vfio: fix memory slot over KVM_USER_MEM_SLOTS

When adding vfio passthrough devices cloud hypervisor allocate a new
memory slot in map_mmio_regions. As we +1 each time as new memory slot,
it will fail as the memory slot over KVM_USER_MEM_SLOTS in kernel when
calling KVM_SET_USER_MEMORY_REGION ioctl.

This patch introduces a vector to save the memory slots allocated, and
return the minimal available memory slot to caller to fix this issue.

Signed-off-by: Yi Wang <foxywang@tencent.com>

pci/src/vfio.rs

@@ -1416,7 +1416,8 @@ pub struct VfioPciDevice {
     container: Arc<VfioContainer>,
     common: VfioCommon,
     iommu_attached: bool,
-    memory_slot: Arc<dyn Fn() -> u32 + Send + Sync>,
+    get_memory_slot: Arc<dyn Fn() -> u32 + Send + Sync>,
+    put_memory_slot: Arc<dyn Fn(u32) + Send + Sync>,
 }
 
 impl VfioPciDevice {
@@ -1431,7 +1432,8 @@ impl VfioPciDevice {
         legacy_interrupt_group: Option<Arc<dyn InterruptSourceGroup>>,
         iommu_attached: bool,
         bdf: PciBdf,
-        memory_slot: Arc<dyn Fn() -> u32 + Send + Sync>,
+        get_memory_slot: Arc<dyn Fn() -> u32 + Send + Sync>,
+        put_memory_slot: Arc<dyn Fn(u32) + Send + Sync>,
         snapshot: Option<Snapshot>,
         x_nv_gpudirect_clique: Option<u8>,
     ) -> Result<Self, VfioPciError> {
@@ -1457,7 +1459,8 @@ impl VfioPciDevice {
             container,
             common,
             iommu_attached,
-            memory_slot,
+            get_memory_slot,
+            put_memory_slot,
         };
 
         Ok(vfio_pci_device)
@@ -1635,7 +1638,7 @@ impl VfioPciDevice {
                     }
 
                     let user_memory_region = UserMemoryRegion {
-                        slot: (self.memory_slot)(),
+                        slot: (self.get_memory_slot)(),
                         start: region.start.0 + area.offset,
                         size: area.size,
                         host_addr: host_addr as u64,
@@ -1699,6 +1702,8 @@ impl VfioPciDevice {
                     error!("Could not remove the userspace memory region: {}", e);
                 }
 
+                (self.put_memory_slot)(user_memory_region.slot);
+
                 // SAFETY: FFI call with correct arguments
                 let ret = unsafe {
                     libc::munmap(

vmm/src/device_manager.rs

@@ -3422,6 +3422,7 @@ impl DeviceManager {
             };
 
         let memory_manager = self.memory_manager.clone();
+        let memory_manager2 = self.memory_manager.clone();
 
         let vfio_pci_device = VfioPciDevice::new(
             vfio_name.clone(),
@@ -3433,6 +3434,7 @@ impl DeviceManager {
             device_cfg.iommu,
             pci_device_bdf,
             Arc::new(move || memory_manager.lock().unwrap().allocate_memory_slot()),
+            Arc::new(move |x| memory_manager2.lock().unwrap().free_memory_slot(x)),
             vm_migration::snapshot_from_id(self.snapshot.as_ref(), vfio_name.as_str()),
             device_cfg.x_nv_gpudirect_clique,
         )

vmm/src/memory_manager.rs

@@ -163,6 +163,7 @@ pub struct MemoryManager {
     boot_guest_memory: GuestMemoryMmap,
     guest_memory: GuestMemoryAtomic<GuestMemoryMmap>,
     next_memory_slot: u32,
+    memory_slots: Vec<u32>,
     start_of_device_area: GuestAddress,
     end_of_device_area: GuestAddress,
     end_of_ram_area: GuestAddress,
@@ -1206,6 +1207,7 @@ impl MemoryManager {
             boot_guest_memory,
             guest_memory,
             next_memory_slot,
+            memory_slots: Vec::new(),
             start_of_device_area,
             end_of_device_area,
             end_of_ram_area,
@@ -1729,12 +1731,29 @@ impl MemoryManager {
         self.end_of_device_area
     }
 
+    pub fn get_available_memory_slot(&self) -> u32 {
+        for i in &self.memory_slots {
+            if !self.memory_slots.contains(&(i + 1)) {
+                return *i + 1;
+            }
+        }
+
+        *self.memory_slots.last().unwrap_or(&0)
+    }
+
     pub fn allocate_memory_slot(&mut self) -> u32 {
-        let slot_id = self.next_memory_slot;
-        self.next_memory_slot += 1;
+        let slot_id = self.get_available_memory_slot();
+        debug!("got memory slot id {}", slot_id);
+        self.memory_slots.push(slot_id);
+        self.memory_slots.sort();
         slot_id
     }
 
+    pub fn free_memory_slot(&mut self, slot: u32) {
+        debug!("free memory slot {:?}", slot);
+        self.memory_slots.retain(|&x| x != slot);
+    }
+
     pub fn create_userspace_mapping(
         &mut self,
         guest_phys_addr: u64,