[PATCH net-next 0/7] cxgb4: add hash-filter support to tc-flower offload

[PATCH net-next 0/7] cxgb4: add hash-filter support to tc-flower offload

[Rahul Lakkireddy]

This series of patches add support to create hash-filters; a.k.a
exact-match filters, to tc-flower offload.  T6 supports creating
~500K hash-filters in hw and can theoretically be expanded up to
~1 million.

Patch 1 fetches and saves the configured hw filter tuple field shifts
and filter mask.

Patch 2 initializes the driver to use hash-filter configuration.

Patch 3 adds support to create hash filters in hw.

Patch 4 adds support to delete hash filters in hw.

Patch 5 adds support to retrieve filter stats for hash filters.

Patch 6 converts the flower table to use rhashtable instead of
static hlist.

Patch 7 finally adds support to create hash filters via tc-flower
offload.

Thanks,
Rahul

Kumar Sanghvi (7):
  cxgb4: save additional filter tuple field shifts in tp_params
  cxgb4: initialize hash-filter configuration
  cxgb4: add support to create hash filters
  cxgb4: add support to delete hash filter
  cxgb4: add support to retrieve stats for hash filters
  cxgb4: convert flower table to use rhashtable
  cxgb4: add support to create hash-filters via tc-flower offload

 drivers/net/ethernet/chelsio/cxgb4/cxgb4.h         |  32 +-
 drivers/net/ethernet/chelsio/cxgb4/cxgb4_filter.c  | 928 +++++++++++++++++++--
 drivers/net/ethernet/chelsio/cxgb4/cxgb4_filter.h  |   8 +
 drivers/net/ethernet/chelsio/cxgb4/cxgb4_main.c    |  26 +-
 .../net/ethernet/chelsio/cxgb4/cxgb4_tc_flower.c   | 126 ++-
 .../net/ethernet/chelsio/cxgb4/cxgb4_tc_flower.h   |   4 +-
 drivers/net/ethernet/chelsio/cxgb4/cxgb4_tc_u32.c  |   4 +-
 drivers/net/ethernet/chelsio/cxgb4/cxgb4_uld.h     |   6 +-
 drivers/net/ethernet/chelsio/cxgb4/t4_hw.c         |  18 +-
 drivers/net/ethernet/chelsio/cxgb4/t4_msg.h        |   5 +
 drivers/net/ethernet/chelsio/cxgb4/t4_regs.h       |  17 +
 drivers/net/ethernet/chelsio/cxgb4/t4_tcb.h        |  22 +
 drivers/net/ethernet/chelsio/cxgb4/t4fw_api.h      |   1 +
 13 files changed, 1082 insertions(+), 115 deletions(-)

-- 
2.14.1

[PATCH net-next 1/7] cxgb4: save additional filter tuple field shifts in tp_params

[Rahul Lakkireddy]

From: Kumar Sanghvi <kumaras@chelsio.com>

Save additional filter tuple field shifts in tp_params based on
configured filter tuple fields.

Also, save the combined filter tuple mask based on configured
filter tuple fields.

Signed-off-by: Kumar Sanghvi <kumaras@chelsio.com>
Signed-off-by: Rahul Lakkireddy <rahul.lakkireddy@chelsio.com>
Signed-off-by: Ganesh Goudar <ganeshgr@chelsio.com>
---
 drivers/net/ethernet/chelsio/cxgb4/cxgb4.h   | 12 ++++++++++--
 drivers/net/ethernet/chelsio/cxgb4/t4_hw.c   | 18 ++++++++++++++++--
 drivers/net/ethernet/chelsio/cxgb4/t4_regs.h |  3 +++
 3 files changed, 29 insertions(+), 4 deletions(-)

diff --git a/drivers/net/ethernet/chelsio/cxgb4/cxgb4.h b/drivers/net/ethernet/chelsio/cxgb4/cxgb4.h
index a57761b28edc..e2c75b73595f 100644
--- a/drivers/net/ethernet/chelsio/cxgb4/cxgb4.h
+++ b/drivers/net/ethernet/chelsio/cxgb4/cxgb4.h
@@ -287,10 +287,18 @@ struct tp_params {
 	 * places we store their offsets here, or a -1 if the field isn't
 	 * present.
 	 */
-	int vlan_shift;
-	int vnic_shift;
+	int fcoe_shift;
 	int port_shift;
+	int vnic_shift;
+	int vlan_shift;
+	int tos_shift;
 	int protocol_shift;
+	int ethertype_shift;
+	int macmatch_shift;
+	int matchtype_shift;
+	int frag_shift;
+
+	u64 hash_filter_mask;
 };
 
 struct vpd_params {
diff --git a/drivers/net/ethernet/chelsio/cxgb4/t4_hw.c b/drivers/net/ethernet/chelsio/cxgb4/t4_hw.c
index c289ca1efc1b..efe9d3a20135 100644
--- a/drivers/net/ethernet/chelsio/cxgb4/t4_hw.c
+++ b/drivers/net/ethernet/chelsio/cxgb4/t4_hw.c
@@ -8816,11 +8816,21 @@ int t4_init_tp_params(struct adapter *adap, bool sleep_ok)
 	 * shift positions of several elements of the Compressed Filter Tuple
 	 * for this adapter which we need frequently ...
 	 */
-	adap->params.tp.vlan_shift = t4_filter_field_shift(adap, VLAN_F);
-	adap->params.tp.vnic_shift = t4_filter_field_shift(adap, VNIC_ID_F);
+	adap->params.tp.fcoe_shift = t4_filter_field_shift(adap, FCOE_F);
 	adap->params.tp.port_shift = t4_filter_field_shift(adap, PORT_F);
+	adap->params.tp.vnic_shift = t4_filter_field_shift(adap, VNIC_ID_F);
+	adap->params.tp.vlan_shift = t4_filter_field_shift(adap, VLAN_F);
+	adap->params.tp.tos_shift = t4_filter_field_shift(adap, TOS_F);
 	adap->params.tp.protocol_shift = t4_filter_field_shift(adap,
 							       PROTOCOL_F);
+	adap->params.tp.ethertype_shift = t4_filter_field_shift(adap,
+								ETHERTYPE_F);
+	adap->params.tp.macmatch_shift = t4_filter_field_shift(adap,
+							       MACMATCH_F);
+	adap->params.tp.matchtype_shift = t4_filter_field_shift(adap,
+								MPSHITTYPE_F);
+	adap->params.tp.frag_shift = t4_filter_field_shift(adap,
+							   FRAGMENTATION_F);
 
 	/* If TP_INGRESS_CONFIG.VNID == 0, then TP_VLAN_PRI_MAP.VNIC_ID
 	 * represents the presence of an Outer VLAN instead of a VNIC ID.
@@ -8828,6 +8838,10 @@ int t4_init_tp_params(struct adapter *adap, bool sleep_ok)
 	if ((adap->params.tp.ingress_config & VNIC_F) == 0)
 		adap->params.tp.vnic_shift = -1;
 
+	v = t4_read_reg(adap, LE_3_DB_HASH_MASK_GEN_IPV4_T6_A);
+	adap->params.tp.hash_filter_mask = v;
+	v = t4_read_reg(adap, LE_4_DB_HASH_MASK_GEN_IPV4_T6_A);
+	adap->params.tp.hash_filter_mask |= ((u64)v << 32);
 	return 0;
 }
 
diff --git a/drivers/net/ethernet/chelsio/cxgb4/t4_regs.h b/drivers/net/ethernet/chelsio/cxgb4/t4_regs.h
index 483fb7644355..44713bad0045 100644
--- a/drivers/net/ethernet/chelsio/cxgb4/t4_regs.h
+++ b/drivers/net/ethernet/chelsio/cxgb4/t4_regs.h
@@ -2933,6 +2933,9 @@
 #define SSRAMINTPERR_V(x) ((x) << SSRAMINTPERR_S)
 #define SSRAMINTPERR_F    SSRAMINTPERR_V(1U)
 
+#define LE_3_DB_HASH_MASK_GEN_IPV4_T6_A	0x19eac
+#define LE_4_DB_HASH_MASK_GEN_IPV4_T6_A	0x19eb0
+
 #define NCSI_INT_CAUSE_A 0x1a0d8
 
 #define CIM_DM_PRTY_ERR_S    8
-- 
2.14.1

[PATCH net-next 2/7] cxgb4: initialize hash-filter configuration

[Rahul Lakkireddy]

From: Kumar Sanghvi <kumaras@chelsio.com>

Add support for hash-filter configuration on T6. Also, do basic
checks for the related initialization.

Signed-off-by: Kumar Sanghvi <kumaras@chelsio.com>
Signed-off-by: Rahul Lakkireddy <rahul.lakkireddy@chelsio.com>
Signed-off-by: Ganesh Goudar <ganeshgr@chelsio.com>
---
 drivers/net/ethernet/chelsio/cxgb4/cxgb4.h        |  6 ++++++
 drivers/net/ethernet/chelsio/cxgb4/cxgb4_filter.c | 22 ++++++++++++++++++++++
 drivers/net/ethernet/chelsio/cxgb4/cxgb4_filter.h |  1 +
 drivers/net/ethernet/chelsio/cxgb4/cxgb4_main.c   | 14 ++++++++++----
 drivers/net/ethernet/chelsio/cxgb4/t4_regs.h      | 14 ++++++++++++++
 drivers/net/ethernet/chelsio/cxgb4/t4fw_api.h     |  1 +
 6 files changed, 54 insertions(+), 4 deletions(-)

diff --git a/drivers/net/ethernet/chelsio/cxgb4/cxgb4.h b/drivers/net/ethernet/chelsio/cxgb4/cxgb4.h
index e2c75b73595f..5f021e6062b0 100644
--- a/drivers/net/ethernet/chelsio/cxgb4/cxgb4.h
+++ b/drivers/net/ethernet/chelsio/cxgb4/cxgb4.h
@@ -366,6 +366,7 @@ struct adapter_params {
 	unsigned char crypto;		/* HW capability for crypto */
 
 	unsigned char bypass;
+	unsigned char hash_filter;
 
 	unsigned int ofldq_wr_cred;
 	bool ulptx_memwrite_dsgl;          /* use of T5 DSGL allowed */
@@ -1140,6 +1141,11 @@ static inline int is_offload(const struct adapter *adap)
 	return adap->params.offload;
 }
 
+static inline int is_hashfilter(const struct adapter *adap)
+{
+	return adap->params.hash_filter;
+}
+
 static inline int is_pci_uld(const struct adapter *adap)
 {
 	return adap->params.crypto;
diff --git a/drivers/net/ethernet/chelsio/cxgb4/cxgb4_filter.c b/drivers/net/ethernet/chelsio/cxgb4/cxgb4_filter.c
index 89272f29f807..566bd2d3737c 100644
--- a/drivers/net/ethernet/chelsio/cxgb4/cxgb4_filter.c
+++ b/drivers/net/ethernet/chelsio/cxgb4/cxgb4_filter.c
@@ -915,3 +915,25 @@ void filter_rpl(struct adapter *adap, const struct cpl_set_tcb_rpl *rpl)
 			complete(&ctx->completion);
 	}
 }
+
+int init_hash_filter(struct adapter *adap)
+{
+	/* On T6, verify the necessary register configs and warn the user in
+	 * case of improper config
+	 */
+	if (is_t6(adap->params.chip)) {
+		if (TCAM_ACTV_HIT_G(t4_read_reg(adap, LE_DB_RSP_CODE_0_A)) != 4)
+			goto err;
+
+		if (HASH_ACTV_HIT_G(t4_read_reg(adap, LE_DB_RSP_CODE_1_A)) != 4)
+			goto err;
+	} else {
+		dev_err(adap->pdev_dev, "Hash filter supported only on T6\n");
+		return -EINVAL;
+	}
+	adap->params.hash_filter = 1;
+	return 0;
+err:
+	dev_warn(adap->pdev_dev, "Invalid hash filter config!\n");
+	return -EINVAL;
+}
diff --git a/drivers/net/ethernet/chelsio/cxgb4/cxgb4_filter.h b/drivers/net/ethernet/chelsio/cxgb4/cxgb4_filter.h
index 23742cb1c69f..d3c1a8fafd32 100644
--- a/drivers/net/ethernet/chelsio/cxgb4/cxgb4_filter.h
+++ b/drivers/net/ethernet/chelsio/cxgb4/cxgb4_filter.h
@@ -45,4 +45,5 @@ int delete_filter(struct adapter *adapter, unsigned int fidx);
 
 int writable_filter(struct filter_entry *f);
 void clear_all_filters(struct adapter *adapter);
+int init_hash_filter(struct adapter *adap);
 #endif /* __CXGB4_FILTER_H */
diff --git a/drivers/net/ethernet/chelsio/cxgb4/cxgb4_main.c b/drivers/net/ethernet/chelsio/cxgb4/cxgb4_main.c
index e16078ddb39f..4b07cfe8c66c 100644
--- a/drivers/net/ethernet/chelsio/cxgb4/cxgb4_main.c
+++ b/drivers/net/ethernet/chelsio/cxgb4/cxgb4_main.c
@@ -3963,7 +3963,8 @@ static int adap_init0(struct adapter *adap)
 	if (ret < 0)
 		goto bye;
 
-	if (caps_cmd.ofldcaps) {
+	if (caps_cmd.ofldcaps ||
+	    (caps_cmd.niccaps & htons(FW_CAPS_CONFIG_NIC_HASHFILTER))) {
 		/* query offload-related parameters */
 		params[0] = FW_PARAM_DEV(NTID);
 		params[1] = FW_PARAM_PFVF(SERVER_START);
@@ -4000,8 +4001,13 @@ static int adap_init0(struct adapter *adap)
 		adap->vres.ddp.size = val[4] - val[3] + 1;
 		adap->params.ofldq_wr_cred = val[5];
 
-		adap->params.offload = 1;
-		adap->num_ofld_uld += 1;
+		if (caps_cmd.niccaps & htons(FW_CAPS_CONFIG_NIC_HASHFILTER)) {
+			if (init_hash_filter(adap) < 0)
+				goto bye;
+		} else {
+			adap->params.offload = 1;
+			adap->num_ofld_uld += 1;
+		}
 	}
 	if (caps_cmd.rdmacaps) {
 		params[0] = FW_PARAM_PFVF(STAG_START);
@@ -5171,7 +5177,7 @@ static int init_one(struct pci_dev *pdev, const struct pci_device_id *ent)
 		cxgb4_init_tc_flower(adapter);
 	}
 
-	if (is_offload(adapter)) {
+	if (is_offload(adapter) || is_hashfilter(adapter)) {
 		if (t4_read_reg(adapter, LE_DB_CONFIG_A) & HASHEN_F) {
 			u32 hash_base, hash_reg;
 
diff --git a/drivers/net/ethernet/chelsio/cxgb4/t4_regs.h b/drivers/net/ethernet/chelsio/cxgb4/t4_regs.h
index 44713bad0045..623f453bd327 100644
--- a/drivers/net/ethernet/chelsio/cxgb4/t4_regs.h
+++ b/drivers/net/ethernet/chelsio/cxgb4/t4_regs.h
@@ -2933,6 +2933,20 @@
 #define SSRAMINTPERR_V(x) ((x) << SSRAMINTPERR_S)
 #define SSRAMINTPERR_F    SSRAMINTPERR_V(1U)
 
+#define LE_DB_RSP_CODE_0_A	0x19c74
+
+#define TCAM_ACTV_HIT_S		0
+#define TCAM_ACTV_HIT_M		0x1fU
+#define TCAM_ACTV_HIT_V(x)	((x) << TCAM_ACTV_HIT_S)
+#define TCAM_ACTV_HIT_G(x)	(((x) >> TCAM_ACTV_HIT_S) & TCAM_ACTV_HIT_M)
+
+#define LE_DB_RSP_CODE_1_A     0x19c78
+
+#define HASH_ACTV_HIT_S		25
+#define HASH_ACTV_HIT_M		0x1fU
+#define HASH_ACTV_HIT_V(x)	((x) << HASH_ACTV_HIT_S)
+#define HASH_ACTV_HIT_G(x)	(((x) >> HASH_ACTV_HIT_S) & HASH_ACTV_HIT_M)
+
 #define LE_3_DB_HASH_MASK_GEN_IPV4_T6_A	0x19eac
 #define LE_4_DB_HASH_MASK_GEN_IPV4_T6_A	0x19eb0
 
diff --git a/drivers/net/ethernet/chelsio/cxgb4/t4fw_api.h b/drivers/net/ethernet/chelsio/cxgb4/t4fw_api.h
index 2ba890926c73..57eb4ad3485d 100644
--- a/drivers/net/ethernet/chelsio/cxgb4/t4fw_api.h
+++ b/drivers/net/ethernet/chelsio/cxgb4/t4fw_api.h
@@ -1092,6 +1092,7 @@ enum fw_caps_config_switch {
 enum fw_caps_config_nic {
 	FW_CAPS_CONFIG_NIC		= 0x00000001,
 	FW_CAPS_CONFIG_NIC_VM		= 0x00000002,
+	FW_CAPS_CONFIG_NIC_HASHFILTER	= 0x00000020,
 };
 
 enum fw_caps_config_ofld {
-- 
2.14.1

[PATCH net-next 3/7] cxgb4: add support to create hash filters

[Rahul Lakkireddy]

From: Kumar Sanghvi <kumaras@chelsio.com>

Add support to create hash (exact-match) filters based on the value
of 'hash' field in ch_filter_specification.

Allocate SMT/L2T entries if DMAC-rewrite/SMAC-rewrite is requested.

Allocate CLIP entry in case of IPv6 filter.

Use cpl_act_open_req[6] to send hash filter create request to hw.
Also, the filter tuple is calculated as part of sending this request.

Hash-filter reply is processed on getting cpl_act_open_rpl.
In case of success, various bits/fields in filter-tcb are set per
filter requirement, such as enabling filter hitcnts, and/or various
header rewrite operations, such as VLAN-rewrite, NAT or
(L3/L4)-rewrite, and SMAC/DMAC-rewrite. In case of failure, clear the
filter entry and release any hw resources occupied by it.

The patch also moves the functions set_tcb_field, set_tcb_tflag and
configure_filter_smac towards beginning of file.

Signed-off-by: Kumar Sanghvi <kumaras@chelsio.com>
Signed-off-by: Rahul Lakkireddy <rahul.lakkireddy@chelsio.com>
Signed-off-by: Ganesh Goudar <ganeshgr@chelsio.com>
---
 drivers/net/ethernet/chelsio/cxgb4/cxgb4.h        |  10 +-
 drivers/net/ethernet/chelsio/cxgb4/cxgb4_filter.c | 573 ++++++++++++++++++++--
 drivers/net/ethernet/chelsio/cxgb4/cxgb4_filter.h |   3 +
 drivers/net/ethernet/chelsio/cxgb4/cxgb4_main.c   |   4 +
 drivers/net/ethernet/chelsio/cxgb4/t4_msg.h       |   5 +
 drivers/net/ethernet/chelsio/cxgb4/t4_tcb.h       |  17 +
 6 files changed, 562 insertions(+), 50 deletions(-)

diff --git a/drivers/net/ethernet/chelsio/cxgb4/cxgb4.h b/drivers/net/ethernet/chelsio/cxgb4/cxgb4.h
index 5f021e6062b0..bb7f0e4c9a81 100644
--- a/drivers/net/ethernet/chelsio/cxgb4/cxgb4.h
+++ b/drivers/net/ethernet/chelsio/cxgb4/cxgb4.h
@@ -1050,6 +1050,7 @@ struct ch_filter_specification {
 	 * matching that doesn't exist as a (value, mask) tuple.
 	 */
 	uint32_t type:1;        /* 0 => IPv4, 1 => IPv6 */
+	u32 hash:1;		/* 0 => wild-card, 1 => exact-match */
 
 	/* Packet dispatch information.  Ingress packets which match the
 	 * filter rules will be dropped, passed to the host or switched back
@@ -1107,7 +1108,14 @@ enum {
 };
 
 enum {
-	NAT_MODE_ALL = 7,	/* NAT on entire 4-tuple */
+	NAT_MODE_NONE = 0,	/* No NAT performed */
+	NAT_MODE_DIP,		/* NAT on Dst IP */
+	NAT_MODE_DIP_DP,	/* NAT on Dst IP, Dst Port */
+	NAT_MODE_DIP_DP_SIP,	/* NAT on Dst IP, Dst Port and Src IP */
+	NAT_MODE_DIP_DP_SP,	/* NAT on Dst IP, Dst Port and Src Port */
+	NAT_MODE_SIP_SP,	/* NAT on Src IP and Src Port */
+	NAT_MODE_DIP_SIP_SP,	/* NAT on Dst IP, Src IP and Src Port */
+	NAT_MODE_ALL		/* NAT on entire 4-tuple */
 };
 
 /* Host shadow copy of ingress filter entry.  This is in host native format
diff --git a/drivers/net/ethernet/chelsio/cxgb4/cxgb4_filter.c b/drivers/net/ethernet/chelsio/cxgb4/cxgb4_filter.c
index 566bd2d3737c..cf8ca695f27d 100644
--- a/drivers/net/ethernet/chelsio/cxgb4/cxgb4_filter.c
+++ b/drivers/net/ethernet/chelsio/cxgb4/cxgb4_filter.c
@@ -35,6 +35,8 @@
 #include "cxgb4.h"
 #include "t4_regs.h"
 #include "t4_tcb.h"
+#include "t4_values.h"
+#include "clip_tbl.h"
 #include "l2t.h"
 #include "smt.h"
 #include "t4fw_api.h"
@@ -50,6 +52,141 @@ static inline bool unsupported(u32 conf, u32 conf_mask, u32 val, u32 mask)
 	return !(conf & conf_mask) && is_field_set(val, mask);
 }
 
+static int set_tcb_field(struct adapter *adap, struct filter_entry *f,
+			 unsigned int ftid,  u16 word, u64 mask, u64 val,
+			 int no_reply)
+{
+	struct cpl_set_tcb_field *req;
+	struct sk_buff *skb;
+
+	skb = alloc_skb(sizeof(struct cpl_set_tcb_field), GFP_ATOMIC);
+	if (!skb)
+		return -ENOMEM;
+
+	req = (struct cpl_set_tcb_field *)__skb_put(skb, sizeof(*req));
+	memset(req, 0, sizeof(*req));
+	INIT_TP_WR_CPL(req, CPL_SET_TCB_FIELD, ftid);
+	req->reply_ctrl = htons(REPLY_CHAN_V(0) |
+				QUEUENO_V(adap->sge.fw_evtq.abs_id) |
+				NO_REPLY_V(no_reply));
+	req->word_cookie = htons(TCB_WORD_V(word) | TCB_COOKIE_V(ftid));
+	req->mask = cpu_to_be64(mask);
+	req->val = cpu_to_be64(val);
+	set_wr_txq(skb, CPL_PRIORITY_CONTROL, f->fs.val.iport & 0x3);
+	t4_ofld_send(adap, skb);
+	return 0;
+}
+
+/* Set one of the t_flags bits in the TCB.
+ */
+static int set_tcb_tflag(struct adapter *adap, struct filter_entry *f,
+			 unsigned int ftid, unsigned int bit_pos,
+			 unsigned int val, int no_reply)
+{
+	return set_tcb_field(adap, f, ftid,  TCB_T_FLAGS_W, 1ULL << bit_pos,
+			     (unsigned long long)val << bit_pos, no_reply);
+}
+
+static int configure_filter_smac(struct adapter *adap, struct filter_entry *f)
+{
+	int err;
+
+	/* do a set-tcb for smac-sel and CWR bit.. */
+	err = set_tcb_tflag(adap, f, f->tid, TF_CCTRL_CWR_S, 1, 1);
+	if (err)
+		goto smac_err;
+
+	err = set_tcb_field(adap, f, f->tid, TCB_SMAC_SEL_W,
+			    TCB_SMAC_SEL_V(TCB_SMAC_SEL_M),
+			    TCB_SMAC_SEL_V(f->smt->idx), 1);
+	if (!err)
+		return 0;
+
+smac_err:
+	dev_err(adap->pdev_dev, "filter %u smac config failed with error %u\n",
+		f->tid, err);
+	return err;
+}
+
+static void set_nat_params(struct adapter *adap, struct filter_entry *f,
+			   unsigned int tid, bool dip, bool sip, bool dp,
+			   bool sp)
+{
+	if (dip) {
+		if (f->fs.type) {
+			set_tcb_field(adap, f, tid, TCB_SND_UNA_RAW_W,
+				      WORD_MASK, f->fs.nat_lip[15] |
+				      f->fs.nat_lip[14] << 8 |
+				      f->fs.nat_lip[13] << 16 |
+				      f->fs.nat_lip[12] << 24, 1);
+
+			set_tcb_field(adap, f, tid, TCB_SND_UNA_RAW_W + 1,
+				      WORD_MASK, f->fs.nat_lip[11] |
+				      f->fs.nat_lip[10] << 8 |
+				      f->fs.nat_lip[9] << 16 |
+				      f->fs.nat_lip[8] << 24, 1);
+
+			set_tcb_field(adap, f, tid, TCB_SND_UNA_RAW_W + 2,
+				      WORD_MASK, f->fs.nat_lip[7] |
+				      f->fs.nat_lip[6] << 8 |
+				      f->fs.nat_lip[5] << 16 |
+				      f->fs.nat_lip[4] << 24, 1);
+
+			set_tcb_field(adap, f, tid, TCB_SND_UNA_RAW_W + 3,
+				      WORD_MASK, f->fs.nat_lip[3] |
+				      f->fs.nat_lip[2] << 8 |
+				      f->fs.nat_lip[1] << 16 |
+				      f->fs.nat_lip[0] << 24, 1);
+		} else {
+			set_tcb_field(adap, f, tid, TCB_RX_FRAG3_LEN_RAW_W,
+				      WORD_MASK, f->fs.nat_lip[3] |
+				      f->fs.nat_lip[2] << 8 |
+				      f->fs.nat_lip[1] << 16 |
+				      f->fs.nat_lip[0] << 24, 1);
+		}
+	}
+
+	if (sip) {
+		if (f->fs.type) {
+			set_tcb_field(adap, f, tid, TCB_RX_FRAG2_PTR_RAW_W,
+				      WORD_MASK, f->fs.nat_fip[15] |
+				      f->fs.nat_fip[14] << 8 |
+				      f->fs.nat_fip[13] << 16 |
+				      f->fs.nat_fip[12] << 24, 1);
+
+			set_tcb_field(adap, f, tid, TCB_RX_FRAG2_PTR_RAW_W + 1,
+				      WORD_MASK, f->fs.nat_fip[11] |
+				      f->fs.nat_fip[10] << 8 |
+				      f->fs.nat_fip[9] << 16 |
+				      f->fs.nat_fip[8] << 24, 1);
+
+			set_tcb_field(adap, f, tid, TCB_RX_FRAG2_PTR_RAW_W + 2,
+				      WORD_MASK, f->fs.nat_fip[7] |
+				      f->fs.nat_fip[6] << 8 |
+				      f->fs.nat_fip[5] << 16 |
+				      f->fs.nat_fip[4] << 24, 1);
+
+			set_tcb_field(adap, f, tid, TCB_RX_FRAG2_PTR_RAW_W + 3,
+				      WORD_MASK, f->fs.nat_fip[3] |
+				      f->fs.nat_fip[2] << 8 |
+				      f->fs.nat_fip[1] << 16 |
+				      f->fs.nat_fip[0] << 24, 1);
+
+		} else {
+			set_tcb_field(adap, f, tid,
+				      TCB_RX_FRAG3_START_IDX_OFFSET_RAW_W,
+				      WORD_MASK, f->fs.nat_fip[3] |
+				      f->fs.nat_fip[2] << 8 |
+				      f->fs.nat_fip[1] << 16 |
+				      f->fs.nat_fip[0] << 24, 1);
+		}
+	}
+
+	set_tcb_field(adap, f, tid, TCB_PDU_HDR_LEN_W, WORD_MASK,
+		      (dp ? f->fs.nat_lport : 0) |
+		      (sp ? f->fs.nat_fport << 16 : 0), 1);
+}
+
 /* Validate filter spec against configuration done on the card. */
 static int validate_filter(struct net_device *dev,
 			   struct ch_filter_specification *fs)
@@ -484,10 +621,8 @@ int delete_filter(struct adapter *adapter, unsigned int fidx)
 void clear_filter(struct adapter *adap, struct filter_entry *f)
 {
 	/* If the new or old filter have loopback rewriteing rules then we'll
-	 * need to free any existing Layer Two Table (L2T) entries of the old
-	 * filter rule.  The firmware will handle freeing up any Source MAC
-	 * Table (SMT) entries used for rewriting Source MAC Addresses in
-	 * loopback rules.
+	 * need to free any existing L2T, SMT, CLIP entries of filter
+	 * rule.
 	 */
 	if (f->l2t)
 		cxgb4_l2t_release(f->l2t);
@@ -495,6 +630,9 @@ void clear_filter(struct adapter *adap, struct filter_entry *f)
 	if (f->smt)
 		cxgb4_smt_release(f->smt);
 
+	if (f->fs.hash && f->fs.type)
+		cxgb4_clip_release(f->dev, (const u32 *)&f->fs.val.lip, 1);
+
 	/* The zeroing of the filter rule below clears the filter valid,
 	 * pending, locked flags, l2t pointer, etc. so it's all we need for
 	 * this operation.
@@ -564,6 +702,269 @@ static void fill_default_mask(struct ch_filter_specification *fs)
 		fs->mask.fport = ~0;
 }
 
+static u64 hash_filter_ntuple(struct ch_filter_specification *fs,
+			      struct net_device *dev)
+{
+	struct adapter *adap = netdev2adap(dev);
+	struct tp_params *tp = &adap->params.tp;
+	u64 ntuple = 0;
+
+	/* Initialize each of the fields which we care about which are present
+	 * in the Compressed Filter Tuple.
+	 */
+	if (tp->vlan_shift >= 0 && fs->mask.ivlan)
+		ntuple |= (FT_VLAN_VLD_F | fs->val.ivlan) << tp->vlan_shift;
+
+	if (tp->port_shift >= 0 && fs->mask.iport)
+		ntuple |= (u64)fs->val.iport << tp->port_shift;
+
+	if (tp->protocol_shift >= 0) {
+		if (!fs->val.proto)
+			ntuple |= (u64)IPPROTO_TCP << tp->protocol_shift;
+		else
+			ntuple |= (u64)fs->val.proto << tp->protocol_shift;
+	}
+
+	if (tp->tos_shift >= 0 && fs->mask.tos)
+		ntuple |= (u64)(fs->val.tos) << tp->tos_shift;
+
+	if (tp->vnic_shift >= 0) {
+		if ((adap->params.tp.ingress_config & VNIC_F) &&
+		    fs->mask.pfvf_vld)
+			ntuple |= (u64)((fs->val.pfvf_vld << 16) |
+					(fs->val.pf << 13) |
+					(fs->val.vf)) << tp->vnic_shift;
+		else
+			ntuple |= (u64)((fs->val.ovlan_vld << 16) |
+					(fs->val.ovlan)) << tp->vnic_shift;
+	}
+
+	if (tp->macmatch_shift >= 0 && fs->mask.macidx)
+		ntuple |= (u64)(fs->val.macidx) << tp->macmatch_shift;
+
+	if (tp->ethertype_shift >= 0 && fs->mask.ethtype)
+		ntuple |= (u64)(fs->val.ethtype) << tp->ethertype_shift;
+
+	if (tp->matchtype_shift >= 0 && fs->mask.matchtype)
+		ntuple |= (u64)(fs->val.matchtype) << tp->matchtype_shift;
+
+	if (tp->frag_shift >= 0 && fs->mask.frag)
+		ntuple |= (u64)(fs->val.frag) << tp->frag_shift;
+
+	if (tp->fcoe_shift >= 0 && fs->mask.fcoe)
+		ntuple |= (u64)(fs->val.fcoe) << tp->fcoe_shift;
+	return ntuple;
+}
+
+static void mk_act_open_req6(struct filter_entry *f, struct sk_buff *skb,
+			     unsigned int qid_filterid, struct adapter *adap)
+{
+	struct cpl_t6_act_open_req6 *t6req = NULL;
+	struct cpl_act_open_req6 *req = NULL;
+
+	t6req = (struct cpl_t6_act_open_req6 *)__skb_put(skb, sizeof(*t6req));
+	INIT_TP_WR(t6req, 0);
+	req = (struct cpl_act_open_req6 *)t6req;
+	OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_ACT_OPEN_REQ6, qid_filterid));
+	req->local_port = cpu_to_be16(f->fs.val.lport);
+	req->peer_port = cpu_to_be16(f->fs.val.fport);
+	req->local_ip_hi = *(__be64 *)(&f->fs.val.lip);
+	req->local_ip_lo = *(((__be64 *)&f->fs.val.lip) + 1);
+	req->peer_ip_hi = *(__be64 *)(&f->fs.val.fip);
+	req->peer_ip_lo = *(((__be64 *)&f->fs.val.fip) + 1);
+	req->opt0 = cpu_to_be64(NAGLE_V(f->fs.newvlan == VLAN_REMOVE ||
+					f->fs.newvlan == VLAN_REWRITE) |
+				DELACK_V(f->fs.hitcnts) |
+				L2T_IDX_V(f->l2t ? f->l2t->idx : 0) |
+				SMAC_SEL_V((cxgb4_port_viid(f->dev) &
+					    0x7F) << 1) |
+				TX_CHAN_V(f->fs.eport) |
+				NO_CONG_V(f->fs.rpttid) |
+				ULP_MODE_V(f->fs.nat_mode ?
+					   ULP_MODE_TCPDDP : ULP_MODE_NONE) |
+				TCAM_BYPASS_F | NON_OFFLOAD_F);
+	t6req->params = cpu_to_be64(FILTER_TUPLE_V(hash_filter_ntuple(&f->fs,
+								      f->dev)));
+	t6req->opt2 = htonl(RSS_QUEUE_VALID_F |
+			    RSS_QUEUE_V(f->fs.iq) |
+			    TX_QUEUE_V(f->fs.nat_mode) |
+			    T5_OPT_2_VALID_F |
+			    RX_CHANNEL_F |
+			    CONG_CNTRL_V((f->fs.action == FILTER_DROP) |
+					 (f->fs.dirsteer << 1)) |
+			    PACE_V((f->fs.maskhash) |
+				   ((f->fs.dirsteerhash) << 1)) |
+			    CCTRL_ECN_V(f->fs.action == FILTER_SWITCH));
+}
+
+static void mk_act_open_req(struct filter_entry *f, struct sk_buff *skb,
+			    unsigned int qid_filterid, struct adapter *adap)
+{
+	struct cpl_t6_act_open_req *t6req = NULL;
+	struct cpl_act_open_req *req = NULL;
+
+	t6req = (struct cpl_t6_act_open_req *)__skb_put(skb, sizeof(*t6req));
+	INIT_TP_WR(t6req, 0);
+	req = (struct cpl_act_open_req *)t6req;
+	OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_ACT_OPEN_REQ, qid_filterid));
+	req->local_port = cpu_to_be16(f->fs.val.lport);
+	req->peer_port = cpu_to_be16(f->fs.val.fport);
+	req->local_ip = f->fs.val.lip[0] | f->fs.val.lip[1] << 8 |
+		f->fs.val.lip[2] << 16 | f->fs.val.lip[3] << 24;
+	req->peer_ip = f->fs.val.fip[0] | f->fs.val.fip[1] << 8 |
+		f->fs.val.fip[2] << 16 | f->fs.val.fip[3] << 24;
+	req->opt0 = cpu_to_be64(NAGLE_V(f->fs.newvlan == VLAN_REMOVE ||
+					f->fs.newvlan == VLAN_REWRITE) |
+				DELACK_V(f->fs.hitcnts) |
+				L2T_IDX_V(f->l2t ? f->l2t->idx : 0) |
+				SMAC_SEL_V((cxgb4_port_viid(f->dev) &
+					    0x7F) << 1) |
+				TX_CHAN_V(f->fs.eport) |
+				NO_CONG_V(f->fs.rpttid) |
+				ULP_MODE_V(f->fs.nat_mode ?
+					   ULP_MODE_TCPDDP : ULP_MODE_NONE) |
+				TCAM_BYPASS_F | NON_OFFLOAD_F);
+
+	t6req->params = cpu_to_be64(FILTER_TUPLE_V(hash_filter_ntuple(&f->fs,
+								      f->dev)));
+	t6req->opt2 = htonl(RSS_QUEUE_VALID_F |
+			    RSS_QUEUE_V(f->fs.iq) |
+			    TX_QUEUE_V(f->fs.nat_mode) |
+			    T5_OPT_2_VALID_F |
+			    RX_CHANNEL_F |
+			    CONG_CNTRL_V((f->fs.action == FILTER_DROP) |
+					 (f->fs.dirsteer << 1)) |
+			    PACE_V((f->fs.maskhash) |
+				   ((f->fs.dirsteerhash) << 1)) |
+			    CCTRL_ECN_V(f->fs.action == FILTER_SWITCH));
+}
+
+static int cxgb4_set_hash_filter(struct net_device *dev,
+				 struct ch_filter_specification *fs,
+				 struct filter_ctx *ctx)
+{
+	struct adapter *adapter = netdev2adap(dev);
+	struct tid_info *t = &adapter->tids;
+	struct filter_entry *f;
+	struct sk_buff *skb;
+	int iq, atid, size;
+	int ret = 0;
+	u32 iconf;
+
+	fill_default_mask(fs);
+	ret = validate_filter(dev, fs);
+	if (ret)
+		return ret;
+
+	iq = get_filter_steerq(dev, fs);
+	if (iq < 0)
+		return iq;
+
+	f = kzalloc(sizeof(*f), GFP_KERNEL);
+	if (!f)
+		return -ENOMEM;
+
+	f->fs = *fs;
+	f->ctx = ctx;
+	f->dev = dev;
+	f->fs.iq = iq;
+
+	/* If the new filter requires loopback Destination MAC and/or VLAN
+	 * rewriting then we need to allocate a Layer 2 Table (L2T) entry for
+	 * the filter.
+	 */
+	if (f->fs.newdmac || f->fs.newvlan) {
+		/* allocate L2T entry for new filter */
+		f->l2t = t4_l2t_alloc_switching(adapter, f->fs.vlan,
+						f->fs.eport, f->fs.dmac);
+		if (!f->l2t) {
+			ret = -ENOMEM;
+			goto out_err;
+		}
+	}
+
+	/* If the new filter requires loopback Source MAC rewriting then
+	 * we need to allocate a SMT entry for the filter.
+	 */
+	if (f->fs.newsmac) {
+		f->smt = cxgb4_smt_alloc_switching(f->dev, f->fs.smac);
+		if (!f->smt) {
+			if (f->l2t) {
+				cxgb4_l2t_release(f->l2t);
+				f->l2t = NULL;
+			}
+			ret = -ENOMEM;
+			goto free_l2t;
+		}
+	}
+
+	atid = cxgb4_alloc_atid(t, f);
+	if (atid < 0)
+		goto free_smt;
+
+	iconf = adapter->params.tp.ingress_config;
+	if (iconf & VNIC_F) {
+		f->fs.val.ovlan = (fs->val.pf << 13) | fs->val.vf;
+		f->fs.mask.ovlan = (fs->mask.pf << 13) | fs->mask.vf;
+		f->fs.val.ovlan_vld = fs->val.pfvf_vld;
+		f->fs.mask.ovlan_vld = fs->mask.pfvf_vld;
+	}
+
+	size = sizeof(struct cpl_t6_act_open_req);
+	if (f->fs.type) {
+		ret = cxgb4_clip_get(f->dev, (const u32 *)&f->fs.val.lip, 1);
+		if (ret)
+			goto free_atid;
+
+		skb = alloc_skb(size, GFP_KERNEL);
+		if (!skb) {
+			ret = -ENOMEM;
+			goto free_clip;
+		}
+
+		mk_act_open_req6(f, skb,
+				 ((adapter->sge.fw_evtq.abs_id << 14) | atid),
+				 adapter);
+	} else {
+		skb = alloc_skb(size, GFP_KERNEL);
+		if (!skb) {
+			ret = -ENOMEM;
+			goto free_atid;
+		}
+
+		mk_act_open_req(f, skb,
+				((adapter->sge.fw_evtq.abs_id << 14) | atid),
+				adapter);
+	}
+
+	f->pending = 1;
+	set_wr_txq(skb, CPL_PRIORITY_SETUP, f->fs.val.iport & 0x3);
+	t4_ofld_send(adapter, skb);
+	return 0;
+
+free_clip:
+	cxgb4_clip_release(f->dev, (const u32 *)&f->fs.val.lip, 1);
+
+free_atid:
+	cxgb4_free_atid(t, atid);
+
+free_smt:
+	if (f->smt) {
+		cxgb4_smt_release(f->smt);
+		f->smt = NULL;
+	}
+
+free_l2t:
+	if (f->l2t) {
+		cxgb4_l2t_release(f->l2t);
+		f->l2t = NULL;
+	}
+
+out_err:
+	kfree(f);
+	return ret;
+}
+
 /* Check a Chelsio Filter Request for validity, convert it into our internal
  * format and send it to the hardware.  Return 0 on success, an error number
  * otherwise.  We attach any provided filter operation context to the internal
@@ -580,6 +981,14 @@ int __cxgb4_set_filter(struct net_device *dev, int filter_id,
 	u32 iconf;
 	int iq, ret;
 
+	if (fs->hash) {
+		if (is_hashfilter(adapter))
+			return cxgb4_set_hash_filter(dev, fs, ctx);
+		netdev_err(dev, "%s: Exact-match filters only supported with Hash Filter configuration\n",
+			   __func__);
+		return -EINVAL;
+	}
+
 	max_fidx = adapter->tids.nftids;
 	if (filter_id != (max_fidx + adapter->tids.nsftids - 1) &&
 	    filter_id >= max_fidx)
@@ -789,60 +1198,126 @@ int cxgb4_del_filter(struct net_device *dev, int filter_id)
 	return ret;
 }
 
-static int set_tcb_field(struct adapter *adap, struct filter_entry *f,
-			 unsigned int ftid,  u16 word, u64 mask, u64 val,
-			 int no_reply)
+static int configure_filter_tcb(struct adapter *adap, unsigned int tid,
+				struct filter_entry *f)
 {
-	struct cpl_set_tcb_field *req;
-	struct sk_buff *skb;
-
-	skb = alloc_skb(sizeof(struct cpl_set_tcb_field), GFP_ATOMIC);
-	if (!skb)
-		return -ENOMEM;
-
-	req = (struct cpl_set_tcb_field *)__skb_put(skb, sizeof(*req));
-	memset(req, 0, sizeof(*req));
-	INIT_TP_WR_CPL(req, CPL_SET_TCB_FIELD, ftid);
-	req->reply_ctrl = htons(REPLY_CHAN_V(0) |
-				QUEUENO_V(adap->sge.fw_evtq.abs_id) |
-				NO_REPLY_V(no_reply));
-	req->word_cookie = htons(TCB_WORD_V(word) | TCB_COOKIE_V(ftid));
-	req->mask = cpu_to_be64(mask);
-	req->val = cpu_to_be64(val);
-	set_wr_txq(skb, CPL_PRIORITY_CONTROL, f->fs.val.iport & 0x3);
-	t4_ofld_send(adap, skb);
+	if (f->fs.hitcnts)
+		set_tcb_field(adap, f, tid, TCB_TIMESTAMP_W,
+			      TCB_TIMESTAMP_V(TCB_TIMESTAMP_M) |
+			      TCB_RTT_TS_RECENT_AGE_V(TCB_RTT_TS_RECENT_AGE_M),
+			      TCB_TIMESTAMP_V(0ULL) |
+			      TCB_RTT_TS_RECENT_AGE_V(0ULL),
+			      1);
+
+	if (f->fs.newdmac)
+		set_tcb_tflag(adap, f, tid, TF_CCTRL_ECE_S, 1,
+			      1);
+
+	if (f->fs.newvlan == VLAN_INSERT ||
+	    f->fs.newvlan == VLAN_REWRITE)
+		set_tcb_tflag(adap, f, tid, TF_CCTRL_RFR_S, 1,
+			      1);
+	if (f->fs.newsmac)
+		configure_filter_smac(adap, f);
+
+	if (f->fs.nat_mode) {
+		switch (f->fs.nat_mode) {
+		case NAT_MODE_DIP:
+			set_nat_params(adap, f, tid, true, false, false, false);
+			break;
+
+		case NAT_MODE_DIP_DP:
+			set_nat_params(adap, f, tid, true, false, true, false);
+			break;
+
+		case NAT_MODE_DIP_DP_SIP:
+			set_nat_params(adap, f, tid, true, true, true, false);
+			break;
+		case NAT_MODE_DIP_DP_SP:
+			set_nat_params(adap, f, tid, true, false, true, true);
+			break;
+
+		case NAT_MODE_SIP_SP:
+			set_nat_params(adap, f, tid, false, true, false, true);
+			break;
+
+		case NAT_MODE_DIP_SIP_SP:
+			set_nat_params(adap, f, tid, true, true, false, true);
+			break;
+
+		case NAT_MODE_ALL:
+			set_nat_params(adap, f, tid, true, true, true, true);
+			break;
+
+		default:
+			pr_err("%s: Invalid NAT mode: %d\n",
+			       __func__, f->fs.nat_mode);
+			return -EINVAL;
+		}
+	}
 	return 0;
 }
 
-/* Set one of the t_flags bits in the TCB.
- */
-static int set_tcb_tflag(struct adapter *adap, struct filter_entry *f,
-			 unsigned int ftid, unsigned int bit_pos,
-			 unsigned int val, int no_reply)
+void hash_filter_rpl(struct adapter *adap, const struct cpl_act_open_rpl *rpl)
 {
-	return set_tcb_field(adap, f, ftid,  TCB_T_FLAGS_W, 1ULL << bit_pos,
-			     (unsigned long long)val << bit_pos, no_reply);
-}
+	unsigned int ftid = TID_TID_G(AOPEN_ATID_G(ntohl(rpl->atid_status)));
+	unsigned int status  = AOPEN_STATUS_G(ntohl(rpl->atid_status));
+	struct tid_info *t = &adap->tids;
+	unsigned int tid = GET_TID(rpl);
+	struct filter_ctx *ctx = NULL;
+	struct filter_entry *f;
 
-static int configure_filter_smac(struct adapter *adap, struct filter_entry *f)
-{
-	int err;
+	dev_dbg(adap->pdev_dev, "%s: tid = %u; atid = %u; status = %u\n",
+		__func__, tid, ftid, status);
 
-	/* do a set-tcb for smac-sel and CWR bit.. */
-	err = set_tcb_tflag(adap, f, f->tid, TF_CCTRL_CWR_S, 1, 1);
-	if (err)
-		goto smac_err;
+	f = lookup_atid(t, ftid);
+	if (!f) {
+		dev_err(adap->pdev_dev, "%s:could not find filter entry",
+			__func__);
+		return;
+	}
+	ctx = f->ctx;
+	f->ctx = NULL;
+
+	switch (status) {
+	case CPL_ERR_NONE:
+		f->tid = tid;
+		f->pending = 0;
+		f->valid = 1;
+		cxgb4_insert_tid(t, f, f->tid, 0);
+		cxgb4_free_atid(t, ftid);
+		if (ctx) {
+			ctx->tid = f->tid;
+			ctx->result = 0;
+		}
+		if (configure_filter_tcb(adap, tid, f)) {
+			clear_filter(adap, f);
+			cxgb4_remove_tid(t, 0, tid, 0);
+			kfree(f);
+			if (ctx) {
+				ctx->result = -EINVAL;
+				complete(&ctx->completion);
+			}
+			return;
+		}
+		break;
 
-	err = set_tcb_field(adap, f, f->tid, TCB_SMAC_SEL_W,
-			    TCB_SMAC_SEL_V(TCB_SMAC_SEL_M),
-			    TCB_SMAC_SEL_V(f->smt->idx), 1);
-	if (!err)
-		return 0;
+	default:
+		dev_err(adap->pdev_dev, "%s: filter creation PROBLEM; status = %u\n",
+			__func__, status);
 
-smac_err:
-	dev_err(adap->pdev_dev, "filter %u smac config failed with error %u\n",
-		f->tid, err);
-	return err;
+		if (ctx) {
+			if (status == CPL_ERR_TCAM_FULL)
+				ctx->result = -EAGAIN;
+			else
+				ctx->result = -EINVAL;
+		}
+		clear_filter(adap, f);
+		cxgb4_free_atid(t, ftid);
+		kfree(f);
+	}
+	if (ctx)
+		complete(&ctx->completion);
 }
 
 /* Handle a filter write/deletion reply. */
diff --git a/drivers/net/ethernet/chelsio/cxgb4/cxgb4_filter.h b/drivers/net/ethernet/chelsio/cxgb4/cxgb4_filter.h
index d3c1a8fafd32..7480d65550a8 100644
--- a/drivers/net/ethernet/chelsio/cxgb4/cxgb4_filter.h
+++ b/drivers/net/ethernet/chelsio/cxgb4/cxgb4_filter.h
@@ -37,7 +37,10 @@
 
 #include "t4_msg.h"
 
+#define WORD_MASK	0xffffffff
+
 void filter_rpl(struct adapter *adap, const struct cpl_set_tcb_rpl *rpl);
+void hash_filter_rpl(struct adapter *adap, const struct cpl_act_open_rpl *rpl);
 void clear_filter(struct adapter *adap, struct filter_entry *f);
 
 int set_filter_wr(struct adapter *adapter, int fidx);
diff --git a/drivers/net/ethernet/chelsio/cxgb4/cxgb4_main.c b/drivers/net/ethernet/chelsio/cxgb4/cxgb4_main.c
index 4b07cfe8c66c..77b4bd958748 100644
--- a/drivers/net/ethernet/chelsio/cxgb4/cxgb4_main.c
+++ b/drivers/net/ethernet/chelsio/cxgb4/cxgb4_main.c
@@ -572,6 +572,10 @@ static int fwevtq_handler(struct sge_rspq *q, const __be64 *rsp,
 		const struct cpl_set_tcb_rpl *p = (void *)rsp;
 
 		filter_rpl(q->adap, p);
+	} else if (opcode == CPL_ACT_OPEN_RPL) {
+		const struct cpl_act_open_rpl *p = (void *)rsp;
+
+		hash_filter_rpl(q->adap, p);
 	} else
 		dev_err(q->adap->pdev_dev,
 			"unexpected CPL %#x on FW event queue\n", opcode);
diff --git a/drivers/net/ethernet/chelsio/cxgb4/t4_msg.h b/drivers/net/ethernet/chelsio/cxgb4/t4_msg.h
index ce4838d907da..7e12f241145b 100644
--- a/drivers/net/ethernet/chelsio/cxgb4/t4_msg.h
+++ b/drivers/net/ethernet/chelsio/cxgb4/t4_msg.h
@@ -286,6 +286,7 @@ struct work_request_hdr {
 
 #define RX_CHANNEL_S    26
 #define RX_CHANNEL_V(x) ((x) << RX_CHANNEL_S)
+#define RX_CHANNEL_F	RX_CHANNEL_V(1U)
 
 #define WND_SCALE_EN_S    28
 #define WND_SCALE_EN_V(x) ((x) << WND_SCALE_EN_S)
@@ -315,6 +316,10 @@ struct cpl_pass_open_req {
 #define DELACK_V(x) ((x) << DELACK_S)
 #define DELACK_F    DELACK_V(1U)
 
+#define NON_OFFLOAD_S		7
+#define NON_OFFLOAD_V(x)	((x) << NON_OFFLOAD_S)
+#define NON_OFFLOAD_F		NON_OFFLOAD_V(1U)
+
 #define DSCP_S    22
 #define DSCP_M    0x3F
 #define DSCP_V(x) ((x) << DSCP_S)
diff --git a/drivers/net/ethernet/chelsio/cxgb4/t4_tcb.h b/drivers/net/ethernet/chelsio/cxgb4/t4_tcb.h
index c1c76663034d..c7201eb7b14c 100644
--- a/drivers/net/ethernet/chelsio/cxgb4/t4_tcb.h
+++ b/drivers/net/ethernet/chelsio/cxgb4/t4_tcb.h
@@ -42,6 +42,23 @@
 
 #define TCB_T_FLAGS_W		1
 
+#define TF_CCTRL_ECE_S		60
 #define TF_CCTRL_CWR_S		61
+#define TF_CCTRL_RFR_S		62
 
+#define TCB_TIMESTAMP_W		5
+#define TCB_TIMESTAMP_S		0
+#define TCB_TIMESTAMP_M		0xffffffffULL
+#define TCB_TIMESTAMP_V(x)	((x) << TCB_TIMESTAMP_S)
+
+#define TCB_RTT_TS_RECENT_AGE_W		6
+#define TCB_RTT_TS_RECENT_AGE_S		0
+#define TCB_RTT_TS_RECENT_AGE_M		0xffffffffULL
+#define TCB_RTT_TS_RECENT_AGE_V(x)	((x) << TCB_RTT_TS_RECENT_AGE_S)
+
+#define TCB_SND_UNA_RAW_W	10
+#define TCB_RX_FRAG2_PTR_RAW_W	27
+#define TCB_RX_FRAG3_LEN_RAW_W	29
+#define TCB_RX_FRAG3_START_IDX_OFFSET_RAW_W	30
+#define TCB_PDU_HDR_LEN_W	31
 #endif /* __T4_TCB_H */
-- 
2.14.1

[PATCH net-next 4/7] cxgb4: add support to delete hash filter

[Rahul Lakkireddy]

From: Kumar Sanghvi <kumaras@chelsio.com>

Use a combined ulptx work-request to send hash filter deletion
request to hw. Hash filter deletion reply is processed on
getting cpl_abort_rpl_rss.

Release any L2T/SMT/CLIP entries on filter deletion.
Also, free up the corresponding filter entry.

Signed-off-by: Kumar Sanghvi <kumaras@chelsio.com>
Signed-off-by: Rahul Lakkireddy <rahul.lakkireddy@chelsio.com>
Signed-off-by: Ganesh Goudar <ganeshgr@chelsio.com>
---
 drivers/net/ethernet/chelsio/cxgb4/cxgb4_filter.c  | 159 ++++++++++++++++++++-
 drivers/net/ethernet/chelsio/cxgb4/cxgb4_filter.h  |   2 +
 drivers/net/ethernet/chelsio/cxgb4/cxgb4_main.c    |   4 +
 .../net/ethernet/chelsio/cxgb4/cxgb4_tc_flower.c   |   2 +-
 drivers/net/ethernet/chelsio/cxgb4/cxgb4_tc_u32.c  |   4 +-
 drivers/net/ethernet/chelsio/cxgb4/cxgb4_uld.h     |   4 +-
 drivers/net/ethernet/chelsio/cxgb4/t4_tcb.h        |   5 +
 7 files changed, 173 insertions(+), 7 deletions(-)

diff --git a/drivers/net/ethernet/chelsio/cxgb4/cxgb4_filter.c b/drivers/net/ethernet/chelsio/cxgb4/cxgb4_filter.c
index cf8ca695f27d..eb6ba9824501 100644
--- a/drivers/net/ethernet/chelsio/cxgb4/cxgb4_filter.c
+++ b/drivers/net/ethernet/chelsio/cxgb4/cxgb4_filter.c
@@ -87,6 +87,59 @@ static int set_tcb_tflag(struct adapter *adap, struct filter_entry *f,
 			     (unsigned long long)val << bit_pos, no_reply);
 }
 
+static void mk_abort_req_ulp(struct cpl_abort_req *abort_req, unsigned int tid)
+{
+	struct ulp_txpkt *txpkt = (struct ulp_txpkt *)abort_req;
+	struct ulptx_idata *sc = (struct ulptx_idata *)(txpkt + 1);
+
+	txpkt->cmd_dest = htonl(ULPTX_CMD_V(ULP_TX_PKT) | ULP_TXPKT_DEST_V(0));
+	txpkt->len = htonl(DIV_ROUND_UP(sizeof(*abort_req), 16));
+	sc->cmd_more = htonl(ULPTX_CMD_V(ULP_TX_SC_IMM));
+	sc->len = htonl(sizeof(*abort_req) - sizeof(struct work_request_hdr));
+	OPCODE_TID(abort_req) = htonl(MK_OPCODE_TID(CPL_ABORT_REQ, tid));
+	abort_req->rsvd0 = htonl(0);
+	abort_req->rsvd1 = 0;
+	abort_req->cmd = CPL_ABORT_NO_RST;
+}
+
+static void mk_abort_rpl_ulp(struct cpl_abort_rpl *abort_rpl, unsigned int tid)
+{
+	struct ulp_txpkt *txpkt = (struct ulp_txpkt *)abort_rpl;
+	struct ulptx_idata *sc = (struct ulptx_idata *)(txpkt + 1);
+
+	txpkt->cmd_dest = htonl(ULPTX_CMD_V(ULP_TX_PKT) | ULP_TXPKT_DEST_V(0));
+	txpkt->len = htonl(DIV_ROUND_UP(sizeof(*abort_rpl), 16));
+	sc->cmd_more = htonl(ULPTX_CMD_V(ULP_TX_SC_IMM));
+	sc->len = htonl(sizeof(*abort_rpl) - sizeof(struct work_request_hdr));
+	OPCODE_TID(abort_rpl) = htonl(MK_OPCODE_TID(CPL_ABORT_RPL, tid));
+	abort_rpl->rsvd0 = htonl(0);
+	abort_rpl->rsvd1 = 0;
+	abort_rpl->cmd = CPL_ABORT_NO_RST;
+}
+
+static void mk_set_tcb_ulp(struct filter_entry *f,
+			   struct cpl_set_tcb_field *req,
+			   unsigned int word, u64 mask, u64 val,
+			   u8 cookie, int no_reply)
+{
+	struct ulp_txpkt *txpkt = (struct ulp_txpkt *)req;
+	struct ulptx_idata *sc = (struct ulptx_idata *)(txpkt + 1);
+
+	txpkt->cmd_dest = htonl(ULPTX_CMD_V(ULP_TX_PKT) | ULP_TXPKT_DEST_V(0));
+	txpkt->len = htonl(DIV_ROUND_UP(sizeof(*req), 16));
+	sc->cmd_more = htonl(ULPTX_CMD_V(ULP_TX_SC_IMM));
+	sc->len = htonl(sizeof(*req) - sizeof(struct work_request_hdr));
+	OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_SET_TCB_FIELD, f->tid));
+	req->reply_ctrl = htons(NO_REPLY_V(no_reply) | REPLY_CHAN_V(0) |
+				QUEUENO_V(0));
+	req->word_cookie = htons(TCB_WORD_V(word) | TCB_COOKIE_V(cookie));
+	req->mask = cpu_to_be64(mask);
+	req->val = cpu_to_be64(val);
+	sc = (struct ulptx_idata *)(req + 1);
+	sc->cmd_more = htonl(ULPTX_CMD_V(ULP_TX_SC_NOOP));
+	sc->len = htonl(0);
+}
+
 static int configure_filter_smac(struct adapter *adap, struct filter_entry *f)
 {
 	int err;
@@ -1110,12 +1163,74 @@ int __cxgb4_set_filter(struct net_device *dev, int filter_id,
 	return ret;
 }
 
+static int cxgb4_del_hash_filter(struct net_device *dev, int filter_id,
+				 struct filter_ctx *ctx)
+{
+	struct adapter *adapter = netdev2adap(dev);
+	struct tid_info *t = &adapter->tids;
+	struct cpl_abort_req *abort_req;
+	struct cpl_abort_rpl *abort_rpl;
+	struct cpl_set_tcb_field *req;
+	struct ulptx_idata *aligner;
+	struct work_request_hdr *wr;
+	struct filter_entry *f;
+	struct sk_buff *skb;
+	unsigned int wrlen;
+	int ret;
+
+	netdev_dbg(dev, "%s: filter_id = %d ; nftids = %d\n",
+		   __func__, filter_id, adapter->tids.nftids);
+
+	if (filter_id > adapter->tids.ntids)
+		return -E2BIG;
+
+	f = lookup_tid(t, filter_id);
+	if (!f) {
+		netdev_err(dev, "%s: no filter entry for filter_id = %d",
+			   __func__, filter_id);
+		return -EINVAL;
+	}
+
+	ret = writable_filter(f);
+	if (ret)
+		return ret;
+
+	if (!f->valid)
+		return -EINVAL;
+
+	f->ctx = ctx;
+	f->pending = 1;
+	wrlen = roundup(sizeof(*wr) + (sizeof(*req) + sizeof(*aligner))
+			+ sizeof(*abort_req) + sizeof(*abort_rpl), 16);
+	skb = alloc_skb(wrlen, GFP_KERNEL);
+	if (!skb) {
+		netdev_err(dev, "%s: could not allocate skb ..\n", __func__);
+		return -ENOMEM;
+	}
+	set_wr_txq(skb, CPL_PRIORITY_CONTROL, f->fs.val.iport & 0x3);
+	req = (struct cpl_set_tcb_field *)__skb_put(skb, wrlen);
+	INIT_ULPTX_WR(req, wrlen, 0, 0);
+	wr = (struct work_request_hdr *)req;
+	wr++;
+	req = (struct cpl_set_tcb_field *)wr;
+	mk_set_tcb_ulp(f, req, TCB_RSS_INFO_W, TCB_RSS_INFO_V(TCB_RSS_INFO_M),
+		       TCB_RSS_INFO_V(adapter->sge.fw_evtq.abs_id), 0, 1);
+	aligner = (struct ulptx_idata *)(req + 1);
+	abort_req = (struct cpl_abort_req *)(aligner + 1);
+	mk_abort_req_ulp(abort_req, f->tid);
+	abort_rpl = (struct cpl_abort_rpl *)(abort_req + 1);
+	mk_abort_rpl_ulp(abort_rpl, f->tid);
+	t4_ofld_send(adapter, skb);
+	return 0;
+}
+
 /* Check a delete filter request for validity and send it to the hardware.
  * Return 0 on success, an error number otherwise.  We attach any provided
  * filter operation context to the internal filter specification in order to
  * facilitate signaling completion of the operation.
  */
 int __cxgb4_del_filter(struct net_device *dev, int filter_id,
+		       struct ch_filter_specification *fs,
 		       struct filter_ctx *ctx)
 {
 	struct adapter *adapter = netdev2adap(dev);
@@ -1123,6 +1238,14 @@ int __cxgb4_del_filter(struct net_device *dev, int filter_id,
 	unsigned int max_fidx;
 	int ret;
 
+	if (fs && fs->hash) {
+		if (is_hashfilter(adapter))
+			return cxgb4_del_hash_filter(dev, filter_id, ctx);
+		netdev_err(dev, "%s: Exact-match filters only supported with Hash Filter configuration\n",
+			   __func__);
+		return -EINVAL;
+	}
+
 	max_fidx = adapter->tids.nftids;
 	if (filter_id != (max_fidx + adapter->tids.nsftids - 1) &&
 	    filter_id >= max_fidx)
@@ -1173,18 +1296,19 @@ int cxgb4_set_filter(struct net_device *dev, int filter_id,
 	return ret;
 }
 
-int cxgb4_del_filter(struct net_device *dev, int filter_id)
+int cxgb4_del_filter(struct net_device *dev, int filter_id,
+		     struct ch_filter_specification *fs)
 {
 	struct filter_ctx ctx;
 	int ret;
 
 	/* If we are shutting down the adapter do not wait for completion */
 	if (netdev2adap(dev)->flags & SHUTTING_DOWN)
-		return __cxgb4_del_filter(dev, filter_id, NULL);
+		return __cxgb4_del_filter(dev, filter_id, fs, NULL);
 
 	init_completion(&ctx.completion);
 
-	ret = __cxgb4_del_filter(dev, filter_id, &ctx);
+	ret = __cxgb4_del_filter(dev, filter_id, fs, &ctx);
 	if (ret)
 		goto out;
 
@@ -1258,6 +1382,35 @@ static int configure_filter_tcb(struct adapter *adap, unsigned int tid,
 	return 0;
 }
 
+void hash_del_filter_rpl(struct adapter *adap,
+			 const struct cpl_abort_rpl_rss *rpl)
+{
+	unsigned int status = rpl->status;
+	struct tid_info *t = &adap->tids;
+	unsigned int tid = GET_TID(rpl);
+	struct filter_ctx *ctx = NULL;
+	struct filter_entry *f;
+
+	dev_dbg(adap->pdev_dev, "%s: status = %u; tid = %u\n",
+		__func__, status, tid);
+
+	f = lookup_tid(t, tid);
+	if (!f) {
+		dev_err(adap->pdev_dev, "%s:could not find filter entry",
+			__func__);
+		return;
+	}
+	ctx = f->ctx;
+	f->ctx = NULL;
+	clear_filter(adap, f);
+	cxgb4_remove_tid(t, 0, tid, 0);
+	kfree(f);
+	if (ctx) {
+		ctx->result = 0;
+		complete(&ctx->completion);
+	}
+}
+
 void hash_filter_rpl(struct adapter *adap, const struct cpl_act_open_rpl *rpl)
 {
 	unsigned int ftid = TID_TID_G(AOPEN_ATID_G(ntohl(rpl->atid_status)));
diff --git a/drivers/net/ethernet/chelsio/cxgb4/cxgb4_filter.h b/drivers/net/ethernet/chelsio/cxgb4/cxgb4_filter.h
index 7480d65550a8..9475abd3384e 100644
--- a/drivers/net/ethernet/chelsio/cxgb4/cxgb4_filter.h
+++ b/drivers/net/ethernet/chelsio/cxgb4/cxgb4_filter.h
@@ -41,6 +41,8 @@
 
 void filter_rpl(struct adapter *adap, const struct cpl_set_tcb_rpl *rpl);
 void hash_filter_rpl(struct adapter *adap, const struct cpl_act_open_rpl *rpl);
+void hash_del_filter_rpl(struct adapter *adap,
+			 const struct cpl_abort_rpl_rss *rpl);
 void clear_filter(struct adapter *adap, struct filter_entry *f);
 
 int set_filter_wr(struct adapter *adapter, int fidx);
diff --git a/drivers/net/ethernet/chelsio/cxgb4/cxgb4_main.c b/drivers/net/ethernet/chelsio/cxgb4/cxgb4_main.c
index 77b4bd958748..35709c7f7c5b 100644
--- a/drivers/net/ethernet/chelsio/cxgb4/cxgb4_main.c
+++ b/drivers/net/ethernet/chelsio/cxgb4/cxgb4_main.c
@@ -576,6 +576,10 @@ static int fwevtq_handler(struct sge_rspq *q, const __be64 *rsp,
 		const struct cpl_act_open_rpl *p = (void *)rsp;
 
 		hash_filter_rpl(q->adap, p);
+	} else if (opcode == CPL_ABORT_RPL_RSS) {
+		const struct cpl_abort_rpl_rss *p = (void *)rsp;
+
+		hash_del_filter_rpl(q->adap, p);
 	} else
 		dev_err(q->adap->pdev_dev,
 			"unexpected CPL %#x on FW event queue\n", opcode);
diff --git a/drivers/net/ethernet/chelsio/cxgb4/cxgb4_tc_flower.c b/drivers/net/ethernet/chelsio/cxgb4/cxgb4_tc_flower.c
index 614db014ef18..ed377e2e9f8a 100644
--- a/drivers/net/ethernet/chelsio/cxgb4/cxgb4_tc_flower.c
+++ b/drivers/net/ethernet/chelsio/cxgb4/cxgb4_tc_flower.c
@@ -730,7 +730,7 @@ int cxgb4_tc_flower_destroy(struct net_device *dev,
 	if (!ch_flower)
 		return -ENOENT;
 
-	ret = cxgb4_del_filter(dev, ch_flower->filter_id);
+	ret = cxgb4_del_filter(dev, ch_flower->filter_id, &ch_flower->fs);
 	if (ret)
 		goto err;
 
diff --git a/drivers/net/ethernet/chelsio/cxgb4/cxgb4_tc_u32.c b/drivers/net/ethernet/chelsio/cxgb4/cxgb4_tc_u32.c
index 48970ba08bdc..cd0cd13a964d 100644
--- a/drivers/net/ethernet/chelsio/cxgb4/cxgb4_tc_u32.c
+++ b/drivers/net/ethernet/chelsio/cxgb4/cxgb4_tc_u32.c
@@ -380,7 +380,7 @@ int cxgb4_delete_knode(struct net_device *dev, struct tc_cls_u32_offload *cls)
 			return -EINVAL;
 	}
 
-	ret = cxgb4_del_filter(dev, filter_id);
+	ret = cxgb4_del_filter(dev, filter_id, NULL);
 	if (ret)
 		goto out;
 
@@ -399,7 +399,7 @@ int cxgb4_delete_knode(struct net_device *dev, struct tc_cls_u32_offload *cls)
 				if (!test_bit(j, link->tid_map))
 					continue;
 
-				ret = __cxgb4_del_filter(dev, j, NULL);
+				ret = __cxgb4_del_filter(dev, j, NULL, NULL);
 				if (ret)
 					goto out;
 
diff --git a/drivers/net/ethernet/chelsio/cxgb4/cxgb4_uld.h b/drivers/net/ethernet/chelsio/cxgb4/cxgb4_uld.h
index 52324c77a4fe..a1c850861cbf 100644
--- a/drivers/net/ethernet/chelsio/cxgb4/cxgb4_uld.h
+++ b/drivers/net/ethernet/chelsio/cxgb4/cxgb4_uld.h
@@ -217,10 +217,12 @@ int __cxgb4_set_filter(struct net_device *dev, int filter_id,
 		       struct ch_filter_specification *fs,
 		       struct filter_ctx *ctx);
 int __cxgb4_del_filter(struct net_device *dev, int filter_id,
+		       struct ch_filter_specification *fs,
 		       struct filter_ctx *ctx);
 int cxgb4_set_filter(struct net_device *dev, int filter_id,
 		     struct ch_filter_specification *fs);
-int cxgb4_del_filter(struct net_device *dev, int filter_id);
+int cxgb4_del_filter(struct net_device *dev, int filter_id,
+		     struct ch_filter_specification *fs);
 int cxgb4_get_filter_counters(struct net_device *dev, unsigned int fidx,
 			      u64 *hitcnt, u64 *bytecnt);
 
diff --git a/drivers/net/ethernet/chelsio/cxgb4/t4_tcb.h b/drivers/net/ethernet/chelsio/cxgb4/t4_tcb.h
index c7201eb7b14c..3297ce025e8b 100644
--- a/drivers/net/ethernet/chelsio/cxgb4/t4_tcb.h
+++ b/drivers/net/ethernet/chelsio/cxgb4/t4_tcb.h
@@ -46,6 +46,11 @@
 #define TF_CCTRL_CWR_S		61
 #define TF_CCTRL_RFR_S		62
 
+#define TCB_RSS_INFO_W		3
+#define TCB_RSS_INFO_S		0
+#define TCB_RSS_INFO_M		0x3ffULL
+#define TCB_RSS_INFO_V(x)	((x) << TCB_RSS_INFO_S)
+
 #define TCB_TIMESTAMP_W		5
 #define TCB_TIMESTAMP_S		0
 #define TCB_TIMESTAMP_M		0xffffffffULL
-- 
2.14.1

[PATCH net-next 5/7] cxgb4: add support to retrieve stats for hash filters

[Rahul Lakkireddy]

From: Kumar Sanghvi <kumaras@chelsio.com>

Add support to retrieve packet-count and byte-count for hash-filters
by retrieving filter-entry appropriately based on whether the
request is for hash-filter or not.

Signed-off-by: Kumar Sanghvi <kumaras@chelsio.com>
Signed-off-by: Rahul Lakkireddy <rahul.lakkireddy@chelsio.com>
Signed-off-by: Ganesh Goudar <ganeshgr@chelsio.com>
---
 drivers/net/ethernet/chelsio/cxgb4/cxgb4_filter.c  | 30 ++++++++++++++--------
 .../net/ethernet/chelsio/cxgb4/cxgb4_tc_flower.c   |  6 +++--
 drivers/net/ethernet/chelsio/cxgb4/cxgb4_uld.h     |  2 +-
 3 files changed, 25 insertions(+), 13 deletions(-)

diff --git a/drivers/net/ethernet/chelsio/cxgb4/cxgb4_filter.c b/drivers/net/ethernet/chelsio/cxgb4/cxgb4_filter.c
index eb6ba9824501..9b3ff6209eb5 100644
--- a/drivers/net/ethernet/chelsio/cxgb4/cxgb4_filter.c
+++ b/drivers/net/ethernet/chelsio/cxgb4/cxgb4_filter.c
@@ -341,7 +341,7 @@ static int get_filter_steerq(struct net_device *dev,
 }
 
 static int get_filter_count(struct adapter *adapter, unsigned int fidx,
-			    u64 *pkts, u64 *bytes)
+			    u64 *pkts, u64 *bytes, bool hash)
 {
 	unsigned int tcb_base, tcbaddr;
 	unsigned int word_offset;
@@ -350,14 +350,24 @@ static int get_filter_count(struct adapter *adapter, unsigned int fidx,
 	int ret;
 
 	tcb_base = t4_read_reg(adapter, TP_CMM_TCB_BASE_A);
-	if ((fidx != (adapter->tids.nftids + adapter->tids.nsftids - 1)) &&
-	    fidx >= adapter->tids.nftids)
-		return -E2BIG;
-
-	f = &adapter->tids.ftid_tab[fidx];
-	if (!f->valid)
-		return -EINVAL;
+	if (is_hashfilter(adapter) && hash) {
+		if (fidx < adapter->tids.ntids) {
+			f = adapter->tids.tid_tab[fidx];
+			if (!f)
+				return -EINVAL;
+		} else {
+			return -E2BIG;
+		}
+	} else {
+		if ((fidx != (adapter->tids.nftids +
+			      adapter->tids.nsftids - 1)) &&
+		    fidx >= adapter->tids.nftids)
+			return -E2BIG;
 
+		f = &adapter->tids.ftid_tab[fidx];
+		if (!f->valid)
+			return -EINVAL;
+	}
 	tcbaddr = tcb_base + f->tid * TCB_SIZE;
 
 	spin_lock(&adapter->win0_lock);
@@ -409,11 +419,11 @@ static int get_filter_count(struct adapter *adapter, unsigned int fidx,
 }
 
 int cxgb4_get_filter_counters(struct net_device *dev, unsigned int fidx,
-			      u64 *hitcnt, u64 *bytecnt)
+			      u64 *hitcnt, u64 *bytecnt, bool hash)
 {
 	struct adapter *adapter = netdev2adap(dev);
 
-	return get_filter_count(adapter, fidx, hitcnt, bytecnt);
+	return get_filter_count(adapter, fidx, hitcnt, bytecnt, hash);
 }
 
 int cxgb4_get_free_ftid(struct net_device *dev, int family)
diff --git a/drivers/net/ethernet/chelsio/cxgb4/cxgb4_tc_flower.c b/drivers/net/ethernet/chelsio/cxgb4/cxgb4_tc_flower.c
index ed377e2e9f8a..a26acd183eef 100644
--- a/drivers/net/ethernet/chelsio/cxgb4/cxgb4_tc_flower.c
+++ b/drivers/net/ethernet/chelsio/cxgb4/cxgb4_tc_flower.c
@@ -755,7 +755,8 @@ static void ch_flower_stats_cb(struct timer_list *t)
 	hash_for_each_rcu(adap->flower_anymatch_tbl, i, flower_entry, link) {
 		ret = cxgb4_get_filter_counters(adap->port[0],
 						flower_entry->filter_id,
-						&packets, &bytes);
+						&packets, &bytes,
+						flower_entry->fs.hash);
 		if (!ret) {
 			spin_lock(&flower_entry->lock);
 			ofld_stats = &flower_entry->stats;
@@ -788,7 +789,8 @@ int cxgb4_tc_flower_stats(struct net_device *dev,
 	}
 
 	ret = cxgb4_get_filter_counters(dev, ch_flower->filter_id,
-					&packets, &bytes);
+					&packets, &bytes,
+					ch_flower->fs.hash);
 	if (ret < 0)
 		goto err;
 
diff --git a/drivers/net/ethernet/chelsio/cxgb4/cxgb4_uld.h b/drivers/net/ethernet/chelsio/cxgb4/cxgb4_uld.h
index a1c850861cbf..08e709ab6dd4 100644
--- a/drivers/net/ethernet/chelsio/cxgb4/cxgb4_uld.h
+++ b/drivers/net/ethernet/chelsio/cxgb4/cxgb4_uld.h
@@ -224,7 +224,7 @@ int cxgb4_set_filter(struct net_device *dev, int filter_id,
 int cxgb4_del_filter(struct net_device *dev, int filter_id,
 		     struct ch_filter_specification *fs);
 int cxgb4_get_filter_counters(struct net_device *dev, unsigned int fidx,
-			      u64 *hitcnt, u64 *bytecnt);
+			      u64 *hitcnt, u64 *bytecnt, bool hash);
 
 static inline void set_wr_txq(struct sk_buff *skb, int prio, int queue)
 {
-- 
2.14.1

[PATCH net-next 6/7] cxgb4: convert flower table to use rhashtable

[Rahul Lakkireddy]

From: Kumar Sanghvi <kumaras@chelsio.com>

T6 supports ~500K hash filters and can theoretically climb up to
~1 million hash filters. Preallocated hash table is not efficient
in terms of memory usage. So, use rhashtable instead which gives
the flexibility to grow based on usage.

Signed-off-by: Kumar Sanghvi <kumaras@chelsio.com>
Signed-off-by: Rahul Lakkireddy <rahul.lakkireddy@chelsio.com>
Signed-off-by: Ganesh Goudar <ganeshgr@chelsio.com>
---
 drivers/net/ethernet/chelsio/cxgb4/cxgb4.h         |   4 +-
 drivers/net/ethernet/chelsio/cxgb4/cxgb4_main.c    |   4 +-
 .../net/ethernet/chelsio/cxgb4/cxgb4_tc_flower.c   | 106 ++++++++++++++-------
 .../net/ethernet/chelsio/cxgb4/cxgb4_tc_flower.h   |   4 +-
 4 files changed, 82 insertions(+), 36 deletions(-)

diff --git a/drivers/net/ethernet/chelsio/cxgb4/cxgb4.h b/drivers/net/ethernet/chelsio/cxgb4/cxgb4.h
index bb7f0e4c9a81..0c83ceb5a1a6 100644
--- a/drivers/net/ethernet/chelsio/cxgb4/cxgb4.h
+++ b/drivers/net/ethernet/chelsio/cxgb4/cxgb4.h
@@ -918,8 +918,10 @@ struct adapter {
 	struct chcr_stats_debug chcr_stats;
 
 	/* TC flower offload */
-	DECLARE_HASHTABLE(flower_anymatch_tbl, 9);
+	struct rhashtable flower_tbl;
+	struct rhashtable_params flower_ht_params;
 	struct timer_list flower_stats_timer;
+	struct work_struct flower_stats_work;
 
 	/* Ethtool Dump */
 	struct ethtool_dump eth_dump;
diff --git a/drivers/net/ethernet/chelsio/cxgb4/cxgb4_main.c b/drivers/net/ethernet/chelsio/cxgb4/cxgb4_main.c
index 35709c7f7c5b..8fd41917c07a 100644
--- a/drivers/net/ethernet/chelsio/cxgb4/cxgb4_main.c
+++ b/drivers/net/ethernet/chelsio/cxgb4/cxgb4_main.c
@@ -5182,7 +5182,9 @@ static int init_one(struct pci_dev *pdev, const struct pci_device_id *ent)
 			dev_warn(&pdev->dev,
 				 "could not offload tc u32, continuing\n");
 
-		cxgb4_init_tc_flower(adapter);
+		if (cxgb4_init_tc_flower(adapter))
+			dev_warn(&pdev->dev,
+				 "could not offload tc flower, continuing\n");
 	}
 
 	if (is_offload(adapter) || is_hashfilter(adapter)) {
diff --git a/drivers/net/ethernet/chelsio/cxgb4/cxgb4_tc_flower.c b/drivers/net/ethernet/chelsio/cxgb4/cxgb4_tc_flower.c
index a26acd183eef..3953bc1fdc20 100644
--- a/drivers/net/ethernet/chelsio/cxgb4/cxgb4_tc_flower.c
+++ b/drivers/net/ethernet/chelsio/cxgb4/cxgb4_tc_flower.c
@@ -74,13 +74,8 @@ static struct ch_tc_flower_entry *allocate_flower_entry(void)
 static struct ch_tc_flower_entry *ch_flower_lookup(struct adapter *adap,
 						   unsigned long flower_cookie)
 {
-	struct ch_tc_flower_entry *flower_entry;
-
-	hash_for_each_possible_rcu(adap->flower_anymatch_tbl, flower_entry,
-				   link, flower_cookie)
-		if (flower_entry->tc_flower_cookie == flower_cookie)
-			return flower_entry;
-	return NULL;
+	return rhashtable_lookup_fast(&adap->flower_tbl, &flower_cookie,
+				      adap->flower_ht_params);
 }
 
 static void cxgb4_process_flow_match(struct net_device *dev,
@@ -707,12 +702,17 @@ int cxgb4_tc_flower_replace(struct net_device *dev,
 		goto free_entry;
 	}
 
-	INIT_HLIST_NODE(&ch_flower->link);
 	ch_flower->tc_flower_cookie = cls->cookie;
 	ch_flower->filter_id = ctx.tid;
-	hash_add_rcu(adap->flower_anymatch_tbl, &ch_flower->link, cls->cookie);
+	ret = rhashtable_insert_fast(&adap->flower_tbl, &ch_flower->node,
+				     adap->flower_ht_params);
+	if (ret)
+		goto del_filter;
 
-	return ret;
+	return 0;
+
+del_filter:
+	cxgb4_del_filter(dev, ch_flower->filter_id, &ch_flower->fs);
 
 free_entry:
 	kfree(ch_flower);
@@ -734,44 +734,66 @@ int cxgb4_tc_flower_destroy(struct net_device *dev,
 	if (ret)
 		goto err;
 
-	hash_del_rcu(&ch_flower->link);
+	ret = rhashtable_remove_fast(&adap->flower_tbl, &ch_flower->node,
+				     adap->flower_ht_params);
+	if (ret) {
+		netdev_err(dev, "Flow remove from rhashtable failed");
+		goto err;
+	}
 	kfree_rcu(ch_flower, rcu);
 
 err:
 	return ret;
 }
 
-static void ch_flower_stats_cb(struct timer_list *t)
+static void ch_flower_stats_handler(struct work_struct *work)
 {
-	struct adapter *adap = from_timer(adap, t, flower_stats_timer);
+	struct adapter *adap = container_of(work, struct adapter,
+					    flower_stats_work);
 	struct ch_tc_flower_entry *flower_entry;
 	struct ch_tc_flower_stats *ofld_stats;
-	unsigned int i;
+	struct rhashtable_iter iter;
 	u64 packets;
 	u64 bytes;
 	int ret;
 
-	rcu_read_lock();
-	hash_for_each_rcu(adap->flower_anymatch_tbl, i, flower_entry, link) {
-		ret = cxgb4_get_filter_counters(adap->port[0],
-						flower_entry->filter_id,
-						&packets, &bytes,
-						flower_entry->fs.hash);
-		if (!ret) {
-			spin_lock(&flower_entry->lock);
-			ofld_stats = &flower_entry->stats;
-
-			if (ofld_stats->prev_packet_count != packets) {
-				ofld_stats->prev_packet_count = packets;
-				ofld_stats->last_used = jiffies;
+	rhashtable_walk_enter(&adap->flower_tbl, &iter);
+	do {
+		flower_entry = ERR_PTR(rhashtable_walk_start(&iter));
+		if (IS_ERR(flower_entry))
+			goto walk_stop;
+
+		while ((flower_entry = rhashtable_walk_next(&iter)) &&
+		       !IS_ERR(flower_entry)) {
+			ret = cxgb4_get_filter_counters(adap->port[0],
+							flower_entry->filter_id,
+							&packets, &bytes,
+							flower_entry->fs.hash);
+			if (!ret) {
+				spin_lock(&flower_entry->lock);
+				ofld_stats = &flower_entry->stats;
+
+				if (ofld_stats->prev_packet_count != packets) {
+					ofld_stats->prev_packet_count = packets;
+					ofld_stats->last_used = jiffies;
+				}
+				spin_unlock(&flower_entry->lock);
 			}
-			spin_unlock(&flower_entry->lock);
 		}
-	}
-	rcu_read_unlock();
+walk_stop:
+		rhashtable_walk_stop(&iter);
+	} while (flower_entry == ERR_PTR(-EAGAIN));
+	rhashtable_walk_exit(&iter);
 	mod_timer(&adap->flower_stats_timer, jiffies + STATS_CHECK_PERIOD);
 }
 
+static void ch_flower_stats_cb(struct timer_list *t)
+{
+	struct adapter *adap = from_timer(adap, t, flower_stats_timer);
+
+	schedule_work(&adap->flower_stats_work);
+}
+
 int cxgb4_tc_flower_stats(struct net_device *dev,
 			  struct tc_cls_flower_offload *cls)
 {
@@ -814,15 +836,35 @@ int cxgb4_tc_flower_stats(struct net_device *dev,
 	return ret;
 }
 
-void cxgb4_init_tc_flower(struct adapter *adap)
+static const struct rhashtable_params cxgb4_tc_flower_ht_params = {
+	.nelem_hint = 384,
+	.head_offset = offsetof(struct ch_tc_flower_entry, node),
+	.key_offset = offsetof(struct ch_tc_flower_entry, tc_flower_cookie),
+	.key_len = sizeof(((struct ch_tc_flower_entry *)0)->tc_flower_cookie),
+	.max_size = 524288,
+	.min_size = 512,
+	.automatic_shrinking = true
+};
+
+int cxgb4_init_tc_flower(struct adapter *adap)
 {
-	hash_init(adap->flower_anymatch_tbl);
+	int ret;
+
+	adap->flower_ht_params = cxgb4_tc_flower_ht_params;
+	ret = rhashtable_init(&adap->flower_tbl, &adap->flower_ht_params);
+	if (ret)
+		return ret;
+
+	INIT_WORK(&adap->flower_stats_work, ch_flower_stats_handler);
 	timer_setup(&adap->flower_stats_timer, ch_flower_stats_cb, 0);
 	mod_timer(&adap->flower_stats_timer, jiffies + STATS_CHECK_PERIOD);
+	return 0;
 }
 
 void cxgb4_cleanup_tc_flower(struct adapter *adap)
 {
 	if (adap->flower_stats_timer.function)
 		del_timer_sync(&adap->flower_stats_timer);
+	cancel_work_sync(&adap->flower_stats_work);
+	rhashtable_destroy(&adap->flower_tbl);
 }
diff --git a/drivers/net/ethernet/chelsio/cxgb4/cxgb4_tc_flower.h b/drivers/net/ethernet/chelsio/cxgb4/cxgb4_tc_flower.h
index 202d5c9ec303..050c8a50ae41 100644
--- a/drivers/net/ethernet/chelsio/cxgb4/cxgb4_tc_flower.h
+++ b/drivers/net/ethernet/chelsio/cxgb4/cxgb4_tc_flower.h
@@ -48,7 +48,7 @@ struct ch_tc_flower_entry {
 	struct ch_filter_specification fs;
 	struct ch_tc_flower_stats stats;
 	unsigned long tc_flower_cookie;
-	struct hlist_node link;
+	struct rhash_head node;
 	struct rcu_head rcu;
 	spinlock_t lock; /* lock for stats */
 	u32 filter_id;
@@ -115,6 +115,6 @@ int cxgb4_tc_flower_destroy(struct net_device *dev,
 int cxgb4_tc_flower_stats(struct net_device *dev,
 			  struct tc_cls_flower_offload *cls);
 
-void cxgb4_init_tc_flower(struct adapter *adap);
+int cxgb4_init_tc_flower(struct adapter *adap);
 void cxgb4_cleanup_tc_flower(struct adapter *adap);
 #endif /* __CXGB4_TC_FLOWER_H */
-- 
2.14.1

[PATCH net-next 7/7] cxgb4: add support to create hash-filters via tc-flower offload

[Rahul Lakkireddy]

From: Kumar Sanghvi <kumaras@chelsio.com>

Determine whether the flow classifies as exact-match with respect to
4-tuple and configured tuple mask in hw. If successfully classified
as exact-match, offload the flow as hash-filter in hw.

Signed-off-by: Kumar Sanghvi <kumaras@chelsio.com>
Signed-off-by: Rahul Lakkireddy <rahul.lakkireddy@chelsio.com>
Signed-off-by: Ganesh Goudar <ganeshgr@chelsio.com>
---
 drivers/net/ethernet/chelsio/cxgb4/cxgb4_filter.c  | 148 +++++++++++++++++++++
 drivers/net/ethernet/chelsio/cxgb4/cxgb4_filter.h  |   2 +
 .../net/ethernet/chelsio/cxgb4/cxgb4_tc_flower.c   |  16 ++-
 3 files changed, 161 insertions(+), 5 deletions(-)

diff --git a/drivers/net/ethernet/chelsio/cxgb4/cxgb4_filter.c b/drivers/net/ethernet/chelsio/cxgb4/cxgb4_filter.c
index 9b3ff6209eb5..abab67d52edb 100644
--- a/drivers/net/ethernet/chelsio/cxgb4/cxgb4_filter.c
+++ b/drivers/net/ethernet/chelsio/cxgb4/cxgb4_filter.c
@@ -31,6 +31,7 @@
  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  * SOFTWARE.
  */
+#include <net/ipv6.h>
 
 #include "cxgb4.h"
 #include "t4_regs.h"
@@ -765,6 +766,153 @@ static void fill_default_mask(struct ch_filter_specification *fs)
 		fs->mask.fport = ~0;
 }
 
+static bool is_addr_all_mask(u8 *ipmask, int family)
+{
+	if (family == AF_INET) {
+		struct in_addr *addr;
+
+		addr = (struct in_addr *)ipmask;
+		if (addr->s_addr == 0xffffffff)
+			return true;
+	} else if (family == AF_INET6) {
+		struct in6_addr *addr6;
+
+		addr6 = (struct in6_addr *)ipmask;
+		if (addr6->s6_addr32[0] == 0xffffffff &&
+		    addr6->s6_addr32[1] == 0xffffffff &&
+		    addr6->s6_addr32[2] == 0xffffffff &&
+		    addr6->s6_addr32[3] == 0xffffffff)
+			return true;
+	}
+	return false;
+}
+
+static bool is_inaddr_any(u8 *ip, int family)
+{
+	int addr_type;
+
+	if (family == AF_INET) {
+		struct in_addr *addr;
+
+		addr = (struct in_addr *)ip;
+		if (addr->s_addr == htonl(INADDR_ANY))
+			return true;
+	} else if (family == AF_INET6) {
+		struct in6_addr *addr6;
+
+		addr6 = (struct in6_addr *)ip;
+		addr_type = ipv6_addr_type((const struct in6_addr *)
+					   &addr6);
+		if (addr_type == IPV6_ADDR_ANY)
+			return true;
+	}
+	return false;
+}
+
+bool is_filter_exact_match(struct adapter *adap,
+			   struct ch_filter_specification *fs)
+{
+	struct tp_params *tp = &adap->params.tp;
+	u64 hash_filter_mask = tp->hash_filter_mask;
+	u32 mask;
+
+	if (!is_hashfilter(adap))
+		return false;
+
+	if (fs->type) {
+		if (is_inaddr_any(fs->val.fip, AF_INET6) ||
+		    !is_addr_all_mask(fs->mask.fip, AF_INET6))
+			return false;
+
+		if (is_inaddr_any(fs->val.lip, AF_INET6) ||
+		    !is_addr_all_mask(fs->mask.lip, AF_INET6))
+			return false;
+	} else {
+		if (is_inaddr_any(fs->val.fip, AF_INET) ||
+		    !is_addr_all_mask(fs->mask.fip, AF_INET))
+			return false;
+
+		if (is_inaddr_any(fs->val.lip, AF_INET) ||
+		    !is_addr_all_mask(fs->mask.lip, AF_INET))
+			return false;
+	}
+
+	if (!fs->val.lport || fs->mask.lport != 0xffff)
+		return false;
+
+	if (!fs->val.fport || fs->mask.fport != 0xffff)
+		return false;
+
+	if (tp->fcoe_shift >= 0) {
+		mask = (hash_filter_mask >> tp->fcoe_shift) & FT_FCOE_W;
+		if (mask && !fs->mask.fcoe)
+			return false;
+	}
+
+	if (tp->port_shift >= 0) {
+		mask = (hash_filter_mask >> tp->port_shift) & FT_PORT_W;
+		if (mask && !fs->mask.iport)
+			return false;
+	}
+
+	if (tp->vnic_shift >= 0) {
+		mask = (hash_filter_mask >> tp->vnic_shift) & FT_VNIC_ID_W;
+
+		if ((adap->params.tp.ingress_config & VNIC_F)) {
+			if (mask && !fs->mask.pfvf_vld)
+				return false;
+		} else {
+			if (mask && !fs->mask.ovlan_vld)
+				return false;
+		}
+	}
+
+	if (tp->vlan_shift >= 0) {
+		mask = (hash_filter_mask >> tp->vlan_shift) & FT_VLAN_W;
+		if (mask && !fs->mask.ivlan)
+			return false;
+	}
+
+	if (tp->tos_shift >= 0) {
+		mask = (hash_filter_mask >> tp->tos_shift) & FT_TOS_W;
+		if (mask && !fs->mask.tos)
+			return false;
+	}
+
+	if (tp->protocol_shift >= 0) {
+		mask = (hash_filter_mask >> tp->protocol_shift) & FT_PROTOCOL_W;
+		if (mask && !fs->mask.proto)
+			return false;
+	}
+
+	if (tp->ethertype_shift >= 0) {
+		mask = (hash_filter_mask >> tp->ethertype_shift) &
+			FT_ETHERTYPE_W;
+		if (mask && !fs->mask.ethtype)
+			return false;
+	}
+
+	if (tp->macmatch_shift >= 0) {
+		mask = (hash_filter_mask >> tp->macmatch_shift) & FT_MACMATCH_W;
+		if (mask && !fs->mask.macidx)
+			return false;
+	}
+
+	if (tp->matchtype_shift >= 0) {
+		mask = (hash_filter_mask >> tp->matchtype_shift) &
+			FT_MPSHITTYPE_W;
+		if (mask && !fs->mask.matchtype)
+			return false;
+	}
+	if (tp->frag_shift >= 0) {
+		mask = (hash_filter_mask >> tp->frag_shift) &
+			FT_FRAGMENTATION_W;
+		if (mask && !fs->mask.frag)
+			return false;
+	}
+	return true;
+}
+
 static u64 hash_filter_ntuple(struct ch_filter_specification *fs,
 			      struct net_device *dev)
 {
diff --git a/drivers/net/ethernet/chelsio/cxgb4/cxgb4_filter.h b/drivers/net/ethernet/chelsio/cxgb4/cxgb4_filter.h
index 9475abd3384e..8db5fca6dcc9 100644
--- a/drivers/net/ethernet/chelsio/cxgb4/cxgb4_filter.h
+++ b/drivers/net/ethernet/chelsio/cxgb4/cxgb4_filter.h
@@ -51,4 +51,6 @@ int delete_filter(struct adapter *adapter, unsigned int fidx);
 int writable_filter(struct filter_entry *f);
 void clear_all_filters(struct adapter *adapter);
 int init_hash_filter(struct adapter *adap);
+bool is_filter_exact_match(struct adapter *adap,
+			   struct ch_filter_specification *fs);
 #endif /* __CXGB4_FILTER_H */
diff --git a/drivers/net/ethernet/chelsio/cxgb4/cxgb4_tc_flower.c b/drivers/net/ethernet/chelsio/cxgb4/cxgb4_tc_flower.c
index 3953bc1fdc20..d4a548a6a55c 100644
--- a/drivers/net/ethernet/chelsio/cxgb4/cxgb4_tc_flower.c
+++ b/drivers/net/ethernet/chelsio/cxgb4/cxgb4_tc_flower.c
@@ -38,6 +38,7 @@
 #include <net/tc_act/tc_vlan.h>
 
 #include "cxgb4.h"
+#include "cxgb4_filter.h"
 #include "cxgb4_tc_flower.h"
 
 #define STATS_CHECK_PERIOD (HZ / 2)
@@ -672,11 +673,16 @@ int cxgb4_tc_flower_replace(struct net_device *dev,
 	cxgb4_process_flow_match(dev, cls, fs);
 	cxgb4_process_flow_actions(dev, cls, fs);
 
-	fidx = cxgb4_get_free_ftid(dev, fs->type ? PF_INET6 : PF_INET);
-	if (fidx < 0) {
-		netdev_err(dev, "%s: No fidx for offload.\n", __func__);
-		ret = -ENOMEM;
-		goto free_entry;
+	fs->hash = is_filter_exact_match(adap, fs);
+	if (fs->hash) {
+		fidx = 0;
+	} else {
+		fidx = cxgb4_get_free_ftid(dev, fs->type ? PF_INET6 : PF_INET);
+		if (fidx < 0) {
+			netdev_err(dev, "%s: No fidx for offload.\n", __func__);
+			ret = -ENOMEM;
+			goto free_entry;
+		}
 	}
 
 	init_completion(&ctx.completion);
-- 
2.14.1

Re: [PATCH net-next 0/7] cxgb4: add hash-filter support to tc-flower offload

[David Miller]

From: Rahul Lakkireddy <rahul.lakkireddy@chelsio.com>
Date: Wed,  1 Nov 2017 08:52:58 +0530

> This series of patches add support to create hash-filters; a.k.a
> exact-match filters, to tc-flower offload.  T6 supports creating
> ~500K hash-filters in hw and can theoretically be expanded up to
> ~1 million.
> 
> Patch 1 fetches and saves the configured hw filter tuple field shifts
> and filter mask.
> 
> Patch 2 initializes the driver to use hash-filter configuration.
> 
> Patch 3 adds support to create hash filters in hw.
> 
> Patch 4 adds support to delete hash filters in hw.
> 
> Patch 5 adds support to retrieve filter stats for hash filters.
> 
> Patch 6 converts the flower table to use rhashtable instead of
> static hlist.
> 
> Patch 7 finally adds support to create hash filters via tc-flower
> offload.

Series applied, thanks.