From: Kevin Wolf <kwolf@redhat.com>
To: Hanna Czenczek <hreitz@redhat.com>
Cc: qemu-block@nongnu.org, qemu-devel@nongnu.org,
"Stefan Hajnoczi" <stefanha@redhat.com>,
"Paolo Bonzini" <pbonzini@redhat.com>,
"Richard W . M . Jones" <rjones@redhat.com>,
"Ilya Dryomov" <idryomov@gmail.com>,
"Peter Lieven" <pl@dlhnet.de>,
"Philippe Mathieu-Daudé" <philmd@linaro.org>,
"Alex Bennée" <alex.bennee@linaro.org>,
"Fam Zheng" <fam@euphon.net>,
"Ronnie Sahlberg" <ronniesahlberg@gmail.com>
Subject: Re: [PATCH 08/16] nvme: Fix coroutine waking
Date: Wed, 29 Oct 2025 18:43:34 +0100 [thread overview]
Message-ID: <aQJSRicrD7NET-az@redhat.com> (raw)
In-Reply-To: <20251028163343.116249-9-hreitz@redhat.com>
Am 28.10.2025 um 17:33 hat Hanna Czenczek geschrieben:
> nvme wakes the request coroutine via qemu_coroutine_enter() from a BH
> scheduled in the BDS AioContext. This may not be the same context as
> the one in which the request originally ran, which would be wrong:
> - It could mean we enter the coroutine before it yields,
> - We would move the coroutine in to a different context.
>
> (Can be reproduced with multiqueue by adding a usleep(100000) before the
> `while (data.ret == -EINPROGRESS)` loop.)
>
> To fix that, schedule nvme_rw_cb_bh() in the coroutine AioContext.
> (Just like in the preceding iscsi and nfs patches, we could drop the
> trivial nvme_rw_cb_bh() and just use aio_co_wake() directly, but don’t
> do that to keep replay_bh_schedule_oneshot_event().)
>
> With this, we can remove NVMeCoData.ctx.
>
> Note the check of data->co == NULL to bypass the BH/yield combination in
> case nvme_rw_cb() is called from nvme_submit_command(): We probably want
> to keep this fast path for performance reasons, but we have to be quite
> careful about it:
> - We cannot overload .ret for this, but have to use a dedicated
> .skip_yield field. Otherwise, if nvme_rw_cb() runs in a different
> thread than the coroutine, it may see .ret set and skip the yield,
> while nvme_rw_cb() will still schedule a BH for waking. Therefore,
> the signal to skip the yield can only be set in nvme_rw_cb() if waking
> too is skipped, which is independent from communicating the return
> value.
> - We can only skip the yield if nvme_rw_cb() actually runs in the
> request coroutine. Otherwise (specifically if they run in different
> AioContexts), the order between this function’s execution and the
> coroutine yielding (or not yielding) is not reliable.
> - There is no point to yielding in a loop; there are no spurious wakes,
> so once we yield, we will only be re-entered once the command is done.
> Replace `while` by `if`.
>
> Cc: qemu-stable@nongnu.org
> Signed-off-by: Hanna Czenczek <hreitz@redhat.com>
As suggested in an earlier patch, I prefer to keep setting -EINPROGRESS,
even if we don't check for it elsewhere, for better debugability.
Kevin
next prev parent reply other threads:[~2025-10-29 17:44 UTC|newest]
Thread overview: 36+ messages / expand[flat|nested] mbox.gz Atom feed top
2025-10-28 16:33 [PATCH 00/16] block: Some multi-threading fixes Hanna Czenczek
2025-10-28 16:33 ` [PATCH 01/16] block: Note on aio_co_wake use if not yet yielding Hanna Czenczek
2025-10-28 16:33 ` [PATCH 02/16] rbd: Run co BH CB in the coroutine’s AioContext Hanna Czenczek
2025-10-28 16:33 ` [PATCH 03/16] iscsi: " Hanna Czenczek
2025-10-29 14:27 ` Kevin Wolf
2025-10-31 9:07 ` Hanna Czenczek
2025-10-31 13:27 ` Kevin Wolf
2025-10-28 16:33 ` [PATCH 04/16] nfs: " Hanna Czenczek
2025-10-28 16:33 ` [PATCH 05/16] curl: Fix coroutine waking Hanna Czenczek
2025-10-29 16:57 ` Kevin Wolf
2025-10-31 9:15 ` Hanna Czenczek
2025-10-31 13:17 ` Kevin Wolf
2025-10-28 16:33 ` [PATCH 06/16] gluster: Do not move coroutine into BDS context Hanna Czenczek
2025-10-29 17:10 ` Kevin Wolf
2025-10-31 9:16 ` Hanna Czenczek
2025-10-28 16:33 ` [PATCH 07/16] nvme: Kick and check completions in " Hanna Czenczek
2025-10-29 17:23 ` Kevin Wolf
2025-10-29 17:39 ` Kevin Wolf
2025-10-31 9:18 ` Hanna Czenczek
2025-10-31 9:19 ` Hanna Czenczek
2025-10-28 16:33 ` [PATCH 08/16] nvme: Fix coroutine waking Hanna Czenczek
2025-10-29 17:43 ` Kevin Wolf [this message]
2025-10-28 16:33 ` [PATCH 09/16] block/io: Take reqs_lock for tracked_requests Hanna Czenczek
2025-10-28 16:33 ` [PATCH 10/16] qcow2: Fix cache_clean_timer Hanna Czenczek
2025-10-29 20:23 ` Kevin Wolf
2025-10-31 9:29 ` Hanna Czenczek
2025-10-31 13:03 ` Kevin Wolf
2025-11-06 16:08 ` Hanna Czenczek
2025-10-28 16:33 ` [PATCH 11/16] ssh: Run restart_coroutine in current AioContext Hanna Czenczek
2025-10-28 16:33 ` [PATCH 12/16] blkreplay: Run BH in coroutine’s AioContext Hanna Czenczek
2025-10-28 16:33 ` [PATCH 13/16] block: Note in which AioContext AIO CBs are called Hanna Czenczek
2025-10-28 16:33 ` [PATCH 14/16] iscsi: Create AIO BH in original AioContext Hanna Czenczek
2025-10-28 16:33 ` [PATCH 15/16] null-aio: Run CB " Hanna Czenczek
2025-10-28 16:33 ` [PATCH 16/16] win32-aio: Run CB in original context Hanna Czenczek
2025-10-30 14:12 ` Kevin Wolf
2025-10-31 9:31 ` Hanna Czenczek
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=aQJSRicrD7NET-az@redhat.com \
--to=kwolf@redhat.com \
--cc=alex.bennee@linaro.org \
--cc=fam@euphon.net \
--cc=hreitz@redhat.com \
--cc=idryomov@gmail.com \
--cc=pbonzini@redhat.com \
--cc=philmd@linaro.org \
--cc=pl@dlhnet.de \
--cc=qemu-block@nongnu.org \
--cc=qemu-devel@nongnu.org \
--cc=rjones@redhat.com \
--cc=ronniesahlberg@gmail.com \
--cc=stefanha@redhat.com \
/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).