摘要:
A mechanism is provided for sharing resources among logical partitions in a logical partitioned data processing system and for managing the changes to resources in such a way that the sharing operating systems are able to handle the various transitions in a graceful manner. Four hypervisor functions plus a specific return code manage the granting of access of resources owned by one partition to another (client) partition, accepting of granted resources by client partitions, returning of granted resources by client partitions, and rescinding of access by the owning partition. These four hypervisor functions are invoked either explicitly by the owning and client partitions or automatically by the hypervisor in response to partition termination. The hypervisor functions provide the needed infrastructure to manage the sharing of logical resources among partitions.
摘要:
A computer implemented method, apparatus and mechanism for recovery of an I/O fabric that has become terminally congested or deadlocked due to a failure which causes buffers/queues to fill and thereby causes the root complexes to lose access to their I/O subsystems. Upon detection of a terminally congested or deadlocked transmit queue, access to such queue by other root complexes is suspended while each item in the queue is examined and processed accordingly. Store requests and DMA read reply packets in the queue are discarded, and load requests in the queue are processed by returning a special completion package. Access to the queue by the root complexes is then resumed.
摘要:
A computer-implemented method, apparatus, and computer program product are disclosed for managing direct memory access (DMA) write page faults using a pool of substitute pages. A computer system platform resolves a DMA write page fault for a page that is dedicated to an Input/Output (I/O) adapter. The I/O adapter attempts to write DMA data to the page. A determination is made that the page is unavailable for writing. The DMA data is then written to data locations in a substitute page that was selected from the pool of substitute pages. A flag is then set in a flag location for each one of the data locations. The flag locations correspond to the data locations. When a flag is set, the flag indicates that DMA write data is present in the data location that corresponds to that flag's flag location.
摘要:
Method, apparatus and system for isolating input/output adapter error domains in a data processing system. Errors occurring in one input/output adapter are isolated from other input/output adapters of the data processing system by functionality in a host bridge that connects the input/output adapters to a system bus of the data processing system, thus permitting the use of low cost, industry standard switches and bridges external to the host bridge.
摘要:
A computer implemented method, apparatus and mechanism for recovery of an I/O fabric that has become terminally congested or deadlocked due to a failure which causes buffers/queues to fill and thereby causes the root complexes to lose access to their I/O subsystems. Upon detection of a terminally congested or deadlocked transmit queue, access to such queue by other root complexes is suspended while each item in the queue is examined and processed accordingly. Store requests and DMA read reply packets in the queue are discarded, and load requests in the queue are processed by returning a special completion package. Access to the queue by the root complexes is then resumed.
摘要:
A method and implementing computer system is provided in which specific device identification information is acquired when a faulty condition is detected during an information transfer transaction, and the condition is reported to the device driver of the identified device for corrective action without initiating a system shut-down. In one example, PCI adapter sequence information, including tag number, requester bus number, requester device number and requester function number is captured and used in reporting an error condition in order to identify and isolate the adapter in a recovery operation.
摘要:
Method, apparatus and system for controlling input/output adapter data flow operations in a data processing system that includes at least one of a traffic class mechanism in conjunction with virtual channel resources so as to be able to associate Load/Store and DMA flows to/from an input/output adapter, and a relaxed ordering mechanism for associating a relaxed ordering bit to Load/Store operations to an input/output adapter. Functionality for controlling the input/output adapter data flow is provided in a host bridge that connects the input/output adapter to a system bus of the data processing system.
摘要:
A method, system, and apparatus for preventing input/output (I/O) adapters used by an operating system (OS) image, in a logically partitioned data processing system, from fetching or corrupting data from a memory location allocated to another OS image within the data processing system is provided. A hypervisor prevents transmission of data between an input/output adapter in one of the logical partitions and memory locations assigned to other logical partitions during a direct memory access (DMA) operation by assigning each of the input/output adapters a range of I/O bus DMA addresses. The I/O adapters (IOAs) are connected to PCI host bridges via terminal bridges. A single terminal bridge may support multiple IOAs, in which case every terminal bridge has a plurality of sets of range registers, each associated with a respective one of the IOAs to which it is connected. An arbiter is provided which selects one of the input/output adapters to use the PCI bus. The terminal bridge can examine the grant signals from the arbiter to the IOAs, to determine which set of range registers is to be used.
摘要:
A method, system, and apparatus for isolating an input/output (I/O) bus error, received from an I/O adapter, from the other I/O adapters that may be in different partitions within a logically partitioned data process system is provided. In one embodiment, the logically partitioned data processing system includes a system bus, a processing unit, a memory unit, a host bridge, a plurality of terminal bridges, and a plurality of input/output adapters. The processing unit, memory unit, and the host bridge are all coupled to each other through the system bus. Each of the plurality of terminal bridges is coupled to the host bridge through a first bus. Each of the input/output adapters is coupled to one of the plurality of terminal bridges through a one of a plurality of second buses, such that each input/output adapter corresponds to a single terminal bridge. Each of the input/output adapters are assigned to one of a plurality of logical partitions within the data processing system. Each of the terminal bridges isolates errors received from a respective one of the input/output adapters from other input/output adapters, some of which may be within a different one of the plurality of logical partitions.
摘要:
An interrupt subsystem within a data processing system is scalable from low-end uni-processor systems to high-end multi-processor (MP) systems. This interrupt subsystem provides for queueing of interrupts from many sources, and for queueing of interrupts to the best processor in a multi-processor system. The external interrupt mechanism is separated into two layers, an interrupt routing layer and an interrupt presentation layer. The interrupt routing layer routes the interrupt conditions to the appropriate instance of an interrupt management area within the interrupt presentation layer. The interrupt presentation layer communicates the interrupt source to the system software which is to service/process the interrupt. By providing two layers within the interrupt subsystem, application or system software can be written which is independent from the types or sources of interrupts. The interrupt routing layer hides the details of a particular hardware implementation from the software. The interrupt presentation layer interfaces to the system and/or application software, and provides hardware independent functionality.