On 2/12/26 11:58 AM, Alex Williamson wrote:
On Thu, 12 Feb 2026 10:45:47 -0500
Anthony Krowiak <[email protected]> wrote:
I created a sysfs attribute for a mediated device to enable/block migration
of the associated VFIODevice in QEMU. When the sysfs attribute is changed,
an eventfd is signaled to QEMU to enable or block migration. In response,
QEMU adds a migration blocker to the VFIODevice object and sets it's
'enable_migration' field to 'ON_OFF_AUTO_OFF'.
If some userspace entity has an inclination to block migration, why
wouldn't it just coordinate with the VM orchestration tools rather than
take this inherently racy and vfio specific path of the device
signaling up through QEMU? Thanks,
Alex
Hi Alex, thanks for the response. I have never used a VM orchestration
tool, so I have no idea how this would, or even if it would accomplish
what I need. What I need is to prevent migration while changes to
the source and destination devices passed through to the guest
are made, without having to terminate the guests. These changes
can only be made while the devices are online and must be
synchronized, so they cannot occur during a migration, so I need
a way for a sysadmin to dynamically prevent migration while this
synchronization is completed.
