Compaq Computer Corporation, Inc.

Hewlett-packard Development Company, L.p.

Compaq Information Technologies Group, L.p.

Kenneth T Chin

Robert A Lester

Phillip M Jones

Michael J Collins

Gary J Piccirillo

Jerome J Johnson

Clarence K Coffee

John Edward Larson

Jeffrey C Stevens

C Kevin Coffee

Dwight D Riley

Hubert E Brinkmann

Paul V Brownell

David L Matthews

Paul B Rawlins

Hahn Norden

James Xuan Dinh

Clarence Coffee

Jim Blocker

A network node within a network includes a first receive buffer, first buffer management, a second receive buffer and second buffer management. The first buffer management performs link level credit based flow control for network packets that the first buffer management places in the first receive buffer. The second buffer management performs end-to-end credit based flow control for network packets that the second buffer management receives from the first receive buffer and processes before placing data in the second receive buffer.

End-to-end flow control in a network

A system is provided comprising a fabric coupling together a plurality of computing devices, wherein the fabric transfers a stream of packets between the computing devices. Each computing device comprises a Quality of Service ('QOS') filter that monitors incoming packets to filter out packets of a maintenance type and permit transfer of packets of a transaction type.

Quality of service with control flow packet filtering

Methods for improving processor performance. Specifically, by reducing some of the latency cycles within a host controller, request processing speed can be improved. One technique for improving processing speed involves initiating a deferred reply transaction before the data is available from a memory controller. A second technique involves anticipating the need to transition from a block next request (BNR) state to a bus priority request (BPRI) state, thereby eliminating the need to wait for a request check to determine if the BPRI state must be implemented.

Method for improving processor performance

A technique for optimizing cycle time in maintaining cache coherency. Specifically, a method and apparatus are provided to optimize the processing of requests in a multi-processor-bus system which implements a snoop-based coherency scheme. The acts of snooping a bus for a first address and searching a posting queue for the next address to be snooped are performed simultaneously to minimize the request cycle time.

Next snoop predictor in a host controller

A computer system includes an adaptive memory arbiter for prioritizing memory access requests, including a self-adjusting, programmable request-priority ranking system. The memory arbiter adapts during every arbitration cycle, reducing the priority of any request which wins memory arbitration. Thus, a memory request initially holding a low priority ranking may gradually advance in priority until that request wins memory arbitration. Such a scheme prevents lower-priority devices from becoming &ldquo;memory-starved.&rdquo; Because some types of memory requests (such as refresh requests and memory reads) inherently require faster memory access than other requests (such as memory writes), the adaptive memory arbiter additionally integrates a nonadjustable priority structure into the adaptive ranking system which guarantees faster service to the most urgent requests. Also, the adaptive memory arbitration scheme introduces a flexible method of adjustable priority-weighting which permits selected devices to transact a programmable number of consecutive memory accesses without those devices losing request priority.

Computer system with adaptive memory arbitration scheme

A computer is provided having a bus interface unit coupled between a processor bus, a peripheral bus, and a memory bus. The bus interface unit includes a processor controller linked to the processor bus for controlling the transfer of cycles from the processor to the peripheral bus and memory bus. Those cycles are initially forwarded as a request, whereby the processor controller includes a memory request queue separate from a peripheral request queue. Requests from the memory and peripheral request queues can be de-queued concurrently to the memory and peripheral buses. This enhances throughput of read and write requests; however, proper ordering of data returned as a result of read requests and data transferred as a result of write requests must be ensured. An in-order queue is also present in the processor controller which records the order in which the requests are dispatched to the peripheral and memory buses from the peripheral and memory request queues. Data ensuing from the request can be re-ordered and presented to the destination based on the current pointer position within the in-order queue. Thus, the in-order queue keeps track of the order in which data is transferred across the processor bus consistent with the order in which the previous requests were transferred.

System and method for concurrently requesting input/output and memory address space while maintaining order of data sent and returned therefrom

A computer system includes a CPU and a memory device coupled by a bridge logic unit. CPU to memory write requests (including the data to be written) are temporarily stored in a queue in the bridge logic unit. The bridge logic unit preferably begins a write cycle to the memory device before all of the write data has been stored in the queue and available to the memory device. By beginning the memory cycle as early as possible, the total amount of time required to store all of the write data in the queue and then de-queue the data from the queue is reduced. Consequently, many CPU to memory write transactions are performed more efficiently and generally with less latency than previously possible.

Computer system with improved memory access

A computer is provided having a bus interface unit coupled between a CPU bus, a peripheral bus (i.e., PCI bus and/or graphics bus), and a memory bus. The bus interface unit includes controllers linked to the respective buses, and a plurality of queues placed within address and data paths between the various controllers. The peripheral bus controller can decode a write cycle to memory, and the processor controller can thereafter request and be granted ownership of the CPU local bus. The address of the write cycle can then be snooped to determine if valid data exists within the CPU cache storage locations. If so, a writeback operation can occur. Ownership of the CPU bus is maintained by the bus interface unit during the snooping operation, as well as during writeback and the request of the memory bus by the peripheral-derived write cycle. It is not until ownership of the memory bus is granted by the memory arbiter that mastership is terminated by the bus interface unit. Accordingly, the bus interface unit keeps CPU-derived cycles off the CPU bus to ensure memory arbiter grants ownership to a write cycle from the peripheral bus. In this fashion, data from the peripheral bus can be stored in system memory before accessing that data by a CPU read cycle. The number of snoop cycles which the bus interface unit can initiate is determined by configuration registers programmed during power on, reset or boot up of computer.

System and method for maintaining ownership of a processor bus while sending a programmed number of snoop cycles to the processor cache

A computer is provided having a system interface unit coupled between main memory, a CPU bus, and a PCI bus and/or graphics bus. A hard drive is typically coupled to the PCI bus. The system interface unit is configured to perform a data integrity protocol. Also, all bus master devices (CPUs) on the processor bus may perform the same data integrity protocol. When a CPU requests read data from main memory, the bus interface unit forwards the read data and error information unmodified to the processor bus bypassing the data integrity logic within the system interface unit. However, the system interface unit may still perform the data integrity protocol in parallel with the requesting CPU so that the system interface unit may track errors and possibly notify the operating system or other error control software of any errors. In this manner processor read latency is improved without sacrificing data integrity. Furthermore, the system interface unit may still track errors on processor reads. If the read request is from a device on a peripheral bus (AGP or PCI bus), then the system interface unit performs the data integrity protocol on the data and error bits before forwarding the read data to the appropriate bus.

System and method for improving processor read latency in a system employing error checking and correction

A computer system that includes a CPU, a memory and a memory controller for controlling access to the memory. The memory controller generally includes arbitration logic for deciding which memory request among one or more pending requests should win arbitration. When a request wins arbitration, the arbitration logic asserts a &ldquo;won&rdquo; signal corresponding to that memory request. The memory controller also includes synchronizing logic to synchronize memory requests, corresponding to a first group of requests, that win arbitration to a clock signal and an arbitration enable signal. The synchronizing logic includes an AND gate and a latch for synchronizing the won signals. The memory controller also asynchronously arbitrates a second group of memory requests by asserting a won signal associated with the second group requests that is not synchronized to the clock signal. In this manner, the won signals for the second group of requests can be asserted earlier than the synchronized won signals, thereby permitting the asynchronously arbitrated second group memory requests to be performed earlier than otherwise possible.

Computer system with synchronous memory arbiter that permits asynchronous memory requests

A computer system includes a CPU, a memory device, two expansion buses, and a bridge logic unit coupling together the CPU, the memory device and the expansion buses. The CPU couples to the bridge logic unit via a CPU bus and the memory device couples to the bridge logic unit via a memory bus. The bridge logic unit generally routes bus cycle requests from one of the four buses to another of the buses while concurrently routing bus cycle requests to another pair of buses. The bridge logic unit preferably includes four interfaces, one each to the CPU, memory device and the two expansion buses. Each pair of interfaces are coupled by at least one queue; write requests are stored (or &ldquo;posted&rdquo;) in write queues and read data are stored in read queues. Because each interface can communicate concurrently with all other interfaces via the read and write queues, the possibility exists that a first interface cannot access a second interface because the second interface is busy processing read or write requests from a third interface, thus starving the first interface for access to the second interface. To remedy this starvation problem, the bridge logic unit prevents the third interface from posting additional write requests to its write queue, thereby permitting the first interface access to the second interface. Further, read cycles may be retried from one interface to allow another interface to complete its bus transactions.

Computer system employing memory controller and bridge interface permitting concurrent operation

A computer is provided having a bus interface unit coupled between a CPU bus, a peripheral bus, and a memory bus. The bus interface unit includes a processor controller linked to the CPU bus for controlling the transfer of cycles from the CPU to the peripheral bus and memory bus. Those cycles can be arranged in order within the CPU bus pipeline. A subset of cycles destined for a peripheral bus can be stalled within a snoop phase associated with the CPU bus. Snoop stall can continue until a memory cycle is encountered upon the CPU bus pipeline within a phase prior to the snoop phase. Once the memory cycle progresses to the snoop phase, snoop stall can be discontinued and the previous, peripheral cycles can then be deferred and/or retried, allowing the memory cycle to be quickly dispatched through all phases of the CPU bus and onto the memory bus. In this fashion, memory cycles can be completed quickly, yet deferrals or retries are minimized to avoid the throughput penalty associated with deferring or retrying cycles back again through each phase of the CPU bus.

Growing community of inventors

Cypress, TX, United States of America

Kenneth T Chin