摘要:
A mechanism is provided in a logically partitioned data processing system for controlling depth and latency of exit of a virtual processor's idle state. A virtualization layer generates a cede latency setting information (CLSI) data. Responsive to booting a logical partition, the virtualization layer communicates the CLSI data to an operating system (OS) of the logical partition. The OS determines, based on the CLSI data, a particular idle state of a virtual processor under a control of the OS. Responsive to the OS calling the virtualization layer, the OS communicates the particular idle state of the virtual processor to the virtualization layer for assigning the particular idle state and wake-up characteristics to the virtual processor.
摘要:
A mechanism is provided for temporarily allocating dedicated processors to a shared processor pool. A virtual machine monitor determines whether a temporary allocation associated with an identified dedicated processor is long-term or short-term. Responsive to the temporary allocation being long-term, the virtual machine monitor determines whether an operating frequency of the identified dedicated processor is within a predetermined threshold of an operating frequency of one or more operating systems utilizing the shared processor pool. Responsive to the operating frequency of the identified dedicated processor failing to be within the predetermined threshold, the virtual machine monitor either increases or decreases the frequency of the identified dedicated processor to be within the predetermined threshold of the operating frequency of the one or more operating systems utilizing the shared processor pool and temporarily allocates the identified dedicated processor to the shared processor pool.
摘要:
A method, system and computer-usable medium are disclosed for managing power consumption in information processing systems. Processing resources are successively folded, allowing them to be placed into deeper and deeper power saving states while maintaining the ability to respond to new processing loads without exposing the latency of the deeper power saving states as they are unfolded. Before a deeper power saving state can be used, there must be sufficient processing resources in the prior power saving state to mask the latency of bringing a processing resource out of the deeper power saving state.
摘要:
An information handling system includes instruction processing nodes in respective physical partitions. A communications bus couples two information processing nodes together. Each node includes hardware resources such as CPUs, memories and I/O adapters. Prior to a command to merge the physical partitions, the communication bus exhibits a disabled state such that the two information processing nodes are effectively disconnected. After receiving a command to merge the physical partitions, the system enables the communication bus to effectively hot-plug the two nodes together. A modified master hypervisor in one node stores data structures detailing the hardware resources of the two nodes. The modified master may assign resources from one node to a logical partition in another node.
摘要:
Controlled partition shut-down is provided within a shared memory partition data processing system including a shared memory partition, a paging service partition, a hypervisor and a shared memory pool within physical memory. The hypervisor manages access to logical pages within the pool and page-out of pages from the pool to external paging storage via the paging service partition. A respective paging service stream exists between the paging service partition and hypervisor for each shared memory partition, with each stream including a stream state. The control method includes: responsive to a shut-down initiating event, notifying the paging service partition to shut down, and determining whether a shared memory partition is currently active, and if so, signaling the hypervisor to complete paging activity for the active memory partition and waiting for its stream state to enter a suspended or a completed state before automatically shutting down the paging service partition.
摘要:
Transparent hypervisor pinning of critical memory areas is provided for a shared memory partition data processing system. The transparent hypervisor pinning includes receiving at a hypervisor a hypervisor call initiated by a logical partition to register a logical memory area of the logical partition with the hypervisor. Responsive to this hypervisor call, the hypervisor transparently determines whether the logical memory is a critical memory area for access by the hypervisor. If the logical memory area is a critical memory area, then the hypervisor automatically pins the logical memory area to physical memory of the shared memory partition data processing system, thereby ensuring that the memory area will not be paged-out from physical memory to external storage, and thus ensuring availability of the logic memory area to the hypervisor.
摘要:
A method and apparatus for transparently handling recurring correctable errors to prevent costly system shutdowns for correctable memory errors or system failures from uncorrectable memory errors. When a high number of correctable errors are detected for a given memory location, the hypervisor moves the data associated with the memory location to an alternate physical memory location transparently to the partition such that the partition has no knowledge that the physical memory actualizing the memory location has been changed. Similarly, the hypervisor can move direct memory access (DMA) memory locations using an I/O translation table.
摘要:
A multiprocessor system which includes automatic workload distribution. As threads execute in the multiprocessor system, an operating system or hypervisor continuously learns the execution characteristics of the threads and saves the information in thread-specific control blocks. The execution characteristics are used to generate thread performance data. As the thread executes, the operating system continuously uses the performance data to steer the thread to a core that will execute the workload most efficiently.
摘要:
A snoop coherency method, system and program are provided for intervening a requested cache line from a plurality of candidate memory sources in a multiprocessor system on the basis of the sensed temperature or power dissipation value at each memory source. By providing temperature or power dissipation sensors in each of the candidate memory sources (e.g., at cores, cache memories, memory controller, etc.) that share a requested cache line, control logic may be used to determine which memory source should source the cache line by using the power sensor signals to signal only the memory source with acceptable power dissipation to provide the cache line to the requester.
摘要:
An apparatus and method selectively invalidate entries in an address translation cache instead of invalidating all, or nearly all, entries. One or more translation mode bits are provided in each entry in the address translation cache. These translation mode bits may be set according to the addressing mode used to create the cache entry. One or more “hint bits” are defined in an instruction that allow specifying which of the entries in the address translation cache are selectively preserved during an invalidation operation according to the value(s) of the translation mode bit(s). In the alternative, multiple instructions may be defined to preserve entries in the address translation cache that have specified addressing modes. In this manner, more intelligence is used to recognize that some entries in the address translation cache may be valid after a task or partition switch, and may therefore be retained, while other entries in the address translation cache are invalidated.