Non-coherent read in a strongly consistent cache system for frequently read but rarely updated data
A technique relates to enabling a multiprocessor computer system to make a non-coherent request for a cache line. A first processor core sends a non-coherent fetch to a cache. In response to a second processor core having exclusive ownership of the cache line in the cache, the first processor core receives a stale copy of the cache line in the cache based on the non-coherent fetch. The non-coherent fetch is configured to obtain the stale copy for a predefined use. Cache coherency is maintained for the cache, such that the second processor core continues to have exclusive ownership of the cache line while the first processor core receives the stale copy of the cache line.
1. A computer-implemented method for enabling a multiprocessor computer system to make a non-coherent request for a cache line in a cache, the method comprising:
sending by a first processor core a non-coherent fetch to the cache; and
in response to a second processor core having exclusive ownership of the cache line in the cache, receiving by the first processor core a stale copy of the cache line in the cache based on the non-coherent fetch concurrently with maintaining cache coherency, such that the second processor core continues to have the exclusive ownership of the cache line in the cache, the exclusive ownership being different from a shared state, while the first processor core receives the stale copy of the cache line;
wherein the non-coherent fetch is configured to obtain the stale copy for a predefined use;
wherein the predefined use is a single use for the stale copy.
2. The computer-implemented method of claim 1 , wherein the predefined use limits use of the stale copy by the first processor core to comparing a value of the stale copy to a predefined value.
3. The computer-implemented method of claim 1 , wherein the cache is a shared cache system.
4. The computer-implemented method of claim 1 , wherein the predefined use permits the second processor core to store the stale copy in a local cache.
5. The computer-implemented method of claim 4 , wherein a cache line of the local cache having the stale copy is marked invalid, such that any subsequent fetch to the local cache for the stale copy results in a cache miss based on the predefined use.
6. The computer-implemented method of claim 1 , wherein, when the second processor core makes updates to the cache line, the stale copy received by the first processor core is different from the cache line updated by the second processor core.