Few questions regarding RIO

The documentation for RIOSend states that an in-flight buffer cannot be referenced by more than one tx operation concurrently; is this really true? If so, could it be explained as to why? I have a lot of software which relies on the absence of such a restriction as is the case with the usual I/O API.

Additionally, is there any plan to support SGIO with RIO? It appears such is intended given the presence of the familiar count when specifying the number of supplied buffers, however this is restricted to 1 at present.

Note that, I understand RIO is beyond the scope of ntdev in general given it’s just a userland API, however I’ve exhausted all other options for information regarding this still rather esoteric API and I’m aware there’s a few MS/NDIS team members around here who, if anyone, would know.

Thanks,

Matt.

RIO is just about the same as regular IO except that the UM application designates memory regions that get pinned into RAM and so cannot be paged out an certain locks can be elided when making read / write calls.

Obviously you would not pass the exact same buffer to two concurrent ReadFile calls. The same is true with RIO.

WriteFile is more interesting as there is no API level guarantee that the buffer will be const, but many applications would break if the kernel modified data buffers as they are written