Merge remote-tracking branch 'remotes/jnsnow/tags/ide-pull-request' into staging

# gpg: Signature made Fri May 22 20:58:44 2015 BST using RSA key ID AAFC390E
# gpg: Good signature from "John Snow (John Huston) <jsnow@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: FAEB 9711 A12C F475 812F  18F2 88A9 064D 1835 61EB
#      Subkey fingerprint: F9B7 ABDB BCAC DF95 BE76  CBD0 7DEF 8106 AAFC 390E

* remotes/jnsnow/tags/ide-pull-request:
  ahci: do not remap clb/fis unconditionally
  macio: move unaligned DMA write code into separate pmac_dma_write() function
  macio: move unaligned DMA read code into separate pmac_dma_read() function
  qtest: pre-buffer hex nibs
  libqos/ahci: Swap memread/write with bufread/write
  qtest: add memset to qtest protocol
  qtest: Add base64 encoded read/write
  qtest: allow arbitrarily long sends
  qtest/ahci: add migrate halted dma test
  qtest/ahci: add halted dma test
  qtest/ahci: add flush migrate test
  qtest/ahci: add migrate dma test
  qtest/ahci: Add migration test
  ich9/ahci: Enable Migration
  libqos: Add migration helpers
  libqos/ahci: Fix sector set method
  libqos/ahci: Add halted command helpers
  glib: remove stale compat functions
  configure: require glib 2.22

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

ahci: do not remap clb/fis unconditionally

This continues the IOMMU fix from 2.3, where we should not attempt
to remap the CLB or FIS RX buffers if the AHCI device is currently
running.

The same applies to migration: keep our mitts off these registers
unless the device is supposed to be on.

Does not impact backwards compatibility for the AHCI device.

Signed-off-by: John Snow <jsnow@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Message-id: 1431470173-30847-2-git-send-email-jsnow@redhat.com

macio: move unaligned DMA write code into separate pmac_dma_write() function

Similarly switch the macio IDE routines over to use the new function and
tidy-up the remaining code as required.

[Maintainer edit: printf format codes adjusted for 32/64bit. --js]

Signed-off-by: Mark Cave-Ayland <mark.cave-ayland@ilande.co.uk>
Acked-by: John Snow <jsnow@redhat.com>
Message-id: 1425939893-14404-3-git-send-email-mark.cave-ayland@ilande.co.uk
Signed-off-by: John Snow <jsnow@redhat.com>

macio: move unaligned DMA read code into separate pmac_dma_read() function

This considerably helps simplify the complexity of the macio read routines and
by switching macio CDROM accesses to use the new code, fixes the issue with
the CDROM device being detected intermittently by Darwin/OS X.

[Maintainer edit: printf format codes adjusted for 32/64bit. --js]

Signed-off-by: Mark Cave-Ayland <mark.cave-ayland@ailande.co.uk>
Acked-by: John Snow <jsnow@redhat.com>
Message-id: 1425939893-14404-2-git-send-email-mark.cave-ayland@ilande.co.uk
Signed-off-by: John Snow <jsnow@redhat.com>

qtest: pre-buffer hex nibs

Instead of converting each byte one-at-a-time and then sending each byte
over the wire, use sprintf() to pre-compute all of the hex nibs into a
single buffer, then send the entire buffer all at once.

This gives a moderate speed boost to memread() and memwrite() functions.

Signed-off-by: John Snow <jsnow@redhat.com>
Reviewed-by: Markus Armbruster <armbru@redhat.com>
Message-id: 1431021095-7558-2-git-send-email-jsnow@redhat.com

libqos/ahci: Swap memread/write with bufread/write

Where it makes sense, use the new faster primitives.
For generally small reads/writes such as for the PRDT
and FIS packets, stick with the more wasteful but
easier to debug memread/memwrite.

For ahci-test (before migration tests):
With this patch:
real    0m3.675s
user    0m2.582s
sys     0m1.718s

Without any qtest protocol improvements:
real    0m14.171s
user    0m12.072s
sys     0m12.527s

Signed-off-by: John Snow <jsnow@redhat.com>
Message-id: 1430864578-22072-6-git-send-email-jsnow@redhat.com

qtest: add memset to qtest protocol

Previously, memset was just a frontend to write() and only
stupidly sent the pattern many times across the wire.

Let's not discuss who stupidly wrote it like that in the first place.
(Hint: It was me.)

Signed-off-by: John Snow <jsnow@redhat.com>
Message-id: 1430864578-22072-4-git-send-email-jsnow@redhat.com

qtest: Add base64 encoded read/write

For larger pieces of data that won't need to be debugged and
viewing the hex nibbles is unlikely to be useful, we can encode
data using base64 instead of encoding each byte as %02x, which
leads to some space savings and faster reads/writes.

For now, the default is left as hex nibbles in memwrite() and memread().
For the purposes of making qtest io easier to read and debug, some
callers may want to specify using the old encoding format for small
patches of data where the savings from base64 wouldn't be that profound.

memwrite/memread use a data encoding that takes 2x the size of the original
buffer, but base64 uses "only" (4/3)x, so for larger buffers we can save a
decent amount of time and space.

Signed-off-by: John Snow <jsnow@redhat.com>
Message-id: 1430864578-22072-3-git-send-email-jsnow@redhat.com

qtest: allow arbitrarily long sends

qtest currently has a static buffer of size 1024 that if we
overflow, ignores the additional data silently which leads
to hangs or stream failures.

Use glib's string facilities to allow arbitrarily long data,
but split this off into a new function, qtest_sendf.

Static data can still be sent using qtest_send, which avoids
the malloc/copy overhead.

Signed-off-by: John Snow <jsnow@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Message-id: 1430864578-22072-2-git-send-email-jsnow@redhat.com

qtest/ahci: add migrate halted dma test

Test migrating a halted DMA transaction.
Resume, then test data integrity.

Signed-off-by: John Snow <jsnow@redhat.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Message-id: 1430417242-11859-10-git-send-email-jsnow@redhat.com

qtest/ahci: add halted dma test

If we're going to test the migration of halted DMA jobs,
we should probably check to make sure we can resume them
locally as a first step.

Signed-off-by: John Snow <jsnow@redhat.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Message-id: 1430417242-11859-9-git-send-email-jsnow@redhat.com

qtest/ahci: add flush migrate test

Use blkdebug to inject an error on first flush, then attempt to flush
on the first guest. When the error halts the VM, migrate to the
second VM, and attempt to resume the command.

Signed-off-by: John Snow <jsnow@redhat.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Message-id: 1430417242-11859-8-git-send-email-jsnow@redhat.com

qtest/ahci: add migrate dma test

Write to one guest, migrate, and then read from the other.
adjust ahci_io to clear any buffers it creates, so that we
can use ahci_io safely on both guests knowing we are using
empty buffers and not accidentally re-using data.

Signed-off-by: John Snow <jsnow@redhat.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Message-id: 1430417242-11859-7-git-send-email-jsnow@redhat.com

qtest/ahci: Add migration test

Notes:

 * The migration is performed on QOSState objects.

 * The migration is performed in such a way that it does not assume
   consistency between the allocators attached to each. That is to say,
   you can use each QOSState object completely independently and then at
   an arbitrary point decide to migrate, and the destination object will
   now be consistent with the memory within the source guest. The source
   object that was migrated from will have a completely blank allocator.

ahci-test.c:
 - verify_state is added
 - ahci_migrate is added as a frontend to migrate
 - test_migrate_sanity test case is added.

Signed-off-by: John Snow <jsnow@redhat.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Message-id: 1430417242-11859-6-git-send-email-jsnow@redhat.com

ich9/ahci: Enable Migration

Lift the flag preventing the migration of the ICH9/AHCI devices.

Signed-off-by: John Snow <jsnow@redhat.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Message-id: 1430417242-11859-5-git-send-email-jsnow@redhat.com

libqos: Add migration helpers

libqos.c:
    -set_context for addressing which commands go where
    -migrate performs the actual migration

malloc.c:
    - Structure of the allocator is adjusted slightly with
      a second-tier malloc to make swapping around the allocators
      easy when we "migrate" the lists from the source to the destination.

Signed-off-by: John Snow <jsnow@redhat.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Message-id: 1430417242-11859-4-git-send-email-jsnow@redhat.com

libqos/ahci: Fix sector set method

|| probably does not mean the same thing as |.

Additionally, allow users to submit a prd_size of 0
to indicate that they'd like to continue using the default.

Signed-off-by: John Snow <jsnow@redhat.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Message-id: 1430417242-11859-3-git-send-email-jsnow@redhat.com

libqos/ahci: Add halted command helpers

Sometimes we want a command to halt the VM instead
of complete successfully, so it'd be nice to let the
libqos/ahci functions cope with such scenarios.

Signed-off-by: John Snow <jsnow@redhat.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Message-id: 1430417242-11859-2-git-send-email-jsnow@redhat.com

glib: remove stale compat functions

Since we're bumping the version to 2.22+,
remove the now-stale compat functions.

Signed-off-by: John Snow <jsnow@redhat.com>
Reviewed-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Alex Bennée <alex.bennee@linaro.org>
Message-id: 1431469140-22208-2-git-send-email-jsnow@redhat.com

configure: require glib 2.22

This provides g_ptr_array_new_with_free_func, as well as a few
other functions that we've been hacking around in glib-compat.h.
Cleaning up the compatibility headers will come later.

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: John Snow <jsnow@redhat.com>
Reviewed-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Alex Bennée <alex.bennee@linaro.org>
Message-id: 1431469140-22208-2-git-send-email-jsnow@redhat.com

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

Block layer core and image format patches

# gpg: Signature made Fri May 22 16:21:03 2015 BST using RSA key ID C88F2FD6
# gpg: Good signature from "Kevin Wolf <kwolf@redhat.com>"

* remotes/kevin/tags/for-upstream: (22 commits)
  MAINTAINERS: Split "Block QAPI, monitor, command line" off core
  MAINTAINERS: Add header files to Block Layer Core section
  tests: add test case for encrypted qcow2 read/write
  qemu-io: prompt for encryption keys when required
  util: allow \n to terminate password input
  util: move read_password method out of qemu-img into osdep/oslib
  qcow2/qcow: protect against uninitialized encryption key
  qemu-iotests: Make debugging python tests easier
  qemu-iotests: qemu-img info on afl VMDK image with a huge capacity
  block: Detect multiplication overflow in bdrv_getlength
  qemu-io: Use getopt() correctly
  qcow2: style fixes in qcow2-cache.c
  qcow2: make qcow2_cache_put() a void function
  qcow2: use a hash to look for entries in the L2 cache
  qcow2: remove qcow2_cache_find_entry_to_replace()
  qcow2: use an LRU algorithm to replace entries from the L2 cache
  qcow2: simplify qcow2_cache_put() and qcow2_cache_entry_mark_dirty()
  qcow2: use one single memory block for the L2/refcount cache tables
  vmdk: Fix overflow if l1_size is 0x20000000
  vmdk: Fix next_cluster_sector for compressed write
  ...

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

Merge remote-tracking branch 'remotes/bkoppelmann/tags/pull-tricore-20150522' into staging

TriCore v1.6.1 ISA and missing v1.6 instructions

# gpg: Signature made Fri May 22 16:02:45 2015 BST using RSA key ID 6B69CA14
# gpg: Good signature from "Bastian Koppelmann <kbastian@mail.uni-paderborn.de>"

* remotes/bkoppelmann/tags/pull-tricore-20150522:
  target-tricore: add RR_DIV and RR_DIV_U instructions of the v1.6 ISA
  target-tricore: add FRET instructions of the v1.6 ISA
  target-tricore: add FCALL instructions of the v1.6 ISA
  target-tricore: add SYS_RESTORE instruction of the v1.6 ISA
  target-tricore: add RR_CRC32 instruction of the v1.6.1 ISA
  target-tricore: add SWAPMSK instructions of the v1.6.1 ISA
  target-tricore: add CMPSWP instructions of the v1.6.1 ISA
  target-tricore: Add SRC_MOV_E instruction of the v1.6 ISA
  target-tricore: introduce ISA v1.6.1 feature
  target-tricore: Add ISA v1.3.1 cpu and fix tc1796 to using v1.3

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

MAINTAINERS: Split "Block QAPI, monitor, command line" off core

Kevin and Stefan asked me to take care of this part.

Signed-off-by: Markus Armbruster <armbru@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

MAINTAINERS: Add header files to Block Layer Core section

Suggested-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Alberto Garcia <berto@igalia.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

tests: add test case for encrypted qcow2 read/write

Add a simple test case for qemu-iotests that covers read/write
with encrypted qcow2 files.

Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

qemu-io: prompt for encryption keys when required

The qemu-io tool does not check if the image is encrypted so
historically would silently corrupt the sectors by writing
plain text data into them instead of cipher text. The earlier
commit turns this mistake into a fatal abort, so check for
encryption and prompt for key when required.

This enables us to add unit tests to ensure we don't break
the ability of qemu-img to convert existing encrypted qcow2
files into a non-encrypted format.

Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

util: allow \n to terminate password input

The qemu_read_password() method looks for \r to terminate the
reading of the a password. This is what will be seen when
reading the password from a TTY. When scripting though, it is
useful to be able to send the password via a pipe, in which
case we must look for \n to terminate password input.

Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

util: move read_password method out of qemu-img into osdep/oslib

The qemu-img.c file has a read_password() method impl that is
used to prompt for passwords on the console, with impls for
POSIX and Windows. This will be needed by qemu-io.c too, so
move it into the QEMU osdep/oslib files where it can be shared
without code duplication

Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

qcow2/qcow: protect against uninitialized encryption key

When a qcow[2] file is opened, if the header reports an
encryption method, this is used to set the 'crypt_method_header'
field on the BDRVQcow[2]State struct, and the 'encrypted' flag
in the BDRVState struct.

When doing I/O operations, the 'crypt_method' field on the
BDRVQcow[2]State struct is checked to determine if encryption
needs to be applied.

The crypt_method_header value is copied into crypt_method when
the bdrv_set_key() method is called.

The QEMU code which opens a block device is expected to always
do a check

   if (bdrv_is_encrypted(bs)) {
       bdrv_set_key(bs, ....key...);
   }

If code forgets to do this, then 'crypt_method' is never set
and so when I/O is performed, QEMU writes plain text data
into a sector which is expected to contain cipher text, or
when reading, will return cipher text instead of plain
text.

Change the qcow[2] code to consult bs->encrypted when deciding
whether encryption is required, and assert(s->crypt_method)
to protect against cases where the caller forgets to set the
encryption key.

Also put an assert in the set_key methods to protect against
the case where the caller sets an encryption key on a block
device that does not have encryption

Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

qemu-iotests: Make debugging python tests easier

Adding "-d" option. The output goes to "tee" so it appears in your
console. Also, raise the verbosity of unnitest runner.

When testing a topic branch, it's possible that a bug introduced by a
code change makes the python test case hang, with debug output, it is
much easier to locate the problem.

This can also be helpful if you want to watch the progress of a python
test, it offers you a way to sense the speed of each test case method
you're writing.

Note: because there is no easy way to get *both* the verbose output and
the output expected by ./check comparison, the case would always fail
with an "output mismatch". The sole purpose of using this option is
giving developers a quick way to debug when things go wrong.

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

qemu-iotests: qemu-img info on afl VMDK image with a huge capacity

The image is contributed by Richard W.M. Jones.

Cc: Richard W.M. Jones <rjones@redhat.com>
Signed-off-by: Fam Zheng <famz@redhat.com>
Reviewed-by: Alberto Garcia <berto@igalia.com>
Reviewed-by: Markus Armbruster <armbru@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block: Detect multiplication overflow in bdrv_getlength

Bogus image may have a large total_sectors that will overflow the
multiplication. For cleanness, fix the return code so the error message
will be meaningful.

Reported-by: Richard W.M. Jones <rjones@redhat.com>
Signed-off-by: Fam Zheng <famz@redhat.com>
Reviewed-by: Alberto Garcia <berto@igalia.com>
Reviewed-by: Markus Armbruster <armbru@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

qemu-io: Use getopt() correctly

POSIX says getopt() returns -1 on completion.  While Linux happens
to define EOF as -1, this definition is not required by POSIX, and
there may be platforms where checking for EOF instead of -1 would
lead to an infinite loop.

Signed-off-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Alberto Garcia <berto@igalia.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

qcow2: style fixes in qcow2-cache.c

Fix pointer declaration to make it consistent with the rest of the
code.

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

qcow2: make qcow2_cache_put() a void function

This function never receives an invalid table pointer, so we can make
it void and remove all the error checking code.

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

qcow2: use a hash to look for entries in the L2 cache

The current cache algorithm traverses the array starting always from
the beginning, so the average number of comparisons needed to perform
a lookup is proportional to the size of the array.

By using a hash of the offset as the starting point, lookups are
faster and independent from the array size.

The hash is computed using the cluster number of the table, multiplied
by 4 to make it perform better when there are collisions.

In my tests, using a cache with 2048 entries, this reduces the average
number of comparisons per lookup from 430 to 2.5.

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

qcow2: remove qcow2_cache_find_entry_to_replace()

A cache miss means that the whole array was traversed and the entry
we were looking for was not found, so there's no need to traverse it
again in order to select an entry to replace.

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

qcow2: use an LRU algorithm to replace entries from the L2 cache

The current algorithm to evict entries from the cache gives always
preference to those in the lowest positions. As the size of the cache
increases, the chances of the later elements of being removed decrease
exponentially.

In a scenario with random I/O and lots of cache misses, entries in
positions 8 and higher are rarely (if ever) evicted. This can be seen
even with the default cache size, but with larger caches the problem
becomes more obvious.

Using an LRU algorithm makes the chances of being removed from the
cache independent from the position.

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

qcow2: simplify qcow2_cache_put() and qcow2_cache_entry_mark_dirty()

Since all tables are now stored together, it is possible to obtain
the position of a particular table directly from its address, so the
operation becomes O(1).

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

qcow2: use one single memory block for the L2/refcount cache tables

The qcow2 L2/refcount cache contains one separate table for each cache
entry. Doing one allocation per table adds unnecessary overhead and it
also requires us to store the address of each table separately.

Since the size of the cache is constant during its lifetime, it's
better to have an array that contains all the tables using one single
allocation.

In my tests measuring freshly created caches with sizes 128MB (L2) and
32MB (refcount) this uses around 10MB of RAM less.

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

vmdk: Fix overflow if l1_size is 0x20000000

Richard Jones caught this bug with afl fuzzer.

In fact, that's the only possible value to overflow (extent->l1_size =
0x20000000) l1_size:

l1_size = extent->l1_size * sizeof(long) => 0x80000000;

g_try_malloc returns NULL because l1_size is interpreted as negative
during type casting from 'int' to 'gsize', which yields a enormous
value. Hence, by coincidence, we get a "not too bad" behavior:

qemu-img: Could not open '/tmp/afl6.img': Could not open
'/tmp/afl6.img': Cannot allocate memory

Values larger than 0x20000000 will be refused by the validation in
vmdk_add_extent.

Values smaller than 0x20000000 will not overflow l1_size.

Cc: qemu-stable@nongnu.org
Reported-by: Richard W.M. Jones <rjones@redhat.com>
Signed-off-by: Fam Zheng <famz@redhat.com>
Reviewed-by: Max Reitz <mreitz@redhat.com>
Tested-by: Richard W.M. Jones <rjones@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

vmdk: Fix next_cluster_sector for compressed write

This fixes the bug introduced by commit c6ac36e (vmdk: Optimize cluster
allocation).

Sometimes, write_len could be larger than cluster size, because it
contains both data and marker.  We must advance next_cluster_sector in
this case, otherwise the image gets corrupted.

Cc: qemu-stable@nongnu.org
Reported-by: Antoni Villalonga <qemu-list@friki.cat>
Signed-off-by: Fam Zheng <famz@redhat.com>
Reviewed-by: Max Reitz <mreitz@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

nvme: support NVME_VOLATILE_WRITE_CACHE feature

The SCSI emulation in the Linux NVMe driver really wants to know
if a device has a volatile write cache.  Given that qemu has moved
away from a model where we report the backing store WCE bit to
one where the WCE bit is supposed to be part of the migratable
guest-visible state we always return 1 here.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Acked-by: Keith Busch <keith.busch@intel.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

qcow2: Flush pending discards before allocating cluster

Before a freed cluster can be reused, pending discards for this cluster
must be processed.

The original assumption was that this was not a problem because discards
are only cached during discard/write zeroes operations, which are
synchronous so that no concurrent write requests can cause cluster
allocations.

However, the discard/write zeroes operation itself can allocate a new L2
table (and it has to in order to put zero flags there), so make sure we
can cope with the situation.

This fixes https://bugs.launchpad.net/bugs/1349972.

Cc: qemu-stable@nongnu.org
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Max Reitz <mreitz@redhat.com>

target-tricore: add RR_DIV and RR_DIV_U instructions of the v1.6 ISA

Signed-off-by: Bastian Koppelmann <kbastian@mail.uni-paderborn.de>
Reviewed-by: Richard Henderson <rth@twiddle.net>

target-tricore: add FRET instructions of the v1.6 ISA

Signed-off-by: Bastian Koppelmann <kbastian@mail.uni-paderborn.de>
Reviewed-by: Richard Henderson <rth@twiddle.net>

target-tricore: add FCALL instructions of the v1.6 ISA

Signed-off-by: Bastian Koppelmann <kbastian@mail.uni-paderborn.de>
Reviewed-by: Richard Henderson <rth@twiddle.net>

target-tricore: add SYS_RESTORE instruction of the v1.6 ISA

Signed-off-by: Bastian Koppelmann <kbastian@mail.uni-paderborn.de>
Reviewed-by: Richard Henderson <rth@twiddle.net>

target-tricore: add RR_CRC32 instruction of the v1.6.1 ISA

This instruction was introduced by the new Aurix platform.

Signed-off-by: Bastian Koppelmann <kbastian@mail.uni-paderborn.de>
Reviewed-by: Richard Henderson <rth@twiddle.net>

target-tricore: add SWAPMSK instructions of the v1.6.1 ISA

Those instruction were introduced in the new Aurix platform.

Signed-off-by: Bastian Koppelmann <kbastian@mail.uni-paderborn.de>
Reviewed-by: Richard Henderson <rth@twiddle.net>

target-tricore: add CMPSWP instructions of the v1.6.1 ISA

Those instruction were introduced in the new Aurix platform.

Signed-off-by: Bastian Koppelmann <kbastian@mail.uni-paderborn.de>
Reviewed-by: Richard Henderson <rth@twiddle.net>

target-tricore: Add SRC_MOV_E instruction of the v1.6 ISA

Signed-off-by: Bastian Koppelmann <kbastian@mail.uni-paderborn.de>
Reviewed-by: Richard Henderson <rth@twiddle.net>

target-tricore: introduce ISA v1.6.1 feature

The aurix platform contains of several different cpu models and uses
the 1.6.1 ISA. This patch changes the generic aurix model to the more
specific tc27x cpu model and sets specific features.

Signed-off-by: Bastian Koppelmann <kbastian@mail.uni-paderborn.de>
Reviewed-by: Richard Henderson <rth@twiddle.net>

target-tricore: Add ISA v1.3.1 cpu and fix tc1796 to using v1.3

Signed-off-by: Bastian Koppelmann <kbastian@mail.uni-paderborn.de>
Reviewed-by: Richard Henderson <rth@twiddle.net>

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

# gpg: Signature made Fri May 22 10:00:53 2015 BST using RSA key ID 81AB73C8
# gpg: Good signature from "Stefan Hajnoczi <stefanha@redhat.com>"
# gpg:                 aka "Stefan Hajnoczi <stefanha@gmail.com>"

* remotes/stefanha/tags/block-pull-request: (38 commits)
  block: get_block_status: use "else" when testing the opposite condition
  qemu-iotests: Test unaligned sub-block zero write
  block: Fix NULL deference for unaligned write if qiov is NULL
  Revert "block: Fix unaligned zero write"
  block: align bounce buffers to page
  block: minimal bounce buffer alignment
  block: return EPERM on writes or discards to read-only devices
  configure: Add workaround for ccache and clang
  configure: silence glib unknown attribute __alloc_size__
  configure: factor out supported flag check
  configure: handle clang -nopie argument warning
  block/parallels: improve image writing performance further
  block/parallels: optimize linear image expansion
  block/parallels: add prealloc-mode and prealloc-size open paramemets
  block/parallels: delay writing to BAT till bdrv_co_flush_to_os
  block/parallels: create bat_entry_off helper
  block/parallels: improve image reading performance
  iotests, parallels: check for incorrectly closed image in tests
  block/parallels: implement incorrect close detection
  block/parallels: implement parallels_check method of block driver
  ...

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

Revert "target-alpha: Add vector implementation for CMPBGE"

This reverts commit 32ad48abd74a997220b841e4e913edeb267aa362.

Unfortunately the SSE2 code here fails to compile on some versions
of gcc:
 target-alpha/int_helper.c:77:24: error: invalid operands to binary >=
 (have '__vector(16) unsigned char' and '__vector(16) unsigned char')

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

Merge remote-tracking branch 'remotes/rth/tags/pull-axp-20150521' into staging

Rewrite fp exceptions

# gpg: Signature made Thu May 21 18:35:52 2015 BST using RSA key ID 4DD0279B
# gpg: Good signature from "Richard Henderson <rth7680@gmail.com>"
# gpg:                 aka "Richard Henderson <rth@redhat.com>"
# gpg:                 aka "Richard Henderson <rth@twiddle.net>"

* remotes/rth/tags/pull-axp-20150521:
  target-alpha: Add vector implementation for CMPBGE
  target-alpha: Rewrite helper_zapnot
  target-alpha: Raise IOV from CVTQL
  target-alpha: Suppress underflow from CVTTQ if DNZ
  target-alpha: Raise EXC_M_INV properly for fp inputs
  target-alpha: Disallow literal operand to 1C.30 to 1C.37
  target-alpha: Implement WH64EN
  target-alpha: Fix integer overflow checking insns
  target-alpha: Fix cvttq vs inf
  target-alpha: Fix cvttq vs large integers
  target-alpha: Raise IOV from CVTTQ
  target-alpha: Set EXC_M_SWC for exceptions from /S insns
  target-alpha: Set fpcr_exc_status even for disabled exceptions
  target-alpha: Tidy FPCR representation
  target-alpha: Set PC correctly for floating-point exceptions
  target-alpha: Forget installed round mode after MT_FPCR
  target-alpha: Rename floating-point subroutines
  target-alpha: Move VAX helpers to a new file

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

block: get_block_status: use "else" when testing the opposite condition

A bit of Boolean algebra (and common sense) tells us that the
second "if" here is looking for blocks that are not allocated.
This is the opposite of the "if" that sets BDRV_BLOCK_ALLOCATED,
and thus it can use an "else".

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Fam Zheng <famz@redhat.com>
Message-id: 1431599702-10431-1-git-send-email-pbonzini@redhat.com
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>

qemu-iotests: Test unaligned sub-block zero write

Test zero write in byte range 512~1024 for 4k alignment.

Signed-off-by: Fam Zheng <famz@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Message-id: 1431522721-3266-4-git-send-email-famz@redhat.com
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>

block: Fix NULL deference for unaligned write if qiov is NULL

For zero write, callers pass in NULL qiov (qemu-io "write -z" or
scsi-disk "write same").

Commit fc3959e466 fixed bdrv_co_write_zeroes which is the common case
for this bug, but it still exists in bdrv_aio_write_zeroes. A simpler
fix would be in bdrv_co_do_pwritev which is the NULL dereference point
and covers both cases.

So don't access it in bdrv_co_do_pwritev in this case, use three aligned
writes.

[Initialize ret to 0 in bdrv_co_do_zero_pwritev() to avoid uninitialized
variable warning with gcc 4.9.2.
--Stefan]

Signed-off-by: Fam Zheng <famz@redhat.com>
Message-id: 1431522721-3266-3-git-send-email-famz@redhat.com
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>

Revert "block: Fix unaligned zero write"

This reverts commit fc3959e4669a1c2149b91ccb05101cfc7ae1fc05.

The core write code already handles the case, so remove this
duplication.

Because commit 61007b316 moved the touched code from block.c to
block/io.c, the change is manually reverted.

Signed-off-by: Fam Zheng <famz@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Message-id: 1431522721-3266-2-git-send-email-famz@redhat.com
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>

block: align bounce buffers to page

The following sequence
    int fd = open(argv[1], O_RDWR | O_CREAT | O_DIRECT, 0644);
    for (i = 0; i < 100000; i++)
            write(fd, buf, 4096);
performs 5% better if buf is aligned to 4096 bytes.

The difference is quite reliable.

On the other hand we do not want at the moment to enforce bounce
buffering if guest request is aligned to 512 bytes.

The patch changes default bounce buffer optimal alignment to
MAX(page size, 4k). 4k is chosen as maximal known sector size on real
HDD.

The justification of the performance improve is quite interesting.
From the kernel point of view each request to the disk was split
by two. This could be seen by blktrace like this:
  9,0   11  1     0.000000000 11151  Q  WS 312737792 + 1023 [qemu-img]
  9,0   11  2     0.000007938 11151  Q  WS 312738815 + 8 [qemu-img]
  9,0   11  3     0.000030735 11151  Q  WS 312738823 + 1016 [qemu-img]
  9,0   11  4     0.000032482 11151  Q  WS 312739839 + 8 [qemu-img]
  9,0   11  5     0.000041379 11151  Q  WS 312739847 + 1016 [qemu-img]
  9,0   11  6     0.000042818 11151  Q  WS 312740863 + 8 [qemu-img]
  9,0   11  7     0.000051236 11151  Q  WS 312740871 + 1017 [qemu-img]
  9,0    5  1     0.169071519 11151  Q  WS 312741888 + 1023 [qemu-img]
After the patch the pattern becomes normal:
  9,0    6  1     0.000000000 12422  Q  WS 314834944 + 1024 [qemu-img]
  9,0    6  2     0.000038527 12422  Q  WS 314835968 + 1024 [qemu-img]
  9,0    6  3     0.000072849 12422  Q  WS 314836992 + 1024 [qemu-img]
  9,0    6  4     0.000106276 12422  Q  WS 314838016 + 1024 [qemu-img]
and the amount of requests sent to disk (could be calculated counting
number of lines in the output of blktrace) is reduced about 2 times.

Both qemu-img and qemu-io are affected while qemu-kvm is not. The guest
does his job well and real requests comes properly aligned (to page).

Signed-off-by: Denis V. Lunev <den@openvz.org>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Message-id: 1431441056-26198-3-git-send-email-den@openvz.org
CC: Paolo Bonzini <pbonzini@redhat.com>
CC: Kevin Wolf <kwolf@redhat.com>
CC: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>

block: minimal bounce buffer alignment

The patch introduces new concept: minimal memory alignment for bounce
buffers. Original so called "optimal" value is actually minimal required
value for aligment. It should be used for validation that the IOVec
is properly aligned and bounce buffer is not required.

Though, from the performance point of view, it would be better if
bounce buffer or IOVec allocated by QEMU will be aligned stricter.

The patch does not change any alignment value yet.

Signed-off-by: Denis V. Lunev <den@openvz.org>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Message-id: 1431441056-26198-2-git-send-email-den@openvz.org
CC: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
CC: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>

block: return EPERM on writes or discards to read-only devices

This is the behavior in the operating system, for example Linux's
blkdev_write_iter has the following:

        if (bdev_read_only(I_BDEV(bd_inode)))
                return -EPERM;

This does not apply to opening a device for read/write, when the
device only supports read-only operation.  In this case any of
EACCES, EPERM or EROFS is acceptable depending on why writing is
not possible.

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Message-id: 1431013548-22492-1-git-send-email-pbonzini@redhat.com
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>

configure: Add workaround for ccache and clang

Test if ccache is interfering with semantic analysis of macros,
disable its habit of trying to compile already pre-processed
versions of code if so. ccache attempts to save time by compiling
pre-processed versions of code, but this disturbs clang's static
analysis enough to produce false positives.

ccache allows us to disable this feature, opting instead to
compile the original version instead of its preprocessed version.
This makes ccache much slower for cache misses, but at least it
becomes usable with QEMU/clang.

This workaround only activates for users using ccache AND clang,
and only if their configuration is observed to be producing warnings.
You may need to clear your ccache for builds started without -Werror,
as those may continue to produce warnings from the cache.

Thanks to Peter Eisentraut for his writeup on the issue:
http://peter.eisentraut.org/blog/2014/12/01/ccache-and-clang-part-3/

Signed-off-by: John Snow <jsnow@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Message-id: 1427324259-1481-5-git-send-email-jsnow@redhat.com
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>

configure: silence glib unknown attribute __alloc_size__

The glib headers use GCC attributes.  Unfortunately the __GNUC__ and
__GNUC_MINOR__ version macros are also defined by clang, but clang
doesn't support the same attributes as GCC.

clang 3.5.0 does not support the __alloc_size__ attribute:

  https://github.com/llvm-mirror/clang/commit/c047507a9a79e89fc8339e074fa72822a7e7ea73

The following warning is produced:

  gstrfuncs.h:257:44: warning: unknown attribute '__alloc_size__' ignored [-Wunknown-attributes]
        G_GNUC_MALLOC G_GNUC_ALLOC_SIZE(2);
          gmacros.h:67:45: note: expanded from macro 'G_GNUC_ALLOC_SIZE'
                #define G_GNUC_ALLOC_SIZE(x) __attribute__((__alloc_size__(x)))

This patch checks whether glib headers cause warnings and disables
-Wunknown-attributes if it is able to.

Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: John Snow <jsnow@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Message-id: 1427324259-1481-4-git-send-email-jsnow@redhat.com
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>

configure: factor out supported flag check

Factor out the function that checks if a compiler
flag is supported or not.

Signed-off-by: John Snow <jsnow@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Message-id: 1427324259-1481-3-git-send-email-jsnow@redhat.com
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>

configure: handle clang -nopie argument warning

gcc 4.9.2 treats -nopie as an error:

  cc: error: unrecognized command line option ‘-nopie’

clang 3.5.0 treats -nopie as a warning:

  clang: warning: argument unused during compilation: '-nopie'

The causes ./configure to fail with clang:

  ERROR: configure test passed without -Werror but failed with -Werror.

Make the -nopie test use -Werror so that compile_prog works for both gcc
and clang.

Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: John Snow <jsnow@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Message-id: 1427324259-1481-2-git-send-email-jsnow@redhat.com
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>

block/parallels: improve image writing performance further

Try to perform IO for the biggest continuous block possible.
All blocks abscent in the image are accounted in the same type
and preallocation is made for all of them at once.

The performance for sequential write is increased from 200 Mb/sec to
235 Mb/sec on my SSD HDD.

Signed-off-by: Denis V. Lunev <den@openvz.org>
Reviewed-by: Roman Kagan <rkagan@parallels.com>
Signed-off-by: Roman Kagan <rkagan@parallels.com>
Message-id: 1430207220-24458-28-git-send-email-den@openvz.org
CC: Kevin Wolf <kwolf@redhat.com>
CC: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>

block/parallels: optimize linear image expansion

Plain image expansion spends a lot of time to update image file size.
This seriously affects the performance. The following simple test
  qemu_img create -f parallels -o cluster_size=64k ./1.hds 64G
  qemu_io -n -c "write -P 0x11 0 1024M" ./1.hds
could be improved if the format driver will pre-allocate some space
in the image file with a reasonable chunk.

This patch preallocates 128 Mb using bdrv_write_zeroes, which should
normally use fallocate() call inside. Fallback to older truncate()
could be used as a fallback using image open options thanks to the
previous patch.

The benefit is around 15%.

Signed-off-by: Denis V. Lunev <den@openvz.org>
Reviewed-by: Roman Karan <rkagan@parallels.com>
Signed-off-by: Roman Kagan <rkagan@parallels.com>
Message-id: 1430207220-24458-27-git-send-email-den@openvz.org
CC: Kevin Wolf <kwolf@redhat.com>
CC: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>

block/parallels: add prealloc-mode and prealloc-size open paramemets

This is preparational commit for tweaks in Parallels image expansion.
The idea is that enlarge via truncate by one data block is slow. It
would be much better to use fallocate via bdrv_write_zeroes and
expand by some significant amount at once.

Original idea with sequential file writing to the end of the file without
fallocate/truncate would be slower than this approach if the image is
expanded with several operations:
- each image expanding means file metadata update, i.e. filesystem
  journal write. Truncate/write to newly truncated space update file
  metadata twice thus truncate removal helps. With fallocate call
  inside bdrv_write_zeroes file metadata is updated only once and
  this should happen infrequently thus this approach is the best one
  for the image expansion
- tail writes are ordered, i.e. the guest IO queue could not be sent
  immediately to the host introducing additional IO delays

This patch just adds proper parameters into BDRVParallelsState and
performs options parsing in parallels_open.

Signed-off-by: Denis V. Lunev <den@openvz.org>
Reviewed-by: Roman Kagan <rkagan@parallels.com>
Signed-off-by: Roman Kagan <rkagan@parallels.com>
Message-id: 1430207220-24458-26-git-send-email-den@openvz.org
CC: Roman Kagan <rkagan@parallels.com>
CC: Kevin Wolf <kwolf@redhat.com>
CC: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>

block/parallels: delay writing to BAT till bdrv_co_flush_to_os

The idea is that we do not need to immediately sync BAT to the image as
from the guest point of view there is a possibility that IO is lost
even in the physical controller until flush command was finished.
bdrv_co_flush_to_os is exactly the right place for this purpose.

Technically the patch uses loaded BAT data as a cache and performs
actual on-disk metadata updates in parallels_co_flush_to_os callback.

This patch speed ups
  qemu-img create -f parallels -o cluster_size=64k ./1.hds 64G
  qemu-io -f parallels -c "write -P 0x11 0 1024k" 1.hds
writing from 50-60 Mb/sec to 80-90 Mb/sec on rotational media and
from 160 Mb/sec to 190 Mb/sec on SSD disk.

Signed-off-by: Denis V. Lunev <den@openvz.org>
Reviewed-by: Roman Kagan <rkagan@parallels.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Roman Kagan <rkagan@parallels.com>
Message-id: 1430207220-24458-25-git-send-email-den@openvz.org
CC: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>

block/parallels: create bat_entry_off helper

calculate offset of the BAT entry in the image file.

Signed-off-by: Denis V. Lunev <den@openvz.org>
Reviewed-by: Roman Kagan <rkagan@parallels.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Roman Kagan <rkagan@parallels.com>
Message-id: 1430207220-24458-24-git-send-email-den@openvz.org
CC: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>

block/parallels: improve image reading performance

Try to perform IO for the biggest continuous block possible.
The performance for sequential read is increased from 220 Mb/sec to
360 Mb/sec for continous image on my SSD HDD.

Signed-off-by: Denis V. Lunev <den@openvz.org>
Reviewed-by: Roman Kagan <rkagan@parallels.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Roman Kagan <rkagan@parallels.com>
Message-id: 1430207220-24458-23-git-send-email-den@openvz.org
CC: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>

iotests, parallels: check for incorrectly closed image in tests

Signed-off-by: Denis V. Lunev <den@openvz.org>
Reviewed-by: Roman Kagan <rkagan@parallels.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Roman Kagan <rkagan@parallels.com>
Message-id: 1430207220-24458-22-git-send-email-den@openvz.org
CC: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>

block/parallels: implement incorrect close detection

The software driver must set inuse field in Parallels header to
0x746F6E59 when the image is opened in read-write mode. The presence of
this magic in the header on open forces image consistency check.

There is an unfortunate trick here. We can not check for inuse in
parallels_check as this will happen too late. It is possible to do
that for simple check, but during the fix this would always report
an error as the image was opened in BDRV_O_RDWR mode. Thus we save
the flag in BDRVParallelsState for this.

On the other hand, nothing should be done to clear inuse in
parallels_check. Generic close will do the job right.

Signed-off-by: Denis V. Lunev <den@openvz.org>
Reviewed-by: Roman Kagan <rkagan@parallels.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Roman Kagan <rkagan@parallels.com>
Message-id: 1430207220-24458-21-git-send-email-den@openvz.org
CC: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>

block/parallels: implement parallels_check method of block driver

The check is very simple at the moment. It calculates necessary stats
and fix only the following errors:
- space leak at the end of the image. This would happens due to
  preallocation
- clusters outside the image are zeroed. Nothing else could be done here

Signed-off-by: Denis V. Lunev <den@openvz.org>
Reviewed-by: Roman Kagan <rkagan@parallels.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Roman Kagan <rkagan@parallels.com>
Message-id: 1430207220-24458-20-git-send-email-den@openvz.org
CC: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>

block/parallels: move parallels_open/probe to the very end of the file

This will help to avoid forward declarations for upcoming parallels_check

Some very obvious formatting fixes were made to the moved code to make
checkpatch happy.

Signed-off-by: Denis V. Lunev <den@openvz.org>
Reviewed-by: Roman Kagan <rkagan@parallels.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Roman Kagan <rkagan@parallels.com>
Message-id: 1430207220-24458-19-git-send-email-den@openvz.org
CC: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>

block/parallels: read parallels image header and BAT into single buffer

This metadata cache would allow to properly batch BAT updates to disk
in next patches. These updates will be properly aligned to avoid
read-modify-write transactions on block level.

Signed-off-by: Denis V. Lunev <den@openvz.org>
Reviewed-by: Roman Kagan <rkagan@parallels.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Roman Kagan <rkagan@parallels.com>
Message-id: 1430207220-24458-18-git-send-email-den@openvz.org
CC: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>

block/parallels: keep BAT bitmap data in little endian in memory

This will allow to use this data as buffer to BAT update directly
without any intermediate buffers.

Signed-off-by: Denis V. Lunev <den@openvz.org>
Reviewed-by: Roman Kagan <rkagan@parallels.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Roman Kagan <rkagan@parallels.com>
Message-id: 1430207220-24458-17-git-send-email-den@openvz.org
CC: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>

block/parallels: create bat2sect helper

deduplicate copy/paste arithmetcs

Signed-off-by: Denis V. Lunev <den@openvz.org>
Reviewed-by: Roman Kagan <rkagan@parallels.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Roman Kagan <rkagan@parallels.com>
Message-id: 1430207220-24458-16-git-send-email-den@openvz.org
CC: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>

block/parallels: rename catalog_ names to bat_

BAT means 'block allocation table'. Thus this name is clean and shorter
on writing.

Some obvious formatting fixes in the old code were made to make checkpatch
happy.

Signed-off-by: Denis V. Lunev <den@openvz.org>
Reviewed-by: Roman Kagan <rkagan@parallels.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Roman Kagan <rkagan@parallels.com>
Message-id: 1430207220-24458-15-git-send-email-den@openvz.org
CC: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>

parallels: change copyright information in the image header

Signed-off-by: Denis V. Lunev <den@openvz.org>
Reviewed-by: Roman Kagan <rkagan@parallels.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Roman Kagan <rkagan@parallels.com>
Message-id: 1430207220-24458-14-git-send-email-den@openvz.org
CC: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>

iotests, parallels: test for newly created parallels image via qemu-img

Signed-off-by: Denis V. Lunev <den@openvz.org>
Reviewed-by: Roman Kagan <rkagan@parallels.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Roman Kagan <rkagan@parallels.com>
Message-id: 1430207220-24458-13-git-send-email-den@openvz.org
CC: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>

block/parallels: support parallels image creation

Do not even care to create WithoutFreeSpace image, it is obsolete.
Always create WithouFreSpacExt one.

The code also does not spend a lot of efforts to fill cylinders and
heads fields, they are not used actually in a real life neither in
QEMU nor in Parallels products.

Signed-off-by: Denis V. Lunev <den@openvz.org>
Reviewed-by: Roman Kagan <rkagan@parallels.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Roman Kagan <rkagan@parallels.com>
Message-id: 1430207220-24458-12-git-send-email-den@openvz.org
CC: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>

iotests, parallels: test for write into Parallels image

Signed-off-by: Denis V. Lunev <den@openvz.org>
Reviewed-by: Roman Kagan <rkagan@parallels.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Roman Kagan <rkagan@parallels.com>
Message-id: 1430207220-24458-11-git-send-email-den@openvz.org
CC: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>

block/parallels: _co_writev callback for Parallels format

Support write on Parallels images. The code is almost the same as one
in the previous patch implemented scatter-gather IO for read.

Signed-off-by: Denis V. Lunev <den@openvz.org>
Reviewed-by: Roman Kagan <rkagan@parallels.com>
Signed-off-by: Roman Kagan <rkagan@parallels.com>
Message-id: 1430207220-24458-10-git-send-email-den@openvz.org
CC: Roman Kagan <rkagan@parallels.com>
CC: Kevin Wolf <kwolf@redhat.com>
CC: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>

block/parallels: mark parallels format driver as zero inited

From the guest point of view unallocated blocks are zeroed.

Signed-off-by: Denis V. Lunev <den@openvz.org>
Reviewed-by: Roman Kagan <rkagan@parallels.com>
Signed-off-by: Roman Kagan <rkagan@parallels.com>
Message-id: 1430207220-24458-9-git-send-email-den@openvz.org
CC: Roman Kagan <rkagan@parallels.com>
CC: Kevin Wolf <kwolf@redhat.com>
CC: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>

block/parallels: replace magic constants 4, 64 with proper sizeofs

simple purification..

Signed-off-by: Denis V. Lunev <den@openvz.org>
Reviewed-by: Roman Kagan <rkagan@parallels.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Roman Kagan <rkagan@parallels.com>
Message-id: 1430207220-24458-8-git-send-email-den@openvz.org
CC: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>

block/parallels: provide _co_readv routine for parallels format driver

Main approach is taken from qcow2_co_readv.

The patch drops coroutine lock for the duration of IO operation and
peforms normal scatter-gather IO using standard QEMU backend.

The patch also adds comment about locking considerations in the driver.

Signed-off-by: Denis V. Lunev <den@openvz.org>
Reviewed-by: Roman Kagan <rkagan@parallels.com>
Signed-off-by: Roman Kagan <rkagan@parallels.com>
Message-id: 1430207220-24458-7-git-send-email-den@openvz.org
CC: Roman Kagan <rkagan@parallels.com>
CC: Kevin Wolf <kwolf@redhat.com>
CC: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>

block/parallels: add get_block_status

Implement VFS method for get_block_status to Parallels format driver.

qemu_co_mutex_lock is not necessary yet (the driver is read-only) but
will be necessary very soon when write will be supported.

Signed-off-by: Roman Kagan <rkagan@parallels.com>
Reviewed-by: Denis V. Lunev <den@openvz.org>
Signed-off-by: Denis V. Lunev <den@openvz.org>
Message-id: 1430207220-24458-6-git-send-email-den@openvz.org
CC: Kevin Wolf <kwolf@redhat.com>
CC: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>

block/parallels: read up to cluster end in one go

Teach parallels_read() to do reads in coarser granularity than just a
single sector: if requested, read up to the cluster end in one go.

Signed-off-by: Roman Kagan <rkagan@parallels.com>
Reviewed-by: Denis V. Lunev <den@openvz.org>
Signed-off-by: Denis V. Lunev <den@openvz.org>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Message-id: 1430207220-24458-5-git-send-email-den@openvz.org
CC: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>

block/parallels: switch to bdrv_read

Switch the .bdrv_read method implementation from using bdrv_pread() to
bdrv_read() on the underlying file, since the latter is subject to i/o
throttling while the former is not.

Besides, since bdrv_read() operates in sectors rather than bytes, adjust
the helper functions to do so too.

Signed-off-by: Roman Kagan <rkagan@parallels.com>
Reviewed-by: Denis V. Lunev <den@openvz.org>
Signed-off-by: Denis V. Lunev <den@openvz.org>
Message-id: 1430207220-24458-4-git-send-email-den@openvz.org
CC: Kevin Wolf <kwolf@redhat.com>
CC: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>

block/parallels: rename parallels_header to ParallelsHeader

this follows QEMU coding convention

Signed-off-by: Denis V. Lunev <den@openvz.org>
Reviewed-by: Roman Kagan <rkagan@parallels.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Message-id: 1430207220-24458-3-git-send-email-den@openvz.org
CC: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>

iotests, parallels: quote TEST_IMG in 076 test to be path-safe

suggested by Jeff Cody

Signed-off-by: Denis V. Lunev <den@openvz.org>
Reviewed-by: Roman Kagan <rkagan@parallels.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Message-id: 1430207220-24458-2-git-send-email-den@openvz.org
CC: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>

target-alpha: Add vector implementation for CMPBGE

While conditionalized on SSE2, it's a "portable" gcc generic vector
implementation, which could be enabled on other hosts.

Signed-off-by: Richard Henderson <rth@twiddle.net>

target-alpha: Rewrite helper_zapnot

This form produces significantly smaller code on x86_64.

Signed-off-by: Richard Henderson <rth@twiddle.net>

Merge remote-tracking branch 'remotes/kraxel/tags/pull-vnc-20150520-1' into staging

vnc: misc fixes.

# gpg: Signature made Wed May 20 09:32:45 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-vnc-20150520-1:
  qemu-sockets: Report explicit error if unlink fails
  vnc: Tweak error when init fails
  vnc: Don't assert if opening unix socket fails
  ui: remove check for failure of qemu_acl_init()
  Strip brackets from vnc host

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

qemu-sockets: Report explicit error if unlink fails

Consider this case:

$ ls -ld ~/root-owned/
drwx--x--x. 2 root root 4096 Apr 29 12:55 /home/crobinso/root-owned/
$ ls -l ~/root-owned/foo.sock
-rwxrwxrwx. 1 crobinso crobinso 0 Apr 29 12:55 /home/crobinso/root-owned/foo.sock

$ qemu-system-x86_64 -vnc unix:~/root-owned/foo.sock
qemu-system-x86_64: -vnc unix:/home/crobinso/root-owned/foo.sock: Failed to start VNC server: Failed to bind socket to /home/crobinso/root-owned/foo.sock: Address already in use

...which is techinically true, but the real error is that we failed to
unlink. So report it.

This may seem pathological but it's a real possibility via libvirt.

Signed-off-by: Cole Robinson <crobinso@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>

vnc: Tweak error when init fails

Before:
qemu-system-x86_64: -display vnc=unix:/root/foo.sock: Failed to start VNC server on `(null)': Failed to bind socket to /root/foo.sock: Permission denied

After:
qemu-system-x86_64: -display vnc=unix:/root/foo.sock: Failed to start VNC server: Failed to bind socket to /root/foo.sock: Permission denied

Rather than tweak the string possibly show unix: value as well,
just drop the explicit display reporting. We already get the cli
string in the error message, that should be sufficient.

Signed-off-by: Cole Robinson <crobinso@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>