From e5c01c169fb0102022ee7a570c748cb0164ac1bb Mon Sep 17 00:00:00 2001 From: Julia Graham Date: Mon, 23 Sep 2024 17:11:27 -0400 Subject: [PATCH] Max_vcpus_topology: ensure that the VM gets correct topology with maximum vcpus --- libvirt/tests/cfg/cpu/vcpu_max_topology.cfg | 27 +++ libvirt/tests/src/cpu/vcpu_max_topology.py | 209 ++++++++++++++++++++ 2 files changed, 236 insertions(+) create mode 100644 libvirt/tests/cfg/cpu/vcpu_max_topology.cfg create mode 100644 libvirt/tests/src/cpu/vcpu_max_topology.py diff --git a/libvirt/tests/cfg/cpu/vcpu_max_topology.cfg b/libvirt/tests/cfg/cpu/vcpu_max_topology.cfg new file mode 100644 index 0000000000..0d751caf39 --- /dev/null +++ b/libvirt/tests/cfg/cpu/vcpu_max_topology.cfg @@ -0,0 +1,27 @@ +- vcpu.max_topology: + type = vcpu_max_topology + vcpus_placement = "static" + sockets = "" + cores = "" + clusters = "" + variants: + - one_socket: + sockets = "one" + cores = "many" + variants: + - default_clusters: + clusters = "" + - many_clusters: + clusters = "many" + - one_core_per_socket: + sockets = "many" + cores = "one" + clusters = "" + - many_cores_per_socket: + sockets = "many" + cores = "many" + variants: + - default_clusters: + clusters = "" + - many_clusters: + clusters = "many" diff --git a/libvirt/tests/src/cpu/vcpu_max_topology.py b/libvirt/tests/src/cpu/vcpu_max_topology.py new file mode 100644 index 0000000000..0a80e90a13 --- /dev/null +++ b/libvirt/tests/src/cpu/vcpu_max_topology.py @@ -0,0 +1,209 @@ +import logging as log + + +from virttest import virt_vm +from virttest import cpu as cpuutil +from virttest.libvirt_xml import vm_xml + + +# Using as lower capital is not the best way to do, but this is just a +# workaround to avoid changing the entire file. +logging = log.getLogger('avocado.' + __name__) + + +def run(test, params, env): + """ + Test that the vm can start with vcpus which is equal to host online cpu number + and vm topology is consistent to those configured. + + Steps: + 1. Configure the vm topology with specified number of sockets, cores, and clusters + 2. Start configured vm with guest agent + 3. Check that the vm setup is consistent with the topology configured + 3a. Check the lscpu output + 3b. Check the kernel file for core id for each vcpu + 3c. Check the vcpu cluster number + 3d. Check the cluster cpu list + """ + sockets_param = params.get("sockets", "") + cores_param = params.get("cores", "") + clusters_param = params.get("clusters", "") + + vcpus_num = 0 + sockets_list = [] + cores_list = [] + clusters_list = [] + + # Set vcpus_num to the host online cpu number + host_cpu_info = cpuutil.get_cpu_info() + host_online_cpus = int(host_cpu_info["On-line CPU(s) list"].split("-")[1]) + 1 + logging.debug("Host online CPU number: %s", str(host_online_cpus)) + vcpus_num = host_online_cpus + + # Setting number of sockets, cores, and clusters + + # one_socket case + # cores = vcpus_num // number of clusters + if sockets_param == "one": + sockets_list = [1] + + # many_clusters case + if clusters_param == "many": + # Ensure that vcpus_num is evenly divisible by the number of clusters + clusters_list = [clusters for clusters in [2, 4, 6] if vcpus_num % clusters == 0] + cores_list = [vcpus_num // clusters for clusters in clusters_list] + # default_clusters case + else: + cores_list = [vcpus_num] + + # one_core_per_socket case + elif sockets_param == "many" and cores_param == "one": + sockets_list = [vcpus_num] + cores_list = [1] + + # many_cores_per_socket + # Ensure that vcpus_num is evenly divisible by the number of cores + else: + # many_clusters case + # sockets = vcpus_num // number of cores // number of clusters + if clusters_param == "many": + # defaulting to either 2 or 3 cores + cores = 2 if vcpus_num % 2 == 0 else 3 + clusters_list = [clusters for clusters in [2, 4, 6] if (vcpus_num / cores) % clusters == 0] + cores_list = [cores] * (len(clusters_list)) + sockets_list = [(vcpus_num // cores) // clusters for clusters in clusters_list] + # default_clusters case + # sockets * cores = vcpus_num + else: + cores_list = [cores for cores in [2, 4, 6] if vcpus_num % cores == 0] + sockets_list = [vcpus_num // cores for cores in cores_list] + + if not sockets_list or not cores_list: + test.error("The number of sockets or cores is not valid") + elif (len(cores_list) == 1): + # len(sockets_list) will also be 1 + set_and_check_topology(test, params, env, vcpus_num, sockets_list[0], cores_list[0]) + else: + for i, cores in enumerate(cores_list): + if (cores == 0): + continue + if (len(sockets_list) == 1): + if clusters_list: + set_and_check_topology(test, params, env, vcpus_num, sockets_list[0], cores, clusters_list[i]) + else: + set_and_check_topology(test, params, env, vcpus_num, sockets_list[0], cores) + else: + if (sockets_list[i] == 0): + continue + if clusters_list: + set_and_check_topology(test, params, env, vcpus_num, sockets_list[i], cores, clusters_list[i]) + else: + set_and_check_topology(test, params, env, vcpus_num, sockets_list[i], cores) + + +def set_and_check_topology(test, params, env, vcpus_num, sockets, cores, clusters=1): + ''' + Perform steps 2-3 for each vm topology configuration + ''' + vm_name = params.get("main_vm") + vm = env.get_vm(vm_name) + vcpus_placement = params.get("vcpus_placement", "static") + + # Back up domain XML + vmxml = vm_xml.VMXML.new_from_inactive_dumpxml(vm_name) + vmxml_backup = vmxml.copy() + + try: + # Modify vm + if vm.is_alive(): + vm.destroy() + vmxml.placement = vcpus_placement + + if clusters > 1: + # Set manually + vmcpu_xml = vmxml["cpu"] + vmcpu_xml["topology"] = { + "sockets": str(sockets), + "clusters": str(clusters), + "cores": str(cores), + "threads": "1"} + vmxml["cpu"] = vmcpu_xml + vmxml["vcpu"] = vcpus_num + vmxml.sync() + + else: + # Set using set_vm_vcpus helper function + vmxml.set_vm_vcpus( + vm_name, + vcpus_num, + sockets=sockets, + cores=cores, + threads=1, + add_topology=True + ) + logging.debug("Defined guest with '%s' vcpu(s), '%s' socket(s), and '%s' core(s), and '%s' cluster(s)", + str(vcpus_num), str(sockets), str(cores), str(clusters)) + + # Start guest agent in vm + try: + vm.prepare_guest_agent() + except virt_vm.VMStartError as info: + if "not supported" in str(info).lower(): + test.cancel(info) + else: + test.error(info) + + # Check lscpu output within the vm is consistent with the topology configured + + # Check the topology in the guest + session = vm.wait_for_login() + lscpu_output = cpuutil.get_cpu_info(session) + # get_cpu_info() should close the session + session.close() + + lscpu_check_fail = "The configured topology is not consistent with the lscpu output within the vm for " + if (str(vcpus_num) != lscpu_output["CPU(s)"]): + test.fail(lscpu_check_fail + "CPU(s)") + elif (('0' + '-' + str(vcpus_num - 1)) != lscpu_output["On-line CPU(s) list"]): + test.fail(lscpu_check_fail + "on-line CPU(s) list") + elif ("1" != lscpu_output["Thread(s) per core"]): + test.fail(lscpu_check_fail + "thread(s) per core") + elif (str(sockets) != lscpu_output["Socket(s)"]): + test.fail(lscpu_check_fail + "socket(s)") + elif (str(cores * clusters) != lscpu_output["Core(s) per socket"]): + test.fail(lscpu_check_fail + "core(s) per socket") + + session = vm.wait_for_login() + + # Check kernel file for core id for each vcpu in the vm + + for vcpu in range(vcpus_num): + cmd_coreid = f'cat /sys/devices/system/cpu/cpu{vcpu}/topology/core_id' + ret_coreid = session.cmd_output(cmd_coreid, timeout=100).strip() + if (str(vcpu) != ret_coreid): + test.fail("In the vm kernel file, the core id for vcpu %s should not be %s" % (vcpu, ret_coreid)) + + # Check vcpu cluster number + + cmd_clusterid = 'cat /sys/devices/system/cpu/cpu*/topology/cluster_id | sort | uniq -c | wc -l' + ret_clusterid = session.cmd_output(cmd_clusterid, timeout=100).strip() + # The result should be equal to defined clusters' value + if (str(sockets * clusters) != ret_clusterid): + test.fail("In the vm kernel file, the vcpu cluster number should be %s, not %s" % (str(sockets * clusters), ret_clusterid)) + + # Check cluster cpu list + + cmd_cluster_cpu_list = 'cat /sys/devices/system/cpu/cpu*/topology/cluster_cpus_list | sort | uniq -c | wc -l' + ret_cluster_cpu_list = session.cmd_output(cmd_cluster_cpu_list, timeout=100).strip() + # The result should be equal to #sockets * #clusters + if (str(sockets * clusters) != ret_cluster_cpu_list): + test.fail("In the vm kernel file, the cluster cpu list should be %s, not %s" % (str(sockets * clusters), ret_cluster_cpu_list)) + + session.close() + + finally: + # Recover VM + if vm.is_alive(): + vm.destroy(gracefully=False) + logging.info("Restoring vm...") + vmxml_backup.sync()