CPU profiles test

chv.nix

@@ -17,7 +17,8 @@ let
   commonArgs = {
     meta = cloud-hypervisor-meta;
 
-    src = craneLib'.cleanCargoSource cloud-hypervisor-src;
+    # src = craneLib'.cleanCargoSource cloud-hypervisor-src;
+    src = cloud-hypervisor-src;
 
     patches =
       let

flake.nix

@@ -8,14 +8,14 @@
     # A local path can be used for developing or testing local changes. Make
     # sure the submodules in a local libvirt checkout are populated.
     libvirt-src = {
-      # url = "git+file:<path/to/libvirt>?submodules=1";
-      url = "git+https://github.com/cyberus-technology/libvirt?ref=gardenlinux&submodules=1";
-      # url = "git+ssh://git@gitlab.cyberus-technology.de/cyberus/cloud/libvirt?ref=managedsave-fix&submodules=1";
+      # url = "git+file:/home/skober/repos/libvirt?submodules=1";
+      url = "git+https://github.com/hertrste/libvirt?ref=cpu-profiles&submodules=1";
+      # url = "git+ssh://git@gitlab.cyberus-technology.de/shertrampf/libvirt.git?ref=cpu-profiles&submodules=1";
       flake = false;
     };
     cloud-hypervisor-src = {
-      # url = "git+file:<path/to/cloud-hypervisor>";
-      url = "github:cyberus-technology/cloud-hypervisor?ref=gardenlinux";
+      # url = "git+file::<path/to/cloud-hypervisor>";
+      url = "github:cyberus-technology/cloud-hypervisor?ref=cyberus-fork-cpu-profiles-cpuid";
       flake = false;
     };
     edk2-src = {

tests/common.nix

@@ -22,6 +22,7 @@ let
       numa ? false,
       hugepages ? false,
       prefault ? false,
+      cpuModel ? "",
       serial ? "pty",
     }:
     ''
@@ -79,6 +80,16 @@ let
             ''
           else
             ''
+              ${
+                if cpuModel == "" then
+                  ""
+                else
+                  ''
+                    <cpu mode='custom' match='exact' check='full'>
+                      <model fallback='forbid'>${cpuModel}</model>
+                    </cpu>
+                  ''
+              }
               <vcpu placement='static'>2</vcpu>
               ${
                 if hugepages then
@@ -421,6 +432,13 @@ in
             })}";
           };
         };
+        "/etc/domain-chv-cpu-sapphire-rapid.xml" = {
+          "C+" = {
+            argument = "${pkgs.writeText "cirros-sapphire-rapid.xml" (virsh_ch_xml {
+              cpuModel = "sapphire-rapids";
+            })}";
+          };
+        };
         "/etc/new_interface.xml" = {
           "C+" = {
             argument = "${pkgs.writeText "new_interface.xml" new_interface}";

tests/default.nix

@@ -25,4 +25,14 @@
       ;
     testScriptFile = ./testscript_long_migration_with_load.py;
   };
+
+  cpu_profiles = import ./libvirt-test.nix {
+    inherit
+      pkgs
+      libvirt-src
+      nixos-image
+      chv-ovmf
+      ;
+    testScriptFile = ./testscript_cpu_profiles.py;
+  };
 }

tests/testscript.py

@@ -1278,10 +1278,32 @@ def test_live_migration_with_vcpu_pinning(self):
         assert int(taskset_vcpu0_controller, 16) == int(taskset_vcpu0_compute, 16)
         assert int(taskset_vcpu2_controller, 16) == int(taskset_vcpu2_compute, 16)
 
+    def test_cpu_models(self):
+        """
+        This tests checks that cpu-models API call is implemented and returns
+        at least a sapphire-rapids model.
+        Further, we check that the domcapabilities API call returns the
+        expected CPU profile as usable.
+        Both is required to be able to use the specific CPU profile.
+        While the 'virsh cpu-models' call only lists the CPU profiles the VMM
+        supports, the 'virsh domcapabilities' call takes into account the hosts
+        architecture. Thus, the latter reports what CPU profile actually can be
+        used in the current environment.
+        """
+        out = controllerVM.succeed("virsh cpu-models x86_64")
+        self.assertIn("sapphire-rapids", out)
+
+        out = controllerVM.succeed("virsh domcapabilities")
+        self.assertIn(
+            "<model usable='yes' vendor='Intel' canonical='sapphire-rapids'>sapphire-rapids</model>",
+            out,
+        )
+
 
 def suite():
     # Test cases in alphabetical order
     testcases = [
+        LibvirtTests.test_cpu_models,
         LibvirtTests.test_disk_is_locked,
         LibvirtTests.test_disk_resize_qcow2,
         LibvirtTests.test_disk_resize_raw,

tests/testscript_cpu_profiles.py

@@ -0,0 +1,133 @@
+import time
+import unittest
+
+# Following is required to allow proper linting of the python code in IDEs.
+# Because certain functions like start_all() and certain objects like computeVM
+# or other machines are added by Nix, we need to provide certain stub objects
+# in order to allow the IDE to lint the python code successfully.
+if "start_all" not in globals():
+    from nixos_test_stubs import start_all, computeVM, controllerVM  # type: ignore
+
+
+class LibvirtTests(unittest.TestCase):
+    @classmethod
+    def setUpClass(cls):
+        start_all()
+        controllerVM.wait_for_unit("multi-user.target")
+        computeVM.wait_for_unit("multi-user.target")
+        controllerVM.succeed("cp /etc/nixos.img /nfs-root/")
+        controllerVM.succeed("chmod 0666 /nfs-root/nixos.img")
+
+        controllerVM.succeed(
+            'virt-admin -c virtchd:///system daemon-log-outputs "2:journald 1:file:/var/log/libvirt/libvirtd.log"'
+        )
+        controllerVM.succeed(
+            "virt-admin -c virtchd:///system daemon-timeout --timeout 0"
+        )
+
+        computeVM.succeed(
+            'virt-admin -c virtchd:///system daemon-log-outputs "2:journald 1:file:/var/log/libvirt/libvirtd.log"'
+        )
+        computeVM.succeed("virt-admin -c virtchd:///system daemon-timeout --timeout 0")
+
+        controllerVM.succeed("mkdir -p /var/lib/libvirt/storage-pools/nfs-share")
+        computeVM.succeed("mkdir -p /var/lib/libvirt/storage-pools/nfs-share")
+
+        controllerVM.succeed("ssh -o StrictHostKeyChecking=no computeVM echo")
+        computeVM.succeed("ssh -o StrictHostKeyChecking=no controllerVM echo")
+
+        controllerVM.succeed(
+            'virsh pool-define-as --name "nfs-share" --type netfs --source-host "localhost" --source-path "nfs-root" --source-format "nfs" --target "/var/lib/libvirt/storage-pools/nfs-share"'
+        )
+        controllerVM.succeed("virsh pool-start nfs-share")
+
+        computeVM.succeed(
+            'virsh pool-define-as --name "nfs-share" --type netfs --source-host "controllerVM" --source-path "nfs-root" --source-format "nfs" --target "/var/lib/libvirt/storage-pools/nfs-share"'
+        )
+        computeVM.succeed("virsh pool-start nfs-share")
+
+    def setUp(self):
+        print(f"\n\nRunning test: {self._testMethodName}\n\n")
+
+    def tearDown(self):
+        # Destroy and undefine all running and persistent domains
+        controllerVM.execute(
+            'virsh list --name | while read domain; do [[ -n "$domain" ]] && virsh destroy "$domain"; done'
+        )
+        controllerVM.execute(
+            'virsh list --all --name | while read domain; do [[ -n "$domain" ]] && virsh undefine "$domain"; done'
+        )
+        computeVM.execute(
+            'virsh list --name | while read domain; do [[ -n "$domain" ]] && virsh destroy "$domain"; done'
+        )
+        computeVM.execute(
+            'virsh list --all --name | while read domain; do [[ -n "$domain" ]] && virsh undefine "$domain"; done'
+        )
+
+        # After undefining and destroying all domains, there should not be any .xml files left
+        # Any files left here, indicate that we do not clean up properly
+        controllerVM.fail("find /run/libvirt/ch -name *.xml | grep .")
+        controllerVM.fail("find /var/lib/libvirt/ch -name *.xml | grep .")
+        computeVM.fail("find /run/libvirt/ch -name *.xml | grep .")
+        computeVM.fail("find /var/lib/libvirt/ch -name *.xml | grep .")
+
+        # Destroy any remaining huge page allocations.
+        controllerVM.succeed("echo 0 > /proc/sys/vm/nr_hugepages")
+        computeVM.succeed("echo 0 > /proc/sys/vm/nr_hugepages")
+
+        # Remove any remaining vm logs.
+        controllerVM.succeed("rm -f /tmp/*.log")
+        computeVM.succeed("rm -f /tmp/*.log")
+
+    def test_live_cpu_profile_sapphire_rapids(self):
+        """
+        Test using the Sapphire Rapids CPU profile for live migration works as expected.
+        """
+
+        controllerVM.succeed("virsh define /etc/domain-chv-cpu-sapphire-rapid.xml")
+        controllerVM.succeed("virsh start testvm")
+
+        assert wait_for_ssh(controllerVM)
+
+        run_loops = 500
+        for i in range(run_loops):
+            print(f"Run {i + 1}/{run_loops}")
+
+            controllerVM.succeed(
+                "virsh migrate --domain testvm --desturi ch+tcp://computeVM/session --persistent --live --p2p"
+            )
+            assert wait_for_ssh(computeVM)
+
+            computeVM.succeed(
+                "virsh migrate --domain testvm --desturi ch+tcp://controllerVM/session --persistent --live --p2p"
+            )
+            assert wait_for_ssh(controllerVM)
+
+
+def suite():
+    suite = unittest.TestSuite()
+    suite.addTest(LibvirtTests("test_live_cpu_profile_sapphire_rapids"))
+    return suite
+
+
+def wait_for_ssh(machine, user="root", password="root", ip="192.168.1.2"):
+    retries = 100
+    for i in range(retries):
+        print(f"Wait for ssh {i}/{retries}")
+        status, _ = ssh(machine, "echo hello", user, password, ip="192.168.1.2")
+        if status == 0:
+            return True
+        time.sleep(1)
+    return False
+
+
+def ssh(machine, cmd, user="root", password="root", ip="192.168.1.2"):
+    status, out = machine.execute(
+        f"sshpass -p {password} ssh -o UserKnownHostsFile=/dev/null -o StrictHostKeyChecking=no {user}@{ip} {cmd}"
+    )
+    return status, out
+
+
+runner = unittest.TextTestRunner()
+if not runner.run(suite()).wasSuccessful():
+    raise Exception("Test Run unsuccessful")