On 05/09/2024 19:23, Peter Xu wrote:
External email: Use caution opening links or attachments
On Thu, Sep 05, 2024 at 07:07:28PM +0300, Avihai Horon wrote:
So I wonder why stop-size can change from a driver, and whether that can be
reported in a more predictable fashion. Otherwise I see little point in
providing both stop-size and precopy-size, otherwise we'll always add them
up into VFIO's total pending-size. The line on provision which data falls
into which bucket doesn't seem to be clear to me.
Stopcopy-size is reported by VFIO_DEVICE_FEATURE_MIG_DATA_SIZE ioctl, which
states that this is "[...] the estimated data length that will be required
to complete stop copy".
So by this definition, stopcopy-size can change during precopy (it can also
change if device resources are created or destroyed).
Precopy-size is reported by VFIO_MIG_GET_PRECOPY_INFO ioctl, which states
that this is "[...] estimates of data available from the device during the
PRE_COPY states".
Maybe the confusion was caused by this line in vfio_state_pending_exact():
*must_precopy += migration->precopy_init_size +
migration->precopy_dirty_size;
Which I think should be removed, as stopcopy-size should already cover
precopy data.
Would you please help double check on the kernel drivers' side? If that's
true I agree that should be dropped.
Sure.
Does it also mean then that the currently reported stop-size - precopy-size
will be very close to the constant non-iterable data size?
It's not constant, while the VM is running it can change.
Thanks.
If so, is it
cache-able? It doesn't need to be 100% accurate as the ioctls are not
atomic, but still if it's mostly usable (again, will need you all to help
justify that..) as that then it looks like we have some way to report that
without the concern you raised before.
Thanks,
--
Peter Xu
