[Qemu-devel] [RFC] Undeterministic behaviour with icount.

[Qemu-devel] [RFC] Undeterministic behaviour with icount.

[Frederic Konrad]

[-- Attachment #1: Type: text/plain, Size: 547 bytes --]

Hi everybody,

We get some issues with reverse execution caused by indeterminism.

Something catched our attention:
static void icount_warp_rt(void *opaque), cpus.c:276

We have the feeling that icount is synchronized with rt_clock, is that 
possible?

According to this Paolo's series 
<http://lists.gnu.org/archive/html/qemu-devel/2011-04/msg01271.html> 
this must be called when the cpus are sleeping, but
I saw this called frequently during the execution, is that the expected 
behaviour?
If not what's the best way to fix that?

Thanks,
Fred

[-- Attachment #2: Type: text/html, Size: 851 bytes --]

Re: [Qemu-devel] [RFC] Undeterministic behaviour with icount.

[Paolo Bonzini]

Il 08/07/2013 19:32, Frederic Konrad ha scritto:
> Hi everybody,
> 
> We get some issues with reverse execution caused by indeterminism.
> 
> Something catched our attention:
> static void icount_warp_rt(void *opaque), cpus.c:276
> 
> We have the feeling that icount is synchronized with rt_clock, is that
> possible?

When the CPU is idle, yes.

Note that the same thing happened before the series you linked.  This is
the relevant code:

-        /* Wait for either IO to occur or the next
-           timer event.  */
-        add = qemu_next_deadline();
-        /* We advance the timer before checking for IO.
-           Limit the amount we advance so that early IO
-           activity won't get the guest too far ahead.  */
-        if (add > 10000000)
-            add = 10000000;
-        delta += add;
-        qemu_icount += qemu_icount_round (add);

It was adding to the icount while waiting for the next timer event to
happen.

> According to this Paolo's series
> <http://lists.gnu.org/archive/html/qemu-devel/2011-04/msg01271.html>
> this must be called when the cpus are sleeping, but
> I saw this called frequently during the execution, is that the expected
> behaviour?

What workload are you running?  For anything that is not CPU bound I'd
expect that to be the case.

Paolo

> If not what's the best way to fix that?

Re: [Qemu-devel] [RFC] Undeterministic behaviour with icount.

[Frederic Konrad]

On 10/07/2013 15:45, Paolo Bonzini wrote:
> Il 08/07/2013 19:32, Frederic Konrad ha scritto:
>> Hi everybody,
>>
>> We get some issues with reverse execution caused by indeterminism.
>>
>> Something catched our attention:
>> static void icount_warp_rt(void *opaque), cpus.c:276
>>
>> We have the feeling that icount is synchronized with rt_clock, is that
>> possible?
> When the CPU is idle, yes.
>
> Note that the same thing happened before the series you linked.  This is
> the relevant code:
>
> -        /* Wait for either IO to occur or the next
> -           timer event.  */
> -        add = qemu_next_deadline();
> -        /* We advance the timer before checking for IO.
> -           Limit the amount we advance so that early IO
> -           activity won't get the guest too far ahead.  */
> -        if (add > 10000000)
> -            add = 10000000;
> -        delta += add;
> -        qemu_icount += qemu_icount_round (add);
>
> It was adding to the icount while waiting for the next timer event to
> happen.
>
>> According to this Paolo's series
>> <http://lists.gnu.org/archive/html/qemu-devel/2011-04/msg01271.html>
>> this must be called when the cpus are sleeping, but
>> I saw this called frequently during the execution, is that the expected
>> behaviour?
> What workload are you running?  For anything that is not CPU bound I'd
> expect that to be the case.
>
> Paolo
>
>> If not what's the best way to fix that?
Hi Paolo,

Thanks for your answer.

For the moment the workload is a kernel boot. We removed everything from the
board.

So if it's the case, vm_clock will run non deterministic and we can't 
replay.

I just send a series about that.

Fred