| Message ID | 20230112085020.15866-11-avihaih@nvidia.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | vfio/migration: Implement VFIO migration protocol v2 | expand |
On 1/12/23 09:50, Avihai Horon wrote: > During pre-copy phase of migration vfio_save_pending() is called > repeatedly and queries the VFIO device for its pending data size. > > As long as pending RAM size is over the threshold, migration can't > converge and be completed. Therefore, during this time there is no > point in querying the VFIO device pending data size. > > Avoid these unnecessary queries by issuing them in a RAM pre-copy > notifier instead of vfio_save_pending(). > > This way the VFIO device is queried only when RAM pending data is > below the threshold, when there is an actual chance for migration to > converge. > > Signed-off-by: Avihai Horon <avihaih@nvidia.com> Reviewed-by: Cédric Le Goater <clg@redhat.com> Thanks, C. > --- > include/hw/vfio/vfio-common.h | 2 ++ > hw/vfio/migration.c | 56 +++++++++++++++++++++++++++-------- > hw/vfio/trace-events | 1 + > 3 files changed, 46 insertions(+), 13 deletions(-) > > diff --git a/include/hw/vfio/vfio-common.h b/include/hw/vfio/vfio-common.h > index 2ec3346fea..113f8d9208 100644 > --- a/include/hw/vfio/vfio-common.h > +++ b/include/hw/vfio/vfio-common.h > @@ -65,11 +65,13 @@ typedef struct VFIOMigration { > uint32_t device_state_v1; > int vm_running; > Notifier migration_state; > + NotifierWithReturn migration_data; > uint64_t pending_bytes; > enum vfio_device_mig_state device_state; > int data_fd; > void *data_buffer; > size_t data_buffer_size; > + uint64_t stop_copy_size; > bool v2; > } VFIOMigration; > > diff --git a/hw/vfio/migration.c b/hw/vfio/migration.c > index 08f53189fa..04f4397212 100644 > --- a/hw/vfio/migration.c > +++ b/hw/vfio/migration.c > @@ -655,29 +655,19 @@ static void vfio_v1_save_cleanup(void *opaque) > trace_vfio_save_cleanup(vbasedev->name); > } > > -/* > - * Migration size of VFIO devices can be as little as a few KBs or as big as > - * many GBs. This value should be big enough to cover the worst case. > - */ > -#define VFIO_MIG_STOP_COPY_SIZE (100 * GiB) > static void vfio_save_pending(void *opaque, uint64_t threshold_size, > uint64_t *res_precopy_only, > uint64_t *res_compatible, > uint64_t *res_postcopy_only) > { > VFIODevice *vbasedev = opaque; > - uint64_t stop_copy_size = VFIO_MIG_STOP_COPY_SIZE; > + VFIOMigration *migration = vbasedev->migration; > > - /* > - * If getting pending migration size fails, VFIO_MIG_STOP_COPY_SIZE is > - * reported so downtime limit won't be violated. > - */ > - vfio_query_stop_copy_size(vbasedev, &stop_copy_size); > - *res_precopy_only += stop_copy_size; > + *res_precopy_only += migration->stop_copy_size; > > trace_vfio_save_pending(vbasedev->name, *res_precopy_only, > *res_postcopy_only, *res_compatible, > - stop_copy_size); > + migration->stop_copy_size); > } > > static void vfio_v1_save_pending(void *opaque, uint64_t threshold_size, > @@ -1104,6 +1094,40 @@ static void vfio_migration_state_notifier(Notifier *notifier, void *data) > } > } > > +/* > + * Migration size of VFIO devices can be as little as a few KBs or as big as > + * many GBs. This value should be big enough to cover the worst case. > + */ > +#define VFIO_MIG_STOP_COPY_SIZE (100 * GiB) > +static int vfio_migration_data_notifier(NotifierWithReturn *n, void *data) > +{ > + VFIOMigration *migration = container_of(n, VFIOMigration, migration_data); > + VFIODevice *vbasedev = migration->vbasedev; > + PrecopyNotifyData *pnd = data; > + > + if (pnd->reason != PRECOPY_NOTIFY_AFTER_BITMAP_SYNC) { > + return 0; > + } > + > + /* No need to get pending size when finishing migration */ > + if (runstate_check(RUN_STATE_FINISH_MIGRATE)) { > + return 0; > + } > + > + if (vfio_query_stop_copy_size(vbasedev, &migration->stop_copy_size)) { > + /* > + * Failed to get pending migration size. Report big pending size so > + * downtime limit won't be violated. > + */ > + migration->stop_copy_size = VFIO_MIG_STOP_COPY_SIZE; > + } > + > + trace_vfio_migration_data_notifier(vbasedev->name, > + migration->stop_copy_size); > + > + return 0; > +} > + > static void vfio_migration_exit(VFIODevice *vbasedev) > { > VFIOMigration *migration = vbasedev->migration; > @@ -1225,6 +1249,9 @@ static int vfio_migration_init(VFIODevice *vbasedev) > > migration->vm_state = qdev_add_vm_change_state_handler( > vbasedev->dev, vfio_vmstate_change, vbasedev); > + > + migration->migration_data.notify = vfio_migration_data_notifier; > + precopy_add_notifier(&migration->migration_data); > } else { > register_savevm_live(id, VMSTATE_INSTANCE_ID_ANY, 1, > &savevm_vfio_v1_handlers, vbasedev); > @@ -1283,6 +1310,9 @@ void vfio_migration_finalize(VFIODevice *vbasedev) > if (vbasedev->migration) { > VFIOMigration *migration = vbasedev->migration; > > + if (migration->v2) { > + precopy_remove_notifier(&migration->migration_data); > + } > remove_migration_state_change_notifier(&migration->migration_state); > qemu_del_vm_change_state_handler(migration->vm_state); > unregister_savevm(VMSTATE_IF(vbasedev->dev), "vfio", vbasedev); > diff --git a/hw/vfio/trace-events b/hw/vfio/trace-events > index e7b3b4658c..a8a64f0627 100644 > --- a/hw/vfio/trace-events > +++ b/hw/vfio/trace-events > @@ -173,3 +173,4 @@ vfio_load_cleanup(const char *name) " (%s)" > vfio_get_dirty_bitmap(int fd, uint64_t iova, uint64_t size, uint64_t bitmap_size, uint64_t start) "container fd=%d, iova=0x%"PRIx64" size= 0x%"PRIx64" bitmap_size=0x%"PRIx64" start=0x%"PRIx64 > vfio_iommu_map_dirty_notify(uint64_t iova_start, uint64_t iova_end) "iommu dirty @ 0x%"PRIx64" - 0x%"PRIx64 > vfio_save_block(const char *name, int data_size) " (%s) data_size %d" > +vfio_migration_data_notifier(const char *name, uint64_t stopcopy_size) " (%s) stopcopy size 0x%"PRIx64
diff --git a/include/hw/vfio/vfio-common.h b/include/hw/vfio/vfio-common.h index 2ec3346fea..113f8d9208 100644 --- a/include/hw/vfio/vfio-common.h +++ b/include/hw/vfio/vfio-common.h @@ -65,11 +65,13 @@ typedef struct VFIOMigration { uint32_t device_state_v1; int vm_running; Notifier migration_state; + NotifierWithReturn migration_data; uint64_t pending_bytes; enum vfio_device_mig_state device_state; int data_fd; void *data_buffer; size_t data_buffer_size; + uint64_t stop_copy_size; bool v2; } VFIOMigration; diff --git a/hw/vfio/migration.c b/hw/vfio/migration.c index 08f53189fa..04f4397212 100644 --- a/hw/vfio/migration.c +++ b/hw/vfio/migration.c @@ -655,29 +655,19 @@ static void vfio_v1_save_cleanup(void *opaque) trace_vfio_save_cleanup(vbasedev->name); } -/* - * Migration size of VFIO devices can be as little as a few KBs or as big as - * many GBs. This value should be big enough to cover the worst case. - */ -#define VFIO_MIG_STOP_COPY_SIZE (100 * GiB) static void vfio_save_pending(void *opaque, uint64_t threshold_size, uint64_t *res_precopy_only, uint64_t *res_compatible, uint64_t *res_postcopy_only) { VFIODevice *vbasedev = opaque; - uint64_t stop_copy_size = VFIO_MIG_STOP_COPY_SIZE; + VFIOMigration *migration = vbasedev->migration; - /* - * If getting pending migration size fails, VFIO_MIG_STOP_COPY_SIZE is - * reported so downtime limit won't be violated. - */ - vfio_query_stop_copy_size(vbasedev, &stop_copy_size); - *res_precopy_only += stop_copy_size; + *res_precopy_only += migration->stop_copy_size; trace_vfio_save_pending(vbasedev->name, *res_precopy_only, *res_postcopy_only, *res_compatible, - stop_copy_size); + migration->stop_copy_size); } static void vfio_v1_save_pending(void *opaque, uint64_t threshold_size, @@ -1104,6 +1094,40 @@ static void vfio_migration_state_notifier(Notifier *notifier, void *data) } } +/* + * Migration size of VFIO devices can be as little as a few KBs or as big as + * many GBs. This value should be big enough to cover the worst case. + */ +#define VFIO_MIG_STOP_COPY_SIZE (100 * GiB) +static int vfio_migration_data_notifier(NotifierWithReturn *n, void *data) +{ + VFIOMigration *migration = container_of(n, VFIOMigration, migration_data); + VFIODevice *vbasedev = migration->vbasedev; + PrecopyNotifyData *pnd = data; + + if (pnd->reason != PRECOPY_NOTIFY_AFTER_BITMAP_SYNC) { + return 0; + } + + /* No need to get pending size when finishing migration */ + if (runstate_check(RUN_STATE_FINISH_MIGRATE)) { + return 0; + } + + if (vfio_query_stop_copy_size(vbasedev, &migration->stop_copy_size)) { + /* + * Failed to get pending migration size. Report big pending size so + * downtime limit won't be violated. + */ + migration->stop_copy_size = VFIO_MIG_STOP_COPY_SIZE; + } + + trace_vfio_migration_data_notifier(vbasedev->name, + migration->stop_copy_size); + + return 0; +} + static void vfio_migration_exit(VFIODevice *vbasedev) { VFIOMigration *migration = vbasedev->migration; @@ -1225,6 +1249,9 @@ static int vfio_migration_init(VFIODevice *vbasedev) migration->vm_state = qdev_add_vm_change_state_handler( vbasedev->dev, vfio_vmstate_change, vbasedev); + + migration->migration_data.notify = vfio_migration_data_notifier; + precopy_add_notifier(&migration->migration_data); } else { register_savevm_live(id, VMSTATE_INSTANCE_ID_ANY, 1, &savevm_vfio_v1_handlers, vbasedev); @@ -1283,6 +1310,9 @@ void vfio_migration_finalize(VFIODevice *vbasedev) if (vbasedev->migration) { VFIOMigration *migration = vbasedev->migration; + if (migration->v2) { + precopy_remove_notifier(&migration->migration_data); + } remove_migration_state_change_notifier(&migration->migration_state); qemu_del_vm_change_state_handler(migration->vm_state); unregister_savevm(VMSTATE_IF(vbasedev->dev), "vfio", vbasedev); diff --git a/hw/vfio/trace-events b/hw/vfio/trace-events index e7b3b4658c..a8a64f0627 100644 --- a/hw/vfio/trace-events +++ b/hw/vfio/trace-events @@ -173,3 +173,4 @@ vfio_load_cleanup(const char *name) " (%s)" vfio_get_dirty_bitmap(int fd, uint64_t iova, uint64_t size, uint64_t bitmap_size, uint64_t start) "container fd=%d, iova=0x%"PRIx64" size= 0x%"PRIx64" bitmap_size=0x%"PRIx64" start=0x%"PRIx64 vfio_iommu_map_dirty_notify(uint64_t iova_start, uint64_t iova_end) "iommu dirty @ 0x%"PRIx64" - 0x%"PRIx64 vfio_save_block(const char *name, int data_size) " (%s) data_size %d" +vfio_migration_data_notifier(const char *name, uint64_t stopcopy_size) " (%s) stopcopy size 0x%"PRIx64
During pre-copy phase of migration vfio_save_pending() is called repeatedly and queries the VFIO device for its pending data size. As long as pending RAM size is over the threshold, migration can't converge and be completed. Therefore, during this time there is no point in querying the VFIO device pending data size. Avoid these unnecessary queries by issuing them in a RAM pre-copy notifier instead of vfio_save_pending(). This way the VFIO device is queried only when RAM pending data is below the threshold, when there is an actual chance for migration to converge. Signed-off-by: Avihai Horon <avihaih@nvidia.com> --- include/hw/vfio/vfio-common.h | 2 ++ hw/vfio/migration.c | 56 +++++++++++++++++++++++++++-------- hw/vfio/trace-events | 1 + 3 files changed, 46 insertions(+), 13 deletions(-)