Today virtio-blk dataplane uses a 1:1 device-per-thread model. Now that
IOThreads have been introduced we can generalize this to N:M devices per
threads.
This patch drops thread code from dataplane in favor of running inside
an IOThread AioContext.
As a bonus we solve the case where a guest keeps submitting I/O requests
while dataplane is trying to stop. Previously the dataplane thread
would continue to process requests until the request gave it a break.
Now we can shut down in bounded time thanks to
aio_context_acquire/release.
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Add a "iothread" qdev property type so devices can be hooked up to an
IOThread from the comand-line:
qemu -object iothread,id=iothread0 \
-device some-device,x-iothread=iothread0
Note that Paolo Bonzini <pbonzini@redhat.com> has suggested using QOM
links instead. This way the relationship between the objects is
reflected in QOM. There are currently shortcomings of
object_property_add_link() which prevent this use case. I will attempt
to fix them and move to QOM links in a separate series.
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
get_pointer()'s print() callback might return a heap allocated
string, to avoid adding dedicated get_pointer_foo for this case
convert current print() callbacks to return temporary heap
allocated string and make get_pointer() free it.
Reviewed-by: Andreas Färber <afaerber@suse.de>
Signed-off-by: Igor Mammedov <imammedo@redhat.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
This is a stand-in for Michael Roth's QContext. I expect this to be
replaced once QContext is completed.
The IOThread object is an AioContext event loop thread. This patch adds
the concept of multiple event loop threads, allowing users to define
them.
When SMP guests run on SMP hosts it makes sense to instantiate multiple
IOThreads. This spreads event loop processing across multiple cores.
Note that additional patches are required to actually bind a device to
an IOThread.
[Andreas Färber <afaerber@suse.de> pointed out that the embedded parent
object instance should be called "parent_obj" and have a newline
afterwards. This patch has been changed to reflect this.
-- Stefan]
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
It can be useful to run an AioContext from a thread which normally does
not "own" the AioContext. For example, request draining can be
implemented by acquiring the AioContext and looping aio_poll() until all
requests have been completed.
The following pattern should work:
/* Event loop thread */
while (running) {
aio_context_acquire(ctx);
aio_poll(ctx, true);
aio_context_release(ctx);
}
/* Another thread */
aio_context_acquire(ctx);
bdrv_read(bs, 0x1000, buf, 1);
aio_context_release(ctx);
This patch implements aio_context_acquire() and aio_context_release().
Note that existing aio_poll() callers do not need to worry about
acquiring and releasing - it is only needed when multiple threads will
call aio_poll() on the same AioContext.
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
QemuMutex does not guarantee fairness and cannot be acquired
recursively:
Fairness means each locker gets a turn and the scheduler cannot cause
starvation.
Recursive locking is useful for composition, it allows a sequence of
locking operations to be invoked atomically by acquiring the lock around
them.
This patch adds RFifoLock, a recursive lock that guarantees FIFO order.
Its first user is added in the next patch.
RFifoLock has one additional feature: it can be initialized with an
optional contention callback. The callback is invoked whenever a thread
must wait for the lock. For example, it can be used to poke the current
owner so that they release the lock soon.
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
It is often useful to find an object's child property name. Also use
this new function to simplify the implementation of
object_get_canonical_path().
Reviewed-by: Andreas Färber <afaerber@suse.de>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
This patch keep the recursive way of doing things but simplify it by giving
two responsabilities to all block filters implementors.
They will need to do two things:
-Set the is_filter field of their block driver to true.
-Implement the bdrv_recurse_is_first_non_filter method of their block driver like
it is done on the Quorum block driver. (block/quorum.c)
[Paolo Bonzini <pbonzini@redhat.com> pointed out that this patch changes
the semantics of blkverify, which now recurses down both bs->file and
s->test_file.
-- Stefan]
Reported-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Benoit Canet <benoit@irqsave.net>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Extend test file 060 by a test case for corruption occuring concurrently
to a COW request. QEMU should not crash but rather return an appropriate
error message.
Signed-off-by: Max Reitz <mreitz@redhat.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Currently, bdrv_debug_resume() requires every bs->drv in the BDS stack
to be NULL until a bs->drv with an implementation of bdrv_debug_resume()
is found. For a normal function, this would be fine, but this is a
function for debugging purposes and should therefore allow intermediate
BDS not to have a driver (i.e., be "ejected"). Otherwise, it is hard to
debug such situations.
Signed-off-by: Max Reitz <mreitz@redhat.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Before dereferencing bs->drv for a call to its member bdrv_co_readv(),
copy_sectors() should check whether that pointer is indeed valid, since
it may have been set to NULL by e.g. a concurrent write triggering the
corruption prevention mechanism.
Signed-off-by: Max Reitz <mreitz@redhat.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
After migration has completed, we call bdrv_invalidate_cache() so that
drivers which cache some data drop their stale copy of the data and
reread it from the image file to get a new version of data that the
source modified while the migration was running.
Reloading metadata from the image file is useless, though, if the size
of the image file stays stale (this is a value that is cached for all
image formats in block.c). Reads from (meta)data after the old EOF
return only zeroes, causing image corruption.
We need to update bs->total_sectors in all layers that could potentially
have changed their size (i.e. backing files are not a concern - if they
are changed, we're in bigger trouble)
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
When reading the refcount table entry in get_refcount(), only bits which
are actually significant for the refcount block offset should be taken
into account.
Signed-off-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Laszlo Ersek <lersek@redhat.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Older versions of gcc (eg 4.6) can't handle varargs functions declared
inline for anything other than completely trivial uses, and complain:
tests/qom-test.c: In function 'qmp': tests/libqtest.h:359:60: sorry,
unimplemented: function 'qmp' can never be inlined because it uses
variable argument lists
Avoid this problem by putting the functions into libqtest.c instead
of using inline definitions in libqtest.h.
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Andreas Färber <afaerber@suse.de>
This converts the old-style SysBusDevice::init() callback to a new-style
DeviceClass::realize() callback.
As a part of conversion, this replaces fprintf(stderr) with error_setg()
as realize() does not "return" any value, instead it puts the extended
error into **errp.
Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Reviewed-by: Mike Day <ncmike@ncultra.org>
Signed-off-by: Andreas Färber <afaerber@suse.de>
Currently interrupt priorities are set to 0 (highest) at the very
beginning of the guest execution which is not correct and makes the guest
produce random interrupt error messages such as:
"Interrupt 0x1001 (real) is invalid, disabling it".
This also prevents interrupt states from correct migration.
This initializes priority to 0xFF as the emulated XICS does.
Suggested-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Signed-off-by: Andreas Färber <afaerber@suse.de>
The changelog is:
> version: update to 20140304
> Introduce dummy console device
> vio-vscsi: Fix CRQ allocation alignment
> version: update to 20140204
> virtio-9p: disable unused structure
> Make "boot net:dhcp" boot from IPv4 only
> Fix virtio device shutdown
> Change shutdown method name for virtio-scsi
> Add support for 64bit LE ABI v1 and v2 support
> Change representation of string environment variable
> cas: return error when unknown node found
> version: update
> Reset obp-tftp arguments before parsing
> Enable seamless netboot on IPv6 network
> Fix shutdown for virtio devices
> Fix zero checksum in UDP header
> Handle router advertisement message properly
> [oex]hci_exit: Check before freeing/unmapping memory
> Work around missing sc 1 traps on pHyp
> fix print_version() to return where it came from
> usb-xhci: memory freeing and using returns as bool uniformly
> Output banner and initial display output in VNC window
> use VERSION file to generate FW version
> cas: remove warning
> Add support for loading little endian ELF binaries.
> Add bswap_{16,32,64}p
> dhcpv6 and other minor net-snk fixes
> Fix missing drop in virtio-fs setup-alias
> Find next available alias name
> SLOF does not exit if given 1KB disk
> boot: enable support for bootindex
> pci-properties: add properties to enable hotplug for spapr
> e1000: remember node handle
> Increase quiesce tokens array size
> virtio: timeout after 5sec
> Enable IPv6 support in dns
> usb-ohci: fix warnings
> Add ipv6 support in net-snk
> ipv4: fix frame overwriting following arp_send_request
> e1000: fix SLOF_dma_map_out arguments
> Maintain single global packet buffer for tftp
> Increase virtio-net receive queue size
> Increase veth receive queue size
> Fix dprintf macros at various points
> usb-ohci: rewrite done_head processing code
> boot: add net in default boot order
> block 0 address in the allocator
> scsi: make-media-alias fix
> usb-xhci: add xhci host controller support
> usb-xhci: add xhci support
> Avoid veth read/write calls with zero length buffer
> boot: include other aliases
> usb-core: disable xhci
Cc: Andreas Färber <afaerber@suse.de>
Cc: Nikunj A Dadhania <nikunj@linux.vnet.ibm.com>
Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
As reported in commit 9c06a1f79f, xen.h is
not self-contained with regards to its use of QEMUMachine. Fix this.
Reported-by: Alexander Graf <agraf@suse.de>
Suggested-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Andreas Färber <afaerber@suse.de>
This reverts commit 9c06a1f79f.
The new header sysemu/qemumachine.h is undesired.
Suggested-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Andreas Färber <afaerber@suse.de>
'socket_accept' waits for QEMU to init its unix socket.
If QEMU encounters an error during command line parsing,
it can exit before initializing the communication channel.
Using a timeout for sockets fixes the issue.
Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Marcel Apfelbaum <marcel.a@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Andreas Färber <afaerber@suse.de>
Use the realize and unrealize hooks to register and unregister
vmstate_pcibus respectively.
Relocate some stuff to avoid forward declarations.
Signed-off-by: Bandan Das <bsd@redhat.com>
Acked-by: Michael S. Tsirkin <mst@redhat.com>
[AF: Keep using PCI_BUS() cast macro]
Signed-off-by: Andreas Färber <afaerber@suse.de>
Integrate (un)realization of child buses with realization/unrealization
of the device hosting them. Code in device_unparent() is reordered for
unrealization of buses to work as part of device unrealization.
That way no changes need to be made to bus instantiation.
Signed-off-by: Bandan Das <bsd@redhat.com>
Signed-off-by: Andreas Färber <afaerber@suse.de>
Add a "realized" property calling realize/unrealize hooks as for devices.
Signed-off-by: Bandan Das <bsd@redhat.com>
Signed-off-by: Andreas Färber <afaerber@suse.de>
This adds a test whether sPAPR PHB can be added via the command line.
Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Signed-off-by: Andreas Färber <afaerber@suse.de>
Recursively walk all properties under /machine and try to retrieve their
value. This is a regression test for link<> properties and the
DeviceState::hotpluggable property.
Cf. be2f78b6b0 and
1a37eca107
Signed-off-by: Andreas Färber <afaerber@suse.de>
In order to allow attaching machine options to a machine instance,
current_machine is converted into MachineState.
As a first step of deprecating QEMUMachine, some of the functions
were modified to return MachineClass.
Signed-off-by: Marcel Apfelbaum <marcel.a@redhat.com>
Signed-off-by: Andreas Färber <afaerber@suse.de>
The machine registration flow is refactored to use the QOM functionality.
Instead of linking the machines into a list, each machine has a type
and the types can be traversed in the QOM way.
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Marcel Apfelbaum <marcel.a@redhat.com>
Signed-off-by: Andreas Färber <afaerber@suse.de>
The main functional change is to convert QEMUMachine into MachineClass
and QEMUMachineInitArgs into MachineState, instance of MachineClass.
As a first step, in order to make possible an incremental development,
both QEMUMachine and QEMUMachineInitArgs are being embedded into the
new types.
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Marcel Apfelbaum <marcel.a@redhat.com>
Signed-off-by: Andreas Färber <afaerber@suse.de>
Test the error class instead. Expecting a specific message is
fragile. In fact, it broke once already, in commit 75884af. Restore
the test of error member "class" dropped there, and drop the test of
error member "desc".
There are no other tests of "desc" as far as I can tell.
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Andreas Färber <afaerber@suse.de>
Use g_assert_cmpstr() instead of combining g_assert() and strcmp(3).
This simplifies the code since we no longer have to play games to
distinguish NULL from "" using "(null)".
gcc extension haters will also be happy that ?: was dropped.
Suggested-by: Markus Armbruster <armbru@redhat.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Andreas Färber <afaerber@suse.de>
Previously libvirt required the first/default PCI bus to have name "pci".
Since QEMU can support multiple buses now, libvirt wants "pci.0" now.
This removes custom bus name and lets QEMU make up default names.
Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Signed-off-by: Andreas Färber <afaerber@suse.de>
There are no usages left of this legacy cast. Delete.
Signed-off-by: Peter Crosthwaite <peter.crosthwaite@xilinx.com>
[AF: Rename SSISlave parent field]
Signed-off-by: Andreas Färber <afaerber@suse.de>
ssi: Rename parent field
Define and use QOM cast macro. Removes some usages of legacy casting
systems.
Signed-off-by: Peter Crosthwaite <peter.crosthwaite@xilinx.com>
[AF: Rename parent field]
Signed-off-by: Andreas Färber <afaerber@suse.de>
Create an abstract class that encompasses both max111x variants. This is
needed for QOM cast macro creation (and is the right thing to do
anyway). Macroify type-names in the process.
Signed-off-by: Peter Crosthwaite <peter.crosthwaite@xilinx.com>
Signed-off-by: Andreas Färber <afaerber@suse.de>
Remove two legacy ->qbus style casts from TYPE_SSI_BUS to TYPE_BUS in
ssi.c.
Signed-off-by: Peter Crosthwaite <peter.crosthwaite@xilinx.com>
[AF: Convert one missing ->qbus and rename parent field]
Signed-off-by: Andreas Färber <afaerber@suse.de>
Convert legacy ->qdev style casts from TYPE_SSI_SLAVE to TYPE_DEVICE.
Signed-off-by: Peter Crosthwaite <peter.crosthwaite@xilinx.com>
[AF: Introduce local DeviceState variable for transition to QOM realize]
Signed-off-by: Andreas Färber <afaerber@suse.de>
When object_property_add_str() and object_property_add_bool() fail, they
leak their internal StringProperty and BoolProperty structs. Remember
to free the structs on error.
Luckily this is a low-impact memory leak since most QOM properties are
static qdev properties that will never take the error case.
object_property_add() only fails if the property name is already in use.
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Cc: qemu-stable@nongnu.org
Signed-off-by: Andreas Färber <afaerber@suse.de>
Stefan Hajnoczi