]> git.proxmox.com Git - mirror_ubuntu-hirsute-kernel.git/blob - drivers/infiniband/hw/cxgb4/cm.c
projects / mirror_ubuntu-hirsute-kernel.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
regulator: palmas: Fix off-by-one for ramp_delay and register value mapping
[mirror_ubuntu-hirsute-kernel.git] / drivers / infiniband / hw / cxgb4 / cm.c
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 #include <linux/if_vlan.h>
42
43 #include <net/neighbour.h>
44 #include <net/netevent.h>
45 #include <net/route.h>
46 #include <net/tcp.h>
47
48 #include "iw_cxgb4.h"
49
50 static char *states[] = {
51 "idle",
52 "listen",
53 "connecting",
54 "mpa_wait_req",
55 "mpa_req_sent",
56 "mpa_req_rcvd",
57 "mpa_rep_sent",
58 "fpdu_mode",
59 "aborting",
60 "closing",
61 "moribund",
62 "dead",
63 NULL,
64 };
65
66 static int nocong;
67 module_param(nocong, int, 0644);
68 MODULE_PARM_DESC(nocong, "Turn of congestion control (default=0)");
69
70 static int enable_ecn;
71 module_param(enable_ecn, int, 0644);
72 MODULE_PARM_DESC(enable_ecn, "Enable ECN (default=0/disabled)");
73
74 static int dack_mode = 1;
75 module_param(dack_mode, int, 0644);
76 MODULE_PARM_DESC(dack_mode, "Delayed ack mode (default=1)");
77
78 int c4iw_max_read_depth = 8;
79 module_param(c4iw_max_read_depth, int, 0644);
80 MODULE_PARM_DESC(c4iw_max_read_depth, "Per-connection max ORD/IRD (default=8)");
81
82 static int enable_tcp_timestamps;
83 module_param(enable_tcp_timestamps, int, 0644);
84 MODULE_PARM_DESC(enable_tcp_timestamps, "Enable tcp timestamps (default=0)");
85
86 static int enable_tcp_sack;
87 module_param(enable_tcp_sack, int, 0644);
88 MODULE_PARM_DESC(enable_tcp_sack, "Enable tcp SACK (default=0)");
89
90 static int enable_tcp_window_scaling = 1;
91 module_param(enable_tcp_window_scaling, int, 0644);
92 MODULE_PARM_DESC(enable_tcp_window_scaling,
93 "Enable tcp window scaling (default=1)");
94
95 int c4iw_debug;
96 module_param(c4iw_debug, int, 0644);
97 MODULE_PARM_DESC(c4iw_debug, "Enable debug logging (default=0)");
98
99 static int peer2peer;
100 module_param(peer2peer, int, 0644);
101 MODULE_PARM_DESC(peer2peer, "Support peer2peer ULPs (default=0)");
102
103 static int p2p_type = FW_RI_INIT_P2PTYPE_READ_REQ;
104 module_param(p2p_type, int, 0644);
105 MODULE_PARM_DESC(p2p_type, "RDMAP opcode to use for the RTR message: "
106 "1=RDMA_READ 0=RDMA_WRITE (default 1)");
107
108 static int ep_timeout_secs = 60;
109 module_param(ep_timeout_secs, int, 0644);
110 MODULE_PARM_DESC(ep_timeout_secs, "CM Endpoint operation timeout "
111 "in seconds (default=60)");
112
113 static int mpa_rev = 1;
114 module_param(mpa_rev, int, 0644);
115 MODULE_PARM_DESC(mpa_rev, "MPA Revision, 0 supports amso1100, "
116 "1 is RFC0544 spec compliant, 2 is IETF MPA Peer Connect Draft"
117 " compliant (default=1)");
118
119 static int markers_enabled;
120 module_param(markers_enabled, int, 0644);
121 MODULE_PARM_DESC(markers_enabled, "Enable MPA MARKERS (default(0)=disabled)");
122
123 static int crc_enabled = 1;
124 module_param(crc_enabled, int, 0644);
125 MODULE_PARM_DESC(crc_enabled, "Enable MPA CRC (default(1)=enabled)");
126
127 static int rcv_win = 256 * 1024;
128 module_param(rcv_win, int, 0644);
129 MODULE_PARM_DESC(rcv_win, "TCP receive window in bytes (default=256KB)");
130
131 static int snd_win = 128 * 1024;
132 module_param(snd_win, int, 0644);
133 MODULE_PARM_DESC(snd_win, "TCP send window in bytes (default=128KB)");
134
135 static struct workqueue_struct *workq;
136
137 static struct sk_buff_head rxq;
138
139 static struct sk_buff *get_skb(struct sk_buff *skb, int len, gfp_t gfp);
140 static void ep_timeout(unsigned long arg);
141 static void connect_reply_upcall(struct c4iw_ep *ep, int status);
142
143 static LIST_HEAD(timeout_list);
144 static spinlock_t timeout_lock;
145
146 static void deref_qp(struct c4iw_ep *ep)
147 {
148 c4iw_qp_rem_ref(&ep->com.qp->ibqp);
149 clear_bit(QP_REFERENCED, &ep->com.flags);
150 }
151
152 static void ref_qp(struct c4iw_ep *ep)
153 {
154 set_bit(QP_REFERENCED, &ep->com.flags);
155 c4iw_qp_add_ref(&ep->com.qp->ibqp);
156 }
157
158 static void start_ep_timer(struct c4iw_ep *ep)
159 {
160 PDBG("%s ep %p\n", __func__, ep);
161 if (timer_pending(&ep->timer)) {
162 pr_err("%s timer already started! ep %p\n",
163 __func__, ep);
164 return;
165 }
166 clear_bit(TIMEOUT, &ep->com.flags);
167 c4iw_get_ep(&ep->com);
168 ep->timer.expires = jiffies + ep_timeout_secs * HZ;
169 ep->timer.data = (unsigned long)ep;
170 ep->timer.function = ep_timeout;
171 add_timer(&ep->timer);
172 }
173
174 static void stop_ep_timer(struct c4iw_ep *ep)
175 {
176 PDBG("%s ep %p stopping\n", __func__, ep);
177 del_timer_sync(&ep->timer);
178 if (!test_and_set_bit(TIMEOUT, &ep->com.flags))
179 c4iw_put_ep(&ep->com);
180 }
181
182 static int c4iw_l2t_send(struct c4iw_rdev *rdev, struct sk_buff *skb,
183 struct l2t_entry *l2e)
184 {
185 int error = 0;
186
187 if (c4iw_fatal_error(rdev)) {
188 kfree_skb(skb);
189 PDBG("%s - device in error state - dropping\n", __func__);
190 return -EIO;
191 }
192 error = cxgb4_l2t_send(rdev->lldi.ports[0], skb, l2e);
193 if (error < 0)
194 kfree_skb(skb);
195 return error < 0 ? error : 0;
196 }
197
198 int c4iw_ofld_send(struct c4iw_rdev *rdev, struct sk_buff *skb)
199 {
200 int error = 0;
201
202 if (c4iw_fatal_error(rdev)) {
203 kfree_skb(skb);
204 PDBG("%s - device in error state - dropping\n", __func__);
205 return -EIO;
206 }
207 error = cxgb4_ofld_send(rdev->lldi.ports[0], skb);
208 if (error < 0)
209 kfree_skb(skb);
210 return error < 0 ? error : 0;
211 }
212
213 static void release_tid(struct c4iw_rdev *rdev, u32 hwtid, struct sk_buff *skb)
214 {
215 struct cpl_tid_release *req;
216
217 skb = get_skb(skb, sizeof *req, GFP_KERNEL);
218 if (!skb)
219 return;
220 req = (struct cpl_tid_release *) skb_put(skb, sizeof(*req));
221 INIT_TP_WR(req, hwtid);
222 OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_TID_RELEASE, hwtid));
223 set_wr_txq(skb, CPL_PRIORITY_SETUP, 0);
224 c4iw_ofld_send(rdev, skb);
225 return;
226 }
227
228 static void set_emss(struct c4iw_ep *ep, u16 opt)
229 {
230 ep->emss = ep->com.dev->rdev.lldi.mtus[GET_TCPOPT_MSS(opt)] - 40;
231 ep->mss = ep->emss;
232 if (GET_TCPOPT_TSTAMP(opt))
233 ep->emss -= 12;
234 if (ep->emss < 128)
235 ep->emss = 128;
236 PDBG("%s mss_idx %u mss %u emss=%u\n", __func__, GET_TCPOPT_MSS(opt),
237 ep->mss, ep->emss);
238 }
239
240 static enum c4iw_ep_state state_read(struct c4iw_ep_common *epc)
241 {
242 enum c4iw_ep_state state;
243
244 mutex_lock(&epc->mutex);
245 state = epc->state;
246 mutex_unlock(&epc->mutex);
247 return state;
248 }
249
250 static void __state_set(struct c4iw_ep_common *epc, enum c4iw_ep_state new)
251 {
252 epc->state = new;
253 }
254
255 static void state_set(struct c4iw_ep_common *epc, enum c4iw_ep_state new)
256 {
257 mutex_lock(&epc->mutex);
258 PDBG("%s - %s -> %s\n", __func__, states[epc->state], states[new]);
259 __state_set(epc, new);
260 mutex_unlock(&epc->mutex);
261 return;
262 }
263
264 static void *alloc_ep(int size, gfp_t gfp)
265 {
266 struct c4iw_ep_common *epc;
267
268 epc = kzalloc(size, gfp);
269 if (epc) {
270 kref_init(&epc->kref);
271 mutex_init(&epc->mutex);
272 c4iw_init_wr_wait(&epc->wr_wait);
273 }
274 PDBG("%s alloc ep %p\n", __func__, epc);
275 return epc;
276 }
277
278 void _c4iw_free_ep(struct kref *kref)
279 {
280 struct c4iw_ep *ep;
281
282 ep = container_of(kref, struct c4iw_ep, com.kref);
283 PDBG("%s ep %p state %s\n", __func__, ep, states[state_read(&ep->com)]);
284 if (test_bit(QP_REFERENCED, &ep->com.flags))
285 deref_qp(ep);
286 if (test_bit(RELEASE_RESOURCES, &ep->com.flags)) {
287 remove_handle(ep->com.dev, &ep->com.dev->hwtid_idr, ep->hwtid);
288 cxgb4_remove_tid(ep->com.dev->rdev.lldi.tids, 0, ep->hwtid);
289 dst_release(ep->dst);
290 cxgb4_l2t_release(ep->l2t);
291 }
292 kfree(ep);
293 }
294
295 static void release_ep_resources(struct c4iw_ep *ep)
296 {
297 set_bit(RELEASE_RESOURCES, &ep->com.flags);
298 c4iw_put_ep(&ep->com);
299 }
300
301 static int status2errno(int status)
302 {
303 switch (status) {
304 case CPL_ERR_NONE:
305 return 0;
306 case CPL_ERR_CONN_RESET:
307 return -ECONNRESET;
308 case CPL_ERR_ARP_MISS:
309 return -EHOSTUNREACH;
310 case CPL_ERR_CONN_TIMEDOUT:
311 return -ETIMEDOUT;
312 case CPL_ERR_TCAM_FULL:
313 return -ENOMEM;
314 case CPL_ERR_CONN_EXIST:
315 return -EADDRINUSE;
316 default:
317 return -EIO;
318 }
319 }
320
321 /*
322 * Try and reuse skbs already allocated...
323 */
324 static struct sk_buff *get_skb(struct sk_buff *skb, int len, gfp_t gfp)
325 {
326 if (skb && !skb_is_nonlinear(skb) && !skb_cloned(skb)) {
327 skb_trim(skb, 0);
328 skb_get(skb);
329 skb_reset_transport_header(skb);
330 } else {
331 skb = alloc_skb(len, gfp);
332 }
333 return skb;
334 }
335
336 static struct rtable *find_route(struct c4iw_dev *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
351 static void arp_failure_discard(void *handle, struct sk_buff *skb)
352 {
353 PDBG("%s c4iw_dev %p\n", __func__, handle);
354 kfree_skb(skb);
355 }
356
357 /*
358 * Handle an ARP failure for an active open.
359 */
360 static void act_open_req_arp_failure(void *handle, struct sk_buff *skb)
361 {
362 printk(KERN_ERR MOD "ARP failure duing connect\n");
363 kfree_skb(skb);
364 }
365
366 /*
367 * Handle an ARP failure for a CPL_ABORT_REQ. Change it into a no RST variant
368 * and send it along.
369 */
370 static void abort_arp_failure(void *handle, struct sk_buff *skb)
371 {
372 struct c4iw_rdev *rdev = handle;
373 struct cpl_abort_req *req = cplhdr(skb);
374
375 PDBG("%s rdev %p\n", __func__, rdev);
376 req->cmd = CPL_ABORT_NO_RST;
377 c4iw_ofld_send(rdev, skb);
378 }
379
380 static void send_flowc(struct c4iw_ep *ep, struct sk_buff *skb)
381 {
382 unsigned int flowclen = 80;
383 struct fw_flowc_wr *flowc;
384 int i;
385
386 skb = get_skb(skb, flowclen, GFP_KERNEL);
387 flowc = (struct fw_flowc_wr *)__skb_put(skb, flowclen);
388
389 flowc->op_to_nparams = cpu_to_be32(FW_WR_OP(FW_FLOWC_WR) |
390 FW_FLOWC_WR_NPARAMS(8));
391 flowc->flowid_len16 = cpu_to_be32(FW_WR_LEN16(DIV_ROUND_UP(flowclen,
392 16)) | FW_WR_FLOWID(ep->hwtid));
393
394 flowc->mnemval[0].mnemonic = FW_FLOWC_MNEM_PFNVFN;
395 flowc->mnemval[0].val = cpu_to_be32(PCI_FUNC(ep->com.dev->rdev.lldi.pdev->devfn) << 8);
396 flowc->mnemval[1].mnemonic = FW_FLOWC_MNEM_CH;
397 flowc->mnemval[1].val = cpu_to_be32(ep->tx_chan);
398 flowc->mnemval[2].mnemonic = FW_FLOWC_MNEM_PORT;
399 flowc->mnemval[2].val = cpu_to_be32(ep->tx_chan);
400 flowc->mnemval[3].mnemonic = FW_FLOWC_MNEM_IQID;
401 flowc->mnemval[3].val = cpu_to_be32(ep->rss_qid);
402 flowc->mnemval[4].mnemonic = FW_FLOWC_MNEM_SNDNXT;
403 flowc->mnemval[4].val = cpu_to_be32(ep->snd_seq);
404 flowc->mnemval[5].mnemonic = FW_FLOWC_MNEM_RCVNXT;
405 flowc->mnemval[5].val = cpu_to_be32(ep->rcv_seq);
406 flowc->mnemval[6].mnemonic = FW_FLOWC_MNEM_SNDBUF;
407 flowc->mnemval[6].val = cpu_to_be32(snd_win);
408 flowc->mnemval[7].mnemonic = FW_FLOWC_MNEM_MSS;
409 flowc->mnemval[7].val = cpu_to_be32(ep->emss);
410 /* Pad WR to 16 byte boundary */
411 flowc->mnemval[8].mnemonic = 0;
412 flowc->mnemval[8].val = 0;
413 for (i = 0; i < 9; i++) {
414 flowc->mnemval[i].r4[0] = 0;
415 flowc->mnemval[i].r4[1] = 0;
416 flowc->mnemval[i].r4[2] = 0;
417 }
418
419 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
420 c4iw_ofld_send(&ep->com.dev->rdev, skb);
421 }
422
423 static int send_halfclose(struct c4iw_ep *ep, gfp_t gfp)
424 {
425 struct cpl_close_con_req *req;
426 struct sk_buff *skb;
427 int wrlen = roundup(sizeof *req, 16);
428
429 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
430 skb = get_skb(NULL, wrlen, gfp);
431 if (!skb) {
432 printk(KERN_ERR MOD "%s - failed to alloc skb\n", __func__);
433 return -ENOMEM;
434 }
435 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
436 t4_set_arp_err_handler(skb, NULL, arp_failure_discard);
437 req = (struct cpl_close_con_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_CLOSE_CON_REQ,
441 ep->hwtid));
442 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
443 }
444
445 static int send_abort(struct c4iw_ep *ep, struct sk_buff *skb, gfp_t gfp)
446 {
447 struct cpl_abort_req *req;
448 int wrlen = roundup(sizeof *req, 16);
449
450 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
451 skb = get_skb(skb, wrlen, gfp);
452 if (!skb) {
453 printk(KERN_ERR MOD "%s - failed to alloc skb.\n",
454 __func__);
455 return -ENOMEM;
456 }
457 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
458 t4_set_arp_err_handler(skb, &ep->com.dev->rdev, abort_arp_failure);
459 req = (struct cpl_abort_req *) skb_put(skb, wrlen);
460 memset(req, 0, wrlen);
461 INIT_TP_WR(req, ep->hwtid);
462 OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_ABORT_REQ, ep->hwtid));
463 req->cmd = CPL_ABORT_SEND_RST;
464 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
465 }
466
467 #define VLAN_NONE 0xfff
468 #define FILTER_SEL_VLAN_NONE 0xffff
469 #define FILTER_SEL_WIDTH_P_FC (3+1) /* port uses 3 bits, FCoE one bit */
470 #define FILTER_SEL_WIDTH_VIN_P_FC \
471 (6 + 7 + FILTER_SEL_WIDTH_P_FC) /* 6 bits are unused, VF uses 7 bits*/
472 #define FILTER_SEL_WIDTH_TAG_P_FC \
473 (3 + FILTER_SEL_WIDTH_VIN_P_FC) /* PF uses 3 bits */
474 #define FILTER_SEL_WIDTH_VLD_TAG_P_FC (1 + FILTER_SEL_WIDTH_TAG_P_FC)
475
476 static unsigned int select_ntuple(struct c4iw_dev *dev, struct dst_entry *dst,
477 struct l2t_entry *l2t)
478 {
479 unsigned int ntuple = 0;
480 u32 viid;
481
482 switch (dev->rdev.lldi.filt_mode) {
483
484 /* default filter mode */
485 case HW_TPL_FR_MT_PR_IV_P_FC:
486 if (l2t->vlan == VLAN_NONE)
487 ntuple |= FILTER_SEL_VLAN_NONE << FILTER_SEL_WIDTH_P_FC;
488 else {
489 ntuple |= l2t->vlan << FILTER_SEL_WIDTH_P_FC;
490 ntuple |= 1 << FILTER_SEL_WIDTH_VLD_TAG_P_FC;
491 }
492 ntuple |= l2t->lport << S_PORT | IPPROTO_TCP <<
493 FILTER_SEL_WIDTH_VLD_TAG_P_FC;
494 break;
495 case HW_TPL_FR_MT_PR_OV_P_FC: {
496 viid = cxgb4_port_viid(l2t->neigh->dev);
497
498 ntuple |= FW_VIID_VIN_GET(viid) << FILTER_SEL_WIDTH_P_FC;
499 ntuple |= FW_VIID_PFN_GET(viid) << FILTER_SEL_WIDTH_VIN_P_FC;
500 ntuple |= FW_VIID_VIVLD_GET(viid) << FILTER_SEL_WIDTH_TAG_P_FC;
501 ntuple |= l2t->lport << S_PORT | IPPROTO_TCP <<
502 FILTER_SEL_WIDTH_VLD_TAG_P_FC;
503 break;
504 }
505 default:
506 break;
507 }
508 return ntuple;
509 }
510
511 static int send_connect(struct c4iw_ep *ep)
512 {
513 struct cpl_act_open_req *req;
514 struct sk_buff *skb;
515 u64 opt0;
516 u32 opt2;
517 unsigned int mtu_idx;
518 int wscale;
519 int wrlen = roundup(sizeof *req, 16);
520
521 PDBG("%s ep %p atid %u\n", __func__, ep, ep->atid);
522
523 skb = get_skb(NULL, wrlen, GFP_KERNEL);
524 if (!skb) {
525 printk(KERN_ERR MOD "%s - failed to alloc skb.\n",
526 __func__);
527 return -ENOMEM;
528 }
529 set_wr_txq(skb, CPL_PRIORITY_SETUP, ep->ctrlq_idx);
530
531 cxgb4_best_mtu(ep->com.dev->rdev.lldi.mtus, ep->mtu, &mtu_idx);
532 wscale = compute_wscale(rcv_win);
533 opt0 = (nocong ? NO_CONG(1) : 0) |
534 KEEP_ALIVE(1) |
535 DELACK(1) |
536 WND_SCALE(wscale) |
537 MSS_IDX(mtu_idx) |
538 L2T_IDX(ep->l2t->idx) |
539 TX_CHAN(ep->tx_chan) |
540 SMAC_SEL(ep->smac_idx) |
541 DSCP(ep->tos) |
542 ULP_MODE(ULP_MODE_TCPDDP) |
543 RCV_BUFSIZ(rcv_win>>10);
544 opt2 = RX_CHANNEL(0) |
545 CCTRL_ECN(enable_ecn) |
546 RSS_QUEUE_VALID | RSS_QUEUE(ep->rss_qid);
547 if (enable_tcp_timestamps)
548 opt2 |= TSTAMPS_EN(1);
549 if (enable_tcp_sack)
550 opt2 |= SACK_EN(1);
551 if (wscale && enable_tcp_window_scaling)
552 opt2 |= WND_SCALE_EN(1);
553 t4_set_arp_err_handler(skb, NULL, act_open_req_arp_failure);
554
555 req = (struct cpl_act_open_req *) skb_put(skb, wrlen);
556 INIT_TP_WR(req, 0);
557 OPCODE_TID(req) = cpu_to_be32(
558 MK_OPCODE_TID(CPL_ACT_OPEN_REQ, ((ep->rss_qid<<14)|ep->atid)));
559 req->local_port = ep->com.local_addr.sin_port;
560 req->peer_port = ep->com.remote_addr.sin_port;
561 req->local_ip = ep->com.local_addr.sin_addr.s_addr;
562 req->peer_ip = ep->com.remote_addr.sin_addr.s_addr;
563 req->opt0 = cpu_to_be64(opt0);
564 req->params = cpu_to_be32(select_ntuple(ep->com.dev, ep->dst, ep->l2t));
565 req->opt2 = cpu_to_be32(opt2);
566 set_bit(ACT_OPEN_REQ, &ep->com.history);
567 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
568 }
569
570 static void send_mpa_req(struct c4iw_ep *ep, struct sk_buff *skb,
571 u8 mpa_rev_to_use)
572 {
573 int mpalen, wrlen;
574 struct fw_ofld_tx_data_wr *req;
575 struct mpa_message *mpa;
576 struct mpa_v2_conn_params mpa_v2_params;
577
578 PDBG("%s ep %p tid %u pd_len %d\n", __func__, ep, ep->hwtid, ep->plen);
579
580 BUG_ON(skb_cloned(skb));
581
582 mpalen = sizeof(*mpa) + ep->plen;
583 if (mpa_rev_to_use == 2)
584 mpalen += sizeof(struct mpa_v2_conn_params);
585 wrlen = roundup(mpalen + sizeof *req, 16);
586 skb = get_skb(skb, wrlen, GFP_KERNEL);
587 if (!skb) {
588 connect_reply_upcall(ep, -ENOMEM);
589 return;
590 }
591 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
592
593 req = (struct fw_ofld_tx_data_wr *)skb_put(skb, wrlen);
594 memset(req, 0, wrlen);
595 req->op_to_immdlen = cpu_to_be32(
596 FW_WR_OP(FW_OFLD_TX_DATA_WR) |
597 FW_WR_COMPL(1) |
598 FW_WR_IMMDLEN(mpalen));
599 req->flowid_len16 = cpu_to_be32(
600 FW_WR_FLOWID(ep->hwtid) |
601 FW_WR_LEN16(wrlen >> 4));
602 req->plen = cpu_to_be32(mpalen);
603 req->tunnel_to_proxy = cpu_to_be32(
604 FW_OFLD_TX_DATA_WR_FLUSH(1) |
605 FW_OFLD_TX_DATA_WR_SHOVE(1));
606
607 mpa = (struct mpa_message *)(req + 1);
608 memcpy(mpa->key, MPA_KEY_REQ, sizeof(mpa->key));
609 mpa->flags = (crc_enabled ? MPA_CRC : 0) |
610 (markers_enabled ? MPA_MARKERS : 0) |
611 (mpa_rev_to_use == 2 ? MPA_ENHANCED_RDMA_CONN : 0);
612 mpa->private_data_size = htons(ep->plen);
613 mpa->revision = mpa_rev_to_use;
614 if (mpa_rev_to_use == 1) {
615 ep->tried_with_mpa_v1 = 1;
616 ep->retry_with_mpa_v1 = 0;
617 }
618
619 if (mpa_rev_to_use == 2) {
620 mpa->private_data_size = htons(ntohs(mpa->private_data_size) +
621 sizeof (struct mpa_v2_conn_params));
622 mpa_v2_params.ird = htons((u16)ep->ird);
623 mpa_v2_params.ord = htons((u16)ep->ord);
624
625 if (peer2peer) {
626 mpa_v2_params.ird |= htons(MPA_V2_PEER2PEER_MODEL);
627 if (p2p_type == FW_RI_INIT_P2PTYPE_RDMA_WRITE)
628 mpa_v2_params.ord |=
629 htons(MPA_V2_RDMA_WRITE_RTR);
630 else if (p2p_type == FW_RI_INIT_P2PTYPE_READ_REQ)
631 mpa_v2_params.ord |=
632 htons(MPA_V2_RDMA_READ_RTR);
633 }
634 memcpy(mpa->private_data, &mpa_v2_params,
635 sizeof(struct mpa_v2_conn_params));
636
637 if (ep->plen)
638 memcpy(mpa->private_data +
639 sizeof(struct mpa_v2_conn_params),
640 ep->mpa_pkt + sizeof(*mpa), ep->plen);
641 } else
642 if (ep->plen)
643 memcpy(mpa->private_data,
644 ep->mpa_pkt + sizeof(*mpa), ep->plen);
645
646 /*
647 * Reference the mpa skb. This ensures the data area
648 * will remain in memory until the hw acks the tx.
649 * Function fw4_ack() will deref it.
650 */
651 skb_get(skb);
652 t4_set_arp_err_handler(skb, NULL, arp_failure_discard);
653 BUG_ON(ep->mpa_skb);
654 ep->mpa_skb = skb;
655 c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
656 start_ep_timer(ep);
657 state_set(&ep->com, MPA_REQ_SENT);
658 ep->mpa_attr.initiator = 1;
659 return;
660 }
661
662 static int send_mpa_reject(struct c4iw_ep *ep, const void *pdata, u8 plen)
663 {
664 int mpalen, wrlen;
665 struct fw_ofld_tx_data_wr *req;
666 struct mpa_message *mpa;
667 struct sk_buff *skb;
668 struct mpa_v2_conn_params mpa_v2_params;
669
670 PDBG("%s ep %p tid %u pd_len %d\n", __func__, ep, ep->hwtid, ep->plen);
671
672 mpalen = sizeof(*mpa) + plen;
673 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn)
674 mpalen += sizeof(struct mpa_v2_conn_params);
675 wrlen = roundup(mpalen + sizeof *req, 16);
676
677 skb = get_skb(NULL, wrlen, GFP_KERNEL);
678 if (!skb) {
679 printk(KERN_ERR MOD "%s - cannot alloc skb!\n", __func__);
680 return -ENOMEM;
681 }
682 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
683
684 req = (struct fw_ofld_tx_data_wr *)skb_put(skb, wrlen);
685 memset(req, 0, wrlen);
686 req->op_to_immdlen = cpu_to_be32(
687 FW_WR_OP(FW_OFLD_TX_DATA_WR) |
688 FW_WR_COMPL(1) |
689 FW_WR_IMMDLEN(mpalen));
690 req->flowid_len16 = cpu_to_be32(
691 FW_WR_FLOWID(ep->hwtid) |
692 FW_WR_LEN16(wrlen >> 4));
693 req->plen = cpu_to_be32(mpalen);
694 req->tunnel_to_proxy = cpu_to_be32(
695 FW_OFLD_TX_DATA_WR_FLUSH(1) |
696 FW_OFLD_TX_DATA_WR_SHOVE(1));
697
698 mpa = (struct mpa_message *)(req + 1);
699 memset(mpa, 0, sizeof(*mpa));
700 memcpy(mpa->key, MPA_KEY_REP, sizeof(mpa->key));
701 mpa->flags = MPA_REJECT;
702 mpa->revision = ep->mpa_attr.version;
703 mpa->private_data_size = htons(plen);
704
705 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) {
706 mpa->flags |= MPA_ENHANCED_RDMA_CONN;
707 mpa->private_data_size = htons(ntohs(mpa->private_data_size) +
708 sizeof (struct mpa_v2_conn_params));
709 mpa_v2_params.ird = htons(((u16)ep->ird) |
710 (peer2peer ? MPA_V2_PEER2PEER_MODEL :
711 0));
712 mpa_v2_params.ord = htons(((u16)ep->ord) | (peer2peer ?
713 (p2p_type ==
714 FW_RI_INIT_P2PTYPE_RDMA_WRITE ?
715 MPA_V2_RDMA_WRITE_RTR : p2p_type ==
716 FW_RI_INIT_P2PTYPE_READ_REQ ?
717 MPA_V2_RDMA_READ_RTR : 0) : 0));
718 memcpy(mpa->private_data, &mpa_v2_params,
719 sizeof(struct mpa_v2_conn_params));
720
721 if (ep->plen)
722 memcpy(mpa->private_data +
723 sizeof(struct mpa_v2_conn_params), pdata, plen);
724 } else
725 if (plen)
726 memcpy(mpa->private_data, pdata, plen);
727
728 /*
729 * Reference the mpa skb again. This ensures the data area
730 * will remain in memory until the hw acks the tx.
731 * Function fw4_ack() will deref it.
732 */
733 skb_get(skb);
734 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
735 t4_set_arp_err_handler(skb, NULL, arp_failure_discard);
736 BUG_ON(ep->mpa_skb);
737 ep->mpa_skb = skb;
738 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
739 }
740
741 static int send_mpa_reply(struct c4iw_ep *ep, const void *pdata, u8 plen)
742 {
743 int mpalen, wrlen;
744 struct fw_ofld_tx_data_wr *req;
745 struct mpa_message *mpa;
746 struct sk_buff *skb;
747 struct mpa_v2_conn_params mpa_v2_params;
748
749 PDBG("%s ep %p tid %u pd_len %d\n", __func__, ep, ep->hwtid, ep->plen);
750
751 mpalen = sizeof(*mpa) + plen;
752 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn)
753 mpalen += sizeof(struct mpa_v2_conn_params);
754 wrlen = roundup(mpalen + sizeof *req, 16);
755
756 skb = get_skb(NULL, wrlen, GFP_KERNEL);
757 if (!skb) {
758 printk(KERN_ERR MOD "%s - cannot alloc skb!\n", __func__);
759 return -ENOMEM;
760 }
761 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
762
763 req = (struct fw_ofld_tx_data_wr *) skb_put(skb, wrlen);
764 memset(req, 0, wrlen);
765 req->op_to_immdlen = cpu_to_be32(
766 FW_WR_OP(FW_OFLD_TX_DATA_WR) |
767 FW_WR_COMPL(1) |
768 FW_WR_IMMDLEN(mpalen));
769 req->flowid_len16 = cpu_to_be32(
770 FW_WR_FLOWID(ep->hwtid) |
771 FW_WR_LEN16(wrlen >> 4));
772 req->plen = cpu_to_be32(mpalen);
773 req->tunnel_to_proxy = cpu_to_be32(
774 FW_OFLD_TX_DATA_WR_FLUSH(1) |
775 FW_OFLD_TX_DATA_WR_SHOVE(1));
776
777 mpa = (struct mpa_message *)(req + 1);
778 memset(mpa, 0, sizeof(*mpa));
779 memcpy(mpa->key, MPA_KEY_REP, sizeof(mpa->key));
780 mpa->flags = (ep->mpa_attr.crc_enabled ? MPA_CRC : 0) |
781 (markers_enabled ? MPA_MARKERS : 0);
782 mpa->revision = ep->mpa_attr.version;
783 mpa->private_data_size = htons(plen);
784
785 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) {
786 mpa->flags |= MPA_ENHANCED_RDMA_CONN;
787 mpa->private_data_size = htons(ntohs(mpa->private_data_size) +
788 sizeof (struct mpa_v2_conn_params));
789 mpa_v2_params.ird = htons((u16)ep->ird);
790 mpa_v2_params.ord = htons((u16)ep->ord);
791 if (peer2peer && (ep->mpa_attr.p2p_type !=
792 FW_RI_INIT_P2PTYPE_DISABLED)) {
793 mpa_v2_params.ird |= htons(MPA_V2_PEER2PEER_MODEL);
794
795 if (p2p_type == FW_RI_INIT_P2PTYPE_RDMA_WRITE)
796 mpa_v2_params.ord |=
797 htons(MPA_V2_RDMA_WRITE_RTR);
798 else if (p2p_type == FW_RI_INIT_P2PTYPE_READ_REQ)
799 mpa_v2_params.ord |=
800 htons(MPA_V2_RDMA_READ_RTR);
801 }
802
803 memcpy(mpa->private_data, &mpa_v2_params,
804 sizeof(struct mpa_v2_conn_params));
805
806 if (ep->plen)
807 memcpy(mpa->private_data +
808 sizeof(struct mpa_v2_conn_params), pdata, plen);
809 } else
810 if (plen)
811 memcpy(mpa->private_data, pdata, plen);
812
813 /*
814 * Reference the mpa skb. This ensures the data area
815 * will remain in memory until the hw acks the tx.
816 * Function fw4_ack() will deref it.
817 */
818 skb_get(skb);
819 t4_set_arp_err_handler(skb, NULL, arp_failure_discard);
820 ep->mpa_skb = skb;
821 state_set(&ep->com, MPA_REP_SENT);
822 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
823 }
824
825 static int act_establish(struct c4iw_dev *dev, struct sk_buff *skb)
826 {
827 struct c4iw_ep *ep;
828 struct cpl_act_establish *req = cplhdr(skb);
829 unsigned int tid = GET_TID(req);
830 unsigned int atid = GET_TID_TID(ntohl(req->tos_atid));
831 struct tid_info *t = dev->rdev.lldi.tids;
832
833 ep = lookup_atid(t, atid);
834
835 PDBG("%s ep %p tid %u snd_isn %u rcv_isn %u\n", __func__, ep, tid,
836 be32_to_cpu(req->snd_isn), be32_to_cpu(req->rcv_isn));
837
838 dst_confirm(ep->dst);
839
840 /* setup the hwtid for this connection */
841 ep->hwtid = tid;
842 cxgb4_insert_tid(t, ep, tid);
843 insert_handle(dev, &dev->hwtid_idr, ep, ep->hwtid);
844
845 ep->snd_seq = be32_to_cpu(req->snd_isn);
846 ep->rcv_seq = be32_to_cpu(req->rcv_isn);
847
848 set_emss(ep, ntohs(req->tcp_opt));
849
850 /* dealloc the atid */
851 remove_handle(ep->com.dev, &ep->com.dev->atid_idr, atid);
852 cxgb4_free_atid(t, atid);
853 set_bit(ACT_ESTAB, &ep->com.history);
854
855 /* start MPA negotiation */
856 send_flowc(ep, NULL);
857 if (ep->retry_with_mpa_v1)
858 send_mpa_req(ep, skb, 1);
859 else
860 send_mpa_req(ep, skb, mpa_rev);
861
862 return 0;
863 }
864
865 static void close_complete_upcall(struct c4iw_ep *ep)
866 {
867 struct iw_cm_event event;
868
869 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
870 memset(&event, 0, sizeof(event));
871 event.event = IW_CM_EVENT_CLOSE;
872 if (ep->com.cm_id) {
873 PDBG("close complete delivered ep %p cm_id %p tid %u\n",
874 ep, ep->com.cm_id, ep->hwtid);
875 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
876 ep->com.cm_id->rem_ref(ep->com.cm_id);
877 ep->com.cm_id = NULL;
878 set_bit(CLOSE_UPCALL, &ep->com.history);
879 }
880 }
881
882 static int abort_connection(struct c4iw_ep *ep, struct sk_buff *skb, gfp_t gfp)
883 {
884 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
885 close_complete_upcall(ep);
886 state_set(&ep->com, ABORTING);
887 set_bit(ABORT_CONN, &ep->com.history);
888 return send_abort(ep, skb, gfp);
889 }
890
891 static void peer_close_upcall(struct c4iw_ep *ep)
892 {
893 struct iw_cm_event event;
894
895 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
896 memset(&event, 0, sizeof(event));
897 event.event = IW_CM_EVENT_DISCONNECT;
898 if (ep->com.cm_id) {
899 PDBG("peer close delivered ep %p cm_id %p tid %u\n",
900 ep, ep->com.cm_id, ep->hwtid);
901 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
902 set_bit(DISCONN_UPCALL, &ep->com.history);
903 }
904 }
905
906 static void peer_abort_upcall(struct c4iw_ep *ep)
907 {
908 struct iw_cm_event event;
909
910 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
911 memset(&event, 0, sizeof(event));
912 event.event = IW_CM_EVENT_CLOSE;
913 event.status = -ECONNRESET;
914 if (ep->com.cm_id) {
915 PDBG("abort delivered ep %p cm_id %p tid %u\n", ep,
916 ep->com.cm_id, ep->hwtid);
917 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
918 ep->com.cm_id->rem_ref(ep->com.cm_id);
919 ep->com.cm_id = NULL;
920 set_bit(ABORT_UPCALL, &ep->com.history);
921 }
922 }
923
924 static void connect_reply_upcall(struct c4iw_ep *ep, int status)
925 {
926 struct iw_cm_event event;
927
928 PDBG("%s ep %p tid %u status %d\n", __func__, ep, ep->hwtid, status);
929 memset(&event, 0, sizeof(event));
930 event.event = IW_CM_EVENT_CONNECT_REPLY;
931 event.status = status;
932 event.local_addr = ep->com.local_addr;
933 event.remote_addr = ep->com.remote_addr;
934
935 if ((status == 0) || (status == -ECONNREFUSED)) {
936 if (!ep->tried_with_mpa_v1) {
937 /* this means MPA_v2 is used */
938 event.private_data_len = ep->plen -
939 sizeof(struct mpa_v2_conn_params);
940 event.private_data = ep->mpa_pkt +
941 sizeof(struct mpa_message) +
942 sizeof(struct mpa_v2_conn_params);
943 } else {
944 /* this means MPA_v1 is used */
945 event.private_data_len = ep->plen;
946 event.private_data = ep->mpa_pkt +
947 sizeof(struct mpa_message);
948 }
949 }
950
951 PDBG("%s ep %p tid %u status %d\n", __func__, ep,
952 ep->hwtid, status);
953 set_bit(CONN_RPL_UPCALL, &ep->com.history);
954 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
955
956 if (status < 0) {
957 ep->com.cm_id->rem_ref(ep->com.cm_id);
958 ep->com.cm_id = NULL;
959 }
960 }
961
962 static void connect_request_upcall(struct c4iw_ep *ep)
963 {
964 struct iw_cm_event event;
965
966 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
967 memset(&event, 0, sizeof(event));
968 event.event = IW_CM_EVENT_CONNECT_REQUEST;
969 event.local_addr = ep->com.local_addr;
970 event.remote_addr = ep->com.remote_addr;
971 event.provider_data = ep;
972 if (!ep->tried_with_mpa_v1) {
973 /* this means MPA_v2 is used */
974 event.ord = ep->ord;
975 event.ird = ep->ird;
976 event.private_data_len = ep->plen -
977 sizeof(struct mpa_v2_conn_params);
978 event.private_data = ep->mpa_pkt + sizeof(struct mpa_message) +
979 sizeof(struct mpa_v2_conn_params);
980 } else {
981 /* this means MPA_v1 is used. Send max supported */
982 event.ord = c4iw_max_read_depth;
983 event.ird = c4iw_max_read_depth;
984 event.private_data_len = ep->plen;
985 event.private_data = ep->mpa_pkt + sizeof(struct mpa_message);
986 }
987 if (state_read(&ep->parent_ep->com) != DEAD) {
988 c4iw_get_ep(&ep->com);
989 ep->parent_ep->com.cm_id->event_handler(
990 ep->parent_ep->com.cm_id,
991 &event);
992 }
993 set_bit(CONNREQ_UPCALL, &ep->com.history);
994 c4iw_put_ep(&ep->parent_ep->com);
995 ep->parent_ep = NULL;
996 }
997
998 static void established_upcall(struct c4iw_ep *ep)
999 {
1000 struct iw_cm_event event;
1001
1002 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1003 memset(&event, 0, sizeof(event));
1004 event.event = IW_CM_EVENT_ESTABLISHED;
1005 event.ird = ep->ird;
1006 event.ord = ep->ord;
1007 if (ep->com.cm_id) {
1008 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1009 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1010 set_bit(ESTAB_UPCALL, &ep->com.history);
1011 }
1012 }
1013
1014 static int update_rx_credits(struct c4iw_ep *ep, u32 credits)
1015 {
1016 struct cpl_rx_data_ack *req;
1017 struct sk_buff *skb;
1018 int wrlen = roundup(sizeof *req, 16);
1019
1020 PDBG("%s ep %p tid %u credits %u\n", __func__, ep, ep->hwtid, credits);
1021 skb = get_skb(NULL, wrlen, GFP_KERNEL);
1022 if (!skb) {
1023 printk(KERN_ERR MOD "update_rx_credits - cannot alloc skb!\n");
1024 return 0;
1025 }
1026
1027 req = (struct cpl_rx_data_ack *) skb_put(skb, wrlen);
1028 memset(req, 0, wrlen);
1029 INIT_TP_WR(req, ep->hwtid);
1030 OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_RX_DATA_ACK,
1031 ep->hwtid));
1032 req->credit_dack = cpu_to_be32(credits | RX_FORCE_ACK(1) |
1033 F_RX_DACK_CHANGE |
1034 V_RX_DACK_MODE(dack_mode));
1035 set_wr_txq(skb, CPL_PRIORITY_ACK, ep->ctrlq_idx);
1036 c4iw_ofld_send(&ep->com.dev->rdev, skb);
1037 return credits;
1038 }
1039
1040 static void process_mpa_reply(struct c4iw_ep *ep, struct sk_buff *skb)
1041 {
1042 struct mpa_message *mpa;
1043 struct mpa_v2_conn_params *mpa_v2_params;
1044 u16 plen;
1045 u16 resp_ird, resp_ord;
1046 u8 rtr_mismatch = 0, insuff_ird = 0;
1047 struct c4iw_qp_attributes attrs;
1048 enum c4iw_qp_attr_mask mask;
1049 int err;
1050
1051 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1052
1053 /*
1054 * Stop mpa timer. If it expired, then the state has
1055 * changed and we bail since ep_timeout already aborted
1056 * the connection.
1057 */
1058 stop_ep_timer(ep);
1059 if (state_read(&ep->com) != MPA_REQ_SENT)
1060 return;
1061
1062 /*
1063 * If we get more than the supported amount of private data
1064 * then we must fail this connection.
1065 */
1066 if (ep->mpa_pkt_len + skb->len > sizeof(ep->mpa_pkt)) {
1067 err = -EINVAL;
1068 goto err;
1069 }
1070
1071 /*
1072 * copy the new data into our accumulation buffer.
1073 */
1074 skb_copy_from_linear_data(skb, &(ep->mpa_pkt[ep->mpa_pkt_len]),
1075 skb->len);
1076 ep->mpa_pkt_len += skb->len;
1077
1078 /*
1079 * if we don't even have the mpa message, then bail.
1080 */
1081 if (ep->mpa_pkt_len < sizeof(*mpa))
1082 return;
1083 mpa = (struct mpa_message *) ep->mpa_pkt;
1084
1085 /* Validate MPA header. */
1086 if (mpa->revision > mpa_rev) {
1087 printk(KERN_ERR MOD "%s MPA version mismatch. Local = %d,"
1088 " Received = %d\n", __func__, mpa_rev, mpa->revision);
1089 err = -EPROTO;
1090 goto err;
1091 }
1092 if (memcmp(mpa->key, MPA_KEY_REP, sizeof(mpa->key))) {
1093 err = -EPROTO;
1094 goto err;
1095 }
1096
1097 plen = ntohs(mpa->private_data_size);
1098
1099 /*
1100 * Fail if there's too much private data.
1101 */
1102 if (plen > MPA_MAX_PRIVATE_DATA) {
1103 err = -EPROTO;
1104 goto err;
1105 }
1106
1107 /*
1108 * If plen does not account for pkt size
1109 */
1110 if (ep->mpa_pkt_len > (sizeof(*mpa) + plen)) {
1111 err = -EPROTO;
1112 goto err;
1113 }
1114
1115 ep->plen = (u8) plen;
1116
1117 /*
1118 * If we don't have all the pdata yet, then bail.
1119 * We'll continue process when more data arrives.
1120 */
1121 if (ep->mpa_pkt_len < (sizeof(*mpa) + plen))
1122 return;
1123
1124 if (mpa->flags & MPA_REJECT) {
1125 err = -ECONNREFUSED;
1126 goto err;
1127 }
1128
1129 /*
1130 * If we get here we have accumulated the entire mpa
1131 * start reply message including private data. And
1132 * the MPA header is valid.
1133 */
1134 state_set(&ep->com, FPDU_MODE);
1135 ep->mpa_attr.crc_enabled = (mpa->flags & MPA_CRC) | crc_enabled ? 1 : 0;
1136 ep->mpa_attr.recv_marker_enabled = markers_enabled;
1137 ep->mpa_attr.xmit_marker_enabled = mpa->flags & MPA_MARKERS ? 1 : 0;
1138 ep->mpa_attr.version = mpa->revision;
1139 ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED;
1140
1141 if (mpa->revision == 2) {
1142 ep->mpa_attr.enhanced_rdma_conn =
1143 mpa->flags & MPA_ENHANCED_RDMA_CONN ? 1 : 0;
1144 if (ep->mpa_attr.enhanced_rdma_conn) {
1145 mpa_v2_params = (struct mpa_v2_conn_params *)
1146 (ep->mpa_pkt + sizeof(*mpa));
1147 resp_ird = ntohs(mpa_v2_params->ird) &
1148 MPA_V2_IRD_ORD_MASK;
1149 resp_ord = ntohs(mpa_v2_params->ord) &
1150 MPA_V2_IRD_ORD_MASK;
1151
1152 /*
1153 * This is a double-check. Ideally, below checks are
1154 * not required since ird/ord stuff has been taken
1155 * care of in c4iw_accept_cr
1156 */
1157 if ((ep->ird < resp_ord) || (ep->ord > resp_ird)) {
1158 err = -ENOMEM;
1159 ep->ird = resp_ord;
1160 ep->ord = resp_ird;
1161 insuff_ird = 1;
1162 }
1163
1164 if (ntohs(mpa_v2_params->ird) &
1165 MPA_V2_PEER2PEER_MODEL) {
1166 if (ntohs(mpa_v2_params->ord) &
1167 MPA_V2_RDMA_WRITE_RTR)
1168 ep->mpa_attr.p2p_type =
1169 FW_RI_INIT_P2PTYPE_RDMA_WRITE;
1170 else if (ntohs(mpa_v2_params->ord) &
1171 MPA_V2_RDMA_READ_RTR)
1172 ep->mpa_attr.p2p_type =
1173 FW_RI_INIT_P2PTYPE_READ_REQ;
1174 }
1175 }
1176 } else if (mpa->revision == 1)
1177 if (peer2peer)
1178 ep->mpa_attr.p2p_type = p2p_type;
1179
1180 PDBG("%s - crc_enabled=%d, recv_marker_enabled=%d, "
1181 "xmit_marker_enabled=%d, version=%d p2p_type=%d local-p2p_type = "
1182 "%d\n", __func__, ep->mpa_attr.crc_enabled,
1183 ep->mpa_attr.recv_marker_enabled,
1184 ep->mpa_attr.xmit_marker_enabled, ep->mpa_attr.version,
1185 ep->mpa_attr.p2p_type, p2p_type);
1186
1187 /*
1188 * If responder's RTR does not match with that of initiator, assign
1189 * FW_RI_INIT_P2PTYPE_DISABLED in mpa attributes so that RTR is not
1190 * generated when moving QP to RTS state.
1191 * A TERM message will be sent after QP has moved to RTS state
1192 */
1193 if ((ep->mpa_attr.version == 2) && peer2peer &&
1194 (ep->mpa_attr.p2p_type != p2p_type)) {
1195 ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED;
1196 rtr_mismatch = 1;
1197 }
1198
1199 attrs.mpa_attr = ep->mpa_attr;
1200 attrs.max_ird = ep->ird;
1201 attrs.max_ord = ep->ord;
1202 attrs.llp_stream_handle = ep;
1203 attrs.next_state = C4IW_QP_STATE_RTS;
1204
1205 mask = C4IW_QP_ATTR_NEXT_STATE |
1206 C4IW_QP_ATTR_LLP_STREAM_HANDLE | C4IW_QP_ATTR_MPA_ATTR |
1207 C4IW_QP_ATTR_MAX_IRD | C4IW_QP_ATTR_MAX_ORD;
1208
1209 /* bind QP and TID with INIT_WR */
1210 err = c4iw_modify_qp(ep->com.qp->rhp,
1211 ep->com.qp, mask, &attrs, 1);
1212 if (err)
1213 goto err;
1214
1215 /*
1216 * If responder's RTR requirement did not match with what initiator
1217 * supports, generate TERM message
1218 */
1219 if (rtr_mismatch) {
1220 printk(KERN_ERR "%s: RTR mismatch, sending TERM\n", __func__);
1221 attrs.layer_etype = LAYER_MPA | DDP_LLP;
1222 attrs.ecode = MPA_NOMATCH_RTR;
1223 attrs.next_state = C4IW_QP_STATE_TERMINATE;
1224 err = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
1225 C4IW_QP_ATTR_NEXT_STATE, &attrs, 0);
1226 err = -ENOMEM;
1227 goto out;
1228 }
1229
1230 /*
1231 * Generate TERM if initiator IRD is not sufficient for responder
1232 * provided ORD. Currently, we do the same behaviour even when
1233 * responder provided IRD is also not sufficient as regards to
1234 * initiator ORD.
1235 */
1236 if (insuff_ird) {
1237 printk(KERN_ERR "%s: Insufficient IRD, sending TERM\n",
1238 __func__);
1239 attrs.layer_etype = LAYER_MPA | DDP_LLP;
1240 attrs.ecode = MPA_INSUFF_IRD;
1241 attrs.next_state = C4IW_QP_STATE_TERMINATE;
1242 err = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
1243 C4IW_QP_ATTR_NEXT_STATE, &attrs, 0);
1244 err = -ENOMEM;
1245 goto out;
1246 }
1247 goto out;
1248 err:
1249 state_set(&ep->com, ABORTING);
1250 send_abort(ep, skb, GFP_KERNEL);
1251 out:
1252 connect_reply_upcall(ep, err);
1253 return;
1254 }
1255
1256 static void process_mpa_request(struct c4iw_ep *ep, struct sk_buff *skb)
1257 {
1258 struct mpa_message *mpa;
1259 struct mpa_v2_conn_params *mpa_v2_params;
1260 u16 plen;
1261
1262 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1263
1264 if (state_read(&ep->com) != MPA_REQ_WAIT)
1265 return;
1266
1267 /*
1268 * If we get more than the supported amount of private data
1269 * then we must fail this connection.
1270 */
1271 if (ep->mpa_pkt_len + skb->len > sizeof(ep->mpa_pkt)) {
1272 stop_ep_timer(ep);
1273 abort_connection(ep, skb, GFP_KERNEL);
1274 return;
1275 }
1276
1277 PDBG("%s enter (%s line %u)\n", __func__, __FILE__, __LINE__);
1278
1279 /*
1280 * Copy the new data into our accumulation buffer.
1281 */
1282 skb_copy_from_linear_data(skb, &(ep->mpa_pkt[ep->mpa_pkt_len]),
1283 skb->len);
1284 ep->mpa_pkt_len += skb->len;
1285
1286 /*
1287 * If we don't even have the mpa message, then bail.
1288 * We'll continue process when more data arrives.
1289 */
1290 if (ep->mpa_pkt_len < sizeof(*mpa))
1291 return;
1292
1293 PDBG("%s enter (%s line %u)\n", __func__, __FILE__, __LINE__);
1294 stop_ep_timer(ep);
1295 mpa = (struct mpa_message *) ep->mpa_pkt;
1296
1297 /*
1298 * Validate MPA Header.
1299 */
1300 if (mpa->revision > mpa_rev) {
1301 printk(KERN_ERR MOD "%s MPA version mismatch. Local = %d,"
1302 " Received = %d\n", __func__, mpa_rev, mpa->revision);
1303 stop_ep_timer(ep);
1304 abort_connection(ep, skb, GFP_KERNEL);
1305 return;
1306 }
1307
1308 if (memcmp(mpa->key, MPA_KEY_REQ, sizeof(mpa->key))) {
1309 stop_ep_timer(ep);
1310 abort_connection(ep, skb, GFP_KERNEL);
1311 return;
1312 }
1313
1314 plen = ntohs(mpa->private_data_size);
1315
1316 /*
1317 * Fail if there's too much private data.
1318 */
1319 if (plen > MPA_MAX_PRIVATE_DATA) {
1320 stop_ep_timer(ep);
1321 abort_connection(ep, skb, GFP_KERNEL);
1322 return;
1323 }
1324
1325 /*
1326 * If plen does not account for pkt size
1327 */
1328 if (ep->mpa_pkt_len > (sizeof(*mpa) + plen)) {
1329 stop_ep_timer(ep);
1330 abort_connection(ep, skb, GFP_KERNEL);
1331 return;
1332 }
1333 ep->plen = (u8) plen;
1334
1335 /*
1336 * If we don't have all the pdata yet, then bail.
1337 */
1338 if (ep->mpa_pkt_len < (sizeof(*mpa) + plen))
1339 return;
1340
1341 /*
1342 * If we get here we have accumulated the entire mpa
1343 * start reply message including private data.
1344 */
1345 ep->mpa_attr.initiator = 0;
1346 ep->mpa_attr.crc_enabled = (mpa->flags & MPA_CRC) | crc_enabled ? 1 : 0;
1347 ep->mpa_attr.recv_marker_enabled = markers_enabled;
1348 ep->mpa_attr.xmit_marker_enabled = mpa->flags & MPA_MARKERS ? 1 : 0;
1349 ep->mpa_attr.version = mpa->revision;
1350 if (mpa->revision == 1)
1351 ep->tried_with_mpa_v1 = 1;
1352 ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED;
1353
1354 if (mpa->revision == 2) {
1355 ep->mpa_attr.enhanced_rdma_conn =
1356 mpa->flags & MPA_ENHANCED_RDMA_CONN ? 1 : 0;
1357 if (ep->mpa_attr.enhanced_rdma_conn) {
1358 mpa_v2_params = (struct mpa_v2_conn_params *)
1359 (ep->mpa_pkt + sizeof(*mpa));
1360 ep->ird = ntohs(mpa_v2_params->ird) &
1361 MPA_V2_IRD_ORD_MASK;
1362 ep->ord = ntohs(mpa_v2_params->ord) &
1363 MPA_V2_IRD_ORD_MASK;
1364 if (ntohs(mpa_v2_params->ird) & MPA_V2_PEER2PEER_MODEL)
1365 if (peer2peer) {
1366 if (ntohs(mpa_v2_params->ord) &
1367 MPA_V2_RDMA_WRITE_RTR)
1368 ep->mpa_attr.p2p_type =
1369 FW_RI_INIT_P2PTYPE_RDMA_WRITE;
1370 else if (ntohs(mpa_v2_params->ord) &
1371 MPA_V2_RDMA_READ_RTR)
1372 ep->mpa_attr.p2p_type =
1373 FW_RI_INIT_P2PTYPE_READ_REQ;
1374 }
1375 }
1376 } else if (mpa->revision == 1)
1377 if (peer2peer)
1378 ep->mpa_attr.p2p_type = p2p_type;
1379
1380 PDBG("%s - crc_enabled=%d, recv_marker_enabled=%d, "
1381 "xmit_marker_enabled=%d, version=%d p2p_type=%d\n", __func__,
1382 ep->mpa_attr.crc_enabled, ep->mpa_attr.recv_marker_enabled,
1383 ep->mpa_attr.xmit_marker_enabled, ep->mpa_attr.version,
1384 ep->mpa_attr.p2p_type);
1385
1386 state_set(&ep->com, MPA_REQ_RCVD);
1387
1388 /* drive upcall */
1389 connect_request_upcall(ep);
1390 return;
1391 }
1392
1393 static int rx_data(struct c4iw_dev *dev, struct sk_buff *skb)
1394 {
1395 struct c4iw_ep *ep;
1396 struct cpl_rx_data *hdr = cplhdr(skb);
1397 unsigned int dlen = ntohs(hdr->len);
1398 unsigned int tid = GET_TID(hdr);
1399 struct tid_info *t = dev->rdev.lldi.tids;
1400 __u8 status = hdr->status;
1401
1402 ep = lookup_tid(t, tid);
1403 PDBG("%s ep %p tid %u dlen %u\n", __func__, ep, ep->hwtid, dlen);
1404 skb_pull(skb, sizeof(*hdr));
1405 skb_trim(skb, dlen);
1406
1407 /* update RX credits */
1408 update_rx_credits(ep, dlen);
1409
1410 switch (state_read(&ep->com)) {
1411 case MPA_REQ_SENT:
1412 ep->rcv_seq += dlen;
1413 process_mpa_reply(ep, skb);
1414 break;
1415 case MPA_REQ_WAIT:
1416 ep->rcv_seq += dlen;
1417 process_mpa_request(ep, skb);
1418 break;
1419 case FPDU_MODE: {
1420 struct c4iw_qp_attributes attrs;
1421 BUG_ON(!ep->com.qp);
1422 if (status)
1423 pr_err("%s Unexpected streaming data." \
1424 " qpid %u ep %p state %d tid %u status %d\n",
1425 __func__, ep->com.qp->wq.sq.qid, ep,
1426 state_read(&ep->com), ep->hwtid, status);
1427 attrs.next_state = C4IW_QP_STATE_ERROR;
1428 c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
1429 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
1430 c4iw_ep_disconnect(ep, 1, GFP_KERNEL);
1431 break;
1432 }
1433 default:
1434 break;
1435 }
1436 return 0;
1437 }
1438
1439 static int abort_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
1440 {
1441 struct c4iw_ep *ep;
1442 struct cpl_abort_rpl_rss *rpl = cplhdr(skb);
1443 int release = 0;
1444 unsigned int tid = GET_TID(rpl);
1445 struct tid_info *t = dev->rdev.lldi.tids;
1446
1447 ep = lookup_tid(t, tid);
1448 if (!ep) {
1449 printk(KERN_WARNING MOD "Abort rpl to freed endpoint\n");
1450 return 0;
1451 }
1452 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1453 mutex_lock(&ep->com.mutex);
1454 switch (ep->com.state) {
1455 case ABORTING:
1456 c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
1457 __state_set(&ep->com, DEAD);
1458 release = 1;
1459 break;
1460 default:
1461 printk(KERN_ERR "%s ep %p state %d\n",
1462 __func__, ep, ep->com.state);
1463 break;
1464 }
1465 mutex_unlock(&ep->com.mutex);
1466
1467 if (release)
1468 release_ep_resources(ep);
1469 return 0;
1470 }
1471
1472 static void send_fw_act_open_req(struct c4iw_ep *ep, unsigned int atid)
1473 {
1474 struct sk_buff *skb;
1475 struct fw_ofld_connection_wr *req;
1476 unsigned int mtu_idx;
1477 int wscale;
1478
1479 skb = get_skb(NULL, sizeof(*req), GFP_KERNEL);
1480 req = (struct fw_ofld_connection_wr *)__skb_put(skb, sizeof(*req));
1481 memset(req, 0, sizeof(*req));
1482 req->op_compl = htonl(V_WR_OP(FW_OFLD_CONNECTION_WR));
1483 req->len16_pkd = htonl(FW_WR_LEN16(DIV_ROUND_UP(sizeof(*req), 16)));
1484 req->le.filter = cpu_to_be32(select_ntuple(ep->com.dev, ep->dst,
1485 ep->l2t));
1486 req->le.lport = ep->com.local_addr.sin_port;
1487 req->le.pport = ep->com.remote_addr.sin_port;
1488 req->le.u.ipv4.lip = ep->com.local_addr.sin_addr.s_addr;
1489 req->le.u.ipv4.pip = ep->com.remote_addr.sin_addr.s_addr;
1490 req->tcb.t_state_to_astid =
1491 htonl(V_FW_OFLD_CONNECTION_WR_T_STATE(TCP_SYN_SENT) |
1492 V_FW_OFLD_CONNECTION_WR_ASTID(atid));
1493 req->tcb.cplrxdataack_cplpassacceptrpl =
1494 htons(F_FW_OFLD_CONNECTION_WR_CPLRXDATAACK);
1495 req->tcb.tx_max = (__force __be32) jiffies;
1496 req->tcb.rcv_adv = htons(1);
1497 cxgb4_best_mtu(ep->com.dev->rdev.lldi.mtus, ep->mtu, &mtu_idx);
1498 wscale = compute_wscale(rcv_win);
1499 req->tcb.opt0 = (__force __be64) (TCAM_BYPASS(1) |
1500 (nocong ? NO_CONG(1) : 0) |
1501 KEEP_ALIVE(1) |
1502 DELACK(1) |
1503 WND_SCALE(wscale) |
1504 MSS_IDX(mtu_idx) |
1505 L2T_IDX(ep->l2t->idx) |
1506 TX_CHAN(ep->tx_chan) |
1507 SMAC_SEL(ep->smac_idx) |
1508 DSCP(ep->tos) |
1509 ULP_MODE(ULP_MODE_TCPDDP) |
1510 RCV_BUFSIZ(rcv_win >> 10));
1511 req->tcb.opt2 = (__force __be32) (PACE(1) |
1512 TX_QUEUE(ep->com.dev->rdev.lldi.tx_modq[ep->tx_chan]) |
1513 RX_CHANNEL(0) |
1514 CCTRL_ECN(enable_ecn) |
1515 RSS_QUEUE_VALID | RSS_QUEUE(ep->rss_qid));
1516 if (enable_tcp_timestamps)
1517 req->tcb.opt2 |= (__force __be32) TSTAMPS_EN(1);
1518 if (enable_tcp_sack)
1519 req->tcb.opt2 |= (__force __be32) SACK_EN(1);
1520 if (wscale && enable_tcp_window_scaling)
1521 req->tcb.opt2 |= (__force __be32) WND_SCALE_EN(1);
1522 req->tcb.opt0 = cpu_to_be64((__force u64) req->tcb.opt0);
1523 req->tcb.opt2 = cpu_to_be32((__force u32) req->tcb.opt2);
1524 set_wr_txq(skb, CPL_PRIORITY_CONTROL, ep->ctrlq_idx);
1525 set_bit(ACT_OFLD_CONN, &ep->com.history);
1526 c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
1527 }
1528
1529 /*
1530 * Return whether a failed active open has allocated a TID
1531 */
1532 static inline int act_open_has_tid(int status)
1533 {
1534 return status != CPL_ERR_TCAM_FULL && status != CPL_ERR_CONN_EXIST &&
1535 status != CPL_ERR_ARP_MISS;
1536 }
1537
1538 #define ACT_OPEN_RETRY_COUNT 2
1539
1540 static int c4iw_reconnect(struct c4iw_ep *ep)
1541 {
1542 int err = 0;
1543 struct rtable *rt;
1544 struct port_info *pi;
1545 struct net_device *pdev;
1546 int step;
1547 struct neighbour *neigh;
1548
1549 PDBG("%s qp %p cm_id %p\n", __func__, ep->com.qp, ep->com.cm_id);
1550 init_timer(&ep->timer);
1551
1552 /*
1553 * Allocate an active TID to initiate a TCP connection.
1554 */
1555 ep->atid = cxgb4_alloc_atid(ep->com.dev->rdev.lldi.tids, ep);
1556 if (ep->atid == -1) {
1557 pr_err("%s - cannot alloc atid.\n", __func__);
1558 err = -ENOMEM;
1559 goto fail2;
1560 }
1561 insert_handle(ep->com.dev, &ep->com.dev->atid_idr, ep, ep->atid);
1562
1563 /* find a route */
1564 rt = find_route(ep->com.dev,
1565 ep->com.cm_id->local_addr.sin_addr.s_addr,
1566 ep->com.cm_id->remote_addr.sin_addr.s_addr,
1567 ep->com.cm_id->local_addr.sin_port,
1568 ep->com.cm_id->remote_addr.sin_port, 0);
1569 if (!rt) {
1570 pr_err("%s - cannot find route.\n", __func__);
1571 err = -EHOSTUNREACH;
1572 goto fail3;
1573 }
1574 ep->dst = &rt->dst;
1575
1576 neigh = dst_neigh_lookup(ep->dst,
1577 &ep->com.cm_id->remote_addr.sin_addr.s_addr);
1578 /* get a l2t entry */
1579 if (neigh->dev->flags & IFF_LOOPBACK) {
1580 PDBG("%s LOOPBACK\n", __func__);
1581 pdev = ip_dev_find(&init_net,
1582 ep->com.cm_id->remote_addr.sin_addr.s_addr);
1583 ep->l2t = cxgb4_l2t_get(ep->com.dev->rdev.lldi.l2t,
1584 neigh, pdev, 0);
1585 pi = (struct port_info *)netdev_priv(pdev);
1586 ep->mtu = pdev->mtu;
1587 ep->tx_chan = cxgb4_port_chan(pdev);
1588 ep->smac_idx = (cxgb4_port_viid(pdev) & 0x7F) << 1;
1589 dev_put(pdev);
1590 } else {
1591 ep->l2t = cxgb4_l2t_get(ep->com.dev->rdev.lldi.l2t,
1592 neigh, neigh->dev, 0);
1593 pi = (struct port_info *)netdev_priv(neigh->dev);
1594 ep->mtu = dst_mtu(ep->dst);
1595 ep->tx_chan = cxgb4_port_chan(neigh->dev);
1596 ep->smac_idx = (cxgb4_port_viid(neigh->dev) &
1597 0x7F) << 1;
1598 }
1599
1600 step = ep->com.dev->rdev.lldi.ntxq / ep->com.dev->rdev.lldi.nchan;
1601 ep->txq_idx = pi->port_id * step;
1602 ep->ctrlq_idx = pi->port_id;
1603 step = ep->com.dev->rdev.lldi.nrxq / ep->com.dev->rdev.lldi.nchan;
1604 ep->rss_qid = ep->com.dev->rdev.lldi.rxq_ids[pi->port_id * step];
1605
1606 if (!ep->l2t) {
1607 pr_err("%s - cannot alloc l2e.\n", __func__);
1608 err = -ENOMEM;
1609 goto fail4;
1610 }
1611
1612 PDBG("%s txq_idx %u tx_chan %u smac_idx %u rss_qid %u l2t_idx %u\n",
1613 __func__, ep->txq_idx, ep->tx_chan, ep->smac_idx, ep->rss_qid,
1614 ep->l2t->idx);
1615
1616 state_set(&ep->com, CONNECTING);
1617 ep->tos = 0;
1618
1619 /* send connect request to rnic */
1620 err = send_connect(ep);
1621 if (!err)
1622 goto out;
1623
1624 cxgb4_l2t_release(ep->l2t);
1625 fail4:
1626 dst_release(ep->dst);
1627 fail3:
1628 remove_handle(ep->com.dev, &ep->com.dev->atid_idr, ep->atid);
1629 cxgb4_free_atid(ep->com.dev->rdev.lldi.tids, ep->atid);
1630 fail2:
1631 /*
1632 * remember to send notification to upper layer.
1633 * We are in here so the upper layer is not aware that this is
1634 * re-connect attempt and so, upper layer is still waiting for
1635 * response of 1st connect request.
1636 */
1637 connect_reply_upcall(ep, -ECONNRESET);
1638 c4iw_put_ep(&ep->com);
1639 out:
1640 return err;
1641 }
1642
1643 static int act_open_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
1644 {
1645 struct c4iw_ep *ep;
1646 struct cpl_act_open_rpl *rpl = cplhdr(skb);
1647 unsigned int atid = GET_TID_TID(GET_AOPEN_ATID(
1648 ntohl(rpl->atid_status)));
1649 struct tid_info *t = dev->rdev.lldi.tids;
1650 int status = GET_AOPEN_STATUS(ntohl(rpl->atid_status));
1651
1652 ep = lookup_atid(t, atid);
1653
1654 PDBG("%s ep %p atid %u status %u errno %d\n", __func__, ep, atid,
1655 status, status2errno(status));
1656
1657 if (status == CPL_ERR_RTX_NEG_ADVICE) {
1658 printk(KERN_WARNING MOD "Connection problems for atid %u\n",
1659 atid);
1660 return 0;
1661 }
1662
1663 set_bit(ACT_OPEN_RPL, &ep->com.history);
1664
1665 /*
1666 * Log interesting failures.
1667 */
1668 switch (status) {
1669 case CPL_ERR_CONN_RESET:
1670 case CPL_ERR_CONN_TIMEDOUT:
1671 break;
1672 case CPL_ERR_TCAM_FULL:
1673 if (dev->rdev.lldi.enable_fw_ofld_conn) {
1674 mutex_lock(&dev->rdev.stats.lock);
1675 dev->rdev.stats.tcam_full++;
1676 mutex_unlock(&dev->rdev.stats.lock);
1677 send_fw_act_open_req(ep,
1678 GET_TID_TID(GET_AOPEN_ATID(
1679 ntohl(rpl->atid_status))));
1680 return 0;
1681 }
1682 break;
1683 case CPL_ERR_CONN_EXIST:
1684 if (ep->retry_count++ < ACT_OPEN_RETRY_COUNT) {
1685 set_bit(ACT_RETRY_INUSE, &ep->com.history);
1686 remove_handle(ep->com.dev, &ep->com.dev->atid_idr,
1687 atid);
1688 cxgb4_free_atid(t, atid);
1689 dst_release(ep->dst);
1690 cxgb4_l2t_release(ep->l2t);
1691 c4iw_reconnect(ep);
1692 return 0;
1693 }
1694 break;
1695 default:
1696 printk(KERN_INFO MOD "Active open failure - "
1697 "atid %u status %u errno %d %pI4:%u->%pI4:%u\n",
1698 atid, status, status2errno(status),
1699 &ep->com.local_addr.sin_addr.s_addr,
1700 ntohs(ep->com.local_addr.sin_port),
1701 &ep->com.remote_addr.sin_addr.s_addr,
1702 ntohs(ep->com.remote_addr.sin_port));
1703 break;
1704 }
1705
1706 connect_reply_upcall(ep, status2errno(status));
1707 state_set(&ep->com, DEAD);
1708
1709 if (status && act_open_has_tid(status))
1710 cxgb4_remove_tid(ep->com.dev->rdev.lldi.tids, 0, GET_TID(rpl));
1711
1712 remove_handle(ep->com.dev, &ep->com.dev->atid_idr, atid);
1713 cxgb4_free_atid(t, atid);
1714 dst_release(ep->dst);
1715 cxgb4_l2t_release(ep->l2t);
1716 c4iw_put_ep(&ep->com);
1717
1718 return 0;
1719 }
1720
1721 static int pass_open_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
1722 {
1723 struct cpl_pass_open_rpl *rpl = cplhdr(skb);
1724 struct tid_info *t = dev->rdev.lldi.tids;
1725 unsigned int stid = GET_TID(rpl);
1726 struct c4iw_listen_ep *ep = lookup_stid(t, stid);
1727
1728 if (!ep) {
1729 PDBG("%s stid %d lookup failure!\n", __func__, stid);
1730 goto out;
1731 }
1732 PDBG("%s ep %p status %d error %d\n", __func__, ep,
1733 rpl->status, status2errno(rpl->status));
1734 c4iw_wake_up(&ep->com.wr_wait, status2errno(rpl->status));
1735
1736 out:
1737 return 0;
1738 }
1739
1740 static int listen_stop(struct c4iw_listen_ep *ep)
1741 {
1742 struct sk_buff *skb;
1743 struct cpl_close_listsvr_req *req;
1744
1745 PDBG("%s ep %p\n", __func__, ep);
1746 skb = get_skb(NULL, sizeof(*req), GFP_KERNEL);
1747 if (!skb) {
1748 printk(KERN_ERR MOD "%s - failed to alloc skb\n", __func__);
1749 return -ENOMEM;
1750 }
1751 req = (struct cpl_close_listsvr_req *) skb_put(skb, sizeof(*req));
1752 INIT_TP_WR(req, 0);
1753 OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_CLOSE_LISTSRV_REQ,
1754 ep->stid));
1755 req->reply_ctrl = cpu_to_be16(
1756 QUEUENO(ep->com.dev->rdev.lldi.rxq_ids[0]));
1757 set_wr_txq(skb, CPL_PRIORITY_SETUP, 0);
1758 return c4iw_ofld_send(&ep->com.dev->rdev, skb);
1759 }
1760
1761 static int close_listsrv_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
1762 {
1763 struct cpl_close_listsvr_rpl *rpl = cplhdr(skb);
1764 struct tid_info *t = dev->rdev.lldi.tids;
1765 unsigned int stid = GET_TID(rpl);
1766 struct c4iw_listen_ep *ep = lookup_stid(t, stid);
1767
1768 PDBG("%s ep %p\n", __func__, ep);
1769 c4iw_wake_up(&ep->com.wr_wait, status2errno(rpl->status));
1770 return 0;
1771 }
1772
1773 static void accept_cr(struct c4iw_ep *ep, __be32 peer_ip, struct sk_buff *skb,
1774 struct cpl_pass_accept_req *req)
1775 {
1776 struct cpl_pass_accept_rpl *rpl;
1777 unsigned int mtu_idx;
1778 u64 opt0;
1779 u32 opt2;
1780 int wscale;
1781
1782 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1783 BUG_ON(skb_cloned(skb));
1784 skb_trim(skb, sizeof(*rpl));
1785 skb_get(skb);
1786 cxgb4_best_mtu(ep->com.dev->rdev.lldi.mtus, ep->mtu, &mtu_idx);
1787 wscale = compute_wscale(rcv_win);
1788 opt0 = (nocong ? NO_CONG(1) : 0) |
1789 KEEP_ALIVE(1) |
1790 DELACK(1) |
1791 WND_SCALE(wscale) |
1792 MSS_IDX(mtu_idx) |
1793 L2T_IDX(ep->l2t->idx) |
1794 TX_CHAN(ep->tx_chan) |
1795 SMAC_SEL(ep->smac_idx) |
1796 DSCP(ep->tos >> 2) |
1797 ULP_MODE(ULP_MODE_TCPDDP) |
1798 RCV_BUFSIZ(rcv_win>>10);
1799 opt2 = RX_CHANNEL(0) |
1800 RSS_QUEUE_VALID | RSS_QUEUE(ep->rss_qid);
1801
1802 if (enable_tcp_timestamps && req->tcpopt.tstamp)
1803 opt2 |= TSTAMPS_EN(1);
1804 if (enable_tcp_sack && req->tcpopt.sack)
1805 opt2 |= SACK_EN(1);
1806 if (wscale && enable_tcp_window_scaling)
1807 opt2 |= WND_SCALE_EN(1);
1808 if (enable_ecn) {
1809 const struct tcphdr *tcph;
1810 u32 hlen = ntohl(req->hdr_len);
1811
1812 tcph = (const void *)(req + 1) + G_ETH_HDR_LEN(hlen) +
1813 G_IP_HDR_LEN(hlen);
1814 if (tcph->ece && tcph->cwr)
1815 opt2 |= CCTRL_ECN(1);
1816 }
1817
1818 rpl = cplhdr(skb);
1819 INIT_TP_WR(rpl, ep->hwtid);
1820 OPCODE_TID(rpl) = cpu_to_be32(MK_OPCODE_TID(CPL_PASS_ACCEPT_RPL,
1821 ep->hwtid));
1822 rpl->opt0 = cpu_to_be64(opt0);
1823 rpl->opt2 = cpu_to_be32(opt2);
1824 set_wr_txq(skb, CPL_PRIORITY_SETUP, ep->ctrlq_idx);
1825 c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
1826
1827 return;
1828 }
1829
1830 static void reject_cr(struct c4iw_dev *dev, u32 hwtid, __be32 peer_ip,
1831 struct sk_buff *skb)
1832 {
1833 PDBG("%s c4iw_dev %p tid %u peer_ip %x\n", __func__, dev, hwtid,
1834 peer_ip);
1835 BUG_ON(skb_cloned(skb));
1836 skb_trim(skb, sizeof(struct cpl_tid_release));
1837 skb_get(skb);
1838 release_tid(&dev->rdev, hwtid, skb);
1839 return;
1840 }
1841
1842 static void get_4tuple(struct cpl_pass_accept_req *req,
1843 __be32 *local_ip, __be32 *peer_ip,
1844 __be16 *local_port, __be16 *peer_port)
1845 {
1846 int eth_len = G_ETH_HDR_LEN(be32_to_cpu(req->hdr_len));
1847 int ip_len = G_IP_HDR_LEN(be32_to_cpu(req->hdr_len));
1848 struct iphdr *ip = (struct iphdr *)((u8 *)(req + 1) + eth_len);
1849 struct tcphdr *tcp = (struct tcphdr *)
1850 ((u8 *)(req + 1) + eth_len + ip_len);
1851
1852 PDBG("%s saddr 0x%x daddr 0x%x sport %u dport %u\n", __func__,
1853 ntohl(ip->saddr), ntohl(ip->daddr), ntohs(tcp->source),
1854 ntohs(tcp->dest));
1855
1856 *peer_ip = ip->saddr;
1857 *local_ip = ip->daddr;
1858 *peer_port = tcp->source;
1859 *local_port = tcp->dest;
1860
1861 return;
1862 }
1863
1864 static int import_ep(struct c4iw_ep *ep, __be32 peer_ip, struct dst_entry *dst,
1865 struct c4iw_dev *cdev, bool clear_mpa_v1)
1866 {
1867 struct neighbour *n;
1868 int err, step;
1869
1870 n = dst_neigh_lookup(dst, &peer_ip);
1871 if (!n)
1872 return -ENODEV;
1873
1874 rcu_read_lock();
1875 err = -ENOMEM;
1876 if (n->dev->flags & IFF_LOOPBACK) {
1877 struct net_device *pdev;
1878
1879 pdev = ip_dev_find(&init_net, peer_ip);
1880 if (!pdev) {
1881 err = -ENODEV;
1882 goto out;
1883 }
1884 ep->l2t = cxgb4_l2t_get(cdev->rdev.lldi.l2t,
1885 n, pdev, 0);
1886 if (!ep->l2t)
1887 goto out;
1888 ep->mtu = pdev->mtu;
1889 ep->tx_chan = cxgb4_port_chan(pdev);
1890 ep->smac_idx = (cxgb4_port_viid(pdev) & 0x7F) << 1;
1891 step = cdev->rdev.lldi.ntxq /
1892 cdev->rdev.lldi.nchan;
1893 ep->txq_idx = cxgb4_port_idx(pdev) * step;
1894 step = cdev->rdev.lldi.nrxq /
1895 cdev->rdev.lldi.nchan;
1896 ep->ctrlq_idx = cxgb4_port_idx(pdev);
1897 ep->rss_qid = cdev->rdev.lldi.rxq_ids[
1898 cxgb4_port_idx(pdev) * step];
1899 dev_put(pdev);
1900 } else {
1901 ep->l2t = cxgb4_l2t_get(cdev->rdev.lldi.l2t,
1902 n, n->dev, 0);
1903 if (!ep->l2t)
1904 goto out;
1905 ep->mtu = dst_mtu(dst);
1906 ep->tx_chan = cxgb4_port_chan(n->dev);
1907 ep->smac_idx = (cxgb4_port_viid(n->dev) & 0x7F) << 1;
1908 step = cdev->rdev.lldi.ntxq /
1909 cdev->rdev.lldi.nchan;
1910 ep->txq_idx = cxgb4_port_idx(n->dev) * step;
1911 ep->ctrlq_idx = cxgb4_port_idx(n->dev);
1912 step = cdev->rdev.lldi.nrxq /
1913 cdev->rdev.lldi.nchan;
1914 ep->rss_qid = cdev->rdev.lldi.rxq_ids[
1915 cxgb4_port_idx(n->dev) * step];
1916
1917 if (clear_mpa_v1) {
1918 ep->retry_with_mpa_v1 = 0;
1919 ep->tried_with_mpa_v1 = 0;
1920 }
1921 }
1922 err = 0;
1923 out:
1924 rcu_read_unlock();
1925
1926 neigh_release(n);
1927
1928 return err;
1929 }
1930
1931 static int pass_accept_req(struct c4iw_dev *dev, struct sk_buff *skb)
1932 {
1933 struct c4iw_ep *child_ep = NULL, *parent_ep;
1934 struct cpl_pass_accept_req *req = cplhdr(skb);
1935 unsigned int stid = GET_POPEN_TID(ntohl(req->tos_stid));
1936 struct tid_info *t = dev->rdev.lldi.tids;
1937 unsigned int hwtid = GET_TID(req);
1938 struct dst_entry *dst;
1939 struct rtable *rt;
1940 __be32 local_ip, peer_ip = 0;
1941 __be16 local_port, peer_port;
1942 int err;
1943 u16 peer_mss = ntohs(req->tcpopt.mss);
1944
1945 parent_ep = lookup_stid(t, stid);
1946 if (!parent_ep) {
1947 PDBG("%s connect request on invalid stid %d\n", __func__, stid);
1948 goto reject;
1949 }
1950 get_4tuple(req, &local_ip, &peer_ip, &local_port, &peer_port);
1951
1952 PDBG("%s parent ep %p hwtid %u laddr 0x%x raddr 0x%x lport %d " \
1953 "rport %d peer_mss %d\n", __func__, parent_ep, hwtid,
1954 ntohl(local_ip), ntohl(peer_ip), ntohs(local_port),
1955 ntohs(peer_port), peer_mss);
1956
1957 if (state_read(&parent_ep->com) != LISTEN) {
1958 printk(KERN_ERR "%s - listening ep not in LISTEN\n",
1959 __func__);
1960 goto reject;
1961 }
1962
1963 /* Find output route */
1964 rt = find_route(dev, local_ip, peer_ip, local_port, peer_port,
1965 GET_POPEN_TOS(ntohl(req->tos_stid)));
1966 if (!rt) {
1967 printk(KERN_ERR MOD "%s - failed to find dst entry!\n",
1968 __func__);
1969 goto reject;
1970 }
1971 dst = &rt->dst;
1972
1973 child_ep = alloc_ep(sizeof(*child_ep), GFP_KERNEL);
1974 if (!child_ep) {
1975 printk(KERN_ERR MOD "%s - failed to allocate ep entry!\n",
1976 __func__);
1977 dst_release(dst);
1978 goto reject;
1979 }
1980
1981 err = import_ep(child_ep, peer_ip, dst, dev, false);
1982 if (err) {
1983 printk(KERN_ERR MOD "%s - failed to allocate l2t entry!\n",
1984 __func__);
1985 dst_release(dst);
1986 kfree(child_ep);
1987 goto reject;
1988 }
1989
1990 if (peer_mss && child_ep->mtu > (peer_mss + 40))
1991 child_ep->mtu = peer_mss + 40;
1992
1993 state_set(&child_ep->com, CONNECTING);
1994 child_ep->com.dev = dev;
1995 child_ep->com.cm_id = NULL;
1996 child_ep->com.local_addr.sin_family = PF_INET;
1997 child_ep->com.local_addr.sin_port = local_port;
1998 child_ep->com.local_addr.sin_addr.s_addr = local_ip;
1999 child_ep->com.remote_addr.sin_family = PF_INET;
2000 child_ep->com.remote_addr.sin_port = peer_port;
2001 child_ep->com.remote_addr.sin_addr.s_addr = peer_ip;
2002 c4iw_get_ep(&parent_ep->com);
2003 child_ep->parent_ep = parent_ep;
2004 child_ep->tos = GET_POPEN_TOS(ntohl(req->tos_stid));
2005 child_ep->dst = dst;
2006 child_ep->hwtid = hwtid;
2007
2008 PDBG("%s tx_chan %u smac_idx %u rss_qid %u\n", __func__,
2009 child_ep->tx_chan, child_ep->smac_idx, child_ep->rss_qid);
2010
2011 init_timer(&child_ep->timer);
2012 cxgb4_insert_tid(t, child_ep, hwtid);
2013 insert_handle(dev, &dev->hwtid_idr, child_ep, child_ep->hwtid);
2014 accept_cr(child_ep, peer_ip, skb, req);
2015 set_bit(PASS_ACCEPT_REQ, &child_ep->com.history);
2016 goto out;
2017 reject:
2018 reject_cr(dev, hwtid, peer_ip, skb);
2019 out:
2020 return 0;
2021 }
2022
2023 static int pass_establish(struct c4iw_dev *dev, struct sk_buff *skb)
2024 {
2025 struct c4iw_ep *ep;
2026 struct cpl_pass_establish *req = cplhdr(skb);
2027 struct tid_info *t = dev->rdev.lldi.tids;
2028 unsigned int tid = GET_TID(req);
2029
2030 ep = lookup_tid(t, tid);
2031 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
2032 ep->snd_seq = be32_to_cpu(req->snd_isn);
2033 ep->rcv_seq = be32_to_cpu(req->rcv_isn);
2034
2035 PDBG("%s ep %p hwtid %u tcp_opt 0x%02x\n", __func__, ep, tid,
2036 ntohs(req->tcp_opt));
2037
2038 set_emss(ep, ntohs(req->tcp_opt));
2039
2040 dst_confirm(ep->dst);
2041 state_set(&ep->com, MPA_REQ_WAIT);
2042 start_ep_timer(ep);
2043 send_flowc(ep, skb);
2044 set_bit(PASS_ESTAB, &ep->com.history);
2045
2046 return 0;
2047 }
2048
2049 static int peer_close(struct c4iw_dev *dev, struct sk_buff *skb)
2050 {
2051 struct cpl_peer_close *hdr = cplhdr(skb);
2052 struct c4iw_ep *ep;
2053 struct c4iw_qp_attributes attrs;
2054 int disconnect = 1;
2055 int release = 0;
2056 struct tid_info *t = dev->rdev.lldi.tids;
2057 unsigned int tid = GET_TID(hdr);
2058 int ret;
2059
2060 ep = lookup_tid(t, tid);
2061 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
2062 dst_confirm(ep->dst);
2063
2064 set_bit(PEER_CLOSE, &ep->com.history);
2065 mutex_lock(&ep->com.mutex);
2066 switch (ep->com.state) {
2067 case MPA_REQ_WAIT:
2068 __state_set(&ep->com, CLOSING);
2069 break;
2070 case MPA_REQ_SENT:
2071 __state_set(&ep->com, CLOSING);
2072 connect_reply_upcall(ep, -ECONNRESET);
2073 break;
2074 case MPA_REQ_RCVD:
2075
2076 /*
2077 * We're gonna mark this puppy DEAD, but keep
2078 * the reference on it until the ULP accepts or
2079 * rejects the CR. Also wake up anyone waiting
2080 * in rdma connection migration (see c4iw_accept_cr()).
2081 */
2082 __state_set(&ep->com, CLOSING);
2083 PDBG("waking up ep %p tid %u\n", ep, ep->hwtid);
2084 c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
2085 break;
2086 case MPA_REP_SENT:
2087 __state_set(&ep->com, CLOSING);
2088 PDBG("waking up ep %p tid %u\n", ep, ep->hwtid);
2089 c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
2090 break;
2091 case FPDU_MODE:
2092 start_ep_timer(ep);
2093 __state_set(&ep->com, CLOSING);
2094 attrs.next_state = C4IW_QP_STATE_CLOSING;
2095 ret = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
2096 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
2097 if (ret != -ECONNRESET) {
2098 peer_close_upcall(ep);
2099 disconnect = 1;
2100 }
2101 break;
2102 case ABORTING:
2103 disconnect = 0;
2104 break;
2105 case CLOSING:
2106 __state_set(&ep->com, MORIBUND);
2107 disconnect = 0;
2108 break;
2109 case MORIBUND:
2110 stop_ep_timer(ep);
2111 if (ep->com.cm_id && ep->com.qp) {
2112 attrs.next_state = C4IW_QP_STATE_IDLE;
2113 c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
2114 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
2115 }
2116 close_complete_upcall(ep);
2117 __state_set(&ep->com, DEAD);
2118 release = 1;
2119 disconnect = 0;
2120 break;
2121 case DEAD:
2122 disconnect = 0;
2123 break;
2124 default:
2125 BUG_ON(1);
2126 }
2127 mutex_unlock(&ep->com.mutex);
2128 if (disconnect)
2129 c4iw_ep_disconnect(ep, 0, GFP_KERNEL);
2130 if (release)
2131 release_ep_resources(ep);
2132 return 0;
2133 }
2134
2135 /*
2136 * Returns whether an ABORT_REQ_RSS message is a negative advice.
2137 */
2138 static int is_neg_adv_abort(unsigned int status)
2139 {
2140 return status == CPL_ERR_RTX_NEG_ADVICE ||
2141 status == CPL_ERR_PERSIST_NEG_ADVICE;
2142 }
2143
2144 static int peer_abort(struct c4iw_dev *dev, struct sk_buff *skb)
2145 {
2146 struct cpl_abort_req_rss *req = cplhdr(skb);
2147 struct c4iw_ep *ep;
2148 struct cpl_abort_rpl *rpl;
2149 struct sk_buff *rpl_skb;
2150 struct c4iw_qp_attributes attrs;
2151 int ret;
2152 int release = 0;
2153 struct tid_info *t = dev->rdev.lldi.tids;
2154 unsigned int tid = GET_TID(req);
2155
2156 ep = lookup_tid(t, tid);
2157 if (is_neg_adv_abort(req->status)) {
2158 PDBG("%s neg_adv_abort ep %p tid %u\n", __func__, ep,
2159 ep->hwtid);
2160 return 0;
2161 }
2162 PDBG("%s ep %p tid %u state %u\n", __func__, ep, ep->hwtid,
2163 ep->com.state);
2164 set_bit(PEER_ABORT, &ep->com.history);
2165
2166 /*
2167 * Wake up any threads in rdma_init() or rdma_fini().
2168 * However, this is not needed if com state is just
2169 * MPA_REQ_SENT
2170 */
2171 if (ep->com.state != MPA_REQ_SENT)
2172 c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
2173
2174 mutex_lock(&ep->com.mutex);
2175 switch (ep->com.state) {
2176 case CONNECTING:
2177 break;
2178 case MPA_REQ_WAIT:
2179 stop_ep_timer(ep);
2180 break;
2181 case MPA_REQ_SENT:
2182 stop_ep_timer(ep);
2183 if (mpa_rev == 1 || (mpa_rev == 2 && ep->tried_with_mpa_v1))
2184 connect_reply_upcall(ep, -ECONNRESET);
2185 else {
2186 /*
2187 * we just don't send notification upwards because we
2188 * want to retry with mpa_v1 without upper layers even
2189 * knowing it.
2190 *
2191 * do some housekeeping so as to re-initiate the
2192 * connection
2193 */
2194 PDBG("%s: mpa_rev=%d. Retrying with mpav1\n", __func__,
2195 mpa_rev);
2196 ep->retry_with_mpa_v1 = 1;
2197 }
2198 break;
2199 case MPA_REP_SENT:
2200 break;
2201 case MPA_REQ_RCVD:
2202 break;
2203 case MORIBUND:
2204 case CLOSING:
2205 stop_ep_timer(ep);
2206 /*FALLTHROUGH*/
2207 case FPDU_MODE:
2208 if (ep->com.cm_id && ep->com.qp) {
2209 attrs.next_state = C4IW_QP_STATE_ERROR;
2210 ret = c4iw_modify_qp(ep->com.qp->rhp,
2211 ep->com.qp, C4IW_QP_ATTR_NEXT_STATE,
2212 &attrs, 1);
2213 if (ret)
2214 printk(KERN_ERR MOD
2215 "%s - qp <- error failed!\n",
2216 __func__);
2217 }
2218 peer_abort_upcall(ep);
2219 break;
2220 case ABORTING:
2221 break;
2222 case DEAD:
2223 PDBG("%s PEER_ABORT IN DEAD STATE!!!!\n", __func__);
2224 mutex_unlock(&ep->com.mutex);
2225 return 0;
2226 default:
2227 BUG_ON(1);
2228 break;
2229 }
2230 dst_confirm(ep->dst);
2231 if (ep->com.state != ABORTING) {
2232 __state_set(&ep->com, DEAD);
2233 /* we don't release if we want to retry with mpa_v1 */
2234 if (!ep->retry_with_mpa_v1)
2235 release = 1;
2236 }
2237 mutex_unlock(&ep->com.mutex);
2238
2239 rpl_skb = get_skb(skb, sizeof(*rpl), GFP_KERNEL);
2240 if (!rpl_skb) {
2241 printk(KERN_ERR MOD "%s - cannot allocate skb!\n",
2242 __func__);
2243 release = 1;
2244 goto out;
2245 }
2246 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
2247 rpl = (struct cpl_abort_rpl *) skb_put(rpl_skb, sizeof(*rpl));
2248 INIT_TP_WR(rpl, ep->hwtid);
2249 OPCODE_TID(rpl) = cpu_to_be32(MK_OPCODE_TID(CPL_ABORT_RPL, ep->hwtid));
2250 rpl->cmd = CPL_ABORT_NO_RST;
2251 c4iw_ofld_send(&ep->com.dev->rdev, rpl_skb);
2252 out:
2253 if (release)
2254 release_ep_resources(ep);
2255 else if (ep->retry_with_mpa_v1) {
2256 remove_handle(ep->com.dev, &ep->com.dev->hwtid_idr, ep->hwtid);
2257 cxgb4_remove_tid(ep->com.dev->rdev.lldi.tids, 0, ep->hwtid);
2258 dst_release(ep->dst);
2259 cxgb4_l2t_release(ep->l2t);
2260 c4iw_reconnect(ep);
2261 }
2262
2263 return 0;
2264 }
2265
2266 static int close_con_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
2267 {
2268 struct c4iw_ep *ep;
2269 struct c4iw_qp_attributes attrs;
2270 struct cpl_close_con_rpl *rpl = cplhdr(skb);
2271 int release = 0;
2272 struct tid_info *t = dev->rdev.lldi.tids;
2273 unsigned int tid = GET_TID(rpl);
2274
2275 ep = lookup_tid(t, tid);
2276
2277 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
2278 BUG_ON(!ep);
2279
2280 /* The cm_id may be null if we failed to connect */
2281 mutex_lock(&ep->com.mutex);
2282 switch (ep->com.state) {
2283 case CLOSING:
2284 __state_set(&ep->com, MORIBUND);
2285 break;
2286 case MORIBUND:
2287 stop_ep_timer(ep);
2288 if ((ep->com.cm_id) && (ep->com.qp)) {
2289 attrs.next_state = C4IW_QP_STATE_IDLE;
2290 c4iw_modify_qp(ep->com.qp->rhp,
2291 ep->com.qp,
2292 C4IW_QP_ATTR_NEXT_STATE,
2293 &attrs, 1);
2294 }
2295 close_complete_upcall(ep);
2296 __state_set(&ep->com, DEAD);
2297 release = 1;
2298 break;
2299 case ABORTING:
2300 case DEAD:
2301 break;
2302 default:
2303 BUG_ON(1);
2304 break;
2305 }
2306 mutex_unlock(&ep->com.mutex);
2307 if (release)
2308 release_ep_resources(ep);
2309 return 0;
2310 }
2311
2312 static int terminate(struct c4iw_dev *dev, struct sk_buff *skb)
2313 {
2314 struct cpl_rdma_terminate *rpl = cplhdr(skb);
2315 struct tid_info *t = dev->rdev.lldi.tids;
2316 unsigned int tid = GET_TID(rpl);
2317 struct c4iw_ep *ep;
2318 struct c4iw_qp_attributes attrs;
2319
2320 ep = lookup_tid(t, tid);
2321 BUG_ON(!ep);
2322
2323 if (ep && ep->com.qp) {
2324 printk(KERN_WARNING MOD "TERM received tid %u qpid %u\n", tid,
2325 ep->com.qp->wq.sq.qid);
2326 attrs.next_state = C4IW_QP_STATE_TERMINATE;
2327 c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
2328 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
2329 } else
2330 printk(KERN_WARNING MOD "TERM received tid %u no ep/qp\n", tid);
2331
2332 return 0;
2333 }
2334
2335 /*
2336 * Upcall from the adapter indicating data has been transmitted.
2337 * For us its just the single MPA request or reply. We can now free
2338 * the skb holding the mpa message.
2339 */
2340 static int fw4_ack(struct c4iw_dev *dev, struct sk_buff *skb)
2341 {
2342 struct c4iw_ep *ep;
2343 struct cpl_fw4_ack *hdr = cplhdr(skb);
2344 u8 credits = hdr->credits;
2345 unsigned int tid = GET_TID(hdr);
2346 struct tid_info *t = dev->rdev.lldi.tids;
2347
2348
2349 ep = lookup_tid(t, tid);
2350 PDBG("%s ep %p tid %u credits %u\n", __func__, ep, ep->hwtid, credits);
2351 if (credits == 0) {
2352 PDBG("%s 0 credit ack ep %p tid %u state %u\n",
2353 __func__, ep, ep->hwtid, state_read(&ep->com));
2354 return 0;
2355 }
2356
2357 dst_confirm(ep->dst);
2358 if (ep->mpa_skb) {
2359 PDBG("%s last streaming msg ack ep %p tid %u state %u "
2360 "initiator %u freeing skb\n", __func__, ep, ep->hwtid,
2361 state_read(&ep->com), ep->mpa_attr.initiator ? 1 : 0);
2362 kfree_skb(ep->mpa_skb);
2363 ep->mpa_skb = NULL;
2364 }
2365 return 0;
2366 }
2367
2368 int c4iw_reject_cr(struct iw_cm_id *cm_id, const void *pdata, u8 pdata_len)
2369 {
2370 int err;
2371 struct c4iw_ep *ep = to_ep(cm_id);
2372 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
2373
2374 if (state_read(&ep->com) == DEAD) {
2375 c4iw_put_ep(&ep->com);
2376 return -ECONNRESET;
2377 }
2378 set_bit(ULP_REJECT, &ep->com.history);
2379 BUG_ON(state_read(&ep->com) != MPA_REQ_RCVD);
2380 if (mpa_rev == 0)
2381 abort_connection(ep, NULL, GFP_KERNEL);
2382 else {
2383 err = send_mpa_reject(ep, pdata, pdata_len);
2384 err = c4iw_ep_disconnect(ep, 0, GFP_KERNEL);
2385 }
2386 c4iw_put_ep(&ep->com);
2387 return 0;
2388 }
2389
2390 int c4iw_accept_cr(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
2391 {
2392 int err;
2393 struct c4iw_qp_attributes attrs;
2394 enum c4iw_qp_attr_mask mask;
2395 struct c4iw_ep *ep = to_ep(cm_id);
2396 struct c4iw_dev *h = to_c4iw_dev(cm_id->device);
2397 struct c4iw_qp *qp = get_qhp(h, conn_param->qpn);
2398
2399 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
2400 if (state_read(&ep->com) == DEAD) {
2401 err = -ECONNRESET;
2402 goto err;
2403 }
2404
2405 BUG_ON(state_read(&ep->com) != MPA_REQ_RCVD);
2406 BUG_ON(!qp);
2407
2408 set_bit(ULP_ACCEPT, &ep->com.history);
2409 if ((conn_param->ord > c4iw_max_read_depth) ||
2410 (conn_param->ird > c4iw_max_read_depth)) {
2411 abort_connection(ep, NULL, GFP_KERNEL);
2412 err = -EINVAL;
2413 goto err;
2414 }
2415
2416 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) {
2417 if (conn_param->ord > ep->ird) {
2418 ep->ird = conn_param->ird;
2419 ep->ord = conn_param->ord;
2420 send_mpa_reject(ep, conn_param->private_data,
2421 conn_param->private_data_len);
2422 abort_connection(ep, NULL, GFP_KERNEL);
2423 err = -ENOMEM;
2424 goto err;
2425 }
2426 if (conn_param->ird > ep->ord) {
2427 if (!ep->ord)
2428 conn_param->ird = 1;
2429 else {
2430 abort_connection(ep, NULL, GFP_KERNEL);
2431 err = -ENOMEM;
2432 goto err;
2433 }
2434 }
2435
2436 }
2437 ep->ird = conn_param->ird;
2438 ep->ord = conn_param->ord;
2439
2440 if (ep->mpa_attr.version != 2)
2441 if (peer2peer && ep->ird == 0)
2442 ep->ird = 1;
2443
2444 PDBG("%s %d ird %d ord %d\n", __func__, __LINE__, ep->ird, ep->ord);
2445
2446 cm_id->add_ref(cm_id);
2447 ep->com.cm_id = cm_id;
2448 ep->com.qp = qp;
2449 ref_qp(ep);
2450
2451 /* bind QP to EP and move to RTS */
2452 attrs.mpa_attr = ep->mpa_attr;
2453 attrs.max_ird = ep->ird;
2454 attrs.max_ord = ep->ord;
2455 attrs.llp_stream_handle = ep;
2456 attrs.next_state = C4IW_QP_STATE_RTS;
2457
2458 /* bind QP and TID with INIT_WR */
2459 mask = C4IW_QP_ATTR_NEXT_STATE |
2460 C4IW_QP_ATTR_LLP_STREAM_HANDLE |
2461 C4IW_QP_ATTR_MPA_ATTR |
2462 C4IW_QP_ATTR_MAX_IRD |
2463 C4IW_QP_ATTR_MAX_ORD;
2464
2465 err = c4iw_modify_qp(ep->com.qp->rhp,
2466 ep->com.qp, mask, &attrs, 1);
2467 if (err)
2468 goto err1;
2469 err = send_mpa_reply(ep, conn_param->private_data,
2470 conn_param->private_data_len);
2471 if (err)
2472 goto err1;
2473
2474 state_set(&ep->com, FPDU_MODE);
2475 established_upcall(ep);
2476 c4iw_put_ep(&ep->com);
2477 return 0;
2478 err1:
2479 ep->com.cm_id = NULL;
2480 cm_id->rem_ref(cm_id);
2481 err:
2482 c4iw_put_ep(&ep->com);
2483 return err;
2484 }
2485
2486 int c4iw_connect(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
2487 {
2488 struct c4iw_dev *dev = to_c4iw_dev(cm_id->device);
2489 struct c4iw_ep *ep;
2490 struct rtable *rt;
2491 int err = 0;
2492
2493 if ((conn_param->ord > c4iw_max_read_depth) ||
2494 (conn_param->ird > c4iw_max_read_depth)) {
2495 err = -EINVAL;
2496 goto out;
2497 }
2498 ep = alloc_ep(sizeof(*ep), GFP_KERNEL);
2499 if (!ep) {
2500 printk(KERN_ERR MOD "%s - cannot alloc ep.\n", __func__);
2501 err = -ENOMEM;
2502 goto out;
2503 }
2504 init_timer(&ep->timer);
2505 ep->plen = conn_param->private_data_len;
2506 if (ep->plen)
2507 memcpy(ep->mpa_pkt + sizeof(struct mpa_message),
2508 conn_param->private_data, ep->plen);
2509 ep->ird = conn_param->ird;
2510 ep->ord = conn_param->ord;
2511
2512 if (peer2peer && ep->ord == 0)
2513 ep->ord = 1;
2514
2515 cm_id->add_ref(cm_id);
2516 ep->com.dev = dev;
2517 ep->com.cm_id = cm_id;
2518 ep->com.qp = get_qhp(dev, conn_param->qpn);
2519 BUG_ON(!ep->com.qp);
2520 ref_qp(ep);
2521 PDBG("%s qpn 0x%x qp %p cm_id %p\n", __func__, conn_param->qpn,
2522 ep->com.qp, cm_id);
2523
2524 /*
2525 * Allocate an active TID to initiate a TCP connection.
2526 */
2527 ep->atid = cxgb4_alloc_atid(dev->rdev.lldi.tids, ep);
2528 if (ep->atid == -1) {
2529 printk(KERN_ERR MOD "%s - cannot alloc atid.\n", __func__);
2530 err = -ENOMEM;
2531 goto fail2;
2532 }
2533 insert_handle(dev, &dev->atid_idr, ep, ep->atid);
2534
2535 PDBG("%s saddr 0x%x sport 0x%x raddr 0x%x rport 0x%x\n", __func__,
2536 ntohl(cm_id->local_addr.sin_addr.s_addr),
2537 ntohs(cm_id->local_addr.sin_port),
2538 ntohl(cm_id->remote_addr.sin_addr.s_addr),
2539 ntohs(cm_id->remote_addr.sin_port));
2540
2541 /* find a route */
2542 rt = find_route(dev,
2543 cm_id->local_addr.sin_addr.s_addr,
2544 cm_id->remote_addr.sin_addr.s_addr,
2545 cm_id->local_addr.sin_port,
2546 cm_id->remote_addr.sin_port, 0);
2547 if (!rt) {
2548 printk(KERN_ERR MOD "%s - cannot find route.\n", __func__);
2549 err = -EHOSTUNREACH;
2550 goto fail3;
2551 }
2552 ep->dst = &rt->dst;
2553
2554 err = import_ep(ep, cm_id->remote_addr.sin_addr.s_addr,
2555 ep->dst, ep->com.dev, true);
2556 if (err) {
2557 printk(KERN_ERR MOD "%s - cannot alloc l2e.\n", __func__);
2558 goto fail4;
2559 }
2560
2561 PDBG("%s txq_idx %u tx_chan %u smac_idx %u rss_qid %u l2t_idx %u\n",
2562 __func__, ep->txq_idx, ep->tx_chan, ep->smac_idx, ep->rss_qid,
2563 ep->l2t->idx);
2564
2565 state_set(&ep->com, CONNECTING);
2566 ep->tos = 0;
2567 ep->com.local_addr = cm_id->local_addr;
2568 ep->com.remote_addr = cm_id->remote_addr;
2569
2570 /* send connect request to rnic */
2571 err = send_connect(ep);
2572 if (!err)
2573 goto out;
2574
2575 cxgb4_l2t_release(ep->l2t);
2576 fail4:
2577 dst_release(ep->dst);
2578 fail3:
2579 remove_handle(ep->com.dev, &ep->com.dev->atid_idr, ep->atid);
2580 cxgb4_free_atid(ep->com.dev->rdev.lldi.tids, ep->atid);
2581 fail2:
2582 cm_id->rem_ref(cm_id);
2583 c4iw_put_ep(&ep->com);
2584 out:
2585 return err;
2586 }
2587
2588 int c4iw_create_listen(struct iw_cm_id *cm_id, int backlog)
2589 {
2590 int err = 0;
2591 struct c4iw_dev *dev = to_c4iw_dev(cm_id->device);
2592 struct c4iw_listen_ep *ep;
2593
2594 might_sleep();
2595
2596 ep = alloc_ep(sizeof(*ep), GFP_KERNEL);
2597 if (!ep) {
2598 printk(KERN_ERR MOD "%s - cannot alloc ep.\n", __func__);
2599 err = -ENOMEM;
2600 goto fail1;
2601 }
2602 PDBG("%s ep %p\n", __func__, ep);
2603 cm_id->add_ref(cm_id);
2604 ep->com.cm_id = cm_id;
2605 ep->com.dev = dev;
2606 ep->backlog = backlog;
2607 ep->com.local_addr = cm_id->local_addr;
2608
2609 /*
2610 * Allocate a server TID.
2611 */
2612 if (dev->rdev.lldi.enable_fw_ofld_conn)
2613 ep->stid = cxgb4_alloc_sftid(dev->rdev.lldi.tids, PF_INET, ep);
2614 else
2615 ep->stid = cxgb4_alloc_stid(dev->rdev.lldi.tids, PF_INET, ep);
2616
2617 if (ep->stid == -1) {
2618 printk(KERN_ERR MOD "%s - cannot alloc stid.\n", __func__);
2619 err = -ENOMEM;
2620 goto fail2;
2621 }
2622 insert_handle(dev, &dev->stid_idr, ep, ep->stid);
2623 state_set(&ep->com, LISTEN);
2624 if (dev->rdev.lldi.enable_fw_ofld_conn) {
2625 do {
2626 err = cxgb4_create_server_filter(
2627 ep->com.dev->rdev.lldi.ports[0], ep->stid,
2628 ep->com.local_addr.sin_addr.s_addr,
2629 ep->com.local_addr.sin_port,
2630 0,
2631 ep->com.dev->rdev.lldi.rxq_ids[0],
2632 0,
2633 0);
2634 if (err == -EBUSY) {
2635 set_current_state(TASK_UNINTERRUPTIBLE);
2636 schedule_timeout(usecs_to_jiffies(100));
2637 }
2638 } while (err == -EBUSY);
2639 } else {
2640 c4iw_init_wr_wait(&ep->com.wr_wait);
2641 err = cxgb4_create_server(ep->com.dev->rdev.lldi.ports[0],
2642 ep->stid, ep->com.local_addr.sin_addr.s_addr,
2643 ep->com.local_addr.sin_port,
2644 0,
2645 ep->com.dev->rdev.lldi.rxq_ids[0]);
2646 if (!err)
2647 err = c4iw_wait_for_reply(&ep->com.dev->rdev,
2648 &ep->com.wr_wait,
2649 0, 0, __func__);
2650 }
2651 if (!err) {
2652 cm_id->provider_data = ep;
2653 goto out;
2654 }
2655 pr_err("%s cxgb4_create_server/filter failed err %d " \
2656 "stid %d laddr %08x lport %d\n", \
2657 __func__, err, ep->stid,
2658 ntohl(ep->com.local_addr.sin_addr.s_addr),
2659 ntohs(ep->com.local_addr.sin_port));
2660 cxgb4_free_stid(ep->com.dev->rdev.lldi.tids, ep->stid, PF_INET);
2661 fail2:
2662 cm_id->rem_ref(cm_id);
2663 c4iw_put_ep(&ep->com);
2664 fail1:
2665 out:
2666 return err;
2667 }
2668
2669 int c4iw_destroy_listen(struct iw_cm_id *cm_id)
2670 {
2671 int err;
2672 struct c4iw_listen_ep *ep = to_listen_ep(cm_id);
2673
2674 PDBG("%s ep %p\n", __func__, ep);
2675
2676 might_sleep();
2677 state_set(&ep->com, DEAD);
2678 if (ep->com.dev->rdev.lldi.enable_fw_ofld_conn) {
2679 err = cxgb4_remove_server_filter(
2680 ep->com.dev->rdev.lldi.ports[0], ep->stid,
2681 ep->com.dev->rdev.lldi.rxq_ids[0], 0);
2682 } else {
2683 c4iw_init_wr_wait(&ep->com.wr_wait);
2684 err = listen_stop(ep);
2685 if (err)
2686 goto done;
2687 err = c4iw_wait_for_reply(&ep->com.dev->rdev, &ep->com.wr_wait,
2688 0, 0, __func__);
2689 }
2690 remove_handle(ep->com.dev, &ep->com.dev->stid_idr, ep->stid);
2691 cxgb4_free_stid(ep->com.dev->rdev.lldi.tids, ep->stid, PF_INET);
2692 done:
2693 cm_id->rem_ref(cm_id);
2694 c4iw_put_ep(&ep->com);
2695 return err;
2696 }
2697
2698 int c4iw_ep_disconnect(struct c4iw_ep *ep, int abrupt, gfp_t gfp)
2699 {
2700 int ret = 0;
2701 int close = 0;
2702 int fatal = 0;
2703 struct c4iw_rdev *rdev;
2704
2705 mutex_lock(&ep->com.mutex);
2706
2707 PDBG("%s ep %p state %s, abrupt %d\n", __func__, ep,
2708 states[ep->com.state], abrupt);
2709
2710 rdev = &ep->com.dev->rdev;
2711 if (c4iw_fatal_error(rdev)) {
2712 fatal = 1;
2713 close_complete_upcall(ep);
2714 ep->com.state = DEAD;
2715 }
2716 switch (ep->com.state) {
2717 case MPA_REQ_WAIT:
2718 case MPA_REQ_SENT:
2719 case MPA_REQ_RCVD:
2720 case MPA_REP_SENT:
2721 case FPDU_MODE:
2722 close = 1;
2723 if (abrupt)
2724 ep->com.state = ABORTING;
2725 else {
2726 ep->com.state = CLOSING;
2727 start_ep_timer(ep);
2728 }
2729 set_bit(CLOSE_SENT, &ep->com.flags);
2730 break;
2731 case CLOSING:
2732 if (!test_and_set_bit(CLOSE_SENT, &ep->com.flags)) {
2733 close = 1;
2734 if (abrupt) {
2735 stop_ep_timer(ep);
2736 ep->com.state = ABORTING;
2737 } else
2738 ep->com.state = MORIBUND;
2739 }
2740 break;
2741 case MORIBUND:
2742 case ABORTING:
2743 case DEAD:
2744 PDBG("%s ignoring disconnect ep %p state %u\n",
2745 __func__, ep, ep->com.state);
2746 break;
2747 default:
2748 BUG();
2749 break;
2750 }
2751
2752 if (close) {
2753 if (abrupt) {
2754 set_bit(EP_DISC_ABORT, &ep->com.history);
2755 close_complete_upcall(ep);
2756 ret = send_abort(ep, NULL, gfp);
2757 } else {
2758 set_bit(EP_DISC_CLOSE, &ep->com.history);
2759 ret = send_halfclose(ep, gfp);
2760 }
2761 if (ret)
2762 fatal = 1;
2763 }
2764 mutex_unlock(&ep->com.mutex);
2765 if (fatal)
2766 release_ep_resources(ep);
2767 return ret;
2768 }
2769
2770 static void active_ofld_conn_reply(struct c4iw_dev *dev, struct sk_buff *skb,
2771 struct cpl_fw6_msg_ofld_connection_wr_rpl *req)
2772 {
2773 struct c4iw_ep *ep;
2774 int atid = be32_to_cpu(req->tid);
2775
2776 ep = (struct c4iw_ep *)lookup_atid(dev->rdev.lldi.tids,
2777 (__force u32) req->tid);
2778 if (!ep)
2779 return;
2780
2781 switch (req->retval) {
2782 case FW_ENOMEM:
2783 set_bit(ACT_RETRY_NOMEM, &ep->com.history);
2784 if (ep->retry_count++ < ACT_OPEN_RETRY_COUNT) {
2785 send_fw_act_open_req(ep, atid);
2786 return;
2787 }
2788 case FW_EADDRINUSE:
2789 set_bit(ACT_RETRY_INUSE, &ep->com.history);
2790 if (ep->retry_count++ < ACT_OPEN_RETRY_COUNT) {
2791 send_fw_act_open_req(ep, atid);
2792 return;
2793 }
2794 break;
2795 default:
2796 pr_info("%s unexpected ofld conn wr retval %d\n",
2797 __func__, req->retval);
2798 break;
2799 }
2800 pr_err("active ofld_connect_wr failure %d atid %d\n",
2801 req->retval, atid);
2802 mutex_lock(&dev->rdev.stats.lock);
2803 dev->rdev.stats.act_ofld_conn_fails++;
2804 mutex_unlock(&dev->rdev.stats.lock);
2805 connect_reply_upcall(ep, status2errno(req->retval));
2806 state_set(&ep->com, DEAD);
2807 remove_handle(dev, &dev->atid_idr, atid);
2808 cxgb4_free_atid(dev->rdev.lldi.tids, atid);
2809 dst_release(ep->dst);
2810 cxgb4_l2t_release(ep->l2t);
2811 c4iw_put_ep(&ep->com);
2812 }
2813
2814 static void passive_ofld_conn_reply(struct c4iw_dev *dev, struct sk_buff *skb,
2815 struct cpl_fw6_msg_ofld_connection_wr_rpl *req)
2816 {
2817 struct sk_buff *rpl_skb;
2818 struct cpl_pass_accept_req *cpl;
2819 int ret;
2820
2821 rpl_skb = (struct sk_buff *)(unsigned long)req->cookie;
2822 BUG_ON(!rpl_skb);
2823 if (req->retval) {
2824 PDBG("%s passive open failure %d\n", __func__, req->retval);
2825 mutex_lock(&dev->rdev.stats.lock);
2826 dev->rdev.stats.pas_ofld_conn_fails++;
2827 mutex_unlock(&dev->rdev.stats.lock);
2828 kfree_skb(rpl_skb);
2829 } else {
2830 cpl = (struct cpl_pass_accept_req *)cplhdr(rpl_skb);
2831 OPCODE_TID(cpl) = htonl(MK_OPCODE_TID(CPL_PASS_ACCEPT_REQ,
2832 (__force u32) htonl(
2833 (__force u32) req->tid)));
2834 ret = pass_accept_req(dev, rpl_skb);
2835 if (!ret)
2836 kfree_skb(rpl_skb);
2837 }
2838 return;
2839 }
2840
2841 static int deferred_fw6_msg(struct c4iw_dev *dev, struct sk_buff *skb)
2842 {
2843 struct cpl_fw6_msg *rpl = cplhdr(skb);
2844 struct cpl_fw6_msg_ofld_connection_wr_rpl *req;
2845
2846 switch (rpl->type) {
2847 case FW6_TYPE_CQE:
2848 c4iw_ev_dispatch(dev, (struct t4_cqe *)&rpl->data[0]);
2849 break;
2850 case FW6_TYPE_OFLD_CONNECTION_WR_RPL:
2851 req = (struct cpl_fw6_msg_ofld_connection_wr_rpl *)rpl->data;
2852 switch (req->t_state) {
2853 case TCP_SYN_SENT:
2854 active_ofld_conn_reply(dev, skb, req);
2855 break;
2856 case TCP_SYN_RECV:
2857 passive_ofld_conn_reply(dev, skb, req);
2858 break;
2859 default:
2860 pr_err("%s unexpected ofld conn wr state %d\n",
2861 __func__, req->t_state);
2862 break;
2863 }
2864 break;
2865 }
2866 return 0;
2867 }
2868
2869 static void build_cpl_pass_accept_req(struct sk_buff *skb, int stid , u8 tos)
2870 {
2871 u32 l2info;
2872 u16 vlantag, len, hdr_len;
2873 u8 intf;
2874 struct cpl_rx_pkt *cpl = cplhdr(skb);
2875 struct cpl_pass_accept_req *req;
2876 struct tcp_options_received tmp_opt;
2877
2878 /* Store values from cpl_rx_pkt in temporary location. */
2879 vlantag = (__force u16) cpl->vlan;
2880 len = (__force u16) cpl->len;
2881 l2info = (__force u32) cpl->l2info;
2882 hdr_len = (__force u16) cpl->hdr_len;
2883 intf = cpl->iff;
2884
2885 __skb_pull(skb, sizeof(*req) + sizeof(struct rss_header));
2886
2887 /*
2888 * We need to parse the TCP options from SYN packet.
2889 * to generate cpl_pass_accept_req.
2890 */
2891 memset(&tmp_opt, 0, sizeof(tmp_opt));
2892 tcp_clear_options(&tmp_opt);
2893 tcp_parse_options(skb, &tmp_opt, NULL, 0, NULL);
2894
2895 req = (struct cpl_pass_accept_req *)__skb_push(skb, sizeof(*req));
2896 memset(req, 0, sizeof(*req));
2897 req->l2info = cpu_to_be16(V_SYN_INTF(intf) |
2898 V_SYN_MAC_IDX(G_RX_MACIDX(
2899 (__force int) htonl(l2info))) |
2900 F_SYN_XACT_MATCH);
2901 req->hdr_len = cpu_to_be32(V_SYN_RX_CHAN(G_RX_CHAN(
2902 (__force int) htonl(l2info))) |
2903 V_TCP_HDR_LEN(G_RX_TCPHDR_LEN(
2904 (__force int) htons(hdr_len))) |
2905 V_IP_HDR_LEN(G_RX_IPHDR_LEN(
2906 (__force int) htons(hdr_len))) |
2907 V_ETH_HDR_LEN(G_RX_ETHHDR_LEN(
2908 (__force int) htonl(l2info))));
2909 req->vlan = (__force __be16) vlantag;
2910 req->len = (__force __be16) len;
2911 req->tos_stid = cpu_to_be32(PASS_OPEN_TID(stid) |
2912 PASS_OPEN_TOS(tos));
2913 req->tcpopt.mss = htons(tmp_opt.mss_clamp);
2914 if (tmp_opt.wscale_ok)
2915 req->tcpopt.wsf = tmp_opt.snd_wscale;
2916 req->tcpopt.tstamp = tmp_opt.saw_tstamp;
2917 if (tmp_opt.sack_ok)
2918 req->tcpopt.sack = 1;
2919 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_PASS_ACCEPT_REQ, 0));
2920 return;
2921 }
2922
2923 static void send_fw_pass_open_req(struct c4iw_dev *dev, struct sk_buff *skb,
2924 __be32 laddr, __be16 lport,
2925 __be32 raddr, __be16 rport,
2926 u32 rcv_isn, u32 filter, u16 window,
2927 u32 rss_qid, u8 port_id)
2928 {
2929 struct sk_buff *req_skb;
2930 struct fw_ofld_connection_wr *req;
2931 struct cpl_pass_accept_req *cpl = cplhdr(skb);
2932
2933 req_skb = alloc_skb(sizeof(struct fw_ofld_connection_wr), GFP_KERNEL);
2934 req = (struct fw_ofld_connection_wr *)__skb_put(req_skb, sizeof(*req));
2935 memset(req, 0, sizeof(*req));
2936 req->op_compl = htonl(V_WR_OP(FW_OFLD_CONNECTION_WR) | FW_WR_COMPL(1));
2937 req->len16_pkd = htonl(FW_WR_LEN16(DIV_ROUND_UP(sizeof(*req), 16)));
2938 req->le.version_cpl = htonl(F_FW_OFLD_CONNECTION_WR_CPL);
2939 req->le.filter = (__force __be32) filter;
2940 req->le.lport = lport;
2941 req->le.pport = rport;
2942 req->le.u.ipv4.lip = laddr;
2943 req->le.u.ipv4.pip = raddr;
2944 req->tcb.rcv_nxt = htonl(rcv_isn + 1);
2945 req->tcb.rcv_adv = htons(window);
2946 req->tcb.t_state_to_astid =
2947 htonl(V_FW_OFLD_CONNECTION_WR_T_STATE(TCP_SYN_RECV) |
2948 V_FW_OFLD_CONNECTION_WR_RCV_SCALE(cpl->tcpopt.wsf) |
2949 V_FW_OFLD_CONNECTION_WR_ASTID(
2950 GET_PASS_OPEN_TID(ntohl(cpl->tos_stid))));
2951
2952 /*
2953 * We store the qid in opt2 which will be used by the firmware
2954 * to send us the wr response.
2955 */
2956 req->tcb.opt2 = htonl(V_RSS_QUEUE(rss_qid));
2957
2958 /*
2959 * We initialize the MSS index in TCB to 0xF.
2960 * So that when driver sends cpl_pass_accept_rpl
2961 * TCB picks up the correct value. If this was 0
2962 * TP will ignore any value > 0 for MSS index.
2963 */
2964 req->tcb.opt0 = cpu_to_be64(V_MSS_IDX(0xF));
2965 req->cookie = (unsigned long)skb;
2966
2967 set_wr_txq(req_skb, CPL_PRIORITY_CONTROL, port_id);
2968 cxgb4_ofld_send(dev->rdev.lldi.ports[0], req_skb);
2969 }
2970
2971 /*
2972 * Handler for CPL_RX_PKT message. Need to handle cpl_rx_pkt
2973 * messages when a filter is being used instead of server to
2974 * redirect a syn packet. When packets hit filter they are redirected
2975 * to the offload queue and driver tries to establish the connection
2976 * using firmware work request.
2977 */
2978 static int rx_pkt(struct c4iw_dev *dev, struct sk_buff *skb)
2979 {
2980 int stid;
2981 unsigned int filter;
2982 struct ethhdr *eh = NULL;
2983 struct vlan_ethhdr *vlan_eh = NULL;
2984 struct iphdr *iph;
2985 struct tcphdr *tcph;
2986 struct rss_header *rss = (void *)skb->data;
2987 struct cpl_rx_pkt *cpl = (void *)skb->data;
2988 struct cpl_pass_accept_req *req = (void *)(rss + 1);
2989 struct l2t_entry *e;
2990 struct dst_entry *dst;
2991 struct rtable *rt;
2992 struct c4iw_ep *lep;
2993 u16 window;
2994 struct port_info *pi;
2995 struct net_device *pdev;
2996 u16 rss_qid;
2997 int step;
2998 u32 tx_chan;
2999 struct neighbour *neigh;
3000
3001 /* Drop all non-SYN packets */
3002 if (!(cpl->l2info & cpu_to_be32(F_RXF_SYN)))
3003 goto reject;
3004
3005 /*
3006 * Drop all packets which did not hit the filter.
3007 * Unlikely to happen.
3008 */
3009 if (!(rss->filter_hit && rss->filter_tid))
3010 goto reject;
3011
3012 /*
3013 * Calculate the server tid from filter hit index from cpl_rx_pkt.
3014 */
3015 stid = (__force int) cpu_to_be32((__force u32) rss->hash_val)
3016 - dev->rdev.lldi.tids->sftid_base
3017 + dev->rdev.lldi.tids->nstids;
3018
3019 lep = (struct c4iw_ep *)lookup_stid(dev->rdev.lldi.tids, stid);
3020 if (!lep) {
3021 PDBG("%s connect request on invalid stid %d\n", __func__, stid);
3022 goto reject;
3023 }
3024
3025 if (G_RX_ETHHDR_LEN(ntohl(cpl->l2info)) == ETH_HLEN) {
3026 eh = (struct ethhdr *)(req + 1);
3027 iph = (struct iphdr *)(eh + 1);
3028 } else {
3029 vlan_eh = (struct vlan_ethhdr *)(req + 1);
3030 iph = (struct iphdr *)(vlan_eh + 1);
3031 skb->vlan_tci = ntohs(cpl->vlan);
3032 }
3033
3034 if (iph->version != 0x4)
3035 goto reject;
3036
3037 tcph = (struct tcphdr *)(iph + 1);
3038 skb_set_network_header(skb, (void *)iph - (void *)rss);
3039 skb_set_transport_header(skb, (void *)tcph - (void *)rss);
3040 skb_get(skb);
3041
3042 PDBG("%s lip 0x%x lport %u pip 0x%x pport %u tos %d\n", __func__,
3043 ntohl(iph->daddr), ntohs(tcph->dest), ntohl(iph->saddr),
3044 ntohs(tcph->source), iph->tos);
3045
3046 rt = find_route(dev, iph->daddr, iph->saddr, tcph->dest, tcph->source,
3047 iph->tos);
3048 if (!rt) {
3049 pr_err("%s - failed to find dst entry!\n",
3050 __func__);
3051 goto reject;
3052 }
3053 dst = &rt->dst;
3054 neigh = dst_neigh_lookup_skb(dst, skb);
3055
3056 if (neigh->dev->flags & IFF_LOOPBACK) {
3057 pdev = ip_dev_find(&init_net, iph->daddr);
3058 e = cxgb4_l2t_get(dev->rdev.lldi.l2t, neigh,
3059 pdev, 0);
3060 pi = (struct port_info *)netdev_priv(pdev);
3061 tx_chan = cxgb4_port_chan(pdev);
3062 dev_put(pdev);
3063 } else {
3064 e = cxgb4_l2t_get(dev->rdev.lldi.l2t, neigh,
3065 neigh->dev, 0);
3066 pi = (struct port_info *)netdev_priv(neigh->dev);
3067 tx_chan = cxgb4_port_chan(neigh->dev);
3068 }
3069 if (!e) {
3070 pr_err("%s - failed to allocate l2t entry!\n",
3071 __func__);
3072 goto free_dst;
3073 }
3074
3075 step = dev->rdev.lldi.nrxq / dev->rdev.lldi.nchan;
3076 rss_qid = dev->rdev.lldi.rxq_ids[pi->port_id * step];
3077 window = (__force u16) htons((__force u16)tcph->window);
3078
3079 /* Calcuate filter portion for LE region. */
3080 filter = (__force unsigned int) cpu_to_be32(select_ntuple(dev, dst, e));
3081
3082 /*
3083 * Synthesize the cpl_pass_accept_req. We have everything except the
3084 * TID. Once firmware sends a reply with TID we update the TID field
3085 * in cpl and pass it through the regular cpl_pass_accept_req path.
3086 */
3087 build_cpl_pass_accept_req(skb, stid, iph->tos);
3088 send_fw_pass_open_req(dev, skb, iph->daddr, tcph->dest, iph->saddr,
3089 tcph->source, ntohl(tcph->seq), filter, window,
3090 rss_qid, pi->port_id);
3091 cxgb4_l2t_release(e);
3092 free_dst:
3093 dst_release(dst);
3094 reject:
3095 return 0;
3096 }
3097
3098 /*
3099 * These are the real handlers that are called from a
3100 * work queue.
3101 */
3102 static c4iw_handler_func work_handlers[NUM_CPL_CMDS] = {
3103 [CPL_ACT_ESTABLISH] = act_establish,
3104 [CPL_ACT_OPEN_RPL] = act_open_rpl,
3105 [CPL_RX_DATA] = rx_data,
3106 [CPL_ABORT_RPL_RSS] = abort_rpl,
3107 [CPL_ABORT_RPL] = abort_rpl,
3108 [CPL_PASS_OPEN_RPL] = pass_open_rpl,
3109 [CPL_CLOSE_LISTSRV_RPL] = close_listsrv_rpl,
3110 [CPL_PASS_ACCEPT_REQ] = pass_accept_req,
3111 [CPL_PASS_ESTABLISH] = pass_establish,
3112 [CPL_PEER_CLOSE] = peer_close,
3113 [CPL_ABORT_REQ_RSS] = peer_abort,
3114 [CPL_CLOSE_CON_RPL] = close_con_rpl,
3115 [CPL_RDMA_TERMINATE] = terminate,
3116 [CPL_FW4_ACK] = fw4_ack,
3117 [CPL_FW6_MSG] = deferred_fw6_msg,
3118 [CPL_RX_PKT] = rx_pkt
3119 };
3120
3121 static void process_timeout(struct c4iw_ep *ep)
3122 {
3123 struct c4iw_qp_attributes attrs;
3124 int abort = 1;
3125
3126 mutex_lock(&ep->com.mutex);
3127 PDBG("%s ep %p tid %u state %d\n", __func__, ep, ep->hwtid,
3128 ep->com.state);
3129 set_bit(TIMEDOUT, &ep->com.history);
3130 switch (ep->com.state) {
3131 case MPA_REQ_SENT:
3132 __state_set(&ep->com, ABORTING);
3133 connect_reply_upcall(ep, -ETIMEDOUT);
3134 break;
3135 case MPA_REQ_WAIT:
3136 __state_set(&ep->com, ABORTING);
3137 break;
3138 case CLOSING:
3139 case MORIBUND:
3140 if (ep->com.cm_id && ep->com.qp) {
3141 attrs.next_state = C4IW_QP_STATE_ERROR;
3142 c4iw_modify_qp(ep->com.qp->rhp,
3143 ep->com.qp, C4IW_QP_ATTR_NEXT_STATE,
3144 &attrs, 1);
3145 }
3146 __state_set(&ep->com, ABORTING);
3147 break;
3148 default:
3149 WARN(1, "%s unexpected state ep %p tid %u state %u\n",
3150 __func__, ep, ep->hwtid, ep->com.state);
3151 abort = 0;
3152 }
3153 mutex_unlock(&ep->com.mutex);
3154 if (abort)
3155 abort_connection(ep, NULL, GFP_KERNEL);
3156 c4iw_put_ep(&ep->com);
3157 }
3158
3159 static void process_timedout_eps(void)
3160 {
3161 struct c4iw_ep *ep;
3162
3163 spin_lock_irq(&timeout_lock);
3164 while (!list_empty(&timeout_list)) {
3165 struct list_head *tmp;
3166
3167 tmp = timeout_list.next;
3168 list_del(tmp);
3169 spin_unlock_irq(&timeout_lock);
3170 ep = list_entry(tmp, struct c4iw_ep, entry);
3171 process_timeout(ep);
3172 spin_lock_irq(&timeout_lock);
3173 }
3174 spin_unlock_irq(&timeout_lock);
3175 }
3176
3177 static void process_work(struct work_struct *work)
3178 {
3179 struct sk_buff *skb = NULL;
3180 struct c4iw_dev *dev;
3181 struct cpl_act_establish *rpl;
3182 unsigned int opcode;
3183 int ret;
3184
3185 while ((skb = skb_dequeue(&rxq))) {
3186 rpl = cplhdr(skb);
3187 dev = *((struct c4iw_dev **) (skb->cb + sizeof(void *)));
3188 opcode = rpl->ot.opcode;
3189
3190 BUG_ON(!work_handlers[opcode]);
3191 ret = work_handlers[opcode](dev, skb);
3192 if (!ret)
3193 kfree_skb(skb);
3194 }
3195 process_timedout_eps();
3196 }
3197
3198 static DECLARE_WORK(skb_work, process_work);
3199
3200 static void ep_timeout(unsigned long arg)
3201 {
3202 struct c4iw_ep *ep = (struct c4iw_ep *)arg;
3203 int kickit = 0;
3204
3205 spin_lock(&timeout_lock);
3206 if (!test_and_set_bit(TIMEOUT, &ep->com.flags)) {
3207 list_add_tail(&ep->entry, &timeout_list);
3208 kickit = 1;
3209 }
3210 spin_unlock(&timeout_lock);
3211 if (kickit)
3212 queue_work(workq, &skb_work);
3213 }
3214
3215 /*
3216 * All the CM events are handled on a work queue to have a safe context.
3217 */
3218 static int sched(struct c4iw_dev *dev, struct sk_buff *skb)
3219 {
3220
3221 /*
3222 * Save dev in the skb->cb area.
3223 */
3224 *((struct c4iw_dev **) (skb->cb + sizeof(void *))) = dev;
3225
3226 /*
3227 * Queue the skb and schedule the worker thread.
3228 */
3229 skb_queue_tail(&rxq, skb);
3230 queue_work(workq, &skb_work);
3231 return 0;
3232 }
3233
3234 static int set_tcb_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
3235 {
3236 struct cpl_set_tcb_rpl *rpl = cplhdr(skb);
3237
3238 if (rpl->status != CPL_ERR_NONE) {
3239 printk(KERN_ERR MOD "Unexpected SET_TCB_RPL status %u "
3240 "for tid %u\n", rpl->status, GET_TID(rpl));
3241 }
3242 kfree_skb(skb);
3243 return 0;
3244 }
3245
3246 static int fw6_msg(struct c4iw_dev *dev, struct sk_buff *skb)
3247 {
3248 struct cpl_fw6_msg *rpl = cplhdr(skb);
3249 struct c4iw_wr_wait *wr_waitp;
3250 int ret;
3251
3252 PDBG("%s type %u\n", __func__, rpl->type);
3253
3254 switch (rpl->type) {
3255 case FW6_TYPE_WR_RPL:
3256 ret = (int)((be64_to_cpu(rpl->data[0]) >> 8) & 0xff);
3257 wr_waitp = (struct c4iw_wr_wait *)(__force unsigned long) rpl->data[1];
3258 PDBG("%s wr_waitp %p ret %u\n", __func__, wr_waitp, ret);
3259 if (wr_waitp)
3260 c4iw_wake_up(wr_waitp, ret ? -ret : 0);
3261 kfree_skb(skb);
3262 break;
3263 case FW6_TYPE_CQE:
3264 case FW6_TYPE_OFLD_CONNECTION_WR_RPL:
3265 sched(dev, skb);
3266 break;
3267 default:
3268 printk(KERN_ERR MOD "%s unexpected fw6 msg type %u\n", __func__,
3269 rpl->type);
3270 kfree_skb(skb);
3271 break;
3272 }
3273 return 0;
3274 }
3275
3276 static int peer_abort_intr(struct c4iw_dev *dev, struct sk_buff *skb)
3277 {
3278 struct cpl_abort_req_rss *req = cplhdr(skb);
3279 struct c4iw_ep *ep;
3280 struct tid_info *t = dev->rdev.lldi.tids;
3281 unsigned int tid = GET_TID(req);
3282
3283 ep = lookup_tid(t, tid);
3284 if (!ep) {
3285 printk(KERN_WARNING MOD
3286 "Abort on non-existent endpoint, tid %d\n", tid);
3287 kfree_skb(skb);
3288 return 0;
3289 }
3290 if (is_neg_adv_abort(req->status)) {
3291 PDBG("%s neg_adv_abort ep %p tid %u\n", __func__, ep,
3292 ep->hwtid);
3293 kfree_skb(skb);
3294 return 0;
3295 }
3296 PDBG("%s ep %p tid %u state %u\n", __func__, ep, ep->hwtid,
3297 ep->com.state);
3298
3299 /*
3300 * Wake up any threads in rdma_init() or rdma_fini().
3301 * However, if we are on MPAv2 and want to retry with MPAv1
3302 * then, don't wake up yet.
3303 */
3304 if (mpa_rev == 2 && !ep->tried_with_mpa_v1) {
3305 if (ep->com.state != MPA_REQ_SENT)
3306 c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
3307 } else
3308 c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
3309 sched(dev, skb);
3310 return 0;
3311 }
3312
3313 /*
3314 * Most upcalls from the T4 Core go to sched() to
3315 * schedule the processing on a work queue.
3316 */
3317 c4iw_handler_func c4iw_handlers[NUM_CPL_CMDS] = {
3318 [CPL_ACT_ESTABLISH] = sched,
3319 [CPL_ACT_OPEN_RPL] = sched,
3320 [CPL_RX_DATA] = sched,
3321 [CPL_ABORT_RPL_RSS] = sched,
3322 [CPL_ABORT_RPL] = sched,
3323 [CPL_PASS_OPEN_RPL] = sched,
3324 [CPL_CLOSE_LISTSRV_RPL] = sched,
3325 [CPL_PASS_ACCEPT_REQ] = sched,
3326 [CPL_PASS_ESTABLISH] = sched,
3327 [CPL_PEER_CLOSE] = sched,
3328 [CPL_CLOSE_CON_RPL] = sched,
3329 [CPL_ABORT_REQ_RSS] = peer_abort_intr,
3330 [CPL_RDMA_TERMINATE] = sched,
3331 [CPL_FW4_ACK] = sched,
3332 [CPL_SET_TCB_RPL] = set_tcb_rpl,
3333 [CPL_FW6_MSG] = fw6_msg,
3334 [CPL_RX_PKT] = sched
3335 };
3336
3337 int __init c4iw_cm_init(void)
3338 {
3339 spin_lock_init(&timeout_lock);
3340 skb_queue_head_init(&rxq);
3341
3342 workq = create_singlethread_workqueue("iw_cxgb4");
3343 if (!workq)
3344 return -ENOMEM;
3345
3346 return 0;
3347 }
3348
3349 void __exit c4iw_cm_term(void)
3350 {
3351 WARN_ON(!list_empty(&timeout_list));
3352 flush_workqueue(workq);
3353 destroy_workqueue(workq);
3354 }
Ubuntu Hirsute Linux kernel mirror