Have been trying to understand the image below which shows status of spinlock using cache coherence. Reference is Computer Architecture A Quantitative Approach------------------
Could anybody explain why there is a cache miss in step no.5 when P2 tries to execute swap.
$\begingroup$
$\endgroup$
$\begingroup$
$\endgroup$
In step 4 P2 tests the lock and sees the value is 0. But testing the lock is just a read operation, so P2 has the cache line in state shared. In step 5 P2 executes a swap operation, which is both a read and a write, so requires a coherence action (invalidate P1's shared copy) to acquire the cache line in state exclusive. It's a "cache miss" in the sense that the required state in the cache is incorrect, not in the sense that the data is missing.
answered Nov 24 '16 at 4:57
$\begingroup$
$\endgroup$
- In step 2, there is a write invalidate (from P0) of the lock variable.
- In step 3, P1 experiences a cache miss.
- In step 4, P1 is waiting for the bus to become available.
- Finally, in step 5, the bus becomes available and P1's request is satisfied.
answered Nov 23 '16 at 22:51
-
$\begingroup$ I understand, but what i am enquiring about here is-----why is P2 suffering a cache miss when it has the lock in it's cache in shared mode in step 5. $\endgroup$ – Abhishek Dhankar Nov 24 '16 at 4:02
