From: Jens Axboe <axboe@kernel.dk>
To: Benny Halevy <bhalevy@scylladb.com>,
linux-block@vger.kernel.org, linux-fsdevel@vger.kernel.org,
linux-aio@kvack.org
Cc: hch@lst.de, jmoyer@redhat.com
Subject: Re: [PATCH 15/26] aio: support for IO polling
Date: Wed, 5 Dec 2018 06:10:22 -0700 [thread overview]
Message-ID: <a8df7d41-3a3c-370f-33be-1e2954b45362@kernel.dk> (raw)
In-Reply-To: <51ae659719c64d869cae06cdc94748f64132ba3a.camel@scylladb.com>
On 12/5/18 2:56 AM, Benny Halevy wrote:
>> +static int aio_iopoll_getevents(struct kioctx *ctx,
>> + struct io_event __user *event,
>> + unsigned int *nr_events, long min, long max)
>> +{
>> + struct aio_kiocb *iocb;
>> + int to_poll, polled, ret;
>> +
>> + /*
>> + * Check if we already have done events that satisfy what we need
>> + */
>> + if (!list_empty(&ctx->poll_completing)) {
>> + ret = aio_iopoll_reap(ctx, event, nr_events, max);
>> + if (ret < 0)
>> + return ret;
>> + /* if min is zero, still go through a poll check */
>
> A function-level comment about that would be more visible.
Yes, that might be a better idea.
>> + if (min && *nr_events >= min)
>>
>> + return 0;
>
> if ((min && *nr_events >= min) || *nr_events >= max) ?
>
> Otherwise, if poll_completing or poll_submitted are not empty,
> besides getting to the "Shouldn't happen" case in aio_iopoll_reap,
> it looks like we're gonna waste some cycles and never call f_op->iopoll
Agree, that would be a better place to catch it. I'll make those two
changes, thanks!
>> +
>> + /*
>> + * Find up to 'max' worth of events to poll for, including the
>> + * events we already successfully polled
>> + */
>> + polled = to_poll = 0;
>> + list_for_each_entry(iocb, &ctx->poll_completing, ki_list) {
>> + /*
>> + * Poll for needed events with spin == true, anything after
>> + * that we just check if we have more, up to max.
>> + */
>> + bool spin = polled + *nr_events >= min;
>
> I'm confused. Shouldn't spin == true if polled + *nr_events < min?
Heh, that does look off, and it is. I think the correct condition is:
bool spin = !polled || *nr_events < min;
and I'm not sure why that got broken. I just tested it, slight
performance win as expected correcting this. It ties in with the above
nr_events check - it's perfectly valid to pass min_nr == 0 and just do a
poll check. Conversely, if we already spun, just do non-spin checks on
followup poll checks.
>> +static int __aio_iopoll_check(struct kioctx *ctx, struct io_event __user *event,
>> + unsigned int *nr_events, long min_nr, long max_nr)
>> +{
>> + int ret = 0;
>> +
>> + while (!*nr_events || !need_resched()) {
>> + int tmin = 0;
>> +
>> + if (*nr_events < min_nr)
>> + tmin = min_nr - *nr_events;
>
> Otherwise, shouldn't the function just return 0?
>
> Or perhaps you explicitly want to go through the tmin==0 case
> in aio_iopoll_getevents if *nr_events == min_nr (or min_nr==0)?
No, we need to go through poll, if min_nr == 0, but only if min_nr == 0
and !nr_events. But we only get here for nr_events != 0, so should be
fine as-is.
Thanks for your review, very useful! I'll make the above changes right
now.
--
Jens Axboe
next prev parent reply other threads:[~2018-12-05 13:10 UTC|newest]
Thread overview: 54+ messages / expand[flat|nested] mbox.gz Atom feed top
2018-12-04 23:37 [PATCHSET v5] Support for polled aio Jens Axboe
2018-12-04 23:37 ` [PATCH 01/26] fs: add an iopoll method to struct file_operations Jens Axboe
2018-12-06 19:16 ` Jeff Moyer
2018-12-04 23:37 ` [PATCH 02/26] block: add REQ_HIPRI_ASYNC Jens Axboe
2018-12-04 23:37 ` [PATCH 03/26] block: wire up block device iopoll method Jens Axboe
2018-12-06 19:11 ` Jeff Moyer
2018-12-06 19:14 ` Jens Axboe
2018-12-06 19:15 ` Christoph Hellwig
2018-12-06 19:25 ` Jens Axboe
2018-12-04 23:37 ` [PATCH 04/26] block: use REQ_HIPRI_ASYNC for non-sync polled IO Jens Axboe
2018-12-04 23:37 ` [PATCH 05/26] iomap: wire up the iopoll method Jens Axboe
2018-12-04 23:37 ` [PATCH 06/26] aio: use assigned completion handler Jens Axboe
2018-12-04 23:37 ` [PATCH 07/26] aio: separate out ring reservation from req allocation Jens Axboe
2018-12-04 23:37 ` [PATCH 08/26] aio: don't zero entire aio_kiocb aio_get_req() Jens Axboe
2018-12-06 19:27 ` Jeff Moyer
2018-12-06 19:28 ` Jens Axboe
2018-12-06 19:38 ` Jeff Moyer
2018-12-06 19:42 ` Jens Axboe
2018-12-04 23:37 ` [PATCH 09/26] aio: only use blk plugs for > 2 depth submissions Jens Axboe
2018-12-06 19:29 ` Jeff Moyer
2018-12-06 19:31 ` Jens Axboe
2018-12-04 23:37 ` [PATCH 10/26] aio: use iocb_put() instead of open coding it Jens Axboe
2018-12-04 23:37 ` [PATCH 11/26] aio: split out iocb copy from io_submit_one() Jens Axboe
2018-12-04 23:37 ` [PATCH 12/26] aio: abstract out io_event filler helper Jens Axboe
2018-12-04 23:37 ` [PATCH 13/26] aio: add io_setup2() system call Jens Axboe
2018-12-04 23:37 ` [PATCH 14/26] aio: add support for having user mapped iocbs Jens Axboe
2018-12-04 23:37 ` [PATCH 15/26] aio: support for IO polling Jens Axboe
2018-12-05 9:56 ` Benny Halevy
2018-12-05 13:10 ` Jens Axboe [this message]
2018-12-06 9:04 ` Benny Halevy
2018-12-04 23:37 ` [PATCH 16/26] aio: add submission side request cache Jens Axboe
2018-12-04 23:37 ` [PATCH 17/26] fs: add fget_many() and fput_many() Jens Axboe
2018-12-04 23:37 ` [PATCH 18/26] aio: use fget/fput_many() for file references Jens Axboe
2018-12-04 23:37 ` [PATCH 19/26] aio: split iocb init from allocation Jens Axboe
2018-12-04 23:37 ` [PATCH 20/26] aio: batch aio_kiocb allocation Jens Axboe
2018-12-04 23:37 ` [PATCH 21/26] block: add BIO_HOLD_PAGES flag Jens Axboe
2018-12-04 23:37 ` [PATCH 22/26] block: implement bio helper to add iter bvec pages to bio Jens Axboe
2018-12-04 23:37 ` [PATCH 23/26] fs: add support for mapping an ITER_BVEC for O_DIRECT Jens Axboe
2018-12-04 23:37 ` [PATCH 24/26] aio: add support for pre-mapped user IO buffers Jens Axboe
2018-12-04 23:37 ` [PATCH 25/26] aio: split old ring complete out from aio_complete() Jens Axboe
2018-12-04 23:37 ` [PATCH 26/26] aio: add support for submission/completion rings Jens Axboe
2018-12-07 18:48 ` [PATCHSET v5] Support for polled aio Jeff Moyer
2018-12-07 18:52 ` Jens Axboe
2018-12-07 18:55 ` Jens Axboe
2018-12-07 19:34 ` Jeff Moyer
2018-12-07 19:35 ` Jens Axboe
2018-12-07 21:58 ` Jens Axboe
2018-12-07 21:59 ` Jens Axboe
2018-12-07 22:00 ` Jens Axboe
2018-12-07 22:06 ` Jens Axboe
2018-12-07 22:19 [PATCHSET v6] Support for polled aio (and more) Jens Axboe
2018-12-07 22:20 ` [PATCH 15/26] aio: support for IO polling Jens Axboe
2018-12-09 10:02 ` Benny Halevy
2018-12-09 14:36 ` Jens Axboe
2018-12-10 8:10 ` Benny Halevy
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=a8df7d41-3a3c-370f-33be-1e2954b45362@kernel.dk \
--to=axboe@kernel.dk \
--cc=bhalevy@scylladb.com \
--cc=hch@lst.de \
--cc=jmoyer@redhat.com \
--cc=linux-aio@kvack.org \
--cc=linux-block@vger.kernel.org \
--cc=linux-fsdevel@vger.kernel.org \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).