What is the specific flow of prfm operation in ARMv8?

Question

In ARM, prfm supports prefetching data to cache.

If I prefetch data to L2 cache using PLDL2KEEP first, wait until the data reaches L2 cache. Then PLDL1KEEP is used to fetch the same data, does prfm get data from memory or L2 cache?

That is, is data prefetch grading feasible?

Answer 1

Prefetch is entirely timing dependent. You have no guaranteed way to "wait until the data reaches L2 cache". You have no guaranteed way to ensure data stays in the cache until you need it once it gets there.

So ...

does prfm get data from memory or L2 cache?

... it depends. If it's in cache you'll get data from cache. If it's not then you'll get data from memory. Or not at all, given that prefetch is just a hint.