When instantiating XenBus itself, for each NIC which is configured with
either the model unspecified, or set to to "xen" or "xen-net-device",
create a corresponding xen-net-device for it.
Now we can revert the previous more hackish version which relied on the
platform code explicitly registering the NICs on its own XenBus, having
returned the BusState* from xen_bus_init() itself.
This also fixes the setup for Xen PV guests, which was previously broken
in various ways and never actually managed to peer with the netdev.
Signed-off-by: David Woodhouse <dwmw@amazon.co.uk>
Reviewed-by: Paul Durrant <paul@xen.org>
The default NIC creation seems a bit hackish to me. I don't understand
why each platform has to call pci_nic_init_nofail() from a point in the
code where it actually has a pointer to the PCI bus, and then we have
the special cases for things like ne2k_isa.
If qmp_device_add() can *find* the appropriate bus and instantiate
the device on it, why can't we just do that from generic code for
creating the default NICs too?
But that isn't a yak I want to shave today. Add a xenbus field to the
PCMachineState so that it can make its way from pc_basic_device_init()
to pc_nic_init() and be handled as a special case like ne2k_isa is.
Now we can launch emulated Xen guests with '-nic user'.
Signed-off-by: David Woodhouse <dwmw@amazon.co.uk>
Reviewed-by: Paul Durrant <paul@xen.org>
The primary console is special because the toolstack maps a page into
the guest for its ring, and also allocates the guest-side event channel.
The guest's grant table is even primed to export that page using a known
grant ref#. Add support for all that in emulated mode, so that we can
have a primary console.
For reasons unclear, the backends running under real Xen don't just use
a mapping of the well-known GNTTAB_RESERVED_CONSOLE grant ref (which
would also be in the ring-ref node in XenStore). Instead, the toolstack
sets the ring-ref node of the primary console to the GFN of the guest
page. The backend is expected to handle that special case and map it
with foreignmem operations instead.
We don't have an implementation of foreignmem ops for emulated Xen mode,
so just make it map GNTTAB_RESERVED_CONSOLE instead. This would probably
work for real Xen too, but we can't work out how to make real Xen create
a primary console of type "ioemu" to make QEMU drive it, so we can't
test that; might as well leave it as it is for now under Xen.
Now at last we can boot the Xen PV shim and run PV kernels in QEMU.
Signed-off-by: David Woodhouse <dwmw@amazon.co.uk>
Reviewed-by: Paul Durrant <paul@xen.org>
The primary Xen console is special. The guest's side is set up for it by
the toolstack automatically and not by the standard PV init sequence.
Accordingly, its *frontend* doesn't appear in …/device/console/0 either;
instead it appears under …/console in the guest's XenStore node.
To allow the Xen console driver to override the frontend path for the
primary console, add a method to the XenDeviceClass which can be used
instead of the standard xen_device_get_frontend_path()
Signed-off-by: David Woodhouse <dwmw@amazon.co.uk>
Reviewed-by: Paul Durrant <paul@xen.org>
The previous commit introduced redirectable gnttab operations fairly
much like-for-like, with the exception of the extra arguments to the
->open() call which were always NULL/0 anyway.
This *changes* the arguments to the ->unmap() operation to include the
original ref# that was mapped. Under real Xen it isn't necessary; all we
need to do from QEMU is munmap(), then the kernel will release the grant,
and Xen does the tracking/refcounting for the guest.
When we have emulated grant tables though, we need to do all that for
ourselves. So let's have the back ends keep track of what they mapped
and pass it in to the ->unmap() method for us.
Signed-off-by: David Woodhouse <dwmw@amazon.co.uk>
Reviewed-by: Paul Durrant <paul@xen.org>
Move the existing code using libxengnttab to xen-operations.c and allow
the operations to be redirected so that we can add emulation of grant
table mapping for backend drivers.
In emulation, mapping more than one grant ref to be virtually contiguous
would be fairly difficult. The best way to do it might be to make the
ram_block mappings actually backed by a file (shmem or a deleted file,
perhaps) so that we can have multiple *shared* mappings of it. But that
would be fairly intrusive.
Making the backend drivers cope with page *lists* instead of expecting
the mapping to be contiguous is also non-trivial, since some structures
would actually *cross* page boundaries (e.g. the 32-bit blkif responses
which are 12 bytes).
So for now, we'll support only single-page mappings in emulation. Add a
XEN_GNTTAB_OP_FEATURE_MAP_MULTIPLE flag to indicate that the native Xen
implementation *does* support multi-page maps, and a helper function to
query it.
Signed-off-by: David Woodhouse <dwmw@amazon.co.uk>
Reviewed-by: Paul Durrant <paul@xen.org>
The existing implementation calling into the real libxenevtchn moves to
a new file hw/xen/xen-operations.c, and is called via a function table
which in a subsequent commit will also be able to invoke the emulated
event channel support.
Signed-off-by: David Woodhouse <dwmw@amazon.co.uk>
Reviewed-by: Paul Durrant <paul@xen.org>
Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
Acked-by: Anthony PERARD <anthony.perard@citrix.com>
Message-Id: <20221219130205.687815-3-berrange@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
One less qemu-specific macro. It also helps to make some headers/units
only depend on glib, and thus moved in standalone projects eventually.
Signed-off-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Reviewed-by: Richard W.M. Jones <rjones@redhat.com>
Currently a single watch on /local/domain/X/backend is registered by each
QEMU process running in service domain X (where X is usually 0). The purpose
of this watch is to ensure that QEMU is notified when the Xen toolstack
creates a new device backend area.
Such a backend area is specific to a single frontend area created for a
specific guest domain and, since each QEMU process is also created to service
a specfic guest domain, it is unnecessary and inefficient to notify all QEMU
processes.
Only the QEMU process associated with the same guest domain need
receive the notification. This patch re-factors the watch registration code
such that notifications are targetted appropriately.
Reported-by: Jerome Leseinne <jerome.leseinne@gmail.com>
Signed-off-by: Paul Durrant <pdurrant@amazon.com>
Reviewed-by: Anthony PERARD <anthony.perard@citrix.com>
Message-Id: <20201001081500.1026-1-paul@xen.org>
Signed-off-by: Anthony PERARD <anthony.perard@citrix.com>
One of the goals of having less boilerplate on QOM declarations
is to avoid human error. Requiring an extra argument that is
never used is an opportunity for mistakes.
Remove the unused argument from OBJECT_DECLARE_TYPE and
OBJECT_DECLARE_SIMPLE_TYPE.
Coccinelle patch used to convert all users of the macros:
@@
declarer name OBJECT_DECLARE_TYPE;
identifier InstanceType, ClassType, lowercase, UPPERCASE;
@@
OBJECT_DECLARE_TYPE(InstanceType, ClassType,
- lowercase,
UPPERCASE);
@@
declarer name OBJECT_DECLARE_SIMPLE_TYPE;
identifier InstanceType, lowercase, UPPERCASE;
@@
OBJECT_DECLARE_SIMPLE_TYPE(InstanceType,
- lowercase,
UPPERCASE);
Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>
Reviewed-by: Cédric Le Goater <clg@kaod.org>
Acked-by: Cornelia Huck <cohuck@redhat.com>
Acked-by: Igor Mammedov <imammedo@redhat.com>
Acked-by: Paul Durrant <paul@xen.org>
Acked-by: Thomas Huth <thuth@redhat.com>
Message-Id: <20200916182519.415636-4-ehabkost@redhat.com>
Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>
Some typedefs and macros are defined after the type check macros.
This makes it difficult to automatically replace their
definitions with OBJECT_DECLARE_TYPE.
Patch generated using:
$ ./scripts/codeconverter/converter.py -i \
--pattern=QOMStructTypedefSplit $(git grep -l '' -- '*.[ch]')
which will split "typdef struct { ... } TypedefName"
declarations.
Followed by:
$ ./scripts/codeconverter/converter.py -i --pattern=MoveSymbols \
$(git grep -l '' -- '*.[ch]')
which will:
- move the typedefs and #defines above the type check macros
- add missing #include "qom/object.h" lines if necessary
Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>
Reviewed-by: Juan Quintela <quintela@redhat.com>
Message-Id: <20200831210740.126168-9-ehabkost@redhat.com>
Reviewed-by: Juan Quintela <quintela@redhat.com>
Message-Id: <20200831210740.126168-10-ehabkost@redhat.com>
Message-Id: <20200831210740.126168-11-ehabkost@redhat.com>
Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>
Devices may have component devices and buses.
Device realization may fail. Realization is recursive: a device's
realize() method realizes its components, and device_set_realized()
realizes its buses (which should in turn realize the devices on that
bus, except bus_set_realized() doesn't implement that, yet).
When realization of a component or bus fails, we need to roll back:
unrealize everything we realized so far. If any of these unrealizes
failed, the device would be left in an inconsistent state. Must not
happen.
device_set_realized() lets it happen: it ignores errors in the roll
back code starting at label child_realize_fail.
Since realization is recursive, unrealization must be recursive, too.
But how could a partly failed unrealize be rolled back? We'd have to
re-realize, which can fail. This design is fundamentally broken.
device_set_realized() does not roll back at all. Instead, it keeps
unrealizing, ignoring further errors.
It can screw up even for a device with no buses: if the lone
dc->unrealize() fails, it still unregisters vmstate, and calls
listeners' unrealize() callback.
bus_set_realized() does not roll back either. Instead, it stops
unrealizing.
Fortunately, no unrealize method can fail, as we'll see below.
To fix the design error, drop parameter @errp from all the unrealize
methods.
Any unrealize method that uses @errp now needs an update. This leads
us to unrealize() methods that can fail. Merely passing it to another
unrealize method cannot cause failure, though. Here are the ones that
do other things with @errp:
* virtio_serial_device_unrealize()
Fails when qbus_set_hotplug_handler() fails, but still does all the
other work. On failure, the device would stay realized with its
resources completely gone. Oops. Can't happen, because
qbus_set_hotplug_handler() can't actually fail here. Pass
&error_abort to qbus_set_hotplug_handler() instead.
* hw/ppc/spapr_drc.c's unrealize()
Fails when object_property_del() fails, but all the other work is
already done. On failure, the device would stay realized with its
vmstate registration gone. Oops. Can't happen, because
object_property_del() can't actually fail here. Pass &error_abort
to object_property_del() instead.
* spapr_phb_unrealize()
Fails and bails out when remove_drcs() fails, but other work is
already done. On failure, the device would stay realized with some
of its resources gone. Oops. remove_drcs() fails only when
chassis_from_bus()'s object_property_get_uint() fails, and it can't
here. Pass &error_abort to remove_drcs() instead.
Therefore, no unrealize method can fail before this patch.
device_set_realized()'s recursive unrealization via bus uses
object_property_set_bool(). Can't drop @errp there, so pass
&error_abort.
We similarly unrealize with object_property_set_bool() elsewhere,
always ignoring errors. Pass &error_abort instead.
Several unrealize methods no longer handle errors from other unrealize
methods: virtio_9p_device_unrealize(),
virtio_input_device_unrealize(), scsi_qdev_unrealize(), ...
Much of the deleted error handling looks wrong anyway.
One unrealize methods no longer ignore such errors:
usb_ehci_pci_exit().
Several realize methods no longer ignore errors when rolling back:
v9fs_device_realize_common(), pci_qdev_unrealize(),
spapr_phb_realize(), usb_qdev_realize(), vfio_ccw_realize(),
virtio_device_realize().
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Message-Id: <20200505152926.18877-17-armbru@redhat.com>
It is not safe to close an event channel from the QEMU main thread when
that channel's poller is running in IOThread context.
This patch adds a new xen_device_set_event_channel_context() function
to explicitly assign the channel AioContext, and modifies
xen_device_bind_event_channel() to initially assign the channel's poller
to the QEMU main thread context. The code in xen-block's dataplane is
then modified to assign the channel to IOThread context during
xen_block_dataplane_start() and de-assign it during in
xen_block_dataplane_stop(), such that the channel is always assigned
back to main thread context before it is closed. aio_set_fd_handler()
already deals with all the necessary synchronization when moving an fd
between AioContext-s so no extra code is needed to manage this.
Reported-by: Julien Grall <jgrall@amazon.com>
Signed-off-by: Paul Durrant <pdurrant@amazon.com>
Reviewed-by: Anthony PERARD <anthony.perard@citrix.com>
Message-Id: <20191216143451.19024-1-pdurrant@amazon.com>
Signed-off-by: Anthony PERARD <anthony.perard@citrix.com>
Cleaning up offline XenDevice objects directly in
xen_device_backend_changed() is dangerous as xen_device_unrealize() will
modify the watch list that is being walked. Even the QLIST_FOREACH_SAFE()
used in notifier_list_notify() is insufficient as *two* notifiers (for
the frontend and backend watches) are removed, thus potentially rendering
the 'next' pointer unsafe.
The solution is to use the XenBus backend_watch handler to do the clean-up
instead, as it is invoked whilst walking a separate watch list.
This patch therefore adds a new 'inactive_devices' list to XenBus, to which
offline devices are added by xen_device_backend_changed(). The XenBus
backend_watch registration is also changed to not only invoke
xen_bus_enumerate() but also a new xen_bus_cleanup() function, which will
walk 'inactive_devices' and perform the necessary actions.
For safety an extra 'online' check is also added to xen_bus_type_enumerate()
to make sure that no attempt is made to create a new XenDevice object for a
backend that is offline.
NOTE: This patch also includes some cosmetic changes:
- substitute the local variable name 'backend_state'
in xen_bus_type_enumerate() with 'state', since there
is no ambiguity with any other state in that context.
- change xen_device_state_is_active() to
xen_device_frontend_is_active() (and pass a XenDevice directly)
since the state tests contained therein only apply to a frontend.
- use 'state' rather then 'xendev->backend_state' in
xen_device_backend_changed() to shorten the code.
Signed-off-by: Paul Durrant <paul.durrant@citrix.com>
Reviewed-by: Anthony PERARD <anthony.perard@citrix.com>
Message-Id: <20190913082159.31338-4-paul.durrant@citrix.com>
Signed-off-by: Anthony PERARD <anthony.perard@citrix.com>
This patch uses the XenWatchList abstraction to add a separate watch list
for each device. This is more scalable than walking a single notifier
list for all watches and is also necessary to implement a bug-fix in a
subsequent patch.
Signed-off-by: Paul Durrant <paul.durrant@citrix.com>
Reviewed-by: Anthony Perard <anthony.perard@citrix.com>
Message-Id: <20190913082159.31338-3-paul.durrant@citrix.com>
Signed-off-by: Anthony PERARD <anthony.perard@citrix.com>
Xenstore watch call-backs are already abstracted away from XenBus using
the XenWatch data structure but the associated NotifierList manipulation
and file handle registration is still open coded in various xen_bus_...()
functions.
This patch creates a new XenWatchList data structure to allow these
interactions to be abstracted away from XenBus as well. This is in
preparation for a subsequent patch which will introduce separate watch lists
for XenBus and XenDevice objects.
NOTE: This patch also introduces a new notifier_list_empty() helper function
for the purposes of adding an assertion that a XenWatchList is not
freed whilst its associated NotifierList is still occupied.
Signed-off-by: Paul Durrant <paul.durrant@citrix.com>
Reviewed-by: Anthony Perard <anthony.perard@citrix.com>
Message-Id: <20190913082159.31338-2-paul.durrant@citrix.com>
Signed-off-by: Anthony PERARD <anthony.perard@citrix.com>
This patch introduces a poll callback for event channel fd-s and uses
this to invoke the channel callback function.
To properly support polling, it is necessary for the event channel callback
function to return a boolean saying whether it has done any useful work or
not. Thus xen_block_dataplane_event() is modified to directly invoke
xen_block_handle_requests() and the latter only returns true if it actually
processes any requests. This also means that the call to qemu_bh_schedule()
is moved into xen_block_complete_aio(), which is more intuitive since the
only reason for doing a deferred poll of the shared ring should be because
there were previously insufficient resources to fully complete a previous
poll.
Signed-off-by: Paul Durrant <paul.durrant@citrix.com>
Reviewed-by: Anthony PERARD <anthony.perard@citrix.com>
Message-Id: <20190408151617.13025-4-paul.durrant@citrix.com>
Signed-off-by: Anthony PERARD <anthony.perard@citrix.com>
This patch adds an AioContext parameter to xen_device_bind_event_channel()
and then uses aio_set_fd_handler() to set the callback rather than
qemu_set_fd_handler().
Signed-off-by: Paul Durrant <paul.durrant@citrix.com>
Reviewed-by: Anthony PERARD <anthony.perard@citrix.com>
Message-Id: <20190408151617.13025-3-paul.durrant@citrix.com>
[Call aio_set_fd_handler() with is_external=true]
Signed-off-by: Anthony PERARD <anthony.perard@citrix.com>
To better support use of IOThread-s it will be necessary to be able to set
the AioContext for each XenEventChannel and hence it is necessary to open a
separate handle to libxenevtchan for each channel.
This patch stops using NotifierList for event channel callbacks, replacing
that construct by a list of complete XenEventChannel structures. Each of
these now has a xenevtchn_handle pointer in place of the single pointer
previously held in the XenDevice structure. The individual handles are
opened/closed in xen_device_bind/unbind_event_channel(), replacing the
single open/close in xen_device_realize/unrealize().
NOTE: This patch does not add an AioContext parameter to
xen_device_bind_event_channel(). That will be done in a subsequent
patch.
Signed-off-by: Paul Durrant <paul.durrant@citrix.com>
Reviewed-by: Anthony PERARD <anthony.perard@citrix.com>
Message-Id: <20190408151617.13025-2-paul.durrant@citrix.com>
Signed-off-by: Anthony PERARD <anthony.perard@citrix.com>
...that maintains compatibility with existing Xen toolstacks.
Xen toolstacks instantiate PV backends by simply writing information into
xenstore and expecting a backend implementation to be watching for this.
This patch adds a new 'xen-backend' module to allow individual XenDevice
implementations to register create and destroy functions. The creator
will be called when a tool-stack instantiates a new backend in this way,
and the destructor will then be called after the resulting XenDevice
object is unrealized.
To support this it is also necessary to add new watchers into the XenBus
implementation to handle enumeration of new backends and also destruction
of XenDevice-s when the toolstack sets the backend 'online' key to 0.
NOTE: This patch only adds the framework. A subsequent patch will add a
creator function for xen-block devices.
Signed-off-by: Paul Durrant <paul.durrant@citrix.com>
Reviewed-by: Anthony Perard <anthony.perard@citrix.com>
Signed-off-by: Anthony PERARD <anthony.perard@citrix.com>
...and wire in the dataplane.
This patch adds the remaining code to make the xen-block XenDevice
functional. The parameters that a block frontend expects to find are
populated in the backend xenstore area, and the 'ring-ref' and
'event-channel' values specified in the frontend xenstore area are
mapped/bound and used to set up the dataplane.
Signed-off-by: Paul Durrant <paul.durrant@citrix.com>
Reviewed-by: Anthony Perard <anthony.perard@citrix.com>
Signed-off-by: Anthony PERARD <anthony.perard@citrix.com>
The legacy PV backend infrastructure provides functions to bind, unbind
and send notifications to event channnels. Similar functionality will be
required by XenDevice implementations so this patch adds the necessary
support.
Signed-off-by: Paul Durrant <paul.durrant@citrix.com>
Reviewed-by: Anthony Perard <anthony.perard@citrix.com>
Signed-off-by: Anthony PERARD <anthony.perard@citrix.com>
Patch squashed with:
Patch "xen: add event channel interface for XenDevice-s" makes use of
the type xenevtchn_port_or_error_t, but this isn't avaiable before Xen
4.7. Also the function xen_device_bind_event_channel assign the return
value of xenevtchn_bind_interdomain to channel->local_port but check the
result for error with xendev->local_port.
Fix by:
- removing local_port from struct XenDevice as it isn't use anywere.
- adding a compatibility typedef for xenevtchn_port_or_error_t for Xen
4.6 and earlier.
As extra, replace the type of XenEventChannel->local_port by
evtchn_port_t.
Signed-off-by: Anthony PERARD <anthony.perard@citrix.com>
Reviewed-by: Paul Durrant <paul.durrant@citrix.com>
The legacy PV backend infrastructure provides functions to map, unmap and
copy pages granted by frontends. Similar functionality will be required
by XenDevice implementations so this patch adds the necessary support.
Signed-off-by: Paul Durrant <paul.durrant@citrix.com>
Reviewed-by: Anthony Perard <anthony.perard@citrix.com>
Signed-off-by: Anthony PERARD <anthony.perard@citrix.com>
A Xen PV frontend communicates its state to the PV backend by writing to
the 'state' key in the frontend area in xenstore. It is therefore
necessary for a XenDevice implementation to be notified whenever the
value of this key changes.
This patch adds code to do this as follows:
- an 'fd handler' is registered on the libxenstore handle which will be
triggered whenever a 'watch' event occurs
- primitives are added to xen-bus-helper to add or remove watch events
- a list of Notifier objects is added to XenBus to provide a mechanism
to call the appropriate 'watch handler' when its associated event
occurs
The xen-block implementation is extended with a 'frontend_changed' method,
which calls as-yet stub 'connect' and 'disconnect' functions when the
relevant frontend state transitions occur. A subsequent patch will supply
a full implementation for these functions.
Signed-off-by: Paul Durrant <paul.durrant@citrix.com>
Reviewed-by: Anthony Perard <anthony.perard@citrix.com>
Signed-off-by: Anthony PERARD <anthony.perard@citrix.com>
This patch adds a new source module, xen-bus-helper.c, which builds on
basic libxenstore primitives to provide functions to create (setting
permissions appropriately) and destroy xenstore areas, and functions to
'printf' and 'scanf' nodes therein. The main xen-bus code then uses
these primitives [1] to initialize and destroy the frontend and backend
areas for a XenDevice during realize and unrealize respectively.
The 'xen-block' implementation is extended with a 'get_name' method that
returns the VBD number. This number is required to 'name' the xenstore
areas.
NOTE: An exit handler is also added to make sure the xenstore areas are
cleaned up if QEMU terminates without devices being unrealized.
[1] The 'scanf' functions are actually not yet needed, but they will be
needed by code delivered in subsequent patches.
Signed-off-by: Paul Durrant <paul.durrant@citrix.com>
Reviewed-by: Anthony Perard <anthony.perard@citrix.com>
Signed-off-by: Anthony PERARD <anthony.perard@citrix.com>
This patch adds the basic boilerplate for a 'XenBus' object that will act
as a parent to 'XenDevice' PV backends.
A new 'XenBridge' object is also added to connect XenBus to the system bus.
The XenBus object is instantiated by a new xen_bus_init() function called
from the same sites as the legacy xen_be_init() function.
Subsequent patches will flesh-out the functionality of these objects.
Signed-off-by: Paul Durrant <paul.durrant@citrix.com>
Reviewed-by: Anthony Perard <anthony.perard@citrix.com>
Signed-off-by: Anthony PERARD <anthony.perard@citrix.com>
