Merge "testpmd: Simplify GUEST's testpmd configuration"
[vswitchperf.git] / vnfs / qemu / qemu.py
index f292d7d..67dbfab 100644 (file)
@@ -1,4 +1,4 @@
-# Copyright 2015 Intel Corporation.
+# Copyright 2015-2016 Intel Corporation.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
@@ -12,7 +12,7 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
-"""Automation of QEMU hypervisor for launching vhost-cuse enabled guests.
+"""Automation of QEMU hypervisor for launching guests.
 """
 
 import os
@@ -20,6 +20,8 @@ import logging
 import locale
 import re
 import subprocess
+import time
+import pexpect
 
 from conf import settings as S
 from conf import get_test_param
@@ -30,7 +32,7 @@ class IVnfQemu(IVnf):
     Abstract class for controling an instance of QEMU
     """
     _cmd = None
-    _expect = S.getValue('GUEST_PROMPT_LOGIN')
+    _expect = None
     _proc_name = 'qemu'
 
     class GuestCommandFilter(logging.Filter):
@@ -45,24 +47,57 @@ class IVnfQemu(IVnf):
         Initialisation function.
         """
         super(IVnfQemu, self).__init__()
+
+        self._expect = S.getValue('GUEST_PROMPT_LOGIN')[self._number]
         self._logger = logging.getLogger(__name__)
         self._logfile = os.path.join(
             S.getValue('LOG_DIR'),
             S.getValue('LOG_FILE_QEMU')) + str(self._number)
         self._timeout = S.getValue('GUEST_TIMEOUT')[self._number]
         self._monitor = '%s/vm%dmonitor' % ('/tmp', self._number)
-        self._net1 = S.getValue('VANILLA_NIC1_NAME')[self._number]
-        self._net2 = S.getValue('VANILLA_NIC2_NAME')[self._number]
+        # read GUEST NICs configuration and use only defined NR of NICS
+        nics_nr = S.getValue('GUEST_NICS_NR')[self._number]
+        # and inform user about missconfiguration
+        if nics_nr < 1:
+            raise RuntimeError('At least one VM NIC is mandotory, but {} '
+                               'NICs are configured'.format(nics_nr))
+        elif nics_nr > 1 and nics_nr % 2:
+            nics_nr = int(nics_nr / 2) * 2
+            self._logger.warning('Odd number of NICs is configured, only '
+                                 '%s NICs will be used', nics_nr)
+
+        self._nics = S.getValue('GUEST_NICS')[self._number][:nics_nr]
+
+        # set guest loopback application based on VNF configuration
+        # cli option take precedence to config file values
+        self._guest_loopback = S.getValue('GUEST_LOOPBACK')[self._number]
+
+        self._testpmd_fwd_mode = S.getValue('GUEST_TESTPMD_FWD_MODE')[self._number]
+        # in case of SRIOV we must ensure, that MAC addresses are not swapped
+        if S.getValue('SRIOV_ENABLED') and self._testpmd_fwd_mode.startswith('mac') and \
+           not S.getValue('VNF').endswith('PciPassthrough'):
+
+            self._logger.info("SRIOV detected, forwarding mode of testpmd was changed from '%s' to '%s'",
+                              self._testpmd_fwd_mode, 'io')
+            self._testpmd_fwd_mode = 'io'
+
+        guest_smp = int(get_test_param('guest_smp', 0))
+        if guest_smp:
+            override_list = [guest_smp] * (self._number + 1)
+            S.setValue('GUEST_SMP', override_list)
 
         name = 'Client%d' % self._number
         vnc = ':%d' % self._number
-        # don't use taskset to affinize main qemu process; It causes hangup
-        # of 2nd VM in case of DPDK. It also slows down VM responsivnes.
-        self._cmd = ['sudo', '-E', S.getValue('QEMU_BIN'),
+        # NOTE: affinization of main qemu process can cause hangup of 2nd VM
+        # in case of DPDK usage. It can also slow down VM response time.
+        cpumask = ",".join(S.getValue('GUEST_CORE_BINDING')[self._number])
+        self._cmd = ['sudo', '-E', 'taskset', '-c', cpumask,
+                     S.getValue('TOOLS')['qemu-system'],
                      '-m', S.getValue('GUEST_MEMORY')[self._number],
                      '-smp', str(S.getValue('GUEST_SMP')[self._number]),
-                     '-cpu', 'host',
-                     '-drive', 'if=ide,file=' +
+                     '-cpu', 'host,migratable=off',
+                     '-drive', 'if={},file='.format(S.getValue(
+                         'GUEST_BOOT_DRIVE_TYPE')[self._number]) +
                      S.getValue('GUEST_IMAGE')[self._number],
                      '-boot', 'c', '--enable-kvm',
                      '-monitor', 'unix:%s,server,nowait' % self._monitor,
@@ -74,8 +109,9 @@ class IVnfQemu(IVnf):
                      '-nographic', '-vnc', str(vnc), '-name', name,
                      '-snapshot', '-net none', '-no-reboot',
                      '-drive',
-                     'if=ide,file=fat:rw:%s,snapshot=off' %
-                     S.getValue('GUEST_SHARE_DIR')[self._number],
+                     'if=%s,format=raw,file=fat:rw:%s,snapshot=off' %
+                     (S.getValue('GUEST_SHARED_DRIVE_TYPE')[self._number],
+                      S.getValue('GUEST_SHARE_DIR')[self._number]),
                     ]
         self._configure_logging()
 
@@ -104,9 +140,39 @@ class IVnfQemu(IVnf):
         if S.getValue('VNF_AFFINITIZATION_ON'):
             self._affinitize()
 
+        if S.getValue('VSWITCH_VHOST_NET_AFFINITIZATION') and S.getValue(
+                'VNF') == 'QemuVirtioNet':
+            self._affinitize_vhost_net()
+
         if self._timeout:
             self._config_guest_loopback()
 
+    def stop(self):
+        """
+        Stops VNF instance gracefully first.
+        """
+        try:
+            # exit testpmd if needed
+            if self._guest_loopback == 'testpmd':
+                self.execute_and_wait('stop', 120, "Done")
+                self.execute_and_wait('quit', 120, "[bB]ye")
+
+            # turn off VM
+            self.execute_and_wait('poweroff', 120, "Power down")
+
+        except pexpect.TIMEOUT:
+            self.kill()
+
+        # wait until qemu shutdowns
+        self._logger.debug('Wait for QEMU to terminate')
+        for dummy in range(30):
+            time.sleep(1)
+            if not self.is_running():
+                break
+
+        # just for case that graceful shutdown failed
+        super(IVnfQemu, self).stop()
+
     # helper functions
 
     def _login(self, timeout=120):
@@ -126,11 +192,11 @@ class IVnfQemu(IVnf):
         if not self._timeout:
             self._expect_process(timeout=timeout)
 
-        self._child.sendline(S.getValue('GUEST_USERNAME'))
-        self._child.expect(S.getValue('GUEST_PROMPT_PASSWORD'), timeout=5)
-        self._child.sendline(S.getValue('GUEST_PASSWORD'))
+        self._child.sendline(S.getValue('GUEST_USERNAME')[self._number])
+        self._child.expect(S.getValue('GUEST_PROMPT_PASSWORD')[self._number], timeout=5)
+        self._child.sendline(S.getValue('GUEST_PASSWORD')[self._number])
 
-        self._expect_process(S.getValue('GUEST_PROMPT'), timeout=5)
+        self._expect_process(S.getValue('GUEST_PROMPT')[self._number], timeout=5)
 
     def send_and_pass(self, cmd, timeout=30):
         """
@@ -142,10 +208,10 @@ class IVnfQemu(IVnf):
         :returns: None
         """
         self.execute(cmd)
-        self.wait(S.getValue('GUEST_PROMPT'), timeout=timeout)
+        self.wait(S.getValue('GUEST_PROMPT')[self._number], timeout=timeout)
         self.execute('echo $?')
         self._child.expect('^0$', timeout=1)  # expect a 0
-        self.wait(S.getValue('GUEST_PROMPT'), timeout=timeout)
+        self.wait(S.getValue('GUEST_PROMPT')[self._number], timeout=timeout)
 
     def _affinitize(self):
         """
@@ -171,51 +237,82 @@ class IVnfQemu(IVnf):
             stdin=proc.stdout)
         proc.wait()
 
+        guest_core_binding = int(get_test_param('guest_core_binding', 0))
         for cpu in range(0, int(S.getValue('GUEST_SMP')[self._number])):
             match = None
             for line in output.decode(cur_locale).split('\n'):
                 match = re.search(thread_id % cpu, line)
                 if match:
-                    self._affinitize_pid(
-                        S.getValue('GUEST_CORE_BINDING')[self._number][cpu],
-                        match.group(1))
+                    if guest_core_binding:
+                        self._affinitize_pid(guest_core_binding, match.group(1))
+                    else:
+                        self._affinitize_pid(
+                            S.getValue('GUEST_CORE_BINDING')[self._number][cpu],
+                            match.group(1))
                     break
 
             if not match:
                 self._logger.error('Failed to affinitize guest core #%d. Could'
                                    ' not parse tid.', cpu)
 
+    def _affinitize_vhost_net(self):
+        """
+        Affinitize the vhost net threads for Vanilla OVS and guest nic queues.
+
+        :return: None
+        """
+        self._logger.info('Affinitizing VHOST Net threads.')
+        args1 = ['pgrep', 'vhost-']
+        process1 = subprocess.Popen(args1, stdout=subprocess.PIPE,
+                                    shell=False)
+        out = process1.communicate()[0]
+        processes = out.decode(locale.getdefaultlocale()[1]).split('\n')
+        if processes[-1] == '':
+            processes.pop() # pgrep may return an extra line with no data
+        self._logger.info('Found %s vhost net threads...', len(processes))
+
+        cpumap = S.getValue('VSWITCH_VHOST_CPU_MAP')
+        mapcount = 0
+        for proc in processes:
+            self._affinitize_pid(cpumap[mapcount], proc)
+            mapcount += 1
+            if mapcount + 1 > len(cpumap):
+                # Not enough cpus were given in the mapping to cover all the
+                # threads on a 1 to 1 ratio with cpus so reset the list counter
+                #  to 0.
+                mapcount = 0
+
     def _config_guest_loopback(self):
         """
-        Configure VM to run VNF (e.g. port forwarding application)
+        Configure VM to run VNF, e.g. port forwarding application based on the configuration
         """
-        # set guest loopback application based on VNF configuration
-        # cli option take precedence to config file values
-        guest_loopback = get_test_param('guest_loopback', S.getValue('GUEST_LOOPBACK')[self._number])
-        if guest_loopback == 'testpmd':
+        if self._guest_loopback == 'testpmd':
             self._login()
             self._configure_testpmd()
-        elif guest_loopback == 'l2fwd':
+        elif self._guest_loopback == 'l2fwd':
             self._login()
             self._configure_l2fwd()
-        elif guest_loopback == 'linux_bridge':
+        elif self._guest_loopback == 'linux_bridge':
             self._login()
             self._configure_linux_bridge()
-        elif guest_loopback != 'buildin':
+        elif self._guest_loopback != 'buildin':
             self._logger.error('Unsupported guest loopback method "%s" was specified. Option'
-                               ' "buildin" will be used as a fallback.', guest_loopback)
+                               ' "buildin" will be used as a fallback.', self._guest_loopback)
 
-    def wait(self, prompt=S.getValue('GUEST_PROMPT'), timeout=30):
+    def wait(self, prompt=None, timeout=30):
+        if prompt is None:
+            prompt = S.getValue('GUEST_PROMPT')[self._number]
         super(IVnfQemu, self).wait(prompt=prompt, timeout=timeout)
 
-    def execute_and_wait(self, cmd, timeout=30,
-                         prompt=S.getValue('GUEST_PROMPT')):
+    def execute_and_wait(self, cmd, timeout=30, prompt=None):
+        if prompt is None:
+            prompt = S.getValue('GUEST_PROMPT')[self._number]
         super(IVnfQemu, self).execute_and_wait(cmd, timeout=timeout,
                                                prompt=prompt)
 
     def _modify_dpdk_makefile(self):
         """
-        Modifies DPDK makefile in Guest before compilation
+        Modifies DPDK makefile in Guest before compilation if needed
         """
         pass
 
@@ -224,22 +321,36 @@ class IVnfQemu(IVnf):
         Mount shared directory and copy DPDK and l2fwd sources
         """
         # mount shared directory
-        self.execute_and_wait('umount ' + S.getValue('OVS_DPDK_SHARE'))
-        self.execute_and_wait('rm -rf ' + S.getValue('GUEST_OVS_DPDK_DIR'))
-        self.execute_and_wait('mkdir -p ' + S.getValue('OVS_DPDK_SHARE'))
-        self.execute_and_wait('mount -o iocharset=utf8 /dev/sdb1 ' +
-                              S.getValue('OVS_DPDK_SHARE'))
-        self.execute_and_wait('mkdir -p ' + S.getValue('GUEST_OVS_DPDK_DIR'))
-        self.execute_and_wait('cp -ra ' + os.path.join(S.getValue('OVS_DPDK_SHARE'), dirname) +
-                              ' ' + S.getValue('GUEST_OVS_DPDK_DIR'))
+        self.execute_and_wait('umount /dev/sdb1')
+        self.execute_and_wait('rm -rf ' + S.getValue('GUEST_OVS_DPDK_DIR')[self._number])
+        self.execute_and_wait('mkdir -p ' + S.getValue('GUEST_OVS_DPDK_SHARE')[self._number])
+        self.execute_and_wait('mount -o ro,iocharset=utf8 /dev/sdb1 ' +
+                              S.getValue('GUEST_OVS_DPDK_SHARE')[self._number])
+        self.execute_and_wait('mkdir -p ' + S.getValue('GUEST_OVS_DPDK_DIR')[self._number])
+        self.execute_and_wait('cp -r ' + os.path.join(S.getValue('GUEST_OVS_DPDK_SHARE')[self._number], dirname) +
+                              ' ' + S.getValue('GUEST_OVS_DPDK_DIR')[self._number])
+        self.execute_and_wait('umount /dev/sdb1')
 
     def _configure_disable_firewall(self):
         """
         Disable firewall in VM
         """
-        # Disable services (F16)
-        self.execute_and_wait('systemctl status iptables.service')
-        self.execute_and_wait('systemctl stop iptables.service')
+        for iptables in ['iptables', 'ip6tables']:
+            # filter table
+            for chain in ['INPUT', 'FORWARD', 'OUTPUT']:
+                self.execute_and_wait("{} -t filter -P {} ACCEPT".format(iptables, chain))
+            # mangle table
+            for chain in ['PREROUTING', 'INPUT', 'FORWARD', 'OUTPUT', 'POSTROUTING']:
+                self.execute_and_wait("{} -t mangle -P {} ACCEPT".format(iptables, chain))
+            # nat table
+            for chain in ['PREROUTING', 'INPUT', 'OUTPUT', 'POSTROUTING']:
+                self.execute_and_wait("{} -t nat -P {} ACCEPT".format(iptables, chain))
+
+            # flush rules and delete chains created by user
+            for table in ['filter', 'mangle', 'nat']:
+                self.execute_and_wait("{} -t {} -F".format(iptables, table))
+                self.execute_and_wait("{} -t {} -X".format(iptables, table))
+
 
     def _configure_testpmd(self):
         """
@@ -250,7 +361,7 @@ class IVnfQemu(IVnf):
 
         # Guest images _should_ have 1024 hugepages by default,
         # but just in case:'''
-        self.execute_and_wait('sysctl vm.nr_hugepages=1024')
+        self.execute_and_wait('sysctl vm.nr_hugepages={}'.format(S.getValue('GUEST_HUGEPAGES_NR')[self._number]))
 
         # Mount hugepages
         self.execute_and_wait('mkdir -p /dev/hugepages')
@@ -258,74 +369,97 @@ class IVnfQemu(IVnf):
             'mount -t hugetlbfs hugetlbfs /dev/hugepages')
 
         # build and configure system for dpdk
-        self.execute_and_wait('cd ' + S.getValue('GUEST_OVS_DPDK_DIR') +
+        self.execute_and_wait('cd ' + S.getValue('GUEST_OVS_DPDK_DIR')[self._number] +
                               '/DPDK')
         self.execute_and_wait('export CC=gcc')
         self.execute_and_wait('export RTE_SDK=' +
-                              S.getValue('GUEST_OVS_DPDK_DIR') + '/DPDK')
+                              S.getValue('GUEST_OVS_DPDK_DIR')[self._number] + '/DPDK')
         self.execute_and_wait('export RTE_TARGET=%s' % S.getValue('RTE_TARGET'))
 
         # modify makefile if needed
         self._modify_dpdk_makefile()
 
+        # disable network interfaces, so DPDK can take care of them
+        for nic in self._nics:
+            self.execute_and_wait('ifdown ' + nic['device'])
+
+        # build and insert igb_uio and rebind interfaces to it
         self.execute_and_wait('make RTE_OUTPUT=$RTE_SDK/$RTE_TARGET -C '
                               '$RTE_SDK/lib/librte_eal/linuxapp/igb_uio')
         self.execute_and_wait('modprobe uio')
         self.execute_and_wait('insmod %s/kmod/igb_uio.ko' %
                               S.getValue('RTE_TARGET'))
-        self.execute_and_wait('./tools/dpdk_nic_bind.py --status')
-        self.execute_and_wait(
-            './tools/dpdk_nic_bind.py -b igb_uio' ' ' +
-            S.getValue('GUEST_NET1_PCI_ADDRESS')[self._number] + ' ' +
-            S.getValue('GUEST_NET2_PCI_ADDRESS')[self._number])
+        self.execute_and_wait('./tools/dpdk*bind.py --status')
+        pci_list = ' '.join([nic['pci'] for nic in self._nics])
+        self.execute_and_wait('./tools/dpdk*bind.py -u ' + pci_list)
+        self.execute_and_wait('./tools/dpdk*bind.py -b igb_uio ' + pci_list)
+        self.execute_and_wait('./tools/dpdk*bind.py --status')
 
         # build and run 'test-pmd'
-        self.execute_and_wait('cd ' + S.getValue('GUEST_OVS_DPDK_DIR') +
+        self.execute_and_wait('cd ' + S.getValue('GUEST_OVS_DPDK_DIR')[self._number] +
                               '/DPDK/app/test-pmd')
         self.execute_and_wait('make clean')
         self.execute_and_wait('make')
-        self.execute_and_wait('./testpmd -c 0x3 -n 4 --socket-mem 512 --'
-                              ' --burst=64 -i --txqflags=0xf00 ' +
-                              '--disable-hw-vlan', 60, "Done")
-        self.execute('set fwd mac_retry', 1)
-        self.execute_and_wait('start', 20,
-                              'TX RS bit threshold=0 - TXQ flags=0xf00')
+
+        # get testpmd settings from CLI
+        testpmd_params = get_test_param('guest_testpmd_params',
+                                        S.getValue('GUEST_TESTPMD_PARAMS')[self._number])
+
+        self.execute_and_wait( './testpmd {}'.format(testpmd_params), 60, "Done")
+        self.execute('set fwd ' + self._testpmd_fwd_mode, 1)
+        self.execute_and_wait('start', 20, 'testpmd>')
 
     def _configure_l2fwd(self):
         """
         Configure VM to perform L2 forwarding between NICs by l2fwd module
         """
+        if int(S.getValue('GUEST_NIC_QUEUES')[self._number]):
+            self._set_multi_queue_nic()
         self._configure_copy_sources('l2fwd')
         self._configure_disable_firewall()
 
+        # configure all interfaces
+        for nic in self._nics:
+            self.execute('ip addr add ' +
+                         nic['ip'] + ' dev ' + nic['device'])
+            self.execute('ip link set dev ' + nic['device'] + ' up')
+
         # build and configure system for l2fwd
-        self.execute_and_wait('cd ' + S.getValue('GUEST_OVS_DPDK_DIR') +
+        self.execute_and_wait('cd ' + S.getValue('GUEST_OVS_DPDK_DIR')[self._number] +
                               '/l2fwd')
         self.execute_and_wait('export CC=gcc')
 
         self.execute_and_wait('make')
-        self.execute_and_wait('insmod ' + S.getValue('GUEST_OVS_DPDK_DIR') +
-                              '/l2fwd' + '/l2fwd.ko net1=' + self._net1 +
-                              ' net2=' + self._net2)
+        if len(self._nics) == 2:
+            self.execute_and_wait('insmod ' + S.getValue('GUEST_OVS_DPDK_DIR')[self._number] +
+                                  '/l2fwd' + '/l2fwd.ko net1=' + self._nics[0]['device'] +
+                                  ' net2=' + self._nics[1]['device'])
+        else:
+            raise RuntimeError('l2fwd can forward only between 2 NICs, but {} NICs are '
+                               'configured inside GUEST'.format(len(self._nics)))
 
     def _configure_linux_bridge(self):
         """
         Configure VM to perform L2 forwarding between NICs by linux bridge
         """
+        if int(S.getValue('GUEST_NIC_QUEUES')[self._number]):
+            self._set_multi_queue_nic()
         self._configure_disable_firewall()
-        nic1_name = get_test_param('vanilla_nic1_name', self._net1)
-        self.execute('ifconfig ' + nic1_name + ' ' +
-                     S.getValue('VANILLA_NIC1_IP_CIDR')[self._number])
-
-        nic2_name = get_test_param('vanilla_nic2_name', self._net2)
-        self.execute('ifconfig ' + nic2_name + ' ' +
-                     S.getValue('VANILLA_NIC2_IP_CIDR')[self._number])
 
         # configure linux bridge
         self.execute('brctl addbr br0')
-        self.execute('brctl addif br0 ' + self._net1 + ' ' + self._net2)
-        self.execute('ifconfig br0 ' +
-                     S.getValue('VANILLA_BRIDGE_IP')[self._number])
+
+        # add all NICs into the bridge
+        for nic in self._nics:
+            self.execute('ip addr add ' +
+                         nic['ip'] + ' dev ' + nic['device'])
+            self.execute('ip link set dev ' + nic['device'] + ' up')
+            self.execute('brctl addif br0 ' + nic['device'])
+
+        self.execute('ip addr add ' +
+                     S.getValue('GUEST_BRIDGE_IP')[self._number] +
+                     ' dev br0')
+        self.execute('ip link set dev br0 up')
 
         # Add the arp entries for the IXIA ports and the bridge you are using.
         # Use command line values if provided.
@@ -349,5 +483,15 @@ class IVnfQemu(IVnf):
         # Controls source route verification
         # 0 means no source validation
         self.execute('sysctl -w net.ipv4.conf.all.rp_filter=0')
-        self.execute('sysctl -w net.ipv4.conf.' + self._net1 + '.rp_filter=0')
-        self.execute('sysctl -w net.ipv4.conf.' + self._net2 + '.rp_filter=0')
+        for nic in self._nics:
+            self.execute('sysctl -w net.ipv4.conf.' + nic['device'] + '.rp_filter=0')
+
+    def _set_multi_queue_nic(self):
+        """
+        Enable multi-queue in guest kernel with ethool.
+        :return: None
+        """
+        for nic in self._nics:
+            self.execute_and_wait('ethtool -L {} combined {}'.format(
+                nic['device'], S.getValue('GUEST_NIC_QUEUES')[self._number]))
+            self.execute_and_wait('ethtool -l {}'.format(nic['device']))