From: David Wei <dw@davidwei.uk>
To: io-uring@vger.kernel.org, netdev@vger.kernel.org
Cc: Jens Axboe <axboe@kernel.dk>, Pavel Begunkov <asml.silence@gmail.com>
Subject: [PATCH v2 5/5] io_uring/zcrx: free proxy ifqs
Date: Sat, 25 Oct 2025 12:15:04 -0700 [thread overview]
Message-ID: <20251025191504.3024224-6-dw@davidwei.uk> (raw)
In-Reply-To: <20251025191504.3024224-1-dw@davidwei.uk>
Freeing an ifq is a two step process: in io_ring_exit_work(), the ifqs
are first cleaned up via io_shutdown_zcrx_ifqs() while there are still
outstanding ctx->refs. Then once ctx->refs falls to 0, the ifqs are
freed in io_unregister_zcrx_ifqs().
The main thing to note is that io_shutdown_zcrx_ifqs() may be called
multiple times. To ensure each ifq is only cleaned up once, set
ifq->if_rxq to -1 once cleanup is done.
Proxy ifqs hold two refs: one on the src ifq and one on the src ring. In
io_shutdown_zcrx_ifqs(), dec both refs. While these refs are held, the
src ring is looping in io_ring_exit_work(). The active refs on ifq->refs
prevents the src ifqs from being cleaned up, and the active refs on
ctx->refs prevents the objects from being freed. Once all refs are gone,
the src ring proceeds with io_ring_ctx_free().
Signed-off-by: David Wei <dw@davidwei.uk>
---
io_uring/zcrx.c | 18 +++++++++++++++---
1 file changed, 15 insertions(+), 3 deletions(-)
diff --git a/io_uring/zcrx.c b/io_uring/zcrx.c
index 6b9066333fcf..2d0d1ca016c5 100644
--- a/io_uring/zcrx.c
+++ b/io_uring/zcrx.c
@@ -519,6 +519,8 @@ static void io_close_queue(struct io_zcrx_ifq *ifq)
static void io_zcrx_ifq_free(struct io_zcrx_ifq *ifq)
{
+ if (ifq->proxy)
+ goto free;
io_close_queue(ifq);
if (ifq->area)
@@ -528,6 +530,7 @@ static void io_zcrx_ifq_free(struct io_zcrx_ifq *ifq)
io_free_rbuf_ring(ifq);
mutex_destroy(&ifq->pp_lock);
+free:
kfree(ifq);
}
@@ -801,10 +804,19 @@ void io_shutdown_zcrx_ifqs(struct io_ring_ctx *ctx)
lockdep_assert_held(&ctx->uring_lock);
xa_for_each(&ctx->zcrx_ctxs, index, ifq) {
- if (refcount_read(&ifq->refs) > 1)
+ if (ifq->if_rxq == -1)
continue;
- io_zcrx_scrub(ifq);
- io_close_queue(ifq);
+
+ if (!ifq->proxy) {
+ if (refcount_read(&ifq->refs) > 1)
+ continue;
+ io_zcrx_scrub(ifq);
+ io_close_queue(ifq);
+ } else {
+ refcount_dec(&ifq->proxy->refs);
+ percpu_ref_put(&ifq->proxy->ctx->refs);
+ ifq->if_rxq = -1;
+ }
}
}
--
2.47.3
prev parent reply other threads:[~2025-10-25 19:15 UTC|newest]
Thread overview: 13+ messages / expand[flat|nested] mbox.gz Atom feed top
2025-10-25 19:14 [PATCH v2 0/5] io_uring zcrx ifq sharing David Wei
2025-10-25 19:15 ` [PATCH v2 1/5] io_uring/rsrc: rename and export io_lock_two_rings() David Wei
2025-10-25 19:15 ` [PATCH v2 2/5] io_uring/zcrx: add refcount to struct io_zcrx_ifq David Wei
2025-10-25 23:37 ` Jens Axboe
2025-10-26 4:10 ` David Wei
2025-10-25 19:15 ` [PATCH v2 3/5] io_uring/zcrx: share an ifq between rings David Wei
2025-10-25 23:41 ` Jens Axboe
2025-10-26 4:12 ` David Wei
2025-10-26 13:16 ` Jens Axboe
2025-10-26 13:43 ` Jens Axboe
2025-10-26 15:06 ` David Wei
2025-10-25 19:15 ` [PATCH v2 4/5] io_uring/zcrx: redirect io_recvzc on proxy ifq to src ifq David Wei
2025-10-25 19:15 ` David Wei [this message]
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=20251025191504.3024224-6-dw@davidwei.uk \
--to=dw@davidwei.uk \
--cc=asml.silence@gmail.com \
--cc=axboe@kernel.dk \
--cc=io-uring@vger.kernel.org \
--cc=netdev@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