On 10/12/21 6:17 PM, Joanne Koong wrote:
If the bitset data could reside in a global array in the bpf program,
then I
agree - it seems like we could just make a helper function that takes in
an ARG_PTR_TO_MEM where we pass the data in as a ptr, instead of needing
a map.
The main advantage of helpers for bit ops is ease of implementation.
We can add set/test/clear_bit along with ffs fls and others
very quickly.
But programs that need to do atomic bit ops probably care
about performance, so non-inlined helper might be too slow.
Doing the same via new instructions will take a ton more time,
but the long term benefits of insns are worth considering.
