International Business Machines Corporation

Dennis Charles Wilkerson

Richard Gerard Hofmann

Peter Dean LaFauci

Mark Michael Schaffer

Edward Hammond Green, Iii

Jason Michael Hopp

Anthony Correale, Jr

Jeffrey Todd Bridges

Rhonda Gurganious Mitchell

David Otero

A method and system for forwarding interrupt requests from a source device to a destination device. A controller bridge receives data, from a source device, for a destination device and stores the incoming data in a data queue. An interrupt request is received from the source device for the destination device and forwarded to the destination device in response to completing a transfer of the data from the source device to the destination device. If data received from the source device for the destination device are pending in the data queue, the interrupt request is rejected and the source may resubmit the interrupt request at a later time. If additional data are received from the source device for the destination device, the data may be rejected in response to an interrupt pending in the interrupt queue from the source device for the destination device.

System and method for delaying an interrupt request until corresponding data is provided to a destination device

A multi-master computer system having overlapped read and write signal with scalable address pipelining programmable increases the depth of address pipelining independently on two overlapped read and write data busses up to &ldquo;N&rdquo; deep requests. The system includes a local bus having an address bus, a read bus, and a write bus. Master devices are coupled to separate address, read data and write data buses. Slave devices are attached to the data busses through shared, but decoupled address, read and write data buses. An arbiter is coupled to the data bus and allows masters to compete for bus ownership. The arbiter includes read and write pipeline logic for processing and priortizing master and slave read and write data transfers across the data bus. Programming apparatus alters the read and write pipeline logic for address pipelining

Multi-master computer system with overlapped read and write operations and scalable address pipelining

Disclosed is a method and apparatus for optimizing communication between buses operating at different frequencies. A high speed bus provides communication between high speed devices as well as provides communication to a lower speed bus which provides communications between low speed devices and between low speed devices and high speed devices. During transaction reset a bridge device that controls communication between the high and low speed buses determines the ratio of the respective bus clocks. The bridge device sets a logic state machine based on this data and selects from the higher frequency clock selected cycles on which transfers to the lower speed bus take place. The unused cycles of the high speed clock may then be used on the high speed bus for additional transfers. The clock ratios may be determined on each transaction reset and new data on clock ratios stored in the bridge device for subsequent operation.

Method and system for optimizing data transfers between devices interconnected by buses operating at different clocking speeds

A high performance integrated circuit (IC) with independent read and write data busses enables full simultaneous read and write data transfers between devices coupled to the buses. Multiple master and multiple slave devices communicate using the resources of a bus controller and a bus arbiter. Having separate read and write data busses with separate and independent arbitration allows reads and writes from different devices to occur simultaneously. Many high performance IC, like systems on a chip (SOC), have many different functional units communicating with a central processing unit (CPU). Many such CPUs have architectures that may cause in certain applications an unbalance between read and write traffic on the independent busses. Master and slave devices contain auxiliary internal read and write data buses multiplexed such that read or write data may be interchanged. A corresponding Auxiliary_(read or write) command is routed to the slave units to notify the units when to route normal read or write data to an idle bus. The bus controller may use this added feature to optimize the available bandwidth of independent read and write data busses up to the limit where a read or write bandwidth may be two times that available if the read and write data buses were used only for their normal traffic.

Dynamic data bus allocation

The present invention discloses a method and system for managing independent read and write buses by dividing the pending read and write request signals and the read and write request priority level signals. The arbitration for use of the read and write buses are done independently for the read and write operations. A higher priority read, for example, can be concurrent with a corresponding lower priority write. Interruption of in process reads or writes is also done using the split arbitrations of the read and write buses leading the disruption of lower priority operations only if the conflicts are concurrent for the same read or write operation.

Dual burst latency timers for overlapped read and write data transfers

Methods, arbiters, and computer program products determine if a request for an idle bus in a dual bus data processing system is being blocked by one or more pending requests for the other bus. In this circumstance, any such pending request for the other bus is masked by the arbiter so that the request for the idle bus can be granted. Consequently, a more efficient utilization of the dual bus architecture is achieved. In an illustrative embodiment, a bus request is received for a first one of the dual busses. If the address and control busses are unavailable to allow the request to be granted, then an inquiry is made regarding the status of a pending request for the second one of the dual busses that has gained control of the address and control busses. In particular, it is determined whether a primary request has been granted and a secondary request has been pipelined for the second one of the dual busses. If a primary request has been granted and a secondary request has been pipelined, then the priority of the pending requests for the second one of the dual busses are examined. If the priority of the pending requests for the second one of the dual busses are at least as high as the currently pending request for the first one of the dual busses, then these requests are masked so that they no longer appears to be pending, which allows the request for the first one of the dual busses to be granted.

Methods, arbiters, and computer program products that can improve the performance of a pipelined dual bus data processing system

A method and implementing system are provided which includes a DMA controller coupled to a slave bus controller through a processor local bus. The slave bus controller is also coupled to a memory unit. The memory unit is connected directly to a peripheral device. The DMA controller is arranged to receive a data transfer request from the peripheral unit and initiate a transfer cycle with the slave bus controller. The slave bus controller is selectively operable to assert a transfer signal to the memory unit which enables data movement directly between memory and the peripheral device in accordance with the request from the peripheral device. Upon completion of the address transfer and prior to the completion of the data transfer, the slave bus controller generates a transfer complete signal back to the peripheral device. This technique allows for a DMA FlyBy transfer to be overlapped with a subsequent processor local bus transfer.

Processor local bus posted DMA FlyBy burst transfers

A DMA Controller, in response to a data transfer request from a slave device, initiates a memory transfer cycle and informs the slave device when the data transfer has completed. In order to avoid dead clock cycles on internal bus(es), the DMA Controller initiates a speculative data transfer cycle after the notification. The DMA Controller aborts the speculative data transfer cycle if the slave device does not request another data transfer within a predetermined time.

Speculative direct memory access transfer between slave devices and

A method and apparatus is provided which allows overlapping of DMA line read and line write cycles. In an exemplary embodiment, the PLB Line Read Word Address bus is used with a DMA controller sideband signal to detect the conditions required to allow the DMA controller to start the line write one cycle prior to the completion of the line read cycle. A reference bit is set when the first word of a multi-word line transfer has been read. A sideband timing signal is generated one cycle prior to the completion of the read cycle indicating that there is only one read data phase remaining of the line read. If the first word to be written out to memory has been read or is available when the timing signal is generated, the write operation is begun prior to the final phase of the memory read transfer, and the read and write operations are overlapped thereby accomplishing an overlapped read/write transfer in fewer cycles than the sum of read and write transfer cycles if done sequentially.

Growing community of inventors

Durham, NC, United States of America

Dennis Charles Wilkerson