Intel Corporation

Mark A Gonzales

Linda J Rankin

David R Gray

D Michael Bell

Susan S Meredith

Sudarshan Bala Cadambi

Charles B Guy

Thomas J Holman

Patrick F Stolt

A method and system to prefetch data from system memory to a central processing unit (CPU). The system includes a dynamic random access memory (DRAM) connected to a high speed bus, a CPU and a bus interface unit that allows the CPU to communicate with the high speed bus. The bus interface unit contains logic circuitry, so that when the CPU generates a read memory access request for information associated with a first address, the interface unit generates a request packet for the information and prefetch information associated with a prefetch address. The bus interface unit creates the request packet by increasing the number of addresses originally requested by the CPU. The interface then sends the request packet to the system memory device, which retrieves and returns the requested data. The interface may include a pair of buffers which store both the information requested by the CPU and speculative or prefetch information. When the CPU generates a subsequent request, the interface compares the addresses requested with the addresses in the prefetch buffer. If the buffer contains the addresses, the data is sent to the processor. The prefetch buffer is directly addressable so that any line within the buffer can be retrieved.

Method to prefetch data from system memory using a bus interface unit

A computer architecture that includes a high speed, low pin bus that directly couples a microprocessor to the physical memory of the processor. Physical memory typically has a number of dynamic random access memory (DRAM) devices. The bus is a byte wide and has a data rate of approximately 500 Mbytes/sec. The high speed bus may be coupled with a conventional bus, so that conventional devices can communicate with the processor using existing bus protocols. The present invention includes a processor interface that allows the processor to communicate using the protocol of either bus. The interface also allows communication between devices on either bus. Also included is a system that incorporates cache memory on a high speed memory bus and a method for allowing I/O devices to be placed on both a conventional bus and the separate high speed bus.

Method and apparatus of bus interface for a processor

A computer system, a bus interface unit, and a method for prefetching data from system memory to a central processing unit (CPU). The system includes a dynamic random access memory (DRAM) connected to a high speed bus, a CPU and a bus interface unit that allows the CPU to communicate with the high speed bus. The bus interface unit contains logic circuitry, so that when the CPU generates a read memory access request for information associated with a first address, the interface unit generates a request a packet for the information and prefetch information associated with a prefetch address. The bus interface unit creates the request packet by increasing the number of addresses originally requested by the CPU. The interface then sends the request packet to the system memory device, which retrieves and returns the requested data. The interface may include a pair of buffers which store both the information requested by the CPU and speculative or prefetch information. When the CPU generates a subsequent request, the interface compares the addresses requested with addresses in the prefetch buffer. If the buffer contains the addresses, the data is sent to the processor. The prefetch buffer is directly addressable so that any line within the buffer can be retrieved.

Computer system having a bus interface unit for prefetching data from system memory

Method and apparatus for data and address transmission over a bus

The present invention provides a method and apparatus for automatically scrubbing ECC errors in memory upon the detection of a correctable error in data read from memory. This is performed by providing in a memory controller memory control logic for controlling accesses to memory, an ECC error checking and correcting unit for checking data read from memory for errors and correcting any correctable errors found in the read data, a first data buffer for storing the corrected read data output from the ECC error checking and correcting unit and a writeback path having an input end coupled to an output of the first data buffer and an output end coupled to memory. Upon the detection of a correctable error in data read from a particular memory location, the ECC error checking and correcting unit signals to the memory control logic the existence of a correctable error in the read data. The memory control logic then obtains exclusive control over the first data buffer and the writeback path to control writing of the corrected read data onto the writeback path and subsequently to memory.

Method and apparatus for automatically scrubbing ECC errors in memory

A bus bridge situated between two buses includes two queues: an outbound request queue and an inbound request queue. Requests originating on the first bus which target a destination on the second bus are placed into the outbound request queue. Requests originating on the second bus which target a destination on the first bus are placed into the inbound request queue. A transaction arbitration unit (TAU) within the bridge maintains transaction ordering and avoids deadlocks. The TAU determines whether requests can be placed in the inbound request queue. The TAU also determines whether requests originating on the first bus can be responded to immediately or whether the agent originating the request must wait for a reply. In addition, the TAU includes logic for determining whether a request in the outbound request queue can be executed on the second bus. The TAU determines whether posting to the inbound request queue is enabled or disabled; whether any posted transactions exist in the inbound request queue; and whether ownership of the second bus is available.

Method and apparatus for maintaining transaction ordering and

CPU reads data from slow bus if I/O devices connected to fast bus do not

A bus bridge situated between two buses includes two queues: an outbound request queue and an inbound request queue. Requests originating on the first bus which target a destination on the second bus are placed into the outbound request queue. Decoding circuitry within the bridge issues a deferred response if the request can be deferred. This deferred response is returned to the originating agent on the first bus, thereby informing the originating agent that the request will be serviced at a later time. Bus control circuitry coupled to the outbound request queue removes requests from the outbound request queue and executes them on the second bus. The bus control circuitry receives a response from the destination agent on the second bus in response to the execution of the outbound request. This response is returned to the originating agent either immediately or after passing through an inbound request queue. Both the outbound request queue and the inbound request queue have data buffers associated therewith for transferring data between the two buses. In addition, requests may originate on the second bus which target a device on the first bus (i.e., inbound requests). These inbound requests are placed in the inbound request queue and are executed on the first bus when removed from the inbound request queue.

Method and apparatus for maintaining transaction ordering and supporting

A method and apparatus for allowing two or more masters, such as central processing units (CPUs), to read a dynamic random access memory (DRAM) device which includes a cache connected to a main memory block. When a CPU provides a read request, the DRAM has a first logic circuit that compares addresses requested with addresses stored in the cache. If the addresses are the same, the DRAM sends an acknowledge (ACK) signal to that CPU and sends the data to the processor. If the addresses are not the same, the DRAM sends a no acknowledge (NACK) signal to the CPU and transfers the requested data from the main memory block to the cache. The DRAM has a second logic circuit that contains a latch which is set when the DRAM sends a NACK signal and reset when the DRAM sends a subsequent ACK signal. The second logic circuit is connected to the first logic circuit to disable the first logic circuit and prevent a cache fetch from main memory when the latch has been set. The second logic circuit is also connected to a refresh controller of the DRAM to prevent a refresh cycle when the latch is set. Both the first logic circuit and the refresh controller are enabled when the latch is reset.

Memory device and method for avoiding live lock of a DRAM with cache

A method and apparatus for detecting and correcting errors in data transferred between a CPU and system memory. System memory typically has a number of dynamic random access memory (DRAM) devices that each have a block of memory cells. The DRAM also has an internal cache that contains a row of memory from a main memory block. Both the cache and block of memory cells contain vertical and horizontal parity bits. Each byte of data bits has an associated horizontal parity bit. Similarly a group of data bits having the same bit position will have an associated vertical parity bit. The parity bits are used to detect and correct errors in data transmissions between a CPU and system memory The cache includes arrays of exclusive OR (XOR) gates that can update the vertical parity bits when one or more bytes of data are written into the DRAM.

Method and apparatus for error detection and correction of data

A distributed method for arbitrating access to a common bus in a multiple processor environment is described. This method provides for fairness where multiple processors are vying for access to a global memory. An apparatus for arbitrating access to a common bus in a multiple processor environment is also described. This apparatus provides for priority determination of each processor upon system reset and provides for fairness where multiple processors are vying for access to a global memory.

Distributed arbitration method and apparatus for a computer bus using

A method of transferring data in response to a read command in a computer system having a plurality of processors coupled to an address bus, a command bus and a data bus is described. A first processor generates and sends the read command to read a first data from a second processor. The second processor then determines with which one of (1) the first data and (2) a read response command and the first data it desires to respond to the read command. If the second processor determines to respond with the first data, then it acknowledges receipt of the read command and performs an ordered response in which the command and address buses are released and only the first data is later sent to the first processor via the data bus when available. If the second processor determines to respond with the read response command and the first data, then it acknowledges receipt of the read command and performs an out-of-order response in which the access of the command and address buses is first released and gained again by arbitration when the first data is determined to be available in the second processor. The second processor then gains the access of the data bus when the data bus is free of any data transaction. The read response command and its address and the first data are then issued to the first processor.

Growing community of inventors

Portland, OR, United States of America

Mark A Gonzales