Intel T1600 LF80537NF0281MN User Manual

Product codes
LF80537NF0281MN
Page of 40
R
 
 
Specification Update  
15 
 
Errata 
V1. 
I/O Restart in SMM may Fail after Simultaneous Machine Check Exception (MCE) 
Problem: 
If an I/O instruction (IN, INS, REP INS, OUT, OUTS, or REP OUTS) is being executed, and if the data 
for this instruction becomes corrupted, the processor will signal a Machine Check Exception (MCE). If 
the instruction is directed at a device that is powered down, the processor may also receive an assertion 
of SMI#. Since MCEs have higher priority, the processor will call the MCE handler, and the SMI# 
assertion will remain pending. However, upon attempting to execute the first instruction of the MCE 
handler, the SMI# will be recognized and the processor will attempt to execute the SMM handler. If the 
SMM handler is completed successfully, it will attempt to restart the I/O instruction, but will not have 
the correct machine state, due to the call to the MCE handler.
 
Implication:  A simultaneous MCE and SMI# assertion may occur for one of the I/O instructions above. The SMM 
handler may attempt to restart such an I/O instruction, but will have an incorrect state due to the MCE 
handler call, leading to failure of the restart and shutdown of the processor. 
Workaround: If a system implementation must support both SMM and board I/O restart, the first thing the SMM 
handler code should do is check for a pending MCE. If there is an MCE pending, the SMM handler 
should immediately exit via an RSM instruction and allow the MCE handler to execute. If there is no 
MCE pending, the SMM handler may proceed with its normal operation. 
Status: 
For the steppings affected, see the Summary Tables of Changes
 
V2. 
MCA Registers May Contain Invalid Information if RESET# Occurs and 
PWRGOOD Is Not Held Asserted 
Problem: 
This erratum can occur as a result either of the following events:  
•  PWRGOOD is de-asserted during a RESET# assertion causing internal glitches that may result in 
the possibility that the MCA registers latch invalid information. 
•  Or during a reset sequence if the processor’s power remains valid regardless of the state of 
PWRGOOD, and RESET# is re-asserted before the processor has cleared the MCA registers, the 
processor will begin the reset process again but may not clear these registers. 
Implication:  When this erratum occurs, the information in the MCA registers may not be reliable. 
Workaround: Ensure that
 
PWRGOOD remains asserted throughout any RESET# assertion and that RESET# is not re-
asserted while PWRGOOD is de-asserted. 
Status: 
For the steppings affected, see the Summary Tables of Changes