Apple Inc.

Google Inc.

Xiaoming Wang

Harshavardhan Kaushikkar

Srinivasa Rangan Sridharan

Sridhar Kotha

Prashant R Chandra

Weihuang Wang

Gurushankar Rajamani

Michael Roger Marty

Nandita Dukkipati

Srinivas Vaduvatha

Abhishek Agarwal

Jiazhen Zheng

Neeraj Parik

Rong Zhang Hu

Yu Zhang

Gautam Kumar

Hassan Wassel

Nicholas McDonald

Neelesh Bansod

Horia Alexandru Toma

Le Wang

Albert Forte Magyar

Spoorthy Nanjaiah

Nicholas Mcdonald

The mapping of system memory addresses to physical memory addresses is modeled as a two dimensional mapping array. Each element of the mapping array is assigned a system memory address and a physical memory address to which the system memory address is mapped. The mapping array is arranged to facilitate designation of a portion of the physical memory addresses as spareable physical memory addresses that are employed when there is a memory failure.

Memory sparing to improve chip reliability

An application specific integrated circuit (ASIC) is provided for reliable transport of packets. The network interface card may include a reliable transport accelerator (RTA). The RTA may include a cache lookup database. The RTA may be configured to determine, from a received data packet, a connection identifier and query the cache lookup database for a cache entry corresponding to a connection context having the connection identifier. In response to the query, the RTA may receive a cache hit or a cache miss.

Off-chip memory backed reliable transport connection cache hardware architecture

A system includes a first processor configured to analyze packets received over a communication protocol system and determine one or more congestion indicators from the analysis of the data packets, the one or more congestion indicators being indicative of network congestion for data packets transmitted over a reliable transport protocol layer of the communication protocol system. The system also includes a rate update engine separate from the packet datapath and configured to operate a second processor to receive the determined one or more congestion indicators, determine one or more congestion control parameters for controlling transmission of data packets based on the received one or more congestion indicators, and output a congestion control result based on the determined one or more congestion control parameters.

Rate update engine for reliable transport protocol

A system and method for efficiently allocating resources of destinations to sources conveying requests to the destinations. In various embodiments, a computing system includes multiple sources that generate requests and multiple destinations that service the requests. One or more transaction tables store requests received from the multiple sources. Arbitration logic selects requests and stores them in a processing table. When the logic selects a given request from the processing table, and determines resources for the corresponding destination is unavailable, the logic removes the given request from the processing table and allocates the request in a retry handling queue. When the retry handling queue has no data storage for the request, logic updates a transaction table entry and maintains a count of such occurrences. When the count exceeds a threshold, the logic stalls requests for that source. Requests in the retry handling queue have priority over requests in the transaction tables.

Resource access management

Systems, apparatuses, and methods for reducing memory cache control command hops through a fabric are disclosed. A system includes an interconnect fabric, a plurality of transaction processing queues, and a plurality of memory pipelines. Each memory pipeline includes an arbiter, a combined coherence point and memory cache controller unit, and a memory controller coupled to a memory channel. Each combined unit includes a memory cache controller, a memory cache, and a duplicate tag structure. A single arbiter per memory pipeline performs arbitration across the transaction processing queues to select a transaction address to feed the memory pipeline's combined unit. The combined unit performs coherence operations and a memory cache lookup for the selected transaction. Only after processing is completed in the combined unit is the transaction moved out of its transaction processing queue, reducing power consumption caused by data movement through the fabric.

Reducing memory cache control command hops on a fabric

Systems, apparatuses, and methods for implementing a distributed global ordering point are disclosed. A system includes at least a communication fabric, sequencing logic, and a plurality of coherence point pipelines. Each coherence point pipeline receives transactions from the communication fabric and then performs coherence operations and a memory cache lookup for the received transactions. The global ordering point of the system is distributed across the outputs of the separate coherence point pipelines. Device-ordered transactions travelling upstream toward memory are assigned sequence numbers by the sequencing logic. The transactions are speculatively issued from the communication fabric to the coherence point pipelines. Speculatively issuing the transactions to the coherence point pipelines may cause the transactions to pass through the distributed global ordering point out of order. Control logic on the downstream path reorders the transactions based on the assigned sequence numbers.

Systems and methods for providing distributed global ordering

Systems, apparatuses, and methods for implementing coherence flows for dual-processing coherence and memory cache pipelines are disclosed. A dual-processing pipeline includes a coherence processing pipeline and a memory cache processing pipeline. When a transaction is issued to the dual-processing pipeline, the coherence processing pipeline performs a duplicate tag lookup in parallel with the memory cache processing pipeline performing a memory cache tag lookup for the transaction. If the duplicate tag lookup is a hit, then the coherence processing pipeline locks the matching entry, the memory cache processing pipeline discards the original transaction, and a copyback request is sent to a coherent agent identified by the matching entry. When the copyback response is received by a communication fabric, the copyback response is issued to the memory cache processing pipeline. When the copyback response passes the global ordering point, the coherence processing pipeline clears the lock on the matching entry.

Coherence flows for dual-processing pipelines

An embodiment of an apparatus includes a retry queue circuit, a transaction arbiter circuit, and a plurality of transaction buffers. The retry queue circuit may store one or more entries corresponding to one or more memory transactions. A position in the retry queue circuit of an entry of the one or more entries may correspond to a priority for processing a memory transaction corresponding to the entry. The transaction arbiter circuit may receive a real-time memory transaction from a particular transaction buffer. In response to a determination that the real-time memory transaction is unable to be processed, the transaction arbiter circuit may create an entry for the real-time memory transaction in the retry queue circuit. In response to a determination that a bulk memory transaction is scheduled for processing prior to the real-time memory transaction, the transaction arbiter circuit may upgrade the bulk memory transaction to use real-time memory resources.

Real-time resource handling in resource retry queue

Growing community of inventors

Sunnyvale, CA, United States of America

Xiaoming Wang