xref: /linux/drivers/infiniband/hw/cxgb4/device.c (revision 856e7c4b619af622d56b3b454f7bec32a170ac99)
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 #include <linux/math64.h>
37 
38 #include <rdma/ib_verbs.h>
39 
40 #include "iw_cxgb4.h"
41 
42 #define DRV_VERSION "0.1"
43 
44 MODULE_AUTHOR("Steve Wise");
45 MODULE_DESCRIPTION("Chelsio T4/T5 RDMA Driver");
46 MODULE_LICENSE("Dual BSD/GPL");
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 int c4iw_wr_log = 0;
59 module_param(c4iw_wr_log, int, 0444);
60 MODULE_PARM_DESC(c4iw_wr_log, "Enables logging of work request timing data.");
61 
62 static int c4iw_wr_log_size_order = 12;
63 module_param(c4iw_wr_log_size_order, int, 0444);
64 MODULE_PARM_DESC(c4iw_wr_log_size_order,
65 		 "Number of entries (log2) in the work request timing log.");
66 
67 static LIST_HEAD(uld_ctx_list);
68 static DEFINE_MUTEX(dev_mutex);
69 static struct workqueue_struct *reg_workq;
70 
71 #define DB_FC_RESUME_SIZE 64
72 #define DB_FC_RESUME_DELAY 1
73 #define DB_FC_DRAIN_THRESH 0
74 
75 static struct dentry *c4iw_debugfs_root;
76 
77 struct c4iw_debugfs_data {
78 	struct c4iw_dev *devp;
79 	char *buf;
80 	int bufsize;
81 	int pos;
82 };
83 
84 static int count_idrs(int id, void *p, void *data)
85 {
86 	int *countp = data;
87 
88 	*countp = *countp + 1;
89 	return 0;
90 }
91 
92 static ssize_t debugfs_read(struct file *file, char __user *buf, size_t count,
93 			    loff_t *ppos)
94 {
95 	struct c4iw_debugfs_data *d = file->private_data;
96 
97 	return simple_read_from_buffer(buf, count, ppos, d->buf, d->pos);
98 }
99 
100 void c4iw_log_wr_stats(struct t4_wq *wq, struct t4_cqe *cqe)
101 {
102 	struct wr_log_entry le;
103 	int idx;
104 
105 	if (!wq->rdev->wr_log)
106 		return;
107 
108 	idx = (atomic_inc_return(&wq->rdev->wr_log_idx) - 1) &
109 		(wq->rdev->wr_log_size - 1);
110 	le.poll_sge_ts = cxgb4_read_sge_timestamp(wq->rdev->lldi.ports[0]);
111 	le.poll_host_time = ktime_get();
112 	le.valid = 1;
113 	le.cqe_sge_ts = CQE_TS(cqe);
114 	if (SQ_TYPE(cqe)) {
115 		le.qid = wq->sq.qid;
116 		le.opcode = CQE_OPCODE(cqe);
117 		le.post_host_time = wq->sq.sw_sq[wq->sq.cidx].host_time;
118 		le.post_sge_ts = wq->sq.sw_sq[wq->sq.cidx].sge_ts;
119 		le.wr_id = CQE_WRID_SQ_IDX(cqe);
120 	} else {
121 		le.qid = wq->rq.qid;
122 		le.opcode = FW_RI_RECEIVE;
123 		le.post_host_time = wq->rq.sw_rq[wq->rq.cidx].host_time;
124 		le.post_sge_ts = wq->rq.sw_rq[wq->rq.cidx].sge_ts;
125 		le.wr_id = CQE_WRID_MSN(cqe);
126 	}
127 	wq->rdev->wr_log[idx] = le;
128 }
129 
130 static int wr_log_show(struct seq_file *seq, void *v)
131 {
132 	struct c4iw_dev *dev = seq->private;
133 	ktime_t prev_time;
134 	struct wr_log_entry *lep;
135 	int prev_time_set = 0;
136 	int idx, end;
137 
138 #define ts2ns(ts) div64_u64((ts) * dev->rdev.lldi.cclk_ps, 1000)
139 
140 	idx = atomic_read(&dev->rdev.wr_log_idx) &
141 		(dev->rdev.wr_log_size - 1);
142 	end = idx - 1;
143 	if (end < 0)
144 		end = dev->rdev.wr_log_size - 1;
145 	lep = &dev->rdev.wr_log[idx];
146 	while (idx != end) {
147 		if (lep->valid) {
148 			if (!prev_time_set) {
149 				prev_time_set = 1;
150 				prev_time = lep->poll_host_time;
151 			}
152 			seq_printf(seq, "%04u: nsec %llu qid %u opcode "
153 				   "%u %s 0x%x host_wr_delta nsec %llu "
154 				   "post_sge_ts 0x%llx cqe_sge_ts 0x%llx "
155 				   "poll_sge_ts 0x%llx post_poll_delta_ns %llu "
156 				   "cqe_poll_delta_ns %llu\n",
157 				   idx,
158 				   ktime_to_ns(ktime_sub(lep->poll_host_time,
159 							 prev_time)),
160 				   lep->qid, lep->opcode,
161 				   lep->opcode == FW_RI_RECEIVE ?
162 							"msn" : "wrid",
163 				   lep->wr_id,
164 				   ktime_to_ns(ktime_sub(lep->poll_host_time,
165 							 lep->post_host_time)),
166 				   lep->post_sge_ts, lep->cqe_sge_ts,
167 				   lep->poll_sge_ts,
168 				   ts2ns(lep->poll_sge_ts - lep->post_sge_ts),
169 				   ts2ns(lep->poll_sge_ts - lep->cqe_sge_ts));
170 			prev_time = lep->poll_host_time;
171 		}
172 		idx++;
173 		if (idx > (dev->rdev.wr_log_size - 1))
174 			idx = 0;
175 		lep = &dev->rdev.wr_log[idx];
176 	}
177 #undef ts2ns
178 	return 0;
179 }
180 
181 static int wr_log_open(struct inode *inode, struct file *file)
182 {
183 	return single_open(file, wr_log_show, inode->i_private);
184 }
185 
186 static ssize_t wr_log_clear(struct file *file, const char __user *buf,
187 			    size_t count, loff_t *pos)
188 {
189 	struct c4iw_dev *dev = ((struct seq_file *)file->private_data)->private;
190 	int i;
191 
192 	if (dev->rdev.wr_log)
193 		for (i = 0; i < dev->rdev.wr_log_size; i++)
194 			dev->rdev.wr_log[i].valid = 0;
195 	return count;
196 }
197 
198 static const struct file_operations wr_log_debugfs_fops = {
199 	.owner   = THIS_MODULE,
200 	.open    = wr_log_open,
201 	.release = single_release,
202 	.read    = seq_read,
203 	.llseek  = seq_lseek,
204 	.write   = wr_log_clear,
205 };
206 
207 static struct sockaddr_in zero_sin = {
208 	.sin_family = AF_INET,
209 };
210 
211 static struct sockaddr_in6 zero_sin6 = {
212 	.sin6_family = AF_INET6,
213 };
214 
215 static void set_ep_sin_addrs(struct c4iw_ep *ep,
216 			     struct sockaddr_in **lsin,
217 			     struct sockaddr_in **rsin,
218 			     struct sockaddr_in **m_lsin,
219 			     struct sockaddr_in **m_rsin)
220 {
221 	struct iw_cm_id *id = ep->com.cm_id;
222 
223 	*m_lsin = (struct sockaddr_in *)&ep->com.local_addr;
224 	*m_rsin = (struct sockaddr_in *)&ep->com.remote_addr;
225 	if (id) {
226 		*lsin = (struct sockaddr_in *)&id->local_addr;
227 		*rsin = (struct sockaddr_in *)&id->remote_addr;
228 	} else {
229 		*lsin = &zero_sin;
230 		*rsin = &zero_sin;
231 	}
232 }
233 
234 static void set_ep_sin6_addrs(struct c4iw_ep *ep,
235 			      struct sockaddr_in6 **lsin6,
236 			      struct sockaddr_in6 **rsin6,
237 			      struct sockaddr_in6 **m_lsin6,
238 			      struct sockaddr_in6 **m_rsin6)
239 {
240 	struct iw_cm_id *id = ep->com.cm_id;
241 
242 	*m_lsin6 = (struct sockaddr_in6 *)&ep->com.local_addr;
243 	*m_rsin6 = (struct sockaddr_in6 *)&ep->com.remote_addr;
244 	if (id) {
245 		*lsin6 = (struct sockaddr_in6 *)&id->local_addr;
246 		*rsin6 = (struct sockaddr_in6 *)&id->remote_addr;
247 	} else {
248 		*lsin6 = &zero_sin6;
249 		*rsin6 = &zero_sin6;
250 	}
251 }
252 
253 static int dump_qp(int id, void *p, void *data)
254 {
255 	struct c4iw_qp *qp = p;
256 	struct c4iw_debugfs_data *qpd = data;
257 	int space;
258 	int cc;
259 
260 	if (id != qp->wq.sq.qid)
261 		return 0;
262 
263 	space = qpd->bufsize - qpd->pos - 1;
264 	if (space == 0)
265 		return 1;
266 
267 	if (qp->ep) {
268 		struct c4iw_ep *ep = qp->ep;
269 
270 		if (ep->com.local_addr.ss_family == AF_INET) {
271 			struct sockaddr_in *lsin;
272 			struct sockaddr_in *rsin;
273 			struct sockaddr_in *m_lsin;
274 			struct sockaddr_in *m_rsin;
275 
276 			set_ep_sin_addrs(ep, &lsin, &rsin, &m_lsin, &m_rsin);
277 			cc = snprintf(qpd->buf + qpd->pos, space,
278 				      "rc qp sq id %u rq id %u state %u "
279 				      "onchip %u ep tid %u state %u "
280 				      "%pI4:%u/%u->%pI4:%u/%u\n",
281 				      qp->wq.sq.qid, qp->wq.rq.qid,
282 				      (int)qp->attr.state,
283 				      qp->wq.sq.flags & T4_SQ_ONCHIP,
284 				      ep->hwtid, (int)ep->com.state,
285 				      &lsin->sin_addr, ntohs(lsin->sin_port),
286 				      ntohs(m_lsin->sin_port),
287 				      &rsin->sin_addr, ntohs(rsin->sin_port),
288 				      ntohs(m_rsin->sin_port));
289 		} else {
290 			struct sockaddr_in6 *lsin6;
291 			struct sockaddr_in6 *rsin6;
292 			struct sockaddr_in6 *m_lsin6;
293 			struct sockaddr_in6 *m_rsin6;
294 
295 			set_ep_sin6_addrs(ep, &lsin6, &rsin6, &m_lsin6,
296 					  &m_rsin6);
297 			cc = snprintf(qpd->buf + qpd->pos, space,
298 				      "rc qp sq id %u rq id %u state %u "
299 				      "onchip %u ep tid %u state %u "
300 				      "%pI6:%u/%u->%pI6:%u/%u\n",
301 				      qp->wq.sq.qid, qp->wq.rq.qid,
302 				      (int)qp->attr.state,
303 				      qp->wq.sq.flags & T4_SQ_ONCHIP,
304 				      ep->hwtid, (int)ep->com.state,
305 				      &lsin6->sin6_addr,
306 				      ntohs(lsin6->sin6_port),
307 				      ntohs(m_lsin6->sin6_port),
308 				      &rsin6->sin6_addr,
309 				      ntohs(rsin6->sin6_port),
310 				      ntohs(m_rsin6->sin6_port));
311 		}
312 	} else
313 		cc = snprintf(qpd->buf + qpd->pos, space,
314 			     "qp sq id %u rq id %u state %u onchip %u\n",
315 			      qp->wq.sq.qid, qp->wq.rq.qid,
316 			      (int)qp->attr.state,
317 			      qp->wq.sq.flags & T4_SQ_ONCHIP);
318 	if (cc < space)
319 		qpd->pos += cc;
320 	return 0;
321 }
322 
323 static int qp_release(struct inode *inode, struct file *file)
324 {
325 	struct c4iw_debugfs_data *qpd = file->private_data;
326 	if (!qpd) {
327 		pr_info("%s null qpd?\n", __func__);
328 		return 0;
329 	}
330 	vfree(qpd->buf);
331 	kfree(qpd);
332 	return 0;
333 }
334 
335 static int qp_open(struct inode *inode, struct file *file)
336 {
337 	struct c4iw_debugfs_data *qpd;
338 	int count = 1;
339 
340 	qpd = kmalloc(sizeof *qpd, GFP_KERNEL);
341 	if (!qpd)
342 		return -ENOMEM;
343 
344 	qpd->devp = inode->i_private;
345 	qpd->pos = 0;
346 
347 	spin_lock_irq(&qpd->devp->lock);
348 	idr_for_each(&qpd->devp->qpidr, count_idrs, &count);
349 	spin_unlock_irq(&qpd->devp->lock);
350 
351 	qpd->bufsize = count * 180;
352 	qpd->buf = vmalloc(qpd->bufsize);
353 	if (!qpd->buf) {
354 		kfree(qpd);
355 		return -ENOMEM;
356 	}
357 
358 	spin_lock_irq(&qpd->devp->lock);
359 	idr_for_each(&qpd->devp->qpidr, dump_qp, qpd);
360 	spin_unlock_irq(&qpd->devp->lock);
361 
362 	qpd->buf[qpd->pos++] = 0;
363 	file->private_data = qpd;
364 	return 0;
365 }
366 
367 static const struct file_operations qp_debugfs_fops = {
368 	.owner   = THIS_MODULE,
369 	.open    = qp_open,
370 	.release = qp_release,
371 	.read    = debugfs_read,
372 	.llseek  = default_llseek,
373 };
374 
375 static int dump_stag(int id, void *p, void *data)
376 {
377 	struct c4iw_debugfs_data *stagd = data;
378 	int space;
379 	int cc;
380 	struct fw_ri_tpte tpte;
381 	int ret;
382 
383 	space = stagd->bufsize - stagd->pos - 1;
384 	if (space == 0)
385 		return 1;
386 
387 	ret = cxgb4_read_tpte(stagd->devp->rdev.lldi.ports[0], (u32)id<<8,
388 			      (__be32 *)&tpte);
389 	if (ret) {
390 		dev_err(&stagd->devp->rdev.lldi.pdev->dev,
391 			"%s cxgb4_read_tpte err %d\n", __func__, ret);
392 		return ret;
393 	}
394 	cc = snprintf(stagd->buf + stagd->pos, space,
395 		      "stag: idx 0x%x valid %d key 0x%x state %d pdid %d "
396 		      "perm 0x%x ps %d len 0x%llx va 0x%llx\n",
397 		      (u32)id<<8,
398 		      FW_RI_TPTE_VALID_G(ntohl(tpte.valid_to_pdid)),
399 		      FW_RI_TPTE_STAGKEY_G(ntohl(tpte.valid_to_pdid)),
400 		      FW_RI_TPTE_STAGSTATE_G(ntohl(tpte.valid_to_pdid)),
401 		      FW_RI_TPTE_PDID_G(ntohl(tpte.valid_to_pdid)),
402 		      FW_RI_TPTE_PERM_G(ntohl(tpte.locread_to_qpid)),
403 		      FW_RI_TPTE_PS_G(ntohl(tpte.locread_to_qpid)),
404 		      ((u64)ntohl(tpte.len_hi) << 32) | ntohl(tpte.len_lo),
405 		      ((u64)ntohl(tpte.va_hi) << 32) | ntohl(tpte.va_lo_fbo));
406 	if (cc < space)
407 		stagd->pos += cc;
408 	return 0;
409 }
410 
411 static int stag_release(struct inode *inode, struct file *file)
412 {
413 	struct c4iw_debugfs_data *stagd = file->private_data;
414 	if (!stagd) {
415 		pr_info("%s null stagd?\n", __func__);
416 		return 0;
417 	}
418 	vfree(stagd->buf);
419 	kfree(stagd);
420 	return 0;
421 }
422 
423 static int stag_open(struct inode *inode, struct file *file)
424 {
425 	struct c4iw_debugfs_data *stagd;
426 	int ret = 0;
427 	int count = 1;
428 
429 	stagd = kmalloc(sizeof *stagd, GFP_KERNEL);
430 	if (!stagd) {
431 		ret = -ENOMEM;
432 		goto out;
433 	}
434 	stagd->devp = inode->i_private;
435 	stagd->pos = 0;
436 
437 	spin_lock_irq(&stagd->devp->lock);
438 	idr_for_each(&stagd->devp->mmidr, count_idrs, &count);
439 	spin_unlock_irq(&stagd->devp->lock);
440 
441 	stagd->bufsize = count * 256;
442 	stagd->buf = vmalloc(stagd->bufsize);
443 	if (!stagd->buf) {
444 		ret = -ENOMEM;
445 		goto err1;
446 	}
447 
448 	spin_lock_irq(&stagd->devp->lock);
449 	idr_for_each(&stagd->devp->mmidr, dump_stag, stagd);
450 	spin_unlock_irq(&stagd->devp->lock);
451 
452 	stagd->buf[stagd->pos++] = 0;
453 	file->private_data = stagd;
454 	goto out;
455 err1:
456 	kfree(stagd);
457 out:
458 	return ret;
459 }
460 
461 static const struct file_operations stag_debugfs_fops = {
462 	.owner   = THIS_MODULE,
463 	.open    = stag_open,
464 	.release = stag_release,
465 	.read    = debugfs_read,
466 	.llseek  = default_llseek,
467 };
468 
469 static char *db_state_str[] = {"NORMAL", "FLOW_CONTROL", "RECOVERY", "STOPPED"};
470 
471 static int stats_show(struct seq_file *seq, void *v)
472 {
473 	struct c4iw_dev *dev = seq->private;
474 
475 	seq_printf(seq, "   Object: %10s %10s %10s %10s\n", "Total", "Current",
476 		   "Max", "Fail");
477 	seq_printf(seq, "     PDID: %10llu %10llu %10llu %10llu\n",
478 			dev->rdev.stats.pd.total, dev->rdev.stats.pd.cur,
479 			dev->rdev.stats.pd.max, dev->rdev.stats.pd.fail);
480 	seq_printf(seq, "      QID: %10llu %10llu %10llu %10llu\n",
481 			dev->rdev.stats.qid.total, dev->rdev.stats.qid.cur,
482 			dev->rdev.stats.qid.max, dev->rdev.stats.qid.fail);
483 	seq_printf(seq, "   TPTMEM: %10llu %10llu %10llu %10llu\n",
484 			dev->rdev.stats.stag.total, dev->rdev.stats.stag.cur,
485 			dev->rdev.stats.stag.max, dev->rdev.stats.stag.fail);
486 	seq_printf(seq, "   PBLMEM: %10llu %10llu %10llu %10llu\n",
487 			dev->rdev.stats.pbl.total, dev->rdev.stats.pbl.cur,
488 			dev->rdev.stats.pbl.max, dev->rdev.stats.pbl.fail);
489 	seq_printf(seq, "   RQTMEM: %10llu %10llu %10llu %10llu\n",
490 			dev->rdev.stats.rqt.total, dev->rdev.stats.rqt.cur,
491 			dev->rdev.stats.rqt.max, dev->rdev.stats.rqt.fail);
492 	seq_printf(seq, "  OCQPMEM: %10llu %10llu %10llu %10llu\n",
493 			dev->rdev.stats.ocqp.total, dev->rdev.stats.ocqp.cur,
494 			dev->rdev.stats.ocqp.max, dev->rdev.stats.ocqp.fail);
495 	seq_printf(seq, "  DB FULL: %10llu\n", dev->rdev.stats.db_full);
496 	seq_printf(seq, " DB EMPTY: %10llu\n", dev->rdev.stats.db_empty);
497 	seq_printf(seq, "  DB DROP: %10llu\n", dev->rdev.stats.db_drop);
498 	seq_printf(seq, " DB State: %s Transitions %llu FC Interruptions %llu\n",
499 		   db_state_str[dev->db_state],
500 		   dev->rdev.stats.db_state_transitions,
501 		   dev->rdev.stats.db_fc_interruptions);
502 	seq_printf(seq, "TCAM_FULL: %10llu\n", dev->rdev.stats.tcam_full);
503 	seq_printf(seq, "ACT_OFLD_CONN_FAILS: %10llu\n",
504 		   dev->rdev.stats.act_ofld_conn_fails);
505 	seq_printf(seq, "PAS_OFLD_CONN_FAILS: %10llu\n",
506 		   dev->rdev.stats.pas_ofld_conn_fails);
507 	seq_printf(seq, "NEG_ADV_RCVD: %10llu\n", dev->rdev.stats.neg_adv);
508 	seq_printf(seq, "AVAILABLE IRD: %10u\n", dev->avail_ird);
509 	return 0;
510 }
511 
512 static int stats_open(struct inode *inode, struct file *file)
513 {
514 	return single_open(file, stats_show, inode->i_private);
515 }
516 
517 static ssize_t stats_clear(struct file *file, const char __user *buf,
518 		size_t count, loff_t *pos)
519 {
520 	struct c4iw_dev *dev = ((struct seq_file *)file->private_data)->private;
521 
522 	mutex_lock(&dev->rdev.stats.lock);
523 	dev->rdev.stats.pd.max = 0;
524 	dev->rdev.stats.pd.fail = 0;
525 	dev->rdev.stats.qid.max = 0;
526 	dev->rdev.stats.qid.fail = 0;
527 	dev->rdev.stats.stag.max = 0;
528 	dev->rdev.stats.stag.fail = 0;
529 	dev->rdev.stats.pbl.max = 0;
530 	dev->rdev.stats.pbl.fail = 0;
531 	dev->rdev.stats.rqt.max = 0;
532 	dev->rdev.stats.rqt.fail = 0;
533 	dev->rdev.stats.ocqp.max = 0;
534 	dev->rdev.stats.ocqp.fail = 0;
535 	dev->rdev.stats.db_full = 0;
536 	dev->rdev.stats.db_empty = 0;
537 	dev->rdev.stats.db_drop = 0;
538 	dev->rdev.stats.db_state_transitions = 0;
539 	dev->rdev.stats.tcam_full = 0;
540 	dev->rdev.stats.act_ofld_conn_fails = 0;
541 	dev->rdev.stats.pas_ofld_conn_fails = 0;
542 	mutex_unlock(&dev->rdev.stats.lock);
543 	return count;
544 }
545 
546 static const struct file_operations stats_debugfs_fops = {
547 	.owner   = THIS_MODULE,
548 	.open    = stats_open,
549 	.release = single_release,
550 	.read    = seq_read,
551 	.llseek  = seq_lseek,
552 	.write   = stats_clear,
553 };
554 
555 static int dump_ep(int id, void *p, void *data)
556 {
557 	struct c4iw_ep *ep = p;
558 	struct c4iw_debugfs_data *epd = data;
559 	int space;
560 	int cc;
561 
562 	space = epd->bufsize - epd->pos - 1;
563 	if (space == 0)
564 		return 1;
565 
566 	if (ep->com.local_addr.ss_family == AF_INET) {
567 		struct sockaddr_in *lsin;
568 		struct sockaddr_in *rsin;
569 		struct sockaddr_in *m_lsin;
570 		struct sockaddr_in *m_rsin;
571 
572 		set_ep_sin_addrs(ep, &lsin, &rsin, &m_lsin, &m_rsin);
573 		cc = snprintf(epd->buf + epd->pos, space,
574 			      "ep %p cm_id %p qp %p state %d flags 0x%lx "
575 			      "history 0x%lx hwtid %d atid %d "
576 			      "conn_na %u abort_na %u "
577 			      "%pI4:%d/%d <-> %pI4:%d/%d\n",
578 			      ep, ep->com.cm_id, ep->com.qp,
579 			      (int)ep->com.state, ep->com.flags,
580 			      ep->com.history, ep->hwtid, ep->atid,
581 			      ep->stats.connect_neg_adv,
582 			      ep->stats.abort_neg_adv,
583 			      &lsin->sin_addr, ntohs(lsin->sin_port),
584 			      ntohs(m_lsin->sin_port),
585 			      &rsin->sin_addr, ntohs(rsin->sin_port),
586 			      ntohs(m_rsin->sin_port));
587 	} else {
588 		struct sockaddr_in6 *lsin6;
589 		struct sockaddr_in6 *rsin6;
590 		struct sockaddr_in6 *m_lsin6;
591 		struct sockaddr_in6 *m_rsin6;
592 
593 		set_ep_sin6_addrs(ep, &lsin6, &rsin6, &m_lsin6, &m_rsin6);
594 		cc = snprintf(epd->buf + epd->pos, space,
595 			      "ep %p cm_id %p qp %p state %d flags 0x%lx "
596 			      "history 0x%lx hwtid %d atid %d "
597 			      "conn_na %u abort_na %u "
598 			      "%pI6:%d/%d <-> %pI6:%d/%d\n",
599 			      ep, ep->com.cm_id, ep->com.qp,
600 			      (int)ep->com.state, ep->com.flags,
601 			      ep->com.history, ep->hwtid, ep->atid,
602 			      ep->stats.connect_neg_adv,
603 			      ep->stats.abort_neg_adv,
604 			      &lsin6->sin6_addr, ntohs(lsin6->sin6_port),
605 			      ntohs(m_lsin6->sin6_port),
606 			      &rsin6->sin6_addr, ntohs(rsin6->sin6_port),
607 			      ntohs(m_rsin6->sin6_port));
608 	}
609 	if (cc < space)
610 		epd->pos += cc;
611 	return 0;
612 }
613 
614 static int dump_listen_ep(int id, void *p, void *data)
615 {
616 	struct c4iw_listen_ep *ep = p;
617 	struct c4iw_debugfs_data *epd = data;
618 	int space;
619 	int cc;
620 
621 	space = epd->bufsize - epd->pos - 1;
622 	if (space == 0)
623 		return 1;
624 
625 	if (ep->com.local_addr.ss_family == AF_INET) {
626 		struct sockaddr_in *lsin = (struct sockaddr_in *)
627 			&ep->com.cm_id->local_addr;
628 		struct sockaddr_in *m_lsin = (struct sockaddr_in *)
629 			&ep->com.cm_id->m_local_addr;
630 
631 		cc = snprintf(epd->buf + epd->pos, space,
632 			      "ep %p cm_id %p state %d flags 0x%lx stid %d "
633 			      "backlog %d %pI4:%d/%d\n",
634 			      ep, ep->com.cm_id, (int)ep->com.state,
635 			      ep->com.flags, ep->stid, ep->backlog,
636 			      &lsin->sin_addr, ntohs(lsin->sin_port),
637 			      ntohs(m_lsin->sin_port));
638 	} else {
639 		struct sockaddr_in6 *lsin6 = (struct sockaddr_in6 *)
640 			&ep->com.cm_id->local_addr;
641 		struct sockaddr_in6 *m_lsin6 = (struct sockaddr_in6 *)
642 			&ep->com.cm_id->m_local_addr;
643 
644 		cc = snprintf(epd->buf + epd->pos, space,
645 			      "ep %p cm_id %p state %d flags 0x%lx stid %d "
646 			      "backlog %d %pI6:%d/%d\n",
647 			      ep, ep->com.cm_id, (int)ep->com.state,
648 			      ep->com.flags, ep->stid, ep->backlog,
649 			      &lsin6->sin6_addr, ntohs(lsin6->sin6_port),
650 			      ntohs(m_lsin6->sin6_port));
651 	}
652 	if (cc < space)
653 		epd->pos += cc;
654 	return 0;
655 }
656 
657 static int ep_release(struct inode *inode, struct file *file)
658 {
659 	struct c4iw_debugfs_data *epd = file->private_data;
660 	if (!epd) {
661 		pr_info("%s null qpd?\n", __func__);
662 		return 0;
663 	}
664 	vfree(epd->buf);
665 	kfree(epd);
666 	return 0;
667 }
668 
669 static int ep_open(struct inode *inode, struct file *file)
670 {
671 	struct c4iw_debugfs_data *epd;
672 	int ret = 0;
673 	int count = 1;
674 
675 	epd = kmalloc(sizeof(*epd), GFP_KERNEL);
676 	if (!epd) {
677 		ret = -ENOMEM;
678 		goto out;
679 	}
680 	epd->devp = inode->i_private;
681 	epd->pos = 0;
682 
683 	spin_lock_irq(&epd->devp->lock);
684 	idr_for_each(&epd->devp->hwtid_idr, count_idrs, &count);
685 	idr_for_each(&epd->devp->atid_idr, count_idrs, &count);
686 	idr_for_each(&epd->devp->stid_idr, count_idrs, &count);
687 	spin_unlock_irq(&epd->devp->lock);
688 
689 	epd->bufsize = count * 240;
690 	epd->buf = vmalloc(epd->bufsize);
691 	if (!epd->buf) {
692 		ret = -ENOMEM;
693 		goto err1;
694 	}
695 
696 	spin_lock_irq(&epd->devp->lock);
697 	idr_for_each(&epd->devp->hwtid_idr, dump_ep, epd);
698 	idr_for_each(&epd->devp->atid_idr, dump_ep, epd);
699 	idr_for_each(&epd->devp->stid_idr, dump_listen_ep, epd);
700 	spin_unlock_irq(&epd->devp->lock);
701 
702 	file->private_data = epd;
703 	goto out;
704 err1:
705 	kfree(epd);
706 out:
707 	return ret;
708 }
709 
710 static const struct file_operations ep_debugfs_fops = {
711 	.owner   = THIS_MODULE,
712 	.open    = ep_open,
713 	.release = ep_release,
714 	.read    = debugfs_read,
715 };
716 
717 static int setup_debugfs(struct c4iw_dev *devp)
718 {
719 	if (!devp->debugfs_root)
720 		return -1;
721 
722 	debugfs_create_file_size("qps", S_IWUSR, devp->debugfs_root,
723 				 (void *)devp, &qp_debugfs_fops, 4096);
724 
725 	debugfs_create_file_size("stags", S_IWUSR, devp->debugfs_root,
726 				 (void *)devp, &stag_debugfs_fops, 4096);
727 
728 	debugfs_create_file_size("stats", S_IWUSR, devp->debugfs_root,
729 				 (void *)devp, &stats_debugfs_fops, 4096);
730 
731 	debugfs_create_file_size("eps", S_IWUSR, devp->debugfs_root,
732 				 (void *)devp, &ep_debugfs_fops, 4096);
733 
734 	if (c4iw_wr_log)
735 		debugfs_create_file_size("wr_log", S_IWUSR, devp->debugfs_root,
736 					 (void *)devp, &wr_log_debugfs_fops, 4096);
737 	return 0;
738 }
739 
740 void c4iw_release_dev_ucontext(struct c4iw_rdev *rdev,
741 			       struct c4iw_dev_ucontext *uctx)
742 {
743 	struct list_head *pos, *nxt;
744 	struct c4iw_qid_list *entry;
745 
746 	mutex_lock(&uctx->lock);
747 	list_for_each_safe(pos, nxt, &uctx->qpids) {
748 		entry = list_entry(pos, struct c4iw_qid_list, entry);
749 		list_del_init(&entry->entry);
750 		if (!(entry->qid & rdev->qpmask)) {
751 			c4iw_put_resource(&rdev->resource.qid_table,
752 					  entry->qid);
753 			mutex_lock(&rdev->stats.lock);
754 			rdev->stats.qid.cur -= rdev->qpmask + 1;
755 			mutex_unlock(&rdev->stats.lock);
756 		}
757 		kfree(entry);
758 	}
759 
760 	list_for_each_safe(pos, nxt, &uctx->cqids) {
761 		entry = list_entry(pos, struct c4iw_qid_list, entry);
762 		list_del_init(&entry->entry);
763 		kfree(entry);
764 	}
765 	mutex_unlock(&uctx->lock);
766 }
767 
768 void c4iw_init_dev_ucontext(struct c4iw_rdev *rdev,
769 			    struct c4iw_dev_ucontext *uctx)
770 {
771 	INIT_LIST_HEAD(&uctx->qpids);
772 	INIT_LIST_HEAD(&uctx->cqids);
773 	mutex_init(&uctx->lock);
774 }
775 
776 /* Caller takes care of locking if needed */
777 static int c4iw_rdev_open(struct c4iw_rdev *rdev)
778 {
779 	int err;
780 
781 	c4iw_init_dev_ucontext(rdev, &rdev->uctx);
782 
783 	/*
784 	 * This implementation assumes udb_density == ucq_density!  Eventually
785 	 * we might need to support this but for now fail the open. Also the
786 	 * cqid and qpid range must match for now.
787 	 */
788 	if (rdev->lldi.udb_density != rdev->lldi.ucq_density) {
789 		pr_err("%s: unsupported udb/ucq densities %u/%u\n",
790 		       pci_name(rdev->lldi.pdev), rdev->lldi.udb_density,
791 		       rdev->lldi.ucq_density);
792 		return -EINVAL;
793 	}
794 	if (rdev->lldi.vr->qp.start != rdev->lldi.vr->cq.start ||
795 	    rdev->lldi.vr->qp.size != rdev->lldi.vr->cq.size) {
796 		pr_err("%s: unsupported qp and cq id ranges qp start %u size %u cq start %u size %u\n",
797 		       pci_name(rdev->lldi.pdev), rdev->lldi.vr->qp.start,
798 		       rdev->lldi.vr->qp.size, rdev->lldi.vr->cq.size,
799 		       rdev->lldi.vr->cq.size);
800 		return -EINVAL;
801 	}
802 
803 	rdev->qpmask = rdev->lldi.udb_density - 1;
804 	rdev->cqmask = rdev->lldi.ucq_density - 1;
805 	pr_debug("dev %s stag start 0x%0x size 0x%0x num stags %d pbl start 0x%0x size 0x%0x rq start 0x%0x size 0x%0x qp qid start %u size %u cq qid start %u size %u\n",
806 		 pci_name(rdev->lldi.pdev), rdev->lldi.vr->stag.start,
807 		 rdev->lldi.vr->stag.size, c4iw_num_stags(rdev),
808 		 rdev->lldi.vr->pbl.start,
809 		 rdev->lldi.vr->pbl.size, rdev->lldi.vr->rq.start,
810 		 rdev->lldi.vr->rq.size,
811 		 rdev->lldi.vr->qp.start,
812 		 rdev->lldi.vr->qp.size,
813 		 rdev->lldi.vr->cq.start,
814 		 rdev->lldi.vr->cq.size);
815 	pr_debug("udb %pR db_reg %p gts_reg %p qpmask 0x%x cqmask 0x%x\n",
816 		 &rdev->lldi.pdev->resource[2],
817 		 rdev->lldi.db_reg, rdev->lldi.gts_reg,
818 		 rdev->qpmask, rdev->cqmask);
819 
820 	if (c4iw_num_stags(rdev) == 0)
821 		return -EINVAL;
822 
823 	rdev->stats.pd.total = T4_MAX_NUM_PD;
824 	rdev->stats.stag.total = rdev->lldi.vr->stag.size;
825 	rdev->stats.pbl.total = rdev->lldi.vr->pbl.size;
826 	rdev->stats.rqt.total = rdev->lldi.vr->rq.size;
827 	rdev->stats.ocqp.total = rdev->lldi.vr->ocq.size;
828 	rdev->stats.qid.total = rdev->lldi.vr->qp.size;
829 
830 	err = c4iw_init_resource(rdev, c4iw_num_stags(rdev), T4_MAX_NUM_PD);
831 	if (err) {
832 		pr_err("error %d initializing resources\n", err);
833 		return err;
834 	}
835 	err = c4iw_pblpool_create(rdev);
836 	if (err) {
837 		pr_err("error %d initializing pbl pool\n", err);
838 		goto destroy_resource;
839 	}
840 	err = c4iw_rqtpool_create(rdev);
841 	if (err) {
842 		pr_err("error %d initializing rqt pool\n", err);
843 		goto destroy_pblpool;
844 	}
845 	err = c4iw_ocqp_pool_create(rdev);
846 	if (err) {
847 		pr_err("error %d initializing ocqp pool\n", err);
848 		goto destroy_rqtpool;
849 	}
850 	rdev->status_page = (struct t4_dev_status_page *)
851 			    __get_free_page(GFP_KERNEL);
852 	if (!rdev->status_page) {
853 		err = -ENOMEM;
854 		goto destroy_ocqp_pool;
855 	}
856 	rdev->status_page->qp_start = rdev->lldi.vr->qp.start;
857 	rdev->status_page->qp_size = rdev->lldi.vr->qp.size;
858 	rdev->status_page->cq_start = rdev->lldi.vr->cq.start;
859 	rdev->status_page->cq_size = rdev->lldi.vr->cq.size;
860 
861 	if (c4iw_wr_log) {
862 		rdev->wr_log = kcalloc(1 << c4iw_wr_log_size_order,
863 				       sizeof(*rdev->wr_log),
864 				       GFP_KERNEL);
865 		if (rdev->wr_log) {
866 			rdev->wr_log_size = 1 << c4iw_wr_log_size_order;
867 			atomic_set(&rdev->wr_log_idx, 0);
868 		}
869 	}
870 
871 	rdev->free_workq = create_singlethread_workqueue("iw_cxgb4_free");
872 	if (!rdev->free_workq) {
873 		err = -ENOMEM;
874 		goto err_free_status_page_and_wr_log;
875 	}
876 
877 	rdev->status_page->db_off = 0;
878 
879 	init_completion(&rdev->rqt_compl);
880 	init_completion(&rdev->pbl_compl);
881 	kref_init(&rdev->rqt_kref);
882 	kref_init(&rdev->pbl_kref);
883 
884 	return 0;
885 err_free_status_page_and_wr_log:
886 	if (c4iw_wr_log && rdev->wr_log)
887 		kfree(rdev->wr_log);
888 	free_page((unsigned long)rdev->status_page);
889 destroy_ocqp_pool:
890 	c4iw_ocqp_pool_destroy(rdev);
891 destroy_rqtpool:
892 	c4iw_rqtpool_destroy(rdev);
893 destroy_pblpool:
894 	c4iw_pblpool_destroy(rdev);
895 destroy_resource:
896 	c4iw_destroy_resource(&rdev->resource);
897 	return err;
898 }
899 
900 static void c4iw_rdev_close(struct c4iw_rdev *rdev)
901 {
902 	kfree(rdev->wr_log);
903 	c4iw_release_dev_ucontext(rdev, &rdev->uctx);
904 	free_page((unsigned long)rdev->status_page);
905 	c4iw_pblpool_destroy(rdev);
906 	c4iw_rqtpool_destroy(rdev);
907 	wait_for_completion(&rdev->pbl_compl);
908 	wait_for_completion(&rdev->rqt_compl);
909 	c4iw_ocqp_pool_destroy(rdev);
910 	destroy_workqueue(rdev->free_workq);
911 	c4iw_destroy_resource(&rdev->resource);
912 }
913 
914 void c4iw_dealloc(struct uld_ctx *ctx)
915 {
916 	c4iw_rdev_close(&ctx->dev->rdev);
917 	WARN_ON_ONCE(!idr_is_empty(&ctx->dev->cqidr));
918 	idr_destroy(&ctx->dev->cqidr);
919 	WARN_ON_ONCE(!idr_is_empty(&ctx->dev->qpidr));
920 	idr_destroy(&ctx->dev->qpidr);
921 	WARN_ON_ONCE(!idr_is_empty(&ctx->dev->mmidr));
922 	idr_destroy(&ctx->dev->mmidr);
923 	wait_event(ctx->dev->wait, idr_is_empty(&ctx->dev->hwtid_idr));
924 	idr_destroy(&ctx->dev->hwtid_idr);
925 	idr_destroy(&ctx->dev->stid_idr);
926 	idr_destroy(&ctx->dev->atid_idr);
927 	if (ctx->dev->rdev.bar2_kva)
928 		iounmap(ctx->dev->rdev.bar2_kva);
929 	if (ctx->dev->rdev.oc_mw_kva)
930 		iounmap(ctx->dev->rdev.oc_mw_kva);
931 	ib_dealloc_device(&ctx->dev->ibdev);
932 	ctx->dev = NULL;
933 }
934 
935 static void c4iw_remove(struct uld_ctx *ctx)
936 {
937 	pr_debug("c4iw_dev %p\n", ctx->dev);
938 	c4iw_unregister_device(ctx->dev);
939 	c4iw_dealloc(ctx);
940 }
941 
942 static int rdma_supported(const struct cxgb4_lld_info *infop)
943 {
944 	return infop->vr->stag.size > 0 && infop->vr->pbl.size > 0 &&
945 	       infop->vr->rq.size > 0 && infop->vr->qp.size > 0 &&
946 	       infop->vr->cq.size > 0;
947 }
948 
949 static struct c4iw_dev *c4iw_alloc(const struct cxgb4_lld_info *infop)
950 {
951 	struct c4iw_dev *devp;
952 	int ret;
953 
954 	if (!rdma_supported(infop)) {
955 		pr_info("%s: RDMA not supported on this device\n",
956 			pci_name(infop->pdev));
957 		return ERR_PTR(-ENOSYS);
958 	}
959 	if (!ocqp_supported(infop))
960 		pr_info("%s: On-Chip Queues not supported on this device\n",
961 			pci_name(infop->pdev));
962 
963 	devp = (struct c4iw_dev *)ib_alloc_device(sizeof(*devp));
964 	if (!devp) {
965 		pr_err("Cannot allocate ib device\n");
966 		return ERR_PTR(-ENOMEM);
967 	}
968 	devp->rdev.lldi = *infop;
969 
970 	/* init various hw-queue params based on lld info */
971 	pr_debug("Ing. padding boundary is %d, egrsstatuspagesize = %d\n",
972 		 devp->rdev.lldi.sge_ingpadboundary,
973 		 devp->rdev.lldi.sge_egrstatuspagesize);
974 
975 	devp->rdev.hw_queue.t4_eq_status_entries =
976 		devp->rdev.lldi.sge_egrstatuspagesize / 64;
977 	devp->rdev.hw_queue.t4_max_eq_size = 65520;
978 	devp->rdev.hw_queue.t4_max_iq_size = 65520;
979 	devp->rdev.hw_queue.t4_max_rq_size = 8192 -
980 		devp->rdev.hw_queue.t4_eq_status_entries - 1;
981 	devp->rdev.hw_queue.t4_max_sq_size =
982 		devp->rdev.hw_queue.t4_max_eq_size -
983 		devp->rdev.hw_queue.t4_eq_status_entries - 1;
984 	devp->rdev.hw_queue.t4_max_qp_depth =
985 		devp->rdev.hw_queue.t4_max_rq_size;
986 	devp->rdev.hw_queue.t4_max_cq_depth =
987 		devp->rdev.hw_queue.t4_max_iq_size - 2;
988 	devp->rdev.hw_queue.t4_stat_len =
989 		devp->rdev.lldi.sge_egrstatuspagesize;
990 
991 	/*
992 	 * For T5/T6 devices, we map all of BAR2 with WC.
993 	 * For T4 devices with onchip qp mem, we map only that part
994 	 * of BAR2 with WC.
995 	 */
996 	devp->rdev.bar2_pa = pci_resource_start(devp->rdev.lldi.pdev, 2);
997 	if (!is_t4(devp->rdev.lldi.adapter_type)) {
998 		devp->rdev.bar2_kva = ioremap_wc(devp->rdev.bar2_pa,
999 			pci_resource_len(devp->rdev.lldi.pdev, 2));
1000 		if (!devp->rdev.bar2_kva) {
1001 			pr_err("Unable to ioremap BAR2\n");
1002 			ib_dealloc_device(&devp->ibdev);
1003 			return ERR_PTR(-EINVAL);
1004 		}
1005 	} else if (ocqp_supported(infop)) {
1006 		devp->rdev.oc_mw_pa =
1007 			pci_resource_start(devp->rdev.lldi.pdev, 2) +
1008 			pci_resource_len(devp->rdev.lldi.pdev, 2) -
1009 			roundup_pow_of_two(devp->rdev.lldi.vr->ocq.size);
1010 		devp->rdev.oc_mw_kva = ioremap_wc(devp->rdev.oc_mw_pa,
1011 			devp->rdev.lldi.vr->ocq.size);
1012 		if (!devp->rdev.oc_mw_kva) {
1013 			pr_err("Unable to ioremap onchip mem\n");
1014 			ib_dealloc_device(&devp->ibdev);
1015 			return ERR_PTR(-EINVAL);
1016 		}
1017 	}
1018 
1019 	pr_debug("ocq memory: hw_start 0x%x size %u mw_pa 0x%lx mw_kva %p\n",
1020 		 devp->rdev.lldi.vr->ocq.start, devp->rdev.lldi.vr->ocq.size,
1021 		 devp->rdev.oc_mw_pa, devp->rdev.oc_mw_kva);
1022 
1023 	ret = c4iw_rdev_open(&devp->rdev);
1024 	if (ret) {
1025 		pr_err("Unable to open CXIO rdev err %d\n", ret);
1026 		ib_dealloc_device(&devp->ibdev);
1027 		return ERR_PTR(ret);
1028 	}
1029 
1030 	idr_init(&devp->cqidr);
1031 	idr_init(&devp->qpidr);
1032 	idr_init(&devp->mmidr);
1033 	idr_init(&devp->hwtid_idr);
1034 	idr_init(&devp->stid_idr);
1035 	idr_init(&devp->atid_idr);
1036 	spin_lock_init(&devp->lock);
1037 	mutex_init(&devp->rdev.stats.lock);
1038 	mutex_init(&devp->db_mutex);
1039 	INIT_LIST_HEAD(&devp->db_fc_list);
1040 	init_waitqueue_head(&devp->wait);
1041 	devp->avail_ird = devp->rdev.lldi.max_ird_adapter;
1042 
1043 	if (c4iw_debugfs_root) {
1044 		devp->debugfs_root = debugfs_create_dir(
1045 					pci_name(devp->rdev.lldi.pdev),
1046 					c4iw_debugfs_root);
1047 		setup_debugfs(devp);
1048 	}
1049 
1050 
1051 	return devp;
1052 }
1053 
1054 static void *c4iw_uld_add(const struct cxgb4_lld_info *infop)
1055 {
1056 	struct uld_ctx *ctx;
1057 	static int vers_printed;
1058 	int i;
1059 
1060 	if (!vers_printed++)
1061 		pr_info("Chelsio T4/T5 RDMA Driver - version %s\n",
1062 			DRV_VERSION);
1063 
1064 	ctx = kzalloc(sizeof *ctx, GFP_KERNEL);
1065 	if (!ctx) {
1066 		ctx = ERR_PTR(-ENOMEM);
1067 		goto out;
1068 	}
1069 	ctx->lldi = *infop;
1070 
1071 	pr_debug("found device %s nchan %u nrxq %u ntxq %u nports %u\n",
1072 		 pci_name(ctx->lldi.pdev),
1073 		 ctx->lldi.nchan, ctx->lldi.nrxq,
1074 		 ctx->lldi.ntxq, ctx->lldi.nports);
1075 
1076 	mutex_lock(&dev_mutex);
1077 	list_add_tail(&ctx->entry, &uld_ctx_list);
1078 	mutex_unlock(&dev_mutex);
1079 
1080 	for (i = 0; i < ctx->lldi.nrxq; i++)
1081 		pr_debug("rxqid[%u] %u\n", i, ctx->lldi.rxq_ids[i]);
1082 out:
1083 	return ctx;
1084 }
1085 
1086 static inline struct sk_buff *copy_gl_to_skb_pkt(const struct pkt_gl *gl,
1087 						 const __be64 *rsp,
1088 						 u32 pktshift)
1089 {
1090 	struct sk_buff *skb;
1091 
1092 	/*
1093 	 * Allocate space for cpl_pass_accept_req which will be synthesized by
1094 	 * driver. Once the driver synthesizes the request the skb will go
1095 	 * through the regular cpl_pass_accept_req processing.
1096 	 * The math here assumes sizeof cpl_pass_accept_req >= sizeof
1097 	 * cpl_rx_pkt.
1098 	 */
1099 	skb = alloc_skb(gl->tot_len + sizeof(struct cpl_pass_accept_req) +
1100 			sizeof(struct rss_header) - pktshift, GFP_ATOMIC);
1101 	if (unlikely(!skb))
1102 		return NULL;
1103 
1104 	__skb_put(skb, gl->tot_len + sizeof(struct cpl_pass_accept_req) +
1105 		  sizeof(struct rss_header) - pktshift);
1106 
1107 	/*
1108 	 * This skb will contain:
1109 	 *   rss_header from the rspq descriptor (1 flit)
1110 	 *   cpl_rx_pkt struct from the rspq descriptor (2 flits)
1111 	 *   space for the difference between the size of an
1112 	 *      rx_pkt and pass_accept_req cpl (1 flit)
1113 	 *   the packet data from the gl
1114 	 */
1115 	skb_copy_to_linear_data(skb, rsp, sizeof(struct cpl_pass_accept_req) +
1116 				sizeof(struct rss_header));
1117 	skb_copy_to_linear_data_offset(skb, sizeof(struct rss_header) +
1118 				       sizeof(struct cpl_pass_accept_req),
1119 				       gl->va + pktshift,
1120 				       gl->tot_len - pktshift);
1121 	return skb;
1122 }
1123 
1124 static inline int recv_rx_pkt(struct c4iw_dev *dev, const struct pkt_gl *gl,
1125 			   const __be64 *rsp)
1126 {
1127 	unsigned int opcode = *(u8 *)rsp;
1128 	struct sk_buff *skb;
1129 
1130 	if (opcode != CPL_RX_PKT)
1131 		goto out;
1132 
1133 	skb = copy_gl_to_skb_pkt(gl , rsp, dev->rdev.lldi.sge_pktshift);
1134 	if (skb == NULL)
1135 		goto out;
1136 
1137 	if (c4iw_handlers[opcode] == NULL) {
1138 		pr_info("%s no handler opcode 0x%x...\n", __func__, opcode);
1139 		kfree_skb(skb);
1140 		goto out;
1141 	}
1142 	c4iw_handlers[opcode](dev, skb);
1143 	return 1;
1144 out:
1145 	return 0;
1146 }
1147 
1148 static int c4iw_uld_rx_handler(void *handle, const __be64 *rsp,
1149 			const struct pkt_gl *gl)
1150 {
1151 	struct uld_ctx *ctx = handle;
1152 	struct c4iw_dev *dev = ctx->dev;
1153 	struct sk_buff *skb;
1154 	u8 opcode;
1155 
1156 	if (gl == NULL) {
1157 		/* omit RSS and rsp_ctrl at end of descriptor */
1158 		unsigned int len = 64 - sizeof(struct rsp_ctrl) - 8;
1159 
1160 		skb = alloc_skb(256, GFP_ATOMIC);
1161 		if (!skb)
1162 			goto nomem;
1163 		__skb_put(skb, len);
1164 		skb_copy_to_linear_data(skb, &rsp[1], len);
1165 	} else if (gl == CXGB4_MSG_AN) {
1166 		const struct rsp_ctrl *rc = (void *)rsp;
1167 
1168 		u32 qid = be32_to_cpu(rc->pldbuflen_qid);
1169 		c4iw_ev_handler(dev, qid);
1170 		return 0;
1171 	} else if (unlikely(*(u8 *)rsp != *(u8 *)gl->va)) {
1172 		if (recv_rx_pkt(dev, gl, rsp))
1173 			return 0;
1174 
1175 		pr_info("%s: unexpected FL contents at %p, RSS %#llx, FL %#llx, len %u\n",
1176 			pci_name(ctx->lldi.pdev), gl->va,
1177 			be64_to_cpu(*rsp),
1178 			be64_to_cpu(*(__force __be64 *)gl->va),
1179 			gl->tot_len);
1180 
1181 		return 0;
1182 	} else {
1183 		skb = cxgb4_pktgl_to_skb(gl, 128, 128);
1184 		if (unlikely(!skb))
1185 			goto nomem;
1186 	}
1187 
1188 	opcode = *(u8 *)rsp;
1189 	if (c4iw_handlers[opcode]) {
1190 		c4iw_handlers[opcode](dev, skb);
1191 	} else {
1192 		pr_info("%s no handler opcode 0x%x...\n", __func__, opcode);
1193 		kfree_skb(skb);
1194 	}
1195 
1196 	return 0;
1197 nomem:
1198 	return -1;
1199 }
1200 
1201 static int c4iw_uld_state_change(void *handle, enum cxgb4_state new_state)
1202 {
1203 	struct uld_ctx *ctx = handle;
1204 
1205 	pr_debug("new_state %u\n", new_state);
1206 	switch (new_state) {
1207 	case CXGB4_STATE_UP:
1208 		pr_info("%s: Up\n", pci_name(ctx->lldi.pdev));
1209 		if (!ctx->dev) {
1210 			ctx->dev = c4iw_alloc(&ctx->lldi);
1211 			if (IS_ERR(ctx->dev)) {
1212 				pr_err("%s: initialization failed: %ld\n",
1213 				       pci_name(ctx->lldi.pdev),
1214 				       PTR_ERR(ctx->dev));
1215 				ctx->dev = NULL;
1216 				break;
1217 			}
1218 
1219 			INIT_WORK(&ctx->reg_work, c4iw_register_device);
1220 			queue_work(reg_workq, &ctx->reg_work);
1221 		}
1222 		break;
1223 	case CXGB4_STATE_DOWN:
1224 		pr_info("%s: Down\n", pci_name(ctx->lldi.pdev));
1225 		if (ctx->dev)
1226 			c4iw_remove(ctx);
1227 		break;
1228 	case CXGB4_STATE_FATAL_ERROR:
1229 	case CXGB4_STATE_START_RECOVERY:
1230 		pr_info("%s: Fatal Error\n", pci_name(ctx->lldi.pdev));
1231 		if (ctx->dev) {
1232 			struct ib_event event;
1233 
1234 			ctx->dev->rdev.flags |= T4_FATAL_ERROR;
1235 			memset(&event, 0, sizeof event);
1236 			event.event  = IB_EVENT_DEVICE_FATAL;
1237 			event.device = &ctx->dev->ibdev;
1238 			ib_dispatch_event(&event);
1239 			c4iw_remove(ctx);
1240 		}
1241 		break;
1242 	case CXGB4_STATE_DETACH:
1243 		pr_info("%s: Detach\n", pci_name(ctx->lldi.pdev));
1244 		if (ctx->dev)
1245 			c4iw_remove(ctx);
1246 		break;
1247 	}
1248 	return 0;
1249 }
1250 
1251 static int disable_qp_db(int id, void *p, void *data)
1252 {
1253 	struct c4iw_qp *qp = p;
1254 
1255 	t4_disable_wq_db(&qp->wq);
1256 	return 0;
1257 }
1258 
1259 static void stop_queues(struct uld_ctx *ctx)
1260 {
1261 	unsigned long flags;
1262 
1263 	spin_lock_irqsave(&ctx->dev->lock, flags);
1264 	ctx->dev->rdev.stats.db_state_transitions++;
1265 	ctx->dev->db_state = STOPPED;
1266 	if (ctx->dev->rdev.flags & T4_STATUS_PAGE_DISABLED)
1267 		idr_for_each(&ctx->dev->qpidr, disable_qp_db, NULL);
1268 	else
1269 		ctx->dev->rdev.status_page->db_off = 1;
1270 	spin_unlock_irqrestore(&ctx->dev->lock, flags);
1271 }
1272 
1273 static int enable_qp_db(int id, void *p, void *data)
1274 {
1275 	struct c4iw_qp *qp = p;
1276 
1277 	t4_enable_wq_db(&qp->wq);
1278 	return 0;
1279 }
1280 
1281 static void resume_rc_qp(struct c4iw_qp *qp)
1282 {
1283 	spin_lock(&qp->lock);
1284 	t4_ring_sq_db(&qp->wq, qp->wq.sq.wq_pidx_inc, NULL);
1285 	qp->wq.sq.wq_pidx_inc = 0;
1286 	t4_ring_rq_db(&qp->wq, qp->wq.rq.wq_pidx_inc, NULL);
1287 	qp->wq.rq.wq_pidx_inc = 0;
1288 	spin_unlock(&qp->lock);
1289 }
1290 
1291 static void resume_a_chunk(struct uld_ctx *ctx)
1292 {
1293 	int i;
1294 	struct c4iw_qp *qp;
1295 
1296 	for (i = 0; i < DB_FC_RESUME_SIZE; i++) {
1297 		qp = list_first_entry(&ctx->dev->db_fc_list, struct c4iw_qp,
1298 				      db_fc_entry);
1299 		list_del_init(&qp->db_fc_entry);
1300 		resume_rc_qp(qp);
1301 		if (list_empty(&ctx->dev->db_fc_list))
1302 			break;
1303 	}
1304 }
1305 
1306 static void resume_queues(struct uld_ctx *ctx)
1307 {
1308 	spin_lock_irq(&ctx->dev->lock);
1309 	if (ctx->dev->db_state != STOPPED)
1310 		goto out;
1311 	ctx->dev->db_state = FLOW_CONTROL;
1312 	while (1) {
1313 		if (list_empty(&ctx->dev->db_fc_list)) {
1314 			WARN_ON(ctx->dev->db_state != FLOW_CONTROL);
1315 			ctx->dev->db_state = NORMAL;
1316 			ctx->dev->rdev.stats.db_state_transitions++;
1317 			if (ctx->dev->rdev.flags & T4_STATUS_PAGE_DISABLED) {
1318 				idr_for_each(&ctx->dev->qpidr, enable_qp_db,
1319 					     NULL);
1320 			} else {
1321 				ctx->dev->rdev.status_page->db_off = 0;
1322 			}
1323 			break;
1324 		} else {
1325 			if (cxgb4_dbfifo_count(ctx->dev->rdev.lldi.ports[0], 1)
1326 			    < (ctx->dev->rdev.lldi.dbfifo_int_thresh <<
1327 			       DB_FC_DRAIN_THRESH)) {
1328 				resume_a_chunk(ctx);
1329 			}
1330 			if (!list_empty(&ctx->dev->db_fc_list)) {
1331 				spin_unlock_irq(&ctx->dev->lock);
1332 				if (DB_FC_RESUME_DELAY) {
1333 					set_current_state(TASK_UNINTERRUPTIBLE);
1334 					schedule_timeout(DB_FC_RESUME_DELAY);
1335 				}
1336 				spin_lock_irq(&ctx->dev->lock);
1337 				if (ctx->dev->db_state != FLOW_CONTROL)
1338 					break;
1339 			}
1340 		}
1341 	}
1342 out:
1343 	if (ctx->dev->db_state != NORMAL)
1344 		ctx->dev->rdev.stats.db_fc_interruptions++;
1345 	spin_unlock_irq(&ctx->dev->lock);
1346 }
1347 
1348 struct qp_list {
1349 	unsigned idx;
1350 	struct c4iw_qp **qps;
1351 };
1352 
1353 static int add_and_ref_qp(int id, void *p, void *data)
1354 {
1355 	struct qp_list *qp_listp = data;
1356 	struct c4iw_qp *qp = p;
1357 
1358 	c4iw_qp_add_ref(&qp->ibqp);
1359 	qp_listp->qps[qp_listp->idx++] = qp;
1360 	return 0;
1361 }
1362 
1363 static int count_qps(int id, void *p, void *data)
1364 {
1365 	unsigned *countp = data;
1366 	(*countp)++;
1367 	return 0;
1368 }
1369 
1370 static void deref_qps(struct qp_list *qp_list)
1371 {
1372 	int idx;
1373 
1374 	for (idx = 0; idx < qp_list->idx; idx++)
1375 		c4iw_qp_rem_ref(&qp_list->qps[idx]->ibqp);
1376 }
1377 
1378 static void recover_lost_dbs(struct uld_ctx *ctx, struct qp_list *qp_list)
1379 {
1380 	int idx;
1381 	int ret;
1382 
1383 	for (idx = 0; idx < qp_list->idx; idx++) {
1384 		struct c4iw_qp *qp = qp_list->qps[idx];
1385 
1386 		spin_lock_irq(&qp->rhp->lock);
1387 		spin_lock(&qp->lock);
1388 		ret = cxgb4_sync_txq_pidx(qp->rhp->rdev.lldi.ports[0],
1389 					  qp->wq.sq.qid,
1390 					  t4_sq_host_wq_pidx(&qp->wq),
1391 					  t4_sq_wq_size(&qp->wq));
1392 		if (ret) {
1393 			pr_err("%s: Fatal error - DB overflow recovery failed - error syncing SQ qid %u\n",
1394 			       pci_name(ctx->lldi.pdev), qp->wq.sq.qid);
1395 			spin_unlock(&qp->lock);
1396 			spin_unlock_irq(&qp->rhp->lock);
1397 			return;
1398 		}
1399 		qp->wq.sq.wq_pidx_inc = 0;
1400 
1401 		ret = cxgb4_sync_txq_pidx(qp->rhp->rdev.lldi.ports[0],
1402 					  qp->wq.rq.qid,
1403 					  t4_rq_host_wq_pidx(&qp->wq),
1404 					  t4_rq_wq_size(&qp->wq));
1405 
1406 		if (ret) {
1407 			pr_err("%s: Fatal error - DB overflow recovery failed - error syncing RQ qid %u\n",
1408 			       pci_name(ctx->lldi.pdev), qp->wq.rq.qid);
1409 			spin_unlock(&qp->lock);
1410 			spin_unlock_irq(&qp->rhp->lock);
1411 			return;
1412 		}
1413 		qp->wq.rq.wq_pidx_inc = 0;
1414 		spin_unlock(&qp->lock);
1415 		spin_unlock_irq(&qp->rhp->lock);
1416 
1417 		/* Wait for the dbfifo to drain */
1418 		while (cxgb4_dbfifo_count(qp->rhp->rdev.lldi.ports[0], 1) > 0) {
1419 			set_current_state(TASK_UNINTERRUPTIBLE);
1420 			schedule_timeout(usecs_to_jiffies(10));
1421 		}
1422 	}
1423 }
1424 
1425 static void recover_queues(struct uld_ctx *ctx)
1426 {
1427 	int count = 0;
1428 	struct qp_list qp_list;
1429 	int ret;
1430 
1431 	/* slow everybody down */
1432 	set_current_state(TASK_UNINTERRUPTIBLE);
1433 	schedule_timeout(usecs_to_jiffies(1000));
1434 
1435 	/* flush the SGE contexts */
1436 	ret = cxgb4_flush_eq_cache(ctx->dev->rdev.lldi.ports[0]);
1437 	if (ret) {
1438 		pr_err("%s: Fatal error - DB overflow recovery failed\n",
1439 		       pci_name(ctx->lldi.pdev));
1440 		return;
1441 	}
1442 
1443 	/* Count active queues so we can build a list of queues to recover */
1444 	spin_lock_irq(&ctx->dev->lock);
1445 	WARN_ON(ctx->dev->db_state != STOPPED);
1446 	ctx->dev->db_state = RECOVERY;
1447 	idr_for_each(&ctx->dev->qpidr, count_qps, &count);
1448 
1449 	qp_list.qps = kcalloc(count, sizeof(*qp_list.qps), GFP_ATOMIC);
1450 	if (!qp_list.qps) {
1451 		spin_unlock_irq(&ctx->dev->lock);
1452 		return;
1453 	}
1454 	qp_list.idx = 0;
1455 
1456 	/* add and ref each qp so it doesn't get freed */
1457 	idr_for_each(&ctx->dev->qpidr, add_and_ref_qp, &qp_list);
1458 
1459 	spin_unlock_irq(&ctx->dev->lock);
1460 
1461 	/* now traverse the list in a safe context to recover the db state*/
1462 	recover_lost_dbs(ctx, &qp_list);
1463 
1464 	/* we're almost done!  deref the qps and clean up */
1465 	deref_qps(&qp_list);
1466 	kfree(qp_list.qps);
1467 
1468 	spin_lock_irq(&ctx->dev->lock);
1469 	WARN_ON(ctx->dev->db_state != RECOVERY);
1470 	ctx->dev->db_state = STOPPED;
1471 	spin_unlock_irq(&ctx->dev->lock);
1472 }
1473 
1474 static int c4iw_uld_control(void *handle, enum cxgb4_control control, ...)
1475 {
1476 	struct uld_ctx *ctx = handle;
1477 
1478 	switch (control) {
1479 	case CXGB4_CONTROL_DB_FULL:
1480 		stop_queues(ctx);
1481 		ctx->dev->rdev.stats.db_full++;
1482 		break;
1483 	case CXGB4_CONTROL_DB_EMPTY:
1484 		resume_queues(ctx);
1485 		mutex_lock(&ctx->dev->rdev.stats.lock);
1486 		ctx->dev->rdev.stats.db_empty++;
1487 		mutex_unlock(&ctx->dev->rdev.stats.lock);
1488 		break;
1489 	case CXGB4_CONTROL_DB_DROP:
1490 		recover_queues(ctx);
1491 		mutex_lock(&ctx->dev->rdev.stats.lock);
1492 		ctx->dev->rdev.stats.db_drop++;
1493 		mutex_unlock(&ctx->dev->rdev.stats.lock);
1494 		break;
1495 	default:
1496 		pr_warn("%s: unknown control cmd %u\n",
1497 			pci_name(ctx->lldi.pdev), control);
1498 		break;
1499 	}
1500 	return 0;
1501 }
1502 
1503 static struct cxgb4_uld_info c4iw_uld_info = {
1504 	.name = DRV_NAME,
1505 	.nrxq = MAX_ULD_QSETS,
1506 	.ntxq = MAX_ULD_QSETS,
1507 	.rxq_size = 511,
1508 	.ciq = true,
1509 	.lro = false,
1510 	.add = c4iw_uld_add,
1511 	.rx_handler = c4iw_uld_rx_handler,
1512 	.state_change = c4iw_uld_state_change,
1513 	.control = c4iw_uld_control,
1514 };
1515 
1516 void _c4iw_free_wr_wait(struct kref *kref)
1517 {
1518 	struct c4iw_wr_wait *wr_waitp;
1519 
1520 	wr_waitp = container_of(kref, struct c4iw_wr_wait, kref);
1521 	pr_debug("Free wr_wait %p\n", wr_waitp);
1522 	kfree(wr_waitp);
1523 }
1524 
1525 struct c4iw_wr_wait *c4iw_alloc_wr_wait(gfp_t gfp)
1526 {
1527 	struct c4iw_wr_wait *wr_waitp;
1528 
1529 	wr_waitp = kzalloc(sizeof(*wr_waitp), gfp);
1530 	if (wr_waitp) {
1531 		kref_init(&wr_waitp->kref);
1532 		pr_debug("wr_wait %p\n", wr_waitp);
1533 	}
1534 	return wr_waitp;
1535 }
1536 
1537 static int __init c4iw_init_module(void)
1538 {
1539 	int err;
1540 
1541 	err = c4iw_cm_init();
1542 	if (err)
1543 		return err;
1544 
1545 	c4iw_debugfs_root = debugfs_create_dir(DRV_NAME, NULL);
1546 	if (!c4iw_debugfs_root)
1547 		pr_warn("could not create debugfs entry, continuing\n");
1548 
1549 	reg_workq = create_singlethread_workqueue("Register_iWARP_device");
1550 	if (!reg_workq) {
1551 		pr_err("Failed creating workqueue to register iwarp device\n");
1552 		return -ENOMEM;
1553 	}
1554 
1555 	cxgb4_register_uld(CXGB4_ULD_RDMA, &c4iw_uld_info);
1556 
1557 	return 0;
1558 }
1559 
1560 static void __exit c4iw_exit_module(void)
1561 {
1562 	struct uld_ctx *ctx, *tmp;
1563 
1564 	mutex_lock(&dev_mutex);
1565 	list_for_each_entry_safe(ctx, tmp, &uld_ctx_list, entry) {
1566 		if (ctx->dev)
1567 			c4iw_remove(ctx);
1568 		kfree(ctx);
1569 	}
1570 	mutex_unlock(&dev_mutex);
1571 	flush_workqueue(reg_workq);
1572 	destroy_workqueue(reg_workq);
1573 	cxgb4_unregister_uld(CXGB4_ULD_RDMA);
1574 	c4iw_cm_term();
1575 	debugfs_remove_recursive(c4iw_debugfs_root);
1576 }
1577 
1578 module_init(c4iw_init_module);
1579 module_exit(c4iw_exit_module);
1580