Netdev List
 help / color / mirror / Atom feed
From: Jeremy Kerr <jk@codeconstruct.com.au>
To: Matt Johnston <matt@codeconstruct.com.au>,
	 Andrew Lunn <andrew+netdev@lunn.ch>,
	 "David S. Miller" <davem@davemloft.net>,
	Eric Dumazet <edumazet@google.com>,
	 Jakub Kicinski <kuba@kernel.org>,
	Paolo Abeni <pabeni@redhat.com>,
	 Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Cc: netdev@vger.kernel.org, linux-usb@vger.kernel.org
Subject: [PATCH net-next v2 12/12] net: mctp: usb: Allow multiple urbs in flight
Date: Fri, 03 Jul 2026 13:47:32 +0800	[thread overview]
Message-ID: <20260703-dev-mctp-usb-1-1-v2-12-60367b861b33@codeconstruct.com.au> (raw)
In-Reply-To: <20260703-dev-mctp-usb-1-1-v2-0-60367b861b33@codeconstruct.com.au>

Currently, we stop tx queues when we have one urb submitted. This means
we will immediately hit dev_hard_start_xmit's tx-queues-off ->
NETDEV_TX_BUSY case, and revert to the requeue -> gso_skb single-dequeue
path, and no longer be able to pack skbs without an xmit_more
indication.

Instead, allow a few urbs to be in-flight, with a limit of 16kB of data
outstanding (after which we will disable queues). With this, the tx path
will cause fewer requeues (and therefore non-packed transfers) under
normal loads.

Signed-off-by: Jeremy Kerr <jk@codeconstruct.com.au>
---
v2:
 - rework tx_qmem locking, prevent race between completion and
   submission for stopping/waking the tx queues; perform the wake while
   locked
 - only wake if netif_running()
---
 drivers/net/mctp/mctp-usb.c | 33 +++++++++++++++++++++++++++++----
 1 file changed, 29 insertions(+), 4 deletions(-)

diff --git a/drivers/net/mctp/mctp-usb.c b/drivers/net/mctp/mctp-usb.c
index 4bf5546d520c..fd33f1b97e23 100644
--- a/drivers/net/mctp/mctp-usb.c
+++ b/drivers/net/mctp/mctp-usb.c
@@ -39,6 +39,9 @@ struct mctp_usb {
 
 	struct mctp_usblib_tx tx;
 	struct usb_anchor tx_anchor;
+	/* serialises tx_qmem updates to netdev queue states */
+	spinlock_t tx_qmem_lock;
+	int tx_qmem;
 };
 
 enum {
@@ -46,23 +49,41 @@ enum {
 	MCTP_USB_SUBCLASS_SPAN = 0x02,
 };
 
+/* We use a total-size limit for outstanding URBs, as the transfer counts
+ * may vary a lot between spanning- and non-spanning modes. In spanning mode,
+ * this will allow for a couple of max-sized transfers to be in flight. In
+ * non-spanning mode, 32.
+ *
+ * We want to avoid disabling the tx queue if possible; doing so will end up
+ * requeueing to gso_skb, and we only dequeue from that one skb at a time,
+ * so can no longer perform transfer packing.
+ */
+static const unsigned int TX_QMEM_MAX = 16384;
+
 static void mctp_usb_out_complete(struct urb *urb)
 {
 	struct mctp_usblib_tx_ctx *tx_ctx = urb->context;
 	struct mctp_usb *mctp_usb = mctp_usblib_tx_ctx_priv(tx_ctx);
+	unsigned int len = urb->transfer_buffer_length;
 	struct net_device *netdev = mctp_usb->netdev;
+	unsigned long flags;
 
 	mctp_usblib_tx_send_complete(tx_ctx, netdev, urb->status == 0);
 
 	usb_free_urb(urb);
 
-	netif_wake_queue(netdev);
+	spin_lock_irqsave(&mctp_usb->tx_qmem_lock, flags);
+	mctp_usb->tx_qmem -= len;
+	if (mctp_usb->tx_qmem < TX_QMEM_MAX && netif_running(netdev))
+		netif_wake_queue(netdev);
+	spin_unlock_irqrestore(&mctp_usb->tx_qmem_lock, flags);
 }
 
 static int mctp_usb_tx_send(struct mctp_usblib_tx_ctx *tx_ctx,
 			    void *data, size_t len)
 {
 	struct mctp_usb *mctp_usb = mctp_usblib_tx_ctx_priv(tx_ctx);
+	unsigned long flags;
 	struct urb *urb;
 	int rc;
 
@@ -77,8 +98,6 @@ static int mctp_usb_tx_send(struct mctp_usblib_tx_ctx *tx_ctx,
 	if (mctp_usb->span)
 		urb->transfer_flags |= URB_ZERO_PACKET;
 
-	netif_stop_queue(mctp_usb->netdev);
-
 	usb_anchor_urb(urb, &mctp_usb->tx_anchor);
 
 	rc = usb_submit_urb(urb, GFP_ATOMIC);
@@ -86,7 +105,12 @@ static int mctp_usb_tx_send(struct mctp_usblib_tx_ctx *tx_ctx,
 		netdev_dbg(mctp_usb->netdev, "TX urb submit failed, %d\n", rc);
 		usb_unanchor_urb(urb);
 		usb_free_urb(urb);
-		netif_start_queue(mctp_usb->netdev);
+	} else {
+		spin_lock_irqsave(&mctp_usb->tx_qmem_lock, flags);
+		mctp_usb->tx_qmem += len;
+		if (mctp_usb->tx_qmem >= TX_QMEM_MAX)
+			netif_stop_queue(mctp_usb->netdev);
+		spin_unlock_irqrestore(&mctp_usb->tx_qmem_lock, flags);
 	}
 
 	return rc;
@@ -283,6 +307,7 @@ static int mctp_usb_probe(struct usb_interface *intf,
 	if (dev->span)
 		netdev->max_mtu = MCTP_USB_1_1_MTU_MAX;
 	spin_lock_init(&dev->rx_lock);
+	spin_lock_init(&dev->tx_qmem_lock);
 	usb_set_intfdata(intf, dev);
 
 	mctp_usblib_rx_init(&dev->rx, dev->span);

-- 
2.47.3


      parent reply	other threads:[~2026-07-03  5:48 UTC|newest]

Thread overview: 13+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2026-07-03  5:47 [PATCH net-next v2 00/12] net: mctp: usb: Add support for MCTP-over-USB v1.1 Jeremy Kerr
2026-07-03  5:47 ` [PATCH net-next v2 01/12] net: mctp: usb: Include version indicator in max packet size defines Jeremy Kerr
2026-07-03  5:47 ` [PATCH net-next v2 02/12] net: mctp: usb: Use packet-length max for maximum packet-size check Jeremy Kerr
2026-07-03  5:47 ` [PATCH net-next v2 03/12] net: mctp: usblib: Move RX transfer processing to a new mctp-usblib Jeremy Kerr
2026-07-03  5:47 ` [PATCH net-next v2 04/12] net: mctp: usblib: Move TX transfer processing to mctp-usblib Jeremy Kerr
2026-07-03  5:47 ` [PATCH net-next v2 05/12] net: mctp: usb: Allow for multiple urb submissions from a packet tx Jeremy Kerr
2026-07-03  5:47 ` [PATCH net-next v2 06/12] net: mctp: usblib: Add support for multi-packet transmit Jeremy Kerr
2026-07-03  5:47 ` [PATCH net-next v2 07/12] net: mctp: usb: Accommodate DSP0283 v1.1 header format Jeremy Kerr
2026-07-03  5:47 ` [PATCH net-next v2 08/12] net: mctp: usblib: Implement receive-side packet spanning Jeremy Kerr
2026-07-03  5:47 ` [PATCH net-next v2 09/12] net: mctp: usblib: Implement transmit-side " Jeremy Kerr
2026-07-03  5:47 ` [PATCH net-next v2 10/12] net: mctp: usblib: Add initial kunit tests Jeremy Kerr
2026-07-03  5:47 ` [PATCH net-next v2 11/12] net: mctp: usb: enable v1.1 packet spanning Jeremy Kerr
2026-07-03  5:47 ` Jeremy Kerr [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=20260703-dev-mctp-usb-1-1-v2-12-60367b861b33@codeconstruct.com.au \
    --to=jk@codeconstruct.com.au \
    --cc=andrew+netdev@lunn.ch \
    --cc=davem@davemloft.net \
    --cc=edumazet@google.com \
    --cc=gregkh@linuxfoundation.org \
    --cc=kuba@kernel.org \
    --cc=linux-usb@vger.kernel.org \
    --cc=matt@codeconstruct.com.au \
    --cc=netdev@vger.kernel.org \
    --cc=pabeni@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