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/moduleparam.h>
34 #include <linux/debugfs.h>
35 #include <linux/vmalloc.h>
36 
37 #include <rdma/ib_verbs.h>
38 
39 #include "iw_cxgb4.h"
40 
41 #define DRV_VERSION "0.1"
42 
43 MODULE_AUTHOR("Steve Wise");
44 MODULE_DESCRIPTION("Chelsio T4/T5 RDMA Driver");
45 MODULE_LICENSE("Dual BSD/GPL");
46 MODULE_VERSION(DRV_VERSION);
47 
48 static int allow_db_fc_on_t5;
49 module_param(allow_db_fc_on_t5, int, 0644);
50 MODULE_PARM_DESC(allow_db_fc_on_t5,
51 		 "Allow DB Flow Control on T5 (default = 0)");
52 
53 static int allow_db_coalescing_on_t5;
54 module_param(allow_db_coalescing_on_t5, int, 0644);
55 MODULE_PARM_DESC(allow_db_coalescing_on_t5,
56 		 "Allow DB Coalescing on T5 (default = 0)");
57 
58 struct uld_ctx {
59 	struct list_head entry;
60 	struct cxgb4_lld_info lldi;
61 	struct c4iw_dev *dev;
62 };
63 
64 static LIST_HEAD(uld_ctx_list);
65 static DEFINE_MUTEX(dev_mutex);
66 
67 #define DB_FC_RESUME_SIZE 64
68 #define DB_FC_RESUME_DELAY 1
69 #define DB_FC_DRAIN_THRESH 0
70 
71 static struct dentry *c4iw_debugfs_root;
72 
73 struct c4iw_debugfs_data {
74 	struct c4iw_dev *devp;
75 	char *buf;
76 	int bufsize;
77 	int pos;
78 };
79 
80 /* registered cxgb4 netlink callbacks */
81 static struct ibnl_client_cbs c4iw_nl_cb_table[] = {
82 	[RDMA_NL_IWPM_REG_PID] = {.dump = iwpm_register_pid_cb},
83 	[RDMA_NL_IWPM_ADD_MAPPING] = {.dump = iwpm_add_mapping_cb},
84 	[RDMA_NL_IWPM_QUERY_MAPPING] = {.dump = iwpm_add_and_query_mapping_cb},
85 	[RDMA_NL_IWPM_HANDLE_ERR] = {.dump = iwpm_mapping_error_cb},
86 	[RDMA_NL_IWPM_MAPINFO] = {.dump = iwpm_mapping_info_cb},
87 	[RDMA_NL_IWPM_MAPINFO_NUM] = {.dump = iwpm_ack_mapping_info_cb}
88 };
89 
90 static int count_idrs(int id, void *p, void *data)
91 {
92 	int *countp = data;
93 
94 	*countp = *countp + 1;
95 	return 0;
96 }
97 
98 static ssize_t debugfs_read(struct file *file, char __user *buf, size_t count,
99 			    loff_t *ppos)
100 {
101 	struct c4iw_debugfs_data *d = file->private_data;
102 
103 	return simple_read_from_buffer(buf, count, ppos, d->buf, d->pos);
104 }
105 
106 static int dump_qp(int id, void *p, void *data)
107 {
108 	struct c4iw_qp *qp = p;
109 	struct c4iw_debugfs_data *qpd = data;
110 	int space;
111 	int cc;
112 
113 	if (id != qp->wq.sq.qid)
114 		return 0;
115 
116 	space = qpd->bufsize - qpd->pos - 1;
117 	if (space == 0)
118 		return 1;
119 
120 	if (qp->ep) {
121 		if (qp->ep->com.local_addr.ss_family == AF_INET) {
122 			struct sockaddr_in *lsin = (struct sockaddr_in *)
123 				&qp->ep->com.local_addr;
124 			struct sockaddr_in *rsin = (struct sockaddr_in *)
125 				&qp->ep->com.remote_addr;
126 			struct sockaddr_in *mapped_lsin = (struct sockaddr_in *)
127 				&qp->ep->com.mapped_local_addr;
128 			struct sockaddr_in *mapped_rsin = (struct sockaddr_in *)
129 				&qp->ep->com.mapped_remote_addr;
130 
131 			cc = snprintf(qpd->buf + qpd->pos, space,
132 				      "rc qp sq id %u rq id %u state %u "
133 				      "onchip %u ep tid %u state %u "
134 				      "%pI4:%u/%u->%pI4:%u/%u\n",
135 				      qp->wq.sq.qid, qp->wq.rq.qid,
136 				      (int)qp->attr.state,
137 				      qp->wq.sq.flags & T4_SQ_ONCHIP,
138 				      qp->ep->hwtid, (int)qp->ep->com.state,
139 				      &lsin->sin_addr, ntohs(lsin->sin_port),
140 				      ntohs(mapped_lsin->sin_port),
141 				      &rsin->sin_addr, ntohs(rsin->sin_port),
142 				      ntohs(mapped_rsin->sin_port));
143 		} else {
144 			struct sockaddr_in6 *lsin6 = (struct sockaddr_in6 *)
145 				&qp->ep->com.local_addr;
146 			struct sockaddr_in6 *rsin6 = (struct sockaddr_in6 *)
147 				&qp->ep->com.remote_addr;
148 			struct sockaddr_in6 *mapped_lsin6 =
149 				(struct sockaddr_in6 *)
150 				&qp->ep->com.mapped_local_addr;
151 			struct sockaddr_in6 *mapped_rsin6 =
152 				(struct sockaddr_in6 *)
153 				&qp->ep->com.mapped_remote_addr;
154 
155 			cc = snprintf(qpd->buf + qpd->pos, space,
156 				      "rc qp sq id %u rq id %u state %u "
157 				      "onchip %u ep tid %u state %u "
158 				      "%pI6:%u/%u->%pI6:%u/%u\n",
159 				      qp->wq.sq.qid, qp->wq.rq.qid,
160 				      (int)qp->attr.state,
161 				      qp->wq.sq.flags & T4_SQ_ONCHIP,
162 				      qp->ep->hwtid, (int)qp->ep->com.state,
163 				      &lsin6->sin6_addr,
164 				      ntohs(lsin6->sin6_port),
165 				      ntohs(mapped_lsin6->sin6_port),
166 				      &rsin6->sin6_addr,
167 				      ntohs(rsin6->sin6_port),
168 				      ntohs(mapped_rsin6->sin6_port));
169 		}
170 	} else
171 		cc = snprintf(qpd->buf + qpd->pos, space,
172 			     "qp sq id %u rq id %u state %u onchip %u\n",
173 			      qp->wq.sq.qid, qp->wq.rq.qid,
174 			      (int)qp->attr.state,
175 			      qp->wq.sq.flags & T4_SQ_ONCHIP);
176 	if (cc < space)
177 		qpd->pos += cc;
178 	return 0;
179 }
180 
181 static int qp_release(struct inode *inode, struct file *file)
182 {
183 	struct c4iw_debugfs_data *qpd = file->private_data;
184 	if (!qpd) {
185 		printk(KERN_INFO "%s null qpd?\n", __func__);
186 		return 0;
187 	}
188 	vfree(qpd->buf);
189 	kfree(qpd);
190 	return 0;
191 }
192 
193 static int qp_open(struct inode *inode, struct file *file)
194 {
195 	struct c4iw_debugfs_data *qpd;
196 	int ret = 0;
197 	int count = 1;
198 
199 	qpd = kmalloc(sizeof *qpd, GFP_KERNEL);
200 	if (!qpd) {
201 		ret = -ENOMEM;
202 		goto out;
203 	}
204 	qpd->devp = inode->i_private;
205 	qpd->pos = 0;
206 
207 	spin_lock_irq(&qpd->devp->lock);
208 	idr_for_each(&qpd->devp->qpidr, count_idrs, &count);
209 	spin_unlock_irq(&qpd->devp->lock);
210 
211 	qpd->bufsize = count * 128;
212 	qpd->buf = vmalloc(qpd->bufsize);
213 	if (!qpd->buf) {
214 		ret = -ENOMEM;
215 		goto err1;
216 	}
217 
218 	spin_lock_irq(&qpd->devp->lock);
219 	idr_for_each(&qpd->devp->qpidr, dump_qp, qpd);
220 	spin_unlock_irq(&qpd->devp->lock);
221 
222 	qpd->buf[qpd->pos++] = 0;
223 	file->private_data = qpd;
224 	goto out;
225 err1:
226 	kfree(qpd);
227 out:
228 	return ret;
229 }
230 
231 static const struct file_operations qp_debugfs_fops = {
232 	.owner   = THIS_MODULE,
233 	.open    = qp_open,
234 	.release = qp_release,
235 	.read    = debugfs_read,
236 	.llseek  = default_llseek,
237 };
238 
239 static int dump_stag(int id, void *p, void *data)
240 {
241 	struct c4iw_debugfs_data *stagd = data;
242 	int space;
243 	int cc;
244 
245 	space = stagd->bufsize - stagd->pos - 1;
246 	if (space == 0)
247 		return 1;
248 
249 	cc = snprintf(stagd->buf + stagd->pos, space, "0x%x\n", id<<8);
250 	if (cc < space)
251 		stagd->pos += cc;
252 	return 0;
253 }
254 
255 static int stag_release(struct inode *inode, struct file *file)
256 {
257 	struct c4iw_debugfs_data *stagd = file->private_data;
258 	if (!stagd) {
259 		printk(KERN_INFO "%s null stagd?\n", __func__);
260 		return 0;
261 	}
262 	kfree(stagd->buf);
263 	kfree(stagd);
264 	return 0;
265 }
266 
267 static int stag_open(struct inode *inode, struct file *file)
268 {
269 	struct c4iw_debugfs_data *stagd;
270 	int ret = 0;
271 	int count = 1;
272 
273 	stagd = kmalloc(sizeof *stagd, GFP_KERNEL);
274 	if (!stagd) {
275 		ret = -ENOMEM;
276 		goto out;
277 	}
278 	stagd->devp = inode->i_private;
279 	stagd->pos = 0;
280 
281 	spin_lock_irq(&stagd->devp->lock);
282 	idr_for_each(&stagd->devp->mmidr, count_idrs, &count);
283 	spin_unlock_irq(&stagd->devp->lock);
284 
285 	stagd->bufsize = count * sizeof("0x12345678\n");
286 	stagd->buf = kmalloc(stagd->bufsize, GFP_KERNEL);
287 	if (!stagd->buf) {
288 		ret = -ENOMEM;
289 		goto err1;
290 	}
291 
292 	spin_lock_irq(&stagd->devp->lock);
293 	idr_for_each(&stagd->devp->mmidr, dump_stag, stagd);
294 	spin_unlock_irq(&stagd->devp->lock);
295 
296 	stagd->buf[stagd->pos++] = 0;
297 	file->private_data = stagd;
298 	goto out;
299 err1:
300 	kfree(stagd);
301 out:
302 	return ret;
303 }
304 
305 static const struct file_operations stag_debugfs_fops = {
306 	.owner   = THIS_MODULE,
307 	.open    = stag_open,
308 	.release = stag_release,
309 	.read    = debugfs_read,
310 	.llseek  = default_llseek,
311 };
312 
313 static char *db_state_str[] = {"NORMAL", "FLOW_CONTROL", "RECOVERY", "STOPPED"};
314 
315 static int stats_show(struct seq_file *seq, void *v)
316 {
317 	struct c4iw_dev *dev = seq->private;
318 
319 	seq_printf(seq, "   Object: %10s %10s %10s %10s\n", "Total", "Current",
320 		   "Max", "Fail");
321 	seq_printf(seq, "     PDID: %10llu %10llu %10llu %10llu\n",
322 			dev->rdev.stats.pd.total, dev->rdev.stats.pd.cur,
323 			dev->rdev.stats.pd.max, dev->rdev.stats.pd.fail);
324 	seq_printf(seq, "      QID: %10llu %10llu %10llu %10llu\n",
325 			dev->rdev.stats.qid.total, dev->rdev.stats.qid.cur,
326 			dev->rdev.stats.qid.max, dev->rdev.stats.qid.fail);
327 	seq_printf(seq, "   TPTMEM: %10llu %10llu %10llu %10llu\n",
328 			dev->rdev.stats.stag.total, dev->rdev.stats.stag.cur,
329 			dev->rdev.stats.stag.max, dev->rdev.stats.stag.fail);
330 	seq_printf(seq, "   PBLMEM: %10llu %10llu %10llu %10llu\n",
331 			dev->rdev.stats.pbl.total, dev->rdev.stats.pbl.cur,
332 			dev->rdev.stats.pbl.max, dev->rdev.stats.pbl.fail);
333 	seq_printf(seq, "   RQTMEM: %10llu %10llu %10llu %10llu\n",
334 			dev->rdev.stats.rqt.total, dev->rdev.stats.rqt.cur,
335 			dev->rdev.stats.rqt.max, dev->rdev.stats.rqt.fail);
336 	seq_printf(seq, "  OCQPMEM: %10llu %10llu %10llu %10llu\n",
337 			dev->rdev.stats.ocqp.total, dev->rdev.stats.ocqp.cur,
338 			dev->rdev.stats.ocqp.max, dev->rdev.stats.ocqp.fail);
339 	seq_printf(seq, "  DB FULL: %10llu\n", dev->rdev.stats.db_full);
340 	seq_printf(seq, " DB EMPTY: %10llu\n", dev->rdev.stats.db_empty);
341 	seq_printf(seq, "  DB DROP: %10llu\n", dev->rdev.stats.db_drop);
342 	seq_printf(seq, " DB State: %s Transitions %llu FC Interruptions %llu\n",
343 		   db_state_str[dev->db_state],
344 		   dev->rdev.stats.db_state_transitions,
345 		   dev->rdev.stats.db_fc_interruptions);
346 	seq_printf(seq, "TCAM_FULL: %10llu\n", dev->rdev.stats.tcam_full);
347 	seq_printf(seq, "ACT_OFLD_CONN_FAILS: %10llu\n",
348 		   dev->rdev.stats.act_ofld_conn_fails);
349 	seq_printf(seq, "PAS_OFLD_CONN_FAILS: %10llu\n",
350 		   dev->rdev.stats.pas_ofld_conn_fails);
351 	return 0;
352 }
353 
354 static int stats_open(struct inode *inode, struct file *file)
355 {
356 	return single_open(file, stats_show, inode->i_private);
357 }
358 
359 static ssize_t stats_clear(struct file *file, const char __user *buf,
360 		size_t count, loff_t *pos)
361 {
362 	struct c4iw_dev *dev = ((struct seq_file *)file->private_data)->private;
363 
364 	mutex_lock(&dev->rdev.stats.lock);
365 	dev->rdev.stats.pd.max = 0;
366 	dev->rdev.stats.pd.fail = 0;
367 	dev->rdev.stats.qid.max = 0;
368 	dev->rdev.stats.qid.fail = 0;
369 	dev->rdev.stats.stag.max = 0;
370 	dev->rdev.stats.stag.fail = 0;
371 	dev->rdev.stats.pbl.max = 0;
372 	dev->rdev.stats.pbl.fail = 0;
373 	dev->rdev.stats.rqt.max = 0;
374 	dev->rdev.stats.rqt.fail = 0;
375 	dev->rdev.stats.ocqp.max = 0;
376 	dev->rdev.stats.ocqp.fail = 0;
377 	dev->rdev.stats.db_full = 0;
378 	dev->rdev.stats.db_empty = 0;
379 	dev->rdev.stats.db_drop = 0;
380 	dev->rdev.stats.db_state_transitions = 0;
381 	dev->rdev.stats.tcam_full = 0;
382 	dev->rdev.stats.act_ofld_conn_fails = 0;
383 	dev->rdev.stats.pas_ofld_conn_fails = 0;
384 	mutex_unlock(&dev->rdev.stats.lock);
385 	return count;
386 }
387 
388 static const struct file_operations stats_debugfs_fops = {
389 	.owner   = THIS_MODULE,
390 	.open    = stats_open,
391 	.release = single_release,
392 	.read    = seq_read,
393 	.llseek  = seq_lseek,
394 	.write   = stats_clear,
395 };
396 
397 static int dump_ep(int id, void *p, void *data)
398 {
399 	struct c4iw_ep *ep = p;
400 	struct c4iw_debugfs_data *epd = data;
401 	int space;
402 	int cc;
403 
404 	space = epd->bufsize - epd->pos - 1;
405 	if (space == 0)
406 		return 1;
407 
408 	if (ep->com.local_addr.ss_family == AF_INET) {
409 		struct sockaddr_in *lsin = (struct sockaddr_in *)
410 			&ep->com.local_addr;
411 		struct sockaddr_in *rsin = (struct sockaddr_in *)
412 			&ep->com.remote_addr;
413 		struct sockaddr_in *mapped_lsin = (struct sockaddr_in *)
414 			&ep->com.mapped_local_addr;
415 		struct sockaddr_in *mapped_rsin = (struct sockaddr_in *)
416 			&ep->com.mapped_remote_addr;
417 
418 		cc = snprintf(epd->buf + epd->pos, space,
419 			      "ep %p cm_id %p qp %p state %d flags 0x%lx "
420 			      "history 0x%lx hwtid %d atid %d "
421 			      "%pI4:%d/%d <-> %pI4:%d/%d\n",
422 			      ep, ep->com.cm_id, ep->com.qp,
423 			      (int)ep->com.state, ep->com.flags,
424 			      ep->com.history, ep->hwtid, ep->atid,
425 			      &lsin->sin_addr, ntohs(lsin->sin_port),
426 			      ntohs(mapped_lsin->sin_port),
427 			      &rsin->sin_addr, ntohs(rsin->sin_port),
428 			      ntohs(mapped_rsin->sin_port));
429 	} else {
430 		struct sockaddr_in6 *lsin6 = (struct sockaddr_in6 *)
431 			&ep->com.local_addr;
432 		struct sockaddr_in6 *rsin6 = (struct sockaddr_in6 *)
433 			&ep->com.remote_addr;
434 		struct sockaddr_in6 *mapped_lsin6 = (struct sockaddr_in6 *)
435 			&ep->com.mapped_local_addr;
436 		struct sockaddr_in6 *mapped_rsin6 = (struct sockaddr_in6 *)
437 			&ep->com.mapped_remote_addr;
438 
439 		cc = snprintf(epd->buf + epd->pos, space,
440 			      "ep %p cm_id %p qp %p state %d flags 0x%lx "
441 			      "history 0x%lx hwtid %d atid %d "
442 			      "%pI6:%d/%d <-> %pI6:%d/%d\n",
443 			      ep, ep->com.cm_id, ep->com.qp,
444 			      (int)ep->com.state, ep->com.flags,
445 			      ep->com.history, ep->hwtid, ep->atid,
446 			      &lsin6->sin6_addr, ntohs(lsin6->sin6_port),
447 			      ntohs(mapped_lsin6->sin6_port),
448 			      &rsin6->sin6_addr, ntohs(rsin6->sin6_port),
449 			      ntohs(mapped_rsin6->sin6_port));
450 	}
451 	if (cc < space)
452 		epd->pos += cc;
453 	return 0;
454 }
455 
456 static int dump_listen_ep(int id, void *p, void *data)
457 {
458 	struct c4iw_listen_ep *ep = p;
459 	struct c4iw_debugfs_data *epd = data;
460 	int space;
461 	int cc;
462 
463 	space = epd->bufsize - epd->pos - 1;
464 	if (space == 0)
465 		return 1;
466 
467 	if (ep->com.local_addr.ss_family == AF_INET) {
468 		struct sockaddr_in *lsin = (struct sockaddr_in *)
469 			&ep->com.local_addr;
470 		struct sockaddr_in *mapped_lsin = (struct sockaddr_in *)
471 			&ep->com.mapped_local_addr;
472 
473 		cc = snprintf(epd->buf + epd->pos, space,
474 			      "ep %p cm_id %p state %d flags 0x%lx stid %d "
475 			      "backlog %d %pI4:%d/%d\n",
476 			      ep, ep->com.cm_id, (int)ep->com.state,
477 			      ep->com.flags, ep->stid, ep->backlog,
478 			      &lsin->sin_addr, ntohs(lsin->sin_port),
479 			      ntohs(mapped_lsin->sin_port));
480 	} else {
481 		struct sockaddr_in6 *lsin6 = (struct sockaddr_in6 *)
482 			&ep->com.local_addr;
483 		struct sockaddr_in6 *mapped_lsin6 = (struct sockaddr_in6 *)
484 			&ep->com.mapped_local_addr;
485 
486 		cc = snprintf(epd->buf + epd->pos, space,
487 			      "ep %p cm_id %p state %d flags 0x%lx stid %d "
488 			      "backlog %d %pI6:%d/%d\n",
489 			      ep, ep->com.cm_id, (int)ep->com.state,
490 			      ep->com.flags, ep->stid, ep->backlog,
491 			      &lsin6->sin6_addr, ntohs(lsin6->sin6_port),
492 			      ntohs(mapped_lsin6->sin6_port));
493 	}
494 	if (cc < space)
495 		epd->pos += cc;
496 	return 0;
497 }
498 
499 static int ep_release(struct inode *inode, struct file *file)
500 {
501 	struct c4iw_debugfs_data *epd = file->private_data;
502 	if (!epd) {
503 		pr_info("%s null qpd?\n", __func__);
504 		return 0;
505 	}
506 	vfree(epd->buf);
507 	kfree(epd);
508 	return 0;
509 }
510 
511 static int ep_open(struct inode *inode, struct file *file)
512 {
513 	struct c4iw_debugfs_data *epd;
514 	int ret = 0;
515 	int count = 1;
516 
517 	epd = kmalloc(sizeof(*epd), GFP_KERNEL);
518 	if (!epd) {
519 		ret = -ENOMEM;
520 		goto out;
521 	}
522 	epd->devp = inode->i_private;
523 	epd->pos = 0;
524 
525 	spin_lock_irq(&epd->devp->lock);
526 	idr_for_each(&epd->devp->hwtid_idr, count_idrs, &count);
527 	idr_for_each(&epd->devp->atid_idr, count_idrs, &count);
528 	idr_for_each(&epd->devp->stid_idr, count_idrs, &count);
529 	spin_unlock_irq(&epd->devp->lock);
530 
531 	epd->bufsize = count * 160;
532 	epd->buf = vmalloc(epd->bufsize);
533 	if (!epd->buf) {
534 		ret = -ENOMEM;
535 		goto err1;
536 	}
537 
538 	spin_lock_irq(&epd->devp->lock);
539 	idr_for_each(&epd->devp->hwtid_idr, dump_ep, epd);
540 	idr_for_each(&epd->devp->atid_idr, dump_ep, epd);
541 	idr_for_each(&epd->devp->stid_idr, dump_listen_ep, epd);
542 	spin_unlock_irq(&epd->devp->lock);
543 
544 	file->private_data = epd;
545 	goto out;
546 err1:
547 	kfree(epd);
548 out:
549 	return ret;
550 }
551 
552 static const struct file_operations ep_debugfs_fops = {
553 	.owner   = THIS_MODULE,
554 	.open    = ep_open,
555 	.release = ep_release,
556 	.read    = debugfs_read,
557 };
558 
559 static int setup_debugfs(struct c4iw_dev *devp)
560 {
561 	struct dentry *de;
562 
563 	if (!devp->debugfs_root)
564 		return -1;
565 
566 	de = debugfs_create_file("qps", S_IWUSR, devp->debugfs_root,
567 				 (void *)devp, &qp_debugfs_fops);
568 	if (de && de->d_inode)
569 		de->d_inode->i_size = 4096;
570 
571 	de = debugfs_create_file("stags", S_IWUSR, devp->debugfs_root,
572 				 (void *)devp, &stag_debugfs_fops);
573 	if (de && de->d_inode)
574 		de->d_inode->i_size = 4096;
575 
576 	de = debugfs_create_file("stats", S_IWUSR, devp->debugfs_root,
577 			(void *)devp, &stats_debugfs_fops);
578 	if (de && de->d_inode)
579 		de->d_inode->i_size = 4096;
580 
581 	de = debugfs_create_file("eps", S_IWUSR, devp->debugfs_root,
582 			(void *)devp, &ep_debugfs_fops);
583 	if (de && de->d_inode)
584 		de->d_inode->i_size = 4096;
585 
586 	return 0;
587 }
588 
589 void c4iw_release_dev_ucontext(struct c4iw_rdev *rdev,
590 			       struct c4iw_dev_ucontext *uctx)
591 {
592 	struct list_head *pos, *nxt;
593 	struct c4iw_qid_list *entry;
594 
595 	mutex_lock(&uctx->lock);
596 	list_for_each_safe(pos, nxt, &uctx->qpids) {
597 		entry = list_entry(pos, struct c4iw_qid_list, entry);
598 		list_del_init(&entry->entry);
599 		if (!(entry->qid & rdev->qpmask)) {
600 			c4iw_put_resource(&rdev->resource.qid_table,
601 					  entry->qid);
602 			mutex_lock(&rdev->stats.lock);
603 			rdev->stats.qid.cur -= rdev->qpmask + 1;
604 			mutex_unlock(&rdev->stats.lock);
605 		}
606 		kfree(entry);
607 	}
608 
609 	list_for_each_safe(pos, nxt, &uctx->qpids) {
610 		entry = list_entry(pos, struct c4iw_qid_list, entry);
611 		list_del_init(&entry->entry);
612 		kfree(entry);
613 	}
614 	mutex_unlock(&uctx->lock);
615 }
616 
617 void c4iw_init_dev_ucontext(struct c4iw_rdev *rdev,
618 			    struct c4iw_dev_ucontext *uctx)
619 {
620 	INIT_LIST_HEAD(&uctx->qpids);
621 	INIT_LIST_HEAD(&uctx->cqids);
622 	mutex_init(&uctx->lock);
623 }
624 
625 /* Caller takes care of locking if needed */
626 static int c4iw_rdev_open(struct c4iw_rdev *rdev)
627 {
628 	int err;
629 
630 	c4iw_init_dev_ucontext(rdev, &rdev->uctx);
631 
632 	/*
633 	 * qpshift is the number of bits to shift the qpid left in order
634 	 * to get the correct address of the doorbell for that qp.
635 	 */
636 	rdev->qpshift = PAGE_SHIFT - ilog2(rdev->lldi.udb_density);
637 	rdev->qpmask = rdev->lldi.udb_density - 1;
638 	rdev->cqshift = PAGE_SHIFT - ilog2(rdev->lldi.ucq_density);
639 	rdev->cqmask = rdev->lldi.ucq_density - 1;
640 	PDBG("%s dev %s stag start 0x%0x size 0x%0x num stags %d "
641 	     "pbl start 0x%0x size 0x%0x rq start 0x%0x size 0x%0x "
642 	     "qp qid start %u size %u cq qid start %u size %u\n",
643 	     __func__, pci_name(rdev->lldi.pdev), rdev->lldi.vr->stag.start,
644 	     rdev->lldi.vr->stag.size, c4iw_num_stags(rdev),
645 	     rdev->lldi.vr->pbl.start,
646 	     rdev->lldi.vr->pbl.size, rdev->lldi.vr->rq.start,
647 	     rdev->lldi.vr->rq.size,
648 	     rdev->lldi.vr->qp.start,
649 	     rdev->lldi.vr->qp.size,
650 	     rdev->lldi.vr->cq.start,
651 	     rdev->lldi.vr->cq.size);
652 	PDBG("udb len 0x%x udb base %llx db_reg %p gts_reg %p qpshift %lu "
653 	     "qpmask 0x%x cqshift %lu cqmask 0x%x\n",
654 	     (unsigned)pci_resource_len(rdev->lldi.pdev, 2),
655 	     (u64)pci_resource_start(rdev->lldi.pdev, 2),
656 	     rdev->lldi.db_reg,
657 	     rdev->lldi.gts_reg,
658 	     rdev->qpshift, rdev->qpmask,
659 	     rdev->cqshift, rdev->cqmask);
660 
661 	if (c4iw_num_stags(rdev) == 0) {
662 		err = -EINVAL;
663 		goto err1;
664 	}
665 
666 	rdev->stats.pd.total = T4_MAX_NUM_PD;
667 	rdev->stats.stag.total = rdev->lldi.vr->stag.size;
668 	rdev->stats.pbl.total = rdev->lldi.vr->pbl.size;
669 	rdev->stats.rqt.total = rdev->lldi.vr->rq.size;
670 	rdev->stats.ocqp.total = rdev->lldi.vr->ocq.size;
671 	rdev->stats.qid.total = rdev->lldi.vr->qp.size;
672 
673 	err = c4iw_init_resource(rdev, c4iw_num_stags(rdev), T4_MAX_NUM_PD);
674 	if (err) {
675 		printk(KERN_ERR MOD "error %d initializing resources\n", err);
676 		goto err1;
677 	}
678 	err = c4iw_pblpool_create(rdev);
679 	if (err) {
680 		printk(KERN_ERR MOD "error %d initializing pbl pool\n", err);
681 		goto err2;
682 	}
683 	err = c4iw_rqtpool_create(rdev);
684 	if (err) {
685 		printk(KERN_ERR MOD "error %d initializing rqt pool\n", err);
686 		goto err3;
687 	}
688 	err = c4iw_ocqp_pool_create(rdev);
689 	if (err) {
690 		printk(KERN_ERR MOD "error %d initializing ocqp pool\n", err);
691 		goto err4;
692 	}
693 	rdev->status_page = (struct t4_dev_status_page *)
694 			    __get_free_page(GFP_KERNEL);
695 	if (!rdev->status_page) {
696 		pr_err(MOD "error allocating status page\n");
697 		goto err4;
698 	}
699 	return 0;
700 err4:
701 	c4iw_rqtpool_destroy(rdev);
702 err3:
703 	c4iw_pblpool_destroy(rdev);
704 err2:
705 	c4iw_destroy_resource(&rdev->resource);
706 err1:
707 	return err;
708 }
709 
710 static void c4iw_rdev_close(struct c4iw_rdev *rdev)
711 {
712 	free_page((unsigned long)rdev->status_page);
713 	c4iw_pblpool_destroy(rdev);
714 	c4iw_rqtpool_destroy(rdev);
715 	c4iw_destroy_resource(&rdev->resource);
716 }
717 
718 static void c4iw_dealloc(struct uld_ctx *ctx)
719 {
720 	c4iw_rdev_close(&ctx->dev->rdev);
721 	idr_destroy(&ctx->dev->cqidr);
722 	idr_destroy(&ctx->dev->qpidr);
723 	idr_destroy(&ctx->dev->mmidr);
724 	idr_destroy(&ctx->dev->hwtid_idr);
725 	idr_destroy(&ctx->dev->stid_idr);
726 	idr_destroy(&ctx->dev->atid_idr);
727 	if (ctx->dev->rdev.bar2_kva)
728 		iounmap(ctx->dev->rdev.bar2_kva);
729 	if (ctx->dev->rdev.oc_mw_kva)
730 		iounmap(ctx->dev->rdev.oc_mw_kva);
731 	ib_dealloc_device(&ctx->dev->ibdev);
732 	iwpm_exit(RDMA_NL_C4IW);
733 	ctx->dev = NULL;
734 }
735 
736 static void c4iw_remove(struct uld_ctx *ctx)
737 {
738 	PDBG("%s c4iw_dev %p\n", __func__,  ctx->dev);
739 	c4iw_unregister_device(ctx->dev);
740 	c4iw_dealloc(ctx);
741 }
742 
743 static int rdma_supported(const struct cxgb4_lld_info *infop)
744 {
745 	return infop->vr->stag.size > 0 && infop->vr->pbl.size > 0 &&
746 	       infop->vr->rq.size > 0 && infop->vr->qp.size > 0 &&
747 	       infop->vr->cq.size > 0;
748 }
749 
750 static struct c4iw_dev *c4iw_alloc(const struct cxgb4_lld_info *infop)
751 {
752 	struct c4iw_dev *devp;
753 	int ret;
754 
755 	if (!rdma_supported(infop)) {
756 		printk(KERN_INFO MOD "%s: RDMA not supported on this device.\n",
757 		       pci_name(infop->pdev));
758 		return ERR_PTR(-ENOSYS);
759 	}
760 	if (!ocqp_supported(infop))
761 		pr_info("%s: On-Chip Queues not supported on this device.\n",
762 			pci_name(infop->pdev));
763 
764 	devp = (struct c4iw_dev *)ib_alloc_device(sizeof(*devp));
765 	if (!devp) {
766 		printk(KERN_ERR MOD "Cannot allocate ib device\n");
767 		return ERR_PTR(-ENOMEM);
768 	}
769 	devp->rdev.lldi = *infop;
770 
771 	/*
772 	 * For T5 devices, we map all of BAR2 with WC.
773 	 * For T4 devices with onchip qp mem, we map only that part
774 	 * of BAR2 with WC.
775 	 */
776 	devp->rdev.bar2_pa = pci_resource_start(devp->rdev.lldi.pdev, 2);
777 	if (is_t5(devp->rdev.lldi.adapter_type)) {
778 		devp->rdev.bar2_kva = ioremap_wc(devp->rdev.bar2_pa,
779 			pci_resource_len(devp->rdev.lldi.pdev, 2));
780 		if (!devp->rdev.bar2_kva) {
781 			pr_err(MOD "Unable to ioremap BAR2\n");
782 			ib_dealloc_device(&devp->ibdev);
783 			return ERR_PTR(-EINVAL);
784 		}
785 	} else if (ocqp_supported(infop)) {
786 		devp->rdev.oc_mw_pa =
787 			pci_resource_start(devp->rdev.lldi.pdev, 2) +
788 			pci_resource_len(devp->rdev.lldi.pdev, 2) -
789 			roundup_pow_of_two(devp->rdev.lldi.vr->ocq.size);
790 		devp->rdev.oc_mw_kva = ioremap_wc(devp->rdev.oc_mw_pa,
791 			devp->rdev.lldi.vr->ocq.size);
792 		if (!devp->rdev.oc_mw_kva) {
793 			pr_err(MOD "Unable to ioremap onchip mem\n");
794 			ib_dealloc_device(&devp->ibdev);
795 			return ERR_PTR(-EINVAL);
796 		}
797 	}
798 
799 	PDBG(KERN_INFO MOD "ocq memory: "
800 	       "hw_start 0x%x size %u mw_pa 0x%lx mw_kva %p\n",
801 	       devp->rdev.lldi.vr->ocq.start, devp->rdev.lldi.vr->ocq.size,
802 	       devp->rdev.oc_mw_pa, devp->rdev.oc_mw_kva);
803 
804 	ret = c4iw_rdev_open(&devp->rdev);
805 	if (ret) {
806 		printk(KERN_ERR MOD "Unable to open CXIO rdev err %d\n", ret);
807 		ib_dealloc_device(&devp->ibdev);
808 		return ERR_PTR(ret);
809 	}
810 
811 	idr_init(&devp->cqidr);
812 	idr_init(&devp->qpidr);
813 	idr_init(&devp->mmidr);
814 	idr_init(&devp->hwtid_idr);
815 	idr_init(&devp->stid_idr);
816 	idr_init(&devp->atid_idr);
817 	spin_lock_init(&devp->lock);
818 	mutex_init(&devp->rdev.stats.lock);
819 	mutex_init(&devp->db_mutex);
820 	INIT_LIST_HEAD(&devp->db_fc_list);
821 
822 	if (c4iw_debugfs_root) {
823 		devp->debugfs_root = debugfs_create_dir(
824 					pci_name(devp->rdev.lldi.pdev),
825 					c4iw_debugfs_root);
826 		setup_debugfs(devp);
827 	}
828 
829 	ret = iwpm_init(RDMA_NL_C4IW);
830 	if (ret) {
831 		pr_err("port mapper initialization failed with %d\n", ret);
832 		ib_dealloc_device(&devp->ibdev);
833 		return ERR_PTR(ret);
834 	}
835 
836 	return devp;
837 }
838 
839 static void *c4iw_uld_add(const struct cxgb4_lld_info *infop)
840 {
841 	struct uld_ctx *ctx;
842 	static int vers_printed;
843 	int i;
844 
845 	if (!vers_printed++)
846 		pr_info("Chelsio T4/T5 RDMA Driver - version %s\n",
847 			DRV_VERSION);
848 
849 	ctx = kzalloc(sizeof *ctx, GFP_KERNEL);
850 	if (!ctx) {
851 		ctx = ERR_PTR(-ENOMEM);
852 		goto out;
853 	}
854 	ctx->lldi = *infop;
855 
856 	PDBG("%s found device %s nchan %u nrxq %u ntxq %u nports %u\n",
857 	     __func__, pci_name(ctx->lldi.pdev),
858 	     ctx->lldi.nchan, ctx->lldi.nrxq,
859 	     ctx->lldi.ntxq, ctx->lldi.nports);
860 
861 	mutex_lock(&dev_mutex);
862 	list_add_tail(&ctx->entry, &uld_ctx_list);
863 	mutex_unlock(&dev_mutex);
864 
865 	for (i = 0; i < ctx->lldi.nrxq; i++)
866 		PDBG("rxqid[%u] %u\n", i, ctx->lldi.rxq_ids[i]);
867 out:
868 	return ctx;
869 }
870 
871 static inline struct sk_buff *copy_gl_to_skb_pkt(const struct pkt_gl *gl,
872 						 const __be64 *rsp,
873 						 u32 pktshift)
874 {
875 	struct sk_buff *skb;
876 
877 	/*
878 	 * Allocate space for cpl_pass_accept_req which will be synthesized by
879 	 * driver. Once the driver synthesizes the request the skb will go
880 	 * through the regular cpl_pass_accept_req processing.
881 	 * The math here assumes sizeof cpl_pass_accept_req >= sizeof
882 	 * cpl_rx_pkt.
883 	 */
884 	skb = alloc_skb(gl->tot_len + sizeof(struct cpl_pass_accept_req) +
885 			sizeof(struct rss_header) - pktshift, GFP_ATOMIC);
886 	if (unlikely(!skb))
887 		return NULL;
888 
889 	 __skb_put(skb, gl->tot_len + sizeof(struct cpl_pass_accept_req) +
890 		   sizeof(struct rss_header) - pktshift);
891 
892 	/*
893 	 * This skb will contain:
894 	 *   rss_header from the rspq descriptor (1 flit)
895 	 *   cpl_rx_pkt struct from the rspq descriptor (2 flits)
896 	 *   space for the difference between the size of an
897 	 *      rx_pkt and pass_accept_req cpl (1 flit)
898 	 *   the packet data from the gl
899 	 */
900 	skb_copy_to_linear_data(skb, rsp, sizeof(struct cpl_pass_accept_req) +
901 				sizeof(struct rss_header));
902 	skb_copy_to_linear_data_offset(skb, sizeof(struct rss_header) +
903 				       sizeof(struct cpl_pass_accept_req),
904 				       gl->va + pktshift,
905 				       gl->tot_len - pktshift);
906 	return skb;
907 }
908 
909 static inline int recv_rx_pkt(struct c4iw_dev *dev, const struct pkt_gl *gl,
910 			   const __be64 *rsp)
911 {
912 	unsigned int opcode = *(u8 *)rsp;
913 	struct sk_buff *skb;
914 
915 	if (opcode != CPL_RX_PKT)
916 		goto out;
917 
918 	skb = copy_gl_to_skb_pkt(gl , rsp, dev->rdev.lldi.sge_pktshift);
919 	if (skb == NULL)
920 		goto out;
921 
922 	if (c4iw_handlers[opcode] == NULL) {
923 		pr_info("%s no handler opcode 0x%x...\n", __func__,
924 		       opcode);
925 		kfree_skb(skb);
926 		goto out;
927 	}
928 	c4iw_handlers[opcode](dev, skb);
929 	return 1;
930 out:
931 	return 0;
932 }
933 
934 static int c4iw_uld_rx_handler(void *handle, const __be64 *rsp,
935 			const struct pkt_gl *gl)
936 {
937 	struct uld_ctx *ctx = handle;
938 	struct c4iw_dev *dev = ctx->dev;
939 	struct sk_buff *skb;
940 	u8 opcode;
941 
942 	if (gl == NULL) {
943 		/* omit RSS and rsp_ctrl at end of descriptor */
944 		unsigned int len = 64 - sizeof(struct rsp_ctrl) - 8;
945 
946 		skb = alloc_skb(256, GFP_ATOMIC);
947 		if (!skb)
948 			goto nomem;
949 		__skb_put(skb, len);
950 		skb_copy_to_linear_data(skb, &rsp[1], len);
951 	} else if (gl == CXGB4_MSG_AN) {
952 		const struct rsp_ctrl *rc = (void *)rsp;
953 
954 		u32 qid = be32_to_cpu(rc->pldbuflen_qid);
955 		c4iw_ev_handler(dev, qid);
956 		return 0;
957 	} else if (unlikely(*(u8 *)rsp != *(u8 *)gl->va)) {
958 		if (recv_rx_pkt(dev, gl, rsp))
959 			return 0;
960 
961 		pr_info("%s: unexpected FL contents at %p, " \
962 		       "RSS %#llx, FL %#llx, len %u\n",
963 		       pci_name(ctx->lldi.pdev), gl->va,
964 		       (unsigned long long)be64_to_cpu(*rsp),
965 		       (unsigned long long)be64_to_cpu(
966 		       *(__force __be64 *)gl->va),
967 		       gl->tot_len);
968 
969 		return 0;
970 	} else {
971 		skb = cxgb4_pktgl_to_skb(gl, 128, 128);
972 		if (unlikely(!skb))
973 			goto nomem;
974 	}
975 
976 	opcode = *(u8 *)rsp;
977 	if (c4iw_handlers[opcode]) {
978 		c4iw_handlers[opcode](dev, skb);
979 	} else {
980 		pr_info("%s no handler opcode 0x%x...\n", __func__,
981 		       opcode);
982 		kfree_skb(skb);
983 	}
984 
985 	return 0;
986 nomem:
987 	return -1;
988 }
989 
990 static int c4iw_uld_state_change(void *handle, enum cxgb4_state new_state)
991 {
992 	struct uld_ctx *ctx = handle;
993 
994 	PDBG("%s new_state %u\n", __func__, new_state);
995 	switch (new_state) {
996 	case CXGB4_STATE_UP:
997 		printk(KERN_INFO MOD "%s: Up\n", pci_name(ctx->lldi.pdev));
998 		if (!ctx->dev) {
999 			int ret;
1000 
1001 			ctx->dev = c4iw_alloc(&ctx->lldi);
1002 			if (IS_ERR(ctx->dev)) {
1003 				printk(KERN_ERR MOD
1004 				       "%s: initialization failed: %ld\n",
1005 				       pci_name(ctx->lldi.pdev),
1006 				       PTR_ERR(ctx->dev));
1007 				ctx->dev = NULL;
1008 				break;
1009 			}
1010 			ret = c4iw_register_device(ctx->dev);
1011 			if (ret) {
1012 				printk(KERN_ERR MOD
1013 				       "%s: RDMA registration failed: %d\n",
1014 				       pci_name(ctx->lldi.pdev), ret);
1015 				c4iw_dealloc(ctx);
1016 			}
1017 		}
1018 		break;
1019 	case CXGB4_STATE_DOWN:
1020 		printk(KERN_INFO MOD "%s: Down\n",
1021 		       pci_name(ctx->lldi.pdev));
1022 		if (ctx->dev)
1023 			c4iw_remove(ctx);
1024 		break;
1025 	case CXGB4_STATE_START_RECOVERY:
1026 		printk(KERN_INFO MOD "%s: Fatal Error\n",
1027 		       pci_name(ctx->lldi.pdev));
1028 		if (ctx->dev) {
1029 			struct ib_event event;
1030 
1031 			ctx->dev->rdev.flags |= T4_FATAL_ERROR;
1032 			memset(&event, 0, sizeof event);
1033 			event.event  = IB_EVENT_DEVICE_FATAL;
1034 			event.device = &ctx->dev->ibdev;
1035 			ib_dispatch_event(&event);
1036 			c4iw_remove(ctx);
1037 		}
1038 		break;
1039 	case CXGB4_STATE_DETACH:
1040 		printk(KERN_INFO MOD "%s: Detach\n",
1041 		       pci_name(ctx->lldi.pdev));
1042 		if (ctx->dev)
1043 			c4iw_remove(ctx);
1044 		break;
1045 	}
1046 	return 0;
1047 }
1048 
1049 static int disable_qp_db(int id, void *p, void *data)
1050 {
1051 	struct c4iw_qp *qp = p;
1052 
1053 	t4_disable_wq_db(&qp->wq);
1054 	return 0;
1055 }
1056 
1057 static void stop_queues(struct uld_ctx *ctx)
1058 {
1059 	unsigned long flags;
1060 
1061 	spin_lock_irqsave(&ctx->dev->lock, flags);
1062 	ctx->dev->rdev.stats.db_state_transitions++;
1063 	ctx->dev->db_state = STOPPED;
1064 	if (ctx->dev->rdev.flags & T4_STATUS_PAGE_DISABLED)
1065 		idr_for_each(&ctx->dev->qpidr, disable_qp_db, NULL);
1066 	else
1067 		ctx->dev->rdev.status_page->db_off = 1;
1068 	spin_unlock_irqrestore(&ctx->dev->lock, flags);
1069 }
1070 
1071 static int enable_qp_db(int id, void *p, void *data)
1072 {
1073 	struct c4iw_qp *qp = p;
1074 
1075 	t4_enable_wq_db(&qp->wq);
1076 	return 0;
1077 }
1078 
1079 static void resume_rc_qp(struct c4iw_qp *qp)
1080 {
1081 	spin_lock(&qp->lock);
1082 	t4_ring_sq_db(&qp->wq, qp->wq.sq.wq_pidx_inc,
1083 		      is_t5(qp->rhp->rdev.lldi.adapter_type), NULL);
1084 	qp->wq.sq.wq_pidx_inc = 0;
1085 	t4_ring_rq_db(&qp->wq, qp->wq.rq.wq_pidx_inc,
1086 		      is_t5(qp->rhp->rdev.lldi.adapter_type), NULL);
1087 	qp->wq.rq.wq_pidx_inc = 0;
1088 	spin_unlock(&qp->lock);
1089 }
1090 
1091 static void resume_a_chunk(struct uld_ctx *ctx)
1092 {
1093 	int i;
1094 	struct c4iw_qp *qp;
1095 
1096 	for (i = 0; i < DB_FC_RESUME_SIZE; i++) {
1097 		qp = list_first_entry(&ctx->dev->db_fc_list, struct c4iw_qp,
1098 				      db_fc_entry);
1099 		list_del_init(&qp->db_fc_entry);
1100 		resume_rc_qp(qp);
1101 		if (list_empty(&ctx->dev->db_fc_list))
1102 			break;
1103 	}
1104 }
1105 
1106 static void resume_queues(struct uld_ctx *ctx)
1107 {
1108 	spin_lock_irq(&ctx->dev->lock);
1109 	if (ctx->dev->db_state != STOPPED)
1110 		goto out;
1111 	ctx->dev->db_state = FLOW_CONTROL;
1112 	while (1) {
1113 		if (list_empty(&ctx->dev->db_fc_list)) {
1114 			WARN_ON(ctx->dev->db_state != FLOW_CONTROL);
1115 			ctx->dev->db_state = NORMAL;
1116 			ctx->dev->rdev.stats.db_state_transitions++;
1117 			if (ctx->dev->rdev.flags & T4_STATUS_PAGE_DISABLED) {
1118 				idr_for_each(&ctx->dev->qpidr, enable_qp_db,
1119 					     NULL);
1120 			} else {
1121 				ctx->dev->rdev.status_page->db_off = 0;
1122 			}
1123 			break;
1124 		} else {
1125 			if (cxgb4_dbfifo_count(ctx->dev->rdev.lldi.ports[0], 1)
1126 			    < (ctx->dev->rdev.lldi.dbfifo_int_thresh <<
1127 			       DB_FC_DRAIN_THRESH)) {
1128 				resume_a_chunk(ctx);
1129 			}
1130 			if (!list_empty(&ctx->dev->db_fc_list)) {
1131 				spin_unlock_irq(&ctx->dev->lock);
1132 				if (DB_FC_RESUME_DELAY) {
1133 					set_current_state(TASK_UNINTERRUPTIBLE);
1134 					schedule_timeout(DB_FC_RESUME_DELAY);
1135 				}
1136 				spin_lock_irq(&ctx->dev->lock);
1137 				if (ctx->dev->db_state != FLOW_CONTROL)
1138 					break;
1139 			}
1140 		}
1141 	}
1142 out:
1143 	if (ctx->dev->db_state != NORMAL)
1144 		ctx->dev->rdev.stats.db_fc_interruptions++;
1145 	spin_unlock_irq(&ctx->dev->lock);
1146 }
1147 
1148 struct qp_list {
1149 	unsigned idx;
1150 	struct c4iw_qp **qps;
1151 };
1152 
1153 static int add_and_ref_qp(int id, void *p, void *data)
1154 {
1155 	struct qp_list *qp_listp = data;
1156 	struct c4iw_qp *qp = p;
1157 
1158 	c4iw_qp_add_ref(&qp->ibqp);
1159 	qp_listp->qps[qp_listp->idx++] = qp;
1160 	return 0;
1161 }
1162 
1163 static int count_qps(int id, void *p, void *data)
1164 {
1165 	unsigned *countp = data;
1166 	(*countp)++;
1167 	return 0;
1168 }
1169 
1170 static void deref_qps(struct qp_list *qp_list)
1171 {
1172 	int idx;
1173 
1174 	for (idx = 0; idx < qp_list->idx; idx++)
1175 		c4iw_qp_rem_ref(&qp_list->qps[idx]->ibqp);
1176 }
1177 
1178 static void recover_lost_dbs(struct uld_ctx *ctx, struct qp_list *qp_list)
1179 {
1180 	int idx;
1181 	int ret;
1182 
1183 	for (idx = 0; idx < qp_list->idx; idx++) {
1184 		struct c4iw_qp *qp = qp_list->qps[idx];
1185 
1186 		spin_lock_irq(&qp->rhp->lock);
1187 		spin_lock(&qp->lock);
1188 		ret = cxgb4_sync_txq_pidx(qp->rhp->rdev.lldi.ports[0],
1189 					  qp->wq.sq.qid,
1190 					  t4_sq_host_wq_pidx(&qp->wq),
1191 					  t4_sq_wq_size(&qp->wq));
1192 		if (ret) {
1193 			pr_err(KERN_ERR MOD "%s: Fatal error - "
1194 			       "DB overflow recovery failed - "
1195 			       "error syncing SQ qid %u\n",
1196 			       pci_name(ctx->lldi.pdev), qp->wq.sq.qid);
1197 			spin_unlock(&qp->lock);
1198 			spin_unlock_irq(&qp->rhp->lock);
1199 			return;
1200 		}
1201 		qp->wq.sq.wq_pidx_inc = 0;
1202 
1203 		ret = cxgb4_sync_txq_pidx(qp->rhp->rdev.lldi.ports[0],
1204 					  qp->wq.rq.qid,
1205 					  t4_rq_host_wq_pidx(&qp->wq),
1206 					  t4_rq_wq_size(&qp->wq));
1207 
1208 		if (ret) {
1209 			pr_err(KERN_ERR MOD "%s: Fatal error - "
1210 			       "DB overflow recovery failed - "
1211 			       "error syncing RQ qid %u\n",
1212 			       pci_name(ctx->lldi.pdev), qp->wq.rq.qid);
1213 			spin_unlock(&qp->lock);
1214 			spin_unlock_irq(&qp->rhp->lock);
1215 			return;
1216 		}
1217 		qp->wq.rq.wq_pidx_inc = 0;
1218 		spin_unlock(&qp->lock);
1219 		spin_unlock_irq(&qp->rhp->lock);
1220 
1221 		/* Wait for the dbfifo to drain */
1222 		while (cxgb4_dbfifo_count(qp->rhp->rdev.lldi.ports[0], 1) > 0) {
1223 			set_current_state(TASK_UNINTERRUPTIBLE);
1224 			schedule_timeout(usecs_to_jiffies(10));
1225 		}
1226 	}
1227 }
1228 
1229 static void recover_queues(struct uld_ctx *ctx)
1230 {
1231 	int count = 0;
1232 	struct qp_list qp_list;
1233 	int ret;
1234 
1235 	/* slow everybody down */
1236 	set_current_state(TASK_UNINTERRUPTIBLE);
1237 	schedule_timeout(usecs_to_jiffies(1000));
1238 
1239 	/* flush the SGE contexts */
1240 	ret = cxgb4_flush_eq_cache(ctx->dev->rdev.lldi.ports[0]);
1241 	if (ret) {
1242 		printk(KERN_ERR MOD "%s: Fatal error - DB overflow recovery failed\n",
1243 		       pci_name(ctx->lldi.pdev));
1244 		return;
1245 	}
1246 
1247 	/* Count active queues so we can build a list of queues to recover */
1248 	spin_lock_irq(&ctx->dev->lock);
1249 	WARN_ON(ctx->dev->db_state != STOPPED);
1250 	ctx->dev->db_state = RECOVERY;
1251 	idr_for_each(&ctx->dev->qpidr, count_qps, &count);
1252 
1253 	qp_list.qps = kzalloc(count * sizeof *qp_list.qps, GFP_ATOMIC);
1254 	if (!qp_list.qps) {
1255 		printk(KERN_ERR MOD "%s: Fatal error - DB overflow recovery failed\n",
1256 		       pci_name(ctx->lldi.pdev));
1257 		spin_unlock_irq(&ctx->dev->lock);
1258 		return;
1259 	}
1260 	qp_list.idx = 0;
1261 
1262 	/* add and ref each qp so it doesn't get freed */
1263 	idr_for_each(&ctx->dev->qpidr, add_and_ref_qp, &qp_list);
1264 
1265 	spin_unlock_irq(&ctx->dev->lock);
1266 
1267 	/* now traverse the list in a safe context to recover the db state*/
1268 	recover_lost_dbs(ctx, &qp_list);
1269 
1270 	/* we're almost done!  deref the qps and clean up */
1271 	deref_qps(&qp_list);
1272 	kfree(qp_list.qps);
1273 
1274 	spin_lock_irq(&ctx->dev->lock);
1275 	WARN_ON(ctx->dev->db_state != RECOVERY);
1276 	ctx->dev->db_state = STOPPED;
1277 	spin_unlock_irq(&ctx->dev->lock);
1278 }
1279 
1280 static int c4iw_uld_control(void *handle, enum cxgb4_control control, ...)
1281 {
1282 	struct uld_ctx *ctx = handle;
1283 
1284 	switch (control) {
1285 	case CXGB4_CONTROL_DB_FULL:
1286 		stop_queues(ctx);
1287 		ctx->dev->rdev.stats.db_full++;
1288 		break;
1289 	case CXGB4_CONTROL_DB_EMPTY:
1290 		resume_queues(ctx);
1291 		mutex_lock(&ctx->dev->rdev.stats.lock);
1292 		ctx->dev->rdev.stats.db_empty++;
1293 		mutex_unlock(&ctx->dev->rdev.stats.lock);
1294 		break;
1295 	case CXGB4_CONTROL_DB_DROP:
1296 		recover_queues(ctx);
1297 		mutex_lock(&ctx->dev->rdev.stats.lock);
1298 		ctx->dev->rdev.stats.db_drop++;
1299 		mutex_unlock(&ctx->dev->rdev.stats.lock);
1300 		break;
1301 	default:
1302 		printk(KERN_WARNING MOD "%s: unknown control cmd %u\n",
1303 		       pci_name(ctx->lldi.pdev), control);
1304 		break;
1305 	}
1306 	return 0;
1307 }
1308 
1309 static struct cxgb4_uld_info c4iw_uld_info = {
1310 	.name = DRV_NAME,
1311 	.add = c4iw_uld_add,
1312 	.rx_handler = c4iw_uld_rx_handler,
1313 	.state_change = c4iw_uld_state_change,
1314 	.control = c4iw_uld_control,
1315 };
1316 
1317 static int __init c4iw_init_module(void)
1318 {
1319 	int err;
1320 
1321 	err = c4iw_cm_init();
1322 	if (err)
1323 		return err;
1324 
1325 	c4iw_debugfs_root = debugfs_create_dir(DRV_NAME, NULL);
1326 	if (!c4iw_debugfs_root)
1327 		printk(KERN_WARNING MOD
1328 		       "could not create debugfs entry, continuing\n");
1329 
1330 	if (ibnl_add_client(RDMA_NL_C4IW, RDMA_NL_IWPM_NUM_OPS,
1331 			    c4iw_nl_cb_table))
1332 		pr_err("%s[%u]: Failed to add netlink callback\n"
1333 		       , __func__, __LINE__);
1334 
1335 	cxgb4_register_uld(CXGB4_ULD_RDMA, &c4iw_uld_info);
1336 
1337 	return 0;
1338 }
1339 
1340 static void __exit c4iw_exit_module(void)
1341 {
1342 	struct uld_ctx *ctx, *tmp;
1343 
1344 	mutex_lock(&dev_mutex);
1345 	list_for_each_entry_safe(ctx, tmp, &uld_ctx_list, entry) {
1346 		if (ctx->dev)
1347 			c4iw_remove(ctx);
1348 		kfree(ctx);
1349 	}
1350 	mutex_unlock(&dev_mutex);
1351 	cxgb4_unregister_uld(CXGB4_ULD_RDMA);
1352 	ibnl_remove_client(RDMA_NL_C4IW);
1353 	c4iw_cm_term();
1354 	debugfs_remove_recursive(c4iw_debugfs_root);
1355 }
1356 
1357 module_init(c4iw_init_module);
1358 module_exit(c4iw_exit_module);
1359