[dts] [PATCH] framework: enable AArch64 and add more options for libvirt
Herbert Guan
Herbert.Guan at arm.com
Fri Dec 22 10:53:28 CET 2017
Marvin,
> -----Original Message-----
> From: Liu, Yong [mailto:yong.liu at intel.com]
> Sent: Friday, December 22, 2017 16:10
> To: Herbert Guan <Herbert.Guan at arm.com>; dts at dpdk.org
> Cc: Phil Yang <Phil.Yang at arm.com>; Jianbo Liu <Jianbo.Liu at arm.com>
> Subject: RE: [dts] [PATCH] framework: enable AArch64 and add more options
> for libvirt
>
> Herbert, thanks for your patch. I'm not very clear about what is must for
> AArch64 support and what is for enabling your environment.
> Please separate them in different functions.
To make it more clear, are you requesting to split this patch into smaller ones (maybe around 6 patches)?
>
> Thanks,
> Marvin
>
> > -----Original Message-----
> > From: dts [mailto:dts-bounces at dpdk.org] On Behalf Of Herbert Guan
> > Sent: Tuesday, December 05, 2017 1:46 PM
> > To: dts at dpdk.org
> > Cc: phil.yang at arm.com; jianbo.liu at arm.com; Herbert Guan
> > <herbert.guan at arm.com>
> > Subject: [dts] [PATCH] framework: enable AArch64 and add more options
> > for libvirt
> >
> > 1) Add the libvirt support on AArch64 platforms
> > 2) Add 'loader' and 'nvram' option support in 'os' section
> > 3) Add 'opt_bus', 'opt_dev' and 'opt_controller' option support in
> > 'disk' section. Change 'type' to 'opt_format' to get aligined
> > with option naming in qemu_kvm.
> > 4) Add serial 'serial_option' option support for serial console
> > 5) Add domain parsing in __parse_pci()
> > 6) Function add_vm_device() supports both VF (new) and PF devices now
> >
> > Signed-off-by: Herbert Guan <herbert.guan at arm.com>
> > ---
> > conf/vm_power_manager.cfg | 4 +-
> > framework/qemu_libvirt.py | 152
> > ++++++++++++++++++++++++++++++++++++++---
> > -----
> > 2 files changed, 128 insertions(+), 28 deletions(-) mode change
> > 100644 => 100755 conf/vm_power_manager.cfg mode change 100644 =>
> > 100755 framework/qemu_libvirt.py
> >
> > diff --git a/conf/vm_power_manager.cfg b/conf/vm_power_manager.cfg
> old
> > mode 100644 new mode 100755 index 9c8f87a..7d5d9b6
> > --- a/conf/vm_power_manager.cfg
> > +++ b/conf/vm_power_manager.cfg
> > @@ -7,7 +7,7 @@
> > # size: 4096
> > # disk
> > # file: absolute path to disk image
> > -# type: disk image format
> > +# opt_format: [ raw | qcow2 | ... ] #disk image format
> > # login
> > # user: user name to login into VM
> > # password: passwork to login into VM
> > @@ -25,7 +25,7 @@ cpu =
> > mem =
> > size=4096;
> > disk =
> > - file=/storage/vm-image/vm0.img,type=raw;
> > + file=/storage/vm-image/vm0.img,opt_format=raw;
> > login =
> > user=root,password=tester;
> > [vm1]
> > diff --git a/framework/qemu_libvirt.py b/framework/qemu_libvirt.py old
> > mode 100644 new mode 100755 index ed8e0e6..18dcd8b
> > --- a/framework/qemu_libvirt.py
> > +++ b/framework/qemu_libvirt.py
> > @@ -57,6 +57,7 @@ class LibvirtKvm(VirtBase):
> >
> > # disk and pci device default index
> > self.diskindex = 'a'
> > + self.controllerindex = 0
> > self.pciindex = 10
> >
> > # configure root element
> > @@ -82,7 +83,12 @@ class LibvirtKvm(VirtBase):
> >
> > # set some default values for vm,
> > # if there is not the values of the specified options
> > - self.set_vm_default()
> > + arch = self.host_session.send_expect('uname -m', '# ')
> > + set_default_func = getattr(self, 'set_vm_default_' + arch)
> > + if callable(set_default_func):
> > + set_default_func()
> > + else:
> > + self.set_vm_default()
>
> I'm not sure how much is the difference between x86 and AArch64 platform.
> Look like somethings are common for example graphic and control interface.
> Could we separate arch related setting in separated function and keep those
> common in another function?
>
How about this: call into common function set_vm_default() first, and then call arch specific function set_vm_default_'x86_64/aarch64/...'()?
> >
> > def get_qemu_emulator(self):
> > """
> > @@ -98,6 +104,13 @@ class LibvirtKvm(VirtBase):
> > """
> > check and setup host virtual ability
> > """
> > + arch = self.host_session.send_expect('uname -m', '# ')
> > + if arch == 'aarch64':
> > + out = self.host_session.send_expect('service libvirtd
> > + status',
> > "# ")
> > + if 'active (running)' not in out:
> > + return False
> > + return True
> > +
>
> Herbert, libvirt status has been checked in function has_virtual_ability when
> initialize LibvirtKvm object.
> Why do this check again?
>
This is not double check but just modifying the logic of "has_virtual_ability ". It defines the specific checking logic for arm platforms. And if it's not 'aarch64' platform, will go with the original logic.
> > out = self.host_session.send_expect('cat /proc/cpuinfo | grep
> > flags',
> > '# ')
> > rgx = re.search(' vmx ', out) @@ -201,6 +214,47 @@ class
> > LibvirtKvm(VirtBase):
> > ',name=org.qemu.guest_agent.0'})
> > self.qga_sock_path = '/tmp/%s_qga0.sock' % self.vm_name
> >
> > + def add_vm_os(self, **options):
> > + os = self.domain.find('os')
> > + if 'loader' in options.keys():
> > + loader = ET.SubElement(
> > + os, 'loader', {'readonly': 'yes', 'type': 'pflash'})
> > + loader.text = options['loader']
> > + if 'nvram' in options.keys():
> > + nvram = ET.SubElement(os, 'nvram')
> > + nvram.text = options['nvram']
> > +
>
> Where this function is called?
It's called by load_local_config() in virt_base.py. There'll be an entry added in case config file, such as:
os =
loader=/usr/share/AAVMF/AAVMF_CODE.fd,nvram=/var/lib/libvirt/qemu/nvram/vm0_VARS.fd;
>
> > +
> > + def set_vm_default_aarch64(self):
> > + os = ET.SubElement(self.domain, 'os')
> > + type = ET.SubElement(
> > + os, 'type', {'arch': 'aarch64', 'machine': 'virt'})
> > + type.text = 'hvm'
> > + ET.SubElement(os, 'boot', {'dev': 'hd'})
> > +
> > + features = ET.SubElement(self.domain, 'features')
> > + ET.SubElement(features, 'acpi')
> > + ET.SubElement(features, 'gic', {'version': '3'})
> > +
> > + ET.SubElement(self.domain, 'cpu',
> > + {'mode': 'host-passthrough', 'check': 'none'})
> > +
> > + # qemu-kvm for emulator
> > + device = ET.SubElement(self.domain, 'devices')
> > + ET.SubElement(device, 'emulator').text = self.qemu_emulator
> > +
> > + # graphic device
> > + #ET.SubElement(device, 'graphics', {
> > + # 'type': 'vnc', 'port': '-1', 'autoport': 'yes'})
> > + # qemu guest agent
> > + self.add_vm_qga(None)
> > +
> > + # add default control interface
> > + if not self.__default_nic:
> > + def_nic = {'type': 'nic', 'opt_hostfwd': ''}
> > + self.add_vm_net(**def_nic)
> > + self.__default_nic = True
> > +
> > def set_vm_default(self):
> > os = ET.SubElement(self.domain, 'os')
> > type = ET.SubElement(
> > @@ -273,17 +327,44 @@ class LibvirtKvm(VirtBase):
> > return False
> >
> > ET.SubElement(disk, 'source', {'file': options['file']})
> > - if 'type' not in options:
> > + if 'opt_format' not in options:
> > disk_type = 'raw'
> > else:
> > - disk_type = options['type']
> > + disk_type = options['opt_format']
> >
> > ET.SubElement(disk, 'driver', {'name': 'qemu', 'type':
> > disk_type})
> >
> > + if 'opt_bus' not in options:
> > + bus = 'virtio'
> > + else:
> > + bus = options['opt_bus']
> > + if 'opt_dev' not in options:
> > + dev = 'vd%c' % self.diskindex
> > + self.diskindex = chr(ord(self.diskindex) + 1)
> > + else:
> > + dev = options['opt_dev']
> > ET.SubElement(
> > - disk, 'target', {'dev': 'vd%c' % self.diskindex, 'bus':
> > 'virtio'})
> > + disk, 'target', {'dev': dev, 'bus': bus})
> > +
> > + if 'opt_controller' in options:
> > + controller = ET.SubElement(devices, 'controller',
> > + {'type': bus,
> > + 'index': hex(self.controllerindex)[2:],
> > + 'model': options['opt_controller']})
> > + self.controllerindex += 1
> > + ET.SubElement(controller, 'address',
> > + {'type': 'pci', 'domain': '0x0000', 'bus':
> > hex(self.pciindex),
> > + 'slot': '0x00', 'function': '0x00'})
> > + self.pciindex += 1
> >
> > - self.diskindex = chr(ord(self.diskindex) + 1)
> > + def add_vm_serial_port(self, **options):
> > + if 'enable' in options.keys():
> > + if options['enable'].lower() == 'yes':
> > + devices = self.domain.find('devices')
> > + serial = ET.SubElement(devices, 'serial', {'type': 'pty'})
> > + ET.SubElement(serial, 'target', {'port': '0'})
> > + console = ET.SubElement(devices, 'console', {'type':
> > 'pty'})
> > + ET.SubElement(console, 'target', {'type': 'serial',
> > 'port': '0'})
> >
>
> In qemu kvm module, we are using unix domain socket for serial device. Can
> you align to this?
> Please let device type should be configurable, we plan to add more types
> later.
OK, I'll add the unix socket support and make type configurable in the new version.
>
> > def add_vm_login(self, **options):
> > """
> > @@ -305,14 +386,23 @@ class LibvirtKvm(VirtBase):
> > def __parse_pci(self, pci_address):
> > pci_regex = r"([0-9a-fA-F]{1,2}):([0-9a-fA-F]{1,2})" + \
> > ".([0-9a-fA-F]{1,2})"
> > + pci_regex_domain = r"([0-9a-fA-F]{1,4}):([0-9a-fA-F]{1,2}):" + \
> > + "([0-9a-fA-F]{1,2}).([0-9a-fA-F]{1,2})"
> > m = re.match(pci_regex, pci_address)
> > - if m is None:
> > - return None
> > - bus = m.group(1)
> > - slot = m.group(2)
> > - func = m.group(3)
> > -
> > - return (bus, slot, func)
> > + if m is not None:
> > + bus = m.group(1)
> > + slot = m.group(2)
> > + func = m.group(3)
> > + dom = '0'
> > + return (bus, slot, func, dom)
> > + m = re.match(pci_regex_domain, pci_address)
> > + if m is not None:
> > + bus = m.group(2)
> > + slot = m.group(3)
> > + func = m.group(4)
> > + dom = m.group(1)
> > + return (bus, slot, func, dom)
> > + return None
> >
> > def add_vm_device(self, **options):
> > """
> > @@ -325,35 +415,45 @@ class LibvirtKvm(VirtBase):
> > 'mode': 'subsystem', 'type': 'pci',
> > 'managed': 'yes'})
> >
> > - if 'pf_idx' not in options.keys():
> > - print utils.RED("Missing device index for device option!!!")
> > - return False
> > -
> > - pf = int(options['pf_idx'])
> > - if pf > len(self.host_dut.ports_info):
> > - print utils.RED("PF device index over size!!!")
> > + if 'pf_idx' in options.keys():
> > + pf = int(options['pf_idx'])
> > + if pf > len(self.host_dut.ports_info):
> > + print utils.RED("PF device index over size!!!")
> > + return False
>
> Previous check sentence is incorrect:) Should be "if pf >=
> len(self.host_dut.ports_info):".
> Most of suites are just using option "opt_host" to specify the pci address of
> device. So I'm planning to remove pf_idx option, what's your idea for that?
>
Fully agree switching to "opt_host" to specify the PCI address. This can match the function name "add_vm_device" better, not limited to pf/vf NIC devices. 'vf_idx' was added just because it only support pf before, but this is quite complex and not aligned with vf_xx suites. I can make such change if decided to go. Could you confirm that?
> > + pci_addr = self.host_dut.ports_info[pf]['pci']
> > + elif 'vf_idx' in options.keys():
> > + vf = int(options['vf_idx'])
> > + if vf > len(self.host_dut.ports_info):
> > + print utils.RED("VF device index over size!!!")
> > + return False
> > + if 'vf_id' in options.keys():
> > + vf_id = int(options['vf_id'])
> > + else:
> > + vf_id = 0
> > + pci_addr =
> > self.host_dut.ports_info[vf]['sriov_vfs_pci'][vf_id]
> > + else:
> > + print utils.RED("Missing pf/vf device index for device
> > option!!!")
> > return False
> >
> > - pci_addr = self.host_dut.ports_info[pf]['pci']
> >
> > pci = self.__parse_pci(pci_addr)
> > if pci is None:
> > return False
> > - bus, slot, func = pci
> > + bus, slot, func, dom = pci
> >
> > source = ET.SubElement(hostdevice, 'source')
> > ET.SubElement(source, 'address', {
> > - 'domain': '0x0', 'bus': '0x%s' % bus,
> > + 'domain': '0x%s' % dom, 'bus': '0x%s' % bus,
> > 'slot': '0x%s' % slot,
> > 'function': '0x%s' % func})
> > if 'guestpci' in options.keys():
> > pci = self.__parse_pci(options['guestpci'])
> > if pci is None:
> > return False
> > - bus, slot, func = pci
> > + bus, slot, func, dom = pci
> > ET.SubElement(hostdevice, 'address', {
> > - 'type': 'pci', 'domain': '0x0', 'bus': '0x%s' %
> > bus,
> > - 'slot': '0x%s' % slot, 'function': '0x%s' %
> > func})
> > + 'type': 'pci', 'domain': '0x%s' % dom, 'bus':
> > + '0x%s' %
> > bus,
> > + 'slot': '0x%s' % slot, 'function': '0x%s' % func})
> > # save host and guest pci address mapping
> > pci_map = {}
> > pci_map['hostpci'] = pci_addr @@ -397,7 +497,7 @@ class
> > LibvirtKvm(VirtBase):
> > pci = self.__parse_pci(options['opt_addr'])
> > if pci is None:
> > return False
> > - bus, slot, func = pci
> > + bus, slot, func, dom = pci
> > ET.SubElement(qemu, 'qemu:arg',
> > {'value': 'nic,model=e1000,addr=0x%s' % slot})
> > else:
> > --
> > 1.8.3.1
Thanks,
Herbert
More information about the dts
mailing list