# US 2003/0204679 A1 · Methods and arrangements to enhance an upbound path

> Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt

- Number: US 2003/0204679 A1
- Document ID: US20030204679A1
- Status: Published application
- Kind: A1
- Application number: 10/135,552
- Filed: 2002-04-30
- Published: 2003-10-30
- Earliest claimed: 2002-04-30
- Assignees: Intel Corporation
- Inventors: Robert G. Blankenship
- Web page: https://patentgrep.com/patent/US20030204679A1

## CPC

- G06F 12/0835: {for main memory peripheral accesses (e.g. I/O or DMA)}
- G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N)

## Abstract

A cache coherency arrangement to enhance an upbound path for input-output interfaces is disclosed. Several embodiments may enhance upbound write bandwidth and buffer utilization. Some embodiments may comprise requesting content of a memory granule and merging the content with data associated with a write request for the memory granule prior to satisfaction of an ordering rule associated with the write request. Many embodiments may comprise ownership stealing to enhance inbound bandwidth and to prevent or attenuate starvation and/or deadlock of transactions or of an input-output interface for transactions. Such embodiments may also comprise invalidating merged content of the memory granule. Further embodiments may comprise reverting the merged content to the data associated with the write request.

## Claims

### Claim 1

An apparatus, comprising:
interface memory to store a write request and to receive a content of a memory granule associated with the write request from an unordered interface;
logic circuitry coupled with the unordered interface to request the content in response to the write request; and
a merge controller coupled with said interface memory to merge the content with the write request prior to satisfaction of an ordering rule associated with the write request.

### Claim 2 (depends on 1)

The apparatus of claim 1, further comprising a snoop filter coupled with the interface memory to invalidate the content based upon a conflict.

### Claim 3 (depends on 2)

The apparatus of claim 2, wherein said merge controller comprises derivation circuitry to derive the write data from a merger of the content with the write request in response to an invalidation of the content.

### Claim 4 (depends on 2)

The apparatus of claim 2, further comprising a coherency interface coupled with the snoop filter to invalidate the content in response to a transaction of a peer hub interface.

### Claim 5 (depends on 1)

The apparatus of claim 1, said interface memory comprises a queue coupled with an ordered interface to maintain a transaction order for a header of the write request based upon the ordering rule.

### Claim 6 (depends on 1)

The apparatus of claim 1, wherein said interface memory comprises a buffer coupled with said merge controller to store a merger of write data associated with the write request and the content.

### Claim 7 (depends on 1)

The apparatus of claim 1, wherein said logic circuitry comprises content request circuitry to forward a header associated with the write request to the unordered interface.

### Claim 8 (depends on 1)

The apparatus of claim 1, wherein said logic circuitry comprises merge request circuitry coupled with said merge controller to request a merger of the content with the write request.

### Claim 9 (depends on 1)

The apparatus of claim 1, wherein said merge controller comprises overwrite circuitry coupled with said interface circuitry to overwrite part of the content in accordance with the write request.

### Claim 10

A method, comprising:
receiving a write request to store write data in a memory granule;
pre-fetching ownership of the memory granule for the write request;
receiving a content of the memory granule in response to said pre-fetching ownership; and
combining the content with the write data before a transaction order of the write request satisfies an ordering rule.

### Claim 11 (depends on 10)

The method of claim 10, further comprising determining a conflict associated with the memory granule to invalidate the ownership of the memory granule by the write request.

### Claim 12 (depends on 11)

The method of claim 11, further comprising deriving the write data from a combination of the content with the write data in response to determining a conflict.

### Claim 13 (depends on 10)

The method of claim 10, wherein said receiving a write request comprises receiving a request to overwrite part of the content in the memory granule.

### Claim 14 (depends on 10)

The method of claim 10, wherein said receiving a write request comprises storing a header associated with the write request in a queue.

### Claim 15 (depends on 10)

The method of claim 10, wherein said pre-fetching ownership comprises forwarding a header associated with the write request to an unordered interface.

### Claim 16 (depends on 10)

The method of claim 10, wherein said receiving a content of the memory granule comprises receiving the content of a memory line.

### Claim 17 (depends on 10)

The method of claim 10, wherein said combining the content comprises storing a combination of the content and the write data in interface memory.

### Claim 18 (depends on 17)

The method of claim 17, wherein storing a combination comprises overwriting the content in a buffer with a subsequent content of the memory granule.

### Claim 19

A system, comprising:
an input-output device to initiate a write request comprising write data to modify a content of a memory granule;
a hub interface coupled with said input-output device to maintain a transaction order for the write request according to an ordering rule; and
an upbound path coupled with said hub interface to obtain the content and to merge the content with write data prior to satisfaction of the ordering rule.

### Claim 20 (depends on 19)

The system of claim 19, further comprising a snoop filter to receive a snoop request to revert a merger of the content with the write data in response to the snoop request.

### Claim 21 (depends on 19)

The system of claim 19, wherein said hub interface comprises a queue to maintain the transaction order to forward the write request to an unordered interface after the write request is at the top of the queue.

### Claim 22 (depends on 19)

The system of claim 19, wherein said upbound path comprises logic circuitry to forward a merger of the content with the write request to the unordered interface after the transaction order of the write request satisfies the ordering rule.

### Claim 23 (depends on 19)

The system of claim 19, wherein said upbound path comprises merge controller in communication with a buffer to combine the write data with the content.

### Claim 24

A system apparatus, comprising:
a queue coupled with an ordered interface to store a write request;
logic circuitry responsive to said queue to request a content of a memory granule from the unordered interface;
a merge controller in communication with the unordered interface to receive the content and responsive to said logic circuitry to combine the content with write data of the write request;
a buffer in communication with said merge controller to store a combination of the content and the write data; and
an unordered interface responsive to said logic circuitry and communicatively coupled with said buffer to forward the combination to the memory granule.

### Claim 25 (depends on 24)

The system of claim 24, further comprising a snoop filter responsive to said unordered interface to invalidate an ownership of the memory granule in response to a snoop request.

### Claim 26 (depends on 24)

The system of claim 24, further comprising a snoop filter responsive to another hub interface to invalidate an ownership of the memory granule.

### Claim 27 (depends on 26)

The system of claim 26, wherein said merge controller comprises derivation circuitry responsive to said snoop filter to derive the write data from the merger.

### Claim 28

A machine-readable medium containing instructions, which when executed by a machine, cause said machine to perform operations, comprising:
requesting ownership of the memory granule in response to a write request;
storing a content of the memory granule based upon a response to said requesting ownership; and
merging the content with write data of the write request before satisfaction of an ordering rule associated with the write request.

### Claim 29 (depends on 28)

The machine-readable medium of claim 28, further comprising reverting said merging to determine the write data.

### Claim 30 (depends on 29)

The machine-readable medium of claim 29, wherein reverting said merging comprises removing the content from the merger of the content with the write data.
