Merge remote-tracking branch 'remotes/afaerber/tags/qom-cpu-for-peter' into staging

QOM CPUState and X86CPU

* Include exception state in CPU VMState
* Fix -cpu *,migratable=foo
* Error out on unknown -cpu *,+foo,-bar

# gpg: Signature made Fri 05 Sep 2014 15:38:14 BST using RSA key ID 3E7E013F
# gpg: Good signature from "Andreas Färber <[email protected]>"
# gpg:                 aka "Andreas Färber <[email protected]>"

* remotes/afaerber/tags/qom-cpu-for-peter:
  target-i386: Reject invalid CPU feature names on the command-line
  target-i386: Support migratable=no properly
  exec: Save CPUState::exception_index field

Signed-off-by: Peter Maydell <[email protected]>

target-i386: Reject invalid CPU feature names on the command-line

Instead of simply printing a warning, report an error when invalid CPU
options are provided on the CPU model string.

Signed-off-by: Eduardo Habkost <[email protected]>
Reviewed-by: Igor Mammedov <[email protected]>
Signed-off-by: Andreas Färber <[email protected]>

target-i386: Support migratable=no properly

When the "migratable" property was implemented, the behavior was tested
by changing the default on the code, but actually using the option on
the command-line (e.g. "-cpu host,migratable=false") doesn't work as
expected. This is a regression for a common use case of "-cpu host",
which is to enable features that are supported by the host CPU + kernel
before feature-specific code is added to QEMU.

Fix this by initializing the feature words for "-cpu host" on
x86_cpu_parse_featurestr(), right after parsing the CPU options.

Signed-off-by: Eduardo Habkost <[email protected]>
Reviewed-by: Michael Roth <[email protected]>
Cc: [email protected]
Signed-off-by: Andreas Färber <[email protected]>

exec: Save CPUState::exception_index field

This patch adds a subsection with exception_index field to the VMState for
correct saving the CPU state.
Without this patch, simulator could miss the pending exception in the saved
virtual machine state.

Signed-off-by: Pavel Dovgalyuk <[email protected]>
Cc: [email protected]
Signed-off-by: Andreas Färber <[email protected]>

console: Remove unused QEMU_BIG_ENDIAN_FLAG

If we need to, we should use the pixman formats instead but for
now this is unused except in commented out code so take it out
to avoid further confusion about surface endianness.

Signed-off-by: Benjamin Herrenschmidt <[email protected]>
Signed-off-by: Gerd Hoffmann <[email protected]>

Merge remote-tracking branch 'remotes/kraxel/tags/pull-chardev-20140905-1' into staging

pty: Fix byte loss bug when connecting to pty

# gpg: Signature made Fri 05 Sep 2014 12:57:32 BST using RSA key ID D3E87138
# gpg: Good signature from "Gerd Hoffmann (work) <[email protected]>"
# gpg:                 aka "Gerd Hoffmann <[email protected]>"
# gpg:                 aka "Gerd Hoffmann (private) <[email protected]>"

* remotes/kraxel/tags/pull-chardev-20140905-1:
  pty: Fix byte loss bug when connecting to pty

Signed-off-by: Peter Maydell <[email protected]>

console: add qemu_pixman_linebuf_copy

Helper function for copying data from linebuf to framebuffer using
pixman, possibly converting in case src and dst formats differ.

Signed-off-by: Gerd Hoffmann <[email protected]>

console: add dpy_gfx_update_dirty

Calls dpy_gfx_update for all dirty scanlines. Works for
DisplaySurfaces backed by guest memory (i.e. the ones created
using qemu_create_displaysurface_guestmem).

Signed-off-by: Gerd Hoffmann <[email protected]>

console: add qemu_create_displaysurface_guestmem

This patch adds a qemu_create_displaysurface_guestmem helper function.
Works simliar to qemu_create_displaysurface_from, but accepts a
guest address instead of a host pointer and it handles
cpu_physical_memory_{map,unmap} for you.

Signed-off-by: Gerd Hoffmann <[email protected]>

console: stop using PixelFormat

With this patch the qemu console core stops using PixelFormat and pixman
format codes side-by-side, pixman format code is the primary way to
specify the DisplaySurface format:

 * DisplaySurface stops carrying a PixelFormat field.
 * qemu_create_displaysurface_from() expects a pixman format now.

Functions to convert PixelFormat to pixman_format_code_t (and back)
exist for those who still use PixelFormat.   As PixelFormat allows
easy access to masks and shifts it will probably continue to exist.

[ xenfb added by Benjamin Herrenschmidt ]

Signed-off-by: Gerd Hoffmann <[email protected]>

console: reimplement qemu_default_pixelformat

Use the new qemu_pixelformat_from_pixman and qemu_default_pixman_format
functions to reimplement qemu_default_pixelformat
(qemu_different_endianness_pixelformat too).

Signed-off-by: Gerd Hoffmann <[email protected]>

console: add qemu_default_pixman_format

Function returning the default pixman format for a given depth.

Signed-off-by: Gerd Hoffmann <[email protected]>

console: add qemu_pixelformat_from_pixman

Function to convert pixman format codes to qemu PixelFormat.

[ Benjamin Herrenschmidt: fix BGRA+RGBA shifts ]

Signed-off-by: Gerd Hoffmann <[email protected]>

pty: Fix byte loss bug when connecting to pty

When trying to print data to the pty, we first check if it is connected.
If not, we try to reconnect, but we drop the pending data even if we
have successfully reconnected; this makes us lose the first byte of the very
first transmission.
This small fix addresses the issue by checking once more if the pty is connected
after having tried to reconnect.

Signed-off-by: Sebastian Tanase <[email protected]>
Signed-off-by: Gerd Hoffmann <[email protected]>

Merge remote-tracking branch 'remotes/kraxel/tags/pull-cve-2014-3615-20140905-1' into staging

CVE-2014-3615: fix sanity checks in vbe (bochs dispi) and spice.

# gpg: Signature made Fri 05 Sep 2014 12:18:04 BST using RSA key ID D3E87138
# gpg: Good signature from "Gerd Hoffmann (work) <[email protected]>"
# gpg:                 aka "Gerd Hoffmann <[email protected]>"
# gpg:                 aka "Gerd Hoffmann (private) <[email protected]>"

* remotes/kraxel/tags/pull-cve-2014-3615-20140905-1:
  spice: make sure we don't overflow ssd->buf
  vbe: rework sanity checks
  vbe: make bochs dispi interface return the correct memory size with qxl

Signed-off-by: Peter Maydell <[email protected]>

spice: make sure we don't overflow ssd->buf

Related spice-only bug.  We have a fixed 16 MB buffer here, being
presented to the spice-server as qxl video memory in case spice is
used with a non-qxl card.  It's also used with qxl in vga mode.

When using display resolutions requiring more than 16 MB of memory we
are going to overflow that buffer.  In theory the guest can write,
indirectly via spice-server.  The spice-server clears the memory after
setting a new video mode though, triggering a segfault in the overflow
case, so qemu crashes before the guest has a chance to do something
evil.

Fix that by switching to dynamic allocation for the buffer.

CVE-2014-3615

Cc: [email protected]
Cc: [email protected]
Signed-off-by: Gerd Hoffmann <[email protected]>
Reviewed-by: Laszlo Ersek <[email protected]>

Merge remote-tracking branch 'remotes/afaerber/tags/qom-devices-for-peter' into staging

QOM infrastructure fixes and device conversions

* Cleanups for recursive device unrealization

# gpg: Signature made Thu 04 Sep 2014 18:17:35 BST using RSA key ID 3E7E013F
# gpg: Good signature from "Andreas Färber <[email protected]>"
# gpg:                 aka "Andreas Färber <[email protected]>"

* remotes/afaerber/tags/qom-devices-for-peter:
  qdev: Add cleanup logic in device_set_realized() to avoid resource leak
  qdev: Use NULL instead of local_err for qbus_child unrealize
  qdev: Use error_abort instead of using local_err
  memory: Remove object_property_add_child_array()
  qom: Add automatic arrayification to object_property_add()
  machine: Clean up -machine handling
  qom: Make object_child_foreach() safe for objects removal

Signed-off-by: Peter Maydell <[email protected]>

Merge remote-tracking branch 'remotes/kvaneesh/for-upstream' into staging

* remotes/kvaneesh/for-upstream:
  hw/9pfs: Don't return type from host in readdir on local 9p filesystem
  hw/9pfs: Use little-endian format for xattr values

Signed-off-by: Peter Maydell <[email protected]>

qdev: Add cleanup logic in device_set_realized() to avoid resource leak

At present, this function doesn't have partial cleanup implemented,
which will cause resource leaks in some scenarios.

Example:

1. Assume that "dc->realize(dev, &local_err)" executes successful
   and local_err == NULL;
2. device hotplug in hotplug_handler_plug() executes but fails
   (it is prone to occur). Then local_err != NULL;
3. error_propagate(errp, local_err) and return. But the resources
   which have been allocated in dc->realize() will be leaked.
Simple backtrace:
  dc->realize()
   |->device_realize
            |->pci_qdev_init()
                |->do_pci_register_device()
                |->etc.

Add fuller cleanup logic which assures that function can
goto appropriate error label as local_err population is
detected at each relevant point.

Signed-off-by: Gonglei <[email protected]>
Reviewed-by: Peter Crosthwaite <[email protected]>
Cc: [email protected]
Signed-off-by: Andreas Färber <[email protected]>

qdev: Use NULL instead of local_err for qbus_child unrealize

Forcefully unrealize all children regardless of errors in earlier
iterations (if any). We should keep going with cleanup operation
rather than report an error immediately. Therefore store the first
child unrealization failure and propagate it at the end. We also
forcefully unregister vmsd and unrealize actual object, too.

Signed-off-by: Gonglei <[email protected]>
Reviewed-by: Peter Crosthwaite <[email protected]>
Cc: [email protected]
Signed-off-by: Andreas Färber <[email protected]>

Merge remote-tracking branch 'remotes/stefanha/tags/net-pull-request' into staging

Net patches

# gpg: Signature made Thu 04 Sep 2014 17:32:44 BST using RSA key ID 81AB73C8
# gpg: Good signature from "Stefan Hajnoczi <[email protected]>"
# gpg:                 aka "Stefan Hajnoczi <[email protected]>"

* remotes/stefanha/tags/net-pull-request:
  virtio-net: purge outstanding packets when starting vhost
  net: complete all queued packets on VM stop
  net: invoke callback when purging queue
  virtio: don't call device on !vm_running
  virtio-net: don't run bh on vm stopped
  net: Forbid dealing with packets when VM is not running

Signed-off-by: Peter Maydell <[email protected]>

virtio-net: purge outstanding packets when starting vhost

whenever we start vhost, virtio could have outstanding packets
queued, when they complete later we'll modify the ring
while vhost is processing it.

To prevent this, purge outstanding packets on vhost start.

Cc: [email protected]
Cc: Jason Wang <[email protected]>
Signed-off-by: Michael S. Tsirkin <[email protected]>
Signed-off-by: Stefan Hajnoczi <[email protected]>

net: complete all queued packets on VM stop

This completes all packets, ensuring that callbacks
will not run when VM is stopped.

Cc: [email protected]
Cc: Jason Wang <[email protected]>
Signed-off-by: Michael S. Tsirkin <[email protected]>
Signed-off-by: Stefan Hajnoczi <[email protected]>

net: invoke callback when purging queue

devices rely on packet callbacks eventually running,
but we violate this rule whenever we purge the queue.
To fix, invoke callbacks on all packets on purge.
Set length to 0, this way callers can detect that
this happened and re-queue if necessary.

Cc: [email protected]
Cc: Jason Wang <[email protected]>
Signed-off-by: Michael S. Tsirkin <[email protected]>
Acked-by: Jason Wang <[email protected]>
Signed-off-by: Stefan Hajnoczi <[email protected]>

virtio: don't call device on !vm_running

On vm stop, virtio changes vm_running state
too soon, so callbacks can get envoked with
vm_running = false;

Cc: [email protected]
Cc: Jason Wang <[email protected]>
Signed-off-by: Michael S. Tsirkin <[email protected]>
Signed-off-by: Stefan Hajnoczi <[email protected]>

virtio-net: don't run bh on vm stopped

commit 783e7706937fe15523b609b545587a028a2bdd03
    virtio-net: stop/start bh when appropriate

is incomplete: BH might execute within the same main loop iteration but
after vmstop, so in theory, we might trigger an assertion.
I was unable to reproduce this in practice,
but it seems clear enough that the potential is there, so worth fixing.

Cc: [email protected]
Reported-by: Stefan Hajnoczi <[email protected]>
Signed-off-by: Michael S. Tsirkin <[email protected]>
Signed-off-by: Stefan Hajnoczi <[email protected]>

hw/9pfs: Don't return type from host in readdir on local 9p filesystem

When using mapped mode in 9pfs, readdir implementation
should not return file type in d_type from the host
readdir, instead, it should use the type stored in
the extended attributes.  Since d_type is optional
and reading ext attrs for every readdir is expensive,
it should be sufficient to just set d_type to DT_UNKNOWN,
so guest will know to look it up separately.

This is a -stable material.

Signed-off-by: Bastian Blank <[email protected]>
Signed-off-by: Michael Tokarev <[email protected]>
Signed-off-by: Aneesh Kumar K.V <[email protected]>

qdev: Use error_abort instead of using local_err

This error can not happen normally. If it happens, it indicates
something very wrong, we should abort QEMU. Moreover, the
user can only refer to /machine/peripheral or /objects, not
/machine/unattached.

While at it, remove superfluous check about local_err.

Signed-off-by: Gonglei <[email protected]>
Reviewed-by: Peter Crosthwaite <[email protected]>
Signed-off-by: Andreas Färber <[email protected]>

memory: Remove object_property_add_child_array()

Obsoleted by automatic object_property_add() arrayification.

Reviewed-by: Paolo Bonzini <[email protected]>
Signed-off-by: Peter Crosthwaite <[email protected]>
Signed-off-by: Andreas Färber <[email protected]>

qom: Add automatic arrayification to object_property_add()

If "[*]" is given as the last part of a QOM property name, treat that
as an array property. The added property is given the first available
name, replacing the * with a decimal number counting from 0.

First add with name "foo[*]" will be "foo[0]". Second "foo[1]" and so
on.

Callers may inspect the ObjectProperty * return value to see what
number the added property was given.

Signed-off-by: Peter Crosthwaite <[email protected]>
Signed-off-by: Andreas Färber <[email protected]>

machine: Clean up -machine handling

Since commit c4090f8, -object options are no longer handled through
object_set_property(), so clean up -object leftovers by renaming the
function and dropping special-casing of qom-type and id properties.

Cc: Paolo Bonzini <[email protected]>
Reviewed-by: Marcel Apfelbaum <[email protected]>
Signed-off-by: Andreas Färber <[email protected]>

qom: Make object_child_foreach() safe for objects removal

Current object_child_foreach() uses QTAILQ_FOREACH() to walk
through children and that makes children removal from the callback
impossible.

This makes object_child_foreach() use QTAILQ_FOREACH_SAFE().

Signed-off-by: Alexey Kardashevskiy <[email protected]>
Reviewed-by: Hu Tao <[email protected]>
Signed-off-by: Andreas Färber <[email protected]>

net: Forbid dealing with packets when VM is not running

For all NICs(except virtio-net) emulated by qemu,
Such as e1000, rtl8139, pcnet and ne2k_pci,
Qemu can still receive packets when VM is not running.

If this happened in *migration's* last PAUSE VM stage, but
before the end of the migration, the new receiving packets will possibly dirty
parts of RAM which has been cached in *iovec*(will be sent asynchronously) and
dirty parts of new RAM which will be missed.
This will lead serious network fault in VM.

To avoid this, we forbid receiving packets in generic net code when
VM is not running.

Bug reproduction steps:
(1) Start a VM which configured at least one NIC
(2) In VM, open several Terminal and do *Ping IP -i 0.1*
(3) Migrate the VM repeatedly between two Hosts
And the *PING* command in VM will very likely fail with message:
'Destination HOST Unreachable', the NIC in VM will stay unavailable unless you
run 'service network restart'

Signed-off-by: zhanghailiang <[email protected]>
Reviewed-by: Jason Wang <[email protected]>
Reviewed-by: Juan Quintela <[email protected]>
Reviewed-by: Michael S. Tsirkin <[email protected]>
Signed-off-by: Stefan Hajnoczi <[email protected]>

Merge remote-tracking branch 'remotes/mjt/tags/trivial-patches-2014-09-03' into staging

trivial patches for 2014-09-03

# gpg: Signature made Wed 03 Sep 2014 06:53:42 BST using RSA key ID A4C3D7DB
# gpg: Good signature from "Michael Tokarev <[email protected]>"
# gpg:                 aka "Michael Tokarev <[email protected]>"
# gpg:                 aka "Michael Tokarev <[email protected]>"
# gpg: WARNING: This key is not certified with a trusted signature!
# gpg:          There is no indication that the signature belongs to the owner.
# Primary key fingerprint: 6EE1 95D1 886E 8FFB 810D  4324 457C E0A0 8044 65C5
#      Subkey fingerprint: 6F67 E18E 7C91 C5B1 5514  66A7 BEE5 9D74 A4C3 D7DB

* remotes/mjt/tags/trivial-patches-2014-09-03:
  slirp: Honour vlan/stack in hostfwd_remove commands
  hmp: fix MemdevList memory leak
  qom/object.c, hmp.c: fix string_output_get_string() memory leak
  query-memdev: fix potential memory leaks
  MAINTAINERS: Add VMWare devices maintainer
  device_tree.c: dump all err mesages with error_report
  device_tree.c: redirect load_device_tree err message to stderr
  scripts: Remove scripts/qtest
  Fix debug print warning
  curl: The macro that you have to uncomment to get debugging is DEBUG_CURL.

Signed-off-by: Peter Maydell <[email protected]>

Merge remote-tracking branch 'remotes/mst/tags/for_upstream' into staging

pci, pc fixes, features

A bunch of bugfixes - these will make sense for 2.1.1

Initial Intel IOMMU support.

Signed-off-by: Michael S. Tsirkin <[email protected]>
# gpg: Signature made Wed 03 Sep 2014 14:41:23 BST using RSA key ID D28D5469
# gpg: Good signature from "Michael S. Tsirkin <[email protected]>"
# gpg:                 aka "Michael S. Tsirkin <[email protected]>"

* remotes/mst/tags/for_upstream:
  acpi-build: Set FORCE_APIC_CLUSTER_MODEL bit for FADT flags
  vhost-scsi: init backend features earlier
  vhost_net: init acked_features to backend_features
  vhost_net: start/stop guest notifiers properly

Signed-off-by: Peter Maydell <[email protected]>

Revert "vhost_net: start/stop guest notifiers properly"

This reverts commit aad4dce934649b3a398396fc2a76f215bb194ea4.

I accidentally merged the wrong version of a pull request
which had a buggy version of this patch. Reverting the
buggy version means we can then cleanly merge in the correct
pull with the corrected change.

Signed-off-by: Peter Maydell <[email protected]>

vbe: rework sanity checks

Plug a bunch of holes in the bochs dispi interface parameter checking.
Add a function doing verification on all registers.  Call that
unconditionally on every register write.  That way we should catch
everything, even changing one register affecting the valid range of
another register.

Some of the holes have been added by commit
e9c6149f6ae6873f14a12eea554925b6aa4c4dec.  Before that commit the
maximum possible framebuffer (VBE_DISPI_MAX_XRES * VBE_DISPI_MAX_YRES *
32 bpp) has been smaller than the qemu vga memory (8MB) and the checking
for VBE_DISPI_MAX_XRES + VBE_DISPI_MAX_YRES + VBE_DISPI_MAX_BPP was ok.

Some of the holes have been there forever, such as
VBE_DISPI_INDEX_X_OFFSET and VBE_DISPI_INDEX_Y_OFFSET register writes
lacking any verification.

Security impact:

(1) Guest can make the ui (gtk/vnc/...) use memory rages outside the vga
frame buffer as source  ->  host memory leak.  Memory isn't leaked to
the guest but to the vnc client though.

(2) Qemu will segfault in case the memory range happens to include
unmapped areas  ->  Guest can DoS itself.

The guest can not modify host memory, so I don't think this can be used
by the guest to escape.

CVE-2014-3615

Cc: [email protected]
Cc: [email protected]
Signed-off-by: Gerd Hoffmann <[email protected]>
Reviewed-by: Laszlo Ersek <[email protected]>

vbe: make bochs dispi interface return the correct memory size with qxl

VgaState->vram_size is the size of the pci bar.  In case of qxl not the
whole pci bar can be used as vga framebuffer.  Add a new variable
vbe_size to handle that case.  By default (if unset) it equals
vram_size, but qxl can set vbe_size to something else.

This makes sure VBE_DISPI_INDEX_VIDEO_MEMORY_64K returns correct results
and sanity checks are done with the correct size too.

Cc: [email protected]
Signed-off-by: Gerd Hoffmann <[email protected]>
Reviewed-by: Laszlo Ersek <[email protected]>

acpi-build: Set FORCE_APIC_CLUSTER_MODEL bit for FADT flags

If we start Windows 2008 R2 DataCenter with number of cpu less than 8,
The system will use APIC Flat Logical destination mode as default configuration,
Which has an upper limit of 8 CPUs.

The fault is that VM can not show all processors within Task Manager if
we hot-add cpus when the number of cpus in VM extends the limit of 8.

If we use cluster destination model, the problem will be solved.

Note:
This flag was introduced later than ACPI v1.0 specification while QEMU
generates v1.0 tables only, but...

linux kernel ignores this flag, so patch has no influence on it.

Tested with Win[XPsp3|Srv2003EE|Srv2008DC|Srv2008R2|Srv2012R2], there
isn't BSODs and guests boot just fine. In cases guest doesn't support
cpu-hotplug, cpu becomes visible after reboot and in case the guest
supports cpu-hotplug, it works as expected with this patch.

Cc: [email protected]
Signed-off-by: huangzhichao <[email protected]>
Signed-off-by: zhanghailiang <[email protected]>
Reviewed-by: Michael S. Tsirkin <[email protected]>
Signed-off-by: Michael S. Tsirkin <[email protected]>
Reviewed-By: Igor Mammedov <[email protected]>

vhost-scsi: init backend features earlier

As vhost core can use backend_features during init, clear it earlier to
avoid using uninitialized memory.
This use would be harmless since vhost scsi ignores the result
anyway, but initializing earlier will help prevent valgrind errors,
and make scsi and net behave similarly.

Cc: [email protected]
Acked-by: Paolo Bonzini <[email protected]>
Acked-by: Jason Wang <[email protected]>
Signed-off-by: Michael S. Tsirkin <[email protected]>

vhost_net: init acked_features to backend_features

commit 2e6d46d77ed328d34a94688da8371bcbe243479b (vhost: add
vhost_get_features and vhost_ack_features) removes the step that
initializes the acked_features to backend_features.

As this field is now uninitialized, vhost initialization will sometimes
fail.

To fix, initialize acked_features on each ack.

Tested-by: Andrey Korolyov <[email protected]>
Cc: Nikolay Nikolaev <[email protected]>
Cc: [email protected]
Signed-off-by: Jason Wang <[email protected]>
Reviewed-by: Michael S. Tsirkin <[email protected]>
Signed-off-by: Michael S. Tsirkin <[email protected]>

vhost_net: start/stop guest notifiers properly

commit a9f98bb5ebe6fb1869321dcc58e72041ae626ad8 "vhost: multiqueue
support" changed the order of stopping the device. Previously
vhost_dev_stop would disable backend and only afterwards, unset guest
notifiers. We now unset guest notifiers while vhost is still
active. This can lose interrupts causing guest networking to fail. In
particular, this has been observed during migration.

To fix this, several other changes are needed:
- remove the hdev->started assertion in vhost.c since we may want to
start the guest notifiers before vhost starts and stop the guest
notifiers after vhost is stopped.
- introduce the vhost_net_set_vq_index() and call it before setting
guest notifiers. This is to guarantee vhost_net has the correct
virtqueue index when setting guest notifiers.

MST: fix up error handling.

Cc: [email protected]
Cc: Jason Wang <[email protected]>
Signed-off-by: Michael S. Tsirkin <[email protected]>
Tested-by: Andrey Korolyov <[email protected]>
Reported-by: "Zhangjie (HZ)" <[email protected]>
Tested-by: William Dauchy <[email protected]>
Signed-off-by: Jason Wang <[email protected]>
Reviewed-by: Michael S. Tsirkin <[email protected]>

hw/9pfs: Use little-endian format for xattr values

With security_model=mapped-xattr, we encode the uid,gid and other file
attributes as extended attributes of the file. We save them under
user.virtfs.* namespace.

Use little-endian encoding for on-disk values. This enables us to export
the same directory from both little-endian and big-endian hosts.

NOTE: This will break big-endian host that have virtFS exports
using security model mapped-xattr. They will have to use external tools
to convert the xattr to little-endian format.

Signed-off-by: Aneesh Kumar K.V <[email protected]>

slirp: Honour vlan/stack in hostfwd_remove commands

The hostfwd_add and hostfwd_remove monitor commands allow the user
to optionally specify a vlan/stack tuple. hostfwd_add honours this,
but hostfwd_remove does not (it looks up the tuple but then ignores
the SlirpState it has looked up and always uses the first stack
in the list anyway). Correct this to honour what the user requested.

Signed-off-by: Peter Maydell <[email protected]>
Signed-off-by: Michael Tokarev <[email protected]>

hmp: fix MemdevList memory leak

the memdev_list in hmp_info_memdev() is never freed.
so we use existent method qapi_free_MemdevList() to free it.
and also we can use qapi_free_MemdevList() to replace list loops
to clean up the memdev list in error path.

Signed-off-by: Chen Fan <[email protected]>
Reviewed-by: Peter Crosthwaite <[email protected]>
Reviewed-by: Hu Tao <[email protected]>
Signed-off-by: Michael Tokarev <[email protected]>

qom/object.c, hmp.c: fix string_output_get_string() memory leak

string_output_get_string() uses g_string_free(str, false) to
transfer the 'str' pointer to callers and never free it.

Signed-off-by: Chen Fan <[email protected]>
Reviewed-by: Peter Crosthwaite <[email protected]>
Reviewed-by: Hu Tao <[email protected]>
Signed-off-by: Michael Tokarev <[email protected]>

query-memdev: fix potential memory leaks

Signed-off-by: Chen Fan <[email protected]>
Reviewed-by: Peter Crosthwaite <[email protected]>
Reviewed-by: Hu Tao <[email protected]>
Signed-off-by: Michael Tokarev <[email protected]>

MAINTAINERS: Add VMWare devices maintainer

Signed-off-by: Dmitry Fleytman <[email protected]>
Reviewed-by: Stefan Hajnoczi <[email protected]>
Signed-off-by: Michael Tokarev <[email protected]>

device_tree.c: dump all err mesages with error_report

Signed-off-by: Li Liu <[email protected]>
Signed-off-by: Michael Tokarev <[email protected]>

device_tree.c: redirect load_device_tree err message to stderr

Reviewed-by: Peter Crosthwaite <[email protected]>
Signed-off-by: Li Liu <[email protected]>
Signed-off-by: Michael Tokarev <[email protected]>

scripts: Remove scripts/qtest

This is a dummy file with no user, drop it.

Signed-off-by: Fam Zheng <[email protected]>
Reviewed-by: Stefan Hajnoczi <[email protected]>
Signed-off-by: Michael Tokarev <[email protected]>

Fix debug print warning

Steps:

1.enable qemu debug print, using simply scprit as below:
 grep "//#define DEBUG" * -rl | xargs sed -i "s/\/\/#define DEBUG/#define DEBUG/g"
2. make -j
3. get some warning:
hw/i2c/pm_smbus.c: In function 'smb_ioport_writeb':
hw/i2c/pm_smbus.c:142: warning: format '%04x' expects type 'unsigned int', but argument 2 has type 'hwaddr'
hw/i2c/pm_smbus.c:142: warning: format '%02x' expects type 'unsigned int', but argument 3 has type 'uint64_t'
hw/i2c/pm_smbus.c: In function 'smb_ioport_readb':
hw/i2c/pm_smbus.c:209: warning: format '%04x' expects type 'unsigned int', but argument 2 has type 'hwaddr'
hw/intc/i8259.c: In function 'pic_ioport_read':
hw/intc/i8259.c:373: warning: format '%02x' expects type 'unsigned int', but argument 2 has type 'hwaddr'
hw/input/pckbd.c: In function 'kbd_write_command':
hw/input/pckbd.c:232: warning: format '%02x' expects type 'unsigned int', but argument 2 has type 'uint64_t'
hw/input/pckbd.c: In function 'kbd_write_data':
hw/input/pckbd.c:333: warning: format '%02x' expects type 'unsigned int', but argument 2 has type 'uint64_t'
hw/isa/apm.c: In function 'apm_ioport_writeb':
hw/isa/apm.c:44: warning: format '%x' expects type 'unsigned int', but argument 2 has type 'hwaddr'
hw/isa/apm.c:44: warning: format '%02x' expects type 'unsigned int', but argument 3 has type 'uint64_t'
hw/isa/apm.c: In function 'apm_ioport_readb':
hw/isa/apm.c:67: warning: format '%x' expects type 'unsigned int', but argument 2 has type 'hwaddr'
hw/timer/mc146818rtc.c: In function 'cmos_ioport_write':
hw/timer/mc146818rtc.c:394: warning: format '%02x' expects type 'unsigned int', but argument 3 has type 'uint64_t'
hw/i386/pc.c: In function 'port92_write':
hw/i386/pc.c:479: warning: format '%02x' expects type 'unsigned int', but argument 2 has type 'uint64_t'

Fix them.

Cc: [email protected]
Signed-off-by: Gonglei <[email protected]>
Signed-off-by: Michael Tokarev <[email protected]>

curl: The macro that you have to uncomment to get debugging is DEBUG_CURL.

Signed-off-by: Richard W.M. Jones <[email protected]>
Signed-off-by: Michael Tokarev <[email protected]>

Merge remote-tracking branch 'remotes/mst/tags/for_upstream' into staging

pci, pc fixes, features

A bunch of bugfixes - these will make sense for 2.1.1

Initial Intel IOMMU support.

Signed-off-by: Michael S. Tsirkin <[email protected]>
# gpg: Signature made Tue 02 Sep 2014 16:05:04 BST using RSA key ID D28D5469
# gpg: Good signature from "Michael S. Tsirkin <[email protected]>"
# gpg:                 aka "Michael S. Tsirkin <[email protected]>"

* remotes/mst/tags/for_upstream:
  vhost_net: start/stop guest notifiers properly
  pci: avoid losing config updates to MSI/MSIX cap regs
  virtio-net: don't run bh on vm stopped
  ioh3420: remove unused ioh3420_init() declaration
  vhost_net: cleanup start/stop condition
  intel-iommu: add IOTLB using hash table
  intel-iommu: add context-cache to cache context-entry
  intel-iommu: add supports for queued invalidation interface
  intel-iommu: fix coding style issues around in q35.c and machine.c
  intel-iommu: add Intel IOMMU emulation to q35 and add a machine option "iommu" as a switch
  intel-iommu: add DMAR table to ACPI tables
  intel-iommu: introduce Intel IOMMU (VT-d) emulation
  iommu: add is_write as a parameter to the translate function of MemoryRegionIOMMUOps

Signed-off-by: Peter Maydell <[email protected]>

vhost_net: start/stop guest notifiers properly

commit a9f98bb5ebe6fb1869321dcc58e72041ae626ad8 vhost: multiqueue
support changed the order of stopping the device. Previously
vhost_dev_stop would disable backend and only afterwards, unset guest
notifiers. We now unset guest notifiers while vhost is still
active. This can lose interrupts causing guest networking to fail. In
particular, this has been observed during migration.

To adapt this, several other changes are needed:
- remove the hdev->started assertion in vhost.c since we may want to
start the guest notifiers before vhost starts and stop the guest
notifiers after vhost is stopped.
- introduce the vhost_net_set_vq_index() and call it before setting
guest notifiers. This is used to guarantee vhost_net has the correct
virtqueue index when setting guest notifiers.

Cc: [email protected]
Reported-by: "Zhangjie (HZ)" <[email protected]>
Tested-by: William Dauchy <[email protected]>
Signed-off-by: Michael S. Tsirkin <[email protected]>
Signed-off-by: Jason Wang <[email protected]>
Reviewed-by: Michael S. Tsirkin <[email protected]>
Signed-off-by: Michael S. Tsirkin <[email protected]>

pci: avoid losing config updates to MSI/MSIX cap regs

Since
commit 95d658002401e2e47a5404298ebe9508846e8a39
    msi: Invoke msi/msix_write_config from PCI core
msix config writes are lost, the value written is always 0.

Fix pci_default_write_config to avoid this.

Cc: [email protected]
Signed-off-by: Knut Omang <[email protected]>
Reviewed-by: Michael S. Tsirkin <[email protected]>
Signed-off-by: Michael S. Tsirkin <[email protected]>

virtio-net: don't run bh on vm stopped

commit 783e7706937fe15523b609b545587a028a2bdd03
    virtio-net: stop/start bh when appropriate

is incomplete: BH might execute within the same main loop iteration but
after vmstop, so in theory, we might trigger an assertion.
I was unable to reproduce this in practice,
but it seems clear enough that the potential is there, so worth fixing.

Cc: [email protected]
Reported-by: Stefan Hajnoczi <[email protected]>
Signed-off-by: Michael S. Tsirkin <[email protected]>

ioh3420: remove unused ioh3420_init() declaration

commit 0f9b1771ccc65873a8376c81200a437aa58c2f6d
    ioh3420: Remove obsoleted, unused ioh3420_init function
removed the implementation of ioh3420_init

Drop the declaration from the header file as well.

Signed-off-by: Gonglei <[email protected]>
Reviewed-by: Knut Omang <[email protected]>
Reviewed-by: Michael S. Tsirkin <[email protected]>
Signed-off-by: Michael S. Tsirkin <[email protected]>

vhost_net: cleanup start/stop condition

Checking vhost device internal state in vhost_net looks like
a layering violation since vhost_net does not
set this flag: it is set and tested by vhost.c.
There seems to be no reason to check this:
caller in virtio net uses its own flag,
vhost_started, to ensure vhost is started/stopped
as appropriate.

Signed-off-by: Michael S. Tsirkin <[email protected]>
Reviewed-by: Amos Kong <[email protected]>

Merge remote-tracking branch 'remotes/spice/tags/pull-spice-20140902-1' into staging

sanity check for qxl, minor spice display channel tweak.

# gpg: Signature made Tue 02 Sep 2014 09:53:39 BST using RSA key ID D3E87138
# gpg: Good signature from "Gerd Hoffmann (work) <[email protected]>"
# gpg:                 aka "Gerd Hoffmann <[email protected]>"
# gpg:                 aka "Gerd Hoffmann (private) <[email protected]>"

* remotes/spice/tags/pull-spice-20140902-1:
  spice: use console index as display id
  qxl-render: add more sanity checks

Signed-off-by: Peter Maydell <[email protected]>

implementing victim TLB for QEMU system emulated TLB

QEMU system mode page table walks are expensive. Taken by running QEMU
qemu-system-x86_64 system mode on Intel PIN , a TLB miss and walking a
4-level page tables in guest Linux OS takes ~450 X86 instructions on
average.

QEMU system mode TLB is implemented using a directly-mapped hashtable.
This structure suffers from conflict misses. Increasing the
associativity of the TLB may not be the solution to conflict misses as
all the ways may have to be walked in serial.

A victim TLB is a TLB used to hold translations evicted from the
primary TLB upon replacement. The victim TLB lies between the main TLB
and its refill path. Victim TLB is of greater associativity (fully
associative in this patch). It takes longer to lookup the victim TLB,
but its likely better than a full page table walk. The memory
translation path is changed as follows :

Before Victim TLB:
1. Inline TLB lookup
2. Exit code cache on TLB miss.
3. Check for unaligned, IO accesses
4. TLB refill.
5. Do the memory access.
6. Return to code cache.

After Victim TLB:
1. Inline TLB lookup
2. Exit code cache on TLB miss.
3. Check for unaligned, IO accesses
4. Victim TLB lookup.
5. If victim TLB misses, TLB refill
6. Do the memory access.
7. Return to code cache

The advantage is that victim TLB can offer more associativity to a
directly mapped TLB and thus potentially fewer page table walks while
still keeping the time taken to flush within reasonable limits.
However, placing a victim TLB before the refill path increase TLB
refill path as the victim TLB is consulted before the TLB refill. The
performance results demonstrate that the pros outweigh the cons.

some performance results taken on SPECINT2006 train
datasets and kernel boot and qemu configure script on an
Intel(R) Xeon(R) CPU  E5620  @ 2.40GHz Linux machine are shown in the
Google Doc link below.

https://docs.google.com/spreadsheets/d/1eiItzekZwNQOal_h-5iJmC4tMDi051m9qidi5_nwvH4/edit?usp=sharing

In summary, victim TLB improves the performance of qemu-system-x86_64 by
11% on average on SPECINT2006, kernelboot and qemu configscript and with
highest improvement of in 26% in 456.hmmer. And victim TLB does not result
in any performance degradation in any of the measured benchmarks. Furthermore,
the implemented victim TLB is architecture independent and is expected to
benefit other architectures in QEMU as well.

Although there are measurement fluctuations, the performance
improvement is very significant and by no means in the range of
noises.

Signed-off-by: Xin Tong <[email protected]>
Message-id: 1407202523[email protected]
Reviewed-by: Peter Maydell <[email protected]>
Signed-off-by: Peter Maydell <[email protected]>

target-tricore: Add instructions of SR opcode format

Add instructions of SR opcode format.
Add micro-op generator functions for saturate.
Add helper return from exception (rfe).

Signed-off-by: Bastian Koppelmann <[email protected]>
Message-id: 1409572800[email protected]
Signed-off-by: Peter Maydell <[email protected]>

target-tricore: Add instructions of SLR, SSRO and SRO opcode format

Add instructions of SLR, SSRO and SRO opcode format.

Signed-off-by: Bastian Koppelmann <[email protected]>
Reviewed-by: Richard Henderson <[email protected]>
Message-id: 1409572800[email protected]
Signed-off-by: Peter Maydell <[email protected]>

target-tricore: Add instructions of SC opcode format

Add instructions of SC opcode format.
Add helper for begin interrupt service routine.

Signed-off-by: Bastian Koppelmann <[email protected]>
Message-id: 1409572800[email protected]
Signed-off-by: Peter Maydell <[email protected]>

target-tricore: Add instructions of SBR opcode format

Add instructions of SBR opcode format.
Add gen_loop micro-op generator function.

Signed-off-by: Bastian Koppelmann <[email protected]>
Reviewed-by: Richard Henderson <[email protected]>
Message-id: 1409572800[email protected]
Signed-off-by: Peter Maydell <[email protected]>

target-tricore: Add instructions of SBC and SBRN opcode format

Add instructions of SBC and SBRN opcode format.

Signed-off-by: Bastian Koppelmann <[email protected]>
Reviewed-by: Richard Henderson <[email protected]>
Message-id: 1409572800[email protected]
Signed-off-by: Peter Maydell <[email protected]>

target-tricore: Add instructions of SB opcode format

Add instructions of SB opcode format.
Add helper call/ret.
Add micro-op generator functions for branches.
Add makro to generate helper functions.

Signed-off-by: Bastian Koppelmann <[email protected]>
Message-id: 1409572800[email protected]
Signed-off-by: Peter Maydell <[email protected]>

target-tricore: Add instructions of SRRS and SLRO opcode format

Add instructions of SSRS and SLRO opcode format.
Add micro-op generator functions for offset loads.

Signed-off-by: Bastian Koppelmann <[email protected]>
Reviewed-by: Richard Henderson <[email protected]>
Message-id: 1409572800[email protected]
Signed-off-by: Peter Maydell <[email protected]>

target-tricore: Add instructions of SSR opcode format

Add instructions of SSR opcode format.

Signed-off-by: Bastian Koppelmann <[email protected]>
Reviewed-by: Richard Henderson <[email protected]>
Message-id: 1409572800[email protected]
Signed-off-by: Peter Maydell <[email protected]>

target-tricore: Add instructions of SRR opcode format

Add instructions of SRR opcode format.
Add helper for add/sub_ssov.

Signed-off-by: Bastian Koppelmann <[email protected]>
Message-id: 1409572800[email protected]
Signed-off-by: Peter Maydell <[email protected]>

target-tricore: Add instructions of SRC opcode format

Add instructions of SRC opcode format.
Add micro-op generator functions for add, conditional add/sub and shi/shai.

Signed-off-by: Bastian Koppelmann <[email protected]>
Reviewed-by: Richard Henderson <[email protected]>
Message-id: 1409572800[email protected]
Signed-off-by: Peter Maydell <[email protected]>

target-tricore: Add masks and opcodes for decoding

Add masks and opcodes for decoding TriCore instructions.

Signed-off-by: Bastian Koppelmann <[email protected]>
Reviewed-by: Richard Henderson <[email protected]>
Message-id: 1409572800[email protected]
Signed-off-by: Peter Maydell <[email protected]>

target-tricore: Add initialization for translation and activate target

Add tcg and cpu model initialization.
Add gen_intermediate_code function.
Activate target in configure and add softmmu config.

Signed-off-by: Bastian Koppelmann <[email protected]>
Message-id: 1409572800[email protected]
Signed-off-by: Peter Maydell <[email protected]>

target-tricore: Add softmmu support

Add basic softmmu support for TriCore

Signed-off-by: Bastian Koppelmann <[email protected]>
Message-id: 1409572800[email protected]
Signed-off-by: Peter Maydell <[email protected]>

target-tricore: Add board for systemmode

Add basic board to allow systemmode emulation

Signed-off-by: Bastian Koppelmann <[email protected]>
Message-id: 1409572800[email protected]
Signed-off-by: Peter Maydell <[email protected]>

target-tricore: Add target stubs and qom-cpu

Add TriCore target stubs, and QOM cpu, and Maintainer

Signed-off-by: Bastian Koppelmann <[email protected]>
Message-id: 1409572800[email protected]
Signed-off-by: Peter Maydell <[email protected]>

Merge remote-tracking branch 'remotes/borntraeger/tags/kvm-s390-20140901' into staging

s390x/kvm: Several updates/fixes/features

1. s390x/kvm: avoid synchronize_rcu's in kernel
----------------------------------------------
The first patches change s390x/kvm code to issue VCPU specific ioctls
from the VCPU thread. This will avoid unnecessary synchronize_rcu in
the kernel, which caused a noticably slowdown with many guest CPUs.
It speeds up all start/restart/reset operations involving cpus
drastically.

2. s390-ccw.img: block size and DASD format support
---------------------------------------------------
The second part changes the s390-ccw bios to IPL (boot)  more disk
formats than before. Furthermore a small fix is made to the console
output of the bios.

3. s390: Support for Hotplug of Standby Memory
----------------------------------------------
The third part adds support in s390 for a pool of standby memory,
which can be set online/offline by the guest (ie, via chmem).
The standby pool of memory is allocated as the difference between
the initial memory setting and the maxmem setting.
As part of this work, additional results are provided for the
Read SCP Information SCLP, and new implentation is added for the
Read Storage Element Information, Attach Storage Element,
Assign Storage and Unassign Storage SCLPs, which enables the s390
guest to manipulate the standby memory pool.

This patchset is based on work originally done by Jeng-Fang (Nick)
Wang.

Sample qemu command snippet:

qemu -machine s390-ccw-virtio  -m 1024M,maxmem=2048M,slots=32 -enable-kvm

This will allocate 1024M of active memory, and another 1024M
of standby memory.  Example output from s390-tools lsmem:
=============================================================================
0x0000000000000000-0x000000000fffffff        256  online   no         0-127
0x0000000010000000-0x000000001fffffff        256  online   yes        128-255
0x0000000020000000-0x000000003fffffff        512  online   no         256-511
0x0000000040000000-0x000000007fffffff       1024  offline  -          512-1023

Memory device size  : 2 MB
Memory block size   : 256 MB
Total online memory : 1024 MB
Total offline memory: 1024 MB

The guest can dynamically enable part or all of the standby pool
via the s390-tools chmem, for example:

chmem -e 512M

And can attempt to dynamically disable:

chmem -d 512M

4. s390x/gdb: various fixes
---------------------------
* Patch 1 fixes a bug where the cc was changed accidentally.
* Patch 2 adds the gdb feature XML files for s390x
* Patch 3 Define acr and fpr registers as coprocessor registers. This allows us
   to reuse the feature XML files.
* Patch 4 whitespace fixes

# gpg: Signature made Mon 01 Sep 2014 12:53:39 BST using RSA key ID B5A61C7C
# gpg: Can't check signature: public key not found

* remotes/borntraeger/tags/kvm-s390-20140901:
  s390x/gdb: coding style fixes
  s390x/gdb: generate target.xml and handle fp/ac as coprocessors
  s390x/gdb: add the feature xml files for s390x
  s390x/gdb: don't touch the cc if tcg is not enabled
  sclp-s390: Add memory hotplug SCLPs
  s390-virtio: Apply same memory boundaries as virtio-ccw
  virtio-ccw: Include standby memory when calculating storage increment
  sclp-s390: Add device to manage s390 memory hotplug
  pc-bios/s390-ccw.img binary update
  pc-bios/s390-ccw: Do proper console setup
  pc-bios/s390-ccw: IPL from DASD with format variations
  pc-bios/s390-ccw Really big EAV ECKD DASD handling
  pc-bios/s390-ccw Improve ECKD informational message
  pc-bios/s390-ccw: handle more ECKD DASD block sizes
  pc-bios/s390-ccw: support all virtio block size
  s390x/kvm: execute the first cpu reset on the vcpu thread
  s390x/kvm: execute "system reset" cpu resets on the vcpu thread
  s390x/kvm: execute sigp orders on the target vcpu thread
  s390x/kvm: run guest triggered resets on the target vcpu thread

Signed-off-by: Peter Maydell <[email protected]>

spice: use console index as display id

... instead of maintaining our own numbering.

Signed-off-by: Gerd Hoffmann <[email protected]>

qxl-render: add more sanity checks

Damn, the dirty rectangle values are signed integers.  So the checks
added by commit 788fbf042fc6d5aaeab56757e6dad622ac5f0c21 are not good
enough, we also have to make sure they are not negative.

[ Note: There must be something broken in spice-server so we get
  negative values in the first place.  Bug opened:
  https://bugzilla.redhat.com/show_bug.cgi?id=1135372 ]

Cc: [email protected]
Signed-off-by: Gerd Hoffmann <[email protected]>
Reviewed-by: Dr. David Alan Gilbert <[email protected]>

s390x/gdb: coding style fixes

This patch cleanes up two coding style issues (missing whitespaces).

Signed-off-by: David Hildenbrand <[email protected]>
Acked-by: Cornelia Huck <[email protected]>
Acked-by: Christian Borntraeger <[email protected]>
Signed-off-by: Jens Freimann <[email protected]>
Signed-off-by: Christian Borntraeger <[email protected]>

s390x/gdb: generate target.xml and handle fp/ac as coprocessors

This patch reduces the core registers to the psw and the general purpose
registers. The fpc and ac registers are handled as coprocessors registers by gdb.
This allows to reuse the feature xml files taken from gdb without further
modification and is what other architectures do.

The target.xml is now generated and provided to the gdb client. Therefore, the
client doesn't have to guess which registers are available at which logical
register number.

Signed-off-by: David Hildenbrand <[email protected]>
Acked-by: Cornelia Huck <[email protected]>
Acked-by: Christian Borntraeger <[email protected]>
Signed-off-by: Jens Freimann <[email protected]>
Signed-off-by: Christian Borntraeger <[email protected]>

s390x/gdb: add the feature xml files for s390x

This patch adds the relevant s390x feature xml files taken from gdb.

Signed-off-by: David Hildenbrand <[email protected]>
Acked-by: Cornelia Huck <[email protected]>
Acked-by: Christian Borntraeger <[email protected]>
Signed-off-by: Jens Freimann <[email protected]>
Signed-off-by: Christian Borntraeger <[email protected]>

s390x/gdb: don't touch the cc if tcg is not enabled

When reading/writing the psw mask, the condition code may only be touched if
running on tcg.

Signed-off-by: David Hildenbrand <[email protected]>
Signed-off-by: Jens Freimann <[email protected]>
Signed-off-by: Christian Borntraeger <[email protected]>

sclp-s390: Add memory hotplug SCLPs

Add memory information to read SCP info and add handlers for
Read Storage Element Information, Attach Storage Element,
Assign Storage and Unassign Storage.

Signed-off-by: Matthew Rosato <[email protected]>
Signed-off-by: Christian Borntraeger <[email protected]>

s390-virtio: Apply same memory boundaries as virtio-ccw

Although s390-virtio won't support memory hotplug, it should
enforce the same memory boundaries so that it can use shared codepaths
(like read_SCP_info).

Signed-off-by: Matthew Rosato <[email protected]>
Signed-off-by: Christian Borntraeger <[email protected]>

virtio-ccw: Include standby memory when calculating storage increment

When determining the memory increment size, use the maxmem size if
it was specified.

Signed-off-by: Matthew Rosato <[email protected]>
Signed-off-by: Christian Borntraeger <[email protected]>

sclp-s390: Add device to manage s390 memory hotplug

Add sclpMemoryHotplugDev to contain associated data structures, etc.

Signed-off-by: Matthew Rosato <[email protected]>
Signed-off-by: Christian Borntraeger <[email protected]>

pc-bios/s390-ccw.img binary update

Rebuild of s390-ccw.img containing these patches:

  pc-bios/s390-ccw: Do proper console setup
  pc-bios/s390-ccw: support all virtio block size
  pc-bios/s390-ccw: handle more ECKD DASD block sizes
  pc-bios/s390-ccw Improve ECKD informational message
  pc-bios/s390-ccw Really big EAV ECKD DASD handling
  pc-bios/s390-ccw: IPL from DASD with format variations

Signed-off-by: Jens Freimann <[email protected]>
Signed-off-by: Christian Borntraeger <[email protected]>

pc-bios/s390-ccw: Do proper console setup

The final newline/return must happen before we reset the sclp via
diag 308.

Signed-off-by: Christian Borntraeger <[email protected]>
Reviewed-by: David Hildenbrand <[email protected]>
Acked-by: Cornelia Huck <[email protected]>
Signed-off-by: Jens Freimann <[email protected]>
Signed-off-by: Christian Borntraeger <[email protected]>

pc-bios/s390-ccw: IPL from DASD with format variations

There are two known cases of DASD format where signatures are
incomplete or absent:

1. result of <dasdfmt -d ldl -L ...> (ECKD_LDL_UNLABELED)
2. CDL with zero keys in IPL1 and IPL2 records

Now the code attempts to
1. find zIPL and use SCSI layout
2. find IPL1 and use CDL layout
3. find CMS1 and use LDL layout
3. find LNX1 and use LDL layout
4. find zIPL and use unlabeled LDL layout
5. find zIPL and use CDL layout
6. die
in this sequence.

Signed-off-by: Eugene (jno) Dvurechenski <[email protected]>
Acked-by: Cornelia Huck <[email protected]>
Acked-by: Christian Borntraeger <[email protected]>
Reviewed-by: David Hildenbrand <[email protected]>
Signed-off-by: Jens Freimann <[email protected]>
Signed-off-by: Christian Borntraeger <[email protected]>

pc-bios/s390-ccw Really big EAV ECKD DASD handling

For EAV ECKD DASD, the cylinder count will have the magic value
0xfffeU. Therefore, use the block number to test for valid eckd
addresses instead.

Signed-off-by: Eugene (jno) Dvurechenski <[email protected]>
Acked-by: Cornelia Huck <[email protected]>
Acked-by: Christian Borntraeger <[email protected]>
Reviewed-by: David Hildenbrand <[email protected]>
Signed-off-by: Jens Freimann <[email protected]>
Signed-off-by: Christian Borntraeger <[email protected]>

pc-bios/s390-ccw Improve ECKD informational message

Add block size display to ECKD scheme report.

Signed-off-by: Eugene (jno) Dvurechenski <[email protected]>
Reviewed-by: David Hildenbrand <[email protected]>
Acked-by: Christian Borntraeger <[email protected]>
Acked-by: Cornelia Huck <[email protected]>
Signed-off-by: Jens Freimann <[email protected]>
Signed-off-by: Christian Borntraeger <[email protected]>

pc-bios/s390-ccw: handle more ECKD DASD block sizes

Using dasdfmt(8) to format a DASD allows to choose a block size.
There are four supported values: 512, 1024, 2048, and 4096 bytes
per block. Each block size leads to selection of new count of
sectors per track. The head count remains always the same: 15.

This empiric knowledge is used to detect ECKD DASD to IPL from.

Signed-off-by: Eugene (jno) Dvurechenski <[email protected]>
Reviewed-by: David Hildenbrand <[email protected]>
Acked-by: Christian Borntraeger <[email protected]>
Signed-off-by: Jens Freimann <[email protected]>
Signed-off-by: Christian Borntraeger <[email protected]>

pc-bios/s390-ccw: support all virtio block size

The block size value may be given "as is" OR as a base value and
a shift count (exponent). So, we have to use calculation to get
the proper number in the code.

The main expression reads as
        (blk_cfg.blk_size << blk_cfg.physical_block_exp)

E.g., various combinations between blk_size=1/physical_block_exp=12
and blk_size=4096/physical_block_exp=0 are valid for 4K blocks.

Signed-off-by: Eugene (jno) Dvurechenski <[email protected]>
Acked-by: Christian Borntraeger <[email protected]>
Reviewed-by: David Hildenbrand <[email protected]>
Signed-off-by: Jens Freimann <[email protected]>
Signed-off-by: Christian Borntraeger <[email protected]>

s390x/kvm: execute the first cpu reset on the vcpu thread

As all full cpu resets currently call into the kernel to do initial cpu reset,
let's run this reset (triggered by cpu_s390x_init()) on the proper vcpu thread.

Reviewed-by: Cornelia Huck <[email protected]>
Acked-by: Christian Borntraeger <[email protected]>
Signed-off-by: David Hildenbrand <[email protected]>
Signed-off-by: Jens Freimann <[email protected]>
Signed-off-by: Christian Borntraeger <[email protected]>

s390x/kvm: execute "system reset" cpu resets on the vcpu thread

Let's execute resets triggered by qemu system resets on the target vcpu thread.
This will avoid synchronize_rcu's in the kernel.

Reviewed-by: Cornelia Huck <[email protected]>
Acked-by: Christian Borntraeger <[email protected]>
Signed-off-by: David Hildenbrand <[email protected]>
Signed-off-by: Jens Freimann <[email protected]>
Signed-off-by: Christian Borntraeger <[email protected]>

s390x/kvm: execute sigp orders on the target vcpu thread

All sigp orders that can result in ioctls on the target vcpu should be executed
on the associated vcpu thread.

Reviewed-by: Cornelia Huck <[email protected]>
Acked-by: Christian Borntraeger <[email protected]>
Signed-off-by: David Hildenbrand <[email protected]>
Signed-off-by: Jens Freimann <[email protected]>
Signed-off-by: Christian Borntraeger <[email protected]>

s390x/kvm: run guest triggered resets on the target vcpu thread

Currently, load_normal_reset() and modified_clear_reset() as triggered
by a guest vcpu will initiate cpu resets on the current vcpu thread for
all cpus. The reset should happen on the individual vcpu thread
instead, so let's use run_on_cpu() for this.

This avoids calls to synchronize_rcu() in the kernel.

Reviewed-by: Cornelia Huck <[email protected]>
Acked-by: Christian Borntraeger <[email protected]>
Signed-off-by: David Hildenbrand <[email protected]>
Signed-off-by: Jens Freimann <[email protected]>
Signed-off-by: Christian Borntraeger <[email protected]>

Merge remote-tracking branch 'remotes/stefanha/tags/block-pull-request' into staging

Block pull request

# gpg: Signature made Fri 29 Aug 2014 17:25:58 BST using RSA key ID 81AB73C8
# gpg: Good signature from "Stefan Hajnoczi <[email protected]>"
# gpg:                 aka "Stefan Hajnoczi <[email protected]>"

* remotes/stefanha/tags/block-pull-request: (35 commits)
  quorum: Fix leak of opts in quorum_open
  blkverify: Fix leak of opts in blkverify_open
  nfs: Fix leak of opts in nfs_file_open
  curl: Don't deref NULL pointer in call to aio_poll.
  curl: Allow a cookie or cookies to be sent with http/https requests.
  virtio-blk: allow drive_del with dataplane
  block: acquire AioContext in do_drive_del()
  linux-aio: avoid deadlock in nested aio_poll() calls
  qemu-iotests: add multiwrite test cases
  block: fix overlapping multiwrite requests
  nbd: Follow the BDS' AIO context
  block: Add AIO context notifiers
  nbd: Drop nbd_can_read()
  sheepdog: fix a core dump while do auto-reconnecting
  aio-win32: add support for sockets
  qemu-coroutine-io: fix for Win32
  AioContext: introduce aio_prepare
  aio-win32: add aio_set_dispatching optimization
  test-aio: test timers on Windows too
  AioContext: export and use aio_dispatch
  ...

Signed-off-by: Peter Maydell <[email protected]>

quorum: Fix leak of opts in quorum_open

Signed-off-by: Fam Zheng <[email protected]>
Reviewed-by: Benoît Canet <[email protected]>
Signed-off-by: Stefan Hajnoczi <[email protected]>