International Business Machines Corporation

Qualcomm Incorporated

Kabushiki Kaisha Toshiba

International Business Machines Corpation

Thuong Quang Truong

Charles Ray Johns

Michael Norman Day

James Allan Kahle

David James Shippy

Peichun Peter Liu

Harm Peter Hofstee

Eric Francis Robinson

Takeshi Yamazaki

Hien Le

Jonathan James DeMent

Jason Alan Cox

Todd Swanson

Peichum Peter Liu

Robert John Dorsey

Shigehiro Asano

Russell Dean Hoover

Srinivasan Ramani

Jeffrey Douglas Brown

John Samuel Liberty

Richard Nicholas

Jaya Prakash Subramaniam Ganasan

Alvan Wing Ng

Etai Adar

Roy Moonseuk Kim

Bernard Charles Drerup

Gary Dale Carpenter

Brian Mitchell Bass

Kun Xu

David John Krolak

Michael Fan Wang

Scott Douglas Clark

Prasanna Srinivasan

Yasukichi Okawa

Kerey Michelle Tassin

Cesar Aaron Ramirez

Sang Hoo Dhong

Mark Richard Nutter

Thomas Philip Speier

Ronald P Hall

Richard William Doing

Joseph Gerald McDonald

Thomas L Jeremiah

Gurushankar Rajamani

Mark J Wolski

Steven Robert Testa

Randall John Pascarella

Kevin C Stelzer

Keith A DeWeese

Asano Shigehiro

John R Patty

Joseph Gerald Mcdonald

Hien Minh Le

Kevin Stelzer

Aspects disclosed herein include avoiding deadlocks in processor-based systems employing retry and in-order-response non-retry bus coherency protocols. In this regard, an interface bridge circuit is communicatively coupled to a first core device that implements a retry bus coherency protocol, and a second core device that implements an in-order-response non-retry bus coherency protocol. The interface bridge circuit receives a snoop command from the first core device, and forwards the snoop command to the second core device. While the snoop command is pending, the interface bridge circuit detects a potential deadlock condition between the first core device and the second core device. In response to detecting the potential deadlock condition, the interface bridge circuit is configured to send a retry response to the first core device. This enables the first core device to continue processing, thereby eliminating the potential deadlock condition.

Avoiding deadlocks in processor-based systems employing retry and in-order-response non-retry bus coherency protocols

Maintaining cache coherency using conditional intervention among multiple master devices is disclosed. In one aspect, a conditional intervention circuit is configured to receive intervention responses from multiple snooping master devices. To select a snooping master device to provide intervention data, the conditional intervention circuit determines how many snooping master devices have a cache line granule size the same as or larger than a requesting master device. If one snooping master device has a same or larger cache line granule size, that snooping master device is selected. If more than one snooping master device has a same or larger cache line granule size, a snooping master device is selected based on an alternate criteria. The intervention responses provided by the unselected snooping master devices are canceled by the conditional intervention circuit, and intervention data from the selected snooping master device is provided to the requesting master device.

Maintaining cache coherency using conditional intervention among multiple master devices

Bridging strongly ordered write transactions to devices in weakly ordered domains, and related apparatuses, methods, and computer-readable media are disclosed. In one aspect, a host bridge device is configured to receive strongly ordered write transactions from one or more strongly ordered producer devices. The host bridge device issues the strongly ordered write transactions to one or more consumer devices within a weakly ordered domain. The host bridge device detects a first write transaction that is not accepted by a first consumer device of the one or more consumer devices. For each of one or more write transactions issued subsequent to the first write transaction and accepted by a respective consumer device, the host bridge device sends a cancellation message to the respective consumer device. The host bridge device replays the first write transaction and the one or more write transactions that were issued subsequent to the first write transaction.

Bridging strongly ordered write transactions to devices in weakly ordered domains, and related apparatuses, methods, and computer-readable media

Counter registers are shared among multiple threads executing on multiple processor cores. An event within the processor core is selected. A multiplexer in front of each of a number of counters is configured to route the event to a counter. A number of counters are assigned for the event to each of a plurality of threads running for a plurality of applications on a plurality of processor cores, wherein each of the counters includes a thread identifier in the interrupt thread identification field and a processor identifier in the processor identification field. The number of counters is configured to have a number of interrupt thread identification fields and a number of processor identification fields to identify a thread that will receive a number of interrupts.

Performance monitor design for instruction profiling using shared counters

A number of hypervisor register fields are set to specify which processor cores are allowed to generate a number of performance events for a particular thread group. A plurality of threads for an application running in the computing environment to a plurality of thread groups are configured by a plurality of thread group fields in a plurality of control registers. A number of counter sets are allowed to count a number of thread group events originating from one of a shared resource and a shared cache are specified by a number of additional hypervisor register fields.

Performance monitor design for counting events generated by thread groups

Sampled instruction address registers are shared among multiple threads executing on a plurality of processor cores. Each of a plurality of sampled instruction address registers are assigned to a particular thread running for an application on the plurality of processor cores. Each of the sampled instruction address registers are configured by storing in each of the sampled instruction address registers a thread identification of the particular thread in a thread identification field and a processor identification of a particular processor on which the particular thread is running in a processor identification field.

Sharing sampled instruction address registers for efficient instruction sampling in massively multithreaded processors

A method for maintaining cache coherency operates in a data processing system with a system memory and a plurality of processing units (PUs), each PU having a cache, and each PU coupled to at least another one of the plurality of PUs. A first PU receives a first data block for storage in a first cache of the first PU. The first PU stores the first data block in the first cache. The first PU assigns a first coherency state and a first tag to the first data block, wherein the first coherency state is one of a plurality of coherency states that indicate whether the first PU has accessed the first data block. The plurality of coherency states further indicate whether, in the event the first PU has not accessed the first data block, the first PU received the first data block from a neighboring PU.

System and method for cache coherency in a multiprocessor system

A method for managing data operates in a data processing system with a system memory and a plurality of processing units (PUs), each PU having a cache comprising a plurality of cache lines, each cache line having one of a plurality of coherency states, and each PU coupled to at least another one of the plurality of PUs. A first PU selects a castout cache line of a plurality of cache lines in a first cache of the first PU to be castout of the first cache. The first PU sends a request to a second PU, wherein the second PU is a neighboring PU of the first PU, and the request comprises a first address and first coherency state of the selected castout cache line. The second PU determines whether the first address matches an address of any cache line in the second PU. The second PU sends a response to the first PU based on a coherency state of each of a plurality of cache lines in the second cache and whether there is an address hit. The first PU determines whether to transmit the castout cache line to the second PU based on the response. And, in the event the first PU determines to transmit the castout cache line to the second PU, the first PU transmits the castout cache line to the second PU.

System and method for optimizing neighboring cache usage in a multiprocessor environment

A system for communicating command parameters between a processor and a memory flow controller is provided. The system makes use of a channel interface as the primary mechanism for communicating between the processor and a memory flow controller. The channel interface provides channels for communicating with processor facilities, memory flow control facilities, machine state registers, and external processor interrupt facilities, for example. These channels may be designated as blocking or non-blocking. With blocking channels, when no data is available to be read from the corresponding registers, or there is no space available to write to the corresponding registers, the processor is placed in a low power 'stall' state. The processor is automatically awakened, via communication across the blocking channel, when data becomes available or space is freed. Thus, the channels of the present invention permit the processor to stay in a low power state.

System for communicating command parameters between a processor and a memory flow controller

A method, and apparatus are provided for preventing livelocks in processor selection of load requests in a multiprocessor (MP) system. On random occasions a selection mechanism is changed for first holding up all requests and then a random selection is made. Then a round robin selection mechanism is used for further requests. A livelock-preventing selection mechanism includes a pair of linear feedback shift registers (LFSRs), each LFSR for generating pseudo random values.

Preventing livelocks in processor selection of load requests

A write-through cache scheme is created. A store data command is sent to a cache line of a cache array from a processing unit. It is then determined whether the address of the store data is valid, wherein the original data from the store's address has been previously loaded into the cache. A write-through command is sent to a system bus as a function of whether the address of the store data is valid. The bus controller is employed to sense the write-through command. If the write-through command is sensed, a clean command is generated by the bus controller. If the write-through command is sensed, the store data is written into the cache array, and the data is marked as modified. If the write-through command is sensed, the clean command is sent onto the system bus by the bus controller, thereby causing modified data to be written to memory.

Bus controller initiated write-through mechanism with hardware automatically generated clean command

A mechanism for communicating instructions and data between a processor and external devices are provided. The mechanism makes use of a channel interface as the primary mechanism for communicating between the processor and a memory flow controller. The channel interface provides channels for communicating with processor facilities, memory flow control facilities, machine state registers, and external processor interrupt facilities, for example. These channels may be designated as blocking or non-blocking. With blocking channels, when no data is available to be read from the corresponding registers, or there is no space available to write to the corresponding registers, the processor is placed in a low power 'stall' state. The processor is automatically awakened, via communication across the blocking channel, when data becomes available or space is freed. Thus, the channels of the present invention permit the processor to stay in a low power state.

Communicating instructions and data between a processor and external devices

A method for managing a cache operates in a data processing system with a system memory and a plurality of processing units (PUs). A first PU determines that one of a plurality of cache lines in a first cache of the first PU must be replaced with a first data block, and determines whether the first data block is a victim cache line from another one of the plurality of PUs. In the event the first data block is not a victim cache line from another one of the plurality of PUs, the first cache does not contain a cache line in coherency state invalid, and the first cache contains a cache line in coherency state moved, the first PU selects a cache line in coherency state moved, stores the first data block in the selected cache line and updates the coherency state of the first data block.

System and method for cache line replacement selection in a multiprocessor environment

A cache coherency technique used in a multi-node symmetric multi-processor system that reduces the number of message phases of a read request from 5 to 4, canceling the combined response phase for read requests in most cases, thereby improving system performance and reducing the overall system power consumption.

Combined response cancellation for load command

The present invention provides for atomic update primitives in an asymmetric single-chip heterogeneous multiprocessor computer system having a shared memory with DMA transfers. At least one lock line command is generated from a set comprising a get lock line command with reservation, a put lock line conditional command, and a put lock line unconditional command.

Method to provide atomic update primitives in an asymmetric heterogeneous multiprocessor environment

Growing community of inventors

Austin, TX, United States of America

Thuong Quang Truong