Re: [PATCH v2] KVM: halt-polling: poll if emulated lapic timer will fire soon

[Yang Zhang]

On 2016/5/24 9:16, David Matlack wrote:
> On Mon, May 23, 2016 at 6:13 PM, Yang Zhang <yang.zhang.wz@xxxxxxxxx> wrote:
> > On 2016/5/24 2:04, David Matlack wrote:
> > > On Sun, May 22, 2016 at 6:26 PM, Yang Zhang <yang.zhang.wz@xxxxxxxxx>
> > > wrote:
> > > > On 2016/5/21 2:37, David Matlack wrote:
> > > > > It's not obvious to me why polling for a timer interrupt would improve
> > > > > context switch latency. Can you explain a bit more?
> > > >
> > > >
> > > >
> > > > We have a workload which using high resolution timer(less than 1ms)
> > > > inside
> > > > guest. It rely on the timer to wakeup itself. Sometimes the timer is
> > > > expected to fired just after the VCPU is blocked due to execute halt
> > > > instruction. But the thread who is running in the CPU will turn off the
> > > > hardware interrupt for long time due to disk access. This will cause the
> > > > timer interrupt been blocked until the interrupt is re-open.
> > >
> > >
> > > Does this happen on the idle thread (swapper)? If not, halt-polling
> > > may not help; it only polls if there are no other runnable threads.
> >
> >
> > Yes, there is no runnable task inside guest.
>
> Sorry for the confusion, my question was about the host, not the
> guest. Halt-polling only polls if there are no other runnable threads
> on the host CPU (see the check for single_task_running() in
> kvm_vcpu_block()).

ok, i see your point. I didn't check it before. But i have observed that 
the host CPU may wake up immediately after entering the power idle. Then 
another task takes the CPU until the schedule signal arrived. And this 
will cause timer injection delay by several ms.

> > > > For optimization, we let VCPU to poll for a while if the next timer will
> > > > arrive soon before schedule out. And the result shows good when running
> > > > several workloads inside guest.
> > >
> > >
> > > Thanks for the explanation, I appreciate it.
> > >
> > > > --
> > > > best regards
> > > > yang
> >
> >
> >
> > --
> > best regards
> > yang


--
best regards
yang