Description

SPARC T3-4 systems running Solaris 10 with all the processors in a single CPU partition may report system hangs with a large backlog of clock cyclic. T3-4 systems have 512 VCPUs and by default all of the CPUs belong to a single CPU partition.

Occurrence

This issue can occur in the following releases:

SPARC Platform

• Solaris 10 with patch 137111-03 or later

Note 1: This issue only occurs on SPARC T3-4 systems with all the processors in a single CPU partition. To determine the number of CPU partitions in the system, the following command can be used:

    > echo ::cpupart | mdb -k
    ID ADDR #CPU CPUS
    0 198fcb0 256 0-256

Note 2: ZFS Storage Appliance (ZFSSA) is not impacted by this issue.

Note 3: Solaris 11 and Solaris on the x86 platform are not impacted by this issue.

Note 4: Solaris 8 and Solaris 9 will not be evaluated regarding the potential impact of the issue described in this document.

Symptoms

 If the described issue occurs, the system will hang. The ::cpuinfo on CPU0 and the thread stack from the crash dump can be verified as shown below:

   >  0::cpuinfo -v 
   ID ADDR        FLG NRUN BSPL PRI RNRN KRNRN SWITCH THREAD      PROC 
    0 0000180c000  1d    1    0 169   no    no t-175623 2a100047c80 sched 
                    |    |    | 
         RUNNING <--+    |    +--> PIL THREAD 
        QUIESCED         |          10 2a100047c80 
          EXISTS         |           - 3028684f500 java 
          ENABLE         | 
                         +-->  PRI THREAD      PROC 
                                99 2a1005d7c80 sched 
 
    > 2a100047c80::findstack 
    stack pointer for thread 2a100047c80: 2a100046e31 
    [ 000002a100046e31 panic_idle+0x14() ] 
      000002a100046ee1 ktl0+0x64() 
      000002a100047031 clock+0x154() 
      000002a100047121 cyclic_softint+0xac() 
      000002a1000471e1 cbe_level10+8() 
      000002a100047291 intr_thread+0x25c() 
      000002a11c3ad181 lwp_unpark+0x44() 
      000002a11c3ad231 syslwp_park+0x64() 
      000002a11c3ad2e1 syscall_trap+0xac()

Workaround

 To work around this issue, create a processor set on cpu0, using the following command:

    # psrset -c 0

The workaround above is the final resolution to this issue. 

History

 06-Jun-2017: Document released, status Resolved

Internal Section: Comments:

Questions regarding this document should be addressed to
sunalertpublication_us_grp@oracle.com and copy the
submitter/responsible engineer listed below:

Internal Contributor/Submitter: vamsi.nagineni@oracle.com
Internal Eng Responsible Engineer: vamsi.nagineni@oracle.com
Oracle Knowledge Analyst: jeff.folla@oracle.com
Internal Eng Business Unit Group: Systems RPE
Internal Associated SR IDs: 3-13808674621
Internal Pending Patches:
Internal Resolution Patches:

References

This solution has no attachment