pci: Support injecting interrupts from externally-provided irqfds

The virtio vhost-user device backend prefers to use externally-provided
eventfds as irqfds.  This allows the frontend VM to notify the backend
VM directly, without the need for a userspace proxy process.  Since the
frontend can provide irqfds at any time, the backend needs to register
and unregister irqfds dynamically.

This is tricky because the functions that access the irqfd table all
take `&self`, not `&mut self`.  The obvious solution to this problem is
to wrap the table in a mutex.  Most of these functions are not called on
hot paths, but `.notifier()` is called whenever Cloud Hypervisor needs
to inject an interrupt into a guest.  Most devices don't need to
register irqfds at runtime, and for them, slowing down interrupt
injection would be wasteful.

Instead, require devices to opt-in to irqfd registration.  The irqfd
table now comes in two forms: one that contains a mutex and one that
does not.  The one containing a mutex can be mutated freely, while
attempting to mutate the one that does not will panic.

Right now, no code registeres irqfds at runtime, but this will change in
subsequent commits.

Signed-off-by: Demi Marie Obenour <demiobenour@gmail.com>

pci/src/lib.rs

@@ -32,7 +32,10 @@ pub use self::device::{
     BarReprogrammingParams, DeviceRelocation, Error as PciDeviceError, PciDevice,
 };
 pub use self::msi::{MsiCap, MsiConfig, msi_num_enabled_vectors};
-pub use self::msix::{MSIX_CONFIG_ID, MSIX_TABLE_ENTRY_SIZE, MsixCap, MsixConfig, MsixTableEntry};
+pub use self::msix::{
+    MSIX_CONFIG_ID, MSIX_TABLE_ENTRY_SIZE, MaybeMutInterruptSourceGroup, MsixCap, MsixConfig,
+    MsixTableEntry,
+};
 pub use self::vfio::{MmioRegion, VfioDmaMapping, VfioPciDevice, VfioPciError};
 pub use self::vfio_user::{VfioUserDmaMapping, VfioUserPciDevice, VfioUserPciDeviceError};
 

pci/src/msix.rs

@@ -3,7 +3,7 @@
 // SPDX-License-Identifier: Apache-2.0 OR BSD-3-Clause
 //
 
-use std::sync::Arc;
+use std::sync::{Arc, Mutex};
 use std::{io, result};
 
 use byteorder::{ByteOrder, LittleEndian};
@@ -15,6 +15,7 @@ use vm_device::interrupt::{
 };
 use vm_memory::ByteValued;
 use vm_migration::{MigratableError, Pausable, Snapshot, Snapshottable};
+use vmm_sys_util::eventfd::EventFd;
 
 use crate::{PciCapability, PciCapabilityId};
 
@@ -72,11 +73,66 @@ pub struct MsixConfigState {
     enabled: bool,
 }
 
+#[derive(Clone)]
+pub enum MaybeMutInterruptSourceGroup {
+    Immutable(Arc<dyn InterruptSourceGroup>),
+    Mutable(Arc<Mutex<dyn InterruptSourceGroup>>),
+}
+
+macro_rules! impl_method {
+    ($(
+        fn $i: ident(&self $(,$index:ident : $InterruptIndex:ty)*$(,)?) -> $r: ty;
+    )*) => {
+        $(
+            fn $i(&self $(,$index: $InterruptIndex)*) -> $r {
+                match self {
+                    Self::Immutable(source) => source.$i($($index),*),
+                    Self::Mutable(source) => source.lock().unwrap().$i($($index),*),
+                }
+            }
+        )*
+    };
+}
+
+impl InterruptSourceGroup for MaybeMutInterruptSourceGroup {
+    impl_method! {
+        fn trigger(&self, index: InterruptIndex) -> vm_device::interrupt::Result<()>;
+
+        fn notifier(&self, index: InterruptIndex) -> Option<vmm_sys_util::eventfd::EventFd>;
+
+        fn update(
+            &self,
+            index: InterruptIndex,
+            config: InterruptSourceConfig,
+            masked: bool,
+            set_gsi: bool,
+        ) -> vm_device::interrupt::Result<()>;
+
+        fn set_gsi(&self) -> vm_device::interrupt::Result<()>;
+    }
+}
+
+impl MaybeMutInterruptSourceGroup {
+    pub fn set_notifier(
+        &self,
+        index: InterruptIndex,
+        eventfd: Option<EventFd>,
+        vm: &dyn hypervisor::Vm,
+    ) -> std::io::Result<()> {
+        match self {
+            Self::Immutable(_) => panic!(
+                "Attempted to set a notifier of an immutable source.  You must mark your device as needing a mutable source by having sets_irqfd() return true."
+            ),
+            Self::Mutable(source) => source.lock().unwrap().set_notifier(index, eventfd, vm),
+        }
+    }
+}
+
 pub struct MsixConfig {
     pub table_entries: Vec<MsixTableEntry>,
     pub pba_entries: Vec<u64>,
     pub devid: u32,
-    interrupt_source_group: Arc<dyn InterruptSourceGroup>,
+    interrupt_source_group: MaybeMutInterruptSourceGroup,
     masked: bool,
     enabled: bool,
 }
@@ -84,7 +140,7 @@ pub struct MsixConfig {
 impl MsixConfig {
     pub fn new(
         msix_vectors: u16,
-        interrupt_source_group: Arc<dyn InterruptSourceGroup>,
+        interrupt_source_group: MaybeMutInterruptSourceGroup,
         devid: u32,
         state: Option<MsixConfigState>,
     ) -> result::Result<Self, Error> {

pci/src/vfio.rs

@@ -37,7 +37,7 @@ use vmm_sys_util::eventfd::EventFd;
 
 use crate::mmap::MmapRegion;
 use crate::msi::{MSI_CONFIG_ID, MsiConfigState};
-use crate::msix::MsixConfigState;
+use crate::msix::{MaybeMutInterruptSourceGroup, MsixConfigState};
 use crate::{
     BarReprogrammingParams, MSIX_CONFIG_ID, MSIX_TABLE_ENTRY_SIZE, MsiCap, MsiConfig, MsixCap,
     MsixConfig, PCI_CONFIGURATION_ID, PciBarConfiguration, PciBarPrefetchable, PciBarRegionType,
@@ -863,7 +863,7 @@ impl VfioCommon {
 
         let msix_config = MsixConfig::new(
             msix_cap.table_size(),
-            interrupt_source_group.clone(),
+            MaybeMutInterruptSourceGroup::Immutable(interrupt_source_group.clone()),
             bdf.into(),
             state,
         )

virtio-devices/src/device.rs

@@ -80,6 +80,18 @@ pub trait VirtioDevice: Send {
     /// The maximum size of each queue that this device supports.
     fn queue_max_sizes(&self) -> &[u16];
 
+    /// Whether the device needs to register extra irqfds at runtime
+    /// from external sources.
+    /// The default is false.  If this is true, locking is required for
+    /// most operations involving interrupts (but not for sending)
+    /// interrupts from external irqfds).
+    ///
+    /// If the device claims to not need to register irqfds, but
+    /// attempts to do so, a panic will ensue.
+    fn interrupt_source_mutable(&self) -> bool {
+        false
+    }
+
     /// The set of feature bits that this device supports.
     fn features(&self) -> u64 {
         0

virtio-devices/src/transport/pci_device.rs

@@ -17,9 +17,10 @@ use anyhow::anyhow;
 use libc::EFD_NONBLOCK;
 use log::{error, info};
 use pci::{
-    BarReprogrammingParams, MsixCap, MsixConfig, PciBarConfiguration, PciBarRegionType,
-    PciCapability, PciCapabilityId, PciClassCode, PciConfiguration, PciDevice, PciDeviceError,
-    PciHeaderType, PciMassStorageSubclass, PciNetworkControllerSubclass, PciSubclass,
+    BarReprogrammingParams, MaybeMutInterruptSourceGroup, MsixCap, MsixConfig, PciBarConfiguration,
+    PciBarRegionType, PciCapability, PciCapabilityId, PciClassCode, PciConfiguration, PciDevice,
+    PciDeviceError, PciHeaderType, PciMassStorageSubclass, PciNetworkControllerSubclass,
+    PciSubclass,
 };
 use serde::{Deserialize, Serialize};
 use thiserror::Error;
@@ -359,7 +360,7 @@ pub struct VirtioPciDevice {
     // PCI interrupts.
     interrupt_status: Arc<AtomicUsize>,
     virtio_interrupt: Option<Arc<dyn VirtioInterrupt>>,
-    interrupt_source_group: Arc<dyn InterruptSourceGroup>,
+    interrupt_source_group: MaybeMutInterruptSourceGroup,
 
     // virtio queues
     queues: Vec<Queue>,
@@ -433,17 +434,26 @@ impl VirtioPciDevice {
 
         let pci_device_id = VIRTIO_PCI_DEVICE_ID_BASE + locked_device.device_type() as u16;
 
-        let interrupt_source_group = interrupt_manager
-            .create_group(MsiIrqGroupConfig {
+        let interrupt_source_group: MaybeMutInterruptSourceGroup = {
+            let config = MsiIrqGroupConfig {
                 base: 0,
                 count: msix_num as InterruptIndex,
+            };
+            (if locked_device.interrupt_source_mutable() {
+                interrupt_manager
+                    .create_group_mut(config)
+                    .map(MaybeMutInterruptSourceGroup::Mutable)
+            } else {
+                interrupt_manager
+                    .create_group(config)
+                    .map(MaybeMutInterruptSourceGroup::Immutable)
             })
             .map_err(|e| {
                 VirtioPciDeviceError::CreateVirtioPciDevice(anyhow!(
                     "Failed creating MSI interrupt group: {e}"
                 ))
-            })?;
-
+            })?
+        };
         let msix_state =
             vm_migration::state_from_id(snapshot, pci::MSIX_CONFIG_ID).map_err(|e| {
                 VirtioPciDeviceError::CreateVirtioPciDevice(anyhow!(
@@ -452,14 +462,11 @@ impl VirtioPciDevice {
             })?;
 
         let (msix_config, msix_config_clone) = if msix_num > 0 {
+            let interrupt_source_group: MaybeMutInterruptSourceGroup =
+                interrupt_source_group.clone();
             let msix_config = Arc::new(Mutex::new(
-                MsixConfig::new(
-                    msix_num,
-                    interrupt_source_group.clone(),
-                    pci_device_bdf,
-                    msix_state,
-                )
-                .unwrap(),
+                MsixConfig::new(msix_num, interrupt_source_group, pci_device_bdf, msix_state)
+                    .unwrap(),
             ));
             let msix_config_clone = msix_config.clone();
             (Some(msix_config), Some(msix_config_clone))
@@ -598,7 +605,7 @@ impl VirtioPciDevice {
             memory,
             settings_bar: 0,
             use_64bit_bar,
-            interrupt_source_group,
+            interrupt_source_group: interrupt_source_group.clone(),
             cap_pci_cfg_info,
             bar_regions: vec![],
             activate_evt,
@@ -855,7 +862,7 @@ pub struct VirtioInterruptMsix {
     msix_config: Arc<Mutex<MsixConfig>>,
     config_vector: Arc<AtomicU16>,
     queues_vectors: Arc<Mutex<Vec<u16>>>,
-    interrupt_source_group: Arc<dyn InterruptSourceGroup>,
+    interrupt_source_group: MaybeMutInterruptSourceGroup,
 }
 
 impl VirtioInterruptMsix {
@@ -863,7 +870,7 @@ impl VirtioInterruptMsix {
         msix_config: Arc<Mutex<MsixConfig>>,
         config_vector: Arc<AtomicU16>,
         queues_vectors: Arc<Mutex<Vec<u16>>>,
-        interrupt_source_group: Arc<dyn InterruptSourceGroup>,
+        interrupt_source_group: MaybeMutInterruptSourceGroup,
     ) -> Self {
         VirtioInterruptMsix {
             msix_config,
@@ -917,11 +924,12 @@ impl VirtioInterrupt for VirtioInterruptMsix {
 
     fn set_notifier(
         &self,
-        _interrupt: u32,
-        _eventfd: Option<EventFd>,
-        _vm: &dyn hypervisor::Vm,
+        interrupt: u32,
+        eventfd: Option<EventFd>,
+        vm: &dyn hypervisor::Vm,
     ) -> std::io::Result<()> {
-        unimplemented!()
+        self.interrupt_source_group
+            .set_notifier(interrupt, eventfd, vm)
     }
 }
 

vm-device/src/interrupt/mod.rs

@@ -57,7 +57,8 @@
 //! * The virtual device backend requests the interrupt manager to create an interrupt group
 //!   according to guest configuration information
 
-use std::sync::Arc;
+use std::io::{Error, ErrorKind};
+use std::sync::{Arc, Mutex};
 
 pub use hypervisor::{InterruptSourceConfig, LegacyIrqSourceConfig, MsiIrqSourceConfig};
 use vmm_sys_util::eventfd::EventFd;
@@ -107,6 +108,30 @@ pub trait InterruptManager: Send + Sync {
     /// * count: number of Interrupt Sources to be managed by the group object.
     fn create_group(&self, config: Self::GroupConfig) -> Result<Arc<dyn InterruptSourceGroup>>;
 
+    /// Create an [InterruptSourceGroup](trait.InterruptSourceGroup.html) object to manage
+    /// interrupt sources for a virtual device
+    ///
+    /// An [InterruptSourceGroup](trait.InterruptSourceGroup.html) object manages all interrupt
+    /// sources of the same type for a virtual device.
+    ///
+    /// This is the same as [`Self::create_group`], except that the returned
+    /// [`InterruptSourceGroup`] allows setting the irqfd used as notifier via
+    /// [`InterruptSourceGroup::set_notifier`].
+    ///
+    /// # Arguments
+    /// * interrupt_type: type of interrupt source.
+    /// * base: base Interrupt Source ID to be managed by the group object.
+    /// * count: number of Interrupt Sources to be managed by the group object.
+    fn create_group_mut(
+        &self,
+        _config: Self::GroupConfig,
+    ) -> Result<Arc<Mutex<dyn InterruptSourceGroup>>> {
+        Err(Error::new(
+            ErrorKind::Unsupported,
+            "setting notifiers not supported",
+        ))
+    }
+
     /// Destroy an [InterruptSourceGroup](trait.InterruptSourceGroup.html) object created by
     /// [create_group()](trait.InterruptManager.html#tymethod.create_group).
     ///
@@ -137,7 +162,7 @@ pub trait InterruptSourceGroup: Send + Sync {
     /// Returns an interrupt notifier from this interrupt.
     ///
     /// An interrupt notifier allows for external components and processes
-    /// to inject interrupts into a guest, by writing to the file returned
+    /// to inject interrupts into a guest, by writing to the [`EventFd`] returned
     /// by this method.
     #[allow(unused_variables)]
     fn notifier(&self, index: InterruptIndex) -> Option<EventFd>;
@@ -159,4 +184,17 @@ pub trait InterruptSourceGroup: Send + Sync {
 
     /// Set the interrupt group GSI routing table.
     fn set_gsi(&self) -> Result<()>;
+
+    /// Sets the [`EventFd`] used to trigger interrupts.
+    fn set_notifier(
+        &mut self,
+        _index: InterruptIndex,
+        _eventfd: Option<EventFd>,
+        _vm: &dyn hypervisor::Vm,
+    ) -> Result<()> {
+        Err(Error::new(
+            ErrorKind::Unsupported,
+            "setting notifiers not supported",
+        ))
+    }
 }

vmm/src/device_manager.rs

@@ -4201,8 +4201,8 @@ impl DeviceManager {
             return Err(DeviceManagerError::MissingNode);
         }
 
-        // Allows support for one MSI-X vector per queue. It also adds 1
-        // as we need to take into account the dedicated vector to notify
+        // Allows support for one MSI-X vector per interrupt needed by the device.
+        // It also adds 1 as we need to take into account the dedicated vector to notify
         // about a virtio config change.
         let msix_num = (virtio_device.lock().unwrap().queue_max_sizes().len() + 1) as u16;
 

vmm/src/interrupt.rs

@@ -86,6 +86,9 @@ impl InterruptRoute {
         )
     }
 
+    // This is currently not used, but the upcoming vhost-guest feature
+    // will use it. Use #[allow(dead_code)] to suppress a compiler
+    // warning.
     #[allow(dead_code)]
     pub fn set_notifier(
         &mut self,
@@ -96,7 +99,7 @@ impl InterruptRoute {
         if self.registered {
             if let Some(ref irq_fd) = self.irq_fd {
                 vm.register_irqfd(irq_fd, self.gsi)
-                    .map_err(|e| io::Error::other(format!("Failed registering irq_fd: {e}")))?
+                    .map_err(|e| io::Error::other(format!("Failed registering irq_fd: {e}")))?;
             }
             // If the irqfd cannot be unregistered, what to do?  Spin?
             // Returning an error isn't helpful as the new irqfd is already registered.
@@ -235,6 +238,19 @@ impl InterruptSourceGroup for MsiInterruptGroup {
         let routes = self.gsi_msi_routes.lock().unwrap();
         self.set_gsi_routes(&routes)
     }
+
+    fn set_notifier(
+        &mut self,
+        index: InterruptIndex,
+        eventfd: Option<EventFd>,
+        vm: &dyn hypervisor::Vm,
+    ) -> Result<()> {
+        if let Some(route) = self.irq_routes.get(&index) {
+            return route.lock().unwrap().set_notifier(eventfd, vm);
+        }
+
+        Ok(())
+    }
 }
 
 pub struct LegacyUserspaceInterruptGroup {
@@ -323,6 +339,26 @@ impl InterruptManager for LegacyUserspaceInterruptManager {
     }
 }
 
+impl MsiInterruptManager {
+    fn create_group_raw(
+        &self,
+        config: <Self as InterruptManager>::GroupConfig,
+    ) -> Result<MsiInterruptGroup> {
+        let mut allocator = self.allocator.lock().unwrap();
+        let mut irq_routes: HashMap<InterruptIndex, Mutex<InterruptRoute>> =
+            HashMap::with_capacity(config.count as usize);
+        for i in config.base..config.base + config.count {
+            irq_routes.insert(i, Mutex::new(InterruptRoute::new(&mut allocator)?));
+        }
+
+        Ok(MsiInterruptGroup::new(
+            self.vm.clone(),
+            self.gsi_msi_routes.clone(),
+            irq_routes,
+        ))
+    }
+}
+
 impl InterruptManager for MsiInterruptManager {
     type GroupConfig = MsiIrqGroupConfig;
 
@@ -341,6 +377,14 @@ impl InterruptManager for MsiInterruptManager {
         )))
     }
 
+    fn create_group_mut(
+        &self,
+        config: Self::GroupConfig,
+    ) -> vm_device::interrupt::Result<Arc<Mutex<dyn InterruptSourceGroup>>> {
+        let r = self.create_group_raw(config)?;
+        Ok(Arc::new(Mutex::new(r)))
+    }
+
     fn destroy_group(&self, _group: Arc<dyn InterruptSourceGroup>) -> Result<()> {
         Ok(())
     }