xref: /openbmc/linux/drivers/scsi/cxgbi/cxgb4i/cxgb4i.c (revision 12eb4683)
1 /*
2  * cxgb4i.c: Chelsio T4 iSCSI driver.
3  *
4  * Copyright (c) 2010 Chelsio Communications, Inc.
5  *
6  * This program is free software; you can redistribute it and/or modify
7  * it under the terms of the GNU General Public License as published by
8  * the Free Software Foundation.
9  *
10  * Written by:	Karen Xie (kxie@chelsio.com)
11  *		Rakesh Ranjan (rranjan@chelsio.com)
12  */
13 
14 #define pr_fmt(fmt) KBUILD_MODNAME ":%s: " fmt, __func__
15 
16 #include <linux/module.h>
17 #include <linux/moduleparam.h>
18 #include <scsi/scsi_host.h>
19 #include <net/tcp.h>
20 #include <net/dst.h>
21 #include <linux/netdevice.h>
22 
23 #include "t4_regs.h"
24 #include "t4_msg.h"
25 #include "cxgb4.h"
26 #include "cxgb4_uld.h"
27 #include "t4fw_api.h"
28 #include "l2t.h"
29 #include "cxgb4i.h"
30 
31 static unsigned int dbg_level;
32 
33 #include "../libcxgbi.h"
34 
35 #define	DRV_MODULE_NAME		"cxgb4i"
36 #define DRV_MODULE_DESC		"Chelsio T4/T5 iSCSI Driver"
37 #define	DRV_MODULE_VERSION	"0.9.4"
38 
39 static char version[] =
40 	DRV_MODULE_DESC " " DRV_MODULE_NAME
41 	" v" DRV_MODULE_VERSION "\n";
42 
43 MODULE_AUTHOR("Chelsio Communications, Inc.");
44 MODULE_DESCRIPTION(DRV_MODULE_DESC);
45 MODULE_VERSION(DRV_MODULE_VERSION);
46 MODULE_LICENSE("GPL");
47 
48 module_param(dbg_level, uint, 0644);
49 MODULE_PARM_DESC(dbg_level, "Debug flag (default=0)");
50 
51 static int cxgb4i_rcv_win = 256 * 1024;
52 module_param(cxgb4i_rcv_win, int, 0644);
53 MODULE_PARM_DESC(cxgb4i_rcv_win, "TCP reveive window in bytes");
54 
55 static int cxgb4i_snd_win = 128 * 1024;
56 module_param(cxgb4i_snd_win, int, 0644);
57 MODULE_PARM_DESC(cxgb4i_snd_win, "TCP send window in bytes");
58 
59 static int cxgb4i_rx_credit_thres = 10 * 1024;
60 module_param(cxgb4i_rx_credit_thres, int, 0644);
61 MODULE_PARM_DESC(cxgb4i_rx_credit_thres,
62 		"RX credits return threshold in bytes (default=10KB)");
63 
64 static unsigned int cxgb4i_max_connect = (8 * 1024);
65 module_param(cxgb4i_max_connect, uint, 0644);
66 MODULE_PARM_DESC(cxgb4i_max_connect, "Maximum number of connections");
67 
68 static unsigned short cxgb4i_sport_base = 20000;
69 module_param(cxgb4i_sport_base, ushort, 0644);
70 MODULE_PARM_DESC(cxgb4i_sport_base, "Starting port number (default 20000)");
71 
72 typedef void (*cxgb4i_cplhandler_func)(struct cxgbi_device *, struct sk_buff *);
73 
74 static void *t4_uld_add(const struct cxgb4_lld_info *);
75 static int t4_uld_rx_handler(void *, const __be64 *, const struct pkt_gl *);
76 static int t4_uld_state_change(void *, enum cxgb4_state state);
77 
78 static const struct cxgb4_uld_info cxgb4i_uld_info = {
79 	.name = DRV_MODULE_NAME,
80 	.add = t4_uld_add,
81 	.rx_handler = t4_uld_rx_handler,
82 	.state_change = t4_uld_state_change,
83 };
84 
85 static struct scsi_host_template cxgb4i_host_template = {
86 	.module		= THIS_MODULE,
87 	.name		= DRV_MODULE_NAME,
88 	.proc_name	= DRV_MODULE_NAME,
89 	.can_queue	= CXGB4I_SCSI_HOST_QDEPTH,
90 	.queuecommand	= iscsi_queuecommand,
91 	.change_queue_depth = iscsi_change_queue_depth,
92 	.sg_tablesize	= SG_ALL,
93 	.max_sectors	= 0xFFFF,
94 	.cmd_per_lun	= ISCSI_DEF_CMD_PER_LUN,
95 	.eh_abort_handler = iscsi_eh_abort,
96 	.eh_device_reset_handler = iscsi_eh_device_reset,
97 	.eh_target_reset_handler = iscsi_eh_recover_target,
98 	.target_alloc	= iscsi_target_alloc,
99 	.use_clustering	= DISABLE_CLUSTERING,
100 	.this_id	= -1,
101 };
102 
103 static struct iscsi_transport cxgb4i_iscsi_transport = {
104 	.owner		= THIS_MODULE,
105 	.name		= DRV_MODULE_NAME,
106 	.caps		= CAP_RECOVERY_L0 | CAP_MULTI_R2T | CAP_HDRDGST |
107 				CAP_DATADGST | CAP_DIGEST_OFFLOAD |
108 				CAP_PADDING_OFFLOAD | CAP_TEXT_NEGO,
109 	.attr_is_visible	= cxgbi_attr_is_visible,
110 	.get_host_param	= cxgbi_get_host_param,
111 	.set_host_param	= cxgbi_set_host_param,
112 	/* session management */
113 	.create_session	= cxgbi_create_session,
114 	.destroy_session	= cxgbi_destroy_session,
115 	.get_session_param = iscsi_session_get_param,
116 	/* connection management */
117 	.create_conn	= cxgbi_create_conn,
118 	.bind_conn		= cxgbi_bind_conn,
119 	.destroy_conn	= iscsi_tcp_conn_teardown,
120 	.start_conn		= iscsi_conn_start,
121 	.stop_conn		= iscsi_conn_stop,
122 	.get_conn_param	= iscsi_conn_get_param,
123 	.set_param	= cxgbi_set_conn_param,
124 	.get_stats	= cxgbi_get_conn_stats,
125 	/* pdu xmit req from user space */
126 	.send_pdu	= iscsi_conn_send_pdu,
127 	/* task */
128 	.init_task	= iscsi_tcp_task_init,
129 	.xmit_task	= iscsi_tcp_task_xmit,
130 	.cleanup_task	= cxgbi_cleanup_task,
131 	/* pdu */
132 	.alloc_pdu	= cxgbi_conn_alloc_pdu,
133 	.init_pdu	= cxgbi_conn_init_pdu,
134 	.xmit_pdu	= cxgbi_conn_xmit_pdu,
135 	.parse_pdu_itt	= cxgbi_parse_pdu_itt,
136 	/* TCP connect/disconnect */
137 	.get_ep_param	= cxgbi_get_ep_param,
138 	.ep_connect	= cxgbi_ep_connect,
139 	.ep_poll	= cxgbi_ep_poll,
140 	.ep_disconnect	= cxgbi_ep_disconnect,
141 	/* Error recovery timeout call */
142 	.session_recovery_timedout = iscsi_session_recovery_timedout,
143 };
144 
145 static struct scsi_transport_template *cxgb4i_stt;
146 
147 /*
148  * CPL (Chelsio Protocol Language) defines a message passing interface between
149  * the host driver and Chelsio asic.
150  * The section below implments CPLs that related to iscsi tcp connection
151  * open/close/abort and data send/receive.
152  */
153 #define DIV_ROUND_UP(n, d)	(((n) + (d) - 1) / (d))
154 #define RCV_BUFSIZ_MASK		0x3FFU
155 #define MAX_IMM_TX_PKT_LEN	128
156 
157 static inline void set_queue(struct sk_buff *skb, unsigned int queue,
158 				const struct cxgbi_sock *csk)
159 {
160 	skb->queue_mapping = queue;
161 }
162 
163 static int push_tx_frames(struct cxgbi_sock *, int);
164 
165 /*
166  * is_ofld_imm - check whether a packet can be sent as immediate data
167  * @skb: the packet
168  *
169  * Returns true if a packet can be sent as an offload WR with immediate
170  * data.  We currently use the same limit as for Ethernet packets.
171  */
172 static inline int is_ofld_imm(const struct sk_buff *skb)
173 {
174 	return skb->len <= (MAX_IMM_TX_PKT_LEN -
175 			sizeof(struct fw_ofld_tx_data_wr));
176 }
177 
178 
179 #define VLAN_NONE 0xfff
180 #define FILTER_SEL_VLAN_NONE 0xffff
181 #define FILTER_SEL_WIDTH_P_FC (3+1) /* port uses 3 bits, FCoE one bit */
182 #define FILTER_SEL_WIDTH_VIN_P_FC \
183 	(6 + 7 + FILTER_SEL_WIDTH_P_FC) /* 6 bits are unused, VF uses 7 bits*/
184 #define FILTER_SEL_WIDTH_TAG_P_FC \
185 	(3 + FILTER_SEL_WIDTH_VIN_P_FC) /* PF uses 3 bits */
186 #define FILTER_SEL_WIDTH_VLD_TAG_P_FC (1 + FILTER_SEL_WIDTH_TAG_P_FC)
187 
188 static unsigned int select_ntuple(struct cxgbi_device *cdev,
189 				struct l2t_entry *l2t)
190 {
191 	struct cxgb4_lld_info *lldi = cxgbi_cdev_priv(cdev);
192 	unsigned int ntuple = 0;
193 	u32 viid;
194 
195 	switch (lldi->filt_mode) {
196 
197 	/* default filter mode */
198 	case HW_TPL_FR_MT_PR_IV_P_FC:
199 		if (l2t->vlan == VLAN_NONE)
200 			ntuple |= FILTER_SEL_VLAN_NONE << FILTER_SEL_WIDTH_P_FC;
201 		else {
202 			ntuple |= l2t->vlan << FILTER_SEL_WIDTH_P_FC;
203 			ntuple |= 1 << FILTER_SEL_WIDTH_VLD_TAG_P_FC;
204 		}
205 		ntuple |= l2t->lport << S_PORT | IPPROTO_TCP <<
206 			  FILTER_SEL_WIDTH_VLD_TAG_P_FC;
207 		break;
208 	case HW_TPL_FR_MT_PR_OV_P_FC: {
209 		viid = cxgb4_port_viid(l2t->neigh->dev);
210 
211 		ntuple |= FW_VIID_VIN_GET(viid) << FILTER_SEL_WIDTH_P_FC;
212 		ntuple |= FW_VIID_PFN_GET(viid) << FILTER_SEL_WIDTH_VIN_P_FC;
213 		ntuple |= FW_VIID_VIVLD_GET(viid) << FILTER_SEL_WIDTH_TAG_P_FC;
214 		ntuple |= l2t->lport << S_PORT | IPPROTO_TCP <<
215 			  FILTER_SEL_WIDTH_VLD_TAG_P_FC;
216 		break;
217 	}
218 	default:
219 		break;
220 	}
221 	return ntuple;
222 }
223 
224 static void send_act_open_req(struct cxgbi_sock *csk, struct sk_buff *skb,
225 				struct l2t_entry *e)
226 {
227 	struct cxgb4_lld_info *lldi = cxgbi_cdev_priv(csk->cdev);
228 	int wscale = cxgbi_sock_compute_wscale(csk->mss_idx);
229 	unsigned long long opt0;
230 	unsigned int opt2;
231 	unsigned int qid_atid = ((unsigned int)csk->atid) |
232 				 (((unsigned int)csk->rss_qid) << 14);
233 
234 	opt0 = KEEP_ALIVE(1) |
235 		WND_SCALE(wscale) |
236 		MSS_IDX(csk->mss_idx) |
237 		L2T_IDX(((struct l2t_entry *)csk->l2t)->idx) |
238 		TX_CHAN(csk->tx_chan) |
239 		SMAC_SEL(csk->smac_idx) |
240 		ULP_MODE(ULP_MODE_ISCSI) |
241 		RCV_BUFSIZ(cxgb4i_rcv_win >> 10);
242 	opt2 = RX_CHANNEL(0) |
243 		RSS_QUEUE_VALID |
244 		(1 << 20) |
245 		RSS_QUEUE(csk->rss_qid);
246 
247 	if (is_t4(lldi->adapter_type)) {
248 		struct cpl_act_open_req *req =
249 				(struct cpl_act_open_req *)skb->head;
250 
251 		req = (struct cpl_act_open_req *)skb->head;
252 
253 		INIT_TP_WR(req, 0);
254 		OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_ACT_OPEN_REQ,
255 					qid_atid));
256 		req->local_port = csk->saddr.sin_port;
257 		req->peer_port = csk->daddr.sin_port;
258 		req->local_ip = csk->saddr.sin_addr.s_addr;
259 		req->peer_ip = csk->daddr.sin_addr.s_addr;
260 		req->opt0 = cpu_to_be64(opt0);
261 		req->params = cpu_to_be32(select_ntuple(csk->cdev, csk->l2t));
262 		opt2 |= 1 << 22;
263 		req->opt2 = cpu_to_be32(opt2);
264 
265 		log_debug(1 << CXGBI_DBG_TOE | 1 << CXGBI_DBG_SOCK,
266 			"csk t4 0x%p, %pI4:%u-%pI4:%u, atid %d, qid %u.\n",
267 			csk, &req->local_ip, ntohs(req->local_port),
268 			&req->peer_ip, ntohs(req->peer_port),
269 			csk->atid, csk->rss_qid);
270 	} else {
271 		struct cpl_t5_act_open_req *req =
272 				(struct cpl_t5_act_open_req *)skb->head;
273 
274 		req = (struct cpl_t5_act_open_req *)skb->head;
275 
276 		INIT_TP_WR(req, 0);
277 		OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_ACT_OPEN_REQ,
278 					qid_atid));
279 		req->local_port = csk->saddr.sin_port;
280 		req->peer_port = csk->daddr.sin_port;
281 		req->local_ip = csk->saddr.sin_addr.s_addr;
282 		req->peer_ip = csk->daddr.sin_addr.s_addr;
283 		req->opt0 = cpu_to_be64(opt0);
284 		req->params = cpu_to_be32(select_ntuple(csk->cdev, csk->l2t));
285 		opt2 |= 1 << 31;
286 		req->opt2 = cpu_to_be32(opt2);
287 
288 		log_debug(1 << CXGBI_DBG_TOE | 1 << CXGBI_DBG_SOCK,
289 			"csk t5 0x%p, %pI4:%u-%pI4:%u, atid %d, qid %u.\n",
290 			csk, &req->local_ip, ntohs(req->local_port),
291 			&req->peer_ip, ntohs(req->peer_port),
292 			csk->atid, csk->rss_qid);
293 	}
294 
295 	set_wr_txq(skb, CPL_PRIORITY_SETUP, csk->port_id);
296 	cxgb4_l2t_send(csk->cdev->ports[csk->port_id], skb, csk->l2t);
297 }
298 
299 static void send_close_req(struct cxgbi_sock *csk)
300 {
301 	struct sk_buff *skb = csk->cpl_close;
302 	struct cpl_close_con_req *req = (struct cpl_close_con_req *)skb->head;
303 	unsigned int tid = csk->tid;
304 
305 	log_debug(1 << CXGBI_DBG_TOE | 1 << CXGBI_DBG_SOCK,
306 		"csk 0x%p,%u,0x%lx, tid %u.\n",
307 		csk, csk->state, csk->flags, csk->tid);
308 	csk->cpl_close = NULL;
309 	set_wr_txq(skb, CPL_PRIORITY_DATA, csk->port_id);
310 	INIT_TP_WR(req, tid);
311 	OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_CLOSE_CON_REQ, tid));
312 	req->rsvd = 0;
313 
314 	cxgbi_sock_skb_entail(csk, skb);
315 	if (csk->state >= CTP_ESTABLISHED)
316 		push_tx_frames(csk, 1);
317 }
318 
319 static void abort_arp_failure(void *handle, struct sk_buff *skb)
320 {
321 	struct cxgbi_sock *csk = (struct cxgbi_sock *)handle;
322 	struct cpl_abort_req *req;
323 
324 	log_debug(1 << CXGBI_DBG_TOE | 1 << CXGBI_DBG_SOCK,
325 		"csk 0x%p,%u,0x%lx, tid %u, abort.\n",
326 		csk, csk->state, csk->flags, csk->tid);
327 	req = (struct cpl_abort_req *)skb->data;
328 	req->cmd = CPL_ABORT_NO_RST;
329 	cxgb4_ofld_send(csk->cdev->ports[csk->port_id], skb);
330 }
331 
332 static void send_abort_req(struct cxgbi_sock *csk)
333 {
334 	struct cpl_abort_req *req;
335 	struct sk_buff *skb = csk->cpl_abort_req;
336 
337 	if (unlikely(csk->state == CTP_ABORTING) || !skb || !csk->cdev)
338 		return;
339 	cxgbi_sock_set_state(csk, CTP_ABORTING);
340 	cxgbi_sock_set_flag(csk, CTPF_ABORT_RPL_PENDING);
341 	cxgbi_sock_purge_write_queue(csk);
342 
343 	csk->cpl_abort_req = NULL;
344 	req = (struct cpl_abort_req *)skb->head;
345 	set_queue(skb, CPL_PRIORITY_DATA, csk);
346 	req->cmd = CPL_ABORT_SEND_RST;
347 	t4_set_arp_err_handler(skb, csk, abort_arp_failure);
348 	INIT_TP_WR(req, csk->tid);
349 	OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_ABORT_REQ, csk->tid));
350 	req->rsvd0 = htonl(csk->snd_nxt);
351 	req->rsvd1 = !cxgbi_sock_flag(csk, CTPF_TX_DATA_SENT);
352 
353 	log_debug(1 << CXGBI_DBG_TOE | 1 << CXGBI_DBG_SOCK,
354 		"csk 0x%p,%u,0x%lx,%u, snd_nxt %u, 0x%x.\n",
355 		csk, csk->state, csk->flags, csk->tid, csk->snd_nxt,
356 		req->rsvd1);
357 
358 	cxgb4_l2t_send(csk->cdev->ports[csk->port_id], skb, csk->l2t);
359 }
360 
361 static void send_abort_rpl(struct cxgbi_sock *csk, int rst_status)
362 {
363 	struct sk_buff *skb = csk->cpl_abort_rpl;
364 	struct cpl_abort_rpl *rpl = (struct cpl_abort_rpl *)skb->head;
365 
366 	log_debug(1 << CXGBI_DBG_TOE | 1 << CXGBI_DBG_SOCK,
367 		"csk 0x%p,%u,0x%lx,%u, status %d.\n",
368 		csk, csk->state, csk->flags, csk->tid, rst_status);
369 
370 	csk->cpl_abort_rpl = NULL;
371 	set_queue(skb, CPL_PRIORITY_DATA, csk);
372 	INIT_TP_WR(rpl, csk->tid);
373 	OPCODE_TID(rpl) = cpu_to_be32(MK_OPCODE_TID(CPL_ABORT_RPL, csk->tid));
374 	rpl->cmd = rst_status;
375 	cxgb4_ofld_send(csk->cdev->ports[csk->port_id], skb);
376 }
377 
378 /*
379  * CPL connection rx data ack: host ->
380  * Send RX credits through an RX_DATA_ACK CPL message. Returns the number of
381  * credits sent.
382  */
383 static u32 send_rx_credits(struct cxgbi_sock *csk, u32 credits)
384 {
385 	struct sk_buff *skb;
386 	struct cpl_rx_data_ack *req;
387 
388 	log_debug(1 << CXGBI_DBG_TOE | 1 << CXGBI_DBG_PDU_RX,
389 		"csk 0x%p,%u,0x%lx,%u, credit %u.\n",
390 		csk, csk->state, csk->flags, csk->tid, credits);
391 
392 	skb = alloc_wr(sizeof(*req), 0, GFP_ATOMIC);
393 	if (!skb) {
394 		pr_info("csk 0x%p, credit %u, OOM.\n", csk, credits);
395 		return 0;
396 	}
397 	req = (struct cpl_rx_data_ack *)skb->head;
398 
399 	set_wr_txq(skb, CPL_PRIORITY_ACK, csk->port_id);
400 	INIT_TP_WR(req, csk->tid);
401 	OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_RX_DATA_ACK,
402 				      csk->tid));
403 	req->credit_dack = cpu_to_be32(RX_CREDITS(credits) | RX_FORCE_ACK(1));
404 	cxgb4_ofld_send(csk->cdev->ports[csk->port_id], skb);
405 	return credits;
406 }
407 
408 /*
409  * sgl_len - calculates the size of an SGL of the given capacity
410  * @n: the number of SGL entries
411  * Calculates the number of flits needed for a scatter/gather list that
412  * can hold the given number of entries.
413  */
414 static inline unsigned int sgl_len(unsigned int n)
415 {
416 	n--;
417 	return (3 * n) / 2 + (n & 1) + 2;
418 }
419 
420 /*
421  * calc_tx_flits_ofld - calculate # of flits for an offload packet
422  * @skb: the packet
423  *
424  * Returns the number of flits needed for the given offload packet.
425  * These packets are already fully constructed and no additional headers
426  * will be added.
427  */
428 static inline unsigned int calc_tx_flits_ofld(const struct sk_buff *skb)
429 {
430 	unsigned int flits, cnt;
431 
432 	if (is_ofld_imm(skb))
433 		return DIV_ROUND_UP(skb->len, 8);
434 	flits = skb_transport_offset(skb) / 8;
435 	cnt = skb_shinfo(skb)->nr_frags;
436 	if (skb_tail_pointer(skb) != skb_transport_header(skb))
437 		cnt++;
438 	return flits + sgl_len(cnt);
439 }
440 
441 static inline void send_tx_flowc_wr(struct cxgbi_sock *csk)
442 {
443 	struct sk_buff *skb;
444 	struct fw_flowc_wr *flowc;
445 	int flowclen, i;
446 
447 	flowclen = 80;
448 	skb = alloc_wr(flowclen, 0, GFP_ATOMIC);
449 	flowc = (struct fw_flowc_wr *)skb->head;
450 	flowc->op_to_nparams =
451 		htonl(FW_WR_OP(FW_FLOWC_WR) | FW_FLOWC_WR_NPARAMS(8));
452 	flowc->flowid_len16 =
453 		htonl(FW_WR_LEN16(DIV_ROUND_UP(72, 16)) |
454 				FW_WR_FLOWID(csk->tid));
455 	flowc->mnemval[0].mnemonic = FW_FLOWC_MNEM_PFNVFN;
456 	flowc->mnemval[0].val = htonl(csk->cdev->pfvf);
457 	flowc->mnemval[1].mnemonic = FW_FLOWC_MNEM_CH;
458 	flowc->mnemval[1].val = htonl(csk->tx_chan);
459 	flowc->mnemval[2].mnemonic = FW_FLOWC_MNEM_PORT;
460 	flowc->mnemval[2].val = htonl(csk->tx_chan);
461 	flowc->mnemval[3].mnemonic = FW_FLOWC_MNEM_IQID;
462 	flowc->mnemval[3].val = htonl(csk->rss_qid);
463 	flowc->mnemval[4].mnemonic = FW_FLOWC_MNEM_SNDNXT;
464 	flowc->mnemval[4].val = htonl(csk->snd_nxt);
465 	flowc->mnemval[5].mnemonic = FW_FLOWC_MNEM_RCVNXT;
466 	flowc->mnemval[5].val = htonl(csk->rcv_nxt);
467 	flowc->mnemval[6].mnemonic = FW_FLOWC_MNEM_SNDBUF;
468 	flowc->mnemval[6].val = htonl(cxgb4i_snd_win);
469 	flowc->mnemval[7].mnemonic = FW_FLOWC_MNEM_MSS;
470 	flowc->mnemval[7].val = htonl(csk->advmss);
471 	flowc->mnemval[8].mnemonic = 0;
472 	flowc->mnemval[8].val = 0;
473 	for (i = 0; i < 9; i++) {
474 		flowc->mnemval[i].r4[0] = 0;
475 		flowc->mnemval[i].r4[1] = 0;
476 		flowc->mnemval[i].r4[2] = 0;
477 	}
478 	set_queue(skb, CPL_PRIORITY_DATA, csk);
479 
480 	log_debug(1 << CXGBI_DBG_TOE | 1 << CXGBI_DBG_SOCK,
481 		"csk 0x%p, tid 0x%x, %u,%u,%u,%u,%u,%u,%u.\n",
482 		csk, csk->tid, 0, csk->tx_chan, csk->rss_qid,
483 		csk->snd_nxt, csk->rcv_nxt, cxgb4i_snd_win,
484 		csk->advmss);
485 
486 	cxgb4_ofld_send(csk->cdev->ports[csk->port_id], skb);
487 }
488 
489 static inline void make_tx_data_wr(struct cxgbi_sock *csk, struct sk_buff *skb,
490 				   int dlen, int len, u32 credits, int compl)
491 {
492 	struct fw_ofld_tx_data_wr *req;
493 	unsigned int submode = cxgbi_skcb_ulp_mode(skb) & 3;
494 	unsigned int wr_ulp_mode = 0;
495 
496 	req = (struct fw_ofld_tx_data_wr *)__skb_push(skb, sizeof(*req));
497 
498 	if (is_ofld_imm(skb)) {
499 		req->op_to_immdlen = htonl(FW_WR_OP(FW_OFLD_TX_DATA_WR) |
500 					FW_WR_COMPL(1) |
501 					FW_WR_IMMDLEN(dlen));
502 		req->flowid_len16 = htonl(FW_WR_FLOWID(csk->tid) |
503 						FW_WR_LEN16(credits));
504 	} else {
505 		req->op_to_immdlen =
506 			cpu_to_be32(FW_WR_OP(FW_OFLD_TX_DATA_WR) |
507 					FW_WR_COMPL(1) |
508 					FW_WR_IMMDLEN(0));
509 		req->flowid_len16 =
510 			cpu_to_be32(FW_WR_FLOWID(csk->tid) |
511 					FW_WR_LEN16(credits));
512 	}
513 	if (submode)
514 		wr_ulp_mode = FW_OFLD_TX_DATA_WR_ULPMODE(ULP2_MODE_ISCSI) |
515 				FW_OFLD_TX_DATA_WR_ULPSUBMODE(submode);
516 	req->tunnel_to_proxy = htonl(wr_ulp_mode |
517 		 FW_OFLD_TX_DATA_WR_SHOVE(skb_peek(&csk->write_queue) ? 0 : 1));
518 	req->plen = htonl(len);
519 	if (!cxgbi_sock_flag(csk, CTPF_TX_DATA_SENT))
520 		cxgbi_sock_set_flag(csk, CTPF_TX_DATA_SENT);
521 }
522 
523 static void arp_failure_skb_discard(void *handle, struct sk_buff *skb)
524 {
525 	kfree_skb(skb);
526 }
527 
528 static int push_tx_frames(struct cxgbi_sock *csk, int req_completion)
529 {
530 	int total_size = 0;
531 	struct sk_buff *skb;
532 
533 	if (unlikely(csk->state < CTP_ESTABLISHED ||
534 		csk->state == CTP_CLOSE_WAIT_1 || csk->state >= CTP_ABORTING)) {
535 		log_debug(1 << CXGBI_DBG_TOE | 1 << CXGBI_DBG_SOCK |
536 			 1 << CXGBI_DBG_PDU_TX,
537 			"csk 0x%p,%u,0x%lx,%u, in closing state.\n",
538 			csk, csk->state, csk->flags, csk->tid);
539 		return 0;
540 	}
541 
542 	while (csk->wr_cred && (skb = skb_peek(&csk->write_queue)) != NULL) {
543 		int dlen = skb->len;
544 		int len = skb->len;
545 		unsigned int credits_needed;
546 
547 		skb_reset_transport_header(skb);
548 		if (is_ofld_imm(skb))
549 			credits_needed = DIV_ROUND_UP(dlen +
550 					sizeof(struct fw_ofld_tx_data_wr), 16);
551 		else
552 			credits_needed = DIV_ROUND_UP(8*calc_tx_flits_ofld(skb)
553 					+ sizeof(struct fw_ofld_tx_data_wr),
554 					16);
555 
556 		if (csk->wr_cred < credits_needed) {
557 			log_debug(1 << CXGBI_DBG_PDU_TX,
558 				"csk 0x%p, skb %u/%u, wr %d < %u.\n",
559 				csk, skb->len, skb->data_len,
560 				credits_needed, csk->wr_cred);
561 			break;
562 		}
563 		__skb_unlink(skb, &csk->write_queue);
564 		set_queue(skb, CPL_PRIORITY_DATA, csk);
565 		skb->csum = credits_needed;
566 		csk->wr_cred -= credits_needed;
567 		csk->wr_una_cred += credits_needed;
568 		cxgbi_sock_enqueue_wr(csk, skb);
569 
570 		log_debug(1 << CXGBI_DBG_PDU_TX,
571 			"csk 0x%p, skb %u/%u, wr %d, left %u, unack %u.\n",
572 			csk, skb->len, skb->data_len, credits_needed,
573 			csk->wr_cred, csk->wr_una_cred);
574 
575 		if (likely(cxgbi_skcb_test_flag(skb, SKCBF_TX_NEED_HDR))) {
576 			if (!cxgbi_sock_flag(csk, CTPF_TX_DATA_SENT)) {
577 				send_tx_flowc_wr(csk);
578 				skb->csum += 5;
579 				csk->wr_cred -= 5;
580 				csk->wr_una_cred += 5;
581 			}
582 			len += cxgbi_ulp_extra_len(cxgbi_skcb_ulp_mode(skb));
583 			make_tx_data_wr(csk, skb, dlen, len, credits_needed,
584 					req_completion);
585 			csk->snd_nxt += len;
586 			cxgbi_skcb_clear_flag(skb, SKCBF_TX_NEED_HDR);
587 		}
588 		total_size += skb->truesize;
589 		t4_set_arp_err_handler(skb, csk, arp_failure_skb_discard);
590 
591 		log_debug(1 << CXGBI_DBG_TOE | 1 << CXGBI_DBG_PDU_TX,
592 			"csk 0x%p,%u,0x%lx,%u, skb 0x%p, %u.\n",
593 			csk, csk->state, csk->flags, csk->tid, skb, len);
594 
595 		cxgb4_l2t_send(csk->cdev->ports[csk->port_id], skb, csk->l2t);
596 	}
597 	return total_size;
598 }
599 
600 static inline void free_atid(struct cxgbi_sock *csk)
601 {
602 	struct cxgb4_lld_info *lldi = cxgbi_cdev_priv(csk->cdev);
603 
604 	if (cxgbi_sock_flag(csk, CTPF_HAS_ATID)) {
605 		cxgb4_free_atid(lldi->tids, csk->atid);
606 		cxgbi_sock_clear_flag(csk, CTPF_HAS_ATID);
607 		cxgbi_sock_put(csk);
608 	}
609 }
610 
611 static void do_act_establish(struct cxgbi_device *cdev, struct sk_buff *skb)
612 {
613 	struct cxgbi_sock *csk;
614 	struct cpl_act_establish *req = (struct cpl_act_establish *)skb->data;
615 	unsigned short tcp_opt = ntohs(req->tcp_opt);
616 	unsigned int tid = GET_TID(req);
617 	unsigned int atid = GET_TID_TID(ntohl(req->tos_atid));
618 	struct cxgb4_lld_info *lldi = cxgbi_cdev_priv(cdev);
619 	struct tid_info *t = lldi->tids;
620 	u32 rcv_isn = be32_to_cpu(req->rcv_isn);
621 
622 	csk = lookup_atid(t, atid);
623 	if (unlikely(!csk)) {
624 		pr_err("NO conn. for atid %u, cdev 0x%p.\n", atid, cdev);
625 		goto rel_skb;
626 	}
627 
628 	if (csk->atid != atid) {
629 		pr_err("bad conn atid %u, csk 0x%p,%u,0x%lx,tid %u, atid %u.\n",
630 			atid, csk, csk->state, csk->flags, csk->tid, csk->atid);
631 		goto rel_skb;
632 	}
633 
634 	log_debug(1 << CXGBI_DBG_TOE | 1 << CXGBI_DBG_SOCK,
635 		"csk 0x%p,%u,0x%lx, tid %u, atid %u, rseq %u.\n",
636 		csk, csk->state, csk->flags, tid, atid, rcv_isn);
637 
638 	cxgbi_sock_get(csk);
639 	csk->tid = tid;
640 	cxgb4_insert_tid(lldi->tids, csk, tid);
641 	cxgbi_sock_set_flag(csk, CTPF_HAS_TID);
642 
643 	free_atid(csk);
644 
645 	spin_lock_bh(&csk->lock);
646 	if (unlikely(csk->state != CTP_ACTIVE_OPEN))
647 		pr_info("csk 0x%p,%u,0x%lx,%u, got EST.\n",
648 			csk, csk->state, csk->flags, csk->tid);
649 
650 	if (csk->retry_timer.function) {
651 		del_timer(&csk->retry_timer);
652 		csk->retry_timer.function = NULL;
653 	}
654 
655 	csk->copied_seq = csk->rcv_wup = csk->rcv_nxt = rcv_isn;
656 	/*
657 	 * Causes the first RX_DATA_ACK to supply any Rx credits we couldn't
658 	 * pass through opt0.
659 	 */
660 	if (cxgb4i_rcv_win > (RCV_BUFSIZ_MASK << 10))
661 		csk->rcv_wup -= cxgb4i_rcv_win - (RCV_BUFSIZ_MASK << 10);
662 
663 	csk->advmss = lldi->mtus[GET_TCPOPT_MSS(tcp_opt)] - 40;
664 	if (GET_TCPOPT_TSTAMP(tcp_opt))
665 		csk->advmss -= 12;
666 	if (csk->advmss < 128)
667 		csk->advmss = 128;
668 
669 	log_debug(1 << CXGBI_DBG_TOE | 1 << CXGBI_DBG_SOCK,
670 		"csk 0x%p, mss_idx %u, advmss %u.\n",
671 			csk, GET_TCPOPT_MSS(tcp_opt), csk->advmss);
672 
673 	cxgbi_sock_established(csk, ntohl(req->snd_isn), ntohs(req->tcp_opt));
674 
675 	if (unlikely(cxgbi_sock_flag(csk, CTPF_ACTIVE_CLOSE_NEEDED)))
676 		send_abort_req(csk);
677 	else {
678 		if (skb_queue_len(&csk->write_queue))
679 			push_tx_frames(csk, 0);
680 		cxgbi_conn_tx_open(csk);
681 	}
682 	spin_unlock_bh(&csk->lock);
683 
684 rel_skb:
685 	__kfree_skb(skb);
686 }
687 
688 static int act_open_rpl_status_to_errno(int status)
689 {
690 	switch (status) {
691 	case CPL_ERR_CONN_RESET:
692 		return -ECONNREFUSED;
693 	case CPL_ERR_ARP_MISS:
694 		return -EHOSTUNREACH;
695 	case CPL_ERR_CONN_TIMEDOUT:
696 		return -ETIMEDOUT;
697 	case CPL_ERR_TCAM_FULL:
698 		return -ENOMEM;
699 	case CPL_ERR_CONN_EXIST:
700 		return -EADDRINUSE;
701 	default:
702 		return -EIO;
703 	}
704 }
705 
706 static void csk_act_open_retry_timer(unsigned long data)
707 {
708 	struct sk_buff *skb;
709 	struct cxgbi_sock *csk = (struct cxgbi_sock *)data;
710 	struct cxgb4_lld_info *lldi = cxgbi_cdev_priv(csk->cdev);
711 
712 	log_debug(1 << CXGBI_DBG_TOE | 1 << CXGBI_DBG_SOCK,
713 		"csk 0x%p,%u,0x%lx,%u.\n",
714 		csk, csk->state, csk->flags, csk->tid);
715 
716 	cxgbi_sock_get(csk);
717 	spin_lock_bh(&csk->lock);
718 	skb = alloc_wr(is_t4(lldi->adapter_type) ?
719 				sizeof(struct cpl_act_open_req) :
720 				sizeof(struct cpl_t5_act_open_req),
721 			0, GFP_ATOMIC);
722 	if (!skb)
723 		cxgbi_sock_fail_act_open(csk, -ENOMEM);
724 	else {
725 		skb->sk = (struct sock *)csk;
726 		t4_set_arp_err_handler(skb, csk,
727 					cxgbi_sock_act_open_req_arp_failure);
728 		send_act_open_req(csk, skb, csk->l2t);
729 	}
730 	spin_unlock_bh(&csk->lock);
731 	cxgbi_sock_put(csk);
732 }
733 
734 static void do_act_open_rpl(struct cxgbi_device *cdev, struct sk_buff *skb)
735 {
736 	struct cxgbi_sock *csk;
737 	struct cpl_act_open_rpl *rpl = (struct cpl_act_open_rpl *)skb->data;
738 	unsigned int tid = GET_TID(rpl);
739 	unsigned int atid =
740 		GET_TID_TID(GET_AOPEN_ATID(be32_to_cpu(rpl->atid_status)));
741 	unsigned int status = GET_AOPEN_STATUS(be32_to_cpu(rpl->atid_status));
742 	struct cxgb4_lld_info *lldi = cxgbi_cdev_priv(cdev);
743 	struct tid_info *t = lldi->tids;
744 
745 	csk = lookup_atid(t, atid);
746 	if (unlikely(!csk)) {
747 		pr_err("NO matching conn. atid %u, tid %u.\n", atid, tid);
748 		goto rel_skb;
749 	}
750 
751 	pr_info("%pI4:%u-%pI4:%u, atid %u,%u, status %u, csk 0x%p,%u,0x%lx.\n",
752 		&csk->saddr.sin_addr.s_addr, ntohs(csk->saddr.sin_port),
753 		&csk->daddr.sin_addr.s_addr, ntohs(csk->daddr.sin_port),
754 		atid, tid, status, csk, csk->state, csk->flags);
755 
756 	if (status == CPL_ERR_RTX_NEG_ADVICE)
757 		goto rel_skb;
758 
759 	if (status && status != CPL_ERR_TCAM_FULL &&
760 	    status != CPL_ERR_CONN_EXIST &&
761 	    status != CPL_ERR_ARP_MISS)
762 		cxgb4_remove_tid(lldi->tids, csk->port_id, GET_TID(rpl));
763 
764 	cxgbi_sock_get(csk);
765 	spin_lock_bh(&csk->lock);
766 
767 	if (status == CPL_ERR_CONN_EXIST &&
768 	    csk->retry_timer.function != csk_act_open_retry_timer) {
769 		csk->retry_timer.function = csk_act_open_retry_timer;
770 		mod_timer(&csk->retry_timer, jiffies + HZ / 2);
771 	} else
772 		cxgbi_sock_fail_act_open(csk,
773 					act_open_rpl_status_to_errno(status));
774 
775 	spin_unlock_bh(&csk->lock);
776 	cxgbi_sock_put(csk);
777 rel_skb:
778 	__kfree_skb(skb);
779 }
780 
781 static void do_peer_close(struct cxgbi_device *cdev, struct sk_buff *skb)
782 {
783 	struct cxgbi_sock *csk;
784 	struct cpl_peer_close *req = (struct cpl_peer_close *)skb->data;
785 	unsigned int tid = GET_TID(req);
786 	struct cxgb4_lld_info *lldi = cxgbi_cdev_priv(cdev);
787 	struct tid_info *t = lldi->tids;
788 
789 	csk = lookup_tid(t, tid);
790 	if (unlikely(!csk)) {
791 		pr_err("can't find connection for tid %u.\n", tid);
792 		goto rel_skb;
793 	}
794 	log_debug(1 << CXGBI_DBG_TOE | 1 << CXGBI_DBG_SOCK,
795 		"csk 0x%p,%u,0x%lx,%u.\n",
796 		csk, csk->state, csk->flags, csk->tid);
797 	cxgbi_sock_rcv_peer_close(csk);
798 rel_skb:
799 	__kfree_skb(skb);
800 }
801 
802 static void do_close_con_rpl(struct cxgbi_device *cdev, struct sk_buff *skb)
803 {
804 	struct cxgbi_sock *csk;
805 	struct cpl_close_con_rpl *rpl = (struct cpl_close_con_rpl *)skb->data;
806 	unsigned int tid = GET_TID(rpl);
807 	struct cxgb4_lld_info *lldi = cxgbi_cdev_priv(cdev);
808 	struct tid_info *t = lldi->tids;
809 
810 	csk = lookup_tid(t, tid);
811 	if (unlikely(!csk)) {
812 		pr_err("can't find connection for tid %u.\n", tid);
813 		goto rel_skb;
814 	}
815 	log_debug(1 << CXGBI_DBG_TOE | 1 << CXGBI_DBG_SOCK,
816 		"csk 0x%p,%u,0x%lx,%u.\n",
817 		csk, csk->state, csk->flags, csk->tid);
818 	cxgbi_sock_rcv_close_conn_rpl(csk, ntohl(rpl->snd_nxt));
819 rel_skb:
820 	__kfree_skb(skb);
821 }
822 
823 static int abort_status_to_errno(struct cxgbi_sock *csk, int abort_reason,
824 								int *need_rst)
825 {
826 	switch (abort_reason) {
827 	case CPL_ERR_BAD_SYN: /* fall through */
828 	case CPL_ERR_CONN_RESET:
829 		return csk->state > CTP_ESTABLISHED ?
830 			-EPIPE : -ECONNRESET;
831 	case CPL_ERR_XMIT_TIMEDOUT:
832 	case CPL_ERR_PERSIST_TIMEDOUT:
833 	case CPL_ERR_FINWAIT2_TIMEDOUT:
834 	case CPL_ERR_KEEPALIVE_TIMEDOUT:
835 		return -ETIMEDOUT;
836 	default:
837 		return -EIO;
838 	}
839 }
840 
841 static void do_abort_req_rss(struct cxgbi_device *cdev, struct sk_buff *skb)
842 {
843 	struct cxgbi_sock *csk;
844 	struct cpl_abort_req_rss *req = (struct cpl_abort_req_rss *)skb->data;
845 	unsigned int tid = GET_TID(req);
846 	struct cxgb4_lld_info *lldi = cxgbi_cdev_priv(cdev);
847 	struct tid_info *t = lldi->tids;
848 	int rst_status = CPL_ABORT_NO_RST;
849 
850 	csk = lookup_tid(t, tid);
851 	if (unlikely(!csk)) {
852 		pr_err("can't find connection for tid %u.\n", tid);
853 		goto rel_skb;
854 	}
855 
856 	log_debug(1 << CXGBI_DBG_TOE | 1 << CXGBI_DBG_SOCK,
857 		"csk 0x%p,%u,0x%lx, tid %u, status 0x%x.\n",
858 		csk, csk->state, csk->flags, csk->tid, req->status);
859 
860 	if (req->status == CPL_ERR_RTX_NEG_ADVICE ||
861 	    req->status == CPL_ERR_PERSIST_NEG_ADVICE)
862 		goto rel_skb;
863 
864 	cxgbi_sock_get(csk);
865 	spin_lock_bh(&csk->lock);
866 
867 	if (!cxgbi_sock_flag(csk, CTPF_ABORT_REQ_RCVD)) {
868 		cxgbi_sock_set_flag(csk, CTPF_ABORT_REQ_RCVD);
869 		cxgbi_sock_set_state(csk, CTP_ABORTING);
870 		goto done;
871 	}
872 
873 	cxgbi_sock_clear_flag(csk, CTPF_ABORT_REQ_RCVD);
874 	send_abort_rpl(csk, rst_status);
875 
876 	if (!cxgbi_sock_flag(csk, CTPF_ABORT_RPL_PENDING)) {
877 		csk->err = abort_status_to_errno(csk, req->status, &rst_status);
878 		cxgbi_sock_closed(csk);
879 	}
880 done:
881 	spin_unlock_bh(&csk->lock);
882 	cxgbi_sock_put(csk);
883 rel_skb:
884 	__kfree_skb(skb);
885 }
886 
887 static void do_abort_rpl_rss(struct cxgbi_device *cdev, struct sk_buff *skb)
888 {
889 	struct cxgbi_sock *csk;
890 	struct cpl_abort_rpl_rss *rpl = (struct cpl_abort_rpl_rss *)skb->data;
891 	unsigned int tid = GET_TID(rpl);
892 	struct cxgb4_lld_info *lldi = cxgbi_cdev_priv(cdev);
893 	struct tid_info *t = lldi->tids;
894 
895 	csk = lookup_tid(t, tid);
896 	if (!csk)
897 		goto rel_skb;
898 
899 	log_debug(1 << CXGBI_DBG_TOE | 1 << CXGBI_DBG_SOCK,
900 		"status 0x%x, csk 0x%p, s %u, 0x%lx.\n",
901 		rpl->status, csk, csk ? csk->state : 0,
902 		csk ? csk->flags : 0UL);
903 
904 	if (rpl->status == CPL_ERR_ABORT_FAILED)
905 		goto rel_skb;
906 
907 	cxgbi_sock_rcv_abort_rpl(csk);
908 rel_skb:
909 	__kfree_skb(skb);
910 }
911 
912 static void do_rx_iscsi_hdr(struct cxgbi_device *cdev, struct sk_buff *skb)
913 {
914 	struct cxgbi_sock *csk;
915 	struct cpl_iscsi_hdr *cpl = (struct cpl_iscsi_hdr *)skb->data;
916 	unsigned short pdu_len_ddp = be16_to_cpu(cpl->pdu_len_ddp);
917 	unsigned int tid = GET_TID(cpl);
918 	struct cxgb4_lld_info *lldi = cxgbi_cdev_priv(cdev);
919 	struct tid_info *t = lldi->tids;
920 
921 	csk = lookup_tid(t, tid);
922 	if (unlikely(!csk)) {
923 		pr_err("can't find conn. for tid %u.\n", tid);
924 		goto rel_skb;
925 	}
926 
927 	log_debug(1 << CXGBI_DBG_TOE | 1 << CXGBI_DBG_PDU_RX,
928 		"csk 0x%p,%u,0x%lx, tid %u, skb 0x%p,%u, 0x%x.\n",
929 		csk, csk->state, csk->flags, csk->tid, skb, skb->len,
930 		pdu_len_ddp);
931 
932 	spin_lock_bh(&csk->lock);
933 
934 	if (unlikely(csk->state >= CTP_PASSIVE_CLOSE)) {
935 		log_debug(1 << CXGBI_DBG_TOE | 1 << CXGBI_DBG_SOCK,
936 			"csk 0x%p,%u,0x%lx,%u, bad state.\n",
937 			csk, csk->state, csk->flags, csk->tid);
938 		if (csk->state != CTP_ABORTING)
939 			goto abort_conn;
940 		else
941 			goto discard;
942 	}
943 
944 	cxgbi_skcb_tcp_seq(skb) = ntohl(cpl->seq);
945 	cxgbi_skcb_flags(skb) = 0;
946 
947 	skb_reset_transport_header(skb);
948 	__skb_pull(skb, sizeof(*cpl));
949 	__pskb_trim(skb, ntohs(cpl->len));
950 
951 	if (!csk->skb_ulp_lhdr) {
952 		unsigned char *bhs;
953 		unsigned int hlen, dlen, plen;
954 
955 		log_debug(1 << CXGBI_DBG_TOE | 1 << CXGBI_DBG_PDU_RX,
956 			"csk 0x%p,%u,0x%lx, tid %u, skb 0x%p header.\n",
957 			csk, csk->state, csk->flags, csk->tid, skb);
958 		csk->skb_ulp_lhdr = skb;
959 		cxgbi_skcb_set_flag(skb, SKCBF_RX_HDR);
960 
961 		if (cxgbi_skcb_tcp_seq(skb) != csk->rcv_nxt) {
962 			pr_info("tid %u, CPL_ISCSI_HDR, bad seq, 0x%x/0x%x.\n",
963 				csk->tid, cxgbi_skcb_tcp_seq(skb),
964 				csk->rcv_nxt);
965 			goto abort_conn;
966 		}
967 
968 		bhs = skb->data;
969 		hlen = ntohs(cpl->len);
970 		dlen = ntohl(*(unsigned int *)(bhs + 4)) & 0xFFFFFF;
971 
972 		plen = ISCSI_PDU_LEN(pdu_len_ddp);
973 		if (is_t4(lldi->adapter_type))
974 			plen -= 40;
975 
976 		if ((hlen + dlen) != plen) {
977 			pr_info("tid 0x%x, CPL_ISCSI_HDR, pdu len "
978 				"mismatch %u != %u + %u, seq 0x%x.\n",
979 				csk->tid, plen, hlen, dlen,
980 				cxgbi_skcb_tcp_seq(skb));
981 			goto abort_conn;
982 		}
983 
984 		cxgbi_skcb_rx_pdulen(skb) = (hlen + dlen + 3) & (~0x3);
985 		if (dlen)
986 			cxgbi_skcb_rx_pdulen(skb) += csk->dcrc_len;
987 		csk->rcv_nxt += cxgbi_skcb_rx_pdulen(skb);
988 
989 		log_debug(1 << CXGBI_DBG_TOE | 1 << CXGBI_DBG_PDU_RX,
990 			"csk 0x%p, skb 0x%p, 0x%x,%u+%u,0x%x,0x%x.\n",
991 			csk, skb, *bhs, hlen, dlen,
992 			ntohl(*((unsigned int *)(bhs + 16))),
993 			ntohl(*((unsigned int *)(bhs + 24))));
994 
995 	} else {
996 		struct sk_buff *lskb = csk->skb_ulp_lhdr;
997 
998 		cxgbi_skcb_set_flag(lskb, SKCBF_RX_DATA);
999 		log_debug(1 << CXGBI_DBG_TOE | 1 << CXGBI_DBG_PDU_RX,
1000 			"csk 0x%p,%u,0x%lx, skb 0x%p data, 0x%p.\n",
1001 			csk, csk->state, csk->flags, skb, lskb);
1002 	}
1003 
1004 	__skb_queue_tail(&csk->receive_queue, skb);
1005 	spin_unlock_bh(&csk->lock);
1006 	return;
1007 
1008 abort_conn:
1009 	send_abort_req(csk);
1010 discard:
1011 	spin_unlock_bh(&csk->lock);
1012 rel_skb:
1013 	__kfree_skb(skb);
1014 }
1015 
1016 static void do_rx_data_ddp(struct cxgbi_device *cdev,
1017 				  struct sk_buff *skb)
1018 {
1019 	struct cxgbi_sock *csk;
1020 	struct sk_buff *lskb;
1021 	struct cpl_rx_data_ddp *rpl = (struct cpl_rx_data_ddp *)skb->data;
1022 	unsigned int tid = GET_TID(rpl);
1023 	struct cxgb4_lld_info *lldi = cxgbi_cdev_priv(cdev);
1024 	struct tid_info *t = lldi->tids;
1025 	unsigned int status = ntohl(rpl->ddpvld);
1026 
1027 	csk = lookup_tid(t, tid);
1028 	if (unlikely(!csk)) {
1029 		pr_err("can't find connection for tid %u.\n", tid);
1030 		goto rel_skb;
1031 	}
1032 
1033 	log_debug(1 << CXGBI_DBG_TOE | 1 << CXGBI_DBG_PDU_RX,
1034 		"csk 0x%p,%u,0x%lx, skb 0x%p,0x%x, lhdr 0x%p.\n",
1035 		csk, csk->state, csk->flags, skb, status, csk->skb_ulp_lhdr);
1036 
1037 	spin_lock_bh(&csk->lock);
1038 
1039 	if (unlikely(csk->state >= CTP_PASSIVE_CLOSE)) {
1040 		log_debug(1 << CXGBI_DBG_TOE | 1 << CXGBI_DBG_SOCK,
1041 			"csk 0x%p,%u,0x%lx,%u, bad state.\n",
1042 			csk, csk->state, csk->flags, csk->tid);
1043 		if (csk->state != CTP_ABORTING)
1044 			goto abort_conn;
1045 		else
1046 			goto discard;
1047 	}
1048 
1049 	if (!csk->skb_ulp_lhdr) {
1050 		pr_err("tid 0x%x, rcv RX_DATA_DDP w/o pdu bhs.\n", csk->tid);
1051 		goto abort_conn;
1052 	}
1053 
1054 	lskb = csk->skb_ulp_lhdr;
1055 	csk->skb_ulp_lhdr = NULL;
1056 
1057 	cxgbi_skcb_rx_ddigest(lskb) = ntohl(rpl->ulp_crc);
1058 
1059 	if (ntohs(rpl->len) != cxgbi_skcb_rx_pdulen(lskb))
1060 		pr_info("tid 0x%x, RX_DATA_DDP pdulen %u != %u.\n",
1061 			csk->tid, ntohs(rpl->len), cxgbi_skcb_rx_pdulen(lskb));
1062 
1063 	if (status & (1 << CPL_RX_DDP_STATUS_HCRC_SHIFT)) {
1064 		pr_info("csk 0x%p, lhdr 0x%p, status 0x%x, hcrc bad 0x%lx.\n",
1065 			csk, lskb, status, cxgbi_skcb_flags(lskb));
1066 		cxgbi_skcb_set_flag(lskb, SKCBF_RX_HCRC_ERR);
1067 	}
1068 	if (status & (1 << CPL_RX_DDP_STATUS_DCRC_SHIFT)) {
1069 		pr_info("csk 0x%p, lhdr 0x%p, status 0x%x, dcrc bad 0x%lx.\n",
1070 			csk, lskb, status, cxgbi_skcb_flags(lskb));
1071 		cxgbi_skcb_set_flag(lskb, SKCBF_RX_DCRC_ERR);
1072 	}
1073 	if (status & (1 << CPL_RX_DDP_STATUS_PAD_SHIFT)) {
1074 		log_debug(1 << CXGBI_DBG_PDU_RX,
1075 			"csk 0x%p, lhdr 0x%p, status 0x%x, pad bad.\n",
1076 			csk, lskb, status);
1077 		cxgbi_skcb_set_flag(lskb, SKCBF_RX_PAD_ERR);
1078 	}
1079 	if ((status & (1 << CPL_RX_DDP_STATUS_DDP_SHIFT)) &&
1080 		!cxgbi_skcb_test_flag(lskb, SKCBF_RX_DATA)) {
1081 		log_debug(1 << CXGBI_DBG_PDU_RX,
1082 			"csk 0x%p, lhdr 0x%p, 0x%x, data ddp'ed.\n",
1083 			csk, lskb, status);
1084 		cxgbi_skcb_set_flag(lskb, SKCBF_RX_DATA_DDPD);
1085 	}
1086 	log_debug(1 << CXGBI_DBG_PDU_RX,
1087 		"csk 0x%p, lskb 0x%p, f 0x%lx.\n",
1088 		csk, lskb, cxgbi_skcb_flags(lskb));
1089 
1090 	cxgbi_skcb_set_flag(lskb, SKCBF_RX_STATUS);
1091 	cxgbi_conn_pdu_ready(csk);
1092 	spin_unlock_bh(&csk->lock);
1093 	goto rel_skb;
1094 
1095 abort_conn:
1096 	send_abort_req(csk);
1097 discard:
1098 	spin_unlock_bh(&csk->lock);
1099 rel_skb:
1100 	__kfree_skb(skb);
1101 }
1102 
1103 static void do_fw4_ack(struct cxgbi_device *cdev, struct sk_buff *skb)
1104 {
1105 	struct cxgbi_sock *csk;
1106 	struct cpl_fw4_ack *rpl = (struct cpl_fw4_ack *)skb->data;
1107 	unsigned int tid = GET_TID(rpl);
1108 	struct cxgb4_lld_info *lldi = cxgbi_cdev_priv(cdev);
1109 	struct tid_info *t = lldi->tids;
1110 
1111 	csk = lookup_tid(t, tid);
1112 	if (unlikely(!csk))
1113 		pr_err("can't find connection for tid %u.\n", tid);
1114 	else {
1115 		log_debug(1 << CXGBI_DBG_TOE | 1 << CXGBI_DBG_SOCK,
1116 			"csk 0x%p,%u,0x%lx,%u.\n",
1117 			csk, csk->state, csk->flags, csk->tid);
1118 		cxgbi_sock_rcv_wr_ack(csk, rpl->credits, ntohl(rpl->snd_una),
1119 					rpl->seq_vld);
1120 	}
1121 	__kfree_skb(skb);
1122 }
1123 
1124 static void do_set_tcb_rpl(struct cxgbi_device *cdev, struct sk_buff *skb)
1125 {
1126 	struct cpl_set_tcb_rpl *rpl = (struct cpl_set_tcb_rpl *)skb->data;
1127 	unsigned int tid = GET_TID(rpl);
1128 	struct cxgb4_lld_info *lldi = cxgbi_cdev_priv(cdev);
1129 	struct tid_info *t = lldi->tids;
1130 	struct cxgbi_sock *csk;
1131 
1132 	csk = lookup_tid(t, tid);
1133 	if (!csk)
1134 		pr_err("can't find conn. for tid %u.\n", tid);
1135 
1136 	log_debug(1 << CXGBI_DBG_TOE | 1 << CXGBI_DBG_SOCK,
1137 		"csk 0x%p,%u,%lx,%u, status 0x%x.\n",
1138 		csk, csk->state, csk->flags, csk->tid, rpl->status);
1139 
1140 	if (rpl->status != CPL_ERR_NONE)
1141 		pr_err("csk 0x%p,%u, SET_TCB_RPL status %u.\n",
1142 			csk, tid, rpl->status);
1143 
1144 	__kfree_skb(skb);
1145 }
1146 
1147 static int alloc_cpls(struct cxgbi_sock *csk)
1148 {
1149 	csk->cpl_close = alloc_wr(sizeof(struct cpl_close_con_req),
1150 					0, GFP_KERNEL);
1151 	if (!csk->cpl_close)
1152 		return -ENOMEM;
1153 
1154 	csk->cpl_abort_req = alloc_wr(sizeof(struct cpl_abort_req),
1155 					0, GFP_KERNEL);
1156 	if (!csk->cpl_abort_req)
1157 		goto free_cpls;
1158 
1159 	csk->cpl_abort_rpl = alloc_wr(sizeof(struct cpl_abort_rpl),
1160 					0, GFP_KERNEL);
1161 	if (!csk->cpl_abort_rpl)
1162 		goto free_cpls;
1163 	return 0;
1164 
1165 free_cpls:
1166 	cxgbi_sock_free_cpl_skbs(csk);
1167 	return -ENOMEM;
1168 }
1169 
1170 static inline void l2t_put(struct cxgbi_sock *csk)
1171 {
1172 	if (csk->l2t) {
1173 		cxgb4_l2t_release(csk->l2t);
1174 		csk->l2t = NULL;
1175 		cxgbi_sock_put(csk);
1176 	}
1177 }
1178 
1179 static void release_offload_resources(struct cxgbi_sock *csk)
1180 {
1181 	struct cxgb4_lld_info *lldi;
1182 
1183 	log_debug(1 << CXGBI_DBG_TOE | 1 << CXGBI_DBG_SOCK,
1184 		"csk 0x%p,%u,0x%lx,%u.\n",
1185 		csk, csk->state, csk->flags, csk->tid);
1186 
1187 	cxgbi_sock_free_cpl_skbs(csk);
1188 	if (csk->wr_cred != csk->wr_max_cred) {
1189 		cxgbi_sock_purge_wr_queue(csk);
1190 		cxgbi_sock_reset_wr_list(csk);
1191 	}
1192 
1193 	l2t_put(csk);
1194 	if (cxgbi_sock_flag(csk, CTPF_HAS_ATID))
1195 		free_atid(csk);
1196 	else if (cxgbi_sock_flag(csk, CTPF_HAS_TID)) {
1197 		lldi = cxgbi_cdev_priv(csk->cdev);
1198 		cxgb4_remove_tid(lldi->tids, 0, csk->tid);
1199 		cxgbi_sock_clear_flag(csk, CTPF_HAS_TID);
1200 		cxgbi_sock_put(csk);
1201 	}
1202 	csk->dst = NULL;
1203 	csk->cdev = NULL;
1204 }
1205 
1206 static int init_act_open(struct cxgbi_sock *csk)
1207 {
1208 	struct cxgbi_device *cdev = csk->cdev;
1209 	struct cxgb4_lld_info *lldi = cxgbi_cdev_priv(cdev);
1210 	struct net_device *ndev = cdev->ports[csk->port_id];
1211 	struct port_info *pi = netdev_priv(ndev);
1212 	struct sk_buff *skb = NULL;
1213 	struct neighbour *n;
1214 	unsigned int step;
1215 
1216 	log_debug(1 << CXGBI_DBG_TOE | 1 << CXGBI_DBG_SOCK,
1217 		"csk 0x%p,%u,0x%lx,%u.\n",
1218 		csk, csk->state, csk->flags, csk->tid);
1219 
1220 	csk->atid = cxgb4_alloc_atid(lldi->tids, csk);
1221 	if (csk->atid < 0) {
1222 		pr_err("%s, NO atid available.\n", ndev->name);
1223 		return -EINVAL;
1224 	}
1225 	cxgbi_sock_set_flag(csk, CTPF_HAS_ATID);
1226 	cxgbi_sock_get(csk);
1227 
1228 	n = dst_neigh_lookup(csk->dst, &csk->daddr.sin_addr.s_addr);
1229 	if (!n) {
1230 		pr_err("%s, can't get neighbour of csk->dst.\n", ndev->name);
1231 		goto rel_resource;
1232 	}
1233 	csk->l2t = cxgb4_l2t_get(lldi->l2t, n, ndev, 0);
1234 	if (!csk->l2t) {
1235 		pr_err("%s, cannot alloc l2t.\n", ndev->name);
1236 		goto rel_resource;
1237 	}
1238 	cxgbi_sock_get(csk);
1239 
1240 	skb = alloc_wr(is_t4(lldi->adapter_type) ?
1241 				sizeof(struct cpl_act_open_req) :
1242 				sizeof(struct cpl_t5_act_open_req),
1243 			0, GFP_ATOMIC);
1244 	if (!skb)
1245 		goto rel_resource;
1246 	skb->sk = (struct sock *)csk;
1247 	t4_set_arp_err_handler(skb, csk, cxgbi_sock_act_open_req_arp_failure);
1248 
1249 	if (!csk->mtu)
1250 		csk->mtu = dst_mtu(csk->dst);
1251 	cxgb4_best_mtu(lldi->mtus, csk->mtu, &csk->mss_idx);
1252 	csk->tx_chan = cxgb4_port_chan(ndev);
1253 	/* SMT two entries per row */
1254 	csk->smac_idx = ((cxgb4_port_viid(ndev) & 0x7F)) << 1;
1255 	step = lldi->ntxq / lldi->nchan;
1256 	csk->txq_idx = cxgb4_port_idx(ndev) * step;
1257 	step = lldi->nrxq / lldi->nchan;
1258 	csk->rss_qid = lldi->rxq_ids[cxgb4_port_idx(ndev) * step];
1259 	csk->wr_max_cred = csk->wr_cred = lldi->wr_cred;
1260 	csk->wr_una_cred = 0;
1261 	cxgbi_sock_reset_wr_list(csk);
1262 	csk->err = 0;
1263 	log_debug(1 << CXGBI_DBG_TOE | 1 << CXGBI_DBG_SOCK,
1264 		"csk 0x%p,p%d,%s, %u,%u,%u, mss %u,%u, smac %u.\n",
1265 		csk, pi->port_id, ndev->name, csk->tx_chan,
1266 		csk->txq_idx, csk->rss_qid, csk->mtu, csk->mss_idx,
1267 		csk->smac_idx);
1268 
1269 	cxgbi_sock_set_state(csk, CTP_ACTIVE_OPEN);
1270 	send_act_open_req(csk, skb, csk->l2t);
1271 	neigh_release(n);
1272 	return 0;
1273 
1274 rel_resource:
1275 	if (n)
1276 		neigh_release(n);
1277 	if (skb)
1278 		__kfree_skb(skb);
1279 	return -EINVAL;
1280 }
1281 
1282 #define CPL_ISCSI_DATA		0xB2
1283 #define CPL_RX_ISCSI_DDP	0x49
1284 cxgb4i_cplhandler_func cxgb4i_cplhandlers[NUM_CPL_CMDS] = {
1285 	[CPL_ACT_ESTABLISH] = do_act_establish,
1286 	[CPL_ACT_OPEN_RPL] = do_act_open_rpl,
1287 	[CPL_PEER_CLOSE] = do_peer_close,
1288 	[CPL_ABORT_REQ_RSS] = do_abort_req_rss,
1289 	[CPL_ABORT_RPL_RSS] = do_abort_rpl_rss,
1290 	[CPL_CLOSE_CON_RPL] = do_close_con_rpl,
1291 	[CPL_FW4_ACK] = do_fw4_ack,
1292 	[CPL_ISCSI_HDR] = do_rx_iscsi_hdr,
1293 	[CPL_ISCSI_DATA] = do_rx_iscsi_hdr,
1294 	[CPL_SET_TCB_RPL] = do_set_tcb_rpl,
1295 	[CPL_RX_DATA_DDP] = do_rx_data_ddp,
1296 	[CPL_RX_ISCSI_DDP] = do_rx_data_ddp,
1297 };
1298 
1299 int cxgb4i_ofld_init(struct cxgbi_device *cdev)
1300 {
1301 	int rc;
1302 
1303 	if (cxgb4i_max_connect > CXGB4I_MAX_CONN)
1304 		cxgb4i_max_connect = CXGB4I_MAX_CONN;
1305 
1306 	rc = cxgbi_device_portmap_create(cdev, cxgb4i_sport_base,
1307 					cxgb4i_max_connect);
1308 	if (rc < 0)
1309 		return rc;
1310 
1311 	cdev->csk_release_offload_resources = release_offload_resources;
1312 	cdev->csk_push_tx_frames = push_tx_frames;
1313 	cdev->csk_send_abort_req = send_abort_req;
1314 	cdev->csk_send_close_req = send_close_req;
1315 	cdev->csk_send_rx_credits = send_rx_credits;
1316 	cdev->csk_alloc_cpls = alloc_cpls;
1317 	cdev->csk_init_act_open = init_act_open;
1318 
1319 	pr_info("cdev 0x%p, offload up, added.\n", cdev);
1320 	return 0;
1321 }
1322 
1323 /*
1324  * functions to program the pagepod in h/w
1325  */
1326 #define ULPMEM_IDATA_MAX_NPPODS	4 /* 256/PPOD_SIZE */
1327 static inline void ulp_mem_io_set_hdr(struct cxgb4_lld_info *lldi,
1328 				struct ulp_mem_io *req,
1329 				unsigned int wr_len, unsigned int dlen,
1330 				unsigned int pm_addr)
1331 {
1332 	struct ulptx_idata *idata = (struct ulptx_idata *)(req + 1);
1333 
1334 	INIT_ULPTX_WR(req, wr_len, 0, 0);
1335 	if (is_t4(lldi->adapter_type))
1336 		req->cmd = htonl(ULPTX_CMD(ULP_TX_MEM_WRITE) |
1337 					(ULP_MEMIO_ORDER(1)));
1338 	else
1339 		req->cmd = htonl(ULPTX_CMD(ULP_TX_MEM_WRITE) |
1340 					(V_T5_ULP_MEMIO_IMM(1)));
1341 	req->dlen = htonl(ULP_MEMIO_DATA_LEN(dlen >> 5));
1342 	req->lock_addr = htonl(ULP_MEMIO_ADDR(pm_addr >> 5));
1343 	req->len16 = htonl(DIV_ROUND_UP(wr_len - sizeof(req->wr), 16));
1344 
1345 	idata->cmd_more = htonl(ULPTX_CMD(ULP_TX_SC_IMM));
1346 	idata->len = htonl(dlen);
1347 }
1348 
1349 static int ddp_ppod_write_idata(struct cxgbi_device *cdev, unsigned int port_id,
1350 				struct cxgbi_pagepod_hdr *hdr, unsigned int idx,
1351 				unsigned int npods,
1352 				struct cxgbi_gather_list *gl,
1353 				unsigned int gl_pidx)
1354 {
1355 	struct cxgbi_ddp_info *ddp = cdev->ddp;
1356 	struct cxgb4_lld_info *lldi = cxgbi_cdev_priv(cdev);
1357 	struct sk_buff *skb;
1358 	struct ulp_mem_io *req;
1359 	struct ulptx_idata *idata;
1360 	struct cxgbi_pagepod *ppod;
1361 	unsigned int pm_addr = idx * PPOD_SIZE + ddp->llimit;
1362 	unsigned int dlen = PPOD_SIZE * npods;
1363 	unsigned int wr_len = roundup(sizeof(struct ulp_mem_io) +
1364 				sizeof(struct ulptx_idata) + dlen, 16);
1365 	unsigned int i;
1366 
1367 	skb = alloc_wr(wr_len, 0, GFP_ATOMIC);
1368 	if (!skb) {
1369 		pr_err("cdev 0x%p, idx %u, npods %u, OOM.\n",
1370 			cdev, idx, npods);
1371 		return -ENOMEM;
1372 	}
1373 	req = (struct ulp_mem_io *)skb->head;
1374 	set_queue(skb, CPL_PRIORITY_CONTROL, NULL);
1375 
1376 	ulp_mem_io_set_hdr(lldi, req, wr_len, dlen, pm_addr);
1377 	idata = (struct ulptx_idata *)(req + 1);
1378 	ppod = (struct cxgbi_pagepod *)(idata + 1);
1379 
1380 	for (i = 0; i < npods; i++, ppod++, gl_pidx += PPOD_PAGES_MAX) {
1381 		if (!hdr && !gl)
1382 			cxgbi_ddp_ppod_clear(ppod);
1383 		else
1384 			cxgbi_ddp_ppod_set(ppod, hdr, gl, gl_pidx);
1385 	}
1386 
1387 	cxgb4_ofld_send(cdev->ports[port_id], skb);
1388 	return 0;
1389 }
1390 
1391 static int ddp_set_map(struct cxgbi_sock *csk, struct cxgbi_pagepod_hdr *hdr,
1392 			unsigned int idx, unsigned int npods,
1393 			struct cxgbi_gather_list *gl)
1394 {
1395 	unsigned int i, cnt;
1396 	int err = 0;
1397 
1398 	for (i = 0; i < npods; i += cnt, idx += cnt) {
1399 		cnt = npods - i;
1400 		if (cnt > ULPMEM_IDATA_MAX_NPPODS)
1401 			cnt = ULPMEM_IDATA_MAX_NPPODS;
1402 		err = ddp_ppod_write_idata(csk->cdev, csk->port_id, hdr,
1403 					idx, cnt, gl, 4 * i);
1404 		if (err < 0)
1405 			break;
1406 	}
1407 	return err;
1408 }
1409 
1410 static void ddp_clear_map(struct cxgbi_hba *chba, unsigned int tag,
1411 			  unsigned int idx, unsigned int npods)
1412 {
1413 	unsigned int i, cnt;
1414 	int err;
1415 
1416 	for (i = 0; i < npods; i += cnt, idx += cnt) {
1417 		cnt = npods - i;
1418 		if (cnt > ULPMEM_IDATA_MAX_NPPODS)
1419 			cnt = ULPMEM_IDATA_MAX_NPPODS;
1420 		err = ddp_ppod_write_idata(chba->cdev, chba->port_id, NULL,
1421 					idx, cnt, NULL, 0);
1422 		if (err < 0)
1423 			break;
1424 	}
1425 }
1426 
1427 static int ddp_setup_conn_pgidx(struct cxgbi_sock *csk, unsigned int tid,
1428 				int pg_idx, bool reply)
1429 {
1430 	struct sk_buff *skb;
1431 	struct cpl_set_tcb_field *req;
1432 
1433 	if (!pg_idx || pg_idx >= DDP_PGIDX_MAX)
1434 		return 0;
1435 
1436 	skb = alloc_wr(sizeof(*req), 0, GFP_KERNEL);
1437 	if (!skb)
1438 		return -ENOMEM;
1439 
1440 	/*  set up ulp page size */
1441 	req = (struct cpl_set_tcb_field *)skb->head;
1442 	INIT_TP_WR(req, csk->tid);
1443 	OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_SET_TCB_FIELD, csk->tid));
1444 	req->reply_ctrl = htons(NO_REPLY(reply) | QUEUENO(csk->rss_qid));
1445 	req->word_cookie = htons(0);
1446 	req->mask = cpu_to_be64(0x3 << 8);
1447 	req->val = cpu_to_be64(pg_idx << 8);
1448 	set_wr_txq(skb, CPL_PRIORITY_CONTROL, csk->port_id);
1449 
1450 	log_debug(1 << CXGBI_DBG_TOE | 1 << CXGBI_DBG_SOCK,
1451 		"csk 0x%p, tid 0x%x, pg_idx %u.\n", csk, csk->tid, pg_idx);
1452 
1453 	cxgb4_ofld_send(csk->cdev->ports[csk->port_id], skb);
1454 	return 0;
1455 }
1456 
1457 static int ddp_setup_conn_digest(struct cxgbi_sock *csk, unsigned int tid,
1458 				 int hcrc, int dcrc, int reply)
1459 {
1460 	struct sk_buff *skb;
1461 	struct cpl_set_tcb_field *req;
1462 
1463 	if (!hcrc && !dcrc)
1464 		return 0;
1465 
1466 	skb = alloc_wr(sizeof(*req), 0, GFP_KERNEL);
1467 	if (!skb)
1468 		return -ENOMEM;
1469 
1470 	csk->hcrc_len = (hcrc ? 4 : 0);
1471 	csk->dcrc_len = (dcrc ? 4 : 0);
1472 	/*  set up ulp submode */
1473 	req = (struct cpl_set_tcb_field *)skb->head;
1474 	INIT_TP_WR(req, tid);
1475 	OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_SET_TCB_FIELD, tid));
1476 	req->reply_ctrl = htons(NO_REPLY(reply) | QUEUENO(csk->rss_qid));
1477 	req->word_cookie = htons(0);
1478 	req->mask = cpu_to_be64(0x3 << 4);
1479 	req->val = cpu_to_be64(((hcrc ? ULP_CRC_HEADER : 0) |
1480 				(dcrc ? ULP_CRC_DATA : 0)) << 4);
1481 	set_wr_txq(skb, CPL_PRIORITY_CONTROL, csk->port_id);
1482 
1483 	log_debug(1 << CXGBI_DBG_TOE | 1 << CXGBI_DBG_SOCK,
1484 		"csk 0x%p, tid 0x%x, crc %d,%d.\n", csk, csk->tid, hcrc, dcrc);
1485 
1486 	cxgb4_ofld_send(csk->cdev->ports[csk->port_id], skb);
1487 	return 0;
1488 }
1489 
1490 static int cxgb4i_ddp_init(struct cxgbi_device *cdev)
1491 {
1492 	struct cxgb4_lld_info *lldi = cxgbi_cdev_priv(cdev);
1493 	struct cxgbi_ddp_info *ddp = cdev->ddp;
1494 	unsigned int tagmask, pgsz_factor[4];
1495 	int err;
1496 
1497 	if (ddp) {
1498 		kref_get(&ddp->refcnt);
1499 		pr_warn("cdev 0x%p, ddp 0x%p already set up.\n",
1500 			cdev, cdev->ddp);
1501 		return -EALREADY;
1502 	}
1503 
1504 	err = cxgbi_ddp_init(cdev, lldi->vr->iscsi.start,
1505 			lldi->vr->iscsi.start + lldi->vr->iscsi.size - 1,
1506 			lldi->iscsi_iolen, lldi->iscsi_iolen);
1507 	if (err < 0)
1508 		return err;
1509 
1510 	ddp = cdev->ddp;
1511 
1512 	tagmask = ddp->idx_mask << PPOD_IDX_SHIFT;
1513 	cxgbi_ddp_page_size_factor(pgsz_factor);
1514 	cxgb4_iscsi_init(lldi->ports[0], tagmask, pgsz_factor);
1515 
1516 	cdev->csk_ddp_setup_digest = ddp_setup_conn_digest;
1517 	cdev->csk_ddp_setup_pgidx = ddp_setup_conn_pgidx;
1518 	cdev->csk_ddp_set = ddp_set_map;
1519 	cdev->csk_ddp_clear = ddp_clear_map;
1520 
1521 	pr_info("cxgb4i 0x%p tag: sw %u, rsvd %u,%u, mask 0x%x.\n",
1522 		cdev, cdev->tag_format.sw_bits, cdev->tag_format.rsvd_bits,
1523 		cdev->tag_format.rsvd_shift, cdev->tag_format.rsvd_mask);
1524 	pr_info("cxgb4i 0x%p, nppods %u, bits %u, mask 0x%x,0x%x pkt %u/%u, "
1525 		" %u/%u.\n",
1526 		cdev, ddp->nppods, ddp->idx_bits, ddp->idx_mask,
1527 		ddp->rsvd_tag_mask, ddp->max_txsz, lldi->iscsi_iolen,
1528 		ddp->max_rxsz, lldi->iscsi_iolen);
1529 	pr_info("cxgb4i 0x%p max payload size: %u/%u, %u/%u.\n",
1530 		cdev, cdev->tx_max_size, ddp->max_txsz, cdev->rx_max_size,
1531 		ddp->max_rxsz);
1532 	return 0;
1533 }
1534 
1535 static void *t4_uld_add(const struct cxgb4_lld_info *lldi)
1536 {
1537 	struct cxgbi_device *cdev;
1538 	struct port_info *pi;
1539 	int i, rc;
1540 
1541 	cdev = cxgbi_device_register(sizeof(*lldi), lldi->nports);
1542 	if (!cdev) {
1543 		pr_info("t4 device 0x%p, register failed.\n", lldi);
1544 		return NULL;
1545 	}
1546 	pr_info("0x%p,0x%x, ports %u,%s, chan %u, q %u,%u, wr %u.\n",
1547 		cdev, lldi->adapter_type, lldi->nports,
1548 		lldi->ports[0]->name, lldi->nchan, lldi->ntxq,
1549 		lldi->nrxq, lldi->wr_cred);
1550 	for (i = 0; i < lldi->nrxq; i++)
1551 		log_debug(1 << CXGBI_DBG_DEV,
1552 			"t4 0x%p, rxq id #%d: %u.\n",
1553 			cdev, i, lldi->rxq_ids[i]);
1554 
1555 	memcpy(cxgbi_cdev_priv(cdev), lldi, sizeof(*lldi));
1556 	cdev->flags = CXGBI_FLAG_DEV_T4;
1557 	cdev->pdev = lldi->pdev;
1558 	cdev->ports = lldi->ports;
1559 	cdev->nports = lldi->nports;
1560 	cdev->mtus = lldi->mtus;
1561 	cdev->nmtus = NMTUS;
1562 	cdev->snd_win = cxgb4i_snd_win;
1563 	cdev->rcv_win = cxgb4i_rcv_win;
1564 	cdev->rx_credit_thres = cxgb4i_rx_credit_thres;
1565 	cdev->skb_tx_rsvd = CXGB4I_TX_HEADER_LEN;
1566 	cdev->skb_rx_extra = sizeof(struct cpl_iscsi_hdr);
1567 	cdev->itp = &cxgb4i_iscsi_transport;
1568 
1569 	cdev->pfvf = FW_VIID_PFN_GET(cxgb4_port_viid(lldi->ports[0])) << 8;
1570 	pr_info("cdev 0x%p,%s, pfvf %u.\n",
1571 		cdev, lldi->ports[0]->name, cdev->pfvf);
1572 
1573 	rc = cxgb4i_ddp_init(cdev);
1574 	if (rc) {
1575 		pr_info("t4 0x%p ddp init failed.\n", cdev);
1576 		goto err_out;
1577 	}
1578 	rc = cxgb4i_ofld_init(cdev);
1579 	if (rc) {
1580 		pr_info("t4 0x%p ofld init failed.\n", cdev);
1581 		goto err_out;
1582 	}
1583 
1584 	rc = cxgbi_hbas_add(cdev, CXGB4I_MAX_LUN, CXGBI_MAX_CONN,
1585 				&cxgb4i_host_template, cxgb4i_stt);
1586 	if (rc)
1587 		goto err_out;
1588 
1589 	for (i = 0; i < cdev->nports; i++) {
1590 		pi = netdev_priv(lldi->ports[i]);
1591 		cdev->hbas[i]->port_id = pi->port_id;
1592 	}
1593 	return cdev;
1594 
1595 err_out:
1596 	cxgbi_device_unregister(cdev);
1597 	return ERR_PTR(-ENOMEM);
1598 }
1599 
1600 #define RX_PULL_LEN	128
1601 static int t4_uld_rx_handler(void *handle, const __be64 *rsp,
1602 				const struct pkt_gl *pgl)
1603 {
1604 	const struct cpl_act_establish *rpl;
1605 	struct sk_buff *skb;
1606 	unsigned int opc;
1607 	struct cxgbi_device *cdev = handle;
1608 
1609 	if (pgl == NULL) {
1610 		unsigned int len = 64 - sizeof(struct rsp_ctrl) - 8;
1611 
1612 		skb = alloc_wr(len, 0, GFP_ATOMIC);
1613 		if (!skb)
1614 			goto nomem;
1615 		skb_copy_to_linear_data(skb, &rsp[1], len);
1616 	} else {
1617 		if (unlikely(*(u8 *)rsp != *(u8 *)pgl->va)) {
1618 			pr_info("? FL 0x%p,RSS%#llx,FL %#llx,len %u.\n",
1619 				pgl->va, be64_to_cpu(*rsp),
1620 				be64_to_cpu(*(u64 *)pgl->va),
1621 				pgl->tot_len);
1622 			return 0;
1623 		}
1624 		skb = cxgb4_pktgl_to_skb(pgl, RX_PULL_LEN, RX_PULL_LEN);
1625 		if (unlikely(!skb))
1626 			goto nomem;
1627 	}
1628 
1629 	rpl = (struct cpl_act_establish *)skb->data;
1630 	opc = rpl->ot.opcode;
1631 	log_debug(1 << CXGBI_DBG_TOE,
1632 		"cdev %p, opcode 0x%x(0x%x,0x%x), skb %p.\n",
1633 		 cdev, opc, rpl->ot.opcode_tid, ntohl(rpl->ot.opcode_tid), skb);
1634 	if (cxgb4i_cplhandlers[opc])
1635 		cxgb4i_cplhandlers[opc](cdev, skb);
1636 	else {
1637 		pr_err("No handler for opcode 0x%x.\n", opc);
1638 		__kfree_skb(skb);
1639 	}
1640 	return 0;
1641 nomem:
1642 	log_debug(1 << CXGBI_DBG_TOE, "OOM bailing out.\n");
1643 	return 1;
1644 }
1645 
1646 static int t4_uld_state_change(void *handle, enum cxgb4_state state)
1647 {
1648 	struct cxgbi_device *cdev = handle;
1649 
1650 	switch (state) {
1651 	case CXGB4_STATE_UP:
1652 		pr_info("cdev 0x%p, UP.\n", cdev);
1653 		/* re-initialize */
1654 		break;
1655 	case CXGB4_STATE_START_RECOVERY:
1656 		pr_info("cdev 0x%p, RECOVERY.\n", cdev);
1657 		/* close all connections */
1658 		break;
1659 	case CXGB4_STATE_DOWN:
1660 		pr_info("cdev 0x%p, DOWN.\n", cdev);
1661 		break;
1662 	case CXGB4_STATE_DETACH:
1663 		pr_info("cdev 0x%p, DETACH.\n", cdev);
1664 		cxgbi_device_unregister(cdev);
1665 		break;
1666 	default:
1667 		pr_info("cdev 0x%p, unknown state %d.\n", cdev, state);
1668 		break;
1669 	}
1670 	return 0;
1671 }
1672 
1673 static int __init cxgb4i_init_module(void)
1674 {
1675 	int rc;
1676 
1677 	printk(KERN_INFO "%s", version);
1678 
1679 	rc = cxgbi_iscsi_init(&cxgb4i_iscsi_transport, &cxgb4i_stt);
1680 	if (rc < 0)
1681 		return rc;
1682 	cxgb4_register_uld(CXGB4_ULD_ISCSI, &cxgb4i_uld_info);
1683 	return 0;
1684 }
1685 
1686 static void __exit cxgb4i_exit_module(void)
1687 {
1688 	cxgb4_unregister_uld(CXGB4_ULD_ISCSI);
1689 	cxgbi_device_unregister_all(CXGBI_FLAG_DEV_T4);
1690 	cxgbi_iscsi_cleanup(&cxgb4i_iscsi_transport, &cxgb4i_stt);
1691 }
1692 
1693 module_init(cxgb4i_init_module);
1694 module_exit(cxgb4i_exit_module);
1695