Re: [Qemu-devel] [RFC PATCH v4 17/25] cpus: make icount warp deterministic in replay mode

[Paolo Bonzini <pbonzini@redhat.com>]

On 07/11/2014 12:45, Pavel Dovgaluk wrote:
>> From: Paolo Bonzini [mailto:pbonzini@redhat.com]
>> On 07/11/2014 11:33, Pavel Dovgalyuk wrote:
>>> This patch adds saving and replaying warping parameters in record and replay
>>> modes. These parameters affect on virtual clock values and therefore should
>>> be deterministic.
>>
>> Why are QEMU_CLOCK_REALTIME timers not recorded/replayed like
>> QEMU_CLOCK_HOST is?
> 
> We cannot record all QEMU_CLOCK_REALTIME timers, because they are simulator-related timers,
> not the VM-related ones. If we'll record them then we'll have to replay simulator
> execution. And we want just to replay VM allowing different realtime events to be
> non-replayable (like screen refresh, monitor events handling, and so on).

Hmm... I think you want the icount_rt timer to use what
QEMU_CLOCK_VIRTUAL does in !use_icount mode.  Maybe you can add a fourth
clock, QEMU_CLOCK_VIRTUAL_RT and use this one independent of the replay
mode.  This clock can then be traced like QEMU_CLOCK_HOST.

The problem is that QEMU_CLOCK_HOST can move backwards or forwards,
sometimes by as much as 1 hour, so it's not the best match for this case.

Paolo