Errata
32
Specification
Update
3.
Uncacheable (UC) Code in Same Line As Write Back (WB) Data May
Lead to Data Corruption
Problem:
When both code (being accessed as UC or WC) and data (being accessed as WB) are
aliased into the same cache line, the UC fetch will cause the processor to self-snoop
and generate an implicit writeback. The data supplied by this implicit writeback may
be corrupted due to the way the processor handles self-modifying code.
Implication:
UC or WC code located in the same cache line as WB data may lead to data
corruption
Workaround:
UC or WC code should not be located in the same physical 64 byte cache line as
any location that is being stored to with WB data.
Status:
For the steppings affected, see the
Summary Tables of Changes.
4.
Transaction Is Not Retried after BINIT#
Problem:
If the first transaction of a locked sequence receives a HITM# and DEFER# during the
snoop phase it should be retried and the locked sequence restarted. However, if
BINIT# is also asserted during this transaction, it will not be retried.
Implication:
When this erratum occurs, locked transactions will unexpectedly not be retried.
Workaround:
None identified.
Status:
For the steppings affected, see the
Summary Tables of Changes.
5.
Invalid Opcode 0FFFh Requires a ModRM Byte
Problem:
Some invalid opcodes require a ModRM byte (or other following bytes), while others
do not. The invalid opcode 0FFFh did not require a ModRM byte in previous
generation Intel architecture processors, but does in the Pentium 4 processor.
Implication:
The use of an invalid opcode 0FFFh without the ModRM byte may result in a page or
limit fault on the Pentium 4 processor.
Workaround:
Use a ModRM byte with invalid 0FFFh opcode.
Status:
For the steppings affected, see the
Summary Tables of Changes.
6.
RFO-ECC-Snoop-MCA Combination Can Result in Two Lines Being
Corrupted in Main Memory
Problem:
When a snoop comes into the processor between global observation and data return
for a Read-for-Ownership (RFO) request that hits an E or M state in the L2 cache that
contains a correctable error, two lines in system memory may be corrupted. One of
the corrupted lines is the one that contained the correctable error. The second
corrupted line is unrelated to the first line. When a snoop comes into the processor
between global observation and data return for a Read-for-Ownership (RFO) request
that hits an E or M state in the L2 cache that contains a correctable error, two lines in
system memory may be corrupted. One of the corrupted lines is the one that
contained the correctable error. The second corrupted line is unrelated to the first
line.
