5 Migration of virtual machine involves saving the state for each device that
6 the guest is running on source host and restoring this saved state on the
7 destination host. This document details how saving and restoring of VFIO
8 devices is done in QEMU.
10 Migration of VFIO devices consists of two phases: the optional pre-copy phase,
11 and the stop-and-copy phase. The pre-copy phase is iterative and allows to
12 accommodate VFIO devices that have a large amount of data that needs to be
13 transferred. The iterative pre-copy phase of migration allows for the guest to
14 continue whilst the VFIO device state is transferred to the destination, this
15 helps to reduce the total downtime of the VM. VFIO devices can choose to skip
16 the pre-copy phase of migration by returning pending_bytes as zero during the
19 A detailed description of the UAPI for VFIO device migration can be found in
20 the comment for the ``vfio_device_migration_info`` structure in the header
21 file linux-headers/linux/vfio.h.
23 VFIO implements the device hooks for the iterative approach as follows:
25 * A ``save_setup`` function that sets up the migration region and sets _SAVING
26 flag in the VFIO device state.
28 * A ``load_setup`` function that sets up the migration region on the
29 destination and sets _RESUMING flag in the VFIO device state.
31 * A ``save_live_pending`` function that reads pending_bytes from the vendor
32 driver, which indicates the amount of data that the vendor driver has yet to
33 save for the VFIO device.
35 * A ``save_live_iterate`` function that reads the VFIO device's data from the
36 vendor driver through the migration region during iterative phase.
38 * A ``save_state`` function to save the device config space if it is present.
40 * A ``save_live_complete_precopy`` function that resets _RUNNING flag from the
41 VFIO device state and iteratively copies the remaining data for the VFIO
42 device until the vendor driver indicates that no data remains (pending bytes
45 * A ``load_state`` function that loads the config section and the data
46 sections that are generated by the save functions above
48 * ``cleanup`` functions for both save and load that perform any migration
49 related cleanup, including unmapping the migration region
52 The VFIO migration code uses a VM state change handler to change the VFIO
53 device state when the VM state changes from running to not-running, and
56 Similarly, a migration state change handler is used to trigger a transition of
57 the VFIO device state when certain changes of the migration state occur. For
58 example, the VFIO device state is transitioned back to _RUNNING in case a
59 migration failed or was canceled.
61 System memory dirty pages tracking
62 ----------------------------------
64 A ``log_global_start`` and ``log_global_stop`` memory listener callback informs
65 the VFIO IOMMU module to start and stop dirty page tracking. A ``log_sync``
66 memory listener callback marks those system memory pages as dirty which are
67 used for DMA by the VFIO device. The dirty pages bitmap is queried per
68 container. All pages pinned by the vendor driver through external APIs have to
69 be marked as dirty during migration. When there are CPU writes, CPU dirty page
70 tracking can identify dirtied pages, but any page pinned by the vendor driver
71 can also be written by the device. There is currently no device or IOMMU
72 support for dirty page tracking in hardware.
74 By default, dirty pages are tracked when the device is in pre-copy as well as
75 stop-and-copy phase. So, a page pinned by the vendor driver will be copied to
76 the destination in both phases. Copying dirty pages in pre-copy phase helps
77 QEMU to predict if it can achieve its downtime tolerances. If QEMU during
78 pre-copy phase keeps finding dirty pages continuously, then it understands
79 that even in stop-and-copy phase, it is likely to find dirty pages and can
80 predict the downtime accordingly.
82 QEMU also provides a per device opt-out option ``pre-copy-dirty-page-tracking``
83 which disables querying the dirty bitmap during pre-copy phase. If it is set to
84 off, all dirty pages will be copied to the destination in stop-and-copy phase
87 System memory dirty pages tracking when vIOMMU is enabled
88 ---------------------------------------------------------
90 With vIOMMU, an IO virtual address range can get unmapped while in pre-copy
91 phase of migration. In that case, the unmap ioctl returns any dirty pages in
92 that range and QEMU reports corresponding guest physical pages dirty. During
93 stop-and-copy phase, an IOMMU notifier is used to get a callback for mapped
94 pages and then dirty pages bitmap is fetched from VFIO IOMMU modules for those
97 Flow of state changes during Live migration
98 ===========================================
100 Below is the flow of state change during live migration.
101 The values in the brackets represent the VM state, the migration state, and
102 the VFIO device state, respectively.
104 Live migration save path
105 ------------------------
109 QEMU normal running state
110 (RUNNING, _NONE, _RUNNING)
112 migrate_init spawns migration_thread
113 Migration thread then calls each device's .save_setup()
114 (RUNNING, _SETUP, _RUNNING|_SAVING)
116 (RUNNING, _ACTIVE, _RUNNING|_SAVING)
117 If device is active, get pending_bytes by .save_live_pending()
118 If total pending_bytes >= threshold_size, call .save_live_iterate()
119 Data of VFIO device for pre-copy phase is copied
120 Iterate till total pending bytes converge and are less than threshold
122 On migration completion, vCPU stops and calls .save_live_complete_precopy for
123 each active device. The VFIO device is then transitioned into _SAVING state
124 (FINISH_MIGRATE, _DEVICE, _SAVING)
126 For the VFIO device, iterate in .save_live_complete_precopy until
128 (FINISH_MIGRATE, _DEVICE, _STOPPED)
130 (FINISH_MIGRATE, _COMPLETED, _STOPPED)
131 Migraton thread schedules cleanup bottom half and exits
133 Live migration resume path
134 --------------------------
138 Incoming migration calls .load_setup for each device
139 (RESTORE_VM, _ACTIVE, _STOPPED)
141 For each device, .load_state is called for that device section data
142 (RESTORE_VM, _ACTIVE, _RESUMING)
144 At the end, .load_cleanup is called for each device and vCPUs are started
145 (RUNNING, _NONE, _RUNNING)
150 Postcopy migration is currently not supported for VFIO devices.