[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <4739CD03.90406@qumranet.com>
Date: Tue, 13 Nov 2007 18:12:51 +0200
From: Avi Kivity <avi@...ranet.com>
To: "Dong, Eddie" <eddie.dong@...el.com>
CC: Glauber de Oliveira Costa <gcosta@...hat.com>,
linux-kernel@...r.kernel.org, jeremy@...p.org, hollisb@...ibm.com,
kvm-devel@...ts.sourceforge.net
Subject: Re: [kvm-devel] [PATCH 2/3] kvmclock - the host part.
Dong, Eddie wrote:
>>>
>>> After thinking for a little while, you are theoretically right.
>>> In the current state, we could even be preempted between all
>>> operations ;-) Maybe after avi's suggestion of moving the call to it
>>> it will end up in a preempt safe region, but anyway, it's safer to
>>> add the preempt markers here. I'll put it in next version, thanks
>>>
>>>
>>>
>> Well, you can't kvm_write_guest() with preemption enabled.
>>
>> preempt notifiers to the rescue! We have a callout during preemption,
>> so you can just zero out a flag there, and when we're scheduled again
>> retry the whole thing.
>>
>>
>
> The preemption issue is within following code which need to be done in a
> short enough period.
>
> + kvm_get_msr(vcpu, MSR_IA32_TIME_STAMP_COUNTER,
> + &vcpu->hv_clock.last_tsc);
> +
> + ktime_get_ts(&ts);
> + vcpu->hv_clock.now_ns = ts.tv_nsec + (NSEC_PER_SEC *
> (u64)ts.tv_sec);
> + vcpu->hv_clock.wc_sec = get_seconds();
>
> I am even thinking we have to disable interrupt between these lines,
> otherwise
> guest wall clock may see backward time source when calculating the
> delta TSC since last vcpu->hv_clock.now_ns update.
>
That's true. While we do need to handle vcpu migration and
descheduling, the code sequence you note needs to be as atomic as possible.
--
error compiling committee.c: too many arguments to function
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
Powered by blists - more mailing lists