Separate IOMMU and PCI device checks. (#40)

* split PCI tests out of IOMMU

* added to changelog

* rename pci to pci devices

* markups done

* updated tests

* Markups round 2

* line length fix

* line length fix

* Update tests/test_host_check.py

Co-authored-by: tjohnes <tjohnes@cisco.com>

---------

Co-authored-by: tjohnes <tjohnes@cisco.com>

Author: ebeaty-cisco

CHANGELOG.md

@@ -5,6 +5,10 @@
 
 The v1 release supports Cisco IOS-XR release versions from 7.7.1 to 7.10.1.
 
+### v1.1.15 (2024-03-04)
+
+- `host-check` now has separate checks for IOMMU being enabled and PCI device availability.
+
 ### v1.1.14 (2023-11-27)
 
 - `host-check` now summarizes failures and warnings separately. When testing both XR platforms, a failure for either is now treated as a failure overall.

scripts/host-check

@@ -1104,6 +1104,45 @@ def check_iommu() -> CheckFuncReturn:
             f"Unable to check if IOMMU is enabled by listing {iommu_dev_filepath}.\n"
             + recommendation,
         )
+    return (CheckState.SUCCESS, "IOMMU enabled for vfio-pci.")
+
+
+def check_pci_devices() -> CheckFuncReturn:
+    """Check PCI devices are available."""
+    # If only vfio-pci is supported (and it is not in no-IOMMU mode), then
+    # check for PCI network devices in IOMMU group, otherwise check for any
+    # PCI network devices.
+    if (
+        PCIDriver("vfio-pci").is_supported()
+        and not PCIDriver("igb_uio").is_supported()
+    ):
+        require_iommu = True
+        # Overide require_iommu if no-IOMMU mode is enabled
+        noiommu_filepath = (
+            "/sys/module/vfio/parameters/enable_unsafe_noiommu_mode"
+        )
+        try:
+            with open(noiommu_filepath, "r", encoding="utf-8") as f:
+                if f.read().strip().upper() in ("Y", "1"):
+                    require_iommu = False
+        except OSError:
+            # If the no-IOMMU file does not exist, then no-IOMMU mode can't be
+            # enabled.
+            pass
+    else:
+        require_iommu = False
+
+    if require_iommu:
+        # Get IOMMU devices
+        iommu_dev_filepath = "/sys/class/iommu/*/devices/*"
+        try:
+            iommu_devices = [
+                os.path.basename(p) for p in glob.glob(iommu_dev_filepath)
+            ]
+        except Exception:
+            # If getting the directory fails, then assume IOMMU is not enabled
+            iommu_devices = []
+
     # List the network PCI devices
     try:
         cmd = "lshw -businfo -c network"
@@ -1112,8 +1151,6 @@ def check_iommu() -> CheckFuncReturn:
             r"pci@([\da-f]{4}:[\da-f]{2}:[\da-f]{2}\.[\da-f])\s+(\S+)", output
         )
         if not matches:
-            # This should always be a warning - we need PCI devices whether
-            # we're using igb_uio or vfio-pci.
             return CheckState.WARNING, "no PCI network devices found"
         network_devices = set()
         net_devices_dict = {}
@@ -1126,31 +1163,32 @@ def check_iommu() -> CheckFuncReturn:
         return (
             CheckState.ERROR,
             f"The cmd {cmd!r} failed - unable to\n"
-            "determine the network devices on the host. IOMMU is enabled.",
-        )
-    net_iommu_devices = network_devices & set(iommu_devices)
-    if not net_iommu_devices:
-        return (
-            warning_state,
-            "IOMMU enabled for vfio-pci, but no network PCI devices found.",
-        )
-    net_iommu_devs_list = []
-    for device in sorted(net_iommu_devices):
-        net_iommu_devs_list.append(
-            net_devices_dict[device] + " (" + device + ")"
+            "determine the network devices on the host.",
         )
+
+    # If in IOMMU mode then check if the network devices are in the IOMMU group
+    if require_iommu:
+        network_devices &= set(iommu_devices)
+        if not network_devices:
+            return (
+                CheckState.WARNING,
+                "No PCI network devices found with IOMMU enabled for vfio-pci.",
+            )
+
+    net_devs_list = []
+    for device in sorted(network_devices):
+        net_devs_list.append(net_devices_dict[device] + " (" + device + ")")
     dev_output = ""
-    for count, dev in enumerate(net_iommu_devs_list, start=1):
-        if count == len(net_iommu_devs_list):
+    for count, dev in enumerate(net_devs_list, start=1):
+        if count == len(net_devs_list):
             dev_output = dev_output + f"{dev}"
         elif count % 3 != 0:
             dev_output = dev_output + f"{dev}, "
         else:
             dev_output = dev_output + f"{dev},\n"
     return (
         CheckState.SUCCESS,
-        "IOMMU enabled for vfio-pci with the following PCI device(s):\n"
-        + dev_output,
+        "The following PCI device(s) are available:\n" + dev_output,
     )
 
 
@@ -1505,6 +1543,7 @@ VROUTER_CHECKS = [
     Check("Hugepages", check_hugepages, []),
     Check("Interface kernel driver", check_interface_kernel_driver, []),
     Check("IOMMU", check_iommu, ["Interface kernel driver"]),
+    Check("PCI devices", check_pci_devices, ["Interface kernel driver"]),
     Check("Shared memory pages max size", check_shmem_pages_max_size, []),
     Check("Real-time Group Scheduling", check_realtime_group_sched, []),
 ]