Intel NETWORK PROCESSOR IXP2800 - Manual (Page 128)

128 Hardware Reference Manual
Intel
®
IXP2800 Network Processor
Intel XScale ® Core
3.11.4 Atomic Operations
The Intel XScale
®
core has Swap (SWP) and Swap Byte (SWPB) instructions that generate an
atomic read-write pair to a single address. These instructions are supported for the SRAM and
Scratch space, and also to any other address space if it is done by a Read command followed by
Write command.
cbiIO is asserted when a data cache request is initiated to a memory region with cacheable and
bufferable bits in the translation table first-level descriptor set to 0. Also, if cbiIO is asserted during
the CMB read portion of the SWP, then it also does a Read Command followed by Write
Command, regardless of address. In those cases, the SWP/SWPB is atomic with respect to
processes running on the Intel XScale
®
core, but not with respect to the Microengines.
The following summarizes the Atomic operation.
When the Intel XScale ® core presents the read command portion of the SWP, it will assert the
cbiLock signal. The gasket will “ack” the read and save the BufID in the push_ff. It will not
arbitrate for the command bus at that time; rather it will wait for the corresponding write of the
SWP (which will also have cbiLock asserted). At that time the gasket will arbitrate for the
command bus to send a command with the atomic operation in the command field (the command is
based on the address space chosen for the SRAM/Scratch, which has multiple aliased address
ranges).
The SRAM or Scratch controller will pull the data, do the atomic read-modify-write, and then push
the read data back. The gasket will use the saved BufID when returning the data to CMB.
Note: Unrelated reads, such as instruction and Page Table fetches, can come in the interval between the
read-lock and write-unlock, and will be handled by the gasket. No other data reads or writes will
come in that interval. Also, the Intel XScale
®
core will not wait for the SWP read data before
presenting the write data.
The gasket uses address aliases to generate the following atomic operations.
• Bit Set
• Bit Clear
• Add
• Subtract
• Swap
For the alias address type of atomic operation, the Intel XScale
®
core will issue a SWP command
with an alias address if it needs data return. The Intel XScale
®
core will use the write command
with an alias address if it does not need data return.
Xscale_IF will not check the second address, as long as it detects one write after one read, both
with cbiLock enabled. It will take the write
address and put it in the command.
Address Space cbiIO Operation
SRAM/Scratch 0 RMW Command
Not SRAM/Scratch x Read Command followed by Write Command
Any 1 Read Command followed by Write Command