Re: [kvm-devel] [PATCH 0/2][KVM] guest time accounting

Avi Kivity wrote:

Laurent Vivier wrote:

quoted

The aim of these two patches is to measure the CPU time used by a virtual
machine. All comments are welcome... I'm not sure it's the good way to
do that.

[PATCH 1/2] introduce a new field, "guest", in cpustat to store the
time used by
the CPU to run virtual CPU. Modify /proc/stat to display this new field.

[PATCH 2/2] modify account_system_time() to add cputime to
cpustat->guest if we
are running a VCPU. We add this cputime to  cpustat->user instead of
cpustat->system because this part of KVM code is in fact user code
although it
is executed in the kernel. We duplicate VCPU time between guest and
user to
allow an unmodified "top(1)" to display correct value. A modified
"top(1)" is
able to display good cpu user time and cpu guest time by subtracting
cpu guest
time from cpu user time.
  

[copying Ingo and Rusty]

The patches look good.  A couple of comments:

- perhaps the new fields should be guarded by a #ifdef CONFIG_HYPERVISOR
(selected by CONFIG_KVM)?  that way the (minor) additional overhead is
only incurred if it can possibly be used.  I imagine that our canine
cousin will want to use this as well.

There is also a CONFIG_VIRTUALIZATION and a CONFIG_VIRT_CPU_ACCOUNTING (from
s390 and powerpc) Which one to use ?

I'm wondering if we can have a more accurate accounting:

- For the moment we add all system time since the previous entering to the VCPU
to the guest time (and I guess there is some real system time in it ???)

- Perhaps we can sum nanoseconds spent in the VCPU and add it to cpustat when
these ns are greater than 1 ms ? (I'm trying to make something in this way)

- I think that there is per-task accounting of user time and system
time; that should be extended as well.

it should be easy to do too...

Laurent
-- 
------------- Laurent.Vivier@bull.net  --------------
          "Software is hard" - Donald Knuth