xref: /linux/drivers/scsi/cxgbi/cxgb4i/cxgb4i.c (revision 4fd18fc38757217c746aa063ba9e4729814dc737)
1 /*
2  * cxgb4i.c: Chelsio T4 iSCSI driver.
3  *
4  * Copyright (c) 2010-2015 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/kernel.h>
17 #include <linux/module.h>
18 #include <linux/moduleparam.h>
19 #include <scsi/scsi_host.h>
20 #include <net/tcp.h>
21 #include <net/dst.h>
22 #include <linux/netdevice.h>
23 #include <net/addrconf.h>
24 
25 #include "t4_regs.h"
26 #include "t4_msg.h"
27 #include "cxgb4.h"
28 #include "cxgb4_uld.h"
29 #include "t4fw_api.h"
30 #include "l2t.h"
31 #include "cxgb4i.h"
32 #include "clip_tbl.h"
33 
34 static unsigned int dbg_level;
35 
36 #include "../libcxgbi.h"
37 
38 #ifdef CONFIG_CHELSIO_T4_DCB
39 #include <net/dcbevent.h>
40 #include "cxgb4_dcb.h"
41 #endif
42 
43 #define	DRV_MODULE_NAME		"cxgb4i"
44 #define DRV_MODULE_DESC		"Chelsio T4-T6 iSCSI Driver"
45 #define	DRV_MODULE_VERSION	"0.9.5-ko"
46 #define DRV_MODULE_RELDATE	"Apr. 2015"
47 
48 static char version[] =
49 	DRV_MODULE_DESC " " DRV_MODULE_NAME
50 	" v" DRV_MODULE_VERSION " (" DRV_MODULE_RELDATE ")\n";
51 
52 MODULE_AUTHOR("Chelsio Communications, Inc.");
53 MODULE_DESCRIPTION(DRV_MODULE_DESC);
54 MODULE_VERSION(DRV_MODULE_VERSION);
55 MODULE_LICENSE("GPL");
56 
57 module_param(dbg_level, uint, 0644);
58 MODULE_PARM_DESC(dbg_level, "Debug flag (default=0)");
59 
60 #define CXGB4I_DEFAULT_10G_RCV_WIN (256 * 1024)
61 static int cxgb4i_rcv_win = -1;
62 module_param(cxgb4i_rcv_win, int, 0644);
63 MODULE_PARM_DESC(cxgb4i_rcv_win, "TCP receive window in bytes");
64 
65 #define CXGB4I_DEFAULT_10G_SND_WIN (128 * 1024)
66 static int cxgb4i_snd_win = -1;
67 module_param(cxgb4i_snd_win, int, 0644);
68 MODULE_PARM_DESC(cxgb4i_snd_win, "TCP send window in bytes");
69 
70 static int cxgb4i_rx_credit_thres = 10 * 1024;
71 module_param(cxgb4i_rx_credit_thres, int, 0644);
72 MODULE_PARM_DESC(cxgb4i_rx_credit_thres,
73 		"RX credits return threshold in bytes (default=10KB)");
74 
75 static unsigned int cxgb4i_max_connect = (8 * 1024);
76 module_param(cxgb4i_max_connect, uint, 0644);
77 MODULE_PARM_DESC(cxgb4i_max_connect, "Maximum number of connections");
78 
79 static unsigned short cxgb4i_sport_base = 20000;
80 module_param(cxgb4i_sport_base, ushort, 0644);
81 MODULE_PARM_DESC(cxgb4i_sport_base, "Starting port number (default 20000)");
82 
83 typedef void (*cxgb4i_cplhandler_func)(struct cxgbi_device *, struct sk_buff *);
84 
85 static void *t4_uld_add(const struct cxgb4_lld_info *);
86 static int t4_uld_rx_handler(void *, const __be64 *, const struct pkt_gl *);
87 static int t4_uld_state_change(void *, enum cxgb4_state state);
88 static inline int send_tx_flowc_wr(struct cxgbi_sock *);
89 
90 static const struct cxgb4_uld_info cxgb4i_uld_info = {
91 	.name = DRV_MODULE_NAME,
92 	.nrxq = MAX_ULD_QSETS,
93 	.ntxq = MAX_ULD_QSETS,
94 	.rxq_size = 1024,
95 	.lro = false,
96 	.add = t4_uld_add,
97 	.rx_handler = t4_uld_rx_handler,
98 	.state_change = t4_uld_state_change,
99 };
100 
101 static struct scsi_host_template cxgb4i_host_template = {
102 	.module		= THIS_MODULE,
103 	.name		= DRV_MODULE_NAME,
104 	.proc_name	= DRV_MODULE_NAME,
105 	.can_queue	= CXGB4I_SCSI_HOST_QDEPTH,
106 	.queuecommand	= iscsi_queuecommand,
107 	.change_queue_depth = scsi_change_queue_depth,
108 	.sg_tablesize	= SG_ALL,
109 	.max_sectors	= 0xFFFF,
110 	.cmd_per_lun	= ISCSI_DEF_CMD_PER_LUN,
111 	.eh_timed_out	= iscsi_eh_cmd_timed_out,
112 	.eh_abort_handler = iscsi_eh_abort,
113 	.eh_device_reset_handler = iscsi_eh_device_reset,
114 	.eh_target_reset_handler = iscsi_eh_recover_target,
115 	.target_alloc	= iscsi_target_alloc,
116 	.dma_boundary	= PAGE_SIZE - 1,
117 	.this_id	= -1,
118 	.track_queue_depth = 1,
119 };
120 
121 static struct iscsi_transport cxgb4i_iscsi_transport = {
122 	.owner		= THIS_MODULE,
123 	.name		= DRV_MODULE_NAME,
124 	.caps		= CAP_RECOVERY_L0 | CAP_MULTI_R2T | CAP_HDRDGST |
125 				CAP_DATADGST | CAP_DIGEST_OFFLOAD |
126 				CAP_PADDING_OFFLOAD | CAP_TEXT_NEGO,
127 	.attr_is_visible	= cxgbi_attr_is_visible,
128 	.get_host_param	= cxgbi_get_host_param,
129 	.set_host_param	= cxgbi_set_host_param,
130 	/* session management */
131 	.create_session	= cxgbi_create_session,
132 	.destroy_session	= cxgbi_destroy_session,
133 	.get_session_param = iscsi_session_get_param,
134 	/* connection management */
135 	.create_conn	= cxgbi_create_conn,
136 	.bind_conn		= cxgbi_bind_conn,
137 	.destroy_conn	= iscsi_tcp_conn_teardown,
138 	.start_conn		= iscsi_conn_start,
139 	.stop_conn		= iscsi_conn_stop,
140 	.get_conn_param	= iscsi_conn_get_param,
141 	.set_param	= cxgbi_set_conn_param,
142 	.get_stats	= cxgbi_get_conn_stats,
143 	/* pdu xmit req from user space */
144 	.send_pdu	= iscsi_conn_send_pdu,
145 	/* task */
146 	.init_task	= iscsi_tcp_task_init,
147 	.xmit_task	= iscsi_tcp_task_xmit,
148 	.cleanup_task	= cxgbi_cleanup_task,
149 	/* pdu */
150 	.alloc_pdu	= cxgbi_conn_alloc_pdu,
151 	.init_pdu	= cxgbi_conn_init_pdu,
152 	.xmit_pdu	= cxgbi_conn_xmit_pdu,
153 	.parse_pdu_itt	= cxgbi_parse_pdu_itt,
154 	/* TCP connect/disconnect */
155 	.get_ep_param	= cxgbi_get_ep_param,
156 	.ep_connect	= cxgbi_ep_connect,
157 	.ep_poll	= cxgbi_ep_poll,
158 	.ep_disconnect	= cxgbi_ep_disconnect,
159 	/* Error recovery timeout call */
160 	.session_recovery_timedout = iscsi_session_recovery_timedout,
161 };
162 
163 #ifdef CONFIG_CHELSIO_T4_DCB
164 static int
165 cxgb4_dcb_change_notify(struct notifier_block *, unsigned long, void *);
166 
167 static struct notifier_block cxgb4_dcb_change = {
168 	.notifier_call = cxgb4_dcb_change_notify,
169 };
170 #endif
171 
172 static struct scsi_transport_template *cxgb4i_stt;
173 
174 /*
175  * CPL (Chelsio Protocol Language) defines a message passing interface between
176  * the host driver and Chelsio asic.
177  * The section below implments CPLs that related to iscsi tcp connection
178  * open/close/abort and data send/receive.
179  */
180 
181 #define RCV_BUFSIZ_MASK		0x3FFU
182 #define MAX_IMM_TX_PKT_LEN	256
183 
184 static int push_tx_frames(struct cxgbi_sock *, int);
185 
186 /*
187  * is_ofld_imm - check whether a packet can be sent as immediate data
188  * @skb: the packet
189  *
190  * Returns true if a packet can be sent as an offload WR with immediate
191  * data.  We currently use the same limit as for Ethernet packets.
192  */
193 static inline bool is_ofld_imm(const struct sk_buff *skb)
194 {
195 	int len = skb->len;
196 
197 	if (likely(cxgbi_skcb_test_flag(skb, SKCBF_TX_NEED_HDR)))
198 		len += sizeof(struct fw_ofld_tx_data_wr);
199 
200 	if  (likely(cxgbi_skcb_test_flag((struct sk_buff *)skb, SKCBF_TX_ISO)))
201 		len += sizeof(struct cpl_tx_data_iso);
202 
203 	return (len <= MAX_IMM_OFLD_TX_DATA_WR_LEN);
204 }
205 
206 static void send_act_open_req(struct cxgbi_sock *csk, struct sk_buff *skb,
207 				struct l2t_entry *e)
208 {
209 	struct cxgb4_lld_info *lldi = cxgbi_cdev_priv(csk->cdev);
210 	int wscale = cxgbi_sock_compute_wscale(csk->mss_idx);
211 	unsigned long long opt0;
212 	unsigned int opt2;
213 	unsigned int qid_atid = ((unsigned int)csk->atid) |
214 				 (((unsigned int)csk->rss_qid) << 14);
215 
216 	opt0 = KEEP_ALIVE_F |
217 		WND_SCALE_V(wscale) |
218 		MSS_IDX_V(csk->mss_idx) |
219 		L2T_IDX_V(((struct l2t_entry *)csk->l2t)->idx) |
220 		TX_CHAN_V(csk->tx_chan) |
221 		SMAC_SEL_V(csk->smac_idx) |
222 		ULP_MODE_V(ULP_MODE_ISCSI) |
223 		RCV_BUFSIZ_V(csk->rcv_win >> 10);
224 
225 	opt2 = RX_CHANNEL_V(0) |
226 		RSS_QUEUE_VALID_F |
227 		RSS_QUEUE_V(csk->rss_qid);
228 
229 	if (is_t4(lldi->adapter_type)) {
230 		struct cpl_act_open_req *req =
231 				(struct cpl_act_open_req *)skb->head;
232 
233 		INIT_TP_WR(req, 0);
234 		OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_ACT_OPEN_REQ,
235 					qid_atid));
236 		req->local_port = csk->saddr.sin_port;
237 		req->peer_port = csk->daddr.sin_port;
238 		req->local_ip = csk->saddr.sin_addr.s_addr;
239 		req->peer_ip = csk->daddr.sin_addr.s_addr;
240 		req->opt0 = cpu_to_be64(opt0);
241 		req->params = cpu_to_be32(cxgb4_select_ntuple(
242 					csk->cdev->ports[csk->port_id],
243 					csk->l2t));
244 		opt2 |= RX_FC_VALID_F;
245 		req->opt2 = cpu_to_be32(opt2);
246 
247 		log_debug(1 << CXGBI_DBG_TOE | 1 << CXGBI_DBG_SOCK,
248 			"csk t4 0x%p, %pI4:%u-%pI4:%u, atid %d, qid %u.\n",
249 			csk, &req->local_ip, ntohs(req->local_port),
250 			&req->peer_ip, ntohs(req->peer_port),
251 			csk->atid, csk->rss_qid);
252 	} else if (is_t5(lldi->adapter_type)) {
253 		struct cpl_t5_act_open_req *req =
254 				(struct cpl_t5_act_open_req *)skb->head;
255 		u32 isn = (prandom_u32() & ~7UL) - 1;
256 
257 		INIT_TP_WR(req, 0);
258 		OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_ACT_OPEN_REQ,
259 					qid_atid));
260 		req->local_port = csk->saddr.sin_port;
261 		req->peer_port = csk->daddr.sin_port;
262 		req->local_ip = csk->saddr.sin_addr.s_addr;
263 		req->peer_ip = csk->daddr.sin_addr.s_addr;
264 		req->opt0 = cpu_to_be64(opt0);
265 		req->params = cpu_to_be64(FILTER_TUPLE_V(
266 				cxgb4_select_ntuple(
267 					csk->cdev->ports[csk->port_id],
268 					csk->l2t)));
269 		req->rsvd = cpu_to_be32(isn);
270 		opt2 |= T5_ISS_VALID;
271 		opt2 |= T5_OPT_2_VALID_F;
272 
273 		req->opt2 = cpu_to_be32(opt2);
274 
275 		log_debug(1 << CXGBI_DBG_TOE | 1 << CXGBI_DBG_SOCK,
276 			"csk t5 0x%p, %pI4:%u-%pI4:%u, atid %d, qid %u.\n",
277 			csk, &req->local_ip, ntohs(req->local_port),
278 			&req->peer_ip, ntohs(req->peer_port),
279 			csk->atid, csk->rss_qid);
280 	} else {
281 		struct cpl_t6_act_open_req *req =
282 				(struct cpl_t6_act_open_req *)skb->head;
283 		u32 isn = (prandom_u32() & ~7UL) - 1;
284 
285 		INIT_TP_WR(req, 0);
286 		OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_ACT_OPEN_REQ,
287 							    qid_atid));
288 		req->local_port = csk->saddr.sin_port;
289 		req->peer_port = csk->daddr.sin_port;
290 		req->local_ip = csk->saddr.sin_addr.s_addr;
291 		req->peer_ip = csk->daddr.sin_addr.s_addr;
292 		req->opt0 = cpu_to_be64(opt0);
293 		req->params = cpu_to_be64(FILTER_TUPLE_V(
294 				cxgb4_select_ntuple(
295 					csk->cdev->ports[csk->port_id],
296 					csk->l2t)));
297 		req->rsvd = cpu_to_be32(isn);
298 
299 		opt2 |= T5_ISS_VALID;
300 		opt2 |= RX_FC_DISABLE_F;
301 		opt2 |= T5_OPT_2_VALID_F;
302 
303 		req->opt2 = cpu_to_be32(opt2);
304 		req->rsvd2 = cpu_to_be32(0);
305 		req->opt3 = cpu_to_be32(0);
306 
307 		log_debug(1 << CXGBI_DBG_TOE | 1 << CXGBI_DBG_SOCK,
308 			  "csk t6 0x%p, %pI4:%u-%pI4:%u, atid %d, qid %u.\n",
309 			  csk, &req->local_ip, ntohs(req->local_port),
310 			  &req->peer_ip, ntohs(req->peer_port),
311 			  csk->atid, csk->rss_qid);
312 	}
313 
314 	set_wr_txq(skb, CPL_PRIORITY_SETUP, csk->port_id);
315 
316 	pr_info_ipaddr("t%d csk 0x%p,%u,0x%lx,%u, rss_qid %u.\n",
317 		       (&csk->saddr), (&csk->daddr),
318 		       CHELSIO_CHIP_VERSION(lldi->adapter_type), csk,
319 		       csk->state, csk->flags, csk->atid, csk->rss_qid);
320 
321 	cxgb4_l2t_send(csk->cdev->ports[csk->port_id], skb, csk->l2t);
322 }
323 
324 #if IS_ENABLED(CONFIG_IPV6)
325 static void send_act_open_req6(struct cxgbi_sock *csk, struct sk_buff *skb,
326 			       struct l2t_entry *e)
327 {
328 	struct cxgb4_lld_info *lldi = cxgbi_cdev_priv(csk->cdev);
329 	int wscale = cxgbi_sock_compute_wscale(csk->mss_idx);
330 	unsigned long long opt0;
331 	unsigned int opt2;
332 	unsigned int qid_atid = ((unsigned int)csk->atid) |
333 				 (((unsigned int)csk->rss_qid) << 14);
334 
335 	opt0 = KEEP_ALIVE_F |
336 		WND_SCALE_V(wscale) |
337 		MSS_IDX_V(csk->mss_idx) |
338 		L2T_IDX_V(((struct l2t_entry *)csk->l2t)->idx) |
339 		TX_CHAN_V(csk->tx_chan) |
340 		SMAC_SEL_V(csk->smac_idx) |
341 		ULP_MODE_V(ULP_MODE_ISCSI) |
342 		RCV_BUFSIZ_V(csk->rcv_win >> 10);
343 
344 	opt2 = RX_CHANNEL_V(0) |
345 		RSS_QUEUE_VALID_F |
346 		RSS_QUEUE_V(csk->rss_qid);
347 
348 	if (is_t4(lldi->adapter_type)) {
349 		struct cpl_act_open_req6 *req =
350 			    (struct cpl_act_open_req6 *)skb->head;
351 
352 		INIT_TP_WR(req, 0);
353 		OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_ACT_OPEN_REQ6,
354 							    qid_atid));
355 		req->local_port = csk->saddr6.sin6_port;
356 		req->peer_port = csk->daddr6.sin6_port;
357 
358 		req->local_ip_hi = *(__be64 *)(csk->saddr6.sin6_addr.s6_addr);
359 		req->local_ip_lo = *(__be64 *)(csk->saddr6.sin6_addr.s6_addr +
360 								    8);
361 		req->peer_ip_hi = *(__be64 *)(csk->daddr6.sin6_addr.s6_addr);
362 		req->peer_ip_lo = *(__be64 *)(csk->daddr6.sin6_addr.s6_addr +
363 								    8);
364 
365 		req->opt0 = cpu_to_be64(opt0);
366 
367 		opt2 |= RX_FC_VALID_F;
368 		req->opt2 = cpu_to_be32(opt2);
369 
370 		req->params = cpu_to_be32(cxgb4_select_ntuple(
371 					  csk->cdev->ports[csk->port_id],
372 					  csk->l2t));
373 	} else if (is_t5(lldi->adapter_type)) {
374 		struct cpl_t5_act_open_req6 *req =
375 				(struct cpl_t5_act_open_req6 *)skb->head;
376 
377 		INIT_TP_WR(req, 0);
378 		OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_ACT_OPEN_REQ6,
379 							    qid_atid));
380 		req->local_port = csk->saddr6.sin6_port;
381 		req->peer_port = csk->daddr6.sin6_port;
382 		req->local_ip_hi = *(__be64 *)(csk->saddr6.sin6_addr.s6_addr);
383 		req->local_ip_lo = *(__be64 *)(csk->saddr6.sin6_addr.s6_addr +
384 									8);
385 		req->peer_ip_hi = *(__be64 *)(csk->daddr6.sin6_addr.s6_addr);
386 		req->peer_ip_lo = *(__be64 *)(csk->daddr6.sin6_addr.s6_addr +
387 									8);
388 		req->opt0 = cpu_to_be64(opt0);
389 
390 		opt2 |= T5_OPT_2_VALID_F;
391 		req->opt2 = cpu_to_be32(opt2);
392 
393 		req->params = cpu_to_be64(FILTER_TUPLE_V(cxgb4_select_ntuple(
394 					  csk->cdev->ports[csk->port_id],
395 					  csk->l2t)));
396 	} else {
397 		struct cpl_t6_act_open_req6 *req =
398 				(struct cpl_t6_act_open_req6 *)skb->head;
399 
400 		INIT_TP_WR(req, 0);
401 		OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_ACT_OPEN_REQ6,
402 							    qid_atid));
403 		req->local_port = csk->saddr6.sin6_port;
404 		req->peer_port = csk->daddr6.sin6_port;
405 		req->local_ip_hi = *(__be64 *)(csk->saddr6.sin6_addr.s6_addr);
406 		req->local_ip_lo = *(__be64 *)(csk->saddr6.sin6_addr.s6_addr +
407 									8);
408 		req->peer_ip_hi = *(__be64 *)(csk->daddr6.sin6_addr.s6_addr);
409 		req->peer_ip_lo = *(__be64 *)(csk->daddr6.sin6_addr.s6_addr +
410 									8);
411 		req->opt0 = cpu_to_be64(opt0);
412 
413 		opt2 |= RX_FC_DISABLE_F;
414 		opt2 |= T5_OPT_2_VALID_F;
415 
416 		req->opt2 = cpu_to_be32(opt2);
417 
418 		req->params = cpu_to_be64(FILTER_TUPLE_V(cxgb4_select_ntuple(
419 					  csk->cdev->ports[csk->port_id],
420 					  csk->l2t)));
421 
422 		req->rsvd2 = cpu_to_be32(0);
423 		req->opt3 = cpu_to_be32(0);
424 	}
425 
426 	set_wr_txq(skb, CPL_PRIORITY_SETUP, csk->port_id);
427 
428 	pr_info("t%d csk 0x%p,%u,0x%lx,%u, [%pI6]:%u-[%pI6]:%u, rss_qid %u.\n",
429 		CHELSIO_CHIP_VERSION(lldi->adapter_type), csk, csk->state,
430 		csk->flags, csk->atid,
431 		&csk->saddr6.sin6_addr, ntohs(csk->saddr.sin_port),
432 		&csk->daddr6.sin6_addr, ntohs(csk->daddr.sin_port),
433 		csk->rss_qid);
434 
435 	cxgb4_l2t_send(csk->cdev->ports[csk->port_id], skb, csk->l2t);
436 }
437 #endif
438 
439 static void send_close_req(struct cxgbi_sock *csk)
440 {
441 	struct sk_buff *skb = csk->cpl_close;
442 	struct cpl_close_con_req *req = (struct cpl_close_con_req *)skb->head;
443 	unsigned int tid = csk->tid;
444 
445 	log_debug(1 << CXGBI_DBG_TOE | 1 << CXGBI_DBG_SOCK,
446 		"csk 0x%p,%u,0x%lx, tid %u.\n",
447 		csk, csk->state, csk->flags, csk->tid);
448 	csk->cpl_close = NULL;
449 	set_wr_txq(skb, CPL_PRIORITY_DATA, csk->port_id);
450 	INIT_TP_WR(req, tid);
451 	OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_CLOSE_CON_REQ, tid));
452 	req->rsvd = 0;
453 
454 	cxgbi_sock_skb_entail(csk, skb);
455 	if (csk->state >= CTP_ESTABLISHED)
456 		push_tx_frames(csk, 1);
457 }
458 
459 static void abort_arp_failure(void *handle, struct sk_buff *skb)
460 {
461 	struct cxgbi_sock *csk = (struct cxgbi_sock *)handle;
462 	struct cpl_abort_req *req;
463 
464 	log_debug(1 << CXGBI_DBG_TOE | 1 << CXGBI_DBG_SOCK,
465 		"csk 0x%p,%u,0x%lx, tid %u, abort.\n",
466 		csk, csk->state, csk->flags, csk->tid);
467 	req = (struct cpl_abort_req *)skb->data;
468 	req->cmd = CPL_ABORT_NO_RST;
469 	cxgb4_ofld_send(csk->cdev->ports[csk->port_id], skb);
470 }
471 
472 static void send_abort_req(struct cxgbi_sock *csk)
473 {
474 	struct cpl_abort_req *req;
475 	struct sk_buff *skb = csk->cpl_abort_req;
476 
477 	if (unlikely(csk->state == CTP_ABORTING) || !skb || !csk->cdev)
478 		return;
479 
480 	if (!cxgbi_sock_flag(csk, CTPF_TX_DATA_SENT)) {
481 		send_tx_flowc_wr(csk);
482 		cxgbi_sock_set_flag(csk, CTPF_TX_DATA_SENT);
483 	}
484 
485 	cxgbi_sock_set_state(csk, CTP_ABORTING);
486 	cxgbi_sock_set_flag(csk, CTPF_ABORT_RPL_PENDING);
487 	cxgbi_sock_purge_write_queue(csk);
488 
489 	csk->cpl_abort_req = NULL;
490 	req = (struct cpl_abort_req *)skb->head;
491 	set_wr_txq(skb, CPL_PRIORITY_DATA, csk->port_id);
492 	req->cmd = CPL_ABORT_SEND_RST;
493 	t4_set_arp_err_handler(skb, csk, abort_arp_failure);
494 	INIT_TP_WR(req, csk->tid);
495 	OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_ABORT_REQ, csk->tid));
496 	req->rsvd0 = htonl(csk->snd_nxt);
497 	req->rsvd1 = !cxgbi_sock_flag(csk, CTPF_TX_DATA_SENT);
498 
499 	log_debug(1 << CXGBI_DBG_TOE | 1 << CXGBI_DBG_SOCK,
500 		"csk 0x%p,%u,0x%lx,%u, snd_nxt %u, 0x%x.\n",
501 		csk, csk->state, csk->flags, csk->tid, csk->snd_nxt,
502 		req->rsvd1);
503 
504 	cxgb4_l2t_send(csk->cdev->ports[csk->port_id], skb, csk->l2t);
505 }
506 
507 static void send_abort_rpl(struct cxgbi_sock *csk, int rst_status)
508 {
509 	struct sk_buff *skb = csk->cpl_abort_rpl;
510 	struct cpl_abort_rpl *rpl = (struct cpl_abort_rpl *)skb->head;
511 
512 	log_debug(1 << CXGBI_DBG_TOE | 1 << CXGBI_DBG_SOCK,
513 		"csk 0x%p,%u,0x%lx,%u, status %d.\n",
514 		csk, csk->state, csk->flags, csk->tid, rst_status);
515 
516 	csk->cpl_abort_rpl = NULL;
517 	set_wr_txq(skb, CPL_PRIORITY_DATA, csk->port_id);
518 	INIT_TP_WR(rpl, csk->tid);
519 	OPCODE_TID(rpl) = cpu_to_be32(MK_OPCODE_TID(CPL_ABORT_RPL, csk->tid));
520 	rpl->cmd = rst_status;
521 	cxgb4_ofld_send(csk->cdev->ports[csk->port_id], skb);
522 }
523 
524 /*
525  * CPL connection rx data ack: host ->
526  * Send RX credits through an RX_DATA_ACK CPL message. Returns the number of
527  * credits sent.
528  */
529 static u32 send_rx_credits(struct cxgbi_sock *csk, u32 credits)
530 {
531 	struct sk_buff *skb;
532 	struct cpl_rx_data_ack *req;
533 
534 	log_debug(1 << CXGBI_DBG_TOE | 1 << CXGBI_DBG_PDU_RX,
535 		"csk 0x%p,%u,0x%lx,%u, credit %u.\n",
536 		csk, csk->state, csk->flags, csk->tid, credits);
537 
538 	skb = alloc_wr(sizeof(*req), 0, GFP_ATOMIC);
539 	if (!skb) {
540 		pr_info("csk 0x%p, credit %u, OOM.\n", csk, credits);
541 		return 0;
542 	}
543 	req = (struct cpl_rx_data_ack *)skb->head;
544 
545 	set_wr_txq(skb, CPL_PRIORITY_ACK, csk->port_id);
546 	INIT_TP_WR(req, csk->tid);
547 	OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_RX_DATA_ACK,
548 				      csk->tid));
549 	req->credit_dack = cpu_to_be32(RX_CREDITS_V(credits)
550 				       | RX_FORCE_ACK_F);
551 	cxgb4_ofld_send(csk->cdev->ports[csk->port_id], skb);
552 	return credits;
553 }
554 
555 /*
556  * sgl_len - calculates the size of an SGL of the given capacity
557  * @n: the number of SGL entries
558  * Calculates the number of flits needed for a scatter/gather list that
559  * can hold the given number of entries.
560  */
561 static inline unsigned int sgl_len(unsigned int n)
562 {
563 	n--;
564 	return (3 * n) / 2 + (n & 1) + 2;
565 }
566 
567 /*
568  * calc_tx_flits_ofld - calculate # of flits for an offload packet
569  * @skb: the packet
570  *
571  * Returns the number of flits needed for the given offload packet.
572  * These packets are already fully constructed and no additional headers
573  * will be added.
574  */
575 static inline unsigned int calc_tx_flits_ofld(const struct sk_buff *skb)
576 {
577 	unsigned int flits, cnt;
578 
579 	if (is_ofld_imm(skb))
580 		return DIV_ROUND_UP(skb->len, 8);
581 	flits = skb_transport_offset(skb) / 8;
582 	cnt = skb_shinfo(skb)->nr_frags;
583 	if (skb_tail_pointer(skb) != skb_transport_header(skb))
584 		cnt++;
585 	return flits + sgl_len(cnt);
586 }
587 
588 #define FLOWC_WR_NPARAMS_MIN	9
589 static inline int tx_flowc_wr_credits(int *nparamsp, int *flowclenp)
590 {
591 	int nparams, flowclen16, flowclen;
592 
593 	nparams = FLOWC_WR_NPARAMS_MIN;
594 #ifdef CONFIG_CHELSIO_T4_DCB
595 	nparams++;
596 #endif
597 	flowclen = offsetof(struct fw_flowc_wr, mnemval[nparams]);
598 	flowclen16 = DIV_ROUND_UP(flowclen, 16);
599 	flowclen = flowclen16 * 16;
600 	/*
601 	 * Return the number of 16-byte credits used by the FlowC request.
602 	 * Pass back the nparams and actual FlowC length if requested.
603 	 */
604 	if (nparamsp)
605 		*nparamsp = nparams;
606 	if (flowclenp)
607 		*flowclenp = flowclen;
608 
609 	return flowclen16;
610 }
611 
612 static inline int send_tx_flowc_wr(struct cxgbi_sock *csk)
613 {
614 	struct sk_buff *skb;
615 	struct fw_flowc_wr *flowc;
616 	int nparams, flowclen16, flowclen;
617 
618 #ifdef CONFIG_CHELSIO_T4_DCB
619 	u16 vlan = ((struct l2t_entry *)csk->l2t)->vlan;
620 #endif
621 	flowclen16 = tx_flowc_wr_credits(&nparams, &flowclen);
622 	skb = alloc_wr(flowclen, 0, GFP_ATOMIC);
623 	flowc = (struct fw_flowc_wr *)skb->head;
624 	flowc->op_to_nparams =
625 		htonl(FW_WR_OP_V(FW_FLOWC_WR) | FW_FLOWC_WR_NPARAMS_V(nparams));
626 	flowc->flowid_len16 =
627 		htonl(FW_WR_LEN16_V(flowclen16) | FW_WR_FLOWID_V(csk->tid));
628 	flowc->mnemval[0].mnemonic = FW_FLOWC_MNEM_PFNVFN;
629 	flowc->mnemval[0].val = htonl(csk->cdev->pfvf);
630 	flowc->mnemval[1].mnemonic = FW_FLOWC_MNEM_CH;
631 	flowc->mnemval[1].val = htonl(csk->tx_chan);
632 	flowc->mnemval[2].mnemonic = FW_FLOWC_MNEM_PORT;
633 	flowc->mnemval[2].val = htonl(csk->tx_chan);
634 	flowc->mnemval[3].mnemonic = FW_FLOWC_MNEM_IQID;
635 	flowc->mnemval[3].val = htonl(csk->rss_qid);
636 	flowc->mnemval[4].mnemonic = FW_FLOWC_MNEM_SNDNXT;
637 	flowc->mnemval[4].val = htonl(csk->snd_nxt);
638 	flowc->mnemval[5].mnemonic = FW_FLOWC_MNEM_RCVNXT;
639 	flowc->mnemval[5].val = htonl(csk->rcv_nxt);
640 	flowc->mnemval[6].mnemonic = FW_FLOWC_MNEM_SNDBUF;
641 	flowc->mnemval[6].val = htonl(csk->snd_win);
642 	flowc->mnemval[7].mnemonic = FW_FLOWC_MNEM_MSS;
643 	flowc->mnemval[7].val = htonl(csk->advmss);
644 	flowc->mnemval[8].mnemonic = 0;
645 	flowc->mnemval[8].val = 0;
646 	flowc->mnemval[8].mnemonic = FW_FLOWC_MNEM_TXDATAPLEN_MAX;
647 	if (csk->cdev->skb_iso_txhdr)
648 		flowc->mnemval[8].val = cpu_to_be32(CXGBI_MAX_ISO_DATA_IN_SKB);
649 	else
650 		flowc->mnemval[8].val = cpu_to_be32(16128);
651 #ifdef CONFIG_CHELSIO_T4_DCB
652 	flowc->mnemval[9].mnemonic = FW_FLOWC_MNEM_DCBPRIO;
653 	if (vlan == CPL_L2T_VLAN_NONE) {
654 		pr_warn_ratelimited("csk %u without VLAN Tag on DCB Link\n",
655 				    csk->tid);
656 		flowc->mnemval[9].val = cpu_to_be32(0);
657 	} else {
658 		flowc->mnemval[9].val = cpu_to_be32((vlan & VLAN_PRIO_MASK) >>
659 					VLAN_PRIO_SHIFT);
660 	}
661 #endif
662 
663 	set_wr_txq(skb, CPL_PRIORITY_DATA, csk->port_id);
664 
665 	log_debug(1 << CXGBI_DBG_TOE | 1 << CXGBI_DBG_SOCK,
666 		"csk 0x%p, tid 0x%x, %u,%u,%u,%u,%u,%u,%u.\n",
667 		csk, csk->tid, 0, csk->tx_chan, csk->rss_qid,
668 		csk->snd_nxt, csk->rcv_nxt, csk->snd_win,
669 		csk->advmss);
670 
671 	cxgb4_ofld_send(csk->cdev->ports[csk->port_id], skb);
672 
673 	return flowclen16;
674 }
675 
676 static void
677 cxgb4i_make_tx_iso_cpl(struct sk_buff *skb, struct cpl_tx_data_iso *cpl)
678 {
679 	struct cxgbi_iso_info *info = (struct cxgbi_iso_info *)skb->head;
680 	u32 imm_en = !!(info->flags & CXGBI_ISO_INFO_IMM_ENABLE);
681 	u32 fslice = !!(info->flags & CXGBI_ISO_INFO_FSLICE);
682 	u32 lslice = !!(info->flags & CXGBI_ISO_INFO_LSLICE);
683 	u32 pdu_type = (info->op == ISCSI_OP_SCSI_CMD) ? 0 : 1;
684 	u32 submode = cxgbi_skcb_tx_ulp_mode(skb) & 0x3;
685 
686 	cpl->op_to_scsi = cpu_to_be32(CPL_TX_DATA_ISO_OP_V(CPL_TX_DATA_ISO) |
687 				CPL_TX_DATA_ISO_FIRST_V(fslice) |
688 				CPL_TX_DATA_ISO_LAST_V(lslice) |
689 				CPL_TX_DATA_ISO_CPLHDRLEN_V(0) |
690 				CPL_TX_DATA_ISO_HDRCRC_V(submode & 1) |
691 				CPL_TX_DATA_ISO_PLDCRC_V(((submode >> 1) & 1)) |
692 				CPL_TX_DATA_ISO_IMMEDIATE_V(imm_en) |
693 				CPL_TX_DATA_ISO_SCSI_V(pdu_type));
694 
695 	cpl->ahs_len = info->ahs;
696 	cpl->mpdu = cpu_to_be16(DIV_ROUND_UP(info->mpdu, 4));
697 	cpl->burst_size = cpu_to_be32(info->burst_size);
698 	cpl->len = cpu_to_be32(info->len);
699 	cpl->reserved2_seglen_offset =
700 	     cpu_to_be32(CPL_TX_DATA_ISO_SEGLEN_OFFSET_V(info->segment_offset));
701 	cpl->datasn_offset = cpu_to_be32(info->datasn_offset);
702 	cpl->buffer_offset = cpu_to_be32(info->buffer_offset);
703 	cpl->reserved3 = cpu_to_be32(0);
704 	log_debug(1 << CXGBI_DBG_ISCSI | 1 << CXGBI_DBG_PDU_TX,
705 		  "iso: flags 0x%x, op %u, ahs %u, num_pdu %u, mpdu %u, "
706 		  "burst_size %u, iso_len %u\n",
707 		  info->flags, info->op, info->ahs, info->num_pdu,
708 		  info->mpdu, info->burst_size << 2, info->len);
709 }
710 
711 static void
712 cxgb4i_make_tx_data_wr(struct cxgbi_sock *csk, struct sk_buff *skb, int dlen,
713 		       int len, u32 credits, int compl)
714 {
715 	struct cxgbi_device *cdev = csk->cdev;
716 	struct cxgb4_lld_info *lldi = cxgbi_cdev_priv(cdev);
717 	struct fw_ofld_tx_data_wr *req;
718 	struct cpl_tx_data_iso *cpl;
719 	u32 submode = cxgbi_skcb_tx_ulp_mode(skb) & 0x3;
720 	u32 wr_ulp_mode = 0;
721 	u32 hdr_size = sizeof(*req);
722 	u32 opcode = FW_OFLD_TX_DATA_WR;
723 	u32 immlen = 0;
724 	u32 force = is_t5(lldi->adapter_type) ? TX_FORCE_V(!submode) :
725 						T6_TX_FORCE_F;
726 
727 	if (cxgbi_skcb_test_flag(skb, SKCBF_TX_ISO)) {
728 		hdr_size += sizeof(struct cpl_tx_data_iso);
729 		opcode = FW_ISCSI_TX_DATA_WR;
730 		immlen += sizeof(struct cpl_tx_data_iso);
731 		submode |= 8;
732 	}
733 
734 	if (is_ofld_imm(skb))
735 		immlen += dlen;
736 
737 	req = (struct fw_ofld_tx_data_wr *)__skb_push(skb, hdr_size);
738 	req->op_to_immdlen = cpu_to_be32(FW_WR_OP_V(opcode) |
739 					 FW_WR_COMPL_V(compl) |
740 					 FW_WR_IMMDLEN_V(immlen));
741 	req->flowid_len16 = cpu_to_be32(FW_WR_FLOWID_V(csk->tid) |
742 					FW_WR_LEN16_V(credits));
743 	req->plen = cpu_to_be32(len);
744 	cpl =  (struct cpl_tx_data_iso *)(req + 1);
745 
746 	if (likely(cxgbi_skcb_test_flag(skb, SKCBF_TX_ISO)))
747 		cxgb4i_make_tx_iso_cpl(skb, cpl);
748 
749 	if (submode)
750 		wr_ulp_mode = FW_OFLD_TX_DATA_WR_ULPMODE_V(ULP2_MODE_ISCSI) |
751 			      FW_OFLD_TX_DATA_WR_ULPSUBMODE_V(submode);
752 
753 	req->tunnel_to_proxy = cpu_to_be32(wr_ulp_mode | force |
754 					   FW_OFLD_TX_DATA_WR_SHOVE_V(1U));
755 
756 	if (!cxgbi_sock_flag(csk, CTPF_TX_DATA_SENT))
757 		cxgbi_sock_set_flag(csk, CTPF_TX_DATA_SENT);
758 }
759 
760 static void arp_failure_skb_discard(void *handle, struct sk_buff *skb)
761 {
762 	kfree_skb(skb);
763 }
764 
765 static int push_tx_frames(struct cxgbi_sock *csk, int req_completion)
766 {
767 	int total_size = 0;
768 	struct sk_buff *skb;
769 
770 	if (unlikely(csk->state < CTP_ESTABLISHED ||
771 		csk->state == CTP_CLOSE_WAIT_1 || csk->state >= CTP_ABORTING)) {
772 		log_debug(1 << CXGBI_DBG_TOE | 1 << CXGBI_DBG_SOCK |
773 			  1 << CXGBI_DBG_PDU_TX,
774 			  "csk 0x%p,%u,0x%lx,%u, in closing state.\n",
775 			  csk, csk->state, csk->flags, csk->tid);
776 		return 0;
777 	}
778 
779 	while (csk->wr_cred && ((skb = skb_peek(&csk->write_queue)) != NULL)) {
780 		struct cxgbi_iso_info *iso_cpl;
781 		u32 dlen = skb->len;
782 		u32 len = skb->len;
783 		u32 iso_cpl_len = 0;
784 		u32 flowclen16 = 0;
785 		u32 credits_needed;
786 		u32 num_pdu = 1, hdr_len;
787 
788 		if (cxgbi_skcb_test_flag(skb, SKCBF_TX_ISO))
789 			iso_cpl_len = sizeof(struct cpl_tx_data_iso);
790 
791 		if (is_ofld_imm(skb))
792 			credits_needed = DIV_ROUND_UP(dlen + iso_cpl_len, 16);
793 		else
794 			credits_needed =
795 				DIV_ROUND_UP((8 * calc_tx_flits_ofld(skb)) +
796 					     iso_cpl_len, 16);
797 
798 		if (likely(cxgbi_skcb_test_flag(skb, SKCBF_TX_NEED_HDR)))
799 			credits_needed +=
800 			   DIV_ROUND_UP(sizeof(struct fw_ofld_tx_data_wr), 16);
801 
802 		/*
803 		 * Assumes the initial credits is large enough to support
804 		 * fw_flowc_wr plus largest possible first payload
805 		 */
806 		if (!cxgbi_sock_flag(csk, CTPF_TX_DATA_SENT)) {
807 			flowclen16 = send_tx_flowc_wr(csk);
808 			csk->wr_cred -= flowclen16;
809 			csk->wr_una_cred += flowclen16;
810 			cxgbi_sock_set_flag(csk, CTPF_TX_DATA_SENT);
811 		}
812 
813 		if (csk->wr_cred < credits_needed) {
814 			log_debug(1 << CXGBI_DBG_PDU_TX,
815 				  "csk 0x%p, skb %u/%u, wr %d < %u.\n",
816 				  csk, skb->len, skb->data_len,
817 				  credits_needed, csk->wr_cred);
818 
819 			csk->no_tx_credits++;
820 			break;
821 		}
822 
823 		csk->no_tx_credits = 0;
824 
825 		__skb_unlink(skb, &csk->write_queue);
826 		set_wr_txq(skb, CPL_PRIORITY_DATA, csk->port_id);
827 		skb->csum = (__force __wsum)(credits_needed + flowclen16);
828 		csk->wr_cred -= credits_needed;
829 		csk->wr_una_cred += credits_needed;
830 		cxgbi_sock_enqueue_wr(csk, skb);
831 
832 		log_debug(1 << CXGBI_DBG_PDU_TX,
833 			"csk 0x%p, skb %u/%u, wr %d, left %u, unack %u.\n",
834 			csk, skb->len, skb->data_len, credits_needed,
835 			csk->wr_cred, csk->wr_una_cred);
836 
837 		if (!req_completion &&
838 		    ((csk->wr_una_cred >= (csk->wr_max_cred / 2)) ||
839 		     after(csk->write_seq, (csk->snd_una + csk->snd_win / 2))))
840 			req_completion = 1;
841 
842 		if (likely(cxgbi_skcb_test_flag(skb, SKCBF_TX_NEED_HDR))) {
843 			u32 ulp_mode = cxgbi_skcb_tx_ulp_mode(skb);
844 
845 			if (cxgbi_skcb_test_flag(skb, SKCBF_TX_ISO)) {
846 				iso_cpl = (struct cxgbi_iso_info *)skb->head;
847 				num_pdu = iso_cpl->num_pdu;
848 				hdr_len = cxgbi_skcb_tx_iscsi_hdrlen(skb);
849 				len += (cxgbi_ulp_extra_len(ulp_mode) * num_pdu) +
850 				       (hdr_len * (num_pdu - 1));
851 			} else {
852 				len += cxgbi_ulp_extra_len(ulp_mode);
853 			}
854 
855 			cxgb4i_make_tx_data_wr(csk, skb, dlen, len,
856 					       credits_needed, req_completion);
857 			csk->snd_nxt += len;
858 			cxgbi_skcb_clear_flag(skb, SKCBF_TX_NEED_HDR);
859 		} else if (cxgbi_skcb_test_flag(skb, SKCBF_TX_FLAG_COMPL) &&
860 			   (csk->wr_una_cred >= (csk->wr_max_cred / 2))) {
861 			struct cpl_close_con_req *req =
862 				(struct cpl_close_con_req *)skb->data;
863 
864 			req->wr.wr_hi |= cpu_to_be32(FW_WR_COMPL_F);
865 		}
866 
867 		total_size += skb->truesize;
868 		t4_set_arp_err_handler(skb, csk, arp_failure_skb_discard);
869 
870 		log_debug(1 << CXGBI_DBG_TOE | 1 << CXGBI_DBG_PDU_TX,
871 			  "csk 0x%p,%u,0x%lx,%u, skb 0x%p, %u.\n",
872 			  csk, csk->state, csk->flags, csk->tid, skb, len);
873 		cxgb4_l2t_send(csk->cdev->ports[csk->port_id], skb, csk->l2t);
874 	}
875 	return total_size;
876 }
877 
878 static inline void free_atid(struct cxgbi_sock *csk)
879 {
880 	struct cxgb4_lld_info *lldi = cxgbi_cdev_priv(csk->cdev);
881 
882 	if (cxgbi_sock_flag(csk, CTPF_HAS_ATID)) {
883 		cxgb4_free_atid(lldi->tids, csk->atid);
884 		cxgbi_sock_clear_flag(csk, CTPF_HAS_ATID);
885 		cxgbi_sock_put(csk);
886 	}
887 }
888 
889 static void do_act_establish(struct cxgbi_device *cdev, struct sk_buff *skb)
890 {
891 	struct cxgbi_sock *csk;
892 	struct cpl_act_establish *req = (struct cpl_act_establish *)skb->data;
893 	unsigned short tcp_opt = ntohs(req->tcp_opt);
894 	unsigned int tid = GET_TID(req);
895 	unsigned int atid = TID_TID_G(ntohl(req->tos_atid));
896 	struct cxgb4_lld_info *lldi = cxgbi_cdev_priv(cdev);
897 	struct tid_info *t = lldi->tids;
898 	u32 rcv_isn = be32_to_cpu(req->rcv_isn);
899 
900 	csk = lookup_atid(t, atid);
901 	if (unlikely(!csk)) {
902 		pr_err("NO conn. for atid %u, cdev 0x%p.\n", atid, cdev);
903 		goto rel_skb;
904 	}
905 
906 	if (csk->atid != atid) {
907 		pr_err("bad conn atid %u, csk 0x%p,%u,0x%lx,tid %u, atid %u.\n",
908 			atid, csk, csk->state, csk->flags, csk->tid, csk->atid);
909 		goto rel_skb;
910 	}
911 
912 	pr_info_ipaddr("atid 0x%x, tid 0x%x, csk 0x%p,%u,0x%lx, isn %u.\n",
913 		       (&csk->saddr), (&csk->daddr),
914 		       atid, tid, csk, csk->state, csk->flags, rcv_isn);
915 
916 	module_put(cdev->owner);
917 
918 	cxgbi_sock_get(csk);
919 	csk->tid = tid;
920 	cxgb4_insert_tid(lldi->tids, csk, tid, csk->csk_family);
921 	cxgbi_sock_set_flag(csk, CTPF_HAS_TID);
922 
923 	free_atid(csk);
924 
925 	spin_lock_bh(&csk->lock);
926 	if (unlikely(csk->state != CTP_ACTIVE_OPEN))
927 		pr_info("csk 0x%p,%u,0x%lx,%u, got EST.\n",
928 			csk, csk->state, csk->flags, csk->tid);
929 
930 	if (csk->retry_timer.function) {
931 		del_timer(&csk->retry_timer);
932 		csk->retry_timer.function = NULL;
933 	}
934 
935 	csk->copied_seq = csk->rcv_wup = csk->rcv_nxt = rcv_isn;
936 	/*
937 	 * Causes the first RX_DATA_ACK to supply any Rx credits we couldn't
938 	 * pass through opt0.
939 	 */
940 	if (csk->rcv_win > (RCV_BUFSIZ_MASK << 10))
941 		csk->rcv_wup -= csk->rcv_win - (RCV_BUFSIZ_MASK << 10);
942 
943 	csk->advmss = lldi->mtus[TCPOPT_MSS_G(tcp_opt)] - 40;
944 	if (TCPOPT_TSTAMP_G(tcp_opt))
945 		csk->advmss -= 12;
946 	if (csk->advmss < 128)
947 		csk->advmss = 128;
948 
949 	log_debug(1 << CXGBI_DBG_TOE | 1 << CXGBI_DBG_SOCK,
950 		"csk 0x%p, mss_idx %u, advmss %u.\n",
951 			csk, TCPOPT_MSS_G(tcp_opt), csk->advmss);
952 
953 	cxgbi_sock_established(csk, ntohl(req->snd_isn), ntohs(req->tcp_opt));
954 
955 	if (unlikely(cxgbi_sock_flag(csk, CTPF_ACTIVE_CLOSE_NEEDED)))
956 		send_abort_req(csk);
957 	else {
958 		if (skb_queue_len(&csk->write_queue))
959 			push_tx_frames(csk, 0);
960 		cxgbi_conn_tx_open(csk);
961 	}
962 	spin_unlock_bh(&csk->lock);
963 
964 rel_skb:
965 	__kfree_skb(skb);
966 }
967 
968 static int act_open_rpl_status_to_errno(int status)
969 {
970 	switch (status) {
971 	case CPL_ERR_CONN_RESET:
972 		return -ECONNREFUSED;
973 	case CPL_ERR_ARP_MISS:
974 		return -EHOSTUNREACH;
975 	case CPL_ERR_CONN_TIMEDOUT:
976 		return -ETIMEDOUT;
977 	case CPL_ERR_TCAM_FULL:
978 		return -ENOMEM;
979 	case CPL_ERR_CONN_EXIST:
980 		return -EADDRINUSE;
981 	default:
982 		return -EIO;
983 	}
984 }
985 
986 static void csk_act_open_retry_timer(struct timer_list *t)
987 {
988 	struct sk_buff *skb = NULL;
989 	struct cxgbi_sock *csk = from_timer(csk, t, retry_timer);
990 	struct cxgb4_lld_info *lldi = cxgbi_cdev_priv(csk->cdev);
991 	void (*send_act_open_func)(struct cxgbi_sock *, struct sk_buff *,
992 				   struct l2t_entry *);
993 	int t4 = is_t4(lldi->adapter_type), size, size6;
994 
995 	log_debug(1 << CXGBI_DBG_TOE | 1 << CXGBI_DBG_SOCK,
996 		"csk 0x%p,%u,0x%lx,%u.\n",
997 		csk, csk->state, csk->flags, csk->tid);
998 
999 	cxgbi_sock_get(csk);
1000 	spin_lock_bh(&csk->lock);
1001 
1002 	if (t4) {
1003 		size = sizeof(struct cpl_act_open_req);
1004 		size6 = sizeof(struct cpl_act_open_req6);
1005 	} else {
1006 		size = sizeof(struct cpl_t5_act_open_req);
1007 		size6 = sizeof(struct cpl_t5_act_open_req6);
1008 	}
1009 
1010 	if (csk->csk_family == AF_INET) {
1011 		send_act_open_func = send_act_open_req;
1012 		skb = alloc_wr(size, 0, GFP_ATOMIC);
1013 #if IS_ENABLED(CONFIG_IPV6)
1014 	} else {
1015 		send_act_open_func = send_act_open_req6;
1016 		skb = alloc_wr(size6, 0, GFP_ATOMIC);
1017 #endif
1018 	}
1019 
1020 	if (!skb)
1021 		cxgbi_sock_fail_act_open(csk, -ENOMEM);
1022 	else {
1023 		skb->sk = (struct sock *)csk;
1024 		t4_set_arp_err_handler(skb, csk,
1025 				       cxgbi_sock_act_open_req_arp_failure);
1026 		send_act_open_func(csk, skb, csk->l2t);
1027 	}
1028 
1029 	spin_unlock_bh(&csk->lock);
1030 	cxgbi_sock_put(csk);
1031 
1032 }
1033 
1034 static inline bool is_neg_adv(unsigned int status)
1035 {
1036 	return status == CPL_ERR_RTX_NEG_ADVICE ||
1037 		status == CPL_ERR_KEEPALV_NEG_ADVICE ||
1038 		status == CPL_ERR_PERSIST_NEG_ADVICE;
1039 }
1040 
1041 static void do_act_open_rpl(struct cxgbi_device *cdev, struct sk_buff *skb)
1042 {
1043 	struct cxgbi_sock *csk;
1044 	struct cpl_act_open_rpl *rpl = (struct cpl_act_open_rpl *)skb->data;
1045 	unsigned int tid = GET_TID(rpl);
1046 	unsigned int atid =
1047 		TID_TID_G(AOPEN_ATID_G(be32_to_cpu(rpl->atid_status)));
1048 	unsigned int status = AOPEN_STATUS_G(be32_to_cpu(rpl->atid_status));
1049 	struct cxgb4_lld_info *lldi = cxgbi_cdev_priv(cdev);
1050 	struct tid_info *t = lldi->tids;
1051 
1052 	csk = lookup_atid(t, atid);
1053 	if (unlikely(!csk)) {
1054 		pr_err("NO matching conn. atid %u, tid %u.\n", atid, tid);
1055 		goto rel_skb;
1056 	}
1057 
1058 	pr_info_ipaddr("tid %u/%u, status %u.\n"
1059 		       "csk 0x%p,%u,0x%lx. ", (&csk->saddr), (&csk->daddr),
1060 		       atid, tid, status, csk, csk->state, csk->flags);
1061 
1062 	if (is_neg_adv(status))
1063 		goto rel_skb;
1064 
1065 	module_put(cdev->owner);
1066 
1067 	if (status && status != CPL_ERR_TCAM_FULL &&
1068 	    status != CPL_ERR_CONN_EXIST &&
1069 	    status != CPL_ERR_ARP_MISS)
1070 		cxgb4_remove_tid(lldi->tids, csk->port_id, GET_TID(rpl),
1071 				 csk->csk_family);
1072 
1073 	cxgbi_sock_get(csk);
1074 	spin_lock_bh(&csk->lock);
1075 
1076 	if (status == CPL_ERR_CONN_EXIST &&
1077 	    csk->retry_timer.function != csk_act_open_retry_timer) {
1078 		csk->retry_timer.function = csk_act_open_retry_timer;
1079 		mod_timer(&csk->retry_timer, jiffies + HZ / 2);
1080 	} else
1081 		cxgbi_sock_fail_act_open(csk,
1082 					act_open_rpl_status_to_errno(status));
1083 
1084 	spin_unlock_bh(&csk->lock);
1085 	cxgbi_sock_put(csk);
1086 rel_skb:
1087 	__kfree_skb(skb);
1088 }
1089 
1090 static void do_peer_close(struct cxgbi_device *cdev, struct sk_buff *skb)
1091 {
1092 	struct cxgbi_sock *csk;
1093 	struct cpl_peer_close *req = (struct cpl_peer_close *)skb->data;
1094 	unsigned int tid = GET_TID(req);
1095 	struct cxgb4_lld_info *lldi = cxgbi_cdev_priv(cdev);
1096 	struct tid_info *t = lldi->tids;
1097 
1098 	csk = lookup_tid(t, tid);
1099 	if (unlikely(!csk)) {
1100 		pr_err("can't find connection for tid %u.\n", tid);
1101 		goto rel_skb;
1102 	}
1103 	pr_info_ipaddr("csk 0x%p,%u,0x%lx,%u.\n",
1104 		       (&csk->saddr), (&csk->daddr),
1105 		       csk, csk->state, csk->flags, csk->tid);
1106 	cxgbi_sock_rcv_peer_close(csk);
1107 rel_skb:
1108 	__kfree_skb(skb);
1109 }
1110 
1111 static void do_close_con_rpl(struct cxgbi_device *cdev, struct sk_buff *skb)
1112 {
1113 	struct cxgbi_sock *csk;
1114 	struct cpl_close_con_rpl *rpl = (struct cpl_close_con_rpl *)skb->data;
1115 	unsigned int tid = GET_TID(rpl);
1116 	struct cxgb4_lld_info *lldi = cxgbi_cdev_priv(cdev);
1117 	struct tid_info *t = lldi->tids;
1118 
1119 	csk = lookup_tid(t, tid);
1120 	if (unlikely(!csk)) {
1121 		pr_err("can't find connection for tid %u.\n", tid);
1122 		goto rel_skb;
1123 	}
1124 	pr_info_ipaddr("csk 0x%p,%u,0x%lx,%u.\n",
1125 		       (&csk->saddr), (&csk->daddr),
1126 		       csk, csk->state, csk->flags, csk->tid);
1127 	cxgbi_sock_rcv_close_conn_rpl(csk, ntohl(rpl->snd_nxt));
1128 rel_skb:
1129 	__kfree_skb(skb);
1130 }
1131 
1132 static int abort_status_to_errno(struct cxgbi_sock *csk, int abort_reason,
1133 								int *need_rst)
1134 {
1135 	switch (abort_reason) {
1136 	case CPL_ERR_BAD_SYN:
1137 	case CPL_ERR_CONN_RESET:
1138 		return csk->state > CTP_ESTABLISHED ?
1139 			-EPIPE : -ECONNRESET;
1140 	case CPL_ERR_XMIT_TIMEDOUT:
1141 	case CPL_ERR_PERSIST_TIMEDOUT:
1142 	case CPL_ERR_FINWAIT2_TIMEDOUT:
1143 	case CPL_ERR_KEEPALIVE_TIMEDOUT:
1144 		return -ETIMEDOUT;
1145 	default:
1146 		return -EIO;
1147 	}
1148 }
1149 
1150 static void do_abort_req_rss(struct cxgbi_device *cdev, struct sk_buff *skb)
1151 {
1152 	struct cxgbi_sock *csk;
1153 	struct cpl_abort_req_rss *req = (struct cpl_abort_req_rss *)skb->data;
1154 	unsigned int tid = GET_TID(req);
1155 	struct cxgb4_lld_info *lldi = cxgbi_cdev_priv(cdev);
1156 	struct tid_info *t = lldi->tids;
1157 	int rst_status = CPL_ABORT_NO_RST;
1158 
1159 	csk = lookup_tid(t, tid);
1160 	if (unlikely(!csk)) {
1161 		pr_err("can't find connection for tid %u.\n", tid);
1162 		goto rel_skb;
1163 	}
1164 
1165 	pr_info_ipaddr("csk 0x%p,%u,0x%lx,%u, status %u.\n",
1166 		       (&csk->saddr), (&csk->daddr),
1167 		       csk, csk->state, csk->flags, csk->tid, req->status);
1168 
1169 	if (is_neg_adv(req->status))
1170 		goto rel_skb;
1171 
1172 	cxgbi_sock_get(csk);
1173 	spin_lock_bh(&csk->lock);
1174 
1175 	cxgbi_sock_clear_flag(csk, CTPF_ABORT_REQ_RCVD);
1176 
1177 	if (!cxgbi_sock_flag(csk, CTPF_TX_DATA_SENT)) {
1178 		send_tx_flowc_wr(csk);
1179 		cxgbi_sock_set_flag(csk, CTPF_TX_DATA_SENT);
1180 	}
1181 
1182 	cxgbi_sock_set_flag(csk, CTPF_ABORT_REQ_RCVD);
1183 	cxgbi_sock_set_state(csk, CTP_ABORTING);
1184 
1185 	send_abort_rpl(csk, rst_status);
1186 
1187 	if (!cxgbi_sock_flag(csk, CTPF_ABORT_RPL_PENDING)) {
1188 		csk->err = abort_status_to_errno(csk, req->status, &rst_status);
1189 		cxgbi_sock_closed(csk);
1190 	}
1191 
1192 	spin_unlock_bh(&csk->lock);
1193 	cxgbi_sock_put(csk);
1194 rel_skb:
1195 	__kfree_skb(skb);
1196 }
1197 
1198 static void do_abort_rpl_rss(struct cxgbi_device *cdev, struct sk_buff *skb)
1199 {
1200 	struct cxgbi_sock *csk;
1201 	struct cpl_abort_rpl_rss *rpl = (struct cpl_abort_rpl_rss *)skb->data;
1202 	unsigned int tid = GET_TID(rpl);
1203 	struct cxgb4_lld_info *lldi = cxgbi_cdev_priv(cdev);
1204 	struct tid_info *t = lldi->tids;
1205 
1206 	csk = lookup_tid(t, tid);
1207 	if (!csk)
1208 		goto rel_skb;
1209 
1210 	pr_info_ipaddr("csk 0x%p,%u,0x%lx,%u, status %u.\n",
1211 		       (&csk->saddr), (&csk->daddr), csk,
1212 		       csk->state, csk->flags, csk->tid, rpl->status);
1213 
1214 	if (rpl->status == CPL_ERR_ABORT_FAILED)
1215 		goto rel_skb;
1216 
1217 	cxgbi_sock_rcv_abort_rpl(csk);
1218 rel_skb:
1219 	__kfree_skb(skb);
1220 }
1221 
1222 static void do_rx_data(struct cxgbi_device *cdev, struct sk_buff *skb)
1223 {
1224 	struct cxgbi_sock *csk;
1225 	struct cpl_rx_data *cpl = (struct cpl_rx_data *)skb->data;
1226 	unsigned int tid = GET_TID(cpl);
1227 	struct cxgb4_lld_info *lldi = cxgbi_cdev_priv(cdev);
1228 	struct tid_info *t = lldi->tids;
1229 
1230 	csk = lookup_tid(t, tid);
1231 	if (!csk) {
1232 		pr_err("can't find connection for tid %u.\n", tid);
1233 	} else {
1234 		/* not expecting this, reset the connection. */
1235 		pr_err("csk 0x%p, tid %u, rcv cpl_rx_data.\n", csk, tid);
1236 		spin_lock_bh(&csk->lock);
1237 		send_abort_req(csk);
1238 		spin_unlock_bh(&csk->lock);
1239 	}
1240 	__kfree_skb(skb);
1241 }
1242 
1243 static void do_rx_iscsi_hdr(struct cxgbi_device *cdev, struct sk_buff *skb)
1244 {
1245 	struct cxgbi_sock *csk;
1246 	struct cpl_iscsi_hdr *cpl = (struct cpl_iscsi_hdr *)skb->data;
1247 	unsigned short pdu_len_ddp = be16_to_cpu(cpl->pdu_len_ddp);
1248 	unsigned int tid = GET_TID(cpl);
1249 	struct cxgb4_lld_info *lldi = cxgbi_cdev_priv(cdev);
1250 	struct tid_info *t = lldi->tids;
1251 
1252 	csk = lookup_tid(t, tid);
1253 	if (unlikely(!csk)) {
1254 		pr_err("can't find conn. for tid %u.\n", tid);
1255 		goto rel_skb;
1256 	}
1257 
1258 	log_debug(1 << CXGBI_DBG_TOE | 1 << CXGBI_DBG_PDU_RX,
1259 		"csk 0x%p,%u,0x%lx, tid %u, skb 0x%p,%u, 0x%x.\n",
1260 		csk, csk->state, csk->flags, csk->tid, skb, skb->len,
1261 		pdu_len_ddp);
1262 
1263 	spin_lock_bh(&csk->lock);
1264 
1265 	if (unlikely(csk->state >= CTP_PASSIVE_CLOSE)) {
1266 		log_debug(1 << CXGBI_DBG_TOE | 1 << CXGBI_DBG_SOCK,
1267 			"csk 0x%p,%u,0x%lx,%u, bad state.\n",
1268 			csk, csk->state, csk->flags, csk->tid);
1269 		if (csk->state != CTP_ABORTING)
1270 			goto abort_conn;
1271 		else
1272 			goto discard;
1273 	}
1274 
1275 	cxgbi_skcb_tcp_seq(skb) = ntohl(cpl->seq);
1276 	cxgbi_skcb_flags(skb) = 0;
1277 
1278 	skb_reset_transport_header(skb);
1279 	__skb_pull(skb, sizeof(*cpl));
1280 	__pskb_trim(skb, ntohs(cpl->len));
1281 
1282 	if (!csk->skb_ulp_lhdr) {
1283 		unsigned char *bhs;
1284 		unsigned int hlen, dlen, plen;
1285 
1286 		log_debug(1 << CXGBI_DBG_TOE | 1 << CXGBI_DBG_PDU_RX,
1287 			"csk 0x%p,%u,0x%lx, tid %u, skb 0x%p header.\n",
1288 			csk, csk->state, csk->flags, csk->tid, skb);
1289 		csk->skb_ulp_lhdr = skb;
1290 		cxgbi_skcb_set_flag(skb, SKCBF_RX_HDR);
1291 
1292 		if ((CHELSIO_CHIP_VERSION(lldi->adapter_type) <= CHELSIO_T5) &&
1293 		    (cxgbi_skcb_tcp_seq(skb) != csk->rcv_nxt)) {
1294 			pr_info("tid %u, CPL_ISCSI_HDR, bad seq, 0x%x/0x%x.\n",
1295 				csk->tid, cxgbi_skcb_tcp_seq(skb),
1296 				csk->rcv_nxt);
1297 			goto abort_conn;
1298 		}
1299 
1300 		bhs = skb->data;
1301 		hlen = ntohs(cpl->len);
1302 		dlen = ntohl(*(unsigned int *)(bhs + 4)) & 0xFFFFFF;
1303 
1304 		plen = ISCSI_PDU_LEN_G(pdu_len_ddp);
1305 		if (is_t4(lldi->adapter_type))
1306 			plen -= 40;
1307 
1308 		if ((hlen + dlen) != plen) {
1309 			pr_info("tid 0x%x, CPL_ISCSI_HDR, pdu len "
1310 				"mismatch %u != %u + %u, seq 0x%x.\n",
1311 				csk->tid, plen, hlen, dlen,
1312 				cxgbi_skcb_tcp_seq(skb));
1313 			goto abort_conn;
1314 		}
1315 
1316 		cxgbi_skcb_rx_pdulen(skb) = (hlen + dlen + 3) & (~0x3);
1317 		if (dlen)
1318 			cxgbi_skcb_rx_pdulen(skb) += csk->dcrc_len;
1319 		csk->rcv_nxt += cxgbi_skcb_rx_pdulen(skb);
1320 
1321 		log_debug(1 << CXGBI_DBG_TOE | 1 << CXGBI_DBG_PDU_RX,
1322 			"csk 0x%p, skb 0x%p, 0x%x,%u+%u,0x%x,0x%x.\n",
1323 			csk, skb, *bhs, hlen, dlen,
1324 			ntohl(*((unsigned int *)(bhs + 16))),
1325 			ntohl(*((unsigned int *)(bhs + 24))));
1326 
1327 	} else {
1328 		struct sk_buff *lskb = csk->skb_ulp_lhdr;
1329 
1330 		cxgbi_skcb_set_flag(lskb, SKCBF_RX_DATA);
1331 		log_debug(1 << CXGBI_DBG_TOE | 1 << CXGBI_DBG_PDU_RX,
1332 			"csk 0x%p,%u,0x%lx, skb 0x%p data, 0x%p.\n",
1333 			csk, csk->state, csk->flags, skb, lskb);
1334 	}
1335 
1336 	__skb_queue_tail(&csk->receive_queue, skb);
1337 	spin_unlock_bh(&csk->lock);
1338 	return;
1339 
1340 abort_conn:
1341 	send_abort_req(csk);
1342 discard:
1343 	spin_unlock_bh(&csk->lock);
1344 rel_skb:
1345 	__kfree_skb(skb);
1346 }
1347 
1348 static void do_rx_iscsi_data(struct cxgbi_device *cdev, struct sk_buff *skb)
1349 {
1350 	struct cxgbi_sock *csk;
1351 	struct cpl_iscsi_hdr *cpl = (struct cpl_iscsi_hdr *)skb->data;
1352 	struct cxgb4_lld_info *lldi = cxgbi_cdev_priv(cdev);
1353 	struct tid_info *t = lldi->tids;
1354 	struct sk_buff *lskb;
1355 	u32 tid = GET_TID(cpl);
1356 	u16 pdu_len_ddp = be16_to_cpu(cpl->pdu_len_ddp);
1357 
1358 	csk = lookup_tid(t, tid);
1359 	if (unlikely(!csk)) {
1360 		pr_err("can't find conn. for tid %u.\n", tid);
1361 		goto rel_skb;
1362 	}
1363 
1364 	log_debug(1 << CXGBI_DBG_TOE | 1 << CXGBI_DBG_PDU_RX,
1365 		  "csk 0x%p,%u,0x%lx, tid %u, skb 0x%p,%u, 0x%x.\n",
1366 		  csk, csk->state, csk->flags, csk->tid, skb,
1367 		  skb->len, pdu_len_ddp);
1368 
1369 	spin_lock_bh(&csk->lock);
1370 
1371 	if (unlikely(csk->state >= CTP_PASSIVE_CLOSE)) {
1372 		log_debug(1 << CXGBI_DBG_TOE | 1 << CXGBI_DBG_SOCK,
1373 			  "csk 0x%p,%u,0x%lx,%u, bad state.\n",
1374 			  csk, csk->state, csk->flags, csk->tid);
1375 
1376 		if (csk->state != CTP_ABORTING)
1377 			goto abort_conn;
1378 		else
1379 			goto discard;
1380 	}
1381 
1382 	cxgbi_skcb_tcp_seq(skb) = be32_to_cpu(cpl->seq);
1383 	cxgbi_skcb_flags(skb) = 0;
1384 
1385 	skb_reset_transport_header(skb);
1386 	__skb_pull(skb, sizeof(*cpl));
1387 	__pskb_trim(skb, ntohs(cpl->len));
1388 
1389 	if (!csk->skb_ulp_lhdr)
1390 		csk->skb_ulp_lhdr = skb;
1391 
1392 	lskb = csk->skb_ulp_lhdr;
1393 	cxgbi_skcb_set_flag(lskb, SKCBF_RX_DATA);
1394 
1395 	log_debug(1 << CXGBI_DBG_TOE | 1 << CXGBI_DBG_PDU_RX,
1396 		  "csk 0x%p,%u,0x%lx, skb 0x%p data, 0x%p.\n",
1397 		  csk, csk->state, csk->flags, skb, lskb);
1398 
1399 	__skb_queue_tail(&csk->receive_queue, skb);
1400 	spin_unlock_bh(&csk->lock);
1401 	return;
1402 
1403 abort_conn:
1404 	send_abort_req(csk);
1405 discard:
1406 	spin_unlock_bh(&csk->lock);
1407 rel_skb:
1408 	__kfree_skb(skb);
1409 }
1410 
1411 static void
1412 cxgb4i_process_ddpvld(struct cxgbi_sock *csk,
1413 		      struct sk_buff *skb, u32 ddpvld)
1414 {
1415 	if (ddpvld & (1 << CPL_RX_DDP_STATUS_HCRC_SHIFT)) {
1416 		pr_info("csk 0x%p, lhdr 0x%p, status 0x%x, hcrc bad 0x%lx.\n",
1417 			csk, skb, ddpvld, cxgbi_skcb_flags(skb));
1418 		cxgbi_skcb_set_flag(skb, SKCBF_RX_HCRC_ERR);
1419 	}
1420 
1421 	if (ddpvld & (1 << CPL_RX_DDP_STATUS_DCRC_SHIFT)) {
1422 		pr_info("csk 0x%p, lhdr 0x%p, status 0x%x, dcrc bad 0x%lx.\n",
1423 			csk, skb, ddpvld, cxgbi_skcb_flags(skb));
1424 		cxgbi_skcb_set_flag(skb, SKCBF_RX_DCRC_ERR);
1425 	}
1426 
1427 	if (ddpvld & (1 << CPL_RX_DDP_STATUS_PAD_SHIFT)) {
1428 		log_debug(1 << CXGBI_DBG_PDU_RX,
1429 			  "csk 0x%p, lhdr 0x%p, status 0x%x, pad bad.\n",
1430 			  csk, skb, ddpvld);
1431 		cxgbi_skcb_set_flag(skb, SKCBF_RX_PAD_ERR);
1432 	}
1433 
1434 	if ((ddpvld & (1 << CPL_RX_DDP_STATUS_DDP_SHIFT)) &&
1435 	    !cxgbi_skcb_test_flag(skb, SKCBF_RX_DATA)) {
1436 		log_debug(1 << CXGBI_DBG_PDU_RX,
1437 			  "csk 0x%p, lhdr 0x%p, 0x%x, data ddp'ed.\n",
1438 			  csk, skb, ddpvld);
1439 		cxgbi_skcb_set_flag(skb, SKCBF_RX_DATA_DDPD);
1440 	}
1441 }
1442 
1443 static void do_rx_data_ddp(struct cxgbi_device *cdev,
1444 				  struct sk_buff *skb)
1445 {
1446 	struct cxgbi_sock *csk;
1447 	struct sk_buff *lskb;
1448 	struct cpl_rx_data_ddp *rpl = (struct cpl_rx_data_ddp *)skb->data;
1449 	unsigned int tid = GET_TID(rpl);
1450 	struct cxgb4_lld_info *lldi = cxgbi_cdev_priv(cdev);
1451 	struct tid_info *t = lldi->tids;
1452 	u32 ddpvld = be32_to_cpu(rpl->ddpvld);
1453 
1454 	csk = lookup_tid(t, tid);
1455 	if (unlikely(!csk)) {
1456 		pr_err("can't find connection for tid %u.\n", tid);
1457 		goto rel_skb;
1458 	}
1459 
1460 	log_debug(1 << CXGBI_DBG_TOE | 1 << CXGBI_DBG_PDU_RX,
1461 		"csk 0x%p,%u,0x%lx, skb 0x%p,0x%x, lhdr 0x%p.\n",
1462 		csk, csk->state, csk->flags, skb, ddpvld, csk->skb_ulp_lhdr);
1463 
1464 	spin_lock_bh(&csk->lock);
1465 
1466 	if (unlikely(csk->state >= CTP_PASSIVE_CLOSE)) {
1467 		log_debug(1 << CXGBI_DBG_TOE | 1 << CXGBI_DBG_SOCK,
1468 			"csk 0x%p,%u,0x%lx,%u, bad state.\n",
1469 			csk, csk->state, csk->flags, csk->tid);
1470 		if (csk->state != CTP_ABORTING)
1471 			goto abort_conn;
1472 		else
1473 			goto discard;
1474 	}
1475 
1476 	if (!csk->skb_ulp_lhdr) {
1477 		pr_err("tid 0x%x, rcv RX_DATA_DDP w/o pdu bhs.\n", csk->tid);
1478 		goto abort_conn;
1479 	}
1480 
1481 	lskb = csk->skb_ulp_lhdr;
1482 	csk->skb_ulp_lhdr = NULL;
1483 
1484 	cxgbi_skcb_rx_ddigest(lskb) = ntohl(rpl->ulp_crc);
1485 
1486 	if (ntohs(rpl->len) != cxgbi_skcb_rx_pdulen(lskb))
1487 		pr_info("tid 0x%x, RX_DATA_DDP pdulen %u != %u.\n",
1488 			csk->tid, ntohs(rpl->len), cxgbi_skcb_rx_pdulen(lskb));
1489 
1490 	cxgb4i_process_ddpvld(csk, lskb, ddpvld);
1491 
1492 	log_debug(1 << CXGBI_DBG_PDU_RX,
1493 		"csk 0x%p, lskb 0x%p, f 0x%lx.\n",
1494 		csk, lskb, cxgbi_skcb_flags(lskb));
1495 
1496 	cxgbi_skcb_set_flag(lskb, SKCBF_RX_STATUS);
1497 	cxgbi_conn_pdu_ready(csk);
1498 	spin_unlock_bh(&csk->lock);
1499 	goto rel_skb;
1500 
1501 abort_conn:
1502 	send_abort_req(csk);
1503 discard:
1504 	spin_unlock_bh(&csk->lock);
1505 rel_skb:
1506 	__kfree_skb(skb);
1507 }
1508 
1509 static void
1510 do_rx_iscsi_cmp(struct cxgbi_device *cdev, struct sk_buff *skb)
1511 {
1512 	struct cxgbi_sock *csk;
1513 	struct cpl_rx_iscsi_cmp *rpl = (struct cpl_rx_iscsi_cmp *)skb->data;
1514 	struct cxgb4_lld_info *lldi = cxgbi_cdev_priv(cdev);
1515 	struct tid_info *t = lldi->tids;
1516 	struct sk_buff *data_skb = NULL;
1517 	u32 tid = GET_TID(rpl);
1518 	u32 ddpvld = be32_to_cpu(rpl->ddpvld);
1519 	u32 seq = be32_to_cpu(rpl->seq);
1520 	u16 pdu_len_ddp = be16_to_cpu(rpl->pdu_len_ddp);
1521 
1522 	csk = lookup_tid(t, tid);
1523 	if (unlikely(!csk)) {
1524 		pr_err("can't find connection for tid %u.\n", tid);
1525 		goto rel_skb;
1526 	}
1527 
1528 	log_debug(1 << CXGBI_DBG_TOE | 1 << CXGBI_DBG_PDU_RX,
1529 		  "csk 0x%p,%u,0x%lx, skb 0x%p,0x%x, lhdr 0x%p, len %u, "
1530 		  "pdu_len_ddp %u, status %u.\n",
1531 		  csk, csk->state, csk->flags, skb, ddpvld, csk->skb_ulp_lhdr,
1532 		  ntohs(rpl->len), pdu_len_ddp,  rpl->status);
1533 
1534 	spin_lock_bh(&csk->lock);
1535 
1536 	if (unlikely(csk->state >= CTP_PASSIVE_CLOSE)) {
1537 		log_debug(1 << CXGBI_DBG_TOE | 1 << CXGBI_DBG_SOCK,
1538 			  "csk 0x%p,%u,0x%lx,%u, bad state.\n",
1539 			  csk, csk->state, csk->flags, csk->tid);
1540 
1541 		if (csk->state != CTP_ABORTING)
1542 			goto abort_conn;
1543 		else
1544 			goto discard;
1545 	}
1546 
1547 	cxgbi_skcb_tcp_seq(skb) = seq;
1548 	cxgbi_skcb_flags(skb) = 0;
1549 	cxgbi_skcb_rx_pdulen(skb) = 0;
1550 
1551 	skb_reset_transport_header(skb);
1552 	__skb_pull(skb, sizeof(*rpl));
1553 	__pskb_trim(skb, be16_to_cpu(rpl->len));
1554 
1555 	csk->rcv_nxt = seq + pdu_len_ddp;
1556 
1557 	if (csk->skb_ulp_lhdr) {
1558 		data_skb = skb_peek(&csk->receive_queue);
1559 		if (!data_skb ||
1560 		    !cxgbi_skcb_test_flag(data_skb, SKCBF_RX_DATA)) {
1561 			pr_err("Error! freelist data not found 0x%p, tid %u\n",
1562 			       data_skb, tid);
1563 
1564 			goto abort_conn;
1565 		}
1566 		__skb_unlink(data_skb, &csk->receive_queue);
1567 
1568 		cxgbi_skcb_set_flag(skb, SKCBF_RX_DATA);
1569 
1570 		__skb_queue_tail(&csk->receive_queue, skb);
1571 		__skb_queue_tail(&csk->receive_queue, data_skb);
1572 	} else {
1573 		 __skb_queue_tail(&csk->receive_queue, skb);
1574 	}
1575 
1576 	csk->skb_ulp_lhdr = NULL;
1577 
1578 	cxgbi_skcb_set_flag(skb, SKCBF_RX_HDR);
1579 	cxgbi_skcb_set_flag(skb, SKCBF_RX_STATUS);
1580 	cxgbi_skcb_set_flag(skb, SKCBF_RX_ISCSI_COMPL);
1581 	cxgbi_skcb_rx_ddigest(skb) = be32_to_cpu(rpl->ulp_crc);
1582 
1583 	cxgb4i_process_ddpvld(csk, skb, ddpvld);
1584 
1585 	log_debug(1 << CXGBI_DBG_PDU_RX, "csk 0x%p, skb 0x%p, f 0x%lx.\n",
1586 		  csk, skb, cxgbi_skcb_flags(skb));
1587 
1588 	cxgbi_conn_pdu_ready(csk);
1589 	spin_unlock_bh(&csk->lock);
1590 
1591 	return;
1592 
1593 abort_conn:
1594 	send_abort_req(csk);
1595 discard:
1596 	spin_unlock_bh(&csk->lock);
1597 rel_skb:
1598 	__kfree_skb(skb);
1599 }
1600 
1601 static void do_fw4_ack(struct cxgbi_device *cdev, struct sk_buff *skb)
1602 {
1603 	struct cxgbi_sock *csk;
1604 	struct cpl_fw4_ack *rpl = (struct cpl_fw4_ack *)skb->data;
1605 	unsigned int tid = GET_TID(rpl);
1606 	struct cxgb4_lld_info *lldi = cxgbi_cdev_priv(cdev);
1607 	struct tid_info *t = lldi->tids;
1608 
1609 	csk = lookup_tid(t, tid);
1610 	if (unlikely(!csk))
1611 		pr_err("can't find connection for tid %u.\n", tid);
1612 	else {
1613 		log_debug(1 << CXGBI_DBG_TOE | 1 << CXGBI_DBG_SOCK,
1614 			"csk 0x%p,%u,0x%lx,%u.\n",
1615 			csk, csk->state, csk->flags, csk->tid);
1616 		cxgbi_sock_rcv_wr_ack(csk, rpl->credits, ntohl(rpl->snd_una),
1617 					rpl->seq_vld);
1618 	}
1619 	__kfree_skb(skb);
1620 }
1621 
1622 static void do_set_tcb_rpl(struct cxgbi_device *cdev, struct sk_buff *skb)
1623 {
1624 	struct cpl_set_tcb_rpl *rpl = (struct cpl_set_tcb_rpl *)skb->data;
1625 	unsigned int tid = GET_TID(rpl);
1626 	struct cxgb4_lld_info *lldi = cxgbi_cdev_priv(cdev);
1627 	struct tid_info *t = lldi->tids;
1628 	struct cxgbi_sock *csk;
1629 
1630 	csk = lookup_tid(t, tid);
1631 	if (!csk) {
1632 		pr_err("can't find conn. for tid %u.\n", tid);
1633 		return;
1634 	}
1635 
1636 	log_debug(1 << CXGBI_DBG_TOE | 1 << CXGBI_DBG_SOCK,
1637 		"csk 0x%p,%u,%lx,%u, status 0x%x.\n",
1638 		csk, csk->state, csk->flags, csk->tid, rpl->status);
1639 
1640 	if (rpl->status != CPL_ERR_NONE) {
1641 		pr_err("csk 0x%p,%u, SET_TCB_RPL status %u.\n",
1642 			csk, tid, rpl->status);
1643 		csk->err = -EINVAL;
1644 	}
1645 
1646 	complete(&csk->cmpl);
1647 
1648 	__kfree_skb(skb);
1649 }
1650 
1651 static int alloc_cpls(struct cxgbi_sock *csk)
1652 {
1653 	csk->cpl_close = alloc_wr(sizeof(struct cpl_close_con_req),
1654 					0, GFP_KERNEL);
1655 	if (!csk->cpl_close)
1656 		return -ENOMEM;
1657 
1658 	csk->cpl_abort_req = alloc_wr(sizeof(struct cpl_abort_req),
1659 					0, GFP_KERNEL);
1660 	if (!csk->cpl_abort_req)
1661 		goto free_cpls;
1662 
1663 	csk->cpl_abort_rpl = alloc_wr(sizeof(struct cpl_abort_rpl),
1664 					0, GFP_KERNEL);
1665 	if (!csk->cpl_abort_rpl)
1666 		goto free_cpls;
1667 	return 0;
1668 
1669 free_cpls:
1670 	cxgbi_sock_free_cpl_skbs(csk);
1671 	return -ENOMEM;
1672 }
1673 
1674 static inline void l2t_put(struct cxgbi_sock *csk)
1675 {
1676 	if (csk->l2t) {
1677 		cxgb4_l2t_release(csk->l2t);
1678 		csk->l2t = NULL;
1679 		cxgbi_sock_put(csk);
1680 	}
1681 }
1682 
1683 static void release_offload_resources(struct cxgbi_sock *csk)
1684 {
1685 	struct cxgb4_lld_info *lldi;
1686 #if IS_ENABLED(CONFIG_IPV6)
1687 	struct net_device *ndev = csk->cdev->ports[csk->port_id];
1688 #endif
1689 
1690 	log_debug(1 << CXGBI_DBG_TOE | 1 << CXGBI_DBG_SOCK,
1691 		"csk 0x%p,%u,0x%lx,%u.\n",
1692 		csk, csk->state, csk->flags, csk->tid);
1693 
1694 	cxgbi_sock_free_cpl_skbs(csk);
1695 	cxgbi_sock_purge_write_queue(csk);
1696 	if (csk->wr_cred != csk->wr_max_cred) {
1697 		cxgbi_sock_purge_wr_queue(csk);
1698 		cxgbi_sock_reset_wr_list(csk);
1699 	}
1700 
1701 	l2t_put(csk);
1702 #if IS_ENABLED(CONFIG_IPV6)
1703 	if (csk->csk_family == AF_INET6)
1704 		cxgb4_clip_release(ndev,
1705 				   (const u32 *)&csk->saddr6.sin6_addr, 1);
1706 #endif
1707 
1708 	if (cxgbi_sock_flag(csk, CTPF_HAS_ATID))
1709 		free_atid(csk);
1710 	else if (cxgbi_sock_flag(csk, CTPF_HAS_TID)) {
1711 		lldi = cxgbi_cdev_priv(csk->cdev);
1712 		cxgb4_remove_tid(lldi->tids, 0, csk->tid,
1713 				 csk->csk_family);
1714 		cxgbi_sock_clear_flag(csk, CTPF_HAS_TID);
1715 		cxgbi_sock_put(csk);
1716 	}
1717 	csk->dst = NULL;
1718 }
1719 
1720 #ifdef CONFIG_CHELSIO_T4_DCB
1721 static inline u8 get_iscsi_dcb_state(struct net_device *ndev)
1722 {
1723 	return ndev->dcbnl_ops->getstate(ndev);
1724 }
1725 
1726 static int select_priority(int pri_mask)
1727 {
1728 	if (!pri_mask)
1729 		return 0;
1730 	return (ffs(pri_mask) - 1);
1731 }
1732 
1733 static u8 get_iscsi_dcb_priority(struct net_device *ndev)
1734 {
1735 	int rv;
1736 	u8 caps;
1737 
1738 	struct dcb_app iscsi_dcb_app = {
1739 		.protocol = 3260
1740 	};
1741 
1742 	rv = (int)ndev->dcbnl_ops->getcap(ndev, DCB_CAP_ATTR_DCBX, &caps);
1743 	if (rv)
1744 		return 0;
1745 
1746 	if (caps & DCB_CAP_DCBX_VER_IEEE) {
1747 		iscsi_dcb_app.selector = IEEE_8021QAZ_APP_SEL_STREAM;
1748 		rv = dcb_ieee_getapp_mask(ndev, &iscsi_dcb_app);
1749 		if (!rv) {
1750 			iscsi_dcb_app.selector = IEEE_8021QAZ_APP_SEL_ANY;
1751 			rv = dcb_ieee_getapp_mask(ndev, &iscsi_dcb_app);
1752 		}
1753 	} else if (caps & DCB_CAP_DCBX_VER_CEE) {
1754 		iscsi_dcb_app.selector = DCB_APP_IDTYPE_PORTNUM;
1755 		rv = dcb_getapp(ndev, &iscsi_dcb_app);
1756 	}
1757 
1758 	log_debug(1 << CXGBI_DBG_ISCSI,
1759 		  "iSCSI priority is set to %u\n", select_priority(rv));
1760 	return select_priority(rv);
1761 }
1762 #endif
1763 
1764 static int init_act_open(struct cxgbi_sock *csk)
1765 {
1766 	struct cxgbi_device *cdev = csk->cdev;
1767 	struct cxgb4_lld_info *lldi = cxgbi_cdev_priv(cdev);
1768 	struct net_device *ndev = cdev->ports[csk->port_id];
1769 	struct sk_buff *skb = NULL;
1770 	struct neighbour *n = NULL;
1771 	void *daddr;
1772 	unsigned int step;
1773 	unsigned int rxq_idx;
1774 	unsigned int size, size6;
1775 	unsigned int linkspeed;
1776 	unsigned int rcv_winf, snd_winf;
1777 #ifdef CONFIG_CHELSIO_T4_DCB
1778 	u8 priority = 0;
1779 #endif
1780 	log_debug(1 << CXGBI_DBG_TOE | 1 << CXGBI_DBG_SOCK,
1781 		"csk 0x%p,%u,0x%lx,%u.\n",
1782 		csk, csk->state, csk->flags, csk->tid);
1783 
1784 	if (csk->csk_family == AF_INET)
1785 		daddr = &csk->daddr.sin_addr.s_addr;
1786 #if IS_ENABLED(CONFIG_IPV6)
1787 	else if (csk->csk_family == AF_INET6)
1788 		daddr = &csk->daddr6.sin6_addr;
1789 #endif
1790 	else {
1791 		pr_err("address family 0x%x not supported\n", csk->csk_family);
1792 		goto rel_resource;
1793 	}
1794 
1795 	n = dst_neigh_lookup(csk->dst, daddr);
1796 
1797 	if (!n) {
1798 		pr_err("%s, can't get neighbour of csk->dst.\n", ndev->name);
1799 		goto rel_resource;
1800 	}
1801 
1802 	if (!(n->nud_state & NUD_VALID))
1803 		neigh_event_send(n, NULL);
1804 
1805 	csk->atid = cxgb4_alloc_atid(lldi->tids, csk);
1806 	if (csk->atid < 0) {
1807 		pr_err("%s, NO atid available.\n", ndev->name);
1808 		goto rel_resource_without_clip;
1809 	}
1810 	cxgbi_sock_set_flag(csk, CTPF_HAS_ATID);
1811 	cxgbi_sock_get(csk);
1812 
1813 #ifdef CONFIG_CHELSIO_T4_DCB
1814 	if (get_iscsi_dcb_state(ndev))
1815 		priority = get_iscsi_dcb_priority(ndev);
1816 
1817 	csk->dcb_priority = priority;
1818 	csk->l2t = cxgb4_l2t_get(lldi->l2t, n, ndev, priority);
1819 #else
1820 	csk->l2t = cxgb4_l2t_get(lldi->l2t, n, ndev, 0);
1821 #endif
1822 	if (!csk->l2t) {
1823 		pr_err("%s, cannot alloc l2t.\n", ndev->name);
1824 		goto rel_resource_without_clip;
1825 	}
1826 	cxgbi_sock_get(csk);
1827 
1828 #if IS_ENABLED(CONFIG_IPV6)
1829 	if (csk->csk_family == AF_INET6)
1830 		cxgb4_clip_get(ndev, (const u32 *)&csk->saddr6.sin6_addr, 1);
1831 #endif
1832 
1833 	if (is_t4(lldi->adapter_type)) {
1834 		size = sizeof(struct cpl_act_open_req);
1835 		size6 = sizeof(struct cpl_act_open_req6);
1836 	} else if (is_t5(lldi->adapter_type)) {
1837 		size = sizeof(struct cpl_t5_act_open_req);
1838 		size6 = sizeof(struct cpl_t5_act_open_req6);
1839 	} else {
1840 		size = sizeof(struct cpl_t6_act_open_req);
1841 		size6 = sizeof(struct cpl_t6_act_open_req6);
1842 	}
1843 
1844 	if (csk->csk_family == AF_INET)
1845 		skb = alloc_wr(size, 0, GFP_NOIO);
1846 #if IS_ENABLED(CONFIG_IPV6)
1847 	else
1848 		skb = alloc_wr(size6, 0, GFP_NOIO);
1849 #endif
1850 
1851 	if (!skb)
1852 		goto rel_resource;
1853 	skb->sk = (struct sock *)csk;
1854 	t4_set_arp_err_handler(skb, csk, cxgbi_sock_act_open_req_arp_failure);
1855 
1856 	if (!csk->mtu)
1857 		csk->mtu = dst_mtu(csk->dst);
1858 	cxgb4_best_mtu(lldi->mtus, csk->mtu, &csk->mss_idx);
1859 	csk->tx_chan = cxgb4_port_chan(ndev);
1860 	csk->smac_idx = ((struct port_info *)netdev_priv(ndev))->smt_idx;
1861 	step = lldi->ntxq / lldi->nchan;
1862 	csk->txq_idx = cxgb4_port_idx(ndev) * step;
1863 	step = lldi->nrxq / lldi->nchan;
1864 	rxq_idx = (cxgb4_port_idx(ndev) * step) + (cdev->rxq_idx_cntr % step);
1865 	cdev->rxq_idx_cntr++;
1866 	csk->rss_qid = lldi->rxq_ids[rxq_idx];
1867 	linkspeed = ((struct port_info *)netdev_priv(ndev))->link_cfg.speed;
1868 	csk->snd_win = cxgb4i_snd_win;
1869 	csk->rcv_win = cxgb4i_rcv_win;
1870 	if (cxgb4i_rcv_win <= 0) {
1871 		csk->rcv_win = CXGB4I_DEFAULT_10G_RCV_WIN;
1872 		rcv_winf = linkspeed / SPEED_10000;
1873 		if (rcv_winf)
1874 			csk->rcv_win *= rcv_winf;
1875 	}
1876 	if (cxgb4i_snd_win <= 0) {
1877 		csk->snd_win = CXGB4I_DEFAULT_10G_SND_WIN;
1878 		snd_winf = linkspeed / SPEED_10000;
1879 		if (snd_winf)
1880 			csk->snd_win *= snd_winf;
1881 	}
1882 	csk->wr_cred = lldi->wr_cred -
1883 		       DIV_ROUND_UP(sizeof(struct cpl_abort_req), 16);
1884 	csk->wr_max_cred = csk->wr_cred;
1885 	csk->wr_una_cred = 0;
1886 	cxgbi_sock_reset_wr_list(csk);
1887 	csk->err = 0;
1888 
1889 	pr_info_ipaddr("csk 0x%p,%u,0x%lx,%u,%u,%u, mtu %u,%u, smac %u.\n",
1890 		       (&csk->saddr), (&csk->daddr), csk, csk->state,
1891 		       csk->flags, csk->tx_chan, csk->txq_idx, csk->rss_qid,
1892 		       csk->mtu, csk->mss_idx, csk->smac_idx);
1893 
1894 	/* must wait for either a act_open_rpl or act_open_establish */
1895 	if (!try_module_get(cdev->owner)) {
1896 		pr_err("%s, try_module_get failed.\n", ndev->name);
1897 		goto rel_resource;
1898 	}
1899 
1900 	cxgbi_sock_set_state(csk, CTP_ACTIVE_OPEN);
1901 	if (csk->csk_family == AF_INET)
1902 		send_act_open_req(csk, skb, csk->l2t);
1903 #if IS_ENABLED(CONFIG_IPV6)
1904 	else
1905 		send_act_open_req6(csk, skb, csk->l2t);
1906 #endif
1907 	neigh_release(n);
1908 
1909 	return 0;
1910 
1911 rel_resource:
1912 #if IS_ENABLED(CONFIG_IPV6)
1913 	if (csk->csk_family == AF_INET6)
1914 		cxgb4_clip_release(ndev,
1915 				   (const u32 *)&csk->saddr6.sin6_addr, 1);
1916 #endif
1917 rel_resource_without_clip:
1918 	if (n)
1919 		neigh_release(n);
1920 	if (skb)
1921 		__kfree_skb(skb);
1922 	return -EINVAL;
1923 }
1924 
1925 static cxgb4i_cplhandler_func cxgb4i_cplhandlers[NUM_CPL_CMDS] = {
1926 	[CPL_ACT_ESTABLISH] = do_act_establish,
1927 	[CPL_ACT_OPEN_RPL] = do_act_open_rpl,
1928 	[CPL_PEER_CLOSE] = do_peer_close,
1929 	[CPL_ABORT_REQ_RSS] = do_abort_req_rss,
1930 	[CPL_ABORT_RPL_RSS] = do_abort_rpl_rss,
1931 	[CPL_CLOSE_CON_RPL] = do_close_con_rpl,
1932 	[CPL_FW4_ACK] = do_fw4_ack,
1933 	[CPL_ISCSI_HDR] = do_rx_iscsi_hdr,
1934 	[CPL_ISCSI_DATA] = do_rx_iscsi_data,
1935 	[CPL_SET_TCB_RPL] = do_set_tcb_rpl,
1936 	[CPL_RX_DATA_DDP] = do_rx_data_ddp,
1937 	[CPL_RX_ISCSI_DDP] = do_rx_data_ddp,
1938 	[CPL_RX_ISCSI_CMP] = do_rx_iscsi_cmp,
1939 	[CPL_RX_DATA] = do_rx_data,
1940 };
1941 
1942 static int cxgb4i_ofld_init(struct cxgbi_device *cdev)
1943 {
1944 	int rc;
1945 
1946 	if (cxgb4i_max_connect > CXGB4I_MAX_CONN)
1947 		cxgb4i_max_connect = CXGB4I_MAX_CONN;
1948 
1949 	rc = cxgbi_device_portmap_create(cdev, cxgb4i_sport_base,
1950 					cxgb4i_max_connect);
1951 	if (rc < 0)
1952 		return rc;
1953 
1954 	cdev->csk_release_offload_resources = release_offload_resources;
1955 	cdev->csk_push_tx_frames = push_tx_frames;
1956 	cdev->csk_send_abort_req = send_abort_req;
1957 	cdev->csk_send_close_req = send_close_req;
1958 	cdev->csk_send_rx_credits = send_rx_credits;
1959 	cdev->csk_alloc_cpls = alloc_cpls;
1960 	cdev->csk_init_act_open = init_act_open;
1961 
1962 	pr_info("cdev 0x%p, offload up, added.\n", cdev);
1963 	return 0;
1964 }
1965 
1966 static inline void
1967 ulp_mem_io_set_hdr(struct cxgbi_device *cdev,
1968 		   struct ulp_mem_io *req,
1969 		   unsigned int wr_len, unsigned int dlen,
1970 		   unsigned int pm_addr,
1971 		   int tid)
1972 {
1973 	struct cxgb4_lld_info *lldi = cxgbi_cdev_priv(cdev);
1974 	struct ulptx_idata *idata = (struct ulptx_idata *)(req + 1);
1975 
1976 	INIT_ULPTX_WR(req, wr_len, 0, tid);
1977 	req->wr.wr_hi = htonl(FW_WR_OP_V(FW_ULPTX_WR) |
1978 		FW_WR_ATOMIC_V(0));
1979 	req->cmd = htonl(ULPTX_CMD_V(ULP_TX_MEM_WRITE) |
1980 		ULP_MEMIO_ORDER_V(is_t4(lldi->adapter_type)) |
1981 		T5_ULP_MEMIO_IMM_V(!is_t4(lldi->adapter_type)));
1982 	req->dlen = htonl(ULP_MEMIO_DATA_LEN_V(dlen >> 5));
1983 	req->lock_addr = htonl(ULP_MEMIO_ADDR_V(pm_addr >> 5));
1984 	req->len16 = htonl(DIV_ROUND_UP(wr_len - sizeof(req->wr), 16));
1985 
1986 	idata->cmd_more = htonl(ULPTX_CMD_V(ULP_TX_SC_IMM));
1987 	idata->len = htonl(dlen);
1988 }
1989 
1990 static struct sk_buff *
1991 ddp_ppod_init_idata(struct cxgbi_device *cdev,
1992 		    struct cxgbi_ppm *ppm,
1993 		    unsigned int idx, unsigned int npods,
1994 		    unsigned int tid)
1995 {
1996 	unsigned int pm_addr = (idx << PPOD_SIZE_SHIFT) + ppm->llimit;
1997 	unsigned int dlen = npods << PPOD_SIZE_SHIFT;
1998 	unsigned int wr_len = roundup(sizeof(struct ulp_mem_io) +
1999 				sizeof(struct ulptx_idata) + dlen, 16);
2000 	struct sk_buff *skb = alloc_wr(wr_len, 0, GFP_ATOMIC);
2001 
2002 	if (!skb) {
2003 		pr_err("%s: %s idx %u, npods %u, OOM.\n",
2004 		       __func__, ppm->ndev->name, idx, npods);
2005 		return NULL;
2006 	}
2007 
2008 	ulp_mem_io_set_hdr(cdev, (struct ulp_mem_io *)skb->head, wr_len, dlen,
2009 			   pm_addr, tid);
2010 
2011 	return skb;
2012 }
2013 
2014 static int ddp_ppod_write_idata(struct cxgbi_ppm *ppm, struct cxgbi_sock *csk,
2015 				struct cxgbi_task_tag_info *ttinfo,
2016 				unsigned int idx, unsigned int npods,
2017 				struct scatterlist **sg_pp,
2018 				unsigned int *sg_off)
2019 {
2020 	struct cxgbi_device *cdev = csk->cdev;
2021 	struct sk_buff *skb = ddp_ppod_init_idata(cdev, ppm, idx, npods,
2022 						  csk->tid);
2023 	struct ulp_mem_io *req;
2024 	struct ulptx_idata *idata;
2025 	struct cxgbi_pagepod *ppod;
2026 	int i;
2027 
2028 	if (!skb)
2029 		return -ENOMEM;
2030 
2031 	req = (struct ulp_mem_io *)skb->head;
2032 	idata = (struct ulptx_idata *)(req + 1);
2033 	ppod = (struct cxgbi_pagepod *)(idata + 1);
2034 
2035 	for (i = 0; i < npods; i++, ppod++)
2036 		cxgbi_ddp_set_one_ppod(ppod, ttinfo, sg_pp, sg_off);
2037 
2038 	cxgbi_skcb_set_flag(skb, SKCBF_TX_MEM_WRITE);
2039 	cxgbi_skcb_set_flag(skb, SKCBF_TX_FLAG_COMPL);
2040 	set_wr_txq(skb, CPL_PRIORITY_DATA, csk->port_id);
2041 
2042 	spin_lock_bh(&csk->lock);
2043 	cxgbi_sock_skb_entail(csk, skb);
2044 	spin_unlock_bh(&csk->lock);
2045 
2046 	return 0;
2047 }
2048 
2049 static int ddp_set_map(struct cxgbi_ppm *ppm, struct cxgbi_sock *csk,
2050 		       struct cxgbi_task_tag_info *ttinfo)
2051 {
2052 	unsigned int pidx = ttinfo->idx;
2053 	unsigned int npods = ttinfo->npods;
2054 	unsigned int i, cnt;
2055 	int err = 0;
2056 	struct scatterlist *sg = ttinfo->sgl;
2057 	unsigned int offset = 0;
2058 
2059 	ttinfo->cid = csk->port_id;
2060 
2061 	for (i = 0; i < npods; i += cnt, pidx += cnt) {
2062 		cnt = npods - i;
2063 
2064 		if (cnt > ULPMEM_IDATA_MAX_NPPODS)
2065 			cnt = ULPMEM_IDATA_MAX_NPPODS;
2066 		err = ddp_ppod_write_idata(ppm, csk, ttinfo, pidx, cnt,
2067 					   &sg, &offset);
2068 		if (err < 0)
2069 			break;
2070 	}
2071 
2072 	return err;
2073 }
2074 
2075 static int ddp_setup_conn_pgidx(struct cxgbi_sock *csk, unsigned int tid,
2076 				int pg_idx)
2077 {
2078 	struct sk_buff *skb;
2079 	struct cpl_set_tcb_field *req;
2080 
2081 	if (!pg_idx || pg_idx >= DDP_PGIDX_MAX)
2082 		return 0;
2083 
2084 	skb = alloc_wr(sizeof(*req), 0, GFP_KERNEL);
2085 	if (!skb)
2086 		return -ENOMEM;
2087 
2088 	/*  set up ulp page size */
2089 	req = (struct cpl_set_tcb_field *)skb->head;
2090 	INIT_TP_WR(req, csk->tid);
2091 	OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_SET_TCB_FIELD, csk->tid));
2092 	req->reply_ctrl = htons(NO_REPLY_V(0) | QUEUENO_V(csk->rss_qid));
2093 	req->word_cookie = htons(0);
2094 	req->mask = cpu_to_be64(0x3 << 8);
2095 	req->val = cpu_to_be64(pg_idx << 8);
2096 	set_wr_txq(skb, CPL_PRIORITY_CONTROL, csk->port_id);
2097 
2098 	log_debug(1 << CXGBI_DBG_TOE | 1 << CXGBI_DBG_SOCK,
2099 		"csk 0x%p, tid 0x%x, pg_idx %u.\n", csk, csk->tid, pg_idx);
2100 
2101 	reinit_completion(&csk->cmpl);
2102 	cxgb4_ofld_send(csk->cdev->ports[csk->port_id], skb);
2103 	wait_for_completion(&csk->cmpl);
2104 
2105 	return csk->err;
2106 }
2107 
2108 static int ddp_setup_conn_digest(struct cxgbi_sock *csk, unsigned int tid,
2109 				 int hcrc, int dcrc)
2110 {
2111 	struct sk_buff *skb;
2112 	struct cpl_set_tcb_field *req;
2113 
2114 	if (!hcrc && !dcrc)
2115 		return 0;
2116 
2117 	skb = alloc_wr(sizeof(*req), 0, GFP_KERNEL);
2118 	if (!skb)
2119 		return -ENOMEM;
2120 
2121 	csk->hcrc_len = (hcrc ? 4 : 0);
2122 	csk->dcrc_len = (dcrc ? 4 : 0);
2123 	/*  set up ulp submode */
2124 	req = (struct cpl_set_tcb_field *)skb->head;
2125 	INIT_TP_WR(req, tid);
2126 	OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_SET_TCB_FIELD, tid));
2127 	req->reply_ctrl = htons(NO_REPLY_V(0) | QUEUENO_V(csk->rss_qid));
2128 	req->word_cookie = htons(0);
2129 	req->mask = cpu_to_be64(0x3 << 4);
2130 	req->val = cpu_to_be64(((hcrc ? ULP_CRC_HEADER : 0) |
2131 				(dcrc ? ULP_CRC_DATA : 0)) << 4);
2132 	set_wr_txq(skb, CPL_PRIORITY_CONTROL, csk->port_id);
2133 
2134 	log_debug(1 << CXGBI_DBG_TOE | 1 << CXGBI_DBG_SOCK,
2135 		"csk 0x%p, tid 0x%x, crc %d,%d.\n", csk, csk->tid, hcrc, dcrc);
2136 
2137 	reinit_completion(&csk->cmpl);
2138 	cxgb4_ofld_send(csk->cdev->ports[csk->port_id], skb);
2139 	wait_for_completion(&csk->cmpl);
2140 
2141 	return csk->err;
2142 }
2143 
2144 static struct cxgbi_ppm *cdev2ppm(struct cxgbi_device *cdev)
2145 {
2146 	return (struct cxgbi_ppm *)(*((struct cxgb4_lld_info *)
2147 				       (cxgbi_cdev_priv(cdev)))->iscsi_ppm);
2148 }
2149 
2150 static int cxgb4i_ddp_init(struct cxgbi_device *cdev)
2151 {
2152 	struct cxgb4_lld_info *lldi = cxgbi_cdev_priv(cdev);
2153 	struct net_device *ndev = cdev->ports[0];
2154 	struct cxgbi_tag_format tformat;
2155 	int i, err;
2156 
2157 	if (!lldi->vr->iscsi.size) {
2158 		pr_warn("%s, iscsi NOT enabled, check config!\n", ndev->name);
2159 		return -EACCES;
2160 	}
2161 
2162 	cdev->flags |= CXGBI_FLAG_USE_PPOD_OFLDQ;
2163 
2164 	memset(&tformat, 0, sizeof(struct cxgbi_tag_format));
2165 	for (i = 0; i < 4; i++)
2166 		tformat.pgsz_order[i] = (lldi->iscsi_pgsz_order >> (i << 3))
2167 					 & 0xF;
2168 	cxgbi_tagmask_check(lldi->iscsi_tagmask, &tformat);
2169 
2170 	pr_info("iscsi_edram.start 0x%x iscsi_edram.size 0x%x",
2171 		lldi->vr->ppod_edram.start, lldi->vr->ppod_edram.size);
2172 
2173 	err = cxgbi_ddp_ppm_setup(lldi->iscsi_ppm, cdev, &tformat,
2174 				  lldi->vr->iscsi.size, lldi->iscsi_llimit,
2175 				  lldi->vr->iscsi.start, 2,
2176 				  lldi->vr->ppod_edram.start,
2177 				  lldi->vr->ppod_edram.size);
2178 
2179 	if (err < 0)
2180 		return err;
2181 
2182 	cdev->csk_ddp_setup_digest = ddp_setup_conn_digest;
2183 	cdev->csk_ddp_setup_pgidx = ddp_setup_conn_pgidx;
2184 	cdev->csk_ddp_set_map = ddp_set_map;
2185 	cdev->tx_max_size = min_t(unsigned int, ULP2_MAX_PDU_PAYLOAD,
2186 				  lldi->iscsi_iolen - ISCSI_PDU_NONPAYLOAD_LEN);
2187 	cdev->rx_max_size = min_t(unsigned int, ULP2_MAX_PDU_PAYLOAD,
2188 				  lldi->iscsi_iolen - ISCSI_PDU_NONPAYLOAD_LEN);
2189 	cdev->cdev2ppm = cdev2ppm;
2190 
2191 	return 0;
2192 }
2193 
2194 static bool is_memfree(struct adapter *adap)
2195 {
2196 	u32 io;
2197 
2198 	io = t4_read_reg(adap, MA_TARGET_MEM_ENABLE_A);
2199 	if (is_t5(adap->params.chip)) {
2200 		if ((io & EXT_MEM0_ENABLE_F) || (io & EXT_MEM1_ENABLE_F))
2201 			return false;
2202 	} else if (io & EXT_MEM_ENABLE_F) {
2203 		return false;
2204 	}
2205 
2206 	return true;
2207 }
2208 
2209 static void *t4_uld_add(const struct cxgb4_lld_info *lldi)
2210 {
2211 	struct cxgbi_device *cdev;
2212 	struct port_info *pi;
2213 	struct net_device *ndev;
2214 	struct adapter *adap;
2215 	struct tid_info *t;
2216 	u32 max_cmds = CXGB4I_SCSI_HOST_QDEPTH;
2217 	u32 max_conn = CXGBI_MAX_CONN;
2218 	int i, rc;
2219 
2220 	cdev = cxgbi_device_register(sizeof(*lldi), lldi->nports);
2221 	if (!cdev) {
2222 		pr_info("t4 device 0x%p, register failed.\n", lldi);
2223 		return NULL;
2224 	}
2225 	pr_info("0x%p,0x%x, ports %u,%s, chan %u, q %u,%u, wr %u.\n",
2226 		cdev, lldi->adapter_type, lldi->nports,
2227 		lldi->ports[0]->name, lldi->nchan, lldi->ntxq,
2228 		lldi->nrxq, lldi->wr_cred);
2229 	for (i = 0; i < lldi->nrxq; i++)
2230 		log_debug(1 << CXGBI_DBG_DEV,
2231 			"t4 0x%p, rxq id #%d: %u.\n",
2232 			cdev, i, lldi->rxq_ids[i]);
2233 
2234 	memcpy(cxgbi_cdev_priv(cdev), lldi, sizeof(*lldi));
2235 	cdev->flags = CXGBI_FLAG_DEV_T4;
2236 	cdev->pdev = lldi->pdev;
2237 	cdev->ports = lldi->ports;
2238 	cdev->nports = lldi->nports;
2239 	cdev->mtus = lldi->mtus;
2240 	cdev->nmtus = NMTUS;
2241 	cdev->rx_credit_thres = (CHELSIO_CHIP_VERSION(lldi->adapter_type) <=
2242 				 CHELSIO_T5) ? cxgb4i_rx_credit_thres : 0;
2243 	cdev->skb_tx_rsvd = CXGB4I_TX_HEADER_LEN;
2244 	cdev->skb_rx_extra = sizeof(struct cpl_iscsi_hdr);
2245 	cdev->itp = &cxgb4i_iscsi_transport;
2246 	cdev->owner = THIS_MODULE;
2247 
2248 	cdev->pfvf = FW_PFVF_CMD_PFN_V(lldi->pf);
2249 	pr_info("cdev 0x%p,%s, pfvf %u.\n",
2250 		cdev, lldi->ports[0]->name, cdev->pfvf);
2251 
2252 	rc = cxgb4i_ddp_init(cdev);
2253 	if (rc) {
2254 		pr_info("t4 0x%p ddp init failed %d.\n", cdev, rc);
2255 		goto err_out;
2256 	}
2257 
2258 	ndev = cdev->ports[0];
2259 	adap = netdev2adap(ndev);
2260 	if (adap) {
2261 		t = &adap->tids;
2262 		if (t->ntids <= CXGBI_MAX_CONN)
2263 			max_conn = t->ntids;
2264 
2265 		if (is_memfree(adap)) {
2266 			cdev->flags |=	CXGBI_FLAG_DEV_ISO_OFF;
2267 			max_cmds = CXGB4I_SCSI_HOST_QDEPTH >> 2;
2268 
2269 			pr_info("%s: 0x%p, tid %u, SO adapter.\n",
2270 				ndev->name, cdev, t->ntids);
2271 		}
2272 	} else {
2273 		pr_info("%s, 0x%p, NO adapter struct.\n", ndev->name, cdev);
2274 	}
2275 
2276 	/* ISO is enabled in T5/T6 firmware version >= 1.13.43.0 */
2277 	if (!is_t4(lldi->adapter_type) &&
2278 	    (lldi->fw_vers >= 0x10d2b00) &&
2279 	    !(cdev->flags & CXGBI_FLAG_DEV_ISO_OFF))
2280 		cdev->skb_iso_txhdr = sizeof(struct cpl_tx_data_iso);
2281 
2282 	rc = cxgb4i_ofld_init(cdev);
2283 	if (rc) {
2284 		pr_info("t4 0x%p ofld init failed.\n", cdev);
2285 		goto err_out;
2286 	}
2287 
2288 	cxgb4i_host_template.can_queue = max_cmds;
2289 	rc = cxgbi_hbas_add(cdev, CXGB4I_MAX_LUN, max_conn,
2290 			    &cxgb4i_host_template, cxgb4i_stt);
2291 	if (rc)
2292 		goto err_out;
2293 
2294 	for (i = 0; i < cdev->nports; i++) {
2295 		pi = netdev_priv(lldi->ports[i]);
2296 		cdev->hbas[i]->port_id = pi->port_id;
2297 	}
2298 	return cdev;
2299 
2300 err_out:
2301 	cxgbi_device_unregister(cdev);
2302 	return ERR_PTR(-ENOMEM);
2303 }
2304 
2305 #define RX_PULL_LEN	128
2306 static int t4_uld_rx_handler(void *handle, const __be64 *rsp,
2307 				const struct pkt_gl *pgl)
2308 {
2309 	const struct cpl_act_establish *rpl;
2310 	struct sk_buff *skb;
2311 	unsigned int opc;
2312 	struct cxgbi_device *cdev = handle;
2313 
2314 	if (pgl == NULL) {
2315 		unsigned int len = 64 - sizeof(struct rsp_ctrl) - 8;
2316 
2317 		skb = alloc_wr(len, 0, GFP_ATOMIC);
2318 		if (!skb)
2319 			goto nomem;
2320 		skb_copy_to_linear_data(skb, &rsp[1], len);
2321 	} else {
2322 		if (unlikely(*(u8 *)rsp != *(u8 *)pgl->va)) {
2323 			pr_info("? FL 0x%p,RSS%#llx,FL %#llx,len %u.\n",
2324 				pgl->va, be64_to_cpu(*rsp),
2325 				be64_to_cpu(*(u64 *)pgl->va),
2326 				pgl->tot_len);
2327 			return 0;
2328 		}
2329 		skb = cxgb4_pktgl_to_skb(pgl, RX_PULL_LEN, RX_PULL_LEN);
2330 		if (unlikely(!skb))
2331 			goto nomem;
2332 	}
2333 
2334 	rpl = (struct cpl_act_establish *)skb->data;
2335 	opc = rpl->ot.opcode;
2336 	log_debug(1 << CXGBI_DBG_TOE,
2337 		"cdev %p, opcode 0x%x(0x%x,0x%x), skb %p.\n",
2338 		 cdev, opc, rpl->ot.opcode_tid, ntohl(rpl->ot.opcode_tid), skb);
2339 	if (opc >= ARRAY_SIZE(cxgb4i_cplhandlers) || !cxgb4i_cplhandlers[opc]) {
2340 		pr_err("No handler for opcode 0x%x.\n", opc);
2341 		__kfree_skb(skb);
2342 	} else
2343 		cxgb4i_cplhandlers[opc](cdev, skb);
2344 
2345 	return 0;
2346 nomem:
2347 	log_debug(1 << CXGBI_DBG_TOE, "OOM bailing out.\n");
2348 	return 1;
2349 }
2350 
2351 static int t4_uld_state_change(void *handle, enum cxgb4_state state)
2352 {
2353 	struct cxgbi_device *cdev = handle;
2354 
2355 	switch (state) {
2356 	case CXGB4_STATE_UP:
2357 		pr_info("cdev 0x%p, UP.\n", cdev);
2358 		break;
2359 	case CXGB4_STATE_START_RECOVERY:
2360 		pr_info("cdev 0x%p, RECOVERY.\n", cdev);
2361 		/* close all connections */
2362 		break;
2363 	case CXGB4_STATE_DOWN:
2364 		pr_info("cdev 0x%p, DOWN.\n", cdev);
2365 		break;
2366 	case CXGB4_STATE_DETACH:
2367 		pr_info("cdev 0x%p, DETACH.\n", cdev);
2368 		cxgbi_device_unregister(cdev);
2369 		break;
2370 	default:
2371 		pr_info("cdev 0x%p, unknown state %d.\n", cdev, state);
2372 		break;
2373 	}
2374 	return 0;
2375 }
2376 
2377 #ifdef CONFIG_CHELSIO_T4_DCB
2378 static int
2379 cxgb4_dcb_change_notify(struct notifier_block *self, unsigned long val,
2380 			void *data)
2381 {
2382 	int i, port = 0xFF;
2383 	struct net_device *ndev;
2384 	struct cxgbi_device *cdev = NULL;
2385 	struct dcb_app_type *iscsi_app = data;
2386 	struct cxgbi_ports_map *pmap;
2387 	u8 priority;
2388 
2389 	if (iscsi_app->dcbx & DCB_CAP_DCBX_VER_IEEE) {
2390 		if ((iscsi_app->app.selector != IEEE_8021QAZ_APP_SEL_STREAM) &&
2391 		    (iscsi_app->app.selector != IEEE_8021QAZ_APP_SEL_ANY))
2392 			return NOTIFY_DONE;
2393 
2394 		priority = iscsi_app->app.priority;
2395 	} else if (iscsi_app->dcbx & DCB_CAP_DCBX_VER_CEE) {
2396 		if (iscsi_app->app.selector != DCB_APP_IDTYPE_PORTNUM)
2397 			return NOTIFY_DONE;
2398 
2399 		if (!iscsi_app->app.priority)
2400 			return NOTIFY_DONE;
2401 
2402 		priority = ffs(iscsi_app->app.priority) - 1;
2403 	} else {
2404 		return NOTIFY_DONE;
2405 	}
2406 
2407 	if (iscsi_app->app.protocol != 3260)
2408 		return NOTIFY_DONE;
2409 
2410 	log_debug(1 << CXGBI_DBG_ISCSI, "iSCSI priority for ifid %d is %u\n",
2411 		  iscsi_app->ifindex, priority);
2412 
2413 	ndev = dev_get_by_index(&init_net, iscsi_app->ifindex);
2414 	if (!ndev)
2415 		return NOTIFY_DONE;
2416 
2417 	cdev = cxgbi_device_find_by_netdev_rcu(ndev, &port);
2418 
2419 	dev_put(ndev);
2420 	if (!cdev)
2421 		return NOTIFY_DONE;
2422 
2423 	pmap = &cdev->pmap;
2424 
2425 	for (i = 0; i < pmap->used; i++) {
2426 		if (pmap->port_csk[i]) {
2427 			struct cxgbi_sock *csk = pmap->port_csk[i];
2428 
2429 			if (csk->dcb_priority != priority) {
2430 				iscsi_conn_failure(csk->user_data,
2431 						   ISCSI_ERR_CONN_FAILED);
2432 				pr_info("Restarting iSCSI connection %p with "
2433 					"priority %u->%u.\n", csk,
2434 					csk->dcb_priority, priority);
2435 			}
2436 		}
2437 	}
2438 	return NOTIFY_OK;
2439 }
2440 #endif
2441 
2442 static int __init cxgb4i_init_module(void)
2443 {
2444 	int rc;
2445 
2446 	printk(KERN_INFO "%s", version);
2447 
2448 	rc = cxgbi_iscsi_init(&cxgb4i_iscsi_transport, &cxgb4i_stt);
2449 	if (rc < 0)
2450 		return rc;
2451 	cxgb4_register_uld(CXGB4_ULD_ISCSI, &cxgb4i_uld_info);
2452 
2453 #ifdef CONFIG_CHELSIO_T4_DCB
2454 	pr_info("%s dcb enabled.\n", DRV_MODULE_NAME);
2455 	register_dcbevent_notifier(&cxgb4_dcb_change);
2456 #endif
2457 	return 0;
2458 }
2459 
2460 static void __exit cxgb4i_exit_module(void)
2461 {
2462 #ifdef CONFIG_CHELSIO_T4_DCB
2463 	unregister_dcbevent_notifier(&cxgb4_dcb_change);
2464 #endif
2465 	cxgb4_unregister_uld(CXGB4_ULD_ISCSI);
2466 	cxgbi_device_unregister_all(CXGBI_FLAG_DEV_T4);
2467 	cxgbi_iscsi_cleanup(&cxgb4i_iscsi_transport, &cxgb4i_stt);
2468 }
2469 
2470 module_init(cxgb4i_init_module);
2471 module_exit(cxgb4i_exit_module);
2472