With virtual machines becoming larger (more CPUs, more memory) the
memory needed by the SMM code in OVMF to manage page tables and vcpu
state grows too.
Default SMM memory (aka TSEG) size is 16 MB, and this often is not
enough. Bump it to 64 MB for new machine types.
Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Message-Id: <20251106105640.1642109-1-kraxel@redhat.com>
Kernel allows user to switch IOMMU domain, e.g., switch between DMA
and identity domain. When this happen in IOMMU scalable mode, a pasid
cache invalidation request is sent, this request is ignored by vIOMMU
which leads to device binding to wrong address space, then DMA fails.
This issue exists in scalable mode with both first stage and second
stage translations, both emulated and passthrough devices.
Take network device for example, below sequence trigger issue:
1. start a guest with iommu=pt
2. echo 0000:01:00.0 > /sys/bus/pci/drivers/virtio-pci/unbind
3. echo DMA > /sys/kernel/iommu_groups/6/type
4. echo 0000:01:00.0 > /sys/bus/pci/drivers/virtio-pci/bind
5. Ping test
Fix it by switching address space in invalidation handler.
Fixes: 4a4f219e8a ("intel_iommu: add scalable-mode option to make scalable mode work")
Signed-off-by: Zhenzhong Duan <zhenzhong.duan@intel.com>
Reviewed-by: Yi Liu <yi.l.liu@intel.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Message-Id: <20251017093602.525338-4-zhenzhong.duan@intel.com>
Adds a new entry VTDPASIDCacheEntry in VTDAddressSpace to cache the pasid
entry and track PASID usage and future PASID tagged DMA address translation
support in vIOMMU.
When guest triggers pasid cache invalidation, QEMU will capture it and
update or invalidate pasid cache.
vIOMMU emulator could figure out the reason by fetching latest guest pasid
entry in memory and compare it with cached PASID entry if it's valid.
Signed-off-by: Yi Liu <yi.l.liu@intel.com>
Signed-off-by: Zhenzhong Duan <zhenzhong.duan@intel.com>
Reviewed-by: Yi Liu <yi.l.liu@intel.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Message-Id: <20251017093602.525338-2-zhenzhong.duan@intel.com>
The physical AMD IOMMU supports up to 64 bits of IOVA. When a device tries
to read or write from a given DMA address, the IOMMU translates the address
using the I/O page tables assigned to that device. Since the emulated IOMMU
uses per-device page tables, an ideal cache tag would need to be 68 bits
(64-bit address - 12-bit page alignment + 16-bit device ID).
The current software IOTLB implementation uses a GLib hash table with a
64-bit key to hash both the IOVA and device ID, which limits the IOVA to 60
bits. This causes a failure while setting up the device when a guest is
booted with "iommu.forcedac=1", which forces the use of DMA addresses at the
top of the 64-bit address space.
To address this issue, construct the 64-bit hash key using the upper 52 bits
of IOVA (GFN) and lower 12 bits of the device ID to avoid truncation as much
as possible (reducing hash collisions).
Fixes: d29a09ca68 ("hw/i386: Introduce AMD IOMMU")
Signed-off-by: Sairaj Kodilkar <sarunkod@amd.com>
Reviewed-by: Vasant Hegde <vasant.hegde@amd.com>
Reviewed-by: Alejandro Jimenez <alejandro.j.jimenez@oracle.com>
Tested-by: Alejandro Jimenez <alejandro.j.jimenez@oracle.com>
Signed-off-by: Alejandro Jimenez <alejandro.j.jimenez@oracle.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Message-Id: <20251103203209.645434-4-alejandro.j.jimenez@oracle.com>
The AMD IOMMU is set up at boot time and uses PCI bus numbers + devfn for
indexing into the DTE. The problem is that before the guest starts, all PCI
bus numbers are 0 as no PCI discovery has happened yet (BIOS and/or kernel
will do that later), so relying on the bus number is wrong.
The immediate effect is that emulated devices cannot do DMA when placed on a
bus other than 0.
Replace the static address_space array with a hash table keyed by devfn and
PCIBus*, since these values do not change after the guest boots.
Co-developed-by: Alexey Kardashevskiy <aik@amd.com>
Signed-off-by: Alexey Kardashevskiy <aik@amd.com>
Signed-off-by: Sairaj Kodilkar <sarunkod@amd.com>
Reviewed-by: Vasant Hegde <vasant.hegde@amd.com>
Reviewed-by: Alejandro Jimenez <alejandro.j.jimenez@oracle.com>
Tested-by: Alejandro Jimenez <alejandro.j.jimenez@oracle.com>
Signed-off-by: Alejandro Jimenez <alejandro.j.jimenez@oracle.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Message-Id: <20251103203209.645434-3-alejandro.j.jimenez@oracle.com>
MachineState already provides an fdt field, remove it from
MicrovmMachineState and use that instead.
This is useful when using the `dumpdtb` option, as QEMU expects the
device tree to be stored in the MachineState struct, otherwise it
will return this error:
qemu-system-x86_64: This machine doesn't have an FDT
Signed-off-by: Luigi Leonardi <leonardi@redhat.com>
Reviewed-by: Stefano Garzarella <sgarzare@redhat.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Message-ID: <20251030-fix_microvm-v1-1-f89544a04be3@redhat.com>
Signed-off-by: Philippe Mathieu-Daudé <philmd@linaro.org>
* char: rename CharBackend->CharFrontend
* esp: fix esp_cdb_ready() FIFO wraparound limit calculation
* isapc: warn rather than reject modern x86 CPU models
* mshv: fix Coverity issues
* qdev: Change PropertyInfo method print() to return malloc'ed string
* qobject: make refcount atomic
* rcu: make synchronize_rcu() more efficient
* rust: cleanup glib_sys bindings
* rust: Convert bit value to u8 within #[property]
* rust: only leave leaf crates as workspace members
* scripts: clean up meson-buildoptions.py
* scsi: make refcount atomic
* target/i386: Init SMM cpu address space for hotplugged CPUs
# -----BEGIN PGP SIGNATURE-----
#
# iQFIBAABCgAyFiEE8TM4V0tmI4mGbHaCv/vSX3jHroMFAmkB3UUUHHBib256aW5p
# QHJlZGhhdC5jb20ACgkQv/vSX3jHroMrRAgAidqHD3mBOEPhcz3Xh5xuJGd+fFnr
# wD5/zi5DPy2ZmPKY/buMv/92R6GKKQwZVJrKKdj2+yXFayp0LZNs+ZXNhOl8+EWT
# uZayJJt1Wx5E8BB31NKDBLSnvIjVnP+0QDN3pn7ihoIYtdy7ziUd2sS955z+42Vx
# ewLwZMfIodSykYfeUsjA7A4GgXtHZWv1bGOZ4qVX/Bgr06efKui3WUaVAkN6i6T+
# WJBn4Wb3fBiHLhmFzdIxB91hZjyP9athOQb5kiTL1g+2uYsuQPSfWRhqfgy3EWGf
# K72bpnxnMSzhjz0YXKz5S5kjpe/3g7OCiPJm0Jf5Aq7KqGyIkUSSdgo59Q==
# =KCqH
# -----END PGP SIGNATURE-----
# gpg: Signature made Wed 29 Oct 2025 10:24:21 AM CET
# gpg: using RSA key F13338574B662389866C7682BFFBD25F78C7AE83
# gpg: issuer "pbonzini@redhat.com"
# gpg: Good signature from "Paolo Bonzini <bonzini@gnu.org>" [unknown]
# gpg: aka "Paolo Bonzini <pbonzini@redhat.com>" [unknown]
# gpg: WARNING: The key's User ID is not certified with a trusted signature!
# gpg: There is no indication that the signature belongs to the owner.
# Primary key fingerprint: 46F5 9FBD 57D6 12E7 BFD4 E2F7 7E15 100C CD36 69B1
# Subkey fingerprint: F133 3857 4B66 2389 866C 7682 BFFB D25F 78C7 AE83
* tag 'for-upstream' of https://gitlab.com/bonzini/qemu:
rust: migration: allow passing ParentField<> to vmstate_of!
target/i386: clear CPU_INTERRUPT_SIPI for all accelerators
docs/about/deprecated.rst: document isapc deprecation for modern x86 CPU models
hw/i386/isapc.c: warn rather than reject modern x86 CPU models
qdev: Change PropertyInfo method print() to return malloc'ed string
scsi: make SCSIRequest refcount atomic
rust/qemu-macros: Convert bit value to u8 within #[property]
qtest/am53c974-test: add additional test for cmdfifo overflow
esp.c: fix esp_cdb_ready() FIFO wraparound limit calculation
accel/mshv: use return value of handle_pio_str_read
accel/mshv: initialize thread name
char: rename CharBackend->CharFrontend
qobject: make refcount atomic
rust: only leave leaf crates as workspace members
rust: remove useless glib_sys bindings
rcu: Unify force quiescent state
i386/kvm/cpu: Init SMM cpu address space for hotplugged CPUs
scripts: clean up meson-buildoptions.py
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Commit e1e2909f8e ("hw/i386/pc_piix.c: restrict isapc machine to 32-bit CPUs")
restricted the isapc machine to 32-bit CPUs, but subsequent concern has been
expressed as to the effect this could have on users.
The outcome of the latest discussion is that we should exercise more caution
and follow the official deprecation route, so instead of rejecting modern x86
CPUs issue a deprecation warning but allow the user to continue.
Signed-off-by: Mark Cave-Ayland <mark.caveayland@nutanix.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Link: https://lore.kernel.org/r/20251023142926.964718-2-mark.caveayland@nutanix.com
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Makes the APIC API more type-safe by resolving quite a few APIC_COMMON
downcasts.
Like PICCommonState, the APICCommonState is now a public typedef while staying
an abstract datatype.
Signed-off-by: Bernhard Beschow <shentey@gmail.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Message-ID: <20251019210303.104718-8-shentey@gmail.com>
Signed-off-by: Philippe Mathieu-Daudé <philmd@linaro.org>
The pdcm feature is supposed to be disabled when PMU is not
available. Up until v10.1, pdcm feature is enabled even when PMU
is off. This behavior has been fixed but this change breaks the
migration of VMs that are run with QEMU < 10.0 and expect the pdcm
feature to be enabled on the destination host.
This commit restores the legacy behavior for machines with version
prior to 10.1 to allow the migration from older QEMU to QEMU 10.1.
Signed-off-by: Hector Cao <hector.cao@canonical.com>
Link: https://lore.kernel.org/r/20250910115733.21149-3-hector.cao@canonical.com
Fixes: e68ec29809 ("i386/cpu: Move adjustment of CPUID_EXT_PDCM before feature_dependencies[] check", 2025-06-20)
[Move property from migration object to CPU. - Paolo]
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Prior to v10.1, if requested by user, arch-capabilities is always on
despite the fact that CPUID advertises it to be off/unvailable.
This causes a migration issue for VMs that are run on a machine
without arch-capabilities and expect this feature to be present
on the destination host with QEMU 10.1.
Add a compatibility property to restore the legacy behavior for all
machines with version prior to 10.1.
To preserve the functionality (added by 10.1) of turning off
ARCH_CAPABILITIES where Windows does not like it, use directly
the guest CPU vendor: x86_cpu_get_supported_feature_word is not
KVM-specific and therefore should not necessarily use the host
CPUID.
Co-authored-by: Hector Cao <hector.cao@canonical.com>
Signed-off-by: Hector Cao <hector.cao@canonical.com>
Fixes: d3a24134e3 ("target/i386: do not expose ARCH_CAPABILITIES on AMD CPU", 2025-07-17)
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
In early days, we had different tricks to ensure caching-mode=on with VFIO
device:
28cf553afe ("intel_iommu: Sanity check vfio-pci config on machine init done")
c6cbc29d36 ("pc/q35: Disallow vfio-pci hotplug without VT-d caching mode")
There is also a patch with the same purpose but for VDPA device:
b8d78277c0 ("intel-iommu: fail MAP notifier without caching mode")
Because without caching mode, MAP notifier won't work correctly since guest
won't send IOTLB update event when it establishes new mappings in the I/O page
tables.
Now with host IOMMU device interface between VFIO and vIOMMU, we can simplify
first two commits above with a small check in set_iommu_device(). This also
works for future IOMMUFD backed VDPA implementation which may also need caching
mode on. But for legacy VDPA we still need commit b8d78277c0 as it doesn't
use the host IOMMU device interface.
For coldplug VFIO device:
qemu-system-x86_64: -device vfio-pci,host=0000:3b:00.0,id=hostdev3,bus=root0,iommufd=iommufd0: vfio 0000:3b:00.0: Failed to set vIOMMU: Device assignment is not allowed without enabling caching-mode=on for Intel IOMMU.
For hotplug VFIO device:
if "iommu=off" is configured in guest,
Error: vfio 0000:3b:00.0: Failed to set vIOMMU: Device assignment is not allowed without enabling caching-mode=on for Intel IOMMU.
else
Error: vfio 0000:3b:00.0: memory listener initialization failed: Region vtd-00.0-dmar: device 01.00.0 requires caching mode: Operation not supported
The specialty for hotplug is due to the check in commit b8d78277c0 happen before
the check in set_iommu_device.
Signed-off-by: Zhenzhong Duan <zhenzhong.duan@intel.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Message-ID: <20250929034206.439266-3-zhenzhong.duan@intel.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
According to VTD spec rev 4.1 section 6.6:
"For implementations reporting the Enhanced Set Root Table Pointer Support
(ESRTPS) field as Clear, on a 'Set Root Table Pointer' operation, software
must perform a global invalidate of the context cache, PASID-cache (if
applicable), and IOTLB, in that order. This is required to ensure hardware
references only the remapping structures referenced by the new root table
pointer and not stale cached entries.
For implementations reporting the Enhanced Set Root Table Pointer Support
(ESRTPS) field as Set, as part of 'Set Root Table Pointer' operation,
hardware performs global invalidation on all DMA remapping translation
caches and hence software is not required to perform additional
invalidations"
We already implemented ESRTPS capability in vtd_handle_gcmd_srtp() by
calling vtd_reset_caches(), just set ESRTPS in DMAR_CAP_REG to avoid
unnecessary global invalidation requests of context, PASID-cache and
IOTLB from guest.
This change doesn't impact migration as the content of DMAR_CAP_REG is
migrated too.
Signed-off-by: Zhenzhong Duan <zhenzhong.duan@intel.com>
Reviewed-by: Clément Mathieu--Drif <clement.mathieu--drif@eviden.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Message-ID: <20250929034206.439266-2-zhenzhong.duan@intel.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
The AMD I/O Virtualization Technology (IOMMU) Specification (see Table
8: V, TV, and GV Fields in Device Table Entry), specifies that a DTE
with V=1, TV=0 does not contain a valid address translation information.
If a request requires a table walk, the walk is terminated when this
condition is encountered.
Do not assume that addresses for a device with DTE[TV]=0 are passed
through (i.e. not remapped) and instead terminate the page table walk
early.
Fixes: d29a09ca68 ("hw/i386: Introduce AMD IOMMU")
Signed-off-by: Alejandro Jimenez <alejandro.j.jimenez@oracle.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Message-ID: <20250919213515.917111-20-alejandro.j.jimenez@oracle.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
A guest must issue an INVALIDATE_DEVTAB_ENTRY command after changing a
Device Table entry (DTE) e.g. after attaching a device and setting up its
DTE. When intercepting this event, determine if the DTE has been configured
for paging or not, and toggle the appropriate memory regions to allow DMA
address translation for the address space if needed. Requires dma-remap=on.
Signed-off-by: Alejandro Jimenez <alejandro.j.jimenez@oracle.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Message-ID: <20250919213515.917111-19-alejandro.j.jimenez@oracle.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
In order to enable device assignment with IOMMU protection and guest DMA
address translation, IOMMU MAP notifier support is necessary to allow users
like VFIO to synchronize the shadow page tables i.e. to receive
notifications when the guest updates its I/O page tables and replay the
mappings onto host I/O page tables.
Provide a new dma-remap property to govern the ability to register for MAP
notifications, effectively providing global control over the DMA address
translation functionality that was implemented in previous changes.
Note that DMA remapping support also requires the vIOMMU is configured with
the NpCache capability, so a guest driver issues IOMMU invalidations for
both map() and unmap() operations. This capability is already set by default
and written to the configuration in amdvi_pci_realize() as part of
AMDVI_CAPAB_FEATURES.
Signed-off-by: Alejandro Jimenez <alejandro.j.jimenez@oracle.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Message-ID: <20250919213515.917111-18-alejandro.j.jimenez@oracle.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
