Message-ID: <a0114fa234fcf08208c3ba388beb1fe571df1c6c.camel@kernel.org>
Date:   Wed, 16 Sep 2020 12:22:38 -0700
From:   Saeed Mahameed <saeed@...nel.org>
To:     Yangbo Lu <yangbo.lu@....com>, netdev@...r.kernel.org
Cc:     "David S . Miller" <davem@...emloft.net>,
        Jakub Kicinski <kuba@...nel.org>,
        Ioana Ciornei <ioana.ciornei@....com>,
        Ioana Radulescu <ruxandra.radulescu@....com>,
        Richard Cochran <richardcochran@...il.com>
Subject: Re: [v3, 5/6] dpaa2-eth: support PTP Sync packet one-step
 timestamping

On Wed, 2020-09-16 at 20:08 +0800, Yangbo Lu wrote:
> This patch is to add PTP sync packet one-step timestamping support.
> Before egress, one-step timestamping enablement needs,
> 
> - Enabling timestamp and FAS (Frame Annotation Status) in
>   dpni buffer layout.
> 
> - Write timestamp to frame annotation and set PTP bit in
>   FAS to mark as one-step timestamping event.
> 
> - Enabling one-step timestamping by dpni_set_single_step_cfg()
>   API, with offset provided to insert correction time on frame.
>   The offset must respect all MAC headers, VLAN tags and other
>   protocol headers accordingly. The correction field update can
>   consider delays up to one second. So PTP frame needs to be
>   filtered and parsed, and written timestamp into Sync frame
>   originTimestamp field.
> 
> The operation of API dpni_set_single_step_cfg() has to be done
> when no one-step timestamping frames are in flight. So we have
> to make sure the last one-step timestamping frame has already
> been transmitted on hardware before starting to send the current
> one. The resolution is,
> 
> - Utilize skb->cb[0] to mark timestamping request per packet.
>   If it is one-step timestamping PTP sync packet, queue to skb queue.
>   If not, transmit immediately.
> 
> - Schedule a work to transmit skbs in skb queue.
> 
> - mutex lock is used to ensure the last one-step timestamping packet
>   has already been transmitted on hardware through TX confirmation
> queue
>   before transmitting current packet.
> 
> Signed-off-by: Yangbo Lu <yangbo.lu@....com>
> ---
> Changes for v2:
> 	- None.
> Changes for v3:
> 	- Fixed build issue on 32-bit.
> 	- Converted to use ptp_parse_header.
> ---
>  drivers/net/ethernet/freescale/dpaa2/dpaa2-eth.c   | 191
> +++++++++++++++++++--
>  drivers/net/ethernet/freescale/dpaa2/dpaa2-eth.h   |  32 +++-
>  .../net/ethernet/freescale/dpaa2/dpaa2-ethtool.c   |   4 +-
>  3 files changed, 211 insertions(+), 16 deletions(-)
> 
> diff --git a/drivers/net/ethernet/freescale/dpaa2/dpaa2-eth.c
> b/drivers/net/ethernet/freescale/dpaa2/dpaa2-eth.c
> index eab9470..ba570f6 100644
> --- a/drivers/net/ethernet/freescale/dpaa2/dpaa2-eth.c
> +++ b/drivers/net/ethernet/freescale/dpaa2/dpaa2-eth.c
> @@ -15,6 +15,7 @@
>  #include <linux/bpf.h>
>  #include <linux/bpf_trace.h>
>  #include <linux/fsl/ptp_qoriq.h>
> +#include <linux/ptp_classify.h>
>  #include <net/pkt_cls.h>
>  #include <net/sock.h>
>  
> @@ -563,11 +564,57 @@ static int dpaa2_eth_consume_frames(struct
> dpaa2_eth_channel *ch,
>  	return cleaned;
>  }
>  
> +static int dpaa2_eth_ptp_parse(struct sk_buff *skb,
> +			       u8 *msgtype, u8 *twostep, u8 *udp,
> +			       u16 *correction_offset,
> +			       u16 *origintimestamp_offset)
> +{
> +	unsigned int ptp_class;
> +	struct ptp_header *hdr;
> +	unsigned int type;
> +	u8 *base;
> +
> +	ptp_class = ptp_classify_raw(skb);
> +	if (ptp_class == PTP_CLASS_NONE)
> +		return -EINVAL;
> +
> +	hdr = ptp_parse_header(skb, ptp_class);
> +	if (!hdr)
> +		return -EINVAL;
> +
> +	*msgtype = ptp_get_msgtype(hdr, ptp_class);
> +	*twostep = hdr->flag_field[0] & 0x2;
> +
> +	type = ptp_class & PTP_CLASS_PMASK;
> +	if (type == PTP_CLASS_IPV4 ||
> +	    type == PTP_CLASS_IPV6)
> +		*udp = 1;
> +	else
> +		*udp = 0;
> +
> +	base = skb_mac_header(skb);
> +	*correction_offset = (u8 *)&hdr->correction - base;
> +	*origintimestamp_offset = (u8 *)hdr + sizeof(struct ptp_header)
> - base;
> +
> +	return 0;
> +}
> +
>  /* Configure the egress frame annotation for timestamp update */
> -static void dpaa2_eth_enable_tx_tstamp(struct dpaa2_fd *fd, void
> *buf_start)
> +static void dpaa2_eth_enable_tx_tstamp(struct dpaa2_eth_priv *priv,
> +				       struct dpaa2_fd *fd,
> +				       void *buf_start,
> +				       struct sk_buff *skb)
>  {
> +	struct ptp_tstamp origin_timestamp;
> +	struct dpni_single_step_cfg cfg;
> +	u8 msgtype, twostep, udp;
>  	struct dpaa2_faead *faead;
> +	struct dpaa2_fas *fas;
> +	struct timespec64 ts;
> +	u16 offset1, offset2;
>  	u32 ctrl, frc;
> +	__le64 *ns;
> +	u8 *data;
>  
>  	/* Mark the egress frame annotation area as valid */
>  	frc = dpaa2_fd_get_frc(fd);
> @@ -583,6 +630,58 @@ static void dpaa2_eth_enable_tx_tstamp(struct
> dpaa2_fd *fd, void *buf_start)
>  	ctrl = DPAA2_FAEAD_A2V | DPAA2_FAEAD_UPDV | DPAA2_FAEAD_UPD;
>  	faead = dpaa2_get_faead(buf_start, true);
>  	faead->ctrl = cpu_to_le32(ctrl);
> +
> +	if (skb->cb[0] == TX_TSTAMP_ONESTEP_SYNC) {
> +		if (dpaa2_eth_ptp_parse(skb, &msgtype, &twostep, &udp,
> +					&offset1, &offset2)) {
> +			netdev_err(priv->net_dev,
> +				   "bad packet for one-step
> timestamping\n");

don't use netdev_err in data path, have a counter or
WARN_ONCE/netdev_warn_once if you know that this shouldn't happen.

> +			return;
> +		}
> +
> +		if (msgtype != 0 || twostep) {
> +			netdev_err(priv->net_dev,
> +				   "bad packet for one-step
> timestamping\n");
> +			return;
> +		}
> +
> +		/* Mark the frame annotation status as valid */
> +		frc = dpaa2_fd_get_frc(fd);
> +		dpaa2_fd_set_frc(fd, frc | DPAA2_FD_FRC_FASV);
> +
> +		/* Mark the PTP flag for one step timestamping */
> +		fas = dpaa2_get_fas(buf_start, true);
> +		fas->status = cpu_to_le32(DPAA2_FAS_PTP);
> +
> +		/* Write current time to FA timestamp field */
> +		if (!dpaa2_ptp) {
> +			netdev_err(priv->net_dev,
> +				   "ptp driver may not loaded for one-
> step timestamping\n");

why are you even here if dpaa2_ptp not valid ? isn't it too late to
abort here ? some of the tx descriptor fields are already populated.

Just don't allow ptp and avoid marking skbs for timestamping as early
as possible if your driver is not ready for it .. 

> +			return;
> +		}
> +		dpaa2_ptp->caps.gettime64(&dpaa2_ptp->caps, &ts);
> +		ns = dpaa2_get_ts(buf_start, true);
> +		*ns = cpu_to_le64(timespec64_to_ns(&ts) /
> +				  DPAA2_PTP_CLK_PERIOD_NS);
> +
> +		/* Update current time to PTP message originTimestamp
> field */
> +		ns_to_ptp_tstamp(&origin_timestamp, le64_to_cpup(ns));
> +		data = skb_mac_header(skb);
> +		*(__be16 *)(data + offset2) =
> htons(origin_timestamp.sec_msb);
> +		*(__be32 *)(data + offset2 + 2) =
> +			htonl(origin_timestamp.sec_lsb);
> +		*(__be32 *)(data + offset2 + 6) =
> htonl(origin_timestamp.nsec);
> +
> +		cfg.en = 1;
> +		cfg.ch_update = udp;
> +		cfg.offset = offset1;
> +		cfg.peer_delay = 0;
> +
> +		if (dpni_set_single_step_cfg(priv->mc_io, 0, priv-
> >mc_token,
> +					     &cfg))
> +			netdev_err(priv->net_dev,
> +				   "dpni_set_single_step_cfg
> failed\n");

Again too much error prints on data path, without any real useful debug
information 

> +	}
>  }
>  
>  /* Create a frame descriptor based on a fragmented skb */
> @@ -820,7 +919,7 @@ static int dpaa2_eth_build_single_fd(struct
> dpaa2_eth_priv *priv,
>   * This can be called either from dpaa2_eth_tx_conf() or on the
> error path of
>   * dpaa2_eth_tx().
>   */
> -static void dpaa2_eth_free_tx_fd(const struct dpaa2_eth_priv *priv,
> +static void dpaa2_eth_free_tx_fd(struct dpaa2_eth_priv *priv,
>  				 struct dpaa2_eth_fq *fq,
>  				 const struct dpaa2_fd *fd, bool
> in_napi)
>  {
> @@ -903,6 +1002,8 @@ static void dpaa2_eth_free_tx_fd(const struct
> dpaa2_eth_priv *priv,
>  		ns = DPAA2_PTP_CLK_PERIOD_NS * le64_to_cpup(ts);
>  		shhwtstamps.hwtstamp = ns_to_ktime(ns);
>  		skb_tstamp_tx(skb, &shhwtstamps);
> +	} else if (skb->cb[0] == TX_TSTAMP_ONESTEP_SYNC) {
> +		mutex_unlock(&priv->onestep_tstamp_lock);
>  	}
>  
>  	/* Free SGT buffer allocated on tx */
> @@ -922,7 +1023,8 @@ static void dpaa2_eth_free_tx_fd(const struct
> dpaa2_eth_priv *priv,
>  	napi_consume_skb(skb, in_napi);
>  }
>  
> -static netdev_tx_t dpaa2_eth_tx(struct sk_buff *skb, struct
> net_device *net_dev)
> +static netdev_tx_t __dpaa2_eth_tx(struct sk_buff *skb,
> +				  struct net_device *net_dev)
>  {
>  	struct dpaa2_eth_priv *priv = netdev_priv(net_dev);
>  	struct dpaa2_fd fd;
> @@ -937,13 +1039,6 @@ static netdev_tx_t dpaa2_eth_tx(struct sk_buff
> *skb, struct net_device *net_dev)
>  	int err, i;
>  	void *swa;
>  
> -	/* Utilize skb->cb[0] for timestamping request per skb */
> -	skb->cb[0] = 0;
> -
> -	if (priv->tx_tstamp_type == HWTSTAMP_TX_ON &&
> -	    skb_shinfo(skb)->tx_flags & SKBTX_HW_TSTAMP)
> -		skb->cb[0] = TX_TSTAMP;
> -
>  	percpu_stats = this_cpu_ptr(priv->percpu_stats);
>  	percpu_extras = this_cpu_ptr(priv->percpu_extras);
>  
> @@ -981,8 +1076,8 @@ static netdev_tx_t dpaa2_eth_tx(struct sk_buff
> *skb, struct net_device *net_dev)
>  		goto err_build_fd;
>  	}
>  
> -	if (skb->cb[0] == TX_TSTAMP)
> -		dpaa2_eth_enable_tx_tstamp(&fd, swa);
> +	if (skb->cb[0])
> +		dpaa2_eth_enable_tx_tstamp(priv, &fd, swa, skb);
>  
>  	/* Tracing point */
>  	trace_dpaa2_tx_fd(net_dev, &fd);
> @@ -1037,6 +1132,58 @@ static netdev_tx_t dpaa2_eth_tx(struct sk_buff
> *skb, struct net_device *net_dev)
>  	return NETDEV_TX_OK;
>  }
>  
> +static void dpaa2_eth_tx_onestep_tstamp(struct work_struct *work)
> +{
> +	struct dpaa2_eth_priv *priv = container_of(work, struct
> dpaa2_eth_priv,
> +						   tx_onestep_tstamp);
> +	struct sk_buff *skb;
> +
> +	while (true) {
> +		skb = skb_dequeue(&priv->tx_skbs);
> +		if (!skb)
> +			return;
> +
> +		mutex_lock(&priv->onestep_tstamp_lock);
> +		__dpaa2_eth_tx(skb, priv->net_dev);

Very weird approach, at least document here where the lock is being
released, in addition to the comment that you have on the mutex
declaration.

> +	}
> +}
> +
> +static netdev_tx_t dpaa2_eth_tx(struct sk_buff *skb, struct
> net_device *net_dev)
> +{
> +	struct dpaa2_eth_priv *priv = netdev_priv(net_dev);
> +	u8 msgtype, twostep, udp;
> +	u16 offset1, offset2;
> +
> +	/* Utilize skb->cb[0] for timestamping request per skb */
> +	skb->cb[0] = 0;
> +
> +	if (skb_shinfo(skb)->tx_flags & SKBTX_HW_TSTAMP) {
> +		if (priv->tx_tstamp_type == HWTSTAMP_TX_ON)
> +			skb->cb[0] = TX_TSTAMP;
> +		else if (priv->tx_tstamp_type ==
> HWTSTAMP_TX_ONESTEP_SYNC)
> +			skb->cb[0] = TX_TSTAMP_ONESTEP_SYNC;
> +	}
> +
> +	/* TX for one-step timestamping PTP Sync packet */
> +	if (skb->cb[0] == TX_TSTAMP_ONESTEP_SYNC) {
> +		if (!dpaa2_eth_ptp_parse(skb, &msgtype, &twostep, &udp,
> +					 &offset1, &offset2))
> +			if (msgtype == 0 && twostep == 0) {
> +				skb_queue_tail(&priv->tx_skbs, skb);
> +				queue_work(priv->dpaa2_ptp_wq,
> +					   &priv->tx_onestep_tstamp);
> +				return NETDEV_TX_OK;
> +			}
> +		/* Use two-step timestamping if not one-step
> timestamping
> +		 * PTP Sync packet
> +		 */
> +		skb->cb[0] = TX_TSTAMP;
> +	}
> +
> +	/* TX for other packets */
> +	return __dpaa2_eth_tx(skb, net_dev);
> +}
> +
>  /* Tx confirmation frame processing routine */
>  static void dpaa2_eth_tx_conf(struct dpaa2_eth_priv *priv,
>  			      struct dpaa2_eth_channel *ch
> __always_unused,
> @@ -1906,6 +2053,7 @@ static int dpaa2_eth_ts_ioctl(struct net_device
> *dev, struct ifreq *rq, int cmd)
>  	switch (config.tx_type) {
>  	case HWTSTAMP_TX_OFF:
>  	case HWTSTAMP_TX_ON:
> +	case HWTSTAMP_TX_ONESTEP_SYNC:
>  		priv->tx_tstamp_type = config.tx_type;
>  		break;
>  	default:
> @@ -2731,8 +2879,10 @@ static int dpaa2_eth_set_buffer_layout(struct
> dpaa2_eth_priv *priv)
>  	/* tx buffer */
>  	buf_layout.private_data_size = DPAA2_ETH_SWA_SIZE;
>  	buf_layout.pass_timestamp = true;
> +	buf_layout.pass_frame_status = true;
>  	buf_layout.options = DPNI_BUF_LAYOUT_OPT_PRIVATE_DATA_SIZE |
> -			     DPNI_BUF_LAYOUT_OPT_TIMESTAMP;
> +			     DPNI_BUF_LAYOUT_OPT_TIMESTAMP |
> +			     DPNI_BUF_LAYOUT_OPT_FRAME_STATUS;
>  	err = dpni_set_buffer_layout(priv->mc_io, 0, priv->mc_token,
>  				     DPNI_QUEUE_TX, &buf_layout);
>  	if (err) {
> @@ -2741,7 +2891,8 @@ static int dpaa2_eth_set_buffer_layout(struct
> dpaa2_eth_priv *priv)
>  	}
>  
>  	/* tx-confirm buffer */
> -	buf_layout.options = DPNI_BUF_LAYOUT_OPT_TIMESTAMP;
> +	buf_layout.options = DPNI_BUF_LAYOUT_OPT_TIMESTAMP |
> +			     DPNI_BUF_LAYOUT_OPT_FRAME_STATUS;
>  	err = dpni_set_buffer_layout(priv->mc_io, 0, priv->mc_token,
>  				     DPNI_QUEUE_TX_CONFIRM,
> &buf_layout);
>  	if (err) {
> @@ -3969,6 +4120,16 @@ static int dpaa2_eth_probe(struct
> fsl_mc_device *dpni_dev)
>  	priv->tx_tstamp_type = HWTSTAMP_TX_OFF;
>  	priv->rx_tstamp = false;
>  
> +	priv->dpaa2_ptp_wq = alloc_workqueue("dpaa2_ptp_wq", 0, 0);
> +	if (!priv->dpaa2_ptp_wq) {
> +		err = -ENOMEM;
> +		goto err_wq_alloc;
> +	}
> +
> +	INIT_WORK(&priv->tx_onestep_tstamp,
> dpaa2_eth_tx_onestep_tstamp);
> +
> +	skb_queue_head_init(&priv->tx_skbs);
> +
>  	/* Obtain a MC portal */
>  	err = fsl_mc_portal_allocate(dpni_dev,
> FSL_MC_IO_ATOMIC_CONTEXT_PORTAL,
>  				     &priv->mc_io);
> @@ -4107,6 +4268,8 @@ static int dpaa2_eth_probe(struct fsl_mc_device
> *dpni_dev)
>  err_dpni_setup:
>  	fsl_mc_portal_free(priv->mc_io);
>  err_portal_alloc:
> +	destroy_workqueue(priv->dpaa2_ptp_wq);
> +err_wq_alloc:
>  	dev_set_drvdata(dev, NULL);
>  	free_netdev(net_dev);
>  
> diff --git a/drivers/net/ethernet/freescale/dpaa2/dpaa2-eth.h
> b/drivers/net/ethernet/freescale/dpaa2/dpaa2-eth.h
> index e33d79e..6436fa3 100644
> --- a/drivers/net/ethernet/freescale/dpaa2/dpaa2-eth.h
> +++ b/drivers/net/ethernet/freescale/dpaa2/dpaa2-eth.h
> @@ -195,6 +195,24 @@ struct dpaa2_faead {
>  #define DPAA2_FAEAD_EBDDV		0x00002000
>  #define DPAA2_FAEAD_UPD			0x00000010
>  
> +struct ptp_tstamp {
> +	u16 sec_msb;
> +	u32 sec_lsb;
> +	u32 nsec;
> +};
> +
> +static inline void ns_to_ptp_tstamp(struct ptp_tstamp *tstamp, u64
> ns)
> +{
> +	u64 sec, nsec;
> +
> +	sec = ns;
> +	nsec = do_div(sec, 1000000000);
> +
> +	tstamp->sec_lsb = sec & 0xFFFFFFFF;
> +	tstamp->sec_msb = (sec >> 32) & 0xFFFF;
> +	tstamp->nsec = nsec;
> +}
> +
>  /* Accessors for the hardware annotation fields that we use */
>  static inline void *dpaa2_get_hwa(void *buf_addr, bool swa)
>  {
> @@ -474,9 +492,21 @@ struct dpaa2_eth_priv {
>  #endif
>  
>  	struct dpaa2_mac *mac;
> +	struct workqueue_struct	*dpaa2_ptp_wq;
> +	struct work_struct	tx_onestep_tstamp;
> +	struct sk_buff_head	tx_skbs;
> +	/* The one-step timestamping configuration on hardware
> +	 * registers could only be done when no one-step
> +	 * timestamping frames are in flight. So we use a mutex
> +	 * lock here to make sure the lock is released by last
> +	 * one-step timestamping packet through TX confirmation
> +	 * queue before transmit current packet.
> +	 */
> +	struct mutex		onestep_tstamp_lock;
>  };
>  
>  #define TX_TSTAMP		0x1
> +#define TX_TSTAMP_ONESTEP_SYNC	0x2
>  
>  #define DPAA2_RXH_SUPPORTED	(RXH_L2DA | RXH_VLAN | RXH_L3_PROTO \
>  				| RXH_IP_SRC | RXH_IP_DST |
> RXH_L4_B_0_1 \
> @@ -581,7 +611,7 @@ static inline unsigned int
> dpaa2_eth_needed_headroom(struct sk_buff *skb)
>  		return 0;
>  
>  	/* If we have Tx timestamping, need 128B hardware annotation */
> -	if (skb->cb[0] == TX_TSTAMP)
> +	if (skb->cb[0])
>  		headroom += DPAA2_ETH_TX_HWA_SIZE;
>  
>  	return headroom;
> diff --git a/drivers/net/ethernet/freescale/dpaa2/dpaa2-ethtool.c
> b/drivers/net/ethernet/freescale/dpaa2/dpaa2-ethtool.c
> index 26bd99b..bf3baf6 100644
> --- a/drivers/net/ethernet/freescale/dpaa2/dpaa2-ethtool.c
> +++ b/drivers/net/ethernet/freescale/dpaa2/dpaa2-ethtool.c
> @@ -1,6 +1,7 @@
>  // SPDX-License-Identifier: (GPL-2.0+ OR BSD-3-Clause)
>  /* Copyright 2014-2016 Freescale Semiconductor Inc.
>   * Copyright 2016 NXP
> + * Copyright 2020 NXP
>   */
>  
>  #include <linux/net_tstamp.h>
> @@ -770,7 +771,8 @@ static int dpaa2_eth_get_ts_info(struct
> net_device *dev,
>  	info->phc_index = dpaa2_phc_index;
>  
>  	info->tx_types = (1 << HWTSTAMP_TX_OFF) |
> -			 (1 << HWTSTAMP_TX_ON);
> +			 (1 << HWTSTAMP_TX_ON) |
> +			 (1 << HWTSTAMP_TX_ONESTEP_SYNC);
>  
>  	info->rx_filters = (1 << HWTSTAMP_FILTER_NONE) |
>  			   (1 << HWTSTAMP_FILTER_ALL);

Powered by blists - more mailing lists