iommu skip bridge devices

pkg/capacity/pci.go

@@ -214,8 +214,20 @@ func GPU(p *PCI) bool {
 	return p.Class == 0x030000
 }
 
+func PCIBridge(p *PCI) bool {
+	// this will include 0x060000 and 0x060400
+	// or any other pci bridge device
+	return p.Class>>16 == 0x06
+}
+
+func Not(f Filter) Filter {
+	return func(pci *PCI) bool {
+		return !f(pci)
+	}
+}
+
 // given a pci device, return all devices in the same iommu group
-func IoMMUGroup(pci PCI) ([]PCI, error) {
+func IoMMUGroup(pci PCI, filter ...Filter) ([]PCI, error) {
 	path := filepath.Join(pciDir, pci.Slot, "iommu_group", "devices")
 	entries, err := os.ReadDir(path)
 	if os.IsNotExist(err) {
@@ -226,11 +238,17 @@ func IoMMUGroup(pci PCI) ([]PCI, error) {
 	}
 
 	var devices []PCI
+next:
 	for _, entry := range entries {
 		pci, err := pciDeviceFromSlot(entry.Name())
 		if err != nil {
 			return nil, err
 		}
+		for _, f := range filter {
+			if !f(&pci) {
+				continue next
+			}
+		}
 
 		devices = append(devices, pci)
 	}

pkg/primitives/vm/gpu.go

@@ -68,17 +68,16 @@ func (m *Manager) initGPUs() error {
 		return errors.Wrap(err, "failed to list system GPUs")
 	}
 
-	disconnectVT := false
 	for _, gpu := range gpus {
 		bootVga, err := gpu.Flag("boot_vga")
 		if err != nil && !os.IsNotExist(err) {
 			return errors.Wrapf(err, "failed to read GPU '%s' boot_vga flag", gpu.Slot)
 		}
 		if bootVga > 0 {
-			disconnectVT = true
+			m.unbindBootVga()
 		}
 
-		devices, err := capacity.IoMMUGroup(gpu)
+		devices, err := capacity.IoMMUGroup(gpu, capacity.Not(capacity.PCIBridge))
 		if err != nil {
 			return errors.Wrapf(err, "failed to list devices in iommu group for '%s'", gpu.Slot)
 		}
@@ -117,10 +116,6 @@ func (m *Manager) initGPUs() error {
 		}
 	}
 
-	if disconnectVT {
-		return m.unbindBootVga()
-	}
-
 	return nil
 }
 
@@ -143,7 +138,7 @@ func (m *Manager) expandGPUs(gpus []zos.GPU) ([]capacity.PCI, error) {
 			return nil, fmt.Errorf("unknown GPU id '%s'", gpu)
 		}
 
-		sub, err := capacity.IoMMUGroup(device)
+		sub, err := capacity.IoMMUGroup(device, capacity.Not(capacity.PCIBridge))
 		if err != nil {
 			return nil, errors.Wrapf(err, "failed to list all devices belonging to '%s'", device.Slot)
 		}