xref: /linux/drivers/scsi/cxgbi/cxgb3i/cxgb3i.c (revision ca55b2fef3a9373fcfc30f82fd26bc7fccbda732)
1 /*
2  * cxgb3i_offload.c: Chelsio S3xx iscsi offloaded tcp connection management
3  *
4  * Copyright (C) 2003-2015 Chelsio Communications.  All rights reserved.
5  *
6  * This program is distributed in the hope that it will be useful, but WITHOUT
7  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
8  * FITNESS FOR A PARTICULAR PURPOSE.  See the LICENSE file included in this
9  * release for licensing terms and conditions.
10  *
11  * Written by:	Dimitris Michailidis (dm@chelsio.com)
12  *		Karen Xie (kxie@chelsio.com)
13  */
14 
15 #define pr_fmt(fmt) KBUILD_MODNAME ":%s: " fmt, __func__
16 
17 #include <linux/module.h>
18 #include <linux/moduleparam.h>
19 #include <scsi/scsi_host.h>
20 
21 #include "common.h"
22 #include "t3_cpl.h"
23 #include "t3cdev.h"
24 #include "cxgb3_defs.h"
25 #include "cxgb3_ctl_defs.h"
26 #include "cxgb3_offload.h"
27 #include "firmware_exports.h"
28 #include "cxgb3i.h"
29 
30 static unsigned int dbg_level;
31 #include "../libcxgbi.h"
32 
33 #define DRV_MODULE_NAME         "cxgb3i"
34 #define DRV_MODULE_DESC         "Chelsio T3 iSCSI Driver"
35 #define DRV_MODULE_VERSION	"2.0.1-ko"
36 #define DRV_MODULE_RELDATE	"Apr. 2015"
37 
38 static char version[] =
39 	DRV_MODULE_DESC " " DRV_MODULE_NAME
40 	" v" DRV_MODULE_VERSION " (" DRV_MODULE_RELDATE ")\n";
41 
42 MODULE_AUTHOR("Chelsio Communications, Inc.");
43 MODULE_DESCRIPTION(DRV_MODULE_DESC);
44 MODULE_VERSION(DRV_MODULE_VERSION);
45 MODULE_LICENSE("GPL");
46 
47 module_param(dbg_level, uint, 0644);
48 MODULE_PARM_DESC(dbg_level, "debug flag (default=0)");
49 
50 static int cxgb3i_rcv_win = 256 * 1024;
51 module_param(cxgb3i_rcv_win, int, 0644);
52 MODULE_PARM_DESC(cxgb3i_rcv_win, "TCP receive window in bytes (default=256KB)");
53 
54 static int cxgb3i_snd_win = 128 * 1024;
55 module_param(cxgb3i_snd_win, int, 0644);
56 MODULE_PARM_DESC(cxgb3i_snd_win, "TCP send window in bytes (default=128KB)");
57 
58 static int cxgb3i_rx_credit_thres = 10 * 1024;
59 module_param(cxgb3i_rx_credit_thres, int, 0644);
60 MODULE_PARM_DESC(rx_credit_thres,
61 		 "RX credits return threshold in bytes (default=10KB)");
62 
63 static unsigned int cxgb3i_max_connect = 8 * 1024;
64 module_param(cxgb3i_max_connect, uint, 0644);
65 MODULE_PARM_DESC(cxgb3i_max_connect, "Max. # of connections (default=8092)");
66 
67 static unsigned int cxgb3i_sport_base = 20000;
68 module_param(cxgb3i_sport_base, uint, 0644);
69 MODULE_PARM_DESC(cxgb3i_sport_base, "starting port number (default=20000)");
70 
71 static void cxgb3i_dev_open(struct t3cdev *);
72 static void cxgb3i_dev_close(struct t3cdev *);
73 static void cxgb3i_dev_event_handler(struct t3cdev *, u32, u32);
74 
75 static struct cxgb3_client t3_client = {
76 	.name = DRV_MODULE_NAME,
77 	.handlers = cxgb3i_cpl_handlers,
78 	.add = cxgb3i_dev_open,
79 	.remove = cxgb3i_dev_close,
80 	.event_handler = cxgb3i_dev_event_handler,
81 };
82 
83 static struct scsi_host_template cxgb3i_host_template = {
84 	.module		= THIS_MODULE,
85 	.name		= DRV_MODULE_NAME,
86 	.proc_name	= DRV_MODULE_NAME,
87 	.can_queue	= CXGB3I_SCSI_HOST_QDEPTH,
88 	.queuecommand	= iscsi_queuecommand,
89 	.change_queue_depth = scsi_change_queue_depth,
90 	.sg_tablesize	= SG_ALL,
91 	.max_sectors	= 0xFFFF,
92 	.cmd_per_lun	= ISCSI_DEF_CMD_PER_LUN,
93 	.eh_abort_handler = iscsi_eh_abort,
94 	.eh_device_reset_handler = iscsi_eh_device_reset,
95 	.eh_target_reset_handler = iscsi_eh_recover_target,
96 	.target_alloc	= iscsi_target_alloc,
97 	.use_clustering	= DISABLE_CLUSTERING,
98 	.this_id	= -1,
99 	.track_queue_depth = 1,
100 };
101 
102 static struct iscsi_transport cxgb3i_iscsi_transport = {
103 	.owner		= THIS_MODULE,
104 	.name		= DRV_MODULE_NAME,
105 	/* owner and name should be set already */
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 *cxgb3i_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 
154 static int push_tx_frames(struct cxgbi_sock *csk, int req_completion);
155 
156 static void send_act_open_req(struct cxgbi_sock *csk, struct sk_buff *skb,
157 			      const struct l2t_entry *e)
158 {
159 	unsigned int wscale = cxgbi_sock_compute_wscale(csk->rcv_win);
160 	struct cpl_act_open_req *req = (struct cpl_act_open_req *)skb->head;
161 
162 	skb->priority = CPL_PRIORITY_SETUP;
163 
164 	req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
165 	OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_ACT_OPEN_REQ, csk->atid));
166 	req->local_port = csk->saddr.sin_port;
167 	req->peer_port = csk->daddr.sin_port;
168 	req->local_ip = csk->saddr.sin_addr.s_addr;
169 	req->peer_ip = csk->daddr.sin_addr.s_addr;
170 
171 	req->opt0h = htonl(V_KEEP_ALIVE(1) | F_TCAM_BYPASS |
172 			V_WND_SCALE(wscale) | V_MSS_IDX(csk->mss_idx) |
173 			V_L2T_IDX(e->idx) | V_TX_CHANNEL(e->smt_idx));
174 	req->opt0l = htonl(V_ULP_MODE(ULP2_MODE_ISCSI) |
175 			V_RCV_BUFSIZ(csk->rcv_win >> 10));
176 
177 	log_debug(1 << CXGBI_DBG_TOE | 1 << CXGBI_DBG_SOCK,
178 		"csk 0x%p,%u,0x%lx,%u, %pI4:%u-%pI4:%u, %u,%u,%u.\n",
179 		csk, csk->state, csk->flags, csk->atid,
180 		&req->local_ip, ntohs(req->local_port),
181 		&req->peer_ip, ntohs(req->peer_port),
182 		csk->mss_idx, e->idx, e->smt_idx);
183 
184 	l2t_send(csk->cdev->lldev, skb, csk->l2t);
185 }
186 
187 static inline void act_open_arp_failure(struct t3cdev *dev, struct sk_buff *skb)
188 {
189 	cxgbi_sock_act_open_req_arp_failure(NULL, skb);
190 }
191 
192 /*
193  * CPL connection close request: host ->
194  *
195  * Close a connection by sending a CPL_CLOSE_CON_REQ message and queue it to
196  * the write queue (i.e., after any unsent txt data).
197  */
198 static void send_close_req(struct cxgbi_sock *csk)
199 {
200 	struct sk_buff *skb = csk->cpl_close;
201 	struct cpl_close_con_req *req = (struct cpl_close_con_req *)skb->head;
202 	unsigned int tid = csk->tid;
203 
204 	log_debug(1 << CXGBI_DBG_TOE | 1 << CXGBI_DBG_SOCK,
205 		"csk 0x%p,%u,0x%lx,%u.\n",
206 		csk, csk->state, csk->flags, csk->tid);
207 
208 	csk->cpl_close = NULL;
209 	req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_OFLD_CLOSE_CON));
210 	req->wr.wr_lo = htonl(V_WR_TID(tid));
211 	OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_CLOSE_CON_REQ, tid));
212 	req->rsvd = htonl(csk->write_seq);
213 
214 	cxgbi_sock_skb_entail(csk, skb);
215 	if (csk->state >= CTP_ESTABLISHED)
216 		push_tx_frames(csk, 1);
217 }
218 
219 /*
220  * CPL connection abort request: host ->
221  *
222  * Send an ABORT_REQ message. Makes sure we do not send multiple ABORT_REQs
223  * for the same connection and also that we do not try to send a message
224  * after the connection has closed.
225  */
226 static void abort_arp_failure(struct t3cdev *tdev, struct sk_buff *skb)
227 {
228 	struct cpl_abort_req *req = cplhdr(skb);
229 
230 	log_debug(1 << CXGBI_DBG_TOE | 1 << CXGBI_DBG_SOCK,
231 		"t3dev 0x%p, tid %u, skb 0x%p.\n",
232 		tdev, GET_TID(req), skb);
233 	req->cmd = CPL_ABORT_NO_RST;
234 	cxgb3_ofld_send(tdev, skb);
235 }
236 
237 static void send_abort_req(struct cxgbi_sock *csk)
238 {
239 	struct sk_buff *skb = csk->cpl_abort_req;
240 	struct cpl_abort_req *req;
241 
242 	if (unlikely(csk->state == CTP_ABORTING || !skb))
243 		return;
244 	cxgbi_sock_set_state(csk, CTP_ABORTING);
245 	cxgbi_sock_set_flag(csk, CTPF_ABORT_RPL_PENDING);
246 	/* Purge the send queue so we don't send anything after an abort. */
247 	cxgbi_sock_purge_write_queue(csk);
248 
249 	csk->cpl_abort_req = NULL;
250 	req = (struct cpl_abort_req *)skb->head;
251 	skb->priority = CPL_PRIORITY_DATA;
252 	set_arp_failure_handler(skb, abort_arp_failure);
253 	req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_OFLD_HOST_ABORT_CON_REQ));
254 	req->wr.wr_lo = htonl(V_WR_TID(csk->tid));
255 	OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_ABORT_REQ, csk->tid));
256 	req->rsvd0 = htonl(csk->snd_nxt);
257 	req->rsvd1 = !cxgbi_sock_flag(csk, CTPF_TX_DATA_SENT);
258 	req->cmd = CPL_ABORT_SEND_RST;
259 
260 	log_debug(1 << CXGBI_DBG_TOE | 1 << CXGBI_DBG_SOCK,
261 		"csk 0x%p,%u,0x%lx,%u, snd_nxt %u, 0x%x.\n",
262 		csk, csk->state, csk->flags, csk->tid, csk->snd_nxt,
263 		req->rsvd1);
264 
265 	l2t_send(csk->cdev->lldev, skb, csk->l2t);
266 }
267 
268 /*
269  * CPL connection abort reply: host ->
270  *
271  * Send an ABORT_RPL message in response of the ABORT_REQ received.
272  */
273 static void send_abort_rpl(struct cxgbi_sock *csk, int rst_status)
274 {
275 	struct sk_buff *skb = csk->cpl_abort_rpl;
276 	struct cpl_abort_rpl *rpl = (struct cpl_abort_rpl *)skb->head;
277 
278 	log_debug(1 << CXGBI_DBG_TOE | 1 << CXGBI_DBG_SOCK,
279 		"csk 0x%p,%u,0x%lx,%u, status %d.\n",
280 		csk, csk->state, csk->flags, csk->tid, rst_status);
281 
282 	csk->cpl_abort_rpl = NULL;
283 	skb->priority = CPL_PRIORITY_DATA;
284 	rpl->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_OFLD_HOST_ABORT_CON_RPL));
285 	rpl->wr.wr_lo = htonl(V_WR_TID(csk->tid));
286 	OPCODE_TID(rpl) = htonl(MK_OPCODE_TID(CPL_ABORT_RPL, csk->tid));
287 	rpl->cmd = rst_status;
288 	cxgb3_ofld_send(csk->cdev->lldev, skb);
289 }
290 
291 /*
292  * CPL connection rx data ack: host ->
293  * Send RX credits through an RX_DATA_ACK CPL message. Returns the number of
294  * credits sent.
295  */
296 static u32 send_rx_credits(struct cxgbi_sock *csk, u32 credits)
297 {
298 	struct sk_buff *skb;
299 	struct cpl_rx_data_ack *req;
300 	u32 dack = F_RX_DACK_CHANGE | V_RX_DACK_MODE(1);
301 
302 	log_debug(1 << CXGBI_DBG_TOE | 1 << CXGBI_DBG_PDU_RX,
303 		"csk 0x%p,%u,0x%lx,%u, credit %u, dack %u.\n",
304 		csk, csk->state, csk->flags, csk->tid, credits, dack);
305 
306 	skb = alloc_wr(sizeof(*req), 0, GFP_ATOMIC);
307 	if (!skb) {
308 		pr_info("csk 0x%p, credit %u, OOM.\n", csk, credits);
309 		return 0;
310 	}
311 	req = (struct cpl_rx_data_ack *)skb->head;
312 	req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
313 	OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_RX_DATA_ACK, csk->tid));
314 	req->credit_dack = htonl(F_RX_DACK_CHANGE | V_RX_DACK_MODE(1) |
315 				V_RX_CREDITS(credits));
316 	skb->priority = CPL_PRIORITY_ACK;
317 	cxgb3_ofld_send(csk->cdev->lldev, skb);
318 	return credits;
319 }
320 
321 /*
322  * CPL connection tx data: host ->
323  *
324  * Send iscsi PDU via TX_DATA CPL message. Returns the number of
325  * credits sent.
326  * Each TX_DATA consumes work request credit (wrs), so we need to keep track of
327  * how many we've used so far and how many are pending (i.e., yet ack'ed by T3).
328  */
329 
330 static unsigned int wrlen __read_mostly;
331 static unsigned int skb_wrs[SKB_WR_LIST_SIZE] __read_mostly;
332 
333 static void init_wr_tab(unsigned int wr_len)
334 {
335 	int i;
336 
337 	if (skb_wrs[1])		/* already initialized */
338 		return;
339 	for (i = 1; i < SKB_WR_LIST_SIZE; i++) {
340 		int sgl_len = (3 * i) / 2 + (i & 1);
341 
342 		sgl_len += 3;
343 		skb_wrs[i] = (sgl_len <= wr_len
344 			      ? 1 : 1 + (sgl_len - 2) / (wr_len - 1));
345 	}
346 	wrlen = wr_len * 8;
347 }
348 
349 static inline void make_tx_data_wr(struct cxgbi_sock *csk, struct sk_buff *skb,
350 				   int len, int req_completion)
351 {
352 	struct tx_data_wr *req;
353 	struct l2t_entry *l2t = csk->l2t;
354 
355 	skb_reset_transport_header(skb);
356 	req = (struct tx_data_wr *)__skb_push(skb, sizeof(*req));
357 	req->wr_hi = htonl(V_WR_OP(FW_WROPCODE_OFLD_TX_DATA) |
358 			(req_completion ? F_WR_COMPL : 0));
359 	req->wr_lo = htonl(V_WR_TID(csk->tid));
360 	/* len includes the length of any HW ULP additions */
361 	req->len = htonl(len);
362 	/* V_TX_ULP_SUBMODE sets both the mode and submode */
363 	req->flags = htonl(V_TX_ULP_SUBMODE(cxgbi_skcb_ulp_mode(skb)) |
364 			   V_TX_SHOVE((skb_peek(&csk->write_queue) ? 0 : 1)));
365 	req->sndseq = htonl(csk->snd_nxt);
366 	req->param = htonl(V_TX_PORT(l2t->smt_idx));
367 
368 	if (!cxgbi_sock_flag(csk, CTPF_TX_DATA_SENT)) {
369 		req->flags |= htonl(V_TX_ACK_PAGES(2) | F_TX_INIT |
370 				    V_TX_CPU_IDX(csk->rss_qid));
371 		/* sendbuffer is in units of 32KB. */
372 		req->param |= htonl(V_TX_SNDBUF(csk->snd_win >> 15));
373 		cxgbi_sock_set_flag(csk, CTPF_TX_DATA_SENT);
374 	}
375 }
376 
377 /**
378  * push_tx_frames -- start transmit
379  * @c3cn: the offloaded connection
380  * @req_completion: request wr_ack or not
381  *
382  * Prepends TX_DATA_WR or CPL_CLOSE_CON_REQ headers to buffers waiting in a
383  * connection's send queue and sends them on to T3.  Must be called with the
384  * connection's lock held.  Returns the amount of send buffer space that was
385  * freed as a result of sending queued data to T3.
386  */
387 
388 static void arp_failure_skb_discard(struct t3cdev *dev, struct sk_buff *skb)
389 {
390 	kfree_skb(skb);
391 }
392 
393 static int push_tx_frames(struct cxgbi_sock *csk, int req_completion)
394 {
395 	int total_size = 0;
396 	struct sk_buff *skb;
397 
398 	if (unlikely(csk->state < CTP_ESTABLISHED ||
399 		csk->state == CTP_CLOSE_WAIT_1 || csk->state >= CTP_ABORTING)) {
400 			log_debug(1 << CXGBI_DBG_TOE | 1 << CXGBI_DBG_PDU_TX,
401 				"csk 0x%p,%u,0x%lx,%u, in closing state.\n",
402 				csk, csk->state, csk->flags, csk->tid);
403 		return 0;
404 	}
405 
406 	while (csk->wr_cred && (skb = skb_peek(&csk->write_queue)) != NULL) {
407 		int len = skb->len;	/* length before skb_push */
408 		int frags = skb_shinfo(skb)->nr_frags + (len != skb->data_len);
409 		int wrs_needed = skb_wrs[frags];
410 
411 		if (wrs_needed > 1 && len + sizeof(struct tx_data_wr) <= wrlen)
412 			wrs_needed = 1;
413 
414 		WARN_ON(frags >= SKB_WR_LIST_SIZE || wrs_needed < 1);
415 
416 		if (csk->wr_cred < wrs_needed) {
417 			log_debug(1 << CXGBI_DBG_PDU_TX,
418 				"csk 0x%p, skb len %u/%u, frag %u, wr %d<%u.\n",
419 				csk, skb->len, skb->data_len, frags,
420 				wrs_needed, csk->wr_cred);
421 			break;
422 		}
423 
424 		__skb_unlink(skb, &csk->write_queue);
425 		skb->priority = CPL_PRIORITY_DATA;
426 		skb->csum = wrs_needed;	/* remember this until the WR_ACK */
427 		csk->wr_cred -= wrs_needed;
428 		csk->wr_una_cred += wrs_needed;
429 		cxgbi_sock_enqueue_wr(csk, skb);
430 
431 		log_debug(1 << CXGBI_DBG_TOE | 1 << CXGBI_DBG_PDU_TX,
432 			"csk 0x%p, enqueue, skb len %u/%u, frag %u, wr %d, "
433 			"left %u, unack %u.\n",
434 			csk, skb->len, skb->data_len, frags, skb->csum,
435 			csk->wr_cred, csk->wr_una_cred);
436 
437 		if (likely(cxgbi_skcb_test_flag(skb, SKCBF_TX_NEED_HDR))) {
438 			if ((req_completion &&
439 				csk->wr_una_cred == wrs_needed) ||
440 			     csk->wr_una_cred >= csk->wr_max_cred / 2) {
441 				req_completion = 1;
442 				csk->wr_una_cred = 0;
443 			}
444 			len += cxgbi_ulp_extra_len(cxgbi_skcb_ulp_mode(skb));
445 			make_tx_data_wr(csk, skb, len, req_completion);
446 			csk->snd_nxt += len;
447 			cxgbi_skcb_clear_flag(skb, SKCBF_TX_NEED_HDR);
448 		}
449 		total_size += skb->truesize;
450 		log_debug(1 << CXGBI_DBG_TOE | 1 << CXGBI_DBG_PDU_TX,
451 			"csk 0x%p, tid 0x%x, send skb 0x%p.\n",
452 			csk, csk->tid, skb);
453 		set_arp_failure_handler(skb, arp_failure_skb_discard);
454 		l2t_send(csk->cdev->lldev, skb, csk->l2t);
455 	}
456 	return total_size;
457 }
458 
459 /*
460  * Process a CPL_ACT_ESTABLISH message: -> host
461  * Updates connection state from an active establish CPL message.  Runs with
462  * the connection lock held.
463  */
464 
465 static inline void free_atid(struct cxgbi_sock *csk)
466 {
467 	if (cxgbi_sock_flag(csk, CTPF_HAS_ATID)) {
468 		cxgb3_free_atid(csk->cdev->lldev, csk->atid);
469 		cxgbi_sock_clear_flag(csk, CTPF_HAS_ATID);
470 		cxgbi_sock_put(csk);
471 	}
472 }
473 
474 static int do_act_establish(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
475 {
476 	struct cxgbi_sock *csk = ctx;
477 	struct cpl_act_establish *req = cplhdr(skb);
478 	unsigned int tid = GET_TID(req);
479 	unsigned int atid = G_PASS_OPEN_TID(ntohl(req->tos_tid));
480 	u32 rcv_isn = ntohl(req->rcv_isn);	/* real RCV_ISN + 1 */
481 
482 	log_debug(1 << CXGBI_DBG_TOE | 1 << CXGBI_DBG_SOCK,
483 		"atid 0x%x,tid 0x%x, csk 0x%p,%u,0x%lx, isn %u.\n",
484 		atid, atid, csk, csk->state, csk->flags, rcv_isn);
485 
486 	cxgbi_sock_get(csk);
487 	cxgbi_sock_set_flag(csk, CTPF_HAS_TID);
488 	csk->tid = tid;
489 	cxgb3_insert_tid(csk->cdev->lldev, &t3_client, csk, tid);
490 
491 	free_atid(csk);
492 
493 	csk->rss_qid = G_QNUM(ntohs(skb->csum));
494 
495 	spin_lock_bh(&csk->lock);
496 	if (csk->retry_timer.function) {
497 		del_timer(&csk->retry_timer);
498 		csk->retry_timer.function = NULL;
499 	}
500 
501 	if (unlikely(csk->state != CTP_ACTIVE_OPEN))
502 		pr_info("csk 0x%p,%u,0x%lx,%u, got EST.\n",
503 			csk, csk->state, csk->flags, csk->tid);
504 
505 	csk->copied_seq = csk->rcv_wup = csk->rcv_nxt = rcv_isn;
506 	if (csk->rcv_win > (M_RCV_BUFSIZ << 10))
507 		csk->rcv_wup -= csk->rcv_win - (M_RCV_BUFSIZ << 10);
508 
509 	cxgbi_sock_established(csk, ntohl(req->snd_isn), ntohs(req->tcp_opt));
510 
511 	if (unlikely(cxgbi_sock_flag(csk, CTPF_ACTIVE_CLOSE_NEEDED)))
512 		/* upper layer has requested closing */
513 		send_abort_req(csk);
514 	else {
515 		if (skb_queue_len(&csk->write_queue))
516 			push_tx_frames(csk, 1);
517 		cxgbi_conn_tx_open(csk);
518 	}
519 
520 	spin_unlock_bh(&csk->lock);
521 	__kfree_skb(skb);
522 	return 0;
523 }
524 
525 /*
526  * Process a CPL_ACT_OPEN_RPL message: -> host
527  * Handle active open failures.
528  */
529 static int act_open_rpl_status_to_errno(int status)
530 {
531 	switch (status) {
532 	case CPL_ERR_CONN_RESET:
533 		return -ECONNREFUSED;
534 	case CPL_ERR_ARP_MISS:
535 		return -EHOSTUNREACH;
536 	case CPL_ERR_CONN_TIMEDOUT:
537 		return -ETIMEDOUT;
538 	case CPL_ERR_TCAM_FULL:
539 		return -ENOMEM;
540 	case CPL_ERR_CONN_EXIST:
541 		return -EADDRINUSE;
542 	default:
543 		return -EIO;
544 	}
545 }
546 
547 static void act_open_retry_timer(unsigned long data)
548 {
549 	struct sk_buff *skb;
550 	struct cxgbi_sock *csk = (struct cxgbi_sock *)data;
551 
552 	log_debug(1 << CXGBI_DBG_TOE | 1 << CXGBI_DBG_SOCK,
553 		"csk 0x%p,%u,0x%lx,%u.\n",
554 		csk, csk->state, csk->flags, csk->tid);
555 
556 	cxgbi_sock_get(csk);
557 	spin_lock_bh(&csk->lock);
558 	skb = alloc_wr(sizeof(struct cpl_act_open_req), 0, GFP_ATOMIC);
559 	if (!skb)
560 		cxgbi_sock_fail_act_open(csk, -ENOMEM);
561 	else {
562 		skb->sk = (struct sock *)csk;
563 		set_arp_failure_handler(skb, act_open_arp_failure);
564 		send_act_open_req(csk, skb, csk->l2t);
565 	}
566 	spin_unlock_bh(&csk->lock);
567 	cxgbi_sock_put(csk);
568 }
569 
570 static int do_act_open_rpl(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
571 {
572 	struct cxgbi_sock *csk = ctx;
573 	struct cpl_act_open_rpl *rpl = cplhdr(skb);
574 
575 	pr_info("csk 0x%p,%u,0x%lx,%u, status %u, %pI4:%u-%pI4:%u.\n",
576 		csk, csk->state, csk->flags, csk->atid, rpl->status,
577 		&csk->saddr.sin_addr.s_addr, ntohs(csk->saddr.sin_port),
578 		&csk->daddr.sin_addr.s_addr, ntohs(csk->daddr.sin_port));
579 
580 	if (rpl->status != CPL_ERR_TCAM_FULL &&
581 	    rpl->status != CPL_ERR_CONN_EXIST &&
582 	    rpl->status != CPL_ERR_ARP_MISS)
583 		cxgb3_queue_tid_release(tdev, GET_TID(rpl));
584 
585 	cxgbi_sock_get(csk);
586 	spin_lock_bh(&csk->lock);
587 	if (rpl->status == CPL_ERR_CONN_EXIST &&
588 	    csk->retry_timer.function != act_open_retry_timer) {
589 		csk->retry_timer.function = act_open_retry_timer;
590 		mod_timer(&csk->retry_timer, jiffies + HZ / 2);
591 	} else
592 		cxgbi_sock_fail_act_open(csk,
593 				act_open_rpl_status_to_errno(rpl->status));
594 
595 	spin_unlock_bh(&csk->lock);
596 	cxgbi_sock_put(csk);
597 	__kfree_skb(skb);
598 	return 0;
599 }
600 
601 /*
602  * Process PEER_CLOSE CPL messages: -> host
603  * Handle peer FIN.
604  */
605 static int do_peer_close(struct t3cdev *cdev, struct sk_buff *skb, void *ctx)
606 {
607 	struct cxgbi_sock *csk = ctx;
608 
609 	log_debug(1 << CXGBI_DBG_TOE | 1 << CXGBI_DBG_SOCK,
610 		"csk 0x%p,%u,0x%lx,%u.\n",
611 		csk, csk->state, csk->flags, csk->tid);
612 
613 	cxgbi_sock_rcv_peer_close(csk);
614 	__kfree_skb(skb);
615 	return 0;
616 }
617 
618 /*
619  * Process CLOSE_CONN_RPL CPL message: -> host
620  * Process a peer ACK to our FIN.
621  */
622 static int do_close_con_rpl(struct t3cdev *cdev, struct sk_buff *skb,
623 			    void *ctx)
624 {
625 	struct cxgbi_sock *csk = ctx;
626 	struct cpl_close_con_rpl *rpl = cplhdr(skb);
627 
628 	log_debug(1 << CXGBI_DBG_TOE | 1 << CXGBI_DBG_SOCK,
629 		"csk 0x%p,%u,0x%lx,%u, snxt %u.\n",
630 		csk, csk->state, csk->flags, csk->tid, ntohl(rpl->snd_nxt));
631 
632 	cxgbi_sock_rcv_close_conn_rpl(csk, ntohl(rpl->snd_nxt));
633 	__kfree_skb(skb);
634 	return 0;
635 }
636 
637 /*
638  * Process ABORT_REQ_RSS CPL message: -> host
639  * Process abort requests.  If we are waiting for an ABORT_RPL we ignore this
640  * request except that we need to reply to it.
641  */
642 
643 static int abort_status_to_errno(struct cxgbi_sock *csk, int abort_reason,
644 				 int *need_rst)
645 {
646 	switch (abort_reason) {
647 	case CPL_ERR_BAD_SYN: /* fall through */
648 	case CPL_ERR_CONN_RESET:
649 		return csk->state > CTP_ESTABLISHED ? -EPIPE : -ECONNRESET;
650 	case CPL_ERR_XMIT_TIMEDOUT:
651 	case CPL_ERR_PERSIST_TIMEDOUT:
652 	case CPL_ERR_FINWAIT2_TIMEDOUT:
653 	case CPL_ERR_KEEPALIVE_TIMEDOUT:
654 		return -ETIMEDOUT;
655 	default:
656 		return -EIO;
657 	}
658 }
659 
660 static int do_abort_req(struct t3cdev *cdev, struct sk_buff *skb, void *ctx)
661 {
662 	const struct cpl_abort_req_rss *req = cplhdr(skb);
663 	struct cxgbi_sock *csk = ctx;
664 	int rst_status = CPL_ABORT_NO_RST;
665 
666 	log_debug(1 << CXGBI_DBG_TOE | 1 << CXGBI_DBG_SOCK,
667 		"csk 0x%p,%u,0x%lx,%u.\n",
668 		csk, csk->state, csk->flags, csk->tid);
669 
670 	if (req->status == CPL_ERR_RTX_NEG_ADVICE ||
671 	    req->status == CPL_ERR_PERSIST_NEG_ADVICE) {
672 		goto done;
673 	}
674 
675 	cxgbi_sock_get(csk);
676 	spin_lock_bh(&csk->lock);
677 
678 	if (!cxgbi_sock_flag(csk, CTPF_ABORT_REQ_RCVD)) {
679 		cxgbi_sock_set_flag(csk, CTPF_ABORT_REQ_RCVD);
680 		cxgbi_sock_set_state(csk, CTP_ABORTING);
681 		goto out;
682 	}
683 
684 	cxgbi_sock_clear_flag(csk, CTPF_ABORT_REQ_RCVD);
685 	send_abort_rpl(csk, rst_status);
686 
687 	if (!cxgbi_sock_flag(csk, CTPF_ABORT_RPL_PENDING)) {
688 		csk->err = abort_status_to_errno(csk, req->status, &rst_status);
689 		cxgbi_sock_closed(csk);
690 	}
691 
692 out:
693 	spin_unlock_bh(&csk->lock);
694 	cxgbi_sock_put(csk);
695 done:
696 	__kfree_skb(skb);
697 	return 0;
698 }
699 
700 /*
701  * Process ABORT_RPL_RSS CPL message: -> host
702  * Process abort replies.  We only process these messages if we anticipate
703  * them as the coordination between SW and HW in this area is somewhat lacking
704  * and sometimes we get ABORT_RPLs after we are done with the connection that
705  * originated the ABORT_REQ.
706  */
707 static int do_abort_rpl(struct t3cdev *cdev, struct sk_buff *skb, void *ctx)
708 {
709 	struct cpl_abort_rpl_rss *rpl = cplhdr(skb);
710 	struct cxgbi_sock *csk = ctx;
711 
712 	log_debug(1 << CXGBI_DBG_TOE | 1 << CXGBI_DBG_SOCK,
713 		"status 0x%x, csk 0x%p, s %u, 0x%lx.\n",
714 		rpl->status, csk, csk ? csk->state : 0,
715 		csk ? csk->flags : 0UL);
716 	/*
717 	 * Ignore replies to post-close aborts indicating that the abort was
718 	 * requested too late.  These connections are terminated when we get
719 	 * PEER_CLOSE or CLOSE_CON_RPL and by the time the abort_rpl_rss
720 	 * arrives the TID is either no longer used or it has been recycled.
721 	 */
722 	if (rpl->status == CPL_ERR_ABORT_FAILED)
723 		goto rel_skb;
724 	/*
725 	 * Sometimes we've already closed the connection, e.g., a post-close
726 	 * abort races with ABORT_REQ_RSS, the latter frees the connection
727 	 * expecting the ABORT_REQ will fail with CPL_ERR_ABORT_FAILED,
728 	 * but FW turns the ABORT_REQ into a regular one and so we get
729 	 * ABORT_RPL_RSS with status 0 and no connection.
730 	 */
731 	if (csk)
732 		cxgbi_sock_rcv_abort_rpl(csk);
733 rel_skb:
734 	__kfree_skb(skb);
735 	return 0;
736 }
737 
738 /*
739  * Process RX_ISCSI_HDR CPL message: -> host
740  * Handle received PDUs, the payload could be DDP'ed. If not, the payload
741  * follow after the bhs.
742  */
743 static int do_iscsi_hdr(struct t3cdev *t3dev, struct sk_buff *skb, void *ctx)
744 {
745 	struct cxgbi_sock *csk = ctx;
746 	struct cpl_iscsi_hdr *hdr_cpl = cplhdr(skb);
747 	struct cpl_iscsi_hdr_norss data_cpl;
748 	struct cpl_rx_data_ddp_norss ddp_cpl;
749 	unsigned int hdr_len, data_len, status;
750 	unsigned int len;
751 	int err;
752 
753 	log_debug(1 << CXGBI_DBG_TOE | 1 << CXGBI_DBG_PDU_RX,
754 		"csk 0x%p,%u,0x%lx,%u, skb 0x%p,%u.\n",
755 		csk, csk->state, csk->flags, csk->tid, skb, skb->len);
756 
757 	spin_lock_bh(&csk->lock);
758 
759 	if (unlikely(csk->state >= CTP_PASSIVE_CLOSE)) {
760 		log_debug(1 << CXGBI_DBG_TOE | 1 << CXGBI_DBG_SOCK,
761 			"csk 0x%p,%u,0x%lx,%u, bad state.\n",
762 			csk, csk->state, csk->flags, csk->tid);
763 		if (csk->state != CTP_ABORTING)
764 			goto abort_conn;
765 		else
766 			goto discard;
767 	}
768 
769 	cxgbi_skcb_tcp_seq(skb) = ntohl(hdr_cpl->seq);
770 	cxgbi_skcb_flags(skb) = 0;
771 
772 	skb_reset_transport_header(skb);
773 	__skb_pull(skb, sizeof(struct cpl_iscsi_hdr));
774 
775 	len = hdr_len = ntohs(hdr_cpl->len);
776 	/* msg coalesce is off or not enough data received */
777 	if (skb->len <= hdr_len) {
778 		pr_err("%s: tid %u, CPL_ISCSI_HDR, skb len %u < %u.\n",
779 			csk->cdev->ports[csk->port_id]->name, csk->tid,
780 			skb->len, hdr_len);
781 		goto abort_conn;
782 	}
783 	cxgbi_skcb_set_flag(skb, SKCBF_RX_COALESCED);
784 
785 	err = skb_copy_bits(skb, skb->len - sizeof(ddp_cpl), &ddp_cpl,
786 			    sizeof(ddp_cpl));
787 	if (err < 0) {
788 		pr_err("%s: tid %u, copy cpl_ddp %u-%zu failed %d.\n",
789 			csk->cdev->ports[csk->port_id]->name, csk->tid,
790 			skb->len, sizeof(ddp_cpl), err);
791 		goto abort_conn;
792 	}
793 
794 	cxgbi_skcb_set_flag(skb, SKCBF_RX_STATUS);
795 	cxgbi_skcb_rx_pdulen(skb) = ntohs(ddp_cpl.len);
796 	cxgbi_skcb_rx_ddigest(skb) = ntohl(ddp_cpl.ulp_crc);
797 	status = ntohl(ddp_cpl.ddp_status);
798 
799 	log_debug(1 << CXGBI_DBG_TOE | 1 << CXGBI_DBG_PDU_RX,
800 		"csk 0x%p, skb 0x%p,%u, pdulen %u, status 0x%x.\n",
801 		csk, skb, skb->len, cxgbi_skcb_rx_pdulen(skb), status);
802 
803 	if (status & (1 << CPL_RX_DDP_STATUS_HCRC_SHIFT))
804 		cxgbi_skcb_set_flag(skb, SKCBF_RX_HCRC_ERR);
805 	if (status & (1 << CPL_RX_DDP_STATUS_DCRC_SHIFT))
806 		cxgbi_skcb_set_flag(skb, SKCBF_RX_DCRC_ERR);
807 	if (status & (1 << CPL_RX_DDP_STATUS_PAD_SHIFT))
808 		cxgbi_skcb_set_flag(skb, SKCBF_RX_PAD_ERR);
809 
810 	if (skb->len > (hdr_len + sizeof(ddp_cpl))) {
811 		err = skb_copy_bits(skb, hdr_len, &data_cpl, sizeof(data_cpl));
812 		if (err < 0) {
813 			pr_err("%s: tid %u, cp %zu/%u failed %d.\n",
814 				csk->cdev->ports[csk->port_id]->name,
815 				csk->tid, sizeof(data_cpl), skb->len, err);
816 			goto abort_conn;
817 		}
818 		data_len = ntohs(data_cpl.len);
819 		log_debug(1 << CXGBI_DBG_DDP | 1 << CXGBI_DBG_PDU_RX,
820 			"skb 0x%p, pdu not ddp'ed %u/%u, status 0x%x.\n",
821 			skb, data_len, cxgbi_skcb_rx_pdulen(skb), status);
822 		len += sizeof(data_cpl) + data_len;
823 	} else if (status & (1 << CPL_RX_DDP_STATUS_DDP_SHIFT))
824 		cxgbi_skcb_set_flag(skb, SKCBF_RX_DATA_DDPD);
825 
826 	csk->rcv_nxt = ntohl(ddp_cpl.seq) + cxgbi_skcb_rx_pdulen(skb);
827 	__pskb_trim(skb, len);
828 	__skb_queue_tail(&csk->receive_queue, skb);
829 	cxgbi_conn_pdu_ready(csk);
830 
831 	spin_unlock_bh(&csk->lock);
832 	return 0;
833 
834 abort_conn:
835 	send_abort_req(csk);
836 discard:
837 	spin_unlock_bh(&csk->lock);
838 	__kfree_skb(skb);
839 	return 0;
840 }
841 
842 /*
843  * Process TX_DATA_ACK CPL messages: -> host
844  * Process an acknowledgment of WR completion.  Advance snd_una and send the
845  * next batch of work requests from the write queue.
846  */
847 static int do_wr_ack(struct t3cdev *cdev, struct sk_buff *skb, void *ctx)
848 {
849 	struct cxgbi_sock *csk = ctx;
850 	struct cpl_wr_ack *hdr = cplhdr(skb);
851 
852 	log_debug(1 << CXGBI_DBG_TOE | 1 << CXGBI_DBG_PDU_RX,
853 		"csk 0x%p,%u,0x%lx,%u, cr %u.\n",
854 		csk, csk->state, csk->flags, csk->tid, ntohs(hdr->credits));
855 
856 	cxgbi_sock_rcv_wr_ack(csk, ntohs(hdr->credits), ntohl(hdr->snd_una), 1);
857 	__kfree_skb(skb);
858 	return 0;
859 }
860 
861 /*
862  * for each connection, pre-allocate skbs needed for close/abort requests. So
863  * that we can service the request right away.
864  */
865 static int alloc_cpls(struct cxgbi_sock *csk)
866 {
867 	csk->cpl_close = alloc_wr(sizeof(struct cpl_close_con_req), 0,
868 					GFP_KERNEL);
869 	if (!csk->cpl_close)
870 		return -ENOMEM;
871 	csk->cpl_abort_req = alloc_wr(sizeof(struct cpl_abort_req), 0,
872 					GFP_KERNEL);
873 	if (!csk->cpl_abort_req)
874 		goto free_cpl_skbs;
875 
876 	csk->cpl_abort_rpl = alloc_wr(sizeof(struct cpl_abort_rpl), 0,
877 					GFP_KERNEL);
878 	if (!csk->cpl_abort_rpl)
879 		goto free_cpl_skbs;
880 
881 	return 0;
882 
883 free_cpl_skbs:
884 	cxgbi_sock_free_cpl_skbs(csk);
885 	return -ENOMEM;
886 }
887 
888 /**
889  * release_offload_resources - release offload resource
890  * @c3cn: the offloaded iscsi tcp connection.
891  * Release resources held by an offload connection (TID, L2T entry, etc.)
892  */
893 static void l2t_put(struct cxgbi_sock *csk)
894 {
895 	struct t3cdev *t3dev = (struct t3cdev *)csk->cdev->lldev;
896 
897 	if (csk->l2t) {
898 		l2t_release(t3dev, csk->l2t);
899 		csk->l2t = NULL;
900 		cxgbi_sock_put(csk);
901 	}
902 }
903 
904 static void release_offload_resources(struct cxgbi_sock *csk)
905 {
906 	struct t3cdev *t3dev = (struct t3cdev *)csk->cdev->lldev;
907 
908 	log_debug(1 << CXGBI_DBG_TOE | 1 << CXGBI_DBG_SOCK,
909 		"csk 0x%p,%u,0x%lx,%u.\n",
910 		csk, csk->state, csk->flags, csk->tid);
911 
912 	csk->rss_qid = 0;
913 	cxgbi_sock_free_cpl_skbs(csk);
914 
915 	if (csk->wr_cred != csk->wr_max_cred) {
916 		cxgbi_sock_purge_wr_queue(csk);
917 		cxgbi_sock_reset_wr_list(csk);
918 	}
919 	l2t_put(csk);
920 	if (cxgbi_sock_flag(csk, CTPF_HAS_ATID))
921 		free_atid(csk);
922 	else if (cxgbi_sock_flag(csk, CTPF_HAS_TID)) {
923 		cxgb3_remove_tid(t3dev, (void *)csk, csk->tid);
924 		cxgbi_sock_clear_flag(csk, CTPF_HAS_TID);
925 		cxgbi_sock_put(csk);
926 	}
927 	csk->dst = NULL;
928 	csk->cdev = NULL;
929 }
930 
931 static void update_address(struct cxgbi_hba *chba)
932 {
933 	if (chba->ipv4addr) {
934 		if (chba->vdev &&
935 		    chba->ipv4addr != cxgb3i_get_private_ipv4addr(chba->vdev)) {
936 			cxgb3i_set_private_ipv4addr(chba->vdev, chba->ipv4addr);
937 			cxgb3i_set_private_ipv4addr(chba->ndev, 0);
938 			pr_info("%s set %pI4.\n",
939 				chba->vdev->name, &chba->ipv4addr);
940 		} else if (chba->ipv4addr !=
941 				cxgb3i_get_private_ipv4addr(chba->ndev)) {
942 			cxgb3i_set_private_ipv4addr(chba->ndev, chba->ipv4addr);
943 			pr_info("%s set %pI4.\n",
944 				chba->ndev->name, &chba->ipv4addr);
945 		}
946 	} else if (cxgb3i_get_private_ipv4addr(chba->ndev)) {
947 		if (chba->vdev)
948 			cxgb3i_set_private_ipv4addr(chba->vdev, 0);
949 		cxgb3i_set_private_ipv4addr(chba->ndev, 0);
950 	}
951 }
952 
953 static int init_act_open(struct cxgbi_sock *csk)
954 {
955 	struct dst_entry *dst = csk->dst;
956 	struct cxgbi_device *cdev = csk->cdev;
957 	struct t3cdev *t3dev = (struct t3cdev *)cdev->lldev;
958 	struct net_device *ndev = cdev->ports[csk->port_id];
959 	struct cxgbi_hba *chba = cdev->hbas[csk->port_id];
960 	struct sk_buff *skb = NULL;
961 
962 	log_debug(1 << CXGBI_DBG_TOE | 1 << CXGBI_DBG_SOCK,
963 		"csk 0x%p,%u,0x%lx.\n", csk, csk->state, csk->flags);
964 
965 	update_address(chba);
966 	if (chba->ipv4addr)
967 		csk->saddr.sin_addr.s_addr = chba->ipv4addr;
968 
969 	csk->rss_qid = 0;
970 	csk->l2t = t3_l2t_get(t3dev, dst, ndev,
971 			      &csk->daddr.sin_addr.s_addr);
972 	if (!csk->l2t) {
973 		pr_err("NO l2t available.\n");
974 		return -EINVAL;
975 	}
976 	cxgbi_sock_get(csk);
977 
978 	csk->atid = cxgb3_alloc_atid(t3dev, &t3_client, csk);
979 	if (csk->atid < 0) {
980 		pr_err("NO atid available.\n");
981 		goto rel_resource;
982 	}
983 	cxgbi_sock_set_flag(csk, CTPF_HAS_ATID);
984 	cxgbi_sock_get(csk);
985 
986 	skb = alloc_wr(sizeof(struct cpl_act_open_req), 0, GFP_KERNEL);
987 	if (!skb)
988 		goto rel_resource;
989 	skb->sk = (struct sock *)csk;
990 	set_arp_failure_handler(skb, act_open_arp_failure);
991 	csk->snd_win = cxgb3i_snd_win;
992 	csk->rcv_win = cxgb3i_rcv_win;
993 
994 	csk->wr_max_cred = csk->wr_cred = T3C_DATA(t3dev)->max_wrs - 1;
995 	csk->wr_una_cred = 0;
996 	csk->mss_idx = cxgbi_sock_select_mss(csk, dst_mtu(dst));
997 	cxgbi_sock_reset_wr_list(csk);
998 	csk->err = 0;
999 
1000 	log_debug(1 << CXGBI_DBG_TOE | 1 << CXGBI_DBG_SOCK,
1001 		"csk 0x%p,%u,0x%lx, %pI4:%u-%pI4:%u.\n",
1002 		csk, csk->state, csk->flags,
1003 		&csk->saddr.sin_addr.s_addr, ntohs(csk->saddr.sin_port),
1004 		&csk->daddr.sin_addr.s_addr, ntohs(csk->daddr.sin_port));
1005 
1006 	cxgbi_sock_set_state(csk, CTP_ACTIVE_OPEN);
1007 	send_act_open_req(csk, skb, csk->l2t);
1008 	return 0;
1009 
1010 rel_resource:
1011 	if (skb)
1012 		__kfree_skb(skb);
1013 	return -EINVAL;
1014 }
1015 
1016 cxgb3_cpl_handler_func cxgb3i_cpl_handlers[NUM_CPL_CMDS] = {
1017 	[CPL_ACT_ESTABLISH] = do_act_establish,
1018 	[CPL_ACT_OPEN_RPL] = do_act_open_rpl,
1019 	[CPL_PEER_CLOSE] = do_peer_close,
1020 	[CPL_ABORT_REQ_RSS] = do_abort_req,
1021 	[CPL_ABORT_RPL_RSS] = do_abort_rpl,
1022 	[CPL_CLOSE_CON_RPL] = do_close_con_rpl,
1023 	[CPL_TX_DMA_ACK] = do_wr_ack,
1024 	[CPL_ISCSI_HDR] = do_iscsi_hdr,
1025 };
1026 
1027 /**
1028  * cxgb3i_ofld_init - allocate and initialize resources for each adapter found
1029  * @cdev:	cxgbi adapter
1030  */
1031 int cxgb3i_ofld_init(struct cxgbi_device *cdev)
1032 {
1033 	struct t3cdev *t3dev = (struct t3cdev *)cdev->lldev;
1034 	struct adap_ports port;
1035 	struct ofld_page_info rx_page_info;
1036 	unsigned int wr_len;
1037 	int rc;
1038 
1039 	if (t3dev->ctl(t3dev, GET_WR_LEN, &wr_len) < 0 ||
1040 	    t3dev->ctl(t3dev, GET_PORTS, &port) < 0 ||
1041 	    t3dev->ctl(t3dev, GET_RX_PAGE_INFO, &rx_page_info) < 0) {
1042 		pr_warn("t3 0x%p, offload up, ioctl failed.\n", t3dev);
1043 		return -EINVAL;
1044 	}
1045 
1046 	if (cxgb3i_max_connect > CXGBI_MAX_CONN)
1047 		cxgb3i_max_connect = CXGBI_MAX_CONN;
1048 
1049 	rc = cxgbi_device_portmap_create(cdev, cxgb3i_sport_base,
1050 					cxgb3i_max_connect);
1051 	if (rc < 0)
1052 		return rc;
1053 
1054 	init_wr_tab(wr_len);
1055 	cdev->csk_release_offload_resources = release_offload_resources;
1056 	cdev->csk_push_tx_frames = push_tx_frames;
1057 	cdev->csk_send_abort_req = send_abort_req;
1058 	cdev->csk_send_close_req = send_close_req;
1059 	cdev->csk_send_rx_credits = send_rx_credits;
1060 	cdev->csk_alloc_cpls = alloc_cpls;
1061 	cdev->csk_init_act_open = init_act_open;
1062 
1063 	pr_info("cdev 0x%p, offload up, added.\n", cdev);
1064 	return 0;
1065 }
1066 
1067 /*
1068  * functions to program the pagepod in h/w
1069  */
1070 static inline void ulp_mem_io_set_hdr(struct sk_buff *skb, unsigned int addr)
1071 {
1072 	struct ulp_mem_io *req = (struct ulp_mem_io *)skb->head;
1073 
1074 	memset(req, 0, sizeof(*req));
1075 
1076 	req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_BYPASS));
1077 	req->cmd_lock_addr = htonl(V_ULP_MEMIO_ADDR(addr >> 5) |
1078 				   V_ULPTX_CMD(ULP_MEM_WRITE));
1079 	req->len = htonl(V_ULP_MEMIO_DATA_LEN(PPOD_SIZE >> 5) |
1080 			 V_ULPTX_NFLITS((PPOD_SIZE >> 3) + 1));
1081 }
1082 
1083 static int ddp_set_map(struct cxgbi_sock *csk, struct cxgbi_pagepod_hdr *hdr,
1084 			unsigned int idx, unsigned int npods,
1085 				struct cxgbi_gather_list *gl)
1086 {
1087 	struct cxgbi_device *cdev = csk->cdev;
1088 	struct cxgbi_ddp_info *ddp = cdev->ddp;
1089 	unsigned int pm_addr = (idx << PPOD_SIZE_SHIFT) + ddp->llimit;
1090 	int i;
1091 
1092 	log_debug(1 << CXGBI_DBG_DDP,
1093 		"csk 0x%p, idx %u, npods %u, gl 0x%p.\n",
1094 		csk, idx, npods, gl);
1095 
1096 	for (i = 0; i < npods; i++, idx++, pm_addr += PPOD_SIZE) {
1097 		struct sk_buff *skb = alloc_wr(sizeof(struct ulp_mem_io) +
1098 						PPOD_SIZE, 0, GFP_ATOMIC);
1099 
1100 		if (!skb)
1101 			return -ENOMEM;
1102 
1103 		ulp_mem_io_set_hdr(skb, pm_addr);
1104 		cxgbi_ddp_ppod_set((struct cxgbi_pagepod *)(skb->head +
1105 					sizeof(struct ulp_mem_io)),
1106 				   hdr, gl, i * PPOD_PAGES_MAX);
1107 		skb->priority = CPL_PRIORITY_CONTROL;
1108 		cxgb3_ofld_send(cdev->lldev, skb);
1109 	}
1110 	return 0;
1111 }
1112 
1113 static void ddp_clear_map(struct cxgbi_hba *chba, unsigned int tag,
1114 			  unsigned int idx, unsigned int npods)
1115 {
1116 	struct cxgbi_device *cdev = chba->cdev;
1117 	struct cxgbi_ddp_info *ddp = cdev->ddp;
1118 	unsigned int pm_addr = (idx << PPOD_SIZE_SHIFT) + ddp->llimit;
1119 	int i;
1120 
1121 	log_debug(1 << CXGBI_DBG_DDP,
1122 		"cdev 0x%p, idx %u, npods %u, tag 0x%x.\n",
1123 		cdev, idx, npods, tag);
1124 
1125 	for (i = 0; i < npods; i++, idx++, pm_addr += PPOD_SIZE) {
1126 		struct sk_buff *skb = alloc_wr(sizeof(struct ulp_mem_io) +
1127 						PPOD_SIZE, 0, GFP_ATOMIC);
1128 
1129 		if (!skb) {
1130 			pr_err("tag 0x%x, 0x%x, %d/%u, skb OOM.\n",
1131 				tag, idx, i, npods);
1132 			continue;
1133 		}
1134 		ulp_mem_io_set_hdr(skb, pm_addr);
1135 		skb->priority = CPL_PRIORITY_CONTROL;
1136 		cxgb3_ofld_send(cdev->lldev, skb);
1137 	}
1138 }
1139 
1140 static int ddp_setup_conn_pgidx(struct cxgbi_sock *csk,
1141 				       unsigned int tid, int pg_idx, bool reply)
1142 {
1143 	struct sk_buff *skb = alloc_wr(sizeof(struct cpl_set_tcb_field), 0,
1144 					GFP_KERNEL);
1145 	struct cpl_set_tcb_field *req;
1146 	u64 val = pg_idx < DDP_PGIDX_MAX ? pg_idx : 0;
1147 
1148 	log_debug(1 << CXGBI_DBG_DDP,
1149 		"csk 0x%p, tid %u, pg_idx %d.\n", csk, tid, pg_idx);
1150 	if (!skb)
1151 		return -ENOMEM;
1152 
1153 	/* set up ulp submode and page size */
1154 	req = (struct cpl_set_tcb_field *)skb->head;
1155 	req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
1156 	OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_SET_TCB_FIELD, tid));
1157 	req->reply = V_NO_REPLY(reply ? 0 : 1);
1158 	req->cpu_idx = 0;
1159 	req->word = htons(31);
1160 	req->mask = cpu_to_be64(0xF0000000);
1161 	req->val = cpu_to_be64(val << 28);
1162 	skb->priority = CPL_PRIORITY_CONTROL;
1163 
1164 	cxgb3_ofld_send(csk->cdev->lldev, skb);
1165 	return 0;
1166 }
1167 
1168 /**
1169  * cxgb3i_setup_conn_digest - setup conn. digest setting
1170  * @csk: cxgb tcp socket
1171  * @tid: connection id
1172  * @hcrc: header digest enabled
1173  * @dcrc: data digest enabled
1174  * @reply: request reply from h/w
1175  * set up the iscsi digest settings for a connection identified by tid
1176  */
1177 static int ddp_setup_conn_digest(struct cxgbi_sock *csk, unsigned int tid,
1178 			     int hcrc, int dcrc, int reply)
1179 {
1180 	struct sk_buff *skb = alloc_wr(sizeof(struct cpl_set_tcb_field), 0,
1181 					GFP_KERNEL);
1182 	struct cpl_set_tcb_field *req;
1183 	u64 val = (hcrc ? 1 : 0) | (dcrc ? 2 : 0);
1184 
1185 	log_debug(1 << CXGBI_DBG_DDP,
1186 		"csk 0x%p, tid %u, crc %d,%d.\n", csk, tid, hcrc, dcrc);
1187 	if (!skb)
1188 		return -ENOMEM;
1189 
1190 	/* set up ulp submode and page size */
1191 	req = (struct cpl_set_tcb_field *)skb->head;
1192 	req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
1193 	OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_SET_TCB_FIELD, tid));
1194 	req->reply = V_NO_REPLY(reply ? 0 : 1);
1195 	req->cpu_idx = 0;
1196 	req->word = htons(31);
1197 	req->mask = cpu_to_be64(0x0F000000);
1198 	req->val = cpu_to_be64(val << 24);
1199 	skb->priority = CPL_PRIORITY_CONTROL;
1200 
1201 	cxgb3_ofld_send(csk->cdev->lldev, skb);
1202 	return 0;
1203 }
1204 
1205 /**
1206  * t3_ddp_cleanup - release the cxgb3 adapter's ddp resource
1207  * @cdev: cxgb3i adapter
1208  * release all the resource held by the ddp pagepod manager for a given
1209  * adapter if needed
1210  */
1211 
1212 static void t3_ddp_cleanup(struct cxgbi_device *cdev)
1213 {
1214 	struct t3cdev *tdev = (struct t3cdev *)cdev->lldev;
1215 
1216 	if (cxgbi_ddp_cleanup(cdev)) {
1217 		pr_info("t3dev 0x%p, ulp_iscsi no more user.\n", tdev);
1218 		tdev->ulp_iscsi = NULL;
1219 	}
1220 }
1221 
1222 /**
1223  * ddp_init - initialize the cxgb3 adapter's ddp resource
1224  * @cdev: cxgb3i adapter
1225  * initialize the ddp pagepod manager for a given adapter
1226  */
1227 static int cxgb3i_ddp_init(struct cxgbi_device *cdev)
1228 {
1229 	struct t3cdev *tdev = (struct t3cdev *)cdev->lldev;
1230 	struct cxgbi_ddp_info *ddp = tdev->ulp_iscsi;
1231 	struct ulp_iscsi_info uinfo;
1232 	unsigned int pgsz_factor[4];
1233 	int i, err;
1234 
1235 	if (ddp) {
1236 		kref_get(&ddp->refcnt);
1237 		pr_warn("t3dev 0x%p, ddp 0x%p already set up.\n",
1238 			tdev, tdev->ulp_iscsi);
1239 		cdev->ddp = ddp;
1240 		return -EALREADY;
1241 	}
1242 
1243 	err = tdev->ctl(tdev, ULP_ISCSI_GET_PARAMS, &uinfo);
1244 	if (err < 0) {
1245 		pr_err("%s, failed to get iscsi param err=%d.\n",
1246 			 tdev->name, err);
1247 		return err;
1248 	}
1249 
1250 	err = cxgbi_ddp_init(cdev, uinfo.llimit, uinfo.ulimit,
1251 			uinfo.max_txsz, uinfo.max_rxsz);
1252 	if (err < 0)
1253 		return err;
1254 
1255 	ddp = cdev->ddp;
1256 
1257 	uinfo.tagmask = ddp->idx_mask << PPOD_IDX_SHIFT;
1258 	cxgbi_ddp_page_size_factor(pgsz_factor);
1259 	for (i = 0; i < 4; i++)
1260 		uinfo.pgsz_factor[i] = pgsz_factor[i];
1261 	uinfo.ulimit = uinfo.llimit + (ddp->nppods << PPOD_SIZE_SHIFT);
1262 
1263 	err = tdev->ctl(tdev, ULP_ISCSI_SET_PARAMS, &uinfo);
1264 	if (err < 0) {
1265 		pr_warn("%s unable to set iscsi param err=%d, ddp disabled.\n",
1266 			tdev->name, err);
1267 		cxgbi_ddp_cleanup(cdev);
1268 		return err;
1269 	}
1270 	tdev->ulp_iscsi = ddp;
1271 
1272 	cdev->csk_ddp_setup_digest = ddp_setup_conn_digest;
1273 	cdev->csk_ddp_setup_pgidx = ddp_setup_conn_pgidx;
1274 	cdev->csk_ddp_set = ddp_set_map;
1275 	cdev->csk_ddp_clear = ddp_clear_map;
1276 
1277 	pr_info("tdev 0x%p, nppods %u, bits %u, mask 0x%x,0x%x pkt %u/%u, "
1278 		"%u/%u.\n",
1279 		tdev, ddp->nppods, ddp->idx_bits, ddp->idx_mask,
1280 		ddp->rsvd_tag_mask, ddp->max_txsz, uinfo.max_txsz,
1281 		ddp->max_rxsz, uinfo.max_rxsz);
1282 	return 0;
1283 }
1284 
1285 static void cxgb3i_dev_close(struct t3cdev *t3dev)
1286 {
1287 	struct cxgbi_device *cdev = cxgbi_device_find_by_lldev(t3dev);
1288 
1289 	if (!cdev || cdev->flags & CXGBI_FLAG_ADAPTER_RESET) {
1290 		pr_info("0x%p close, f 0x%x.\n", cdev, cdev ? cdev->flags : 0);
1291 		return;
1292 	}
1293 
1294 	cxgbi_device_unregister(cdev);
1295 }
1296 
1297 /**
1298  * cxgb3i_dev_open - init a t3 adapter structure and any h/w settings
1299  * @t3dev: t3cdev adapter
1300  */
1301 static void cxgb3i_dev_open(struct t3cdev *t3dev)
1302 {
1303 	struct cxgbi_device *cdev = cxgbi_device_find_by_lldev(t3dev);
1304 	struct adapter *adapter = tdev2adap(t3dev);
1305 	int i, err;
1306 
1307 	if (cdev) {
1308 		pr_info("0x%p, updating.\n", cdev);
1309 		return;
1310 	}
1311 
1312 	cdev = cxgbi_device_register(0, adapter->params.nports);
1313 	if (!cdev) {
1314 		pr_warn("device 0x%p register failed.\n", t3dev);
1315 		return;
1316 	}
1317 
1318 	cdev->flags = CXGBI_FLAG_DEV_T3 | CXGBI_FLAG_IPV4_SET;
1319 	cdev->lldev = t3dev;
1320 	cdev->pdev = adapter->pdev;
1321 	cdev->ports = adapter->port;
1322 	cdev->nports = adapter->params.nports;
1323 	cdev->mtus = adapter->params.mtus;
1324 	cdev->nmtus = NMTUS;
1325 	cdev->rx_credit_thres = cxgb3i_rx_credit_thres;
1326 	cdev->skb_tx_rsvd = CXGB3I_TX_HEADER_LEN;
1327 	cdev->skb_rx_extra = sizeof(struct cpl_iscsi_hdr_norss);
1328 	cdev->dev_ddp_cleanup = t3_ddp_cleanup;
1329 	cdev->itp = &cxgb3i_iscsi_transport;
1330 
1331 	err = cxgb3i_ddp_init(cdev);
1332 	if (err) {
1333 		pr_info("0x%p ddp init failed\n", cdev);
1334 		goto err_out;
1335 	}
1336 
1337 	err = cxgb3i_ofld_init(cdev);
1338 	if (err) {
1339 		pr_info("0x%p offload init failed\n", cdev);
1340 		goto err_out;
1341 	}
1342 
1343 	err = cxgbi_hbas_add(cdev, CXGB3I_MAX_LUN, CXGBI_MAX_CONN,
1344 				&cxgb3i_host_template, cxgb3i_stt);
1345 	if (err)
1346 		goto err_out;
1347 
1348 	for (i = 0; i < cdev->nports; i++)
1349 		cdev->hbas[i]->ipv4addr =
1350 			cxgb3i_get_private_ipv4addr(cdev->ports[i]);
1351 
1352 	pr_info("cdev 0x%p, f 0x%x, t3dev 0x%p open, err %d.\n",
1353 		cdev, cdev ? cdev->flags : 0, t3dev, err);
1354 	return;
1355 
1356 err_out:
1357 	cxgbi_device_unregister(cdev);
1358 }
1359 
1360 static void cxgb3i_dev_event_handler(struct t3cdev *t3dev, u32 event, u32 port)
1361 {
1362 	struct cxgbi_device *cdev = cxgbi_device_find_by_lldev(t3dev);
1363 
1364 	log_debug(1 << CXGBI_DBG_TOE,
1365 		"0x%p, cdev 0x%p, event 0x%x, port 0x%x.\n",
1366 		t3dev, cdev, event, port);
1367 	if (!cdev)
1368 		return;
1369 
1370 	switch (event) {
1371 	case OFFLOAD_STATUS_DOWN:
1372 		cdev->flags |= CXGBI_FLAG_ADAPTER_RESET;
1373 		break;
1374 	case OFFLOAD_STATUS_UP:
1375 		cdev->flags &= ~CXGBI_FLAG_ADAPTER_RESET;
1376 		break;
1377 	}
1378 }
1379 
1380 /**
1381  * cxgb3i_init_module - module init entry point
1382  *
1383  * initialize any driver wide global data structures and register itself
1384  *	with the cxgb3 module
1385  */
1386 static int __init cxgb3i_init_module(void)
1387 {
1388 	int rc;
1389 
1390 	printk(KERN_INFO "%s", version);
1391 
1392 	rc = cxgbi_iscsi_init(&cxgb3i_iscsi_transport, &cxgb3i_stt);
1393 	if (rc < 0)
1394 		return rc;
1395 
1396 	cxgb3_register_client(&t3_client);
1397 	return 0;
1398 }
1399 
1400 /**
1401  * cxgb3i_exit_module - module cleanup/exit entry point
1402  *
1403  * go through the driver hba list and for each hba, release any resource held.
1404  *	and unregisters iscsi transport and the cxgb3 module
1405  */
1406 static void __exit cxgb3i_exit_module(void)
1407 {
1408 	cxgb3_unregister_client(&t3_client);
1409 	cxgbi_device_unregister_all(CXGBI_FLAG_DEV_T3);
1410 	cxgbi_iscsi_cleanup(&cxgb3i_iscsi_transport, &cxgb3i_stt);
1411 }
1412 
1413 module_init(cxgb3i_init_module);
1414 module_exit(cxgb3i_exit_module);
1415