Hi Bart, this isn't just about performance - it's about understanability of the I/O path. The legacy request path already has the prep_fn, which is intended for exactly this sort of prep work, but even that led to a lot of confusion. For blk-mq we decided to not add it but let the called driver in control. I really don't want to move away from that. The passthrough requests using blk_get_requst are a special case as the caller allocates the requests, stuffs data into them and only then hands control to the driver, and thus we need some way to initialize the request before handing controller to the driver in this particular special case. And if needed would could actually do that with explicit calls instead of the callback, although you changed it to save a bit of code.
