1 files changed, 184 insertions, 58 deletions

diff --git a/drivers/scsi/cxgbi/cxgb4i/cxgb4i.c b/drivers/scsi/cxgbi/cxgb4i/cxgb4i.c
index 8ce8592f6a64..4e82c14cb795 100644
--- a/drivers/scsi/cxgbi/cxgb4i/cxgb4i.c
+++ b/drivers/scsi/cxgbi/cxgb4i/cxgb4i.c
@@ -197,7 +197,10 @@ static inline bool is_ofld_imm(const struct sk_buff *skb)
 	if (likely(cxgbi_skcb_test_flag(skb, SKCBF_TX_NEED_HDR)))
 		len += sizeof(struct fw_ofld_tx_data_wr);
 
-	return len <= MAX_IMM_TX_PKT_LEN;
+	if  (likely(cxgbi_skcb_test_flag((struct sk_buff *)skb, SKCBF_TX_ISO)))
+		len += sizeof(struct cpl_tx_data_iso);
+
+	return (len <= MAX_IMM_OFLD_TX_DATA_WR_LEN);
 }
 
 static void send_act_open_req(struct cxgbi_sock *csk, struct sk_buff *skb,
@@ -641,7 +644,10 @@ static inline int send_tx_flowc_wr(struct cxgbi_sock *csk)
 	flowc->mnemval[8].mnemonic = 0;
 	flowc->mnemval[8].val = 0;
 	flowc->mnemval[8].mnemonic = FW_FLOWC_MNEM_TXDATAPLEN_MAX;
-	flowc->mnemval[8].val = 16384;
+	if (csk->cdev->skb_iso_txhdr)
+		flowc->mnemval[8].val = cpu_to_be32(CXGBI_MAX_ISO_DATA_IN_SKB);
+	else
+		flowc->mnemval[8].val = cpu_to_be32(16128);
 #ifdef CONFIG_CHELSIO_T4_DCB
 	flowc->mnemval[9].mnemonic = FW_FLOWC_MNEM_DCBPRIO;
 	if (vlan == CPL_L2T_VLAN_NONE) {
@@ -667,38 +673,86 @@ static inline int send_tx_flowc_wr(struct cxgbi_sock *csk)
 	return flowclen16;
 }
 
-static inline void make_tx_data_wr(struct cxgbi_sock *csk, struct sk_buff *skb,
-				   int dlen, int len, u32 credits, int compl)
+static void
+cxgb4i_make_tx_iso_cpl(struct sk_buff *skb, struct cpl_tx_data_iso *cpl)
+{
+	struct cxgbi_iso_info *info = (struct cxgbi_iso_info *)skb->head;
+	u32 imm_en = !!(info->flags & CXGBI_ISO_INFO_IMM_ENABLE);
+	u32 fslice = !!(info->flags & CXGBI_ISO_INFO_FSLICE);
+	u32 lslice = !!(info->flags & CXGBI_ISO_INFO_LSLICE);
+	u32 pdu_type = (info->op == ISCSI_OP_SCSI_CMD) ? 0 : 1;
+	u32 submode = cxgbi_skcb_tx_ulp_mode(skb) & 0x3;
+
+	cpl->op_to_scsi = cpu_to_be32(CPL_TX_DATA_ISO_OP_V(CPL_TX_DATA_ISO) |
+				CPL_TX_DATA_ISO_FIRST_V(fslice) |
+				CPL_TX_DATA_ISO_LAST_V(lslice) |
+				CPL_TX_DATA_ISO_CPLHDRLEN_V(0) |
+				CPL_TX_DATA_ISO_HDRCRC_V(submode & 1) |
+				CPL_TX_DATA_ISO_PLDCRC_V(((submode >> 1) & 1)) |
+				CPL_TX_DATA_ISO_IMMEDIATE_V(imm_en) |
+				CPL_TX_DATA_ISO_SCSI_V(pdu_type));
+
+	cpl->ahs_len = info->ahs;
+	cpl->mpdu = cpu_to_be16(DIV_ROUND_UP(info->mpdu, 4));
+	cpl->burst_size = cpu_to_be32(info->burst_size);
+	cpl->len = cpu_to_be32(info->len);
+	cpl->reserved2_seglen_offset =
+	     cpu_to_be32(CPL_TX_DATA_ISO_SEGLEN_OFFSET_V(info->segment_offset));
+	cpl->datasn_offset = cpu_to_be32(info->datasn_offset);
+	cpl->buffer_offset = cpu_to_be32(info->buffer_offset);
+	cpl->reserved3 = cpu_to_be32(0);
+	log_debug(1 << CXGBI_DBG_ISCSI | 1 << CXGBI_DBG_PDU_TX,
+		  "iso: flags 0x%x, op %u, ahs %u, num_pdu %u, mpdu %u, "
+		  "burst_size %u, iso_len %u\n",
+		  info->flags, info->op, info->ahs, info->num_pdu,
+		  info->mpdu, info->burst_size << 2, info->len);
+}
+
+static void
+cxgb4i_make_tx_data_wr(struct cxgbi_sock *csk, struct sk_buff *skb, int dlen,
+		       int len, u32 credits, int compl)
 {
+	struct cxgbi_device *cdev = csk->cdev;
+	struct cxgb4_lld_info *lldi = cxgbi_cdev_priv(cdev);
 	struct fw_ofld_tx_data_wr *req;
-	unsigned int submode = cxgbi_skcb_ulp_mode(skb) & 3;
-	unsigned int wr_ulp_mode = 0, val;
-	bool imm = is_ofld_imm(skb);
-
-	req = __skb_push(skb, sizeof(*req));
-
-	if (imm) {
-		req->op_to_immdlen = htonl(FW_WR_OP_V(FW_OFLD_TX_DATA_WR) |
-					FW_WR_COMPL_F |
-					FW_WR_IMMDLEN_V(dlen));
-		req->flowid_len16 = htonl(FW_WR_FLOWID_V(csk->tid) |
-						FW_WR_LEN16_V(credits));
-	} else {
-		req->op_to_immdlen =
-			cpu_to_be32(FW_WR_OP_V(FW_OFLD_TX_DATA_WR) |
-					FW_WR_COMPL_F |
-					FW_WR_IMMDLEN_V(0));
-		req->flowid_len16 =
-			cpu_to_be32(FW_WR_FLOWID_V(csk->tid) |
-					FW_WR_LEN16_V(credits));
+	struct cpl_tx_data_iso *cpl;
+	u32 submode = cxgbi_skcb_tx_ulp_mode(skb) & 0x3;
+	u32 wr_ulp_mode = 0;
+	u32 hdr_size = sizeof(*req);
+	u32 opcode = FW_OFLD_TX_DATA_WR;
+	u32 immlen = 0;
+	u32 force = is_t5(lldi->adapter_type) ? TX_FORCE_V(!submode) :
+						T6_TX_FORCE_F;
+
+	if (cxgbi_skcb_test_flag(skb, SKCBF_TX_ISO)) {
+		hdr_size += sizeof(struct cpl_tx_data_iso);
+		opcode = FW_ISCSI_TX_DATA_WR;
+		immlen += sizeof(struct cpl_tx_data_iso);
+		submode |= 8;
 	}
+
+	if (is_ofld_imm(skb))
+		immlen += dlen;
+
+	req = (struct fw_ofld_tx_data_wr *)__skb_push(skb, hdr_size);
+	req->op_to_immdlen = cpu_to_be32(FW_WR_OP_V(opcode) |
+					 FW_WR_COMPL_V(compl) |
+					 FW_WR_IMMDLEN_V(immlen));
+	req->flowid_len16 = cpu_to_be32(FW_WR_FLOWID_V(csk->tid) |
+					FW_WR_LEN16_V(credits));
+	req->plen = cpu_to_be32(len);
+	cpl =  (struct cpl_tx_data_iso *)(req + 1);
+
+	if (likely(cxgbi_skcb_test_flag(skb, SKCBF_TX_ISO)))
+		cxgb4i_make_tx_iso_cpl(skb, cpl);
+
 	if (submode)
 		wr_ulp_mode = FW_OFLD_TX_DATA_WR_ULPMODE_V(ULP2_MODE_ISCSI) |
-				FW_OFLD_TX_DATA_WR_ULPSUBMODE_V(submode);
-	val = skb_peek(&csk->write_queue) ? 0 : 1;
-	req->tunnel_to_proxy = htonl(wr_ulp_mode |
-				     FW_OFLD_TX_DATA_WR_SHOVE_V(val));
-	req->plen = htonl(len);
+			      FW_OFLD_TX_DATA_WR_ULPSUBMODE_V(submode);
+
+	req->tunnel_to_proxy = cpu_to_be32(wr_ulp_mode | force |
+					   FW_OFLD_TX_DATA_WR_SHOVE_V(1U));
+
 	if (!cxgbi_sock_flag(csk, CTPF_TX_DATA_SENT))
 		cxgbi_sock_set_flag(csk, CTPF_TX_DATA_SENT);
 }
@@ -716,30 +770,34 @@ static int push_tx_frames(struct cxgbi_sock *csk, int req_completion)
 	if (unlikely(csk->state < CTP_ESTABLISHED ||
 		csk->state == CTP_CLOSE_WAIT_1 || csk->state >= CTP_ABORTING)) {
 		log_debug(1 << CXGBI_DBG_TOE | 1 << CXGBI_DBG_SOCK |
-			 1 << CXGBI_DBG_PDU_TX,
-			"csk 0x%p,%u,0x%lx,%u, in closing state.\n",
-			csk, csk->state, csk->flags, csk->tid);
+			  1 << CXGBI_DBG_PDU_TX,
+			  "csk 0x%p,%u,0x%lx,%u, in closing state.\n",
+			  csk, csk->state, csk->flags, csk->tid);
 		return 0;
 	}
 
-	while (csk->wr_cred && (skb = skb_peek(&csk->write_queue)) != NULL) {
-		int dlen = skb->len;
-		int len = skb->len;
-		unsigned int credits_needed;
-		int flowclen16 = 0;
+	while (csk->wr_cred && ((skb = skb_peek(&csk->write_queue)) != NULL)) {
+		struct cxgbi_iso_info *iso_cpl;
+		u32 dlen = skb->len;
+		u32 len = skb->len;
+		u32 iso_cpl_len = 0;
+		u32 flowclen16 = 0;
+		u32 credits_needed;
+		u32 num_pdu = 1, hdr_len;
+
+		if (cxgbi_skcb_test_flag(skb, SKCBF_TX_ISO))
+			iso_cpl_len = sizeof(struct cpl_tx_data_iso);
 
-		skb_reset_transport_header(skb);
 		if (is_ofld_imm(skb))
-			credits_needed = DIV_ROUND_UP(dlen, 16);
+			credits_needed = DIV_ROUND_UP(dlen + iso_cpl_len, 16);
 		else
-			credits_needed = DIV_ROUND_UP(
-						8 * calc_tx_flits_ofld(skb),
-						16);
+			credits_needed =
+				DIV_ROUND_UP((8 * calc_tx_flits_ofld(skb)) +
+					     iso_cpl_len, 16);
 
 		if (likely(cxgbi_skcb_test_flag(skb, SKCBF_TX_NEED_HDR)))
-			credits_needed += DIV_ROUND_UP(
-					sizeof(struct fw_ofld_tx_data_wr),
-					16);
+			credits_needed +=
+			   DIV_ROUND_UP(sizeof(struct fw_ofld_tx_data_wr), 16);
 
 		/*
 		 * Assumes the initial credits is large enough to support
@@ -754,14 +812,19 @@ static int push_tx_frames(struct cxgbi_sock *csk, int req_completion)
 
 		if (csk->wr_cred < credits_needed) {
 			log_debug(1 << CXGBI_DBG_PDU_TX,
-				"csk 0x%p, skb %u/%u, wr %d < %u.\n",
-				csk, skb->len, skb->data_len,
-				credits_needed, csk->wr_cred);
+				  "csk 0x%p, skb %u/%u, wr %d < %u.\n",
+				  csk, skb->len, skb->data_len,
+				  credits_needed, csk->wr_cred);
+
+			csk->no_tx_credits++;
 			break;
 		}
+
+		csk->no_tx_credits = 0;
+
 		__skb_unlink(skb, &csk->write_queue);
 		set_wr_txq(skb, CPL_PRIORITY_DATA, csk->port_id);
-		skb->csum = credits_needed + flowclen16;
+		skb->csum = (__force __wsum)(credits_needed + flowclen16);
 		csk->wr_cred -= credits_needed;
 		csk->wr_una_cred += credits_needed;
 		cxgbi_sock_enqueue_wr(csk, skb);
@@ -771,25 +834,42 @@ static int push_tx_frames(struct cxgbi_sock *csk, int req_completion)
 			csk, skb->len, skb->data_len, credits_needed,
 			csk->wr_cred, csk->wr_una_cred);
 
+		if (!req_completion &&
+		    ((csk->wr_una_cred >= (csk->wr_max_cred / 2)) ||
+		     after(csk->write_seq, (csk->snd_una + csk->snd_win / 2))))
+			req_completion = 1;
+
 		if (likely(cxgbi_skcb_test_flag(skb, SKCBF_TX_NEED_HDR))) {
-			len += cxgbi_ulp_extra_len(cxgbi_skcb_ulp_mode(skb));
-			make_tx_data_wr(csk, skb, dlen, len, credits_needed,
-					req_completion);
+			u32 ulp_mode = cxgbi_skcb_tx_ulp_mode(skb);
+
+			if (cxgbi_skcb_test_flag(skb, SKCBF_TX_ISO)) {
+				iso_cpl = (struct cxgbi_iso_info *)skb->head;
+				num_pdu = iso_cpl->num_pdu;
+				hdr_len = cxgbi_skcb_tx_iscsi_hdrlen(skb);
+				len += (cxgbi_ulp_extra_len(ulp_mode) * num_pdu) +
+				       (hdr_len * (num_pdu - 1));
+			} else {
+				len += cxgbi_ulp_extra_len(ulp_mode);
+			}
+
+			cxgb4i_make_tx_data_wr(csk, skb, dlen, len,
+					       credits_needed, req_completion);
 			csk->snd_nxt += len;
 			cxgbi_skcb_clear_flag(skb, SKCBF_TX_NEED_HDR);
 		} else if (cxgbi_skcb_test_flag(skb, SKCBF_TX_FLAG_COMPL) &&
 			   (csk->wr_una_cred >= (csk->wr_max_cred / 2))) {
 			struct cpl_close_con_req *req =
 				(struct cpl_close_con_req *)skb->data;
-			req->wr.wr_hi |= htonl(FW_WR_COMPL_F);
+
+			req->wr.wr_hi |= cpu_to_be32(FW_WR_COMPL_F);
 		}
+
 		total_size += skb->truesize;
 		t4_set_arp_err_handler(skb, csk, arp_failure_skb_discard);
 
 		log_debug(1 << CXGBI_DBG_TOE | 1 << CXGBI_DBG_PDU_TX,
-			"csk 0x%p,%u,0x%lx,%u, skb 0x%p, %u.\n",
-			csk, csk->state, csk->flags, csk->tid, skb, len);
-
+			  "csk 0x%p,%u,0x%lx,%u, skb 0x%p, %u.\n",
+			  csk, csk->state, csk->flags, csk->tid, skb, len);
 		cxgb4_l2t_send(csk->cdev->ports[csk->port_id], skb, csk->l2t);
 	}
 	return total_size;
@@ -2111,10 +2191,30 @@ static int cxgb4i_ddp_init(struct cxgbi_device *cdev)
 	return 0;
 }
 
+static bool is_memfree(struct adapter *adap)
+{
+	u32 io;
+
+	io = t4_read_reg(adap, MA_TARGET_MEM_ENABLE_A);
+	if (is_t5(adap->params.chip)) {
+		if ((io & EXT_MEM0_ENABLE_F) || (io & EXT_MEM1_ENABLE_F))
+			return false;
+	} else if (io & EXT_MEM_ENABLE_F) {
+		return false;
+	}
+
+	return true;
+}
+
 static void *t4_uld_add(const struct cxgb4_lld_info *lldi)
 {
 	struct cxgbi_device *cdev;
 	struct port_info *pi;
+	struct net_device *ndev;
+	struct adapter *adap;
+	struct tid_info *t;
+	u32 max_cmds = CXGB4I_SCSI_HOST_QDEPTH;
+	u32 max_conn = CXGBI_MAX_CONN;
 	int i, rc;
 
 	cdev = cxgbi_device_register(sizeof(*lldi), lldi->nports);
@@ -2154,14 +2254,40 @@ static void *t4_uld_add(const struct cxgb4_lld_info *lldi)
 		pr_info("t4 0x%p ddp init failed %d.\n", cdev, rc);
 		goto err_out;
 	}
+
+	ndev = cdev->ports[0];
+	adap = netdev2adap(ndev);
+	if (adap) {
+		t = &adap->tids;
+		if (t->ntids <= CXGBI_MAX_CONN)
+			max_conn = t->ntids;
+
+		if (is_memfree(adap)) {
+			cdev->flags |=	CXGBI_FLAG_DEV_ISO_OFF;
+			max_cmds = CXGB4I_SCSI_HOST_QDEPTH >> 2;
+
+			pr_info("%s: 0x%p, tid %u, SO adapter.\n",
+				ndev->name, cdev, t->ntids);
+		}
+	} else {
+		pr_info("%s, 0x%p, NO adapter struct.\n", ndev->name, cdev);
+	}
+
+	/* ISO is enabled in T5/T6 firmware version >= 1.13.43.0 */
+	if (!is_t4(lldi->adapter_type) &&
+	    (lldi->fw_vers >= 0x10d2b00) &&
+	    !(cdev->flags & CXGBI_FLAG_DEV_ISO_OFF))
+		cdev->skb_iso_txhdr = sizeof(struct cpl_tx_data_iso);
+
 	rc = cxgb4i_ofld_init(cdev);
 	if (rc) {
 		pr_info("t4 0x%p ofld init failed.\n", cdev);
 		goto err_out;
 	}
 
-	rc = cxgbi_hbas_add(cdev, CXGB4I_MAX_LUN, CXGBI_MAX_CONN,
-				&cxgb4i_host_template, cxgb4i_stt);
+	cxgb4i_host_template.can_queue = max_cmds;
+	rc = cxgbi_hbas_add(cdev, CXGB4I_MAX_LUN, max_conn,
+			    &cxgb4i_host_template, cxgb4i_stt);
 	if (rc)
 		goto err_out;