Note: This bug is displayed in read-only format because the product is no longer active in Red Hat Bugzilla.

Bug 1071015

Summary: RHEL7 CPUScaling Turbo Boost Failure
Product: [Retired] Red Hat Hardware Certification Program Reporter: HP Enterprise Cert team <redhat_cert>
Component: Test Suite (tests)Assignee: Greg Nichols <gnichols>
Status: CLOSED DUPLICATE QA Contact: Red Hat Kernel QE team <kernel-qe>
Severity: unspecified Docs Contact:
Priority: unspecified    
Version: 1.7.0CC: qcai, rlandry
Target Milestone: ---   
Target Release: ---   
Hardware: Unspecified   
OS: Unspecified   
Whiteboard:
Fixed In Version: Doc Type: Bug Fix
Doc Text:
Story Points: ---
Clone Of: Environment:
Last Closed: 2014-02-28 14:41:52 UTC Type: Bug
Regression: --- Mount Type: ---
Documentation: --- CRM:
Verified Versions: Category: ---
oVirt Team: --- RHEL 7.3 requirements from Atomic Host:
Cloudforms Team: --- Target Upstream Version:
Embargoed:
Attachments:
Description Flags
fail48.xml.gz
none
fail24.xml.gz
none
pass16.xml.gz
none
cpuinfo.log none

Description HP Enterprise Cert team 2014-02-27 22:45:44 UTC
Created attachment 868771 [details]
fail48.xml.gz

Description of problem:
On RHEL7 snapshot 7 with the hwcert-client-1.7.0-20140212 cert kit the cpuscaling turbo boost frequency portion of the test fails b/c 3.20GHz is not achieved on the 24 threads tested. According to the Intel doc for the E5-2697v2 2.70GHz processor the 3.20GHz turbo boost frequency can only be achieved on 4 cores (8 threads) which is reported in the cpuscaling log as follows:
  boost state support:
    Supported: yes
    Active: yes
    3200 MHz max turbo 4 active cores
    3300 MHz max turbo 3 active cores
    3400 MHz max turbo 2 active cores
    3500 MHz max turbo 1 active cores
Per the Intel E5-2697v2 2.70GHz processor doc the above turbo boost frequency and core counts are correct.

However, the test is trying to achieve the 3.20GHz turbo boost frequency on 24 threads seen by the following log errors on each of the 24 threads: (not all listed here, see fail48.xml.gz)
    User Space (max): Error: cpu0 in package 0 has a measured frequency of 2993 MHz vs. a requirement of 3200 MHz (- 50 MHz tolerance)
    User Space (max): Error: cpu1 in package 0 has a measured frequency of 2993 MHz vs. a requirement of 3200 MHz (- 50 MHz tolerance)
    User Space (max): Error: cpu2 in package 0 has a measured frequency of 2992 MHz vs. a requirement of 3200 MHz (- 50 MHz tolerance)
    User Space (max): Error: cpu3 ...

If the core count is reduced via RBSU so the OS only sees 16 threads then the cpuscaling test will pass (see pass16.xml.gz), as the 3.20GHz turbo boost frequency test is performed on 8 threads which is valid.

If the core count is increased where the OS sees 24 threads, the cpuscaling test still fails b/c the 3.20GHz turbo boost frequency test is being done on 12 threads which is not valid. (see fail24.xml.gz)

During the investigation of this issue I also found the max frequency to be incorrect for the E5-2697v2 2.70GHz 12c processor.
The cpuscaling log reports the following for each thread:
    CPU: 0
    ----------
               min: 1200 MHz
               max: 3200 MHz
         min boost: 3200 MHz
         max boost: 3500 MHz
         governors: performance, powersave, performance, powersave

However, according to the Intel doc the max core frequency (P1) is 2.70GHz not 3.20GHz.

Version-Release number of selected component (if applicable):
hwcert-client-1.7.0-20140212 cert kit
RHEL7 snapshot 7

How reproducible: always


Steps to Reproduce:
1. Use an Intel server with procs that support turbo boost and a minimum of 24
   threads.
2. Install RHEL7 and the cert kit mentioned above.
3. Run the cpuscaling test

Actual results: The cpuscaling test fails b/c it cannot achieve the 3.2GHz turbo boost frequency on 24 threads, but according to the Intel doc 3.2GHz can only be achieved on 4 cores (8 threads).

Expected results: The 3.2GHz turbo boost frequency test should only be done on 8 threads.

Additional info: For details see the attached fail48.xml.gz, fail24.xml.gz, pass16.xml.gz and the cpuinfo.log files.

Comment 1 HP Enterprise Cert team 2014-02-27 22:46:33 UTC
Created attachment 868772 [details]
fail24.xml.gz

Comment 2 HP Enterprise Cert team 2014-02-27 22:47:20 UTC
Created attachment 868773 [details]
pass16.xml.gz

Comment 3 HP Enterprise Cert team 2014-02-27 22:48:01 UTC
Created attachment 868774 [details]
cpuinfo.log

Comment 4 Greg Nichols 2014-02-28 14:41:52 UTC
The patches for bug 1065571 change the tests' expected frequency for the fully-loaded case to that stated in the "model" string of /proc/cpuinfo, rather than the output of cpupower.

*** This bug has been marked as a duplicate of bug 1065571 ***