block: Rewrite bdrv_close_all()

This patch rewrites bdrv_close_all(): Until now, all root BDSs have been
force-closed. This is bad because it can lead to cached data not being
flushed to disk.

Instead, try to make all reference holders relinquish their reference
voluntarily:

1. All BlockBackend users are handled by making all BBs simply eject
   their BDS tree. Since a BDS can never be on top of a BB, this will
   not cause any of the issues as seen with the force-closing of BDSs.
   The references will be relinquished and any further access to the BB
   will fail gracefully.
2. All BDSs which are owned by the monitor itself (because they do not
   have a BB) are relinquished next.
3. Besides BBs and the monitor, block jobs and other BDSs are the only
   things left that can hold a reference to BDSs. After every remaining
   block job has been canceled, there should not be any BDSs left (and
   the loop added here will always terminate (as long as NDEBUG is not
   defined), because either all_bdrv_states will be empty or there will
   not be any block job left to cancel, failing the assertion).

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

block: Add blk_remove_all_bs()

When bdrv_close_all() is called, instead of force-closing all root
BlockDriverStates, it is better to just drop the reference from all
BlockBackends and let them be closed automatically. This prevents BDS
from getting closed that are still referenced by other BDS, which may
result in loss of cached data.

This patch adds a function for doing that, but does not yet incorporate
it in bdrv_close_all().

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

blockdev: Keep track of monitor-owned BDS

As a side effect, we can now make x-blockdev-del's check whether a BDS
is actually owned by the monitor explicit.

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

block: Add list of all BlockDriverStates

We need this list so that bdrv_close_all() can keep track of which BDSs
are still open after having removed the BDSs from all of the BBs and
having released all monitor BDS references.

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

block: Make bdrv_close() static

There are no users of bdrv_close() left, except for one of bdrv_open()'s
failure paths, bdrv_close_all() and bdrv_delete(), and that is good.
Make bdrv_close() static so nobody makes the mistake of directly using
bdrv_close() again.

Signed-off-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Alberto Garcia <berto@igalia.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Fam Zheng <famz@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

blockdev: Use blk_remove_bs() in do_drive_del()

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

block: Use blk_remove_bs() in blk_delete()

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

block: Remove BDS close notifier

It is unused now, so we can remove it.

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

nbd: Switch from close to eject notifier

The NBD code uses the BDS close notifier to determine when a medium is
ejected. However, now it should use the BB's BDS removal notifier for
that instead of the BDS's close notifier.

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

virtio-scsi: Catch BDS-BB removal/insertion

Make use of the BDS-BB removal and insertion notifiers to remove or set
up, respectively, virtio-scsi's op blockers.

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

virtio-blk: Functions for op blocker management

Put the code for setting up and removing op blockers into an own
function, respectively. Then, we can invoke those functions whenever a
BDS is removed from an virtio-blk BB or inserted into it.

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

block: Add BB-BDS remove/insert notifiers

bdrv_close() no longer signifies ejection of a medium, this is now done
by removing the BDS from the BB. Therefore, we want to have a notifier
for that in the BB instead of a close notifier in the BDS. The former is
added now, the latter is removed later.

Symmetrically, another notifier list is added that is invoked whenever a
BDS is inserted. We will need that for virtio-blk and virtio-scsi, which
can then remove their op blockers on BDS ejection and set them up on
insertion.

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

iotests: Add test for eject under NBD server

This patch adds a test for ejecting the BlockBackend an NBD server is
connected to (the NBD server is supposed to stop).

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

block: Release named dirty bitmaps in bdrv_close()

bdrv_delete() is not very happy about deleting BlockDriverStates with
dirty bitmaps still attached to them. In the past, we got around that
very easily by relying on bdrv_close_all() bypassing bdrv_delete(), and
bdrv_close() simply ignoring that condition. We should fix that by
releasing all named dirty bitmaps in bdrv_close() (there should not be
any unnamed bitmaps left) and moving the assertion from bdrv_delete()
there.

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

block: Remove unused struct definition BlockFinishData

Unused since 94db6d2d3.

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

iotests: Add test for a nonexistent NBD export

Trying to connect to a nonexistent NBD export should not crash the
server.

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

iotests: Make redirecting qemu's stderr optional

Redirecting qemu's stderr to stdout makes working with the stderr output
difficult due to the other file descriptor magic performed in
_launch_qemu ("ambiguous redirect").

Add an option which specifies whether stderr should be redirected to
stdout or not (allowing for other modes to be added in the future).

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

iotests: Make _filter_nbd support more URL types

This function should support URLs of the "nbd://" format (without
swallowing the export name), and for "nbd:///" URLs it should replace
"?socket=$TEST_DIR" by "?socket=TEST_DIR" because putting the Unix
socket files into the test directory makes sense.

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

iotests: Make _filter_nbd drop log lines

The NBD log lines ("/your/source/dir/nbd/xyz.c:function():line: error")
should not be converted to empty lines but removed altogether.

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

iotests: Move _filter_nbd into common.filter

_filter_nbd can be useful for other NBD tests, too, therefore it should
reside in common.filter.

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

iotests: Change coding style of _filter_nbd in 083

In order to be able to move _filter_nbd to common.filter in the next
patch, its coding style needs to be adapted to that of common.filter.
That means, we have to convert tabs to four spaces, adjust the alignment
of the last line (done with spaces already, assuming one tab equals
eight spaces), fix the line length of the comment, and add a line break
before the opening brace.

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

iotests: Rename filter_nbd to _filter_nbd in 083

In the patch after the next, this function is moved to common.filter.
Therefore, its name should be preceded by an underscore to signify its
global availability.

To keep the code motion patch clean, we cannot rename it in the same
patch, so we need to choose some order of renaming vs. motion. It is
better to keep a supposedly global function used by only a single test
in that test than to keep a supposedly local function in a common* file
and use it from a test, so we should rename the function before moving
it.

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

nbd: client_close on error in nbd_co_client_start

Use client_close() if an error in nbd_co_client_start() occurs instead
of manually inlining parts of it. This fixes an assertion error on the
server side if nbd_negotiate() fails.

Signed-off-by: Max Reitz <mreitz@redhat.com>
Acked-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

iotests: Limit supported formats for 118

Image formats used in test 118 need to support image creation.

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

vmdk: Fix converting to streamOptimized

Commit d62d9dc4b8 lifted streamOptimized images's version to 3, but we
now refuse to open version 3 images read-write.  We need to make
streamOptimized an exception to allow converting to it. This fixes the
accidentally broken iotests case 059 for the same reason.

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

block/qapi: Emit tray_open only if there is a tray

Signed-off-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Alberto Garcia <berto@igalia.com>
Message-id: 1454096953-31773-5-git-send-email-mreitz@redhat.com

Revert "hw/block/fdc: Implement tray status"

This reverts the changes that commit
2e1280e8ff95b3145bc6262accc9d447718e5318 applied to hw/block/fdc.c;
also, an additional case of drv->media_inserted use has crept in since,
which is replaced by a call to blk_is_inserted().

That commit changed tests/fdc-test.c, too, because after it, one less
TRAY_MOVED event would be emitted when executing 'change' on an empty
drive. However, now, no TRAY_MOVED events will be emitted at all, and
the tray_open status returned by query-block will always be false,
necessitating (different) changes to tests/fdc-test.c and iotest 118,
which is why this patch is not a pure revert of said commit.

Signed-off-by: Max Reitz <mreitz@redhat.com>
Message-id: 1454096953-31773-4-git-send-email-mreitz@redhat.com
Reviewed-by: Eric Blake <eblake@redhat.com>

blockdev: Fix 'change' for slot devices

'change' and related operations did not work when used on guest devices
featuring removable media but no actual tray, because
blk_dev_is_tray_open() always returned false for them and the
blockdev-{insert,remove}-medium commands required it to return true.

Fix this by making blockdev-{insert,remove}-medium work on tray-less
devices. Also, blockdev-{open,close}-tray are now explicitly no-ops when
invoked on such devices, and blk_dev_change_media_cb() is instead
called by blockdev-{insert,remove}-medium (for tray-less devices only).

Reported-by: Peter Maydell <peter.maydell@linaro.org>
Cc: qemu-stable <qemu-stable@nongnu.org>
Signed-off-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Alberto Garcia <berto@igalia.com>
Message-id: 1454096953-31773-3-git-send-email-mreitz@redhat.com
Reviewed-by: Eric Blake <eblake@redhat.com>

block: Add blk_dev_has_tray()

Pull out the check whether a block device has a tray from
blk_dev_is_tray_open() into its own function so both attributes (whether
there is a tray vs. whether that tray is open) can be queried
independently.

Cc: qemu-stable <qemu-stable@nongnu.org>
Signed-off-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Alberto Garcia <berto@igalia.com>
Message-id: 1454096953-31773-2-git-send-email-mreitz@redhat.com

Merge remote-tracking branch 'remotes/berrange/tags/pull-qcrypto-next-2016-02-02-1' into staging

Merge qcrypto-next 2016/2/2 v1

# gpg: Signature made Tue 02 Feb 2016 13:13:05 GMT using RSA key ID 15104FDF
# gpg: Good signature from "Daniel P. Berrange <dan@berrange.com>"
# gpg:                 aka "Daniel P. Berrange <berrange@redhat.com>"

* remotes/berrange/tags/pull-qcrypto-next-2016-02-02-1:
  crypto: ensure qcrypto_hash_digest_len is always defined
  crypto: register properties against the class instead of object
  crypto: fix description of @errp parameter initialization

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

Merge remote-tracking branch 'remotes/kraxel/tags/pull-ui-20160202-1' into staging

ui: gtk vc fix, adaptive sdl refresh.

# gpg: Signature made Tue 02 Feb 2016 13:06:07 GMT 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-ui-20160202-1:
  sdl: shorten the GUI refresh interval when mouse or keyboard is active
  gtk: use qemu_chr_alloc() to allocate CharDriverState

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

Merge remote-tracking branch 'remotes/kraxel/tags/pull-audio-20160202-1' into staging

audio: Clean up includes

# gpg: Signature made Tue 02 Feb 2016 12:58:06 GMT 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-audio-20160202-1:
  audio: Clean up includes

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

Merge remote-tracking branch 'remotes/kraxel/tags/pull-fwcfg-20160202-1' into staging

nvme: generate OpenFirmware device path in the "bootorder" fw_cfg file

# gpg: Signature made Tue 02 Feb 2016 12:54:04 GMT 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-fwcfg-20160202-1:
  nvme: generate OpenFirmware device path in the "bootorder" fw_cfg file

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

Merge remote-tracking branch 'remotes/elmarco/tags/ivshmem-pull-request' into staging

# gpg: Signature made Tue 02 Feb 2016 12:43:03 GMT using RSA key ID 75969CE5
# gpg: Good signature from "Marc-André Lureau <marcandre.lureau@redhat.com>"
# gpg:                 aka "Marc-André Lureau <marcandre.lureau@gmail.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: 87A9 BD93 3F87 C606 D276  F62D DAE8 E109 7596 9CE5

* remotes/elmarco/tags/ivshmem-pull-request:
  char: remove qemu_chr_open_eventfd
  ivshmem: use a single eventfd callback, get rid of CharDriver
  ivshmem: generalize ivshmem_setup_interrupts
  ivshmem-test: test both msi & irq cases
  libqos: remove some leaks
  ivshmem-test: leak fixes
  ivshmem: remove redundant assignment, fix crash with msi=off
  ivshmem: no need for opaque argument

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

sdl: shorten the GUI refresh interval when mouse or keyboard is active

Signed-off-by: Jindřich Makovička <makovick@gmail.com>
Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>

gtk: use qemu_chr_alloc() to allocate CharDriverState

The gd_vc_handler() callback is using g_malloc0() to
allocate the CharDriverState struct. As a result the
logfd field is getting initialized to 0, instead of
-1 when no logfile is requested.

The result is that when running

 $ qemu-system-i386 -nodefaults -chardev vc,id=mon0 -mon chardev=mon0

qemu duplicates all monitor output to stdout as well
as the GTK window.

Not using qemu_chr_alloc() was already a bug, but harmless
until this commit

  commit d0d7708ba29cbcc343364a46bff981e0ff88366f
  Author: Daniel P. Berrange <berrange@redhat.com>
  Date:   Mon Jan 11 12:44:41 2016 +0000

    qemu-char: add logfile facility to all chardev backends

which exposed the problem as a behaviour regression

Reported-by: Hervé Poussineau <hpoussin@reactos.org>
Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Tested-by: Hervé Poussineau <hpoussin@reactos.org>
Message-id: 1453377386-10190-1-git-send-email-berrange@redhat.com
Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>

crypto: ensure qcrypto_hash_digest_len is always defined

The qcrypto_hash_digest_len method was accidentally inside
a CONFIG_GNUTLS_HASH block, even though it doesn't depend
on gnutls. Re-arrange it to be unconditionally defined.

Reviewed-by: Fam Zheng <famz@redhat.com>
Signed-off-by: Daniel P. Berrange <berrange@redhat.com>

audio: Clean up includes

Clean up includes so that osdep.h is included first and headers
which it implies are not included manually.

This commit was created with scripts/clean-includes.

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Message-id: 1453138432-8324-1-git-send-email-peter.maydell@linaro.org
Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>

char: remove qemu_chr_open_eventfd

Broken since d0d7708ba29cbc, since the backend is NULL.

And now no longer needed by ivshmem.

Signed-off-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Reviewed-by: Markus Armbruster <armbru@redhat.com>

ivshmem: use a single eventfd callback, get rid of CharDriver

Simplify the interrupt handling by having a single callback on irq&msi
cases. Remove usage of CharDriver, replace it with
qemu_set_fd_handler(). Use event_notifier_test_and_clear() to read the
eventfd.

Before this patch, ivshmem writes the first byte received to
s->intrstatus. But ivshmem_device_spec.txt says "The status register is
set to 1 when an interrupt occurs." Fortunately, the byte usually comes
from another ivshmem device, and those always write 1.

After this commit, follows the specification, set to 1 when an interrupt
occurs.

Signed-off-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Acked-by: Markus Armbruster <armbru@redhat.com>

ivshmem: generalize ivshmem_setup_interrupts

Call ivshmem_setup_interrupts() with or without MSI, always allocate
msi_vectors that is going to be used in all case in the following patch.

Signed-off-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Reviewed-by: Markus Armbruster <armbru@redhat.com>

ivshmem-test: test both msi & irq cases

Recent commit 660c97ee introduced a regression in irq case, make
sure this code path is also tested.

Signed-off-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Reviewed-by: Markus Armbruster <armbru@redhat.com>

libqos: remove some leaks

qpci_device_find() returns allocated data, don't leak it.

Signed-off-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Reviewed-by: Markus Armbruster <armbru@redhat.com>

ivshmem-test: leak fixes

Add a cleanup_vm() function to free QPCIDevice & QPCIBus when cleaning
up the IVState.

Signed-off-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Reviewed-by: Markus Armbruster <armbru@redhat.com>

ivshmem: remove redundant assignment, fix crash with msi=off

Fix crash when msi=false introduced in 660c97ee (msi_vectors is NULL in
this case)

Signed-off-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Reviewed-by: Markus Armbruster <armbru@redhat.com>

ivshmem: no need for opaque argument

Signed-off-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Reviewed-by: Markus Armbruster <armbru@redhat.com>

nvme: generate OpenFirmware device path in the "bootorder" fw_cfg file

Background on QEMU boot indices
-------------------------------

Normally, the "bootindex" property is configured for bootable devices
with:

  DEVICE_instance_init()
    device_add_bootindex_property(..., "bootindex", ...)
      object_property_add(..., device_get_bootindex,
                          device_set_bootindex, ...)

and when the bootindex is set on the QEMU command line, with

  -device DEVICE,...,bootindex=N

the setter that was configured above is invoked:

  device_set_bootindex()
    /* parse boot index */
    visit_type_int32()

    /* verify unicity */
    check_boot_index()

    /* store parsed boot index */
    ...

    /* insert device path to boot order */
    add_boot_device_path()

In the last step, add_boot_device_path() ensures that an OpenFirmware
device path will show up in the "bootorder" fw_cfg file, at a position
corresponding to the device's boot index. Thus guest firmware (SeaBIOS and
OVMF) can try to boot off the device with the right priority.

NVMe boot index
---------------

In QEMU commit 33739c712982,

  nvma: ide: add bootindex to qom property

the following generic setters / getters:
- device_set_bootindex()
- device_get_bootindex()

were open-coded for NVMe, under the names
- nvme_set_bootindex()
- nvme_get_bootindex()

Plus nvme_instance_init() was added to configure the "bootindex" property
manually, designating the open-coded getter & setter, rather than calling
device_add_bootindex_property().

Crucially, nvme_set_bootindex() avoided the final add_boot_device_path()
call. This fact is spelled out in the message of commit 33739c712982, and
it was presumably the entire reason for all of the code duplication.

Now, Vladislav filed an RFE for OVMF
<https://github.com/tianocore/edk2/issues/48>; OVMF should boot off NVMe
devices. It is simple to build edk2's existent NvmExpressDxe driver into
OVMF, but the boot order matching logic in OVMF can only handle NVMe if
the "bootorder" fw_cfg file includes such devices.

Therefore this patch converts the NVMe device model to
device_set_bootindex() all the way.

Device paths
------------

device_set_bootindex() accepts an optional parameter called "suffix". When
present, it is expected to take the form of an OpenFirmware device path
node, and it gets appended as last node to the otherwise auto-generated
OFW path.

For NVMe, the auto-generated part is

  /pci@i0cf8/pci8086,5845@6[,1]
       ^     ^            ^  ^
       |     |            PCI slot and (present when nonzero)
       |     |            function of the NVMe controller, both hex
       |     "driver name" component, built from PCI vendor & device IDs
       PCI root at system bus port, PIO

to which here we append the suffix

  /namespace@1,0
             ^ ^
             | big endian (MSB at lowest address) numeric interpretation
             | of the 64-bit IEEE Extended Unique Identifier, aka EUI-64,
             | hex
             32-bit NVMe namespace identifier, aka NSID, hex

resulting in the OFW device path

  /pci@i0cf8/pci8086,5845@6[,1]/namespace@1,0

The reason for including the NSID and the EUI-64 is that an NVMe device
can in theory produce several different namespaces (distinguished by
NSID). Additionally, each of those may (optionally) have an EUI-64 value.

For now, QEMU only provides namespace 1.

Furthermore, QEMU doesn't even represent the EUI-64 as a standalone field;
it is embedded (and left unused) inside the "NvmeIdNs.res30" array, at the
last eight bytes. (Which is fine, since EUI-64 can be left zero-filled if
unsupported by the device.)

Based on the above, we set the "unit address" part of the last
("namespace") node to fixed "1,0".

OVMF will then map the above OFW device path to the following UEFI device
path fragment, for boot order processing:

  PciRoot(0x0)/Pci(0x6,0x1)/NVMe(0x1,00-00-00-00-00-00-00-00)
          ^        ^   ^    ^    ^   ^
          |        |   |    |    |   octets of the EUI-64 in address order
          |        |   |    |    NSID
          |        |   |    NVMe namespace messaging device path node
          |        PCI slot and function
          PCI root bridge

Cc: Keith Busch <keith.busch@intel.com> (supporter:nvme)
Cc: Kevin Wolf <kwolf@redhat.com> (supporter:Block layer core)
Cc: qemu-block@nongnu.org (open list:nvme)
Cc: Gonglei <arei.gonglei@huawei.com>
Cc: Vladislav Vovchenko <vladislav.vovchenko@sk.com>
Cc: Feng Tian <feng.tian@intel.com>
Cc: Gerd Hoffmann <kraxel@redhat.com>
Cc: Kevin O'Connor <kevin@koconnor.net>
Signed-off-by: Laszlo Ersek <lersek@redhat.com>
Acked-by: Gonglei <arei.gonglei@huawei.com>
Acked-by: Keith Busch <keith.busch@intel.com>
Tested-by: Vladislav Vovchenko <vladislav.vovchenko@sk.com>
Message-id: 1453850483-27511-1-git-send-email-lersek@redhat.com
Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>

Merge remote-tracking branch 'remotes/dgibson/tags/ppc-for-2.6-20160201' into staging

ppc patch queue for 2016-02-01

Currently accumulated patches for target-ppc, pseries machine type and
related devices.
  * Cleanup of error handling code in spapr
  * A number of fixes for Macintosh devices for the benefit of MacOS 9 and X
  * Remove some abuses of the RTAS memory access functions in spapr
  * Fixes for the gdbstub (and monitor debug) for VMX and VSX extensions.
  * Fix pseries machine hotplug memory under TCG
  * Clean up and extend handling of multiple page sizes with 64-bit hash MMUs
  * Fix to the TCG implementation of mcrfs

# gpg: Signature made Mon 01 Feb 2016 02:28:34 GMT using RSA key ID 20D9B392
# gpg: Good signature from "David Gibson <david@gibson.dropbear.id.au>"
# gpg:                 aka "David Gibson (Red Hat) <dgibson@redhat.com>"
# gpg:                 aka "David Gibson (ozlabs.org) <dgibson@ozlabs.org>"
# 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: 75F4 6586 AE61 A66C C44E  87DC 6C38 CACA 20D9 B392

* remotes/dgibson/tags/ppc-for-2.6-20160201: (40 commits)
  target-ppc: mcrfs should always update FEX/VX and only clear exception bits
  target-ppc: Make every FPSCR_ macro have a corresponding FP_ macro
  target-ppc: Allow more page sizes for POWER7 & POWER8 in TCG
  target-ppc: Helper to determine page size information from hpte alone
  target-ppc: Add new TLB invalidate by HPTE call for hash64 MMUs
  target-ppc: Split 44x tlbiva from ppc_tlb_invalidate_one()
  target-ppc: Remove unused mmu models from ppc_tlb_invalidate_one
  target-ppc: Use actual page size encodings from HPTE
  target-ppc: Rework SLB page size lookup
  target-ppc: Rework ppc_store_slb
  target-ppc: Convert mmu-hash{32,64}.[ch] from CPUPPCState to PowerPCCPU
  target-ppc: Remove unused kvmppc_read_segment_page_sizes() stub
  uninorth.c: add support for UniNorth kMacRISCPCIAddressSelect (0x48) register
  cuda.c: return error for unknown commands
  pseries: Allow TCG h_enter to work with hotplugged memory
  target-ppc: gdbstub: Add VSX support
  target-ppc: gdbstub: fix spe registers for little-endian guests
  target-ppc: gdbstub: fix altivec registers for little-endian guests
  target-ppc: gdbstub: introduce avr_need_swap()
  target-ppc: gdbstub: fix float registers for little-endian guests
  ...

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

crypto: register properties against the class instead of object

This converts the tlscredsx509, tlscredsanon and secret objects
to register their properties against the class rather than object.

Signed-off-by: Daniel P. Berrange <berrange@redhat.com>

crypto: fix description of @errp parameter initialization

The "Error **errp" parameters must be NULL initialized
not uninitialized.

Signed-off-by: Daniel P. Berrange <berrange@redhat.com>

target-ppc: mcrfs should always update FEX/VX and only clear exception bits

Here is the description of the mcrfs instruction from the PowerPC Architecture
Book, Version 2.02, Book I: PowerPC User Instruction Set Architecture
(http://www.ibm.com/developerworks/systems/library/es-archguide-v2.html), found
on page 120:

    The contents of FPSCR field BFA are copied to Condition Register field BF.
    All exception bits copied are set to 0 in the FPSCR. If the FX bit is
    copied, it is set to 0 in the FPSCR.

    Special Registers Altered:
        CR field BF
        FX OX                        (if BFA=0)
        UX ZX XX VXSNAN              (if BFA=1)
        VXISI VXIDI VXZDZ VXIMZ      (if BFA=2)
        VXVC                         (if BFA=3)
        VXSOFT VXSQRT VXCVI          (if BFA=5)

However, currently every bit in FPSCR field BFA is set to 0, including ones not
on that list.

This can be seen in the following simple C program:

    #include <fenv.h>
    #include <stdio.h>

    int main(int argc, char **argv) {
        int ret;
        ret = fegetround();
        printf("Current rounding: %d\n", ret);
        ret = fesetround(FE_UPWARD);
        printf("Setting to FE_UPWARD (%d): %d\n", FE_UPWARD, ret);
        ret = fegetround();
        printf("Current rounding: %d\n", ret);
        ret = fegetround();
        printf("Current rounding: %d\n", ret);
        return 0;
    }

which gave the output (before this commit):

    Current rounding: 0
    Setting to FE_UPWARD (2): 0
    Current rounding: 2
    Current rounding: 0

instead of (after this commit):

    Current rounding: 0
    Setting to FE_UPWARD (2): 0
    Current rounding: 2
    Current rounding: 2

The relevant disassembly is in fegetround(), which, on my system, is:

    __GI___fegetround:
    <+0>:   mcrfs  cr7, cr7
    <+4>:   mfcr   r3
    <+8>:   clrldi r3, r3, 62
    <+12>:  blr

What happens is that, the first time fegetround() is called, FPSCR field 7 is
retrieved. However, because of the bug in mcrfs, the entirety of field 7 is set
to 0, which includes the rounding mode.

There are other issues this will fix, such as condition flags not persisting
when they should if read, and if you were to read a specific field with some
exception bits set, but no others were set in the entire register, then the
bits would be cleared correctly, but FEX/VX would not be updated to 0 as they
should be.

Signed-off-by: James Clarke <jrtc27@jrtc27.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>

target-ppc: Make every FPSCR_ macro have a corresponding FP_ macro

Signed-off-by: James Clarke <jrtc27@jrtc27.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>

target-ppc: Allow more page sizes for POWER7 & POWER8 in TCG

Now that the TCG and spapr code has been extended to allow (semi-)
arbitrary page encodings in the CPU's 'sps' table, we can add the many
page sizes supported by real POWER7 and POWER8 hardware that we previously
didn't support in TCG.

Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Acked-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Reviewed-by: Alexander Graf <agraf@suse.de>

target-ppc: Helper to determine page size information from hpte alone

h_enter() in the spapr code needs to know the page size of the HPTE it's
about to insert.  Unlike other paths that do this, it doesn't have access
to the SLB, so at the moment it determines this with some open-coded
tests which assume POWER7 or POWER8 page size encodings.

To make this more flexible add ppc_hash64_hpte_page_shift_noslb() to
determine both the "base" page size per segment, and the individual
effective page size from an HPTE alone.

This means that the spapr code should now be able to handle any page size
listed in the env->sps table.

Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Acked-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Reviewed-by: Alexander Graf <agraf@suse.de>

target-ppc: Add new TLB invalidate by HPTE call for hash64 MMUs

When HPTEs are removed or modified by hypercalls on spapr, we need to
invalidate the relevant pages in the qemu TLB.

Currently we do that by doing some complicated calculations to work out the
right encoding for the tlbie instruction, then passing that to
ppc_tlb_invalidate_one()... which totally ignores the argument and flushes
the whole tlb.

Avoid that by adding a new flush-by-hpte helper in mmu-hash64.c.

Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Acked-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Reviewed-by: Alexander Graf <agraf@suse.de>

target-ppc: Split 44x tlbiva from ppc_tlb_invalidate_one()

Currently both the tlbiva instruction (used on 44x chips) and the tlbie
instruction (used on hash MMU chips) are both handled via
ppc_tlb_invalidate_one().  This is silly, because they're invoked from
different places, and do different things.

Clean this up by separating out the tlbiva instruction into its own
handling.  In fact the implementation is only a stub anyway.

Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: Laurent Vivier <lvivier@redhat.com>
Acked-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Reviewed-by: Alexander Graf <agraf@suse.de>

target-ppc: Remove unused mmu models from ppc_tlb_invalidate_one

ppc_tlb_invalidate_one() has a big switch handling many different MMU
types.  However, most of those branches can never be reached:

It is called from 3 places: from remove_hpte() and h_protect() in
spapr_hcall.c (which always has a 64-bit hash MMU type), and from
helper_tlbie() in mmu_helper.c.

Calls to helper_tlbie() are generated from gen_tlbiel, gen_tlbiel and
gen_tlbiva.  The first two are only used with the PPC_MEM_TLBIE flag,
set only with 32-bit or 64-bit hash MMU models, and gen_tlbiva() is
used only on 440 and 460 models with the BookE mmu model.

These means the exhaustive list of MMU types which may call
ppc_tlb_invalidate_one() is: POWERPC_MMU_SOFT_6xx, POWERPC_MMU_601,
POWERPC_MMU_32B, POWERPC_MMU_SOFT_74xx, POWERPC_MMU_64B, POWERPC_MMU_2_03,
POWERPC_MMU_2_06, POWERPC_MMU_2_07 and POWERPC_MMU_BOOKE.

Clean up by removing logic for all other MMU types from
ppc_tlb_invalidate_one().

This means that ppc4xx_tlb_invalidate_virt() now has no callers, or rather,
makes it obvious that it has no callers.  So, we remove that function as
well.

Signed-off-by: David Gibson <david@gibson.dropbear.id.au>

target-ppc: Use actual page size encodings from HPTE

At present the 64-bit hash MMU code uses information from the SLB to
determine the page size of a translation.  We do need that information to
correctly look up the hash table.  However the MMU also allows a
possibly larger page size to be encoded into the HPTE itself, which is used
to populate the TLB.  At present qemu doesn't check that, and so doesn't
support the MPSS "Multiple Page Size per Segment" feature.

This makes a start on allowing this, by adding an hpte_page_shift()
function which looks up the page size of an HPTE.  We use this to validate
page sizes encodings on faults, and populate the qemu TLB with larger
page sizes when appropriate.

Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Acked-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Reviewed-by: Alexander Graf <agraf@suse.de>

target-ppc: Rework SLB page size lookup

Currently, the ppc_hash64_page_shift() function looks up a page size based
on information in an SLB entry.  It open codes the bit translation for
existing CPUs, however different CPU models can have different SLB
encodings.  We already store those in the 'sps' table in CPUPPCState, but
we don't currently enforce that that actually matches the logic in
ppc_hash64_page_shift.

This patch reworks lookup of page size from SLB in several ways:
  * ppc_store_slb() will now fail (triggering an illegal instruction
    exception) if given a bad SLB page size encoding
  * On success ppc_store_slb() stores a pointer to the relevant entry in
    the page size table in the SLB entry.  This is looked up directly from
    the published table of page size encodings, so can't get out ot sync.
  * ppc_hash64_htab_lookup() and others now use this precached page size
    information rather than decoding the SLB values
  * Now that callers have easy access to the page_shift,
    ppc_hash64_pte_raddr() amounts to just a deposit64(), so remove it and
    have the callers use deposit64() directly.

Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Acked-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Reviewed-by: Alexander Graf <agraf@suse.de>

target-ppc: Rework ppc_store_slb

ppc_store_slb updates the SLB for PPC cpus with 64-bit hash MMUs.
Currently it takes two parameters, which contain values encoded as the
register arguments to the slbmte instruction, one register contains the
ESID portion of the SLBE and also the slot number, the other contains the
VSID portion of the SLBE.

We're shortly going to want to do some SLB updates from other code where
it is more convenient to supply the slot number and ESID separately, so
rework this function and its callers to work this way.

As a bonus, this slightly simplifies the emulation of segment registers for
when running a 32-bit OS on a 64-bit CPU.

Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: Laurent Vivier <lvivier@redhat.com>
Acked-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Reviewed-by: Alexander Graf <agraf@suse.de>

target-ppc: Convert mmu-hash{32,64}.[ch] from CPUPPCState to PowerPCCPU

Like a lot of places these files include a mixture of functions taking
both the older CPUPPCState *env and newer PowerPCCPU *cpu.  Move a step
closer to cleaning this up by standardizing on PowerPCCPU, except for the
helper_* functions which are called with the CPUPPCState * from tcg.

Callers and some related functions are updated as well, the boundaries of
what's changed here are a bit arbitrary.

Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: Laurent Vivier <lvivier@redhat.com>
Reviewed-by: Alexander Graf <agraf@suse.de>

target-ppc: Remove unused kvmppc_read_segment_page_sizes() stub

This stub function is in the !KVM ifdef in target-ppc/kvm_ppc.h.  However
no such function exists on the KVM side, or is ever used.

I think this originally referenced a function which read host page size
information from /proc, for we we now use the KVM GET_SMMU_INFO extension
instead.

In any case, it has no function now, so remove it.

Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: Thomas Huth <thuth@redhat.com>
Reviewed-by: Laurent Vivier <lvivier@redhat.com>
Reviewed-by: Alexander Graf <agraf@suse.de>

uninorth.c: add support for UniNorth kMacRISCPCIAddressSelect (0x48) register

Darwin/OS X use the undocumented kMacRISCPCIAddressSelect (0x48) to
configure PCI memory space size for mac99 machines. Without this
register, warnings similar to below are emitted to the console during boot:

AppleMacRiscPCI: bad range 2(80000000:01000000)
AppleMacRiscPCI: bad range 2(81000000:00001000)
AppleMacRiscPCI: bad range 2(81080000:00080000)

Based upon the algorithm in Darwin's AppleMacRiscPCI.cpp driver, set the
kMacRISCPCIAddressSelect register so that Darwin considers the PCI
memory space to be at 0x80000000 (size 0x10000000) which matches that
currently used by QEMU and OpenBIOS.

Signed-off-by: John Arbuckle <programmingkidx@gmail.com>
Tested-by: Mark Cave-Ayland <mark.cave-ayland@ilande.co.uk>
[commit message and comment revised as suggested by Mark Cave-Ayland]
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>

cuda.c: return error for unknown commands

This avoids MacsBug hanging at startup in the absence of ADB mouse
input, by replying with an error (which is also what MOL does) when
it sends an unknown command (0x1c).

Signed-off-by: Alyssa Milburn <fuzzie@fuzzie.org>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>

pseries: Allow TCG h_enter to work with hotplugged memory

The implementation of the H_ENTER hypercall for PAPR guests needs to
enforce correct access attributes on the inserted HPTE.  This means
determining if the HPTE's real address is a regular RAM address (which
requires attributes for coherent access) or an IO address (which requires
attributes for cache-inhibited access).

At the moment this check is implemented with (raddr < machine->ram_size),
but that only handles addresses in the base RAM area, not any hotplugged
RAM.

This patch corrects the problem with a new helper.

Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: Alexey Kardashevskiy <aik@ozlabs.ru>

target-ppc: gdbstub: Add VSX support

Add the XML and functions to get and set VSX registers.

Signed-off-by: Anton Blanchard <anton@samba.org>
(fixed little-endian guests)
Signed-off-by: Greg Kurz <gkurz@linux.vnet.ibm.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>

target-ppc: gdbstub: fix spe registers for little-endian guests

Let's reuse the ppc_maybe_bswap_register() helper, like we already do
with the general registers.

Signed-off-by: Greg Kurz <gkurz@linux.vnet.ibm.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>

target-ppc: gdbstub: fix altivec registers for little-endian guests

Altivec registers are 128-bit wide. They are stored in memory as two
64-bit values that must be byteswapped when the guest is little-endian.
Let's reuse the ppc_maybe_bswap_register() helper for this.

We also need to fix the ordering of the 64-bit elements according to
the target endianness, for both system and user mode.

Signed-off-by: Greg Kurz <gkurz@linux.vnet.ibm.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>

target-ppc: gdbstub: introduce avr_need_swap()

This helper will be used to support Altivec registers in little-endian guests.
This patch does not change functionnality.

Note: I had to put the helper some lines away from the gdb_*_avr_reg()
routines to get a more readable patch.

Signed-off-by: Greg Kurz <gkurz@linux.vnet.ibm.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>

target-ppc: gdbstub: fix float registers for little-endian guests

Let's reuse the ppc_maybe_bswap_register() helper, like we already do
with the general registers.

Signed-off-by: Greg Kurz <gkurz@linux.vnet.ibm.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>

target-ppc: rename and export maybe_bswap_register()

This helper will be used to support FP, Altivec and VSX registers when
the guest is little-endian.

Signed-off-by: Greg Kurz <gkurz@linux.vnet.ibm.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>

target-ppc: kvm: fix floating point registers sync on little-endian hosts

On VSX capable CPUs, the 32 FP registers are mapped to the high-bits
of the 32 first VSX registers. So if you have:

VSR31 = (uint128) 0x0102030405060708090a0b0c0d0e0f00

then

FPR31 = (uint64) 0x0102030405060708

The kernel stores the VSX registers in the fp_state struct following the
host endian element ordering.

On big-endian:

fp_state.fpr[31][0] = 0x0102030405060708
fp_state.fpr[31][1] = 0x090a0b0c0d0e0f00

On little-endian:

fp_state.fpr[31][0] = 0x090a0b0c0d0e0f00
fp_state.fpr[31][1] = 0x0102030405060708

The KVM_GET_ONE_REG and KVM_SET_ONE_REG ioctls preserve this ordering, but
QEMU considers it as big-endian and always copies element [0] to the
fpr[] array and element [1] to the vsr[] array. This does not work with
little-endian hosts, and you will get:

(qemu) p $f31
0x90a0b0c0d0e0f00

instead of:

(qemu) p $f31
0x102030405060708

This patch fixes the element ordering for little-endian hosts.

Signed-off-by: Greg Kurz <gkurz@linux.vnet.ibm.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>

pseries: Clean up error reporting in htab migration functions

The functions for migrating the hash page table on pseries machine type
(htab_save_setup() and htab_load()) can report some errors with an
explicit fprintf() before returning an appropriate error code.  Change some
of these to use error_report() instead. htab_save_setup() is omitted for
now to avoid conflicts with some other in-progress work.

Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: Thomas Huth <thuth@redhat.com>
Reviewed-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Reviewed-by: Markus Armbruster <armbru@redhat.com>

pseries: Clean up error reporting in ppc_spapr_init()

This function includes a number of explicit fprintf()s for errors.
Change these to use error_report() instead.

Also replace the single exit(EXIT_FAILURE) with an explicit exit(1), since
the latter is the more usual idiom in qemu by a large margin.

Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Reviewed-by: Markus Armbruster <armbru@redhat.com>

pseries: Clean up error handling in xics_system_init()

Use the error handling infrastructure to pass an error out from
try_create_xics() instead of assuming &error_abort - the caller is in a
better position to decide on error handling policy.

Also change the error handling from an &error_abort to &error_fatal, since
this occurs during the initial machine construction and could be triggered
by bad configuration rather than a program error.

Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: Thomas Huth <thuth@redhat.com>
Reviewed-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Reviewed-by: Markus Armbruster <armbru@redhat.com>

pseries: Clean up error handling in spapr_rtas_register()

The errors detected in this function necessarily indicate bugs in the rest
of the qemu code, rather than an external or configuration problem.

So, a simple assert() is more appropriate than any more complex error
reporting.

Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: Thomas Huth <thuth@redhat.com>
Reviewed-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Reviewed-by: Markus Armbruster <armbru@redhat.com>

pseries: Clean up error handling in spapr_vga_init()

Use error_setg() to return an error rather than an explicit exit().
Previously it was an exit(0) instead of a non-zero exit code, which was
simply a bug.  Also improve the error message.

While we're at it change the type of spapr_vga_init() to bool since that's
how we're using it anyway.

Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: Thomas Huth <thuth@redhat.com>
Reviewed-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Reviewed-by: Markus Armbruster <armbru@redhat.com>

pseries: Clean up error handling in spapr_validate_node_memory()

Use error_setg() and return an error, rather than using an explicit exit().

Also improve messages, and be more explicit about which constraint failed.

Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: Bharata B Rao <bharata@linux.vnet.ibm.com>
Reviewed-by: Thomas Huth <thuth@redhat.com>
Reviewed-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Reviewed-by: Markus Armbruster <armbru@redhat.com>

pseries: Clean up error handling of spapr_cpu_init()

Currently spapr_cpu_init() is hardcoded to handle any errors as fatal.
That works for now, since it's only called from initial setup where an
error here means we really can't proceed.

However, we'll want to handle this more flexibly for cpu hotplug in future
so generalize this using the error reporting infrastructure.  While we're
at it make a small cleanup in a related part of ppc_spapr_init() to use
error_report() instead of an old-style explicit fprintf().

Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: Bharata B Rao <bharata@linux.vnet.ibm.com>
Reviewed-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Reviewed-by: Markus Armbruster <armbru@redhat.com>

ppc: Clean up error handling in ppc_set_compat()

Current ppc_set_compat() returns -1 for errors, and also (unconditionally)
reports an error message.  The caller in h_client_architecture_support()
may then report it again using an outdated fprintf().

Clean this up by using the modern error reporting mechanisms.  Also add
strerror(errno) to the error message.

Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: Thomas Huth <thuth@redhat.com>
Reviewed-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Reviewed-by: Markus Armbruster <armbru@redhat.com>

spapr: Don't create ibm,dynamic-reconfiguration-memory w/o DR LMBs

If guest doesn't have any dynamically reconfigurable (DR) logical memory
blocks (LMB), then we shouldn't create ibm,dynamic-reconfiguration-memory
device tree node.

Signed-off-by: Bharata B Rao <bharata@linux.vnet.ibm.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>

spapr: Remove abuse of rtas_ld() in h_client_architecture_support

h_client_architecture_support() uses rtas_ld() for general purpose memory
access, despite the fact that it's not an RTAS routine at all and rtas_ld
makes things more awkward.

Clean this up by replacing rtas_ld() calls with appropriate ldXX_phys()
calls.

Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: Alexey Kardashevskiy <aik@ozlabs.ru>

spapr: Remove rtas_st_buffer_direct()

rtas_st_buffer_direct() is a not particularly useful wrapper around
cpu_physical_memory_write().  All the callers are in
rtas_ibm_configure_connector, where it's better handled by local helper.

Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: Alexey Kardashevskiy <aik@ozlabs.ru>

spapr: Small fixes to rtas_ibm_get_system_parameter, remove rtas_st_buffer

rtas_st_buffer() appears in spapr.h as though it were a widely used helper,
but in fact it is only used for saving data in a format used by
rtas_ibm_get_system_parameter().  This changes it to a local helper more
specifically for that function.

While we're there fix a couple of small defects in
rtas_ibm_get_system_parameter:
  - For the string value SPLPAR_CHARACTERISTICS, it wasn't including the
    terminating \0 in the length which it should according to LoPAPR
    7.3.16.1
  - It now checks that the supplied buffer has at least enough space for
    the length of the returned data, and returns an error if it does not.

Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: Alexey Kardashevskiy <aik@ozlabs.ru>

cuda: add missing fields to VMStateDescription

Include some fields missed from the previous VMState conversion to the
migration stream, as well as the new SR_INT delay timer.

Signed-off-by: Mark Cave-Ayland <mark.cave-ayland@ilande.co.uk>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>

mac_dbdma: add DBDMA controller state to VMStateDescription

Make sure that we include the DBDMA controller state in the migration
stream.

Signed-off-by: Mark Cave-Ayland <mark.cave-ayland@ilande.co.uk>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>

macio: add dma_active to VMStateDescription

Make sure that we include the value of dma_active in the migration stream.

Signed-off-by: Mark Cave-Ayland <mark.cave-ayland@ilande.co.uk>
Acked-by: John Snow <jsnow@redhat.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>

macio: use the existing IDEDMA aiocb to hold the active DMA aiocb

Currently the aiocb is held within MACIOIDEState, however the IDE core code
assumes that the current actvie DMA aiocb is held in aiocb in a few places,
e.g. ide_bus_reset() and ide_reset().

Switch over to using IDEDMA aiocb to store the aiocb for the current active
DMA request so that bus resets and restarts are handled correctly. As a
consequence we can now use ide_set_inactive() rather than handling its
functionality ourselves.

Signed-off-by: Mark Cave-Ayland <mark.cave-ayland@ilande.co.uk>
Reviewed-by: John Snow <jsnow@redhat.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>

target-ppc: use cpu_write_xer() helper in cpu_post_load

Otherwise some internal xer variables fail to get set post-migration.

Signed-off-by: Mark Cave-Ayland <mark.cave-ayland@ilande.co.uk>
Reviewed-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>

target-ppc: Use sensible POWER8/POWER8E versions

We never released anything older than POWER8 DD2.0 and POWER8E DD2.1,
so let's use these versions, without that some firmware or Linux code
might fail to use some HW features that were non functional in earlier
internal only spins of the chip.

Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>

hw: Clean up includes

Clean up includes so that osdep.h is included first and headers
which it implies are not included manually.

This commit was created with scripts/clean-includes.

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Message-id: 1453832250-766-38-git-send-email-peter.maydell@linaro.org

hw/core: Clean up includes

Clean up includes so that osdep.h is included first and headers
which it implies are not included manually.

This commit was created with scripts/clean-includes.

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Message-id: 1453832250-766-37-git-send-email-peter.maydell@linaro.org

arm devices: Clean up includes

Clean up includes so that osdep.h is included first and headers
which it implies are not included manually.

This commit was created with scripts/clean-includes.

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Message-id: 1453832250-766-36-git-send-email-peter.maydell@linaro.org

tilegx: Clean up includes

Clean up includes so that osdep.h is included first and headers
which it implies are not included manually.

This commit was created with scripts/clean-includes.

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Message-id: 1453832250-766-35-git-send-email-peter.maydell@linaro.org

tricore: Clean up includes

Clean up includes so that osdep.h is included first and headers
which it implies are not included manually.

This commit was created with scripts/clean-includes.

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Message-id: 1453832250-766-34-git-send-email-peter.maydell@linaro.org

moxie: Clean up includes

Clean up includes so that osdep.h is included first and headers
which it implies are not included manually.

This commit was created with scripts/clean-includes.

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Message-id: 1453832250-766-33-git-send-email-peter.maydell@linaro.org

cris: Clean up includes

Clean up includes so that osdep.h is included first and headers
which it implies are not included manually.

This commit was created with scripts/clean-includes.

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Message-id: 1453832250-766-32-git-send-email-peter.maydell@linaro.org

m68k: Clean up includes

Clean up includes so that osdep.h is included first and headers
which it implies are not included manually.

This commit was created with scripts/clean-includes.

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Message-id: 1453832250-766-31-git-send-email-peter.maydell@linaro.org

openrisc: Clean up includes

Clean up includes so that osdep.h is included first and headers
which it implies are not included manually.

This commit was created with scripts/clean-includes.

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Message-id: 1453832250-766-30-git-send-email-peter.maydell@linaro.org

xtensa: Clean up includes

Clean up includes so that osdep.h is included first and headers
which it implies are not included manually.

This commit was created with scripts/clean-includes.

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Message-id: 1453832250-766-29-git-send-email-peter.maydell@linaro.org