Merge git://git.kernel.org/pub/scm/linux/kernel/git/davem/net-next-2.6
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / drivers / infiniband / hw / cxgb4 / cm.c
blobf660cd04ec2f31774ed678b16dbb85d3a8ed6b39
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 static int dack_mode = 1;
65 module_param(dack_mode, int, 0644);
66 MODULE_PARM_DESC(dack_mode, "Delayed ack mode (default=1)");
68 int c4iw_max_read_depth = 8;
69 module_param(c4iw_max_read_depth, int, 0644);
70 MODULE_PARM_DESC(c4iw_max_read_depth, "Per-connection max ORD/IRD (default=8)");
72 static int enable_tcp_timestamps;
73 module_param(enable_tcp_timestamps, int, 0644);
74 MODULE_PARM_DESC(enable_tcp_timestamps, "Enable tcp timestamps (default=0)");
76 static int enable_tcp_sack;
77 module_param(enable_tcp_sack, int, 0644);
78 MODULE_PARM_DESC(enable_tcp_sack, "Enable tcp SACK (default=0)");
80 static int enable_tcp_window_scaling = 1;
81 module_param(enable_tcp_window_scaling, int, 0644);
82 MODULE_PARM_DESC(enable_tcp_window_scaling,
83 "Enable tcp window scaling (default=1)");
85 int c4iw_debug;
86 module_param(c4iw_debug, int, 0644);
87 MODULE_PARM_DESC(c4iw_debug, "Enable debug logging (default=0)");
89 static int peer2peer;
90 module_param(peer2peer, int, 0644);
91 MODULE_PARM_DESC(peer2peer, "Support peer2peer ULPs (default=0)");
93 static int p2p_type = FW_RI_INIT_P2PTYPE_READ_REQ;
94 module_param(p2p_type, int, 0644);
95 MODULE_PARM_DESC(p2p_type, "RDMAP opcode to use for the RTR message: "
96 "1=RDMA_READ 0=RDMA_WRITE (default 1)");
98 static int ep_timeout_secs = 60;
99 module_param(ep_timeout_secs, int, 0644);
100 MODULE_PARM_DESC(ep_timeout_secs, "CM Endpoint operation timeout "
101 "in seconds (default=60)");
103 static int mpa_rev = 1;
104 module_param(mpa_rev, int, 0644);
105 MODULE_PARM_DESC(mpa_rev, "MPA Revision, 0 supports amso1100, "
106 "1 is spec compliant. (default=1)");
108 static int markers_enabled;
109 module_param(markers_enabled, int, 0644);
110 MODULE_PARM_DESC(markers_enabled, "Enable MPA MARKERS (default(0)=disabled)");
112 static int crc_enabled = 1;
113 module_param(crc_enabled, int, 0644);
114 MODULE_PARM_DESC(crc_enabled, "Enable MPA CRC (default(1)=enabled)");
116 static int rcv_win = 256 * 1024;
117 module_param(rcv_win, int, 0644);
118 MODULE_PARM_DESC(rcv_win, "TCP receive window in bytes (default=256KB)");
120 static int snd_win = 128 * 1024;
121 module_param(snd_win, int, 0644);
122 MODULE_PARM_DESC(snd_win, "TCP send window in bytes (default=128KB)");
124 static struct workqueue_struct *workq;
126 static struct sk_buff_head rxq;
128 static struct sk_buff *get_skb(struct sk_buff *skb, int len, gfp_t gfp);
129 static void ep_timeout(unsigned long arg);
130 static void connect_reply_upcall(struct c4iw_ep *ep, int status);
132 static LIST_HEAD(timeout_list);
133 static spinlock_t timeout_lock;
135 static void start_ep_timer(struct c4iw_ep *ep)
137 PDBG("%s ep %p\n", __func__, ep);
138 if (timer_pending(&ep->timer)) {
139 PDBG("%s stopped / restarted timer ep %p\n", __func__, ep);
140 del_timer_sync(&ep->timer);
141 } else
142 c4iw_get_ep(&ep->com);
143 ep->timer.expires = jiffies + ep_timeout_secs * HZ;
144 ep->timer.data = (unsigned long)ep;
145 ep->timer.function = ep_timeout;
146 add_timer(&ep->timer);
149 static void stop_ep_timer(struct c4iw_ep *ep)
151 PDBG("%s ep %p\n", __func__, ep);
152 if (!timer_pending(&ep->timer)) {
153 printk(KERN_ERR "%s timer stopped when its not running! "
154 "ep %p state %u\n", __func__, ep, ep->com.state);
155 WARN_ON(1);
156 return;
158 del_timer_sync(&ep->timer);
159 c4iw_put_ep(&ep->com);
162 static int c4iw_l2t_send(struct c4iw_rdev *rdev, struct sk_buff *skb,
163 struct l2t_entry *l2e)
165 int error = 0;
167 if (c4iw_fatal_error(rdev)) {
168 kfree_skb(skb);
169 PDBG("%s - device in error state - dropping\n", __func__);
170 return -EIO;
172 error = cxgb4_l2t_send(rdev->lldi.ports[0], skb, l2e);
173 if (error < 0)
174 kfree_skb(skb);
175 return error < 0 ? error : 0;
178 int c4iw_ofld_send(struct c4iw_rdev *rdev, struct sk_buff *skb)
180 int error = 0;
182 if (c4iw_fatal_error(rdev)) {
183 kfree_skb(skb);
184 PDBG("%s - device in error state - dropping\n", __func__);
185 return -EIO;
187 error = cxgb4_ofld_send(rdev->lldi.ports[0], skb);
188 if (error < 0)
189 kfree_skb(skb);
190 return error < 0 ? error : 0;
193 static void release_tid(struct c4iw_rdev *rdev, u32 hwtid, struct sk_buff *skb)
195 struct cpl_tid_release *req;
197 skb = get_skb(skb, sizeof *req, GFP_KERNEL);
198 if (!skb)
199 return;
200 req = (struct cpl_tid_release *) skb_put(skb, sizeof(*req));
201 INIT_TP_WR(req, hwtid);
202 OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_TID_RELEASE, hwtid));
203 set_wr_txq(skb, CPL_PRIORITY_SETUP, 0);
204 c4iw_ofld_send(rdev, skb);
205 return;
208 static void set_emss(struct c4iw_ep *ep, u16 opt)
210 ep->emss = ep->com.dev->rdev.lldi.mtus[GET_TCPOPT_MSS(opt)] - 40;
211 ep->mss = ep->emss;
212 if (GET_TCPOPT_TSTAMP(opt))
213 ep->emss -= 12;
214 if (ep->emss < 128)
215 ep->emss = 128;
216 PDBG("%s mss_idx %u mss %u emss=%u\n", __func__, GET_TCPOPT_MSS(opt),
217 ep->mss, ep->emss);
220 static enum c4iw_ep_state state_read(struct c4iw_ep_common *epc)
222 enum c4iw_ep_state state;
224 mutex_lock(&epc->mutex);
225 state = epc->state;
226 mutex_unlock(&epc->mutex);
227 return state;
230 static void __state_set(struct c4iw_ep_common *epc, enum c4iw_ep_state new)
232 epc->state = new;
235 static void state_set(struct c4iw_ep_common *epc, enum c4iw_ep_state new)
237 mutex_lock(&epc->mutex);
238 PDBG("%s - %s -> %s\n", __func__, states[epc->state], states[new]);
239 __state_set(epc, new);
240 mutex_unlock(&epc->mutex);
241 return;
244 static void *alloc_ep(int size, gfp_t gfp)
246 struct c4iw_ep_common *epc;
248 epc = kzalloc(size, gfp);
249 if (epc) {
250 kref_init(&epc->kref);
251 mutex_init(&epc->mutex);
252 c4iw_init_wr_wait(&epc->wr_wait);
254 PDBG("%s alloc ep %p\n", __func__, epc);
255 return epc;
258 void _c4iw_free_ep(struct kref *kref)
260 struct c4iw_ep *ep;
262 ep = container_of(kref, struct c4iw_ep, com.kref);
263 PDBG("%s ep %p state %s\n", __func__, ep, states[state_read(&ep->com)]);
264 if (test_bit(RELEASE_RESOURCES, &ep->com.flags)) {
265 cxgb4_remove_tid(ep->com.dev->rdev.lldi.tids, 0, ep->hwtid);
266 dst_release(ep->dst);
267 cxgb4_l2t_release(ep->l2t);
269 kfree(ep);
272 static void release_ep_resources(struct c4iw_ep *ep)
274 set_bit(RELEASE_RESOURCES, &ep->com.flags);
275 c4iw_put_ep(&ep->com);
278 static int status2errno(int status)
280 switch (status) {
281 case CPL_ERR_NONE:
282 return 0;
283 case CPL_ERR_CONN_RESET:
284 return -ECONNRESET;
285 case CPL_ERR_ARP_MISS:
286 return -EHOSTUNREACH;
287 case CPL_ERR_CONN_TIMEDOUT:
288 return -ETIMEDOUT;
289 case CPL_ERR_TCAM_FULL:
290 return -ENOMEM;
291 case CPL_ERR_CONN_EXIST:
292 return -EADDRINUSE;
293 default:
294 return -EIO;
299 * Try and reuse skbs already allocated...
301 static struct sk_buff *get_skb(struct sk_buff *skb, int len, gfp_t gfp)
303 if (skb && !skb_is_nonlinear(skb) && !skb_cloned(skb)) {
304 skb_trim(skb, 0);
305 skb_get(skb);
306 skb_reset_transport_header(skb);
307 } else {
308 skb = alloc_skb(len, gfp);
310 return skb;
313 static struct rtable *find_route(struct c4iw_dev *dev, __be32 local_ip,
314 __be32 peer_ip, __be16 local_port,
315 __be16 peer_port, u8 tos)
317 struct rtable *rt;
318 struct flowi4 fl4;
320 rt = ip_route_output_ports(&init_net, &fl4, NULL, peer_ip, local_ip,
321 peer_port, local_port, IPPROTO_TCP,
322 tos, 0);
323 if (IS_ERR(rt))
324 return NULL;
325 return rt;
328 static void arp_failure_discard(void *handle, struct sk_buff *skb)
330 PDBG("%s c4iw_dev %p\n", __func__, handle);
331 kfree_skb(skb);
335 * Handle an ARP failure for an active open.
337 static void act_open_req_arp_failure(void *handle, struct sk_buff *skb)
339 printk(KERN_ERR MOD "ARP failure duing connect\n");
340 kfree_skb(skb);
344 * Handle an ARP failure for a CPL_ABORT_REQ. Change it into a no RST variant
345 * and send it along.
347 static void abort_arp_failure(void *handle, struct sk_buff *skb)
349 struct c4iw_rdev *rdev = handle;
350 struct cpl_abort_req *req = cplhdr(skb);
352 PDBG("%s rdev %p\n", __func__, rdev);
353 req->cmd = CPL_ABORT_NO_RST;
354 c4iw_ofld_send(rdev, skb);
357 static void send_flowc(struct c4iw_ep *ep, struct sk_buff *skb)
359 unsigned int flowclen = 80;
360 struct fw_flowc_wr *flowc;
361 int i;
363 skb = get_skb(skb, flowclen, GFP_KERNEL);
364 flowc = (struct fw_flowc_wr *)__skb_put(skb, flowclen);
366 flowc->op_to_nparams = cpu_to_be32(FW_WR_OP(FW_FLOWC_WR) |
367 FW_FLOWC_WR_NPARAMS(8));
368 flowc->flowid_len16 = cpu_to_be32(FW_WR_LEN16(DIV_ROUND_UP(flowclen,
369 16)) | FW_WR_FLOWID(ep->hwtid));
371 flowc->mnemval[0].mnemonic = FW_FLOWC_MNEM_PFNVFN;
372 flowc->mnemval[0].val = cpu_to_be32(PCI_FUNC(ep->com.dev->rdev.lldi.pdev->devfn) << 8);
373 flowc->mnemval[1].mnemonic = FW_FLOWC_MNEM_CH;
374 flowc->mnemval[1].val = cpu_to_be32(ep->tx_chan);
375 flowc->mnemval[2].mnemonic = FW_FLOWC_MNEM_PORT;
376 flowc->mnemval[2].val = cpu_to_be32(ep->tx_chan);
377 flowc->mnemval[3].mnemonic = FW_FLOWC_MNEM_IQID;
378 flowc->mnemval[3].val = cpu_to_be32(ep->rss_qid);
379 flowc->mnemval[4].mnemonic = FW_FLOWC_MNEM_SNDNXT;
380 flowc->mnemval[4].val = cpu_to_be32(ep->snd_seq);
381 flowc->mnemval[5].mnemonic = FW_FLOWC_MNEM_RCVNXT;
382 flowc->mnemval[5].val = cpu_to_be32(ep->rcv_seq);
383 flowc->mnemval[6].mnemonic = FW_FLOWC_MNEM_SNDBUF;
384 flowc->mnemval[6].val = cpu_to_be32(snd_win);
385 flowc->mnemval[7].mnemonic = FW_FLOWC_MNEM_MSS;
386 flowc->mnemval[7].val = cpu_to_be32(ep->emss);
387 /* Pad WR to 16 byte boundary */
388 flowc->mnemval[8].mnemonic = 0;
389 flowc->mnemval[8].val = 0;
390 for (i = 0; i < 9; i++) {
391 flowc->mnemval[i].r4[0] = 0;
392 flowc->mnemval[i].r4[1] = 0;
393 flowc->mnemval[i].r4[2] = 0;
396 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
397 c4iw_ofld_send(&ep->com.dev->rdev, skb);
400 static int send_halfclose(struct c4iw_ep *ep, gfp_t gfp)
402 struct cpl_close_con_req *req;
403 struct sk_buff *skb;
404 int wrlen = roundup(sizeof *req, 16);
406 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
407 skb = get_skb(NULL, wrlen, gfp);
408 if (!skb) {
409 printk(KERN_ERR MOD "%s - failed to alloc skb\n", __func__);
410 return -ENOMEM;
412 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
413 t4_set_arp_err_handler(skb, NULL, arp_failure_discard);
414 req = (struct cpl_close_con_req *) skb_put(skb, wrlen);
415 memset(req, 0, wrlen);
416 INIT_TP_WR(req, ep->hwtid);
417 OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_CLOSE_CON_REQ,
418 ep->hwtid));
419 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
422 static int send_abort(struct c4iw_ep *ep, struct sk_buff *skb, gfp_t gfp)
424 struct cpl_abort_req *req;
425 int wrlen = roundup(sizeof *req, 16);
427 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
428 skb = get_skb(skb, wrlen, gfp);
429 if (!skb) {
430 printk(KERN_ERR MOD "%s - failed to alloc skb.\n",
431 __func__);
432 return -ENOMEM;
434 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
435 t4_set_arp_err_handler(skb, &ep->com.dev->rdev, abort_arp_failure);
436 req = (struct cpl_abort_req *) skb_put(skb, wrlen);
437 memset(req, 0, wrlen);
438 INIT_TP_WR(req, ep->hwtid);
439 OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_ABORT_REQ, ep->hwtid));
440 req->cmd = CPL_ABORT_SEND_RST;
441 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
444 static int send_connect(struct c4iw_ep *ep)
446 struct cpl_act_open_req *req;
447 struct sk_buff *skb;
448 u64 opt0;
449 u32 opt2;
450 unsigned int mtu_idx;
451 int wscale;
452 int wrlen = roundup(sizeof *req, 16);
454 PDBG("%s ep %p atid %u\n", __func__, ep, ep->atid);
456 skb = get_skb(NULL, wrlen, GFP_KERNEL);
457 if (!skb) {
458 printk(KERN_ERR MOD "%s - failed to alloc skb.\n",
459 __func__);
460 return -ENOMEM;
462 set_wr_txq(skb, CPL_PRIORITY_SETUP, ep->ctrlq_idx);
464 cxgb4_best_mtu(ep->com.dev->rdev.lldi.mtus, ep->mtu, &mtu_idx);
465 wscale = compute_wscale(rcv_win);
466 opt0 = KEEP_ALIVE(1) |
467 DELACK(1) |
468 WND_SCALE(wscale) |
469 MSS_IDX(mtu_idx) |
470 L2T_IDX(ep->l2t->idx) |
471 TX_CHAN(ep->tx_chan) |
472 SMAC_SEL(ep->smac_idx) |
473 DSCP(ep->tos) |
474 ULP_MODE(ULP_MODE_TCPDDP) |
475 RCV_BUFSIZ(rcv_win>>10);
476 opt2 = RX_CHANNEL(0) |
477 RSS_QUEUE_VALID | RSS_QUEUE(ep->rss_qid);
478 if (enable_tcp_timestamps)
479 opt2 |= TSTAMPS_EN(1);
480 if (enable_tcp_sack)
481 opt2 |= SACK_EN(1);
482 if (wscale && enable_tcp_window_scaling)
483 opt2 |= WND_SCALE_EN(1);
484 t4_set_arp_err_handler(skb, NULL, act_open_req_arp_failure);
486 req = (struct cpl_act_open_req *) skb_put(skb, wrlen);
487 INIT_TP_WR(req, 0);
488 OPCODE_TID(req) = cpu_to_be32(
489 MK_OPCODE_TID(CPL_ACT_OPEN_REQ, ((ep->rss_qid<<14)|ep->atid)));
490 req->local_port = ep->com.local_addr.sin_port;
491 req->peer_port = ep->com.remote_addr.sin_port;
492 req->local_ip = ep->com.local_addr.sin_addr.s_addr;
493 req->peer_ip = ep->com.remote_addr.sin_addr.s_addr;
494 req->opt0 = cpu_to_be64(opt0);
495 req->params = 0;
496 req->opt2 = cpu_to_be32(opt2);
497 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
500 static void send_mpa_req(struct c4iw_ep *ep, struct sk_buff *skb)
502 int mpalen, wrlen;
503 struct fw_ofld_tx_data_wr *req;
504 struct mpa_message *mpa;
506 PDBG("%s ep %p tid %u pd_len %d\n", __func__, ep, ep->hwtid, ep->plen);
508 BUG_ON(skb_cloned(skb));
510 mpalen = sizeof(*mpa) + ep->plen;
511 wrlen = roundup(mpalen + sizeof *req, 16);
512 skb = get_skb(skb, wrlen, GFP_KERNEL);
513 if (!skb) {
514 connect_reply_upcall(ep, -ENOMEM);
515 return;
517 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
519 req = (struct fw_ofld_tx_data_wr *)skb_put(skb, wrlen);
520 memset(req, 0, wrlen);
521 req->op_to_immdlen = cpu_to_be32(
522 FW_WR_OP(FW_OFLD_TX_DATA_WR) |
523 FW_WR_COMPL(1) |
524 FW_WR_IMMDLEN(mpalen));
525 req->flowid_len16 = cpu_to_be32(
526 FW_WR_FLOWID(ep->hwtid) |
527 FW_WR_LEN16(wrlen >> 4));
528 req->plen = cpu_to_be32(mpalen);
529 req->tunnel_to_proxy = cpu_to_be32(
530 FW_OFLD_TX_DATA_WR_FLUSH(1) |
531 FW_OFLD_TX_DATA_WR_SHOVE(1));
533 mpa = (struct mpa_message *)(req + 1);
534 memcpy(mpa->key, MPA_KEY_REQ, sizeof(mpa->key));
535 mpa->flags = (crc_enabled ? MPA_CRC : 0) |
536 (markers_enabled ? MPA_MARKERS : 0);
537 mpa->private_data_size = htons(ep->plen);
538 mpa->revision = mpa_rev;
540 if (ep->plen)
541 memcpy(mpa->private_data, ep->mpa_pkt + sizeof(*mpa), ep->plen);
544 * Reference the mpa skb. This ensures the data area
545 * will remain in memory until the hw acks the tx.
546 * Function fw4_ack() will deref it.
548 skb_get(skb);
549 t4_set_arp_err_handler(skb, NULL, arp_failure_discard);
550 BUG_ON(ep->mpa_skb);
551 ep->mpa_skb = skb;
552 c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
553 start_ep_timer(ep);
554 state_set(&ep->com, MPA_REQ_SENT);
555 ep->mpa_attr.initiator = 1;
556 return;
559 static int send_mpa_reject(struct c4iw_ep *ep, const void *pdata, u8 plen)
561 int mpalen, wrlen;
562 struct fw_ofld_tx_data_wr *req;
563 struct mpa_message *mpa;
564 struct sk_buff *skb;
566 PDBG("%s ep %p tid %u pd_len %d\n", __func__, ep, ep->hwtid, ep->plen);
568 mpalen = sizeof(*mpa) + plen;
569 wrlen = roundup(mpalen + sizeof *req, 16);
571 skb = get_skb(NULL, wrlen, GFP_KERNEL);
572 if (!skb) {
573 printk(KERN_ERR MOD "%s - cannot alloc skb!\n", __func__);
574 return -ENOMEM;
576 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
578 req = (struct fw_ofld_tx_data_wr *)skb_put(skb, wrlen);
579 memset(req, 0, wrlen);
580 req->op_to_immdlen = cpu_to_be32(
581 FW_WR_OP(FW_OFLD_TX_DATA_WR) |
582 FW_WR_COMPL(1) |
583 FW_WR_IMMDLEN(mpalen));
584 req->flowid_len16 = cpu_to_be32(
585 FW_WR_FLOWID(ep->hwtid) |
586 FW_WR_LEN16(wrlen >> 4));
587 req->plen = cpu_to_be32(mpalen);
588 req->tunnel_to_proxy = cpu_to_be32(
589 FW_OFLD_TX_DATA_WR_FLUSH(1) |
590 FW_OFLD_TX_DATA_WR_SHOVE(1));
592 mpa = (struct mpa_message *)(req + 1);
593 memset(mpa, 0, sizeof(*mpa));
594 memcpy(mpa->key, MPA_KEY_REP, sizeof(mpa->key));
595 mpa->flags = MPA_REJECT;
596 mpa->revision = mpa_rev;
597 mpa->private_data_size = htons(plen);
598 if (plen)
599 memcpy(mpa->private_data, pdata, plen);
602 * Reference the mpa skb again. This ensures the data area
603 * will remain in memory until the hw acks the tx.
604 * Function fw4_ack() will deref it.
606 skb_get(skb);
607 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
608 t4_set_arp_err_handler(skb, NULL, arp_failure_discard);
609 BUG_ON(ep->mpa_skb);
610 ep->mpa_skb = skb;
611 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
614 static int send_mpa_reply(struct c4iw_ep *ep, const void *pdata, u8 plen)
616 int mpalen, wrlen;
617 struct fw_ofld_tx_data_wr *req;
618 struct mpa_message *mpa;
619 struct sk_buff *skb;
621 PDBG("%s ep %p tid %u pd_len %d\n", __func__, ep, ep->hwtid, ep->plen);
623 mpalen = sizeof(*mpa) + plen;
624 wrlen = roundup(mpalen + sizeof *req, 16);
626 skb = get_skb(NULL, wrlen, GFP_KERNEL);
627 if (!skb) {
628 printk(KERN_ERR MOD "%s - cannot alloc skb!\n", __func__);
629 return -ENOMEM;
631 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
633 req = (struct fw_ofld_tx_data_wr *) skb_put(skb, wrlen);
634 memset(req, 0, wrlen);
635 req->op_to_immdlen = cpu_to_be32(
636 FW_WR_OP(FW_OFLD_TX_DATA_WR) |
637 FW_WR_COMPL(1) |
638 FW_WR_IMMDLEN(mpalen));
639 req->flowid_len16 = cpu_to_be32(
640 FW_WR_FLOWID(ep->hwtid) |
641 FW_WR_LEN16(wrlen >> 4));
642 req->plen = cpu_to_be32(mpalen);
643 req->tunnel_to_proxy = cpu_to_be32(
644 FW_OFLD_TX_DATA_WR_FLUSH(1) |
645 FW_OFLD_TX_DATA_WR_SHOVE(1));
647 mpa = (struct mpa_message *)(req + 1);
648 memset(mpa, 0, sizeof(*mpa));
649 memcpy(mpa->key, MPA_KEY_REP, sizeof(mpa->key));
650 mpa->flags = (ep->mpa_attr.crc_enabled ? MPA_CRC : 0) |
651 (markers_enabled ? MPA_MARKERS : 0);
652 mpa->revision = mpa_rev;
653 mpa->private_data_size = htons(plen);
654 if (plen)
655 memcpy(mpa->private_data, pdata, plen);
658 * Reference the mpa skb. This ensures the data area
659 * will remain in memory until the hw acks the tx.
660 * Function fw4_ack() will deref it.
662 skb_get(skb);
663 t4_set_arp_err_handler(skb, NULL, arp_failure_discard);
664 ep->mpa_skb = skb;
665 state_set(&ep->com, MPA_REP_SENT);
666 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
669 static int act_establish(struct c4iw_dev *dev, struct sk_buff *skb)
671 struct c4iw_ep *ep;
672 struct cpl_act_establish *req = cplhdr(skb);
673 unsigned int tid = GET_TID(req);
674 unsigned int atid = GET_TID_TID(ntohl(req->tos_atid));
675 struct tid_info *t = dev->rdev.lldi.tids;
677 ep = lookup_atid(t, atid);
679 PDBG("%s ep %p tid %u snd_isn %u rcv_isn %u\n", __func__, ep, tid,
680 be32_to_cpu(req->snd_isn), be32_to_cpu(req->rcv_isn));
682 dst_confirm(ep->dst);
684 /* setup the hwtid for this connection */
685 ep->hwtid = tid;
686 cxgb4_insert_tid(t, ep, tid);
688 ep->snd_seq = be32_to_cpu(req->snd_isn);
689 ep->rcv_seq = be32_to_cpu(req->rcv_isn);
691 set_emss(ep, ntohs(req->tcp_opt));
693 /* dealloc the atid */
694 cxgb4_free_atid(t, atid);
696 /* start MPA negotiation */
697 send_flowc(ep, NULL);
698 send_mpa_req(ep, skb);
700 return 0;
703 static void close_complete_upcall(struct c4iw_ep *ep)
705 struct iw_cm_event event;
707 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
708 memset(&event, 0, sizeof(event));
709 event.event = IW_CM_EVENT_CLOSE;
710 if (ep->com.cm_id) {
711 PDBG("close complete delivered ep %p cm_id %p tid %u\n",
712 ep, ep->com.cm_id, ep->hwtid);
713 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
714 ep->com.cm_id->rem_ref(ep->com.cm_id);
715 ep->com.cm_id = NULL;
716 ep->com.qp = NULL;
720 static int abort_connection(struct c4iw_ep *ep, struct sk_buff *skb, gfp_t gfp)
722 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
723 close_complete_upcall(ep);
724 state_set(&ep->com, ABORTING);
725 return send_abort(ep, skb, gfp);
728 static void peer_close_upcall(struct c4iw_ep *ep)
730 struct iw_cm_event event;
732 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
733 memset(&event, 0, sizeof(event));
734 event.event = IW_CM_EVENT_DISCONNECT;
735 if (ep->com.cm_id) {
736 PDBG("peer close delivered ep %p cm_id %p tid %u\n",
737 ep, ep->com.cm_id, ep->hwtid);
738 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
742 static void peer_abort_upcall(struct c4iw_ep *ep)
744 struct iw_cm_event event;
746 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
747 memset(&event, 0, sizeof(event));
748 event.event = IW_CM_EVENT_CLOSE;
749 event.status = -ECONNRESET;
750 if (ep->com.cm_id) {
751 PDBG("abort delivered ep %p cm_id %p tid %u\n", ep,
752 ep->com.cm_id, ep->hwtid);
753 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
754 ep->com.cm_id->rem_ref(ep->com.cm_id);
755 ep->com.cm_id = NULL;
756 ep->com.qp = NULL;
760 static void connect_reply_upcall(struct c4iw_ep *ep, int status)
762 struct iw_cm_event event;
764 PDBG("%s ep %p tid %u status %d\n", __func__, ep, ep->hwtid, status);
765 memset(&event, 0, sizeof(event));
766 event.event = IW_CM_EVENT_CONNECT_REPLY;
767 event.status = status;
768 event.local_addr = ep->com.local_addr;
769 event.remote_addr = ep->com.remote_addr;
771 if ((status == 0) || (status == -ECONNREFUSED)) {
772 event.private_data_len = ep->plen;
773 event.private_data = ep->mpa_pkt + sizeof(struct mpa_message);
776 PDBG("%s ep %p tid %u status %d\n", __func__, ep,
777 ep->hwtid, status);
778 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
780 if (status < 0) {
781 ep->com.cm_id->rem_ref(ep->com.cm_id);
782 ep->com.cm_id = NULL;
783 ep->com.qp = NULL;
787 static void connect_request_upcall(struct c4iw_ep *ep)
789 struct iw_cm_event event;
791 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
792 memset(&event, 0, sizeof(event));
793 event.event = IW_CM_EVENT_CONNECT_REQUEST;
794 event.local_addr = ep->com.local_addr;
795 event.remote_addr = ep->com.remote_addr;
796 event.private_data_len = ep->plen;
797 event.private_data = ep->mpa_pkt + sizeof(struct mpa_message);
798 event.provider_data = ep;
799 if (state_read(&ep->parent_ep->com) != DEAD) {
800 c4iw_get_ep(&ep->com);
801 ep->parent_ep->com.cm_id->event_handler(
802 ep->parent_ep->com.cm_id,
803 &event);
805 c4iw_put_ep(&ep->parent_ep->com);
806 ep->parent_ep = NULL;
809 static void established_upcall(struct c4iw_ep *ep)
811 struct iw_cm_event event;
813 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
814 memset(&event, 0, sizeof(event));
815 event.event = IW_CM_EVENT_ESTABLISHED;
816 if (ep->com.cm_id) {
817 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
818 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
822 static int update_rx_credits(struct c4iw_ep *ep, u32 credits)
824 struct cpl_rx_data_ack *req;
825 struct sk_buff *skb;
826 int wrlen = roundup(sizeof *req, 16);
828 PDBG("%s ep %p tid %u credits %u\n", __func__, ep, ep->hwtid, credits);
829 skb = get_skb(NULL, wrlen, GFP_KERNEL);
830 if (!skb) {
831 printk(KERN_ERR MOD "update_rx_credits - cannot alloc skb!\n");
832 return 0;
835 req = (struct cpl_rx_data_ack *) skb_put(skb, wrlen);
836 memset(req, 0, wrlen);
837 INIT_TP_WR(req, ep->hwtid);
838 OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_RX_DATA_ACK,
839 ep->hwtid));
840 req->credit_dack = cpu_to_be32(credits | RX_FORCE_ACK(1) |
841 F_RX_DACK_CHANGE |
842 V_RX_DACK_MODE(dack_mode));
843 set_wr_txq(skb, CPL_PRIORITY_ACK, ep->ctrlq_idx);
844 c4iw_ofld_send(&ep->com.dev->rdev, skb);
845 return credits;
848 static void process_mpa_reply(struct c4iw_ep *ep, struct sk_buff *skb)
850 struct mpa_message *mpa;
851 u16 plen;
852 struct c4iw_qp_attributes attrs;
853 enum c4iw_qp_attr_mask mask;
854 int err;
856 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
859 * Stop mpa timer. If it expired, then the state has
860 * changed and we bail since ep_timeout already aborted
861 * the connection.
863 stop_ep_timer(ep);
864 if (state_read(&ep->com) != MPA_REQ_SENT)
865 return;
868 * If we get more than the supported amount of private data
869 * then we must fail this connection.
871 if (ep->mpa_pkt_len + skb->len > sizeof(ep->mpa_pkt)) {
872 err = -EINVAL;
873 goto err;
877 * copy the new data into our accumulation buffer.
879 skb_copy_from_linear_data(skb, &(ep->mpa_pkt[ep->mpa_pkt_len]),
880 skb->len);
881 ep->mpa_pkt_len += skb->len;
884 * if we don't even have the mpa message, then bail.
886 if (ep->mpa_pkt_len < sizeof(*mpa))
887 return;
888 mpa = (struct mpa_message *) ep->mpa_pkt;
890 /* Validate MPA header. */
891 if (mpa->revision != mpa_rev) {
892 err = -EPROTO;
893 goto err;
895 if (memcmp(mpa->key, MPA_KEY_REP, sizeof(mpa->key))) {
896 err = -EPROTO;
897 goto err;
900 plen = ntohs(mpa->private_data_size);
903 * Fail if there's too much private data.
905 if (plen > MPA_MAX_PRIVATE_DATA) {
906 err = -EPROTO;
907 goto err;
911 * If plen does not account for pkt size
913 if (ep->mpa_pkt_len > (sizeof(*mpa) + plen)) {
914 err = -EPROTO;
915 goto err;
918 ep->plen = (u8) plen;
921 * If we don't have all the pdata yet, then bail.
922 * We'll continue process when more data arrives.
924 if (ep->mpa_pkt_len < (sizeof(*mpa) + plen))
925 return;
927 if (mpa->flags & MPA_REJECT) {
928 err = -ECONNREFUSED;
929 goto err;
933 * If we get here we have accumulated the entire mpa
934 * start reply message including private data. And
935 * the MPA header is valid.
937 state_set(&ep->com, FPDU_MODE);
938 ep->mpa_attr.crc_enabled = (mpa->flags & MPA_CRC) | crc_enabled ? 1 : 0;
939 ep->mpa_attr.recv_marker_enabled = markers_enabled;
940 ep->mpa_attr.xmit_marker_enabled = mpa->flags & MPA_MARKERS ? 1 : 0;
941 ep->mpa_attr.version = mpa_rev;
942 ep->mpa_attr.p2p_type = peer2peer ? p2p_type :
943 FW_RI_INIT_P2PTYPE_DISABLED;
944 PDBG("%s - crc_enabled=%d, recv_marker_enabled=%d, "
945 "xmit_marker_enabled=%d, version=%d\n", __func__,
946 ep->mpa_attr.crc_enabled, ep->mpa_attr.recv_marker_enabled,
947 ep->mpa_attr.xmit_marker_enabled, ep->mpa_attr.version);
949 attrs.mpa_attr = ep->mpa_attr;
950 attrs.max_ird = ep->ird;
951 attrs.max_ord = ep->ord;
952 attrs.llp_stream_handle = ep;
953 attrs.next_state = C4IW_QP_STATE_RTS;
955 mask = C4IW_QP_ATTR_NEXT_STATE |
956 C4IW_QP_ATTR_LLP_STREAM_HANDLE | C4IW_QP_ATTR_MPA_ATTR |
957 C4IW_QP_ATTR_MAX_IRD | C4IW_QP_ATTR_MAX_ORD;
959 /* bind QP and TID with INIT_WR */
960 err = c4iw_modify_qp(ep->com.qp->rhp,
961 ep->com.qp, mask, &attrs, 1);
962 if (err)
963 goto err;
964 goto out;
965 err:
966 state_set(&ep->com, ABORTING);
967 send_abort(ep, skb, GFP_KERNEL);
968 out:
969 connect_reply_upcall(ep, err);
970 return;
973 static void process_mpa_request(struct c4iw_ep *ep, struct sk_buff *skb)
975 struct mpa_message *mpa;
976 u16 plen;
978 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
980 if (state_read(&ep->com) != MPA_REQ_WAIT)
981 return;
984 * If we get more than the supported amount of private data
985 * then we must fail this connection.
987 if (ep->mpa_pkt_len + skb->len > sizeof(ep->mpa_pkt)) {
988 stop_ep_timer(ep);
989 abort_connection(ep, skb, GFP_KERNEL);
990 return;
993 PDBG("%s enter (%s line %u)\n", __func__, __FILE__, __LINE__);
996 * Copy the new data into our accumulation buffer.
998 skb_copy_from_linear_data(skb, &(ep->mpa_pkt[ep->mpa_pkt_len]),
999 skb->len);
1000 ep->mpa_pkt_len += skb->len;
1003 * If we don't even have the mpa message, then bail.
1004 * We'll continue process when more data arrives.
1006 if (ep->mpa_pkt_len < sizeof(*mpa))
1007 return;
1009 PDBG("%s enter (%s line %u)\n", __func__, __FILE__, __LINE__);
1010 stop_ep_timer(ep);
1011 mpa = (struct mpa_message *) ep->mpa_pkt;
1014 * Validate MPA Header.
1016 if (mpa->revision != mpa_rev) {
1017 abort_connection(ep, skb, GFP_KERNEL);
1018 return;
1021 if (memcmp(mpa->key, MPA_KEY_REQ, sizeof(mpa->key))) {
1022 abort_connection(ep, skb, GFP_KERNEL);
1023 return;
1026 plen = ntohs(mpa->private_data_size);
1029 * Fail if there's too much private data.
1031 if (plen > MPA_MAX_PRIVATE_DATA) {
1032 abort_connection(ep, skb, GFP_KERNEL);
1033 return;
1037 * If plen does not account for pkt size
1039 if (ep->mpa_pkt_len > (sizeof(*mpa) + plen)) {
1040 abort_connection(ep, skb, GFP_KERNEL);
1041 return;
1043 ep->plen = (u8) plen;
1046 * If we don't have all the pdata yet, then bail.
1048 if (ep->mpa_pkt_len < (sizeof(*mpa) + plen))
1049 return;
1052 * If we get here we have accumulated the entire mpa
1053 * start reply message including private data.
1055 ep->mpa_attr.initiator = 0;
1056 ep->mpa_attr.crc_enabled = (mpa->flags & MPA_CRC) | crc_enabled ? 1 : 0;
1057 ep->mpa_attr.recv_marker_enabled = markers_enabled;
1058 ep->mpa_attr.xmit_marker_enabled = mpa->flags & MPA_MARKERS ? 1 : 0;
1059 ep->mpa_attr.version = mpa_rev;
1060 ep->mpa_attr.p2p_type = peer2peer ? p2p_type :
1061 FW_RI_INIT_P2PTYPE_DISABLED;
1062 PDBG("%s - crc_enabled=%d, recv_marker_enabled=%d, "
1063 "xmit_marker_enabled=%d, version=%d p2p_type=%d\n", __func__,
1064 ep->mpa_attr.crc_enabled, ep->mpa_attr.recv_marker_enabled,
1065 ep->mpa_attr.xmit_marker_enabled, ep->mpa_attr.version,
1066 ep->mpa_attr.p2p_type);
1068 state_set(&ep->com, MPA_REQ_RCVD);
1070 /* drive upcall */
1071 connect_request_upcall(ep);
1072 return;
1075 static int rx_data(struct c4iw_dev *dev, struct sk_buff *skb)
1077 struct c4iw_ep *ep;
1078 struct cpl_rx_data *hdr = cplhdr(skb);
1079 unsigned int dlen = ntohs(hdr->len);
1080 unsigned int tid = GET_TID(hdr);
1081 struct tid_info *t = dev->rdev.lldi.tids;
1083 ep = lookup_tid(t, tid);
1084 PDBG("%s ep %p tid %u dlen %u\n", __func__, ep, ep->hwtid, dlen);
1085 skb_pull(skb, sizeof(*hdr));
1086 skb_trim(skb, dlen);
1088 ep->rcv_seq += dlen;
1089 BUG_ON(ep->rcv_seq != (ntohl(hdr->seq) + dlen));
1091 /* update RX credits */
1092 update_rx_credits(ep, dlen);
1094 switch (state_read(&ep->com)) {
1095 case MPA_REQ_SENT:
1096 process_mpa_reply(ep, skb);
1097 break;
1098 case MPA_REQ_WAIT:
1099 process_mpa_request(ep, skb);
1100 break;
1101 case MPA_REP_SENT:
1102 break;
1103 default:
1104 printk(KERN_ERR MOD "%s Unexpected streaming data."
1105 " ep %p state %d tid %u\n",
1106 __func__, ep, state_read(&ep->com), ep->hwtid);
1109 * The ep will timeout and inform the ULP of the failure.
1110 * See ep_timeout().
1112 break;
1114 return 0;
1117 static int abort_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
1119 struct c4iw_ep *ep;
1120 struct cpl_abort_rpl_rss *rpl = cplhdr(skb);
1121 int release = 0;
1122 unsigned int tid = GET_TID(rpl);
1123 struct tid_info *t = dev->rdev.lldi.tids;
1125 ep = lookup_tid(t, tid);
1126 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1127 BUG_ON(!ep);
1128 mutex_lock(&ep->com.mutex);
1129 switch (ep->com.state) {
1130 case ABORTING:
1131 __state_set(&ep->com, DEAD);
1132 release = 1;
1133 break;
1134 default:
1135 printk(KERN_ERR "%s ep %p state %d\n",
1136 __func__, ep, ep->com.state);
1137 break;
1139 mutex_unlock(&ep->com.mutex);
1141 if (release)
1142 release_ep_resources(ep);
1143 return 0;
1147 * Return whether a failed active open has allocated a TID
1149 static inline int act_open_has_tid(int status)
1151 return status != CPL_ERR_TCAM_FULL && status != CPL_ERR_CONN_EXIST &&
1152 status != CPL_ERR_ARP_MISS;
1155 static int act_open_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
1157 struct c4iw_ep *ep;
1158 struct cpl_act_open_rpl *rpl = cplhdr(skb);
1159 unsigned int atid = GET_TID_TID(GET_AOPEN_ATID(
1160 ntohl(rpl->atid_status)));
1161 struct tid_info *t = dev->rdev.lldi.tids;
1162 int status = GET_AOPEN_STATUS(ntohl(rpl->atid_status));
1164 ep = lookup_atid(t, atid);
1166 PDBG("%s ep %p atid %u status %u errno %d\n", __func__, ep, atid,
1167 status, status2errno(status));
1169 if (status == CPL_ERR_RTX_NEG_ADVICE) {
1170 printk(KERN_WARNING MOD "Connection problems for atid %u\n",
1171 atid);
1172 return 0;
1175 connect_reply_upcall(ep, status2errno(status));
1176 state_set(&ep->com, DEAD);
1178 if (status && act_open_has_tid(status))
1179 cxgb4_remove_tid(ep->com.dev->rdev.lldi.tids, 0, GET_TID(rpl));
1181 cxgb4_free_atid(t, atid);
1182 dst_release(ep->dst);
1183 cxgb4_l2t_release(ep->l2t);
1184 c4iw_put_ep(&ep->com);
1186 return 0;
1189 static int pass_open_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
1191 struct cpl_pass_open_rpl *rpl = cplhdr(skb);
1192 struct tid_info *t = dev->rdev.lldi.tids;
1193 unsigned int stid = GET_TID(rpl);
1194 struct c4iw_listen_ep *ep = lookup_stid(t, stid);
1196 if (!ep) {
1197 printk(KERN_ERR MOD "stid %d lookup failure!\n", stid);
1198 return 0;
1200 PDBG("%s ep %p status %d error %d\n", __func__, ep,
1201 rpl->status, status2errno(rpl->status));
1202 c4iw_wake_up(&ep->com.wr_wait, status2errno(rpl->status));
1204 return 0;
1207 static int listen_stop(struct c4iw_listen_ep *ep)
1209 struct sk_buff *skb;
1210 struct cpl_close_listsvr_req *req;
1212 PDBG("%s ep %p\n", __func__, ep);
1213 skb = get_skb(NULL, sizeof(*req), GFP_KERNEL);
1214 if (!skb) {
1215 printk(KERN_ERR MOD "%s - failed to alloc skb\n", __func__);
1216 return -ENOMEM;
1218 req = (struct cpl_close_listsvr_req *) skb_put(skb, sizeof(*req));
1219 INIT_TP_WR(req, 0);
1220 OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_CLOSE_LISTSRV_REQ,
1221 ep->stid));
1222 req->reply_ctrl = cpu_to_be16(
1223 QUEUENO(ep->com.dev->rdev.lldi.rxq_ids[0]));
1224 set_wr_txq(skb, CPL_PRIORITY_SETUP, 0);
1225 return c4iw_ofld_send(&ep->com.dev->rdev, skb);
1228 static int close_listsrv_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
1230 struct cpl_close_listsvr_rpl *rpl = cplhdr(skb);
1231 struct tid_info *t = dev->rdev.lldi.tids;
1232 unsigned int stid = GET_TID(rpl);
1233 struct c4iw_listen_ep *ep = lookup_stid(t, stid);
1235 PDBG("%s ep %p\n", __func__, ep);
1236 c4iw_wake_up(&ep->com.wr_wait, status2errno(rpl->status));
1237 return 0;
1240 static void accept_cr(struct c4iw_ep *ep, __be32 peer_ip, struct sk_buff *skb,
1241 struct cpl_pass_accept_req *req)
1243 struct cpl_pass_accept_rpl *rpl;
1244 unsigned int mtu_idx;
1245 u64 opt0;
1246 u32 opt2;
1247 int wscale;
1249 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1250 BUG_ON(skb_cloned(skb));
1251 skb_trim(skb, sizeof(*rpl));
1252 skb_get(skb);
1253 cxgb4_best_mtu(ep->com.dev->rdev.lldi.mtus, ep->mtu, &mtu_idx);
1254 wscale = compute_wscale(rcv_win);
1255 opt0 = KEEP_ALIVE(1) |
1256 DELACK(1) |
1257 WND_SCALE(wscale) |
1258 MSS_IDX(mtu_idx) |
1259 L2T_IDX(ep->l2t->idx) |
1260 TX_CHAN(ep->tx_chan) |
1261 SMAC_SEL(ep->smac_idx) |
1262 DSCP(ep->tos) |
1263 ULP_MODE(ULP_MODE_TCPDDP) |
1264 RCV_BUFSIZ(rcv_win>>10);
1265 opt2 = RX_CHANNEL(0) |
1266 RSS_QUEUE_VALID | RSS_QUEUE(ep->rss_qid);
1268 if (enable_tcp_timestamps && req->tcpopt.tstamp)
1269 opt2 |= TSTAMPS_EN(1);
1270 if (enable_tcp_sack && req->tcpopt.sack)
1271 opt2 |= SACK_EN(1);
1272 if (wscale && enable_tcp_window_scaling)
1273 opt2 |= WND_SCALE_EN(1);
1275 rpl = cplhdr(skb);
1276 INIT_TP_WR(rpl, ep->hwtid);
1277 OPCODE_TID(rpl) = cpu_to_be32(MK_OPCODE_TID(CPL_PASS_ACCEPT_RPL,
1278 ep->hwtid));
1279 rpl->opt0 = cpu_to_be64(opt0);
1280 rpl->opt2 = cpu_to_be32(opt2);
1281 set_wr_txq(skb, CPL_PRIORITY_SETUP, ep->ctrlq_idx);
1282 c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
1284 return;
1287 static void reject_cr(struct c4iw_dev *dev, u32 hwtid, __be32 peer_ip,
1288 struct sk_buff *skb)
1290 PDBG("%s c4iw_dev %p tid %u peer_ip %x\n", __func__, dev, hwtid,
1291 peer_ip);
1292 BUG_ON(skb_cloned(skb));
1293 skb_trim(skb, sizeof(struct cpl_tid_release));
1294 skb_get(skb);
1295 release_tid(&dev->rdev, hwtid, skb);
1296 return;
1299 static void get_4tuple(struct cpl_pass_accept_req *req,
1300 __be32 *local_ip, __be32 *peer_ip,
1301 __be16 *local_port, __be16 *peer_port)
1303 int eth_len = G_ETH_HDR_LEN(be32_to_cpu(req->hdr_len));
1304 int ip_len = G_IP_HDR_LEN(be32_to_cpu(req->hdr_len));
1305 struct iphdr *ip = (struct iphdr *)((u8 *)(req + 1) + eth_len);
1306 struct tcphdr *tcp = (struct tcphdr *)
1307 ((u8 *)(req + 1) + eth_len + ip_len);
1309 PDBG("%s saddr 0x%x daddr 0x%x sport %u dport %u\n", __func__,
1310 ntohl(ip->saddr), ntohl(ip->daddr), ntohs(tcp->source),
1311 ntohs(tcp->dest));
1313 *peer_ip = ip->saddr;
1314 *local_ip = ip->daddr;
1315 *peer_port = tcp->source;
1316 *local_port = tcp->dest;
1318 return;
1321 static int pass_accept_req(struct c4iw_dev *dev, struct sk_buff *skb)
1323 struct c4iw_ep *child_ep, *parent_ep;
1324 struct cpl_pass_accept_req *req = cplhdr(skb);
1325 unsigned int stid = GET_POPEN_TID(ntohl(req->tos_stid));
1326 struct tid_info *t = dev->rdev.lldi.tids;
1327 unsigned int hwtid = GET_TID(req);
1328 struct dst_entry *dst;
1329 struct l2t_entry *l2t;
1330 struct rtable *rt;
1331 __be32 local_ip, peer_ip;
1332 __be16 local_port, peer_port;
1333 struct net_device *pdev;
1334 u32 tx_chan, smac_idx;
1335 u16 rss_qid;
1336 u32 mtu;
1337 int step;
1338 int txq_idx, ctrlq_idx;
1340 parent_ep = lookup_stid(t, stid);
1341 PDBG("%s parent ep %p tid %u\n", __func__, parent_ep, hwtid);
1343 get_4tuple(req, &local_ip, &peer_ip, &local_port, &peer_port);
1345 if (state_read(&parent_ep->com) != LISTEN) {
1346 printk(KERN_ERR "%s - listening ep not in LISTEN\n",
1347 __func__);
1348 goto reject;
1351 /* Find output route */
1352 rt = find_route(dev, local_ip, peer_ip, local_port, peer_port,
1353 GET_POPEN_TOS(ntohl(req->tos_stid)));
1354 if (!rt) {
1355 printk(KERN_ERR MOD "%s - failed to find dst entry!\n",
1356 __func__);
1357 goto reject;
1359 dst = &rt->dst;
1360 if (dst->neighbour->dev->flags & IFF_LOOPBACK) {
1361 pdev = ip_dev_find(&init_net, peer_ip);
1362 BUG_ON(!pdev);
1363 l2t = cxgb4_l2t_get(dev->rdev.lldi.l2t, dst->neighbour,
1364 pdev, 0);
1365 mtu = pdev->mtu;
1366 tx_chan = cxgb4_port_chan(pdev);
1367 smac_idx = (cxgb4_port_viid(pdev) & 0x7F) << 1;
1368 step = dev->rdev.lldi.ntxq / dev->rdev.lldi.nchan;
1369 txq_idx = cxgb4_port_idx(pdev) * step;
1370 ctrlq_idx = cxgb4_port_idx(pdev);
1371 step = dev->rdev.lldi.nrxq / dev->rdev.lldi.nchan;
1372 rss_qid = dev->rdev.lldi.rxq_ids[cxgb4_port_idx(pdev) * step];
1373 dev_put(pdev);
1374 } else {
1375 l2t = cxgb4_l2t_get(dev->rdev.lldi.l2t, dst->neighbour,
1376 dst->neighbour->dev, 0);
1377 mtu = dst_mtu(dst);
1378 tx_chan = cxgb4_port_chan(dst->neighbour->dev);
1379 smac_idx = (cxgb4_port_viid(dst->neighbour->dev) & 0x7F) << 1;
1380 step = dev->rdev.lldi.ntxq / dev->rdev.lldi.nchan;
1381 txq_idx = cxgb4_port_idx(dst->neighbour->dev) * step;
1382 ctrlq_idx = cxgb4_port_idx(dst->neighbour->dev);
1383 step = dev->rdev.lldi.nrxq / dev->rdev.lldi.nchan;
1384 rss_qid = dev->rdev.lldi.rxq_ids[
1385 cxgb4_port_idx(dst->neighbour->dev) * step];
1387 if (!l2t) {
1388 printk(KERN_ERR MOD "%s - failed to allocate l2t entry!\n",
1389 __func__);
1390 dst_release(dst);
1391 goto reject;
1394 child_ep = alloc_ep(sizeof(*child_ep), GFP_KERNEL);
1395 if (!child_ep) {
1396 printk(KERN_ERR MOD "%s - failed to allocate ep entry!\n",
1397 __func__);
1398 cxgb4_l2t_release(l2t);
1399 dst_release(dst);
1400 goto reject;
1402 state_set(&child_ep->com, CONNECTING);
1403 child_ep->com.dev = dev;
1404 child_ep->com.cm_id = NULL;
1405 child_ep->com.local_addr.sin_family = PF_INET;
1406 child_ep->com.local_addr.sin_port = local_port;
1407 child_ep->com.local_addr.sin_addr.s_addr = local_ip;
1408 child_ep->com.remote_addr.sin_family = PF_INET;
1409 child_ep->com.remote_addr.sin_port = peer_port;
1410 child_ep->com.remote_addr.sin_addr.s_addr = peer_ip;
1411 c4iw_get_ep(&parent_ep->com);
1412 child_ep->parent_ep = parent_ep;
1413 child_ep->tos = GET_POPEN_TOS(ntohl(req->tos_stid));
1414 child_ep->l2t = l2t;
1415 child_ep->dst = dst;
1416 child_ep->hwtid = hwtid;
1417 child_ep->tx_chan = tx_chan;
1418 child_ep->smac_idx = smac_idx;
1419 child_ep->rss_qid = rss_qid;
1420 child_ep->mtu = mtu;
1421 child_ep->txq_idx = txq_idx;
1422 child_ep->ctrlq_idx = ctrlq_idx;
1424 PDBG("%s tx_chan %u smac_idx %u rss_qid %u\n", __func__,
1425 tx_chan, smac_idx, rss_qid);
1427 init_timer(&child_ep->timer);
1428 cxgb4_insert_tid(t, child_ep, hwtid);
1429 accept_cr(child_ep, peer_ip, skb, req);
1430 goto out;
1431 reject:
1432 reject_cr(dev, hwtid, peer_ip, skb);
1433 out:
1434 return 0;
1437 static int pass_establish(struct c4iw_dev *dev, struct sk_buff *skb)
1439 struct c4iw_ep *ep;
1440 struct cpl_pass_establish *req = cplhdr(skb);
1441 struct tid_info *t = dev->rdev.lldi.tids;
1442 unsigned int tid = GET_TID(req);
1444 ep = lookup_tid(t, tid);
1445 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1446 ep->snd_seq = be32_to_cpu(req->snd_isn);
1447 ep->rcv_seq = be32_to_cpu(req->rcv_isn);
1449 set_emss(ep, ntohs(req->tcp_opt));
1451 dst_confirm(ep->dst);
1452 state_set(&ep->com, MPA_REQ_WAIT);
1453 start_ep_timer(ep);
1454 send_flowc(ep, skb);
1456 return 0;
1459 static int peer_close(struct c4iw_dev *dev, struct sk_buff *skb)
1461 struct cpl_peer_close *hdr = cplhdr(skb);
1462 struct c4iw_ep *ep;
1463 struct c4iw_qp_attributes attrs;
1464 int disconnect = 1;
1465 int release = 0;
1466 int abort = 0;
1467 struct tid_info *t = dev->rdev.lldi.tids;
1468 unsigned int tid = GET_TID(hdr);
1470 ep = lookup_tid(t, tid);
1471 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1472 dst_confirm(ep->dst);
1474 mutex_lock(&ep->com.mutex);
1475 switch (ep->com.state) {
1476 case MPA_REQ_WAIT:
1477 __state_set(&ep->com, CLOSING);
1478 break;
1479 case MPA_REQ_SENT:
1480 __state_set(&ep->com, CLOSING);
1481 connect_reply_upcall(ep, -ECONNRESET);
1482 break;
1483 case MPA_REQ_RCVD:
1486 * We're gonna mark this puppy DEAD, but keep
1487 * the reference on it until the ULP accepts or
1488 * rejects the CR. Also wake up anyone waiting
1489 * in rdma connection migration (see c4iw_accept_cr()).
1491 __state_set(&ep->com, CLOSING);
1492 PDBG("waking up ep %p tid %u\n", ep, ep->hwtid);
1493 c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
1494 break;
1495 case MPA_REP_SENT:
1496 __state_set(&ep->com, CLOSING);
1497 PDBG("waking up ep %p tid %u\n", ep, ep->hwtid);
1498 c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
1499 break;
1500 case FPDU_MODE:
1501 start_ep_timer(ep);
1502 __state_set(&ep->com, CLOSING);
1503 attrs.next_state = C4IW_QP_STATE_CLOSING;
1504 abort = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
1505 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
1506 peer_close_upcall(ep);
1507 disconnect = 1;
1508 break;
1509 case ABORTING:
1510 disconnect = 0;
1511 break;
1512 case CLOSING:
1513 __state_set(&ep->com, MORIBUND);
1514 disconnect = 0;
1515 break;
1516 case MORIBUND:
1517 stop_ep_timer(ep);
1518 if (ep->com.cm_id && ep->com.qp) {
1519 attrs.next_state = C4IW_QP_STATE_IDLE;
1520 c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
1521 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
1523 close_complete_upcall(ep);
1524 __state_set(&ep->com, DEAD);
1525 release = 1;
1526 disconnect = 0;
1527 break;
1528 case DEAD:
1529 disconnect = 0;
1530 break;
1531 default:
1532 BUG_ON(1);
1534 mutex_unlock(&ep->com.mutex);
1535 if (disconnect)
1536 c4iw_ep_disconnect(ep, 0, GFP_KERNEL);
1537 if (release)
1538 release_ep_resources(ep);
1539 return 0;
1543 * Returns whether an ABORT_REQ_RSS message is a negative advice.
1545 static int is_neg_adv_abort(unsigned int status)
1547 return status == CPL_ERR_RTX_NEG_ADVICE ||
1548 status == CPL_ERR_PERSIST_NEG_ADVICE;
1551 static int peer_abort(struct c4iw_dev *dev, struct sk_buff *skb)
1553 struct cpl_abort_req_rss *req = cplhdr(skb);
1554 struct c4iw_ep *ep;
1555 struct cpl_abort_rpl *rpl;
1556 struct sk_buff *rpl_skb;
1557 struct c4iw_qp_attributes attrs;
1558 int ret;
1559 int release = 0;
1560 struct tid_info *t = dev->rdev.lldi.tids;
1561 unsigned int tid = GET_TID(req);
1563 ep = lookup_tid(t, tid);
1564 if (is_neg_adv_abort(req->status)) {
1565 PDBG("%s neg_adv_abort ep %p tid %u\n", __func__, ep,
1566 ep->hwtid);
1567 return 0;
1569 PDBG("%s ep %p tid %u state %u\n", __func__, ep, ep->hwtid,
1570 ep->com.state);
1573 * Wake up any threads in rdma_init() or rdma_fini().
1575 c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
1577 mutex_lock(&ep->com.mutex);
1578 switch (ep->com.state) {
1579 case CONNECTING:
1580 break;
1581 case MPA_REQ_WAIT:
1582 stop_ep_timer(ep);
1583 break;
1584 case MPA_REQ_SENT:
1585 stop_ep_timer(ep);
1586 connect_reply_upcall(ep, -ECONNRESET);
1587 break;
1588 case MPA_REP_SENT:
1589 break;
1590 case MPA_REQ_RCVD:
1591 break;
1592 case MORIBUND:
1593 case CLOSING:
1594 stop_ep_timer(ep);
1595 /*FALLTHROUGH*/
1596 case FPDU_MODE:
1597 if (ep->com.cm_id && ep->com.qp) {
1598 attrs.next_state = C4IW_QP_STATE_ERROR;
1599 ret = c4iw_modify_qp(ep->com.qp->rhp,
1600 ep->com.qp, C4IW_QP_ATTR_NEXT_STATE,
1601 &attrs, 1);
1602 if (ret)
1603 printk(KERN_ERR MOD
1604 "%s - qp <- error failed!\n",
1605 __func__);
1607 peer_abort_upcall(ep);
1608 break;
1609 case ABORTING:
1610 break;
1611 case DEAD:
1612 PDBG("%s PEER_ABORT IN DEAD STATE!!!!\n", __func__);
1613 mutex_unlock(&ep->com.mutex);
1614 return 0;
1615 default:
1616 BUG_ON(1);
1617 break;
1619 dst_confirm(ep->dst);
1620 if (ep->com.state != ABORTING) {
1621 __state_set(&ep->com, DEAD);
1622 release = 1;
1624 mutex_unlock(&ep->com.mutex);
1626 rpl_skb = get_skb(skb, sizeof(*rpl), GFP_KERNEL);
1627 if (!rpl_skb) {
1628 printk(KERN_ERR MOD "%s - cannot allocate skb!\n",
1629 __func__);
1630 release = 1;
1631 goto out;
1633 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
1634 rpl = (struct cpl_abort_rpl *) skb_put(rpl_skb, sizeof(*rpl));
1635 INIT_TP_WR(rpl, ep->hwtid);
1636 OPCODE_TID(rpl) = cpu_to_be32(MK_OPCODE_TID(CPL_ABORT_RPL, ep->hwtid));
1637 rpl->cmd = CPL_ABORT_NO_RST;
1638 c4iw_ofld_send(&ep->com.dev->rdev, rpl_skb);
1639 out:
1640 if (release)
1641 release_ep_resources(ep);
1642 return 0;
1645 static int close_con_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
1647 struct c4iw_ep *ep;
1648 struct c4iw_qp_attributes attrs;
1649 struct cpl_close_con_rpl *rpl = cplhdr(skb);
1650 int release = 0;
1651 struct tid_info *t = dev->rdev.lldi.tids;
1652 unsigned int tid = GET_TID(rpl);
1654 ep = lookup_tid(t, tid);
1656 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1657 BUG_ON(!ep);
1659 /* The cm_id may be null if we failed to connect */
1660 mutex_lock(&ep->com.mutex);
1661 switch (ep->com.state) {
1662 case CLOSING:
1663 __state_set(&ep->com, MORIBUND);
1664 break;
1665 case MORIBUND:
1666 stop_ep_timer(ep);
1667 if ((ep->com.cm_id) && (ep->com.qp)) {
1668 attrs.next_state = C4IW_QP_STATE_IDLE;
1669 c4iw_modify_qp(ep->com.qp->rhp,
1670 ep->com.qp,
1671 C4IW_QP_ATTR_NEXT_STATE,
1672 &attrs, 1);
1674 close_complete_upcall(ep);
1675 __state_set(&ep->com, DEAD);
1676 release = 1;
1677 break;
1678 case ABORTING:
1679 case DEAD:
1680 break;
1681 default:
1682 BUG_ON(1);
1683 break;
1685 mutex_unlock(&ep->com.mutex);
1686 if (release)
1687 release_ep_resources(ep);
1688 return 0;
1691 static int terminate(struct c4iw_dev *dev, struct sk_buff *skb)
1693 struct cpl_rdma_terminate *rpl = cplhdr(skb);
1694 struct tid_info *t = dev->rdev.lldi.tids;
1695 unsigned int tid = GET_TID(rpl);
1696 struct c4iw_ep *ep;
1697 struct c4iw_qp_attributes attrs;
1699 ep = lookup_tid(t, tid);
1700 BUG_ON(!ep);
1702 if (ep && ep->com.qp) {
1703 printk(KERN_WARNING MOD "TERM received tid %u qpid %u\n", tid,
1704 ep->com.qp->wq.sq.qid);
1705 attrs.next_state = C4IW_QP_STATE_TERMINATE;
1706 c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
1707 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
1708 } else
1709 printk(KERN_WARNING MOD "TERM received tid %u no ep/qp\n", tid);
1711 return 0;
1715 * Upcall from the adapter indicating data has been transmitted.
1716 * For us its just the single MPA request or reply. We can now free
1717 * the skb holding the mpa message.
1719 static int fw4_ack(struct c4iw_dev *dev, struct sk_buff *skb)
1721 struct c4iw_ep *ep;
1722 struct cpl_fw4_ack *hdr = cplhdr(skb);
1723 u8 credits = hdr->credits;
1724 unsigned int tid = GET_TID(hdr);
1725 struct tid_info *t = dev->rdev.lldi.tids;
1728 ep = lookup_tid(t, tid);
1729 PDBG("%s ep %p tid %u credits %u\n", __func__, ep, ep->hwtid, credits);
1730 if (credits == 0) {
1731 PDBG("%s 0 credit ack ep %p tid %u state %u\n",
1732 __func__, ep, ep->hwtid, state_read(&ep->com));
1733 return 0;
1736 dst_confirm(ep->dst);
1737 if (ep->mpa_skb) {
1738 PDBG("%s last streaming msg ack ep %p tid %u state %u "
1739 "initiator %u freeing skb\n", __func__, ep, ep->hwtid,
1740 state_read(&ep->com), ep->mpa_attr.initiator ? 1 : 0);
1741 kfree_skb(ep->mpa_skb);
1742 ep->mpa_skb = NULL;
1744 return 0;
1747 int c4iw_reject_cr(struct iw_cm_id *cm_id, const void *pdata, u8 pdata_len)
1749 int err;
1750 struct c4iw_ep *ep = to_ep(cm_id);
1751 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1753 if (state_read(&ep->com) == DEAD) {
1754 c4iw_put_ep(&ep->com);
1755 return -ECONNRESET;
1757 BUG_ON(state_read(&ep->com) != MPA_REQ_RCVD);
1758 if (mpa_rev == 0)
1759 abort_connection(ep, NULL, GFP_KERNEL);
1760 else {
1761 err = send_mpa_reject(ep, pdata, pdata_len);
1762 err = c4iw_ep_disconnect(ep, 0, GFP_KERNEL);
1764 c4iw_put_ep(&ep->com);
1765 return 0;
1768 int c4iw_accept_cr(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
1770 int err;
1771 struct c4iw_qp_attributes attrs;
1772 enum c4iw_qp_attr_mask mask;
1773 struct c4iw_ep *ep = to_ep(cm_id);
1774 struct c4iw_dev *h = to_c4iw_dev(cm_id->device);
1775 struct c4iw_qp *qp = get_qhp(h, conn_param->qpn);
1777 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1778 if (state_read(&ep->com) == DEAD) {
1779 err = -ECONNRESET;
1780 goto err;
1783 BUG_ON(state_read(&ep->com) != MPA_REQ_RCVD);
1784 BUG_ON(!qp);
1786 if ((conn_param->ord > c4iw_max_read_depth) ||
1787 (conn_param->ird > c4iw_max_read_depth)) {
1788 abort_connection(ep, NULL, GFP_KERNEL);
1789 err = -EINVAL;
1790 goto err;
1793 cm_id->add_ref(cm_id);
1794 ep->com.cm_id = cm_id;
1795 ep->com.qp = qp;
1797 ep->ird = conn_param->ird;
1798 ep->ord = conn_param->ord;
1800 if (peer2peer && ep->ird == 0)
1801 ep->ird = 1;
1803 PDBG("%s %d ird %d ord %d\n", __func__, __LINE__, ep->ird, ep->ord);
1805 /* bind QP to EP and move to RTS */
1806 attrs.mpa_attr = ep->mpa_attr;
1807 attrs.max_ird = ep->ird;
1808 attrs.max_ord = ep->ord;
1809 attrs.llp_stream_handle = ep;
1810 attrs.next_state = C4IW_QP_STATE_RTS;
1812 /* bind QP and TID with INIT_WR */
1813 mask = C4IW_QP_ATTR_NEXT_STATE |
1814 C4IW_QP_ATTR_LLP_STREAM_HANDLE |
1815 C4IW_QP_ATTR_MPA_ATTR |
1816 C4IW_QP_ATTR_MAX_IRD |
1817 C4IW_QP_ATTR_MAX_ORD;
1819 err = c4iw_modify_qp(ep->com.qp->rhp,
1820 ep->com.qp, mask, &attrs, 1);
1821 if (err)
1822 goto err1;
1823 err = send_mpa_reply(ep, conn_param->private_data,
1824 conn_param->private_data_len);
1825 if (err)
1826 goto err1;
1828 state_set(&ep->com, FPDU_MODE);
1829 established_upcall(ep);
1830 c4iw_put_ep(&ep->com);
1831 return 0;
1832 err1:
1833 ep->com.cm_id = NULL;
1834 ep->com.qp = NULL;
1835 cm_id->rem_ref(cm_id);
1836 err:
1837 c4iw_put_ep(&ep->com);
1838 return err;
1841 int c4iw_connect(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
1843 int err = 0;
1844 struct c4iw_dev *dev = to_c4iw_dev(cm_id->device);
1845 struct c4iw_ep *ep;
1846 struct rtable *rt;
1847 struct net_device *pdev;
1848 int step;
1850 if ((conn_param->ord > c4iw_max_read_depth) ||
1851 (conn_param->ird > c4iw_max_read_depth)) {
1852 err = -EINVAL;
1853 goto out;
1855 ep = alloc_ep(sizeof(*ep), GFP_KERNEL);
1856 if (!ep) {
1857 printk(KERN_ERR MOD "%s - cannot alloc ep.\n", __func__);
1858 err = -ENOMEM;
1859 goto out;
1861 init_timer(&ep->timer);
1862 ep->plen = conn_param->private_data_len;
1863 if (ep->plen)
1864 memcpy(ep->mpa_pkt + sizeof(struct mpa_message),
1865 conn_param->private_data, ep->plen);
1866 ep->ird = conn_param->ird;
1867 ep->ord = conn_param->ord;
1869 if (peer2peer && ep->ord == 0)
1870 ep->ord = 1;
1872 cm_id->add_ref(cm_id);
1873 ep->com.dev = dev;
1874 ep->com.cm_id = cm_id;
1875 ep->com.qp = get_qhp(dev, conn_param->qpn);
1876 BUG_ON(!ep->com.qp);
1877 PDBG("%s qpn 0x%x qp %p cm_id %p\n", __func__, conn_param->qpn,
1878 ep->com.qp, cm_id);
1881 * Allocate an active TID to initiate a TCP connection.
1883 ep->atid = cxgb4_alloc_atid(dev->rdev.lldi.tids, ep);
1884 if (ep->atid == -1) {
1885 printk(KERN_ERR MOD "%s - cannot alloc atid.\n", __func__);
1886 err = -ENOMEM;
1887 goto fail2;
1890 PDBG("%s saddr 0x%x sport 0x%x raddr 0x%x rport 0x%x\n", __func__,
1891 ntohl(cm_id->local_addr.sin_addr.s_addr),
1892 ntohs(cm_id->local_addr.sin_port),
1893 ntohl(cm_id->remote_addr.sin_addr.s_addr),
1894 ntohs(cm_id->remote_addr.sin_port));
1896 /* find a route */
1897 rt = find_route(dev,
1898 cm_id->local_addr.sin_addr.s_addr,
1899 cm_id->remote_addr.sin_addr.s_addr,
1900 cm_id->local_addr.sin_port,
1901 cm_id->remote_addr.sin_port, 0);
1902 if (!rt) {
1903 printk(KERN_ERR MOD "%s - cannot find route.\n", __func__);
1904 err = -EHOSTUNREACH;
1905 goto fail3;
1907 ep->dst = &rt->dst;
1909 /* get a l2t entry */
1910 if (ep->dst->neighbour->dev->flags & IFF_LOOPBACK) {
1911 PDBG("%s LOOPBACK\n", __func__);
1912 pdev = ip_dev_find(&init_net,
1913 cm_id->remote_addr.sin_addr.s_addr);
1914 ep->l2t = cxgb4_l2t_get(ep->com.dev->rdev.lldi.l2t,
1915 ep->dst->neighbour,
1916 pdev, 0);
1917 ep->mtu = pdev->mtu;
1918 ep->tx_chan = cxgb4_port_chan(pdev);
1919 ep->smac_idx = (cxgb4_port_viid(pdev) & 0x7F) << 1;
1920 step = ep->com.dev->rdev.lldi.ntxq /
1921 ep->com.dev->rdev.lldi.nchan;
1922 ep->txq_idx = cxgb4_port_idx(pdev) * step;
1923 step = ep->com.dev->rdev.lldi.nrxq /
1924 ep->com.dev->rdev.lldi.nchan;
1925 ep->ctrlq_idx = cxgb4_port_idx(pdev);
1926 ep->rss_qid = ep->com.dev->rdev.lldi.rxq_ids[
1927 cxgb4_port_idx(pdev) * step];
1928 dev_put(pdev);
1929 } else {
1930 ep->l2t = cxgb4_l2t_get(ep->com.dev->rdev.lldi.l2t,
1931 ep->dst->neighbour,
1932 ep->dst->neighbour->dev, 0);
1933 ep->mtu = dst_mtu(ep->dst);
1934 ep->tx_chan = cxgb4_port_chan(ep->dst->neighbour->dev);
1935 ep->smac_idx = (cxgb4_port_viid(ep->dst->neighbour->dev) &
1936 0x7F) << 1;
1937 step = ep->com.dev->rdev.lldi.ntxq /
1938 ep->com.dev->rdev.lldi.nchan;
1939 ep->txq_idx = cxgb4_port_idx(ep->dst->neighbour->dev) * step;
1940 ep->ctrlq_idx = cxgb4_port_idx(ep->dst->neighbour->dev);
1941 step = ep->com.dev->rdev.lldi.nrxq /
1942 ep->com.dev->rdev.lldi.nchan;
1943 ep->rss_qid = ep->com.dev->rdev.lldi.rxq_ids[
1944 cxgb4_port_idx(ep->dst->neighbour->dev) * step];
1946 if (!ep->l2t) {
1947 printk(KERN_ERR MOD "%s - cannot alloc l2e.\n", __func__);
1948 err = -ENOMEM;
1949 goto fail4;
1952 PDBG("%s txq_idx %u tx_chan %u smac_idx %u rss_qid %u l2t_idx %u\n",
1953 __func__, ep->txq_idx, ep->tx_chan, ep->smac_idx, ep->rss_qid,
1954 ep->l2t->idx);
1956 state_set(&ep->com, CONNECTING);
1957 ep->tos = 0;
1958 ep->com.local_addr = cm_id->local_addr;
1959 ep->com.remote_addr = cm_id->remote_addr;
1961 /* send connect request to rnic */
1962 err = send_connect(ep);
1963 if (!err)
1964 goto out;
1966 cxgb4_l2t_release(ep->l2t);
1967 fail4:
1968 dst_release(ep->dst);
1969 fail3:
1970 cxgb4_free_atid(ep->com.dev->rdev.lldi.tids, ep->atid);
1971 fail2:
1972 cm_id->rem_ref(cm_id);
1973 c4iw_put_ep(&ep->com);
1974 out:
1975 return err;
1978 int c4iw_create_listen(struct iw_cm_id *cm_id, int backlog)
1980 int err = 0;
1981 struct c4iw_dev *dev = to_c4iw_dev(cm_id->device);
1982 struct c4iw_listen_ep *ep;
1985 might_sleep();
1987 ep = alloc_ep(sizeof(*ep), GFP_KERNEL);
1988 if (!ep) {
1989 printk(KERN_ERR MOD "%s - cannot alloc ep.\n", __func__);
1990 err = -ENOMEM;
1991 goto fail1;
1993 PDBG("%s ep %p\n", __func__, ep);
1994 cm_id->add_ref(cm_id);
1995 ep->com.cm_id = cm_id;
1996 ep->com.dev = dev;
1997 ep->backlog = backlog;
1998 ep->com.local_addr = cm_id->local_addr;
2001 * Allocate a server TID.
2003 ep->stid = cxgb4_alloc_stid(dev->rdev.lldi.tids, PF_INET, ep);
2004 if (ep->stid == -1) {
2005 printk(KERN_ERR MOD "%s - cannot alloc stid.\n", __func__);
2006 err = -ENOMEM;
2007 goto fail2;
2010 state_set(&ep->com, LISTEN);
2011 c4iw_init_wr_wait(&ep->com.wr_wait);
2012 err = cxgb4_create_server(ep->com.dev->rdev.lldi.ports[0], ep->stid,
2013 ep->com.local_addr.sin_addr.s_addr,
2014 ep->com.local_addr.sin_port,
2015 ep->com.dev->rdev.lldi.rxq_ids[0]);
2016 if (err)
2017 goto fail3;
2019 /* wait for pass_open_rpl */
2020 err = c4iw_wait_for_reply(&ep->com.dev->rdev, &ep->com.wr_wait, 0, 0,
2021 __func__);
2022 if (!err) {
2023 cm_id->provider_data = ep;
2024 goto out;
2026 fail3:
2027 cxgb4_free_stid(ep->com.dev->rdev.lldi.tids, ep->stid, PF_INET);
2028 fail2:
2029 cm_id->rem_ref(cm_id);
2030 c4iw_put_ep(&ep->com);
2031 fail1:
2032 out:
2033 return err;
2036 int c4iw_destroy_listen(struct iw_cm_id *cm_id)
2038 int err;
2039 struct c4iw_listen_ep *ep = to_listen_ep(cm_id);
2041 PDBG("%s ep %p\n", __func__, ep);
2043 might_sleep();
2044 state_set(&ep->com, DEAD);
2045 c4iw_init_wr_wait(&ep->com.wr_wait);
2046 err = listen_stop(ep);
2047 if (err)
2048 goto done;
2049 err = c4iw_wait_for_reply(&ep->com.dev->rdev, &ep->com.wr_wait, 0, 0,
2050 __func__);
2051 cxgb4_free_stid(ep->com.dev->rdev.lldi.tids, ep->stid, PF_INET);
2052 done:
2053 cm_id->rem_ref(cm_id);
2054 c4iw_put_ep(&ep->com);
2055 return err;
2058 int c4iw_ep_disconnect(struct c4iw_ep *ep, int abrupt, gfp_t gfp)
2060 int ret = 0;
2061 int close = 0;
2062 int fatal = 0;
2063 struct c4iw_rdev *rdev;
2065 mutex_lock(&ep->com.mutex);
2067 PDBG("%s ep %p state %s, abrupt %d\n", __func__, ep,
2068 states[ep->com.state], abrupt);
2070 rdev = &ep->com.dev->rdev;
2071 if (c4iw_fatal_error(rdev)) {
2072 fatal = 1;
2073 close_complete_upcall(ep);
2074 ep->com.state = DEAD;
2076 switch (ep->com.state) {
2077 case MPA_REQ_WAIT:
2078 case MPA_REQ_SENT:
2079 case MPA_REQ_RCVD:
2080 case MPA_REP_SENT:
2081 case FPDU_MODE:
2082 close = 1;
2083 if (abrupt)
2084 ep->com.state = ABORTING;
2085 else {
2086 ep->com.state = CLOSING;
2087 start_ep_timer(ep);
2089 set_bit(CLOSE_SENT, &ep->com.flags);
2090 break;
2091 case CLOSING:
2092 if (!test_and_set_bit(CLOSE_SENT, &ep->com.flags)) {
2093 close = 1;
2094 if (abrupt) {
2095 stop_ep_timer(ep);
2096 ep->com.state = ABORTING;
2097 } else
2098 ep->com.state = MORIBUND;
2100 break;
2101 case MORIBUND:
2102 case ABORTING:
2103 case DEAD:
2104 PDBG("%s ignoring disconnect ep %p state %u\n",
2105 __func__, ep, ep->com.state);
2106 break;
2107 default:
2108 BUG();
2109 break;
2112 mutex_unlock(&ep->com.mutex);
2113 if (close) {
2114 if (abrupt)
2115 ret = abort_connection(ep, NULL, gfp);
2116 else
2117 ret = send_halfclose(ep, gfp);
2118 if (ret)
2119 fatal = 1;
2121 if (fatal)
2122 release_ep_resources(ep);
2123 return ret;
2126 static int async_event(struct c4iw_dev *dev, struct sk_buff *skb)
2128 struct cpl_fw6_msg *rpl = cplhdr(skb);
2129 c4iw_ev_dispatch(dev, (struct t4_cqe *)&rpl->data[0]);
2130 return 0;
2134 * These are the real handlers that are called from a
2135 * work queue.
2137 static c4iw_handler_func work_handlers[NUM_CPL_CMDS] = {
2138 [CPL_ACT_ESTABLISH] = act_establish,
2139 [CPL_ACT_OPEN_RPL] = act_open_rpl,
2140 [CPL_RX_DATA] = rx_data,
2141 [CPL_ABORT_RPL_RSS] = abort_rpl,
2142 [CPL_ABORT_RPL] = abort_rpl,
2143 [CPL_PASS_OPEN_RPL] = pass_open_rpl,
2144 [CPL_CLOSE_LISTSRV_RPL] = close_listsrv_rpl,
2145 [CPL_PASS_ACCEPT_REQ] = pass_accept_req,
2146 [CPL_PASS_ESTABLISH] = pass_establish,
2147 [CPL_PEER_CLOSE] = peer_close,
2148 [CPL_ABORT_REQ_RSS] = peer_abort,
2149 [CPL_CLOSE_CON_RPL] = close_con_rpl,
2150 [CPL_RDMA_TERMINATE] = terminate,
2151 [CPL_FW4_ACK] = fw4_ack,
2152 [CPL_FW6_MSG] = async_event
2155 static void process_timeout(struct c4iw_ep *ep)
2157 struct c4iw_qp_attributes attrs;
2158 int abort = 1;
2160 mutex_lock(&ep->com.mutex);
2161 PDBG("%s ep %p tid %u state %d\n", __func__, ep, ep->hwtid,
2162 ep->com.state);
2163 switch (ep->com.state) {
2164 case MPA_REQ_SENT:
2165 __state_set(&ep->com, ABORTING);
2166 connect_reply_upcall(ep, -ETIMEDOUT);
2167 break;
2168 case MPA_REQ_WAIT:
2169 __state_set(&ep->com, ABORTING);
2170 break;
2171 case CLOSING:
2172 case MORIBUND:
2173 if (ep->com.cm_id && ep->com.qp) {
2174 attrs.next_state = C4IW_QP_STATE_ERROR;
2175 c4iw_modify_qp(ep->com.qp->rhp,
2176 ep->com.qp, C4IW_QP_ATTR_NEXT_STATE,
2177 &attrs, 1);
2179 __state_set(&ep->com, ABORTING);
2180 break;
2181 default:
2182 printk(KERN_ERR "%s unexpected state ep %p tid %u state %u\n",
2183 __func__, ep, ep->hwtid, ep->com.state);
2184 WARN_ON(1);
2185 abort = 0;
2187 mutex_unlock(&ep->com.mutex);
2188 if (abort)
2189 abort_connection(ep, NULL, GFP_KERNEL);
2190 c4iw_put_ep(&ep->com);
2193 static void process_timedout_eps(void)
2195 struct c4iw_ep *ep;
2197 spin_lock_irq(&timeout_lock);
2198 while (!list_empty(&timeout_list)) {
2199 struct list_head *tmp;
2201 tmp = timeout_list.next;
2202 list_del(tmp);
2203 spin_unlock_irq(&timeout_lock);
2204 ep = list_entry(tmp, struct c4iw_ep, entry);
2205 process_timeout(ep);
2206 spin_lock_irq(&timeout_lock);
2208 spin_unlock_irq(&timeout_lock);
2211 static void process_work(struct work_struct *work)
2213 struct sk_buff *skb = NULL;
2214 struct c4iw_dev *dev;
2215 struct cpl_act_establish *rpl;
2216 unsigned int opcode;
2217 int ret;
2219 while ((skb = skb_dequeue(&rxq))) {
2220 rpl = cplhdr(skb);
2221 dev = *((struct c4iw_dev **) (skb->cb + sizeof(void *)));
2222 opcode = rpl->ot.opcode;
2224 BUG_ON(!work_handlers[opcode]);
2225 ret = work_handlers[opcode](dev, skb);
2226 if (!ret)
2227 kfree_skb(skb);
2229 process_timedout_eps();
2232 static DECLARE_WORK(skb_work, process_work);
2234 static void ep_timeout(unsigned long arg)
2236 struct c4iw_ep *ep = (struct c4iw_ep *)arg;
2238 spin_lock(&timeout_lock);
2239 list_add_tail(&ep->entry, &timeout_list);
2240 spin_unlock(&timeout_lock);
2241 queue_work(workq, &skb_work);
2245 * All the CM events are handled on a work queue to have a safe context.
2247 static int sched(struct c4iw_dev *dev, struct sk_buff *skb)
2251 * Save dev in the skb->cb area.
2253 *((struct c4iw_dev **) (skb->cb + sizeof(void *))) = dev;
2256 * Queue the skb and schedule the worker thread.
2258 skb_queue_tail(&rxq, skb);
2259 queue_work(workq, &skb_work);
2260 return 0;
2263 static int set_tcb_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
2265 struct cpl_set_tcb_rpl *rpl = cplhdr(skb);
2267 if (rpl->status != CPL_ERR_NONE) {
2268 printk(KERN_ERR MOD "Unexpected SET_TCB_RPL status %u "
2269 "for tid %u\n", rpl->status, GET_TID(rpl));
2271 kfree_skb(skb);
2272 return 0;
2275 static int fw6_msg(struct c4iw_dev *dev, struct sk_buff *skb)
2277 struct cpl_fw6_msg *rpl = cplhdr(skb);
2278 struct c4iw_wr_wait *wr_waitp;
2279 int ret;
2281 PDBG("%s type %u\n", __func__, rpl->type);
2283 switch (rpl->type) {
2284 case 1:
2285 ret = (int)((be64_to_cpu(rpl->data[0]) >> 8) & 0xff);
2286 wr_waitp = (struct c4iw_wr_wait *)(__force unsigned long) rpl->data[1];
2287 PDBG("%s wr_waitp %p ret %u\n", __func__, wr_waitp, ret);
2288 if (wr_waitp)
2289 c4iw_wake_up(wr_waitp, ret ? -ret : 0);
2290 kfree_skb(skb);
2291 break;
2292 case 2:
2293 sched(dev, skb);
2294 break;
2295 default:
2296 printk(KERN_ERR MOD "%s unexpected fw6 msg type %u\n", __func__,
2297 rpl->type);
2298 kfree_skb(skb);
2299 break;
2301 return 0;
2305 * Most upcalls from the T4 Core go to sched() to
2306 * schedule the processing on a work queue.
2308 c4iw_handler_func c4iw_handlers[NUM_CPL_CMDS] = {
2309 [CPL_ACT_ESTABLISH] = sched,
2310 [CPL_ACT_OPEN_RPL] = sched,
2311 [CPL_RX_DATA] = sched,
2312 [CPL_ABORT_RPL_RSS] = sched,
2313 [CPL_ABORT_RPL] = sched,
2314 [CPL_PASS_OPEN_RPL] = sched,
2315 [CPL_CLOSE_LISTSRV_RPL] = sched,
2316 [CPL_PASS_ACCEPT_REQ] = sched,
2317 [CPL_PASS_ESTABLISH] = sched,
2318 [CPL_PEER_CLOSE] = sched,
2319 [CPL_CLOSE_CON_RPL] = sched,
2320 [CPL_ABORT_REQ_RSS] = sched,
2321 [CPL_RDMA_TERMINATE] = sched,
2322 [CPL_FW4_ACK] = sched,
2323 [CPL_SET_TCB_RPL] = set_tcb_rpl,
2324 [CPL_FW6_MSG] = fw6_msg
2327 int __init c4iw_cm_init(void)
2329 spin_lock_init(&timeout_lock);
2330 skb_queue_head_init(&rxq);
2332 workq = create_singlethread_workqueue("iw_cxgb4");
2333 if (!workq)
2334 return -ENOMEM;
2336 return 0;
2339 void __exit c4iw_cm_term(void)
2341 WARN_ON(!list_empty(&timeout_list));
2342 flush_workqueue(workq);
2343 destroy_workqueue(workq);