xref: /linux/drivers/scsi/cxgbi/cxgb3i/cxgb3i.c (revision a4eb44a6435d6d8f9e642407a4a06f65eb90ca04)
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 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 };
102 
103 static struct iscsi_transport cxgb3i_iscsi_transport = {
104 	.owner		= THIS_MODULE,
105 	.name		= DRV_MODULE_NAME,
106 	/* owner and name should be set already */
107 	.caps		= CAP_RECOVERY_L0 | CAP_MULTI_R2T | CAP_HDRDGST
108 				| CAP_DATADGST | CAP_DIGEST_OFFLOAD |
109 				CAP_PADDING_OFFLOAD | CAP_TEXT_NEGO,
110 	.attr_is_visible	= cxgbi_attr_is_visible,
111 	.get_host_param	= cxgbi_get_host_param,
112 	.set_host_param	= cxgbi_set_host_param,
113 	/* session management */
114 	.create_session	= cxgbi_create_session,
115 	.destroy_session	= cxgbi_destroy_session,
116 	.get_session_param = iscsi_session_get_param,
117 	/* connection management */
118 	.create_conn	= cxgbi_create_conn,
119 	.bind_conn	= cxgbi_bind_conn,
120 	.unbind_conn	= iscsi_conn_unbind,
121 	.destroy_conn	= iscsi_tcp_conn_teardown,
122 	.start_conn	= iscsi_conn_start,
123 	.stop_conn	= iscsi_conn_stop,
124 	.get_conn_param	= iscsi_conn_get_param,
125 	.set_param	= cxgbi_set_conn_param,
126 	.get_stats	= cxgbi_get_conn_stats,
127 	/* pdu xmit req from user space */
128 	.send_pdu	= iscsi_conn_send_pdu,
129 	/* task */
130 	.init_task	= iscsi_tcp_task_init,
131 	.xmit_task	= iscsi_tcp_task_xmit,
132 	.cleanup_task	= cxgbi_cleanup_task,
133 	/* pdu */
134 	.alloc_pdu	= cxgbi_conn_alloc_pdu,
135 	.init_pdu	= cxgbi_conn_init_pdu,
136 	.xmit_pdu	= cxgbi_conn_xmit_pdu,
137 	.parse_pdu_itt	= cxgbi_parse_pdu_itt,
138 	/* TCP connect/disconnect */
139 	.get_ep_param	= cxgbi_get_ep_param,
140 	.ep_connect	= cxgbi_ep_connect,
141 	.ep_poll	= cxgbi_ep_poll,
142 	.ep_disconnect	= cxgbi_ep_disconnect,
143 	/* Error recovery timeout call */
144 	.session_recovery_timedout = iscsi_session_recovery_timedout,
145 };
146 
147 static struct scsi_transport_template *cxgb3i_stt;
148 
149 /*
150  * CPL (Chelsio Protocol Language) defines a message passing interface between
151  * the host driver and Chelsio asic.
152  * The section below implments CPLs that related to iscsi tcp connection
153  * open/close/abort and data send/receive.
154  */
155 
156 static int push_tx_frames(struct cxgbi_sock *csk, int req_completion);
157 
158 static void send_act_open_req(struct cxgbi_sock *csk, struct sk_buff *skb,
159 			      const struct l2t_entry *e)
160 {
161 	unsigned int wscale = cxgbi_sock_compute_wscale(csk->rcv_win);
162 	struct cpl_act_open_req *req = (struct cpl_act_open_req *)skb->head;
163 
164 	skb->priority = CPL_PRIORITY_SETUP;
165 
166 	req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
167 	OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_ACT_OPEN_REQ, csk->atid));
168 	req->local_port = csk->saddr.sin_port;
169 	req->peer_port = csk->daddr.sin_port;
170 	req->local_ip = csk->saddr.sin_addr.s_addr;
171 	req->peer_ip = csk->daddr.sin_addr.s_addr;
172 
173 	req->opt0h = htonl(V_KEEP_ALIVE(1) | F_TCAM_BYPASS |
174 			V_WND_SCALE(wscale) | V_MSS_IDX(csk->mss_idx) |
175 			V_L2T_IDX(e->idx) | V_TX_CHANNEL(e->smt_idx));
176 	req->opt0l = htonl(V_ULP_MODE(ULP2_MODE_ISCSI) |
177 			V_RCV_BUFSIZ(csk->rcv_win >> 10));
178 
179 	log_debug(1 << CXGBI_DBG_TOE | 1 << CXGBI_DBG_SOCK,
180 		"csk 0x%p,%u,0x%lx,%u, %pI4:%u-%pI4:%u, %u,%u,%u.\n",
181 		csk, csk->state, csk->flags, csk->atid,
182 		&req->local_ip, ntohs(req->local_port),
183 		&req->peer_ip, ntohs(req->peer_port),
184 		csk->mss_idx, e->idx, e->smt_idx);
185 
186 	l2t_send(csk->cdev->lldev, skb, csk->l2t);
187 }
188 
189 static inline void act_open_arp_failure(struct t3cdev *dev, struct sk_buff *skb)
190 {
191 	cxgbi_sock_act_open_req_arp_failure(NULL, skb);
192 }
193 
194 /*
195  * CPL connection close request: host ->
196  *
197  * Close a connection by sending a CPL_CLOSE_CON_REQ message and queue it to
198  * the write queue (i.e., after any unsent txt data).
199  */
200 static void send_close_req(struct cxgbi_sock *csk)
201 {
202 	struct sk_buff *skb = csk->cpl_close;
203 	struct cpl_close_con_req *req = (struct cpl_close_con_req *)skb->head;
204 	unsigned int tid = csk->tid;
205 
206 	log_debug(1 << CXGBI_DBG_TOE | 1 << CXGBI_DBG_SOCK,
207 		"csk 0x%p,%u,0x%lx,%u.\n",
208 		csk, csk->state, csk->flags, csk->tid);
209 
210 	csk->cpl_close = NULL;
211 	req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_OFLD_CLOSE_CON));
212 	req->wr.wr_lo = htonl(V_WR_TID(tid));
213 	OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_CLOSE_CON_REQ, tid));
214 	req->rsvd = htonl(csk->write_seq);
215 
216 	cxgbi_sock_skb_entail(csk, skb);
217 	if (csk->state >= CTP_ESTABLISHED)
218 		push_tx_frames(csk, 1);
219 }
220 
221 /*
222  * CPL connection abort request: host ->
223  *
224  * Send an ABORT_REQ message. Makes sure we do not send multiple ABORT_REQs
225  * for the same connection and also that we do not try to send a message
226  * after the connection has closed.
227  */
228 static void abort_arp_failure(struct t3cdev *tdev, struct sk_buff *skb)
229 {
230 	struct cpl_abort_req *req = cplhdr(skb);
231 
232 	log_debug(1 << CXGBI_DBG_TOE | 1 << CXGBI_DBG_SOCK,
233 		"t3dev 0x%p, tid %u, skb 0x%p.\n",
234 		tdev, GET_TID(req), skb);
235 	req->cmd = CPL_ABORT_NO_RST;
236 	cxgb3_ofld_send(tdev, skb);
237 }
238 
239 static void send_abort_req(struct cxgbi_sock *csk)
240 {
241 	struct sk_buff *skb = csk->cpl_abort_req;
242 	struct cpl_abort_req *req;
243 
244 	if (unlikely(csk->state == CTP_ABORTING || !skb))
245 		return;
246 	cxgbi_sock_set_state(csk, CTP_ABORTING);
247 	cxgbi_sock_set_flag(csk, CTPF_ABORT_RPL_PENDING);
248 	/* Purge the send queue so we don't send anything after an abort. */
249 	cxgbi_sock_purge_write_queue(csk);
250 
251 	csk->cpl_abort_req = NULL;
252 	req = (struct cpl_abort_req *)skb->head;
253 	skb->priority = CPL_PRIORITY_DATA;
254 	set_arp_failure_handler(skb, abort_arp_failure);
255 	req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_OFLD_HOST_ABORT_CON_REQ));
256 	req->wr.wr_lo = htonl(V_WR_TID(csk->tid));
257 	OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_ABORT_REQ, csk->tid));
258 	req->rsvd0 = htonl(csk->snd_nxt);
259 	req->rsvd1 = !cxgbi_sock_flag(csk, CTPF_TX_DATA_SENT);
260 	req->cmd = CPL_ABORT_SEND_RST;
261 
262 	log_debug(1 << CXGBI_DBG_TOE | 1 << CXGBI_DBG_SOCK,
263 		"csk 0x%p,%u,0x%lx,%u, snd_nxt %u, 0x%x.\n",
264 		csk, csk->state, csk->flags, csk->tid, csk->snd_nxt,
265 		req->rsvd1);
266 
267 	l2t_send(csk->cdev->lldev, skb, csk->l2t);
268 }
269 
270 /*
271  * CPL connection abort reply: host ->
272  *
273  * Send an ABORT_RPL message in response of the ABORT_REQ received.
274  */
275 static void send_abort_rpl(struct cxgbi_sock *csk, int rst_status)
276 {
277 	struct sk_buff *skb = csk->cpl_abort_rpl;
278 	struct cpl_abort_rpl *rpl = (struct cpl_abort_rpl *)skb->head;
279 
280 	log_debug(1 << CXGBI_DBG_TOE | 1 << CXGBI_DBG_SOCK,
281 		"csk 0x%p,%u,0x%lx,%u, status %d.\n",
282 		csk, csk->state, csk->flags, csk->tid, rst_status);
283 
284 	csk->cpl_abort_rpl = NULL;
285 	skb->priority = CPL_PRIORITY_DATA;
286 	rpl->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_OFLD_HOST_ABORT_CON_RPL));
287 	rpl->wr.wr_lo = htonl(V_WR_TID(csk->tid));
288 	OPCODE_TID(rpl) = htonl(MK_OPCODE_TID(CPL_ABORT_RPL, csk->tid));
289 	rpl->cmd = rst_status;
290 	cxgb3_ofld_send(csk->cdev->lldev, skb);
291 }
292 
293 /*
294  * CPL connection rx data ack: host ->
295  * Send RX credits through an RX_DATA_ACK CPL message. Returns the number of
296  * credits sent.
297  */
298 static u32 send_rx_credits(struct cxgbi_sock *csk, u32 credits)
299 {
300 	struct sk_buff *skb;
301 	struct cpl_rx_data_ack *req;
302 	u32 dack = F_RX_DACK_CHANGE | V_RX_DACK_MODE(1);
303 
304 	log_debug(1 << CXGBI_DBG_TOE | 1 << CXGBI_DBG_PDU_RX,
305 		"csk 0x%p,%u,0x%lx,%u, credit %u, dack %u.\n",
306 		csk, csk->state, csk->flags, csk->tid, credits, dack);
307 
308 	skb = alloc_wr(sizeof(*req), 0, GFP_ATOMIC);
309 	if (!skb) {
310 		pr_info("csk 0x%p, credit %u, OOM.\n", csk, credits);
311 		return 0;
312 	}
313 	req = (struct cpl_rx_data_ack *)skb->head;
314 	req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
315 	OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_RX_DATA_ACK, csk->tid));
316 	req->credit_dack = htonl(F_RX_DACK_CHANGE | V_RX_DACK_MODE(1) |
317 				V_RX_CREDITS(credits));
318 	skb->priority = CPL_PRIORITY_ACK;
319 	cxgb3_ofld_send(csk->cdev->lldev, skb);
320 	return credits;
321 }
322 
323 /*
324  * CPL connection tx data: host ->
325  *
326  * Send iscsi PDU via TX_DATA CPL message. Returns the number of
327  * credits sent.
328  * Each TX_DATA consumes work request credit (wrs), so we need to keep track of
329  * how many we've used so far and how many are pending (i.e., yet ack'ed by T3).
330  */
331 
332 static unsigned int wrlen __read_mostly;
333 static unsigned int skb_wrs[SKB_WR_LIST_SIZE] __read_mostly;
334 
335 static void init_wr_tab(unsigned int wr_len)
336 {
337 	int i;
338 
339 	if (skb_wrs[1])		/* already initialized */
340 		return;
341 	for (i = 1; i < SKB_WR_LIST_SIZE; i++) {
342 		int sgl_len = (3 * i) / 2 + (i & 1);
343 
344 		sgl_len += 3;
345 		skb_wrs[i] = (sgl_len <= wr_len
346 			      ? 1 : 1 + (sgl_len - 2) / (wr_len - 1));
347 	}
348 	wrlen = wr_len * 8;
349 }
350 
351 static inline void make_tx_data_wr(struct cxgbi_sock *csk, struct sk_buff *skb,
352 				   int len, int req_completion)
353 {
354 	struct tx_data_wr *req;
355 	struct l2t_entry *l2t = csk->l2t;
356 
357 	skb_reset_transport_header(skb);
358 	req = __skb_push(skb, sizeof(*req));
359 	req->wr_hi = htonl(V_WR_OP(FW_WROPCODE_OFLD_TX_DATA) |
360 			(req_completion ? F_WR_COMPL : 0));
361 	req->wr_lo = htonl(V_WR_TID(csk->tid));
362 	/* len includes the length of any HW ULP additions */
363 	req->len = htonl(len);
364 	/* V_TX_ULP_SUBMODE sets both the mode and submode */
365 	req->flags = htonl(V_TX_ULP_SUBMODE(cxgbi_skcb_tx_ulp_mode(skb)) |
366 			   V_TX_SHOVE((skb_peek(&csk->write_queue) ? 0 : 1)));
367 	req->sndseq = htonl(csk->snd_nxt);
368 	req->param = htonl(V_TX_PORT(l2t->smt_idx));
369 
370 	if (!cxgbi_sock_flag(csk, CTPF_TX_DATA_SENT)) {
371 		req->flags |= htonl(V_TX_ACK_PAGES(2) | F_TX_INIT |
372 				    V_TX_CPU_IDX(csk->rss_qid));
373 		/* sendbuffer is in units of 32KB. */
374 		req->param |= htonl(V_TX_SNDBUF(csk->snd_win >> 15));
375 		cxgbi_sock_set_flag(csk, CTPF_TX_DATA_SENT);
376 	}
377 }
378 
379 /*
380  * push_tx_frames -- start transmit
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_tx_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(struct timer_list *t)
548 {
549 	struct cxgbi_sock *csk = from_timer(csk, t, retry_timer);
550 	struct sk_buff *skb;
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:
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 static void l2t_put(struct cxgbi_sock *csk)
889 {
890 	struct t3cdev *t3dev = (struct t3cdev *)csk->cdev->lldev;
891 
892 	if (csk->l2t) {
893 		l2t_release(t3dev, csk->l2t);
894 		csk->l2t = NULL;
895 		cxgbi_sock_put(csk);
896 	}
897 }
898 
899 /*
900  * release_offload_resources - release offload resource
901  * Release resources held by an offload connection (TID, L2T entry, etc.)
902  */
903 static void release_offload_resources(struct cxgbi_sock *csk)
904 {
905 	struct t3cdev *t3dev = (struct t3cdev *)csk->cdev->lldev;
906 
907 	log_debug(1 << CXGBI_DBG_TOE | 1 << CXGBI_DBG_SOCK,
908 		"csk 0x%p,%u,0x%lx,%u.\n",
909 		csk, csk->state, csk->flags, csk->tid);
910 
911 	csk->rss_qid = 0;
912 	cxgbi_sock_free_cpl_skbs(csk);
913 
914 	if (csk->wr_cred != csk->wr_max_cred) {
915 		cxgbi_sock_purge_wr_queue(csk);
916 		cxgbi_sock_reset_wr_list(csk);
917 	}
918 	l2t_put(csk);
919 	if (cxgbi_sock_flag(csk, CTPF_HAS_ATID))
920 		free_atid(csk);
921 	else if (cxgbi_sock_flag(csk, CTPF_HAS_TID)) {
922 		cxgb3_remove_tid(t3dev, (void *)csk, csk->tid);
923 		cxgbi_sock_clear_flag(csk, CTPF_HAS_TID);
924 		cxgbi_sock_put(csk);
925 	}
926 	csk->dst = NULL;
927 	csk->cdev = NULL;
928 }
929 
930 static void update_address(struct cxgbi_hba *chba)
931 {
932 	if (chba->ipv4addr) {
933 		if (chba->vdev &&
934 		    chba->ipv4addr != cxgb3i_get_private_ipv4addr(chba->vdev)) {
935 			cxgb3i_set_private_ipv4addr(chba->vdev, chba->ipv4addr);
936 			cxgb3i_set_private_ipv4addr(chba->ndev, 0);
937 			pr_info("%s set %pI4.\n",
938 				chba->vdev->name, &chba->ipv4addr);
939 		} else if (chba->ipv4addr !=
940 				cxgb3i_get_private_ipv4addr(chba->ndev)) {
941 			cxgb3i_set_private_ipv4addr(chba->ndev, chba->ipv4addr);
942 			pr_info("%s set %pI4.\n",
943 				chba->ndev->name, &chba->ipv4addr);
944 		}
945 	} else if (cxgb3i_get_private_ipv4addr(chba->ndev)) {
946 		if (chba->vdev)
947 			cxgb3i_set_private_ipv4addr(chba->vdev, 0);
948 		cxgb3i_set_private_ipv4addr(chba->ndev, 0);
949 	}
950 }
951 
952 static int init_act_open(struct cxgbi_sock *csk)
953 {
954 	struct dst_entry *dst = csk->dst;
955 	struct cxgbi_device *cdev = csk->cdev;
956 	struct t3cdev *t3dev = (struct t3cdev *)cdev->lldev;
957 	struct net_device *ndev = cdev->ports[csk->port_id];
958 	struct cxgbi_hba *chba = cdev->hbas[csk->port_id];
959 	struct sk_buff *skb = NULL;
960 	int ret;
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 		ret = -EINVAL;
982 		goto put_sock;
983 	}
984 	cxgbi_sock_set_flag(csk, CTPF_HAS_ATID);
985 	cxgbi_sock_get(csk);
986 
987 	skb = alloc_wr(sizeof(struct cpl_act_open_req), 0, GFP_KERNEL);
988 	if (!skb) {
989 		ret = -ENOMEM;
990 		goto free_atid;
991 	}
992 	skb->sk = (struct sock *)csk;
993 	set_arp_failure_handler(skb, act_open_arp_failure);
994 	csk->snd_win = cxgb3i_snd_win;
995 	csk->rcv_win = cxgb3i_rcv_win;
996 
997 	csk->wr_max_cred = csk->wr_cred = T3C_DATA(t3dev)->max_wrs - 1;
998 	csk->wr_una_cred = 0;
999 	csk->mss_idx = cxgbi_sock_select_mss(csk, dst_mtu(dst));
1000 	cxgbi_sock_reset_wr_list(csk);
1001 	csk->err = 0;
1002 
1003 	log_debug(1 << CXGBI_DBG_TOE | 1 << CXGBI_DBG_SOCK,
1004 		"csk 0x%p,%u,0x%lx, %pI4:%u-%pI4:%u.\n",
1005 		csk, csk->state, csk->flags,
1006 		&csk->saddr.sin_addr.s_addr, ntohs(csk->saddr.sin_port),
1007 		&csk->daddr.sin_addr.s_addr, ntohs(csk->daddr.sin_port));
1008 
1009 	cxgbi_sock_set_state(csk, CTP_ACTIVE_OPEN);
1010 	send_act_open_req(csk, skb, csk->l2t);
1011 	return 0;
1012 
1013 free_atid:
1014 	cxgb3_free_atid(t3dev, csk->atid);
1015 put_sock:
1016 	cxgbi_sock_put(csk);
1017 	l2t_release(t3dev, csk->l2t);
1018 	csk->l2t = NULL;
1019 
1020 	return ret;
1021 }
1022 
1023 cxgb3_cpl_handler_func cxgb3i_cpl_handlers[NUM_CPL_CMDS] = {
1024 	[CPL_ACT_ESTABLISH] = do_act_establish,
1025 	[CPL_ACT_OPEN_RPL] = do_act_open_rpl,
1026 	[CPL_PEER_CLOSE] = do_peer_close,
1027 	[CPL_ABORT_REQ_RSS] = do_abort_req,
1028 	[CPL_ABORT_RPL_RSS] = do_abort_rpl,
1029 	[CPL_CLOSE_CON_RPL] = do_close_con_rpl,
1030 	[CPL_TX_DMA_ACK] = do_wr_ack,
1031 	[CPL_ISCSI_HDR] = do_iscsi_hdr,
1032 };
1033 
1034 /**
1035  * cxgb3i_ofld_init - allocate and initialize resources for each adapter found
1036  * @cdev:	cxgbi adapter
1037  */
1038 static int cxgb3i_ofld_init(struct cxgbi_device *cdev)
1039 {
1040 	struct t3cdev *t3dev = (struct t3cdev *)cdev->lldev;
1041 	struct adap_ports port;
1042 	struct ofld_page_info rx_page_info;
1043 	unsigned int wr_len;
1044 	int rc;
1045 
1046 	if (t3dev->ctl(t3dev, GET_WR_LEN, &wr_len) < 0 ||
1047 	    t3dev->ctl(t3dev, GET_PORTS, &port) < 0 ||
1048 	    t3dev->ctl(t3dev, GET_RX_PAGE_INFO, &rx_page_info) < 0) {
1049 		pr_warn("t3 0x%p, offload up, ioctl failed.\n", t3dev);
1050 		return -EINVAL;
1051 	}
1052 
1053 	if (cxgb3i_max_connect > CXGBI_MAX_CONN)
1054 		cxgb3i_max_connect = CXGBI_MAX_CONN;
1055 
1056 	rc = cxgbi_device_portmap_create(cdev, cxgb3i_sport_base,
1057 					cxgb3i_max_connect);
1058 	if (rc < 0)
1059 		return rc;
1060 
1061 	init_wr_tab(wr_len);
1062 	cdev->csk_release_offload_resources = release_offload_resources;
1063 	cdev->csk_push_tx_frames = push_tx_frames;
1064 	cdev->csk_send_abort_req = send_abort_req;
1065 	cdev->csk_send_close_req = send_close_req;
1066 	cdev->csk_send_rx_credits = send_rx_credits;
1067 	cdev->csk_alloc_cpls = alloc_cpls;
1068 	cdev->csk_init_act_open = init_act_open;
1069 
1070 	pr_info("cdev 0x%p, offload up, added.\n", cdev);
1071 	return 0;
1072 }
1073 
1074 /*
1075  * functions to program the pagepod in h/w
1076  */
1077 static inline void ulp_mem_io_set_hdr(struct sk_buff *skb, unsigned int addr)
1078 {
1079 	struct ulp_mem_io *req = (struct ulp_mem_io *)skb->head;
1080 
1081 	memset(req, 0, sizeof(*req));
1082 
1083 	req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_BYPASS));
1084 	req->cmd_lock_addr = htonl(V_ULP_MEMIO_ADDR(addr >> 5) |
1085 				   V_ULPTX_CMD(ULP_MEM_WRITE));
1086 	req->len = htonl(V_ULP_MEMIO_DATA_LEN(IPPOD_SIZE >> 5) |
1087 			 V_ULPTX_NFLITS((IPPOD_SIZE >> 3) + 1));
1088 }
1089 
1090 static struct cxgbi_ppm *cdev2ppm(struct cxgbi_device *cdev)
1091 {
1092 	return ((struct t3cdev *)cdev->lldev)->ulp_iscsi;
1093 }
1094 
1095 static int ddp_set_map(struct cxgbi_ppm *ppm, struct cxgbi_sock *csk,
1096 		       struct cxgbi_task_tag_info *ttinfo)
1097 {
1098 	unsigned int idx = ttinfo->idx;
1099 	unsigned int npods = ttinfo->npods;
1100 	struct scatterlist *sg = ttinfo->sgl;
1101 	struct cxgbi_pagepod *ppod;
1102 	struct ulp_mem_io *req;
1103 	unsigned int sg_off;
1104 	unsigned int pm_addr = (idx << PPOD_SIZE_SHIFT) + ppm->llimit;
1105 	int i;
1106 
1107 	for (i = 0; i < npods; i++, idx++, pm_addr += IPPOD_SIZE) {
1108 		struct sk_buff *skb = alloc_wr(sizeof(struct ulp_mem_io) +
1109 					       IPPOD_SIZE, 0, GFP_ATOMIC);
1110 
1111 		if (!skb)
1112 			return -ENOMEM;
1113 		ulp_mem_io_set_hdr(skb, pm_addr);
1114 		req = (struct ulp_mem_io *)skb->head;
1115 		ppod = (struct cxgbi_pagepod *)(req + 1);
1116 		sg_off = i * PPOD_PAGES_MAX;
1117 		cxgbi_ddp_set_one_ppod(ppod, ttinfo, &sg,
1118 				       &sg_off);
1119 		skb->priority = CPL_PRIORITY_CONTROL;
1120 		cxgb3_ofld_send(ppm->lldev, skb);
1121 	}
1122 	return 0;
1123 }
1124 
1125 static void ddp_clear_map(struct cxgbi_device *cdev, struct cxgbi_ppm *ppm,
1126 			  struct cxgbi_task_tag_info *ttinfo)
1127 {
1128 	unsigned int idx = ttinfo->idx;
1129 	unsigned int pm_addr = (idx << PPOD_SIZE_SHIFT) + ppm->llimit;
1130 	unsigned int npods = ttinfo->npods;
1131 	int i;
1132 
1133 	log_debug(1 << CXGBI_DBG_DDP,
1134 		  "cdev 0x%p, clear idx %u, npods %u.\n",
1135 		  cdev, idx, npods);
1136 
1137 	for (i = 0; i < npods; i++, idx++, pm_addr += IPPOD_SIZE) {
1138 		struct sk_buff *skb = alloc_wr(sizeof(struct ulp_mem_io) +
1139 					       IPPOD_SIZE, 0, GFP_ATOMIC);
1140 
1141 		if (!skb) {
1142 			pr_err("cdev 0x%p, clear ddp, %u,%d/%u, skb OOM.\n",
1143 			       cdev, idx, i, npods);
1144 			continue;
1145 		}
1146 		ulp_mem_io_set_hdr(skb, pm_addr);
1147 		skb->priority = CPL_PRIORITY_CONTROL;
1148 		cxgb3_ofld_send(ppm->lldev, skb);
1149 	}
1150 }
1151 
1152 static int ddp_setup_conn_pgidx(struct cxgbi_sock *csk,
1153 				unsigned int tid, int pg_idx)
1154 {
1155 	struct sk_buff *skb = alloc_wr(sizeof(struct cpl_set_tcb_field), 0,
1156 					GFP_KERNEL);
1157 	struct cpl_set_tcb_field *req;
1158 	u64 val = pg_idx < DDP_PGIDX_MAX ? pg_idx : 0;
1159 
1160 	log_debug(1 << CXGBI_DBG_DDP,
1161 		"csk 0x%p, tid %u, pg_idx %d.\n", csk, tid, pg_idx);
1162 	if (!skb)
1163 		return -ENOMEM;
1164 
1165 	/* set up ulp submode and page size */
1166 	req = (struct cpl_set_tcb_field *)skb->head;
1167 	req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
1168 	OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_SET_TCB_FIELD, tid));
1169 	req->reply = V_NO_REPLY(1);
1170 	req->cpu_idx = 0;
1171 	req->word = htons(31);
1172 	req->mask = cpu_to_be64(0xF0000000);
1173 	req->val = cpu_to_be64(val << 28);
1174 	skb->priority = CPL_PRIORITY_CONTROL;
1175 
1176 	cxgb3_ofld_send(csk->cdev->lldev, skb);
1177 	return 0;
1178 }
1179 
1180 /**
1181  * ddp_setup_conn_digest - setup conn. digest setting
1182  * @csk: cxgb tcp socket
1183  * @tid: connection id
1184  * @hcrc: header digest enabled
1185  * @dcrc: data digest enabled
1186  * set up the iscsi digest settings for a connection identified by tid
1187  */
1188 static int ddp_setup_conn_digest(struct cxgbi_sock *csk, unsigned int tid,
1189 				 int hcrc, int dcrc)
1190 {
1191 	struct sk_buff *skb = alloc_wr(sizeof(struct cpl_set_tcb_field), 0,
1192 					GFP_KERNEL);
1193 	struct cpl_set_tcb_field *req;
1194 	u64 val = (hcrc ? 1 : 0) | (dcrc ? 2 : 0);
1195 
1196 	log_debug(1 << CXGBI_DBG_DDP,
1197 		"csk 0x%p, tid %u, crc %d,%d.\n", csk, tid, hcrc, dcrc);
1198 	if (!skb)
1199 		return -ENOMEM;
1200 
1201 	/* set up ulp submode and page size */
1202 	req = (struct cpl_set_tcb_field *)skb->head;
1203 	req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
1204 	OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_SET_TCB_FIELD, tid));
1205 	req->reply = V_NO_REPLY(1);
1206 	req->cpu_idx = 0;
1207 	req->word = htons(31);
1208 	req->mask = cpu_to_be64(0x0F000000);
1209 	req->val = cpu_to_be64(val << 24);
1210 	skb->priority = CPL_PRIORITY_CONTROL;
1211 
1212 	cxgb3_ofld_send(csk->cdev->lldev, skb);
1213 	return 0;
1214 }
1215 
1216 /**
1217  * cxgb3i_ddp_init - initialize the cxgb3 adapter's ddp resource
1218  * @cdev: cxgb3i adapter
1219  * initialize the ddp pagepod manager for a given adapter
1220  */
1221 static int cxgb3i_ddp_init(struct cxgbi_device *cdev)
1222 {
1223 	struct t3cdev *tdev = (struct t3cdev *)cdev->lldev;
1224 	struct net_device *ndev = cdev->ports[0];
1225 	struct cxgbi_tag_format tformat;
1226 	unsigned int ppmax, tagmask = 0;
1227 	struct ulp_iscsi_info uinfo;
1228 	int i, err;
1229 
1230 	err = tdev->ctl(tdev, ULP_ISCSI_GET_PARAMS, &uinfo);
1231 	if (err < 0) {
1232 		pr_err("%s, failed to get iscsi param %d.\n",
1233 		       ndev->name, err);
1234 		return err;
1235 	}
1236 	if (uinfo.llimit >= uinfo.ulimit) {
1237 		pr_warn("T3 %s, iscsi NOT enabled %u ~ %u!\n",
1238 			ndev->name, uinfo.llimit, uinfo.ulimit);
1239 		return -EACCES;
1240 	}
1241 
1242 	ppmax = (uinfo.ulimit - uinfo.llimit + 1) >> PPOD_SIZE_SHIFT;
1243 	tagmask = cxgbi_tagmask_set(ppmax);
1244 
1245 	pr_info("T3 %s: 0x%x~0x%x, 0x%x, tagmask 0x%x -> 0x%x.\n",
1246 		ndev->name, uinfo.llimit, uinfo.ulimit, ppmax, uinfo.tagmask,
1247 		tagmask);
1248 
1249 	memset(&tformat, 0, sizeof(struct cxgbi_tag_format));
1250 	for (i = 0; i < 4; i++)
1251 		tformat.pgsz_order[i] = uinfo.pgsz_factor[i];
1252 	cxgbi_tagmask_check(tagmask, &tformat);
1253 
1254 	err = cxgbi_ddp_ppm_setup(&tdev->ulp_iscsi, cdev, &tformat,
1255 				  (uinfo.ulimit - uinfo.llimit + 1),
1256 				  uinfo.llimit, uinfo.llimit, 0, 0, 0);
1257 	if (err)
1258 		return err;
1259 
1260 	if (!(cdev->flags & CXGBI_FLAG_DDP_OFF)) {
1261 		uinfo.tagmask = tagmask;
1262 		uinfo.ulimit = uinfo.llimit + (ppmax << PPOD_SIZE_SHIFT);
1263 
1264 		err = tdev->ctl(tdev, ULP_ISCSI_SET_PARAMS, &uinfo);
1265 		if (err < 0) {
1266 			pr_err("T3 %s fail to set iscsi param %d.\n",
1267 			       ndev->name, err);
1268 			cdev->flags |= CXGBI_FLAG_DDP_OFF;
1269 		}
1270 		err = 0;
1271 	}
1272 
1273 	cdev->csk_ddp_setup_digest = ddp_setup_conn_digest;
1274 	cdev->csk_ddp_setup_pgidx = ddp_setup_conn_pgidx;
1275 	cdev->csk_ddp_set_map = ddp_set_map;
1276 	cdev->csk_ddp_clear_map = ddp_clear_map;
1277 	cdev->cdev2ppm = cdev2ppm;
1278 	cdev->tx_max_size = min_t(unsigned int, ULP2_MAX_PDU_PAYLOAD,
1279 				  uinfo.max_txsz - ISCSI_PDU_NONPAYLOAD_LEN);
1280 	cdev->rx_max_size = min_t(unsigned int, ULP2_MAX_PDU_PAYLOAD,
1281 				  uinfo.max_rxsz - ISCSI_PDU_NONPAYLOAD_LEN);
1282 
1283 	return 0;
1284 }
1285 
1286 static void cxgb3i_dev_close(struct t3cdev *t3dev)
1287 {
1288 	struct cxgbi_device *cdev = cxgbi_device_find_by_lldev(t3dev);
1289 
1290 	if (!cdev || cdev->flags & CXGBI_FLAG_ADAPTER_RESET) {
1291 		pr_info("0x%p close, f 0x%x.\n", cdev, cdev ? cdev->flags : 0);
1292 		return;
1293 	}
1294 
1295 	cxgbi_device_unregister(cdev);
1296 }
1297 
1298 /**
1299  * cxgb3i_dev_open - init a t3 adapter structure and any h/w settings
1300  * @t3dev: t3cdev adapter
1301  */
1302 static void cxgb3i_dev_open(struct t3cdev *t3dev)
1303 {
1304 	struct cxgbi_device *cdev = cxgbi_device_find_by_lldev(t3dev);
1305 	struct adapter *adapter = tdev2adap(t3dev);
1306 	int i, err;
1307 
1308 	if (cdev) {
1309 		pr_info("0x%p, updating.\n", cdev);
1310 		return;
1311 	}
1312 
1313 	cdev = cxgbi_device_register(0, adapter->params.nports);
1314 	if (!cdev) {
1315 		pr_warn("device 0x%p register failed.\n", t3dev);
1316 		return;
1317 	}
1318 
1319 	cdev->flags = CXGBI_FLAG_DEV_T3 | CXGBI_FLAG_IPV4_SET;
1320 	cdev->lldev = t3dev;
1321 	cdev->pdev = adapter->pdev;
1322 	cdev->ports = adapter->port;
1323 	cdev->nports = adapter->params.nports;
1324 	cdev->mtus = adapter->params.mtus;
1325 	cdev->nmtus = NMTUS;
1326 	cdev->rx_credit_thres = cxgb3i_rx_credit_thres;
1327 	cdev->skb_tx_rsvd = CXGB3I_TX_HEADER_LEN;
1328 	cdev->skb_rx_extra = sizeof(struct cpl_iscsi_hdr_norss);
1329 	cdev->itp = &cxgb3i_iscsi_transport;
1330 
1331 	err = cxgb3i_ddp_init(cdev);
1332 	if (err) {
1333 		pr_info("0x%p ddp init failed %d\n", cdev, err);
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