# US 8,279,231 B1 · Bandwidth impedance matching and starvation avoidance by read completion buffer allocation

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

- Number: US 8,279,231 B1
- Document ID: US8279231B1
- Status: Granted
- Kind: B1
- Application number: 12/260,985
- Filed: 2008-10-29
- Granted: 2012-10-02
- Earliest claimed: 2008-10-29
- Assignees: NVIDIA Corporation
- Inventors: Samuel Hammond Duncan; John H. Edmondson; Raymond Hoi Man Wong; Lukito Muliadi
- Web page: https://patentgrep.com/patent/US8279231B1

## CPC

- G06F 12/1081: for peripheral access to main memory, e.g. direct memory access [DMA]
- G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N)

## Abstract

Read completion buffer space is allocated in accordance with a preset limit. When a read request is received from a client, the sum of a current allocation of the read completion buffer space and a new allocation of the read completion buffer space required by the read request is compared with the preset limit. If the preset limit is not exceeded, read completion buffer space is allocated to the read request. If the preset limit is exceeded, the read request is suspended until sufficient data is read out from the read completion buffer.

## Claims

### Claim 1

A method for allocating read completion buffer space, comprising:
receiving a read request from a client;
determining that a read completion buffer space allocated to said client will not exceed a preset upper bound for said client as a result of allocating read completion buffer space required by the read request from said client; and
allocating read completion buffer space for the read request from said client,
wherein the preset upper bound comprises a first value when the read request is directed to system memory or a second value when the read request is directed to graphics memory, and wherein the preset upper bound is established by a limit that is computed as latency of reads from a memory multiplied by size of read completion interface of the client.

### Claim 2 (depends on 1)

The method according to claim 1, wherein the read request from said client includes a request to read data from system memory, and the upper bound for said client is preset based on latency of reads from system memory and size of read completion interface of the client.

### Claim 3 (depends on 1)

The method according to claim 1, wherein the read request from said client includes a request to read data from graphics memory, and the upper bound for said client is preset based on latency of reads from graphics memory and size of read completion interface of the client.

### Claim 4 (depends on 1)

The method according to claim 1, further comprising:
receiving another read request from a client;
determining that a read completion buffer space allocated to said client will exceed a preset upper bound for said client as a result of allocating read completion buffer space required by the read request from said client; and suspending the read request from said client.

### Claim 5 (depends on 4)

The method according to claim 4, further comprising:
completing one or more read requests for said client;
determining that the read completion buffer space allocated to said client will not exceed the preset upper bound for said client as a result of allocating read completion buffer space required by the suspended read request from said client;
allocating read completion buffer space for the suspended read request from said client; and
resuming the suspended read request from said client.

### Claim 6

In an input/output unit for a processing unit that is interfaced with local memory and system memory, a method for allocating read completion buffer space for multiple clients, comprising:
receiving a read request from a client;
retrieving a first preset limit for said client from memory if the read request is to system memory and a second preset limit for said client from memory if the read request is to local memory; and
allocating read completion buffer space for the read request based on a comparison with at least one of the first and second preset limits,
wherein the first preset limit is computed as latency of reads from the system memory multiplied by size of read completion interface of the client, and the second preset limit is computed as latency of reads from the local memory multiplied by size of read completion interface of the client.

### Claim 7 (depends on 6)

The method according to claim 6, further comprising:
determining that the read request is a read request to system memory,
wherein the comparison includes a comparison of a current allocation of read completion buffer space for said client and a new allocation of read completion buffer space required by the read request with the first preset limit.

### Claim 8 (depends on 7)

The method according to claim 7, wherein the comparison is a comparison of the sum of the current allocation of read completion buffer space for read requests to system memory from said client and the new allocation of read completion buffer space required by the read request with the first preset limit.

### Claim 9 (depends on 6)

The method according to claim 6, further comprising:
determining that the read request is a read request to local memory,
wherein the comparison includes a comparison of a current allocation of read completion buffer space for said client and a new allocation of read completion buffer space required by the read request with the second preset limit.

### Claim 10 (depends on 9)

The method according to claim 9, wherein the comparison is a comparison of the sum of the current allocation of read completion buffer space for read requests to local memory from said client and the new allocation of read completion buffer space required by the read request with the second preset limit.

### Claim 11 (depends on 6)

The method according to claim 6, further comprising:
receiving another read request from a client;
retrieving a first preset limit for said client from memory if the read request is to system memory and a second preset limit for said client from memory if the read request is to local memory; and
allocating read completion buffer space for the read request from said client based on a comparison with at least one of the first and second preset limits for said client.

### Claim 12

In an input/output unit for a processing unit that is interfaced with local memory and system memory, a method for allocating read completion buffer space for multiple clients, comprising:
receiving a read request from a client;
retrieving a first preset limit for said client from memory if previous read requests to system memory are not pending for said client, and a second preset limit for said client from memory if previous read requests to system memory are pending for said client; and
allocating read completion buffer space for the read request based on a comparison with at least one of the first and second preset limits,
wherein the first preset limit and the second preset limit are computed as latency of reads from the system memory multiplied by size of read completion interface of the client.

### Claim 13 (depends on 12)

The method according to claim 12, wherein the comparison is a comparison of the sum of the current allocation of read completion buffer space for read requests to system memory and to local memory from said client, and the new allocation of read completion buffer space required by the read request.

### Claim 14

A computer system comprising:
a first memory unit;
a second memory unit;
a plurality of clients that generate requests to read data from the first memory unit and the second memory unit; and
an input/output unit that processes the requests to read data from the clients,
wherein the input/output unit includes a read completion buffer and stores read completion buffer allocation limits for the clients, and wherein the read completion buffer allocation limits for each of the clients comprises a first preset limit that is computed as latency of reads from the first memory unit multiplied by size of read completion interface of the client, and a second preset limit is computed as latency of reads from the second memory unit multiplied by size of read completion interface of the client.

### Claim 15 (depends on 14)

The computer system according to claim 14, wherein the input/output unit stores first and second read completion buffer allocation limits for each of the clients.

### Claim 16 (depends on 15)

The computer system according to claim 15, wherein the input/output unit is configured to limit the amount of read completion buffer that is allocated to requests from a client to read data from the first memory unit in accordance with the first read completion buffer allocation limit for said client.

### Claim 17 (depends on 15)

The computer system according to claim 15, wherein the input/output unit is configured to limit the amount of read completion buffer that is allocated to requests from a client to read data from the second memory unit in accordance with the second read completion buffer allocation limit for said client.

### Claim 18 (depends on 15)

The computer system according to claim 15, wherein the input/output unit is configured to limit the amount of read completion buffer that is allocated to requests from a client to read data from either the first memory unit or the second memory unit by comparing the total amount of read completion buffer allocated to the client, to the first read completion buffer allocation limit when the client does not have pending read requests to the first memory unit, and to the second read completion buffer allocation limit when the client has pending read requests to the first memory unit.

### Claim 19 (depends on 14)

The computer system according to claim 14, further comprising a graphics processing unit, wherein the first memory unit is system memory and the second memory unit is local memory for the graphics processing unit.

### Claim 20 (depends on 19)

The computer system according to claim 19, wherein the input/output unit is a non-isochronous hub for the graphics processing unit.
