International Business Machines Corporation

Kabushiki Kaisha Toshiba

Sony Computer Entertainment Inc.

Brad William Michael

Brian King Flachs

Mark Richard Nutter

Kathryn Mary O'Brien

Tong Chen

John Steven Dodson

Marc Avery Gollub

Tao Zhang

John K P O'Brien

Warren Edward Maule

Charles Ray Johns

John Samuel Liberty

Sang Hoo Dhong

Patrick James Meaney

Harm Peter Hofstee

Eric Eugene Retter

Stephen J Powell

Gary A Van Huben

Glenn David Gilda

Matthew Edward King

Silvia Melitta Mueller

Diyanesh B Chinnakkonda Vidyapoornachary

James A O'Connor

Barry Marshall Trager

Gordon Clyde Fossum

Tony Emile Sawan

Makoto Aikawa

Gilles Gervais

Hwa-Joon Oh

Seiji Maeda

Mack Wayne Riley

Daniel Alan Brokenshire

John Kevin Patrick O'Brien

Nicolas Maeding

Iwao Takiguchi

Bruce David D'Amora

Yaping Zhou

Tetsuji Tamura

Hiroo Hayashi

John Fred Spannaus

Shigeaki Iwasa

Paul Marlan Harvey

Michael Alan Kutner

Kevin Duc Tran

Nicholas Steven Rolfe

Naxin Jason Wang

Susan D Wright

Joseph L Bradley

Mark E VanNostrand

Lilith Hale

A memory controller comprises a system bus interface that connects the MC to a system processor, a system memory interface that connects the MC to a system memory, a read buffer comprising a plurality of entries constituting storage areas, the entries comprising at least one read buffer entry (RBE) and at least one extended prefetch read buffer entry (EPRBE), read buffer logic, dynamic controls that are used by the read buffer logic, and an MC processor comprising at least one extended prefetch machine (EPM), each corresponding to one of the at least EPRBEs, where the MC processor is configured to allocate and deallocate EPRBEs and RBEs according to an allocation method using the dynamic controls.

Dynamically allocating memory controller resources for extended prefetching

A memory controller receives, via a first interface, a first read request requesting a requested data granule. Based on receipt of the first read request, the memory controller transmits, via a second interface, a second read request to initiate access of the requested data granule from a system memory. Based on a determination to schedule accelerated data delivery and receipt by the memory controller of a data scheduling indication that indicates a timing of future delivery of the requested data granule, the memory controller requests, prior to receipt of the requested data granule, permission to transmit the requested data granule on the system interconnect fabric. Based on receipt of the requested data granule at the indicated timing and a grant of the permission to transmit, the memory controller initiates transmission of the requested data granule on the system interconnect fabric and transmits an error indication for the requested data granule.

Reduced system memory latency via a variable latency interface

Tracking address ranges for computer memory errors including detecting, by memory logic, an error at a memory address, the memory address representing one or more memory cells at a physical location of computer memory; reporting, by the memory logic to memory firmware, the detected error including providing the memory firmware with the memory address; identifying, by the memory firmware, an address range affected by the detected error including scanning the computer memory in dependence upon the memory address; determining, by the memory firmware, a region size based on the address range affected by the detected error; and populating an entry in a mark table corresponding to the detected error, including populating a field specifying the region size and a field specifying a match address corresponding to the memory address.

Tracking address ranges for computer memory errors

Performing error correction in computer memory including receiving a read request targeting a read address within the computer memory; accessing a mark table comprising a plurality of entries, each entry including a field specifying a region size, a field specifying a match address, and a field specifying a mark location; performing a lookup of the mark table using the read address including, for each entry in the mark table: generating a mask based on the region size stored in the entry; determining, based on the mask, whether the read address is within a memory region specified by the match address and region size stored in the entry; and if the read address is within the memory region specified by the match address and region size stored in the entry, performing error correction using the mark location stored in the entry.

Performing error correction in computer memory

Embodiments of the present invention include a memory module that includes a plurality of memory devices and a memory buffer device. The memory devices are characterized as one of a high or low random bit error rate (RBER) memory device. The memory buffer device includes a read data interface to receive data read from a memory address on one of the memory devices, and common error correction logic to detect and correct error conditions in data read from both high RBER and low RBER memory devices. The memory buffer device also includes refresh rate logic configured to adjust a refresh rate based on the detected error conditions.

Common high and low random bit error correction logic

Embodiments of the present invention include a memory module that includes a plurality of memory devices and a memory buffer device. Each of the memory devices are characterized as one of a high random bit error rate (RBER) and a low RBER memory device. The memory buffer device includes a read data interface to receive data read from a memory address on one of the memory devices. The memory buffer device also includes common error correction logic to detect and correct error conditions in data read from both high RBER and low RBER memory devices. The common error correction logic includes a plurality of error correction units which provide different complexity levels of error correction and have different latencies. The error correction units include a first fast path error correction unit for isolating and correcting random symbol errors.

Managing entries in a mark table of computer memory errors including identifying at least two mark table entries as candidates for merger, wherein each mark table entry indicates an error at a location in a computer memory; and merging the identified mark table entries into a single mark table entry, including removing one of the identified mark table entries from the mark table.

Managing entries in a mark table of computer memory errors

Confirming memory marks indicating an error in computer memory including detecting, by memory logic responsive to a memory read operation, an error in at a memory location; marking, by the memory logic in an entry in a hardware mark table, the memory location as containing the error, the entry including one or more parameters for correcting the error; and retrying, by the memory logic, the memory read operation, including: responsive to again detecting the error in the memory location, determining whether the error is correctable at the memory location using the parameters included in the entry; and if the error is correctable at the memory location using the one or more parameters included in the entry, confirming the error in the entry of the hardware mark table.

Confirming memory marks indicating an error in computer memory

Mechanisms are provided for arranging binary code to reduce instruction cache conflict misses. These mechanisms generate a call graph of a portion of code. Nodes and edges in the call graph are weighted to generate a weighted call graph. The weighted call graph is then partitioned according to the weights, affinities between nodes of the call graph, and the size of cache lines in an instruction cache of the data processing system, so that binary code associated with one or more subsets of nodes in the call graph are combined into individual cache lines based on the partitioning. The binary code corresponding to the partitioned call graph is then output for execution in a computing device.

Arranging binary code based on call graph partitioning

An aspect includes memory error recovery in a memory system includes detecting an error condition within a memory chip of the memory system. A chip mark is applied to the memory chip to flag the error condition. An address range of the memory chip associated with the error condition is determined. Data are written from the address range of the memory chip to a cache memory. The chip mark is removed based on determining that all of the data from the address range have been written to the cache memory.

Memory error recovery

Growing community of inventors

Cedar Park, TX, United States of America

Brad William Michael