ionic: add tx/rx-push support with device Component Memory Buffers

The ionic device has on-board memory (CMB) that can be used
for descriptors as a way to speed descriptor access for faster
packet processing.  It is rumored to improve latency and/or
packets-per-second for some profiles of small packet traffic,
although your mileage may vary.
Signed-off-by: Shannon Nelson <shannon.nelson@amd.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

drivers/net/ethernet/pensando/ionic/ionic_bus_pci.c

@@ -352,6 +352,7 @@ static int ionic_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
 err_out_reset:
 	ionic_reset(ionic);
 err_out_teardown:
+	ionic_dev_teardown(ionic);
 	pci_clear_master(pdev);
 	/* Don't fail the probe for these errors, keep
 	 * the hw interface around for inspection
@@ -390,6 +391,7 @@ static void ionic_remove(struct pci_dev *pdev)
 
 	ionic_port_reset(ionic);
 	ionic_reset(ionic);
+	ionic_dev_teardown(ionic);
 	pci_clear_master(pdev);
 	ionic_unmap_bars(ionic);
 	pci_release_regions(pdev);

drivers/net/ethernet/pensando/ionic/ionic_dev.c

@@ -92,6 +92,7 @@ int ionic_dev_setup(struct ionic *ionic)
 	unsigned int num_bars = ionic->num_bars;
 	struct ionic_dev *idev = &ionic->idev;
 	struct device *dev = ionic->dev;
+	int size;
 	u32 sig;
 
 	/* BAR0: dev_cmd and interrupts */
@@ -133,9 +134,36 @@ int ionic_dev_setup(struct ionic *ionic)
 	idev->db_pages = bar->vaddr;
 	idev->phy_db_pages = bar->bus_addr;
 
+	/* BAR2: optional controller memory mapping */
+	bar++;
+	mutex_init(&idev->cmb_inuse_lock);
+	if (num_bars < 3 || !ionic->bars[IONIC_PCI_BAR_CMB].len) {
+		idev->cmb_inuse = NULL;
+		return 0;
+	}
+
+	idev->phy_cmb_pages = bar->bus_addr;
+	idev->cmb_npages = bar->len / PAGE_SIZE;
+	size = BITS_TO_LONGS(idev->cmb_npages) * sizeof(long);
+	idev->cmb_inuse = kzalloc(size, GFP_KERNEL);
+	if (!idev->cmb_inuse)
+		dev_warn(dev, "No memory for CMB, disabling\n");
+
 	return 0;
 }
 
+void ionic_dev_teardown(struct ionic *ionic)
+{
+	struct ionic_dev *idev = &ionic->idev;
+
+	kfree(idev->cmb_inuse);
+	idev->cmb_inuse = NULL;
+	idev->phy_cmb_pages = 0;
+	idev->cmb_npages = 0;
+
+	mutex_destroy(&idev->cmb_inuse_lock);
+}
+
 /* Devcmd Interface */
 bool ionic_is_fw_running(struct ionic_dev *idev)
 {
@@ -571,6 +599,33 @@ int ionic_db_page_num(struct ionic_lif *lif, int pid)
 	return (lif->hw_index * lif->dbid_count) + pid;
 }
 
+int ionic_get_cmb(struct ionic_lif *lif, u32 *pgid, phys_addr_t *pgaddr, int order)
+{
+	struct ionic_dev *idev = &lif->ionic->idev;
+	int ret;
+
+	mutex_lock(&idev->cmb_inuse_lock);
+	ret = bitmap_find_free_region(idev->cmb_inuse, idev->cmb_npages, order);
+	mutex_unlock(&idev->cmb_inuse_lock);
+
+	if (ret < 0)
+		return ret;
+
+	*pgid = ret;
+	*pgaddr = idev->phy_cmb_pages + ret * PAGE_SIZE;
+
+	return 0;
+}
+
+void ionic_put_cmb(struct ionic_lif *lif, u32 pgid, int order)
+{
+	struct ionic_dev *idev = &lif->ionic->idev;
+
+	mutex_lock(&idev->cmb_inuse_lock);
+	bitmap_release_region(idev->cmb_inuse, pgid, order);
+	mutex_unlock(&idev->cmb_inuse_lock);
+}
+
 int ionic_cq_init(struct ionic_lif *lif, struct ionic_cq *cq,
 		  struct ionic_intr_info *intr,
 		  unsigned int num_descs, size_t desc_size)
@@ -679,6 +734,18 @@ void ionic_q_map(struct ionic_queue *q, void *base, dma_addr_t base_pa)
 		cur->desc = base + (i * q->desc_size);
 }
 
+void ionic_q_cmb_map(struct ionic_queue *q, void __iomem *base, dma_addr_t base_pa)
+{
+	struct ionic_desc_info *cur;
+	unsigned int i;
+
+	q->cmb_base = base;
+	q->cmb_base_pa = base_pa;
+
+	for (i = 0, cur = q->info; i < q->num_descs; i++, cur++)
+		cur->cmb_desc = base + (i * q->desc_size);
+}
+
 void ionic_q_sg_map(struct ionic_queue *q, void *base, dma_addr_t base_pa)
 {
 	struct ionic_desc_info *cur;

drivers/net/ethernet/pensando/ionic/ionic_dev.h

@@ -159,6 +159,11 @@ struct ionic_dev {
 	struct ionic_intr __iomem *intr_ctrl;
 	u64 __iomem *intr_status;
 
+	struct mutex cmb_inuse_lock; /* for cmb_inuse */
+	unsigned long *cmb_inuse;
+	dma_addr_t phy_cmb_pages;
+	u32 cmb_npages;
+
 	u32 port_info_sz;
 	struct ionic_port_info *port_info;
 	dma_addr_t port_info_pa;
@@ -203,6 +208,7 @@ struct ionic_desc_info {
 		struct ionic_rxq_desc *rxq_desc;
 		struct ionic_admin_cmd *adminq_desc;
 	};
+	void __iomem *cmb_desc;
 	union {
 		void *sg_desc;
 		struct ionic_txq_sg_desc *txq_sg_desc;
@@ -241,12 +247,14 @@ struct ionic_queue {
 		struct ionic_rxq_desc *rxq;
 		struct ionic_admin_cmd *adminq;
 	};
+	void __iomem *cmb_base;
 	union {
 		void *sg_base;
 		struct ionic_txq_sg_desc *txq_sgl;
 		struct ionic_rxq_sg_desc *rxq_sgl;
 	};
 	dma_addr_t base_pa;
+	dma_addr_t cmb_base_pa;
 	dma_addr_t sg_base_pa;
 	unsigned int desc_size;
 	unsigned int sg_desc_size;
@@ -309,6 +317,7 @@ static inline bool ionic_q_has_space(struct ionic_queue *q, unsigned int want)
 
 void ionic_init_devinfo(struct ionic *ionic);
 int ionic_dev_setup(struct ionic *ionic);
+void ionic_dev_teardown(struct ionic *ionic);
 
 void ionic_dev_cmd_go(struct ionic_dev *idev, union ionic_dev_cmd *cmd);
 u8 ionic_dev_cmd_status(struct ionic_dev *idev);
@@ -344,6 +353,9 @@ void ionic_dev_cmd_adminq_init(struct ionic_dev *idev, struct ionic_qcq *qcq,
 
 int ionic_db_page_num(struct ionic_lif *lif, int pid);
 
+int ionic_get_cmb(struct ionic_lif *lif, u32 *pgid, phys_addr_t *pgaddr, int order);
+void ionic_put_cmb(struct ionic_lif *lif, u32 pgid, int order);
+
 int ionic_cq_init(struct ionic_lif *lif, struct ionic_cq *cq,
 		  struct ionic_intr_info *intr,
 		  unsigned int num_descs, size_t desc_size);
@@ -360,6 +372,7 @@ int ionic_q_init(struct ionic_lif *lif, struct ionic_dev *idev,
 		 unsigned int num_descs, size_t desc_size,
 		 size_t sg_desc_size, unsigned int pid);
 void ionic_q_map(struct ionic_queue *q, void *base, dma_addr_t base_pa);
+void ionic_q_cmb_map(struct ionic_queue *q, void __iomem *base, dma_addr_t base_pa);
 void ionic_q_sg_map(struct ionic_queue *q, void *base, dma_addr_t base_pa);
 void ionic_q_post(struct ionic_queue *q, bool ring_doorbell, ionic_desc_cb cb,
 		  void *cb_arg);

drivers/net/ethernet/pensando/ionic/ionic_ethtool.c

@@ -511,6 +511,87 @@ static int ionic_set_coalesce(struct net_device *netdev,
 	return 0;
 }
 
+static int ionic_validate_cmb_config(struct ionic_lif *lif,
+				     struct ionic_queue_params *qparam)
+{
+	int pages_have, pages_required = 0;
+	unsigned long sz;
+
+	if (!lif->ionic->idev.cmb_inuse &&
+	    (qparam->cmb_tx || qparam->cmb_rx)) {
+		netdev_info(lif->netdev, "CMB rings are not supported on this device\n");
+		return -EOPNOTSUPP;
+	}
+
+	if (qparam->cmb_tx) {
+		if (!(lif->qtype_info[IONIC_QTYPE_TXQ].features & IONIC_QIDENT_F_CMB)) {
+			netdev_info(lif->netdev,
+				    "CMB rings for tx-push are not supported on this device\n");
+			return -EOPNOTSUPP;
+		}
+
+		sz = sizeof(struct ionic_txq_desc) * qparam->ntxq_descs * qparam->nxqs;
+		pages_required += ALIGN(sz, PAGE_SIZE) / PAGE_SIZE;
+	}
+
+	if (qparam->cmb_rx) {
+		if (!(lif->qtype_info[IONIC_QTYPE_RXQ].features & IONIC_QIDENT_F_CMB)) {
+			netdev_info(lif->netdev,
+				    "CMB rings for rx-push are not supported on this device\n");
+			return -EOPNOTSUPP;
+		}
+
+		sz = sizeof(struct ionic_rxq_desc) * qparam->nrxq_descs * qparam->nxqs;
+		pages_required += ALIGN(sz, PAGE_SIZE) / PAGE_SIZE;
+	}
+
+	pages_have = lif->ionic->bars[IONIC_PCI_BAR_CMB].len / PAGE_SIZE;
+	if (pages_required > pages_have) {
+		netdev_info(lif->netdev,
+			    "Not enough CMB pages for number of queues and size of descriptor rings, need %d have %d",
+			    pages_required, pages_have);
+		return -ENOMEM;
+	}
+
+	return pages_required;
+}
+
+static int ionic_cmb_rings_toggle(struct ionic_lif *lif, bool cmb_tx, bool cmb_rx)
+{
+	struct ionic_queue_params qparam;
+	int pages_used;
+
+	if (netif_running(lif->netdev)) {
+		netdev_info(lif->netdev, "Please stop device to toggle CMB for tx/rx-push\n");
+		return -EBUSY;
+	}
+
+	ionic_init_queue_params(lif, &qparam);
+	qparam.cmb_tx = cmb_tx;
+	qparam.cmb_rx = cmb_rx;
+	pages_used = ionic_validate_cmb_config(lif, &qparam);
+	if (pages_used < 0)
+		return pages_used;
+
+	if (cmb_tx)
+		set_bit(IONIC_LIF_F_CMB_TX_RINGS, lif->state);
+	else
+		clear_bit(IONIC_LIF_F_CMB_TX_RINGS, lif->state);
+
+	if (cmb_rx)
+		set_bit(IONIC_LIF_F_CMB_RX_RINGS, lif->state);
+	else
+		clear_bit(IONIC_LIF_F_CMB_RX_RINGS, lif->state);
+
+	if (cmb_tx || cmb_rx)
+		netdev_info(lif->netdev, "Enabling CMB %s %s rings - %d pages\n",
+			    cmb_tx ? "TX" : "", cmb_rx ? "RX" : "", pages_used);
+	else
+		netdev_info(lif->netdev, "Disabling CMB rings\n");
+
+	return 0;
+}
+
 static void ionic_get_ringparam(struct net_device *netdev,
 				struct ethtool_ringparam *ring,
 				struct kernel_ethtool_ringparam *kernel_ring,
@@ -522,6 +603,8 @@ static void ionic_get_ringparam(struct net_device *netdev,
 	ring->tx_pending = lif->ntxq_descs;
 	ring->rx_max_pending = IONIC_MAX_RX_DESC;
 	ring->rx_pending = lif->nrxq_descs;
+	kernel_ring->tx_push = test_bit(IONIC_LIF_F_CMB_TX_RINGS, lif->state);
+	kernel_ring->rx_push = test_bit(IONIC_LIF_F_CMB_RX_RINGS, lif->state);
 }
 
 static int ionic_set_ringparam(struct net_device *netdev,
@@ -551,9 +634,28 @@ static int ionic_set_ringparam(struct net_device *netdev,
 
 	/* if nothing to do return success */
 	if (ring->tx_pending == lif->ntxq_descs &&
-	    ring->rx_pending == lif->nrxq_descs)
+	    ring->rx_pending == lif->nrxq_descs &&
+	    kernel_ring->tx_push == test_bit(IONIC_LIF_F_CMB_TX_RINGS, lif->state) &&
+	    kernel_ring->rx_push == test_bit(IONIC_LIF_F_CMB_RX_RINGS, lif->state))
 		return 0;
 
+	qparam.ntxq_descs = ring->tx_pending;
+	qparam.nrxq_descs = ring->rx_pending;
+	qparam.cmb_tx = kernel_ring->tx_push;
+	qparam.cmb_rx = kernel_ring->rx_push;
+
+	err = ionic_validate_cmb_config(lif, &qparam);
+	if (err < 0)
+		return err;
+
+	if (kernel_ring->tx_push != test_bit(IONIC_LIF_F_CMB_TX_RINGS, lif->state) ||
+	    kernel_ring->rx_push != test_bit(IONIC_LIF_F_CMB_RX_RINGS, lif->state)) {
+		err = ionic_cmb_rings_toggle(lif, kernel_ring->tx_push,
+					     kernel_ring->rx_push);
+		if (err < 0)
+			return err;
+	}
+
 	if (ring->tx_pending != lif->ntxq_descs)
 		netdev_info(netdev, "Changing Tx ring size from %d to %d\n",
 			    lif->ntxq_descs, ring->tx_pending);
@@ -569,9 +671,6 @@ static int ionic_set_ringparam(struct net_device *netdev,
 		return 0;
 	}
 
-	qparam.ntxq_descs = ring->tx_pending;
-	qparam.nrxq_descs = ring->rx_pending;
-
 	mutex_lock(&lif->queue_lock);
 	err = ionic_reconfigure_queues(lif, &qparam);
 	mutex_unlock(&lif->queue_lock);
@@ -638,7 +737,7 @@ static int ionic_set_channels(struct net_device *netdev,
 				    lif->nxqs, ch->combined_count);
 
 		qparam.nxqs = ch->combined_count;
-		qparam.intr_split = 0;
+		qparam.intr_split = false;
 	} else {
 		max_cnt /= 2;
 		if (ch->rx_count > max_cnt)
@@ -654,9 +753,13 @@ static int ionic_set_channels(struct net_device *netdev,
 				    lif->nxqs, ch->rx_count);
 
 		qparam.nxqs = ch->rx_count;
-		qparam.intr_split = 1;
+		qparam.intr_split = true;
 	}
 
+	err = ionic_validate_cmb_config(lif, &qparam);
+	if (err < 0)
+		return err;
+
 	/* if we're not running, just set the values and return */
 	if (!netif_running(lif->netdev)) {
 		lif->nxqs = qparam.nxqs;
@@ -965,6 +1068,8 @@ static const struct ethtool_ops ionic_ethtool_ops = {
 	.supported_coalesce_params = ETHTOOL_COALESCE_USECS |
 				     ETHTOOL_COALESCE_USE_ADAPTIVE_RX |
 				     ETHTOOL_COALESCE_USE_ADAPTIVE_TX,
+	.supported_ring_params = ETHTOOL_RING_USE_TX_PUSH |
+				 ETHTOOL_RING_USE_RX_PUSH,
 	.get_drvinfo		= ionic_get_drvinfo,
 	.get_regs_len		= ionic_get_regs_len,
 	.get_regs		= ionic_get_regs,

drivers/net/ethernet/pensando/ionic/ionic_if.h

@@ -3073,9 +3073,10 @@ union ionic_adminq_comp {
 
 #define IONIC_BARS_MAX			6
 #define IONIC_PCI_BAR_DBELL		1
+#define IONIC_PCI_BAR_CMB		2
 
-/* BAR0 */
 #define IONIC_BAR0_SIZE				0x8000
+#define IONIC_BAR2_SIZE				0x800000
 
 #define IONIC_BAR0_DEV_INFO_REGS_OFFSET		0x0000
 #define IONIC_BAR0_DEV_CMD_REGS_OFFSET		0x0800

drivers/net/ethernet/pensando/ionic/ionic_lif.h

@@ -59,6 +59,7 @@ struct ionic_rx_stats {
 #define IONIC_QCQ_F_TX_STATS		BIT(3)
 #define IONIC_QCQ_F_RX_STATS		BIT(4)
 #define IONIC_QCQ_F_NOTIFYQ		BIT(5)
+#define IONIC_QCQ_F_CMB_RINGS		BIT(6)
 
 struct ionic_qcq {
 	void *q_base;
@@ -70,6 +71,11 @@ struct ionic_qcq {
 	void *sg_base;
 	dma_addr_t sg_base_pa;
 	u32 sg_size;
+	void __iomem *cmb_q_base;
+	phys_addr_t cmb_q_base_pa;
+	u32 cmb_q_size;
+	u32 cmb_pgid;
+	u32 cmb_order;
 	struct dim dim;
 	struct ionic_queue q;
 	struct ionic_cq cq;
@@ -142,6 +148,8 @@ enum ionic_lif_state_flags {
 	IONIC_LIF_F_BROKEN,
 	IONIC_LIF_F_TX_DIM_INTR,
 	IONIC_LIF_F_RX_DIM_INTR,
+	IONIC_LIF_F_CMB_TX_RINGS,
+	IONIC_LIF_F_CMB_RX_RINGS,
 
 	/* leave this as last */
 	IONIC_LIF_F_STATE_SIZE
@@ -245,8 +253,10 @@ struct ionic_queue_params {
 	unsigned int nxqs;
 	unsigned int ntxq_descs;
 	unsigned int nrxq_descs;
-	unsigned int intr_split;
 	u64 rxq_features;
+	bool intr_split;
+	bool cmb_tx;
+	bool cmb_rx;
 };
 
 static inline void ionic_init_queue_params(struct ionic_lif *lif,
@@ -255,8 +265,34 @@ static inline void ionic_init_queue_params(struct ionic_lif *lif,
 	qparam->nxqs = lif->nxqs;
 	qparam->ntxq_descs = lif->ntxq_descs;
 	qparam->nrxq_descs = lif->nrxq_descs;
-	qparam->intr_split = test_bit(IONIC_LIF_F_SPLIT_INTR, lif->state);
 	qparam->rxq_features = lif->rxq_features;
+	qparam->intr_split = test_bit(IONIC_LIF_F_SPLIT_INTR, lif->state);
+	qparam->cmb_tx = test_bit(IONIC_LIF_F_CMB_TX_RINGS, lif->state);
+	qparam->cmb_rx = test_bit(IONIC_LIF_F_CMB_RX_RINGS, lif->state);
+}
+
+static inline void ionic_set_queue_params(struct ionic_lif *lif,
+					  struct ionic_queue_params *qparam)
+{
+	lif->nxqs = qparam->nxqs;
+	lif->ntxq_descs = qparam->ntxq_descs;
+	lif->nrxq_descs = qparam->nrxq_descs;
+	lif->rxq_features = qparam->rxq_features;
+
+	if (qparam->intr_split)
+		set_bit(IONIC_LIF_F_SPLIT_INTR, lif->state);
+	else
+		clear_bit(IONIC_LIF_F_SPLIT_INTR, lif->state);
+
+	if (qparam->cmb_tx)
+		set_bit(IONIC_LIF_F_CMB_TX_RINGS, lif->state);
+	else
+		clear_bit(IONIC_LIF_F_CMB_TX_RINGS, lif->state);
+
+	if (qparam->cmb_rx)
+		set_bit(IONIC_LIF_F_CMB_RX_RINGS, lif->state);
+	else
+		clear_bit(IONIC_LIF_F_CMB_RX_RINGS, lif->state);
 }
 
 static inline u32 ionic_coal_usec_to_hw(struct ionic *ionic, u32 usecs)

drivers/net/ethernet/pensando/ionic/ionic_txrx.c

@@ -402,6 +402,14 @@ bool ionic_rx_service(struct ionic_cq *cq, struct ionic_cq_info *cq_info)
 	return true;
 }
 
+static inline void ionic_write_cmb_desc(struct ionic_queue *q,
+					void __iomem *cmb_desc,
+					void *desc)
+{
+	if (q_to_qcq(q)->flags & IONIC_QCQ_F_CMB_RINGS)
+		memcpy_toio(cmb_desc, desc, q->desc_size);
+}
+
 void ionic_rx_fill(struct ionic_queue *q)
 {
 	struct net_device *netdev = q->lif->netdev;
@@ -480,6 +488,8 @@ void ionic_rx_fill(struct ionic_queue *q)
 					      IONIC_RXQ_DESC_OPCODE_SIMPLE;
 		desc_info->nbufs = nfrags;
 
+		ionic_write_cmb_desc(q, desc_info->cmb_desc, desc);
+
 		ionic_rxq_post(q, false, ionic_rx_clean, NULL);
 	}
 
@@ -943,7 +953,8 @@ static int ionic_tx_tcp_pseudo_csum(struct sk_buff *skb)
 	return 0;
 }
 
-static void ionic_tx_tso_post(struct ionic_queue *q, struct ionic_txq_desc *desc,
+static void ionic_tx_tso_post(struct ionic_queue *q,
+			      struct ionic_desc_info *desc_info,
 			      struct sk_buff *skb,
 			      dma_addr_t addr, u8 nsge, u16 len,
 			      unsigned int hdrlen, unsigned int mss,
@@ -951,6 +962,7 @@ static void ionic_tx_tso_post(struct ionic_queue *q, struct ionic_txq_desc *desc
 			      u16 vlan_tci, bool has_vlan,
 			      bool start, bool done)
 {
+	struct ionic_txq_desc *desc = desc_info->desc;
 	u8 flags = 0;
 	u64 cmd;
 
@@ -966,6 +978,8 @@ static void ionic_tx_tso_post(struct ionic_queue *q, struct ionic_txq_desc *desc
 	desc->hdr_len = cpu_to_le16(hdrlen);
 	desc->mss = cpu_to_le16(mss);
 
+	ionic_write_cmb_desc(q, desc_info->cmb_desc, desc);
+
 	if (start) {
 		skb_tx_timestamp(skb);
 		if (!unlikely(q->features & IONIC_TXQ_F_HWSTAMP))
@@ -1084,7 +1098,7 @@ static int ionic_tx_tso(struct ionic_queue *q, struct sk_buff *skb)
 		seg_rem = min(tso_rem, mss);
 		done = (tso_rem == 0);
 		/* post descriptor */
-		ionic_tx_tso_post(q, desc, skb,
+		ionic_tx_tso_post(q, desc_info, skb,
 				  desc_addr, desc_nsge, desc_len,
 				  hdrlen, mss, outer_csum, vlan_tci, has_vlan,
 				  start, done);
@@ -1133,6 +1147,8 @@ static void ionic_tx_calc_csum(struct ionic_queue *q, struct sk_buff *skb,
 	desc->csum_start = cpu_to_le16(skb_checksum_start_offset(skb));
 	desc->csum_offset = cpu_to_le16(skb->csum_offset);
 
+	ionic_write_cmb_desc(q, desc_info->cmb_desc, desc);
+
 	if (skb_csum_is_sctp(skb))
 		stats->crc32_csum++;
 	else
@@ -1170,6 +1186,8 @@ static void ionic_tx_calc_no_csum(struct ionic_queue *q, struct sk_buff *skb,
 	desc->csum_start = 0;
 	desc->csum_offset = 0;
 
+	ionic_write_cmb_desc(q, desc_info->cmb_desc, desc);
+
 	stats->csum_none++;
 }