hypervisor, vmm: Add dynamic control of logging dirty pages

This patch extends slightly the current live-migration code path with
the ability to dynamically start and stop logging dirty-pages, which
relies on two new methods added to the `hypervisor::vm::Vm` Trait. This
patch also contains a complete implementation of the two new methods
based on `kvm` and placeholders for `mshv` in the `hypervisor` crate.

Fixes: #2858

Signed-off-by: Bo Chen <chen.bo@intel.com>

hypervisor/src/kvm/mod.rs

@@ -383,6 +383,59 @@ impl vm::Vm for KvmVm {
         Ok(())
     }
 
+    ///
+    /// Start logging dirty pages
+    ///
+    fn start_dirty_log(
+        &self,
+        slot: u32,
+        guest_phys_addr: u64,
+        memory_size: u64,
+        userspace_addr: u64,
+    ) -> vm::Result<()> {
+        let region = self.make_user_memory_region(
+            slot,
+            guest_phys_addr,
+            memory_size,
+            userspace_addr,
+            false,
+            true,
+        );
+        // Safe because guest regions are guaranteed not to overlap.
+        unsafe {
+            self.fd
+                .set_user_memory_region(region)
+                .map_err(|e| vm::HypervisorVmError::StartDirtyLog(e.into()))
+        }
+    }
+
+    ///
+    /// Stop logging dirty pages
+    ///
+    fn stop_dirty_log(
+        &self,
+        slot: u32,
+        guest_phys_addr: u64,
+        memory_size: u64,
+        userspace_addr: u64,
+    ) -> vm::Result<()> {
+        let region = self.make_user_memory_region(
+            slot,
+            guest_phys_addr,
+            memory_size,
+            userspace_addr,
+            false,
+            false,
+        );
+
+        // Safe because guest regions are guaranteed not to overlap.
+        unsafe {
+            self.fd
+                .set_user_memory_region(region)
+                .map_err(|e| vm::HypervisorVmError::StopDirtyLog(e.into()))
+        }
+    }
+
     ///
     /// Get dirty pages bitmap (one bit per page)
     ///

hypervisor/src/mshv/mod.rs

@@ -877,6 +877,34 @@ impl vm::Vm for MshvVm {
         Ok(())
     }
     ///
+    /// Start logging dirty pages
+    ///
+    fn start_dirty_log(
+        &self,
+        _slot: u32,
+        _guest_phys_addr: u64,
+        _memory_size: u64,
+        _userspace_addr: u64,
+    ) -> vm::Result<()> {
+        Err(vm::HypervisorVmError::StartDirtyLog(anyhow!(
+            "functionality not implemented"
+        )))
+    }
+    ///
+    /// Stop logging dirty pages
+    ///
+    fn stop_dirty_log(
+        &self,
+        _slot: u32,
+        _guest_phys_addr: u64,
+        _memory_size: u64,
+        _userspace_addr: u64,
+    ) -> vm::Result<()> {
+        Err(vm::HypervisorVmError::StopDirtyLog(anyhow!(
+            "functionality not implemented"
+        )))
+    }
+    ///
     /// Get dirty pages bitmap (one bit per page)
     ///
     fn get_dirty_log(&self, _slot: u32, _memory_size: u64) -> vm::Result<Vec<u64>> {

hypervisor/src/vm.rs

@@ -168,6 +168,16 @@ pub enum HypervisorVmError {
     #[error("Failed to write to IO Bus: {0}")]
     IoBusWrite(#[source] anyhow::Error),
     ///
+    /// Start dirty log error
+    ///
+    #[error("Failed to get dirty log: {0}")]
+    StartDirtyLog(#[source] anyhow::Error),
+    ///
+    /// Stop dirty log error
+    ///
+    #[error("Failed to get dirty log: {0}")]
+    StopDirtyLog(#[source] anyhow::Error),
+    ///
     /// Get dirty log error
     ///
     #[error("Failed to get dirty log: {0}")]
@@ -270,6 +280,22 @@ pub trait Vm: Send + Sync {
     fn state(&self) -> Result<VmState>;
     /// Set the VM state
     fn set_state(&self, state: VmState) -> Result<()>;
+    /// Start logging dirty pages
+    fn start_dirty_log(
+        &self,
+        slot: u32,
+        guest_phys_addr: u64,
+        memory_size: u64,
+        userspace_addr: u64,
+    ) -> Result<()>;
+    /// Stop logging dirty pages
+    fn stop_dirty_log(
+        &self,
+        slot: u32,
+        guest_phys_addr: u64,
+        memory_size: u64,
+        userspace_addr: u64,
+    ) -> Result<()>;
     /// Get dirty pages bitmap
     fn get_dirty_log(&self, slot: u32, memory_size: u64) -> Result<Vec<u64>>;
     #[cfg(feature = "tdx")]

vmm/src/lib.rs

@@ -1049,6 +1049,9 @@ impl Vmm {
             // Send last batch of dirty pages
             Self::vm_maybe_send_dirty_pages(vm, &mut socket)?;
 
+            // Stop logging dirty pages
+            vm.stop_memory_dirty_log()?;
+
             // Capture snapshot and send it
             let vm_snapshot = vm.snapshot()?;
             let snapshot_data = serde_json::to_vec(&vm_snapshot).unwrap();

vmm/src/memory_manager.rs

@@ -1561,13 +1561,21 @@ impl MemoryManager {
         Ok(table)
     }
 
-    // The dirty log is cleared by the kernel by calling the KVM_GET_DIRTY_LOG ioctl.
-    // Just before we do a bulk copy we want to clear the dirty log so that
+    // Start the dirty log on guest RAM in the hypervisor (kvm/mshv).
+    // Also, reset the dirty bitmap logged by the vmm.
+    // Just before we do a bulk copy we want to start/clear the dirty log so that
     // pages touched during our bulk copy are tracked.
     pub fn start_memory_dirty_log(&self) -> std::result::Result<(), MigratableError> {
         for r in &self.guest_ram_mappings {
-            self.vm.get_dirty_log(r.slot, r.size).map_err(|e| {
-                MigratableError::MigrateSend(anyhow!("Error getting VM dirty log {}", e))
+            let user_addr = self
+                .guest_memory()
+                .memory()
+                .get_host_address(GuestAddress(r.gpa))
+                .unwrap();
+            self.vm
+                .start_dirty_log(r.slot, r.gpa, r.size, user_addr as u64)
+                .map_err(|e| {
+                    MigratableError::MigrateSend(anyhow!("Error starting VM dirty log {}", e))
                 })?;
         }
 
@@ -1577,6 +1585,23 @@ impl MemoryManager {
 
         Ok(())
     }
+
+    pub fn stop_memory_dirty_log(&self) -> std::result::Result<(), MigratableError> {
+        for r in &self.guest_ram_mappings {
+            let user_addr = self
+                .guest_memory()
+                .memory()
+                .get_host_address(GuestAddress(r.gpa))
+                .unwrap();
+            self.vm
+                .stop_dirty_log(r.slot, r.gpa, r.size, user_addr as u64)
+                .map_err(|e| {
+                    MigratableError::MigrateSend(anyhow!("Error stopping VM dirty log {}", e))
+                })?;
+        }
+
+        Ok(())
+    }
 }
 
 #[cfg(feature = "acpi")]

vmm/src/vm.rs

@@ -2156,6 +2156,10 @@ impl Vm {
         self.memory_manager.lock().unwrap().start_memory_dirty_log()
     }
 
+    pub fn stop_memory_dirty_log(&self) -> std::result::Result<(), MigratableError> {
+        self.memory_manager.lock().unwrap().stop_memory_dirty_log()
+    }
+
     pub fn dirty_memory_range_table(
         &self,
     ) -> std::result::Result<MemoryRangeTable, MigratableError> {