sdk/emulator/qemu.git
11 years agoblock: fix bdrv_read_unthrottled()
Peter Lieven [Thu, 18 Jul 2013 08:37:32 +0000 (10:37 +0200)]
block: fix bdrv_read_unthrottled()

Signed-off-by: Peter Lieven <pl@kamp.de>
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
11 years agocpus: Let vm_stop[_force_state]() always flush block devices
Kevin Wolf [Thu, 18 Jul 2013 12:52:19 +0000 (14:52 +0200)]
cpus: Let vm_stop[_force_state]() always flush block devices

Even if the VM is already stopped, we cannot assume that all data has
already been successfully flushed to disk. The flush during the previous
vm_stop() could have failed.

Run bdrv_flush_all() unconditionally so that we get an error each time
if the block device isn't really flushed.

Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
11 years agoblock-migration: efficiently encode zero blocks
Peter Lieven [Thu, 18 Jul 2013 07:48:50 +0000 (09:48 +0200)]
block-migration: efficiently encode zero blocks

this patch adds a efficient encoding for zero blocks by
adding a new flag indicating a block is completely zero.

additionally bdrv_write_zeros() is used at the destination
to efficiently write these zeroes. depending on the implementation
this avoids that the destination target gets fully provisioned.

Signed-off-by: Peter Lieven <pl@kamp.de>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
11 years agoblock/raw: add bdrv_co_write_zeroes
Peter Lieven [Thu, 11 Jul 2013 12:16:23 +0000 (14:16 +0200)]
block/raw: add bdrv_co_write_zeroes

Signed-off-by: Peter Lieven <pl@kamp.de>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
11 years agoblock: add bdrv_write_zeroes()
Peter Lieven [Thu, 11 Jul 2013 12:16:22 +0000 (14:16 +0200)]
block: add bdrv_write_zeroes()

Signed-off-by: Peter Lieven <pl@kamp.de>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
11 years agoblock: fix vvfat error path for enable_write_target
Fam Zheng [Wed, 17 Jul 2013 09:57:37 +0000 (17:57 +0800)]
block: fix vvfat error path for enable_write_target

s->qcow and s->qcow_filename are allocated but not freed on error. Fix the
possible leaks, remove unnecessary check for bdrv_new(), propagate ret code of
bdrv_create() and also the one of enable_write_target().

Signed-off-by: Fam Zheng <famz@redhat.com>
Reviewed-by: Laszlo Ersek <lersek@redhat.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
11 years agoQEMUBH: make AioContext's bh re-entrant
Liu Ping Fan [Tue, 16 Jul 2013 04:28:58 +0000 (12:28 +0800)]
QEMUBH: make AioContext's bh re-entrant

BH will be used outside big lock, so introduce lock to protect
between the writers, ie, bh's adders and deleter. The lock only
affects the writers and bh's callback does not take this extra lock.
Note that for the same AioContext, aio_bh_poll() can not run in
parallel yet.

Signed-off-by: Liu Ping Fan <pingfank@linux.vnet.ibm.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
11 years agodataplane: sync virtio.c and vring.c virtqueue state
Stefan Hajnoczi [Mon, 1 Jul 2013 11:29:17 +0000 (13:29 +0200)]
dataplane: sync virtio.c and vring.c virtqueue state

Load the virtio.c state into vring.c when we start dataplane mode and
vice versa when stopping dataplane mode.  This patch makes it possible
to start and stop dataplane any time while the guest is running.

This will eventually allow us to go back to QEMU main loop for
bdrv_drain_all() and live migration.  In the meantime, this patch makes
the dataplane lifecycle more robust but should make no visible
difference.  It may be useful in the virtio-net dataplane effort.

Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
11 years agogluster: Add discard support for GlusterFS block driver.
Bharata B Rao [Tue, 16 Jul 2013 16:17:42 +0000 (21:47 +0530)]
gluster: Add discard support for GlusterFS block driver.

Implement bdrv_aio_discard for gluster.

Signed-off-by: Bharata B Rao <bharata@linux.vnet.ibm.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
11 years agogluster: Use pkg-config to configure GlusterFS block driver
Bharata B Rao [Tue, 16 Jul 2013 16:17:41 +0000 (21:47 +0530)]
gluster: Use pkg-config to configure GlusterFS block driver

Use pkg-config to determine the version and library dependency
for GlusterFS block driver.

Signed-off-by: Bharata B Rao <bharata@linux.vnet.ibm.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
11 years agoboot-order-test: Add tests for Sun4u
Markus Armbruster [Wed, 26 Jun 2013 13:52:23 +0000 (15:52 +0200)]
boot-order-test: Add tests for Sun4u

Cc: Blue Swirl <blauwirbel@gmail.com>
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Message-id: 1372254743-15808-13-git-send-email-armbru@redhat.com
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
11 years agolibqos: Generalize I/O-mapped fw_cfg
Markus Armbruster [Wed, 26 Jun 2013 13:52:22 +0000 (15:52 +0200)]
libqos: Generalize I/O-mapped fw_cfg

Provide a constructor that takes the base address in addition to the
PC-specific one.

Signed-off-by: Markus Armbruster <armbru@redhat.com>
Message-id: 1372254743-15808-12-git-send-email-armbru@redhat.com
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
11 years agoboot-order-test: Add tests for Sun4m
Markus Armbruster [Wed, 26 Jun 2013 13:52:21 +0000 (15:52 +0200)]
boot-order-test: Add tests for Sun4m

Cc: Blue Swirl <blauwirbel@gmail.com>
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Message-id: 1372254743-15808-11-git-send-email-armbru@redhat.com
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
11 years agoboot-order-test: Add tests for PowerPC PREP
Markus Armbruster [Wed, 26 Jun 2013 13:52:20 +0000 (15:52 +0200)]
boot-order-test: Add tests for PowerPC PREP

Cc: Andreas Färber <afaerber@suse.de>
Cc: Alexander Graf <agraf@suse.de>
Cc: qemu-ppc@nongnu.org
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Message-id: 1372254743-15808-10-git-send-email-armbru@redhat.com
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
11 years agoboot-order-test: Code motion for better readability
Markus Armbruster [Wed, 26 Jun 2013 13:52:19 +0000 (15:52 +0200)]
boot-order-test: Code motion for better readability

Cc: Andreas Färber <afaerber@suse.de>
Cc: Alexander Graf <agraf@suse.de>
Cc: qemu-ppc@nongnu.org
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Message-id: 1372254743-15808-9-git-send-email-armbru@redhat.com
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
11 years agoboot-order-test: Better separate target-specific and generic parts
Markus Armbruster [Wed, 26 Jun 2013 13:52:18 +0000 (15:52 +0200)]
boot-order-test: Better separate target-specific and generic parts

The initial version did just PC.  I didn't bother to separate out
generic parts, because I don't like to abstract from a single case.

Now we have two cases, PC and PowerMac, and I'm about to add more.
Time to do it right.

To ease review, this commit changes the code in-place, and the next
commit reorders it for better readability.

Cc: Andreas Färber <afaerber@suse.de>
Cc: Alexander Graf <agraf@suse.de>
Cc: qemu-ppc@nongnu.org
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Message-id: 1372254743-15808-8-git-send-email-armbru@redhat.com
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
11 years agoboot-order-test: Cover -boot once in ppc tests
Markus Armbruster [Wed, 26 Jun 2013 13:52:17 +0000 (15:52 +0200)]
boot-order-test: Cover -boot once in ppc tests

Cc: Andreas Färber <afaerber@suse.de>
Cc: Alexander Graf <agraf@suse.de>
Cc: qemu-ppc@nongnu.org
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Message-id: 1372254743-15808-7-git-send-email-armbru@redhat.com
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
11 years agoboot-order-test: Add tests for PowerMacs
Andreas Färber [Wed, 26 Jun 2013 13:52:16 +0000 (15:52 +0200)]
boot-order-test: Add tests for PowerMacs

They set the boot device via fw_cfg, which is then translated to a boot
path of "hd" or "cd" in OpenBIOS.

Signed-off-by: Andreas Färber <afaerber@suse.de>
Signed-off-by: Andreas Färber <afaerber@suse.de>
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Message-id: 1372254743-15808-6-git-send-email-armbru@redhat.com
Cc: Alexander Graf <agraf@suse.de>
Cc: qemu-ppc@nongnu.org
Converted to libqos/fw_cfg on Anthony's request.
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
11 years agolibqos: Add support for memory-mapped fw_cfg
Markus Armbruster [Wed, 26 Jun 2013 13:52:15 +0000 (15:52 +0200)]
libqos: Add support for memory-mapped fw_cfg

Signed-off-by: Markus Armbruster <armbru@redhat.com>
Message-id: 1372254743-15808-5-git-send-email-armbru@redhat.com
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
11 years agolibqos: include dependencies
Anthony Liguori [Wed, 26 Jun 2013 13:52:14 +0000 (15:52 +0200)]
libqos: include dependencies

Otherwise rebuilds can fail when libqos is modified.

Reported-by: Markus Armbruster <armbru@redhat.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Message-id: 1372254743-15808-4-git-send-email-armbru@redhat.com
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
11 years agoboot-order-test: New; covering just PC for now
Markus Armbruster [Wed, 26 Jun 2013 13:52:13 +0000 (15:52 +0200)]
boot-order-test: New; covering just PC for now

Signed-off-by: Markus Armbruster <armbru@redhat.com>
Message-id: 1372254743-15808-3-git-send-email-armbru@redhat.com
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
11 years agoqtest: Don't reset on qtest chardev connect
Markus Armbruster [Wed, 26 Jun 2013 13:52:12 +0000 (15:52 +0200)]
qtest: Don't reset on qtest chardev connect

libqtest's qtest_init() connecting to the qtest socket triggers reset.
This was coded in the hope we could use the same QEMU process for
multiple tests that way.  Never used.  Injects an extra reset even
when it's not used, and that can mess up tests such as the one of
-boot once I'm about to add.  Drop it.

Signed-off-by: Markus Armbruster <armbru@redhat.com>
Message-id: 1372254743-15808-2-git-send-email-armbru@redhat.com
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
11 years agolibqtest: New qtest_end() to go with qtest_start()
Markus Armbruster [Thu, 20 Jun 2013 06:55:29 +0000 (08:55 +0200)]
libqtest: New qtest_end() to go with qtest_start()

Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Andreas Färber <afaerber@suse.de>
Reviewed-by: Anthony Liguori <aliguori@us.ibm.com>
Message-id: 1371711329-9144-3-git-send-email-armbru@redhat.com
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
11 years agolibqtest: Plug fd and memory leaks in qtest_quit()
Markus Armbruster [Thu, 20 Jun 2013 06:55:28 +0000 (08:55 +0200)]
libqtest: Plug fd and memory leaks in qtest_quit()

Reviewed-by: Anthony Liguori <aliguori@us.ibm.com>
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Andreas Färber <afaerber@suse.de>
Message-id: 1371711329-9144-2-git-send-email-armbru@redhat.com
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
11 years agomonitor: maintain at most one G_IO_OUT watch
Laszlo Ersek [Tue, 16 Jul 2013 18:19:41 +0000 (20:19 +0200)]
monitor: maintain at most one G_IO_OUT watch

When monitor_flush() is invoked repeatedly outside the monitor_unblocked()
callback, for example from tlb_info() -> ... -> print_pte(), several
watches may be added for the same event.

This is no problem per se because the extra monitor_unblocked() callbacks
are harmless if mon->outbuf is empty, the watches will be removed
gradually. However a big number of watches can grow "gpollfds" without
limit in glib_pollfds_fill(), triggering a -1/EINVAL condition in
g_poll().

Keep at most one such watch, by following the pattern observable in eg.
commits c874ea97 and c3d6b96e. The change has no effect when
monitor_unblocked() calls monitor_flush() (when the watch can either be
removed or renewed 1-for-1), but non-callback contexts won't create an
additional watch when the monitor already has one.

Related RHBZ: https://bugzilla.redhat.com/show_bug.cgi?id=970047

Signed-off-by: Laszlo Ersek <lersek@redhat.com>
Reviewed-by: Amit Shah <amit.shah@redhat.com>
Reviewed-by: Anthony Liguori <aliguori@us.ibm.com>
Message-id: 1373998781-29561-3-git-send-email-lersek@redhat.com
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
11 years agochar: io_channel_send: don't lose written bytes
Laszlo Ersek [Tue, 16 Jul 2013 18:19:40 +0000 (20:19 +0200)]
char: io_channel_send: don't lose written bytes

The g_io_channel_write_chars() documentation states,

  bytes_written: The number of bytes written. This can be nonzero even if
                 the return value is not G_IO_STATUS_NORMAL. [...]

io_channel_send() could lose such bytes before.

Furthermore, the (status == G_IO_STATUS_EOF) condition used to evaluate to
constant false whenever it was reached. When that condition actually held,
it always led to -1 / EINVAL. This patch (almost) distinguishes
G_IO_STATUS_EOF only when no bytes have been written, and then treats it
as an error.

Signed-off-by: Laszlo Ersek <lersek@redhat.com>
Reviewed-by: Amit Shah <amit.shah@redhat.com>
Reviewed-by: Anthony Liguori <aliguori@us.ibm.com>
Message-id: 1373998781-29561-2-git-send-email-lersek@redhat.com
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
11 years agoconfigure: Provide more helpful message if libvte not present
Peter Maydell [Thu, 18 Jul 2013 15:42:01 +0000 (16:42 +0100)]
configure: Provide more helpful message if libvte not present

If the system has GTK but not libvte, it's more helpful to
tell the user that libvte is missing than to simply say that
GTK is not present.

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Message-id: 1374162121-31582-1-git-send-email-peter.maydell@linaro.org
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
11 years agoMerge remote-tracking branch 'awilliam/tags/vfio-for-qemu-1.6' into staging
Anthony Liguori [Thu, 18 Jul 2013 13:14:22 +0000 (08:14 -0500)]
Merge remote-tracking branch 'awilliam/tags/vfio-for-qemu-1.6' into staging

vfio: enhanced VGA quirks + AER error containment

# gpg: Signature made Wed 17 Jul 2013 05:39:38 PM CDT using RSA key ID 3BB08B22
# gpg: Can't check signature: public key not found

# By Alex Williamson (1) and Vijay Mohan Pandarathil (1)
# Via Alex Williamson
* awilliam/tags/vfio-for-qemu-1.6:
  vfio: QEMU-AER: Qemu changes to support AER for VFIO-PCI devices
  vfio-pci: VGA quirk update

Message-id: 20130717224939.4763.87264.stgit@bling.home
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
11 years agoMerge remote-tracking branch 'luiz/queue/qmp' into staging
Anthony Liguori [Thu, 18 Jul 2013 13:14:10 +0000 (08:14 -0500)]
Merge remote-tracking branch 'luiz/queue/qmp' into staging

# By Amos Kong (1) and Luiz Capitulino (1)
# Via Luiz Capitulino
* luiz/queue/qmp:
  qmp: update send-key document
  qapi: qapi-commands: fix possible leaks on visitor dealloc

Message-id: 1374093679-29213-1-git-send-email-lcapitulino@redhat.com
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
11 years agoMerge remote-tracking branch 'bonzini/scsi-next' into staging
Anthony Liguori [Thu, 18 Jul 2013 13:13:50 +0000 (08:13 -0500)]
Merge remote-tracking branch 'bonzini/scsi-next' into staging

# By Peter Lieven (4) and Ronnie Sahlberg (1)
# Via Paolo Bonzini
* bonzini/scsi-next:
  iscsi: factor out sector conversions
  iscsi: assert that sectors are aligned to LUN blocksize
  iscsi: remove support for misaligned nb_sectors in aio_readv
  iscsi: fix -ENOSPC in iscsi_create()
  Fix iSCSI crash on SG_IO with an iovector

Message-id: 1374073524-8469-1-git-send-email-pbonzini@redhat.com
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
11 years agoMerge remote-tracking branch 'mst/tags/for_anthony' into staging
Anthony Liguori [Thu, 18 Jul 2013 13:12:47 +0000 (08:12 -0500)]
Merge remote-tracking branch 'mst/tags/for_anthony' into staging

pci,net,pc enhancements

This includes some fixes and enhancements that accumulated in my tree:
pci fixes by dkoch, virtio-net enhancements by akong and mst,
and a fix for xen pc by mst.

Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
# gpg: Signature made Wed 17 Jul 2013 04:44:45 AM CDT using RSA key ID D28D5469
# gpg: Can't check signature: public key not found

# By Don Koch (2) and others
# Via Michael S. Tsirkin
* mst/tags/for_anthony:
  pc: don't access fw cfg if NULL
  virtio-net: add feature bit for any header s/g
  net: add support of mac-programming over macvtap in QEMU side
  pci: fix BRDIGE typo
  pci-bridge: update mappings for migration/restore

Message-id: 1374054430-21966-1-git-send-email-mst@redhat.com
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
11 years agoqmp: update send-key document
Amos Kong [Tue, 16 Jul 2013 11:52:14 +0000 (19:52 +0800)]
qmp: update send-key document

commit 9f328977 changes qmp_send_key() to accept key codes in hex,
but the document wasn't updated. The items of keys list is union
now, not enum.

Signed-off-by: Amos Kong <akong@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Luiz Capitulino <lcapitulino@redhat.com>
11 years agoiscsi: factor out sector conversions
Peter Lieven [Thu, 11 Jul 2013 12:16:25 +0000 (14:16 +0200)]
iscsi: factor out sector conversions

Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Peter Lieven <pl@kamp.de>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
11 years agoiscsi: assert that sectors are aligned to LUN blocksize
Peter Lieven [Thu, 11 Jul 2013 12:16:27 +0000 (14:16 +0200)]
iscsi: assert that sectors are aligned to LUN blocksize

if the blocksize of an iSCSI LUN is bigger than the BDRV_SECTOR_SIZE
it is possible that sector_num or nb_sectors are not correctly
aligned.

to avoid corruption we fail requests which are misaligned.

Signed-off-by: Peter Lieven <pl@kamp.de>
Cc: qemu-stable@nongnu.org
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
11 years agoiscsi: remove support for misaligned nb_sectors in aio_readv
Peter Lieven [Thu, 11 Jul 2013 12:16:26 +0000 (14:16 +0200)]
iscsi: remove support for misaligned nb_sectors in aio_readv

this hask is not working (anymore). support for misaligned offsets should
be handled at the block layer.

Signed-off-by: Peter Lieven <pl@kamp.de>
Cc: qemu-stable@nongnu.org
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
11 years agoiscsi: fix -ENOSPC in iscsi_create()
Peter Lieven [Thu, 11 Jul 2013 12:16:24 +0000 (14:16 +0200)]
iscsi: fix -ENOSPC in iscsi_create()

the -ENOPSC case did not work due to the missing goto.

Reported-by: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Peter Lieven <pl@kamp.de>
Cc: qemu-stable@nongnu.org
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
11 years agoFix iSCSI crash on SG_IO with an iovector
Ronnie Sahlberg [Sun, 23 Jun 2013 15:07:08 +0000 (08:07 -0700)]
Fix iSCSI crash on SG_IO with an iovector

Don't assume that SG_IO is always invoked with a simple buffer,
check the iovec_count and if it is >= 1 then we need to pass an array
of iovectors to libiscsi instead of just a plain buffer.

Signed-off-by: Ronnie Sahlberg <ronniesahlberg@gmail.com>
Cc: qemu-stable@nongnu.org
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
11 years agoqapi: qapi-commands: fix possible leaks on visitor dealloc
Luiz Capitulino [Thu, 11 Jul 2013 18:26:56 +0000 (14:26 -0400)]
qapi: qapi-commands: fix possible leaks on visitor dealloc

In qmp-marshal.c the dealloc visitor calls use the same errp
pointer of the input visitor calls. This means that if any of
the input visitor calls fails, then the dealloc visitor will
return early, before freeing the object's memory.

Here's an example, consider this code:

int qmp_marshal_input_block_passwd(Monitor *mon, const QDict *qdict, QObject **ret)
{
[...]

    char * device = NULL;
    char * password = NULL;

    mi = qmp_input_visitor_new_strict(QOBJECT(args));
    v = qmp_input_get_visitor(mi);
    visit_type_str(v, &device, "device", errp);
    visit_type_str(v, &password, "password", errp);
    qmp_input_visitor_cleanup(mi);

    if (error_is_set(errp)) {
        goto out;
    }
    qmp_block_passwd(device, password, errp);

out:
    md = qapi_dealloc_visitor_new();
    v = qapi_dealloc_get_visitor(md);
    visit_type_str(v, &device, "device", errp);
    visit_type_str(v, &password, "password", errp);
    qapi_dealloc_visitor_cleanup(md);

[...]

    return 0;
}

Consider errp != NULL when the out label is reached, we're going
to leak device and password.

This patch fixes this by always passing errp=NULL for dealloc
visitors, meaning that we always try to free them regardless of
any previous failure. The above example would then be:

out:
    md = qapi_dealloc_visitor_new();
    v = qapi_dealloc_get_visitor(md);
    visit_type_str(v, &device, "device", NULL);
    visit_type_str(v, &password, "password", NULL);
    qapi_dealloc_visitor_cleanup(md);

Signed-off-by: Luiz Capitulino <lcapitulino@redhat.com>
Reviewed-by: Laszlo Ersek <lersek@redhat.com>
Reviewed-by: Michael Roth <mdroth@linux.vnet.ibm.com>
11 years agovfio: QEMU-AER: Qemu changes to support AER for VFIO-PCI devices
Vijay Mohan Pandarathil [Mon, 15 Jul 2013 21:49:49 +0000 (15:49 -0600)]
vfio: QEMU-AER: Qemu changes to support AER for VFIO-PCI devices

Add support for error containment when a VFIO device assigned to a KVM
guest encounters an error. This is for PCIe devices/drivers that support AER
functionality. When the host OS is notified of an error in a device either
through the firmware first approach or through an interrupt handled by the AER
root port driver, the error handler registered by the vfio-pci driver gets
invoked. The qemu process is signaled through an eventfd registered per
VFIO device by the qemu process. In the eventfd handler, qemu decides on
what action to take. In this implementation, guest is brought down to
contain the error.

The kernel patches for the above functionality has been already accepted.

This is a refresh of the QEMU patch which was reviewed earlier.
http://marc.info/?l=linux-kernel&m=136281557608087&w=2
This patch has the same contents and has been built after refreshing
to latest upstream and after the linux headers have been updated in qemu.

- Create eventfd per vfio device assigned to a guest and register an
          event handler

- This fd is passed to the vfio_pci driver through the SET_IRQ ioctl

- When the device encounters an error, the eventfd is signalled
          and the qemu eventfd handler gets invoked.

- In the handler decide what action to take. Current action taken
          is to stop the guest.

Signed-off-by: Vijay Mohan Pandarathil <vijaymohan.pandarathil@hp.com>
Signed-off-by: Alex Williamson <alex.williamson@redhat.com>
11 years agovfio-pci: VGA quirk update
Alex Williamson [Mon, 15 Jul 2013 21:48:11 +0000 (15:48 -0600)]
vfio-pci: VGA quirk update

Turns out all the suspicions for AMD devices were correct, everywhere
we read a BAR address that the address matches the config space offset,
there's full access to PCI config space.  Attempt to generalize some
helpers to allow quirks to easily be added for mirrors and windows.
Also fill in complete config space for AMD.

Signed-off-by: Alex Williamson <alex.williamson@redhat.com>
11 years agoMerge remote-tracking branch 'quintela/migration.next' into staging
Anthony Liguori [Mon, 15 Jul 2013 19:49:15 +0000 (14:49 -0500)]
Merge remote-tracking branch 'quintela/migration.next' into staging

# By Chegu Vinod
# Via Juan Quintela
* quintela/migration.next:
  Force auto-convegence of live migration
  Add 'auto-converge' migration capability
  Introduce async_run_on_cpu()

Message-id: 1373664508-5404-1-git-send-email-quintela@redhat.com
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
11 years agoMerge remote-tracking branch 'cohuck/virtio-ccw-upstr' into staging
Anthony Liguori [Mon, 15 Jul 2013 19:02:41 +0000 (14:02 -0500)]
Merge remote-tracking branch 'cohuck/virtio-ccw-upstr' into staging

# By Dominik Dingel
# Via Cornelia Huck
* cohuck/virtio-ccw-upstr:
  virtio-ccw: Enable x-data-plane for virtio-ccw-blk

Message-id: 1373903207-27085-1-git-send-email-cornelia.huck@de.ibm.com
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
11 years agoMerge remote-tracking branch 'pmaydell/tags/pull-target-arm-20130715-1' into staging
Anthony Liguori [Mon, 15 Jul 2013 19:02:32 +0000 (14:02 -0500)]
Merge remote-tracking branch 'pmaydell/tags/pull-target-arm-20130715-1' into staging

target-arm queue

# gpg: Signature made Mon 15 Jul 2013 11:15:13 AM CDT using RSA key ID 14360CDE
# gpg: Can't check signature: public key not found

# By Mans Rullgard (3) and others
# Via Peter Maydell
* pmaydell/tags/pull-target-arm-20130715-1:
  target-arm: Avoid g_hash_table_get_keys()
  target-arm: avoid undefined behaviour when writing TTBCR
  target-arm/helper.c: Allow const opaques in arm CP
  target-arm/helper.c: Implement MIDR aliases
  target-arm/helper.c: OMAP/StrongARM cp15 crn=0 cleanup
  target-arm: explicitly decode SEVL instruction
  target-arm: implement LDA/STL instructions
  target-arm: add feature flag for ARMv8

Message-id: 1373905022-27735-1-git-send-email-peter.maydell@linaro.org
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
11 years agoMerge remote-tracking branch 'pmaydell/tags/pull-arm-devs-20130715' into staging
Anthony Liguori [Mon, 15 Jul 2013 19:02:12 +0000 (14:02 -0500)]
Merge remote-tracking branch 'pmaydell/tags/pull-arm-devs-20130715' into staging

arm-devs queue

# gpg: Signature made Mon 15 Jul 2013 10:53:44 AM CDT using RSA key ID 14360CDE
# gpg: Can't check signature: public key not found

# By Peter Maydell (4) and others
# Via Peter Maydell
* pmaydell/tags/pull-arm-devs-20130715:
  ARM/highbank: add support for Calxeda ECX-2000 / Midway
  ARM/highbank: prepare for adding similar machines
  hw/arm/vexpress: Add alias for flash at address 0 on A15 board
  hw/dma/omap_dma: Fix bugs with DMA requests above 32
  sd/pl181.c: Avoid undefined shift behaviour in RWORD macro
  hw/cpu/a15mpcore: Correct default value for num-irq
  char/cadence_uart: Fix reset for unattached instances

Message-id: 1373904095-27592-1-git-send-email-peter.maydell@linaro.org
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
11 years agoMerge remote-tracking branch 'rth/axp-next' into staging
Anthony Liguori [Mon, 15 Jul 2013 19:00:57 +0000 (14:00 -0500)]
Merge remote-tracking branch 'rth/axp-next' into staging

# By Richard Henderson
# Via Richard Henderson
* rth/axp-next:
  hw/alpha: Use SRM epoch
  hw/alpha: Drop latch_tmp hack
  exec: Support 64-bit operations in address_space_rw
  hw/alpha: Don't machine check on missing pci i/o
  hw/alpha: Don't use get_system_io

Message-id: 1373840171-25556-1-git-send-email-rth@twiddle.net
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
11 years agoMerge remote-tracking branch 'kwolf/for-anthony' into staging
Anthony Liguori [Mon, 15 Jul 2013 19:00:32 +0000 (14:00 -0500)]
Merge remote-tracking branch 'kwolf/for-anthony' into staging

# By Kevin Wolf (6) and Stefan Hajnoczi (2)
# Via Kevin Wolf
* kwolf/for-anthony:
  ahci: Fix FLUSH command
  migration: Fail migration on bdrv_flush_all() error
  cpus: Add return value for vm_stop()
  block: Add return value for bdrv_flush_all()
  qemu-iotests: Update 051 reference output
  block: Don't parse protocol from file.filename
  block: add drive_backup HMP command
  blockdev: add sync mode to drive-backup QMP command

Message-id: 1373887000-4488-1-git-send-email-kwolf@redhat.com
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
11 years agopc: don't access fw cfg if NULL
Michael S. Tsirkin [Wed, 10 Jul 2013 15:28:04 +0000 (18:28 +0300)]
pc: don't access fw cfg if NULL

commit f8c457b88d72a48989f190bc3d7b79f4f3b7d11c
     "pc: pass PCI hole ranges to Guests"
broke Xen as it has no fw_cfg.
Check for this configuration and boil out.

Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Tested-by: Stefano Stabellini <stefano.stabellini@eu.citrix.com>
11 years agovirtio-net: add feature bit for any header s/g
Michael S. Tsirkin [Thu, 11 Jul 2013 13:06:46 +0000 (16:06 +0300)]
virtio-net: add feature bit for any header s/g

Old qemu versions required that 1st s/g entry is the header.

Since QEMU 1.5, patchset titled "virtio-net: iovec handling cleanup"
removed this limitation but a feature bit is needed so guests know it's
safe to lay out header differently.

This patch applies on top and adds such a feature bit to QEMU.
It is set by default for virtio-net.
virtio net header inline with the data is beneficial
for latency and small packet bandwidth - guest driver
code utilizing this feature has been acked but missed 3.11
by a narrow margin, it's pending for 3.12.

This feature bit is cleared by default when compatibility with old
machine types is requested.

Other performance-sensitive devices (blk and scsi)
don't yet support arbitrary s/g layouts, so
we only set this bit for virtio-net for now.
There are plans to allow arbitrary layouts there, but
no code has been posted yet.

Cc: Rusty Russell <rusty@rustcorp.com.au>
Reviewed-by: Laszlo Ersek <lersek@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
11 years agonet: add support of mac-programming over macvtap in QEMU side
Amos Kong [Fri, 14 Jun 2013 07:45:52 +0000 (15:45 +0800)]
net: add support of mac-programming over macvtap in QEMU side

Currently macvtap based macvlan device is working in promiscuous
mode, we want to implement mac-programming over macvtap through
Libvirt for better performance.

Design:
 QEMU notifies Libvirt when rx-filter config is changed in guest,
 then Libvirt query the rx-filter information by a monitor command,
 and sync the change to macvtap device. Related rx-filter config
 of the nic contains main mac, rx-mode items and vlan table.

This patch adds a QMP event to notify management of rx-filter change,
and adds a monitor command for management to query rx-filter
information.

Test:
 If we repeatedly add/remove vlan, and change macaddr of vlan
 interfaces in guest by a loop script.

Result:
 The events will flood the QMP client(management), management takes
 too much resource to process the events.

 Event_throttle API (set rate to 1 ms) can avoid the events to flood
 QMP client, but it could cause an unexpected delay (~1ms), guests
 guests normally expect rx-filter updates immediately.

 So we use a flag for each nic to avoid events flooding, the event
 is emitted once until the query command is executed. The flag
 implementation could not introduce unexpected delay.

There maybe exist an uncontrollable delay if we let Libvirt do the
real change, guests normally expect rx-filter updates immediately.
But it's another separate issue, we can investigate it when the
work in Libvirt side is done.

Michael S. Tsirkin: tweaked to enable events on start
Michael S. Tsirkin: fixed not to crash when no id
Michael S. Tsirkin: fold in patch:
   "additional fixes for mac-programming feature"
Amos Kong: always notify QMP client if mactable is changed
Amos Kong: return NULL list if no net client supports rx-filter query

Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Markus Armbruster <armbru@redhat.com>
Signed-off-by: Amos Kong <akong@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
11 years agotarget-arm: Avoid g_hash_table_get_keys()
Peter Maydell [Mon, 1 Jul 2013 11:40:19 +0000 (12:40 +0100)]
target-arm: Avoid g_hash_table_get_keys()

g_hash_table_get_keys() was only introduced in glib 2.14, and we're
still targeting a minimum version of 2.12.  Rewrite the offending
code (introduced in commit 721fae1) to use g_hash_table_foreach()
to build the list of keys.

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Tested-by: Laurent Desnogues <laurent.desnogues@gmail.com>
Tested-by: Peter Crosthwaite <peter.crosthwaite@xilinx.com>
Message-id: 1372678819-8633-1-git-send-email-peter.maydell@linaro.org

11 years agotarget-arm: avoid undefined behaviour when writing TTBCR
Peter Maydell [Thu, 27 Jun 2013 15:38:47 +0000 (16:38 +0100)]
target-arm: avoid undefined behaviour when writing TTBCR

LPAE CPUs have more potentially valid bits in the TTBCR, and so the
simple masking out of invalid bits is no longer sufficient to obtain
the base address width field of the register, which is what we use to
precalculate c2_mask and c2_base_mask.  Explicitly extract the
relevant register field rather than simply shifting by the register
value.

This bug would have had no ill effects in practice, since if the
EAE bit (TTBCR bit 31) is set then we don't use the precalculated
masks, and if EAE is zero then bits 30..3 are all UNK/SBZP, so
well-behaved guests won't set them. However the shift is undefined
behaviour, so we should avoid it.

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Message-id: 1372347527-4428-1-git-send-email-peter.maydell@linaro.org

11 years agotarget-arm/helper.c: Allow const opaques in arm CP
Peter Crosthwaite [Wed, 10 Jul 2013 04:22:59 +0000 (14:22 +1000)]
target-arm/helper.c: Allow const opaques in arm CP

Allow for defining const opaque data in ARM CP register definitions by
setting .opaque = foo. If non null opaque is passed into
define_one_arm_cp_reg_with_opaque then that opaque will take
precedence, otherwise if null opaque is passed, the original opaque
data will be used.

Signed-off-by: Peter Crosthwaite <peter.crosthwaite@xilinx.com>
Message-id: cf0a3ac3438d97464240db9f5f4ef1585cbc1d77.1373429432.git.peter.crosthwaite@xilinx.com
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
11 years agotarget-arm/helper.c: Implement MIDR aliases
Peter Crosthwaite [Wed, 10 Jul 2013 04:22:21 +0000 (14:22 +1000)]
target-arm/helper.c: Implement MIDR aliases

Unimplemented registers in the cp15, CRn=0, opc1=0, CRm=0 space default
to aliasing the MIDR register. Set all registers in the space to access
MIDR by default.

Signed-off-by: Peter Crosthwaite <peter.crosthwaite@xilinx.com>
Message-id: 6127846712b7ad2727354a4f5e1d809451f1e859.1373429432.git.peter.crosthwaite@xilinx.com
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
11 years agotarget-arm/helper.c: OMAP/StrongARM cp15 crn=0 cleanup
Peter Crosthwaite [Wed, 10 Jul 2013 04:21:42 +0000 (14:21 +1000)]
target-arm/helper.c: OMAP/StrongARM cp15 crn=0 cleanup

The if block detecting OMAP/StrongARM modifies the id_cp_reginfo
.access fields in place. So there is no need to replicate the call
to define_arm_cp_reg(). Dropped, and let the OMAP case fall through
to the normal behaviour after the in-place modification.

Signed-off-by: Peter Crosthwaite <peter.crosthwaite@xilinx.com>
Message-id: 72aae9b8ebbc9a76d2b06faf8666ef8a4b34b92a.1373429432.git.peter.crosthwaite@xilinx.com
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
11 years agotarget-arm: explicitly decode SEVL instruction
Mans Rullgard [Mon, 15 Jul 2013 13:35:25 +0000 (14:35 +0100)]
target-arm: explicitly decode SEVL instruction

The ARMv8 SEVL instruction is in the architectural hint space already
emulated as nop.  This makes the decoding of SEVL explicit for clarity.

Signed-off-by: Mans Rullgard <mans@mansr.com>
Message-id: 1370606786-5650-3-git-send-email-mans@mansr.com
[PMM: added 'SEVL' to the TODO comment]
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
11 years agotarget-arm: implement LDA/STL instructions
Mans Rullgard [Mon, 15 Jul 2013 13:35:25 +0000 (14:35 +0100)]
target-arm: implement LDA/STL instructions

This adds support for the ARMv8 load acquire/store release instructions.
Since qemu does nothing special for memory barriers, these can be
emulated like their non-acquire/release counterparts.

Signed-off-by: Mans Rullgard <mans@mansr.com>
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
11 years agovirtio-ccw: Enable x-data-plane for virtio-ccw-blk
Dominik Dingel [Thu, 11 Jul 2013 11:22:19 +0000 (13:22 +0200)]
virtio-ccw: Enable x-data-plane for virtio-ccw-blk

Add property x-data-plane to virtio-ccw-blk devices.

Signed-off-by: Dominik Dingel <dingel@linux.vnet.ibm.com>
Signed-off-by: Cornelia Huck <cornelia.huck@de.ibm.com>
11 years agoARM/highbank: add support for Calxeda ECX-2000 / Midway
Andre Przywara [Fri, 5 Jul 2013 12:21:37 +0000 (14:21 +0200)]
ARM/highbank: add support for Calxeda ECX-2000 / Midway

The Calxeda ECX-2000 chip (aka. Midway) is model-wise quite similar
to the Highbank. The most prominent difference is the Cortex-A15 CPU
core in it, together with the associated core peripherals.

Add a new ARM machine type called "midway".
Move the L2 cache controller device into the Highbank specific part,
since Midway does not have (and need) it.

Signed-off-by: Andre Przywara <andre.przywara@calxeda.com>
Message-id: 1373026897-12085-3-git-send-email-andre.przywara@calxeda.com
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
11 years agoARM/highbank: prepare for adding similar machines
Andre Przywara [Fri, 5 Jul 2013 12:21:36 +0000 (14:21 +0200)]
ARM/highbank: prepare for adding similar machines

To allow the modelling of machines similar to Calxeda Highbank,
introduce a parameter to the init function and call it from a
wrapper. This allows to tweak the definition for individual machines
later on.

Signed-off-by: Andre Przywara <andre.przywara@calxeda.com>
Message-id: 1373026897-12085-2-git-send-email-andre.przywara@calxeda.com
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
11 years agohw/arm/vexpress: Add alias for flash at address 0 on A15 board
Peter Maydell [Tue, 9 Jul 2013 12:49:40 +0000 (13:49 +0100)]
hw/arm/vexpress: Add alias for flash at address 0 on A15 board

The A15 Versatile Express board can remap a variety of things at address
0. We don't currently emulate the Serial Configuration Controller which
is how the guest can control this remapping, but we can provide the
initial default mapping of the first flash device into this space.
In particular this allows QEMU to boot flash images such as UEFI which
expect to include an exception vector table.

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Tested-by: Grant Likely <grant.likely@linaro.org>
Message-id: 1373374180-19884-1-git-send-email-peter.maydell@linaro.org

11 years agohw/dma/omap_dma: Fix bugs with DMA requests above 32
Peter Maydell [Fri, 28 Jun 2013 12:51:59 +0000 (13:51 +0100)]
hw/dma/omap_dma: Fix bugs with DMA requests above 32

The drqbmp field of struct soc_dma_s is a uint64_t; however several
places in the code attempt to set bits in it using "(1 << drq)",
which will fail if drq is large enough that the 1 bit gets shifted
off the top of a 32 bit integer.  Change these to "(1ULL << drq)" so
that the promotion to 64 bit happens before the shift rather than
afterwards.

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Message-id: 1372423919-5669-1-git-send-email-peter.maydell@linaro.org

11 years agosd/pl181.c: Avoid undefined shift behaviour in RWORD macro
Peter Maydell [Thu, 27 Jun 2013 14:03:51 +0000 (15:03 +0100)]
sd/pl181.c: Avoid undefined shift behaviour in RWORD macro

Add a cast to avoid potentially shifting into the sign bit of
a signed value, which is undefined behaviour in C.

(Detected with clang's -fsanitize=undefined.)

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Message-id: 1372341831-4264-1-git-send-email-peter.maydell@linaro.org

11 years agohw/cpu/a15mpcore: Correct default value for num-irq
Peter Maydell [Fri, 5 Jul 2013 13:54:41 +0000 (14:54 +0100)]
hw/cpu/a15mpcore: Correct default value for num-irq

The a15mpcore device claims that its default value for num-irq
is the number of interrupts used by the A15MP in the vexpress-a15
board. However that chip has 128 external interrupts, not 64.
Since there is only one A15 based model in QEMU currently, we
can fix this by simply changing the default value.

This error was causing recent (3.10) Linux kernels to print
warnings/backtraces when the number of interrupts reported
by the GIC was smaller than an interrupt number they wanted
to use.

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Message-id: 1373032481-15280-1-git-send-email-peter.maydell@linaro.org

11 years agotarget-arm: add feature flag for ARMv8
Mans Rullgard [Mon, 15 Jul 2013 13:35:25 +0000 (14:35 +0100)]
target-arm: add feature flag for ARMv8

Signed-off-by: Mans Rullgard <mans@mansr.com>
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
11 years agochar/cadence_uart: Fix reset for unattached instances
Peter Crosthwaite [Mon, 15 Jul 2013 11:28:07 +0000 (12:28 +0100)]
char/cadence_uart: Fix reset for unattached instances

commit 1db8b5efe0c2b5000e50691eea61264a615f43de introduced an issue
where QEMU would segfault if you have an unattached Cadence UART.

Fix by guarding the flush-on-reset logic on there being a qemu_chr
attachment.

Reported-by: Soren Brinkmann <soren.brinkmann@xilinx.com>
Signed-off-by: Peter Crosthwaite <peter.crosthwaite@xilinx.com>
Tested-by: Soren Brinkmann <soren.brinkmann@xilinx.com>
Message-id: 9009578ee10a50d994b2e10aa2840d73765f5968.1370577272.git.peter.crosthwaite@xilinx.com
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
11 years agoahci: Fix FLUSH command
Kevin Wolf [Mon, 15 Jul 2013 09:25:55 +0000 (11:25 +0200)]
ahci: Fix FLUSH command

AHCI couldn't cope with asynchronous commands that aren't doing DMA, it
simply wouldn't complete them. Due to the bug fixed in commit f68ec837,
FLUSH commands would seem to have completed immediately even if they
were still running on the host. After the commit, they would simply hang
and never unset the BSY bit, rendering AHCI unusable on any OS sending
flushes.

This patch adds another callback for the completion of asynchronous
commands. This is what AHCI really wants to use for its command
completion logic rather than an DMA completion callback.

Cc: qemu-stable@nongnu.org
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
11 years agomigration: Fail migration on bdrv_flush_all() error
Kevin Wolf [Fri, 5 Jul 2013 11:54:55 +0000 (13:54 +0200)]
migration: Fail migration on bdrv_flush_all() error

If bdrv_flush_all() returns an error, there is an inconsistency in the
view of an image file between the source and the destination host.
Completing the migration would lead to corruption. Better abort
migration in this case.

To reproduce this case, try the following (ensures that there is
something to flush, and then fails that flush):

$ qemu-img create -f qcow2 test.qcow2 1G
$ cat blkdebug.cfg
[inject-error]
event = "flush_to_os"
errno = "5"
$ qemu-system-x86_64 -hda blkdebug:blkdebug.cfg:test.qcow2 -monitor stdio
(qemu) qemu-io ide0-hd0 "write 0 4k"
(qemu) migrate ...

Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
11 years agocpus: Add return value for vm_stop()
Kevin Wolf [Fri, 5 Jul 2013 11:49:54 +0000 (13:49 +0200)]
cpus: Add return value for vm_stop()

If flushing the block devices fails, return an error. The VM is stopped
anyway.

Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
11 years agoblock: Add return value for bdrv_flush_all()
Kevin Wolf [Fri, 5 Jul 2013 11:48:01 +0000 (13:48 +0200)]
block: Add return value for bdrv_flush_all()

bdrv_flush() can fail, and bdrv_flush_all() should return an error as
well if this happens for a block device. It returns the first error
return now, but still at least tries to flush the remaining devices even
in error cases.

Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
11 years agoqemu-iotests: Update 051 reference output
Kevin Wolf [Wed, 10 Jul 2013 15:30:26 +0000 (17:30 +0200)]
qemu-iotests: Update 051 reference output

This has been broken by commit bd5c51ee.

Signed-off-by: Kevin Wolf <kwolf@redhat.com>
11 years agoblock: Don't parse protocol from file.filename
Kevin Wolf [Wed, 10 Jul 2013 13:47:39 +0000 (15:47 +0200)]
block: Don't parse protocol from file.filename

One of the major reasons for doing something new for -blockdev and
blockdev-add was that the old block layer code parses filenames instead
of just taking them literally. So we should really leave it untouched
when it's passing using the new interfaces (like -drive
file.filename=...).

This allows opening relative file names that contain a colon.

Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
11 years agoblock: add drive_backup HMP command
Stefan Hajnoczi [Wed, 26 Jun 2013 12:11:58 +0000 (14:11 +0200)]
block: add drive_backup HMP command

Make "drive_backup" available on the HMP monitor:

  drive_backup [-n] [-f] device target [format]

The -n flag requests QEMU to reuse the image found in new-image-file,
instead of recreating it from scratch.

The -f flag requests QEMU to copy the whole disk, so that the result
does not need a backing file.  Note that this flag *must* currently be
passed since the other sync modes ('none' and 'top') have not been
implemented yet.  Requiring it ensures that "drive_backup" behaves like
"drive_mirror".

Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
11 years agoblockdev: add sync mode to drive-backup QMP command
Stefan Hajnoczi [Wed, 26 Jun 2013 12:11:57 +0000 (14:11 +0200)]
blockdev: add sync mode to drive-backup QMP command

The drive-backup command is similar to the drive-mirror command, except
no guest data written after the command executes gets copied.  Add a
sync mode argument which determines whether the entire disk is copied,
just allocated clusters, or only clusters being written to by the guest.

Currently only sync mode 'full' is supported - it copies the entire disk.
For read-only point-in-time snapshots we may only need sync mode 'none'
since the target can be a qcow2 file using the guest's disk as its
backing file (no need to copy the entire disk).  Finally, sync mode
'top' is useful if we wish to preserve the backing chain.

Note that this patch just adds the sync mode argument to drive-backup.
It does not implement sync modes 'top' or 'none'.  This patch is
necessary so we can add a drive-backup HMP command that behaves like the
existing drive-mirror HMP command and takes a sync mode.

Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
11 years agohw/alpha: Use SRM epoch
Richard Henderson [Sun, 14 Jul 2013 00:23:37 +0000 (17:23 -0700)]
hw/alpha: Use SRM epoch

The 1980 epoch is used by the ARC PALcode for NT.  But we're emulating
a system using the SRM PALcode.  Using the proper epoch results in less
confusion in the guest userland.

Signed-off-by: Richard Henderson <rth@twiddle.net>
11 years agohw/alpha: Drop latch_tmp hack
Richard Henderson [Mon, 8 Jul 2013 21:57:39 +0000 (14:57 -0700)]
hw/alpha: Drop latch_tmp hack

The memory and i/o core now support passing 64-bit accesses along
from the guest, so we no longer need to emulate them.

Signed-off-by: Richard Henderson <rth@twiddle.net>
11 years agoexec: Support 64-bit operations in address_space_rw
Richard Henderson [Mon, 8 Jul 2013 21:55:59 +0000 (14:55 -0700)]
exec: Support 64-bit operations in address_space_rw

Honor the implementation maximum access size, and at least check
the minimum access size.

Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Richard Henderson <rth@twiddle.net>
11 years agohw/alpha: Don't machine check on missing pci i/o
Richard Henderson [Mon, 8 Jul 2013 22:46:37 +0000 (15:46 -0700)]
hw/alpha: Don't machine check on missing pci i/o

Not really correct, but we don't implement all of the random devices
that the kernel looks for.  This is good enough to keep us booting.

Signed-off-by: Richard Henderson <rth@twiddle.net>
11 years agohw/alpha: Don't use get_system_io
Richard Henderson [Mon, 8 Jul 2013 20:55:30 +0000 (13:55 -0700)]
hw/alpha: Don't use get_system_io

Advancements in the ioport subsystem mean that we need no longer
thunk memory-mapped i/o through the system-io address space.

Signed-off-by: Richard Henderson <rth@twiddle.net>
11 years agoioport: remove LITTLE_ENDIAN mark for portio
Anthony Liguori [Fri, 12 Jul 2013 19:37:47 +0000 (14:37 -0500)]
ioport: remove LITTLE_ENDIAN mark for portio

Setting it to LE forces a byte swap when host != guest endian but
this makes no sense at all.

Herve made the suggestion upon observing that word writes/reads
were broken into byte writes/reads in such a way as to assume
devices are interpret registers as LE.

However, even if this were a problem, marking the region as LE is
not useful because what's essentially happening here is that LE is
open coded.  So by marking it LE in MemoryRegionOps, we're doing a
superflous swap.

Now, the portio code is suspicious to begin with.  The dispatch
layer really has no purpose in splitting I/O requests in the first
place...

Cc: Hervé Poussineau <hpoussin@reactos.org>
Cc: Alex Graf <agraf@suse.de>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
11 years agoForce auto-convegence of live migration
Chegu Vinod [Mon, 24 Jun 2013 09:47:39 +0000 (03:47 -0600)]
Force auto-convegence of live migration

If a user chooses to turn on the auto-converge migration capability
these changes detect the lack of convergence and throttle down the
guest. i.e. force the VCPUs out of the guest for some duration
and let the migration thread catchup and help converge.

Verified the convergence using the following :
 - Java Warehouse workload running on a 20VCPU/256G guest(~80% busy)
 - OLTP like workload running on a 80VCPU/512G guest (~80% busy)

Sample results with Java warehouse workload : (migrate speed set to 20Gb and
migrate downtime set to 4seconds).

 (qemu) info migrate
 capabilities: xbzrle: off auto-converge: off  <----
 Migration status: active
 total time: 1487503 milliseconds
 expected downtime: 519 milliseconds
 transferred ram: 383749347 kbytes
 remaining ram: 2753372 kbytes
 total ram: 268444224 kbytes
 duplicate: 65461532 pages
 skipped: 64901568 pages
 normal: 95750218 pages
 normal bytes: 383000872 kbytes
 dirty pages rate: 67551 pages

 ---

 (qemu) info migrate
 capabilities: xbzrle: off auto-converge: on   <----
 Migration status: completed
 total time: 241161 milliseconds
 downtime: 6373 milliseconds
 transferred ram: 28235307 kbytes
 remaining ram: 0 kbytes
 total ram: 268444224 kbytes
 duplicate: 64946416 pages
 skipped: 64903523 pages
 normal: 7044971 pages
 normal bytes: 28179884 kbytes

Signed-off-by: Chegu Vinod <chegu_vinod@hp.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>
11 years agoAdd 'auto-converge' migration capability
Chegu Vinod [Mon, 24 Jun 2013 09:49:42 +0000 (03:49 -0600)]
Add 'auto-converge' migration capability

The auto-converge migration capability allows the user to specify if they
choose live migration seqeunce to automatically detect and force convergence.

Signed-off-by: Chegu Vinod <chegu_vinod@hp.com>
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>
11 years agoIntroduce async_run_on_cpu()
Chegu Vinod [Mon, 24 Jun 2013 09:49:41 +0000 (03:49 -0600)]
Introduce async_run_on_cpu()

Introduce an asynchronous version of run_on_cpu() i.e. the caller
doesn't have to block till the call back routine finishes execution
on the target vcpu.

Signed-off-by: Chegu Vinod <chegu_vinod@hp.com>
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>
11 years agoMerge remote-tracking branch 'agraf/ppc-for-upstream' into staging
Anthony Liguori [Fri, 12 Jul 2013 12:58:31 +0000 (07:58 -0500)]
Merge remote-tracking branch 'agraf/ppc-for-upstream' into staging

# By Alexander Graf (16) and others
# Via Alexander Graf
* agraf/ppc-for-upstream: (22 commits)
  PPC: dbdma: Support more multi-issue DMA requests
  PPC: Add timer handler for newworld mac-io
  PPC: dbdma: Support unaligned DMA access
  PPC: dbdma: Wait for DMA until we have data
  PPC: dbdma: Move processing to io
  PPC: dbdma: macio: Add DMA callback
  PPC: dbdma: Move static bh variable to device struct
  PPC: dbdma: Introduce kick function
  PPC: dbdma: Move defines into header file
  PPC: dbdma: Allow new commands in RUN state
  PPC: dbdma: Fix debug print
  PPC: Mac: Add debug prints in macio and dbdma code
  PPC: dbdma: Replace tabs with spaces
  PPC: Macio: Replace tabs with spaces
  PPC: g3beige: Move secondary IDE bus to mac-io
  PPC: Mac: Fix guest exported tbfreq values
  target-ppc: Add POWER8 v1.0 CPU model
  pseries: move interrupt controllers to hw/intc/
  spapr: Respect -bios command line option for SLOF
  spapr: Use named enum for function remove_hpte
  ...

Message-id: 1373562085-29728-1-git-send-email-agraf@suse.de
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
11 years agoPPC: dbdma: Support more multi-issue DMA requests
Alexander Graf [Sun, 30 Jun 2013 13:29:13 +0000 (15:29 +0200)]
PPC: dbdma: Support more multi-issue DMA requests

A DMA request can happen for data that hasn't been completely been
provided by the IDE core yet. For example

  - DBDMA request for 0x1000 bytes
  - IDE request for 1 sector
  - DBDMA wants to read 0x1000 bytes (8 sectors) from bdrv
  - breakage

Instead, we should truncate our bdrv request to the maximum number
of sectors we're allowed to read at that given time. Once that transfer
is through, we will fall into our recently introduced waiting logic.

  - DBDMA requests for 0x1000 bytes
  - IDE request for 1 sector
  - DBDMA wants to read MIN(0x1000, 1 * 512) bytes
  - DBDMA finishes reading, indicates to IDE core that transfer is complete
  - IDE request for 7 sectors
  - DBDMA finishes the DMA

Reported-by: Mark Cave-Ayland <mark.cave-ayland@ilande.co.uk>
Signed-off-by: Alexander Graf <agraf@suse.de>
11 years agoPPC: Add timer handler for newworld mac-io
Alexander Graf [Sun, 30 Jun 2013 03:15:14 +0000 (05:15 +0200)]
PPC: Add timer handler for newworld mac-io

Mac OS X accesses fancy timer registers inside of the mac-io on bootup.

These really should be ticking at the mac-io bus frequency, but I don't
see anyone upset when we just make them as fast as we want to.

With this patch on top of my previous patch queue and latest OpenBIOS
I am able to boot Mac OS X 10.4 with -M mac99.

Signed-off-by: Alexander Graf <agraf@suse.de>
11 years agoPPC: dbdma: Support unaligned DMA access
Alexander Graf [Fri, 28 Jun 2013 11:30:01 +0000 (13:30 +0200)]
PPC: dbdma: Support unaligned DMA access

The DBDMA engine really just reads bytes from a producing device (IDE
in our case) and shoves these bytes into memory. It doesn't care whether
any alignment takes place or not.

Our code today however assumes that block accesses always happen on
sector (512 byte) boundaries. This is a fair assumption for most cases.

However, Mac OS X really likes to do unaligned, incomplete accesses
that it finishes with the next DMA request.

So we need to read / write the unaligned bits independent of the actual
asynchronous request, because that one can only handle 512-byte-aligned
data. We also need to cache these unaligned sectors until the next DMA
request, at which point the data might be successfully flushed from the
pipe.

Signed-off-by: Alexander Graf <agraf@suse.de>
11 years agoPPC: dbdma: Wait for DMA until we have data
Alexander Graf [Sun, 30 Jun 2013 00:54:35 +0000 (02:54 +0200)]
PPC: dbdma: Wait for DMA until we have data

We should only start processing DMA requests when we have data to process.
Hold off working through the DMA shuffling until the IDE core told us that
it's ready.

This is required because the guest can program the DMA engine or the IDE
transfer first. Both are legal.

Signed-off-by: Alexander Graf <agraf@suse.de>
11 years agoPPC: dbdma: Move processing to io
Alexander Graf [Sun, 30 Jun 2013 00:47:20 +0000 (02:47 +0200)]
PPC: dbdma: Move processing to io

Soon we will introduce intermediate processing pauses which will
allow the bottom half to restart a DMA request that couldn't be
fulfilled yet.

For that to work, move the processing variable into the io struct
which is what DMA providers work with.

While touching it, also change it into a bool

Signed-off-by: Alexander Graf <agraf@suse.de>
11 years agoPPC: dbdma: macio: Add DMA callback
Alexander Graf [Sun, 30 Jun 2013 00:36:14 +0000 (02:36 +0200)]
PPC: dbdma: macio: Add DMA callback

We need to know when the IDE core starts a DMA transfer. Add a notifier
function so we have the chance to start transmitting data.

Signed-off-by: Alexander Graf <agraf@suse.de>
11 years agoPPC: dbdma: Move static bh variable to device struct
Alexander Graf [Sun, 30 Jun 2013 00:22:41 +0000 (02:22 +0200)]
PPC: dbdma: Move static bh variable to device struct

The DBDMA controller has a bottom half to asynchronously process DMA
request queues.

This bh was stored as a gross static variable. Move it into the device
struct instead.

While at it, move all users of it to the new generic kick function.

Signed-off-by: Alexander Graf <agraf@suse.de>
11 years agoPPC: dbdma: Introduce kick function
Alexander Graf [Sun, 30 Jun 2013 00:18:54 +0000 (02:18 +0200)]
PPC: dbdma: Introduce kick function

The DBDMA engine really is running all the time, waiting for input. However
we don't want to waste cycles constantly polling.

So introduce a kick function that data providers can call to notify the
DBDMA controller of new input.

Signed-off-by: Alexander Graf <agraf@suse.de>
11 years agoPPC: dbdma: Move defines into header file
Alexander Graf [Sun, 30 Jun 2013 00:06:35 +0000 (02:06 +0200)]
PPC: dbdma: Move defines into header file

We usually keep struct and constant definitions in header files. Move
them there to stay consistent and to make access to fields easier.

Signed-off-by: Alexander Graf <agraf@suse.de>
11 years agoPPC: dbdma: Allow new commands in RUN state
Alexander Graf [Sat, 29 Jun 2013 23:53:51 +0000 (01:53 +0200)]
PPC: dbdma: Allow new commands in RUN state

The DBDMA controller can not change its command stream while it's
actively streaming data, true. But the fact that it's in RUN state
doesn't actually indicate anything. It could just as well be in
WAIT while in RUN. And then it's legal to change commands.

This fixes a real world issue I've encountered with Mac OS X.

Signed-off-by: Alexander Graf <agraf@suse.de>
11 years agoPPC: dbdma: Fix debug print
Alexander Graf [Sat, 29 Jun 2013 23:53:05 +0000 (01:53 +0200)]
PPC: dbdma: Fix debug print

There was a debug print that didn't compile for me because the format
and the arguments weren't in sync. Fix it up.

Signed-off-by: Alexander Graf <agraf@suse.de>
11 years agoPPC: Mac: Add debug prints in macio and dbdma code
Alexander Graf [Sat, 29 Jun 2013 23:23:45 +0000 (01:23 +0200)]
PPC: Mac: Add debug prints in macio and dbdma code

The macio code is basically undebuggable as it stands today, with no
debug prints anywhere whatsoever. DBDMA was better, but I needed a
few more to create reasonable logs that tell me where breakage is.

Add a DPRINTF macro in the macio source file and add a bunch of debug
prints that are all disabled by default of course.

Signed-off-by: Alexander Graf <agraf@suse.de>
11 years agoPPC: dbdma: Replace tabs with spaces
Alexander Graf [Sat, 29 Jun 2013 23:45:22 +0000 (01:45 +0200)]
PPC: dbdma: Replace tabs with spaces

s/^I/        /g on the file with a few manual tweaks to align things.

Signed-off-by: Alexander Graf <agraf@suse.de>
11 years agoPPC: Macio: Replace tabs with spaces
Alexander Graf [Sat, 29 Jun 2013 23:43:17 +0000 (01:43 +0200)]
PPC: Macio: Replace tabs with spaces

s/^I/        /g on the file.

Signed-off-by: Alexander Graf <agraf@suse.de>
11 years agoPPC: g3beige: Move secondary IDE bus to mac-io
Alexander Graf [Mon, 24 Jun 2013 19:40:50 +0000 (21:40 +0200)]
PPC: g3beige: Move secondary IDE bus to mac-io

On a real G3 Beige the secondary IDE bus lives on the mac-io chip, not
on some random PCI device. Move it there to become more compatible.

While at it, also clean up the IDE channel connection logic.

Signed-off-by: Alexander Graf <agraf@suse.de>
11 years agoPPC: Mac: Fix guest exported tbfreq values
Alexander Graf [Sat, 29 Jun 2013 15:34:58 +0000 (17:34 +0200)]
PPC: Mac: Fix guest exported tbfreq values

We can tell the guest the frequency of its time base through fwcfg.

However, we tell it a different value from the speed tb actually runs
at. Let's fix it and make the tbfreq initialization and the fwcfg exposure
use the same values.

Signed-off-by: Alexander Graf <agraf@suse.de>
11 years agotarget-ppc: Add POWER8 v1.0 CPU model
Prerna Saxena [Thu, 4 Jul 2013 06:42:32 +0000 (12:12 +0530)]
target-ppc: Add POWER8 v1.0 CPU model

This patch adds CPU PVR definition for POWER8,
and enables QEMU to launch guests on POWER8 hardware.

Signed-off-by: Prerna Saxena <prerna@linux.vnet.ibm.com>
Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Reviewed-by: Paul Mackerras <paulus@samba.org>
Reviewed-by: Andreas Farber <afaerber@suse.de>
Signed-off-by: Alexander Graf <agraf@suse.de>