If we use the request table to resolve the issue, this actually introduces another coherence problem: how to make sure that at any time, each cache has the same/most recent copy of the request table.
However, this may be an easy coherence problem (maybe can be managed by only allow atomic update of the table).
Azerty
Is there special hardware for handling updates of the request tables? Or it is somehow done through the bus shared with the different caches, adding overheads on it?
grose
@meatie, well, the request table is just updated by bus transactions, and the cache will see each one of those in order, so it should just need to pay attention, right?
meatie
@grose what do you mean by "pay attention"?
cube
@meatie The cache should just be able to read the data that's being passed on the bus to always guarantee that the request table is up to date, because the cache is always connected and watching the bus.
If we use the request table to resolve the issue, this actually introduces another coherence problem: how to make sure that at any time, each cache has the same/most recent copy of the request table.
However, this may be an easy coherence problem (maybe can be managed by only allow atomic update of the table).
Is there special hardware for handling updates of the request tables? Or it is somehow done through the bus shared with the different caches, adding overheads on it?
@meatie, well, the request table is just updated by bus transactions, and the cache will see each one of those in order, so it should just need to pay attention, right?
@grose what do you mean by "pay attention"?
@meatie The cache should just be able to read the data that's being passed on the bus to always guarantee that the request table is up to date, because the cache is always connected and watching the bus.