From: Jonah Palmer <jonah.palmer@oracle.com>
To: qemu-devel@nongnu.org
Cc: jonah.palmer@oracle.com, eperezma@redhat.com, peterx@redhat.com,
mst@redhat.com, jasowang@redhat.com, lvivier@redhat.com,
dtatulea@nvidia.com, leiyang@redhat.com, parav@mellanox.com,
sgarzare@redhat.com, si-wei.liu@oracle.com,
lingshan.zhu@intel.com, boris.ostrovsky@oracle.com
Subject: [PATCH v4 7/7] vdpa: move memory listener register to vhost_vdpa_init
Date: Wed, 7 May 2025 14:46:47 -0400 [thread overview]
Message-ID: <20250507184647.15580-8-jonah.palmer@oracle.com> (raw)
In-Reply-To: <20250507184647.15580-1-jonah.palmer@oracle.com>
From: Eugenio Pérez <eperezma@redhat.com>
Current memory operations like pinning may take a lot of time at the
destination. Currently they are done after the source of the migration is
stopped, and before the workload is resumed at the destination. This is a
period where neigher traffic can flow, nor the VM workload can continue
(downtime).
We can do better as we know the memory layout of the guest RAM at the
destination from the moment that all devices are initializaed. So
moving that operation allows QEMU to communicate the kernel the maps
while the workload is still running in the source, so Linux can start
mapping them.
As a small drawback, there is a time in the initialization where QEMU
cannot respond to QMP etc. By some testing, this time is about
0.2seconds. This may be further reduced (or increased) depending on the
vdpa driver and the platform hardware, and it is dominated by the cost
of memory pinning.
This matches the time that we move out of the called downtime window.
The downtime is measured as checking the trace timestamp from the moment
the source suspend the device to the moment the destination starts the
eight and last virtqueue pair. For a 39G guest, it goes from ~2.2526
secs to 2.0949.
Signed-off-by: Eugenio Pérez <eperezma@redhat.com>
Signed-off-by: Jonah Palmer <jonah.palmer@oracle.com>
v3:
---
Move memory listener unregistration from vhost_vdpa_reset_status to
vhost_vdpa_reset_device. By unregistering the listener here, we can
guarantee that every reset leaves the device in an expected state.
Also remove the duplicate call in vhost_vdpa_reset_status.
Reported-by: Lei Yang <leiyang@redhat.com>
Suggested-by: Si-Wei Liu <si-wei.liu@oracle.com>
--
v2:
Move the memory listener registration to vhost_vdpa_set_owner function.
In case of hotplug the vdpa device, the memory is already set up, and
leaving memory listener register call in the init function made maps
occur before set owner call.
To be 100% safe, let's put it right after set_owner call.
Reported-by: Lei Yang <leiyang@redhat.com>
---
hw/virtio/vhost-vdpa.c | 35 ++++++++++++++++++++++++++++-------
1 file changed, 28 insertions(+), 7 deletions(-)
diff --git a/hw/virtio/vhost-vdpa.c b/hw/virtio/vhost-vdpa.c
index de834f2ebd..e20da95f30 100644
--- a/hw/virtio/vhost-vdpa.c
+++ b/hw/virtio/vhost-vdpa.c
@@ -894,8 +894,14 @@ static int vhost_vdpa_reset_device(struct vhost_dev *dev)
ret = vhost_vdpa_call(dev, VHOST_VDPA_SET_STATUS, &status);
trace_vhost_vdpa_reset_device(dev);
+ if (ret) {
+ return ret;
+ }
+
+ memory_listener_unregister(&v->shared->listener);
+ v->shared->listener_registered = false;
v->suspended = false;
- return ret;
+ return 0;
}
static int vhost_vdpa_get_vq_index(struct vhost_dev *dev, int idx)
@@ -1379,6 +1385,11 @@ static int vhost_vdpa_dev_start(struct vhost_dev *dev, bool started)
"IOMMU and try again");
return -1;
}
+ if (v->shared->listener_registered &&
+ dev->vdev->dma_as != v->shared->listener.address_space) {
+ memory_listener_unregister(&v->shared->listener);
+ v->shared->listener_registered = false;
+ }
if (!v->shared->listener_registered) {
memory_listener_register(&v->shared->listener, dev->vdev->dma_as);
v->shared->listener_registered = true;
@@ -1392,8 +1403,6 @@ static int vhost_vdpa_dev_start(struct vhost_dev *dev, bool started)
static void vhost_vdpa_reset_status(struct vhost_dev *dev)
{
- struct vhost_vdpa *v = dev->opaque;
-
if (!vhost_vdpa_last_dev(dev)) {
return;
}
@@ -1401,9 +1410,6 @@ static void vhost_vdpa_reset_status(struct vhost_dev *dev)
vhost_vdpa_reset_device(dev);
vhost_vdpa_add_status(dev, VIRTIO_CONFIG_S_ACKNOWLEDGE |
VIRTIO_CONFIG_S_DRIVER);
- memory_listener_unregister(&v->shared->listener);
- v->shared->listener_registered = false;
-
}
static int vhost_vdpa_set_log_base(struct vhost_dev *dev, uint64_t base,
@@ -1537,12 +1543,27 @@ static int vhost_vdpa_get_features(struct vhost_dev *dev,
static int vhost_vdpa_set_owner(struct vhost_dev *dev)
{
+ int r;
+ struct vhost_vdpa *v;
+
if (!vhost_vdpa_first_dev(dev)) {
return 0;
}
trace_vhost_vdpa_set_owner(dev);
- return vhost_vdpa_call(dev, VHOST_SET_OWNER, NULL);
+ r = vhost_vdpa_call(dev, VHOST_SET_OWNER, NULL);
+ if (unlikely(r < 0)) {
+ return r;
+ }
+
+ /*
+ * Being optimistic and listening address space memory. If the device
+ * uses vIOMMU, it is changed at vhost_vdpa_dev_start.
+ */
+ v = dev->opaque;
+ memory_listener_register(&v->shared->listener, &address_space_memory);
+ v->shared->listener_registered = true;
+ return 0;
}
static int vhost_vdpa_vq_get_addr(struct vhost_dev *dev,
--
2.43.5
next prev parent reply other threads:[~2025-05-07 18:48 UTC|newest]
Thread overview: 44+ messages / expand[flat|nested] mbox.gz Atom feed top
2025-05-07 18:46 [PATCH v4 0/7] Move memory listener register to vhost_vdpa_init Jonah Palmer
2025-05-07 18:46 ` [PATCH v4 1/7] vdpa: check for iova tree initialized at net_client_start Jonah Palmer
2025-05-16 1:52 ` Jason Wang
2025-05-07 18:46 ` [PATCH v4 2/7] vdpa: reorder vhost_vdpa_set_backend_cap Jonah Palmer
2025-05-16 1:53 ` Jason Wang
2025-05-16 1:56 ` Jason Wang
2025-05-07 18:46 ` [PATCH v4 3/7] vdpa: set backend capabilities at vhost_vdpa_init Jonah Palmer
2025-05-16 1:57 ` Jason Wang
2025-05-07 18:46 ` [PATCH v4 4/7] vdpa: add listener_registered Jonah Palmer
2025-05-16 2:00 ` Jason Wang
2025-05-07 18:46 ` [PATCH v4 5/7] vdpa: reorder listener assignment Jonah Palmer
2025-05-16 2:01 ` Jason Wang
2025-05-07 18:46 ` [PATCH v4 6/7] vdpa: move iova_tree allocation to net_vhost_vdpa_init Jonah Palmer
2025-05-16 2:07 ` Jason Wang
2025-05-07 18:46 ` Jonah Palmer [this message]
2025-05-15 5:42 ` [PATCH v4 7/7] vdpa: move memory listener register to vhost_vdpa_init Michael S. Tsirkin
2025-05-15 17:36 ` Si-Wei Liu
2025-05-20 13:23 ` Jonah Palmer
2025-05-14 1:42 ` [PATCH v4 0/7] Move " Lei Yang
2025-05-14 15:49 ` Eugenio Perez Martin
2025-05-15 0:17 ` Si-Wei Liu
2025-05-15 5:43 ` Michael S. Tsirkin
2025-05-15 17:41 ` Si-Wei Liu
2025-05-16 10:45 ` Michael S. Tsirkin
2025-05-15 8:30 ` Eugenio Perez Martin
2025-05-16 1:49 ` Jason Wang
2025-05-20 13:27 ` Jonah Palmer
2025-05-14 23:00 ` Si-Wei Liu
2025-05-16 1:47 ` Jason Wang
2025-05-16 1:51 ` Jason Wang
2025-05-16 6:40 ` Markus Armbruster
2025-05-16 19:09 ` Si-Wei Liu
2025-05-26 9:16 ` Markus Armbruster
2025-05-29 7:57 ` Si-Wei Liu
2025-06-02 8:08 ` Markus Armbruster
2025-06-02 8:29 ` Markus Armbruster
2025-06-06 16:21 ` Jonah Palmer
2025-06-26 12:08 ` Markus Armbruster
2025-07-02 19:31 ` Jonah Palmer
2025-07-04 15:00 ` Markus Armbruster
2025-07-07 13:21 ` Jonah Palmer
2025-07-08 8:17 ` Markus Armbruster
2025-07-09 19:57 ` Jonah Palmer
2025-07-10 5:31 ` Markus Armbruster
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=20250507184647.15580-8-jonah.palmer@oracle.com \
--to=jonah.palmer@oracle.com \
--cc=boris.ostrovsky@oracle.com \
--cc=dtatulea@nvidia.com \
--cc=eperezma@redhat.com \
--cc=jasowang@redhat.com \
--cc=leiyang@redhat.com \
--cc=lingshan.zhu@intel.com \
--cc=lvivier@redhat.com \
--cc=mst@redhat.com \
--cc=parav@mellanox.com \
--cc=peterx@redhat.com \
--cc=qemu-devel@nongnu.org \
--cc=sgarzare@redhat.com \
--cc=si-wei.liu@oracle.com \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).