Re: [PATCH v4 00/33] Multifd 🔀 device state transfer support with VFIO consumer

[Fabiano Rosas]

"Maciej S. Szmigiero" <mail@maciej.szmigiero.name> writes:

> On 30.01.2025 21:19, Fabiano Rosas wrote:
>> "Maciej S. Szmigiero" <mail@maciej.szmigiero.name> writes:
>> 
>>> From: "Maciej S. Szmigiero" <maciej.szmigiero@oracle.com>
>>>
>>> This is an updated v4 patch series of the v3 series located here:
>>> https://lore.kernel.org/qemu-devel/cover.1731773021.git.maciej.szmigiero@oracle.com/
>>>
>>> Changes from v3:
>>> * MigrationLoadThread now returns bool and an Error complex error type
>>> instead of just an int.
>>>
>>> * qemu_loadvm_load_thread_pool now reports error via migrate_set_error()
>>> instead of dedicated load_threads_ret variable.
>>>
>>> * Since the change above uncovered an issue with respect to multifd send
>>> channels not terminating TLS session properly QIOChannelTLS now allows
>>> gracefully handling this situation.
>>>
>>> * qemu_loadvm_load_thread_pool state is now part of MigrationIncomingState
>>> instead of being stored in global variables.
>>> This state now also has its own init/cleanup helpers.
>>>
>>> * qemu_loadvm_load_thread_pool code is now moved into a separate section
>>> of the savevm.c file, marked by an appropriate comment.
>>>
>>> * thread_pool_free() is now documented to have wait-before-free semantic,
>>> which allowed removal of explicit waits from thread pool cleanup paths.
>>>
>>> * thread_pool_submit_immediate() method was added since this functionality
>>> is used by both generic thread pool users in this patch set.
>>>
>>> * postcopy_ram_listen_thread() now takes BQL around function calls that
>>> ultimately call migration methods requiring BQL.
>>> This fixes one of QEMU tests failing when explicitly BQL-sensitive code
>>> is added later to these methods.
>>>
>>> * qemu_loadvm_load_state_buffer() now returns a bool value instead of int.
>>>
>>> * "Send final SYNC only after device state is complete" patch was
>>> dropped since Peter implemented equivalent functionality upstream.
>>>
>>> * "Document the BQL behavior of load SaveVMHandlers" patch was dropped
>>> since that's something better done later, separately from this patch set.
>>>
>>> * Header size is now added to mig_stats.multifd_bytes where it is actually
>>> sent in the zero copy case - in multifd_nocomp_send_prepare().
>>>
>>> * Spurious wakeups from qemu_cond_wait() are now handled properly as
>>> pointed out by Avihai.
>>>
>>> * VFIO migration FD now allows partial write() completion as pointed out
>>> by Avihai.
>>>
>>> * Patch "vfio/migration: Don't run load cleanup if load setup didn't run"
>>> was dropped, instead all objects related to multifd load are now located in
>>> their own VFIOMultifd struct which is allocated only if multifd device state
>>> transfer is actually in use.
>>>
>>> * Intermediate VFIOStateBuffers API as suggested by Avihai is now introduced
>>> to simplify vfio_load_state_buffer() and vfio_load_bufs_thread().
>>>
>>> * Optional VFIO device config state loading interlocking with loading
>>> other iterables is now possible due to ARM64 platform VFIO dependency on
>>> interrupt controller being loaded first as pointed out by Avihai.
>>>
>>> * Patch "Multifd device state transfer support - receive side" was split
>>> into a few smaller patches as suggested by Cédric.
>>>
>>> * x-migration-multifd-transfer VFIO property compat changes were moved
>>> into a separate patch as suggested by Cédric.
>>>
>>> * Other small changes, like renamed functions and variables/members, added
>>> review tags, code formatting, moved QEMU_LOCK_GUARD() instances closer to
>>> actual protected blocks, etc.
>>>
>>> ========================================================================
>>>
>>> This patch set is targeting QEMU 10.0.
>>>
>>> What's not yet present is documentation update under docs/devel/migration
>>> but I didn't want to delay posting the code any longer.
>>> Such doc can still be merged later when the design is 100% finalized.
>>>
>>> ========================================================================
>>>
>>> Maciej S. Szmigiero (32):
>>>    migration: Clarify that {load,save}_cleanup handlers can run without
>>>      setup
>>>    thread-pool: Remove thread_pool_submit() function
>>>    thread-pool: Rename AIO pool functions to *_aio() and data types to
>>>      *Aio
>>>    thread-pool: Implement generic (non-AIO) pool support
>>>    migration: Add MIG_CMD_SWITCHOVER_START and its load handler
>>>    migration: Add qemu_loadvm_load_state_buffer() and its handler
>>>    io: tls: Allow terminating the TLS session gracefully with EOF
>>>    migration/multifd: Allow premature EOF on TLS incoming channels
>>>    migration: postcopy_ram_listen_thread() needs to take BQL for some
>>>      calls
>>>    error: define g_autoptr() cleanup function for the Error type
>>>    migration: Add thread pool of optional load threads
>>>    migration/multifd: Split packet into header and RAM data
>>>    migration/multifd: Device state transfer support - receive side
>>>    migration/multifd: Make multifd_send() thread safe
>>>    migration/multifd: Add an explicit MultiFDSendData destructor
>>>    migration/multifd: Device state transfer support - send side
>>>    migration/multifd: Add multifd_device_state_supported()
>>>    migration: Add save_live_complete_precopy_thread handler
>>>    vfio/migration: Add x-migration-load-config-after-iter VFIO property
>>>    vfio/migration: Add load_device_config_state_start trace event
>>>    vfio/migration: Convert bytes_transferred counter to atomic
>>>    vfio/migration: Multifd device state transfer support - basic types
>>>    vfio/migration: Multifd device state transfer support -
>>>      VFIOStateBuffer(s)
>>>    vfio/migration: Multifd device state transfer - add support checking
>>>      function
>>>    vfio/migration: Multifd device state transfer support - receive
>>>      init/cleanup
>>>    vfio/migration: Multifd device state transfer support - received
>>>      buffers queuing
>>>    vfio/migration: Multifd device state transfer support - load thread
>>>    vfio/migration: Multifd device state transfer support - config loading
>>>      support
>>>    migration/qemu-file: Define g_autoptr() cleanup function for QEMUFile
>>>    vfio/migration: Multifd device state transfer support - send side
>>>    vfio/migration: Add x-migration-multifd-transfer VFIO property
>>>    hw/core/machine: Add compat for x-migration-multifd-transfer VFIO
>>>      property
>>>
>>> Peter Xu (1):
>>>    migration/multifd: Make MultiFDSendData a struct
>>>
>>>   hw/core/machine.c                  |   2 +
>>>   hw/vfio/migration.c                | 754 ++++++++++++++++++++++++++++-
>>>   hw/vfio/pci.c                      |  14 +
>>>   hw/vfio/trace-events               |  11 +-
>>>   include/block/aio.h                |   8 +-
>>>   include/block/thread-pool.h        |  62 ++-
>>>   include/hw/vfio/vfio-common.h      |   7 +
>>>   include/io/channel-tls.h           |  11 +
>>>   include/migration/client-options.h |   4 +
>>>   include/migration/misc.h           |  16 +
>>>   include/migration/register.h       |  54 ++-
>>>   include/qapi/error.h               |   2 +
>>>   include/qemu/typedefs.h            |   6 +
>>>   io/channel-tls.c                   |   6 +
>>>   migration/colo.c                   |   3 +
>>>   migration/meson.build              |   1 +
>>>   migration/migration-hmp-cmds.c     |   2 +
>>>   migration/migration.c              |   6 +-
>>>   migration/migration.h              |   7 +
>>>   migration/multifd-device-state.c   | 192 ++++++++
>>>   migration/multifd-nocomp.c         |  30 +-
>>>   migration/multifd.c                | 248 ++++++++--
>>>   migration/multifd.h                |  74 ++-
>>>   migration/options.c                |   9 +
>>>   migration/qemu-file.h              |   2 +
>>>   migration/savevm.c                 | 195 +++++++-
>>>   migration/savevm.h                 |   6 +-
>>>   migration/trace-events             |   1 +
>>>   scripts/analyze-migration.py       |  11 +
>>>   tests/unit/test-thread-pool.c      |   6 +-
>>>   util/async.c                       |   6 +-
>>>   util/thread-pool.c                 | 184 +++++--
>>>   util/trace-events                  |   6 +-
>>>   33 files changed, 1814 insertions(+), 132 deletions(-)
>>>   create mode 100644 migration/multifd-device-state.c
>> 
>> Hi!
>> 
>> We have build issues:
>> 
>> https://gitlab.com/farosas/qemu/-/pipelines/1649146958
>> 
>
> Looks like that's an issue that qatomics on 64-bit
> VFIO bytes transferred counters aren't available on
> 32-bit host platforms.
>
> The easiest way would be probably to change these to
> 32-bit counters on 32-bit platforms since they can't
> realistically address more memory anyway.
>
>> And the postcopy/recovery test is failing. It seems the migration
>> finishes before the test can issue migrate-pause:
>> 
>> QTEST_QEMU_BINARY=./qemu-system-x86_64  ./tests/qtest/migration-test -p
>> /x86_64/migration/postcopy/recovery/plain
>> ...
>> {"execute": "migrate-start-postcopy"}
>> {"return": {}}
>> {"secs": 1738267018, "usecs": 860991}, "event": "MIGRATION", "data": 
>> {"status": "postcopy-active"}
>> {"secs": 1738267018, "usecs": 861284}, "event": "STOP"
>> {"secs": 1738267017, "usecs": 960322}, "event": "MIGRATION", "data": 
>> {"status": "active"}
>> {"secs": 1738267018, "usecs": 865589}, "event": "MIGRATION", "data": 
>> {"status": "postcopy-active"}
>> {"secs": 1738267099, "usecs": 120971}, "event": "MIGRATION", "data": 
>> {"status": "completed"}
>> {"secs": 1738267099, "usecs": 121154}, "event": "RESUME"
>> {"execute": "query-migrate"}
>> 
>> ERROR:../tests/qtest/migration/migration-qmp.c:172:check_migration_status:
>> assertion failed (current_status != "completed"): ("completed" !=
>> "completed")
>> 
>
> Hmm, it looks like this failure wasn't showing
> in my tests because the test was skipped due to
> missing userfaultfd support:
>
> $ QTEST_QEMU_BINARY=./qemu-system-x86_64 ./tests/qtest/migration-test -p 
> /x86_64/migration/postcopy/recovery/plain
> TAP version 14
> # random seed: R02Sc99a7d93274064bb87f3e0789fbf8326
> # Skipping test: userfaultfd not available
> # Start of x86_64 tests
> # Start of migration tests
> # End of migration tests
> # End of x86_64 tests
> 1..0
>
> Will try to make this test run and investigate the reason for
> failure.

This will probably help:

sysctl -w vm.unprivileged_userfaultfd=1

I also had broken userfaultfd detection in the tests a while back. But
it's fixed now.

>
> Thanks,
> Maciej