xref: /openbmc/linux/drivers/scsi/cxgbi/libcxgbi.c (revision 63dc02bd)
1 /*
2  * libcxgbi.c: Chelsio common library for T3/T4 iSCSI driver.
3  *
4  * Copyright (c) 2010 Chelsio Communications, Inc.
5  *
6  * This program is free software; you can redistribute it and/or modify
7  * it under the terms of the GNU General Public License as published by
8  * the Free Software Foundation.
9  *
10  * Written by: Karen Xie (kxie@chelsio.com)
11  * Written by: Rakesh Ranjan (rranjan@chelsio.com)
12  */
13 
14 #define pr_fmt(fmt)	KBUILD_MODNAME ":%s: " fmt, __func__
15 
16 #include <linux/skbuff.h>
17 #include <linux/crypto.h>
18 #include <linux/scatterlist.h>
19 #include <linux/pci.h>
20 #include <scsi/scsi.h>
21 #include <scsi/scsi_cmnd.h>
22 #include <scsi/scsi_host.h>
23 #include <linux/if_vlan.h>
24 #include <linux/inet.h>
25 #include <net/dst.h>
26 #include <net/route.h>
27 #include <linux/inetdevice.h>	/* ip_dev_find */
28 #include <linux/module.h>
29 #include <net/tcp.h>
30 
31 static unsigned int dbg_level;
32 
33 #include "libcxgbi.h"
34 
35 #define DRV_MODULE_NAME		"libcxgbi"
36 #define DRV_MODULE_DESC		"Chelsio iSCSI driver library"
37 #define DRV_MODULE_VERSION	"0.9.0"
38 #define DRV_MODULE_RELDATE	"Jun. 2010"
39 
40 MODULE_AUTHOR("Chelsio Communications, Inc.");
41 MODULE_DESCRIPTION(DRV_MODULE_DESC);
42 MODULE_VERSION(DRV_MODULE_VERSION);
43 MODULE_LICENSE("GPL");
44 
45 module_param(dbg_level, uint, 0644);
46 MODULE_PARM_DESC(dbg_level, "libiscsi debug level (default=0)");
47 
48 
49 /*
50  * cxgbi device management
51  * maintains a list of the cxgbi devices
52  */
53 static LIST_HEAD(cdev_list);
54 static DEFINE_MUTEX(cdev_mutex);
55 
56 int cxgbi_device_portmap_create(struct cxgbi_device *cdev, unsigned int base,
57 				unsigned int max_conn)
58 {
59 	struct cxgbi_ports_map *pmap = &cdev->pmap;
60 
61 	pmap->port_csk = cxgbi_alloc_big_mem(max_conn *
62 					     sizeof(struct cxgbi_sock *),
63 					     GFP_KERNEL);
64 	if (!pmap->port_csk) {
65 		pr_warn("cdev 0x%p, portmap OOM %u.\n", cdev, max_conn);
66 		return -ENOMEM;
67 	}
68 
69 	pmap->max_connect = max_conn;
70 	pmap->sport_base = base;
71 	spin_lock_init(&pmap->lock);
72 	return 0;
73 }
74 EXPORT_SYMBOL_GPL(cxgbi_device_portmap_create);
75 
76 void cxgbi_device_portmap_cleanup(struct cxgbi_device *cdev)
77 {
78 	struct cxgbi_ports_map *pmap = &cdev->pmap;
79 	struct cxgbi_sock *csk;
80 	int i;
81 
82 	for (i = 0; i < pmap->max_connect; i++) {
83 		if (pmap->port_csk[i]) {
84 			csk = pmap->port_csk[i];
85 			pmap->port_csk[i] = NULL;
86 			log_debug(1 << CXGBI_DBG_SOCK,
87 				"csk 0x%p, cdev 0x%p, offload down.\n",
88 				csk, cdev);
89 			spin_lock_bh(&csk->lock);
90 			cxgbi_sock_set_flag(csk, CTPF_OFFLOAD_DOWN);
91 			cxgbi_sock_closed(csk);
92 			spin_unlock_bh(&csk->lock);
93 			cxgbi_sock_put(csk);
94 		}
95 	}
96 }
97 EXPORT_SYMBOL_GPL(cxgbi_device_portmap_cleanup);
98 
99 static inline void cxgbi_device_destroy(struct cxgbi_device *cdev)
100 {
101 	log_debug(1 << CXGBI_DBG_DEV,
102 		"cdev 0x%p, p# %u.\n", cdev, cdev->nports);
103 	cxgbi_hbas_remove(cdev);
104 	cxgbi_device_portmap_cleanup(cdev);
105 	if (cdev->dev_ddp_cleanup)
106 		cdev->dev_ddp_cleanup(cdev);
107 	else
108 		cxgbi_ddp_cleanup(cdev);
109 	if (cdev->ddp)
110 		cxgbi_ddp_cleanup(cdev);
111 	if (cdev->pmap.max_connect)
112 		cxgbi_free_big_mem(cdev->pmap.port_csk);
113 	kfree(cdev);
114 }
115 
116 struct cxgbi_device *cxgbi_device_register(unsigned int extra,
117 					   unsigned int nports)
118 {
119 	struct cxgbi_device *cdev;
120 
121 	cdev = kzalloc(sizeof(*cdev) + extra + nports *
122 			(sizeof(struct cxgbi_hba *) +
123 			 sizeof(struct net_device *)),
124 			GFP_KERNEL);
125 	if (!cdev) {
126 		pr_warn("nport %d, OOM.\n", nports);
127 		return NULL;
128 	}
129 	cdev->ports = (struct net_device **)(cdev + 1);
130 	cdev->hbas = (struct cxgbi_hba **)(((char*)cdev->ports) + nports *
131 						sizeof(struct net_device *));
132 	if (extra)
133 		cdev->dd_data = ((char *)cdev->hbas) +
134 				nports * sizeof(struct cxgbi_hba *);
135 	spin_lock_init(&cdev->pmap.lock);
136 
137 	mutex_lock(&cdev_mutex);
138 	list_add_tail(&cdev->list_head, &cdev_list);
139 	mutex_unlock(&cdev_mutex);
140 
141 	log_debug(1 << CXGBI_DBG_DEV,
142 		"cdev 0x%p, p# %u.\n", cdev, nports);
143 	return cdev;
144 }
145 EXPORT_SYMBOL_GPL(cxgbi_device_register);
146 
147 void cxgbi_device_unregister(struct cxgbi_device *cdev)
148 {
149 	log_debug(1 << CXGBI_DBG_DEV,
150 		"cdev 0x%p, p# %u,%s.\n",
151 		cdev, cdev->nports, cdev->nports ? cdev->ports[0]->name : "");
152 	mutex_lock(&cdev_mutex);
153 	list_del(&cdev->list_head);
154 	mutex_unlock(&cdev_mutex);
155 	cxgbi_device_destroy(cdev);
156 }
157 EXPORT_SYMBOL_GPL(cxgbi_device_unregister);
158 
159 void cxgbi_device_unregister_all(unsigned int flag)
160 {
161 	struct cxgbi_device *cdev, *tmp;
162 
163 	mutex_lock(&cdev_mutex);
164 	list_for_each_entry_safe(cdev, tmp, &cdev_list, list_head) {
165 		if ((cdev->flags & flag) == flag) {
166 			log_debug(1 << CXGBI_DBG_DEV,
167 				"cdev 0x%p, p# %u,%s.\n",
168 				cdev, cdev->nports, cdev->nports ?
169 				 cdev->ports[0]->name : "");
170 			list_del(&cdev->list_head);
171 			cxgbi_device_destroy(cdev);
172 		}
173 	}
174 	mutex_unlock(&cdev_mutex);
175 }
176 EXPORT_SYMBOL_GPL(cxgbi_device_unregister_all);
177 
178 struct cxgbi_device *cxgbi_device_find_by_lldev(void *lldev)
179 {
180 	struct cxgbi_device *cdev, *tmp;
181 
182 	mutex_lock(&cdev_mutex);
183 	list_for_each_entry_safe(cdev, tmp, &cdev_list, list_head) {
184 		if (cdev->lldev == lldev) {
185 			mutex_unlock(&cdev_mutex);
186 			return cdev;
187 		}
188 	}
189 	mutex_unlock(&cdev_mutex);
190 	log_debug(1 << CXGBI_DBG_DEV,
191 		"lldev 0x%p, NO match found.\n", lldev);
192 	return NULL;
193 }
194 EXPORT_SYMBOL_GPL(cxgbi_device_find_by_lldev);
195 
196 static struct cxgbi_device *cxgbi_device_find_by_netdev(struct net_device *ndev,
197 							int *port)
198 {
199 	struct net_device *vdev = NULL;
200 	struct cxgbi_device *cdev, *tmp;
201 	int i;
202 
203 	if (ndev->priv_flags & IFF_802_1Q_VLAN) {
204 		vdev = ndev;
205 		ndev = vlan_dev_real_dev(ndev);
206 		log_debug(1 << CXGBI_DBG_DEV,
207 			"vlan dev %s -> %s.\n", vdev->name, ndev->name);
208 	}
209 
210 	mutex_lock(&cdev_mutex);
211 	list_for_each_entry_safe(cdev, tmp, &cdev_list, list_head) {
212 		for (i = 0; i < cdev->nports; i++) {
213 			if (ndev == cdev->ports[i]) {
214 				cdev->hbas[i]->vdev = vdev;
215 				mutex_unlock(&cdev_mutex);
216 				if (port)
217 					*port = i;
218 				return cdev;
219 			}
220 		}
221 	}
222 	mutex_unlock(&cdev_mutex);
223 	log_debug(1 << CXGBI_DBG_DEV,
224 		"ndev 0x%p, %s, NO match found.\n", ndev, ndev->name);
225 	return NULL;
226 }
227 
228 void cxgbi_hbas_remove(struct cxgbi_device *cdev)
229 {
230 	int i;
231 	struct cxgbi_hba *chba;
232 
233 	log_debug(1 << CXGBI_DBG_DEV,
234 		"cdev 0x%p, p#%u.\n", cdev, cdev->nports);
235 
236 	for (i = 0; i < cdev->nports; i++) {
237 		chba = cdev->hbas[i];
238 		if (chba) {
239 			cdev->hbas[i] = NULL;
240 			iscsi_host_remove(chba->shost);
241 			pci_dev_put(cdev->pdev);
242 			iscsi_host_free(chba->shost);
243 		}
244 	}
245 }
246 EXPORT_SYMBOL_GPL(cxgbi_hbas_remove);
247 
248 int cxgbi_hbas_add(struct cxgbi_device *cdev, unsigned int max_lun,
249 		unsigned int max_id, struct scsi_host_template *sht,
250 		struct scsi_transport_template *stt)
251 {
252 	struct cxgbi_hba *chba;
253 	struct Scsi_Host *shost;
254 	int i, err;
255 
256 	log_debug(1 << CXGBI_DBG_DEV, "cdev 0x%p, p#%u.\n", cdev, cdev->nports);
257 
258 	for (i = 0; i < cdev->nports; i++) {
259 		shost = iscsi_host_alloc(sht, sizeof(*chba), 1);
260 		if (!shost) {
261 			pr_info("0x%p, p%d, %s, host alloc failed.\n",
262 				cdev, i, cdev->ports[i]->name);
263 			err = -ENOMEM;
264 			goto err_out;
265 		}
266 
267 		shost->transportt = stt;
268 		shost->max_lun = max_lun;
269 		shost->max_id = max_id;
270 		shost->max_channel = 0;
271 		shost->max_cmd_len = 16;
272 
273 		chba = iscsi_host_priv(shost);
274 		chba->cdev = cdev;
275 		chba->ndev = cdev->ports[i];
276 		chba->shost = shost;
277 
278 		log_debug(1 << CXGBI_DBG_DEV,
279 			"cdev 0x%p, p#%d %s: chba 0x%p.\n",
280 			cdev, i, cdev->ports[i]->name, chba);
281 
282 		pci_dev_get(cdev->pdev);
283 		err = iscsi_host_add(shost, &cdev->pdev->dev);
284 		if (err) {
285 			pr_info("cdev 0x%p, p#%d %s, host add failed.\n",
286 				cdev, i, cdev->ports[i]->name);
287 			pci_dev_put(cdev->pdev);
288 			scsi_host_put(shost);
289 			goto  err_out;
290 		}
291 
292 		cdev->hbas[i] = chba;
293 	}
294 
295 	return 0;
296 
297 err_out:
298 	cxgbi_hbas_remove(cdev);
299 	return err;
300 }
301 EXPORT_SYMBOL_GPL(cxgbi_hbas_add);
302 
303 /*
304  * iSCSI offload
305  *
306  * - source port management
307  *   To find a free source port in the port allocation map we use a very simple
308  *   rotor scheme to look for the next free port.
309  *
310  *   If a source port has been specified make sure that it doesn't collide with
311  *   our normal source port allocation map.  If it's outside the range of our
312  *   allocation/deallocation scheme just let them use it.
313  *
314  *   If the source port is outside our allocation range, the caller is
315  *   responsible for keeping track of their port usage.
316  */
317 static int sock_get_port(struct cxgbi_sock *csk)
318 {
319 	struct cxgbi_device *cdev = csk->cdev;
320 	struct cxgbi_ports_map *pmap = &cdev->pmap;
321 	unsigned int start;
322 	int idx;
323 
324 	if (!pmap->max_connect) {
325 		pr_err("cdev 0x%p, p#%u %s, NO port map.\n",
326 			   cdev, csk->port_id, cdev->ports[csk->port_id]->name);
327 		return -EADDRNOTAVAIL;
328 	}
329 
330 	if (csk->saddr.sin_port) {
331 		pr_err("source port NON-ZERO %u.\n",
332 			ntohs(csk->saddr.sin_port));
333 		return -EADDRINUSE;
334 	}
335 
336 	spin_lock_bh(&pmap->lock);
337 	if (pmap->used >= pmap->max_connect) {
338 		spin_unlock_bh(&pmap->lock);
339 		pr_info("cdev 0x%p, p#%u %s, ALL ports used.\n",
340 			cdev, csk->port_id, cdev->ports[csk->port_id]->name);
341 		return -EADDRNOTAVAIL;
342 	}
343 
344 	start = idx = pmap->next;
345 	do {
346 		if (++idx >= pmap->max_connect)
347 			idx = 0;
348 		if (!pmap->port_csk[idx]) {
349 			pmap->used++;
350 			csk->saddr.sin_port =
351 				htons(pmap->sport_base + idx);
352 			pmap->next = idx;
353 			pmap->port_csk[idx] = csk;
354 			spin_unlock_bh(&pmap->lock);
355 			cxgbi_sock_get(csk);
356 			log_debug(1 << CXGBI_DBG_SOCK,
357 				"cdev 0x%p, p#%u %s, p %u, %u.\n",
358 				cdev, csk->port_id,
359 				cdev->ports[csk->port_id]->name,
360 				pmap->sport_base + idx, pmap->next);
361 			return 0;
362 		}
363 	} while (idx != start);
364 	spin_unlock_bh(&pmap->lock);
365 
366 	/* should not happen */
367 	pr_warn("cdev 0x%p, p#%u %s, next %u?\n",
368 		cdev, csk->port_id, cdev->ports[csk->port_id]->name,
369 		pmap->next);
370 	return -EADDRNOTAVAIL;
371 }
372 
373 static void sock_put_port(struct cxgbi_sock *csk)
374 {
375 	struct cxgbi_device *cdev = csk->cdev;
376 	struct cxgbi_ports_map *pmap = &cdev->pmap;
377 
378 	if (csk->saddr.sin_port) {
379 		int idx = ntohs(csk->saddr.sin_port) - pmap->sport_base;
380 
381 		csk->saddr.sin_port = 0;
382 		if (idx < 0 || idx >= pmap->max_connect) {
383 			pr_err("cdev 0x%p, p#%u %s, port %u OOR.\n",
384 				cdev, csk->port_id,
385 				cdev->ports[csk->port_id]->name,
386 				ntohs(csk->saddr.sin_port));
387 			return;
388 		}
389 
390 		spin_lock_bh(&pmap->lock);
391 		pmap->port_csk[idx] = NULL;
392 		pmap->used--;
393 		spin_unlock_bh(&pmap->lock);
394 
395 		log_debug(1 << CXGBI_DBG_SOCK,
396 			"cdev 0x%p, p#%u %s, release %u.\n",
397 			cdev, csk->port_id, cdev->ports[csk->port_id]->name,
398 			pmap->sport_base + idx);
399 
400 		cxgbi_sock_put(csk);
401 	}
402 }
403 
404 /*
405  * iscsi tcp connection
406  */
407 void cxgbi_sock_free_cpl_skbs(struct cxgbi_sock *csk)
408 {
409 	if (csk->cpl_close) {
410 		kfree_skb(csk->cpl_close);
411 		csk->cpl_close = NULL;
412 	}
413 	if (csk->cpl_abort_req) {
414 		kfree_skb(csk->cpl_abort_req);
415 		csk->cpl_abort_req = NULL;
416 	}
417 	if (csk->cpl_abort_rpl) {
418 		kfree_skb(csk->cpl_abort_rpl);
419 		csk->cpl_abort_rpl = NULL;
420 	}
421 }
422 EXPORT_SYMBOL_GPL(cxgbi_sock_free_cpl_skbs);
423 
424 static struct cxgbi_sock *cxgbi_sock_create(struct cxgbi_device *cdev)
425 {
426 	struct cxgbi_sock *csk = kzalloc(sizeof(*csk), GFP_NOIO);
427 
428 	if (!csk) {
429 		pr_info("alloc csk %zu failed.\n", sizeof(*csk));
430 		return NULL;
431 	}
432 
433 	if (cdev->csk_alloc_cpls(csk) < 0) {
434 		pr_info("csk 0x%p, alloc cpls failed.\n", csk);
435 		kfree(csk);
436 		return NULL;
437 	}
438 
439 	spin_lock_init(&csk->lock);
440 	kref_init(&csk->refcnt);
441 	skb_queue_head_init(&csk->receive_queue);
442 	skb_queue_head_init(&csk->write_queue);
443 	setup_timer(&csk->retry_timer, NULL, (unsigned long)csk);
444 	rwlock_init(&csk->callback_lock);
445 	csk->cdev = cdev;
446 	csk->flags = 0;
447 	cxgbi_sock_set_state(csk, CTP_CLOSED);
448 
449 	log_debug(1 << CXGBI_DBG_SOCK, "cdev 0x%p, new csk 0x%p.\n", cdev, csk);
450 
451 	return csk;
452 }
453 
454 static struct rtable *find_route_ipv4(struct flowi4 *fl4,
455 				      __be32 saddr, __be32 daddr,
456 				      __be16 sport, __be16 dport, u8 tos)
457 {
458 	struct rtable *rt;
459 
460 	rt = ip_route_output_ports(&init_net, fl4, NULL, daddr, saddr,
461 				   dport, sport, IPPROTO_TCP, tos, 0);
462 	if (IS_ERR(rt))
463 		return NULL;
464 
465 	return rt;
466 }
467 
468 static struct cxgbi_sock *cxgbi_check_route(struct sockaddr *dst_addr)
469 {
470 	struct sockaddr_in *daddr = (struct sockaddr_in *)dst_addr;
471 	struct dst_entry *dst;
472 	struct net_device *ndev;
473 	struct cxgbi_device *cdev;
474 	struct rtable *rt = NULL;
475 	struct neighbour *n;
476 	struct flowi4 fl4;
477 	struct cxgbi_sock *csk = NULL;
478 	unsigned int mtu = 0;
479 	int port = 0xFFFF;
480 	int err = 0;
481 
482 	if (daddr->sin_family != AF_INET) {
483 		pr_info("address family 0x%x NOT supported.\n",
484 			daddr->sin_family);
485 		err = -EAFNOSUPPORT;
486 		goto err_out;
487 	}
488 
489 	rt = find_route_ipv4(&fl4, 0, daddr->sin_addr.s_addr, 0, daddr->sin_port, 0);
490 	if (!rt) {
491 		pr_info("no route to ipv4 0x%x, port %u.\n",
492 			daddr->sin_addr.s_addr, daddr->sin_port);
493 		err = -ENETUNREACH;
494 		goto err_out;
495 	}
496 	dst = &rt->dst;
497 	n = dst_get_neighbour_noref(dst);
498 	if (!n) {
499 		err = -ENODEV;
500 		goto rel_rt;
501 	}
502 	ndev = n->dev;
503 
504 	if (rt->rt_flags & (RTCF_MULTICAST | RTCF_BROADCAST)) {
505 		pr_info("multi-cast route %pI4, port %u, dev %s.\n",
506 			&daddr->sin_addr.s_addr, ntohs(daddr->sin_port),
507 			ndev->name);
508 		err = -ENETUNREACH;
509 		goto rel_rt;
510 	}
511 
512 	if (ndev->flags & IFF_LOOPBACK) {
513 		ndev = ip_dev_find(&init_net, daddr->sin_addr.s_addr);
514 		mtu = ndev->mtu;
515 		pr_info("rt dev %s, loopback -> %s, mtu %u.\n",
516 			n->dev->name, ndev->name, mtu);
517 	}
518 
519 	cdev = cxgbi_device_find_by_netdev(ndev, &port);
520 	if (!cdev) {
521 		pr_info("dst %pI4, %s, NOT cxgbi device.\n",
522 			&daddr->sin_addr.s_addr, ndev->name);
523 		err = -ENETUNREACH;
524 		goto rel_rt;
525 	}
526 	log_debug(1 << CXGBI_DBG_SOCK,
527 		"route to %pI4 :%u, ndev p#%d,%s, cdev 0x%p.\n",
528 		&daddr->sin_addr.s_addr, ntohs(daddr->sin_port),
529 			   port, ndev->name, cdev);
530 
531 	csk = cxgbi_sock_create(cdev);
532 	if (!csk) {
533 		err = -ENOMEM;
534 		goto rel_rt;
535 	}
536 	csk->cdev = cdev;
537 	csk->port_id = port;
538 	csk->mtu = mtu;
539 	csk->dst = dst;
540 	csk->daddr.sin_addr.s_addr = daddr->sin_addr.s_addr;
541 	csk->daddr.sin_port = daddr->sin_port;
542 	csk->daddr.sin_family = daddr->sin_family;
543 	csk->saddr.sin_addr.s_addr = fl4.saddr;
544 
545 	return csk;
546 
547 rel_rt:
548 	ip_rt_put(rt);
549 	if (csk)
550 		cxgbi_sock_closed(csk);
551 err_out:
552 	return ERR_PTR(err);
553 }
554 
555 void cxgbi_sock_established(struct cxgbi_sock *csk, unsigned int snd_isn,
556 			unsigned int opt)
557 {
558 	csk->write_seq = csk->snd_nxt = csk->snd_una = snd_isn;
559 	dst_confirm(csk->dst);
560 	smp_mb();
561 	cxgbi_sock_set_state(csk, CTP_ESTABLISHED);
562 }
563 EXPORT_SYMBOL_GPL(cxgbi_sock_established);
564 
565 static void cxgbi_inform_iscsi_conn_closing(struct cxgbi_sock *csk)
566 {
567 	log_debug(1 << CXGBI_DBG_SOCK,
568 		"csk 0x%p, state %u, flags 0x%lx, conn 0x%p.\n",
569 		csk, csk->state, csk->flags, csk->user_data);
570 
571 	if (csk->state != CTP_ESTABLISHED) {
572 		read_lock_bh(&csk->callback_lock);
573 		if (csk->user_data)
574 			iscsi_conn_failure(csk->user_data,
575 					ISCSI_ERR_CONN_FAILED);
576 		read_unlock_bh(&csk->callback_lock);
577 	}
578 }
579 
580 void cxgbi_sock_closed(struct cxgbi_sock *csk)
581 {
582 	log_debug(1 << CXGBI_DBG_SOCK, "csk 0x%p,%u,0x%lx,%u.\n",
583 		csk, (csk)->state, (csk)->flags, (csk)->tid);
584 	cxgbi_sock_set_flag(csk, CTPF_ACTIVE_CLOSE_NEEDED);
585 	if (csk->state == CTP_ACTIVE_OPEN || csk->state == CTP_CLOSED)
586 		return;
587 	if (csk->saddr.sin_port)
588 		sock_put_port(csk);
589 	if (csk->dst)
590 		dst_release(csk->dst);
591 	csk->cdev->csk_release_offload_resources(csk);
592 	cxgbi_sock_set_state(csk, CTP_CLOSED);
593 	cxgbi_inform_iscsi_conn_closing(csk);
594 	cxgbi_sock_put(csk);
595 }
596 EXPORT_SYMBOL_GPL(cxgbi_sock_closed);
597 
598 static void need_active_close(struct cxgbi_sock *csk)
599 {
600 	int data_lost;
601 	int close_req = 0;
602 
603 	log_debug(1 << CXGBI_DBG_SOCK, "csk 0x%p,%u,0x%lx,%u.\n",
604 		csk, (csk)->state, (csk)->flags, (csk)->tid);
605 	spin_lock_bh(&csk->lock);
606 	dst_confirm(csk->dst);
607 	data_lost = skb_queue_len(&csk->receive_queue);
608 	__skb_queue_purge(&csk->receive_queue);
609 
610 	if (csk->state == CTP_ACTIVE_OPEN)
611 		cxgbi_sock_set_flag(csk, CTPF_ACTIVE_CLOSE_NEEDED);
612 	else if (csk->state == CTP_ESTABLISHED) {
613 		close_req = 1;
614 		cxgbi_sock_set_state(csk, CTP_ACTIVE_CLOSE);
615 	} else if (csk->state == CTP_PASSIVE_CLOSE) {
616 		close_req = 1;
617 		cxgbi_sock_set_state(csk, CTP_CLOSE_WAIT_2);
618 	}
619 
620 	if (close_req) {
621 		if (data_lost)
622 			csk->cdev->csk_send_abort_req(csk);
623 		else
624 			csk->cdev->csk_send_close_req(csk);
625 	}
626 
627 	spin_unlock_bh(&csk->lock);
628 }
629 
630 void cxgbi_sock_fail_act_open(struct cxgbi_sock *csk, int errno)
631 {
632 	pr_info("csk 0x%p,%u,%lx, %pI4:%u-%pI4:%u, err %d.\n",
633 			csk, csk->state, csk->flags,
634 			&csk->saddr.sin_addr.s_addr, csk->saddr.sin_port,
635 			&csk->daddr.sin_addr.s_addr, csk->daddr.sin_port,
636 			errno);
637 
638 	cxgbi_sock_set_state(csk, CTP_CONNECTING);
639 	csk->err = errno;
640 	cxgbi_sock_closed(csk);
641 }
642 EXPORT_SYMBOL_GPL(cxgbi_sock_fail_act_open);
643 
644 void cxgbi_sock_act_open_req_arp_failure(void *handle, struct sk_buff *skb)
645 {
646 	struct cxgbi_sock *csk = (struct cxgbi_sock *)skb->sk;
647 
648 	log_debug(1 << CXGBI_DBG_SOCK, "csk 0x%p,%u,0x%lx,%u.\n",
649 		csk, (csk)->state, (csk)->flags, (csk)->tid);
650 	cxgbi_sock_get(csk);
651 	spin_lock_bh(&csk->lock);
652 	if (csk->state == CTP_ACTIVE_OPEN)
653 		cxgbi_sock_fail_act_open(csk, -EHOSTUNREACH);
654 	spin_unlock_bh(&csk->lock);
655 	cxgbi_sock_put(csk);
656 	__kfree_skb(skb);
657 }
658 EXPORT_SYMBOL_GPL(cxgbi_sock_act_open_req_arp_failure);
659 
660 void cxgbi_sock_rcv_abort_rpl(struct cxgbi_sock *csk)
661 {
662 	cxgbi_sock_get(csk);
663 	spin_lock_bh(&csk->lock);
664 	if (cxgbi_sock_flag(csk, CTPF_ABORT_RPL_PENDING)) {
665 		if (!cxgbi_sock_flag(csk, CTPF_ABORT_RPL_RCVD))
666 			cxgbi_sock_set_flag(csk, CTPF_ABORT_RPL_RCVD);
667 		else {
668 			cxgbi_sock_clear_flag(csk, CTPF_ABORT_RPL_RCVD);
669 			cxgbi_sock_clear_flag(csk, CTPF_ABORT_RPL_PENDING);
670 			if (cxgbi_sock_flag(csk, CTPF_ABORT_REQ_RCVD))
671 				pr_err("csk 0x%p,%u,0x%lx,%u,ABT_RPL_RSS.\n",
672 					csk, csk->state, csk->flags, csk->tid);
673 			cxgbi_sock_closed(csk);
674 		}
675 	}
676 	spin_unlock_bh(&csk->lock);
677 	cxgbi_sock_put(csk);
678 }
679 EXPORT_SYMBOL_GPL(cxgbi_sock_rcv_abort_rpl);
680 
681 void cxgbi_sock_rcv_peer_close(struct cxgbi_sock *csk)
682 {
683 	log_debug(1 << CXGBI_DBG_SOCK, "csk 0x%p,%u,0x%lx,%u.\n",
684 		csk, (csk)->state, (csk)->flags, (csk)->tid);
685 	cxgbi_sock_get(csk);
686 	spin_lock_bh(&csk->lock);
687 
688 	if (cxgbi_sock_flag(csk, CTPF_ABORT_RPL_PENDING))
689 		goto done;
690 
691 	switch (csk->state) {
692 	case CTP_ESTABLISHED:
693 		cxgbi_sock_set_state(csk, CTP_PASSIVE_CLOSE);
694 		break;
695 	case CTP_ACTIVE_CLOSE:
696 		cxgbi_sock_set_state(csk, CTP_CLOSE_WAIT_2);
697 		break;
698 	case CTP_CLOSE_WAIT_1:
699 		cxgbi_sock_closed(csk);
700 		break;
701 	case CTP_ABORTING:
702 		break;
703 	default:
704 		pr_err("csk 0x%p,%u,0x%lx,%u, bad state.\n",
705 			csk, csk->state, csk->flags, csk->tid);
706 	}
707 	cxgbi_inform_iscsi_conn_closing(csk);
708 done:
709 	spin_unlock_bh(&csk->lock);
710 	cxgbi_sock_put(csk);
711 }
712 EXPORT_SYMBOL_GPL(cxgbi_sock_rcv_peer_close);
713 
714 void cxgbi_sock_rcv_close_conn_rpl(struct cxgbi_sock *csk, u32 snd_nxt)
715 {
716 	log_debug(1 << CXGBI_DBG_SOCK, "csk 0x%p,%u,0x%lx,%u.\n",
717 		csk, (csk)->state, (csk)->flags, (csk)->tid);
718 	cxgbi_sock_get(csk);
719 	spin_lock_bh(&csk->lock);
720 
721 	csk->snd_una = snd_nxt - 1;
722 	if (cxgbi_sock_flag(csk, CTPF_ABORT_RPL_PENDING))
723 		goto done;
724 
725 	switch (csk->state) {
726 	case CTP_ACTIVE_CLOSE:
727 		cxgbi_sock_set_state(csk, CTP_CLOSE_WAIT_1);
728 		break;
729 	case CTP_CLOSE_WAIT_1:
730 	case CTP_CLOSE_WAIT_2:
731 		cxgbi_sock_closed(csk);
732 		break;
733 	case CTP_ABORTING:
734 		break;
735 	default:
736 		pr_err("csk 0x%p,%u,0x%lx,%u, bad state.\n",
737 			csk, csk->state, csk->flags, csk->tid);
738 	}
739 done:
740 	spin_unlock_bh(&csk->lock);
741 	cxgbi_sock_put(csk);
742 }
743 EXPORT_SYMBOL_GPL(cxgbi_sock_rcv_close_conn_rpl);
744 
745 void cxgbi_sock_rcv_wr_ack(struct cxgbi_sock *csk, unsigned int credits,
746 			   unsigned int snd_una, int seq_chk)
747 {
748 	log_debug(1 << CXGBI_DBG_TOE | 1 << CXGBI_DBG_SOCK,
749 			"csk 0x%p,%u,0x%lx,%u, cr %u,%u+%u, snd_una %u,%d.\n",
750 			csk, csk->state, csk->flags, csk->tid, credits,
751 			csk->wr_cred, csk->wr_una_cred, snd_una, seq_chk);
752 
753 	spin_lock_bh(&csk->lock);
754 
755 	csk->wr_cred += credits;
756 	if (csk->wr_una_cred > csk->wr_max_cred - csk->wr_cred)
757 		csk->wr_una_cred = csk->wr_max_cred - csk->wr_cred;
758 
759 	while (credits) {
760 		struct sk_buff *p = cxgbi_sock_peek_wr(csk);
761 
762 		if (unlikely(!p)) {
763 			pr_err("csk 0x%p,%u,0x%lx,%u, cr %u,%u+%u, empty.\n",
764 				csk, csk->state, csk->flags, csk->tid, credits,
765 				csk->wr_cred, csk->wr_una_cred);
766 			break;
767 		}
768 
769 		if (unlikely(credits < p->csum)) {
770 			pr_warn("csk 0x%p,%u,0x%lx,%u, cr %u,%u+%u, < %u.\n",
771 				csk, csk->state, csk->flags, csk->tid,
772 				credits, csk->wr_cred, csk->wr_una_cred,
773 				p->csum);
774 			p->csum -= credits;
775 			break;
776 		} else {
777 			cxgbi_sock_dequeue_wr(csk);
778 			credits -= p->csum;
779 			kfree_skb(p);
780 		}
781 	}
782 
783 	cxgbi_sock_check_wr_invariants(csk);
784 
785 	if (seq_chk) {
786 		if (unlikely(before(snd_una, csk->snd_una))) {
787 			pr_warn("csk 0x%p,%u,0x%lx,%u, snd_una %u/%u.",
788 				csk, csk->state, csk->flags, csk->tid, snd_una,
789 				csk->snd_una);
790 			goto done;
791 		}
792 
793 		if (csk->snd_una != snd_una) {
794 			csk->snd_una = snd_una;
795 			dst_confirm(csk->dst);
796 		}
797 	}
798 
799 	if (skb_queue_len(&csk->write_queue)) {
800 		if (csk->cdev->csk_push_tx_frames(csk, 0))
801 			cxgbi_conn_tx_open(csk);
802 	} else
803 		cxgbi_conn_tx_open(csk);
804 done:
805 	spin_unlock_bh(&csk->lock);
806 }
807 EXPORT_SYMBOL_GPL(cxgbi_sock_rcv_wr_ack);
808 
809 static unsigned int cxgbi_sock_find_best_mtu(struct cxgbi_sock *csk,
810 					     unsigned short mtu)
811 {
812 	int i = 0;
813 
814 	while (i < csk->cdev->nmtus - 1 && csk->cdev->mtus[i + 1] <= mtu)
815 		++i;
816 
817 	return i;
818 }
819 
820 unsigned int cxgbi_sock_select_mss(struct cxgbi_sock *csk, unsigned int pmtu)
821 {
822 	unsigned int idx;
823 	struct dst_entry *dst = csk->dst;
824 
825 	csk->advmss = dst_metric_advmss(dst);
826 
827 	if (csk->advmss > pmtu - 40)
828 		csk->advmss = pmtu - 40;
829 	if (csk->advmss < csk->cdev->mtus[0] - 40)
830 		csk->advmss = csk->cdev->mtus[0] - 40;
831 	idx = cxgbi_sock_find_best_mtu(csk, csk->advmss + 40);
832 
833 	return idx;
834 }
835 EXPORT_SYMBOL_GPL(cxgbi_sock_select_mss);
836 
837 void cxgbi_sock_skb_entail(struct cxgbi_sock *csk, struct sk_buff *skb)
838 {
839 	cxgbi_skcb_tcp_seq(skb) = csk->write_seq;
840 	__skb_queue_tail(&csk->write_queue, skb);
841 }
842 EXPORT_SYMBOL_GPL(cxgbi_sock_skb_entail);
843 
844 void cxgbi_sock_purge_wr_queue(struct cxgbi_sock *csk)
845 {
846 	struct sk_buff *skb;
847 
848 	while ((skb = cxgbi_sock_dequeue_wr(csk)) != NULL)
849 		kfree_skb(skb);
850 }
851 EXPORT_SYMBOL_GPL(cxgbi_sock_purge_wr_queue);
852 
853 void cxgbi_sock_check_wr_invariants(const struct cxgbi_sock *csk)
854 {
855 	int pending = cxgbi_sock_count_pending_wrs(csk);
856 
857 	if (unlikely(csk->wr_cred + pending != csk->wr_max_cred))
858 		pr_err("csk 0x%p, tid %u, credit %u + %u != %u.\n",
859 			csk, csk->tid, csk->wr_cred, pending, csk->wr_max_cred);
860 }
861 EXPORT_SYMBOL_GPL(cxgbi_sock_check_wr_invariants);
862 
863 static int cxgbi_sock_send_pdus(struct cxgbi_sock *csk, struct sk_buff *skb)
864 {
865 	struct cxgbi_device *cdev = csk->cdev;
866 	struct sk_buff *next;
867 	int err, copied = 0;
868 
869 	spin_lock_bh(&csk->lock);
870 
871 	if (csk->state != CTP_ESTABLISHED) {
872 		log_debug(1 << CXGBI_DBG_PDU_TX,
873 			"csk 0x%p,%u,0x%lx,%u, EAGAIN.\n",
874 			csk, csk->state, csk->flags, csk->tid);
875 		err = -EAGAIN;
876 		goto out_err;
877 	}
878 
879 	if (csk->err) {
880 		log_debug(1 << CXGBI_DBG_PDU_TX,
881 			"csk 0x%p,%u,0x%lx,%u, EPIPE %d.\n",
882 			csk, csk->state, csk->flags, csk->tid, csk->err);
883 		err = -EPIPE;
884 		goto out_err;
885 	}
886 
887 	if (csk->write_seq - csk->snd_una >= cdev->snd_win) {
888 		log_debug(1 << CXGBI_DBG_PDU_TX,
889 			"csk 0x%p,%u,0x%lx,%u, FULL %u-%u >= %u.\n",
890 			csk, csk->state, csk->flags, csk->tid, csk->write_seq,
891 			csk->snd_una, cdev->snd_win);
892 		err = -ENOBUFS;
893 		goto out_err;
894 	}
895 
896 	while (skb) {
897 		int frags = skb_shinfo(skb)->nr_frags +
898 				(skb->len != skb->data_len);
899 
900 		if (unlikely(skb_headroom(skb) < cdev->skb_tx_rsvd)) {
901 			pr_err("csk 0x%p, skb head %u < %u.\n",
902 				csk, skb_headroom(skb), cdev->skb_tx_rsvd);
903 			err = -EINVAL;
904 			goto out_err;
905 		}
906 
907 		if (frags >= SKB_WR_LIST_SIZE) {
908 			pr_err("csk 0x%p, frags %d, %u,%u >%u.\n",
909 				csk, skb_shinfo(skb)->nr_frags, skb->len,
910 				skb->data_len, (uint)(SKB_WR_LIST_SIZE));
911 			err = -EINVAL;
912 			goto out_err;
913 		}
914 
915 		next = skb->next;
916 		skb->next = NULL;
917 		cxgbi_skcb_set_flag(skb, SKCBF_TX_NEED_HDR);
918 		cxgbi_sock_skb_entail(csk, skb);
919 		copied += skb->len;
920 		csk->write_seq += skb->len +
921 				cxgbi_ulp_extra_len(cxgbi_skcb_ulp_mode(skb));
922 		skb = next;
923 	}
924 done:
925 	if (likely(skb_queue_len(&csk->write_queue)))
926 		cdev->csk_push_tx_frames(csk, 1);
927 	spin_unlock_bh(&csk->lock);
928 	return copied;
929 
930 out_err:
931 	if (copied == 0 && err == -EPIPE)
932 		copied = csk->err ? csk->err : -EPIPE;
933 	else
934 		copied = err;
935 	goto done;
936 }
937 
938 /*
939  * Direct Data Placement -
940  * Directly place the iSCSI Data-In or Data-Out PDU's payload into pre-posted
941  * final destination host-memory buffers based on the Initiator Task Tag (ITT)
942  * in Data-In or Target Task Tag (TTT) in Data-Out PDUs.
943  * The host memory address is programmed into h/w in the format of pagepod
944  * entries.
945  * The location of the pagepod entry is encoded into ddp tag which is used as
946  * the base for ITT/TTT.
947  */
948 
949 static unsigned char ddp_page_order[DDP_PGIDX_MAX] = {0, 1, 2, 4};
950 static unsigned char ddp_page_shift[DDP_PGIDX_MAX] = {12, 13, 14, 16};
951 static unsigned char page_idx = DDP_PGIDX_MAX;
952 
953 static unsigned char sw_tag_idx_bits;
954 static unsigned char sw_tag_age_bits;
955 
956 /*
957  * Direct-Data Placement page size adjustment
958  */
959 static int ddp_adjust_page_table(void)
960 {
961 	int i;
962 	unsigned int base_order, order;
963 
964 	if (PAGE_SIZE < (1UL << ddp_page_shift[0])) {
965 		pr_info("PAGE_SIZE 0x%lx too small, min 0x%lx\n",
966 			PAGE_SIZE, 1UL << ddp_page_shift[0]);
967 		return -EINVAL;
968 	}
969 
970 	base_order = get_order(1UL << ddp_page_shift[0]);
971 	order = get_order(1UL << PAGE_SHIFT);
972 
973 	for (i = 0; i < DDP_PGIDX_MAX; i++) {
974 		/* first is the kernel page size, then just doubling */
975 		ddp_page_order[i] = order - base_order + i;
976 		ddp_page_shift[i] = PAGE_SHIFT + i;
977 	}
978 	return 0;
979 }
980 
981 static int ddp_find_page_index(unsigned long pgsz)
982 {
983 	int i;
984 
985 	for (i = 0; i < DDP_PGIDX_MAX; i++) {
986 		if (pgsz == (1UL << ddp_page_shift[i]))
987 			return i;
988 	}
989 	pr_info("ddp page size %lu not supported.\n", pgsz);
990 	return DDP_PGIDX_MAX;
991 }
992 
993 static void ddp_setup_host_page_size(void)
994 {
995 	if (page_idx == DDP_PGIDX_MAX) {
996 		page_idx = ddp_find_page_index(PAGE_SIZE);
997 
998 		if (page_idx == DDP_PGIDX_MAX) {
999 			pr_info("system PAGE %lu, update hw.\n", PAGE_SIZE);
1000 			if (ddp_adjust_page_table() < 0) {
1001 				pr_info("PAGE %lu, disable ddp.\n", PAGE_SIZE);
1002 				return;
1003 			}
1004 			page_idx = ddp_find_page_index(PAGE_SIZE);
1005 		}
1006 		pr_info("system PAGE %lu, ddp idx %u.\n", PAGE_SIZE, page_idx);
1007 	}
1008 }
1009 
1010 void cxgbi_ddp_page_size_factor(int *pgsz_factor)
1011 {
1012 	int i;
1013 
1014 	for (i = 0; i < DDP_PGIDX_MAX; i++)
1015 		pgsz_factor[i] = ddp_page_order[i];
1016 }
1017 EXPORT_SYMBOL_GPL(cxgbi_ddp_page_size_factor);
1018 
1019 /*
1020  * DDP setup & teardown
1021  */
1022 
1023 void cxgbi_ddp_ppod_set(struct cxgbi_pagepod *ppod,
1024 			struct cxgbi_pagepod_hdr *hdr,
1025 			struct cxgbi_gather_list *gl, unsigned int gidx)
1026 {
1027 	int i;
1028 
1029 	memcpy(ppod, hdr, sizeof(*hdr));
1030 	for (i = 0; i < (PPOD_PAGES_MAX + 1); i++, gidx++) {
1031 		ppod->addr[i] = gidx < gl->nelem ?
1032 				cpu_to_be64(gl->phys_addr[gidx]) : 0ULL;
1033 	}
1034 }
1035 EXPORT_SYMBOL_GPL(cxgbi_ddp_ppod_set);
1036 
1037 void cxgbi_ddp_ppod_clear(struct cxgbi_pagepod *ppod)
1038 {
1039 	memset(ppod, 0, sizeof(*ppod));
1040 }
1041 EXPORT_SYMBOL_GPL(cxgbi_ddp_ppod_clear);
1042 
1043 static inline int ddp_find_unused_entries(struct cxgbi_ddp_info *ddp,
1044 					unsigned int start, unsigned int max,
1045 					unsigned int count,
1046 					struct cxgbi_gather_list *gl)
1047 {
1048 	unsigned int i, j, k;
1049 
1050 	/*  not enough entries */
1051 	if ((max - start) < count) {
1052 		log_debug(1 << CXGBI_DBG_DDP,
1053 			"NOT enough entries %u+%u < %u.\n", start, count, max);
1054 		return -EBUSY;
1055 	}
1056 
1057 	max -= count;
1058 	spin_lock(&ddp->map_lock);
1059 	for (i = start; i < max;) {
1060 		for (j = 0, k = i; j < count; j++, k++) {
1061 			if (ddp->gl_map[k])
1062 				break;
1063 		}
1064 		if (j == count) {
1065 			for (j = 0, k = i; j < count; j++, k++)
1066 				ddp->gl_map[k] = gl;
1067 			spin_unlock(&ddp->map_lock);
1068 			return i;
1069 		}
1070 		i += j + 1;
1071 	}
1072 	spin_unlock(&ddp->map_lock);
1073 	log_debug(1 << CXGBI_DBG_DDP,
1074 		"NO suitable entries %u available.\n", count);
1075 	return -EBUSY;
1076 }
1077 
1078 static inline void ddp_unmark_entries(struct cxgbi_ddp_info *ddp,
1079 						int start, int count)
1080 {
1081 	spin_lock(&ddp->map_lock);
1082 	memset(&ddp->gl_map[start], 0,
1083 		count * sizeof(struct cxgbi_gather_list *));
1084 	spin_unlock(&ddp->map_lock);
1085 }
1086 
1087 static inline void ddp_gl_unmap(struct pci_dev *pdev,
1088 					struct cxgbi_gather_list *gl)
1089 {
1090 	int i;
1091 
1092 	for (i = 0; i < gl->nelem; i++)
1093 		dma_unmap_page(&pdev->dev, gl->phys_addr[i], PAGE_SIZE,
1094 				PCI_DMA_FROMDEVICE);
1095 }
1096 
1097 static inline int ddp_gl_map(struct pci_dev *pdev,
1098 				    struct cxgbi_gather_list *gl)
1099 {
1100 	int i;
1101 
1102 	for (i = 0; i < gl->nelem; i++) {
1103 		gl->phys_addr[i] = dma_map_page(&pdev->dev, gl->pages[i], 0,
1104 						PAGE_SIZE,
1105 						PCI_DMA_FROMDEVICE);
1106 		if (unlikely(dma_mapping_error(&pdev->dev, gl->phys_addr[i]))) {
1107 			log_debug(1 << CXGBI_DBG_DDP,
1108 				"page %d 0x%p, 0x%p dma mapping err.\n",
1109 				i, gl->pages[i], pdev);
1110 			goto unmap;
1111 		}
1112 	}
1113 	return i;
1114 unmap:
1115 	if (i) {
1116 		unsigned int nelem = gl->nelem;
1117 
1118 		gl->nelem = i;
1119 		ddp_gl_unmap(pdev, gl);
1120 		gl->nelem = nelem;
1121 	}
1122 	return -EINVAL;
1123 }
1124 
1125 static void ddp_release_gl(struct cxgbi_gather_list *gl,
1126 				  struct pci_dev *pdev)
1127 {
1128 	ddp_gl_unmap(pdev, gl);
1129 	kfree(gl);
1130 }
1131 
1132 static struct cxgbi_gather_list *ddp_make_gl(unsigned int xferlen,
1133 						    struct scatterlist *sgl,
1134 						    unsigned int sgcnt,
1135 						    struct pci_dev *pdev,
1136 						    gfp_t gfp)
1137 {
1138 	struct cxgbi_gather_list *gl;
1139 	struct scatterlist *sg = sgl;
1140 	struct page *sgpage = sg_page(sg);
1141 	unsigned int sglen = sg->length;
1142 	unsigned int sgoffset = sg->offset;
1143 	unsigned int npages = (xferlen + sgoffset + PAGE_SIZE - 1) >>
1144 				PAGE_SHIFT;
1145 	int i = 1, j = 0;
1146 
1147 	if (xferlen < DDP_THRESHOLD) {
1148 		log_debug(1 << CXGBI_DBG_DDP,
1149 			"xfer %u < threshold %u, no ddp.\n",
1150 			xferlen, DDP_THRESHOLD);
1151 		return NULL;
1152 	}
1153 
1154 	gl = kzalloc(sizeof(struct cxgbi_gather_list) +
1155 		     npages * (sizeof(dma_addr_t) +
1156 		     sizeof(struct page *)), gfp);
1157 	if (!gl) {
1158 		log_debug(1 << CXGBI_DBG_DDP,
1159 			"xfer %u, %u pages, OOM.\n", xferlen, npages);
1160 		return NULL;
1161 	}
1162 
1163 	 log_debug(1 << CXGBI_DBG_DDP,
1164 		"xfer %u, sgl %u, gl max %u.\n", xferlen, sgcnt, npages);
1165 
1166 	gl->pages = (struct page **)&gl->phys_addr[npages];
1167 	gl->nelem = npages;
1168 	gl->length = xferlen;
1169 	gl->offset = sgoffset;
1170 	gl->pages[0] = sgpage;
1171 
1172 	for (i = 1, sg = sg_next(sgl), j = 0; i < sgcnt;
1173 		i++, sg = sg_next(sg)) {
1174 		struct page *page = sg_page(sg);
1175 
1176 		if (sgpage == page && sg->offset == sgoffset + sglen)
1177 			sglen += sg->length;
1178 		else {
1179 			/*  make sure the sgl is fit for ddp:
1180 			 *  each has the same page size, and
1181 			 *  all of the middle pages are used completely
1182 			 */
1183 			if ((j && sgoffset) || ((i != sgcnt - 1) &&
1184 			    ((sglen + sgoffset) & ~PAGE_MASK))) {
1185 				log_debug(1 << CXGBI_DBG_DDP,
1186 					"page %d/%u, %u + %u.\n",
1187 					i, sgcnt, sgoffset, sglen);
1188 				goto error_out;
1189 			}
1190 
1191 			j++;
1192 			if (j == gl->nelem || sg->offset) {
1193 				log_debug(1 << CXGBI_DBG_DDP,
1194 					"page %d/%u, offset %u.\n",
1195 					j, gl->nelem, sg->offset);
1196 				goto error_out;
1197 			}
1198 			gl->pages[j] = page;
1199 			sglen = sg->length;
1200 			sgoffset = sg->offset;
1201 			sgpage = page;
1202 		}
1203 	}
1204 	gl->nelem = ++j;
1205 
1206 	if (ddp_gl_map(pdev, gl) < 0)
1207 		goto error_out;
1208 
1209 	return gl;
1210 
1211 error_out:
1212 	kfree(gl);
1213 	return NULL;
1214 }
1215 
1216 static void ddp_tag_release(struct cxgbi_hba *chba, u32 tag)
1217 {
1218 	struct cxgbi_device *cdev = chba->cdev;
1219 	struct cxgbi_ddp_info *ddp = cdev->ddp;
1220 	u32 idx;
1221 
1222 	idx = (tag >> PPOD_IDX_SHIFT) & ddp->idx_mask;
1223 	if (idx < ddp->nppods) {
1224 		struct cxgbi_gather_list *gl = ddp->gl_map[idx];
1225 		unsigned int npods;
1226 
1227 		if (!gl || !gl->nelem) {
1228 			pr_warn("tag 0x%x, idx %u, gl 0x%p, %u.\n",
1229 				tag, idx, gl, gl ? gl->nelem : 0);
1230 			return;
1231 		}
1232 		npods = (gl->nelem + PPOD_PAGES_MAX - 1) >> PPOD_PAGES_SHIFT;
1233 		log_debug(1 << CXGBI_DBG_DDP,
1234 			"tag 0x%x, release idx %u, npods %u.\n",
1235 			tag, idx, npods);
1236 		cdev->csk_ddp_clear(chba, tag, idx, npods);
1237 		ddp_unmark_entries(ddp, idx, npods);
1238 		ddp_release_gl(gl, ddp->pdev);
1239 	} else
1240 		pr_warn("tag 0x%x, idx %u > max %u.\n", tag, idx, ddp->nppods);
1241 }
1242 
1243 static int ddp_tag_reserve(struct cxgbi_sock *csk, unsigned int tid,
1244 			   u32 sw_tag, u32 *tagp, struct cxgbi_gather_list *gl,
1245 			   gfp_t gfp)
1246 {
1247 	struct cxgbi_device *cdev = csk->cdev;
1248 	struct cxgbi_ddp_info *ddp = cdev->ddp;
1249 	struct cxgbi_tag_format *tformat = &cdev->tag_format;
1250 	struct cxgbi_pagepod_hdr hdr;
1251 	unsigned int npods;
1252 	int idx = -1;
1253 	int err = -ENOMEM;
1254 	u32 tag;
1255 
1256 	npods = (gl->nelem + PPOD_PAGES_MAX - 1) >> PPOD_PAGES_SHIFT;
1257 	if (ddp->idx_last == ddp->nppods)
1258 		idx = ddp_find_unused_entries(ddp, 0, ddp->nppods,
1259 							npods, gl);
1260 	else {
1261 		idx = ddp_find_unused_entries(ddp, ddp->idx_last + 1,
1262 							ddp->nppods, npods,
1263 							gl);
1264 		if (idx < 0 && ddp->idx_last >= npods) {
1265 			idx = ddp_find_unused_entries(ddp, 0,
1266 				min(ddp->idx_last + npods, ddp->nppods),
1267 							npods, gl);
1268 		}
1269 	}
1270 	if (idx < 0) {
1271 		log_debug(1 << CXGBI_DBG_DDP,
1272 			"xferlen %u, gl %u, npods %u NO DDP.\n",
1273 			gl->length, gl->nelem, npods);
1274 		return idx;
1275 	}
1276 
1277 	tag = cxgbi_ddp_tag_base(tformat, sw_tag);
1278 	tag |= idx << PPOD_IDX_SHIFT;
1279 
1280 	hdr.rsvd = 0;
1281 	hdr.vld_tid = htonl(PPOD_VALID_FLAG | PPOD_TID(tid));
1282 	hdr.pgsz_tag_clr = htonl(tag & ddp->rsvd_tag_mask);
1283 	hdr.max_offset = htonl(gl->length);
1284 	hdr.page_offset = htonl(gl->offset);
1285 
1286 	err = cdev->csk_ddp_set(csk, &hdr, idx, npods, gl);
1287 	if (err < 0)
1288 		goto unmark_entries;
1289 
1290 	ddp->idx_last = idx;
1291 	log_debug(1 << CXGBI_DBG_DDP,
1292 		"xfer %u, gl %u,%u, tid 0x%x, tag 0x%x->0x%x(%u,%u).\n",
1293 		gl->length, gl->nelem, gl->offset, tid, sw_tag, tag, idx,
1294 		npods);
1295 	*tagp = tag;
1296 	return 0;
1297 
1298 unmark_entries:
1299 	ddp_unmark_entries(ddp, idx, npods);
1300 	return err;
1301 }
1302 
1303 int cxgbi_ddp_reserve(struct cxgbi_sock *csk, unsigned int *tagp,
1304 			unsigned int sw_tag, unsigned int xferlen,
1305 			struct scatterlist *sgl, unsigned int sgcnt, gfp_t gfp)
1306 {
1307 	struct cxgbi_device *cdev = csk->cdev;
1308 	struct cxgbi_tag_format *tformat = &cdev->tag_format;
1309 	struct cxgbi_gather_list *gl;
1310 	int err;
1311 
1312 	if (page_idx >= DDP_PGIDX_MAX || !cdev->ddp ||
1313 	    xferlen < DDP_THRESHOLD) {
1314 		log_debug(1 << CXGBI_DBG_DDP,
1315 			"pgidx %u, xfer %u, NO ddp.\n", page_idx, xferlen);
1316 		return -EINVAL;
1317 	}
1318 
1319 	if (!cxgbi_sw_tag_usable(tformat, sw_tag)) {
1320 		log_debug(1 << CXGBI_DBG_DDP,
1321 			"sw_tag 0x%x NOT usable.\n", sw_tag);
1322 		return -EINVAL;
1323 	}
1324 
1325 	gl = ddp_make_gl(xferlen, sgl, sgcnt, cdev->pdev, gfp);
1326 	if (!gl)
1327 		return -ENOMEM;
1328 
1329 	err = ddp_tag_reserve(csk, csk->tid, sw_tag, tagp, gl, gfp);
1330 	if (err < 0)
1331 		ddp_release_gl(gl, cdev->pdev);
1332 
1333 	return err;
1334 }
1335 
1336 static void ddp_destroy(struct kref *kref)
1337 {
1338 	struct cxgbi_ddp_info *ddp = container_of(kref,
1339 						struct cxgbi_ddp_info,
1340 						refcnt);
1341 	struct cxgbi_device *cdev = ddp->cdev;
1342 	int i = 0;
1343 
1344 	pr_info("kref 0, destroy ddp 0x%p, cdev 0x%p.\n", ddp, cdev);
1345 
1346 	while (i < ddp->nppods) {
1347 		struct cxgbi_gather_list *gl = ddp->gl_map[i];
1348 
1349 		if (gl) {
1350 			int npods = (gl->nelem + PPOD_PAGES_MAX - 1)
1351 					>> PPOD_PAGES_SHIFT;
1352 			pr_info("cdev 0x%p, ddp %d + %d.\n", cdev, i, npods);
1353 			kfree(gl);
1354 			i += npods;
1355 		} else
1356 			i++;
1357 	}
1358 	cxgbi_free_big_mem(ddp);
1359 }
1360 
1361 int cxgbi_ddp_cleanup(struct cxgbi_device *cdev)
1362 {
1363 	struct cxgbi_ddp_info *ddp = cdev->ddp;
1364 
1365 	log_debug(1 << CXGBI_DBG_DDP,
1366 		"cdev 0x%p, release ddp 0x%p.\n", cdev, ddp);
1367 	cdev->ddp = NULL;
1368 	if (ddp)
1369 		return kref_put(&ddp->refcnt, ddp_destroy);
1370 	return 0;
1371 }
1372 EXPORT_SYMBOL_GPL(cxgbi_ddp_cleanup);
1373 
1374 int cxgbi_ddp_init(struct cxgbi_device *cdev,
1375 		   unsigned int llimit, unsigned int ulimit,
1376 		   unsigned int max_txsz, unsigned int max_rxsz)
1377 {
1378 	struct cxgbi_ddp_info *ddp;
1379 	unsigned int ppmax, bits;
1380 
1381 	ppmax = (ulimit - llimit + 1) >> PPOD_SIZE_SHIFT;
1382 	bits = __ilog2_u32(ppmax) + 1;
1383 	if (bits > PPOD_IDX_MAX_SIZE)
1384 		bits = PPOD_IDX_MAX_SIZE;
1385 	ppmax = (1 << (bits - 1)) - 1;
1386 
1387 	ddp = cxgbi_alloc_big_mem(sizeof(struct cxgbi_ddp_info) +
1388 				ppmax * (sizeof(struct cxgbi_gather_list *) +
1389 					 sizeof(struct sk_buff *)),
1390 				GFP_KERNEL);
1391 	if (!ddp) {
1392 		pr_warn("cdev 0x%p, ddp ppmax %u OOM.\n", cdev, ppmax);
1393 		return -ENOMEM;
1394 	}
1395 	ddp->gl_map = (struct cxgbi_gather_list **)(ddp + 1);
1396 	cdev->ddp = ddp;
1397 
1398 	spin_lock_init(&ddp->map_lock);
1399 	kref_init(&ddp->refcnt);
1400 
1401 	ddp->cdev = cdev;
1402 	ddp->pdev = cdev->pdev;
1403 	ddp->llimit = llimit;
1404 	ddp->ulimit = ulimit;
1405 	ddp->max_txsz = min_t(unsigned int, max_txsz, ULP2_MAX_PKT_SIZE);
1406 	ddp->max_rxsz = min_t(unsigned int, max_rxsz, ULP2_MAX_PKT_SIZE);
1407 	ddp->nppods = ppmax;
1408 	ddp->idx_last = ppmax;
1409 	ddp->idx_bits = bits;
1410 	ddp->idx_mask = (1 << bits) - 1;
1411 	ddp->rsvd_tag_mask = (1 << (bits + PPOD_IDX_SHIFT)) - 1;
1412 
1413 	cdev->tag_format.sw_bits = sw_tag_idx_bits + sw_tag_age_bits;
1414 	cdev->tag_format.rsvd_bits = ddp->idx_bits;
1415 	cdev->tag_format.rsvd_shift = PPOD_IDX_SHIFT;
1416 	cdev->tag_format.rsvd_mask = (1 << cdev->tag_format.rsvd_bits) - 1;
1417 
1418 	pr_info("%s tag format, sw %u, rsvd %u,%u, mask 0x%x.\n",
1419 		cdev->ports[0]->name, cdev->tag_format.sw_bits,
1420 		cdev->tag_format.rsvd_bits, cdev->tag_format.rsvd_shift,
1421 		cdev->tag_format.rsvd_mask);
1422 
1423 	cdev->tx_max_size = min_t(unsigned int, ULP2_MAX_PDU_PAYLOAD,
1424 				ddp->max_txsz - ISCSI_PDU_NONPAYLOAD_LEN);
1425 	cdev->rx_max_size = min_t(unsigned int, ULP2_MAX_PDU_PAYLOAD,
1426 				ddp->max_rxsz - ISCSI_PDU_NONPAYLOAD_LEN);
1427 
1428 	log_debug(1 << CXGBI_DBG_DDP,
1429 		"%s max payload size: %u/%u, %u/%u.\n",
1430 		cdev->ports[0]->name, cdev->tx_max_size, ddp->max_txsz,
1431 		cdev->rx_max_size, ddp->max_rxsz);
1432 	return 0;
1433 }
1434 EXPORT_SYMBOL_GPL(cxgbi_ddp_init);
1435 
1436 /*
1437  * APIs interacting with open-iscsi libraries
1438  */
1439 
1440 static unsigned char padding[4];
1441 
1442 static void task_release_itt(struct iscsi_task *task, itt_t hdr_itt)
1443 {
1444 	struct scsi_cmnd *sc = task->sc;
1445 	struct iscsi_tcp_conn *tcp_conn = task->conn->dd_data;
1446 	struct cxgbi_conn *cconn = tcp_conn->dd_data;
1447 	struct cxgbi_hba *chba = cconn->chba;
1448 	struct cxgbi_tag_format *tformat = &chba->cdev->tag_format;
1449 	u32 tag = ntohl((__force u32)hdr_itt);
1450 
1451 	log_debug(1 << CXGBI_DBG_DDP,
1452 		   "cdev 0x%p, release tag 0x%x.\n", chba->cdev, tag);
1453 	if (sc &&
1454 	    (scsi_bidi_cmnd(sc) || sc->sc_data_direction == DMA_FROM_DEVICE) &&
1455 	    cxgbi_is_ddp_tag(tformat, tag))
1456 		ddp_tag_release(chba, tag);
1457 }
1458 
1459 static int task_reserve_itt(struct iscsi_task *task, itt_t *hdr_itt)
1460 {
1461 	struct scsi_cmnd *sc = task->sc;
1462 	struct iscsi_conn *conn = task->conn;
1463 	struct iscsi_session *sess = conn->session;
1464 	struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
1465 	struct cxgbi_conn *cconn = tcp_conn->dd_data;
1466 	struct cxgbi_hba *chba = cconn->chba;
1467 	struct cxgbi_tag_format *tformat = &chba->cdev->tag_format;
1468 	u32 sw_tag = (sess->age << cconn->task_idx_bits) | task->itt;
1469 	u32 tag = 0;
1470 	int err = -EINVAL;
1471 
1472 	if (sc &&
1473 	    (scsi_bidi_cmnd(sc) || sc->sc_data_direction == DMA_FROM_DEVICE)) {
1474 		err = cxgbi_ddp_reserve(cconn->cep->csk, &tag, sw_tag,
1475 					scsi_in(sc)->length,
1476 					scsi_in(sc)->table.sgl,
1477 					scsi_in(sc)->table.nents,
1478 					GFP_ATOMIC);
1479 		if (err < 0)
1480 			log_debug(1 << CXGBI_DBG_DDP,
1481 				"csk 0x%p, R task 0x%p, %u,%u, no ddp.\n",
1482 				cconn->cep->csk, task, scsi_in(sc)->length,
1483 				scsi_in(sc)->table.nents);
1484 	}
1485 
1486 	if (err < 0)
1487 		tag = cxgbi_set_non_ddp_tag(tformat, sw_tag);
1488 	/*  the itt need to sent in big-endian order */
1489 	*hdr_itt = (__force itt_t)htonl(tag);
1490 
1491 	log_debug(1 << CXGBI_DBG_DDP,
1492 		"cdev 0x%p, task 0x%p, 0x%x(0x%x,0x%x)->0x%x/0x%x.\n",
1493 		chba->cdev, task, sw_tag, task->itt, sess->age, tag, *hdr_itt);
1494 	return 0;
1495 }
1496 
1497 void cxgbi_parse_pdu_itt(struct iscsi_conn *conn, itt_t itt, int *idx, int *age)
1498 {
1499 	struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
1500 	struct cxgbi_conn *cconn = tcp_conn->dd_data;
1501 	struct cxgbi_device *cdev = cconn->chba->cdev;
1502 	u32 tag = ntohl((__force u32) itt);
1503 	u32 sw_bits;
1504 
1505 	sw_bits = cxgbi_tag_nonrsvd_bits(&cdev->tag_format, tag);
1506 	if (idx)
1507 		*idx = sw_bits & ((1 << cconn->task_idx_bits) - 1);
1508 	if (age)
1509 		*age = (sw_bits >> cconn->task_idx_bits) & ISCSI_AGE_MASK;
1510 
1511 	log_debug(1 << CXGBI_DBG_DDP,
1512 		"cdev 0x%p, tag 0x%x/0x%x, -> 0x%x(0x%x,0x%x).\n",
1513 		cdev, tag, itt, sw_bits, idx ? *idx : 0xFFFFF,
1514 		age ? *age : 0xFF);
1515 }
1516 EXPORT_SYMBOL_GPL(cxgbi_parse_pdu_itt);
1517 
1518 void cxgbi_conn_tx_open(struct cxgbi_sock *csk)
1519 {
1520 	struct iscsi_conn *conn = csk->user_data;
1521 
1522 	if (conn) {
1523 		log_debug(1 << CXGBI_DBG_SOCK,
1524 			"csk 0x%p, cid %d.\n", csk, conn->id);
1525 		iscsi_conn_queue_work(conn);
1526 	}
1527 }
1528 EXPORT_SYMBOL_GPL(cxgbi_conn_tx_open);
1529 
1530 /*
1531  * pdu receive, interact with libiscsi_tcp
1532  */
1533 static inline int read_pdu_skb(struct iscsi_conn *conn,
1534 			       struct sk_buff *skb,
1535 			       unsigned int offset,
1536 			       int offloaded)
1537 {
1538 	int status = 0;
1539 	int bytes_read;
1540 
1541 	bytes_read = iscsi_tcp_recv_skb(conn, skb, offset, offloaded, &status);
1542 	switch (status) {
1543 	case ISCSI_TCP_CONN_ERR:
1544 		pr_info("skb 0x%p, off %u, %d, TCP_ERR.\n",
1545 			  skb, offset, offloaded);
1546 		return -EIO;
1547 	case ISCSI_TCP_SUSPENDED:
1548 		log_debug(1 << CXGBI_DBG_PDU_RX,
1549 			"skb 0x%p, off %u, %d, TCP_SUSPEND, rc %d.\n",
1550 			skb, offset, offloaded, bytes_read);
1551 		/* no transfer - just have caller flush queue */
1552 		return bytes_read;
1553 	case ISCSI_TCP_SKB_DONE:
1554 		pr_info("skb 0x%p, off %u, %d, TCP_SKB_DONE.\n",
1555 			skb, offset, offloaded);
1556 		/*
1557 		 * pdus should always fit in the skb and we should get
1558 		 * segment done notifcation.
1559 		 */
1560 		iscsi_conn_printk(KERN_ERR, conn, "Invalid pdu or skb.");
1561 		return -EFAULT;
1562 	case ISCSI_TCP_SEGMENT_DONE:
1563 		log_debug(1 << CXGBI_DBG_PDU_RX,
1564 			"skb 0x%p, off %u, %d, TCP_SEG_DONE, rc %d.\n",
1565 			skb, offset, offloaded, bytes_read);
1566 		return bytes_read;
1567 	default:
1568 		pr_info("skb 0x%p, off %u, %d, invalid status %d.\n",
1569 			skb, offset, offloaded, status);
1570 		return -EINVAL;
1571 	}
1572 }
1573 
1574 static int skb_read_pdu_bhs(struct iscsi_conn *conn, struct sk_buff *skb)
1575 {
1576 	struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
1577 
1578 	log_debug(1 << CXGBI_DBG_PDU_RX,
1579 		"conn 0x%p, skb 0x%p, len %u, flag 0x%lx.\n",
1580 		conn, skb, skb->len, cxgbi_skcb_flags(skb));
1581 
1582 	if (!iscsi_tcp_recv_segment_is_hdr(tcp_conn)) {
1583 		pr_info("conn 0x%p, skb 0x%p, not hdr.\n", conn, skb);
1584 		iscsi_conn_failure(conn, ISCSI_ERR_PROTO);
1585 		return -EIO;
1586 	}
1587 
1588 	if (conn->hdrdgst_en &&
1589 	    cxgbi_skcb_test_flag(skb, SKCBF_RX_HCRC_ERR)) {
1590 		pr_info("conn 0x%p, skb 0x%p, hcrc.\n", conn, skb);
1591 		iscsi_conn_failure(conn, ISCSI_ERR_HDR_DGST);
1592 		return -EIO;
1593 	}
1594 
1595 	return read_pdu_skb(conn, skb, 0, 0);
1596 }
1597 
1598 static int skb_read_pdu_data(struct iscsi_conn *conn, struct sk_buff *lskb,
1599 			     struct sk_buff *skb, unsigned int offset)
1600 {
1601 	struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
1602 	bool offloaded = 0;
1603 	int opcode = tcp_conn->in.hdr->opcode & ISCSI_OPCODE_MASK;
1604 
1605 	log_debug(1 << CXGBI_DBG_PDU_RX,
1606 		"conn 0x%p, skb 0x%p, len %u, flag 0x%lx.\n",
1607 		conn, skb, skb->len, cxgbi_skcb_flags(skb));
1608 
1609 	if (conn->datadgst_en &&
1610 	    cxgbi_skcb_test_flag(lskb, SKCBF_RX_DCRC_ERR)) {
1611 		pr_info("conn 0x%p, skb 0x%p, dcrc 0x%lx.\n",
1612 			conn, lskb, cxgbi_skcb_flags(lskb));
1613 		iscsi_conn_failure(conn, ISCSI_ERR_DATA_DGST);
1614 		return -EIO;
1615 	}
1616 
1617 	if (iscsi_tcp_recv_segment_is_hdr(tcp_conn))
1618 		return 0;
1619 
1620 	/* coalesced, add header digest length */
1621 	if (lskb == skb && conn->hdrdgst_en)
1622 		offset += ISCSI_DIGEST_SIZE;
1623 
1624 	if (cxgbi_skcb_test_flag(lskb, SKCBF_RX_DATA_DDPD))
1625 		offloaded = 1;
1626 
1627 	if (opcode == ISCSI_OP_SCSI_DATA_IN)
1628 		log_debug(1 << CXGBI_DBG_PDU_RX,
1629 			"skb 0x%p, op 0x%x, itt 0x%x, %u %s ddp'ed.\n",
1630 			skb, opcode, ntohl(tcp_conn->in.hdr->itt),
1631 			tcp_conn->in.datalen, offloaded ? "is" : "not");
1632 
1633 	return read_pdu_skb(conn, skb, offset, offloaded);
1634 }
1635 
1636 static void csk_return_rx_credits(struct cxgbi_sock *csk, int copied)
1637 {
1638 	struct cxgbi_device *cdev = csk->cdev;
1639 	int must_send;
1640 	u32 credits;
1641 
1642 	log_debug(1 << CXGBI_DBG_PDU_RX,
1643 		"csk 0x%p,%u,0x%lu,%u, seq %u, wup %u, thre %u, %u.\n",
1644 		csk, csk->state, csk->flags, csk->tid, csk->copied_seq,
1645 		csk->rcv_wup, cdev->rx_credit_thres,
1646 		cdev->rcv_win);
1647 
1648 	if (csk->state != CTP_ESTABLISHED)
1649 		return;
1650 
1651 	credits = csk->copied_seq - csk->rcv_wup;
1652 	if (unlikely(!credits))
1653 		return;
1654 	if (unlikely(cdev->rx_credit_thres == 0))
1655 		return;
1656 
1657 	must_send = credits + 16384 >= cdev->rcv_win;
1658 	if (must_send || credits >= cdev->rx_credit_thres)
1659 		csk->rcv_wup += cdev->csk_send_rx_credits(csk, credits);
1660 }
1661 
1662 void cxgbi_conn_pdu_ready(struct cxgbi_sock *csk)
1663 {
1664 	struct cxgbi_device *cdev = csk->cdev;
1665 	struct iscsi_conn *conn = csk->user_data;
1666 	struct sk_buff *skb;
1667 	unsigned int read = 0;
1668 	int err = 0;
1669 
1670 	log_debug(1 << CXGBI_DBG_PDU_RX,
1671 		"csk 0x%p, conn 0x%p.\n", csk, conn);
1672 
1673 	if (unlikely(!conn || conn->suspend_rx)) {
1674 		log_debug(1 << CXGBI_DBG_PDU_RX,
1675 			"csk 0x%p, conn 0x%p, id %d, suspend_rx %lu!\n",
1676 			csk, conn, conn ? conn->id : 0xFF,
1677 			conn ? conn->suspend_rx : 0xFF);
1678 		return;
1679 	}
1680 
1681 	while (!err) {
1682 		skb = skb_peek(&csk->receive_queue);
1683 		if (!skb ||
1684 		    !(cxgbi_skcb_test_flag(skb, SKCBF_RX_STATUS))) {
1685 			if (skb)
1686 				log_debug(1 << CXGBI_DBG_PDU_RX,
1687 					"skb 0x%p, NOT ready 0x%lx.\n",
1688 					skb, cxgbi_skcb_flags(skb));
1689 			break;
1690 		}
1691 		__skb_unlink(skb, &csk->receive_queue);
1692 
1693 		read += cxgbi_skcb_rx_pdulen(skb);
1694 		log_debug(1 << CXGBI_DBG_PDU_RX,
1695 			"csk 0x%p, skb 0x%p,%u,f 0x%lx, pdu len %u.\n",
1696 			csk, skb, skb->len, cxgbi_skcb_flags(skb),
1697 			cxgbi_skcb_rx_pdulen(skb));
1698 
1699 		if (cxgbi_skcb_test_flag(skb, SKCBF_RX_COALESCED)) {
1700 			err = skb_read_pdu_bhs(conn, skb);
1701 			if (err < 0) {
1702 				pr_err("coalesced bhs, csk 0x%p, skb 0x%p,%u, "
1703 					"f 0x%lx, plen %u.\n",
1704 					csk, skb, skb->len,
1705 					cxgbi_skcb_flags(skb),
1706 					cxgbi_skcb_rx_pdulen(skb));
1707 				goto skb_done;
1708 			}
1709 			err = skb_read_pdu_data(conn, skb, skb,
1710 						err + cdev->skb_rx_extra);
1711 			if (err < 0)
1712 				pr_err("coalesced data, csk 0x%p, skb 0x%p,%u, "
1713 					"f 0x%lx, plen %u.\n",
1714 					csk, skb, skb->len,
1715 					cxgbi_skcb_flags(skb),
1716 					cxgbi_skcb_rx_pdulen(skb));
1717 		} else {
1718 			err = skb_read_pdu_bhs(conn, skb);
1719 			if (err < 0) {
1720 				pr_err("bhs, csk 0x%p, skb 0x%p,%u, "
1721 					"f 0x%lx, plen %u.\n",
1722 					csk, skb, skb->len,
1723 					cxgbi_skcb_flags(skb),
1724 					cxgbi_skcb_rx_pdulen(skb));
1725 				goto skb_done;
1726 			}
1727 
1728 			if (cxgbi_skcb_test_flag(skb, SKCBF_RX_DATA)) {
1729 				struct sk_buff *dskb;
1730 
1731 				dskb = skb_peek(&csk->receive_queue);
1732 				if (!dskb) {
1733 					pr_err("csk 0x%p, skb 0x%p,%u, f 0x%lx,"
1734 						" plen %u, NO data.\n",
1735 						csk, skb, skb->len,
1736 						cxgbi_skcb_flags(skb),
1737 						cxgbi_skcb_rx_pdulen(skb));
1738 					err = -EIO;
1739 					goto skb_done;
1740 				}
1741 				__skb_unlink(dskb, &csk->receive_queue);
1742 
1743 				err = skb_read_pdu_data(conn, skb, dskb, 0);
1744 				if (err < 0)
1745 					pr_err("data, csk 0x%p, skb 0x%p,%u, "
1746 						"f 0x%lx, plen %u, dskb 0x%p,"
1747 						"%u.\n",
1748 						csk, skb, skb->len,
1749 						cxgbi_skcb_flags(skb),
1750 						cxgbi_skcb_rx_pdulen(skb),
1751 						dskb, dskb->len);
1752 				__kfree_skb(dskb);
1753 			} else
1754 				err = skb_read_pdu_data(conn, skb, skb, 0);
1755 		}
1756 skb_done:
1757 		__kfree_skb(skb);
1758 
1759 		if (err < 0)
1760 			break;
1761 	}
1762 
1763 	log_debug(1 << CXGBI_DBG_PDU_RX, "csk 0x%p, read %u.\n", csk, read);
1764 	if (read) {
1765 		csk->copied_seq += read;
1766 		csk_return_rx_credits(csk, read);
1767 		conn->rxdata_octets += read;
1768 	}
1769 
1770 	if (err < 0) {
1771 		pr_info("csk 0x%p, 0x%p, rx failed %d, read %u.\n",
1772 			csk, conn, err, read);
1773 		iscsi_conn_failure(conn, ISCSI_ERR_CONN_FAILED);
1774 	}
1775 }
1776 EXPORT_SYMBOL_GPL(cxgbi_conn_pdu_ready);
1777 
1778 static int sgl_seek_offset(struct scatterlist *sgl, unsigned int sgcnt,
1779 				unsigned int offset, unsigned int *off,
1780 				struct scatterlist **sgp)
1781 {
1782 	int i;
1783 	struct scatterlist *sg;
1784 
1785 	for_each_sg(sgl, sg, sgcnt, i) {
1786 		if (offset < sg->length) {
1787 			*off = offset;
1788 			*sgp = sg;
1789 			return 0;
1790 		}
1791 		offset -= sg->length;
1792 	}
1793 	return -EFAULT;
1794 }
1795 
1796 static int sgl_read_to_frags(struct scatterlist *sg, unsigned int sgoffset,
1797 				unsigned int dlen, struct page_frag *frags,
1798 				int frag_max)
1799 {
1800 	unsigned int datalen = dlen;
1801 	unsigned int sglen = sg->length - sgoffset;
1802 	struct page *page = sg_page(sg);
1803 	int i;
1804 
1805 	i = 0;
1806 	do {
1807 		unsigned int copy;
1808 
1809 		if (!sglen) {
1810 			sg = sg_next(sg);
1811 			if (!sg) {
1812 				pr_warn("sg %d NULL, len %u/%u.\n",
1813 					i, datalen, dlen);
1814 				return -EINVAL;
1815 			}
1816 			sgoffset = 0;
1817 			sglen = sg->length;
1818 			page = sg_page(sg);
1819 
1820 		}
1821 		copy = min(datalen, sglen);
1822 		if (i && page == frags[i - 1].page &&
1823 		    sgoffset + sg->offset ==
1824 			frags[i - 1].offset + frags[i - 1].size) {
1825 			frags[i - 1].size += copy;
1826 		} else {
1827 			if (i >= frag_max) {
1828 				pr_warn("too many pages %u, dlen %u.\n",
1829 					frag_max, dlen);
1830 				return -EINVAL;
1831 			}
1832 
1833 			frags[i].page = page;
1834 			frags[i].offset = sg->offset + sgoffset;
1835 			frags[i].size = copy;
1836 			i++;
1837 		}
1838 		datalen -= copy;
1839 		sgoffset += copy;
1840 		sglen -= copy;
1841 	} while (datalen);
1842 
1843 	return i;
1844 }
1845 
1846 int cxgbi_conn_alloc_pdu(struct iscsi_task *task, u8 opcode)
1847 {
1848 	struct iscsi_tcp_conn *tcp_conn = task->conn->dd_data;
1849 	struct cxgbi_conn *cconn = tcp_conn->dd_data;
1850 	struct cxgbi_device *cdev = cconn->chba->cdev;
1851 	struct iscsi_conn *conn = task->conn;
1852 	struct iscsi_tcp_task *tcp_task = task->dd_data;
1853 	struct cxgbi_task_data *tdata = iscsi_task_cxgbi_data(task);
1854 	struct scsi_cmnd *sc = task->sc;
1855 	int headroom = SKB_TX_ISCSI_PDU_HEADER_MAX;
1856 
1857 	tcp_task->dd_data = tdata;
1858 	task->hdr = NULL;
1859 
1860 	if (SKB_MAX_HEAD(cdev->skb_tx_rsvd) > (512 * MAX_SKB_FRAGS) &&
1861 	    (opcode == ISCSI_OP_SCSI_DATA_OUT ||
1862 	     (opcode == ISCSI_OP_SCSI_CMD &&
1863 	      (scsi_bidi_cmnd(sc) || sc->sc_data_direction == DMA_TO_DEVICE))))
1864 		/* data could goes into skb head */
1865 		headroom += min_t(unsigned int,
1866 				SKB_MAX_HEAD(cdev->skb_tx_rsvd),
1867 				conn->max_xmit_dlength);
1868 
1869 	tdata->skb = alloc_skb(cdev->skb_tx_rsvd + headroom, GFP_ATOMIC);
1870 	if (!tdata->skb) {
1871 		struct cxgbi_sock *csk = cconn->cep->csk;
1872 		struct net_device *ndev = cdev->ports[csk->port_id];
1873 		ndev->stats.tx_dropped++;
1874 		return -ENOMEM;
1875 	}
1876 
1877 	skb_reserve(tdata->skb, cdev->skb_tx_rsvd);
1878 	task->hdr = (struct iscsi_hdr *)tdata->skb->data;
1879 	task->hdr_max = SKB_TX_ISCSI_PDU_HEADER_MAX; /* BHS + AHS */
1880 
1881 	/* data_out uses scsi_cmd's itt */
1882 	if (opcode != ISCSI_OP_SCSI_DATA_OUT)
1883 		task_reserve_itt(task, &task->hdr->itt);
1884 
1885 	log_debug(1 << CXGBI_DBG_ISCSI | 1 << CXGBI_DBG_PDU_TX,
1886 		"task 0x%p, op 0x%x, skb 0x%p,%u+%u/%u, itt 0x%x.\n",
1887 		task, opcode, tdata->skb, cdev->skb_tx_rsvd, headroom,
1888 		conn->max_xmit_dlength, ntohl(task->hdr->itt));
1889 
1890 	return 0;
1891 }
1892 EXPORT_SYMBOL_GPL(cxgbi_conn_alloc_pdu);
1893 
1894 static inline void tx_skb_setmode(struct sk_buff *skb, int hcrc, int dcrc)
1895 {
1896 	if (hcrc || dcrc) {
1897 		u8 submode = 0;
1898 
1899 		if (hcrc)
1900 			submode |= 1;
1901 		if (dcrc)
1902 			submode |= 2;
1903 		cxgbi_skcb_ulp_mode(skb) = (ULP2_MODE_ISCSI << 4) | submode;
1904 	} else
1905 		cxgbi_skcb_ulp_mode(skb) = 0;
1906 }
1907 
1908 int cxgbi_conn_init_pdu(struct iscsi_task *task, unsigned int offset,
1909 			      unsigned int count)
1910 {
1911 	struct iscsi_conn *conn = task->conn;
1912 	struct cxgbi_task_data *tdata = iscsi_task_cxgbi_data(task);
1913 	struct sk_buff *skb = tdata->skb;
1914 	unsigned int datalen = count;
1915 	int i, padlen = iscsi_padding(count);
1916 	struct page *pg;
1917 
1918 	log_debug(1 << CXGBI_DBG_ISCSI | 1 << CXGBI_DBG_PDU_TX,
1919 		"task 0x%p,0x%p, skb 0x%p, 0x%x,0x%x,0x%x, %u+%u.\n",
1920 		task, task->sc, skb, (*skb->data) & ISCSI_OPCODE_MASK,
1921 		ntohl(task->cmdsn), ntohl(task->hdr->itt), offset, count);
1922 
1923 	skb_put(skb, task->hdr_len);
1924 	tx_skb_setmode(skb, conn->hdrdgst_en, datalen ? conn->datadgst_en : 0);
1925 	if (!count)
1926 		return 0;
1927 
1928 	if (task->sc) {
1929 		struct scsi_data_buffer *sdb = scsi_out(task->sc);
1930 		struct scatterlist *sg = NULL;
1931 		int err;
1932 
1933 		tdata->offset = offset;
1934 		tdata->count = count;
1935 		err = sgl_seek_offset(
1936 					sdb->table.sgl, sdb->table.nents,
1937 					tdata->offset, &tdata->sgoffset, &sg);
1938 		if (err < 0) {
1939 			pr_warn("tpdu, sgl %u, bad offset %u/%u.\n",
1940 				sdb->table.nents, tdata->offset, sdb->length);
1941 			return err;
1942 		}
1943 		err = sgl_read_to_frags(sg, tdata->sgoffset, tdata->count,
1944 					tdata->frags, MAX_PDU_FRAGS);
1945 		if (err < 0) {
1946 			pr_warn("tpdu, sgl %u, bad offset %u + %u.\n",
1947 				sdb->table.nents, tdata->offset, tdata->count);
1948 			return err;
1949 		}
1950 		tdata->nr_frags = err;
1951 
1952 		if (tdata->nr_frags > MAX_SKB_FRAGS ||
1953 		    (padlen && tdata->nr_frags == MAX_SKB_FRAGS)) {
1954 			char *dst = skb->data + task->hdr_len;
1955 			struct page_frag *frag = tdata->frags;
1956 
1957 			/* data fits in the skb's headroom */
1958 			for (i = 0; i < tdata->nr_frags; i++, frag++) {
1959 				char *src = kmap_atomic(frag->page);
1960 
1961 				memcpy(dst, src+frag->offset, frag->size);
1962 				dst += frag->size;
1963 				kunmap_atomic(src);
1964 			}
1965 			if (padlen) {
1966 				memset(dst, 0, padlen);
1967 				padlen = 0;
1968 			}
1969 			skb_put(skb, count + padlen);
1970 		} else {
1971 			/* data fit into frag_list */
1972 			for (i = 0; i < tdata->nr_frags; i++) {
1973 				__skb_fill_page_desc(skb, i,
1974 						tdata->frags[i].page,
1975 						tdata->frags[i].offset,
1976 						tdata->frags[i].size);
1977 				skb_frag_ref(skb, i);
1978 			}
1979 			skb_shinfo(skb)->nr_frags = tdata->nr_frags;
1980 			skb->len += count;
1981 			skb->data_len += count;
1982 			skb->truesize += count;
1983 		}
1984 
1985 	} else {
1986 		pg = virt_to_page(task->data);
1987 
1988 		get_page(pg);
1989 		skb_fill_page_desc(skb, 0, pg, offset_in_page(task->data),
1990 					count);
1991 		skb->len += count;
1992 		skb->data_len += count;
1993 		skb->truesize += count;
1994 	}
1995 
1996 	if (padlen) {
1997 		i = skb_shinfo(skb)->nr_frags;
1998 		skb_fill_page_desc(skb, skb_shinfo(skb)->nr_frags,
1999 				virt_to_page(padding), offset_in_page(padding),
2000 				padlen);
2001 
2002 		skb->data_len += padlen;
2003 		skb->truesize += padlen;
2004 		skb->len += padlen;
2005 	}
2006 
2007 	return 0;
2008 }
2009 EXPORT_SYMBOL_GPL(cxgbi_conn_init_pdu);
2010 
2011 int cxgbi_conn_xmit_pdu(struct iscsi_task *task)
2012 {
2013 	struct iscsi_tcp_conn *tcp_conn = task->conn->dd_data;
2014 	struct cxgbi_conn *cconn = tcp_conn->dd_data;
2015 	struct cxgbi_task_data *tdata = iscsi_task_cxgbi_data(task);
2016 	struct sk_buff *skb = tdata->skb;
2017 	unsigned int datalen;
2018 	int err;
2019 
2020 	if (!skb) {
2021 		log_debug(1 << CXGBI_DBG_ISCSI | 1 << CXGBI_DBG_PDU_TX,
2022 			"task 0x%p, skb NULL.\n", task);
2023 		return 0;
2024 	}
2025 
2026 	datalen = skb->data_len;
2027 	tdata->skb = NULL;
2028 	err = cxgbi_sock_send_pdus(cconn->cep->csk, skb);
2029 	if (err > 0) {
2030 		int pdulen = err;
2031 
2032 		log_debug(1 << CXGBI_DBG_PDU_TX,
2033 			"task 0x%p,0x%p, skb 0x%p, len %u/%u, rv %d.\n",
2034 			task, task->sc, skb, skb->len, skb->data_len, err);
2035 
2036 		if (task->conn->hdrdgst_en)
2037 			pdulen += ISCSI_DIGEST_SIZE;
2038 
2039 		if (datalen && task->conn->datadgst_en)
2040 			pdulen += ISCSI_DIGEST_SIZE;
2041 
2042 		task->conn->txdata_octets += pdulen;
2043 		return 0;
2044 	}
2045 
2046 	if (err == -EAGAIN || err == -ENOBUFS) {
2047 		log_debug(1 << CXGBI_DBG_PDU_TX,
2048 			"task 0x%p, skb 0x%p, len %u/%u, %d EAGAIN.\n",
2049 			task, skb, skb->len, skb->data_len, err);
2050 		/* reset skb to send when we are called again */
2051 		tdata->skb = skb;
2052 		return err;
2053 	}
2054 
2055 	kfree_skb(skb);
2056 	log_debug(1 << CXGBI_DBG_ISCSI | 1 << CXGBI_DBG_PDU_TX,
2057 		"itt 0x%x, skb 0x%p, len %u/%u, xmit err %d.\n",
2058 		task->itt, skb, skb->len, skb->data_len, err);
2059 	iscsi_conn_printk(KERN_ERR, task->conn, "xmit err %d.\n", err);
2060 	iscsi_conn_failure(task->conn, ISCSI_ERR_XMIT_FAILED);
2061 	return err;
2062 }
2063 EXPORT_SYMBOL_GPL(cxgbi_conn_xmit_pdu);
2064 
2065 void cxgbi_cleanup_task(struct iscsi_task *task)
2066 {
2067 	struct cxgbi_task_data *tdata = iscsi_task_cxgbi_data(task);
2068 
2069 	log_debug(1 << CXGBI_DBG_ISCSI,
2070 		"task 0x%p, skb 0x%p, itt 0x%x.\n",
2071 		task, tdata->skb, task->hdr_itt);
2072 
2073 	/*  never reached the xmit task callout */
2074 	if (tdata->skb)
2075 		__kfree_skb(tdata->skb);
2076 	memset(tdata, 0, sizeof(*tdata));
2077 
2078 	task_release_itt(task, task->hdr_itt);
2079 	iscsi_tcp_cleanup_task(task);
2080 }
2081 EXPORT_SYMBOL_GPL(cxgbi_cleanup_task);
2082 
2083 void cxgbi_get_conn_stats(struct iscsi_cls_conn *cls_conn,
2084 				struct iscsi_stats *stats)
2085 {
2086 	struct iscsi_conn *conn = cls_conn->dd_data;
2087 
2088 	stats->txdata_octets = conn->txdata_octets;
2089 	stats->rxdata_octets = conn->rxdata_octets;
2090 	stats->scsicmd_pdus = conn->scsicmd_pdus_cnt;
2091 	stats->dataout_pdus = conn->dataout_pdus_cnt;
2092 	stats->scsirsp_pdus = conn->scsirsp_pdus_cnt;
2093 	stats->datain_pdus = conn->datain_pdus_cnt;
2094 	stats->r2t_pdus = conn->r2t_pdus_cnt;
2095 	stats->tmfcmd_pdus = conn->tmfcmd_pdus_cnt;
2096 	stats->tmfrsp_pdus = conn->tmfrsp_pdus_cnt;
2097 	stats->digest_err = 0;
2098 	stats->timeout_err = 0;
2099 	stats->custom_length = 1;
2100 	strcpy(stats->custom[0].desc, "eh_abort_cnt");
2101 	stats->custom[0].value = conn->eh_abort_cnt;
2102 }
2103 EXPORT_SYMBOL_GPL(cxgbi_get_conn_stats);
2104 
2105 static int cxgbi_conn_max_xmit_dlength(struct iscsi_conn *conn)
2106 {
2107 	struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
2108 	struct cxgbi_conn *cconn = tcp_conn->dd_data;
2109 	struct cxgbi_device *cdev = cconn->chba->cdev;
2110 	unsigned int headroom = SKB_MAX_HEAD(cdev->skb_tx_rsvd);
2111 	unsigned int max_def = 512 * MAX_SKB_FRAGS;
2112 	unsigned int max = max(max_def, headroom);
2113 
2114 	max = min(cconn->chba->cdev->tx_max_size, max);
2115 	if (conn->max_xmit_dlength)
2116 		conn->max_xmit_dlength = min(conn->max_xmit_dlength, max);
2117 	else
2118 		conn->max_xmit_dlength = max;
2119 	cxgbi_align_pdu_size(conn->max_xmit_dlength);
2120 
2121 	return 0;
2122 }
2123 
2124 static int cxgbi_conn_max_recv_dlength(struct iscsi_conn *conn)
2125 {
2126 	struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
2127 	struct cxgbi_conn *cconn = tcp_conn->dd_data;
2128 	unsigned int max = cconn->chba->cdev->rx_max_size;
2129 
2130 	cxgbi_align_pdu_size(max);
2131 
2132 	if (conn->max_recv_dlength) {
2133 		if (conn->max_recv_dlength > max) {
2134 			pr_err("MaxRecvDataSegmentLength %u > %u.\n",
2135 				conn->max_recv_dlength, max);
2136 			return -EINVAL;
2137 		}
2138 		conn->max_recv_dlength = min(conn->max_recv_dlength, max);
2139 		cxgbi_align_pdu_size(conn->max_recv_dlength);
2140 	} else
2141 		conn->max_recv_dlength = max;
2142 
2143 	return 0;
2144 }
2145 
2146 int cxgbi_set_conn_param(struct iscsi_cls_conn *cls_conn,
2147 			enum iscsi_param param, char *buf, int buflen)
2148 {
2149 	struct iscsi_conn *conn = cls_conn->dd_data;
2150 	struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
2151 	struct cxgbi_conn *cconn = tcp_conn->dd_data;
2152 	struct cxgbi_sock *csk = cconn->cep->csk;
2153 	int err;
2154 
2155 	log_debug(1 << CXGBI_DBG_ISCSI,
2156 		"cls_conn 0x%p, param %d, buf(%d) %s.\n",
2157 		cls_conn, param, buflen, buf);
2158 
2159 	switch (param) {
2160 	case ISCSI_PARAM_HDRDGST_EN:
2161 		err = iscsi_set_param(cls_conn, param, buf, buflen);
2162 		if (!err && conn->hdrdgst_en)
2163 			err = csk->cdev->csk_ddp_setup_digest(csk, csk->tid,
2164 							conn->hdrdgst_en,
2165 							conn->datadgst_en, 0);
2166 		break;
2167 	case ISCSI_PARAM_DATADGST_EN:
2168 		err = iscsi_set_param(cls_conn, param, buf, buflen);
2169 		if (!err && conn->datadgst_en)
2170 			err = csk->cdev->csk_ddp_setup_digest(csk, csk->tid,
2171 							conn->hdrdgst_en,
2172 							conn->datadgst_en, 0);
2173 		break;
2174 	case ISCSI_PARAM_MAX_R2T:
2175 		return iscsi_tcp_set_max_r2t(conn, buf);
2176 	case ISCSI_PARAM_MAX_RECV_DLENGTH:
2177 		err = iscsi_set_param(cls_conn, param, buf, buflen);
2178 		if (!err)
2179 			err = cxgbi_conn_max_recv_dlength(conn);
2180 		break;
2181 	case ISCSI_PARAM_MAX_XMIT_DLENGTH:
2182 		err = iscsi_set_param(cls_conn, param, buf, buflen);
2183 		if (!err)
2184 			err = cxgbi_conn_max_xmit_dlength(conn);
2185 		break;
2186 	default:
2187 		return iscsi_set_param(cls_conn, param, buf, buflen);
2188 	}
2189 	return err;
2190 }
2191 EXPORT_SYMBOL_GPL(cxgbi_set_conn_param);
2192 
2193 int cxgbi_get_ep_param(struct iscsi_endpoint *ep, enum iscsi_param param,
2194 		       char *buf)
2195 {
2196 	struct cxgbi_endpoint *cep = ep->dd_data;
2197 	struct cxgbi_sock *csk;
2198 	int len;
2199 
2200 	log_debug(1 << CXGBI_DBG_ISCSI,
2201 		"cls_conn 0x%p, param %d.\n", ep, param);
2202 
2203 	switch (param) {
2204 	case ISCSI_PARAM_CONN_PORT:
2205 	case ISCSI_PARAM_CONN_ADDRESS:
2206 		if (!cep)
2207 			return -ENOTCONN;
2208 
2209 		csk = cep->csk;
2210 		if (!csk)
2211 			return -ENOTCONN;
2212 
2213 		return iscsi_conn_get_addr_param((struct sockaddr_storage *)
2214 						 &csk->daddr, param, buf);
2215 	default:
2216 		return -ENOSYS;
2217 	}
2218 	return len;
2219 }
2220 EXPORT_SYMBOL_GPL(cxgbi_get_ep_param);
2221 
2222 struct iscsi_cls_conn *
2223 cxgbi_create_conn(struct iscsi_cls_session *cls_session, u32 cid)
2224 {
2225 	struct iscsi_cls_conn *cls_conn;
2226 	struct iscsi_conn *conn;
2227 	struct iscsi_tcp_conn *tcp_conn;
2228 	struct cxgbi_conn *cconn;
2229 
2230 	cls_conn = iscsi_tcp_conn_setup(cls_session, sizeof(*cconn), cid);
2231 	if (!cls_conn)
2232 		return NULL;
2233 
2234 	conn = cls_conn->dd_data;
2235 	tcp_conn = conn->dd_data;
2236 	cconn = tcp_conn->dd_data;
2237 	cconn->iconn = conn;
2238 
2239 	log_debug(1 << CXGBI_DBG_ISCSI,
2240 		"cid %u(0x%x), cls 0x%p,0x%p, conn 0x%p,0x%p,0x%p.\n",
2241 		cid, cid, cls_session, cls_conn, conn, tcp_conn, cconn);
2242 
2243 	return cls_conn;
2244 }
2245 EXPORT_SYMBOL_GPL(cxgbi_create_conn);
2246 
2247 int cxgbi_bind_conn(struct iscsi_cls_session *cls_session,
2248 				struct iscsi_cls_conn *cls_conn,
2249 				u64 transport_eph, int is_leading)
2250 {
2251 	struct iscsi_conn *conn = cls_conn->dd_data;
2252 	struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
2253 	struct cxgbi_conn *cconn = tcp_conn->dd_data;
2254 	struct iscsi_endpoint *ep;
2255 	struct cxgbi_endpoint *cep;
2256 	struct cxgbi_sock *csk;
2257 	int err;
2258 
2259 	ep = iscsi_lookup_endpoint(transport_eph);
2260 	if (!ep)
2261 		return -EINVAL;
2262 
2263 	/*  setup ddp pagesize */
2264 	cep = ep->dd_data;
2265 	csk = cep->csk;
2266 	err = csk->cdev->csk_ddp_setup_pgidx(csk, csk->tid, page_idx, 0);
2267 	if (err < 0)
2268 		return err;
2269 
2270 	err = iscsi_conn_bind(cls_session, cls_conn, is_leading);
2271 	if (err)
2272 		return -EINVAL;
2273 
2274 	/*  calculate the tag idx bits needed for this conn based on cmds_max */
2275 	cconn->task_idx_bits = (__ilog2_u32(conn->session->cmds_max - 1)) + 1;
2276 
2277 	write_lock_bh(&csk->callback_lock);
2278 	csk->user_data = conn;
2279 	cconn->chba = cep->chba;
2280 	cconn->cep = cep;
2281 	cep->cconn = cconn;
2282 	write_unlock_bh(&csk->callback_lock);
2283 
2284 	cxgbi_conn_max_xmit_dlength(conn);
2285 	cxgbi_conn_max_recv_dlength(conn);
2286 
2287 	log_debug(1 << CXGBI_DBG_ISCSI,
2288 		"cls 0x%p,0x%p, ep 0x%p, cconn 0x%p, csk 0x%p.\n",
2289 		cls_session, cls_conn, ep, cconn, csk);
2290 	/*  init recv engine */
2291 	iscsi_tcp_hdr_recv_prep(tcp_conn);
2292 
2293 	return 0;
2294 }
2295 EXPORT_SYMBOL_GPL(cxgbi_bind_conn);
2296 
2297 struct iscsi_cls_session *cxgbi_create_session(struct iscsi_endpoint *ep,
2298 						u16 cmds_max, u16 qdepth,
2299 						u32 initial_cmdsn)
2300 {
2301 	struct cxgbi_endpoint *cep;
2302 	struct cxgbi_hba *chba;
2303 	struct Scsi_Host *shost;
2304 	struct iscsi_cls_session *cls_session;
2305 	struct iscsi_session *session;
2306 
2307 	if (!ep) {
2308 		pr_err("missing endpoint.\n");
2309 		return NULL;
2310 	}
2311 
2312 	cep = ep->dd_data;
2313 	chba = cep->chba;
2314 	shost = chba->shost;
2315 
2316 	BUG_ON(chba != iscsi_host_priv(shost));
2317 
2318 	cls_session = iscsi_session_setup(chba->cdev->itp, shost,
2319 					cmds_max, 0,
2320 					sizeof(struct iscsi_tcp_task) +
2321 					sizeof(struct cxgbi_task_data),
2322 					initial_cmdsn, ISCSI_MAX_TARGET);
2323 	if (!cls_session)
2324 		return NULL;
2325 
2326 	session = cls_session->dd_data;
2327 	if (iscsi_tcp_r2tpool_alloc(session))
2328 		goto remove_session;
2329 
2330 	log_debug(1 << CXGBI_DBG_ISCSI,
2331 		"ep 0x%p, cls sess 0x%p.\n", ep, cls_session);
2332 	return cls_session;
2333 
2334 remove_session:
2335 	iscsi_session_teardown(cls_session);
2336 	return NULL;
2337 }
2338 EXPORT_SYMBOL_GPL(cxgbi_create_session);
2339 
2340 void cxgbi_destroy_session(struct iscsi_cls_session *cls_session)
2341 {
2342 	log_debug(1 << CXGBI_DBG_ISCSI,
2343 		"cls sess 0x%p.\n", cls_session);
2344 
2345 	iscsi_tcp_r2tpool_free(cls_session->dd_data);
2346 	iscsi_session_teardown(cls_session);
2347 }
2348 EXPORT_SYMBOL_GPL(cxgbi_destroy_session);
2349 
2350 int cxgbi_set_host_param(struct Scsi_Host *shost, enum iscsi_host_param param,
2351 			char *buf, int buflen)
2352 {
2353 	struct cxgbi_hba *chba = iscsi_host_priv(shost);
2354 
2355 	if (!chba->ndev) {
2356 		shost_printk(KERN_ERR, shost, "Could not get host param. "
2357 				"netdev for host not set.\n");
2358 		return -ENODEV;
2359 	}
2360 
2361 	log_debug(1 << CXGBI_DBG_ISCSI,
2362 		"shost 0x%p, hba 0x%p,%s, param %d, buf(%d) %s.\n",
2363 		shost, chba, chba->ndev->name, param, buflen, buf);
2364 
2365 	switch (param) {
2366 	case ISCSI_HOST_PARAM_IPADDRESS:
2367 	{
2368 		__be32 addr = in_aton(buf);
2369 		log_debug(1 << CXGBI_DBG_ISCSI,
2370 			"hba %s, req. ipv4 %pI4.\n", chba->ndev->name, &addr);
2371 		cxgbi_set_iscsi_ipv4(chba, addr);
2372 		return 0;
2373 	}
2374 	case ISCSI_HOST_PARAM_HWADDRESS:
2375 	case ISCSI_HOST_PARAM_NETDEV_NAME:
2376 		return 0;
2377 	default:
2378 		return iscsi_host_set_param(shost, param, buf, buflen);
2379 	}
2380 }
2381 EXPORT_SYMBOL_GPL(cxgbi_set_host_param);
2382 
2383 int cxgbi_get_host_param(struct Scsi_Host *shost, enum iscsi_host_param param,
2384 			char *buf)
2385 {
2386 	struct cxgbi_hba *chba = iscsi_host_priv(shost);
2387 	int len = 0;
2388 
2389 	if (!chba->ndev) {
2390 		shost_printk(KERN_ERR, shost, "Could not get host param. "
2391 				"netdev for host not set.\n");
2392 		return -ENODEV;
2393 	}
2394 
2395 	log_debug(1 << CXGBI_DBG_ISCSI,
2396 		"shost 0x%p, hba 0x%p,%s, param %d.\n",
2397 		shost, chba, chba->ndev->name, param);
2398 
2399 	switch (param) {
2400 	case ISCSI_HOST_PARAM_HWADDRESS:
2401 		len = sysfs_format_mac(buf, chba->ndev->dev_addr, 6);
2402 		break;
2403 	case ISCSI_HOST_PARAM_NETDEV_NAME:
2404 		len = sprintf(buf, "%s\n", chba->ndev->name);
2405 		break;
2406 	case ISCSI_HOST_PARAM_IPADDRESS:
2407 	{
2408 		__be32 addr;
2409 
2410 		addr = cxgbi_get_iscsi_ipv4(chba);
2411 		len = sprintf(buf, "%pI4", &addr);
2412 		log_debug(1 << CXGBI_DBG_ISCSI,
2413 			"hba %s, ipv4 %pI4.\n", chba->ndev->name, &addr);
2414 		break;
2415 	}
2416 	default:
2417 		return iscsi_host_get_param(shost, param, buf);
2418 	}
2419 
2420 	return len;
2421 }
2422 EXPORT_SYMBOL_GPL(cxgbi_get_host_param);
2423 
2424 struct iscsi_endpoint *cxgbi_ep_connect(struct Scsi_Host *shost,
2425 					struct sockaddr *dst_addr,
2426 					int non_blocking)
2427 {
2428 	struct iscsi_endpoint *ep;
2429 	struct cxgbi_endpoint *cep;
2430 	struct cxgbi_hba *hba = NULL;
2431 	struct cxgbi_sock *csk;
2432 	int err = -EINVAL;
2433 
2434 	log_debug(1 << CXGBI_DBG_ISCSI | 1 << CXGBI_DBG_SOCK,
2435 		"shost 0x%p, non_blocking %d, dst_addr 0x%p.\n",
2436 		shost, non_blocking, dst_addr);
2437 
2438 	if (shost) {
2439 		hba = iscsi_host_priv(shost);
2440 		if (!hba) {
2441 			pr_info("shost 0x%p, priv NULL.\n", shost);
2442 			goto err_out;
2443 		}
2444 	}
2445 
2446 	csk = cxgbi_check_route(dst_addr);
2447 	if (IS_ERR(csk))
2448 		return (struct iscsi_endpoint *)csk;
2449 	cxgbi_sock_get(csk);
2450 
2451 	if (!hba)
2452 		hba = csk->cdev->hbas[csk->port_id];
2453 	else if (hba != csk->cdev->hbas[csk->port_id]) {
2454 		pr_info("Could not connect through requested host %u"
2455 			"hba 0x%p != 0x%p (%u).\n",
2456 			shost->host_no, hba,
2457 			csk->cdev->hbas[csk->port_id], csk->port_id);
2458 		err = -ENOSPC;
2459 		goto release_conn;
2460 	}
2461 
2462 	err = sock_get_port(csk);
2463 	if (err)
2464 		goto release_conn;
2465 
2466 	cxgbi_sock_set_state(csk, CTP_CONNECTING);
2467 	err = csk->cdev->csk_init_act_open(csk);
2468 	if (err)
2469 		goto release_conn;
2470 
2471 	if (cxgbi_sock_is_closing(csk)) {
2472 		err = -ENOSPC;
2473 		pr_info("csk 0x%p is closing.\n", csk);
2474 		goto release_conn;
2475 	}
2476 
2477 	ep = iscsi_create_endpoint(sizeof(*cep));
2478 	if (!ep) {
2479 		err = -ENOMEM;
2480 		pr_info("iscsi alloc ep, OOM.\n");
2481 		goto release_conn;
2482 	}
2483 
2484 	cep = ep->dd_data;
2485 	cep->csk = csk;
2486 	cep->chba = hba;
2487 
2488 	log_debug(1 << CXGBI_DBG_ISCSI | 1 << CXGBI_DBG_SOCK,
2489 		"ep 0x%p, cep 0x%p, csk 0x%p, hba 0x%p,%s.\n",
2490 		ep, cep, csk, hba, hba->ndev->name);
2491 	return ep;
2492 
2493 release_conn:
2494 	cxgbi_sock_put(csk);
2495 	cxgbi_sock_closed(csk);
2496 err_out:
2497 	return ERR_PTR(err);
2498 }
2499 EXPORT_SYMBOL_GPL(cxgbi_ep_connect);
2500 
2501 int cxgbi_ep_poll(struct iscsi_endpoint *ep, int timeout_ms)
2502 {
2503 	struct cxgbi_endpoint *cep = ep->dd_data;
2504 	struct cxgbi_sock *csk = cep->csk;
2505 
2506 	if (!cxgbi_sock_is_established(csk))
2507 		return 0;
2508 	return 1;
2509 }
2510 EXPORT_SYMBOL_GPL(cxgbi_ep_poll);
2511 
2512 void cxgbi_ep_disconnect(struct iscsi_endpoint *ep)
2513 {
2514 	struct cxgbi_endpoint *cep = ep->dd_data;
2515 	struct cxgbi_conn *cconn = cep->cconn;
2516 	struct cxgbi_sock *csk = cep->csk;
2517 
2518 	log_debug(1 << CXGBI_DBG_ISCSI | 1 << CXGBI_DBG_SOCK,
2519 		"ep 0x%p, cep 0x%p, cconn 0x%p, csk 0x%p,%u,0x%lx.\n",
2520 		ep, cep, cconn, csk, csk->state, csk->flags);
2521 
2522 	if (cconn && cconn->iconn) {
2523 		iscsi_suspend_tx(cconn->iconn);
2524 		write_lock_bh(&csk->callback_lock);
2525 		cep->csk->user_data = NULL;
2526 		cconn->cep = NULL;
2527 		write_unlock_bh(&csk->callback_lock);
2528 	}
2529 	iscsi_destroy_endpoint(ep);
2530 
2531 	if (likely(csk->state >= CTP_ESTABLISHED))
2532 		need_active_close(csk);
2533 	else
2534 		cxgbi_sock_closed(csk);
2535 
2536 	cxgbi_sock_put(csk);
2537 }
2538 EXPORT_SYMBOL_GPL(cxgbi_ep_disconnect);
2539 
2540 int cxgbi_iscsi_init(struct iscsi_transport *itp,
2541 			struct scsi_transport_template **stt)
2542 {
2543 	*stt = iscsi_register_transport(itp);
2544 	if (*stt == NULL) {
2545 		pr_err("unable to register %s transport 0x%p.\n",
2546 			itp->name, itp);
2547 		return -ENODEV;
2548 	}
2549 	log_debug(1 << CXGBI_DBG_ISCSI,
2550 		"%s, registered iscsi transport 0x%p.\n",
2551 		itp->name, stt);
2552 	return 0;
2553 }
2554 EXPORT_SYMBOL_GPL(cxgbi_iscsi_init);
2555 
2556 void cxgbi_iscsi_cleanup(struct iscsi_transport *itp,
2557 			struct scsi_transport_template **stt)
2558 {
2559 	if (*stt) {
2560 		log_debug(1 << CXGBI_DBG_ISCSI,
2561 			"de-register transport 0x%p, %s, stt 0x%p.\n",
2562 			itp, itp->name, *stt);
2563 		*stt = NULL;
2564 		iscsi_unregister_transport(itp);
2565 	}
2566 }
2567 EXPORT_SYMBOL_GPL(cxgbi_iscsi_cleanup);
2568 
2569 umode_t cxgbi_attr_is_visible(int param_type, int param)
2570 {
2571 	switch (param_type) {
2572 	case ISCSI_HOST_PARAM:
2573 		switch (param) {
2574 		case ISCSI_HOST_PARAM_NETDEV_NAME:
2575 		case ISCSI_HOST_PARAM_HWADDRESS:
2576 		case ISCSI_HOST_PARAM_IPADDRESS:
2577 		case ISCSI_HOST_PARAM_INITIATOR_NAME:
2578 			return S_IRUGO;
2579 		default:
2580 			return 0;
2581 		}
2582 	case ISCSI_PARAM:
2583 		switch (param) {
2584 		case ISCSI_PARAM_MAX_RECV_DLENGTH:
2585 		case ISCSI_PARAM_MAX_XMIT_DLENGTH:
2586 		case ISCSI_PARAM_HDRDGST_EN:
2587 		case ISCSI_PARAM_DATADGST_EN:
2588 		case ISCSI_PARAM_CONN_ADDRESS:
2589 		case ISCSI_PARAM_CONN_PORT:
2590 		case ISCSI_PARAM_EXP_STATSN:
2591 		case ISCSI_PARAM_PERSISTENT_ADDRESS:
2592 		case ISCSI_PARAM_PERSISTENT_PORT:
2593 		case ISCSI_PARAM_PING_TMO:
2594 		case ISCSI_PARAM_RECV_TMO:
2595 		case ISCSI_PARAM_INITIAL_R2T_EN:
2596 		case ISCSI_PARAM_MAX_R2T:
2597 		case ISCSI_PARAM_IMM_DATA_EN:
2598 		case ISCSI_PARAM_FIRST_BURST:
2599 		case ISCSI_PARAM_MAX_BURST:
2600 		case ISCSI_PARAM_PDU_INORDER_EN:
2601 		case ISCSI_PARAM_DATASEQ_INORDER_EN:
2602 		case ISCSI_PARAM_ERL:
2603 		case ISCSI_PARAM_TARGET_NAME:
2604 		case ISCSI_PARAM_TPGT:
2605 		case ISCSI_PARAM_USERNAME:
2606 		case ISCSI_PARAM_PASSWORD:
2607 		case ISCSI_PARAM_USERNAME_IN:
2608 		case ISCSI_PARAM_PASSWORD_IN:
2609 		case ISCSI_PARAM_FAST_ABORT:
2610 		case ISCSI_PARAM_ABORT_TMO:
2611 		case ISCSI_PARAM_LU_RESET_TMO:
2612 		case ISCSI_PARAM_TGT_RESET_TMO:
2613 		case ISCSI_PARAM_IFACE_NAME:
2614 		case ISCSI_PARAM_INITIATOR_NAME:
2615 			return S_IRUGO;
2616 		default:
2617 			return 0;
2618 		}
2619 	}
2620 
2621 	return 0;
2622 }
2623 EXPORT_SYMBOL_GPL(cxgbi_attr_is_visible);
2624 
2625 static int __init libcxgbi_init_module(void)
2626 {
2627 	sw_tag_idx_bits = (__ilog2_u32(ISCSI_ITT_MASK)) + 1;
2628 	sw_tag_age_bits = (__ilog2_u32(ISCSI_AGE_MASK)) + 1;
2629 
2630 	pr_info("tag itt 0x%x, %u bits, age 0x%x, %u bits.\n",
2631 		ISCSI_ITT_MASK, sw_tag_idx_bits,
2632 		ISCSI_AGE_MASK, sw_tag_age_bits);
2633 
2634 	ddp_setup_host_page_size();
2635 	return 0;
2636 }
2637 
2638 static void __exit libcxgbi_exit_module(void)
2639 {
2640 	cxgbi_device_unregister_all(0xFF);
2641 	return;
2642 }
2643 
2644 module_init(libcxgbi_init_module);
2645 module_exit(libcxgbi_exit_module);
2646