Bug 814294

Summary: numa_all_cpus_ptr reports the wrong thing when bound to cpu0
Product: Red Hat Enterprise Linux 6 Reporter: Travis Gummels <tgummels>
Component: numactlAssignee: Petr Holasek <pholasek>
Status: CLOSED ERRATA QA Contact: qe-baseos-daemons
Severity: medium Docs Contact:
Priority: medium    
Version: 6.2CC: asersen, azelinka, dhoward, mpoole, woodard
Target Milestone: rc   
Target Release: ---   
Hardware: All   
OS: Linux   
Whiteboard:
Fixed In Version: numactl-2.0.7-4.el6 Doc Type: Bug Fix
Doc Text:
Cause: global pointer numa_all_cpus_ptr contained all possible CPU0 when taskset was set to cpu 0. Consequence: Wrong condition in code parsing /proc/self/stat Fix: Codepath fixed. Result: numa_all_cpus_ptr contains only specified CPUs, when taskset contains cpu 0.
 Story Points: ---
Clone Of: Environment:
Last Closed: 2013-02-21 10:11:27 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:
Bug Depends On:    
Bug Blocks: 782183, 787802, 840699    
Attachments:
Description Flags
Reproducer
none
Proposed Patch none

Description Travis Gummels 2012-04-19 14:27:59 UTC 
Description of problem:

There is a global pointer named 'numa_all_cpus_ptr' which is broken in that it doesn't function as documented.  It is supposed to be set to ' a bitmask that is allocated by the library with bits representing all cpus on which the calling  thread  may  execute.  However, this seems to not work when the bitmask is set to only CPU0

$./t
numa_all_cpus_ptr: 1111111111111111111111111111111100
Number of CPUS in numa_all_cpus_ptr: 32

$ taskset -c 4 ./t
numa_all_cpus_ptr: 0000100000000000000000000000000000
Number of CPUS in numa_all_cpus_ptr: 1

$ taskset -c 0 ./t
numa_all_cpus_ptr: 1111111111111111111111111111111100
Number of CPUS in numa_all_cpus_ptr: 32

$ taskset -c 1 ./t
numa_all_cpus_ptr: 0100000000000000000000000000000000
Number of CPUS in numa_all_cpus_ptr: 1

Note that the taskset -c0 ./t case should only have 1 bit set in the 0th position.  So it must be the code that parses /proc/self/stat Cpus_allowed.

Version-Release number of selected component (if applicable):
numactl-2.0.3-9.el6

How reproducible:
100%

Steps to Reproduce:
1.  See above
  
Actual results:
Broken

Expected results:
Not Broken

Additional info:
LLNL has requested this bug be public.
Comment 1 Travis Gummels 2012-04-19 14:32:52 UTC 
Created attachment 578663 [details]
Reproducer

gcc -o t -lnuma t.c
Comment 2 Travis Gummels 2012-04-19 14:34:30 UTC 
Created attachment 578665 [details]
Proposed Patch

commit 346e427204fba80e86d6ebd8965610b89dc5c7d6
Author: Mark A. Grondona <mgrondona>
Date:   Tue Mar 27 15:51:28 2012 -0700

    libnuma: fix numa_all_cpus_ptr when bound to cpu0
    
    A bug in the test for invalid maxproccpu leads to resetting the
    numa_all_cpus_ptr mask when the cpu affinity mask of the current
    task only contains CPU0. This leads to numa_all_cpus_ptr having
    all CPUs instead of just one.
    
    Fix the test to be < 0 instead of <= 0 -- 0 is a valid CPU number.
Comment 8 errata-xmlrpc 2013-02-21 10:11:27 UTC 
Since the problem described in this bug report should be
resolved in a recent advisory, it has been closed with a
resolution of ERRATA.

For information on the advisory, and where to find the updated
files, follow the link below.

If the solution does not work for you, open a new bug report.

http://rhn.redhat.com/errata/RHBA-2013-0401.html