cxgb4/cxgb4vf: Add support for SGE doorbell queue timer

T6 introduced a Timer Mechanism in SGE called the
SGE Doorbell Queue Timer. With this we can now configure
TX Queues to get CIDX Updates when:

    Time(CIDX == PIDX) >= Timer

Previously we rely on TX Queue Status Page updates by hardware
for DMA completions. This will make Hardware/Firmware actually
deliver the CIDX Updates as Ingress Queue messages with
commensurate Interrupts.

So we now have a new RX Path component for processing CIDX Updates
and reclaiming TX Descriptors faster.

Original work by: Casey Leedom <leedom@chelsio.com>
Signed-off-by: Vishal Kulkarni <vishal@chelsio.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

drivers/net/ethernet/chelsio/cxgb4/cxgb4.h

@@ -617,6 +617,7 @@ enum {                                 /* adapter flags */
 	FW_OFLD_CONN       = (1 << 9),
 	ROOT_NO_RELAXED_ORDERING = (1 << 10),
 	SHUTTING_DOWN	   = (1 << 11),
+	SGE_DBQ_TIMER      = (1 << 12),
 };
 
 enum {
@@ -756,6 +757,8 @@ struct sge_eth_txq {                /* state for an SGE Ethernet Tx queue */
 #ifdef CONFIG_CHELSIO_T4_DCB
 	u8 dcb_prio;		    /* DCB Priority bound to queue */
 #endif
+	u8 dbqt;                    /* SGE Doorbell Queue Timer in use */
+	unsigned int dbqtimerix;    /* SGE Doorbell Queue Timer Index */
 	unsigned long tso;          /* # of TSO requests */
 	unsigned long tx_cso;       /* # of Tx checksum offloads */
 	unsigned long vlan_ins;     /* # of Tx VLAN insertions */
@@ -816,6 +819,7 @@ struct sge {
 	u16 nqs_per_uld;	    /* # of Rx queues per ULD */
 	u16 timer_val[SGE_NTIMERS];
 	u8 counter_val[SGE_NCOUNTERS];
+	u16 dbqtimer_val[SGE_NDBQTIMERS];
 	u32 fl_pg_order;            /* large page allocation size */
 	u32 stat_len;               /* length of status page at ring end */
 	u32 pktshift;               /* padding between CPL & packet data */
@@ -1402,7 +1406,7 @@ int t4_sge_alloc_rxq(struct adapter *adap, struct sge_rspq *iq, bool fwevtq,
 		     rspq_flush_handler_t flush_handler, int cong);
 int t4_sge_alloc_eth_txq(struct adapter *adap, struct sge_eth_txq *txq,
 			 struct net_device *dev, struct netdev_queue *netdevq,
-			 unsigned int iqid);
+			 unsigned int iqid, u8 dbqt);
 int t4_sge_alloc_ctrl_txq(struct adapter *adap, struct sge_ctrl_txq *txq,
 			  struct net_device *dev, unsigned int iqid,
 			  unsigned int cmplqid);
@@ -1415,6 +1419,8 @@ irqreturn_t t4_sge_intr_msix(int irq, void *cookie);
 int t4_sge_init(struct adapter *adap);
 void t4_sge_start(struct adapter *adap);
 void t4_sge_stop(struct adapter *adap);
+int t4_sge_eth_txq_egress_update(struct adapter *adap, struct sge_eth_txq *q,
+				 int maxreclaim);
 void cxgb4_set_ethtool_ops(struct net_device *netdev);
 int cxgb4_write_rss(const struct port_info *pi, const u16 *queues);
 enum cpl_tx_tnl_lso_type cxgb_encap_offload_supported(struct sk_buff *skb);
@@ -1821,6 +1827,8 @@ int t4_ctrl_eq_free(struct adapter *adap, unsigned int mbox, unsigned int pf,
 int t4_ofld_eq_free(struct adapter *adap, unsigned int mbox, unsigned int pf,
 		    unsigned int vf, unsigned int eqid);
 int t4_sge_ctxt_flush(struct adapter *adap, unsigned int mbox, int ctxt_type);
+int t4_read_sge_dbqtimers(struct adapter *adap, unsigned int ndbqtimers,
+			  u16 *dbqtimers);
 void t4_handle_get_port_info(struct port_info *pi, const __be64 *rpl);
 int t4_update_port_info(struct port_info *pi);
 int t4_get_link_params(struct port_info *pi, unsigned int *link_okp,

drivers/net/ethernet/chelsio/cxgb4/cxgb4_main.c

@@ -575,7 +575,7 @@ static int fwevtq_handler(struct sge_rspq *q, const __be64 *rsp,
 			struct sge_eth_txq *eq;
 
 			eq = container_of(txq, struct sge_eth_txq, q);
-			netif_tx_wake_queue(eq->txq);
+			t4_sge_eth_txq_egress_update(q->adap, eq, -1);
 		} else {
 			struct sge_uld_txq *oq;
 
@@ -933,10 +933,13 @@ static int setup_sge_queues(struct adapter *adap)
 			q->rspq.idx = j;
 			memset(&q->stats, 0, sizeof(q->stats));
 		}
-		for (j = 0; j < pi->nqsets; j++, t++) {
+
+		q = &s->ethrxq[pi->first_qset];
+		for (j = 0; j < pi->nqsets; j++, t++, q++) {
 			err = t4_sge_alloc_eth_txq(adap, t, dev,
 					netdev_get_tx_queue(dev, j),
-					s->fw_evtq.cntxt_id);
+					q->rspq.cntxt_id,
+					!!(adap->flags & SGE_DBQ_TIMER));
 			if (err)
 				goto freeout;
 		}
@@ -958,7 +961,7 @@ static int setup_sge_queues(struct adapter *adap)
 	if (!is_t4(adap->params.chip)) {
 		err = t4_sge_alloc_eth_txq(adap, &s->ptptxq, adap->port[0],
 					   netdev_get_tx_queue(adap->port[0], 0)
-					   , s->fw_evtq.cntxt_id);
+					   , s->fw_evtq.cntxt_id, false);
 		if (err)
 			goto freeout;
 	}
@@ -4325,6 +4328,14 @@ static int adap_init0(struct adapter *adap)
 	if (ret < 0)
 		goto bye;
 
+	/* Grab the SGE Doorbell Queue Timer values.  If successful, that
+	 * indicates that the Firmware and Hardware support this.
+	 */
+	ret = t4_read_sge_dbqtimers(adap, ARRAY_SIZE(adap->sge.dbqtimer_val),
+				    adap->sge.dbqtimer_val);
+	if (!ret)
+		adap->flags |= SGE_DBQ_TIMER;
+
 	if (is_bypass_device(adap->pdev->device))
 		adap->params.bypass = 1;
 

drivers/net/ethernet/chelsio/cxgb4/t4_hw.c

@@ -6712,6 +6712,47 @@ int t4_sge_ctxt_flush(struct adapter *adap, unsigned int mbox, int ctxt_type)
 	return ret;
 }
 
+/**
+ *	t4_read_sge_dbqtimers - reag SGE Doorbell Queue Timer values
+ *	@adap - the adapter
+ *	@ndbqtimers: size of the provided SGE Doorbell Queue Timer table
+ *	@dbqtimers: SGE Doorbell Queue Timer table
+ *
+ *	Reads the SGE Doorbell Queue Timer values into the provided table.
+ *	Returns 0 on success (Firmware and Hardware support this feature),
+ *	an error on failure.
+ */
+int t4_read_sge_dbqtimers(struct adapter *adap, unsigned int ndbqtimers,
+			  u16 *dbqtimers)
+{
+	int ret, dbqtimerix;
+
+	ret = 0;
+	dbqtimerix = 0;
+	while (dbqtimerix < ndbqtimers) {
+		int nparams, param;
+		u32 params[7], vals[7];
+
+		nparams = ndbqtimers - dbqtimerix;
+		if (nparams > ARRAY_SIZE(params))
+			nparams = ARRAY_SIZE(params);
+
+		for (param = 0; param < nparams; param++)
+			params[param] =
+			  (FW_PARAMS_MNEM_V(FW_PARAMS_MNEM_DEV) |
+			   FW_PARAMS_PARAM_X_V(FW_PARAMS_PARAM_DEV_DBQ_TIMER) |
+			   FW_PARAMS_PARAM_Y_V(dbqtimerix + param));
+		ret = t4_query_params(adap, adap->mbox, adap->pf, 0,
+				      nparams, params, vals);
+		if (ret)
+			break;
+
+		for (param = 0; param < nparams; param++)
+			dbqtimers[dbqtimerix++] = vals[param];
+	}
+	return ret;
+}
+
 /**
  *      t4_fw_hello - establish communication with FW
  *      @adap: the adapter

drivers/net/ethernet/chelsio/cxgb4/t4_hw.h

@@ -91,6 +91,7 @@ enum {
 	SGE_CTXT_SIZE = 24,       /* size of SGE context */
 	SGE_NTIMERS = 6,          /* # of interrupt holdoff timer values */
 	SGE_NCOUNTERS = 4,        /* # of interrupt packet counter values */
+	SGE_NDBQTIMERS = 8,       /* # of Doorbell Queue Timer values */
 	SGE_MAX_IQ_SIZE = 65520,
 
 	SGE_TIMER_RSTRT_CNTR = 6, /* restart RX packet threshold counter */

drivers/net/ethernet/chelsio/cxgb4/t4_values.h

@@ -71,12 +71,18 @@
 #define FETCHBURSTMIN_64B_X		2
 #define FETCHBURSTMIN_128B_X		3
 
+/* T6 and later use a single-bit encoding for FetchBurstMin */
+#define FETCHBURSTMIN_64B_T6_X		0
+#define FETCHBURSTMIN_128B_T6_X		1
+
 #define FETCHBURSTMAX_256B_X		2
 #define FETCHBURSTMAX_512B_X		3
 
+#define HOSTFCMODE_INGRESS_QUEUE_X	1
 #define HOSTFCMODE_STATUS_PAGE_X	2
 
 #define CIDXFLUSHTHRESH_32_X		5
+#define CIDXFLUSHTHRESH_128_X		7
 
 #define UPDATEDELIVERY_INTERRUPT_X	1
 

drivers/net/ethernet/chelsio/cxgb4/t4fw_api.h

@@ -1254,6 +1254,8 @@ enum fw_params_param_dev {
 	FW_PARAMS_PARAM_DEV_RDMA_WRITE_WITH_IMM = 0x21,
 	FW_PARAMS_PARAM_DEV_RI_WRITE_CMPL_WR    = 0x24,
 	FW_PARAMS_PARAM_DEV_OPAQUE_VIID_SMT_EXTN = 0x27,
+	FW_PARAMS_PARAM_DEV_DBQ_TIMER	= 0x29,
+	FW_PARAMS_PARAM_DEV_DBQ_TIMERTICK = 0x2A,
 };
 
 /*
@@ -1322,6 +1324,7 @@ enum fw_params_param_dmaq {
 	FW_PARAMS_PARAM_DMAQ_EQ_CMPLIQID_CTRL = 0x11,
 	FW_PARAMS_PARAM_DMAQ_EQ_SCHEDCLASS_ETH = 0x12,
 	FW_PARAMS_PARAM_DMAQ_EQ_DCBPRIO_ETH = 0x13,
+	FW_PARAMS_PARAM_DMAQ_EQ_TIMERIX	= 0x15,
 	FW_PARAMS_PARAM_DMAQ_CONM_CTXT = 0x20,
 };
 
@@ -1751,8 +1754,8 @@ struct fw_eq_eth_cmd {
 	__be32 fetchszm_to_iqid;
 	__be32 dcaen_to_eqsize;
 	__be64 eqaddr;
-	__be32 viid_pkd;
-	__be32 r8_lo;
+	__be32 autoequiqe_to_viid;
+	__be32 timeren_timerix;
 	__be64 r9;
 };
 
@@ -1847,6 +1850,10 @@ struct fw_eq_eth_cmd {
 #define FW_EQ_ETH_CMD_EQSIZE_S		0
 #define FW_EQ_ETH_CMD_EQSIZE_V(x)	((x) << FW_EQ_ETH_CMD_EQSIZE_S)
 
+#define FW_EQ_ETH_CMD_AUTOEQUIQE_S	31
+#define FW_EQ_ETH_CMD_AUTOEQUIQE_V(x)	((x) << FW_EQ_ETH_CMD_AUTOEQUIQE_S)
+#define FW_EQ_ETH_CMD_AUTOEQUIQE_F	FW_EQ_ETH_CMD_AUTOEQUIQE_V(1U)
+
 #define FW_EQ_ETH_CMD_AUTOEQUEQE_S	30
 #define FW_EQ_ETH_CMD_AUTOEQUEQE_V(x)	((x) << FW_EQ_ETH_CMD_AUTOEQUEQE_S)
 #define FW_EQ_ETH_CMD_AUTOEQUEQE_F	FW_EQ_ETH_CMD_AUTOEQUEQE_V(1U)
@@ -1854,6 +1861,19 @@ struct fw_eq_eth_cmd {
 #define FW_EQ_ETH_CMD_VIID_S	16
 #define FW_EQ_ETH_CMD_VIID_V(x)	((x) << FW_EQ_ETH_CMD_VIID_S)
 
+#define FW_EQ_ETH_CMD_TIMEREN_S		3
+#define FW_EQ_ETH_CMD_TIMEREN_M		0x1
+#define FW_EQ_ETH_CMD_TIMEREN_V(x)	((x) << FW_EQ_ETH_CMD_TIMEREN_S)
+#define FW_EQ_ETH_CMD_TIMEREN_G(x)	\
+    (((x) >> FW_EQ_ETH_CMD_TIMEREN_S) & FW_EQ_ETH_CMD_TIMEREN_M)
+#define FW_EQ_ETH_CMD_TIMEREN_F	FW_EQ_ETH_CMD_TIMEREN_V(1U)
+
+#define FW_EQ_ETH_CMD_TIMERIX_S		0
+#define FW_EQ_ETH_CMD_TIMERIX_M		0x7
+#define FW_EQ_ETH_CMD_TIMERIX_V(x)	((x) << FW_EQ_ETH_CMD_TIMERIX_S)
+#define FW_EQ_ETH_CMD_TIMERIX_G(x)	\
+    (((x) >> FW_EQ_ETH_CMD_TIMERIX_S) & FW_EQ_ETH_CMD_TIMERIX_M)
+
 struct fw_eq_ctrl_cmd {
 	__be32 op_to_vfn;
 	__be32 alloc_to_len16;

drivers/net/ethernet/chelsio/cxgb4vf/sge.c

@@ -2268,7 +2268,7 @@ int t4vf_sge_alloc_rxq(struct adapter *adapter, struct sge_rspq *rspq,
 	cmd.iqaddr = cpu_to_be64(rspq->phys_addr);
 
 	if (fl) {
-		enum chip_type chip =
+		unsigned int chip_ver =
 			CHELSIO_CHIP_VERSION(adapter->params.chip);
 		/*
 		 * Allocate the ring for the hardware free list (with space
@@ -2319,10 +2319,10 @@ int t4vf_sge_alloc_rxq(struct adapter *adapter, struct sge_rspq *rspq,
 		 */
 		cmd.fl0dcaen_to_fl0cidxfthresh =
 			cpu_to_be16(
-				FW_IQ_CMD_FL0FBMIN_V(chip <= CHELSIO_T5 ?
-						     FETCHBURSTMIN_128B_X :
-						     FETCHBURSTMIN_64B_X) |
-				FW_IQ_CMD_FL0FBMAX_V((chip <= CHELSIO_T5) ?
+				FW_IQ_CMD_FL0FBMIN_V(chip_ver <= CHELSIO_T5
+						     ? FETCHBURSTMIN_128B_X
+						     : FETCHBURSTMIN_64B_T6_X) |
+				FW_IQ_CMD_FL0FBMAX_V((chip_ver <= CHELSIO_T5) ?
 						     FETCHBURSTMAX_512B_X :
 						     FETCHBURSTMAX_256B_X));
 		cmd.fl0size = cpu_to_be16(flsz);
@@ -2411,10 +2411,11 @@ int t4vf_sge_alloc_eth_txq(struct adapter *adapter, struct sge_eth_txq *txq,
 			   struct net_device *dev, struct netdev_queue *devq,
 			   unsigned int iqid)
 {
+	unsigned int chip_ver = CHELSIO_CHIP_VERSION(adapter->params.chip);
+	struct port_info *pi = netdev_priv(dev);
+	struct fw_eq_eth_cmd cmd, rpl;
 	struct sge *s = &adapter->sge;
 	int ret, nentries;
-	struct fw_eq_eth_cmd cmd, rpl;
-	struct port_info *pi = netdev_priv(dev);
 
 	/*
 	 * Calculate the size of the hardware TX Queue (including the Status
@@ -2448,17 +2449,19 @@ int t4vf_sge_alloc_eth_txq(struct adapter *adapter, struct sge_eth_txq *txq,
 	cmd.alloc_to_len16 = cpu_to_be32(FW_EQ_ETH_CMD_ALLOC_F |
 					 FW_EQ_ETH_CMD_EQSTART_F |
 					 FW_LEN16(cmd));
-	cmd.viid_pkd = cpu_to_be32(FW_EQ_ETH_CMD_AUTOEQUEQE_F |
-				   FW_EQ_ETH_CMD_VIID_V(pi->viid));
+	cmd.autoequiqe_to_viid = cpu_to_be32(FW_EQ_ETH_CMD_AUTOEQUEQE_F |
+					     FW_EQ_ETH_CMD_VIID_V(pi->viid));
 	cmd.fetchszm_to_iqid =
 		cpu_to_be32(FW_EQ_ETH_CMD_HOSTFCMODE_V(SGE_HOSTFCMODE_STPG) |
 			    FW_EQ_ETH_CMD_PCIECHN_V(pi->port_id) |
 			    FW_EQ_ETH_CMD_IQID_V(iqid));
 	cmd.dcaen_to_eqsize =
-		cpu_to_be32(FW_EQ_ETH_CMD_FBMIN_V(SGE_FETCHBURSTMIN_64B) |
-			    FW_EQ_ETH_CMD_FBMAX_V(SGE_FETCHBURSTMAX_512B) |
+		cpu_to_be32(FW_EQ_ETH_CMD_FBMIN_V(chip_ver <= CHELSIO_T5
+						  ? FETCHBURSTMIN_64B_X
+						  : FETCHBURSTMIN_64B_T6_X) |
+			    FW_EQ_ETH_CMD_FBMAX_V(FETCHBURSTMAX_512B_X) |
 			    FW_EQ_ETH_CMD_CIDXFTHRESH_V(
-						SGE_CIDXFLUSHTHRESH_32) |
+						CIDXFLUSHTHRESH_32_X) |
 			    FW_EQ_ETH_CMD_EQSIZE_V(nentries));
 	cmd.eqaddr = cpu_to_be64(txq->q.phys_addr);