A driver may wish to iterate every tagged request, not just ones that
satisfy blk_mq_request_started(). The intended use is so a driver may
terminate entered requests on quiesced queues.
How about we just move the started check into the handler passed in for
those that care about it? Much saner to make the interface iterate
everything, and leave whatever state check to the callback.
So we used to do that, and I changed it back in May to test for
MQ_RQ_IN_FLIGHT, and then Ming changed it to check
blk_mq_request_started. So this is clearly a minefield of sorts..
Note that at least mtip32xx, nbd, skd and the various nvme transports
want to use the function to terminate all requests in the error
path, and it would be great to have one single understood, documented
and debugged helper for that in the core, so this is a vote for moving
more of the logic in your second helper into the core code. skd
will need actually use ->complete to release resources for that, though
and mtip plays some odd abort bits. If it weren't for the interesting
abort behavior in nvme-fc that means we could even unexport the
low-level interface.
Yes, I'm very much in favour of this, too.
We always have this IMO slightly weird notion of stopping the queue, set
some error flags in the driver, then _restarting_ the queue, just so
that the driver then sees the error flag and terminates the requests.
Which I always found quite counter-intuitive.
What about requests that come in after the iteration runs? how are those
terminated?
