Collect real process and processor utilization values when virtualization is enabled.

Jeff Scheel scheel at vnet.ibm.com
Wed Jan 12 19:47:23 EST 2005


On Wed, 2005-01-12 at 04:06, Olof Johansson wrote:
> > The value is needed by CKRM to do further calculations.
> 
> How will CKRM use this? Does it have architecture-specific code to
> dig this out of the thread_struct again? Could they use the cputime
> interface if we hooked into that instead?
The only interface which exists today to retrieve purr is
/proc/ppc64/lparconfig which provides PURR summed across all processors.
We are working on other means to retrieve more specific data in the near
future.

> Finally: There's two ways to read PURR on our platform: One is to read
> the SPR value, the other to get it from the hypervisor via the H_PURR
> call. Do they measure the same thing and stay consistent?
Olof, you are correct.  We'll want to go directly to the hardware and
avoid the overhead of a hypervisor call.  Only is the instance where the
hypervisor is emulating PURR will we want to use the hypervisor call. 
The "art" is detecting when/if that is occurring.  Dave E. should be
able to help us with this.
-- 
Jeff Scheel (scheel at vnet.ibm.com)




More information about the Linuxppc64-dev mailing list