P1 has the lock the entire orange time but only has the cache line at the beginning and the end of the orange section

Immediately after acquiring the lock, the cacheline P1's cacheline is invalidated because other processors are fighting for the cacheline and keep on acquiring exclusively and then having theirs invalidated. Once P1 is done, it reacquires and releases the lock.

The use of a ticket lock would decrease the coherency traffic of test-and-set.