gus: Do not manually free the state, qdev does it for us

Signed-off-by: malc <av1474@comtv.ru>

oss: Unbreak mmaping the ability to mmap oss fd on Linux

Signed-off-by: malc <av1474@comtv.ru>

kvm_arch_get_registers() shouldn't be called directly

Direct call to kvm_arch_get_registers() bypass logic in
cpu_synchronize_state()

Signed-off-by: Gleb Natapov <gleb@redhat.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>

vmstate: port cirrus_vga device

Signed-off-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>

vmstate: port serial device

Signed-off-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>

vmstate: port hpet device

Signed-off-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>

hpet: it is imposible that qemu_timer field is NULL at this point

Signed-off-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>

vmstate: port ioapic device

Signed-off-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>

vmstate: add uint64 array support

Signed-off-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>

vmstate: port pckbd device

Signed-off-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>

vmstate: port vmmouse device

Signed-off-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>

vmstate: port dma device

Signed-off-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>

vmstate: add support for arrays of uint16_t

Signed-off-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>

vmstate: port fdc device

We can't move fifo back to an embeded array because it needs to be aligned

Signed-off-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>

vmstate: add support for uint8_t equal

Signed-off-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>

vmstate: port i8259 device

Signed-off-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>

vmstate: port fw_cfg device

Signed-off-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>

vmstate: port cpu_comon

Signed-off-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>

vmstate: Add pre/post_save() hooks

Signed-off-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>

vmstate: Add pre_load() hook

Signed-off-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>

vmstate: rename run_after_load() -> post_load()

This naming was used in kvm tree, and is easier to remember

Signed-off-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>

vmstate: add sensible arguments to vmstate_unregister()

vmsd alone is not enugh, because we can have several structs saved with the same description (vmsd).

Signed-off-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>

Make get_ticks_per_sec() a static inline

ticks_per_sec is a constant.  There's no need to store it as a variable as it
never changes since our time is based on units.

Convert get_ticks_per_sec() to a static inline and move the constant into
qemu-timer.h.  Remove all references to QEMU_TIMER_BASE so that we consistently
use this interface.

Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>

timers: move them to VMState

Signed-off-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>

timers: Createt TimersState and put all timers state there

Signed-off-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>

Unexport ticks_per_sec variable. Create get_ticks_per_sec() function

Signed-off-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>

timers: remove useless check

loadvm_state is called from: vl.c during startup, vmstart() is called after finishing loading.  The other caller do_loadvm() does the call after a vm_stop().  At both places where we can be saving state we are stoped a few lines before

Signed-off-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>

Remove SaveVM v2 support

In previosu series I remove v2 support for RAM (that was the version that was
supported when SaveVM v3 appeared).  Now we can't load RAM for any image saved in SaveVM v2, we can as well remove SaveVM v2 entirely.

Note: That SaveVM RAM was at v2 when General SaveVM support went from v2 to v3 makes talking about versions confusing at least

Signed-off-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>

ram: Remove SaveVM Version 2 support

It don't work.  It fails in this check

        if (qemu_get_be32(f) != last_ram_offset)

With 512MB of ram, values were for me:

v = 20c00000 last_ram_offset = 20840000

Last time that some code changed that was this one.

    commit 94a6b54fd6d2d3321066cb4db7abeeb417af9365
    Implement dynamic guest ram allocation.
    (I.e. it has been broken since at least April)

Going back to the previous commit, ram load correctly, but vga screen gets
corrupted and ide don't load correctly.  At this point I decide that removing
support is the only viable thing.

The last user of the ram_compress_* were RAM_SAVE_FLAG_FULL flag, but
that flag was never ever been stored in an image. Mark the flag obsolete
and remove the functions.

Signed-off-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>

ram: remove support for loading v1

Signed-off-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>

move mux focus field from CharDriverState to MuxDriver

Now that monitor stopped using focus we can make it internal
to the mux driver.

Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>

monitor: fix muxing

make the mux driver send mux_in and mux_out events when switching
focus while hooking up more handlers.

stop using CharDriverState->focus in monitor.c, track state using
the mux events instead.  This also removes the implicit assumtion
that a muxed monitor allways has mux channel 0.

Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>

qdev: add parser for chardev properties

Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>

Allow -serial chardev:<name>

Lets put -chardev into use now.  With this patch applied chardev:name is
accepted as chardev specification everywhere, i.e. now you can:

-chardev stdio,id=ttyS0
-serial chardev:ttyS0

which does the same as '-serial stdio".

Muxing can be done this way:

-chardev stdio,id=mux,mux=on
-serial chardev:mux
-monitor chardev:mux

You can mux more than two streams.

Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>

convert udp chardev to QemuOpts.

While being at it: create a new inet_dgram_opts() function for udp setup,
so udp can handle IPv6 now.

new cmd line syntax:
    -chardev udp,id=name,host=remotehost,port=remoteport,\
localaddr=bindaddr,localport=bindport

Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>

convert mux chardev to QemuOpts.

new cmd line syntax:  you can add mux=1 to any chardev to enable muxing,
then attach it multiple times, like this:

-chardev pty,name=mux,mux=on

Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>

convert vc chardev to QemuOpts.

new cmd line syntax:
    -chardev vc,id=name
    -chardev vc,id=name,width=pixels,height=pixels
    -chardev vc,id=name,cols=chars,rows=chars

Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>

convert tty + parport chardevs to QemuOpts.

new cmd line syntax:
    -chardev tty,id=name,path=/dev/tty*
    -chardev parport,id=name,path=/dev/parport*

Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>

convert windows console chardev to QemuOpts.

Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>

convert braille chardev to QemuOpts.

Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>

convert msmouse chardev to QemuOpts.

Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>

convert stdio chardev to QemuOpts.

Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>

convert pty chardev to QemuOpts.

Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>

convert unix+tcp chardevs to QemuOpts.

new cmd line syntax:
  unix socket:
    -chardev socket,id=name,path=/path/to/socket
  tcp socket:
    -chardev socket,id=name,host=hostaddr|ipaddr,port=portnr

server and nowait options work as usual.  Alternatively you can use
server=[on|off] + wait=[on|off] syntax.

Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>

sockets: add inet_listen_opts

Add inet_listen_opts().  Does the same as inet_listen(), but uses
QemuOpts.  inet_listen() is a compatibility wrapper for
inet_listen_opts() now and should go away some day.

Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>

sockets: add inet_connect_opts

Add inet_connect_opts().  Does the same as inet_connect(), but uses
QemuOpts.  inet_connect() is a compatibility wrapper for
inet_connect_opts() now and should go away some day.

Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>

sockets: add unix_*_opts for windows.

Add unix_*_opts function dummys for windows.

Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>

sockets: add unix_listen_opts

Add unix_listen_opts().  Does the same as unix_listen(), but uses
QemuOpts.  unix_listen() is a compatibility wrapper for
unix_listen_opts() now and should go away some day.

Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>

sockets: add unix_connect_opts

Add unix_connect_opts().  Does the same as unix_connect(), but uses
QemuOpts.  unix_connect() is a compatibility wrapper for
unix_connect_opts() now and should go away some day.

Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>

convert file+pipe chardevs to QemuOpts.

new cmd line syntax:
    -chardev file,id=name,path=/path/to/file
    -chardev pipe,id=name,path=/path/to/pipe

Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>

switch chardev to QemuOpts: infrastructure, null device

start switching chardevs to QemuOpts.  This patch adds the
infrastructure and converts the null device.

The patch brings two new functions:

qemu_chr_open_opts()
same as qemu_chr_open(), but uses QemuOpts instead of a
option char string.

qemu_chr_parse_compat()
accepts a traditional chardev option string, returns the
corresponding QemuOpts instance, to handle backward
compatibility.

The patch also adds a new -chardev switch which can be used to create
named+unconnected chardevs, like this:

-chardev null,id=test

This uses the new qemu_chr_open_opts.  Thus with this patch alone only
the null device works.  The other devices will follow ...

Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>

qemu-option.h include protectors

qemu-option.h has no protection against including it twice.
This patch adds the usual "#ifndef header" bits.

Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>

QemuOpts: split option parser into two functions.

looking for id= and creating a new QemuOpts instance is splitted from
the actual option parser code now, so the parser can be called from
other contexts too.

Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>

qdev: Fix i6300 upcast

Use DO_UPCAST() instead of container_of() to go from PCIDevice to
I6300State.  This ensures that PCIDevice is the first member of struct
I6300State.

Signed-off-by: Markus Armbruster <armbru@redhat.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>

VirtIO: Fix QEMU crash during Windows PNP tests

Hello,

In some cases bus driver can deassert "bus master" bit in PCI command
register. The driver will no longer be able to update related registers in
the device. Eventually it will cause QEMU to exit in "virtqueue_num_heads"
function.

Attached path that fixes the described issue.

Best regards,
Yan Vugenfirer.

>From 3fdafbdfad676ec8479dc073cff70bf356868bfe Mon Sep 17 00:00:00 2001
From: Yan Vugenfirer <yvugenfi@redhat.com>
Date: Tue, 8 Sep 2009 10:08:14 -0400
Subject: [PATCH] VirtIO: Fix QEMU crash during Windows PNP tests

Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>

virtio-blk: add volatile writecache feature

Add a new VIRTIO_BLK_F_WCACHE feature to virtio-blk to indicate that we have
a volatile write cache that needs controlled flushing.  Implement a
VIRTIO_BLK_T_FLUSH operation to flush it.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>

ide: use bdrv_aio_flush

Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>

block: add aio_flush operation

Instead stalling the VCPU while serving a cache flush try to do it
asynchronously.  Use our good old helper thread pool to issue an
asynchronous fdatasync for raw-posix.  Note that while Linux AIO
implements a fdatasync operation it is not useful for us because
it isn't actually implement in asynchronous fashion.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>

block: use fdatasync instead of fsync if possible

If we are flushing the caches for our image files we only care about the
data (including the metadata required for accessing it) but not things
like timestamp updates.  So try to use fdatasync instead of fsync to
implement the flush operations.

Unfortunately many operating systems still do not support fdatasync,
so we add a qemu_fdatasync wrapper that uses fdatasync if available
as per the _POSIX_SYNCHRONIZED_IO feature macro or fsync otherwise.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>

block: add enable_write_cache flag

Add a enable_write_cache flag in the block driver state, and use it to
decide if we claim to have a volatile write cache that needs controlled
flushing from the guest.  The flag is off if cache=writethrough is
defined because O_DSYNC guarantees that every write goes to stable
storage, and it is on for cache=none and cache=writeback.

Both scsi-disk and ide now use the new flage, changing from their
defaults of always off (ide) or always on (scsi-disk).

Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>

qemu: init all queues to NO_VECTOR value

initialize vectors for all vqs to VIRTIO_NO_VECTOR rather than 0 which
is a valid vector. This fixes migration which happened before driver
was loaded.

Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Reported-by: Amit Shah <amit.shah@redhat.com>
Tested-by: Amit Shah <amit.shah@redhat.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>

qemu: make virtio-blk PCI compliant by default

commit bf011293faaa7f87e4de83185931e7411b794128 made virtio-blk-pci not
PCI-compliant, since it makes region 0 (which is an i/o region)
size > 256, and, since PCI 2.1, i/o regions are limited to 256 bytes size.

When the ATA serial number feature is off, which is the default,
make the device spec compliant again, by making region 0 smaller.

Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Reported-by: Vadim Rozenfeld <vrozenfe@redhat.com>
Tested-by: Vadim Rozenfeld <vrozenfe@redhat.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>

usb-linux.c: fix buffer overflow

In usb-linux.c:usb_host_handle_control, we pass a 1024-byte buffer and
length to the kernel.  However, the length was provided by the caller
of dev->handle_packet, and is not checked, so the kernel might provide
too much data and overflow our buffer.

For example, hw/usb-uhci.c could set the length to 2047.
hw/usb-ohci.c looks like it might go up to 4096 or 8192.

This causes a qemu crash, as reported here:
  http://www.mail-archive.com/kvm@vger.kernel.org/msg18447.html

This patch increases the usb-linux.c buffer size to 2048 to fix the
specific device reported, and adds a check to avoid the overflow in
any case.

Signed-off-by: Jim Paris <jim@jtan.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>

Fix conditional compilation (MIPS host)

Compilation for MIPS host (not part of official QEMU)
checks __mips_isa_rev which is not always defined.

Signed-off-by: Stefan Weil <weil@mail.berlios.de>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>

virtio-blk: Use bdrv_aio_multiwrite

It is quite common for virtio-blk to submit more than one write request in a
row to the qemu block layer. Use bdrv_aio_multiwrite to allow block drivers to
optimize its handling of the requests.

Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>

Add bdrv_aio_multiwrite

One performance problem of qcow2 during the initial image growth are
sequential writes that are not cluster aligned. In this case, when a first
requests requires to allocate a new cluster but writes only to the first
couple of sectors in that cluster, the rest of the cluster is zeroed - just
to be overwritten by the following second request that fills up the cluster.

Let's try to merge sequential write requests to the same cluster, so we can
avoid to write the zero padding to the disk in the first place.

As a nice side effect, also other formats take advantage of dealing with less
and larger requests.

Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>

Revert "don't call cpu_sychronize_state from reset handlers"

This reverts commit 733318ea9c6d846a6a047b87619e7d9d6e9707d1.

Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>

microblaze: Trap if QEMU finds an unknown insns.

If PVR settings enable illegal insn trap, trap when QEMU finds an
insn it knows nothing about.

Signed-off-by: Edgar E. Iglesias <edgar.iglesias@gmail.com>

microblaze: Correct prio between MMU and unaligned exceptions.

The microblaze gives MMU faults priority. For stores we still
have a flaw that the value leaks to memory in the case of an
unaligned exception.

Signed-off-by: Edgar E. Iglesias <edgar.iglesias@gmail.com>

microblaze: HW Exception fixes.

* Correct PVR checks for masking off individual exceptions.
* Correct FPU exception code.
* Set EAR on unaligned and unassigned exceptions.

Signed-off-by: Edgar E. Iglesias <edgar.iglesias@gmail.com>

Update OpenBIOS images to r577

Signed-off-by: Aurelien Jarno <aurelien@aurel32.net>

X86_64: Use proper jumps/calls when displacement exceeds +-2G

Signed-off-by: malc <av1474@comtv.ru>

Remove bit-rotten threshold handling

Thanks to Toshiya Takeda for bringing up an unrelated issue which led
to this.

Signed-off-by: malc <av1474@comtv.ru>

F_DUPFD_CLOEXEC is not universally available

The same issue (and the same patch to the byte) was experienced/proposed
by Vince Weaver.

Signed-off-by: malc <av1474@comtv.ru>

Add information w.r.t default GUS IRQ assigment

Fix formatting

Fix formatting, get rid of conf and fix description

Fix formatting and and description field

qdev/isa: convert real time clock

Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>

qdev/isa: finish pckbd conversion

drop old init path and switch remaining users to
isa_create_simple().

Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>

qdev/isa: convert ne2000

Also split the isa bits into a separate source file, so we don't drag in
a dependency for isa-bus.o for machines which want ne2k_pci only.

Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>

qdev/isa: convert gravis ultrasound

Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>

qdev/isa: convert cs4231a sound card

Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>

qdev/isa: convert soundblaster

Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>

qdev: add isa_create() function

Like isa_create_simple, but doesn't call qdev_init, so one can set
properties after creating and before initializing the device.

Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>

qdev: tag isabus-bridge as no-user

isabus-bridge isn't supposed to be added via -device ...

Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>

qdev: simplify isa irq assignments

isa-bus owns the isa irqs now, so it can hand them out directly.
There is no need for the separate isa_connect_irqs step, drop it.

Also hard-code isa interrupts which can't be configured anyway.

Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>

qdev: drop iobase properties from isa bus

Lot of ISA devices work at fixed addresses, so having iobase
as bus property doesn't make much sense.  Devices which can
have different iobases will get a device property.

Also simply hard-code stuff which can't be configured anyway.

Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>

qdev: add property type for 32bit signed integers.

Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>

isapc: pick a more sane default cpu for such old hardware.

Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>

isapc: Fix irq routing

Only send irqs to ioapic in case we have one.
Fixes qemu segfault.

Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>

Allow to unscale the output window with a Ctrl-Alt-u hotkey

Signed-off-by: malc <av1474@comtv.ru>

Fix VMSTATE_PCI_DEVICE version

PCI device entries have to have a default version, not 2, because they are
used in the midle of other structures that can have _any_ version number.

We can't use proper versioning here until we have SubSections support.
Why we didn't noticed before?  Because in a PC, the only device ported with
a version less that 2 is piix_pm, and for that one, default pci values are
right.  If you use a virtio-console, you will see that its state it is not
loaded back.

Thanks to Amit Shah for reporting the problem and help debug the fix.

Signed-off-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>

don't call cpu_sychronize_state from reset handlers

Doing this will make the vcpu ioctl be issued from the I/O thread, instead
of cpu thread. The correct behaviour is to call it from within the cpu thread,
as soon as we are ready to go.

Signed-off-by: Glauber Costa <glommer@redhat.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>

RTC polling mode broken

The RTC emulation does not set the IRQ flags independent of the IRQ enable bits.

The original MC146818A datasheet from 1984 notes:
        "flag bits in Register C [...] are set independent of the
        state of the corresponding enable bits in Register B"
Similar sections can be found in newer documentation e.g. in rtc82885.

Qemu and Bochs set the IRQ flags only if they are enabled,
which breaks drivers polling on them.

The following patch corrects this for the update-ended-flag in Qemu only.
It does not fix the handling of the other flags.

Signed-off-by: Bernhard Kauer <kauer@tudos.org>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>

qemu-kvm: fix segfault when running kvm without /dev/kvm, falling back to non-accelerated mode

qemu-kvm: fix segfault when running kvm without /dev/kvm, falling back
to non-accelerated mode

We're seeing segfaults on systems without access to /dev/kvm.  It
looks like the global kvm_allowed is being set just a little too late
in vl.c.  This patch moves the kvm initialization a bit higher in the
vl.c main, just after options processing, and solves the segfaults.
We're carrying this patch in Ubuntu 9.10 Alpha.  Please apply
upstream, or advise if and why this might not be the optimal solution.

Signed-off-by: Dustin Kirkland <kirkland@canonical.com>
Move the kvm_init() call a bit higher to fix a segfault when
/dev/kvm is not available.  The kvm_allowed global needs
to be set correctly a little earlier.

Signed-off-by: Dustin Kirkland <kirkland@canonical.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>

configure: fix Linux AIO detection

We should set $linux_aio to 'no' if detection failed, otherwise
its contents will be empty, which is a bug as we test for 'yes'
or 'no'.

Signed-off-by: Luiz Capitulino <lcapitulino@redhat.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>

Fix compilation warnings when DEBUG_BUFFERED_FILE is defined

gcc 4.3.2 throws warnings when DEBUG_BUFFERED_FILE is defined, because
we are using the wrong format specifiers to print size_t/ssize_t values.

Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>

qcow2: Order concurrent AIO requests on the same unallocated cluster

When two AIO requests write to the same cluster, and this cluster is
unallocated, currently both requests allocate a new cluster and the second one
merges the first one when it is completed. This means an cluster allocation, a
read and a cluster deallocation which cause some overhead. If we simply let the
second request wait until the first one is done, we improve overall performance
with AIO requests (specifially, qcow2/virtio combinations).

This patch maintains a list of in-flight requests that have allocated new
clusters. A second request touching the same cluster is limited so that it
either doesn't touch the allocation of the first request (so it can have a
non-overlapping allocation) or it waits for the first request to complete.

Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>

qcow2: Fix metadata preallocation

The wrong version of the preallocation patch has been applied, so this is the
remaining diff.

We can't use truncate to grow the image file to the right size because we don't
know if metadata has been written after the last data cluster. In this case
truncate would shrink the file and destroy its metadata. Write a zero sector at
the end of the virtual disk instead to ensure that the file is big enough.

Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>