Techniques for cache injection in a processor system from a remote node
A technique for performing cache injection in a processor system includes monitoring, by a cache, addresses on a bus. Input/output data associated with an address of a data block stored in the cache is then requested from a remote node, via a network controller. Ownership of the input/output data is acquired by the cache when an address on the bus that is associated with the input/output data corresponds to the address of the data block stored in the cache.
1. A method of performing cache injection in a processor system, comprising:
detecting, by a cache, a specific instruction sequence that includes multiple instructions, wherein one of the multiple instructions is a load instruction to a first address and another one of the multiple instructions is a compare instruction;
monitoring, by the cache, addresses on a bus in response to detecting the specific instruction sequence multiple times;
requesting, from a remote node via a network controller, input/output data associated with the first address of a data block stored in the cache; and
acquiring, by the cache, ownership of the input/output data when an address on the bus that is associated with the input/output data corresponds to the first address of the data block stored in the cache.
2. The method of claim 1 , wherein the acquiring further comprises:
storing the input/output data associated with the address on the bus in the cache at the first address; and
acknowledging, by a cache controller associated with the cache that stores the input/output data, receipt of the input/output data.
3. The method of claim 2 , further comprising:
cancelling, by a memory controller, a direct memory access write to main memory that is associated with the input/output data based on the acknowledging.
4. The method of claim 1 , further comprising:
placing an active thread that is a consumer of the input/output data to sleep until the input/output data is received by the network controller.
5. The method of claim 1 , further comprising:
placing an active thread that is a consumer of the input/output data in a low-priority state until the input/output data is received by the network controller.
6. The method of claim 1 , wherein the input/output data is associated with a direct memory access write to main memory.
7. A processor system, comprising:
a cache, wherein the cache is configured to:
detect a specific instruction sequence that includes multiple instructions, wherein one of the multiple instructions is a load instruction to a first address and another one of the multiple instructions is a compare instruction;
monitor addresses on a bus in response to detecting the specific instruction sequence-multiple times; and
acquire ownership of input/output data when an address on the bus that is associated with the input/output data corresponds to the first address of a data block stored in the cache; and
a network controller coupled to the cache and configured to request, from a remote node, the input/output data associated with the first address of the data block stored in the cache responsive to a request from an associated processor thread.
8. The processor system of claim 7 , wherein the cache is further configured to:
store the input/output data associated with the address on the bus in the cache at the first address; and
acknowledge receipt of the input/output data from the network controller.
9. The processor system of claim 8 , further comprising:
a memory controller configured to cancel a direct memory access write to main memory that is associated with the input/output data based on the cache acknowledging receipt of the input/output data.
10. The processor system of claim 7 , further comprising:
an operating system in communication with the processor thread, wherein the operating system is configured to place the processor thread which is a consumer of the input/output data to sleep until the input/output data is received by the network controller.
11. The processor system of claim 7 , further comprising:
an operating system in communication with the processor thread, wherein the operating system is configured to place the processor thread which is a consumer of the input/output data in a low-priority state until the input/output data is received by the network controller.
12. The processor system of claim 7 , wherein the input/output data is associated with a direct memory access write to main memory.
13. A method of performing cache injection in a processor system, comprising:
detecting, by a cache, a specific instruction sequence that includes multiple instructions, wherein one of the multiple instructions is a load instruction to a first address and another one of the multiple instructions is a compare instruction;
monitoring, by the cache, addresses on a bus in response to detecting the specific instruction sequence multiple times;
requesting, from a remote node via a network controller, input/output data associated with the first address of a data block stored in the cache;
placing an active thread that is a consumer of the input/output data to sleep or in a low-priority state until the input/output data is received by the network controller; and
acquiring, by the cache, ownership of the input/output data when an address on the bus that is associated with the input/output data corresponds to the first address of the data block stored in the cache.
14. The method of claim 13 , wherein the acquiring further comprises:
storing the input/output data associated with the address on the bus in the cache at the first address; and
acknowledging, by a cache controller associated with the cache that stores the input/output data, receipt of the input/output data.