Atmel SAM9X25, Manual

Page: 43 / 1145

43
11054A–ATARM–27-Jul-11
SAM9X25
43
11054A–ATARM–27-Jul-11
SAM9X25
9.7.2.2 Write Buffer
The ARM926EJ-S contains a write buffer that has a 16-word data buffer and a four- address buf-
fer. The write buffer is used for all writes to a bufferable region, write-through region and write-
back region. It also allows to avoid stalling the processor when writes to external memory are
performed. When a store occurs, data is written to the write buffer at core speed (high speed).
The write buffer then completes the store to external memory at bus speed (typically slower than
the core speed). During this time, the ARM9EJ-S processor can preform other tasks.
DCache and Write Buffer support write-back and write-through memory regions, controlled by C
and B bits in each section and page descriptor within the MMU translation tables.
Write-though Operation
When a cache write hit occurs, the DCache line is updated. The updated data is then written to
the write buffer which transfers it to external memory.
When a cache write miss occurs, a line, chosen by round robin or another algorithm, is stored in
the write buffer which transfers it to external memory.
Write-back Operation
When a cache write hit occurs, the cache line or half line is marked as dirty, meaning that its
contents are not up-to-date with those in the external memory.
When a cache write miss occurs, a line, chosen by round robin or another algorithm, is stored in
the write buffer which transfers it to external memory.
9.8 Bus Interface Unit
The ARM926EJ-S features a Bus Interface Unit (BIU) that arbitrates and schedules AHB
requests. The BIU implements a multi-layer AHB, based on the AHB-Lite protocol, that enables
parallel access paths between multiple AHB masters and slaves in a system. This is achieved by
using a more complex interconnection matrix and gives the benefit of increased overall bus
bandwidth, and a more flexible system architecture.
The multi-master bus architecture has a number of benefits:
• It allows the development of multi-master systems with an increased bus bandwidth and a
flexible architecture.
• Each AHB layer becomes simple because it only has one master, so no arbitration or master-
to-slave muxing is required. AHB layers, implementing AHB-Lite protocol, do not have to
support request and grant, nor do they have to support retry and split transactions.
• The arbitration becomes effective when more than one master wants to access the same
slave simultaneously.
9.8.1 Supported Transfers
The ARM926EJ-S processor performs all AHB accesses as single word, bursts of four words, or
bursts of eight words. Any ARM9EJ-S core request that is not 1, 4, 8 words in size is split into
packets of these sizes. Note that the Atmel bus is AHB-Lite protocol compliant, hence it does not
support split and retry requests.