1 /*
2 * Copyright (c) 2006 Chelsio, Inc. All rights reserved.
3 *
4 * This software is available to you under a choice of one of two
5 * licenses. You may choose to be licensed under the terms of the GNU
6 * General Public License (GPL) Version 2, available from the file
7 * COPYING in the main directory of this source tree, or the
8 * OpenIB.org BSD license below:
9 *
10 * Redistribution and use in source and binary forms, with or
11 * without modification, are permitted provided that the following
12 * conditions are met:
13 *
14 * - Redistributions of source code must retain the above
15 * copyright notice, this list of conditions and the following
16 * disclaimer.
17 *
18 * - Redistributions in binary form must reproduce the above
19 * copyright notice, this list of conditions and the following
20 * disclaimer in the documentation and/or other materials
21 * provided with the distribution.
22 *
23 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
24 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
25 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
26 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
27 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
28 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
29 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
30 * SOFTWARE.
31 */
32 #include <linux/module.h>
33 #include <linux/list.h>
34 #include <linux/slab.h>
35 #include <linux/workqueue.h>
36 #include <linux/skbuff.h>
37 #include <linux/timer.h>
38 #include <linux/notifier.h>
39 #include <linux/inetdevice.h>
40
41 #include <net/neighbour.h>
42 #include <net/netevent.h>
43 #include <net/route.h>
44
45 #include "tcb.h"
46 #include "cxgb3_offload.h"
47 #include "iwch.h"
48 #include "iwch_provider.h"
49 #include "iwch_cm.h"
50
51 static char *states[] = {
52 "idle",
53 "listen",
54 "connecting",
55 "mpa_wait_req",
56 "mpa_req_sent",
57 "mpa_req_rcvd",
58 "mpa_rep_sent",
59 "fpdu_mode",
60 "aborting",
61 "closing",
62 "moribund",
63 "dead",
64 NULL,
65 };
66
67 int peer2peer = 0;
68 module_param(peer2peer, int, 0644);
69 MODULE_PARM_DESC(peer2peer, "Support peer2peer ULPs (default=0)");
70
71 static int ep_timeout_secs = 60;
72 module_param(ep_timeout_secs, int, 0644);
73 MODULE_PARM_DESC(ep_timeout_secs, "CM Endpoint operation timeout "
74 "in seconds (default=60)");
75
76 static int mpa_rev = 1;
77 module_param(mpa_rev, int, 0644);
78 MODULE_PARM_DESC(mpa_rev, "MPA Revision, 0 supports amso1100, "
79 "1 is spec compliant. (default=1)");
80
81 static int markers_enabled = 0;
82 module_param(markers_enabled, int, 0644);
83 MODULE_PARM_DESC(markers_enabled, "Enable MPA MARKERS (default(0)=disabled)");
84
85 static int crc_enabled = 1;
86 module_param(crc_enabled, int, 0644);
87 MODULE_PARM_DESC(crc_enabled, "Enable MPA CRC (default(1)=enabled)");
88
89 static int rcv_win = 256 * 1024;
90 module_param(rcv_win, int, 0644);
91 MODULE_PARM_DESC(rcv_win, "TCP receive window in bytes (default=256)");
92
93 static int snd_win = 32 * 1024;
94 module_param(snd_win, int, 0644);
95 MODULE_PARM_DESC(snd_win, "TCP send window in bytes (default=32KB)");
96
97 static unsigned int nocong = 0;
98 module_param(nocong, uint, 0644);
99 MODULE_PARM_DESC(nocong, "Turn off congestion control (default=0)");
100
101 static unsigned int cong_flavor = 1;
102 module_param(cong_flavor, uint, 0644);
103 MODULE_PARM_DESC(cong_flavor, "TCP Congestion control flavor (default=1)");
104
105 static struct workqueue_struct *workq;
106
107 static struct sk_buff_head rxq;
108
109 static struct sk_buff *get_skb(struct sk_buff *skb, int len, gfp_t gfp);
110 static void ep_timeout(unsigned long arg);
111 static void connect_reply_upcall(struct iwch_ep *ep, int status);
112
start_ep_timer(struct iwch_ep * ep)113 static void start_ep_timer(struct iwch_ep *ep)
114 {
115 PDBG("%s ep %p\n", __func__, ep);
116 if (timer_pending(&ep->timer)) {
117 PDBG("%s stopped / restarted timer ep %p\n", __func__, ep);
118 del_timer_sync(&ep->timer);
119 } else
120 get_ep(&ep->com);
121 ep->timer.expires = jiffies + ep_timeout_secs * HZ;
122 ep->timer.data = (unsigned long)ep;
123 ep->timer.function = ep_timeout;
124 add_timer(&ep->timer);
125 }
126
stop_ep_timer(struct iwch_ep * ep)127 static void stop_ep_timer(struct iwch_ep *ep)
128 {
129 PDBG("%s ep %p\n", __func__, ep);
130 if (!timer_pending(&ep->timer)) {
131 printk(KERN_ERR "%s timer stopped when its not running! ep %p state %u\n",
132 __func__, ep, ep->com.state);
133 WARN_ON(1);
134 return;
135 }
136 del_timer_sync(&ep->timer);
137 put_ep(&ep->com);
138 }
139
iwch_l2t_send(struct t3cdev * tdev,struct sk_buff * skb,struct l2t_entry * l2e)140 static int iwch_l2t_send(struct t3cdev *tdev, struct sk_buff *skb, struct l2t_entry *l2e)
141 {
142 int error = 0;
143 struct cxio_rdev *rdev;
144
145 rdev = (struct cxio_rdev *)tdev->ulp;
146 if (cxio_fatal_error(rdev)) {
147 kfree_skb(skb);
148 return -EIO;
149 }
150 error = l2t_send(tdev, skb, l2e);
151 if (error < 0)
152 kfree_skb(skb);
153 return error;
154 }
155
iwch_cxgb3_ofld_send(struct t3cdev * tdev,struct sk_buff * skb)156 int iwch_cxgb3_ofld_send(struct t3cdev *tdev, struct sk_buff *skb)
157 {
158 int error = 0;
159 struct cxio_rdev *rdev;
160
161 rdev = (struct cxio_rdev *)tdev->ulp;
162 if (cxio_fatal_error(rdev)) {
163 kfree_skb(skb);
164 return -EIO;
165 }
166 error = cxgb3_ofld_send(tdev, skb);
167 if (error < 0)
168 kfree_skb(skb);
169 return error;
170 }
171
release_tid(struct t3cdev * tdev,u32 hwtid,struct sk_buff * skb)172 static void release_tid(struct t3cdev *tdev, u32 hwtid, struct sk_buff *skb)
173 {
174 struct cpl_tid_release *req;
175
176 skb = get_skb(skb, sizeof *req, GFP_KERNEL);
177 if (!skb)
178 return;
179 req = (struct cpl_tid_release *) skb_put(skb, sizeof(*req));
180 req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
181 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_TID_RELEASE, hwtid));
182 skb->priority = CPL_PRIORITY_SETUP;
183 iwch_cxgb3_ofld_send(tdev, skb);
184 return;
185 }
186
iwch_quiesce_tid(struct iwch_ep * ep)187 int iwch_quiesce_tid(struct iwch_ep *ep)
188 {
189 struct cpl_set_tcb_field *req;
190 struct sk_buff *skb = get_skb(NULL, sizeof(*req), GFP_KERNEL);
191
192 if (!skb)
193 return -ENOMEM;
194 req = (struct cpl_set_tcb_field *) skb_put(skb, sizeof(*req));
195 req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
196 req->wr.wr_lo = htonl(V_WR_TID(ep->hwtid));
197 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_SET_TCB_FIELD, ep->hwtid));
198 req->reply = 0;
199 req->cpu_idx = 0;
200 req->word = htons(W_TCB_RX_QUIESCE);
201 req->mask = cpu_to_be64(1ULL << S_TCB_RX_QUIESCE);
202 req->val = cpu_to_be64(1 << S_TCB_RX_QUIESCE);
203
204 skb->priority = CPL_PRIORITY_DATA;
205 return iwch_cxgb3_ofld_send(ep->com.tdev, skb);
206 }
207
iwch_resume_tid(struct iwch_ep * ep)208 int iwch_resume_tid(struct iwch_ep *ep)
209 {
210 struct cpl_set_tcb_field *req;
211 struct sk_buff *skb = get_skb(NULL, sizeof(*req), GFP_KERNEL);
212
213 if (!skb)
214 return -ENOMEM;
215 req = (struct cpl_set_tcb_field *) skb_put(skb, sizeof(*req));
216 req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
217 req->wr.wr_lo = htonl(V_WR_TID(ep->hwtid));
218 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_SET_TCB_FIELD, ep->hwtid));
219 req->reply = 0;
220 req->cpu_idx = 0;
221 req->word = htons(W_TCB_RX_QUIESCE);
222 req->mask = cpu_to_be64(1ULL << S_TCB_RX_QUIESCE);
223 req->val = 0;
224
225 skb->priority = CPL_PRIORITY_DATA;
226 return iwch_cxgb3_ofld_send(ep->com.tdev, skb);
227 }
228
set_emss(struct iwch_ep * ep,u16 opt)229 static void set_emss(struct iwch_ep *ep, u16 opt)
230 {
231 PDBG("%s ep %p opt %u\n", __func__, ep, opt);
232 ep->emss = T3C_DATA(ep->com.tdev)->mtus[G_TCPOPT_MSS(opt)] - 40;
233 if (G_TCPOPT_TSTAMP(opt))
234 ep->emss -= 12;
235 if (ep->emss < 128)
236 ep->emss = 128;
237 PDBG("emss=%d\n", ep->emss);
238 }
239
state_read(struct iwch_ep_common * epc)240 static enum iwch_ep_state state_read(struct iwch_ep_common *epc)
241 {
242 unsigned long flags;
243 enum iwch_ep_state state;
244
245 spin_lock_irqsave(&epc->lock, flags);
246 state = epc->state;
247 spin_unlock_irqrestore(&epc->lock, flags);
248 return state;
249 }
250
__state_set(struct iwch_ep_common * epc,enum iwch_ep_state new)251 static void __state_set(struct iwch_ep_common *epc, enum iwch_ep_state new)
252 {
253 epc->state = new;
254 }
255
state_set(struct iwch_ep_common * epc,enum iwch_ep_state new)256 static void state_set(struct iwch_ep_common *epc, enum iwch_ep_state new)
257 {
258 unsigned long flags;
259
260 spin_lock_irqsave(&epc->lock, flags);
261 PDBG("%s - %s -> %s\n", __func__, states[epc->state], states[new]);
262 __state_set(epc, new);
263 spin_unlock_irqrestore(&epc->lock, flags);
264 return;
265 }
266
alloc_ep(int size,gfp_t gfp)267 static void *alloc_ep(int size, gfp_t gfp)
268 {
269 struct iwch_ep_common *epc;
270
271 epc = kzalloc(size, gfp);
272 if (epc) {
273 kref_init(&epc->kref);
274 spin_lock_init(&epc->lock);
275 init_waitqueue_head(&epc->waitq);
276 }
277 PDBG("%s alloc ep %p\n", __func__, epc);
278 return epc;
279 }
280
__free_ep(struct kref * kref)281 void __free_ep(struct kref *kref)
282 {
283 struct iwch_ep *ep;
284 ep = container_of(container_of(kref, struct iwch_ep_common, kref),
285 struct iwch_ep, com);
286 PDBG("%s ep %p state %s\n", __func__, ep, states[state_read(&ep->com)]);
287 if (test_bit(RELEASE_RESOURCES, &ep->com.flags)) {
288 cxgb3_remove_tid(ep->com.tdev, (void *)ep, ep->hwtid);
289 dst_release(ep->dst);
290 l2t_release(ep->com.tdev, ep->l2t);
291 }
292 kfree(ep);
293 }
294
release_ep_resources(struct iwch_ep * ep)295 static void release_ep_resources(struct iwch_ep *ep)
296 {
297 PDBG("%s ep %p tid %d\n", __func__, ep, ep->hwtid);
298 set_bit(RELEASE_RESOURCES, &ep->com.flags);
299 put_ep(&ep->com);
300 }
301
status2errno(int status)302 static int status2errno(int status)
303 {
304 switch (status) {
305 case CPL_ERR_NONE:
306 return 0;
307 case CPL_ERR_CONN_RESET:
308 return -ECONNRESET;
309 case CPL_ERR_ARP_MISS:
310 return -EHOSTUNREACH;
311 case CPL_ERR_CONN_TIMEDOUT:
312 return -ETIMEDOUT;
313 case CPL_ERR_TCAM_FULL:
314 return -ENOMEM;
315 case CPL_ERR_CONN_EXIST:
316 return -EADDRINUSE;
317 default:
318 return -EIO;
319 }
320 }
321
322 /*
323 * Try and reuse skbs already allocated...
324 */
get_skb(struct sk_buff * skb,int len,gfp_t gfp)325 static struct sk_buff *get_skb(struct sk_buff *skb, int len, gfp_t gfp)
326 {
327 if (skb && !skb_is_nonlinear(skb) && !skb_cloned(skb)) {
328 skb_trim(skb, 0);
329 skb_get(skb);
330 } else {
331 skb = alloc_skb(len, gfp);
332 }
333 return skb;
334 }
335
find_route(struct t3cdev * dev,__be32 local_ip,__be32 peer_ip,__be16 local_port,__be16 peer_port,u8 tos)336 static struct rtable *find_route(struct t3cdev *dev, __be32 local_ip,
337 __be32 peer_ip, __be16 local_port,
338 __be16 peer_port, u8 tos)
339 {
340 struct rtable *rt;
341 struct flowi4 fl4;
342
343 rt = ip_route_output_ports(&init_net, &fl4, NULL, peer_ip, local_ip,
344 peer_port, local_port, IPPROTO_TCP,
345 tos, 0);
346 if (IS_ERR(rt))
347 return NULL;
348 return rt;
349 }
350
find_best_mtu(const struct t3c_data * d,unsigned short mtu)351 static unsigned int find_best_mtu(const struct t3c_data *d, unsigned short mtu)
352 {
353 int i = 0;
354
355 while (i < d->nmtus - 1 && d->mtus[i + 1] <= mtu)
356 ++i;
357 return i;
358 }
359
arp_failure_discard(struct t3cdev * dev,struct sk_buff * skb)360 static void arp_failure_discard(struct t3cdev *dev, struct sk_buff *skb)
361 {
362 PDBG("%s t3cdev %p\n", __func__, dev);
363 kfree_skb(skb);
364 }
365
366 /*
367 * Handle an ARP failure for an active open.
368 */
act_open_req_arp_failure(struct t3cdev * dev,struct sk_buff * skb)369 static void act_open_req_arp_failure(struct t3cdev *dev, struct sk_buff *skb)
370 {
371 printk(KERN_ERR MOD "ARP failure duing connect\n");
372 kfree_skb(skb);
373 }
374
375 /*
376 * Handle an ARP failure for a CPL_ABORT_REQ. Change it into a no RST variant
377 * and send it along.
378 */
abort_arp_failure(struct t3cdev * dev,struct sk_buff * skb)379 static void abort_arp_failure(struct t3cdev *dev, struct sk_buff *skb)
380 {
381 struct cpl_abort_req *req = cplhdr(skb);
382
383 PDBG("%s t3cdev %p\n", __func__, dev);
384 req->cmd = CPL_ABORT_NO_RST;
385 iwch_cxgb3_ofld_send(dev, skb);
386 }
387
send_halfclose(struct iwch_ep * ep,gfp_t gfp)388 static int send_halfclose(struct iwch_ep *ep, gfp_t gfp)
389 {
390 struct cpl_close_con_req *req;
391 struct sk_buff *skb;
392
393 PDBG("%s ep %p\n", __func__, ep);
394 skb = get_skb(NULL, sizeof(*req), gfp);
395 if (!skb) {
396 printk(KERN_ERR MOD "%s - failed to alloc skb\n", __func__);
397 return -ENOMEM;
398 }
399 skb->priority = CPL_PRIORITY_DATA;
400 set_arp_failure_handler(skb, arp_failure_discard);
401 req = (struct cpl_close_con_req *) skb_put(skb, sizeof(*req));
402 req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_OFLD_CLOSE_CON));
403 req->wr.wr_lo = htonl(V_WR_TID(ep->hwtid));
404 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_CLOSE_CON_REQ, ep->hwtid));
405 return iwch_l2t_send(ep->com.tdev, skb, ep->l2t);
406 }
407
send_abort(struct iwch_ep * ep,struct sk_buff * skb,gfp_t gfp)408 static int send_abort(struct iwch_ep *ep, struct sk_buff *skb, gfp_t gfp)
409 {
410 struct cpl_abort_req *req;
411
412 PDBG("%s ep %p\n", __func__, ep);
413 skb = get_skb(skb, sizeof(*req), gfp);
414 if (!skb) {
415 printk(KERN_ERR MOD "%s - failed to alloc skb.\n",
416 __func__);
417 return -ENOMEM;
418 }
419 skb->priority = CPL_PRIORITY_DATA;
420 set_arp_failure_handler(skb, abort_arp_failure);
421 req = (struct cpl_abort_req *) skb_put(skb, sizeof(*req));
422 req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_OFLD_HOST_ABORT_CON_REQ));
423 req->wr.wr_lo = htonl(V_WR_TID(ep->hwtid));
424 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_ABORT_REQ, ep->hwtid));
425 req->cmd = CPL_ABORT_SEND_RST;
426 return iwch_l2t_send(ep->com.tdev, skb, ep->l2t);
427 }
428
send_connect(struct iwch_ep * ep)429 static int send_connect(struct iwch_ep *ep)
430 {
431 struct cpl_act_open_req *req;
432 struct sk_buff *skb;
433 u32 opt0h, opt0l, opt2;
434 unsigned int mtu_idx;
435 int wscale;
436
437 PDBG("%s ep %p\n", __func__, ep);
438
439 skb = get_skb(NULL, sizeof(*req), GFP_KERNEL);
440 if (!skb) {
441 printk(KERN_ERR MOD "%s - failed to alloc skb.\n",
442 __func__);
443 return -ENOMEM;
444 }
445 mtu_idx = find_best_mtu(T3C_DATA(ep->com.tdev), dst_mtu(ep->dst));
446 wscale = compute_wscale(rcv_win);
447 opt0h = V_NAGLE(0) |
448 V_NO_CONG(nocong) |
449 V_KEEP_ALIVE(1) |
450 F_TCAM_BYPASS |
451 V_WND_SCALE(wscale) |
452 V_MSS_IDX(mtu_idx) |
453 V_L2T_IDX(ep->l2t->idx) | V_TX_CHANNEL(ep->l2t->smt_idx);
454 opt0l = V_TOS((ep->tos >> 2) & M_TOS) | V_RCV_BUFSIZ(rcv_win>>10);
455 opt2 = F_RX_COALESCE_VALID | V_RX_COALESCE(0) | V_FLAVORS_VALID(1) |
456 V_CONG_CONTROL_FLAVOR(cong_flavor);
457 skb->priority = CPL_PRIORITY_SETUP;
458 set_arp_failure_handler(skb, act_open_req_arp_failure);
459
460 req = (struct cpl_act_open_req *) skb_put(skb, sizeof(*req));
461 req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
462 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_ACT_OPEN_REQ, ep->atid));
463 req->local_port = ep->com.local_addr.sin_port;
464 req->peer_port = ep->com.remote_addr.sin_port;
465 req->local_ip = ep->com.local_addr.sin_addr.s_addr;
466 req->peer_ip = ep->com.remote_addr.sin_addr.s_addr;
467 req->opt0h = htonl(opt0h);
468 req->opt0l = htonl(opt0l);
469 req->params = 0;
470 req->opt2 = htonl(opt2);
471 return iwch_l2t_send(ep->com.tdev, skb, ep->l2t);
472 }
473
send_mpa_req(struct iwch_ep * ep,struct sk_buff * skb)474 static void send_mpa_req(struct iwch_ep *ep, struct sk_buff *skb)
475 {
476 int mpalen;
477 struct tx_data_wr *req;
478 struct mpa_message *mpa;
479 int len;
480
481 PDBG("%s ep %p pd_len %d\n", __func__, ep, ep->plen);
482
483 BUG_ON(skb_cloned(skb));
484
485 mpalen = sizeof(*mpa) + ep->plen;
486 if (skb->data + mpalen + sizeof(*req) > skb_end_pointer(skb)) {
487 kfree_skb(skb);
488 skb=alloc_skb(mpalen + sizeof(*req), GFP_KERNEL);
489 if (!skb) {
490 connect_reply_upcall(ep, -ENOMEM);
491 return;
492 }
493 }
494 skb_trim(skb, 0);
495 skb_reserve(skb, sizeof(*req));
496 skb_put(skb, mpalen);
497 skb->priority = CPL_PRIORITY_DATA;
498 mpa = (struct mpa_message *) skb->data;
499 memset(mpa, 0, sizeof(*mpa));
500 memcpy(mpa->key, MPA_KEY_REQ, sizeof(mpa->key));
501 mpa->flags = (crc_enabled ? MPA_CRC : 0) |
502 (markers_enabled ? MPA_MARKERS : 0);
503 mpa->private_data_size = htons(ep->plen);
504 mpa->revision = mpa_rev;
505
506 if (ep->plen)
507 memcpy(mpa->private_data, ep->mpa_pkt + sizeof(*mpa), ep->plen);
508
509 /*
510 * Reference the mpa skb. This ensures the data area
511 * will remain in memory until the hw acks the tx.
512 * Function tx_ack() will deref it.
513 */
514 skb_get(skb);
515 set_arp_failure_handler(skb, arp_failure_discard);
516 skb_reset_transport_header(skb);
517 len = skb->len;
518 req = (struct tx_data_wr *) skb_push(skb, sizeof(*req));
519 req->wr_hi = htonl(V_WR_OP(FW_WROPCODE_OFLD_TX_DATA)|F_WR_COMPL);
520 req->wr_lo = htonl(V_WR_TID(ep->hwtid));
521 req->len = htonl(len);
522 req->param = htonl(V_TX_PORT(ep->l2t->smt_idx) |
523 V_TX_SNDBUF(snd_win>>15));
524 req->flags = htonl(F_TX_INIT);
525 req->sndseq = htonl(ep->snd_seq);
526 BUG_ON(ep->mpa_skb);
527 ep->mpa_skb = skb;
528 iwch_l2t_send(ep->com.tdev, skb, ep->l2t);
529 start_ep_timer(ep);
530 state_set(&ep->com, MPA_REQ_SENT);
531 return;
532 }
533
send_mpa_reject(struct iwch_ep * ep,const void * pdata,u8 plen)534 static int send_mpa_reject(struct iwch_ep *ep, const void *pdata, u8 plen)
535 {
536 int mpalen;
537 struct tx_data_wr *req;
538 struct mpa_message *mpa;
539 struct sk_buff *skb;
540
541 PDBG("%s ep %p plen %d\n", __func__, ep, plen);
542
543 mpalen = sizeof(*mpa) + plen;
544
545 skb = get_skb(NULL, mpalen + sizeof(*req), GFP_KERNEL);
546 if (!skb) {
547 printk(KERN_ERR MOD "%s - cannot alloc skb!\n", __func__);
548 return -ENOMEM;
549 }
550 skb_reserve(skb, sizeof(*req));
551 mpa = (struct mpa_message *) skb_put(skb, mpalen);
552 memset(mpa, 0, sizeof(*mpa));
553 memcpy(mpa->key, MPA_KEY_REP, sizeof(mpa->key));
554 mpa->flags = MPA_REJECT;
555 mpa->revision = mpa_rev;
556 mpa->private_data_size = htons(plen);
557 if (plen)
558 memcpy(mpa->private_data, pdata, plen);
559
560 /*
561 * Reference the mpa skb again. This ensures the data area
562 * will remain in memory until the hw acks the tx.
563 * Function tx_ack() will deref it.
564 */
565 skb_get(skb);
566 skb->priority = CPL_PRIORITY_DATA;
567 set_arp_failure_handler(skb, arp_failure_discard);
568 skb_reset_transport_header(skb);
569 req = (struct tx_data_wr *) skb_push(skb, sizeof(*req));
570 req->wr_hi = htonl(V_WR_OP(FW_WROPCODE_OFLD_TX_DATA)|F_WR_COMPL);
571 req->wr_lo = htonl(V_WR_TID(ep->hwtid));
572 req->len = htonl(mpalen);
573 req->param = htonl(V_TX_PORT(ep->l2t->smt_idx) |
574 V_TX_SNDBUF(snd_win>>15));
575 req->flags = htonl(F_TX_INIT);
576 req->sndseq = htonl(ep->snd_seq);
577 BUG_ON(ep->mpa_skb);
578 ep->mpa_skb = skb;
579 return iwch_l2t_send(ep->com.tdev, skb, ep->l2t);
580 }
581
send_mpa_reply(struct iwch_ep * ep,const void * pdata,u8 plen)582 static int send_mpa_reply(struct iwch_ep *ep, const void *pdata, u8 plen)
583 {
584 int mpalen;
585 struct tx_data_wr *req;
586 struct mpa_message *mpa;
587 int len;
588 struct sk_buff *skb;
589
590 PDBG("%s ep %p plen %d\n", __func__, ep, plen);
591
592 mpalen = sizeof(*mpa) + plen;
593
594 skb = get_skb(NULL, mpalen + sizeof(*req), GFP_KERNEL);
595 if (!skb) {
596 printk(KERN_ERR MOD "%s - cannot alloc skb!\n", __func__);
597 return -ENOMEM;
598 }
599 skb->priority = CPL_PRIORITY_DATA;
600 skb_reserve(skb, sizeof(*req));
601 mpa = (struct mpa_message *) skb_put(skb, mpalen);
602 memset(mpa, 0, sizeof(*mpa));
603 memcpy(mpa->key, MPA_KEY_REP, sizeof(mpa->key));
604 mpa->flags = (ep->mpa_attr.crc_enabled ? MPA_CRC : 0) |
605 (markers_enabled ? MPA_MARKERS : 0);
606 mpa->revision = mpa_rev;
607 mpa->private_data_size = htons(plen);
608 if (plen)
609 memcpy(mpa->private_data, pdata, plen);
610
611 /*
612 * Reference the mpa skb. This ensures the data area
613 * will remain in memory until the hw acks the tx.
614 * Function tx_ack() will deref it.
615 */
616 skb_get(skb);
617 set_arp_failure_handler(skb, arp_failure_discard);
618 skb_reset_transport_header(skb);
619 len = skb->len;
620 req = (struct tx_data_wr *) skb_push(skb, sizeof(*req));
621 req->wr_hi = htonl(V_WR_OP(FW_WROPCODE_OFLD_TX_DATA)|F_WR_COMPL);
622 req->wr_lo = htonl(V_WR_TID(ep->hwtid));
623 req->len = htonl(len);
624 req->param = htonl(V_TX_PORT(ep->l2t->smt_idx) |
625 V_TX_SNDBUF(snd_win>>15));
626 req->flags = htonl(F_TX_INIT);
627 req->sndseq = htonl(ep->snd_seq);
628 ep->mpa_skb = skb;
629 state_set(&ep->com, MPA_REP_SENT);
630 return iwch_l2t_send(ep->com.tdev, skb, ep->l2t);
631 }
632
act_establish(struct t3cdev * tdev,struct sk_buff * skb,void * ctx)633 static int act_establish(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
634 {
635 struct iwch_ep *ep = ctx;
636 struct cpl_act_establish *req = cplhdr(skb);
637 unsigned int tid = GET_TID(req);
638
639 PDBG("%s ep %p tid %d\n", __func__, ep, tid);
640
641 dst_confirm(ep->dst);
642
643 /* setup the hwtid for this connection */
644 ep->hwtid = tid;
645 cxgb3_insert_tid(ep->com.tdev, &t3c_client, ep, tid);
646
647 ep->snd_seq = ntohl(req->snd_isn);
648 ep->rcv_seq = ntohl(req->rcv_isn);
649
650 set_emss(ep, ntohs(req->tcp_opt));
651
652 /* dealloc the atid */
653 cxgb3_free_atid(ep->com.tdev, ep->atid);
654
655 /* start MPA negotiation */
656 send_mpa_req(ep, skb);
657
658 return 0;
659 }
660
abort_connection(struct iwch_ep * ep,struct sk_buff * skb,gfp_t gfp)661 static void abort_connection(struct iwch_ep *ep, struct sk_buff *skb, gfp_t gfp)
662 {
663 PDBG("%s ep %p\n", __FILE__, ep);
664 state_set(&ep->com, ABORTING);
665 send_abort(ep, skb, gfp);
666 }
667
close_complete_upcall(struct iwch_ep * ep)668 static void close_complete_upcall(struct iwch_ep *ep)
669 {
670 struct iw_cm_event event;
671
672 PDBG("%s ep %p\n", __func__, ep);
673 memset(&event, 0, sizeof(event));
674 event.event = IW_CM_EVENT_CLOSE;
675 if (ep->com.cm_id) {
676 PDBG("close complete delivered ep %p cm_id %p tid %d\n",
677 ep, ep->com.cm_id, ep->hwtid);
678 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
679 ep->com.cm_id->rem_ref(ep->com.cm_id);
680 ep->com.cm_id = NULL;
681 ep->com.qp = NULL;
682 }
683 }
684
peer_close_upcall(struct iwch_ep * ep)685 static void peer_close_upcall(struct iwch_ep *ep)
686 {
687 struct iw_cm_event event;
688
689 PDBG("%s ep %p\n", __func__, ep);
690 memset(&event, 0, sizeof(event));
691 event.event = IW_CM_EVENT_DISCONNECT;
692 if (ep->com.cm_id) {
693 PDBG("peer close delivered ep %p cm_id %p tid %d\n",
694 ep, ep->com.cm_id, ep->hwtid);
695 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
696 }
697 }
698
peer_abort_upcall(struct iwch_ep * ep)699 static void peer_abort_upcall(struct iwch_ep *ep)
700 {
701 struct iw_cm_event event;
702
703 PDBG("%s ep %p\n", __func__, ep);
704 memset(&event, 0, sizeof(event));
705 event.event = IW_CM_EVENT_CLOSE;
706 event.status = -ECONNRESET;
707 if (ep->com.cm_id) {
708 PDBG("abort delivered ep %p cm_id %p tid %d\n", ep,
709 ep->com.cm_id, ep->hwtid);
710 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
711 ep->com.cm_id->rem_ref(ep->com.cm_id);
712 ep->com.cm_id = NULL;
713 ep->com.qp = NULL;
714 }
715 }
716
connect_reply_upcall(struct iwch_ep * ep,int status)717 static void connect_reply_upcall(struct iwch_ep *ep, int status)
718 {
719 struct iw_cm_event event;
720
721 PDBG("%s ep %p status %d\n", __func__, ep, status);
722 memset(&event, 0, sizeof(event));
723 event.event = IW_CM_EVENT_CONNECT_REPLY;
724 event.status = status;
725 event.local_addr = ep->com.local_addr;
726 event.remote_addr = ep->com.remote_addr;
727
728 if ((status == 0) || (status == -ECONNREFUSED)) {
729 event.private_data_len = ep->plen;
730 event.private_data = ep->mpa_pkt + sizeof(struct mpa_message);
731 }
732 if (ep->com.cm_id) {
733 PDBG("%s ep %p tid %d status %d\n", __func__, ep,
734 ep->hwtid, status);
735 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
736 }
737 if (status < 0) {
738 ep->com.cm_id->rem_ref(ep->com.cm_id);
739 ep->com.cm_id = NULL;
740 ep->com.qp = NULL;
741 }
742 }
743
connect_request_upcall(struct iwch_ep * ep)744 static void connect_request_upcall(struct iwch_ep *ep)
745 {
746 struct iw_cm_event event;
747
748 PDBG("%s ep %p tid %d\n", __func__, ep, ep->hwtid);
749 memset(&event, 0, sizeof(event));
750 event.event = IW_CM_EVENT_CONNECT_REQUEST;
751 event.local_addr = ep->com.local_addr;
752 event.remote_addr = ep->com.remote_addr;
753 event.private_data_len = ep->plen;
754 event.private_data = ep->mpa_pkt + sizeof(struct mpa_message);
755 event.provider_data = ep;
756 /*
757 * Until ird/ord negotiation via MPAv2 support is added, send max
758 * supported values
759 */
760 event.ird = event.ord = 8;
761 if (state_read(&ep->parent_ep->com) != DEAD) {
762 get_ep(&ep->com);
763 ep->parent_ep->com.cm_id->event_handler(
764 ep->parent_ep->com.cm_id,
765 &event);
766 }
767 put_ep(&ep->parent_ep->com);
768 ep->parent_ep = NULL;
769 }
770
established_upcall(struct iwch_ep * ep)771 static void established_upcall(struct iwch_ep *ep)
772 {
773 struct iw_cm_event event;
774
775 PDBG("%s ep %p\n", __func__, ep);
776 memset(&event, 0, sizeof(event));
777 event.event = IW_CM_EVENT_ESTABLISHED;
778 /*
779 * Until ird/ord negotiation via MPAv2 support is added, send max
780 * supported values
781 */
782 event.ird = event.ord = 8;
783 if (ep->com.cm_id) {
784 PDBG("%s ep %p tid %d\n", __func__, ep, ep->hwtid);
785 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
786 }
787 }
788
update_rx_credits(struct iwch_ep * ep,u32 credits)789 static int update_rx_credits(struct iwch_ep *ep, u32 credits)
790 {
791 struct cpl_rx_data_ack *req;
792 struct sk_buff *skb;
793
794 PDBG("%s ep %p credits %u\n", __func__, ep, credits);
795 skb = get_skb(NULL, sizeof(*req), GFP_KERNEL);
796 if (!skb) {
797 printk(KERN_ERR MOD "update_rx_credits - cannot alloc skb!\n");
798 return 0;
799 }
800
801 req = (struct cpl_rx_data_ack *) skb_put(skb, sizeof(*req));
802 req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
803 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_RX_DATA_ACK, ep->hwtid));
804 req->credit_dack = htonl(V_RX_CREDITS(credits) | V_RX_FORCE_ACK(1));
805 skb->priority = CPL_PRIORITY_ACK;
806 iwch_cxgb3_ofld_send(ep->com.tdev, skb);
807 return credits;
808 }
809
process_mpa_reply(struct iwch_ep * ep,struct sk_buff * skb)810 static void process_mpa_reply(struct iwch_ep *ep, struct sk_buff *skb)
811 {
812 struct mpa_message *mpa;
813 u16 plen;
814 struct iwch_qp_attributes attrs;
815 enum iwch_qp_attr_mask mask;
816 int err;
817
818 PDBG("%s ep %p\n", __func__, ep);
819
820 /*
821 * Stop mpa timer. If it expired, then the state has
822 * changed and we bail since ep_timeout already aborted
823 * the connection.
824 */
825 stop_ep_timer(ep);
826 if (state_read(&ep->com) != MPA_REQ_SENT)
827 return;
828
829 /*
830 * If we get more than the supported amount of private data
831 * then we must fail this connection.
832 */
833 if (ep->mpa_pkt_len + skb->len > sizeof(ep->mpa_pkt)) {
834 err = -EINVAL;
835 goto err;
836 }
837
838 /*
839 * copy the new data into our accumulation buffer.
840 */
841 skb_copy_from_linear_data(skb, &(ep->mpa_pkt[ep->mpa_pkt_len]),
842 skb->len);
843 ep->mpa_pkt_len += skb->len;
844
845 /*
846 * if we don't even have the mpa message, then bail.
847 */
848 if (ep->mpa_pkt_len < sizeof(*mpa))
849 return;
850 mpa = (struct mpa_message *) ep->mpa_pkt;
851
852 /* Validate MPA header. */
853 if (mpa->revision != mpa_rev) {
854 err = -EPROTO;
855 goto err;
856 }
857 if (memcmp(mpa->key, MPA_KEY_REP, sizeof(mpa->key))) {
858 err = -EPROTO;
859 goto err;
860 }
861
862 plen = ntohs(mpa->private_data_size);
863
864 /*
865 * Fail if there's too much private data.
866 */
867 if (plen > MPA_MAX_PRIVATE_DATA) {
868 err = -EPROTO;
869 goto err;
870 }
871
872 /*
873 * If plen does not account for pkt size
874 */
875 if (ep->mpa_pkt_len > (sizeof(*mpa) + plen)) {
876 err = -EPROTO;
877 goto err;
878 }
879
880 ep->plen = (u8) plen;
881
882 /*
883 * If we don't have all the pdata yet, then bail.
884 * We'll continue process when more data arrives.
885 */
886 if (ep->mpa_pkt_len < (sizeof(*mpa) + plen))
887 return;
888
889 if (mpa->flags & MPA_REJECT) {
890 err = -ECONNREFUSED;
891 goto err;
892 }
893
894 /*
895 * If we get here we have accumulated the entire mpa
896 * start reply message including private data. And
897 * the MPA header is valid.
898 */
899 state_set(&ep->com, FPDU_MODE);
900 ep->mpa_attr.initiator = 1;
901 ep->mpa_attr.crc_enabled = (mpa->flags & MPA_CRC) | crc_enabled ? 1 : 0;
902 ep->mpa_attr.recv_marker_enabled = markers_enabled;
903 ep->mpa_attr.xmit_marker_enabled = mpa->flags & MPA_MARKERS ? 1 : 0;
904 ep->mpa_attr.version = mpa_rev;
905 PDBG("%s - crc_enabled=%d, recv_marker_enabled=%d, "
906 "xmit_marker_enabled=%d, version=%d\n", __func__,
907 ep->mpa_attr.crc_enabled, ep->mpa_attr.recv_marker_enabled,
908 ep->mpa_attr.xmit_marker_enabled, ep->mpa_attr.version);
909
910 attrs.mpa_attr = ep->mpa_attr;
911 attrs.max_ird = ep->ird;
912 attrs.max_ord = ep->ord;
913 attrs.llp_stream_handle = ep;
914 attrs.next_state = IWCH_QP_STATE_RTS;
915
916 mask = IWCH_QP_ATTR_NEXT_STATE |
917 IWCH_QP_ATTR_LLP_STREAM_HANDLE | IWCH_QP_ATTR_MPA_ATTR |
918 IWCH_QP_ATTR_MAX_IRD | IWCH_QP_ATTR_MAX_ORD;
919
920 /* bind QP and TID with INIT_WR */
921 err = iwch_modify_qp(ep->com.qp->rhp,
922 ep->com.qp, mask, &attrs, 1);
923 if (err)
924 goto err;
925
926 if (peer2peer && iwch_rqes_posted(ep->com.qp) == 0) {
927 iwch_post_zb_read(ep);
928 }
929
930 goto out;
931 err:
932 abort_connection(ep, skb, GFP_KERNEL);
933 out:
934 connect_reply_upcall(ep, err);
935 return;
936 }
937
process_mpa_request(struct iwch_ep * ep,struct sk_buff * skb)938 static void process_mpa_request(struct iwch_ep *ep, struct sk_buff *skb)
939 {
940 struct mpa_message *mpa;
941 u16 plen;
942
943 PDBG("%s ep %p\n", __func__, ep);
944
945 /*
946 * Stop mpa timer. If it expired, then the state has
947 * changed and we bail since ep_timeout already aborted
948 * the connection.
949 */
950 stop_ep_timer(ep);
951 if (state_read(&ep->com) != MPA_REQ_WAIT)
952 return;
953
954 /*
955 * If we get more than the supported amount of private data
956 * then we must fail this connection.
957 */
958 if (ep->mpa_pkt_len + skb->len > sizeof(ep->mpa_pkt)) {
959 abort_connection(ep, skb, GFP_KERNEL);
960 return;
961 }
962
963 PDBG("%s enter (%s line %u)\n", __func__, __FILE__, __LINE__);
964
965 /*
966 * Copy the new data into our accumulation buffer.
967 */
968 skb_copy_from_linear_data(skb, &(ep->mpa_pkt[ep->mpa_pkt_len]),
969 skb->len);
970 ep->mpa_pkt_len += skb->len;
971
972 /*
973 * If we don't even have the mpa message, then bail.
974 * We'll continue process when more data arrives.
975 */
976 if (ep->mpa_pkt_len < sizeof(*mpa))
977 return;
978 PDBG("%s enter (%s line %u)\n", __func__, __FILE__, __LINE__);
979 mpa = (struct mpa_message *) ep->mpa_pkt;
980
981 /*
982 * Validate MPA Header.
983 */
984 if (mpa->revision != mpa_rev) {
985 abort_connection(ep, skb, GFP_KERNEL);
986 return;
987 }
988
989 if (memcmp(mpa->key, MPA_KEY_REQ, sizeof(mpa->key))) {
990 abort_connection(ep, skb, GFP_KERNEL);
991 return;
992 }
993
994 plen = ntohs(mpa->private_data_size);
995
996 /*
997 * Fail if there's too much private data.
998 */
999 if (plen > MPA_MAX_PRIVATE_DATA) {
1000 abort_connection(ep, skb, GFP_KERNEL);
1001 return;
1002 }
1003
1004 /*
1005 * If plen does not account for pkt size
1006 */
1007 if (ep->mpa_pkt_len > (sizeof(*mpa) + plen)) {
1008 abort_connection(ep, skb, GFP_KERNEL);
1009 return;
1010 }
1011 ep->plen = (u8) plen;
1012
1013 /*
1014 * If we don't have all the pdata yet, then bail.
1015 */
1016 if (ep->mpa_pkt_len < (sizeof(*mpa) + plen))
1017 return;
1018
1019 /*
1020 * If we get here we have accumulated the entire mpa
1021 * start reply message including private data.
1022 */
1023 ep->mpa_attr.initiator = 0;
1024 ep->mpa_attr.crc_enabled = (mpa->flags & MPA_CRC) | crc_enabled ? 1 : 0;
1025 ep->mpa_attr.recv_marker_enabled = markers_enabled;
1026 ep->mpa_attr.xmit_marker_enabled = mpa->flags & MPA_MARKERS ? 1 : 0;
1027 ep->mpa_attr.version = mpa_rev;
1028 PDBG("%s - crc_enabled=%d, recv_marker_enabled=%d, "
1029 "xmit_marker_enabled=%d, version=%d\n", __func__,
1030 ep->mpa_attr.crc_enabled, ep->mpa_attr.recv_marker_enabled,
1031 ep->mpa_attr.xmit_marker_enabled, ep->mpa_attr.version);
1032
1033 state_set(&ep->com, MPA_REQ_RCVD);
1034
1035 /* drive upcall */
1036 connect_request_upcall(ep);
1037 return;
1038 }
1039
rx_data(struct t3cdev * tdev,struct sk_buff * skb,void * ctx)1040 static int rx_data(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1041 {
1042 struct iwch_ep *ep = ctx;
1043 struct cpl_rx_data *hdr = cplhdr(skb);
1044 unsigned int dlen = ntohs(hdr->len);
1045
1046 PDBG("%s ep %p dlen %u\n", __func__, ep, dlen);
1047
1048 skb_pull(skb, sizeof(*hdr));
1049 skb_trim(skb, dlen);
1050
1051 ep->rcv_seq += dlen;
1052 BUG_ON(ep->rcv_seq != (ntohl(hdr->seq) + dlen));
1053
1054 switch (state_read(&ep->com)) {
1055 case MPA_REQ_SENT:
1056 process_mpa_reply(ep, skb);
1057 break;
1058 case MPA_REQ_WAIT:
1059 process_mpa_request(ep, skb);
1060 break;
1061 case MPA_REP_SENT:
1062 break;
1063 default:
1064 printk(KERN_ERR MOD "%s Unexpected streaming data."
1065 " ep %p state %d tid %d\n",
1066 __func__, ep, state_read(&ep->com), ep->hwtid);
1067
1068 /*
1069 * The ep will timeout and inform the ULP of the failure.
1070 * See ep_timeout().
1071 */
1072 break;
1073 }
1074
1075 /* update RX credits */
1076 update_rx_credits(ep, dlen);
1077
1078 return CPL_RET_BUF_DONE;
1079 }
1080
1081 /*
1082 * Upcall from the adapter indicating data has been transmitted.
1083 * For us its just the single MPA request or reply. We can now free
1084 * the skb holding the mpa message.
1085 */
tx_ack(struct t3cdev * tdev,struct sk_buff * skb,void * ctx)1086 static int tx_ack(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1087 {
1088 struct iwch_ep *ep = ctx;
1089 struct cpl_wr_ack *hdr = cplhdr(skb);
1090 unsigned int credits = ntohs(hdr->credits);
1091 unsigned long flags;
1092 int post_zb = 0;
1093
1094 PDBG("%s ep %p credits %u\n", __func__, ep, credits);
1095
1096 if (credits == 0) {
1097 PDBG("%s 0 credit ack ep %p state %u\n",
1098 __func__, ep, state_read(&ep->com));
1099 return CPL_RET_BUF_DONE;
1100 }
1101
1102 spin_lock_irqsave(&ep->com.lock, flags);
1103 BUG_ON(credits != 1);
1104 dst_confirm(ep->dst);
1105 if (!ep->mpa_skb) {
1106 PDBG("%s rdma_init wr_ack ep %p state %u\n",
1107 __func__, ep, ep->com.state);
1108 if (ep->mpa_attr.initiator) {
1109 PDBG("%s initiator ep %p state %u\n",
1110 __func__, ep, ep->com.state);
1111 if (peer2peer && ep->com.state == FPDU_MODE)
1112 post_zb = 1;
1113 } else {
1114 PDBG("%s responder ep %p state %u\n",
1115 __func__, ep, ep->com.state);
1116 if (ep->com.state == MPA_REQ_RCVD) {
1117 ep->com.rpl_done = 1;
1118 wake_up(&ep->com.waitq);
1119 }
1120 }
1121 } else {
1122 PDBG("%s lsm ack ep %p state %u freeing skb\n",
1123 __func__, ep, ep->com.state);
1124 kfree_skb(ep->mpa_skb);
1125 ep->mpa_skb = NULL;
1126 }
1127 spin_unlock_irqrestore(&ep->com.lock, flags);
1128 if (post_zb)
1129 iwch_post_zb_read(ep);
1130 return CPL_RET_BUF_DONE;
1131 }
1132
abort_rpl(struct t3cdev * tdev,struct sk_buff * skb,void * ctx)1133 static int abort_rpl(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1134 {
1135 struct iwch_ep *ep = ctx;
1136 unsigned long flags;
1137 int release = 0;
1138
1139 PDBG("%s ep %p\n", __func__, ep);
1140 BUG_ON(!ep);
1141
1142 /*
1143 * We get 2 abort replies from the HW. The first one must
1144 * be ignored except for scribbling that we need one more.
1145 */
1146 if (!test_and_set_bit(ABORT_REQ_IN_PROGRESS, &ep->com.flags)) {
1147 return CPL_RET_BUF_DONE;
1148 }
1149
1150 spin_lock_irqsave(&ep->com.lock, flags);
1151 switch (ep->com.state) {
1152 case ABORTING:
1153 close_complete_upcall(ep);
1154 __state_set(&ep->com, DEAD);
1155 release = 1;
1156 break;
1157 default:
1158 printk(KERN_ERR "%s ep %p state %d\n",
1159 __func__, ep, ep->com.state);
1160 break;
1161 }
1162 spin_unlock_irqrestore(&ep->com.lock, flags);
1163
1164 if (release)
1165 release_ep_resources(ep);
1166 return CPL_RET_BUF_DONE;
1167 }
1168
1169 /*
1170 * Return whether a failed active open has allocated a TID
1171 */
act_open_has_tid(int status)1172 static inline int act_open_has_tid(int status)
1173 {
1174 return status != CPL_ERR_TCAM_FULL && status != CPL_ERR_CONN_EXIST &&
1175 status != CPL_ERR_ARP_MISS;
1176 }
1177
act_open_rpl(struct t3cdev * tdev,struct sk_buff * skb,void * ctx)1178 static int act_open_rpl(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1179 {
1180 struct iwch_ep *ep = ctx;
1181 struct cpl_act_open_rpl *rpl = cplhdr(skb);
1182
1183 PDBG("%s ep %p status %u errno %d\n", __func__, ep, rpl->status,
1184 status2errno(rpl->status));
1185 connect_reply_upcall(ep, status2errno(rpl->status));
1186 state_set(&ep->com, DEAD);
1187 if (ep->com.tdev->type != T3A && act_open_has_tid(rpl->status))
1188 release_tid(ep->com.tdev, GET_TID(rpl), NULL);
1189 cxgb3_free_atid(ep->com.tdev, ep->atid);
1190 dst_release(ep->dst);
1191 l2t_release(ep->com.tdev, ep->l2t);
1192 put_ep(&ep->com);
1193 return CPL_RET_BUF_DONE;
1194 }
1195
listen_start(struct iwch_listen_ep * ep)1196 static int listen_start(struct iwch_listen_ep *ep)
1197 {
1198 struct sk_buff *skb;
1199 struct cpl_pass_open_req *req;
1200
1201 PDBG("%s ep %p\n", __func__, ep);
1202 skb = get_skb(NULL, sizeof(*req), GFP_KERNEL);
1203 if (!skb) {
1204 printk(KERN_ERR MOD "t3c_listen_start failed to alloc skb!\n");
1205 return -ENOMEM;
1206 }
1207
1208 req = (struct cpl_pass_open_req *) skb_put(skb, sizeof(*req));
1209 req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
1210 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_PASS_OPEN_REQ, ep->stid));
1211 req->local_port = ep->com.local_addr.sin_port;
1212 req->local_ip = ep->com.local_addr.sin_addr.s_addr;
1213 req->peer_port = 0;
1214 req->peer_ip = 0;
1215 req->peer_netmask = 0;
1216 req->opt0h = htonl(F_DELACK | F_TCAM_BYPASS);
1217 req->opt0l = htonl(V_RCV_BUFSIZ(rcv_win>>10));
1218 req->opt1 = htonl(V_CONN_POLICY(CPL_CONN_POLICY_ASK));
1219
1220 skb->priority = 1;
1221 return iwch_cxgb3_ofld_send(ep->com.tdev, skb);
1222 }
1223
pass_open_rpl(struct t3cdev * tdev,struct sk_buff * skb,void * ctx)1224 static int pass_open_rpl(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1225 {
1226 struct iwch_listen_ep *ep = ctx;
1227 struct cpl_pass_open_rpl *rpl = cplhdr(skb);
1228
1229 PDBG("%s ep %p status %d error %d\n", __func__, ep,
1230 rpl->status, status2errno(rpl->status));
1231 ep->com.rpl_err = status2errno(rpl->status);
1232 ep->com.rpl_done = 1;
1233 wake_up(&ep->com.waitq);
1234
1235 return CPL_RET_BUF_DONE;
1236 }
1237
listen_stop(struct iwch_listen_ep * ep)1238 static int listen_stop(struct iwch_listen_ep *ep)
1239 {
1240 struct sk_buff *skb;
1241 struct cpl_close_listserv_req *req;
1242
1243 PDBG("%s ep %p\n", __func__, ep);
1244 skb = get_skb(NULL, sizeof(*req), GFP_KERNEL);
1245 if (!skb) {
1246 printk(KERN_ERR MOD "%s - failed to alloc skb\n", __func__);
1247 return -ENOMEM;
1248 }
1249 req = (struct cpl_close_listserv_req *) skb_put(skb, sizeof(*req));
1250 req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
1251 req->cpu_idx = 0;
1252 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_CLOSE_LISTSRV_REQ, ep->stid));
1253 skb->priority = 1;
1254 return iwch_cxgb3_ofld_send(ep->com.tdev, skb);
1255 }
1256
close_listsrv_rpl(struct t3cdev * tdev,struct sk_buff * skb,void * ctx)1257 static int close_listsrv_rpl(struct t3cdev *tdev, struct sk_buff *skb,
1258 void *ctx)
1259 {
1260 struct iwch_listen_ep *ep = ctx;
1261 struct cpl_close_listserv_rpl *rpl = cplhdr(skb);
1262
1263 PDBG("%s ep %p\n", __func__, ep);
1264 ep->com.rpl_err = status2errno(rpl->status);
1265 ep->com.rpl_done = 1;
1266 wake_up(&ep->com.waitq);
1267 return CPL_RET_BUF_DONE;
1268 }
1269
accept_cr(struct iwch_ep * ep,__be32 peer_ip,struct sk_buff * skb)1270 static void accept_cr(struct iwch_ep *ep, __be32 peer_ip, struct sk_buff *skb)
1271 {
1272 struct cpl_pass_accept_rpl *rpl;
1273 unsigned int mtu_idx;
1274 u32 opt0h, opt0l, opt2;
1275 int wscale;
1276
1277 PDBG("%s ep %p\n", __func__, ep);
1278 BUG_ON(skb_cloned(skb));
1279 skb_trim(skb, sizeof(*rpl));
1280 skb_get(skb);
1281 mtu_idx = find_best_mtu(T3C_DATA(ep->com.tdev), dst_mtu(ep->dst));
1282 wscale = compute_wscale(rcv_win);
1283 opt0h = V_NAGLE(0) |
1284 V_NO_CONG(nocong) |
1285 V_KEEP_ALIVE(1) |
1286 F_TCAM_BYPASS |
1287 V_WND_SCALE(wscale) |
1288 V_MSS_IDX(mtu_idx) |
1289 V_L2T_IDX(ep->l2t->idx) | V_TX_CHANNEL(ep->l2t->smt_idx);
1290 opt0l = V_TOS((ep->tos >> 2) & M_TOS) | V_RCV_BUFSIZ(rcv_win>>10);
1291 opt2 = F_RX_COALESCE_VALID | V_RX_COALESCE(0) | V_FLAVORS_VALID(1) |
1292 V_CONG_CONTROL_FLAVOR(cong_flavor);
1293
1294 rpl = cplhdr(skb);
1295 rpl->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
1296 OPCODE_TID(rpl) = htonl(MK_OPCODE_TID(CPL_PASS_ACCEPT_RPL, ep->hwtid));
1297 rpl->peer_ip = peer_ip;
1298 rpl->opt0h = htonl(opt0h);
1299 rpl->opt0l_status = htonl(opt0l | CPL_PASS_OPEN_ACCEPT);
1300 rpl->opt2 = htonl(opt2);
1301 rpl->rsvd = rpl->opt2; /* workaround for HW bug */
1302 skb->priority = CPL_PRIORITY_SETUP;
1303 iwch_l2t_send(ep->com.tdev, skb, ep->l2t);
1304
1305 return;
1306 }
1307
reject_cr(struct t3cdev * tdev,u32 hwtid,__be32 peer_ip,struct sk_buff * skb)1308 static void reject_cr(struct t3cdev *tdev, u32 hwtid, __be32 peer_ip,
1309 struct sk_buff *skb)
1310 {
1311 PDBG("%s t3cdev %p tid %u peer_ip %x\n", __func__, tdev, hwtid,
1312 peer_ip);
1313 BUG_ON(skb_cloned(skb));
1314 skb_trim(skb, sizeof(struct cpl_tid_release));
1315 skb_get(skb);
1316
1317 if (tdev->type != T3A)
1318 release_tid(tdev, hwtid, skb);
1319 else {
1320 struct cpl_pass_accept_rpl *rpl;
1321
1322 rpl = cplhdr(skb);
1323 skb->priority = CPL_PRIORITY_SETUP;
1324 rpl->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
1325 OPCODE_TID(rpl) = htonl(MK_OPCODE_TID(CPL_PASS_ACCEPT_RPL,
1326 hwtid));
1327 rpl->peer_ip = peer_ip;
1328 rpl->opt0h = htonl(F_TCAM_BYPASS);
1329 rpl->opt0l_status = htonl(CPL_PASS_OPEN_REJECT);
1330 rpl->opt2 = 0;
1331 rpl->rsvd = rpl->opt2;
1332 iwch_cxgb3_ofld_send(tdev, skb);
1333 }
1334 }
1335
pass_accept_req(struct t3cdev * tdev,struct sk_buff * skb,void * ctx)1336 static int pass_accept_req(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1337 {
1338 struct iwch_ep *child_ep, *parent_ep = ctx;
1339 struct cpl_pass_accept_req *req = cplhdr(skb);
1340 unsigned int hwtid = GET_TID(req);
1341 struct dst_entry *dst;
1342 struct l2t_entry *l2t;
1343 struct rtable *rt;
1344 struct iff_mac tim;
1345
1346 PDBG("%s parent ep %p tid %u\n", __func__, parent_ep, hwtid);
1347
1348 if (state_read(&parent_ep->com) != LISTEN) {
1349 printk(KERN_ERR "%s - listening ep not in LISTEN\n",
1350 __func__);
1351 goto reject;
1352 }
1353
1354 /*
1355 * Find the netdev for this connection request.
1356 */
1357 tim.mac_addr = req->dst_mac;
1358 tim.vlan_tag = ntohs(req->vlan_tag);
1359 if (tdev->ctl(tdev, GET_IFF_FROM_MAC, &tim) < 0 || !tim.dev) {
1360 printk(KERN_ERR "%s bad dst mac %pM\n",
1361 __func__, req->dst_mac);
1362 goto reject;
1363 }
1364
1365 /* Find output route */
1366 rt = find_route(tdev,
1367 req->local_ip,
1368 req->peer_ip,
1369 req->local_port,
1370 req->peer_port, G_PASS_OPEN_TOS(ntohl(req->tos_tid)));
1371 if (!rt) {
1372 printk(KERN_ERR MOD "%s - failed to find dst entry!\n",
1373 __func__);
1374 goto reject;
1375 }
1376 dst = &rt->dst;
1377 l2t = t3_l2t_get(tdev, dst, NULL);
1378 if (!l2t) {
1379 printk(KERN_ERR MOD "%s - failed to allocate l2t entry!\n",
1380 __func__);
1381 dst_release(dst);
1382 goto reject;
1383 }
1384 child_ep = alloc_ep(sizeof(*child_ep), GFP_KERNEL);
1385 if (!child_ep) {
1386 printk(KERN_ERR MOD "%s - failed to allocate ep entry!\n",
1387 __func__);
1388 l2t_release(tdev, l2t);
1389 dst_release(dst);
1390 goto reject;
1391 }
1392 state_set(&child_ep->com, CONNECTING);
1393 child_ep->com.tdev = tdev;
1394 child_ep->com.cm_id = NULL;
1395 child_ep->com.local_addr.sin_family = PF_INET;
1396 child_ep->com.local_addr.sin_port = req->local_port;
1397 child_ep->com.local_addr.sin_addr.s_addr = req->local_ip;
1398 child_ep->com.remote_addr.sin_family = PF_INET;
1399 child_ep->com.remote_addr.sin_port = req->peer_port;
1400 child_ep->com.remote_addr.sin_addr.s_addr = req->peer_ip;
1401 get_ep(&parent_ep->com);
1402 child_ep->parent_ep = parent_ep;
1403 child_ep->tos = G_PASS_OPEN_TOS(ntohl(req->tos_tid));
1404 child_ep->l2t = l2t;
1405 child_ep->dst = dst;
1406 child_ep->hwtid = hwtid;
1407 init_timer(&child_ep->timer);
1408 cxgb3_insert_tid(tdev, &t3c_client, child_ep, hwtid);
1409 accept_cr(child_ep, req->peer_ip, skb);
1410 goto out;
1411 reject:
1412 reject_cr(tdev, hwtid, req->peer_ip, skb);
1413 out:
1414 return CPL_RET_BUF_DONE;
1415 }
1416
pass_establish(struct t3cdev * tdev,struct sk_buff * skb,void * ctx)1417 static int pass_establish(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1418 {
1419 struct iwch_ep *ep = ctx;
1420 struct cpl_pass_establish *req = cplhdr(skb);
1421
1422 PDBG("%s ep %p\n", __func__, ep);
1423 ep->snd_seq = ntohl(req->snd_isn);
1424 ep->rcv_seq = ntohl(req->rcv_isn);
1425
1426 set_emss(ep, ntohs(req->tcp_opt));
1427
1428 dst_confirm(ep->dst);
1429 state_set(&ep->com, MPA_REQ_WAIT);
1430 start_ep_timer(ep);
1431
1432 return CPL_RET_BUF_DONE;
1433 }
1434
peer_close(struct t3cdev * tdev,struct sk_buff * skb,void * ctx)1435 static int peer_close(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1436 {
1437 struct iwch_ep *ep = ctx;
1438 struct iwch_qp_attributes attrs;
1439 unsigned long flags;
1440 int disconnect = 1;
1441 int release = 0;
1442
1443 PDBG("%s ep %p\n", __func__, ep);
1444 dst_confirm(ep->dst);
1445
1446 spin_lock_irqsave(&ep->com.lock, flags);
1447 switch (ep->com.state) {
1448 case MPA_REQ_WAIT:
1449 __state_set(&ep->com, CLOSING);
1450 break;
1451 case MPA_REQ_SENT:
1452 __state_set(&ep->com, CLOSING);
1453 connect_reply_upcall(ep, -ECONNRESET);
1454 break;
1455 case MPA_REQ_RCVD:
1456
1457 /*
1458 * We're gonna mark this puppy DEAD, but keep
1459 * the reference on it until the ULP accepts or
1460 * rejects the CR. Also wake up anyone waiting
1461 * in rdma connection migration (see iwch_accept_cr()).
1462 */
1463 __state_set(&ep->com, CLOSING);
1464 ep->com.rpl_done = 1;
1465 ep->com.rpl_err = -ECONNRESET;
1466 PDBG("waking up ep %p\n", ep);
1467 wake_up(&ep->com.waitq);
1468 break;
1469 case MPA_REP_SENT:
1470 __state_set(&ep->com, CLOSING);
1471 ep->com.rpl_done = 1;
1472 ep->com.rpl_err = -ECONNRESET;
1473 PDBG("waking up ep %p\n", ep);
1474 wake_up(&ep->com.waitq);
1475 break;
1476 case FPDU_MODE:
1477 start_ep_timer(ep);
1478 __state_set(&ep->com, CLOSING);
1479 attrs.next_state = IWCH_QP_STATE_CLOSING;
1480 iwch_modify_qp(ep->com.qp->rhp, ep->com.qp,
1481 IWCH_QP_ATTR_NEXT_STATE, &attrs, 1);
1482 peer_close_upcall(ep);
1483 break;
1484 case ABORTING:
1485 disconnect = 0;
1486 break;
1487 case CLOSING:
1488 __state_set(&ep->com, MORIBUND);
1489 disconnect = 0;
1490 break;
1491 case MORIBUND:
1492 stop_ep_timer(ep);
1493 if (ep->com.cm_id && ep->com.qp) {
1494 attrs.next_state = IWCH_QP_STATE_IDLE;
1495 iwch_modify_qp(ep->com.qp->rhp, ep->com.qp,
1496 IWCH_QP_ATTR_NEXT_STATE, &attrs, 1);
1497 }
1498 close_complete_upcall(ep);
1499 __state_set(&ep->com, DEAD);
1500 release = 1;
1501 disconnect = 0;
1502 break;
1503 case DEAD:
1504 disconnect = 0;
1505 break;
1506 default:
1507 BUG_ON(1);
1508 }
1509 spin_unlock_irqrestore(&ep->com.lock, flags);
1510 if (disconnect)
1511 iwch_ep_disconnect(ep, 0, GFP_KERNEL);
1512 if (release)
1513 release_ep_resources(ep);
1514 return CPL_RET_BUF_DONE;
1515 }
1516
1517 /*
1518 * Returns whether an ABORT_REQ_RSS message is a negative advice.
1519 */
is_neg_adv_abort(unsigned int status)1520 static int is_neg_adv_abort(unsigned int status)
1521 {
1522 return status == CPL_ERR_RTX_NEG_ADVICE ||
1523 status == CPL_ERR_PERSIST_NEG_ADVICE;
1524 }
1525
peer_abort(struct t3cdev * tdev,struct sk_buff * skb,void * ctx)1526 static int peer_abort(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1527 {
1528 struct cpl_abort_req_rss *req = cplhdr(skb);
1529 struct iwch_ep *ep = ctx;
1530 struct cpl_abort_rpl *rpl;
1531 struct sk_buff *rpl_skb;
1532 struct iwch_qp_attributes attrs;
1533 int ret;
1534 int release = 0;
1535 unsigned long flags;
1536
1537 if (is_neg_adv_abort(req->status)) {
1538 PDBG("%s neg_adv_abort ep %p tid %d\n", __func__, ep,
1539 ep->hwtid);
1540 t3_l2t_send_event(ep->com.tdev, ep->l2t);
1541 return CPL_RET_BUF_DONE;
1542 }
1543
1544 /*
1545 * We get 2 peer aborts from the HW. The first one must
1546 * be ignored except for scribbling that we need one more.
1547 */
1548 if (!test_and_set_bit(PEER_ABORT_IN_PROGRESS, &ep->com.flags)) {
1549 return CPL_RET_BUF_DONE;
1550 }
1551
1552 spin_lock_irqsave(&ep->com.lock, flags);
1553 PDBG("%s ep %p state %u\n", __func__, ep, ep->com.state);
1554 switch (ep->com.state) {
1555 case CONNECTING:
1556 break;
1557 case MPA_REQ_WAIT:
1558 stop_ep_timer(ep);
1559 break;
1560 case MPA_REQ_SENT:
1561 stop_ep_timer(ep);
1562 connect_reply_upcall(ep, -ECONNRESET);
1563 break;
1564 case MPA_REP_SENT:
1565 ep->com.rpl_done = 1;
1566 ep->com.rpl_err = -ECONNRESET;
1567 PDBG("waking up ep %p\n", ep);
1568 wake_up(&ep->com.waitq);
1569 break;
1570 case MPA_REQ_RCVD:
1571
1572 /*
1573 * We're gonna mark this puppy DEAD, but keep
1574 * the reference on it until the ULP accepts or
1575 * rejects the CR. Also wake up anyone waiting
1576 * in rdma connection migration (see iwch_accept_cr()).
1577 */
1578 ep->com.rpl_done = 1;
1579 ep->com.rpl_err = -ECONNRESET;
1580 PDBG("waking up ep %p\n", ep);
1581 wake_up(&ep->com.waitq);
1582 break;
1583 case MORIBUND:
1584 case CLOSING:
1585 stop_ep_timer(ep);
1586 /*FALLTHROUGH*/
1587 case FPDU_MODE:
1588 if (ep->com.cm_id && ep->com.qp) {
1589 attrs.next_state = IWCH_QP_STATE_ERROR;
1590 ret = iwch_modify_qp(ep->com.qp->rhp,
1591 ep->com.qp, IWCH_QP_ATTR_NEXT_STATE,
1592 &attrs, 1);
1593 if (ret)
1594 printk(KERN_ERR MOD
1595 "%s - qp <- error failed!\n",
1596 __func__);
1597 }
1598 peer_abort_upcall(ep);
1599 break;
1600 case ABORTING:
1601 break;
1602 case DEAD:
1603 PDBG("%s PEER_ABORT IN DEAD STATE!!!!\n", __func__);
1604 spin_unlock_irqrestore(&ep->com.lock, flags);
1605 return CPL_RET_BUF_DONE;
1606 default:
1607 BUG_ON(1);
1608 break;
1609 }
1610 dst_confirm(ep->dst);
1611 if (ep->com.state != ABORTING) {
1612 __state_set(&ep->com, DEAD);
1613 release = 1;
1614 }
1615 spin_unlock_irqrestore(&ep->com.lock, flags);
1616
1617 rpl_skb = get_skb(skb, sizeof(*rpl), GFP_KERNEL);
1618 if (!rpl_skb) {
1619 printk(KERN_ERR MOD "%s - cannot allocate skb!\n",
1620 __func__);
1621 release = 1;
1622 goto out;
1623 }
1624 rpl_skb->priority = CPL_PRIORITY_DATA;
1625 rpl = (struct cpl_abort_rpl *) skb_put(rpl_skb, sizeof(*rpl));
1626 rpl->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_OFLD_HOST_ABORT_CON_RPL));
1627 rpl->wr.wr_lo = htonl(V_WR_TID(ep->hwtid));
1628 OPCODE_TID(rpl) = htonl(MK_OPCODE_TID(CPL_ABORT_RPL, ep->hwtid));
1629 rpl->cmd = CPL_ABORT_NO_RST;
1630 iwch_cxgb3_ofld_send(ep->com.tdev, rpl_skb);
1631 out:
1632 if (release)
1633 release_ep_resources(ep);
1634 return CPL_RET_BUF_DONE;
1635 }
1636
close_con_rpl(struct t3cdev * tdev,struct sk_buff * skb,void * ctx)1637 static int close_con_rpl(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1638 {
1639 struct iwch_ep *ep = ctx;
1640 struct iwch_qp_attributes attrs;
1641 unsigned long flags;
1642 int release = 0;
1643
1644 PDBG("%s ep %p\n", __func__, ep);
1645 BUG_ON(!ep);
1646
1647 /* The cm_id may be null if we failed to connect */
1648 spin_lock_irqsave(&ep->com.lock, flags);
1649 switch (ep->com.state) {
1650 case CLOSING:
1651 __state_set(&ep->com, MORIBUND);
1652 break;
1653 case MORIBUND:
1654 stop_ep_timer(ep);
1655 if ((ep->com.cm_id) && (ep->com.qp)) {
1656 attrs.next_state = IWCH_QP_STATE_IDLE;
1657 iwch_modify_qp(ep->com.qp->rhp,
1658 ep->com.qp,
1659 IWCH_QP_ATTR_NEXT_STATE,
1660 &attrs, 1);
1661 }
1662 close_complete_upcall(ep);
1663 __state_set(&ep->com, DEAD);
1664 release = 1;
1665 break;
1666 case ABORTING:
1667 case DEAD:
1668 break;
1669 default:
1670 BUG_ON(1);
1671 break;
1672 }
1673 spin_unlock_irqrestore(&ep->com.lock, flags);
1674 if (release)
1675 release_ep_resources(ep);
1676 return CPL_RET_BUF_DONE;
1677 }
1678
1679 /*
1680 * T3A does 3 things when a TERM is received:
1681 * 1) send up a CPL_RDMA_TERMINATE message with the TERM packet
1682 * 2) generate an async event on the QP with the TERMINATE opcode
1683 * 3) post a TERMINATE opcde cqe into the associated CQ.
1684 *
1685 * For (1), we save the message in the qp for later consumer consumption.
1686 * For (2), we move the QP into TERMINATE, post a QP event and disconnect.
1687 * For (3), we toss the CQE in cxio_poll_cq().
1688 *
1689 * terminate() handles case (1)...
1690 */
terminate(struct t3cdev * tdev,struct sk_buff * skb,void * ctx)1691 static int terminate(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1692 {
1693 struct iwch_ep *ep = ctx;
1694
1695 if (state_read(&ep->com) != FPDU_MODE)
1696 return CPL_RET_BUF_DONE;
1697
1698 PDBG("%s ep %p\n", __func__, ep);
1699 skb_pull(skb, sizeof(struct cpl_rdma_terminate));
1700 PDBG("%s saving %d bytes of term msg\n", __func__, skb->len);
1701 skb_copy_from_linear_data(skb, ep->com.qp->attr.terminate_buffer,
1702 skb->len);
1703 ep->com.qp->attr.terminate_msg_len = skb->len;
1704 ep->com.qp->attr.is_terminate_local = 0;
1705 return CPL_RET_BUF_DONE;
1706 }
1707
ec_status(struct t3cdev * tdev,struct sk_buff * skb,void * ctx)1708 static int ec_status(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1709 {
1710 struct cpl_rdma_ec_status *rep = cplhdr(skb);
1711 struct iwch_ep *ep = ctx;
1712
1713 PDBG("%s ep %p tid %u status %d\n", __func__, ep, ep->hwtid,
1714 rep->status);
1715 if (rep->status) {
1716 struct iwch_qp_attributes attrs;
1717
1718 printk(KERN_ERR MOD "%s BAD CLOSE - Aborting tid %u\n",
1719 __func__, ep->hwtid);
1720 stop_ep_timer(ep);
1721 attrs.next_state = IWCH_QP_STATE_ERROR;
1722 iwch_modify_qp(ep->com.qp->rhp,
1723 ep->com.qp, IWCH_QP_ATTR_NEXT_STATE,
1724 &attrs, 1);
1725 abort_connection(ep, NULL, GFP_KERNEL);
1726 }
1727 return CPL_RET_BUF_DONE;
1728 }
1729
ep_timeout(unsigned long arg)1730 static void ep_timeout(unsigned long arg)
1731 {
1732 struct iwch_ep *ep = (struct iwch_ep *)arg;
1733 struct iwch_qp_attributes attrs;
1734 unsigned long flags;
1735 int abort = 1;
1736
1737 spin_lock_irqsave(&ep->com.lock, flags);
1738 PDBG("%s ep %p tid %u state %d\n", __func__, ep, ep->hwtid,
1739 ep->com.state);
1740 switch (ep->com.state) {
1741 case MPA_REQ_SENT:
1742 __state_set(&ep->com, ABORTING);
1743 connect_reply_upcall(ep, -ETIMEDOUT);
1744 break;
1745 case MPA_REQ_WAIT:
1746 __state_set(&ep->com, ABORTING);
1747 break;
1748 case CLOSING:
1749 case MORIBUND:
1750 if (ep->com.cm_id && ep->com.qp) {
1751 attrs.next_state = IWCH_QP_STATE_ERROR;
1752 iwch_modify_qp(ep->com.qp->rhp,
1753 ep->com.qp, IWCH_QP_ATTR_NEXT_STATE,
1754 &attrs, 1);
1755 }
1756 __state_set(&ep->com, ABORTING);
1757 break;
1758 default:
1759 printk(KERN_ERR "%s unexpected state ep %p state %u\n",
1760 __func__, ep, ep->com.state);
1761 WARN_ON(1);
1762 abort = 0;
1763 }
1764 spin_unlock_irqrestore(&ep->com.lock, flags);
1765 if (abort)
1766 abort_connection(ep, NULL, GFP_ATOMIC);
1767 put_ep(&ep->com);
1768 }
1769
iwch_reject_cr(struct iw_cm_id * cm_id,const void * pdata,u8 pdata_len)1770 int iwch_reject_cr(struct iw_cm_id *cm_id, const void *pdata, u8 pdata_len)
1771 {
1772 int err;
1773 struct iwch_ep *ep = to_ep(cm_id);
1774 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1775
1776 if (state_read(&ep->com) == DEAD) {
1777 put_ep(&ep->com);
1778 return -ECONNRESET;
1779 }
1780 BUG_ON(state_read(&ep->com) != MPA_REQ_RCVD);
1781 if (mpa_rev == 0)
1782 abort_connection(ep, NULL, GFP_KERNEL);
1783 else {
1784 err = send_mpa_reject(ep, pdata, pdata_len);
1785 err = iwch_ep_disconnect(ep, 0, GFP_KERNEL);
1786 }
1787 put_ep(&ep->com);
1788 return 0;
1789 }
1790
iwch_accept_cr(struct iw_cm_id * cm_id,struct iw_cm_conn_param * conn_param)1791 int iwch_accept_cr(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
1792 {
1793 int err;
1794 struct iwch_qp_attributes attrs;
1795 enum iwch_qp_attr_mask mask;
1796 struct iwch_ep *ep = to_ep(cm_id);
1797 struct iwch_dev *h = to_iwch_dev(cm_id->device);
1798 struct iwch_qp *qp = get_qhp(h, conn_param->qpn);
1799
1800 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1801 if (state_read(&ep->com) == DEAD) {
1802 err = -ECONNRESET;
1803 goto err;
1804 }
1805
1806 BUG_ON(state_read(&ep->com) != MPA_REQ_RCVD);
1807 BUG_ON(!qp);
1808
1809 if ((conn_param->ord > qp->rhp->attr.max_rdma_read_qp_depth) ||
1810 (conn_param->ird > qp->rhp->attr.max_rdma_reads_per_qp)) {
1811 abort_connection(ep, NULL, GFP_KERNEL);
1812 err = -EINVAL;
1813 goto err;
1814 }
1815
1816 cm_id->add_ref(cm_id);
1817 ep->com.cm_id = cm_id;
1818 ep->com.qp = qp;
1819
1820 ep->ird = conn_param->ird;
1821 ep->ord = conn_param->ord;
1822
1823 if (peer2peer && ep->ird == 0)
1824 ep->ird = 1;
1825
1826 PDBG("%s %d ird %d ord %d\n", __func__, __LINE__, ep->ird, ep->ord);
1827
1828 /* bind QP to EP and move to RTS */
1829 attrs.mpa_attr = ep->mpa_attr;
1830 attrs.max_ird = ep->ird;
1831 attrs.max_ord = ep->ord;
1832 attrs.llp_stream_handle = ep;
1833 attrs.next_state = IWCH_QP_STATE_RTS;
1834
1835 /* bind QP and TID with INIT_WR */
1836 mask = IWCH_QP_ATTR_NEXT_STATE |
1837 IWCH_QP_ATTR_LLP_STREAM_HANDLE |
1838 IWCH_QP_ATTR_MPA_ATTR |
1839 IWCH_QP_ATTR_MAX_IRD |
1840 IWCH_QP_ATTR_MAX_ORD;
1841
1842 err = iwch_modify_qp(ep->com.qp->rhp,
1843 ep->com.qp, mask, &attrs, 1);
1844 if (err)
1845 goto err1;
1846
1847 /* if needed, wait for wr_ack */
1848 if (iwch_rqes_posted(qp)) {
1849 wait_event(ep->com.waitq, ep->com.rpl_done);
1850 err = ep->com.rpl_err;
1851 if (err)
1852 goto err1;
1853 }
1854
1855 err = send_mpa_reply(ep, conn_param->private_data,
1856 conn_param->private_data_len);
1857 if (err)
1858 goto err1;
1859
1860
1861 state_set(&ep->com, FPDU_MODE);
1862 established_upcall(ep);
1863 put_ep(&ep->com);
1864 return 0;
1865 err1:
1866 ep->com.cm_id = NULL;
1867 ep->com.qp = NULL;
1868 cm_id->rem_ref(cm_id);
1869 err:
1870 put_ep(&ep->com);
1871 return err;
1872 }
1873
is_loopback_dst(struct iw_cm_id * cm_id)1874 static int is_loopback_dst(struct iw_cm_id *cm_id)
1875 {
1876 struct net_device *dev;
1877
1878 dev = ip_dev_find(&init_net, cm_id->remote_addr.sin_addr.s_addr);
1879 if (!dev)
1880 return 0;
1881 dev_put(dev);
1882 return 1;
1883 }
1884
iwch_connect(struct iw_cm_id * cm_id,struct iw_cm_conn_param * conn_param)1885 int iwch_connect(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
1886 {
1887 struct iwch_dev *h = to_iwch_dev(cm_id->device);
1888 struct iwch_ep *ep;
1889 struct rtable *rt;
1890 int err = 0;
1891
1892 if (is_loopback_dst(cm_id)) {
1893 err = -ENOSYS;
1894 goto out;
1895 }
1896
1897 ep = alloc_ep(sizeof(*ep), GFP_KERNEL);
1898 if (!ep) {
1899 printk(KERN_ERR MOD "%s - cannot alloc ep.\n", __func__);
1900 err = -ENOMEM;
1901 goto out;
1902 }
1903 init_timer(&ep->timer);
1904 ep->plen = conn_param->private_data_len;
1905 if (ep->plen)
1906 memcpy(ep->mpa_pkt + sizeof(struct mpa_message),
1907 conn_param->private_data, ep->plen);
1908 ep->ird = conn_param->ird;
1909 ep->ord = conn_param->ord;
1910
1911 if (peer2peer && ep->ord == 0)
1912 ep->ord = 1;
1913
1914 ep->com.tdev = h->rdev.t3cdev_p;
1915
1916 cm_id->add_ref(cm_id);
1917 ep->com.cm_id = cm_id;
1918 ep->com.qp = get_qhp(h, conn_param->qpn);
1919 BUG_ON(!ep->com.qp);
1920 PDBG("%s qpn 0x%x qp %p cm_id %p\n", __func__, conn_param->qpn,
1921 ep->com.qp, cm_id);
1922
1923 /*
1924 * Allocate an active TID to initiate a TCP connection.
1925 */
1926 ep->atid = cxgb3_alloc_atid(h->rdev.t3cdev_p, &t3c_client, ep);
1927 if (ep->atid == -1) {
1928 printk(KERN_ERR MOD "%s - cannot alloc atid.\n", __func__);
1929 err = -ENOMEM;
1930 goto fail2;
1931 }
1932
1933 /* find a route */
1934 rt = find_route(h->rdev.t3cdev_p,
1935 cm_id->local_addr.sin_addr.s_addr,
1936 cm_id->remote_addr.sin_addr.s_addr,
1937 cm_id->local_addr.sin_port,
1938 cm_id->remote_addr.sin_port, IPTOS_LOWDELAY);
1939 if (!rt) {
1940 printk(KERN_ERR MOD "%s - cannot find route.\n", __func__);
1941 err = -EHOSTUNREACH;
1942 goto fail3;
1943 }
1944 ep->dst = &rt->dst;
1945 ep->l2t = t3_l2t_get(ep->com.tdev, ep->dst, NULL);
1946 if (!ep->l2t) {
1947 printk(KERN_ERR MOD "%s - cannot alloc l2e.\n", __func__);
1948 err = -ENOMEM;
1949 goto fail4;
1950 }
1951
1952 state_set(&ep->com, CONNECTING);
1953 ep->tos = IPTOS_LOWDELAY;
1954 ep->com.local_addr = cm_id->local_addr;
1955 ep->com.remote_addr = cm_id->remote_addr;
1956
1957 /* send connect request to rnic */
1958 err = send_connect(ep);
1959 if (!err)
1960 goto out;
1961
1962 l2t_release(h->rdev.t3cdev_p, ep->l2t);
1963 fail4:
1964 dst_release(ep->dst);
1965 fail3:
1966 cxgb3_free_atid(ep->com.tdev, ep->atid);
1967 fail2:
1968 cm_id->rem_ref(cm_id);
1969 put_ep(&ep->com);
1970 out:
1971 return err;
1972 }
1973
iwch_create_listen(struct iw_cm_id * cm_id,int backlog)1974 int iwch_create_listen(struct iw_cm_id *cm_id, int backlog)
1975 {
1976 int err = 0;
1977 struct iwch_dev *h = to_iwch_dev(cm_id->device);
1978 struct iwch_listen_ep *ep;
1979
1980
1981 might_sleep();
1982
1983 ep = alloc_ep(sizeof(*ep), GFP_KERNEL);
1984 if (!ep) {
1985 printk(KERN_ERR MOD "%s - cannot alloc ep.\n", __func__);
1986 err = -ENOMEM;
1987 goto fail1;
1988 }
1989 PDBG("%s ep %p\n", __func__, ep);
1990 ep->com.tdev = h->rdev.t3cdev_p;
1991 cm_id->add_ref(cm_id);
1992 ep->com.cm_id = cm_id;
1993 ep->backlog = backlog;
1994 ep->com.local_addr = cm_id->local_addr;
1995
1996 /*
1997 * Allocate a server TID.
1998 */
1999 ep->stid = cxgb3_alloc_stid(h->rdev.t3cdev_p, &t3c_client, ep);
2000 if (ep->stid == -1) {
2001 printk(KERN_ERR MOD "%s - cannot alloc atid.\n", __func__);
2002 err = -ENOMEM;
2003 goto fail2;
2004 }
2005
2006 state_set(&ep->com, LISTEN);
2007 err = listen_start(ep);
2008 if (err)
2009 goto fail3;
2010
2011 /* wait for pass_open_rpl */
2012 wait_event(ep->com.waitq, ep->com.rpl_done);
2013 err = ep->com.rpl_err;
2014 if (!err) {
2015 cm_id->provider_data = ep;
2016 goto out;
2017 }
2018 fail3:
2019 cxgb3_free_stid(ep->com.tdev, ep->stid);
2020 fail2:
2021 cm_id->rem_ref(cm_id);
2022 put_ep(&ep->com);
2023 fail1:
2024 out:
2025 return err;
2026 }
2027
iwch_destroy_listen(struct iw_cm_id * cm_id)2028 int iwch_destroy_listen(struct iw_cm_id *cm_id)
2029 {
2030 int err;
2031 struct iwch_listen_ep *ep = to_listen_ep(cm_id);
2032
2033 PDBG("%s ep %p\n", __func__, ep);
2034
2035 might_sleep();
2036 state_set(&ep->com, DEAD);
2037 ep->com.rpl_done = 0;
2038 ep->com.rpl_err = 0;
2039 err = listen_stop(ep);
2040 if (err)
2041 goto done;
2042 wait_event(ep->com.waitq, ep->com.rpl_done);
2043 cxgb3_free_stid(ep->com.tdev, ep->stid);
2044 done:
2045 err = ep->com.rpl_err;
2046 cm_id->rem_ref(cm_id);
2047 put_ep(&ep->com);
2048 return err;
2049 }
2050
iwch_ep_disconnect(struct iwch_ep * ep,int abrupt,gfp_t gfp)2051 int iwch_ep_disconnect(struct iwch_ep *ep, int abrupt, gfp_t gfp)
2052 {
2053 int ret=0;
2054 unsigned long flags;
2055 int close = 0;
2056 int fatal = 0;
2057 struct t3cdev *tdev;
2058 struct cxio_rdev *rdev;
2059
2060 spin_lock_irqsave(&ep->com.lock, flags);
2061
2062 PDBG("%s ep %p state %s, abrupt %d\n", __func__, ep,
2063 states[ep->com.state], abrupt);
2064
2065 tdev = (struct t3cdev *)ep->com.tdev;
2066 rdev = (struct cxio_rdev *)tdev->ulp;
2067 if (cxio_fatal_error(rdev)) {
2068 fatal = 1;
2069 close_complete_upcall(ep);
2070 ep->com.state = DEAD;
2071 }
2072 switch (ep->com.state) {
2073 case MPA_REQ_WAIT:
2074 case MPA_REQ_SENT:
2075 case MPA_REQ_RCVD:
2076 case MPA_REP_SENT:
2077 case FPDU_MODE:
2078 close = 1;
2079 if (abrupt)
2080 ep->com.state = ABORTING;
2081 else {
2082 ep->com.state = CLOSING;
2083 start_ep_timer(ep);
2084 }
2085 set_bit(CLOSE_SENT, &ep->com.flags);
2086 break;
2087 case CLOSING:
2088 if (!test_and_set_bit(CLOSE_SENT, &ep->com.flags)) {
2089 close = 1;
2090 if (abrupt) {
2091 stop_ep_timer(ep);
2092 ep->com.state = ABORTING;
2093 } else
2094 ep->com.state = MORIBUND;
2095 }
2096 break;
2097 case MORIBUND:
2098 case ABORTING:
2099 case DEAD:
2100 PDBG("%s ignoring disconnect ep %p state %u\n",
2101 __func__, ep, ep->com.state);
2102 break;
2103 default:
2104 BUG();
2105 break;
2106 }
2107
2108 spin_unlock_irqrestore(&ep->com.lock, flags);
2109 if (close) {
2110 if (abrupt)
2111 ret = send_abort(ep, NULL, gfp);
2112 else
2113 ret = send_halfclose(ep, gfp);
2114 if (ret)
2115 fatal = 1;
2116 }
2117 if (fatal)
2118 release_ep_resources(ep);
2119 return ret;
2120 }
2121
iwch_ep_redirect(void * ctx,struct dst_entry * old,struct dst_entry * new,struct l2t_entry * l2t)2122 int iwch_ep_redirect(void *ctx, struct dst_entry *old, struct dst_entry *new,
2123 struct l2t_entry *l2t)
2124 {
2125 struct iwch_ep *ep = ctx;
2126
2127 if (ep->dst != old)
2128 return 0;
2129
2130 PDBG("%s ep %p redirect to dst %p l2t %p\n", __func__, ep, new,
2131 l2t);
2132 dst_hold(new);
2133 l2t_release(ep->com.tdev, ep->l2t);
2134 ep->l2t = l2t;
2135 dst_release(old);
2136 ep->dst = new;
2137 return 1;
2138 }
2139
2140 /*
2141 * All the CM events are handled on a work queue to have a safe context.
2142 * These are the real handlers that are called from the work queue.
2143 */
2144 static const cxgb3_cpl_handler_func work_handlers[NUM_CPL_CMDS] = {
2145 [CPL_ACT_ESTABLISH] = act_establish,
2146 [CPL_ACT_OPEN_RPL] = act_open_rpl,
2147 [CPL_RX_DATA] = rx_data,
2148 [CPL_TX_DMA_ACK] = tx_ack,
2149 [CPL_ABORT_RPL_RSS] = abort_rpl,
2150 [CPL_ABORT_RPL] = abort_rpl,
2151 [CPL_PASS_OPEN_RPL] = pass_open_rpl,
2152 [CPL_CLOSE_LISTSRV_RPL] = close_listsrv_rpl,
2153 [CPL_PASS_ACCEPT_REQ] = pass_accept_req,
2154 [CPL_PASS_ESTABLISH] = pass_establish,
2155 [CPL_PEER_CLOSE] = peer_close,
2156 [CPL_ABORT_REQ_RSS] = peer_abort,
2157 [CPL_CLOSE_CON_RPL] = close_con_rpl,
2158 [CPL_RDMA_TERMINATE] = terminate,
2159 [CPL_RDMA_EC_STATUS] = ec_status,
2160 };
2161
process_work(struct work_struct * work)2162 static void process_work(struct work_struct *work)
2163 {
2164 struct sk_buff *skb = NULL;
2165 void *ep;
2166 struct t3cdev *tdev;
2167 int ret;
2168
2169 while ((skb = skb_dequeue(&rxq))) {
2170 ep = *((void **) (skb->cb));
2171 tdev = *((struct t3cdev **) (skb->cb + sizeof(void *)));
2172 ret = work_handlers[G_OPCODE(ntohl((__force __be32)skb->csum))](tdev, skb, ep);
2173 if (ret & CPL_RET_BUF_DONE)
2174 kfree_skb(skb);
2175
2176 /*
2177 * ep was referenced in sched(), and is freed here.
2178 */
2179 put_ep((struct iwch_ep_common *)ep);
2180 }
2181 }
2182
2183 static DECLARE_WORK(skb_work, process_work);
2184
sched(struct t3cdev * tdev,struct sk_buff * skb,void * ctx)2185 static int sched(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
2186 {
2187 struct iwch_ep_common *epc = ctx;
2188
2189 get_ep(epc);
2190
2191 /*
2192 * Save ctx and tdev in the skb->cb area.
2193 */
2194 *((void **) skb->cb) = ctx;
2195 *((struct t3cdev **) (skb->cb + sizeof(void *))) = tdev;
2196
2197 /*
2198 * Queue the skb and schedule the worker thread.
2199 */
2200 skb_queue_tail(&rxq, skb);
2201 queue_work(workq, &skb_work);
2202 return 0;
2203 }
2204
set_tcb_rpl(struct t3cdev * tdev,struct sk_buff * skb,void * ctx)2205 static int set_tcb_rpl(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
2206 {
2207 struct cpl_set_tcb_rpl *rpl = cplhdr(skb);
2208
2209 if (rpl->status != CPL_ERR_NONE) {
2210 printk(KERN_ERR MOD "Unexpected SET_TCB_RPL status %u "
2211 "for tid %u\n", rpl->status, GET_TID(rpl));
2212 }
2213 return CPL_RET_BUF_DONE;
2214 }
2215
2216 /*
2217 * All upcalls from the T3 Core go to sched() to schedule the
2218 * processing on a work queue.
2219 */
2220 cxgb3_cpl_handler_func t3c_handlers[NUM_CPL_CMDS] = {
2221 [CPL_ACT_ESTABLISH] = sched,
2222 [CPL_ACT_OPEN_RPL] = sched,
2223 [CPL_RX_DATA] = sched,
2224 [CPL_TX_DMA_ACK] = sched,
2225 [CPL_ABORT_RPL_RSS] = sched,
2226 [CPL_ABORT_RPL] = sched,
2227 [CPL_PASS_OPEN_RPL] = sched,
2228 [CPL_CLOSE_LISTSRV_RPL] = sched,
2229 [CPL_PASS_ACCEPT_REQ] = sched,
2230 [CPL_PASS_ESTABLISH] = sched,
2231 [CPL_PEER_CLOSE] = sched,
2232 [CPL_CLOSE_CON_RPL] = sched,
2233 [CPL_ABORT_REQ_RSS] = sched,
2234 [CPL_RDMA_TERMINATE] = sched,
2235 [CPL_RDMA_EC_STATUS] = sched,
2236 [CPL_SET_TCB_RPL] = set_tcb_rpl,
2237 };
2238
iwch_cm_init(void)2239 int __init iwch_cm_init(void)
2240 {
2241 skb_queue_head_init(&rxq);
2242
2243 workq = create_singlethread_workqueue("iw_cxgb3");
2244 if (!workq)
2245 return -ENOMEM;
2246
2247 return 0;
2248 }
2249
iwch_cm_term(void)2250 void __exit iwch_cm_term(void)
2251 {
2252 flush_workqueue(workq);
2253 destroy_workqueue(workq);
2254 }
2255