Merge branch 'net-lan966x-add-support-for-fdma'

Horatiu Vultur says:

====================
net: lan966x: Add support for FDMA

Currently when injecting or extracting a frame from CPU, the frame
is given to the HW each word at a time. There is another way to
inject/extract frames from CPU using FDMA(Frame Direct Memory Access).
In this way the entire frame is given to the HW. This improves both
RX and TX bitrate.
====================

Tested-by: Michael Walle <michael@walle.cc> # on kontron-kswitch-d10
Link: https://lore.kernel.org/r/20220408070357.559899-1-horatiu.vultur@microchip.comSigned-off-by: Jakub Kicinski <kuba@kernel.org>

drivers/net/ethernet/microchip/lan966x/Makefile

@@ -8,4 +8,4 @@ obj-$(CONFIG_LAN966X_SWITCH) += lan966x-switch.o
 lan966x-switch-objs  := lan966x_main.o lan966x_phylink.o lan966x_port.o \
 			lan966x_mac.o lan966x_ethtool.o lan966x_switchdev.o \
 			lan966x_vlan.o lan966x_fdb.o lan966x_mdb.o \
-			lan966x_ptp.o
+			lan966x_ptp.o lan966x_fdma.o

drivers/net/ethernet/microchip/lan966x/lan966x_main.c

@@ -24,9 +24,6 @@
 #define XTR_NOT_READY			0x07000080U
 #define XTR_VALID_BYTES(x)		(4 - (((x) >> 24) & 3))
 
-#define READL_SLEEP_US			10
-#define READL_TIMEOUT_US		100000000
-
 #define IO_RANGES 2
 
 static const struct of_device_id lan966x_match[] = {
@@ -43,6 +40,7 @@ struct lan966x_main_io_resource {
 
 static const struct lan966x_main_io_resource lan966x_main_iomap[] =  {
 	{ TARGET_CPU,                   0xc0000, 0 }, /* 0xe00c0000 */
+	{ TARGET_FDMA,                  0xc0400, 0 }, /* 0xe00c0400 */
 	{ TARGET_ORG,                         0, 1 }, /* 0xe2000000 */
 	{ TARGET_GCB,                    0x4000, 1 }, /* 0xe2004000 */
 	{ TARGET_QS,                     0x8000, 1 }, /* 0xe2008000 */
@@ -343,7 +341,10 @@ static int lan966x_port_xmit(struct sk_buff *skb, struct net_device *dev)
 	}
 
 	spin_lock(&lan966x->tx_lock);
-	err = lan966x_port_ifh_xmit(skb, ifh, dev);
+	if (port->lan966x->fdma)
+		err = lan966x_fdma_xmit(skb, ifh, dev);
+	else
+		err = lan966x_port_ifh_xmit(skb, ifh, dev);
 	spin_unlock(&lan966x->tx_lock);
 
 	return err;
@@ -353,12 +354,24 @@ static int lan966x_port_change_mtu(struct net_device *dev, int new_mtu)
 {
 	struct lan966x_port *port = netdev_priv(dev);
 	struct lan966x *lan966x = port->lan966x;
+	int old_mtu = dev->mtu;
+	int err;
 
 	lan_wr(DEV_MAC_MAXLEN_CFG_MAX_LEN_SET(new_mtu),
 	       lan966x, DEV_MAC_MAXLEN_CFG(port->chip_port));
 	dev->mtu = new_mtu;
 
-	return 0;
+	if (!lan966x->fdma)
+		return 0;
+
+	err = lan966x_fdma_change_mtu(lan966x);
+	if (err) {
+		lan_wr(DEV_MAC_MAXLEN_CFG_MAX_LEN_SET(old_mtu),
+		       lan966x, DEV_MAC_MAXLEN_CFG(port->chip_port));
+		dev->mtu = old_mtu;
+	}
+
+	return err;
 }
 
 static int lan966x_mc_unsync(struct net_device *dev, const unsigned char *addr)
@@ -432,8 +445,7 @@ bool lan966x_netdevice_check(const struct net_device *dev)
 	return dev->netdev_ops == &lan966x_port_netdev_ops;
 }
 
-static bool lan966x_hw_offload(struct lan966x *lan966x, u32 port,
-			       struct sk_buff *skb)
+bool lan966x_hw_offload(struct lan966x *lan966x, u32 port, struct sk_buff *skb)
 {
 	u32 val;
 
@@ -514,7 +526,7 @@ static int lan966x_rx_frame_word(struct lan966x *lan966x, u8 grp, u32 *rval)
 	}
 }
 
-static void lan966x_ifh_get_src_port(void *ifh, u64 *src_port)
+void lan966x_ifh_get_src_port(void *ifh, u64 *src_port)
 {
 	packing(ifh, src_port, IFH_POS_SRCPORT + IFH_WID_SRCPORT - 1,
 		IFH_POS_SRCPORT, IFH_LEN * 4, UNPACK, 0);
@@ -526,7 +538,7 @@ static void lan966x_ifh_get_len(void *ifh, u64 *len)
 		IFH_POS_LEN, IFH_LEN * 4, UNPACK, 0);
 }
 
-static void lan966x_ifh_get_timestamp(void *ifh, u64 *timestamp)
+void lan966x_ifh_get_timestamp(void *ifh, u64 *timestamp)
 {
 	packing(ifh, timestamp, IFH_POS_TIMESTAMP + IFH_WID_TIMESTAMP - 1,
 		IFH_POS_TIMESTAMP, IFH_LEN * 4, UNPACK, 0);
@@ -646,6 +658,9 @@ static void lan966x_cleanup_ports(struct lan966x *lan966x)
 		if (port->dev)
 			unregister_netdev(port->dev);
 
+		if (lan966x->fdma && lan966x->fdma_ndev == port->dev)
+			lan966x_fdma_netdev_deinit(lan966x, port->dev);
+
 		if (port->phylink) {
 			rtnl_lock();
 			lan966x_port_stop(port->dev);
@@ -665,6 +680,9 @@ static void lan966x_cleanup_ports(struct lan966x *lan966x)
 		disable_irq(lan966x->ana_irq);
 		lan966x->ana_irq = -ENXIO;
 	}
+
+	if (lan966x->fdma)
+		devm_free_irq(lan966x->dev, lan966x->fdma_irq, lan966x);
 }
 
 static int lan966x_probe_port(struct lan966x *lan966x, u32 p,
@@ -793,12 +811,12 @@ static void lan966x_init(struct lan966x *lan966x)
 	/* Do byte-swap and expect status after last data word
 	 * Extraction: Mode: manual extraction) | Byte_swap
 	 */
-	lan_wr(QS_XTR_GRP_CFG_MODE_SET(1) |
+	lan_wr(QS_XTR_GRP_CFG_MODE_SET(lan966x->fdma ? 2 : 1) |
 	       QS_XTR_GRP_CFG_BYTE_SWAP_SET(1),
 	       lan966x, QS_XTR_GRP_CFG(0));
 
 	/* Injection: Mode: manual injection | Byte_swap */
-	lan_wr(QS_INJ_GRP_CFG_MODE_SET(1) |
+	lan_wr(QS_INJ_GRP_CFG_MODE_SET(lan966x->fdma ? 2 : 1) |
 	       QS_INJ_GRP_CFG_BYTE_SWAP_SET(1),
 	       lan966x, QS_INJ_GRP_CFG(0));
 
@@ -1020,6 +1038,17 @@ static int lan966x_probe(struct platform_device *pdev)
 		lan966x->ptp = 1;
 	}
 
+	lan966x->fdma_irq = platform_get_irq_byname(pdev, "fdma");
+	if (lan966x->fdma_irq > 0) {
+		err = devm_request_irq(&pdev->dev, lan966x->fdma_irq,
+				       lan966x_fdma_irq_handler, 0,
+				       "fdma irq", lan966x);
+		if (err)
+			return dev_err_probe(&pdev->dev, err, "Unable to use fdma irq");
+
+		lan966x->fdma = true;
+	}
+
 	/* init switch */
 	lan966x_init(lan966x);
 	lan966x_stats_init(lan966x);
@@ -1058,8 +1087,15 @@ static int lan966x_probe(struct platform_device *pdev)
 	if (err)
 		goto cleanup_fdb;
 
+	err = lan966x_fdma_init(lan966x);
+	if (err)
+		goto cleanup_ptp;
+
 	return 0;
 
+cleanup_ptp:
+	lan966x_ptp_deinit(lan966x);
+
 cleanup_fdb:
 	lan966x_fdb_deinit(lan966x);
 
@@ -1079,6 +1115,7 @@ static int lan966x_remove(struct platform_device *pdev)
 {
 	struct lan966x *lan966x = platform_get_drvdata(pdev);
 
+	lan966x_fdma_deinit(lan966x);
 	lan966x_cleanup_ports(lan966x);
 
 	cancel_delayed_work_sync(&lan966x->stats_work);

drivers/net/ethernet/microchip/lan966x/lan966x_main.h

@@ -17,6 +17,9 @@
 #define TABLE_UPDATE_SLEEP_US		10
 #define TABLE_UPDATE_TIMEOUT_US		100000
 
+#define READL_SLEEP_US			10
+#define READL_TIMEOUT_US		100000000
+
 #define LAN966X_BUFFER_CELL_SZ		64
 #define LAN966X_BUFFER_MEMORY		(160 * 1024)
 #define LAN966X_BUFFER_MIN_SZ		60
@@ -58,6 +61,22 @@
 #define IFH_REW_OP_ONE_STEP_PTP		0x3
 #define IFH_REW_OP_TWO_STEP_PTP		0x4
 
+#define FDMA_RX_DCB_MAX_DBS		1
+#define FDMA_TX_DCB_MAX_DBS		1
+#define FDMA_DCB_INFO_DATAL(x)		((x) & GENMASK(15, 0))
+
+#define FDMA_DCB_STATUS_BLOCKL(x)	((x) & GENMASK(15, 0))
+#define FDMA_DCB_STATUS_SOF		BIT(16)
+#define FDMA_DCB_STATUS_EOF		BIT(17)
+#define FDMA_DCB_STATUS_INTR		BIT(18)
+#define FDMA_DCB_STATUS_DONE		BIT(19)
+#define FDMA_DCB_STATUS_BLOCKO(x)	(((x) << 20) & GENMASK(31, 20))
+#define FDMA_DCB_INVALID_DATA		0x1
+
+#define FDMA_XTR_CHANNEL		6
+#define FDMA_INJ_CHANNEL		0
+#define FDMA_DCB_MAX			512
+
 /* MAC table entry types.
  * ENTRYTYPE_NORMAL is subject to aging.
  * ENTRYTYPE_LOCKED is not subject to aging.
@@ -73,6 +92,83 @@ enum macaccess_entry_type {
 
 struct lan966x_port;
 
+struct lan966x_db {
+	u64 dataptr;
+	u64 status;
+};
+
+struct lan966x_rx_dcb {
+	u64 nextptr;
+	u64 info;
+	struct lan966x_db db[FDMA_RX_DCB_MAX_DBS];
+};
+
+struct lan966x_tx_dcb {
+	u64 nextptr;
+	u64 info;
+	struct lan966x_db db[FDMA_TX_DCB_MAX_DBS];
+};
+
+struct lan966x_rx {
+	struct lan966x *lan966x;
+
+	/* Pointer to the array of hardware dcbs. */
+	struct lan966x_rx_dcb *dcbs;
+
+	/* Pointer to the last address in the dcbs. */
+	struct lan966x_rx_dcb *last_entry;
+
+	/* For each DB, there is a page */
+	struct page *page[FDMA_DCB_MAX][FDMA_RX_DCB_MAX_DBS];
+
+	/* Represents the db_index, it can have a value between 0 and
+	 * FDMA_RX_DCB_MAX_DBS, once it reaches the value of FDMA_RX_DCB_MAX_DBS
+	 * it means that the DCB can be reused.
+	 */
+	int db_index;
+
+	/* Represents the index in the dcbs. It has a value between 0 and
+	 * FDMA_DCB_MAX
+	 */
+	int dcb_index;
+
+	/* Represents the dma address to the dcbs array */
+	dma_addr_t dma;
+
+	/* Represents the page order that is used to allocate the pages for the
+	 * RX buffers. This value is calculated based on max MTU of the devices.
+	 */
+	u8 page_order;
+
+	u8 channel_id;
+};
+
+struct lan966x_tx_dcb_buf {
+	struct net_device *dev;
+	struct sk_buff *skb;
+	dma_addr_t dma_addr;
+	bool used;
+	bool ptp;
+};
+
+struct lan966x_tx {
+	struct lan966x *lan966x;
+
+	/* Pointer to the dcb list */
+	struct lan966x_tx_dcb *dcbs;
+	u16 last_in_use;
+
+	/* Represents the DMA address to the first entry of the dcb entries. */
+	dma_addr_t dma;
+
+	/* Array of dcbs that are given to the HW */
+	struct lan966x_tx_dcb_buf *dcbs_buf;
+
+	u8 channel_id;
+
+	bool activated;
+};
+
 struct lan966x_stat_layout {
 	u32 offset;
 	char name[ETH_GSTRING_LEN];
@@ -134,6 +230,7 @@ struct lan966x {
 	int xtr_irq;
 	int ana_irq;
 	int ptp_irq;
+	int fdma_irq;
 
 	/* worqueue for fdb */
 	struct workqueue_struct *fdb_work;
@@ -150,6 +247,13 @@ struct lan966x {
 	spinlock_t ptp_ts_id_lock; /* lock for ts_id */
 	struct mutex ptp_lock; /* lock for ptp interface state */
 	u16 ptp_skbs;
+
+	/* fdma */
+	bool fdma;
+	struct net_device *fdma_ndev;
+	struct lan966x_rx rx;
+	struct lan966x_tx tx;
+	struct napi_struct napi;
 };
 
 struct lan966x_port_config {
@@ -195,6 +299,11 @@ bool lan966x_netdevice_check(const struct net_device *dev);
 void lan966x_register_notifier_blocks(void);
 void lan966x_unregister_notifier_blocks(void);
 
+bool lan966x_hw_offload(struct lan966x *lan966x, u32 port, struct sk_buff *skb);
+
+void lan966x_ifh_get_src_port(void *ifh, u64 *src_port);
+void lan966x_ifh_get_timestamp(void *ifh, u64 *timestamp);
+
 void lan966x_stats_get(struct net_device *dev,
 		       struct rtnl_link_stats64 *stats);
 int lan966x_stats_init(struct lan966x *lan966x);
@@ -284,6 +393,14 @@ void lan966x_ptp_txtstamp_release(struct lan966x_port *port,
 				  struct sk_buff *skb);
 irqreturn_t lan966x_ptp_irq_handler(int irq, void *args);
 
+int lan966x_fdma_xmit(struct sk_buff *skb, __be32 *ifh, struct net_device *dev);
+int lan966x_fdma_change_mtu(struct lan966x *lan966x);
+void lan966x_fdma_netdev_init(struct lan966x *lan966x, struct net_device *dev);
+void lan966x_fdma_netdev_deinit(struct lan966x *lan966x, struct net_device *dev);
+int lan966x_fdma_init(struct lan966x *lan966x);
+void lan966x_fdma_deinit(struct lan966x *lan966x);
+irqreturn_t lan966x_fdma_irq_handler(int irq, void *args);
+
 static inline void __iomem *lan_addr(void __iomem *base[],
 				     int id, int tinst, int tcnt,
 				     int gbase, int ginst,

drivers/net/ethernet/microchip/lan966x/lan966x_port.c

@@ -393,6 +393,9 @@ void lan966x_port_init(struct lan966x_port *port)
 
 	lan966x_port_config_down(port);
 
+	if (lan966x->fdma)
+		lan966x_fdma_netdev_init(lan966x, port->dev);
+
 	if (config->portmode != PHY_INTERFACE_MODE_QSGMII)
 		return;
 

drivers/net/ethernet/microchip/lan966x/lan966x_regs.h

@@ -17,6 +17,7 @@ enum lan966x_target {
 	TARGET_CHIP_TOP = 5,
 	TARGET_CPU = 6,
 	TARGET_DEV = 13,
+	TARGET_FDMA = 21,
 	TARGET_GCB = 27,
 	TARGET_ORG = 36,
 	TARGET_PTP = 41,
@@ -578,6 +579,111 @@ enum lan966x_target {
 #define DEV_PCS1G_STICKY_LINK_DOWN_STICKY_GET(x)\
 	FIELD_GET(DEV_PCS1G_STICKY_LINK_DOWN_STICKY, x)
 
+/*      FDMA:FDMA:FDMA_CH_ACTIVATE */
+#define FDMA_CH_ACTIVATE          __REG(TARGET_FDMA, 0, 1, 8, 0, 1, 428, 0, 0, 1, 4)
+
+#define FDMA_CH_ACTIVATE_CH_ACTIVATE             GENMASK(7, 0)
+#define FDMA_CH_ACTIVATE_CH_ACTIVATE_SET(x)\
+	FIELD_PREP(FDMA_CH_ACTIVATE_CH_ACTIVATE, x)
+#define FDMA_CH_ACTIVATE_CH_ACTIVATE_GET(x)\
+	FIELD_GET(FDMA_CH_ACTIVATE_CH_ACTIVATE, x)
+
+/*      FDMA:FDMA:FDMA_CH_RELOAD */
+#define FDMA_CH_RELOAD            __REG(TARGET_FDMA, 0, 1, 8, 0, 1, 428, 4, 0, 1, 4)
+
+#define FDMA_CH_RELOAD_CH_RELOAD                 GENMASK(7, 0)
+#define FDMA_CH_RELOAD_CH_RELOAD_SET(x)\
+	FIELD_PREP(FDMA_CH_RELOAD_CH_RELOAD, x)
+#define FDMA_CH_RELOAD_CH_RELOAD_GET(x)\
+	FIELD_GET(FDMA_CH_RELOAD_CH_RELOAD, x)
+
+/*      FDMA:FDMA:FDMA_CH_DISABLE */
+#define FDMA_CH_DISABLE           __REG(TARGET_FDMA, 0, 1, 8, 0, 1, 428, 8, 0, 1, 4)
+
+#define FDMA_CH_DISABLE_CH_DISABLE               GENMASK(7, 0)
+#define FDMA_CH_DISABLE_CH_DISABLE_SET(x)\
+	FIELD_PREP(FDMA_CH_DISABLE_CH_DISABLE, x)
+#define FDMA_CH_DISABLE_CH_DISABLE_GET(x)\
+	FIELD_GET(FDMA_CH_DISABLE_CH_DISABLE, x)
+
+/*      FDMA:FDMA:FDMA_CH_DB_DISCARD */
+#define FDMA_CH_DB_DISCARD        __REG(TARGET_FDMA, 0, 1, 8, 0, 1, 428, 16, 0, 1, 4)
+
+#define FDMA_CH_DB_DISCARD_DB_DISCARD            GENMASK(7, 0)
+#define FDMA_CH_DB_DISCARD_DB_DISCARD_SET(x)\
+	FIELD_PREP(FDMA_CH_DB_DISCARD_DB_DISCARD, x)
+#define FDMA_CH_DB_DISCARD_DB_DISCARD_GET(x)\
+	FIELD_GET(FDMA_CH_DB_DISCARD_DB_DISCARD, x)
+
+/*      FDMA:FDMA:FDMA_DCB_LLP */
+#define FDMA_DCB_LLP(r)           __REG(TARGET_FDMA, 0, 1, 8, 0, 1, 428, 52, r, 8, 4)
+
+/*      FDMA:FDMA:FDMA_DCB_LLP1 */
+#define FDMA_DCB_LLP1(r)          __REG(TARGET_FDMA, 0, 1, 8, 0, 1, 428, 84, r, 8, 4)
+
+/*      FDMA:FDMA:FDMA_CH_ACTIVE */
+#define FDMA_CH_ACTIVE            __REG(TARGET_FDMA, 0, 1, 8, 0, 1, 428, 180, 0, 1, 4)
+
+/*      FDMA:FDMA:FDMA_CH_CFG */
+#define FDMA_CH_CFG(r)            __REG(TARGET_FDMA, 0, 1, 8, 0, 1, 428, 224, r, 8, 4)
+
+#define FDMA_CH_CFG_CH_INTR_DB_EOF_ONLY          BIT(4)
+#define FDMA_CH_CFG_CH_INTR_DB_EOF_ONLY_SET(x)\
+	FIELD_PREP(FDMA_CH_CFG_CH_INTR_DB_EOF_ONLY, x)
+#define FDMA_CH_CFG_CH_INTR_DB_EOF_ONLY_GET(x)\
+	FIELD_GET(FDMA_CH_CFG_CH_INTR_DB_EOF_ONLY, x)
+
+#define FDMA_CH_CFG_CH_INJ_PORT                  BIT(3)
+#define FDMA_CH_CFG_CH_INJ_PORT_SET(x)\
+	FIELD_PREP(FDMA_CH_CFG_CH_INJ_PORT, x)
+#define FDMA_CH_CFG_CH_INJ_PORT_GET(x)\
+	FIELD_GET(FDMA_CH_CFG_CH_INJ_PORT, x)
+
+#define FDMA_CH_CFG_CH_DCB_DB_CNT                GENMASK(2, 1)
+#define FDMA_CH_CFG_CH_DCB_DB_CNT_SET(x)\
+	FIELD_PREP(FDMA_CH_CFG_CH_DCB_DB_CNT, x)
+#define FDMA_CH_CFG_CH_DCB_DB_CNT_GET(x)\
+	FIELD_GET(FDMA_CH_CFG_CH_DCB_DB_CNT, x)
+
+#define FDMA_CH_CFG_CH_MEM                       BIT(0)
+#define FDMA_CH_CFG_CH_MEM_SET(x)\
+	FIELD_PREP(FDMA_CH_CFG_CH_MEM, x)
+#define FDMA_CH_CFG_CH_MEM_GET(x)\
+	FIELD_GET(FDMA_CH_CFG_CH_MEM, x)
+
+/*      FDMA:FDMA:FDMA_PORT_CTRL */
+#define FDMA_PORT_CTRL(r)         __REG(TARGET_FDMA, 0, 1, 8, 0, 1, 428, 376, r, 2, 4)
+
+#define FDMA_PORT_CTRL_INJ_STOP                  BIT(4)
+#define FDMA_PORT_CTRL_INJ_STOP_SET(x)\
+	FIELD_PREP(FDMA_PORT_CTRL_INJ_STOP, x)
+#define FDMA_PORT_CTRL_INJ_STOP_GET(x)\
+	FIELD_GET(FDMA_PORT_CTRL_INJ_STOP, x)
+
+#define FDMA_PORT_CTRL_XTR_STOP                  BIT(2)
+#define FDMA_PORT_CTRL_XTR_STOP_SET(x)\
+	FIELD_PREP(FDMA_PORT_CTRL_XTR_STOP, x)
+#define FDMA_PORT_CTRL_XTR_STOP_GET(x)\
+	FIELD_GET(FDMA_PORT_CTRL_XTR_STOP, x)
+
+/*      FDMA:FDMA:FDMA_INTR_DB */
+#define FDMA_INTR_DB              __REG(TARGET_FDMA, 0, 1, 8, 0, 1, 428, 392, 0, 1, 4)
+
+/*      FDMA:FDMA:FDMA_INTR_DB_ENA */
+#define FDMA_INTR_DB_ENA          __REG(TARGET_FDMA, 0, 1, 8, 0, 1, 428, 396, 0, 1, 4)
+
+#define FDMA_INTR_DB_ENA_INTR_DB_ENA             GENMASK(7, 0)
+#define FDMA_INTR_DB_ENA_INTR_DB_ENA_SET(x)\
+	FIELD_PREP(FDMA_INTR_DB_ENA_INTR_DB_ENA, x)
+#define FDMA_INTR_DB_ENA_INTR_DB_ENA_GET(x)\
+	FIELD_GET(FDMA_INTR_DB_ENA_INTR_DB_ENA, x)
+
+/*      FDMA:FDMA:FDMA_INTR_ERR */
+#define FDMA_INTR_ERR             __REG(TARGET_FDMA, 0, 1, 8, 0, 1, 428, 400, 0, 1, 4)
+
+/*      FDMA:FDMA:FDMA_ERRORS */
+#define FDMA_ERRORS               __REG(TARGET_FDMA, 0, 1, 8, 0, 1, 428, 412, 0, 1, 4)
+
 /*      PTP:PTP_CFG:PTP_DOM_CFG */
 #define PTP_DOM_CFG               __REG(TARGET_PTP, 0, 1, 512, 0, 1, 16, 12, 0, 1, 4)