International Business Machines Corporation

Globalfoundries Inc.

Charles Francis Marino

William John Starke

Jeffrey A Stuecheli

Michael Steven Siegel

Praveen S Reddy

Guy L Guthrie

David John Krolak

John Thomas Hollaway, Jr

Bartholomew Blaner

Paul Allen Ganfield

Robert Christopher Dixon

Lonny Lambrecht

Daniel C Howe

Lakshminarayana Baba Arimilli

Srinivasan Ramani

Jeff A Stuecheli

Srinivas Bangalore Purushotham

Derek E Williams

Bernard Charles Drerup

Eric Eugene Retter

Prasanna Jayaraman

John David Irish

Mark L Rudquist

Harrison M McCreary

John Thomas Holloway, Jr

Ravi Kumar Arimilli

David Allen Hrusecky

Robert Alan Cargnoni

Chuck Hong Ngai

Daniel J Buerkle

Sooraj R Nair

Gary Alan Gorman

Julie Ann Rosser

John William Urda

A symmetric multiprocessor includes with a hierarchical ring-based interconnection network is disclosed. The symmetric processor includes a plurality of buses comprised on the symmetric multiprocessor, wherein each of the buses are configured in a circular topology. The symmetric multiprocessor also includes a plurality of multi-processing nodes interconnected by the buses to make a hierarchical ring-based interconnection network for conveying commands between the multi-processing nodes. The interconnection network includes a command network configured to transport commands based on command tokens, wherein the tokens dictate a destination of the command, a partial response network configured to transport partial responses generated by the multi-processing nodes, and a combined response network configured to combine the partial responses generated by the multi-processing nodes using combined response tokens.

Hierarchical ring-based interconnection network for symmetric multiprocessors

Data processing in a data processing system including a plurality of processing nodes coupled to an interconnect includes receiving, by a fabric controller, a first command from a remote processing node via the interconnect. The fabric controller determines that the command includes a replay indication, the replay indication indicative of a replay event at one or more processing nodes of the plurality of processing nodes. The first command is dropped from a deskew buffer of the fabric controller responsive to the determining that the command includes the replay indication.

Link-level cyclic redundancy check replay for non-blocking coherence flow

A technique for operating a data processing system includes determining, by an arbiter of a processing unit of the data processing system, whether an over-commit has occurred. In response to determining that the over-commit has occurred, the arbiter selects a broadcast command to be dropped based on a number of hops traversed through the data processing system by the broadcast command.

Techniques for command arbitation in symmetric multiprocessor systems

Data processing in a data processing system including a plurality of processing nodes coupled by a communication link includes receiving a first command from a first processing node. A link stall of the communication link is detected by a first link layer of the first processing node. A stop command is received at a first transaction layer of the first processing node from the first link layer. The first command is truncated by the first transaction layer into a first truncated command responsive to receiving the stop command. A command arbiter is instructed to stop issuing new commands. The first truncated command is forwarded to an asynchronous crossing buffer of the first processing node.

Dropped command truncation for efficient queue utilization in multiprocessor data processing system

Managing starvation in a distributed arbitration scheme including sending, by a starved local arbiter, a starvation message toward a head arbiter, wherein the starvation message comprises an identifier of the starved local arbiter and a request for resources to transfer data to a destination, wherein the requested resources comprise a destination token and a bus slot; receiving, by a neighboring local arbiter, the starvation message, wherein the neighboring local arbiter is between the starved local arbiter and the head arbiter; if the neighboring local arbiter currently has the resources requested in the starvation message: marking, by the neighboring local arbiter, the requested resources with the identifier of the starved local arbiter.

Managing starvation in a distributed arbitration scheme

A technique for operating a data processing system includes detecting that a processing unit within a first group of processing units in the data processing system has a hang condition. In response to detecting that the processing unit has a hang condition, a command issue rate for the first group of processing units is reduced. One or more other groups of processing units in the data processing system are notified that the first group of processing units has reduced the command issue rate for the first group of processing units. In response to the notifying, respective command issue rates of the other groups of processing units are reduced to reduce a number of commands received by the first group of processing units from the other groups of processing units.

Techniques for managing a hang condition in a data processing system with shared memory

A coherency error detection and reporting mechanism monitors for coherency errors in a processor and between processors. When a requestor broadcasts a memory address in a command and a coherency error is detected, information regarding the command that caused the coherency error is logged, and the coherency error is reported a system error handler. The information logged for the coherency error may include the address of the coherency error, the requestor, the command, the response to the command, the scope of the coherency error, the error type, etc. Logging information relating to the coherency error provides more information to a person analyzing the processor for failures to more easily track down the cause of coherency errors.

Coherency error detection and reporting in a processor

A method for monitoring performance of events occurring in a multiprocessor system is provided where the performance monitoring units (PMUs) are globally synchronized. The global synchronization is carried out with a dedicated bit field set to any of pause, stop, restart, or reset command. The command is sent across the scan communications interface (SCOM) of all chips by using existing fabric connecting all nest units to control the PMUs in the system. A pre-scale counter before a main counter may be used to buffer event counts until a reset or a restart command is sent to the SCOM in the system.

Processor performance monitoring unit synchronization

A request to send a message from a first component, located on a first processor, to a second component, located on a second processor, is received. It is determined that the second processor can be communicated with via a first bidirectional communication path. It is determined that bandwidth is available on the first bidirectional communication path. It is determined that bandwidth is available on a second bidirectional communication path. In response to a determination that bandwidth is available on the second bidirectional communication path, a data path is created between the first component and the second bidirectional communication path and the request to send the message to the second component is granted. In response to a determination that bandwidth is not available on the first bidirectional communication path or on the second bidirectional communication path, the grant of the request to send the message to the second component is delayed.

Intercomponent data communication between different processors

In one or more embodiments, one or more systems, devices, methods, and/or processes described can continually increase a command rate of an interconnect if one or more requests to lower the command rate are not received within one or more periods of time. In one example, the command rate can be set to a fastest level. In another example, the command rate can be incrementally increased over periods of time. If a request to lower the command rate is received, the command rate can be set to a reference level or can be decremented to one slower rate level. In one or more embodiments, the one or more requests to lower the command rate can be based on at least one of an issue rate of speculative commands and a number of overcommit failures, among others.

Command rate configuration in data processing system

A request to send a first message from a first component to a second component is received at an arbiter. The first component is located in a first time zone and the second component is located in a second time zone. The arbiter determines that the second component is located in the second time zone. It is determined that the second time zone can be communicated with via one or more communications channels in a first direction. It is determined whether bandwidth is available on the one or more communications channels in the first direction. If bandwidth is available on the one or more communications channels in the first direction, a data path between the first component and the one or more communications channels in the first direction is created and the request is granted. Otherwise, the grant of the request is delayed.

Intercomponent data communication

Intercomponent data communication between multiple time zones

A method of configuring a prescaling circuit in a performance counter circuit for a computer processing system can include receiving a first number of signaled events at a prescaling circuit configured to generate event counts for a performance counter circuit. The method can include generating event counts at a current event-count rate for the first number of signaled events and determining a detected event-count rate for the signaled events based on a rate at which the first number of signaled events are received at the prescaling circuit. The method can include determining that the detected event-count rate is greater than the current event-count rate. The method can include increasing the current event-count rate in response to determining that the detected event-count rate is greater than the current event-count rate.

Dynamic prescaling for performance counters

Growing community of inventors

Round Rock, TX, United States of America

Charles Francis Marino