Re: [PATCH 2/2] x86/idle: use dynamic halt poll
From: Yang Zhang
Date: Thu Jun 22 2017 - 23:58:41 EST
On 2017/6/22 19:51, Paolo Bonzini wrote:
On 22/06/2017 13:22, root wrote:
==============================================================
+poll_grow: (X86 only)
+
+This parameter is multiplied in the grow_poll_ns() to increase the poll time.
+By default, the values is 2.
+
+==============================================================
+poll_shrink: (X86 only)
+
+This parameter is divided in the shrink_poll_ns() to reduce the poll time.
+By default, the values is 2.
Even before starting the debate on whether this is a good idea or a bad
idea, KVM reduces the polling value to the minimum (10 us) by default
I noticed it. It looks like the logic inside KVM is more reasonable. I
will do more testing to compare the two.
when polling fails. Also, it shouldn't be bound to
CONFIG_HYPERVISOR_GUEST, since there's nothing specific to virtual
machines here.
Yes. The original idea to use CONFIG_HYPERVISOR_GUEST because this
mechanism will only helpful inside VM. But as Thomas mentioned on other
thread it is wrong to use it since most distribution kernel will set it
to yes and still affect the bare metal. I will integrate it with
paravirtualizaion part as you suggested in below.
Regarding the good/bad idea part, KVM's polling is made much more
acceptable by single_task_running(). At least you need to integrate it
with paravirtualization. If the VM is scheduled out, you shrink the
polling period. There is already vcpu_is_preempted for this, it is used
by mutexes.
I have considered single_task_running() before. But since there is no
such paravirtual interface currently and i am not sure whether it is a
information leak from host if introducing such interface, so i didn't do
it. Do you mean vcpu_is_preempted can do the same thing? I check the
code and seems it only tells whether the VCPU is scheduled out or not
which cannot satisfy the needs.
Paolo
--
Yang
Alibaba Cloud Computing