Hi Oliver,
On Wed, 08 Mar 2023 07:46:00 +0000,
Oliver Upton <oliver.upton@xxxxxxxxx> wrote:
[...]
> I certainly agree that (2a) is highly desirable to get existing VMMs to
> 'do the right thing' for free. Playing devil's advocate, would this not
> also break the tracing example you've given of correlating timestamps
> between the host and guest? I wouldn't expect a userspace + VM tracing
> contraption to live migrate but restoring from a snapshot seems
> plausible.
It really depends when this VM was saved.
If you saved it on an old host that doesn't expose CNTPCT and restore
it on a new host, the physical offset is still zero (this is already
special-cased), and there is no difference in behaviour.
If you saved it from a host that does expose CNTPCT, then the
behaviour changes. But should we really care?
> Regardless, I like the general direction you've proposed. IIUC, you'll
> want to go ahead with ignoring writes to CNT{P,V}CT if the offset was
> written by userspace, right?
That'd be my preference in order not to break the "blind restore"
behaviour that QEMU already uses for about everything.
I'll repost the series shortly, as it has grown some extra goodies
such as moving the PPI settings out of the way...
Thanks,
M.
--
Without deviation from the norm, progress is not possible.
