Extensive refactor of IO,PCI,Virtio

src/devices/irq_event.rs

@@ -0,0 +1,45 @@
+use std::{io, result};
+use vmm_sys_util::eventfd::EventFd;
+use crate::vm::KvmVm;
+
+pub struct IrqLevelEvent {
+    trigger_event: EventFd,
+    resample_event: EventFd,
+}
+
+type Result<T> = result::Result<T, io::Error>;
+
+impl IrqLevelEvent {
+    pub fn register(kvm_vm: &KvmVm, irq: u8) -> Result<Self> {
+        let ev = Self::new()?;
+        kvm_vm.vm_fd()
+            .register_irqfd_with_resample(&ev.trigger_event, &ev.resample_event, irq as u32)?;
+        Ok(ev)
+    }
+
+    pub fn new() -> Result<Self> {
+        let trigger_event = EventFd::new(0)?;
+        let resample_event = EventFd::new(0)?;
+        Ok(IrqLevelEvent {
+            trigger_event, resample_event,
+        })
+    }
+
+    pub fn try_clone(&self) -> Result<IrqLevelEvent> {
+        let trigger_event = self.trigger_event.try_clone()?;
+        let resample_event = self.resample_event.try_clone()?;
+        Ok(IrqLevelEvent {
+            trigger_event,
+            resample_event,
+        })
+    }
+
+    pub fn trigger(&self) -> Result<()> {
+        self.trigger_event.write(1)
+    }
+
+    pub fn wait_resample(&self) -> Result<()> {
+        let _ = self.resample_event.read()?;
+        Ok(())
+    }
+}

src/devices/rtc.rs

@@ -1,8 +1,7 @@
-use std::sync::{Arc,RwLock};
 use std::mem;
 use libc;
-
+use crate::io::bus::BusDevice;
-use crate::vm::io::{IoDispatcher,IoPortOps};
+use crate::io::ReadableInt;
 
 const RTC_SECONDS: u8 = 0x00;
 const RTC_MINUTES: u8 = 0x02;
@@ -21,31 +20,30 @@ pub struct Rtc {
     data: [u8; 128]
 }
 
-impl IoPortOps for Rtc {
+impl BusDevice for Rtc {
-    fn io_in(&mut self, port: u16, _size: usize) -> u32 {
+    fn read(&mut self, offset: u64, data: &mut [u8]) {
-        if port == 0x0071 {
+        if offset == 1 && data.len() == 1 {
-            self.data_in() as u32
+            ReadableInt::new_byte(self.data_in())
+                .read(data);
         } else {
-            0
+            data.fill(0);
         }
     }
 
-    fn io_out(&mut self, port: u16, _size: usize, val: u32) {
+    fn write(&mut self, offset: u64, data: &[u8]) {
-        if port == 0x0070 {
+        if data.len() == 1 {
-            self.index_out(val as u8);
+            match offset {
-        } else if port == 0x0071 {
+                0 => self.index_out(data[0]),
-            self.data_out(val as u8)
+                1 => self.data_out(data[0]),
+                _ => {},
+            }
         }
     }
 }
 
 impl Rtc {
-    pub fn register(io: Arc<IoDispatcher>) {
-        let rtc = Arc::new(RwLock::new(Rtc::new()));
-        io.register_ioports(0x0070, 2, rtc);
-    }
 
-    fn new() -> Rtc {
+    pub fn new() -> Rtc {
         Rtc {
             idx:0,
             data: [0; 128]

src/devices/serial.rs

@@ -1,7 +1,6 @@
-use std::sync::{Arc, RwLock};
 use std::io::{self, Write};
+use crate::io::bus::BusDevice;
 
-use crate::vm::io::{IoPortOps,IoDispatcher};
 use crate::vm::KvmVm;
 
 const UART_TX: u16 = 0;
@@ -43,7 +42,29 @@ const UART_SCR: u16 = 7;
 
 const FIFO_LEN: usize = 64;
 
+pub enum SerialPort {
+    COM1,
+    COM2,
+    COM3,
+    COM4,
+}
 
+impl SerialPort {
+    pub fn io_port(&self) -> u16 {
+        match self {
+            SerialPort::COM1 => 0x3f8,
+            SerialPort::COM2 => 0x2f8,
+            SerialPort::COM3 => 0x3e8,
+            SerialPort::COM4 => 0x2e8,
+        }
+    }
+    pub fn irq(&self) -> u8 {
+        match self {
+            SerialPort::COM1|SerialPort::COM3 => 4,
+            SerialPort::COM2|SerialPort::COM4 => 4,
+        }
+    }
+}
 
 trait Bits {
     fn set(&mut self, flag: Self);
@@ -66,7 +87,6 @@ impl Bits for u8 {
 }
 
 pub struct SerialDevice {
-    iobase: u16,
     kvm_vm: KvmVm,
     irq: u8,
     irq_state: u8,
@@ -87,15 +107,17 @@ pub struct SerialDevice {
     scr: u8,
 }
 
-impl IoPortOps for SerialDevice {
+impl BusDevice for SerialDevice {
-    fn io_in(&mut self, port: u16, _size: usize) -> u32 {
+    fn read(&mut self, offset: u64, data: &mut [u8]) {
-        let off = port - self.iobase;
+        if data.len() == 1 {
-        self.serial_in(off) as u32
+            data[0] = self.serial_in(offset as u16);
+        }
     }
 
-    fn io_out(&mut self, port: u16, _size: usize, val: u32) {
+    fn write(&mut self, offset: u64, data: &[u8]) {
-        let off = port - self.iobase;
+        if data.len() == 1 {
-        self.serial_out(off, val as u8);
+            self.serial_out(offset as u16, data[0])
+        }
     }
 }
 
@@ -270,6 +292,7 @@ impl SerialDevice {
         }
     }
 
+    /*
     pub fn register(kvm_vm: KvmVm, io: Arc<IoDispatcher>, id: u8) {
         if let Some((base,irq)) = SerialDevice::base_irq_for_id(id) {
             let dev = SerialDevice::new(kvm_vm, base, irq);
@@ -287,9 +310,11 @@ impl SerialDevice {
         }
     }
 
-    fn new(kvm_vm: KvmVm, iobase: u16, irq: u8) -> SerialDevice {
+     */
+
+    pub fn new(kvm_vm: KvmVm, irq: u8) -> SerialDevice {
         SerialDevice {
-            iobase,
+//            iobase,
             kvm_vm,
             irq,
             irq_state: 0,

src/devices/virtio_9p/mod.rs

@@ -1,10 +1,8 @@
-use std::sync::{Arc,RwLock};
 use std::thread;
 
 use std::path::{PathBuf, Path};
 
-use crate::memory::{GuestRam, MemoryManager};
+use crate::memory::GuestRam;
-use crate::virtio::{self,VirtioBus,VirtioDeviceOps, VirtQueue, Result};
 use crate::devices::virtio_9p::server::Server;
 use crate::devices::virtio_9p::filesystem::{FileSystem, FileSystemOps};
 use self::pdu::PduParser;
@@ -16,16 +14,15 @@ mod filesystem;
 mod server;
 mod synthetic;
 
-
-const VIRTIO_ID_9P: u16 = 9;
 const VIRTIO_9P_MOUNT_TAG: u64 = 0x1;
 
 pub use synthetic::SyntheticFS;
+use crate::io::{FeatureBits, Queues, VirtioDevice, VirtioDeviceType, VirtQueue};
 
 pub struct VirtioP9<T: FileSystemOps> {
     filesystem: T,
     root_dir: PathBuf,
-    feature_bits: u64,
+    features: FeatureBits,
     debug: bool,
     config: Vec<u8>,
 }
@@ -41,52 +38,54 @@ impl <T: FileSystemOps+'static> VirtioP9<T> {
         config
     }
 
-    fn new(filesystem: T, tag_name: &str, root_dir: &str, debug: bool) -> Arc<RwLock<Self>> {
+    pub fn new(filesystem: T, tag_name: &str, root_dir: &str, debug: bool) -> Self {
-        Arc::new(RwLock::new(VirtioP9 {
+        VirtioP9 {
             filesystem,
             root_dir: PathBuf::from(root_dir),
-            feature_bits: 0,
+            features: FeatureBits::new_default(VIRTIO_9P_MOUNT_TAG),
             debug,
             config: VirtioP9::<T>::create_config(tag_name),
-        }))
+        }
     }
 
-    pub fn create_with_filesystem(filesystem: T, vbus: &mut VirtioBus, tag_name: &str, root_dir: &str, debug: bool) -> Result<()> {
-        vbus.new_virtio_device(VIRTIO_ID_9P, VirtioP9::new(filesystem, tag_name, root_dir, debug))
-            .set_num_queues(1)
-            .set_features(VIRTIO_9P_MOUNT_TAG)
-            .set_config_size(tag_name.len() + 3)
-            .register()
-    }
 }
 
 impl VirtioP9<FileSystem> {
-
+    pub fn new_filesystem(tag_name: &str, root_dir: &str, read_only: bool, debug: bool) -> Self {
-    pub fn create(vbus: &mut VirtioBus, tag_name: &str, root_dir: &str, read_only: bool, debug: bool) -> Result<()> {
         let filesystem = FileSystem::new(PathBuf::from(root_dir), read_only);
-        Self::create_with_filesystem(filesystem, vbus, tag_name, root_dir, debug)
+        Self::new(filesystem, tag_name, root_dir, debug)
     }
 }
 
-impl <T: FileSystemOps+'static> VirtioDeviceOps for VirtioP9<T> {
+impl <T: FileSystemOps+'static> VirtioDevice for VirtioP9<T> {
-    fn reset(&mut self) {
+    fn features(&self) -> &FeatureBits {
-        println!("Reset called");
+        &self.features
     }
 
-    fn enable_features(&mut self, bits: u64) -> bool {
+    fn queue_sizes(&self) -> &[u16] {
-        self.feature_bits = bits;
+        &[VirtQueue::DEFAULT_QUEUE_SIZE]
-        true
     }
 
-    fn read_config(&mut self, offset: usize, size: usize) -> u64 {
+    fn device_type(&self) -> VirtioDeviceType {
-        virtio::read_config_buffer(&self.config, offset, size)
+        VirtioDeviceType::NineP
     }
 
-    fn start(&mut self, memory: &MemoryManager, mut queues: Vec<VirtQueue>) {
+    fn config_size(&self) -> usize {
-        let vq = queues.pop().unwrap();
+        self.config.len()
+    }
+
+    fn read_config(&self, offset: u64, data: &mut [u8]) {
+        let offset = offset as usize;
+        if offset + data.len() <= self.config.len() {
+            data.copy_from_slice(&self.config[offset..offset+data.len()])
+        }
+    }
+
+    fn start(&mut self, queues: &Queues) {
+        let vq = queues.get_queue(0);
         let root_dir = self.root_dir.clone();
         let filesystem = self.filesystem.clone();
-        let ram = memory.guest_ram().clone();
+        let ram = queues.memory().guest_ram().clone();
         let debug = self.debug;
         thread::spawn(move || run_device(ram, vq, &root_dir, filesystem, debug));
     }

src/devices/virtio_9p/pdu.rs

@@ -6,8 +6,8 @@ use libc;
 use byteorder::{LittleEndian,ReadBytesExt,WriteBytesExt};
 
 use crate::devices::virtio_9p::file::Qid;
+use crate::io::Chain;
 use crate::memory::GuestRam;
-use crate::virtio::Chain;
 
 const P9_HEADER_LEN: usize = 7;
 const P9_RLERROR: u8 = 7;

src/devices/virtio_block.rs

@@ -1,13 +1,12 @@
 use std::io::Write;
-use std::sync::{RwLock, Arc};
 use std::{result, io, thread};
 
-use crate::{disk, virtio};
+use crate::disk;
-use crate::virtio::{VirtioBus, VirtioDeviceOps, VirtQueue, DeviceConfigArea, Chain};
-use crate::memory::MemoryManager;
 use crate::disk::DiskImage;
 
 use thiserror::Error;
+use crate::io::{Chain, FeatureBits, Queues, VirtioDevice, VirtioDeviceType, VirtioError, VirtQueue};
+use crate::io::virtio::DeviceConfigArea;
 
 const VIRTIO_BLK_F_RO: u64 = 1 << 5;
 const VIRTIO_BLK_F_BLK_SIZE: u64 = 1 << 6;
@@ -39,7 +38,7 @@ enum Error {
     #[error("error flushing disk image: {0}")]
     DiskFlush(disk::Error),
     #[error("error waiting on virtqueue: {0}")]
-    VirtQueueWait(virtio::Error),
+    VirtQueueWait(VirtioError),
     #[error("virtqueue read descriptor size ({0}) is invalid. Not a multiple of sector size")]
     InvalidReadDescriptor(usize),
 }
@@ -49,66 +48,66 @@ type Result<T> = result::Result<T, Error>;
 pub struct VirtioBlock<D: DiskImage+'static> {
     disk_image: Option<D>,
     config: DeviceConfigArea,
-    enabled_features: u64,
+    features: FeatureBits,
 }
 
 const HEADER_SIZE: usize = 16;
 
-const VIRTIO_ID_BLOCK: u16 = 2;
 const CAPACITY_OFFSET: usize = 0;
 const SEG_MAX_OFFSET: usize = 12;
 const BLK_SIZE_OFFSET: usize = 20;
 const CONFIG_SIZE: usize = 24;
 impl <D: DiskImage + 'static> VirtioBlock<D> {
 
-    fn new(disk_image: D) -> Self {
+    pub fn new(disk_image: D) -> Self {
         let mut config = DeviceConfigArea::new(CONFIG_SIZE);
         config.write_u64(CAPACITY_OFFSET, disk_image.sector_count());
         config.write_u32(SEG_MAX_OFFSET, QUEUE_SIZE as u32 - 2);
         config.write_u32(BLK_SIZE_OFFSET, 1024);
-        VirtioBlock {
+        let features = FeatureBits::new_default( VIRTIO_BLK_F_FLUSH |
-            disk_image: Some(disk_image),
-            config,
-            enabled_features: 0,
-        }
-    }
-
-    pub fn create(vbus: &mut VirtioBus, disk_image: D) -> virtio::Result<()> {
-        let feature_bits = VIRTIO_BLK_F_FLUSH |
                 VIRTIO_BLK_F_BLK_SIZE |
                 VIRTIO_BLK_F_SEG_MAX  |
                 if disk_image.read_only() {
                     VIRTIO_BLK_F_RO
                 } else {
                     0
-            };
+                }
-
+        );
-        let dev = Arc::new(RwLock::new(VirtioBlock::new(disk_image)));
+        VirtioBlock {
-
+            disk_image: Some(disk_image),
-        vbus.new_virtio_device(VIRTIO_ID_BLOCK, dev)
+            config,
-            .set_queue_sizes(&[QUEUE_SIZE])
+            features,
-            .set_config_size(CONFIG_SIZE)
+        }
-            .set_features(feature_bits)
-            .register()
     }
 }
 
-impl <D: DiskImage> VirtioDeviceOps for VirtioBlock<D> {
+impl <D: DiskImage> VirtioDevice for VirtioBlock<D> {
-    fn enable_features(&mut self, bits: u64) -> bool {
+    fn features(&self) -> &FeatureBits {
-        self.enabled_features = bits;
+        &self.features
-        true
     }
 
-    fn write_config(&mut self, offset: usize, size: usize, val: u64) {
+    fn queue_sizes(&self) -> &[u16] {
-        self.config.write_config(offset, size, val);
+        &[QUEUE_SIZE as u16]
     }
 
-    fn read_config(&mut self, offset: usize, size: usize) -> u64 {
+    fn device_type(&self) -> VirtioDeviceType {
-        self.config.read_config(offset, size)
+        VirtioDeviceType::Block
     }
 
-    fn start(&mut self, _: &MemoryManager, mut queues: Vec<VirtQueue>) {
+    fn config_size(&self) -> usize {
-        let vq = queues.pop().unwrap();
+        CONFIG_SIZE
+    }
+
+    fn read_config(&self, offset: u64, data: &mut [u8]) {
+        self.config.read_config(offset, data);
+    }
+
+    fn write_config(&mut self, offset: u64, data: &[u8]) {
+        self.config.write_config(offset, data);
+    }
+
+    fn start(&mut self, queues: &Queues) {
+        let vq = queues.get_queue(0);
 
         let mut disk = self.disk_image.take().expect("No disk image?");
         if let Err(err) = disk.open() {

src/devices/virtio_net.rs

@@ -1,7 +1,4 @@
-use crate::virtio::{VirtioDeviceOps, VirtQueue, VirtioBus, Chain};
+use crate::system;
-use crate::memory::MemoryManager;
-use crate::{system, virtio};
-use std::sync::{RwLock, Arc};
 use std::{result, thread, io};
 use crate::system::{EPoll,Event};
 use std::io::{Read, Write};
@@ -9,8 +6,8 @@ use std::os::unix::io::AsRawFd;
 use crate::system::Tap;
 
 use thiserror::Error;
+use crate::io::{Chain, FeatureBits, Queues, VirtioDevice, VirtioDeviceType, VirtQueue};
 
-const VIRTIO_ID_NET: u16 = 1;
 const MAC_ADDR_LEN: usize = 6;
 
 #[derive(Debug,Error)]
@@ -46,19 +43,12 @@ const VIRTIO_NET_F_HOST_ECN: u64 = 1 << 13;
 const VIRTIO_NET_HDR_SIZE: i32 = 12;
 
 pub struct VirtioNet {
-    _features_supported: u64,
+    features: FeatureBits,
     tap: Option<Tap>,
 }
 
 impl VirtioNet {
-    fn new(tap: Tap, features_supported: u64) -> Self {
+    pub fn new(tap: Tap) -> Self {
-        VirtioNet{
-            _features_supported: features_supported,
-            tap: Some(tap)
-        }
-    }
-
-    pub fn create(vbus: &mut VirtioBus, tap: Tap) -> virtio::Result<()> {
         tap.set_offload(TUN_F_CSUM | TUN_F_TSO4 | TUN_F_TSO6| TUN_F_TSO_ECN).unwrap();
         tap.set_vnet_hdr_size(VIRTIO_NET_HDR_SIZE).unwrap();
         let feature_bits =
@@ -70,25 +60,44 @@ impl VirtioNet {
                 VIRTIO_NET_F_HOST_TSO4 |
                 VIRTIO_NET_F_HOST_TSO6 |
                 VIRTIO_NET_F_HOST_ECN;
-
+        let features = FeatureBits::new_default(feature_bits);
-        let dev = Arc::new(RwLock::new(VirtioNet::new(tap, feature_bits)));
+        VirtioNet{
-        vbus.new_virtio_device(VIRTIO_ID_NET, dev)
+            features,
-            .set_queue_sizes(&[256, 256])
+            tap: Some(tap)
-            .set_config_size(MAC_ADDR_LEN)
-            .set_features(feature_bits)
-            .register()
         }
     }
 
-pub const TUN_F_CSUM: u32 = 1;
+}
-pub const TUN_F_TSO4: u32 = 2;
+
-pub const TUN_F_TSO6: u32 = 4;
+impl VirtioDevice for VirtioNet {
-pub const TUN_F_TSO_ECN: u32 = 8;
+    fn features(&self) -> &FeatureBits {
+        &self.features
+    }
+
+    fn queue_sizes(&self) -> &[u16] {
+        &[256, 256]
+    }
+
+    fn device_type(&self) -> VirtioDeviceType {
+        VirtioDeviceType::Net
+    }
+
+    fn config_size(&self) -> usize {
+        MAC_ADDR_LEN
+    }
+
+    fn read_config(&self, offset: u64, data: &mut [u8]) {
+        let (_,_) = (offset, data);
+    }
+
+    fn write_config(&mut self, offset: u64, data: &[u8]) {
+        let (_,_) = (offset, data);
+    }
+
+    fn start(&mut self, queues: &Queues) {
+        let rx = queues.get_queue(0);
+        let tx = queues.get_queue(1);
 
-impl VirtioDeviceOps for VirtioNet {
-    fn start(&mut self, _memory: &MemoryManager, mut queues: Vec<VirtQueue>) {
-        let tx = queues.pop().unwrap();
-        let rx = queues.pop().unwrap();
         let tap = self.tap.take().unwrap();
         let poll = match EPoll::new() {
             Ok(poll) => poll,
@@ -105,6 +114,11 @@ impl VirtioDeviceOps for VirtioNet {
         });
     }
 }
+pub const TUN_F_CSUM: u32 = 1;
+pub const TUN_F_TSO4: u32 = 2;
+pub const TUN_F_TSO6: u32 = 4;
+pub const TUN_F_TSO_ECN: u32 = 8;
+
 
 const MAX_BUFFER_SIZE: usize = 65562;
 const RX_VQ_TOKEN:u64 = 1;

src/devices/virtio_rng.rs

@@ -1,34 +1,18 @@
 
-use std::sync::{Arc,RwLock};
 use std::thread;
 use std::fs::File;
+use crate::io::{FeatureBits, Queues, VirtioDevice, VirtioDeviceType, VirtQueue};
 
-use crate::virtio::{VirtioDeviceOps,VirtioBus,VirtQueue,Result};
+pub struct VirtioRandom {
-use crate::memory::MemoryManager;
+    features: FeatureBits,
-
+}
-
-const VIRTIO_ID_RANDOM: u16 = 4;
-
-pub struct VirtioRandom;
 
 impl VirtioRandom {
-    fn new() -> VirtioRandom { VirtioRandom }
+    pub fn new() -> VirtioRandom {
-
+        VirtioRandom {
-    pub fn create(vbus: &mut VirtioBus) -> Result<()> {
+            features: FeatureBits::new_default(0),
-        let dev = Arc::new(RwLock::new(VirtioRandom::new()));
-        vbus.new_virtio_device(VIRTIO_ID_RANDOM, dev)
-            .set_num_queues(1)
-            .register()
         }
     }
-
-impl VirtioDeviceOps for VirtioRandom {
-
-    fn start(&mut self, _memory: &MemoryManager, mut queues: Vec<VirtQueue>) {
-        thread::spawn(move|| {
-            run(queues.pop().unwrap())
-        });
-    }
 }
 
 fn run(q: VirtQueue) {
@@ -42,3 +26,24 @@ fn run(q: VirtQueue) {
         });
     }
 }
+
+impl VirtioDevice for VirtioRandom {
+    fn features(&self) -> &FeatureBits {
+        &self.features
+    }
+
+    fn queue_sizes(&self) -> &[u16] {
+        &[VirtQueue::DEFAULT_QUEUE_SIZE]
+    }
+
+    fn device_type(&self) -> VirtioDeviceType {
+        VirtioDeviceType::Rng
+    }
+
+    fn start(&mut self, queues: &Queues) {
+        let vq = queues.get_queue(0);
+        thread::spawn(move|| {
+            run(vq)
+        });
+    }
+}

src/devices/virtio_serial.rs

@@ -1,12 +1,8 @@
-use std::sync::{Arc,RwLock};
 use std::io::{self,Write,Read};
 use std::thread::spawn;
 use termios::*;
 
-use crate::virtio::{VirtioDeviceOps,VirtioBus, VirtQueue,Result};
+use crate::io::{VirtioDevice, VirtioDeviceType, FeatureBits, VirtQueue, ReadableInt, Queues};
-use crate::memory::MemoryManager;
-
-const VIRTIO_ID_CONSOLE: u16 = 3;
 
 const VIRTIO_CONSOLE_F_SIZE: u64 = 0x1;
 const VIRTIO_CONSOLE_F_MULTIPORT: u64 = 0x2;
@@ -21,25 +17,18 @@ const VIRTIO_CONSOLE_PORT_OPEN: u16     = 6;
 const _VIRTIO_CONSOLE_PORT_NAME: u16     = 7;
 
 pub struct VirtioSerial {
-    feature_bits: u64,
+    features: FeatureBits,
 }
 
 impl VirtioSerial {
-    fn new() -> VirtioSerial {
+    pub fn new() -> VirtioSerial {
-        VirtioSerial{feature_bits:0}
+        let features = FeatureBits::new_default(VIRTIO_CONSOLE_F_MULTIPORT|VIRTIO_CONSOLE_F_SIZE);
+        VirtioSerial{
+            features,
+        }
     }
 
-    pub fn create(vbus: &mut VirtioBus) -> Result<()> {
+    fn start_console(&self, q: VirtQueue) {
-        let dev = Arc::new(RwLock::new(VirtioSerial::new()));
-        vbus.new_virtio_device(VIRTIO_ID_CONSOLE, dev)
-            .set_num_queues(4)
-            .set_device_class(0x0700)
-            .set_config_size(12)
-            .set_features(VIRTIO_CONSOLE_F_MULTIPORT|VIRTIO_CONSOLE_F_SIZE)
-            .register()
-    }
-
-    fn start_console(&self, _memory: &MemoryManager, q: VirtQueue) {
         spawn(move || {
             loop {
                 q.wait_ready().unwrap();
@@ -52,7 +41,7 @@ impl VirtioSerial {
     }
 
     fn multiport(&self) -> bool {
-        self.feature_bits & VIRTIO_CONSOLE_F_MULTIPORT != 0
+        self.features.has_guest_bit(VIRTIO_CONSOLE_F_MULTIPORT)
     }
 }
 
@@ -69,39 +58,50 @@ struct WinSz {
 
 const TIOCGWINSZ: u64 = 0x5413;
 
-impl VirtioDeviceOps for VirtioSerial {
+impl VirtioDevice for VirtioSerial {
-    fn reset(&mut self) {
+    fn features(&self) -> &FeatureBits {
-        println!("Reset called");
+        &self.features
     }
 
-    fn enable_features(&mut self, bits: u64) -> bool {
+
-        self.feature_bits = bits;
+    fn queue_sizes(&self) -> &[u16] {
-        true
+        &[
+            VirtQueue::DEFAULT_QUEUE_SIZE,
+            VirtQueue::DEFAULT_QUEUE_SIZE,
+            VirtQueue::DEFAULT_QUEUE_SIZE,
+            VirtQueue::DEFAULT_QUEUE_SIZE,
+        ]
     }
 
-    fn read_config(&mut self, offset: usize, _size: usize) -> u64 {
+    fn device_type(&self) -> VirtioDeviceType {
-        if offset == 4 {
+        VirtioDeviceType::Console
-            return 1;
-        }
-        0
     }
 
-    fn start(&mut self, memory: &MemoryManager, mut queues: Vec<VirtQueue>) {
+    fn config_size(&self) -> usize {
-        let mut term = Terminal::create(queues.remove(0));
+        12
-        self.start_console(memory, queues.remove(0));
+    }
 
+    fn read_config(&self, offset: u64, data: &mut [u8]) {
+        if offset == 4 && data.len() == 4 {
+            ReadableInt::new_dword(1).read(data);
+        } else {
+            data.fill(0);
+        }
+    }
+
+    fn start(&mut self, queues: &Queues) {
+        let mut term = Terminal::create(queues.get_queue(0));
+        self.start_console(queues.get_queue(1));
         spawn( move || {
             term.read_loop();
         });
-
         if self.multiport() {
-            let mut control = Control::new(queues.remove(0), queues.remove(0));
+            let mut control = Control::new(queues.get_queue(2), queues.get_queue(3));
             spawn(move || {
                 control.run();
             });
         }
     }
-
 }
 
 struct Control {

src/devices/virtio_wl/device.rs

@@ -1,16 +1,15 @@
 use std::os::unix::io::{AsRawFd, RawFd};
-use std::sync::{RwLock, Arc};
 use std::thread;
 
-use crate::{system, virtio};
+use crate::system;
 use crate::system::EPoll;
 use crate::memory::{MemoryManager, DrmDescriptor};
-use crate::virtio::{VirtQueue, VirtioBus, VirtioDeviceOps, Chain};
 
 use crate::devices::virtio_wl::{vfd::VfdManager, consts::*, Error, Result, VfdObject};
 use crate::system::ioctl::ioctl_with_ref;
 use std::os::raw::{c_ulong, c_uint, c_ulonglong};
 use vmm_sys_util::eventfd::EventFd;
+use crate::io::{Chain, FeatureBits, Queues, VirtioDevice, VirtioDeviceType, VirtQueue};
 
 #[repr(C)]
 struct dma_buf_sync {
@@ -20,25 +19,21 @@ const DMA_BUF_IOCTL_BASE: c_uint = 0x62;
 const DMA_BUF_IOCTL_SYNC: c_ulong = iow!(DMA_BUF_IOCTL_BASE, 0, ::std::mem::size_of::<dma_buf_sync>() as i32);
 
 pub struct VirtioWayland {
-    feature_bits: u64,
+    features: FeatureBits,
     enable_dmabuf: bool,
 }
 
 impl VirtioWayland {
-    fn new(enable_dmabuf: bool) -> Self {
+    pub fn new(enable_dmabuf: bool) -> Self {
-        VirtioWayland { feature_bits: 0, enable_dmabuf }
+        let features = FeatureBits::new_default(VIRTIO_WL_F_TRANS_FLAGS as u64);
+        VirtioWayland {
+            features,
+            enable_dmabuf
         }
-
-    pub fn create(vbus: &mut VirtioBus, dmabuf: bool) -> virtio::Result<()> {
-        let dev = Arc::new(RwLock::new(VirtioWayland::new(dmabuf)));
-        vbus.new_virtio_device(VIRTIO_ID_WL, dev)
-            .set_num_queues(2)
-            .set_features(VIRTIO_WL_F_TRANS_FLAGS as u64)
-            .register()
     }
 
     fn transition_flags(&self) -> bool {
-        self.feature_bits & VIRTIO_WL_F_TRANS_FLAGS as u64 != 0
+        self.features.has_guest_bit(VIRTIO_WL_F_TRANS_FLAGS as u64)
     }
 
     fn create_device(memory: MemoryManager, in_vq: VirtQueue, out_vq: VirtQueue, transition: bool, enable_dmabuf: bool) -> Result<WaylandDevice> {
@@ -48,21 +43,28 @@ impl VirtioWayland {
     }
 }
 
-impl VirtioDeviceOps for VirtioWayland {
+impl VirtioDevice for VirtioWayland {
-    fn enable_features(&mut self, bits: u64) -> bool {
+    fn features(&self) -> &FeatureBits {
-        self.feature_bits = bits;
+        &self.features
-        true
     }
 
-    fn start(&mut self, memory: &MemoryManager, mut queues: Vec<VirtQueue>) {
+    fn queue_sizes(&self) -> &[u16] {
+        &[VirtQueue::DEFAULT_QUEUE_SIZE, VirtQueue::DEFAULT_QUEUE_SIZE]
+    }
+
+    fn device_type(&self) -> VirtioDeviceType {
+        VirtioDeviceType::Wl
+    }
+
+    fn start(&mut self, queues: &Queues) {
         thread::spawn({
-            let memory = memory.clone();
             let transition = self.transition_flags();
             let enable_dmabuf = self.enable_dmabuf;
+            let in_vq = queues.get_queue(0);
+            let out_vq = queues.get_queue(1);
+            let memory = queues.memory().clone();
             move || {
-                let out_vq = queues.pop().unwrap();
+                let mut dev = match Self::create_device(memory, in_vq, out_vq,transition, enable_dmabuf) {
-                let in_vq = queues.pop().unwrap();
-                let mut dev = match Self::create_device(memory.clone(), in_vq, out_vq,transition, enable_dmabuf) {
                     Err(e) => {
                         warn!("Error creating virtio wayland device: {}", e);
                         return;

src/devices/virtio_wl/mod.rs

@@ -16,7 +16,6 @@ mod device;
 mod consts {
     use std::mem;
 
-    pub const VIRTIO_ID_WL: u16 = 63;
     pub const VIRTWL_SEND_MAX_ALLOCS: usize = 28;
     pub const VIRTIO_WL_CMD_VFD_NEW: u32 = 256;
     pub const VIRTIO_WL_CMD_VFD_CLOSE: u32 = 257;

src/devices/virtio_wl/vfd.rs

@@ -6,11 +6,11 @@ use std::time::Duration;
 
 use crate::memory::{MemoryManager, DrmDescriptor};
 use crate::system::{FileDesc, FileFlags,EPoll,MemoryFd};
-use crate::virtio::{VirtQueue, Chain};
 
 use crate::devices::virtio_wl::{
     consts::*, Error, Result, shm::VfdSharedMemory, pipe::VfdPipe, socket::VfdSocket, VfdObject
 };
+use crate::io::{Chain, VirtQueue};
 
 pub struct VfdManager {
     wayland_path: PathBuf,

src/io/bus.rs

@@ -0,0 +1,144 @@
+use std::cmp::Ordering;
+use std::collections::BTreeMap;
+use std::result;
+use std::sync::{Arc, Mutex};
+
+use thiserror::Error;
+
+
+#[derive(Debug,Error)]
+pub enum Error {
+    #[error("New device overlaps with an old device.")]
+    Overlap,
+}
+
+pub type Result<T> = result::Result<T, Error>;
+
+pub trait BusDevice {
+    fn read(&mut self, offset: u64, data: &mut [u8]) {
+        let (_,_) = (offset, data);
+
+    }
+    fn write(&mut self, offset: u64, data: &[u8]) {
+        let (_,_) = (offset, data);
+    }
+}
+
+#[derive(Debug,Copy,Clone)]
+struct BusRange(u64, u64);
+
+impl Eq for BusRange {}
+
+impl PartialEq for BusRange {
+    fn eq(&self, other: &Self) -> bool {
+        self.0 == other.0
+    }
+}
+
+impl Ord for BusRange {
+    fn cmp(&self, other: &Self) -> Ordering {
+        self.0.cmp(&other.0)
+    }
+}
+
+impl PartialOrd for BusRange {
+    fn partial_cmp(&self, other: &Self) -> Option<Ordering> {
+        self.0.partial_cmp(&other.0)
+    }
+}
+
+/// A device container for routing reads and writes over some address space.
+///
+/// This doesn't have any restrictions on what kind of device or address space this applies to. The
+/// only restriction is that no two devices can overlap in this address space.
+#[derive(Clone,Default)]
+pub struct Bus {
+    devices: BTreeMap<BusRange, Arc<Mutex<dyn BusDevice + Send>>>,
+}
+
+impl Bus {
+    /// Constructs an a bus with an empty address space.
+    pub fn new() -> Bus {
+        Bus {
+            devices: BTreeMap::new(),
+        }
+    }
+    fn first_before(&self, addr: u64) -> Option<(BusRange, &Arc<Mutex<dyn BusDevice+Send>>)> {
+        for (range, dev) in self.devices.iter().rev() {
+            if range.0 <= addr {
+                return Some((*range, dev))
+            }
+        }
+        None
+    }
+
+    /// Puts the given device at the given address space.
+    pub fn get_device(&self, addr: u64) -> Option<(u64, &Arc<Mutex<dyn BusDevice+Send>>)> {
+        if let Some((BusRange(start, len), dev)) = self.first_before(addr) {
+            let offset = addr - start;
+            if offset < len {
+                return Some((offset, dev))
+            }
+        }
+        None
+    }
+    /// Puts the given device at the given address space.
+    pub fn insert(&mut self, device: Arc<Mutex<dyn BusDevice+Send>>, base: u64, len: u64) -> Result<()> {
+        if len == 0 {
+            return Err(Error::Overlap);
+        }
+
+        // Reject all cases where the new device's base is within an old device's range.
+        if self.get_device(base).is_some() {
+            return Err(Error::Overlap);
+        }
+
+        // The above check will miss an overlap in which the new device's base address is before the
+        // range of another device. To catch that case, we search for a device with a range before
+        // the new device's range's end. If there is no existing device in that range that starts
+        // after the new device, then there will be no overlap.
+        if let Some((BusRange(start, _), _)) = self.first_before(base + len - 1) {
+            // Such a device only conflicts with the new device if it also starts after the new
+            // device because of our initial `get_device` check above.
+            if start >= base {
+                return Err(Error::Overlap);
+            }
+        }
+
+        if self.devices.insert(BusRange(base, len), device).is_some() {
+            return Err(Error::Overlap);
+        }
+
+        Ok(())
+    }
+
+    /// Reads data from the device that owns the range containing `addr` and puts it into `data`.
+    ///
+    /// Returns true on success, otherwise `data` is untouched.
+    pub fn read(&self, addr: u64, data: &mut [u8]) -> bool {
+        if let Some((offset, dev)) = self.get_device(addr) {
+            // OK to unwrap as lock() failing is a serious error condition and should panic.
+            dev.lock()
+                .expect("Failed to acquire device lock")
+                .read(offset, data);
+            true
+        } else {
+            false
+        }
+    }
+
+    /// Writes `data` to the device that owns the range containing `addr`.
+    ///
+    /// Returns true on success, otherwise `data` is untouched.
+    pub fn write(&self, addr: u64, data: &[u8]) -> bool {
+        if let Some((offset, dev)) = self.get_device(addr) {
+            // OK to unwrap as lock() failing is a serious error condition and should panic.
+            dev.lock()
+                .expect("Failed to acquire device lock")
+                .write(offset, data);
+            true
+        } else {
+            false
+        }
+    }
+}

src/io/busdata.rs

@@ -0,0 +1,131 @@
+use std::convert::TryInto;
+use std::fmt::{Debug, Formatter};
+
+/*
+pub enum IoInt {
+    Byte(u8, [u8; 1]),
+    Word(u16, [u8; 2]),
+    DWord(u32, [u8; 4]),
+    QWord(u64, [u8; 8]),
+    Data(Vec<u8>),
+}
+
+impl IoInt {
+    pub fn new_byte(n: u8) -> Self {
+        Self::Byte(n, [n])
+    }
+    pub fn new_word(n: u16) -> Self {
+        Self::Word(n, n.to_le_bytes())
+    }
+    pub fn new_dword(n: u32) -> Self {
+        Self::DWord(n, n.to_le_bytes())
+    }
+    pub fn new_qword(n: u64) -> Self {
+        Self::QWord(n, n.to_le_bytes())
+    }
+
+}
+
+impl From<&[u8]> for IoInt {
+    fn from(bytes: &[u8]) -> Self {
+        match bytes.len() {
+            1 => Self::Byte(bytes[0], [bytes[0]]),
+            2 => {
+                let n = u16::from_le_bytes(bytes.try_into().unwrap());
+                Self::Word(n, n.to_le_bytes())
+            },
+            4 => {
+                let n = u32::from_le_bytes(bytes.try_into().unwrap());
+                Self::DWord(n, n.to_le_bytes())
+            },
+            8 => {
+                let n = u64::from_le_bytes(bytes.try_into().unwrap());
+                Self::QWord(n, n.to_le_bytes())
+            },
+            _ => Self::Data(bytes.to_vec()),
+        }
+    }
+}
+
+ */
+
+pub enum WriteableInt {
+    Byte(u8),
+    Word(u16),
+    DWord(u32),
+    QWord(u64),
+    Data(Vec<u8>),
+}
+
+impl From<&[u8]> for WriteableInt {
+    fn from(bytes: &[u8]) -> Self {
+        match bytes.len() {
+            1 => Self::Byte(bytes[0]),
+            2 => Self::Word(u16::from_le_bytes(bytes.try_into().unwrap())),
+            4 => Self::DWord(u32::from_le_bytes(bytes.try_into().unwrap())),
+            8 => Self::QWord(u64::from_le_bytes(bytes.try_into().unwrap())),
+            _ => Self::Data(bytes.to_vec()),
+        }
+    }
+}
+
+pub enum ReadableInt {
+    Byte(u8, [u8; 1]),
+    Word(u16, [u8; 2]),
+    DWord(u32, [u8; 4]),
+}
+
+impl ReadableInt {
+
+    pub fn new_byte(n: u8) -> Self {
+        Self::Byte(n, [n])
+    }
+    pub fn new_word(n: u16) -> Self {
+        Self::Word(n, n.to_le_bytes())
+    }
+    pub fn new_dword(n: u32) -> Self {
+        Self::DWord(n, n.to_le_bytes())
+    }
+
+    fn as_bytes(&self) -> &[u8] {
+        match self {
+            ReadableInt::Byte(_, bs) => bs,
+            ReadableInt::Word(_, bs) => bs,
+            ReadableInt::DWord(_, bs) => bs,
+        }
+    }
+
+    pub fn read(&self, buffer: &mut [u8]) {
+        let bs = self.as_bytes();
+        if buffer.len() >= bs.len() {
+            buffer[..bs.len()].copy_from_slice(bs);
+        }
+    }
+}
+
+impl From<u8> for ReadableInt {
+    fn from(value: u8) -> Self {
+        ReadableInt::new_byte(value)
+    }
+}
+
+impl From<u16> for ReadableInt {
+    fn from(value: u16) -> Self {
+        ReadableInt::new_word(value)
+    }
+}
+impl From<u32> for ReadableInt {
+    fn from(value: u32) -> Self {
+        ReadableInt::new_dword(value)
+    }
+}
+
+impl Debug for ReadableInt {
+    fn fmt(&self, f: &mut Formatter<'_>) -> std::fmt::Result {
+        match self {
+            ReadableInt::Byte(n, _) => write!(f, "Byte({})", n),
+            ReadableInt::Word(n, _) => write!(f, "Word({})", n),
+            ReadableInt::DWord(n, _) => write!(f, "DWord({})", n),
+        }
+    }
+}

src/io/manager.rs

@@ -0,0 +1,192 @@
+use std::sync::{Arc, Mutex, MutexGuard};
+use vm_allocator::{AddressAllocator, AllocPolicy, IdAllocator, RangeInclusive};
+use vmm_sys_util::eventfd::EventFd;
+use crate::devices::rtc::Rtc;
+use crate::devices::serial::{SerialDevice, SerialPort};
+use crate::io::bus::{Bus, BusDevice};
+use crate::io::pci::{MmioHandler, PciBarAllocation, PciBus, PciDevice};
+use crate::io::{PciIrq, virtio};
+use crate::io::virtio::{VirtioDeviceState,VirtioDevice};
+use crate::memory::{AddressRange, MemoryManager};
+use crate::vm::arch;
+
+#[derive(Clone)]
+pub struct IoAllocator {
+    mmio_allocator: Arc<Mutex<AddressAllocator>>,
+    irq_allocator: Arc<Mutex<IdAllocator>>,
+}
+
+impl IoAllocator {
+    fn new() -> Self {
+        let mmio_allocator = AddressAllocator::new(arch::PCI_MMIO_RESERVED_BASE, arch::PCI_MMIO_RESERVED_SIZE as u64)
+            .expect("Failed to create address allocator");
+        let irq_allocator = IdAllocator::new(arch::IRQ_BASE, arch::IRQ_MAX)
+            .expect("Failed to create IRQ allocator");
+        IoAllocator {
+            mmio_allocator: Arc::new(Mutex::new(mmio_allocator)),
+            irq_allocator: Arc::new(Mutex::new(irq_allocator)),
+        }
+    }
+
+    pub fn allocate_mmio(&self, size: usize) -> RangeInclusive {
+        let mut allocator = self.mmio_allocator.lock().unwrap();
+        allocator.allocate(size as u64, 4096, AllocPolicy::FirstMatch).unwrap()
+    }
+
+    pub fn allocate_irq(&self) -> u8 {
+        let mut allocator = self.irq_allocator.lock().unwrap();
+        allocator.allocate_id().unwrap() as u8
+    }
+}
+
+#[derive(Clone)]
+pub struct IoManager {
+    memory: MemoryManager,
+    pio_bus: Bus,
+    mmio_bus: Bus,
+    pci_bus: Arc<Mutex<PciBus>>,
+    allocator: IoAllocator,
+}
+
+impl IoManager {
+    pub fn new(memory: MemoryManager) -> IoManager {
+        let pci_bus = Arc::new(Mutex::new(PciBus::new()));
+        let mut pio_bus = Bus::new();
+        pio_bus.insert(pci_bus.clone(), PciBus::PCI_CONFIG_ADDRESS as u64, 8)
+            .expect("Failed to add PCI configuration to PIO");
+        IoManager {
+            memory,
+            pio_bus,
+            mmio_bus: Bus::new(),
+            pci_bus,
+            allocator: IoAllocator::new(),
+        }
+    }
+
+    pub fn register_legacy_devices(&mut self, reset_evt: EventFd) {
+        let rtc = Arc::new(Mutex::new(Rtc::new()));
+        self.pio_bus.insert(rtc, 0x0070, 2).unwrap();
+
+        let i8042 = Arc::new(Mutex::new(I8042Device::new(reset_evt)));
+        self.pio_bus.insert(i8042, 0x0060, 8).unwrap();
+    }
+
+    pub fn register_serial_port(&mut self, port: SerialPort) {
+        let serial = SerialDevice::new(self.memory.kvm_vm().clone(), port.irq());
+        let serial = Arc::new(Mutex::new(serial));
+        self.pio_bus.insert(serial, port.io_port() as u64, 8).unwrap();
+
+    }
+
+    pub fn allocator(&self) -> IoAllocator {
+        self.allocator.clone()
+    }
+
+    pub fn mmio_read(&self, addr: u64, data: &mut [u8]) -> bool {
+        self.mmio_bus.read(addr, data)
+    }
+
+    pub fn mmio_write(&self, addr: u64, data: &[u8]) -> bool {
+        self.mmio_bus.write(addr, data)
+    }
+
+    pub fn pio_read(&self, port: u16, data: &mut [u8]) -> bool {
+        self.pio_bus.read(port as u64, data)
+    }
+
+    pub fn pio_write(&self, port: u16, data: &[u8]) -> bool {
+        self.pio_bus.write(port as u64, data)
+    }
+
+    fn pci_bus(&self) -> MutexGuard<PciBus> {
+        self.pci_bus.lock().unwrap()
+    }
+
+    pub fn pci_irqs(&self) -> Vec<PciIrq> {
+        self.pci_bus().pci_irqs()
+    }
+
+    fn allocate_pci_bars(&mut self, dev: &Arc<Mutex<dyn PciDevice+Send>>) {
+        let allocations = dev.lock().unwrap().bar_allocations();
+        if allocations.is_empty() {
+            return;
+        }
+
+        for a in allocations {
+            let mut allocated = Vec::new();
+            match a {
+                PciBarAllocation::Mmio(bar, size) => {
+                    let range = self.allocator.allocate_mmio(size);
+                    let mmio = AddressRange::new(range.start(), range.len() as usize);
+                    dev.lock().unwrap().config_mut().set_mmio_bar(bar, mmio);
+                    allocated.push((bar,range.start()));
+                    let handler = Arc::new(Mutex::new(MmioHandler::new(bar, dev.clone())));
+                    self.mmio_bus.insert(handler, range.start(), range.len()).unwrap();
+                }
+            }
+            dev.lock().unwrap().configure_bars(allocated);
+        }
+    }
+
+    pub fn add_pci_device(&mut self, device: Arc<Mutex<dyn PciDevice+Send>>) {
+        self.allocate_pci_bars(&device);
+        let mut pci = self.pci_bus.lock().unwrap();
+        pci.add_device(device);
+    }
+
+    pub fn add_virtio_device<D: VirtioDevice+'static>(&mut self, dev: D) -> virtio::Result<()> {
+        //let devtype = dev.device_type();
+        //let dev = Arc::new(Mutex::new(dev));
+        //let devstate = VirtioDeviceState::new(dev.clone(), self.memory.clone(), self.allocator.clone())?;
+        let irq = self.allocator.allocate_irq();
+        //let devstate = VirtioDeviceState::new(dev, self.memory.clone(), self.allocator.clone())?;
+        let devstate = VirtioDeviceState::new(dev, self.memory.clone(), irq)?;
+        self.add_pci_device(Arc::new(Mutex::new(devstate)));
+
+       // let mmio_range = devstate.mmio_range();
+
+        //let mut pci = self.pci_bus.lock().unwrap();
+        //pci.add_device(devstate);
+//        let mut pci_device = pci.new_device(devstate.irq() as u8, PCI_VENDOR_ID_REDHAT, devtype.device_id(), devtype.class_id());
+        // XXX add mmio bar
+        //pci_device.set_mmio_bar(0, AddressRange::new(mmio_range.start(), mmio_range.len() as usize));
+
+//        devstate.add_pci_capabilities(&mut pci_device);
+        // XXX add devstate to mmio bus
+        //self.mmio_bus.insert(Arc::new(Mutex::new(devstate)), mmio_range.start(), mmio_range.len())?;
+
+        //pci.add_device(pci_device);
+        Ok(())
+    }
+}
+
+pub struct I8042Device {
+    reset_evt: EventFd,
+}
+impl I8042Device {
+    fn new(reset_evt: EventFd) -> Self {
+        I8042Device { reset_evt }
+    }
+}
+
+impl BusDevice for I8042Device {
+    fn read(&mut self, offset: u64, data: &mut [u8]) {
+        if data.len() == 1 {
+            match offset {
+                0 => data[0] = 0x20,
+                1 => data[0] = 0x00,
+                _ => {},
+            }
+        }
+    }
+
+    fn write(&mut self, offset: u64, data: &[u8]) {
+        if data.len() == 1 {
+            if offset == 3 && data[0] == 0xfe {
+                if let Err(err) = self.reset_evt.write(1) {
+                    warn!("Error triggering i8042 reset event: {}", err);
+                }
+            }
+        }
+    }
+}

src/io/mod.rs

@@ -0,0 +1,12 @@
+pub mod bus;
+pub mod busdata;
+pub mod pci;
+pub mod manager;
+pub mod virtio;
+pub use virtio::{VirtioDevice,FeatureBits,VirtioDeviceType,VirtQueue,Chain,Queues};
+pub use virtio::Error as VirtioError;
+pub use busdata::{ReadableInt,WriteableInt};
+pub use pci::PciIrq;
+// PCI Vendor id for Virtio devices
+
+pub const PCI_VENDOR_ID_REDHAT: u16 = 0x1af4;

src/io/pci/address.rs

@@ -0,0 +1,32 @@
+
+#[derive(Copy,Clone,Debug,PartialEq,Eq,PartialOrd,Ord,Hash)]
+pub struct PciAddress(u16);
+
+impl PciAddress {
+    pub fn empty() -> Self {
+        Self::new(0,0,0)
+    }
+
+    pub fn new(bus: u8, device: u8, function: u8) -> Self {
+        const DEVICE_MASK: u16 = 0x1f;
+        const FUNCTION_MASK: u16 = 0x07;
+
+        let bus = bus as u16;
+        let device = device as u16;
+        let function = function as u16;
+
+        let addr = bus << 8
+            | (device & DEVICE_MASK) << 3
+            | (function & FUNCTION_MASK);
+
+        PciAddress(addr)
+    }
+
+    pub fn device(&self) -> u8 {
+        ((self.0 & 0xF) >> 3) as u8
+    }
+
+    pub fn address(&self) -> u16 {
+        self.0
+    }
+}

src/io/pci/bus.rs

@@ -0,0 +1,198 @@
+use std::collections::BTreeMap;
+use std::sync::{Arc, Mutex};
+use crate::io::bus::BusDevice;
+use crate::io::pci::address::PciAddress;
+use crate::io::pci::config::PciConfiguration;
+use crate::io::pci::consts::{PCI_CLASS_BRIDGE_HOST, PCI_MAX_DEVICES, PCI_VENDOR_ID_INTEL};
+use crate::io::pci::PciDevice;
+
+/// Current address to read/write from (io port 0xcf8)
+struct PciConfigAddress([u8; 4]);
+
+impl PciConfigAddress {
+    fn new() -> Self {
+        PciConfigAddress([0u8; 4])
+    }
+    fn bus(&self) -> u8 {
+        self.0[2]
+    }
+
+    fn function(&self) -> u8 {
+        self.0[1] & 0x7
+    }
+    fn device(&self) -> u8 {
+        self.0[1] >> 3
+    }
+    fn offset(&self) -> u8 {
+        self.0[0] & !0x3
+    }
+
+    fn enabled(&self) -> bool {
+        self.0[3] & 0x80 != 0
+    }
+
+    fn pci_address(&self) -> PciAddress {
+        PciAddress::new(self.bus(), self.device(), self.function())
+    }
+
+    fn write(&mut self, offset: u64, data: &[u8]) {
+        let offset = offset as usize;
+        if offset + data.len() <= 4 {
+            self.0[offset..offset+data.len()]
+                .copy_from_slice(data)
+        }
+    }
+
+    fn read(&self, offset: u64, data: &mut [u8]) {
+        let offset = offset as usize;
+        if offset + data.len() <= 4 {
+            data.copy_from_slice(&self.0[offset..offset+data.len()])
+        }
+    }
+}
+
+struct PciRootDevice(PciConfiguration);
+
+impl PciRootDevice {
+    fn new() -> Self {
+        let config = PciConfiguration::new(0, PCI_VENDOR_ID_INTEL, 0, PCI_CLASS_BRIDGE_HOST);
+        PciRootDevice(config)
+    }
+}
+impl PciDevice for PciRootDevice {
+
+    fn config(&self) -> &PciConfiguration {
+        &self.0
+    }
+
+    fn config_mut(&mut self) -> &mut PciConfiguration {
+        &mut self.0
+    }
+}
+
+pub struct PciBus {
+    devices: BTreeMap<PciAddress, Arc<Mutex<dyn PciDevice>>>,
+
+    config_address: PciConfigAddress,
+    used_device_ids: Vec<bool>,
+
+}
+
+impl PciBus {
+    pub const PCI_CONFIG_ADDRESS: u16 = 0xcf8;
+
+    pub fn new() -> PciBus {
+        let mut pci = PciBus {
+            devices: BTreeMap::new(),
+            config_address: PciConfigAddress::new(),
+            used_device_ids: vec![false; PCI_MAX_DEVICES],
+        };
+
+        let root = PciRootDevice::new();
+        pci.add_device(Arc::new(Mutex::new(root)));
+        pci
+
+    }
+
+    pub fn add_device(&mut self, device: Arc<Mutex<dyn PciDevice>>) {
+        let id = self.allocate_id().unwrap();
+        let address = PciAddress::new(0, id, 0);
+        device.lock().unwrap().config_mut().set_address(address);
+        self.devices.insert(address, device);
+    }
+
+    pub fn pci_irqs(&self) -> Vec<PciIrq> {
+        let mut irqs = Vec::new();
+        for (addr, dev)  in &self.devices {
+            let lock = dev.lock().unwrap();
+            if let Some(irq) = lock.irq() {
+                irqs.push(PciIrq::new(addr.device(), irq));
+            }
+        }
+        irqs
+    }
+
+    fn allocate_id(&mut self) -> Option<u8> {
+        for i in 0..PCI_MAX_DEVICES {
+            if !self.used_device_ids[i] {
+                self.used_device_ids[i] = true;
+                return Some(i as u8)
+            }
+        }
+        None
+    }
+
+    fn is_in_range(base: u64, offset: u64, len: usize) -> bool {
+        let end = offset + len as u64;
+        offset >= base && end <= (base + 4)
+    }
+
+    fn is_config_address(offset: u64, len: usize) -> bool {
+        Self::is_in_range(0, offset, len)
+    }
+
+    fn is_config_data(offset: u64, len: usize) -> bool {
+        Self::is_in_range(4, offset, len)
+    }
+
+    fn current_config_device(&self) -> Option<Arc<Mutex<dyn PciDevice>>> {
+        if self.config_address.enabled() {
+            let addr = self.config_address.pci_address();
+            self.devices.get(&addr).cloned()
+        } else {
+            None
+        }
+    }
+}
+
+impl BusDevice for PciBus {
+    fn read(&mut self, offset: u64, data: &mut [u8]) {
+        if PciBus::is_config_address(offset, data.len()) {
+            self.config_address.read(offset, data);
+        } else if PciBus::is_config_data(offset, data.len()) {
+            if let Some(dev) = self.current_config_device() {
+                let lock = dev.lock().unwrap();
+                let offset = (offset - 4) + self.config_address.offset() as u64;
+                lock.config().read(offset, data)
+            } else {
+                data.fill(0xff)
+            }
+        }
+    }
+    fn write(&mut self, offset: u64, data: &[u8]) {
+        if PciBus::is_config_address(offset, data.len()) {
+            self.config_address.write(offset, data)
+        } else if PciBus::is_config_data(offset, data.len()) {
+            if let Some(dev) = self.current_config_device() {
+                let mut lock = dev.lock().unwrap();
+                let offset = (offset - 4) + self.config_address.offset() as u64;
+                lock.config_mut().write(offset, data)
+            }
+        }
+    }
+}
+
+#[derive(Debug)]
+pub struct PciIrq {
+    pci_id: u8,
+    int_pin: u8,
+    irq: u8,
+}
+
+impl PciIrq {
+    fn new(pci_id: u8, irq: u8) -> PciIrq {
+        PciIrq {
+            pci_id,
+            int_pin: 1,
+            irq,
+        }
+    }
+
+    pub fn src_bus_irq(&self) -> u8 {
+        (self.pci_id << 2) | (self.int_pin - 1)
+    }
+
+    pub fn irq_line(&self) -> u8 {
+        self.irq
+    }
+}

src/io/pci/config.rs

@@ -0,0 +1,248 @@
+use crate::io::pci::address::PciAddress;
+use crate::io::pci::consts::{PCI_BAR0, PCI_BAR5, PCI_CACHE_LINE_SIZE, PCI_CAP_BASE_OFFSET, PCI_CAP_ID_VENDOR, PCI_CAPABILITY_LIST, PCI_CLASS_DEVICE, PCI_CLASS_REVISION, PCI_COMMAND, PCI_COMMAND_IO, PCI_COMMAND_MEMORY, PCI_DEVICE_ID, PCI_INTERRUPT_LINE, PCI_INTERRUPT_PIN, PCI_STATUS, PCI_STATUS_CAP_LIST, PCI_SUBSYSTEM_ID, PCI_VENDOR_ID};
+use crate::io::pci::device::PciBar;
+use crate::memory::AddressRange;
+use crate::util::{ByteBuffer,Writeable};
+
+const PCI_CONFIG_SPACE_SIZE: usize = 256;
+const MAX_CAPABILITY_COUNT:usize  = 16; // arbitrary
+
+pub struct PciCapability<'a> {
+    config: &'a mut PciConfiguration,
+    buffer: ByteBuffer<Vec<u8>>,
+}
+
+impl <'a> PciCapability<'a> {
+    pub fn new_vendor_capability(config: &'a mut PciConfiguration) -> Self {
+        let mut buffer = ByteBuffer::new_empty();
+        buffer.write(PCI_CAP_ID_VENDOR);
+        buffer.write(0u8);
+        PciCapability { config, buffer }
+    }
+
+    pub fn write<V: Writeable>(&mut self, val: V) {
+        self.buffer.write(val);
+    }
+
+    pub fn store(&mut self) {
+        let offset = self.config.next_capability_offset;
+        self.config.update_capability_chain(self.buffer.len());
+        self.config.write_bytes(offset, self.buffer.as_ref());
+    }
+
+}
+
+pub struct PciConfiguration {
+    address: PciAddress,
+    irq: u8,
+    bytes: [u8; PCI_CONFIG_SPACE_SIZE],
+    bar_write_masks: [u32; 6],
+    next_capability_offset: usize,
+}
+
+impl PciConfiguration {
+    pub fn new(irq: u8, vendor: u16, device: u16, class_id: u16) -> Self {
+        let mut config = PciConfiguration {
+            address: PciAddress::empty(),
+            irq,
+            bytes: [0; PCI_CONFIG_SPACE_SIZE],
+            bar_write_masks: [0; 6],
+            next_capability_offset: PCI_CAP_BASE_OFFSET,
+        };
+
+        config.buffer()
+            .write_at(PCI_VENDOR_ID, vendor)
+            .write_at(PCI_DEVICE_ID, device)
+            .write_at(PCI_COMMAND, PCI_COMMAND_IO | PCI_COMMAND_MEMORY)
+            .write_at(PCI_CLASS_REVISION, u8::from(1))
+            .write_at(PCI_CLASS_DEVICE, class_id)
+            .write_at(PCI_INTERRUPT_PIN, u8::from(1))
+            .write_at(PCI_INTERRUPT_LINE, irq)
+            .write_at(PCI_SUBSYSTEM_ID, 0x40u16);
+
+        config
+    }
+
+    pub fn address(&self) -> PciAddress {
+        self.address
+    }
+
+    pub fn set_address(&mut self, address: PciAddress) {
+        self.address = address;
+    }
+
+    pub fn irq(&self) -> u8 {
+        self.irq
+    }
+
+    fn buffer(&mut self) -> ByteBuffer<&mut[u8]> {
+        ByteBuffer::from_bytes_mut(&mut self.bytes).little_endian()
+    }
+
+    fn write_bytes(&mut self, offset: usize, bytes: &[u8]) {
+        (&mut self.bytes[offset..offset+bytes.len()])
+            .copy_from_slice(bytes)
+    }
+
+    fn read_bytes(&self, offset: usize, bytes: &mut [u8]) {
+        bytes.copy_from_slice(&self.bytes[offset..offset+bytes.len()]);
+    }
+
+
+    fn bar_mask(&self, offset: usize) -> Option<u32> {
+
+        fn is_bar_offset(offset: usize) -> bool {
+            offset >= PCI_BAR0 && offset < (PCI_BAR5 + 4)
+        }
+
+        fn bar_idx(offset: usize) -> usize {
+            (offset - PCI_BAR0) / 4
+        }
+
+        if is_bar_offset(offset) {
+            Some(self.bar_write_masks[bar_idx(offset)])
+        } else {
+            None
+        }
+    }
+
+    fn write_masked_byte(&mut self, offset: usize, mask: u8, new_byte: u8) {
+        let orig = self.bytes[offset];
+
+        self.bytes[offset] = (orig & !mask) | (new_byte & mask);
+    }
+
+    fn write_bar(&mut self, offset: usize, data: &[u8]) {
+        let mask_bytes = match self.bar_mask(offset) {
+            Some(mask) if mask != 0 => mask.to_le_bytes(),
+            _ => return,
+        };
+        let mod4 = offset % 4;
+        let mask_bytes = &mask_bytes[mod4..];
+        assert!(mask_bytes.len() >= data.len());
+
+
+        for idx in 0..data.len() {
+            self.write_masked_byte(offset + idx, mask_bytes[idx], data[idx])
+        }
+    }
+
+    fn write_config(&mut self, offset: usize, data: &[u8]) {
+        let size = data.len();
+        match offset {
+            PCI_COMMAND | PCI_STATUS if size == 2 => {
+                self.write_bytes(offset, data)
+            },
+            PCI_CACHE_LINE_SIZE if size == 1 => {
+                self.write_bytes(offset, data)
+            },
+            PCI_BAR0..=0x27 => {
+                self.write_bar(offset, data)
+            }, // bars
+            _ => {},
+
+        }
+    }
+
+    fn is_valid_access(offset: u64, size: usize) -> bool {
+        fn check_aligned_range(offset: u64, size: usize) -> bool {
+            let offset = offset as usize;
+            offset + size <= PCI_CONFIG_SPACE_SIZE && offset % size == 0
+        }
+
+        match size {
+            4 => check_aligned_range(offset, 4),
+            2 => check_aligned_range(offset, 2),
+            1 => check_aligned_range(offset, 1),
+            _ => false,
+        }
+
+    }
+
+    fn next_capability(&self, offset: usize) -> Option<usize> {
+        fn is_valid_cap_offset(offset: usize) -> bool {
+            offset < 254 && offset >= PCI_CAP_BASE_OFFSET
+        }
+
+        if is_valid_cap_offset(offset) {
+            Some(self.bytes[offset + 1] as usize)
+        } else {
+            None
+        }
+    }
+
+    fn update_next_capability_offset(&mut self, caplen: usize) {
+        let aligned = (caplen + 3) & !3;
+        self.next_capability_offset += aligned;
+        assert!(self.next_capability_offset < PCI_CONFIG_SPACE_SIZE);
+    }
+
+    fn update_capability_chain(&mut self, caplen: usize)  {
+
+        let next_offset = self.next_capability_offset as u8;
+        self.update_next_capability_offset(caplen);
+
+        let mut cap_ptr = self.bytes[PCI_CAPABILITY_LIST] as usize;
+
+        if cap_ptr == 0 {
+            self.bytes[PCI_CAPABILITY_LIST] = next_offset;
+            self.bytes[PCI_STATUS] |= PCI_STATUS_CAP_LIST as u8;
+            return;
+        }
+
+        for _ in 0..MAX_CAPABILITY_COUNT {
+            if let Some(next) = self.next_capability(cap_ptr) {
+                if next == 0 {
+                    self.bytes[cap_ptr + 1] = next_offset;
+                    return;
+                }
+                cap_ptr = next;
+            }
+        }
+    }
+
+    pub fn new_capability(&mut self) -> PciCapability {
+        PciCapability::new_vendor_capability(self)
+    }
+
+    pub fn set_mmio_bar(&mut self, bar: PciBar, range: AddressRange) {
+        assert!(range.is_naturally_aligned(), "cannot set_mmio_bar() because mmio range is not naturally aligned");
+        self.bar_write_masks[bar.idx()] = !((range.size() as u32) - 1);
+        let offset = PCI_BAR0 + (bar.idx() * 4);
+        let address = (range.base() as u32).to_le_bytes();
+        self.write_bytes(offset, &address);
+    }
+
+    pub fn read(&self, offset: u64, data: &mut [u8]) {
+        if Self::is_valid_access(offset, data.len()) {
+            self.read_bytes(offset as usize, data)
+        } else {
+            data.fill(0xff)
+        }
+    }
+
+    pub fn write(&mut self, offset: u64, data: &[u8]) {
+        if Self::is_valid_access(offset, data.len()) {
+            self.write_config(offset as usize, data);
+        }
+    }
+}
+
+/*
+impl BusDevice for PciConfiguration {
+    fn read(&mut self, offset: u64, data: &mut [u8]) {
+        if Self::is_valid_access(offset, data.len()) {
+            self.read_bytes(offset as usize, data)
+        } else {
+            data.fill(0xff)
+        }
+    }
+
+    fn write(&mut self, offset: u64, data: &[u8]) {
+        if Self::is_valid_access(offset, data.len()) {
+            self.write_config(offset as usize, data);
+        }
+    }
+}
+
+ */

src/io/pci/consts.rs

@@ -0,0 +1,33 @@
+
+// Maximum number of logical devices on a PCI bus
+
+pub const PCI_MAX_DEVICES: usize = 32;
+
+// Vendor specific PCI capabilities
+
+pub const PCI_CAP_ID_VENDOR: u8 = 0x09;
+
+pub const PCI_CAP_BASE_OFFSET: usize = 0x40;
+
+pub const PCI_VENDOR_ID: usize = 0x00;
+pub const PCI_DEVICE_ID: usize = 0x02;
+pub const PCI_COMMAND: usize = 0x04;
+pub const PCI_COMMAND_IO: u16 = 0x01;
+pub const PCI_COMMAND_MEMORY: u16 = 0x02;
+pub const PCI_STATUS: usize = 0x06;
+pub const PCI_BAR0: usize = 0x10;
+pub const PCI_BAR5: usize = 0x24;
+pub const PCI_STATUS_CAP_LIST: u16 = 0x10;
+pub const PCI_CLASS_REVISION: usize = 0x08;
+pub const PCI_CLASS_DEVICE: usize = 0x0a;
+pub const PCI_CACHE_LINE_SIZE: usize = 0x0c;
+
+pub const _PCI_SUBSYSTEM_VENDOR_ID: usize = 0x2c;
+pub const PCI_SUBSYSTEM_ID: usize = 0x2e;
+pub const PCI_CAPABILITY_LIST: usize = 0x34;
+pub const PCI_INTERRUPT_LINE: usize = 0x3C;
+pub const PCI_INTERRUPT_PIN: usize = 0x3D;
+
+pub const PCI_VENDOR_ID_INTEL: u16 = 0x8086;
+pub const PCI_CLASS_BRIDGE_HOST: u16 = 0x0600;
+

src/io/pci/device.rs

@@ -0,0 +1,68 @@
+use std::sync::{Arc, Mutex};
+use crate::io::bus::BusDevice;
+use crate::io::pci::PciConfiguration;
+
+#[derive(Copy,Clone,Eq,PartialEq)]
+#[repr(u8)]
+pub enum PciBar {
+    Bar0 = 0,
+    Bar1 = 1,
+    Bar2 = 2,
+    Bar3 = 3,
+    Bar4 = 4,
+    Bar5 = 5,
+}
+
+impl PciBar {
+    pub fn idx(&self) -> usize {
+        *self as usize
+    }
+}
+
+pub enum PciBarAllocation {
+    Mmio(PciBar, usize),
+}
+
+pub trait PciDevice: Send {
+    fn config(&self) -> &PciConfiguration;
+    fn config_mut(&mut self) -> &mut PciConfiguration;
+
+    fn read_bar(&mut self, bar: PciBar, offset: u64, data: &mut [u8]) {
+        let (_,_,_) = (bar, offset, data);
+    }
+
+    fn write_bar(&mut self, bar: PciBar, offset: u64, data: &[u8]) {
+        let (_,_,_) = (bar,offset, data);
+    }
+
+    fn irq(&self) -> Option<u8> { None }
+
+    fn bar_allocations(&self) -> Vec<PciBarAllocation> { vec![] }
+
+    fn configure_bars(&mut self, allocations: Vec<(PciBar, u64)>) { let _ = allocations; }
+}
+
+pub struct MmioHandler {
+    bar: PciBar,
+    device: Arc<Mutex<dyn PciDevice+Send>>
+}
+
+impl MmioHandler {
+    pub fn new(bar: PciBar, device: Arc<Mutex<dyn PciDevice+Send>>) -> Self {
+        MmioHandler {
+            bar, device,
+        }
+    }
+}
+
+impl BusDevice for MmioHandler {
+    fn read(&mut self, offset: u64, data: &mut [u8]) {
+        let mut lock = self.device.lock().unwrap();
+        lock.read_bar(self.bar, offset, data)
+    }
+
+    fn write(&mut self, offset: u64, data: &[u8]) {
+        let mut lock = self.device.lock().unwrap();
+        lock.write_bar(self.bar, offset, data)
+    }
+}

src/io/pci/mod.rs

@@ -0,0 +1,10 @@
+
+mod address;
+mod bus;
+mod config;
+mod consts;
+mod device;
+pub use bus::{PciBus,PciIrq};
+pub use config::{PciCapability,PciConfiguration};
+pub use address::PciAddress;
+pub use device::{PciDevice,PciBar,PciBarAllocation,MmioHandler};

src/io/virtio/consts.rs

@@ -0,0 +1,74 @@
+
+#[derive(Copy,Clone,Eq,PartialEq,Debug)]
+#[repr(u32)]
+pub enum VirtioDeviceType {
+    Net = 1,
+    Block = 2,
+    Console = 3,
+    Rng = 4,
+    NineP = 9,
+    Wl = 63,
+}
+
+impl VirtioDeviceType {
+    // Base PCI device id for Virtio devices
+    const PCI_VIRTIO_DEVICE_ID_BASE: u16 = 0x1040;
+
+    const PCI_CLASS_NETWORK_ETHERNET: u16 = 0x0200;
+    const PCI_CLASS_STORAGE_SCSI: u16 = 0x0100;
+    const PCI_CLASS_COMMUNICATION_OTHER: u16 = 0x0780;
+    const PCI_CLASS_OTHERS: u16 = 0xff;
+    const PCI_CLASS_STORAGE_OTHER: u16 = 0x0180;
+
+    pub fn device_id(&self) -> u16 {
+        Self::PCI_VIRTIO_DEVICE_ID_BASE + (*self as u16)
+    }
+
+    pub fn class_id(&self) -> u16 {
+        match self {
+            VirtioDeviceType::Net => Self::PCI_CLASS_NETWORK_ETHERNET,
+            VirtioDeviceType::Block => Self::PCI_CLASS_STORAGE_SCSI,
+            VirtioDeviceType::Console => Self::PCI_CLASS_COMMUNICATION_OTHER,
+            VirtioDeviceType::Rng => Self::PCI_CLASS_OTHERS,
+            VirtioDeviceType::NineP => Self::PCI_CLASS_STORAGE_OTHER,
+            VirtioDeviceType::Wl => Self::PCI_CLASS_OTHERS,
+        }
+    }
+}
+
+pub const VIRTIO_MMIO_AREA_SIZE: usize = 4096;
+
+// Offsets and sizes for each structure in MMIO area
+
+pub const VIRTIO_MMIO_OFFSET_COMMON_CFG : u64 = 0;     // Common configuration offset
+pub const VIRTIO_MMIO_OFFSET_ISR        : u64 = 56;    // ISR register offset
+pub const VIRTIO_MMIO_OFFSET_NOTIFY     : u64 = 0x400; // Notify area offset
+pub const VIRTIO_MMIO_OFFSET_DEV_CFG    : u64 = 0x800; // Device specific configuration offset
+
+pub const VIRTIO_MMIO_COMMON_CFG_SIZE: u64 = 56;       // Common configuration size
+pub const VIRTIO_MMIO_NOTIFY_SIZE    : u64 = 0x400;    // Notify area size
+pub const VIRTIO_MMIO_ISR_SIZE       : u64 = 4;        // ISR register size
+
+
+// Common configuration status bits
+
+pub const _VIRTIO_CONFIG_S_ACKNOWLEDGE : u8 = 1;
+pub const _VIRTIO_CONFIG_S_DRIVER      : u8 = 2;
+pub const VIRTIO_CONFIG_S_DRIVER_OK   : u8 = 4;
+pub const VIRTIO_CONFIG_S_FEATURES_OK : u8 = 8;
+pub const VIRTIO_CONFIG_S_FAILED      : u8 = 0x80;
+
+pub const MAX_QUEUE_SIZE: u16 = 1024;
+
+pub const VIRTIO_NO_MSI_VECTOR: u16 = 0xFFFF;
+
+// Bar number 0 is used for Virtio MMIO area
+
+pub const VIRTIO_MMIO_BAR: usize = 0;
+
+// Virtio PCI capability types
+
+pub const VIRTIO_PCI_CAP_COMMON_CFG : u8 = 1;
+pub const VIRTIO_PCI_CAP_NOTIFY_CFG : u8 = 2;
+pub const VIRTIO_PCI_CAP_ISR_CFG    : u8 = 3;
+pub const VIRTIO_PCI_CAP_DEVICE_CFG : u8 = 4;

src/io/virtio/device.rs

@@ -0,0 +1,430 @@
+use std::ops::Range;
+use std::sync::{Arc, Mutex, MutexGuard};
+use byteorder::{ByteOrder, LittleEndian};
+
+use crate::io::busdata::{ReadableInt, WriteableInt};
+use crate::io::pci::{PciBar, PciBarAllocation, PciConfiguration, PciDevice};
+use crate::io::virtio::consts::*;
+use crate::io::virtio::features::FeatureBits;
+use crate::io::virtio::queues::Queues;
+use crate::io::virtio::Result;
+use crate::io::PCI_VENDOR_ID_REDHAT;
+use crate::memory::{AddressRange, MemoryManager};
+
+pub trait VirtioDevice: Send {
+
+    fn features(&self) -> &FeatureBits;
+    fn features_ok(&self) -> bool { true }
+
+    fn queue_sizes(&self) -> &[u16];
+    fn device_type(&self) -> VirtioDeviceType;
+
+    fn config_size(&self) -> usize { 0 }
+
+    fn read_config(&self, offset: u64, data: &mut [u8]) {
+        let (_,_) = (offset, data);
+    }
+
+    fn write_config(&mut self, offset: u64, data: &[u8]) {
+        let (_,_) = (offset, data);
+    }
+
+    fn start(&mut self, queues: &Queues);
+}
+
+pub struct VirtioDeviceState {
+    pci_config: PciConfiguration,
+    device: Arc<Mutex<dyn VirtioDevice>>,
+    status: u8,
+    queues: Queues,
+}
+
+impl VirtioDeviceState {
+
+    pub fn new<T: VirtioDevice+'static>(device: T, memory: MemoryManager, irq: u8) -> Result<Self> {
+        let devtype = device.device_type();
+        let config_size = device.config_size();
+
+        let device = Arc::new(Mutex::new(device));
+        let queues = Queues::new(memory, irq)?;
+        let mut pci_config = PciConfiguration::new(queues.irq(), PCI_VENDOR_ID_REDHAT, devtype.device_id(), devtype.class_id());
+        Self::add_pci_capabilities::<T>(&mut pci_config, config_size);
+
+        Ok(VirtioDeviceState {
+            pci_config,
+            device,
+            status: 0,
+            queues,
+        })
+    }
+
+    fn add_pci_capabilities<T: VirtioDevice>(pci_config: &mut PciConfiguration, config_size: usize) {
+        VirtioPciCapability::new(VIRTIO_PCI_CAP_COMMON_CFG)
+            .set_mmio_range(VIRTIO_MMIO_OFFSET_COMMON_CFG, VIRTIO_MMIO_COMMON_CFG_SIZE)
+            .store(pci_config);
+
+        VirtioPciCapability::new(VIRTIO_PCI_CAP_ISR_CFG)
+            .set_mmio_range(VIRTIO_MMIO_OFFSET_ISR, VIRTIO_MMIO_ISR_SIZE)
+            .store(pci_config);
+
+        VirtioPciCapability::new(VIRTIO_PCI_CAP_NOTIFY_CFG)
+            .set_mmio_range(VIRTIO_MMIO_OFFSET_NOTIFY, VIRTIO_MMIO_NOTIFY_SIZE)
+            .set_extra_word(4)
+            .store(pci_config);
+
+        if config_size > 0 {
+            VirtioPciCapability::new(VIRTIO_PCI_CAP_DEVICE_CFG)
+                .set_mmio_range(VIRTIO_MMIO_OFFSET_DEV_CFG, config_size as u64)
+                .store(pci_config);
+        }
+    }
+
+    fn device(&self) -> MutexGuard<dyn VirtioDevice + 'static> {
+        self.device.lock().unwrap()
+    }
+
+    fn reset(&mut self) {
+        self.queues.reset();
+        self.device().features().reset();
+        self.status = 0;
+    }
+
+    fn status_write(&mut self, val: u8) {
+        let new_bits = val & !self.status;
+
+        let has_new_bit = |bit| -> bool {
+            new_bits & bit != 0
+        };
+
+        self.status |= new_bits;
+
+        if val == 0 {
+            self.reset();
+        } else if has_new_bit(VIRTIO_CONFIG_S_FEATURES_OK) {
+            // 2.2.2: The device SHOULD accept any valid subset of features the driver accepts,
+            // otherwise it MUST fail to set the FEATURES_OK device status bit when the driver
+            // writes it.
+            if !self.device().features_ok() {
+                self.status &= VIRTIO_CONFIG_S_FEATURES_OK;
+            }
+        } else if has_new_bit(VIRTIO_CONFIG_S_DRIVER_OK) {
+            let features = self.device().features().guest_value();
+            if let Err(err) = self.queues.configure_queues(features) {
+                warn!("Error configuring virtqueue: {}", err);
+            } else {
+                self.device().start(&self.queues)
+            }
+        } else if has_new_bit(VIRTIO_CONFIG_S_FAILED) {
+            // XXX print a warning
+        }
+    }
+
+    fn common_config_write(&mut self, offset: u64, val: WriteableInt) {
+        match val {
+            WriteableInt::Byte(n) => match offset {
+                /* device_status */
+                20 => self.status_write(n),
+                _ => warn!("VirtioDeviceState: common_config_write: unhandled byte offset {}", offset),
+            },
+            WriteableInt::Word(n) => match offset {
+                /* queue_select */
+                22 => self.queues.select(n),
+                /* queue_size */
+                24 => self.queues.set_size(n),
+                /* queue_enable */
+                28 => self.queues.enable_current(),
+                _ => warn!("VirtioDeviceState: common_config_write: unhandled word offset {}", offset),
+            }
+            WriteableInt::DWord(n) => match offset {
+                /* device_feature_select */
+                0 => self.device().features().set_device_selected(n),
+                /* guest_feature_select */
+                8 => self.device().features().set_guest_selected(n),
+                /* guest_feature */
+                12 => self.device().features().write_guest_word(n),
+                /* queue_desc_lo */
+                32 => self.queues.set_current_descriptor_area(n, false),
+                /* queue_desc_hi */
+                36 => self.queues.set_current_descriptor_area(n, true),
+                /* queue_avail_lo */
+                40 => self.queues.set_avail_area(n, false),
+                /* queue_avail_hi */
+                44 => self.queues.set_avail_area(n, true),
+                /* queue_used_lo */
+                48 => self.queues.set_used_area(n, false),
+                /* queue_used_hi */
+                52 => self.queues.set_used_area(n, true),
+                _ => warn!("VirtioDeviceState: common_config_write: unhandled dword offset {}", offset),
+            },
+            WriteableInt::QWord(_) => warn!("VirtioDeviceState: common_config_write: unhandled qword offset {}", offset),
+            WriteableInt::Data(bs) => warn!("VirtioDeviceState: common_config_write: unhandled raw bytes offset {}, len {}", offset, bs.len()),
+        }
+    }
+
+    fn common_config_read(&self, offset: u64) -> ReadableInt {
+        match offset {
+            /* device_feature_select */
+            0 => self.device().features().device_selected().into(),
+            /* device_feature */
+            4 => self.device().features().read_device_word().into(),
+            /* guest_feature_select */
+            8 => self.device().features().guest_selected().into(),
+            /* guest_feature */
+            12 => self.device().features().read_guest_word().into(),
+            /* msix_config */
+            16 => VIRTIO_NO_MSI_VECTOR.into(),
+            /* num_queues */
+            18 => self.queues.num_queues().into(),
+            /* device_status */
+            20 => self.status.into(),
+            /* config_generation */
+            21 => (0u8).into(),
+            /* queue_select */
+            22 => self.queues.selected_queue().into(),
+            /* queue_size */
+            24 => self.queues.queue_size().into(),
+            /* queue_msix_vector */
+            26 => VIRTIO_NO_MSI_VECTOR.into(),
+            /* queue_enable */
+            28 => if self.queues.is_current_enabled() { 1u16.into() } else { 0u16.into() },
+            /* queue_notify_off */
+            30 => self.queues.selected_queue().into(),
+            /* queue_desc_lo */
+            32 => self.queues.get_current_descriptor_area(false).into(),
+            /* queue_desc_hi */
+            36 => self.queues.get_current_descriptor_area(true).into(),
+            /* queue_avail_lo */
+            40 => self.queues.get_avail_area(false).into(),
+            /* queue_avail_hi */
+            44 => self.queues.get_avail_area(true).into(),
+            /* queue_used_lo */
+            48 => self.queues.get_used_area(false).into(),
+            /* queue_used_hi */
+            52 => self.queues.get_used_area(true).into(),
+            _ => ReadableInt::new_dword(0),
+        }
+    }
+
+    fn isr_read(&self) -> u8 {
+        self.queues.isr_read() as u8
+    }
+
+    fn is_device_config_range(&self, offset: u64, len: usize) -> bool {
+        let dev = self.device();
+        if dev.config_size() > 0 {
+            let range = AddressRange::new(VIRTIO_MMIO_OFFSET_DEV_CFG, dev.config_size());
+            range.contains(offset, len)
+        } else {
+            false
+        }
+    }
+
+    fn is_common_cfg_range(&self, offset: u64, len: usize) -> bool {
+        AddressRange::new(VIRTIO_MMIO_OFFSET_COMMON_CFG, VIRTIO_MMIO_COMMON_CFG_SIZE as usize)
+            .contains(offset, len)
+    }
+}
+
+impl PciDevice for VirtioDeviceState {
+
+    fn config(&self) -> &PciConfiguration {
+        &self.pci_config
+    }
+
+    fn config_mut(&mut self) -> &mut PciConfiguration {
+        &mut self.pci_config
+    }
+
+    fn read_bar(&mut self, bar: PciBar, offset: u64, data: &mut [u8]) {
+        if bar != PciBar::Bar0 {
+            warn!("Virtio PciDevice: read_bar() expected bar0!");
+            return;
+
+        }
+
+        if self.is_common_cfg_range(offset, data.len()) {
+            let v = self.common_config_read(offset);
+            v.read(data);
+        } else if offset == VIRTIO_MMIO_OFFSET_ISR && data.len() == 1 {
+            data[0] = self.isr_read();
+        } else if self.is_device_config_range(offset, data.len()) {
+            let dev = self.device();
+            dev.read_config(offset - VIRTIO_MMIO_OFFSET_DEV_CFG, data);
+        }
+    }
+
+    fn write_bar(&mut self, bar: PciBar, offset: u64, data: &[u8]) {
+        if bar != PciBar::Bar0 {
+            warn!("Virtio PciDevice: write_bar() expected bar0!");
+            return;
+        }
+        if self.is_common_cfg_range(offset, data.len()) {
+            let data = WriteableInt::from(data);
+            self.common_config_write(offset, data);
+        } else if self.is_device_config_range(offset, data.len()) {
+            let mut dev = self.device();
+            dev.write_config(offset - VIRTIO_MMIO_OFFSET_DEV_CFG, data);
+        }
+    }
+
+    fn irq(&self) -> Option<u8> {
+        Some(self.queues.irq())
+    }
+
+    fn bar_allocations(&self) -> Vec<PciBarAllocation> {
+        vec![PciBarAllocation::Mmio(PciBar::Bar0, VIRTIO_MMIO_AREA_SIZE)]
+    }
+
+    fn configure_bars(&mut self, allocations: Vec<(PciBar, u64)>) {
+        for (bar,base) in allocations {
+            if bar == PciBar::Bar0 {
+                let queue_sizes = self.device().queue_sizes().to_vec();
+                if let Err(e) = self.queues.create_queues(base, &queue_sizes) {
+                    warn!("Error creating queues: {}", e);
+                }
+            } else {
+                warn!("Virtio PciDevice: Cannot configure unexpected PCI bar: {}", bar.idx());
+            }
+        }
+    }
+}
+
+struct VirtioPciCapability {
+    vtype: u8,
+    size: u8,
+    mmio_offset: u32,
+    mmio_len: u32,
+    extra_word: Option<u32>,
+}
+
+impl VirtioPciCapability {
+    fn new(vtype: u8) -> VirtioPciCapability{
+        VirtioPciCapability {
+            vtype,
+            size: 16,
+            mmio_offset: 0,
+            mmio_len: 0,
+            extra_word: None
+        }
+    }
+    fn set_mmio_range(&mut self, offset: u64, len: u64) -> &mut VirtioPciCapability {
+        self.mmio_offset = offset as u32;
+        self.mmio_len = len as u32;
+        self
+    }
+
+    fn set_extra_word(&mut self, val: u32) -> &mut VirtioPciCapability {
+        self.size += 4;
+        self.extra_word = Some(val);
+        self
+    }
+
+    fn store(&self, pci_config: &mut PciConfiguration) {
+        /*
+         * struct virtio_pci_cap {
+         *     u8 cap_vndr; /* Generic PCI field: PCI_CAP_ID_VNDR */
+         *     u8 cap_next; /* Generic PCI field: next ptr. */
+         *     u8 cap_len; /* Generic PCI field: capability length */
+         *     u8 cfg_type; /* Identifies the structure. */
+         *     u8 bar; /* Where to find it. */
+         *     u8 padding[3]; /* Pad to full dword. */
+         *     le32 offset; /* Offset within bar. */
+         *     le32 length; /* Length of the structure, in bytes. */
+         * };
+         */
+        let mut cap = pci_config.new_capability();
+        cap.write(self.size);
+        cap.write(self.vtype);
+        // Also fills the padding bytes
+        cap.write(VIRTIO_MMIO_BAR as u32);
+        cap.write(self.mmio_offset);
+        cap.write(self.mmio_len);
+        if let Some(word) = self.extra_word {
+            cap.write(word);
+        }
+        cap.store();
+    }
+}
+
+pub struct DeviceConfigArea {
+    buffer: Vec<u8>,
+    write_filter: DeviceConfigWriteFilter,
+}
+
+
+#[allow(dead_code)]
+impl DeviceConfigArea {
+    pub fn new(size: usize) -> Self {
+        DeviceConfigArea{
+            buffer: vec![0u8; size],
+            write_filter: DeviceConfigWriteFilter::new(size),
+        }
+    }
+
+    pub fn read_config(&self, offset: u64, data: &mut [u8]) {
+        let offset = offset as usize;
+        if offset + data.len() <= self.buffer.len() {
+            data.copy_from_slice(&self.buffer[offset..offset+data.len()]);
+        }
+    }
+    pub fn write_config(&mut self, offset: u64, data: &[u8]) {
+        let offset = offset as usize;
+        if self.write_filter.is_writeable(offset, data.len()) {
+            self.buffer[offset..offset+data.len()].copy_from_slice(data);
+        }
+    }
+
+    pub fn set_writeable(&mut self, offset: usize, size: usize) {
+        self.write_filter.set_writable(offset, size)
+    }
+
+    pub fn write_u8(&mut self, offset: usize, val: u8) {
+        assert!(offset + 1 <= self.buffer.len());
+        self.buffer[offset] = val;
+    }
+
+    pub fn write_u16(&mut self, offset: usize, val: u16) {
+        assert!(offset + 2 <= self.buffer.len());
+        LittleEndian::write_u16(&mut self.buffer[offset..], val);
+    }
+
+    pub fn write_u32(&mut self, offset: usize, val: u32) {
+        assert!(offset + 4 <= self.buffer.len());
+        LittleEndian::write_u32(&mut self.buffer[offset..], val);
+    }
+
+    pub fn write_u64(&mut self, offset: usize, val: u64) {
+        assert!(offset + 8 <= self.buffer.len());
+        LittleEndian::write_u64(&mut self.buffer[offset..], val);
+    }
+
+    pub fn write_bytes(&mut self, offset: usize, bytes: &[u8]) {
+        assert!(offset + bytes.len() <= self.buffer.len());
+        self.buffer[offset..offset + bytes.len()].copy_from_slice(bytes);
+    }
+}
+
+struct DeviceConfigWriteFilter {
+    size: usize,
+    ranges: Vec<Range<usize>>,
+}
+
+impl DeviceConfigWriteFilter {
+    fn new(size: usize) -> Self {
+        DeviceConfigWriteFilter { size, ranges: Vec::new() }
+    }
+
+    fn set_writable(&mut self, offset: usize, size: usize) {
+        let end = offset + size;
+        self.ranges.push(offset..end);
+    }
+
+    fn is_writeable(&self, offset: usize, size: usize) -> bool {
+        if offset + size > self.size {
+            false
+        } else {
+            let last = offset + size - 1;
+            self.ranges.iter().any(|r| r.contains(&offset) && r.contains(&last))
+        }
+    }
+}

src/io/virtio/features.rs

@@ -0,0 +1,110 @@
+use std::sync::{Arc, Mutex, MutexGuard};
+
+#[derive(Copy,Clone)]
+#[repr(u64)]
+pub enum ReservedFeatureBit {
+    _IndirectDesc = 1 << 28,
+    EventIdx = 1 << 29,
+    Version1 = 1 << 32,
+}
+
+impl ReservedFeatureBit {
+    pub fn is_set_in(&self, flags: u64) -> bool {
+        flags & (*self as u64) != 0
+    }
+}
+
+#[derive(Clone)]
+pub struct FeatureBits {
+    device_bits: Arc<Mutex<Inner>>,
+    guest_bits: Arc<Mutex<Inner>>,
+}
+
+struct Inner {
+    bits: u64,
+    selected: u32,
+}
+
+impl Inner {
+    fn new(bits: u64) -> Arc<Mutex<Self>> {
+        Arc::new(Mutex::new(Inner { bits, selected: 0 }))
+    }
+}
+
+impl FeatureBits {
+
+    pub fn new_default(device_bits: u64) -> Self {
+        FeatureBits {
+            guest_bits: Inner::new(0),
+            device_bits: Inner::new(ReservedFeatureBit::Version1 as u64 | device_bits),
+        }
+    }
+
+    pub fn reset(&self) {
+        let mut guest = self.guest();
+        guest.bits = 0;
+        guest.selected = 0;
+    }
+
+    fn guest(&self) -> MutexGuard<Inner> {
+        self.guest_bits.lock().unwrap()
+    }
+    fn device(&self) -> MutexGuard<Inner> {
+        self.device_bits.lock().unwrap()
+    }
+
+    pub fn guest_selected(&self) -> u32 {
+        self.guest().selected
+    }
+
+    pub fn guest_value(&self) -> u64 {
+        self.guest().bits
+    }
+
+    pub fn has_guest_bit(&self, bit: u64) -> bool {
+        self.guest_value() & bit == bit
+    }
+
+    pub fn set_guest_selected(&self, val: u32) {
+        self.guest().selected = val;
+    }
+
+    pub fn write_guest_word(&self, val: u32) {
+        const MASK_LOW_32: u64 = (1u64 << 32) - 1;
+        const MASK_HI_32: u64 = MASK_LOW_32 << 32;
+        let mut inner = self.guest();
+        let val = u64::from(val);
+
+        match inner.selected {
+            0 => inner.bits = (inner.bits & MASK_HI_32) | val,
+            1 => inner.bits = val << 32 | (inner.bits & MASK_LOW_32),
+            _ => (),
+        }
+    }
+
+    pub fn read_guest_word(&self) -> u32 {
+        let inner = self.guest();
+        match inner.selected {
+            0 => inner.bits as u32,
+            1 => (inner.bits >> 32) as u32,
+            _ => 0,
+        }
+    }
+
+    pub fn set_device_selected(&self, val: u32) {
+        self.device().selected = val;
+    }
+
+    pub fn device_selected(&self) -> u32 {
+        self.device().selected
+    }
+
+    pub fn read_device_word(&self) -> u32 {
+        let inner = self.device();
+        match inner.selected {
+            0 => inner.bits as u32,
+            1 => (inner.bits >> 32) as u32,
+            _ => 0,
+        }
+    }
+}

src/io/virtio/mod.rs

@@ -0,0 +1,41 @@
+mod device;
+mod consts;
+mod vq;
+mod queues;
+mod features;
+
+use std::result;
+pub use device::{VirtioDeviceState, VirtioDevice, DeviceConfigArea};
+pub use queues::Queues;
+pub use features::FeatureBits;
+pub use consts::VirtioDeviceType;
+pub use vq::virtqueue::VirtQueue;
+pub use vq::chain::Chain;
+use crate::io::bus::Error as BusError;
+
+use thiserror::Error;
+use vmm_sys_util::errno;
+
+pub type Result<T> = result::Result<T, Error>;
+
+#[derive(Debug,Error)]
+pub enum Error {
+    #[error("failed to create EventFd for VirtQueue: {0}")]
+    CreateEventFd(std::io::Error),
+    #[error("failed to create IoEventFd for VirtQueue: {0}")]
+    CreateIoEventFd(kvm_ioctls::Error),
+    #[error("failed to read from IoEventFd: {0}")]
+    ReadIoEventFd(std::io::Error),
+    #[error("VirtQueue not enabled")]
+    QueueNotEnabled,
+    #[error("VirtQueue descriptor table range is invalid 0x{0:x}")]
+    RangeInvalid(u64),
+    #[error("VirtQueue avail ring range range is invalid 0x{0:x}")]
+    AvailInvalid(u64),
+    #[error("VirtQueue used ring range is invalid 0x{0:x}")]
+    UsedInvalid(u64),
+    #[error("{0}")]
+    BusInsert(#[from]BusError),
+    #[error("Error registering irqfd: {0}")]
+    IrqFd(errno::Error),
+}

src/io/virtio/queues.rs

@@ -0,0 +1,247 @@
+use std::sync::Arc;
+use std::sync::atomic::{AtomicUsize, Ordering};
+use kvm_ioctls::{IoEventAddress, NoDatamatch};
+use vmm_sys_util::eventfd::EventFd;
+use crate::memory::MemoryManager;
+use crate::io::virtio::{Error, Result};
+use crate::io::virtio::consts::VIRTIO_MMIO_OFFSET_NOTIFY;
+use crate::io::VirtQueue;
+use crate::vm::KvmVm;
+
+pub struct InterruptLine {
+    irqfd: EventFd,
+    irq: u8,
+    isr: AtomicUsize,
+}
+
+impl InterruptLine {
+    fn new(kvm_vm: &KvmVm, irq: u8) -> Result<InterruptLine> {
+        let irqfd = EventFd::new(0)
+            .map_err(Error::CreateEventFd)?;
+        kvm_vm.vm_fd().register_irqfd(&irqfd, irq as u32)
+            .map_err(Error::IrqFd)?;
+        Ok(InterruptLine{
+            irqfd,
+            irq,
+            isr: AtomicUsize::new(0)
+        })
+
+    }
+
+    fn irq(&self) -> u8 {
+        self.irq
+    }
+
+
+    fn isr_read(&self) -> u64 {
+        self.isr.swap(0, Ordering::SeqCst) as u64
+    }
+
+    pub fn notify_queue(&self) {
+        self.isr.fetch_or(0x1, Ordering::SeqCst);
+        self.irqfd.write(1).unwrap();
+    }
+
+    pub fn notify_config(&self) {
+        self.isr.fetch_or(0x2, Ordering::SeqCst);
+        self.irqfd.write(1).unwrap();
+    }
+}
+
+pub struct Queues {
+    memory: MemoryManager,
+    selected_queue: u16,
+    queues: Vec<VirtQueue>,
+    interrupt: Arc<InterruptLine>,
+}
+
+impl Queues {
+    pub fn new(memory: MemoryManager, irq: u8) -> Result<Self> {
+        let interrupt = InterruptLine::new(memory.kvm_vm(), irq)?;
+        let queues = Queues {
+            memory,
+            selected_queue: 0,
+            queues: Vec::new(),
+            interrupt: Arc::new(interrupt),
+        };
+        Ok(queues)
+    }
+
+    pub fn get_queue(&self, idx: usize) -> VirtQueue {
+        self.queues
+            .get(idx)
+            .cloned()
+            .expect(&format!("Virtio device requested VQ index {} that does not exist", idx))
+    }
+
+    pub fn queues(&self) -> Vec<VirtQueue> {
+        self.queues.clone()
+    }
+
+    pub fn memory(&self) -> &MemoryManager {
+        &self.memory
+    }
+
+    pub fn configure_queues(&self, features: u64) -> Result<()> {
+        for q in &self.queues {
+            q.configure(features)?;
+        }
+        Ok(())
+    }
+
+    pub fn reset(&mut self) {
+        self.selected_queue = 0;
+        let _ = self.isr_read();
+        for vr in &mut self.queues {
+            vr.reset();
+        }
+    }
+
+    pub fn irq(&self) -> u8 {
+        self.interrupt.irq()
+    }
+
+    pub fn isr_read(&self) -> u64 {
+        self.interrupt.isr_read()
+    }
+
+    pub fn num_queues(&self) -> u16 {
+        self.queues.len() as u16
+    }
+
+    pub fn create_queues(&mut self, mmio_base: u64, queue_sizes: &[u16]) -> Result<()> {
+        let mut idx = 0;
+        for &sz in queue_sizes {
+            let ioevent = self.create_ioevent(idx, mmio_base)?;
+            let vq = VirtQueue::new(self.memory.guest_ram().clone(), sz, self.interrupt.clone(), ioevent);
+            self.queues.push(vq);
+            idx += 1;
+        }
+        Ok(())
+    }
+
+    fn create_ioevent(&self, index: usize, mmio_base: u64) -> Result<Arc<EventFd>> {
+        let evt = EventFd::new(0)
+            .map_err(Error::CreateEventFd)?;
+
+        let notify_address = mmio_base +
+            VIRTIO_MMIO_OFFSET_NOTIFY +
+            (4 * index as u64);
+
+        let addr = IoEventAddress::Mmio(notify_address);
+
+        self.memory.kvm_vm().vm_fd().register_ioevent(&evt, &addr, NoDatamatch)
+            .map_err(Error::CreateIoEventFd)?;
+
+        Ok(Arc::new(evt))
+    }
+
+
+    fn current_queue(&self) -> Option<&VirtQueue> {
+        self.queues.get(self.selected_queue as usize)
+    }
+
+    fn with_current<F>(&mut self, f: F)
+        where F: FnOnce(&mut VirtQueue)
+    {
+        if let Some(vq) = self.queues.get_mut(self.selected_queue as usize) {
+            if !vq.is_enabled() {
+                f(vq)
+            }
+        }
+    }
+
+    pub fn selected_queue(&self) -> u16 {
+        self.selected_queue
+    }
+
+    pub fn select(&mut self, index: u16) {
+        self.selected_queue = index;
+    }
+
+    pub fn is_current_enabled(&self) -> bool {
+        self.current_queue()
+            .map(|q| q.is_enabled())
+            .unwrap_or(false)
+    }
+
+    pub fn queue_size(&self) -> u16 {
+        self.current_queue()
+            .map(|q| q.size())
+            .unwrap_or(0)
+    }
+
+    pub fn set_size(&mut self, size: u16) {
+        self.with_current(|q| q.set_size(size))
+    }
+
+    pub fn enable_current(&mut self) {
+        self.with_current(|q| q.enable())
+    }
+
+    pub fn get_current_descriptor_area(&self, hi_word: bool) -> u32 {
+        self.current_queue().map(|q| if hi_word {
+            Self::get_hi32(q.descriptor_area())
+        } else {
+            Self::get_lo32(q.descriptor_area())
+        }).unwrap_or(0)
+
+    }
+    pub fn set_current_descriptor_area(&mut self, val: u32, hi_word: bool) {
+        self.with_current(|q| {
+            let mut addr = q.descriptor_area();
+            if hi_word { Self::set_hi32(&mut addr, val) } else { Self::set_lo32(&mut addr, val) }
+            q.set_descriptor_area(addr);
+        });
+    }
+
+    pub fn get_avail_area(&self, hi_word: bool) -> u32 {
+       self.current_queue().map(|q| if hi_word {
+           Self::get_hi32(q.driver_area())
+       } else {
+           Self::get_lo32(q.driver_area())
+       }).unwrap_or(0)
+    }
+
+    fn set_hi32(val: &mut u64, dword: u32) {
+        const MASK_LO_32: u64 = (1u64 << 32) - 1;
+        *val = (*val & MASK_LO_32) | (u64::from(dword) << 32)
+    }
+
+    fn set_lo32(val: &mut u64, dword: u32) {
+        const MASK_HI_32: u64 = ((1u64 << 32) - 1) << 32;
+        *val = (*val & MASK_HI_32) | u64::from(dword)
+    }
+
+    fn get_hi32(val: u64) -> u32 {
+        (val >> 32) as u32
+    }
+
+    fn get_lo32(val: u64) -> u32 {
+        val as u32
+    }
+
+    pub fn set_avail_area(&mut self, val: u32, hi_word: bool) {
+        self.with_current(|q| {
+            let mut addr = q.driver_area();
+            if hi_word { Self::set_hi32(&mut addr, val) } else { Self::set_lo32(&mut addr, val) }
+            q.set_driver_area(addr);
+        });
+    }
+
+    pub fn set_used_area(&mut self, val: u32, hi_word: bool) {
+        self.with_current(|q| {
+            let mut addr = q.device_area();
+            if hi_word { Self::set_hi32(&mut addr, val) } else { Self::set_lo32(&mut addr, val) }
+            q.set_device_area(addr);
+        });
+    }
+
+    pub fn get_used_area(&self, hi_word: bool) -> u32 {
+        self.current_queue().map(|q| if hi_word {
+            Self::get_hi32(q.device_area())
+        } else {
+            Self::get_lo32(q.device_area())
+        }).unwrap_or(0)
+    }
+}

src/io/virtio/vq/chain.rs

@@ -1,11 +1,11 @@
-use std::fmt;
+use std::{fmt, io};
-use std::io::{self,Read,Write};
+use std::io::{Read, Write};
-
+use std::sync::{Arc, Mutex};
+use crate::io::virtio::vq::descriptor::Descriptor;
+use crate::io::virtio::vq::virtqueue::QueueBackend;
 use crate::memory::GuestRam;
-use crate::virtio::VirtQueue;
-use crate::virtio::vring::Descriptor;
 
-struct DescriptorList {
+pub struct DescriptorList {
     memory: GuestRam,
     descriptors: Vec<Descriptor>,
     offset: usize,
@@ -14,7 +14,7 @@ struct DescriptorList {
 }
 
 impl DescriptorList {
-    fn new(memory: GuestRam) -> Self {
+    pub fn new(memory: GuestRam) -> Self {
         DescriptorList {
             memory,
             descriptors: Vec::new(),
@@ -24,12 +24,12 @@ impl DescriptorList {
         }
     }
 
-    fn add_descriptor(&mut self, d: Descriptor) {
+    pub fn add_descriptor(&mut self, d: Descriptor) {
-        self.total_size += d.len as usize;
+        self.total_size += d.length();
         self.descriptors.push(d)
     }
 
-    fn reverse(&mut self) {
+    pub fn reverse(&mut self) {
         self.descriptors.reverse();
     }
 
@@ -38,7 +38,7 @@ impl DescriptorList {
         self.offset = 0;
     }
 
-    fn is_empty(&self) -> bool {
+    pub fn is_empty(&self) -> bool {
         self.descriptors.is_empty()
     }
 
@@ -49,7 +49,7 @@ impl DescriptorList {
     fn current_address(&self, size: usize) -> Option<u64> {
         self.current().and_then(|d| {
             if d.remaining(self.offset) >= size {
-                Some(d.addr + self.offset as u64)
+                Some(d.address() + self.offset as u64)
             } else {
                 None
             }
@@ -111,7 +111,7 @@ impl DescriptorList {
     fn current_slice(&self) -> &[u8] {
         if let Some(d) = self.current() {
             let size = d.remaining(self.offset);
-            let addr = d.addr + self.offset as u64;
+            let addr = d.address() + self.offset as u64;
             self.memory.slice(addr, size).unwrap_or(&[])
         } else {
             &[]
@@ -121,7 +121,7 @@ impl DescriptorList {
     fn current_mut_slice(&self) -> &mut [u8] {
         if let Some(d) = self.current() {
             let size = d.remaining(self.offset);
-            let addr = d.addr + self.offset as u64;
+            let addr = d.address() + self.offset as u64;
             self.memory.mut_slice(addr, size).unwrap_or(&mut [])
         } else {
             &mut []
@@ -135,64 +135,31 @@ impl DescriptorList {
 
 impl fmt::Debug for DescriptorList {
     fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
-        write!(f, "DList[size={}, [", self.total_size)?;
+        write!(f, "[size={}, [", self.total_size)?;
         for d in self.descriptors.iter().rev() {
-            write!(f, "(#{}, 0x{:08x}, [{}]),", d.idx, d.addr, d.len)?;
+            write!(f, "(0x{:08x}, [{}]),", d.address(), d.length())?;
         }
-        write!(f, "]")
+        write!(f, "] ]")
     }
 }
 
 pub struct Chain {
+    backend: Arc<Mutex<dyn QueueBackend>>,
     head: Option<u16>,
-    vq: VirtQueue,
     readable: DescriptorList,
     writeable: DescriptorList,
 }
 
 impl Chain {
-    pub fn new(memory: GuestRam, vq: VirtQueue, head: u16, ttl: u16) -> Self {
+    pub fn new(backend: Arc<Mutex<dyn QueueBackend>>, head: u16, readable: DescriptorList, writeable: DescriptorList) -> Self {
-        let (readable,writeable) = Self::load_descriptors(memory, &vq, head, ttl);
         Chain {
+            backend,
             head: Some(head),
-            vq,
             readable,
             writeable,
         }
     }
 
-    fn load_descriptors(memory: GuestRam, vq: &VirtQueue, head: u16, ttl: u16) -> (DescriptorList, DescriptorList) {
-        let mut readable = DescriptorList::new(memory.clone());
-        let mut writeable = DescriptorList::new(memory);
-        let mut idx = head;
-        let mut ttl = ttl;
-
-        while let Some(d) = vq.load_descriptor(idx) {
-            if ttl == 0 {
-                warn!("Descriptor chain length exceeded ttl");
-                break;
-            } else {
-                ttl -= 1;
-            }
-
-            if d.is_write() {
-                writeable.add_descriptor(d);
-            } else {
-                if !writeable.is_empty() {
-                    warn!("Guest sent readable virtqueue descriptor after writeable descriptor in violation of specification");
-                }
-                readable.add_descriptor(d);
-            }
-            if !d.has_next() {
-                break;
-            }
-            idx = d.next;
-        }
-        readable.reverse();
-        writeable.reverse();
-        return (readable, writeable);
-    }
-
     pub fn w8(&mut self, n: u8) -> io::Result<()> {
         self.write_all(&[n])?;
         Ok(())
@@ -205,6 +172,7 @@ impl Chain {
         self.write_all(&n.to_le_bytes())?;
         Ok(())
     }
+
     pub fn w64(&mut self, n: u64) -> io::Result<()> {
         self.write_all(&n.to_le_bytes())?;
         Ok(())
@@ -231,7 +199,8 @@ impl Chain {
         if let Some(head) = self.head.take() {
             self.readable.clear();
             self.writeable.clear();
-            self.vq.put_used(head, self.writeable.consumed_size as u32);
+            let backend = self.backend.lock().unwrap();
+            backend.put_used(head, self.writeable.consumed_size as u32);
         }
     }
 

src/io/virtio/vq/descriptor.rs

@@ -0,0 +1,111 @@
+use std::{cmp, io};
+use std::io::Read;
+use crate::memory::GuestRam;
+
+#[repr(u16)]
+enum DescriptorFlag {
+    Next = 1,
+    Write = 2,
+    Indirect = 4,
+    PackedAvail = 1<<7,
+    PackedUsed = 1<<15,
+}
+
+#[derive(Copy,Clone)]
+pub struct Descriptor {
+    address: u64,
+    length: u32,
+    flags: u16,
+    // 'next' field for split virtqueue, 'buffer_id' for packed virtqueue
+    extra: u16,
+}
+
+impl Descriptor {
+
+    pub fn new(address: u64, length: u32, flags: u16, extra: u16) -> Self {
+        Descriptor {
+            address, length, flags, extra
+        }
+    }
+
+    pub fn length(&self) -> usize {
+        self.length as usize
+    }
+
+    pub fn address(&self) -> u64 {
+        self.address
+    }
+
+    ///
+    /// Test if `flag` is set in `self.flags`
+    ///
+    fn has_flag(&self, flag: DescriptorFlag) -> bool {
+        self.flags & (flag as u16) != 0
+    }
+
+    ///
+    /// Is VRING_DESC_F_NEXT set in `self.flags`?
+    ///
+    pub fn has_next(&self) -> bool {
+        self.has_flag(DescriptorFlag::Next)
+    }
+
+    pub fn next(&self) -> u16 {
+        self.extra
+    }
+
+    ///
+    /// Is VRING_DESC_F_WRITE set in `self.flags`?
+    ///
+    pub fn is_write(&self) -> bool {
+        self.has_flag(DescriptorFlag::Write)
+    }
+
+    ///
+    /// Is VRING_DESC_F_INDIRECT set in `self.flags`?
+    ///
+    pub fn is_indirect(&self) -> bool {
+        self.has_flag(DescriptorFlag::Indirect)
+    }
+
+
+    pub fn remaining(&self, offset: usize) -> usize {
+        if offset >= self.length as usize {
+            0
+        } else {
+            self.length as usize - offset
+        }
+    }
+
+    pub fn is_desc_avail(&self, wrap_counter: bool) -> bool {
+        let used = self.has_flag(DescriptorFlag::PackedUsed);
+        let avail = self.has_flag(DescriptorFlag::PackedAvail);
+        (used != avail) && (avail == wrap_counter)
+    }
+
+    pub fn read_from(&self, memory: &GuestRam, offset: usize, buf: &mut[u8]) -> usize {
+        let sz = cmp::min(buf.len(), self.remaining(offset));
+        if sz > 0 {
+            memory.read_bytes(self.address + offset as u64, &mut buf[..sz]).unwrap();
+        }
+        sz
+    }
+
+    pub fn write_to(&self, memory: &GuestRam, offset: usize, buf: &[u8]) -> usize {
+        let sz = cmp::min(buf.len(), self.remaining(offset));
+        if sz > 0 {
+            memory.write_bytes(self.address + offset as u64, &buf[..sz]).unwrap();
+        }
+        sz
+    }
+
+    pub fn write_from_reader<R: Read+Sized>(&self, memory: &GuestRam, offset: usize, mut r: R, size: usize) -> io::Result<usize> {
+        let sz = cmp::min(size, self.remaining(offset));
+        if sz > 0 {
+            let slice = memory.mut_slice(self.address + offset as u64, sz).unwrap();
+            return r.read(slice);
+        }
+        Ok(0)
+    }
+
+}

src/io/virtio/vq/mod.rs

@@ -0,0 +1,28 @@
+use std::sync::Arc;
+use std::sync::atomic::{AtomicUsize, Ordering};
+
+pub mod chain;
+mod descriptor;
+mod splitqueue;
+pub mod virtqueue;
+
+///
+/// A convenience wrapper around `AtomicUsize`
+///
+#[derive(Clone)]
+pub struct SharedIndex(Arc<AtomicUsize>);
+
+impl SharedIndex {
+    fn new() -> SharedIndex {
+        SharedIndex(Arc::new(AtomicUsize::new(0)))
+    }
+    fn get(&self) -> u16 {
+        self.0.load(Ordering::SeqCst) as u16
+    }
+    fn inc(&self) {
+        self.0.fetch_add(1, Ordering::SeqCst);
+    }
+    fn set(&self, v: u16) {
+        self.0.store(v as usize, Ordering::SeqCst);
+    }
+}

src/io/virtio/vq/splitqueue.rs

@@ -0,0 +1,278 @@
+use std::sync::{Arc, atomic};
+use std::sync::atomic::Ordering;
+use crate::io::virtio::Error;
+use crate::io::virtio::features::ReservedFeatureBit;
+use crate::io::virtio::queues::InterruptLine;
+use crate::io::virtio::vq::chain::DescriptorList;
+use crate::io::virtio::vq::descriptor::Descriptor;
+use crate::io::virtio::vq::SharedIndex;
+use crate::io::virtio::vq::virtqueue::QueueBackend;
+use crate::memory::GuestRam;
+
+
+pub struct SplitQueue {
+    memory: GuestRam,
+    interrupt: Arc<InterruptLine>,
+
+    queue_size: u16,
+    features: u64,
+
+    descriptor_base: u64,
+    avail_base: u64,
+    used_base: u64,
+    /// last seen avail_idx loaded from guest memory
+    cached_avail_idx: SharedIndex,
+    /// The index in the avail ring where the next available entry will be read
+    next_avail: SharedIndex,
+    /// The index in the used ring where the next used entry will be placed
+    next_used_idx: SharedIndex,
+}
+
+impl SplitQueue {
+    pub fn new(memory: GuestRam, interrupt: Arc<InterruptLine>) -> Self {
+        SplitQueue {
+            memory,
+            interrupt,
+            queue_size: 0,
+            features: 0,
+            descriptor_base: 0,
+            avail_base: 0,
+            used_base: 0,
+
+            cached_avail_idx: SharedIndex::new(),
+            next_avail: SharedIndex::new(),
+            next_used_idx: SharedIndex::new(),
+        }
+    }
+
+    ///
+    /// Load the descriptor table entry at `idx` from guest memory and return it.
+    ///
+    fn load_descriptor(&self, idx: u16) -> Option<Descriptor> {
+        if idx >= self.queue_size {
+            panic!("load_descriptor called with index larger than queue size");
+        }
+        let head = self.descriptor_base + (idx as u64 * 16);
+
+        let addr = self.memory.read_int::<u64>(head).unwrap();
+        let len= self.memory.read_int::<u32>(head + 8).unwrap();
+        let flags = self.memory.read_int::<u16>(head + 12).unwrap();
+        let next = self.memory.read_int::<u16>(head + 14).unwrap();
+
+        if self.memory.is_valid_range(addr, len as usize) && next < self.queue_size {
+            return Some(Descriptor::new(addr, len, flags, next));
+        }
+        None
+    }
+
+    fn load_descriptor_lists(&self, head: u16) -> (DescriptorList,DescriptorList) {
+        let mut readable = DescriptorList::new(self.memory.clone());
+        let mut writeable = DescriptorList::new(self.memory.clone());
+        let mut idx = head;
+        let mut ttl = self.queue_size;
+
+        while let Some(d) = self.load_descriptor(idx) {
+            if ttl == 0 {
+                warn!("Descriptor chain length exceeded ttl");
+                break;
+            } else {
+                ttl -= 1;
+            }
+            if d.is_write() {
+                writeable.add_descriptor(d);
+            } else {
+                if !writeable.is_empty() {
+                    warn!("Guest sent readable virtqueue descriptor after writeable descriptor in violation of specification");
+                }
+                readable.add_descriptor(d);
+            }
+            if !d.has_next() {
+                break;
+            }
+            idx = d.next();
+        }
+
+        readable.reverse();
+        writeable.reverse();
+        return (readable, writeable)
+    }
+
+    ///
+    /// Load `avail_ring.idx` from guest memory and store it in `cached_avail_idx`.
+    ///
+    fn load_avail_idx(&self) -> u16 {
+        let avail_idx = self.memory.read_int::<u16>(self.avail_base + 2).unwrap();
+        self.cached_avail_idx.set(avail_idx);
+        avail_idx
+    }
+
+    ///
+    /// Read from guest memory and return the Avail ring entry at
+    /// index `ring_idx % queue_size`.
+    ///
+    fn load_avail_entry(&self, ring_idx: u16) -> u16 {
+        let offset = (4 + (ring_idx % self.queue_size) * 2) as u64;
+        self.memory.read_int(self.avail_base + offset).unwrap()
+    }
+
+    /// Queue is empty if `next_avail` is same value as
+    /// `avail_ring.idx` value in guest memory If `cached_avail_idx`
+    /// currently matches `next_avail` it is reloaded from
+    /// memory in case guest has updated field since last
+    /// time it was loaded.
+    ///
+    fn is_empty(&self) -> bool {
+        let next_avail = self.next_avail.get();
+        if self.cached_avail_idx.get() != next_avail {
+            return false;
+        }
+        next_avail == self.load_avail_idx()
+    }
+
+    ///
+    /// If queue is not empty, read and return the next Avail ring entry
+    /// and increment `next_avail`.  If queue is empty return `None`
+    ///
+    fn pop_avail_entry(&self) -> Option<u16> {
+        if self.is_empty() {
+            return None
+        }
+        let next_avail = self.next_avail.get();
+        let avail_entry = self.load_avail_entry(next_avail);
+        self.next_avail.inc();
+        if self.has_event_idx() {
+            self.write_avail_event(self.next_avail.get());
+        }
+        Some(avail_entry)
+    }
+
+    fn read_avail_flags(&self) -> u16 {
+        self.memory.read_int::<u16>(self.avail_base).unwrap()
+    }
+
+    ///
+    /// Write an entry into the Used ring.
+    ///
+    /// The entry is written into the ring structure at offset
+    /// `next_used_idx % queue_size`.  The value of `next_used_idx`
+    /// is then incremented and the new value is written into
+    /// guest memory into the `used_ring.idx` field.
+    ///
+    fn put_used_entry(&self, idx: u16, len: u32) {
+        if idx >= self.queue_size {
+            return;
+        }
+
+        let used_idx = (self.next_used_idx.get() % self.queue_size) as u64;
+        let elem_addr = self.used_base + (4 + used_idx * 8);
+        // write descriptor index to 'next used' slot in used ring
+        self.memory.write_int(elem_addr, idx as u32).unwrap();
+        // write length to 'next used' slot in ring
+        self.memory.write_int(elem_addr + 4, len as u32).unwrap();
+
+        self.next_used_idx.inc();
+        atomic::fence(Ordering::Release);
+        // write updated next_used
+        self.memory.write_int(self.used_base + 2, self.next_used_idx.get()).unwrap();
+    }
+
+    ///
+    /// Write `val` to the `avail_event` field of Used ring.
+    ///
+    /// If `val` is not a valid index for this virtqueue this
+    /// function does nothing.
+    ///
+    pub fn write_avail_event(&self, val: u16) {
+        if val > self.queue_size {
+            return;
+        }
+        let addr = self.used_base + 4 + (self.queue_size as u64 * 8);
+        self.memory.write_int::<u16>(addr, val).unwrap();
+        atomic::fence(Ordering::Release);
+    }
+
+    fn has_event_idx(&self) -> bool {
+        ReservedFeatureBit::EventIdx.is_set_in(self.features)
+    }
+
+    ///
+    /// Read and return the `used_event` field from the Avail ring
+    fn read_used_event(&self) -> u16 {
+        let addr = self.avail_base + 4 + (self.queue_size as u64  * 2);
+        self.memory.read_int::<u16>(addr).unwrap()
+    }
+
+    fn need_interrupt(&self, first_used: u16) -> bool {
+        if self.has_event_idx() {
+            first_used == self.read_used_event()
+        } else {
+            self.read_avail_flags() & 0x1 == 0
+        }
+    }
+}
+
+impl QueueBackend for SplitQueue {
+    fn configure(&mut self, descriptor_area: u64, driver_area: u64, device_area: u64, size: u16, features: u64) -> crate::io::virtio::Result<()> {
+        let desc_table_sz = 16 * size as usize;
+        let avail_ring_sz = 6 + 2 * size as usize;
+        let used_ring_sz = 6 + 8 * size as usize;
+
+        if !self.memory.is_valid_range(descriptor_area, desc_table_sz) {
+            return Err(Error::RangeInvalid(descriptor_area));
+        }
+        if !self.memory.is_valid_range(driver_area, avail_ring_sz) {
+            return Err(Error::AvailInvalid(driver_area));
+        }
+        if !self.memory.is_valid_range(device_area, used_ring_sz) {
+            return Err(Error::UsedInvalid(device_area));
+        }
+
+        self.descriptor_base = descriptor_area;
+        self.avail_base = driver_area;
+        self.used_base = device_area;
+        self.queue_size = size;
+        self.features = features;
+
+        Ok(())
+    }
+
+    fn reset(&mut self) {
+        self.queue_size = 0;
+        self.features = 0;
+        self.descriptor_base = 0;
+        self.avail_base = 0;
+        self.used_base = 0;
+        self.next_avail.set(0);
+        self.cached_avail_idx.set(0);
+        self.next_used_idx.set(0);
+    }
+
+    /// Queue is empty if `next_avail` is same value as
+    /// `avail_ring.idx` value in guest memory If `cached_avail_idx`
+    /// currently matches `next_avail` it is reloaded from
+    /// memory in case guest has updated field since last
+    /// time it was loaded.
+    ///
+    fn is_empty(&self) -> bool {
+        let next_avail = self.next_avail.get();
+        if self.cached_avail_idx.get() != next_avail {
+            return false;
+        }
+        next_avail == self.load_avail_idx()
+    }
+
+    fn next_descriptors(&self) -> Option<(u16, DescriptorList, DescriptorList)> {
+        self.pop_avail_entry().map(|head| {
+            let (r,w) = self.load_descriptor_lists(head);
+            (head, r, w)
+        })
+    }
+
+    fn put_used(&self, id: u16, size: u32) {
+        let used = self.next_used_idx.get();
+        self.put_used_entry(id, size);
+        if self.need_interrupt(used) {
+            self.interrupt.notify_queue();
+        }
+    }
+}

src/io/virtio/vq/virtqueue.rs

@@ -0,0 +1,195 @@
+use std::sync::{Arc, Mutex, MutexGuard};
+
+use vmm_sys_util::eventfd::EventFd;
+
+use crate::io::virtio::{Error, Result};
+use crate::io::virtio::consts::MAX_QUEUE_SIZE;
+use crate::io::virtio::queues::InterruptLine;
+use crate::io::virtio::vq::chain::{Chain, DescriptorList};
+use crate::io::virtio::vq::splitqueue::SplitQueue;
+use crate::memory::GuestRam;
+
+pub trait QueueBackend: Send {
+
+    fn configure(&mut self, descriptor_area: u64, driver_area: u64, device_area: u64, size: u16, features: u64) -> Result<()>;
+
+    fn reset(&mut self);
+    fn is_empty(&self) -> bool;
+
+
+    fn next_descriptors(&self) -> Option<(u16, DescriptorList,DescriptorList)>;
+    fn put_used(&self, id: u16, size: u32);
+}
+
+#[derive(Clone)]
+pub struct VirtQueue {
+    ioeventfd: Arc<EventFd>,
+
+    /// Default queue_size for this virtqueue
+    default_size: u16,
+
+    /// Number of elements in the virtqueue ring
+    queue_size: u16,
+
+    descriptor_area: u64,
+    driver_area: u64,
+    device_area: u64,
+
+    backend: Arc<Mutex<dyn QueueBackend>>,
+
+    /// Has this virtqueue been enabled?
+    enabled: bool,
+}
+
+impl VirtQueue {
+    pub const DEFAULT_QUEUE_SIZE: u16 = 128;
+
+    pub fn new(memory: GuestRam, default_size: u16, interrupt: Arc<InterruptLine>, ioeventfd: Arc<EventFd>) -> Self {
+        let backend = Arc::new(Mutex::new(SplitQueue::new(memory, interrupt)));
+        VirtQueue {
+            ioeventfd,
+            default_size,
+            queue_size: default_size,
+            descriptor_area: 0,
+            driver_area: 0,
+            device_area: 0,
+            backend,
+            enabled: false,
+        }
+    }
+
+    fn backend(&self) -> MutexGuard<dyn QueueBackend+'static> {
+        self.backend.lock().unwrap()
+    }
+
+    pub fn descriptor_area(&self) -> u64 {
+        self.descriptor_area
+    }
+
+    pub fn set_descriptor_area(&mut self, address: u64) {
+        self.descriptor_area = address;
+    }
+
+    pub fn driver_area(&self) -> u64 {
+        self.driver_area
+    }
+
+    pub fn set_driver_area(&mut self, address: u64) {
+        self.driver_area = address;
+    }
+
+    pub fn device_area(&self) -> u64 {
+        self.device_area
+    }
+
+    pub fn set_device_area(&mut self, address: u64) {
+        self.device_area = address
+    }
+
+    pub fn is_enabled(&self) -> bool {
+        self.enabled
+    }
+
+    pub fn enable(&mut self) {
+        self.enabled = true
+    }
+
+    ///
+    /// Set the queue size of this `VirtQueue`.  If `sz` is an invalid value
+    /// ignore the request.  It is illegal to change the queue size after
+    /// a virtqueue has been enabled, so ignore requests if enabled.
+    ///
+    /// Valid sizes are less than or equal to `MAX_QUEUE_SIZE` and must
+    /// be a power of 2.
+    ///
+    pub fn set_size(&mut self, sz: u16) {
+        if self.is_enabled() || sz > MAX_QUEUE_SIZE || (sz & (sz - 1) != 0) {
+            return;
+        }
+        self.queue_size = sz;
+    }
+
+    pub fn size(&self) -> u16 {
+        self.queue_size
+    }
+
+    ///
+    /// Reset `VirtQueue` to the initial state.  `queue_size` is set to the `default_size`
+    /// and all other fields are cleared.  `enabled` is set to false.
+    ///
+    pub fn reset(&mut self) {
+        self.queue_size = self.default_size;
+        self.descriptor_area = 0;
+        self.driver_area = 0;
+        self.device_area = 0;
+        self.enabled = false;
+        self.backend().reset();
+    }
+
+    pub fn configure(&self, features: u64) -> Result<()> {
+        if !self.enabled {
+            return Err(Error::QueueNotEnabled);
+        }
+        self.backend().configure(self.descriptor_area, self.driver_area, self.device_area, self.size(), features)
+    }
+
+    ///
+    /// Does `VirtQueue` currently have available entries?
+    ///
+    pub fn is_empty(&self) -> bool {
+        self.backend().is_empty()
+    }
+
+    pub fn wait_ready(&self) -> Result<()> {
+        if self.is_empty() {
+            let _ = self.ioeventfd.read()
+                .map_err(Error::ReadIoEventFd)?;
+        }
+        Ok(())
+    }
+
+    pub fn wait_next_chain(&self) -> Result<Chain> {
+        loop {
+            self.wait_ready()?;
+            if let Some(chain) = self.next_chain() {
+                return Ok(chain)
+            }
+        }
+    }
+
+    pub fn next_chain(&self) -> Option<Chain> {
+        self.backend().next_descriptors().map(|(id, r, w)| {
+            Chain::new(self.backend.clone(), id, r, w)
+        })
+    }
+
+    pub fn on_each_chain<F>(&self, mut f: F)
+        where F: FnMut(Chain) {
+        loop {
+            self.wait_ready().unwrap();
+            for chain in self.iter() {
+                f(chain);
+            }
+        }
+    }
+
+    pub fn iter(&self) -> QueueIter {
+        QueueIter { vq: self.clone() }
+    }
+
+    pub fn ioevent(&self) -> &EventFd {
+        &self.ioeventfd
+    }
+}
+
+pub struct QueueIter {
+    vq: VirtQueue
+}
+
+impl Iterator for QueueIter {
+    type Item = Chain;
+
+    fn next(&mut self) -> Option<Self::Item> {
+        self.vq.next_chain()
+    }
+}

src/system/mod.rs

@@ -5,6 +5,7 @@ mod socket;
 mod filedesc;
 mod memfd;
 mod tap;
+//pub mod priority;
 pub mod netlink;
 
 pub use filedesc::{FileDesc, FileFlags};

src/util/mod.rs

@@ -4,5 +4,5 @@ mod buffer;
 mod log;
 
 pub use bitvec::BitSet;
-pub use buffer::ByteBuffer;
+pub use buffer::{ByteBuffer,Writeable};
 pub use log::{Logger,LogLevel};

src/virtio/bus.rs

@@ -1,161 +0,0 @@
-use std::sync::{Arc,RwLock};
-use crate::vm::io::IoDispatcher;
-use crate::memory::{AddressRange, MemoryManager};
-use super::{VirtioDevice,VirtioDeviceOps,PciIrq};
-use super::consts::*;
-use super::pci::PciBus;
-use crate::virtio::Result;
-use std::iter;
-use crate::vm::KvmVm;
-
-
-pub struct VirtioBus {
-    kvm_vm: KvmVm,
-    memory: MemoryManager,
-    io_dispatcher: Arc<IoDispatcher>,
-    pci_bus: Arc<RwLock<PciBus>>,
-    devices: Vec<Arc<RwLock<VirtioDevice>>>,
-}
-
-impl VirtioBus {
-    pub fn new(memory: MemoryManager, io_dispatcher: Arc<IoDispatcher>, kvm_vm: KvmVm) -> VirtioBus {
-        VirtioBus {
-            kvm_vm,
-            memory,
-            io_dispatcher: io_dispatcher.clone(),
-            pci_bus: PciBus::new(&io_dispatcher),
-            devices: Vec::new(),
-        }
-    }
-
-    pub fn new_virtio_device(&mut self, device_type: u16, ops: Arc<RwLock<dyn VirtioDeviceOps>>) -> VirtioDeviceConfig {
-        VirtioDeviceConfig::new(self, device_type, ops)
-    }
-
-    pub fn pci_irqs(&self) -> Vec<PciIrq> {
-        self.pci_bus.read().unwrap().pci_irqs()
-    }
-}
-
-pub struct VirtioDeviceConfig<'a> {
-    virtio_bus: &'a mut VirtioBus,
-    device_type: u16,
-    irq: u8,
-    kvm_vm: KvmVm,
-    ops: Arc<RwLock<dyn VirtioDeviceOps>>,
-    mmio: AddressRange,
-    queue_sizes: Vec<usize>,
-    config_size: usize,
-    device_class: u16,
-    features: u64,
-
-}
-
-impl <'a> VirtioDeviceConfig<'a> {
-    fn new(virtio_bus: &mut VirtioBus, device_type: u16, ops: Arc<RwLock<dyn VirtioDeviceOps>>) -> VirtioDeviceConfig {
-        let kvm_vm = virtio_bus.kvm_vm.clone();
-        let mmio = virtio_bus.pci_bus.write().unwrap().allocate_mmio_space(VIRTIO_MMIO_AREA_SIZE);
-        VirtioDeviceConfig {
-            virtio_bus,
-            device_type,
-            irq: 0,
-            kvm_vm,
-            ops,
-            mmio,
-            queue_sizes: Vec::new(),
-            config_size: 0,
-            features: 0,
-            device_class: 0x0880,
-        }
-    }
-
-    pub fn kvm_vm(&self) -> &KvmVm { &self.kvm_vm }
-
-    pub fn ops(&self) -> Arc<RwLock<dyn VirtioDeviceOps>> {
-        self.ops.clone()
-    }
-    pub fn irq(&self) -> u8 { self.irq }
-
-    pub fn common_cfg_mmio(&self) -> AddressRange {
-        self.mmio.subrange(VIRTIO_MMIO_OFFSET_COMMON_CFG, VIRTIO_MMIO_COMMON_CFG_SIZE).unwrap()
-    }
-
-    pub fn notify_mmio(&self) -> AddressRange {
-        self.mmio.subrange(VIRTIO_MMIO_OFFSET_NOTIFY, VIRTIO_MMIO_NOTIFY_SIZE).unwrap()
-    }
-
-    pub fn isr_mmio(&self) -> AddressRange {
-        self.mmio.subrange(VIRTIO_MMIO_OFFSET_ISR, VIRTIO_MMIO_ISR_SIZE).unwrap()
-    }
-
-    pub fn device_cfg_mmio(&self) -> Option<AddressRange> {
-        if self.config_size > 0 {
-            Some(self.mmio.subrange(VIRTIO_MMIO_OFFSET_DEV_CFG, self.config_size).unwrap())
-        } else {
-            None
-        }
-    }
-
-    pub fn feature_bits(&self) -> u64 {
-        self.features
-    }
-
-    pub fn num_queues(&self) -> usize {
-        self.queue_sizes.len()
-    }
-
-    pub fn queue_sizes(&self) -> &[usize] {
-        &self.queue_sizes
-    }
-
-    #[allow(dead_code)]
-    pub fn config_size(&self) -> usize {
-        self.config_size
-    }
-
-    pub fn set_queue_sizes(&mut self, sizes: &[usize]) -> &'a mut VirtioDeviceConfig {
-        self.queue_sizes.clear();
-        self.queue_sizes.extend_from_slice(sizes);
-        self
-    }
-
-    pub fn set_num_queues(&mut self, n: usize) -> &'a mut VirtioDeviceConfig {
-        self.queue_sizes.clear();
-        self.queue_sizes.extend(iter::repeat(DEFAULT_QUEUE_SIZE as usize).take(n));
-        self
-    }
-
-    pub fn set_config_size(&mut self, sz: usize) -> &'a mut VirtioDeviceConfig {
-        self.config_size = sz;
-        self
-    }
-
-    pub fn set_device_class(&mut self, cl: u16) -> &'a mut VirtioDeviceConfig {
-        self.device_class = cl;
-        self
-    }
-
-    pub fn set_features(&mut self, features: u64) -> &'a mut VirtioDeviceConfig {
-        self.features = features;
-        self
-    }
-
-    pub fn register(&mut self) -> Result<()> {
-        self.create_pci_device();
-        self.features |= VIRTIO_F_VERSION_1;
-        //self.features |= VIRTIO_F_EVENT_IDX;
-        let dev = VirtioDevice::new(self.virtio_bus.memory.clone(), &self)?;
-        self.virtio_bus.io_dispatcher.register_mmio(self.mmio, dev.clone());
-        self.virtio_bus.devices.push(dev);
-        Ok(())
-    }
-
-    fn create_pci_device(&mut self) {
-        let mut pci_bus = self.virtio_bus.pci_bus.write().unwrap();
-        let mut pci = pci_bus.create_device(PCI_VENDOR_ID_REDHAT, PCI_VIRTIO_DEVICE_ID_BASE + self.device_type, self.device_class);
-        pci.add_virtio_caps(self.config_size);
-        pci.set_mmio_bar(VIRTIO_MMIO_BAR, self.mmio);
-        self.irq = pci.get_irq();
-        pci_bus.store_device(pci);
-    }
-}

src/virtio/config.rs

@@ -1,135 +0,0 @@
-use crate::memory::GuestRam;
-use std::sync::Arc;
-use kvm_ioctls::{IoEventAddress, NoDatamatch};
-use vmm_sys_util::eventfd::EventFd;
-
-use super::VirtQueue;
-use super::vring::Vring;
-use super::virtqueue::InterruptLine;
-use super::bus::VirtioDeviceConfig;
-use crate::virtio::{Error, Result};
-
-///
-/// Manages a set of virtqueues during device intitialization.
-///
-pub struct VirtQueueConfig {
-    num_queues: usize,
-    selected_queue: u16,
-    enabled_features: u64,
-    vrings: Vec<Vring>,
-    interrupt: Arc<InterruptLine>,
-    events: Vec<Arc<EventFd>>,
-}
-
-impl VirtQueueConfig {
-    pub fn new(memory: &GuestRam, dev_config: &VirtioDeviceConfig) -> Result<VirtQueueConfig> {
-        Ok(VirtQueueConfig {
-            num_queues: dev_config.num_queues(),
-            selected_queue: 0,
-            enabled_features: 0,
-            vrings: create_vrings(memory,dev_config.queue_sizes()),
-            interrupt: InterruptLine::from_config(&dev_config)?,
-            events: create_ioeventfds(&dev_config)?,
-        })
-    }
-
-    pub fn isr_read(&self) -> u64 {
-        self.interrupt.isr_read()
-    }
-
-    pub fn notify_config(&self) {
-        self.interrupt.notify_config();
-    }
-
-
-    pub fn enable_features(&mut self, features: u64) {
-        self.enabled_features = features;
-    }
-
-    pub fn reset(&mut self) {
-        self.selected_queue = 0;
-        let _ = self.interrupt.isr_read();
-        for vr in &mut self.vrings {
-            vr.reset();
-        }
-    }
-
-    pub fn num_queues(&self) -> u16 {
-        self.num_queues as u16
-    }
-
-    pub fn selected_queue(&self) -> u16 {
-        self.selected_queue
-    }
-
-    pub fn select_queue(&mut self, q: u16) {
-        self.selected_queue = q;
-    }
-
-    pub fn with_vring<U,F>(&self, d: U, f: F) -> U
-        where F: FnOnce(&Vring) -> U
-    {
-        match self.vrings.get(self.selected_queue as usize) {
-            Some(vr) => f(vr),
-            None => d,
-        }
-    }
-
-    pub fn with_vring_mut<F>(&mut self, f: F)
-        where F: FnOnce(&mut Vring)
-    {
-        match self.vrings.get_mut(self.selected_queue as usize) {
-            Some(vr) => if !vr.is_enabled() { f(vr) },
-            None => (),
-        }
-    }
-
-    pub fn vring_get_size(&self) -> u16 { self.with_vring(0, |vr| vr.size() ) }
-    pub fn vring_set_size(&mut self, sz: u16) { self.with_vring_mut(|vr| vr.set_size(sz)) }
-    pub fn vring_enable(&mut self) { self.with_vring_mut(|vr| vr.enable() ) }
-    pub fn vring_is_enabled(&self) -> bool { self.with_vring(false, |vr| vr.is_enabled() ) }
-
-    pub fn notify(&self, vq: u16) {
-        match self.events.get(vq as usize) {
-            Some(ref ev) => ev.write(1).expect("ioeventfd write failed in notify"),
-            None => (),
-        }
-    }
-
-    fn create_vq(&self, memory: &GuestRam, idx: usize) -> Result<VirtQueue> {
-        let vring = self.vrings[idx].clone();
-        vring.validate()?;
-        Ok(VirtQueue::new(memory.clone(), vring, self.interrupt.clone(), self.events[idx].clone()))
-    }
-
-    pub fn create_queues(&self, memory: &GuestRam) -> Result<Vec<VirtQueue>> {
-        let mut v = Vec::with_capacity(self.num_queues);
-        for i in 0..self.num_queues {
-            v.push(self.create_vq(memory, i)?);
-        }
-        Ok(v)
-    }
-}
-
-fn create_ioeventfds(conf: &VirtioDeviceConfig) -> Result<Vec<Arc<EventFd>>> {
-    let mut v = Vec::with_capacity(conf.num_queues());
-    let notify_base = conf.notify_mmio().base();
-
-    for i in 0..conf.num_queues() {
-        let evt = EventFd::new(0)
-            .map_err(Error::CreateEventFd)?;
-        let addr = IoEventAddress::Mmio(notify_base + (4 * i as u64));
-        conf.kvm_vm().vm_fd().register_ioevent(&evt, &addr, NoDatamatch)
-            .map_err(Error::CreateIoEventFd)?;
-        v.push(Arc::new(evt));
-    }
-    Ok(v)
-}
-
-fn create_vrings(memory: &GuestRam, queue_sizes: &[usize]) -> Vec<Vring> {
-    let mut v = Vec::with_capacity(queue_sizes.len());
-    for &sz in queue_sizes {
-        v.push(Vring::new(memory.clone(), sz as u16))
-    }
-    v
-}

src/virtio/consts.rs

@@ -1,120 +0,0 @@
-
-// Maximum number of logical devices on a PCI bus
-
-pub const PCI_MAX_DEVICES: usize = 32;
-
-// IO Port addresses for PCI configuration access
-
-pub const PCI_CONFIG_ADDRESS: u16 = 0xcf8;
-pub const PCI_CONFIG_DATA:    u16 = 0xcfc;
-
-// Vendor specific PCI capabilities
-
-pub const PCI_CAP_ID_VENDOR: u8 = 0x09;
-
-pub const PCI_CONFIG_SPACE_SIZE: usize = 256;
-pub const PCI_CAP_BASE_OFFSET: usize = 0x40;
-
-pub const PCI_VENDOR_ID: usize = 0x00;
-pub const PCI_DEVICE_ID: usize = 0x02;
-pub const PCI_COMMAND: usize = 0x04;
-pub const PCI_COMMAND_IO: u16 = 0x01;
-pub const PCI_COMMAND_MEMORY: u16 = 0x02;
-pub const PCI_COMMAND_INTX_DISABLE: u16 = 0x400;
-pub const PCI_STATUS: usize = 0x06;
-pub const PCI_STATUS_CAP_LIST: u16 = 0x10;
-pub const PCI_CLASS_REVISION: usize = 0x08;
-pub const PCI_CLASS_DEVICE: usize = 0x0a;
-pub const PCI_CACHE_LINE_SIZE: usize = 0x0c;
-pub const PCI_LATENCY_TIMER: usize = 0x0d;
-
-pub const _PCI_SUBSYSTEM_VENDOR_ID: usize = 0x2c;
-pub const PCI_SUBSYSTEM_ID: usize = 0x2e;
-pub const PCI_CAPABILITY_LIST: usize = 0x34;
-pub const PCI_INTERRUPT_LINE: usize = 0x3C;
-pub const PCI_INTERRUPT_PIN: usize = 0x3D;
-
-// Virtio PCI capability types
-
-pub const VIRTIO_PCI_CAP_COMMON_CFG : u8 = 1;
-pub const VIRTIO_PCI_CAP_NOTIFY_CFG : u8 = 2;
-pub const VIRTIO_PCI_CAP_ISR_CFG    : u8 = 3;
-pub const VIRTIO_PCI_CAP_DEVICE_CFG : u8 = 4;
-
-// Indicates that no MSIX vector is configured
-
-pub const VIRTIO_NO_MSI_VECTOR: u16 = 0xFFFF;
-
-// Bar number 0 is used for Virtio MMIO area
-
-pub const VIRTIO_MMIO_BAR: usize = 0;
-
-// Virtio MMIO area is one page
-
-pub const VIRTIO_MMIO_AREA_SIZE: usize = 4096;
-
-// Offsets and sizes for each structure in MMIO area
-
-pub const VIRTIO_MMIO_OFFSET_COMMON_CFG : usize = 0;     // Common configuration offset
-pub const VIRTIO_MMIO_OFFSET_ISR        : usize = 56;    // ISR register offset
-pub const VIRTIO_MMIO_OFFSET_NOTIFY     : usize = 0x400; // Notify area offset
-pub const VIRTIO_MMIO_OFFSET_DEV_CFG    : usize = 0x800; // Device specific configuration offset
-
-pub const VIRTIO_MMIO_COMMON_CFG_SIZE: usize = 56;       // Common configuration size
-pub const VIRTIO_MMIO_NOTIFY_SIZE    : usize = 0x400;    // Notify area size
-pub const VIRTIO_MMIO_ISR_SIZE       : usize = 4;        // ISR register size
-
-// Common configuration header offsets
-
-pub const VIRTIO_PCI_COMMON_DFSELECT      : usize = 0;
-pub const VIRTIO_PCI_COMMON_DF            : usize = 4;
-pub const VIRTIO_PCI_COMMON_GFSELECT      : usize = 8;
-pub const VIRTIO_PCI_COMMON_GF            : usize = 12;
-pub const VIRTIO_PCI_COMMON_MSIX          : usize = 16;
-pub const VIRTIO_PCI_COMMON_NUMQ          : usize = 18;
-pub const VIRTIO_PCI_COMMON_STATUS        : usize = 20;
-pub const VIRTIO_PCI_COMMON_CFGGENERATION : usize = 21;
-pub const VIRTIO_PCI_COMMON_Q_SELECT      : usize = 22;
-pub const VIRTIO_PCI_COMMON_Q_SIZE        : usize = 24;
-pub const VIRTIO_PCI_COMMON_Q_MSIX        : usize = 26;
-pub const VIRTIO_PCI_COMMON_Q_ENABLE      : usize = 28;
-pub const VIRTIO_PCI_COMMON_Q_NOFF        : usize = 30;
-pub const VIRTIO_PCI_COMMON_Q_DESCLO      : usize = 32;
-pub const VIRTIO_PCI_COMMON_Q_DESCHI      : usize = 36;
-pub const VIRTIO_PCI_COMMON_Q_AVAILLO     : usize = 40;
-pub const VIRTIO_PCI_COMMON_Q_AVAILHI     : usize = 44;
-pub const VIRTIO_PCI_COMMON_Q_USEDLO      : usize = 48;
-pub const VIRTIO_PCI_COMMON_Q_USEDHI      : usize = 52;
-
-// Common configuration status bits
-
-pub const _VIRTIO_CONFIG_S_ACKNOWLEDGE : u8 = 1;
-pub const _VIRTIO_CONFIG_S_DRIVER      : u8 = 2;
-pub const VIRTIO_CONFIG_S_DRIVER_OK   : u8 = 4;
-pub const VIRTIO_CONFIG_S_FEATURES_OK : u8 = 8;
-pub const VIRTIO_CONFIG_S_NEEDS_RESET : u8 = 0x40;
-pub const _VIRTIO_CONFIG_S_FAILED      : u8 = 0x80;
-
-pub const _VRING_USED_F_NO_NOTIFY: u16 = 1;
-pub const _VRING_AVAIL_F_NO_INTERRUPT: u16 = 1;
-pub const _VIRTIO_F_INDIRECT_DESC: u64 = 1 << 28;
-pub const VIRTIO_F_EVENT_IDX: u64 = 1 << 29;
-pub const VIRTIO_F_VERSION_1: u64 = 1 << 32;
-
-pub const VRING_DESC_F_NEXT: u16     = 1;
-pub const VRING_DESC_F_WRITE: u16    = 2;
-pub const VRING_DESC_F_INDIRECT: u16 = 4;
-
-pub const DEFAULT_QUEUE_SIZE: u16 = 128;
-pub const MAX_QUEUE_SIZE: u16 = 1024;
-
-// PCI Vendor id for Virtio devices
-
-pub const PCI_VENDOR_ID_REDHAT: u16 = 0x1af4;
-
-// Base PCI device id for Virtio devices
-
-pub const PCI_VIRTIO_DEVICE_ID_BASE: u16 = 0x1040;
-
-pub const PCI_VENDOR_ID_INTEL: u16 = 0x8086;
-pub const PCI_CLASS_BRIDGE_HOST: u16 = 0x0600;

src/virtio/device.rs

@@ -1,228 +0,0 @@
-use std::sync::{Arc,RwLock};
-use std::ops::DerefMut;
-
-use crate::memory::{AddressRange, MemoryManager};
-use super::bus::VirtioDeviceConfig;
-use super::VirtQueue;
-use super::config::VirtQueueConfig;
-use super::consts::*;
-use crate::vm::io::MmioOps;
-use crate::virtio::Result;
-
-pub trait VirtioDeviceOps: Send+Sync {
-    fn reset(&mut self) {}
-    fn enable_features(&mut self, bits: u64) -> bool { let _ = bits; true }
-    fn write_config(&mut self, offset: usize, size: usize, val: u64) { let (_,_,_) = (offset, size, val); }
-    fn read_config(&mut self, offset: usize, size: usize) -> u64 { let (_,_) = (offset, size); 0 }
-    fn start(&mut self, memory: &MemoryManager, queues: Vec<VirtQueue>);
-}
-
-pub struct VirtioDevice {
-    memory: MemoryManager,
-    vq_config: VirtQueueConfig,
-    common_cfg_mmio: AddressRange,
-    isr_mmio: AddressRange,
-    notify_mmio: AddressRange,
-    device_cfg_mmio: Option<AddressRange>,
-    device_ops: Arc<RwLock<dyn VirtioDeviceOps>>,
-    dfselect: u32,
-    gfselect: u32,
-    device_features: u64,
-    guest_features: u64,
-    status: u8,
-}
-
-const MASK_LOW_32: u64 = (1u64 << 32) - 1;
-const MASK_HI_32: u64 = MASK_LOW_32 << 32;
-
-fn set_lo32(val: &mut u64, low32: u32)  { *val = (*val & MASK_HI_32) | (low32 as u64) }
-fn set_hi32(val: &mut u64, hi32: u32)  { *val = ((hi32 as u64) << 32) | (*val & MASK_LOW_32) }
-fn get_lo32(val: u64) -> u32 { val as u32 }
-fn get_hi32(val: u64) -> u32 { (val >> 32) as u32 }
-
-
-
-impl VirtioDevice {
-    pub fn new(memory: MemoryManager, config: &VirtioDeviceConfig) -> Result<Arc<RwLock<VirtioDevice>>> {
-        Ok(Arc::new(RwLock::new(VirtioDevice {
-            memory: memory.clone(),
-            vq_config: VirtQueueConfig::new(memory.guest_ram(),&config)?,
-            common_cfg_mmio: config.common_cfg_mmio(),
-            isr_mmio: config.isr_mmio(),
-            notify_mmio: config.notify_mmio(),
-            device_cfg_mmio: config.device_cfg_mmio(),
-
-            device_ops: config.ops(),
-            dfselect: 0,
-            gfselect: 0,
-
-            device_features: config.feature_bits(),
-            guest_features: 0,
-            status: 0,
-        })))
-    }
-
-    fn reset(&mut self) {
-        self.dfselect = 0;
-        self.gfselect = 0;
-        self.guest_features = 0;
-        self.status = 0;
-        self.vq_config.reset();
-    }
-
-    fn status_write(&mut self, val: u8) {
-
-        // 4.1.4.3.1 The device MUST reset when 0 is written to device status
-        if val == 0 {
-            self.reset();
-            return;
-        }
-        // 2.1.1 The driver MUST NOT clear a device status bit
-        if self.status & !val != 0 {
-            return;
-        }
-
-        let new_bits = val & !self.status;
-
-        if new_bits & VIRTIO_CONFIG_S_DRIVER_OK != 0 {
-            match self.vq_config.create_queues(self.memory.guest_ram()) {
-                Ok(queues) => self.with_ops(|ops| ops.start(&self.memory, queues)),
-                Err(e) => {
-                    println!("creating virtqueues failed {}", e);
-                    self.status |= VIRTIO_CONFIG_S_NEEDS_RESET;
-                    self.vq_config.notify_config();
-                    return;
-                }
-            }
-        }
-
-        if new_bits & VIRTIO_CONFIG_S_FEATURES_OK != 0 {
-            if !self.with_ops(|ops| ops.enable_features(self.guest_features)) {
-                self.vq_config.enable_features(self.guest_features);
-               return;
-            }
-        }
-
-        self.status |= new_bits;
-    }
-
-    fn common_config_write(&mut self, offset: usize, _size: usize, val: u32) {
-        match offset {
-            VIRTIO_PCI_COMMON_DFSELECT => self.dfselect = val,
-            VIRTIO_PCI_COMMON_GFSELECT => self.gfselect = val,
-            VIRTIO_PCI_COMMON_GF => {
-                match self.gfselect {
-                    0 => set_lo32(&mut self.guest_features, val),
-                    1 => set_hi32(&mut self.guest_features, val),
-                    _ => {},
-                }
-                // 2.2.1
-                //   The driver MUST NOT accept a feature which the device did
-                //   not offer.
-                self.guest_features &= self.device_features;
-            },
-            VIRTIO_PCI_COMMON_STATUS => self.status_write(val as u8),
-            VIRTIO_PCI_COMMON_Q_SELECT=> self.vq_config.select_queue(val as u16),
-            VIRTIO_PCI_COMMON_Q_SIZE => self.vq_config.vring_set_size(val as u16),
-            VIRTIO_PCI_COMMON_Q_ENABLE=> if val == 1 { self.vq_config.vring_enable() } ,
-            VIRTIO_PCI_COMMON_Q_DESCLO=> self.vq_config.with_vring_mut(|vr| set_lo32(&mut vr.descriptors, val)),
-            VIRTIO_PCI_COMMON_Q_DESCHI=> self.vq_config.with_vring_mut(|vr| set_hi32(&mut vr.descriptors, val)),
-            VIRTIO_PCI_COMMON_Q_AVAILLO=> self.vq_config.with_vring_mut(|vr| set_lo32(&mut vr.avail_ring, val)),
-            VIRTIO_PCI_COMMON_Q_AVAILHI=> self.vq_config.with_vring_mut(|vr| set_hi32(&mut vr.avail_ring, val)),
-            VIRTIO_PCI_COMMON_Q_USEDLO=> self.vq_config.with_vring_mut(|vr| set_lo32(&mut vr.used_ring, val)),
-            VIRTIO_PCI_COMMON_Q_USEDHI=> self.vq_config.with_vring_mut(|vr| set_hi32(&mut vr.used_ring, val)),
-            _ => {},
-        }
-    }
-
-    fn common_config_read(&mut self, offset: usize, _size: usize) -> u32 {
-        match offset {
-            VIRTIO_PCI_COMMON_DFSELECT => self.dfselect,
-            VIRTIO_PCI_COMMON_DF=> match self.dfselect {
-                0 => get_lo32(self.device_features),
-                1 => get_hi32(self.device_features),
-                _ => 0,
-            },
-            VIRTIO_PCI_COMMON_GFSELECT => { self.gfselect },
-            VIRTIO_PCI_COMMON_GF => match self.gfselect {
-                0 => get_lo32(self.guest_features),
-                1 => get_hi32(self.guest_features),
-                _ => 0,
-            },
-            VIRTIO_PCI_COMMON_MSIX => VIRTIO_NO_MSI_VECTOR as u32,
-            VIRTIO_PCI_COMMON_NUMQ => self.vq_config.num_queues() as u32,
-            VIRTIO_PCI_COMMON_STATUS => self.status as u32,
-            VIRTIO_PCI_COMMON_CFGGENERATION => 0,
-            VIRTIO_PCI_COMMON_Q_SELECT => self.vq_config.selected_queue() as u32,
-            VIRTIO_PCI_COMMON_Q_SIZE => self.vq_config.vring_get_size() as u32,
-            VIRTIO_PCI_COMMON_Q_MSIX => VIRTIO_NO_MSI_VECTOR as u32,
-            VIRTIO_PCI_COMMON_Q_ENABLE => if self.vq_config.vring_is_enabled() {1} else {0},
-            VIRTIO_PCI_COMMON_Q_NOFF => self.vq_config.selected_queue() as u32,
-            VIRTIO_PCI_COMMON_Q_DESCLO => self.vq_config.with_vring(0, |vr| get_lo32(vr.descriptors)),
-            VIRTIO_PCI_COMMON_Q_DESCHI => self.vq_config.with_vring(0, |vr| get_hi32(vr.descriptors)),
-            VIRTIO_PCI_COMMON_Q_AVAILLO => self.vq_config.with_vring(0, |vr| get_lo32(vr.avail_ring)),
-            VIRTIO_PCI_COMMON_Q_AVAILHI => self.vq_config.with_vring(0, |vr| get_hi32(vr.avail_ring)),
-            VIRTIO_PCI_COMMON_Q_USEDLO => self.vq_config.with_vring(0, |vr| get_lo32(vr.used_ring)),
-            VIRTIO_PCI_COMMON_Q_USEDHI => self.vq_config.with_vring(0, |vr| get_hi32(vr.used_ring)),
-            _ => 0,
-        }
-    }
-
-    fn notify_read(&mut self, _offset: usize, _size: usize) -> u64 {
-        0
-    }
-
-    fn notify_write(&mut self, offset: usize, _size: usize, _val: u64) {
-        let vq = (offset / 4) as u16;
-        self.vq_config.notify(vq);
-    }
-
-    fn isr_read(&mut self) -> u64 {
-        self.vq_config.isr_read()
-    }
-
-    fn with_ops<U,F>(&self, f: F) -> U
-      where F: FnOnce(&mut dyn VirtioDeviceOps) -> U {
-        let mut ops = self.device_ops.write().unwrap();
-        f(ops.deref_mut())
-    }
-}
-
-impl MmioOps for VirtioDevice {
-    fn mmio_read(&mut self, address: u64, size: usize) -> u64 {
-        if self.common_cfg_mmio.contains(address, size) {
-            let offset = self.common_cfg_mmio.offset_of(address);
-            self.common_config_read(offset,size) as u64
-
-        } else if self.notify_mmio.contains(address, size) {
-            let offset = self.notify_mmio.offset_of(address);
-            self.notify_read(offset, size) as u64
-
-        } else if self.isr_mmio.contains(address, size) {
-            self.isr_read()
-
-        } else if let Some(ref dev_cfg_mmio) = self.device_cfg_mmio {
-            let offset = dev_cfg_mmio.offset_of(address);
-            self.with_ops(|ops| ops.read_config(offset, size))
-
-        } else {
-            0
-        }
-    }
-
-    fn mmio_write(&mut self, address: u64, size: usize, val: u64) {
-        if self.common_cfg_mmio.contains(address, size) {
-            let offset = self.common_cfg_mmio.offset_of(address);
-            self.common_config_write(offset,size, val as u32)
-
-        } else if self.notify_mmio.contains(address, size) {
-            let offset = self.notify_mmio.offset_of(address);
-            self.notify_write(offset, size, val)
-
-        } else if let Some(ref dev_cfg_mmio) = self.device_cfg_mmio {
-            let offset = dev_cfg_mmio.offset_of(address);
-            self.with_ops(|ops| ops.write_config(offset, size, val))
-        }
-    }
-}
-

src/virtio/device_config.rs

@@ -1,98 +0,0 @@
-
-use byteorder::{ByteOrder,LittleEndian};
-use std::ops::Range;
-
-pub struct DeviceConfigArea {
-    buffer: Vec<u8>,
-    write_filter: DeviceConfigWriteFilter,
-}
-
-
-#[allow(dead_code)]
-impl DeviceConfigArea {
-    pub fn new(size: usize) -> Self {
-        DeviceConfigArea{
-            buffer: vec![0u8; size],
-            write_filter: DeviceConfigWriteFilter::new(size),
-        }
-    }
-
-    pub fn read_config(&self, offset: usize, size: usize) -> u64 {
-        if offset + size > self.buffer.len() {
-            return 0;
-        }
-        match size {
-            1 => self.buffer[offset] as u64,
-            2 => LittleEndian::read_u16(&self.buffer[offset..]) as u64,
-            4 => LittleEndian::read_u32(&self.buffer[offset..]) as u64,
-            8 => LittleEndian::read_u64(&self.buffer[offset..]),
-            _ => 0,
-        }
-    }
-
-    pub fn write_config(&mut self, offset: usize, size: usize, val: u64) {
-        if self.write_filter.is_writeable(offset, size) {
-            match size {
-                1 => self.write_u8(offset, val as u8),
-                2 => self.write_u16(offset, val as u16),
-                4 => self.write_u32(offset, val as u32),
-                8 => self.write_u64(offset, val as u64),
-                _ => {},
-            }
-        }
-    }
-
-    pub fn set_writeable(&mut self, offset: usize, size: usize) {
-        self.write_filter.set_writable(offset, size)
-    }
-
-    pub fn write_u8(&mut self, offset: usize, val: u8) {
-        assert!(offset + 1 <= self.buffer.len());
-        self.buffer[offset] = val;
-    }
-
-    pub fn write_u16(&mut self, offset: usize, val: u16) {
-        assert!(offset + 2 <= self.buffer.len());
-        LittleEndian::write_u16(&mut self.buffer[offset..], val);
-    }
-
-    pub fn write_u32(&mut self, offset: usize, val: u32) {
-        assert!(offset + 4 <= self.buffer.len());
-        LittleEndian::write_u32(&mut self.buffer[offset..], val);
-    }
-
-    pub fn write_u64(&mut self, offset: usize, val: u64) {
-        assert!(offset + 8 <= self.buffer.len());
-        LittleEndian::write_u64(&mut self.buffer[offset..], val);
-    }
-
-    pub fn write_bytes(&mut self, offset: usize, bytes: &[u8]) {
-        assert!(offset + bytes.len() <= self.buffer.len());
-        self.buffer[offset..offset + bytes.len()].copy_from_slice(bytes);
-    }
-}
-
-struct DeviceConfigWriteFilter {
-    size: usize,
-    ranges: Vec<Range<usize>>,
-}
-
-impl DeviceConfigWriteFilter {
-    fn new(size: usize) -> Self {
-        DeviceConfigWriteFilter { size, ranges: Vec::new() }
-    }
-
-    fn set_writable(&mut self, offset: usize, size: usize) {
-        let end = offset + size;
-        self.ranges.push(offset..end);
-    }
-
-    fn is_writeable(&self, offset: usize, size: usize) -> bool {
-        if offset + size > self.size {
-            false
-        } else {
-            let last = offset + size - 1;
-            self.ranges.iter().any(|r| r.contains(&offset) && r.contains(&last))
-        }
-    }
-}

src/virtio/mod.rs

@@ -1,56 +0,0 @@
-mod bus;
-mod chain;
-mod config;
-mod consts;
-mod device;
-mod pci;
-mod virtqueue;
-mod vring;
-mod device_config;
-
-pub use self::virtqueue::VirtQueue;
-pub use self::pci::PciIrq;
-pub use self::bus::VirtioBus;
-pub use self::device::{VirtioDevice,VirtioDeviceOps};
-pub use self::chain::Chain;
-pub use self::device_config::DeviceConfigArea;
-
-use byteorder::{ByteOrder,LittleEndian};
-use std::result;
-
-use thiserror::Error;
-
-pub type Result<T> = result::Result<T, Error>;
-
-#[derive(Debug,Error)]
-pub enum Error {
-    #[error("failed to create EventFd for VirtQueue: {0}")]
-    CreateEventFd(std::io::Error),
-    #[error("failed to create IoEventFd for VirtQueue: {0}")]
-    CreateIoEventFd(kvm_ioctls::Error),
-    #[error("failed to read from IoEventFd: {0}")]
-    ReadIoEventFd(std::io::Error),
-    #[error("VirtQueue: {0}")]
-    IrqFd(kvm_ioctls::Error),
-    #[error("vring not enabled")]
-    VringNotEnabled,
-    #[error("vring descriptor table range is invalid 0x{0:x}")]
-    VringRangeInvalid(u64),
-    #[error("vring avail ring range range is invalid 0x{0:x}")]
-    VringAvailInvalid(u64),
-    #[error("vring used ring range is invalid 0x{0:x}")]
-    VringUsedInvalid(u64),
-}
-
-pub fn read_config_buffer(config: &[u8], offset: usize, size: usize) -> u64 {
-    if offset + size > config.len() {
-        return 0;
-    }
-    match size {
-        1 => config[offset] as u64,
-        2 => LittleEndian::read_u16(&config[offset..]) as u64,
-        4 => LittleEndian::read_u32(&config[offset..]) as u64,
-        8 => LittleEndian::read_u64(&config[offset..]) as u64,
-        _ => 0,
-    }
-}

src/virtio/pci.rs

@@ -1,436 +0,0 @@
-use std::sync::{Arc,RwLock};
-use byteorder::{ByteOrder,LittleEndian};
-
-use crate::vm::io::{IoDispatcher,IoPortOps};
-use crate::vm::arch::PCI_MMIO_RESERVED_BASE;
-use crate::memory::AddressRange;
-use super::consts::*;
-
-struct PciConfigAddress(u32);
-
-impl PciConfigAddress {
-    fn new() -> PciConfigAddress { PciConfigAddress(0) }
-    fn set(&mut self, n: u32) { self.0 = n }
-    fn get(&self) -> u32 { self.0 }
-    fn bus(&self) -> u32 { self.bits(16, 8) }
-    fn function(&self) -> u32 { self.bits(8, 3) }
-    fn device(&self) -> usize { self.bits(11, 5) as usize }
-    fn offset(&self) -> usize { (self.bits(0, 8) & !0x3) as usize }
-    fn bits(&self, offset: u32, size: u32) -> u32 {
-        let mask = (1u32 << size) - 1;
-        (self.0 >> offset) & mask
-    }
-}
-
-pub struct PciIrq {
-    pci_id: u8,
-    int_pin: u8,
-    irq: u8,
-}
-
-impl PciIrq {
-    fn new(pci: &PciDevice) -> PciIrq {
-        PciIrq {
-            pci_id: pci.id,
-            int_pin: 1,
-            irq: pci.irq,
-        }
-    }
-
-    pub fn src_bus_irq(&self) -> u8 {
-        (self.pci_id << 2) | (self.int_pin - 1)
-    }
-
-    pub fn irq_line(&self) -> u8 {
-        self.irq
-    }
-}
-
-pub struct PciBus {
-    devices: Vec<Option<PciDevice>>,
-    mmio_next_alloc: u32,
-    next_irq: u8,
-    next_dev: u8,
-    config_address: PciConfigAddress,
-}
-
-impl PciBus {
-    pub fn new(io: &IoDispatcher) -> Arc<RwLock<PciBus>> {
-        let bus = Arc::new(RwLock::new(PciBus {
-            devices: PciBus::create_device_vec(PCI_MAX_DEVICES),
-            mmio_next_alloc: PCI_MMIO_RESERVED_BASE as u32,
-            next_irq: 5,
-            next_dev: 1,
-            config_address: PciConfigAddress::new(),
-        }));
-
-        io.register_ioports(PCI_CONFIG_ADDRESS, 8, bus.clone());
-        let pci = PciDevice::new(0, 0, PCI_VENDOR_ID_INTEL, 0, PCI_CLASS_BRIDGE_HOST);
-        bus.write().unwrap().store_device(pci);
-        bus
-    }
-
-    pub fn pci_irqs(&self) -> Vec<PciIrq> {
-        let mut v = Vec::new();
-        for d in &self.devices {
-            match *d {
-                Some(ref dev) => v.push(PciIrq::new(dev)),
-                None => (),
-            }
-        }
-        v
-    }
-
-    fn allocate_irq(&mut self) -> u8 {
-        let irq = self.next_irq;
-        self.next_irq += 1;
-        irq
-    }
-
-    fn allocate_id(&mut self) -> u8 {
-        let id = self.next_dev;
-        self.next_dev += 1;
-        id
-    }
-
-    pub fn create_device(&mut self, vendor: u16, device: u16, class_id: u16) -> PciDevice {
-        let irq = self.allocate_irq();
-        let id = self.allocate_id();
-        let pci = PciDevice::new(id, irq, vendor, device, class_id);
-        pci
-    }
-
-    pub fn store_device(&mut self, pci: PciDevice) {
-        let id = pci.id as usize;
-        self.devices[id] = Some(pci)
-    }
-
-    fn create_device_vec(sz: usize) -> Vec<Option<PciDevice>> {
-        let mut v = Vec::with_capacity(sz);
-        for _ in 0..sz {
-            v.push(None)
-        }
-        v
-    }
-
-    pub fn allocate_mmio_space(&mut self, sz: usize) -> AddressRange {
-        let mask = (sz - 1) as u32;
-        let aligned = (self.mmio_next_alloc + mask) & !mask;
-        self.mmio_next_alloc = aligned + (sz as u32);
-        AddressRange::new(aligned as u64, sz)
-    }
-
-    fn is_in_range(base: u16, port: u16, len: usize) -> bool {
-        let end = port + len as u16;
-        port >= base && end <= (base + 4)
-    }
-
-    fn is_config_address(&self, port: u16, len: usize) -> bool {
-        PciBus::is_in_range(PCI_CONFIG_ADDRESS, port, len)
-    }
-
-    fn is_config_data(&self, port: u16, len: usize) -> bool {
-        PciBus::is_in_range(PCI_CONFIG_DATA, port, len)
-    }
-
-    fn config_address_in(&self, _: usize) -> u32 {
-        self.config_address.get()
-    }
-
-    fn current_config_device(&mut self) -> Option<&mut PciDevice> {
-        let b = self.config_address.bus();
-        let d = self.config_address.device();
-        let f = self.config_address.function();
-
-        if b != 0 || f != 0 || d >= self.devices.len() {
-            return None;
-        }
-
-        self.devices[d].as_mut()
-    }
-
-    fn config_address_out(&mut self, _offset: u16, size: usize, data: u32) {
-        if size == 4 {
-            self.config_address.set(data);
-        }
-    }
-
-    #[allow(dead_code)]
-    fn valid_config_access(&self, offset: u16, len: usize) -> bool {
-        (offset as usize) + len <= 4
-    }
-
-    fn config_data_in(&mut self, offset: usize, size: usize) -> u32 {
-        let off = self.config_address.offset() + offset;
-        match self.current_config_device() {
-            Some(dev) => { dev.read_config(off, size)},
-            None => 0xFFFFFFFF,
-        }
-    }
-
-    fn config_data_out(&mut self, offset: u16, size: usize, data: u32) {
-        let off = self.config_address.offset() + offset as usize;
-        if let Some(dev) = self.current_config_device() {
-            dev.write_config(off, size,data)
-        }
-    }
-}
-
-impl IoPortOps for PciBus {
-    fn io_in(&mut self, port: u16, size: usize) -> u32 {
-        if self.is_config_address(port, size) {
-            return self.config_address_in(size)
-        }
-        if self.is_config_data(port, size) {
-            return self.config_data_in((port - PCI_CONFIG_DATA) as usize, size)
-        }
-        return 0;
-    }
-
-    fn io_out(&mut self, port: u16, size: usize, val: u32) {
-        if self.is_config_address(port, size) {
-            self.config_address_out(port - PCI_CONFIG_ADDRESS,size,  val)
-        }
-        if self.is_config_data(port, size) {
-            self.config_data_out(port - PCI_CONFIG_DATA, size, val)
-        }
-    }
-}
-
-
-pub struct PciDevice {
-    next_cap: usize,
-    last_cap: usize,
-    id: u8,
-    irq: u8,
-    config_buffer: [u8; PCI_CONFIG_SPACE_SIZE],
-    bar_write_masks: [u32; 6],
-}
-
-impl PciDevice {
-    pub fn new(id: u8, irq: u8, vendor: u16, device: u16, class_id: u16) -> PciDevice {
-        let mut d = PciDevice {
-            next_cap: PCI_CAP_BASE_OFFSET,
-            last_cap: 0,
-            id,
-            irq,
-            config_buffer: [0; PCI_CONFIG_SPACE_SIZE],
-            bar_write_masks: [0; 6],
-        };
-        d.w16(PCI_VENDOR_ID, vendor);
-        d.w16(PCI_DEVICE_ID, device);
-        d.w16(PCI_COMMAND, PCI_COMMAND_IO | PCI_COMMAND_MEMORY);
-        d.w8(PCI_CLASS_REVISION, 1);
-        d.w16(PCI_CLASS_DEVICE, class_id);
-        d.w8(PCI_INTERRUPT_PIN, 1);
-        d.w8(PCI_INTERRUPT_LINE, irq);
-        d.w16(PCI_SUBSYSTEM_ID, 0x40);
-        d
-    }
-
-    pub fn get_irq(&self) -> u8 {
-        self.irq
-    }
-
-    fn is_valid_write(&self, offset: usize, size: usize) -> bool {
-        if offset + size > PCI_CONFIG_SPACE_SIZE {
-            return false;
-        }
-        // check alignment of write
-        let mod4 = offset % 4;
-        match size {
-            4 if mod4 == 0 => true,
-            2 if mod4 == 0 || mod4 == 2 => true,
-            1 => true,
-            _ => false,
-        }
-    }
-
-    fn write_bar(&mut self, offset: usize, size: usize, data: u32) {
-        assert!(is_bar_offset(offset), "not a bar offset in write_bar()");
-
-        let bar = offset_to_bar(offset);
-        let write_mask = self.bar_write_masks[bar];
-
-        if write_mask == 0 {
-            // no writable bits
-            return;
-        }
-
-        let mod4 = offset % 4;
-
-        match size {
-            4 => self.w32(offset, data),
-            2 => self.w16(offset+ mod4, data as u16),
-            1 => self.w8(offset+ mod4, data as u8),
-            _ => (),
-        };
-
-        // apply write mask to whatever was written
-        let v = self.r32(offset);
-        self.w32(offset, v & write_mask);
-    }
-
-    fn write_config(&mut self, offset: usize, size: usize, data: u32) {
-        if !self.is_valid_write(offset, size) {
-            return;
-        }
-
-        if is_bar_offset(offset) {
-            self.write_bar(offset, size, data);
-            return;
-        }
-
-        match offset {
-            PCI_COMMAND if size == 2 => self.w16(PCI_COMMAND, data as u16),
-            PCI_STATUS if size == 2 => self.w16(PCI_STATUS, data as u16),
-            PCI_CACHE_LINE_SIZE if size == 1 => self.w8(PCI_CACHE_LINE_SIZE, data as u8),
-            PCI_LATENCY_TIMER if size == 1 => self.w8(PCI_LATENCY_TIMER, data as u8),
-            _ => (),
-        }
-    }
-
-    fn w32(&mut self, off: usize, val: u32) { LittleEndian::write_u32(&mut self.config_buffer[off..], val); }
-    fn w16(&mut self, off: usize, val: u16) { LittleEndian::write_u16(&mut self.config_buffer[off..], val); }
-    fn w8(&mut self, off: usize, val: u8) { self.config_buffer[off] = val; }
-
-    fn r32(&self, off: usize) -> u32 { LittleEndian::read_u32(&self.config_buffer[off..]) }
-    fn r16(&self, off: usize) -> u16 { LittleEndian::read_u16(&self.config_buffer[off..]) }
-    fn r8(&self, off: usize) -> u8 { self.config_buffer[off] }
-
-    fn read_config(&self, offset: usize, size: usize) -> u32 {
-        if offset + size > PCI_CONFIG_SPACE_SIZE {
-            return 0xFFFFFFFF;
-        }
-        match size {
-            1 => self.r8(offset) as u32,
-            2 => self.r16(offset) as u32,
-            4 => self.r32(offset),
-            _ => 0xFFFFFFFF
-        }
-    }
-
-    #[allow(dead_code)]
-    pub fn is_irq_disabled(&self) -> bool {
-        self.r16(PCI_COMMAND) & PCI_COMMAND_INTX_DISABLE != 0
-    }
-
-    pub fn set_mmio_bar(&mut self, bar: usize, range: AddressRange) {
-        assert!(range.is_naturally_aligned(), "cannot set_mmio_bar() because mmio range is not naturally aligned");
-        assert!(bar < 5, "bar is invalid value in set_mmio_bar()");
-        self.bar_write_masks[bar] = !((range.size() as u32) - 1);
-        self.w32(bar_to_offset(bar), range.base() as u32);
-    }
-
-    pub fn add_virtio_caps(&mut self, config_size: usize) {
-        self.new_virtio_cap(VIRTIO_PCI_CAP_COMMON_CFG)
-            .set_mmio_range(VIRTIO_MMIO_OFFSET_COMMON_CFG, VIRTIO_MMIO_COMMON_CFG_SIZE).add(self);
-
-        self.new_virtio_cap(VIRTIO_PCI_CAP_ISR_CFG)
-            .set_mmio_range(VIRTIO_MMIO_OFFSET_ISR, VIRTIO_MMIO_ISR_SIZE).add(self);
-
-        self.new_virtio_cap(VIRTIO_PCI_CAP_NOTIFY_CFG)
-            .set_mmio_range(VIRTIO_MMIO_OFFSET_NOTIFY, VIRTIO_MMIO_NOTIFY_SIZE)
-            .set_extra_word(4).add(self);
-
-        if config_size > 0 {
-            self.new_virtio_cap(VIRTIO_PCI_CAP_DEVICE_CFG)
-                .set_mmio_range(VIRTIO_MMIO_OFFSET_DEV_CFG,config_size).add(self);
-        }
-    }
-
-    pub fn new_virtio_cap(&mut self, vtype: u8) -> VirtioCap {
-        VirtioCap::new(self.next_cap, vtype)
-    }
-
-    fn inc_cap(&mut self, size: usize) {
-        let next = self.next_cap as u8;
-        let last = self.last_cap;
-        if self.last_cap == 0 {
-            self.w8(PCI_CAPABILITY_LIST, next);
-            let status = self.r16(PCI_STATUS) | PCI_STATUS_CAP_LIST;
-            self.w16(PCI_STATUS, status);
-        } else {
-            self.w8(last + 1, next);
-        }
-        self.last_cap = self.next_cap;
-        let aligned = (size + 3) & !3;
-        self.next_cap += aligned;
-    }
-}
-
-fn is_bar_offset(offset: usize) -> bool {
-    offset >= 0x10 && offset < 0x28
-}
-
-fn bar_to_offset(bar: usize) -> usize {
-    0x10 + (bar * 4)
-}
-
-fn offset_to_bar(offset: usize) -> usize {
-    assert!(offset >= 0x10 && offset < 0x28, "not a valid bar offset");
-    (offset - 0x10) / 4
-}
-
-
-pub struct VirtioCap {
-    offset: usize,
-    vtype: u8,
-    size: u8,
-    mmio_offset: u32,
-    mmio_len: u32,
-    extra_word: Option<u32>,
-}
-
-impl VirtioCap {
-    fn new(offset: usize, vtype: u8) -> VirtioCap {
-        VirtioCap {
-            vtype,
-            offset,
-            size: 16,
-            mmio_offset: 0,
-            mmio_len: 0,
-            extra_word: None,
-        }
-    }
-
-    pub fn set_mmio_range(&mut self, offset: usize, len: usize) -> &mut VirtioCap {
-        self.mmio_offset = offset as u32;
-        self.mmio_len = len as u32;
-        self
-    }
-
-    pub fn set_extra_word(&mut self, val: u32) -> &mut VirtioCap {
-        self.size += 4;
-        self.extra_word = Some(val);
-        self
-    }
-
-    pub fn add(&mut self, dev: &mut PciDevice) {
-        /*
-         * struct virtio_pci_cap {
-         *     u8 cap_vndr; /* Generic PCI field: PCI_CAP_ID_VNDR */
-         *     u8 cap_next; /* Generic PCI field: next ptr. */
-         *     u8 cap_len; /* Generic PCI field: capability length */
-         *     u8 cfg_type; /* Identifies the structure. */
-         *     u8 bar; /* Where to find it. */
-         *     u8 padding[3]; /* Pad to full dword. */
-         *     le32 offset; /* Offset within bar. */
-         *     le32 length; /* Length of the structure, in bytes. */
-         * };
-         */
-        dev.w8(self.offset, PCI_CAP_ID_VENDOR);
-        dev.w8(self.offset + 2, self.size);
-        dev.w8(self.offset + 3, self.vtype);
-        dev.w8(self.offset + 4, VIRTIO_MMIO_BAR as u8);
-        if self.mmio_len > 0 {
-            dev.w32(self.offset + 8, self.mmio_offset);
-            dev.w32(self.offset + 12, self.mmio_len);
-        }
-        if let Some(word) = self.extra_word {
-            dev.w32(self.offset + 16, word);
-        }
-
-        dev.inc_cap(self.size as usize);
-    }
-}

src/virtio/virtqueue.rs

@@ -1,177 +0,0 @@
-use std::sync::atomic::{Ordering, AtomicUsize, AtomicBool};
-use std::sync::Arc;
-use vmm_sys_util::eventfd::EventFd;
-
-use crate::memory::GuestRam;
-use crate::virtio::{Result,Error};
-use super::consts::*;
-use super::vring::{Vring,Descriptor};
-use super::bus::VirtioDeviceConfig;
-use crate::virtio::chain::Chain;
-use crate::vm::KvmVm;
-
-#[derive(Clone)]
-pub struct VirtQueue {
-    memory: GuestRam,
-    vring: Vring,
-    features: u64,
-    ioeventfd: Arc<EventFd>,
-    interrupt: Arc<InterruptLine>,
-    closed: Arc<AtomicBool>,
-}
-
-impl VirtQueue {
-    pub fn new(memory: GuestRam, vring: Vring, interrupt: Arc<InterruptLine>, ioeventfd: Arc<EventFd>) -> VirtQueue {
-        VirtQueue {
-            memory,
-            vring,
-            features: 0,
-            ioeventfd,
-            interrupt,
-            closed: Arc::new(AtomicBool::new(false)),
-        }
-    }
-
-    #[allow(dead_code)]
-    pub fn set_closed(&self) {
-        self.closed.store(true, Ordering::SeqCst);
-        self.ioeventfd.write(1).unwrap();
-    }
-
-    #[allow(dead_code)]
-    pub fn is_closed(&self) -> bool {
-        self.closed.load(Ordering::SeqCst)
-    }
-
-    fn use_event_idx(&self) -> bool {
-        self.features & VIRTIO_F_EVENT_IDX != 0
-    }
-
-    pub fn wait_ready(&self) -> Result<()> {
-        if self.vring.is_empty() {
-            let _ = self.ioeventfd.read()
-                .map_err(Error::ReadIoEventFd)?;
-        }
-        Ok(())
-    }
-
-    pub fn wait_next_chain(&self) -> Result<Chain> {
-        loop {
-            self.wait_ready()?;
-            if let Some(idx) = self.pop_avail_entry() {
-                return Ok(Chain::new(self.memory.clone(), self.clone(), idx, self.vring.size()));
-            }
-        }
-    }
-
-    pub fn next_chain(&self) -> Option<Chain> {
-        self.pop_avail_entry()
-            .map(|idx| Chain::new(self.memory.clone(), self.clone(), idx, self.vring.size()))
-    }
-
-    pub fn on_each_chain<F>(&self, mut f: F)
-        where F: FnMut(Chain) {
-        loop {
-            self.wait_ready().unwrap();
-            for chain in self.iter() {
-                f(chain);
-            }
-        }
-    }
-
-    pub fn iter(&self) -> QueueIter {
-        QueueIter { vq: self.clone() }
-    }
-
-    fn need_interrupt(&self, first_used: u16, used_count: usize) -> bool {
-        if used_count == 0 {
-            return false;
-        }
-        if self.use_event_idx() {
-            let event = self.vring.read_used_event();
-            // Minimum count needed to traverse event idx
-            let span = ((event - first_used) + 1) as usize;
-            return used_count >= span;
-        }
-        !self.vring.read_avail_no_interrupt()
-    }
-
-    pub fn put_used(&self, idx: u16, len: u32) {
-        let used = self.vring.next_used();
-        self.vring.put_used(idx, len);
-        if self.need_interrupt(used, 1) {
-            self.interrupt.notify_queue();
-        }
-    }
-
-    fn pop_avail_entry(&self) -> Option<u16> {
-        if let Some(idx) = self.vring.pop_avail_entry() {
-            if self.use_event_idx() {
-                self.vring.write_avail_event(self.vring.next_avail());
-            }
-            return Some(idx)
-        }
-        None
-    }
-
-    pub fn load_descriptor(&self, idx: u16) -> Option<Descriptor> {
-        self.vring.load_descriptor(idx)
-    }
-
-    pub fn ioevent(&self) -> &EventFd {
-        &self.ioeventfd
-    }
-}
-
-pub struct QueueIter {
-    vq: VirtQueue,
-}
-
-impl Iterator for QueueIter {
-    type Item = Chain;
-
-    fn next(&mut self) -> Option<Self::Item> {
-        self.vq.pop_avail_entry().map(|idx| {
-            Chain::new(self.vq.memory.clone(), self.vq.clone(), idx, self.vq.vring.size())
-        })
-    }
-}
-
-
-pub struct InterruptLine {
-    irqfd: EventFd,
-    isr: AtomicUsize,
-}
-
-impl InterruptLine {
-    pub fn from_config(conf: &VirtioDeviceConfig) -> Result<Arc<InterruptLine>> {
-        InterruptLine::new(conf.kvm_vm(), conf.irq())
-    }
-
-    fn new(kvm_vm: &KvmVm, irq: u8) -> Result<Arc<InterruptLine>> {
-        let irqfd = EventFd::new(0)
-            .map_err(Error::CreateEventFd)?;
-        kvm_vm.vm_fd().register_irqfd(&irqfd, irq as u32)
-            .map_err(Error::IrqFd)?;
-        Ok(Arc::new(InterruptLine{
-            irqfd,
-            isr: AtomicUsize::new(0)
-        }))
-    }
-
-    pub fn isr_read(&self) -> u64 {
-        self.isr.swap(0, Ordering::SeqCst) as u64
-    }
-
-    pub fn notify_queue(&self) {
-        self.isr.fetch_or(0x1, Ordering::SeqCst);
-        self.irqfd.write(1).unwrap();
-    }
-
-    pub fn notify_config(&self) {
-        self.isr.fetch_or(0x2, Ordering::SeqCst);
-        self.irqfd.write(1).unwrap();
-    }
-}
-
-

src/virtio/vring.rs

@@ -1,384 +0,0 @@
-use std::sync::atomic::{self,AtomicUsize,Ordering};
-use std::sync::Arc;
-use std::fmt;
-use std::cmp;
-use std::io::{self, Read};
-
-use crate::memory::GuestRam;
-use super::consts::*;
-
-use crate::virtio::{Result,Error};
-
-///
-/// A convenience wrapper around `AtomicUsize`
-///
-#[derive(Clone)]
-struct SharedIndex(Arc<AtomicUsize>);
-
-impl SharedIndex {
-    fn new() -> SharedIndex {
-        SharedIndex(Arc::new(AtomicUsize::new(0)))
-    }
-    fn get(&self) -> u16 {
-        self.0.load(Ordering::SeqCst) as u16
-    }
-    fn inc(&self) {
-        self.0.fetch_add(1, Ordering::SeqCst);
-    }
-    fn set(&self, v: u16) {
-        self.0.store(v as usize, Ordering::SeqCst);
-    }
-}
-
-///
-/// Access to the low-level memory structure of a Virtqueue.
-///
-#[derive(Clone)]
-pub struct Vring {
-    memory: GuestRam,
-    /// Default queue_size for this virtqueue
-    default_size: u16,
-    /// Number of elements in the virtqueue ring
-    queue_size: u16,
-    /// Guest address for base of descriptor table
-    pub descriptors: u64,
-    /// Guest address for base of available ring
-    pub avail_ring: u64,
-    /// Guest address for base of used ring
-    pub used_ring: u64,
-
-    /// Has this virtqueue been enabled?
-    enabled: bool,
-
-    /// The index in the used ring where the next used entry will be placed
-    next_used_idx: SharedIndex,
-    /// last seen avail_idx loaded from guest memory
-    cached_avail_idx: SharedIndex,
-    /// The index in the avail ring where the next available entry will be read
-    next_avail: SharedIndex,
-}
-
-impl Vring {
-
-    pub fn new(memory: GuestRam, default_size: u16) -> Vring {
-        Vring {
-            memory,
-            default_size,
-            queue_size: default_size,
-            descriptors:0,
-            avail_ring: 0,
-            used_ring: 0,
-            enabled: false,
-
-            next_used_idx: SharedIndex::new(),
-            cached_avail_idx: SharedIndex::new(),
-            next_avail: SharedIndex::new(),
-        }
-
-    }
-
-    ///
-    /// Set `Vring` into the enabled state.
-    ///
-    pub fn enable(&mut self) {
-        self.enabled = true;
-    }
-
-    ///
-    /// Return `true` if this `Vring` has been enabled.
-    ///
-    pub fn is_enabled(&self) -> bool {
-        self.enabled
-    }
-
-    ///
-    /// Queue size of this `Vring`
-    ///
-    pub fn size(&self) -> u16 {
-        self.queue_size
-    }
-
-    ///
-    /// Set the queue size of this `Vring`.  If `sz` is an invalid value
-    /// ignore the request.  It is illegal to change the queue size after
-    /// a virtqueue has been enabled, so ignore requests if enabled.
-    ///
-    /// Valid sizes are less than or equal to `MAX_QUEUE_SIZE` and must
-    /// be a power of 2.
-    ///
-    pub fn set_size(&mut self, sz: u16) {
-        if self.enabled || sz > MAX_QUEUE_SIZE || (sz & (sz - 1) != 0) {
-            return;
-        }
-        self.queue_size = sz;
-    }
-
-    ///
-    /// Reset `Vring` to the initial state.  `queue_size` is set to the `default_size`
-    /// and all other fields are cleared.  `enabled` is set to false.
-    ///
-    pub fn reset(&mut self) {
-        self.queue_size = self.default_size;
-        self.descriptors = 0;
-        self.avail_ring = 0;
-        self.used_ring = 0;
-        self.enabled = false;
-        self.next_used_idx.set(0);
-        self.cached_avail_idx.set(0);
-        self.next_avail.set(0);
-    }
-
-
-    ///
-    /// Does `Vring` currently have available entries?
-    ///
-    /// Queue is empty if `next_avail` is same value as
-    /// `avail_ring.idx` value in guest memory If `cached_avail_idx`
-    /// currently matches `next_avail` it is reloaded from
-    /// memory in case guest has updated field since last
-    /// time it was loaded.
-    ///
-    pub fn is_empty(&self) -> bool {
-        let next_avail = self.next_avail.get();
-        if self.cached_avail_idx.get() != next_avail {
-            return false;
-        }
-        next_avail == self.load_avail_idx()
-    }
-
-    ///
-    /// Write an entry into the Used ring.
-    ///
-    /// The entry is written into the ring structure at offset
-    /// `next_used_idx % queue_size`.  The value of `next_used_idx`
-    /// is then incremented and the new value is written into
-    /// guest memory into the `used_ring.idx` field.
-    ///
-    pub fn put_used(&self, idx: u16, len: u32) {
-        if idx >= self.queue_size {
-            return;
-        }
-
-        let used_idx = (self.next_used_idx.get() % self.queue_size) as u64;
-        let elem_addr = self.used_ring + (4 + used_idx * 8);
-        // write descriptor index to 'next used' slot in used ring
-        self.memory.write_int(elem_addr, idx as u32).unwrap();
-        // write length to 'next used' slot in ring
-        self.memory.write_int(elem_addr + 4, len as u32).unwrap();
-
-        self.next_used_idx.inc();
-        atomic::fence(Ordering::Release);
-        // write updated next_used
-        self.memory.write_int(self.used_ring + 2, self.next_used_idx.get()).unwrap();
-    }
-
-
-    ///
-    /// Load `avail_ring.idx` from guest memory and store it in `cached_avail_idx`.
-    ///
-    pub fn load_avail_idx(&self) -> u16 {
-        let avail_idx = self.memory.read_int::<u16>(self.avail_ring + 2).unwrap();
-        self.cached_avail_idx.set(avail_idx);
-        avail_idx
-    }
-
-    ///
-    /// Read from guest memory and return the Avail ring entry at
-    /// index `ring_idx % queue_size`.
-    ///
-    fn load_avail_entry(&self, ring_idx: u16) -> u16 {
-        let offset = (4 + (ring_idx % self.queue_size) * 2) as u64;
-        self.memory.read_int(self.avail_ring + offset).unwrap()
-    }
-
-    ///
-    /// If queue is not empty, read and return the next Avail ring entry
-    /// and increment `next_avail`.  If queue is empty return `None`
-    ///
-    pub fn pop_avail_entry(&self) -> Option<u16> {
-        if self.is_empty() {
-            return None
-        }
-        let next_avail = self.next_avail.get();
-        let avail_entry = self.load_avail_entry(next_avail);
-        self.next_avail.inc();
-        Some(avail_entry)
-    }
-
-    pub fn next_avail(&self) -> u16 {
-        self.next_avail.get() % self.queue_size
-    }
-
-    ///
-    /// Read and return the `used_event` field from the Avail ring.
-    ///
-    pub fn read_used_event(&self) -> u16 {
-        let addr = self.avail_ring + 4 + (self.queue_size as u64 * 2);
-        self.memory.read_int::<u16>(addr).unwrap()
-    }
-
-    ///
-    /// Read the `flags` field from the Avail ring and return `true` if
-    /// `NO_INTERRUPT` bit is set.
-    ///
-    pub fn read_avail_no_interrupt(&self) -> bool {
-        let flags = self.memory.read_int::<u16>(self.avail_ring).unwrap();
-        flags & 0x01 != 0
-    }
-
-    ///
-    /// Write `val` to the `avail_event` field of Used ring.
-    ///
-    /// If `val` is not a valid index for this virtqueue this
-    /// function does nothing.
-    ///
-    pub fn write_avail_event(&self, val: u16) {
-        if val > self.queue_size {
-            return;
-        }
-        let addr = self.used_ring + 4 + (self.queue_size as u64 * 8);
-        self.memory.write_int::<u16>(addr, val).unwrap();
-        atomic::fence(Ordering::Release);
-    }
-
-    ///
-    /// Set or clear the `NO_NOTIFY` bit in flags field of Used ring
-    ///
-    #[allow(dead_code)]
-    pub fn write_used_no_notify(&self, val: bool) {
-        let flag = if val { 0x1 } else { 0x0 };
-        self.memory.write_int::<u16>(self.used_ring, flag).unwrap();
-    }
-
-    ///
-    /// Load the descriptor table entry at `idx` from guest memory and return it.
-    ///
-    pub fn load_descriptor(&self, idx: u16) -> Option<Descriptor> {
-        if idx >= self.queue_size {
-            panic!("load_descriptor called with index larger than queue size");
-        }
-        let head = self.descriptors + (idx as u64 * 16);
-
-        let addr = self.memory.read_int::<u64>(head).unwrap();
-        let len= self.memory.read_int::<u32>(head + 8).unwrap();
-        let flags = self.memory.read_int::<u16>(head + 12).unwrap();
-        let next = self.memory.read_int::<u16>(head + 14).unwrap();
-
-        if self.memory.is_valid_range(addr, len as usize) && next < self.queue_size {
-            return Some(Descriptor::new(idx, addr, len, flags, next));
-        }
-        None
-    }
-
-    pub fn next_used(&self) -> u16 {
-        self.next_used_idx.get()
-    }
-
-    pub fn validate(&self) -> Result<()> {
-        if !self.enabled {
-            return Err(Error::VringNotEnabled);
-        }
-        let qsz = self.queue_size as usize;
-        let desc_table_sz = 16 * qsz;
-        let avail_ring_sz = 6 + 2 * qsz;
-        let used_ring_sz = 6 + 8 * qsz;
-        if !self.memory.is_valid_range(self.descriptors, desc_table_sz) {
-            return Err(Error::VringRangeInvalid(self.descriptors));
-        }
-        if !self.memory.is_valid_range(self.avail_ring, avail_ring_sz) {
-            return Err(Error::VringAvailInvalid(self.avail_ring));
-        }
-        if !self.memory.is_valid_range(self.used_ring, used_ring_sz) {
-            return Err(Error::VringUsedInvalid(self.used_ring));
-        }
-        Ok(())
-    }
-}
-
-///
-/// An entry read from the descriptor table
-///
-#[derive(Copy,Clone)]
-pub struct Descriptor {
-    pub idx: u16,
-    pub addr: u64,
-    pub len: u32,
-    pub flags: u16,
-    pub next: u16,
-}
-
-impl Descriptor {
-    fn new(idx: u16, addr: u64, len: u32, flags: u16, next:u16) -> Descriptor {
-        Descriptor{ idx, addr, len, flags, next }
-    }
-
-    ///
-    /// Test if `flag` is set in `self.flags`
-    ///
-    fn has_flag(&self, flag: u16) -> bool {
-        self.flags & flag == flag
-    }
-
-    ///
-    /// Is VRING_DESC_F_NEXT set in `self.flags`?
-    ///
-    pub fn has_next(&self) -> bool {
-        self.has_flag(VRING_DESC_F_NEXT)
-    }
-
-    ///
-    /// Is VRING_DESC_F_WRITE set in `self.flags`?
-    ///
-    pub fn is_write(&self) -> bool {
-        self.has_flag(VRING_DESC_F_WRITE)
-    }
-
-    ///
-    /// Is VRING_DESC_F_INDIRECT set in `self.flags`?
-    ///
-    #[allow(dead_code)]
-    pub fn is_indirect(&self) -> bool {
-        self.has_flag(VRING_DESC_F_INDIRECT)
-    }
-
-    pub fn remaining(&self, offset: usize) -> usize {
-        if offset >= self.len as usize {
-            0
-        } else {
-            self.len as usize - offset
-        }
-    }
-
-    pub fn read_from(&self, memory: &GuestRam, offset: usize, buf: &mut[u8]) -> usize {
-        let sz = cmp::min(buf.len(), self.remaining(offset));
-        if sz > 0 {
-            memory.read_bytes(self.addr + offset as u64, &mut buf[..sz]).unwrap();
-        }
-        sz
-    }
-
-    pub fn write_to(&self, memory: &GuestRam, offset: usize, buf: &[u8]) -> usize {
-        let sz = cmp::min(buf.len(), self.remaining(offset));
-        if sz > 0 {
-            memory.write_bytes(self.addr + offset as u64, &buf[..sz]).unwrap();
-        }
-        sz
-    }
-
-    pub fn write_from_reader<R: Read+Sized>(&self, memory: &GuestRam, offset: usize, mut r: R, size: usize) -> io::Result<usize> {
-        let sz = cmp::min(size, self.remaining(offset));
-        if sz > 0 {
-            let slice = memory.mut_slice(self.addr + offset as u64, sz).unwrap();
-            return r.read(slice);
-        }
-        Ok(0)
-    }
-}
-
-impl fmt::Debug for Descriptor {
-    fn fmt(&self, f: &mut fmt::Formatter) -> fmt::Result {
-        write!(f, "Descriptor{{ idx: {} addr: {:x} len: {} flags: {:x} next: {} }}", self.idx, self.addr, self.len, self.flags, self.next)
-    }
-}
-
-

src/vm/arch/mod.rs

@@ -6,12 +6,13 @@ use crate::memory::MemoryManager;
 mod error;
 mod x86;
 
-pub use x86::PCI_MMIO_RESERVED_BASE;
+pub use x86::{PCI_MMIO_RESERVED_BASE,PCI_MMIO_RESERVED_SIZE,IRQ_BASE,IRQ_MAX};
+
 
 pub use error::{Error,Result};
+use crate::io::PciIrq;
 use crate::vm::kernel_cmdline::KernelCmdLine;
 use crate::vm::VmConfig;
-use crate::virtio::PciIrq;
 use crate::vm::kvm_vm::KvmVm;
 
 pub fn create_setup(config: &VmConfig) -> X86ArchSetup {

src/vm/arch/x86/memory.rs

@@ -1,16 +1,18 @@
 use crate::memory::{MemoryManager, MemoryRegion, GuestRam};
 use crate::vm::arch::{Error, Result};
 use std::cmp;
+use crate::io::PciIrq;
 use crate::vm::kernel_cmdline::KernelCmdLine;
 use crate::vm::arch::x86::kernel::{load_pm_kernel, KERNEL_CMDLINE_ADDRESS};
 use crate::system;
 use crate::vm::arch::x86::mptable::setup_mptable;
-use crate::virtio::PciIrq;
 use crate::vm::KvmVm;
 
 pub const HIMEM_BASE: u64 = 1 << 32;
 pub const PCI_MMIO_RESERVED_SIZE: usize = 512 << 20;
 pub const PCI_MMIO_RESERVED_BASE: u64 = HIMEM_BASE - PCI_MMIO_RESERVED_SIZE as u64;
+pub const IRQ_BASE: u32 = 5;
+pub const IRQ_MAX: u32 = 23;
 
 
 pub fn x86_setup_memory_regions(memory: &mut MemoryManager, ram_size: usize) -> Result<()> {

src/vm/arch/x86/mod.rs

@@ -8,4 +8,4 @@ mod kernel;
 mod setup;
 
 pub use setup::X86ArchSetup;
-pub use memory::PCI_MMIO_RESERVED_BASE;
+pub use memory::{PCI_MMIO_RESERVED_BASE,PCI_MMIO_RESERVED_SIZE,IRQ_BASE,IRQ_MAX};

src/vm/kvm_vm.rs

@@ -4,9 +4,9 @@ use std::sync::atomic::AtomicBool;
 use kvm_bindings::{CpuId, KVM_MAX_CPUID_ENTRIES, kvm_pit_config, KVM_PIT_SPEAKER_DUMMY, kvm_userspace_memory_region};
 use kvm_ioctls::{Cap, Kvm, VmFd};
 use kvm_ioctls::Cap::*;
+use crate::io::manager::IoManager;
 use crate::vm::vcpu::Vcpu;
 use crate::vm::{Result, Error, ArchSetup};
-use crate::vm::io::IoDispatcher;
 
 const KVM_API_VERSION: i32 = 12;
 type KvmResult<T> = result::Result<T, kvm_ioctls::Error>;
@@ -17,6 +17,7 @@ static REQUIRED_EXTENSIONS: &[Cap] = &[
     ExtCpuid,
     Hlt,
     Ioeventfd,
+    IoeventfdNoLength,
     Irqchip,
     MpState,
     Pit2,
@@ -115,10 +116,10 @@ impl KvmVm {
             .map_err(Error::VmSetup)
     }
 
-    pub fn create_vcpu<A: ArchSetup>(&self, id: u64, io: Arc<IoDispatcher>, shutdown: Arc<AtomicBool>, arch: &mut A) -> Result<Vcpu> {
+    pub fn create_vcpu<A: ArchSetup>(&self, id: u64, io_manager: IoManager, shutdown: Arc<AtomicBool>, arch: &mut A) -> Result<Vcpu> {
         let vcpu_fd = self.vm_fd.create_vcpu(id)
             .map_err(Error::CreateVcpu)?;
-        let vcpu = Vcpu::new(vcpu_fd, io, shutdown);
+        let vcpu = Vcpu::new(vcpu_fd, io_manager, shutdown);
         arch.setup_vcpu(vcpu.vcpu_fd(), self.supported_cpuid().clone()).map_err(Error::ArchError)?;
         Ok(vcpu)
     }

src/vm/mod.rs

@@ -3,7 +3,6 @@ static PHINIT: &[u8] = include_bytes!("../../ph-init/target/release/ph-init");
 static SOMMELIER: &[u8] = include_bytes!("../../sommelier/build/sommelier");
 
 pub mod arch;
-pub mod io;
 mod setup;
 mod error;
 mod kernel_cmdline;

src/vm/setup.rs

@@ -1,13 +1,9 @@
 use crate::vm::{VmConfig, Result, Error, PHINIT, SOMMELIER};
 use crate::vm::arch::ArchSetup;
 use crate::vm::kernel_cmdline::KernelCmdLine;
-use crate::vm::io::IoDispatcher;
-use crate::devices;
 use termios::Termios;
-use crate::virtio::VirtioBus;
+use crate::devices::{SyntheticFS, VirtioBlock, VirtioNet, VirtioP9, VirtioRandom, VirtioSerial, VirtioWayland};
-use crate::virtio;
+use std::{env, fs, thread};
-use crate::devices::SyntheticFS;
-use std::{fs, thread};
 use crate::system::{Tap, NetlinkSocket};
 use crate::disk::DiskImage;
 use std::sync::Arc;
@@ -15,6 +11,10 @@ use crate::memory::MemoryManager;
 use std::sync::atomic::AtomicBool;
 use kvm_ioctls::VmFd;
 use vmm_sys_util::eventfd::EventFd;
+use crate::devices::ac97::{Ac97Dev, Ac97Parameters};
+use crate::devices::serial::SerialPort;
+use crate::io::manager::IoManager;
+use crate::{Logger, LogLevel};
 use crate::vm::kvm_vm::KvmVm;
 use crate::vm::vcpu::Vcpu;
 
@@ -22,12 +22,12 @@ pub struct Vm {
     kvm_vm: KvmVm,
     vcpus: Vec<Vcpu>,
     memory: MemoryManager,
-    io_dispatch: Arc<IoDispatcher>,
+    io_manager: IoManager,
     termios: Option<Termios>,
 }
 
 impl Vm {
-    fn create<A: ArchSetup>(arch: &mut A, reset_evt: EventFd) -> Result<Self> {
+    fn create<A: ArchSetup>(arch: &mut A) -> Result<Self> {
         let kvm_vm = KvmVm::open()?;
         kvm_vm.create_irqchip()?;
         kvm_vm.vm_fd().set_tss_address(0xfffbd000)
@@ -36,11 +36,13 @@ impl Vm {
         let memory = arch.create_memory(kvm_vm.clone())
             .map_err(Error::ArchError)?;
 
+        let io_manager = IoManager::new(memory.clone());
+
         Ok(Vm {
             kvm_vm,
             memory,
+            io_manager,
             vcpus: Vec::new(),
-            io_dispatch: IoDispatcher::new(reset_evt),
             termios: None,
         })
     }
@@ -89,14 +91,16 @@ impl <T: ArchSetup> VmSetup <T> {
 
     pub fn create_vm(&mut self) -> Result<Vm> {
         let exit_evt = EventFd::new(libc::EFD_NONBLOCK)?;
-        let reset_evt = exit_evt.try_clone()?;
+        let mut vm = Vm::create(&mut self.arch)?;
-        let mut vm = Vm::create(&mut self.arch, reset_evt)?;
+
+        let reset_evt = exit_evt.try_clone()?;
+        vm.io_manager.register_legacy_devices(reset_evt);
 
-        devices::rtc::Rtc::register(vm.io_dispatch.clone());
 
         if self.config.verbose() {
+            Logger::set_log_level(LogLevel::Info);
             self.cmdline.push("earlyprintk=serial");
-            devices::serial::SerialDevice::register(vm.kvm_vm.clone(),vm.io_dispatch.clone(), 0);
+            vm.io_manager.register_serial_port(SerialPort::COM1);
         } else {
             self.cmdline.push("quiet");
         }
@@ -115,36 +119,50 @@ impl <T: ArchSetup> VmSetup <T> {
             .map_err(Error::TerminalTermios)?;
         vm.termios = Some(saved);
 
-        let mut virtio = VirtioBus::new(vm.memory.clone(), vm.io_dispatch.clone(), vm.kvm_vm.clone());
+        self.setup_synthetic_bootfs(&mut vm.io_manager)?;
-        self.setup_synthetic_bootfs(&mut virtio)?;
+        self.setup_virtio(&mut vm.io_manager)?;
-        self.setup_virtio(&mut virtio)
+
-            .map_err(Error::SetupVirtio)?;
+        if self.config.is_audio_enable() {
+
+            if unsafe { libc::geteuid() } == 0 {
+                self.drop_privs();
+            }
+            env::set_var("HOME", "/home/citadel");
+            env::set_var("XDG_RUNTIME_DIR", "/run/user/1000");
+            let irq = vm.io_manager.allocator().allocate_irq();
+            let mem = vm.memory.guest_ram().clone();
+            // XXX expect()
+            let ac97 = Ac97Dev::try_new(&vm.kvm_vm, irq, mem, Ac97Parameters::new_pulseaudio()).expect("audio initialize error");
+            vm.io_manager.add_pci_device(Arc::new(Mutex::new(ac97)));
+
+        }
 
         if let Some(init_cmd) = self.config.get_init_cmdline() {
             self.cmdline.push_set_val("init", init_cmd);
         }
 
-        self.arch.setup_memory(&self.cmdline, &virtio.pci_irqs())
+        let pci_irqs = vm.io_manager.pci_irqs();
+        self.arch.setup_memory(&self.cmdline, &pci_irqs)
             .map_err(Error::ArchError)?;
 
         let shutdown = Arc::new(AtomicBool::new(false));
         for id in 0..self.config.ncpus() {
-            let vcpu = vm.kvm_vm.create_vcpu(id as u64, vm.io_dispatch.clone(), shutdown.clone(), &mut self.arch)?;
+            let vcpu = vm.kvm_vm.create_vcpu(id as u64, vm.io_manager.clone(), shutdown.clone(), &mut self.arch)?;
             vm.vcpus.push(vcpu);
         }
         Ok(vm)
     }
 
-    fn setup_virtio(&mut self, virtio: &mut VirtioBus) -> virtio::Result<()> {
+    fn setup_virtio(&mut self, io_manager: &mut IoManager) -> Result<()> {
-        devices::VirtioSerial::create(virtio)?;
+        io_manager.add_virtio_device(VirtioSerial::new())?;
-        devices::VirtioRandom::create(virtio)?;
+        io_manager.add_virtio_device(VirtioRandom::new())?;
 
         if self.config.is_wayland_enabled() {
-            devices::VirtioWayland::create(virtio, self.config.is_dmabuf_enabled())?;
+            io_manager.add_virtio_device(VirtioWayland::new(self.config.is_dmabuf_enabled()))?;
         }
 
         let homedir = self.config.homedir();
-        devices::VirtioP9::create(virtio, "home", homedir, false, false)?;
+        io_manager.add_virtio_device(VirtioP9::new_filesystem("home", homedir, false, false))?;
         if homedir != "/home/user" && !self.config.is_realm() {
             self.cmdline.push_set_val("phinit.home", homedir);
         }
@@ -155,14 +173,14 @@ impl <T: ArchSetup> VmSetup <T> {
             if block_root == None {
                 block_root = Some(disk.read_only());
             }
-            devices::VirtioBlock::create(virtio, disk)?;
+            io_manager.add_virtio_device(VirtioBlock::new(disk))?;
         }
 
         for disk in self.config.get_raw_disk_images() {
             if block_root == None {
                 block_root = Some(disk.read_only());
             }
-            devices::VirtioBlock::create(virtio, disk)?;
+            io_manager.add_virtio_device(VirtioBlock::new(disk))?;
         }
 
         if let Some(read_only) = block_root {
@@ -172,14 +190,14 @@ impl <T: ArchSetup> VmSetup <T> {
             self.cmdline.push("phinit.root=/dev/vda");
             self.cmdline.push("phinit.rootfstype=ext4");
         } else {
-            devices::VirtioP9::create(virtio, "9proot", "/", true, false)?;
+            io_manager.add_virtio_device(VirtioP9::new_filesystem("9proot", "/", true, false))?;
             self.cmdline.push_set_val("phinit.root", "9proot");
             self.cmdline.push_set_val("phinit.rootfstype", "9p");
             self.cmdline.push_set_val("phinit.rootflags", "trans=virtio");
         }
 
         if self.config.network() {
-            self.setup_network(virtio)?;
+            self.setup_network(io_manager)?;
             self.drop_privs();
 
         }
@@ -196,12 +214,11 @@ impl <T: ArchSetup> VmSetup <T> {
 
     }
 
-    fn setup_synthetic_bootfs(&mut self, virtio: &mut VirtioBus) -> Result<()> {
+    fn setup_synthetic_bootfs(&mut self, io_manager: &mut IoManager) -> Result<()> {
         let bootfs = self.create_bootfs()
             .map_err(Error::SetupBootFs)?;
 
-        devices::VirtioP9::create_with_filesystem(bootfs, virtio, "/dev/root", "/", false)
+        io_manager.add_virtio_device(VirtioP9::new(bootfs, "/dev/root", "/", false))?;
-            .map_err(Error::SetupVirtio)?;
 
         self.cmdline.push_set_val("init", "/usr/bin/ph-init");
         self.cmdline.push_set_val("root", "/dev/root");
@@ -211,7 +228,7 @@ impl <T: ArchSetup> VmSetup <T> {
         Ok(())
     }
 
-    fn create_bootfs(&self) -> ::std::io::Result<SyntheticFS> {
+    fn create_bootfs(&self) -> std::io::Result<SyntheticFS> {
         let mut s = SyntheticFS::new();
         s.mkdirs(&["/tmp", "/proc", "/sys", "/dev", "/home/user", "/bin", "/etc"]);
 
@@ -226,7 +243,7 @@ impl <T: ArchSetup> VmSetup <T> {
         Ok(s)
     }
 
-    fn setup_network(&mut self, virtio: &mut VirtioBus) -> virtio::Result<()> {
+    fn setup_network(&mut self, io_manager: &mut IoManager) -> Result<()> {
         let tap = match self.setup_tap() {
             Ok(tap) => tap,
             Err(e) => {
@@ -234,7 +251,7 @@ impl <T: ArchSetup> VmSetup <T> {
                 return Ok(());
             }
         };
-        devices::VirtioNet::create(virtio, tap)?;
+        io_manager.add_virtio_device(VirtioNet::new(tap))?;
         self.cmdline.push("phinit.ip=172.17.0.22");
         Ok(())
     }

src/vm/vcpu.rs

@@ -1,34 +1,22 @@
-use std::convert::TryInto;
+use std::sync::{Arc, Barrier};
-use std::sync::Arc;
 use std::sync::atomic::{AtomicBool,Ordering};
 use kvm_ioctls::{VcpuExit, VcpuFd};
-use vmm_sys_util::sock_ctrl_msg::IntoIovec;
+use crate::io::manager::IoManager;
-use crate::vm::io::IoDispatcher;
 
 
-/*
-pub enum VcpuEvent {
-    Exit,
-}
-
-pub struct VcpuHandle {
-    sender: Sender<VcpuEvent>,
-    thread: thread::JoinHandle<()>,
-}
-
- */
-
 pub struct Vcpu {
     vcpu_fd: VcpuFd,
-    io: Arc<IoDispatcher>,
+    io_manager: IoManager,
     shutdown: Arc<AtomicBool>,
 }
 
 
 impl Vcpu {
-    pub fn new(vcpu_fd: VcpuFd, io: Arc<IoDispatcher>, shutdown: Arc<AtomicBool>) -> Self {
+    pub fn new(vcpu_fd: VcpuFd, io_manager: IoManager, shutdown: Arc<AtomicBool>) -> Self {
         Vcpu {
-            vcpu_fd, io, shutdown,
+            vcpu_fd,
+            io_manager,
+            shutdown,
         }
     }
 
@@ -36,60 +24,45 @@ impl Vcpu {
         &self.vcpu_fd
     }
 
-    fn data_to_int(data: &[u8]) -> u64 {
-        match data.len() {
-            1 =>  data[0] as u64,
-            2 =>  u16::from_le_bytes(data.try_into().unwrap()) as u64,
-            4 =>  u32::from_le_bytes(data.try_into().unwrap()) as u64,
-            8 =>  u64::from_le_bytes(data.try_into().unwrap()),
-            _ => 0,
-        }
-    }
-
-    fn int_to_data(n: u64, data: &mut[u8]) {
-        match data.len() {
-            1 => data[0] = n as u8,
-            2 => data.copy_from_slice((n as u16).to_le_bytes().as_slice()),
-            4 => data.copy_from_slice((n as u32).to_le_bytes().as_slice()),
-            8 => data.copy_from_slice((n as u64).to_le_bytes().as_slice()),
-            _ => {},
-        }
-    }
 
     fn handle_io_out(&self, port: u16, data: &[u8]) {
-        let val = Self::data_to_int(data) as u32;
+        let _ok = self.io_manager.pio_write(port, data);
-        self.io.emulate_io_out(port, data.size(), val);
     }
 
     fn handle_io_in(&self, port: u16, data: &mut [u8]) {
-        let val = self.io.emulate_io_in(port, data.len());
+        let _ok = self.io_manager.pio_read(port, data);
-        Self::int_to_data(val as u64, data);
     }
 
     fn handle_mmio_read(&self, addr: u64, data: &mut [u8]) {
-        let val = self.io.emulate_mmio_read(addr, data.len());
+        let _ok = self.io_manager.mmio_read(addr, data);
-        Self::int_to_data(val, data);
     }
 
     fn handle_mmio_write(&self, addr: u64, data: &[u8]) {
-        let val = Self::data_to_int(data);
+        let _ok = self.io_manager.mmio_write(addr,data);
-        self.io.emulate_mmio_write(addr, data.size(), val);
     }
 
     fn handle_shutdown(&self) {
         self.shutdown.store(true, Ordering::Relaxed);
     }
 
-    pub fn run(&self) {
+    pub fn run(&self, barrier: &Arc<Barrier>) {
+        barrier.wait();
         loop {
-            match self.vcpu_fd.run().expect("fail") {
+            match self.vcpu_fd.run() {
-                VcpuExit::IoOut(port, data) => self.handle_io_out(port, data),
+                Ok(VcpuExit::IoOut(port, data)) => self.handle_io_out(port, data),
-                VcpuExit::IoIn(port, data) => self.handle_io_in(port, data),
+                Ok(VcpuExit::IoIn(port, data)) => self.handle_io_in(port, data),
-                VcpuExit::MmioRead(addr, data) => self.handle_mmio_read(addr, data),
+                Ok(VcpuExit::MmioRead(addr, data)) => self.handle_mmio_read(addr, data),
-                VcpuExit::MmioWrite(addr, data) => self.handle_mmio_write(addr, data),
+                Ok(VcpuExit::MmioWrite(addr, data)) => self.handle_mmio_write(addr, data),
-                VcpuExit::Shutdown => self.handle_shutdown(),
+                Ok(VcpuExit::Shutdown) => self.handle_shutdown(),
-                exit => {
+                Ok(exit) => {
                     println!("unhandled exit: {:?}", exit);
+                },
+                Err(err) => {
+                    if err.errno() == libc::EAGAIN {}
+                    else {
+                        warn!("VCPU run() returned error: {}", err);
+                        return;
+                    }
                 }
             }
             if self.shutdown.load(Ordering::Relaxed) {