[dpdk-dev] [PATCH 5/9] net/cxgbe: add support to delete flows in HASH region

[Rahul Lakkireddy <rahul.lakkireddy@chelsio.com>]

From: Shagun Agrawal <shaguna@chelsio.com>

Add interface to delete offloaded flows in HASH region. Use the
hash index saved during insertion to delete the corresponding flow.

Signed-off-by: Shagun Agrawal <shaguna@chelsio.com>
Signed-off-by: Rahul Lakkireddy <rahul.lakkireddy@chelsio.com>
---
 drivers/net/cxgbe/base/t4_msg.h         |  72 ++++++++++++
 drivers/net/cxgbe/base/t4_tcb.h         |  15 +++
 drivers/net/cxgbe/base/t4fw_interface.h |   6 +
 drivers/net/cxgbe/cxgbe_filter.c        | 193 ++++++++++++++++++++++++++++++++
 drivers/net/cxgbe/cxgbe_filter.h        |   2 +
 drivers/net/cxgbe/cxgbe_main.c          |  48 ++++++++
 drivers/net/cxgbe/cxgbe_ofld.h          |  15 +++
 7 files changed, 351 insertions(+)
 create mode 100644 drivers/net/cxgbe/base/t4_tcb.h

diff --git a/drivers/net/cxgbe/base/t4_msg.h b/drivers/net/cxgbe/base/t4_msg.h
index 4112ff212..7f4c98fb6 100644
--- a/drivers/net/cxgbe/base/t4_msg.h
+++ b/drivers/net/cxgbe/base/t4_msg.h
@@ -8,7 +8,12 @@
 
 enum {
 	CPL_ACT_OPEN_REQ      = 0x3,
+	CPL_SET_TCB_FIELD     = 0x5,
+	CPL_ABORT_REQ         = 0xA,
+	CPL_ABORT_RPL         = 0xB,
+	CPL_TID_RELEASE       = 0x1A,
 	CPL_ACT_OPEN_RPL      = 0x25,
+	CPL_ABORT_RPL_RSS     = 0x2D,
 	CPL_SET_TCB_RPL       = 0x3A,
 	CPL_ACT_OPEN_REQ6     = 0x83,
 	CPL_SGE_EGR_UPDATE    = 0xA5,
@@ -27,6 +32,11 @@ enum {
 	ULP_MODE_NONE          = 0,
 };
 
+enum {
+	CPL_ABORT_SEND_RST = 0,
+	CPL_ABORT_NO_RST,
+};
+
 enum {                     /* TX_PKT_XT checksum types */
 	TX_CSUM_TCPIP  = 8,
 	TX_CSUM_UDPIP  = 9,
@@ -189,6 +199,29 @@ struct cpl_act_open_rpl {
 #define M_AOPEN_ATID    0xFFFFFF
 #define G_AOPEN_ATID(x) (((x) >> S_AOPEN_ATID) & M_AOPEN_ATID)
 
+struct cpl_set_tcb_field {
+	WR_HDR;
+	union opcode_tid ot;
+	__be16 reply_ctrl;
+	__be16 word_cookie;
+	__be64 mask;
+	__be64 val;
+};
+
+/* cpl_set_tcb_field.word_cookie fields */
+#define S_WORD    0
+#define V_WORD(x) ((x) << S_WORD)
+
+/* cpl_get_tcb.reply_ctrl fields */
+#define S_QUEUENO    0
+#define V_QUEUENO(x) ((x) << S_QUEUENO)
+
+#define S_REPLY_CHAN    14
+#define V_REPLY_CHAN(x) ((x) << S_REPLY_CHAN)
+
+#define S_NO_REPLY    15
+#define V_NO_REPLY(x) ((x) << S_NO_REPLY)
+
 struct cpl_set_tcb_rpl {
 	RSS_HDR
 	union opcode_tid ot;
@@ -198,6 +231,39 @@ struct cpl_set_tcb_rpl {
 	__be64 oldval;
 };
 
+/* cpl_abort_req status command code
+ */
+struct cpl_abort_req {
+	WR_HDR;
+	union opcode_tid ot;
+	__be32 rsvd0;
+	__u8  rsvd1;
+	__u8  cmd;
+	__u8  rsvd2[6];
+};
+
+struct cpl_abort_rpl_rss {
+	RSS_HDR
+	union opcode_tid ot;
+	__u8  rsvd[3];
+	__u8  status;
+};
+
+struct cpl_abort_rpl {
+	WR_HDR;
+	union opcode_tid ot;
+	__be32 rsvd0;
+	__u8  rsvd1;
+	__u8  cmd;
+	__u8  rsvd2[6];
+};
+
+struct cpl_tid_release {
+	WR_HDR;
+	union opcode_tid ot;
+	__be32 rsvd;
+};
+
 struct cpl_tx_data {
 	union opcode_tid ot;
 	__be32 len;
@@ -403,7 +469,13 @@ struct cpl_fw6_msg {
 	__be64 data[4];
 };
 
+/* ULP_TX opcodes */
+enum {
+	ULP_TX_PKT = 4
+};
+
 enum {
+	ULP_TX_SC_NOOP = 0x80,
 	ULP_TX_SC_IMM  = 0x81,
 	ULP_TX_SC_DSGL = 0x82,
 	ULP_TX_SC_ISGL = 0x83
diff --git a/drivers/net/cxgbe/base/t4_tcb.h b/drivers/net/cxgbe/base/t4_tcb.h
new file mode 100644
index 000000000..6d7f5e8c1
--- /dev/null
+++ b/drivers/net/cxgbe/base/t4_tcb.h
@@ -0,0 +1,15 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(c) 2018 Chelsio Communications.
+ * All rights reserved.
+ */
+
+#ifndef _T4_TCB_DEFS_H
+#define _T4_TCB_DEFS_H
+
+/* 105:96 */
+#define W_TCB_RSS_INFO    3
+#define S_TCB_RSS_INFO    0
+#define M_TCB_RSS_INFO    0x3ffULL
+#define V_TCB_RSS_INFO(x) ((x) << S_TCB_RSS_INFO)
+
+#endif /* _T4_TCB_DEFS_H */
diff --git a/drivers/net/cxgbe/base/t4fw_interface.h b/drivers/net/cxgbe/base/t4fw_interface.h
index 19bcfc124..7b2f2d37f 100644
--- a/drivers/net/cxgbe/base/t4fw_interface.h
+++ b/drivers/net/cxgbe/base/t4fw_interface.h
@@ -55,6 +55,7 @@ enum fw_memtype {
 
 enum fw_wr_opcodes {
 	FW_FILTER_WR		= 0x02,
+	FW_ULPTX_WR		= 0x04,
 	FW_TP_WR		= 0x05,
 	FW_ETH_TX_PKT_WR	= 0x08,
 	FW_ETH_TX_PKTS_WR	= 0x09,
@@ -78,6 +79,11 @@ struct fw_wr_hdr {
 #define V_FW_WR_OP(x)		((x) << S_FW_WR_OP)
 #define G_FW_WR_OP(x)		(((x) >> S_FW_WR_OP) & M_FW_WR_OP)
 
+/* atomic flag (hi) - firmware encapsulates CPLs in CPL_BARRIER
+ */
+#define S_FW_WR_ATOMIC		23
+#define V_FW_WR_ATOMIC(x)	((x) << S_FW_WR_ATOMIC)
+
 /* work request immediate data length (hi)
  */
 #define S_FW_WR_IMMDLEN	0
diff --git a/drivers/net/cxgbe/cxgbe_filter.c b/drivers/net/cxgbe/cxgbe_filter.c
index bac7aa291..7759b8acf 100644
--- a/drivers/net/cxgbe/cxgbe_filter.c
+++ b/drivers/net/cxgbe/cxgbe_filter.c
@@ -4,6 +4,7 @@
  */
 #include <rte_net.h>
 #include "common.h"
+#include "t4_tcb.h"
 #include "t4_regs.h"
 #include "cxgbe_filter.h"
 #include "clip_tbl.h"
@@ -116,6 +117,34 @@ int writable_filter(struct filter_entry *f)
 	return 0;
 }
 
+/**
+ * Build a CPL_SET_TCB_FIELD message as payload of a ULP_TX_PKT command.
+ */
+static inline void mk_set_tcb_field_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 = cpu_to_be32(V_ULPTX_CMD(ULP_TX_PKT) |
+				      V_ULP_TXPKT_DEST(0));
+	txpkt->len = cpu_to_be32(DIV_ROUND_UP(sizeof(*req), 16));
+	sc->cmd_more = cpu_to_be32(V_ULPTX_CMD(ULP_TX_SC_IMM));
+	sc->len = cpu_to_be32(sizeof(*req) - sizeof(struct work_request_hdr));
+	OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_SET_TCB_FIELD, f->tid));
+	req->reply_ctrl = cpu_to_be16(V_NO_REPLY(no_reply) | V_REPLY_CHAN(0) |
+				      V_QUEUENO(0));
+	req->word_cookie = cpu_to_be16(V_WORD(word) | V_COOKIE(cookie));
+	req->mask = cpu_to_be64(mask);
+	req->val = cpu_to_be64(val);
+	sc = (struct ulptx_idata *)(req + 1);
+	sc->cmd_more = cpu_to_be32(V_ULPTX_CMD(ULP_TX_SC_NOOP));
+	sc->len = cpu_to_be32(0);
+}
+
 /**
  * Check if entry already filled.
  */
@@ -185,6 +214,132 @@ static u64 hash_filter_ntuple(const struct filter_entry *f)
 	return ntuple;
 }
 
+/**
+ * Build a CPL_ABORT_REQ message as payload of a ULP_TX_PKT command.
+ */
+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 = cpu_to_be32(V_ULPTX_CMD(ULP_TX_PKT) |
+				      V_ULP_TXPKT_DEST(0));
+	txpkt->len = cpu_to_be32(DIV_ROUND_UP(sizeof(*abort_req), 16));
+	sc->cmd_more = cpu_to_be32(V_ULPTX_CMD(ULP_TX_SC_IMM));
+	sc->len = cpu_to_be32(sizeof(*abort_req) -
+			      sizeof(struct work_request_hdr));
+	OPCODE_TID(abort_req) = cpu_to_be32(MK_OPCODE_TID(CPL_ABORT_REQ, tid));
+	abort_req->rsvd0 = cpu_to_be32(0);
+	abort_req->rsvd1 = 0;
+	abort_req->cmd = CPL_ABORT_NO_RST;
+	sc = (struct ulptx_idata *)(abort_req + 1);
+	sc->cmd_more = cpu_to_be32(V_ULPTX_CMD(ULP_TX_SC_NOOP));
+	sc->len = cpu_to_be32(0);
+}
+
+/**
+ * Build a CPL_ABORT_RPL message as payload of a ULP_TX_PKT command.
+ */
+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 = cpu_to_be32(V_ULPTX_CMD(ULP_TX_PKT) |
+				      V_ULP_TXPKT_DEST(0));
+	txpkt->len = cpu_to_be32(DIV_ROUND_UP(sizeof(*abort_rpl), 16));
+	sc->cmd_more = cpu_to_be32(V_ULPTX_CMD(ULP_TX_SC_IMM));
+	sc->len = cpu_to_be32(sizeof(*abort_rpl) -
+			      sizeof(struct work_request_hdr));
+	OPCODE_TID(abort_rpl) = cpu_to_be32(MK_OPCODE_TID(CPL_ABORT_RPL, tid));
+	abort_rpl->rsvd0 = cpu_to_be32(0);
+	abort_rpl->rsvd1 = 0;
+	abort_rpl->cmd = CPL_ABORT_NO_RST;
+	sc = (struct ulptx_idata *)(abort_rpl + 1);
+	sc->cmd_more = cpu_to_be32(V_ULPTX_CMD(ULP_TX_SC_NOOP));
+	sc->len = cpu_to_be32(0);
+}
+
+/**
+ * Delete the specified hash filter.
+ */
+static int cxgbe_del_hash_filter(struct rte_eth_dev *dev,
+				 unsigned int filter_id,
+				 struct filter_ctx *ctx)
+{
+	struct adapter *adapter = ethdev2adap(dev);
+	struct tid_info *t = &adapter->tids;
+	struct filter_entry *f;
+	struct sge_ctrl_txq *ctrlq;
+	unsigned int port_id = ethdev2pinfo(dev)->port_id;
+	int ret;
+
+	if (filter_id > adapter->tids.ntids)
+		return -E2BIG;
+
+	f = lookup_tid(t, filter_id);
+	if (!f) {
+		dev_err(adapter, "%s: no filter entry for filter_id = %d\n",
+			__func__, filter_id);
+		return -EINVAL;
+	}
+
+	ret = writable_filter(f);
+	if (ret)
+		return ret;
+
+	if (f->valid) {
+		unsigned int wrlen;
+		struct rte_mbuf *mbuf;
+		struct work_request_hdr *wr;
+		struct ulptx_idata *aligner;
+		struct cpl_set_tcb_field *req;
+		struct cpl_abort_req *abort_req;
+		struct cpl_abort_rpl *abort_rpl;
+
+		f->ctx = ctx;
+		f->pending = 1;
+
+		wrlen = cxgbe_roundup(sizeof(*wr) +
+				      (sizeof(*req) + sizeof(*aligner)) +
+				      sizeof(*abort_req) + sizeof(*abort_rpl),
+				      16);
+
+		ctrlq = &adapter->sge.ctrlq[port_id];
+		mbuf = rte_pktmbuf_alloc(ctrlq->mb_pool);
+		if (!mbuf) {
+			dev_err(adapter, "%s: could not allocate skb ..\n",
+				__func__);
+			goto out_err;
+		}
+
+		mbuf->data_len = wrlen;
+		mbuf->pkt_len = mbuf->data_len;
+
+		req = rte_pktmbuf_mtod(mbuf, struct cpl_set_tcb_field *);
+		INIT_ULPTX_WR(req, wrlen, 0, 0);
+		wr = (struct work_request_hdr *)req;
+		wr++;
+		req = (struct cpl_set_tcb_field *)wr;
+		mk_set_tcb_field_ulp(f, req, W_TCB_RSS_INFO,
+				V_TCB_RSS_INFO(M_TCB_RSS_INFO),
+				V_TCB_RSS_INFO(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_mgmt_tx(ctrlq, mbuf);
+	}
+	return 0;
+
+out_err:
+	return -ENOMEM;
+}
+
 /**
  * Build a ACT_OPEN_REQ6 message for setting IPv6 hash filter.
  */
@@ -560,6 +715,9 @@ int cxgbe_del_filter(struct rte_eth_dev *dev, unsigned int filter_id,
 	unsigned int chip_ver;
 	int ret;
 
+	if (is_hashfilter(adapter) && fs->cap)
+		return cxgbe_del_hash_filter(dev, filter_id, ctx);
+
 	if (filter_id >= adapter->tids.nftids)
 		return -ERANGE;
 
@@ -967,3 +1125,38 @@ int cxgbe_get_filter_count(struct adapter *adapter, unsigned int fidx,
 	}
 	return 0;
 }
+
+/**
+ * Handle a Hash filter delete reply.
+ */
+void hash_del_filter_rpl(struct adapter *adap,
+			 const struct cpl_abort_rpl_rss *rpl)
+{
+	struct tid_info *t = &adap->tids;
+	struct filter_entry *f;
+	struct filter_ctx *ctx = NULL;
+	unsigned int tid = GET_TID(rpl);
+
+	f = lookup_tid(t, tid);
+	if (!f) {
+		dev_warn(adap, "%s: could not find filter entry: %u\n",
+			 __func__, tid);
+		return;
+	}
+
+	ctx = f->ctx;
+	f->ctx = NULL;
+
+	f->valid = 0;
+
+	if (f->clipt)
+		cxgbe_clip_release(f->dev, f->clipt);
+
+	cxgbe_remove_tid(t, 0, tid, 0);
+	t4_os_free(f);
+
+	if (ctx) {
+		ctx->result = 0;
+		t4_complete(&ctx->completion);
+	}
+}
diff --git a/drivers/net/cxgbe/cxgbe_filter.h b/drivers/net/cxgbe/cxgbe_filter.h
index 7c469c895..c51efea7d 100644
--- a/drivers/net/cxgbe/cxgbe_filter.h
+++ b/drivers/net/cxgbe/cxgbe_filter.h
@@ -224,6 +224,8 @@ int cxgbe_del_filter(struct rte_eth_dev *dev, unsigned int filter_id,
 int cxgbe_alloc_ftid(struct adapter *adap, unsigned int family);
 int init_hash_filter(struct adapter *adap);
 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);
 int validate_filter(struct adapter *adap, struct ch_filter_specification *fs);
 int cxgbe_get_filter_count(struct adapter *adapter, unsigned int fidx,
 			   u64 *c, bool get_byte);
diff --git a/drivers/net/cxgbe/cxgbe_main.c b/drivers/net/cxgbe/cxgbe_main.c
index c550dd5be..08b2a42d1 100644
--- a/drivers/net/cxgbe/cxgbe_main.c
+++ b/drivers/net/cxgbe/cxgbe_main.c
@@ -87,6 +87,10 @@ static int fwevtq_handler(struct sge_rspq *q, const __be64 *rsp,
 		const struct cpl_fw6_msg *msg = (const void *)rsp;
 
 		t4_handle_fw_rpl(q->adapter, msg->data);
+	} else if (opcode == CPL_ABORT_RPL_RSS) {
+		const struct cpl_abort_rpl_rss *p = (const void *)rsp;
+
+		hash_del_filter_rpl(q->adapter, p);
 	} else if (opcode == CPL_SET_TCB_RPL) {
 		const struct cpl_set_tcb_rpl *p = (const void *)rsp;
 
@@ -301,6 +305,50 @@ void cxgbe_free_atid(struct tid_info *t, unsigned int atid)
 	t4_os_unlock(&t->atid_lock);
 }
 
+/**
+ * Populate a TID_RELEASE WR.  Caller must properly size the skb.
+ */
+static void mk_tid_release(struct rte_mbuf *mbuf, unsigned int tid)
+{
+	struct cpl_tid_release *req;
+
+	req = rte_pktmbuf_mtod(mbuf, struct cpl_tid_release *);
+	INIT_TP_WR_MIT_CPL(req, CPL_TID_RELEASE, tid);
+}
+
+/**
+ * Release a TID and inform HW.  If we are unable to allocate the release
+ * message we defer to a work queue.
+ */
+void cxgbe_remove_tid(struct tid_info *t, unsigned int chan, unsigned int tid,
+		      unsigned short family)
+{
+	struct rte_mbuf *mbuf;
+	struct adapter *adap = container_of(t, struct adapter, tids);
+
+	WARN_ON(tid >= t->ntids);
+
+	if (t->tid_tab[tid]) {
+		t->tid_tab[tid] = NULL;
+		rte_atomic32_dec(&t->conns_in_use);
+		if (t->hash_base && tid >= t->hash_base) {
+			if (family == FILTER_TYPE_IPV4)
+				rte_atomic32_dec(&t->hash_tids_in_use);
+		} else {
+			if (family == FILTER_TYPE_IPV4)
+				rte_atomic32_dec(&t->tids_in_use);
+		}
+	}
+
+	mbuf = rte_pktmbuf_alloc((&adap->sge.ctrlq[chan])->mb_pool);
+	if (mbuf) {
+		mbuf->data_len = sizeof(struct cpl_tid_release);
+		mbuf->pkt_len = mbuf->data_len;
+		mk_tid_release(mbuf, tid);
+		t4_mgmt_tx(&adap->sge.ctrlq[chan], mbuf);
+	}
+}
+
 /**
  * Insert a TID.
  */
diff --git a/drivers/net/cxgbe/cxgbe_ofld.h b/drivers/net/cxgbe/cxgbe_ofld.h
index 798e39828..50931ed04 100644
--- a/drivers/net/cxgbe/cxgbe_ofld.h
+++ b/drivers/net/cxgbe/cxgbe_ofld.h
@@ -19,6 +19,19 @@
 	(w)->wr.wr_lo = cpu_to_be64(0); \
 } while (0)
 
+#define INIT_TP_WR_MIT_CPL(w, cpl, tid) do { \
+	INIT_TP_WR(w, tid); \
+	OPCODE_TID(w) = cpu_to_be32(MK_OPCODE_TID(cpl, tid)); \
+} while (0)
+
+#define INIT_ULPTX_WR(w, wrlen, atomic, tid) do { \
+	(w)->wr.wr_hi = cpu_to_be32(V_FW_WR_OP(FW_ULPTX_WR) | \
+				    V_FW_WR_ATOMIC(atomic)); \
+	(w)->wr.wr_mid = cpu_to_be32(V_FW_WR_LEN16(DIV_ROUND_UP(wrlen, 16)) | \
+				     V_FW_WR_FLOWID(tid)); \
+	(w)->wr.wr_lo = cpu_to_be64(0); \
+} while (0)
+
 /*
  * Max # of ATIDs.  The absolute HW max is 16K but we keep it lower.
  */
@@ -68,6 +81,8 @@ static inline void *lookup_atid(const struct tid_info *t, unsigned int atid)
 
 int cxgbe_alloc_atid(struct tid_info *t, void *data);
 void cxgbe_free_atid(struct tid_info *t, unsigned int atid);
+void cxgbe_remove_tid(struct tid_info *t, unsigned int qid, unsigned int tid,
+		      unsigned short family);
 void cxgbe_insert_tid(struct tid_info *t, void *data, unsigned int tid,
 		      unsigned short family);
 
-- 
2.14.1