[Qemu-devel] Deterministic replay

[Qemu-devel] Deterministic replay

[Pavel Dovgaluk]

Hello.

 I am working on implementation of deteministic execution replay technology
for Qemu. It should be similar to VMWare's replay debugging.
 
 To make alarm timer (which invokes host_alarm_handler function)
determinisic, I changed it's behavior: it sets flag, that execution should
be stopped and this flag is checked before every instruction in the
translated code.
 To allow breaking the execution in any moment, I added flag check and
exit_tb call before every instruction.

 But there is one problem occurred: when I make the same actions like
gen_eob() function, Windows XP cannot load on emulated machine (it reboots
with GPF).
 But when I remove two of the actions, everything is OK. These actions are
the following:
    if (s->tb->flags & HF_INHIBIT_IRQ_MASK) {
        gen_helper_reset_inhibit_irq();
    }
    if (s->tb->flags & HF_RF_MASK) {
        gen_helper_reset_rf();
    }

 What are these actions for? When I remove any of them the booting process
is OK.
 Please help me to understand interrupts mechanism or give links where to
read about it.

 I apply my modifications to version 0.12.3 of Qemu. Please refer to it.

Pavel Dovgaluk

Re: [Qemu-devel] Deterministic replay

[Paul Brook]

> Hello.
> 
>  I am working on implementation of deteministic execution replay technology
> for Qemu. It should be similar to VMWare's replay debugging.
> 
>  To make alarm timer (which invokes host_alarm_handler function)
> determinisic, I changed it's behavior: it sets flag, that execution should
> be stopped and this flag is checked before every instruction in the
> translated code.

You don't need to do this. A much better solution is to not use the host timer 
at all. See -icount.

Paul

RE: [Qemu-devel] Deterministic replay

[Pavel Dovgaluk]

> >  I am working on implementation of deteministic execution replay
> technology
> > for Qemu. It should be similar to VMWare's replay debugging.
> >
> >  To make alarm timer (which invokes host_alarm_handler function)
> > determinisic, I changed it's behavior: it sets flag, that execution
> should
> > be stopped and this flag is checked before every instruction in the
> > translated code.
> 
> You don't need to do this. A much better solution is to not use the
> host timer
> at all. See -icount.

 Thank you for your reply. 
 I know, that there is a virtual timers, that are enabled by -icount option
and can be used to get rid of host timers usage.
 But the problem is different - I need to synchronize alarm thread, which
breaks the execution of guest code to allow processing interrupts and
interaction with VNC/GDB/...
 Events caused by alarm thread are non-deterministic and asynchronous. To
save these events in the execution log (for latter replay) I need to
synchronize them with execution of guest code.
 The way which I am using for it is allowing execution stop at any point of
the guest code.


Pavel Dovgaluk