Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
Date:   Wed, 3 Nov 2021 15:22:29 +0100
From:   =?UTF-8?B?Q2zDqW1lbnQgTMOpZ2Vy?= <clement.leger@bootlin.com>
To:     Vladimir Oltean <vladimir.oltean@nxp.com>
Cc:     "David S. Miller" <davem@davemloft.net>,
        Jakub Kicinski <kuba@kernel.org>,
        Rob Herring <robh+dt@kernel.org>,
        Claudiu Manoil <claudiu.manoil@nxp.com>,
        Alexandre Belloni <alexandre.belloni@bootlin.com>,
        "UNGLinuxDriver@microchip.com" <UNGLinuxDriver@microchip.com>,
        Andrew Lunn <andrew@lunn.ch>,
        "netdev@vger.kernel.org" <netdev@vger.kernel.org>,
        "devicetree@vger.kernel.org" <devicetree@vger.kernel.org>,
        "linux-kernel@vger.kernel.org" <linux-kernel@vger.kernel.org>,
        Thomas Petazzoni <thomas.petazzoni@bootlin.com>
Subject: Re: [PATCH v2 5/6] net: ocelot: add FDMA support
Message-ID: <20211103152229.6066e053@fixe.home>
In-Reply-To: <20211103123153.li3rvt4753rm5w66@skbuf>
References: <20211103091943.3878621-1-clement.leger@bootlin.com>
        <20211103091943.3878621-6-clement.leger@bootlin.com>
        <20211103123153.li3rvt4753rm5w66@skbuf>
Organization: Bootlin
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8BIT
Precedence: bulk

Le Wed, 3 Nov 2021 12:31:54 +0000,
Vladimir Oltean <vladimir.oltean@nxp.com> a écrit :

> On Wed, Nov 03, 2021 at 10:19:42AM +0100, Clément Léger wrote:
> > Ethernet frames can be extracted or injected autonomously to or
> > from the device’s DDR3/DDR3L memory and/or PCIe memory space.
> > Linked list data structures in memory are used for injecting or
> > extracting Ethernet frames. The FDMA generates interrupts when
> > frame extraction or injection is done and when the linked lists
> > need updating.
> > 
> > The FDMA is shared between all the ethernet ports of the switch and
> > uses a linked list of descriptors (DCB) to inject and extract
> > packets. Before adding descriptors, the FDMA channels must be
> > stopped. It would be inefficient to do that each time a descriptor
> > would be added,
> > 
> > TX path uses multiple lists to handle descriptors. tx_ongoing is
> > the list of DCB currently owned by the hardware, tx_queued is a
> > list of DCB that will be given to the hardware when tx_ongoing is
> > done and finally tx_free_dcb is the list of DCB available for TX.
> > 
> > RX path uses two list, rx_hw is the list of DCB currently given to
> > the hardware and rx_sw is the list of descriptors that have been
> > completed by the FDMA and will be reinjected when the DMA hits the
> > end of the linked list.
> > 
> > Co-developed-by: Alexandre Belloni <alexandre.belloni@bootlin.com>
> > Signed-off-by: Alexandre Belloni <alexandre.belloni@bootlin.com>
> > Signed-off-by: Clément Léger <clement.leger@bootlin.com>
> > ---  
> 
> Honestly, my mind exploded when I saw locking between TX and TX
> confirmation. Can you not constrain the list of TX DCBs to act like a
> ring-based device? Meaning that the linked list is always constant in
> size, and you just update the Linked List Pointer of the last entry
> populated by software to be NULL, to make the hardware stop processing
> beyond that point. This could help you avoid keeping a list in
> software, and a DMA pool for the DCBs, just have a contiguous memory
> mapping of all the DCBs for TX, and then you shouldn't need a
> spin_lock for a list you no longer keep.

Indeed, this will simplify DCB handling. I will do these modifications.

> 
> I haven't even gotten to reviewing RX properly...
> 
> >  drivers/net/ethernet/mscc/Makefile         |   1 +
> >  drivers/net/ethernet/mscc/ocelot.h         |   1 +
> >  drivers/net/ethernet/mscc/ocelot_fdma.c    | 693
> > +++++++++++++++++++++ drivers/net/ethernet/mscc/ocelot_fdma.h    |
> > 59 ++ drivers/net/ethernet/mscc/ocelot_net.c     |  11 +-
> >  drivers/net/ethernet/mscc/ocelot_vsc7514.c |  15 +
> >  include/soc/mscc/ocelot.h                  |   2 +
> >  7 files changed, 779 insertions(+), 3 deletions(-)
> >  create mode 100644 drivers/net/ethernet/mscc/ocelot_fdma.c
> >  create mode 100644 drivers/net/ethernet/mscc/ocelot_fdma.h
> > 
> > diff --git a/drivers/net/ethernet/mscc/Makefile
> > b/drivers/net/ethernet/mscc/Makefile index
> > 722c27694b21..d76a9b78b6ca 100644 ---
> > a/drivers/net/ethernet/mscc/Makefile +++
> > b/drivers/net/ethernet/mscc/Makefile @@ -11,5 +11,6 @@
> > mscc_ocelot_switch_lib-y := \
> > mscc_ocelot_switch_lib-$(CONFIG_BRIDGE_MRP) += ocelot_mrp.o
> > obj-$(CONFIG_MSCC_OCELOT_SWITCH) += mscc_ocelot.o mscc_ocelot-y := \
> > +	ocelot_fdma.o \
> >  	ocelot_vsc7514.o \
> >  	ocelot_net.o
> > diff --git a/drivers/net/ethernet/mscc/ocelot.h
> > b/drivers/net/ethernet/mscc/ocelot.h index
> > ba0dec7dd64f..ad85ad1079ad 100644 ---
> > a/drivers/net/ethernet/mscc/ocelot.h +++
> > b/drivers/net/ethernet/mscc/ocelot.h @@ -9,6 +9,7 @@
> >  #define _MSCC_OCELOT_H_
> >  
> >  #include <linux/bitops.h>
> > +#include <linux/dsa/ocelot.h>
> >  #include <linux/etherdevice.h>
> >  #include <linux/if_vlan.h>
> >  #include <linux/net_tstamp.h>
> > diff --git a/drivers/net/ethernet/mscc/ocelot_fdma.c
> > b/drivers/net/ethernet/mscc/ocelot_fdma.c new file mode 100644
> > index 000000000000..d8cdf022bbee
> > --- /dev/null
> > +++ b/drivers/net/ethernet/mscc/ocelot_fdma.c
> > @@ -0,0 +1,693 @@
> > +// SPDX-License-Identifier: (GPL-2.0 OR MIT)
> > +/*
> > + * Microsemi SoCs FDMA driver
> > + *
> > + * Copyright (c) 2021 Microchip
> > + */
> > +
> > +#include <linux/bitops.h>
> > +#include <linux/dmapool.h>
> > +#include <linux/dsa/ocelot.h>
> > +#include <linux/netdevice.h>
> > +#include <linux/of_platform.h>
> > +#include <linux/skbuff.h>
> > +
> > +#include "ocelot_fdma.h"
> > +#include "ocelot_qs.h"
> > +
> > +#define MSCC_FDMA_DCB_LLP(x)			((x) * 4 + 0x0)
> > +
> > +#define MSCC_FDMA_DCB_STAT_BLOCKO(x)		(((x) << 20) &
> > GENMASK(31, 20)) +#define MSCC_FDMA_DCB_STAT_BLOCKO_M
> > 	GENMASK(31, 20) +#define
> > MSCC_FDMA_DCB_STAT_BLOCKO_X(x)		(((x) & GENMASK(31,
> > 20)) >> 20) +#define MSCC_FDMA_DCB_STAT_PD
> > BIT(19) +#define MSCC_FDMA_DCB_STAT_ABORT		BIT(18)
> > +#define MSCC_FDMA_DCB_STAT_EOF			BIT(17)
> > +#define MSCC_FDMA_DCB_STAT_SOF			BIT(16)
> > +#define MSCC_FDMA_DCB_STAT_BLOCKL_M		GENMASK(15, 0)
> > +#define MSCC_FDMA_DCB_STAT_BLOCKL(x)		((x) &
> > GENMASK(15, 0)) + +#define MSCC_FDMA_CH_SAFE
> > 0xcc +
> > +#define MSCC_FDMA_CH_ACTIVATE			0xd0
> > +
> > +#define MSCC_FDMA_CH_DISABLE			0xd4
> > +
> > +#define MSCC_FDMA_EVT_ERR			0x164
> > +
> > +#define MSCC_FDMA_EVT_ERR_CODE			0x168
> > +
> > +#define MSCC_FDMA_INTR_LLP			0x16c
> > +
> > +#define MSCC_FDMA_INTR_LLP_ENA			0x170
> > +
> > +#define MSCC_FDMA_INTR_FRM			0x174
> > +
> > +#define MSCC_FDMA_INTR_FRM_ENA			0x178
> > +
> > +#define MSCC_FDMA_INTR_ENA			0x184
> > +
> > +#define MSCC_FDMA_INTR_IDENT			0x188
> > +
> > +#define MSCC_FDMA_INJ_CHAN			2
> > +#define MSCC_FDMA_XTR_CHAN			0
> > +
> > +#define FDMA_MAX_SKB				256
> > +#define FDMA_WEIGHT				32
> > +
> > +#define OCELOT_TAG_WORD_LEN			(OCELOT_TAG_LEN
> > / 4) +
> > +/* Add 4 for possible misalignment when mapping the data */
> > +#define FDMA_RX_EXTRA_SIZE			\
> > +	(OCELOT_TAG_LEN + ETH_FCS_LEN + ETH_HLEN + 4)
> > +
> > +struct ocelot_fdma_dcb_hw_v2 {
> > +	u32 llp;
> > +	u32 datap;
> > +	u32 datal;
> > +	u32 stat;
> > +};
> > +
> > +struct ocelot_fdma_dcb {
> > +	struct ocelot_fdma_dcb_hw_v2	hw;
> > +	struct list_head		node;
> > +	struct sk_buff			*skb;
> > +	dma_addr_t			mapping;
> > +	size_t				mapped_size;
> > +	dma_addr_t			phys;
> > +};
> > +
> > +static int fdma_rx_compute_buffer_size(int mtu)
> > +{
> > +	return ALIGN(mtu + FDMA_RX_EXTRA_SIZE, 4);
> > +}
> > +
> > +static void fdma_writel(struct ocelot_fdma *fdma, u32 reg, u32
> > data) +{
> > +	writel(data, fdma->base + reg);
> > +}
> > +
> > +static u32 fdma_readl(struct ocelot_fdma *fdma, u32 reg)
> > +{
> > +	return readl(fdma->base + reg);
> > +}
> > +
> > +static void fdma_activate_chan(struct ocelot_fdma *fdma,
> > +			       struct ocelot_fdma_dcb *dcb, int
> > chan) +{
> > +	fdma_writel(fdma, MSCC_FDMA_DCB_LLP(chan), dcb->phys);
> > +	fdma_writel(fdma, MSCC_FDMA_CH_ACTIVATE, BIT(chan));
> > +}
> > +
> > +static void fdma_stop_channel(struct ocelot_fdma *fdma, int chan)
> > +{
> > +	u32 safe;
> > +
> > +	fdma_writel(fdma, MSCC_FDMA_CH_DISABLE, BIT(chan));
> > +	do {
> > +		safe = fdma_readl(fdma, MSCC_FDMA_CH_SAFE);
> > +	} while (!(safe & BIT(chan)));
> > +}
> > +
> > +static bool ocelot_fdma_dcb_set_data(struct ocelot_fdma *fdma,
> > +				     struct ocelot_fdma_dcb *dcb,
> > void *data,
> > +				     size_t size, enum
> > dma_data_direction dir) +{
> > +	u32 offset;
> > +
> > +	dcb->mapped_size = size;
> > +	dcb->mapping = dma_map_single(fdma->dev, data, size, dir);
> > +	if (unlikely(dma_mapping_error(fdma->dev, dcb->mapping)))
> > +		return false;
> > +
> > +	offset = dcb->mapping & 0x3;
> > +
> > +	dcb->hw.llp = 0;
> > +	dcb->hw.datap = dcb->mapping & ~0x3;
> > +	/* DATAL must be a multiple of word size */
> > +	dcb->hw.datal = ALIGN_DOWN(size - offset, 4);
> > +	dcb->hw.stat = MSCC_FDMA_DCB_STAT_BLOCKO(offset);
> > +
> > +	return true;
> > +}
> > +
> > +static bool ocelot_fdma_dcb_set_rx_skb(struct ocelot_fdma *fdma,
> > +				       struct ocelot_fdma_dcb *dcb,
> > +				       struct sk_buff *skb, size_t
> > size) +{
> > +	dcb->skb = skb;
> > +	return ocelot_fdma_dcb_set_data(fdma, dcb, skb->data, size,
> > +				      DMA_FROM_DEVICE);
> > +}
> > +
> > +static bool ocelot_fdma_dcb_set_tx_skb(struct ocelot_fdma *fdma,
> > +				       struct ocelot_fdma_dcb *dcb,
> > +				       struct sk_buff *skb)
> > +{
> > +	if (!ocelot_fdma_dcb_set_data(fdma, dcb, skb->data,
> > skb->len,
> > +				      DMA_TO_DEVICE))
> > +		return false;
> > +
> > +	dcb->skb = skb;
> > +	dcb->hw.stat |= MSCC_FDMA_DCB_STAT_BLOCKL(skb->len);
> > +	dcb->hw.stat |= MSCC_FDMA_DCB_STAT_SOF |
> > MSCC_FDMA_DCB_STAT_EOF; +
> > +	return true;
> > +}
> > +
> > +static struct ocelot_fdma_dcb *fdma_dcb_alloc(struct ocelot_fdma
> > *fdma) +{
> > +	struct ocelot_fdma_dcb *dcb;
> > +	dma_addr_t phys;
> > +
> > +	dcb = dma_pool_zalloc(fdma->dcb_pool, GFP_KERNEL, &phys);
> > +	if (unlikely(!dcb))
> > +		return NULL;
> > +
> > +	dcb->phys = phys;
> > +
> > +	return dcb;
> > +}
> > +
> > +static struct net_device *fdma_get_port_netdev(struct ocelot_fdma
> > *fdma,
> > +					       int port_num)
> > +{
> > +	struct ocelot_port_private *port_priv;
> > +	struct ocelot *ocelot = fdma->ocelot;
> > +	struct ocelot_port *port;
> > +
> > +	if (port_num >= ocelot->num_phys_ports)
> > +		return NULL;
> > +
> > +	port = ocelot->ports[port_num];
> > +
> > +	if (!port)
> > +		return NULL;
> > +
> > +	port_priv = container_of(port, struct ocelot_port_private,
> > port); +
> > +	return port_priv->dev;
> > +}
> > +
> > +static bool ocelot_fdma_rx_process_skb(struct ocelot_fdma *fdma,
> > +				       struct ocelot_fdma_dcb *dcb,
> > +				       int budget)
> > +{
> > +	struct sk_buff *skb = dcb->skb;
> > +	struct net_device *ndev;
> > +	u64 src_port;
> > +	void *xfh;
> > +
> > +	dma_unmap_single(fdma->dev, dcb->mapping, dcb->mapped_size,
> > +			 DMA_FROM_DEVICE);
> > +
> > +	xfh = skb->data;
> > +	ocelot_xfh_get_src_port(xfh, &src_port);
> > +
> > +	skb_put(skb, MSCC_FDMA_DCB_STAT_BLOCKL(dcb->hw.stat));
> > +	skb_pull(skb, OCELOT_TAG_LEN);
> > +
> > +	ndev = fdma_get_port_netdev(fdma, src_port);
> > +	if (unlikely(!ndev)) {
> > +		napi_consume_skb(dcb->skb, budget);
> > +		return false;
> > +	}
> > +
> > +	skb->dev = ndev;
> > +	skb->protocol = eth_type_trans(skb, skb->dev);
> > +	skb->dev->stats.rx_bytes += skb->len;
> > +	skb->dev->stats.rx_packets++;
> > +
> > +	netif_receive_skb(skb);
> > +
> > +	return true;
> > +}
> > +
> > +static void ocelot_fdma_rx_refill(struct ocelot_fdma *fdma)
> > +{
> > +	struct ocelot_fdma_dcb *dcb, *last_dcb;
> > +
> > +	WARN_ON(list_empty(&fdma->rx_sw));
> > +
> > +	dcb = list_first_entry(&fdma->rx_sw, struct
> > ocelot_fdma_dcb, node);
> > +	/* Splice old hardware DCB list + new one */
> > +	if (!list_empty(&fdma->rx_hw)) {
> > +		last_dcb = list_last_entry(&fdma->rx_hw, struct
> > ocelot_fdma_dcb,
> > +					   node);
> > +		last_dcb->hw.llp = dcb->phys;
> > +	}
> > +
> > +	/* Move software list to hardware list */
> > +	list_splice_tail_init(&fdma->rx_sw, &fdma->rx_hw);
> > +
> > +	/* Finally reactivate the channel */
> > +	fdma_activate_chan(fdma, dcb, MSCC_FDMA_XTR_CHAN);
> > +}
> > +
> > +static void ocelot_fdma_list_add_dcb(struct list_head *list,
> > +				     struct ocelot_fdma_dcb *dcb)
> > +{
> > +	struct ocelot_fdma_dcb *last_dcb;
> > +
> > +	if (!list_empty(list)) {
> > +		last_dcb = list_last_entry(list, struct
> > ocelot_fdma_dcb, node);
> > +		last_dcb->hw.llp = dcb->phys;
> > +	}
> > +
> > +	list_add_tail(&dcb->node, list);
> > +}
> > +
> > +static bool ocelot_fdma_rx_add_dcb_sw(struct ocelot_fdma *fdma,
> > +				      struct ocelot_fdma_dcb *dcb)
> > +{
> > +	struct sk_buff *new_skb;
> > +
> > +	/* Add DCB to end of list with new SKB */
> > +	new_skb = napi_alloc_skb(&fdma->napi, fdma->rx_buf_size);
> > +	if (unlikely(!new_skb)) {
> > +		pr_err("skb_alloc failed\n");
> > +		return false;
> > +	}
> > +
> > +	ocelot_fdma_dcb_set_rx_skb(fdma, dcb, new_skb,
> > fdma->rx_buf_size);
> > +	ocelot_fdma_list_add_dcb(&fdma->rx_sw, dcb);
> > +
> > +	return true;
> > +}
> > +
> > +static bool ocelot_fdma_rx_get(struct ocelot_fdma *fdma, int
> > budget) +{
> > +	struct ocelot_fdma_dcb *dcb;
> > +	bool valid = true;
> > +	u32 stat;
> > +
> > +	dcb = list_first_entry_or_null(&fdma->rx_hw, struct
> > ocelot_fdma_dcb,
> > +				       node);
> > +	if (!dcb || MSCC_FDMA_DCB_STAT_BLOCKL(dcb->hw.stat) == 0)
> > +		return false;
> > +
> > +	list_del(&dcb->node);
> > +
> > +	stat = dcb->hw.stat;
> > +	if (stat & MSCC_FDMA_DCB_STAT_ABORT || stat &
> > MSCC_FDMA_DCB_STAT_PD)
> > +		valid = false;
> > +
> > +	if (!(stat & MSCC_FDMA_DCB_STAT_SOF) ||
> > +	    !(stat & MSCC_FDMA_DCB_STAT_EOF))
> > +		valid = false;
> > +
> > +	if (likely(valid)) {
> > +		if (!ocelot_fdma_rx_process_skb(fdma, dcb, budget))
> > +			pr_err("Process skb failed, stat %x\n",
> > stat);
> > +	} else {
> > +		napi_consume_skb(dcb->skb, budget);
> > +	}
> > +
> > +	return ocelot_fdma_rx_add_dcb_sw(fdma, dcb);
> > +}
> > +
> > +static void ocelot_fdma_rx_check_stopped(struct ocelot_fdma *fdma)
> > +{
> > +	u32 llp = fdma_readl(fdma,
> > MSCC_FDMA_DCB_LLP(MSCC_FDMA_XTR_CHAN));
> > +	/* LLP is non NULL, FDMA is still fetching packets */
> > +	if (llp)
> > +		return;
> > +
> > +	fdma_stop_channel(fdma, MSCC_FDMA_XTR_CHAN);
> > +	ocelot_fdma_rx_refill(fdma);
> > +}
> > +
> > +static void ocelot_fdma_tx_free_dcb(struct ocelot_fdma *fdma,
> > +				    struct list_head *list)
> > +{
> > +	struct ocelot_fdma_dcb *dcb;
> > +
> > +	if (list_empty(list))
> > +		return;
> > +
> > +	/* Free all SKBs that have been used for TX */
> > +	list_for_each_entry(dcb, list, node) {
> > +		dma_unmap_single(fdma->dev, dcb->mapping,
> > dcb->mapped_size,
> > +				 DMA_TO_DEVICE);
> > +		dev_consume_skb_any(dcb->skb);
> > +		dcb->skb = NULL;
> > +	}
> > +
> > +	/* All DCBs can now be given to free list */
> > +	spin_lock(&fdma->tx_free_lock);
> > +	list_splice_tail_init(list, &fdma->tx_free_dcb);
> > +	spin_unlock(&fdma->tx_free_lock);
> > +}
> > +
> > +static void ocelot_fdma_tx_cleanup(struct ocelot_fdma *fdma)
> > +{
> > +	struct list_head tx_done = LIST_HEAD_INIT(tx_done);
> > +	struct ocelot_fdma_dcb *dcb, *temp;
> > +
> > +	spin_lock(&fdma->tx_enqueue_lock);
> > +	if (list_empty(&fdma->tx_ongoing))
> > +		goto out_unlock;
> > +
> > +	list_for_each_entry_safe(dcb, temp, &fdma->tx_ongoing,
> > node) {
> > +		if (!(dcb->hw.stat & MSCC_FDMA_DCB_STAT_PD))
> > +			break;
> > +
> > +		list_move_tail(&dcb->node, &tx_done);
> > +	}
> > +
> > +out_unlock:
> > +	spin_unlock(&fdma->tx_enqueue_lock);
> > +
> > +	ocelot_fdma_tx_free_dcb(fdma, &tx_done);
> > +}
> > +
> > +static void ocelot_fdma_tx_restart(struct ocelot_fdma *fdma)
> > +{
> > +	struct ocelot_fdma_dcb *dcb;
> > +	u32 safe;
> > +
> > +	spin_lock(&fdma->tx_enqueue_lock);
> > +
> > +	if (!list_empty(&fdma->tx_ongoing) ||
> > list_empty(&fdma->tx_queued))
> > +		goto out_unlock;
> > +
> > +	/* Ongoing list is empty, channel should be in safe mode */
> > +	do {
> > +		safe = fdma_readl(fdma, MSCC_FDMA_CH_SAFE);
> > +	} while (!(safe & BIT(MSCC_FDMA_INJ_CHAN)));
> > +
> > +	/* Move queued DCB to ongoing and restart the DMA */
> > +	list_splice_tail_init(&fdma->tx_queued, &fdma->tx_ongoing);
> > +	/* List can't be empty, no need to check */
> > +	dcb = list_first_entry(&fdma->tx_ongoing, struct
> > ocelot_fdma_dcb, node); +
> > +	fdma_activate_chan(fdma, dcb, MSCC_FDMA_INJ_CHAN);
> > +
> > +out_unlock:
> > +	spin_unlock(&fdma->tx_enqueue_lock);
> > +}
> > +
> > +static int ocelot_fdma_napi_poll(struct napi_struct *napi, int
> > budget) +{
> > +	struct ocelot_fdma *fdma = container_of(napi, struct
> > ocelot_fdma, napi);
> > +	int work_done = 0;
> > +
> > +	ocelot_fdma_tx_cleanup(fdma);
> > +	ocelot_fdma_tx_restart(fdma);
> > +
> > +	while (work_done < budget) {
> > +		if (!ocelot_fdma_rx_get(fdma, budget))
> > +			break;
> > +
> > +		work_done++;
> > +	}
> > +
> > +	ocelot_fdma_rx_check_stopped(fdma);
> > +
> > +	if (work_done < budget) {
> > +		napi_complete(&fdma->napi);  
> 
> Documentation says you should consider calling
> napi_complete_done(&fdma->napi, work_done);

Acked.

> 
> > +		fdma_writel(fdma, MSCC_FDMA_INTR_ENA,
> > +			    BIT(MSCC_FDMA_INJ_CHAN) |
> > BIT(MSCC_FDMA_XTR_CHAN));
> > +	}
> > +
> > +	return work_done;
> > +}
> > +
> > +static irqreturn_t ocelot_fdma_interrupt(int irq, void *dev_id)
> > +{
> > +	u32 ident, llp, frm, err, err_code;
> > +	struct ocelot_fdma *fdma = dev_id;
> > +
> > +	ident = fdma_readl(fdma, MSCC_FDMA_INTR_IDENT);
> > +	frm = fdma_readl(fdma, MSCC_FDMA_INTR_FRM);
> > +	llp = fdma_readl(fdma, MSCC_FDMA_INTR_LLP);
> > +
> > +	fdma_writel(fdma, MSCC_FDMA_INTR_LLP, llp & ident);
> > +	fdma_writel(fdma, MSCC_FDMA_INTR_FRM, frm & ident);
> > +	if (frm | llp) {  
> 
> Bitwise OR? Strange.

I will use a logic OR even if both works here.

> 
> > +		fdma_writel(fdma, MSCC_FDMA_INTR_ENA, 0);
> > +		napi_schedule(&fdma->napi);
> > +	}
> > +
> > +	err = fdma_readl(fdma, MSCC_FDMA_EVT_ERR);
> > +	if (unlikely(err)) {
> > +		err_code = fdma_readl(fdma,
> > MSCC_FDMA_EVT_ERR_CODE);
> > +		dev_err_ratelimited(fdma->dev,
> > +				    "Error ! chans mask: %#x,
> > code: %#x\n",
> > +				    err, err_code);
> > +
> > +		fdma_writel(fdma, MSCC_FDMA_EVT_ERR, err);
> > +		fdma_writel(fdma, MSCC_FDMA_EVT_ERR_CODE,
> > err_code);
> > +	}
> > +
> > +	return IRQ_HANDLED;
> > +}
> > +
> > +static struct ocelot_fdma_dcb *fdma_tx_get_dcb(struct ocelot_fdma
> > *fdma)  
> 
> Please name these functions consistently and make them start with
> ocelot_.

Acked

> 
> > +{
> > +	struct ocelot_fdma_dcb *dcb = NULL;
> > +
> > +	spin_lock_bh(&fdma->tx_free_lock);
> > +	dcb = list_first_entry_or_null(&fdma->tx_free_dcb,
> > +				       struct ocelot_fdma_dcb,
> > node);
> > +	if (dcb)
> > +		list_del(&dcb->node);
> > +
> > +	spin_unlock_bh(&fdma->tx_free_lock);
> > +
> > +	return dcb;
> > +}
> > +
> > +int ocelot_fdma_inject_frame(struct ocelot_fdma *fdma, int port,
> > u32 rew_op,
> > +			     struct sk_buff *skb, struct
> > net_device *dev) +{
> > +	struct ocelot_port *port_s = fdma->ocelot->ports[port];
> > +	struct ocelot_fdma_dcb *dcb;
> > +	struct sk_buff *new_skb;
> > +	void *ifh;
> > +
> > +	if (unlikely(skb_shinfo(skb)->nr_frags != 0)) {
> > +		netdev_err(dev, "Unsupported fragmented packet");
> > +		dev_kfree_skb_any(skb);  
> 
> skb_linearize()

I can use multiple TX DCBs to send that packet but I wanted to do so in
another series.

> 
> Also please don't print stuff from the hot path without
> net_ratelimited()

Ok.

> 
> > +		return NETDEV_TX_OK;
> > +	}
> > +
> > +	if (skb_headroom(skb) < OCELOT_TAG_LEN ||
> > +	    skb_tailroom(skb) < ETH_FCS_LEN) {  
> 
> Don't you also need to copy the skb (to ensure it's writable) if it's
> cloned (like would be the case for packets with two-step PTP TX
> timestamping)? I don't see any calls to skb_unshare().
> 
> You can test with:
> 
> ptp4l -i swp0 -2 -P -m --tx_timestamp_timeout 20
> 
> on two back-to-back boards.
> 
> > +		new_skb = skb_copy_expand(skb, OCELOT_TAG_LEN,
> > ETH_FCS_LEN,
> > +					  GFP_ATOMIC);
> > +		dev_consume_skb_any(skb);  
> 
> I think you can use pskb_expand_head() and avoid creating a new_skb.
> Look at dsa_realloc_skb().

Ok thanks.

> 
> > +		if (!new_skb)
> > +			return NETDEV_TX_OK;
> > +
> > +		skb = new_skb;
> > +	}
> > +
> > +	ifh = skb_push(skb, OCELOT_TAG_LEN);
> > +	skb_put(skb, ETH_FCS_LEN);
> > +	ocelot_ifh_port_set(ifh, port_s, rew_op,
> > skb_vlan_tag_get(skb)); +
> > +	dcb = fdma_tx_get_dcb(fdma);
> > +	if (unlikely(!dcb))
> > +		return NETDEV_TX_BUSY;
> > +
> > +	if (!ocelot_fdma_dcb_set_tx_skb(fdma, dcb, skb)) {
> > +		dev_kfree_skb_any(skb);
> > +		spin_lock_bh(&fdma->tx_free_lock);
> > +		list_add_tail(&dcb->node, &fdma->tx_free_dcb);
> > +		spin_unlock_bh(&fdma->tx_free_lock);
> > +		return NETDEV_TX_OK;
> > +	}
> > +
> > +	spin_lock_bh(&fdma->tx_enqueue_lock);
> > +
> > +	if (list_empty(&fdma->tx_ongoing)) {
> > +		ocelot_fdma_list_add_dcb(&fdma->tx_ongoing, dcb);
> > +		fdma_activate_chan(fdma, dcb, MSCC_FDMA_INJ_CHAN);
> > +	} else {
> > +		ocelot_fdma_list_add_dcb(&fdma->tx_queued, dcb);
> > +	}
> > +
> > +	spin_unlock_bh(&fdma->tx_enqueue_lock);  
> 
> I think you don't need _bh locking from ndo_start_xmit() context.
> __dev_queue_xmit() calls rcu_read_lock_bh(). On the other hand, I
> think you might need to use spin_lock_bh from
> ocelot_fdma_napi_poll(), since that runs from NET_RX softirq, and
> ndo_start_xmit() can run from loads of other contexts.

Ok, I'll rework that.
> 
> > +	return NETDEV_TX_OK;
> > +}
> > +
> > +static void fdma_free_skbs_list(struct ocelot_fdma *fdma,
> > +				struct list_head *list,
> > +				enum dma_data_direction dir)
> > +{
> > +	struct ocelot_fdma_dcb *dcb;
> > +
> > +	if (list_empty(list))
> > +		return;  
> 
> I'm not sure this is really needed.
> 
> > +
> > +	list_for_each_entry(dcb, list, node) {
> > +		if (dcb->skb) {
> > +			dma_unmap_single(fdma->dev, dcb->mapping,
> > +					 dcb->mapped_size, dir);
> > +			dev_kfree_skb_any(dcb->skb);  
> 
> dcb->skb = NULL?
> 
> > +		}
> > +	}
> > +}
> > +
> > +static int fdma_init_tx(struct ocelot_fdma *fdma)
> > +{
> > +	int i;
> > +	struct ocelot_fdma_dcb *dcb;
> > +
> > +	for (i = 0; i < FDMA_MAX_SKB; i++) {
> > +		dcb = fdma_dcb_alloc(fdma);
> > +		if (!dcb)
> > +			return -ENOMEM;
> > +
> > +		list_add_tail(&dcb->node, &fdma->tx_free_dcb);
> > +	}
> > +
> > +	return 0;
> > +}
> > +
> > +static int fdma_init_rx(struct ocelot_fdma *fdma)
> > +{
> > +	struct ocelot_port_private *port_priv;
> > +	struct ocelot *ocelot = fdma->ocelot;
> > +	struct ocelot_fdma_dcb *dcb;
> > +	struct ocelot_port *port;
> > +	struct net_device *ndev;
> > +	int max_mtu = 0;
> > +	int i;
> > +	u8 port_num;  
> 
> Please declare variables in the order of descending line length (aka
> "reverse Christmas tree"). Here, and in fdma_init_rx(), and in other
> places.


Acked.

> 
> > +
> > +	for (port_num = 0; port_num < ocelot->num_phys_ports;
> > port_num++) {  
> 
> The naming convention is "int port", "struct ocelot_port
> *ocelot_port". Please keep it. Thanks.

Ok.

> 
> > +		port = ocelot->ports[port_num];
> > +		if (!port)
> > +			continue;
> > +
> > +		port_priv = container_of(port, struct
> > ocelot_port_private,
> > +					 port);
> > +		ndev = port_priv->dev;
> > +
> > +		ndev->needed_headroom = OCELOT_TAG_LEN;
> > +		ndev->needed_tailroom = ETH_FCS_LEN;
> > +
> > +		if (READ_ONCE(ndev->mtu) > max_mtu)
> > +			max_mtu = READ_ONCE(ndev->mtu);  
> 
> This seems silly, you use READ_ONCE twice... what's the point?
> Also, what is this racing with?
> 
> > +	}
> > +
> > +	if (!ndev)
> > +		return -ENODEV;
> > +
> > +	fdma->rx_buf_size = fdma_rx_compute_buffer_size(max_mtu);
> > +	netif_napi_add(ndev, &fdma->napi, ocelot_fdma_napi_poll,
> > +		       FDMA_WEIGHT);
> > +
> > +	for (i = 0; i < FDMA_MAX_SKB; i++) {
> > +		dcb = fdma_dcb_alloc(fdma);
> > +		if (!dcb)
> > +			return -ENOMEM;
> > +
> > +		ocelot_fdma_rx_add_dcb_sw(fdma, dcb);
> > +	}
> > +
> > +	napi_enable(&fdma->napi);
> > +
> > +	return 0;
> > +}
> > +
> > +struct ocelot_fdma *ocelot_fdma_init(struct platform_device *pdev,
> > +				     struct ocelot *ocelot)
> > +{
> > +	struct ocelot_fdma *fdma;
> > +	int ret;
> > +
> > +	fdma = devm_kzalloc(&pdev->dev, sizeof(*fdma), GFP_KERNEL);
> > +	if (!fdma)
> > +		return ERR_PTR(-ENOMEM);
> > +
> > +	fdma->ocelot = ocelot;
> > +	fdma->base = devm_platform_ioremap_resource_byname(pdev,
> > "fdma");  
> 
> Don't you want to look up the resource by name before allocating
> stuff? Maybe the allocation won't be needed, then you'll have to live
> with it, since you use devres. Although my personal recommendation
> would be to just not use devres, it makes you think more.

Ok, I will see if it's simpler without all devm_ variants.

> 
> > +	if (IS_ERR_OR_NULL(fdma->base))
> > +		return fdma->base;  
> 
> Just return NULL and simplify the caller, you aren't using the ERR
> value anyway (or do something with the error value at the call site).

Acked, error path will be reworked.

> 
> > +
> > +	fdma->dev = &pdev->dev;
> > +	fdma->dev->coherent_dma_mask = DMA_BIT_MASK(32);
> > +
> > +	spin_lock_init(&fdma->tx_enqueue_lock);
> > +	spin_lock_init(&fdma->tx_free_lock);
> > +
> > +	fdma_writel(fdma, MSCC_FDMA_INTR_ENA, 0);
> > +
> > +	fdma->irq = platform_get_irq_byname(pdev, "fdma");
> > +	ret = devm_request_irq(&pdev->dev, fdma->irq,
> > ocelot_fdma_interrupt, 0,
> > +			       dev_name(&pdev->dev), fdma);
> > +	if (ret)
> > +		return ERR_PTR(ret);
> > +
> > +	/* Create a pool of consistent memory blocks for hardware
> > descriptors */
> > +	fdma->dcb_pool = dmam_pool_create("ocelot_fdma",
> > &pdev->dev,
> > +					  sizeof(struct
> > ocelot_fdma_dcb),
> > +					  __alignof__(struct
> > ocelot_fdma_dcb),
> > +					  0);
> > +	if (!fdma->dcb_pool) {
> > +		dev_err(&pdev->dev, "unable to allocate DMA
> > descriptor pool\n");
> > +		return ERR_PTR(-ENOMEM);
> > +	}
> > +
> > +	INIT_LIST_HEAD(&fdma->tx_ongoing);
> > +	INIT_LIST_HEAD(&fdma->tx_free_dcb);
> > +	INIT_LIST_HEAD(&fdma->tx_queued);
> > +	INIT_LIST_HEAD(&fdma->rx_sw);
> > +	INIT_LIST_HEAD(&fdma->rx_hw);
> > +
> > +	return fdma;
> > +}
> > +
> > +int ocelot_fdma_start(struct ocelot_fdma *fdma)
> > +{
> > +	struct ocelot *ocelot = fdma->ocelot;
> > +	int ret;
> > +
> > +	ret = fdma_init_tx(fdma);
> > +	if (ret)
> > +		return ret;
> > +
> > +	ret = fdma_init_rx(fdma);
> > +	if (ret)  
> 
> Don't you want to undo the fdma_dcb_alloc() from fdma_init_tx() if
> this fails?

Indeed. Since I will switch to your proposal of using a single large
area of coherent descriptor, this will be removed anyway.

> 
> > +		return ret;
> > +
> > +	/* Reconfigure for extraction and injection using DMA */
> > +	ocelot_write_rix(ocelot, QS_INJ_GRP_CFG_MODE(2),
> > QS_INJ_GRP_CFG, 0);
> > +	ocelot_write_rix(ocelot, QS_INJ_CTRL_GAP_SIZE(0),
> > QS_INJ_CTRL, 0); +
> > +	ocelot_write_rix(ocelot, QS_XTR_GRP_CFG_MODE(2),
> > QS_XTR_GRP_CFG, 0); +
> > +	fdma_writel(fdma, MSCC_FDMA_INTR_LLP, 0xffffffff);
> > +	fdma_writel(fdma, MSCC_FDMA_INTR_FRM, 0xffffffff);
> > +
> > +	fdma_writel(fdma, MSCC_FDMA_INTR_LLP_ENA,
> > +		    BIT(MSCC_FDMA_INJ_CHAN) |
> > BIT(MSCC_FDMA_XTR_CHAN));
> > +	fdma_writel(fdma, MSCC_FDMA_INTR_FRM_ENA,
> > BIT(MSCC_FDMA_XTR_CHAN));
> > +	fdma_writel(fdma, MSCC_FDMA_INTR_ENA,
> > +		    BIT(MSCC_FDMA_INJ_CHAN) |
> > BIT(MSCC_FDMA_XTR_CHAN)); +
> > +	ocelot_fdma_rx_refill(fdma);
> > +
> > +	return 0;
> > +}
> > +
> > +int ocelot_fdma_stop(struct ocelot_fdma *fdma)
> > +{
> > +	fdma_writel(fdma, MSCC_FDMA_INTR_ENA, 0);
> > +
> > +	fdma_stop_channel(fdma, MSCC_FDMA_XTR_CHAN);
> > +	fdma_stop_channel(fdma, MSCC_FDMA_INJ_CHAN);
> > +
> > +	/* Free potentially pending SKBs in DCB lists */
> > +	fdma_free_skbs_list(fdma, &fdma->rx_hw, DMA_FROM_DEVICE);
> > +	fdma_free_skbs_list(fdma, &fdma->rx_sw, DMA_FROM_DEVICE);
> > +	fdma_free_skbs_list(fdma, &fdma->tx_ongoing,
> > DMA_TO_DEVICE);
> > +	fdma_free_skbs_list(fdma, &fdma->tx_queued, DMA_TO_DEVICE);
> > +
> > +	netif_napi_del(&fdma->napi);
> > +
> > +	return 0;
> > +}
> > diff --git a/drivers/net/ethernet/mscc/ocelot_fdma.h
> > b/drivers/net/ethernet/mscc/ocelot_fdma.h new file mode 100644
> > index 000000000000..6c5c5872abf5
> > --- /dev/null
> > +++ b/drivers/net/ethernet/mscc/ocelot_fdma.h
> > @@ -0,0 +1,59 @@
> > +/* SPDX-License-Identifier: (GPL-2.0 OR MIT) */
> > +/*
> > + * Microsemi SoCs FDMA driver
> > + *
> > + * Copyright (c) 2021 Microchip
> > + */
> > +#ifndef _MSCC_OCELOT_FDMA_H_
> > +#define _MSCC_OCELOT_FDMA_H_
> > +
> > +#include "ocelot.h"
> > +
> > +/**
> > + * struct ocelot_fdma - FMDA struct
> > + *
> > + * @ocelot: Pointer to ocelot struct
> > + * @base: base address of FDMA registers
> > + * @dcb_pool: Pool used for DCB allocation
> > + * @irq: FDMA interrupt
> > + * @dev: Ocelot device
> > + * @napi: napi handle
> > + * @rx_buf_size: Size of RX buffer
> > + * @tx_ongoing: List of DCB handed out to the FDMA
> > + * @tx_queued: pending list of DCBs to be given to the hardware
> > + * @tx_enqueue_lock: Lock used for tx_queued and tx_ongoing
> > + * @tx_free_dcb: List of DCB available for TX
> > + * @tx_free_lock: Lock used to access tx_free_dcb list
> > + * @rx_hw: RX DCBs currently owned by the hardware and not
> > completed
> > + * @rx_sw: RX DCBs completed
> > + */
> > +struct ocelot_fdma {
> > +	struct ocelot		*ocelot;
> > +	void __iomem		*base;
> > +	struct dma_pool		*dcb_pool;
> > +	int			irq;
> > +	struct device		*dev;
> > +	struct napi_struct	napi;
> > +	size_t			rx_buf_size;
> > +
> > +	struct list_head	tx_ongoing;
> > +	struct list_head	tx_queued;
> > +	/* Lock for tx_queued and tx_ongoing lists */
> > +	spinlock_t		tx_enqueue_lock;
> > +
> > +	struct list_head	tx_free_dcb;
> > +	/* Lock for tx_free_dcb list */
> > +	spinlock_t		tx_free_lock;
> > +
> > +	struct list_head	rx_hw;
> > +	struct list_head	rx_sw;
> > +};
> > +
> > +struct ocelot_fdma *ocelot_fdma_init(struct platform_device *pdev,
> > +				     struct ocelot *ocelot);
> > +int ocelot_fdma_start(struct ocelot_fdma *fdma);
> > +int ocelot_fdma_stop(struct ocelot_fdma *fdma);
> > +int ocelot_fdma_inject_frame(struct ocelot_fdma *fdma, int port,
> > u32 rew_op,
> > +			     struct sk_buff *skb, struct
> > net_device *dev); +
> > +#endif
> > diff --git a/drivers/net/ethernet/mscc/ocelot_net.c
> > b/drivers/net/ethernet/mscc/ocelot_net.c index
> > 5916492fd6d0..3971b810c5b4 100644 ---
> > a/drivers/net/ethernet/mscc/ocelot_net.c +++
> > b/drivers/net/ethernet/mscc/ocelot_net.c @@ -15,6 +15,7 @@
> >  #include <net/pkt_cls.h>
> >  #include "ocelot.h"
> >  #include "ocelot_vcap.h"
> > +#include "ocelot_fdma.h"
> >  
> >  #define OCELOT_MAC_QUIRKS
> > OCELOT_QUIRK_QSGMII_PORTS_MUST_BE_UP 
> > @@ -457,7 +458,7 @@ static netdev_tx_t ocelot_port_xmit(struct
> > sk_buff *skb, struct net_device *dev) int port = priv->chip_port;
> >  	u32 rew_op = 0;
> >  
> > -	if (!ocelot_can_inject(ocelot, 0))
> > +	if (!ocelot->fdma && !ocelot_can_inject(ocelot, 0))
> >  		return NETDEV_TX_BUSY;
> >  
> >  	/* Check if timestamping is needed */
> > @@ -475,9 +476,13 @@ static netdev_tx_t ocelot_port_xmit(struct
> > sk_buff *skb, struct net_device *dev) rew_op =
> > ocelot_ptp_rew_op(skb); }
> >  
> > -	ocelot_port_inject_frame(ocelot, port, 0, rew_op, skb);
> > +	if (ocelot->fdma) {
> > +		ocelot_fdma_inject_frame(ocelot->fdma, port,
> > rew_op, skb, dev);
> > +	} else {
> > +		ocelot_port_inject_frame(ocelot, port, 0, rew_op,
> > skb); 
> > -	kfree_skb(skb);
> > +		kfree_skb(skb);  
> 
> I know this is unrelated, but.. consume_skb maybe?

Acked.

> 
> > +	}
> >  
> >  	return NETDEV_TX_OK;
> >  }
> > diff --git a/drivers/net/ethernet/mscc/ocelot_vsc7514.c
> > b/drivers/net/ethernet/mscc/ocelot_vsc7514.c index
> > 38103b0255b0..985d584db3a1 100644 ---
> > a/drivers/net/ethernet/mscc/ocelot_vsc7514.c +++
> > b/drivers/net/ethernet/mscc/ocelot_vsc7514.c @@ -18,6 +18,7 @@
> >  
> >  #include <soc/mscc/ocelot_vcap.h>
> >  #include <soc/mscc/ocelot_hsio.h>
> > +#include "ocelot_fdma.h"
> >  #include "ocelot.h"
> >  
> >  static const u32 ocelot_ana_regmap[] = {
> > @@ -1080,6 +1081,10 @@ static int mscc_ocelot_probe(struct
> > platform_device *pdev) ocelot->targets[io_target[i].id] = target;
> >  	}
> >  
> > +	ocelot->fdma = ocelot_fdma_init(pdev, ocelot);
> > +	if (IS_ERR(ocelot->fdma))
> > +		ocelot->fdma = NULL;
> > +
> >  	hsio =
> > syscon_regmap_lookup_by_compatible("mscc,ocelot-hsio"); if
> > (IS_ERR(hsio)) { dev_err(&pdev->dev, "missing hsio syscon\n");
> > @@ -1139,6 +1144,12 @@ static int mscc_ocelot_probe(struct
> > platform_device *pdev) if (err)
> >  		goto out_ocelot_devlink_unregister;
> >  
> > +	if (ocelot->fdma) {
> > +		err = ocelot_fdma_start(ocelot->fdma);
> > +		if (err)
> > +			goto out_ocelot_devlink_unregister;
> > +	}
> > +
> >  	err = ocelot_devlink_sb_register(ocelot);
> >  	if (err)
> >  		goto out_ocelot_release_ports;
> > @@ -1166,6 +1177,8 @@ static int mscc_ocelot_probe(struct
> > platform_device *pdev) out_ocelot_release_ports:
> >  	mscc_ocelot_release_ports(ocelot);
> >  	mscc_ocelot_teardown_devlink_ports(ocelot);
> > +	if (ocelot->fdma)
> > +		ocelot_fdma_stop(ocelot->fdma);
> >  out_ocelot_devlink_unregister:
> >  	ocelot_deinit(ocelot);
> >  out_put_ports:
> > @@ -1179,6 +1192,8 @@ static int mscc_ocelot_remove(struct
> > platform_device *pdev) {
> >  	struct ocelot *ocelot = platform_get_drvdata(pdev);
> >  
> > +	if (ocelot->fdma)
> > +		ocelot_fdma_stop(ocelot->fdma);  
> 
> Are you sure you want to call netif_napi_del() while the net devices
> are still registered? :-/

Indeed, the napi removal must be done later, I will split fdma_stop().

> 
> >  	devlink_unregister(ocelot->devlink);
> >  	ocelot_deinit_timestamp(ocelot);
> >  	ocelot_devlink_sb_unregister(ocelot);
> > diff --git a/include/soc/mscc/ocelot.h b/include/soc/mscc/ocelot.h
> > index b3381c90ff3e..33e1559bdea3 100644
> > --- a/include/soc/mscc/ocelot.h
> > +++ b/include/soc/mscc/ocelot.h
> > @@ -695,6 +695,8 @@ struct ocelot {
> >  	/* Protects the PTP clock */
> >  	spinlock_t			ptp_clock_lock;
> >  	struct ptp_pin_desc
> > ptp_pins[OCELOT_PTP_PINS_NUM]; +
> > +	struct ocelot_fdma		*fdma;
> >  };
> >  
> >  struct ocelot_policer {
> > -- 
> > 2.33.0
>   


-- 
Clément Léger,
Embedded Linux and Kernel engineer at Bootlin
https://bootlin.com