[OE-core] [PATCH V3 8/8] runqemu: refactor it and remove machine knowledge
Robert Yang
liezhi.yang at windriver.com
Fri Aug 19 09:34:00 UTC 2016
On 08/19/2016 04:52 PM, Nathan Rossi wrote:
> On Fri, Aug 19, 2016 at 6:20 PM, Robert Yang <liezhi.yang at windriver.com> wrote:
>>
>>
>> On 08/19/2016 03:52 PM, Nathan Rossi wrote:
>>>
>>> On Fri, Aug 19, 2016 at 4:55 PM, Robert Yang <liezhi.yang at windriver.com>
>>> wrote:
>>>>
>>>> Previously, runqemu had hard coded machine knowledge, which limited its
>>>> usage, for example, qemu can boot genericx86, but runqemu can't, we need
>>>> edit runqemu/runqemu-internal a lot if we want to boot genericx86.
>>>>
>>>> Now bsp conf files can set vars to make it can be boot by runqemu, and
>>>> qemuboot.bbclass will save these info to DEPLOY_DIR_IMAGE/qemuboot.py.
>>>> Please see qemuboot.bbclass' comments on how to set the vars.
>>>>
>>>
>>> -- snip --
>>>
>>>> +
>>>> +
>>>> + def setup_slirp(self):
>>>> + if self.fstype == 'nfs':
>>>> + self.setup_nfs()
>>>> + if self.get('TUNE_ARCH') == 'aarch64':
>>>> + raise Exception("aarch64 doesn't support slirp mode")
>>>
>>>
>>> This doesn't seem quite right? because slirp works fine with the
>>> qemuzynqmp machine which is aarch64 (based on the patch you provided
>>> for meta-xilinx). This looks to be specific to the qemuarm64 machine?
>>
>>
>> Yes, thanks for the testing.
>>
>> When I first tested run qemuzynqmp, the following patch didn't get merged,
>> so it didn't work with slirp:
>>
>> commit e5c6a78db46192800669f1b392351f6b52f3e20c
>> Author: Alistair Francis <alistair.francis at xilinx.com>
>> AuthorDate: Wed Jun 15 13:43:23 2016 -0700
>> Commit: Richard Purdie <richard.purdie at linuxfoundation.org>
>> CommitDate: Wed Aug 10 10:45:32 2016 +0100
>>
>> runqemu: qemuzynqmp: Add Linux boot support
>>
>> Add support to direct boot Linux instead of just booting u-boot.
>>
>>
>> And for qemuarm64, the old code:
>> QEMU_NETWORK_CMD="-netdev tap,id=net0,ifname=$TAP,script=no,downscript=no
>> -device virtio-net-device,netdev=net0 "
>>
>> It hardcoded TAP, and didn't work with slirp when I tested it, so I
>> added the checking to avoid unexpected errors. Does anyone know how
>> to boot qemuarm64 by slirp, please ?
>
> I was able to boot qemuarm64 with slirp, here is a patch (its a quick
> hack specific for qemuarm64).
>
> diff --git a/scripts/runqemu b/scripts/runqemu
> index 7633bcb..ccf46a9 100755
> --- a/scripts/runqemu
> +++ b/scripts/runqemu
> @@ -552,8 +552,9 @@ class BaseConfig(object):
> if self.fstype == 'nfs':
> self.setup_nfs()
> self.append('KERNEL_CMDLINE_SCRIPT', ' ip=dhcp')
> - self.rootfs_options = '-drive file=%s,if=virtio,format=raw' %
> self.rootfs
> - self.set('QB_NETWORK_CMD', '')
> + qemu_network_cmd = '-netdev user,id=net0 -device
> virtio-net-device,netdev=net0'
> + self.rootfs_options = '-drive
> id=disk0,file=%s,if=none,format=raw -device
> virtio-blk-device,drive=disk0' % self.rootfs
> + self.set('QB_NETWORK_CMD', qemu_network_cmd)
Cool, this can boot aarch64 bsp, though it fails on other archs'
bsps such as qemuarm, qemux86, we can use QB_SLIRP_CMD as
you suggested to fix the problem.
$ runqemu nographic tmp/deploy/images/qemuarm slirp
[snip]
qemu-system-arm: -device virtio-net-device,netdev=net0: No 'virtio-bus' bus
found for device 'virtio-net-device'
[snip]
>
> def setup_tap(self):
> """Setup tap"""
>
> Note the rootfs_options set, it appears this is in the setup_tap as
> well, and is required to get the disk to appear. Might need to
> refactor the drive setup into the "setup_final"? or a
> setup_disk/rootfs function?
Set rootfs_options in setup_final() was what I had did before, but it
would make code more complicated, the problem is slirp, tap, cpio,
nfs, ext2/3/4, iso and vmdk have different rootfs_options, and plus
aarch64 is special, so if we don't set it when we check each of them,
and move the set to setup_final(), then we have to do the check again.
>
>>
>> I tested "runqemu qemuzynqmp slirp" just now, it works, so I will remove
>> this checking in the repo.
>>
>> For further work, if there are more bsps which can not be boot by slirp,
>> maybe we can add a var like QB_SLIRP_SUPPORTED = "0" or "1", and default
>> to 1, if the bsp sets '0' then error out.
>
> It might be worth having a QB_SLIRP_CMD = "", this would allow
> machines to setup the args for slirp/user mode networking and tap
> networking separately.
See above, sounds good, will do.
>
>>
>> Removed the checking, and updated in the repo:
>>
>> if self.get('TUNE_ARCH') == 'aarch64':
>> raise Exception("aarch64 doesn't support slirp mode")
>>
>> // Robert
>>
>>
>>>
>>> Regards,
>>> Nathan
>>>
>>>> + self.append('KERNEL_CMDLINE_SCRIPT', ' ip=dhcp')
>>>> + self.rootfs_options = '-drive file=%s,if=virtio,format=raw' %
>>>> self.rootfs
>>>> + self.set('QB_NETWORK_CMD', '')
>>>> +
>>>> + def setup_tap(self):
>>>> + """Setup tap"""
>>>> +
>>>> + # This file is created when runqemu-gen-tapdevs creates a bank
>>>> of tap
>>>> + # devices, indicating that the user should not bring up new ones
>>>> using
>>>> + # sudo.
>>>> + nosudo_flag = '/etc/runqemu-nosudo'
>>>> + self.qemuifup = shutil.which('runqemu-ifup')
>>>> + self.qemuifdown = shutil.which('runqemu-ifdown')
>>>> + ip = shutil.which('ip')
>>>> + lockdir = "/tmp/qemu-tap-locks"
>>>> +
>>>> + if not (self.qemuifup and self.qemuifdown and ip):
>>>> + raise Exception("runqemu-ifup, runqemu-ifdown or ip not
>>>> found")
>>>> +
>>>> + if not os.path.exists(lockdir):
>>>> + os.mkdir(lockdir)
>>>> +
>>>> + cmd = '%s link' % ip
>>>> + logger.info('Running %s...' % cmd)
>>>> + ip_link = subprocess.Popen(cmd, shell=True,
>>>> stdout=subprocess.PIPE).stdout.read().decode('utf-8')
>>>> + # Matches line like: 6: tap0: <foo>
>>>> + possibles = re.findall('^[1-9]+: +(tap[0-9]+): <.*', ip_link,
>>>> re.M)
>>>> + tap = ""
>>>> + use_preconf_tap = False
>>>> + for p in possibles:
>>>> + lockfile = os.path.join(lockdir, p)
>>>> + if os.path.exists('%s.skip' % lockfile):
>>>> + logger.info('Found %s.skip, skipping %s' % (lockfile,
>>>> p))
>>>> + continue
>>>> + if acquire_lock(lockfile):
>>>> + self.lockfile = lockfile
>>>> + tap = p
>>>> + logger.info("Using preconfigured tap device %s" % tap)
>>>> + logger.info("If this is not intended, touch %s.skip to
>>>> make runqemu skip %s." %(lockfile, tap))
>>>> + use_preconf_tap = True
>>>> + break
>>>> +
>>>> + if not tap:
>>>> + if os.path.exists(nosudo_flag):
>>>> + logger.error("Error: There are no available tap devices
>>>> to use for networking,")
>>>> + logger.error("and I see %s exists, so I am not going to
>>>> try creating" % nosudo_flag)
>>>> + raise Exception("a new one with sudo.")
>>>> +
>>>> + gid = os.getgid()
>>>> + uid = os.getuid()
>>>> + logger.info("Setting up tap interface under sudo")
>>>> + cmd = 'sudo %s %s %s %s' % (self.qemuifup, uid, gid,
>>>> self.get('STAGING_DIR_NATIVE'))
>>>> + tap = subprocess.Popen(cmd, shell=True,
>>>> stdout=subprocess.PIPE).stdout.read().decode('utf-8')
>>>> + self.cleantap = True
>>>> + logger.info('Created tap: %s' % tap)
>>>> +
>>>> + self.tap = tap
>>>> + n0 = tap[3:]
>>>> + n1 = int(n0) * 2 + 1
>>>> + n2 = n1 + 1
>>>> + self.nfs_instance = n0
>>>> + if self.fstype == 'nfs':
>>>> + self.setup_nfs()
>>>> + self.append("KERNEL_CMDLINE_SCRIPT", "
>>>> ip=192.168.7.%s::192.168.7.%s:255.255.255.0" % (n2, n1))
>>>> + qemu_tap_cmd = "-net
>>>> tap,vlan=0,ifname=%s,script=no,downscript=no" % self.tap
>>>> + self.rootfs_options = '-drive file=%s,if=virtio,format=raw' %
>>>> self.rootfs
>>>> + if self.get('TUNE_ARCH') == 'aarch64':
>>>> + qemu_network_cmd = '-netdev
>>>> tap,id=net0,ifname=%s,script=no,downscript=no -device
>>>> virtio-net-device,netdev=net0' % self.tap
>>>> + self.rootfs_options = '-drive
>>>> id=disk0,file=%s,if=none,format=raw -device virtio-blk-device,drive=disk0' %
>>>> self.rootfs
>>>> + elif self.get('TUNE_ARCH') == 'powerpc':
>>>> + qemu_network_cmd = '%s ' % qemu_tap_cmd
>>>> + else:
>>>> + qemu_network_cmd = "-net nic,model=virtio %s " %
>>>> qemu_tap_cmd
>>>> + if self.vhost_enabled:
>>>> + qemu_network_cmd += ',vhost=on'
>>>> + self.weak_assign('QB_NETWORK_CMD', qemu_network_cmd)
>>>> +
>>>> + def setup_network(self):
>>>> + cmd = "stty -g"
>>>> + self.saved_stty = subprocess.Popen(cmd, shell=True,
>>>> stdout=subprocess.PIPE).stdout.read().decode('utf-8')
>>>> + if self.slirp_enabled:
>>>> + self.setup_slirp()
>>>> + else:
>>>> + self.setup_tap()
>>>> +
>>>> + if self.fstype in ('cpio.gz', 'cpio'):
>>>> + self.set('QB_NETWORK_CMD', '')
>>>> + self.kernel_cmdline = 'root=/dev/ram0 rw debugshell'
>>>> + self.rootfs_options = '-initrd %s' % self.rootfs
>>>> + else:
>>>> + if self.fstype in self.vmtypes:
>>>> + if self.fstype == 'iso':
>>>> + vm_drive = '-cdrom %s' % self.rootfs
>>>> + else:
>>>> + cmd1 = "grep -q 'root=/dev/sd' %s" % self.rootfs
>>>> + cmd2 = "grep -q 'root=/dev/hd' %s" % self.rootfs
>>>> + if subprocess.call(cmd1, shell=True) == 0:
>>>> + logger.info('Using scsi drive')
>>>> + vm_drive = '-drive if=none,id=hd,file=%s -device
>>>> virtio-scsi-pci,id=scsi -device scsi-hd,drive=hd' % self.rootfs
>>>> + elif subprocess.call(cmd2, shell=True) == 0:
>>>> + logger.info('Using scsi drive')
>>>> + vm_drive = self.rootfs
>>>> + else:
>>>> + logger.warn("Can't detect drive type %s" %
>>>> self.rootfs)
>>>> + logger.warn('Tring to use virtio block drive')
>>>> + vm_drive = '-drive if=virtio,file=%s' %
>>>> self.rootfs
>>>> + self.rootfs_options = '%s -no-reboot' % vm_drive
>>>> + self.kernel_cmdline = 'root=%s rw highres=off' %
>>>> (self.get('KERNEL_ROOT'))
>>>> +
>>>> + if self.fstype == 'nfs':
>>>> + self.rootfs_options = ''
>>>> + k_root = '/dev/nfs nfsroot=%s:%s,%s' % (self.nfs_server,
>>>> self.nfs_dir, self.unfs_opts)
>>>> + self.kernel_cmdline = 'root=%s rw highres=off' % k_root
>>>> +
>>>> + self.weak_assign('QB_ROOTFS_OPTIONS', self.rootfs_options)
>>>> +
>>>> + def setup_final(self):
>>>> + qemu_system = self.get('QB_SYSTEM_NAME')
>>>> + if not qemu_system:
>>>> + raise Exception("Failed to boot, QB_SYSTEM_NAME is NULL!")
>>>> +
>>>> + qemu_bin = '%s/%s' % (self.get('STAGING_BINDIR_NATIVE'),
>>>> qemu_system)
>>>> + if not os.access(qemu_bin, os.X_OK):
>>>> + raise Exception("No QEMU binary '%s' could be found" %
>>>> qemu_bin)
>>>> +
>>>> + self.qemu_opt = "%s %s %s %s %s %s" % (qemu_bin,
>>>> self.get('QB_NETWORK_CMD'), self.get('QEMU_OPT_SCRIPT'),
>>>> self.get('QB_ROOTFS_OPTIONS'), self.get('QB_DTB'),
>>>> self.get('QB_OPT_APPEND'))
>>>> +
>>>> + if self.serialstdio:
>>>> + logger.info("Interrupt character is '^]'")
>>>> + cmd = "stty intr ^]"
>>>> + subprocess.call(cmd, shell=True)
>>>> +
>>>> + first_serial = ""
>>>> + if not re.search("-nographic", self.qemu_opt):
>>>> + first_serial = "-serial mon:vc"
>>>> + # We always want a ttyS1. Since qemu by default adds a serial
>>>> + # port when nodefaults is not specified, it seems that all that
>>>> + # would be needed is to make sure a "-serial" is there. However,
>>>> + # it appears that when "-serial" is specified, it ignores the
>>>> + # default serial port that is normally added. So here we make
>>>> + # sure to add two -serial if there are none. And only one if
>>>> + # there is one -serial already.
>>>> + serial_num = len(re.findall("-serial", self.qemu_opt))
>>>> + if serial_num == 0:
>>>> + self.qemu_opt += " %s %s" % (first_serial,
>>>> self.get("QB_SERIAL_OPT"))
>>>> + elif serial_num == 1:
>>>> + self.qemu_opt += " %s" % self.get("QB_SERIAL_OPT")
>>>> +
>>>> + def start_qemu(self):
>>>> + if self.kernel:
>>>> + kernel_opts = "-kernel %s -append '%s %s %s'" %
>>>> (self.kernel, self.kernel_cmdline, self.get('KERNEL_CMDLINE_SCRIPT'),
>>>> self.get('QB_KERNEL_CMDLINE_APPEND'))
>>>> + else:
>>>> + kernel_opts = ""
>>>> + cmd = "%s %s" % (self.qemu_opt, kernel_opts)
>>>> + logger.info('Running %s' % cmd)
>>>> + if subprocess.call(cmd, shell=True) != 0:
>>>> + raise Exception('Failed to run %s' % cmd)
>>>> +
>>>> + def cleanup(self):
>>>> + if self.cleantap:
>>>> + cmd = 'sudo %s %s %s' % (self.qemuifdown, self.tap,
>>>> self.get('STAGING_DIR_NATIVE'))
>>>> + logger.info('Running %s' % cmd)
>>>> + subprocess.call(cmd, shell=True)
>>>> + if self.lockfile:
>>>> + logger.info("Releasing lockfile of preconfigured tap device
>>>> '%s'" % self.tap)
>>>> + release_lock(self.lockfile)
>>>> +
>>>> + if self.nfs_running:
>>>> + logger.info("Shutting down the userspace NFS server...")
>>>> + cmd = "runqemu-export-rootfs stop %s" % self.nfs_dir
>>>> + logger.info('Running %s' % cmd)
>>>> + subprocess.call(cmd, shell=True)
>>>> +
>>>> + if self.saved_stty:
>>>> + cmd = "stty %s" % self.saved_stty
>>>> + subprocess.call(cmd, shell=True)
>>>> +
>>>> + if self.clean_nfs_dir:
>>>> + logger.info('Removing %s' % self.nfs_dir)
>>>> + shutil.rmtree(self.nfs_dir)
>>>> + shutil.rmtree('%s.pseudo_state' % self.nfs_dir)
>>>> +
>>>> +def main():
>>>> + if len(sys.argv) == 1 or "help" in sys.argv:
>>>> + print_usage()
>>>> + return 0
>>>> + config = BaseConfig()
>>>> + config.check_args()
>>>
>>>
>>> When it fails to parse the args instead of showing the python
>>> exception info (with call stack, etc), it should show the exception
>>> message and the runqemu usage.
>>>
>>> Regards,
>>> Nathan
>>>
>
> You may not have seen this comment, since I accidentally put two
> "Regards," in my previous email.
Yes, I had missed this, print the full help text seems too more,
I will print "See 'runqemu help' on how to use it"
// Robert
>
> Regards,
> Nathan
>
More information about the Openembedded-core
mailing list