AR# 71915

|

Design Advisory for Alveo Data Center Accelerator Card - Golden Image can become corrupted

Description

The Alveo Data Center Accelerator Card is manufactured with a Golden Image in the QSPI device. 

This enables the card to initially load in a system where no Shell is programmed on the card yet and provides a safe failback image should the Shell or Customer Image become corrupted.   

 

The QSPI is write-protected and locked in this region to ensure that the Golden Image is always available. 

On Alveo Data Center Accelerator Cards (U200 / U250) manufactured up to the "Version Fixed" part numbers, the QPSI write-protection was not covering all required sectors and allowed partial overwrite of the Golden Image.

This can cause the following problems:


Problem 1:

If the Card is updated with an image written to the wrong location, both images will be corrupted.  

Upon cold power cycle, the Alveo Data Card Accelerator Card will no longer be functional. 


Version Found:

Alveo U200 and U250 Alveo Data Center Accelerator Cards.


Problem 2:

When loading the Shell onto cards per the instructions in (UG1301) Getting Started with Alveo Accelerator Cards (v1.1 or earlier) via Xilinx Runtime (XRT), if multiple cards are updated via a single command, cards 2 to N will become corrupted. 

This can occur under the following conditions:

1) Two or more U200 or U250 cards are present in the system, and the following command is used:

sudo /opt/xilinx/xrt/bin/xbutil flash -a <shell_name> -t <timestamp>

 2) Mixed U200 or U250 cards are present in the system, both DSA packages are installed, and the following command is used:

sudo /opt/xilinx/xrt/bin/xbutil flash -a all 

 XRT will output that all updates are successful; however, after a cold boot of the system, cards 2 N will no longer be present in the system.

Single card updates are not affected by this issue.

 

Version Found:

  • XRT_201830.2.1.1695
  • XRT_201802.2.1.83



For more information on how to sign up to receive notifications of new Design Advisories, see (Xilinx Answer 18683).

For additional Known Issues related to Xilinx Runtime (XRT), see (Xilinx Answer 71628)

For additional Known Issues related to Alveo Data Center Accelerator Card, see (Xilinx Answer 71752)

Solution

A fix is now available to update the Alveo Data Center Accelerator Card QSPI write-protection, in order to ensure that the Golden Image cannot be corrupted in a future upgrade. 

Additionally, if you have experienced corruption of the Alveo Data Center Accelerator Card, this is the procedure for recovery.



Problem 1:

The procedure and downloads to correctly lock the QSPI device for a card manufactured prior to the fix, and to recover a corrupted Alveo Data Center Accelerator Card are available in (Xilinx Answer 71916).  

Card Version Fixed:

Board
Current Revision (043 Sticker)
AU200A64G0432129-04
AU200P64G0432130-03
AU250A64G
0432132-03
AU250P64G0432133-02
 

Problem 2:

The following work-around and/or fix will not provide the write-protect correction to the Alveo Data Center Accelerator Card QSPI, but it will prevent overwrite of the unprotected region of Golden Image, and ensure that the Shell / Custom Image is programmed at the correct location. 

Please use the work-around with all affected versions of Xilinx Runtime (XRT) until a version is released where the issue is fixed.

Alveo Data Center Accelerator Card Deployment Shell or Development Shell flow:

Work-around for Problem 2:

For use with XRT_201830.2.1.1695 and XRT_201802.2.1.83.

Use only the following command to update your cards via XRT, updating each card individually:

sudo /opt/xilinx/xrt/bin/xbutil flash -a <shell_name> -d <card_number> 

To determine you card numbers, the following command can be used:

sudo /opt/xilinx/xrt/bin/xbutil flash scan 

Fix for Problem 2:

Xilinx Runtime (XRT) version xrt_201830.2.1.1712 has been released with this problem resolved.  

Please use this version of XRT to do any multi-card flashing.  If an older version must be used for compatibility, please continue to use the work-around for Problem 2. 


Custom Flow (Vivado Board Aware or Custom Shell)

Ensure that the QSPI image is created correctly by following the guidance in (Xilinx Answer 71756)


If you have any questions on the above, please contact Xilinx Technical Support.

https://www.xilinx.com/support.html

Revision History:

  • 01/20/2019 - Initial release
  • 02/18/2019 - Updated with XRT version for Problem 2 fix

Linked Answer Records

Master Answer Records

Associated Answer Records

Answer Number Answer Title Version Found Version Resolved
71629 Xilinx Runtime version xrt_201802.2.1.83 Known Issues N/A N/A
71861 Xilinx Runtime version xrt_201830.2.1.1695 Known Issues N/A N/A
AR# 71915
Date 05/15/2019
Status Active
Type Design Advisory
Boards & Kits
People Also Viewed