net-next: remove useless union keyword
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / drivers / infiniband / hw / cxgb4 / cm.c
blob8b693c8c25e2971d0dd663a1c1755da7494d752a
1 /*
2 * Copyright (c) 2009-2010 Chelsio, Inc. All rights reserved.
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:
10 * Redistribution and use in source and binary forms, with or
11 * without modification, are permitted provided that the following
12 * conditions are met:
14 * - Redistributions of source code must retain the above
15 * copyright notice, this list of conditions and the following
16 * disclaimer.
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.
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.
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>
42 #include <net/neighbour.h>
43 #include <net/netevent.h>
44 #include <net/route.h>
46 #include "iw_cxgb4.h"
48 static char *states[] = {
49 "idle",
50 "listen",
51 "connecting",
52 "mpa_wait_req",
53 "mpa_req_sent",
54 "mpa_req_rcvd",
55 "mpa_rep_sent",
56 "fpdu_mode",
57 "aborting",
58 "closing",
59 "moribund",
60 "dead",
61 NULL,
64 int c4iw_max_read_depth = 8;
65 module_param(c4iw_max_read_depth, int, 0644);
66 MODULE_PARM_DESC(c4iw_max_read_depth, "Per-connection max ORD/IRD (default=8)");
68 static int enable_tcp_timestamps;
69 module_param(enable_tcp_timestamps, int, 0644);
70 MODULE_PARM_DESC(enable_tcp_timestamps, "Enable tcp timestamps (default=0)");
72 static int enable_tcp_sack;
73 module_param(enable_tcp_sack, int, 0644);
74 MODULE_PARM_DESC(enable_tcp_sack, "Enable tcp SACK (default=0)");
76 static int enable_tcp_window_scaling = 1;
77 module_param(enable_tcp_window_scaling, int, 0644);
78 MODULE_PARM_DESC(enable_tcp_window_scaling,
79 "Enable tcp window scaling (default=1)");
81 int c4iw_debug;
82 module_param(c4iw_debug, int, 0644);
83 MODULE_PARM_DESC(c4iw_debug, "Enable debug logging (default=0)");
85 static int peer2peer;
86 module_param(peer2peer, int, 0644);
87 MODULE_PARM_DESC(peer2peer, "Support peer2peer ULPs (default=0)");
89 static int p2p_type = FW_RI_INIT_P2PTYPE_READ_REQ;
90 module_param(p2p_type, int, 0644);
91 MODULE_PARM_DESC(p2p_type, "RDMAP opcode to use for the RTR message: "
92 "1=RDMA_READ 0=RDMA_WRITE (default 1)");
94 static int ep_timeout_secs = 60;
95 module_param(ep_timeout_secs, int, 0644);
96 MODULE_PARM_DESC(ep_timeout_secs, "CM Endpoint operation timeout "
97 "in seconds (default=60)");
99 static int mpa_rev = 1;
100 module_param(mpa_rev, int, 0644);
101 MODULE_PARM_DESC(mpa_rev, "MPA Revision, 0 supports amso1100, "
102 "1 is spec compliant. (default=1)");
104 static int markers_enabled;
105 module_param(markers_enabled, int, 0644);
106 MODULE_PARM_DESC(markers_enabled, "Enable MPA MARKERS (default(0)=disabled)");
108 static int crc_enabled = 1;
109 module_param(crc_enabled, int, 0644);
110 MODULE_PARM_DESC(crc_enabled, "Enable MPA CRC (default(1)=enabled)");
112 static int rcv_win = 256 * 1024;
113 module_param(rcv_win, int, 0644);
114 MODULE_PARM_DESC(rcv_win, "TCP receive window in bytes (default=256KB)");
116 static int snd_win = 32 * 1024;
117 module_param(snd_win, int, 0644);
118 MODULE_PARM_DESC(snd_win, "TCP send window in bytes (default=32KB)");
120 static struct workqueue_struct *workq;
122 static struct sk_buff_head rxq;
124 static struct sk_buff *get_skb(struct sk_buff *skb, int len, gfp_t gfp);
125 static void ep_timeout(unsigned long arg);
126 static void connect_reply_upcall(struct c4iw_ep *ep, int status);
128 static LIST_HEAD(timeout_list);
129 static spinlock_t timeout_lock;
131 static void start_ep_timer(struct c4iw_ep *ep)
133 PDBG("%s ep %p\n", __func__, ep);
134 if (timer_pending(&ep->timer)) {
135 PDBG("%s stopped / restarted timer ep %p\n", __func__, ep);
136 del_timer_sync(&ep->timer);
137 } else
138 c4iw_get_ep(&ep->com);
139 ep->timer.expires = jiffies + ep_timeout_secs * HZ;
140 ep->timer.data = (unsigned long)ep;
141 ep->timer.function = ep_timeout;
142 add_timer(&ep->timer);
145 static void stop_ep_timer(struct c4iw_ep *ep)
147 PDBG("%s ep %p\n", __func__, ep);
148 if (!timer_pending(&ep->timer)) {
149 printk(KERN_ERR "%s timer stopped when its not running! "
150 "ep %p state %u\n", __func__, ep, ep->com.state);
151 WARN_ON(1);
152 return;
154 del_timer_sync(&ep->timer);
155 c4iw_put_ep(&ep->com);
158 static int c4iw_l2t_send(struct c4iw_rdev *rdev, struct sk_buff *skb,
159 struct l2t_entry *l2e)
161 int error = 0;
163 if (c4iw_fatal_error(rdev)) {
164 kfree_skb(skb);
165 PDBG("%s - device in error state - dropping\n", __func__);
166 return -EIO;
168 error = cxgb4_l2t_send(rdev->lldi.ports[0], skb, l2e);
169 if (error < 0)
170 kfree_skb(skb);
171 return error;
174 int c4iw_ofld_send(struct c4iw_rdev *rdev, struct sk_buff *skb)
176 int error = 0;
178 if (c4iw_fatal_error(rdev)) {
179 kfree_skb(skb);
180 PDBG("%s - device in error state - dropping\n", __func__);
181 return -EIO;
183 error = cxgb4_ofld_send(rdev->lldi.ports[0], skb);
184 if (error < 0)
185 kfree_skb(skb);
186 return error;
189 static void release_tid(struct c4iw_rdev *rdev, u32 hwtid, struct sk_buff *skb)
191 struct cpl_tid_release *req;
193 skb = get_skb(skb, sizeof *req, GFP_KERNEL);
194 if (!skb)
195 return;
196 req = (struct cpl_tid_release *) skb_put(skb, sizeof(*req));
197 INIT_TP_WR(req, hwtid);
198 OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_TID_RELEASE, hwtid));
199 set_wr_txq(skb, CPL_PRIORITY_SETUP, 0);
200 c4iw_ofld_send(rdev, skb);
201 return;
204 static void set_emss(struct c4iw_ep *ep, u16 opt)
206 ep->emss = ep->com.dev->rdev.lldi.mtus[GET_TCPOPT_MSS(opt)] - 40;
207 ep->mss = ep->emss;
208 if (GET_TCPOPT_TSTAMP(opt))
209 ep->emss -= 12;
210 if (ep->emss < 128)
211 ep->emss = 128;
212 PDBG("%s mss_idx %u mss %u emss=%u\n", __func__, GET_TCPOPT_MSS(opt),
213 ep->mss, ep->emss);
216 static enum c4iw_ep_state state_read(struct c4iw_ep_common *epc)
218 unsigned long flags;
219 enum c4iw_ep_state state;
221 spin_lock_irqsave(&epc->lock, flags);
222 state = epc->state;
223 spin_unlock_irqrestore(&epc->lock, flags);
224 return state;
227 static void __state_set(struct c4iw_ep_common *epc, enum c4iw_ep_state new)
229 epc->state = new;
232 static void state_set(struct c4iw_ep_common *epc, enum c4iw_ep_state new)
234 unsigned long flags;
236 spin_lock_irqsave(&epc->lock, flags);
237 PDBG("%s - %s -> %s\n", __func__, states[epc->state], states[new]);
238 __state_set(epc, new);
239 spin_unlock_irqrestore(&epc->lock, flags);
240 return;
243 static void *alloc_ep(int size, gfp_t gfp)
245 struct c4iw_ep_common *epc;
247 epc = kzalloc(size, gfp);
248 if (epc) {
249 kref_init(&epc->kref);
250 spin_lock_init(&epc->lock);
251 init_waitqueue_head(&epc->waitq);
253 PDBG("%s alloc ep %p\n", __func__, epc);
254 return epc;
257 void _c4iw_free_ep(struct kref *kref)
259 struct c4iw_ep *ep;
261 ep = container_of(kref, struct c4iw_ep, com.kref);
262 PDBG("%s ep %p state %s\n", __func__, ep, states[state_read(&ep->com)]);
263 if (test_bit(RELEASE_RESOURCES, &ep->com.flags)) {
264 cxgb4_remove_tid(ep->com.dev->rdev.lldi.tids, 0, ep->hwtid);
265 dst_release(ep->dst);
266 cxgb4_l2t_release(ep->l2t);
268 kfree(ep);
271 static void release_ep_resources(struct c4iw_ep *ep)
273 set_bit(RELEASE_RESOURCES, &ep->com.flags);
274 c4iw_put_ep(&ep->com);
277 static int status2errno(int status)
279 switch (status) {
280 case CPL_ERR_NONE:
281 return 0;
282 case CPL_ERR_CONN_RESET:
283 return -ECONNRESET;
284 case CPL_ERR_ARP_MISS:
285 return -EHOSTUNREACH;
286 case CPL_ERR_CONN_TIMEDOUT:
287 return -ETIMEDOUT;
288 case CPL_ERR_TCAM_FULL:
289 return -ENOMEM;
290 case CPL_ERR_CONN_EXIST:
291 return -EADDRINUSE;
292 default:
293 return -EIO;
298 * Try and reuse skbs already allocated...
300 static struct sk_buff *get_skb(struct sk_buff *skb, int len, gfp_t gfp)
302 if (skb && !skb_is_nonlinear(skb) && !skb_cloned(skb)) {
303 skb_trim(skb, 0);
304 skb_get(skb);
305 skb_reset_transport_header(skb);
306 } else {
307 skb = alloc_skb(len, gfp);
309 return skb;
312 static struct rtable *find_route(struct c4iw_dev *dev, __be32 local_ip,
313 __be32 peer_ip, __be16 local_port,
314 __be16 peer_port, u8 tos)
316 struct rtable *rt;
317 struct flowi fl = {
318 .oif = 0,
319 .nl_u = {
320 .ip4_u = {
321 .daddr = peer_ip,
322 .saddr = local_ip,
323 .tos = tos}
325 .proto = IPPROTO_TCP,
326 .uli_u = {
327 .ports = {
328 .sport = local_port,
329 .dport = peer_port}
333 if (ip_route_output_flow(&init_net, &rt, &fl, NULL, 0))
334 return NULL;
335 return rt;
338 static void arp_failure_discard(void *handle, struct sk_buff *skb)
340 PDBG("%s c4iw_dev %p\n", __func__, handle);
341 kfree_skb(skb);
345 * Handle an ARP failure for an active open.
347 static void act_open_req_arp_failure(void *handle, struct sk_buff *skb)
349 printk(KERN_ERR MOD "ARP failure duing connect\n");
350 kfree_skb(skb);
354 * Handle an ARP failure for a CPL_ABORT_REQ. Change it into a no RST variant
355 * and send it along.
357 static void abort_arp_failure(void *handle, struct sk_buff *skb)
359 struct c4iw_rdev *rdev = handle;
360 struct cpl_abort_req *req = cplhdr(skb);
362 PDBG("%s rdev %p\n", __func__, rdev);
363 req->cmd = CPL_ABORT_NO_RST;
364 c4iw_ofld_send(rdev, skb);
367 static void send_flowc(struct c4iw_ep *ep, struct sk_buff *skb)
369 unsigned int flowclen = 80;
370 struct fw_flowc_wr *flowc;
371 int i;
373 skb = get_skb(skb, flowclen, GFP_KERNEL);
374 flowc = (struct fw_flowc_wr *)__skb_put(skb, flowclen);
376 flowc->op_to_nparams = cpu_to_be32(FW_WR_OP(FW_FLOWC_WR) |
377 FW_FLOWC_WR_NPARAMS(8));
378 flowc->flowid_len16 = cpu_to_be32(FW_WR_LEN16(DIV_ROUND_UP(flowclen,
379 16)) | FW_WR_FLOWID(ep->hwtid));
381 flowc->mnemval[0].mnemonic = FW_FLOWC_MNEM_PFNVFN;
382 flowc->mnemval[0].val = cpu_to_be32(0);
383 flowc->mnemval[1].mnemonic = FW_FLOWC_MNEM_CH;
384 flowc->mnemval[1].val = cpu_to_be32(ep->tx_chan);
385 flowc->mnemval[2].mnemonic = FW_FLOWC_MNEM_PORT;
386 flowc->mnemval[2].val = cpu_to_be32(ep->tx_chan);
387 flowc->mnemval[3].mnemonic = FW_FLOWC_MNEM_IQID;
388 flowc->mnemval[3].val = cpu_to_be32(ep->rss_qid);
389 flowc->mnemval[4].mnemonic = FW_FLOWC_MNEM_SNDNXT;
390 flowc->mnemval[4].val = cpu_to_be32(ep->snd_seq);
391 flowc->mnemval[5].mnemonic = FW_FLOWC_MNEM_RCVNXT;
392 flowc->mnemval[5].val = cpu_to_be32(ep->rcv_seq);
393 flowc->mnemval[6].mnemonic = FW_FLOWC_MNEM_SNDBUF;
394 flowc->mnemval[6].val = cpu_to_be32(snd_win);
395 flowc->mnemval[7].mnemonic = FW_FLOWC_MNEM_MSS;
396 flowc->mnemval[7].val = cpu_to_be32(ep->emss);
397 /* Pad WR to 16 byte boundary */
398 flowc->mnemval[8].mnemonic = 0;
399 flowc->mnemval[8].val = 0;
400 for (i = 0; i < 9; i++) {
401 flowc->mnemval[i].r4[0] = 0;
402 flowc->mnemval[i].r4[1] = 0;
403 flowc->mnemval[i].r4[2] = 0;
406 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
407 c4iw_ofld_send(&ep->com.dev->rdev, skb);
410 static int send_halfclose(struct c4iw_ep *ep, gfp_t gfp)
412 struct cpl_close_con_req *req;
413 struct sk_buff *skb;
414 int wrlen = roundup(sizeof *req, 16);
416 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
417 skb = get_skb(NULL, wrlen, gfp);
418 if (!skb) {
419 printk(KERN_ERR MOD "%s - failed to alloc skb\n", __func__);
420 return -ENOMEM;
422 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
423 t4_set_arp_err_handler(skb, NULL, arp_failure_discard);
424 req = (struct cpl_close_con_req *) skb_put(skb, wrlen);
425 memset(req, 0, wrlen);
426 INIT_TP_WR(req, ep->hwtid);
427 OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_CLOSE_CON_REQ,
428 ep->hwtid));
429 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
432 static int send_abort(struct c4iw_ep *ep, struct sk_buff *skb, gfp_t gfp)
434 struct cpl_abort_req *req;
435 int wrlen = roundup(sizeof *req, 16);
437 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
438 skb = get_skb(skb, wrlen, gfp);
439 if (!skb) {
440 printk(KERN_ERR MOD "%s - failed to alloc skb.\n",
441 __func__);
442 return -ENOMEM;
444 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
445 t4_set_arp_err_handler(skb, &ep->com.dev->rdev, abort_arp_failure);
446 req = (struct cpl_abort_req *) skb_put(skb, wrlen);
447 memset(req, 0, wrlen);
448 INIT_TP_WR(req, ep->hwtid);
449 OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_ABORT_REQ, ep->hwtid));
450 req->cmd = CPL_ABORT_SEND_RST;
451 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
454 static int send_connect(struct c4iw_ep *ep)
456 struct cpl_act_open_req *req;
457 struct sk_buff *skb;
458 u64 opt0;
459 u32 opt2;
460 unsigned int mtu_idx;
461 int wscale;
462 int wrlen = roundup(sizeof *req, 16);
464 PDBG("%s ep %p atid %u\n", __func__, ep, ep->atid);
466 skb = get_skb(NULL, wrlen, GFP_KERNEL);
467 if (!skb) {
468 printk(KERN_ERR MOD "%s - failed to alloc skb.\n",
469 __func__);
470 return -ENOMEM;
472 set_wr_txq(skb, CPL_PRIORITY_SETUP, ep->txq_idx);
474 cxgb4_best_mtu(ep->com.dev->rdev.lldi.mtus, ep->mtu, &mtu_idx);
475 wscale = compute_wscale(rcv_win);
476 opt0 = KEEP_ALIVE(1) |
477 WND_SCALE(wscale) |
478 MSS_IDX(mtu_idx) |
479 L2T_IDX(ep->l2t->idx) |
480 TX_CHAN(ep->tx_chan) |
481 SMAC_SEL(ep->smac_idx) |
482 DSCP(ep->tos) |
483 RCV_BUFSIZ(rcv_win>>10);
484 opt2 = RX_CHANNEL(0) |
485 RSS_QUEUE_VALID | RSS_QUEUE(ep->rss_qid);
486 if (enable_tcp_timestamps)
487 opt2 |= TSTAMPS_EN(1);
488 if (enable_tcp_sack)
489 opt2 |= SACK_EN(1);
490 if (wscale && enable_tcp_window_scaling)
491 opt2 |= WND_SCALE_EN(1);
492 t4_set_arp_err_handler(skb, NULL, act_open_req_arp_failure);
494 req = (struct cpl_act_open_req *) skb_put(skb, wrlen);
495 INIT_TP_WR(req, 0);
496 OPCODE_TID(req) = cpu_to_be32(
497 MK_OPCODE_TID(CPL_ACT_OPEN_REQ, ((ep->rss_qid<<14)|ep->atid)));
498 req->local_port = ep->com.local_addr.sin_port;
499 req->peer_port = ep->com.remote_addr.sin_port;
500 req->local_ip = ep->com.local_addr.sin_addr.s_addr;
501 req->peer_ip = ep->com.remote_addr.sin_addr.s_addr;
502 req->opt0 = cpu_to_be64(opt0);
503 req->params = 0;
504 req->opt2 = cpu_to_be32(opt2);
505 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
508 static void send_mpa_req(struct c4iw_ep *ep, struct sk_buff *skb)
510 int mpalen, wrlen;
511 struct fw_ofld_tx_data_wr *req;
512 struct mpa_message *mpa;
514 PDBG("%s ep %p tid %u pd_len %d\n", __func__, ep, ep->hwtid, ep->plen);
516 BUG_ON(skb_cloned(skb));
518 mpalen = sizeof(*mpa) + ep->plen;
519 wrlen = roundup(mpalen + sizeof *req, 16);
520 skb = get_skb(skb, wrlen, GFP_KERNEL);
521 if (!skb) {
522 connect_reply_upcall(ep, -ENOMEM);
523 return;
525 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
527 req = (struct fw_ofld_tx_data_wr *)skb_put(skb, wrlen);
528 memset(req, 0, wrlen);
529 req->op_to_immdlen = cpu_to_be32(
530 FW_WR_OP(FW_OFLD_TX_DATA_WR) |
531 FW_WR_COMPL(1) |
532 FW_WR_IMMDLEN(mpalen));
533 req->flowid_len16 = cpu_to_be32(
534 FW_WR_FLOWID(ep->hwtid) |
535 FW_WR_LEN16(wrlen >> 4));
536 req->plen = cpu_to_be32(mpalen);
537 req->tunnel_to_proxy = cpu_to_be32(
538 FW_OFLD_TX_DATA_WR_FLUSH(1) |
539 FW_OFLD_TX_DATA_WR_SHOVE(1));
541 mpa = (struct mpa_message *)(req + 1);
542 memcpy(mpa->key, MPA_KEY_REQ, sizeof(mpa->key));
543 mpa->flags = (crc_enabled ? MPA_CRC : 0) |
544 (markers_enabled ? MPA_MARKERS : 0);
545 mpa->private_data_size = htons(ep->plen);
546 mpa->revision = mpa_rev;
548 if (ep->plen)
549 memcpy(mpa->private_data, ep->mpa_pkt + sizeof(*mpa), ep->plen);
552 * Reference the mpa skb. This ensures the data area
553 * will remain in memory until the hw acks the tx.
554 * Function fw4_ack() will deref it.
556 skb_get(skb);
557 t4_set_arp_err_handler(skb, NULL, arp_failure_discard);
558 BUG_ON(ep->mpa_skb);
559 ep->mpa_skb = skb;
560 c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
561 start_ep_timer(ep);
562 state_set(&ep->com, MPA_REQ_SENT);
563 ep->mpa_attr.initiator = 1;
564 return;
567 static int send_mpa_reject(struct c4iw_ep *ep, const void *pdata, u8 plen)
569 int mpalen, wrlen;
570 struct fw_ofld_tx_data_wr *req;
571 struct mpa_message *mpa;
572 struct sk_buff *skb;
574 PDBG("%s ep %p tid %u pd_len %d\n", __func__, ep, ep->hwtid, ep->plen);
576 mpalen = sizeof(*mpa) + plen;
577 wrlen = roundup(mpalen + sizeof *req, 16);
579 skb = get_skb(NULL, wrlen, GFP_KERNEL);
580 if (!skb) {
581 printk(KERN_ERR MOD "%s - cannot alloc skb!\n", __func__);
582 return -ENOMEM;
584 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
586 req = (struct fw_ofld_tx_data_wr *)skb_put(skb, wrlen);
587 memset(req, 0, wrlen);
588 req->op_to_immdlen = cpu_to_be32(
589 FW_WR_OP(FW_OFLD_TX_DATA_WR) |
590 FW_WR_COMPL(1) |
591 FW_WR_IMMDLEN(mpalen));
592 req->flowid_len16 = cpu_to_be32(
593 FW_WR_FLOWID(ep->hwtid) |
594 FW_WR_LEN16(wrlen >> 4));
595 req->plen = cpu_to_be32(mpalen);
596 req->tunnel_to_proxy = cpu_to_be32(
597 FW_OFLD_TX_DATA_WR_FLUSH(1) |
598 FW_OFLD_TX_DATA_WR_SHOVE(1));
600 mpa = (struct mpa_message *)(req + 1);
601 memset(mpa, 0, sizeof(*mpa));
602 memcpy(mpa->key, MPA_KEY_REP, sizeof(mpa->key));
603 mpa->flags = MPA_REJECT;
604 mpa->revision = mpa_rev;
605 mpa->private_data_size = htons(plen);
606 if (plen)
607 memcpy(mpa->private_data, pdata, plen);
610 * Reference the mpa skb again. This ensures the data area
611 * will remain in memory until the hw acks the tx.
612 * Function fw4_ack() will deref it.
614 skb_get(skb);
615 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
616 t4_set_arp_err_handler(skb, NULL, arp_failure_discard);
617 BUG_ON(ep->mpa_skb);
618 ep->mpa_skb = skb;
619 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
622 static int send_mpa_reply(struct c4iw_ep *ep, const void *pdata, u8 plen)
624 int mpalen, wrlen;
625 struct fw_ofld_tx_data_wr *req;
626 struct mpa_message *mpa;
627 struct sk_buff *skb;
629 PDBG("%s ep %p tid %u pd_len %d\n", __func__, ep, ep->hwtid, ep->plen);
631 mpalen = sizeof(*mpa) + plen;
632 wrlen = roundup(mpalen + sizeof *req, 16);
634 skb = get_skb(NULL, wrlen, GFP_KERNEL);
635 if (!skb) {
636 printk(KERN_ERR MOD "%s - cannot alloc skb!\n", __func__);
637 return -ENOMEM;
639 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
641 req = (struct fw_ofld_tx_data_wr *) skb_put(skb, wrlen);
642 memset(req, 0, wrlen);
643 req->op_to_immdlen = cpu_to_be32(
644 FW_WR_OP(FW_OFLD_TX_DATA_WR) |
645 FW_WR_COMPL(1) |
646 FW_WR_IMMDLEN(mpalen));
647 req->flowid_len16 = cpu_to_be32(
648 FW_WR_FLOWID(ep->hwtid) |
649 FW_WR_LEN16(wrlen >> 4));
650 req->plen = cpu_to_be32(mpalen);
651 req->tunnel_to_proxy = cpu_to_be32(
652 FW_OFLD_TX_DATA_WR_FLUSH(1) |
653 FW_OFLD_TX_DATA_WR_SHOVE(1));
655 mpa = (struct mpa_message *)(req + 1);
656 memset(mpa, 0, sizeof(*mpa));
657 memcpy(mpa->key, MPA_KEY_REP, sizeof(mpa->key));
658 mpa->flags = (ep->mpa_attr.crc_enabled ? MPA_CRC : 0) |
659 (markers_enabled ? MPA_MARKERS : 0);
660 mpa->revision = mpa_rev;
661 mpa->private_data_size = htons(plen);
662 if (plen)
663 memcpy(mpa->private_data, pdata, plen);
666 * Reference the mpa skb. This ensures the data area
667 * will remain in memory until the hw acks the tx.
668 * Function fw4_ack() will deref it.
670 skb_get(skb);
671 t4_set_arp_err_handler(skb, NULL, arp_failure_discard);
672 ep->mpa_skb = skb;
673 state_set(&ep->com, MPA_REP_SENT);
674 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
677 static int act_establish(struct c4iw_dev *dev, struct sk_buff *skb)
679 struct c4iw_ep *ep;
680 struct cpl_act_establish *req = cplhdr(skb);
681 unsigned int tid = GET_TID(req);
682 unsigned int atid = GET_TID_TID(ntohl(req->tos_atid));
683 struct tid_info *t = dev->rdev.lldi.tids;
685 ep = lookup_atid(t, atid);
687 PDBG("%s ep %p tid %u snd_isn %u rcv_isn %u\n", __func__, ep, tid,
688 be32_to_cpu(req->snd_isn), be32_to_cpu(req->rcv_isn));
690 dst_confirm(ep->dst);
692 /* setup the hwtid for this connection */
693 ep->hwtid = tid;
694 cxgb4_insert_tid(t, ep, tid);
696 ep->snd_seq = be32_to_cpu(req->snd_isn);
697 ep->rcv_seq = be32_to_cpu(req->rcv_isn);
699 set_emss(ep, ntohs(req->tcp_opt));
701 /* dealloc the atid */
702 cxgb4_free_atid(t, atid);
704 /* start MPA negotiation */
705 send_flowc(ep, NULL);
706 send_mpa_req(ep, skb);
708 return 0;
711 static void close_complete_upcall(struct c4iw_ep *ep)
713 struct iw_cm_event event;
715 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
716 memset(&event, 0, sizeof(event));
717 event.event = IW_CM_EVENT_CLOSE;
718 if (ep->com.cm_id) {
719 PDBG("close complete delivered ep %p cm_id %p tid %u\n",
720 ep, ep->com.cm_id, ep->hwtid);
721 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
722 ep->com.cm_id->rem_ref(ep->com.cm_id);
723 ep->com.cm_id = NULL;
724 ep->com.qp = NULL;
728 static int abort_connection(struct c4iw_ep *ep, struct sk_buff *skb, gfp_t gfp)
730 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
731 close_complete_upcall(ep);
732 state_set(&ep->com, ABORTING);
733 return send_abort(ep, skb, gfp);
736 static void peer_close_upcall(struct c4iw_ep *ep)
738 struct iw_cm_event event;
740 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
741 memset(&event, 0, sizeof(event));
742 event.event = IW_CM_EVENT_DISCONNECT;
743 if (ep->com.cm_id) {
744 PDBG("peer close delivered ep %p cm_id %p tid %u\n",
745 ep, ep->com.cm_id, ep->hwtid);
746 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
750 static void peer_abort_upcall(struct c4iw_ep *ep)
752 struct iw_cm_event event;
754 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
755 memset(&event, 0, sizeof(event));
756 event.event = IW_CM_EVENT_CLOSE;
757 event.status = -ECONNRESET;
758 if (ep->com.cm_id) {
759 PDBG("abort delivered ep %p cm_id %p tid %u\n", ep,
760 ep->com.cm_id, ep->hwtid);
761 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
762 ep->com.cm_id->rem_ref(ep->com.cm_id);
763 ep->com.cm_id = NULL;
764 ep->com.qp = NULL;
768 static void connect_reply_upcall(struct c4iw_ep *ep, int status)
770 struct iw_cm_event event;
772 PDBG("%s ep %p tid %u status %d\n", __func__, ep, ep->hwtid, status);
773 memset(&event, 0, sizeof(event));
774 event.event = IW_CM_EVENT_CONNECT_REPLY;
775 event.status = status;
776 event.local_addr = ep->com.local_addr;
777 event.remote_addr = ep->com.remote_addr;
779 if ((status == 0) || (status == -ECONNREFUSED)) {
780 event.private_data_len = ep->plen;
781 event.private_data = ep->mpa_pkt + sizeof(struct mpa_message);
783 if (ep->com.cm_id) {
784 PDBG("%s ep %p tid %u status %d\n", __func__, ep,
785 ep->hwtid, status);
786 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
788 if (status < 0) {
789 ep->com.cm_id->rem_ref(ep->com.cm_id);
790 ep->com.cm_id = NULL;
791 ep->com.qp = NULL;
795 static void connect_request_upcall(struct c4iw_ep *ep)
797 struct iw_cm_event event;
799 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
800 memset(&event, 0, sizeof(event));
801 event.event = IW_CM_EVENT_CONNECT_REQUEST;
802 event.local_addr = ep->com.local_addr;
803 event.remote_addr = ep->com.remote_addr;
804 event.private_data_len = ep->plen;
805 event.private_data = ep->mpa_pkt + sizeof(struct mpa_message);
806 event.provider_data = ep;
807 if (state_read(&ep->parent_ep->com) != DEAD) {
808 c4iw_get_ep(&ep->com);
809 ep->parent_ep->com.cm_id->event_handler(
810 ep->parent_ep->com.cm_id,
811 &event);
813 c4iw_put_ep(&ep->parent_ep->com);
814 ep->parent_ep = NULL;
817 static void established_upcall(struct c4iw_ep *ep)
819 struct iw_cm_event event;
821 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
822 memset(&event, 0, sizeof(event));
823 event.event = IW_CM_EVENT_ESTABLISHED;
824 if (ep->com.cm_id) {
825 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
826 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
830 static int update_rx_credits(struct c4iw_ep *ep, u32 credits)
832 struct cpl_rx_data_ack *req;
833 struct sk_buff *skb;
834 int wrlen = roundup(sizeof *req, 16);
836 PDBG("%s ep %p tid %u credits %u\n", __func__, ep, ep->hwtid, credits);
837 skb = get_skb(NULL, wrlen, GFP_KERNEL);
838 if (!skb) {
839 printk(KERN_ERR MOD "update_rx_credits - cannot alloc skb!\n");
840 return 0;
843 req = (struct cpl_rx_data_ack *) skb_put(skb, wrlen);
844 memset(req, 0, wrlen);
845 INIT_TP_WR(req, ep->hwtid);
846 OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_RX_DATA_ACK,
847 ep->hwtid));
848 req->credit_dack = cpu_to_be32(credits);
849 set_wr_txq(skb, CPL_PRIORITY_ACK, ep->txq_idx);
850 c4iw_ofld_send(&ep->com.dev->rdev, skb);
851 return credits;
854 static void process_mpa_reply(struct c4iw_ep *ep, struct sk_buff *skb)
856 struct mpa_message *mpa;
857 u16 plen;
858 struct c4iw_qp_attributes attrs;
859 enum c4iw_qp_attr_mask mask;
860 int err;
862 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
865 * Stop mpa timer. If it expired, then the state has
866 * changed and we bail since ep_timeout already aborted
867 * the connection.
869 stop_ep_timer(ep);
870 if (state_read(&ep->com) != MPA_REQ_SENT)
871 return;
874 * If we get more than the supported amount of private data
875 * then we must fail this connection.
877 if (ep->mpa_pkt_len + skb->len > sizeof(ep->mpa_pkt)) {
878 err = -EINVAL;
879 goto err;
883 * copy the new data into our accumulation buffer.
885 skb_copy_from_linear_data(skb, &(ep->mpa_pkt[ep->mpa_pkt_len]),
886 skb->len);
887 ep->mpa_pkt_len += skb->len;
890 * if we don't even have the mpa message, then bail.
892 if (ep->mpa_pkt_len < sizeof(*mpa))
893 return;
894 mpa = (struct mpa_message *) ep->mpa_pkt;
896 /* Validate MPA header. */
897 if (mpa->revision != mpa_rev) {
898 err = -EPROTO;
899 goto err;
901 if (memcmp(mpa->key, MPA_KEY_REP, sizeof(mpa->key))) {
902 err = -EPROTO;
903 goto err;
906 plen = ntohs(mpa->private_data_size);
909 * Fail if there's too much private data.
911 if (plen > MPA_MAX_PRIVATE_DATA) {
912 err = -EPROTO;
913 goto err;
917 * If plen does not account for pkt size
919 if (ep->mpa_pkt_len > (sizeof(*mpa) + plen)) {
920 err = -EPROTO;
921 goto err;
924 ep->plen = (u8) plen;
927 * If we don't have all the pdata yet, then bail.
928 * We'll continue process when more data arrives.
930 if (ep->mpa_pkt_len < (sizeof(*mpa) + plen))
931 return;
933 if (mpa->flags & MPA_REJECT) {
934 err = -ECONNREFUSED;
935 goto err;
939 * If we get here we have accumulated the entire mpa
940 * start reply message including private data. And
941 * the MPA header is valid.
943 state_set(&ep->com, FPDU_MODE);
944 ep->mpa_attr.crc_enabled = (mpa->flags & MPA_CRC) | crc_enabled ? 1 : 0;
945 ep->mpa_attr.recv_marker_enabled = markers_enabled;
946 ep->mpa_attr.xmit_marker_enabled = mpa->flags & MPA_MARKERS ? 1 : 0;
947 ep->mpa_attr.version = mpa_rev;
948 ep->mpa_attr.p2p_type = peer2peer ? p2p_type :
949 FW_RI_INIT_P2PTYPE_DISABLED;
950 PDBG("%s - crc_enabled=%d, recv_marker_enabled=%d, "
951 "xmit_marker_enabled=%d, version=%d\n", __func__,
952 ep->mpa_attr.crc_enabled, ep->mpa_attr.recv_marker_enabled,
953 ep->mpa_attr.xmit_marker_enabled, ep->mpa_attr.version);
955 attrs.mpa_attr = ep->mpa_attr;
956 attrs.max_ird = ep->ird;
957 attrs.max_ord = ep->ord;
958 attrs.llp_stream_handle = ep;
959 attrs.next_state = C4IW_QP_STATE_RTS;
961 mask = C4IW_QP_ATTR_NEXT_STATE |
962 C4IW_QP_ATTR_LLP_STREAM_HANDLE | C4IW_QP_ATTR_MPA_ATTR |
963 C4IW_QP_ATTR_MAX_IRD | C4IW_QP_ATTR_MAX_ORD;
965 /* bind QP and TID with INIT_WR */
966 err = c4iw_modify_qp(ep->com.qp->rhp,
967 ep->com.qp, mask, &attrs, 1);
968 if (err)
969 goto err;
970 goto out;
971 err:
972 abort_connection(ep, skb, GFP_KERNEL);
973 out:
974 connect_reply_upcall(ep, err);
975 return;
978 static void process_mpa_request(struct c4iw_ep *ep, struct sk_buff *skb)
980 struct mpa_message *mpa;
981 u16 plen;
983 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
985 if (state_read(&ep->com) != MPA_REQ_WAIT)
986 return;
989 * If we get more than the supported amount of private data
990 * then we must fail this connection.
992 if (ep->mpa_pkt_len + skb->len > sizeof(ep->mpa_pkt)) {
993 stop_ep_timer(ep);
994 abort_connection(ep, skb, GFP_KERNEL);
995 return;
998 PDBG("%s enter (%s line %u)\n", __func__, __FILE__, __LINE__);
1001 * Copy the new data into our accumulation buffer.
1003 skb_copy_from_linear_data(skb, &(ep->mpa_pkt[ep->mpa_pkt_len]),
1004 skb->len);
1005 ep->mpa_pkt_len += skb->len;
1008 * If we don't even have the mpa message, then bail.
1009 * We'll continue process when more data arrives.
1011 if (ep->mpa_pkt_len < sizeof(*mpa))
1012 return;
1014 PDBG("%s enter (%s line %u)\n", __func__, __FILE__, __LINE__);
1015 stop_ep_timer(ep);
1016 mpa = (struct mpa_message *) ep->mpa_pkt;
1019 * Validate MPA Header.
1021 if (mpa->revision != mpa_rev) {
1022 abort_connection(ep, skb, GFP_KERNEL);
1023 return;
1026 if (memcmp(mpa->key, MPA_KEY_REQ, sizeof(mpa->key))) {
1027 abort_connection(ep, skb, GFP_KERNEL);
1028 return;
1031 plen = ntohs(mpa->private_data_size);
1034 * Fail if there's too much private data.
1036 if (plen > MPA_MAX_PRIVATE_DATA) {
1037 abort_connection(ep, skb, GFP_KERNEL);
1038 return;
1042 * If plen does not account for pkt size
1044 if (ep->mpa_pkt_len > (sizeof(*mpa) + plen)) {
1045 abort_connection(ep, skb, GFP_KERNEL);
1046 return;
1048 ep->plen = (u8) plen;
1051 * If we don't have all the pdata yet, then bail.
1053 if (ep->mpa_pkt_len < (sizeof(*mpa) + plen))
1054 return;
1057 * If we get here we have accumulated the entire mpa
1058 * start reply message including private data.
1060 ep->mpa_attr.initiator = 0;
1061 ep->mpa_attr.crc_enabled = (mpa->flags & MPA_CRC) | crc_enabled ? 1 : 0;
1062 ep->mpa_attr.recv_marker_enabled = markers_enabled;
1063 ep->mpa_attr.xmit_marker_enabled = mpa->flags & MPA_MARKERS ? 1 : 0;
1064 ep->mpa_attr.version = mpa_rev;
1065 ep->mpa_attr.p2p_type = peer2peer ? p2p_type :
1066 FW_RI_INIT_P2PTYPE_DISABLED;
1067 PDBG("%s - crc_enabled=%d, recv_marker_enabled=%d, "
1068 "xmit_marker_enabled=%d, version=%d p2p_type=%d\n", __func__,
1069 ep->mpa_attr.crc_enabled, ep->mpa_attr.recv_marker_enabled,
1070 ep->mpa_attr.xmit_marker_enabled, ep->mpa_attr.version,
1071 ep->mpa_attr.p2p_type);
1073 state_set(&ep->com, MPA_REQ_RCVD);
1075 /* drive upcall */
1076 connect_request_upcall(ep);
1077 return;
1080 static int rx_data(struct c4iw_dev *dev, struct sk_buff *skb)
1082 struct c4iw_ep *ep;
1083 struct cpl_rx_data *hdr = cplhdr(skb);
1084 unsigned int dlen = ntohs(hdr->len);
1085 unsigned int tid = GET_TID(hdr);
1086 struct tid_info *t = dev->rdev.lldi.tids;
1088 ep = lookup_tid(t, tid);
1089 PDBG("%s ep %p tid %u dlen %u\n", __func__, ep, ep->hwtid, dlen);
1090 skb_pull(skb, sizeof(*hdr));
1091 skb_trim(skb, dlen);
1093 ep->rcv_seq += dlen;
1094 BUG_ON(ep->rcv_seq != (ntohl(hdr->seq) + dlen));
1096 /* update RX credits */
1097 update_rx_credits(ep, dlen);
1099 switch (state_read(&ep->com)) {
1100 case MPA_REQ_SENT:
1101 process_mpa_reply(ep, skb);
1102 break;
1103 case MPA_REQ_WAIT:
1104 process_mpa_request(ep, skb);
1105 break;
1106 case MPA_REP_SENT:
1107 break;
1108 default:
1109 printk(KERN_ERR MOD "%s Unexpected streaming data."
1110 " ep %p state %d tid %u\n",
1111 __func__, ep, state_read(&ep->com), ep->hwtid);
1114 * The ep will timeout and inform the ULP of the failure.
1115 * See ep_timeout().
1117 break;
1119 return 0;
1122 static int abort_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
1124 struct c4iw_ep *ep;
1125 struct cpl_abort_rpl_rss *rpl = cplhdr(skb);
1126 unsigned long flags;
1127 int release = 0;
1128 unsigned int tid = GET_TID(rpl);
1129 struct tid_info *t = dev->rdev.lldi.tids;
1131 ep = lookup_tid(t, tid);
1132 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1133 BUG_ON(!ep);
1134 spin_lock_irqsave(&ep->com.lock, flags);
1135 switch (ep->com.state) {
1136 case ABORTING:
1137 __state_set(&ep->com, DEAD);
1138 release = 1;
1139 break;
1140 default:
1141 printk(KERN_ERR "%s ep %p state %d\n",
1142 __func__, ep, ep->com.state);
1143 break;
1145 spin_unlock_irqrestore(&ep->com.lock, flags);
1147 if (release)
1148 release_ep_resources(ep);
1149 return 0;
1153 * Return whether a failed active open has allocated a TID
1155 static inline int act_open_has_tid(int status)
1157 return status != CPL_ERR_TCAM_FULL && status != CPL_ERR_CONN_EXIST &&
1158 status != CPL_ERR_ARP_MISS;
1161 static int act_open_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
1163 struct c4iw_ep *ep;
1164 struct cpl_act_open_rpl *rpl = cplhdr(skb);
1165 unsigned int atid = GET_TID_TID(GET_AOPEN_ATID(
1166 ntohl(rpl->atid_status)));
1167 struct tid_info *t = dev->rdev.lldi.tids;
1168 int status = GET_AOPEN_STATUS(ntohl(rpl->atid_status));
1170 ep = lookup_atid(t, atid);
1172 PDBG("%s ep %p atid %u status %u errno %d\n", __func__, ep, atid,
1173 status, status2errno(status));
1175 if (status == CPL_ERR_RTX_NEG_ADVICE) {
1176 printk(KERN_WARNING MOD "Connection problems for atid %u\n",
1177 atid);
1178 return 0;
1181 connect_reply_upcall(ep, status2errno(status));
1182 state_set(&ep->com, DEAD);
1184 if (status && act_open_has_tid(status))
1185 cxgb4_remove_tid(ep->com.dev->rdev.lldi.tids, 0, GET_TID(rpl));
1187 cxgb4_free_atid(t, atid);
1188 dst_release(ep->dst);
1189 cxgb4_l2t_release(ep->l2t);
1190 c4iw_put_ep(&ep->com);
1192 return 0;
1195 static int pass_open_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
1197 struct cpl_pass_open_rpl *rpl = cplhdr(skb);
1198 struct tid_info *t = dev->rdev.lldi.tids;
1199 unsigned int stid = GET_TID(rpl);
1200 struct c4iw_listen_ep *ep = lookup_stid(t, stid);
1202 if (!ep) {
1203 printk(KERN_ERR MOD "stid %d lookup failure!\n", stid);
1204 return 0;
1206 PDBG("%s ep %p status %d error %d\n", __func__, ep,
1207 rpl->status, status2errno(rpl->status));
1208 ep->com.rpl_err = status2errno(rpl->status);
1209 ep->com.rpl_done = 1;
1210 wake_up(&ep->com.waitq);
1212 return 0;
1215 static int listen_stop(struct c4iw_listen_ep *ep)
1217 struct sk_buff *skb;
1218 struct cpl_close_listsvr_req *req;
1220 PDBG("%s ep %p\n", __func__, ep);
1221 skb = get_skb(NULL, sizeof(*req), GFP_KERNEL);
1222 if (!skb) {
1223 printk(KERN_ERR MOD "%s - failed to alloc skb\n", __func__);
1224 return -ENOMEM;
1226 req = (struct cpl_close_listsvr_req *) skb_put(skb, sizeof(*req));
1227 INIT_TP_WR(req, 0);
1228 OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_CLOSE_LISTSRV_REQ,
1229 ep->stid));
1230 req->reply_ctrl = cpu_to_be16(
1231 QUEUENO(ep->com.dev->rdev.lldi.rxq_ids[0]));
1232 set_wr_txq(skb, CPL_PRIORITY_SETUP, 0);
1233 return c4iw_ofld_send(&ep->com.dev->rdev, skb);
1236 static int close_listsrv_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
1238 struct cpl_close_listsvr_rpl *rpl = cplhdr(skb);
1239 struct tid_info *t = dev->rdev.lldi.tids;
1240 unsigned int stid = GET_TID(rpl);
1241 struct c4iw_listen_ep *ep = lookup_stid(t, stid);
1243 PDBG("%s ep %p\n", __func__, ep);
1244 ep->com.rpl_err = status2errno(rpl->status);
1245 ep->com.rpl_done = 1;
1246 wake_up(&ep->com.waitq);
1247 return 0;
1250 static void accept_cr(struct c4iw_ep *ep, __be32 peer_ip, struct sk_buff *skb,
1251 struct cpl_pass_accept_req *req)
1253 struct cpl_pass_accept_rpl *rpl;
1254 unsigned int mtu_idx;
1255 u64 opt0;
1256 u32 opt2;
1257 int wscale;
1259 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1260 BUG_ON(skb_cloned(skb));
1261 skb_trim(skb, sizeof(*rpl));
1262 skb_get(skb);
1263 cxgb4_best_mtu(ep->com.dev->rdev.lldi.mtus, ep->mtu, &mtu_idx);
1264 wscale = compute_wscale(rcv_win);
1265 opt0 = KEEP_ALIVE(1) |
1266 WND_SCALE(wscale) |
1267 MSS_IDX(mtu_idx) |
1268 L2T_IDX(ep->l2t->idx) |
1269 TX_CHAN(ep->tx_chan) |
1270 SMAC_SEL(ep->smac_idx) |
1271 DSCP(ep->tos) |
1272 RCV_BUFSIZ(rcv_win>>10);
1273 opt2 = RX_CHANNEL(0) |
1274 RSS_QUEUE_VALID | RSS_QUEUE(ep->rss_qid);
1276 if (enable_tcp_timestamps && req->tcpopt.tstamp)
1277 opt2 |= TSTAMPS_EN(1);
1278 if (enable_tcp_sack && req->tcpopt.sack)
1279 opt2 |= SACK_EN(1);
1280 if (wscale && enable_tcp_window_scaling)
1281 opt2 |= WND_SCALE_EN(1);
1283 rpl = cplhdr(skb);
1284 INIT_TP_WR(rpl, ep->hwtid);
1285 OPCODE_TID(rpl) = cpu_to_be32(MK_OPCODE_TID(CPL_PASS_ACCEPT_RPL,
1286 ep->hwtid));
1287 rpl->opt0 = cpu_to_be64(opt0);
1288 rpl->opt2 = cpu_to_be32(opt2);
1289 set_wr_txq(skb, CPL_PRIORITY_SETUP, ep->txq_idx);
1290 c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
1292 return;
1295 static void reject_cr(struct c4iw_dev *dev, u32 hwtid, __be32 peer_ip,
1296 struct sk_buff *skb)
1298 PDBG("%s c4iw_dev %p tid %u peer_ip %x\n", __func__, dev, hwtid,
1299 peer_ip);
1300 BUG_ON(skb_cloned(skb));
1301 skb_trim(skb, sizeof(struct cpl_tid_release));
1302 skb_get(skb);
1303 release_tid(&dev->rdev, hwtid, skb);
1304 return;
1307 static void get_4tuple(struct cpl_pass_accept_req *req,
1308 __be32 *local_ip, __be32 *peer_ip,
1309 __be16 *local_port, __be16 *peer_port)
1311 int eth_len = G_ETH_HDR_LEN(be32_to_cpu(req->hdr_len));
1312 int ip_len = G_IP_HDR_LEN(be32_to_cpu(req->hdr_len));
1313 struct iphdr *ip = (struct iphdr *)((u8 *)(req + 1) + eth_len);
1314 struct tcphdr *tcp = (struct tcphdr *)
1315 ((u8 *)(req + 1) + eth_len + ip_len);
1317 PDBG("%s saddr 0x%x daddr 0x%x sport %u dport %u\n", __func__,
1318 ntohl(ip->saddr), ntohl(ip->daddr), ntohs(tcp->source),
1319 ntohs(tcp->dest));
1321 *peer_ip = ip->saddr;
1322 *local_ip = ip->daddr;
1323 *peer_port = tcp->source;
1324 *local_port = tcp->dest;
1326 return;
1329 static int pass_accept_req(struct c4iw_dev *dev, struct sk_buff *skb)
1331 struct c4iw_ep *child_ep, *parent_ep;
1332 struct cpl_pass_accept_req *req = cplhdr(skb);
1333 unsigned int stid = GET_POPEN_TID(ntohl(req->tos_stid));
1334 struct tid_info *t = dev->rdev.lldi.tids;
1335 unsigned int hwtid = GET_TID(req);
1336 struct dst_entry *dst;
1337 struct l2t_entry *l2t;
1338 struct rtable *rt;
1339 __be32 local_ip, peer_ip;
1340 __be16 local_port, peer_port;
1341 struct net_device *pdev;
1342 u32 tx_chan, smac_idx;
1343 u16 rss_qid;
1344 u32 mtu;
1345 int step;
1346 int txq_idx;
1348 parent_ep = lookup_stid(t, stid);
1349 PDBG("%s parent ep %p tid %u\n", __func__, parent_ep, hwtid);
1351 get_4tuple(req, &local_ip, &peer_ip, &local_port, &peer_port);
1353 if (state_read(&parent_ep->com) != LISTEN) {
1354 printk(KERN_ERR "%s - listening ep not in LISTEN\n",
1355 __func__);
1356 goto reject;
1359 /* Find output route */
1360 rt = find_route(dev, local_ip, peer_ip, local_port, peer_port,
1361 GET_POPEN_TOS(ntohl(req->tos_stid)));
1362 if (!rt) {
1363 printk(KERN_ERR MOD "%s - failed to find dst entry!\n",
1364 __func__);
1365 goto reject;
1367 dst = &rt->dst;
1368 if (dst->neighbour->dev->flags & IFF_LOOPBACK) {
1369 pdev = ip_dev_find(&init_net, peer_ip);
1370 BUG_ON(!pdev);
1371 l2t = cxgb4_l2t_get(dev->rdev.lldi.l2t, dst->neighbour,
1372 pdev, 0);
1373 mtu = pdev->mtu;
1374 tx_chan = cxgb4_port_chan(pdev);
1375 smac_idx = tx_chan << 1;
1376 step = dev->rdev.lldi.ntxq / dev->rdev.lldi.nchan;
1377 txq_idx = cxgb4_port_idx(pdev) * step;
1378 step = dev->rdev.lldi.nrxq / dev->rdev.lldi.nchan;
1379 rss_qid = dev->rdev.lldi.rxq_ids[cxgb4_port_idx(pdev) * step];
1380 dev_put(pdev);
1381 } else {
1382 l2t = cxgb4_l2t_get(dev->rdev.lldi.l2t, dst->neighbour,
1383 dst->neighbour->dev, 0);
1384 mtu = dst_mtu(dst);
1385 tx_chan = cxgb4_port_chan(dst->neighbour->dev);
1386 smac_idx = tx_chan << 1;
1387 step = dev->rdev.lldi.ntxq / dev->rdev.lldi.nchan;
1388 txq_idx = cxgb4_port_idx(dst->neighbour->dev) * step;
1389 step = dev->rdev.lldi.nrxq / dev->rdev.lldi.nchan;
1390 rss_qid = dev->rdev.lldi.rxq_ids[
1391 cxgb4_port_idx(dst->neighbour->dev) * step];
1393 if (!l2t) {
1394 printk(KERN_ERR MOD "%s - failed to allocate l2t entry!\n",
1395 __func__);
1396 dst_release(dst);
1397 goto reject;
1400 child_ep = alloc_ep(sizeof(*child_ep), GFP_KERNEL);
1401 if (!child_ep) {
1402 printk(KERN_ERR MOD "%s - failed to allocate ep entry!\n",
1403 __func__);
1404 cxgb4_l2t_release(l2t);
1405 dst_release(dst);
1406 goto reject;
1408 state_set(&child_ep->com, CONNECTING);
1409 child_ep->com.dev = dev;
1410 child_ep->com.cm_id = NULL;
1411 child_ep->com.local_addr.sin_family = PF_INET;
1412 child_ep->com.local_addr.sin_port = local_port;
1413 child_ep->com.local_addr.sin_addr.s_addr = local_ip;
1414 child_ep->com.remote_addr.sin_family = PF_INET;
1415 child_ep->com.remote_addr.sin_port = peer_port;
1416 child_ep->com.remote_addr.sin_addr.s_addr = peer_ip;
1417 c4iw_get_ep(&parent_ep->com);
1418 child_ep->parent_ep = parent_ep;
1419 child_ep->tos = GET_POPEN_TOS(ntohl(req->tos_stid));
1420 child_ep->l2t = l2t;
1421 child_ep->dst = dst;
1422 child_ep->hwtid = hwtid;
1423 child_ep->tx_chan = tx_chan;
1424 child_ep->smac_idx = smac_idx;
1425 child_ep->rss_qid = rss_qid;
1426 child_ep->mtu = mtu;
1427 child_ep->txq_idx = txq_idx;
1429 PDBG("%s tx_chan %u smac_idx %u rss_qid %u\n", __func__,
1430 tx_chan, smac_idx, rss_qid);
1432 init_timer(&child_ep->timer);
1433 cxgb4_insert_tid(t, child_ep, hwtid);
1434 accept_cr(child_ep, peer_ip, skb, req);
1435 goto out;
1436 reject:
1437 reject_cr(dev, hwtid, peer_ip, skb);
1438 out:
1439 return 0;
1442 static int pass_establish(struct c4iw_dev *dev, struct sk_buff *skb)
1444 struct c4iw_ep *ep;
1445 struct cpl_pass_establish *req = cplhdr(skb);
1446 struct tid_info *t = dev->rdev.lldi.tids;
1447 unsigned int tid = GET_TID(req);
1449 ep = lookup_tid(t, tid);
1450 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1451 ep->snd_seq = be32_to_cpu(req->snd_isn);
1452 ep->rcv_seq = be32_to_cpu(req->rcv_isn);
1454 set_emss(ep, ntohs(req->tcp_opt));
1456 dst_confirm(ep->dst);
1457 state_set(&ep->com, MPA_REQ_WAIT);
1458 start_ep_timer(ep);
1459 send_flowc(ep, skb);
1461 return 0;
1464 static int peer_close(struct c4iw_dev *dev, struct sk_buff *skb)
1466 struct cpl_peer_close *hdr = cplhdr(skb);
1467 struct c4iw_ep *ep;
1468 struct c4iw_qp_attributes attrs;
1469 unsigned long flags;
1470 int disconnect = 1;
1471 int release = 0;
1472 int closing = 0;
1473 struct tid_info *t = dev->rdev.lldi.tids;
1474 unsigned int tid = GET_TID(hdr);
1475 int start_timer = 0;
1476 int stop_timer = 0;
1478 ep = lookup_tid(t, tid);
1479 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1480 dst_confirm(ep->dst);
1482 spin_lock_irqsave(&ep->com.lock, flags);
1483 switch (ep->com.state) {
1484 case MPA_REQ_WAIT:
1485 __state_set(&ep->com, CLOSING);
1486 break;
1487 case MPA_REQ_SENT:
1488 __state_set(&ep->com, CLOSING);
1489 connect_reply_upcall(ep, -ECONNRESET);
1490 break;
1491 case MPA_REQ_RCVD:
1494 * We're gonna mark this puppy DEAD, but keep
1495 * the reference on it until the ULP accepts or
1496 * rejects the CR. Also wake up anyone waiting
1497 * in rdma connection migration (see c4iw_accept_cr()).
1499 __state_set(&ep->com, CLOSING);
1500 ep->com.rpl_done = 1;
1501 ep->com.rpl_err = -ECONNRESET;
1502 PDBG("waking up ep %p tid %u\n", ep, ep->hwtid);
1503 wake_up(&ep->com.waitq);
1504 break;
1505 case MPA_REP_SENT:
1506 __state_set(&ep->com, CLOSING);
1507 ep->com.rpl_done = 1;
1508 ep->com.rpl_err = -ECONNRESET;
1509 PDBG("waking up ep %p tid %u\n", ep, ep->hwtid);
1510 wake_up(&ep->com.waitq);
1511 break;
1512 case FPDU_MODE:
1513 start_timer = 1;
1514 __state_set(&ep->com, CLOSING);
1515 closing = 1;
1516 peer_close_upcall(ep);
1517 break;
1518 case ABORTING:
1519 disconnect = 0;
1520 break;
1521 case CLOSING:
1522 __state_set(&ep->com, MORIBUND);
1523 disconnect = 0;
1524 break;
1525 case MORIBUND:
1526 stop_timer = 1;
1527 if (ep->com.cm_id && ep->com.qp) {
1528 attrs.next_state = C4IW_QP_STATE_IDLE;
1529 c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
1530 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
1532 close_complete_upcall(ep);
1533 __state_set(&ep->com, DEAD);
1534 release = 1;
1535 disconnect = 0;
1536 break;
1537 case DEAD:
1538 disconnect = 0;
1539 break;
1540 default:
1541 BUG_ON(1);
1543 spin_unlock_irqrestore(&ep->com.lock, flags);
1544 if (closing) {
1545 attrs.next_state = C4IW_QP_STATE_CLOSING;
1546 c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
1547 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
1549 if (start_timer)
1550 start_ep_timer(ep);
1551 if (stop_timer)
1552 stop_ep_timer(ep);
1553 if (disconnect)
1554 c4iw_ep_disconnect(ep, 0, GFP_KERNEL);
1555 if (release)
1556 release_ep_resources(ep);
1557 return 0;
1561 * Returns whether an ABORT_REQ_RSS message is a negative advice.
1563 static int is_neg_adv_abort(unsigned int status)
1565 return status == CPL_ERR_RTX_NEG_ADVICE ||
1566 status == CPL_ERR_PERSIST_NEG_ADVICE;
1569 static int peer_abort(struct c4iw_dev *dev, struct sk_buff *skb)
1571 struct cpl_abort_req_rss *req = cplhdr(skb);
1572 struct c4iw_ep *ep;
1573 struct cpl_abort_rpl *rpl;
1574 struct sk_buff *rpl_skb;
1575 struct c4iw_qp_attributes attrs;
1576 int ret;
1577 int release = 0;
1578 unsigned long flags;
1579 struct tid_info *t = dev->rdev.lldi.tids;
1580 unsigned int tid = GET_TID(req);
1581 int stop_timer = 0;
1583 ep = lookup_tid(t, tid);
1584 if (is_neg_adv_abort(req->status)) {
1585 PDBG("%s neg_adv_abort ep %p tid %u\n", __func__, ep,
1586 ep->hwtid);
1587 return 0;
1589 spin_lock_irqsave(&ep->com.lock, flags);
1590 PDBG("%s ep %p tid %u state %u\n", __func__, ep, ep->hwtid,
1591 ep->com.state);
1592 switch (ep->com.state) {
1593 case CONNECTING:
1594 break;
1595 case MPA_REQ_WAIT:
1596 stop_timer = 1;
1597 break;
1598 case MPA_REQ_SENT:
1599 stop_timer = 1;
1600 connect_reply_upcall(ep, -ECONNRESET);
1601 break;
1602 case MPA_REP_SENT:
1603 ep->com.rpl_done = 1;
1604 ep->com.rpl_err = -ECONNRESET;
1605 PDBG("waking up ep %p\n", ep);
1606 wake_up(&ep->com.waitq);
1607 break;
1608 case MPA_REQ_RCVD:
1611 * We're gonna mark this puppy DEAD, but keep
1612 * the reference on it until the ULP accepts or
1613 * rejects the CR. Also wake up anyone waiting
1614 * in rdma connection migration (see c4iw_accept_cr()).
1616 ep->com.rpl_done = 1;
1617 ep->com.rpl_err = -ECONNRESET;
1618 PDBG("waking up ep %p tid %u\n", ep, ep->hwtid);
1619 wake_up(&ep->com.waitq);
1620 break;
1621 case MORIBUND:
1622 case CLOSING:
1623 stop_timer = 1;
1624 /*FALLTHROUGH*/
1625 case FPDU_MODE:
1626 if (ep->com.cm_id && ep->com.qp) {
1627 attrs.next_state = C4IW_QP_STATE_ERROR;
1628 ret = c4iw_modify_qp(ep->com.qp->rhp,
1629 ep->com.qp, C4IW_QP_ATTR_NEXT_STATE,
1630 &attrs, 1);
1631 if (ret)
1632 printk(KERN_ERR MOD
1633 "%s - qp <- error failed!\n",
1634 __func__);
1636 peer_abort_upcall(ep);
1637 break;
1638 case ABORTING:
1639 break;
1640 case DEAD:
1641 PDBG("%s PEER_ABORT IN DEAD STATE!!!!\n", __func__);
1642 spin_unlock_irqrestore(&ep->com.lock, flags);
1643 return 0;
1644 default:
1645 BUG_ON(1);
1646 break;
1648 dst_confirm(ep->dst);
1649 if (ep->com.state != ABORTING) {
1650 __state_set(&ep->com, DEAD);
1651 release = 1;
1653 spin_unlock_irqrestore(&ep->com.lock, flags);
1655 rpl_skb = get_skb(skb, sizeof(*rpl), GFP_KERNEL);
1656 if (!rpl_skb) {
1657 printk(KERN_ERR MOD "%s - cannot allocate skb!\n",
1658 __func__);
1659 release = 1;
1660 goto out;
1662 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
1663 rpl = (struct cpl_abort_rpl *) skb_put(rpl_skb, sizeof(*rpl));
1664 INIT_TP_WR(rpl, ep->hwtid);
1665 OPCODE_TID(rpl) = cpu_to_be32(MK_OPCODE_TID(CPL_ABORT_RPL, ep->hwtid));
1666 rpl->cmd = CPL_ABORT_NO_RST;
1667 c4iw_ofld_send(&ep->com.dev->rdev, rpl_skb);
1668 out:
1669 if (stop_timer)
1670 stop_ep_timer(ep);
1671 if (release)
1672 release_ep_resources(ep);
1673 return 0;
1676 static int close_con_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
1678 struct c4iw_ep *ep;
1679 struct c4iw_qp_attributes attrs;
1680 struct cpl_close_con_rpl *rpl = cplhdr(skb);
1681 unsigned long flags;
1682 int release = 0;
1683 struct tid_info *t = dev->rdev.lldi.tids;
1684 unsigned int tid = GET_TID(rpl);
1685 int stop_timer = 0;
1687 ep = lookup_tid(t, tid);
1689 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1690 BUG_ON(!ep);
1692 /* The cm_id may be null if we failed to connect */
1693 spin_lock_irqsave(&ep->com.lock, flags);
1694 switch (ep->com.state) {
1695 case CLOSING:
1696 __state_set(&ep->com, MORIBUND);
1697 break;
1698 case MORIBUND:
1699 stop_timer = 1;
1700 if ((ep->com.cm_id) && (ep->com.qp)) {
1701 attrs.next_state = C4IW_QP_STATE_IDLE;
1702 c4iw_modify_qp(ep->com.qp->rhp,
1703 ep->com.qp,
1704 C4IW_QP_ATTR_NEXT_STATE,
1705 &attrs, 1);
1707 close_complete_upcall(ep);
1708 __state_set(&ep->com, DEAD);
1709 release = 1;
1710 break;
1711 case ABORTING:
1712 case DEAD:
1713 break;
1714 default:
1715 BUG_ON(1);
1716 break;
1718 spin_unlock_irqrestore(&ep->com.lock, flags);
1719 if (stop_timer)
1720 stop_ep_timer(ep);
1721 if (release)
1722 release_ep_resources(ep);
1723 return 0;
1726 static int terminate(struct c4iw_dev *dev, struct sk_buff *skb)
1728 struct c4iw_ep *ep;
1729 struct cpl_rdma_terminate *term = cplhdr(skb);
1730 struct tid_info *t = dev->rdev.lldi.tids;
1731 unsigned int tid = GET_TID(term);
1733 ep = lookup_tid(t, tid);
1735 if (state_read(&ep->com) != FPDU_MODE)
1736 return 0;
1738 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1739 skb_pull(skb, sizeof *term);
1740 PDBG("%s saving %d bytes of term msg\n", __func__, skb->len);
1741 skb_copy_from_linear_data(skb, ep->com.qp->attr.terminate_buffer,
1742 skb->len);
1743 ep->com.qp->attr.terminate_msg_len = skb->len;
1744 ep->com.qp->attr.is_terminate_local = 0;
1745 return 0;
1749 * Upcall from the adapter indicating data has been transmitted.
1750 * For us its just the single MPA request or reply. We can now free
1751 * the skb holding the mpa message.
1753 static int fw4_ack(struct c4iw_dev *dev, struct sk_buff *skb)
1755 struct c4iw_ep *ep;
1756 struct cpl_fw4_ack *hdr = cplhdr(skb);
1757 u8 credits = hdr->credits;
1758 unsigned int tid = GET_TID(hdr);
1759 struct tid_info *t = dev->rdev.lldi.tids;
1762 ep = lookup_tid(t, tid);
1763 PDBG("%s ep %p tid %u credits %u\n", __func__, ep, ep->hwtid, credits);
1764 if (credits == 0) {
1765 PDBG(KERN_ERR "%s 0 credit ack ep %p tid %u state %u\n",
1766 __func__, ep, ep->hwtid, state_read(&ep->com));
1767 return 0;
1770 dst_confirm(ep->dst);
1771 if (ep->mpa_skb) {
1772 PDBG("%s last streaming msg ack ep %p tid %u state %u "
1773 "initiator %u freeing skb\n", __func__, ep, ep->hwtid,
1774 state_read(&ep->com), ep->mpa_attr.initiator ? 1 : 0);
1775 kfree_skb(ep->mpa_skb);
1776 ep->mpa_skb = NULL;
1778 return 0;
1781 int c4iw_reject_cr(struct iw_cm_id *cm_id, const void *pdata, u8 pdata_len)
1783 int err;
1784 struct c4iw_ep *ep = to_ep(cm_id);
1785 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1787 if (state_read(&ep->com) == DEAD) {
1788 c4iw_put_ep(&ep->com);
1789 return -ECONNRESET;
1791 BUG_ON(state_read(&ep->com) != MPA_REQ_RCVD);
1792 if (mpa_rev == 0)
1793 abort_connection(ep, NULL, GFP_KERNEL);
1794 else {
1795 err = send_mpa_reject(ep, pdata, pdata_len);
1796 err = c4iw_ep_disconnect(ep, 0, GFP_KERNEL);
1798 c4iw_put_ep(&ep->com);
1799 return 0;
1802 int c4iw_accept_cr(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
1804 int err;
1805 struct c4iw_qp_attributes attrs;
1806 enum c4iw_qp_attr_mask mask;
1807 struct c4iw_ep *ep = to_ep(cm_id);
1808 struct c4iw_dev *h = to_c4iw_dev(cm_id->device);
1809 struct c4iw_qp *qp = get_qhp(h, conn_param->qpn);
1811 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1812 if (state_read(&ep->com) == DEAD) {
1813 err = -ECONNRESET;
1814 goto err;
1817 BUG_ON(state_read(&ep->com) != MPA_REQ_RCVD);
1818 BUG_ON(!qp);
1820 if ((conn_param->ord > c4iw_max_read_depth) ||
1821 (conn_param->ird > c4iw_max_read_depth)) {
1822 abort_connection(ep, NULL, GFP_KERNEL);
1823 err = -EINVAL;
1824 goto err;
1827 cm_id->add_ref(cm_id);
1828 ep->com.cm_id = cm_id;
1829 ep->com.qp = qp;
1831 ep->ird = conn_param->ird;
1832 ep->ord = conn_param->ord;
1834 if (peer2peer && ep->ird == 0)
1835 ep->ird = 1;
1837 PDBG("%s %d ird %d ord %d\n", __func__, __LINE__, ep->ird, ep->ord);
1839 /* bind QP to EP and move to RTS */
1840 attrs.mpa_attr = ep->mpa_attr;
1841 attrs.max_ird = ep->ird;
1842 attrs.max_ord = ep->ord;
1843 attrs.llp_stream_handle = ep;
1844 attrs.next_state = C4IW_QP_STATE_RTS;
1846 /* bind QP and TID with INIT_WR */
1847 mask = C4IW_QP_ATTR_NEXT_STATE |
1848 C4IW_QP_ATTR_LLP_STREAM_HANDLE |
1849 C4IW_QP_ATTR_MPA_ATTR |
1850 C4IW_QP_ATTR_MAX_IRD |
1851 C4IW_QP_ATTR_MAX_ORD;
1853 err = c4iw_modify_qp(ep->com.qp->rhp,
1854 ep->com.qp, mask, &attrs, 1);
1855 if (err)
1856 goto err1;
1857 err = send_mpa_reply(ep, conn_param->private_data,
1858 conn_param->private_data_len);
1859 if (err)
1860 goto err1;
1862 state_set(&ep->com, FPDU_MODE);
1863 established_upcall(ep);
1864 c4iw_put_ep(&ep->com);
1865 return 0;
1866 err1:
1867 ep->com.cm_id = NULL;
1868 ep->com.qp = NULL;
1869 cm_id->rem_ref(cm_id);
1870 err:
1871 c4iw_put_ep(&ep->com);
1872 return err;
1875 int c4iw_connect(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
1877 int err = 0;
1878 struct c4iw_dev *dev = to_c4iw_dev(cm_id->device);
1879 struct c4iw_ep *ep;
1880 struct rtable *rt;
1881 struct net_device *pdev;
1882 int step;
1884 if ((conn_param->ord > c4iw_max_read_depth) ||
1885 (conn_param->ird > c4iw_max_read_depth)) {
1886 err = -EINVAL;
1887 goto out;
1889 ep = alloc_ep(sizeof(*ep), GFP_KERNEL);
1890 if (!ep) {
1891 printk(KERN_ERR MOD "%s - cannot alloc ep.\n", __func__);
1892 err = -ENOMEM;
1893 goto out;
1895 init_timer(&ep->timer);
1896 ep->plen = conn_param->private_data_len;
1897 if (ep->plen)
1898 memcpy(ep->mpa_pkt + sizeof(struct mpa_message),
1899 conn_param->private_data, ep->plen);
1900 ep->ird = conn_param->ird;
1901 ep->ord = conn_param->ord;
1903 if (peer2peer && ep->ord == 0)
1904 ep->ord = 1;
1906 cm_id->add_ref(cm_id);
1907 ep->com.dev = dev;
1908 ep->com.cm_id = cm_id;
1909 ep->com.qp = get_qhp(dev, conn_param->qpn);
1910 BUG_ON(!ep->com.qp);
1911 PDBG("%s qpn 0x%x qp %p cm_id %p\n", __func__, conn_param->qpn,
1912 ep->com.qp, cm_id);
1915 * Allocate an active TID to initiate a TCP connection.
1917 ep->atid = cxgb4_alloc_atid(dev->rdev.lldi.tids, ep);
1918 if (ep->atid == -1) {
1919 printk(KERN_ERR MOD "%s - cannot alloc atid.\n", __func__);
1920 err = -ENOMEM;
1921 goto fail2;
1924 PDBG("%s saddr 0x%x sport 0x%x raddr 0x%x rport 0x%x\n", __func__,
1925 ntohl(cm_id->local_addr.sin_addr.s_addr),
1926 ntohs(cm_id->local_addr.sin_port),
1927 ntohl(cm_id->remote_addr.sin_addr.s_addr),
1928 ntohs(cm_id->remote_addr.sin_port));
1930 /* find a route */
1931 rt = find_route(dev,
1932 cm_id->local_addr.sin_addr.s_addr,
1933 cm_id->remote_addr.sin_addr.s_addr,
1934 cm_id->local_addr.sin_port,
1935 cm_id->remote_addr.sin_port, 0);
1936 if (!rt) {
1937 printk(KERN_ERR MOD "%s - cannot find route.\n", __func__);
1938 err = -EHOSTUNREACH;
1939 goto fail3;
1941 ep->dst = &rt->dst;
1943 /* get a l2t entry */
1944 if (ep->dst->neighbour->dev->flags & IFF_LOOPBACK) {
1945 PDBG("%s LOOPBACK\n", __func__);
1946 pdev = ip_dev_find(&init_net,
1947 cm_id->remote_addr.sin_addr.s_addr);
1948 ep->l2t = cxgb4_l2t_get(ep->com.dev->rdev.lldi.l2t,
1949 ep->dst->neighbour,
1950 pdev, 0);
1951 ep->mtu = pdev->mtu;
1952 ep->tx_chan = cxgb4_port_chan(pdev);
1953 ep->smac_idx = ep->tx_chan << 1;
1954 step = ep->com.dev->rdev.lldi.ntxq /
1955 ep->com.dev->rdev.lldi.nchan;
1956 ep->txq_idx = cxgb4_port_idx(pdev) * step;
1957 step = ep->com.dev->rdev.lldi.nrxq /
1958 ep->com.dev->rdev.lldi.nchan;
1959 ep->rss_qid = ep->com.dev->rdev.lldi.rxq_ids[
1960 cxgb4_port_idx(pdev) * step];
1961 dev_put(pdev);
1962 } else {
1963 ep->l2t = cxgb4_l2t_get(ep->com.dev->rdev.lldi.l2t,
1964 ep->dst->neighbour,
1965 ep->dst->neighbour->dev, 0);
1966 ep->mtu = dst_mtu(ep->dst);
1967 ep->tx_chan = cxgb4_port_chan(ep->dst->neighbour->dev);
1968 ep->smac_idx = ep->tx_chan << 1;
1969 step = ep->com.dev->rdev.lldi.ntxq /
1970 ep->com.dev->rdev.lldi.nchan;
1971 ep->txq_idx = cxgb4_port_idx(ep->dst->neighbour->dev) * step;
1972 step = ep->com.dev->rdev.lldi.nrxq /
1973 ep->com.dev->rdev.lldi.nchan;
1974 ep->rss_qid = ep->com.dev->rdev.lldi.rxq_ids[
1975 cxgb4_port_idx(ep->dst->neighbour->dev) * step];
1977 if (!ep->l2t) {
1978 printk(KERN_ERR MOD "%s - cannot alloc l2e.\n", __func__);
1979 err = -ENOMEM;
1980 goto fail4;
1983 PDBG("%s txq_idx %u tx_chan %u smac_idx %u rss_qid %u l2t_idx %u\n",
1984 __func__, ep->txq_idx, ep->tx_chan, ep->smac_idx, ep->rss_qid,
1985 ep->l2t->idx);
1987 state_set(&ep->com, CONNECTING);
1988 ep->tos = 0;
1989 ep->com.local_addr = cm_id->local_addr;
1990 ep->com.remote_addr = cm_id->remote_addr;
1992 /* send connect request to rnic */
1993 err = send_connect(ep);
1994 if (!err)
1995 goto out;
1997 cxgb4_l2t_release(ep->l2t);
1998 fail4:
1999 dst_release(ep->dst);
2000 fail3:
2001 cxgb4_free_atid(ep->com.dev->rdev.lldi.tids, ep->atid);
2002 fail2:
2003 cm_id->rem_ref(cm_id);
2004 c4iw_put_ep(&ep->com);
2005 out:
2006 return err;
2009 int c4iw_create_listen(struct iw_cm_id *cm_id, int backlog)
2011 int err = 0;
2012 struct c4iw_dev *dev = to_c4iw_dev(cm_id->device);
2013 struct c4iw_listen_ep *ep;
2016 might_sleep();
2018 ep = alloc_ep(sizeof(*ep), GFP_KERNEL);
2019 if (!ep) {
2020 printk(KERN_ERR MOD "%s - cannot alloc ep.\n", __func__);
2021 err = -ENOMEM;
2022 goto fail1;
2024 PDBG("%s ep %p\n", __func__, ep);
2025 cm_id->add_ref(cm_id);
2026 ep->com.cm_id = cm_id;
2027 ep->com.dev = dev;
2028 ep->backlog = backlog;
2029 ep->com.local_addr = cm_id->local_addr;
2032 * Allocate a server TID.
2034 ep->stid = cxgb4_alloc_stid(dev->rdev.lldi.tids, PF_INET, ep);
2035 if (ep->stid == -1) {
2036 printk(KERN_ERR MOD "%s - cannot alloc stid.\n", __func__);
2037 err = -ENOMEM;
2038 goto fail2;
2041 state_set(&ep->com, LISTEN);
2042 err = cxgb4_create_server(ep->com.dev->rdev.lldi.ports[0], ep->stid,
2043 ep->com.local_addr.sin_addr.s_addr,
2044 ep->com.local_addr.sin_port,
2045 ep->com.dev->rdev.lldi.rxq_ids[0]);
2046 if (err)
2047 goto fail3;
2049 /* wait for pass_open_rpl */
2050 wait_event(ep->com.waitq, ep->com.rpl_done);
2051 err = ep->com.rpl_err;
2052 if (!err) {
2053 cm_id->provider_data = ep;
2054 goto out;
2056 fail3:
2057 cxgb4_free_stid(ep->com.dev->rdev.lldi.tids, ep->stid, PF_INET);
2058 fail2:
2059 cm_id->rem_ref(cm_id);
2060 c4iw_put_ep(&ep->com);
2061 fail1:
2062 out:
2063 return err;
2066 int c4iw_destroy_listen(struct iw_cm_id *cm_id)
2068 int err;
2069 struct c4iw_listen_ep *ep = to_listen_ep(cm_id);
2071 PDBG("%s ep %p\n", __func__, ep);
2073 might_sleep();
2074 state_set(&ep->com, DEAD);
2075 ep->com.rpl_done = 0;
2076 ep->com.rpl_err = 0;
2077 err = listen_stop(ep);
2078 if (err)
2079 goto done;
2080 wait_event(ep->com.waitq, ep->com.rpl_done);
2081 cxgb4_free_stid(ep->com.dev->rdev.lldi.tids, ep->stid, PF_INET);
2082 done:
2083 err = ep->com.rpl_err;
2084 cm_id->rem_ref(cm_id);
2085 c4iw_put_ep(&ep->com);
2086 return err;
2089 int c4iw_ep_disconnect(struct c4iw_ep *ep, int abrupt, gfp_t gfp)
2091 int ret = 0;
2092 unsigned long flags;
2093 int close = 0;
2094 int fatal = 0;
2095 struct c4iw_rdev *rdev;
2096 int start_timer = 0;
2097 int stop_timer = 0;
2099 spin_lock_irqsave(&ep->com.lock, flags);
2101 PDBG("%s ep %p state %s, abrupt %d\n", __func__, ep,
2102 states[ep->com.state], abrupt);
2104 rdev = &ep->com.dev->rdev;
2105 if (c4iw_fatal_error(rdev)) {
2106 fatal = 1;
2107 close_complete_upcall(ep);
2108 ep->com.state = DEAD;
2110 switch (ep->com.state) {
2111 case MPA_REQ_WAIT:
2112 case MPA_REQ_SENT:
2113 case MPA_REQ_RCVD:
2114 case MPA_REP_SENT:
2115 case FPDU_MODE:
2116 close = 1;
2117 if (abrupt)
2118 ep->com.state = ABORTING;
2119 else {
2120 ep->com.state = CLOSING;
2121 start_timer = 1;
2123 set_bit(CLOSE_SENT, &ep->com.flags);
2124 break;
2125 case CLOSING:
2126 if (!test_and_set_bit(CLOSE_SENT, &ep->com.flags)) {
2127 close = 1;
2128 if (abrupt) {
2129 stop_timer = 1;
2130 ep->com.state = ABORTING;
2131 } else
2132 ep->com.state = MORIBUND;
2134 break;
2135 case MORIBUND:
2136 case ABORTING:
2137 case DEAD:
2138 PDBG("%s ignoring disconnect ep %p state %u\n",
2139 __func__, ep, ep->com.state);
2140 break;
2141 default:
2142 BUG();
2143 break;
2146 spin_unlock_irqrestore(&ep->com.lock, flags);
2147 if (start_timer)
2148 start_ep_timer(ep);
2149 if (stop_timer)
2150 stop_ep_timer(ep);
2151 if (close) {
2152 if (abrupt)
2153 ret = abort_connection(ep, NULL, gfp);
2154 else
2155 ret = send_halfclose(ep, gfp);
2156 if (ret)
2157 fatal = 1;
2159 if (fatal)
2160 release_ep_resources(ep);
2161 return ret;
2165 * These are the real handlers that are called from a
2166 * work queue.
2168 static c4iw_handler_func work_handlers[NUM_CPL_CMDS] = {
2169 [CPL_ACT_ESTABLISH] = act_establish,
2170 [CPL_ACT_OPEN_RPL] = act_open_rpl,
2171 [CPL_RX_DATA] = rx_data,
2172 [CPL_ABORT_RPL_RSS] = abort_rpl,
2173 [CPL_ABORT_RPL] = abort_rpl,
2174 [CPL_PASS_OPEN_RPL] = pass_open_rpl,
2175 [CPL_CLOSE_LISTSRV_RPL] = close_listsrv_rpl,
2176 [CPL_PASS_ACCEPT_REQ] = pass_accept_req,
2177 [CPL_PASS_ESTABLISH] = pass_establish,
2178 [CPL_PEER_CLOSE] = peer_close,
2179 [CPL_ABORT_REQ_RSS] = peer_abort,
2180 [CPL_CLOSE_CON_RPL] = close_con_rpl,
2181 [CPL_RDMA_TERMINATE] = terminate,
2182 [CPL_FW4_ACK] = fw4_ack
2185 static void process_timeout(struct c4iw_ep *ep)
2187 struct c4iw_qp_attributes attrs;
2188 int abort = 1;
2190 spin_lock_irq(&ep->com.lock);
2191 PDBG("%s ep %p tid %u state %d\n", __func__, ep, ep->hwtid,
2192 ep->com.state);
2193 switch (ep->com.state) {
2194 case MPA_REQ_SENT:
2195 __state_set(&ep->com, ABORTING);
2196 connect_reply_upcall(ep, -ETIMEDOUT);
2197 break;
2198 case MPA_REQ_WAIT:
2199 __state_set(&ep->com, ABORTING);
2200 break;
2201 case CLOSING:
2202 case MORIBUND:
2203 if (ep->com.cm_id && ep->com.qp) {
2204 attrs.next_state = C4IW_QP_STATE_ERROR;
2205 c4iw_modify_qp(ep->com.qp->rhp,
2206 ep->com.qp, C4IW_QP_ATTR_NEXT_STATE,
2207 &attrs, 1);
2209 __state_set(&ep->com, ABORTING);
2210 break;
2211 default:
2212 printk(KERN_ERR "%s unexpected state ep %p tid %u state %u\n",
2213 __func__, ep, ep->hwtid, ep->com.state);
2214 WARN_ON(1);
2215 abort = 0;
2217 spin_unlock_irq(&ep->com.lock);
2218 if (abort)
2219 abort_connection(ep, NULL, GFP_KERNEL);
2220 c4iw_put_ep(&ep->com);
2223 static void process_timedout_eps(void)
2225 struct c4iw_ep *ep;
2227 spin_lock_irq(&timeout_lock);
2228 while (!list_empty(&timeout_list)) {
2229 struct list_head *tmp;
2231 tmp = timeout_list.next;
2232 list_del(tmp);
2233 spin_unlock_irq(&timeout_lock);
2234 ep = list_entry(tmp, struct c4iw_ep, entry);
2235 process_timeout(ep);
2236 spin_lock_irq(&timeout_lock);
2238 spin_unlock_irq(&timeout_lock);
2241 static void process_work(struct work_struct *work)
2243 struct sk_buff *skb = NULL;
2244 struct c4iw_dev *dev;
2245 struct cpl_act_establish *rpl = cplhdr(skb);
2246 unsigned int opcode;
2247 int ret;
2249 while ((skb = skb_dequeue(&rxq))) {
2250 rpl = cplhdr(skb);
2251 dev = *((struct c4iw_dev **) (skb->cb + sizeof(void *)));
2252 opcode = rpl->ot.opcode;
2254 BUG_ON(!work_handlers[opcode]);
2255 ret = work_handlers[opcode](dev, skb);
2256 if (!ret)
2257 kfree_skb(skb);
2259 process_timedout_eps();
2262 static DECLARE_WORK(skb_work, process_work);
2264 static void ep_timeout(unsigned long arg)
2266 struct c4iw_ep *ep = (struct c4iw_ep *)arg;
2268 spin_lock(&timeout_lock);
2269 list_add_tail(&ep->entry, &timeout_list);
2270 spin_unlock(&timeout_lock);
2271 queue_work(workq, &skb_work);
2275 * All the CM events are handled on a work queue to have a safe context.
2277 static int sched(struct c4iw_dev *dev, struct sk_buff *skb)
2281 * Save dev in the skb->cb area.
2283 *((struct c4iw_dev **) (skb->cb + sizeof(void *))) = dev;
2286 * Queue the skb and schedule the worker thread.
2288 skb_queue_tail(&rxq, skb);
2289 queue_work(workq, &skb_work);
2290 return 0;
2293 static int set_tcb_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
2295 struct cpl_set_tcb_rpl *rpl = cplhdr(skb);
2297 if (rpl->status != CPL_ERR_NONE) {
2298 printk(KERN_ERR MOD "Unexpected SET_TCB_RPL status %u "
2299 "for tid %u\n", rpl->status, GET_TID(rpl));
2301 return 0;
2304 static int fw6_msg(struct c4iw_dev *dev, struct sk_buff *skb)
2306 struct cpl_fw6_msg *rpl = cplhdr(skb);
2307 struct c4iw_wr_wait *wr_waitp;
2308 int ret;
2310 PDBG("%s type %u\n", __func__, rpl->type);
2312 switch (rpl->type) {
2313 case 1:
2314 ret = (int)((be64_to_cpu(rpl->data[0]) >> 8) & 0xff);
2315 wr_waitp = (__force struct c4iw_wr_wait *)rpl->data[1];
2316 PDBG("%s wr_waitp %p ret %u\n", __func__, wr_waitp, ret);
2317 if (wr_waitp) {
2318 wr_waitp->ret = ret;
2319 wr_waitp->done = 1;
2320 wake_up(&wr_waitp->wait);
2322 break;
2323 case 2:
2324 c4iw_ev_dispatch(dev, (struct t4_cqe *)&rpl->data[0]);
2325 break;
2326 default:
2327 printk(KERN_ERR MOD "%s unexpected fw6 msg type %u\n", __func__,
2328 rpl->type);
2329 break;
2331 return 0;
2335 * Most upcalls from the T4 Core go to sched() to
2336 * schedule the processing on a work queue.
2338 c4iw_handler_func c4iw_handlers[NUM_CPL_CMDS] = {
2339 [CPL_ACT_ESTABLISH] = sched,
2340 [CPL_ACT_OPEN_RPL] = sched,
2341 [CPL_RX_DATA] = sched,
2342 [CPL_ABORT_RPL_RSS] = sched,
2343 [CPL_ABORT_RPL] = sched,
2344 [CPL_PASS_OPEN_RPL] = sched,
2345 [CPL_CLOSE_LISTSRV_RPL] = sched,
2346 [CPL_PASS_ACCEPT_REQ] = sched,
2347 [CPL_PASS_ESTABLISH] = sched,
2348 [CPL_PEER_CLOSE] = sched,
2349 [CPL_CLOSE_CON_RPL] = sched,
2350 [CPL_ABORT_REQ_RSS] = sched,
2351 [CPL_RDMA_TERMINATE] = sched,
2352 [CPL_FW4_ACK] = sched,
2353 [CPL_SET_TCB_RPL] = set_tcb_rpl,
2354 [CPL_FW6_MSG] = fw6_msg
2357 int __init c4iw_cm_init(void)
2359 spin_lock_init(&timeout_lock);
2360 skb_queue_head_init(&rxq);
2362 workq = create_singlethread_workqueue("iw_cxgb4");
2363 if (!workq)
2364 return -ENOMEM;
2366 return 0;
2369 void __exit c4iw_cm_term(void)
2371 WARN_ON(!list_empty(&timeout_list));
2372 flush_workqueue(workq);
2373 destroy_workqueue(workq);