Re: [PATCH v5 08/24] replay: Fix migration use of clock

[Nicholas Piggin]

On Wed Mar 20, 2024 at 6:40 AM AEST, Alex Bennée wrote:
> Nicholas Piggin <npiggin@gmail.com> writes:
>
> > Migration reads host clocks when not holding the replay_mutex, which
> > asserts when recording a trace. It seems that these migration times
> > should be host times like other statistics in MigrationState.
>
> s/host/CLOCK_HOST/ and s/host/CLOCK_REALTIME/ but its a confusing
> sentence.

Yes you're right.

> If the MigrationState is guest visible it should be
> QEMU_CLOCK_VIRTUAL surely?

I didn't think it was visible. It was added with ed4fbd10823 and
just exported to QMP.

It was the first and only user of host clock in migration stats,
the rest use rt clock. OTOH it does seem to have deliberately
chosen host... can you see any reason for it?

Thanks,
Nick