Abstract: Main memory operation in a symmetric multiprocessing computer, the computer comprising one or more processors operatively coupled through a cache controller to at least one cache of main memory, the main memory shared among the processors, the computer further comprising input/output (‘I/O’) resources, including receiving, in the cache controller from an issuing resource, a memory instruction for a memory address, the memory instruction requiring writing data to main memory; locking by the cache controller the memory address against further memory operations for the memory address; advising the issuing resource of completion of the memory instruction before the memory instruction completes in main memory; issuing by the cache controller the memory instruction to main memory; and unlocking the memory address only after completion of the memory instruction in main memory.

Abstract: Various embodiments of the present invention mitigate busy time in a hierarchical store-through memory cache structure. In one embodiment, a cache directory associated with a memory cache is divided into a plurality of portions each associated with a portion memory cache. Simultaneous cache lookup operations and cache write operations between the plurality of portions of the cache directory are supported. Two or more store commands are simultaneously processed in a shared cache pipeline communicatively coupled to the plurality of portions of the cache directory.

Abstract: A pipelined processing device includes: a processor configured to receive a request to perform an operation; a plurality of processing controllers configured to receive at least one instruction associated with the operation, each of the plurality of processing controllers including a memory to store at least one instruction therein; a pipeline processor configured to receive and process the at least one instruction, the pipeline processor including shared error detection logic configured to detect a parity error in the at least one instruction as the at least one instruction is processed in a pipeline and generate an error signal; and a pipeline bus connected to each of the plurality of processing controllers and configured to communicate the error signal from the error detection logic.

Abstract: Dynamic pipeline cache error correction includes receiving a request to perform an operation that requires a storage cache slot, the storage cache slot residing in a cache. The dynamic pipeline cache error correction also includes accessing the storage cache slot, determining a cache hit for the storage cache slot, identifying and correcting any correctable soft errors associated with the storage cache slot. The dynamic cache error correction further includes updating the cache with results of corrected data.

Abstract: A memory refresh requestor, a memory request interpreter, a cache memory, and a cache controller on a single chip. The cache controller configured to receive a memory access request, the memory access request for a memory address range in the cache memory, detect that the cache memory located at the memory address range is available, and send the memory access request to the memory request interpreter when the memory address range is available. The memory request interpreter configured to receive the memory access request from the cache controller, determine if the memory access request is a request to refresh a contents of the memory address range, and refresh data in the memory address range when the memory access request is a request to refresh memory.

Abstract: A method of managing a temporary memory includes: receiving a request to transfer data from a source location to a destination location, the data transfer request associated with an operation to be performed, the operation selected from an input into an intermediate temporary memory and an output; checking a two-state indicator associated with the temporary memory, the two-state indicator having a first state indicating that an immediately preceding operation on the temporary memory was an input to the temporary memory and a second state indicating that the immediately preceding operation was an output from the temporary memory; and performing the operation responsive to one of: the operation being an input operation and the two-state indicator being in the second state, indicating that the immediately preceding operation was an output; and the operation being an output operation and the two-state indicator being in the first state, indicating that the immediately preceding operation was an input.

Abstract: A method and apparatus in which the observability of cross-invalidates requests within remote nodes is controlled at the time of a partial response generation, when a remote request initially checks/snoops the directory state of the remote node, but before such the time that the cross-invalidate request is actually sent to the processors on a given node. If all of the remote nodes in the system indicate that the cross-invalidates could be sent during an initial directory snoop, the requesting node is able to return full exclusivity to a given cache line to a requesting processor at the time when it receives all of the partial responses, instead of having to wait for the final responses from each of the remote nodes within the system.

Abstract: A method of preventing lockout and stalling conditions in a multi-node system having a plurality of nodes which includes initiating a processor request to a shared level of cache in a requesting node, performing a fabric coherency establishment sequence on the plurality of nodes, issuing a speculative memory fetch request to a memory, detecting a conflict on one of the plurality of nodes and communicating the conflict back to the requesting node within the system, canceling the speculative memory fetch request issued, and repeating the fabric coherency establishment sequence in the system until the point of conflict is resolved, without issuing another speculative memory fetch request. The subsequent memory fetch request is only issued after determining the state of line within the system, after the successful completion of the multi-node fabric coherency establishment sequence.

Abstract: A modification of rank priority arbitration for access to computer system resources through a shared pipeline that provides more equitable arbitration by allowing a higher ranked request access to the shared resource ahead of a lower ranked requester only one time. If multiple requests are active at the same time, the rank priority will first select the highest priority active request and grant it access to the resource. It will also set a ‘blocking latch’ to prevent that higher priority request from re-gaining access to the resource until the rest of the outstanding lower priority active requesters have had a chance to access the resource.

Abstract: A two pipe pass method for least recently used (LRU) compartment capture in a multiprocessor system. The method includes receiving a fetch request via a requesting processor and accessing a cache directory based on the received fetch request, performing a first pipe pass by determining whether a fetch hit or a fetch miss has occurred in the cache directory, and determining an LRU compartment associated with a specified congruence class of the cache directory based on the fetch request received, when it is determined that a fetch miss has occurred, and performing a second pipe pass by using the LRU compartment determined and the specified congruence class to access the cache directory and to select an LRU address to be cast out of the cache directory.

Abstract: A method, system and computer program product for selectively purging entries in a cache of a computer system. The method includes determining a starting storage address and a length of the storage address range to be purged, determining preset values for a congruence class and a compartment of a cache directory, accessing the cache directory based on the preset value of the congruence class, and selecting an entry in the cache directory based on the preset value of the compartment, determining validity of the entry accessed by examining an ownership tag of the entry, comparing a line address of the entry with the starting storage address and a sum of the starting storage address and the length of the storage address range, and selectively purging the entry based on the comparison result.

Abstract: The method includes initiating a processor request to a cache in a requesting node and broadcasting the processor request to remote nodes when the processor request encounters a local cache miss, performing a directory search of each remote cache to determine a state of a target line's address and an ownership slate of a specified address, returning the state of the target line to the requesting node and forming a combined response, and broadcasting the combined response to each remote node. During a fetch operation, when the directory search indicates an IM or a Target Memory node on a remote node, data is sourced from the respective remote cache and forwarded to the requesting node while protecting the data, and during a store operation, the data is sourced from the requesting node and protected while being forwarded to the IM or the Target Memory node after coherency has been established.

Abstract: A method of preventing lockout and stalling conditions in a multi-node system having a plurality of nodes which includes initiating a processor request to a shared level of cache in a requesting node, performing a fabric coherency establishment sequence on the plurality of nodes, issuing a speculative memory fetch request to a memory, detecting a conflict on one of the plurality of nodes and communicating the conflict back to the requesting node within the system, canceling the speculative memory fetch request issued, and repeating the fabric coherency establishment sequence in the system until the point of conflict is resolved, without issuing another speculative memory fetch request. The subsequent memory fetch request is only issued after determining the state of line within the system, after the successful completion of the multi-node fabric coherency establishment sequence.

Abstract: Cache coherency latency is reduced through a method and apparatus that expedites the return of line exclusivity to a given processor in a multi-node data handling system through enhanced inter-node communications.

Abstract: Methods and means of error correction code (ECC) debugging may comprise detecting whether a bit error has occurred; determining which bit or bits were in error; and using the bit error information for debug. The method may further comprise comparing ECC syndromes against one or more ECC syndrome patterns. The method may allow for accumulating bit error information, comparing error bit failures against a pattern, trapping data, counting errors, determining pick/drop information, or stopping the machine for debug.

Abstract: A method of error correction code (ECC) debugging for a system comprising, receiving data having an ECC, determining whether a data error has occurred, generating a syndrome of an error result, decoding flipped data bits, processing the received data and the decoded flipped bits to correct the data, outputting corrected data having an ECC, receiving a trap update signal, and saving the decoded flipped data bits responsive to receiving the trap update signal.

Abstract: A modification of rank priority arbitration for access to computer system resources through a shared pipeline that provides more equitable arbitration by allowing a higher ranked request access to the shared resource ahead of a lower ranked requester only one time. If multiple requests are active at the same time, the rank priority will first select the highest priority active request and grant it access to the resource. It will also set a ‘blocking latch’ to prevent that higher priority request from re-gaining access to the resource until the rest of the outstanding lower priority active requesters have had a chance to access the resource.

Abstract: Improved access to retained data useful to a system is accomplished by managing data flow through cache associated with the processor(s) of a multi-node system. A data management facility operable with the processors and memory array directs the flow of data from the processors to the memory array by determining the path along which data evicted from a level of cache close to one of the processors is to return to a main memory and directing evicted data to be stored, if possible, in a horizontally associated cache.