sdk/emulator/qemu.git
12 years agotarget-i386: add API to get dump info
Wen Congyang [Mon, 7 May 2012 04:08:56 +0000 (12:08 +0800)]
target-i386: add API to get dump info

Dump info contains: endian, class and architecture. The next
patch will use these information to create vmcore. Note: on
x86 box, the  class is ELFCLASS64 if the memory is larger than 4G.

Signed-off-by: Wen Congyang <wency@cn.fujitsu.com>
Signed-off-by: Luiz Capitulino <lcapitulino@redhat.com>
12 years agotarget-i386: Add API to write cpu status to core file
Wen Congyang [Mon, 7 May 2012 04:08:22 +0000 (12:08 +0800)]
target-i386: Add API to write cpu status to core file

The core file has register's value. But it does not include all registers value.
Store the cpu status into QEMU note, and the user can get more information
from vmcore. If you change QEMUCPUState, please count up QEMUCPUSTATE_VERSION.

Signed-off-by: Wen Congyang <wency@cn.fujitsu.com>
Signed-off-by: Luiz Capitulino <lcapitulino@redhat.com>
12 years agotarget-i386: Add API to write elf notes to core file
Wen Congyang [Mon, 7 May 2012 04:07:48 +0000 (12:07 +0800)]
target-i386: Add API to write elf notes to core file

The core file contains register's value. These APIs write registers to
core file, and them will be called in the following patch.

Signed-off-by: Wen Congyang <wency@cn.fujitsu.com>
Signed-off-by: Luiz Capitulino <lcapitulino@redhat.com>
12 years agoAdd API to get memory mapping without do paging
Wen Congyang [Mon, 7 May 2012 04:07:07 +0000 (12:07 +0800)]
Add API to get memory mapping without do paging

crash does not need the virtual address and physical address mapping, and the
mapping does not include the memory that is not referenced by the page table.
crash does not use the virtual address, so we can create the mapping for all
physical memory(virtual address is always 0). This patch provides a API to do
this thing, and it will be used in the following patch.

Signed-off-by: Wen Congyang <wency@cn.fujitsu.com>
Signed-off-by: Luiz Capitulino <lcapitulino@redhat.com>
12 years agoAdd API to get memory mapping
Wen Congyang [Mon, 7 May 2012 04:06:40 +0000 (12:06 +0800)]
Add API to get memory mapping

Add API to get all virtual address and physical address mapping.
If the guest doesn't use paging, the virtual address is equal to the phyical
address. The virtual address and physical address mapping is for gdb's user, and
it does not include the memory that is not referenced by the page table. So if
you want to use crash to anaylze the vmcore, please do not specify -p option.
the reason why the -p option is not default explicitly: guest machine in a
catastrophic state can have corrupted memory, which we cannot trust.

Signed-off-by: Wen Congyang <wency@cn.fujitsu.com>
Signed-off-by: Luiz Capitulino <lcapitulino@redhat.com>
12 years agoAdd API to check whether paging mode is enabled
Wen Congyang [Mon, 7 May 2012 04:05:42 +0000 (12:05 +0800)]
Add API to check whether paging mode is enabled

This API will be used in the following patch.

Signed-off-by: Wen Congyang <wency@cn.fujitsu.com>
Signed-off-by: Luiz Capitulino <lcapitulino@redhat.com>
12 years agoimplement cpu_get_memory_mapping()
Wen Congyang [Mon, 7 May 2012 04:04:57 +0000 (12:04 +0800)]
implement cpu_get_memory_mapping()

Walk cpu's page table and collect all virtual address and physical address mapping.
Then, add these mapping into memory mapping list. If the guest does not use paging,
it will do nothing. Note: the I/O memory will be skipped.

Signed-off-by: Wen Congyang <wency@cn.fujitsu.com>
Signed-off-by: Luiz Capitulino <lcapitulino@redhat.com>
12 years agoAdd API to check whether a physical address is I/O address
Wen Congyang [Mon, 7 May 2012 04:04:18 +0000 (12:04 +0800)]
Add API to check whether a physical address is I/O address

This API will be used in the following patch.

Signed-off-by: Wen Congyang <wency@cn.fujitsu.com>
Signed-off-by: Luiz Capitulino <lcapitulino@redhat.com>
12 years agoAdd API to create memory mapping list
Wen Congyang [Mon, 7 May 2012 04:03:46 +0000 (12:03 +0800)]
Add API to create memory mapping list

The memory mapping list stores virtual address and physical address mapping.
The virtual address and physical address are contiguous in the mapping.
The folloing patch will use this information to create PT_LOAD in the vmcore.

Signed-off-by: Wen Congyang <wency@cn.fujitsu.com>
Signed-off-by: Luiz Capitulino <lcapitulino@redhat.com>
12 years agotarget-microblaze: lwx/swx: first implementation
Peter A. G. Crosthwaite [Fri, 1 Jun 2012 03:23:28 +0000 (13:23 +1000)]
target-microblaze: lwx/swx: first implementation

Signed-off-by: Peter A. G. Crosthwaite <peter.crosthwaite@petalogix.com>
Signed-off-by: Edgar E. Iglesias <edgar.iglesias@gmail.com>
12 years agoRevert "rtl8139: do the network/host communication only in normal operating mode"
Jason Wang [Thu, 10 May 2012 08:37:22 +0000 (16:37 +0800)]
Revert "rtl8139: do the network/host communication only in normal operating mode"

This reverts commit ff71f2e8cacefae99179993204172bc65e4303df. This is because
the linux 8139cp driver would leave the card in "Config Register Write Enable"
mode after the eeprom were read or write ( which is unexpected in the spec
). Also a physical 8139 card can still DMA into host memory in modes other than
Normal mode, so we need revert this commit to align with the behavior of
physical card.

The issue of 8139cp driver should be fixed in linux seperately.

Signed-off-by: Jason Wang <jasowang@redhat.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
12 years agoMerge remote-tracking branch 'qemu-kvm/uq/master' into staging
Anthony Liguori [Sat, 2 Jun 2012 23:56:23 +0000 (07:56 +0800)]
Merge remote-tracking branch 'qemu-kvm/uq/master' into staging

* qemu-kvm/uq/master:
  virtio/vhost: Add support for KVM in-kernel MSI injection
  msix: Add msix_nr_vectors_allocated
  kvm: Enable use of kvm_irqchip_in_kernel in hwlib code
  kvm: Introduce kvm_irqchip_add/remove_irqfd
  kvm: Make kvm_irqchip_commit_routes an internal service
  kvm: Publicize kvm_irqchip_release_virq
  kvm: Introduce kvm_irqchip_add_msi_route
  kvm: Rename kvm_irqchip_add_route to kvm_irqchip_add_irq_route
  msix: Introduce vector notifiers
  msix: Invoke msix_handle_mask_update on msix_mask_all
  msix: Factor out msix_get_message
  kvm: update vmxcap for EPT A/D, INVPCID, RDRAND, VMFUNC
  kvm: Enable in-kernel irqchip support by default
  kvm: Add support for direct MSI injections
  kvm: Update kernel headers
  kvm: x86: Wire up MSI support for in-kernel irqchip
  pc: Enable MSI support at APIC level
  kvm: Introduce basic MSI support for in-kernel irqchips
  Introduce MSIMessage structure
  kvm: Refactor KVMState::max_gsi to gsi_count

12 years agoMerge remote-tracking branch 'kwolf/for-anthony' into staging
Anthony Liguori [Sat, 2 Jun 2012 23:54:05 +0000 (07:54 +0800)]
Merge remote-tracking branch 'kwolf/for-anthony' into staging

* kwolf/for-anthony:
  ahci: SATA FIS is 20 bytes, not 0x20
  virtio-blk: Fix geometry sector calculation
  block: prevent snapshot mode $TMPDIR symlink attack
  sheepdog: fix return value of do_load_save_vm_state
  virtio: Fix compiler warning for non Linux hosts

12 years agoUpdate version to open the 1.2 development branch
Anthony Liguori [Fri, 1 Jun 2012 08:56:16 +0000 (16:56 +0800)]
Update version to open the 1.2 development branch

Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
12 years agoUpdate version for 1.1.0 release
Anthony Liguori [Fri, 1 Jun 2012 08:55:34 +0000 (16:55 +0800)]
Update version for 1.1.0 release

Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
12 years agoUpdate version for 1.1.0-rc4 release
Anthony Liguori [Thu, 31 May 2012 00:53:39 +0000 (08:53 +0800)]
Update version for 1.1.0-rc4 release

Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
12 years agoMerge remote-tracking branch 'origin/master' into staging
Anthony Liguori [Thu, 31 May 2012 00:49:46 +0000 (08:49 +0800)]
Merge remote-tracking branch 'origin/master' into staging

* origin/master:
  pc-bios: Update OpenBIOS images

12 years agopc-bios: Update OpenBIOS images
Blue Swirl [Wed, 30 May 2012 17:04:41 +0000 (17:04 +0000)]
pc-bios: Update OpenBIOS images

Update OpenBIOS images to r1060 built from submodule.

Signed-off-by: Blue Swirl <blauwirbel@gmail.com>
12 years agoahci: SATA FIS is 20 bytes, not 0x20
Daniel Verkamp [Tue, 22 May 2012 23:26:42 +0000 (16:26 -0700)]
ahci: SATA FIS is 20 bytes, not 0x20

As in the SATA and AHCI specifications, a FIS is 5 Dwords of 4 bytes
each, which comes to 20 bytes (decimal), not 0x20.

Signed-off-by: Daniel Verkamp <daniel@drv.nu>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
12 years agovirtio-blk: Fix geometry sector calculation
Christian Borntraeger [Thu, 24 May 2012 11:22:55 +0000 (13:22 +0200)]
virtio-blk: Fix geometry sector calculation

Currently the sector value for the geometry is masked, even if the
user usesa command line parameter that explicitely gives a number.
This breaks dasd devices on s390. A dasd device can have
a physical block size of 4096 (== same for logical block size)
and a typcial geometry of 15 heads and 12 sectors per cyl.
The ibm partition detection relies on a correct geometry
reported by the device. Unfortunately the current code changes
12 to 8. This would be necessary if the total size is
not a multiple of logical sector size,  but for dasd this
is not the case.

This patch checks the device size and only applies sector
mask if necessary.

Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
CC: Christoph Hellwig <hch@lst.de>
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
12 years agoblock: prevent snapshot mode $TMPDIR symlink attack
Jim Meyering [Mon, 28 May 2012 07:27:54 +0000 (09:27 +0200)]
block: prevent snapshot mode $TMPDIR symlink attack

In snapshot mode, bdrv_open creates an empty temporary file without
checking for mkstemp or close failure, and ignoring the possibility
of a buffer overrun given a surprisingly long $TMPDIR.
Change the get_tmp_filename function to return int (not void),
so that it can inform its two callers of those failures.
Also avoid the risk of buffer overrun and do not ignore mkstemp
or close failure.
Update both callers (in block.c and vvfat.c) to propagate
temp-file-creation failure to their callers.

get_tmp_filename creates and closes an empty file, while its
callers later open that presumed-existing file with O_CREAT.
The problem was that a malicious user could provoke mkstemp failure
and race to create a symlink with the selected temporary file name,
thus causing the qemu process (usually root owned) to open through
the symlink, overwriting an attacker-chosen file.

This addresses CVE-2012-2652.
http://bugzilla.redhat.com/CVE-2012-2652

Signed-off-by: Jim Meyering <meyering@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@linux.vnet.ibm.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
12 years agosheepdog: fix return value of do_load_save_vm_state
MORITA Kazutaka [Tue, 29 May 2012 16:05:15 +0000 (01:05 +0900)]
sheepdog: fix return value of do_load_save_vm_state

bdrv_save_vmstate and bdrv_load_vmstate should return the vmstate size
on success, and -errno on error.

Signed-off-by: MORITA Kazutaka <morita.kazutaka@lab.ntt.co.jp>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
12 years agovirtio: Fix compiler warning for non Linux hosts
Stefan Weil [Tue, 22 May 2012 21:23:32 +0000 (23:23 +0200)]
virtio: Fix compiler warning for non Linux hosts

The local variables ret, i are only used if __linux__ is defined.

Signed-off-by: Stefan Weil <sw@weilnetz.de>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
12 years agoMerge remote-tracking branch 'mdroth/qga-pull-5-29-12-v2' into staging
Anthony Liguori [Wed, 30 May 2012 07:02:03 +0000 (15:02 +0800)]
Merge remote-tracking branch 'mdroth/qga-pull-5-29-12-v2' into staging

* mdroth/qga-pull-5-29-12-v2:
  qemu-ga: avoid blocking on atime update when reading /etc/mtab
  qemu-ga: Fix use of environ on Darwin

12 years agoblock: prevent snapshot mode $TMPDIR symlink attack
Jim Meyering [Mon, 28 May 2012 07:27:54 +0000 (09:27 +0200)]
block: prevent snapshot mode $TMPDIR symlink attack

In snapshot mode, bdrv_open creates an empty temporary file without
checking for mkstemp or close failure, and ignoring the possibility
of a buffer overrun given a surprisingly long $TMPDIR.
Change the get_tmp_filename function to return int (not void),
so that it can inform its two callers of those failures.
Also avoid the risk of buffer overrun and do not ignore mkstemp
or close failure.
Update both callers (in block.c and vvfat.c) to propagate
temp-file-creation failure to their callers.

get_tmp_filename creates and closes an empty file, while its
callers later open that presumed-existing file with O_CREAT.
The problem was that a malicious user could provoke mkstemp failure
and race to create a symlink with the selected temporary file name,
thus causing the qemu process (usually root owned) to open through
the symlink, overwriting an attacker-chosen file.

This addresses CVE-2012-2652.
http://bugzilla.redhat.com/CVE-2012-2652

Reviewed-by: Stefan Hajnoczi <stefanha@linux.vnet.ibm.com>
Signed-off-by: Jim Meyering <meyering@redhat.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
12 years agoxhci: add usage info to docs
Gerd Hoffmann [Tue, 29 May 2012 09:20:31 +0000 (11:20 +0200)]
xhci: add usage info to docs

Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
12 years agovnc: fix segfault in vnc_display_pw_expire()
Gerd Hoffmann [Thu, 24 May 2012 08:55:01 +0000 (10:55 +0200)]
vnc: fix segfault in vnc_display_pw_expire()

NULL pointer dereference in case no vnc server is configured.
Catch this and return -EINVAL like vnc_display_password() does.

Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
12 years agoExpose CPUID leaf 7 only for -cpu host
Eduardo Habkost [Mon, 21 May 2012 14:27:02 +0000 (11:27 -0300)]
Expose CPUID leaf 7 only for -cpu host

Changes v2 -> v3;
  - Check for kvm_enabled() before setting cpuid_7_0_ebx_features

Changes v1 -> v2:
  - Use kvm_arch_get_supported_cpuid() instead of host_cpuid() on
    cpu_x86_fill_host().

  We should use GET_SUPPORTED_CPUID for all bits on "-cpu host"
  eventually, but I am not changing all the other CPUID leaves because
  we may not be able to test such an intrusive change in time for 1.1.

Description of the bug:

Since QEMU 0.15, the CPUID information on CPUID[EAX=7,ECX=0] is being
returned unfiltered to the guest, directly from the GET_SUPPORTED_CPUID
return value.

The problem is that this makes the resulting CPU feature flags
unpredictable and dependent on the host CPU and kernel version. This
breaks live-migration badly if migrating from a host CPU that supports
some features on that CPUID leaf (running a recent kernel) to a kernel
or host CPU that doesn't support it.

Migration also is incorrect (the virtual CPU changes under the guest's
feet) if you migrate in the opposite direction (from an old CPU/kernel
to a new CPU/kernel), but with less serious consequences (guests
normally query CPUID information only once on boot).

Fortunately, the bug affects only users using cpudefs with level >= 7.

The right behavior should be to explicitly enable those features on
[cpudef] config sections or on the "-cpu" command-line arguments. Right
now there is no predefined CPU model on QEMU that has those features:
the latest Intel model we have is Sandy Bridge.

I would like to get this fixed on 1.1, so I am submitting this patch,
that enables those features only if "-cpu host" is being used (as we
don't have any pre-defined CPU model that actually have those features).
After 1.1 is released, we can make those features properly configurable
on [cpudef] and -cpu configuration.

One problem is: with this patch, users with the following setup:
- Running QEMU 1.0;
- Using a cpudef having level >= 7;
- Running a kernel that supports the features on CPUID leaf 7; and
- Running on a CPU that supports some features on CPUID leaf 7
won't be able to live-migrate to QEMU 1.1. But for these users
live-migration is already broken (they can't live-migrate to hosts with
older CPUs or older kernels, already), I don't see how to avoid this
problem.

Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
12 years agoqemu-ga: avoid blocking on atime update when reading /etc/mtab
Michael Roth [Tue, 29 May 2012 15:08:50 +0000 (10:08 -0500)]
qemu-ga: avoid blocking on atime update when reading /etc/mtab

Currently we re-read/re-process /etc/mtab to get an updated list of
mounts when guest-fsfreeze-thaw is called. This can cause an atime
update on /etc/mtab, which will block if we're in a frozen state.

Instead, use /proc's version of mtab, which may not be up-to-date with
options passed via -o remount, but is compatible for our use cases since
we only care about the filesystem type.

Reported-by: Matsuda, Daiki <matsudadik@intellilink.co.jp>
Signed-off-by: Michael Roth <mdroth@linux.vnet.ibm.com>
12 years agoqemu-ga: Fix use of environ on Darwin
Andreas Färber [Sun, 27 May 2012 15:02:20 +0000 (17:02 +0200)]
qemu-ga: Fix use of environ on Darwin

Use _NSGetEnviron() helper to access the environment.

Signed-off-by: Andreas Färber <andreas.faerber@web.de>
Cc: Charlie Somerville <charlie@charliesomerville.com>
Signed-off-by: Michael Roth <mdroth@linux.vnet.ibm.com>
12 years agopci: call object_unparent() before free_qdev()
Amos Kong [Sun, 20 May 2012 09:57:45 +0000 (17:57 +0800)]
pci: call object_unparent() before free_qdev()

Start VM with 8 multiple-function block devs, hot-removing
those block devs by 'device_del ...' would cause qemu abort.

| (qemu) device_del virti0-0-0
| (qemu) **
|ERROR:qom/object.c:389:object_delete: assertion failed: (obj->ref == 0)

It's a regression introduced by commit 57c9fafe

The whole PCI slot should be removed once. Currently only one func
is cleaned in pci_unplug_device(), if you try to remove a single
func by monitor cmd.

free_qdev() are called for all functions in slot,
but unparent_delete() is only called for one
function.

Signed-off-by: XXXX
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
12 years agofix multiboot loading if load_end_addr == 0
Scott Moser [Mon, 26 Mar 2012 19:27:00 +0000 (15:27 -0400)]
fix multiboot loading if load_end_addr == 0

The previous multiboot load code did not treat the case where
load_end_addr was 0 specially.  The multiboot specification says the
following:
 * load_end_addr
   Contains the physical address of the end of the data segment.
   (load_end_addr - load_addr) specifies how much data to load. This
   implies that the text and data segments must be consecutive in the
   OS image; this is true for existing a.out executable formats. If
   this field is zero, the boot loader assumes that the text and data
   segments occupy the whole OS image file.

Signed-off-by: Scott Moser <smoser@ubuntu.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
12 years agovga: fix vram double-mapping with -vga std and -M pc-0.12
Avi Kivity [Wed, 9 May 2012 15:23:06 +0000 (18:23 +0300)]
vga: fix vram double-mapping with -vga std and -M pc-0.12

With pc-0.12, we map the video RAM both through the PCI BAR (the guest does
this) and through a fixed mapping at 0xe0000000.  The memory API doesn't allow
this double map, and aborts.

Fix by using an alias.

Reported-by: Michael Tokarev <mjt@tls.msk.ru>
Signed-off-by: Avi Kivity <avi@redhat.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
12 years agoMerge remote-tracking branch 'afaerber-or/cocoa-for-upstream' into staging
Anthony Liguori [Tue, 29 May 2012 11:54:16 +0000 (06:54 -0500)]
Merge remote-tracking branch 'afaerber-or/cocoa-for-upstream' into staging

* afaerber-or/cocoa-for-upstream:
  cocoa: Suppress Cocoa frontend for -qtest
  arch_init: Fix AltiVec build on Darwin/ppc

12 years agococoa: Suppress Cocoa frontend for -qtest
Andreas Färber [Mon, 28 May 2012 01:18:31 +0000 (03:18 +0200)]
cocoa: Suppress Cocoa frontend for -qtest

Signed-off-by: Andreas Färber <andreas.faerber@web.de>
12 years agoarch_init: Fix AltiVec build on Darwin/ppc
Andreas Färber [Sun, 27 May 2012 14:21:02 +0000 (16:21 +0200)]
arch_init: Fix AltiVec build on Darwin/ppc

Commit f29a56147b66845914d0a645bf9b4c5bb9a6af57 (implement
-no-user-config command-line option (v3)) introduced uses of bool
in arch_init.c. Shortly before that usage is support code for
AltiVec (conditional to __ALTIVEC__).

GCC's altivec.h may in a !__APPLE_ALTIVEC__ code path redefine bool,
leading to type mismatches. altivec.h recommends to #undef for C++
compatibility, but doing so in C leads to bool remaining undefined.

Fix by redefining bool to _Bool as mandated for stdbool.h by POSIX.

Signed-off-by: Andreas Färber <andreas.faerber@web.de>
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
12 years agoMerge remote-tracking branch 'sstabellini/for_1.1_rc3' into staging
Anthony Liguori [Tue, 29 May 2012 09:32:13 +0000 (04:32 -0500)]
Merge remote-tracking branch 'sstabellini/for_1.1_rc3' into staging

* sstabellini/for_1.1_rc3:
  Call xc_domain_shutdown with the reboot flag when the guest requests a reboot.
  xen: Fix PV-on-HVM
  xen_disk: properly update stats in ioreq_release()
  xen_disk: use bdrv_aio_flush instead of bdrv_flush
  xen_disk: remove syncwrite option
  xen: disable rtc_clock
  xen: do not initialize the interval timer and PCSPK emulator

12 years agoMerge remote-tracking branch 'mdroth/qga-pull-5-24-12' into staging
Anthony Liguori [Tue, 29 May 2012 09:31:29 +0000 (04:31 -0500)]
Merge remote-tracking branch 'mdroth/qga-pull-5-24-12' into staging

* mdroth/qga-pull-5-24-12:
  qemu-ga: Fix missing environ declaration
  configure: check if environ is declared

12 years agoMerge remote-tracking branch 'kwolf/for-anthony' into staging
Anthony Liguori [Tue, 29 May 2012 09:30:49 +0000 (04:30 -0500)]
Merge remote-tracking branch 'kwolf/for-anthony' into staging

* kwolf/for-anthony:
  fdc-test: introduced qtest no_media_on_start and cmos qtest for floppy
  fdc: fix media detection
  fdc: floppy drive should be visible after start without media
  qemu-iotests: mark 035 qcow2-only
  qcow2: Check qcow2_alloc_clusters_at() return value
  sheepdog: use heap instead of stack for BDRVSheepdogState
  sheepdog: return -errno on error
  sheepdog: mark image as snapshot when tag is specified
  qemu-img: Explain how rebase operation can be used to perform a 'diff' operation.
  qcow2: don't leak buffer for unexpected qcow_version in header

12 years agoMerge remote-tracking branch 'kiszka/queues/slirp' into staging
Anthony Liguori [Tue, 29 May 2012 09:30:00 +0000 (04:30 -0500)]
Merge remote-tracking branch 'kiszka/queues/slirp' into staging

* kiszka/queues/slirp:
  slirp: Avoid redefining MAX_TCPOPTLEN
  slirp: Avoid statements without effect on Big Endian host
  slirp: Untangle TCPOLEN_* from TCPOPT_*

12 years agoMerge remote-tracking branch 'bonzini/scsi-next' into staging
Anthony Liguori [Tue, 29 May 2012 09:28:59 +0000 (04:28 -0500)]
Merge remote-tracking branch 'bonzini/scsi-next' into staging

* bonzini/scsi-next:
  ISCSI: Switch to using READ16/WRITE16 for I/O to the LUN
  ISCSI: Only call READCAPACITY16 for SBC devices, use READCAPACITY10 for MMC
  ISCSI: get device type at connection time
  ISCSI: change num_blocks to 64-bit
  ISCSI: redo how we set up the events
  scsi: declare vmstate_info_scsi_requests to be static

12 years agoslirp: Avoid redefining MAX_TCPOPTLEN
Andreas Färber [Mon, 28 May 2012 17:52:35 +0000 (19:52 +0200)]
slirp: Avoid redefining MAX_TCPOPTLEN

MAX_TCPOPTLEN is being defined as 32. Darwin already has it as 40,
causing a warning. The value is only used to declare an array,
into which currently 4 bytes are written at most.

Therefore always override MAX_TCPOPTLEN for now.

Suggested-by: Jan Kiszka <jan.kiszka@web.de>
Reviewed-by: Stefan Weil <sw@weilnetz.de>
Signed-off-by: Andreas Färber <andreas.faerber@web.de>
Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
12 years agoslirp: Avoid statements without effect on Big Endian host
Andreas Färber [Mon, 28 May 2012 17:34:20 +0000 (19:34 +0200)]
slirp: Avoid statements without effect on Big Endian host

Darwin has HTON*/NTOH* macros that on BE simply return the argument.
This is incompatible with SLIRP's use of these macros as a statement.

Undefine the macros in the HOST_WORDS_BIGENDIAN code path to redefine
these macros as no-op, as already done when they were undefined.

Suggested-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Andreas Färber <andreas.faerber@web.de>
Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
12 years agoISCSI: Switch to using READ16/WRITE16 for I/O to the LUN
Ronnie Sahlberg [Tue, 22 May 2012 10:10:05 +0000 (20:10 +1000)]
ISCSI: Switch to using READ16/WRITE16 for I/O to the LUN

This allows using LUNs bigger than 2TB.  Keep using READ10 for other
device types such as MMC.

Signed-off-by: Ronnie Sahlberg <ronniesahlberg@gmail.com>
12 years agoISCSI: Only call READCAPACITY16 for SBC devices, use READCAPACITY10 for MMC
Ronnie Sahlberg [Sat, 26 May 2012 04:56:38 +0000 (14:56 +1000)]
ISCSI: Only call READCAPACITY16 for SBC devices, use READCAPACITY10 for MMC

Signed-off-by: Ronnie Sahlberg <ronniesahlberg@gmail.com>
12 years agoISCSI: get device type at connection time
Ronnie Sahlberg [Fri, 25 May 2012 11:59:01 +0000 (21:59 +1000)]
ISCSI: get device type at connection time

This is needed to avoid READ CAPACITY(16) for MMC devices.

Signed-off-by: Ronnie Sahlberg <ronniesahlberg@gmail.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
12 years agoISCSI: change num_blocks to 64-bit
Paolo Bonzini [Sat, 26 May 2012 07:41:13 +0000 (09:41 +0200)]
ISCSI: change num_blocks to 64-bit

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
12 years agoISCSI: redo how we set up the events
Ronnie Sahlberg [Tue, 22 May 2012 09:56:36 +0000 (19:56 +1000)]
ISCSI: redo how we set up the events

Call qemu_notify_event() after updating events.  Otherwise, If we add
an event for -is-writeable but the socket is already writeable there
may be a delay before the event callback is actually triggered.

Those delays would in particular hurt performance during BIOS boot and
when the GRUB bootloader reads the kernel and initrd.

But first call out to the socket write functions directly, and only set up
the write event if the socket is full.  This will happen very rarely and
this improves performance.

Signed-off-by: Ronnie Sahlberg <ronniesahlberg@gmail.com>
12 years agoslirp: Untangle TCPOLEN_* from TCPOPT_*
Andreas Färber [Fri, 27 Apr 2012 22:29:05 +0000 (00:29 +0200)]
slirp: Untangle TCPOLEN_* from TCPOPT_*

Commit b72210568ef0c0fb141a01cffb71a09c4efa0364 (slirp: clean up
conflicts with system headers) enclosed TCPOLEN_MAXSEG with an #ifdef
TCPOPT_EOL. This broke the build on illumos, which has TCPOPT_*
but not TCPOLEN_*.

Move them to their own #ifdef TCPOLEN_MAXSEG section to remedy this.

Cc: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Andreas Färber <andreas.faerber@web.de>
Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
12 years agotcg/ppc: Handle _CALL_DARWIN being undefined on Darwin
Andreas Färber [Sun, 27 May 2012 17:50:47 +0000 (19:50 +0200)]
tcg/ppc: Handle _CALL_DARWIN being undefined on Darwin

powerpc-apple-darwin9-gcc-4.2.1 (GCC) 4.2.1 (Apple Inc. build 5577)
does not define _CALL_DARWIN, leading to unexpected behavior w.r.t.
register clobbering and stack frame layout.

Since _CALL_DARWIN is a reserved identifier, define a custom
TCG_TARGET_CALL_DARWIN based on either _CALL_DARWIN or __APPLE__.

Signed-off-by: Andreas F?rber <andreas.faerber@web.de>
Signed-off-by: malc <av1474@comtv.ru>
12 years agofdc-test: introduced qtest no_media_on_start and cmos qtest for floppy
Pavel Hrdina [Thu, 24 May 2012 09:02:30 +0000 (11:02 +0200)]
fdc-test: introduced qtest no_media_on_start and cmos qtest for floppy

As default a guest has always one floppy drive so 0x10 byte in CMOS
has to have 0x40 value. Higher 4 bits means that the first floppy drive
is 1.44 Mb 3"5 drive and lower 4 bits means the second drive is not present.

After the guest starts DSKCHG bit in DIR register should be set. If there
is no media in drive, this bit should be set all the time.

Because we start the guest without media in drive, we have to swap
'eject' and 'change' in 'test_media_change'.

Signed-off-by: Pavel Hrdina <phrdina@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
12 years agofdc: fix media detection
Pavel Hrdina [Thu, 24 May 2012 09:02:29 +0000 (11:02 +0200)]
fdc: fix media detection

We have to set up 'media_changed' after guest start so floppy driver
could detect that there is no media in drive. For this purpose we call
'fdctrl_change_cb' instead of 'fd_revalidate' in 'fdctrl_connect_drives'.
'fd_revalidate' is called inside 'fdctrl_change_cb'.

We still have to set default drive geometry in 'fd_revalidate' even
if there is no media in drive. When you try to open (windows) or mount (linux)
floppy the driver tries to seek on track 1. Linux guest stuck in loop then
kernel crashes and windows guest prints error message.

Signed-off-by: Pavel Hrdina <phrdina@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
12 years agofdc: floppy drive should be visible after start without media
Pavel Hrdina [Thu, 24 May 2012 09:02:28 +0000 (11:02 +0200)]
fdc: floppy drive should be visible after start without media

If you start guest with floppy drive but without media inserted, guest
still should see floppy drive pressent.

Signed-off-by: Pavel Hrdina <phrdina@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
12 years agoqemu-iotests: mark 035 qcow2-only
Stefan Hajnoczi [Thu, 24 May 2012 15:26:51 +0000 (16:26 +0100)]
qemu-iotests: mark 035 qcow2-only

The 035 parallel aio write test relies on knowledge of qcow2 metadata
layout to stress parallel L2 table accesses.  This only works for qcow2
unless we add additional calculations for qed or other formats.

Mark this test as qcow2-only.

Note that the test is strictly speaking non-deterministic although the
output produced is reliable with qcow2.  This is because the aio_write
command returns before the aio write request has completed.  Completions
can occur at any time afterwards and cause a message to be printed.
Therefore the exact output of this test is not deterministic but we seem
to get away with it for qcow2 (maybe due to coroutine and main loop
scheduling).

Signed-off-by: Stefan Hajnoczi <stefanha@linux.vnet.ibm.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
12 years agoqcow2: Check qcow2_alloc_clusters_at() return value
Kevin Wolf [Thu, 24 May 2012 10:56:32 +0000 (12:56 +0200)]
qcow2: Check qcow2_alloc_clusters_at() return value

When using qcow2_alloc_clusters_at(), the cluster allocation code
checked the wrong variable for an error code.

Signed-off-by: Kevin Wolf <kwolf@redhat.com>
12 years agosheepdog: use heap instead of stack for BDRVSheepdogState
MORITA Kazutaka [Wed, 16 May 2012 18:15:34 +0000 (03:15 +0900)]
sheepdog: use heap instead of stack for BDRVSheepdogState

bdrv_create() is called in coroutine context now, so we cannot use
more stack than 1 MB in the function if we use ucontext coroutine.
This patch allocates BDRVSheepdogState, whose size is 4 MB, on the
heap in sd_create().

Signed-off-by: MORITA Kazutaka <morita.kazutaka@lab.ntt.co.jp>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
12 years agosheepdog: return -errno on error
MORITA Kazutaka [Wed, 16 May 2012 18:15:33 +0000 (03:15 +0900)]
sheepdog: return -errno on error

On error, BlockDriver APIs should return -errno instead of -1.

Signed-off-by: MORITA Kazutaka <morita.kazutaka@lab.ntt.co.jp>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
12 years agosheepdog: mark image as snapshot when tag is specified
MORITA Kazutaka [Wed, 16 May 2012 18:15:31 +0000 (03:15 +0900)]
sheepdog: mark image as snapshot when tag is specified

When a snapshot tag is specified in the filename, the opened image is
a snapshot.

Signed-off-by: MORITA Kazutaka <morita.kazutaka@lab.ntt.co.jp>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
12 years agoqemu-img: Explain how rebase operation can be used to perform a 'diff' operation.
Richard W.M. Jones [Mon, 21 May 2012 13:58:05 +0000 (14:58 +0100)]
qemu-img: Explain how rebase operation can be used to perform a 'diff' operation.

Signed-off-by: Richard W.M. Jones <rjones@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
12 years agoqcow2: don't leak buffer for unexpected qcow_version in header
Jim Meyering [Mon, 21 May 2012 11:06:54 +0000 (13:06 +0200)]
qcow2: don't leak buffer for unexpected qcow_version in header

Signed-off-by: Jim Meyering <meyering@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
12 years agoscsi: declare vmstate_info_scsi_requests to be static
Jim Meyering [Mon, 21 May 2012 10:03:10 +0000 (12:03 +0200)]
scsi: declare vmstate_info_scsi_requests to be static

Signed-off-by: Jim Meyering <meyering@redhat.com>
12 years agoqemu-ga: Fix missing environ declaration
Luiz Capitulino [Wed, 23 May 2012 18:48:05 +0000 (15:48 -0300)]
qemu-ga: Fix missing environ declaration

Commit 3674838cd05268954bb6473239cd7f700a79bf0f uses the environ global
variable, but is relying on environ to be declared somewhere else.

This worked for me because on F16 environ is declared in <unistd.h>, but
that doesn't happen in OpenBSD for example, causing a build failure.

This commit fixes the build error by declaring environ if it hasn't
being declared yet.

Also fixes a build warning due to a missing <sys/wait.h> include.

Signed-off-by: Luiz Capitulino <lcapitulino@redhat.com>
Signed-off-by: Michael Roth <mdroth@linux.vnet.ibm.com>
12 years agoconfigure: check if environ is declared
Luiz Capitulino [Wed, 23 May 2012 18:48:04 +0000 (15:48 -0300)]
configure: check if environ is declared

Some systems may declare environ automatically, others don't. Check for it.

Signed-off-by: Luiz Capitulino <lcapitulino@redhat.com>
Signed-off-by: Michael Roth <mdroth@linux.vnet.ibm.com>
12 years agoaudio: Always call fini on exit
Jan Kiszka [Thu, 24 May 2012 15:05:15 +0000 (12:05 -0300)]
audio: Always call fini on exit

Not only clean up enabled voices but any registered one. Backends like
pulsaudio rely on unconditional fini handler invocations.

This fixes "Memory pool destroyed but not all memory blocks freed!"
warnings on VM shutdowns when pa is used and lockups of QEMU on shutdown
as it got stuck on some pa-internal synchronization point.

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
Signed-off-by: malc <av1474@comtv.ru>
12 years agoes1370: Fix debug code
Stefan Weil [Wed, 23 May 2012 21:26:41 +0000 (23:26 +0200)]
es1370: Fix debug code

When DEBUG_ES1370 is defined, the compiler shows these warnings:

hw/es1370.c: In function ?es1370_update_voices?:
hw/es1370.c:414: warning: format ?%d? expects type ?int?, but argument 3 has type ?size_t?
hw/es1370.c: In function ?es1370_writel?:
hw/es1370.c:582: warning: format ?%d? expects type ?int?, but argument 3 has type ?long int?
hw/es1370.c:592: warning: format ?%d? expects type ?int?, but argument 3 has type ?long int?
hw/es1370.c:609: warning: format ?%d? expects type ?int?, but argument 3 has type ?long int?
hw/es1370.c: In function ?es1370_readl?:
hw/es1370.c:751: warning: suggest braces around empty body in an ?if? statement

Fix the format strings and add the missing braces.

Signed-off-by: Stefan Weil <sw@weilnetz.de>
Signed-off-by: malc <av1474@comtv.ru>
12 years agoUpdate version for 1.1.0-rc3
Anthony Liguori [Tue, 22 May 2012 14:21:01 +0000 (09:21 -0500)]
Update version for 1.1.0-rc3

Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
12 years agoxen: Fix PV-on-HVM
Anthony PERARD [Wed, 16 May 2012 17:50:10 +0000 (18:50 +0100)]
xen: Fix PV-on-HVM

In the context of PV-on-HVM under Xen, the emulated nics are supposed to be
unplug before the guest drivers are initialized, when the guest write to a
specific IO port.

Without this patch, the guest end up with two nics with the same MAC, the
emulated nic and the PV nic.

Acked-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Anthony PERARD <anthony.perard@citrix.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
12 years agoqdev: Fix memory leak
dunrong huang [Fri, 18 May 2012 11:14:13 +0000 (19:14 +0800)]
qdev: Fix memory leak

The str allocated in visit_type_str was not freed.

The visit_type_str function is an input visitor(<QMP/String/etc>-to-native)
here, it will allocate memory for caller, so the caller is responsible for
freeing the memory.

Reviewed-by: Stefan Weil <sw@weilnetz.de>
Signed-off-by: dunrong huang <riegamaths@gmail.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
12 years agovirtio: check virtio_load return code
Orit Wassermann [Wed, 16 May 2012 10:21:35 +0000 (12:21 +0200)]
virtio: check virtio_load return code

Otherwise we crash on error.

Signed-off-by: Ulrich Obergfell <uobergfe@redhat.com>
Signed-off-by: Orit Wassermann <owasserm@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
12 years agovirtio-blk: always enable VIRTIO_BLK_F_SCSI
Paolo Bonzini [Wed, 16 May 2012 10:54:06 +0000 (12:54 +0200)]
virtio-blk: always enable VIRTIO_BLK_F_SCSI

VIRTIO_BLK_F_SCSI is supposed to mean whether the host can *parse*
SCSI requests, not *execute* them.  You could run QEMU with scsi=on
and a file-backed disk, and QEMU would fail all SCSI requests even
though it advertises VIRTIO_BLK_F_SCSI.

Because we need to do this to fix a migration compatibility problem
related to how QEMU is invoked by management, we must do this
unconditionally even on older machine types.  This more or less assumes
that no one ever invoked QEMU with scsi=off.

Here is how testing goes:

- old QEMU, scsi=on -> new QEMU, scsi=on
- new QEMU, scsi=on -> old QEMU, scsi=on
- old QEMU, scsi=off -> new QEMU, scsi=on
- new QEMU, scsi=off -> old QEMU, scsi=on
        ok (new QEMU has VIRTIO_BLK_F_SCSI, adding host features is fine)

- old QEMU, scsi=off -> new QEMU, scsi=off
        ok (new QEMU has VIRTIO_BLK_F_SCSI, adding host features is fine)

- old QEMU, scsi=on -> new QEMU, scsi=off
        ok, bug fixed

- new QEMU, scsi=on -> old QEMU, scsi=off
        doesn't work (same as: old QEMU, scsi=on -> old QEMU, scsi=off)

- new QEMU, scsi=off -> old QEMU, scsi=off
        broken by the patch

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
12 years agovirtio-blk: define VirtIOBlkConf
Paolo Bonzini [Wed, 16 May 2012 10:54:05 +0000 (12:54 +0200)]
virtio-blk: define VirtIOBlkConf

We will have to add another field to the virtio-blk configuration in
the next patch.  Avoid a proliferation of arguments to virtio_blk_init.

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
12 years agovirtio-blk: blockdev_mark_auto_del is transport-independent
Paolo Bonzini [Wed, 16 May 2012 10:54:04 +0000 (12:54 +0200)]
virtio-blk: blockdev_mark_auto_del is transport-independent

Move it from virtio_blk_exit_pci to virtio_blk_exit.

This is included here because the next patch removes proxy->block.

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
12 years agovirtio-blk: report non-zero status when failing SG_IO requests
Paolo Bonzini [Wed, 16 May 2012 10:54:03 +0000 (12:54 +0200)]
virtio-blk: report non-zero status when failing SG_IO requests

Linux really looks only at scsi->errors for SG_IO requests; it does
not look at the virtio request status at all.  Because of this, when
a SG_IO request is failed early with virtio_blk_req_complete(req,
VIRTIO_BLK_S_UNSUPP), without writing hdr.status, it will look like
a success to the guest.

This is their bug, but we can make it safe for older guests now by
forcing scsi->errors to have a non-zero value whenever a request
has to be failed.

But if we fix the bug in the guest driver, we will have another problem
because QEMU returns VIRTIO_BLK_S_IOERR if the status is non-zero, and
Linux translates that to -EIO.  Rather, the guest should succeed the
request and pass the non-zero status via the userspace-provided SG_IO
structure.  So, remove the case where virtio_blk_handle_scsi can
return VIRTIO_BLK_S_IOERR.

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
12 years agouse an uint64_t for the max_sz parameter in load_image_targphys
Mark Langsdorf [Mon, 12 Mar 2012 16:33:25 +0000 (11:33 -0500)]
use an uint64_t for the max_sz parameter in load_image_targphys

Allow load_image_targphys to load files on systems with more than 2G of
emulated memory by changing the max_sz parameter from an int to an
uint64_t.

Reviewed-by: Andreas F=E4rber <afaerber@suse.de>
Acked-by: Alexander Graf <agraf@suse.de>
Signed-off-by: Mark Langsdorf <mark.langsdorf@calxeda.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
12 years agoMerge remote-tracking branch 'mdroth/qga-pull-5-15-12' into staging
Anthony Liguori [Mon, 21 May 2012 20:31:31 +0000 (15:31 -0500)]
Merge remote-tracking branch 'mdroth/qga-pull-5-15-12' into staging

* mdroth/qga-pull-5-15-12:
  qemu-ga: align versioning with QEMU_VERSION
  qemu-ga: fix segv after failure to open log file
  qemu-ga: guest-shutdown: use only async-signal-safe functions
  qemu-ga: guest-shutdown: become synchronous
  qemu-ga: guest-suspend: make the API synchronous
  qemu-ga: become_daemon(): reopen standard fds to /dev/null
  qemu-ga: make reopen_fd_to_null() public
  qemu-ga: guest-suspend-hybrid: don't emit a success response
  qemu-ga: guest-suspend-ram: don't emit a success response
  qemu-ga: guest-suspend-disk: don't emit a success response
  qemu-ga: guest-shutdown: don't emit a success response
  qemu-ga: don't warn on no command return
  qapi: add support for command options

12 years agotests: Add rtc-test (fix test regression)
Stefan Weil [Tue, 15 May 2012 16:19:45 +0000 (18:19 +0200)]
tests: Add rtc-test (fix test regression)

Commit 93e9eb6808c886f5f1c903b7ced1eed65de2ba39 added fdc-test,
but accidentally removed rtc-test because check-qtest-i386-y was
not enhanced but set twice.

This patch adds rtc-test again (and sorts both tests alphabetically).

Signed-off-by: Stefan Weil <sw@weilnetz.de>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
12 years agovirtio/vhost: Add support for KVM in-kernel MSI injection
Jan Kiszka [Thu, 17 May 2012 13:32:39 +0000 (10:32 -0300)]
virtio/vhost: Add support for KVM in-kernel MSI injection

Make use of the new vector notifier to track changes of the MSI-X
configuration of virtio PCI devices. On enabling events, we establish
the required virtual IRQ to MSI-X message route and link the signaling
eventfd file descriptor to this vIRQ line. That way, vhost-generated
interrupts can be directly delivered to an in-kernel MSI-X consumer like
the x86 APIC.

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
12 years agomsix: Add msix_nr_vectors_allocated
Jan Kiszka [Thu, 17 May 2012 13:32:38 +0000 (10:32 -0300)]
msix: Add msix_nr_vectors_allocated

Analogously to msi_nr_vectors_allocated, add a service for MSI-X. Will
be used by the virtio-pci layer.

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
12 years agokvm: Enable use of kvm_irqchip_in_kernel in hwlib code
Jan Kiszka [Thu, 17 May 2012 13:32:37 +0000 (10:32 -0300)]
kvm: Enable use of kvm_irqchip_in_kernel in hwlib code

Provide a dummy kvm_kernel_irqchip so that kvm_irqchip_in_kernel can be
used by code that is not under CONFIG_KVM protection.

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
12 years agokvm: Introduce kvm_irqchip_add/remove_irqfd
Jan Kiszka [Thu, 17 May 2012 13:32:36 +0000 (10:32 -0300)]
kvm: Introduce kvm_irqchip_add/remove_irqfd

Add services to associate an eventfd file descriptor as input with an
IRQ line as output. Such a line can be an input pin of an in-kernel
irqchip or a virtual line returned by kvm_irqchip_add_route.

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
12 years agokvm: Make kvm_irqchip_commit_routes an internal service
Jan Kiszka [Thu, 17 May 2012 13:32:35 +0000 (10:32 -0300)]
kvm: Make kvm_irqchip_commit_routes an internal service

Automatically commit route changes after kvm_add_routing_entry and
kvm_irqchip_release_virq. There is no performance relevant use case for
which collecting multiple route changes is beneficial. This makes
kvm_irqchip_commit_routes an internal service which assert()s that the
corresponding IOCTL will always succeed.

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
12 years agokvm: Publicize kvm_irqchip_release_virq
Jan Kiszka [Thu, 17 May 2012 13:32:34 +0000 (10:32 -0300)]
kvm: Publicize kvm_irqchip_release_virq

This allows to drop routes created by kvm_irqchip_add_irq/msi_route
again.

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
12 years agokvm: Introduce kvm_irqchip_add_msi_route
Jan Kiszka [Thu, 17 May 2012 13:32:33 +0000 (10:32 -0300)]
kvm: Introduce kvm_irqchip_add_msi_route

Add a service that establishes a static route from a virtual IRQ line to
an MSI message. Will be used for IRQFD and device assignment. As we will
use this service outside of CONFIG_KVM protected code, stub it properly.

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
12 years agokvm: Rename kvm_irqchip_add_route to kvm_irqchip_add_irq_route
Jan Kiszka [Thu, 17 May 2012 13:32:32 +0000 (10:32 -0300)]
kvm: Rename kvm_irqchip_add_route to kvm_irqchip_add_irq_route

We will add kvm_irqchip_add_msi_route, so let's make the difference
clearer.

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
12 years agomsix: Introduce vector notifiers
Jan Kiszka [Thu, 17 May 2012 13:32:31 +0000 (10:32 -0300)]
msix: Introduce vector notifiers

Vector notifiers shall be triggered by the MSI/MSI-X core whenever a
relevant configuration change is programmed by the guest. In case of
MSI-X, changes are reported when the effective mask (global &&
per-vector) alters its state. On unmask, the current vector
configuration is included in the event report. This allows users - e.g.
virtio-pci layer - to transfer this information to external MSI-X
routing subsystems - like vhost + KVM in-kernel irqchip.

This implementation only provides MSI-X support, but extension to MSI is
feasible and will be provided later on when adding support for KVM PCI
device assignment.

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
12 years agomsix: Invoke msix_handle_mask_update on msix_mask_all
Jan Kiszka [Thu, 17 May 2012 13:32:30 +0000 (10:32 -0300)]
msix: Invoke msix_handle_mask_update on msix_mask_all

In preparation of firing vector notifiers on mask changes, call
msix_handle_mask_update also from msix_mask_all. So far, this will have
no real effect.

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
12 years agomsix: Factor out msix_get_message
Jan Kiszka [Thu, 17 May 2012 13:32:29 +0000 (10:32 -0300)]
msix: Factor out msix_get_message

This helper will also be used by the upcoming config notifier.

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
12 years agotests: Fix linker failure for fdc-test
Stefan Weil [Tue, 15 May 2012 05:27:19 +0000 (07:27 +0200)]
tests: Fix linker failure for fdc-test

When QEMU was built with the simple trace backend, linking failed:

  LINK  tests/fdc-test
oslib-posix.o: In function `trace_qemu_memalign':
qemu/bin/debug/x86/./trace.h:31: undefined reference to `trace3'
oslib-posix.o: In function `trace_qemu_vmalloc':
qemu/bin/debug/x86/./trace.h:35: undefined reference to `trace2'
oslib-posix.o: In function `trace_qemu_vfree':
qemu/bin/debug/x86/./trace.h:39: undefined reference to `trace1'
collect2: error: ld returned 1 exit status
make: *** [tests/fdc-test] Fehler 1

Signed-off-by: Stefan Weil <sw@weilnetz.de>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Blue Swirl <blauwirbel@gmail.com>
12 years agomips: Fix BC1ANY[24]F instructions
Richard Sandiford [Sat, 26 Nov 2011 03:37:07 +0000 (03:37 +0000)]
mips: Fix BC1ANY[24]F instructions

There's some dodgy application of De Morgan's law in the emulation
of the MIPS BC1ANY[24]F instructions: they end up branching only
if all CCs are false, rather than if one CC is.

Tested on mips64-linux-gnu, where it fixes the GCC MIPS3D tests.

Signed-off-by: Richard Sandiford <rdsandiford@googlemail.com>
Reviewed-by: Richard Henderson <rth@twiddle.net>
Signed-off-by: Blue Swirl <blauwirbel@gmail.com>
12 years agolinux-user: Fix stale tbs after mmap
Alexander Graf [Thu, 10 May 2012 22:40:10 +0000 (22:40 +0000)]
linux-user: Fix stale tbs after mmap

If we execute linux-user code that does the following:

  * A = mmap()
  * execute code in A
  * munmap(A)
  * B = mmap(), but mmap returns the same address as A
  * execute code in B

we end up executing a stale cached tb that contains translated code
from A, while we want new code from B.

This patch adds a TB flush for mmap'ed regions, before we return them,
avoiding the whole issue. It also adds a flush for munmap, so that we
don't execute stale TBs instead of getting a segfault.

Reported-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Acked-by: Riku Voipio <riku.voipio@linaro.org>
Signed-off-by: Blue Swirl <blauwirbel@gmail.com>
12 years agovirtio-pci: add missing 'static'
Blue Swirl [Sat, 12 May 2012 19:45:22 +0000 (19:45 +0000)]
virtio-pci: add missing 'static'

There are no outside references to virtio_portio.
Add missing 'static' specifier.

Reviewed-by: Stefan Weil <sw@weilnetz.de>
Signed-off-by: Blue Swirl <blauwirbel@gmail.com>
12 years agosparc64: fix initrd loading
Blue Swirl [Sat, 12 May 2012 17:20:52 +0000 (17:20 +0000)]
sparc64: fix initrd loading

Initrd load address is too low, it conflicts with kernel load
address:
rom: requested regions overlap (rom phdr #0: /tmp/vmlinux-debian-6.0.4-sparc64. free=0x0000000000742519, addr=0x0000000000400000)
rom loading failed

Fix by making the initrd address variable, load initrd after kernel
image. Use 64 bit variables instead of longs or 32 bit types.

Tested-by: Artyom Tarasenko <atar4qemu@gmail.com>
Signed-off-by: Blue Swirl <blauwirbel@gmail.com>
12 years agoaudio: split IN_T into two separate constants
Roger Pau Monne [Fri, 18 May 2012 11:08:14 +0000 (12:08 +0100)]
audio: split IN_T into two separate constants

Split IN_T into BSIZE and ITYPE, to avoid expansion if the OS has
defined macros for the intX_t and uintX_t types. The IN_T constant is
then defined in mixeng_template.h so it can be used by the
functions/macros on this header file.

This change has been tested successfully under Debian Linux and NetBSD
6.0BETA.

Cc: Vassili Karpov (malc) <av1474@comtv.ru>
Signed-off-by: Roger Pau Monne <roger.pau@citrix.com>
Signed-off-by: malc <av1474@comtv.ru>
12 years agotarget-microblaze: impelemented swapx instructions
Peter A. G. Crosthwaite [Thu, 17 May 2012 05:37:49 +0000 (15:37 +1000)]
target-microblaze: impelemented swapx instructions

Implemented the swapb and swaph byte/halfword reversal instructions added
to microblaze v8.30

Signed-off-by: Peter A. G. Crosthwaite <peter.crosthwaite@petalogix.com>
Signed-off-by: Edgar E. Iglesias <edgar.iglesias@gmail.com>
12 years agokvm: update vmxcap for EPT A/D, INVPCID, RDRAND, VMFUNC
Avi Kivity [Wed, 16 May 2012 11:31:37 +0000 (14:31 +0300)]
kvm: update vmxcap for EPT A/D, INVPCID, RDRAND, VMFUNC

Signed-off-by: Avi Kivity <avi@redhat.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
12 years agoCall xc_domain_shutdown with the reboot flag when the guest requests a reboot.
John V. Baboval [Thu, 17 May 2012 10:33:09 +0000 (10:33 +0000)]
Call xc_domain_shutdown with the reboot flag when the guest requests a reboot.

Signed-off-by: John V. Baboval <john.baboval@virtualcomputer.com>
Signed-off-by: Tom Goetz <tom.goetz@virtualcomputer.com>
Signed-off-by: Anthony PERARD <anthony.perard@citrix.com>
Signed-off-by: Stefano Stabellini <stefano.stabellini@eu.citrix.com>
12 years agoxen: Fix PV-on-HVM
Anthony PERARD [Thu, 17 May 2012 10:31:20 +0000 (10:31 +0000)]
xen: Fix PV-on-HVM

In the context of PV-on-HVM under Xen, the emulated nics are supposed to be
unplug before the guest drivers are initialized, when the guest write to a
specific IO port.

Without this patch, the guest end up with two nics with the same MAC, the
emulated nic and the PV nic.

Signed-off-by: Anthony PERARD <anthony.perard@citrix.com>
Signed-off-by: Stefano Stabellini <stefano.stabellini@eu.citrix.com>
Acked-by: Paolo Bonzini <pbonzini@redhat.com>
12 years agoxen_disk: properly update stats in ioreq_release()
Jan Beulich [Mon, 14 May 2012 16:46:33 +0000 (16:46 +0000)]
xen_disk: properly update stats in ioreq_release()

While for the "normal" case (called from blk_send_response_all())
decrementing requests_finished is correct, doing so in the parse error
case is wrong; requests_inflight needs to be decremented instead.

Signed-off-by: Jan Beulich <jbeulich@suse.com>
Signed-off-by: Stefano Stabellini <stefano.stabellini@eu.citrix.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
12 years agoxen_disk: use bdrv_aio_flush instead of bdrv_flush
Stefano Stabellini [Thu, 26 Apr 2012 16:35:53 +0000 (16:35 +0000)]
xen_disk: use bdrv_aio_flush instead of bdrv_flush

Use bdrv_aio_flush instead of bdrv_flush.

Make sure to call bdrv_aio_writev/readv after the presync bdrv_aio_flush is fully
completed and make sure to call the postsync bdrv_aio_flush after
bdrv_aio_writev/readv is fully completed.

Signed-off-by: Stefano Stabellini <stefano.stabellini@eu.citrix.com>
12 years agoxen_disk: remove syncwrite option
Stefano Stabellini [Fri, 13 Apr 2012 16:44:54 +0000 (16:44 +0000)]
xen_disk: remove syncwrite option

This patch removes a dead option.

The same can be achieved removing BDRV_O_NOCACHE and BDRV_O_CACHE_WB
from the flags passed to bdrv_open.

Signed-off-by: Stefano Stabellini <stefano.stabellini@eu.citrix.com>