xref: /linux/drivers/scsi/cxgbi/libcxgbi.c (revision f2ee442115c9b6219083c019939a9cc0c9abb2f8)
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 flowi4 fl4;
476 	struct cxgbi_sock *csk = NULL;
477 	unsigned int mtu = 0;
478 	int port = 0xFFFF;
479 	int err = 0;
480 
481 	if (daddr->sin_family != AF_INET) {
482 		pr_info("address family 0x%x NOT supported.\n",
483 			daddr->sin_family);
484 		err = -EAFNOSUPPORT;
485 		goto err_out;
486 	}
487 
488 	rt = find_route_ipv4(&fl4, 0, daddr->sin_addr.s_addr, 0, daddr->sin_port, 0);
489 	if (!rt) {
490 		pr_info("no route to ipv4 0x%x, port %u.\n",
491 			daddr->sin_addr.s_addr, daddr->sin_port);
492 		err = -ENETUNREACH;
493 		goto err_out;
494 	}
495 	dst = &rt->dst;
496 	ndev = dst_get_neighbour(dst)->dev;
497 
498 	if (rt->rt_flags & (RTCF_MULTICAST | RTCF_BROADCAST)) {
499 		pr_info("multi-cast route %pI4, port %u, dev %s.\n",
500 			&daddr->sin_addr.s_addr, ntohs(daddr->sin_port),
501 			ndev->name);
502 		err = -ENETUNREACH;
503 		goto rel_rt;
504 	}
505 
506 	if (ndev->flags & IFF_LOOPBACK) {
507 		ndev = ip_dev_find(&init_net, daddr->sin_addr.s_addr);
508 		mtu = ndev->mtu;
509 		pr_info("rt dev %s, loopback -> %s, mtu %u.\n",
510 			dst_get_neighbour(dst)->dev->name, ndev->name, mtu);
511 	}
512 
513 	cdev = cxgbi_device_find_by_netdev(ndev, &port);
514 	if (!cdev) {
515 		pr_info("dst %pI4, %s, NOT cxgbi device.\n",
516 			&daddr->sin_addr.s_addr, ndev->name);
517 		err = -ENETUNREACH;
518 		goto rel_rt;
519 	}
520 	log_debug(1 << CXGBI_DBG_SOCK,
521 		"route to %pI4 :%u, ndev p#%d,%s, cdev 0x%p.\n",
522 		&daddr->sin_addr.s_addr, ntohs(daddr->sin_port),
523 			   port, ndev->name, cdev);
524 
525 	csk = cxgbi_sock_create(cdev);
526 	if (!csk) {
527 		err = -ENOMEM;
528 		goto rel_rt;
529 	}
530 	csk->cdev = cdev;
531 	csk->port_id = port;
532 	csk->mtu = mtu;
533 	csk->dst = dst;
534 	csk->daddr.sin_addr.s_addr = daddr->sin_addr.s_addr;
535 	csk->daddr.sin_port = daddr->sin_port;
536 	csk->daddr.sin_family = daddr->sin_family;
537 	csk->saddr.sin_addr.s_addr = fl4.saddr;
538 
539 	return csk;
540 
541 rel_rt:
542 	ip_rt_put(rt);
543 	if (csk)
544 		cxgbi_sock_closed(csk);
545 err_out:
546 	return ERR_PTR(err);
547 }
548 
549 void cxgbi_sock_established(struct cxgbi_sock *csk, unsigned int snd_isn,
550 			unsigned int opt)
551 {
552 	csk->write_seq = csk->snd_nxt = csk->snd_una = snd_isn;
553 	dst_confirm(csk->dst);
554 	smp_mb();
555 	cxgbi_sock_set_state(csk, CTP_ESTABLISHED);
556 }
557 EXPORT_SYMBOL_GPL(cxgbi_sock_established);
558 
559 static void cxgbi_inform_iscsi_conn_closing(struct cxgbi_sock *csk)
560 {
561 	log_debug(1 << CXGBI_DBG_SOCK,
562 		"csk 0x%p, state %u, flags 0x%lx, conn 0x%p.\n",
563 		csk, csk->state, csk->flags, csk->user_data);
564 
565 	if (csk->state != CTP_ESTABLISHED) {
566 		read_lock_bh(&csk->callback_lock);
567 		if (csk->user_data)
568 			iscsi_conn_failure(csk->user_data,
569 					ISCSI_ERR_CONN_FAILED);
570 		read_unlock_bh(&csk->callback_lock);
571 	}
572 }
573 
574 void cxgbi_sock_closed(struct cxgbi_sock *csk)
575 {
576 	log_debug(1 << CXGBI_DBG_SOCK, "csk 0x%p,%u,0x%lx,%u.\n",
577 		csk, (csk)->state, (csk)->flags, (csk)->tid);
578 	cxgbi_sock_set_flag(csk, CTPF_ACTIVE_CLOSE_NEEDED);
579 	if (csk->state == CTP_ACTIVE_OPEN || csk->state == CTP_CLOSED)
580 		return;
581 	if (csk->saddr.sin_port)
582 		sock_put_port(csk);
583 	if (csk->dst)
584 		dst_release(csk->dst);
585 	csk->cdev->csk_release_offload_resources(csk);
586 	cxgbi_sock_set_state(csk, CTP_CLOSED);
587 	cxgbi_inform_iscsi_conn_closing(csk);
588 	cxgbi_sock_put(csk);
589 }
590 EXPORT_SYMBOL_GPL(cxgbi_sock_closed);
591 
592 static void need_active_close(struct cxgbi_sock *csk)
593 {
594 	int data_lost;
595 	int close_req = 0;
596 
597 	log_debug(1 << CXGBI_DBG_SOCK, "csk 0x%p,%u,0x%lx,%u.\n",
598 		csk, (csk)->state, (csk)->flags, (csk)->tid);
599 	spin_lock_bh(&csk->lock);
600 	dst_confirm(csk->dst);
601 	data_lost = skb_queue_len(&csk->receive_queue);
602 	__skb_queue_purge(&csk->receive_queue);
603 
604 	if (csk->state == CTP_ACTIVE_OPEN)
605 		cxgbi_sock_set_flag(csk, CTPF_ACTIVE_CLOSE_NEEDED);
606 	else if (csk->state == CTP_ESTABLISHED) {
607 		close_req = 1;
608 		cxgbi_sock_set_state(csk, CTP_ACTIVE_CLOSE);
609 	} else if (csk->state == CTP_PASSIVE_CLOSE) {
610 		close_req = 1;
611 		cxgbi_sock_set_state(csk, CTP_CLOSE_WAIT_2);
612 	}
613 
614 	if (close_req) {
615 		if (data_lost)
616 			csk->cdev->csk_send_abort_req(csk);
617 		else
618 			csk->cdev->csk_send_close_req(csk);
619 	}
620 
621 	spin_unlock_bh(&csk->lock);
622 }
623 
624 void cxgbi_sock_fail_act_open(struct cxgbi_sock *csk, int errno)
625 {
626 	pr_info("csk 0x%p,%u,%lx, %pI4:%u-%pI4:%u, err %d.\n",
627 			csk, csk->state, csk->flags,
628 			&csk->saddr.sin_addr.s_addr, csk->saddr.sin_port,
629 			&csk->daddr.sin_addr.s_addr, csk->daddr.sin_port,
630 			errno);
631 
632 	cxgbi_sock_set_state(csk, CTP_CONNECTING);
633 	csk->err = errno;
634 	cxgbi_sock_closed(csk);
635 }
636 EXPORT_SYMBOL_GPL(cxgbi_sock_fail_act_open);
637 
638 void cxgbi_sock_act_open_req_arp_failure(void *handle, struct sk_buff *skb)
639 {
640 	struct cxgbi_sock *csk = (struct cxgbi_sock *)skb->sk;
641 
642 	log_debug(1 << CXGBI_DBG_SOCK, "csk 0x%p,%u,0x%lx,%u.\n",
643 		csk, (csk)->state, (csk)->flags, (csk)->tid);
644 	cxgbi_sock_get(csk);
645 	spin_lock_bh(&csk->lock);
646 	if (csk->state == CTP_ACTIVE_OPEN)
647 		cxgbi_sock_fail_act_open(csk, -EHOSTUNREACH);
648 	spin_unlock_bh(&csk->lock);
649 	cxgbi_sock_put(csk);
650 	__kfree_skb(skb);
651 }
652 EXPORT_SYMBOL_GPL(cxgbi_sock_act_open_req_arp_failure);
653 
654 void cxgbi_sock_rcv_abort_rpl(struct cxgbi_sock *csk)
655 {
656 	cxgbi_sock_get(csk);
657 	spin_lock_bh(&csk->lock);
658 	if (cxgbi_sock_flag(csk, CTPF_ABORT_RPL_PENDING)) {
659 		if (!cxgbi_sock_flag(csk, CTPF_ABORT_RPL_RCVD))
660 			cxgbi_sock_set_flag(csk, CTPF_ABORT_RPL_RCVD);
661 		else {
662 			cxgbi_sock_clear_flag(csk, CTPF_ABORT_RPL_RCVD);
663 			cxgbi_sock_clear_flag(csk, CTPF_ABORT_RPL_PENDING);
664 			if (cxgbi_sock_flag(csk, CTPF_ABORT_REQ_RCVD))
665 				pr_err("csk 0x%p,%u,0x%lx,%u,ABT_RPL_RSS.\n",
666 					csk, csk->state, csk->flags, csk->tid);
667 			cxgbi_sock_closed(csk);
668 		}
669 	}
670 	spin_unlock_bh(&csk->lock);
671 	cxgbi_sock_put(csk);
672 }
673 EXPORT_SYMBOL_GPL(cxgbi_sock_rcv_abort_rpl);
674 
675 void cxgbi_sock_rcv_peer_close(struct cxgbi_sock *csk)
676 {
677 	log_debug(1 << CXGBI_DBG_SOCK, "csk 0x%p,%u,0x%lx,%u.\n",
678 		csk, (csk)->state, (csk)->flags, (csk)->tid);
679 	cxgbi_sock_get(csk);
680 	spin_lock_bh(&csk->lock);
681 
682 	if (cxgbi_sock_flag(csk, CTPF_ABORT_RPL_PENDING))
683 		goto done;
684 
685 	switch (csk->state) {
686 	case CTP_ESTABLISHED:
687 		cxgbi_sock_set_state(csk, CTP_PASSIVE_CLOSE);
688 		break;
689 	case CTP_ACTIVE_CLOSE:
690 		cxgbi_sock_set_state(csk, CTP_CLOSE_WAIT_2);
691 		break;
692 	case CTP_CLOSE_WAIT_1:
693 		cxgbi_sock_closed(csk);
694 		break;
695 	case CTP_ABORTING:
696 		break;
697 	default:
698 		pr_err("csk 0x%p,%u,0x%lx,%u, bad state.\n",
699 			csk, csk->state, csk->flags, csk->tid);
700 	}
701 	cxgbi_inform_iscsi_conn_closing(csk);
702 done:
703 	spin_unlock_bh(&csk->lock);
704 	cxgbi_sock_put(csk);
705 }
706 EXPORT_SYMBOL_GPL(cxgbi_sock_rcv_peer_close);
707 
708 void cxgbi_sock_rcv_close_conn_rpl(struct cxgbi_sock *csk, u32 snd_nxt)
709 {
710 	log_debug(1 << CXGBI_DBG_SOCK, "csk 0x%p,%u,0x%lx,%u.\n",
711 		csk, (csk)->state, (csk)->flags, (csk)->tid);
712 	cxgbi_sock_get(csk);
713 	spin_lock_bh(&csk->lock);
714 
715 	csk->snd_una = snd_nxt - 1;
716 	if (cxgbi_sock_flag(csk, CTPF_ABORT_RPL_PENDING))
717 		goto done;
718 
719 	switch (csk->state) {
720 	case CTP_ACTIVE_CLOSE:
721 		cxgbi_sock_set_state(csk, CTP_CLOSE_WAIT_1);
722 		break;
723 	case CTP_CLOSE_WAIT_1:
724 	case CTP_CLOSE_WAIT_2:
725 		cxgbi_sock_closed(csk);
726 		break;
727 	case CTP_ABORTING:
728 		break;
729 	default:
730 		pr_err("csk 0x%p,%u,0x%lx,%u, bad state.\n",
731 			csk, csk->state, csk->flags, csk->tid);
732 	}
733 done:
734 	spin_unlock_bh(&csk->lock);
735 	cxgbi_sock_put(csk);
736 }
737 EXPORT_SYMBOL_GPL(cxgbi_sock_rcv_close_conn_rpl);
738 
739 void cxgbi_sock_rcv_wr_ack(struct cxgbi_sock *csk, unsigned int credits,
740 			   unsigned int snd_una, int seq_chk)
741 {
742 	log_debug(1 << CXGBI_DBG_TOE | 1 << CXGBI_DBG_SOCK,
743 			"csk 0x%p,%u,0x%lx,%u, cr %u,%u+%u, snd_una %u,%d.\n",
744 			csk, csk->state, csk->flags, csk->tid, credits,
745 			csk->wr_cred, csk->wr_una_cred, snd_una, seq_chk);
746 
747 	spin_lock_bh(&csk->lock);
748 
749 	csk->wr_cred += credits;
750 	if (csk->wr_una_cred > csk->wr_max_cred - csk->wr_cred)
751 		csk->wr_una_cred = csk->wr_max_cred - csk->wr_cred;
752 
753 	while (credits) {
754 		struct sk_buff *p = cxgbi_sock_peek_wr(csk);
755 
756 		if (unlikely(!p)) {
757 			pr_err("csk 0x%p,%u,0x%lx,%u, cr %u,%u+%u, empty.\n",
758 				csk, csk->state, csk->flags, csk->tid, credits,
759 				csk->wr_cred, csk->wr_una_cred);
760 			break;
761 		}
762 
763 		if (unlikely(credits < p->csum)) {
764 			pr_warn("csk 0x%p,%u,0x%lx,%u, cr %u,%u+%u, < %u.\n",
765 				csk, csk->state, csk->flags, csk->tid,
766 				credits, csk->wr_cred, csk->wr_una_cred,
767 				p->csum);
768 			p->csum -= credits;
769 			break;
770 		} else {
771 			cxgbi_sock_dequeue_wr(csk);
772 			credits -= p->csum;
773 			kfree_skb(p);
774 		}
775 	}
776 
777 	cxgbi_sock_check_wr_invariants(csk);
778 
779 	if (seq_chk) {
780 		if (unlikely(before(snd_una, csk->snd_una))) {
781 			pr_warn("csk 0x%p,%u,0x%lx,%u, snd_una %u/%u.",
782 				csk, csk->state, csk->flags, csk->tid, snd_una,
783 				csk->snd_una);
784 			goto done;
785 		}
786 
787 		if (csk->snd_una != snd_una) {
788 			csk->snd_una = snd_una;
789 			dst_confirm(csk->dst);
790 		}
791 	}
792 
793 	if (skb_queue_len(&csk->write_queue)) {
794 		if (csk->cdev->csk_push_tx_frames(csk, 0))
795 			cxgbi_conn_tx_open(csk);
796 	} else
797 		cxgbi_conn_tx_open(csk);
798 done:
799 	spin_unlock_bh(&csk->lock);
800 }
801 EXPORT_SYMBOL_GPL(cxgbi_sock_rcv_wr_ack);
802 
803 static unsigned int cxgbi_sock_find_best_mtu(struct cxgbi_sock *csk,
804 					     unsigned short mtu)
805 {
806 	int i = 0;
807 
808 	while (i < csk->cdev->nmtus - 1 && csk->cdev->mtus[i + 1] <= mtu)
809 		++i;
810 
811 	return i;
812 }
813 
814 unsigned int cxgbi_sock_select_mss(struct cxgbi_sock *csk, unsigned int pmtu)
815 {
816 	unsigned int idx;
817 	struct dst_entry *dst = csk->dst;
818 
819 	csk->advmss = dst_metric_advmss(dst);
820 
821 	if (csk->advmss > pmtu - 40)
822 		csk->advmss = pmtu - 40;
823 	if (csk->advmss < csk->cdev->mtus[0] - 40)
824 		csk->advmss = csk->cdev->mtus[0] - 40;
825 	idx = cxgbi_sock_find_best_mtu(csk, csk->advmss + 40);
826 
827 	return idx;
828 }
829 EXPORT_SYMBOL_GPL(cxgbi_sock_select_mss);
830 
831 void cxgbi_sock_skb_entail(struct cxgbi_sock *csk, struct sk_buff *skb)
832 {
833 	cxgbi_skcb_tcp_seq(skb) = csk->write_seq;
834 	__skb_queue_tail(&csk->write_queue, skb);
835 }
836 EXPORT_SYMBOL_GPL(cxgbi_sock_skb_entail);
837 
838 void cxgbi_sock_purge_wr_queue(struct cxgbi_sock *csk)
839 {
840 	struct sk_buff *skb;
841 
842 	while ((skb = cxgbi_sock_dequeue_wr(csk)) != NULL)
843 		kfree_skb(skb);
844 }
845 EXPORT_SYMBOL_GPL(cxgbi_sock_purge_wr_queue);
846 
847 void cxgbi_sock_check_wr_invariants(const struct cxgbi_sock *csk)
848 {
849 	int pending = cxgbi_sock_count_pending_wrs(csk);
850 
851 	if (unlikely(csk->wr_cred + pending != csk->wr_max_cred))
852 		pr_err("csk 0x%p, tid %u, credit %u + %u != %u.\n",
853 			csk, csk->tid, csk->wr_cred, pending, csk->wr_max_cred);
854 }
855 EXPORT_SYMBOL_GPL(cxgbi_sock_check_wr_invariants);
856 
857 static int cxgbi_sock_send_pdus(struct cxgbi_sock *csk, struct sk_buff *skb)
858 {
859 	struct cxgbi_device *cdev = csk->cdev;
860 	struct sk_buff *next;
861 	int err, copied = 0;
862 
863 	spin_lock_bh(&csk->lock);
864 
865 	if (csk->state != CTP_ESTABLISHED) {
866 		log_debug(1 << CXGBI_DBG_PDU_TX,
867 			"csk 0x%p,%u,0x%lx,%u, EAGAIN.\n",
868 			csk, csk->state, csk->flags, csk->tid);
869 		err = -EAGAIN;
870 		goto out_err;
871 	}
872 
873 	if (csk->err) {
874 		log_debug(1 << CXGBI_DBG_PDU_TX,
875 			"csk 0x%p,%u,0x%lx,%u, EPIPE %d.\n",
876 			csk, csk->state, csk->flags, csk->tid, csk->err);
877 		err = -EPIPE;
878 		goto out_err;
879 	}
880 
881 	if (csk->write_seq - csk->snd_una >= cdev->snd_win) {
882 		log_debug(1 << CXGBI_DBG_PDU_TX,
883 			"csk 0x%p,%u,0x%lx,%u, FULL %u-%u >= %u.\n",
884 			csk, csk->state, csk->flags, csk->tid, csk->write_seq,
885 			csk->snd_una, cdev->snd_win);
886 		err = -ENOBUFS;
887 		goto out_err;
888 	}
889 
890 	while (skb) {
891 		int frags = skb_shinfo(skb)->nr_frags +
892 				(skb->len != skb->data_len);
893 
894 		if (unlikely(skb_headroom(skb) < cdev->skb_tx_rsvd)) {
895 			pr_err("csk 0x%p, skb head %u < %u.\n",
896 				csk, skb_headroom(skb), cdev->skb_tx_rsvd);
897 			err = -EINVAL;
898 			goto out_err;
899 		}
900 
901 		if (frags >= SKB_WR_LIST_SIZE) {
902 			pr_err("csk 0x%p, frags %d, %u,%u >%u.\n",
903 				csk, skb_shinfo(skb)->nr_frags, skb->len,
904 				skb->data_len, (uint)(SKB_WR_LIST_SIZE));
905 			err = -EINVAL;
906 			goto out_err;
907 		}
908 
909 		next = skb->next;
910 		skb->next = NULL;
911 		cxgbi_skcb_set_flag(skb, SKCBF_TX_NEED_HDR);
912 		cxgbi_sock_skb_entail(csk, skb);
913 		copied += skb->len;
914 		csk->write_seq += skb->len +
915 				cxgbi_ulp_extra_len(cxgbi_skcb_ulp_mode(skb));
916 		skb = next;
917 	}
918 done:
919 	if (likely(skb_queue_len(&csk->write_queue)))
920 		cdev->csk_push_tx_frames(csk, 1);
921 	spin_unlock_bh(&csk->lock);
922 	return copied;
923 
924 out_err:
925 	if (copied == 0 && err == -EPIPE)
926 		copied = csk->err ? csk->err : -EPIPE;
927 	else
928 		copied = err;
929 	goto done;
930 }
931 
932 /*
933  * Direct Data Placement -
934  * Directly place the iSCSI Data-In or Data-Out PDU's payload into pre-posted
935  * final destination host-memory buffers based on the Initiator Task Tag (ITT)
936  * in Data-In or Target Task Tag (TTT) in Data-Out PDUs.
937  * The host memory address is programmed into h/w in the format of pagepod
938  * entries.
939  * The location of the pagepod entry is encoded into ddp tag which is used as
940  * the base for ITT/TTT.
941  */
942 
943 static unsigned char ddp_page_order[DDP_PGIDX_MAX] = {0, 1, 2, 4};
944 static unsigned char ddp_page_shift[DDP_PGIDX_MAX] = {12, 13, 14, 16};
945 static unsigned char page_idx = DDP_PGIDX_MAX;
946 
947 static unsigned char sw_tag_idx_bits;
948 static unsigned char sw_tag_age_bits;
949 
950 /*
951  * Direct-Data Placement page size adjustment
952  */
953 static int ddp_adjust_page_table(void)
954 {
955 	int i;
956 	unsigned int base_order, order;
957 
958 	if (PAGE_SIZE < (1UL << ddp_page_shift[0])) {
959 		pr_info("PAGE_SIZE 0x%lx too small, min 0x%lx\n",
960 			PAGE_SIZE, 1UL << ddp_page_shift[0]);
961 		return -EINVAL;
962 	}
963 
964 	base_order = get_order(1UL << ddp_page_shift[0]);
965 	order = get_order(1UL << PAGE_SHIFT);
966 
967 	for (i = 0; i < DDP_PGIDX_MAX; i++) {
968 		/* first is the kernel page size, then just doubling */
969 		ddp_page_order[i] = order - base_order + i;
970 		ddp_page_shift[i] = PAGE_SHIFT + i;
971 	}
972 	return 0;
973 }
974 
975 static int ddp_find_page_index(unsigned long pgsz)
976 {
977 	int i;
978 
979 	for (i = 0; i < DDP_PGIDX_MAX; i++) {
980 		if (pgsz == (1UL << ddp_page_shift[i]))
981 			return i;
982 	}
983 	pr_info("ddp page size %lu not supported.\n", pgsz);
984 	return DDP_PGIDX_MAX;
985 }
986 
987 static void ddp_setup_host_page_size(void)
988 {
989 	if (page_idx == DDP_PGIDX_MAX) {
990 		page_idx = ddp_find_page_index(PAGE_SIZE);
991 
992 		if (page_idx == DDP_PGIDX_MAX) {
993 			pr_info("system PAGE %lu, update hw.\n", PAGE_SIZE);
994 			if (ddp_adjust_page_table() < 0) {
995 				pr_info("PAGE %lu, disable ddp.\n", PAGE_SIZE);
996 				return;
997 			}
998 			page_idx = ddp_find_page_index(PAGE_SIZE);
999 		}
1000 		pr_info("system PAGE %lu, ddp idx %u.\n", PAGE_SIZE, page_idx);
1001 	}
1002 }
1003 
1004 void cxgbi_ddp_page_size_factor(int *pgsz_factor)
1005 {
1006 	int i;
1007 
1008 	for (i = 0; i < DDP_PGIDX_MAX; i++)
1009 		pgsz_factor[i] = ddp_page_order[i];
1010 }
1011 EXPORT_SYMBOL_GPL(cxgbi_ddp_page_size_factor);
1012 
1013 /*
1014  * DDP setup & teardown
1015  */
1016 
1017 void cxgbi_ddp_ppod_set(struct cxgbi_pagepod *ppod,
1018 			struct cxgbi_pagepod_hdr *hdr,
1019 			struct cxgbi_gather_list *gl, unsigned int gidx)
1020 {
1021 	int i;
1022 
1023 	memcpy(ppod, hdr, sizeof(*hdr));
1024 	for (i = 0; i < (PPOD_PAGES_MAX + 1); i++, gidx++) {
1025 		ppod->addr[i] = gidx < gl->nelem ?
1026 				cpu_to_be64(gl->phys_addr[gidx]) : 0ULL;
1027 	}
1028 }
1029 EXPORT_SYMBOL_GPL(cxgbi_ddp_ppod_set);
1030 
1031 void cxgbi_ddp_ppod_clear(struct cxgbi_pagepod *ppod)
1032 {
1033 	memset(ppod, 0, sizeof(*ppod));
1034 }
1035 EXPORT_SYMBOL_GPL(cxgbi_ddp_ppod_clear);
1036 
1037 static inline int ddp_find_unused_entries(struct cxgbi_ddp_info *ddp,
1038 					unsigned int start, unsigned int max,
1039 					unsigned int count,
1040 					struct cxgbi_gather_list *gl)
1041 {
1042 	unsigned int i, j, k;
1043 
1044 	/*  not enough entries */
1045 	if ((max - start) < count) {
1046 		log_debug(1 << CXGBI_DBG_DDP,
1047 			"NOT enough entries %u+%u < %u.\n", start, count, max);
1048 		return -EBUSY;
1049 	}
1050 
1051 	max -= count;
1052 	spin_lock(&ddp->map_lock);
1053 	for (i = start; i < max;) {
1054 		for (j = 0, k = i; j < count; j++, k++) {
1055 			if (ddp->gl_map[k])
1056 				break;
1057 		}
1058 		if (j == count) {
1059 			for (j = 0, k = i; j < count; j++, k++)
1060 				ddp->gl_map[k] = gl;
1061 			spin_unlock(&ddp->map_lock);
1062 			return i;
1063 		}
1064 		i += j + 1;
1065 	}
1066 	spin_unlock(&ddp->map_lock);
1067 	log_debug(1 << CXGBI_DBG_DDP,
1068 		"NO suitable entries %u available.\n", count);
1069 	return -EBUSY;
1070 }
1071 
1072 static inline void ddp_unmark_entries(struct cxgbi_ddp_info *ddp,
1073 						int start, int count)
1074 {
1075 	spin_lock(&ddp->map_lock);
1076 	memset(&ddp->gl_map[start], 0,
1077 		count * sizeof(struct cxgbi_gather_list *));
1078 	spin_unlock(&ddp->map_lock);
1079 }
1080 
1081 static inline void ddp_gl_unmap(struct pci_dev *pdev,
1082 					struct cxgbi_gather_list *gl)
1083 {
1084 	int i;
1085 
1086 	for (i = 0; i < gl->nelem; i++)
1087 		dma_unmap_page(&pdev->dev, gl->phys_addr[i], PAGE_SIZE,
1088 				PCI_DMA_FROMDEVICE);
1089 }
1090 
1091 static inline int ddp_gl_map(struct pci_dev *pdev,
1092 				    struct cxgbi_gather_list *gl)
1093 {
1094 	int i;
1095 
1096 	for (i = 0; i < gl->nelem; i++) {
1097 		gl->phys_addr[i] = dma_map_page(&pdev->dev, gl->pages[i], 0,
1098 						PAGE_SIZE,
1099 						PCI_DMA_FROMDEVICE);
1100 		if (unlikely(dma_mapping_error(&pdev->dev, gl->phys_addr[i]))) {
1101 			log_debug(1 << CXGBI_DBG_DDP,
1102 				"page %d 0x%p, 0x%p dma mapping err.\n",
1103 				i, gl->pages[i], pdev);
1104 			goto unmap;
1105 		}
1106 	}
1107 	return i;
1108 unmap:
1109 	if (i) {
1110 		unsigned int nelem = gl->nelem;
1111 
1112 		gl->nelem = i;
1113 		ddp_gl_unmap(pdev, gl);
1114 		gl->nelem = nelem;
1115 	}
1116 	return -EINVAL;
1117 }
1118 
1119 static void ddp_release_gl(struct cxgbi_gather_list *gl,
1120 				  struct pci_dev *pdev)
1121 {
1122 	ddp_gl_unmap(pdev, gl);
1123 	kfree(gl);
1124 }
1125 
1126 static struct cxgbi_gather_list *ddp_make_gl(unsigned int xferlen,
1127 						    struct scatterlist *sgl,
1128 						    unsigned int sgcnt,
1129 						    struct pci_dev *pdev,
1130 						    gfp_t gfp)
1131 {
1132 	struct cxgbi_gather_list *gl;
1133 	struct scatterlist *sg = sgl;
1134 	struct page *sgpage = sg_page(sg);
1135 	unsigned int sglen = sg->length;
1136 	unsigned int sgoffset = sg->offset;
1137 	unsigned int npages = (xferlen + sgoffset + PAGE_SIZE - 1) >>
1138 				PAGE_SHIFT;
1139 	int i = 1, j = 0;
1140 
1141 	if (xferlen < DDP_THRESHOLD) {
1142 		log_debug(1 << CXGBI_DBG_DDP,
1143 			"xfer %u < threshold %u, no ddp.\n",
1144 			xferlen, DDP_THRESHOLD);
1145 		return NULL;
1146 	}
1147 
1148 	gl = kzalloc(sizeof(struct cxgbi_gather_list) +
1149 		     npages * (sizeof(dma_addr_t) +
1150 		     sizeof(struct page *)), gfp);
1151 	if (!gl) {
1152 		log_debug(1 << CXGBI_DBG_DDP,
1153 			"xfer %u, %u pages, OOM.\n", xferlen, npages);
1154 		return NULL;
1155 	}
1156 
1157 	 log_debug(1 << CXGBI_DBG_DDP,
1158 		"xfer %u, sgl %u, gl max %u.\n", xferlen, sgcnt, npages);
1159 
1160 	gl->pages = (struct page **)&gl->phys_addr[npages];
1161 	gl->nelem = npages;
1162 	gl->length = xferlen;
1163 	gl->offset = sgoffset;
1164 	gl->pages[0] = sgpage;
1165 
1166 	for (i = 1, sg = sg_next(sgl), j = 0; i < sgcnt;
1167 		i++, sg = sg_next(sg)) {
1168 		struct page *page = sg_page(sg);
1169 
1170 		if (sgpage == page && sg->offset == sgoffset + sglen)
1171 			sglen += sg->length;
1172 		else {
1173 			/*  make sure the sgl is fit for ddp:
1174 			 *  each has the same page size, and
1175 			 *  all of the middle pages are used completely
1176 			 */
1177 			if ((j && sgoffset) || ((i != sgcnt - 1) &&
1178 			    ((sglen + sgoffset) & ~PAGE_MASK))) {
1179 				log_debug(1 << CXGBI_DBG_DDP,
1180 					"page %d/%u, %u + %u.\n",
1181 					i, sgcnt, sgoffset, sglen);
1182 				goto error_out;
1183 			}
1184 
1185 			j++;
1186 			if (j == gl->nelem || sg->offset) {
1187 				log_debug(1 << CXGBI_DBG_DDP,
1188 					"page %d/%u, offset %u.\n",
1189 					j, gl->nelem, sg->offset);
1190 				goto error_out;
1191 			}
1192 			gl->pages[j] = page;
1193 			sglen = sg->length;
1194 			sgoffset = sg->offset;
1195 			sgpage = page;
1196 		}
1197 	}
1198 	gl->nelem = ++j;
1199 
1200 	if (ddp_gl_map(pdev, gl) < 0)
1201 		goto error_out;
1202 
1203 	return gl;
1204 
1205 error_out:
1206 	kfree(gl);
1207 	return NULL;
1208 }
1209 
1210 static void ddp_tag_release(struct cxgbi_hba *chba, u32 tag)
1211 {
1212 	struct cxgbi_device *cdev = chba->cdev;
1213 	struct cxgbi_ddp_info *ddp = cdev->ddp;
1214 	u32 idx;
1215 
1216 	idx = (tag >> PPOD_IDX_SHIFT) & ddp->idx_mask;
1217 	if (idx < ddp->nppods) {
1218 		struct cxgbi_gather_list *gl = ddp->gl_map[idx];
1219 		unsigned int npods;
1220 
1221 		if (!gl || !gl->nelem) {
1222 			pr_warn("tag 0x%x, idx %u, gl 0x%p, %u.\n",
1223 				tag, idx, gl, gl ? gl->nelem : 0);
1224 			return;
1225 		}
1226 		npods = (gl->nelem + PPOD_PAGES_MAX - 1) >> PPOD_PAGES_SHIFT;
1227 		log_debug(1 << CXGBI_DBG_DDP,
1228 			"tag 0x%x, release idx %u, npods %u.\n",
1229 			tag, idx, npods);
1230 		cdev->csk_ddp_clear(chba, tag, idx, npods);
1231 		ddp_unmark_entries(ddp, idx, npods);
1232 		ddp_release_gl(gl, ddp->pdev);
1233 	} else
1234 		pr_warn("tag 0x%x, idx %u > max %u.\n", tag, idx, ddp->nppods);
1235 }
1236 
1237 static int ddp_tag_reserve(struct cxgbi_sock *csk, unsigned int tid,
1238 			   u32 sw_tag, u32 *tagp, struct cxgbi_gather_list *gl,
1239 			   gfp_t gfp)
1240 {
1241 	struct cxgbi_device *cdev = csk->cdev;
1242 	struct cxgbi_ddp_info *ddp = cdev->ddp;
1243 	struct cxgbi_tag_format *tformat = &cdev->tag_format;
1244 	struct cxgbi_pagepod_hdr hdr;
1245 	unsigned int npods;
1246 	int idx = -1;
1247 	int err = -ENOMEM;
1248 	u32 tag;
1249 
1250 	npods = (gl->nelem + PPOD_PAGES_MAX - 1) >> PPOD_PAGES_SHIFT;
1251 	if (ddp->idx_last == ddp->nppods)
1252 		idx = ddp_find_unused_entries(ddp, 0, ddp->nppods,
1253 							npods, gl);
1254 	else {
1255 		idx = ddp_find_unused_entries(ddp, ddp->idx_last + 1,
1256 							ddp->nppods, npods,
1257 							gl);
1258 		if (idx < 0 && ddp->idx_last >= npods) {
1259 			idx = ddp_find_unused_entries(ddp, 0,
1260 				min(ddp->idx_last + npods, ddp->nppods),
1261 							npods, gl);
1262 		}
1263 	}
1264 	if (idx < 0) {
1265 		log_debug(1 << CXGBI_DBG_DDP,
1266 			"xferlen %u, gl %u, npods %u NO DDP.\n",
1267 			gl->length, gl->nelem, npods);
1268 		return idx;
1269 	}
1270 
1271 	tag = cxgbi_ddp_tag_base(tformat, sw_tag);
1272 	tag |= idx << PPOD_IDX_SHIFT;
1273 
1274 	hdr.rsvd = 0;
1275 	hdr.vld_tid = htonl(PPOD_VALID_FLAG | PPOD_TID(tid));
1276 	hdr.pgsz_tag_clr = htonl(tag & ddp->rsvd_tag_mask);
1277 	hdr.max_offset = htonl(gl->length);
1278 	hdr.page_offset = htonl(gl->offset);
1279 
1280 	err = cdev->csk_ddp_set(csk, &hdr, idx, npods, gl);
1281 	if (err < 0)
1282 		goto unmark_entries;
1283 
1284 	ddp->idx_last = idx;
1285 	log_debug(1 << CXGBI_DBG_DDP,
1286 		"xfer %u, gl %u,%u, tid 0x%x, tag 0x%x->0x%x(%u,%u).\n",
1287 		gl->length, gl->nelem, gl->offset, tid, sw_tag, tag, idx,
1288 		npods);
1289 	*tagp = tag;
1290 	return 0;
1291 
1292 unmark_entries:
1293 	ddp_unmark_entries(ddp, idx, npods);
1294 	return err;
1295 }
1296 
1297 int cxgbi_ddp_reserve(struct cxgbi_sock *csk, unsigned int *tagp,
1298 			unsigned int sw_tag, unsigned int xferlen,
1299 			struct scatterlist *sgl, unsigned int sgcnt, gfp_t gfp)
1300 {
1301 	struct cxgbi_device *cdev = csk->cdev;
1302 	struct cxgbi_tag_format *tformat = &cdev->tag_format;
1303 	struct cxgbi_gather_list *gl;
1304 	int err;
1305 
1306 	if (page_idx >= DDP_PGIDX_MAX || !cdev->ddp ||
1307 	    xferlen < DDP_THRESHOLD) {
1308 		log_debug(1 << CXGBI_DBG_DDP,
1309 			"pgidx %u, xfer %u, NO ddp.\n", page_idx, xferlen);
1310 		return -EINVAL;
1311 	}
1312 
1313 	if (!cxgbi_sw_tag_usable(tformat, sw_tag)) {
1314 		log_debug(1 << CXGBI_DBG_DDP,
1315 			"sw_tag 0x%x NOT usable.\n", sw_tag);
1316 		return -EINVAL;
1317 	}
1318 
1319 	gl = ddp_make_gl(xferlen, sgl, sgcnt, cdev->pdev, gfp);
1320 	if (!gl)
1321 		return -ENOMEM;
1322 
1323 	err = ddp_tag_reserve(csk, csk->tid, sw_tag, tagp, gl, gfp);
1324 	if (err < 0)
1325 		ddp_release_gl(gl, cdev->pdev);
1326 
1327 	return err;
1328 }
1329 
1330 static void ddp_destroy(struct kref *kref)
1331 {
1332 	struct cxgbi_ddp_info *ddp = container_of(kref,
1333 						struct cxgbi_ddp_info,
1334 						refcnt);
1335 	struct cxgbi_device *cdev = ddp->cdev;
1336 	int i = 0;
1337 
1338 	pr_info("kref 0, destroy ddp 0x%p, cdev 0x%p.\n", ddp, cdev);
1339 
1340 	while (i < ddp->nppods) {
1341 		struct cxgbi_gather_list *gl = ddp->gl_map[i];
1342 
1343 		if (gl) {
1344 			int npods = (gl->nelem + PPOD_PAGES_MAX - 1)
1345 					>> PPOD_PAGES_SHIFT;
1346 			pr_info("cdev 0x%p, ddp %d + %d.\n", cdev, i, npods);
1347 			kfree(gl);
1348 			i += npods;
1349 		} else
1350 			i++;
1351 	}
1352 	cxgbi_free_big_mem(ddp);
1353 }
1354 
1355 int cxgbi_ddp_cleanup(struct cxgbi_device *cdev)
1356 {
1357 	struct cxgbi_ddp_info *ddp = cdev->ddp;
1358 
1359 	log_debug(1 << CXGBI_DBG_DDP,
1360 		"cdev 0x%p, release ddp 0x%p.\n", cdev, ddp);
1361 	cdev->ddp = NULL;
1362 	if (ddp)
1363 		return kref_put(&ddp->refcnt, ddp_destroy);
1364 	return 0;
1365 }
1366 EXPORT_SYMBOL_GPL(cxgbi_ddp_cleanup);
1367 
1368 int cxgbi_ddp_init(struct cxgbi_device *cdev,
1369 		   unsigned int llimit, unsigned int ulimit,
1370 		   unsigned int max_txsz, unsigned int max_rxsz)
1371 {
1372 	struct cxgbi_ddp_info *ddp;
1373 	unsigned int ppmax, bits;
1374 
1375 	ppmax = (ulimit - llimit + 1) >> PPOD_SIZE_SHIFT;
1376 	bits = __ilog2_u32(ppmax) + 1;
1377 	if (bits > PPOD_IDX_MAX_SIZE)
1378 		bits = PPOD_IDX_MAX_SIZE;
1379 	ppmax = (1 << (bits - 1)) - 1;
1380 
1381 	ddp = cxgbi_alloc_big_mem(sizeof(struct cxgbi_ddp_info) +
1382 				ppmax * (sizeof(struct cxgbi_gather_list *) +
1383 					 sizeof(struct sk_buff *)),
1384 				GFP_KERNEL);
1385 	if (!ddp) {
1386 		pr_warn("cdev 0x%p, ddp ppmax %u OOM.\n", cdev, ppmax);
1387 		return -ENOMEM;
1388 	}
1389 	ddp->gl_map = (struct cxgbi_gather_list **)(ddp + 1);
1390 	cdev->ddp = ddp;
1391 
1392 	spin_lock_init(&ddp->map_lock);
1393 	kref_init(&ddp->refcnt);
1394 
1395 	ddp->cdev = cdev;
1396 	ddp->pdev = cdev->pdev;
1397 	ddp->llimit = llimit;
1398 	ddp->ulimit = ulimit;
1399 	ddp->max_txsz = min_t(unsigned int, max_txsz, ULP2_MAX_PKT_SIZE);
1400 	ddp->max_rxsz = min_t(unsigned int, max_rxsz, ULP2_MAX_PKT_SIZE);
1401 	ddp->nppods = ppmax;
1402 	ddp->idx_last = ppmax;
1403 	ddp->idx_bits = bits;
1404 	ddp->idx_mask = (1 << bits) - 1;
1405 	ddp->rsvd_tag_mask = (1 << (bits + PPOD_IDX_SHIFT)) - 1;
1406 
1407 	cdev->tag_format.sw_bits = sw_tag_idx_bits + sw_tag_age_bits;
1408 	cdev->tag_format.rsvd_bits = ddp->idx_bits;
1409 	cdev->tag_format.rsvd_shift = PPOD_IDX_SHIFT;
1410 	cdev->tag_format.rsvd_mask = (1 << cdev->tag_format.rsvd_bits) - 1;
1411 
1412 	pr_info("%s tag format, sw %u, rsvd %u,%u, mask 0x%x.\n",
1413 		cdev->ports[0]->name, cdev->tag_format.sw_bits,
1414 		cdev->tag_format.rsvd_bits, cdev->tag_format.rsvd_shift,
1415 		cdev->tag_format.rsvd_mask);
1416 
1417 	cdev->tx_max_size = min_t(unsigned int, ULP2_MAX_PDU_PAYLOAD,
1418 				ddp->max_txsz - ISCSI_PDU_NONPAYLOAD_LEN);
1419 	cdev->rx_max_size = min_t(unsigned int, ULP2_MAX_PDU_PAYLOAD,
1420 				ddp->max_rxsz - ISCSI_PDU_NONPAYLOAD_LEN);
1421 
1422 	log_debug(1 << CXGBI_DBG_DDP,
1423 		"%s max payload size: %u/%u, %u/%u.\n",
1424 		cdev->ports[0]->name, cdev->tx_max_size, ddp->max_txsz,
1425 		cdev->rx_max_size, ddp->max_rxsz);
1426 	return 0;
1427 }
1428 EXPORT_SYMBOL_GPL(cxgbi_ddp_init);
1429 
1430 /*
1431  * APIs interacting with open-iscsi libraries
1432  */
1433 
1434 static unsigned char padding[4];
1435 
1436 static void task_release_itt(struct iscsi_task *task, itt_t hdr_itt)
1437 {
1438 	struct scsi_cmnd *sc = task->sc;
1439 	struct iscsi_tcp_conn *tcp_conn = task->conn->dd_data;
1440 	struct cxgbi_conn *cconn = tcp_conn->dd_data;
1441 	struct cxgbi_hba *chba = cconn->chba;
1442 	struct cxgbi_tag_format *tformat = &chba->cdev->tag_format;
1443 	u32 tag = ntohl((__force u32)hdr_itt);
1444 
1445 	log_debug(1 << CXGBI_DBG_DDP,
1446 		   "cdev 0x%p, release tag 0x%x.\n", chba->cdev, tag);
1447 	if (sc &&
1448 	    (scsi_bidi_cmnd(sc) || sc->sc_data_direction == DMA_FROM_DEVICE) &&
1449 	    cxgbi_is_ddp_tag(tformat, tag))
1450 		ddp_tag_release(chba, tag);
1451 }
1452 
1453 static int task_reserve_itt(struct iscsi_task *task, itt_t *hdr_itt)
1454 {
1455 	struct scsi_cmnd *sc = task->sc;
1456 	struct iscsi_conn *conn = task->conn;
1457 	struct iscsi_session *sess = conn->session;
1458 	struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
1459 	struct cxgbi_conn *cconn = tcp_conn->dd_data;
1460 	struct cxgbi_hba *chba = cconn->chba;
1461 	struct cxgbi_tag_format *tformat = &chba->cdev->tag_format;
1462 	u32 sw_tag = (sess->age << cconn->task_idx_bits) | task->itt;
1463 	u32 tag = 0;
1464 	int err = -EINVAL;
1465 
1466 	if (sc &&
1467 	    (scsi_bidi_cmnd(sc) || sc->sc_data_direction == DMA_FROM_DEVICE)) {
1468 		err = cxgbi_ddp_reserve(cconn->cep->csk, &tag, sw_tag,
1469 					scsi_in(sc)->length,
1470 					scsi_in(sc)->table.sgl,
1471 					scsi_in(sc)->table.nents,
1472 					GFP_ATOMIC);
1473 		if (err < 0)
1474 			log_debug(1 << CXGBI_DBG_DDP,
1475 				"csk 0x%p, R task 0x%p, %u,%u, no ddp.\n",
1476 				cconn->cep->csk, task, scsi_in(sc)->length,
1477 				scsi_in(sc)->table.nents);
1478 	}
1479 
1480 	if (err < 0)
1481 		tag = cxgbi_set_non_ddp_tag(tformat, sw_tag);
1482 	/*  the itt need to sent in big-endian order */
1483 	*hdr_itt = (__force itt_t)htonl(tag);
1484 
1485 	log_debug(1 << CXGBI_DBG_DDP,
1486 		"cdev 0x%p, task 0x%p, 0x%x(0x%x,0x%x)->0x%x/0x%x.\n",
1487 		chba->cdev, task, sw_tag, task->itt, sess->age, tag, *hdr_itt);
1488 	return 0;
1489 }
1490 
1491 void cxgbi_parse_pdu_itt(struct iscsi_conn *conn, itt_t itt, int *idx, int *age)
1492 {
1493 	struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
1494 	struct cxgbi_conn *cconn = tcp_conn->dd_data;
1495 	struct cxgbi_device *cdev = cconn->chba->cdev;
1496 	u32 tag = ntohl((__force u32) itt);
1497 	u32 sw_bits;
1498 
1499 	sw_bits = cxgbi_tag_nonrsvd_bits(&cdev->tag_format, tag);
1500 	if (idx)
1501 		*idx = sw_bits & ((1 << cconn->task_idx_bits) - 1);
1502 	if (age)
1503 		*age = (sw_bits >> cconn->task_idx_bits) & ISCSI_AGE_MASK;
1504 
1505 	log_debug(1 << CXGBI_DBG_DDP,
1506 		"cdev 0x%p, tag 0x%x/0x%x, -> 0x%x(0x%x,0x%x).\n",
1507 		cdev, tag, itt, sw_bits, idx ? *idx : 0xFFFFF,
1508 		age ? *age : 0xFF);
1509 }
1510 EXPORT_SYMBOL_GPL(cxgbi_parse_pdu_itt);
1511 
1512 void cxgbi_conn_tx_open(struct cxgbi_sock *csk)
1513 {
1514 	struct iscsi_conn *conn = csk->user_data;
1515 
1516 	if (conn) {
1517 		log_debug(1 << CXGBI_DBG_SOCK,
1518 			"csk 0x%p, cid %d.\n", csk, conn->id);
1519 		iscsi_conn_queue_work(conn);
1520 	}
1521 }
1522 EXPORT_SYMBOL_GPL(cxgbi_conn_tx_open);
1523 
1524 /*
1525  * pdu receive, interact with libiscsi_tcp
1526  */
1527 static inline int read_pdu_skb(struct iscsi_conn *conn,
1528 			       struct sk_buff *skb,
1529 			       unsigned int offset,
1530 			       int offloaded)
1531 {
1532 	int status = 0;
1533 	int bytes_read;
1534 
1535 	bytes_read = iscsi_tcp_recv_skb(conn, skb, offset, offloaded, &status);
1536 	switch (status) {
1537 	case ISCSI_TCP_CONN_ERR:
1538 		pr_info("skb 0x%p, off %u, %d, TCP_ERR.\n",
1539 			  skb, offset, offloaded);
1540 		return -EIO;
1541 	case ISCSI_TCP_SUSPENDED:
1542 		log_debug(1 << CXGBI_DBG_PDU_RX,
1543 			"skb 0x%p, off %u, %d, TCP_SUSPEND, rc %d.\n",
1544 			skb, offset, offloaded, bytes_read);
1545 		/* no transfer - just have caller flush queue */
1546 		return bytes_read;
1547 	case ISCSI_TCP_SKB_DONE:
1548 		pr_info("skb 0x%p, off %u, %d, TCP_SKB_DONE.\n",
1549 			skb, offset, offloaded);
1550 		/*
1551 		 * pdus should always fit in the skb and we should get
1552 		 * segment done notifcation.
1553 		 */
1554 		iscsi_conn_printk(KERN_ERR, conn, "Invalid pdu or skb.");
1555 		return -EFAULT;
1556 	case ISCSI_TCP_SEGMENT_DONE:
1557 		log_debug(1 << CXGBI_DBG_PDU_RX,
1558 			"skb 0x%p, off %u, %d, TCP_SEG_DONE, rc %d.\n",
1559 			skb, offset, offloaded, bytes_read);
1560 		return bytes_read;
1561 	default:
1562 		pr_info("skb 0x%p, off %u, %d, invalid status %d.\n",
1563 			skb, offset, offloaded, status);
1564 		return -EINVAL;
1565 	}
1566 }
1567 
1568 static int skb_read_pdu_bhs(struct iscsi_conn *conn, struct sk_buff *skb)
1569 {
1570 	struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
1571 
1572 	log_debug(1 << CXGBI_DBG_PDU_RX,
1573 		"conn 0x%p, skb 0x%p, len %u, flag 0x%lx.\n",
1574 		conn, skb, skb->len, cxgbi_skcb_flags(skb));
1575 
1576 	if (!iscsi_tcp_recv_segment_is_hdr(tcp_conn)) {
1577 		pr_info("conn 0x%p, skb 0x%p, not hdr.\n", conn, skb);
1578 		iscsi_conn_failure(conn, ISCSI_ERR_PROTO);
1579 		return -EIO;
1580 	}
1581 
1582 	if (conn->hdrdgst_en &&
1583 	    cxgbi_skcb_test_flag(skb, SKCBF_RX_HCRC_ERR)) {
1584 		pr_info("conn 0x%p, skb 0x%p, hcrc.\n", conn, skb);
1585 		iscsi_conn_failure(conn, ISCSI_ERR_HDR_DGST);
1586 		return -EIO;
1587 	}
1588 
1589 	return read_pdu_skb(conn, skb, 0, 0);
1590 }
1591 
1592 static int skb_read_pdu_data(struct iscsi_conn *conn, struct sk_buff *lskb,
1593 			     struct sk_buff *skb, unsigned int offset)
1594 {
1595 	struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
1596 	bool offloaded = 0;
1597 	int opcode = tcp_conn->in.hdr->opcode & ISCSI_OPCODE_MASK;
1598 
1599 	log_debug(1 << CXGBI_DBG_PDU_RX,
1600 		"conn 0x%p, skb 0x%p, len %u, flag 0x%lx.\n",
1601 		conn, skb, skb->len, cxgbi_skcb_flags(skb));
1602 
1603 	if (conn->datadgst_en &&
1604 	    cxgbi_skcb_test_flag(lskb, SKCBF_RX_DCRC_ERR)) {
1605 		pr_info("conn 0x%p, skb 0x%p, dcrc 0x%lx.\n",
1606 			conn, lskb, cxgbi_skcb_flags(lskb));
1607 		iscsi_conn_failure(conn, ISCSI_ERR_DATA_DGST);
1608 		return -EIO;
1609 	}
1610 
1611 	if (iscsi_tcp_recv_segment_is_hdr(tcp_conn))
1612 		return 0;
1613 
1614 	/* coalesced, add header digest length */
1615 	if (lskb == skb && conn->hdrdgst_en)
1616 		offset += ISCSI_DIGEST_SIZE;
1617 
1618 	if (cxgbi_skcb_test_flag(lskb, SKCBF_RX_DATA_DDPD))
1619 		offloaded = 1;
1620 
1621 	if (opcode == ISCSI_OP_SCSI_DATA_IN)
1622 		log_debug(1 << CXGBI_DBG_PDU_RX,
1623 			"skb 0x%p, op 0x%x, itt 0x%x, %u %s ddp'ed.\n",
1624 			skb, opcode, ntohl(tcp_conn->in.hdr->itt),
1625 			tcp_conn->in.datalen, offloaded ? "is" : "not");
1626 
1627 	return read_pdu_skb(conn, skb, offset, offloaded);
1628 }
1629 
1630 static void csk_return_rx_credits(struct cxgbi_sock *csk, int copied)
1631 {
1632 	struct cxgbi_device *cdev = csk->cdev;
1633 	int must_send;
1634 	u32 credits;
1635 
1636 	log_debug(1 << CXGBI_DBG_PDU_RX,
1637 		"csk 0x%p,%u,0x%lu,%u, seq %u, wup %u, thre %u, %u.\n",
1638 		csk, csk->state, csk->flags, csk->tid, csk->copied_seq,
1639 		csk->rcv_wup, cdev->rx_credit_thres,
1640 		cdev->rcv_win);
1641 
1642 	if (csk->state != CTP_ESTABLISHED)
1643 		return;
1644 
1645 	credits = csk->copied_seq - csk->rcv_wup;
1646 	if (unlikely(!credits))
1647 		return;
1648 	if (unlikely(cdev->rx_credit_thres == 0))
1649 		return;
1650 
1651 	must_send = credits + 16384 >= cdev->rcv_win;
1652 	if (must_send || credits >= cdev->rx_credit_thres)
1653 		csk->rcv_wup += cdev->csk_send_rx_credits(csk, credits);
1654 }
1655 
1656 void cxgbi_conn_pdu_ready(struct cxgbi_sock *csk)
1657 {
1658 	struct cxgbi_device *cdev = csk->cdev;
1659 	struct iscsi_conn *conn = csk->user_data;
1660 	struct sk_buff *skb;
1661 	unsigned int read = 0;
1662 	int err = 0;
1663 
1664 	log_debug(1 << CXGBI_DBG_PDU_RX,
1665 		"csk 0x%p, conn 0x%p.\n", csk, conn);
1666 
1667 	if (unlikely(!conn || conn->suspend_rx)) {
1668 		log_debug(1 << CXGBI_DBG_PDU_RX,
1669 			"csk 0x%p, conn 0x%p, id %d, suspend_rx %lu!\n",
1670 			csk, conn, conn ? conn->id : 0xFF,
1671 			conn ? conn->suspend_rx : 0xFF);
1672 		return;
1673 	}
1674 
1675 	while (!err) {
1676 		skb = skb_peek(&csk->receive_queue);
1677 		if (!skb ||
1678 		    !(cxgbi_skcb_test_flag(skb, SKCBF_RX_STATUS))) {
1679 			if (skb)
1680 				log_debug(1 << CXGBI_DBG_PDU_RX,
1681 					"skb 0x%p, NOT ready 0x%lx.\n",
1682 					skb, cxgbi_skcb_flags(skb));
1683 			break;
1684 		}
1685 		__skb_unlink(skb, &csk->receive_queue);
1686 
1687 		read += cxgbi_skcb_rx_pdulen(skb);
1688 		log_debug(1 << CXGBI_DBG_PDU_RX,
1689 			"csk 0x%p, skb 0x%p,%u,f 0x%lx, pdu len %u.\n",
1690 			csk, skb, skb->len, cxgbi_skcb_flags(skb),
1691 			cxgbi_skcb_rx_pdulen(skb));
1692 
1693 		if (cxgbi_skcb_test_flag(skb, SKCBF_RX_COALESCED)) {
1694 			err = skb_read_pdu_bhs(conn, skb);
1695 			if (err < 0) {
1696 				pr_err("coalesced bhs, csk 0x%p, skb 0x%p,%u, "
1697 					"f 0x%lx, plen %u.\n",
1698 					csk, skb, skb->len,
1699 					cxgbi_skcb_flags(skb),
1700 					cxgbi_skcb_rx_pdulen(skb));
1701 				goto skb_done;
1702 			}
1703 			err = skb_read_pdu_data(conn, skb, skb,
1704 						err + cdev->skb_rx_extra);
1705 			if (err < 0)
1706 				pr_err("coalesced data, csk 0x%p, skb 0x%p,%u, "
1707 					"f 0x%lx, plen %u.\n",
1708 					csk, skb, skb->len,
1709 					cxgbi_skcb_flags(skb),
1710 					cxgbi_skcb_rx_pdulen(skb));
1711 		} else {
1712 			err = skb_read_pdu_bhs(conn, skb);
1713 			if (err < 0) {
1714 				pr_err("bhs, csk 0x%p, skb 0x%p,%u, "
1715 					"f 0x%lx, plen %u.\n",
1716 					csk, skb, skb->len,
1717 					cxgbi_skcb_flags(skb),
1718 					cxgbi_skcb_rx_pdulen(skb));
1719 				goto skb_done;
1720 			}
1721 
1722 			if (cxgbi_skcb_test_flag(skb, SKCBF_RX_DATA)) {
1723 				struct sk_buff *dskb;
1724 
1725 				dskb = skb_peek(&csk->receive_queue);
1726 				if (!dskb) {
1727 					pr_err("csk 0x%p, skb 0x%p,%u, f 0x%lx,"
1728 						" plen %u, NO data.\n",
1729 						csk, skb, skb->len,
1730 						cxgbi_skcb_flags(skb),
1731 						cxgbi_skcb_rx_pdulen(skb));
1732 					err = -EIO;
1733 					goto skb_done;
1734 				}
1735 				__skb_unlink(dskb, &csk->receive_queue);
1736 
1737 				err = skb_read_pdu_data(conn, skb, dskb, 0);
1738 				if (err < 0)
1739 					pr_err("data, csk 0x%p, skb 0x%p,%u, "
1740 						"f 0x%lx, plen %u, dskb 0x%p,"
1741 						"%u.\n",
1742 						csk, skb, skb->len,
1743 						cxgbi_skcb_flags(skb),
1744 						cxgbi_skcb_rx_pdulen(skb),
1745 						dskb, dskb->len);
1746 				__kfree_skb(dskb);
1747 			} else
1748 				err = skb_read_pdu_data(conn, skb, skb, 0);
1749 		}
1750 skb_done:
1751 		__kfree_skb(skb);
1752 
1753 		if (err < 0)
1754 			break;
1755 	}
1756 
1757 	log_debug(1 << CXGBI_DBG_PDU_RX, "csk 0x%p, read %u.\n", csk, read);
1758 	if (read) {
1759 		csk->copied_seq += read;
1760 		csk_return_rx_credits(csk, read);
1761 		conn->rxdata_octets += read;
1762 	}
1763 
1764 	if (err < 0) {
1765 		pr_info("csk 0x%p, 0x%p, rx failed %d, read %u.\n",
1766 			csk, conn, err, read);
1767 		iscsi_conn_failure(conn, ISCSI_ERR_CONN_FAILED);
1768 	}
1769 }
1770 EXPORT_SYMBOL_GPL(cxgbi_conn_pdu_ready);
1771 
1772 static int sgl_seek_offset(struct scatterlist *sgl, unsigned int sgcnt,
1773 				unsigned int offset, unsigned int *off,
1774 				struct scatterlist **sgp)
1775 {
1776 	int i;
1777 	struct scatterlist *sg;
1778 
1779 	for_each_sg(sgl, sg, sgcnt, i) {
1780 		if (offset < sg->length) {
1781 			*off = offset;
1782 			*sgp = sg;
1783 			return 0;
1784 		}
1785 		offset -= sg->length;
1786 	}
1787 	return -EFAULT;
1788 }
1789 
1790 static int sgl_read_to_frags(struct scatterlist *sg, unsigned int sgoffset,
1791 				unsigned int dlen, struct page_frag *frags,
1792 				int frag_max)
1793 {
1794 	unsigned int datalen = dlen;
1795 	unsigned int sglen = sg->length - sgoffset;
1796 	struct page *page = sg_page(sg);
1797 	int i;
1798 
1799 	i = 0;
1800 	do {
1801 		unsigned int copy;
1802 
1803 		if (!sglen) {
1804 			sg = sg_next(sg);
1805 			if (!sg) {
1806 				pr_warn("sg %d NULL, len %u/%u.\n",
1807 					i, datalen, dlen);
1808 				return -EINVAL;
1809 			}
1810 			sgoffset = 0;
1811 			sglen = sg->length;
1812 			page = sg_page(sg);
1813 
1814 		}
1815 		copy = min(datalen, sglen);
1816 		if (i && page == frags[i - 1].page &&
1817 		    sgoffset + sg->offset ==
1818 			frags[i - 1].offset + frags[i - 1].size) {
1819 			frags[i - 1].size += copy;
1820 		} else {
1821 			if (i >= frag_max) {
1822 				pr_warn("too many pages %u, dlen %u.\n",
1823 					frag_max, dlen);
1824 				return -EINVAL;
1825 			}
1826 
1827 			frags[i].page = page;
1828 			frags[i].offset = sg->offset + sgoffset;
1829 			frags[i].size = copy;
1830 			i++;
1831 		}
1832 		datalen -= copy;
1833 		sgoffset += copy;
1834 		sglen -= copy;
1835 	} while (datalen);
1836 
1837 	return i;
1838 }
1839 
1840 int cxgbi_conn_alloc_pdu(struct iscsi_task *task, u8 opcode)
1841 {
1842 	struct iscsi_tcp_conn *tcp_conn = task->conn->dd_data;
1843 	struct cxgbi_conn *cconn = tcp_conn->dd_data;
1844 	struct cxgbi_device *cdev = cconn->chba->cdev;
1845 	struct iscsi_conn *conn = task->conn;
1846 	struct iscsi_tcp_task *tcp_task = task->dd_data;
1847 	struct cxgbi_task_data *tdata = iscsi_task_cxgbi_data(task);
1848 	struct scsi_cmnd *sc = task->sc;
1849 	int headroom = SKB_TX_ISCSI_PDU_HEADER_MAX;
1850 
1851 	tcp_task->dd_data = tdata;
1852 	task->hdr = NULL;
1853 
1854 	if (SKB_MAX_HEAD(cdev->skb_tx_rsvd) > (512 * MAX_SKB_FRAGS) &&
1855 	    (opcode == ISCSI_OP_SCSI_DATA_OUT ||
1856 	     (opcode == ISCSI_OP_SCSI_CMD &&
1857 	      (scsi_bidi_cmnd(sc) || sc->sc_data_direction == DMA_TO_DEVICE))))
1858 		/* data could goes into skb head */
1859 		headroom += min_t(unsigned int,
1860 				SKB_MAX_HEAD(cdev->skb_tx_rsvd),
1861 				conn->max_xmit_dlength);
1862 
1863 	tdata->skb = alloc_skb(cdev->skb_tx_rsvd + headroom, GFP_ATOMIC);
1864 	if (!tdata->skb) {
1865 		pr_warn("alloc skb %u+%u, opcode 0x%x failed.\n",
1866 			cdev->skb_tx_rsvd, headroom, opcode);
1867 		return -ENOMEM;
1868 	}
1869 
1870 	skb_reserve(tdata->skb, cdev->skb_tx_rsvd);
1871 	task->hdr = (struct iscsi_hdr *)tdata->skb->data;
1872 	task->hdr_max = SKB_TX_ISCSI_PDU_HEADER_MAX; /* BHS + AHS */
1873 
1874 	/* data_out uses scsi_cmd's itt */
1875 	if (opcode != ISCSI_OP_SCSI_DATA_OUT)
1876 		task_reserve_itt(task, &task->hdr->itt);
1877 
1878 	log_debug(1 << CXGBI_DBG_ISCSI | 1 << CXGBI_DBG_PDU_TX,
1879 		"task 0x%p, op 0x%x, skb 0x%p,%u+%u/%u, itt 0x%x.\n",
1880 		task, opcode, tdata->skb, cdev->skb_tx_rsvd, headroom,
1881 		conn->max_xmit_dlength, ntohl(task->hdr->itt));
1882 
1883 	return 0;
1884 }
1885 EXPORT_SYMBOL_GPL(cxgbi_conn_alloc_pdu);
1886 
1887 static inline void tx_skb_setmode(struct sk_buff *skb, int hcrc, int dcrc)
1888 {
1889 	if (hcrc || dcrc) {
1890 		u8 submode = 0;
1891 
1892 		if (hcrc)
1893 			submode |= 1;
1894 		if (dcrc)
1895 			submode |= 2;
1896 		cxgbi_skcb_ulp_mode(skb) = (ULP2_MODE_ISCSI << 4) | submode;
1897 	} else
1898 		cxgbi_skcb_ulp_mode(skb) = 0;
1899 }
1900 
1901 int cxgbi_conn_init_pdu(struct iscsi_task *task, unsigned int offset,
1902 			      unsigned int count)
1903 {
1904 	struct iscsi_conn *conn = task->conn;
1905 	struct cxgbi_task_data *tdata = iscsi_task_cxgbi_data(task);
1906 	struct sk_buff *skb = tdata->skb;
1907 	unsigned int datalen = count;
1908 	int i, padlen = iscsi_padding(count);
1909 	struct page *pg;
1910 
1911 	log_debug(1 << CXGBI_DBG_ISCSI | 1 << CXGBI_DBG_PDU_TX,
1912 		"task 0x%p,0x%p, skb 0x%p, 0x%x,0x%x,0x%x, %u+%u.\n",
1913 		task, task->sc, skb, (*skb->data) & ISCSI_OPCODE_MASK,
1914 		ntohl(task->cmdsn), ntohl(task->hdr->itt), offset, count);
1915 
1916 	skb_put(skb, task->hdr_len);
1917 	tx_skb_setmode(skb, conn->hdrdgst_en, datalen ? conn->datadgst_en : 0);
1918 	if (!count)
1919 		return 0;
1920 
1921 	if (task->sc) {
1922 		struct scsi_data_buffer *sdb = scsi_out(task->sc);
1923 		struct scatterlist *sg = NULL;
1924 		int err;
1925 
1926 		tdata->offset = offset;
1927 		tdata->count = count;
1928 		err = sgl_seek_offset(
1929 					sdb->table.sgl, sdb->table.nents,
1930 					tdata->offset, &tdata->sgoffset, &sg);
1931 		if (err < 0) {
1932 			pr_warn("tpdu, sgl %u, bad offset %u/%u.\n",
1933 				sdb->table.nents, tdata->offset, sdb->length);
1934 			return err;
1935 		}
1936 		err = sgl_read_to_frags(sg, tdata->sgoffset, tdata->count,
1937 					tdata->frags, MAX_PDU_FRAGS);
1938 		if (err < 0) {
1939 			pr_warn("tpdu, sgl %u, bad offset %u + %u.\n",
1940 				sdb->table.nents, tdata->offset, tdata->count);
1941 			return err;
1942 		}
1943 		tdata->nr_frags = err;
1944 
1945 		if (tdata->nr_frags > MAX_SKB_FRAGS ||
1946 		    (padlen && tdata->nr_frags == MAX_SKB_FRAGS)) {
1947 			char *dst = skb->data + task->hdr_len;
1948 			struct page_frag *frag = tdata->frags;
1949 
1950 			/* data fits in the skb's headroom */
1951 			for (i = 0; i < tdata->nr_frags; i++, frag++) {
1952 				char *src = kmap_atomic(frag->page,
1953 							KM_SOFTIRQ0);
1954 
1955 				memcpy(dst, src+frag->offset, frag->size);
1956 				dst += frag->size;
1957 				kunmap_atomic(src, KM_SOFTIRQ0);
1958 			}
1959 			if (padlen) {
1960 				memset(dst, 0, padlen);
1961 				padlen = 0;
1962 			}
1963 			skb_put(skb, count + padlen);
1964 		} else {
1965 			/* data fit into frag_list */
1966 			for (i = 0; i < tdata->nr_frags; i++) {
1967 				__skb_fill_page_desc(skb, i,
1968 						tdata->frags[i].page,
1969 						tdata->frags[i].offset,
1970 						tdata->frags[i].size);
1971 				skb_frag_ref(skb, i);
1972 			}
1973 			skb_shinfo(skb)->nr_frags = tdata->nr_frags;
1974 			skb->len += count;
1975 			skb->data_len += count;
1976 			skb->truesize += count;
1977 		}
1978 
1979 	} else {
1980 		pg = virt_to_page(task->data);
1981 
1982 		get_page(pg);
1983 		skb_fill_page_desc(skb, 0, pg, offset_in_page(task->data),
1984 					count);
1985 		skb->len += count;
1986 		skb->data_len += count;
1987 		skb->truesize += count;
1988 	}
1989 
1990 	if (padlen) {
1991 		i = skb_shinfo(skb)->nr_frags;
1992 		skb_fill_page_desc(skb, skb_shinfo(skb)->nr_frags,
1993 				virt_to_page(padding), offset_in_page(padding),
1994 				padlen);
1995 
1996 		skb->data_len += padlen;
1997 		skb->truesize += padlen;
1998 		skb->len += padlen;
1999 	}
2000 
2001 	return 0;
2002 }
2003 EXPORT_SYMBOL_GPL(cxgbi_conn_init_pdu);
2004 
2005 int cxgbi_conn_xmit_pdu(struct iscsi_task *task)
2006 {
2007 	struct iscsi_tcp_conn *tcp_conn = task->conn->dd_data;
2008 	struct cxgbi_conn *cconn = tcp_conn->dd_data;
2009 	struct cxgbi_task_data *tdata = iscsi_task_cxgbi_data(task);
2010 	struct sk_buff *skb = tdata->skb;
2011 	unsigned int datalen;
2012 	int err;
2013 
2014 	if (!skb) {
2015 		log_debug(1 << CXGBI_DBG_ISCSI | 1 << CXGBI_DBG_PDU_TX,
2016 			"task 0x%p, skb NULL.\n", task);
2017 		return 0;
2018 	}
2019 
2020 	datalen = skb->data_len;
2021 	tdata->skb = NULL;
2022 	err = cxgbi_sock_send_pdus(cconn->cep->csk, skb);
2023 	if (err > 0) {
2024 		int pdulen = err;
2025 
2026 		log_debug(1 << CXGBI_DBG_PDU_TX,
2027 			"task 0x%p,0x%p, skb 0x%p, len %u/%u, rv %d.\n",
2028 			task, task->sc, skb, skb->len, skb->data_len, err);
2029 
2030 		if (task->conn->hdrdgst_en)
2031 			pdulen += ISCSI_DIGEST_SIZE;
2032 
2033 		if (datalen && task->conn->datadgst_en)
2034 			pdulen += ISCSI_DIGEST_SIZE;
2035 
2036 		task->conn->txdata_octets += pdulen;
2037 		return 0;
2038 	}
2039 
2040 	if (err == -EAGAIN || err == -ENOBUFS) {
2041 		log_debug(1 << CXGBI_DBG_PDU_TX,
2042 			"task 0x%p, skb 0x%p, len %u/%u, %d EAGAIN.\n",
2043 			task, skb, skb->len, skb->data_len, err);
2044 		/* reset skb to send when we are called again */
2045 		tdata->skb = skb;
2046 		return err;
2047 	}
2048 
2049 	kfree_skb(skb);
2050 	log_debug(1 << CXGBI_DBG_ISCSI | 1 << CXGBI_DBG_PDU_TX,
2051 		"itt 0x%x, skb 0x%p, len %u/%u, xmit err %d.\n",
2052 		task->itt, skb, skb->len, skb->data_len, err);
2053 	iscsi_conn_printk(KERN_ERR, task->conn, "xmit err %d.\n", err);
2054 	iscsi_conn_failure(task->conn, ISCSI_ERR_XMIT_FAILED);
2055 	return err;
2056 }
2057 EXPORT_SYMBOL_GPL(cxgbi_conn_xmit_pdu);
2058 
2059 void cxgbi_cleanup_task(struct iscsi_task *task)
2060 {
2061 	struct cxgbi_task_data *tdata = iscsi_task_cxgbi_data(task);
2062 
2063 	log_debug(1 << CXGBI_DBG_ISCSI,
2064 		"task 0x%p, skb 0x%p, itt 0x%x.\n",
2065 		task, tdata->skb, task->hdr_itt);
2066 
2067 	/*  never reached the xmit task callout */
2068 	if (tdata->skb)
2069 		__kfree_skb(tdata->skb);
2070 	memset(tdata, 0, sizeof(*tdata));
2071 
2072 	task_release_itt(task, task->hdr_itt);
2073 	iscsi_tcp_cleanup_task(task);
2074 }
2075 EXPORT_SYMBOL_GPL(cxgbi_cleanup_task);
2076 
2077 void cxgbi_get_conn_stats(struct iscsi_cls_conn *cls_conn,
2078 				struct iscsi_stats *stats)
2079 {
2080 	struct iscsi_conn *conn = cls_conn->dd_data;
2081 
2082 	stats->txdata_octets = conn->txdata_octets;
2083 	stats->rxdata_octets = conn->rxdata_octets;
2084 	stats->scsicmd_pdus = conn->scsicmd_pdus_cnt;
2085 	stats->dataout_pdus = conn->dataout_pdus_cnt;
2086 	stats->scsirsp_pdus = conn->scsirsp_pdus_cnt;
2087 	stats->datain_pdus = conn->datain_pdus_cnt;
2088 	stats->r2t_pdus = conn->r2t_pdus_cnt;
2089 	stats->tmfcmd_pdus = conn->tmfcmd_pdus_cnt;
2090 	stats->tmfrsp_pdus = conn->tmfrsp_pdus_cnt;
2091 	stats->digest_err = 0;
2092 	stats->timeout_err = 0;
2093 	stats->custom_length = 1;
2094 	strcpy(stats->custom[0].desc, "eh_abort_cnt");
2095 	stats->custom[0].value = conn->eh_abort_cnt;
2096 }
2097 EXPORT_SYMBOL_GPL(cxgbi_get_conn_stats);
2098 
2099 static int cxgbi_conn_max_xmit_dlength(struct iscsi_conn *conn)
2100 {
2101 	struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
2102 	struct cxgbi_conn *cconn = tcp_conn->dd_data;
2103 	struct cxgbi_device *cdev = cconn->chba->cdev;
2104 	unsigned int headroom = SKB_MAX_HEAD(cdev->skb_tx_rsvd);
2105 	unsigned int max_def = 512 * MAX_SKB_FRAGS;
2106 	unsigned int max = max(max_def, headroom);
2107 
2108 	max = min(cconn->chba->cdev->tx_max_size, max);
2109 	if (conn->max_xmit_dlength)
2110 		conn->max_xmit_dlength = min(conn->max_xmit_dlength, max);
2111 	else
2112 		conn->max_xmit_dlength = max;
2113 	cxgbi_align_pdu_size(conn->max_xmit_dlength);
2114 
2115 	return 0;
2116 }
2117 
2118 static int cxgbi_conn_max_recv_dlength(struct iscsi_conn *conn)
2119 {
2120 	struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
2121 	struct cxgbi_conn *cconn = tcp_conn->dd_data;
2122 	unsigned int max = cconn->chba->cdev->rx_max_size;
2123 
2124 	cxgbi_align_pdu_size(max);
2125 
2126 	if (conn->max_recv_dlength) {
2127 		if (conn->max_recv_dlength > max) {
2128 			pr_err("MaxRecvDataSegmentLength %u > %u.\n",
2129 				conn->max_recv_dlength, max);
2130 			return -EINVAL;
2131 		}
2132 		conn->max_recv_dlength = min(conn->max_recv_dlength, max);
2133 		cxgbi_align_pdu_size(conn->max_recv_dlength);
2134 	} else
2135 		conn->max_recv_dlength = max;
2136 
2137 	return 0;
2138 }
2139 
2140 int cxgbi_set_conn_param(struct iscsi_cls_conn *cls_conn,
2141 			enum iscsi_param param, char *buf, int buflen)
2142 {
2143 	struct iscsi_conn *conn = cls_conn->dd_data;
2144 	struct iscsi_session *session = conn->session;
2145 	struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
2146 	struct cxgbi_conn *cconn = tcp_conn->dd_data;
2147 	struct cxgbi_sock *csk = cconn->cep->csk;
2148 	int value, err = 0;
2149 
2150 	log_debug(1 << CXGBI_DBG_ISCSI,
2151 		"cls_conn 0x%p, param %d, buf(%d) %s.\n",
2152 		cls_conn, param, buflen, buf);
2153 
2154 	switch (param) {
2155 	case ISCSI_PARAM_HDRDGST_EN:
2156 		err = iscsi_set_param(cls_conn, param, buf, buflen);
2157 		if (!err && conn->hdrdgst_en)
2158 			err = csk->cdev->csk_ddp_setup_digest(csk, csk->tid,
2159 							conn->hdrdgst_en,
2160 							conn->datadgst_en, 0);
2161 		break;
2162 	case ISCSI_PARAM_DATADGST_EN:
2163 		err = iscsi_set_param(cls_conn, param, buf, buflen);
2164 		if (!err && conn->datadgst_en)
2165 			err = csk->cdev->csk_ddp_setup_digest(csk, csk->tid,
2166 							conn->hdrdgst_en,
2167 							conn->datadgst_en, 0);
2168 		break;
2169 	case ISCSI_PARAM_MAX_R2T:
2170 		sscanf(buf, "%d", &value);
2171 		if (value <= 0 || !is_power_of_2(value))
2172 			return -EINVAL;
2173 		if (session->max_r2t == value)
2174 			break;
2175 		iscsi_tcp_r2tpool_free(session);
2176 		err = iscsi_set_param(cls_conn, param, buf, buflen);
2177 		if (!err && iscsi_tcp_r2tpool_alloc(session))
2178 			return -ENOMEM;
2179 	case ISCSI_PARAM_MAX_RECV_DLENGTH:
2180 		err = iscsi_set_param(cls_conn, param, buf, buflen);
2181 		if (!err)
2182 			err = cxgbi_conn_max_recv_dlength(conn);
2183 		break;
2184 	case ISCSI_PARAM_MAX_XMIT_DLENGTH:
2185 		err = iscsi_set_param(cls_conn, param, buf, buflen);
2186 		if (!err)
2187 			err = cxgbi_conn_max_xmit_dlength(conn);
2188 		break;
2189 	default:
2190 		return iscsi_set_param(cls_conn, param, buf, buflen);
2191 	}
2192 	return err;
2193 }
2194 EXPORT_SYMBOL_GPL(cxgbi_set_conn_param);
2195 
2196 int cxgbi_get_ep_param(struct iscsi_endpoint *ep, enum iscsi_param param,
2197 		       char *buf)
2198 {
2199 	struct cxgbi_endpoint *cep = ep->dd_data;
2200 	struct cxgbi_sock *csk;
2201 	int len;
2202 
2203 	log_debug(1 << CXGBI_DBG_ISCSI,
2204 		"cls_conn 0x%p, param %d.\n", ep, param);
2205 
2206 	switch (param) {
2207 	case ISCSI_PARAM_CONN_PORT:
2208 	case ISCSI_PARAM_CONN_ADDRESS:
2209 		if (!cep)
2210 			return -ENOTCONN;
2211 
2212 		csk = cep->csk;
2213 		if (!csk)
2214 			return -ENOTCONN;
2215 
2216 		return iscsi_conn_get_addr_param((struct sockaddr_storage *)
2217 						 &csk->daddr, param, buf);
2218 	default:
2219 		return -ENOSYS;
2220 	}
2221 	return len;
2222 }
2223 EXPORT_SYMBOL_GPL(cxgbi_get_ep_param);
2224 
2225 struct iscsi_cls_conn *
2226 cxgbi_create_conn(struct iscsi_cls_session *cls_session, u32 cid)
2227 {
2228 	struct iscsi_cls_conn *cls_conn;
2229 	struct iscsi_conn *conn;
2230 	struct iscsi_tcp_conn *tcp_conn;
2231 	struct cxgbi_conn *cconn;
2232 
2233 	cls_conn = iscsi_tcp_conn_setup(cls_session, sizeof(*cconn), cid);
2234 	if (!cls_conn)
2235 		return NULL;
2236 
2237 	conn = cls_conn->dd_data;
2238 	tcp_conn = conn->dd_data;
2239 	cconn = tcp_conn->dd_data;
2240 	cconn->iconn = conn;
2241 
2242 	log_debug(1 << CXGBI_DBG_ISCSI,
2243 		"cid %u(0x%x), cls 0x%p,0x%p, conn 0x%p,0x%p,0x%p.\n",
2244 		cid, cid, cls_session, cls_conn, conn, tcp_conn, cconn);
2245 
2246 	return cls_conn;
2247 }
2248 EXPORT_SYMBOL_GPL(cxgbi_create_conn);
2249 
2250 int cxgbi_bind_conn(struct iscsi_cls_session *cls_session,
2251 				struct iscsi_cls_conn *cls_conn,
2252 				u64 transport_eph, int is_leading)
2253 {
2254 	struct iscsi_conn *conn = cls_conn->dd_data;
2255 	struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
2256 	struct cxgbi_conn *cconn = tcp_conn->dd_data;
2257 	struct iscsi_endpoint *ep;
2258 	struct cxgbi_endpoint *cep;
2259 	struct cxgbi_sock *csk;
2260 	int err;
2261 
2262 	ep = iscsi_lookup_endpoint(transport_eph);
2263 	if (!ep)
2264 		return -EINVAL;
2265 
2266 	/*  setup ddp pagesize */
2267 	cep = ep->dd_data;
2268 	csk = cep->csk;
2269 	err = csk->cdev->csk_ddp_setup_pgidx(csk, csk->tid, page_idx, 0);
2270 	if (err < 0)
2271 		return err;
2272 
2273 	err = iscsi_conn_bind(cls_session, cls_conn, is_leading);
2274 	if (err)
2275 		return -EINVAL;
2276 
2277 	/*  calculate the tag idx bits needed for this conn based on cmds_max */
2278 	cconn->task_idx_bits = (__ilog2_u32(conn->session->cmds_max - 1)) + 1;
2279 
2280 	write_lock_bh(&csk->callback_lock);
2281 	csk->user_data = conn;
2282 	cconn->chba = cep->chba;
2283 	cconn->cep = cep;
2284 	cep->cconn = cconn;
2285 	write_unlock_bh(&csk->callback_lock);
2286 
2287 	cxgbi_conn_max_xmit_dlength(conn);
2288 	cxgbi_conn_max_recv_dlength(conn);
2289 
2290 	log_debug(1 << CXGBI_DBG_ISCSI,
2291 		"cls 0x%p,0x%p, ep 0x%p, cconn 0x%p, csk 0x%p.\n",
2292 		cls_session, cls_conn, ep, cconn, csk);
2293 	/*  init recv engine */
2294 	iscsi_tcp_hdr_recv_prep(tcp_conn);
2295 
2296 	return 0;
2297 }
2298 EXPORT_SYMBOL_GPL(cxgbi_bind_conn);
2299 
2300 struct iscsi_cls_session *cxgbi_create_session(struct iscsi_endpoint *ep,
2301 						u16 cmds_max, u16 qdepth,
2302 						u32 initial_cmdsn)
2303 {
2304 	struct cxgbi_endpoint *cep;
2305 	struct cxgbi_hba *chba;
2306 	struct Scsi_Host *shost;
2307 	struct iscsi_cls_session *cls_session;
2308 	struct iscsi_session *session;
2309 
2310 	if (!ep) {
2311 		pr_err("missing endpoint.\n");
2312 		return NULL;
2313 	}
2314 
2315 	cep = ep->dd_data;
2316 	chba = cep->chba;
2317 	shost = chba->shost;
2318 
2319 	BUG_ON(chba != iscsi_host_priv(shost));
2320 
2321 	cls_session = iscsi_session_setup(chba->cdev->itp, shost,
2322 					cmds_max, 0,
2323 					sizeof(struct iscsi_tcp_task) +
2324 					sizeof(struct cxgbi_task_data),
2325 					initial_cmdsn, ISCSI_MAX_TARGET);
2326 	if (!cls_session)
2327 		return NULL;
2328 
2329 	session = cls_session->dd_data;
2330 	if (iscsi_tcp_r2tpool_alloc(session))
2331 		goto remove_session;
2332 
2333 	log_debug(1 << CXGBI_DBG_ISCSI,
2334 		"ep 0x%p, cls sess 0x%p.\n", ep, cls_session);
2335 	return cls_session;
2336 
2337 remove_session:
2338 	iscsi_session_teardown(cls_session);
2339 	return NULL;
2340 }
2341 EXPORT_SYMBOL_GPL(cxgbi_create_session);
2342 
2343 void cxgbi_destroy_session(struct iscsi_cls_session *cls_session)
2344 {
2345 	log_debug(1 << CXGBI_DBG_ISCSI,
2346 		"cls sess 0x%p.\n", cls_session);
2347 
2348 	iscsi_tcp_r2tpool_free(cls_session->dd_data);
2349 	iscsi_session_teardown(cls_session);
2350 }
2351 EXPORT_SYMBOL_GPL(cxgbi_destroy_session);
2352 
2353 int cxgbi_set_host_param(struct Scsi_Host *shost, enum iscsi_host_param param,
2354 			char *buf, int buflen)
2355 {
2356 	struct cxgbi_hba *chba = iscsi_host_priv(shost);
2357 
2358 	if (!chba->ndev) {
2359 		shost_printk(KERN_ERR, shost, "Could not get host param. "
2360 				"netdev for host not set.\n");
2361 		return -ENODEV;
2362 	}
2363 
2364 	log_debug(1 << CXGBI_DBG_ISCSI,
2365 		"shost 0x%p, hba 0x%p,%s, param %d, buf(%d) %s.\n",
2366 		shost, chba, chba->ndev->name, param, buflen, buf);
2367 
2368 	switch (param) {
2369 	case ISCSI_HOST_PARAM_IPADDRESS:
2370 	{
2371 		__be32 addr = in_aton(buf);
2372 		log_debug(1 << CXGBI_DBG_ISCSI,
2373 			"hba %s, req. ipv4 %pI4.\n", chba->ndev->name, &addr);
2374 		cxgbi_set_iscsi_ipv4(chba, addr);
2375 		return 0;
2376 	}
2377 	case ISCSI_HOST_PARAM_HWADDRESS:
2378 	case ISCSI_HOST_PARAM_NETDEV_NAME:
2379 		return 0;
2380 	default:
2381 		return iscsi_host_set_param(shost, param, buf, buflen);
2382 	}
2383 }
2384 EXPORT_SYMBOL_GPL(cxgbi_set_host_param);
2385 
2386 int cxgbi_get_host_param(struct Scsi_Host *shost, enum iscsi_host_param param,
2387 			char *buf)
2388 {
2389 	struct cxgbi_hba *chba = iscsi_host_priv(shost);
2390 	int len = 0;
2391 
2392 	if (!chba->ndev) {
2393 		shost_printk(KERN_ERR, shost, "Could not get host param. "
2394 				"netdev for host not set.\n");
2395 		return -ENODEV;
2396 	}
2397 
2398 	log_debug(1 << CXGBI_DBG_ISCSI,
2399 		"shost 0x%p, hba 0x%p,%s, param %d.\n",
2400 		shost, chba, chba->ndev->name, param);
2401 
2402 	switch (param) {
2403 	case ISCSI_HOST_PARAM_HWADDRESS:
2404 		len = sysfs_format_mac(buf, chba->ndev->dev_addr, 6);
2405 		break;
2406 	case ISCSI_HOST_PARAM_NETDEV_NAME:
2407 		len = sprintf(buf, "%s\n", chba->ndev->name);
2408 		break;
2409 	case ISCSI_HOST_PARAM_IPADDRESS:
2410 	{
2411 		__be32 addr;
2412 
2413 		addr = cxgbi_get_iscsi_ipv4(chba);
2414 		len = sprintf(buf, "%pI4", &addr);
2415 		log_debug(1 << CXGBI_DBG_ISCSI,
2416 			"hba %s, ipv4 %pI4.\n", chba->ndev->name, &addr);
2417 		break;
2418 	}
2419 	default:
2420 		return iscsi_host_get_param(shost, param, buf);
2421 	}
2422 
2423 	return len;
2424 }
2425 EXPORT_SYMBOL_GPL(cxgbi_get_host_param);
2426 
2427 struct iscsi_endpoint *cxgbi_ep_connect(struct Scsi_Host *shost,
2428 					struct sockaddr *dst_addr,
2429 					int non_blocking)
2430 {
2431 	struct iscsi_endpoint *ep;
2432 	struct cxgbi_endpoint *cep;
2433 	struct cxgbi_hba *hba = NULL;
2434 	struct cxgbi_sock *csk;
2435 	int err = -EINVAL;
2436 
2437 	log_debug(1 << CXGBI_DBG_ISCSI | 1 << CXGBI_DBG_SOCK,
2438 		"shost 0x%p, non_blocking %d, dst_addr 0x%p.\n",
2439 		shost, non_blocking, dst_addr);
2440 
2441 	if (shost) {
2442 		hba = iscsi_host_priv(shost);
2443 		if (!hba) {
2444 			pr_info("shost 0x%p, priv NULL.\n", shost);
2445 			goto err_out;
2446 		}
2447 	}
2448 
2449 	csk = cxgbi_check_route(dst_addr);
2450 	if (IS_ERR(csk))
2451 		return (struct iscsi_endpoint *)csk;
2452 	cxgbi_sock_get(csk);
2453 
2454 	if (!hba)
2455 		hba = csk->cdev->hbas[csk->port_id];
2456 	else if (hba != csk->cdev->hbas[csk->port_id]) {
2457 		pr_info("Could not connect through requested host %u"
2458 			"hba 0x%p != 0x%p (%u).\n",
2459 			shost->host_no, hba,
2460 			csk->cdev->hbas[csk->port_id], csk->port_id);
2461 		err = -ENOSPC;
2462 		goto release_conn;
2463 	}
2464 
2465 	err = sock_get_port(csk);
2466 	if (err)
2467 		goto release_conn;
2468 
2469 	cxgbi_sock_set_state(csk, CTP_CONNECTING);
2470 	err = csk->cdev->csk_init_act_open(csk);
2471 	if (err)
2472 		goto release_conn;
2473 
2474 	if (cxgbi_sock_is_closing(csk)) {
2475 		err = -ENOSPC;
2476 		pr_info("csk 0x%p is closing.\n", csk);
2477 		goto release_conn;
2478 	}
2479 
2480 	ep = iscsi_create_endpoint(sizeof(*cep));
2481 	if (!ep) {
2482 		err = -ENOMEM;
2483 		pr_info("iscsi alloc ep, OOM.\n");
2484 		goto release_conn;
2485 	}
2486 
2487 	cep = ep->dd_data;
2488 	cep->csk = csk;
2489 	cep->chba = hba;
2490 
2491 	log_debug(1 << CXGBI_DBG_ISCSI | 1 << CXGBI_DBG_SOCK,
2492 		"ep 0x%p, cep 0x%p, csk 0x%p, hba 0x%p,%s.\n",
2493 		ep, cep, csk, hba, hba->ndev->name);
2494 	return ep;
2495 
2496 release_conn:
2497 	cxgbi_sock_put(csk);
2498 	cxgbi_sock_closed(csk);
2499 err_out:
2500 	return ERR_PTR(err);
2501 }
2502 EXPORT_SYMBOL_GPL(cxgbi_ep_connect);
2503 
2504 int cxgbi_ep_poll(struct iscsi_endpoint *ep, int timeout_ms)
2505 {
2506 	struct cxgbi_endpoint *cep = ep->dd_data;
2507 	struct cxgbi_sock *csk = cep->csk;
2508 
2509 	if (!cxgbi_sock_is_established(csk))
2510 		return 0;
2511 	return 1;
2512 }
2513 EXPORT_SYMBOL_GPL(cxgbi_ep_poll);
2514 
2515 void cxgbi_ep_disconnect(struct iscsi_endpoint *ep)
2516 {
2517 	struct cxgbi_endpoint *cep = ep->dd_data;
2518 	struct cxgbi_conn *cconn = cep->cconn;
2519 	struct cxgbi_sock *csk = cep->csk;
2520 
2521 	log_debug(1 << CXGBI_DBG_ISCSI | 1 << CXGBI_DBG_SOCK,
2522 		"ep 0x%p, cep 0x%p, cconn 0x%p, csk 0x%p,%u,0x%lx.\n",
2523 		ep, cep, cconn, csk, csk->state, csk->flags);
2524 
2525 	if (cconn && cconn->iconn) {
2526 		iscsi_suspend_tx(cconn->iconn);
2527 		write_lock_bh(&csk->callback_lock);
2528 		cep->csk->user_data = NULL;
2529 		cconn->cep = NULL;
2530 		write_unlock_bh(&csk->callback_lock);
2531 	}
2532 	iscsi_destroy_endpoint(ep);
2533 
2534 	if (likely(csk->state >= CTP_ESTABLISHED))
2535 		need_active_close(csk);
2536 	else
2537 		cxgbi_sock_closed(csk);
2538 
2539 	cxgbi_sock_put(csk);
2540 }
2541 EXPORT_SYMBOL_GPL(cxgbi_ep_disconnect);
2542 
2543 int cxgbi_iscsi_init(struct iscsi_transport *itp,
2544 			struct scsi_transport_template **stt)
2545 {
2546 	*stt = iscsi_register_transport(itp);
2547 	if (*stt == NULL) {
2548 		pr_err("unable to register %s transport 0x%p.\n",
2549 			itp->name, itp);
2550 		return -ENODEV;
2551 	}
2552 	log_debug(1 << CXGBI_DBG_ISCSI,
2553 		"%s, registered iscsi transport 0x%p.\n",
2554 		itp->name, stt);
2555 	return 0;
2556 }
2557 EXPORT_SYMBOL_GPL(cxgbi_iscsi_init);
2558 
2559 void cxgbi_iscsi_cleanup(struct iscsi_transport *itp,
2560 			struct scsi_transport_template **stt)
2561 {
2562 	if (*stt) {
2563 		log_debug(1 << CXGBI_DBG_ISCSI,
2564 			"de-register transport 0x%p, %s, stt 0x%p.\n",
2565 			itp, itp->name, *stt);
2566 		*stt = NULL;
2567 		iscsi_unregister_transport(itp);
2568 	}
2569 }
2570 EXPORT_SYMBOL_GPL(cxgbi_iscsi_cleanup);
2571 
2572 mode_t cxgbi_attr_is_visible(int param_type, int param)
2573 {
2574 	switch (param_type) {
2575 	case ISCSI_HOST_PARAM:
2576 		switch (param) {
2577 		case ISCSI_HOST_PARAM_NETDEV_NAME:
2578 		case ISCSI_HOST_PARAM_HWADDRESS:
2579 		case ISCSI_HOST_PARAM_IPADDRESS:
2580 		case ISCSI_HOST_PARAM_INITIATOR_NAME:
2581 			return S_IRUGO;
2582 		default:
2583 			return 0;
2584 		}
2585 	case ISCSI_PARAM:
2586 		switch (param) {
2587 		case ISCSI_PARAM_MAX_RECV_DLENGTH:
2588 		case ISCSI_PARAM_MAX_XMIT_DLENGTH:
2589 		case ISCSI_PARAM_HDRDGST_EN:
2590 		case ISCSI_PARAM_DATADGST_EN:
2591 		case ISCSI_PARAM_CONN_ADDRESS:
2592 		case ISCSI_PARAM_CONN_PORT:
2593 		case ISCSI_PARAM_EXP_STATSN:
2594 		case ISCSI_PARAM_PERSISTENT_ADDRESS:
2595 		case ISCSI_PARAM_PERSISTENT_PORT:
2596 		case ISCSI_PARAM_PING_TMO:
2597 		case ISCSI_PARAM_RECV_TMO:
2598 		case ISCSI_PARAM_INITIAL_R2T_EN:
2599 		case ISCSI_PARAM_MAX_R2T:
2600 		case ISCSI_PARAM_IMM_DATA_EN:
2601 		case ISCSI_PARAM_FIRST_BURST:
2602 		case ISCSI_PARAM_MAX_BURST:
2603 		case ISCSI_PARAM_PDU_INORDER_EN:
2604 		case ISCSI_PARAM_DATASEQ_INORDER_EN:
2605 		case ISCSI_PARAM_ERL:
2606 		case ISCSI_PARAM_TARGET_NAME:
2607 		case ISCSI_PARAM_TPGT:
2608 		case ISCSI_PARAM_USERNAME:
2609 		case ISCSI_PARAM_PASSWORD:
2610 		case ISCSI_PARAM_USERNAME_IN:
2611 		case ISCSI_PARAM_PASSWORD_IN:
2612 		case ISCSI_PARAM_FAST_ABORT:
2613 		case ISCSI_PARAM_ABORT_TMO:
2614 		case ISCSI_PARAM_LU_RESET_TMO:
2615 		case ISCSI_PARAM_TGT_RESET_TMO:
2616 		case ISCSI_PARAM_IFACE_NAME:
2617 		case ISCSI_PARAM_INITIATOR_NAME:
2618 			return S_IRUGO;
2619 		default:
2620 			return 0;
2621 		}
2622 	}
2623 
2624 	return 0;
2625 }
2626 EXPORT_SYMBOL_GPL(cxgbi_attr_is_visible);
2627 
2628 static int __init libcxgbi_init_module(void)
2629 {
2630 	sw_tag_idx_bits = (__ilog2_u32(ISCSI_ITT_MASK)) + 1;
2631 	sw_tag_age_bits = (__ilog2_u32(ISCSI_AGE_MASK)) + 1;
2632 
2633 	pr_info("tag itt 0x%x, %u bits, age 0x%x, %u bits.\n",
2634 		ISCSI_ITT_MASK, sw_tag_idx_bits,
2635 		ISCSI_AGE_MASK, sw_tag_age_bits);
2636 
2637 	ddp_setup_host_page_size();
2638 	return 0;
2639 }
2640 
2641 static void __exit libcxgbi_exit_module(void)
2642 {
2643 	cxgbi_device_unregister_all(0xFF);
2644 	return;
2645 }
2646 
2647 module_init(libcxgbi_init_module);
2648 module_exit(libcxgbi_exit_module);
2649