecond timestamp of when the count value was loaded.
    pub count_load_time: u64,
}

// Convenience constructors for IrqRoutes
impl IrqRoute {
    pub fn ioapic_irq_route(irq_num: u32) -> IrqRoute {
        IrqRoute {
            gsi: irq_num,
            source: IrqSource::Irqchip {
                chip: IrqSourceChip::Ioapic,
                pin: irq_num,
            },
        }
    }

    pub fn pic_irq_route(id: IrqSourceChip, irq_num: u32) -> IrqRoute {
        IrqRoute {
            gsi: irq_num,
            source: IrqSource::Irqchip {
                chip: id,
                pin: irq_num % 8,
            },
        }
    }
}

/// State of a VCPU's general purpose registers.
#[repr(C)]
#[derive(Debug, Copy, Clone, Serialize, Deserialize)]
pub struct Regs {
    pub rax: u64,
    pub rbx: u64,
    pub rcx: u64,
    pub rdx: u64,
    pub rsi: u64,
    pub rdi: u64,
    pub rsp: u64,
    pub rbp: u64,
    pub r8: u64,
    pub r9: u64,
    pub r10: u64,
    pub r11: u64,
    pub r12: u64,
    pub r13: u64,
    pub r14: u64,
    pub r15: u64,
    pub rip: u64,
    pub rflags: u64,
}

impl Default for Regs {
    fn default() -> Self {
        Regs {
            rax: 0,
            rbx: 0,
            rcx: 0,
            rdx: 0,
            rsi: 0,
            rdi: 0,
            rsp: 0,
            rbp: 0,
            r8: 0,
            r9: 0,
            r10: 0,
            r11: 0,
            r12: 0,
            r13: 0,
            r14: 0,
            r15: 0,
            rip: 0xfff0, // Reset vector.
            rflags: 0x2, // Bit 1 (0x2) is always 1.
        }
    }
}

/// State of a memory segment.
#[repr(C)]
#[derive(Debug, Default, Copy, Clone, Serialize, Deserialize, PartialEq, Eq)]
pub struct Segment {
    pub base: u64,
    /// Limit of the segment - always in bytes, regardless of granularity (`g`) field.
    pub limit_bytes: u32,
    pub selector: u16,
    pub type_: u8,
    pub present: u8,
    pub dpl: u8,
    pub db: u8,
    pub s: u8,
    pub l: u8,
    pub g: u8,
    pub avl: u8,
}

/// State of a global descriptor table or interrupt descriptor table.
#[repr(C)]
#[derive(Debug, Default, Copy, Clone, Serialize, Deserialize)]
pub struct DescriptorTable {
    pub base: u64,
    pub limit: u16,
}

/// State of a VCPU's special registers.
#[repr(C)]
#[derive(Debug, Copy, Clone, Serialize, Deserialize)]
pub struct Sregs {
    pub cs: Segment,
    pub ds: Segment,
    pub es: Segment,
    pub fs: Segment,
    pub gs: Segment,
    pub ss: Segment,
    pub tr: Segment,
    pub ldt: Segment,
    pub gdt: DescriptorTable,
    pub idt: DescriptorTable,
    pub cr0: u64,
    pub cr2: u64,
    pub cr3: u64,
    pub cr4: u64,
    pub cr8: u64,
    pub efer: u64,
}

impl Default for Sregs {
    fn default() -> Self {
        // Intel SDM Vol. 3A, 3.4.5.1 ("Code- and Data-Segment Descriptor Types")
        const SEG_TYPE_DATA: u8 = 0b0000;
        const SEG_TYPE_DATA_WRITABLE: u8 = 0b0010;

        const SEG_TYPE_CODE: u8 = 0b1000;
        const SEG_TYPE_CODE_READABLE: u8 = 0b0010;

        const SEG_TYPE_ACCESSED: u8 = 0b0001;

        // Intel SDM Vol. 3A, 3.4.5 ("Segment Descriptors")
        const SEG_S_SYSTEM: u8 = 0; // System segment.
        const SEG_S_CODE_OR_DATA: u8 = 1; // Data/code segment.

        // 16-bit real-mode code segment (reset vector).
        let code_seg = Segment {
            base: 0xffff0000,
            limit_bytes: 0xffff,
            selector: 0xf000,
            type_: SEG_TYPE_CODE | SEG_TYPE_CODE_READABLE | SEG_TYPE_ACCESSED, // 11
            present: 1,
            s: SEG_S_CODE_OR_DATA,
            ..Default::default()
        };

        // 16-bit real-mode data segment.
        let data_seg = Segment {
            base: 0,
            limit_bytes: 0xffff,
            selector: 0,
            type_: SEG_TYPE_DATA | SEG_TYPE_DATA_WRITABLE | SEG_TYPE_ACCESSED, // 3
            present: 1,
            s: SEG_S_CODE_OR_DATA,
            ..Default::default()
        };

        // 16-bit TSS segment.
        let task_seg = Segment {
            base: 0,
            limit_bytes: 0xffff,
            selector: 0,
            type_: SEG_TYPE_CODE | SEG_TYPE_CODE_READABLE | SEG_TYPE_ACCESSED, // 11
            present: 1,
            s: SEG_S_SYSTEM,
            ..Default::default()
        };

        // Local descriptor table.
        let ldt = Segment {
            base: 0,
            limit_bytes: 0xffff,
            selector: 0,
            type_: SEG_TYPE_DATA | SEG_TYPE_DATA_WRITABLE, // 2
            present: 1,
            s: SEG_S_SYSTEM,
            ..Default::default()
        };

        // Global descriptor table.
        let gdt = DescriptorTable {
            base: 0,
            limit: 0xffff,
        };

        // Interrupt descriptor table.
        let idt = DescriptorTable {
            base: 0,
            limit: 0xffff,
        };

        let cr0 = (1 << 4) // CR0.ET (reserved, always 1)
                | (1 << 30); // CR0.CD (cache disable)

        Sregs {
            cs: code_seg,
            ds: data_seg,
            es: data_seg,
            fs: data_seg,
            gs: data_seg,
            ss: data_seg,
            tr: task_seg,
            ldt,
            gdt,
            idt,
            cr0,
            cr2: 0,
            cr3: 0,
            cr4: 0,
            cr8: 0,
            efer: 0,
        }
    }
}

/// x87 80-bit floating point value.
#[repr(C)]
#[derive(Copy, Clone, Debug, Default, Eq, PartialEq, Serialize, Deserialize)]
pub struct FpuReg {
    /// 64-bit mantissa.
    pub significand: u64,

    /// 15-bit biased exponent and sign bit.
    pub sign_exp: u16,
}

impl FpuReg {
    /// Convert an array of 8x16-byte arrays to an array of 8 `FpuReg`.
    ///
    /// Ignores any data in the upper 6 bytes of each element; the values represent 80-bit FPU
    /// registers, so the upper 48 bits are unused.
    pub fn from_16byte_arrays(byte_arrays: &[[u8; 16]; 8]) -> [FpuReg; 8] {
        let mut regs = [FpuReg::default(); 8];
        for (dst, src) in regs.iter_mut().zip(byte_arrays.iter()) {
            let tbyte: [u8; 10] = src[0..10].try_into().unwrap();
            *dst = FpuReg::from(tbyte);
        }
        regs
    }

    /// Convert an array of 8 `FpuReg` into 8x16-byte arrays.
    pub fn to_16byte_arrays(regs: &[FpuReg; 8]) -> [[u8; 16]; 8] {
        let mut byte_arrays = [[0u8; 16]; 8];
        for (dst, src) in byte_arrays.iter_mut().zip(regs.iter()) {
            *dst = (*src).into();
        }
        byte_arrays
    }
}

impl From<[u8; 10]> for FpuReg {
    /// Construct a `FpuReg` from an 80-bit representation.
    fn from(value: [u8; 10]) -> FpuReg {
        // These array sub-slices can't fail, but there's no (safe) way to express that in Rust
        // without an `unwrap()`.
        let significand_bytes = value[0..8].try_into().unwrap();
        let significand = u64::from_le_bytes(significand_bytes);
        let sign_exp_bytes = value[8..10].try_into().unwrap();
        let sign_exp = u16::from_le_bytes(sign_exp_bytes);
        FpuReg {
            significand,
            sign_exp,
        }
    }
}

impl From<FpuReg> for [u8; 10] {
    /// Convert an `FpuReg` into its 80-bit "TBYTE" representation.
    fn from(value: FpuReg) -> [u8; 10] {
        let mut bytes = [0u8; 10];
        bytes[0..8].copy_from_slice(&value.significand.to_le_bytes());
        bytes[8..10].copy_from_slice(&value.sign_exp.to_le_bytes());
        bytes
    }
}

impl From<FpuReg> for [u8; 16] {
    /// Convert an `FpuReg` into its 80-bit representation plus 6 unused upper bytes.
    /// This is a convenience function for converting to hypervisor types.
    fn from(value: FpuReg) -> [u8; 16] {
        let mut bytes = [0u8; 16];
        bytes[0..8].copy_from_slice(&value.significand.to_le_bytes());
        bytes[8..10].copy_from_slice(&value.sign_exp.to_le_bytes());
        bytes
    }
}

/// State of a VCPU's floating point unit.
#[repr(C)]
#[derive(Debug, Copy, Clone, Serialize, Deserialize)]
pub struct Fpu {
    pub fpr: [FpuReg; 8],
    pub fcw: u16,
    pub fsw: u16,
    pub ftwx: u8,
    pub last_opcode: u16,
    pub last_ip: u64,
    pub last_dp: u64,
    pub xmm: [[u8; 16usize]; 16usize],
    pub mxcsr: u32,
}

impl Default for Fpu {
    fn default() -> Self {
        Fpu {
            fpr: Default::default(),
            fcw: 0x37f, // Intel SDM Vol. 1, 13.6
            fsw: 0,
            ftwx: 0,
            last_opcode: 0,
            last_ip: 0,
            last_dp: 0,
            xmm: Default::default(),
            mxcsr: 0x1f80, // Intel SDM Vol. 1, 11.6.4
        }
    }
}

/// State of a VCPU's debug registers.
#[repr(C)]
#[derive(Debug, Default, Copy, Clone, Serialize, Deserialize)]
pub struct DebugRegs {
    pub db: [u64; 4usize],
    pub dr6: u64,
    pub dr7: u64,
}

/// The hybrid type for intel hybrid CPU.
#[derive(Clone, Copy, Debug, PartialEq, Eq, Serialize, Deserialize)]
pub enum CpuHybridType {
    /// Intel Atom.
    Atom,
    /// Intel Core.
    Core,
}

/// State of the VCPU's x87 FPU, MMX, XMM, YMM registers.
/// May contain more state depending on enabled extensions.
#[derive(Clone, Debug, Serialize, Deserialize)]
pub struct Xsave {
    data: Vec<u32>,

    // Actual length in bytes. May be smaller than data if a non-u32 multiple of bytes is
    // requested.
    len: usize,
}

impl Xsave {
    /// Create a new buffer to store Xsave data.
    ///
    /// # Argments
    /// * `len` size in bytes.
    pub fn new(len: usize) -> Self {
        Xsave {
            data: vec![0; len.div_ceil(4)],
            len,
        }
    }

    pub fn as_ptr(&self) -> *const c_void {
        self.data.as_ptr() as *const c_void
    }

    pub fn as_mut_ptr(&mut self) -> *mut c_void {
        self.data.as_mut_ptr() as *mut c_void
    }

    /// Length in bytes of the XSAVE data.
    pub fn len(&self) -> usize {
        self.len
    }

    /// Returns true is length of XSAVE data is zero
    pub fn is_empty(&self) -> bool {
        self.len() == 0
    }
}