Re: [Xen-devel] Re: [PATCH 3/5] x86/pvclock: add vsyscall implementation

[Jeremy Fitzhardinge <jeremy@goop.org>]

On 10/12/09 11:29, Avi Kivity wrote:
> Good catch.  Doesn't that invalidate rdtscp based vgettimeofday on
> non-virt as well (assuming p == cpu)?

The tsc clocksource assumes the tsc is (mostly?) synced; it doesn't use
rdtscp or make any attempt at per-cpu corrections.

>> I suppose that works if you assume that:
>>
>>     1. every task->vcpu migration is associated with a hv/guest context
>>        switch, and
>>     2. every hv/guest context switch is a write barrier
>>
>> I guess 2 is a given, but I can at least imagine cases where 1 might not
>> be true.  Maybe.  It all seems very subtle.
>>    
>
> What is 1 exactly?  task switching to another vcpu?  that doesn't
> incur hypervisor involvement.  vcpu moving to another cpu?  That does.
Aie... OK.  So no barrier is required for a task double migration on
vcpus, because it ends up on the same pcpu and the ordering is local; if
there's a vcpu migration to a new pcpu in there too, then we always
expect a barrier.

>> And I don't really see a gain.  You avoid maintaining a second version
>> number, but at the cost of two rdtscps.  In my measurements, the whole
>> vsyscall takes around 100ns to run, and a single rdtsc takes about 30,
>> so 30% of total.  Unlike rdtsc, rdtscp is documented as being ordered in
>> the instruction stream, and so will take at least as long; two of them
>> will completely blow the vsyscall execution time.
>>    
>
> I agree, let's stick with the rdtscpless implementation.

OK, I'll use PeterZ's hint to try and find a more complete set of
migration points.

    J