Re: [PATCH v6 00/16] QEMU cpus.c refactoring part2

[Alberto Garcia]

On Tue 01 Sep 2020 09:21:45 AM CEST, Claudio Fontana wrote:
> * in some cases the virtual clock is queried before an accelerator
>   is set or ticks are enabled with
>
>   qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL)
>   
>   by the qcow2.c code (ending up in 0); maybe this should not happen
>   at all, as it could hurt migrations with the clock jumping up from
>   0? Should it be QEMU_CLOCK_REALTIME? (Berto, Paolo)

As far as I can see the only place in the qcow2 code that uses
QEMU_CLOCK_VIRTUAL is in the timer that clears unused cache entries
periodically.

I used QEMU_CLOCK_VIRTUAL because it didn't make sense to me to expire
cache entries when the VM is stopped.

I'm not sure about what would happen during a migration, is the qcow2
cache migrated at all? And if it is, would the clock jump up suddenly?
If it's just that I understand that the effect would be that the timer
would be fired earlier than expected, but I don't think it's a big deal.

Berto