net: thunderx: Support for upto 96 queues for a VF

This patch adds support for handling multiple qsets assigned to a
single VF. There by increasing no of queues from earlier 8 to max
no of CPUs in the system i.e 48 queues on a single node and 96 on
dual node system. User doesn't have option to assign which Qsets/VFs
 to be merged. Upon request from VF, PF assigns next free Qsets as
secondary qsets. To maintain current behavior no of queues is kept
to 8 by default which can be increased via ethtool.

If user wants to unbind NICVF driver from a secondary Qset then it
should be done after tearing down primary VF's interface.
Signed-off-by: Sunil Goutham <sgoutham@cavium.com>
Signed-off-by: Aleksey Makarov <aleksey.makarov@caviumnetworks.com>
Signed-off-by: Robert Richter <rrichter@cavium.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

drivers/net/ethernet/cavium/thunder/nic.h

@@ -258,13 +258,23 @@ struct nicvf_drv_stats {
 };
 
 struct nicvf {
+	struct nicvf		*pnicvf;
 	struct net_device	*netdev;
 	struct pci_dev		*pdev;
 	u8			vf_id;
 	u8			node;
-	u8			tns_mode;
+	u8			tns_mode:1;
+	u8			sqs_mode:1;
 	u16			mtu;
 	struct queue_set	*qs;
+#define	MAX_SQS_PER_VF_SINGLE_NODE		5
+#define	MAX_SQS_PER_VF				11
+	u8			sqs_id;
+	u8			sqs_count; /* Secondary Qset count */
+	struct nicvf		*snicvf[MAX_SQS_PER_VF];
+	u8			rx_queues;
+	u8			tx_queues;
+	u8			max_queues;
 	void __iomem		*reg_base;
 	bool			link_up;
 	u8			duplex;
@@ -330,14 +340,19 @@ struct nicvf {
 #define	NIC_MBOX_MSG_RQ_SW_SYNC		0x0F	/* Flush inflight pkts to RQ */
 #define	NIC_MBOX_MSG_BGX_STATS		0x10	/* Get stats from BGX */
 #define	NIC_MBOX_MSG_BGX_LINK_CHANGE	0x11	/* BGX:LMAC link status */
-#define NIC_MBOX_MSG_CFG_DONE		0x12	/* VF configuration done */
-#define NIC_MBOX_MSG_SHUTDOWN		0x13	/* VF is being shutdown */
+#define	NIC_MBOX_MSG_ALLOC_SQS		0x12	/* Allocate secondary Qset */
+#define	NIC_MBOX_MSG_NICVF_PTR		0x13	/* Send nicvf ptr to PF */
+#define	NIC_MBOX_MSG_PNICVF_PTR		0x14	/* Get primary qset nicvf ptr */
+#define	NIC_MBOX_MSG_SNICVF_PTR		0x15	/* Send sqet nicvf ptr to PVF */
+#define	NIC_MBOX_MSG_CFG_DONE		0xF0	/* VF configuration done */
+#define	NIC_MBOX_MSG_SHUTDOWN		0xF1	/* VF is being shutdown */
 
 struct nic_cfg_msg {
 	u8    msg;
 	u8    vf_id;
-	u8    tns_mode;
 	u8    node_id;
+	u8    tns_mode:1;
+	u8    sqs_mode:1;
 	u8    mac_addr[ETH_ALEN];
 };
 
@@ -345,6 +360,7 @@ struct nic_cfg_msg {
 struct qs_cfg_msg {
 	u8    msg;
 	u8    num;
+	u8    sqs_count;
 	u64   cfg;
 };
 
@@ -361,6 +377,7 @@ struct sq_cfg_msg {
 	u8    msg;
 	u8    qs_num;
 	u8    sq_num;
+	bool  sqs_mode;
 	u64   cfg;
 };
 
@@ -420,6 +437,21 @@ struct bgx_link_status {
 	u32   speed;
 };
 
+/* Get Extra Qset IDs */
+struct sqs_alloc {
+	u8    msg;
+	u8    vf_id;
+	u8    qs_count;
+};
+
+struct nicvf_ptr {
+	u8    msg;
+	u8    vf_id;
+	bool  sqs_mode;
+	u8    sqs_id;
+	u64   nicvf;
+};
+
 /* 128 bit shared memory between PF and each VF */
 union nic_mbx {
 	struct { u8 msg; }	msg;
@@ -434,6 +466,8 @@ union nic_mbx {
 	struct rss_cfg_msg	rss_cfg;
 	struct bgx_stats_msg    bgx_stats;
 	struct bgx_link_status  link_status;
+	struct sqs_alloc        sqs_alloc;
+	struct nicvf_ptr	nicvf;
 };
 
 #define NIC_NODE_ID_MASK	0x03

drivers/net/ethernet/cavium/thunder/nic_main.c

@@ -28,6 +28,11 @@ struct nicpf {
 	u8			num_vf_en;      /* No of VF enabled */
 	bool			vf_enabled[MAX_NUM_VFS_SUPPORTED];
 	void __iomem		*reg_base;       /* Register start address */
+	u8			num_sqs_en;	/* Secondary qsets enabled */
+	u64			nicvf[MAX_NUM_VFS_SUPPORTED];
+	u8			vf_sqs[MAX_NUM_VFS_SUPPORTED][MAX_SQS_PER_VF];
+	u8			pqs_vf[MAX_NUM_VFS_SUPPORTED];
+	bool			sqs_used[MAX_NUM_VFS_SUPPORTED];
 	struct pkind_cfg	pkind;
 #define	NIC_SET_VF_LMAC_MAP(bgx, lmac)	(((bgx & 0xF) << 4) | (lmac & 0xF))
 #define	NIC_GET_BGX_FROM_VF_LMAC_MAP(map)	((map >> 4) & 0xF)
@@ -139,13 +144,15 @@ static void nic_mbx_send_ready(struct nicpf *nic, int vf)
 
 	mbx.nic_cfg.tns_mode = NIC_TNS_BYPASS_MODE;
 
+	if (vf < MAX_LMAC) {
 		bgx_idx = NIC_GET_BGX_FROM_VF_LMAC_MAP(nic->vf_lmac_map[vf]);
 		lmac = NIC_GET_LMAC_FROM_VF_LMAC_MAP(nic->vf_lmac_map[vf]);
 
 		mac = bgx_get_lmac_mac(nic->node, bgx_idx, lmac);
 		if (mac)
 			ether_addr_copy((u8 *)&mbx.nic_cfg.mac_addr, mac);
-
+	}
+	mbx.nic_cfg.sqs_mode = (vf >= nic->num_vf_en) ? true : false;
 	mbx.nic_cfg.node_id = nic->node;
 	nic_send_msg_to_vf(nic, vf, &mbx);
 }
@@ -433,6 +440,12 @@ static void nic_config_rss(struct nicpf *nic, struct rss_cfg_msg *cfg)
 	qset = cfg->vf_id;
 
 	for (; rssi < (rssi_base + cfg->tbl_len); rssi++) {
+		u8 svf = cfg->ind_tbl[idx] >> 3;
+
+		if (svf)
+			qset = nic->vf_sqs[cfg->vf_id][svf - 1];
+		else
+			qset = cfg->vf_id;
 		nic_reg_write(nic, NIC_PF_RSSI_0_4097_RQ | (rssi << 3),
 			      (qset << 3) | (cfg->ind_tbl[idx] & 0x7));
 		idx++;
@@ -456,19 +469,31 @@ static void nic_config_rss(struct nicpf *nic, struct rss_cfg_msg *cfg)
  * VNIC6-SQ0 -> TL4(528) -> TL3[132] -> TL2[33] -> TL1[1] -> BGX1
  * VNIC7-SQ0 -> TL4(536) -> TL3[134] -> TL2[33] -> TL1[1] -> BGX1
  */
-static void nic_tx_channel_cfg(struct nicpf *nic, u8 vnic, u8 sq_idx)
+static void nic_tx_channel_cfg(struct nicpf *nic, u8 vnic,
+			       struct sq_cfg_msg *sq)
 {
 	u32 bgx, lmac, chan;
 	u32 tl2, tl3, tl4;
 	u32 rr_quantum;
+	u8 sq_idx = sq->sq_num;
+	u8 pqs_vnic;
+
+	if (sq->sqs_mode)
+		pqs_vnic = nic->pqs_vf[vnic];
+	else
+		pqs_vnic = vnic;
+
+	bgx = NIC_GET_BGX_FROM_VF_LMAC_MAP(nic->vf_lmac_map[pqs_vnic]);
+	lmac = NIC_GET_LMAC_FROM_VF_LMAC_MAP(nic->vf_lmac_map[pqs_vnic]);
 
-	bgx = NIC_GET_BGX_FROM_VF_LMAC_MAP(nic->vf_lmac_map[vnic]);
-	lmac = NIC_GET_LMAC_FROM_VF_LMAC_MAP(nic->vf_lmac_map[vnic]);
 	/* 24 bytes for FCS, IPG and preamble */
 	rr_quantum = ((NIC_HW_MAX_FRS + 24) / 4);
 
 	tl4 = (lmac * NIC_TL4_PER_LMAC) + (bgx * NIC_TL4_PER_BGX);
 	tl4 += sq_idx;
+	if (sq->sqs_mode)
+		tl4 += vnic * 8;
+
 	tl3 = tl4 / (NIC_MAX_TL4 / NIC_MAX_TL3);
 	nic_reg_write(nic, NIC_PF_QSET_0_127_SQ_0_7_CFG2 |
 		      ((u64)vnic << NIC_QS_ID_SHIFT) |
@@ -489,6 +514,71 @@ static void nic_tx_channel_cfg(struct nicpf *nic, u8 vnic, u8 sq_idx)
 	nic_reg_write(nic, NIC_PF_TL2_0_63_PRI | (tl2 << 3), 0x00);
 }
 
+/* Send primary nicvf pointer to secondary QS's VF */
+static void nic_send_pnicvf(struct nicpf *nic, int sqs)
+{
+	union nic_mbx mbx = {};
+
+	mbx.nicvf.msg = NIC_MBOX_MSG_PNICVF_PTR;
+	mbx.nicvf.nicvf = nic->nicvf[nic->pqs_vf[sqs]];
+	nic_send_msg_to_vf(nic, sqs, &mbx);
+}
+
+/* Send SQS's nicvf pointer to primary QS's VF */
+static void nic_send_snicvf(struct nicpf *nic, struct nicvf_ptr *nicvf)
+{
+	union nic_mbx mbx = {};
+	int sqs_id = nic->vf_sqs[nicvf->vf_id][nicvf->sqs_id];
+
+	mbx.nicvf.msg = NIC_MBOX_MSG_SNICVF_PTR;
+	mbx.nicvf.sqs_id = nicvf->sqs_id;
+	mbx.nicvf.nicvf = nic->nicvf[sqs_id];
+	nic_send_msg_to_vf(nic, nicvf->vf_id, &mbx);
+}
+
+/* Find next available Qset that can be assigned as a
+ * secondary Qset to a VF.
+ */
+static int nic_nxt_avail_sqs(struct nicpf *nic)
+{
+	int sqs;
+
+	for (sqs = 0; sqs < nic->num_sqs_en; sqs++) {
+		if (!nic->sqs_used[sqs])
+			nic->sqs_used[sqs] = true;
+		else
+			continue;
+		return sqs + nic->num_vf_en;
+	}
+	return -1;
+}
+
+/* Allocate additional Qsets for requested VF */
+static void nic_alloc_sqs(struct nicpf *nic, struct sqs_alloc *sqs)
+{
+	union nic_mbx mbx = {};
+	int idx, alloc_qs = 0;
+	int sqs_id;
+
+	if (!nic->num_sqs_en)
+		goto send_mbox;
+
+	for (idx = 0; idx < sqs->qs_count; idx++) {
+		sqs_id = nic_nxt_avail_sqs(nic);
+		if (sqs_id < 0)
+			break;
+		nic->vf_sqs[sqs->vf_id][idx] = sqs_id;
+		nic->pqs_vf[sqs_id] = sqs->vf_id;
+		alloc_qs++;
+	}
+
+send_mbox:
+	mbx.sqs_alloc.msg = NIC_MBOX_MSG_ALLOC_SQS;
+	mbx.sqs_alloc.vf_id = sqs->vf_id;
+	mbx.sqs_alloc.qs_count = alloc_qs;
+	nic_send_msg_to_vf(nic, sqs->vf_id, &mbx);
+}
+
 /* Interrupt handler to handle mailbox messages from VFs */
 static void nic_handle_mbx_intr(struct nicpf *nic, int vf)
 {
@@ -496,6 +586,7 @@ static void nic_handle_mbx_intr(struct nicpf *nic, int vf)
 	u64 *mbx_data;
 	u64 mbx_addr;
 	u64 reg_addr;
+	u64 cfg;
 	int bgx, lmac;
 	int i;
 	int ret = 0;
@@ -516,15 +607,24 @@ static void nic_handle_mbx_intr(struct nicpf *nic, int vf)
 	switch (mbx.msg.msg) {
 	case NIC_MBOX_MSG_READY:
 		nic_mbx_send_ready(nic, vf);
+		if (vf < MAX_LMAC) {
 			nic->link[vf] = 0;
 			nic->duplex[vf] = 0;
 			nic->speed[vf] = 0;
+		}
 		ret = 1;
 		break;
 	case NIC_MBOX_MSG_QS_CFG:
 		reg_addr = NIC_PF_QSET_0_127_CFG |
 			   (mbx.qs.num << NIC_QS_ID_SHIFT);
-		nic_reg_write(nic, reg_addr, mbx.qs.cfg);
+		cfg = mbx.qs.cfg;
+		/* Check if its a secondary Qset */
+		if (vf >= nic->num_vf_en) {
+			cfg = cfg & (~0x7FULL);
+			/* Assign this Qset to primary Qset's VF */
+			cfg |= nic->pqs_vf[vf];
+		}
+		nic_reg_write(nic, reg_addr, cfg);
 		break;
 	case NIC_MBOX_MSG_RQ_CFG:
 		reg_addr = NIC_PF_QSET_0_127_RQ_0_7_CFG |
@@ -552,9 +652,11 @@ static void nic_handle_mbx_intr(struct nicpf *nic, int vf)
 			   (mbx.sq.qs_num << NIC_QS_ID_SHIFT) |
 			   (mbx.sq.sq_num << NIC_Q_NUM_SHIFT);
 		nic_reg_write(nic, reg_addr, mbx.sq.cfg);
-		nic_tx_channel_cfg(nic, mbx.qs.num, mbx.sq.sq_num);
+		nic_tx_channel_cfg(nic, mbx.qs.num, &mbx.sq);
 		break;
 	case NIC_MBOX_MSG_SET_MAC:
+		if (vf >= nic->num_vf_en)
+			break;
 		lmac = mbx.mac.vf_id;
 		bgx = NIC_GET_BGX_FROM_VF_LMAC_MAP(nic->vf_lmac_map[lmac]);
 		lmac = NIC_GET_LMAC_FROM_VF_LMAC_MAP(nic->vf_lmac_map[lmac]);
@@ -581,7 +683,22 @@ static void nic_handle_mbx_intr(struct nicpf *nic, int vf)
 	case NIC_MBOX_MSG_SHUTDOWN:
 		/* First msg in VF teardown sequence */
 		nic->vf_enabled[vf] = false;
+		if (vf >= nic->num_vf_en)
+			nic->sqs_used[vf - nic->num_vf_en] = false;
+		nic->pqs_vf[vf] = 0;
+		break;
+	case NIC_MBOX_MSG_ALLOC_SQS:
+		nic_alloc_sqs(nic, &mbx.sqs_alloc);
+		goto unlock;
+	case NIC_MBOX_MSG_NICVF_PTR:
+		nic->nicvf[vf] = mbx.nicvf.nicvf;
 		break;
+	case NIC_MBOX_MSG_PNICVF_PTR:
+		nic_send_pnicvf(nic, vf);
+		goto unlock;
+	case NIC_MBOX_MSG_SNICVF_PTR:
+		nic_send_snicvf(nic, &mbx.nicvf);
+		goto unlock;
 	case NIC_MBOX_MSG_BGX_STATS:
 		nic_get_bgx_stats(nic, &mbx.bgx_stats);
 		goto unlock;
@@ -610,8 +727,7 @@ static void nic_mbx_intr_handler (struct nicpf *nic, int mbx)
 		if (intr & (1ULL << vf)) {
 			dev_dbg(&nic->pdev->dev, "Intr from VF %d\n",
 				vf + (mbx * vf_per_mbx_reg));
-			if ((vf + (mbx * vf_per_mbx_reg)) > nic->num_vf_en)
-				break;
+
 			nic_handle_mbx_intr(nic, vf + (mbx * vf_per_mbx_reg));
 			nic_clear_mbx_intr(nic, vf, mbx);
 		}
@@ -717,9 +833,24 @@ static void nic_unregister_interrupts(struct nicpf *nic)
 	nic_disable_msix(nic);
 }
 
+static int nic_num_sqs_en(struct nicpf *nic, int vf_en)
+{
+	int pos, sqs_per_vf = MAX_SQS_PER_VF_SINGLE_NODE;
+	u16 total_vf;
+
+	/* Check if its a multi-node environment */
+	if (nr_node_ids > 1)
+		sqs_per_vf = MAX_SQS_PER_VF;
+
+	pos = pci_find_ext_capability(nic->pdev, PCI_EXT_CAP_ID_SRIOV);
+	pci_read_config_word(nic->pdev, (pos + PCI_SRIOV_TOTAL_VF), &total_vf);
+	return min(total_vf - vf_en, vf_en * sqs_per_vf);
+}
+
 static int nic_sriov_init(struct pci_dev *pdev, struct nicpf *nic)
 {
 	int pos = 0;
+	int vf_en;
 	int err;
 	u16 total_vf_cnt;
 
@@ -736,16 +867,20 @@ static int nic_sriov_init(struct pci_dev *pdev, struct nicpf *nic)
 	if (!total_vf_cnt)
 		return 0;
 
-	err = pci_enable_sriov(pdev, nic->num_vf_en);
+	vf_en = nic->num_vf_en;
+	nic->num_sqs_en = nic_num_sqs_en(nic, nic->num_vf_en);
+	vf_en += nic->num_sqs_en;
+
+	err = pci_enable_sriov(pdev, vf_en);
 	if (err) {
 		dev_err(&pdev->dev, "SRIOV enable failed, num VF is %d\n",
-			nic->num_vf_en);
+			vf_en);
 		nic->num_vf_en = 0;
 		return err;
 	}
 
 	dev_info(&pdev->dev, "SRIOV enabled, number of VF available %d\n",
-		 nic->num_vf_en);
+		 vf_en);
 
 	nic->flags |= NIC_SRIOV_ENABLED;
 	return 0;

drivers/net/ethernet/cavium/thunder/nicvf_ethtool.c

@@ -149,10 +149,33 @@ static void nicvf_set_msglevel(struct net_device *netdev, u32 lvl)
 	nic->msg_enable = lvl;
 }
 
+static void nicvf_get_qset_strings(struct nicvf *nic, u8 **data, int qset)
+{
+	int stats, qidx;
+	int start_qidx = qset * MAX_RCV_QUEUES_PER_QS;
+
+	for (qidx = 0; qidx < nic->qs->rq_cnt; qidx++) {
+		for (stats = 0; stats < nicvf_n_queue_stats; stats++) {
+			sprintf(*data, "rxq%d: %s", qidx + start_qidx,
+				nicvf_queue_stats[stats].name);
+			*data += ETH_GSTRING_LEN;
+		}
+	}
+
+	for (qidx = 0; qidx < nic->qs->sq_cnt; qidx++) {
+		for (stats = 0; stats < nicvf_n_queue_stats; stats++) {
+			sprintf(*data, "txq%d: %s", qidx + start_qidx,
+				nicvf_queue_stats[stats].name);
+			*data += ETH_GSTRING_LEN;
+		}
+	}
+}
+
 static void nicvf_get_strings(struct net_device *netdev, u32 sset, u8 *data)
 {
 	struct nicvf *nic = netdev_priv(netdev);
-	int stats, qidx;
+	int stats;
+	int sqs;
 
 	if (sset != ETH_SS_STATS)
 		return;
@@ -167,20 +190,12 @@ static void nicvf_get_strings(struct net_device *netdev, u32 sset, u8 *data)
 		data += ETH_GSTRING_LEN;
 	}
 
-	for (qidx = 0; qidx < nic->qs->rq_cnt; qidx++) {
-		for (stats = 0; stats < nicvf_n_queue_stats; stats++) {
-			sprintf(data, "rxq%d: %s", qidx,
-				nicvf_queue_stats[stats].name);
-			data += ETH_GSTRING_LEN;
-		}
-	}
+	nicvf_get_qset_strings(nic, &data, 0);
 
-	for (qidx = 0; qidx < nic->qs->sq_cnt; qidx++) {
-		for (stats = 0; stats < nicvf_n_queue_stats; stats++) {
-			sprintf(data, "txq%d: %s", qidx,
-				nicvf_queue_stats[stats].name);
-			data += ETH_GSTRING_LEN;
-		}
+	for (sqs = 0; sqs < nic->sqs_count; sqs++) {
+		if (!nic->snicvf[sqs])
+			continue;
+		nicvf_get_qset_strings(nic->snicvf[sqs], &data, sqs + 1);
 	}
 
 	for (stats = 0; stats < BGX_RX_STATS_COUNT; stats++) {
@@ -197,21 +212,58 @@ static void nicvf_get_strings(struct net_device *netdev, u32 sset, u8 *data)
 static int nicvf_get_sset_count(struct net_device *netdev, int sset)
 {
 	struct nicvf *nic = netdev_priv(netdev);
+	int qstats_count;
+	int sqs;
 
 	if (sset != ETH_SS_STATS)
 		return -EINVAL;
 
+	qstats_count = nicvf_n_queue_stats *
+		       (nic->qs->rq_cnt + nic->qs->sq_cnt);
+	for (sqs = 0; sqs < nic->sqs_count; sqs++) {
+		struct nicvf *snic;
+
+		snic = nic->snicvf[sqs];
+		if (!snic)
+			continue;
+		qstats_count += nicvf_n_queue_stats *
+				(snic->qs->rq_cnt + snic->qs->sq_cnt);
+	}
+
 	return nicvf_n_hw_stats + nicvf_n_drv_stats +
-		(nicvf_n_queue_stats *
-		 (nic->qs->rq_cnt + nic->qs->sq_cnt)) +
+		qstats_count +
 		BGX_RX_STATS_COUNT + BGX_TX_STATS_COUNT;
 }
 
+static void nicvf_get_qset_stats(struct nicvf *nic,
+				 struct ethtool_stats *stats, u64 **data)
+{
+	int stat, qidx;
+
+	if (!nic)
+		return;
+
+	for (qidx = 0; qidx < nic->qs->rq_cnt; qidx++) {
+		nicvf_update_rq_stats(nic, qidx);
+		for (stat = 0; stat < nicvf_n_queue_stats; stat++)
+			*((*data)++) = ((u64 *)&nic->qs->rq[qidx].stats)
+					[nicvf_queue_stats[stat].index];
+	}
+
+	for (qidx = 0; qidx < nic->qs->sq_cnt; qidx++) {
+		nicvf_update_sq_stats(nic, qidx);
+		for (stat = 0; stat < nicvf_n_queue_stats; stat++)
+			*((*data)++) = ((u64 *)&nic->qs->sq[qidx].stats)
+					[nicvf_queue_stats[stat].index];
+	}
+}
+
 static void nicvf_get_ethtool_stats(struct net_device *netdev,
 				    struct ethtool_stats *stats, u64 *data)
 {
 	struct nicvf *nic = netdev_priv(netdev);
-	int stat, qidx;
+	int stat;
+	int sqs;
 
 	nicvf_update_stats(nic);
 
@@ -225,16 +277,12 @@ static void nicvf_get_ethtool_stats(struct net_device *netdev,
 		*(data++) = ((u64 *)&nic->drv_stats)
 				[nicvf_drv_stats[stat].index];
 
-	for (qidx = 0; qidx < nic->qs->rq_cnt; qidx++) {
-		for (stat = 0; stat < nicvf_n_queue_stats; stat++)
-			*(data++) = ((u64 *)&nic->qs->rq[qidx].stats)
-					[nicvf_queue_stats[stat].index];
-	}
+	nicvf_get_qset_stats(nic, stats, &data);
 
-	for (qidx = 0; qidx < nic->qs->sq_cnt; qidx++) {
-		for (stat = 0; stat < nicvf_n_queue_stats; stat++)
-			*(data++) = ((u64 *)&nic->qs->sq[qidx].stats)
-					[nicvf_queue_stats[stat].index];
+	for (sqs = 0; sqs < nic->sqs_count; sqs++) {
+		if (!nic->snicvf[sqs])
+			continue;
+		nicvf_get_qset_stats(nic->snicvf[sqs], stats, &data);
 	}
 
 	for (stat = 0; stat < BGX_RX_STATS_COUNT; stat++)
@@ -393,7 +441,7 @@ static int nicvf_get_rxnfc(struct net_device *dev,
 
 	switch (info->cmd) {
 	case ETHTOOL_GRXRINGS:
-		info->data = nic->qs->rq_cnt;
+		info->data = nic->rx_queues;
 		ret = 0;
 		break;
 	case ETHTOOL_GRXFH:
@@ -556,11 +604,11 @@ static void nicvf_get_channels(struct net_device *dev,
 
 	memset(channel, 0, sizeof(*channel));
 
-	channel->max_rx = MAX_RCV_QUEUES_PER_QS;
-	channel->max_tx = MAX_SND_QUEUES_PER_QS;
+	channel->max_rx = nic->max_queues;
+	channel->max_tx = nic->max_queues;
 
-	channel->rx_count = nic->qs->rq_cnt;
-	channel->tx_count = nic->qs->sq_cnt;
+	channel->rx_count = nic->rx_queues;
+	channel->tx_count = nic->tx_queues;
 }
 
 /* Set no of Tx, Rx queues to be used */
@@ -570,22 +618,34 @@ static int nicvf_set_channels(struct net_device *dev,
 	struct nicvf *nic = netdev_priv(dev);
 	int err = 0;
 	bool if_up = netif_running(dev);
+	int cqcount;
 
 	if (!channel->rx_count || !channel->tx_count)
 		return -EINVAL;
-	if (channel->rx_count > MAX_RCV_QUEUES_PER_QS)
+	if (channel->rx_count > nic->max_queues)
 		return -EINVAL;
-	if (channel->tx_count > MAX_SND_QUEUES_PER_QS)
+	if (channel->tx_count > nic->max_queues)
 		return -EINVAL;
 
 	if (if_up)
 		nicvf_stop(dev);
 
-	nic->qs->rq_cnt = channel->rx_count;
-	nic->qs->sq_cnt = channel->tx_count;
+	cqcount = max(channel->rx_count, channel->tx_count);
+
+	if (cqcount > MAX_CMP_QUEUES_PER_QS) {
+		nic->sqs_count = roundup(cqcount, MAX_CMP_QUEUES_PER_QS);
+		nic->sqs_count = (nic->sqs_count / MAX_CMP_QUEUES_PER_QS) - 1;
+	} else {
+		nic->sqs_count = 0;
+	}
+
+	nic->qs->rq_cnt = min_t(u32, channel->rx_count, MAX_RCV_QUEUES_PER_QS);
+	nic->qs->sq_cnt = min_t(u32, channel->tx_count, MAX_SND_QUEUES_PER_QS);
 	nic->qs->cq_cnt = max(nic->qs->rq_cnt, nic->qs->sq_cnt);
 
-	err = nicvf_set_real_num_queues(dev, nic->qs->sq_cnt, nic->qs->rq_cnt);
+	nic->rx_queues = channel->rx_count;
+	nic->tx_queues = channel->tx_count;
+	err = nicvf_set_real_num_queues(dev, nic->tx_queues, nic->rx_queues);
 	if (err)
 		return err;
 
@@ -593,7 +653,7 @@ static int nicvf_set_channels(struct net_device *dev,
 		nicvf_open(dev);
 
 	netdev_info(dev, "Setting num Tx rings to %d, Rx rings to %d success\n",
-		    nic->qs->sq_cnt, nic->qs->rq_cnt);
+		    nic->tx_queues, nic->rx_queues);
 
 	return err;
 }

drivers/net/ethernet/cavium/thunder/nicvf_main.c

@@ -51,6 +51,14 @@ module_param(cpi_alg, int, S_IRUGO);
 MODULE_PARM_DESC(cpi_alg,
 		 "PFC algorithm (0=none, 1=VLAN, 2=VLAN16, 3=IP Diffserv)");
 
+static inline u8 nicvf_netdev_qidx(struct nicvf *nic, u8 qidx)
+{
+	if (nic->sqs_mode)
+		return qidx + ((nic->sqs_id + 1) * MAX_CMP_QUEUES_PER_QS);
+	else
+		return qidx;
+}
+
 static inline void nicvf_set_rx_frame_cnt(struct nicvf *nic,
 					  struct sk_buff *skb)
 {
@@ -193,6 +201,7 @@ static void  nicvf_handle_mbx_intr(struct nicvf *nic)
 		if (!nic->set_mac_pending)
 			ether_addr_copy(nic->netdev->dev_addr,
 					mbx.nic_cfg.mac_addr);
+		nic->sqs_mode = mbx.nic_cfg.sqs_mode;
 		nic->link_up = false;
 		nic->duplex = 0;
 		nic->speed = 0;
@@ -230,6 +239,26 @@ static void  nicvf_handle_mbx_intr(struct nicvf *nic)
 			netif_tx_stop_all_queues(nic->netdev);
 		}
 		break;
+	case NIC_MBOX_MSG_ALLOC_SQS:
+		nic->sqs_count = mbx.sqs_alloc.qs_count;
+		nic->pf_acked = true;
+		break;
+	case NIC_MBOX_MSG_SNICVF_PTR:
+		/* Primary VF: make note of secondary VF's pointer
+		 * to be used while packet transmission.
+		 */
+		nic->snicvf[mbx.nicvf.sqs_id] =
+			(struct nicvf *)mbx.nicvf.nicvf;
+		nic->pf_acked = true;
+		break;
+	case NIC_MBOX_MSG_PNICVF_PTR:
+		/* Secondary VF/Qset: make note of primary VF's pointer
+		 * to be used while packet reception, to handover packet
+		 * to primary VF's netdev.
+		 */
+		nic->pnicvf = (struct nicvf *)mbx.nicvf.nicvf;
+		nic->pf_acked = true;
+		break;
 	default:
 		netdev_err(nic->netdev,
 			   "Invalid message from PF, msg 0x%x\n", mbx.msg.msg);
@@ -338,11 +367,100 @@ static int nicvf_rss_init(struct nicvf *nic)
 
 	for (idx = 0; idx < rss->rss_size; idx++)
 		rss->ind_tbl[idx] = ethtool_rxfh_indir_default(idx,
-							       nic->qs->rq_cnt);
+							       nic->rx_queues);
 	nicvf_config_rss(nic);
 	return 1;
 }
 
+/* Request PF to allocate additional Qsets */
+static void nicvf_request_sqs(struct nicvf *nic)
+{
+	union nic_mbx mbx = {};
+	int sqs;
+	int sqs_count = nic->sqs_count;
+	int rx_queues = 0, tx_queues = 0;
+
+	/* Only primary VF should request */
+	if (nic->sqs_mode ||  !nic->sqs_count)
+		return;
+
+	mbx.sqs_alloc.msg = NIC_MBOX_MSG_ALLOC_SQS;
+	mbx.sqs_alloc.vf_id = nic->vf_id;
+	mbx.sqs_alloc.qs_count = nic->sqs_count;
+	if (nicvf_send_msg_to_pf(nic, &mbx)) {
+		/* No response from PF */
+		nic->sqs_count = 0;
+		return;
+	}
+
+	/* Return if no Secondary Qsets available */
+	if (!nic->sqs_count)
+		return;
+
+	if (nic->rx_queues > MAX_RCV_QUEUES_PER_QS)
+		rx_queues = nic->rx_queues - MAX_RCV_QUEUES_PER_QS;
+	if (nic->tx_queues > MAX_SND_QUEUES_PER_QS)
+		tx_queues = nic->tx_queues - MAX_SND_QUEUES_PER_QS;
+
+	/* Set no of Rx/Tx queues in each of the SQsets */
+	for (sqs = 0; sqs < nic->sqs_count; sqs++) {
+		mbx.nicvf.msg = NIC_MBOX_MSG_SNICVF_PTR;
+		mbx.nicvf.vf_id = nic->vf_id;
+		mbx.nicvf.sqs_id = sqs;
+		nicvf_send_msg_to_pf(nic, &mbx);
+
+		nic->snicvf[sqs]->sqs_id = sqs;
+		if (rx_queues > MAX_RCV_QUEUES_PER_QS) {
+			nic->snicvf[sqs]->qs->rq_cnt = MAX_RCV_QUEUES_PER_QS;
+			rx_queues -= MAX_RCV_QUEUES_PER_QS;
+		} else {
+			nic->snicvf[sqs]->qs->rq_cnt = rx_queues;
+			rx_queues = 0;
+		}
+
+		if (tx_queues > MAX_SND_QUEUES_PER_QS) {
+			nic->snicvf[sqs]->qs->sq_cnt = MAX_SND_QUEUES_PER_QS;
+			tx_queues -= MAX_SND_QUEUES_PER_QS;
+		} else {
+			nic->snicvf[sqs]->qs->sq_cnt = tx_queues;
+			tx_queues = 0;
+		}
+
+		nic->snicvf[sqs]->qs->cq_cnt =
+		max(nic->snicvf[sqs]->qs->rq_cnt, nic->snicvf[sqs]->qs->sq_cnt);
+
+		/* Initialize secondary Qset's queues and its interrupts */
+		nicvf_open(nic->snicvf[sqs]->netdev);
+	}
+
+	/* Update stack with actual Rx/Tx queue count allocated */
+	if (sqs_count != nic->sqs_count)
+		nicvf_set_real_num_queues(nic->netdev,
+					  nic->tx_queues, nic->rx_queues);
+}
+
+/* Send this Qset's nicvf pointer to PF.
+ * PF inturn sends primary VF's nicvf struct to secondary Qsets/VFs
+ * so that packets received by these Qsets can use primary VF's netdev
+ */
+static void nicvf_send_vf_struct(struct nicvf *nic)
+{
+	union nic_mbx mbx = {};
+
+	mbx.nicvf.msg = NIC_MBOX_MSG_NICVF_PTR;
+	mbx.nicvf.sqs_mode = nic->sqs_mode;
+	mbx.nicvf.nicvf = (u64)nic;
+	nicvf_send_msg_to_pf(nic, &mbx);
+}
+
+static void nicvf_get_primary_vf_struct(struct nicvf *nic)
+{
+	union nic_mbx mbx = {};
+
+	mbx.nicvf.msg = NIC_MBOX_MSG_PNICVF_PTR;
+	nicvf_send_msg_to_pf(nic, &mbx);
+}
+
 int nicvf_set_real_num_queues(struct net_device *netdev,
 			      int tx_queues, int rx_queues)
 {
@@ -453,6 +571,15 @@ static void nicvf_rcv_pkt_handler(struct net_device *netdev,
 	struct sk_buff *skb;
 	struct nicvf *nic = netdev_priv(netdev);
 	int err = 0;
+	int rq_idx;
+
+	rq_idx = nicvf_netdev_qidx(nic, cqe_rx->rq_idx);
+
+	if (nic->sqs_mode) {
+		/* Use primary VF's 'nicvf' struct */
+		nic = nic->pnicvf;
+		netdev = nic->netdev;
+	}
 
 	/* Check for errors */
 	err = nicvf_check_cqe_rx_errs(nic, cq, cqe_rx);
@@ -482,7 +609,7 @@ static void nicvf_rcv_pkt_handler(struct net_device *netdev,
 
 	nicvf_set_rxhash(netdev, cqe_rx, skb);
 
-	skb_record_rx_queue(skb, cqe_rx->rq_idx);
+	skb_record_rx_queue(skb, rq_idx);
 	if (netdev->hw_features & NETIF_F_RXCSUM) {
 		/* HW by default verifies TCP/UDP/SCTP checksums */
 		skb->ip_summed = CHECKSUM_UNNECESSARY;
@@ -578,8 +705,11 @@ static int nicvf_cq_intr_handler(struct net_device *netdev, u8 cq_idx,
 done:
 	/* Wakeup TXQ if its stopped earlier due to SQ full */
 	if (tx_done) {
-		txq = netdev_get_tx_queue(netdev, cq_idx);
-		if (netif_tx_queue_stopped(txq)) {
+		netdev = nic->pnicvf->netdev;
+		txq = netdev_get_tx_queue(netdev,
+					  nicvf_netdev_qidx(nic, cq_idx));
+		nic = nic->pnicvf;
+		if (netif_tx_queue_stopped(txq) && netif_carrier_ok(netdev)) {
 			netif_tx_start_queue(txq);
 			nic->drv_stats.txq_wake++;
 			if (netif_msg_tx_err(nic))
@@ -893,7 +1023,6 @@ static netdev_tx_t nicvf_xmit(struct sk_buff *skb, struct net_device *netdev)
 			netdev_warn(netdev,
 				    "%s: Transmit ring full, stopping SQ%d\n",
 				    netdev->name, qid);
-
 		return NETDEV_TX_BUSY;
 	}
 
@@ -926,6 +1055,17 @@ int nicvf_stop(struct net_device *netdev)
 	nicvf_send_msg_to_pf(nic, &mbx);
 
 	netif_carrier_off(netdev);
+	netif_tx_stop_all_queues(nic->netdev);
+
+	/* Teardown secondary qsets first */
+	if (!nic->sqs_mode) {
+		for (qidx = 0; qidx < nic->sqs_count; qidx++) {
+			if (!nic->snicvf[qidx])
+				continue;
+			nicvf_stop(nic->snicvf[qidx]->netdev);
+			nic->snicvf[qidx] = NULL;
+		}
+	}
 
 	/* Disable RBDR & QS error interrupts */
 	for (qidx = 0; qidx < qs->rbdr_cnt; qidx++) {
@@ -973,6 +1113,10 @@ int nicvf_stop(struct net_device *netdev)
 
 	nicvf_free_cq_poll(nic);
 
+	/* Clear multiqset info */
+	nic->pnicvf = nic;
+	nic->sqs_count = 0;
+
 	return 0;
 }
 
@@ -1028,9 +1172,15 @@ int nicvf_open(struct net_device *netdev)
 
 	/* Configure CPI alorithm */
 	nic->cpi_alg = cpi_alg;
+	if (!nic->sqs_mode)
 		nicvf_config_cpi(nic);
 
+	nicvf_request_sqs(nic);
+	if (nic->sqs_mode)
+		nicvf_get_primary_vf_struct(nic);
+
 	/* Configure receive side scaling */
+	if (!nic->sqs_mode)
 		nicvf_rss_init(nic);
 
 	err = nicvf_register_interrupts(nic);
@@ -1282,8 +1432,7 @@ static int nicvf_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
 	struct device *dev = &pdev->dev;
 	struct net_device *netdev;
 	struct nicvf *nic;
-	struct queue_set *qs;
-	int    err;
+	int    err, qcount;
 
 	err = pci_enable_device(pdev);
 	if (err) {
@@ -1309,9 +1458,17 @@ static int nicvf_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
 		goto err_release_regions;
 	}
 
-	netdev = alloc_etherdev_mqs(sizeof(struct nicvf),
-				    MAX_RCV_QUEUES_PER_QS,
-				    MAX_SND_QUEUES_PER_QS);
+	qcount = MAX_CMP_QUEUES_PER_QS;
+
+	/* Restrict multiqset support only for host bound VFs */
+	if (pdev->is_virtfn) {
+		/* Set max number of queues per VF */
+		qcount = roundup(num_online_cpus(), MAX_CMP_QUEUES_PER_QS);
+		qcount = min(qcount,
+			     (MAX_SQS_PER_VF + 1) * MAX_CMP_QUEUES_PER_QS);
+	}
+
+	netdev = alloc_etherdev_mqs(sizeof(struct nicvf), qcount, qcount);
 	if (!netdev) {
 		err = -ENOMEM;
 		goto err_release_regions;
@@ -1324,6 +1481,8 @@ static int nicvf_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
 	nic = netdev_priv(netdev);
 	nic->netdev = netdev;
 	nic->pdev = pdev;
+	nic->pnicvf = nic;
+	nic->max_queues = qcount;
 
 	/* MAP VF's configuration registers */
 	nic->reg_base = pcim_iomap(pdev, PCI_CFG_REG_BAR_NUM, 0);
@@ -1337,20 +1496,26 @@ static int nicvf_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
 	if (err)
 		goto err_free_netdev;
 
-	qs = nic->qs;
-
-	err = nicvf_set_real_num_queues(netdev, qs->sq_cnt, qs->rq_cnt);
-	if (err)
-		goto err_free_netdev;
-
 	/* Check if PF is alive and get MAC address for this VF */
 	err = nicvf_register_misc_interrupt(nic);
 	if (err)
 		goto err_free_netdev;
 
+	nicvf_send_vf_struct(nic);
+
+	/* Check if this VF is in QS only mode */
+	if (nic->sqs_mode)
+		return 0;
+
+	err = nicvf_set_real_num_queues(netdev, nic->tx_queues, nic->rx_queues);
+	if (err)
+		goto err_unregister_interrupts;
+
 	netdev->hw_features = (NETIF_F_RXCSUM | NETIF_F_IP_CSUM | NETIF_F_SG |
 			       NETIF_F_TSO | NETIF_F_GRO |
-			       NETIF_F_HW_VLAN_CTAG_RX | NETIF_F_RXHASH);
+			       NETIF_F_HW_VLAN_CTAG_RX);
+
+	netdev->hw_features |= NETIF_F_RXHASH;
 
 	netdev->features |= netdev->hw_features;
 
@@ -1389,8 +1554,13 @@ static void nicvf_remove(struct pci_dev *pdev)
 {
 	struct net_device *netdev = pci_get_drvdata(pdev);
 	struct nicvf *nic = netdev_priv(netdev);
+	struct net_device *pnetdev = nic->pnicvf->netdev;
 
-	unregister_netdev(netdev);
+	/* Check if this Qset is assigned to different VF.
+	 * If yes, clean primary and all secondary Qsets.
+	 */
+	if (pnetdev && (pnetdev->reg_state == NETREG_REGISTERED))
+		unregister_netdev(pnetdev);
 	nicvf_unregister_interrupts(nic);
 	pci_set_drvdata(pdev, NULL);
 	free_netdev(netdev);

drivers/net/ethernet/cavium/thunder/nicvf_queues.c

@@ -621,6 +621,7 @@ static void nicvf_snd_queue_config(struct nicvf *nic, struct queue_set *qs,
 	mbx.sq.msg = NIC_MBOX_MSG_SQ_CFG;
 	mbx.sq.qs_num = qs->vnic_id;
 	mbx.sq.sq_num = qidx;
+	mbx.sq.sqs_mode = nic->sqs_mode;
 	mbx.sq.cfg = (sq->cq_qs << 3) | sq->cq_idx;
 	nicvf_send_msg_to_pf(nic, &mbx);
 
@@ -702,6 +703,7 @@ void nicvf_qset_config(struct nicvf *nic, bool enable)
 	/* Send a mailbox msg to PF to config Qset */
 	mbx.qs.msg = NIC_MBOX_MSG_QS_CFG;
 	mbx.qs.num = qs->vnic_id;
+	mbx.qs.sqs_count = nic->sqs_count;
 
 	mbx.qs.cfg = 0;
 	qs_cfg = (struct qs_cfg *)&mbx.qs.cfg;
@@ -782,6 +784,10 @@ int nicvf_set_qset_resources(struct nicvf *nic)
 	qs->rbdr_len = RCV_BUF_COUNT;
 	qs->sq_len = SND_QUEUE_LEN;
 	qs->cq_len = CMP_QUEUE_LEN;
+
+	nic->rx_queues = qs->rq_cnt;
+	nic->tx_queues = qs->sq_cnt;
+
 	return 0;
 }
 
@@ -1025,7 +1031,7 @@ static inline void nicvf_sq_add_gather_subdesc(struct snd_queue *sq, int qentry,
  * them to SQ for transfer
  */
 static int nicvf_sq_append_tso(struct nicvf *nic, struct snd_queue *sq,
-			       int qentry, struct sk_buff *skb)
+			       int sq_num, int qentry, struct sk_buff *skb)
 {
 	struct tso_t tso;
 	int seg_subdescs = 0, desc_cnt = 0;
@@ -1085,7 +1091,7 @@ static int nicvf_sq_append_tso(struct nicvf *nic, struct snd_queue *sq,
 
 	/* Inform HW to xmit all TSO segments */
 	nicvf_queue_reg_write(nic, NIC_QSET_SQ_0_7_DOOR,
-			      skb_get_queue_mapping(skb), desc_cnt);
+			      sq_num, desc_cnt);
 	nic->drv_stats.tx_tso++;
 	return 1;
 }
@@ -1096,10 +1102,24 @@ int nicvf_sq_append_skb(struct nicvf *nic, struct sk_buff *skb)
 	int i, size;
 	int subdesc_cnt;
 	int sq_num, qentry;
-	struct queue_set *qs = nic->qs;
+	struct queue_set *qs;
 	struct snd_queue *sq;
 
 	sq_num = skb_get_queue_mapping(skb);
+	if (sq_num >= MAX_SND_QUEUES_PER_QS) {
+		/* Get secondary Qset's SQ structure */
+		i = sq_num / MAX_SND_QUEUES_PER_QS;
+		if (!nic->snicvf[i - 1]) {
+			netdev_warn(nic->netdev,
+				    "Secondary Qset#%d's ptr not initialized\n",
+				    i - 1);
+			return 1;
+		}
+		nic = (struct nicvf *)nic->snicvf[i - 1];
+		sq_num = sq_num % MAX_SND_QUEUES_PER_QS;
+	}
+
+	qs = nic->qs;
 	sq = &qs->sq[sq_num];
 
 	subdesc_cnt = nicvf_sq_subdesc_required(nic, skb);
@@ -1110,7 +1130,7 @@ int nicvf_sq_append_skb(struct nicvf *nic, struct sk_buff *skb)
 
 	/* Check if its a TSO packet */
 	if (skb_shinfo(skb)->gso_size)
-		return nicvf_sq_append_tso(nic, sq, qentry, skb);
+		return nicvf_sq_append_tso(nic, sq, sq_num, qentry, skb);
 
 	/* Add SQ header subdesc */
 	nicvf_sq_add_hdr_subdesc(sq, qentry, subdesc_cnt - 1, skb, skb->len);
@@ -1146,6 +1166,8 @@ int nicvf_sq_append_skb(struct nicvf *nic, struct sk_buff *skb)
 	return 1;
 
 append_fail:
+	/* Use original PCI dev for debug log */
+	nic = nic->pnicvf;
 	netdev_dbg(nic->netdev, "Not enough SQ descriptors to xmit pkt\n");
 	return 0;
 }