Merge branch 'cxgb4-next'

Hariprasad Shenai says:

====================
cxgb4/cxgb4vf: T5 BAR2 and ethtool related fixes

This series adds new interface to calculate BAR2 SGE queue register address for
cxgb4 and cxgb4vf driver and some more sge related fixes for T5. Also adds a
patch which updates the FW version displayed by ethtool after firmware flash.

The patches series is created against 'net-next' tree.
And includes patches on cxgb4 and cxgb4vf driver.

We have included all the maintainers of respective drivers. Kindly review the
change and let us know in case of any review comments.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>

11 files changed, 656 insertions, 151 deletions

diff --git a/drivers/net/ethernet/chelsio/cxgb4/cxgb4.h b/drivers/net/ethernet/chelsio/cxgb4/cxgb4.h
index e8b09bbdc226..c38a93607ea2 100644
--- a/drivers/net/ethernet/chelsio/cxgb4/cxgb4.h
+++ b/drivers/net/ethernet/chelsio/cxgb4/cxgb4.h
@@ -222,6 +222,12 @@ struct tp_err_stats {
 	u32 ofldCongDefer;
 };
 
+struct sge_params {
+	u32 hps;			/* host page size for our PF/VF */
+	u32 eq_qpp;			/* egress queues/page for our PF/VF */
+	u32 iq_qpp;			/* egress queues/page for our PF/VF */
+};
+
 struct tp_params {
 	unsigned int ntxchan;        /* # of Tx channels */
 	unsigned int tre;            /* log2 of core clocks per TP tick */
@@ -285,6 +291,7 @@ enum chip_type {
 };
 
 struct adapter_params {
+	struct sge_params sge;
 	struct tp_params  tp;
 	struct vpd_params vpd;
 	struct pci_params pci;
@@ -431,7 +438,8 @@ struct sge_fl {                     /* SGE free-buffer queue state */
 	struct rx_sw_desc *sdesc;   /* address of SW Rx descriptor ring */
 	__be64 *desc;               /* address of HW Rx descriptor ring */
 	dma_addr_t addr;            /* bus address of HW ring start */
-	u64 udb;                    /* BAR2 offset of User Doorbell area */
+	void __iomem *bar2_addr;    /* address of BAR2 Queue registers */
+	unsigned int bar2_qid;      /* Queue ID for BAR2 Queue registers */
 };
 
 /* A packet gather list */
@@ -461,7 +469,8 @@ struct sge_rspq {                   /* state for an SGE response queue */
 	u16 abs_id;                 /* absolute SGE id for the response q */
 	__be64 *desc;               /* address of HW response ring */
 	dma_addr_t phys_addr;       /* physical address of the ring */
-	u64 udb;                    /* BAR2 offset of User Doorbell area */
+	void __iomem *bar2_addr;    /* address of BAR2 Queue registers */
+	unsigned int bar2_qid;      /* Queue ID for BAR2 Queue registers */
 	unsigned int iqe_len;       /* entry size */
 	unsigned int size;          /* capacity of response queue */
 	struct adapter *adap;
@@ -519,7 +528,8 @@ struct sge_txq {
 	int db_disabled;
 	unsigned short db_pidx;
 	unsigned short db_pidx_inc;
-	u64 udb;                    /* BAR2 offset of User Doorbell area */
+	void __iomem *bar2_addr;    /* address of BAR2 Queue registers */
+	unsigned int bar2_qid;      /* Queue ID for BAR2 Queue registers */
 };
 
 struct sge_eth_txq {                /* state for an SGE Ethernet Tx queue */
@@ -995,6 +1005,15 @@ int t4_prep_fw(struct adapter *adap, struct fw_info *fw_info,
 	       const u8 *fw_data, unsigned int fw_size,
 	       struct fw_hdr *card_fw, enum dev_state state, int *reset);
 int t4_prep_adapter(struct adapter *adapter);
+
+enum t4_bar2_qtype { T4_BAR2_QTYPE_EGRESS, T4_BAR2_QTYPE_INGRESS };
+int t4_bar2_sge_qregs(struct adapter *adapter,
+		      unsigned int qid,
+		      enum t4_bar2_qtype qtype,
+		      u64 *pbar2_qoffset,
+		      unsigned int *pbar2_qid);
+
+int t4_init_sge_params(struct adapter *adapter);
 int t4_init_tp_params(struct adapter *adap);
 int t4_filter_field_shift(const struct adapter *adap, int filter_sel);
 int t4_port_init(struct adapter *adap, int mbox, int pf, int vf);
diff --git a/drivers/net/ethernet/chelsio/cxgb4/cxgb4_main.c b/drivers/net/ethernet/chelsio/cxgb4/cxgb4_main.c
index e7342bc85026..4c26be97fc9a 100644
--- a/drivers/net/ethernet/chelsio/cxgb4/cxgb4_main.c
+++ b/drivers/net/ethernet/chelsio/cxgb4/cxgb4_main.c
@@ -3805,6 +3805,22 @@ u64 cxgb4_read_sge_timestamp(struct net_device *dev)
 }
 EXPORT_SYMBOL(cxgb4_read_sge_timestamp);
 
+int cxgb4_bar2_sge_qregs(struct net_device *dev,
+			 unsigned int qid,
+			 enum cxgb4_bar2_qtype qtype,
+			 u64 *pbar2_qoffset,
+			 unsigned int *pbar2_qid)
+{
+	return t4_bar2_sge_qregs(netdev2adap(dev),
+				 qid,
+				 (qtype == CXGB4_BAR2_QTYPE_EGRESS
+				  ? T4_BAR2_QTYPE_EGRESS
+				  : T4_BAR2_QTYPE_INGRESS),
+				 pbar2_qoffset,
+				 pbar2_qid);
+}
+EXPORT_SYMBOL(cxgb4_bar2_sge_qregs);
+
 static struct pci_driver cxgb4_driver;
 
 static void check_neigh_update(struct neighbour *neigh)
@@ -3987,31 +4003,18 @@ static void process_db_drop(struct work_struct *work)
 		u32 dropped_db = t4_read_reg(adap, 0x010ac);
 		u16 qid = (dropped_db >> 15) & 0x1ffff;
 		u16 pidx_inc = dropped_db & 0x1fff;
-		unsigned int s_qpp;
-		unsigned short udb_density;
-		unsigned long qpshift;
-		int page;
-		u32 udb;
-
-		dev_warn(adap->pdev_dev,
-			 "Dropped DB 0x%x qid %d bar2 %d coalesce %d pidx %d\n",
-			 dropped_db, qid,
-			 (dropped_db >> 14) & 1,
-			 (dropped_db >> 13) & 1,
-			 pidx_inc);
-
-		drain_db_fifo(adap, 1);
+		u64 bar2_qoffset;
+		unsigned int bar2_qid;
+		int ret;
 
-		s_qpp = QUEUESPERPAGEPF1 * adap->fn;
-		udb_density = 1 << QUEUESPERPAGEPF0_GET(t4_read_reg(adap,
-				SGE_EGRESS_QUEUES_PER_PAGE_PF) >> s_qpp);
-		qpshift = PAGE_SHIFT - ilog2(udb_density);
-		udb = qid << qpshift;
-		udb &= PAGE_MASK;
-		page = udb / PAGE_SIZE;
-		udb += (qid - (page * udb_density)) * 128;
-
-		writel(PIDX(pidx_inc),  adap->bar2 + udb + 8);
+		ret = t4_bar2_sge_qregs(adap, qid, T4_BAR2_QTYPE_EGRESS,
+					&bar2_qoffset, &bar2_qid);
+		if (ret)
+			dev_err(adap->pdev_dev, "doorbell drop recovery: "
+				"qid=%d, pidx_inc=%d\n", qid, pidx_inc);
+		else
+			writel(PIDX_T5(pidx_inc) | QID(bar2_qid),
+			       adap->bar2 + bar2_qoffset + SGE_UDB_KDOORBELL);
 
 		/* Re-enable BAR2 WC */
 		t4_set_reg_field(adap, 0x10b0, 1<<15, 1<<15);
@@ -4069,12 +4072,8 @@ static void uld_attach(struct adapter *adap, unsigned int uld)
 	lli.adapter_type = adap->params.chip;
 	lli.iscsi_iolen = MAXRXDATA_GET(t4_read_reg(adap, TP_PARA_REG2));
 	lli.cclk_ps = 1000000000 / adap->params.vpd.cclk;
-	lli.udb_density = 1 << QUEUESPERPAGEPF0_GET(
-			t4_read_reg(adap, SGE_EGRESS_QUEUES_PER_PAGE_PF) >>
-			(adap->fn * 4));
-	lli.ucq_density = 1 << QUEUESPERPAGEPF0_GET(
-			t4_read_reg(adap, SGE_INGRESS_QUEUES_PER_PAGE_PF) >>
-			(adap->fn * 4));
+	lli.udb_density = 1 << adap->params.sge.eq_qpp;
+	lli.ucq_density = 1 << adap->params.sge.iq_qpp;
 	lli.filt_mode = adap->params.tp.vlan_pri_map;
 	/* MODQ_REQ_MAP sets queues 0-3 to chan 0-3 */
 	for (i = 0; i < NCHAN; i++)
@@ -5926,6 +5925,7 @@ static int adap_init0(struct adapter *adap)
 		t4_load_mtus(adap, adap->params.mtus, adap->params.a_wnd,
 			     adap->params.b_wnd);
 	}
+	t4_init_sge_params(adap);
 	t4_init_tp_params(adap);
 	adap->flags |= FW_OK;
 	return 0;
diff --git a/drivers/net/ethernet/chelsio/cxgb4/cxgb4_uld.h b/drivers/net/ethernet/chelsio/cxgb4/cxgb4_uld.h
index 4eba7cb1b89c..152b4c4c7809 100644
--- a/drivers/net/ethernet/chelsio/cxgb4/cxgb4_uld.h
+++ b/drivers/net/ethernet/chelsio/cxgb4/cxgb4_uld.h
@@ -305,4 +305,11 @@ void cxgb4_enable_db_coalescing(struct net_device *dev);
 int cxgb4_read_tpte(struct net_device *dev, u32 stag, __be32 *tpte);
 u64 cxgb4_read_sge_timestamp(struct net_device *dev);
 
+enum cxgb4_bar2_qtype { CXGB4_BAR2_QTYPE_EGRESS, CXGB4_BAR2_QTYPE_INGRESS };
+int cxgb4_bar2_sge_qregs(struct net_device *dev,
+			 unsigned int qid,
+			 enum cxgb4_bar2_qtype qtype,
+			 u64 *pbar2_qoffset,
+			 unsigned int *pbar2_qid);
+
 #endif  /* !__CXGB4_OFLD_H */
diff --git a/drivers/net/ethernet/chelsio/cxgb4/sge.c b/drivers/net/ethernet/chelsio/cxgb4/sge.c
index 433560b8cb1b..f12debd98dac 100644
--- a/drivers/net/ethernet/chelsio/cxgb4/sge.c
+++ b/drivers/net/ethernet/chelsio/cxgb4/sge.c
@@ -527,14 +527,16 @@ static inline void ring_fl_db(struct adapter *adap, struct sge_fl *q)
 		val |= DBPRIO(1);
 		wmb();
 
-		/* If we're on T4, use the old doorbell mechanism; otherwise
-		 * use the new BAR2 mechanism.
+		/* If we don't have access to the new User Doorbell (T5+), use
+		 * the old doorbell mechanism; otherwise use the new BAR2
+		 * mechanism.
 		 */
-		if (is_t4(adap->params.chip)) {
+		if (unlikely(q->bar2_addr == NULL)) {
 			t4_write_reg(adap, MYPF_REG(SGE_PF_KDOORBELL),
 				     val | QID(q->cntxt_id));
 		} else {
-			writel(val,  adap->bar2 + q->udb + SGE_UDB_KDOORBELL);
+			writel(val | QID(q->bar2_qid),
+			       q->bar2_addr + SGE_UDB_KDOORBELL);
 
 			/* This Write memory Barrier will force the write to
 			 * the User Doorbell area to be flushed.
@@ -850,14 +852,13 @@ static void write_sgl(const struct sk_buff *skb, struct sge_txq *q,
 		*end = 0;
 }
 
-/* This function copies a tx_desc struct to memory mapped BAR2 space(user space
- * writes). For coalesced WR SGE, fetches data from the FIFO instead of from
- * Host.
+/* This function copies 64 byte coalesced work request to
+ * memory mapped BAR2 space. For coalesced WR SGE fetches
+ * data from the FIFO instead of from Host.
  */
-static void cxgb_pio_copy(u64 __iomem *dst, struct tx_desc *desc)
+static void cxgb_pio_copy(u64 __iomem *dst, u64 *src)
 {
-	int count = sizeof(*desc) / sizeof(u64);
-	u64 *src = (u64 *)desc;
+	int count = 8;
 
 	while (count) {
 		writeq(*src, dst);
@@ -879,7 +880,10 @@ static inline void ring_tx_db(struct adapter *adap, struct sge_txq *q, int n)
 {
 	wmb();            /* write descriptors before telling HW */
 
-	if (is_t4(adap->params.chip)) {
+	/* If we don't have access to the new User Doorbell (T5+), use the old
+	 * doorbell mechanism; otherwise use the new BAR2 mechanism.
+	 */
+	if (unlikely(q->bar2_addr == NULL)) {
 		u32 val = PIDX(n);
 		unsigned long flags;
 
@@ -905,21 +909,22 @@ static inline void ring_tx_db(struct adapter *adap, struct sge_txq *q, int n)
 		 */
 		WARN_ON(val & DBPRIO(1));
 
-		/* For T5 and later we use the Write-Combine mapped BAR2 User
-		 * Doorbell mechanism.  If we're only writing a single TX
-		 * Descriptor and TX Write Combining hasn't been disabled, we
-		 * can use the Write Combining Gather Buffer; otherwise we use
-		 * the simple doorbell.
+		/* If we're only writing a single TX Descriptor and we can use
+		 * Inferred QID registers, we can use the Write Combining
+		 * Gather Buffer; otherwise we use the simple doorbell.
 		 */
-		if (n == 1) {
+		if (n == 1 && q->bar2_qid == 0) {
 			int index = (q->pidx
 				     ? (q->pidx - 1)
 				     : (q->size - 1));
+			u64 *wr = (u64 *)&q->desc[index];
 
-			cxgb_pio_copy(adap->bar2 + q->udb + SGE_UDB_WCDOORBELL,
-				      q->desc + index);
+			cxgb_pio_copy((u64 __iomem *)
+				      (q->bar2_addr + SGE_UDB_WCDOORBELL),
+				      wr);
 		} else {
-			writel(val,  adap->bar2 + q->udb + SGE_UDB_KDOORBELL);
+			writel(val | QID(q->bar2_qid),
+			       q->bar2_addr + SGE_UDB_KDOORBELL);
 		}
 
 		/* This Write Memory Barrier will force the write to the User
@@ -1997,11 +2002,16 @@ static int napi_rx_handler(struct napi_struct *napi, int budget)
 		params = QINTR_TIMER_IDX(7);
 
 	val = CIDXINC(work_done) | SEINTARM(params);
-	if (is_t4(q->adap->params.chip)) {
+
+	/* If we don't have access to the new User GTS (T5+), use the old
+	 * doorbell mechanism; otherwise use the new BAR2 mechanism.
+	 */
+	if (unlikely(q->bar2_addr == NULL)) {
 		t4_write_reg(q->adap, MYPF_REG(SGE_PF_GTS),
 			     val | INGRESSQID((u32)q->cntxt_id));
 	} else {
-		writel(val, q->adap->bar2 + q->udb + SGE_UDB_GTS);
+		writel(val | INGRESSQID(q->bar2_qid),
+		       q->bar2_addr + SGE_UDB_GTS);
 		wmb();
 	}
 	return work_done;
@@ -2047,11 +2057,16 @@ static unsigned int process_intrq(struct adapter *adap)
 	}
 
 	val =  CIDXINC(credits) | SEINTARM(q->intr_params);
-	if (is_t4(adap->params.chip)) {
+
+	/* If we don't have access to the new User GTS (T5+), use the old
+	 * doorbell mechanism; otherwise use the new BAR2 mechanism.
+	 */
+	if (unlikely(q->bar2_addr == NULL)) {
 		t4_write_reg(adap, MYPF_REG(SGE_PF_GTS),
 			     val | INGRESSQID(q->cntxt_id));
 	} else {
-		writel(val, adap->bar2 + q->udb + SGE_UDB_GTS);
+		writel(val | INGRESSQID(q->bar2_qid),
+		       q->bar2_addr + SGE_UDB_GTS);
 		wmb();
 	}
 	spin_unlock(&adap->sge.intrq_lock);
@@ -2235,48 +2250,32 @@ static void sge_tx_timer_cb(unsigned long data)
 }
 
 /**
- *      udb_address - return the BAR2 User Doorbell address for a Queue
- *      @adap: the adapter
- *      @cntxt_id: the Queue Context ID
- *      @qpp: Queues Per Page (for all PFs)
+ *	bar2_address - return the BAR2 address for an SGE Queue's Registers
+ *	@adapter: the adapter
+ *	@qid: the SGE Queue ID
+ *	@qtype: the SGE Queue Type (Egress or Ingress)
+ *	@pbar2_qid: BAR2 Queue ID or 0 for Queue ID inferred SGE Queues
  *
- *      Returns the BAR2 address of the user Doorbell associated with the
- *      indicated Queue Context ID.  Note that this is only applicable
- *      for T5 and later.
- */
-static u64 udb_address(struct adapter *adap, unsigned int cntxt_id,
-		       unsigned int qpp)
-{
-	u64 udb;
-	unsigned int s_qpp;
-	unsigned short udb_density;
-	unsigned long qpshift;
-	int page;
-
-	BUG_ON(is_t4(adap->params.chip));
-
-	s_qpp = (QUEUESPERPAGEPF0 +
-		(QUEUESPERPAGEPF1 - QUEUESPERPAGEPF0) * adap->fn);
-	udb_density = 1 << ((qpp >> s_qpp) & QUEUESPERPAGEPF0_MASK);
-	qpshift = PAGE_SHIFT - ilog2(udb_density);
-	udb = (u64)cntxt_id << qpshift;
-	udb &= PAGE_MASK;
-	page = udb / PAGE_SIZE;
-	udb += (cntxt_id - (page * udb_density)) * SGE_UDB_SIZE;
-
-	return udb;
-}
+ *	Returns the BAR2 address for the SGE Queue Registers associated with
+ *	@qid.  If BAR2 SGE Registers aren't available, returns NULL.  Also
+ *	returns the BAR2 Queue ID to be used with writes to the BAR2 SGE
+ *	Queue Registers.  If the BAR2 Queue ID is 0, then "Inferred Queue ID"
+ *	Registers are supported (e.g. the Write Combining Doorbell Buffer).
+ */
+static void __iomem *bar2_address(struct adapter *adapter,
+				  unsigned int qid,
+				  enum t4_bar2_qtype qtype,
+				  unsigned int *pbar2_qid)
+{
+	u64 bar2_qoffset;
+	int ret;
 
-static u64 udb_address_eq(struct adapter *adap, unsigned int cntxt_id)
-{
-	return udb_address(adap, cntxt_id,
-			   t4_read_reg(adap, SGE_EGRESS_QUEUES_PER_PAGE_PF));
-}
+	ret = t4_bar2_sge_qregs(adapter, qid, qtype,
+				&bar2_qoffset, pbar2_qid);
+	if (ret)
+		return NULL;
 
-static u64 udb_address_iq(struct adapter *adap, unsigned int cntxt_id)
-{
-	return udb_address(adap, cntxt_id,
-			   t4_read_reg(adap, SGE_INGRESS_QUEUES_PER_PAGE_PF));
+	return adapter->bar2 + bar2_qoffset;
 }
 
 int t4_sge_alloc_rxq(struct adapter *adap, struct sge_rspq *iq, bool fwevtq,
@@ -2344,8 +2343,10 @@ int t4_sge_alloc_rxq(struct adapter *adap, struct sge_rspq *iq, bool fwevtq,
 	iq->next_intr_params = iq->intr_params;
 	iq->cntxt_id = ntohs(c.iqid);
 	iq->abs_id = ntohs(c.physiqid);
-	if (!is_t4(adap->params.chip))
-		iq->udb = udb_address_iq(adap, iq->cntxt_id);
+	iq->bar2_addr = bar2_address(adap,
+				     iq->cntxt_id,
+				     T4_BAR2_QTYPE_INGRESS,
+				     &iq->bar2_qid);
 	iq->size--;                           /* subtract status entry */
 	iq->netdev = dev;
 	iq->handler = hnd;
@@ -2362,11 +2363,13 @@ int t4_sge_alloc_rxq(struct adapter *adap, struct sge_rspq *iq, bool fwevtq,
 		fl->alloc_failed = fl->large_alloc_failed = fl->starving = 0;
 		adap->sge.egr_map[fl->cntxt_id - adap->sge.egr_start] = fl;
 
-		/* Note, we must initialize the Free List User Doorbell
-		 * address before refilling the Free List!
+		/* Note, we must initialize the BAR2 Free List User Doorbell
+		 * information before refilling the Free List!
 		 */
-		if (!is_t4(adap->params.chip))
-			fl->udb = udb_address_eq(adap, fl->cntxt_id);
+		fl->bar2_addr = bar2_address(adap,
+					     fl->cntxt_id,
+					     T4_BAR2_QTYPE_EGRESS,
+					     &fl->bar2_qid);
 		refill_fl(adap, fl, fl_cap(fl), GFP_KERNEL);
 	}
 	return 0;
@@ -2392,9 +2395,10 @@ err:
 static void init_txq(struct adapter *adap, struct sge_txq *q, unsigned int id)
 {
 	q->cntxt_id = id;
-	if (!is_t4(adap->params.chip))
-		q->udb = udb_address_eq(adap, q->cntxt_id);
-
+	q->bar2_addr = bar2_address(adap,
+				    q->cntxt_id,
+				    T4_BAR2_QTYPE_EGRESS,
+				    &q->bar2_qid);
 	q->in_use = 0;
 	q->cidx = q->pidx = 0;
 	q->stops = q->restarts = 0;
diff --git a/drivers/net/ethernet/chelsio/cxgb4/t4_hw.c b/drivers/net/ethernet/chelsio/cxgb4/t4_hw.c
index 7975d26f50df..a9323bdb3585 100644
--- a/drivers/net/ethernet/chelsio/cxgb4/t4_hw.c
+++ b/drivers/net/ethernet/chelsio/cxgb4/t4_hw.c
@@ -1213,6 +1213,8 @@ out:
 	if (ret)
 		dev_err(adap->pdev_dev, "firmware download failed, error %d\n",
 			ret);
+	else
+		ret = t4_get_fw_version(adap, &adap->params.fw_vers);
 	return ret;
 }
 
@@ -4003,6 +4005,126 @@ int t4_prep_adapter(struct adapter *adapter)
 }
 
 /**
+ *	t4_bar2_sge_qregs - return BAR2 SGE Queue register information
+ *	@adapter: the adapter
+ *	@qid: the Queue ID
+ *	@qtype: the Ingress or Egress type for @qid
+ *	@pbar2_qoffset: BAR2 Queue Offset
+ *	@pbar2_qid: BAR2 Queue ID or 0 for Queue ID inferred SGE Queues
+ *
+ *	Returns the BAR2 SGE Queue Registers information associated with the
+ *	indicated Absolute Queue ID.  These are passed back in return value
+ *	pointers.  @qtype should be T4_BAR2_QTYPE_EGRESS for Egress Queue
+ *	and T4_BAR2_QTYPE_INGRESS for Ingress Queues.
+ *
+ *	This may return an error which indicates that BAR2 SGE Queue
+ *	registers aren't available.  If an error is not returned, then the
+ *	following values are returned:
+ *
+ *	  *@pbar2_qoffset: the BAR2 Offset of the @qid Registers
+ *	  *@pbar2_qid: the BAR2 SGE Queue ID or 0 of @qid
+ *
+ *	If the returned BAR2 Queue ID is 0, then BAR2 SGE registers which
+ *	require the "Inferred Queue ID" ability may be used.  E.g. the
+ *	Write Combining Doorbell Buffer. If the BAR2 Queue ID is not 0,
+ *	then these "Inferred Queue ID" register may not be used.
+ */
+int t4_bar2_sge_qregs(struct adapter *adapter,
+		      unsigned int qid,
+		      enum t4_bar2_qtype qtype,
+		      u64 *pbar2_qoffset,
+		      unsigned int *pbar2_qid)
+{
+	unsigned int page_shift, page_size, qpp_shift, qpp_mask;
+	u64 bar2_page_offset, bar2_qoffset;
+	unsigned int bar2_qid, bar2_qid_offset, bar2_qinferred;
+
+	/* T4 doesn't support BAR2 SGE Queue registers.
+	 */
+	if (is_t4(adapter->params.chip))
+		return -EINVAL;
+
+	/* Get our SGE Page Size parameters.
+	 */
+	page_shift = adapter->params.sge.hps + 10;
+	page_size = 1 << page_shift;
+
+	/* Get the right Queues per Page parameters for our Queue.
+	 */
+	qpp_shift = (qtype == T4_BAR2_QTYPE_EGRESS
+		     ? adapter->params.sge.eq_qpp
+		     : adapter->params.sge.iq_qpp);
+	qpp_mask = (1 << qpp_shift) - 1;
+
+	/*  Calculate the basics of the BAR2 SGE Queue register area:
+	 *  o The BAR2 page the Queue registers will be in.
+	 *  o The BAR2 Queue ID.
+	 *  o The BAR2 Queue ID Offset into the BAR2 page.
+	 */
+	bar2_page_offset = ((qid >> qpp_shift) << page_shift);
+	bar2_qid = qid & qpp_mask;
+	bar2_qid_offset = bar2_qid * SGE_UDB_SIZE;
+
+	/* If the BAR2 Queue ID Offset is less than the Page Size, then the
+	 * hardware will infer the Absolute Queue ID simply from the writes to
+	 * the BAR2 Queue ID Offset within the BAR2 Page (and we need to use a
+	 * BAR2 Queue ID of 0 for those writes).  Otherwise, we'll simply
+	 * write to the first BAR2 SGE Queue Area within the BAR2 Page with
+	 * the BAR2 Queue ID and the hardware will infer the Absolute Queue ID
+	 * from the BAR2 Page and BAR2 Queue ID.
+	 *
+	 * One important censequence of this is that some BAR2 SGE registers
+	 * have a "Queue ID" field and we can write the BAR2 SGE Queue ID
+	 * there.  But other registers synthesize the SGE Queue ID purely
+	 * from the writes to the registers -- the Write Combined Doorbell
+	 * Buffer is a good example.  These BAR2 SGE Registers are only
+	 * available for those BAR2 SGE Register areas where the SGE Absolute
+	 * Queue ID can be inferred from simple writes.
+	 */
+	bar2_qoffset = bar2_page_offset;
+	bar2_qinferred = (bar2_qid_offset < page_size);
+	if (bar2_qinferred) {
+		bar2_qoffset += bar2_qid_offset;
+		bar2_qid = 0;
+	}
+
+	*pbar2_qoffset = bar2_qoffset;
+	*pbar2_qid = bar2_qid;
+	return 0;
+}
+
+/**
+ *	t4_init_sge_params - initialize adap->params.sge
+ *	@adapter: the adapter
+ *
+ *	Initialize various fields of the adapter's SGE Parameters structure.
+ */
+int t4_init_sge_params(struct adapter *adapter)
+{
+	struct sge_params *sge_params = &adapter->params.sge;
+	u32 hps, qpp;
+	unsigned int s_hps, s_qpp;
+
+	/* Extract the SGE Page Size for our PF.
+	 */
+	hps = t4_read_reg(adapter, SGE_HOST_PAGE_SIZE);
+	s_hps = (HOSTPAGESIZEPF0_S +
+		 (HOSTPAGESIZEPF1_S - HOSTPAGESIZEPF0_S) * adapter->fn);
+	sge_params->hps = ((hps >> s_hps) & HOSTPAGESIZEPF0_M);
+
+	/* Extract the SGE Egress and Ingess Queues Per Page for our PF.
+	 */
+	s_qpp = (QUEUESPERPAGEPF0_S +
+		(QUEUESPERPAGEPF1_S - QUEUESPERPAGEPF0_S) * adapter->fn);
+	qpp = t4_read_reg(adapter, SGE_EGRESS_QUEUES_PER_PAGE_PF);
+	sge_params->eq_qpp = ((qpp >> s_qpp) & QUEUESPERPAGEPF0_MASK);
+	qpp = t4_read_reg(adapter, SGE_INGRESS_QUEUES_PER_PAGE_PF);
+	sge_params->iq_qpp = ((qpp >> s_qpp) & QUEUESPERPAGEPF0_MASK);
+
+	return 0;
+}
+
+/**
  *      t4_init_tp_params - initialize adap->params.tp
  *      @adap: the adapter
  *
diff --git a/drivers/net/ethernet/chelsio/cxgb4/t4_regs.h b/drivers/net/ethernet/chelsio/cxgb4/t4_regs.h
index ccdf8a7f4916..d7bd34ee65bd 100644
--- a/drivers/net/ethernet/chelsio/cxgb4/t4_regs.h
+++ b/drivers/net/ethernet/chelsio/cxgb4/t4_regs.h
@@ -155,17 +155,22 @@
 #define  HOSTPAGESIZEPF2_SHIFT  8
 #define  HOSTPAGESIZEPF2(x)     ((x) << HOSTPAGESIZEPF2_SHIFT)
 
-#define  HOSTPAGESIZEPF1_MASK   0x0000000fU
-#define  HOSTPAGESIZEPF1_SHIFT  4
-#define  HOSTPAGESIZEPF1(x)     ((x) << HOSTPAGESIZEPF1_SHIFT)
+#define  HOSTPAGESIZEPF1_M	0x0000000fU
+#define  HOSTPAGESIZEPF1_S	4
+#define  HOSTPAGESIZEPF1(x)     ((x) << HOSTPAGESIZEPF1_S)
 
-#define  HOSTPAGESIZEPF0_MASK   0x0000000fU
-#define  HOSTPAGESIZEPF0_SHIFT  0
-#define  HOSTPAGESIZEPF0(x)     ((x) << HOSTPAGESIZEPF0_SHIFT)
+#define  HOSTPAGESIZEPF0_M	0x0000000fU
+#define  HOSTPAGESIZEPF0_S	0
+#define  HOSTPAGESIZEPF0(x)     ((x) << HOSTPAGESIZEPF0_S)
 
 #define SGE_EGRESS_QUEUES_PER_PAGE_PF 0x1010
-#define  QUEUESPERPAGEPF0_MASK   0x0000000fU
-#define  QUEUESPERPAGEPF0_GET(x) ((x) & QUEUESPERPAGEPF0_MASK)
+#define SGE_EGRESS_QUEUES_PER_PAGE_VF_A 0x1014
+
+#define QUEUESPERPAGEPF1_S    4
+
+#define QUEUESPERPAGEPF0_S    0
+#define QUEUESPERPAGEPF0_MASK   0x0000000fU
+#define QUEUESPERPAGEPF0_GET(x) ((x) & QUEUESPERPAGEPF0_MASK)
 
 #define QUEUESPERPAGEPF0    0
 #define QUEUESPERPAGEPF1    4
@@ -323,6 +328,7 @@
 #define SGE_DEBUG_DATA_LOW_INDEX_3	0x12cc
 #define SGE_DEBUG_DATA_HIGH_INDEX_10	0x12a8
 #define SGE_INGRESS_QUEUES_PER_PAGE_PF 0x10f4
+#define SGE_INGRESS_QUEUES_PER_PAGE_VF_A 0x10f8
 
 #define S_HP_INT_THRESH    28
 #define M_HP_INT_THRESH 0xfU
diff --git a/drivers/net/ethernet/chelsio/cxgb4vf/adapter.h b/drivers/net/ethernet/chelsio/cxgb4vf/adapter.h
index 3d06e77d7121..d00a751f0588 100644
--- a/drivers/net/ethernet/chelsio/cxgb4vf/adapter.h
+++ b/drivers/net/ethernet/chelsio/cxgb4vf/adapter.h
@@ -138,6 +138,8 @@ struct sge_fl {
 	struct rx_sw_desc *sdesc;	/* address of SW RX descriptor ring */
 	__be64 *desc;			/* address of HW RX descriptor ring */
 	dma_addr_t addr;		/* PCI bus address of hardware ring */
+	void __iomem *bar2_addr;	/* address of BAR2 Queue registers */
+	unsigned int bar2_qid;		/* Queue ID for BAR2 Queue registers */
 };
 
 /*
@@ -178,6 +180,8 @@ struct sge_rspq {
 	u16 abs_id;			/* SGE abs QID for the response Q */
 	__be64 *desc;			/* address of hardware response ring */
 	dma_addr_t phys_addr;		/* PCI bus address of ring */
+	void __iomem *bar2_addr;	/* address of BAR2 Queue registers */
+	unsigned int bar2_qid;		/* Queue ID for BAR2 Queue registers */
 	unsigned int iqe_len;		/* entry size */
 	unsigned int size;		/* capcity of response Q */
 	struct adapter *adapter;	/* our adapter */
@@ -240,6 +244,8 @@ struct sge_txq {
 	struct tx_sw_desc *sdesc;	/* address of SW TX descriptor ring */
 	struct sge_qstat *stat;		/* queue status entry */
 	dma_addr_t phys_addr;		/* PCI bus address of hardware ring */
+	void __iomem *bar2_addr;	/* address of BAR2 Queue registers */
+	unsigned int bar2_qid;		/* Queue ID for BAR2 Queue registers */
 };
 
 /*
@@ -345,6 +351,7 @@ struct sge {
 struct adapter {
 	/* PCI resources */
 	void __iomem *regs;
+	void __iomem *bar2;
 	struct pci_dev *pdev;
 	struct device *pdev_dev;
 
diff --git a/drivers/net/ethernet/chelsio/cxgb4vf/cxgb4vf_main.c b/drivers/net/ethernet/chelsio/cxgb4vf/cxgb4vf_main.c
index ad88246a428e..aa74ec34a467 100644
--- a/drivers/net/ethernet/chelsio/cxgb4vf/cxgb4vf_main.c
+++ b/drivers/net/ethernet/chelsio/cxgb4vf/cxgb4vf_main.c
@@ -2095,7 +2095,6 @@ static int adap_init0(struct adapter *adapter)
 	unsigned int ethqsets;
 	int err;
 	u32 param, val = 0;
-	unsigned int chipid;
 
 	/*
 	 * Wait for the device to become ready before proceeding ...
@@ -2123,17 +2122,6 @@ static int adap_init0(struct adapter *adapter)
 		return err;
 	}
 
-	adapter->params.chip = 0;
-	switch (adapter->pdev->device >> 12) {
-	case CHELSIO_T4:
-		adapter->params.chip = CHELSIO_CHIP_CODE(CHELSIO_T4, 0);
-		break;
-	case CHELSIO_T5:
-		chipid = G_REV(t4_read_reg(adapter, A_PL_VF_REV));
-		adapter->params.chip |= CHELSIO_CHIP_CODE(CHELSIO_T5, chipid);
-		break;
-	}
-
 	/*
 	 * Grab basic operational parameters.  These will predominantly have
 	 * been set up by the Physical Function Driver or will be hard coded
@@ -2594,6 +2582,27 @@ static int cxgb4vf_pci_probe(struct pci_dev *pdev,
 		goto err_free_adapter;
 	}
 
+	/* Wait for the device to become ready before proceeding ...
+	 */
+	err = t4vf_prep_adapter(adapter);
+	if (err) {
+		dev_err(adapter->pdev_dev, "device didn't become ready:"
+			" err=%d\n", err);
+		goto err_unmap_bar0;
+	}
+
+	/* For T5 and later we want to use the new BAR-based User Doorbells,
+	 * so we need to map BAR2 here ...
+	 */
+	if (!is_t4(adapter->params.chip)) {
+		adapter->bar2 = ioremap_wc(pci_resource_start(pdev, 2),
+					   pci_resource_len(pdev, 2));
+		if (!adapter->bar2) {
+			dev_err(adapter->pdev_dev, "cannot map BAR2 doorbells\n");
+			err = -ENOMEM;
+			goto err_unmap_bar0;
+		}
+	}
 	/*
 	 * Initialize adapter level features.
 	 */
@@ -2786,6 +2795,10 @@ err_free_dev:
 	}
 
 err_unmap_bar:
+	if (!is_t4(adapter->params.chip))
+		iounmap(adapter->bar2);
+
+err_unmap_bar0:
 	iounmap(adapter->regs);
 
 err_free_adapter:
@@ -2856,6 +2869,8 @@ static void cxgb4vf_pci_remove(struct pci_dev *pdev)
 			free_netdev(netdev);
 		}
 		iounmap(adapter->regs);
+		if (!is_t4(adapter->params.chip))
+			iounmap(adapter->bar2);
 		kfree(adapter);
 	}
 
diff --git a/drivers/net/ethernet/chelsio/cxgb4vf/sge.c b/drivers/net/ethernet/chelsio/cxgb4vf/sge.c
index 045301d336bb..f7fd1317d996 100644
--- a/drivers/net/ethernet/chelsio/cxgb4vf/sge.c
+++ b/drivers/net/ethernet/chelsio/cxgb4vf/sge.c
@@ -525,19 +525,40 @@ static inline void ring_fl_db(struct adapter *adapter, struct sge_fl *fl)
 {
 	u32 val;
 
-	/*
-	 * The SGE keeps track of its Producer and Consumer Indices in terms
+	/* The SGE keeps track of its Producer and Consumer Indices in terms
 	 * of Egress Queue Units so we can only tell it about integral numbers
 	 * of multiples of Free List Entries per Egress Queue Units ...
 	 */
 	if (fl->pend_cred >= FL_PER_EQ_UNIT) {
-		val = PIDX(fl->pend_cred / FL_PER_EQ_UNIT);
-		if (!is_t4(adapter->params.chip))
-			val |= DBTYPE(1);
+		if (is_t4(adapter->params.chip))
+			val = PIDX(fl->pend_cred / FL_PER_EQ_UNIT);
+		else
+			val = PIDX_T5(fl->pend_cred / FL_PER_EQ_UNIT) |
+			      DBTYPE(1);
+		val |= DBPRIO(1);
+
+		/* Make sure all memory writes to the Free List queue are
+		 * committed before we tell the hardware about them.
+		 */
 		wmb();
-		t4_write_reg(adapter, T4VF_SGE_BASE_ADDR + SGE_VF_KDOORBELL,
-			     DBPRIO(1) |
-			     QID(fl->cntxt_id) | val);
+
+		/* If we don't have access to the new User Doorbell (T5+), use
+		 * the old doorbell mechanism; otherwise use the new BAR2
+		 * mechanism.
+		 */
+		if (unlikely(fl->bar2_addr == NULL)) {
+			t4_write_reg(adapter,
+				     T4VF_SGE_BASE_ADDR + SGE_VF_KDOORBELL,
+				     QID(fl->cntxt_id) | val);
+		} else {
+			writel(val | QID(fl->bar2_qid),
+			       fl->bar2_addr + SGE_UDB_KDOORBELL);
+
+			/* This Write memory Barrier will force the write to
+			 * the User Doorbell area to be flushed.
+			 */
+			wmb();
+		}
 		fl->pend_cred %= FL_PER_EQ_UNIT;
 	}
 }
@@ -949,14 +970,74 @@ static void write_sgl(const struct sk_buff *skb, struct sge_txq *tq,
 static inline void ring_tx_db(struct adapter *adapter, struct sge_txq *tq,
 			      int n)
 {
-	/*
-	 * Warn if we write doorbells with the wrong priority and write
-	 * descriptors before telling HW.
+	/* Make sure that all writes to the TX Descriptors are committed
+	 * before we tell the hardware about them.
 	 */
-	WARN_ON((QID(tq->cntxt_id) | PIDX(n)) & DBPRIO(1));
 	wmb();
-	t4_write_reg(adapter, T4VF_SGE_BASE_ADDR + SGE_VF_KDOORBELL,
-		     QID(tq->cntxt_id) | PIDX(n));
+
+	/* If we don't have access to the new User Doorbell (T5+), use the old
+	 * doorbell mechanism; otherwise use the new BAR2 mechanism.
+	 */
+	if (unlikely(tq->bar2_addr == NULL)) {
+		u32 val = PIDX(n);
+
+		t4_write_reg(adapter, T4VF_SGE_BASE_ADDR + SGE_VF_KDOORBELL,
+			     QID(tq->cntxt_id) | val);
+	} else {
+		u32 val = PIDX_T5(n);
+
+		/* T4 and later chips share the same PIDX field offset within
+		 * the doorbell, but T5 and later shrank the field in order to
+		 * gain a bit for Doorbell Priority.  The field was absurdly
+		 * large in the first place (14 bits) so we just use the T5
+		 * and later limits and warn if a Queue ID is too large.
+		 */
+		WARN_ON(val & DBPRIO(1));
+
+		/* If we're only writing a single Egress Unit and the BAR2
+		 * Queue ID is 0, we can use the Write Combining Doorbell
+		 * Gather Buffer; otherwise we use the simple doorbell.
+		 */
+		if (n == 1 && tq->bar2_qid == 0) {
+			unsigned int index = (tq->pidx
+					      ? (tq->pidx - 1)
+					      : (tq->size - 1));
+			__be64 *src = (__be64 *)&tq->desc[index];
+			__be64 __iomem *dst = (__be64 *)(tq->bar2_addr +
+							 SGE_UDB_WCDOORBELL);
+			unsigned int count = EQ_UNIT / sizeof(__be64);
+
+			/* Copy the TX Descriptor in a tight loop in order to
+			 * try to get it to the adapter in a single Write
+			 * Combined transfer on the PCI-E Bus.  If the Write
+			 * Combine fails (say because of an interrupt, etc.)
+			 * the hardware will simply take the last write as a
+			 * simple doorbell write with a PIDX Increment of 1
+			 * and will fetch the TX Descriptor from memory via
+			 * DMA.
+			 */
+			while (count) {
+				writeq(*src, dst);
+				src++;
+				dst++;
+				count--;
+			}
+		} else
+			writel(val | QID(tq->bar2_qid),
+			       tq->bar2_addr + SGE_UDB_KDOORBELL);
+
+		/* This Write Memory Barrier will force the write to the User
+		 * Doorbell area to be flushed.  This is needed to prevent
+		 * writes on different CPUs for the same queue from hitting
+		 * the adapter out of order.  This is required when some Work
+		 * Requests take the Write Combine Gather Buffer path (user
+		 * doorbell area offset [SGE_UDB_WCDOORBELL..+63]) and some
+		 * take the traditional path where we simply increment the
+		 * PIDX (User Doorbell area SGE_UDB_KDOORBELL) and have the
+		 * hardware DMA read the actual Work Request.
+		 */
+		wmb();
+	}
 }
 
 /**
@@ -1782,6 +1863,7 @@ static int napi_rx_handler(struct napi_struct *napi, int budget)
 	unsigned int intr_params;
 	struct sge_rspq *rspq = container_of(napi, struct sge_rspq, napi);
 	int work_done = process_responses(rspq, budget);
+	u32 val;
 
 	if (likely(work_done < budget)) {
 		napi_complete(napi);
@@ -1793,11 +1875,16 @@ static int napi_rx_handler(struct napi_struct *napi, int budget)
 	if (unlikely(work_done == 0))
 		rspq->unhandled_irqs++;
 
-	t4_write_reg(rspq->adapter,
-		     T4VF_SGE_BASE_ADDR + SGE_VF_GTS,
-		     CIDXINC(work_done) |
-		     INGRESSQID((u32)rspq->cntxt_id) |
-		     SEINTARM(intr_params));
+	val = CIDXINC(work_done) | SEINTARM(intr_params);
+	if (is_t4(rspq->adapter->params.chip)) {
+		t4_write_reg(rspq->adapter,
+			     T4VF_SGE_BASE_ADDR + SGE_VF_GTS,
+			     val | INGRESSQID((u32)rspq->cntxt_id));
+	} else {
+		writel(val | INGRESSQID(rspq->bar2_qid),
+		       rspq->bar2_addr + SGE_UDB_GTS);
+		wmb();
+	}
 	return work_done;
 }
 
@@ -1822,6 +1909,7 @@ static unsigned int process_intrq(struct adapter *adapter)
 	struct sge *s = &adapter->sge;
 	struct sge_rspq *intrq = &s->intrq;
 	unsigned int work_done;
+	u32 val;
 
 	spin_lock(&adapter->sge.intrq_lock);
 	for (work_done = 0; ; work_done++) {
@@ -1887,10 +1975,15 @@ static unsigned int process_intrq(struct adapter *adapter)
 		rspq_next(intrq);
 	}
 
-	t4_write_reg(adapter, T4VF_SGE_BASE_ADDR + SGE_VF_GTS,
-		     CIDXINC(work_done) |
-		     INGRESSQID(intrq->cntxt_id) |
-		     SEINTARM(intrq->intr_params));
+	val = CIDXINC(work_done) | SEINTARM(intrq->intr_params);
+	if (is_t4(adapter->params.chip))
+		t4_write_reg(adapter, T4VF_SGE_BASE_ADDR + SGE_VF_GTS,
+			     val | INGRESSQID(intrq->cntxt_id));
+	else {
+		writel(val | INGRESSQID(intrq->bar2_qid),
+		       intrq->bar2_addr + SGE_UDB_GTS);
+		wmb();
+	}
 
 	spin_unlock(&adapter->sge.intrq_lock);
 
@@ -2036,6 +2129,35 @@ static void sge_tx_timer_cb(unsigned long data)
 }
 
 /**
+ *	bar2_address - return the BAR2 address for an SGE Queue's Registers
+ *	@adapter: the adapter
+ *	@qid: the SGE Queue ID
+ *	@qtype: the SGE Queue Type (Egress or Ingress)
+ *	@pbar2_qid: BAR2 Queue ID or 0 for Queue ID inferred SGE Queues
+ *
+ *	Returns the BAR2 address for the SGE Queue Registers associated with
+ *	@qid.  If BAR2 SGE Registers aren't available, returns NULL.  Also
+ *	returns the BAR2 Queue ID to be used with writes to the BAR2 SGE
+ *	Queue Registers.  If the BAR2 Queue ID is 0, then "Inferred Queue ID"
+ *	Registers are supported (e.g. the Write Combining Doorbell Buffer).
+ */
+static void __iomem *bar2_address(struct adapter *adapter,
+				  unsigned int qid,
+				  enum t4_bar2_qtype qtype,
+				  unsigned int *pbar2_qid)
+{
+	u64 bar2_qoffset;
+	int ret;
+
+	ret = t4_bar2_sge_qregs(adapter, qid, qtype,
+				&bar2_qoffset, pbar2_qid);
+	if (ret)
+		return NULL;
+
+	return adapter->bar2 + bar2_qoffset;
+}
+
+/**
  *	t4vf_sge_alloc_rxq - allocate an SGE RX Queue
  *	@adapter: the adapter
  *	@rspq: pointer to to the new rxq's Response Queue to be filled in
@@ -2166,6 +2288,10 @@ int t4vf_sge_alloc_rxq(struct adapter *adapter, struct sge_rspq *rspq,
 	rspq->gen = 1;
 	rspq->next_intr_params = rspq->intr_params;
 	rspq->cntxt_id = be16_to_cpu(rpl.iqid);
+	rspq->bar2_addr = bar2_address(adapter,
+				       rspq->cntxt_id,
+				       T4_BAR2_QTYPE_INGRESS,
+				       &rspq->bar2_qid);
 	rspq->abs_id = be16_to_cpu(rpl.physiqid);
 	rspq->size--;			/* subtract status entry */
 	rspq->adapter = adapter;
@@ -2184,6 +2310,15 @@ int t4vf_sge_alloc_rxq(struct adapter *adapter, struct sge_rspq *rspq,
 		fl->alloc_failed = 0;
 		fl->large_alloc_failed = 0;
 		fl->starving = 0;
+
+		/* Note, we must initialize the BAR2 Free List User Doorbell
+		 * information before refilling the Free List!
+		 */
+		fl->bar2_addr = bar2_address(adapter,
+					     fl->cntxt_id,
+					     T4_BAR2_QTYPE_EGRESS,
+					     &fl->bar2_qid);
+
 		refill_fl(adapter, fl, fl_cap(fl), GFP_KERNEL);
 	}
 
@@ -2296,6 +2431,10 @@ int t4vf_sge_alloc_eth_txq(struct adapter *adapter, struct sge_eth_txq *txq,
 	txq->q.pidx = 0;
 	txq->q.stat = (void *)&txq->q.desc[txq->q.size];
 	txq->q.cntxt_id = FW_EQ_ETH_CMD_EQID_G(be32_to_cpu(rpl.eqid_pkd));
+	txq->q.bar2_addr = bar2_address(adapter,
+					txq->q.cntxt_id,
+					T4_BAR2_QTYPE_EGRESS,
+					&txq->q.bar2_qid);
 	txq->q.abs_id =
 		FW_EQ_ETH_CMD_PHYSEQID_G(be32_to_cpu(rpl.physeqid_pkd));
 	txq->txq = devq;
diff --git a/drivers/net/ethernet/chelsio/cxgb4vf/t4vf_common.h b/drivers/net/ethernet/chelsio/cxgb4vf/t4vf_common.h
index a608c6657d63..8d3237f5e364 100644
--- a/drivers/net/ethernet/chelsio/cxgb4vf/t4vf_common.h
+++ b/drivers/net/ethernet/chelsio/cxgb4vf/t4vf_common.h
@@ -135,9 +135,12 @@ struct dev_params {
 struct sge_params {
 	u32 sge_control;		/* padding, boundaries, lengths, etc. */
 	u32 sge_control2;		/* T5: more of the same */
-	u32 sge_host_page_size;		/* RDMA page sizes */
-	u32 sge_queues_per_page;	/* RDMA queues/page */
-	u32 sge_user_mode_limits;	/* limits for BAR2 user mode accesses */
+	u32 sge_host_page_size;		/* PF0-7 page sizes */
+	u32 sge_egress_queues_per_page;	/* PF0-7 egress queues/page */
+	u32 sge_ingress_queues_per_page;/* PF0-7 ingress queues/page */
+	u32 sge_vf_hps;                 /* host page size for our vf */
+	u32 sge_vf_eq_qpp;		/* egress queues/page for our VF */
+	u32 sge_vf_iq_qpp;		/* ingress queues/page for our VF */
 	u32 sge_fl_buffer_size[16];	/* free list buffer sizes */
 	u32 sge_ingress_rx_threshold;	/* RX counter interrupt threshold[4] */
 	u32 sge_congestion_control;     /* congestion thresholds, etc. */
@@ -267,6 +270,8 @@ static inline int t4vf_wr_mbox_ns(struct adapter *adapter, const void *cmd,
 	return t4vf_wr_mbox_core(adapter, cmd, size, rpl, false);
 }
 
+#define CHELSIO_PCI_ID_VER(dev_id)  ((dev_id) >> 12)
+
 static inline int is_t4(enum chip_type chip)
 {
 	return CHELSIO_CHIP_VERSION(chip) == CHELSIO_T4;
@@ -278,6 +283,13 @@ int t4vf_port_init(struct adapter *, int);
 int t4vf_fw_reset(struct adapter *);
 int t4vf_set_params(struct adapter *, unsigned int, const u32 *, const u32 *);
 
+enum t4_bar2_qtype { T4_BAR2_QTYPE_EGRESS, T4_BAR2_QTYPE_INGRESS };
+int t4_bar2_sge_qregs(struct adapter *adapter,
+		      unsigned int qid,
+		      enum t4_bar2_qtype qtype,
+		      u64 *pbar2_qoffset,
+		      unsigned int *pbar2_qid);
+
 int t4vf_get_sge_params(struct adapter *);
 int t4vf_get_vpd_params(struct adapter *);
 int t4vf_get_dev_params(struct adapter *);
@@ -309,5 +321,6 @@ int t4vf_iq_free(struct adapter *, unsigned int, unsigned int, unsigned int,
 int t4vf_eth_eq_free(struct adapter *, unsigned int);
 
 int t4vf_handle_fw_rpl(struct adapter *, const __be64 *);
+int t4vf_prep_adapter(struct adapter *);
 
 #endif /* __T4VF_COMMON_H__ */
diff --git a/drivers/net/ethernet/chelsio/cxgb4vf/t4vf_hw.c b/drivers/net/ethernet/chelsio/cxgb4vf/t4vf_hw.c
index 624a213dea87..02e8833b7797 100644
--- a/drivers/net/ethernet/chelsio/cxgb4vf/t4vf_hw.c
+++ b/drivers/net/ethernet/chelsio/cxgb4vf/t4vf_hw.c
@@ -430,6 +430,95 @@ int t4vf_set_params(struct adapter *adapter, unsigned int nparams,
 }
 
 /**
+ *	t4_bar2_sge_qregs - return BAR2 SGE Queue register information
+ *	@adapter: the adapter
+ *	@qid: the Queue ID
+ *	@qtype: the Ingress or Egress type for @qid
+ *	@pbar2_qoffset: BAR2 Queue Offset
+ *	@pbar2_qid: BAR2 Queue ID or 0 for Queue ID inferred SGE Queues
+ *
+ *	Returns the BAR2 SGE Queue Registers information associated with the
+ *	indicated Absolute Queue ID.  These are passed back in return value
+ *	pointers.  @qtype should be T4_BAR2_QTYPE_EGRESS for Egress Queue
+ *	and T4_BAR2_QTYPE_INGRESS for Ingress Queues.
+ *
+ *	This may return an error which indicates that BAR2 SGE Queue
+ *	registers aren't available.  If an error is not returned, then the
+ *	following values are returned:
+ *
+ *	  *@pbar2_qoffset: the BAR2 Offset of the @qid Registers
+ *	  *@pbar2_qid: the BAR2 SGE Queue ID or 0 of @qid
+ *
+ *	If the returned BAR2 Queue ID is 0, then BAR2 SGE registers which
+ *	require the "Inferred Queue ID" ability may be used.  E.g. the
+ *	Write Combining Doorbell Buffer. If the BAR2 Queue ID is not 0,
+ *	then these "Inferred Queue ID" register may not be used.
+ */
+int t4_bar2_sge_qregs(struct adapter *adapter,
+		      unsigned int qid,
+		      enum t4_bar2_qtype qtype,
+		      u64 *pbar2_qoffset,
+		      unsigned int *pbar2_qid)
+{
+	unsigned int page_shift, page_size, qpp_shift, qpp_mask;
+	u64 bar2_page_offset, bar2_qoffset;
+	unsigned int bar2_qid, bar2_qid_offset, bar2_qinferred;
+
+	/* T4 doesn't support BAR2 SGE Queue registers.
+	 */
+	if (is_t4(adapter->params.chip))
+		return -EINVAL;
+
+	/* Get our SGE Page Size parameters.
+	 */
+	page_shift = adapter->params.sge.sge_vf_hps + 10;
+	page_size = 1 << page_shift;
+
+	/* Get the right Queues per Page parameters for our Queue.
+	 */
+	qpp_shift = (qtype == T4_BAR2_QTYPE_EGRESS
+		     ? adapter->params.sge.sge_vf_eq_qpp
+		     : adapter->params.sge.sge_vf_iq_qpp);
+	qpp_mask = (1 << qpp_shift) - 1;
+
+	/* Calculate the basics of the BAR2 SGE Queue register area:
+	 *  o The BAR2 page the Queue registers will be in.
+	 *  o The BAR2 Queue ID.
+	 *  o The BAR2 Queue ID Offset into the BAR2 page.
+	 */
+	bar2_page_offset = ((qid >> qpp_shift) << page_shift);
+	bar2_qid = qid & qpp_mask;
+	bar2_qid_offset = bar2_qid * SGE_UDB_SIZE;
+
+	/* If the BAR2 Queue ID Offset is less than the Page Size, then the
+	 * hardware will infer the Absolute Queue ID simply from the writes to
+	 * the BAR2 Queue ID Offset within the BAR2 Page (and we need to use a
+	 * BAR2 Queue ID of 0 for those writes).  Otherwise, we'll simply
+	 * write to the first BAR2 SGE Queue Area within the BAR2 Page with
+	 * the BAR2 Queue ID and the hardware will infer the Absolute Queue ID
+	 * from the BAR2 Page and BAR2 Queue ID.
+	 *
+	 * One important censequence of this is that some BAR2 SGE registers
+	 * have a "Queue ID" field and we can write the BAR2 SGE Queue ID
+	 * there.  But other registers synthesize the SGE Queue ID purely
+	 * from the writes to the registers -- the Write Combined Doorbell
+	 * Buffer is a good example.  These BAR2 SGE Registers are only
+	 * available for those BAR2 SGE Register areas where the SGE Absolute
+	 * Queue ID can be inferred from simple writes.
+	 */
+	bar2_qoffset = bar2_page_offset;
+	bar2_qinferred = (bar2_qid_offset < page_size);
+	if (bar2_qinferred) {
+		bar2_qoffset += bar2_qid_offset;
+		bar2_qid = 0;
+	}
+
+	*pbar2_qoffset = bar2_qoffset;
+	*pbar2_qid = bar2_qid;
+	return 0;
+}
+
+/**
  *	t4vf_get_sge_params - retrieve adapter Scatter gather Engine parameters
  *	@adapter: the adapter
  *
@@ -501,6 +590,55 @@ int t4vf_get_sge_params(struct adapter *adapter)
 	sge_params->sge_ingress_rx_threshold = vals[0];
 	sge_params->sge_congestion_control = vals[1];
 
+	/* For T5 and later we want to use the new BAR2 Doorbells.
+	 * Unfortunately, older firmware didn't allow the this register to be
+	 * read.
+	 */
+	if (!is_t4(adapter->params.chip)) {
+		u32 whoami;
+		unsigned int pf, s_hps, s_qpp;
+
+		params[0] = (FW_PARAMS_MNEM_V(FW_PARAMS_MNEM_REG) |
+			     FW_PARAMS_PARAM_XYZ_V(
+				     SGE_EGRESS_QUEUES_PER_PAGE_VF_A));
+		params[1] = (FW_PARAMS_MNEM_V(FW_PARAMS_MNEM_REG) |
+			     FW_PARAMS_PARAM_XYZ_V(
+				     SGE_INGRESS_QUEUES_PER_PAGE_VF_A));
+		v = t4vf_query_params(adapter, 2, params, vals);
+		if (v != FW_SUCCESS) {
+			dev_warn(adapter->pdev_dev,
+				 "Unable to get VF SGE Queues/Page; "
+				 "probably old firmware.\n");
+			return v;
+		}
+		sge_params->sge_egress_queues_per_page = vals[0];
+		sge_params->sge_ingress_queues_per_page = vals[1];
+
+		/* We need the Queues/Page for our VF.  This is based on the
+		 * PF from which we're instantiated and is indexed in the
+		 * register we just read. Do it once here so other code in
+		 * the driver can just use it.
+		 */
+		whoami = t4_read_reg(adapter,
+				     T4VF_PL_BASE_ADDR + A_PL_VF_WHOAMI);
+		pf = SOURCEPF_GET(whoami);
+
+		s_hps = (HOSTPAGESIZEPF0_S +
+			 (HOSTPAGESIZEPF1_S - HOSTPAGESIZEPF0_S) * pf);
+		sge_params->sge_vf_hps =
+			((sge_params->sge_host_page_size >> s_hps)
+			 & HOSTPAGESIZEPF0_M);
+
+		s_qpp = (QUEUESPERPAGEPF0_S +
+			 (QUEUESPERPAGEPF1_S - QUEUESPERPAGEPF0_S) * pf);
+		sge_params->sge_vf_eq_qpp =
+			((sge_params->sge_egress_queues_per_page >> s_qpp)
+			 & QUEUESPERPAGEPF0_MASK);
+		sge_params->sge_vf_iq_qpp =
+			((sge_params->sge_ingress_queues_per_page >> s_qpp)
+			 & QUEUESPERPAGEPF0_MASK);
+	}
+
 	return 0;
 }
 
@@ -1420,3 +1558,38 @@ int t4vf_handle_fw_rpl(struct adapter *adapter, const __be64 *rpl)
 	}
 	return 0;
 }
+
+/**
+ */
+int t4vf_prep_adapter(struct adapter *adapter)
+{
+	int err;
+	unsigned int chipid;
+
+	/* Wait for the device to become ready before proceeding ...
+	 */
+	err = t4vf_wait_dev_ready(adapter);
+	if (err)
+		return err;
+
+	/* Default port and clock for debugging in case we can't reach
+	 * firmware.
+	 */
+	adapter->params.nports = 1;
+	adapter->params.vfres.pmask = 1;
+	adapter->params.vpd.cclk = 50000;
+
+	adapter->params.chip = 0;
+	switch (CHELSIO_PCI_ID_VER(adapter->pdev->device)) {
+	case CHELSIO_T4:
+		adapter->params.chip |= CHELSIO_CHIP_CODE(CHELSIO_T4, 0);
+		break;
+
+	case CHELSIO_T5:
+		chipid = G_REV(t4_read_reg(adapter, A_PL_VF_REV));
+		adapter->params.chip |= CHELSIO_CHIP_CODE(CHELSIO_T5, chipid);
+		break;
+	}
+
+	return 0;
+}