diff mbox series

[net-next,v6,08/13] net: dsa: microchip: ptp: add packet transmission timestamping

Message ID 20230102050459.31023-9-arun.ramadoss@microchip.com (mailing list archive)
State Superseded
Delegated to: Netdev Maintainers
Headers show
Series net: dsa: microchip: add PTP support for KSZ9563/KSZ8563 and LAN937x | expand

Checks

Context Check Description
netdev/tree_selection success Clearly marked for net-next
netdev/fixes_present success Fixes tag not required for -next series
netdev/subject_prefix success Link
netdev/cover_letter success Series has a cover letter
netdev/patch_count success Link
netdev/header_inline success No static functions without inline keyword in header files
netdev/build_32bit success Errors and warnings before: 0 this patch: 0
netdev/cc_maintainers success CCed 11 of 11 maintainers
netdev/build_clang success Errors and warnings before: 0 this patch: 0
netdev/module_param success Was 0 now: 0
netdev/verify_signedoff success Signed-off-by tag matches author and committer
netdev/check_selftest success No net selftest shell script
netdev/verify_fixes success No Fixes tag
netdev/build_allmodconfig_warn success Errors and warnings before: 0 this patch: 0
netdev/checkpatch success total: 0 errors, 0 warnings, 0 checks, 341 lines checked
netdev/kdoc success Errors and warnings before: 0 this patch: 0
netdev/source_inline success Was 0 now: 0

Commit Message

Arun Ramadoss Jan. 2, 2023, 5:04 a.m. UTC 
From: Christian Eggers <ceggers@arri.de>

This patch adds the routines for transmission of ptp packets. When the
ptp pdelay_req packet to be transmitted, it uses the deferred xmit
worker to schedule the packets.
During irq_setup, interrupt for Sync, Pdelay_req and Pdelay_rsp are
enabled. So interrupt is triggered for all three packets. But for
p2p1step, we require only time stamp of Pdelay_req packet. Hence to
avoid posting of the completion from ISR routine for Sync and
Pdelay_resp packets, ts_en flag is introduced. This controls which
packets need to processed for timestamp.
After the packet is transmitted, ISR is triggered. The time at which
packet transmitted is recorded to separate register.
This value is reconstructed to absolute time and posted to the user
application through socket error queue.

Signed-off-by: Christian Eggers <ceggers@arri.de>
Co-developed-by: Arun Ramadoss <arun.ramadoss@microchip.com>
Signed-off-by: Arun Ramadoss <arun.ramadoss@microchip.com>
Reviewed-by: Vladimir Oltean <olteanv@gmail.com>
---
v1 -> v2
- Declared the deferred xmit and ksz_port_txtstamp function as null in
ptp disabled case

RFC v2 -> Patch v1
- separated the pdelay_rsp message correction update in different patch
---
 drivers/net/dsa/microchip/ksz_common.c |  14 +++
 drivers/net/dsa/microchip/ksz_common.h |   3 +
 drivers/net/dsa/microchip/ksz_ptp.c    | 116 ++++++++++++++++++++++++-
 drivers/net/dsa/microchip/ksz_ptp.h    |   6 ++
 include/linux/dsa/ksz_common.h         |   8 ++
 net/dsa/tag_ksz.c                      |  54 +++++++++++-
 6 files changed, 197 insertions(+), 4 deletions(-)

Comments

Vladimir Oltean Jan. 3, 2023, 5:32 p.m. UTC | #1 
On Mon, Jan 02, 2023 at 10:34:54AM +0530, Arun Ramadoss wrote:
> diff --git a/drivers/net/dsa/microchip/ksz_ptp.c b/drivers/net/dsa/microchip/ksz_ptp.c
> index e72fce54188e..8f5e099b1b99 100644
> --- a/drivers/net/dsa/microchip/ksz_ptp.c
> +++ b/drivers/net/dsa/microchip/ksz_ptp.c
> @@ -18,6 +18,8 @@
>  
>  #define ptp_caps_to_data(d) container_of((d), struct ksz_ptp_data, caps)
>  #define ptp_data_to_ksz_dev(d) container_of((d), struct ksz_device, ptp_data)
> +#define work_to_xmit_work(w) \
> +		container_of((w), struct ksz_deferred_xmit_work, work)
>  
>  /* Sub-nanoseconds-adj,max * sub-nanoseconds / 40ns * 1ns
>   * = (2^30-1) * (2 ^ 32) / 40 ns * 1 ns = 6249999
> @@ -111,9 +113,15 @@ static int ksz_set_hwtstamp_config(struct ksz_device *dev,
>  
>  	switch (config->tx_type) {
>  	case HWTSTAMP_TX_OFF:
> +		prt->ptpmsg_irq[KSZ_SYNC_MSG].ts_en  = 0;
> +		prt->ptpmsg_irq[KSZ_XDREQ_MSG].ts_en = 0;
> +		prt->ptpmsg_irq[KSZ_PDRES_MSG].ts_en = 0;
>  		prt->hwts_tx_en = false;
>  		break;
>  	case HWTSTAMP_TX_ONESTEP_P2P:
> +		prt->ptpmsg_irq[KSZ_SYNC_MSG].ts_en  = 0;
> +		prt->ptpmsg_irq[KSZ_XDREQ_MSG].ts_en = 1;
> +		prt->ptpmsg_irq[KSZ_PDRES_MSG].ts_en = 0;

Please use true/false for boolean types. I believe I've said this before.

>  		prt->hwts_tx_en = true;
>  		break;
>  	default:
> @@ -232,6 +240,88 @@ bool ksz_port_rxtstamp(struct dsa_switch *ds, int port, struct sk_buff *skb,
>  	return false;
>  }
>  
> +void ksz_port_txtstamp(struct dsa_switch *ds, int port,
> +		       struct sk_buff *skb)

Function prototype fits on one line.

> +{
> +	struct ksz_device *dev	= ds->priv;

Again, I believe the tab here is misused, as there is nothing to align
the equal sign to.

> +	struct ptp_header *hdr;
> +	struct sk_buff *clone;
> +	struct ksz_port *prt;
> +	unsigned int type;
> +	u8 ptp_msg_type;
> +
> +	prt = &dev->ports[port];
> +
> +	if (!prt->hwts_tx_en)
> +		return;
> +
> +	type = ptp_classify_raw(skb);
> +	if (type == PTP_CLASS_NONE)
> +		return;
> +
> +	hdr = ptp_parse_header(skb, type);
> +	if (!hdr)
> +		return;
> +
> +	ptp_msg_type = ptp_get_msgtype(hdr, type);
> +
> +	switch (ptp_msg_type) {
> +	case PTP_MSGTYPE_PDELAY_REQ:
> +		break;
> +
> +	default:
> +		return;
> +	}
> +
> +	clone = skb_clone_sk(skb);
> +	if (!clone)
> +		return;
> +
> +	/* caching the value to be used in tag_ksz.c */
> +	KSZ_SKB_CB(skb)->clone = clone;
> +}
> +
> +static void ksz_ptp_txtstamp_skb(struct ksz_device *dev,
> +				 struct ksz_port *prt, struct sk_buff *skb)
> +{
> +	struct skb_shared_hwtstamps hwtstamps = {};
> +	int ret;
> +
> +	skb_shinfo(skb)->tx_flags |= SKBTX_IN_PROGRESS;

It would be good if you set SKBTX_IN_PROGRESS _before_ passing the skb
ownership to the DSA master (dsa_enqueue_skb). It makes a certain class
of bugs much easier to identify, because they will reproduce much more
consistently. Otherwise, as the way things are written, the DSA master
driver will sometimes see this skb as having SKBTX_IN_PROGRESS and
sometimes not in progress (because the assignment races with the
ndo_start_xmit of that other driver).

I know perfectly well that the DSA master driver should not look at the
SKBTX_IN_PROGRESS flag of a skb which it's not support to timestamp itself.
But sometimes drivers aren't all that perfect.

> +
> +	/* timeout must include tstamp latency, IRQ latency and time for
> +	 * reading the time stamp.

FWIW, also the time necessary for the DSA master to transmit the skb,
of course.

> +	 */
> +	ret = wait_for_completion_timeout(&prt->tstamp_msg_comp,
> +					  msecs_to_jiffies(100));
> +	if (!ret)
> +		return;
> +
> +	hwtstamps.hwtstamp = prt->tstamp_msg;
> +	skb_complete_tx_timestamp(skb, &hwtstamps);
> +}
> +
> +void ksz_port_deferred_xmit(struct kthread_work *work)
> +{
> +	struct ksz_deferred_xmit_work *xmit_work = work_to_xmit_work(work);
> +	struct sk_buff *clone, *skb = xmit_work->skb;
> +	struct dsa_switch *ds = xmit_work->dp->ds;
> +	struct ksz_device *dev = ds->priv;
> +	struct ksz_port *prt;
> +
> +	prt = &dev->ports[xmit_work->dp->index];
> +
> +	clone = KSZ_SKB_CB(skb)->clone;
> +
> +	reinit_completion(&prt->tstamp_msg_comp);
> +
> +	dsa_enqueue_skb(skb, skb->dev);
> +
> +	ksz_ptp_txtstamp_skb(dev, prt, clone);
> +
> +	kfree(xmit_work);
> +}
> +
>  static int _ksz_ptp_gettime(struct ksz_device *dev, struct timespec64 *ts)
>  {
>  	u32 nanoseconds;
> @@ -480,7 +570,29 @@ void ksz_ptp_clock_unregister(struct dsa_switch *ds)
>  
>  static irqreturn_t ksz_ptp_msg_thread_fn(int irq, void *dev_id)
>  {
> -	return IRQ_NONE;
> +	struct ksz_ptp_irq *ptpmsg_irq = dev_id;
> +	struct ksz_device *dev;
> +	struct ksz_port *port;
> +	u32 tstamp_raw;
> +	ktime_t tstamp;
> +	int ret;
> +
> +	port = ptpmsg_irq->port;
> +	dev = port->ksz_dev;
> +
> +	if (ptpmsg_irq->ts_en) {
> +		ret = ksz_read32(dev, ptpmsg_irq->ts_reg, &tstamp_raw);
> +		if (ret)
> +			return IRQ_NONE;
> +
> +		tstamp = ksz_decode_tstamp(tstamp_raw);
> +
> +		port->tstamp_msg = ksz_tstamp_reconstruct(dev, tstamp);
> +
> +		complete(&port->tstamp_msg_comp);
> +	}
> +
> +	return IRQ_HANDLED;
>  }
diff mbox series

Patch

diff --git a/drivers/net/dsa/microchip/ksz_common.c b/drivers/net/dsa/microchip/ksz_common.c
index 5e3bb14388b2..f9172b359232 100644
--- a/drivers/net/dsa/microchip/ksz_common.c
+++ b/drivers/net/dsa/microchip/ksz_common.c
@@ -6,6 +6,7 @@ 
  */
 
 #include <linux/delay.h>
+#include <linux/dsa/ksz_common.h>
 #include <linux/export.h>
 #include <linux/gpio/consumer.h>
 #include <linux/kernel.h>
@@ -2539,6 +2540,17 @@  static enum dsa_tag_protocol ksz_get_tag_protocol(struct dsa_switch *ds,
 	return proto;
 }
 
+static int ksz_connect_tag_protocol(struct dsa_switch *ds,
+				    enum dsa_tag_protocol proto)
+{
+	struct ksz_tagger_data *tagger_data;
+
+	tagger_data = ksz_tagger_data(ds);
+	tagger_data->xmit_work_fn = ksz_port_deferred_xmit;
+
+	return 0;
+}
+
 static int ksz_port_vlan_filtering(struct dsa_switch *ds, int port,
 				   bool flag, struct netlink_ext_ack *extack)
 {
@@ -2954,6 +2966,7 @@  static int ksz_switch_detect(struct ksz_device *dev)
 
 static const struct dsa_switch_ops ksz_switch_ops = {
 	.get_tag_protocol	= ksz_get_tag_protocol,
+	.connect_tag_protocol   = ksz_connect_tag_protocol,
 	.get_phy_flags		= ksz_get_phy_flags,
 	.setup			= ksz_setup,
 	.teardown		= ksz_teardown,
@@ -2991,6 +3004,7 @@  static const struct dsa_switch_ops ksz_switch_ops = {
 	.get_ts_info            = ksz_get_ts_info,
 	.port_hwtstamp_get      = ksz_hwtstamp_get,
 	.port_hwtstamp_set      = ksz_hwtstamp_set,
+	.port_txtstamp		= ksz_port_txtstamp,
 	.port_rxtstamp		= ksz_port_rxtstamp,
 };
 
diff --git a/drivers/net/dsa/microchip/ksz_common.h b/drivers/net/dsa/microchip/ksz_common.h
index ec1bceb4efcc..c8b49c86dfe1 100644
--- a/drivers/net/dsa/microchip/ksz_common.h
+++ b/drivers/net/dsa/microchip/ksz_common.h
@@ -87,6 +87,7 @@  struct ksz_irq {
 struct ksz_ptp_irq {
 	struct ksz_port *port;
 	u16 ts_reg;
+	bool ts_en;
 	char name[16];
 	int num;
 };
@@ -116,6 +117,8 @@  struct ksz_port {
 	bool hwts_rx_en;
 	struct ksz_irq ptpirq;
 	struct ksz_ptp_irq ptpmsg_irq[3];
+	ktime_t tstamp_msg;
+	struct completion tstamp_msg_comp;
 #endif
 };
 
diff --git a/drivers/net/dsa/microchip/ksz_ptp.c b/drivers/net/dsa/microchip/ksz_ptp.c
index e72fce54188e..8f5e099b1b99 100644
--- a/drivers/net/dsa/microchip/ksz_ptp.c
+++ b/drivers/net/dsa/microchip/ksz_ptp.c
@@ -18,6 +18,8 @@ 
 
 #define ptp_caps_to_data(d) container_of((d), struct ksz_ptp_data, caps)
 #define ptp_data_to_ksz_dev(d) container_of((d), struct ksz_device, ptp_data)
+#define work_to_xmit_work(w) \
+		container_of((w), struct ksz_deferred_xmit_work, work)
 
 /* Sub-nanoseconds-adj,max * sub-nanoseconds / 40ns * 1ns
  * = (2^30-1) * (2 ^ 32) / 40 ns * 1 ns = 6249999
@@ -111,9 +113,15 @@  static int ksz_set_hwtstamp_config(struct ksz_device *dev,
 
 	switch (config->tx_type) {
 	case HWTSTAMP_TX_OFF:
+		prt->ptpmsg_irq[KSZ_SYNC_MSG].ts_en  = 0;
+		prt->ptpmsg_irq[KSZ_XDREQ_MSG].ts_en = 0;
+		prt->ptpmsg_irq[KSZ_PDRES_MSG].ts_en = 0;
 		prt->hwts_tx_en = false;
 		break;
 	case HWTSTAMP_TX_ONESTEP_P2P:
+		prt->ptpmsg_irq[KSZ_SYNC_MSG].ts_en  = 0;
+		prt->ptpmsg_irq[KSZ_XDREQ_MSG].ts_en = 1;
+		prt->ptpmsg_irq[KSZ_PDRES_MSG].ts_en = 0;
 		prt->hwts_tx_en = true;
 		break;
 	default:
@@ -232,6 +240,88 @@  bool ksz_port_rxtstamp(struct dsa_switch *ds, int port, struct sk_buff *skb,
 	return false;
 }
 
+void ksz_port_txtstamp(struct dsa_switch *ds, int port,
+		       struct sk_buff *skb)
+{
+	struct ksz_device *dev	= ds->priv;
+	struct ptp_header *hdr;
+	struct sk_buff *clone;
+	struct ksz_port *prt;
+	unsigned int type;
+	u8 ptp_msg_type;
+
+	prt = &dev->ports[port];
+
+	if (!prt->hwts_tx_en)
+		return;
+
+	type = ptp_classify_raw(skb);
+	if (type == PTP_CLASS_NONE)
+		return;
+
+	hdr = ptp_parse_header(skb, type);
+	if (!hdr)
+		return;
+
+	ptp_msg_type = ptp_get_msgtype(hdr, type);
+
+	switch (ptp_msg_type) {
+	case PTP_MSGTYPE_PDELAY_REQ:
+		break;
+
+	default:
+		return;
+	}
+
+	clone = skb_clone_sk(skb);
+	if (!clone)
+		return;
+
+	/* caching the value to be used in tag_ksz.c */
+	KSZ_SKB_CB(skb)->clone = clone;
+}
+
+static void ksz_ptp_txtstamp_skb(struct ksz_device *dev,
+				 struct ksz_port *prt, struct sk_buff *skb)
+{
+	struct skb_shared_hwtstamps hwtstamps = {};
+	int ret;
+
+	skb_shinfo(skb)->tx_flags |= SKBTX_IN_PROGRESS;
+
+	/* timeout must include tstamp latency, IRQ latency and time for
+	 * reading the time stamp.
+	 */
+	ret = wait_for_completion_timeout(&prt->tstamp_msg_comp,
+					  msecs_to_jiffies(100));
+	if (!ret)
+		return;
+
+	hwtstamps.hwtstamp = prt->tstamp_msg;
+	skb_complete_tx_timestamp(skb, &hwtstamps);
+}
+
+void ksz_port_deferred_xmit(struct kthread_work *work)
+{
+	struct ksz_deferred_xmit_work *xmit_work = work_to_xmit_work(work);
+	struct sk_buff *clone, *skb = xmit_work->skb;
+	struct dsa_switch *ds = xmit_work->dp->ds;
+	struct ksz_device *dev = ds->priv;
+	struct ksz_port *prt;
+
+	prt = &dev->ports[xmit_work->dp->index];
+
+	clone = KSZ_SKB_CB(skb)->clone;
+
+	reinit_completion(&prt->tstamp_msg_comp);
+
+	dsa_enqueue_skb(skb, skb->dev);
+
+	ksz_ptp_txtstamp_skb(dev, prt, clone);
+
+	kfree(xmit_work);
+}
+
 static int _ksz_ptp_gettime(struct ksz_device *dev, struct timespec64 *ts)
 {
 	u32 nanoseconds;
@@ -480,7 +570,29 @@  void ksz_ptp_clock_unregister(struct dsa_switch *ds)
 
 static irqreturn_t ksz_ptp_msg_thread_fn(int irq, void *dev_id)
 {
-	return IRQ_NONE;
+	struct ksz_ptp_irq *ptpmsg_irq = dev_id;
+	struct ksz_device *dev;
+	struct ksz_port *port;
+	u32 tstamp_raw;
+	ktime_t tstamp;
+	int ret;
+
+	port = ptpmsg_irq->port;
+	dev = port->ksz_dev;
+
+	if (ptpmsg_irq->ts_en) {
+		ret = ksz_read32(dev, ptpmsg_irq->ts_reg, &tstamp_raw);
+		if (ret)
+			return IRQ_NONE;
+
+		tstamp = ksz_decode_tstamp(tstamp_raw);
+
+		port->tstamp_msg = ksz_tstamp_reconstruct(dev, tstamp);
+
+		complete(&port->tstamp_msg_comp);
+	}
+
+	return IRQ_HANDLED;
 }
 
 static irqreturn_t ksz_ptp_irq_thread_fn(int irq, void *dev_id)
@@ -625,6 +737,8 @@  int ksz_ptp_irq_setup(struct dsa_switch *ds, u8 p)
 						REG_PTP_PORT_TX_INT_STATUS__2);
 	snprintf(ptpirq->name, sizeof(ptpirq->name), "ptp-irq-%d", p);
 
+	init_completion(&port->tstamp_msg_comp);
+
 	ptpirq->domain = irq_domain_add_linear(dev->dev->of_node, ptpirq->nirqs,
 					       &ksz_ptp_irq_domain_ops, ptpirq);
 	if (!ptpirq->domain)
diff --git a/drivers/net/dsa/microchip/ksz_ptp.h b/drivers/net/dsa/microchip/ksz_ptp.h
index 9bb8fb059ac2..0b14aed71ec2 100644
--- a/drivers/net/dsa/microchip/ksz_ptp.h
+++ b/drivers/net/dsa/microchip/ksz_ptp.h
@@ -30,6 +30,8 @@  int ksz_get_ts_info(struct dsa_switch *ds, int port,
 		    struct ethtool_ts_info *ts);
 int ksz_hwtstamp_get(struct dsa_switch *ds, int port, struct ifreq *ifr);
 int ksz_hwtstamp_set(struct dsa_switch *ds, int port, struct ifreq *ifr);
+void ksz_port_txtstamp(struct dsa_switch *ds, int port, struct sk_buff *skb);
+void ksz_port_deferred_xmit(struct kthread_work *work);
 bool ksz_port_rxtstamp(struct dsa_switch *ds, int port, struct sk_buff *skb,
 		       unsigned int type);
 int ksz_ptp_irq_setup(struct dsa_switch *ds, u8 p);
@@ -64,6 +66,10 @@  static inline void ksz_ptp_irq_free(struct dsa_switch *ds, u8 p) {}
 
 #define ksz_port_rxtstamp NULL
 
+#define ksz_port_txtstamp NULL
+
+#define ksz_port_deferred_xmit NULL
+
 #endif	/* End of CONFIG_NET_DSA_MICROCHIP_KSZ_PTP */
 
 #endif
diff --git a/include/linux/dsa/ksz_common.h b/include/linux/dsa/ksz_common.h
index a256b08d837d..b91beab5e138 100644
--- a/include/linux/dsa/ksz_common.h
+++ b/include/linux/dsa/ksz_common.h
@@ -23,11 +23,19 @@  static inline ktime_t ksz_decode_tstamp(u32 tstamp)
 	return ns_to_ktime(ns);
 }
 
+struct ksz_deferred_xmit_work {
+	struct dsa_port *dp;
+	struct sk_buff *skb;
+	struct kthread_work work;
+};
+
 struct ksz_tagger_data {
+	void (*xmit_work_fn)(struct kthread_work *work);
 	void (*hwtstamp_set_state)(struct dsa_switch *ds, bool on);
 };
 
 struct ksz_skb_cb {
+	struct sk_buff *clone;
 	u32 tstamp;
 };
 
diff --git a/net/dsa/tag_ksz.c b/net/dsa/tag_ksz.c
index 6603eaa234d2..e14ee26bf6a0 100644
--- a/net/dsa/tag_ksz.c
+++ b/net/dsa/tag_ksz.c
@@ -26,6 +26,7 @@ 
 struct ksz_tagger_private {
 	struct ksz_tagger_data data; /* Must be first */
 	unsigned long state;
+	struct kthread_worker *xmit_worker;
 };
 
 static struct ksz_tagger_private *
@@ -48,6 +49,7 @@  static void ksz_disconnect(struct dsa_switch *ds)
 {
 	struct ksz_tagger_private *priv = ds->tagger_data;
 
+	kthread_destroy_worker(priv->xmit_worker);
 	kfree(priv);
 	ds->tagger_data = NULL;
 }
@@ -55,12 +57,23 @@  static void ksz_disconnect(struct dsa_switch *ds)
 static int ksz_connect(struct dsa_switch *ds)
 {
 	struct ksz_tagger_data *tagger_data;
+	struct kthread_worker *xmit_worker;
 	struct ksz_tagger_private *priv;
+	int ret;
 
 	priv = kzalloc(sizeof(*priv), GFP_KERNEL);
 	if (!priv)
 		return -ENOMEM;
 
+	xmit_worker = kthread_create_worker(0, "dsa%d:%d_xmit",
+					    ds->dst->index, ds->index);
+	if (IS_ERR(xmit_worker)) {
+		ret = PTR_ERR(xmit_worker);
+		kfree(priv);
+		return ret;
+	}
+
+	priv->xmit_worker = xmit_worker;
 	/* Export functions for switch driver use */
 	tagger_data = &priv->data;
 	tagger_data->hwtstamp_set_state = ksz_hwtstamp_set_state;
@@ -191,6 +204,41 @@  static void ksz_xmit_timestamp(struct dsa_port *dp, struct sk_buff *skb)
 	put_unaligned_be32(0, skb_put(skb, KSZ_PTP_TAG_LEN));
 }
 
+/* Defer transmit if waiting for egress time stamp is required.  */
+static struct sk_buff *ksz_defer_xmit(struct dsa_port *dp, struct sk_buff *skb)
+{
+	struct ksz_tagger_data *tagger_data = ksz_tagger_data(dp->ds);
+	struct ksz_tagger_private *priv = ksz_tagger_private(dp->ds);
+	void (*xmit_work_fn)(struct kthread_work *work);
+	struct sk_buff *clone = KSZ_SKB_CB(skb)->clone;
+	struct ksz_deferred_xmit_work *xmit_work;
+	struct kthread_worker *xmit_worker;
+
+	if (!clone)
+		return skb;  /* no deferred xmit for this packet */
+
+	xmit_work_fn = tagger_data->xmit_work_fn;
+	xmit_worker = priv->xmit_worker;
+
+	if (!xmit_work_fn || !xmit_worker)
+		return NULL;
+
+	xmit_work = kzalloc(sizeof(*xmit_work), GFP_ATOMIC);
+	if (!xmit_work)
+		return NULL;
+
+	kthread_init_work(&xmit_work->work, xmit_work_fn);
+	/* Increase refcount so the kfree_skb in dsa_slave_xmit
+	 * won't really free the packet.
+	 */
+	xmit_work->dp = dp;
+	xmit_work->skb = skb_get(skb);
+
+	kthread_queue_work(xmit_worker, &xmit_work->work);
+
+	return NULL;
+}
+
 static struct sk_buff *ksz9477_xmit(struct sk_buff *skb,
 				    struct net_device *dev)
 {
@@ -215,7 +263,7 @@  static struct sk_buff *ksz9477_xmit(struct sk_buff *skb,
 
 	*tag = cpu_to_be16(val);
 
-	return skb;
+	return ksz_defer_xmit(dp, skb);
 }
 
 static struct sk_buff *ksz9477_rcv(struct sk_buff *skb, struct net_device *dev)
@@ -271,7 +319,7 @@  static struct sk_buff *ksz9893_xmit(struct sk_buff *skb,
 	if (is_link_local_ether_addr(addr))
 		*tag |= KSZ9893_TAIL_TAG_OVERRIDE;
 
-	return skb;
+	return ksz_defer_xmit(dp, skb);
 }
 
 static const struct dsa_device_ops ksz9893_netdev_ops = {
@@ -336,7 +384,7 @@  static struct sk_buff *lan937x_xmit(struct sk_buff *skb,
 
 	put_unaligned_be16(val, tag);
 
-	return skb;
+	return ksz_defer_xmit(dp, skb);
 }
 
 static const struct dsa_device_ops lan937x_netdev_ops = {