Update kdump.conf to compress the memory on rhel5 guest, update
logging.info to error.context.
Signed-off-by: NYunping Zheng <yunzheng@redhat.com>

libvirt: add some tests for performance of libvirt.

Sometimes file command retuns upper-case results and it's
possible someone could configure a format w/ mixed case.
Instead, be clear and always compare as lower-case.
Signed-off-by: NChris Evich <cevich@redhat.com>

1. Instead of thread, use os.fork to create a new process,
   because we cannot get any exception in main process from
   another thread.
2. Add "try...finally" part at the end, because when test failed,
   config file won't be recovered.
3. Add 3 test cases for paused VM test.
4. Fix too many 'if' problem
Signed-off-by: NLi Yang <liyang.fnst@cn.fujitsu.com>

Signed-off-by: NLucas Meneghel Rodrigues <lmr@redhat.com>

Step 1: Get the virsh domiflist value.
Step 2: Check for interface in xml file.
Step 3: Check for type in xml file.
Step 4: Check for model inside the guest and xml file.
Step 5: Check for mac id inside the guest and xml file.
Signed-off-by: NSatheesh Rajendran <sathnaga@linux.vnet.ibm.com>

Signed-off-by: NSatheesh Rajendran <sathnaga@linux.vnet.ibm.com>

The parameter "connect_uri" will effect vm's connect_uri.
But it is not useful in remote testcases because what we
need is remote_connect_uri which means connection from
remote to local.
And it shouldn't replace connect_uri of libvirt_vm, otherwise
all operation to local vms will need authorization.
Signed-off-by: NYu Mingfei <yumingfei@cn.fujitsu.com>

With original logic of VirshSession.__init__(), we have
not choice but configuring auto-ssh-login for "local to
remote: virsh -c uri".
While for other types driver like esx, it will fail.
Now support login with password if auto-ssh-login is impossible.
Signed-off-by: NYu Mingfei <yumingfei@cn.fujitsu.com>

It is better to create the temporary GNUPG home dir in
the data_dir.TMP_DIR rather than in the same directory
as the key file, to avoid reuse from different jobs
and also, the gpg mount might be read-only.
Signed-off-by: NLucas Meneghel Rodrigues <lmr@redhat.com>

Signed-off-by: NLucas Meneghel Rodrigues <lmr@redhat.com>

Signed-off-by: NLucas Meneghel Rodrigues <lmr@redhat.com>

This tests the disk hotplug/unplug functionality.
 1) prepares multiple disks to be hotplugged
 2) hotplugs them
 3) verifies that they are in qtree/guest system/...
 4) unplugs them
 5) verifies they are not in qtree/guest system/...
 6) repeats $repeat_times
 *) During the whole test stress_cmd might be executed
Signed-off-by: NLukáš Doktor <ldoktor@redhat.com>

Log also which image_filename failed. It's not easy to find this
information as cleanup messages are logged after this failure.
Signed-off-by: NLukáš Doktor <ldoktor@redhat.com>

There is a hook to create ich9-usb-ehci from multiple devices. Anyway
the device was not plugged into qdev PCI representation thus another
device might use the same address.

qemu_devices still doesn't support x.y PCI address thus plug only
ich9-usb-uhci1 device into this PCI representation into address
0x1d to occupy this slot.
Signed-off-by: NLukáš Doktor <ldoktor@redhat.com>

USB buses behaves realy weird and always assigns different port
when hotplug/unplug is used. Always set port of the device to
make sure it is in the correct slot.
Signed-off-by: NLukáš Doktor <ldoktor@redhat.com>

There were changes in QSCSIBus which now accepts only addr_lengths
instead of full addr_spec. Update the images_define_by_variables()
function to suit current behavior.
Signed-off-by: NLukáš Doktor <ldoktor@redhat.com>

qemu doesn't differentiate between some of the buses, which are
differentiated in autotest (eg. ahci and ide disks belongs simply
to IDE bus type). In order to be able to distinguish between those
buses we need to add atype (=autotest type) of the bus.

This patch makes this change only for images, where were problems
caused by different qemu/autotest behavior. If this concept prooves
to be right, we can use it to distinguish between various kinds of
USBs in order to reflect the qemu behavior while keeping the autotest
standards.
Signed-off-by: NLukáš Doktor <ldoktor@redhat.com>

Conflicts:
	virttest/qemu_devices.py

When we try to match a bus to some specification (dictionary) it makes
sense to ask the bus and not to try to guess. This way we can customize
the function to each bus, if necessarily.

Additionally make it possible to ignore autotest type by setting
atest=False. This could be used to see if this specification matches from
qemu point of view, but doesn't match from autotest point of view.
(from qemu point of view ahci and ide bus is the same, but in autotest
we want to make sure the ide-hd goes into the correct HBA). This feature
will be used in the following patch.
Signed-off-by: NLukáš Doktor <ldoktor@redhat.com>

Conflicts:
	virttest/qemu_devices.py

Since a GPG file is a user provided data (like isos
and guest images), it makes more sense to have it
under shared/data/gpg than shared/gpg. So remove the
later and fix things in the bootstrap module.
Signed-off-by: NLucas Meneghel Rodrigues <lmr@redhat.com>

Test for the performance of libvirt in domain state switching.

Steps:
    * shutdown (or destroy)
    * start
    * suspend
    * resume
Signed-off-by: NDongsheng Yang <yangds.fnst@cn.fujitsu.com>

There might be an occasion where we get the following
error:

TypeError: unsupported operand type(s) for +: 'MonitorProtocolError' and 'str'

So let's try to convert the exception class to a string
first.
Signed-off-by: NLucas Meneghel Rodrigues <lmr@redhat.com>

The problem was that the arguments for the string interpolation
were not passed as a tuple, as they should. When hitting
an unknown migration protocol error, the return would raise
a TypeError. Let's fix it.
Signed-off-by: NLucas Meneghel Rodrigues <lmr@redhat.com>

Signed-off-by: NLucas Meneghel Rodrigues <lmr@redhat.com>

Signed-off-by: NLucas Meneghel Rodrigues <lmr@redhat.com>

Signed-off-by: Nyangdongsheng <yangds.fnst@cn.fujitsu.com>

Lots of times to call vmxml.undefine() and vmxml.define(), to
make our config in vmxml take effect in vm.
Signed-off-by: Nyangdongsheng <yangds.fnst@cn.fujitsu.com>

Signed-off-by: Nyangdongsheng <yangds.fnst@cn.fujitsu.com>

Turns out RHEL5 qemu does not support the 'info qtree'
command, therefore an attempt to parse the output of
this command will fail. So skip that particular check
when running on this particular qemu version.
Signed-off-by: NLucas Meneghel Rodrigues <lmr@redhat.com>

On RHEL5, we were not passing boot=on properly to the
qemu-kvm binary. Fix the heuristics to detect whether
the binary uses this ancient syntax to define virtio
disks as bootable.
Signed-off-by: NLucas Meneghel Rodrigues <lmr@redhat.com>

Signed-off-by: NLucas Meneghel Rodrigues <lmr@redhat.com>

Use the unified configs (scsiid=>unit, lun=>port) and store the images
in the usual location. The location '/tmp/%s' was used before because
python was unable to execute that long command and this prefix shorten
it significantly. Now autotest supports the way to deal with long
commands so we can finally store the images in the right location.
Signed-off-by: NLukáš Doktor <ldoktor@redhat.com>

Some cases create many disks, which takes a long time, where most
time consumes the handling around cmd execution. This patch parallelize
the image processing in case we deal with over 20 disks. It spawns
$no_cpus * 2 threads (or less in case it would be inefficient).

The performance improvement was significant (30=>12s on 4CPU system).
Signed-off-by: NLukáš Doktor <ldoktor@redhat.com>

Current kernels support unlimited argument length, the only limit is
stack size. The default stack size is usually 8KB which is sufficient
for ~2MB commands. Use ulimit to increase these limits. The multi_disk
in some configurations attacks the 2MB barier and in the future we
might see this comming (unless we switch to -readconfig).

NOTE: Only superuser can increase the hard limit, anyway if the command
      fails the test continues and tries to execute the cmd anyway.
Signed-off-by: NLukáš Doktor <ldoktor@redhat.com>

virttest.qemu_devices: Fix floppy issues

Old qemu without -global support must use -drive index=n
to set the floppy slot.
Signed-off-by: NLukáš Doktor <ldoktor@redhat.com>

image_format is usually set for images. Floppies usually use raw so
we'd have to either set image format for floppies or we can simply
ignore it (this solution).
Signed-off-by: NLukáš Doktor <ldoktor@redhat.com>

media accepts only disk or cdrom. Use media='' to ignore it even in
strict_mode. (strict_mode sets media=disk when media is None)
Signed-off-by: NLukáš Doktor <ldoktor@redhat.com>

qemu.test: add reboot step for pci_hotplug testing