xref: /linux/drivers/scsi/cxgbi/cxgb3i/cxgb3i.c (revision 8c994eff8fcfe8ecb1f1dbebed25b4d7bb75be12)
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(cxgb3i_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 const 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_timed_out	= iscsi_eh_cmd_timed_out,
94 	.eh_abort_handler = iscsi_eh_abort,
95 	.eh_device_reset_handler = iscsi_eh_device_reset,
96 	.eh_target_reset_handler = iscsi_eh_recover_target,
97 	.target_alloc	= iscsi_target_alloc,
98 	.dma_boundary	= PAGE_SIZE - 1,
99 	.this_id	= -1,
100 	.track_queue_depth = 1,
101 	.cmd_size	= sizeof(struct iscsi_cmd),
102 };
103 
104 static struct iscsi_transport cxgb3i_iscsi_transport = {
105 	.owner		= THIS_MODULE,
106 	.name		= DRV_MODULE_NAME,
107 	/* owner and name should be set already */
108 	.caps		= CAP_RECOVERY_L0 | CAP_MULTI_R2T | CAP_HDRDGST
109 				| CAP_DATADGST | CAP_DIGEST_OFFLOAD |
110 				CAP_PADDING_OFFLOAD | CAP_TEXT_NEGO,
111 	.attr_is_visible	= cxgbi_attr_is_visible,
112 	.get_host_param	= cxgbi_get_host_param,
113 	.set_host_param	= cxgbi_set_host_param,
114 	/* session management */
115 	.create_session	= cxgbi_create_session,
116 	.destroy_session	= cxgbi_destroy_session,
117 	.get_session_param = iscsi_session_get_param,
118 	/* connection management */
119 	.create_conn	= cxgbi_create_conn,
120 	.bind_conn	= cxgbi_bind_conn,
121 	.unbind_conn	= iscsi_conn_unbind,
122 	.destroy_conn	= iscsi_tcp_conn_teardown,
123 	.start_conn	= iscsi_conn_start,
124 	.stop_conn	= iscsi_conn_stop,
125 	.get_conn_param	= iscsi_conn_get_param,
126 	.set_param	= cxgbi_set_conn_param,
127 	.get_stats	= cxgbi_get_conn_stats,
128 	/* pdu xmit req from user space */
129 	.send_pdu	= iscsi_conn_send_pdu,
130 	/* task */
131 	.init_task	= iscsi_tcp_task_init,
132 	.xmit_task	= iscsi_tcp_task_xmit,
133 	.cleanup_task	= cxgbi_cleanup_task,
134 	/* pdu */
135 	.alloc_pdu	= cxgbi_conn_alloc_pdu,
136 	.init_pdu	= cxgbi_conn_init_pdu,
137 	.xmit_pdu	= cxgbi_conn_xmit_pdu,
138 	.parse_pdu_itt	= cxgbi_parse_pdu_itt,
139 	/* TCP connect/disconnect */
140 	.get_ep_param	= cxgbi_get_ep_param,
141 	.ep_connect	= cxgbi_ep_connect,
142 	.ep_poll	= cxgbi_ep_poll,
143 	.ep_disconnect	= cxgbi_ep_disconnect,
144 	/* Error recovery timeout call */
145 	.session_recovery_timedout = iscsi_session_recovery_timedout,
146 };
147 
148 static struct scsi_transport_template *cxgb3i_stt;
149 
150 /*
151  * CPL (Chelsio Protocol Language) defines a message passing interface between
152  * the host driver and Chelsio asic.
153  * The section below implments CPLs that related to iscsi tcp connection
154  * open/close/abort and data send/receive.
155  */
156 
157 static int push_tx_frames(struct cxgbi_sock *csk, int req_completion);
158 
159 static void send_act_open_req(struct cxgbi_sock *csk, struct sk_buff *skb,
160 			      const struct l2t_entry *e)
161 {
162 	unsigned int wscale = cxgbi_sock_compute_wscale(csk->rcv_win);
163 	struct cpl_act_open_req *req = (struct cpl_act_open_req *)skb->head;
164 
165 	skb->priority = CPL_PRIORITY_SETUP;
166 
167 	req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
168 	OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_ACT_OPEN_REQ, csk->atid));
169 	req->local_port = csk->saddr.sin_port;
170 	req->peer_port = csk->daddr.sin_port;
171 	req->local_ip = csk->saddr.sin_addr.s_addr;
172 	req->peer_ip = csk->daddr.sin_addr.s_addr;
173 
174 	req->opt0h = htonl(V_KEEP_ALIVE(1) | F_TCAM_BYPASS |
175 			V_WND_SCALE(wscale) | V_MSS_IDX(csk->mss_idx) |
176 			V_L2T_IDX(e->idx) | V_TX_CHANNEL(e->smt_idx));
177 	req->opt0l = htonl(V_ULP_MODE(ULP2_MODE_ISCSI) |
178 			V_RCV_BUFSIZ(csk->rcv_win >> 10));
179 
180 	log_debug(1 << CXGBI_DBG_TOE | 1 << CXGBI_DBG_SOCK,
181 		"csk 0x%p,%u,0x%lx,%u, %pI4:%u-%pI4:%u, %u,%u,%u.\n",
182 		csk, csk->state, csk->flags, csk->atid,
183 		&req->local_ip, ntohs(req->local_port),
184 		&req->peer_ip, ntohs(req->peer_port),
185 		csk->mss_idx, e->idx, e->smt_idx);
186 
187 	l2t_send(csk->cdev->lldev, skb, csk->l2t);
188 }
189 
190 static inline void act_open_arp_failure(struct t3cdev *dev, struct sk_buff *skb)
191 {
192 	cxgbi_sock_act_open_req_arp_failure(NULL, skb);
193 }
194 
195 /*
196  * CPL connection close request: host ->
197  *
198  * Close a connection by sending a CPL_CLOSE_CON_REQ message and queue it to
199  * the write queue (i.e., after any unsent txt data).
200  */
201 static void send_close_req(struct cxgbi_sock *csk)
202 {
203 	struct sk_buff *skb = csk->cpl_close;
204 	struct cpl_close_con_req *req = (struct cpl_close_con_req *)skb->head;
205 	unsigned int tid = csk->tid;
206 
207 	log_debug(1 << CXGBI_DBG_TOE | 1 << CXGBI_DBG_SOCK,
208 		"csk 0x%p,%u,0x%lx,%u.\n",
209 		csk, csk->state, csk->flags, csk->tid);
210 
211 	csk->cpl_close = NULL;
212 	req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_OFLD_CLOSE_CON));
213 	req->wr.wr_lo = htonl(V_WR_TID(tid));
214 	OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_CLOSE_CON_REQ, tid));
215 	req->rsvd = htonl(csk->write_seq);
216 
217 	cxgbi_sock_skb_entail(csk, skb);
218 	if (csk->state >= CTP_ESTABLISHED)
219 		push_tx_frames(csk, 1);
220 }
221 
222 /*
223  * CPL connection abort request: host ->
224  *
225  * Send an ABORT_REQ message. Makes sure we do not send multiple ABORT_REQs
226  * for the same connection and also that we do not try to send a message
227  * after the connection has closed.
228  */
229 static void abort_arp_failure(struct t3cdev *tdev, struct sk_buff *skb)
230 {
231 	struct cpl_abort_req *req = cplhdr(skb);
232 
233 	log_debug(1 << CXGBI_DBG_TOE | 1 << CXGBI_DBG_SOCK,
234 		"t3dev 0x%p, tid %u, skb 0x%p.\n",
235 		tdev, GET_TID(req), skb);
236 	req->cmd = CPL_ABORT_NO_RST;
237 	cxgb3_ofld_send(tdev, skb);
238 }
239 
240 static void send_abort_req(struct cxgbi_sock *csk)
241 {
242 	struct sk_buff *skb = csk->cpl_abort_req;
243 	struct cpl_abort_req *req;
244 
245 	if (unlikely(csk->state == CTP_ABORTING || !skb))
246 		return;
247 	cxgbi_sock_set_state(csk, CTP_ABORTING);
248 	cxgbi_sock_set_flag(csk, CTPF_ABORT_RPL_PENDING);
249 	/* Purge the send queue so we don't send anything after an abort. */
250 	cxgbi_sock_purge_write_queue(csk);
251 
252 	csk->cpl_abort_req = NULL;
253 	req = (struct cpl_abort_req *)skb->head;
254 	skb->priority = CPL_PRIORITY_DATA;
255 	set_arp_failure_handler(skb, abort_arp_failure);
256 	req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_OFLD_HOST_ABORT_CON_REQ));
257 	req->wr.wr_lo = htonl(V_WR_TID(csk->tid));
258 	OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_ABORT_REQ, csk->tid));
259 	req->rsvd0 = htonl(csk->snd_nxt);
260 	req->rsvd1 = !cxgbi_sock_flag(csk, CTPF_TX_DATA_SENT);
261 	req->cmd = CPL_ABORT_SEND_RST;
262 
263 	log_debug(1 << CXGBI_DBG_TOE | 1 << CXGBI_DBG_SOCK,
264 		"csk 0x%p,%u,0x%lx,%u, snd_nxt %u, 0x%x.\n",
265 		csk, csk->state, csk->flags, csk->tid, csk->snd_nxt,
266 		req->rsvd1);
267 
268 	l2t_send(csk->cdev->lldev, skb, csk->l2t);
269 }
270 
271 /*
272  * CPL connection abort reply: host ->
273  *
274  * Send an ABORT_RPL message in response of the ABORT_REQ received.
275  */
276 static void send_abort_rpl(struct cxgbi_sock *csk, int rst_status)
277 {
278 	struct sk_buff *skb = csk->cpl_abort_rpl;
279 	struct cpl_abort_rpl *rpl = (struct cpl_abort_rpl *)skb->head;
280 
281 	log_debug(1 << CXGBI_DBG_TOE | 1 << CXGBI_DBG_SOCK,
282 		"csk 0x%p,%u,0x%lx,%u, status %d.\n",
283 		csk, csk->state, csk->flags, csk->tid, rst_status);
284 
285 	csk->cpl_abort_rpl = NULL;
286 	skb->priority = CPL_PRIORITY_DATA;
287 	rpl->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_OFLD_HOST_ABORT_CON_RPL));
288 	rpl->wr.wr_lo = htonl(V_WR_TID(csk->tid));
289 	OPCODE_TID(rpl) = htonl(MK_OPCODE_TID(CPL_ABORT_RPL, csk->tid));
290 	rpl->cmd = rst_status;
291 	cxgb3_ofld_send(csk->cdev->lldev, skb);
292 }
293 
294 /*
295  * CPL connection rx data ack: host ->
296  * Send RX credits through an RX_DATA_ACK CPL message. Returns the number of
297  * credits sent.
298  */
299 static u32 send_rx_credits(struct cxgbi_sock *csk, u32 credits)
300 {
301 	struct sk_buff *skb;
302 	struct cpl_rx_data_ack *req;
303 	u32 dack = F_RX_DACK_CHANGE | V_RX_DACK_MODE(1);
304 
305 	log_debug(1 << CXGBI_DBG_TOE | 1 << CXGBI_DBG_PDU_RX,
306 		"csk 0x%p,%u,0x%lx,%u, credit %u, dack %u.\n",
307 		csk, csk->state, csk->flags, csk->tid, credits, dack);
308 
309 	skb = alloc_wr(sizeof(*req), 0, GFP_ATOMIC);
310 	if (!skb) {
311 		pr_info("csk 0x%p, credit %u, OOM.\n", csk, credits);
312 		return 0;
313 	}
314 	req = (struct cpl_rx_data_ack *)skb->head;
315 	req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
316 	OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_RX_DATA_ACK, csk->tid));
317 	req->credit_dack = htonl(F_RX_DACK_CHANGE | V_RX_DACK_MODE(1) |
318 				V_RX_CREDITS(credits));
319 	skb->priority = CPL_PRIORITY_ACK;
320 	cxgb3_ofld_send(csk->cdev->lldev, skb);
321 	return credits;
322 }
323 
324 /*
325  * CPL connection tx data: host ->
326  *
327  * Send iscsi PDU via TX_DATA CPL message. Returns the number of
328  * credits sent.
329  * Each TX_DATA consumes work request credit (wrs), so we need to keep track of
330  * how many we've used so far and how many are pending (i.e., yet ack'ed by T3).
331  */
332 
333 static unsigned int wrlen __read_mostly;
334 static unsigned int skb_wrs[SKB_WR_LIST_SIZE] __read_mostly;
335 
336 static void init_wr_tab(unsigned int wr_len)
337 {
338 	int i;
339 
340 	if (skb_wrs[1])		/* already initialized */
341 		return;
342 	for (i = 1; i < SKB_WR_LIST_SIZE; i++) {
343 		int sgl_len = (3 * i) / 2 + (i & 1);
344 
345 		sgl_len += 3;
346 		skb_wrs[i] = (sgl_len <= wr_len
347 			      ? 1 : 1 + (sgl_len - 2) / (wr_len - 1));
348 	}
349 	wrlen = wr_len * 8;
350 }
351 
352 static inline void make_tx_data_wr(struct cxgbi_sock *csk, struct sk_buff *skb,
353 				   int len, int req_completion)
354 {
355 	struct tx_data_wr *req;
356 	struct l2t_entry *l2t = csk->l2t;
357 
358 	skb_reset_transport_header(skb);
359 	req = __skb_push(skb, sizeof(*req));
360 	req->wr_hi = htonl(V_WR_OP(FW_WROPCODE_OFLD_TX_DATA) |
361 			(req_completion ? F_WR_COMPL : 0));
362 	req->wr_lo = htonl(V_WR_TID(csk->tid));
363 	/* len includes the length of any HW ULP additions */
364 	req->len = htonl(len);
365 	/* V_TX_ULP_SUBMODE sets both the mode and submode */
366 	req->flags = htonl(V_TX_ULP_SUBMODE(cxgbi_skcb_tx_ulp_mode(skb)) |
367 			   V_TX_SHOVE((skb_peek(&csk->write_queue) ? 0 : 1)));
368 	req->sndseq = htonl(csk->snd_nxt);
369 	req->param = htonl(V_TX_PORT(l2t->smt_idx));
370 
371 	if (!cxgbi_sock_flag(csk, CTPF_TX_DATA_SENT)) {
372 		req->flags |= htonl(V_TX_ACK_PAGES(2) | F_TX_INIT |
373 				    V_TX_CPU_IDX(csk->rss_qid));
374 		/* sendbuffer is in units of 32KB. */
375 		req->param |= htonl(V_TX_SNDBUF(csk->snd_win >> 15));
376 		cxgbi_sock_set_flag(csk, CTPF_TX_DATA_SENT);
377 	}
378 }
379 
380 /*
381  * push_tx_frames -- start transmit
382  *
383  * Prepends TX_DATA_WR or CPL_CLOSE_CON_REQ headers to buffers waiting in a
384  * connection's send queue and sends them on to T3.  Must be called with the
385  * connection's lock held.  Returns the amount of send buffer space that was
386  * freed as a result of sending queued data to T3.
387  */
388 
389 static void arp_failure_skb_discard(struct t3cdev *dev, struct sk_buff *skb)
390 {
391 	kfree_skb(skb);
392 }
393 
394 static int push_tx_frames(struct cxgbi_sock *csk, int req_completion)
395 {
396 	int total_size = 0;
397 	struct sk_buff *skb;
398 
399 	if (unlikely(csk->state < CTP_ESTABLISHED ||
400 		csk->state == CTP_CLOSE_WAIT_1 || csk->state >= CTP_ABORTING)) {
401 			log_debug(1 << CXGBI_DBG_TOE | 1 << CXGBI_DBG_PDU_TX,
402 				"csk 0x%p,%u,0x%lx,%u, in closing state.\n",
403 				csk, csk->state, csk->flags, csk->tid);
404 		return 0;
405 	}
406 
407 	while (csk->wr_cred && (skb = skb_peek(&csk->write_queue)) != NULL) {
408 		int len = skb->len;	/* length before skb_push */
409 		int frags = skb_shinfo(skb)->nr_frags + (len != skb->data_len);
410 		int wrs_needed = skb_wrs[frags];
411 
412 		if (wrs_needed > 1 && len + sizeof(struct tx_data_wr) <= wrlen)
413 			wrs_needed = 1;
414 
415 		WARN_ON(frags >= SKB_WR_LIST_SIZE || wrs_needed < 1);
416 
417 		if (csk->wr_cred < wrs_needed) {
418 			log_debug(1 << CXGBI_DBG_PDU_TX,
419 				"csk 0x%p, skb len %u/%u, frag %u, wr %d<%u.\n",
420 				csk, skb->len, skb->data_len, frags,
421 				wrs_needed, csk->wr_cred);
422 			break;
423 		}
424 
425 		__skb_unlink(skb, &csk->write_queue);
426 		skb->priority = CPL_PRIORITY_DATA;
427 		skb->csum = wrs_needed;	/* remember this until the WR_ACK */
428 		csk->wr_cred -= wrs_needed;
429 		csk->wr_una_cred += wrs_needed;
430 		cxgbi_sock_enqueue_wr(csk, skb);
431 
432 		log_debug(1 << CXGBI_DBG_TOE | 1 << CXGBI_DBG_PDU_TX,
433 			"csk 0x%p, enqueue, skb len %u/%u, frag %u, wr %d, "
434 			"left %u, unack %u.\n",
435 			csk, skb->len, skb->data_len, frags, skb->csum,
436 			csk->wr_cred, csk->wr_una_cred);
437 
438 		if (likely(cxgbi_skcb_test_flag(skb, SKCBF_TX_NEED_HDR))) {
439 			if ((req_completion &&
440 				csk->wr_una_cred == wrs_needed) ||
441 			     csk->wr_una_cred >= csk->wr_max_cred / 2) {
442 				req_completion = 1;
443 				csk->wr_una_cred = 0;
444 			}
445 			len += cxgbi_ulp_extra_len(cxgbi_skcb_tx_ulp_mode(skb));
446 			make_tx_data_wr(csk, skb, len, req_completion);
447 			csk->snd_nxt += len;
448 			cxgbi_skcb_clear_flag(skb, SKCBF_TX_NEED_HDR);
449 		}
450 		total_size += skb->truesize;
451 		log_debug(1 << CXGBI_DBG_TOE | 1 << CXGBI_DBG_PDU_TX,
452 			"csk 0x%p, tid 0x%x, send skb 0x%p.\n",
453 			csk, csk->tid, skb);
454 		set_arp_failure_handler(skb, arp_failure_skb_discard);
455 		l2t_send(csk->cdev->lldev, skb, csk->l2t);
456 	}
457 	return total_size;
458 }
459 
460 /*
461  * Process a CPL_ACT_ESTABLISH message: -> host
462  * Updates connection state from an active establish CPL message.  Runs with
463  * the connection lock held.
464  */
465 
466 static inline void free_atid(struct cxgbi_sock *csk)
467 {
468 	if (cxgbi_sock_flag(csk, CTPF_HAS_ATID)) {
469 		cxgb3_free_atid(csk->cdev->lldev, csk->atid);
470 		cxgbi_sock_clear_flag(csk, CTPF_HAS_ATID);
471 		cxgbi_sock_put(csk);
472 	}
473 }
474 
475 static int do_act_establish(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
476 {
477 	struct cxgbi_sock *csk = ctx;
478 	struct cpl_act_establish *req = cplhdr(skb);
479 	unsigned int tid = GET_TID(req);
480 	unsigned int atid = G_PASS_OPEN_TID(ntohl(req->tos_tid));
481 	u32 rcv_isn = ntohl(req->rcv_isn);	/* real RCV_ISN + 1 */
482 
483 	log_debug(1 << CXGBI_DBG_TOE | 1 << CXGBI_DBG_SOCK,
484 		"atid 0x%x,tid 0x%x, csk 0x%p,%u,0x%lx, isn %u.\n",
485 		atid, atid, csk, csk->state, csk->flags, rcv_isn);
486 
487 	cxgbi_sock_get(csk);
488 	cxgbi_sock_set_flag(csk, CTPF_HAS_TID);
489 	csk->tid = tid;
490 	cxgb3_insert_tid(csk->cdev->lldev, &t3_client, csk, tid);
491 
492 	free_atid(csk);
493 
494 	csk->rss_qid = G_QNUM(ntohs(skb->csum));
495 
496 	spin_lock_bh(&csk->lock);
497 	if (csk->retry_timer.function) {
498 		del_timer(&csk->retry_timer);
499 		csk->retry_timer.function = NULL;
500 	}
501 
502 	if (unlikely(csk->state != CTP_ACTIVE_OPEN))
503 		pr_info("csk 0x%p,%u,0x%lx,%u, got EST.\n",
504 			csk, csk->state, csk->flags, csk->tid);
505 
506 	csk->copied_seq = csk->rcv_wup = csk->rcv_nxt = rcv_isn;
507 	if (csk->rcv_win > (M_RCV_BUFSIZ << 10))
508 		csk->rcv_wup -= csk->rcv_win - (M_RCV_BUFSIZ << 10);
509 
510 	cxgbi_sock_established(csk, ntohl(req->snd_isn), ntohs(req->tcp_opt));
511 
512 	if (unlikely(cxgbi_sock_flag(csk, CTPF_ACTIVE_CLOSE_NEEDED)))
513 		/* upper layer has requested closing */
514 		send_abort_req(csk);
515 	else {
516 		if (skb_queue_len(&csk->write_queue))
517 			push_tx_frames(csk, 1);
518 		cxgbi_conn_tx_open(csk);
519 	}
520 
521 	spin_unlock_bh(&csk->lock);
522 	__kfree_skb(skb);
523 	return 0;
524 }
525 
526 /*
527  * Process a CPL_ACT_OPEN_RPL message: -> host
528  * Handle active open failures.
529  */
530 static int act_open_rpl_status_to_errno(int status)
531 {
532 	switch (status) {
533 	case CPL_ERR_CONN_RESET:
534 		return -ECONNREFUSED;
535 	case CPL_ERR_ARP_MISS:
536 		return -EHOSTUNREACH;
537 	case CPL_ERR_CONN_TIMEDOUT:
538 		return -ETIMEDOUT;
539 	case CPL_ERR_TCAM_FULL:
540 		return -ENOMEM;
541 	case CPL_ERR_CONN_EXIST:
542 		return -EADDRINUSE;
543 	default:
544 		return -EIO;
545 	}
546 }
547 
548 static void act_open_retry_timer(struct timer_list *t)
549 {
550 	struct cxgbi_sock *csk = from_timer(csk, t, retry_timer);
551 	struct sk_buff *skb;
552 
553 	log_debug(1 << CXGBI_DBG_TOE | 1 << CXGBI_DBG_SOCK,
554 		"csk 0x%p,%u,0x%lx,%u.\n",
555 		csk, csk->state, csk->flags, csk->tid);
556 
557 	cxgbi_sock_get(csk);
558 	spin_lock_bh(&csk->lock);
559 	skb = alloc_wr(sizeof(struct cpl_act_open_req), 0, GFP_ATOMIC);
560 	if (!skb)
561 		cxgbi_sock_fail_act_open(csk, -ENOMEM);
562 	else {
563 		skb->sk = (struct sock *)csk;
564 		set_arp_failure_handler(skb, act_open_arp_failure);
565 		send_act_open_req(csk, skb, csk->l2t);
566 	}
567 	spin_unlock_bh(&csk->lock);
568 	cxgbi_sock_put(csk);
569 }
570 
571 static int do_act_open_rpl(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
572 {
573 	struct cxgbi_sock *csk = ctx;
574 	struct cpl_act_open_rpl *rpl = cplhdr(skb);
575 
576 	pr_info("csk 0x%p,%u,0x%lx,%u, status %u, %pI4:%u-%pI4:%u.\n",
577 		csk, csk->state, csk->flags, csk->atid, rpl->status,
578 		&csk->saddr.sin_addr.s_addr, ntohs(csk->saddr.sin_port),
579 		&csk->daddr.sin_addr.s_addr, ntohs(csk->daddr.sin_port));
580 
581 	if (rpl->status != CPL_ERR_TCAM_FULL &&
582 	    rpl->status != CPL_ERR_CONN_EXIST &&
583 	    rpl->status != CPL_ERR_ARP_MISS)
584 		cxgb3_queue_tid_release(tdev, GET_TID(rpl));
585 
586 	cxgbi_sock_get(csk);
587 	spin_lock_bh(&csk->lock);
588 	if (rpl->status == CPL_ERR_CONN_EXIST &&
589 	    csk->retry_timer.function != act_open_retry_timer) {
590 		csk->retry_timer.function = act_open_retry_timer;
591 		mod_timer(&csk->retry_timer, jiffies + HZ / 2);
592 	} else
593 		cxgbi_sock_fail_act_open(csk,
594 				act_open_rpl_status_to_errno(rpl->status));
595 
596 	spin_unlock_bh(&csk->lock);
597 	cxgbi_sock_put(csk);
598 	__kfree_skb(skb);
599 	return 0;
600 }
601 
602 /*
603  * Process PEER_CLOSE CPL messages: -> host
604  * Handle peer FIN.
605  */
606 static int do_peer_close(struct t3cdev *cdev, struct sk_buff *skb, void *ctx)
607 {
608 	struct cxgbi_sock *csk = ctx;
609 
610 	log_debug(1 << CXGBI_DBG_TOE | 1 << CXGBI_DBG_SOCK,
611 		"csk 0x%p,%u,0x%lx,%u.\n",
612 		csk, csk->state, csk->flags, csk->tid);
613 
614 	cxgbi_sock_rcv_peer_close(csk);
615 	__kfree_skb(skb);
616 	return 0;
617 }
618 
619 /*
620  * Process CLOSE_CONN_RPL CPL message: -> host
621  * Process a peer ACK to our FIN.
622  */
623 static int do_close_con_rpl(struct t3cdev *cdev, struct sk_buff *skb,
624 			    void *ctx)
625 {
626 	struct cxgbi_sock *csk = ctx;
627 	struct cpl_close_con_rpl *rpl = cplhdr(skb);
628 
629 	log_debug(1 << CXGBI_DBG_TOE | 1 << CXGBI_DBG_SOCK,
630 		"csk 0x%p,%u,0x%lx,%u, snxt %u.\n",
631 		csk, csk->state, csk->flags, csk->tid, ntohl(rpl->snd_nxt));
632 
633 	cxgbi_sock_rcv_close_conn_rpl(csk, ntohl(rpl->snd_nxt));
634 	__kfree_skb(skb);
635 	return 0;
636 }
637 
638 /*
639  * Process ABORT_REQ_RSS CPL message: -> host
640  * Process abort requests.  If we are waiting for an ABORT_RPL we ignore this
641  * request except that we need to reply to it.
642  */
643 
644 static int abort_status_to_errno(struct cxgbi_sock *csk, int abort_reason,
645 				 int *need_rst)
646 {
647 	switch (abort_reason) {
648 	case CPL_ERR_BAD_SYN:
649 	case CPL_ERR_CONN_RESET:
650 		return csk->state > CTP_ESTABLISHED ? -EPIPE : -ECONNRESET;
651 	case CPL_ERR_XMIT_TIMEDOUT:
652 	case CPL_ERR_PERSIST_TIMEDOUT:
653 	case CPL_ERR_FINWAIT2_TIMEDOUT:
654 	case CPL_ERR_KEEPALIVE_TIMEDOUT:
655 		return -ETIMEDOUT;
656 	default:
657 		return -EIO;
658 	}
659 }
660 
661 static int do_abort_req(struct t3cdev *cdev, struct sk_buff *skb, void *ctx)
662 {
663 	const struct cpl_abort_req_rss *req = cplhdr(skb);
664 	struct cxgbi_sock *csk = ctx;
665 	int rst_status = CPL_ABORT_NO_RST;
666 
667 	log_debug(1 << CXGBI_DBG_TOE | 1 << CXGBI_DBG_SOCK,
668 		"csk 0x%p,%u,0x%lx,%u.\n",
669 		csk, csk->state, csk->flags, csk->tid);
670 
671 	if (req->status == CPL_ERR_RTX_NEG_ADVICE ||
672 	    req->status == CPL_ERR_PERSIST_NEG_ADVICE) {
673 		goto done;
674 	}
675 
676 	cxgbi_sock_get(csk);
677 	spin_lock_bh(&csk->lock);
678 
679 	if (!cxgbi_sock_flag(csk, CTPF_ABORT_REQ_RCVD)) {
680 		cxgbi_sock_set_flag(csk, CTPF_ABORT_REQ_RCVD);
681 		cxgbi_sock_set_state(csk, CTP_ABORTING);
682 		goto out;
683 	}
684 
685 	cxgbi_sock_clear_flag(csk, CTPF_ABORT_REQ_RCVD);
686 	send_abort_rpl(csk, rst_status);
687 
688 	if (!cxgbi_sock_flag(csk, CTPF_ABORT_RPL_PENDING)) {
689 		csk->err = abort_status_to_errno(csk, req->status, &rst_status);
690 		cxgbi_sock_closed(csk);
691 	}
692 
693 out:
694 	spin_unlock_bh(&csk->lock);
695 	cxgbi_sock_put(csk);
696 done:
697 	__kfree_skb(skb);
698 	return 0;
699 }
700 
701 /*
702  * Process ABORT_RPL_RSS CPL message: -> host
703  * Process abort replies.  We only process these messages if we anticipate
704  * them as the coordination between SW and HW in this area is somewhat lacking
705  * and sometimes we get ABORT_RPLs after we are done with the connection that
706  * originated the ABORT_REQ.
707  */
708 static int do_abort_rpl(struct t3cdev *cdev, struct sk_buff *skb, void *ctx)
709 {
710 	struct cpl_abort_rpl_rss *rpl = cplhdr(skb);
711 	struct cxgbi_sock *csk = ctx;
712 
713 	log_debug(1 << CXGBI_DBG_TOE | 1 << CXGBI_DBG_SOCK,
714 		"status 0x%x, csk 0x%p, s %u, 0x%lx.\n",
715 		rpl->status, csk, csk ? csk->state : 0,
716 		csk ? csk->flags : 0UL);
717 	/*
718 	 * Ignore replies to post-close aborts indicating that the abort was
719 	 * requested too late.  These connections are terminated when we get
720 	 * PEER_CLOSE or CLOSE_CON_RPL and by the time the abort_rpl_rss
721 	 * arrives the TID is either no longer used or it has been recycled.
722 	 */
723 	if (rpl->status == CPL_ERR_ABORT_FAILED)
724 		goto rel_skb;
725 	/*
726 	 * Sometimes we've already closed the connection, e.g., a post-close
727 	 * abort races with ABORT_REQ_RSS, the latter frees the connection
728 	 * expecting the ABORT_REQ will fail with CPL_ERR_ABORT_FAILED,
729 	 * but FW turns the ABORT_REQ into a regular one and so we get
730 	 * ABORT_RPL_RSS with status 0 and no connection.
731 	 */
732 	if (csk)
733 		cxgbi_sock_rcv_abort_rpl(csk);
734 rel_skb:
735 	__kfree_skb(skb);
736 	return 0;
737 }
738 
739 /*
740  * Process RX_ISCSI_HDR CPL message: -> host
741  * Handle received PDUs, the payload could be DDP'ed. If not, the payload
742  * follow after the bhs.
743  */
744 static int do_iscsi_hdr(struct t3cdev *t3dev, struct sk_buff *skb, void *ctx)
745 {
746 	struct cxgbi_sock *csk = ctx;
747 	struct cpl_iscsi_hdr *hdr_cpl = cplhdr(skb);
748 	struct cpl_iscsi_hdr_norss data_cpl;
749 	struct cpl_rx_data_ddp_norss ddp_cpl;
750 	unsigned int hdr_len, data_len, status;
751 	unsigned int len;
752 	int err;
753 
754 	log_debug(1 << CXGBI_DBG_TOE | 1 << CXGBI_DBG_PDU_RX,
755 		"csk 0x%p,%u,0x%lx,%u, skb 0x%p,%u.\n",
756 		csk, csk->state, csk->flags, csk->tid, skb, skb->len);
757 
758 	spin_lock_bh(&csk->lock);
759 
760 	if (unlikely(csk->state >= CTP_PASSIVE_CLOSE)) {
761 		log_debug(1 << CXGBI_DBG_TOE | 1 << CXGBI_DBG_SOCK,
762 			"csk 0x%p,%u,0x%lx,%u, bad state.\n",
763 			csk, csk->state, csk->flags, csk->tid);
764 		if (csk->state != CTP_ABORTING)
765 			goto abort_conn;
766 		else
767 			goto discard;
768 	}
769 
770 	cxgbi_skcb_tcp_seq(skb) = ntohl(hdr_cpl->seq);
771 	cxgbi_skcb_flags(skb) = 0;
772 
773 	skb_reset_transport_header(skb);
774 	__skb_pull(skb, sizeof(struct cpl_iscsi_hdr));
775 
776 	len = hdr_len = ntohs(hdr_cpl->len);
777 	/* msg coalesce is off or not enough data received */
778 	if (skb->len <= hdr_len) {
779 		pr_err("%s: tid %u, CPL_ISCSI_HDR, skb len %u < %u.\n",
780 			csk->cdev->ports[csk->port_id]->name, csk->tid,
781 			skb->len, hdr_len);
782 		goto abort_conn;
783 	}
784 	cxgbi_skcb_set_flag(skb, SKCBF_RX_COALESCED);
785 
786 	err = skb_copy_bits(skb, skb->len - sizeof(ddp_cpl), &ddp_cpl,
787 			    sizeof(ddp_cpl));
788 	if (err < 0) {
789 		pr_err("%s: tid %u, copy cpl_ddp %u-%zu failed %d.\n",
790 			csk->cdev->ports[csk->port_id]->name, csk->tid,
791 			skb->len, sizeof(ddp_cpl), err);
792 		goto abort_conn;
793 	}
794 
795 	cxgbi_skcb_set_flag(skb, SKCBF_RX_STATUS);
796 	cxgbi_skcb_rx_pdulen(skb) = ntohs(ddp_cpl.len);
797 	cxgbi_skcb_rx_ddigest(skb) = ntohl(ddp_cpl.ulp_crc);
798 	status = ntohl(ddp_cpl.ddp_status);
799 
800 	log_debug(1 << CXGBI_DBG_TOE | 1 << CXGBI_DBG_PDU_RX,
801 		"csk 0x%p, skb 0x%p,%u, pdulen %u, status 0x%x.\n",
802 		csk, skb, skb->len, cxgbi_skcb_rx_pdulen(skb), status);
803 
804 	if (status & (1 << CPL_RX_DDP_STATUS_HCRC_SHIFT))
805 		cxgbi_skcb_set_flag(skb, SKCBF_RX_HCRC_ERR);
806 	if (status & (1 << CPL_RX_DDP_STATUS_DCRC_SHIFT))
807 		cxgbi_skcb_set_flag(skb, SKCBF_RX_DCRC_ERR);
808 	if (status & (1 << CPL_RX_DDP_STATUS_PAD_SHIFT))
809 		cxgbi_skcb_set_flag(skb, SKCBF_RX_PAD_ERR);
810 
811 	if (skb->len > (hdr_len + sizeof(ddp_cpl))) {
812 		err = skb_copy_bits(skb, hdr_len, &data_cpl, sizeof(data_cpl));
813 		if (err < 0) {
814 			pr_err("%s: tid %u, cp %zu/%u failed %d.\n",
815 				csk->cdev->ports[csk->port_id]->name,
816 				csk->tid, sizeof(data_cpl), skb->len, err);
817 			goto abort_conn;
818 		}
819 		data_len = ntohs(data_cpl.len);
820 		log_debug(1 << CXGBI_DBG_DDP | 1 << CXGBI_DBG_PDU_RX,
821 			"skb 0x%p, pdu not ddp'ed %u/%u, status 0x%x.\n",
822 			skb, data_len, cxgbi_skcb_rx_pdulen(skb), status);
823 		len += sizeof(data_cpl) + data_len;
824 	} else if (status & (1 << CPL_RX_DDP_STATUS_DDP_SHIFT))
825 		cxgbi_skcb_set_flag(skb, SKCBF_RX_DATA_DDPD);
826 
827 	csk->rcv_nxt = ntohl(ddp_cpl.seq) + cxgbi_skcb_rx_pdulen(skb);
828 	__pskb_trim(skb, len);
829 	__skb_queue_tail(&csk->receive_queue, skb);
830 	cxgbi_conn_pdu_ready(csk);
831 
832 	spin_unlock_bh(&csk->lock);
833 	return 0;
834 
835 abort_conn:
836 	send_abort_req(csk);
837 discard:
838 	spin_unlock_bh(&csk->lock);
839 	__kfree_skb(skb);
840 	return 0;
841 }
842 
843 /*
844  * Process TX_DATA_ACK CPL messages: -> host
845  * Process an acknowledgment of WR completion.  Advance snd_una and send the
846  * next batch of work requests from the write queue.
847  */
848 static int do_wr_ack(struct t3cdev *cdev, struct sk_buff *skb, void *ctx)
849 {
850 	struct cxgbi_sock *csk = ctx;
851 	struct cpl_wr_ack *hdr = cplhdr(skb);
852 
853 	log_debug(1 << CXGBI_DBG_TOE | 1 << CXGBI_DBG_PDU_RX,
854 		"csk 0x%p,%u,0x%lx,%u, cr %u.\n",
855 		csk, csk->state, csk->flags, csk->tid, ntohs(hdr->credits));
856 
857 	cxgbi_sock_rcv_wr_ack(csk, ntohs(hdr->credits), ntohl(hdr->snd_una), 1);
858 	__kfree_skb(skb);
859 	return 0;
860 }
861 
862 /*
863  * for each connection, pre-allocate skbs needed for close/abort requests. So
864  * that we can service the request right away.
865  */
866 static int alloc_cpls(struct cxgbi_sock *csk)
867 {
868 	csk->cpl_close = alloc_wr(sizeof(struct cpl_close_con_req), 0,
869 					GFP_KERNEL);
870 	if (!csk->cpl_close)
871 		return -ENOMEM;
872 	csk->cpl_abort_req = alloc_wr(sizeof(struct cpl_abort_req), 0,
873 					GFP_KERNEL);
874 	if (!csk->cpl_abort_req)
875 		goto free_cpl_skbs;
876 
877 	csk->cpl_abort_rpl = alloc_wr(sizeof(struct cpl_abort_rpl), 0,
878 					GFP_KERNEL);
879 	if (!csk->cpl_abort_rpl)
880 		goto free_cpl_skbs;
881 
882 	return 0;
883 
884 free_cpl_skbs:
885 	cxgbi_sock_free_cpl_skbs(csk);
886 	return -ENOMEM;
887 }
888 
889 static void l2t_put(struct cxgbi_sock *csk)
890 {
891 	struct t3cdev *t3dev = (struct t3cdev *)csk->cdev->lldev;
892 
893 	if (csk->l2t) {
894 		l2t_release(t3dev, csk->l2t);
895 		csk->l2t = NULL;
896 		cxgbi_sock_put(csk);
897 	}
898 }
899 
900 /*
901  * release_offload_resources - release offload resource
902  * Release resources held by an offload connection (TID, L2T entry, etc.)
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 	int ret;
962 
963 	log_debug(1 << CXGBI_DBG_TOE | 1 << CXGBI_DBG_SOCK,
964 		"csk 0x%p,%u,0x%lx.\n", csk, csk->state, csk->flags);
965 
966 	update_address(chba);
967 	if (chba->ipv4addr)
968 		csk->saddr.sin_addr.s_addr = chba->ipv4addr;
969 
970 	csk->rss_qid = 0;
971 	csk->l2t = t3_l2t_get(t3dev, dst, ndev,
972 			      &csk->daddr.sin_addr.s_addr);
973 	if (!csk->l2t) {
974 		pr_err("NO l2t available.\n");
975 		return -EINVAL;
976 	}
977 	cxgbi_sock_get(csk);
978 
979 	csk->atid = cxgb3_alloc_atid(t3dev, &t3_client, csk);
980 	if (csk->atid < 0) {
981 		pr_err("NO atid available.\n");
982 		ret = -EINVAL;
983 		goto put_sock;
984 	}
985 	cxgbi_sock_set_flag(csk, CTPF_HAS_ATID);
986 	cxgbi_sock_get(csk);
987 
988 	skb = alloc_wr(sizeof(struct cpl_act_open_req), 0, GFP_KERNEL);
989 	if (!skb) {
990 		ret = -ENOMEM;
991 		goto free_atid;
992 	}
993 	skb->sk = (struct sock *)csk;
994 	set_arp_failure_handler(skb, act_open_arp_failure);
995 	csk->snd_win = cxgb3i_snd_win;
996 	csk->rcv_win = cxgb3i_rcv_win;
997 
998 	csk->wr_max_cred = csk->wr_cred = T3C_DATA(t3dev)->max_wrs - 1;
999 	csk->wr_una_cred = 0;
1000 	csk->mss_idx = cxgbi_sock_select_mss(csk, dst_mtu(dst));
1001 	cxgbi_sock_reset_wr_list(csk);
1002 	csk->err = 0;
1003 
1004 	log_debug(1 << CXGBI_DBG_TOE | 1 << CXGBI_DBG_SOCK,
1005 		"csk 0x%p,%u,0x%lx, %pI4:%u-%pI4:%u.\n",
1006 		csk, csk->state, csk->flags,
1007 		&csk->saddr.sin_addr.s_addr, ntohs(csk->saddr.sin_port),
1008 		&csk->daddr.sin_addr.s_addr, ntohs(csk->daddr.sin_port));
1009 
1010 	cxgbi_sock_set_state(csk, CTP_ACTIVE_OPEN);
1011 	send_act_open_req(csk, skb, csk->l2t);
1012 	return 0;
1013 
1014 free_atid:
1015 	cxgb3_free_atid(t3dev, csk->atid);
1016 put_sock:
1017 	cxgbi_sock_put(csk);
1018 	l2t_release(t3dev, csk->l2t);
1019 	csk->l2t = NULL;
1020 
1021 	return ret;
1022 }
1023 
1024 cxgb3_cpl_handler_func cxgb3i_cpl_handlers[NUM_CPL_CMDS] = {
1025 	[CPL_ACT_ESTABLISH] = do_act_establish,
1026 	[CPL_ACT_OPEN_RPL] = do_act_open_rpl,
1027 	[CPL_PEER_CLOSE] = do_peer_close,
1028 	[CPL_ABORT_REQ_RSS] = do_abort_req,
1029 	[CPL_ABORT_RPL_RSS] = do_abort_rpl,
1030 	[CPL_CLOSE_CON_RPL] = do_close_con_rpl,
1031 	[CPL_TX_DMA_ACK] = do_wr_ack,
1032 	[CPL_ISCSI_HDR] = do_iscsi_hdr,
1033 };
1034 
1035 /**
1036  * cxgb3i_ofld_init - allocate and initialize resources for each adapter found
1037  * @cdev:	cxgbi adapter
1038  */
1039 static int cxgb3i_ofld_init(struct cxgbi_device *cdev)
1040 {
1041 	struct t3cdev *t3dev = (struct t3cdev *)cdev->lldev;
1042 	struct adap_ports port;
1043 	struct ofld_page_info rx_page_info;
1044 	unsigned int wr_len;
1045 	int rc;
1046 
1047 	if (t3dev->ctl(t3dev, GET_WR_LEN, &wr_len) < 0 ||
1048 	    t3dev->ctl(t3dev, GET_PORTS, &port) < 0 ||
1049 	    t3dev->ctl(t3dev, GET_RX_PAGE_INFO, &rx_page_info) < 0) {
1050 		pr_warn("t3 0x%p, offload up, ioctl failed.\n", t3dev);
1051 		return -EINVAL;
1052 	}
1053 
1054 	if (cxgb3i_max_connect > CXGBI_MAX_CONN)
1055 		cxgb3i_max_connect = CXGBI_MAX_CONN;
1056 
1057 	rc = cxgbi_device_portmap_create(cdev, cxgb3i_sport_base,
1058 					cxgb3i_max_connect);
1059 	if (rc < 0)
1060 		return rc;
1061 
1062 	init_wr_tab(wr_len);
1063 	cdev->csk_release_offload_resources = release_offload_resources;
1064 	cdev->csk_push_tx_frames = push_tx_frames;
1065 	cdev->csk_send_abort_req = send_abort_req;
1066 	cdev->csk_send_close_req = send_close_req;
1067 	cdev->csk_send_rx_credits = send_rx_credits;
1068 	cdev->csk_alloc_cpls = alloc_cpls;
1069 	cdev->csk_init_act_open = init_act_open;
1070 
1071 	pr_info("cdev 0x%p, offload up, added.\n", cdev);
1072 	return 0;
1073 }
1074 
1075 /*
1076  * functions to program the pagepod in h/w
1077  */
1078 static inline void ulp_mem_io_set_hdr(struct sk_buff *skb, unsigned int addr)
1079 {
1080 	struct ulp_mem_io *req = (struct ulp_mem_io *)skb->head;
1081 
1082 	memset(req, 0, sizeof(*req));
1083 
1084 	req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_BYPASS));
1085 	req->cmd_lock_addr = htonl(V_ULP_MEMIO_ADDR(addr >> 5) |
1086 				   V_ULPTX_CMD(ULP_MEM_WRITE));
1087 	req->len = htonl(V_ULP_MEMIO_DATA_LEN(IPPOD_SIZE >> 5) |
1088 			 V_ULPTX_NFLITS((IPPOD_SIZE >> 3) + 1));
1089 }
1090 
1091 static struct cxgbi_ppm *cdev2ppm(struct cxgbi_device *cdev)
1092 {
1093 	return ((struct t3cdev *)cdev->lldev)->ulp_iscsi;
1094 }
1095 
1096 static int ddp_set_map(struct cxgbi_ppm *ppm, struct cxgbi_sock *csk,
1097 		       struct cxgbi_task_tag_info *ttinfo)
1098 {
1099 	unsigned int idx = ttinfo->idx;
1100 	unsigned int npods = ttinfo->npods;
1101 	struct scatterlist *sg = ttinfo->sgl;
1102 	struct cxgbi_pagepod *ppod;
1103 	struct ulp_mem_io *req;
1104 	unsigned int sg_off;
1105 	unsigned int pm_addr = (idx << PPOD_SIZE_SHIFT) + ppm->llimit;
1106 	int i;
1107 
1108 	for (i = 0; i < npods; i++, idx++, pm_addr += IPPOD_SIZE) {
1109 		struct sk_buff *skb = alloc_wr(sizeof(struct ulp_mem_io) +
1110 					       IPPOD_SIZE, 0, GFP_ATOMIC);
1111 
1112 		if (!skb)
1113 			return -ENOMEM;
1114 		ulp_mem_io_set_hdr(skb, pm_addr);
1115 		req = (struct ulp_mem_io *)skb->head;
1116 		ppod = (struct cxgbi_pagepod *)(req + 1);
1117 		sg_off = i * PPOD_PAGES_MAX;
1118 		cxgbi_ddp_set_one_ppod(ppod, ttinfo, &sg,
1119 				       &sg_off);
1120 		skb->priority = CPL_PRIORITY_CONTROL;
1121 		cxgb3_ofld_send(ppm->lldev, skb);
1122 	}
1123 	return 0;
1124 }
1125 
1126 static void ddp_clear_map(struct cxgbi_device *cdev, struct cxgbi_ppm *ppm,
1127 			  struct cxgbi_task_tag_info *ttinfo)
1128 {
1129 	unsigned int idx = ttinfo->idx;
1130 	unsigned int pm_addr = (idx << PPOD_SIZE_SHIFT) + ppm->llimit;
1131 	unsigned int npods = ttinfo->npods;
1132 	int i;
1133 
1134 	log_debug(1 << CXGBI_DBG_DDP,
1135 		  "cdev 0x%p, clear idx %u, npods %u.\n",
1136 		  cdev, idx, npods);
1137 
1138 	for (i = 0; i < npods; i++, idx++, pm_addr += IPPOD_SIZE) {
1139 		struct sk_buff *skb = alloc_wr(sizeof(struct ulp_mem_io) +
1140 					       IPPOD_SIZE, 0, GFP_ATOMIC);
1141 
1142 		if (!skb) {
1143 			pr_err("cdev 0x%p, clear ddp, %u,%d/%u, skb OOM.\n",
1144 			       cdev, idx, i, npods);
1145 			continue;
1146 		}
1147 		ulp_mem_io_set_hdr(skb, pm_addr);
1148 		skb->priority = CPL_PRIORITY_CONTROL;
1149 		cxgb3_ofld_send(ppm->lldev, skb);
1150 	}
1151 }
1152 
1153 static int ddp_setup_conn_pgidx(struct cxgbi_sock *csk,
1154 				unsigned int tid, int pg_idx)
1155 {
1156 	struct sk_buff *skb = alloc_wr(sizeof(struct cpl_set_tcb_field), 0,
1157 					GFP_KERNEL);
1158 	struct cpl_set_tcb_field *req;
1159 	u64 val = pg_idx < DDP_PGIDX_MAX ? pg_idx : 0;
1160 
1161 	log_debug(1 << CXGBI_DBG_DDP,
1162 		"csk 0x%p, tid %u, pg_idx %d.\n", csk, tid, pg_idx);
1163 	if (!skb)
1164 		return -ENOMEM;
1165 
1166 	/* set up ulp submode and page size */
1167 	req = (struct cpl_set_tcb_field *)skb->head;
1168 	req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
1169 	OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_SET_TCB_FIELD, tid));
1170 	req->reply = V_NO_REPLY(1);
1171 	req->cpu_idx = 0;
1172 	req->word = htons(31);
1173 	req->mask = cpu_to_be64(0xF0000000);
1174 	req->val = cpu_to_be64(val << 28);
1175 	skb->priority = CPL_PRIORITY_CONTROL;
1176 
1177 	cxgb3_ofld_send(csk->cdev->lldev, skb);
1178 	return 0;
1179 }
1180 
1181 /**
1182  * ddp_setup_conn_digest - setup conn. digest setting
1183  * @csk: cxgb tcp socket
1184  * @tid: connection id
1185  * @hcrc: header digest enabled
1186  * @dcrc: data digest enabled
1187  * set up the iscsi digest settings for a connection identified by tid
1188  */
1189 static int ddp_setup_conn_digest(struct cxgbi_sock *csk, unsigned int tid,
1190 				 int hcrc, int dcrc)
1191 {
1192 	struct sk_buff *skb = alloc_wr(sizeof(struct cpl_set_tcb_field), 0,
1193 					GFP_KERNEL);
1194 	struct cpl_set_tcb_field *req;
1195 	u64 val = (hcrc ? 1 : 0) | (dcrc ? 2 : 0);
1196 
1197 	log_debug(1 << CXGBI_DBG_DDP,
1198 		"csk 0x%p, tid %u, crc %d,%d.\n", csk, tid, hcrc, dcrc);
1199 	if (!skb)
1200 		return -ENOMEM;
1201 
1202 	/* set up ulp submode and page size */
1203 	req = (struct cpl_set_tcb_field *)skb->head;
1204 	req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
1205 	OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_SET_TCB_FIELD, tid));
1206 	req->reply = V_NO_REPLY(1);
1207 	req->cpu_idx = 0;
1208 	req->word = htons(31);
1209 	req->mask = cpu_to_be64(0x0F000000);
1210 	req->val = cpu_to_be64(val << 24);
1211 	skb->priority = CPL_PRIORITY_CONTROL;
1212 
1213 	cxgb3_ofld_send(csk->cdev->lldev, skb);
1214 	return 0;
1215 }
1216 
1217 /**
1218  * cxgb3i_ddp_init - initialize the cxgb3 adapter's ddp resource
1219  * @cdev: cxgb3i adapter
1220  * initialize the ddp pagepod manager for a given adapter
1221  */
1222 static int cxgb3i_ddp_init(struct cxgbi_device *cdev)
1223 {
1224 	struct t3cdev *tdev = (struct t3cdev *)cdev->lldev;
1225 	struct net_device *ndev = cdev->ports[0];
1226 	struct cxgbi_tag_format tformat;
1227 	unsigned int ppmax, tagmask = 0;
1228 	struct ulp_iscsi_info uinfo;
1229 	int i, err;
1230 
1231 	err = tdev->ctl(tdev, ULP_ISCSI_GET_PARAMS, &uinfo);
1232 	if (err < 0) {
1233 		pr_err("%s, failed to get iscsi param %d.\n",
1234 		       ndev->name, err);
1235 		return err;
1236 	}
1237 	if (uinfo.llimit >= uinfo.ulimit) {
1238 		pr_warn("T3 %s, iscsi NOT enabled %u ~ %u!\n",
1239 			ndev->name, uinfo.llimit, uinfo.ulimit);
1240 		return -EACCES;
1241 	}
1242 
1243 	ppmax = (uinfo.ulimit - uinfo.llimit + 1) >> PPOD_SIZE_SHIFT;
1244 	tagmask = cxgbi_tagmask_set(ppmax);
1245 
1246 	pr_info("T3 %s: 0x%x~0x%x, 0x%x, tagmask 0x%x -> 0x%x.\n",
1247 		ndev->name, uinfo.llimit, uinfo.ulimit, ppmax, uinfo.tagmask,
1248 		tagmask);
1249 
1250 	memset(&tformat, 0, sizeof(struct cxgbi_tag_format));
1251 	for (i = 0; i < 4; i++)
1252 		tformat.pgsz_order[i] = uinfo.pgsz_factor[i];
1253 	cxgbi_tagmask_check(tagmask, &tformat);
1254 
1255 	err = cxgbi_ddp_ppm_setup(&tdev->ulp_iscsi, cdev, &tformat,
1256 				  (uinfo.ulimit - uinfo.llimit + 1),
1257 				  uinfo.llimit, uinfo.llimit, 0, 0, 0);
1258 	if (err)
1259 		return err;
1260 
1261 	if (!(cdev->flags & CXGBI_FLAG_DDP_OFF)) {
1262 		uinfo.tagmask = tagmask;
1263 		uinfo.ulimit = uinfo.llimit + (ppmax << PPOD_SIZE_SHIFT);
1264 
1265 		err = tdev->ctl(tdev, ULP_ISCSI_SET_PARAMS, &uinfo);
1266 		if (err < 0) {
1267 			pr_err("T3 %s fail to set iscsi param %d.\n",
1268 			       ndev->name, err);
1269 			cdev->flags |= CXGBI_FLAG_DDP_OFF;
1270 		}
1271 		err = 0;
1272 	}
1273 
1274 	cdev->csk_ddp_setup_digest = ddp_setup_conn_digest;
1275 	cdev->csk_ddp_setup_pgidx = ddp_setup_conn_pgidx;
1276 	cdev->csk_ddp_set_map = ddp_set_map;
1277 	cdev->csk_ddp_clear_map = ddp_clear_map;
1278 	cdev->cdev2ppm = cdev2ppm;
1279 	cdev->tx_max_size = min_t(unsigned int, ULP2_MAX_PDU_PAYLOAD,
1280 				  uinfo.max_txsz - ISCSI_PDU_NONPAYLOAD_LEN);
1281 	cdev->rx_max_size = min_t(unsigned int, ULP2_MAX_PDU_PAYLOAD,
1282 				  uinfo.max_rxsz - ISCSI_PDU_NONPAYLOAD_LEN);
1283 
1284 	return 0;
1285 }
1286 
1287 static void cxgb3i_dev_close(struct t3cdev *t3dev)
1288 {
1289 	struct cxgbi_device *cdev = cxgbi_device_find_by_lldev(t3dev);
1290 
1291 	if (!cdev || cdev->flags & CXGBI_FLAG_ADAPTER_RESET) {
1292 		pr_info("0x%p close, f 0x%x.\n", cdev, cdev ? cdev->flags : 0);
1293 		return;
1294 	}
1295 
1296 	cxgbi_device_unregister(cdev);
1297 }
1298 
1299 /**
1300  * cxgb3i_dev_open - init a t3 adapter structure and any h/w settings
1301  * @t3dev: t3cdev adapter
1302  */
1303 static void cxgb3i_dev_open(struct t3cdev *t3dev)
1304 {
1305 	struct cxgbi_device *cdev = cxgbi_device_find_by_lldev(t3dev);
1306 	struct adapter *adapter = tdev2adap(t3dev);
1307 	int i, err;
1308 
1309 	if (cdev) {
1310 		pr_info("0x%p, updating.\n", cdev);
1311 		return;
1312 	}
1313 
1314 	cdev = cxgbi_device_register(0, adapter->params.nports);
1315 	if (!cdev) {
1316 		pr_warn("device 0x%p register failed.\n", t3dev);
1317 		return;
1318 	}
1319 
1320 	cdev->flags = CXGBI_FLAG_DEV_T3 | CXGBI_FLAG_IPV4_SET;
1321 	cdev->lldev = t3dev;
1322 	cdev->pdev = adapter->pdev;
1323 	cdev->ports = adapter->port;
1324 	cdev->nports = adapter->params.nports;
1325 	cdev->mtus = adapter->params.mtus;
1326 	cdev->nmtus = NMTUS;
1327 	cdev->rx_credit_thres = cxgb3i_rx_credit_thres;
1328 	cdev->skb_tx_rsvd = CXGB3I_TX_HEADER_LEN;
1329 	cdev->skb_rx_extra = sizeof(struct cpl_iscsi_hdr_norss);
1330 	cdev->itp = &cxgb3i_iscsi_transport;
1331 
1332 	err = cxgb3i_ddp_init(cdev);
1333 	if (err) {
1334 		pr_info("0x%p ddp init failed %d\n", cdev, err);
1335 		goto err_out;
1336 	}
1337 
1338 	err = cxgb3i_ofld_init(cdev);
1339 	if (err) {
1340 		pr_info("0x%p offload init failed\n", cdev);
1341 		goto err_out;
1342 	}
1343 
1344 	err = cxgbi_hbas_add(cdev, CXGB3I_MAX_LUN, CXGBI_MAX_CONN,
1345 				&cxgb3i_host_template, cxgb3i_stt);
1346 	if (err)
1347 		goto err_out;
1348 
1349 	for (i = 0; i < cdev->nports; i++)
1350 		cdev->hbas[i]->ipv4addr =
1351 			cxgb3i_get_private_ipv4addr(cdev->ports[i]);
1352 
1353 	pr_info("cdev 0x%p, f 0x%x, t3dev 0x%p open, err %d.\n",
1354 		cdev, cdev ? cdev->flags : 0, t3dev, err);
1355 	return;
1356 
1357 err_out:
1358 	cxgbi_device_unregister(cdev);
1359 }
1360 
1361 static void cxgb3i_dev_event_handler(struct t3cdev *t3dev, u32 event, u32 port)
1362 {
1363 	struct cxgbi_device *cdev = cxgbi_device_find_by_lldev(t3dev);
1364 
1365 	log_debug(1 << CXGBI_DBG_TOE,
1366 		"0x%p, cdev 0x%p, event 0x%x, port 0x%x.\n",
1367 		t3dev, cdev, event, port);
1368 	if (!cdev)
1369 		return;
1370 
1371 	switch (event) {
1372 	case OFFLOAD_STATUS_DOWN:
1373 		cdev->flags |= CXGBI_FLAG_ADAPTER_RESET;
1374 		break;
1375 	case OFFLOAD_STATUS_UP:
1376 		cdev->flags &= ~CXGBI_FLAG_ADAPTER_RESET;
1377 		break;
1378 	}
1379 }
1380 
1381 /**
1382  * cxgb3i_init_module - module init entry point
1383  *
1384  * initialize any driver wide global data structures and register itself
1385  *	with the cxgb3 module
1386  */
1387 static int __init cxgb3i_init_module(void)
1388 {
1389 	int rc;
1390 
1391 	printk(KERN_INFO "%s", version);
1392 
1393 	rc = cxgbi_iscsi_init(&cxgb3i_iscsi_transport, &cxgb3i_stt);
1394 	if (rc < 0)
1395 		return rc;
1396 
1397 	cxgb3_register_client(&t3_client);
1398 	return 0;
1399 }
1400 
1401 /**
1402  * cxgb3i_exit_module - module cleanup/exit entry point
1403  *
1404  * go through the driver hba list and for each hba, release any resource held.
1405  *	and unregisters iscsi transport and the cxgb3 module
1406  */
1407 static void __exit cxgb3i_exit_module(void)
1408 {
1409 	cxgb3_unregister_client(&t3_client);
1410 	cxgbi_device_unregister_all(CXGBI_FLAG_DEV_T3);
1411 	cxgbi_iscsi_cleanup(&cxgb3i_iscsi_transport, &cxgb3i_stt);
1412 }
1413 
1414 module_init(cxgb3i_init_module);
1415 module_exit(cxgb3i_exit_module);
1416