Re: [PATCH v4 27/33] vfio/migration: Multifd device state transfer support - received buffers queuing

["Cédric Le Goater" <clg@redhat.com>]

On 1/30/25 11:08, Maciej S. Szmigiero wrote:
> From: "Maciej S. Szmigiero" <maciej.szmigiero@oracle.com>
> 
> The multifd received data needs to be reassembled since device state
> packets sent via different multifd channels can arrive out-of-order.
> 
> Therefore, each VFIO device state packet carries a header indicating its
> position in the stream.
> The raw device state data is saved into a VFIOStateBuffer for later
> in-order loading into the device.
> 
> The last such VFIO device state packet should have
> VFIO_DEVICE_STATE_CONFIG_STATE flag set and carry the device config state.
> 
> Signed-off-by: Maciej S. Szmigiero <maciej.szmigiero@oracle.com>
> ---
>   hw/vfio/migration.c           | 116 ++++++++++++++++++++++++++++++++++
>   hw/vfio/pci.c                 |   2 +
>   hw/vfio/trace-events          |   1 +
>   include/hw/vfio/vfio-common.h |   1 +
>   4 files changed, 120 insertions(+)
> 
> diff --git a/hw/vfio/migration.c b/hw/vfio/migration.c
> index bcdf204d5cf4..0c0caec1bd64 100644
> --- a/hw/vfio/migration.c
> +++ b/hw/vfio/migration.c
> @@ -301,6 +301,12 @@ typedef struct VFIOStateBuffer {
>   } VFIOStateBuffer;
>   
>   typedef struct VFIOMultifd {
> +    VFIOStateBuffers load_bufs;
> +    QemuCond load_bufs_buffer_ready_cond;
> +    QemuMutex load_bufs_mutex; /* Lock order: this lock -> BQL */
> +    uint32_t load_buf_idx;
> +    uint32_t load_buf_idx_last;
> +    uint32_t load_buf_queued_pending_buffers;
>   } VFIOMultifd;
>   
>   static void vfio_state_buffer_clear(gpointer data)
> @@ -346,6 +352,103 @@ static VFIOStateBuffer *vfio_state_buffers_at(VFIOStateBuffers *bufs, guint idx)
>       return &g_array_index(bufs->array, VFIOStateBuffer, idx);
>   }
>   
Each routine executed from a migration thread should have a preliminary
comment saying from which context it is called: migration or VFIO

> +static bool vfio_load_state_buffer_insert(VFIODevice *vbasedev,
> +                                          VFIODeviceStatePacket *packet,
> +                                          size_t packet_total_size,
> +                                          Error **errp)
> +{
> +    VFIOMigration *migration = vbasedev->migration;
> +    VFIOMultifd *multifd = migration->multifd;
> +    VFIOStateBuffer *lb;
> +
> +    vfio_state_buffers_assert_init(&multifd->load_bufs);
> +    if (packet->idx >= vfio_state_buffers_size_get(&multifd->load_bufs)) {
> +        vfio_state_buffers_size_set(&multifd->load_bufs, packet->idx + 1);
> +    }
> +
> +    lb = vfio_state_buffers_at(&multifd->load_bufs, packet->idx);
> +    if (lb->is_present) {
> +        error_setg(errp, "state buffer %" PRIu32 " already filled",
> +                   packet->idx);
> +        return false;
> +    }
> +
> +    assert(packet->idx >= multifd->load_buf_idx);
> +
> +    multifd->load_buf_queued_pending_buffers++;
> +    if (multifd->load_buf_queued_pending_buffers >
> +        vbasedev->migration_max_queued_buffers) {
> +        error_setg(errp,
> +                   "queuing state buffer %" PRIu32 " would exceed the max of %" PRIu64,
> +                   packet->idx, vbasedev->migration_max_queued_buffers);
> +        return false;
> +    }

AFAICT, attributes multifd->load_buf_queued_pending_buffers and
vbasedev->migration_max_queued_buffers are not strictly necessary.
They allow to count buffers and check an arbitrary limit, which
is UINT64_MAX today. It makes me wonder how useful they are.

Please introduce them in a separate patch at the end of the series,
adding documentation on the "x-migration-max-queued-buffers" property
and also general documentation on why and how to use it.

> +
> +    lb->data = g_memdup2(&packet->data, packet_total_size - sizeof(*packet));
> +    lb->len = packet_total_size - sizeof(*packet);
> +    lb->is_present = true;
> +
> +    return true;
> +}
> +
> +static bool vfio_load_state_buffer(void *opaque, char *data, size_t data_size,
> +                                   Error **errp)
> +{
> +    VFIODevice *vbasedev = opaque;
> +    VFIOMigration *migration = vbasedev->migration;
> +    VFIOMultifd *multifd = migration->multifd;
> +    VFIODeviceStatePacket *packet = (VFIODeviceStatePacket *)data;
> +
> +    /*
> +     * Holding BQL here would violate the lock order and can cause
> +     * a deadlock once we attempt to lock load_bufs_mutex below.
> +     */
> +    assert(!bql_locked());
> +
> +    if (!migration->multifd_transfer) {
> +        error_setg(errp,
> +                   "got device state packet but not doing multifd transfer");
> +        return false;
> +    }
> +
> +    assert(multifd);
> +
> +    if (data_size < sizeof(*packet)) {
> +        error_setg(errp, "packet too short at %zu (min is %zu)",
> +                   data_size, sizeof(*packet));
> +        return false;
> +    }
> +
> +    if (packet->version != 0) {

Please add a define for version, even if 0.

> +        error_setg(errp, "packet has unknown version %" PRIu32,
> +                   packet->version);
> +        return false;
> +    }
> +
> +    if (packet->idx == UINT32_MAX) {
> +        error_setg(errp, "packet has too high idx %" PRIu32,
> +                   packet->idx);

I don't think printing out packet->idx is useful here.

> +        return false;
> +    }
> +
> +    trace_vfio_load_state_device_buffer_incoming(vbasedev->name, packet->idx);

I wonder if we can add thread ids to trace events. It would be useful.

> +
> +    QEMU_LOCK_GUARD(&multifd->load_bufs_mutex);
> +
> +    /* config state packet should be the last one in the stream */
> +    if (packet->flags & VFIO_DEVICE_STATE_CONFIG_STATE) {
> +        multifd->load_buf_idx_last = packet->idx;
> +    }
> +
> +    if (!vfio_load_state_buffer_insert(vbasedev, packet, data_size, errp)) {

So the migration thread calling multifd_device_state_recv() will
exit and the vfio thread loading the state into the device will
hang until its aborted ?

This sequence is expected to be called to release the vfio thread

        while (multifd->load_bufs_thread_running) {
             multifd->load_bufs_thread_want_exit = true;

             qemu_cond_signal(&multifd->load_bufs_buffer_ready_cond);
	    ...
        }

right ?


The way the series is presented makes it a bit complex to follow the
proposition, especially regarding the creation and termination of
threads, something the reader should be aware of.

As an initial step in clarifying the design, I would have preferred
a series of patches introducing the various threads, migration threads
and VFIO threads, without any workload. Once the creation and termination
points are established I would then introduce the work load for each
thread.


Thanks,

C.




> +        return false;
> +    }
> +
> +    qemu_cond_signal(&multifd->load_bufs_buffer_ready_cond);
> +
> +    return true;
> +}
> +
>   static int vfio_save_device_config_state(QEMUFile *f, void *opaque,
>                                            Error **errp)
>   {
> @@ -405,11 +508,23 @@ static VFIOMultifd *vfio_multifd_new(void)
>   {
>       VFIOMultifd *multifd = g_new(VFIOMultifd, 1);
>   
> +    vfio_state_buffers_init(&multifd->load_bufs);
> +
> +    qemu_mutex_init(&multifd->load_bufs_mutex);
> +
> +    multifd->load_buf_idx = 0;
> +    multifd->load_buf_idx_last = UINT32_MAX;
> +    multifd->load_buf_queued_pending_buffers = 0;
> +    qemu_cond_init(&multifd->load_bufs_buffer_ready_cond);
> +
>       return multifd;
>   }
>   
>   static void vfio_multifd_free(VFIOMultifd *multifd)
>   {
> +    qemu_cond_destroy(&multifd->load_bufs_buffer_ready_cond);
> +    qemu_mutex_destroy(&multifd->load_bufs_mutex);
> +
>       g_free(multifd);
>   }
>   
> @@ -940,6 +1055,7 @@ static const SaveVMHandlers savevm_vfio_handlers = {
>       .load_setup = vfio_load_setup,
>       .load_cleanup = vfio_load_cleanup,
>       .load_state = vfio_load_state,
> +    .load_state_buffer = vfio_load_state_buffer,
>       .switchover_ack_needed = vfio_switchover_ack_needed,
>   };
>   
> diff --git a/hw/vfio/pci.c b/hw/vfio/pci.c
> index 83090c544d95..2700b355ecf1 100644
> --- a/hw/vfio/pci.c
> +++ b/hw/vfio/pci.c
> @@ -3380,6 +3380,8 @@ static const Property vfio_pci_dev_properties[] = {
>       DEFINE_PROP_ON_OFF_AUTO("x-migration-load-config-after-iter", VFIOPCIDevice,
>                               vbasedev.migration_load_config_after_iter,
>                               ON_OFF_AUTO_AUTO),
> +    DEFINE_PROP_UINT64("x-migration-max-queued-buffers", VFIOPCIDevice,
> +                       vbasedev.migration_max_queued_buffers, UINT64_MAX),
>       DEFINE_PROP_BOOL("migration-events", VFIOPCIDevice,
>                        vbasedev.migration_events, false),
>       DEFINE_PROP_BOOL("x-no-mmap", VFIOPCIDevice, vbasedev.no_mmap, false),
> diff --git a/hw/vfio/trace-events b/hw/vfio/trace-events
> index 1bebe9877d88..042a3dc54a33 100644
> --- a/hw/vfio/trace-events
> +++ b/hw/vfio/trace-events
> @@ -153,6 +153,7 @@ vfio_load_device_config_state_start(const char *name) " (%s)"
>   vfio_load_device_config_state_end(const char *name) " (%s)"
>   vfio_load_state(const char *name, uint64_t data) " (%s) data 0x%"PRIx64
>   vfio_load_state_device_data(const char *name, uint64_t data_size, int ret) " (%s) size %"PRIu64" ret %d"
> +vfio_load_state_device_buffer_incoming(const char *name, uint32_t idx) " (%s) idx %"PRIu32
>   vfio_migration_realize(const char *name) " (%s)"
>   vfio_migration_set_device_state(const char *name, const char *state) " (%s) state %s"
>   vfio_migration_set_state(const char *name, const char *new_state, const char *recover_state) " (%s) new state %s, recover state %s"
> diff --git a/include/hw/vfio/vfio-common.h b/include/hw/vfio/vfio-common.h
> index c0c9c0b1b263..0e8b0848882e 100644
> --- a/include/hw/vfio/vfio-common.h
> +++ b/include/hw/vfio/vfio-common.h
> @@ -139,6 +139,7 @@ typedef struct VFIODevice {
>       OnOffAuto enable_migration;
>       OnOffAuto migration_multifd_transfer;
>       OnOffAuto migration_load_config_after_iter;
> +    uint64_t migration_max_queued_buffers;
>       bool migration_events;
>       VFIODeviceOps *ops;
>       unsigned int num_irqs;
>