error: Don't use error_report() for assertion msgs.

Use fprintf(stderr instead. This removes dependency of libqemuutil.a
on the monitor.

We can further justify this change, in that this code path should only
trigger under a fatal error condition. fprintf-stderr is probably the
appropriate medium as under a fatal error conidition the monitor itself
may be down and out for the count. So assertion failure messages should
go lowest common denominator - straight to stderr.

Fixes the build as reported by Kevin Wolf. Issue debugged and change
suggested by Luiz Capitulino. Issue introduced by
5d24ee70bcbcf578614193526bcd5ed30a8eb16c.

Signed-off-by: Peter Crosthwaite <[email protected]>
Reviewed-by: Edgar E. Iglesias <[email protected]>
Signed-off-by: Edgar E. Iglesias <[email protected]>

Merge remote branch 'luiz/queue/qmp' into qmpq

* luiz/queue/qmp:
  migration: qmp_migrate(): keep working after syntax error
  qerror: Remove assert_no_error()
  qemu-option: Remove qemu_opts_create_nofail
  target-i386: Remove assert_no_error usage
  hw: Remove assert_no_error usages
  qdev: Delete dead code
  error: Add error_abort
  monitor: add object-add (QMP) and object_add (HMP) command
  monitor: add object-del (QMP) and object_del (HMP) command
  qom: catch errors in object_property_add_child
  qom: fix leak for objects created with -object
  rng: initialize file descriptor to -1
  qemu-monitor: HMP cpu-add wrapper
  vl: add missing transition debug->finish_migrate

Message-Id: 1389045795[email protected]
Signed-off-by: Edgar E. Iglesias <[email protected]>

Microblaze: Convert Microblaze-pic handling to GPIOs

This patch uses inbound GPIO lines (IRQ and FIR) for
interrupts instead of using the old pic_cpu method,
which doesn't correspond to real hardware.

This creates the CPU's inbound IRQ and FIR GPIO lines and
updates the Microblaze boards to use this new method.

Signed-off-by: Alistair Francis <[email protected]>
Suggested-by: Peter Crosthwaite <[email protected]>
Reveiwed-by: Edgar E. Iglesias <[email protected]>
Tested-by: Edgar E. Iglesias <[email protected]>
Signed-off-by: Edgar E. Iglesias <[email protected]>

target-arm: Switch ARMCPUInfo arrays to use terminator entries

Switch the ARMCPUInfo arrays in cpu.c and cpu64.c to use a terminator
entry rather than looping based on ARRAY_SIZE. The latter causes
compile warnings on some versions of gcc if the configure options
happen to result in an empty array.

Signed-off-by: Peter Maydell <[email protected]>
Reviewed-by: Edgar E. Iglesias <[email protected]>
Tested-by: Edgar E. Iglesias <[email protected]>
Signed-off-by: Edgar E. Iglesias <[email protected]>

Merge remote-tracking branch 'quintela/tags/migration/20140113' into staging

migration.next for 20140113

# gpg: Signature made Mon 13 Jan 2014 09:38:27 AM PST using RSA key ID 5872D723
# gpg: Can't check signature: public key not found

* quintela/tags/migration/20140113: (49 commits)
  migration: synchronize memory bitmap 64bits at a time
  ram: split function that synchronizes a range
  memory: syncronize kvm bitmap using bitmaps operations
  memory: move bitmap synchronization to its own function
  kvm: refactor start address calculation
  kvm: use directly cpu_physical_memory_* api for tracking dirty pages
  memory: unfold memory_region_test_and_clear()
  memory: split cpu_physical_memory_* functions to its own include
  memory: cpu_physical_memory_set_dirty_tracking() should return void
  memory: make cpu_physical_memory_reset_dirty() take a length parameter
  memory: s/dirty/clean/ in cpu_physical_memory_is_dirty()
  memory: cpu_physical_memory_clear_dirty_range() now uses bitmap operations
  memory: cpu_physical_memory_set_dirty_range() now uses bitmap operations
  memory: use find_next_bit() to find dirty bits
  memory: s/mask/clear/ cpu_physical_memory_mask_dirty_range
  memory: cpu_physical_memory_get_dirty() is used as returning a bool
  memory: make cpu_physical_memory_get_dirty() the main function
  memory: unfold cpu_physical_memory_set_dirty_flag()
  memory: unfold cpu_physical_memory_set_dirty() in its only user
  memory: unfold cpu_physical_memory_clear_dirty_flag() in its only user
  ...

Message-id: 1389634834[email protected]
Signed-off-by: Anthony Liguori <[email protected]>

migration: synchronize memory bitmap 64bits at a time

We use the old code if the bitmaps are not aligned

Signed-off-by: Juan Quintela <[email protected]>
Reviewed-by: Orit Wasserman <[email protected]>

ram: split function that synchronizes a range

This function is the only bit where we care about speed.

Signed-off-by: Juan Quintela <[email protected]>
Reviewed-by: Orit Wasserman <[email protected]>

memory: syncronize kvm bitmap using bitmaps operations

If bitmaps are aligned properly, use bitmap operations.  If they are
not, just use old bit at a time code.

Signed-off-by: Juan Quintela <[email protected]>
Reviewed-by: Orit Wasserman <[email protected]>

memory: move bitmap synchronization to its own function

We want to have all the functions that handle directly the dirty
bitmap near.  We will change it later.

Signed-off-by: Juan Quintela <[email protected]>
Reviewed-by: Orit Wasserman <[email protected]>

kvm: refactor start address calculation

Signed-off-by: Juan Quintela <[email protected]>
Reviewed-by: Orit Wasserman <[email protected]>

kvm: use directly cpu_physical_memory_* api for tracking dirty pages

Performance is important in this function, and we want to optimize even further.

Signed-off-by: Juan Quintela <[email protected]>
Reviewed-by: Orit Wasserman <[email protected]>

memory: unfold memory_region_test_and_clear()

We are going to update the bitmap directly

Signed-off-by: Juan Quintela <[email protected]>
Reviewed-by: Orit Wasserman <[email protected]>

memory: split cpu_physical_memory_* functions to its own include

All the functions that use ram_addr_t should be here.

Signed-off-by: Juan Quintela <[email protected]>
Reviewed-by: Orit Wasserman <[email protected]>

memory: cpu_physical_memory_set_dirty_tracking() should return void

Result was always 0, and not used anywhere.  Once there, use bool type
for the parameter.

Signed-off-by: Juan Quintela <[email protected]>
Reviewed-by: Eric Blake <[email protected]>
Reviewed-by: Orit Wasserman <[email protected]>

memory: make cpu_physical_memory_reset_dirty() take a length parameter

We have an end parameter in all the callers, and this make it coherent
with the rest of cpu_physical_memory_* functions, that also take a
length parameter.

Once here, move the start/end calculation to
tlb_reset_dirty_range_all() as we don't need it here anymore.

Signed-off-by: Juan Quintela <[email protected]>
Reviewed-by: Eric Blake <[email protected]>
Reviewed-by: Orit Wasserman <[email protected]>

memory: s/dirty/clean/ in cpu_physical_memory_is_dirty()

All uses except one really want the other meaning.

Signed-off-by: Juan Quintela <[email protected]>
Reviewed-by: Eric Blake <[email protected]>
Reviewed-by: Orit Wasserman <[email protected]>

memory: cpu_physical_memory_clear_dirty_range() now uses bitmap operations

We were clearing a range of bits, so use bitmap_clear().

Signed-off-by: Juan Quintela <[email protected]>
Reviewed-by: Eric Blake <[email protected]>
Reviewed-by: Orit Wasserman <[email protected]>

memory: cpu_physical_memory_set_dirty_range() now uses bitmap operations

We were setting a range of bits, so use bitmap_set().

Note: xen has always been wrong, and should have used start instead
of addr from the beginning.

Signed-off-by: Juan Quintela <[email protected]>
Reviewed-by: Eric Blake <[email protected]>
Reviewed-by: Orit Wasserman <[email protected]>

memory: use find_next_bit() to find dirty bits

This operation is way faster than doing it bit by bit.

Signed-off-by: Juan Quintela <[email protected]>
Reviewed-by: Eric Blake <[email protected]>
Reviewed-by: Orit Wasserman <[email protected]>

memory: s/mask/clear/ cpu_physical_memory_mask_dirty_range

Now all functions use the same wording that bitops/bitmap operations

Signed-off-by: Juan Quintela <[email protected]>
Reviewed-by: Eric Blake <[email protected]>
Reviewed-by: Orit Wasserman <[email protected]>

memory: cpu_physical_memory_get_dirty() is used as returning a bool

Signed-off-by: Juan Quintela <[email protected]>
Reviewed-by: Eric Blake <[email protected]>
Reviewed-by: Orit Wasserman <[email protected]>

memory: make cpu_physical_memory_get_dirty() the main function

And make cpu_physical_memory_get_dirty_flag() to use it.  It used to
be the other way around.

Signed-off-by: Juan Quintela <[email protected]>
Reviewed-by: Eric Blake <[email protected]>
Reviewed-by: Orit Wasserman <[email protected]>

memory: unfold cpu_physical_memory_set_dirty_flag()

Signed-off-by: Juan Quintela <[email protected]>
Reviewed-by: Eric Blake <[email protected]>
Reviewed-by: Orit Wasserman <[email protected]>

memory: unfold cpu_physical_memory_set_dirty() in its only user

Signed-off-by: Juan Quintela <[email protected]>
Reviewed-by: Eric Blake <[email protected]>
Reviewed-by: Orit Wasserman <[email protected]>

memory: unfold cpu_physical_memory_clear_dirty_flag() in its only user

Signed-off-by: Juan Quintela <[email protected]>
Reviewed-by: Eric Blake <[email protected]>
Reviewed-by: Orit Wasserman <[email protected]>

memory: split dirty bitmap into three

After all the previous patches, spliting the bitmap gets direct.

Note: For some reason, I have to move DIRTY_MEMORY_* definitions to
the beginning of memory.h to make compilation work.

Signed-off-by: Juan Quintela <[email protected]>
Reviewed-by: Eric Blake <[email protected]>
Reviewed-by: Orit Wasserman <[email protected]>

bitmap: Add bitmap_zero_extend operation

Signed-off-by: Juan Quintela <[email protected]>
Reviewed-by: Eric Blake <[email protected]>
Reviewed-by: Orit Wasserman <[email protected]>

memory: cpu_physical_memory_clear_dirty_flag() result is never used

Signed-off-by: Juan Quintela <[email protected]>
Reviewed-by: Eric Blake <[email protected]>
Reviewed-by: Orit Wasserman <[email protected]>

memory: only resize dirty bitmap when memory size increases

Signed-off-by: Juan Quintela <[email protected]>
Reviewed-by: Eric Blake <[email protected]>
Reviewed-by: Orit Wasserman <[email protected]>

memory: make sure that client is always inside range

Signed-off-by: Juan Quintela <[email protected]>
Reviewed-by: Eric Blake <[email protected]>
Reviewed-by: Orit Wasserman <[email protected]>

memory: use bit 2 for migration

For historical reasons it was bit 3.  Once there, create a constant to
know the number of clients.

Signed-off-by: Juan Quintela <[email protected]>
Reviewed-by: Eric Blake <[email protected]>
Reviewed-by: Orit Wasserman <[email protected]>

memory: cpu_physical_memory_mask_dirty_range() always clears a single flag

Document it

Signed-off-by: Juan Quintela <[email protected]>
Reviewed-by: Eric Blake <[email protected]>
Reviewed-by: Orit Wasserman <[email protected]>

memory: cpu_physical_memory_set_dirty_range() always dirty all flags

So remove the flag argument and do it directly.  After this change,
there is nothing else using cpu_physical_memory_set_dirty_flags() so
remove it.

Signed-off-by: Juan Quintela <[email protected]>
Reviewed-by: Eric Blake <[email protected]>
Reviewed-by: Orit Wasserman <[email protected]>

memory: set single dirty flags when possible

Signed-off-by: Juan Quintela <[email protected]>
Reviewed-by: Eric Blake <[email protected]>
Reviewed-by: Orit Wasserman <[email protected]>

memory: all users of cpu_physical_memory_get_dirty used only one flag

So cpu_physical_memory_get_dirty_flags is not needed anymore

Signed-off-by: Juan Quintela <[email protected]>
Reviewed-by: Eric Blake <[email protected]>
Reviewed-by: Orit Wasserman <[email protected]>

memory: make cpu_physical_memory_is_dirty return bool

Signed-off-by: Juan Quintela <[email protected]>
Reviewed-by: Eric Blake <[email protected]>
Reviewed-by: Orit Wasserman <[email protected]>

exec: create function to get a single dirty bit

Signed-off-by: Juan Quintela <[email protected]>
Reviewed-by: Eric Blake <[email protected]>
Reviewed-by: Orit Wasserman <[email protected]>

memory: create function to set a single dirty bit

Signed-off-by: Juan Quintela <[email protected]>
Reviewed-by: Orit Wasserman <[email protected]>
Reviewed-by: Eric Blake <[email protected]>

exec: use accessor function to know if memory is dirty

Signed-off-by: Juan Quintela <[email protected]>
Reviewed-by: Orit Wasserman <[email protected]>
Reviewed-by: Eric Blake <[email protected]>

memory: cpu_physical_memory_set_dirty_range() return void

Signed-off-by: Juan Quintela <[email protected]>
Reviewed-by: Orit Wasserman <[email protected]>
Reviewed-by: Eric Blake <[email protected]>

memory: cpu_physical_memory_set_dirty_flags() result is never used

So return void.

Signed-off-by: Juan Quintela <[email protected]>
Reviewed-by: Orit Wasserman <[email protected]>
Reviewed-by: Eric Blake <[email protected]>

bitmap: use long as index

Move index and size fields from int to long.  We need that for
migration.  long is 64 bits on sane architectures, and 32bits should
be enough on all the 32bits architectures.

Signed-off-by: Juan Quintela <[email protected]>
Reviewed-by: Eric Blake <[email protected]>
Reviewed-by: Orit Wasserman <[email protected]>

tests: Some unit tests for vmstate.c

 * Basic load/save tests
 * Tests for loading older versions
 * Tests for .field_exists() handling

Signed-off-by: Eduardo Habkost <[email protected]>
Signed-off-by: Juan Quintela <[email protected]>

savevm: Small comment about why timer QEMUFile/VMState code is in savevm.c

Signed-off-by: Eduardo Habkost <[email protected]>
Reviewed-by: Orit Wasserman <[email protected]>
Signed-off-by: Juan Quintela <[email protected]>

qemu-file: Move QEMUFile code to qemu-file.c

Signed-off-by: Eduardo Habkost <[email protected]>
Reviewed-by: Orit Wasserman <[email protected]>
Signed-off-by: Juan Quintela <[email protected]>

vmstate: Move VMState code to vmstate.c

This will allow unit tests to be written for VMState code without
pulling dependencies from the savevm code.

Signed-off-by: Eduardo Habkost <[email protected]>
Reviewed-by: Orit Wasserman <[email protected]>
Signed-off-by: Juan Quintela <[email protected]>

savevm.c: Coding style fix

Signed-off-by: Eduardo Habkost <[email protected]>
Reviewed-by: Orit Wasserman <[email protected]>
Signed-off-by: Juan Quintela <[email protected]>

savevm.c: Coding style fixes

Signed-off-by: Eduardo Habkost <[email protected]>
Reviewed-by: Orit Wasserman <[email protected]>
Signed-off-by: Juan Quintela <[email protected]>

savevm: Convert all tabs to spaces

Signed-off-by: Eduardo Habkost <[email protected]>
Reviewed-by: Orit Wasserman <[email protected]>
Signed-off-by: Juan Quintela <[email protected]>

migration: Move QEMU_VM_* defines to migration/migration.h

The VMState code will be moved to vmstate.c and it uses some of the
QEMU_VM_* constants, so move it to a header.

Signed-off-by: Eduardo Habkost <[email protected]>
Reviewed-by: Orit Wasserman <[email protected]>
Signed-off-by: Juan Quintela <[email protected]>

qemu-file: Make a few functions non-static

The QEMUFile code will be moved to qemu-file.c. This will require making
the following functions non-static because they are used by the savevm.c
code:

 * qemu_peek_byte()
 * qemu_peek_buffer()
 * qemu_file_skip()
 * qemu_file_set_error()

Signed-off-by: Eduardo Habkost <[email protected]>
Reviewed-by: Orit Wasserman <[email protected]>
Signed-off-by: Juan Quintela <[email protected]>

migration: Fix rate limit

The migration thread appears to want to allow writeout to occur at full
speed rather than being rate limited during completion of state saving,
but sets the limit to INT_MAX when xfer_limit is INT64_MAX. This causes
problems if there's more than 2GB of state left to save at this point. It
probably ought to just be INT64_MAX instead.

Signed-off-by: Matthew Garrett <[email protected]>
Reviewed-by: Paolo Bonzini <[email protected]>
Signed-off-by: Juan Quintela <[email protected]>

introduce MIG_STATE_CANCELLING state

Introduce MIG_STATE_CANCELLING state to avoid starting a new migration task while the previous one still exist.

Signed-off-by: Zeng Junliang <[email protected]>
Signed-off-by: Zhang Haoyu <[email protected]>
Reviewed-by: Paolo Bonzini <[email protected]>
Signed-off-by: Juan Quintela <[email protected]>

avoid a bogus COMPLETED->CANCELLED transition

Avoid a bogus COMPLETED->CANCELLED transition.
There is a period of time from the timing of setting COMPLETED state to that of migration thread exits, so during which it's problematic in COMPLETED->CANCELLED transition.

Signed-off-by: Zeng Junliang <[email protected]>
Signed-off-by: Zhang Haoyu <[email protected]>
Reviewed-by: Paolo Bonzini <[email protected]>
Signed-off-by: Juan Quintela <[email protected]>

Merge remote-tracking branch 'pmaydell/tags/pull-cocoa-20140112' into staging

cocoa queue:
 * pass command key to guest when VM has mousegrab
 * add .qcow2 to extension list for image load dialog
 * fix bugs in code for starting QEMU via image load dialog
 * fix resize/redraw interaction
 * draw window black if guest hasn't sent anything to screen
 * minor style/typo fixes
 * add myself as cocoa co-maintainer

# gpg: Signature made Sun 12 Jan 2014 02:45:52 PM PST using RSA key ID 14360CDE
# gpg: Can't check signature: public key not found

* pmaydell/tags/pull-cocoa-20140112:
  MAINTAINERS: add myself as cocoa UI co-maintainer
  ui/cocoa: Remove stray tabs
  ui/cocoa: Draw black rectangle if we have no data yet
  ui/cocoa: Redraw at correct size when switching surface
  ui/cocoa: Fix code for starting QEMU via image file load dialog
  ui/cocoa: Add ".qcow2" to extension list for image load dialog
  ui/cocoa: Send warning message to stderr, not stdout
  ui/cocoa: Correct typos in comments and variable names
  ui/cocoa: Pass command key through to guest when VM has mousegrab

Message-id: 1389567158[email protected]
Signed-off-by: Anthony Liguori <[email protected]>

Merge remote-tracking branch 'pmaydell/tags/pull-target-arm-20140112' into staging

target-arm queue:
 * build fix for bigendian hosts

# gpg: Signature made Sun 12 Jan 2014 01:38:22 PM PST using RSA key ID 14360CDE
# gpg: Can't check signature: public key not found

* pmaydell/tags/pull-target-arm-20140112:
  arm: fix compile on bigendian host

Message-id: 1389562970[email protected]
Signed-off-by: Anthony Liguori <[email protected]>

MAINTAINERS: add myself as cocoa UI co-maintainer

Add myself to the maintainers list for the cocoa UI; status
remains "Odd Fixes".

Signed-off-by: Peter Maydell <[email protected]>
Acked-by: Andreas Färber <[email protected]>
Message-id: 1387207075[email protected]

ui/cocoa: Remove stray tabs

The ui/cocoa.m file has just three lines with hardcoded tabs; fix them.

Signed-off-by: Peter Maydell <[email protected]>
Reviewed-by: Stefan Weil <[email protected]>
Message-id: 1387886052[email protected]

ui/cocoa: Draw black rectangle if we have no data yet

If our redraw method is called before we have any data from the guest,
then draw a black rectangle rather than leaving the window empty.
This mostly only matters when the guest machine has no framebuffer
device, but it is more in line with the behaviour of other QEMU UIs.

Signed-off-by: Peter Maydell <[email protected]>
Message-id: 1387853507[email protected]

ui/cocoa: Redraw at correct size when switching surface

If the surface switch involved a resize, we were doing the redraw
at the old size rather than the new, because the update of
screen.width and screen.height was being done after the setFrame
method calls which triggered a redraw. Normally this isn't very
noticeable because typically after the guest triggers the window
resize it also draws something to it, which will in turn cause
us to redraw. However, the combination of a guest which never
draws to the display and a command line setting of a screen size
larger than the default can reveal odd effects.

Move most of the handling of resizes to the top of the method,
and guard it with a check that the surface size actually changed,
to avoid unnecessary operations (including some user visible ones
like "recenter the window on the screen") if the surface is the
same size as the old one.

Signed-off-by: Peter Maydell <[email protected]>
Message-id: 1387853507[email protected]

ui/cocoa: Fix code for starting QEMU via image file load dialog

Fix a number of bugs in the code for starting QEMU via the image
file load dialog:
 * use the actual argv[0] rather than "qemu": this avoids failures to
   find BIOS image files caused by not looking in the correct directory
   relative to the executable path
 * allocate a large enough argv array to NULL terminate it
 * use g_strdup(X) rather than g_strdup_printf("%s", X) or
   g_strdup_printf(X)
 * disable the printing of the simulated command line argument
   (which is presumably intended for debug only)

Signed-off-by: Peter Maydell <[email protected]>
Message-id: 1386543546[email protected]

ui/cocoa: Add ".qcow2" to extension list for image load dialog

Add ".qcow2" to the list of file extensions which are accepted
by the initial disk image load dialog which is displayed if the
user runs QEMU without any command line arguments.

Signed-off-by: Peter Maydell <[email protected]>
Message-id: 1386543546[email protected]

ui/cocoa: Send warning message to stderr, not stdout

Bring a warning message into line with the others in this file by
sending it to stderr, not stdout.

Signed-off-by: Peter Maydell <[email protected]>
Message-id: 1386543546[email protected]

ui/cocoa: Correct typos in comments and variable names

Fix various non-user-visible typos in comments and variable names.

Signed-off-by: Peter Maydell <[email protected]>
Reviewed-by: Stefan Weil <[email protected]>
Message-id: 1386543546[email protected]

ui/cocoa: Pass command key through to guest when VM has mousegrab

The guest might want to be able to use the command key for its won
purposes (as command if it is MacOS X, or for the Windows key if
it is a PC guest, for instance). In line with other UI frontends,
pass it through if the guest has mousegrab, and only use it for UI
menu accelerators if not grabbed.

Thanks to John Arbuckle for reporting this problem, helping
us work through what the best solution would be and providing
a patch which was the initial inspiration for this one.

Reported-by: John Arbuckle <[email protected]>
Signed-off-by: Peter Maydell <[email protected]>
Message-id: 1386543546[email protected]

arm: fix compile on bigendian host

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

Merge remote-tracking branch 'bonzini/scsi-next' into staging

* bonzini/scsi-next:
  scsi-disk: add UNMAP limits to block limits VPD page
  block/iscsi: use a bh to schedule co reentrance

Message-id: 1387720926[email protected]
Signed-off-by: Anthony Liguori <[email protected]>

Merge remote-tracking branch 'stefanha/block' into staging

* stefanha/block:
  commit: Remove unused check
  qemu-iotests: Update test cases for commit active
  commit: Support commit active layer
  block: Add commit_active_start()
  mirror: Move base to MirrorBlockJob
  mirror: Don't close target
  qemu-iotests: drop duplicate virtio-blk initialization failure
  vmdk: Allow vmdk_create to work with protocol
  vmdk: Check VMFS extent line field number
  docs: updated qemu-img man page and qemu-doc to reflect VHDX support.
  block: vhdx - improve error message, and .bdrv_check implementation
  block/iscsi: Fix compilation for libiscsi 1.4.0 (API change)
  qapi-schema: fix QEMU 1.8 references
  dataplane: replace hostmem with memory_region_find
  dataplane: change vring API to use VirtQueueElement
  vring: factor common code for error exits
  vring: create a common function to parse descriptors
  sheepdog: fix dynamic grow for running qcow2 format

Message-id: 1387554416[email protected]
Signed-off-by: Anthony Liguori <[email protected]>

Merge remote-tracking branch 'mst/tags/for_anthony' into staging

acpi,pci,pc,fedora,virtio fixes and enhancements

This includes some Preparatory patches for cpu hotplug for q25 and memory
hotplug by Igor, tests and memory mapping change
by Laszlo and pci reset cleanup by Paolo.
There are also some fixes for fedora and virtio:
included here since they are test blockers for me.

Signed-off-by: Michael S. Tsirkin <[email protected]>
# gpg: Signature made Mon 23 Dec 2013 08:07:18 AM PST using RSA key ID D28D5469
# gpg: Can't check signature: public key not found

* mst/tags/for_anthony:
  target-arm: fix build with gcc 4.8.2
  virtio: add back call to virtio_bus_device_unplugged
  piix: fix 32bit pci hole
  qdev: switch reset to post-order
  qdev: allow both pre- and post-order vists in qdev walking functions
  pci: clean up resetting of IRQs
  pci: do not export pci_bus_reset
  ACPI/DSDT-CPU: cleanup bogus comment
  ACPI: Q35 DSDT: fix CPU hotplug GPE0.2 handler
  acpi: ich9: allow guest to clear SCI rised by GPE
  acpi: factor out common pm_update_sci() into acpi core
  acpi: piix4: remove not needed GPE0 mask
  i440fx-test: verify firmware under 4G and 1M, both -bios and -pflash
  i440fx-test: generate temporary firmware blob
  i440fx-test: give each GTest case its own qtest
  i440fx-test: qtest_start() should be paired with qtest_end()
  hw/i386/pc_sysfw: support two flash drives
  pc_piix: document gigabyte_align
  piix: gigabyte alignment for ram

Message-id: 1387815007[email protected]
Signed-off-by: Anthony Liguori <[email protected]>

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

QOM CPUState refactorings / X86CPU

* TLB invalidation optimizations
* X86CPU initialization cleanups
* Preparations for X86CPU hot-unplug

# gpg: Signature made Tue 24 Dec 2013 04:51:52 AM PST using RSA key ID 3E7E013F
# gpg: Good signature from "Andreas Färber <[email protected]>"
# gpg:                 aka "Andreas Färber <[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: 174F 0347 1BCC 221A 6175  6F96 FA2E D12D 3E7E 013F

* afaerber/tags/qom-cpu-for-anthony:
  target-i386: Cleanup 'foo=val' feature handling
  target-i386: Cleanup 'foo' feature handling
  target-i386: Convert 'check' and 'enforce' to static properties
  target-i386: Convert 'hv_spinlocks' to static property
  target-i386: Convert 'hv_vapic' to static property
  target-i386: Convert 'hv_relaxed' to static property
  cpu-exec: Optimize X86CPU usage in cpu_exec()
  target-i386: Move apic_state field from CPUX86State to X86CPU
  cputlb: Tidy memset() of arrays
  cputlb: Use memset() when flushing entries

Merge remote-tracking branch 'pmaydell/tags/pull-target-arm-20140107' into staging

target-arm queue:
 * further A64 decoder patches, including enabling the aarch64-linux-user
   target; this includes full floating point support. Neon is not yet
   supported.
 * cadence UART model fixes.
 * some minor bug fixes and cleanups.
 * all the softfloat fixes required by the new A64 instructions;
   several of these will also be used by PPC.

* pmaydell/tags/pull-target-arm-20140107: (61 commits)
  target-arm: A64: Add support for FCVT between half, single and double
  target-arm: A64: Add 1-source 32-to-32 and 64-to-64 FP instructions
  target-arm: A64: Add floating-point<->integer conversion instructions
  target-arm: A64: Add floating-point<->fixed-point instructions
  target-arm: A64: Add extra VFP fixed point conversion helpers
  target-arm: Ignore most exceptions from scalbn when doing fixpoint conversion
  target-arm: Rename A32 VFP conversion helpers
  target-arm: Prepare VFP_CONV_FIX helpers for A64 uses
  softfloat: Add support for ties-away rounding
  softfloat: Refactor code handling various rounding modes
  softfloat: Add float16 <=> float64 conversion functions
  softfloat: Factor out RoundAndPackFloat16 and NormalizeFloat16Subnormal
  softfloat: Provide complete set of accessors for fp state
  softfloat: Fix float64_to_uint32_round_to_zero
  softfloat: Fix float64_to_uint32
  softfloat: Fix float64_to_uint64_round_to_zero
  softfloat: Add float32_to_uint64()
  softfloat: Fix factor 2 error for scalbn on denormal inputs
  softfloat: Only raise Invalid when conversions to int are out of range
  softfloat: Fix float64_to_uint64
  ...

Conflicts:
target-arm/cpu.h

aliguori: resolved trivial conflict

Signed-off-by: Anthony Liguori <[email protected]>

Merge remote-tracking branch 'riku/linux-user-for-upstream' into staging

* riku/linux-user-for-upstream:
  linux-user: Remove regs parameter of load_elf_binary and load_flt_binary
  linux-user: Support the accept4 socketcall

Message-id: 1389364137[email protected]
Signed-off-by: Anthony Liguori <[email protected]>

linux-user: Remove regs parameter of load_elf_binary and load_flt_binary

The regs parameter is not used anywhere, so remove it.

Signed-off-by: Will Newton <[email protected]>
Reviewed-by: Erik de Castro Lopo <[email protected]>
Reviewed-by: Andreas Färber <[email protected]>
Signed-off-by: Riku Voipio <[email protected]>

linux-user: Support the accept4 socketcall

Cc: Riku Voipio <[email protected]>
Signed-off-by: André Hentschel <[email protected]>
Reviewed-by: Peter Maydell <[email protected]>
Reviewed-by: Erik de Castro Lopo <[email protected]>
Reviewed-by: Laurent Vivier <[email protected]>
Signed-off-by: Riku Voipio <[email protected]>

Merge remote-tracking branch 'otubo/seccomp' into staging

* otubo/seccomp:
  seccomp: exit if seccomp_init() fails

Message-id: 1387565447[email protected]
Signed-off-by: Anthony Liguori <[email protected]>

Merge remote-tracking branch 'rth/ldst-i386-2' into staging

* rth/ldst-i386-2: (49 commits)
  target-i386: Tidy ljmp
  target-i386: Rename gen_op_jmp_T0 to gen_op_jmp_v
  target-i386: Tidy some size computation
  target-i386: Remove gen_op_mov_reg_A0
  target-i386: Remove gen_op_mov_TN_reg
  target-i386: Remove gen_op_addl_T0_T1
  target-i386: Remove gen_op_mov_reg_T1
  target-i386: Remove gen_op_mov_reg_T0
  target-i386: Tidy cpu_regs initialization
  target_i386: Clean up gen_pop_T0
  target-i386: Combine gen_push_T* into gen_push_v
  target-i386: Tidy addr16 code in gen_lea_modrm
  target-i386: Change dflag to TCGMemOp
  target-i386: Change gen_op_mov_reg_A0 size parameter to TCGMemOp
  target-i386: Change aflag to TCGMemOp
  target-i386: Change gen_op_j*z_ecx size parameter to TCGMemOp
  target-i386: Change gen_op_add_reg_* size parameter to TCGMemOp
  target-i386: Use TCGMemOp for 'ot' variables
  target-i386: Remove gen_op_andl_A0_ffff
  target-i386: Remove gen_op_movl_T0_T1
  ...

Message-id: 1389128439[email protected]
Signed-off-by: Anthony Liguori <[email protected]>

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

QOM infrastructure fixes and device conversions

* QOM interface fixes and unit test
* Device no_user sanitization and documentation
* Device error reporting improvement
* Conversion of APIC, ICC, IOAPIC to QOM realization model

# gpg: Signature made Tue 24 Dec 2013 09:04:05 AM PST using RSA key ID 3E7E013F
# gpg: Good signature from "Andreas Färber <[email protected]>"
# gpg:                 aka "Andreas Färber <[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: 174F 0347 1BCC 221A 6175  6F96 FA2E D12D 3E7E 013F

* afaerber/tags/qom-devices-for-anthony: (24 commits)
  qdev-monitor: Improve error message for -device nonexistant
  ioapic: QOM'ify ioapic
  ioapic: Cleanup for QOM'ification
  icc_bus: QOM'ify ICC
  apic: QOM'ify APIC
  apic: Cleanup for QOM'ification
  qdev: Drop misleading qbus_free() function
  qom: Detect bad reentrance during object_class_foreach()
  tests: Test QOM interface casting
  qom: Do not register interface "types" in the type table and fix names
  qom: Split out object and class caches
  qdev: Document that pointer properties kill device_add
  hw: cannot_instantiate_with_device_add_yet due to pointer props
  qdev-monitor: Avoid device_add crashing on non-device driver name
  qdev: Do not let the user try to device_add when it cannot work
  isa: Clean up use of cannot_instantiate_with_device_add_yet
  vt82c686: Clean up use of cannot_instantiate_with_device_add_yet
  piix3 piix4: Clean up use of cannot_instantiate_with_device_add_yet
  ich9: Document why cannot_instantiate_with_device_add_yet
  pci-host: Consistently set cannot_instantiate_with_device_add_yet
  ...

Merge remote-tracking branch 'mjt/trivial-patches' into staging

* mjt/trivial-patches:
  acpi unit-test: Remove temporary disk after test
  mainstone: Fix duplicate array values for key 'space'
  pxa27x: Add 'const' attribute to keyboard maps
  pxa27x: Reduce size of keyboard matrix mapping
  doc: Mention chardev:id in available devices for -serial
  configure: Python tests must be done before help message
  configure: Rewrite code for help message
  fix -boot strict regressed in commit 6ef4716
  vl: make boot_strict variable static (not used outside vl.c)
  x86: only allow real mode to access 32bit without LMA
  linux-user: Use macro TARGET_NSIG_WORDS where possible
  exynos4210: Use macro ARRAY_SIZE where possible
  ui/cocoa: Use macro ARRAY_SIZE where possible
  misc: Use macro ARRAY_SIZE where possible
  openrisc: Fix spelling in comment (transaltion -> translation)
  hw/arm/highbank: Simplify code (memory region in device state)

Message-id: 1388182050[email protected]
Signed-off-by: Anthony Liguori <[email protected]>

Merge remote-tracking branch 'pmaydell/tags/pull-target-arm-20140108' into staging

target-arm queue:
 * further A64 decoder patches, including enabling the aarch64-linux-user
   target; this includes full floating point support. Neon is not yet
   supported.
 * cadence UART model fixes.
 * some minor bug fixes and cleanups.
 * all the softfloat fixes required by the new A64 instructions;
   several of these will also be used by PPC.

# gpg: Signature made Wed 08 Jan 2014 11:25:12 AM PST using RSA key ID 14360CDE
# gpg: Can't check signature: public key not found

* pmaydell/tags/pull-target-arm-20140108: (76 commits)
  target-arm: A64: Add support for FCVT between half, single and double
  target-arm: A64: Add 1-source 32-to-32 and 64-to-64 FP instructions
  target-arm: A64: Add floating-point<->integer conversion instructions
  target-arm: A64: Add floating-point<->fixed-point instructions
  target-arm: A64: Add extra VFP fixed point conversion helpers
  target-arm: Ignore most exceptions from scalbn when doing fixpoint conversion
  target-arm: Rename A32 VFP conversion helpers
  target-arm: Prepare VFP_CONV_FIX helpers for A64 uses
  softfloat: Add support for ties-away rounding
  softfloat: Refactor code handling various rounding modes
  softfloat: Add float16 <=> float64 conversion functions
  softfloat: Factor out RoundAndPackFloat16 and NormalizeFloat16Subnormal
  softfloat: Provide complete set of accessors for fp state
  softfloat: Fix float64_to_uint32_round_to_zero
  softfloat: Fix float64_to_uint32
  softfloat: Fix float64_to_uint64_round_to_zero
  softfloat: Add float32_to_uint64()
  softfloat: Fix factor 2 error for scalbn on denormal inputs
  softfloat: Only raise Invalid when conversions to int are out of range
  softfloat: Fix float64_to_uint64
  ...

Message-id: 1389209439[email protected]
Signed-off-by: Anthony Liguori <[email protected]>

target-arm: A64: Add support for FCVT between half, single and double

Add support for FCVT between half, single and double precision.

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

target-arm: A64: Add 1-source 32-to-32 and 64-to-64 FP instructions

This patch adds support for those instructions in the  "Floating-point
data-processing (1 source)" group which are simple 32-bit-to-32-bit
or 64-bit-to-64-bit operations (ie everything except FCVT between
single/double/half precision).

We put the new round-to-int helpers in helper.c because they will
also be used by the new ARMv8 A32/T32 rounding instructions.

Signed-off-by: Alexander Graf <[email protected]>
[WN: Commit message tweak, merged single and double precision patches,
 updated to new infrastructure.]
Signed-off-by: Will Newton <[email protected]>
[PMM: reworked decode, split FCVT out into their own patch]
Signed-off-by: Peter Maydell <[email protected]>
Reviewed-by: Richard Henderson <[email protected]>

target-arm: A64: Add floating-point<->integer conversion instructions

Add support for the AArch64 floating-point <-> integer conversion
instructions to disas_fpintconv. In the process we can rearrange
and simplify the detection of unallocated encodings a little.
We also correct a typo in the instruction encoding diagram for this
instruction group: bit 21 is 1, not 0.

Signed-off-by: Will Newton <[email protected]>
Signed-off-by: Peter Maydell <[email protected]>
Reviewed-by: Richard Henderson <[email protected]>

target-arm: A64: Add floating-point<->fixed-point instructions

This patch adds emulation for the instruction group labeled
"Floating-point <-> fixed-point conversions" in the ARM ARM.

Namely this includes the instructions SCVTF, UCVTF, FCVTZS, FCVTZU
(scalar, fixed-point).

Signed-off-by: Alexander Graf <[email protected]>
[WN: Commit message tweak, rebased, updated to new infrastructure.
 Applied bug fixes from Michael Matz and Janne Grunau.]
Signed-off-by: Will Newton <[email protected]>
[PMM: significant cleanup]
Signed-off-by: Peter Maydell <[email protected]>
Reviewed-by: Richard Henderson <[email protected]>

target-arm: A64: Add extra VFP fixed point conversion helpers

Define the full set of floating point to fixed point conversion
helpers required to support AArch64.

Signed-off-by: Will Newton <[email protected]>
Signed-off-by: Peter Maydell <[email protected]>
Reviewed-by: Richard Henderson <[email protected]>

target-arm: Ignore most exceptions from scalbn when doing fixpoint conversion

The VFP fixed point conversion helpers first call float_scalbn and
then convert the result to an integer. This scalbn operation may
set floating point exception flags for:
 * overflow & inexact (if it overflows to infinity)
 * input denormal squashed to zero
 * output denormal squashed to zero
Of these, we only care about the input-denormal flag, since
the output of the whole scale-and-convert operation will be
an integer (so squashed-output-denormal and overflow don't
apply). Suppress the others by saving the pre-scalb exception
flags and only copying across a potential input-denormal flag.

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

target-arm: Rename A32 VFP conversion helpers

The VFP conversion helpers for A32 round to zero as this is the only
rounding mode supported. Rename these helpers to make it clear that
they round to zero and are not suitable for use in the AArch64 code.

Signed-off-by: Will Newton <[email protected]>
Signed-off-by: Peter Maydell <[email protected]>
Reviewed-by: Richard Henderson <[email protected]>

target-arm: Prepare VFP_CONV_FIX helpers for A64 uses

Make the VFP_CONV_FIX helpers a little more flexible in
preparation for the A64 uses. This requires two changes:
 * use the correct softfloat conversion function based on itype
   rather than always the int32 one; this is possible now that
   softfloat provides int16 versions and necessary for the
   future conversion-to-int64 A64 variants. This also allows
   us to drop the awkward 'sign' macro argument.
 * split the 'fsz' argument which currently controls both
   width of the input float type and width of the output
   integer type into two; this will allow us to specify the
   A64 64-bit-int-to-single conversion function, where the
   two widths are different.

We can also drop the (itype##_t) cast now that softfloat
guarantees that all the itype##_to_float* functions take
an integer argument of exactly the correct type.

Signed-off-by: Will Newton <[email protected]>
Signed-off-by: Peter Maydell <[email protected]>
Reviewed-by: Richard Henderson <[email protected]>

softfloat: Add support for ties-away rounding

IEEE754-2008 specifies a new rounding mode:

"roundTiesToAway: the floating-point number nearest to the infinitely
precise result shall be delivered; if the two nearest floating-point
numbers bracketing an unrepresentable infinitely precise result are
equally near, the one with larger magnitude shall be delivered."

Implement this new mode (it is needed for ARM). The general principle
is that the required code is exactly like the ties-to-even code,
except that we do not need to do the "in case of exact tie clear LSB
to round-to-even", because the rounding operation naturally causes
the exact tie to round up in magnitude.

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

softfloat: Refactor code handling various rounding modes

Refactor the code in various functions which calculates rounding
increments given the current rounding mode, so that instead of a
set of nested if statements we have a simple switch statement.
This will give us a clean place to add the case for the new
tiesAway rounding mode.

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

softfloat: Add float16 <=> float64 conversion functions

Add the conversion functions float16_to_float64() and
float64_to_float16(), which will be needed for the ARM
A64 instruction set.

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

softfloat: Factor out RoundAndPackFloat16 and NormalizeFloat16Subnormal

In preparation for adding conversions between float16 and float64,
factor out code currently done inline in the float16<=>float32
conversion functions into functions RoundAndPackFloat16 and
NormalizeFloat16Subnormal along the lines of the existing versions
for the other float types.

Note that we change the handling of zExp from the inline code
to match the API of the other RoundAndPackFloat functions; however
we leave the positioning of the binary point between bits 22 and 23
rather than shifting it up to the high end of the word.

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

softfloat: Provide complete set of accessors for fp state

Tidy up the get/set accessors for the fp state to add missing ones
and make them all inline in softfloat.h rather than some inline and
some not.

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

softfloat: Fix float64_to_uint32_round_to_zero

The float64_to_uint32_round_to_zero routine is incorrect.

For example, the following test pattern:

    425F81378DC0CD1F / 0x1.f81378dc0cd1fp+38

will erroneously set the inexact flag.

This patch re-implements the routine to use the float64_to_uint64_round_to_zero
routine.  If saturation occurs we ignore any flags set by the
conversion function and raise only Invalid.

This contribution can be licensed under either the softfloat-2a or -2b
license.

Signed-off-by: Tom Musta <[email protected]>
Message-id: 1387397961[email protected]
Signed-off-by: Peter Maydell <[email protected]>
Reviewed-by: Richard Henderson <[email protected]>

softfloat: Fix float64_to_uint32

The float64_to_uint32 has several flaws:

 - for numbers between 2**32 and 2**64, the inexact exception flag
   may get incorrectly set.  In this case, only the invalid flag
   should be set.

       test pattern: 425F81378DC0CD1F / 0x1.f81378dc0cd1fp+38

 - for numbers between 2**63 and 2**64, incorrect results may
   be produced:

       test pattern: 43EAAF73F1F0B8BD / 0x1.aaf73f1f0b8bdp+63

This patch re-implements float64_to_uint32 to re-use the
float64_to_uint64 routine (instead of float64_to_int64).  For the
saturation case, we ignore any flags which the conversion routine
has set and raise only the invalid flag.

This contribution can be licensed under either the softfloat-2a or -2b
license.

Signed-off-by: Tom Musta <[email protected]>
Message-id: 1387397961[email protected]
Signed-off-by: Peter Maydell <[email protected]>
Reviewed-by: Richard Henderson <[email protected]>

softfloat: Fix float64_to_uint64_round_to_zero

The float64_to_uint64_round_to_zero routine is incorrect.

For example, the following test pattern:

    46697351FF4AEC29 / 0x1.97351ff4aec29p+103

currently produces 8000000000000000 instead of FFFFFFFFFFFFFFFF.

This patch re-implements the routine to temporarily force the
rounding mode and use the float64_to_uint64 routine.

This contribution can be licensed under either the softfloat-2a or -2b
license.

Signed-off-by: Tom Musta <[email protected]>
Message-id: 1387397961[email protected]
Signed-off-by: Peter Maydell <[email protected]>
Reviewed-by: Richard Henderson <[email protected]>

softfloat: Add float32_to_uint64()

This patch adds the float32_to_uint64() routine, which converts a
32-bit floating point number to an unsigned 64 bit number.

This contribution can be licensed under either the softfloat-2a or -2b
license.

Signed-off-by: Tom Musta <[email protected]>
Reviewed-by: Peter Maydell <[email protected]>
[PMM: removed harmless but silly int64_t casts]
Signed-off-by: Peter Maydell <[email protected]>
Reviewed-by: Richard Henderson <[email protected]>

softfloat: Fix factor 2 error for scalbn on denormal inputs

If the input to float*_scalbn() is denormal then it represents
a number 0.[mantissabits] * 2^(1-exponentbias) (and the actual
exponent field is all zeroes). This means that when we convert
it to our unpacked encoding the unpacked exponent must be one
greater than for a normal number, which represents
1.[mantissabits] * 2^(e-exponentbias) for an exponent field e.

This meant we were giving answers too small by a factor of 2 for
all denormal inputs.

Note that the float-to-int routines also have this behaviour
of not adjusting the exponent for denormals; however there it is
harmless because denormals will all convert to integer zero anyway.

Signed-off-by: Peter Maydell <[email protected]>
Reviewed-by: Aurelien Jarno <[email protected]>
Reviewed-by: Richard Henderson <[email protected]>

softfloat: Only raise Invalid when conversions to int are out of range

We implement a number of float-to-integer conversions using conversion
to an integer type with a wider range and then a check against the
narrower range we are actually converting to. If we find the result to
be out of range we correctly raise the Invalid exception, but we must
also suppress other exceptions which might have been raised by the
conversion function we called.

This won't throw away exceptions we should have preserved, because for
the 'core' exception flags the IEEE spec mandates that the only valid
combinations of exception that can be raised by a single operation are
Inexact + Overflow and Inexact + Underflow. For the non-IEEE softfloat
flag for input denormals, we can guarantee that that flag won't have
been set for out of range float-to-int conversions because a squashed
denormal by definition goes to plus or minus zero, which is always in
range after conversion to integer zero.

This bug has been fixed for some of the float-to-int conversion routines
by previous patches; fix it for the remaining functions as well, so
that they all restore the pre-conversion status flags prior to raising
Invalid.

Signed-off-by: Peter Maydell <[email protected]>
Reviewed-by: Aurelien Jarno <[email protected]>
Reviewed-by: Richard Henderson <[email protected]>

softfloat: Fix float64_to_uint64

The comment preceding the float64_to_uint64 routine suggests that
the implementation is broken.  And this is, indeed, the case.

This patch properly implements the conversion of a 64-bit floating
point number to an unsigned, 64 bit integer.

This contribution can be licensed under either the softfloat-2a or -2b
license.

Signed-off-by: Tom Musta <[email protected]>
Reviewed-by: Peter Maydell <[email protected]>
Signed-off-by: Peter Maydell <[email protected]>
Reviewed-by: Richard Henderson <[email protected]>

softfloat: Make the int-to-float functions take exact-width types

Currently the int-to-float functions take types which are specified
as "at least X bits wide", rather than "exactly X bits wide". This is
confusing and unhelpful since it means that the callers have to include
an explicit cast to [u]intXX_t to ensure the correct behaviour. Fix
them all to take the exactly-X-bits-wide types instead.

Note that this doesn't change behaviour at all since at the moment
we happen to define the 'int32' and 'uint32' types as exactly 32 bits
wide, and the 'int64' and 'uint64' types as exactly 64 bits wide.

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