require RWF_HIPRI for iopoll reads and writes
Subsystems (2)
PATCH 1/1 require RWF_HIPRI for iopoll reads and writes
This makes the mapping RWF_HIPRI <-> IOCB_HIPRI <-> iopoll more consistent; it also allows supporting iopoll operations without IORING_SETUP_IOPOLL in the future.
--- fs/io_uring.c | 4 +++- 1 file changed, 3 insertions(+), 1 deletion(-)
fs/io_uring.c +3 -1
@@ -881,14 +881,16 @@ static int io_prep_rw(struct io_kiocb *req, const struct sqe_submit *s, kiocb->ki_flags |=IOCB_NOWAIT ;if (ctx->flags &IORING_SETUP_IOPOLL ) { +/* require O_DIRECT (or DAX) and RWF_HIPRI */ if (!(kiocb->ki_flags &IOCB_DIRECT ) || + !(kiocb->ki_flags &IOCB_HIPRI ) || !kiocb->ki_filp->f_op->iopoll)return -EOPNOTSUPP ; req->error =0 ; - kiocb->ki_flags |=IOCB_HIPRI ; kiocb->ki_complete = io_complete_rw_iopoll; }else { +/* RWF_HIPRI not allowed */ if (kiocb->ki_flags &IOCB_HIPRI )return -EINVAL ; kiocb->ki_complete = io_complete_rw;
Signature · 1 line
2.20.1
Jens Axboe
maintainerQuoted 3 lines · Stefan Bühler
This makes the mapping
RWF_HIPRI<->IOCB_HIPRI<-> iopoll more consistent; it also allows supporting iopoll operations withoutIORING_SETUP_IOPOLLin the future.
I don't want to make this change now. Additionally, it's never going to be possible to support polled IO mixed with non-polled IO on an io_uring instance, as that makes the wait part of IO impossible to support without adding tracking of requests.
As we can never mix them, it doesn't make a lot of sense to request RWF_HIPRI for polled IO.
Signature · 1 line
Jens Axboe
Hi,
Quoted 5 lines · Jens Axboe
Quoted 3 lines · Stefan Bühler
This makes the mapping
RWF_HIPRI<->IOCB_HIPRI<-> iopoll more consistent; it also allows supporting iopoll operations withoutIORING_SETUP_IOPOLLin the future.
Quoted 7 lines
I don't want to make this change now. Additionally, it's never going to be possible to support polled IO mixed with non-polled IO on an
io_uringinstance, as that makes the wait part of IO impossible to support without adding tracking of requests.As we can never mix them, it doesn't make a lot of sense to request
RWF_HIPRIfor polled IO.
I'm not just new to memory ordering, I'm also new to kernel internals :)
To me it looks like iopoll is basically a busy-loop interface; it helps making things move forward more quickly, while they still might (or might not) finish on their own.
And io_do_iopoll simply loops over all requests and runs a single iteration for them, or, if there is only one request ("!poll_multi_file"), it tells it to spin internally.
While there are multiple requests it can't spin in a single request anyway, and I don't see why it couldn't also check for completion of non-polled requests after looping over the polled requests (whether by only checking the CQ tail or actively tracking (why would that be bad?) the requests some other way). This only means that as long there are non-polled requests pending it mustn't spin in a single request.
And if there are no polled-requests at all it could use io_cqring_wait.
So I don't see why it would be impossible to mix polled and non-polled IO requests.
Any hints what I'm missing here?
(Even if it turns out to be impossible I still think requiring RWF_HIPRI would be the right way, but well...)
cheers,
Stefan
Jens Axboe
maintainerQuoted 13 lines · Stefan Bühler
Hi,
Quoted 10 lines · Jens Axboe
Quoted 3 lines · Stefan Bühler
This makes the mapping
RWF_HIPRI<->IOCB_HIPRI<-> iopoll more consistent; it also allows supporting iopoll operations withoutIORING_SETUP_IOPOLLin the future.I don't want to make this change now. Additionally, it's never going to be possible to support polled IO mixed with non-polled IO on an
io_uringinstance, as that makes the wait part of IO impossible to support without adding tracking of requests.
Quoted 8 lines
Quoted 2 lines
As we can never mix them, it doesn't make a lot of sense to request
RWF_HIPRIfor polled IO.I'm not just new to memory ordering, I'm also new to kernel internals :)
To me it looks like iopoll is basically a busy-loop interface; it helps making things move forward more quickly, while they still might (or might not) finish on their own.
Right, the key there is that they might not. For NVMe and anything else that has been updated, polled IO will not finish on its own. It must be actively found and reaped.
Quoted 3 lines
And
io_do_iopollsimply loops over all requests and runs a single iteration for them, or, if there is only one request ("!poll_multi_file"), it tells it to spin internally.
Correct
Quoted 11 lines
While there are multiple requests it can't spin in a single request anyway, and I don't see why it couldn't also check for completion of non-polled requests after looping over the polled requests (whether by only checking the CQ tail or actively tracking (why would that be bad?) the requests some other way). This only means that as long there are non-polled requests pending it mustn't spin in a single request.
And if there are no polled-requests at all it could use
io_cqring_wait.So I don't see why it would be impossible to mix polled and non-polled IO requests.
It's not technically impossible, but it would be more inefficient to do so. Adding per-request accounting would cost cycles, and the logic of how to handle waiting/polling for a mixed workload would be interesting. Hence it's simpler to simply disallow mixing polled and non polled IO. There are no real benefits to allowing the mix and match of them.
Signature · 1 line
Jens Axboe