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

Block patches

# gpg: Signature made Tue Apr 28 15:35:05 2015 BST using RSA key ID C88F2FD6
# gpg: Good signature from "Kevin Wolf <kwolf@redhat.com>"

* remotes/kevin/tags/for-upstream: (76 commits)
  block: move I/O request processing to block/io.c
  block: extract bdrv_setup_io_funcs()
  block: add bdrv_set_dirty()/bdrv_reset_dirty() to block_int.h
  block: replace bdrv_states iteration with bdrv_next()
  vmdk: Widen before shifting 32 bit header field
  block/dmg: make it modular
  block/mirror: Always call block_job_sleep_ns()
  iotests: add incremental backup granularity tests
  iotests: add incremental backup failure recovery test
  iotests: add simple incremental backup case
  iotests: add QMP event waiting queue
  iotests: add invalid input incremental backup tests
  hbitmap: truncate tests
  block: Resize bitmaps on bdrv_truncate
  block: Ensure consistent bitmap function prototypes
  block: add BdrvDirtyBitmap documentation
  qmp: Add dirty bitmap status field in query-block
  qmp: add block-dirty-bitmap-clear
  qmp: Add support of "dirty-bitmap" sync mode for drive-backup
  block: Add bitmap successors
  ...

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>

block: move I/O request processing to block/io.c

The block.c file has grown to over 6000 lines.  It is time to split this
file so there are fewer conflicts and the code is easier to maintain.

Extract I/O request processing code:
 * Read
 * Write
 * Zero writes and making the image empty
 * Flush
 * Discard
 * ioctl
 * Tracked requests and queuing
 * Throttling and copy-on-read
 * Block status and allocated functions
 * Refreshing block limits
 * Reading/writing vmstate
 * qemu_blockalign() and friends

The patch simply moves code from block.c into block/io.c.

Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block: extract bdrv_setup_io_funcs()

Move the code to install coroutine and aio emulation function pointers
in a BlockDriver to its own function.

Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block: add bdrv_set_dirty()/bdrv_reset_dirty() to block_int.h

The dirty bitmap functions are called from the block I/O processing
code.  Make them visible to block_int.h users so they can be used
outside block.c.

Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block: replace bdrv_states iteration with bdrv_next()

The bdrv_states list is a static variable in block.c.

bdrv_drain_all() and bdrv_flush_all() use this variable to iterate over
all drives.

The next patch will move bdrv_drain_all() and bdrv_flush_all() out of
block.c so it's necessary to switch to the public bdrv_next() interface.

Reviewed-by: Alberto Garcia <berto@igalia.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

vmdk: Widen before shifting 32 bit header field

Coverity spotted this.

The field is 32 bits, but if it's possible to overflow in 32 bit
left shift.

Signed-off-by: Fam Zheng <famz@redhat.com>
Reviewed-by: John Snow <jsnow@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block/dmg: make it modular

dmg can optionally utilize libbz2, make it modular

Signed-off-by: Michael Tokarev <mjt@tls.msk.ru>
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block/mirror: Always call block_job_sleep_ns()

The mirror block job is trying to take a clever shortcut if delay_ns is
0 and skips block_job_sleep_ns() in that case. But that function must be
called in every block job iteration, because otherwise it is for example
impossible to pause the job.

Signed-off-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Fam Zheng <famz@redhat.com>
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

iotests: add incremental backup granularity tests

Test what happens if you fiddle with the granularity.

Reviewed-by: Max Reitz <mreitz@redhat.com>
Signed-off-by: John Snow <jsnow@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Message-id: 1429314609-29776-22-git-send-email-jsnow@redhat.com
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

iotests: add incremental backup failure recovery test

Test the failure case for incremental backups.

Signed-off-by: John Snow <jsnow@redhat.com>
Reviewed-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Message-id: 1429314609-29776-21-git-send-email-jsnow@redhat.com
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

iotests: add simple incremental backup case

Signed-off-by: John Snow <jsnow@redhat.com>
Reviewed-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Message-id: 1429314609-29776-20-git-send-email-jsnow@redhat.com
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

iotests: add QMP event waiting queue

A filter is added to allow callers to request very specific
events to be pulled from the event queue, while leaving undesired
events still in the stream.

This allows us to poll for completion data for multiple asynchronous
events in any arbitrary order.

A new timeout context is added to the qmp pull_event method's
wait parameter to allow tests to fail if they do not complete
within some expected period of time.

Also fixed is a bug in qmp.pull_event where we try to retrieve an event
from an empty list if we attempt to retrieve an event with wait=False
but no events have occurred.

Signed-off-by: John Snow <jsnow@redhat.com>
Reviewed-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Message-id: 1429314609-29776-19-git-send-email-jsnow@redhat.com
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

iotests: add invalid input incremental backup tests

Signed-off-by: John Snow <jsnow@redhat.com>
Reviewed-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Message-id: 1429314609-29776-18-git-send-email-jsnow@redhat.com
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

hbitmap: truncate tests

The general approach is to set bits close to the boundaries of
where we are truncating and ensure that everything appears to
have gone OK.

We test growing and shrinking by different amounts:
- Less than the granularity
- Less than the granularity, but across a boundary
- Less than sizeof(unsigned long)
- Less than sizeof(unsigned long), but across a ulong boundary
- More than sizeof(unsigned long)

Signed-off-by: John Snow <jsnow@redhat.com>
Reviewed-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Message-id: 1429314609-29776-17-git-send-email-jsnow@redhat.com
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block: Resize bitmaps on bdrv_truncate

Signed-off-by: John Snow <jsnow@redhat.com>
Reviewed-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Message-id: 1429314609-29776-16-git-send-email-jsnow@redhat.com
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block: Ensure consistent bitmap function prototypes

We often don't need the BlockDriverState for functions
that operate on bitmaps. Remove it.

Signed-off-by: John Snow <jsnow@redhat.com>
Reviewed-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Message-id: 1429314609-29776-15-git-send-email-jsnow@redhat.com
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block: add BdrvDirtyBitmap documentation

Signed-off-by: John Snow <jsnow@redhat.com>
Reviewed-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Message-id: 1429314609-29776-14-git-send-email-jsnow@redhat.com
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

qmp: Add dirty bitmap status field in query-block

Add the "frozen" status booleans, to inform clients
when a bitmap is occupied doing a task.

Signed-off-by: Fam Zheng <famz@redhat.com>
Signed-off-by: John Snow <jsnow@redhat.com>
Reviewed-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Message-id: 1429314609-29776-13-git-send-email-jsnow@redhat.com
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

qmp: add block-dirty-bitmap-clear

Add bdrv_clear_dirty_bitmap and a matching QMP command,
qmp_block_dirty_bitmap_clear that enables a user to reset
the bitmap attached to a drive.

This allows us to reset a bitmap in the event of a full
drive backup.

Signed-off-by: John Snow <jsnow@redhat.com>
Reviewed-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Message-id: 1429314609-29776-12-git-send-email-jsnow@redhat.com
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

qmp: Add support of "dirty-bitmap" sync mode for drive-backup

For "dirty-bitmap" sync mode, the block job will iterate through the
given dirty bitmap to decide if a sector needs backup (backup all the
dirty clusters and skip clean ones), just as allocation conditions of
"top" sync mode.

Signed-off-by: Fam Zheng <famz@redhat.com>
Signed-off-by: John Snow <jsnow@redhat.com>
Reviewed-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Message-id: 1429314609-29776-11-git-send-email-jsnow@redhat.com
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block: Add bitmap successors

A bitmap successor is an anonymous BdrvDirtyBitmap that is intended to
be created just prior to a sensitive operation (e.g. Incremental Backup)
that can either succeed or fail, but during the course of which we still
want a bitmap tracking writes.

On creating a successor, we "freeze" the parent bitmap which prevents
its deletion, enabling, anonymization, or creating a bitmap with the
same name.

On success, the parent bitmap can "abdicate" responsibility to the
successor, which will inherit its name. The successor will have been
tracking writes during the course of the backup operation. The parent
will be safely deleted.

On failure, we can "reclaim" the successor from the parent, unifying
them such that the resulting bitmap describes all writes occurring since
the last successful backup, for instance. Reclamation will thaw the
parent, but not explicitly re-enable it.

BdrvDirtyBitmap operations that target a single bitmap are protected
by assertions that the bitmap is not frozen and/or disabled.

BdrvDirtyBitmap operations that target a group of bitmaps, such as
bdrv_{set,reset}_dirty will ignore frozen/disabled drives with a
conditional instead.

Internal functions that enable/disable dirty bitmaps have assertions
added to them to prevent modifying frozen bitmaps.

Signed-off-by: John Snow <jsnow@redhat.com>
Reviewed-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Message-id: 1429314609-29776-10-git-send-email-jsnow@redhat.com
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block: Add bitmap disabled status

Add a status indicating the enabled/disabled state of the bitmap.
A bitmap is by default enabled, but you can lock the bitmap into
a read-only state by setting disabled = true.

A previous version of this patch added a QMP interface for changing
the state of the bitmap, but it has since been removed for now until
a use case emerges where this state must be revealed to the user.

The disabled state WILL be used internally for bitmap migration and
bitmap persistence.

Signed-off-by: Fam Zheng <famz@redhat.com>
Signed-off-by: John Snow <jsnow@redhat.com>
Reviewed-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Message-id: 1429314609-29776-9-git-send-email-jsnow@redhat.com
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

hbitmap: add hbitmap_merge

We add a bitmap merge operation to assist in error cases
where we wish to combine two bitmaps together.

This is algorithmically O(bits) provided HBITMAP_LEVELS remains
constant. For a full bitmap on a 64bit machine:
sum(bits/64^k, k, 0, HBITMAP_LEVELS) ~= 1.01587 * bits

We may be able to improve running speed for particularly sparse
bitmaps by using iterators, but the running time for dense maps
will be worse.

We present the simpler solution first, and we can refine it later
if needed.

Signed-off-by: John Snow <jsnow@redhat.com>
Reviewed-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Message-id: 1429314609-29776-8-git-send-email-jsnow@redhat.com
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

hbitmap: cache array lengths

As a convenience: between incremental backups, bitmap migrations
and bitmap persistence we seem to need to recalculate these a lot.

Because the lengths are a little bit-twiddly, let's just solidly
cache them and be done with it.

Reviewed-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: John Snow <jsnow@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Message-id: 1429314609-29776-7-git-send-email-jsnow@redhat.com
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block: Introduce bdrv_dirty_bitmap_granularity()

This returns the granularity (in bytes) of dirty bitmap,
which matches the QMP interface and the existing query
interface.

Signed-off-by: John Snow <jsnow@redhat.com>
Reviewed-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Message-id: 1429314609-29776-6-git-send-email-jsnow@redhat.com
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

qmp: Add block-dirty-bitmap-add and block-dirty-bitmap-remove

The new command pair is added to manage a user created dirty bitmap. The
dirty bitmap's name is mandatory and must be unique for the same device,
but different devices can have bitmaps with the same names.

The granularity is an optional field. If it is not specified, we will
choose a default granularity based on the cluster size if available,
clamped to between 4K and 64K to mirror how the 'mirror' code was
already choosing granularity. If we do not have cluster size info
available, we choose 64K. This code has been factored out into a helper
shared with block/mirror.

This patch also introduces the 'block_dirty_bitmap_lookup' helper,
which takes a device name and a dirty bitmap name and validates the
lookup, returning NULL and setting errp if there is a problem with
either field. This helper will be re-used in future patches in this
series.

The types added to block-core.json will be re-used in future patches
in this series, see:
'qapi: Add transaction support to block-dirty-bitmap-{add, enable, disable}'

Signed-off-by: John Snow <jsnow@redhat.com>
Reviewed-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Message-id: 1429314609-29776-5-git-send-email-jsnow@redhat.com
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

qmp: Ensure consistent granularity type

We treat this field with a variety of different types everywhere
in the code. Now it's just uint32_t.

Signed-off-by: John Snow <jsnow@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Message-id: 1429314609-29776-4-git-send-email-jsnow@redhat.com
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

qapi: Add optional field "name" to block dirty bitmap

This field will be set for user created dirty bitmap. Also pass in an
error pointer to bdrv_create_dirty_bitmap, so when a name is already
taken on this BDS, it can report an error message. This is not global
check, two BDSes can have dirty bitmap with a common name.

Implemented bdrv_find_dirty_bitmap to find a dirty bitmap by name, will
be used later when other QMP commands want to reference dirty bitmap by
name.

Add bdrv_dirty_bitmap_make_anon. This unsets the name of dirty bitmap.

Signed-off-by: Fam Zheng <famz@redhat.com>
Signed-off-by: John Snow <jsnow@redhat.com>
Reviewed-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Message-id: 1429314609-29776-3-git-send-email-jsnow@redhat.com
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

docs: incremental backup documentation

Signed-off-by: John Snow <jsnow@redhat.com>
Reviewed-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Message-id: 1429314609-29776-2-git-send-email-jsnow@redhat.com
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block/iscsi: use the allocationmap also if cache.direct=on

the allocationmap has only a hint character. The driver always
double checks that blocks marked unallocated in the cache are
still unallocated before taking the fast path and return zeroes.
So using the allocationmap is migration safe and can
also be enabled with cache.direct=on.

Signed-off-by: Peter Lieven <pl@kamp.de>
Message-id: 1429193313-4263-10-git-send-email-pl@kamp.de
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block/iscsi: bump year in copyright notice

Signed-off-by: Peter Lieven <pl@kamp.de>
Message-id: 1429193313-4263-9-git-send-email-pl@kamp.de
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block/iscsi: handle SCSI_STATUS_TASK_SET_FULL

a target may issue a SCSI_STATUS_TASK_SET_FULL status
if there is more than one "BUSY" command queued already.

Signed-off-by: Peter Lieven <pl@kamp.de>
Message-id: 1429193313-4263-8-git-send-email-pl@kamp.de
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block/iscsi: increase retry count

The idea is that a command is retried in a BUSY condition
up a time of approx. 60 seconds before it is failed. This should
be far higher than any command timeout in the guest.

Signed-off-by: Peter Lieven <pl@kamp.de>
Message-id: 1429193313-4263-7-git-send-email-pl@kamp.de
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block/iscsi: optimize WRITE10/16 if cache.writeback is not set

SCSI allowes to tell the target to not return from a write command
if the date is not written to the disk. Use this so called FUA
bit if it is supported to optimize WRITE commands if writeback is
not allowed.

In this case qemu always issues a WRITE followed by a FLUSH. This
is 2 round trip times. If we set the FUA bit we can ignore the
following FLUSH.

Signed-off-by: Peter Lieven <pl@kamp.de>
Message-id: 1429193313-4263-6-git-send-email-pl@kamp.de
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block/iscsi: store DPOFUA bit from the modesense command

Signed-off-by: Peter Lieven <pl@kamp.de>
Message-id: 1429193313-4263-5-git-send-email-pl@kamp.de
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block/iscsi: rename iscsi_write_protected and let it return void

Signed-off-by: Peter Lieven <pl@kamp.de>
Message-id: 1429193313-4263-4-git-send-email-pl@kamp.de
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block/iscsi: change all iscsilun properties from uint8_t to bool

Signed-off-by: Peter Lieven <pl@kamp.de>
Message-id: 1429193313-4263-3-git-send-email-pl@kamp.de
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block/iscsi: do not forget to logout from target

We actually were always impolitely dropping the connection and
not cleanly logging out.

CC: qemu-stable@nongnu.org
Signed-off-by: Peter Lieven <pl@kamp.de>
Message-id: 1429193313-4263-2-git-send-email-pl@kamp.de
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

qmp: fill in the image field in BlockDeviceInfo

The image field in BlockDeviceInfo is supposed to contain an ImageInfo
object. However that is being filled in by bdrv_query_info(), not by
bdrv_block_device_info(), which is where BlockDeviceInfo is actually
created.

Anyone calling bdrv_block_device_info() directly will get a null image
field. As a consequence of this, the HMP command 'info block -n -v'
crashes QEMU.

This patch moves the code that fills in that field from
bdrv_query_info() to bdrv_block_device_info().

Signed-off-by: Alberto Garcia <berto@igalia.com>
Message-id: 1429271563-3765-1-git-send-email-berto@igalia.com
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

Revert "hmp: fix crash in 'info block -n -v'"

This reverts commit 638b8366200130cc7cf7a026630bc6bfb63b0c4c.

Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block: add 'node-name' field to BLOCK_IMAGE_CORRUPTED

Since this event can occur in nodes that cannot have a device name
associated, include also a field with the node name.

Signed-off-by: Alberto Garcia <berto@igalia.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Max Reitz <mreitz@redhat.com>
Message-id: 147cec5b3594f4bec0cb41c98afe5fcbfb67567c.1428485266.git.berto@igalia.com
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block: use bdrv_get_device_or_node_name() in error messages

There are several error messages that identify a BlockDriverState by
its device name. However those errors can be produced in nodes that
don't have a device name associated.

In those cases we should use bdrv_get_device_or_node_name() to fall
back to the node name and produce a more meaningful message. The
messages are also updated to use the more generic term 'node' instead
of 'device'.

Signed-off-by: Alberto Garcia <berto@igalia.com>
Reviewed-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Message-id: 9823a1f0514fdb0692e92868661c38a9e00a12d6.1428485266.git.berto@igalia.com
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block: add bdrv_get_device_or_node_name()

This function gets the device name associated with a BlockDriverState,
or its node name if the device name is empty.

Signed-off-by: Alberto Garcia <berto@igalia.com>
Reviewed-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Message-id: 4fa30aa8d61d9052ce266fd5429a59a14e941255.1428485266.git.berto@igalia.com
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block: document block-stream in qmp-commands.hx

The 'block-stream' QMP command is documented in block-core.json but not
qmp-commands.hx.  Add a summary of the command to qmp-commands.hx
(similar to the documentation for 'block-commit').

Reported-by: Kashyap Chamarthy <kchamart@redhat.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Max Reitz <mreitz@redhat.com>
Message-id: 1429094622-26218-1-git-send-email-stefanha@redhat.com
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

m25p80: fix s->blk usage before assignment

Delay the call to blk_blockalign() until s->blk has been assigned.

This never caused a crash because blk_blockalign(NULL, size) defaults to
4096 alignment but it's technically incorrect.

Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Message-id: 1429091024-25098-1-git-send-email-stefanha@redhat.com
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

m25p80: add missing blk_attach_dev_nofail

Of the block devices that poked into -drive options via drive_get_next,
m25p80 was the only one who also did not attach itself to the BlockBackend.

Since sd does it, and all other devices go through a "drive" property,
with this change all block backends attached to the guest will have a
non-NULL result for blk_get_attached_dev().

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Peter Crosthwaite <peter.crosthwaite@xilinx.com>
Message-id: 1429025387-11077-1-git-send-email-pbonzini@redhat.com
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

virtio_blk: comment fix

update virtio blk header from latest linux, include comment fixups.

Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Message-id: 1428854036-12806-1-git-send-email-mst@redhat.com
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block: avoid unnecessary bottom halves

bdrv_aio_* APIs can use coroutines to achieve asynchronicity.  However,
the coroutine may terminate without having yielded back to the caller
(for example because of something that invokes a nested event loop,
or because the coroutine is doing nothing at all).  In this case,
the bdrv_aio_* API must delay the completion to the next iteration
of the main loop, because bdrv_aio_* will never invoke the callback
before returning.

This can be done with a bottom half, and indeed bdrv_aio_* is always
using one for simplicity.  It is possible to gain some performance
(~3%) by avoiding this in the common case.  A new field in the
BlockAIOCBCoroutine struct is set to true until the first time the
corotine has yielded to its creator, and completion goes through a
new function bdrv_co_complete.  If the flag is false, bdrv_co_complete
invokes the callback immediately.  If it is true, the caller will
notice that the coroutine has completed and schedule the bottom
half itself.

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Message-id: 1427524638-28157-1-git-send-email-pbonzini@redhat.com
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

blockjob: Update function name in comments

Signed-off-by: Fam Zheng <famz@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Alberto Garcia <berto@igalia.com>
Message-id: 1428069921-2957-5-git-send-email-famz@redhat.com
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

qemu-iotests: Test that "stop" doesn't drain block jobs

Signed-off-by: Fam Zheng <famz@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Alberto Garcia <berto@igalia.com>
Message-id: 1428069921-2957-4-git-send-email-famz@redhat.com
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block: Pause block jobs in bdrv_drain_all

This is necessary to suppress more IO requests from being generated from
block job coroutines.

Signed-off-by: Fam Zheng <famz@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Alberto Garcia <berto@igalia.com>
Message-id: 1428069921-2957-3-git-send-email-famz@redhat.com
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

blockjob: Allow nested pause

This patch changes block_job_pause to increase the pause counter and
block_job_resume to decrease it.

The counter will allow calling block_job_pause/block_job_resume
unconditionally on a job when we need to suspend the IO temporarily.

From now on, each block_job_resume must be paired with a block_job_pause
to keep the counter balanced.

The user pause from QMP or HMP will only trigger block_job_pause once
until it's resumed, this is achieved by adding a user_paused flag in
BlockJob.

One occurrence of block_job_resume in mirror_complete is replaced with
block_job_enter which does what is necessary.

In block_job_cancel, the cancel flag is good enough to instruct
coroutines to quit loop, so use block_job_enter to replace the unpaired
block_job_resume.

Upon block job IO error, user is notified about the entering to the
pause state, so this pause belongs to user pause, set the flag
accordingly and expect a matching QMP resume.

[Extended doc comments as suggested by Paolo Bonzini
<pbonzini@redhat.com>.
--Stefan]

Signed-off-by: Fam Zheng <famz@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Alberto Garcia <berto@igalia.com>
Message-id: 1428069921-2957-2-git-send-email-famz@redhat.com
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

MAINTAINERS: Add Fam Zheng as Null block driver maintainer

Signed-off-by: Fam Zheng <famz@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Message-id: 1427852740-24315-4-git-send-email-famz@redhat.com
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block/null: Support reopen

Reopen is used in block-commit. With this always-succeed operation, it
is now possible to test committing to a null drive, by specifying
"null-aio://" or "null-co://" as the backing image when creating the
qcow2 image.

Signed-off-by: Fam Zheng <famz@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Message-id: 1427852740-24315-3-git-send-email-famz@redhat.com
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block/null: Latency simulation by adding new option "latency-ns"

Aio context switch should just work because the requests will be
drained, so the scheduled timer(s) on the old context will be freed.

Signed-off-by: Fam Zheng <famz@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Message-id: 1427852740-24315-2-git-send-email-famz@redhat.com
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

scripts: add 'qemu coroutine' command to qemu-gdb.py

The 'qemu coroutine <coroutine-address>' GDB command prints the
backtrace for a CoroutineUContext.  This is useful for peeking inside
yielded coroutines that are waiting for file descriptor events, timers,
etc.

For example:

  $ gdb tests/test-coroutine
  (gdb) b test_yield
  (gdb) r
  (gdb) b qemu_coroutine_enter
  (gdb) c
  (gdb) c
  Continuing.

  Breakpoint 2, qemu_coroutine_enter (co=0x555555c66520, opaque=0x0) at qemu-coroutine.c:103
  103 {
  (gdb) source scripts/qemu-gdb.py
  (gdb) qemu coroutine 0x555555c66520
  #0  0x000055555557a740 in qemu_coroutine_switch (from_=<optimized out>, to_=0x7ffff7f90a70, action=COROUTINE_YIELD) at coroutine-ucontext.c:177
  #1  0x0000555555566af9 in yield_5_times (opaque=0x7fffffffdbb7) at tests/test-coroutine.c:107
  #2  0x000055555557a7aa in coroutine_trampoline (i0=<optimized out>, i1=<optimized out>) at coroutine-ucontext.c:80
  #3  0x00007ffff08de000 in __start_context () at /lib64/libc.so.6

Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Message-id: 1427409754-8556-1-git-send-email-stefanha@redhat.com
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

thread-pool: clean up thread_pool_completion_bh()

This patch simplifies thread_pool_completion_bh().

The function first checks elem->state:

  if (elem->state != THREAD_DONE) {
      continue;
  }

It then goes on to check elem->state == THREAD_DONE although we already
know this must be the case.

The QLIST_REMOVE() is duplicated down both branches of an if-else
statement so that can be lifted out as well.

Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Message-id: 1427992762-10126-1-git-send-email-stefanha@redhat.com
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

vhdx: Fix zero-fill iov length

Fix the length of the zero-fill for the back, which was accidentally
using the same value as for the front. This is caught by qemu-iotests
033.

For consistency, change the code for the front as well to use the length
stored in the iov (it is the same value, copied four lines above).

Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Acked-by: Jeff Cody <jcody@redhat.com>

blkdebug: Add bdrv_truncate()

This is, amongst others, required for qemu-iotests 033 to run as
intended on VHDX, which uses explicit bdrv_truncate() calls to bs->file
when allocating new blocks.

Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Jeff Cody <jcody@redhat.com>

qemu-iotests: Some qemu-img convert tests

This adds a regression test for some problems that the qemu-img convert
rewrite just fixed.

Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Max Reitz <mreitz@redhat.com>

qemu-img convert: Rewrite copying logic

The implementation of qemu-img convert is (a) messy, (b) buggy, and
(c) less efficient than possible. The changes required to beat some
sense into it are massive enough that incremental changes would only
make my and the reviewers' life harder. So throw it away and reimplement
it from scratch.

Let me give some examples what I mean by messy, buggy and inefficient:

(a) The copying logic of qemu-img convert has two separate branches for
    compressed and normal target images, which roughly do the same -
    except for a little code that handles actual differences between
    compressed and uncompressed images, and much more code that
    implements just a different set of optimisations and bugs. This is
    unnecessary code duplication, and makes the code for compressed
    output (unsurprisingly) suffer from bitrot.

    The code for uncompressed ouput is run twice to count the the total
    length for the progress bar. In the first run it just takes a
    shortcut and runs only half the loop, and when it's done, it toggles
    a boolean, jumps out of the loop with a backwards goto and starts
    over. Works, but pretty is something different.

(b) Converting while keeping a backing file (-B option) is broken in
    several ways. This includes not writing to the image file if the
    input has zero clusters or data filled with zeros (ignoring that the
    backing file will be visible instead).

    It also doesn't correctly limit every iteration of the copy loop to
    sectors of the same status so that too many sectors may be copied to
    in the target image. For -B this gives an unexpected result, for
    other images it just does more work than necessary.

    Conversion with a compressed target completely ignores any target
    backing file.

(c) qemu-img convert skips reading and writing an area if it knows from
    metadata that copying isn't needed (except for the bug mentioned
    above that ignores a status change in some cases). It does, however,
    read from the source even if it knows that it will read zeros, and
    then search for non-zero bytes in the read buffer, if it's possible
    that a write might be needed.

This reimplementation of the copying core reorganises the code to remove
the duplication and have a much more obvious code flow, by essentially
splitting the copy iteration loop into three parts:

1. Find the number of contiguous sectors of the same status at the
   current offset (This can also be called in a separate loop before the
   copying loop in order to determine the total sectors for the progress
   bar.)

2. Read sectors. If the status implies that there is no data there to
   read (zero or unallocated cluster), don't do anything.

3. Write sectors depending on the status. If it's data, write it. If
   we want the backing file to be visible (with -B), don't write it. If
   it's zeroed, skip it if you can, otherwise use bdrv_write_zeroes() to
   optimise the write at least where possible.

Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Max Reitz <mreitz@redhat.com>

block-backend: Expose bdrv_write_zeroes()

Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Max Reitz <mreitz@redhat.com>

iothread: release iothread around aio_poll

This is the first step towards having fine-grained critical sections in
dataplane threads, which resolves lock ordering problems between
address_space_* functions (which need the BQL when doing MMIO, even
after we complete RCU-based dispatch) and the AioContext.

Because AioContext does not use contention callbacks anymore, the
unit test has to be changed.

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Message-id: 1424449612-18215-4-git-send-email-pbonzini@redhat.com
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

AioContext: acquire/release AioContext during aio_poll

This is the first step in pushing down acquire/release, and will let
rfifolock drop the contention callback feature.

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Message-id: 1424449612-18215-3-git-send-email-pbonzini@redhat.com
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

aio-posix: move pollfds to thread-local storage

By using thread-local storage, aio_poll can stop using global data during
g_poll_ns.  This will make it possible to drop callbacks from rfifolock.

[Moved npfd = 0 assignment to end of walking_handlers region as
suggested by Paolo.  This resolves the assert(npfd == 0) assertion
failure in pollfds_cleanup().
--Stefan]

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Message-id: 1424449612-18215-2-git-send-email-pbonzini@redhat.com
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block: Switch to host monotonic clock for IO throttling

Currently, throttle timers won't make any progress when VCPU is not
running, which would stall the request queue in utils, qtest, vm
suspending, and live migration, without special handling.

Block jobs are confusingly inconsistent between with and without
throttling: if user sets a bps limit, stops the vm, then start a block
job, the block job will not make any progress; in contrary, if user
unsets the bps limit, or if it's not set, the block job will run
normally.

After this patch, with the host clock, even if the VCPUs are stopped,
the throttle queues will be processed.

This patch also enables potential to add throttle to bdrv_drain_all.
Currently all requests are drained immediately. In other words whenever
it is called, IO throttling goes ineffective (examples: system reset,
migration and many block job operations.). This is a loophole that guest
could exploit. If we use the host clock, we can later just trust the
nested poll. This could be done on top.

Note that for qemu-iotests case 093, which uses qtest, we still keep vm
clock so the script can control the clock stepping in order to be
deterministic.

Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Alberto Garcia <berto@igalia.com>
Signed-off-by: Fam Zheng <famz@redhat.com>
Message-id: 1427268446-6426-1-git-send-email-famz@redhat.com
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

checkpatch: complain about ffs(3) calls

The ffs(3) family of functions is not portable.  MinGW doesn't always
provide the function.

Use ctz32() or ctz64() instead.

Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Message-id: 1427124571-28598-10-git-send-email-stefanha@redhat.com
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

os-win32: drop ffs(3) prototype

The lack of ffs(3) in the MinGW headers is a hint that we shouldn't rely
on it.  MinGW 4.9.2 does not make it available for linking when QEMU's
./configure --enable-debug is used (release builds are fine though).

Now that all QEMU code has been switched to ctz32() there is no need for
ffs(3).

Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Message-id: 1427124571-28598-9-git-send-email-stefanha@redhat.com
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

omap_intc: convert ffs(3) to ctz32() in omap_inth_sir_update()

Rewrite the loop using level &= level - 1 to clear the least significant
bit after each iteration.  This simplifies the loop and makes it easy to
replace ffs(3) with ctz32().

Cc: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Message-id: 1427124571-28598-8-git-send-email-stefanha@redhat.com
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

sd: convert sd_normal_command() ffs(3) call to ctz32()

ffs() cannot be replaced with ctz32() when the argument might be zero,
because ffs(0) returns 0 while ctz32(0) returns 32.

The ffs(3) call in sd_normal_command() is a special case though.  It can
be converted to ctz32() + 1 because the argument is never zero:

  if (!(req.arg >> 8) || (req.arg >> (ctz32(req.arg & ~0xff) + 1))) {
      ~~~~~~~~~~~~~~~
            ^--------------- req.arg cannot be zero

Cc: Markus Armbruster <armbru@redhat.com>
Cc: Peter Crosthwaite <peter.crosthwaite@xilinx.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Message-id: 1427124571-28598-7-git-send-email-stefanha@redhat.com
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

Convert ffs() != 0 callers to ctz32()

There are a number of ffs(3) callers that do roughly:

  bit = ffs(val);
  if (bit) {
      do_something(bit - 1);
  }

This pattern can be converted to ctz32() like this:

  zeroes = ctz32(val);
  if (zeroes != 32) {
      do_something(zeroes);
  }

Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Message-id: 1427124571-28598-6-git-send-email-stefanha@redhat.com
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

Convert (ffs(val) - 1) to ctz32(val)

This commit was generated mechanically by coccinelle from the following
semantic patch:

@@
expression val;
@@
- (ffs(val) - 1)
+ ctz32(val)

The call sites have been audited to ensure the ffs(0) - 1 == -1 case
never occurs (due to input validation, asserts, etc).  Therefore we
don't need to worry about the fact that ctz32(0) == 32.

Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Message-id: 1427124571-28598-5-git-send-email-stefanha@redhat.com
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

uninorth: convert ffs(3) to ctz32()

It is not clear from the code how a 0 parameter should be handled by the
hardware.  Keep the same behavior as ffs(0) - 1 == -1.

Cc: Alexander Graf <agraf@suse.de>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Message-id: 1427124571-28598-4-git-send-email-stefanha@redhat.com
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

hw/arm/nseries: convert ffs(3) to ctz32()

It is not clear from the code how a 0 parameter should be handled by the
hardware.  Keep the same behavior as ffs(0) - 1 == -1.

Cc: Andrzej Zaborowski <balrog@zabor.org>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Message-id: 1427124571-28598-3-git-send-email-stefanha@redhat.com
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

bt-sdp: fix broken uuids power-of-2 calculation

The binary search in sdp_uuid_match() only works when the number of
elements to search is a power of two.

  lo = record->uuid;
  hi = record->uuids;
  while (hi >>= 1)
      if (lo[hi] <= val)
          lo += hi;

  return *lo == val;

I noticed that the record->uuids calculation in
sdp_service_record_build() was suspect:

  record->uuids = 1 << ffs(record->uuids - 1);

Unlike most ffs(val) - 1 users, the expression is ffs(val - 1)!

Actually ffs() is the wrong function to use for power-of-2.  Use
pow2ceil() to achieve the correct effect.  Now the record->uuid[] array
is sized correctly and the binary search in sdp_uuid_match() should
work.

I'm not sure how to run/test this code.

Cc: Andrzej Zaborowski <balrog@zabor.org>
Cc: qemu-stable@nongnu.org
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Message-id: 1427124571-28598-2-git-send-email-stefanha@redhat.com
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

MAINTAINERS: Add myself as the maintainer of the Quorum driver

Signed-off-by: Alberto Garcia <berto@igalia.com>
Message-id: 1426522925-14444-1-git-send-email-berto@igalia.com
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

savevm: create snapshot failed when id_str already exists

The command "virsh create" will fail in such condition: vm has two
disks: vda and vdb. vda has snapshot s1 with id "1", vdb doesn't have
s1 but has snapshot s2 with id "1".  When we want to run command "virsh
create s1", del_existing_snapshots() only deletes s1 in vda, and
bdrv_snapshot_create() tries to create vdb's snapshot s1 with id "1",
but id "1" alreay exists in vdb with name "s2"!

The simplest way is call find_new_snapshot_id() unconditionally.

Signed-off-by: Yi Wang <up2wing@gmail.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

Merge remote-tracking branch 'remotes/ehabkost/tags/x86-pull-request' into staging

X86 queue, 2015-04-27 (v2)

# gpg: Signature made Mon Apr 27 19:42:39 2015 BST using RSA key ID 984DC5A6
# gpg: Good signature from "Eduardo Habkost <ehabkost@redhat.com>"
# gpg: WARNING: This key is not certified with sufficiently trusted signatures!
# gpg:          It is not certain that the signature belongs to the owner.
# Primary key fingerprint: 5A32 2FD5 ABC4 D3DB ACCF  D1AA 2807 936F 984D C5A6

* remotes/ehabkost/tags/x86-pull-request:
  target-i386: Remove AMD feature flag aliases from CPU model table
  target-i386: X86CPU::xlevel2 QOM property
  target-i386: Make "level" and "xlevel" properties static
  qemu-config: Accept empty option values
  MAINTAINERS: Change status of X86 to Maintained
  MAINTAINERS: Add myself to X86

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>

Merge remote-tracking branch 'remotes/ehabkost/tags/numa-pull-request' into staging

NUMA queue, 2015-04-27

# gpg: Signature made Mon Apr 27 19:02:19 2015 BST using RSA key ID 984DC5A6
# gpg: Good signature from "Eduardo Habkost <ehabkost@redhat.com>"
# gpg: WARNING: This key is not certified with sufficiently trusted signatures!
# gpg:          It is not certain that the signature belongs to the owner.
# Primary key fingerprint: 5A32 2FD5 ABC4 D3DB ACCF  D1AA 2807 936F 984D C5A6

* remotes/ehabkost/tags/numa-pull-request:
  MAINTAINERS: Add myself as NUMA code maintainer

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>

Merge remote-tracking branch 'remotes/pmaydell/tags/pull-target-arm-20150427' into staging

target-arm queue:
 * memory system updates to support transaction attributes
 * set user-mode and secure attributes for accesses made by ARM CPUs
 * rename c1_coproc to cpacr_el1
 * adjust id_aa64pfr0 when has_el3 CPU property disabled
 * allow ARMv8 SCR.SMD updates

# gpg: Signature made Mon Apr 27 16:14:30 2015 BST using RSA key ID 14360CDE
# gpg: Good signature from "Peter Maydell <peter.maydell@linaro.org>"

* remotes/pmaydell/tags/pull-target-arm-20150427:
  Allow ARMv8 SCR.SMD updates
  target-arm: Adjust id_aa64pfr0 when has_el3 CPU property disabled
  target-arm: rename c1_coproc to cpacr_el1
  target-arm: Check watchpoints against CPU security state
  target-arm: Use attribute info to handle user-only watchpoints
  target-arm: Add user-mode transaction attribute
  target-arm: Use correct memory attributes for page table walks
  target-arm: Honour NS bits in page tables
  Switch non-CPU callers from ld/st*_phys to address_space_ld/st*
  exec.c: Capture the memory attributes for a watchpoint hit
  exec.c: Add new address_space_ld*/st* functions
  exec.c: Make address_space_rw take transaction attributes
  exec.c: Convert subpage memory ops to _with_attrs
  Add MemTxAttrs to the IOTLB
  Make CPU iotlb a structure rather than a plain hwaddr
  memory: Replace io_mem_read/write with memory_region_dispatch_read/write
  memory: Define API for MemoryRegionOps to take attrs and return status

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>

Merge remote-tracking branch 'remotes/spice/tags/pull-spice-20150427-1' into staging

spice: misc fixes.

# gpg: Signature made Mon Apr 27 12:03:16 2015 BST using RSA key ID D3E87138
# gpg: Good signature from "Gerd Hoffmann (work) <kraxel@redhat.com>"
# gpg:                 aka "Gerd Hoffmann <gerd@kraxel.org>"
# gpg:                 aka "Gerd Hoffmann (private) <kraxel@gmail.com>"

* remotes/spice/tags/pull-spice-20150427-1:
  spice: learn to hide cursor
  spice: set pointer position on hotspot
  spice: fix mouse cursor position
  spice: fix simple display on bigendian hosts
  monitor: Make client_migrate_info synchronous

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>

target-i386: Remove AMD feature flag aliases from CPU model table

When CPU vendor is AMD, the AMD feature alias bits on
CPUID[0x80000001].EDX are already automatically copied from CPUID[1].EDX
on x86_cpu_realizefn(). When CPU vendor is Intel, those bits are
reserved and should be zero. On either case, those bits shouldn't be set
in the CPU model table.

Reviewed-by: Igor Mammedov <imammedo@redhat.com>
Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>

target-i386: X86CPU::xlevel2 QOM property

We already have "level" and "xlevel", only "xlevel2" is missing.

Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>

target-i386: Make "level" and "xlevel" properties static

Static properties require only 1 line of code, much simpler than the
existing code that requires writing new getters/setters.

As a nice side-effect, this fixes an existing bug where the setters were
incorrectly allowing the properties to be changed after the CPU was
already realized.

Reviewed-by: Igor Mammedov <imammedo@redhat.com>
Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>

qemu-config: Accept empty option values

Currently it is impossible to set an option in a config file to an empty
string, because the parser matches only lines containing non-empty
strings between double-quotes.

As sscanf() "[" conversion specifier only matches non-empty strings, add
a special case for empty strings.

Reviewed-by: Eric Blake <eblake@redhat.com>
Acked-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>

MAINTAINERS: Change status of X86 to Maintained

"Odd Fixes" doesn't reflect the current status of target-i386. We have
people looking after it, now.

Acked-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>

MAINTAINERS: Add myself to X86

Acked-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>

Merge remote-tracking branch 'remotes/kraxel/tags/pull-gtk-20150427-1' into staging

gtk: support text consoles without vte, bugfixes.

# gpg: Signature made Mon Apr 27 14:34:15 2015 BST using RSA key ID D3E87138
# gpg: Good signature from "Gerd Hoffmann (work) <kraxel@redhat.com>"
# gpg:                 aka "Gerd Hoffmann <gerd@kraxel.org>"
# gpg:                 aka "Gerd Hoffmann (private) <kraxel@gmail.com>"

* remotes/kraxel/tags/pull-gtk-20150427-1:
  gtk: Avoid accel key leakage into guest on console switch
  gtk: Fix VTE focus grabbing
  console/gtk: add qemu_console_get_label
  gtk: bind to text terminal consoles too
  gtk: handle switch_surface(NULL) properly

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>

MAINTAINERS: Add myself as NUMA code maintainer

The "srat" and "numa" keywords will help get_maintainer.pl catch
NUMA-related code in other files too.

Acked-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>

Merge remote-tracking branch 'remotes/qmp-unstable/tags/for-upstream' into staging

Four little fixes

# gpg: Signature made Fri Apr 24 19:56:51 2015 BST using RSA key ID E24ED5A7
# gpg: Good signature from "Luiz Capitulino <lcapitulino@gmail.com>"

* remotes/qmp-unstable/tags/for-upstream:
  qmp: Give saner messages related to qmp_capabilities misuse
  qmp-commands: fix incorrect uses of ":O" specifier
  qapi: Drop dead genlist parameter
  balloon: improve error msg when adding second device

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>

spice: learn to hide cursor

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

spice: set pointer position on hotspot

The Spice protocol uses cursor position on hotspot: the client is
applying hotspot offset when drawing the cursor.

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

spice: fix mouse cursor position

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

spice: fix simple display on bigendian hosts

Denis Kirjanov is busy getting spice run on ppc64 and trapped into this
one.  Spice wire format is little endian, so we have to explicitly say
we want little endian when letting pixman convert the data for us.

Reported-by: Denis Kirjanov <kirjanov@gmail.com>
Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>

monitor: Make client_migrate_info synchronous

Live migration with spice works like this today:

  (1) client_migrate_info monitor cmd
  (2) spice server notifies client, client connects to target host.
  (3) qemu waits until spice client connect is finished.
  (4) send over vmstate (i.e. main part of live migration).
  (5) spice handover to target host.

(3) is implemented by making client_migrate_info a async monitor
command.  This is the only async monitor command we have.

The original reason to implement this dance was that qemu did not accept
new tcp connections while the incoming migration was running, so (2) and
(4) could not be done in parallel.  That issue was fixed long ago though.
Qemu version 1.3.0 (released Dec 2012) and newer happily accept tcp
connects while the incoming migration runs.

Time to drop step (3).  This patch does exactly that, by making the
monitor command synchronous and removing the code needed to handle the
async monitor command in ui/spice-core.c

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

gtk: Avoid accel key leakage into guest on console switch

GTK2 sends the accel key to the guest when switching to the graphic
console via that shortcut. Resolve this by ignoring any keys until the
next key-release event. However, do not ignore keys when switching via
the menu or when on GTK3.

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>

gtk: Fix VTE focus grabbing

At least on GTK2, the VTE terminal has to be specified as target of
gtk_widget_grab_focus. Otherwise, switching from one VTE terminal to
another causes the focus to get lost.

CC: John Snow <jsnow@redhat.com>
Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
[ kraxel: fixed build with CONFIG_VTE=n ]

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

Allow ARMv8 SCR.SMD updates

Updated scr_write to always allow updates to the SCR.SMD bit on ARMv8
regardless of whether virtualization (EL2) is enabled or not.

Signed-off-by: Greg Bellows <greg.bellows@linaro.org>
Message-id: 1429888797-4378-1-git-send-email-greg.bellows@linaro.org
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>

target-arm: Adjust id_aa64pfr0 when has_el3 CPU property disabled

Signed-off-by: Sergey Fedorov <serge.fdrv@gmail.com>
Message-id: 1429669112-29835-1-git-send-email-serge.fdrv@gmail.com
Reviewed-by: Greg Bellows <greg.bellows@linaro.org>
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>

target-arm: rename c1_coproc to cpacr_el1

Rename the field holding CPACR_EL1 system register state in AArch64
naming style.

Signed-off-by: Sergey Fedorov <serge.fdrv@gmail.com>
[PMM: also fixed a couple of missed occurrences in cpu.c]
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>