Linux Kernel  3.7.1
 All Data Structures Namespaces Files Functions Variables Typedefs Enumerations Enumerator Macros Groups Pages
libcxgbi.c
Go to the documentation of this file.
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 ([email protected])
11  * Written by: Rakesh Ranjan ([email protected])
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.");
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 }
75 
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;
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 }
98 
99 static inline void cxgbi_device_destroy(struct cxgbi_device *cdev)
100 {
102  "cdev 0x%p, p# %u.\n", cdev, cdev->nports);
103  cxgbi_hbas_remove(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 
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 
142  "cdev 0x%p, p# %u.\n", cdev, nports);
143  return cdev;
144 }
146 
148 {
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 }
158 
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) {
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 }
177 
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);
191  "lldev 0x%p, NO match found.\n", lldev);
192  return NULL;
193 }
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);
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);
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 
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 }
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 
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 }
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);
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 
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  */
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 }
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,
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_neigh_lookup(dst, &daddr->sin_addr.s_addr);
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_neigh;
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_neigh;
525  }
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_neigh;
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  neigh_release(n);
545 
546  return csk;
547 
548 rel_neigh:
549  neigh_release(n);
550 
551 rel_rt:
552  ip_rt_put(rt);
553  if (csk)
554  cxgbi_sock_closed(csk);
555 err_out:
556  return ERR_PTR(err);
557 }
558 
559 void cxgbi_sock_established(struct cxgbi_sock *csk, unsigned int snd_isn,
560  unsigned int opt)
561 {
562  csk->write_seq = csk->snd_nxt = csk->snd_una = snd_isn;
563  dst_confirm(csk->dst);
564  smp_mb();
565  cxgbi_sock_set_state(csk, CTP_ESTABLISHED);
566 }
568 
569 static void cxgbi_inform_iscsi_conn_closing(struct cxgbi_sock *csk)
570 {
572  "csk 0x%p, state %u, flags 0x%lx, conn 0x%p.\n",
573  csk, csk->state, csk->flags, csk->user_data);
574 
575  if (csk->state != CTP_ESTABLISHED) {
577  if (csk->user_data)
581  }
582 }
583 
584 void cxgbi_sock_closed(struct cxgbi_sock *csk)
585 {
586  log_debug(1 << CXGBI_DBG_SOCK, "csk 0x%p,%u,0x%lx,%u.\n",
587  csk, (csk)->state, (csk)->flags, (csk)->tid);
588  cxgbi_sock_set_flag(csk, CTPF_ACTIVE_CLOSE_NEEDED);
589  if (csk->state == CTP_ACTIVE_OPEN || csk->state == CTP_CLOSED)
590  return;
591  if (csk->saddr.sin_port)
592  sock_put_port(csk);
593  if (csk->dst)
594  dst_release(csk->dst);
595  csk->cdev->csk_release_offload_resources(csk);
596  cxgbi_sock_set_state(csk, CTP_CLOSED);
597  cxgbi_inform_iscsi_conn_closing(csk);
598  cxgbi_sock_put(csk);
599 }
601 
602 static void need_active_close(struct cxgbi_sock *csk)
603 {
604  int data_lost;
605  int close_req = 0;
606 
607  log_debug(1 << CXGBI_DBG_SOCK, "csk 0x%p,%u,0x%lx,%u.\n",
608  csk, (csk)->state, (csk)->flags, (csk)->tid);
609  spin_lock_bh(&csk->lock);
610  dst_confirm(csk->dst);
611  data_lost = skb_queue_len(&csk->receive_queue);
612  __skb_queue_purge(&csk->receive_queue);
613 
614  if (csk->state == CTP_ACTIVE_OPEN)
615  cxgbi_sock_set_flag(csk, CTPF_ACTIVE_CLOSE_NEEDED);
616  else if (csk->state == CTP_ESTABLISHED) {
617  close_req = 1;
618  cxgbi_sock_set_state(csk, CTP_ACTIVE_CLOSE);
619  } else if (csk->state == CTP_PASSIVE_CLOSE) {
620  close_req = 1;
621  cxgbi_sock_set_state(csk, CTP_CLOSE_WAIT_2);
622  }
623 
624  if (close_req) {
625  if (data_lost)
626  csk->cdev->csk_send_abort_req(csk);
627  else
628  csk->cdev->csk_send_close_req(csk);
629  }
630 
631  spin_unlock_bh(&csk->lock);
632 }
633 
634 void cxgbi_sock_fail_act_open(struct cxgbi_sock *csk, int errno)
635 {
636  pr_info("csk 0x%p,%u,%lx, %pI4:%u-%pI4:%u, err %d.\n",
637  csk, csk->state, csk->flags,
638  &csk->saddr.sin_addr.s_addr, csk->saddr.sin_port,
639  &csk->daddr.sin_addr.s_addr, csk->daddr.sin_port,
640  errno);
641 
642  cxgbi_sock_set_state(csk, CTP_CONNECTING);
643  csk->err = errno;
644  cxgbi_sock_closed(csk);
645 }
647 
649 {
650  struct cxgbi_sock *csk = (struct cxgbi_sock *)skb->sk;
651 
652  log_debug(1 << CXGBI_DBG_SOCK, "csk 0x%p,%u,0x%lx,%u.\n",
653  csk, (csk)->state, (csk)->flags, (csk)->tid);
654  cxgbi_sock_get(csk);
655  spin_lock_bh(&csk->lock);
656  if (csk->state == CTP_ACTIVE_OPEN)
658  spin_unlock_bh(&csk->lock);
659  cxgbi_sock_put(csk);
660  __kfree_skb(skb);
661 }
663 
665 {
666  cxgbi_sock_get(csk);
667  spin_lock_bh(&csk->lock);
668  if (cxgbi_sock_flag(csk, CTPF_ABORT_RPL_PENDING)) {
669  if (!cxgbi_sock_flag(csk, CTPF_ABORT_RPL_RCVD))
670  cxgbi_sock_set_flag(csk, CTPF_ABORT_RPL_RCVD);
671  else {
672  cxgbi_sock_clear_flag(csk, CTPF_ABORT_RPL_RCVD);
673  cxgbi_sock_clear_flag(csk, CTPF_ABORT_RPL_PENDING);
674  if (cxgbi_sock_flag(csk, CTPF_ABORT_REQ_RCVD))
675  pr_err("csk 0x%p,%u,0x%lx,%u,ABT_RPL_RSS.\n",
676  csk, csk->state, csk->flags, csk->tid);
677  cxgbi_sock_closed(csk);
678  }
679  }
680  spin_unlock_bh(&csk->lock);
681  cxgbi_sock_put(csk);
682 }
684 
686 {
687  log_debug(1 << CXGBI_DBG_SOCK, "csk 0x%p,%u,0x%lx,%u.\n",
688  csk, (csk)->state, (csk)->flags, (csk)->tid);
689  cxgbi_sock_get(csk);
690  spin_lock_bh(&csk->lock);
691 
692  if (cxgbi_sock_flag(csk, CTPF_ABORT_RPL_PENDING))
693  goto done;
694 
695  switch (csk->state) {
696  case CTP_ESTABLISHED:
697  cxgbi_sock_set_state(csk, CTP_PASSIVE_CLOSE);
698  break;
699  case CTP_ACTIVE_CLOSE:
700  cxgbi_sock_set_state(csk, CTP_CLOSE_WAIT_2);
701  break;
702  case CTP_CLOSE_WAIT_1:
703  cxgbi_sock_closed(csk);
704  break;
705  case CTP_ABORTING:
706  break;
707  default:
708  pr_err("csk 0x%p,%u,0x%lx,%u, bad state.\n",
709  csk, csk->state, csk->flags, csk->tid);
710  }
711  cxgbi_inform_iscsi_conn_closing(csk);
712 done:
713  spin_unlock_bh(&csk->lock);
714  cxgbi_sock_put(csk);
715 }
717 
719 {
720  log_debug(1 << CXGBI_DBG_SOCK, "csk 0x%p,%u,0x%lx,%u.\n",
721  csk, (csk)->state, (csk)->flags, (csk)->tid);
722  cxgbi_sock_get(csk);
723  spin_lock_bh(&csk->lock);
724 
725  csk->snd_una = snd_nxt - 1;
726  if (cxgbi_sock_flag(csk, CTPF_ABORT_RPL_PENDING))
727  goto done;
728 
729  switch (csk->state) {
730  case CTP_ACTIVE_CLOSE:
731  cxgbi_sock_set_state(csk, CTP_CLOSE_WAIT_1);
732  break;
733  case CTP_CLOSE_WAIT_1:
734  case CTP_CLOSE_WAIT_2:
735  cxgbi_sock_closed(csk);
736  break;
737  case CTP_ABORTING:
738  break;
739  default:
740  pr_err("csk 0x%p,%u,0x%lx,%u, bad state.\n",
741  csk, csk->state, csk->flags, csk->tid);
742  }
743 done:
744  spin_unlock_bh(&csk->lock);
745  cxgbi_sock_put(csk);
746 }
748 
749 void cxgbi_sock_rcv_wr_ack(struct cxgbi_sock *csk, unsigned int credits,
750  unsigned int snd_una, int seq_chk)
751 {
753  "csk 0x%p,%u,0x%lx,%u, cr %u,%u+%u, snd_una %u,%d.\n",
754  csk, csk->state, csk->flags, csk->tid, credits,
755  csk->wr_cred, csk->wr_una_cred, snd_una, seq_chk);
756 
757  spin_lock_bh(&csk->lock);
758 
759  csk->wr_cred += credits;
760  if (csk->wr_una_cred > csk->wr_max_cred - csk->wr_cred)
761  csk->wr_una_cred = csk->wr_max_cred - csk->wr_cred;
762 
763  while (credits) {
764  struct sk_buff *p = cxgbi_sock_peek_wr(csk);
765 
766  if (unlikely(!p)) {
767  pr_err("csk 0x%p,%u,0x%lx,%u, cr %u,%u+%u, empty.\n",
768  csk, csk->state, csk->flags, csk->tid, credits,
769  csk->wr_cred, csk->wr_una_cred);
770  break;
771  }
772 
773  if (unlikely(credits < p->csum)) {
774  pr_warn("csk 0x%p,%u,0x%lx,%u, cr %u,%u+%u, < %u.\n",
775  csk, csk->state, csk->flags, csk->tid,
776  credits, csk->wr_cred, csk->wr_una_cred,
777  p->csum);
778  p->csum -= credits;
779  break;
780  } else {
781  cxgbi_sock_dequeue_wr(csk);
782  credits -= p->csum;
783  kfree_skb(p);
784  }
785  }
786 
788 
789  if (seq_chk) {
790  if (unlikely(before(snd_una, csk->snd_una))) {
791  pr_warn("csk 0x%p,%u,0x%lx,%u, snd_una %u/%u.",
792  csk, csk->state, csk->flags, csk->tid, snd_una,
793  csk->snd_una);
794  goto done;
795  }
796 
797  if (csk->snd_una != snd_una) {
798  csk->snd_una = snd_una;
799  dst_confirm(csk->dst);
800  }
801  }
802 
803  if (skb_queue_len(&csk->write_queue)) {
804  if (csk->cdev->csk_push_tx_frames(csk, 0))
805  cxgbi_conn_tx_open(csk);
806  } else
807  cxgbi_conn_tx_open(csk);
808 done:
809  spin_unlock_bh(&csk->lock);
810 }
812 
813 static unsigned int cxgbi_sock_find_best_mtu(struct cxgbi_sock *csk,
814  unsigned short mtu)
815 {
816  int i = 0;
817 
818  while (i < csk->cdev->nmtus - 1 && csk->cdev->mtus[i + 1] <= mtu)
819  ++i;
820 
821  return i;
822 }
823 
824 unsigned int cxgbi_sock_select_mss(struct cxgbi_sock *csk, unsigned int pmtu)
825 {
826  unsigned int idx;
827  struct dst_entry *dst = csk->dst;
828 
829  csk->advmss = dst_metric_advmss(dst);
830 
831  if (csk->advmss > pmtu - 40)
832  csk->advmss = pmtu - 40;
833  if (csk->advmss < csk->cdev->mtus[0] - 40)
834  csk->advmss = csk->cdev->mtus[0] - 40;
835  idx = cxgbi_sock_find_best_mtu(csk, csk->advmss + 40);
836 
837  return idx;
838 }
840 
841 void cxgbi_sock_skb_entail(struct cxgbi_sock *csk, struct sk_buff *skb)
842 {
843  cxgbi_skcb_tcp_seq(skb) = csk->write_seq;
844  __skb_queue_tail(&csk->write_queue, skb);
845 }
847 
849 {
850  struct sk_buff *skb;
851 
852  while ((skb = cxgbi_sock_dequeue_wr(csk)) != NULL)
853  kfree_skb(skb);
854 }
856 
858 {
859  int pending = cxgbi_sock_count_pending_wrs(csk);
860 
861  if (unlikely(csk->wr_cred + pending != csk->wr_max_cred))
862  pr_err("csk 0x%p, tid %u, credit %u + %u != %u.\n",
863  csk, csk->tid, csk->wr_cred, pending, csk->wr_max_cred);
864 }
866 
867 static int cxgbi_sock_send_pdus(struct cxgbi_sock *csk, struct sk_buff *skb)
868 {
869  struct cxgbi_device *cdev = csk->cdev;
870  struct sk_buff *next;
871  int err, copied = 0;
872 
873  spin_lock_bh(&csk->lock);
874 
875  if (csk->state != CTP_ESTABLISHED) {
877  "csk 0x%p,%u,0x%lx,%u, EAGAIN.\n",
878  csk, csk->state, csk->flags, csk->tid);
879  err = -EAGAIN;
880  goto out_err;
881  }
882 
883  if (csk->err) {
885  "csk 0x%p,%u,0x%lx,%u, EPIPE %d.\n",
886  csk, csk->state, csk->flags, csk->tid, csk->err);
887  err = -EPIPE;
888  goto out_err;
889  }
890 
891  if (csk->write_seq - csk->snd_una >= cdev->snd_win) {
893  "csk 0x%p,%u,0x%lx,%u, FULL %u-%u >= %u.\n",
894  csk, csk->state, csk->flags, csk->tid, csk->write_seq,
895  csk->snd_una, cdev->snd_win);
896  err = -ENOBUFS;
897  goto out_err;
898  }
899 
900  while (skb) {
901  int frags = skb_shinfo(skb)->nr_frags +
902  (skb->len != skb->data_len);
903 
904  if (unlikely(skb_headroom(skb) < cdev->skb_tx_rsvd)) {
905  pr_err("csk 0x%p, skb head %u < %u.\n",
906  csk, skb_headroom(skb), cdev->skb_tx_rsvd);
907  err = -EINVAL;
908  goto out_err;
909  }
910 
911  if (frags >= SKB_WR_LIST_SIZE) {
912  pr_err("csk 0x%p, frags %d, %u,%u >%u.\n",
913  csk, skb_shinfo(skb)->nr_frags, skb->len,
914  skb->data_len, (uint)(SKB_WR_LIST_SIZE));
915  err = -EINVAL;
916  goto out_err;
917  }
918 
919  next = skb->next;
920  skb->next = NULL;
921  cxgbi_skcb_set_flag(skb, SKCBF_TX_NEED_HDR);
922  cxgbi_sock_skb_entail(csk, skb);
923  copied += skb->len;
924  csk->write_seq += skb->len +
925  cxgbi_ulp_extra_len(cxgbi_skcb_ulp_mode(skb));
926  skb = next;
927  }
928 done:
929  if (likely(skb_queue_len(&csk->write_queue)))
930  cdev->csk_push_tx_frames(csk, 1);
931  spin_unlock_bh(&csk->lock);
932  return copied;
933 
934 out_err:
935  if (copied == 0 && err == -EPIPE)
936  copied = csk->err ? csk->err : -EPIPE;
937  else
938  copied = err;
939  goto done;
940 }
941 
942 /*
943  * Direct Data Placement -
944  * Directly place the iSCSI Data-In or Data-Out PDU's payload into pre-posted
945  * final destination host-memory buffers based on the Initiator Task Tag (ITT)
946  * in Data-In or Target Task Tag (TTT) in Data-Out PDUs.
947  * The host memory address is programmed into h/w in the format of pagepod
948  * entries.
949  * The location of the pagepod entry is encoded into ddp tag which is used as
950  * the base for ITT/TTT.
951  */
952 
953 static unsigned char ddp_page_order[DDP_PGIDX_MAX] = {0, 1, 2, 4};
954 static unsigned char ddp_page_shift[DDP_PGIDX_MAX] = {12, 13, 14, 16};
955 static unsigned char page_idx = DDP_PGIDX_MAX;
956 
957 static unsigned char sw_tag_idx_bits;
958 static unsigned char sw_tag_age_bits;
959 
960 /*
961  * Direct-Data Placement page size adjustment
962  */
963 static int ddp_adjust_page_table(void)
964 {
965  int i;
966  unsigned int base_order, order;
967 
968  if (PAGE_SIZE < (1UL << ddp_page_shift[0])) {
969  pr_info("PAGE_SIZE 0x%lx too small, min 0x%lx\n",
970  PAGE_SIZE, 1UL << ddp_page_shift[0]);
971  return -EINVAL;
972  }
973 
974  base_order = get_order(1UL << ddp_page_shift[0]);
975  order = get_order(1UL << PAGE_SHIFT);
976 
977  for (i = 0; i < DDP_PGIDX_MAX; i++) {
978  /* first is the kernel page size, then just doubling */
979  ddp_page_order[i] = order - base_order + i;
980  ddp_page_shift[i] = PAGE_SHIFT + i;
981  }
982  return 0;
983 }
984 
985 static int ddp_find_page_index(unsigned long pgsz)
986 {
987  int i;
988 
989  for (i = 0; i < DDP_PGIDX_MAX; i++) {
990  if (pgsz == (1UL << ddp_page_shift[i]))
991  return i;
992  }
993  pr_info("ddp page size %lu not supported.\n", pgsz);
994  return DDP_PGIDX_MAX;
995 }
996 
997 static void ddp_setup_host_page_size(void)
998 {
999  if (page_idx == DDP_PGIDX_MAX) {
1000  page_idx = ddp_find_page_index(PAGE_SIZE);
1001 
1002  if (page_idx == DDP_PGIDX_MAX) {
1003  pr_info("system PAGE %lu, update hw.\n", PAGE_SIZE);
1004  if (ddp_adjust_page_table() < 0) {
1005  pr_info("PAGE %lu, disable ddp.\n", PAGE_SIZE);
1006  return;
1007  }
1008  page_idx = ddp_find_page_index(PAGE_SIZE);
1009  }
1010  pr_info("system PAGE %lu, ddp idx %u.\n", PAGE_SIZE, page_idx);
1011  }
1012 }
1013 
1014 void cxgbi_ddp_page_size_factor(int *pgsz_factor)
1015 {
1016  int i;
1017 
1018  for (i = 0; i < DDP_PGIDX_MAX; i++)
1019  pgsz_factor[i] = ddp_page_order[i];
1020 }
1022 
1023 /*
1024  * DDP setup & teardown
1025  */
1026 
1028  struct cxgbi_pagepod_hdr *hdr,
1029  struct cxgbi_gather_list *gl, unsigned int gidx)
1030 {
1031  int i;
1032 
1033  memcpy(ppod, hdr, sizeof(*hdr));
1034  for (i = 0; i < (PPOD_PAGES_MAX + 1); i++, gidx++) {
1035  ppod->addr[i] = gidx < gl->nelem ?
1036  cpu_to_be64(gl->phys_addr[gidx]) : 0ULL;
1037  }
1038 }
1040 
1042 {
1043  memset(ppod, 0, sizeof(*ppod));
1044 }
1046 
1047 static inline int ddp_find_unused_entries(struct cxgbi_ddp_info *ddp,
1048  unsigned int start, unsigned int max,
1049  unsigned int count,
1050  struct cxgbi_gather_list *gl)
1051 {
1052  unsigned int i, j, k;
1053 
1054  /* not enough entries */
1055  if ((max - start) < count) {
1056  log_debug(1 << CXGBI_DBG_DDP,
1057  "NOT enough entries %u+%u < %u.\n", start, count, max);
1058  return -EBUSY;
1059  }
1060 
1061  max -= count;
1062  spin_lock(&ddp->map_lock);
1063  for (i = start; i < max;) {
1064  for (j = 0, k = i; j < count; j++, k++) {
1065  if (ddp->gl_map[k])
1066  break;
1067  }
1068  if (j == count) {
1069  for (j = 0, k = i; j < count; j++, k++)
1070  ddp->gl_map[k] = gl;
1071  spin_unlock(&ddp->map_lock);
1072  return i;
1073  }
1074  i += j + 1;
1075  }
1076  spin_unlock(&ddp->map_lock);
1077  log_debug(1 << CXGBI_DBG_DDP,
1078  "NO suitable entries %u available.\n", count);
1079  return -EBUSY;
1080 }
1081 
1082 static inline void ddp_unmark_entries(struct cxgbi_ddp_info *ddp,
1083  int start, int count)
1084 {
1085  spin_lock(&ddp->map_lock);
1086  memset(&ddp->gl_map[start], 0,
1087  count * sizeof(struct cxgbi_gather_list *));
1088  spin_unlock(&ddp->map_lock);
1089 }
1090 
1091 static inline void ddp_gl_unmap(struct pci_dev *pdev,
1092  struct cxgbi_gather_list *gl)
1093 {
1094  int i;
1095 
1096  for (i = 0; i < gl->nelem; i++)
1097  dma_unmap_page(&pdev->dev, gl->phys_addr[i], PAGE_SIZE,
1099 }
1100 
1101 static inline int ddp_gl_map(struct pci_dev *pdev,
1102  struct cxgbi_gather_list *gl)
1103 {
1104  int i;
1105 
1106  for (i = 0; i < gl->nelem; i++) {
1107  gl->phys_addr[i] = dma_map_page(&pdev->dev, gl->pages[i], 0,
1108  PAGE_SIZE,
1110  if (unlikely(dma_mapping_error(&pdev->dev, gl->phys_addr[i]))) {
1111  log_debug(1 << CXGBI_DBG_DDP,
1112  "page %d 0x%p, 0x%p dma mapping err.\n",
1113  i, gl->pages[i], pdev);
1114  goto unmap;
1115  }
1116  }
1117  return i;
1118 unmap:
1119  if (i) {
1120  unsigned int nelem = gl->nelem;
1121 
1122  gl->nelem = i;
1123  ddp_gl_unmap(pdev, gl);
1124  gl->nelem = nelem;
1125  }
1126  return -EINVAL;
1127 }
1128 
1129 static void ddp_release_gl(struct cxgbi_gather_list *gl,
1130  struct pci_dev *pdev)
1131 {
1132  ddp_gl_unmap(pdev, gl);
1133  kfree(gl);
1134 }
1135 
1136 static struct cxgbi_gather_list *ddp_make_gl(unsigned int xferlen,
1137  struct scatterlist *sgl,
1138  unsigned int sgcnt,
1139  struct pci_dev *pdev,
1140  gfp_t gfp)
1141 {
1142  struct cxgbi_gather_list *gl;
1143  struct scatterlist *sg = sgl;
1144  struct page *sgpage = sg_page(sg);
1145  unsigned int sglen = sg->length;
1146  unsigned int sgoffset = sg->offset;
1147  unsigned int npages = (xferlen + sgoffset + PAGE_SIZE - 1) >>
1148  PAGE_SHIFT;
1149  int i = 1, j = 0;
1150 
1151  if (xferlen < DDP_THRESHOLD) {
1152  log_debug(1 << CXGBI_DBG_DDP,
1153  "xfer %u < threshold %u, no ddp.\n",
1154  xferlen, DDP_THRESHOLD);
1155  return NULL;
1156  }
1157 
1158  gl = kzalloc(sizeof(struct cxgbi_gather_list) +
1159  npages * (sizeof(dma_addr_t) +
1160  sizeof(struct page *)), gfp);
1161  if (!gl) {
1162  log_debug(1 << CXGBI_DBG_DDP,
1163  "xfer %u, %u pages, OOM.\n", xferlen, npages);
1164  return NULL;
1165  }
1166 
1167  log_debug(1 << CXGBI_DBG_DDP,
1168  "xfer %u, sgl %u, gl max %u.\n", xferlen, sgcnt, npages);
1169 
1170  gl->pages = (struct page **)&gl->phys_addr[npages];
1171  gl->nelem = npages;
1172  gl->length = xferlen;
1173  gl->offset = sgoffset;
1174  gl->pages[0] = sgpage;
1175 
1176  for (i = 1, sg = sg_next(sgl), j = 0; i < sgcnt;
1177  i++, sg = sg_next(sg)) {
1178  struct page *page = sg_page(sg);
1179 
1180  if (sgpage == page && sg->offset == sgoffset + sglen)
1181  sglen += sg->length;
1182  else {
1183  /* make sure the sgl is fit for ddp:
1184  * each has the same page size, and
1185  * all of the middle pages are used completely
1186  */
1187  if ((j && sgoffset) || ((i != sgcnt - 1) &&
1188  ((sglen + sgoffset) & ~PAGE_MASK))) {
1189  log_debug(1 << CXGBI_DBG_DDP,
1190  "page %d/%u, %u + %u.\n",
1191  i, sgcnt, sgoffset, sglen);
1192  goto error_out;
1193  }
1194 
1195  j++;
1196  if (j == gl->nelem || sg->offset) {
1197  log_debug(1 << CXGBI_DBG_DDP,
1198  "page %d/%u, offset %u.\n",
1199  j, gl->nelem, sg->offset);
1200  goto error_out;
1201  }
1202  gl->pages[j] = page;
1203  sglen = sg->length;
1204  sgoffset = sg->offset;
1205  sgpage = page;
1206  }
1207  }
1208  gl->nelem = ++j;
1209 
1210  if (ddp_gl_map(pdev, gl) < 0)
1211  goto error_out;
1212 
1213  return gl;
1214 
1215 error_out:
1216  kfree(gl);
1217  return NULL;
1218 }
1219 
1220 static void ddp_tag_release(struct cxgbi_hba *chba, u32 tag)
1221 {
1222  struct cxgbi_device *cdev = chba->cdev;
1223  struct cxgbi_ddp_info *ddp = cdev->ddp;
1224  u32 idx;
1225 
1226  idx = (tag >> PPOD_IDX_SHIFT) & ddp->idx_mask;
1227  if (idx < ddp->nppods) {
1228  struct cxgbi_gather_list *gl = ddp->gl_map[idx];
1229  unsigned int npods;
1230 
1231  if (!gl || !gl->nelem) {
1232  pr_warn("tag 0x%x, idx %u, gl 0x%p, %u.\n",
1233  tag, idx, gl, gl ? gl->nelem : 0);
1234  return;
1235  }
1236  npods = (gl->nelem + PPOD_PAGES_MAX - 1) >> PPOD_PAGES_SHIFT;
1237  log_debug(1 << CXGBI_DBG_DDP,
1238  "tag 0x%x, release idx %u, npods %u.\n",
1239  tag, idx, npods);
1240  cdev->csk_ddp_clear(chba, tag, idx, npods);
1241  ddp_unmark_entries(ddp, idx, npods);
1242  ddp_release_gl(gl, ddp->pdev);
1243  } else
1244  pr_warn("tag 0x%x, idx %u > max %u.\n", tag, idx, ddp->nppods);
1245 }
1246 
1247 static int ddp_tag_reserve(struct cxgbi_sock *csk, unsigned int tid,
1248  u32 sw_tag, u32 *tagp, struct cxgbi_gather_list *gl,
1249  gfp_t gfp)
1250 {
1251  struct cxgbi_device *cdev = csk->cdev;
1252  struct cxgbi_ddp_info *ddp = cdev->ddp;
1253  struct cxgbi_tag_format *tformat = &cdev->tag_format;
1254  struct cxgbi_pagepod_hdr hdr;
1255  unsigned int npods;
1256  int idx = -1;
1257  int err = -ENOMEM;
1258  u32 tag;
1259 
1260  npods = (gl->nelem + PPOD_PAGES_MAX - 1) >> PPOD_PAGES_SHIFT;
1261  if (ddp->idx_last == ddp->nppods)
1262  idx = ddp_find_unused_entries(ddp, 0, ddp->nppods,
1263  npods, gl);
1264  else {
1265  idx = ddp_find_unused_entries(ddp, ddp->idx_last + 1,
1266  ddp->nppods, npods,
1267  gl);
1268  if (idx < 0 && ddp->idx_last >= npods) {
1269  idx = ddp_find_unused_entries(ddp, 0,
1270  min(ddp->idx_last + npods, ddp->nppods),
1271  npods, gl);
1272  }
1273  }
1274  if (idx < 0) {
1275  log_debug(1 << CXGBI_DBG_DDP,
1276  "xferlen %u, gl %u, npods %u NO DDP.\n",
1277  gl->length, gl->nelem, npods);
1278  return idx;
1279  }
1280 
1281  tag = cxgbi_ddp_tag_base(tformat, sw_tag);
1282  tag |= idx << PPOD_IDX_SHIFT;
1283 
1284  hdr.rsvd = 0;
1285  hdr.vld_tid = htonl(PPOD_VALID_FLAG | PPOD_TID(tid));
1286  hdr.pgsz_tag_clr = htonl(tag & ddp->rsvd_tag_mask);
1287  hdr.max_offset = htonl(gl->length);
1288  hdr.page_offset = htonl(gl->offset);
1289 
1290  err = cdev->csk_ddp_set(csk, &hdr, idx, npods, gl);
1291  if (err < 0)
1292  goto unmark_entries;
1293 
1294  ddp->idx_last = idx;
1295  log_debug(1 << CXGBI_DBG_DDP,
1296  "xfer %u, gl %u,%u, tid 0x%x, tag 0x%x->0x%x(%u,%u).\n",
1297  gl->length, gl->nelem, gl->offset, tid, sw_tag, tag, idx,
1298  npods);
1299  *tagp = tag;
1300  return 0;
1301 
1302 unmark_entries:
1303  ddp_unmark_entries(ddp, idx, npods);
1304  return err;
1305 }
1306 
1307 int cxgbi_ddp_reserve(struct cxgbi_sock *csk, unsigned int *tagp,
1308  unsigned int sw_tag, unsigned int xferlen,
1309  struct scatterlist *sgl, unsigned int sgcnt, gfp_t gfp)
1310 {
1311  struct cxgbi_device *cdev = csk->cdev;
1312  struct cxgbi_tag_format *tformat = &cdev->tag_format;
1313  struct cxgbi_gather_list *gl;
1314  int err;
1315 
1316  if (page_idx >= DDP_PGIDX_MAX || !cdev->ddp ||
1317  xferlen < DDP_THRESHOLD) {
1318  log_debug(1 << CXGBI_DBG_DDP,
1319  "pgidx %u, xfer %u, NO ddp.\n", page_idx, xferlen);
1320  return -EINVAL;
1321  }
1322 
1323  if (!cxgbi_sw_tag_usable(tformat, sw_tag)) {
1324  log_debug(1 << CXGBI_DBG_DDP,
1325  "sw_tag 0x%x NOT usable.\n", sw_tag);
1326  return -EINVAL;
1327  }
1328 
1329  gl = ddp_make_gl(xferlen, sgl, sgcnt, cdev->pdev, gfp);
1330  if (!gl)
1331  return -ENOMEM;
1332 
1333  err = ddp_tag_reserve(csk, csk->tid, sw_tag, tagp, gl, gfp);
1334  if (err < 0)
1335  ddp_release_gl(gl, cdev->pdev);
1336 
1337  return err;
1338 }
1339 
1340 static void ddp_destroy(struct kref *kref)
1341 {
1342  struct cxgbi_ddp_info *ddp = container_of(kref,
1343  struct cxgbi_ddp_info,
1344  refcnt);
1345  struct cxgbi_device *cdev = ddp->cdev;
1346  int i = 0;
1347 
1348  pr_info("kref 0, destroy ddp 0x%p, cdev 0x%p.\n", ddp, cdev);
1349 
1350  while (i < ddp->nppods) {
1351  struct cxgbi_gather_list *gl = ddp->gl_map[i];
1352 
1353  if (gl) {
1354  int npods = (gl->nelem + PPOD_PAGES_MAX - 1)
1355  >> PPOD_PAGES_SHIFT;
1356  pr_info("cdev 0x%p, ddp %d + %d.\n", cdev, i, npods);
1357  kfree(gl);
1358  i += npods;
1359  } else
1360  i++;
1361  }
1362  cxgbi_free_big_mem(ddp);
1363 }
1364 
1366 {
1367  struct cxgbi_ddp_info *ddp = cdev->ddp;
1368 
1369  log_debug(1 << CXGBI_DBG_DDP,
1370  "cdev 0x%p, release ddp 0x%p.\n", cdev, ddp);
1371  cdev->ddp = NULL;
1372  if (ddp)
1373  return kref_put(&ddp->refcnt, ddp_destroy);
1374  return 0;
1375 }
1377 
1378 int cxgbi_ddp_init(struct cxgbi_device *cdev,
1379  unsigned int llimit, unsigned int ulimit,
1380  unsigned int max_txsz, unsigned int max_rxsz)
1381 {
1382  struct cxgbi_ddp_info *ddp;
1383  unsigned int ppmax, bits;
1384 
1385  ppmax = (ulimit - llimit + 1) >> PPOD_SIZE_SHIFT;
1386  bits = __ilog2_u32(ppmax) + 1;
1387  if (bits > PPOD_IDX_MAX_SIZE)
1388  bits = PPOD_IDX_MAX_SIZE;
1389  ppmax = (1 << (bits - 1)) - 1;
1390 
1391  ddp = cxgbi_alloc_big_mem(sizeof(struct cxgbi_ddp_info) +
1392  ppmax * (sizeof(struct cxgbi_gather_list *) +
1393  sizeof(struct sk_buff *)),
1394  GFP_KERNEL);
1395  if (!ddp) {
1396  pr_warn("cdev 0x%p, ddp ppmax %u OOM.\n", cdev, ppmax);
1397  return -ENOMEM;
1398  }
1399  ddp->gl_map = (struct cxgbi_gather_list **)(ddp + 1);
1400  cdev->ddp = ddp;
1401 
1402  spin_lock_init(&ddp->map_lock);
1403  kref_init(&ddp->refcnt);
1404 
1405  ddp->cdev = cdev;
1406  ddp->pdev = cdev->pdev;
1407  ddp->llimit = llimit;
1408  ddp->ulimit = ulimit;
1409  ddp->max_txsz = min_t(unsigned int, max_txsz, ULP2_MAX_PKT_SIZE);
1410  ddp->max_rxsz = min_t(unsigned int, max_rxsz, ULP2_MAX_PKT_SIZE);
1411  ddp->nppods = ppmax;
1412  ddp->idx_last = ppmax;
1413  ddp->idx_bits = bits;
1414  ddp->idx_mask = (1 << bits) - 1;
1415  ddp->rsvd_tag_mask = (1 << (bits + PPOD_IDX_SHIFT)) - 1;
1416 
1417  cdev->tag_format.sw_bits = sw_tag_idx_bits + sw_tag_age_bits;
1418  cdev->tag_format.rsvd_bits = ddp->idx_bits;
1419  cdev->tag_format.rsvd_shift = PPOD_IDX_SHIFT;
1420  cdev->tag_format.rsvd_mask = (1 << cdev->tag_format.rsvd_bits) - 1;
1421 
1422  pr_info("%s tag format, sw %u, rsvd %u,%u, mask 0x%x.\n",
1423  cdev->ports[0]->name, cdev->tag_format.sw_bits,
1424  cdev->tag_format.rsvd_bits, cdev->tag_format.rsvd_shift,
1425  cdev->tag_format.rsvd_mask);
1426 
1427  cdev->tx_max_size = min_t(unsigned int, ULP2_MAX_PDU_PAYLOAD,
1429  cdev->rx_max_size = min_t(unsigned int, ULP2_MAX_PDU_PAYLOAD,
1431 
1432  log_debug(1 << CXGBI_DBG_DDP,
1433  "%s max payload size: %u/%u, %u/%u.\n",
1434  cdev->ports[0]->name, cdev->tx_max_size, ddp->max_txsz,
1435  cdev->rx_max_size, ddp->max_rxsz);
1436  return 0;
1437 }
1439 
1440 /*
1441  * APIs interacting with open-iscsi libraries
1442  */
1443 
1444 static unsigned char padding[4];
1445 
1446 static void task_release_itt(struct iscsi_task *task, itt_t hdr_itt)
1447 {
1448  struct scsi_cmnd *sc = task->sc;
1449  struct iscsi_tcp_conn *tcp_conn = task->conn->dd_data;
1450  struct cxgbi_conn *cconn = tcp_conn->dd_data;
1451  struct cxgbi_hba *chba = cconn->chba;
1452  struct cxgbi_tag_format *tformat = &chba->cdev->tag_format;
1453  u32 tag = ntohl((__force u32)hdr_itt);
1454 
1455  log_debug(1 << CXGBI_DBG_DDP,
1456  "cdev 0x%p, release tag 0x%x.\n", chba->cdev, tag);
1457  if (sc &&
1458  (scsi_bidi_cmnd(sc) || sc->sc_data_direction == DMA_FROM_DEVICE) &&
1459  cxgbi_is_ddp_tag(tformat, tag))
1460  ddp_tag_release(chba, tag);
1461 }
1462 
1463 static int task_reserve_itt(struct iscsi_task *task, itt_t *hdr_itt)
1464 {
1465  struct scsi_cmnd *sc = task->sc;
1466  struct iscsi_conn *conn = task->conn;
1467  struct iscsi_session *sess = conn->session;
1468  struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
1469  struct cxgbi_conn *cconn = tcp_conn->dd_data;
1470  struct cxgbi_hba *chba = cconn->chba;
1471  struct cxgbi_tag_format *tformat = &chba->cdev->tag_format;
1472  u32 sw_tag = (sess->age << cconn->task_idx_bits) | task->itt;
1473  u32 tag = 0;
1474  int err = -EINVAL;
1475 
1476  if (sc &&
1477  (scsi_bidi_cmnd(sc) || sc->sc_data_direction == DMA_FROM_DEVICE)) {
1478  err = cxgbi_ddp_reserve(cconn->cep->csk, &tag, sw_tag,
1479  scsi_in(sc)->length,
1480  scsi_in(sc)->table.sgl,
1481  scsi_in(sc)->table.nents,
1482  GFP_ATOMIC);
1483  if (err < 0)
1484  log_debug(1 << CXGBI_DBG_DDP,
1485  "csk 0x%p, R task 0x%p, %u,%u, no ddp.\n",
1486  cconn->cep->csk, task, scsi_in(sc)->length,
1487  scsi_in(sc)->table.nents);
1488  }
1489 
1490  if (err < 0)
1491  tag = cxgbi_set_non_ddp_tag(tformat, sw_tag);
1492  /* the itt need to sent in big-endian order */
1493  *hdr_itt = (__force itt_t)htonl(tag);
1494 
1495  log_debug(1 << CXGBI_DBG_DDP,
1496  "cdev 0x%p, task 0x%p, 0x%x(0x%x,0x%x)->0x%x/0x%x.\n",
1497  chba->cdev, task, sw_tag, task->itt, sess->age, tag, *hdr_itt);
1498  return 0;
1499 }
1500 
1501 void cxgbi_parse_pdu_itt(struct iscsi_conn *conn, itt_t itt, int *idx, int *age)
1502 {
1503  struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
1504  struct cxgbi_conn *cconn = tcp_conn->dd_data;
1505  struct cxgbi_device *cdev = cconn->chba->cdev;
1506  u32 tag = ntohl((__force u32) itt);
1507  u32 sw_bits;
1508 
1509  sw_bits = cxgbi_tag_nonrsvd_bits(&cdev->tag_format, tag);
1510  if (idx)
1511  *idx = sw_bits & ((1 << cconn->task_idx_bits) - 1);
1512  if (age)
1513  *age = (sw_bits >> cconn->task_idx_bits) & ISCSI_AGE_MASK;
1514 
1515  log_debug(1 << CXGBI_DBG_DDP,
1516  "cdev 0x%p, tag 0x%x/0x%x, -> 0x%x(0x%x,0x%x).\n",
1517  cdev, tag, itt, sw_bits, idx ? *idx : 0xFFFFF,
1518  age ? *age : 0xFF);
1519 }
1521 
1523 {
1524  struct iscsi_conn *conn = csk->user_data;
1525 
1526  if (conn) {
1528  "csk 0x%p, cid %d.\n", csk, conn->id);
1529  iscsi_conn_queue_work(conn);
1530  }
1531 }
1533 
1534 /*
1535  * pdu receive, interact with libiscsi_tcp
1536  */
1537 static inline int read_pdu_skb(struct iscsi_conn *conn,
1538  struct sk_buff *skb,
1539  unsigned int offset,
1540  int offloaded)
1541 {
1542  int status = 0;
1543  int bytes_read;
1544 
1545  bytes_read = iscsi_tcp_recv_skb(conn, skb, offset, offloaded, &status);
1546  switch (status) {
1547  case ISCSI_TCP_CONN_ERR:
1548  pr_info("skb 0x%p, off %u, %d, TCP_ERR.\n",
1549  skb, offset, offloaded);
1550  return -EIO;
1551  case ISCSI_TCP_SUSPENDED:
1553  "skb 0x%p, off %u, %d, TCP_SUSPEND, rc %d.\n",
1554  skb, offset, offloaded, bytes_read);
1555  /* no transfer - just have caller flush queue */
1556  return bytes_read;
1557  case ISCSI_TCP_SKB_DONE:
1558  pr_info("skb 0x%p, off %u, %d, TCP_SKB_DONE.\n",
1559  skb, offset, offloaded);
1560  /*
1561  * pdus should always fit in the skb and we should get
1562  * segment done notifcation.
1563  */
1564  iscsi_conn_printk(KERN_ERR, conn, "Invalid pdu or skb.");
1565  return -EFAULT;
1568  "skb 0x%p, off %u, %d, TCP_SEG_DONE, rc %d.\n",
1569  skb, offset, offloaded, bytes_read);
1570  return bytes_read;
1571  default:
1572  pr_info("skb 0x%p, off %u, %d, invalid status %d.\n",
1573  skb, offset, offloaded, status);
1574  return -EINVAL;
1575  }
1576 }
1577 
1578 static int skb_read_pdu_bhs(struct iscsi_conn *conn, struct sk_buff *skb)
1579 {
1580  struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
1581 
1583  "conn 0x%p, skb 0x%p, len %u, flag 0x%lx.\n",
1584  conn, skb, skb->len, cxgbi_skcb_flags(skb));
1585 
1586  if (!iscsi_tcp_recv_segment_is_hdr(tcp_conn)) {
1587  pr_info("conn 0x%p, skb 0x%p, not hdr.\n", conn, skb);
1589  return -EIO;
1590  }
1591 
1592  if (conn->hdrdgst_en &&
1593  cxgbi_skcb_test_flag(skb, SKCBF_RX_HCRC_ERR)) {
1594  pr_info("conn 0x%p, skb 0x%p, hcrc.\n", conn, skb);
1596  return -EIO;
1597  }
1598 
1599  return read_pdu_skb(conn, skb, 0, 0);
1600 }
1601 
1602 static int skb_read_pdu_data(struct iscsi_conn *conn, struct sk_buff *lskb,
1603  struct sk_buff *skb, unsigned int offset)
1604 {
1605  struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
1606  bool offloaded = 0;
1607  int opcode = tcp_conn->in.hdr->opcode & ISCSI_OPCODE_MASK;
1608 
1610  "conn 0x%p, skb 0x%p, len %u, flag 0x%lx.\n",
1611  conn, skb, skb->len, cxgbi_skcb_flags(skb));
1612 
1613  if (conn->datadgst_en &&
1614  cxgbi_skcb_test_flag(lskb, SKCBF_RX_DCRC_ERR)) {
1615  pr_info("conn 0x%p, skb 0x%p, dcrc 0x%lx.\n",
1616  conn, lskb, cxgbi_skcb_flags(lskb));
1618  return -EIO;
1619  }
1620 
1621  if (iscsi_tcp_recv_segment_is_hdr(tcp_conn))
1622  return 0;
1623 
1624  /* coalesced, add header digest length */
1625  if (lskb == skb && conn->hdrdgst_en)
1626  offset += ISCSI_DIGEST_SIZE;
1627 
1628  if (cxgbi_skcb_test_flag(lskb, SKCBF_RX_DATA_DDPD))
1629  offloaded = 1;
1630 
1631  if (opcode == ISCSI_OP_SCSI_DATA_IN)
1633  "skb 0x%p, op 0x%x, itt 0x%x, %u %s ddp'ed.\n",
1634  skb, opcode, ntohl(tcp_conn->in.hdr->itt),
1635  tcp_conn->in.datalen, offloaded ? "is" : "not");
1636 
1637  return read_pdu_skb(conn, skb, offset, offloaded);
1638 }
1639 
1640 static void csk_return_rx_credits(struct cxgbi_sock *csk, int copied)
1641 {
1642  struct cxgbi_device *cdev = csk->cdev;
1643  int must_send;
1644  u32 credits;
1645 
1647  "csk 0x%p,%u,0x%lu,%u, seq %u, wup %u, thre %u, %u.\n",
1648  csk, csk->state, csk->flags, csk->tid, csk->copied_seq,
1649  csk->rcv_wup, cdev->rx_credit_thres,
1650  cdev->rcv_win);
1651 
1652  if (csk->state != CTP_ESTABLISHED)
1653  return;
1654 
1655  credits = csk->copied_seq - csk->rcv_wup;
1656  if (unlikely(!credits))
1657  return;
1658  if (unlikely(cdev->rx_credit_thres == 0))
1659  return;
1660 
1661  must_send = credits + 16384 >= cdev->rcv_win;
1662  if (must_send || credits >= cdev->rx_credit_thres)
1663  csk->rcv_wup += cdev->csk_send_rx_credits(csk, credits);
1664 }
1665 
1667 {
1668  struct cxgbi_device *cdev = csk->cdev;
1669  struct iscsi_conn *conn = csk->user_data;
1670  struct sk_buff *skb;
1671  unsigned int read = 0;
1672  int err = 0;
1673 
1675  "csk 0x%p, conn 0x%p.\n", csk, conn);
1676 
1677  if (unlikely(!conn || conn->suspend_rx)) {
1679  "csk 0x%p, conn 0x%p, id %d, suspend_rx %lu!\n",
1680  csk, conn, conn ? conn->id : 0xFF,
1681  conn ? conn->suspend_rx : 0xFF);
1682  return;
1683  }
1684 
1685  while (!err) {
1686  skb = skb_peek(&csk->receive_queue);
1687  if (!skb ||
1688  !(cxgbi_skcb_test_flag(skb, SKCBF_RX_STATUS))) {
1689  if (skb)
1691  "skb 0x%p, NOT ready 0x%lx.\n",
1692  skb, cxgbi_skcb_flags(skb));
1693  break;
1694  }
1695  __skb_unlink(skb, &csk->receive_queue);
1696 
1697  read += cxgbi_skcb_rx_pdulen(skb);
1699  "csk 0x%p, skb 0x%p,%u,f 0x%lx, pdu len %u.\n",
1700  csk, skb, skb->len, cxgbi_skcb_flags(skb),
1701  cxgbi_skcb_rx_pdulen(skb));
1702 
1703  if (cxgbi_skcb_test_flag(skb, SKCBF_RX_COALESCED)) {
1704  err = skb_read_pdu_bhs(conn, skb);
1705  if (err < 0) {
1706  pr_err("coalesced bhs, 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  goto skb_done;
1712  }
1713  err = skb_read_pdu_data(conn, skb, skb,
1714  err + cdev->skb_rx_extra);
1715  if (err < 0)
1716  pr_err("coalesced data, csk 0x%p, skb 0x%p,%u, "
1717  "f 0x%lx, plen %u.\n",
1718  csk, skb, skb->len,
1719  cxgbi_skcb_flags(skb),
1720  cxgbi_skcb_rx_pdulen(skb));
1721  } else {
1722  err = skb_read_pdu_bhs(conn, skb);
1723  if (err < 0) {
1724  pr_err("bhs, csk 0x%p, skb 0x%p,%u, "
1725  "f 0x%lx, plen %u.\n",
1726  csk, skb, skb->len,
1727  cxgbi_skcb_flags(skb),
1728  cxgbi_skcb_rx_pdulen(skb));
1729  goto skb_done;
1730  }
1731 
1732  if (cxgbi_skcb_test_flag(skb, SKCBF_RX_DATA)) {
1733  struct sk_buff *dskb;
1734 
1735  dskb = skb_peek(&csk->receive_queue);
1736  if (!dskb) {
1737  pr_err("csk 0x%p, skb 0x%p,%u, f 0x%lx,"
1738  " plen %u, NO data.\n",
1739  csk, skb, skb->len,
1740  cxgbi_skcb_flags(skb),
1741  cxgbi_skcb_rx_pdulen(skb));
1742  err = -EIO;
1743  goto skb_done;
1744  }
1745  __skb_unlink(dskb, &csk->receive_queue);
1746 
1747  err = skb_read_pdu_data(conn, skb, dskb, 0);
1748  if (err < 0)
1749  pr_err("data, csk 0x%p, skb 0x%p,%u, "
1750  "f 0x%lx, plen %u, dskb 0x%p,"
1751  "%u.\n",
1752  csk, skb, skb->len,
1753  cxgbi_skcb_flags(skb),
1754  cxgbi_skcb_rx_pdulen(skb),
1755  dskb, dskb->len);
1756  __kfree_skb(dskb);
1757  } else
1758  err = skb_read_pdu_data(conn, skb, skb, 0);
1759  }
1760 skb_done:
1761  __kfree_skb(skb);
1762 
1763  if (err < 0)
1764  break;
1765  }
1766 
1767  log_debug(1 << CXGBI_DBG_PDU_RX, "csk 0x%p, read %u.\n", csk, read);
1768  if (read) {
1769  csk->copied_seq += read;
1770  csk_return_rx_credits(csk, read);
1771  conn->rxdata_octets += read;
1772  }
1773 
1774  if (err < 0) {
1775  pr_info("csk 0x%p, 0x%p, rx failed %d, read %u.\n",
1776  csk, conn, err, read);
1778  }
1779 }
1781 
1782 static int sgl_seek_offset(struct scatterlist *sgl, unsigned int sgcnt,
1783  unsigned int offset, unsigned int *off,
1784  struct scatterlist **sgp)
1785 {
1786  int i;
1787  struct scatterlist *sg;
1788 
1789  for_each_sg(sgl, sg, sgcnt, i) {
1790  if (offset < sg->length) {
1791  *off = offset;
1792  *sgp = sg;
1793  return 0;
1794  }
1795  offset -= sg->length;
1796  }
1797  return -EFAULT;
1798 }
1799 
1800 static int sgl_read_to_frags(struct scatterlist *sg, unsigned int sgoffset,
1801  unsigned int dlen, struct page_frag *frags,
1802  int frag_max)
1803 {
1804  unsigned int datalen = dlen;
1805  unsigned int sglen = sg->length - sgoffset;
1806  struct page *page = sg_page(sg);
1807  int i;
1808 
1809  i = 0;
1810  do {
1811  unsigned int copy;
1812 
1813  if (!sglen) {
1814  sg = sg_next(sg);
1815  if (!sg) {
1816  pr_warn("sg %d NULL, len %u/%u.\n",
1817  i, datalen, dlen);
1818  return -EINVAL;
1819  }
1820  sgoffset = 0;
1821  sglen = sg->length;
1822  page = sg_page(sg);
1823 
1824  }
1825  copy = min(datalen, sglen);
1826  if (i && page == frags[i - 1].page &&
1827  sgoffset + sg->offset ==
1828  frags[i - 1].offset + frags[i - 1].size) {
1829  frags[i - 1].size += copy;
1830  } else {
1831  if (i >= frag_max) {
1832  pr_warn("too many pages %u, dlen %u.\n",
1833  frag_max, dlen);
1834  return -EINVAL;
1835  }
1836 
1837  frags[i].page = page;
1838  frags[i].offset = sg->offset + sgoffset;
1839  frags[i].size = copy;
1840  i++;
1841  }
1842  datalen -= copy;
1843  sgoffset += copy;
1844  sglen -= copy;
1845  } while (datalen);
1846 
1847  return i;
1848 }
1849 
1850 int cxgbi_conn_alloc_pdu(struct iscsi_task *task, u8 opcode)
1851 {
1852  struct iscsi_tcp_conn *tcp_conn = task->conn->dd_data;
1853  struct cxgbi_conn *cconn = tcp_conn->dd_data;
1854  struct cxgbi_device *cdev = cconn->chba->cdev;
1855  struct iscsi_conn *conn = task->conn;
1856  struct iscsi_tcp_task *tcp_task = task->dd_data;
1857  struct cxgbi_task_data *tdata = iscsi_task_cxgbi_data(task);
1858  struct scsi_cmnd *sc = task->sc;
1859  int headroom = SKB_TX_ISCSI_PDU_HEADER_MAX;
1860 
1861  tcp_task->dd_data = tdata;
1862  task->hdr = NULL;
1863 
1864  if (SKB_MAX_HEAD(cdev->skb_tx_rsvd) > (512 * MAX_SKB_FRAGS) &&
1865  (opcode == ISCSI_OP_SCSI_DATA_OUT ||
1866  (opcode == ISCSI_OP_SCSI_CMD &&
1867  (scsi_bidi_cmnd(sc) || sc->sc_data_direction == DMA_TO_DEVICE))))
1868  /* data could goes into skb head */
1869  headroom += min_t(unsigned int,
1870  SKB_MAX_HEAD(cdev->skb_tx_rsvd),
1871  conn->max_xmit_dlength);
1872 
1873  tdata->skb = alloc_skb(cdev->skb_tx_rsvd + headroom, GFP_ATOMIC);
1874  if (!tdata->skb) {
1875  struct cxgbi_sock *csk = cconn->cep->csk;
1876  struct net_device *ndev = cdev->ports[csk->port_id];
1877  ndev->stats.tx_dropped++;
1878  return -ENOMEM;
1879  }
1880 
1881  skb_reserve(tdata->skb, cdev->skb_tx_rsvd);
1882  task->hdr = (struct iscsi_hdr *)tdata->skb->data;
1883  task->hdr_max = SKB_TX_ISCSI_PDU_HEADER_MAX; /* BHS + AHS */
1884 
1885  /* data_out uses scsi_cmd's itt */
1886  if (opcode != ISCSI_OP_SCSI_DATA_OUT)
1887  task_reserve_itt(task, &task->hdr->itt);
1888 
1890  "task 0x%p, op 0x%x, skb 0x%p,%u+%u/%u, itt 0x%x.\n",
1891  task, opcode, tdata->skb, cdev->skb_tx_rsvd, headroom,
1892  conn->max_xmit_dlength, ntohl(task->hdr->itt));
1893 
1894  return 0;
1895 }
1897 
1898 static inline void tx_skb_setmode(struct sk_buff *skb, int hcrc, int dcrc)
1899 {
1900  if (hcrc || dcrc) {
1901  u8 submode = 0;
1902 
1903  if (hcrc)
1904  submode |= 1;
1905  if (dcrc)
1906  submode |= 2;
1907  cxgbi_skcb_ulp_mode(skb) = (ULP2_MODE_ISCSI << 4) | submode;
1908  } else
1909  cxgbi_skcb_ulp_mode(skb) = 0;
1910 }
1911 
1912 int cxgbi_conn_init_pdu(struct iscsi_task *task, unsigned int offset,
1913  unsigned int count)
1914 {
1915  struct iscsi_conn *conn = task->conn;
1916  struct cxgbi_task_data *tdata = iscsi_task_cxgbi_data(task);
1917  struct sk_buff *skb = tdata->skb;
1918  unsigned int datalen = count;
1919  int i, padlen = iscsi_padding(count);
1920  struct page *pg;
1921 
1923  "task 0x%p,0x%p, skb 0x%p, 0x%x,0x%x,0x%x, %u+%u.\n",
1924  task, task->sc, skb, (*skb->data) & ISCSI_OPCODE_MASK,
1925  ntohl(task->cmdsn), ntohl(task->hdr->itt), offset, count);
1926 
1927  skb_put(skb, task->hdr_len);
1928  tx_skb_setmode(skb, conn->hdrdgst_en, datalen ? conn->datadgst_en : 0);
1929  if (!count)
1930  return 0;
1931 
1932  if (task->sc) {
1933  struct scsi_data_buffer *sdb = scsi_out(task->sc);
1934  struct scatterlist *sg = NULL;
1935  int err;
1936 
1937  tdata->offset = offset;
1938  tdata->count = count;
1939  err = sgl_seek_offset(
1940  sdb->table.sgl, sdb->table.nents,
1941  tdata->offset, &tdata->sgoffset, &sg);
1942  if (err < 0) {
1943  pr_warn("tpdu, sgl %u, bad offset %u/%u.\n",
1944  sdb->table.nents, tdata->offset, sdb->length);
1945  return err;
1946  }
1947  err = sgl_read_to_frags(sg, tdata->sgoffset, tdata->count,
1948  tdata->frags, MAX_PDU_FRAGS);
1949  if (err < 0) {
1950  pr_warn("tpdu, sgl %u, bad offset %u + %u.\n",
1951  sdb->table.nents, tdata->offset, tdata->count);
1952  return err;
1953  }
1954  tdata->nr_frags = err;
1955 
1956  if (tdata->nr_frags > MAX_SKB_FRAGS ||
1957  (padlen && tdata->nr_frags == MAX_SKB_FRAGS)) {
1958  char *dst = skb->data + task->hdr_len;
1959  struct page_frag *frag = tdata->frags;
1960 
1961  /* data fits in the skb's headroom */
1962  for (i = 0; i < tdata->nr_frags; i++, frag++) {
1963  char *src = kmap_atomic(frag->page);
1964 
1965  memcpy(dst, src+frag->offset, frag->size);
1966  dst += frag->size;
1967  kunmap_atomic(src);
1968  }
1969  if (padlen) {
1970  memset(dst, 0, padlen);
1971  padlen = 0;
1972  }
1973  skb_put(skb, count + padlen);
1974  } else {
1975  /* data fit into frag_list */
1976  for (i = 0; i < tdata->nr_frags; i++) {
1977  __skb_fill_page_desc(skb, i,
1978  tdata->frags[i].page,
1979  tdata->frags[i].offset,
1980  tdata->frags[i].size);
1981  skb_frag_ref(skb, i);
1982  }
1983  skb_shinfo(skb)->nr_frags = tdata->nr_frags;
1984  skb->len += count;
1985  skb->data_len += count;
1986  skb->truesize += count;
1987  }
1988 
1989  } else {
1990  pg = virt_to_page(task->data);
1991 
1992  get_page(pg);
1993  skb_fill_page_desc(skb, 0, pg, offset_in_page(task->data),
1994  count);
1995  skb->len += count;
1996  skb->data_len += count;
1997  skb->truesize += count;
1998  }
1999 
2000  if (padlen) {
2001  i = skb_shinfo(skb)->nr_frags;
2002  skb_fill_page_desc(skb, skb_shinfo(skb)->nr_frags,
2003  virt_to_page(padding), offset_in_page(padding),
2004  padlen);
2005 
2006  skb->data_len += padlen;
2007  skb->truesize += padlen;
2008  skb->len += padlen;
2009  }
2010 
2011  return 0;
2012 }
2014 
2016 {
2017  struct iscsi_tcp_conn *tcp_conn = task->conn->dd_data;
2018  struct cxgbi_conn *cconn = tcp_conn->dd_data;
2019  struct cxgbi_task_data *tdata = iscsi_task_cxgbi_data(task);
2020  struct sk_buff *skb = tdata->skb;
2021  unsigned int datalen;
2022  int err;
2023 
2024  if (!skb) {
2026  "task 0x%p, skb NULL.\n", task);
2027  return 0;
2028  }
2029 
2030  datalen = skb->data_len;
2031  tdata->skb = NULL;
2032  err = cxgbi_sock_send_pdus(cconn->cep->csk, skb);
2033  if (err > 0) {
2034  int pdulen = err;
2035 
2037  "task 0x%p,0x%p, skb 0x%p, len %u/%u, rv %d.\n",
2038  task, task->sc, skb, skb->len, skb->data_len, err);
2039 
2040  if (task->conn->hdrdgst_en)
2041  pdulen += ISCSI_DIGEST_SIZE;
2042 
2043  if (datalen && task->conn->datadgst_en)
2044  pdulen += ISCSI_DIGEST_SIZE;
2045 
2046  task->conn->txdata_octets += pdulen;
2047  return 0;
2048  }
2049 
2050  if (err == -EAGAIN || err == -ENOBUFS) {
2052  "task 0x%p, skb 0x%p, len %u/%u, %d EAGAIN.\n",
2053  task, skb, skb->len, skb->data_len, err);
2054  /* reset skb to send when we are called again */
2055  tdata->skb = skb;
2056  return err;
2057  }
2058 
2059  kfree_skb(skb);
2061  "itt 0x%x, skb 0x%p, len %u/%u, xmit err %d.\n",
2062  task->itt, skb, skb->len, skb->data_len, err);
2063  iscsi_conn_printk(KERN_ERR, task->conn, "xmit err %d.\n", err);
2065  return err;
2066 }
2068 
2069 void cxgbi_cleanup_task(struct iscsi_task *task)
2070 {
2071  struct cxgbi_task_data *tdata = iscsi_task_cxgbi_data(task);
2072 
2074  "task 0x%p, skb 0x%p, itt 0x%x.\n",
2075  task, tdata->skb, task->hdr_itt);
2076 
2077  /* never reached the xmit task callout */
2078  if (tdata->skb)
2079  __kfree_skb(tdata->skb);
2080  memset(tdata, 0, sizeof(*tdata));
2081 
2082  task_release_itt(task, task->hdr_itt);
2083  iscsi_tcp_cleanup_task(task);
2084 }
2086 
2087 void cxgbi_get_conn_stats(struct iscsi_cls_conn *cls_conn,
2088  struct iscsi_stats *stats)
2089 {
2090  struct iscsi_conn *conn = cls_conn->dd_data;
2091 
2092  stats->txdata_octets = conn->txdata_octets;
2093  stats->rxdata_octets = conn->rxdata_octets;
2094  stats->scsicmd_pdus = conn->scsicmd_pdus_cnt;
2095  stats->dataout_pdus = conn->dataout_pdus_cnt;
2096  stats->scsirsp_pdus = conn->scsirsp_pdus_cnt;
2097  stats->datain_pdus = conn->datain_pdus_cnt;
2098  stats->r2t_pdus = conn->r2t_pdus_cnt;
2099  stats->tmfcmd_pdus = conn->tmfcmd_pdus_cnt;
2100  stats->tmfrsp_pdus = conn->tmfrsp_pdus_cnt;
2101  stats->digest_err = 0;
2102  stats->timeout_err = 0;
2103  stats->custom_length = 1;
2104  strcpy(stats->custom[0].desc, "eh_abort_cnt");
2105  stats->custom[0].value = conn->eh_abort_cnt;
2106 }
2108 
2109 static int cxgbi_conn_max_xmit_dlength(struct iscsi_conn *conn)
2110 {
2111  struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
2112  struct cxgbi_conn *cconn = tcp_conn->dd_data;
2113  struct cxgbi_device *cdev = cconn->chba->cdev;
2114  unsigned int headroom = SKB_MAX_HEAD(cdev->skb_tx_rsvd);
2115  unsigned int max_def = 512 * MAX_SKB_FRAGS;
2116  unsigned int max = max(max_def, headroom);
2117 
2118  max = min(cconn->chba->cdev->tx_max_size, max);
2119  if (conn->max_xmit_dlength)
2120  conn->max_xmit_dlength = min(conn->max_xmit_dlength, max);
2121  else
2122  conn->max_xmit_dlength = max;
2124 
2125  return 0;
2126 }
2127 
2128 static int cxgbi_conn_max_recv_dlength(struct iscsi_conn *conn)
2129 {
2130  struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
2131  struct cxgbi_conn *cconn = tcp_conn->dd_data;
2132  unsigned int max = cconn->chba->cdev->rx_max_size;
2133 
2134  cxgbi_align_pdu_size(max);
2135 
2136  if (conn->max_recv_dlength) {
2137  if (conn->max_recv_dlength > max) {
2138  pr_err("MaxRecvDataSegmentLength %u > %u.\n",
2139  conn->max_recv_dlength, max);
2140  return -EINVAL;
2141  }
2142  conn->max_recv_dlength = min(conn->max_recv_dlength, max);
2144  } else
2145  conn->max_recv_dlength = max;
2146 
2147  return 0;
2148 }
2149 
2151  enum iscsi_param param, char *buf, int buflen)
2152 {
2153  struct iscsi_conn *conn = cls_conn->dd_data;
2154  struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
2155  struct cxgbi_conn *cconn = tcp_conn->dd_data;
2156  struct cxgbi_sock *csk = cconn->cep->csk;
2157  int err;
2158 
2160  "cls_conn 0x%p, param %d, buf(%d) %s.\n",
2161  cls_conn, param, buflen, buf);
2162 
2163  switch (param) {
2165  err = iscsi_set_param(cls_conn, param, buf, buflen);
2166  if (!err && conn->hdrdgst_en)
2167  err = csk->cdev->csk_ddp_setup_digest(csk, csk->tid,
2168  conn->hdrdgst_en,
2169  conn->datadgst_en, 0);
2170  break;
2172  err = iscsi_set_param(cls_conn, param, buf, buflen);
2173  if (!err && conn->datadgst_en)
2174  err = csk->cdev->csk_ddp_setup_digest(csk, csk->tid,
2175  conn->hdrdgst_en,
2176  conn->datadgst_en, 0);
2177  break;
2178  case ISCSI_PARAM_MAX_R2T:
2179  return iscsi_tcp_set_max_r2t(conn, buf);
2181  err = iscsi_set_param(cls_conn, param, buf, buflen);
2182  if (!err)
2183  err = cxgbi_conn_max_recv_dlength(conn);
2184  break;
2186  err = iscsi_set_param(cls_conn, param, buf, buflen);
2187  if (!err)
2188  err = cxgbi_conn_max_xmit_dlength(conn);
2189  break;
2190  default:
2191  return iscsi_set_param(cls_conn, param, buf, buflen);
2192  }
2193  return err;
2194 }
2196 
2198  char *buf)
2199 {
2200  struct cxgbi_endpoint *cep = ep->dd_data;
2201  struct cxgbi_sock *csk;
2202  int len;
2203 
2205  "cls_conn 0x%p, param %d.\n", ep, param);
2206 
2207  switch (param) {
2208  case ISCSI_PARAM_CONN_PORT:
2210  if (!cep)
2211  return -ENOTCONN;
2212 
2213  csk = cep->csk;
2214  if (!csk)
2215  return -ENOTCONN;
2216 
2218  &csk->daddr, param, buf);
2219  default:
2220  return -ENOSYS;
2221  }
2222  return len;
2223 }
2225 
2226 struct iscsi_cls_conn *
2228 {
2229  struct iscsi_cls_conn *cls_conn;
2230  struct iscsi_conn *conn;
2231  struct iscsi_tcp_conn *tcp_conn;
2232  struct cxgbi_conn *cconn;
2233 
2234  cls_conn = iscsi_tcp_conn_setup(cls_session, sizeof(*cconn), cid);
2235  if (!cls_conn)
2236  return NULL;
2237 
2238  conn = cls_conn->dd_data;
2239  tcp_conn = conn->dd_data;
2240  cconn = tcp_conn->dd_data;
2241  cconn->iconn = conn;
2242 
2244  "cid %u(0x%x), cls 0x%p,0x%p, conn 0x%p,0x%p,0x%p.\n",
2245  cid, cid, cls_session, cls_conn, conn, tcp_conn, cconn);
2246 
2247  return cls_conn;
2248 }
2250 
2251 int cxgbi_bind_conn(struct iscsi_cls_session *cls_session,
2252  struct iscsi_cls_conn *cls_conn,
2254 {
2255  struct iscsi_conn *conn = cls_conn->dd_data;
2256  struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
2257  struct cxgbi_conn *cconn = tcp_conn->dd_data;
2258  struct iscsi_endpoint *ep;
2259  struct cxgbi_endpoint *cep;
2260  struct cxgbi_sock *csk;
2261  int err;
2262 
2263  ep = iscsi_lookup_endpoint(transport_eph);
2264  if (!ep)
2265  return -EINVAL;
2266 
2267  /* setup ddp pagesize */
2268  cep = ep->dd_data;
2269  csk = cep->csk;
2270  err = csk->cdev->csk_ddp_setup_pgidx(csk, csk->tid, page_idx, 0);
2271  if (err < 0)
2272  return err;
2273 
2274  err = iscsi_conn_bind(cls_session, cls_conn, is_leading);
2275  if (err)
2276  return -EINVAL;
2277 
2278  /* calculate the tag idx bits needed for this conn based on cmds_max */
2279  cconn->task_idx_bits = (__ilog2_u32(conn->session->cmds_max - 1)) + 1;
2280 
2282  csk->user_data = conn;
2283  cconn->chba = cep->chba;
2284  cconn->cep = cep;
2285  cep->cconn = cconn;
2287 
2288  cxgbi_conn_max_xmit_dlength(conn);
2289  cxgbi_conn_max_recv_dlength(conn);
2290 
2292  "cls 0x%p,0x%p, ep 0x%p, cconn 0x%p, csk 0x%p.\n",
2293  cls_session, cls_conn, ep, cconn, csk);
2294  /* init recv engine */
2295  iscsi_tcp_hdr_recv_prep(tcp_conn);
2296 
2297  return 0;
2298 }
2300 
2302  u16 cmds_max, u16 qdepth,
2304 {
2305  struct cxgbi_endpoint *cep;
2306  struct cxgbi_hba *chba;
2307  struct Scsi_Host *shost;
2308  struct iscsi_cls_session *cls_session;
2309  struct iscsi_session *session;
2310 
2311  if (!ep) {
2312  pr_err("missing endpoint.\n");
2313  return NULL;
2314  }
2315 
2316  cep = ep->dd_data;
2317  chba = cep->chba;
2318  shost = chba->shost;
2319 
2320  BUG_ON(chba != iscsi_host_priv(shost));
2321 
2322  cls_session = iscsi_session_setup(chba->cdev->itp, shost,
2323  cmds_max, 0,
2324  sizeof(struct iscsi_tcp_task) +
2325  sizeof(struct cxgbi_task_data),
2326  initial_cmdsn, ISCSI_MAX_TARGET);
2327  if (!cls_session)
2328  return NULL;
2329 
2330  session = cls_session->dd_data;
2331  if (iscsi_tcp_r2tpool_alloc(session))
2332  goto remove_session;
2333 
2335  "ep 0x%p, cls sess 0x%p.\n", ep, cls_session);
2336  return cls_session;
2337 
2338 remove_session:
2339  iscsi_session_teardown(cls_session);
2340  return NULL;
2341 }
2343 
2345 {
2347  "cls sess 0x%p.\n", cls_session);
2348 
2349  iscsi_tcp_r2tpool_free(cls_session->dd_data);
2350  iscsi_session_teardown(cls_session);
2351 }
2353 
2355  char *buf, int buflen)
2356 {
2357  struct cxgbi_hba *chba = iscsi_host_priv(shost);
2358 
2359  if (!chba->ndev) {
2360  shost_printk(KERN_ERR, shost, "Could not get host param. "
2361  "netdev for host not set.\n");
2362  return -ENODEV;
2363  }
2364 
2366  "shost 0x%p, hba 0x%p,%s, param %d, buf(%d) %s.\n",
2367  shost, chba, chba->ndev->name, param, buflen, buf);
2368 
2369  switch (param) {
2371  {
2372  __be32 addr = in_aton(buf);
2374  "hba %s, req. ipv4 %pI4.\n", chba->ndev->name, &addr);
2375  cxgbi_set_iscsi_ipv4(chba, addr);
2376  return 0;
2377  }
2380  return 0;
2381  default:
2382  return iscsi_host_set_param(shost, param, buf, buflen);
2383  }
2384 }
2386 
2388  char *buf)
2389 {
2390  struct cxgbi_hba *chba = iscsi_host_priv(shost);
2391  int len = 0;
2392 
2393  if (!chba->ndev) {
2394  shost_printk(KERN_ERR, shost, "Could not get host param. "
2395  "netdev for host not set.\n");
2396  return -ENODEV;
2397  }
2398 
2400  "shost 0x%p, hba 0x%p,%s, param %d.\n",
2401  shost, chba, chba->ndev->name, param);
2402 
2403  switch (param) {
2405  len = sysfs_format_mac(buf, chba->ndev->dev_addr, 6);
2406  break;
2408  len = sprintf(buf, "%s\n", chba->ndev->name);
2409  break;
2411  {
2412  __be32 addr;
2413 
2414  addr = cxgbi_get_iscsi_ipv4(chba);
2415  len = sprintf(buf, "%pI4", &addr);
2417  "hba %s, ipv4 %pI4.\n", chba->ndev->name, &addr);
2418  break;
2419  }
2420  default:
2421  return iscsi_host_get_param(shost, param, buf);
2422  }
2423 
2424  return len;
2425 }
2427 
2429  struct sockaddr *dst_addr,
2430  int non_blocking)
2431 {
2432  struct iscsi_endpoint *ep;
2433  struct cxgbi_endpoint *cep;
2434  struct cxgbi_hba *hba = NULL;
2435  struct cxgbi_sock *csk;
2436  int err = -EINVAL;
2437 
2439  "shost 0x%p, non_blocking %d, dst_addr 0x%p.\n",
2440  shost, non_blocking, dst_addr);
2441 
2442  if (shost) {
2443  hba = iscsi_host_priv(shost);
2444  if (!hba) {
2445  pr_info("shost 0x%p, priv NULL.\n", shost);
2446  goto err_out;
2447  }
2448  }
2449 
2450  csk = cxgbi_check_route(dst_addr);
2451  if (IS_ERR(csk))
2452  return (struct iscsi_endpoint *)csk;
2453  cxgbi_sock_get(csk);
2454 
2455  if (!hba)
2456  hba = csk->cdev->hbas[csk->port_id];
2457  else if (hba != csk->cdev->hbas[csk->port_id]) {
2458  pr_info("Could not connect through requested host %u"
2459  "hba 0x%p != 0x%p (%u).\n",
2460  shost->host_no, hba,
2461  csk->cdev->hbas[csk->port_id], csk->port_id);
2462  err = -ENOSPC;
2463  goto release_conn;
2464  }
2465 
2466  err = sock_get_port(csk);
2467  if (err)
2468  goto release_conn;
2469 
2470  cxgbi_sock_set_state(csk, CTP_CONNECTING);
2471  err = csk->cdev->csk_init_act_open(csk);
2472  if (err)
2473  goto release_conn;
2474 
2475  if (cxgbi_sock_is_closing(csk)) {
2476  err = -ENOSPC;
2477  pr_info("csk 0x%p is closing.\n", csk);
2478  goto release_conn;
2479  }
2480 
2481  ep = iscsi_create_endpoint(sizeof(*cep));
2482  if (!ep) {
2483  err = -ENOMEM;
2484  pr_info("iscsi alloc ep, OOM.\n");
2485  goto release_conn;
2486  }
2487 
2488  cep = ep->dd_data;
2489  cep->csk = csk;
2490  cep->chba = hba;
2491 
2493  "ep 0x%p, cep 0x%p, csk 0x%p, hba 0x%p,%s.\n",
2494  ep, cep, csk, hba, hba->ndev->name);
2495  return ep;
2496 
2497 release_conn:
2498  cxgbi_sock_put(csk);
2499  cxgbi_sock_closed(csk);
2500 err_out:
2501  return ERR_PTR(err);
2502 }
2504 
2506 {
2507  struct cxgbi_endpoint *cep = ep->dd_data;
2508  struct cxgbi_sock *csk = cep->csk;
2509 
2510  if (!cxgbi_sock_is_established(csk))
2511  return 0;
2512  return 1;
2513 }
2515 
2517 {
2518  struct cxgbi_endpoint *cep = ep->dd_data;
2519  struct cxgbi_conn *cconn = cep->cconn;
2520  struct cxgbi_sock *csk = cep->csk;
2521 
2523  "ep 0x%p, cep 0x%p, cconn 0x%p, csk 0x%p,%u,0x%lx.\n",
2524  ep, cep, cconn, csk, csk->state, csk->flags);
2525 
2526  if (cconn && cconn->iconn) {
2527  iscsi_suspend_tx(cconn->iconn);
2529  cep->csk->user_data = NULL;
2530  cconn->cep = NULL;
2532  }
2534 
2535  if (likely(csk->state >= CTP_ESTABLISHED))
2536  need_active_close(csk);
2537  else
2538  cxgbi_sock_closed(csk);
2539 
2540  cxgbi_sock_put(csk);
2541 }
2543 
2545  struct scsi_transport_template **stt)
2546 {
2547  *stt = iscsi_register_transport(itp);
2548  if (*stt == NULL) {
2549  pr_err("unable to register %s transport 0x%p.\n",
2550  itp->name, itp);
2551  return -ENODEV;
2552  }
2554  "%s, registered iscsi transport 0x%p.\n",
2555  itp->name, stt);
2556  return 0;
2557 }
2559 
2561  struct scsi_transport_template **stt)
2562 {
2563  if (*stt) {
2565  "de-register transport 0x%p, %s, stt 0x%p.\n",
2566  itp, itp->name, *stt);
2567  *stt = NULL;
2569  }
2570 }
2572 
2574 {
2575  switch (param_type) {
2576  case ISCSI_HOST_PARAM:
2577  switch (param) {
2582  return S_IRUGO;
2583  default:
2584  return 0;
2585  }
2586  case ISCSI_PARAM:
2587  switch (param) {
2593  case ISCSI_PARAM_CONN_PORT:
2597  case ISCSI_PARAM_PING_TMO:
2598  case ISCSI_PARAM_RECV_TMO:
2600  case ISCSI_PARAM_MAX_R2T:
2603  case ISCSI_PARAM_MAX_BURST:
2606  case ISCSI_PARAM_ERL:
2608  case ISCSI_PARAM_TPGT:
2609  case ISCSI_PARAM_USERNAME:
2610  case ISCSI_PARAM_PASSWORD:
2614  case ISCSI_PARAM_ABORT_TMO:
2619  return S_IRUGO;
2620  default:
2621  return 0;
2622  }
2623  }
2624 
2625  return 0;
2626 }
2628 
2629 static int __init libcxgbi_init_module(void)
2630 {
2631  sw_tag_idx_bits = (__ilog2_u32(ISCSI_ITT_MASK)) + 1;
2632  sw_tag_age_bits = (__ilog2_u32(ISCSI_AGE_MASK)) + 1;
2633 
2634  pr_info("tag itt 0x%x, %u bits, age 0x%x, %u bits.\n",
2635  ISCSI_ITT_MASK, sw_tag_idx_bits,
2636  ISCSI_AGE_MASK, sw_tag_age_bits);
2637 
2638  ddp_setup_host_page_size();
2639  return 0;
2640 }
2641 
2642 static void __exit libcxgbi_exit_module(void)
2643 {
2645  return;
2646 }
2647 
2648 module_init(libcxgbi_init_module);
2649 module_exit(libcxgbi_exit_module);