International Business Machines Corporation

Philip E Milling

Patrick M Bland

Mark Edward Dean

Louis Bennie Capps, Jr

Ralph M Begun

Robert Earl Kruse

Warren E Price

A distributed system structure for a large-way, symmetric multiprocessor system using a bus-based cache-coherence protocol is provided. The distributed system structure contains an address switch, multiple memory subsystems, and multiple master devices, either processors, I/O agents, or coherent memory adapters, organized into a set of nodes supported by a node controller. The node controller receives commands from a master device, communicates with a master device as another master device or as a slave device, and queues commands received from a master device. Since the achievement of coherency is distributed in time and space, the node controller helps to eliminate certain types of snoop collisions by introducing delays into the snoops of commands selected from its queues in certain circumstances. If the system is lightly loaded, the introduced delay is configured to be the minimum amount necessary to eliminate failed snoops with particular known bus timing conflicts. If the system is more heavily loaded, no delays are experienced in the selection of commands for snoop.

Method and apparatus to eliminate failed snoops of transactions caused by bus timing conflicts in a distributed symmetric multiprocessor system

The present invention provides a method and apparatus for providing memory coherency among an L1 and an L2 cache memory devices and a main memory device. In an embodiment of the invention, a memory controller generates an address snoop for locating modified copies of a data residing in the main memory. As a result of the snoop, the memory controller is notified as to whether a modified copy of the data is in the cache memory devices. If both cache memory devices have a modified copy of the data, the modified copy in the L2 cache will not be allowed to be transferred since modified copies of data in the L1 cache memory are considered to always be the most recent copies of the data. Thus, if the L1 cache memory is unable to transfer the data, The memory controller will continue to snoop the address until the L1 cache memory transfers the data. If, however, there is only one copy of the modified data and it resides in the L2 cache memory device, the copy will be allowed to be transferred to the memory controller.

Apparatus and method of snooping processors and look-aside caches

A logic circuit mechanism for inducing a processing unit to release a LOCK signal that the processing unit uses to secure continuous access to a memory system during read modify write operations requiring 'atomic' (continuous) access. The processing unit has an internal cache enabling it to set up consecutive memory access operations at a pace such that the LOCK signal could be held continuously active while a string of atomic memory accesses is carried out. The present circuit mechanism prevents premature release of the processing unit's LOCK signal, by asserting a Hold signal which requires the processing unit to release its LOCK signal but only after that unit has fully completed its current atomic access operation. The logic circuit reduces its impact on processing unit performance by detecting when the LOCK signal has been active continuously for N consecutive atomic operations coinciding with external contention, and calling for release of the CPUs LOCK signal only while the Nth such operation is being conducted.

Mechanism for efficiently releasing memory lock, after allowing

A dual bus microcomputer system including a cache subsystem improves performance under certain circumstances by allowing programmable control over the LOCK function. More particularly, additional logic is coupled between the LOCK output of the CPU and the LOCK input of the cache controller. A control bit from an I/O port is a second input to the additional logic. With the control bit in one state, the logic allows the LOCK input to follow the LOCK output. In the other state of the control bit, the LOCK input is disabled regardless of the state of the LOCK output.

Dual bus microcomputer system with programmable control of lock function

A multi-bus microcomputer system includes a cache subsystem and an arbitration supervisor. A CPU is provided with a PREEMPT signal source which generates a preempt signal in CPU cycles extending beyond a specified duration. The preempt signal is effective at any device having access to the bus to initiate an orderly termination of the bus usage. When that device signals its termination of bus usage, the arbitration supervisor changes the state of an ARB/GRANT conductor, which had been in the grant phase, to the arbitration phase. During the arbitration phase each of the devices (other than the CPU) cooperates in an arbitration mechanism for bus usage during the next grant phase. On the other hand, the CPU, having asserted preempt, responds to a signal indicating initiation of the arbitration phase by immediately accessing the system bus.

System bus preempt for 80386 when running in an 80386/82385

A network arbitration period following the termination of each transmitted frame is divided into a plurality of access windows which are assigned to respective stations in the network. Any station desiring to gain access to the network can acquire the network by transmitting during its assigned access window if no station assigned an earlier access window has already begun transmitting. The access window assignments can be rotated to equitably distribute access to the network.

Growing community of inventors

Delray Beach, FL, United States of America

Philip E Milling