The virtio-balloon device's verification of the address given to it by the
guest has a number of faults:
* The addresses here are guest physical addresses, which should be
'hwaddr' rather than 'ram_addr_t' (the distinction is admittedly
pretty subtle and confusing)
* We don't check for section.mr being NULL, which is the main way that
memory_region_find() reports basic failures. We really need to check
that before looking at any other section fields, because
memory_region_find() doesn't initialize them on the failure path
* We're passing a length of '1' to memory_region_find(), but really the
guest is requesting that we put the entire page into the balloon,
so it makes more sense to call it with BALLOON_PAGE_SIZE
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: David Hildenbrand <david@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Message-Id: <20190214043916.22128-3-david@gibson.dropbear.id.au>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
When the balloon is inflated, we discard memory place in it using madvise()
with MADV_DONTNEED. And when we deflate it we use MADV_WILLNEED, which
sounds like it makes sense but is actually unnecessary.
The misleadingly named MADV_DONTNEED just discards the memory in question,
it doesn't set any persistent state on it in-kernel; all that's necessary
to bring the memory back is to touch it. MADV_WILLNEED in contrast
specifically says that the memory will be used soon and faults it in.
This patch simplify's the balloon operation by dropping the madvise()
on deflate. This might have an impact on performance - it will move a
delay at deflate time until that memory is actually touched, which
might be more latency sensitive. However:
* Memory that's being given back to the guest by deflating the
balloon *might* be used soon, but it equally could just sit around
in the guest's pools until needed (or even be faulted out again if
the host is under memory pressure).
* Usually, the timescale over which you'll be adjusting the balloon
is long enough that a few extra faults after deflation aren't
going to make a difference.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: David Hildenbrand <david@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Message-Id: <20190214043916.22128-2-david@gibson.dropbear.id.au>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
When we are with intel-iommu device and with IR on, KVM will register
an IEC notifier to detect interrupt updates from the guest and we'll
kick off kvm_update_msi_routes_all() when it happens to make sure
kernel IRQ cache is matching the latest.
Though, kvm_update_msi_routes_all() is buggy in that it ignored the
mask bit of either MSI/MSIX messages and it tries to translate the
message even if the corresponding message was already masked by the
guest driver (hence the MSI/MSIX message will be invalid).
Without this patch, we can receive an error message when we reboot a
guest with both an assigned vfio-pci device and intel-iommu enabled:
qemu-system-x86_64: vtd_interrupt_remap_msi: MSI address low 32 bit invalid: 0x0
The error does not affect functionality of the guest since when we
failed to translate we'll just silently continue (which makes sense
since crashing the VM for this seems even worse), but still it's
better to fix it up.
Signed-off-by: Peter Xu <peterx@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Message-Id: <20190116030815.27273-5-peterx@redhat.com>
[PMD: this patch was first (incorrectly) introduced as a56de056c9]
Signed-off-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Message-Id: <20190212140621.17009-4-philmd@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Reviewed-by: Peter Xu <peterx@redhat.com>
[get|set]_addr are two counterpart to access PCDIMMDevice.addr.
Since we have already set up a property PC_DIMM_ADDR_PROP for this
field and use this mechanism in set_addr, it would be more proper to use
the same mechanism in get_addr.
This patch uses object_property_get_uint() to replace the direct memory
access to make [get|set]_addr with the same mechanism.
Signed-off-by: Wei Yang <richardw.yang@linux.intel.com>
Message-Id: <20190211064629.20186-1-richardw.yang@linux.intel.com>
Reviewed-by: Igor Mammedov <imammedo@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Introduce the following build scripts under "tests/uefi-test-tools":
* "build.sh" builds a single module (a UEFI application) from
UefiTestToolsPkg, for a single QEMU emulation target.
"build.sh" relies on cross-compilers when the emulation target and the
build host architecture don't match. The cross-compiler prefix is
computed according to a fixed, Linux-specific pattern. No attempt is
made to copy or reimplement the GNU Make magic from "qemu/roms/Makefile"
for cross-compiler prefix determination. The reason is that the build
host OSes that are officially supported by edk2, and those that are
supported by QEMU, intersect only in Linux. (Note that the UNIXGCC
toolchain is being removed from edk2,
<https://bugzilla.tianocore.org/show_bug.cgi?id=1377>.)
* "Makefile" currently builds the "UefiTestToolsPkg/BiosTablesTest"
application, for arm, aarch64, i386, and x86_64, with the help of
"build.sh".
"Makefile" turns each resultant UEFI executable into a UEFI-bootable,
qcow2-compressed ISO image. The ISO images are output as
"tests/data/uefi-boot-images/bios-tables-test.<TARGET>.iso.qcow2".
Each ISO image should be passed to QEMU as follows:
-drive id=boot-cd,if=none,readonly,format=qcow2,file=$ISO \
-device virtio-scsi-pci,id=scsi0 \
-device scsi-cd,drive=boot-cd,bus=scsi0.0,bootindex=0 \
"Makefile" assumes that "mkdosfs", "mtools", and "genisoimage" are
present.
Cc: "Michael S. Tsirkin" <mst@redhat.com>
Cc: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Cc: Gerd Hoffmann <kraxel@redhat.com>
Cc: Igor Mammedov <imammedo@redhat.com>
Cc: Philippe Mathieu-Daudé <philmd@redhat.com>
Cc: Shannon Zhao <shannon.zhaosl@gmail.com>
Signed-off-by: Laszlo Ersek <lersek@redhat.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Tested-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Message-Id: <20190204160325.4914-5-lersek@redhat.com>
Reviewed-by: Igor Mammedov <imammedo@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
The "bios-tables-test" program in QEMU's test suite locates the RSD PTR
ACPI table in guest RAM, and (chasing pointers to other ACPI tables)
performs various sanity checks on the QEMU-generated and
firmware-installed tables.
Currently this set of test cases doesn't work with UEFI guests. The ACPI
spec defines distinct methods for OSPM to locate the RSD PTR on
traditional BIOS vs. UEFI platforms, and the UEFI method is more difficult
to implement from the hypervisor side with just raw guest memory access.
Add a UEFI application (to be booted in the UEFI guest) that populates a
small, MB-aligned structure in guest RAM. The structure begins with a
signature GUID. The hypervisor should loop over all MB-aligned pages in
guest RAM until one matches the signature GUID at offset 0, at which point
the hypervisor can fetch the RSDP address field(s) from the structure.
QEMU's test logic currently spins on a pre-determined guest address, until
that address assumes a magic value. The method described in this patch is
conceptually the same ("busy loop until match is found"), except there is
no hard-coded address. This plays a lot more nicely with UEFI guest
firmware (we'll be able to use the normal page allocation UEFI service).
Given the size of EFI_GUID (16 bytes -- 128 bits), mismatches should be
astronomically unlikely. In addition, given the typical guest RAM size for
such tests (128 MB), there are 128 locations to check in one iteration of
the "outer" loop, which shouldn't introduce an intolerable delay after the
guest stores the RSDP address(es), and then the GUID.
The GUID that the hypervisor should search for is
AB87A6B1-2034-BDA0-71BD-375007757785
Expressed as a byte array:
{
0xb1, 0xa6, 0x87, 0xab,
0x34, 0x20,
0xa0, 0xbd,
0x71, 0xbd, 0x37, 0x50, 0x07, 0x75, 0x77, 0x85
}
Note that in the patch, we define "gBiosTablesTestGuid" with all bits
inverted. This is a simple method to prevent the UEFI binary, which
incorporates "gBiosTablesTestGuid", from matching the actual GUID in guest
RAM.
The UEFI application is written against the edk2 framework, which was
introduced earlier as a git submodule. The next patch will provide build
scripts for maintainers.
The source code follows the edk2 coding style, and is licensed under the
2-clause BSDL (in case someone would like to include UefiTestToolsPkg
content in a different edk2 platform).
The "UefiTestToolsPkg.dsc" platform description file resolves the used
edk2 library classes to instances (= library implementations) such that
the UEFI binaries inherit no platform dependencies. They are expected to
run on any system that conforms to the UEFI-2.3.1 spec (which was released
in 2012). The arch-specific build options are carried over from edk2's
ArmVirtPkg and OvmfPkg platforms.
Cc: "Michael S. Tsirkin" <mst@redhat.com>
Cc: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Cc: Gerd Hoffmann <kraxel@redhat.com>
Cc: Igor Mammedov <imammedo@redhat.com>
Cc: Philippe Mathieu-Daudé <philmd@redhat.com>
Cc: Shannon Zhao <shannon.zhaosl@gmail.com>
Signed-off-by: Laszlo Ersek <lersek@redhat.com>
Message-Id: <20190204160325.4914-4-lersek@redhat.com>
Reviewed-by: Igor Mammedov <imammedo@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
The roms/edk2 submodule can help with three goals:
- build the OVMF and ArmVirtQemu virtual UEFI firmware platforms (to be
implemented later),
- build the EfiRom tool on the fly, which is used in roms/Makefile, for
building the "efirom" target,
- build UEFI test applications (to be run in guests), for qtest support.
Edk2 commit 85588389222a3636baf0f9ed8227f2434af4c3f9 stands for the latest
"stable tag", namely "edk2-stable201811".
The edk2 repository tracks some binary files that should not be removed by
QEMU's top-level "make clean"; exempt the full pathnames from the "find"
command.
Cc: "Michael S. Tsirkin" <mst@redhat.com>
Cc: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Cc: Gerd Hoffmann <kraxel@redhat.com>
Cc: Igor Mammedov <imammedo@redhat.com>
Cc: Philippe Mathieu-Daudé <philmd@redhat.com>
Cc: Shannon Zhao <shannon.zhaosl@gmail.com>
Signed-off-by: Laszlo Ersek <lersek@redhat.com>
Reviewed-by: Gerd Hoffmann <kraxel@redhat.com>
Message-Id: <20190204160325.4914-2-lersek@redhat.com>
Reviewed-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: Philippe Mathieu-Daudé <philmd@redhat.com>
Tested-by: Philippe Mathieu-Daudé <philmd@redhat.com>
hw/net/vhost_net.c needs functions that are declared in net/vhost-user.c: the
vhost-user code is always compiled into QEMU, only the constructor
net_init_vhost_user is unreachable. Also, net/vhost-user.c needs functions
declared in hw/virtio/vhost-stub.c even if no virtio device exists.
Break this dependency. First, add a minimal version of net/vhost-user.c,
with no functionality and no dependency on vhost code. Second, #ifdef out
the calls back to net/vhost-user.c from hw/net/vhost_net.c.
While at it, this patch fixes the CONFIG_VHOST_NET_USE*D* typo.
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Message-Id: <1543851204-41186-3-git-send-email-pbonzini@redhat.com>
Reviewed-by: Thomas Huth <thuth@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Message-Id: <1550165756-21617-3-git-send-email-pbonzini@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
There is no reason for CONFIG_VHOST_NET to be specific to a single target;
it is a host feature that can be add to all targets, as long as they support
the virtio-net device. Currently CONFIG_VHOST_NET depends on CONFIG_KVM,
but ioeventfd support is present in the core memory API and works with
other accelerators as well.
As a first step, move the vhost-net stubs to a separate file. Later, they
will become conditional on CONFIG_VIRTIO_NET, which is not available in .c
files.
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Reviewed-by: Thomas Huth <thuth@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Message-Id: <1543851204-41186-2-git-send-email-pbonzini@redhat.com>
Message-Id: <1550165756-21617-2-git-send-email-pbonzini@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
spice_server_vm_start/stop() was added to help migration state (commit
f5bb039c6d).
However, a paused VM could keep running the spice server. This will
allow a Spice client to keep sending commands to a spice chardev. This
allows to stop/cont a VM from a Spice monitor port. Character
devices (vdagent/usb/smartcard/..) should not read from Spice when the
VM is paused.
Signed-off-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Tested-by: Victor Toso <victortoso@redhat.com>
Message-id: 20190221110703.5775-6-marcandre.lureau@redhat.com
Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
Passing several -spice options to qemu command line, or calling
several time qemu_opts_set() will ignore all but the first option
list. Since the spice server is a singleton, it makes sense to merge
all the options, the last value being the one taken into account.
This changes the behaviour from, for ex:
$ qemu... -spice port=5900 -spice port=5901 -> port: 5900
to:
$ qemu... -spice port=5900 -spice port=5901 -> port: 5901
(if necessary we could instead produce an error when an option is
given twice, although this makes handling default values and such more
complicated)
Signed-off-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Tested-by: Victor Toso <victortoso@redhat.com>
Message-id: 20190221110703.5775-5-marcandre.lureau@redhat.com
Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
The Spice server doesn't like to be started or stopped twice . It
aborts with:
(process:6191): Spice-ERROR **: 19:29:35.912: red-worker.c:623:handle_dev_start: assertion `!worker->running' failed
It's easy to avoid that situation since qemu spice_display_is_running
tracks the server state.
After the commit "spice: do not stop spice if VM is paused", it will
be possible to pause and resume the VM, and this will call
qemu_spice_display_start() twice. The easiest is to add a check for
spice_display_is_running with this patch to avoid the assert.
Signed-off-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Tested-by: Victor Toso <victortoso@redhat.com>
Message-id: 20190221110703.5775-4-marcandre.lureau@redhat.com
Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
Most chardev backend handle write() as discarded data if underlying
system is disconnected. For unknown historical reasons, the Spice
backend has "reliable" write: it will wait until the client end is
reconnected to do further successful write().
To decide whether it make sense to wait until the client is
reconnected (or queue the writes), let's review Spice chardev usage
and handling of a disconnected client:
* spice vdagent
The agents reopen the virtio port on disconnect. In qemu side,
virtio_serial_close() will also discard pending data.
* usb redirection
A disconnect creates a device disconnection.
* smartcard emulation
Data is discarded in passthru_apdu_from_guest().
(Spice doesn't explicitly open the smartcard char device until
upcoming 0.14.2, commit 69a5cfc74131ec0459f2eb5a231139f5a69a8037)
* spice webdavd
The daemon will restart the service, and reopen the virtio port.
* spice ports (serial console, qemu monitor..)
Depends on the associated device or usage.
- serial, may be throttled or discarded on write, depending on
device
- QMP/HMP monitor have some CLOSED event handling, but want to
flush the write, which will finish when a new client connects.
On disconnect/reconnect, the client starts with fresh sessions. If it
is a seamless migration, the client disconnects after the source
migrated. The handling of source disconnect in qemu is thus irrelevant
for the Spice session migration.
For all these use cases, it is better to discard writes when the
client is disconnected, and require the vm-side device/agent to behave
correctly on CHR_EVENT_CLOSED, to stop reading and writing from
the spice chardev.
Signed-off-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Tested-by: Victor Toso <victortoso@redhat.com>
Message-id: 20190221110703.5775-3-marcandre.lureau@redhat.com
Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
The license information in this file is very messy. A short note at
the beginning says GPL first, but the long boilerplate code then
talks about "GNU Lesser General Public License version 2.0". First,
there is no such version of the "GNU Lesser GPL", it only started with
version 2.1. In version 2.0, it was still called "GNU Library GPL"
instead. Second, you can easily get the license of this file wrong
if you only quickly glance at the long boilerplate code.
Anyway, looking at the text of the LGPL (see COPYING.LIB in the top
directory), the license clearly states in section "3." that one should
rather replace the license information with the GPL information in
such a case of a mixture instead. Thus let's clean up the confusing
statements and use the proper GPL text only.
Signed-off-by: Thomas Huth <thuth@redhat.com>
Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>
Message-id: 1550731902-28842-1-git-send-email-thuth@redhat.com
[ kraxel: s/v2/v2+/ as requested by Daniel ]
Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
Calls the new SPICE QXL interface function spice_qxl_set_device_info to
set the hardware address of the graphics device represented by the QXL
interface (e.g. a PCI path) and the device display IDs (the IDs of the
device's monitors that belong to this QXL interface).
Also stops using the deprecated spice_qxl_set_max_monitors, the new
interface function replaces it.
Signed-off-by: Lukáš Hrázký <lhrazky@redhat.com>
Message-Id: <20190215150919.8263-1-lhrazky@redhat.com>
Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
In musb_packet(), the call to usb_find_device() can return NULL
if it doesn't find a device matching 'addr' so explicitly check
the return value before passing it to usb_ep_get(). This then
allows the subsequent calculation of 'id' to be streamlined.
Signed-off-by: Liam Merwick <liam.merwick@oracle.com>
Message-id: 1549460216-25808-8-git-send-email-liam.merwick@oracle.com
Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
Most callers of xhci_port_update() and xhci_wakeup() pass in a pointer
to an array entry and can never be NULL but add two defensive asserts
to protect against future changes (e.g. adding a new port speed, etc.)
adding a path through xhci_lookup_port() that could result in the
return of a NULL XHCIPort.
Signed-off-by: Liam Merwick <liam.merwick@oracle.com>
Message-id: 1549460216-25808-3-git-send-email-liam.merwick@oracle.com
Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
When bitmaps are persistent, they may incur a disk read or write when bitmaps
are added or removed. For configurations like virtio-dataplane, failing to
acquire this lock will abort QEMU when disk IO occurs.
We used to acquire aio_context as part of the bitmap lookup, so re-introduce
the lock for just the cases that have an IO penalty. Commit 2119882c removed
these locks, and I failed to notice this when we committed fd5ae4cc, so this
has been broken since persistent bitmaps were introduced.
Fixes: https://bugzilla.redhat.com/show_bug.cgi?id=1672010
Reported-By: Aihua Liang <aliang@redhat.com>
Signed-off-by: John Snow <jsnow@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Message-id: 20190218233154.19303-1-jsnow@redhat.com
Signed-off-by: John Snow <jsnow@redhat.com>
Since qemu currently doesn't flush persistent bitmaps to disk until
shutdown (which might be MUCH later), it's useful if 'query-block'
at least shows WHICH bitmaps will (eventually) make it to persistent
storage. Update affected iotests.
Signed-off-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Reviewed-by: John Snow <jsnow@redhat.com>
Message-id: 20190204210512.27458-1-eblake@redhat.com
Signed-off-by: John Snow <jsnow@redhat.com>
FLIX adds branch and loop instruction variants with 15- and 18-bit wide
target offset. Implement them as additional names for the ordinary
branch/loop opcodes.
Signed-off-by: Max Filippov <jcmvbkbc@gmail.com>
There are opcodes that differ only in encoding or possible range of
immediate arguments. Allow multiple names for single opcode translation
table entry to reduce code duplication in that case.
Signed-off-by: Max Filippov <jcmvbkbc@gmail.com>
David Gibson