qemu_system_reset() function always performs the same basic actions on
all machines. This includes running all the reset handler hooks,
however the order in which these will run is not always easily predictable.
This patch splits the core of qemu_system_reset() - the invocation of
the reset handlers - out into a new qemu_devices_reset() function.
qemu_system_reset() will usually call qemu_devices_reset(), but that
can be now overriden by a new reset method in the QEMUMachine
structure.
Individual machines can use this reset method, if necessary, to
perform any extra, machine specific initializations which have to
occur before or after the bulk of the reset handlers. It's expected
that the method will call qemu_devices_reset() at some point, but if
the machine has really strange ordering requirements between devices
resets it could even override that with it's own reset sequence (with
great care, obviously).
For a specific example of when this might be needed: a number of
machines (but not PC) load images specified with -kernel or -initrd
directly into the machine RAM before booting the guest. This mostly
works at the moment, but to make this actually safe requires that this
load occurs after peripheral devices are reset - otherwise they could
have active DMAs in progress which would clobber the in memory images.
Some machines (notably pseries) also have other entry conditions which
need to be set up as the last thing before executing in guest space -
some of this could be considered "emulated firmware" in the sense that
the actions of the firmware are emulated directly by qemu rather than
by executing a firmware image within the guest. When the platform's
firmware to OS interface is sufficiently well specified, this saves
time both in implementing the "firmware" and executing it.
aliguori: don't unconditionally dereference current_machine
Reviewed-by: Andreas Färber <afaerber@suse.de>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
The pseries machine already supports the -vga std option, creating a
graphics adapter. However, this is not very useful without being able to
add a keyboard and mouse as well. This patch addresses this by adding
a USB interface when requested, and automatically adding a USB keyboard
and mouse when VGA is enabled.
This is a stop gap measure to get usable graphics mode on pseries while
waiting for Li Zhang's rework of USB options to go in after 1.2.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
Fix compilation failure on BSD systems (which don't have
O_DIRECT or O_NOATIME:
osdep.c:116: error: ‘O_DIRECT’ undeclared (first use in this function)
osdep.c:116: error: (Each undeclared identifier is reported only once
osdep.c:116: error: for each function it appears in.)
osdep.c:116: error: ‘O_NOATIME’ undeclared (first use in this function)
Reviewed-by: Stefan Weil <sw@weilnetz.de>
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
Signed-off-by: Eduardo Otubo <otubo@linux.vnet.ibm.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
---
v1:
- Full seccomp calls and data included in vl.c
v1 -> v2:
- Full seccomp calls and data removed from vl.c and put into separate
qemu-seccomp.[ch] file.
Signed-off-by: Eduardo Otubo <otubo@linux.vnet.ibm.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
---
v1:
- I added a syscall struct using priority levels as described in the
libseccomp man page. The priority numbers are based to the frequency
they appear in a sample strace from a regular qemu guest run under
libvirt.
Libseccomp generates linear BPF code to filter system calls, those rules
are read one after another. The priority system places the most common
rules first in order to reduce the overhead when processing them.
v1 -> v2:
- Fixed some style issues
- Removed code from vl.c and created qemu-seccomp.[ch]
- Now using ARRAY_SIZE macro
- Added more syscalls without priority/frequency set yet
v2 -> v3:
- Adding copyright and license information
- Replacing seccomp_whitelist_count just by ARRAY_SIZE
- Adding header protection to qemu-seccomp.h
- Moving QemuSeccompSyscall definition to qemu-seccomp.c
- Negative return from seccomp_start is fatal now.
- Adding open() and execve() to the whitelis
v3 -> v4:
- Tests revealed a bigger set of syscalls.
- seccomp_start() now has an argument to set the mode according to the
configure option trap or kill.
v4 -> v5:
- Tests on x86_64 required a new specific set of system calls.
- libseccomp release 1.0.0: part of the API have changed in this last
release, had to adapt to the new function signatures.
Adding basic options to the configure script to use libseccomp or not.
The default is set to 'no'. If the flag --enable-libseccomp is used, the
script will check for its existence using pkg-config.
Signed-off-by: Eduardo Otubo <otubo@linux.vnet.ibm.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
---
v1 -> v2:
- As I removed all the code related to seccomp from vl.c, I created
qemu-seccomp.[ch].
- Also making the configure script to add the specific line to
Makefile.obj in order to compile with appropriate support to seccomp.
v2 -> v3:
- Removing the line from Makefile.obj and adding it to Makefile.objs.
- Marking libseccomp default option to 'yes' in the configure script.
v3 -> v8:
- fix configure probe if libseccomp isn't available (aliguori)
Hi hard a brain fart when coding that function, it will
fail to "set" the memory beyond the first 512 bytes. This
is in turn causing guest crashes in ibmveth (spapr_llan.c
on the qemu side) due to the receive queue not being
properly initialized.
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
Add a new '[,dump-guest-core=on|off]' option to the '-machine' option. When
'dump-guest-core=off' is specified, guest memory is omitted from the core dump.
The default behavior continues to be to include guest memory when a core dump is
triggered. In my testing, this brought the core dump size down from 384MB to 6MB
on a 2GB guest.
Is anything additional required to preserve this setting for migration or
savevm? I don't believe so.
Changelog:
v3:
Eliminate globals as per Anthony's suggestion
set no dump from qemu_ram_remap() as well
v2:
move the option from -m to -machine, rename option dump -> dump-guest-core
Signed-off-by: Jason Baron <jbaron@redhat.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
MacOSX 10.8 ("Mountain Lion") requires us to compile our one
Objective-C source file with clang even if the rest of QEMU
requires a real gcc, because the system headers we use make
use of Apple's "Blocks" extension to C/ObjC, and mainline
gcc doesn't support that. Since we only need to use a true
gcc for the parts of QEMU that use the fixed-register
env variable, we can simply use clang to build the ObjC
file: it will link to the gcc-built objects with no problems.
Add the necessary support for an OBJCC variable in the
makefile and configure machinery; we default to clang
if we have it, otherwise whatever CC is (since gcc
might be the Apple gcc which does support Blocks).
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
MacOSX 10.8 ("Mountain Lion") defaults to trying to use automated
reference counting on certain objects. This means that the system
header files will use some Objective C syntax constructs even when
compiling pure C, which confuses mainline gcc. Suppress this by
setting OS_OBJECT_USE_OBJC=0. This avoids a compile error like this:
In file included from
/System/Library/Frameworks/Foundation.framework/Headers/NSObject.h:5:0,
from /usr/include/os/object.h:74,
from /usr/include/dispatch/dispatch.h:48,
from /System/Library/Frameworks/IOKit.framework/Headers/IOKitLib.h:56,
from block/raw-posix.c:35:
/System/Library/Frameworks/Foundation.framework/Headers/NSObjCRuntime.h:409:1: error: stray ‘@’ in program
[with a large number of further run-on errors]
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
Both MacOS and Solaris have special case handling for the CPU
type, because the check_define probes will return i386 even if
the hardware is 64 bit and x86_64 would be preferable. Move
these checks earlier in the configure probing so that we can
do them only if the user didn't specify a CPU with --cpu. This
fixes a bug where the user's command line argument was being
ignored.
Reviewed-by: Andreas F=E4rber <afaerber@suse.de>
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
Currently, when parsing a stream of tokens we make a copy of the token
list at the beginning of each level of recursion so that we do not
modify the original list in cases where we need to fall back to an
earlier state.
In the worst case, we will only read 1 or 2 tokens off the list before
recursing again, which means an upper bound of roughly N^2 token allocations.
For a "reasonably" sized QMP request (in this a QMP representation of
cirrus_vga's device state, generated via QIDL, being passed in via
qom-set), this caused my 16GB's of memory to be exhausted before any
noticeable progress was made by the parser.
This patch works around the issue by using single copy of the token list
in the form of an indexable array so that we can save/restore state by
manipulating indices.
A subsequent commit adds a "large_dict" test case which exhibits the
same behavior as above. With this patch applied the test case successfully
completes in under a second.
Tested with valgrind, make check, and QMP.
Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Michael Roth <mdroth@linux.vnet.ibm.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
Signed-off-by: Michael Roth <mdroth@linux.vnet.ibm.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Michael Roth <mdroth@linux.vnet.ibm.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
[ The following text is in the "ISO-8859-1" character set. ]
[ Your display is set for the "KOI8-R" character set. ]
[ Some special characters may be displayed incorrectly. ]
Since we moved pcspk into hwlib, CONFIG_PCSPK is no longer defined per
target. Therefore, statically built soundhw array in arch_init.c stopped
including this card.
Work around this by re-adding this define to config-target.mak.
Long-term, a dynamic creation of this soundhw list will be necessary.
Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
Signed-off-by: malc <av1474@comtv.ru>
* agraf/s390-for-upstream:
s390: provide interface for service interrupt/introduce interrupt.c
s390: Fix error handling and condition code of service call
* agraf/ppc-for-upstream: (24 commits)
openpic: Added BRR1 register
pseries: Update SLOF firmware image
pseries dma: DMA window params added to PHB and DT population changed
pseries: Add PCI MSI/MSI-X support
pseries: Add trace event for PCI irqs
pseries: Export find_phb() utility function for PCI code
pseries: added allocator for a block of IRQs
pseries: Separate PCI RTAS setup from common from emulation specific PCI setup
pseries: Rework irq assignment to avoid carrying qemu_irqs around
pseries: Remove extraneous prints
pseries: Update SLOF
PPC: spapr: Remove global variable
PPC: spapr: Rework VGA select logic
xbzrle: fix compilation on ppc32
spapr: Add support for -vga option
Add one new file vga-pci.h and cleanup on all platforms
Revert "PPC: e500: Use new MPIC dt format"
ppc: Fix bug in handling of PAPR hypercall exits
PPC: e500: add generic e500 platform
PPC: e500: split mpc8544ds machine from generic e500 code
...
mingw32 seems to want the declaration to also carry the weak attribute.
Strangely, gcc on Linux absolutely does not want the declaration to be marked
as weak. This may not be the right fix, but it seems to do the trick.
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
* stefanha/trivial-patches:
ivshmem, qdev-monitor: fix order of qerror parameters
iov_send_recv(): Handle zero bytes case even if OS does not
framebuffer: Fix spelling in comment (leight -> height)
Spelling fix in comment (peripherans -> peripherals)
docs: Fix spelling (propery -> property)
trace: Fix "Qemu" -> "QEMU"
cputlb.c: Fix out of date comment
ehci: fix assertion typo
Makefile: Avoid explicit list of directories in clean target
* kwolf/for-anthony:
qemu-iotests: Fix 030 after switch to GenericError
block: Flush parent to OS with cache=unsafe
iscsi: Fix NULL dereferences / races between task completion and abort
monitor: Clean up fd sets on monitor disconnect
block: Enable qemu_open/close to work with fd sets
block: Convert close calls to qemu_close
block: Convert open calls to qemu_open
block: Prevent detection of /dev/fdset/ as floppy
qapi: Introduce add-fd, remove-fd, query-fdsets
qemu-char: Add MSG_CMSG_CLOEXEC flag to recvmsg
Linux mpic driver uses (changes may be in pipeline to get upstreamed soon)
BRR1. This patch adds the support to emulate readonly FSL BRR1 register.
Currently QEMU does not fully emulate any version on MPIC, so the MPIC
Major number and Minor number are set to 0.
Signed-off-by: Bharat Bhushan <bharat.bhushan@freescale.com>
Signed-off-by: Alexander Graf <agraf@suse.de>
This updates SLOF to handle the necessary device tree properties for MSI
and MSI-X.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Alexander Graf <agraf@suse.de>
Previously the only PCI bus supported was the emulated PCI bus with
fixed DMA window with start at 0 and size 1GB. As we are going to support
PCI pass through which DMA window properties are set by the host
kernel, we have to support DMA windows with parameters other than default.
This patch adds:
1. DMA window properties to sPAPRPHBState: LIOBN (bus id), start,
size of the window.
2. An additional function spapr_dma_dt() to populate DMA window
properties in the device tree which simply accepts all the parameters
and does not try to guess what kind of IOMMU is given to it.
The original spapr_dma_dt() is renamed to spapr_tcet_dma_dt().
Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Alexander Graf <agraf@suse.de>
This patch implements MSI and MSI-X support for the pseries PCI host
bridge. To do this it adds:
* A "config_space_address to msi_table" map, since the MSI RTAS calls
take a PCI config space address as an identifier.
* A MSIX memory region to catch msi_notify()/msix_notiry() from
virtio-pci and pass them to the guest via qemu_irq_pulse().
* RTAS call "ibm,change-msi" which sets up MSI vectors for a
device. Note that this call may configure and return lesser number of
vectors than requested.
* RTAS call "ibm,query-interrupt-source-number" which translates MSI
vector to interrupt controller (XICS) IRQ number.
Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
[agraf: fix error case ndev < 0]
Signed-off-by: Alexander Graf <agraf@suse.de>
This adds a trace event in the pseries PCI specific set_irq() function to
assist in debugging.
Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
[agraf: add trace.h include]
Signed-off-by: Alexander Graf <agraf@suse.de>
The pseries PCI code makes use of an internal find_dev() function which
locates a PCIDevice * given a (platform specific) bus ID and device
address. Internally this needs to first locate the host bridge on which
the device resides based on the bus ID. This patch exposes that host
bridge lookup as a separate function, which we will need later in the MSI
and VFIO code.
Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
[agraf: drop trace.h inclusion]
Signed-off-by: Alexander Graf <agraf@suse.de>
The patch adds a simple helper which allocates a consecutive sequence
of IRQs calling spapr_allocate_irq for each and checks that allocated
IRQs go consequently.
The patch is required for upcoming support of MSI/MSIX on POWER.
Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Alexander Graf <agraf@suse.de>
Currently the RTAS functions for handling PCI are registered from the
class init code for the PCI host bridge. That sort of makes sense
now, but will break in the future when vfio gives us multiple types of
host bridge for pseries (emulated and pass-through, at least). The
RTAS functions will be common across all host bridge types (and will
call out to different places internally depending on the type).
So, this patch moves the RTAS registration into its own function
called direct from the machine setup code.
Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Alexander Graf <agraf@suse.de>
Currently, the interfaces in the pseries machine code for assignment
and setup of interrupts pass around qemu_irq objects. That was done
in an attempt not to be too closely linked to the specific XICS
interrupt controller. However interactions with the device tree setup
made that attempt rather futile, and XICS is part of the PAPR spec
anyway, so this really just meant we had to carry both the qemu_irq
pointers and the XICS irq numbers around.
This mess will just get worse when we add upcoming PCI MSI support,
since that will require tracking a bunch more interrupt. Therefore,
this patch reworks the spapr code to just use XICS irq numbers
(roughly equivalent to GSIs on x86) and only retrieve the qemu_irq
pointers from the XICS code when we need them (a trivial lookup).
This is a reworked and generalized version of an earlier spapr_pci
specific patch from Alexey Kardashevskiy.
Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
[agraf: fix checkpath warning]
Signed-off-by: Alexander Graf <agraf@suse.de>
The pseries machine prints several messages to stderr whenever it starts up
and another whenever the vm is reset. It's not normal for qemu machines to
do this though, so this patch removes them. We can put them back
conditional on a DEBUG symbol if we really need them in future.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Alexander Graf <agraf@suse.de>
This patch updates the SLOF version, introducing a number of fixes:
* add proper graphics support
* fix bugs with graphical terminal under grub2
* fix bugs in handling of 64-bit unit addresses
* fix VSCSI representation to be closer to PowerVM
* fix bugs which caused grub2 to crash
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Alexander Graf <agraf@suse.de>
When selecting our VGA adapter, we want to:
* fail completely when we can't satisfy the user's request
* support -nographic where no VGA adapter should be spawned
This patch reworks the logic so we fulfill the two conditions above.
Signed-off-by: Alexander Graf <agraf@suse.de>
When compiling the xbzrle code on my ppc32 user space, I hit the following
gcc compiler warning (treated as an error):
cc1: warnings being treated as errors
savevm.c: In function ‘xbzrle_encode_buffer’:
savevm.c:2476: error: overflow in implicit constant conversion
Fix this by making the cast explicit, rather than implicit.
Signed-off-by: Alexander Graf <agraf@suse.de>
Also instanciate the USB keyboard and mouse when that option is used
(you can still use -device to create individual devices without all
the defaults)
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Signed-off-by: Li Zhang <zhlcindy@linux.vnet.ibm.com>
[agraf: remove USB bits]
Signed-off-by: Alexander Graf <agraf@suse.de>
Functions pci_vga_init() and pci_cirrus_vga_init() are declared
in pc.h. That prevents other platforms (e.g. sPAPR) to use them.
This patch is to create one new file vga-pci.h and move the
declarations to vga-pci.h, so that they can be shared by
all platforms. This patch also cleans up on all platforms.
Signed-off-by: Li Zhang <zhlcindy@linux.vnet.ibm.com>
Signed-off-by: Alexander Graf <agraf@suse.de>
This reverts commit 518c7fb44f. It breaks
new Linux guests with SMP, because IPIs get mapped to large vectors which
our MPIC emulation does not implement.
Conflicts:
hw/ppc/e500.c
Currently for powerpc, kvm_arch_handle_exit() always returns 1, meaning
that its caller - kvm_cpu_exec() - will always exit immediately afterwards
to the loop in qemu_kvm_cpu_thread_fn().
There's no need to do this. Once we've handled the hypercall there's no
reason we can't go straight around and KVM_RUN again, which is what ret = 0
will signal. The only exception might be for hypercalls which affect the
state of cpu_can_run(), however the only one that might do this is H_CEDE
and for kvm that is always handled in the kernel, not qemu.
Furtherm setting ret = 0 means that when exit_requested is set from a
hypercall, we will enter KVM_RUN once more with a signal which lets the
the kernel do its internal logic to complete the hypercall with out
actually executing any more guest code. This is important if our hypercall
also triggered a reset, which previously would re-initialize everything
without completing the hypercall. This caused the kernel to get confused
because it thought the guest was still in the middle of a hypercall when
it has actually been reset.
This patch therefore changes to ret = 0, which is both a bugfix and a small
optimization.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Alexander Graf <agraf@suse.de>
This gives the kernel a paravirtualized machine to target, without
requiring both sides to pretend to be targeting a specific board
that likely has little to do with the host in KVM scenarios. This
avoids the need to add new boards to QEMU, just to be able to
run KVM on new CPUs.
Signed-off-by: Scott Wood <scottwood@freescale.com>
[agraf: conditionalize on CONFIG_FDT]
Signed-off-by: Alexander Graf <agraf@suse.de>
Currently the only mpc8544ds-ism that is factored out is
toplevel compatible and model. In the future the generic e500
code is expected to become more generic.
Signed-off-by: Scott Wood <scottwood@freescale.com>
[agraf: conditionalize on CONFIG_FDT]
Signed-off-by: Alexander Graf <agraf@suse.de>
No functional changes -- machine is still outwardly mpc8544ds.
The references that are not changed contain mpc8544 hardware details that
need to be parameterized if/when a different e500 platform wants to
change them.
Signed-off-by: Scott Wood <scottwood@freescale.com>
Signed-off-by: Alexander Graf <agraf@suse.de>
Rename the file (with no changes other than fixing up the header paths)
in preparation for refactoring into a generic e500 platform. Also move
it into the newly created ppc/ directory.
Signed-off-by: Scott Wood <scottwood@freescale.com>
Reviewed-by: Andreas Färber <afaerber@suse.de>
[agraf: conditionalize on CONFIG_FDT]
Signed-off-by: Alexander Graf <agraf@suse.de>
spapr_populate_pci_devices() populates the device tree only with bus
properties and has nothing to do with the devices on it as PCI BAR
allocation is done by the system firmware (SLOF).
New name - spapr_populate_pci_dt() - describes the functionality better.
Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Signed-off-by: Alexander Graf <agraf@suse.de>
The PCIHostState struct already contains SysBusDevice so
the one in sPAPRPHBState has to go.
Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Signed-off-by: Alexander Graf <agraf@suse.de>
* qemu-kvm/uq/master:
update-linux-headers.sh: Pull in asm-generic/kvm_para.h
kvmvapic: Disable if there is insufficient memory
kvm: i8254: Finish time conversion fix
kvm: i8254: Cache kernel clock offset in KVMPITState
This patch creates interrupt.c. The first user is a callback for hw/*
code to trigger an service interrupt for a given sccb value. Several
interrupt types for s390 are floating (can be delivered to all CPUs).
so this code does not belong to a specific CPU.
Other interrupts (like the virtio one) are also floating and can be
moved here later on.
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
Signed-off-by: Alexander Graf <agraf@suse.de>
Invalid sccb addresses will cause specification or addressing exception.
Lets add those checks. Furthermore, the good case (cc=0) was incorrect
for KVM, we did not set the CC at all. We now use return codes < 0
as program checks and return codes > 0 as condition code values.
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
Signed-off-by: Alexander Graf <agraf@suse.de>
Now that the QERR_ macros no longer contain a json dictionary,
the order of some parameters needs to be fixed for them to appear
correctly.
Signed-off-by: Alberto Garcia <agarcia@igalia.com>
Reviewed-by: Luiz Capitulino <lcapitulino@redhat.com>
Signed-off-by: Stefan Hajnoczi <stefanha@linux.vnet.ibm.com>
