AMD Geode™ LX Processors Data Book 161
CPU Core Register Descriptions 33234H
7 SPCDEC Decrease Number of Speculative Reads of Data Cache.
0: Actively resync cache tag and data arrays so that loads can be speculatively handled
in one clock if the MRU way is hit. (Default)
1: Do not attempt to resync cache tag and data arrays.
This is a performance optimization bit and the preferred value may have to be empirically
determined. The cache tag and data arrays get “out of sync” when there is a miss to the
MRU way or if the data array is busy with a store, linefill, or eviction. While the arrays are
out of sync, all hits take 2 clocks. When they are in sync, hits to the MRU way take 1
clock while hits to other ways take 3.
6WTBRST Write-Through Bursting.
0: Writes are sent unmodified to the bus on write-through operations. (Default)
1: Writes may be combined using write-burstable semantics on write-through operations.
5WBINVD Convert INVD to WBINVD Instruction.
0: INVD instruction invalidates cache without writeback. (Default)
1: INVD instruction writes back any dirty cache lines
4NOSMC Snoop Detecting on Self-Modified Code. Generates snoops on stores for detecting
self-modified code.
0: Generate snoops. (Default)
1: Disable snoops.
3NOFWD Forward Data from Bus Controller. Enable forwarding of data directly from bus control-
ler if a new request hits a line fill in progress.
0: Forward data from bus controller if possible. (Default)
1: Wait for valid data in cache, then read cache array.
2BLOCKC Blocking Cache.
0: New request overlapped with linefill. (Default)
1: Linefill must complete before starting new request.
1 MISSER Serialize Load Misses. Stall everything but snoops on a load miss. Set this bit if part of
PCI space is marked as cacheable (e.g., for a ROM), data accesses will be made from
that cacheable space, and there is a PCI master device which must complete a master
request before it will complete a slave read.
0: Load misses are treated the same as load hits. (Default)
1: Load misses prevent non-snoop requests from being handled until the miss data is
returned by the bus controller.
0 LDSER Serialize Loads vs Stores. All loads are serialized versus stores in the store queue, but
a load that hits the cache completes without affecting any pending stores in the write
buffers.
0: Loads bypass stores based on region properties. (Default)
1: All loads and stores are executed in program order.
DM_CONFIG0_MSR Bit Descriptions (Continued)
Bits Name Description