Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
From:   Pavel Begunkov <asml.silence@gmail.com>
To:     io-uring@vger.kernel.org, netdev@vger.kernel.org
Cc:     Jens Axboe <axboe@kernel.dk>,
        "David S . Miller" <davem@davemloft.net>,
        Jakub Kicinski <kuba@kernel.org>, kernel-team@fb.com,
        linux-kernel@vger.kernel.org, xen-devel@lists.xenproject.org,
        Wei Liu <wei.liu@kernel.org>, Paul Durrant <paul@xen.org>,
        kvm@vger.kernel.org, virtualization@lists.linux-foundation.org,
        "Michael S . Tsirkin" <mst@redhat.com>,
        Jason Wang <jasowang@redhat.com>,
        Pavel Begunkov <asml.silence@gmail.com>
Subject: [RFC net-next io_uring 11/11] io_uring/notif: add ubuf_info ref caching
Date:   Wed, 10 Aug 2022 16:49:19 +0100
Message-Id: <9514ddeea83902c36887ce764bc0306a518af73c.1660124059.git.asml.silence@gmail.com>
In-Reply-To: <cover.1660124059.git.asml.silence@gmail.com>
References: <cover.1660124059.git.asml.silence@gmail.com>
MIME-Version: 1.0
Content-Transfer-Encoding: 8bit
Precedence: bulk

Cache some active notifier references at the io_uring side and get them
in batches, so the ammortised cost is low. Then these references can be
given away to the network layer using UARGFL_GIFT_REF.

Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
---
 io_uring/net.c   |  8 +++++++-
 io_uring/notif.c |  6 ++++--
 io_uring/notif.h | 22 +++++++++++++++++++++-
 3 files changed, 32 insertions(+), 4 deletions(-)

diff --git a/io_uring/net.c b/io_uring/net.c
index e6fc9748fbd2..bdaf9b10bd1b 100644
--- a/io_uring/net.c
+++ b/io_uring/net.c
@@ -949,6 +949,7 @@ int io_sendzc(struct io_kiocb *req, unsigned int issue_flags)
 	struct io_sendzc *zc = io_kiocb_to_cmd(req);
 	struct io_notif_slot *notif_slot;
 	struct io_kiocb *notif;
+	struct ubuf_info *ubuf;
 	struct msghdr msg;
 	struct iovec iov;
 	struct socket *sock;
@@ -1007,10 +1008,15 @@ int io_sendzc(struct io_kiocb *req, unsigned int issue_flags)
 		min_ret = iov_iter_count(&msg.msg_iter);
 
 	msg.msg_flags = msg_flags;
-	msg.msg_ubuf = &io_notif_to_data(notif)->uarg;
 	msg.sg_from_iter = io_sg_from_iter;
+	msg.msg_ubuf = ubuf = &io_notif_to_data(notif)->uarg;
+	ubuf->flags |= UARGFL_GIFT_REF;
 	ret = sock_sendmsg(sock, &msg);
 
+	/* check if the send consumed an additional ref */
+	if (likely(!(ubuf->flags & UARGFL_GIFT_REF)))
+		io_notif_consume_ref(notif);
+
 	if (unlikely(ret < min_ret)) {
 		if (ret == -EAGAIN && (issue_flags & IO_URING_F_NONBLOCK))
 			return -EAGAIN;
diff --git a/io_uring/notif.c b/io_uring/notif.c
index dd346ea67580..73bbda5de07d 100644
--- a/io_uring/notif.c
+++ b/io_uring/notif.c
@@ -68,15 +68,17 @@ struct io_kiocb *io_alloc_notif(struct io_ring_ctx *ctx,
 	nd->uarg.skb_flags = SKBFL_ZEROCOPY_FRAG | SKBFL_DONT_ORPHAN;
 	nd->uarg.flags = UARGFL_CALLER_PINNED;
 	nd->uarg.callback = io_uring_tx_zerocopy_callback;
+	nd->cached_refs = IO_NOTIF_REF_CACHE_NR;
 	/* master ref owned by io_notif_slot, will be dropped on flush */
-	refcount_set(&nd->uarg.refcnt, 1);
+	refcount_set(&nd->uarg.refcnt, IO_NOTIF_REF_CACHE_NR + 1);
 	return notif;
 }
 
 static inline bool io_notif_drop_refs(struct io_notif_data *nd)
 {
-	int refs = 1;
+	int refs = nd->cached_refs + 1;
 
+	nd->cached_refs = 0;
 	return refcount_sub_and_test(refs, &nd->uarg.refcnt);
 }
 
diff --git a/io_uring/notif.h b/io_uring/notif.h
index 0819304d7e00..2a263055a53b 100644
--- a/io_uring/notif.h
+++ b/io_uring/notif.h
@@ -9,11 +9,14 @@
 
 #define IO_NOTIF_SPLICE_BATCH	32
 #define IORING_MAX_NOTIF_SLOTS (1U << 10)
+#define IO_NOTIF_REF_CACHE_NR	64
 
 struct io_notif_data {
 	struct file		*file;
-	struct ubuf_info	uarg;
 	unsigned long		account_pages;
+	/* extra uarg->refcnt refs */
+	int			cached_refs;
+	struct ubuf_info	uarg;
 };
 
 struct io_notif_slot {
@@ -88,3 +91,20 @@ static inline int io_notif_account_mem(struct io_kiocb *notif, unsigned len)
 	}
 	return 0;
 }
+
+static inline void io_notif_consume_ref(struct io_kiocb *notif)
+	__must_hold(&ctx->uring_lock)
+{
+	struct io_notif_data *nd = io_notif_to_data(notif);
+
+	nd->cached_refs--;
+
+	/*
+	* Issue sends without looking at notif->cached_refs first, so we
+	* always have to have at least one ref cached
+	*/
+	if (unlikely(!nd->cached_refs)) {
+		refcount_add(IO_NOTIF_REF_CACHE_NR, &nd->uarg.refcnt);
+		nd->cached_refs += IO_NOTIF_REF_CACHE_NR;
+	}
+}
-- 
2.37.0