International Business Machines Corporation

Curtis Carl Wollbrink

Ronald Ernest Freking

Elizabeth A McGlone

Steven Mark Thurber

David R Engebretsen

Daniel Ray Spach

Adalberto Guillermo Yanes

William Joseph Armstrong

Gregory Michael Nordstrom

Scott Neil Dunham

William John Starke

Jeffrey A Stuecheli

Eric J Campbell

Michael Steven Siegel

Lakshminarayana Baba Arimilli

Bartholomew Blaner

Christopher W Steffen

John David Irish

Kenneth Michael Valk

Brian Thomas Vanderpool

Chad M Albertson

Nicholas Ollerich

Jeffrey Buchanan Williams

Nicholas V Tram

Daniel Mark Dreps

Daniel Frank Moertl

Danny Marvin Neal

David John Krolak

Mehul M Shah

Philip Rogers Hillier, Iii

Scott Michael Willenborg

Yiftach Benjamini

Lonny Lambrecht

Pat Allen Buckland

John L Pike

An integrated circuit includes a first communication interface for communicatively coupling the integrated circuit with a coherent data processing system, a second communication interface for communicatively coupling the integrated circuit with an accelerator unit including an effective address-based accelerator cache for buffering copies of data from a system memory, and a real address-based directory inclusive of contents of the accelerator cache. The real address-based directory assigns entries based on real addresses utilized to identify storage locations in the system memory. The integrated circuit further includes directory control logic that configures at least a number of congruence classes utilized in the real address-based directory based on configuration parameters specified on behalf of or by the accelerator unit.

Integrated circuit and data processing system having a configurable cache directory for an accelerator

Low latency corrupt data tagging on a cross-chip link including receiving, from the cross-chip link, a control flit comprising a virtual channel identifier for an incoming data flit; storing the virtual channel identifier in a data pipeline and a bad data indicator (BDI) pipeline; receiving, from the cross-chip link, the incoming data flit into the data pipeline; moving, based on the virtual channel identifier in the data pipeline, the data flit from the data pipeline into an entry in a virtual channel queue corresponding to the virtual channel identifier; receiving, from the cross-chip link, a BDI for the data flit into the BDI pipeline; and moving, based on the virtual channel identifier in the BDI pipeline, the BDI for the data flit from the BDI pipeline into an entry in a BDI array corresponding to the entry in the virtual channel queue storing the data flit.

Low latency corrupt data tagging on a cross-chip link

A serial communication system includes a transmitting circuit for serially transmitting data via a serial communication link including N channels where N is an integer greater than 1. The transmitting circuit includes an input buffer having storage for input data frames each including M bytes forming N segments of M/N contiguous bytes. The transmitting circuit additionally includes a reordering circuit coupled to the input buffer. The reordering circuit includes a reorder buffer including multiple entries. The reordering circuit buffers, in each of multiple entries of the reorder buffer, a byte in a common byte position in each of the N segments of an input data frame. The reordering circuit sequentially outputs the contents of the entries of the reorder buffer via the N channels of the serial communication link.

Pre-transmission data reordering for a serial interface

A sideband PCI Express (PCIe) packet initiator in a distributed PCIe switch fabric verifies a PCIe connection between a host device and a PCIe endpoint device without having to power on the host device. The packet initiator assembles a PCIe test packet that acts as a ping for testing reachability of the endpoint device, from the perspective of the host device. The test packet may also verify configurations and settings of the path to the endpoint device. The distributed switch fabric is configured to compare completion data with expected results to verify the PCIe connection, without having to boot the host device.

Peripheral component interconnect express (PCIe) ping in a switch-based environment

Techniques for broadcasting a command in a distributed switch, at a first switch module within the distributed switch. Embodiments receive a request to reset a PCIe link for a first host device, the first host device connected to a plurality of downstream PCIe devices through the distributed switch. A routing table specifying a plurality of downstream switch modules, connected to the first switch modules by one or more ports of the first switch module, is identified. Embodiments suspend PCIe traffic for the first host device on the one or more ports of the first switch module. Broadcast messages are transmitted to the plurality of downstream switch modules, specifying a first reset operation. Upon receiving an acknowledgement message from each of the plurality of downstream switch modules specified in the routing table, embodiments resume PCIe traffic for the first host device on the one or more ports.

Broadcast for a distributed switch network

Each PCIe device may include a media access control (MAC) interface and a physical (PHY) interface that support a plurality of different lane configurations. These interfaces may include hardware modules that support 1&#xd7;32, 2&#xd7;16, 4&#xd7;8, 8&#xd7;4, 16&#xd7;2, and 32&#xd7;1 communication. Instead of physically connecting each of the hardware modules in the MAC interface to respective hardware modules in the PHY interface using dedicated traces, the device may include two bus controllers that arbitrate which hardware modules are connected to a internal bus coupling the two interfaces. When a different lane configuration is desired, the bus controller couples the corresponding hardware module to the internal bus. In this manner, the different lane configurations share the same lanes (and wires) of the bus as the other lane configurations. Accordingly, the shared bus only needs to include enough lanes (and wires) necessary to accommodate the widest lane configuration.

Versatile lane configuration using a PCIe PIE-8 interface

Method for performing an operation to maintain data integrity in a parallel computing system, the operation comprising providing a lookup table specifying a plurality of predefined destinations for data packets, receiving a first data packet comprising a destination address specifying a first destination, wherein the first data packet has an error of a first type, identifying, from the lookup table, an entry specifying a second destination for data packets having errors of the first type, and sending the first data packet to the second destination.

External settings that reconfigure the error handling behavior of a distributed PCIe switch

The standard hot-plug controller (SHPC) specification may be used to generate PCI messages in a distributed switch to disconnect and/or connect virtual hierarchies of an endpoint from hosts that are connected based on multi-root input/output virtualization (MR-IOV). A management controller may instruct a SHPC to generate a PCI packet that specifies a particular virtual hierarchy to disconnect from a particular host. An upstream port connected to the host and the SHPC receives the PCI packet and uses a header that identifies the virtual endpoint in the packet to index into a routing table to identify a downstream port in the distributed switch that is connected to the endpoint. Once the PCI packet traverses the switch and arrives at the downstream port, the downstream port changes routing logic which logically disconnects the host from the specified virtual hierarchy.

Using a PCI standard hot plug controller to modify the hierarchy of a distributed switch

Versatile lane configuration using a PCIe PIe-8 interface

Method, circuit, and system for performing an operation for regulating bandwidth, the operation comprising receiving at a memory, debug data packets and functional data packets for transmittal on a shared bus. The operation then transmits, via the shared bus, the functional data packets and one or more of the debug data packets according to a predefined ratio of debug data packets to functional data packets. The operation then drops one or more of the received debug data packets at the memory, and maintains a count of the one or more dropped debug data packets. The operation then updates the predefined ratio based on the count, and uses the updated predefined ratio to transmit the functional data packets and one or more of the debug data packets.

Bandwidth limiting on generated PCIe packets from debug source

Method, computer program product, and system for performing an operation to maintain data integrity in a parallel computing system, the operation comprising providing a lookup table specifying a plurality of predefined destinations for data packets, receiving a first data packet comprising a destination address specifying a first destination, wherein the first data packet has an error of a first type, identifying, from the lookup table, an entry specifying a second destination for data packets having errors of the first type, and sending the first data packet to the second destination.

Bandwidth limiting on generated PCIE packets from debug source

Growing community of inventors

Rochester, MN, United States of America

Curtis Carl Wollbrink