projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
cxgb4: Add CLIP support to store compressed IPv6 address
[linux-2.6-block.git] / drivers / infiniband / hw / cxgb4 / cm.c
CommitLineData
cfdda9d7
SW		 1/*
2 * Copyright (c) 2009-2010 Chelsio, Inc. All rights reserved.
3 *
4 * This software is available to you under a choice of one of two
5 * licenses. You may choose to be licensed under the terms of the GNU
6 * General Public License (GPL) Version 2, available from the file
7 * COPYING in the main directory of this source tree, or the
8 * OpenIB.org BSD license below:
9 *
10 * Redistribution and use in source and binary forms, with or
11 * without modification, are permitted provided that the following
12 * conditions are met:
13 *
14 * - Redistributions of source code must retain the above
15 * copyright notice, this list of conditions and the following
16 * disclaimer.
17 *
18 * - Redistributions in binary form must reproduce the above
19 * copyright notice, this list of conditions and the following
20 * disclaimer in the documentation and/or other materials
21 * provided with the distribution.
22 *
23 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
24 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
25 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
26 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
27 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
28 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
29 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
30 * SOFTWARE.
31 */
32#include <linux/module.h>
33#include <linux/list.h>
34#include <linux/workqueue.h>
35#include <linux/skbuff.h>
36#include <linux/timer.h>
37#include <linux/notifier.h>
38#include <linux/inetdevice.h>
39#include <linux/ip.h>
40#include <linux/tcp.h>
1cab775c 41#include <linux/if_vlan.h>
cfdda9d7
SW		 42
43#include <net/neighbour.h>
44#include <net/netevent.h>
45#include <net/route.h>
1cab775c 46#include <net/tcp.h>
cfdda9d7
SW		 47
48#include "iw_cxgb4.h"
49
50static char *states[] = {
51 "idle",
52 "listen",
53 "connecting",
54 "mpa_wait_req",
55 "mpa_req_sent",
56 "mpa_req_rcvd",
57 "mpa_rep_sent",
58 "fpdu_mode",
59 "aborting",
60 "closing",
61 "moribund",
62 "dead",
63 NULL,
64};
65
5be78ee9
VP		 66static int nocong;
67module_param(nocong, int, 0644);
68MODULE_PARM_DESC(nocong, "Turn of congestion control (default=0)");
69
70static int enable_ecn;
71module_param(enable_ecn, int, 0644);
72MODULE_PARM_DESC(enable_ecn, "Enable ECN (default=0/disabled)");
73
b52fe09e 74static int dack_mode = 1;
ba6d3925 75module_param(dack_mode, int, 0644);
b52fe09e 76MODULE_PARM_DESC(dack_mode, "Delayed ack mode (default=1)");
ba6d3925 77
be4c9bad
RD		 78int c4iw_max_read_depth = 8;
79module_param(c4iw_max_read_depth, int, 0644);
80MODULE_PARM_DESC(c4iw_max_read_depth, "Per-connection max ORD/IRD (default=8)");
81
cfdda9d7
SW		 82static int enable_tcp_timestamps;
83module_param(enable_tcp_timestamps, int, 0644);
84MODULE_PARM_DESC(enable_tcp_timestamps, "Enable tcp timestamps (default=0)");
85
86static int enable_tcp_sack;
87module_param(enable_tcp_sack, int, 0644);
88MODULE_PARM_DESC(enable_tcp_sack, "Enable tcp SACK (default=0)");
89
90static int enable_tcp_window_scaling = 1;
91module_param(enable_tcp_window_scaling, int, 0644);
92MODULE_PARM_DESC(enable_tcp_window_scaling,
93 "Enable tcp window scaling (default=1)");
94
95int c4iw_debug;
96module_param(c4iw_debug, int, 0644);
97MODULE_PARM_DESC(c4iw_debug, "Enable debug logging (default=0)");
98
99static int peer2peer;
100module_param(peer2peer, int, 0644);
101MODULE_PARM_DESC(peer2peer, "Support peer2peer ULPs (default=0)");
102
103static int p2p_type = FW_RI_INIT_P2PTYPE_READ_REQ;
104module_param(p2p_type, int, 0644);
105MODULE_PARM_DESC(p2p_type, "RDMAP opcode to use for the RTR message: "
106 "1=RDMA_READ 0=RDMA_WRITE (default 1)");
107
108static int ep_timeout_secs = 60;
109module_param(ep_timeout_secs, int, 0644);
110MODULE_PARM_DESC(ep_timeout_secs, "CM Endpoint operation timeout "
111 "in seconds (default=60)");
112
113static int mpa_rev = 1;
114module_param(mpa_rev, int, 0644);
115MODULE_PARM_DESC(mpa_rev, "MPA Revision, 0 supports amso1100, "
d2fe99e8
KS		 116 "1 is RFC0544 spec compliant, 2 is IETF MPA Peer Connect Draft"
117 " compliant (default=1)");
cfdda9d7
SW		 118
119static int markers_enabled;
120module_param(markers_enabled, int, 0644);
121MODULE_PARM_DESC(markers_enabled, "Enable MPA MARKERS (default(0)=disabled)");
122
123static int crc_enabled = 1;
124module_param(crc_enabled, int, 0644);
125MODULE_PARM_DESC(crc_enabled, "Enable MPA CRC (default(1)=enabled)");
126
127static int rcv_win = 256 * 1024;
128module_param(rcv_win, int, 0644);
129MODULE_PARM_DESC(rcv_win, "TCP receive window in bytes (default=256KB)");
130
98ae68b7 131static int snd_win = 128 * 1024;
cfdda9d7 132module_param(snd_win, int, 0644);
98ae68b7 133MODULE_PARM_DESC(snd_win, "TCP send window in bytes (default=128KB)");
cfdda9d7 134
cfdda9d7 135static struct workqueue_struct *workq;
cfdda9d7
SW		 136
137static struct sk_buff_head rxq;
cfdda9d7
SW		 138
139static struct sk_buff *get_skb(struct sk_buff *skb, int len, gfp_t gfp);
140static void ep_timeout(unsigned long arg);
141static void connect_reply_upcall(struct c4iw_ep *ep, int status);
142
be4c9bad
RD		 143static LIST_HEAD(timeout_list);
144static spinlock_t timeout_lock;
145
325abead
VP		 146static void deref_qp(struct c4iw_ep *ep)
147{
148 c4iw_qp_rem_ref(&ep->com.qp->ibqp);
149 clear_bit(QP_REFERENCED, &ep->com.flags);
150}
151
152static void ref_qp(struct c4iw_ep *ep)
153{
154 set_bit(QP_REFERENCED, &ep->com.flags);
155 c4iw_qp_add_ref(&ep->com.qp->ibqp);
156}
157
cfdda9d7
SW		 158static void start_ep_timer(struct c4iw_ep *ep)
159{
160 PDBG("%s ep %p\n", __func__, ep);
161 if (timer_pending(&ep->timer)) {
1ec779cc
VP		 162 pr_err("%s timer already started! ep %p\n",
163 __func__, ep);
164 return;
165 }
166 clear_bit(TIMEOUT, &ep->com.flags);
167 c4iw_get_ep(&ep->com);
cfdda9d7
SW		 168 ep->timer.expires = jiffies + ep_timeout_secs * HZ;
169 ep->timer.data = (unsigned long)ep;
170 ep->timer.function = ep_timeout;
171 add_timer(&ep->timer);
172}
173
174static void stop_ep_timer(struct c4iw_ep *ep)
175{
1ec779cc 176 PDBG("%s ep %p stopping\n", __func__, ep);
cfdda9d7 177 del_timer_sync(&ep->timer);
1ec779cc
VP		 178 if (!test_and_set_bit(TIMEOUT, &ep->com.flags))
179 c4iw_put_ep(&ep->com);
cfdda9d7
SW		 180}
181
182static int c4iw_l2t_send(struct c4iw_rdev *rdev, struct sk_buff *skb,
183 struct l2t_entry *l2e)
184{
185 int error = 0;
186
187 if (c4iw_fatal_error(rdev)) {
188 kfree_skb(skb);
189 PDBG("%s - device in error state - dropping\n", __func__);
190 return -EIO;
191 }
192 error = cxgb4_l2t_send(rdev->lldi.ports[0], skb, l2e);
193 if (error < 0)
194 kfree_skb(skb);
74594861 195 return error < 0 ? error : 0;
cfdda9d7
SW		 196}
197
198int c4iw_ofld_send(struct c4iw_rdev *rdev, struct sk_buff *skb)
199{
200 int error = 0;
201
202 if (c4iw_fatal_error(rdev)) {
203 kfree_skb(skb);
204 PDBG("%s - device in error state - dropping\n", __func__);
205 return -EIO;
206 }
207 error = cxgb4_ofld_send(rdev->lldi.ports[0], skb);
208 if (error < 0)
209 kfree_skb(skb);
74594861 210 return error < 0 ? error : 0;
cfdda9d7
SW		 211}
212
213static void release_tid(struct c4iw_rdev *rdev, u32 hwtid, struct sk_buff *skb)
214{
215 struct cpl_tid_release *req;
216
217 skb = get_skb(skb, sizeof *req, GFP_KERNEL);
218 if (!skb)
219 return;
220 req = (struct cpl_tid_release *) skb_put(skb, sizeof(*req));
221 INIT_TP_WR(req, hwtid);
222 OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_TID_RELEASE, hwtid));
223 set_wr_txq(skb, CPL_PRIORITY_SETUP, 0);
224 c4iw_ofld_send(rdev, skb);
225 return;
226}
227
228static void set_emss(struct c4iw_ep *ep, u16 opt)
229{
230 ep->emss = ep->com.dev->rdev.lldi.mtus[GET_TCPOPT_MSS(opt)] - 40;
231 ep->mss = ep->emss;
232 if (GET_TCPOPT_TSTAMP(opt))
233 ep->emss -= 12;
234 if (ep->emss < 128)
235 ep->emss = 128;
236 PDBG("%s mss_idx %u mss %u emss=%u\n", __func__, GET_TCPOPT_MSS(opt),
237 ep->mss, ep->emss);
238}
239
240static enum c4iw_ep_state state_read(struct c4iw_ep_common *epc)
241{
cfdda9d7
SW		 242 enum c4iw_ep_state state;
243
2f5b48c3 244 mutex_lock(&epc->mutex);
cfdda9d7 245 state = epc->state;
2f5b48c3 246 mutex_unlock(&epc->mutex);
cfdda9d7
SW		 247 return state;
248}
249
250static void __state_set(struct c4iw_ep_common *epc, enum c4iw_ep_state new)
251{
252 epc->state = new;
253}
254
255static void state_set(struct c4iw_ep_common *epc, enum c4iw_ep_state new)
256{
2f5b48c3 257 mutex_lock(&epc->mutex);
cfdda9d7
SW		 258 PDBG("%s - %s -> %s\n", __func__, states[epc->state], states[new]);
259 __state_set(epc, new);
2f5b48c3 260 mutex_unlock(&epc->mutex);
cfdda9d7
SW		 261 return;
262}
263
264static void *alloc_ep(int size, gfp_t gfp)
265{
266 struct c4iw_ep_common *epc;
267
268 epc = kzalloc(size, gfp);
269 if (epc) {
270 kref_init(&epc->kref);
2f5b48c3 271 mutex_init(&epc->mutex);
aadc4df3 272 c4iw_init_wr_wait(&epc->wr_wait);
cfdda9d7
SW		 273 }
274 PDBG("%s alloc ep %p\n", __func__, epc);
275 return epc;
276}
277
278void _c4iw_free_ep(struct kref *kref)
279{
280 struct c4iw_ep *ep;
281
282 ep = container_of(kref, struct c4iw_ep, com.kref);
283 PDBG("%s ep %p state %s\n", __func__, ep, states[state_read(&ep->com)]);
325abead
VP		 284 if (test_bit(QP_REFERENCED, &ep->com.flags))
285 deref_qp(ep);
cfdda9d7 286 if (test_bit(RELEASE_RESOURCES, &ep->com.flags)) {
fe7e0a4d 287 remove_handle(ep->com.dev, &ep->com.dev->hwtid_idr, ep->hwtid);
cfdda9d7
SW		 288 cxgb4_remove_tid(ep->com.dev->rdev.lldi.tids, 0, ep->hwtid);
289 dst_release(ep->dst);
290 cxgb4_l2t_release(ep->l2t);
291 }
292 kfree(ep);
293}
294
295static void release_ep_resources(struct c4iw_ep *ep)
296{
297 set_bit(RELEASE_RESOURCES, &ep->com.flags);
298 c4iw_put_ep(&ep->com);
299}
300
cfdda9d7
SW		 301static int status2errno(int status)
302{
303 switch (status) {
304 case CPL_ERR_NONE:
305 return 0;
306 case CPL_ERR_CONN_RESET:
307 return -ECONNRESET;
308 case CPL_ERR_ARP_MISS:
309 return -EHOSTUNREACH;
310 case CPL_ERR_CONN_TIMEDOUT:
311 return -ETIMEDOUT;
312 case CPL_ERR_TCAM_FULL:
313 return -ENOMEM;
314 case CPL_ERR_CONN_EXIST:
315 return -EADDRINUSE;
316 default:
317 return -EIO;
318 }
319}
320
321/*
322 * Try and reuse skbs already allocated...
323 */
324static struct sk_buff *get_skb(struct sk_buff *skb, int len, gfp_t gfp)
325{
326 if (skb && !skb_is_nonlinear(skb) && !skb_cloned(skb)) {
327 skb_trim(skb, 0);
328 skb_get(skb);
329 skb_reset_transport_header(skb);
330 } else {
331 skb = alloc_skb(len, gfp);
332 }
333 return skb;
334}
335
336static struct rtable *find_route(struct c4iw_dev *dev, __be32 local_ip,
337 __be32 peer_ip, __be16 local_port,
338 __be16 peer_port, u8 tos)
339{
340 struct rtable *rt;
31e4543d 341 struct flowi4 fl4;
78fbfd8a 342
31e4543d 343 rt = ip_route_output_ports(&init_net, &fl4, NULL, peer_ip, local_ip,
78fbfd8a
DM		 344 peer_port, local_port, IPPROTO_TCP,
345 tos, 0);
b23dd4fe 346 if (IS_ERR(rt))
cfdda9d7
SW		 347 return NULL;
348 return rt;
349}
350
351static void arp_failure_discard(void *handle, struct sk_buff *skb)
352{
353 PDBG("%s c4iw_dev %p\n", __func__, handle);
354 kfree_skb(skb);
355}
356
357/*
358 * Handle an ARP failure for an active open.
359 */
360static void act_open_req_arp_failure(void *handle, struct sk_buff *skb)
361{
362 printk(KERN_ERR MOD "ARP failure duing connect\n");
363 kfree_skb(skb);
364}
365
366/*
367 * Handle an ARP failure for a CPL_ABORT_REQ. Change it into a no RST variant
368 * and send it along.
369 */
370static void abort_arp_failure(void *handle, struct sk_buff *skb)
371{
372 struct c4iw_rdev *rdev = handle;
373 struct cpl_abort_req *req = cplhdr(skb);
374
375 PDBG("%s rdev %p\n", __func__, rdev);
376 req->cmd = CPL_ABORT_NO_RST;
377 c4iw_ofld_send(rdev, skb);
378}
379
380static void send_flowc(struct c4iw_ep *ep, struct sk_buff *skb)
381{
382 unsigned int flowclen = 80;
383 struct fw_flowc_wr *flowc;
384 int i;
385
386 skb = get_skb(skb, flowclen, GFP_KERNEL);
387 flowc = (struct fw_flowc_wr *)__skb_put(skb, flowclen);
388
389 flowc->op_to_nparams = cpu_to_be32(FW_WR_OP(FW_FLOWC_WR) |
390 FW_FLOWC_WR_NPARAMS(8));
391 flowc->flowid_len16 = cpu_to_be32(FW_WR_LEN16(DIV_ROUND_UP(flowclen,
392 16)) | FW_WR_FLOWID(ep->hwtid));
393
394 flowc->mnemval[0].mnemonic = FW_FLOWC_MNEM_PFNVFN;
94788657 395 flowc->mnemval[0].val = cpu_to_be32(PCI_FUNC(ep->com.dev->rdev.lldi.pdev->devfn) << 8);
cfdda9d7
SW		 396 flowc->mnemval[1].mnemonic = FW_FLOWC_MNEM_CH;
397 flowc->mnemval[1].val = cpu_to_be32(ep->tx_chan);
398 flowc->mnemval[2].mnemonic = FW_FLOWC_MNEM_PORT;
399 flowc->mnemval[2].val = cpu_to_be32(ep->tx_chan);
400 flowc->mnemval[3].mnemonic = FW_FLOWC_MNEM_IQID;
401 flowc->mnemval[3].val = cpu_to_be32(ep->rss_qid);
402 flowc->mnemval[4].mnemonic = FW_FLOWC_MNEM_SNDNXT;
403 flowc->mnemval[4].val = cpu_to_be32(ep->snd_seq);
404 flowc->mnemval[5].mnemonic = FW_FLOWC_MNEM_RCVNXT;
405 flowc->mnemval[5].val = cpu_to_be32(ep->rcv_seq);
406 flowc->mnemval[6].mnemonic = FW_FLOWC_MNEM_SNDBUF;
407 flowc->mnemval[6].val = cpu_to_be32(snd_win);
408 flowc->mnemval[7].mnemonic = FW_FLOWC_MNEM_MSS;
409 flowc->mnemval[7].val = cpu_to_be32(ep->emss);
410 /* Pad WR to 16 byte boundary */
411 flowc->mnemval[8].mnemonic = 0;
412 flowc->mnemval[8].val = 0;
413 for (i = 0; i < 9; i++) {
414 flowc->mnemval[i].r4[0] = 0;
415 flowc->mnemval[i].r4[1] = 0;
416 flowc->mnemval[i].r4[2] = 0;
417 }
418
419 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
420 c4iw_ofld_send(&ep->com.dev->rdev, skb);
421}
422
423static int send_halfclose(struct c4iw_ep *ep, gfp_t gfp)
424{
425 struct cpl_close_con_req *req;
426 struct sk_buff *skb;
427 int wrlen = roundup(sizeof *req, 16);
428
429 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
430 skb = get_skb(NULL, wrlen, gfp);
431 if (!skb) {
432 printk(KERN_ERR MOD "%s - failed to alloc skb\n", __func__);
433 return -ENOMEM;
434 }
435 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
436 t4_set_arp_err_handler(skb, NULL, arp_failure_discard);
437 req = (struct cpl_close_con_req *) skb_put(skb, wrlen);
438 memset(req, 0, wrlen);
439 INIT_TP_WR(req, ep->hwtid);
440 OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_CLOSE_CON_REQ,
441 ep->hwtid));
442 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
443}
444
445static int send_abort(struct c4iw_ep *ep, struct sk_buff *skb, gfp_t gfp)
446{
447 struct cpl_abort_req *req;
448 int wrlen = roundup(sizeof *req, 16);
449
450 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
451 skb = get_skb(skb, wrlen, gfp);
452 if (!skb) {
453 printk(KERN_ERR MOD "%s - failed to alloc skb.\n",
454 __func__);
455 return -ENOMEM;
456 }
457 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
458 t4_set_arp_err_handler(skb, &ep->com.dev->rdev, abort_arp_failure);
459 req = (struct cpl_abort_req *) skb_put(skb, wrlen);
460 memset(req, 0, wrlen);
461 INIT_TP_WR(req, ep->hwtid);
462 OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_ABORT_REQ, ep->hwtid));
463 req->cmd = CPL_ABORT_SEND_RST;
464 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
465}
466
5be78ee9
VP		 467#define VLAN_NONE 0xfff
468#define FILTER_SEL_VLAN_NONE 0xffff
469#define FILTER_SEL_WIDTH_P_FC (3+1) /* port uses 3 bits, FCoE one bit */
470#define FILTER_SEL_WIDTH_VIN_P_FC \
471 (6 + 7 + FILTER_SEL_WIDTH_P_FC) /* 6 bits are unused, VF uses 7 bits*/
472#define FILTER_SEL_WIDTH_TAG_P_FC \
473 (3 + FILTER_SEL_WIDTH_VIN_P_FC) /* PF uses 3 bits */
474#define FILTER_SEL_WIDTH_VLD_TAG_P_FC (1 + FILTER_SEL_WIDTH_TAG_P_FC)
475
476static unsigned int select_ntuple(struct c4iw_dev *dev, struct dst_entry *dst,
477 struct l2t_entry *l2t)
478{
479 unsigned int ntuple = 0;
480 u32 viid;
481
482 switch (dev->rdev.lldi.filt_mode) {
483
484 /* default filter mode */
485 case HW_TPL_FR_MT_PR_IV_P_FC:
486 if (l2t->vlan == VLAN_NONE)
487 ntuple |= FILTER_SEL_VLAN_NONE << FILTER_SEL_WIDTH_P_FC;
488 else {
489 ntuple |= l2t->vlan << FILTER_SEL_WIDTH_P_FC;
490 ntuple |= 1 << FILTER_SEL_WIDTH_VLD_TAG_P_FC;
491 }
492 ntuple |= l2t->lport << S_PORT | IPPROTO_TCP <<
493 FILTER_SEL_WIDTH_VLD_TAG_P_FC;
494 break;
495 case HW_TPL_FR_MT_PR_OV_P_FC: {
496 viid = cxgb4_port_viid(l2t->neigh->dev);
497
498 ntuple |= FW_VIID_VIN_GET(viid) << FILTER_SEL_WIDTH_P_FC;
499 ntuple |= FW_VIID_PFN_GET(viid) << FILTER_SEL_WIDTH_VIN_P_FC;
500 ntuple |= FW_VIID_VIVLD_GET(viid) << FILTER_SEL_WIDTH_TAG_P_FC;
501 ntuple |= l2t->lport << S_PORT | IPPROTO_TCP <<
502 FILTER_SEL_WIDTH_VLD_TAG_P_FC;
503 break;
504 }
505 default:
506 break;
507 }
508 return ntuple;
509}
510
cfdda9d7
SW		 511static int send_connect(struct c4iw_ep *ep)
512{
513 struct cpl_act_open_req *req;
f079af7a 514 struct cpl_t5_act_open_req *t5_req;
cfdda9d7
SW		 515 struct sk_buff *skb;
516 u64 opt0;
517 u32 opt2;
518 unsigned int mtu_idx;
519 int wscale;
f079af7a
VP		 520 int size = is_t4(ep->com.dev->rdev.lldi.adapter_type) ?
521 sizeof(struct cpl_act_open_req) :
522 sizeof(struct cpl_t5_act_open_req);
523 int wrlen = roundup(size, 16);
cfdda9d7
SW		 524
525 PDBG("%s ep %p atid %u\n", __func__, ep, ep->atid);
526
527 skb = get_skb(NULL, wrlen, GFP_KERNEL);
528 if (!skb) {
529 printk(KERN_ERR MOD "%s - failed to alloc skb.\n",
530 __func__);
531 return -ENOMEM;
532 }
d4f1a5c6 533 set_wr_txq(skb, CPL_PRIORITY_SETUP, ep->ctrlq_idx);
cfdda9d7
SW		 534
535 cxgb4_best_mtu(ep->com.dev->rdev.lldi.mtus, ep->mtu, &mtu_idx);
536 wscale = compute_wscale(rcv_win);
5be78ee9
VP		 537 opt0 = (nocong ? NO_CONG(1) : 0) |
538 KEEP_ALIVE(1) |
ba6d3925 539 DELACK(1) |
cfdda9d7
SW		 540 WND_SCALE(wscale) |
541 MSS_IDX(mtu_idx) |
542 L2T_IDX(ep->l2t->idx) |
543 TX_CHAN(ep->tx_chan) |
544 SMAC_SEL(ep->smac_idx) |
545 DSCP(ep->tos) |
b48f3b9c 546 ULP_MODE(ULP_MODE_TCPDDP) |
cfdda9d7
SW		 547 RCV_BUFSIZ(rcv_win>>10);
548 opt2 = RX_CHANNEL(0) |
5be78ee9 549 CCTRL_ECN(enable_ecn) |
cfdda9d7
SW		 550 RSS_QUEUE_VALID | RSS_QUEUE(ep->rss_qid);
551 if (enable_tcp_timestamps)
552 opt2 |= TSTAMPS_EN(1);
553 if (enable_tcp_sack)
554 opt2 |= SACK_EN(1);
555 if (wscale && enable_tcp_window_scaling)
556 opt2 |= WND_SCALE_EN(1);
557 t4_set_arp_err_handler(skb, NULL, act_open_req_arp_failure);
558
f079af7a
VP		 559 if (is_t4(ep->com.dev->rdev.lldi.adapter_type)) {
560 req = (struct cpl_act_open_req *) skb_put(skb, wrlen);
561 INIT_TP_WR(req, 0);
562 OPCODE_TID(req) = cpu_to_be32(
563 MK_OPCODE_TID(CPL_ACT_OPEN_REQ,
564 ((ep->rss_qid << 14) | ep->atid)));
565 req->local_port = ep->com.local_addr.sin_port;
566 req->peer_port = ep->com.remote_addr.sin_port;
567 req->local_ip = ep->com.local_addr.sin_addr.s_addr;
568 req->peer_ip = ep->com.remote_addr.sin_addr.s_addr;
569 req->opt0 = cpu_to_be64(opt0);
570 req->params = cpu_to_be32(select_ntuple(ep->com.dev,
571 ep->dst, ep->l2t));
572 req->opt2 = cpu_to_be32(opt2);
573 } else {
574 t5_req = (struct cpl_t5_act_open_req *) skb_put(skb, wrlen);
575 INIT_TP_WR(t5_req, 0);
576 OPCODE_TID(t5_req) = cpu_to_be32(
577 MK_OPCODE_TID(CPL_ACT_OPEN_REQ,
578 ((ep->rss_qid << 14) | ep->atid)));
579 t5_req->local_port = ep->com.local_addr.sin_port;
580 t5_req->peer_port = ep->com.remote_addr.sin_port;
581 t5_req->local_ip = ep->com.local_addr.sin_addr.s_addr;
582 t5_req->peer_ip = ep->com.remote_addr.sin_addr.s_addr;
583 t5_req->opt0 = cpu_to_be64(opt0);
584 t5_req->params = cpu_to_be64(V_FILTER_TUPLE(
585 select_ntuple(ep->com.dev, ep->dst, ep->l2t)));
586 t5_req->opt2 = cpu_to_be32(opt2);
587 }
588
793dad94 589 set_bit(ACT_OPEN_REQ, &ep->com.history);
cfdda9d7
SW		 590 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
591}
592
d2fe99e8
KS		 593static void send_mpa_req(struct c4iw_ep *ep, struct sk_buff *skb,
594 u8 mpa_rev_to_use)
cfdda9d7
SW		 595{
596 int mpalen, wrlen;
597 struct fw_ofld_tx_data_wr *req;
598 struct mpa_message *mpa;
d2fe99e8 599 struct mpa_v2_conn_params mpa_v2_params;
cfdda9d7
SW		 600
601 PDBG("%s ep %p tid %u pd_len %d\n", __func__, ep, ep->hwtid, ep->plen);
602
603 BUG_ON(skb_cloned(skb));
604
605 mpalen = sizeof(*mpa) + ep->plen;
d2fe99e8
KS		 606 if (mpa_rev_to_use == 2)
607 mpalen += sizeof(struct mpa_v2_conn_params);
cfdda9d7
SW		 608 wrlen = roundup(mpalen + sizeof *req, 16);
609 skb = get_skb(skb, wrlen, GFP_KERNEL);
610 if (!skb) {
611 connect_reply_upcall(ep, -ENOMEM);
612 return;
613 }
614 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
615
616 req = (struct fw_ofld_tx_data_wr *)skb_put(skb, wrlen);
617 memset(req, 0, wrlen);
618 req->op_to_immdlen = cpu_to_be32(
619 FW_WR_OP(FW_OFLD_TX_DATA_WR) |
620 FW_WR_COMPL(1) |
621 FW_WR_IMMDLEN(mpalen));
622 req->flowid_len16 = cpu_to_be32(
623 FW_WR_FLOWID(ep->hwtid) |
624 FW_WR_LEN16(wrlen >> 4));
625 req->plen = cpu_to_be32(mpalen);
626 req->tunnel_to_proxy = cpu_to_be32(
627 FW_OFLD_TX_DATA_WR_FLUSH(1) |
628 FW_OFLD_TX_DATA_WR_SHOVE(1));
629
630 mpa = (struct mpa_message *)(req + 1);
631 memcpy(mpa->key, MPA_KEY_REQ, sizeof(mpa->key));
632 mpa->flags = (crc_enabled ? MPA_CRC : 0) |
d2fe99e8
KS		 633 (markers_enabled ? MPA_MARKERS : 0) |
634 (mpa_rev_to_use == 2 ? MPA_ENHANCED_RDMA_CONN : 0);
cfdda9d7 635 mpa->private_data_size = htons(ep->plen);
d2fe99e8 636 mpa->revision = mpa_rev_to_use;
01b225e1 637 if (mpa_rev_to_use == 1) {
d2fe99e8 638 ep->tried_with_mpa_v1 = 1;
01b225e1
KS		 639 ep->retry_with_mpa_v1 = 0;
640 }
d2fe99e8
KS		 641
642 if (mpa_rev_to_use == 2) {
f747c34a
RD		 643 mpa->private_data_size = htons(ntohs(mpa->private_data_size) +
644 sizeof (struct mpa_v2_conn_params));
d2fe99e8
KS		 645 mpa_v2_params.ird = htons((u16)ep->ird);
646 mpa_v2_params.ord = htons((u16)ep->ord);
647
648 if (peer2peer) {
649 mpa_v2_params.ird |= htons(MPA_V2_PEER2PEER_MODEL);
650 if (p2p_type == FW_RI_INIT_P2PTYPE_RDMA_WRITE)
651 mpa_v2_params.ord |=
652 htons(MPA_V2_RDMA_WRITE_RTR);
653 else if (p2p_type == FW_RI_INIT_P2PTYPE_READ_REQ)
654 mpa_v2_params.ord |=
655 htons(MPA_V2_RDMA_READ_RTR);
656 }
657 memcpy(mpa->private_data, &mpa_v2_params,
658 sizeof(struct mpa_v2_conn_params));
cfdda9d7 659
d2fe99e8
KS		 660 if (ep->plen)
661 memcpy(mpa->private_data +
662 sizeof(struct mpa_v2_conn_params),
663 ep->mpa_pkt + sizeof(*mpa), ep->plen);
664 } else
665 if (ep->plen)
666 memcpy(mpa->private_data,
667 ep->mpa_pkt + sizeof(*mpa), ep->plen);
cfdda9d7
SW		 668
669 /*
670 * Reference the mpa skb. This ensures the data area
671 * will remain in memory until the hw acks the tx.
672 * Function fw4_ack() will deref it.
673 */
674 skb_get(skb);
675 t4_set_arp_err_handler(skb, NULL, arp_failure_discard);
676 BUG_ON(ep->mpa_skb);
677 ep->mpa_skb = skb;
678 c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
679 start_ep_timer(ep);
680 state_set(&ep->com, MPA_REQ_SENT);
681 ep->mpa_attr.initiator = 1;
682 return;
683}
684
685static int send_mpa_reject(struct c4iw_ep *ep, const void *pdata, u8 plen)
686{
687 int mpalen, wrlen;
688 struct fw_ofld_tx_data_wr *req;
689 struct mpa_message *mpa;
690 struct sk_buff *skb;
d2fe99e8 691 struct mpa_v2_conn_params mpa_v2_params;
cfdda9d7
SW		 692
693 PDBG("%s ep %p tid %u pd_len %d\n", __func__, ep, ep->hwtid, ep->plen);
694
695 mpalen = sizeof(*mpa) + plen;
d2fe99e8
KS		 696 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn)
697 mpalen += sizeof(struct mpa_v2_conn_params);
cfdda9d7
SW		 698 wrlen = roundup(mpalen + sizeof *req, 16);
699
700 skb = get_skb(NULL, wrlen, GFP_KERNEL);
701 if (!skb) {
702 printk(KERN_ERR MOD "%s - cannot alloc skb!\n", __func__);
703 return -ENOMEM;
704 }
705 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
706
707 req = (struct fw_ofld_tx_data_wr *)skb_put(skb, wrlen);
708 memset(req, 0, wrlen);
709 req->op_to_immdlen = cpu_to_be32(
710 FW_WR_OP(FW_OFLD_TX_DATA_WR) |
711 FW_WR_COMPL(1) |
712 FW_WR_IMMDLEN(mpalen));
713 req->flowid_len16 = cpu_to_be32(
714 FW_WR_FLOWID(ep->hwtid) |
715 FW_WR_LEN16(wrlen >> 4));
716 req->plen = cpu_to_be32(mpalen);
717 req->tunnel_to_proxy = cpu_to_be32(
718 FW_OFLD_TX_DATA_WR_FLUSH(1) |
719 FW_OFLD_TX_DATA_WR_SHOVE(1));
720
721 mpa = (struct mpa_message *)(req + 1);
722 memset(mpa, 0, sizeof(*mpa));
723 memcpy(mpa->key, MPA_KEY_REP, sizeof(mpa->key));
724 mpa->flags = MPA_REJECT;
fe7e0a4d 725 mpa->revision = ep->mpa_attr.version;
cfdda9d7 726 mpa->private_data_size = htons(plen);
d2fe99e8
KS		 727
728 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) {
729 mpa->flags |= MPA_ENHANCED_RDMA_CONN;
f747c34a
RD		 730 mpa->private_data_size = htons(ntohs(mpa->private_data_size) +
731 sizeof (struct mpa_v2_conn_params));
d2fe99e8
KS		 732 mpa_v2_params.ird = htons(((u16)ep->ird) |
733 (peer2peer ? MPA_V2_PEER2PEER_MODEL :
734 0));
735 mpa_v2_params.ord = htons(((u16)ep->ord) | (peer2peer ?
736 (p2p_type ==
737 FW_RI_INIT_P2PTYPE_RDMA_WRITE ?
738 MPA_V2_RDMA_WRITE_RTR : p2p_type ==
739 FW_RI_INIT_P2PTYPE_READ_REQ ?
740 MPA_V2_RDMA_READ_RTR : 0) : 0));
741 memcpy(mpa->private_data, &mpa_v2_params,
742 sizeof(struct mpa_v2_conn_params));
743
744 if (ep->plen)
745 memcpy(mpa->private_data +
746 sizeof(struct mpa_v2_conn_params), pdata, plen);
747 } else
748 if (plen)
749 memcpy(mpa->private_data, pdata, plen);
cfdda9d7
SW		 750
751 /*
752 * Reference the mpa skb again. This ensures the data area
753 * will remain in memory until the hw acks the tx.
754 * Function fw4_ack() will deref it.
755 */
756 skb_get(skb);
757 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
758 t4_set_arp_err_handler(skb, NULL, arp_failure_discard);
759 BUG_ON(ep->mpa_skb);
760 ep->mpa_skb = skb;
761 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
762}
763
764static int send_mpa_reply(struct c4iw_ep *ep, const void *pdata, u8 plen)
765{
766 int mpalen, wrlen;
767 struct fw_ofld_tx_data_wr *req;
768 struct mpa_message *mpa;
769 struct sk_buff *skb;
d2fe99e8 770 struct mpa_v2_conn_params mpa_v2_params;
cfdda9d7
SW		 771
772 PDBG("%s ep %p tid %u pd_len %d\n", __func__, ep, ep->hwtid, ep->plen);
773
774 mpalen = sizeof(*mpa) + plen;
d2fe99e8
KS		 775 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn)
776 mpalen += sizeof(struct mpa_v2_conn_params);
cfdda9d7
SW		 777 wrlen = roundup(mpalen + sizeof *req, 16);
778
779 skb = get_skb(NULL, wrlen, GFP_KERNEL);
780 if (!skb) {
781 printk(KERN_ERR MOD "%s - cannot alloc skb!\n", __func__);
782 return -ENOMEM;
783 }
784 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
785
786 req = (struct fw_ofld_tx_data_wr *) skb_put(skb, wrlen);
787 memset(req, 0, wrlen);
788 req->op_to_immdlen = cpu_to_be32(
789 FW_WR_OP(FW_OFLD_TX_DATA_WR) |
790 FW_WR_COMPL(1) |
791 FW_WR_IMMDLEN(mpalen));
792 req->flowid_len16 = cpu_to_be32(
793 FW_WR_FLOWID(ep->hwtid) |
794 FW_WR_LEN16(wrlen >> 4));
795 req->plen = cpu_to_be32(mpalen);
796 req->tunnel_to_proxy = cpu_to_be32(
797 FW_OFLD_TX_DATA_WR_FLUSH(1) |
798 FW_OFLD_TX_DATA_WR_SHOVE(1));
799
800 mpa = (struct mpa_message *)(req + 1);
801 memset(mpa, 0, sizeof(*mpa));
802 memcpy(mpa->key, MPA_KEY_REP, sizeof(mpa->key));
803 mpa->flags = (ep->mpa_attr.crc_enabled ? MPA_CRC : 0) |
804 (markers_enabled ? MPA_MARKERS : 0);
d2fe99e8 805 mpa->revision = ep->mpa_attr.version;
cfdda9d7 806 mpa->private_data_size = htons(plen);
d2fe99e8
KS		 807
808 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) {
809 mpa->flags |= MPA_ENHANCED_RDMA_CONN;
f747c34a
RD		 810 mpa->private_data_size = htons(ntohs(mpa->private_data_size) +
811 sizeof (struct mpa_v2_conn_params));
d2fe99e8
KS		 812 mpa_v2_params.ird = htons((u16)ep->ird);
813 mpa_v2_params.ord = htons((u16)ep->ord);
814 if (peer2peer && (ep->mpa_attr.p2p_type !=
815 FW_RI_INIT_P2PTYPE_DISABLED)) {
816 mpa_v2_params.ird |= htons(MPA_V2_PEER2PEER_MODEL);
817
818 if (p2p_type == FW_RI_INIT_P2PTYPE_RDMA_WRITE)
819 mpa_v2_params.ord |=
820 htons(MPA_V2_RDMA_WRITE_RTR);
821 else if (p2p_type == FW_RI_INIT_P2PTYPE_READ_REQ)
822 mpa_v2_params.ord |=
823 htons(MPA_V2_RDMA_READ_RTR);
824 }
825
826 memcpy(mpa->private_data, &mpa_v2_params,
827 sizeof(struct mpa_v2_conn_params));
828
829 if (ep->plen)
830 memcpy(mpa->private_data +
831 sizeof(struct mpa_v2_conn_params), pdata, plen);
832 } else
833 if (plen)
834 memcpy(mpa->private_data, pdata, plen);
cfdda9d7
SW		 835
836 /*
837 * Reference the mpa skb. This ensures the data area
838 * will remain in memory until the hw acks the tx.
839 * Function fw4_ack() will deref it.
840 */
841 skb_get(skb);
842 t4_set_arp_err_handler(skb, NULL, arp_failure_discard);
843 ep->mpa_skb = skb;
844 state_set(&ep->com, MPA_REP_SENT);
845 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
846}
847
848static int act_establish(struct c4iw_dev *dev, struct sk_buff *skb)
849{
850 struct c4iw_ep *ep;
851 struct cpl_act_establish *req = cplhdr(skb);
852 unsigned int tid = GET_TID(req);
853 unsigned int atid = GET_TID_TID(ntohl(req->tos_atid));
854 struct tid_info *t = dev->rdev.lldi.tids;
855
856 ep = lookup_atid(t, atid);
857
858 PDBG("%s ep %p tid %u snd_isn %u rcv_isn %u\n", __func__, ep, tid,
859 be32_to_cpu(req->snd_isn), be32_to_cpu(req->rcv_isn));
860
861 dst_confirm(ep->dst);
862
863 /* setup the hwtid for this connection */
864 ep->hwtid = tid;
865 cxgb4_insert_tid(t, ep, tid);
793dad94 866 insert_handle(dev, &dev->hwtid_idr, ep, ep->hwtid);
cfdda9d7
SW		 867
868 ep->snd_seq = be32_to_cpu(req->snd_isn);
869 ep->rcv_seq = be32_to_cpu(req->rcv_isn);
870
871 set_emss(ep, ntohs(req->tcp_opt));
872
873 /* dealloc the atid */
793dad94 874 remove_handle(ep->com.dev, &ep->com.dev->atid_idr, atid);
cfdda9d7 875 cxgb4_free_atid(t, atid);
793dad94 876 set_bit(ACT_ESTAB, &ep->com.history);
cfdda9d7
SW		 877
878 /* start MPA negotiation */
879 send_flowc(ep, NULL);
d2fe99e8
KS		 880 if (ep->retry_with_mpa_v1)
881 send_mpa_req(ep, skb, 1);
882 else
883 send_mpa_req(ep, skb, mpa_rev);
cfdda9d7
SW		 884
885 return 0;
886}
887
888static void close_complete_upcall(struct c4iw_ep *ep)
889{
890 struct iw_cm_event event;
891
892 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
893 memset(&event, 0, sizeof(event));
894 event.event = IW_CM_EVENT_CLOSE;
895 if (ep->com.cm_id) {
896 PDBG("close complete delivered ep %p cm_id %p tid %u\n",
897 ep, ep->com.cm_id, ep->hwtid);
898 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
899 ep->com.cm_id->rem_ref(ep->com.cm_id);
900 ep->com.cm_id = NULL;
793dad94 901 set_bit(CLOSE_UPCALL, &ep->com.history);
cfdda9d7
SW		 902 }
903}
904
905static int abort_connection(struct c4iw_ep *ep, struct sk_buff *skb, gfp_t gfp)
906{
907 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
908 close_complete_upcall(ep);
909 state_set(&ep->com, ABORTING);
793dad94 910 set_bit(ABORT_CONN, &ep->com.history);
cfdda9d7
SW		 911 return send_abort(ep, skb, gfp);
912}
913
914static void peer_close_upcall(struct c4iw_ep *ep)
915{
916 struct iw_cm_event event;
917
918 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
919 memset(&event, 0, sizeof(event));
920 event.event = IW_CM_EVENT_DISCONNECT;
921 if (ep->com.cm_id) {
922 PDBG("peer close delivered ep %p cm_id %p tid %u\n",
923 ep, ep->com.cm_id, ep->hwtid);
924 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
793dad94 925 set_bit(DISCONN_UPCALL, &ep->com.history);
cfdda9d7
SW		 926 }
927}
928
929static void peer_abort_upcall(struct c4iw_ep *ep)
930{
931 struct iw_cm_event event;
932
933 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
934 memset(&event, 0, sizeof(event));
935 event.event = IW_CM_EVENT_CLOSE;
936 event.status = -ECONNRESET;
937 if (ep->com.cm_id) {
938 PDBG("abort delivered ep %p cm_id %p tid %u\n", ep,
939 ep->com.cm_id, ep->hwtid);
940 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
941 ep->com.cm_id->rem_ref(ep->com.cm_id);
942 ep->com.cm_id = NULL;
793dad94 943 set_bit(ABORT_UPCALL, &ep->com.history);
cfdda9d7
SW		 944 }
945}
946
947static void connect_reply_upcall(struct c4iw_ep *ep, int status)
948{
949 struct iw_cm_event event;
950
951 PDBG("%s ep %p tid %u status %d\n", __func__, ep, ep->hwtid, status);
952 memset(&event, 0, sizeof(event));
953 event.event = IW_CM_EVENT_CONNECT_REPLY;
954 event.status = status;
24d44a39
SW		 955 memcpy(&event.local_addr, &ep->com.local_addr,
956 sizeof(ep->com.local_addr));
957 memcpy(&event.remote_addr, &ep->com.remote_addr,
958 sizeof(ep->com.remote_addr));
cfdda9d7
SW		 959
960 if ((status == 0) || (status == -ECONNREFUSED)) {
d2fe99e8
KS		 961 if (!ep->tried_with_mpa_v1) {
962 /* this means MPA_v2 is used */
963 event.private_data_len = ep->plen -
964 sizeof(struct mpa_v2_conn_params);
965 event.private_data = ep->mpa_pkt +
966 sizeof(struct mpa_message) +
967 sizeof(struct mpa_v2_conn_params);
968 } else {
969 /* this means MPA_v1 is used */
970 event.private_data_len = ep->plen;
971 event.private_data = ep->mpa_pkt +
972 sizeof(struct mpa_message);
973 }
cfdda9d7 974 }
85963e4c
RD		 975
976 PDBG("%s ep %p tid %u status %d\n", __func__, ep,
977 ep->hwtid, status);
793dad94 978 set_bit(CONN_RPL_UPCALL, &ep->com.history);
85963e4c
RD		 979 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
980
cfdda9d7
SW		 981 if (status < 0) {
982 ep->com.cm_id->rem_ref(ep->com.cm_id);
983 ep->com.cm_id = NULL;
cfdda9d7
SW		 984 }
985}
986
987static void connect_request_upcall(struct c4iw_ep *ep)
988{
989 struct iw_cm_event event;
990
991 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
992 memset(&event, 0, sizeof(event));
993 event.event = IW_CM_EVENT_CONNECT_REQUEST;
24d44a39
SW		 994 memcpy(&event.local_addr, &ep->com.local_addr,
995 sizeof(ep->com.local_addr));
996 memcpy(&event.remote_addr, &ep->com.remote_addr,
997 sizeof(ep->com.remote_addr));
cfdda9d7 998 event.provider_data = ep;
d2fe99e8
KS		 999 if (!ep->tried_with_mpa_v1) {
1000 /* this means MPA_v2 is used */
1001 event.ord = ep->ord;
1002 event.ird = ep->ird;
1003 event.private_data_len = ep->plen -
1004 sizeof(struct mpa_v2_conn_params);
1005 event.private_data = ep->mpa_pkt + sizeof(struct mpa_message) +
1006 sizeof(struct mpa_v2_conn_params);
1007 } else {
1008 /* this means MPA_v1 is used. Send max supported */
1009 event.ord = c4iw_max_read_depth;
1010 event.ird = c4iw_max_read_depth;
1011 event.private_data_len = ep->plen;
1012 event.private_data = ep->mpa_pkt + sizeof(struct mpa_message);
1013 }
cfdda9d7
SW		 1014 if (state_read(&ep->parent_ep->com) != DEAD) {
1015 c4iw_get_ep(&ep->com);
1016 ep->parent_ep->com.cm_id->event_handler(
1017 ep->parent_ep->com.cm_id,
1018 &event);
1019 }
793dad94 1020 set_bit(CONNREQ_UPCALL, &ep->com.history);
cfdda9d7
SW		 1021 c4iw_put_ep(&ep->parent_ep->com);
1022 ep->parent_ep = NULL;
1023}
1024
1025static void established_upcall(struct c4iw_ep *ep)
1026{
1027 struct iw_cm_event event;
1028
1029 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1030 memset(&event, 0, sizeof(event));
1031 event.event = IW_CM_EVENT_ESTABLISHED;
d2fe99e8
KS		 1032 event.ird = ep->ird;
1033 event.ord = ep->ord;
cfdda9d7
SW		 1034 if (ep->com.cm_id) {
1035 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1036 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
793dad94 1037 set_bit(ESTAB_UPCALL, &ep->com.history);
cfdda9d7
SW		 1038 }
1039}
1040
1041static int update_rx_credits(struct c4iw_ep *ep, u32 credits)
1042{
1043 struct cpl_rx_data_ack *req;
1044 struct sk_buff *skb;
1045 int wrlen = roundup(sizeof *req, 16);
1046
1047 PDBG("%s ep %p tid %u credits %u\n", __func__, ep, ep->hwtid, credits);
1048 skb = get_skb(NULL, wrlen, GFP_KERNEL);
1049 if (!skb) {
1050 printk(KERN_ERR MOD "update_rx_credits - cannot alloc skb!\n");
1051 return 0;
1052 }
1053
1054 req = (struct cpl_rx_data_ack *) skb_put(skb, wrlen);
1055 memset(req, 0, wrlen);
1056 INIT_TP_WR(req, ep->hwtid);
1057 OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_RX_DATA_ACK,
1058 ep->hwtid));
ba6d3925
SW		 1059 req->credit_dack = cpu_to_be32(credits | RX_FORCE_ACK(1) |
1060 F_RX_DACK_CHANGE |
1061 V_RX_DACK_MODE(dack_mode));
d4f1a5c6 1062 set_wr_txq(skb, CPL_PRIORITY_ACK, ep->ctrlq_idx);
cfdda9d7
SW		 1063 c4iw_ofld_send(&ep->com.dev->rdev, skb);
1064 return credits;
1065}
1066
1067static void process_mpa_reply(struct c4iw_ep *ep, struct sk_buff *skb)
1068{
1069 struct mpa_message *mpa;
d2fe99e8 1070 struct mpa_v2_conn_params *mpa_v2_params;
cfdda9d7 1071 u16 plen;
d2fe99e8
KS		 1072 u16 resp_ird, resp_ord;
1073 u8 rtr_mismatch = 0, insuff_ird = 0;
cfdda9d7
SW		 1074 struct c4iw_qp_attributes attrs;
1075 enum c4iw_qp_attr_mask mask;
1076 int err;
1077
1078 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1079
1080 /*
1081 * Stop mpa timer. If it expired, then the state has
1082 * changed and we bail since ep_timeout already aborted
1083 * the connection.
1084 */
1085 stop_ep_timer(ep);
1086 if (state_read(&ep->com) != MPA_REQ_SENT)
1087 return;
1088
1089 /*
1090 * If we get more than the supported amount of private data
1091 * then we must fail this connection.
1092 */
1093 if (ep->mpa_pkt_len + skb->len > sizeof(ep->mpa_pkt)) {
1094 err = -EINVAL;
1095 goto err;
1096 }
1097
1098 /*
1099 * copy the new data into our accumulation buffer.
1100 */
1101 skb_copy_from_linear_data(skb, &(ep->mpa_pkt[ep->mpa_pkt_len]),
1102 skb->len);
1103 ep->mpa_pkt_len += skb->len;
1104
1105 /*
1106 * if we don't even have the mpa message, then bail.
1107 */
1108 if (ep->mpa_pkt_len < sizeof(*mpa))
1109 return;
1110 mpa = (struct mpa_message *) ep->mpa_pkt;
1111
1112 /* Validate MPA header. */
d2fe99e8
KS		 1113 if (mpa->revision > mpa_rev) {
1114 printk(KERN_ERR MOD "%s MPA version mismatch. Local = %d,"
1115 " Received = %d\n", __func__, mpa_rev, mpa->revision);
cfdda9d7
SW		 1116 err = -EPROTO;
1117 goto err;
1118 }
1119 if (memcmp(mpa->key, MPA_KEY_REP, sizeof(mpa->key))) {
1120 err = -EPROTO;
1121 goto err;
1122 }
1123
1124 plen = ntohs(mpa->private_data_size);
1125
1126 /*
1127 * Fail if there's too much private data.
1128 */
1129 if (plen > MPA_MAX_PRIVATE_DATA) {
1130 err = -EPROTO;
1131 goto err;
1132 }
1133
1134 /*
1135 * If plen does not account for pkt size
1136 */
1137 if (ep->mpa_pkt_len > (sizeof(*mpa) + plen)) {
1138 err = -EPROTO;
1139 goto err;
1140 }
1141
1142 ep->plen = (u8) plen;
1143
1144 /*
1145 * If we don't have all the pdata yet, then bail.
1146 * We'll continue process when more data arrives.
1147 */
1148 if (ep->mpa_pkt_len < (sizeof(*mpa) + plen))
1149 return;
1150
1151 if (mpa->flags & MPA_REJECT) {
1152 err = -ECONNREFUSED;
1153 goto err;
1154 }
1155
1156 /*
1157 * If we get here we have accumulated the entire mpa
1158 * start reply message including private data. And
1159 * the MPA header is valid.
1160 */
1161 state_set(&ep->com, FPDU_MODE);
1162 ep->mpa_attr.crc_enabled = (mpa->flags & MPA_CRC) | crc_enabled ? 1 : 0;
1163 ep->mpa_attr.recv_marker_enabled = markers_enabled;
1164 ep->mpa_attr.xmit_marker_enabled = mpa->flags & MPA_MARKERS ? 1 : 0;
d2fe99e8
KS		 1165 ep->mpa_attr.version = mpa->revision;
1166 ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED;
1167
1168 if (mpa->revision == 2) {
1169 ep->mpa_attr.enhanced_rdma_conn =
1170 mpa->flags & MPA_ENHANCED_RDMA_CONN ? 1 : 0;
1171 if (ep->mpa_attr.enhanced_rdma_conn) {
1172 mpa_v2_params = (struct mpa_v2_conn_params *)
1173 (ep->mpa_pkt + sizeof(*mpa));
1174 resp_ird = ntohs(mpa_v2_params->ird) &
1175 MPA_V2_IRD_ORD_MASK;
1176 resp_ord = ntohs(mpa_v2_params->ord) &
1177 MPA_V2_IRD_ORD_MASK;
1178
1179 /*
1180 * This is a double-check. Ideally, below checks are
1181 * not required since ird/ord stuff has been taken
1182 * care of in c4iw_accept_cr
1183 */
1184 if ((ep->ird < resp_ord) || (ep->ord > resp_ird)) {
1185 err = -ENOMEM;
1186 ep->ird = resp_ord;
1187 ep->ord = resp_ird;
1188 insuff_ird = 1;
1189 }
1190
1191 if (ntohs(mpa_v2_params->ird) &
1192 MPA_V2_PEER2PEER_MODEL) {
1193 if (ntohs(mpa_v2_params->ord) &
1194 MPA_V2_RDMA_WRITE_RTR)
1195 ep->mpa_attr.p2p_type =
1196 FW_RI_INIT_P2PTYPE_RDMA_WRITE;
1197 else if (ntohs(mpa_v2_params->ord) &
1198 MPA_V2_RDMA_READ_RTR)
1199 ep->mpa_attr.p2p_type =
1200 FW_RI_INIT_P2PTYPE_READ_REQ;
1201 }
1202 }
1203 } else if (mpa->revision == 1)
1204 if (peer2peer)
1205 ep->mpa_attr.p2p_type = p2p_type;
1206
cfdda9d7 1207 PDBG("%s - crc_enabled=%d, recv_marker_enabled=%d, "
d2fe99e8
KS		 1208 "xmit_marker_enabled=%d, version=%d p2p_type=%d local-p2p_type = "
1209 "%d\n", __func__, ep->mpa_attr.crc_enabled,
1210 ep->mpa_attr.recv_marker_enabled,
1211 ep->mpa_attr.xmit_marker_enabled, ep->mpa_attr.version,
1212 ep->mpa_attr.p2p_type, p2p_type);
1213
1214 /*
1215 * If responder's RTR does not match with that of initiator, assign
1216 * FW_RI_INIT_P2PTYPE_DISABLED in mpa attributes so that RTR is not
1217 * generated when moving QP to RTS state.
1218 * A TERM message will be sent after QP has moved to RTS state
1219 */
91018f86 1220 if ((ep->mpa_attr.version == 2) && peer2peer &&
d2fe99e8
KS		 1221 (ep->mpa_attr.p2p_type != p2p_type)) {
1222 ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED;
1223 rtr_mismatch = 1;
1224 }
cfdda9d7
SW		 1225
1226 attrs.mpa_attr = ep->mpa_attr;
1227 attrs.max_ird = ep->ird;
1228 attrs.max_ord = ep->ord;
1229 attrs.llp_stream_handle = ep;
1230 attrs.next_state = C4IW_QP_STATE_RTS;
1231
1232 mask = C4IW_QP_ATTR_NEXT_STATE |
1233 C4IW_QP_ATTR_LLP_STREAM_HANDLE | C4IW_QP_ATTR_MPA_ATTR |
1234 C4IW_QP_ATTR_MAX_IRD | C4IW_QP_ATTR_MAX_ORD;
1235
1236 /* bind QP and TID with INIT_WR */
1237 err = c4iw_modify_qp(ep->com.qp->rhp,
1238 ep->com.qp, mask, &attrs, 1);
1239 if (err)
1240 goto err;
d2fe99e8
KS		 1241
1242 /*
1243 * If responder's RTR requirement did not match with what initiator
1244 * supports, generate TERM message
1245 */
1246 if (rtr_mismatch) {
1247 printk(KERN_ERR "%s: RTR mismatch, sending TERM\n", __func__);
1248 attrs.layer_etype = LAYER_MPA | DDP_LLP;
1249 attrs.ecode = MPA_NOMATCH_RTR;
1250 attrs.next_state = C4IW_QP_STATE_TERMINATE;
1251 err = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
1252 C4IW_QP_ATTR_NEXT_STATE, &attrs, 0);
1253 err = -ENOMEM;
1254 goto out;
1255 }
1256
1257 /*
1258 * Generate TERM if initiator IRD is not sufficient for responder
1259 * provided ORD. Currently, we do the same behaviour even when
1260 * responder provided IRD is also not sufficient as regards to
1261 * initiator ORD.
1262 */
1263 if (insuff_ird) {
1264 printk(KERN_ERR "%s: Insufficient IRD, sending TERM\n",
1265 __func__);
1266 attrs.layer_etype = LAYER_MPA | DDP_LLP;
1267 attrs.ecode = MPA_INSUFF_IRD;
1268 attrs.next_state = C4IW_QP_STATE_TERMINATE;
1269 err = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
1270 C4IW_QP_ATTR_NEXT_STATE, &attrs, 0);
1271 err = -ENOMEM;
1272 goto out;
1273 }
cfdda9d7
SW		 1274 goto out;
1275err:
b21ef16a
SW		 1276 state_set(&ep->com, ABORTING);
1277 send_abort(ep, skb, GFP_KERNEL);
cfdda9d7
SW		 1278out:
1279 connect_reply_upcall(ep, err);
1280 return;
1281}
1282
1283static void process_mpa_request(struct c4iw_ep *ep, struct sk_buff *skb)
1284{
1285 struct mpa_message *mpa;
d2fe99e8 1286 struct mpa_v2_conn_params *mpa_v2_params;
cfdda9d7
SW		 1287 u16 plen;
1288
1289 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1290
1291 if (state_read(&ep->com) != MPA_REQ_WAIT)
1292 return;
1293
1294 /*
1295 * If we get more than the supported amount of private data
1296 * then we must fail this connection.
1297 */
1298 if (ep->mpa_pkt_len + skb->len > sizeof(ep->mpa_pkt)) {
1299 stop_ep_timer(ep);
1300 abort_connection(ep, skb, GFP_KERNEL);
1301 return;
1302 }
1303
1304 PDBG("%s enter (%s line %u)\n", __func__, __FILE__, __LINE__);
1305
1306 /*
1307 * Copy the new data into our accumulation buffer.
1308 */
1309 skb_copy_from_linear_data(skb, &(ep->mpa_pkt[ep->mpa_pkt_len]),
1310 skb->len);
1311 ep->mpa_pkt_len += skb->len;
1312
1313 /*
1314 * If we don't even have the mpa message, then bail.
1315 * We'll continue process when more data arrives.
1316 */
1317 if (ep->mpa_pkt_len < sizeof(*mpa))
1318 return;
1319
1320 PDBG("%s enter (%s line %u)\n", __func__, __FILE__, __LINE__);
1321 stop_ep_timer(ep);
1322 mpa = (struct mpa_message *) ep->mpa_pkt;
1323
1324 /*
1325 * Validate MPA Header.
1326 */
d2fe99e8
KS		 1327 if (mpa->revision > mpa_rev) {
1328 printk(KERN_ERR MOD "%s MPA version mismatch. Local = %d,"
1329 " Received = %d\n", __func__, mpa_rev, mpa->revision);
7c0a33d6 1330 stop_ep_timer(ep);
cfdda9d7
SW		 1331 abort_connection(ep, skb, GFP_KERNEL);
1332 return;
1333 }
1334
1335 if (memcmp(mpa->key, MPA_KEY_REQ, sizeof(mpa->key))) {
7c0a33d6 1336 stop_ep_timer(ep);
cfdda9d7
SW		 1337 abort_connection(ep, skb, GFP_KERNEL);
1338 return;
1339 }
1340
1341 plen = ntohs(mpa->private_data_size);
1342
1343 /*
1344 * Fail if there's too much private data.
1345 */
1346 if (plen > MPA_MAX_PRIVATE_DATA) {
7c0a33d6 1347 stop_ep_timer(ep);
cfdda9d7
SW		 1348 abort_connection(ep, skb, GFP_KERNEL);
1349 return;
1350 }
1351
1352 /*
1353 * If plen does not account for pkt size
1354 */
1355 if (ep->mpa_pkt_len > (sizeof(*mpa) + plen)) {
7c0a33d6 1356 stop_ep_timer(ep);
cfdda9d7
SW		 1357 abort_connection(ep, skb, GFP_KERNEL);
1358 return;
1359 }
1360 ep->plen = (u8) plen;
1361
1362 /*
1363 * If we don't have all the pdata yet, then bail.
1364 */
1365 if (ep->mpa_pkt_len < (sizeof(*mpa) + plen))
1366 return;
1367
1368 /*
1369 * If we get here we have accumulated the entire mpa
1370 * start reply message including private data.
1371 */
1372 ep->mpa_attr.initiator = 0;
1373 ep->mpa_attr.crc_enabled = (mpa->flags & MPA_CRC) | crc_enabled ? 1 : 0;
1374 ep->mpa_attr.recv_marker_enabled = markers_enabled;
1375 ep->mpa_attr.xmit_marker_enabled = mpa->flags & MPA_MARKERS ? 1 : 0;
d2fe99e8
KS		 1376 ep->mpa_attr.version = mpa->revision;
1377 if (mpa->revision == 1)
1378 ep->tried_with_mpa_v1 = 1;
1379 ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED;
1380
1381 if (mpa->revision == 2) {
1382 ep->mpa_attr.enhanced_rdma_conn =
1383 mpa->flags & MPA_ENHANCED_RDMA_CONN ? 1 : 0;
1384 if (ep->mpa_attr.enhanced_rdma_conn) {
1385 mpa_v2_params = (struct mpa_v2_conn_params *)
1386 (ep->mpa_pkt + sizeof(*mpa));
1387 ep->ird = ntohs(mpa_v2_params->ird) &
1388 MPA_V2_IRD_ORD_MASK;
1389 ep->ord = ntohs(mpa_v2_params->ord) &
1390 MPA_V2_IRD_ORD_MASK;
1391 if (ntohs(mpa_v2_params->ird) & MPA_V2_PEER2PEER_MODEL)
1392 if (peer2peer) {
1393 if (ntohs(mpa_v2_params->ord) &
1394 MPA_V2_RDMA_WRITE_RTR)
1395 ep->mpa_attr.p2p_type =
1396 FW_RI_INIT_P2PTYPE_RDMA_WRITE;
1397 else if (ntohs(mpa_v2_params->ord) &
1398 MPA_V2_RDMA_READ_RTR)
1399 ep->mpa_attr.p2p_type =
1400 FW_RI_INIT_P2PTYPE_READ_REQ;
1401 }
1402 }
1403 } else if (mpa->revision == 1)
1404 if (peer2peer)
1405 ep->mpa_attr.p2p_type = p2p_type;
1406
cfdda9d7
SW		 1407 PDBG("%s - crc_enabled=%d, recv_marker_enabled=%d, "
1408 "xmit_marker_enabled=%d, version=%d p2p_type=%d\n", __func__,
1409 ep->mpa_attr.crc_enabled, ep->mpa_attr.recv_marker_enabled,
1410 ep->mpa_attr.xmit_marker_enabled, ep->mpa_attr.version,
1411 ep->mpa_attr.p2p_type);
1412
1413 state_set(&ep->com, MPA_REQ_RCVD);
1414
1415 /* drive upcall */
1416 connect_request_upcall(ep);
1417 return;
1418}
1419
1420static int rx_data(struct c4iw_dev *dev, struct sk_buff *skb)
1421{
1422 struct c4iw_ep *ep;
1423 struct cpl_rx_data *hdr = cplhdr(skb);
1424 unsigned int dlen = ntohs(hdr->len);
1425 unsigned int tid = GET_TID(hdr);
1426 struct tid_info *t = dev->rdev.lldi.tids;
793dad94 1427 __u8 status = hdr->status;
cfdda9d7
SW		 1428
1429 ep = lookup_tid(t, tid);
1430 PDBG("%s ep %p tid %u dlen %u\n", __func__, ep, ep->hwtid, dlen);
1431 skb_pull(skb, sizeof(*hdr));
1432 skb_trim(skb, dlen);
1433
cfdda9d7
SW		 1434 /* update RX credits */
1435 update_rx_credits(ep, dlen);
1436
1437 switch (state_read(&ep->com)) {
1438 case MPA_REQ_SENT:
55abf8df 1439 ep->rcv_seq += dlen;
cfdda9d7
SW		 1440 process_mpa_reply(ep, skb);
1441 break;
1442 case MPA_REQ_WAIT:
55abf8df 1443 ep->rcv_seq += dlen;
cfdda9d7
SW		 1444 process_mpa_request(ep, skb);
1445 break;
1557967b
VP		 1446 case FPDU_MODE: {
1447 struct c4iw_qp_attributes attrs;
1448 BUG_ON(!ep->com.qp);
e8e5b927 1449 if (status)
1557967b 1450 pr_err("%s Unexpected streaming data." \
04236df2
VP		 1451 " qpid %u ep %p state %d tid %u status %d\n",
1452 __func__, ep->com.qp->wq.sq.qid, ep,
1453 state_read(&ep->com), ep->hwtid, status);
1557967b
VP		 1454 attrs.next_state = C4IW_QP_STATE_ERROR;
1455 c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
1456 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
55abf8df 1457 c4iw_ep_disconnect(ep, 1, GFP_KERNEL);
cfdda9d7
SW		 1458 break;
1459 }
1557967b
VP		 1460 default:
1461 break;
1462 }
cfdda9d7
SW		 1463 return 0;
1464}
1465
1466static int abort_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
1467{
1468 struct c4iw_ep *ep;
1469 struct cpl_abort_rpl_rss *rpl = cplhdr(skb);
cfdda9d7
SW		 1470 int release = 0;
1471 unsigned int tid = GET_TID(rpl);
1472 struct tid_info *t = dev->rdev.lldi.tids;
1473
1474 ep = lookup_tid(t, tid);
4984037b
VP		 1475 if (!ep) {
1476 printk(KERN_WARNING MOD "Abort rpl to freed endpoint\n");
1477 return 0;
1478 }
92dd6c3d 1479 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
2f5b48c3 1480 mutex_lock(&ep->com.mutex);
cfdda9d7
SW		 1481 switch (ep->com.state) {
1482 case ABORTING:
91e9c071 1483 c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
cfdda9d7
SW		 1484 __state_set(&ep->com, DEAD);
1485 release = 1;
1486 break;
1487 default:
1488 printk(KERN_ERR "%s ep %p state %d\n",
1489 __func__, ep, ep->com.state);
1490 break;
1491 }
2f5b48c3 1492 mutex_unlock(&ep->com.mutex);
cfdda9d7
SW		 1493
1494 if (release)
1495 release_ep_resources(ep);
1496 return 0;
1497}
1498
5be78ee9
VP		 1499static void send_fw_act_open_req(struct c4iw_ep *ep, unsigned int atid)
1500{
1501 struct sk_buff *skb;
1502 struct fw_ofld_connection_wr *req;
1503 unsigned int mtu_idx;
1504 int wscale;
1505
1506 skb = get_skb(NULL, sizeof(*req), GFP_KERNEL);
1507 req = (struct fw_ofld_connection_wr *)__skb_put(skb, sizeof(*req));
1508 memset(req, 0, sizeof(*req));
1509 req->op_compl = htonl(V_WR_OP(FW_OFLD_CONNECTION_WR));
1510 req->len16_pkd = htonl(FW_WR_LEN16(DIV_ROUND_UP(sizeof(*req), 16)));
1511 req->le.filter = cpu_to_be32(select_ntuple(ep->com.dev, ep->dst,
1512 ep->l2t));
1513 req->le.lport = ep->com.local_addr.sin_port;
1514 req->le.pport = ep->com.remote_addr.sin_port;
1515 req->le.u.ipv4.lip = ep->com.local_addr.sin_addr.s_addr;
1516 req->le.u.ipv4.pip = ep->com.remote_addr.sin_addr.s_addr;
1517 req->tcb.t_state_to_astid =
1518 htonl(V_FW_OFLD_CONNECTION_WR_T_STATE(TCP_SYN_SENT) |
1519 V_FW_OFLD_CONNECTION_WR_ASTID(atid));
1520 req->tcb.cplrxdataack_cplpassacceptrpl =
1521 htons(F_FW_OFLD_CONNECTION_WR_CPLRXDATAACK);
ef5d6355 1522 req->tcb.tx_max = (__force __be32) jiffies;
793dad94 1523 req->tcb.rcv_adv = htons(1);
5be78ee9
VP		 1524 cxgb4_best_mtu(ep->com.dev->rdev.lldi.mtus, ep->mtu, &mtu_idx);
1525 wscale = compute_wscale(rcv_win);
ef5d6355 1526 req->tcb.opt0 = (__force __be64) (TCAM_BYPASS(1) |
5be78ee9
VP		 1527 (nocong ? NO_CONG(1) : 0) |
1528 KEEP_ALIVE(1) |
1529 DELACK(1) |
1530 WND_SCALE(wscale) |
1531 MSS_IDX(mtu_idx) |
1532 L2T_IDX(ep->l2t->idx) |
1533 TX_CHAN(ep->tx_chan) |
1534 SMAC_SEL(ep->smac_idx) |
1535 DSCP(ep->tos) |
1536 ULP_MODE(ULP_MODE_TCPDDP) |
ef5d6355
VP		 1537 RCV_BUFSIZ(rcv_win >> 10));
1538 req->tcb.opt2 = (__force __be32) (PACE(1) |
5be78ee9
VP		 1539 TX_QUEUE(ep->com.dev->rdev.lldi.tx_modq[ep->tx_chan]) |
1540 RX_CHANNEL(0) |
1541 CCTRL_ECN(enable_ecn) |
ef5d6355 1542 RSS_QUEUE_VALID | RSS_QUEUE(ep->rss_qid));
5be78ee9 1543 if (enable_tcp_timestamps)
ef5d6355 1544 req->tcb.opt2 |= (__force __be32) TSTAMPS_EN(1);
5be78ee9 1545 if (enable_tcp_sack)
ef5d6355 1546 req->tcb.opt2 |= (__force __be32) SACK_EN(1);
5be78ee9 1547 if (wscale && enable_tcp_window_scaling)
ef5d6355
VP		 1548 req->tcb.opt2 |= (__force __be32) WND_SCALE_EN(1);
1549 req->tcb.opt0 = cpu_to_be64((__force u64) req->tcb.opt0);
1550 req->tcb.opt2 = cpu_to_be32((__force u32) req->tcb.opt2);
793dad94
VP		 1551 set_wr_txq(skb, CPL_PRIORITY_CONTROL, ep->ctrlq_idx);
1552 set_bit(ACT_OFLD_CONN, &ep->com.history);
5be78ee9
VP		 1553 c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
1554}
1555
cfdda9d7
SW		 1556/*
1557 * Return whether a failed active open has allocated a TID
1558 */
1559static inline int act_open_has_tid(int status)
1560{
1561 return status != CPL_ERR_TCAM_FULL && status != CPL_ERR_CONN_EXIST &&
1562 status != CPL_ERR_ARP_MISS;
1563}
1564
793dad94
VP		 1565#define ACT_OPEN_RETRY_COUNT 2
1566
1567static int c4iw_reconnect(struct c4iw_ep *ep)
1568{
1569 int err = 0;
1570 struct rtable *rt;
1571 struct port_info *pi;
1572 struct net_device *pdev;
1573 int step;
1574 struct neighbour *neigh;
24d44a39
SW		 1575 struct sockaddr_in *laddr = (struct sockaddr_in *)
1576 &ep->com.cm_id->local_addr;
1577 struct sockaddr_in *raddr = (struct sockaddr_in *)
1578 &ep->com.cm_id->remote_addr;
793dad94
VP		 1579
1580 PDBG("%s qp %p cm_id %p\n", __func__, ep->com.qp, ep->com.cm_id);
1581 init_timer(&ep->timer);
1582
1583 /*
1584 * Allocate an active TID to initiate a TCP connection.
1585 */
1586 ep->atid = cxgb4_alloc_atid(ep->com.dev->rdev.lldi.tids, ep);
1587 if (ep->atid == -1) {
1588 pr_err("%s - cannot alloc atid.\n", __func__);
1589 err = -ENOMEM;
1590 goto fail2;
1591 }
1592 insert_handle(ep->com.dev, &ep->com.dev->atid_idr, ep, ep->atid);
1593
1594 /* find a route */
1595 rt = find_route(ep->com.dev,
24d44a39
SW		 1596 laddr->sin_addr.s_addr, raddr->sin_addr.s_addr,
1597 laddr->sin_port, raddr->sin_port, 0);
793dad94
VP		 1598 if (!rt) {
1599 pr_err("%s - cannot find route.\n", __func__);
1600 err = -EHOSTUNREACH;
1601 goto fail3;
1602 }
1603 ep->dst = &rt->dst;
1604
24d44a39 1605 neigh = dst_neigh_lookup(ep->dst, &raddr->sin_addr.s_addr);
aaa0c23c
ZZ		 1606 if (!neigh) {
1607 pr_err("%s - cannot alloc neigh.\n", __func__);
1608 err = -ENOMEM;
1609 goto fail4;
1610 }
1611
793dad94
VP		 1612 /* get a l2t entry */
1613 if (neigh->dev->flags & IFF_LOOPBACK) {
1614 PDBG("%s LOOPBACK\n", __func__);
24d44a39 1615 pdev = ip_dev_find(&init_net, raddr->sin_addr.s_addr);
793dad94
VP		 1616 ep->l2t = cxgb4_l2t_get(ep->com.dev->rdev.lldi.l2t,
1617 neigh, pdev, 0);
1618 pi = (struct port_info *)netdev_priv(pdev);
1619 ep->mtu = pdev->mtu;
1620 ep->tx_chan = cxgb4_port_chan(pdev);
1621 ep->smac_idx = (cxgb4_port_viid(pdev) & 0x7F) << 1;
1622 dev_put(pdev);
1623 } else {
1624 ep->l2t = cxgb4_l2t_get(ep->com.dev->rdev.lldi.l2t,
1625 neigh, neigh->dev, 0);
1626 pi = (struct port_info *)netdev_priv(neigh->dev);
1627 ep->mtu = dst_mtu(ep->dst);
1628 ep->tx_chan = cxgb4_port_chan(neigh->dev);
1629 ep->smac_idx = (cxgb4_port_viid(neigh->dev) &
1630 0x7F) << 1;
1631 }
1632
1633 step = ep->com.dev->rdev.lldi.ntxq / ep->com.dev->rdev.lldi.nchan;
1634 ep->txq_idx = pi->port_id * step;
1635 ep->ctrlq_idx = pi->port_id;
1636 step = ep->com.dev->rdev.lldi.nrxq / ep->com.dev->rdev.lldi.nchan;
1637 ep->rss_qid = ep->com.dev->rdev.lldi.rxq_ids[pi->port_id * step];
1638
1639 if (!ep->l2t) {
1640 pr_err("%s - cannot alloc l2e.\n", __func__);
1641 err = -ENOMEM;
1642 goto fail4;
1643 }
1644
1645 PDBG("%s txq_idx %u tx_chan %u smac_idx %u rss_qid %u l2t_idx %u\n",
1646 __func__, ep->txq_idx, ep->tx_chan, ep->smac_idx, ep->rss_qid,
1647 ep->l2t->idx);
1648
1649 state_set(&ep->com, CONNECTING);
1650 ep->tos = 0;
1651
1652 /* send connect request to rnic */
1653 err = send_connect(ep);
1654 if (!err)
1655 goto out;
1656
1657 cxgb4_l2t_release(ep->l2t);
1658fail4:
1659 dst_release(ep->dst);
1660fail3:
1661 remove_handle(ep->com.dev, &ep->com.dev->atid_idr, ep->atid);
1662 cxgb4_free_atid(ep->com.dev->rdev.lldi.tids, ep->atid);
1663fail2:
1664 /*
1665 * remember to send notification to upper layer.
1666 * We are in here so the upper layer is not aware that this is
1667 * re-connect attempt and so, upper layer is still waiting for
1668 * response of 1st connect request.
1669 */
1670 connect_reply_upcall(ep, -ECONNRESET);
1671 c4iw_put_ep(&ep->com);
1672out:
1673 return err;
1674}
1675
cfdda9d7
SW		 1676static int act_open_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
1677{
1678 struct c4iw_ep *ep;
1679 struct cpl_act_open_rpl *rpl = cplhdr(skb);
1680 unsigned int atid = GET_TID_TID(GET_AOPEN_ATID(
1681 ntohl(rpl->atid_status)));
1682 struct tid_info *t = dev->rdev.lldi.tids;
1683 int status = GET_AOPEN_STATUS(ntohl(rpl->atid_status));
1684
1685 ep = lookup_atid(t, atid);
1686
1687 PDBG("%s ep %p atid %u status %u errno %d\n", __func__, ep, atid,
1688 status, status2errno(status));
1689
1690 if (status == CPL_ERR_RTX_NEG_ADVICE) {
1691 printk(KERN_WARNING MOD "Connection problems for atid %u\n",
1692 atid);
1693 return 0;
1694 }
1695
793dad94
VP		 1696 set_bit(ACT_OPEN_RPL, &ep->com.history);
1697
d716a2a0
VP		 1698 /*
1699 * Log interesting failures.
1700 */
1701 switch (status) {
1702 case CPL_ERR_CONN_RESET:
1703 case CPL_ERR_CONN_TIMEDOUT:
1704 break;
5be78ee9 1705 case CPL_ERR_TCAM_FULL:
3b174d94 1706 dev->rdev.stats.tcam_full++;
793dad94
VP		 1707 if (dev->rdev.lldi.enable_fw_ofld_conn) {
1708 mutex_lock(&dev->rdev.stats.lock);
793dad94
VP		 1709 mutex_unlock(&dev->rdev.stats.lock);
1710 send_fw_act_open_req(ep,
1711 GET_TID_TID(GET_AOPEN_ATID(
1712 ntohl(rpl->atid_status))));
1713 return 0;
1714 }
1715 break;
1716 case CPL_ERR_CONN_EXIST:
1717 if (ep->retry_count++ < ACT_OPEN_RETRY_COUNT) {
1718 set_bit(ACT_RETRY_INUSE, &ep->com.history);
1719 remove_handle(ep->com.dev, &ep->com.dev->atid_idr,
1720 atid);
1721 cxgb4_free_atid(t, atid);
1722 dst_release(ep->dst);
1723 cxgb4_l2t_release(ep->l2t);
1724 c4iw_reconnect(ep);
1725 return 0;
1726 }
5be78ee9 1727 break;
d716a2a0
VP		 1728 default:
1729 printk(KERN_INFO MOD "Active open failure - "
1730 "atid %u status %u errno %d %pI4:%u->%pI4:%u\n",
1731 atid, status, status2errno(status),
1732 &ep->com.local_addr.sin_addr.s_addr,
1733 ntohs(ep->com.local_addr.sin_port),
1734 &ep->com.remote_addr.sin_addr.s_addr,
1735 ntohs(ep->com.remote_addr.sin_port));
1736 break;
1737 }
1738
cfdda9d7
SW		 1739 connect_reply_upcall(ep, status2errno(status));
1740 state_set(&ep->com, DEAD);
1741
1742 if (status && act_open_has_tid(status))
1743 cxgb4_remove_tid(ep->com.dev->rdev.lldi.tids, 0, GET_TID(rpl));
1744
793dad94 1745 remove_handle(ep->com.dev, &ep->com.dev->atid_idr, atid);
cfdda9d7
SW		 1746 cxgb4_free_atid(t, atid);
1747 dst_release(ep->dst);
1748 cxgb4_l2t_release(ep->l2t);
1749 c4iw_put_ep(&ep->com);
1750
1751 return 0;
1752}
1753
1754static int pass_open_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
1755{
1756 struct cpl_pass_open_rpl *rpl = cplhdr(skb);
1757 struct tid_info *t = dev->rdev.lldi.tids;
1758 unsigned int stid = GET_TID(rpl);
1759 struct c4iw_listen_ep *ep = lookup_stid(t, stid);
1760
1761 if (!ep) {
1cab775c
VP		 1762 PDBG("%s stid %d lookup failure!\n", __func__, stid);
1763 goto out;
cfdda9d7
SW		 1764 }
1765 PDBG("%s ep %p status %d error %d\n", __func__, ep,
1766 rpl->status, status2errno(rpl->status));
d9594d99 1767 c4iw_wake_up(&ep->com.wr_wait, status2errno(rpl->status));
cfdda9d7 1768
1cab775c 1769out:
cfdda9d7
SW		 1770 return 0;
1771}
1772
1773static int listen_stop(struct c4iw_listen_ep *ep)
1774{
1775 struct sk_buff *skb;
1776 struct cpl_close_listsvr_req *req;
1777
1778 PDBG("%s ep %p\n", __func__, ep);
1779 skb = get_skb(NULL, sizeof(*req), GFP_KERNEL);
1780 if (!skb) {
1781 printk(KERN_ERR MOD "%s - failed to alloc skb\n", __func__);
1782 return -ENOMEM;
1783 }
1784 req = (struct cpl_close_listsvr_req *) skb_put(skb, sizeof(*req));
1785 INIT_TP_WR(req, 0);
1786 OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_CLOSE_LISTSRV_REQ,
1787 ep->stid));
1788 req->reply_ctrl = cpu_to_be16(
1789 QUEUENO(ep->com.dev->rdev.lldi.rxq_ids[0]));
1790 set_wr_txq(skb, CPL_PRIORITY_SETUP, 0);
1791 return c4iw_ofld_send(&ep->com.dev->rdev, skb);
1792}
1793
1794static int close_listsrv_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
1795{
1796 struct cpl_close_listsvr_rpl *rpl = cplhdr(skb);
1797 struct tid_info *t = dev->rdev.lldi.tids;
1798 unsigned int stid = GET_TID(rpl);
1799 struct c4iw_listen_ep *ep = lookup_stid(t, stid);
1800
1801 PDBG("%s ep %p\n", __func__, ep);
d9594d99 1802 c4iw_wake_up(&ep->com.wr_wait, status2errno(rpl->status));
cfdda9d7
SW		 1803 return 0;
1804}
1805
1806static void accept_cr(struct c4iw_ep *ep, __be32 peer_ip, struct sk_buff *skb,
1807 struct cpl_pass_accept_req *req)
1808{
1809 struct cpl_pass_accept_rpl *rpl;
1810 unsigned int mtu_idx;
1811 u64 opt0;
1812 u32 opt2;
1813 int wscale;
1814
1815 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1816 BUG_ON(skb_cloned(skb));
1817 skb_trim(skb, sizeof(*rpl));
1818 skb_get(skb);
1819 cxgb4_best_mtu(ep->com.dev->rdev.lldi.mtus, ep->mtu, &mtu_idx);
1820 wscale = compute_wscale(rcv_win);
5be78ee9
VP		 1821 opt0 = (nocong ? NO_CONG(1) : 0) |
1822 KEEP_ALIVE(1) |
ba6d3925 1823 DELACK(1) |
cfdda9d7
SW		 1824 WND_SCALE(wscale) |
1825 MSS_IDX(mtu_idx) |
1826 L2T_IDX(ep->l2t->idx) |
1827 TX_CHAN(ep->tx_chan) |
1828 SMAC_SEL(ep->smac_idx) |
5be78ee9 1829 DSCP(ep->tos >> 2) |
b48f3b9c 1830 ULP_MODE(ULP_MODE_TCPDDP) |
cfdda9d7
SW		 1831 RCV_BUFSIZ(rcv_win>>10);
1832 opt2 = RX_CHANNEL(0) |
1833 RSS_QUEUE_VALID | RSS_QUEUE(ep->rss_qid);
1834
1835 if (enable_tcp_timestamps && req->tcpopt.tstamp)
1836 opt2 |= TSTAMPS_EN(1);
1837 if (enable_tcp_sack && req->tcpopt.sack)
1838 opt2 |= SACK_EN(1);
1839 if (wscale && enable_tcp_window_scaling)
1840 opt2 |= WND_SCALE_EN(1);
5be78ee9
VP		 1841 if (enable_ecn) {
1842 const struct tcphdr *tcph;
1843 u32 hlen = ntohl(req->hdr_len);
1844
1845 tcph = (const void *)(req + 1) + G_ETH_HDR_LEN(hlen) +
1846 G_IP_HDR_LEN(hlen);
1847 if (tcph->ece && tcph->cwr)
1848 opt2 |= CCTRL_ECN(1);
1849 }
cfdda9d7
SW		 1850
1851 rpl = cplhdr(skb);
1852 INIT_TP_WR(rpl, ep->hwtid);
1853 OPCODE_TID(rpl) = cpu_to_be32(MK_OPCODE_TID(CPL_PASS_ACCEPT_RPL,
1854 ep->hwtid));
1855 rpl->opt0 = cpu_to_be64(opt0);
1856 rpl->opt2 = cpu_to_be32(opt2);
d4f1a5c6 1857 set_wr_txq(skb, CPL_PRIORITY_SETUP, ep->ctrlq_idx);
cfdda9d7
SW		 1858 c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
1859
1860 return;
1861}
1862
1863static void reject_cr(struct c4iw_dev *dev, u32 hwtid, __be32 peer_ip,
1864 struct sk_buff *skb)
1865{
1866 PDBG("%s c4iw_dev %p tid %u peer_ip %x\n", __func__, dev, hwtid,
1867 peer_ip);
1868 BUG_ON(skb_cloned(skb));
1869 skb_trim(skb, sizeof(struct cpl_tid_release));
1870 skb_get(skb);
1871 release_tid(&dev->rdev, hwtid, skb);
1872 return;
1873}
1874
1875static void get_4tuple(struct cpl_pass_accept_req *req,
1876 __be32 *local_ip, __be32 *peer_ip,
1877 __be16 *local_port, __be16 *peer_port)
1878{
1879 int eth_len = G_ETH_HDR_LEN(be32_to_cpu(req->hdr_len));
1880 int ip_len = G_IP_HDR_LEN(be32_to_cpu(req->hdr_len));
1881 struct iphdr *ip = (struct iphdr *)((u8 *)(req + 1) + eth_len);
1882 struct tcphdr *tcp = (struct tcphdr *)
1883 ((u8 *)(req + 1) + eth_len + ip_len);
1884
1885 PDBG("%s saddr 0x%x daddr 0x%x sport %u dport %u\n", __func__,
1886 ntohl(ip->saddr), ntohl(ip->daddr), ntohs(tcp->source),
1887 ntohs(tcp->dest));
1888
1889 *peer_ip = ip->saddr;
1890 *local_ip = ip->daddr;
1891 *peer_port = tcp->source;
1892 *local_port = tcp->dest;
1893
1894 return;
1895}
1896
3786cf18
DM		 1897static int import_ep(struct c4iw_ep *ep, __be32 peer_ip, struct dst_entry *dst,
1898 struct c4iw_dev *cdev, bool clear_mpa_v1)
1899{
1900 struct neighbour *n;
1901 int err, step;
1902
64b7007e 1903 n = dst_neigh_lookup(dst, &peer_ip);
3786cf18 1904 if (!n)
64b7007e
DM		 1905 return -ENODEV;
1906
1907 rcu_read_lock();
3786cf18
DM		 1908 err = -ENOMEM;
1909 if (n->dev->flags & IFF_LOOPBACK) {
1910 struct net_device *pdev;
1911
1912 pdev = ip_dev_find(&init_net, peer_ip);
71b43fd5
TLSC		 1913 if (!pdev) {
1914 err = -ENODEV;
1915 goto out;
1916 }
3786cf18
DM		 1917 ep->l2t = cxgb4_l2t_get(cdev->rdev.lldi.l2t,
1918 n, pdev, 0);
1919 if (!ep->l2t)
1920 goto out;
1921 ep->mtu = pdev->mtu;
1922 ep->tx_chan = cxgb4_port_chan(pdev);
1923 ep->smac_idx = (cxgb4_port_viid(pdev) & 0x7F) << 1;
1924 step = cdev->rdev.lldi.ntxq /
1925 cdev->rdev.lldi.nchan;
1926 ep->txq_idx = cxgb4_port_idx(pdev) * step;
1927 step = cdev->rdev.lldi.nrxq /
1928 cdev->rdev.lldi.nchan;
1929 ep->ctrlq_idx = cxgb4_port_idx(pdev);
1930 ep->rss_qid = cdev->rdev.lldi.rxq_ids[
1931 cxgb4_port_idx(pdev) * step];
1932 dev_put(pdev);
1933 } else {
1934 ep->l2t = cxgb4_l2t_get(cdev->rdev.lldi.l2t,
1935 n, n->dev, 0);
1936 if (!ep->l2t)
1937 goto out;
bd61baaf 1938 ep->mtu = dst_mtu(dst);
3786cf18
DM		 1939 ep->tx_chan = cxgb4_port_chan(n->dev);
1940 ep->smac_idx = (cxgb4_port_viid(n->dev) & 0x7F) << 1;
1941 step = cdev->rdev.lldi.ntxq /
1942 cdev->rdev.lldi.nchan;
1943 ep->txq_idx = cxgb4_port_idx(n->dev) * step;
1944 ep->ctrlq_idx = cxgb4_port_idx(n->dev);
1945 step = cdev->rdev.lldi.nrxq /
1946 cdev->rdev.lldi.nchan;
1947 ep->rss_qid = cdev->rdev.lldi.rxq_ids[
1948 cxgb4_port_idx(n->dev) * step];
1949
1950 if (clear_mpa_v1) {
1951 ep->retry_with_mpa_v1 = 0;
1952 ep->tried_with_mpa_v1 = 0;
1953 }
1954 }
1955 err = 0;
1956out:
1957 rcu_read_unlock();
1958
64b7007e
DM		 1959 neigh_release(n);
1960
3786cf18
DM		 1961 return err;
1962}
1963
cfdda9d7
SW		 1964static int pass_accept_req(struct c4iw_dev *dev, struct sk_buff *skb)
1965{
793dad94 1966 struct c4iw_ep *child_ep = NULL, *parent_ep;
cfdda9d7
SW		 1967 struct cpl_pass_accept_req *req = cplhdr(skb);
1968 unsigned int stid = GET_POPEN_TID(ntohl(req->tos_stid));
1969 struct tid_info *t = dev->rdev.lldi.tids;
1970 unsigned int hwtid = GET_TID(req);
1971 struct dst_entry *dst;
cfdda9d7 1972 struct rtable *rt;
1cab775c 1973 __be32 local_ip, peer_ip = 0;
cfdda9d7 1974 __be16 local_port, peer_port;
3786cf18 1975 int err;
1cab775c 1976 u16 peer_mss = ntohs(req->tcpopt.mss);
cfdda9d7
SW		 1977
1978 parent_ep = lookup_stid(t, stid);
1cab775c
VP		 1979 if (!parent_ep) {
1980 PDBG("%s connect request on invalid stid %d\n", __func__, stid);
1981 goto reject;
1982 }
cfdda9d7
SW		 1983 get_4tuple(req, &local_ip, &peer_ip, &local_port, &peer_port);
1984
1cab775c
VP		 1985 PDBG("%s parent ep %p hwtid %u laddr 0x%x raddr 0x%x lport %d " \
1986 "rport %d peer_mss %d\n", __func__, parent_ep, hwtid,
1987 ntohl(local_ip), ntohl(peer_ip), ntohs(local_port),
1988 ntohs(peer_port), peer_mss);
1989
cfdda9d7
SW		 1990 if (state_read(&parent_ep->com) != LISTEN) {
1991 printk(KERN_ERR "%s - listening ep not in LISTEN\n",
1992 __func__);
1993 goto reject;
1994 }
1995
1996 /* Find output route */
1997 rt = find_route(dev, local_ip, peer_ip, local_port, peer_port,
1998 GET_POPEN_TOS(ntohl(req->tos_stid)));
1999 if (!rt) {
2000 printk(KERN_ERR MOD "%s - failed to find dst entry!\n",
2001 __func__);
2002 goto reject;
2003 }
d8d1f30b 2004 dst = &rt->dst;
3786cf18
DM		 2005
2006 child_ep = alloc_ep(sizeof(*child_ep), GFP_KERNEL);
2007 if (!child_ep) {
2008 printk(KERN_ERR MOD "%s - failed to allocate ep entry!\n",
cfdda9d7
SW		 2009 __func__);
2010 dst_release(dst);
2011 goto reject;
2012 }
2013
3786cf18
DM		 2014 err = import_ep(child_ep, peer_ip, dst, dev, false);
2015 if (err) {
2016 printk(KERN_ERR MOD "%s - failed to allocate l2t entry!\n",
cfdda9d7 2017 __func__);
cfdda9d7 2018 dst_release(dst);
3786cf18 2019 kfree(child_ep);
cfdda9d7
SW		 2020 goto reject;
2021 }
3786cf18 2022
1cab775c
VP		 2023 if (peer_mss && child_ep->mtu > (peer_mss + 40))
2024 child_ep->mtu = peer_mss + 40;
2025
cfdda9d7
SW		 2026 state_set(&child_ep->com, CONNECTING);
2027 child_ep->com.dev = dev;
2028 child_ep->com.cm_id = NULL;
2029 child_ep->com.local_addr.sin_family = PF_INET;
2030 child_ep->com.local_addr.sin_port = local_port;
2031 child_ep->com.local_addr.sin_addr.s_addr = local_ip;
2032 child_ep->com.remote_addr.sin_family = PF_INET;
2033 child_ep->com.remote_addr.sin_port = peer_port;
2034 child_ep->com.remote_addr.sin_addr.s_addr = peer_ip;
2035 c4iw_get_ep(&parent_ep->com);
2036 child_ep->parent_ep = parent_ep;
2037 child_ep->tos = GET_POPEN_TOS(ntohl(req->tos_stid));
cfdda9d7
SW		 2038 child_ep->dst = dst;
2039 child_ep->hwtid = hwtid;
cfdda9d7
SW		 2040
2041 PDBG("%s tx_chan %u smac_idx %u rss_qid %u\n", __func__,
3786cf18 2042 child_ep->tx_chan, child_ep->smac_idx, child_ep->rss_qid);
cfdda9d7
SW		 2043
2044 init_timer(&child_ep->timer);
2045 cxgb4_insert_tid(t, child_ep, hwtid);
b3de6cfe 2046 insert_handle(dev, &dev->hwtid_idr, child_ep, child_ep->hwtid);
cfdda9d7 2047 accept_cr(child_ep, peer_ip, skb, req);
793dad94 2048 set_bit(PASS_ACCEPT_REQ, &child_ep->com.history);
cfdda9d7
SW		 2049 goto out;
2050reject:
2051 reject_cr(dev, hwtid, peer_ip, skb);
2052out:
2053 return 0;
2054}
2055
2056static int pass_establish(struct c4iw_dev *dev, struct sk_buff *skb)
2057{
2058 struct c4iw_ep *ep;
2059 struct cpl_pass_establish *req = cplhdr(skb);
2060 struct tid_info *t = dev->rdev.lldi.tids;
2061 unsigned int tid = GET_TID(req);
2062
2063 ep = lookup_tid(t, tid);
2064 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
2065 ep->snd_seq = be32_to_cpu(req->snd_isn);
2066 ep->rcv_seq = be32_to_cpu(req->rcv_isn);
2067
1cab775c
VP		 2068 PDBG("%s ep %p hwtid %u tcp_opt 0x%02x\n", __func__, ep, tid,
2069 ntohs(req->tcp_opt));
2070
cfdda9d7
SW		 2071 set_emss(ep, ntohs(req->tcp_opt));
2072
2073 dst_confirm(ep->dst);
2074 state_set(&ep->com, MPA_REQ_WAIT);
2075 start_ep_timer(ep);
2076 send_flowc(ep, skb);
793dad94 2077 set_bit(PASS_ESTAB, &ep->com.history);
cfdda9d7
SW		 2078
2079 return 0;
2080}
2081
2082static int peer_close(struct c4iw_dev *dev, struct sk_buff *skb)
2083{
2084 struct cpl_peer_close *hdr = cplhdr(skb);
2085 struct c4iw_ep *ep;
2086 struct c4iw_qp_attributes attrs;
cfdda9d7
SW		 2087 int disconnect = 1;
2088 int release = 0;
cfdda9d7
SW		 2089 struct tid_info *t = dev->rdev.lldi.tids;
2090 unsigned int tid = GET_TID(hdr);
8da7e7a5 2091 int ret;
cfdda9d7
SW		 2092
2093 ep = lookup_tid(t, tid);
2094 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
2095 dst_confirm(ep->dst);
2096
793dad94 2097 set_bit(PEER_CLOSE, &ep->com.history);
2f5b48c3 2098 mutex_lock(&ep->com.mutex);
cfdda9d7
SW		 2099 switch (ep->com.state) {
2100 case MPA_REQ_WAIT:
2101 __state_set(&ep->com, CLOSING);
2102 break;
2103 case MPA_REQ_SENT:
2104 __state_set(&ep->com, CLOSING);
2105 connect_reply_upcall(ep, -ECONNRESET);
2106 break;
2107 case MPA_REQ_RCVD:
2108
2109 /*
2110 * We're gonna mark this puppy DEAD, but keep
2111 * the reference on it until the ULP accepts or
2112 * rejects the CR. Also wake up anyone waiting
2113 * in rdma connection migration (see c4iw_accept_cr()).
2114 */
2115 __state_set(&ep->com, CLOSING);
cfdda9d7 2116 PDBG("waking up ep %p tid %u\n", ep, ep->hwtid);
d9594d99 2117 c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
cfdda9d7
SW		 2118 break;
2119 case MPA_REP_SENT:
2120 __state_set(&ep->com, CLOSING);
cfdda9d7 2121 PDBG("waking up ep %p tid %u\n", ep, ep->hwtid);
d9594d99 2122 c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
cfdda9d7
SW		 2123 break;
2124 case FPDU_MODE:
ca5a2202 2125 start_ep_timer(ep);
cfdda9d7 2126 __state_set(&ep->com, CLOSING);
30c95c2d 2127 attrs.next_state = C4IW_QP_STATE_CLOSING;
8da7e7a5 2128 ret = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
30c95c2d 2129 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
8da7e7a5
SW		 2130 if (ret != -ECONNRESET) {
2131 peer_close_upcall(ep);
2132 disconnect = 1;
2133 }
cfdda9d7
SW		 2134 break;
2135 case ABORTING:
2136 disconnect = 0;
2137 break;
2138 case CLOSING:
2139 __state_set(&ep->com, MORIBUND);
2140 disconnect = 0;
2141 break;
2142 case MORIBUND:
ca5a2202 2143 stop_ep_timer(ep);
cfdda9d7
SW		 2144 if (ep->com.cm_id && ep->com.qp) {
2145 attrs.next_state = C4IW_QP_STATE_IDLE;
2146 c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
2147 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
2148 }
2149 close_complete_upcall(ep);
2150 __state_set(&ep->com, DEAD);
2151 release = 1;
2152 disconnect = 0;
2153 break;
2154 case DEAD:
2155 disconnect = 0;
2156 break;
2157 default:
2158 BUG_ON(1);
2159 }
2f5b48c3 2160 mutex_unlock(&ep->com.mutex);
cfdda9d7
SW		 2161 if (disconnect)
2162 c4iw_ep_disconnect(ep, 0, GFP_KERNEL);
2163 if (release)
2164 release_ep_resources(ep);
2165 return 0;
2166}
2167
2168/*
2169 * Returns whether an ABORT_REQ_RSS message is a negative advice.
2170 */
2171static int is_neg_adv_abort(unsigned int status)
2172{
2173 return status == CPL_ERR_RTX_NEG_ADVICE ||
2174 status == CPL_ERR_PERSIST_NEG_ADVICE;
2175}
2176
2177static int peer_abort(struct c4iw_dev *dev, struct sk_buff *skb)
2178{
2179 struct cpl_abort_req_rss *req = cplhdr(skb);
2180 struct c4iw_ep *ep;
2181 struct cpl_abort_rpl *rpl;
2182 struct sk_buff *rpl_skb;
2183 struct c4iw_qp_attributes attrs;
2184 int ret;
2185 int release = 0;
cfdda9d7
SW		 2186 struct tid_info *t = dev->rdev.lldi.tids;
2187 unsigned int tid = GET_TID(req);
cfdda9d7
SW		 2188
2189 ep = lookup_tid(t, tid);
2190 if (is_neg_adv_abort(req->status)) {
2191 PDBG("%s neg_adv_abort ep %p tid %u\n", __func__, ep,
2192 ep->hwtid);
2193 return 0;
2194 }
cfdda9d7
SW		 2195 PDBG("%s ep %p tid %u state %u\n", __func__, ep, ep->hwtid,
2196 ep->com.state);
793dad94 2197 set_bit(PEER_ABORT, &ep->com.history);
2f5b48c3
SW		 2198
2199 /*
2200 * Wake up any threads in rdma_init() or rdma_fini().
d2fe99e8
KS		 2201 * However, this is not needed if com state is just
2202 * MPA_REQ_SENT
2f5b48c3 2203 */
d2fe99e8
KS		 2204 if (ep->com.state != MPA_REQ_SENT)
2205 c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
2f5b48c3
SW		 2206
2207 mutex_lock(&ep->com.mutex);
cfdda9d7
SW		 2208 switch (ep->com.state) {
2209 case CONNECTING:
2210 break;
2211 case MPA_REQ_WAIT:
ca5a2202 2212 stop_ep_timer(ep);
cfdda9d7
SW		 2213 break;
2214 case MPA_REQ_SENT:
ca5a2202 2215 stop_ep_timer(ep);
fe7e0a4d 2216 if (mpa_rev == 1 || (mpa_rev == 2 && ep->tried_with_mpa_v1))
d2fe99e8
KS		 2217 connect_reply_upcall(ep, -ECONNRESET);
2218 else {
2219 /*
2220 * we just don't send notification upwards because we
2221 * want to retry with mpa_v1 without upper layers even
2222 * knowing it.
2223 *
2224 * do some housekeeping so as to re-initiate the
2225 * connection
2226 */
2227 PDBG("%s: mpa_rev=%d. Retrying with mpav1\n", __func__,
2228 mpa_rev);
2229 ep->retry_with_mpa_v1 = 1;
2230 }
cfdda9d7
SW		 2231 break;
2232 case MPA_REP_SENT:
cfdda9d7
SW		 2233 break;
2234 case MPA_REQ_RCVD:
cfdda9d7
SW		 2235 break;
2236 case MORIBUND:
2237 case CLOSING:
ca5a2202 2238 stop_ep_timer(ep);
cfdda9d7
SW		 2239 /*FALLTHROUGH*/
2240 case FPDU_MODE:
2241 if (ep->com.cm_id && ep->com.qp) {
2242 attrs.next_state = C4IW_QP_STATE_ERROR;
2243 ret = c4iw_modify_qp(ep->com.qp->rhp,
2244 ep->com.qp, C4IW_QP_ATTR_NEXT_STATE,
2245 &attrs, 1);
2246 if (ret)
2247 printk(KERN_ERR MOD
2248 "%s - qp <- error failed!\n",
2249 __func__);
2250 }
2251 peer_abort_upcall(ep);
2252 break;
2253 case ABORTING:
2254 break;
2255 case DEAD:
2256 PDBG("%s PEER_ABORT IN DEAD STATE!!!!\n", __func__);
2f5b48c3 2257 mutex_unlock(&ep->com.mutex);
cfdda9d7
SW		 2258 return 0;
2259 default:
2260 BUG_ON(1);
2261 break;
2262 }
2263 dst_confirm(ep->dst);
2264 if (ep->com.state != ABORTING) {
2265 __state_set(&ep->com, DEAD);
d2fe99e8
KS		 2266 /* we don't release if we want to retry with mpa_v1 */
2267 if (!ep->retry_with_mpa_v1)
2268 release = 1;
cfdda9d7 2269 }
2f5b48c3 2270 mutex_unlock(&ep->com.mutex);
cfdda9d7
SW		 2271
2272 rpl_skb = get_skb(skb, sizeof(*rpl), GFP_KERNEL);
2273 if (!rpl_skb) {
2274 printk(KERN_ERR MOD "%s - cannot allocate skb!\n",
2275 __func__);
2276 release = 1;
2277 goto out;
2278 }
2279 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
2280 rpl = (struct cpl_abort_rpl *) skb_put(rpl_skb, sizeof(*rpl));
2281 INIT_TP_WR(rpl, ep->hwtid);
2282 OPCODE_TID(rpl) = cpu_to_be32(MK_OPCODE_TID(CPL_ABORT_RPL, ep->hwtid));
2283 rpl->cmd = CPL_ABORT_NO_RST;
2284 c4iw_ofld_send(&ep->com.dev->rdev, rpl_skb);
2285out:
cfdda9d7
SW		 2286 if (release)
2287 release_ep_resources(ep);
fe7e0a4d
VP		 2288 else if (ep->retry_with_mpa_v1) {
2289 remove_handle(ep->com.dev, &ep->com.dev->hwtid_idr, ep->hwtid);
d2fe99e8
KS		 2290 cxgb4_remove_tid(ep->com.dev->rdev.lldi.tids, 0, ep->hwtid);
2291 dst_release(ep->dst);
2292 cxgb4_l2t_release(ep->l2t);
2293 c4iw_reconnect(ep);
2294 }
2295
cfdda9d7
SW		 2296 return 0;
2297}
2298
2299static int close_con_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
2300{
2301 struct c4iw_ep *ep;
2302 struct c4iw_qp_attributes attrs;
2303 struct cpl_close_con_rpl *rpl = cplhdr(skb);
cfdda9d7
SW		 2304 int release = 0;
2305 struct tid_info *t = dev->rdev.lldi.tids;
2306 unsigned int tid = GET_TID(rpl);
cfdda9d7
SW		 2307
2308 ep = lookup_tid(t, tid);
2309
2310 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
2311 BUG_ON(!ep);
2312
2313 /* The cm_id may be null if we failed to connect */
2f5b48c3 2314 mutex_lock(&ep->com.mutex);
cfdda9d7
SW		 2315 switch (ep->com.state) {
2316 case CLOSING:
2317 __state_set(&ep->com, MORIBUND);
2318 break;
2319 case MORIBUND:
ca5a2202 2320 stop_ep_timer(ep);
cfdda9d7
SW		 2321 if ((ep->com.cm_id) && (ep->com.qp)) {
2322 attrs.next_state = C4IW_QP_STATE_IDLE;
2323 c4iw_modify_qp(ep->com.qp->rhp,
2324 ep->com.qp,
2325 C4IW_QP_ATTR_NEXT_STATE,
2326 &attrs, 1);
2327 }
2328 close_complete_upcall(ep);
2329 __state_set(&ep->com, DEAD);
2330 release = 1;
2331 break;
2332 case ABORTING:
2333 case DEAD:
2334 break;
2335 default:
2336 BUG_ON(1);
2337 break;
2338 }
2f5b48c3 2339 mutex_unlock(&ep->com.mutex);
cfdda9d7
SW		 2340 if (release)
2341 release_ep_resources(ep);
2342 return 0;
2343}
2344
2345static int terminate(struct c4iw_dev *dev, struct sk_buff *skb)
2346{
0e42c1f4 2347 struct cpl_rdma_terminate *rpl = cplhdr(skb);
cfdda9d7 2348 struct tid_info *t = dev->rdev.lldi.tids;
0e42c1f4
SW		 2349 unsigned int tid = GET_TID(rpl);
2350 struct c4iw_ep *ep;
2351 struct c4iw_qp_attributes attrs;
cfdda9d7
SW		 2352
2353 ep = lookup_tid(t, tid);
0e42c1f4 2354 BUG_ON(!ep);
cfdda9d7 2355
30c95c2d 2356 if (ep && ep->com.qp) {
0e42c1f4
SW		 2357 printk(KERN_WARNING MOD "TERM received tid %u qpid %u\n", tid,
2358 ep->com.qp->wq.sq.qid);
2359 attrs.next_state = C4IW_QP_STATE_TERMINATE;
2360 c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
2361 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
2362 } else
30c95c2d 2363 printk(KERN_WARNING MOD "TERM received tid %u no ep/qp\n", tid);
cfdda9d7 2364
cfdda9d7
SW		 2365 return 0;
2366}
2367
2368/*
2369 * Upcall from the adapter indicating data has been transmitted.
2370 * For us its just the single MPA request or reply. We can now free
2371 * the skb holding the mpa message.
2372 */
2373static int fw4_ack(struct c4iw_dev *dev, struct sk_buff *skb)
2374{
2375 struct c4iw_ep *ep;
2376 struct cpl_fw4_ack *hdr = cplhdr(skb);
2377 u8 credits = hdr->credits;
2378 unsigned int tid = GET_TID(hdr);
2379 struct tid_info *t = dev->rdev.lldi.tids;
2380
2381
2382 ep = lookup_tid(t, tid);
2383 PDBG("%s ep %p tid %u credits %u\n", __func__, ep, ep->hwtid, credits);
2384 if (credits == 0) {
aa1ad260
JP		 2385 PDBG("%s 0 credit ack ep %p tid %u state %u\n",
2386 __func__, ep, ep->hwtid, state_read(&ep->com));
cfdda9d7
SW		 2387 return 0;
2388 }
2389
2390 dst_confirm(ep->dst);
2391 if (ep->mpa_skb) {
2392 PDBG("%s last streaming msg ack ep %p tid %u state %u "
2393 "initiator %u freeing skb\n", __func__, ep, ep->hwtid,
2394 state_read(&ep->com), ep->mpa_attr.initiator ? 1 : 0);
2395 kfree_skb(ep->mpa_skb);
2396 ep->mpa_skb = NULL;
2397 }
2398 return 0;
2399}
2400
cfdda9d7
SW		 2401int c4iw_reject_cr(struct iw_cm_id *cm_id, const void *pdata, u8 pdata_len)
2402{
2403 int err;
2404 struct c4iw_ep *ep = to_ep(cm_id);
2405 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
2406
2407 if (state_read(&ep->com) == DEAD) {
2408 c4iw_put_ep(&ep->com);
2409 return -ECONNRESET;
2410 }
793dad94 2411 set_bit(ULP_REJECT, &ep->com.history);
cfdda9d7
SW		 2412 BUG_ON(state_read(&ep->com) != MPA_REQ_RCVD);
2413 if (mpa_rev == 0)
2414 abort_connection(ep, NULL, GFP_KERNEL);
2415 else {
2416 err = send_mpa_reject(ep, pdata, pdata_len);
2417 err = c4iw_ep_disconnect(ep, 0, GFP_KERNEL);
2418 }
2419 c4iw_put_ep(&ep->com);
2420 return 0;
2421}
2422
2423int c4iw_accept_cr(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
2424{
2425 int err;
2426 struct c4iw_qp_attributes attrs;
2427 enum c4iw_qp_attr_mask mask;
2428 struct c4iw_ep *ep = to_ep(cm_id);
2429 struct c4iw_dev *h = to_c4iw_dev(cm_id->device);
2430 struct c4iw_qp *qp = get_qhp(h, conn_param->qpn);
2431
2432 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
2433 if (state_read(&ep->com) == DEAD) {
2434 err = -ECONNRESET;
2435 goto err;
2436 }
2437
2438 BUG_ON(state_read(&ep->com) != MPA_REQ_RCVD);
2439 BUG_ON(!qp);
2440
793dad94 2441 set_bit(ULP_ACCEPT, &ep->com.history);
be4c9bad
RD		 2442 if ((conn_param->ord > c4iw_max_read_depth) ||
2443 (conn_param->ird > c4iw_max_read_depth)) {
cfdda9d7
SW		 2444 abort_connection(ep, NULL, GFP_KERNEL);
2445 err = -EINVAL;
2446 goto err;
2447 }
2448
d2fe99e8
KS		 2449 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) {
2450 if (conn_param->ord > ep->ird) {
2451 ep->ird = conn_param->ird;
2452 ep->ord = conn_param->ord;
2453 send_mpa_reject(ep, conn_param->private_data,
2454 conn_param->private_data_len);
2455 abort_connection(ep, NULL, GFP_KERNEL);
2456 err = -ENOMEM;
2457 goto err;
2458 }
2459 if (conn_param->ird > ep->ord) {
2460 if (!ep->ord)
2461 conn_param->ird = 1;
2462 else {
2463 abort_connection(ep, NULL, GFP_KERNEL);
2464 err = -ENOMEM;
2465 goto err;
2466 }
2467 }
cfdda9d7 2468
d2fe99e8 2469 }
cfdda9d7
SW		 2470 ep->ird = conn_param->ird;
2471 ep->ord = conn_param->ord;
2472
d2fe99e8
KS		 2473 if (ep->mpa_attr.version != 2)
2474 if (peer2peer && ep->ird == 0)
2475 ep->ird = 1;
cfdda9d7
SW		 2476
2477 PDBG("%s %d ird %d ord %d\n", __func__, __LINE__, ep->ird, ep->ord);
2478
d2fe99e8
KS		 2479 cm_id->add_ref(cm_id);
2480 ep->com.cm_id = cm_id;
2481 ep->com.qp = qp;
325abead 2482 ref_qp(ep);
d2fe99e8 2483
cfdda9d7
SW		 2484 /* bind QP to EP and move to RTS */
2485 attrs.mpa_attr = ep->mpa_attr;
2486 attrs.max_ird = ep->ird;
2487 attrs.max_ord = ep->ord;
2488 attrs.llp_stream_handle = ep;
2489 attrs.next_state = C4IW_QP_STATE_RTS;
2490
2491 /* bind QP and TID with INIT_WR */
2492 mask = C4IW_QP_ATTR_NEXT_STATE |
2493 C4IW_QP_ATTR_LLP_STREAM_HANDLE |
2494 C4IW_QP_ATTR_MPA_ATTR |
2495 C4IW_QP_ATTR_MAX_IRD |
2496 C4IW_QP_ATTR_MAX_ORD;
2497
2498 err = c4iw_modify_qp(ep->com.qp->rhp,
2499 ep->com.qp, mask, &attrs, 1);
2500 if (err)
2501 goto err1;
2502 err = send_mpa_reply(ep, conn_param->private_data,
2503 conn_param->private_data_len);
2504 if (err)
2505 goto err1;
2506
2507 state_set(&ep->com, FPDU_MODE);
2508 established_upcall(ep);
2509 c4iw_put_ep(&ep->com);
2510 return 0;
2511err1:
2512 ep->com.cm_id = NULL;
cfdda9d7
SW		 2513 cm_id->rem_ref(cm_id);
2514err:
2515 c4iw_put_ep(&ep->com);
2516 return err;
2517}
2518
2519int c4iw_connect(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
2520{
cfdda9d7
SW		 2521 struct c4iw_dev *dev = to_c4iw_dev(cm_id->device);
2522 struct c4iw_ep *ep;
2523 struct rtable *rt;
3786cf18 2524 int err = 0;
24d44a39
SW		 2525 struct sockaddr_in *laddr = (struct sockaddr_in *)&cm_id->local_addr;
2526 struct sockaddr_in *raddr = (struct sockaddr_in *)&cm_id->remote_addr;
cfdda9d7 2527
be4c9bad
RD		 2528 if ((conn_param->ord > c4iw_max_read_depth) ||
2529 (conn_param->ird > c4iw_max_read_depth)) {
2530 err = -EINVAL;
2531 goto out;
2532 }
cfdda9d7
SW		 2533 ep = alloc_ep(sizeof(*ep), GFP_KERNEL);
2534 if (!ep) {
2535 printk(KERN_ERR MOD "%s - cannot alloc ep.\n", __func__);
2536 err = -ENOMEM;
2537 goto out;
2538 }
2539 init_timer(&ep->timer);
2540 ep->plen = conn_param->private_data_len;
2541 if (ep->plen)
2542 memcpy(ep->mpa_pkt + sizeof(struct mpa_message),
2543 conn_param->private_data, ep->plen);
2544 ep->ird = conn_param->ird;
2545 ep->ord = conn_param->ord;
2546
2547 if (peer2peer && ep->ord == 0)
2548 ep->ord = 1;
2549
2550 cm_id->add_ref(cm_id);
2551 ep->com.dev = dev;
2552 ep->com.cm_id = cm_id;
2553 ep->com.qp = get_qhp(dev, conn_param->qpn);
2554 BUG_ON(!ep->com.qp);
325abead 2555 ref_qp(ep);
cfdda9d7
SW		 2556 PDBG("%s qpn 0x%x qp %p cm_id %p\n", __func__, conn_param->qpn,
2557 ep->com.qp, cm_id);
2558
2559 /*
2560 * Allocate an active TID to initiate a TCP connection.
2561 */
2562 ep->atid = cxgb4_alloc_atid(dev->rdev.lldi.tids, ep);
2563 if (ep->atid == -1) {
2564 printk(KERN_ERR MOD "%s - cannot alloc atid.\n", __func__);
2565 err = -ENOMEM;
2566 goto fail2;
2567 }
793dad94 2568 insert_handle(dev, &dev->atid_idr, ep, ep->atid);
cfdda9d7
SW		 2569
2570 PDBG("%s saddr 0x%x sport 0x%x raddr 0x%x rport 0x%x\n", __func__,
24d44a39
SW		 2571 ntohl(laddr->sin_addr.s_addr), ntohs(laddr->sin_port),
2572 ntohl(raddr->sin_addr.s_addr), ntohs(raddr->sin_port));
cfdda9d7
SW		 2573
2574 /* find a route */
24d44a39
SW		 2575 rt = find_route(dev, laddr->sin_addr.s_addr, raddr->sin_addr.s_addr,
2576 laddr->sin_port, raddr->sin_port, 0);
cfdda9d7
SW		 2577 if (!rt) {
2578 printk(KERN_ERR MOD "%s - cannot find route.\n", __func__);
2579 err = -EHOSTUNREACH;
2580 goto fail3;
2581 }
d8d1f30b 2582 ep->dst = &rt->dst;
cfdda9d7 2583
24d44a39 2584 err = import_ep(ep, raddr->sin_addr.s_addr, ep->dst, ep->com.dev, true);
3786cf18 2585 if (err) {
cfdda9d7 2586 printk(KERN_ERR MOD "%s - cannot alloc l2e.\n", __func__);
cfdda9d7
SW		 2587 goto fail4;
2588 }
2589
2590 PDBG("%s txq_idx %u tx_chan %u smac_idx %u rss_qid %u l2t_idx %u\n",
2591 __func__, ep->txq_idx, ep->tx_chan, ep->smac_idx, ep->rss_qid,
2592 ep->l2t->idx);
2593
2594 state_set(&ep->com, CONNECTING);
2595 ep->tos = 0;
24d44a39
SW		 2596 memcpy(&ep->com.local_addr, &cm_id->local_addr,
2597 sizeof(ep->com.local_addr));
2598 memcpy(&ep->com.remote_addr, &cm_id->remote_addr,
2599 sizeof(ep->com.remote_addr));
cfdda9d7
SW		 2600
2601 /* send connect request to rnic */
2602 err = send_connect(ep);
2603 if (!err)
2604 goto out;
2605
2606 cxgb4_l2t_release(ep->l2t);
2607fail4:
2608 dst_release(ep->dst);
2609fail3:
793dad94 2610 remove_handle(ep->com.dev, &ep->com.dev->atid_idr, ep->atid);
cfdda9d7
SW		 2611 cxgb4_free_atid(ep->com.dev->rdev.lldi.tids, ep->atid);
2612fail2:
2613 cm_id->rem_ref(cm_id);
2614 c4iw_put_ep(&ep->com);
2615out:
2616 return err;
2617}
2618
2619int c4iw_create_listen(struct iw_cm_id *cm_id, int backlog)
2620{
2621 int err = 0;
2622 struct c4iw_dev *dev = to_c4iw_dev(cm_id->device);
2623 struct c4iw_listen_ep *ep;
2624
cfdda9d7
SW		 2625 might_sleep();
2626
2627 ep = alloc_ep(sizeof(*ep), GFP_KERNEL);
2628 if (!ep) {
2629 printk(KERN_ERR MOD "%s - cannot alloc ep.\n", __func__);
2630 err = -ENOMEM;
2631 goto fail1;
2632 }
2633 PDBG("%s ep %p\n", __func__, ep);
2634 cm_id->add_ref(cm_id);
2635 ep->com.cm_id = cm_id;
2636 ep->com.dev = dev;
2637 ep->backlog = backlog;
24d44a39
SW		 2638 memcpy(&ep->com.local_addr, &cm_id->local_addr,
2639 sizeof(ep->com.local_addr));
cfdda9d7
SW		 2640
2641 /*
2642 * Allocate a server TID.
2643 */
1cab775c
VP		 2644 if (dev->rdev.lldi.enable_fw_ofld_conn)
2645 ep->stid = cxgb4_alloc_sftid(dev->rdev.lldi.tids, PF_INET, ep);
2646 else
2647 ep->stid = cxgb4_alloc_stid(dev->rdev.lldi.tids, PF_INET, ep);
2648
cfdda9d7 2649 if (ep->stid == -1) {
be4c9bad 2650 printk(KERN_ERR MOD "%s - cannot alloc stid.\n", __func__);
cfdda9d7
SW		 2651 err = -ENOMEM;
2652 goto fail2;
2653 }
793dad94 2654 insert_handle(dev, &dev->stid_idr, ep, ep->stid);
cfdda9d7 2655 state_set(&ep->com, LISTEN);
1cab775c
VP		 2656 if (dev->rdev.lldi.enable_fw_ofld_conn) {
2657 do {
2658 err = cxgb4_create_server_filter(
2659 ep->com.dev->rdev.lldi.ports[0], ep->stid,
2660 ep->com.local_addr.sin_addr.s_addr,
2661 ep->com.local_addr.sin_port,
793dad94
VP		 2662 0,
2663 ep->com.dev->rdev.lldi.rxq_ids[0],
2664 0,
2665 0);
1cab775c
VP		 2666 if (err == -EBUSY) {
2667 set_current_state(TASK_UNINTERRUPTIBLE);
2668 schedule_timeout(usecs_to_jiffies(100));
2669 }
2670 } while (err == -EBUSY);
2671 } else {
2672 c4iw_init_wr_wait(&ep->com.wr_wait);
2673 err = cxgb4_create_server(ep->com.dev->rdev.lldi.ports[0],
2674 ep->stid, ep->com.local_addr.sin_addr.s_addr,
2675 ep->com.local_addr.sin_port,
793dad94 2676 0,
1cab775c
VP		 2677 ep->com.dev->rdev.lldi.rxq_ids[0]);
2678 if (!err)
2679 err = c4iw_wait_for_reply(&ep->com.dev->rdev,
2680 &ep->com.wr_wait,
2681 0, 0, __func__);
2682 }
cfdda9d7
SW		 2683 if (!err) {
2684 cm_id->provider_data = ep;
2685 goto out;
2686 }
1cab775c
VP		 2687 pr_err("%s cxgb4_create_server/filter failed err %d " \
2688 "stid %d laddr %08x lport %d\n", \
2689 __func__, err, ep->stid,
2690 ntohl(ep->com.local_addr.sin_addr.s_addr),
2691 ntohs(ep->com.local_addr.sin_port));
cfdda9d7
SW		 2692 cxgb4_free_stid(ep->com.dev->rdev.lldi.tids, ep->stid, PF_INET);
2693fail2:
2694 cm_id->rem_ref(cm_id);
2695 c4iw_put_ep(&ep->com);
2696fail1:
2697out:
2698 return err;
2699}
2700
2701int c4iw_destroy_listen(struct iw_cm_id *cm_id)
2702{
2703 int err;
2704 struct c4iw_listen_ep *ep = to_listen_ep(cm_id);
2705
2706 PDBG("%s ep %p\n", __func__, ep);
2707
2708 might_sleep();
2709 state_set(&ep->com, DEAD);
1cab775c
VP		 2710 if (ep->com.dev->rdev.lldi.enable_fw_ofld_conn) {
2711 err = cxgb4_remove_server_filter(
2712 ep->com.dev->rdev.lldi.ports[0], ep->stid,
2713 ep->com.dev->rdev.lldi.rxq_ids[0], 0);
2714 } else {
2715 c4iw_init_wr_wait(&ep->com.wr_wait);
2716 err = listen_stop(ep);
2717 if (err)
2718 goto done;
2719 err = c4iw_wait_for_reply(&ep->com.dev->rdev, &ep->com.wr_wait,
2720 0, 0, __func__);
2721 }
793dad94 2722 remove_handle(ep->com.dev, &ep->com.dev->stid_idr, ep->stid);
cfdda9d7
SW		 2723 cxgb4_free_stid(ep->com.dev->rdev.lldi.tids, ep->stid, PF_INET);
2724done:
cfdda9d7
SW		 2725 cm_id->rem_ref(cm_id);
2726 c4iw_put_ep(&ep->com);
2727 return err;
2728}
2729
2730int c4iw_ep_disconnect(struct c4iw_ep *ep, int abrupt, gfp_t gfp)
2731{
2732 int ret = 0;
cfdda9d7
SW		 2733 int close = 0;
2734 int fatal = 0;
2735 struct c4iw_rdev *rdev;
cfdda9d7 2736
2f5b48c3 2737 mutex_lock(&ep->com.mutex);
cfdda9d7
SW		 2738
2739 PDBG("%s ep %p state %s, abrupt %d\n", __func__, ep,
2740 states[ep->com.state], abrupt);
2741
2742 rdev = &ep->com.dev->rdev;
2743 if (c4iw_fatal_error(rdev)) {
2744 fatal = 1;
2745 close_complete_upcall(ep);
2746 ep->com.state = DEAD;
2747 }
2748 switch (ep->com.state) {
2749 case MPA_REQ_WAIT:
2750 case MPA_REQ_SENT:
2751 case MPA_REQ_RCVD:
2752 case MPA_REP_SENT:
2753 case FPDU_MODE:
2754 close = 1;
2755 if (abrupt)
2756 ep->com.state = ABORTING;
2757 else {
2758 ep->com.state = CLOSING;
ca5a2202 2759 start_ep_timer(ep);
cfdda9d7
SW		 2760 }
2761 set_bit(CLOSE_SENT, &ep->com.flags);
2762 break;
2763 case CLOSING:
2764 if (!test_and_set_bit(CLOSE_SENT, &ep->com.flags)) {
2765 close = 1;
2766 if (abrupt) {
ca5a2202 2767 stop_ep_timer(ep);
cfdda9d7
SW		 2768 ep->com.state = ABORTING;
2769 } else
2770 ep->com.state = MORIBUND;
2771 }
2772 break;
2773 case MORIBUND:
2774 case ABORTING:
2775 case DEAD:
2776 PDBG("%s ignoring disconnect ep %p state %u\n",
2777 __func__, ep, ep->com.state);
2778 break;
2779 default:
2780 BUG();
2781 break;
2782 }
2783
cfdda9d7 2784 if (close) {
8da7e7a5 2785 if (abrupt) {
793dad94 2786 set_bit(EP_DISC_ABORT, &ep->com.history);
8da7e7a5
SW		 2787 close_complete_upcall(ep);
2788 ret = send_abort(ep, NULL, gfp);
793dad94
VP		 2789 } else {
2790 set_bit(EP_DISC_CLOSE, &ep->com.history);
cfdda9d7 2791 ret = send_halfclose(ep, gfp);
793dad94 2792 }
cfdda9d7
SW		 2793 if (ret)
2794 fatal = 1;
2795 }
8da7e7a5 2796 mutex_unlock(&ep->com.mutex);
cfdda9d7
SW		 2797 if (fatal)
2798 release_ep_resources(ep);
2799 return ret;
2800}
2801
1cab775c
VP		 2802static void active_ofld_conn_reply(struct c4iw_dev *dev, struct sk_buff *skb,
2803 struct cpl_fw6_msg_ofld_connection_wr_rpl *req)
2804{
2805 struct c4iw_ep *ep;
793dad94 2806 int atid = be32_to_cpu(req->tid);
1cab775c 2807
ef5d6355
VP		 2808 ep = (struct c4iw_ep *)lookup_atid(dev->rdev.lldi.tids,
2809 (__force u32) req->tid);
1cab775c
VP		 2810 if (!ep)
2811 return;
2812
2813 switch (req->retval) {
2814 case FW_ENOMEM:
793dad94
VP		 2815 set_bit(ACT_RETRY_NOMEM, &ep->com.history);
2816 if (ep->retry_count++ < ACT_OPEN_RETRY_COUNT) {
2817 send_fw_act_open_req(ep, atid);
2818 return;
2819 }
1cab775c 2820 case FW_EADDRINUSE:
793dad94
VP		 2821 set_bit(ACT_RETRY_INUSE, &ep->com.history);
2822 if (ep->retry_count++ < ACT_OPEN_RETRY_COUNT) {
2823 send_fw_act_open_req(ep, atid);
2824 return;
2825 }
1cab775c
VP		 2826 break;
2827 default:
2828 pr_info("%s unexpected ofld conn wr retval %d\n",
2829 __func__, req->retval);
2830 break;
2831 }
793dad94
VP		 2832 pr_err("active ofld_connect_wr failure %d atid %d\n",
2833 req->retval, atid);
2834 mutex_lock(&dev->rdev.stats.lock);
2835 dev->rdev.stats.act_ofld_conn_fails++;
2836 mutex_unlock(&dev->rdev.stats.lock);
1cab775c 2837 connect_reply_upcall(ep, status2errno(req->retval));
793dad94
VP		 2838 state_set(&ep->com, DEAD);
2839 remove_handle(dev, &dev->atid_idr, atid);
2840 cxgb4_free_atid(dev->rdev.lldi.tids, atid);
2841 dst_release(ep->dst);
2842 cxgb4_l2t_release(ep->l2t);
2843 c4iw_put_ep(&ep->com);
1cab775c
VP		 2844}
2845
2846static void passive_ofld_conn_reply(struct c4iw_dev *dev, struct sk_buff *skb,
2847 struct cpl_fw6_msg_ofld_connection_wr_rpl *req)
2848{
2849 struct sk_buff *rpl_skb;
2850 struct cpl_pass_accept_req *cpl;
2851 int ret;
2852
710a3110 2853 rpl_skb = (struct sk_buff *)(unsigned long)req->cookie;
1cab775c
VP		 2854 BUG_ON(!rpl_skb);
2855 if (req->retval) {
2856 PDBG("%s passive open failure %d\n", __func__, req->retval);
793dad94
VP		 2857 mutex_lock(&dev->rdev.stats.lock);
2858 dev->rdev.stats.pas_ofld_conn_fails++;
2859 mutex_unlock(&dev->rdev.stats.lock);
1cab775c
VP		 2860 kfree_skb(rpl_skb);
2861 } else {
2862 cpl = (struct cpl_pass_accept_req *)cplhdr(rpl_skb);
2863 OPCODE_TID(cpl) = htonl(MK_OPCODE_TID(CPL_PASS_ACCEPT_REQ,
ef5d6355
VP		 2864 (__force u32) htonl(
2865 (__force u32) req->tid)));
1cab775c
VP		 2866 ret = pass_accept_req(dev, rpl_skb);
2867 if (!ret)
2868 kfree_skb(rpl_skb);
2869 }
2870 return;
2871}
2872
2873static int deferred_fw6_msg(struct c4iw_dev *dev, struct sk_buff *skb)
2f5b48c3
SW		 2874{
2875 struct cpl_fw6_msg *rpl = cplhdr(skb);
1cab775c
VP		 2876 struct cpl_fw6_msg_ofld_connection_wr_rpl *req;
2877
2878 switch (rpl->type) {
2879 case FW6_TYPE_CQE:
2880 c4iw_ev_dispatch(dev, (struct t4_cqe *)&rpl->data[0]);
2881 break;
2882 case FW6_TYPE_OFLD_CONNECTION_WR_RPL:
2883 req = (struct cpl_fw6_msg_ofld_connection_wr_rpl *)rpl->data;
2884 switch (req->t_state) {
2885 case TCP_SYN_SENT:
2886 active_ofld_conn_reply(dev, skb, req);
2887 break;
2888 case TCP_SYN_RECV:
2889 passive_ofld_conn_reply(dev, skb, req);
2890 break;
2891 default:
2892 pr_err("%s unexpected ofld conn wr state %d\n",
2893 __func__, req->t_state);
2894 break;
2895 }
2896 break;
2897 }
2898 return 0;
2899}
2900
2901static void build_cpl_pass_accept_req(struct sk_buff *skb, int stid , u8 tos)
2902{
2903 u32 l2info;
f079af7a 2904 u16 vlantag, len, hdr_len, eth_hdr_len;
1cab775c
VP		 2905 u8 intf;
2906 struct cpl_rx_pkt *cpl = cplhdr(skb);
2907 struct cpl_pass_accept_req *req;
2908 struct tcp_options_received tmp_opt;
f079af7a 2909 struct c4iw_dev *dev;
1cab775c 2910
f079af7a 2911 dev = *((struct c4iw_dev **) (skb->cb + sizeof(void *)));
1cab775c 2912 /* Store values from cpl_rx_pkt in temporary location. */
ef5d6355
VP		 2913 vlantag = (__force u16) cpl->vlan;
2914 len = (__force u16) cpl->len;
2915 l2info = (__force u32) cpl->l2info;
2916 hdr_len = (__force u16) cpl->hdr_len;
1cab775c
VP		 2917 intf = cpl->iff;
2918
2919 __skb_pull(skb, sizeof(*req) + sizeof(struct rss_header));
2920
2921 /*
2922 * We need to parse the TCP options from SYN packet.
2923 * to generate cpl_pass_accept_req.
2924 */
2925 memset(&tmp_opt, 0, sizeof(tmp_opt));
2926 tcp_clear_options(&tmp_opt);
1a2c6181 2927 tcp_parse_options(skb, &tmp_opt, 0, NULL);
1cab775c
VP		 2928
2929 req = (struct cpl_pass_accept_req *)__skb_push(skb, sizeof(*req));
2930 memset(req, 0, sizeof(*req));
2931 req->l2info = cpu_to_be16(V_SYN_INTF(intf) |
ef5d6355
VP		 2932 V_SYN_MAC_IDX(G_RX_MACIDX(
2933 (__force int) htonl(l2info))) |
1cab775c 2934 F_SYN_XACT_MATCH);
f079af7a
VP		 2935 eth_hdr_len = is_t4(dev->rdev.lldi.adapter_type) ?
2936 G_RX_ETHHDR_LEN((__force int) htonl(l2info)) :
2937 G_RX_T5_ETHHDR_LEN((__force int) htonl(l2info));
ef5d6355
VP		 2938 req->hdr_len = cpu_to_be32(V_SYN_RX_CHAN(G_RX_CHAN(
2939 (__force int) htonl(l2info))) |
2940 V_TCP_HDR_LEN(G_RX_TCPHDR_LEN(
2941 (__force int) htons(hdr_len))) |
2942 V_IP_HDR_LEN(G_RX_IPHDR_LEN(
2943 (__force int) htons(hdr_len))) |
f079af7a 2944 V_ETH_HDR_LEN(G_RX_ETHHDR_LEN(eth_hdr_len)));
ef5d6355
VP		 2945 req->vlan = (__force __be16) vlantag;
2946 req->len = (__force __be16) len;
1cab775c
VP		 2947 req->tos_stid = cpu_to_be32(PASS_OPEN_TID(stid) |
2948 PASS_OPEN_TOS(tos));
2949 req->tcpopt.mss = htons(tmp_opt.mss_clamp);
2950 if (tmp_opt.wscale_ok)
2951 req->tcpopt.wsf = tmp_opt.snd_wscale;
2952 req->tcpopt.tstamp = tmp_opt.saw_tstamp;
2953 if (tmp_opt.sack_ok)
2954 req->tcpopt.sack = 1;
2955 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_PASS_ACCEPT_REQ, 0));
2956 return;
2957}
2958
2959static void send_fw_pass_open_req(struct c4iw_dev *dev, struct sk_buff *skb,
2960 __be32 laddr, __be16 lport,
2961 __be32 raddr, __be16 rport,
2962 u32 rcv_isn, u32 filter, u16 window,
2963 u32 rss_qid, u8 port_id)
2964{
2965 struct sk_buff *req_skb;
2966 struct fw_ofld_connection_wr *req;
2967 struct cpl_pass_accept_req *cpl = cplhdr(skb);
2968
2969 req_skb = alloc_skb(sizeof(struct fw_ofld_connection_wr), GFP_KERNEL);
2970 req = (struct fw_ofld_connection_wr *)__skb_put(req_skb, sizeof(*req));
2971 memset(req, 0, sizeof(*req));
2972 req->op_compl = htonl(V_WR_OP(FW_OFLD_CONNECTION_WR) | FW_WR_COMPL(1));
2973 req->len16_pkd = htonl(FW_WR_LEN16(DIV_ROUND_UP(sizeof(*req), 16)));
2974 req->le.version_cpl = htonl(F_FW_OFLD_CONNECTION_WR_CPL);
ef5d6355 2975 req->le.filter = (__force __be32) filter;
1cab775c
VP		 2976 req->le.lport = lport;
2977 req->le.pport = rport;
2978 req->le.u.ipv4.lip = laddr;
2979 req->le.u.ipv4.pip = raddr;
2980 req->tcb.rcv_nxt = htonl(rcv_isn + 1);
2981 req->tcb.rcv_adv = htons(window);
2982 req->tcb.t_state_to_astid =
2983 htonl(V_FW_OFLD_CONNECTION_WR_T_STATE(TCP_SYN_RECV) |
2984 V_FW_OFLD_CONNECTION_WR_RCV_SCALE(cpl->tcpopt.wsf) |
2985 V_FW_OFLD_CONNECTION_WR_ASTID(
2986 GET_PASS_OPEN_TID(ntohl(cpl->tos_stid))));
2987
2988 /*
2989 * We store the qid in opt2 which will be used by the firmware
2990 * to send us the wr response.
2991 */
2992 req->tcb.opt2 = htonl(V_RSS_QUEUE(rss_qid));
2993
2994 /*
2995 * We initialize the MSS index in TCB to 0xF.
2996 * So that when driver sends cpl_pass_accept_rpl
2997 * TCB picks up the correct value. If this was 0
2998 * TP will ignore any value > 0 for MSS index.
2999 */
3000 req->tcb.opt0 = cpu_to_be64(V_MSS_IDX(0xF));
710a3110 3001 req->cookie = (unsigned long)skb;
1cab775c
VP		 3002
3003 set_wr_txq(req_skb, CPL_PRIORITY_CONTROL, port_id);
3004 cxgb4_ofld_send(dev->rdev.lldi.ports[0], req_skb);
3005}
3006
3007/*
3008 * Handler for CPL_RX_PKT message. Need to handle cpl_rx_pkt
3009 * messages when a filter is being used instead of server to
3010 * redirect a syn packet. When packets hit filter they are redirected
3011 * to the offload queue and driver tries to establish the connection
3012 * using firmware work request.
3013 */
3014static int rx_pkt(struct c4iw_dev *dev, struct sk_buff *skb)
3015{
3016 int stid;
3017 unsigned int filter;
3018 struct ethhdr *eh = NULL;
3019 struct vlan_ethhdr *vlan_eh = NULL;
3020 struct iphdr *iph;
3021 struct tcphdr *tcph;
3022 struct rss_header *rss = (void *)skb->data;
3023 struct cpl_rx_pkt *cpl = (void *)skb->data;
3024 struct cpl_pass_accept_req *req = (void *)(rss + 1);
3025 struct l2t_entry *e;
3026 struct dst_entry *dst;
3027 struct rtable *rt;
3028 struct c4iw_ep *lep;
3029 u16 window;
3030 struct port_info *pi;
3031 struct net_device *pdev;
f079af7a 3032 u16 rss_qid, eth_hdr_len;
1cab775c
VP		 3033 int step;
3034 u32 tx_chan;
3035 struct neighbour *neigh;
3036
3037 /* Drop all non-SYN packets */
3038 if (!(cpl->l2info & cpu_to_be32(F_RXF_SYN)))
3039 goto reject;
3040
3041 /*
3042 * Drop all packets which did not hit the filter.
3043 * Unlikely to happen.
3044 */
3045 if (!(rss->filter_hit && rss->filter_tid))
3046 goto reject;
3047
3048 /*
3049 * Calculate the server tid from filter hit index from cpl_rx_pkt.
3050 */
ef5d6355
VP		 3051 stid = (__force int) cpu_to_be32((__force u32) rss->hash_val)
3052 - dev->rdev.lldi.tids->sftid_base
1cab775c
VP		 3053 + dev->rdev.lldi.tids->nstids;
3054
3055 lep = (struct c4iw_ep *)lookup_stid(dev->rdev.lldi.tids, stid);
3056 if (!lep) {
3057 PDBG("%s connect request on invalid stid %d\n", __func__, stid);
3058 goto reject;
3059 }
3060
f079af7a
VP		 3061 eth_hdr_len = is_t4(dev->rdev.lldi.adapter_type) ?
3062 G_RX_ETHHDR_LEN(htonl(cpl->l2info)) :
3063 G_RX_T5_ETHHDR_LEN(htonl(cpl->l2info));
3064 if (eth_hdr_len == ETH_HLEN) {
1cab775c
VP		 3065 eh = (struct ethhdr *)(req + 1);
3066 iph = (struct iphdr *)(eh + 1);
3067 } else {
3068 vlan_eh = (struct vlan_ethhdr *)(req + 1);
3069 iph = (struct iphdr *)(vlan_eh + 1);
3070 skb->vlan_tci = ntohs(cpl->vlan);
3071 }
3072
3073 if (iph->version != 0x4)
3074 goto reject;
3075
3076 tcph = (struct tcphdr *)(iph + 1);
3077 skb_set_network_header(skb, (void *)iph - (void *)rss);
3078 skb_set_transport_header(skb, (void *)tcph - (void *)rss);
3079 skb_get(skb);
3080
3081 PDBG("%s lip 0x%x lport %u pip 0x%x pport %u tos %d\n", __func__,
3082 ntohl(iph->daddr), ntohs(tcph->dest), ntohl(iph->saddr),
3083 ntohs(tcph->source), iph->tos);
3084
3085 rt = find_route(dev, iph->daddr, iph->saddr, tcph->dest, tcph->source,
3086 iph->tos);
3087 if (!rt) {
3088 pr_err("%s - failed to find dst entry!\n",
3089 __func__);
3090 goto reject;
3091 }
3092 dst = &rt->dst;
3093 neigh = dst_neigh_lookup_skb(dst, skb);
3094
aaa0c23c
ZZ		 3095 if (!neigh) {
3096 pr_err("%s - failed to allocate neigh!\n",
3097 __func__);
3098 goto free_dst;
3099 }
3100
1cab775c
VP		 3101 if (neigh->dev->flags & IFF_LOOPBACK) {
3102 pdev = ip_dev_find(&init_net, iph->daddr);
3103 e = cxgb4_l2t_get(dev->rdev.lldi.l2t, neigh,
3104 pdev, 0);
3105 pi = (struct port_info *)netdev_priv(pdev);
3106 tx_chan = cxgb4_port_chan(pdev);
3107 dev_put(pdev);
3108 } else {
3109 e = cxgb4_l2t_get(dev->rdev.lldi.l2t, neigh,
3110 neigh->dev, 0);
3111 pi = (struct port_info *)netdev_priv(neigh->dev);
3112 tx_chan = cxgb4_port_chan(neigh->dev);
3113 }
3114 if (!e) {
3115 pr_err("%s - failed to allocate l2t entry!\n",
3116 __func__);
3117 goto free_dst;
3118 }
3119
3120 step = dev->rdev.lldi.nrxq / dev->rdev.lldi.nchan;
3121 rss_qid = dev->rdev.lldi.rxq_ids[pi->port_id * step];
ef5d6355 3122 window = (__force u16) htons((__force u16)tcph->window);
1cab775c
VP		 3123
3124 /* Calcuate filter portion for LE region. */
ef5d6355 3125 filter = (__force unsigned int) cpu_to_be32(select_ntuple(dev, dst, e));
1cab775c
VP		 3126
3127 /*
3128 * Synthesize the cpl_pass_accept_req. We have everything except the
3129 * TID. Once firmware sends a reply with TID we update the TID field
3130 * in cpl and pass it through the regular cpl_pass_accept_req path.
3131 */
3132 build_cpl_pass_accept_req(skb, stid, iph->tos);
3133 send_fw_pass_open_req(dev, skb, iph->daddr, tcph->dest, iph->saddr,
3134 tcph->source, ntohl(tcph->seq), filter, window,
3135 rss_qid, pi->port_id);
3136 cxgb4_l2t_release(e);
3137free_dst:
3138 dst_release(dst);
3139reject:
2f5b48c3
SW		 3140 return 0;
3141}
3142
be4c9bad
RD		 3143/*
3144 * These are the real handlers that are called from a
3145 * work queue.
3146 */
3147static c4iw_handler_func work_handlers[NUM_CPL_CMDS] = {
3148 [CPL_ACT_ESTABLISH] = act_establish,
3149 [CPL_ACT_OPEN_RPL] = act_open_rpl,
3150 [CPL_RX_DATA] = rx_data,
3151 [CPL_ABORT_RPL_RSS] = abort_rpl,
3152 [CPL_ABORT_RPL] = abort_rpl,
3153 [CPL_PASS_OPEN_RPL] = pass_open_rpl,
3154 [CPL_CLOSE_LISTSRV_RPL] = close_listsrv_rpl,
3155 [CPL_PASS_ACCEPT_REQ] = pass_accept_req,
3156 [CPL_PASS_ESTABLISH] = pass_establish,
3157 [CPL_PEER_CLOSE] = peer_close,
3158 [CPL_ABORT_REQ_RSS] = peer_abort,
3159 [CPL_CLOSE_CON_RPL] = close_con_rpl,
3160 [CPL_RDMA_TERMINATE] = terminate,
2f5b48c3 3161 [CPL_FW4_ACK] = fw4_ack,
1cab775c
VP		 3162 [CPL_FW6_MSG] = deferred_fw6_msg,
3163 [CPL_RX_PKT] = rx_pkt
be4c9bad
RD		 3164};
3165
3166static void process_timeout(struct c4iw_ep *ep)
3167{
3168 struct c4iw_qp_attributes attrs;
3169 int abort = 1;
3170
2f5b48c3 3171 mutex_lock(&ep->com.mutex);
be4c9bad
RD		 3172 PDBG("%s ep %p tid %u state %d\n", __func__, ep, ep->hwtid,
3173 ep->com.state);
793dad94 3174 set_bit(TIMEDOUT, &ep->com.history);
be4c9bad
RD		 3175 switch (ep->com.state) {
3176 case MPA_REQ_SENT:
3177 __state_set(&ep->com, ABORTING);
3178 connect_reply_upcall(ep, -ETIMEDOUT);
3179 break;
3180 case MPA_REQ_WAIT:
3181 __state_set(&ep->com, ABORTING);
3182 break;
3183 case CLOSING:
3184 case MORIBUND:
3185 if (ep->com.cm_id && ep->com.qp) {
3186 attrs.next_state = C4IW_QP_STATE_ERROR;
3187 c4iw_modify_qp(ep->com.qp->rhp,
3188 ep->com.qp, C4IW_QP_ATTR_NEXT_STATE,
3189 &attrs, 1);
3190 }
3191 __state_set(&ep->com, ABORTING);
3192 break;
3193 default:
76f267b7 3194 WARN(1, "%s unexpected state ep %p tid %u state %u\n",
be4c9bad 3195 __func__, ep, ep->hwtid, ep->com.state);
be4c9bad
RD		 3196 abort = 0;
3197 }
2f5b48c3 3198 mutex_unlock(&ep->com.mutex);
be4c9bad
RD		 3199 if (abort)
3200 abort_connection(ep, NULL, GFP_KERNEL);
3201 c4iw_put_ep(&ep->com);
3202}
3203
3204static void process_timedout_eps(void)
3205{
3206 struct c4iw_ep *ep;
3207
3208 spin_lock_irq(&timeout_lock);
3209 while (!list_empty(&timeout_list)) {
3210 struct list_head *tmp;
3211
3212 tmp = timeout_list.next;
3213 list_del(tmp);
3214 spin_unlock_irq(&timeout_lock);
3215 ep = list_entry(tmp, struct c4iw_ep, entry);
3216 process_timeout(ep);
3217 spin_lock_irq(&timeout_lock);
3218 }
3219 spin_unlock_irq(&timeout_lock);
3220}
3221
3222static void process_work(struct work_struct *work)
3223{
3224 struct sk_buff *skb = NULL;
3225 struct c4iw_dev *dev;
c1d7356c 3226 struct cpl_act_establish *rpl;
be4c9bad
RD		 3227 unsigned int opcode;
3228 int ret;
3229
3230 while ((skb = skb_dequeue(&rxq))) {
3231 rpl = cplhdr(skb);
3232 dev = *((struct c4iw_dev **) (skb->cb + sizeof(void *)));
3233 opcode = rpl->ot.opcode;
3234
3235 BUG_ON(!work_handlers[opcode]);
3236 ret = work_handlers[opcode](dev, skb);
3237 if (!ret)
3238 kfree_skb(skb);
3239 }
3240 process_timedout_eps();
3241}
3242
3243static DECLARE_WORK(skb_work, process_work);
3244
3245static void ep_timeout(unsigned long arg)
3246{
3247 struct c4iw_ep *ep = (struct c4iw_ep *)arg;
1ec779cc 3248 int kickit = 0;
be4c9bad
RD		 3249
3250 spin_lock(&timeout_lock);
1ec779cc
VP		 3251 if (!test_and_set_bit(TIMEOUT, &ep->com.flags)) {
3252 list_add_tail(&ep->entry, &timeout_list);
3253 kickit = 1;
3254 }
be4c9bad 3255 spin_unlock(&timeout_lock);
1ec779cc
VP		 3256 if (kickit)
3257 queue_work(workq, &skb_work);
be4c9bad
RD		 3258}
3259
cfdda9d7
SW		 3260/*
3261 * All the CM events are handled on a work queue to have a safe context.
3262 */
3263static int sched(struct c4iw_dev *dev, struct sk_buff *skb)
3264{
3265
3266 /*
3267 * Save dev in the skb->cb area.
3268 */
3269 *((struct c4iw_dev **) (skb->cb + sizeof(void *))) = dev;
3270
3271 /*
3272 * Queue the skb and schedule the worker thread.
3273 */
3274 skb_queue_tail(&rxq, skb);
3275 queue_work(workq, &skb_work);
3276 return 0;
3277}
3278
3279static int set_tcb_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
3280{
3281 struct cpl_set_tcb_rpl *rpl = cplhdr(skb);
3282
3283 if (rpl->status != CPL_ERR_NONE) {
3284 printk(KERN_ERR MOD "Unexpected SET_TCB_RPL status %u "
3285 "for tid %u\n", rpl->status, GET_TID(rpl));
3286 }
2f5b48c3 3287 kfree_skb(skb);
cfdda9d7
SW		 3288 return 0;
3289}
3290
be4c9bad
RD		 3291static int fw6_msg(struct c4iw_dev *dev, struct sk_buff *skb)
3292{
3293 struct cpl_fw6_msg *rpl = cplhdr(skb);
3294 struct c4iw_wr_wait *wr_waitp;
3295 int ret;
3296
3297 PDBG("%s type %u\n", __func__, rpl->type);
3298
3299 switch (rpl->type) {
5be78ee9 3300 case FW6_TYPE_WR_RPL:
be4c9bad 3301 ret = (int)((be64_to_cpu(rpl->data[0]) >> 8) & 0xff);
c8e081a1 3302 wr_waitp = (struct c4iw_wr_wait *)(__force unsigned long) rpl->data[1];
be4c9bad 3303 PDBG("%s wr_waitp %p ret %u\n", __func__, wr_waitp, ret);
d9594d99
SW		 3304 if (wr_waitp)
3305 c4iw_wake_up(wr_waitp, ret ? -ret : 0);
2f5b48c3 3306 kfree_skb(skb);
be4c9bad 3307 break;
5be78ee9 3308 case FW6_TYPE_CQE:
5be78ee9 3309 case FW6_TYPE_OFLD_CONNECTION_WR_RPL:
1cab775c 3310 sched(dev, skb);
5be78ee9 3311 break;
be4c9bad
RD		 3312 default:
3313 printk(KERN_ERR MOD "%s unexpected fw6 msg type %u\n", __func__,
3314 rpl->type);
2f5b48c3 3315 kfree_skb(skb);
be4c9bad
RD		 3316 break;
3317 }
3318 return 0;
3319}
3320
8da7e7a5
SW		 3321static int peer_abort_intr(struct c4iw_dev *dev, struct sk_buff *skb)
3322{
3323 struct cpl_abort_req_rss *req = cplhdr(skb);
3324 struct c4iw_ep *ep;
3325 struct tid_info *t = dev->rdev.lldi.tids;
3326 unsigned int tid = GET_TID(req);
3327
3328 ep = lookup_tid(t, tid);
14b92228
SW		 3329 if (!ep) {
3330 printk(KERN_WARNING MOD
3331 "Abort on non-existent endpoint, tid %d\n", tid);
3332 kfree_skb(skb);
3333 return 0;
3334 }
8da7e7a5
SW		 3335 if (is_neg_adv_abort(req->status)) {
3336 PDBG("%s neg_adv_abort ep %p tid %u\n", __func__, ep,
3337 ep->hwtid);
3338 kfree_skb(skb);
3339 return 0;
3340 }
3341 PDBG("%s ep %p tid %u state %u\n", __func__, ep, ep->hwtid,
3342 ep->com.state);
3343
3344 /*
3345 * Wake up any threads in rdma_init() or rdma_fini().
7c0a33d6
VP		 3346 * However, if we are on MPAv2 and want to retry with MPAv1
3347 * then, don't wake up yet.
8da7e7a5 3348 */
7c0a33d6
VP		 3349 if (mpa_rev == 2 && !ep->tried_with_mpa_v1) {
3350 if (ep->com.state != MPA_REQ_SENT)
3351 c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
3352 } else
3353 c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
8da7e7a5
SW		 3354 sched(dev, skb);
3355 return 0;
3356}
3357
be4c9bad
RD		 3358/*
3359 * Most upcalls from the T4 Core go to sched() to
3360 * schedule the processing on a work queue.
3361 */
3362c4iw_handler_func c4iw_handlers[NUM_CPL_CMDS] = {
3363 [CPL_ACT_ESTABLISH] = sched,
3364 [CPL_ACT_OPEN_RPL] = sched,
3365 [CPL_RX_DATA] = sched,
3366 [CPL_ABORT_RPL_RSS] = sched,
3367 [CPL_ABORT_RPL] = sched,
3368 [CPL_PASS_OPEN_RPL] = sched,
3369 [CPL_CLOSE_LISTSRV_RPL] = sched,
3370 [CPL_PASS_ACCEPT_REQ] = sched,
3371 [CPL_PASS_ESTABLISH] = sched,
3372 [CPL_PEER_CLOSE] = sched,
3373 [CPL_CLOSE_CON_RPL] = sched,
8da7e7a5 3374 [CPL_ABORT_REQ_RSS] = peer_abort_intr,
be4c9bad
RD		 3375 [CPL_RDMA_TERMINATE] = sched,
3376 [CPL_FW4_ACK] = sched,
3377 [CPL_SET_TCB_RPL] = set_tcb_rpl,
1cab775c
VP		 3378 [CPL_FW6_MSG] = fw6_msg,
3379 [CPL_RX_PKT] = sched
be4c9bad
RD		 3380};
3381
cfdda9d7
SW		 3382int __init c4iw_cm_init(void)
3383{
be4c9bad 3384 spin_lock_init(&timeout_lock);
cfdda9d7
SW		 3385 skb_queue_head_init(&rxq);
3386
3387 workq = create_singlethread_workqueue("iw_cxgb4");
3388 if (!workq)
3389 return -ENOMEM;
3390
cfdda9d7
SW		 3391 return 0;
3392}
3393
3394void __exit c4iw_cm_term(void)
3395{
be4c9bad 3396 WARN_ON(!list_empty(&timeout_list));
cfdda9d7
SW		 3397 flush_workqueue(workq);
3398 destroy_workqueue(workq);
3399}
Linux 6.x block layer and io_uring tree(s)