Intel E7-2820 AT80615007245AA User Manual

20

Intel

®

 Xeon

®

 Processor E7-8800/4800/2800 Product Families

September 2012

Specification Update

Workaround:

A BIOS code change has been identified and may be implemented as a workaround for 

this erratum.

Status:

For the steppings affected, see the 

Problem:

A read from the mirror master may complete before the write-back from the IOH 

completes. This will result in the IOH write-data not being immediately visible and can 

lead to the IOH write-data never becoming visible. In the case of a RdInvOwn 

transaction, the reading caching agent will take ownership which can then overwrite 

the IOH data. This is especially visible in false-sharing of cache lines which involve the 

IOH.

Implication:

Due to this erratum, correct data is not delivered by the mirror slave. This erratum only 

occurs during mirror failover.

Workaround:

It is possible for the BIOS to contain a workaround for this erratum.

Status:

For the steppings affected, see the 

Problem:

The UU_CR_U_MSR_PMON_GLOL_OVF_CTL MSR (C02H) is access type RW1C (Read 

Write 1 Clear) and when written with 1's should clear the corresponding bit in the 

UU_CR_U_MSR_PMON_ GLOL_STATUS MSR (C01H). Due to this erratum, a read of the 

UU_CR_U_MSR_PMON_GLOL_OVF_CTL MSR does not return zeros however a read of 

the UU_CR_U_MSR_PMON_ GLOL_STATUS MSR will show appropriate clearing.

Implication:

The UU_CR_U_MSR_PMON_GLOL_OVF_CTL MSR does not return zeros on a read.

Workaround:

None identified.

Status:

For the steppings affected, see the 

Problem:

The HNID (Home Node ID) field used in the PMON (Performance Monitoring) match/

mask is incorrect for CMP (complete) messages from PrefetchHint. The same incorrect 

HNID is logged in the event of an error condition on a CMP for a PrefetchHint in the 

caching agent. The logging of the RNID (Requester Node ID) in the error logs for NDR 

(Non Data Response) messages is incorrect and impacts the caching agent system 

bound errors.

Implication:

There are two implications:

1. Incorrect HNID is filtered or matched for CMPs using the caching agent PMON 

match/mask. 

2. The incorrect RNID will be logged only for errors on NDR messages.

Workaround:

There are two potential workarounds:

1. The Intel

®

 QPI performance monitor match/mask can be used to count different 

types of CMP messages from the caching agent.

2. Ignore the RNID field for NDR system bound messages.

Status:

For the steppings affected, see the 

Problem:

An unexpected Page Fault may occur when a logical processor transitions from C6 to 

C0, with IA-32e mode enabled.