?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Merge tag 'v3.3.7' into 3.3/master
[zen-stable.git] / drivers / infiniband / hw / cxgb4 / cm.c
blob0668bb3472d096ba1c0889bbc999b2076932ffe7
1 /*
2 * Copyright (c) 2009-2010 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/workqueue.h>
35 #include <linux/skbuff.h>
36 #include <linux/timer.h>
37 #include <linux/notifier.h>
38 #include <linux/inetdevice.h>
39 #include <linux/ip.h>
40 #include <linux/tcp.h>
41
42 #include <net/neighbour.h>
43 #include <net/netevent.h>
44 #include <net/route.h>
45
46 #include "iw_cxgb4.h"
47
48 static char *states[] = {
49 "idle",
50 "listen",
51 "connecting",
52 "mpa_wait_req",
53 "mpa_req_sent",
54 "mpa_req_rcvd",
55 "mpa_rep_sent",
56 "fpdu_mode",
57 "aborting",
58 "closing",
59 "moribund",
60 "dead",
61 NULL,
62 };
63
64 static int dack_mode = 1;
65 module_param(dack_mode, int, 0644);
66 MODULE_PARM_DESC(dack_mode, "Delayed ack mode (default=1)");
67
68 int c4iw_max_read_depth = 8;
69 module_param(c4iw_max_read_depth, int, 0644);
70 MODULE_PARM_DESC(c4iw_max_read_depth, "Per-connection max ORD/IRD (default=8)");
71
72 static int enable_tcp_timestamps;
73 module_param(enable_tcp_timestamps, int, 0644);
74 MODULE_PARM_DESC(enable_tcp_timestamps, "Enable tcp timestamps (default=0)");
75
76 static int enable_tcp_sack;
77 module_param(enable_tcp_sack, int, 0644);
78 MODULE_PARM_DESC(enable_tcp_sack, "Enable tcp SACK (default=0)");
79
80 static int enable_tcp_window_scaling = 1;
81 module_param(enable_tcp_window_scaling, int, 0644);
82 MODULE_PARM_DESC(enable_tcp_window_scaling,
83 "Enable tcp window scaling (default=1)");
84
85 int c4iw_debug;
86 module_param(c4iw_debug, int, 0644);
87 MODULE_PARM_DESC(c4iw_debug, "Enable debug logging (default=0)");
88
89 static int peer2peer;
90 module_param(peer2peer, int, 0644);
91 MODULE_PARM_DESC(peer2peer, "Support peer2peer ULPs (default=0)");
92
93 static int p2p_type = FW_RI_INIT_P2PTYPE_READ_REQ;
94 module_param(p2p_type, int, 0644);
95 MODULE_PARM_DESC(p2p_type, "RDMAP opcode to use for the RTR message: "
96 "1=RDMA_READ 0=RDMA_WRITE (default 1)");
97
98 static int ep_timeout_secs = 60;
99 module_param(ep_timeout_secs, int, 0644);
100 MODULE_PARM_DESC(ep_timeout_secs, "CM Endpoint operation timeout "
101 "in seconds (default=60)");
102
103 static int mpa_rev = 1;
104 module_param(mpa_rev, int, 0644);
105 MODULE_PARM_DESC(mpa_rev, "MPA Revision, 0 supports amso1100, "
106 "1 is RFC0544 spec compliant, 2 is IETF MPA Peer Connect Draft"
107 " compliant (default=1)");
108
109 static int markers_enabled;
110 module_param(markers_enabled, int, 0644);
111 MODULE_PARM_DESC(markers_enabled, "Enable MPA MARKERS (default(0)=disabled)");
112
113 static int crc_enabled = 1;
114 module_param(crc_enabled, int, 0644);
115 MODULE_PARM_DESC(crc_enabled, "Enable MPA CRC (default(1)=enabled)");
116
117 static int rcv_win = 256 * 1024;
118 module_param(rcv_win, int, 0644);
119 MODULE_PARM_DESC(rcv_win, "TCP receive window in bytes (default=256KB)");
120
121 static int snd_win = 128 * 1024;
122 module_param(snd_win, int, 0644);
123 MODULE_PARM_DESC(snd_win, "TCP send window in bytes (default=128KB)");
124
125 static struct workqueue_struct *workq;
126
127 static struct sk_buff_head rxq;
128
129 static struct sk_buff *get_skb(struct sk_buff *skb, int len, gfp_t gfp);
130 static void ep_timeout(unsigned long arg);
131 static void connect_reply_upcall(struct c4iw_ep *ep, int status);
132
133 static LIST_HEAD(timeout_list);
134 static spinlock_t timeout_lock;
135
136 static void start_ep_timer(struct c4iw_ep *ep)
137 {
138 PDBG("%s ep %p\n", __func__, ep);
139 if (timer_pending(&ep->timer)) {
140 PDBG("%s stopped / restarted timer ep %p\n", __func__, ep);
141 del_timer_sync(&ep->timer);
142 } else
143 c4iw_get_ep(&ep->com);
144 ep->timer.expires = jiffies + ep_timeout_secs * HZ;
145 ep->timer.data = (unsigned long)ep;
146 ep->timer.function = ep_timeout;
147 add_timer(&ep->timer);
148 }
149
150 static void stop_ep_timer(struct c4iw_ep *ep)
151 {
152 PDBG("%s ep %p\n", __func__, ep);
153 if (!timer_pending(&ep->timer)) {
154 printk(KERN_ERR "%s timer stopped when its not running! "
155 "ep %p state %u\n", __func__, ep, ep->com.state);
156 WARN_ON(1);
157 return;
158 }
159 del_timer_sync(&ep->timer);
160 c4iw_put_ep(&ep->com);
161 }
162
163 static int c4iw_l2t_send(struct c4iw_rdev *rdev, struct sk_buff *skb,
164 struct l2t_entry *l2e)
165 {
166 int error = 0;
167
168 if (c4iw_fatal_error(rdev)) {
169 kfree_skb(skb);
170 PDBG("%s - device in error state - dropping\n", __func__);
171 return -EIO;
172 }
173 error = cxgb4_l2t_send(rdev->lldi.ports[0], skb, l2e);
174 if (error < 0)
175 kfree_skb(skb);
176 return error < 0 ? error : 0;
177 }
178
179 int c4iw_ofld_send(struct c4iw_rdev *rdev, struct sk_buff *skb)
180 {
181 int error = 0;
182
183 if (c4iw_fatal_error(rdev)) {
184 kfree_skb(skb);
185 PDBG("%s - device in error state - dropping\n", __func__);
186 return -EIO;
187 }
188 error = cxgb4_ofld_send(rdev->lldi.ports[0], skb);
189 if (error < 0)
190 kfree_skb(skb);
191 return error < 0 ? error : 0;
192 }
193
194 static void release_tid(struct c4iw_rdev *rdev, u32 hwtid, struct sk_buff *skb)
195 {
196 struct cpl_tid_release *req;
197
198 skb = get_skb(skb, sizeof *req, GFP_KERNEL);
199 if (!skb)
200 return;
201 req = (struct cpl_tid_release *) skb_put(skb, sizeof(*req));
202 INIT_TP_WR(req, hwtid);
203 OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_TID_RELEASE, hwtid));
204 set_wr_txq(skb, CPL_PRIORITY_SETUP, 0);
205 c4iw_ofld_send(rdev, skb);
206 return;
207 }
208
209 static void set_emss(struct c4iw_ep *ep, u16 opt)
210 {
211 ep->emss = ep->com.dev->rdev.lldi.mtus[GET_TCPOPT_MSS(opt)] - 40;
212 ep->mss = ep->emss;
213 if (GET_TCPOPT_TSTAMP(opt))
214 ep->emss -= 12;
215 if (ep->emss < 128)
216 ep->emss = 128;
217 PDBG("%s mss_idx %u mss %u emss=%u\n", __func__, GET_TCPOPT_MSS(opt),
218 ep->mss, ep->emss);
219 }
220
221 static enum c4iw_ep_state state_read(struct c4iw_ep_common *epc)
222 {
223 enum c4iw_ep_state state;
224
225 mutex_lock(&epc->mutex);
226 state = epc->state;
227 mutex_unlock(&epc->mutex);
228 return state;
229 }
230
231 static void __state_set(struct c4iw_ep_common *epc, enum c4iw_ep_state new)
232 {
233 epc->state = new;
234 }
235
236 static void state_set(struct c4iw_ep_common *epc, enum c4iw_ep_state new)
237 {
238 mutex_lock(&epc->mutex);
239 PDBG("%s - %s -> %s\n", __func__, states[epc->state], states[new]);
240 __state_set(epc, new);
241 mutex_unlock(&epc->mutex);
242 return;
243 }
244
245 static void *alloc_ep(int size, gfp_t gfp)
246 {
247 struct c4iw_ep_common *epc;
248
249 epc = kzalloc(size, gfp);
250 if (epc) {
251 kref_init(&epc->kref);
252 mutex_init(&epc->mutex);
253 c4iw_init_wr_wait(&epc->wr_wait);
254 }
255 PDBG("%s alloc ep %p\n", __func__, epc);
256 return epc;
257 }
258
259 void _c4iw_free_ep(struct kref *kref)
260 {
261 struct c4iw_ep *ep;
262
263 ep = container_of(kref, struct c4iw_ep, com.kref);
264 PDBG("%s ep %p state %s\n", __func__, ep, states[state_read(&ep->com)]);
265 if (test_bit(RELEASE_RESOURCES, &ep->com.flags)) {
266 cxgb4_remove_tid(ep->com.dev->rdev.lldi.tids, 0, ep->hwtid);
267 dst_release(ep->dst);
268 cxgb4_l2t_release(ep->l2t);
269 }
270 kfree(ep);
271 }
272
273 static void release_ep_resources(struct c4iw_ep *ep)
274 {
275 set_bit(RELEASE_RESOURCES, &ep->com.flags);
276 c4iw_put_ep(&ep->com);
277 }
278
279 static int status2errno(int status)
280 {
281 switch (status) {
282 case CPL_ERR_NONE:
283 return 0;
284 case CPL_ERR_CONN_RESET:
285 return -ECONNRESET;
286 case CPL_ERR_ARP_MISS:
287 return -EHOSTUNREACH;
288 case CPL_ERR_CONN_TIMEDOUT:
289 return -ETIMEDOUT;
290 case CPL_ERR_TCAM_FULL:
291 return -ENOMEM;
292 case CPL_ERR_CONN_EXIST:
293 return -EADDRINUSE;
294 default:
295 return -EIO;
296 }
297 }
298
299 /*
300 * Try and reuse skbs already allocated...
301 */
302 static struct sk_buff *get_skb(struct sk_buff *skb, int len, gfp_t gfp)
303 {
304 if (skb && !skb_is_nonlinear(skb) && !skb_cloned(skb)) {
305 skb_trim(skb, 0);
306 skb_get(skb);
307 skb_reset_transport_header(skb);
308 } else {
309 skb = alloc_skb(len, gfp);
310 }
311 return skb;
312 }
313
314 static struct rtable *find_route(struct c4iw_dev *dev, __be32 local_ip,
315 __be32 peer_ip, __be16 local_port,
316 __be16 peer_port, u8 tos)
317 {
318 struct rtable *rt;
319 struct flowi4 fl4;
320
321 rt = ip_route_output_ports(&init_net, &fl4, NULL, peer_ip, local_ip,
322 peer_port, local_port, IPPROTO_TCP,
323 tos, 0);
324 if (IS_ERR(rt))
325 return NULL;
326 return rt;
327 }
328
329 static void arp_failure_discard(void *handle, struct sk_buff *skb)
330 {
331 PDBG("%s c4iw_dev %p\n", __func__, handle);
332 kfree_skb(skb);
333 }
334
335 /*
336 * Handle an ARP failure for an active open.
337 */
338 static void act_open_req_arp_failure(void *handle, struct sk_buff *skb)
339 {
340 printk(KERN_ERR MOD "ARP failure duing connect\n");
341 kfree_skb(skb);
342 }
343
344 /*
345 * Handle an ARP failure for a CPL_ABORT_REQ. Change it into a no RST variant
346 * and send it along.
347 */
348 static void abort_arp_failure(void *handle, struct sk_buff *skb)
349 {
350 struct c4iw_rdev *rdev = handle;
351 struct cpl_abort_req *req = cplhdr(skb);
352
353 PDBG("%s rdev %p\n", __func__, rdev);
354 req->cmd = CPL_ABORT_NO_RST;
355 c4iw_ofld_send(rdev, skb);
356 }
357
358 static void send_flowc(struct c4iw_ep *ep, struct sk_buff *skb)
359 {
360 unsigned int flowclen = 80;
361 struct fw_flowc_wr *flowc;
362 int i;
363
364 skb = get_skb(skb, flowclen, GFP_KERNEL);
365 flowc = (struct fw_flowc_wr *)__skb_put(skb, flowclen);
366
367 flowc->op_to_nparams = cpu_to_be32(FW_WR_OP(FW_FLOWC_WR) |
368 FW_FLOWC_WR_NPARAMS(8));
369 flowc->flowid_len16 = cpu_to_be32(FW_WR_LEN16(DIV_ROUND_UP(flowclen,
370 16)) | FW_WR_FLOWID(ep->hwtid));
371
372 flowc->mnemval[0].mnemonic = FW_FLOWC_MNEM_PFNVFN;
373 flowc->mnemval[0].val = cpu_to_be32(PCI_FUNC(ep->com.dev->rdev.lldi.pdev->devfn) << 8);
374 flowc->mnemval[1].mnemonic = FW_FLOWC_MNEM_CH;
375 flowc->mnemval[1].val = cpu_to_be32(ep->tx_chan);
376 flowc->mnemval[2].mnemonic = FW_FLOWC_MNEM_PORT;
377 flowc->mnemval[2].val = cpu_to_be32(ep->tx_chan);
378 flowc->mnemval[3].mnemonic = FW_FLOWC_MNEM_IQID;
379 flowc->mnemval[3].val = cpu_to_be32(ep->rss_qid);
380 flowc->mnemval[4].mnemonic = FW_FLOWC_MNEM_SNDNXT;
381 flowc->mnemval[4].val = cpu_to_be32(ep->snd_seq);
382 flowc->mnemval[5].mnemonic = FW_FLOWC_MNEM_RCVNXT;
383 flowc->mnemval[5].val = cpu_to_be32(ep->rcv_seq);
384 flowc->mnemval[6].mnemonic = FW_FLOWC_MNEM_SNDBUF;
385 flowc->mnemval[6].val = cpu_to_be32(snd_win);
386 flowc->mnemval[7].mnemonic = FW_FLOWC_MNEM_MSS;
387 flowc->mnemval[7].val = cpu_to_be32(ep->emss);
388 /* Pad WR to 16 byte boundary */
389 flowc->mnemval[8].mnemonic = 0;
390 flowc->mnemval[8].val = 0;
391 for (i = 0; i < 9; i++) {
392 flowc->mnemval[i].r4[0] = 0;
393 flowc->mnemval[i].r4[1] = 0;
394 flowc->mnemval[i].r4[2] = 0;
395 }
396
397 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
398 c4iw_ofld_send(&ep->com.dev->rdev, skb);
399 }
400
401 static int send_halfclose(struct c4iw_ep *ep, gfp_t gfp)
402 {
403 struct cpl_close_con_req *req;
404 struct sk_buff *skb;
405 int wrlen = roundup(sizeof *req, 16);
406
407 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
408 skb = get_skb(NULL, wrlen, gfp);
409 if (!skb) {
410 printk(KERN_ERR MOD "%s - failed to alloc skb\n", __func__);
411 return -ENOMEM;
412 }
413 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
414 t4_set_arp_err_handler(skb, NULL, arp_failure_discard);
415 req = (struct cpl_close_con_req *) skb_put(skb, wrlen);
416 memset(req, 0, wrlen);
417 INIT_TP_WR(req, ep->hwtid);
418 OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_CLOSE_CON_REQ,
419 ep->hwtid));
420 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
421 }
422
423 static int send_abort(struct c4iw_ep *ep, struct sk_buff *skb, gfp_t gfp)
424 {
425 struct cpl_abort_req *req;
426 int wrlen = roundup(sizeof *req, 16);
427
428 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
429 skb = get_skb(skb, wrlen, gfp);
430 if (!skb) {
431 printk(KERN_ERR MOD "%s - failed to alloc skb.\n",
432 __func__);
433 return -ENOMEM;
434 }
435 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
436 t4_set_arp_err_handler(skb, &ep->com.dev->rdev, abort_arp_failure);
437 req = (struct cpl_abort_req *) skb_put(skb, wrlen);
438 memset(req, 0, wrlen);
439 INIT_TP_WR(req, ep->hwtid);
440 OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_ABORT_REQ, ep->hwtid));
441 req->cmd = CPL_ABORT_SEND_RST;
442 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
443 }
444
445 static int send_connect(struct c4iw_ep *ep)
446 {
447 struct cpl_act_open_req *req;
448 struct sk_buff *skb;
449 u64 opt0;
450 u32 opt2;
451 unsigned int mtu_idx;
452 int wscale;
453 int wrlen = roundup(sizeof *req, 16);
454
455 PDBG("%s ep %p atid %u\n", __func__, ep, ep->atid);
456
457 skb = get_skb(NULL, wrlen, GFP_KERNEL);
458 if (!skb) {
459 printk(KERN_ERR MOD "%s - failed to alloc skb.\n",
460 __func__);
461 return -ENOMEM;
462 }
463 set_wr_txq(skb, CPL_PRIORITY_SETUP, ep->ctrlq_idx);
464
465 cxgb4_best_mtu(ep->com.dev->rdev.lldi.mtus, ep->mtu, &mtu_idx);
466 wscale = compute_wscale(rcv_win);
467 opt0 = KEEP_ALIVE(1) |
468 DELACK(1) |
469 WND_SCALE(wscale) |
470 MSS_IDX(mtu_idx) |
471 L2T_IDX(ep->l2t->idx) |
472 TX_CHAN(ep->tx_chan) |
473 SMAC_SEL(ep->smac_idx) |
474 DSCP(ep->tos) |
475 ULP_MODE(ULP_MODE_TCPDDP) |
476 RCV_BUFSIZ(rcv_win>>10);
477 opt2 = RX_CHANNEL(0) |
478 RSS_QUEUE_VALID | RSS_QUEUE(ep->rss_qid);
479 if (enable_tcp_timestamps)
480 opt2 |= TSTAMPS_EN(1);
481 if (enable_tcp_sack)
482 opt2 |= SACK_EN(1);
483 if (wscale && enable_tcp_window_scaling)
484 opt2 |= WND_SCALE_EN(1);
485 t4_set_arp_err_handler(skb, NULL, act_open_req_arp_failure);
486
487 req = (struct cpl_act_open_req *) skb_put(skb, wrlen);
488 INIT_TP_WR(req, 0);
489 OPCODE_TID(req) = cpu_to_be32(
490 MK_OPCODE_TID(CPL_ACT_OPEN_REQ, ((ep->rss_qid<<14)|ep->atid)));
491 req->local_port = ep->com.local_addr.sin_port;
492 req->peer_port = ep->com.remote_addr.sin_port;
493 req->local_ip = ep->com.local_addr.sin_addr.s_addr;
494 req->peer_ip = ep->com.remote_addr.sin_addr.s_addr;
495 req->opt0 = cpu_to_be64(opt0);
496 req->params = 0;
497 req->opt2 = cpu_to_be32(opt2);
498 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
499 }
500
501 static void send_mpa_req(struct c4iw_ep *ep, struct sk_buff *skb,
502 u8 mpa_rev_to_use)
503 {
504 int mpalen, wrlen;
505 struct fw_ofld_tx_data_wr *req;
506 struct mpa_message *mpa;
507 struct mpa_v2_conn_params mpa_v2_params;
508
509 PDBG("%s ep %p tid %u pd_len %d\n", __func__, ep, ep->hwtid, ep->plen);
510
511 BUG_ON(skb_cloned(skb));
512
513 mpalen = sizeof(*mpa) + ep->plen;
514 if (mpa_rev_to_use == 2)
515 mpalen += sizeof(struct mpa_v2_conn_params);
516 wrlen = roundup(mpalen + sizeof *req, 16);
517 skb = get_skb(skb, wrlen, GFP_KERNEL);
518 if (!skb) {
519 connect_reply_upcall(ep, -ENOMEM);
520 return;
521 }
522 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
523
524 req = (struct fw_ofld_tx_data_wr *)skb_put(skb, wrlen);
525 memset(req, 0, wrlen);
526 req->op_to_immdlen = cpu_to_be32(
527 FW_WR_OP(FW_OFLD_TX_DATA_WR) |
528 FW_WR_COMPL(1) |
529 FW_WR_IMMDLEN(mpalen));
530 req->flowid_len16 = cpu_to_be32(
531 FW_WR_FLOWID(ep->hwtid) |
532 FW_WR_LEN16(wrlen >> 4));
533 req->plen = cpu_to_be32(mpalen);
534 req->tunnel_to_proxy = cpu_to_be32(
535 FW_OFLD_TX_DATA_WR_FLUSH(1) |
536 FW_OFLD_TX_DATA_WR_SHOVE(1));
537
538 mpa = (struct mpa_message *)(req + 1);
539 memcpy(mpa->key, MPA_KEY_REQ, sizeof(mpa->key));
540 mpa->flags = (crc_enabled ? MPA_CRC : 0) |
541 (markers_enabled ? MPA_MARKERS : 0) |
542 (mpa_rev_to_use == 2 ? MPA_ENHANCED_RDMA_CONN : 0);
543 mpa->private_data_size = htons(ep->plen);
544 mpa->revision = mpa_rev_to_use;
545 if (mpa_rev_to_use == 1) {
546 ep->tried_with_mpa_v1 = 1;
547 ep->retry_with_mpa_v1 = 0;
548 }
549
550 if (mpa_rev_to_use == 2) {
551 mpa->private_data_size +=
552 htons(sizeof(struct mpa_v2_conn_params));
553 mpa_v2_params.ird = htons((u16)ep->ird);
554 mpa_v2_params.ord = htons((u16)ep->ord);
555
556 if (peer2peer) {
557 mpa_v2_params.ird |= htons(MPA_V2_PEER2PEER_MODEL);
558 if (p2p_type == FW_RI_INIT_P2PTYPE_RDMA_WRITE)
559 mpa_v2_params.ord |=
560 htons(MPA_V2_RDMA_WRITE_RTR);
561 else if (p2p_type == FW_RI_INIT_P2PTYPE_READ_REQ)
562 mpa_v2_params.ord |=
563 htons(MPA_V2_RDMA_READ_RTR);
564 }
565 memcpy(mpa->private_data, &mpa_v2_params,
566 sizeof(struct mpa_v2_conn_params));
567
568 if (ep->plen)
569 memcpy(mpa->private_data +
570 sizeof(struct mpa_v2_conn_params),
571 ep->mpa_pkt + sizeof(*mpa), ep->plen);
572 } else
573 if (ep->plen)
574 memcpy(mpa->private_data,
575 ep->mpa_pkt + sizeof(*mpa), ep->plen);
576
577 /*
578 * Reference the mpa skb. This ensures the data area
579 * will remain in memory until the hw acks the tx.
580 * Function fw4_ack() will deref it.
581 */
582 skb_get(skb);
583 t4_set_arp_err_handler(skb, NULL, arp_failure_discard);
584 BUG_ON(ep->mpa_skb);
585 ep->mpa_skb = skb;
586 c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
587 start_ep_timer(ep);
588 state_set(&ep->com, MPA_REQ_SENT);
589 ep->mpa_attr.initiator = 1;
590 return;
591 }
592
593 static int send_mpa_reject(struct c4iw_ep *ep, const void *pdata, u8 plen)
594 {
595 int mpalen, wrlen;
596 struct fw_ofld_tx_data_wr *req;
597 struct mpa_message *mpa;
598 struct sk_buff *skb;
599 struct mpa_v2_conn_params mpa_v2_params;
600
601 PDBG("%s ep %p tid %u pd_len %d\n", __func__, ep, ep->hwtid, ep->plen);
602
603 mpalen = sizeof(*mpa) + plen;
604 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn)
605 mpalen += sizeof(struct mpa_v2_conn_params);
606 wrlen = roundup(mpalen + sizeof *req, 16);
607
608 skb = get_skb(NULL, wrlen, GFP_KERNEL);
609 if (!skb) {
610 printk(KERN_ERR MOD "%s - cannot alloc skb!\n", __func__);
611 return -ENOMEM;
612 }
613 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
614
615 req = (struct fw_ofld_tx_data_wr *)skb_put(skb, wrlen);
616 memset(req, 0, wrlen);
617 req->op_to_immdlen = cpu_to_be32(
618 FW_WR_OP(FW_OFLD_TX_DATA_WR) |
619 FW_WR_COMPL(1) |
620 FW_WR_IMMDLEN(mpalen));
621 req->flowid_len16 = cpu_to_be32(
622 FW_WR_FLOWID(ep->hwtid) |
623 FW_WR_LEN16(wrlen >> 4));
624 req->plen = cpu_to_be32(mpalen);
625 req->tunnel_to_proxy = cpu_to_be32(
626 FW_OFLD_TX_DATA_WR_FLUSH(1) |
627 FW_OFLD_TX_DATA_WR_SHOVE(1));
628
629 mpa = (struct mpa_message *)(req + 1);
630 memset(mpa, 0, sizeof(*mpa));
631 memcpy(mpa->key, MPA_KEY_REP, sizeof(mpa->key));
632 mpa->flags = MPA_REJECT;
633 mpa->revision = mpa_rev;
634 mpa->private_data_size = htons(plen);
635
636 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) {
637 mpa->flags |= MPA_ENHANCED_RDMA_CONN;
638 mpa->private_data_size +=
639 htons(sizeof(struct mpa_v2_conn_params));
640 mpa_v2_params.ird = htons(((u16)ep->ird) |
641 (peer2peer ? MPA_V2_PEER2PEER_MODEL :
642 0));
643 mpa_v2_params.ord = htons(((u16)ep->ord) | (peer2peer ?
644 (p2p_type ==
645 FW_RI_INIT_P2PTYPE_RDMA_WRITE ?
646 MPA_V2_RDMA_WRITE_RTR : p2p_type ==
647 FW_RI_INIT_P2PTYPE_READ_REQ ?
648 MPA_V2_RDMA_READ_RTR : 0) : 0));
649 memcpy(mpa->private_data, &mpa_v2_params,
650 sizeof(struct mpa_v2_conn_params));
651
652 if (ep->plen)
653 memcpy(mpa->private_data +
654 sizeof(struct mpa_v2_conn_params), pdata, plen);
655 } else
656 if (plen)
657 memcpy(mpa->private_data, pdata, plen);
658
659 /*
660 * Reference the mpa skb again. This ensures the data area
661 * will remain in memory until the hw acks the tx.
662 * Function fw4_ack() will deref it.
663 */
664 skb_get(skb);
665 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
666 t4_set_arp_err_handler(skb, NULL, arp_failure_discard);
667 BUG_ON(ep->mpa_skb);
668 ep->mpa_skb = skb;
669 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
670 }
671
672 static int send_mpa_reply(struct c4iw_ep *ep, const void *pdata, u8 plen)
673 {
674 int mpalen, wrlen;
675 struct fw_ofld_tx_data_wr *req;
676 struct mpa_message *mpa;
677 struct sk_buff *skb;
678 struct mpa_v2_conn_params mpa_v2_params;
679
680 PDBG("%s ep %p tid %u pd_len %d\n", __func__, ep, ep->hwtid, ep->plen);
681
682 mpalen = sizeof(*mpa) + plen;
683 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn)
684 mpalen += sizeof(struct mpa_v2_conn_params);
685 wrlen = roundup(mpalen + sizeof *req, 16);
686
687 skb = get_skb(NULL, wrlen, GFP_KERNEL);
688 if (!skb) {
689 printk(KERN_ERR MOD "%s - cannot alloc skb!\n", __func__);
690 return -ENOMEM;
691 }
692 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
693
694 req = (struct fw_ofld_tx_data_wr *) skb_put(skb, wrlen);
695 memset(req, 0, wrlen);
696 req->op_to_immdlen = cpu_to_be32(
697 FW_WR_OP(FW_OFLD_TX_DATA_WR) |
698 FW_WR_COMPL(1) |
699 FW_WR_IMMDLEN(mpalen));
700 req->flowid_len16 = cpu_to_be32(
701 FW_WR_FLOWID(ep->hwtid) |
702 FW_WR_LEN16(wrlen >> 4));
703 req->plen = cpu_to_be32(mpalen);
704 req->tunnel_to_proxy = cpu_to_be32(
705 FW_OFLD_TX_DATA_WR_FLUSH(1) |
706 FW_OFLD_TX_DATA_WR_SHOVE(1));
707
708 mpa = (struct mpa_message *)(req + 1);
709 memset(mpa, 0, sizeof(*mpa));
710 memcpy(mpa->key, MPA_KEY_REP, sizeof(mpa->key));
711 mpa->flags = (ep->mpa_attr.crc_enabled ? MPA_CRC : 0) |
712 (markers_enabled ? MPA_MARKERS : 0);
713 mpa->revision = ep->mpa_attr.version;
714 mpa->private_data_size = htons(plen);
715
716 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) {
717 mpa->flags |= MPA_ENHANCED_RDMA_CONN;
718 mpa->private_data_size +=
719 htons(sizeof(struct mpa_v2_conn_params));
720 mpa_v2_params.ird = htons((u16)ep->ird);
721 mpa_v2_params.ord = htons((u16)ep->ord);
722 if (peer2peer && (ep->mpa_attr.p2p_type !=
723 FW_RI_INIT_P2PTYPE_DISABLED)) {
724 mpa_v2_params.ird |= htons(MPA_V2_PEER2PEER_MODEL);
725
726 if (p2p_type == FW_RI_INIT_P2PTYPE_RDMA_WRITE)
727 mpa_v2_params.ord |=
728 htons(MPA_V2_RDMA_WRITE_RTR);
729 else if (p2p_type == FW_RI_INIT_P2PTYPE_READ_REQ)
730 mpa_v2_params.ord |=
731 htons(MPA_V2_RDMA_READ_RTR);
732 }
733
734 memcpy(mpa->private_data, &mpa_v2_params,
735 sizeof(struct mpa_v2_conn_params));
736
737 if (ep->plen)
738 memcpy(mpa->private_data +
739 sizeof(struct mpa_v2_conn_params), pdata, plen);
740 } else
741 if (plen)
742 memcpy(mpa->private_data, pdata, plen);
743
744 /*
745 * Reference the mpa skb. This ensures the data area
746 * will remain in memory until the hw acks the tx.
747 * Function fw4_ack() will deref it.
748 */
749 skb_get(skb);
750 t4_set_arp_err_handler(skb, NULL, arp_failure_discard);
751 ep->mpa_skb = skb;
752 state_set(&ep->com, MPA_REP_SENT);
753 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
754 }
755
756 static int act_establish(struct c4iw_dev *dev, struct sk_buff *skb)
757 {
758 struct c4iw_ep *ep;
759 struct cpl_act_establish *req = cplhdr(skb);
760 unsigned int tid = GET_TID(req);
761 unsigned int atid = GET_TID_TID(ntohl(req->tos_atid));
762 struct tid_info *t = dev->rdev.lldi.tids;
763
764 ep = lookup_atid(t, atid);
765
766 PDBG("%s ep %p tid %u snd_isn %u rcv_isn %u\n", __func__, ep, tid,
767 be32_to_cpu(req->snd_isn), be32_to_cpu(req->rcv_isn));
768
769 dst_confirm(ep->dst);
770
771 /* setup the hwtid for this connection */
772 ep->hwtid = tid;
773 cxgb4_insert_tid(t, ep, tid);
774
775 ep->snd_seq = be32_to_cpu(req->snd_isn);
776 ep->rcv_seq = be32_to_cpu(req->rcv_isn);
777
778 set_emss(ep, ntohs(req->tcp_opt));
779
780 /* dealloc the atid */
781 cxgb4_free_atid(t, atid);
782
783 /* start MPA negotiation */
784 send_flowc(ep, NULL);
785 if (ep->retry_with_mpa_v1)
786 send_mpa_req(ep, skb, 1);
787 else
788 send_mpa_req(ep, skb, mpa_rev);
789
790 return 0;
791 }
792
793 static void close_complete_upcall(struct c4iw_ep *ep)
794 {
795 struct iw_cm_event event;
796
797 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
798 memset(&event, 0, sizeof(event));
799 event.event = IW_CM_EVENT_CLOSE;
800 if (ep->com.cm_id) {
801 PDBG("close complete delivered ep %p cm_id %p tid %u\n",
802 ep, ep->com.cm_id, ep->hwtid);
803 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
804 ep->com.cm_id->rem_ref(ep->com.cm_id);
805 ep->com.cm_id = NULL;
806 ep->com.qp = NULL;
807 }
808 }
809
810 static int abort_connection(struct c4iw_ep *ep, struct sk_buff *skb, gfp_t gfp)
811 {
812 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
813 close_complete_upcall(ep);
814 state_set(&ep->com, ABORTING);
815 return send_abort(ep, skb, gfp);
816 }
817
818 static void peer_close_upcall(struct c4iw_ep *ep)
819 {
820 struct iw_cm_event event;
821
822 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
823 memset(&event, 0, sizeof(event));
824 event.event = IW_CM_EVENT_DISCONNECT;
825 if (ep->com.cm_id) {
826 PDBG("peer close delivered ep %p cm_id %p tid %u\n",
827 ep, ep->com.cm_id, ep->hwtid);
828 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
829 }
830 }
831
832 static void peer_abort_upcall(struct c4iw_ep *ep)
833 {
834 struct iw_cm_event event;
835
836 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
837 memset(&event, 0, sizeof(event));
838 event.event = IW_CM_EVENT_CLOSE;
839 event.status = -ECONNRESET;
840 if (ep->com.cm_id) {
841 PDBG("abort delivered ep %p cm_id %p tid %u\n", ep,
842 ep->com.cm_id, ep->hwtid);
843 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
844 ep->com.cm_id->rem_ref(ep->com.cm_id);
845 ep->com.cm_id = NULL;
846 ep->com.qp = NULL;
847 }
848 }
849
850 static void connect_reply_upcall(struct c4iw_ep *ep, int status)
851 {
852 struct iw_cm_event event;
853
854 PDBG("%s ep %p tid %u status %d\n", __func__, ep, ep->hwtid, status);
855 memset(&event, 0, sizeof(event));
856 event.event = IW_CM_EVENT_CONNECT_REPLY;
857 event.status = status;
858 event.local_addr = ep->com.local_addr;
859 event.remote_addr = ep->com.remote_addr;
860
861 if ((status == 0) || (status == -ECONNREFUSED)) {
862 if (!ep->tried_with_mpa_v1) {
863 /* this means MPA_v2 is used */
864 event.private_data_len = ep->plen -
865 sizeof(struct mpa_v2_conn_params);
866 event.private_data = ep->mpa_pkt +
867 sizeof(struct mpa_message) +
868 sizeof(struct mpa_v2_conn_params);
869 } else {
870 /* this means MPA_v1 is used */
871 event.private_data_len = ep->plen;
872 event.private_data = ep->mpa_pkt +
873 sizeof(struct mpa_message);
874 }
875 }
876
877 PDBG("%s ep %p tid %u status %d\n", __func__, ep,
878 ep->hwtid, status);
879 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
880
881 if (status < 0) {
882 ep->com.cm_id->rem_ref(ep->com.cm_id);
883 ep->com.cm_id = NULL;
884 ep->com.qp = NULL;
885 }
886 }
887
888 static void connect_request_upcall(struct c4iw_ep *ep)
889 {
890 struct iw_cm_event event;
891
892 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
893 memset(&event, 0, sizeof(event));
894 event.event = IW_CM_EVENT_CONNECT_REQUEST;
895 event.local_addr = ep->com.local_addr;
896 event.remote_addr = ep->com.remote_addr;
897 event.provider_data = ep;
898 if (!ep->tried_with_mpa_v1) {
899 /* this means MPA_v2 is used */
900 event.ord = ep->ord;
901 event.ird = ep->ird;
902 event.private_data_len = ep->plen -
903 sizeof(struct mpa_v2_conn_params);
904 event.private_data = ep->mpa_pkt + sizeof(struct mpa_message) +
905 sizeof(struct mpa_v2_conn_params);
906 } else {
907 /* this means MPA_v1 is used. Send max supported */
908 event.ord = c4iw_max_read_depth;
909 event.ird = c4iw_max_read_depth;
910 event.private_data_len = ep->plen;
911 event.private_data = ep->mpa_pkt + sizeof(struct mpa_message);
912 }
913 if (state_read(&ep->parent_ep->com) != DEAD) {
914 c4iw_get_ep(&ep->com);
915 ep->parent_ep->com.cm_id->event_handler(
916 ep->parent_ep->com.cm_id,
917 &event);
918 }
919 c4iw_put_ep(&ep->parent_ep->com);
920 ep->parent_ep = NULL;
921 }
922
923 static void established_upcall(struct c4iw_ep *ep)
924 {
925 struct iw_cm_event event;
926
927 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
928 memset(&event, 0, sizeof(event));
929 event.event = IW_CM_EVENT_ESTABLISHED;
930 event.ird = ep->ird;
931 event.ord = ep->ord;
932 if (ep->com.cm_id) {
933 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
934 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
935 }
936 }
937
938 static int update_rx_credits(struct c4iw_ep *ep, u32 credits)
939 {
940 struct cpl_rx_data_ack *req;
941 struct sk_buff *skb;
942 int wrlen = roundup(sizeof *req, 16);
943
944 PDBG("%s ep %p tid %u credits %u\n", __func__, ep, ep->hwtid, credits);
945 skb = get_skb(NULL, wrlen, GFP_KERNEL);
946 if (!skb) {
947 printk(KERN_ERR MOD "update_rx_credits - cannot alloc skb!\n");
948 return 0;
949 }
950
951 req = (struct cpl_rx_data_ack *) skb_put(skb, wrlen);
952 memset(req, 0, wrlen);
953 INIT_TP_WR(req, ep->hwtid);
954 OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_RX_DATA_ACK,
955 ep->hwtid));
956 req->credit_dack = cpu_to_be32(credits | RX_FORCE_ACK(1) |
957 F_RX_DACK_CHANGE |
958 V_RX_DACK_MODE(dack_mode));
959 set_wr_txq(skb, CPL_PRIORITY_ACK, ep->ctrlq_idx);
960 c4iw_ofld_send(&ep->com.dev->rdev, skb);
961 return credits;
962 }
963
964 static void process_mpa_reply(struct c4iw_ep *ep, struct sk_buff *skb)
965 {
966 struct mpa_message *mpa;
967 struct mpa_v2_conn_params *mpa_v2_params;
968 u16 plen;
969 u16 resp_ird, resp_ord;
970 u8 rtr_mismatch = 0, insuff_ird = 0;
971 struct c4iw_qp_attributes attrs;
972 enum c4iw_qp_attr_mask mask;
973 int err;
974
975 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
976
977 /*
978 * Stop mpa timer. If it expired, then the state has
979 * changed and we bail since ep_timeout already aborted
980 * the connection.
981 */
982 stop_ep_timer(ep);
983 if (state_read(&ep->com) != MPA_REQ_SENT)
984 return;
985
986 /*
987 * If we get more than the supported amount of private data
988 * then we must fail this connection.
989 */
990 if (ep->mpa_pkt_len + skb->len > sizeof(ep->mpa_pkt)) {
991 err = -EINVAL;
992 goto err;
993 }
994
995 /*
996 * copy the new data into our accumulation buffer.
997 */
998 skb_copy_from_linear_data(skb, &(ep->mpa_pkt[ep->mpa_pkt_len]),
999 skb->len);
1000 ep->mpa_pkt_len += skb->len;
1001
1002 /*
1003 * if we don't even have the mpa message, then bail.
1004 */
1005 if (ep->mpa_pkt_len < sizeof(*mpa))
1006 return;
1007 mpa = (struct mpa_message *) ep->mpa_pkt;
1008
1009 /* Validate MPA header. */
1010 if (mpa->revision > mpa_rev) {
1011 printk(KERN_ERR MOD "%s MPA version mismatch. Local = %d,"
1012 " Received = %d\n", __func__, mpa_rev, mpa->revision);
1013 err = -EPROTO;
1014 goto err;
1015 }
1016 if (memcmp(mpa->key, MPA_KEY_REP, sizeof(mpa->key))) {
1017 err = -EPROTO;
1018 goto err;
1019 }
1020
1021 plen = ntohs(mpa->private_data_size);
1022
1023 /*
1024 * Fail if there's too much private data.
1025 */
1026 if (plen > MPA_MAX_PRIVATE_DATA) {
1027 err = -EPROTO;
1028 goto err;
1029 }
1030
1031 /*
1032 * If plen does not account for pkt size
1033 */
1034 if (ep->mpa_pkt_len > (sizeof(*mpa) + plen)) {
1035 err = -EPROTO;
1036 goto err;
1037 }
1038
1039 ep->plen = (u8) plen;
1040
1041 /*
1042 * If we don't have all the pdata yet, then bail.
1043 * We'll continue process when more data arrives.
1044 */
1045 if (ep->mpa_pkt_len < (sizeof(*mpa) + plen))
1046 return;
1047
1048 if (mpa->flags & MPA_REJECT) {
1049 err = -ECONNREFUSED;
1050 goto err;
1051 }
1052
1053 /*
1054 * If we get here we have accumulated the entire mpa
1055 * start reply message including private data. And
1056 * the MPA header is valid.
1057 */
1058 state_set(&ep->com, FPDU_MODE);
1059 ep->mpa_attr.crc_enabled = (mpa->flags & MPA_CRC) | crc_enabled ? 1 : 0;
1060 ep->mpa_attr.recv_marker_enabled = markers_enabled;
1061 ep->mpa_attr.xmit_marker_enabled = mpa->flags & MPA_MARKERS ? 1 : 0;
1062 ep->mpa_attr.version = mpa->revision;
1063 ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED;
1064
1065 if (mpa->revision == 2) {
1066 ep->mpa_attr.enhanced_rdma_conn =
1067 mpa->flags & MPA_ENHANCED_RDMA_CONN ? 1 : 0;
1068 if (ep->mpa_attr.enhanced_rdma_conn) {
1069 mpa_v2_params = (struct mpa_v2_conn_params *)
1070 (ep->mpa_pkt + sizeof(*mpa));
1071 resp_ird = ntohs(mpa_v2_params->ird) &
1072 MPA_V2_IRD_ORD_MASK;
1073 resp_ord = ntohs(mpa_v2_params->ord) &
1074 MPA_V2_IRD_ORD_MASK;
1075
1076 /*
1077 * This is a double-check. Ideally, below checks are
1078 * not required since ird/ord stuff has been taken
1079 * care of in c4iw_accept_cr
1080 */
1081 if ((ep->ird < resp_ord) || (ep->ord > resp_ird)) {
1082 err = -ENOMEM;
1083 ep->ird = resp_ord;
1084 ep->ord = resp_ird;
1085 insuff_ird = 1;
1086 }
1087
1088 if (ntohs(mpa_v2_params->ird) &
1089 MPA_V2_PEER2PEER_MODEL) {
1090 if (ntohs(mpa_v2_params->ord) &
1091 MPA_V2_RDMA_WRITE_RTR)
1092 ep->mpa_attr.p2p_type =
1093 FW_RI_INIT_P2PTYPE_RDMA_WRITE;
1094 else if (ntohs(mpa_v2_params->ord) &
1095 MPA_V2_RDMA_READ_RTR)
1096 ep->mpa_attr.p2p_type =
1097 FW_RI_INIT_P2PTYPE_READ_REQ;
1098 }
1099 }
1100 } else if (mpa->revision == 1)
1101 if (peer2peer)
1102 ep->mpa_attr.p2p_type = p2p_type;
1103
1104 PDBG("%s - crc_enabled=%d, recv_marker_enabled=%d, "
1105 "xmit_marker_enabled=%d, version=%d p2p_type=%d local-p2p_type = "
1106 "%d\n", __func__, ep->mpa_attr.crc_enabled,
1107 ep->mpa_attr.recv_marker_enabled,
1108 ep->mpa_attr.xmit_marker_enabled, ep->mpa_attr.version,
1109 ep->mpa_attr.p2p_type, p2p_type);
1110
1111 /*
1112 * If responder's RTR does not match with that of initiator, assign
1113 * FW_RI_INIT_P2PTYPE_DISABLED in mpa attributes so that RTR is not
1114 * generated when moving QP to RTS state.
1115 * A TERM message will be sent after QP has moved to RTS state
1116 */
1117 if ((ep->mpa_attr.version == 2) &&
1118 (ep->mpa_attr.p2p_type != p2p_type)) {
1119 ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED;
1120 rtr_mismatch = 1;
1121 }
1122
1123 attrs.mpa_attr = ep->mpa_attr;
1124 attrs.max_ird = ep->ird;
1125 attrs.max_ord = ep->ord;
1126 attrs.llp_stream_handle = ep;
1127 attrs.next_state = C4IW_QP_STATE_RTS;
1128
1129 mask = C4IW_QP_ATTR_NEXT_STATE |
1130 C4IW_QP_ATTR_LLP_STREAM_HANDLE | C4IW_QP_ATTR_MPA_ATTR |
1131 C4IW_QP_ATTR_MAX_IRD | C4IW_QP_ATTR_MAX_ORD;
1132
1133 /* bind QP and TID with INIT_WR */
1134 err = c4iw_modify_qp(ep->com.qp->rhp,
1135 ep->com.qp, mask, &attrs, 1);
1136 if (err)
1137 goto err;
1138
1139 /*
1140 * If responder's RTR requirement did not match with what initiator
1141 * supports, generate TERM message
1142 */
1143 if (rtr_mismatch) {
1144 printk(KERN_ERR "%s: RTR mismatch, sending TERM\n", __func__);
1145 attrs.layer_etype = LAYER_MPA | DDP_LLP;
1146 attrs.ecode = MPA_NOMATCH_RTR;
1147 attrs.next_state = C4IW_QP_STATE_TERMINATE;
1148 err = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
1149 C4IW_QP_ATTR_NEXT_STATE, &attrs, 0);
1150 err = -ENOMEM;
1151 goto out;
1152 }
1153
1154 /*
1155 * Generate TERM if initiator IRD is not sufficient for responder
1156 * provided ORD. Currently, we do the same behaviour even when
1157 * responder provided IRD is also not sufficient as regards to
1158 * initiator ORD.
1159 */
1160 if (insuff_ird) {
1161 printk(KERN_ERR "%s: Insufficient IRD, sending TERM\n",
1162 __func__);
1163 attrs.layer_etype = LAYER_MPA | DDP_LLP;
1164 attrs.ecode = MPA_INSUFF_IRD;
1165 attrs.next_state = C4IW_QP_STATE_TERMINATE;
1166 err = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
1167 C4IW_QP_ATTR_NEXT_STATE, &attrs, 0);
1168 err = -ENOMEM;
1169 goto out;
1170 }
1171 goto out;
1172 err:
1173 state_set(&ep->com, ABORTING);
1174 send_abort(ep, skb, GFP_KERNEL);
1175 out:
1176 connect_reply_upcall(ep, err);
1177 return;
1178 }
1179
1180 static void process_mpa_request(struct c4iw_ep *ep, struct sk_buff *skb)
1181 {
1182 struct mpa_message *mpa;
1183 struct mpa_v2_conn_params *mpa_v2_params;
1184 u16 plen;
1185
1186 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1187
1188 if (state_read(&ep->com) != MPA_REQ_WAIT)
1189 return;
1190
1191 /*
1192 * If we get more than the supported amount of private data
1193 * then we must fail this connection.
1194 */
1195 if (ep->mpa_pkt_len + skb->len > sizeof(ep->mpa_pkt)) {
1196 stop_ep_timer(ep);
1197 abort_connection(ep, skb, GFP_KERNEL);
1198 return;
1199 }
1200
1201 PDBG("%s enter (%s line %u)\n", __func__, __FILE__, __LINE__);
1202
1203 /*
1204 * Copy the new data into our accumulation buffer.
1205 */
1206 skb_copy_from_linear_data(skb, &(ep->mpa_pkt[ep->mpa_pkt_len]),
1207 skb->len);
1208 ep->mpa_pkt_len += skb->len;
1209
1210 /*
1211 * If we don't even have the mpa message, then bail.
1212 * We'll continue process when more data arrives.
1213 */
1214 if (ep->mpa_pkt_len < sizeof(*mpa))
1215 return;
1216
1217 PDBG("%s enter (%s line %u)\n", __func__, __FILE__, __LINE__);
1218 stop_ep_timer(ep);
1219 mpa = (struct mpa_message *) ep->mpa_pkt;
1220
1221 /*
1222 * Validate MPA Header.
1223 */
1224 if (mpa->revision > mpa_rev) {
1225 printk(KERN_ERR MOD "%s MPA version mismatch. Local = %d,"
1226 " Received = %d\n", __func__, mpa_rev, mpa->revision);
1227 abort_connection(ep, skb, GFP_KERNEL);
1228 return;
1229 }
1230
1231 if (memcmp(mpa->key, MPA_KEY_REQ, sizeof(mpa->key))) {
1232 abort_connection(ep, skb, GFP_KERNEL);
1233 return;
1234 }
1235
1236 plen = ntohs(mpa->private_data_size);
1237
1238 /*
1239 * Fail if there's too much private data.
1240 */
1241 if (plen > MPA_MAX_PRIVATE_DATA) {
1242 abort_connection(ep, skb, GFP_KERNEL);
1243 return;
1244 }
1245
1246 /*
1247 * If plen does not account for pkt size
1248 */
1249 if (ep->mpa_pkt_len > (sizeof(*mpa) + plen)) {
1250 abort_connection(ep, skb, GFP_KERNEL);
1251 return;
1252 }
1253 ep->plen = (u8) plen;
1254
1255 /*
1256 * If we don't have all the pdata yet, then bail.
1257 */
1258 if (ep->mpa_pkt_len < (sizeof(*mpa) + plen))
1259 return;
1260
1261 /*
1262 * If we get here we have accumulated the entire mpa
1263 * start reply message including private data.
1264 */
1265 ep->mpa_attr.initiator = 0;
1266 ep->mpa_attr.crc_enabled = (mpa->flags & MPA_CRC) | crc_enabled ? 1 : 0;
1267 ep->mpa_attr.recv_marker_enabled = markers_enabled;
1268 ep->mpa_attr.xmit_marker_enabled = mpa->flags & MPA_MARKERS ? 1 : 0;
1269 ep->mpa_attr.version = mpa->revision;
1270 if (mpa->revision == 1)
1271 ep->tried_with_mpa_v1 = 1;
1272 ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED;
1273
1274 if (mpa->revision == 2) {
1275 ep->mpa_attr.enhanced_rdma_conn =
1276 mpa->flags & MPA_ENHANCED_RDMA_CONN ? 1 : 0;
1277 if (ep->mpa_attr.enhanced_rdma_conn) {
1278 mpa_v2_params = (struct mpa_v2_conn_params *)
1279 (ep->mpa_pkt + sizeof(*mpa));
1280 ep->ird = ntohs(mpa_v2_params->ird) &
1281 MPA_V2_IRD_ORD_MASK;
1282 ep->ord = ntohs(mpa_v2_params->ord) &
1283 MPA_V2_IRD_ORD_MASK;
1284 if (ntohs(mpa_v2_params->ird) & MPA_V2_PEER2PEER_MODEL)
1285 if (peer2peer) {
1286 if (ntohs(mpa_v2_params->ord) &
1287 MPA_V2_RDMA_WRITE_RTR)
1288 ep->mpa_attr.p2p_type =
1289 FW_RI_INIT_P2PTYPE_RDMA_WRITE;
1290 else if (ntohs(mpa_v2_params->ord) &
1291 MPA_V2_RDMA_READ_RTR)
1292 ep->mpa_attr.p2p_type =
1293 FW_RI_INIT_P2PTYPE_READ_REQ;
1294 }
1295 }
1296 } else if (mpa->revision == 1)
1297 if (peer2peer)
1298 ep->mpa_attr.p2p_type = p2p_type;
1299
1300 PDBG("%s - crc_enabled=%d, recv_marker_enabled=%d, "
1301 "xmit_marker_enabled=%d, version=%d p2p_type=%d\n", __func__,
1302 ep->mpa_attr.crc_enabled, ep->mpa_attr.recv_marker_enabled,
1303 ep->mpa_attr.xmit_marker_enabled, ep->mpa_attr.version,
1304 ep->mpa_attr.p2p_type);
1305
1306 state_set(&ep->com, MPA_REQ_RCVD);
1307
1308 /* drive upcall */
1309 connect_request_upcall(ep);
1310 return;
1311 }
1312
1313 static int rx_data(struct c4iw_dev *dev, struct sk_buff *skb)
1314 {
1315 struct c4iw_ep *ep;
1316 struct cpl_rx_data *hdr = cplhdr(skb);
1317 unsigned int dlen = ntohs(hdr->len);
1318 unsigned int tid = GET_TID(hdr);
1319 struct tid_info *t = dev->rdev.lldi.tids;
1320
1321 ep = lookup_tid(t, tid);
1322 PDBG("%s ep %p tid %u dlen %u\n", __func__, ep, ep->hwtid, dlen);
1323 skb_pull(skb, sizeof(*hdr));
1324 skb_trim(skb, dlen);
1325
1326 ep->rcv_seq += dlen;
1327 BUG_ON(ep->rcv_seq != (ntohl(hdr->seq) + dlen));
1328
1329 /* update RX credits */
1330 update_rx_credits(ep, dlen);
1331
1332 switch (state_read(&ep->com)) {
1333 case MPA_REQ_SENT:
1334 process_mpa_reply(ep, skb);
1335 break;
1336 case MPA_REQ_WAIT:
1337 process_mpa_request(ep, skb);
1338 break;
1339 case MPA_REP_SENT:
1340 break;
1341 default:
1342 printk(KERN_ERR MOD "%s Unexpected streaming data."
1343 " ep %p state %d tid %u\n",
1344 __func__, ep, state_read(&ep->com), ep->hwtid);
1345
1346 /*
1347 * The ep will timeout and inform the ULP of the failure.
1348 * See ep_timeout().
1349 */
1350 break;
1351 }
1352 return 0;
1353 }
1354
1355 static int abort_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
1356 {
1357 struct c4iw_ep *ep;
1358 struct cpl_abort_rpl_rss *rpl = cplhdr(skb);
1359 int release = 0;
1360 unsigned int tid = GET_TID(rpl);
1361 struct tid_info *t = dev->rdev.lldi.tids;
1362
1363 ep = lookup_tid(t, tid);
1364 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1365 BUG_ON(!ep);
1366 mutex_lock(&ep->com.mutex);
1367 switch (ep->com.state) {
1368 case ABORTING:
1369 __state_set(&ep->com, DEAD);
1370 release = 1;
1371 break;
1372 default:
1373 printk(KERN_ERR "%s ep %p state %d\n",
1374 __func__, ep, ep->com.state);
1375 break;
1376 }
1377 mutex_unlock(&ep->com.mutex);
1378
1379 if (release)
1380 release_ep_resources(ep);
1381 return 0;
1382 }
1383
1384 /*
1385 * Return whether a failed active open has allocated a TID
1386 */
1387 static inline int act_open_has_tid(int status)
1388 {
1389 return status != CPL_ERR_TCAM_FULL && status != CPL_ERR_CONN_EXIST &&
1390 status != CPL_ERR_ARP_MISS;
1391 }
1392
1393 static int act_open_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
1394 {
1395 struct c4iw_ep *ep;
1396 struct cpl_act_open_rpl *rpl = cplhdr(skb);
1397 unsigned int atid = GET_TID_TID(GET_AOPEN_ATID(
1398 ntohl(rpl->atid_status)));
1399 struct tid_info *t = dev->rdev.lldi.tids;
1400 int status = GET_AOPEN_STATUS(ntohl(rpl->atid_status));
1401
1402 ep = lookup_atid(t, atid);
1403
1404 PDBG("%s ep %p atid %u status %u errno %d\n", __func__, ep, atid,
1405 status, status2errno(status));
1406
1407 if (status == CPL_ERR_RTX_NEG_ADVICE) {
1408 printk(KERN_WARNING MOD "Connection problems for atid %u\n",
1409 atid);
1410 return 0;
1411 }
1412
1413 connect_reply_upcall(ep, status2errno(status));
1414 state_set(&ep->com, DEAD);
1415
1416 if (status && act_open_has_tid(status))
1417 cxgb4_remove_tid(ep->com.dev->rdev.lldi.tids, 0, GET_TID(rpl));
1418
1419 cxgb4_free_atid(t, atid);
1420 dst_release(ep->dst);
1421 cxgb4_l2t_release(ep->l2t);
1422 c4iw_put_ep(&ep->com);
1423
1424 return 0;
1425 }
1426
1427 static int pass_open_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
1428 {
1429 struct cpl_pass_open_rpl *rpl = cplhdr(skb);
1430 struct tid_info *t = dev->rdev.lldi.tids;
1431 unsigned int stid = GET_TID(rpl);
1432 struct c4iw_listen_ep *ep = lookup_stid(t, stid);
1433
1434 if (!ep) {
1435 printk(KERN_ERR MOD "stid %d lookup failure!\n", stid);
1436 return 0;
1437 }
1438 PDBG("%s ep %p status %d error %d\n", __func__, ep,
1439 rpl->status, status2errno(rpl->status));
1440 c4iw_wake_up(&ep->com.wr_wait, status2errno(rpl->status));
1441
1442 return 0;
1443 }
1444
1445 static int listen_stop(struct c4iw_listen_ep *ep)
1446 {
1447 struct sk_buff *skb;
1448 struct cpl_close_listsvr_req *req;
1449
1450 PDBG("%s ep %p\n", __func__, ep);
1451 skb = get_skb(NULL, sizeof(*req), GFP_KERNEL);
1452 if (!skb) {
1453 printk(KERN_ERR MOD "%s - failed to alloc skb\n", __func__);
1454 return -ENOMEM;
1455 }
1456 req = (struct cpl_close_listsvr_req *) skb_put(skb, sizeof(*req));
1457 INIT_TP_WR(req, 0);
1458 OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_CLOSE_LISTSRV_REQ,
1459 ep->stid));
1460 req->reply_ctrl = cpu_to_be16(
1461 QUEUENO(ep->com.dev->rdev.lldi.rxq_ids[0]));
1462 set_wr_txq(skb, CPL_PRIORITY_SETUP, 0);
1463 return c4iw_ofld_send(&ep->com.dev->rdev, skb);
1464 }
1465
1466 static int close_listsrv_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
1467 {
1468 struct cpl_close_listsvr_rpl *rpl = cplhdr(skb);
1469 struct tid_info *t = dev->rdev.lldi.tids;
1470 unsigned int stid = GET_TID(rpl);
1471 struct c4iw_listen_ep *ep = lookup_stid(t, stid);
1472
1473 PDBG("%s ep %p\n", __func__, ep);
1474 c4iw_wake_up(&ep->com.wr_wait, status2errno(rpl->status));
1475 return 0;
1476 }
1477
1478 static void accept_cr(struct c4iw_ep *ep, __be32 peer_ip, struct sk_buff *skb,
1479 struct cpl_pass_accept_req *req)
1480 {
1481 struct cpl_pass_accept_rpl *rpl;
1482 unsigned int mtu_idx;
1483 u64 opt0;
1484 u32 opt2;
1485 int wscale;
1486
1487 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1488 BUG_ON(skb_cloned(skb));
1489 skb_trim(skb, sizeof(*rpl));
1490 skb_get(skb);
1491 cxgb4_best_mtu(ep->com.dev->rdev.lldi.mtus, ep->mtu, &mtu_idx);
1492 wscale = compute_wscale(rcv_win);
1493 opt0 = KEEP_ALIVE(1) |
1494 DELACK(1) |
1495 WND_SCALE(wscale) |
1496 MSS_IDX(mtu_idx) |
1497 L2T_IDX(ep->l2t->idx) |
1498 TX_CHAN(ep->tx_chan) |
1499 SMAC_SEL(ep->smac_idx) |
1500 DSCP(ep->tos) |
1501 ULP_MODE(ULP_MODE_TCPDDP) |
1502 RCV_BUFSIZ(rcv_win>>10);
1503 opt2 = RX_CHANNEL(0) |
1504 RSS_QUEUE_VALID | RSS_QUEUE(ep->rss_qid);
1505
1506 if (enable_tcp_timestamps && req->tcpopt.tstamp)
1507 opt2 |= TSTAMPS_EN(1);
1508 if (enable_tcp_sack && req->tcpopt.sack)
1509 opt2 |= SACK_EN(1);
1510 if (wscale && enable_tcp_window_scaling)
1511 opt2 |= WND_SCALE_EN(1);
1512
1513 rpl = cplhdr(skb);
1514 INIT_TP_WR(rpl, ep->hwtid);
1515 OPCODE_TID(rpl) = cpu_to_be32(MK_OPCODE_TID(CPL_PASS_ACCEPT_RPL,
1516 ep->hwtid));
1517 rpl->opt0 = cpu_to_be64(opt0);
1518 rpl->opt2 = cpu_to_be32(opt2);
1519 set_wr_txq(skb, CPL_PRIORITY_SETUP, ep->ctrlq_idx);
1520 c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
1521
1522 return;
1523 }
1524
1525 static void reject_cr(struct c4iw_dev *dev, u32 hwtid, __be32 peer_ip,
1526 struct sk_buff *skb)
1527 {
1528 PDBG("%s c4iw_dev %p tid %u peer_ip %x\n", __func__, dev, hwtid,
1529 peer_ip);
1530 BUG_ON(skb_cloned(skb));
1531 skb_trim(skb, sizeof(struct cpl_tid_release));
1532 skb_get(skb);
1533 release_tid(&dev->rdev, hwtid, skb);
1534 return;
1535 }
1536
1537 static void get_4tuple(struct cpl_pass_accept_req *req,
1538 __be32 *local_ip, __be32 *peer_ip,
1539 __be16 *local_port, __be16 *peer_port)
1540 {
1541 int eth_len = G_ETH_HDR_LEN(be32_to_cpu(req->hdr_len));
1542 int ip_len = G_IP_HDR_LEN(be32_to_cpu(req->hdr_len));
1543 struct iphdr *ip = (struct iphdr *)((u8 *)(req + 1) + eth_len);
1544 struct tcphdr *tcp = (struct tcphdr *)
1545 ((u8 *)(req + 1) + eth_len + ip_len);
1546
1547 PDBG("%s saddr 0x%x daddr 0x%x sport %u dport %u\n", __func__,
1548 ntohl(ip->saddr), ntohl(ip->daddr), ntohs(tcp->source),
1549 ntohs(tcp->dest));
1550
1551 *peer_ip = ip->saddr;
1552 *local_ip = ip->daddr;
1553 *peer_port = tcp->source;
1554 *local_port = tcp->dest;
1555
1556 return;
1557 }
1558
1559 static int import_ep(struct c4iw_ep *ep, __be32 peer_ip, struct dst_entry *dst,
1560 struct c4iw_dev *cdev, bool clear_mpa_v1)
1561 {
1562 struct neighbour *n;
1563 int err, step;
1564
1565 rcu_read_lock();
1566 n = dst_get_neighbour_noref(dst);
1567 err = -ENODEV;
1568 if (!n)
1569 goto out;
1570 err = -ENOMEM;
1571 if (n->dev->flags & IFF_LOOPBACK) {
1572 struct net_device *pdev;
1573
1574 pdev = ip_dev_find(&init_net, peer_ip);
1575 ep->l2t = cxgb4_l2t_get(cdev->rdev.lldi.l2t,
1576 n, pdev, 0);
1577 if (!ep->l2t)
1578 goto out;
1579 ep->mtu = pdev->mtu;
1580 ep->tx_chan = cxgb4_port_chan(pdev);
1581 ep->smac_idx = (cxgb4_port_viid(pdev) & 0x7F) << 1;
1582 step = cdev->rdev.lldi.ntxq /
1583 cdev->rdev.lldi.nchan;
1584 ep->txq_idx = cxgb4_port_idx(pdev) * step;
1585 step = cdev->rdev.lldi.nrxq /
1586 cdev->rdev.lldi.nchan;
1587 ep->ctrlq_idx = cxgb4_port_idx(pdev);
1588 ep->rss_qid = cdev->rdev.lldi.rxq_ids[
1589 cxgb4_port_idx(pdev) * step];
1590 dev_put(pdev);
1591 } else {
1592 ep->l2t = cxgb4_l2t_get(cdev->rdev.lldi.l2t,
1593 n, n->dev, 0);
1594 if (!ep->l2t)
1595 goto out;
1596 ep->mtu = dst_mtu(ep->dst);
1597 ep->tx_chan = cxgb4_port_chan(n->dev);
1598 ep->smac_idx = (cxgb4_port_viid(n->dev) & 0x7F) << 1;
1599 step = cdev->rdev.lldi.ntxq /
1600 cdev->rdev.lldi.nchan;
1601 ep->txq_idx = cxgb4_port_idx(n->dev) * step;
1602 ep->ctrlq_idx = cxgb4_port_idx(n->dev);
1603 step = cdev->rdev.lldi.nrxq /
1604 cdev->rdev.lldi.nchan;
1605 ep->rss_qid = cdev->rdev.lldi.rxq_ids[
1606 cxgb4_port_idx(n->dev) * step];
1607
1608 if (clear_mpa_v1) {
1609 ep->retry_with_mpa_v1 = 0;
1610 ep->tried_with_mpa_v1 = 0;
1611 }
1612 }
1613 err = 0;
1614 out:
1615 rcu_read_unlock();
1616
1617 return err;
1618 }
1619
1620 static int pass_accept_req(struct c4iw_dev *dev, struct sk_buff *skb)
1621 {
1622 struct c4iw_ep *child_ep, *parent_ep;
1623 struct cpl_pass_accept_req *req = cplhdr(skb);
1624 unsigned int stid = GET_POPEN_TID(ntohl(req->tos_stid));
1625 struct tid_info *t = dev->rdev.lldi.tids;
1626 unsigned int hwtid = GET_TID(req);
1627 struct dst_entry *dst;
1628 struct rtable *rt;
1629 __be32 local_ip, peer_ip;
1630 __be16 local_port, peer_port;
1631 int err;
1632
1633 parent_ep = lookup_stid(t, stid);
1634 PDBG("%s parent ep %p tid %u\n", __func__, parent_ep, hwtid);
1635
1636 get_4tuple(req, &local_ip, &peer_ip, &local_port, &peer_port);
1637
1638 if (state_read(&parent_ep->com) != LISTEN) {
1639 printk(KERN_ERR "%s - listening ep not in LISTEN\n",
1640 __func__);
1641 goto reject;
1642 }
1643
1644 /* Find output route */
1645 rt = find_route(dev, local_ip, peer_ip, local_port, peer_port,
1646 GET_POPEN_TOS(ntohl(req->tos_stid)));
1647 if (!rt) {
1648 printk(KERN_ERR MOD "%s - failed to find dst entry!\n",
1649 __func__);
1650 goto reject;
1651 }
1652 dst = &rt->dst;
1653
1654 child_ep = alloc_ep(sizeof(*child_ep), GFP_KERNEL);
1655 if (!child_ep) {
1656 printk(KERN_ERR MOD "%s - failed to allocate ep entry!\n",
1657 __func__);
1658 dst_release(dst);
1659 goto reject;
1660 }
1661
1662 err = import_ep(child_ep, peer_ip, dst, dev, false);
1663 if (err) {
1664 printk(KERN_ERR MOD "%s - failed to allocate l2t entry!\n",
1665 __func__);
1666 dst_release(dst);
1667 kfree(child_ep);
1668 goto reject;
1669 }
1670
1671 state_set(&child_ep->com, CONNECTING);
1672 child_ep->com.dev = dev;
1673 child_ep->com.cm_id = NULL;
1674 child_ep->com.local_addr.sin_family = PF_INET;
1675 child_ep->com.local_addr.sin_port = local_port;
1676 child_ep->com.local_addr.sin_addr.s_addr = local_ip;
1677 child_ep->com.remote_addr.sin_family = PF_INET;
1678 child_ep->com.remote_addr.sin_port = peer_port;
1679 child_ep->com.remote_addr.sin_addr.s_addr = peer_ip;
1680 c4iw_get_ep(&parent_ep->com);
1681 child_ep->parent_ep = parent_ep;
1682 child_ep->tos = GET_POPEN_TOS(ntohl(req->tos_stid));
1683 child_ep->dst = dst;
1684 child_ep->hwtid = hwtid;
1685
1686 PDBG("%s tx_chan %u smac_idx %u rss_qid %u\n", __func__,
1687 child_ep->tx_chan, child_ep->smac_idx, child_ep->rss_qid);
1688
1689 init_timer(&child_ep->timer);
1690 cxgb4_insert_tid(t, child_ep, hwtid);
1691 accept_cr(child_ep, peer_ip, skb, req);
1692 goto out;
1693 reject:
1694 reject_cr(dev, hwtid, peer_ip, skb);
1695 out:
1696 return 0;
1697 }
1698
1699 static int pass_establish(struct c4iw_dev *dev, struct sk_buff *skb)
1700 {
1701 struct c4iw_ep *ep;
1702 struct cpl_pass_establish *req = cplhdr(skb);
1703 struct tid_info *t = dev->rdev.lldi.tids;
1704 unsigned int tid = GET_TID(req);
1705
1706 ep = lookup_tid(t, tid);
1707 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1708 ep->snd_seq = be32_to_cpu(req->snd_isn);
1709 ep->rcv_seq = be32_to_cpu(req->rcv_isn);
1710
1711 set_emss(ep, ntohs(req->tcp_opt));
1712
1713 dst_confirm(ep->dst);
1714 state_set(&ep->com, MPA_REQ_WAIT);
1715 start_ep_timer(ep);
1716 send_flowc(ep, skb);
1717
1718 return 0;
1719 }
1720
1721 static int peer_close(struct c4iw_dev *dev, struct sk_buff *skb)
1722 {
1723 struct cpl_peer_close *hdr = cplhdr(skb);
1724 struct c4iw_ep *ep;
1725 struct c4iw_qp_attributes attrs;
1726 int disconnect = 1;
1727 int release = 0;
1728 struct tid_info *t = dev->rdev.lldi.tids;
1729 unsigned int tid = GET_TID(hdr);
1730 int ret;
1731
1732 ep = lookup_tid(t, tid);
1733 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1734 dst_confirm(ep->dst);
1735
1736 mutex_lock(&ep->com.mutex);
1737 switch (ep->com.state) {
1738 case MPA_REQ_WAIT:
1739 __state_set(&ep->com, CLOSING);
1740 break;
1741 case MPA_REQ_SENT:
1742 __state_set(&ep->com, CLOSING);
1743 connect_reply_upcall(ep, -ECONNRESET);
1744 break;
1745 case MPA_REQ_RCVD:
1746
1747 /*
1748 * We're gonna mark this puppy DEAD, but keep
1749 * the reference on it until the ULP accepts or
1750 * rejects the CR. Also wake up anyone waiting
1751 * in rdma connection migration (see c4iw_accept_cr()).
1752 */
1753 __state_set(&ep->com, CLOSING);
1754 PDBG("waking up ep %p tid %u\n", ep, ep->hwtid);
1755 c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
1756 break;
1757 case MPA_REP_SENT:
1758 __state_set(&ep->com, CLOSING);
1759 PDBG("waking up ep %p tid %u\n", ep, ep->hwtid);
1760 c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
1761 break;
1762 case FPDU_MODE:
1763 start_ep_timer(ep);
1764 __state_set(&ep->com, CLOSING);
1765 attrs.next_state = C4IW_QP_STATE_CLOSING;
1766 ret = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
1767 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
1768 if (ret != -ECONNRESET) {
1769 peer_close_upcall(ep);
1770 disconnect = 1;
1771 }
1772 break;
1773 case ABORTING:
1774 disconnect = 0;
1775 break;
1776 case CLOSING:
1777 __state_set(&ep->com, MORIBUND);
1778 disconnect = 0;
1779 break;
1780 case MORIBUND:
1781 stop_ep_timer(ep);
1782 if (ep->com.cm_id && ep->com.qp) {
1783 attrs.next_state = C4IW_QP_STATE_IDLE;
1784 c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
1785 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
1786 }
1787 close_complete_upcall(ep);
1788 __state_set(&ep->com, DEAD);
1789 release = 1;
1790 disconnect = 0;
1791 break;
1792 case DEAD:
1793 disconnect = 0;
1794 break;
1795 default:
1796 BUG_ON(1);
1797 }
1798 mutex_unlock(&ep->com.mutex);
1799 if (disconnect)
1800 c4iw_ep_disconnect(ep, 0, GFP_KERNEL);
1801 if (release)
1802 release_ep_resources(ep);
1803 return 0;
1804 }
1805
1806 /*
1807 * Returns whether an ABORT_REQ_RSS message is a negative advice.
1808 */
1809 static int is_neg_adv_abort(unsigned int status)
1810 {
1811 return status == CPL_ERR_RTX_NEG_ADVICE ||
1812 status == CPL_ERR_PERSIST_NEG_ADVICE;
1813 }
1814
1815 static int c4iw_reconnect(struct c4iw_ep *ep)
1816 {
1817 struct rtable *rt;
1818 int err = 0;
1819
1820 PDBG("%s qp %p cm_id %p\n", __func__, ep->com.qp, ep->com.cm_id);
1821 init_timer(&ep->timer);
1822
1823 /*
1824 * Allocate an active TID to initiate a TCP connection.
1825 */
1826 ep->atid = cxgb4_alloc_atid(ep->com.dev->rdev.lldi.tids, ep);
1827 if (ep->atid == -1) {
1828 printk(KERN_ERR MOD "%s - cannot alloc atid.\n", __func__);
1829 err = -ENOMEM;
1830 goto fail2;
1831 }
1832
1833 /* find a route */
1834 rt = find_route(ep->com.dev,
1835 ep->com.cm_id->local_addr.sin_addr.s_addr,
1836 ep->com.cm_id->remote_addr.sin_addr.s_addr,
1837 ep->com.cm_id->local_addr.sin_port,
1838 ep->com.cm_id->remote_addr.sin_port, 0);
1839 if (!rt) {
1840 printk(KERN_ERR MOD "%s - cannot find route.\n", __func__);
1841 err = -EHOSTUNREACH;
1842 goto fail3;
1843 }
1844 ep->dst = &rt->dst;
1845
1846 err = import_ep(ep, ep->com.cm_id->remote_addr.sin_addr.s_addr,
1847 ep->dst, ep->com.dev, false);
1848 if (err) {
1849 printk(KERN_ERR MOD "%s - cannot alloc l2e.\n", __func__);
1850 goto fail4;
1851 }
1852
1853 PDBG("%s txq_idx %u tx_chan %u smac_idx %u rss_qid %u l2t_idx %u\n",
1854 __func__, ep->txq_idx, ep->tx_chan, ep->smac_idx, ep->rss_qid,
1855 ep->l2t->idx);
1856
1857 state_set(&ep->com, CONNECTING);
1858 ep->tos = 0;
1859
1860 /* send connect request to rnic */
1861 err = send_connect(ep);
1862 if (!err)
1863 goto out;
1864
1865 cxgb4_l2t_release(ep->l2t);
1866 fail4:
1867 dst_release(ep->dst);
1868 fail3:
1869 cxgb4_free_atid(ep->com.dev->rdev.lldi.tids, ep->atid);
1870 fail2:
1871 /*
1872 * remember to send notification to upper layer.
1873 * We are in here so the upper layer is not aware that this is
1874 * re-connect attempt and so, upper layer is still waiting for
1875 * response of 1st connect request.
1876 */
1877 connect_reply_upcall(ep, -ECONNRESET);
1878 c4iw_put_ep(&ep->com);
1879 out:
1880 return err;
1881 }
1882
1883 static int peer_abort(struct c4iw_dev *dev, struct sk_buff *skb)
1884 {
1885 struct cpl_abort_req_rss *req = cplhdr(skb);
1886 struct c4iw_ep *ep;
1887 struct cpl_abort_rpl *rpl;
1888 struct sk_buff *rpl_skb;
1889 struct c4iw_qp_attributes attrs;
1890 int ret;
1891 int release = 0;
1892 struct tid_info *t = dev->rdev.lldi.tids;
1893 unsigned int tid = GET_TID(req);
1894
1895 ep = lookup_tid(t, tid);
1896 if (is_neg_adv_abort(req->status)) {
1897 PDBG("%s neg_adv_abort ep %p tid %u\n", __func__, ep,
1898 ep->hwtid);
1899 return 0;
1900 }
1901 PDBG("%s ep %p tid %u state %u\n", __func__, ep, ep->hwtid,
1902 ep->com.state);
1903
1904 /*
1905 * Wake up any threads in rdma_init() or rdma_fini().
1906 * However, this is not needed if com state is just
1907 * MPA_REQ_SENT
1908 */
1909 if (ep->com.state != MPA_REQ_SENT)
1910 c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
1911
1912 mutex_lock(&ep->com.mutex);
1913 switch (ep->com.state) {
1914 case CONNECTING:
1915 break;
1916 case MPA_REQ_WAIT:
1917 stop_ep_timer(ep);
1918 break;
1919 case MPA_REQ_SENT:
1920 stop_ep_timer(ep);
1921 if (mpa_rev == 2 && ep->tried_with_mpa_v1)
1922 connect_reply_upcall(ep, -ECONNRESET);
1923 else {
1924 /*
1925 * we just don't send notification upwards because we
1926 * want to retry with mpa_v1 without upper layers even
1927 * knowing it.
1928 *
1929 * do some housekeeping so as to re-initiate the
1930 * connection
1931 */
1932 PDBG("%s: mpa_rev=%d. Retrying with mpav1\n", __func__,
1933 mpa_rev);
1934 ep->retry_with_mpa_v1 = 1;
1935 }
1936 break;
1937 case MPA_REP_SENT:
1938 break;
1939 case MPA_REQ_RCVD:
1940 break;
1941 case MORIBUND:
1942 case CLOSING:
1943 stop_ep_timer(ep);
1944 /*FALLTHROUGH*/
1945 case FPDU_MODE:
1946 if (ep->com.cm_id && ep->com.qp) {
1947 attrs.next_state = C4IW_QP_STATE_ERROR;
1948 ret = c4iw_modify_qp(ep->com.qp->rhp,
1949 ep->com.qp, C4IW_QP_ATTR_NEXT_STATE,
1950 &attrs, 1);
1951 if (ret)
1952 printk(KERN_ERR MOD
1953 "%s - qp <- error failed!\n",
1954 __func__);
1955 }
1956 peer_abort_upcall(ep);
1957 break;
1958 case ABORTING:
1959 break;
1960 case DEAD:
1961 PDBG("%s PEER_ABORT IN DEAD STATE!!!!\n", __func__);
1962 mutex_unlock(&ep->com.mutex);
1963 return 0;
1964 default:
1965 BUG_ON(1);
1966 break;
1967 }
1968 dst_confirm(ep->dst);
1969 if (ep->com.state != ABORTING) {
1970 __state_set(&ep->com, DEAD);
1971 /* we don't release if we want to retry with mpa_v1 */
1972 if (!ep->retry_with_mpa_v1)
1973 release = 1;
1974 }
1975 mutex_unlock(&ep->com.mutex);
1976
1977 rpl_skb = get_skb(skb, sizeof(*rpl), GFP_KERNEL);
1978 if (!rpl_skb) {
1979 printk(KERN_ERR MOD "%s - cannot allocate skb!\n",
1980 __func__);
1981 release = 1;
1982 goto out;
1983 }
1984 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
1985 rpl = (struct cpl_abort_rpl *) skb_put(rpl_skb, sizeof(*rpl));
1986 INIT_TP_WR(rpl, ep->hwtid);
1987 OPCODE_TID(rpl) = cpu_to_be32(MK_OPCODE_TID(CPL_ABORT_RPL, ep->hwtid));
1988 rpl->cmd = CPL_ABORT_NO_RST;
1989 c4iw_ofld_send(&ep->com.dev->rdev, rpl_skb);
1990 out:
1991 if (release)
1992 release_ep_resources(ep);
1993
1994 /* retry with mpa-v1 */
1995 if (ep && ep->retry_with_mpa_v1) {
1996 cxgb4_remove_tid(ep->com.dev->rdev.lldi.tids, 0, ep->hwtid);
1997 dst_release(ep->dst);
1998 cxgb4_l2t_release(ep->l2t);
1999 c4iw_reconnect(ep);
2000 }
2001
2002 return 0;
2003 }
2004
2005 static int close_con_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
2006 {
2007 struct c4iw_ep *ep;
2008 struct c4iw_qp_attributes attrs;
2009 struct cpl_close_con_rpl *rpl = cplhdr(skb);
2010 int release = 0;
2011 struct tid_info *t = dev->rdev.lldi.tids;
2012 unsigned int tid = GET_TID(rpl);
2013
2014 ep = lookup_tid(t, tid);
2015
2016 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
2017 BUG_ON(!ep);
2018
2019 /* The cm_id may be null if we failed to connect */
2020 mutex_lock(&ep->com.mutex);
2021 switch (ep->com.state) {
2022 case CLOSING:
2023 __state_set(&ep->com, MORIBUND);
2024 break;
2025 case MORIBUND:
2026 stop_ep_timer(ep);
2027 if ((ep->com.cm_id) && (ep->com.qp)) {
2028 attrs.next_state = C4IW_QP_STATE_IDLE;
2029 c4iw_modify_qp(ep->com.qp->rhp,
2030 ep->com.qp,
2031 C4IW_QP_ATTR_NEXT_STATE,
2032 &attrs, 1);
2033 }
2034 close_complete_upcall(ep);
2035 __state_set(&ep->com, DEAD);
2036 release = 1;
2037 break;
2038 case ABORTING:
2039 case DEAD:
2040 break;
2041 default:
2042 BUG_ON(1);
2043 break;
2044 }
2045 mutex_unlock(&ep->com.mutex);
2046 if (release)
2047 release_ep_resources(ep);
2048 return 0;
2049 }
2050
2051 static int terminate(struct c4iw_dev *dev, struct sk_buff *skb)
2052 {
2053 struct cpl_rdma_terminate *rpl = cplhdr(skb);
2054 struct tid_info *t = dev->rdev.lldi.tids;
2055 unsigned int tid = GET_TID(rpl);
2056 struct c4iw_ep *ep;
2057 struct c4iw_qp_attributes attrs;
2058
2059 ep = lookup_tid(t, tid);
2060 BUG_ON(!ep);
2061
2062 if (ep && ep->com.qp) {
2063 printk(KERN_WARNING MOD "TERM received tid %u qpid %u\n", tid,
2064 ep->com.qp->wq.sq.qid);
2065 attrs.next_state = C4IW_QP_STATE_TERMINATE;
2066 c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
2067 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
2068 } else
2069 printk(KERN_WARNING MOD "TERM received tid %u no ep/qp\n", tid);
2070
2071 return 0;
2072 }
2073
2074 /*
2075 * Upcall from the adapter indicating data has been transmitted.
2076 * For us its just the single MPA request or reply. We can now free
2077 * the skb holding the mpa message.
2078 */
2079 static int fw4_ack(struct c4iw_dev *dev, struct sk_buff *skb)
2080 {
2081 struct c4iw_ep *ep;
2082 struct cpl_fw4_ack *hdr = cplhdr(skb);
2083 u8 credits = hdr->credits;
2084 unsigned int tid = GET_TID(hdr);
2085 struct tid_info *t = dev->rdev.lldi.tids;
2086
2087
2088 ep = lookup_tid(t, tid);
2089 PDBG("%s ep %p tid %u credits %u\n", __func__, ep, ep->hwtid, credits);
2090 if (credits == 0) {
2091 PDBG("%s 0 credit ack ep %p tid %u state %u\n",
2092 __func__, ep, ep->hwtid, state_read(&ep->com));
2093 return 0;
2094 }
2095
2096 dst_confirm(ep->dst);
2097 if (ep->mpa_skb) {
2098 PDBG("%s last streaming msg ack ep %p tid %u state %u "
2099 "initiator %u freeing skb\n", __func__, ep, ep->hwtid,
2100 state_read(&ep->com), ep->mpa_attr.initiator ? 1 : 0);
2101 kfree_skb(ep->mpa_skb);
2102 ep->mpa_skb = NULL;
2103 }
2104 return 0;
2105 }
2106
2107 int c4iw_reject_cr(struct iw_cm_id *cm_id, const void *pdata, u8 pdata_len)
2108 {
2109 int err;
2110 struct c4iw_ep *ep = to_ep(cm_id);
2111 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
2112
2113 if (state_read(&ep->com) == DEAD) {
2114 c4iw_put_ep(&ep->com);
2115 return -ECONNRESET;
2116 }
2117 BUG_ON(state_read(&ep->com) != MPA_REQ_RCVD);
2118 if (mpa_rev == 0)
2119 abort_connection(ep, NULL, GFP_KERNEL);
2120 else {
2121 err = send_mpa_reject(ep, pdata, pdata_len);
2122 err = c4iw_ep_disconnect(ep, 0, GFP_KERNEL);
2123 }
2124 c4iw_put_ep(&ep->com);
2125 return 0;
2126 }
2127
2128 int c4iw_accept_cr(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
2129 {
2130 int err;
2131 struct c4iw_qp_attributes attrs;
2132 enum c4iw_qp_attr_mask mask;
2133 struct c4iw_ep *ep = to_ep(cm_id);
2134 struct c4iw_dev *h = to_c4iw_dev(cm_id->device);
2135 struct c4iw_qp *qp = get_qhp(h, conn_param->qpn);
2136
2137 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
2138 if (state_read(&ep->com) == DEAD) {
2139 err = -ECONNRESET;
2140 goto err;
2141 }
2142
2143 BUG_ON(state_read(&ep->com) != MPA_REQ_RCVD);
2144 BUG_ON(!qp);
2145
2146 if ((conn_param->ord > c4iw_max_read_depth) ||
2147 (conn_param->ird > c4iw_max_read_depth)) {
2148 abort_connection(ep, NULL, GFP_KERNEL);
2149 err = -EINVAL;
2150 goto err;
2151 }
2152
2153 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) {
2154 if (conn_param->ord > ep->ird) {
2155 ep->ird = conn_param->ird;
2156 ep->ord = conn_param->ord;
2157 send_mpa_reject(ep, conn_param->private_data,
2158 conn_param->private_data_len);
2159 abort_connection(ep, NULL, GFP_KERNEL);
2160 err = -ENOMEM;
2161 goto err;
2162 }
2163 if (conn_param->ird > ep->ord) {
2164 if (!ep->ord)
2165 conn_param->ird = 1;
2166 else {
2167 abort_connection(ep, NULL, GFP_KERNEL);
2168 err = -ENOMEM;
2169 goto err;
2170 }
2171 }
2172
2173 }
2174 ep->ird = conn_param->ird;
2175 ep->ord = conn_param->ord;
2176
2177 if (ep->mpa_attr.version != 2)
2178 if (peer2peer && ep->ird == 0)
2179 ep->ird = 1;
2180
2181 PDBG("%s %d ird %d ord %d\n", __func__, __LINE__, ep->ird, ep->ord);
2182
2183 cm_id->add_ref(cm_id);
2184 ep->com.cm_id = cm_id;
2185 ep->com.qp = qp;
2186
2187 /* bind QP to EP and move to RTS */
2188 attrs.mpa_attr = ep->mpa_attr;
2189 attrs.max_ird = ep->ird;
2190 attrs.max_ord = ep->ord;
2191 attrs.llp_stream_handle = ep;
2192 attrs.next_state = C4IW_QP_STATE_RTS;
2193
2194 /* bind QP and TID with INIT_WR */
2195 mask = C4IW_QP_ATTR_NEXT_STATE |
2196 C4IW_QP_ATTR_LLP_STREAM_HANDLE |
2197 C4IW_QP_ATTR_MPA_ATTR |
2198 C4IW_QP_ATTR_MAX_IRD |
2199 C4IW_QP_ATTR_MAX_ORD;
2200
2201 err = c4iw_modify_qp(ep->com.qp->rhp,
2202 ep->com.qp, mask, &attrs, 1);
2203 if (err)
2204 goto err1;
2205 err = send_mpa_reply(ep, conn_param->private_data,
2206 conn_param->private_data_len);
2207 if (err)
2208 goto err1;
2209
2210 state_set(&ep->com, FPDU_MODE);
2211 established_upcall(ep);
2212 c4iw_put_ep(&ep->com);
2213 return 0;
2214 err1:
2215 ep->com.cm_id = NULL;
2216 ep->com.qp = NULL;
2217 cm_id->rem_ref(cm_id);
2218 err:
2219 c4iw_put_ep(&ep->com);
2220 return err;
2221 }
2222
2223 int c4iw_connect(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
2224 {
2225 struct c4iw_dev *dev = to_c4iw_dev(cm_id->device);
2226 struct c4iw_ep *ep;
2227 struct rtable *rt;
2228 int err = 0;
2229
2230 if ((conn_param->ord > c4iw_max_read_depth) ||
2231 (conn_param->ird > c4iw_max_read_depth)) {
2232 err = -EINVAL;
2233 goto out;
2234 }
2235 ep = alloc_ep(sizeof(*ep), GFP_KERNEL);
2236 if (!ep) {
2237 printk(KERN_ERR MOD "%s - cannot alloc ep.\n", __func__);
2238 err = -ENOMEM;
2239 goto out;
2240 }
2241 init_timer(&ep->timer);
2242 ep->plen = conn_param->private_data_len;
2243 if (ep->plen)
2244 memcpy(ep->mpa_pkt + sizeof(struct mpa_message),
2245 conn_param->private_data, ep->plen);
2246 ep->ird = conn_param->ird;
2247 ep->ord = conn_param->ord;
2248
2249 if (peer2peer && ep->ord == 0)
2250 ep->ord = 1;
2251
2252 cm_id->add_ref(cm_id);
2253 ep->com.dev = dev;
2254 ep->com.cm_id = cm_id;
2255 ep->com.qp = get_qhp(dev, conn_param->qpn);
2256 BUG_ON(!ep->com.qp);
2257 PDBG("%s qpn 0x%x qp %p cm_id %p\n", __func__, conn_param->qpn,
2258 ep->com.qp, cm_id);
2259
2260 /*
2261 * Allocate an active TID to initiate a TCP connection.
2262 */
2263 ep->atid = cxgb4_alloc_atid(dev->rdev.lldi.tids, ep);
2264 if (ep->atid == -1) {
2265 printk(KERN_ERR MOD "%s - cannot alloc atid.\n", __func__);
2266 err = -ENOMEM;
2267 goto fail2;
2268 }
2269
2270 PDBG("%s saddr 0x%x sport 0x%x raddr 0x%x rport 0x%x\n", __func__,
2271 ntohl(cm_id->local_addr.sin_addr.s_addr),
2272 ntohs(cm_id->local_addr.sin_port),
2273 ntohl(cm_id->remote_addr.sin_addr.s_addr),
2274 ntohs(cm_id->remote_addr.sin_port));
2275
2276 /* find a route */
2277 rt = find_route(dev,
2278 cm_id->local_addr.sin_addr.s_addr,
2279 cm_id->remote_addr.sin_addr.s_addr,
2280 cm_id->local_addr.sin_port,
2281 cm_id->remote_addr.sin_port, 0);
2282 if (!rt) {
2283 printk(KERN_ERR MOD "%s - cannot find route.\n", __func__);
2284 err = -EHOSTUNREACH;
2285 goto fail3;
2286 }
2287 ep->dst = &rt->dst;
2288
2289 err = import_ep(ep, cm_id->remote_addr.sin_addr.s_addr,
2290 ep->dst, ep->com.dev, true);
2291 if (err) {
2292 printk(KERN_ERR MOD "%s - cannot alloc l2e.\n", __func__);
2293 goto fail4;
2294 }
2295
2296 PDBG("%s txq_idx %u tx_chan %u smac_idx %u rss_qid %u l2t_idx %u\n",
2297 __func__, ep->txq_idx, ep->tx_chan, ep->smac_idx, ep->rss_qid,
2298 ep->l2t->idx);
2299
2300 state_set(&ep->com, CONNECTING);
2301 ep->tos = 0;
2302 ep->com.local_addr = cm_id->local_addr;
2303 ep->com.remote_addr = cm_id->remote_addr;
2304
2305 /* send connect request to rnic */
2306 err = send_connect(ep);
2307 if (!err)
2308 goto out;
2309
2310 cxgb4_l2t_release(ep->l2t);
2311 fail4:
2312 dst_release(ep->dst);
2313 fail3:
2314 cxgb4_free_atid(ep->com.dev->rdev.lldi.tids, ep->atid);
2315 fail2:
2316 cm_id->rem_ref(cm_id);
2317 c4iw_put_ep(&ep->com);
2318 out:
2319 return err;
2320 }
2321
2322 int c4iw_create_listen(struct iw_cm_id *cm_id, int backlog)
2323 {
2324 int err = 0;
2325 struct c4iw_dev *dev = to_c4iw_dev(cm_id->device);
2326 struct c4iw_listen_ep *ep;
2327
2328
2329 might_sleep();
2330
2331 ep = alloc_ep(sizeof(*ep), GFP_KERNEL);
2332 if (!ep) {
2333 printk(KERN_ERR MOD "%s - cannot alloc ep.\n", __func__);
2334 err = -ENOMEM;
2335 goto fail1;
2336 }
2337 PDBG("%s ep %p\n", __func__, ep);
2338 cm_id->add_ref(cm_id);
2339 ep->com.cm_id = cm_id;
2340 ep->com.dev = dev;
2341 ep->backlog = backlog;
2342 ep->com.local_addr = cm_id->local_addr;
2343
2344 /*
2345 * Allocate a server TID.
2346 */
2347 ep->stid = cxgb4_alloc_stid(dev->rdev.lldi.tids, PF_INET, ep);
2348 if (ep->stid == -1) {
2349 printk(KERN_ERR MOD "%s - cannot alloc stid.\n", __func__);
2350 err = -ENOMEM;
2351 goto fail2;
2352 }
2353
2354 state_set(&ep->com, LISTEN);
2355 c4iw_init_wr_wait(&ep->com.wr_wait);
2356 err = cxgb4_create_server(ep->com.dev->rdev.lldi.ports[0], ep->stid,
2357 ep->com.local_addr.sin_addr.s_addr,
2358 ep->com.local_addr.sin_port,
2359 ep->com.dev->rdev.lldi.rxq_ids[0]);
2360 if (err)
2361 goto fail3;
2362
2363 /* wait for pass_open_rpl */
2364 err = c4iw_wait_for_reply(&ep->com.dev->rdev, &ep->com.wr_wait, 0, 0,
2365 __func__);
2366 if (!err) {
2367 cm_id->provider_data = ep;
2368 goto out;
2369 }
2370 fail3:
2371 cxgb4_free_stid(ep->com.dev->rdev.lldi.tids, ep->stid, PF_INET);
2372 fail2:
2373 cm_id->rem_ref(cm_id);
2374 c4iw_put_ep(&ep->com);
2375 fail1:
2376 out:
2377 return err;
2378 }
2379
2380 int c4iw_destroy_listen(struct iw_cm_id *cm_id)
2381 {
2382 int err;
2383 struct c4iw_listen_ep *ep = to_listen_ep(cm_id);
2384
2385 PDBG("%s ep %p\n", __func__, ep);
2386
2387 might_sleep();
2388 state_set(&ep->com, DEAD);
2389 c4iw_init_wr_wait(&ep->com.wr_wait);
2390 err = listen_stop(ep);
2391 if (err)
2392 goto done;
2393 err = c4iw_wait_for_reply(&ep->com.dev->rdev, &ep->com.wr_wait, 0, 0,
2394 __func__);
2395 cxgb4_free_stid(ep->com.dev->rdev.lldi.tids, ep->stid, PF_INET);
2396 done:
2397 cm_id->rem_ref(cm_id);
2398 c4iw_put_ep(&ep->com);
2399 return err;
2400 }
2401
2402 int c4iw_ep_disconnect(struct c4iw_ep *ep, int abrupt, gfp_t gfp)
2403 {
2404 int ret = 0;
2405 int close = 0;
2406 int fatal = 0;
2407 struct c4iw_rdev *rdev;
2408
2409 mutex_lock(&ep->com.mutex);
2410
2411 PDBG("%s ep %p state %s, abrupt %d\n", __func__, ep,
2412 states[ep->com.state], abrupt);
2413
2414 rdev = &ep->com.dev->rdev;
2415 if (c4iw_fatal_error(rdev)) {
2416 fatal = 1;
2417 close_complete_upcall(ep);
2418 ep->com.state = DEAD;
2419 }
2420 switch (ep->com.state) {
2421 case MPA_REQ_WAIT:
2422 case MPA_REQ_SENT:
2423 case MPA_REQ_RCVD:
2424 case MPA_REP_SENT:
2425 case FPDU_MODE:
2426 close = 1;
2427 if (abrupt)
2428 ep->com.state = ABORTING;
2429 else {
2430 ep->com.state = CLOSING;
2431 start_ep_timer(ep);
2432 }
2433 set_bit(CLOSE_SENT, &ep->com.flags);
2434 break;
2435 case CLOSING:
2436 if (!test_and_set_bit(CLOSE_SENT, &ep->com.flags)) {
2437 close = 1;
2438 if (abrupt) {
2439 stop_ep_timer(ep);
2440 ep->com.state = ABORTING;
2441 } else
2442 ep->com.state = MORIBUND;
2443 }
2444 break;
2445 case MORIBUND:
2446 case ABORTING:
2447 case DEAD:
2448 PDBG("%s ignoring disconnect ep %p state %u\n",
2449 __func__, ep, ep->com.state);
2450 break;
2451 default:
2452 BUG();
2453 break;
2454 }
2455
2456 if (close) {
2457 if (abrupt) {
2458 close_complete_upcall(ep);
2459 ret = send_abort(ep, NULL, gfp);
2460 } else
2461 ret = send_halfclose(ep, gfp);
2462 if (ret)
2463 fatal = 1;
2464 }
2465 mutex_unlock(&ep->com.mutex);
2466 if (fatal)
2467 release_ep_resources(ep);
2468 return ret;
2469 }
2470
2471 static int async_event(struct c4iw_dev *dev, struct sk_buff *skb)
2472 {
2473 struct cpl_fw6_msg *rpl = cplhdr(skb);
2474 c4iw_ev_dispatch(dev, (struct t4_cqe *)&rpl->data[0]);
2475 return 0;
2476 }
2477
2478 /*
2479 * These are the real handlers that are called from a
2480 * work queue.
2481 */
2482 static c4iw_handler_func work_handlers[NUM_CPL_CMDS] = {
2483 [CPL_ACT_ESTABLISH] = act_establish,
2484 [CPL_ACT_OPEN_RPL] = act_open_rpl,
2485 [CPL_RX_DATA] = rx_data,
2486 [CPL_ABORT_RPL_RSS] = abort_rpl,
2487 [CPL_ABORT_RPL] = abort_rpl,
2488 [CPL_PASS_OPEN_RPL] = pass_open_rpl,
2489 [CPL_CLOSE_LISTSRV_RPL] = close_listsrv_rpl,
2490 [CPL_PASS_ACCEPT_REQ] = pass_accept_req,
2491 [CPL_PASS_ESTABLISH] = pass_establish,
2492 [CPL_PEER_CLOSE] = peer_close,
2493 [CPL_ABORT_REQ_RSS] = peer_abort,
2494 [CPL_CLOSE_CON_RPL] = close_con_rpl,
2495 [CPL_RDMA_TERMINATE] = terminate,
2496 [CPL_FW4_ACK] = fw4_ack,
2497 [CPL_FW6_MSG] = async_event
2498 };
2499
2500 static void process_timeout(struct c4iw_ep *ep)
2501 {
2502 struct c4iw_qp_attributes attrs;
2503 int abort = 1;
2504
2505 mutex_lock(&ep->com.mutex);
2506 PDBG("%s ep %p tid %u state %d\n", __func__, ep, ep->hwtid,
2507 ep->com.state);
2508 switch (ep->com.state) {
2509 case MPA_REQ_SENT:
2510 __state_set(&ep->com, ABORTING);
2511 connect_reply_upcall(ep, -ETIMEDOUT);
2512 break;
2513 case MPA_REQ_WAIT:
2514 __state_set(&ep->com, ABORTING);
2515 break;
2516 case CLOSING:
2517 case MORIBUND:
2518 if (ep->com.cm_id && ep->com.qp) {
2519 attrs.next_state = C4IW_QP_STATE_ERROR;
2520 c4iw_modify_qp(ep->com.qp->rhp,
2521 ep->com.qp, C4IW_QP_ATTR_NEXT_STATE,
2522 &attrs, 1);
2523 }
2524 __state_set(&ep->com, ABORTING);
2525 break;
2526 default:
2527 printk(KERN_ERR "%s unexpected state ep %p tid %u state %u\n",
2528 __func__, ep, ep->hwtid, ep->com.state);
2529 WARN_ON(1);
2530 abort = 0;
2531 }
2532 mutex_unlock(&ep->com.mutex);
2533 if (abort)
2534 abort_connection(ep, NULL, GFP_KERNEL);
2535 c4iw_put_ep(&ep->com);
2536 }
2537
2538 static void process_timedout_eps(void)
2539 {
2540 struct c4iw_ep *ep;
2541
2542 spin_lock_irq(&timeout_lock);
2543 while (!list_empty(&timeout_list)) {
2544 struct list_head *tmp;
2545
2546 tmp = timeout_list.next;
2547 list_del(tmp);
2548 spin_unlock_irq(&timeout_lock);
2549 ep = list_entry(tmp, struct c4iw_ep, entry);
2550 process_timeout(ep);
2551 spin_lock_irq(&timeout_lock);
2552 }
2553 spin_unlock_irq(&timeout_lock);
2554 }
2555
2556 static void process_work(struct work_struct *work)
2557 {
2558 struct sk_buff *skb = NULL;
2559 struct c4iw_dev *dev;
2560 struct cpl_act_establish *rpl;
2561 unsigned int opcode;
2562 int ret;
2563
2564 while ((skb = skb_dequeue(&rxq))) {
2565 rpl = cplhdr(skb);
2566 dev = *((struct c4iw_dev **) (skb->cb + sizeof(void *)));
2567 opcode = rpl->ot.opcode;
2568
2569 BUG_ON(!work_handlers[opcode]);
2570 ret = work_handlers[opcode](dev, skb);
2571 if (!ret)
2572 kfree_skb(skb);
2573 }
2574 process_timedout_eps();
2575 }
2576
2577 static DECLARE_WORK(skb_work, process_work);
2578
2579 static void ep_timeout(unsigned long arg)
2580 {
2581 struct c4iw_ep *ep = (struct c4iw_ep *)arg;
2582
2583 spin_lock(&timeout_lock);
2584 list_add_tail(&ep->entry, &timeout_list);
2585 spin_unlock(&timeout_lock);
2586 queue_work(workq, &skb_work);
2587 }
2588
2589 /*
2590 * All the CM events are handled on a work queue to have a safe context.
2591 */
2592 static int sched(struct c4iw_dev *dev, struct sk_buff *skb)
2593 {
2594
2595 /*
2596 * Save dev in the skb->cb area.
2597 */
2598 *((struct c4iw_dev **) (skb->cb + sizeof(void *))) = dev;
2599
2600 /*
2601 * Queue the skb and schedule the worker thread.
2602 */
2603 skb_queue_tail(&rxq, skb);
2604 queue_work(workq, &skb_work);
2605 return 0;
2606 }
2607
2608 static int set_tcb_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
2609 {
2610 struct cpl_set_tcb_rpl *rpl = cplhdr(skb);
2611
2612 if (rpl->status != CPL_ERR_NONE) {
2613 printk(KERN_ERR MOD "Unexpected SET_TCB_RPL status %u "
2614 "for tid %u\n", rpl->status, GET_TID(rpl));
2615 }
2616 kfree_skb(skb);
2617 return 0;
2618 }
2619
2620 static int fw6_msg(struct c4iw_dev *dev, struct sk_buff *skb)
2621 {
2622 struct cpl_fw6_msg *rpl = cplhdr(skb);
2623 struct c4iw_wr_wait *wr_waitp;
2624 int ret;
2625
2626 PDBG("%s type %u\n", __func__, rpl->type);
2627
2628 switch (rpl->type) {
2629 case 1:
2630 ret = (int)((be64_to_cpu(rpl->data[0]) >> 8) & 0xff);
2631 wr_waitp = (struct c4iw_wr_wait *)(__force unsigned long) rpl->data[1];
2632 PDBG("%s wr_waitp %p ret %u\n", __func__, wr_waitp, ret);
2633 if (wr_waitp)
2634 c4iw_wake_up(wr_waitp, ret ? -ret : 0);
2635 kfree_skb(skb);
2636 break;
2637 case 2:
2638 sched(dev, skb);
2639 break;
2640 default:
2641 printk(KERN_ERR MOD "%s unexpected fw6 msg type %u\n", __func__,
2642 rpl->type);
2643 kfree_skb(skb);
2644 break;
2645 }
2646 return 0;
2647 }
2648
2649 static int peer_abort_intr(struct c4iw_dev *dev, struct sk_buff *skb)
2650 {
2651 struct cpl_abort_req_rss *req = cplhdr(skb);
2652 struct c4iw_ep *ep;
2653 struct tid_info *t = dev->rdev.lldi.tids;
2654 unsigned int tid = GET_TID(req);
2655
2656 ep = lookup_tid(t, tid);
2657 if (is_neg_adv_abort(req->status)) {
2658 PDBG("%s neg_adv_abort ep %p tid %u\n", __func__, ep,
2659 ep->hwtid);
2660 kfree_skb(skb);
2661 return 0;
2662 }
2663 PDBG("%s ep %p tid %u state %u\n", __func__, ep, ep->hwtid,
2664 ep->com.state);
2665
2666 /*
2667 * Wake up any threads in rdma_init() or rdma_fini().
2668 * However, this is not needed if com state is just
2669 * MPA_REQ_SENT
2670 */
2671 if (ep->com.state != MPA_REQ_SENT)
2672 c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
2673 sched(dev, skb);
2674 return 0;
2675 }
2676
2677 /*
2678 * Most upcalls from the T4 Core go to sched() to
2679 * schedule the processing on a work queue.
2680 */
2681 c4iw_handler_func c4iw_handlers[NUM_CPL_CMDS] = {
2682 [CPL_ACT_ESTABLISH] = sched,
2683 [CPL_ACT_OPEN_RPL] = sched,
2684 [CPL_RX_DATA] = sched,
2685 [CPL_ABORT_RPL_RSS] = sched,
2686 [CPL_ABORT_RPL] = sched,
2687 [CPL_PASS_OPEN_RPL] = sched,
2688 [CPL_CLOSE_LISTSRV_RPL] = sched,
2689 [CPL_PASS_ACCEPT_REQ] = sched,
2690 [CPL_PASS_ESTABLISH] = sched,
2691 [CPL_PEER_CLOSE] = sched,
2692 [CPL_CLOSE_CON_RPL] = sched,
2693 [CPL_ABORT_REQ_RSS] = peer_abort_intr,
2694 [CPL_RDMA_TERMINATE] = sched,
2695 [CPL_FW4_ACK] = sched,
2696 [CPL_SET_TCB_RPL] = set_tcb_rpl,
2697 [CPL_FW6_MSG] = fw6_msg
2698 };
2699
2700 int __init c4iw_cm_init(void)
2701 {
2702 spin_lock_init(&timeout_lock);
2703 skb_queue_head_init(&rxq);
2704
2705 workq = create_singlethread_workqueue("iw_cxgb4");
2706 if (!workq)
2707 return -ENOMEM;
2708
2709 return 0;
2710 }
2711
2712 void __exit c4iw_cm_term(void)
2713 {
2714 WARN_ON(!list_empty(&timeout_list));
2715 flush_workqueue(workq);
2716 destroy_workqueue(workq);
2717 }