[PATCH v6 07/27] vfio/pci: Preserve vfio-pci device files across Live Update

Samiullah Khawaja skhawaja at google.com
Fri Nov 6 14:35:27 PST 2026


On Mon, Nov 02, 2026 at 07:36:36AM -0800, Vipin Sharma wrote:
>Implement the Live Update file handler callbacks to preserve a vfio-pci
>device in noiommu mode across a Live Update while keeping it running
>across kexec. Subsequent commits will enable userspace to retrieve and
>take over this file after the Live Update.
>
>State about each device is serialized into struct vfio_pci_core_device_ser
>using Kexec-Handover (KHO) to preserve the pages holding the struct and
>the Live Update Orchestrator (LUO) to preserve the physical address of
>the struct.
>
>Key design notes and constraints:
>
> * Noiommu-only policy and bound cdev requirement: can_preserve() verifies
>   that the preserved file is a bound vfio-pci cdev file descriptor via
>   vfio_file_is_bound_cdev(), and preserve() requires the device to be in
>   noiommu mode via vfio_device_is_noiommu() (returning -EOPNOTSUPP
>   otherwise). CONFIG_VFIO_PCI_LIVEUPDATE now depends on
>   CONFIG_VFIO_DEVICE_CDEV and CONFIG_VFIO_NOIOMMU.
>
> * Serialized state: During preserve(), serialize the PCI domain, BDF,
>   config space size, reset_works flag, and the initial pci_saved_state
>   captured when the device was first opened. During freeze(), snapshot
>   the virtualized PCI config space (vconfig) and capability map
>   (pci_config_map).
>
> * Keep-running freeze and unfreeze: Do not reset the device, clear
>   PCI_COMMAND_MASTER, zap BAR mappings, or revoke dma-bufs during
>   freeze(). Validate that VFIO interrupts are disabled, that the device
>   is in PCI_D0 (and not in VFIO runtime PM low-power mode), and that
>   SR-IOV is not enabled, as VFs are not preserved and the next kernel
>   disables SR-IOV on the running device. Then mask physical INTx
>   (recording the prior state) and mark the device frozen. If the
>   Live Update is aborted after freeze(), unfreeze() restores the physical
>   INTx state and clears liveupdate_frozen so the existing FD remains
>   usable.
>
> * TOCTOU synchronization: vfio_pci_dev_is_frozen() is checked locklessly
>   on entry to the ioctl, read/write and mmap paths, and re-checked under
>   a lock that freeze() holds before any operation that could change the
>   device or vconfig: under vdev->igate in vfio_pci_ioctl_set_irqs() and
>   around every config-space write in vfio_pci_config_rw_single(), under
>   vdev->memory_lock in vfio_pci_runtime_pm_entry(),
>   vfio_pci_ioctl_reset() and the SR-IOV enable path of
>   vfio_pci_core_sriov_configure(), and under dev_set->lock in
>   vfio_pci_dev_set_hot_reset(). A racing operation thus either completes
>   before freeze() takes its snapshot or fails, while existing BAR mmaps,
>   ioeventfds and dma-bufs keep working. Serializing config writes on
>   igate covers the PCI_COMMAND, INTx disable emulation, power state and
>   FLR paths at one point (the INTx disable emulation now uses the
>   igate-held __vfio_pci_intx_mask() and __vfio_pci_intx_unmask()
>   helpers), and also serializes the irq_type and virq_disabled accesses
>   of config writes with VFIO_DEVICE_SET_IRQS.
>
>Co-developed-by: David Matlack <dmatlack at google.com>
>Signed-off-by: David Matlack <dmatlack at google.com>
>Signed-off-by: Vipin Sharma <vipinsh at google.com>
>---
> drivers/vfio/pci/Kconfig               |   2 +-
> drivers/vfio/pci/vfio_pci_config.c     |  19 +-
> drivers/vfio/pci/vfio_pci_core.c       |  49 +++-
> drivers/vfio/pci/vfio_pci_intrs.c      |   4 +-
> drivers/vfio/pci/vfio_pci_liveupdate.c | 303 ++++++++++++++++++++++++-
> drivers/vfio/pci/vfio_pci_priv.h       |  13 ++
> include/linux/kho/abi/vfio_pci.h       |  52 ++++-
> include/linux/vfio_pci_core.h          |   4 +
> 8 files changed, 434 insertions(+), 12 deletions(-)
>

[snip]

>diff --git a/include/linux/kho/abi/vfio_pci.h b/include/linux/kho/abi/vfio_pci.h
>index 67e8199eec59..725043a44b55 100644
>--- a/include/linux/kho/abi/vfio_pci.h
>+++ b/include/linux/kho/abi/vfio_pci.h
>@@ -9,14 +9,19 @@
> #ifndef _LINUX_LIVEUPDATE_ABI_VFIO_PCI_H
> #define _LINUX_LIVEUPDATE_ABI_VFIO_PCI_H
>
>+#include <linux/bits.h>
>+#include <linux/compiler.h>
>+#include <linux/types.h>
>+
> /**
>  * DOC: VFIO PCI Live Update ABI
>  *
>  * VFIO uses the ABI defined below for preserving device files across a kexec
>  * reboot using LUO.
>  *
>- * Device metadata is serialized into memory which is then handed to the next
>- * kernel via KHO.
>+ * Device metadata, virtualized PCI configuration space, VFIO's capability map,
>+ * and the initial PCI saved state captured when the device was first opened are
>+ * serialized into memory which is then handed to the next kernel via KHO.
>  *
>  * This interface is a contract. Any modification to any of the serialization
>  * structs defined here constitutes a breaking change. Such changes require
>@@ -25,4 +30,47 @@
>
> #define VFIO_PCI_LUO_FH_COMPATIBLE "vfio-pci-v1"
>
>+/**
>+ * define VFIO_PCI_SER_RESET_WORKS - Device supports function reset.
>+ *
>+ * Set in &vfio_pci_core_device_ser.flags when the preserved device had
>+ * ``vdev->reset_works == true`` prior to kexec.
>+ */
>+#define VFIO_PCI_SER_RESET_WORKS	BIT(0)
>+
>+#define VFIO_PCI_SER_KNOWN_FLAGS	VFIO_PCI_SER_RESET_WORKS
>+
>+/**
>+ * struct vfio_pci_core_device_ser - Serialized state of a single VFIO PCI
>+ * device.
>+ *
>+ * @domain: The device's PCI domain number (segment).
>+ * @bdf: The device's PCI bus, device, and function number.
>+ * @cfg_size: Size in bytes of the device's PCI configuration space
>+ *            (``PCI_CFG_SPACE_SIZE`` or ``PCI_CFG_SPACE_EXP_SIZE``).
>+ * @size: Total size in bytes of this serialized buffer, equal to
>+ *        ``sizeof(struct vfio_pci_core_device_ser) + 2 * @cfg_size + @saved_state_size``.
>+ * @flags: Bitmask of ``VFIO_PCI_SER_*`` flags; unknown bits are rejected.
>+ * @saved_state_size: Size in bytes of the serialized &struct pci_saved_state_ser
>+ *                    blob stored in @data after ``config_map``, or 0 if none.
>+ * @reserved: Reserved for future use; must be 0.
>+ * @data: Variable-length payload containing, in order:
>+ *        1. ``u8 vconfig[@cfg_size]`` - VFIO virtualized config space snapshot
>+ *           captured during ``freeze()``.
>+ *        2. ``u8 config_map[@cfg_size]`` - VFIO capability map snapshot
>+ *           captured during ``freeze()``.
>+ *        3. ``u8 saved_state[@saved_state_size]`` - serialized
>+ *           &struct pci_saved_state_ser captured during ``preserve()``.
>+ */
>+struct vfio_pci_core_device_ser {
>+	u32 domain;
>+	u16 bdf;
>+	u16 cfg_size;
>+	u32 size;
>+	u32 flags;
>+	u32 saved_state_size;

We will need space for iommufd_token when we preserve that, we should
probably add it now as otherwise the compatibility would be broken when
we add it? Is the reserved here for that?
>+	u32 reserved;
>+	u8 data[];
>+} __packed;
>+
> #endif /* _LINUX_LIVEUPDATE_ABI_VFIO_PCI_H */
>diff --git a/include/linux/vfio_pci_core.h b/include/linux/vfio_pci_core.h
>index 9a1674c152aa..6bae7581f4f4 100644
>--- a/include/linux/vfio_pci_core.h
>+++ b/include/linux/vfio_pci_core.h
>@@ -149,6 +149,10 @@ struct vfio_pci_core_device {
> 	struct notifier_block	nb;
> 	struct rw_semaphore	memory_lock;
> 	struct list_head	dmabufs;
>+#ifdef CONFIG_VFIO_PCI_LIVEUPDATE
>+	bool			liveupdate_frozen;
>+	bool			liveupdate_intx_disabled;
>+#endif
> };
>
> enum vfio_pci_io_width {
>-- 
>2.56.0.317.g454f8302e1-goog
>

Thanks,
Sami



More information about the kexec mailing list