
Chapter 3. L1 and L2 Cache Operation
3-9
Memory and Cache Coherency
has a large negative effect on load miss bandwidth performance. For this reason, it is not
recommended to have guarded loads in code streams that require high system bandwidth
utilization.
3.4.2 Coherency Support
The MPC7400 provides full hardware support for PowerPC cache coherency and ordering
instructions (
dcbz
,
dcbi
,
dcbf
,
sync
,
icbi
, and
eieio
) and full hardware implementation of
the TLB management instructions (
tlbie
, and
tlbsync
). Snooping, described in
Section 3.9.3, òSnooping,ó is integral to the memory subsystem design and operation. The
MPC7400 is self-snooping and can ARTRY its own
tlbie
,
tlbsync
,
icbi
, and
sync
broadcasts.
Each 32-byte cache block in the data cache contains 6 status bits (CDMRSV). The
MPC7400 uses these bits to support the coherency protocols and to direct castout and
reload operations. The L1 data cache status bits and the conditions that cause them to be set
or cleared are deTned in Table 3-1.
Table 3-1. Data Cache Status Bits
Status
Bit
Name
Meaning
Set Conditions
Clear Conditions
C
Castout
The cache block should be
castout from the L1 data cache to
the L2 cache when selected for
replacement
Non-transient reload
from BIU
Transient hit
D
Dirty
The cache block has been stored
to since it was reloaded into the
L1 data cache
Store miss reload from
BIU or L2
Writeback store hit on S
& R
dcbst
hit
M
ModiTed
The cache block is modiTed with
respect to the external system
interface
Store miss reload from
BIU or L2
Writeback store hit on S
& R
dcbst
hit
Snoop clean hit
Snoop read hit
R
Recent
This is the most recent processor
to perform a read transaction to
the cache block while other
processors have a shared copy
Load miss reload from
BIU with SHD response
Load miss reload from
L2 cache with L2 cache
status = R
Snoop read hit