As the comment in qapi/error, passing @errp to error_prepend() requires
ERRP_GUARD():
* = Why, when and how to use ERRP_GUARD() =
*
* Without ERRP_GUARD(), use of the @errp parameter is restricted:
...
* - It should not be passed to error_prepend(), error_vprepend() or
* error_append_hint(), because that doesn't work with &error_fatal.
* ERRP_GUARD() lifts these restrictions.
*
* To use ERRP_GUARD(), add it right at the beginning of the function.
* @errp can then be used without worrying about the argument being
* NULL or &error_fatal.
ERRP_GUARD() could avoid the case when @errp is &error_fatal, the user
can't see this additional information, because exit() happens in
error_setg earlier than information is added [1].
In hw/vfio/pci.c, there are 2 functions passing @errp to error_prepend()
without ERRP_GUARD():
- vfio_add_std_cap()
- vfio_realize()
The @errp of vfio_add_std_cap() is also from vfio_realize(). And
vfio_realize(), as a PCIDeviceClass.realize method, its @errp is from
DeviceClass.realize so that there is no guarantee that the @errp won't
point to @error_fatal.
To avoid the issue like [1] said, add missing ERRP_GUARD() at their
beginning.
[1]: Issue description in the commit message of commit ae7c80a7bd
("error: New macro ERRP_GUARD()").
Cc: Alex Williamson <alex.williamson@redhat.com>
Cc: Cédric Le Goater <clg@redhat.com>
Signed-off-by: Zhao Liu <zhao1.liu@intel.com>
Reviewed-by: Cédric Le Goater <clg@redhat.com>
Message-ID: <20240311033822.3142585-24-zhao1.liu@linux.intel.com>
Signed-off-by: Thomas Huth <thuth@redhat.com>
Not used outside C module.
Signed-off-by: Frediano Ziglio <frediano.ziglio@cloud.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Reviewed-by Tokarev <mjt@tls.msk.ru>
Signed-off-by: Michael Tokarev <mjt@tls.msk.ru>
When doing device assignment of a physical device, MSI-X can be
enabled with no vectors enabled and this sets the IRQ index to
VFIO_PCI_MSIX_IRQ_INDEX. However, when MSI-X is disabled, the IRQ
index is left untouched if no vectors are in use. Then, when INTx
is enabled, the IRQ index value is considered incompatible (set to
MSI-X) and VFIO_DEVICE_SET_IRQS fails. QEMU complains with :
qemu-system-x86_64: vfio 0000:08:00.0: Failed to set up TRIGGER eventfd signaling for interrupt INTX-0: VFIO_DEVICE_SET_IRQS failure: Invalid argument
To avoid that, unconditionaly clear the IRQ index when MSI-X is
disabled.
Buglink: https://issues.redhat.com/browse/RHEL-21293
Fixes: 5ebffa4e87 ("vfio/pci: use an invalid fd to enable MSI-X")
Cc: Jing Liu <jing2.liu@intel.com>
Cc: Alex Williamson <alex.williamson@redhat.com>
Reviewed-by: Alex Williamson <alex.williamson@redhat.com>
Signed-off-by: Cédric Le Goater <clg@redhat.com>
VFIOContainerBase was not introduced as an abstract QOM object because
it felt unnecessary to expose all the IOMMU backends to the QEMU
machine and human interface. However, we can still abstract the IOMMU
backend handlers using a QOM interface class. This provides more
flexibility when referencing the various implementations.
Simply transform the VFIOIOMMUOps struct in an InterfaceClass and do
some initial name replacements. Next changes will start converting
VFIOIOMMUOps.
Reviewed-by: Zhenzhong Duan <zhenzhong.duan@intel.com>
Tested-by: Eric Farman <farman@linux.ibm.com>
Signed-off-by: Cédric Le Goater <clg@redhat.com>
Introduce a helper function to replace the common code to initialize
VFIODevice in pci, platform, ap and ccw VFIO device.
No functional change intended.
Suggested-by: Cédric Le Goater <clg@redhat.com>
Signed-off-by: Zhenzhong Duan <zhenzhong.duan@intel.com>
Tested-by: Nicolin Chen <nicolinc@nvidia.com>
Signed-off-by: Cédric Le Goater <clg@redhat.com>
Some of the VFIODevice initializations is in vfio_realize,
move all of them in vfio_instance_init.
No functional change intended.
Suggested-by: Cédric Le Goater <clg@redhat.com>
Signed-off-by: Zhenzhong Duan <zhenzhong.duan@intel.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Tested-by: Nicolin Chen <nicolinc@nvidia.com>
Signed-off-by: Cédric Le Goater <clg@redhat.com>
This gives management tools like libvirt a chance to open the vfio
cdev with privilege and pass FD to qemu. This way qemu never needs
to have privilege to open a VFIO or iommu cdev node.
Together with the earlier support of pre-opening /dev/iommu device,
now we have full support of passing a vfio device to unprivileged
qemu by management tool. This mode is no more considered for the
legacy backend. So let's remove the "TODO" comment.
Add helper functions vfio_device_set_fd() and vfio_device_get_name()
to set fd and get device name, they will also be used by other vfio
devices.
There is no easy way to check if a device is mdev with FD passing,
so fail the x-balloon-allowed check unconditionally in this case.
There is also no easy way to get BDF as name with FD passing, so
we fake a name by VFIO_FD[fd].
Signed-off-by: Zhenzhong Duan <zhenzhong.duan@intel.com>
Reviewed-by: Cédric Le Goater <clg@redhat.com>
Tested-by: Eric Auger <eric.auger@redhat.com>
Tested-by: Nicolin Chen <nicolinc@nvidia.com>
Signed-off-by: Cédric Le Goater <clg@redhat.com>
Now we support two types of iommu backends, let's add the capability
to select one of them. This depends on whether an iommufd object has
been linked with the vfio-pci device:
If the user wants to use the legacy backend, it shall not
link the vfio-pci device with any iommufd object:
-device vfio-pci,host=0000:02:00.0
This is called the legacy mode/backend.
If the user wants to use the iommufd backend (/dev/iommu) it
shall pass an iommufd object id in the vfio-pci device options:
-object iommufd,id=iommufd0
-device vfio-pci,host=0000:02:00.0,iommufd=iommufd0
Suggested-by: Alex Williamson <alex.williamson@redhat.com>
Signed-off-by: Eric Auger <eric.auger@redhat.com>
Signed-off-by: Yi Liu <yi.l.liu@intel.com>
Signed-off-by: Zhenzhong Duan <zhenzhong.duan@intel.com>
Reviewed-by: Cédric Le Goater <clg@redhat.com>
Tested-by: Eric Auger <eric.auger@redhat.com>
Tested-by: Nicolin Chen <nicolinc@nvidia.com>
Signed-off-by: Cédric Le Goater <clg@redhat.com>
Legacy vfio pci and iommufd cdev have different process to hot reset
vfio device, expand current code to abstract out pci_hot_reset callback
for legacy vfio, this same interface will also be used by iommufd
cdev vfio device.
Rename vfio_pci_hot_reset to vfio_legacy_pci_hot_reset and move it
into container.c.
vfio_pci_[pre/post]_reset and vfio_pci_host_match are exported so
they could be called in legacy and iommufd pci_hot_reset callback.
Suggested-by: Cédric Le Goater <clg@redhat.com>
Signed-off-by: Zhenzhong Duan <zhenzhong.duan@intel.com>
Reviewed-by: Eric Auger <eric.auger@redhat.com>
Tested-by: Eric Auger <eric.auger@redhat.com>
Tested-by: Nicolin Chen <nicolinc@nvidia.com>
Signed-off-by: Cédric Le Goater <clg@redhat.com>
This helper will be used by both legacy and iommufd backends.
No functional changes intended.
Signed-off-by: Zhenzhong Duan <zhenzhong.duan@intel.com>
Reviewed-by: Cédric Le Goater <clg@redhat.com>
Reviewed-by: Eric Auger <eric.auger@redhat.com>
Tested-by: Eric Auger <eric.auger@redhat.com>
Tested-by: Nicolin Chen <nicolinc@nvidia.com>
Signed-off-by: Cédric Le Goater <clg@redhat.com>
qemu_uuid_unparse() includes a trailing NUL when writing the uuid
string and the buffer size should be UUID_FMT_LEN + 1 bytes. Use the
recently added UUID_STR_LEN which defines the correct size.
Fixes: CID 1522913
Fixes: 2dca1b37a7 ("vfio/pci: add support for VF token")
Cc: Alex Williamson <alex.williamson@redhat.com>
Reviewed-by: Alex Williamson <alex.williamson@redhat.com>
Reviewed-by: Juan Quintela <quintela@redhat.com>
Reviewed-by: "Denis V. Lunev" <den@openvz.org>
Signed-off-by: Cédric Le Goater <clg@redhat.com>
Add a "VFIODisplay" subsection whenever "x-ramfb-migrate" is turned on.
Turn it off by default on machines <= 8.1 for compatibility reasons.
Signed-off-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Reviewed-by: Laszlo Ersek <lersek@redhat.com>
Acked-by: Gerd Hoffmann <kraxel@redhat.com>
[ clg: - checkpatch fixes
- improved warn_report() in vfio_realize() ]
Signed-off-by: Cédric Le Goater <clg@redhat.com>
In vfio_realize, on the error path, we currently call
vfio_detach_device() after a successful vfio_attach_device.
While this looks natural, vfio_instance_finalize also induces
a vfio_detach_device(), and it seems to be the right place
instead as other resources are released there which happen
to be a prerequisite to a successful UNSET_CONTAINER.
So let's rely on the finalize vfio_detach_device call to free
all the relevant resources.
Fixes: a28e06621170 ("vfio/pci: Introduce vfio_[attach/detach]_device")
Reported-by: Zhenzhong Duan <zhenzhong.duan@intel.com>
Signed-off-by: Eric Auger <eric.auger@redhat.com>
Tested-by: Zhenzhong Duan <zhenzhong.duan@intel.com>
Reviewed-by: Zhenzhong Duan <zhenzhong.duan@intel.com>
Signed-off-by: Cédric Le Goater <clg@redhat.com>
We want the VFIO devices to be able to use two different
IOMMU backends, the legacy VFIO one and the new iommufd one.
Introduce vfio_[attach/detach]_device which aim at hiding the
underlying IOMMU backend (IOCTLs, datatypes, ...).
Once vfio_attach_device completes, the device is attached
to a security context and its fd can be used. Conversely
When vfio_detach_device completes, the device has been
detached from the security context.
At the moment only the implementation based on the legacy
container/group exists. Let's use it from the vfio-pci device.
Subsequent patches will handle other devices.
We also take benefit of this patch to properly free
vbasedev->name on failure.
Signed-off-by: Eric Auger <eric.auger@redhat.com>
Signed-off-by: Yi Liu <yi.l.liu@intel.com>
Signed-off-by: Zhenzhong Duan <zhenzhong.duan@intel.com>
Reviewed-by: Cédric Le Goater <clg@redhat.com>
Signed-off-by: Cédric Le Goater <clg@redhat.com>
During migration restoring, vfio_enable_vectors() is called to restore
enabling MSI-X interrupts for assigned devices. It sets the range from
0 to nr_vectors to kernel to enable MSI-X and the vectors unmasked in
guest. During the MSI-X enabling, all the vectors within the range are
allocated according to the VFIO_DEVICE_SET_IRQS ioctl.
When dynamic MSI-X allocation is supported, we only want the guest
unmasked vectors being allocated and enabled. Use vector 0 with an
invalid fd to get MSI-X enabled, after that, all the vectors can be
allocated in need.
Signed-off-by: Jing Liu <jing2.liu@intel.com>
Reviewed-by: Cédric Le Goater <clg@redhat.com>
Reviewed-by: Alex Williamson <alex.williamson@redhat.com>
Signed-off-by: Cédric Le Goater <clg@redhat.com>
Guests typically enable MSI-X with all of the vectors masked in the MSI-X
vector table. To match the guest state of device, QEMU enables MSI-X by
enabling vector 0 with userspace triggering and immediately release.
However the release function actually does not release it due to already
using userspace mode.
It is no need to enable triggering on host and rely on the mask bit to
avoid spurious interrupts. Use an invalid fd (i.e. fd = -1) is enough
to get MSI-X enabled.
After dynamic MSI-X allocation is supported, the interrupt restoring
also need use such way to enable MSI-X, therefore, create a function
for that.
Suggested-by: Alex Williamson <alex.williamson@redhat.com>
Signed-off-by: Jing Liu <jing2.liu@intel.com>
Reviewed-by: Cédric Le Goater <clg@redhat.com>
Reviewed-by: Alex Williamson <alex.williamson@redhat.com>
Signed-off-by: Cédric Le Goater <clg@redhat.com>
The vector_use callback is used to enable vector that is unmasked in
guest. The kernel used to only support static MSI-X allocation. When
allocating a new interrupt using "static MSI-X allocation" kernels,
QEMU first disables all previously allocated vectors and then
re-allocates all including the new one. The nr_vectors of VFIOPCIDevice
indicates that all vectors from 0 to nr_vectors are allocated (and may
be enabled), which is used to loop all the possibly used vectors when
e.g., disabling MSI-X interrupts.
Extend the vector_use function to support dynamic MSI-X allocation when
host supports the capability. QEMU therefore can individually allocate
and enable a new interrupt without affecting others or causing interrupts
lost during runtime.
Utilize nr_vectors to calculate the upper bound of enabled vectors in
dynamic MSI-X allocation mode since looping all msix_entries_nr is not
efficient and unnecessary.
Signed-off-by: Jing Liu <jing2.liu@intel.com>
Tested-by: Reinette Chatre <reinette.chatre@intel.com>
Reviewed-by: Alex Williamson <alex.williamson@redhat.com>
Signed-off-by: Cédric Le Goater <clg@redhat.com>
Kernel provides the guidance of dynamic MSI-X allocation support of
passthrough device, by clearing the VFIO_IRQ_INFO_NORESIZE flag to
guide user space.
Fetch the flags from host to determine if dynamic MSI-X allocation is
supported.
Originally-by: Reinette Chatre <reinette.chatre@intel.com>
Signed-off-by: Jing Liu <jing2.liu@intel.com>
Reviewed-by: Cédric Le Goater <clg@redhat.com>
Reviewed-by: Alex Williamson <alex.williamson@redhat.com>
Signed-off-by: Cédric Le Goater <clg@redhat.com>
vfio_put_device() is a VFIO PCI specific function, rename it with
'vfio_pci' prefix to avoid confusing.
No functional change.
Suggested-by: Cédric Le Goater <clg@redhat.com>
Signed-off-by: Zhenzhong Duan <zhenzhong.duan@intel.com>
Signed-off-by: Cédric Le Goater <clg@redhat.com>
NVLink2 support was removed from the PPC PowerNV platform and VFIO in
Linux 5.13 with commits :
562d1e207d32 ("powerpc/powernv: remove the nvlink support")
b392a1989170 ("vfio/pci: remove vfio_pci_nvlink2")
This was 2.5 years ago. Do the same in QEMU with a revert of commit
ec132efaa8 ("spapr: Support NVIDIA V100 GPU with NVLink2"). Some
adjustements are required on the NUMA part.
Cc: Alexey Kardashevskiy <aik@ozlabs.ru>
Reviewed-by: Daniel Henrique Barboza <danielhb413@gmail.com>
Acked-by: Alex Williamson <alex.williamson@redhat.com>
Signed-off-by: Cédric Le Goater <clg@redhat.com>
Message-ID: <20230918091717.149950-1-clg@kaod.org>
Signed-off-by: Daniel Henrique Barboza <danielhb413@gmail.com>
Dynamically enable Atomic Ops completer support around realize/exit of
vfio-pci devices reporting host support for these accesses and adhering
to a minimal configuration standard. While the Atomic Ops completer
bits in the root port device capabilities2 register are read-only, the
PCIe spec does allow RO bits to change to reflect hardware state. We
take advantage of that here around the realize and exit functions of
the vfio-pci device.
Signed-off-by: Alex Williamson <alex.williamson@redhat.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Reviewed-by: Robin Voetter <robin@streamhpc.com>
Tested-by: Robin Voetter <robin@streamhpc.com>
Signed-off-by: Cédric Le Goater <clg@redhat.com>
vfio_realize() has the following flow:
1. vfio_bars_prepare() -- sets VFIOBAR->size.
2. msix_early_setup().
3. vfio_bars_register() -- allocates VFIOBAR->mr.
After vfio_bars_prepare() is called msix_early_setup() can fail. If it
does fail, vfio_bars_register() is never called and VFIOBAR->mr is not
allocated.
In this case, vfio_bars_finalize() is called as part of the error flow
to free the bars' resources. However, vfio_bars_finalize() calls
object_unparent() for VFIOBAR->mr after checking only VFIOBAR->size, and
thus we get a null pointer dereference.
Fix it by checking VFIOBAR->mr in vfio_bars_finalize().
Fixes: 89d5202edc ("vfio/pci: Allow relocating MSI-X MMIO")
Signed-off-by: Avihai Horon <avihaih@nvidia.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Reviewed-by: Cédric Le Goater <clg@redhat.com>
Reviewed-by: Alex Williamson <alex.williamson@redhat.com>
Signed-off-by: Cédric Le Goater <clg@redhat.com>
Make vfio_migration_realize() adhere to the convention of other realize()
callbacks(like qdev_realize) by returning bool instead of int.
Suggested-by: Cédric Le Goater <clg@redhat.com>
Suggested-by: Joao Martins <joao.m.martins@oracle.com>
Signed-off-by: Zhenzhong Duan <zhenzhong.duan@intel.com>
Reviewed-by: Joao Martins <joao.m.martins@oracle.com>
Reviewed-by: Cédric Le Goater <clg@redhat.com>
Signed-off-by: Cédric Le Goater <clg@redhat.com>
Property enable_migration supports [on/off/auto].
In ON mode, error pointer is passed to errp and logged.
In OFF mode, we doesn't need to log "Migration disabled" as it's intentional.
In AUTO mode, we should only ever see errors or warnings if the device
supports migration and an error or incompatibility occurs while further
probing or configuring it. Lack of support for migration shoundn't
generate an error or warning.
Signed-off-by: Zhenzhong Duan <zhenzhong.duan@intel.com>
Reviewed-by: Joao Martins <joao.m.martins@oracle.com>
Reviewed-by: Cédric Le Goater <clg@redhat.com>
Signed-off-by: Cédric Le Goater <clg@redhat.com>
When vfio_realize() succeeds, hot unplug will call vfio_exitfn()
to free resources allocated in vfio_realize(); when vfio_realize()
fails, vfio_exitfn() is never called and we need to free resources
in vfio_realize().
In the case that vfio_migration_realize() fails,
e.g: with -only-migratable & enable-migration=off, we see below:
(qemu) device_add vfio-pci,host=81:11.1,id=vfio1,bus=root1,enable-migration=off
0000:81:11.1: Migration disabled
Error: disallowing migration blocker (--only-migratable) for: 0000:81:11.1: Migration is disabled for VFIO device
If we hotplug again we should see same log as above, but we see:
(qemu) device_add vfio-pci,host=81:11.1,id=vfio1,bus=root1,enable-migration=off
Error: vfio 0000:81:11.1: device is already attached
That's because some references to VFIO device isn't released.
For resources allocated in vfio_migration_realize(), free them by
jumping to out_deinit path with calling a new function
vfio_migration_deinit(). For resources allocated in vfio_realize(),
free them by jumping to de-register path in vfio_realize().
Signed-off-by: Zhenzhong Duan <zhenzhong.duan@intel.com>
Fixes: a22651053b ("vfio: Make vfio-pci device migration capable")
Reviewed-by: Cédric Le Goater <clg@redhat.com>
Reviewed-by: Joao Martins <joao.m.martins@oracle.com>
Signed-off-by: Cédric Le Goater <clg@redhat.com>
Contrary to multiple device blocker which needs to consider already-attached
devices to unblock/block dynamically, the vIOMMU migration blocker is a device
specific config. Meaning it only needs to know whether the device is bypassing
or not the vIOMMU (via machine property, or per pxb-pcie::bypass_iommu), and
does not need the state of currently present devices. For this reason, the
vIOMMU global migration blocker can be consolidated into the per-device
migration blocker, allowing us to remove some unnecessary code.
This change also makes vfio_mig_active() more accurate as it doesn't check for
global blocker.
Signed-off-by: Zhenzhong Duan <zhenzhong.duan@intel.com>
Reviewed-by: Joao Martins <joao.m.martins@oracle.com>
Reviewed-by: Cédric Le Goater <clg@redhat.com>
Signed-off-by: Cédric Le Goater <clg@redhat.com>
When vfio realize fails, INTx isn't disabled if it has been enabled.
This may confuse host side with unhandled interrupt report.
Fixes: c5478fea27 ("vfio/pci: Respond to KVM irqchip change notifier")
Signed-off-by: Zhenzhong Duan <zhenzhong.duan@intel.com>
Reviewed-by: Joao Martins <joao.m.martins@oracle.com>
Reviewed-by: Cédric Le Goater <clg@redhat.com>
Signed-off-by: Cédric Le Goater <clg@redhat.com>
When vfio_realize fails, the mmap_timer used for INTx optimization
isn't freed. As this timer isn't activated yet, the potential impact
is just a piece of leaked memory.
Fixes: ea486926b0 ("vfio-pci: Update slow path INTx algorithm timer related")
Signed-off-by: Zhenzhong Duan <zhenzhong.duan@intel.com>
Reviewed-by: Cédric Le Goater <clg@redhat.com>
Reviewed-by: Joao Martins <joao.m.martins@oracle.com>
Signed-off-by: Cédric Le Goater <clg@redhat.com>
The kvm irqchip notifier is only registered if the device supports
INTx, however it's unconditionally removed in vfio realize error
path. If the assigned device does not support INTx, this will cause
QEMU to crash when vfio realize fails. Change it to conditionally
remove the notifier only if the notify hook is setup.
Before fix:
(qemu) device_add vfio-pci,host=81:11.1,id=vfio1,bus=root1,xres=1
Connection closed by foreign host.
After fix:
(qemu) device_add vfio-pci,host=81:11.1,id=vfio1,bus=root1,xres=1
Error: vfio 0000:81:11.1: xres and yres properties require display=on
(qemu)
Fixes: c5478fea27 ("vfio/pci: Respond to KVM irqchip change notifier")
Signed-off-by: Zhenzhong Duan <zhenzhong.duan@intel.com>
Reviewed-by: Cédric Le Goater <clg@redhat.com>
Reviewed-by: Joao Martins <joao.m.martins@oracle.com>
Signed-off-by: Cédric Le Goater <clg@redhat.com>
The major parts of VFIO migration are supported today in QEMU. This
includes basic VFIO migration, device dirty page tracking and precopy
support.
Thus, at this point in time, it seems appropriate to make VFIO migration
non-experimental: remove the x prefix from enable_migration property,
change it to ON_OFF_AUTO and let the default value be AUTO.
In addition, make the following adjustments:
1. When enable_migration is ON and migration is not supported, fail VFIO
device realization.
2. When enable_migration is AUTO (i.e., not explicitly enabled), require
device dirty tracking support. This is because device dirty tracking
is currently the only method to do dirty page tracking, which is
essential for migrating in a reasonable downtime. Setting
enable_migration to ON will not require device dirty tracking.
3. Make migration error and blocker messages more elaborate.
4. Remove error prints in vfio_migration_query_flags().
5. Rename trace_vfio_migration_probe() to
trace_vfio_migration_realize().
Signed-off-by: Avihai Horon <avihaih@nvidia.com>
Reviewed-by: Joao Martins <joao.m.martins@oracle.com>
Reviewed-by: Cédric Le Goater <clg@redhat.com>
Reviewed-by: Alex Williamson <alex.williamson@redhat.com>
Signed-off-by: Cédric Le Goater <clg@redhat.com>
When vfio_enable_vectors() returns with less than requested nr_vectors
we retry with what kernel reported back. But the retry path doesn't
call vfio_prepare_kvm_msi_virq_batch() and this results in,
qemu-system-aarch64: vfio: Error: Failed to enable 4 MSI vectors, retry with 1
qemu-system-aarch64: ../hw/vfio/pci.c:602: vfio_commit_kvm_msi_virq_batch: Assertion `vdev->defer_kvm_irq_routing' failed
Fixes: dc580d51f7 ("vfio: defer to commit kvm irq routing when enable msi/msix")
Reviewed-by: Longpeng <longpeng2@huawei.com>
Signed-off-by: Shameer Kolothum <shameerali.kolothum.thodi@huawei.com>
Reviewed-by: Cédric Le Goater <clg@redhat.com>
Signed-off-by: Cédric Le Goater <clg@redhat.com>
vbasedev->name is freed wrongly which leads to garbage VFIO trace log.
Fix it by allocating a dup of vbasedev->name and then free the dup.
Fixes: 2dca1b37a7 ("vfio/pci: add support for VF token")
Suggested-by: Alex Williamson <alex.williamson@redhat.com>
Signed-off-by: Zhenzhong Duan <zhenzhong.duan@intel.com>
Reviewed-by: Cédric Le Goater <clg@redhat.com>
Reviewed-by: Matthew Rosato <mjrosato@linux.ibm.com>
Acked-by: Alex Williamson <alex.williamson@redhat.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Signed-off-by: Cédric Le Goater <clg@redhat.com>
The PCI Resizable BAR (ReBAR) capability is currently hidden from the
VM because the protocol for interacting with the capability does not
support a mechanism for the device to reject an advertised supported
BAR size. However, when assigned to a VM, the act of resizing the
BAR requires adjustment of host resources for the device, which
absolutely can fail. Linux does not currently allow us to reserve
resources for the device independent of the current usage.
The only writable field within the ReBAR capability is the BAR Size
register. The PCIe spec indicates that when written, the device
should immediately begin to operate with the provided BAR size. The
spec however also notes that software must only write values
corresponding to supported sizes as indicated in the capability and
control registers. Writing unsupported sizes produces undefined
results. Therefore, if the hypervisor were to virtualize the
capability and control registers such that the current size is the
only indicated available size, then a write of anything other than
the current size falls into the category of undefined behavior,
where we can essentially expose the modified ReBAR capability as
read-only.
This may seem pointless, but users have reported that virtualizing
the capability in this way not only allows guest software to expose
related features as available (even if only cosmetic), but in some
scenarios can resolve guest driver issues. Additionally, no
regressions in behavior have been reported for this change.
A caveat here is that the PCIe spec requires for compatibility that
devices report support for a size in the range of 1MB to 512GB,
therefore if the current BAR size falls outside that range we revert
to hiding the capability.
Reviewed-by: Cédric Le Goater <clg@redhat.com>
Link: https://lore.kernel.org/r/20230505232308.2869912-1-alex.williamson@redhat.com
Signed-off-by: Alex Williamson <alex.williamson@redhat.com>
Pick names that align with the section drivers should use them from,
avoiding the confusion of calling a _finalize() function from _exit()
and generalizing the actual _finalize() to handle removing the viommu
blocker.
Reviewed-by: Cédric Le Goater <clg@redhat.com>
Reviewed-by: Joao Martins <joao.m.martins@oracle.com>
Link: https://lore.kernel.org/r/167820912978.606734.12740287349119694623.stgit@omen
Signed-off-by: Alex Williamson <alex.williamson@redhat.com>
Migrating with vIOMMU will require either tracking maximum
IOMMU supported address space (e.g. 39/48 address width on Intel)
or range-track current mappings and dirty track the new ones
post starting dirty tracking. This will be done as a separate
series, so add a live migration blocker until that is fixed.
Signed-off-by: Joao Martins <joao.m.martins@oracle.com>
Reviewed-by: Cédric Le Goater <clg@redhat.com>
Link: https://lore.kernel.org/r/20230307125450.62409-14-joao.m.martins@oracle.com
Signed-off-by: Alex Williamson <alex.williamson@redhat.com>
Using a VFIODevice handle local variable to improve the code readability.
no functional change intended
Signed-off-by: Eric Auger <eric.auger@redhat.com>
Signed-off-by: Yi Liu <yi.l.liu@intel.com>
Link: https://lore.kernel.org/r/20220502094223.36384-3-yi.l.liu@intel.com
Signed-off-by: Alex Williamson <alex.williamson@redhat.com>
"%m" format specifier is not interpreted by the trace infrastructure
and thus "%m" is output instead of the actual errno string. Fix it by
outputting strerror(errno).
Signed-off-by: Eric Auger <eric.auger@redhat.com>
Signed-off-by: Yi Liu <yi.l.liu@intel.com>
Link: https://lore.kernel.org/r/20220502094223.36384-2-yi.l.liu@intel.com
[aw: replace commit log as provided by Eric]
Signed-off-by: Alex Williamson <alex.williamson@redhat.com>
In migration resume phase, all unmasked msix vectors need to be
setup when loading the VF state. However, the setup operation would
take longer if the VM has more VFs and each VF has more unmasked
vectors.
The hot spot is kvm_irqchip_commit_routes, it'll scan and update
all irqfds that are already assigned each invocation, so more
vectors means need more time to process them.
vfio_pci_load_config
vfio_msix_enable
msix_set_vector_notifiers
for (vector = 0; vector < dev->msix_entries_nr; vector++) {
vfio_msix_vector_do_use
vfio_add_kvm_msi_virq
kvm_irqchip_commit_routes <-- expensive
}
We can reduce the cost by only committing once outside the loop.
The routes are cached in kvm_state, we commit them first and then
bind irqfd for each vector.
The test VM has 128 vcpus and 8 VF (each one has 65 vectors),
we measure the cost of the vfio_msix_enable for each VF, and
we can see 90+% costs can be reduce.
VF Count of irqfds[*] Original With this patch
1st 65 8 2
2nd 130 15 2
3rd 195 22 2
4th 260 24 3
5th 325 36 2
6th 390 44 3
7th 455 51 3
8th 520 58 4
Total 258ms 21ms
[*] Count of irqfds
How many irqfds that already assigned and need to process in this
round.
The optimization can be applied to msi type too.
Signed-off-by: Longpeng(Mike) <longpeng2@huawei.com>
Link: https://lore.kernel.org/r/20220326060226.1892-6-longpeng2@huawei.com
Signed-off-by: Alex Williamson <alex.williamson@redhat.com>
Commit ecebe53fe9 ("vfio: Avoid disabling and enabling vectors
repeatedly in VFIO migration") avoids inefficiently disabling and
enabling vectors repeatedly and lets the unmasked vectors be enabled
one by one.
But we want to batch multiple routes and defer the commit, and only
commit once outside the loop of setting vector notifiers, so we
cannot enable the vectors one by one in the loop now.
Revert that commit and we will take another way in the next patch,
it can not only avoid disabling/enabling vectors repeatedly, but
also satisfy our requirement of defer to commit.
Signed-off-by: Longpeng(Mike) <longpeng2@huawei.com>
Link: https://lore.kernel.org/r/20220326060226.1892-5-longpeng2@huawei.com
Signed-off-by: Alex Williamson <alex.williamson@redhat.com>
It's unnecessary to test against the specific return value of
VFIO_DEVICE_SET_IRQS, since any positive return is an error
indicating the number of vectors we should retry with.
Signed-off-by: Longpeng(Mike) <longpeng2@huawei.com>
Link: https://lore.kernel.org/r/20220326060226.1892-2-longpeng2@huawei.com
Signed-off-by: Alex Williamson <alex.williamson@redhat.com>
Replace the global variables with inlined helper functions. getpagesize() is very
likely annotated with a "const" function attribute (at least with glibc), and thus
optimization should apply even better.
This avoids the need for a constructor initialization too.
Signed-off-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Message-Id: <20220323155743.1585078-12-marcandre.lureau@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
g_new(T, n) is neater than g_malloc(sizeof(T) * n). It's also safer,
for two reasons. One, it catches multiplication overflowing size_t.
Two, it returns T * rather than void *, which lets the compiler catch
more type errors.
This commit only touches allocations with size arguments of the form
sizeof(T).
Patch created mechanically with:
$ spatch --in-place --sp-file scripts/coccinelle/use-g_new-etc.cocci \
--macro-file scripts/cocci-macro-file.h FILES...
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Reviewed-by: Cédric Le Goater <clg@kaod.org>
Reviewed-by: Alex Bennée <alex.bennee@linaro.org>
Acked-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Message-Id: <20220315144156.1595462-4-armbru@redhat.com>
Reviewed-by: Pavel Dovgalyuk <Pavel.Dovgalyuk@ispras.ru>
We invoke the kvm_irqchip_commit_routes() for each addition to MSI route
table, which is not efficient if we are adding lots of routes in some cases.
This patch lets callers invoke the kvm_irqchip_commit_routes(), so the
callers can decide how to optimize.
[1] https://lists.gnu.org/archive/html/qemu-devel/2021-11/msg00967.html
Signed-off-by: Longpeng <longpeng2@huawei.com>
Message-Id: <20220222141116.2091-3-longpeng2@huawei.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
We can expand MemoryRegions of sub-page MMIO BARs in
vfio_pci_write_config() to improve IO performance for some
devices. However, the MemoryRegions of destination VM are
not expanded any more after live migration. Because their
addresses have been updated in vmstate_load_state()
(vfio_pci_load_config) and vfio_sub_page_bar_update_mapping()
will not be called.
This may result in poor performance after live migration.
So iterate BARs in vfio_pci_load_config() and try to update
sub-page BARs.
Reported-by: Nianyao Tang <tangnianyao@huawei.com>
Reported-by: Qixin Gan <ganqixin@huawei.com>
Signed-off-by: Kunkun Jiang <jiangkunkun@huawei.com>
Link: https://lore.kernel.org/r/20211027090406.761-2-jiangkunkun@huawei.com
Signed-off-by: Alex Williamson <alex.williamson@redhat.com>
QDicts are both what QMP natively uses and what the keyval parser
produces. Going through QemuOpts isn't useful for either one, so switch
the main device creation function to QDicts. By sharing more code with
the -object/object-add code path, we can even reduce the code size a
bit.
This commit doesn't remove the detour through QemuOpts from any code
path yet, but it allows the following commits to do so.
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Message-Id: <20211008133442.141332-15-kwolf@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Tested-by: Peter Krempa <pkrempa@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
