Sun Microsystems, Inc.  Sun System Handbook - ISO 3.4 June 2011 Internal/Partner Edition
   Home | Current Systems | Former STK Products | EOL Systems | Components | General Info | Search | Feedback

Asset ID: 1-73-1001194.1
Update Date:2010-08-25
Keywords:

Solution Type  FAB (standard) Sure

Solution  1001194.1 :   Applications using large memory page sizes on UltraSPARC IV+ CPUs may hang or core dump.  


Related Items
  • Sun Fire E6900 Server
  •  
  • Sun Fire E25K Server
  •  
  • Sun Fire E20K Server
  •  
  • Sun Fire V890 Server
  •  
  • Sun Fire E4900 Server
  •  
  • Sun Fire E2900 Server
  •  
  • Sun Fire V490 Server
  •  
Related Categories
  • GCS>Sun Microsystems>Sun FAB>Standard>Controlled Proactive
  •  

PreviouslyPublishedAs
201589


Product
Sun Fire E20K Server
Sun Fire E6900 Server
Sun Fire V490 Server
Sun Fire V890 Server
Sun Fire E2900 Server
Sun Fire E4900 Server
Sun Fire E25K Server

Bug Id
<SUNBUG: 6287398> <SUNBUG: 6313025>

Xoption
  • Xoption Number: X7274A
  • Xoption Description: V490/890 Processor module, 16GB
Xoption
  • Xoption Number: X7273A
  • Xoption Description: V490/890 Processor module, 8GB
Xoption
  • Xoption Number: A53-CLH8C264GTD
  • Xoption Description: A53,8X1500MHZ,64GB,4X146GB
Xoption
  • Xoption Number: A53-CLH8C232GTD
  • Xoption Description: A53,8X1500MHZ,32GB,4X146GB
Xoption
  • Xoption Number: A53-CLH4C216GTD
  • Xoption Description: A53,4X1500MHZ,16GB,4X146GB
Xoption
  • Xoption Number: A53-CLH2C216GTD
  • Xoption Description: A53,2X1500MHZ,16GB,4X146GB
Xoption
  • Xoption Number: A53-CLH2C208GTD
  • Xoption Description: A53,2X1500MHZ,8GB,4X146GB
Xoption
  • Xoption Number: A52-CLH4C232GTB
  • Xoption Description: A52,4X1500MHZ,32GB,2X146GB
Xoption
  • Xoption Number: A52-CLH4C216GTB
  • Xoption Description: A52,4X1500MHZ,16GB,2X146GB
Xoption
  • Xoption Number: A52-CLH2C216GTB
  • Xoption Description: A52,2X1500MHZ,16GB,2X146GB
Xoption
  • Xoption Number: A52-CLH2C208GTB
  • Xoption Description: A52,2X1500MHZ,8GB,2X146GB
XoptionXoption
  • Xoption Number: XE29BRD-482-1500C
  • Xoption Description: E2900 COD BRD 1.5GHZ 4P16GB
XoptionXoption
  • Xoption Number: XE29BRD-484-1500C
  • Xoption Description: E2900 COD BRD 1.5GHZ 4P32GB
Xoption
  • Xoption Number: XUS4BRD-484-1500C
  • Xoption Description: COD,CPU,4PROC@1500/32GB USS400
Xoption
  • Xoption Number: XUS4BRD-482-1500C
  • Xoption Description: COD,CPU,4PROC@1500/16GB USS400
Xoption
  • Xoption Number: XUS4BRD-484-1500
  • Xoption Description: ASY,CPU,4PROC@1500/32GB USS400
Xoption
  • Xoption Number: XUS4BRD-482-1500
  • Xoption Description: ASY,CPU,4PROC@1500/16GB USS400

Impact

Software applications using 32M or 256M memory page sizes, running on UltraSPARC IV+ CPUs, can hang or core dump.


Contributing Factors

This issue can occur if all of the following are true.

  • The system contains an UltraSPARC IV+ System Board
  • The installed OS is S10 3/05 HW1 without patch 118822-18 (KU18)
  • A software application utilizes 32M or 256M memory page sizes

This issue can be triggered by one of the following actions.

  1. An application explicitly uses memcntl() to take advantage of the 32M/256M page sizes for either heap or mmap'ed segments (/dev/zero), generates code into these segments, and tries to execute the code.

  2. The issue can also occur if an application places code into an ISM segment and tries to execute from there.  Since historically ISM segments have PROT_EXEC permission, a user might expect it to work.

In diagnosing this issue, it helps to identify if the application has used any Multiple Page Size Support (MPSS) options during invocation, or if the application calls memcntl(2) to use large pages internally.


Root Cause

The UltraSPARC IV+ IMMU does not support the 32M or 256M page sizes. If the processor cannot translate a virtual address to a physical address, it takes a trap known as a tlb miss.  Solaris will execute the appropriate trap handler and try to load the correct translation for that virtual address.  If a tlb miss occurs on an instruction which came from a segment mapped with either of these large pages, the corresponding application will execute improperly, most likely resulting in a core dump or hang.  

The final resolution for this issue (CR 6287398) is now available in patch 118822-20 or higher.  A partial resolution is available in patch 118822-18, which turns off USIV+ 32M and 256M page size support.


Resolution

If an application is core dumping on an UltraSPARC IV+ system, the following steps should be taken.

  1. Determine the OS and patches that are installed. (use "uname -a" and "showrev -p")
  2. Look at how the application was invoked, and also if any MPSS options were used during invocation (use "pmap -x -s <PID>").  It also helps to know if memcntl() is used inside the application to configure large pages.

  3. KU20 (118822-20) or later patches should be applied to resolve the issue.  KU18 (118822-18) will provide a partial fix by turning off large page size support.  Until either of these patches is applied, the following workaround can be used.

Workaround

Disable 32M/256M pages by setting the following variables in the /etc/system file.  You must reboot the system for the changes to take affect.

   set disable_large_pages = 0x30
   set disable_ism_large_pages = 0x34
   set mmu_ism_pagesize = 4194304

 

Note: If the above contributing factors are not present, this could be a different application issue.  During diagnosis, it can be helpful to look at the /etc/system file on the customer's system to see if any changes were made to this file.


Modification History
Date: 20-SEP-2005
  • Added E2900, E4900, E6900 platforms to Affected Products.

Date: 29-SEP-2005
  • Added E20K and E25K platforms to Affected Products.

Date: 27-OCT-2005
  • Added patch 118822-20 as the final resolution.


Previously Published As
101909
Internal Comments


None.


Internal Eng Business Unit Group
KE Authors

Internal Resolution Patches
118822-18 118822-20

Internal Kasp FAB Legacy ID
101909

Internal Sun Alert & FAB Admin Info
Critical Category:
Significant Change Date:
Avoidance: Patch
Responsible Manager: null
Original Admin Info: null

Product_uuid
1404a2d3-059a-11d8-84cb-080020a9ed93|Sun Fire E20K Server
4fe39727-0599-11d8-84cb-080020a9ed93|Sun Fire E6900 Server
5c71fc02-5e51-11d7-8add-8938754df22a|Sun Fire V490 Server
5d2816fe-5e51-11d7-8de2-d7bc0dd226fc|Sun Fire V890 Server
63ff680d-0598-11d8-84cb-080020a9ed93|Sun Fire E2900 Server
bed24aa9-0598-11d8-84cb-080020a9ed93|Sun Fire E4900 Server
d842dd03-059b-11d8-84cb-080020a9ed93|Sun Fire E25K Server

Attachments
This solution has no attachment
  Copyright © 2011 Sun Microsystems, Inc.  All rights reserved.
 Feedback