International Business Machines Corporation

Harmony Lynn Helterhoff

Guy L Guthrie

Phillip G Williams

William John Starke

Jeffrey A Stuecheli

Kevin Franklin Reick

Robert Alan Cargnoni

Derek E Williams

Michael Andrew Blake

Craig Raymond Walters

Vesselina K Papazova

Arthur J O'Neill

Alvan Wing Ng

David W Cummings

Thomas L Jeremiah

Philip G Williams

In response to a data request, a victim cache line is selected for castout from a lower level cache, and a target lower level cache of one of the plurality of processing units is selected. A determination is made whether the selected target lower level cache has provided more than a threshold number of retry responses to lateral castout (LCO) commands of the first lower level cache, and if so, a different target lower level cache is selected. The first processing unit thereafter issues a LCO command on the interconnect fabric. The LCO command identifies the victim cache line to be castout and indicates that the target lower level cache is an intended destination of the victim cache line. In response to a successful coherence response to the LCO command, the victim cache line is removed from the first lower level cache and held in the second lower level cache.

Empirically based dynamic control of transmission of victim cache lateral castouts

A second lower level cache receives an LCO command issued by a first lower level cache on an interconnect fabric. The LCO command indicates an address of a victim cache line to be castout from the first lower level cache and indicates that the second lower level cache is an intended destination of the victim cache line. The second lower level cache determines whether to accept the victim cache line from the first lower level cache based at least in part on the address of the victim cache line indicated by the LCO command. In response to determining not to accept the victim cache line, the second lower level cache provides a coherence response to the LCO command refusing the identified victim cache line. In response to determining to accept the victim cache line, the second lower level cache updates an entry corresponding to the identified victim cache line.

Empirically based dynamic control of acceptance of victim cache lateral castouts

In response to a data request of a first of a plurality of processing units, the first processing unit selects a victim cache line to be castout from the lower level cache of the first processing unit and determines whether a mode is set. If not, the first processing unit issues on the interconnect fabric an LCO command identifying the victim cache line and indicating that a lower level cache is the intended destination. If the mode is set, the first processing unit issues a castout command with an alternative intended destination. In response to a coherence response to the LCO command indicating success of the LCO command, the first processing unit removes the victim cache line from its lower level cache, and the victim cache line is held elsewhere in the data processing system. The mode can be set to inhibit castouts to system memory, for example, for testing.

Mode-based castout destination selection

A processing unit includes a processor core and a cache memory coupled to the processor core. The cache memory includes a data array, a directory of the data array, error detection logic that sequentially detects a first, second and third correctable errors in the data array of the cache memory and provides indications of detection of the first, second and third correctable errors, and control circuitry that, responsive to the indication of the third correctable error and an indication that the first and second correctable errors occurred at too high a frequency, marks an entry of the data array containing a cache line having the third correctable error as deleted in the directory of the cache memory regardless of which entry of the data array contains a cache line having the second correctable error.

Delete of cache line with correctable error

In response to a data request of a first processing unit among a plurality of processing units, the first processing unit selects a victim cache line to be castout from the lower level cache of the first processing unit and selects the lower level cache of a second of the plurality of processing units as an intended destination of a lateral castout (LCO) command by randomized round-robin selection. The first processing unit issues on the interconnect fabric an LCO command identifying the victim cache line and the intended destination. In response to a coherence response to the LCO command indicating success of the LCO command, the first processing unit removes the victim cache line from its lower level cache, and the victim cache line is held in the lower level cache of one of the plurality of processing units other than the first processing unit.

Lateral castout target selection

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.

Apparatus and method for improved data persistence within a multi-node system

A data processing system employing a weakly ordered storage architecture includes first and second sets of processing units coupled to each other and data storage by an interconnect fabric. Each processing unit has a processor core having an associated cache hierarchy including at least a level one, level two and level three cache memories. A request to perform an update to a portion of a first image of memory contained in the level three cache memory of a first processing unit while at last one kill-type command is pending at the first processing unit, the cache hierarchy of the first processing unit permitting the update to be exposed to any first processor core only after the at least one kill-type command is complete.

Barriers processing in a multiprocessor system having a weakly ordered storage architecture without broadcast of a synchronizing operation

A method of reducing errors in a cache memory of a computer system (e.g., an L2 cache) by periodically issuing a series of purge commands to the L2 cache, sequentially flushing cache lines from the L2 cache to an L3 cache in response to the purge commands, and correcting errors (single-bit) in the cache lines as they are flushed to the L3 cache. Purge commands are issued only when the processor cores associated with the L2 cache have an idle cycle available in a store pipe to the cache. The flush rate of the purge commands can be programmably set, and the purge mechanism can be implemented either in software running on the computer system, or in hardware integrated with the L2 cache. In the case of the software, the purge mechanism can be incorporated into the operating system. In the case of hardware, a purge engine can be provided which advantageously utilizes the store pipe that is provided between the L1 and L2 caches. The L2 cache can be forced to victimize cache lines, by setting tag bits for the cache lines to a value that misses in the L2 cache (e.g., cache-inhibited space). With the eviction mechanism of the cache placed in a direct-mapped mode, the address misses will result in eviction of the cache lines, thereby flushing them to the L3 cache.

Growing community of inventors

Austin, TX, United States of America

Harmony Lynn Helterhoff