diff options
| author | Eugenio Pérez <eperezma@redhat.com> | 2021-08-12 16:09:33 +0200 |
|---|---|---|
| committer | Michael S. Tsirkin <mst@redhat.com> | 2021-09-04 09:07:45 -0400 |
| commit | e6db5df7e2c6e64846f47a6785a9a67f08574790 (patch) | |
| tree | 5db9674e04eb8a82508f934462baf9e779914edb /scripts/qapi/types.py | |
| parent | 8880cc4362fde4ecdac0b2092318893118206fcf (diff) | |
| download | focaccia-qemu-e6db5df7e2c6e64846f47a6785a9a67f08574790.tar.gz focaccia-qemu-e6db5df7e2c6e64846f47a6785a9a67f08574790.zip | |
vhost-vdpa: Do not send empty IOTLB update batches
With the introduction of the batch hinting, meaningless batches can be
created with no IOTLB updates if the memory region was skipped by
vhost_vdpa_listener_skipped_section. This is the case of host notifiers
memory regions, device un/realize, and others. This causes the vdpa
device to receive dma mapping settings with no changes, a possibly
expensive operation for nothing.
To avoid that, VHOST_IOTLB_BATCH_BEGIN hint is delayed until we have a
meaningful (not skipped section) mapping or unmapping operation, and
VHOST_IOTLB_BATCH_END is not written unless at least one of _UPDATE /
_INVALIDATE has been issued.
v3:
* Use a bool instead of a counter avoiding potential number wrapping
* Fix bad check on _commit
* Move VHOST_BACKEND_F_IOTLB_BATCH check to
vhost_vdpa_iotlb_batch_begin_once
v2 (from RFC):
* Rename misleading name
* Abstract start batching function for listener_add/del
Signed-off-by: Eugenio Pérez <eperezma@redhat.com>
Message-Id: <20210812140933.226288-1-eperezma@redhat.com>
Acked-by: Jason Wang <jasowang@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Diffstat (limited to 'scripts/qapi/types.py')
0 files changed, 0 insertions, 0 deletions