There is no generic acceleration, we have to use specific
implementations. Make the base class abstract.
Fixes: b14a0b7469 ("accel: Use QOM classes for accel types")
Signed-off-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Reviewed-by: Cornelia Huck <cohuck@redhat.com>
Message-Id: <20200129212345.20547-3-philmd@redhat.com>
Signed-off-by: Philippe Mathieu-Daudé <philmd@linaro.org>
cpu_common_initfn() is our target agnostic initializer,
while cpu_exec_initfn() is the target specific one.
The %as and %num_ases fields are not target specific,
so initialize them in the common helper.
Signed-off-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Message-Id: <20250123234415.59850-6-philmd@linaro.org>
Cache CPUClass as early as possible, when the instance
is initialized.
Signed-off-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Message-Id: <20250123234415.59850-5-philmd@linaro.org>
Use the tcg_enabled() check so the compiler can elide
the call when TCG isn't available, allowing to remove
the tb_flush() stub.
Signed-off-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Reviewed-by: Alex Bennée <alex.bennee@linaro.org>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Message-Id: <20250123234415.59850-4-philmd@linaro.org>
Both CPUClass::gdb_read_register() and CPUClass::gdb_write_register()
handlers are called from common gdbstub code, and won't be called with
register index over CPUClass::gdb_num_core_regs:
int gdb_read_register(CPUState *cpu, GByteArray *buf, int reg)
{
CPUClass *cc = CPU_GET_CLASS(cpu);
if (reg < cc->gdb_num_core_regs) {
return cc->gdb_read_register(cpu, buf, reg);
}
...
}
static int gdb_write_register(CPUState *cpu, uint8_t *mem_buf, int reg)
{
CPUClass *cc = CPU_GET_CLASS(cpu);
if (reg < cc->gdb_num_core_regs) {
return cc->gdb_write_register(cpu, mem_buf, reg);
}
...
}
Clarify that in CPUClass docstring, and remove unreachable code on
the microblaze and openrisc implementations.
Signed-off-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Reviewed-by: Alex Bennée <alex.bennee@linaro.org>
Message-Id: <20250122093028.52416-3-philmd@linaro.org>
We already have "qemu/compiler.h" for compiler-specific arrangements,
automatically included by "qemu/osdep.h" for each source file. No
need to explicitly include a header for a Clang particularity.
Suggested-by: Pierrick Bouvier <pierrick.bouvier@linaro.org>
Reviewed-by: Pierrick Bouvier <pierrick.bouvier@linaro.org>
Reviewed-by: Alex Bennée <alex.bennee@linaro.org>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Message-Id: <20250117170201.91182-1-philmd@linaro.org>
The dubious casts of mutable references to objects are not used
anymore: the wrappers for qdev_init_clock_in and for IRQ and MMIO
initialization can be called directly on the subclasses, without
casts, plus they take a shared reference so they can just use
"upcast()" instead of "upcast_mut()". Remove them.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Generalize timer_and_addr() to decode all registers into a single enum
HPETRegister, and use the TryInto derive to separate valid and
invalid values.
The main advantage lies in checking that all registers are enumerated
in the "match" statements.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
The values stored in the Fifo are instances of the bitfield-struct
registers::Data. Convert as soon as possible the value written
into DR, and always refer to the bitfield struct; it's generally
cleaner other than PL011State::receive having to do a double
conversion u8=>u32=>registers::Data.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Switch bindings::CharBackend with chardev::CharBackend. This removes
occurrences of "unsafe" due to FFI and switches the wrappers for receive,
can_receive and event callbacks to the common ones implemented by
chardev::CharBackend.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Most of the character device API is pretty simple, with "0 or -errno"
or "number of bytes or -errno" as the convention for return codes.
Add safe wrappers for the API to the CharBackend bindgen-generated
struct.
The API is not complete, but it covers the parts that are used
by the PL011 device, plus qemu_chr_fe_write which is needed to
implement the standard library Write trait.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Send and Sync are now implemented on the opaque wrappers. Remove them
from the bindings module, unless the structs are pure data containers
and/or have no C functions defined on them.
Reviewed-by: Zhao Liu <zhao1.liu@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Timers must be pinned in memory, because modify() stores a pointer to them
in the TimerList. To express this requirement, change init_full() to take
a pinned reference. Because the only way to obtain a Timer is through
Timer::new(), which is unsafe, modify() can assume that the timer it got
was later initialized; and because the initialization takes a Pin<&mut
Timer> modify() can assume that the timer is pinned. In the future the
pinning requirement will be expressed through the pin_init crate instead.
Note that Timer is a bit different from other users of Opaque, in that
it is created in Rust code rather than C code. This is why it has to
use the unsafe constructors provided by Opaque; and in fact Timer::new()
is also unsafe, because it leaves it to the caller to invoke init_full()
before modify(). Without a call to init_full(), modify() will cause a
NULL pointer dereference.
An alternative could be to combine new() + init_full() by returning a
pinned box; however, using a reference makes it easier to express
the requirement that the opaque outlives the timer.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
This simplifies things for migration, since Option<Box<QEMUTimer>> does not
implement VMState.
This also shows a soundness issue because Timer::new() will leave a NULL
timer list pointer, which can then be dereferenced by Timer::modify(). It
will be fixed shortly.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Add a derive macro that makes it easy to peel off all the layers of
specialness (UnsafeCell, MaybeUninit, etc.) and just get a pointer
to the wrapped type; and likewise add them back starting from a
*mut.
Reviewed-by: Zhao Liu <zhao1.liu@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Inspired by the same-named type in Linux. This type provides the compiler
with a correct view of what goes on with FFI types. In addition, it
separates the glue code from the bindgen-generated code, allowing
traits such as Send, Sync or Zeroable to be specified independently
for C and Rust structs.
Reviewed-by: Zhao Liu <zhao1.liu@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Dump sys.stdin when it errors on meson-buildoptions.py, letting us debug
the build errors instead of just saying "Couldn't parse"
Signed-off-by: Nabih Estefan <nabihestefan@google.com>
Signed-off-by: Patrick Venture <venture@google.com>
Link: https://lore.kernel.org/r/20250227180454.2006757-1-venture@google.com
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
chardev is using qio functions, so express that in the Meson internal
dependency. (I found this when adding character devices bindings for
Rust; they initially needed the io dependency added by hand).
Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
QAPI's 'prefix' feature can make the connection between enumeration
type and its constants less than obvious. It's best used with
restraint. Commit 7bbadc60b5..64f5e9db77 eliminated most uses.
Discourage new ones.
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Message-ID: <20250228134335.132278-1-armbru@redhat.com>
Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Message-ID: <20250227085601.4140852-7-armbru@redhat.com>
Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>
Consistently use format "DESCRIPTION (VALUE/VALUE...)".
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Message-ID: <20250227085601.4140852-6-armbru@redhat.com>
Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>
PropertyInfo member @type is externally visible via QMP
device-list-properties and qom-list-properies.
Its meaning is not documented at its definition.
It gets passed as @type argument to object_property_add() and
object_class_property_add(). This argument's documentation isn't of
much help, either:
* @type: the type name of the property. This namespace is pretty loosely
* defined. Sub namespaces are constructed by using a prefix and then
* to angle brackets. For instance, the type 'virtio-net-pci' in the
* 'link' namespace would be 'link<virtio-net-pci>'.
The two QMP commands document it as
# @type: the type of the property. This will typically come in one of
# four forms:
#
# 1) A primitive type such as 'u8', 'u16', 'bool', 'str', or
# 'double'. These types are mapped to the appropriate JSON
# type.
#
# 2) A child type in the form 'child<subtype>' where subtype is a
# qdev device type name. Child properties create the
# composition tree.
#
# 3) A link type in the form 'link<subtype>' where subtype is a
# qdev device type name. Link properties form the device model
# graph.
"Typically come in one of four forms" followed by three items inspires
the level of trust that is appropriate here.
Clean up a bunch of funnies:
* qdev_prop_fdc_drive_type.type is "FdcDriveType". Its .enum_table
refers to QAPI type "FloppyDriveType". So use that.
* qdev_prop_reserved_region is "reserved_region". Its only user is an
array property called "reserved-regions". Its .set() visits str.
So change @type to "str".
* trng_prop_fault_event_set.type is "uint32:bits". Its .set() visits
uint32, so change @type to "uint32". If we believe mentioning it's
actually bits is useful, the proper place would be .description.
* ccw_loadparm.type is "ccw_loadparm". It's users are properties
called "loadparm". Its .set() visits str. So change @type to
"str".
* qdev_prop_nv_gpudirect_clique.type is "uint4". Its set() visits
uint8, so change @type to "uint8". If we believe mentioning the
range is useful, the proper place would be .description.
* s390_pci_fid_propinfo.type is "zpci_fid". Its .set() visits uint32.
So change type to that, and move the "zpci_fid" to .description.
This is admittedly a lousy description, but it's still an
improvement; for instance, output of -device zpci,help changes from
fid=<zpci_fid>
to
fid=<uint32> - zpci_fid
* Similarly for a raft of PropertyInfo in target/riscv/cpu.c.
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Message-ID: <20250227085601.4140852-5-armbru@redhat.com>
Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>
[Commit message typo fixed]
PropertyInfo member @name becomes ObjectProperty member @type, while
Property member @name becomes ObjectProperty member @name. Rename the
former.
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Message-ID: <20250227085601.4140852-4-armbru@redhat.com>
Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>
[One missed instance of @type fixed]
Properties using qdev_prop_pci_devfn initially accepted a string of
the form "DEV.FN" or "DEV" where DEV and FN are in hexadecimal.
Member @name was "pci-devfn" initially.
Commit b403298adb (qdev: make the non-legacy pci address property
accept an integer) changed them to additionally accept integers: bits
3..7 are DEV, and bits 0..2 are FN. This is inaccessible externally
in device_add so far.
The commit also changed @name to "int32", and set member @legacy-name
to "pci-devfn". Together, this kept QMP command
device-list-properties unaffected: it used @name only when
@legacy_name was null.
Commit 07d09c58db (qmp: Print descriptions of object properties)
quietly dumbed that down to use @name always, and the next commit
18b91a3e082q (qdev: Drop legacy_name from qdev properties) dropped
member @legacy_name. This changed the value of @type reported by QMP
command device-list-properties from "pci-devfn" to "int32".
But "int32" is misleading: device_add actually wants QAPI type "str".
So change @name to that.
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Message-ID: <20250227085601.4140852-3-armbru@redhat.com>
Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Message-ID: <20250227085601.4140852-2-armbru@redhat.com>
Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>
A TODO comment in class Annotated reminds us to simplify it once we
can use @dataclass, new in Python 3.7. We have that now, so do it.
There's a similar comment in scripts/qapi/source.py, but I can't
figure out how to use @dataclass there. Left for another day.
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Message-ID: <20250227080757.3978333-4-armbru@redhat.com>
Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>
We use OrderedDict to ensure dictionary order is insertion order.
Plain dict does that since Python 3.6, but it wasn't guaranteed until
3.7. Since we have 3.7 now, replace OrderedDict by dict.
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Message-ID: <20250227080757.3978333-3-armbru@redhat.com>
Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>
Fixes: ca056f4499 (Python: Drop support for Python 3.7)
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Message-ID: <20250227080757.3978333-2-armbru@redhat.com>
Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>
maxmem=4G is too large to address on 32-bit hosts, so reduce it
to 2G since the tuxrun tests don't actually need such an elevated
memory limit.
Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
Reviewed-by: Thomas Huth <thuth@redhat.com>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Message-ID: <20250228102738.3064045-5-berrange@redhat.com>
Signed-off-by: Thomas Huth <thuth@redhat.com>
The 'qemu_bin' field is currently set on the class, despite being
accessed as if it were an object instance field with 'self.qemu_bin'.
This is no obvious need to have it as a class field, so move it into
the object instance.
Reviewed-by: Thomas Huth <thuth@redhat.com>
Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Message-ID: <20250228102738.3064045-3-berrange@redhat.com>
Signed-off-by: Thomas Huth <thuth@redhat.com>
This was copied over from avocado but has not been used in the new
functional tests.
Reviewed-by: Thomas Huth <thuth@redhat.com>
Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Message-ID: <20250228102738.3064045-2-berrange@redhat.com>
Signed-off-by: Thomas Huth <thuth@redhat.com>
DEFINE_PROP_ON_OFF_AUTO() property isn't runtime-mutable so using it
would mean that the source VM would need to decide upfront at startup
time whether it wants to do a multifd device state transfer at some
point.
Source VM can run for a long time before being migrated so it is
desirable to have a fallback mechanism to the old way of transferring
VFIO device state if it turns to be necessary.
This brings this property to the same mutability level as ordinary
migration parameters, which too can be adjusted at the run time.
Signed-off-by: Maciej S. Szmigiero <maciej.szmigiero@oracle.com>
Reviewed-by: Cédric Le Goater <clg@redhat.com>
Link: https://lore.kernel.org/qemu-devel/f2f2d66bda477da3e6cb8c0311006cff36e8651d.1741124640.git.maciej.szmigiero@oracle.com
Signed-off-by: Cédric Le Goater <clg@redhat.com>
This property allows configuring whether to transfer the particular device
state via multifd channels when live migrating that device.
It defaults to AUTO, which means that VFIO device state transfer via
multifd channels is attempted in configurations that otherwise support it.
Signed-off-by: Maciej S. Szmigiero <maciej.szmigiero@oracle.com>
Reviewed-by: Cédric Le Goater <clg@redhat.com>
Link: https://lore.kernel.org/qemu-devel/d6dbb326e3d53c7104d62c96c9e3dd64e1c7b940.1741124640.git.maciej.szmigiero@oracle.com
[ clg: Added documentation ]
Signed-off-by: Cédric Le Goater <clg@redhat.com>
Implement the multifd device state transfer via additional per-device
thread inside save_live_complete_precopy_thread handler.
Switch between doing the data transfer in the new handler and doing it
in the old save_state handler depending if VFIO multifd transfer is enabled
or not.
Signed-off-by: Maciej S. Szmigiero <maciej.szmigiero@oracle.com>
Reviewed-by: Cédric Le Goater <clg@redhat.com>
Link: https://lore.kernel.org/qemu-devel/4d727e2e0435e0022d50004e474077632830e08d.1741124640.git.maciej.szmigiero@oracle.com
[ clg: - Reordered savevm_vfio_handlers
- Updated save_live_complete_precopy* documentation ]
Signed-off-by: Cédric Le Goater <clg@redhat.com>
Add a thread which loads the VFIO device state buffers that were received
via multifd.
Each VFIO device that has multifd device state transfer enabled has one
such thread, which is created using migration core API
qemu_loadvm_start_load_thread().
Since it's important to finish loading device state transferred via the
main migration channel (via save_live_iterate SaveVMHandler) before
starting loading the data asynchronously transferred via multifd the thread
doing the actual loading of the multifd transferred data is only started
from switchover_start SaveVMHandler.
switchover_start handler is called when MIG_CMD_SWITCHOVER_START
sub-command of QEMU_VM_COMMAND is received via the main migration channel.
This sub-command is only sent after all save_live_iterate data have already
been posted so it is safe to commence loading of the multifd-transferred
device state upon receiving it - loading of save_live_iterate data happens
synchronously in the main migration thread (much like the processing of
MIG_CMD_SWITCHOVER_START) so by the time MIG_CMD_SWITCHOVER_START is
processed all the proceeding data must have already been loaded.
Signed-off-by: Maciej S. Szmigiero <maciej.szmigiero@oracle.com>
Reviewed-by: Cédric Le Goater <clg@redhat.com>
Link: https://lore.kernel.org/qemu-devel/9abe612d775aaf42e31646796acd2363c723a57a.1741124640.git.maciej.szmigiero@oracle.com
[ clg: - Reordered savevm_vfio_handlers
- Added switchover_start documentation ]
Signed-off-by: Cédric Le Goater <clg@redhat.com>
The multifd received data needs to be reassembled since device state
packets sent via different multifd channels can arrive out-of-order.
Therefore, each VFIO device state packet carries a header indicating its
position in the stream.
The raw device state data is saved into a VFIOStateBuffer for later
in-order loading into the device.
The last such VFIO device state packet should have
VFIO_DEVICE_STATE_CONFIG_STATE flag set and carry the device config state.
Signed-off-by: Maciej S. Szmigiero <maciej.szmigiero@oracle.com>
Reviewed-by: Cédric Le Goater <clg@redhat.com>
Link: https://lore.kernel.org/qemu-devel/e3bff515a8d61c582b94b409eb12a45b1a143a69.1741124640.git.maciej.szmigiero@oracle.com
[ clg: - Reordered savevm_vfio_handlers
- Added load_state_buffer documentation ]
Signed-off-by: Cédric Le Goater <clg@redhat.com>
