| Commit message (Collapse) | Author | Files | Lines |
|---|
|
These const functions make it possible to use enums easily together
with the bitfield-struct crate.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
|
|
This avoids the repeated ".0" when using the Interrupt struct.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
|
|
One common thing that device emulation does is manipulate bitmasks, for example
to check whether two bitmaps have common bits. One example in the pl011 crate
is the checks for pending interrupts, where an interrupt cause corresponds to
at least one interrupt source from a fixed set.
Unfortunately, this is one case where Rust *can* provide some kind of
abstraction but it does so with a rather Perl-ish There Is More Way To
Do It. It is not something where a crate like "bilge" helps, because
it only covers the packing of bits in a structure; operations like "are
all bits of Y set in X" almost never make sense for bit-packed structs;
you need something else, there are several crates that do it and of course
we're going to roll our own.
In particular I examined three:
- bitmask (https://docs.rs/bitmask/0.5.0/bitmask/) does not support const
at all. This is a showstopper because one of the ugly things in the
current pl011 code is the ugliness of code that defines interrupt masks
at compile time:
pub const E: Self = Self(Self::OE.0 | Self::BE.0 | Self::PE.0 | Self::FE.0);
or even worse:
const IRQMASK: [u32; 6] = [
Interrupt::E.0 | Interrupt::MS.0 | Interrupt::RT.0 | Interrupt::TX.0 | Interrupt::RX.0,
...
}
You would have to use roughly the same code---"bitmask" only helps with
defining the struct.
- bitmask_enum (https://docs.rs/bitmask-enum/2.2.5/bitmask_enum/) does not
have a good separation of "valid" and "invalid" bits, so for example "!x"
will invert all 16 bits if you choose u16 as the representation -- even if
you only defined 10 bits. This makes it easier to introduce subtle bugs
in comparisons.
- bitflags (https://docs.rs/bitflags/2.6.0/bitflags/) is generally the most
used such crate and is the one that I took most inspiration from with
respect to the syntax. It's a pretty sophisticated implementation,
with a lot of bells and whistles such as an implementation of "Iter"
that returns the bits one at a time.
The main thing that all of them lack, however, is a way to simplify the
ugly definitions like the above. "bitflags" includes const methods that
perform AND/OR/XOR of masks (these are necessary because Rust operator
overloading does not support const yet, and therefore overloaded operators
cannot be used in the definition of a "static" variable), but they become
even more verbose and unmanageable, like
Interrupt::E.union(Interrupt::MS).union(Interrupt::RT).union(Interrupt::TX).union(Interrupt::RX)
This was the main reason to create "bits", which allows something like
bits!(Interrupt: E | MS | RT | TX | RX)
and expands it 1) add "Interrupt::" in front of all identifiers 2) convert
operators to the wordy const functions like "union". It supports boolean
operators "&", "|", "^", "!" and parentheses, with a relatively simple
recursive descent parser that's implemented in qemu_api_macros.
Since I don't remember exactly how the macro was developed, I cannot exclude
that it contains code from "bitflags". Therefore, I am conservatively leaving
in the MIT and Apache 2.0 licenses from bitflags. In fact, I think there
would be a benefit in being able to push code back to "bitflags" anyway
whenever applicable, so that the two libraries do not diverge too much,
so that's another reason to use this.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
|
|
Use PRI formats where required.
Cc: Isaku Yamahata <isaku.yamahata@intel.com>
Signed-off-by: Cédric Le Goater <clg@redhat.com>
Link: https://lore.kernel.org/r/20250602173101.1052983-3-clg@redhat.com
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
|
|
Use PRI formats where required and fix pointer cast.
Cc: Xiaoyao Li <xiaoyao.li@intel.com>
Signed-off-by: Cédric Le Goater <clg@redhat.com>
Link: https://lore.kernel.org/r/20250602173101.1052983-2-clg@redhat.com
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
|
|
Fixed commit introduced common dependencies for target libraries. Alas,
it wrongly reused the 'target' variable, which was previously set from
another loop.
Thus, some dependencies were missing depending on order of target list,
as found here [1].
The fix is to use the correct config_base_arch instead.
Kudos to Thomas Huth who had this right, before I reimplement it, and
introduce this bug.
[1] https://lore.kernel.org/qemu-devel/c54469ce-0385-4aea-b345-47711e9e61de@linaro.org/
Fixes: 4fb54de823e9 (meson: build target libraries with common dependencies)
Signed-off-by: Pierrick Bouvier <pierrick.bouvier@linaro.org>
Reviewed-by: Thomas Huth <thuth@redhat.com>
Tested-by: Cédric Le Goater <clg@redhat.com>
Link: https://lore.kernel.org/r/20250602233801.2699961-1-pierrick.bouvier@linaro.org
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
|
|
The immediate form of MSR access instructions are primarily motivated by
performance, not code size: by having the MSR number in an immediate, it
is available *much* earlier in the pipeline, which allows the hardware
much more leeway about how a particular MSR is handled.
Signed-off-by: Xin Li (Intel) <xin@zytor.com>
Link: https://lore.kernel.org/r/20250103084827.1820007-4-xin@zytor.com
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
|
|
The immediate form of MSR access instructions will use this new CPU
feature word.
Signed-off-by: Xin Li (Intel) <xin@zytor.com>
Link: https://lore.kernel.org/r/20250103084827.1820007-3-xin@zytor.com
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
|
|
WRMSRNS doesn't become a required feature for FERD, and Linux has
removed the dependency, as such remove it from Qemu.
Cc: qemu-stable@nongnu.org
Signed-off-by: Xin Li (Intel) <xin@zytor.com>
Reviewed-by: Xiaoyao Li <xiaoyao.li@intel.com>
Link: https://lore.kernel.org/r/20250103084827.1820007-2-xin@zytor.com
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
|
|
Meson has support for invoking clippy and rustdoc on all crates (1.7.0 for
clippy, 1.8.0 for rustdoc). Use it instead of the homegrown version; this
requires disabling the multiple_crate_versions lint (the only one that was
enabled from the "cargo" group)---which was not particularly useful anyway
because all dependencies are converted by hand into Meson subprojects.
rustfmt is still not supported.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
|
|
This is not needed anymore now that tests link with libqemuutil.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
|
|
Doctests are weird. They are essentially integration tests, but they're
"ran" by executing rustdoc --test, which takes a compiler-ish
command line. This is supported by Meson 1.8.0.
Because they run the linker and need all the .o files, run them in the
build jobs rather than the test jobs.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
|
|
rustdoc is effectively a custom version of rustc, and it is necessary to
specify it in order to run doctests from Meson. Add the relevant configure
option and environment variables.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
|
|
libqemuutil is not meant be linked as a whole; if modules are enabled, doing
so results in undefined symbols (corresponding to QMP commands) in
rust/qemu-api/rust-qemu-api-integration.
Support for "objects" in Rust executables is available in Meson 1.8.0; use it
to switching to the same dependencies that C targets use: link_with for
libqemuutil, and objects for everything else.
Reported-by: Bernhard Beschow <shentey@gmail.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
|
|
This adds several improvements to Rust support, including
native clippy and rustdoc targets, the "objects" keyword,
and running doctests.
Require it only when Rust support is requested, to avoid
putting a strict requirement on all build platforms for the
sake of an experimental feature.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
|
|
This is produced by recent versions of bindgen:
warning: use of `offset` with a `usize` casted to an `isize`
--> /builds/bonzini/qemu/rust/target/debug/build/qemu_api-35cb647f4db404b8/out/bindings.inc.rs:39:21
|
39 | let byte = *(core::ptr::addr_of!((*this).storage) as *const u8).offset(byte_index as isize);
| ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ help: try: `(core::ptr::addr_of!((*this).storage) as *const u8).add(byte_index)`
|
= help: for further information visit https://rust-lang.github.io/rust-clippy/master/index.html#ptr_offset_with_cast
= note: `#[warn(clippy::ptr_offset_with_cast)]` on by default
warning: use of `offset` with a `usize` casted to an `isize`
--> /builds/bonzini/qemu/rust/target/debug/build/qemu_api-35cb647f4db404b8/out/bindings.inc.rs:68:13
|
68 | (core::ptr::addr_of_mut!((*this).storage) as *mut u8).offset(byte_index as isize);
| ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ help: try: `(core::ptr::addr_of_mut!((*this).storage) as *mut u8).add(byte_index)`
|
= help: for further information visit https://rust-lang.github.io/rust-clippy/master/index.html#ptr_offset_with_cast
This seems to be new in bindgen 0.71.0, possibly related to bindgen
commit 33006185b7878 ("Add raw_ref_macros feature", 2024-11-22).
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
|
|
Commit 56b1f50e3c10 ("hw/i386/pc: Wire RTC ISA IRQs in south bridges")
attempted to refactor RTC IRQ wiring which was previously done in
pc_basic_device_init() but forgot about the isapc machine. Fix this by
wiring in the code section dedicated exclusively to the isapc machine.
Resolves: https://gitlab.com/qemu-project/qemu/-/issues/2961
Fixes: 56b1f50e3c10 ("hw/i386/pc: Wire RTC ISA IRQs in south bridges")
cc: qemu-stable
Signed-off-by: Bernhard Beschow <shentey@gmail.com>
Reviewed-by: Mark Cave-Ayland <mark.caveayland@nutanix.com>
Message-Id: <20250526203820.1853-1-shentey@gmail.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
|
|
Current memory operations like pinning may take a lot of time at the
destination. Currently they are done after the source of the migration is
stopped, and before the workload is resumed at the destination. This is a
period where neigher traffic can flow, nor the VM workload can continue
(downtime).
We can do better as we know the memory layout of the guest RAM at the
destination from the moment that all devices are initializaed. So
moving that operation allows QEMU to communicate the kernel the maps
while the workload is still running in the source, so Linux can start
mapping them.
As a small drawback, there is a time in the initialization where QEMU
cannot respond to QMP etc. By some testing, this time is about
0.2seconds. This may be further reduced (or increased) depending on the
vdpa driver and the platform hardware, and it is dominated by the cost
of memory pinning.
This matches the time that we move out of the called downtime window.
The downtime is measured as the elapsed trace time between the last
vhost_vdpa_suspend on the source and the last vhost_vdpa_set_vring_enable_one
on the destination. In other words, from "guest CPUs freeze" to the
instant the final Rx/Tx queue-pair is able to start moving data.
Using ConnectX-6 Dx (MLX5) NICs in vhost-vDPA mode with 8 queue-pairs,
the series reduces guest-visible downtime during back-to-back live
migrations by more than half:
- 39G VM: 4.72s -> 2.09s (-2.63s, ~56% improvement)
- 128G VM: 14.72s -> 5.83s (-8.89s, ~60% improvement)
Tested-by: Lei Yang <leiyang@redhat.com>
Reviewed-by: Si-Wei Liu <si-wei.liu@oracle.com>
Acked-by: Jason Wang <jasowang@redhat.com>
Signed-off-by: Eugenio Pérez <eperezma@redhat.com>
Signed-off-by: Jonah Palmer <jonah.palmer@oracle.com>
Message-Id: <20250522145839.59974-8-jonah.palmer@oracle.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
|
|
As we are moving to keep the mapping through all the vdpa device life
instead of resetting it at VirtIO reset, we need to move all its
dependencies to the initialization too. In particular devices with
x-svq=on need a valid iova_tree from the beginning.
Simplify the code also consolidating the two creation points: the first
data vq in case of SVQ active and CVQ start in case only CVQ uses it.
Tested-by: Lei Yang <leiyang@redhat.com>
Reviewed-by: Si-Wei Liu <si-wei.liu@oracle.com>
Acked-by: Jason Wang <jasowang@redhat.com>
Suggested-by: Si-Wei Liu <si-wei.liu@oracle.com>
Signed-off-by: Eugenio Pérez <eperezma@redhat.com>
Signed-off-by: Jonah Palmer <jonah.palmer@oracle.com>
Message-Id: <20250522145839.59974-7-jonah.palmer@oracle.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
|
|
Since commit f6fe3e333f ("vdpa: move memory listener to
vhost_vdpa_shared") this piece of code repeatedly assign
shared->listener members. This was not a problem as it was not used
until device start.
However next patches move the listener registration to this
vhost_vdpa_init function. When the listener is registered it is added
to an embedded linked list, so setting its members again will cause
memory corruption to the linked list node.
Do the right thing and only set it in the first vdpa device.
Tested-by: Lei Yang <leiyang@redhat.com>
Reviewed-by: Si-Wei Liu <si-wei.liu@oracle.com>
Acked-by: Jason Wang <jasowang@redhat.com>
Signed-off-by: Eugenio Pérez <eperezma@redhat.com>
Signed-off-by: Jonah Palmer <jonah.palmer@oracle.com>
Message-Id: <20250522145839.59974-6-jonah.palmer@oracle.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
|
|
Check if the listener has been registered or not, so it needs to be
registered again at start.
Tested-by: Lei Yang <leiyang@redhat.com>
Reviewed-by: Si-Wei Liu <si-wei.liu@oracle.com>
Acked-by: Jason Wang <jasowang@redhat.com>
Signed-off-by: Eugenio Pérez <eperezma@redhat.com>
Signed-off-by: Jonah Palmer <jonah.palmer@oracle.com>
Message-Id: <20250522145839.59974-5-jonah.palmer@oracle.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
|
|
The backend does not reset them until the vdpa file descriptor is closed
so there is no harm in doing it only once.
This allows the destination of a live migration to premap memory in
batches, using VHOST_BACKEND_F_IOTLB_BATCH.
Tested-by: Lei Yang <leiyang@redhat.com>
Reviewed-by: Si-Wei Liu <si-wei.liu@oracle.com>
Acked-by: Jason Wang <jasowang@redhat.com>
Signed-off-by: Eugenio Pérez <eperezma@redhat.com>
Signed-off-by: Jonah Palmer <jonah.palmer@oracle.com>
Message-Id: <20250522145839.59974-4-jonah.palmer@oracle.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
|
|
It will be used directly by vhost_vdpa_init.
Tested-by: Lei Yang <leiyang@redhat.com>
Reviewed-by: Si-Wei Liu <si-wei.liu@oracle.com>
Acked-by: Jason Wang <jasowang@redhat.com>
Signed-off-by: Eugenio Pérez <eperezma@redhat.com>
Signed-off-by: Jonah Palmer <jonah.palmer@oracle.com>
Message-Id: <20250522145839.59974-3-jonah.palmer@oracle.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
|
|
To map the guest memory while it is migrating we need to create the
iova_tree, as long as the destination uses x-svq=on. Checking to not
override it.
The function vhost_vdpa_net_client_stop clear it if the device is
stopped. If the guest starts the device again, the iova tree is
recreated by vhost_vdpa_net_data_start_first or vhost_vdpa_net_cvq_start
if needed, so old behavior is kept.
Tested-by: Lei Yang <leiyang@redhat.com>
Reviewed-by: Si-Wei Liu <si-wei.liu@oracle.com>
Acked-by: Jason Wang <jasowang@redhat.com>
Signed-off-by: Eugenio Pérez <eperezma@redhat.com>
Signed-off-by: Jonah Palmer <jonah.palmer@oracle.com>
Message-Id: <20250522145839.59974-2-jonah.palmer@oracle.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
|
|
The vring call fd is set even when the guest does not use MSI-X (e.g., in the
case of virtio PMD), leading to unnecessary CPU overhead for processing
interrupts.
The commit 96a3d98d2c("vhost: don't set vring call if no vector") optimized the
case where MSI-X is enabled but the queue vector is unset. However, there's an
additional case |