Rafael Vanoni wrote: > Edward Shu wrote: >> Randy Fishel: >>> I am sponsoring this case for Rafael Vanoni and Aubrey Li. It proposes >>> a kstat module for observing turbo mode on Intel processors. It >>> proposes >>> a micro/patch binding and a commited module. I believe that this case >>> qualifies as a self-review, should someone request, it can be turned >>> into a fasttrack. >>> >>> >>> Template Version: @(#)sac_nextcase %I% %G% SMI >>> This information is Copyright 2009 Sun Microsystems >>> 1. Introduction >>> 1.1. Project/Component Working Name: >>> Turbo mode observability >>> 1.2. Name of Document Author/Supplier: >>> Author: Rafael Vanoni >>> 1.3 Date of This Document: >>> 12 February, 2009 >>> 4. Technical Description >>> This RFE proposes to integrate observability features for Intel's >>> "Turbo Mode" feature. This functionality improves system performance >>> by raising the operating frequency of a subset of the system's cores >>> when at least half of the total amount of cores are idle and there's >>> is thermal headroom to do so. >> I think the transition is controlled by some firmware, and the OS can >> only read the state instead of controlling this behavior. How often >> does the OS read the register to refresh the kstats? >> > > Yes, the paragraph is describing the behavior of the feature, not > implying that the OS controls it. The registers are read on demand by > kstat or its consumers.
Correction, Aubrey's implementation reads the registers at every P-state transition if the system supports turbo mode. A kstat consumer may also cause the registers to be read if the CPU is in the highest P-state (P0), so that we have the most up to date value. Rafael
