sdk/emulator/qemu.git
11 years agoMerge remote-tracking branch 'kraxel/testdev.1' into staging
Anthony Liguori [Fri, 4 Jan 2013 19:25:07 +0000 (13:25 -0600)]
Merge remote-tracking branch 'kraxel/testdev.1' into staging

* kraxel/testdev.1:
  pc: remove bochs bios debug ports
  hw: Add test device for unittests execution
  add isa-debug-exit device.
  switch debugcon to memory api

Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
11 years agoMerge remote-tracking branch 'kraxel/acpi.2' into staging
Anthony Liguori [Fri, 4 Jan 2013 19:24:49 +0000 (13:24 -0600)]
Merge remote-tracking branch 'kraxel/acpi.2' into staging

* kraxel/acpi.2:
  apci: assign memory regions to ich9 lpc device
  apci: assign memory regions to piix4 acpi device
  acpi: autoload dsdt
  configure: also symlink *.aml files

Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
11 years agopc: remove bochs bios debug ports
Gerd Hoffmann [Tue, 11 Dec 2012 08:59:55 +0000 (09:59 +0100)]
pc: remove bochs bios debug ports

Prehistoric leftover, zap it.  We poweroff via acpi these days.

And having a port (0x501,0x502) where any random guest write will make
qemu exit -- with no way to turn it off -- is a bad joke anyway.

Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
11 years agohw: Add test device for unittests execution
Lucas Meneghel Rodrigues [Thu, 13 Dec 2012 14:48:53 +0000 (12:48 -0200)]
hw: Add test device for unittests execution

Add a test device which supports the kvmctl ioports,
so one can run the KVM unittest suite.

Intended Usage:

qemu-system-x86_64 -nographic \
    -device pc-testdev \
    -device isa-debug-exit,iobase=0xf4,iosize=0x04 \
    -kernel /path/to/kvm/unittests/msr.flat

Where msr.flat is one of the KVM unittests, present on a
separate repo,

git://git.kernel.org/pub/scm/virt/kvm/kvm-unit-tests.git

[ kraxel: more memory api + qom fixes ]

CC: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Alexander Graf <agraf@suse.de>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
Signed-off-by: Lucas Meneghel Rodrigues <lmr@redhat.com>
Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
11 years agoadd isa-debug-exit device.
Gerd Hoffmann [Wed, 12 Dec 2012 14:54:59 +0000 (15:54 +0100)]
add isa-debug-exit device.

When present it makes qemu exit on any write.
Mapped to port 0x501 by default.

Without this patch Anthony doesn't allow me to
remove the bochs bios debug ports because his
test suite uses this.

Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
11 years agoswitch debugcon to memory api
Gerd Hoffmann [Wed, 12 Dec 2012 14:43:35 +0000 (15:43 +0100)]
switch debugcon to memory api

Also some QOM glue while being at it.

Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
11 years agoapci: assign memory regions to ich9 lpc device
Gerd Hoffmann [Tue, 11 Dec 2012 08:42:18 +0000 (09:42 +0100)]
apci: assign memory regions to ich9 lpc device

Get rid of get_system_io() usage.

Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
11 years agoapci: assign memory regions to piix4 acpi device
Gerd Hoffmann [Tue, 11 Dec 2012 08:40:45 +0000 (09:40 +0100)]
apci: assign memory regions to piix4 acpi device

Get rid of get_system_io() usage.

Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
11 years agoacpi: autoload dsdt
Gerd Hoffmann [Mon, 3 Dec 2012 09:47:27 +0000 (10:47 +0100)]
acpi: autoload dsdt

Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
11 years agoconfigure: also symlink *.aml files
Gerd Hoffmann [Mon, 3 Dec 2012 09:45:49 +0000 (10:45 +0100)]
configure: also symlink *.aml files

Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
11 years agopty: unbreak libvirt
Gerd Hoffmann [Thu, 3 Jan 2013 13:23:03 +0000 (14:23 +0100)]
pty: unbreak libvirt

Commit 586502189edf9fd0f89a83de96717a2ea826fdb0 breaks libvirt pty
support because it tried to figure the pts name from stderr output.

Fix this by moving the label to the end of the line, this way the
libvirt parser does still recognise the message.  libvirt looks
for "char device redirected to ${ptsname}<whitespace>".

Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
11 years agodataplane: use linux-headers/ for virtio includes
Stefan Hajnoczi [Thu, 3 Jan 2013 10:56:16 +0000 (11:56 +0100)]
dataplane: use linux-headers/ for virtio includes

The hw/dataplane/vring.c code includes linux/virtio_ring.h.  Ensure that
we use linux-headers/ instead of the system-wide headers, which may be
out-of-date on older distros.

This resolves the following build error on Debian 6:

  CC    hw/dataplane/vring.o
cc1: warnings being treated as errors
hw/dataplane/vring.c: In function 'vring_enable_notification':
hw/dataplane/vring.c:71: error: implicit declaration of function 'vring_avail_event'
hw/dataplane/vring.c:71: error: nested extern declaration of 'vring_avail_event'
hw/dataplane/vring.c:71: error: lvalue required as left operand of assignment

Note that we now build dataplane/ for each target instead of only once.
There is no way around this since linux-headers/ is only available for
per-target objects - and it's how virtio, vfio, kvm, and friends are
built.

Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
11 years agoconfigure: Write new file "config-all-disas.mak" when running configure
Stefan Weil [Tue, 1 Jan 2013 17:33:44 +0000 (18:33 +0100)]
configure: Write new file "config-all-disas.mak" when running configure

Incremental builds added new lines to that file each time when configure
was run.

Now a new file with a comment line is written.

Signed-off-by: Stefan Weil <sw@weilnetz.de>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
11 years agotci: Fix broken builds with TCG interpreter
Stefan Weil [Tue, 1 Jan 2013 17:43:56 +0000 (18:43 +0100)]
tci: Fix broken builds with TCG interpreter

TCI no longer compiled after commit 76cad71136b7eb371cf2a2a4e1621cfe8d9c769a.

The TCI disassembler depends on data structures which are different for
each QEMU target, so it cannot be compiled as a universal-obj today.

Signed-off-by: Stefan Weil <sw@weilnetz.de>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
11 years agosavevm.c: cleanup system includes
Michael Tokarev [Mon, 31 Dec 2012 11:30:31 +0000 (15:30 +0400)]
savevm.c: cleanup system includes

savevm.c suffers from the same problem as some other files.
Some years ago savevm.c was created from vl.c, moving some
code from there into a separate file.  At that time, all
includes were just copied from vl.c to savevm.c, without
checking which ones are needed and which are not.

But actually most of that stuff is _not_ needed.  More, some
stuff is wrong, for example, *BSD #ifdef'ery around <util.h>
vs <libutil.h> - for one, it fails to build on Debian/kFreebsd.

Just remove all this.  Maybe there's a possibility to clean
it up further - like removing <windows.h> (and maybe including
winsock.h for htons etc), and maybe it's possible to remove
some internal #includes too, but I didn't check this.

While at it, remove duplicate #include of qemu/timer.h.

Signed-off-by: Michael Tokarev <mjt@tls.msk.ru>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
11 years agodisallow -daemonize usage of stdio (curses display, -nographic, -serial stdio etc)
Michael Tokarev [Sun, 30 Dec 2012 08:48:14 +0000 (12:48 +0400)]
disallow -daemonize usage of stdio (curses display, -nographic, -serial stdio etc)

Curses display requires stdin/out to stay on the terminal,
so -daemonize makes no sense in this case.  Instead of
leaving display uninitialized like is done since 995ee2bf469de6bb,
explicitly detect this case earlier and error out.

-nographic can actually be used with -daemonize, by redirecting
everything to a null device, but the problem is that according
to documentation and historical behavour, -nographic redirects
guest ports to stdin/out, which, again, makes no sense in case
of -daemonize.  Since -nographic is a legacy option, don't bother
fixing this case (to allow -nographic and -daemonize by redirecting
guest ports to null instead of stdin/out in this case), but disallow
it completely instead, to stop garbling host terminal.

If no display display needed and user wants to use -nographic,
the right way to go is to use
  -serial null -parallel null -monitor none -display none -vga none
instead of -nographic.

Also prevent the same issue -- it was possible to get garbled
host tty after

  -nographic -daemonize

and it is still possible to have it by using

  -serial stdio -daemonize

Fix this by disallowing opening stdio chardev when -daemonize
is specified.

Signed-off-by: Michael Tokarev <mjt@tls.msk.ru>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
11 years agoMerge remote-tracking branch 'stefanha/block' into staging
Anthony Liguori [Wed, 2 Jan 2013 18:19:27 +0000 (12:19 -0600)]
Merge remote-tracking branch 'stefanha/block' into staging

* stefanha/block:
  sheepdog: pass oid directly to send_pending_req()
  sheepdog: don't update inode when create_and_write fails
  block/raw-win32: Fix compiler warnings (wrong format specifiers)
  qemu-img: report size overflow error message
  cutils: change strtosz_suffix_unit function
  virtio-blk: Return UNSUPP for unknown request types
  virtio-blk: add x-data-plane=on|off performance feature
  dataplane: add virtio-blk data plane code
  virtio-blk: restore VirtIOBlkConf->config_wce flag
  iov: add qemu_iovec_concat_iov()
  test-iov: add iov_discard_front/back() testcases
  iov: add iov_discard_front/back() to remove data
  dataplane: add Linux AIO request queue
  dataplane: add event loop
  dataplane: add virtqueue vring code
  dataplane: add host memory mapping code
  configure: add CONFIG_VIRTIO_BLK_DATA_PLANE
  raw-posix: add raw_get_aio_fd() for virtio-blk-data-plane

Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
11 years agotcg: Remove unneeded assertion
Stefan Weil [Sun, 30 Dec 2012 07:20:13 +0000 (08:20 +0100)]
tcg: Remove unneeded assertion

Commit 7f6f0ae5b95adfa76e10eabe2c34424a955fd10c added two assertions.

One of these assertions is not needed:
The pointer ts is never NULL because it is initialized with the
address of an array element.

Reviewed-by: Richard Henderson <rth@twiddle.net>
Signed-off-by: Stefan Weil <sw@weilnetz.de>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
11 years agosheepdog: pass oid directly to send_pending_req()
Liu Yuan [Mon, 17 Dec 2012 06:17:27 +0000 (14:17 +0800)]
sheepdog: pass oid directly to send_pending_req()

Cc: MORITA Kazutaka <morita.kazutaka@lab.ntt.co.jp>
Cc: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Liu Yuan <tailai.ly@taobao.com>
Reviewed-by: MORITA Kazutaka <morita.kazutaka@lab.ntt.co.jp>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
11 years agosheepdog: don't update inode when create_and_write fails
Liu Yuan [Mon, 17 Dec 2012 06:17:26 +0000 (14:17 +0800)]
sheepdog: don't update inode when create_and_write fails

For the error case such as SD_RES_NO_SPACE, we shouldn't update the inode bitmap
to avoid the scenario that the object is allocated but wasn't created at the
server side. This will result in VM's IO error on the failed object.

Cc: MORITA Kazutaka <morita.kazutaka@lab.ntt.co.jp>
Cc: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Liu Yuan <tailai.ly@taobao.com>
Reviewed-by: MORITA Kazutaka <morita.kazutaka@lab.ntt.co.jp>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
11 years agoblock/raw-win32: Fix compiler warnings (wrong format specifiers)
Stefan Weil [Mon, 17 Dec 2012 19:40:01 +0000 (20:40 +0100)]
block/raw-win32: Fix compiler warnings (wrong format specifiers)

Commit fbcad04d6bfdff937536eb23088a01a280a1a3af added fprintf statements
with wrong format specifiers.

GetLastError() returns a DWORD which is unsigned long, so %lu must be used.

Signed-off-by: Stefan Weil <sw@weilnetz.de>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
11 years agoqemu-img: report size overflow error message
liguang [Mon, 17 Dec 2012 01:49:23 +0000 (09:49 +0800)]
qemu-img: report size overflow error message

qemu-img will complain when qcow or qcow2
size overflow for 64 bits, report the right
message in this condition.

$./qemu-img create -f qcow2 /tmp/foo 0x10000000000000000
before change:
qemu-img: Invalid image size specified! You may use k, M, G or T suffixes for
qemu-img: kilobytes, megabytes, gigabytes and terabytes.

after change:
qemu-img: Image size must be less than 8 EiB!

[Resolved conflict with a9300911 goto removal -- Stefan]

Signed-off-by: liguang <lig.fnst@cn.fujitsu.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
11 years agocutils: change strtosz_suffix_unit function
liguang [Mon, 17 Dec 2012 01:49:22 +0000 (09:49 +0800)]
cutils: change strtosz_suffix_unit function

if value to be translated is larger than INT64_MAX,
this function will not be convenient for caller to
be aware of it, so change a little for this.

Signed-off-by: liguang <lig.fnst@cn.fujitsu.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
11 years agovirtio-blk: Return UNSUPP for unknown request types
Alexey Zaytsev [Thu, 13 Dec 2012 07:03:43 +0000 (09:03 +0200)]
virtio-blk: Return UNSUPP for unknown request types

Currently, all unknown requests are treated as VIRTIO_BLK_T_IN

Signed-off-by: Alexey Zaytsev <alexey.zaytsev@gmail.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
11 years agovirtio-blk: add x-data-plane=on|off performance feature
Stefan Hajnoczi [Wed, 14 Nov 2012 14:45:38 +0000 (15:45 +0100)]
virtio-blk: add x-data-plane=on|off performance feature

The virtio-blk-data-plane feature is easy to integrate into
hw/virtio-blk.c.  The data plane can be started and stopped similar to
vhost-net.

Users can take advantage of the virtio-blk-data-plane feature using the
new -device virtio-blk-pci,x-data-plane=on property.

The x-data-plane name was chosen because at this stage the feature is
experimental and likely to see changes in the future.

If the VM configuration does not support virtio-blk-data-plane an error
message is printed.  Although we could fall back to regular virtio-blk,
I prefer the explicit approach since it prompts the user to fix their
configuration if they want the performance benefit of
virtio-blk-data-plane.

Limitations:
 * Only format=raw is supported
 * Live migration is not supported
 * Block jobs, hot unplug, and other operations fail with -EBUSY
 * I/O throttling limits are ignored
 * Only Linux hosts are supported due to Linux AIO usage

Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
11 years agodataplane: add virtio-blk data plane code
Stefan Hajnoczi [Wed, 14 Nov 2012 14:39:30 +0000 (15:39 +0100)]
dataplane: add virtio-blk data plane code

virtio-blk-data-plane is a subset implementation of virtio-blk.  It only
handles read, write, and flush requests.  It does this using a dedicated
thread that executes an epoll(2)-based event loop and processes I/O
using Linux AIO.

This approach performs very well but can be used for raw image files
only.  The number of IOPS achieved has been reported to be several times
higher than the existing virtio-blk implementation.

Eventually it should be possible to unify virtio-blk-data-plane with the
main body of QEMU code once the block layer and hardware emulation is
able to run outside the global mutex.

Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
11 years agovirtio-blk: restore VirtIOBlkConf->config_wce flag
Stefan Hajnoczi [Mon, 10 Dec 2012 12:14:39 +0000 (13:14 +0100)]
virtio-blk: restore VirtIOBlkConf->config_wce flag

Two slightly different versions of a patch to conditionally set
VIRTIO_BLK_F_CONFIG_WCE through the "config-wce" qdev property have been
applied (ea776abca and eec7f96c2).  David Gibson
<david@gibson.dropbear.id.au> noticed that the "config-wce"
property is broken as a result and fixed it recently.

The fix sets the host_features VIRTIO_BLK_F_CONFIG_WCE bit from a qdev
property.  Unfortunately, the virtio device then has no chance to test
for the presence of the feature bit during virtio_blk_init().

Therefore, reinstate the VirtIOBlkConf->config_wce flag.  Drop the
duplicate qdev property to set the host_features bit.  The
VirtIOBlkConf->config_wce flag will be used by virtio-blk-data-plane in
a later patch.

Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
11 years agoiov: add qemu_iovec_concat_iov()
Stefan Hajnoczi [Thu, 22 Nov 2012 15:06:06 +0000 (16:06 +0100)]
iov: add qemu_iovec_concat_iov()

The qemu_iovec_concat() function copies a subset of a QEMUIOVector.  The
new qemu_iovec_concat_iov() function does the same for a iov/cnt pair.

It is easy to define qemu_iovec_concat() in terms of
qemu_iovec_concat_iov().  The existing code is mostly unchanged, except
for the assertion src->size >= soffset, which cannot be efficiently
checked upfront on a iov/cnt pair.  Instead we assert upon hitting the
end of src with an unsatisfied soffset.

Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
11 years agotest-iov: add iov_discard_front/back() testcases
Stefan Hajnoczi [Wed, 21 Nov 2012 18:18:26 +0000 (19:18 +0100)]
test-iov: add iov_discard_front/back() testcases

Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
11 years agoiov: add iov_discard_front/back() to remove data
Stefan Hajnoczi [Wed, 21 Nov 2012 16:41:10 +0000 (17:41 +0100)]
iov: add iov_discard_front/back() to remove data

The iov_discard_front/back() functions remove data from the front or
back of the vector.  This is useful when peeling off header/footer
structs.

Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
11 years agodataplane: add Linux AIO request queue
Stefan Hajnoczi [Wed, 14 Nov 2012 14:30:09 +0000 (15:30 +0100)]
dataplane: add Linux AIO request queue

The IOQueue has a pool of iocb structs and a function to add new
read/write requests.  Multiple requests can be added before calling the
submit function to actually tell the host kernel to begin I/O.  This
allows callers to batch requests and submit them in one go.

The actual I/O is performed using Linux AIO.

Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
11 years agodataplane: add event loop
Stefan Hajnoczi [Wed, 14 Nov 2012 14:23:00 +0000 (15:23 +0100)]
dataplane: add event loop

Outside the safety of the global mutex we need to poll on file
descriptors.  I found epoll(2) is a convenient way to do that, although
other options could replace this module in the future (such as an
AioContext-based loop or glib's GMainLoop).

One important feature of this small event loop implementation is that
the loop can be terminated in a thread-safe way.  This allows QEMU to
stop the data plane thread cleanly.

Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
11 years agodataplane: add virtqueue vring code
Stefan Hajnoczi [Wed, 14 Nov 2012 14:15:50 +0000 (15:15 +0100)]
dataplane: add virtqueue vring code

The virtio-blk-data-plane cannot access memory using the usual QEMU
functions since it executes outside the global mutex and the memory APIs
are this time are not thread-safe.

This patch introduces a virtqueue module based on the kernel's vhost
vring code.  The trick is that we map guest memory ahead of time and
access it cheaply outside the global mutex.

Once the hardware emulation code can execute outside the global mutex it
will be possible to drop this code.

Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
11 years agodataplane: add host memory mapping code
Stefan Hajnoczi [Tue, 20 Nov 2012 09:30:08 +0000 (10:30 +0100)]
dataplane: add host memory mapping code

The data plane thread needs to map guest physical addresses to host
pointers.  Normally this is done with cpu_physical_memory_map() but the
function assumes the global mutex is held.  The data plane thread does
not touch the global mutex and therefore needs a thread-safe memory
mapping mechanism.

Hostmem registers a MemoryListener similar to how vhost collects and
pushes memory region information into the kernel.  There is a
fine-grained lock on the regions list which is held during lookup and
when installing a new regions list.

When the physical memory map changes the MemoryListener callbacks are
invoked.  They build up a new list of memory regions which is finally
installed when the list has been completed.

Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
11 years agoconfigure: add CONFIG_VIRTIO_BLK_DATA_PLANE
Stefan Hajnoczi [Wed, 14 Nov 2012 14:04:15 +0000 (15:04 +0100)]
configure: add CONFIG_VIRTIO_BLK_DATA_PLANE

The virtio-blk-data-plane feature only works with Linux AIO.  Therefore
add a ./configure option and necessary checks to implement this
dependency.

Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
11 years agoraw-posix: add raw_get_aio_fd() for virtio-blk-data-plane
Stefan Hajnoczi [Wed, 14 Nov 2012 10:43:23 +0000 (11:43 +0100)]
raw-posix: add raw_get_aio_fd() for virtio-blk-data-plane

The raw_get_aio_fd() function allows virtio-blk-data-plane to get the
file descriptor of a raw image file with Linux AIO enabled.  This
interface is really a layering violation that can be resolved once the
block layer is able to run outside the global mutex - at that point
virtio-blk-data-plane will switch from custom Linux AIO code to using
the block layer.

Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
11 years agoMerge remote-tracking branch 'qemu-kvm/uq/master' into staging
Anthony Liguori [Wed, 2 Jan 2013 14:01:54 +0000 (08:01 -0600)]
Merge remote-tracking branch 'qemu-kvm/uq/master' into staging

* qemu-kvm/uq/master:
  qemu-kvm/pci-assign: 64 bits bar emulation
  target-i386: Enabling IA32_TSC_ADJUST for QEMU KVM guest VMs

Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
11 years agoMerge remote-tracking branch 'mst/tags/for_anthony' into staging
Anthony Liguori [Wed, 2 Jan 2013 14:01:36 +0000 (08:01 -0600)]
Merge remote-tracking branch 'mst/tags/for_anthony' into staging

pci,virtio

This optimizes MSIX handling in virtio-pci.
Also included is pci express capability bugfix.

Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
* mst/tags/for_anthony:
  virtio-pci: don't poll masked vectors
  msix: expose access to masked/pending state
  msi: add API to get notified about pending bit poll
  pcie: Fix bug in pcie_ext_cap_set_next
  virtio: make bindings typesafe

11 years agoMerge remote-tracking branch 'kraxel/seabios-a810e4e' into staging
Anthony Liguori [Wed, 2 Jan 2013 14:01:22 +0000 (08:01 -0600)]
Merge remote-tracking branch 'kraxel/seabios-a810e4e' into staging

* kraxel/seabios-a810e4e:
  Update seabios to a810e4e72a0d42c7bc04eda57382f8e019add901

Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
11 years agopflash_cfi01: Suppress warning when Linux probes for AMD flash
Stefan Weil [Sat, 24 Nov 2012 22:03:13 +0000 (23:03 +0100)]
pflash_cfi01: Suppress warning when Linux probes for AMD flash

There are several ARM and MIPS boards which are manufactured with
either Intel (pflash_cfi01.c) or AMD (pflash_cfi02.c) flash memory.

The Linux kernel supports both and first probes for AMD flash which
resulted in one or two warnings from the Intel flash emulation:

pflash_write: Unimplemented flash cmd sequence (offset 0000000000000000, wcycle 0x0 cmd 0x0 value 0xf000f0)
pflash_write: Unimplemented flash cmd sequence (offset 0000000000000000, wcycle 0x0 cmd 0x0 value 0xf0)

These warnings confuse users, so suppress them.

Signed-off-by: Stefan Weil <sw@weilnetz.de>
Signed-off-by: Aurelien Jarno <aurelien@aurel32.net>
11 years agotarget-mips: Use EXCP_SC rather than a magic number
陳韋任 (Wei-Ren Chen) [Mon, 10 Dec 2012 16:15:55 +0000 (00:15 +0800)]
target-mips: Use EXCP_SC rather than a magic number

  From the discussion on the ML [1], the exception limit defined by
magic number 0x100 is actually EXCP_SC defined in cpu.h. Replace the
magic number with EXCP_SC. Remove "#if 1 .. #endif" as well.

[1] http://lists.gnu.org/archive/html/qemu-devel/2012-11/msg03080.html

Signed-off-by: Chen Wei-Ren <chenwj@iis.sinica.edu.tw>
Signed-off-by: Aurelien Jarno <aurelien@aurel32.net>
11 years agotarget-mips: Make repl_ph to sign extend to target-long
Jovanovic, Petar [Tue, 11 Dec 2012 15:06:35 +0000 (15:06 +0000)]
target-mips: Make repl_ph to sign extend to target-long

The immediate value is 9bits, should sign-extend to 16bits. The return value to
register should sign-extend to target_long, as Richard says, removing an
unnecessary cast works fun.

Signed-off-by: Dongxue Zhang <elta.era@gmail.com>
Signed-off-by: Aurelien Jarno <aurelien@aurel32.net>
11 years agoFix my email address
Dongxue Zhang [Tue, 11 Dec 2012 14:28:28 +0000 (22:28 +0800)]
Fix my email address

Fix my email address, last time it's wrong.

Signed-off-by: Dongxue Zhang <elta.era@gmail.com>
Signed-off-by: Aurelien Jarno <aurelien@aurel32.net>
11 years agotarget-mips: Remove semicolon from macro definition
Stefan Weil [Sat, 22 Dec 2012 12:38:19 +0000 (13:38 +0100)]
target-mips: Remove semicolon from macro definition

Macro RESTORE_FLUSH_MODE is similar to RESTORE_ROUNDING_MODE
but included a semicolon.

The code which uses that macro also includes a semicolon,
so the result was an empty statement.

Remove the superfluous semicolon from the macro definition.

Signed-off-by: Stefan Weil <sw@weilnetz.de>
Signed-off-by: Aurelien Jarno <aurelien@aurel32.net>
11 years agotarget-mips: Fix for helpers for EXTR_* instructions
Petar Jovanovic [Mon, 10 Dec 2012 15:28:17 +0000 (16:28 +0100)]
target-mips: Fix for helpers for EXTR_* instructions

The change removes some unnecessary and incorrect code for EXTR_S.H.
Further, it corrects the mask for shift value in the EXTR_ instructions. It also
extends the existing tests so they trigger the issues corrected with the change.

Signed-off-by: Petar Jovanovic <petarj@mips.com>
Signed-off-by: Aurelien Jarno <aurelien@aurel32.net>
11 years agotarget-mips: Fix incorrect reads and writes to DSPControl register
Petar Jovanovic [Thu, 6 Dec 2012 19:30:35 +0000 (20:30 +0100)]
target-mips: Fix incorrect reads and writes to DSPControl register

Upper 4 bits of ccond (bits 31..28 ) of DSPControl register are not used in
the MIPS32 architecture. They are used in the MIPS64 architecture. For MIPS32
these bits must be written as zero, and return zero on read.

The change fixes writes (WRDSP) and reads (RDDSP) to the register. It also fixes
the tests that use these instructions, and makes them smaller and simpler.

Signed-off-by: Petar Jovanovic <petarj@mips.com>
Signed-off-by: Aurelien Jarno <aurelien@aurel32.net>
11 years agoFix semaphores fallback code
Brad Smith [Fri, 28 Dec 2012 06:00:26 +0000 (01:00 -0500)]
Fix semaphores fallback code

As reported in bug 1087114 the semaphores fallback code is broken which
results in QEMU crashing and making QEMU unusable.

This patch is from Paolo.

This needs to be back ported to the 1.3 stable tree as well.

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Brad Smith <brad@comstyle.com>
Signed-off-by: Blue Swirl <blauwirbel@gmail.com>
11 years agotcg-hppa: Fix typo in brcond2
Richard Henderson [Fri, 28 Dec 2012 22:19:35 +0000 (14:19 -0800)]
tcg-hppa: Fix typo in brcond2

Reported-by: Stuart Brady <sdb@zubnet.me.uk>
Signed-off-by: Richard Henderson <rth@twiddle.net>
Signed-off-by: Blue Swirl <blauwirbel@gmail.com>
11 years agotcg-i386: Perform cmov detection at runtime for 32-bit.
Richard Henderson [Fri, 28 Dec 2012 22:17:02 +0000 (14:17 -0800)]
tcg-i386: Perform cmov detection at runtime for 32-bit.

Existing compile-time detection is spotty at best.  Convert
it all to runtime detection instead.

Signed-off-by: Richard Henderson <rth@twiddle.net>
Signed-off-by: Blue Swirl <blauwirbel@gmail.com>
11 years agotcg: Add TCGV_IS_UNUSED_*
Richard Henderson [Fri, 7 Dec 2012 21:07:17 +0000 (15:07 -0600)]
tcg: Add TCGV_IS_UNUSED_*

Cc: Aurelien Jarno <aurelien@aurel32.net>
Signed-off-by: Richard Henderson <rth@twiddle.net>
Reviewed-by: Andreas Färber <afaerber@suse.de>
Signed-off-by: Blue Swirl <blauwirbel@gmail.com>
12 years agoDisable semaphores fallback code for OpenBSD
Brad Smith [Fri, 28 Dec 2012 06:38:11 +0000 (01:38 -0500)]
Disable semaphores fallback code for OpenBSD

Disable the semaphores fallback code for OpenBSD as modern OpenBSD
releases now have sem_timedwait().

Signed-off-by: Brad Smith <brad@comstyle.com>
Signed-off-by: Blue Swirl <blauwirbel@gmail.com>
12 years agousb/redirect.c: unbreak compilation due to include/char/char.h
Alon Levy [Sun, 23 Dec 2012 21:25:09 +0000 (23:25 +0200)]
usb/redirect.c: unbreak compilation due to include/char/char.h

Broken since:

commit 927d4878b0ff319ed87fed9363f314613b0a5ed9
Author: Paolo Bonzini <pbonzini@redhat.com>
Date:   Mon Dec 17 18:20:05 2012 +0100

    softmmu: move remaining include files to include/ subdirectories

Signed-off-by: Alon Levy <alevy@redhat.com>
Signed-off-by: Blue Swirl <blauwirbel@gmail.com>
12 years agoMerge branch 'qom-cpu' of git://repo.or.cz/qemu/afaerber
Blue Swirl [Fri, 28 Dec 2012 16:08:23 +0000 (16:08 +0000)]
Merge branch 'qom-cpu' of git://repo.or.cz/qemu/afaerber

* 'qom-cpu' of git://repo.or.cz/qemu/afaerber:
  MAINTAINERS: Include X86CPU in CPU maintenance area
  cpu: Move kvm_run into CPUState
  cpu: Move kvm_state field into CPUState
  ppc_booke: Pass PowerPCCPU to ppc_booke_timers_init()
  ppc4xx_devs: Return PowerPCCPU from ppc4xx_init()
  ppc_booke: Pass PowerPCCPU to {decr,fit,wdt} timer callbacks
  ppc: Pass PowerPCCPU to [h]decr timer callbacks
  ppc: Pass PowerPCCPU to [h]decr callbacks
  ppc: Pass PowerPCCPU to ppc_set_irq()
  kvm: Pass CPUState to kvm_vcpu_ioctl()
  kvm: Pass CPUState to kvm_arch_*
  cpu: Move kvm_fd into CPUState
  qdev-properties.c: Separate core from the code used only by qemu-system-*
  qdev: Coding style fixes
  cpu: Introduce CPUListState struct
  target-alpha: Add support for -cpu ?
  target-alpha: Turn CPU definitions into subclasses
  target-alpha: Avoid leaking the alarm timer over reset
  alpha: Pass AlphaCPU array to Typhoon
  target-alpha: Let cpu_alpha_init() return AlphaCPU

12 years agovirtio-pci: don't poll masked vectors
Michael S. Tsirkin [Tue, 18 Dec 2012 12:02:46 +0000 (14:02 +0200)]
virtio-pci: don't poll masked vectors

At the moment, when irqfd is in use but a vector is masked,
qemu will poll it and handle vector masks in userspace.
Since almost no one ever looks at the pending bits,
it is better to defer this until pending bits
are actually read.
Implement this optimization using the new poll notifier.

Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
12 years agomsix: expose access to masked/pending state
Michael S. Tsirkin [Tue, 18 Dec 2012 11:54:32 +0000 (13:54 +0200)]
msix: expose access to masked/pending state

For use by poll handler.

Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
12 years agomsi: add API to get notified about pending bit poll
Michael S. Tsirkin [Wed, 12 Dec 2012 14:10:02 +0000 (16:10 +0200)]
msi: add API to get notified about pending bit poll

Update all users.

Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
12 years agopcie: Fix bug in pcie_ext_cap_set_next
Knut Omang [Tue, 18 Dec 2012 21:36:29 +0000 (22:36 +0100)]
pcie: Fix bug in pcie_ext_cap_set_next

Upper 16 bits of the PCIe Extended Capability Header was truncated during update,
also breaking pcie_add_capability.

Signed-off-by: Knut Omang <knut.omang@oracle.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
12 years agoqemu-kvm/pci-assign: 64 bits bar emulation
Xudong Hao [Thu, 20 Dec 2012 03:07:23 +0000 (11:07 +0800)]
qemu-kvm/pci-assign: 64 bits bar emulation

Enable 64 bits bar emulation.

Test pass with the current seabios which already support 64bit pci bars.

Signed-off-by: Xudong Hao <xudong.hao@intel.com>
Reviewed-by: Alex Williamson <alex.williamson@redhat.com>
Signed-off-by: Gleb Natapov <gleb@redhat.com>
12 years agoMAINTAINERS: Take over kvm maintenance
Gleb Natapov [Wed, 19 Dec 2012 15:46:15 +0000 (17:46 +0200)]
MAINTAINERS: Take over kvm maintenance

Replace Avi with myself as kvm maintainer.

Signed-off-by: Gleb Natapov <gleb@redhat.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
12 years agoconfigure: allow disabling pixman if not needed
Robert Schiele [Tue, 4 Dec 2012 15:58:08 +0000 (16:58 +0100)]
configure: allow disabling pixman if not needed

When we build neither any system emulation targets nor the tools there
is actually no need for pixman library.  In that case do not enforce
presence of that library on the system.

Reviewed-by: Andreas F=E4rber <afaerber@suse.de>
Signed-off-by: Robert Schiele <rschiele@gmail.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
12 years agotmp105: Create API for TMP105 temperature sensor.
Alex Horn [Wed, 5 Dec 2012 12:34:06 +0000 (12:34 +0000)]
tmp105: Create API for TMP105 temperature sensor.

* Define enum for TMP105 registers
* Move tmp105_set() from I2C to TMP105 header
* Document units and range of temperature as preconditions

Reviewed-by: Andreas Färber <afaerber@suse.de>
Signed-off-by: Alex Horn <alex.horn@cs.ox.ac.uk>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
12 years agobuild: Use separate makefile for "trace/"
Lluís Vilanova [Fri, 14 Dec 2012 19:13:09 +0000 (20:13 +0100)]
build: Use separate makefile for "trace/"

Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Lluís Vilanova <vilanova@ac.upc.edu>
--
Changes in v2:

* Do not depend on "qemu-timer-common.o".
* Use "$(obj)" in rules to refer to the build sub-directory.
* Remove dependencies against "$(GENERATED_HEADERS)".

Cc: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
12 years agowin32: Switch thread abstraction to us TLS variable internally
Jan Kiszka [Fri, 23 Nov 2012 11:12:01 +0000 (12:12 +0100)]
win32: Switch thread abstraction to us TLS variable internally

We already depend on working __thread support for coroutines, so this
complication here is no longer needed.

Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
12 years agotarget-i386: CPUID: return highest basic leaf if eax > cpuid_xlevel
Eduardo Habkost [Thu, 20 Dec 2012 18:43:48 +0000 (16:43 -0200)]
target-i386: CPUID: return highest basic leaf if eax > cpuid_xlevel

This fixes a subtle bug. A bug that probably won't cause trouble for any
existing OS, but a bug anyway:

Intel SDM Volume 2, CPUID Instruction states:

> Two types of information are returned: basic and extended function
> information. If a value entered for CPUID.EAX is higher than the maximum
> input value for basic or extended function for that processor then the
> data for the highest basic information leaf is returned. For example,
> using the Intel Core i7 processor, the following is true:
>
>   CPUID.EAX = 05H (* Returns MONITOR/MWAIT leaf. *)
>   CPUID.EAX = 0AH (* Returns Architectural Performance Monitoring leaf. *)
>   CPUID.EAX = 0BH (* Returns Extended Topology Enumeration leaf. *)
>   CPUID.EAX = 0CH (* INVALID: Returns the same information as CPUID.EAX = 0BH. *)
>   CPUID.EAX = 80000008H (* Returns linear/physical address size data. *)
>   CPUID.EAX = 8000000AH (* INVALID: Returns same information as CPUID.EAX = 0BH. *)

AMD's CPUID Specification, on the other hand, is less specific:

> The CPUID instruction supports two sets or ranges of functions,
> standard and extended.
>
> • The smallest function number of the standard function range is
>   Fn0000_0000. The largest function num- ber of the standard function
>   range, for a particular implementation, is returned in CPUID
>   Fn0000_0000_EAX.
>
> • The smallest function number of the extended function range is
>   Fn8000_0000. The largest function num- ber of the extended function
>   range, for a particular implementation, is returned in CPUID
>   Fn8000_0000_EAX.
>
> Functions that are neither standard nor extended are undefined and
> should not be relied upon.

QEMU's behavior matched Intel's specification before, but this was
changed by commit b3baa152aaef1905876670590275c2dd0bbb088c. This patch
restores the behavior documented by Intel when cpuid_xlevel2 is 0.

The existing behavior when cpuid_xlevel2 is set (falling back to
level=cpuid_xlevel) is being kept, as I couldn't find any public
documentation on the CPUID 0xC0000000 function range on Centaur CPUs.

Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
12 years agoqemu-char: Inherit ptys and improve output from -serial pty
Lei Li [Fri, 21 Dec 2012 04:26:38 +0000 (12:26 +0800)]
qemu-char: Inherit ptys and improve output from -serial pty

Changes since V1:
  - Avoid crashing since qemu_opts_id() may return null on some
    systems according to Markus's suggestion.

When controlling a qemu instance from another program, it's
hard to know which serial port or monitor device is redirected
to which pty. With more than one device using "pty" a lot of
guesswork is involved.

$ ./x86_64-softmmu/qemu-system-x86_64 -serial pty -serial pty -monitor pty
char device redirected to /dev/pts/5
char device redirected to /dev/pts/6
char device redirected to /dev/pts/7

Although we can find out what everything else is connected to
by the "info chardev" with "-monitor stdio" in the command line,
It'd be very useful to be able to have qemu inherit pseudo-tty
file descriptors so they could just be specified on the command
line like:

$ ./x86_64-softmmu/qemu-system-x86_64 -serial pty -serial pty -monitor pty
char device compat_monitor0 redirected to /dev/pts/5
char device serial0 redirected to /dev/pts/6
char device serial1 redirected to /dev/pts/7

link: https://bugs.launchpad.net/qemu/+bug/938552
Signed-off-by: Lei Li <lilei@linux.vnet.ibm.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
12 years agoMerge branch 'master' of git://git.qemu.org/qemu into qom-cpu
Andreas Färber [Sat, 22 Dec 2012 23:39:34 +0000 (00:39 +0100)]
Merge branch 'master' of git://git.qemu.org/qemu into qom-cpu

Adapt header include paths.

Signed-off-by: Andreas Färber <afaerber@suse.de>
12 years agotarget-xtensa: fix search_pc for the last TB opcode
Max Filippov [Wed, 19 Dec 2012 20:04:09 +0000 (00:04 +0400)]
target-xtensa: fix search_pc for the last TB opcode

Zero out tcg_ctx.gen_opc_instr_start for instructions representing the
last guest opcode in the TB.

Cc: qemu-stable@nongnu.org
Signed-off-by: Max Filippov <jcmvbkbc@gmail.com>
Signed-off-by: Blue Swirl <blauwirbel@gmail.com>
12 years agonet: add missing include file
Liming Wang [Fri, 21 Dec 2012 08:56:58 +0000 (16:56 +0800)]
net: add missing include file

To fix building error:

CC    net/vde.o
net/vde.c: In function ‘vde_cleanup’:
net/vde.c:65:5: error: implicit declaration of function ‘qemu_set_fd_handler’ [-Werror=implicit-function-declaration]
net/vde.c:65:5: error: nested extern declaration of ‘qemu_set_fd_handler’ [-Werror=nested-externs]
cc1: all warnings being treated as errors

Signed-off-by: Liming Wang <walimisdev@gmail.com>
Signed-off-by: Blue Swirl <blauwirbel@gmail.com>
12 years agotranslate-all.c: Use tb1->phys_hash_next directly in tb_remove
陳韋任 (Wei-Ren Chen) [Thu, 20 Dec 2012 01:39:16 +0000 (09:39 +0800)]
translate-all.c: Use tb1->phys_hash_next directly in tb_remove

  When tb_remove was first commited at fd6ce8f6, there were three different
calls pass different names to offsetof. In current codebase, the other two
calls are replaced with tb_page_remove. There is no need to have a general
tb_remove. Omit passing the third parameter and using tb1->phys_hash_next
directly.

Signed-off-by: Chen Wei-Ren <chenwj@iis.sinica.edu.tw>
Signed-off-by: Blue Swirl <blauwirbel@gmail.com>
12 years agonet: Add missing include statement (fix compiler warnings for MinGW)
Stefan Weil [Thu, 20 Dec 2012 22:09:53 +0000 (23:09 +0100)]
net: Add missing include statement (fix compiler warnings for MinGW)

These and some more compiler warnings were caused by a recent commit:

net/tap-win32.c:724: warning: no previous prototype for ‘tap_has_ufo’
net/tap-win32.c:729: warning: no previous prototype for ‘tap_has_vnet_hdr’
...

Signed-off-by: Stefan Weil <sw@weilnetz.de>
Signed-off-by: Blue Swirl <blauwirbel@gmail.com>
12 years agoxen: add missing include
Paolo Bonzini [Thu, 20 Dec 2012 11:29:20 +0000 (12:29 +0100)]
xen: add missing include

xen-all needs to access CharDeviceState's filename field, so
it needs to include char/char.h.

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Tested-by: Andreas Färber <afaerber@suse.de>
Signed-off-by: Blue Swirl <blauwirbel@gmail.com>
12 years agobuild: fix includes for VNC
Paolo Bonzini [Thu, 20 Dec 2012 11:29:19 +0000 (12:29 +0100)]
build: fix includes for VNC

vnc-tls.h is included by vnc.h, and it includes gnutls/gnutls.h.
Hence, GnuTLS header files are needed by all files that include
vnc.h, most notably qmp.c.  Move these flags to QEMU_CFLAGS for
simplicity.

Reported-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Blue Swirl <blauwirbel@gmail.com>
12 years agoMerge remote-tracking branch 'quintela/thread-20121220.next' into staging
Anthony Liguori [Fri, 21 Dec 2012 13:53:48 +0000 (07:53 -0600)]
Merge remote-tracking branch 'quintela/thread-20121220.next' into staging

* quintela/thread-20121220.next: (79 commits)
  migration: merge QEMUFileBuffered into MigrationState
  migration: fix qemu_get_fd for BufferedFile
  ram: refactor ram_save_block() return value
  ram: account the amount of transferred ram better
  ram: optimize migration bitmap walking
  ram: Use memory_region_test_and_clear_dirty
  memory: introduce memory_region_test_and_clear_dirty
  ram: Add last_sent_block
  ram: rename last_block to last_seen_block
  migration: move migration notifier
  migration: Inline qemu_fopen_ops_buffered into migrate_fd_connect
  migration: move migration_fd_put_ready()
  migration: add XFER_LIMIT_RATIO
  migration: move buffered_file.c code into migration.c
  savevm: New save live migration method: pending
  buffered_file: unfold buffered_append in buffered_put_buffer
  buffered_file: don't flush on put buffer
  buffered_file: Unfold the trick to restart generating migration data
  migration: just lock migrate_fd_put_ready
  migration: remove unfreeze logic
  ...

Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
12 years agomigration: merge QEMUFileBuffered into MigrationState
Juan Quintela [Wed, 19 Dec 2012 08:55:50 +0000 (09:55 +0100)]
migration: merge QEMUFileBuffered into MigrationState

Avoid splitting the state of outgoing migration, more or less arbitrarily,
between two data structures.  QEMUFileBuffered anyway is used only during
migration.

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>
12 years agomigration: fix qemu_get_fd for BufferedFile
Paolo Bonzini [Wed, 7 Nov 2012 23:42:50 +0000 (00:42 +0100)]
migration: fix qemu_get_fd for BufferedFile

Not really used, but nice to have it correct. :)

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>
12 years agoram: refactor ram_save_block() return value
Juan Quintela [Mon, 10 Dec 2012 12:27:50 +0000 (13:27 +0100)]
ram: refactor ram_save_block() return value

It could only return 0 if we only found dirty xbzrle pages that hadn't
changed (i.e. they were written with the same content).  We don't care
about that case, it is the same than nothing dirty.

So now the return of the function is how much have it written, nothing
else. Adjust callers.

And we also made ram_save_iterate() return the number of transferred
bytes, not the number of transferred pages.

Signed-off-by: Juan Quintela <quintela@redhat.com>
12 years agoram: account the amount of transferred ram better
Juan Quintela [Thu, 18 Oct 2012 11:56:35 +0000 (13:56 +0200)]
ram: account the amount of transferred ram better

Signed-off-by: Juan Quintela <quintela@redhat.com>
12 years agoram: optimize migration bitmap walking
Juan Quintela [Wed, 17 Oct 2012 22:00:59 +0000 (00:00 +0200)]
ram: optimize migration bitmap walking

Instead of testing each page individually, we search what is the next
dirty page with a bitmap operation.  We have to reorganize the code to
move from a "for" loop, to a while(dirty) loop.

Signed-off-by: Juan Quintela <quintela@redhat.com>
12 years agoram: Use memory_region_test_and_clear_dirty
Juan Quintela [Wed, 17 Oct 2012 18:27:15 +0000 (20:27 +0200)]
ram: Use memory_region_test_and_clear_dirty

This avoids having to do two walks over the dirty bitmap, once reading
the dirty bits, and anthoer cleaning them.

Signed-off-by: Juan Quintela <quintela@redhat.com>
12 years agomemory: introduce memory_region_test_and_clear_dirty
Juan Quintela [Wed, 17 Oct 2012 18:24:28 +0000 (20:24 +0200)]
memory: introduce memory_region_test_and_clear_dirty

This function avoids having to do two calls, one to test the dirty bit, and
other to reset it.

Signed-off-by: Juan Quintela <quintela@redhat.com>
12 years agoram: Add last_sent_block
Juan Quintela [Wed, 17 Oct 2012 18:10:55 +0000 (20:10 +0200)]
ram: Add last_sent_block

This is the last block from where we have sent data.

Signed-off-by: Orit Wasserman <owasserm@redhat.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>
12 years agoram: rename last_block to last_seen_block
Juan Quintela [Wed, 17 Oct 2012 18:08:04 +0000 (20:08 +0200)]
ram: rename last_block to last_seen_block

Signed-off-by: Juan Quintela <quintela@redhat.com>
12 years agomigration: move migration notifier
Juan Quintela [Wed, 3 Oct 2012 18:04:41 +0000 (20:04 +0200)]
migration: move migration notifier

At this point, it is waranteed that state is ACTIVE.  Old position
didn't assured hat.

Signed-off-by: Juan Quintela <quintela@redhat.com>
12 years agomigration: Inline qemu_fopen_ops_buffered into migrate_fd_connect
Juan Quintela [Wed, 3 Oct 2012 16:23:03 +0000 (18:23 +0200)]
migration: Inline qemu_fopen_ops_buffered into migrate_fd_connect

Signed-off-by: Juan Quintela <quintela@redhat.com>
12 years agomigration: move migration_fd_put_ready()
Juan Quintela [Wed, 3 Oct 2012 15:23:59 +0000 (17:23 +0200)]
migration: move migration_fd_put_ready()

Put it near its use and un-export it.

Signed-off-by: Juan Quintela <quintela@redhat.com>
12 years agomigration: add XFER_LIMIT_RATIO
Juan Quintela [Wed, 19 Dec 2012 09:40:48 +0000 (10:40 +0100)]
migration: add XFER_LIMIT_RATIO

The "magic" divisions by 10 are there because of the value of BUFFER_DELAY.
Introduce a constant to explain them better.

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>
12 years agomigration: move buffered_file.c code into migration.c
Juan Quintela [Wed, 3 Oct 2012 12:18:33 +0000 (14:18 +0200)]
migration: move buffered_file.c code into migration.c

This only moves the code (also from buffered_file.h to migration.h).
Fix whitespace until checkpatch is happy.

Signed-off-by: Juan Quintela <quintela@redhat.com>
12 years agosavevm: New save live migration method: pending
Juan Quintela [Fri, 21 Sep 2012 09:18:18 +0000 (11:18 +0200)]
savevm: New save live migration method: pending

Code just now does (simplified for clarity)

    if (qemu_savevm_state_iterate(s->file) == 1) {
       vm_stop_force_state(RUN_STATE_FINISH_MIGRATE);
       qemu_savevm_state_complete(s->file);
    }

Problem here is that qemu_savevm_state_iterate() returns 1 when it
knows that remaining memory to sent takes less than max downtime.

But this means that we could end spending 2x max_downtime, one
downtime in qemu_savevm_iterate, and the other in
qemu_savevm_state_complete.

Changed code to:

    pending_size = qemu_savevm_state_pending(s->file, max_size);
    DPRINTF("pending size %lu max %lu\n", pending_size, max_size);
    if (pending_size >= max_size) {
        ret = qemu_savevm_state_iterate(s->file);
     } else {
        vm_stop_force_state(RUN_STATE_FINISH_MIGRATE);
        qemu_savevm_state_complete(s->file);
     }

So what we do is: at current network speed, we calculate the maximum
number of bytes we can sent: max_size.

Then we ask every save_live section how much they have pending.  If
they are less than max_size, we move to complete phase, otherwise we
do an iterate one.

This makes things much simpler, because now individual sections don't
have to caluclate the bandwidth (it was implossible to do right from
there).

Signed-off-by: Juan Quintela <quintela@redhat.com>
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
12 years agobuffered_file: unfold buffered_append in buffered_put_buffer
Juan Quintela [Tue, 24 Jul 2012 12:24:08 +0000 (14:24 +0200)]
buffered_file: unfold buffered_append in buffered_put_buffer

It was the only user, and now buffered_put_buffer just do the append

Signed-off-by: Juan Quintela <quintela@redhat.com>
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
12 years agobuffered_file: don't flush on put buffer
Juan Quintela [Tue, 24 Jul 2012 12:00:13 +0000 (14:00 +0200)]
buffered_file: don't flush on put buffer

We call buffered_put_buffer with iothread held, and buffered_flush() does
synchronous writes.  We only want to do the synchronous writes outside.

Signed-off-by: Juan Quintela <quintela@redhat.com>
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
12 years agobuffered_file: Unfold the trick to restart generating migration data
Juan Quintela [Tue, 24 Jul 2012 11:22:18 +0000 (13:22 +0200)]
buffered_file: Unfold the trick to restart generating migration data

This was needed before due to the way that the callbacks worked.

Signed-off-by: Juan Quintela <quintela@redhat.com>
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
12 years agomigration: just lock migrate_fd_put_ready
Juan Quintela [Mon, 23 Jul 2012 04:31:30 +0000 (06:31 +0200)]
migration: just lock migrate_fd_put_ready

Signed-off-by: Juan Quintela <quintela@redhat.com>
12 years agomigration: remove unfreeze logic
Juan Quintela [Mon, 23 Jul 2012 04:24:03 +0000 (06:24 +0200)]
migration: remove unfreeze logic

Now that we have a thread, and blocking writes, we don't need it.

Signed-off-by: Juan Quintela <quintela@redhat.com>
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
12 years agomigration: make writes blocking
Juan Quintela [Mon, 23 Jul 2012 04:15:02 +0000 (06:15 +0200)]
migration: make writes blocking

Move all the writes to the migration_thread, and make writings
blocking.  Notice that are still using the iothread for everything
that we do.

Signed-off-by: Juan Quintela <quintela@redhat.com>
12 years agomigration: move migration thread init code to migrate_fd_put_ready
Juan Quintela [Mon, 23 Jul 2012 03:45:29 +0000 (05:45 +0200)]
migration: move migration thread init code to migrate_fd_put_ready

This way everything related with migration is run on the migration
thread and no locking is needed.

Signed-off-by: Juan Quintela <quintela@redhat.com>
12 years agomigration: make qemu_fopen_ops_buffered() return void
Juan Quintela [Mon, 23 Jul 2012 00:13:23 +0000 (02:13 +0200)]
migration: make qemu_fopen_ops_buffered() return void

We want the file assignment to happen before the thread is created to
avoid locking, so we just do it before creating the thread.

Signed-off-by: Juan Quintela <quintela@redhat.com>
Reviewed-by: Orit Wasserman <owasserm@redhat.com>
12 years agobuffered_file: Move from using a timer to use a thread
Juan Quintela [Sun, 22 Jul 2012 23:56:50 +0000 (01:56 +0200)]
buffered_file: Move from using a timer to use a thread

We still protect everything except the wait with the iothread lock.
But we moved from a timer to a thread.  Steps one by one.

We also need to detect when we have finished with a variable "complete".

Signed-off-by: Juan Quintela <quintela@redhat.com>
12 years agoprotect the ramlist with a separate mutex
Umesh Deshpande [Wed, 17 Aug 2011 07:01:33 +0000 (00:01 -0700)]
protect the ramlist with a separate mutex

Add the new mutex that protects shared state between ram_save_live
and the iothread.  If the iothread mutex has to be taken together
with the ramlist mutex, the iothread shall always be _outside_.

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Umesh Deshpande <udeshpan@redhat.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>
Reviewed-by: Orit Wasserman <owasserm@redhat.com>
12 years agoadd a version number to ram_list
Umesh Deshpande [Thu, 18 Aug 2011 18:41:17 +0000 (11:41 -0700)]
add a version number to ram_list

This will be used to detect if last_block might have become invalid
across different calls to ram_save_live.

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Umesh Deshpande <udeshpan@redhat.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>
Reviewed-by: Orit Wasserman <owasserm@redhat.com>
12 years agoexec: sort the memory from biggest to smallest
Paolo Bonzini [Wed, 14 Nov 2012 15:00:51 +0000 (16:00 +0100)]
exec: sort the memory from biggest to smallest

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>