qemu.git
6 years agohw/arm/vexpress: Alias NOR flash at 0 for vexpress-a9
Peter Maydell [Wed, 2 Jul 2014 14:07:50 +0000 (15:07 +0100)] 
hw/arm/vexpress: Alias NOR flash at 0 for vexpress-a9

Make the vexpress-a9 board alias the first NOR flash region at
address zero, like vexpress-a15. This makes "-bios" actually usable
on this board.

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Message-id: 1404310070-3561-1-git-send-email-peter.maydell@linaro.org
Reviewed-by: Greg Bellows <greg.bellows@linaro.org>
6 years agodisas/libvixl: prepend the include path of libvixl header files
Stefano Stabellini [Mon, 7 Jul 2014 15:25:07 +0000 (16:25 +0100)] 
disas/libvixl: prepend the include path of libvixl header files

Currently the Makefile of disas/libvixl appends
-I$(SRC_PATH)/disas/libvixl to QEMU_CFLAGS. As a consequence C++ files
that #include "utils.h", such as disas/libvixl/a64/instructions-a64.cc,
are going to look for utils.h on all the other include paths first.

When building QEMU as part of the Xen make system, another unrelated
utils.h file is going to be chosen for inclusion, causing a build
failure:

In file included from disas/libvixl/a64/instructions-a64.cc:27:0:
/qemu/disas/libvixl/a64/instructions-a64.h:88:64: error:
'rawbits_to_float' was not declared in this scope
 const float kFP32PositiveInfinity = rawbits_to_float(0x7f800000);

Fix the problem by prepending (rather than appending) the libvixl
include path to QEMU_CFLAGS.

Signed-off-by: Stefano Stabellini <stefano.stabellini@eu.citrix.com>
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
6 years agoMerge remote-tracking branch 'remotes/agraf/tags/signed-ppc-for-upstream' into staging
Peter Maydell [Tue, 8 Jul 2014 10:38:12 +0000 (11:38 +0100)] 
Merge remote-tracking branch 'remotes/agraf/tags/signed-ppc-for-upstream' into staging

Patch queue for ppc - 2014-07-08

A few bug fixes for 2.1:

  - Fix e500* TLB emulation with qemu-system-ppc
  - Update SLOF to current upstream (good number of bugfixes)
  - Make POWER7 / POWER8 PVR match more agnostic (needed in 2.1 for cmdline compat)
  - Fix u-boot.e500 install (how did that happen?)
  - Fix H_CAS on LE hosts
  - ppc64le-linux-user fixes

# gpg: Signature made Tue 08 Jul 2014 11:18:58 BST using RSA key ID 03FEDC60
# gpg: Can't check signature: public key not found

* remotes/agraf/tags/signed-ppc-for-upstream:
  PPC: e500: Actually install u-boot.e500
  target-ppc: Remove POWER7+ and POWER8E families
  target-ppc: Add pvr_match() callback
  pseries: Update SLOF firmware image to qemu-slof-20140630
  PPC: Fix booke206 TLB with phys addrs > 32bit
  target-ppc: Fix gdbstub for ppc64le-linux-user
  target-ppc: Change default cpu for ppc64le-linux-user
  target-ppc: KVMPPC_H_CAS fix cpu-version endianess

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
6 years agoPPC: e500: Actually install u-boot.e500
Cole Robinson [Fri, 4 Jul 2014 19:43:18 +0000 (15:43 -0400)] 
PPC: e500: Actually install u-boot.e500

Signed-off-by: Cole Robinson <crobinso@redhat.com>
Signed-off-by: Alexander Graf <agraf@suse.de>
6 years agotarget-ppc: Remove POWER7+ and POWER8E families
Alexey Kardashevskiy [Mon, 30 Jun 2014 14:30:18 +0000 (00:30 +1000)] 
target-ppc: Remove POWER7+ and POWER8E families

POWER8E is architecturally equal to POWER8 and POWER7+ is equal to
POWER7. Also no user space tool makes any difference for CPU node name
in the device tree (such as PowerPC,POWER7@0 vs. PowerPC,POWER7+@0).
So there is no point in emulating POWER7+ and POWER8E apart from POWER7
and POWER8. Also, the previos patch implemented multiple PVR mask support
per CPU class so POWER7 class now covers both POWER7 and POWER7+ CPUs,
same is valid for POWER8/8E.

This removes POWER7+ and POWER8E classes. This replaces references
to POWER7P/POWER8E families with POWER7/POWER8 families.

Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Signed-off-by: Alexander Graf <agraf@suse.de>
6 years agotarget-ppc: Add pvr_match() callback
Alexey Kardashevskiy [Thu, 3 Jul 2014 14:48:55 +0000 (00:48 +1000)] 
target-ppc: Add pvr_match() callback

So far it was enough to have a base PVR value and mask per CPU
family such as POWER7 or POWER8. However there CPUs which are
completely architecturally compatible but have different PVRs such
as POWER7/POWER7+ and POWER8/POWER8E. For these CPUs, top 16 bits
are CPU family and low 16 bits are the version. The families have
PVR base values different enough so defining a mask which
would cover both (or potentially more) CPUs within the family is
not possible.

This adds a pvr_match() callback to PowerPCCPUClass. The default
handler simply compares PVR defined in the class.

This implements ppc_pvr_match_power7/ppc_pvr_match_power8 callbacks
for POWER7/8 families. These check for POWER7/POWER7+ and POWER8/POWER8E.

This changes ppc_cpu_compare_class_pvr_mask() not to check masks but
use the pvr_match() callback.

Since all server CPUs use the same mask, this defines one mask
value - CPU_POWERPC_POWER_SERVER_MASK - which is used everywhere now.
This removes other mask definitions.

This removes pvr_mask from PowerPCCPUClass as it is not used anymore.
This removes pvr initialization for POWER7/8 families as it is not used
to find the class, the pvr_match() callback is used instead.

Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Signed-off-by: Alexander Graf <agraf@suse.de>
6 years agopseries: Update SLOF firmware image to qemu-slof-20140630
Alexey Kardashevskiy [Thu, 3 Jul 2014 13:25:56 +0000 (23:25 +1000)] 
pseries: Update SLOF firmware image to qemu-slof-20140630

The changelog is:
  > Quieten the grub warning
  > Add boot menu support
  > boot from disk having chrp-boot file
  > fat16: fix read and remove debug messages
  > dhcparch define missing in compilation
  > pci-scan: reserve memory for pci-bridge without devices
  > pci-bridge: Fix ranges when no device beyond the bridge
  > Set dhcp arch in board-qemu config file
  > xhci: fix controller stop
  > dhcp: support client architecture code 93
  > virtio-blk: support variable block size
  > usb: use common pci dma alloc/mapping routines
  > Remove unused SLOF code
  > pci-bridge: generic bridge needs to support pci dma functions
  > pci: extract dma functions as separate file
  > e1000: fix usage of multiple nics

Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Signed-off-by: Alexander Graf <agraf@suse.de>
6 years agoPPC: Fix booke206 TLB with phys addrs > 32bit
Alexander Graf [Wed, 2 Jul 2014 17:09:47 +0000 (19:09 +0200)] 
PPC: Fix booke206 TLB with phys addrs > 32bit

We were truncating physical addresses to 32bit when using qemu-system-ppc
with a booke206 TLB implementation. This patch fixes that and makes the full
address space available.

Signed-off-by: Alexander Graf <agraf@suse.de>
6 years agotarget-ppc: Fix gdbstub for ppc64le-linux-user
Richard Henderson [Sat, 28 Jun 2014 16:45:28 +0000 (09:45 -0700)] 
target-ppc: Fix gdbstub for ppc64le-linux-user

The bswap that's needed for system mode isn't required for
user mode, and in fact breaks debugging.

Signed-off-by: Richard Henderson <rth@twiddle.net>
[agraf: fix apple gdbstub implementation]
Signed-off-by: Alexander Graf <agraf@suse.de>
6 years agotarget-ppc: Change default cpu for ppc64le-linux-user
Richard Henderson [Sat, 28 Jun 2014 16:45:27 +0000 (09:45 -0700)] 
target-ppc: Change default cpu for ppc64le-linux-user

The default, 970fx, doesn't support MSR_LE.  So even though we set LE in
ppc_cpu_reset, it gets cleared again in hreg_store_msr.  Error out if a
user-selected cpu model doesn't support LE.

Signed-off-by: Richard Henderson <rth@twiddle.net>
[agraf: switch to POWER7 as default for BE and LE]
Signed-off-by: Alexander Graf <agraf@suse.de>
6 years agotarget-ppc: KVMPPC_H_CAS fix cpu-version endianess
Laurent Dufour [Fri, 27 Jun 2014 13:47:37 +0000 (15:47 +0200)] 
target-ppc: KVMPPC_H_CAS fix cpu-version endianess

During KVMPPC_H_CAS processing, the cpu-version updated value is stored
without taking care of the current endianess. As a consequence, the guest
may not switch to the right CPU model, leading to unexpected results.

If needed, the value is now converted.

Fixes: 6d9412ea8132 ("target-ppc: Implement "compat" CPU option")
Signed-off-by: Laurent Dufour <ldufour@linux.vnet.ibm.com>
Reviewed-by: Greg Kurz <gkurz@linux.vnet.ibm.com>
Signed-off-by: Alexander Graf <agraf@suse.de>
6 years agoMerge remote-tracking branch 'remotes/afaerber/tags/prep-for-2.1' into staging
Peter Maydell [Mon, 7 Jul 2014 18:06:55 +0000 (19:06 +0100)] 
Merge remote-tracking branch 'remotes/afaerber/tags/prep-for-2.1' into staging

PowerPC Reference Platform (PReP)

* Update OpenHack'Ware firmware to replace QEMU-side workarounds

# gpg: Signature made Mon 07 Jul 2014 15:49:42 BST using RSA key ID 3E7E013F
# gpg: Good signature from "Andreas Färber <afaerber@suse.de>"
# gpg:                 aka "Andreas Färber <afaerber@suse.com>"

* remotes/afaerber/tags/prep-for-2.1:
  prep: Update ppc_rom.bin
  prep: Remove CPU reset entry point hack related to OpenHack'Ware
  prep: Remove PCI memory hack related to OpenHack'Ware

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
6 years agoMerge remote-tracking branch 'remotes/mst/tags/for_upstream' into staging
Peter Maydell [Mon, 7 Jul 2014 15:30:14 +0000 (16:30 +0100)] 
Merge remote-tracking branch 'remotes/mst/tags/for_upstream' into staging

pc,vhost,virtio fixes, test

Bugfixes all over the place.

There's a  non bugfix here: re-enabling the vhost-user test,
though the patch just brings back functionality that
I disabled earlier to fix mingw build failures.
This is now sorted, and keeping the unit test enabled
seems important since the feature relies on an external
server to work, so isn't easy to test.

Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
# gpg: Signature made Sun 06 Jul 2014 11:01:35 BST using RSA key ID D28D5469
# gpg: Good signature from "Michael S. Tsirkin <mst@kernel.org>"
# gpg:                 aka "Michael S. Tsirkin <mst@redhat.com>"

* remotes/mst/tags/for_upstream:
  qemu-char: add chr_add_watch support in mux chardev
  virtio-pci: fix MSI memory region use after free
  qdev: Fix crash when using non-device class name on -global
  qdev: Don't abort() in case globals can't be set
  hw/virtio: enable common virtio feature for mmio device
  acpi: fix typo in memory hotplug MMIO region name
  pci: assign devfn to pci_dev before calling pci_device_iommu_address_space()
  Handle G_IO_HUP in tcp_chr_read for tcp chardev
  virtio: move common virtio properties to bus class device
  pc-dimm: error out if memory hotplug is not enabled
  numa: check for busy memory backend
  qtest: enable vhost-user-test

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
6 years agoprep: Update ppc_rom.bin
Andreas Färber [Fri, 4 Jul 2014 15:35:32 +0000 (17:35 +0200)] 
prep: Update ppc_rom.bin

This replaces QEMU-side workarounds for PCI BARs and CPU reset.

Signed-off-by: Andreas Färber <andreas.faerber@web.de>
6 years agoprep: Remove CPU reset entry point hack related to OpenHack'Ware
Hervé Poussineau [Tue, 1 Apr 2014 21:19:16 +0000 (23:19 +0200)] 
prep: Remove CPU reset entry point hack related to OpenHack'Ware

Signed-off-by: Hervé Poussineau <hpoussin@reactos.org>
Signed-off-by: Andreas Färber <andreas.faerber@web.de>
6 years agoprep: Remove PCI memory hack related to OpenHack'Ware
Hervé Poussineau [Tue, 1 Apr 2014 21:19:15 +0000 (23:19 +0200)] 
prep: Remove PCI memory hack related to OpenHack'Ware

Signed-off-by: Hervé Poussineau <hpoussin@reactos.org>
Signed-off-by: Andreas Färber <andreas.faerber@web.de>
6 years agoMerge remote-tracking branch 'remotes/stefanha/tags/block-pull-request' into staging
Peter Maydell [Mon, 7 Jul 2014 14:02:36 +0000 (15:02 +0100)] 
Merge remote-tracking branch 'remotes/stefanha/tags/block-pull-request' into staging

Block pull request

# gpg: Signature made Mon 07 Jul 2014 13:27:20 BST using RSA key ID 81AB73C8
# gpg: Good signature from "Stefan Hajnoczi <stefanha@redhat.com>"
# gpg:                 aka "Stefan Hajnoczi <stefanha@gmail.com>"

* remotes/stefanha/tags/block-pull-request:
  qmp: show QOM properties in device-list-properties
  dataplane: submit I/O as a batch
  linux-aio: implement io plug, unplug and flush io queue
  block: block: introduce APIs for submitting IO as a batch
  ahci: map memory via device's address space instead of address_space_memory
  raw-posix: Fix raw_getlength() to always return -errno on error
  qemu-iotests: Disable Quorum testing in 041 when Quorum is not builtin
  ahci.c: mask unused flags when reading size PRDT DBC
  MAINTAINERS: add Stefan Hajnoczi to IDE maintainers
  mirror: Fix qiov size for short requests
  Fix nocow typos in manpage

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
6 years agoMerge remote-tracking branch 'remotes/sstabellini/xen_arm_20140707' into staging
Peter Maydell [Mon, 7 Jul 2014 12:43:03 +0000 (13:43 +0100)] 
Merge remote-tracking branch 'remotes/sstabellini/xen_arm_20140707' into staging

* remotes/sstabellini/xen_arm_20140707:
  xen: build on ARM
  xen_backend: introduce xenstore_read_uint64 and xenstore_read_fe_uint64

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
6 years agoxen: build on ARM
Stefano Stabellini [Wed, 18 Dec 2013 19:17:32 +0000 (19:17 +0000)] 
xen: build on ARM

Collection of fixes to build QEMU with Xen support on ARM:
- use xenstore_read_fe_uint64 to retrieve the page-ref (xenfb);
- use xen_pfn_t instead of unsigned long in xenfb;
- unsigned long/xenpfn_t in xen_remove_from_physmap;
- in xen-mapcache.c use HOST_LONG_BITS to check for QEMU's address space
size.

Signed-off-by: Stefano Stabellini <stefano.stabellini@eu.citrix.com>
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
6 years agoxen_backend: introduce xenstore_read_uint64 and xenstore_read_fe_uint64
Stefano Stabellini [Wed, 18 Dec 2013 19:17:31 +0000 (19:17 +0000)] 
xen_backend: introduce xenstore_read_uint64 and xenstore_read_fe_uint64

Signed-off-by: Stefano Stabellini <stefano.stabellini@eu.citrix.com>
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
6 years agoqmp: show QOM properties in device-list-properties
Stefan Hajnoczi [Tue, 20 May 2014 12:29:01 +0000 (14:29 +0200)] 
qmp: show QOM properties in device-list-properties

Devices can use a mix of qdev and QOM properties.  Currently only the
qdev properties are displayed by device-list-properties.

This patch extends the property enumeration algorithm to also display
QOM properties (excluding the implicit "type", "realized",
"hotpluggable", and "parent_bus" properties).

When a qdev property exists, use the qdev type name to preserve
backwards compatibility.  QOM type names can be different for bool (qdev
on/off) and str (used by qdev pointers).

Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
6 years agodataplane: submit I/O as a batch
Ming Lei [Fri, 4 Jul 2014 10:04:35 +0000 (18:04 +0800)] 
dataplane: submit I/O as a batch

Before commit 580b6b2aa2(dataplane: use the QEMU block
layer for I/O), dataplane for virtio-blk submits block
I/O as a batch.

This commit 580b6b2aa2 replaces the custom linux AIO
implementation(including submit I/O as a batch) with QEMU
block layer, but this commit causes ~40% throughput regression
on virtio-blk performance, and removing submitting I/O
as a batch is one of the causes.

This patch applies the newly introduced bdrv_io_plug() and
bdrv_io_unplug() interfaces to support submitting I/O
at batch for Qemu block layer, and in my test, the change
can improve throughput by ~30% with 'aio=native'.

Following my fio test script:

[global]
direct=1
size=4G
bsrange=4k-4k
timeout=40
numjobs=4
ioengine=libaio
iodepth=64
filename=/dev/vdc
group_reporting=1

[f]
rw=randread

Result on one of my small machine(host: x86_64, 2cores, 4thread, guest: 4cores):
- qemu master: 65K IOPS
- qemu master with these patches: 92K IOPS
- 2.0.0 release(dataplane using custom linux aio): 104K IOPS

Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Ming Lei <ming.lei@canonical.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
6 years agolinux-aio: implement io plug, unplug and flush io queue
Ming Lei [Fri, 4 Jul 2014 10:04:34 +0000 (18:04 +0800)] 
linux-aio: implement io plug, unplug and flush io queue

This patch implements .bdrv_io_plug, .bdrv_io_unplug and
.bdrv_flush_io_queue callbacks for linux-aio Block Drivers,
so that submitting I/O as a batch can be supported on linux-aio.

[Unprocessed requests are completed with -EIO instead of a bogus ret
value.
--Stefan]

Signed-off-by: Ming Lei <ming.lei@canonical.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
6 years agoblock: block: introduce APIs for submitting IO as a batch
Ming Lei [Fri, 4 Jul 2014 10:04:33 +0000 (18:04 +0800)] 
block: block: introduce APIs for submitting IO as a batch

This patch introduces three APIs so that following
patches can support queuing I/O requests and submitting them
as a batch for improving I/O performance.

Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Ming Lei <ming.lei@canonical.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
6 years agoahci: map memory via device's address space instead of address_space_memory
Le Tan [Thu, 3 Jul 2014 08:26:27 +0000 (16:26 +0800)] 
ahci: map memory via device's address space instead of address_space_memory

In map_page() in hw/ide/ahci.c, replace cpu_physical_memory_map() and
cpu_physical_memory_unmap() with dma_memory_map() and dma_memory_unmap(),
because ahci devices should not access memory directly but via their address
space. Add an AddressSpace parameter to map_page(). In order to call
map_page(), we should pass the AHCIState.as as the AddressSpace argument.

Signed-off-by: Le Tan <tamlokveer@gmail.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
6 years agoraw-posix: Fix raw_getlength() to always return -errno on error
Markus Armbruster [Thu, 26 Jun 2014 11:23:16 +0000 (13:23 +0200)] 
raw-posix: Fix raw_getlength() to always return -errno on error

We got a merry mix of -1 and -errno here.

Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Benoit Canet <benoit@irqsave.net>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
6 years agoqemu-iotests: Disable Quorum testing in 041 when Quorum is not builtin
Benoît Canet [Tue, 1 Jul 2014 11:34:12 +0000 (13:34 +0200)] 
qemu-iotests: Disable Quorum testing in 041 when Quorum is not builtin

This avoid breaking tests on RHEL6 where gnutls is too old for quorum to be
built by default.

Signed-off-by: Benoit Canet <benoit@irqsave.net>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
6 years agoahci.c: mask unused flags when reading size PRDT DBC
Reza Jelveh [Tue, 1 Jul 2014 11:13:27 +0000 (13:13 +0200)] 
ahci.c: mask unused flags when reading size PRDT DBC

The data byte count(DBC) read from the description information is defined for
bits 21:00. Bits 30:22 are reserved and bit 31 is the Interrupt on Completion
(I) flag.

Completion interrupts are triggered after every transaction instead of on
I-flag in QEMU. tbl_entry_size is a signed integer and improperly reading the
DBC leads to a negative offset that causes sglist allocation to fail.

Signed-off-by: Reza Jelveh <reza.jelveh@tuhh.de>
Reviewed-by: Alexander Graf <agraf@suse.de>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: John Snow <jsnow@redhat.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
6 years agoMAINTAINERS: add Stefan Hajnoczi to IDE maintainers
Stefan Hajnoczi [Tue, 1 Jul 2014 12:04:31 +0000 (14:04 +0200)] 
MAINTAINERS: add Stefan Hajnoczi to IDE maintainers

Make Stefan officially co-maintain hw/ide/ with Kevin.

Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Acked-by: Kevin Wolf <kwolf@redhat.com>
6 years agomirror: Fix qiov size for short requests
Kevin Wolf [Tue, 1 Jul 2014 14:52:21 +0000 (16:52 +0200)] 
mirror: Fix qiov size for short requests

When mirroring an image of a size that is not a multiple of the
mirror job granularity, the last request would have the right nb_sectors
argument, but a qiov that is rounded up to the next multiple of the
granularity. Don't do this.

This fixes a segfault that is caused by raw-posix being confused by this
and allocating a buffer with request length, but operating on it with
qiov length.

[s/Driver/Drive/ in qemu-iotests 041 as suggested by Eric
--Stefan]

Reported-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Tested-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
6 years agoFix nocow typos in manpage
Chunyan Liu [Wed, 2 Jul 2014 04:27:29 +0000 (12:27 +0800)] 
Fix nocow typos in manpage

Signed-off-by: Chunyan Liu <cyliu@suse.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
6 years agoqemu-char: add chr_add_watch support in mux chardev
Kirill Batuzov [Fri, 4 Jul 2014 12:43:15 +0000 (16:43 +0400)] 
qemu-char: add chr_add_watch support in mux chardev

Forward chr_add_watch call from mux chardev to underlying
implementation.

This should fix bug #1335444

Signed-off-by: Kirill Batuzov <batuzovk@ispras.ru>
Acked-by: Paolo Bonzini <pbonzini@redhat.com>
Acked-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
6 years agovirtio-pci: fix MSI memory region use after free
Paolo Bonzini [Fri, 4 Jul 2014 09:43:49 +0000 (11:43 +0200)] 
virtio-pci: fix MSI memory region use after free

After memory region QOMification QEMU is stricter in detecting
wrong usage of the memory region API.  Here it detected a
memory_region_destroy done before the corresponding
memory_region_del_subregion; the memory_region_destroy is
done by msix_uninit_exclusive_bar, the memory_region_del_subregion
is done by the PCI core's pci_unregister_io_regions before
pc->exit is called.

The problem was introduced by
commit 06a1307379fcd6c551185ad87679cd7ed896b9ea
    virtio-pci: add device_unplugged callback
As noted in that commit log, virtio device kick callbacks need to be
stopped before generic virtio is cleaned up. This is because these are
notifications from pci proxy to the generic virtio device so they need
to be stopped in the unplug call before the virtio device is unrealized.
However interrupts are notifications from the virtio device to
the pci proxy so they need to stay around while the device
is realized.

The memory API misuse caused an assertion when hot-unplugging virtio
devices.  Using the API correctly fixes the assertion.

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
6 years agoqdev: Fix crash when using non-device class name on -global
Eduardo Habkost [Thu, 3 Jul 2014 19:45:35 +0000 (16:45 -0300)] 
qdev: Fix crash when using non-device class name on -global

This fixes the following crash:

    $ qemu-system-x86_64 -global container.xxx=y
    hw/core/qdev-properties-system.c:399:qdev_add_one_global: Object 0x7f7eff234100 is not an instance of type device
    Aborted (core dumped)

New behavior will be to just warn, just like when non-existing clas
names are used:

    $ qemu-system-x86_64 -global container.xxx=y
    qemu-system-x86_64: Warning: "-global container.xxx=y" not used

Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>
Tested-by: Don Slutz <dslutz@verizon.com>
6 years agoqdev: Don't abort() in case globals can't be set
Eduardo Habkost [Thu, 3 Jul 2014 19:45:34 +0000 (16:45 -0300)] 
qdev: Don't abort() in case globals can't be set

It would be much better if we didn't terminate QEMU inside
device_post_init(), but at least exiting cleanly is better than aborting
and dumping core.

Before this patch:

    $ qemu-system-x86_64 -global cpu.xxx=y
    qemu-system-x86_64: Property '.xxx' not found
    Aborted (core dumped)

After this patch:

    $ qemu-system-x86_64 -global cpu.xxx=y
    qemu-system-x86_64: Property '.xxx' not found

Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Reviewed-By: Igor Mammedov <imammedo@redhat.com>
Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>
6 years agohw/virtio: enable common virtio feature for mmio device
Ming Lei [Wed, 4 Jun 2014 02:05:55 +0000 (10:05 +0800)] 
hw/virtio: enable common virtio feature for mmio device

Both 'indirect_desc' and 'event_idx' are bus independent features,
and they should be enabled for mmio devices too.

On arm64 quad core VM(qemu-kvm), the patch can increase block I/O
performance a lot with latest linux tree:
        - without the patch: 14K IOPS
        - with the patch: 34K IOPS

fio script:
        [global]
        direct=1
        bsrange=4k-4k
        timeout=10
        numjobs=4
        ioengine=libaio
        iodepth=64

        filename=/dev/vdc
        group_reporting=1

        [f1]
        rw=randread

Cc: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Ming Lei <ming.lei@canonical.com>
Acked-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
6 years agoacpi: fix typo in memory hotplug MMIO region name
Igor Mammedov [Wed, 2 Jul 2014 09:16:33 +0000 (11:16 +0200)] 
acpi: fix typo in memory hotplug MMIO region name

Reported-by: Sergey Fionov <fionov@gmail.com>
Signed-off-by: Igor Mammedov <imammedo@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Reviewed-by: Peter Crosthwaite <peter.crosthwaite@xilinx.com>
6 years agopci: assign devfn to pci_dev before calling pci_device_iommu_address_space()
Le Tan [Wed, 2 Jul 2014 00:06:35 +0000 (08:06 +0800)] 
pci: assign devfn to pci_dev before calling pci_device_iommu_address_space()

In function do_pci_register_device() in file hw/pci/pci.c, move the assignment
of pci_dev->devfn to the position before the call to
pci_device_iommu_address_space(pci_dev) which will use the value of
pci_dev->devfn.

Fixes: 9eda7d373e9c691c070eddcbe3467b991f67f6bd
    pci: Introduce helper to retrieve a PCI device's DMA address space

Cc: qemu-stable@nongnu.org
Signed-off-by: Le Tan <tamlokveer@gmail.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
6 years agoHandle G_IO_HUP in tcp_chr_read for tcp chardev
Kirill Batuzov [Tue, 1 Jul 2014 11:52:32 +0000 (15:52 +0400)] 
Handle G_IO_HUP in tcp_chr_read for tcp chardev

Since commit cdaa86a54b232572bba594bf87a7416e527e460c
("Add G_IO_HUP handler for socket chardev")
GLib limitation results in a bug on Windows host. Steps to reproduce:

Start qemu: qemu-system-i386 -qmp tcp:127.0.0.1:4444:server:nowait
Connect with telnet: telnet 127.0.0.1 4444
Try sending some data from telnet.
Expected result: answers from QEMU.
Observed result: no answers (actually tcp_chr_read is not called at all).

Due to GLib limitations it is not possible to create several watches on one
channel on Windows hosts. See bug #338943 in GNOME bugzilla for details:
https://bugzilla.gnome.org/show_bug.cgi?id=338943

This reimplements commit cdaa86a54b232572bba594bf87a7416e527e460c
("Add G_IO_HUP handler for socket chardev") using a single watch:

Handle G_IO_HUP in tcp_chr_read instead. It is already watched by a
corresponding watch.  Remove the second watch with its handler.

Cc: Antonios Motakis <a.motakis@virtualopensystems.com>
Cc: Nikolay Nikolaev <n.nikolaev@virtualopensystems.com>
Cc: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Kirill Batuzov <batuzovk@ispras.ru>
Signed-off-by: Nikita Belov <zodiac@ispras.ru>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
6 years agovirtio: move common virtio properties to bus class device
Ming Lei [Mon, 30 Jun 2014 16:01:58 +0000 (00:01 +0800)] 
virtio: move common virtio properties to bus class device

The two common virtio features can be defined per bus, so move all
into bus class device to make code more clean.

As discussed with cornelia, s390-virtio-blk doesn't support
the two features at all, so keep s390-virtio as it.

Acked-by: Cornelia Huck <cornelia.huck@de.ibm.com> #for s390 ccw
Suggested-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Ming Lei <ming.lei@canonical.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
MST: rebase and resolve conflicts

6 years agopc-dimm: error out if memory hotplug is not enabled
Igor Mammedov [Mon, 30 Jun 2014 10:43:29 +0000 (12:43 +0200)] 
pc-dimm: error out if memory hotplug is not enabled

fixes QEMU abort in case it's started without memory
hotplug enabled.

as result of fix it will print following messages:
"
-device pc-dimm,id=d1,memdev=m1: memory hotplug is not enabled, enable it on startup
-device pc-dimm,id=d1,memdev=m1: Device 'pc-dimm' could not be initialized
"

Also fixup assert condition to detect hotplug address
space overflow.

Signed-off-by: Igor Mammedov <imammedo@redhat.com>
Reported-by: Hu Tao <hutao@cn.fujitsu.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
6 years agonuma: check for busy memory backend
Hu Tao [Mon, 30 Jun 2014 10:28:15 +0000 (18:28 +0800)] 
numa: check for busy memory backend

Specifying the same memory backend twice leads to an assert:

./x86_64-softmmu/qemu-system-x86_64 -m 512M -enable-kvm -object
memory-backend-ram,size=256M,id=ram0 -numa node,nodeid=0,memdev=ram0
-numa node,nodeid=1,memdev=ram0
qemu-system-x86_64: /scm/qemu/memory.c:1506:
memory_region_add_subregion_common: Assertion `!subregion->container'
failed.
Aborted (core dumped)

Detect and exit with an error message instead.

Reviewed-by: Igor Mammedov <imammedo@redhat.com>
Signed-off-by: Hu Tao <hutao@cn.fujitsu.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
6 years agoqtest: enable vhost-user-test
Nikolay Nikolaev [Thu, 19 Jun 2014 17:35:42 +0000 (20:35 +0300)] 
qtest: enable vhost-user-test

Use qtest-obj-y to get the right library order. CONFIG_POSIX ensures
mingw compilation won't break.

Signed-off-by: Nikolay Nikolaev <n.nikolaev@virtualopensystems.com>
Acked-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
MST: whitespace tweak

6 years agoMAINTAINERS: seccomp: change email contact for Eduardo Otubo
Eduardo Otubo [Wed, 2 Jul 2014 08:04:30 +0000 (10:04 +0200)] 
MAINTAINERS: seccomp: change email contact for Eduardo Otubo

Signed-off-by: Eduardo Otubo <eduardo.otubo@profitbricks.com>
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
6 years agoUpdate version for v2.1.0-rc0 release v2.1.0-rc0
Peter Maydell [Tue, 1 Jul 2014 17:48:01 +0000 (18:48 +0100)] 
Update version for v2.1.0-rc0 release

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
6 years agonet: add mmsghdr struct check for L2TPV3
Gonglei [Tue, 1 Jul 2014 12:58:08 +0000 (20:58 +0800)] 
net: add mmsghdr struct check for L2TPV3

The mmsghdr struct is only introduced in Linux 2.6.32; add a
configure check for it and disable L2TPV3 on hosts which are
too old to provide it, rather than simply failing to compile.

Reported-by: chenliang <chenliang88@huawei.com>
Signed-off-by: Gonglei <arei.gonglei@huawei.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Message-id: 1404219488-11196-1-git-send-email-arei.gonglei@huawei.com
[PMM: cleaned up commit message and corrected kernel version number]
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
6 years agoMerge remote-tracking branch 'remotes/kraxel/tags/pull-usb-20140701-1' into staging
Peter Maydell [Tue, 1 Jul 2014 15:16:19 +0000 (16:16 +0100)] 
Merge remote-tracking branch 'remotes/kraxel/tags/pull-usb-20140701-1' into staging

usb bugfixes.

# gpg: Signature made Tue 01 Jul 2014 14:51:19 BST using RSA key ID D3E87138
# gpg: Good signature from "Gerd Hoffmann (work) <kraxel@redhat.com>"
# gpg:                 aka "Gerd Hoffmann <gerd@kraxel.org>"
# gpg:                 aka "Gerd Hoffmann (private) <kraxel@gmail.com>"

* remotes/kraxel/tags/pull-usb-20140701-1:
  ccid-card-emulated: use EventNotifier
  usb: initialize libusb_device to avoid crash
  usb: Fix usb-bt-dongle initialization.
  input: fix jumpy mouse cursor with USB mouse emulation

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
6 years agoMerge remote-tracking branch 'remotes/kraxel/tags/pull-vnc-20140701-1' into staging
Peter Maydell [Tue, 1 Jul 2014 14:12:05 +0000 (15:12 +0100)] 
Merge remote-tracking branch 'remotes/kraxel/tags/pull-vnc-20140701-1' into staging

vnc: two bugfixes (by Peter Lieven).

# gpg: Signature made Tue 01 Jul 2014 12:32:19 BST using RSA key ID D3E87138
# gpg: Good signature from "Gerd Hoffmann (work) <kraxel@redhat.com>"
# gpg:                 aka "Gerd Hoffmann <gerd@kraxel.org>"
# gpg:                 aka "Gerd Hoffmann (private) <kraxel@gmail.com>"

* remotes/kraxel/tags/pull-vnc-20140701-1:
  ui/vnc: fix potential memory corruption issues
  ui/vnc: limit client_cut_text msg payload size

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
6 years agoccid-card-emulated: use EventNotifier
Paolo Bonzini [Mon, 23 Jun 2014 10:30:36 +0000 (12:30 +0200)] 
ccid-card-emulated: use EventNotifier

Shut up Coverity's complaint about unchecked fcntl return values,
and especially make the code simpler and more efficient.

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
6 years agoMerge remote-tracking branch 'remotes/stefanha/tags/tracing-pull-request' into staging
Peter Maydell [Tue, 1 Jul 2014 13:21:50 +0000 (14:21 +0100)] 
Merge remote-tracking branch 'remotes/stefanha/tags/tracing-pull-request' into staging

Tracing pull request

# gpg: Signature made Tue 01 Jul 2014 09:56:27 BST using RSA key ID 81AB73C8
# gpg: Good signature from "Stefan Hajnoczi <stefanha@redhat.com>"
# gpg:                 aka "Stefan Hajnoczi <stefanha@gmail.com>"

* remotes/stefanha/tags/tracing-pull-request:
  trace: add qemu_system_powerdown_request and qemu_system_shutdown_request trace events

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
6 years agoMerge remote-tracking branch 'remotes/stefanha/tags/block-pull-request' into staging
Peter Maydell [Tue, 1 Jul 2014 12:13:04 +0000 (13:13 +0100)] 
Merge remote-tracking branch 'remotes/stefanha/tags/block-pull-request' into staging

Block pull request

# gpg: Signature made Tue 01 Jul 2014 09:47:15 BST using RSA key ID 81AB73C8
# gpg: Good signature from "Stefan Hajnoczi <stefanha@redhat.com>"
# gpg:                 aka "Stefan Hajnoczi <stefanha@gmail.com>"

* remotes/stefanha/tags/block-pull-request: (23 commits)
  block: add backing-file option to block-stream
  block: extend block-commit to accept a string for the backing file
  block: add helper function to determine if a BDS is in a chain
  block: add QAPI command to allow live backing file change
  qapi: Change back sector-count to sectors-count in quorum QAPI events.
  block/cow: Avoid use of uninitialized cow_bs in error path
  block: simplify bdrv_find_base() and bdrv_find_overlay()
  block: make 'top' argument to block-commit optional
  iotests: Add more tests to quick group
  iotests: Add qemu tests to quick group
  iotests: Simplify qemu-iotests-quick.sh
  qemu-img create: add 'nocow' option
  virtio-blk: remove need for explicit x-data-plane=on option
  qdev: drop iothread property type
  virtio-blk: replace x-iothread with iothread link property
  virtio-blk: move qdev properties into virtio-blk.c
  virtio: fix virtio-blk child refcount in transports
  virtio-blk: drop virtio_blk_set_conf()
  virtio-blk: use aliases instead of duplicate qdev properties
  qdev: add qdev_alias_all_properties()
  ...

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
6 years agoui/vnc: fix potential memory corruption issues
Peter Lieven [Mon, 30 Jun 2014 08:57:51 +0000 (10:57 +0200)] 
ui/vnc: fix potential memory corruption issues

this patch makes the VNC server work correctly if the
server surface and the guest surface have different sizes.

Basically the server surface is adjusted to not exceed VNC_MAX_WIDTH
x VNC_MAX_HEIGHT and additionally the width is rounded up to multiple of
VNC_DIRTY_PIXELS_PER_BIT.

If we have a resolution whose width is not dividable by VNC_DIRTY_PIXELS_PER_BIT
we now get a small black bar on the right of the screen.

If the surface is too big to fit the limits only the upper left area is shown.

On top of that this fixes 2 memory corruption issues:

The first was actually discovered during playing
around with a Windows 7 vServer. During resolution
change in Windows 7 it happens sometimes that Windows
changes to an intermediate resolution where
server_stride % cmp_bytes != 0 (in vnc_refresh_server_surface).
This happens only if width % VNC_DIRTY_PIXELS_PER_BIT != 0.

The second is a theoretical issue, but is maybe exploitable
by the guest. If for some reason the guest surface size is bigger
than VNC_MAX_WIDTH x VNC_MAX_HEIGHT we end up in severe corruption since
this limit is nowhere enforced.

Signed-off-by: Peter Lieven <pl@kamp.de>
Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
6 years agoui/vnc: limit client_cut_text msg payload size
Peter Lieven [Mon, 30 Jun 2014 08:07:54 +0000 (10:07 +0200)] 
ui/vnc: limit client_cut_text msg payload size

currently a malicious client could define a payload
size of 2^32 - 1 bytes and send up to that size of
data to the vnc server. The server would allocated
that amount of memory which could easily create an
out of memory condition.

This patch limits the payload size to 1MB max.

Please note that client_cut_text messages are currently
silently ignored.

Signed-off-by: Peter Lieven <pl@kamp.de>
Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
6 years agousb: initialize libusb_device to avoid crash
Jincheng Miao [Fri, 20 Jun 2014 06:12:52 +0000 (14:12 +0800)] 
usb: initialize libusb_device to avoid crash

If libusb_get_device_list() fails, the uninitialized local variable
libusb_device would be passed to libusb_free_device_list(), that
will cause a crash, like:
(gdb) bt
 #0  0x00007fbbb4bafc10 in pthread_mutex_lock () from /lib64/libpthread.so.0
 #1  0x00007fbbb233e653 in libusb_unref_device (dev=0x6275682d627375)
     at core.c:902
 #2  0x00007fbbb233e739 in libusb_free_device_list (list=0x7fbbb6e8436e,
     unref_devices=<optimized out>) at core.c:653
 #3  0x00007fbbb6cd80a4 in usb_host_auto_check (unused=unused@entry=0x0)
     at hw/usb/host-libusb.c:1446
 #4  0x00007fbbb6cd8525 in usb_host_initfn (udev=0x7fbbbd3c5670)
     at hw/usb/host-libusb.c:912
 #5  0x00007fbbb6cc123b in usb_device_init (dev=0x7fbbbd3c5670)
     at hw/usb/bus.c:106
 ...

So initialize libusb_device at the begin time.

Signed-off-by: Jincheng Miao <jmiao@redhat.com>
Reviewed-by: Gonglei <arei.gonglei@huawei.com>
Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
6 years agousb: Fix usb-bt-dongle initialization.
Hani Benhabiles [Tue, 17 Jun 2014 23:23:34 +0000 (00:23 +0100)] 
usb: Fix usb-bt-dongle initialization.

Due to an incomplete initialization, adding a usb-bt-dongle device through HMP
or QMP will cause a segmentation fault.

Signed-off-by: Hani Benhabiles <hani@linux.com>
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
6 years agoinput: fix jumpy mouse cursor with USB mouse emulation
Christian Burger [Sat, 14 Jun 2014 19:19:41 +0000 (20:19 +0100)] 
input: fix jumpy mouse cursor with USB mouse emulation

Guest mouse pointer was jumpy, when moving host mouse in the vertical direction (see bug #1327800).

Signed-off-by: Christian Burger <christian@krikkel.de>
Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
6 years agoMerge remote-tracking branch 'remotes/bonzini/memory' into staging
Peter Maydell [Tue, 1 Jul 2014 10:55:48 +0000 (11:55 +0100)] 
Merge remote-tracking branch 'remotes/bonzini/memory' into staging

* remotes/bonzini/memory:
  qdev: correctly send DEVICE_DELETED for recursively-deleted devices
  memory: do not give a name to the internal exec.c regions
  memory: MemoryRegion: Add size property
  memory: MemoryRegion: Add may-overlap and priority props
  memory: MemoryRegion: Add container and addr props
  memory: MemoryRegion: replace owner field with QOM parent
  memory: MemoryRegion: QOMify
  memory: MemoryRegion: use /machine as default owner
  libqtest: escape strings in QMP commands, fix leak
  qom: object: Ignore refs/unrefs of NULL
  qom: object: remove parent pointer when unparenting
  mc146818rtc: add "rtc-time" link to "/machine/rtc"
  qom: allow creating an alias of a child<> property
  qom: add a generic mechanism to resolve paths
  qom: add object_property_add_alias()

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
6 years agoMerge remote-tracking branch 'remotes/afaerber/tags/qom-devices-for-2.1' into staging
Peter Maydell [Tue, 1 Jul 2014 10:00:52 +0000 (11:00 +0100)] 
Merge remote-tracking branch 'remotes/afaerber/tags/qom-devices-for-2.1' into staging

QOM and device refactorings

* QOM unparenting cleanup
* IRQ conversion to QOM

# gpg: Signature made Tue 01 Jul 2014 04:03:23 BST using RSA key ID 3E7E013F
# gpg: Good signature from "Andreas Färber <afaerber@suse.de>"
# gpg:                 aka "Andreas Färber <afaerber@suse.com>"

* remotes/afaerber/tags/qom-devices-for-2.1:
  irq: Slim conversion of qemu_irq to QOM
  irq: Allocate IRQs individually
  hw: Fix qemu_allocate_irqs() leaks
  sdhci: Fix misuse of qemu_free_irqs()
  qom: Remove parent pointer when unparenting

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
6 years agoMerge remote-tracking branch 'remotes/bonzini/scsi-next' into staging
Peter Maydell [Tue, 1 Jul 2014 09:28:52 +0000 (10:28 +0100)] 
Merge remote-tracking branch 'remotes/bonzini/scsi-next' into staging

* remotes/bonzini/scsi-next:
  configure: Fix -lm test, so that tools can be compiled on hosts that require -lm
  virtio-scsi: scsi events must be converted to target endianness
  virtio-scsi: virtio_scsi_push_event() lacks VirtIOSCSIReq parsing

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
6 years agotrace: add qemu_system_powerdown_request and qemu_system_shutdown_request trace events
Yang Zhiyong [Sat, 21 Jun 2014 18:43:03 +0000 (02:43 +0800)] 
trace: add qemu_system_powerdown_request and qemu_system_shutdown_request trace events

We have the experience that the guest doesn't stop successfully
though it was instructed to shut down.

The root cause may be not in QEMU mostly.  However, QEMU is often
suspected at the beginning just because the issue occurred in
virtualization environment.

Therefore, we need to affirm that QEMU received the shutdown
request and raised ACPI irq from "virsh shutdown" command,
virt-manger or stopping QEMU process to the VM .
So that we can affirm the problems was belonged to the Guset OS
rather than the QEMU itself.

When we stop guests by "virsh shutdown" command or virt-manger,
or stopping QEMU process, qemu_system_powerdown_request() or
qemu_system_shutdown_request() is called. Then the below functions
in main_loop_should_exit() of Vl.c are called roughly in the
following order.

if (qemu_powerdown_requested())
qemu_system_powerdown()
monitor_protocol_event(QEVENT_POWERDOWN, NULL)

OR

if(qemu_shutdown_requested()}
monitor_protocol_event(QEVENT_SHUTDOWN, NULL);

The tracepoint of monitor_protocol_event() already exists, but no
tracepoints are defined for qemu_system_powerdown_request() and
qemu_system_shutdown_request(). So this patch adds two tracepoints for
the two functions. We believe that it will become much easier to
isolate the problem mentioned above by these tracepoints.

Signed-off-by: Yang Zhiyong <yangzy.fnst@cn.fujitsu.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
6 years agoblock: add backing-file option to block-stream
Jeff Cody [Wed, 25 Jun 2014 19:40:11 +0000 (15:40 -0400)] 
block: add backing-file option to block-stream

On some image chains, QEMU may not always be able to resolve the
filenames properly, when updating the backing file of an image
after a block job.

For instance, certain relative pathnames may fail, or drives may
have been specified originally by file descriptor (e.g. /dev/fd/???),
or a relative protocol pathname may have been used.

In these instances, QEMU may lack the information to be able to make
the correct choice, but the user or management layer most likely does
have that knowledge.

With this extension to the block-stream api, the user is able to change
the backing file of the active layer as part of the block-stream
operation.

This allows the change to be 'safe', in the sense that if the attempt
to write the active image metadata fails, then the block-stream
operation returns failure, without disrupting the guest.

If a backing file string is not specified in the command, the backing
file string to use is determined in the same manner as it was
previously.

Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Jeff Cody <jcody@redhat.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
6 years agoblock: extend block-commit to accept a string for the backing file
Jeff Cody [Wed, 25 Jun 2014 19:40:10 +0000 (15:40 -0400)] 
block: extend block-commit to accept a string for the backing file

On some image chains, QEMU may not always be able to resolve the
filenames properly, when updating the backing file of an image
after a block commit.

For instance, certain relative pathnames may fail, or drives may
have been specified originally by file descriptor (e.g. /dev/fd/???),
or a relative protocol pathname may have been used.

In these instances, QEMU may lack the information to be able to make
the correct choice, but the user or management layer most likely does
have that knowledge.

With this extension to the block-commit api, the user is able to change
the backing file of the overlay image as part of the block-commit
operation.

This allows the change to be 'safe', in the sense that if the attempt
to write the overlay image metadata fails, then the block-commit
operation returns failure, without disrupting the guest.

If the commit top is the active layer, then specifying the backing
file string will be treated as an error (there is no overlay image
to modify in that case).

If a backing file string is not specified in the command, the backing
file string to use is determined in the same manner as it was
previously.

Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Jeff Cody <jcody@redhat.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
6 years agoblock: add helper function to determine if a BDS is in a chain
Jeff Cody [Wed, 25 Jun 2014 19:40:09 +0000 (15:40 -0400)] 
block: add helper function to determine if a BDS is in a chain

This is a small helper function, to determine if 'base' is in the
chain of BlockDriverState 'top'.  It returns true if it is in the chain,
and false otherwise.

If either argument is NULL, it will also return false.

Reviewed-by: Benoit Canet <benoit@irqsave.net>
Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Jeff Cody <jcody@redhat.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
6 years agoblock: add QAPI command to allow live backing file change
Jeff Cody [Tue, 1 Jul 2014 07:52:16 +0000 (09:52 +0200)] 
block: add QAPI command to allow live backing file change

This allows a user to make a live change to the backing file recorded in
an open image.

The image file to modify can be specified 2 ways:

1) image filename
2) image node-name

Note: this does not cause the backing file itself to be reopened; it
merely changes the backing filename in the image file structure, and
in internal BDS structures.

It is the responsibility of the user to pass a filename string that
can be resolved when the image chain is reopened, and the filename
string is not validated.

A good analogy for this command is that it is a live version of
'qemu-img rebase -u', with respect to changing the backing file string.

[Jeff is offline so I respun this patch in his absence.  Dropped image
filename since using node-name is preferred and this is a new command.
No need to introduce the limitations of finding images by filename.
--Stefan]

Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Jeff Cody <jcody@redhat.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
6 years agoconfigure: Fix -lm test, so that tools can be compiled on hosts that require -lm
Alexey Kardashevskiy [Tue, 1 Jul 2014 07:30:27 +0000 (17:30 +1000)] 
configure: Fix -lm test, so that tools can be compiled on hosts that require -lm

The existing test whether "-lm" needs to be included or not is
insufficient as it reports false negative on Fedora20/ppc64.
This happens because sin(0.0) is a constant value which compiler
can safely throw away and therefore there is no need to add "-lm".
As the result, qemu-nbd/qemu-io/qemu-img tools cannot compile.

This adds a global variable and uses it in the test to prevent
from optimization.

Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
[Use Peter's improvement on the test to fool LTO, and remove the
 now useless -lm addition in Makefile.target. - Paolo]
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
6 years agoqdev: correctly send DEVICE_DELETED for recursively-deleted devices
Paolo Bonzini [Thu, 26 Jun 2014 13:10:03 +0000 (15:10 +0200)] 
qdev: correctly send DEVICE_DELETED for recursively-deleted devices

When a device is unparented (i.e. made completely hidden from management)
we want to send a DEVICE_DELETED event only if the device actually was
realized.  This avoids raising DEVICE_DELETED events when device_add
fails.

However, this does not work right for recursively-deleted
devices: the whole tree is _first_ unrealized, _then_ unparented.
Then device_unparent sees realized==false and fails to trigger
the event.  The solution is simply to move have_realized into
the DeviceState struct.  If device_add fails, we never set the
new field to true and DEVICE_DELETED is not sent.

Fixes qemu-iotests testcase 067 (broken by commit 5942a19, though that
commit in turn fixed a possible segfault in the same test).

Reported-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Markus Armbruster <armbru@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
6 years agomemory: do not give a name to the internal exec.c regions
Paolo Bonzini [Fri, 13 Jun 2014 08:48:06 +0000 (10:48 +0200)] 
memory: do not give a name to the internal exec.c regions

There is no need to have them visible under /machine.

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
6 years agomemory: MemoryRegion: Add size property
Peter Crosthwaite [Fri, 6 Jun 2014 06:17:35 +0000 (23:17 -0700)] 
memory: MemoryRegion: Add size property

To allow devices to dynamically resize the device. The motivation is
to allow devices with variable size to init their memory_region
without size early and then correctly populate size at realize() time.

Signed-off-by: Peter Crosthwaite <peter.crosthwaite@xilinx.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
6 years agomemory: MemoryRegion: Add may-overlap and priority props
Peter Crosthwaite [Fri, 6 Jun 2014 06:17:01 +0000 (23:17 -0700)] 
memory: MemoryRegion: Add may-overlap and priority props

QOM propertyify the .may-overlap and .priority fields. The setters
will re-add the memory as a subregion if needed (i.e. the values change
when the memory region is already contained).

Signed-off-by: Peter Crosthwaite <peter.crosthwaite@xilinx.com>
[Remove setters. - Paolo]
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
6 years agomemory: MemoryRegion: Add container and addr props
Peter Crosthwaite [Fri, 6 Jun 2014 06:16:27 +0000 (23:16 -0700)] 
memory: MemoryRegion: Add container and addr props

Expose the already existing .parent and .addr fields as QOM properties.
.parent (i.e. the field describing the memory region that contains this
one in Memory hierachy) is renamed "container". This is to avoid
confusion with the QOM parent.

Signed-off-by: Peter Crosthwaite <peter.crosthwaite@xilinx.com>
[Remove setters.  Do not unref parent on releasing the property. Clean
 up error propagation. - Paolo]
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
6 years agomemory: MemoryRegion: replace owner field with QOM parent
Paolo Bonzini [Wed, 11 Jun 2014 08:58:06 +0000 (10:58 +0200)] 
memory: MemoryRegion: replace owner field with QOM parent

The two are now the same.

Reviewed-by: Peter Crosthwaite <peter.crosthwaite@xilinx.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
6 years agomemory: MemoryRegion: QOMify
Peter Crosthwaite [Fri, 6 Jun 2014 06:15:52 +0000 (23:15 -0700)] 
memory: MemoryRegion: QOMify

QOMify memory regions as an Object. The former init() and destroy()
routines become instance_init() and instance_finalize() resp.

memory_region_init() is re-implemented to be:
object_initialize() + set fields

memory_region_destroy() is re-implemented to call unparent().

Signed-off-by: Peter Crosthwaite <peter.crosthwaite@xilinx.com>
[Add newly-created MR as child, unparent on destruction. - Paolo]
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
6 years agomemory: MemoryRegion: use /machine as default owner
Paolo Bonzini [Wed, 11 Jun 2014 08:52:32 +0000 (10:52 +0200)] 
memory: MemoryRegion: use /machine as default owner

This will be added (after QOMification) as the QOM parent.

Reviewed-by: Peter Crosthwaite <peter.crosthwaite@xilinx.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
6 years agolibqtest: escape strings in QMP commands, fix leak
Paolo Bonzini [Fri, 13 Jun 2014 07:35:18 +0000 (09:35 +0200)] 
libqtest: escape strings in QMP commands, fix leak

libqtest is using g_strdup_printf to format QMP commands, but
this does not work if the argument strings need to be escaped.
Instead, use the fancy %-formatting functionality of QObject.
The only change required in tests is that strings have to be
formatted as %s, not '%s' or \"%s\".  Luckily this usage of
parameterized QMP commands is not that frequent.

The leak is in socket_sendf.  Since we are extracting the send
loop to a new function, fix it now.

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
6 years agoqom: object: Ignore refs/unrefs of NULL
Peter Crosthwaite [Fri, 6 Jun 2014 06:13:36 +0000 (23:13 -0700)] 
qom: object: Ignore refs/unrefs of NULL

Just do nothing if passed NULL for a ref or unref. This avoids
call sites that manage a combination of NULL or non-NULL pointers
having to add iffery around every ref and unref.

Signed-off-by: Peter Crosthwaite <peter.crosthwaite@xilinx.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
6 years agoqom: object: remove parent pointer when unparenting
Peter Crosthwaite [Tue, 27 May 2014 00:39:51 +0000 (17:39 -0700)] 
qom: object: remove parent pointer when unparenting

Certain parts of the QOM framework test this pointer to determine if
an object is parented. Nuke it when the object is unparented to allow
for reuse of an object after unparenting.

Signed-off-by: Peter Crosthwaite <peter.crosthwaite@xilinx.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
6 years agomc146818rtc: add "rtc-time" link to "/machine/rtc"
Marcelo Tosatti [Wed, 4 Jun 2014 17:52:03 +0000 (14:52 -0300)] 
mc146818rtc: add "rtc-time" link to "/machine/rtc"

Add a link to rtc under /machine providing a stable
location for management apps to query the value of the
time.  The link should be added by any object that sends
RTC_TIME_CHANGE events.

{"execute":"qom-get","arguments":{"path":"/machine","property":"rtc-time"} }

Suggested by Paolo Bonzini and Andreas Faerber.

Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
6 years agoqom: allow creating an alias of a child<> property
Paolo Bonzini [Tue, 10 Jun 2014 09:17:35 +0000 (11:17 +0200)] 
qom: allow creating an alias of a child<> property

Child properties must be unique.  Fix this problem by
turning their aliases into links.

The resolve function that forwards to the target property
does not have any knowledge of the target property's type,
so it works fine.

Reviewed-by: Peter Crosthwaite <peter.crosthwaite@xilinx.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
6 years agoqom: add a generic mechanism to resolve paths
Paolo Bonzini [Thu, 5 Jun 2014 11:11:51 +0000 (13:11 +0200)] 
qom: add a generic mechanism to resolve paths

It may be desirable to have custom link<> properties that do more
than just store an object.  Even the addition of a "check"
function is not enough if setting the link has side effects
or if a non-standard reference counting is preferrable.

Avoid the assumption that the opaque field of a link<> is a
LinkProperty struct, by adding a generic "resolve" callback
to ObjectProperty.  This fixes aliases of link properties.

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
6 years agoqapi: Change back sector-count to sectors-count in quorum QAPI events.
Benoît Canet [Mon, 30 Jun 2014 15:05:41 +0000 (17:05 +0200)] 
qapi: Change back sector-count to sectors-count in quorum QAPI events.

fe069d9d had aligned code and documentation while dropping the s from the
actual JSON output. Fix that.

This also fix test/qemu-iotest/081 since the missing s was causing a permutation.

Signed-off-by: Benoit Canet <benoit@irqsave.net>
Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
6 years agoblock/cow: Avoid use of uninitialized cow_bs in error path
Peter Maydell [Mon, 30 Jun 2014 18:03:37 +0000 (19:03 +0100)] 
block/cow: Avoid use of uninitialized cow_bs in error path

Commit 25814e8987 introduced an error-exit code path which does
a "goto exit" before the cow_bs variable is initialized, meaning
we would call bdrv_unref() on an uninitialized variable and
likely segfault. Fix this by moving the NULL-initialization
to the top of the function and making the exit code path handle
the case where it is NULL.

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
6 years agoblock: simplify bdrv_find_base() and bdrv_find_overlay()
Jeff Cody [Wed, 25 Jun 2014 19:35:26 +0000 (15:35 -0400)] 
block: simplify bdrv_find_base() and bdrv_find_overlay()

This simplifies the function bdrv_find_overlay().  With this change,
bdrv_find_base() is just a subset of usage of bdrv_find_overlay(),
so this also takes advantage of that.

Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Benoit Canet <benoit@irqsave.net>
Signed-off-by: Jeff Cody <jcody@redhat.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
6 years agoblock: make 'top' argument to block-commit optional
Jeff Cody [Mon, 30 Jun 2014 13:14:15 +0000 (15:14 +0200)] 
block: make 'top' argument to block-commit optional

Now that active layer block-commit is supported, the 'top' argument
no longer needs to be mandatory.

Change it to optional, with the default being the active layer in the
device chain.

[kwolf: Rebased and resolved conflict in tests/qemu-iotests/040]

Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Benoit Canet <benoit@irqsave.net>
Signed-off-by: Jeff Cody <jcody@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
6 years agoiotests: Add more tests to quick group
Max Reitz [Fri, 27 Jun 2014 20:47:48 +0000 (22:47 +0200)] 
iotests: Add more tests to quick group

While at it, add some more tests to the quick group (those that run with
-nocache in under three seconds on my HDD).

Signed-off-by: Max Reitz <mreitz@redhat.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
6 years agoiotests: Add qemu tests to quick group
Max Reitz [Fri, 27 Jun 2014 20:47:47 +0000 (22:47 +0200)] 
iotests: Add qemu tests to quick group

Now that qemu-iotests-quick.sh supports tests using the qemu binary, we
are free to add such tests to the quick group.

Signed-off-by: Max Reitz <mreitz@redhat.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
6 years agoiotests: Simplify qemu-iotests-quick.sh
Max Reitz [Fri, 27 Jun 2014 20:47:46 +0000 (22:47 +0200)] 
iotests: Simplify qemu-iotests-quick.sh

As of the "iotests: Allow out-of-tree run" series, the qemu-iotests may
(and should) be run directly in the build tree and will then guess the
binary paths themselves. Therefore, qemu-iotests-quick.sh does not need
to (and should not) enter the source path anymore; also, it does not
need to specify the binaries because "check" will guess them
automatically.

As a side-effect, tests using qemu may now be added to the quick group.

Signed-off-by: Max Reitz <mreitz@redhat.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
6 years agoqemu-img create: add 'nocow' option
Chunyan Liu [Mon, 30 Jun 2014 06:29:58 +0000 (14:29 +0800)] 
qemu-img create: add 'nocow' option

Add 'nocow' option so that users could have a chance to set NOCOW flag to
newly created files. It's useful on btrfs file system to enhance performance.

Btrfs has low performance when hosting VM images, even more when the guest
in those VM are also using btrfs as file system. One way to mitigate this bad
performance is to turn off COW attributes on VM files. Generally, there are
two ways to turn off NOCOW on btrfs: a) by mounting fs with nodatacow, then
all newly created files will be NOCOW. b) per file. Add the NOCOW file
attribute. It could only be done to empty or new files.

This patch tries the second way, according to the option, it could add NOCOW
per file.

For most block drivers, since the create file step is in raw-posix.c, so we
can do setting NOCOW flag ioctl in raw-posix.c only.

But there are some exceptions, like block/vpc.c and block/vdi.c, they are
creating file by calling qemu_open directly. For them, do the same setting
NOCOW flag ioctl work in them separately.

[Fixed up 082.out due to the new 'nocow' creation option
--Stefan]

Signed-off-by: Chunyan Liu <cyliu@suse.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
6 years agovirtio-scsi: scsi events must be converted to target endianness
Cédric Le Goater [Mon, 30 Jun 2014 15:17:17 +0000 (17:17 +0200)] 
virtio-scsi: scsi events must be converted to target endianness

Virtio SCSI Events need to be byteswapped before being pushed
when host and guest have a different endianness. Not doing so
breaks hotplug of virtio scsi disks, with the following error
message being printed in the guest console:

virtio_scsi: Unsupport virtio scsi event 1000000

This issue got uncovered while testing disk hotplug with a PowerKVM
ppc64le guest. I have checked that this issue also affects a x86_64
guest run on a ppc64 host.

Signed-off-by: Cédric Le Goater <clg@fr.ibm.com>
[ Ported from PowerKVM,
  Greg Kurz <gkurz@linux.vnet.ibm.com> ]
Signed-off-by: Greg Kurz <gkurz@linux.vnet.ibm.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
6 years agovirtio-scsi: virtio_scsi_push_event() lacks VirtIOSCSIReq parsing
Greg Kurz [Mon, 30 Jun 2014 15:33:18 +0000 (17:33 +0200)] 
virtio-scsi: virtio_scsi_push_event() lacks VirtIOSCSIReq parsing

Hotplug of a virtio scsi disk is currently broken: no disk appears in the
guest (verified with a fedora 20 host running a fedora 20 guest with KVM).
Bisect leeds to Paolo's patches to support any_layout, especially this
commit:

commit 36b15c79aa1bef5fe7543f9f2629b6413720bbfb
Author: Paolo Bonzini <pbonzini@redhat.com>
Date:   Tue Jun 10 16:21:18 2014 +0200

    virtio-scsi: start preparing for any_layout

It modifies virtio_scsi_pop_req() so that it is up to the callers to parse
the virtio scsi request. It seems that virtio_scsi_push_event() was not
modified accordingly...

This patch adds a call to virtio_scsi_parse_req(). It also drops some
sanity checks that are already performed by virtio_scsi_parse_req().

Signed-off-by: Greg Kurz <gkurz@linux.vnet.ibm.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
6 years agoqom: add object_property_add_alias()
Stefan Hajnoczi [Wed, 18 Jun 2014 09:58:28 +0000 (17:58 +0800)] 
qom: add object_property_add_alias()

Sometimes an object needs to present a property which is actually on
another object, or it needs to provide an alias name for an existing
property.

Examples:
  a.foo -> b.foo
  a.old_name -> a.new_name

The new object_property_add_alias() API allows objects to alias a
property on the same object or another object.  The source and target
names can be different.

Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Reviewed-by: Peter Crosthwaite <peter.crosthwaite@xilinx.com>
6 years agovirtio-blk: remove need for explicit x-data-plane=on option
Stefan Hajnoczi [Tue, 10 Jun 2014 07:03:22 +0000 (09:03 +0200)] 
virtio-blk: remove need for explicit x-data-plane=on option

The x-data-plane=on|off option is no longer useful because the
iothread=<iothread> option conveys the same information plus which
IOThread to use.

Do not delete x-data-plane=on|off yet as a convenience to people using
this legacy experimental option.  We will drop it in QEMU 2.2.

Instead, turn on data-plane when either x-data-plane=on or
iothread=<iothread> are used.  The following command-line uses
data-plane:

  qemu -device virtio-blk-pci,iothread=foo,drive=drive0

Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
6 years agoqdev: drop iothread property type
Stefan Hajnoczi [Tue, 10 Jun 2014 07:03:21 +0000 (09:03 +0200)] 
qdev: drop iothread property type

The iothread property type is no longer used and can be removed.

Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
6 years agovirtio-blk: replace x-iothread with iothread link property
Stefan Hajnoczi [Tue, 10 Jun 2014 07:03:20 +0000 (09:03 +0200)] 
virtio-blk: replace x-iothread with iothread link property

Up until now -device virtio-blk-pci,x-iothread=<id> was used to assign
an IOThread.  This was a temporary solution while we cleaned up QOM link
properties.

This patch switches over to a QOM link property since it is now possible
to restrict the setter to unrealized instances and automatically unref
the IOThread when the virtio-blk-pci device is freed.

Since the "iothread" property is a QOM property and not a qdev property,
we must alias it explicitly for virtio-blk-pci, as well as CCW and
s390-virtio.

Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
6 years agovirtio-blk: move qdev properties into virtio-blk.c
Stefan Hajnoczi [Wed, 18 Jun 2014 09:58:36 +0000 (17:58 +0800)] 
virtio-blk: move qdev properties into virtio-blk.c

There is no need to make DEFINE_VIRTIO_BLK_PROPERTIES() public.  Inline
it into virtio-blk.c so it cannot be used by mistake from other source
files.

Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Reviewed-by: Peter Crosthwaite <peter.crosthwaite@xilinx.com>
6 years agovirtio: fix virtio-blk child refcount in transports
Stefan Hajnoczi [Wed, 18 Jun 2014 09:58:35 +0000 (17:58 +0800)] 
virtio: fix virtio-blk child refcount in transports

object_initialize() leaves the object with a refcount of 1.
object_property_add_child() adds its own reference which is dropped
again when the property is deleted.

The upshot of this is that we always have a refcount >= 1.  Upon hot
unplug the virtio-blk child is not finalized!

Drop our reference after the child property has been added to the
parent.

Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Reviewed-by: Peter Crosthwaite <peter.crosthwaite@xilinx.com>
6 years agovirtio-blk: drop virtio_blk_set_conf()
Stefan Hajnoczi [Wed, 18 Jun 2014 09:58:34 +0000 (17:58 +0800)] 
virtio-blk: drop virtio_blk_set_conf()

This function is no longer used since parent objects now use child
aliases to set the VirtIOBlkConf directly.

Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Reviewed-by: Peter Crosthwaite <peter.crosthwaite@xilinx.com>
6 years agovirtio-blk: use aliases instead of duplicate qdev properties
Stefan Hajnoczi [Wed, 18 Jun 2014 09:58:33 +0000 (17:58 +0800)] 
virtio-blk: use aliases instead of duplicate qdev properties

virtio-blk-pci, virtio-blk-s390, and virtio-blk-ccw all duplicate the
qdev properties of their VirtIOBlock child.  This approach does not work
well with string or pointer properties since we must be careful about
leaking or double-freeing them.

Use the QOM alias property to forward property accesses to the
VirtIOBlock child.  This way no duplication is necessary.

Remember to stop calling virtio_blk_set_conf() so that we don't clobber
the values already set on the VirtIOBlock instance.

Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Reviewed-by: Peter Crosthwaite <peter.crosthwaite@xilinx.com>
6 years agoqdev: add qdev_alias_all_properties()
Stefan Hajnoczi [Wed, 18 Jun 2014 09:58:32 +0000 (17:58 +0800)] 
qdev: add qdev_alias_all_properties()

The qdev_alias_all_properties() function creates QOM alias properties
for each qdev property on a DeviceState.  This is useful for parent
objects that wish to forward property accesses to their children.

Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Reviewed-by: Peter Crosthwaite <peter.crosthwaite@xilinx.com>
6 years agovirtio-blk: move x-data-plane qdev property to virtio-blk.h
Stefan Hajnoczi [Wed, 18 Jun 2014 09:58:31 +0000 (17:58 +0800)] 
virtio-blk: move x-data-plane qdev property to virtio-blk.h

Move the x-data-plane property.  Originally it was outside since not
every transport may wish to support dataplane.  But that makes little
sense when we have a dedicated CONFIG_VIRTIO_BLK_DATA_PLANE ifdef
already.

This move makes it easier to switch to property aliases in the next
patch.

Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Reviewed-by: Peter Crosthwaite <peter.crosthwaite@xilinx.com>
6 years agodataplane: bail out on unsupported transport
Cornelia Huck [Wed, 18 Jun 2014 09:58:30 +0000 (17:58 +0800)] 
dataplane: bail out on unsupported transport

If the virtio transport does not support notifiers (like s390-virtio),
we can't use dataplane. Bail out early and let the user know what is
wrong.

Signed-off-by: Cornelia Huck <cornelia.huck@de.ibm.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>