Not so much "talking" as directly reading/writing memory outside of your source scope that is also mutated by other chips--like a DMA buffer.
Or for a traditional kernel, reading/writing directly into the address space of a user process. Wouldn't have to worry so much about race conditions there since the process could be suspended, but there is still an opportunity to fudge the indexing.
Or for a traditional kernel, reading/writing directly into the address space of a user process. Wouldn't have to worry so much about race conditions there since the process could be suspended, but there is still an opportunity to fudge the indexing.