xref: /linux/drivers/infiniband/hw/cxgb4/device.c (revision 2ba9268dd603d23e17643437b2246acb6844953b)
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 MODULE_VERSION(DRV_VERSION);
48 
49 static int allow_db_fc_on_t5;
50 module_param(allow_db_fc_on_t5, int, 0644);
51 MODULE_PARM_DESC(allow_db_fc_on_t5,
52 		 "Allow DB Flow Control on T5 (default = 0)");
53 
54 static int allow_db_coalescing_on_t5;
55 module_param(allow_db_coalescing_on_t5, int, 0644);
56 MODULE_PARM_DESC(allow_db_coalescing_on_t5,
57 		 "Allow DB Coalescing on T5 (default = 0)");
58 
59 int c4iw_wr_log = 0;
60 module_param(c4iw_wr_log, int, 0444);
61 MODULE_PARM_DESC(c4iw_wr_log, "Enables logging of work request timing data.");
62 
63 static int c4iw_wr_log_size_order = 12;
64 module_param(c4iw_wr_log_size_order, int, 0444);
65 MODULE_PARM_DESC(c4iw_wr_log_size_order,
66 		 "Number of entries (log2) in the work request timing log.");
67 
68 struct uld_ctx {
69 	struct list_head entry;
70 	struct cxgb4_lld_info lldi;
71 	struct c4iw_dev *dev;
72 };
73 
74 static LIST_HEAD(uld_ctx_list);
75 static DEFINE_MUTEX(dev_mutex);
76 
77 #define DB_FC_RESUME_SIZE 64
78 #define DB_FC_RESUME_DELAY 1
79 #define DB_FC_DRAIN_THRESH 0
80 
81 static struct dentry *c4iw_debugfs_root;
82 
83 struct c4iw_debugfs_data {
84 	struct c4iw_dev *devp;
85 	char *buf;
86 	int bufsize;
87 	int pos;
88 };
89 
90 /* registered cxgb4 netlink callbacks */
91 static struct ibnl_client_cbs c4iw_nl_cb_table[] = {
92 	[RDMA_NL_IWPM_REG_PID] = {.dump = iwpm_register_pid_cb},
93 	[RDMA_NL_IWPM_ADD_MAPPING] = {.dump = iwpm_add_mapping_cb},
94 	[RDMA_NL_IWPM_QUERY_MAPPING] = {.dump = iwpm_add_and_query_mapping_cb},
95 	[RDMA_NL_IWPM_HANDLE_ERR] = {.dump = iwpm_mapping_error_cb},
96 	[RDMA_NL_IWPM_MAPINFO] = {.dump = iwpm_mapping_info_cb},
97 	[RDMA_NL_IWPM_MAPINFO_NUM] = {.dump = iwpm_ack_mapping_info_cb}
98 };
99 
100 static int count_idrs(int id, void *p, void *data)
101 {
102 	int *countp = data;
103 
104 	*countp = *countp + 1;
105 	return 0;
106 }
107 
108 static ssize_t debugfs_read(struct file *file, char __user *buf, size_t count,
109 			    loff_t *ppos)
110 {
111 	struct c4iw_debugfs_data *d = file->private_data;
112 
113 	return simple_read_from_buffer(buf, count, ppos, d->buf, d->pos);
114 }
115 
116 void c4iw_log_wr_stats(struct t4_wq *wq, struct t4_cqe *cqe)
117 {
118 	struct wr_log_entry le;
119 	int idx;
120 
121 	if (!wq->rdev->wr_log)
122 		return;
123 
124 	idx = (atomic_inc_return(&wq->rdev->wr_log_idx) - 1) &
125 		(wq->rdev->wr_log_size - 1);
126 	le.poll_sge_ts = cxgb4_read_sge_timestamp(wq->rdev->lldi.ports[0]);
127 	getnstimeofday(&le.poll_host_ts);
128 	le.valid = 1;
129 	le.cqe_sge_ts = CQE_TS(cqe);
130 	if (SQ_TYPE(cqe)) {
131 		le.qid = wq->sq.qid;
132 		le.opcode = CQE_OPCODE(cqe);
133 		le.post_host_ts = wq->sq.sw_sq[wq->sq.cidx].host_ts;
134 		le.post_sge_ts = wq->sq.sw_sq[wq->sq.cidx].sge_ts;
135 		le.wr_id = CQE_WRID_SQ_IDX(cqe);
136 	} else {
137 		le.qid = wq->rq.qid;
138 		le.opcode = FW_RI_RECEIVE;
139 		le.post_host_ts = wq->rq.sw_rq[wq->rq.cidx].host_ts;
140 		le.post_sge_ts = wq->rq.sw_rq[wq->rq.cidx].sge_ts;
141 		le.wr_id = CQE_WRID_MSN(cqe);
142 	}
143 	wq->rdev->wr_log[idx] = le;
144 }
145 
146 static int wr_log_show(struct seq_file *seq, void *v)
147 {
148 	struct c4iw_dev *dev = seq->private;
149 	struct timespec prev_ts = {0, 0};
150 	struct wr_log_entry *lep;
151 	int prev_ts_set = 0;
152 	int idx, end;
153 
154 #define ts2ns(ts) div64_ul((ts) * dev->rdev.lldi.cclk_ps, 1000)
155 
156 	idx = atomic_read(&dev->rdev.wr_log_idx) &
157 		(dev->rdev.wr_log_size - 1);
158 	end = idx - 1;
159 	if (end < 0)
160 		end = dev->rdev.wr_log_size - 1;
161 	lep = &dev->rdev.wr_log[idx];
162 	while (idx != end) {
163 		if (lep->valid) {
164 			if (!prev_ts_set) {
165 				prev_ts_set = 1;
166 				prev_ts = lep->poll_host_ts;
167 			}
168 			seq_printf(seq, "%04u: sec %lu nsec %lu qid %u opcode "
169 				   "%u %s 0x%x host_wr_delta sec %lu nsec %lu "
170 				   "post_sge_ts 0x%llx cqe_sge_ts 0x%llx "
171 				   "poll_sge_ts 0x%llx post_poll_delta_ns %llu "
172 				   "cqe_poll_delta_ns %llu\n",
173 				   idx,
174 				   timespec_sub(lep->poll_host_ts,
175 						prev_ts).tv_sec,
176 				   timespec_sub(lep->poll_host_ts,
177 						prev_ts).tv_nsec,
178 				   lep->qid, lep->opcode,
179 				   lep->opcode == FW_RI_RECEIVE ?
180 							"msn" : "wrid",
181 				   lep->wr_id,
182 				   timespec_sub(lep->poll_host_ts,
183 						lep->post_host_ts).tv_sec,
184 				   timespec_sub(lep->poll_host_ts,
185 						lep->post_host_ts).tv_nsec,
186 				   lep->post_sge_ts, lep->cqe_sge_ts,
187 				   lep->poll_sge_ts,
188 				   ts2ns(lep->poll_sge_ts - lep->post_sge_ts),
189 				   ts2ns(lep->poll_sge_ts - lep->cqe_sge_ts));
190 			prev_ts = lep->poll_host_ts;
191 		}
192 		idx++;
193 		if (idx > (dev->rdev.wr_log_size - 1))
194 			idx = 0;
195 		lep = &dev->rdev.wr_log[idx];
196 	}
197 #undef ts2ns
198 	return 0;
199 }
200 
201 static int wr_log_open(struct inode *inode, struct file *file)
202 {
203 	return single_open(file, wr_log_show, inode->i_private);
204 }
205 
206 static ssize_t wr_log_clear(struct file *file, const char __user *buf,
207 			    size_t count, loff_t *pos)
208 {
209 	struct c4iw_dev *dev = ((struct seq_file *)file->private_data)->private;
210 	int i;
211 
212 	if (dev->rdev.wr_log)
213 		for (i = 0; i < dev->rdev.wr_log_size; i++)
214 			dev->rdev.wr_log[i].valid = 0;
215 	return count;
216 }
217 
218 static const struct file_operations wr_log_debugfs_fops = {
219 	.owner   = THIS_MODULE,
220 	.open    = wr_log_open,
221 	.release = single_release,
222 	.read    = seq_read,
223 	.llseek  = seq_lseek,
224 	.write   = wr_log_clear,
225 };
226 
227 static int dump_qp(int id, void *p, void *data)
228 {
229 	struct c4iw_qp *qp = p;
230 	struct c4iw_debugfs_data *qpd = data;
231 	int space;
232 	int cc;
233 
234 	if (id != qp->wq.sq.qid)
235 		return 0;
236 
237 	space = qpd->bufsize - qpd->pos - 1;
238 	if (space == 0)
239 		return 1;
240 
241 	if (qp->ep) {
242 		if (qp->ep->com.local_addr.ss_family == AF_INET) {
243 			struct sockaddr_in *lsin = (struct sockaddr_in *)
244 				&qp->ep->com.local_addr;
245 			struct sockaddr_in *rsin = (struct sockaddr_in *)
246 				&qp->ep->com.remote_addr;
247 			struct sockaddr_in *mapped_lsin = (struct sockaddr_in *)
248 				&qp->ep->com.mapped_local_addr;
249 			struct sockaddr_in *mapped_rsin = (struct sockaddr_in *)
250 				&qp->ep->com.mapped_remote_addr;
251 
252 			cc = snprintf(qpd->buf + qpd->pos, space,
253 				      "rc qp sq id %u rq id %u state %u "
254 				      "onchip %u ep tid %u state %u "
255 				      "%pI4:%u/%u->%pI4:%u/%u\n",
256 				      qp->wq.sq.qid, qp->wq.rq.qid,
257 				      (int)qp->attr.state,
258 				      qp->wq.sq.flags & T4_SQ_ONCHIP,
259 				      qp->ep->hwtid, (int)qp->ep->com.state,
260 				      &lsin->sin_addr, ntohs(lsin->sin_port),
261 				      ntohs(mapped_lsin->sin_port),
262 				      &rsin->sin_addr, ntohs(rsin->sin_port),
263 				      ntohs(mapped_rsin->sin_port));
264 		} else {
265 			struct sockaddr_in6 *lsin6 = (struct sockaddr_in6 *)
266 				&qp->ep->com.local_addr;
267 			struct sockaddr_in6 *rsin6 = (struct sockaddr_in6 *)
268 				&qp->ep->com.remote_addr;
269 			struct sockaddr_in6 *mapped_lsin6 =
270 				(struct sockaddr_in6 *)
271 				&qp->ep->com.mapped_local_addr;
272 			struct sockaddr_in6 *mapped_rsin6 =
273 				(struct sockaddr_in6 *)
274 				&qp->ep->com.mapped_remote_addr;
275 
276 			cc = snprintf(qpd->buf + qpd->pos, space,
277 				      "rc qp sq id %u rq id %u state %u "
278 				      "onchip %u ep tid %u state %u "
279 				      "%pI6:%u/%u->%pI6:%u/%u\n",
280 				      qp->wq.sq.qid, qp->wq.rq.qid,
281 				      (int)qp->attr.state,
282 				      qp->wq.sq.flags & T4_SQ_ONCHIP,
283 				      qp->ep->hwtid, (int)qp->ep->com.state,
284 				      &lsin6->sin6_addr,
285 				      ntohs(lsin6->sin6_port),
286 				      ntohs(mapped_lsin6->sin6_port),
287 				      &rsin6->sin6_addr,
288 				      ntohs(rsin6->sin6_port),
289 				      ntohs(mapped_rsin6->sin6_port));
290 		}
291 	} else
292 		cc = snprintf(qpd->buf + qpd->pos, space,
293 			     "qp sq id %u rq id %u state %u onchip %u\n",
294 			      qp->wq.sq.qid, qp->wq.rq.qid,
295 			      (int)qp->attr.state,
296 			      qp->wq.sq.flags & T4_SQ_ONCHIP);
297 	if (cc < space)
298 		qpd->pos += cc;
299 	return 0;
300 }
301 
302 static int qp_release(struct inode *inode, struct file *file)
303 {
304 	struct c4iw_debugfs_data *qpd = file->private_data;
305 	if (!qpd) {
306 		printk(KERN_INFO "%s null qpd?\n", __func__);
307 		return 0;
308 	}
309 	vfree(qpd->buf);
310 	kfree(qpd);
311 	return 0;
312 }
313 
314 static int qp_open(struct inode *inode, struct file *file)
315 {
316 	struct c4iw_debugfs_data *qpd;
317 	int ret = 0;
318 	int count = 1;
319 
320 	qpd = kmalloc(sizeof *qpd, GFP_KERNEL);
321 	if (!qpd) {
322 		ret = -ENOMEM;
323 		goto out;
324 	}
325 	qpd->devp = inode->i_private;
326 	qpd->pos = 0;
327 
328 	spin_lock_irq(&qpd->devp->lock);
329 	idr_for_each(&qpd->devp->qpidr, count_idrs, &count);
330 	spin_unlock_irq(&qpd->devp->lock);
331 
332 	qpd->bufsize = count * 128;
333 	qpd->buf = vmalloc(qpd->bufsize);
334 	if (!qpd->buf) {
335 		ret = -ENOMEM;
336 		goto err1;
337 	}
338 
339 	spin_lock_irq(&qpd->devp->lock);
340 	idr_for_each(&qpd->devp->qpidr, dump_qp, qpd);
341 	spin_unlock_irq(&qpd->devp->lock);
342 
343 	qpd->buf[qpd->pos++] = 0;
344 	file->private_data = qpd;
345 	goto out;
346 err1:
347 	kfree(qpd);
348 out:
349 	return ret;
350 }
351 
352 static const struct file_operations qp_debugfs_fops = {
353 	.owner   = THIS_MODULE,
354 	.open    = qp_open,
355 	.release = qp_release,
356 	.read    = debugfs_read,
357 	.llseek  = default_llseek,
358 };
359 
360 static int dump_stag(int id, void *p, void *data)
361 {
362 	struct c4iw_debugfs_data *stagd = data;
363 	int space;
364 	int cc;
365 	struct fw_ri_tpte tpte;
366 	int ret;
367 
368 	space = stagd->bufsize - stagd->pos - 1;
369 	if (space == 0)
370 		return 1;
371 
372 	ret = cxgb4_read_tpte(stagd->devp->rdev.lldi.ports[0], (u32)id<<8,
373 			      (__be32 *)&tpte);
374 	if (ret) {
375 		dev_err(&stagd->devp->rdev.lldi.pdev->dev,
376 			"%s cxgb4_read_tpte err %d\n", __func__, ret);
377 		return ret;
378 	}
379 	cc = snprintf(stagd->buf + stagd->pos, space,
380 		      "stag: idx 0x%x valid %d key 0x%x state %d pdid %d "
381 		      "perm 0x%x ps %d len 0x%llx va 0x%llx\n",
382 		      (u32)id<<8,
383 		      FW_RI_TPTE_VALID_G(ntohl(tpte.valid_to_pdid)),
384 		      FW_RI_TPTE_STAGKEY_G(ntohl(tpte.valid_to_pdid)),
385 		      FW_RI_TPTE_STAGSTATE_G(ntohl(tpte.valid_to_pdid)),
386 		      FW_RI_TPTE_PDID_G(ntohl(tpte.valid_to_pdid)),
387 		      FW_RI_TPTE_PERM_G(ntohl(tpte.locread_to_qpid)),
388 		      FW_RI_TPTE_PS_G(ntohl(tpte.locread_to_qpid)),
389 		      ((u64)ntohl(tpte.len_hi) << 32) | ntohl(tpte.len_lo),
390 		      ((u64)ntohl(tpte.va_hi) << 32) | ntohl(tpte.va_lo_fbo));
391 	if (cc < space)
392 		stagd->pos += cc;
393 	return 0;
394 }
395 
396 static int stag_release(struct inode *inode, struct file *file)
397 {
398 	struct c4iw_debugfs_data *stagd = file->private_data;
399 	if (!stagd) {
400 		printk(KERN_INFO "%s null stagd?\n", __func__);
401 		return 0;
402 	}
403 	vfree(stagd->buf);
404 	kfree(stagd);
405 	return 0;
406 }
407 
408 static int stag_open(struct inode *inode, struct file *file)
409 {
410 	struct c4iw_debugfs_data *stagd;
411 	int ret = 0;
412 	int count = 1;
413 
414 	stagd = kmalloc(sizeof *stagd, GFP_KERNEL);
415 	if (!stagd) {
416 		ret = -ENOMEM;
417 		goto out;
418 	}
419 	stagd->devp = inode->i_private;
420 	stagd->pos = 0;
421 
422 	spin_lock_irq(&stagd->devp->lock);
423 	idr_for_each(&stagd->devp->mmidr, count_idrs, &count);
424 	spin_unlock_irq(&stagd->devp->lock);
425 
426 	stagd->bufsize = count * 256;
427 	stagd->buf = vmalloc(stagd->bufsize);
428 	if (!stagd->buf) {
429 		ret = -ENOMEM;
430 		goto err1;
431 	}
432 
433 	spin_lock_irq(&stagd->devp->lock);
434 	idr_for_each(&stagd->devp->mmidr, dump_stag, stagd);
435 	spin_unlock_irq(&stagd->devp->lock);
436 
437 	stagd->buf[stagd->pos++] = 0;
438 	file->private_data = stagd;
439 	goto out;
440 err1:
441 	kfree(stagd);
442 out:
443 	return ret;
444 }
445 
446 static const struct file_operations stag_debugfs_fops = {
447 	.owner   = THIS_MODULE,
448 	.open    = stag_open,
449 	.release = stag_release,
450 	.read    = debugfs_read,
451 	.llseek  = default_llseek,
452 };
453 
454 static char *db_state_str[] = {"NORMAL", "FLOW_CONTROL", "RECOVERY", "STOPPED"};
455 
456 static int stats_show(struct seq_file *seq, void *v)
457 {
458 	struct c4iw_dev *dev = seq->private;
459 
460 	seq_printf(seq, "   Object: %10s %10s %10s %10s\n", "Total", "Current",
461 		   "Max", "Fail");
462 	seq_printf(seq, "     PDID: %10llu %10llu %10llu %10llu\n",
463 			dev->rdev.stats.pd.total, dev->rdev.stats.pd.cur,
464 			dev->rdev.stats.pd.max, dev->rdev.stats.pd.fail);
465 	seq_printf(seq, "      QID: %10llu %10llu %10llu %10llu\n",
466 			dev->rdev.stats.qid.total, dev->rdev.stats.qid.cur,
467 			dev->rdev.stats.qid.max, dev->rdev.stats.qid.fail);
468 	seq_printf(seq, "   TPTMEM: %10llu %10llu %10llu %10llu\n",
469 			dev->rdev.stats.stag.total, dev->rdev.stats.stag.cur,
470 			dev->rdev.stats.stag.max, dev->rdev.stats.stag.fail);
471 	seq_printf(seq, "   PBLMEM: %10llu %10llu %10llu %10llu\n",
472 			dev->rdev.stats.pbl.total, dev->rdev.stats.pbl.cur,
473 			dev->rdev.stats.pbl.max, dev->rdev.stats.pbl.fail);
474 	seq_printf(seq, "   RQTMEM: %10llu %10llu %10llu %10llu\n",
475 			dev->rdev.stats.rqt.total, dev->rdev.stats.rqt.cur,
476 			dev->rdev.stats.rqt.max, dev->rdev.stats.rqt.fail);
477 	seq_printf(seq, "  OCQPMEM: %10llu %10llu %10llu %10llu\n",
478 			dev->rdev.stats.ocqp.total, dev->rdev.stats.ocqp.cur,
479 			dev->rdev.stats.ocqp.max, dev->rdev.stats.ocqp.fail);
480 	seq_printf(seq, "  DB FULL: %10llu\n", dev->rdev.stats.db_full);
481 	seq_printf(seq, " DB EMPTY: %10llu\n", dev->rdev.stats.db_empty);
482 	seq_printf(seq, "  DB DROP: %10llu\n", dev->rdev.stats.db_drop);
483 	seq_printf(seq, " DB State: %s Transitions %llu FC Interruptions %llu\n",
484 		   db_state_str[dev->db_state],
485 		   dev->rdev.stats.db_state_transitions,
486 		   dev->rdev.stats.db_fc_interruptions);
487 	seq_printf(seq, "TCAM_FULL: %10llu\n", dev->rdev.stats.tcam_full);
488 	seq_printf(seq, "ACT_OFLD_CONN_FAILS: %10llu\n",
489 		   dev->rdev.stats.act_ofld_conn_fails);
490 	seq_printf(seq, "PAS_OFLD_CONN_FAILS: %10llu\n",
491 		   dev->rdev.stats.pas_ofld_conn_fails);
492 	seq_printf(seq, "AVAILABLE IRD: %10u\n", dev->avail_ird);
493 	return 0;
494 }
495 
496 static int stats_open(struct inode *inode, struct file *file)
497 {
498 	return single_open(file, stats_show, inode->i_private);
499 }
500 
501 static ssize_t stats_clear(struct file *file, const char __user *buf,
502 		size_t count, loff_t *pos)
503 {
504 	struct c4iw_dev *dev = ((struct seq_file *)file->private_data)->private;
505 
506 	mutex_lock(&dev->rdev.stats.lock);
507 	dev->rdev.stats.pd.max = 0;
508 	dev->rdev.stats.pd.fail = 0;
509 	dev->rdev.stats.qid.max = 0;
510 	dev->rdev.stats.qid.fail = 0;
511 	dev->rdev.stats.stag.max = 0;
512 	dev->rdev.stats.stag.fail = 0;
513 	dev->rdev.stats.pbl.max = 0;
514 	dev->rdev.stats.pbl.fail = 0;
515 	dev->rdev.stats.rqt.max = 0;
516 	dev->rdev.stats.rqt.fail = 0;
517 	dev->rdev.stats.ocqp.max = 0;
518 	dev->rdev.stats.ocqp.fail = 0;
519 	dev->rdev.stats.db_full = 0;
520 	dev->rdev.stats.db_empty = 0;
521 	dev->rdev.stats.db_drop = 0;
522 	dev->rdev.stats.db_state_transitions = 0;
523 	dev->rdev.stats.tcam_full = 0;
524 	dev->rdev.stats.act_ofld_conn_fails = 0;
525 	dev->rdev.stats.pas_ofld_conn_fails = 0;
526 	mutex_unlock(&dev->rdev.stats.lock);
527 	return count;
528 }
529 
530 static const struct file_operations stats_debugfs_fops = {
531 	.owner   = THIS_MODULE,
532 	.open    = stats_open,
533 	.release = single_release,
534 	.read    = seq_read,
535 	.llseek  = seq_lseek,
536 	.write   = stats_clear,
537 };
538 
539 static int dump_ep(int id, void *p, void *data)
540 {
541 	struct c4iw_ep *ep = p;
542 	struct c4iw_debugfs_data *epd = data;
543 	int space;
544 	int cc;
545 
546 	space = epd->bufsize - epd->pos - 1;
547 	if (space == 0)
548 		return 1;
549 
550 	if (ep->com.local_addr.ss_family == AF_INET) {
551 		struct sockaddr_in *lsin = (struct sockaddr_in *)
552 			&ep->com.local_addr;
553 		struct sockaddr_in *rsin = (struct sockaddr_in *)
554 			&ep->com.remote_addr;
555 		struct sockaddr_in *mapped_lsin = (struct sockaddr_in *)
556 			&ep->com.mapped_local_addr;
557 		struct sockaddr_in *mapped_rsin = (struct sockaddr_in *)
558 			&ep->com.mapped_remote_addr;
559 
560 		cc = snprintf(epd->buf + epd->pos, space,
561 			      "ep %p cm_id %p qp %p state %d flags 0x%lx "
562 			      "history 0x%lx hwtid %d atid %d "
563 			      "%pI4:%d/%d <-> %pI4:%d/%d\n",
564 			      ep, ep->com.cm_id, ep->com.qp,
565 			      (int)ep->com.state, ep->com.flags,
566 			      ep->com.history, ep->hwtid, ep->atid,
567 			      &lsin->sin_addr, ntohs(lsin->sin_port),
568 			      ntohs(mapped_lsin->sin_port),
569 			      &rsin->sin_addr, ntohs(rsin->sin_port),
570 			      ntohs(mapped_rsin->sin_port));
571 	} else {
572 		struct sockaddr_in6 *lsin6 = (struct sockaddr_in6 *)
573 			&ep->com.local_addr;
574 		struct sockaddr_in6 *rsin6 = (struct sockaddr_in6 *)
575 			&ep->com.remote_addr;
576 		struct sockaddr_in6 *mapped_lsin6 = (struct sockaddr_in6 *)
577 			&ep->com.mapped_local_addr;
578 		struct sockaddr_in6 *mapped_rsin6 = (struct sockaddr_in6 *)
579 			&ep->com.mapped_remote_addr;
580 
581 		cc = snprintf(epd->buf + epd->pos, space,
582 			      "ep %p cm_id %p qp %p state %d flags 0x%lx "
583 			      "history 0x%lx hwtid %d atid %d "
584 			      "%pI6:%d/%d <-> %pI6:%d/%d\n",
585 			      ep, ep->com.cm_id, ep->com.qp,
586 			      (int)ep->com.state, ep->com.flags,
587 			      ep->com.history, ep->hwtid, ep->atid,
588 			      &lsin6->sin6_addr, ntohs(lsin6->sin6_port),
589 			      ntohs(mapped_lsin6->sin6_port),
590 			      &rsin6->sin6_addr, ntohs(rsin6->sin6_port),
591 			      ntohs(mapped_rsin6->sin6_port));
592 	}
593 	if (cc < space)
594 		epd->pos += cc;
595 	return 0;
596 }
597 
598 static int dump_listen_ep(int id, void *p, void *data)
599 {
600 	struct c4iw_listen_ep *ep = p;
601 	struct c4iw_debugfs_data *epd = data;
602 	int space;
603 	int cc;
604 
605 	space = epd->bufsize - epd->pos - 1;
606 	if (space == 0)
607 		return 1;
608 
609 	if (ep->com.local_addr.ss_family == AF_INET) {
610 		struct sockaddr_in *lsin = (struct sockaddr_in *)
611 			&ep->com.local_addr;
612 		struct sockaddr_in *mapped_lsin = (struct sockaddr_in *)
613 			&ep->com.mapped_local_addr;
614 
615 		cc = snprintf(epd->buf + epd->pos, space,
616 			      "ep %p cm_id %p state %d flags 0x%lx stid %d "
617 			      "backlog %d %pI4:%d/%d\n",
618 			      ep, ep->com.cm_id, (int)ep->com.state,
619 			      ep->com.flags, ep->stid, ep->backlog,
620 			      &lsin->sin_addr, ntohs(lsin->sin_port),
621 			      ntohs(mapped_lsin->sin_port));
622 	} else {
623 		struct sockaddr_in6 *lsin6 = (struct sockaddr_in6 *)
624 			&ep->com.local_addr;
625 		struct sockaddr_in6 *mapped_lsin6 = (struct sockaddr_in6 *)
626 			&ep->com.mapped_local_addr;
627 
628 		cc = snprintf(epd->buf + epd->pos, space,
629 			      "ep %p cm_id %p state %d flags 0x%lx stid %d "
630 			      "backlog %d %pI6:%d/%d\n",
631 			      ep, ep->com.cm_id, (int)ep->com.state,
632 			      ep->com.flags, ep->stid, ep->backlog,
633 			      &lsin6->sin6_addr, ntohs(lsin6->sin6_port),
634 			      ntohs(mapped_lsin6->sin6_port));
635 	}
636 	if (cc < space)
637 		epd->pos += cc;
638 	return 0;
639 }
640 
641 static int ep_release(struct inode *inode, struct file *file)
642 {
643 	struct c4iw_debugfs_data *epd = file->private_data;
644 	if (!epd) {
645 		pr_info("%s null qpd?\n", __func__);
646 		return 0;
647 	}
648 	vfree(epd->buf);
649 	kfree(epd);
650 	return 0;
651 }
652 
653 static int ep_open(struct inode *inode, struct file *file)
654 {
655 	struct c4iw_debugfs_data *epd;
656 	int ret = 0;
657 	int count = 1;
658 
659 	epd = kmalloc(sizeof(*epd), GFP_KERNEL);
660 	if (!epd) {
661 		ret = -ENOMEM;
662 		goto out;
663 	}
664 	epd->devp = inode->i_private;
665 	epd->pos = 0;
666 
667 	spin_lock_irq(&epd->devp->lock);
668 	idr_for_each(&epd->devp->hwtid_idr, count_idrs, &count);
669 	idr_for_each(&epd->devp->atid_idr, count_idrs, &count);
670 	idr_for_each(&epd->devp->stid_idr, count_idrs, &count);
671 	spin_unlock_irq(&epd->devp->lock);
672 
673 	epd->bufsize = count * 240;
674 	epd->buf = vmalloc(epd->bufsize);
675 	if (!epd->buf) {
676 		ret = -ENOMEM;
677 		goto err1;
678 	}
679 
680 	spin_lock_irq(&epd->devp->lock);
681 	idr_for_each(&epd->devp->hwtid_idr, dump_ep, epd);
682 	idr_for_each(&epd->devp->atid_idr, dump_ep, epd);
683 	idr_for_each(&epd->devp->stid_idr, dump_listen_ep, epd);
684 	spin_unlock_irq(&epd->devp->lock);
685 
686 	file->private_data = epd;
687 	goto out;
688 err1:
689 	kfree(epd);
690 out:
691 	return ret;
692 }
693 
694 static const struct file_operations ep_debugfs_fops = {
695 	.owner   = THIS_MODULE,
696 	.open    = ep_open,
697 	.release = ep_release,
698 	.read    = debugfs_read,
699 };
700 
701 static int setup_debugfs(struct c4iw_dev *devp)
702 {
703 	if (!devp->debugfs_root)
704 		return -1;
705 
706 	debugfs_create_file_size("qps", S_IWUSR, devp->debugfs_root,
707 				 (void *)devp, &qp_debugfs_fops, 4096);
708 
709 	debugfs_create_file_size("stags", S_IWUSR, devp->debugfs_root,
710 				 (void *)devp, &stag_debugfs_fops, 4096);
711 
712 	debugfs_create_file_size("stats", S_IWUSR, devp->debugfs_root,
713 				 (void *)devp, &stats_debugfs_fops, 4096);
714 
715 	debugfs_create_file_size("eps", S_IWUSR, devp->debugfs_root,
716 				 (void *)devp, &ep_debugfs_fops, 4096);
717 
718 	if (c4iw_wr_log)
719 		debugfs_create_file_size("wr_log", S_IWUSR, devp->debugfs_root,
720 					 (void *)devp, &wr_log_debugfs_fops, 4096);
721 	return 0;
722 }
723 
724 void c4iw_release_dev_ucontext(struct c4iw_rdev *rdev,
725 			       struct c4iw_dev_ucontext *uctx)
726 {
727 	struct list_head *pos, *nxt;
728 	struct c4iw_qid_list *entry;
729 
730 	mutex_lock(&uctx->lock);
731 	list_for_each_safe(pos, nxt, &uctx->qpids) {
732 		entry = list_entry(pos, struct c4iw_qid_list, entry);
733 		list_del_init(&entry->entry);
734 		if (!(entry->qid & rdev->qpmask)) {
735 			c4iw_put_resource(&rdev->resource.qid_table,
736 					  entry->qid);
737 			mutex_lock(&rdev->stats.lock);
738 			rdev->stats.qid.cur -= rdev->qpmask + 1;
739 			mutex_unlock(&rdev->stats.lock);
740 		}
741 		kfree(entry);
742 	}
743 
744 	list_for_each_safe(pos, nxt, &uctx->qpids) {
745 		entry = list_entry(pos, struct c4iw_qid_list, entry);
746 		list_del_init(&entry->entry);
747 		kfree(entry);
748 	}
749 	mutex_unlock(&uctx->lock);
750 }
751 
752 void c4iw_init_dev_ucontext(struct c4iw_rdev *rdev,
753 			    struct c4iw_dev_ucontext *uctx)
754 {
755 	INIT_LIST_HEAD(&uctx->qpids);
756 	INIT_LIST_HEAD(&uctx->cqids);
757 	mutex_init(&uctx->lock);
758 }
759 
760 /* Caller takes care of locking if needed */
761 static int c4iw_rdev_open(struct c4iw_rdev *rdev)
762 {
763 	int err;
764 
765 	c4iw_init_dev_ucontext(rdev, &rdev->uctx);
766 
767 	/*
768 	 * qpshift is the number of bits to shift the qpid left in order
769 	 * to get the correct address of the doorbell for that qp.
770 	 */
771 	rdev->qpshift = PAGE_SHIFT - ilog2(rdev->lldi.udb_density);
772 	rdev->qpmask = rdev->lldi.udb_density - 1;
773 	rdev->cqshift = PAGE_SHIFT - ilog2(rdev->lldi.ucq_density);
774 	rdev->cqmask = rdev->lldi.ucq_density - 1;
775 	PDBG("%s dev %s stag start 0x%0x size 0x%0x num stags %d "
776 	     "pbl start 0x%0x size 0x%0x rq start 0x%0x size 0x%0x "
777 	     "qp qid start %u size %u cq qid start %u size %u\n",
778 	     __func__, pci_name(rdev->lldi.pdev), rdev->lldi.vr->stag.start,
779 	     rdev->lldi.vr->stag.size, c4iw_num_stags(rdev),
780 	     rdev->lldi.vr->pbl.start,
781 	     rdev->lldi.vr->pbl.size, rdev->lldi.vr->rq.start,
782 	     rdev->lldi.vr->rq.size,
783 	     rdev->lldi.vr->qp.start,
784 	     rdev->lldi.vr->qp.size,
785 	     rdev->lldi.vr->cq.start,
786 	     rdev->lldi.vr->cq.size);
787 	PDBG("udb len 0x%x udb base %llx db_reg %p gts_reg %p qpshift %lu "
788 	     "qpmask 0x%x cqshift %lu cqmask 0x%x\n",
789 	     (unsigned)pci_resource_len(rdev->lldi.pdev, 2),
790 	     (u64)pci_resource_start(rdev->lldi.pdev, 2),
791 	     rdev->lldi.db_reg,
792 	     rdev->lldi.gts_reg,
793 	     rdev->qpshift, rdev->qpmask,
794 	     rdev->cqshift, rdev->cqmask);
795 
796 	if (c4iw_num_stags(rdev) == 0) {
797 		err = -EINVAL;
798 		goto err1;
799 	}
800 
801 	rdev->stats.pd.total = T4_MAX_NUM_PD;
802 	rdev->stats.stag.total = rdev->lldi.vr->stag.size;
803 	rdev->stats.pbl.total = rdev->lldi.vr->pbl.size;
804 	rdev->stats.rqt.total = rdev->lldi.vr->rq.size;
805 	rdev->stats.ocqp.total = rdev->lldi.vr->ocq.size;
806 	rdev->stats.qid.total = rdev->lldi.vr->qp.size;
807 
808 	err = c4iw_init_resource(rdev, c4iw_num_stags(rdev), T4_MAX_NUM_PD);
809 	if (err) {
810 		printk(KERN_ERR MOD "error %d initializing resources\n", err);
811 		goto err1;
812 	}
813 	err = c4iw_pblpool_create(rdev);
814 	if (err) {
815 		printk(KERN_ERR MOD "error %d initializing pbl pool\n", err);
816 		goto err2;
817 	}
818 	err = c4iw_rqtpool_create(rdev);
819 	if (err) {
820 		printk(KERN_ERR MOD "error %d initializing rqt pool\n", err);
821 		goto err3;
822 	}
823 	err = c4iw_ocqp_pool_create(rdev);
824 	if (err) {
825 		printk(KERN_ERR MOD "error %d initializing ocqp pool\n", err);
826 		goto err4;
827 	}
828 	rdev->status_page = (struct t4_dev_status_page *)
829 			    __get_free_page(GFP_KERNEL);
830 	if (!rdev->status_page) {
831 		pr_err(MOD "error allocating status page\n");
832 		goto err4;
833 	}
834 
835 	if (c4iw_wr_log) {
836 		rdev->wr_log = kzalloc((1 << c4iw_wr_log_size_order) *
837 				       sizeof(*rdev->wr_log), GFP_KERNEL);
838 		if (rdev->wr_log) {
839 			rdev->wr_log_size = 1 << c4iw_wr_log_size_order;
840 			atomic_set(&rdev->wr_log_idx, 0);
841 		} else {
842 			pr_err(MOD "error allocating wr_log. Logging disabled\n");
843 		}
844 	}
845 
846 	rdev->status_page->db_off = 0;
847 
848 	return 0;
849 err4:
850 	c4iw_rqtpool_destroy(rdev);
851 err3:
852 	c4iw_pblpool_destroy(rdev);
853 err2:
854 	c4iw_destroy_resource(&rdev->resource);
855 err1:
856 	return err;
857 }
858 
859 static void c4iw_rdev_close(struct c4iw_rdev *rdev)
860 {
861 	kfree(rdev->wr_log);
862 	free_page((unsigned long)rdev->status_page);
863 	c4iw_pblpool_destroy(rdev);
864 	c4iw_rqtpool_destroy(rdev);
865 	c4iw_destroy_resource(&rdev->resource);
866 }
867 
868 static void c4iw_dealloc(struct uld_ctx *ctx)
869 {
870 	c4iw_rdev_close(&ctx->dev->rdev);
871 	idr_destroy(&ctx->dev->cqidr);
872 	idr_destroy(&ctx->dev->qpidr);
873 	idr_destroy(&ctx->dev->mmidr);
874 	idr_destroy(&ctx->dev->hwtid_idr);
875 	idr_destroy(&ctx->dev->stid_idr);
876 	idr_destroy(&ctx->dev->atid_idr);
877 	if (ctx->dev->rdev.bar2_kva)
878 		iounmap(ctx->dev->rdev.bar2_kva);
879 	if (ctx->dev->rdev.oc_mw_kva)
880 		iounmap(ctx->dev->rdev.oc_mw_kva);
881 	ib_dealloc_device(&ctx->dev->ibdev);
882 	ctx->dev = NULL;
883 }
884 
885 static void c4iw_remove(struct uld_ctx *ctx)
886 {
887 	PDBG("%s c4iw_dev %p\n", __func__,  ctx->dev);
888 	c4iw_unregister_device(ctx->dev);
889 	c4iw_dealloc(ctx);
890 }
891 
892 static int rdma_supported(const struct cxgb4_lld_info *infop)
893 {
894 	return infop->vr->stag.size > 0 && infop->vr->pbl.size > 0 &&
895 	       infop->vr->rq.size > 0 && infop->vr->qp.size > 0 &&
896 	       infop->vr->cq.size > 0;
897 }
898 
899 static struct c4iw_dev *c4iw_alloc(const struct cxgb4_lld_info *infop)
900 {
901 	struct c4iw_dev *devp;
902 	int ret;
903 
904 	if (!rdma_supported(infop)) {
905 		printk(KERN_INFO MOD "%s: RDMA not supported on this device.\n",
906 		       pci_name(infop->pdev));
907 		return ERR_PTR(-ENOSYS);
908 	}
909 	if (!ocqp_supported(infop))
910 		pr_info("%s: On-Chip Queues not supported on this device.\n",
911 			pci_name(infop->pdev));
912 
913 	devp = (struct c4iw_dev *)ib_alloc_device(sizeof(*devp));
914 	if (!devp) {
915 		printk(KERN_ERR MOD "Cannot allocate ib device\n");
916 		return ERR_PTR(-ENOMEM);
917 	}
918 	devp->rdev.lldi = *infop;
919 
920 	/* init various hw-queue params based on lld info */
921 	PDBG("%s: Ing. padding boundary is %d, egrsstatuspagesize = %d\n",
922 	     __func__, devp->rdev.lldi.sge_ingpadboundary,
923 	     devp->rdev.lldi.sge_egrstatuspagesize);
924 
925 	devp->rdev.hw_queue.t4_eq_status_entries =
926 		devp->rdev.lldi.sge_ingpadboundary > 64 ? 2 : 1;
927 	devp->rdev.hw_queue.t4_max_eq_size = 65520;
928 	devp->rdev.hw_queue.t4_max_iq_size = 65520;
929 	devp->rdev.hw_queue.t4_max_rq_size = 8192 -
930 		devp->rdev.hw_queue.t4_eq_status_entries - 1;
931 	devp->rdev.hw_queue.t4_max_sq_size =
932 		devp->rdev.hw_queue.t4_max_eq_size -
933 		devp->rdev.hw_queue.t4_eq_status_entries - 1;
934 	devp->rdev.hw_queue.t4_max_qp_depth =
935 		devp->rdev.hw_queue.t4_max_rq_size;
936 	devp->rdev.hw_queue.t4_max_cq_depth =
937 		devp->rdev.hw_queue.t4_max_iq_size - 2;
938 	devp->rdev.hw_queue.t4_stat_len =
939 		devp->rdev.lldi.sge_egrstatuspagesize;
940 
941 	/*
942 	 * For T5 devices, we map all of BAR2 with WC.
943 	 * For T4 devices with onchip qp mem, we map only that part
944 	 * of BAR2 with WC.
945 	 */
946 	devp->rdev.bar2_pa = pci_resource_start(devp->rdev.lldi.pdev, 2);
947 	if (is_t5(devp->rdev.lldi.adapter_type)) {
948 		devp->rdev.bar2_kva = ioremap_wc(devp->rdev.bar2_pa,
949 			pci_resource_len(devp->rdev.lldi.pdev, 2));
950 		if (!devp->rdev.bar2_kva) {
951 			pr_err(MOD "Unable to ioremap BAR2\n");
952 			ib_dealloc_device(&devp->ibdev);
953 			return ERR_PTR(-EINVAL);
954 		}
955 	} else if (ocqp_supported(infop)) {
956 		devp->rdev.oc_mw_pa =
957 			pci_resource_start(devp->rdev.lldi.pdev, 2) +
958 			pci_resource_len(devp->rdev.lldi.pdev, 2) -
959 			roundup_pow_of_two(devp->rdev.lldi.vr->ocq.size);
960 		devp->rdev.oc_mw_kva = ioremap_wc(devp->rdev.oc_mw_pa,
961 			devp->rdev.lldi.vr->ocq.size);
962 		if (!devp->rdev.oc_mw_kva) {
963 			pr_err(MOD "Unable to ioremap onchip mem\n");
964 			ib_dealloc_device(&devp->ibdev);
965 			return ERR_PTR(-EINVAL);
966 		}
967 	}
968 
969 	PDBG(KERN_INFO MOD "ocq memory: "
970 	       "hw_start 0x%x size %u mw_pa 0x%lx mw_kva %p\n",
971 	       devp->rdev.lldi.vr->ocq.start, devp->rdev.lldi.vr->ocq.size,
972 	       devp->rdev.oc_mw_pa, devp->rdev.oc_mw_kva);
973 
974 	ret = c4iw_rdev_open(&devp->rdev);
975 	if (ret) {
976 		printk(KERN_ERR MOD "Unable to open CXIO rdev err %d\n", ret);
977 		ib_dealloc_device(&devp->ibdev);
978 		return ERR_PTR(ret);
979 	}
980 
981 	idr_init(&devp->cqidr);
982 	idr_init(&devp->qpidr);
983 	idr_init(&devp->mmidr);
984 	idr_init(&devp->hwtid_idr);
985 	idr_init(&devp->stid_idr);
986 	idr_init(&devp->atid_idr);
987 	spin_lock_init(&devp->lock);
988 	mutex_init(&devp->rdev.stats.lock);
989 	mutex_init(&devp->db_mutex);
990 	INIT_LIST_HEAD(&devp->db_fc_list);
991 	devp->avail_ird = devp->rdev.lldi.max_ird_adapter;
992 
993 	if (c4iw_debugfs_root) {
994 		devp->debugfs_root = debugfs_create_dir(
995 					pci_name(devp->rdev.lldi.pdev),
996 					c4iw_debugfs_root);
997 		setup_debugfs(devp);
998 	}
999 
1000 
1001 	return devp;
1002 }
1003 
1004 static void *c4iw_uld_add(const struct cxgb4_lld_info *infop)
1005 {
1006 	struct uld_ctx *ctx;
1007 	static int vers_printed;
1008 	int i;
1009 
1010 	if (!vers_printed++)
1011 		pr_info("Chelsio T4/T5 RDMA Driver - version %s\n",
1012 			DRV_VERSION);
1013 
1014 	ctx = kzalloc(sizeof *ctx, GFP_KERNEL);
1015 	if (!ctx) {
1016 		ctx = ERR_PTR(-ENOMEM);
1017 		goto out;
1018 	}
1019 	ctx->lldi = *infop;
1020 
1021 	PDBG("%s found device %s nchan %u nrxq %u ntxq %u nports %u\n",
1022 	     __func__, pci_name(ctx->lldi.pdev),
1023 	     ctx->lldi.nchan, ctx->lldi.nrxq,
1024 	     ctx->lldi.ntxq, ctx->lldi.nports);
1025 
1026 	mutex_lock(&dev_mutex);
1027 	list_add_tail(&ctx->entry, &uld_ctx_list);
1028 	mutex_unlock(&dev_mutex);
1029 
1030 	for (i = 0; i < ctx->lldi.nrxq; i++)
1031 		PDBG("rxqid[%u] %u\n", i, ctx->lldi.rxq_ids[i]);
1032 out:
1033 	return ctx;
1034 }
1035 
1036 static inline struct sk_buff *copy_gl_to_skb_pkt(const struct pkt_gl *gl,
1037 						 const __be64 *rsp,
1038 						 u32 pktshift)
1039 {
1040 	struct sk_buff *skb;
1041 
1042 	/*
1043 	 * Allocate space for cpl_pass_accept_req which will be synthesized by
1044 	 * driver. Once the driver synthesizes the request the skb will go
1045 	 * through the regular cpl_pass_accept_req processing.
1046 	 * The math here assumes sizeof cpl_pass_accept_req >= sizeof
1047 	 * cpl_rx_pkt.
1048 	 */
1049 	skb = alloc_skb(gl->tot_len + sizeof(struct cpl_pass_accept_req) +
1050 			sizeof(struct rss_header) - pktshift, GFP_ATOMIC);
1051 	if (unlikely(!skb))
1052 		return NULL;
1053 
1054 	 __skb_put(skb, gl->tot_len + sizeof(struct cpl_pass_accept_req) +
1055 		   sizeof(struct rss_header) - pktshift);
1056 
1057 	/*
1058 	 * This skb will contain:
1059 	 *   rss_header from the rspq descriptor (1 flit)
1060 	 *   cpl_rx_pkt struct from the rspq descriptor (2 flits)
1061 	 *   space for the difference between the size of an
1062 	 *      rx_pkt and pass_accept_req cpl (1 flit)
1063 	 *   the packet data from the gl
1064 	 */
1065 	skb_copy_to_linear_data(skb, rsp, sizeof(struct cpl_pass_accept_req) +
1066 				sizeof(struct rss_header));
1067 	skb_copy_to_linear_data_offset(skb, sizeof(struct rss_header) +
1068 				       sizeof(struct cpl_pass_accept_req),
1069 				       gl->va + pktshift,
1070 				       gl->tot_len - pktshift);
1071 	return skb;
1072 }
1073 
1074 static inline int recv_rx_pkt(struct c4iw_dev *dev, const struct pkt_gl *gl,
1075 			   const __be64 *rsp)
1076 {
1077 	unsigned int opcode = *(u8 *)rsp;
1078 	struct sk_buff *skb;
1079 
1080 	if (opcode != CPL_RX_PKT)
1081 		goto out;
1082 
1083 	skb = copy_gl_to_skb_pkt(gl , rsp, dev->rdev.lldi.sge_pktshift);
1084 	if (skb == NULL)
1085 		goto out;
1086 
1087 	if (c4iw_handlers[opcode] == NULL) {
1088 		pr_info("%s no handler opcode 0x%x...\n", __func__,
1089 		       opcode);
1090 		kfree_skb(skb);
1091 		goto out;
1092 	}
1093 	c4iw_handlers[opcode](dev, skb);
1094 	return 1;
1095 out:
1096 	return 0;
1097 }
1098 
1099 static int c4iw_uld_rx_handler(void *handle, const __be64 *rsp,
1100 			const struct pkt_gl *gl)
1101 {
1102 	struct uld_ctx *ctx = handle;
1103 	struct c4iw_dev *dev = ctx->dev;
1104 	struct sk_buff *skb;
1105 	u8 opcode;
1106 
1107 	if (gl == NULL) {
1108 		/* omit RSS and rsp_ctrl at end of descriptor */
1109 		unsigned int len = 64 - sizeof(struct rsp_ctrl) - 8;
1110 
1111 		skb = alloc_skb(256, GFP_ATOMIC);
1112 		if (!skb)
1113 			goto nomem;
1114 		__skb_put(skb, len);
1115 		skb_copy_to_linear_data(skb, &rsp[1], len);
1116 	} else if (gl == CXGB4_MSG_AN) {
1117 		const struct rsp_ctrl *rc = (void *)rsp;
1118 
1119 		u32 qid = be32_to_cpu(rc->pldbuflen_qid);
1120 		c4iw_ev_handler(dev, qid);
1121 		return 0;
1122 	} else if (unlikely(*(u8 *)rsp != *(u8 *)gl->va)) {
1123 		if (recv_rx_pkt(dev, gl, rsp))
1124 			return 0;
1125 
1126 		pr_info("%s: unexpected FL contents at %p, " \
1127 		       "RSS %#llx, FL %#llx, len %u\n",
1128 		       pci_name(ctx->lldi.pdev), gl->va,
1129 		       (unsigned long long)be64_to_cpu(*rsp),
1130 		       (unsigned long long)be64_to_cpu(
1131 		       *(__force __be64 *)gl->va),
1132 		       gl->tot_len);
1133 
1134 		return 0;
1135 	} else {
1136 		skb = cxgb4_pktgl_to_skb(gl, 128, 128);
1137 		if (unlikely(!skb))
1138 			goto nomem;
1139 	}
1140 
1141 	opcode = *(u8 *)rsp;
1142 	if (c4iw_handlers[opcode]) {
1143 		c4iw_handlers[opcode](dev, skb);
1144 	} else {
1145 		pr_info("%s no handler opcode 0x%x...\n", __func__,
1146 		       opcode);
1147 		kfree_skb(skb);
1148 	}
1149 
1150 	return 0;
1151 nomem:
1152 	return -1;
1153 }
1154 
1155 static int c4iw_uld_state_change(void *handle, enum cxgb4_state new_state)
1156 {
1157 	struct uld_ctx *ctx = handle;
1158 
1159 	PDBG("%s new_state %u\n", __func__, new_state);
1160 	switch (new_state) {
1161 	case CXGB4_STATE_UP:
1162 		printk(KERN_INFO MOD "%s: Up\n", pci_name(ctx->lldi.pdev));
1163 		if (!ctx->dev) {
1164 			int ret;
1165 
1166 			ctx->dev = c4iw_alloc(&ctx->lldi);
1167 			if (IS_ERR(ctx->dev)) {
1168 				printk(KERN_ERR MOD
1169 				       "%s: initialization failed: %ld\n",
1170 				       pci_name(ctx->lldi.pdev),
1171 				       PTR_ERR(ctx->dev));
1172 				ctx->dev = NULL;
1173 				break;
1174 			}
1175 			ret = c4iw_register_device(ctx->dev);
1176 			if (ret) {
1177 				printk(KERN_ERR MOD
1178 				       "%s: RDMA registration failed: %d\n",
1179 				       pci_name(ctx->lldi.pdev), ret);
1180 				c4iw_dealloc(ctx);
1181 			}
1182 		}
1183 		break;
1184 	case CXGB4_STATE_DOWN:
1185 		printk(KERN_INFO MOD "%s: Down\n",
1186 		       pci_name(ctx->lldi.pdev));
1187 		if (ctx->dev)
1188 			c4iw_remove(ctx);
1189 		break;
1190 	case CXGB4_STATE_START_RECOVERY:
1191 		printk(KERN_INFO MOD "%s: Fatal Error\n",
1192 		       pci_name(ctx->lldi.pdev));
1193 		if (ctx->dev) {
1194 			struct ib_event event;
1195 
1196 			ctx->dev->rdev.flags |= T4_FATAL_ERROR;
1197 			memset(&event, 0, sizeof event);
1198 			event.event  = IB_EVENT_DEVICE_FATAL;
1199 			event.device = &ctx->dev->ibdev;
1200 			ib_dispatch_event(&event);
1201 			c4iw_remove(ctx);
1202 		}
1203 		break;
1204 	case CXGB4_STATE_DETACH:
1205 		printk(KERN_INFO MOD "%s: Detach\n",
1206 		       pci_name(ctx->lldi.pdev));
1207 		if (ctx->dev)
1208 			c4iw_remove(ctx);
1209 		break;
1210 	}
1211 	return 0;
1212 }
1213 
1214 static int disable_qp_db(int id, void *p, void *data)
1215 {
1216 	struct c4iw_qp *qp = p;
1217 
1218 	t4_disable_wq_db(&qp->wq);
1219 	return 0;
1220 }
1221 
1222 static void stop_queues(struct uld_ctx *ctx)
1223 {
1224 	unsigned long flags;
1225 
1226 	spin_lock_irqsave(&ctx->dev->lock, flags);
1227 	ctx->dev->rdev.stats.db_state_transitions++;
1228 	ctx->dev->db_state = STOPPED;
1229 	if (ctx->dev->rdev.flags & T4_STATUS_PAGE_DISABLED)
1230 		idr_for_each(&ctx->dev->qpidr, disable_qp_db, NULL);
1231 	else
1232 		ctx->dev->rdev.status_page->db_off = 1;
1233 	spin_unlock_irqrestore(&ctx->dev->lock, flags);
1234 }
1235 
1236 static int enable_qp_db(int id, void *p, void *data)
1237 {
1238 	struct c4iw_qp *qp = p;
1239 
1240 	t4_enable_wq_db(&qp->wq);
1241 	return 0;
1242 }
1243 
1244 static void resume_rc_qp(struct c4iw_qp *qp)
1245 {
1246 	spin_lock(&qp->lock);
1247 	t4_ring_sq_db(&qp->wq, qp->wq.sq.wq_pidx_inc,
1248 		      is_t5(qp->rhp->rdev.lldi.adapter_type), NULL);
1249 	qp->wq.sq.wq_pidx_inc = 0;
1250 	t4_ring_rq_db(&qp->wq, qp->wq.rq.wq_pidx_inc,
1251 		      is_t5(qp->rhp->rdev.lldi.adapter_type), NULL);
1252 	qp->wq.rq.wq_pidx_inc = 0;
1253 	spin_unlock(&qp->lock);
1254 }
1255 
1256 static void resume_a_chunk(struct uld_ctx *ctx)
1257 {
1258 	int i;
1259 	struct c4iw_qp *qp;
1260 
1261 	for (i = 0; i < DB_FC_RESUME_SIZE; i++) {
1262 		qp = list_first_entry(&ctx->dev->db_fc_list, struct c4iw_qp,
1263 				      db_fc_entry);
1264 		list_del_init(&qp->db_fc_entry);
1265 		resume_rc_qp(qp);
1266 		if (list_empty(&ctx->dev->db_fc_list))
1267 			break;
1268 	}
1269 }
1270 
1271 static void resume_queues(struct uld_ctx *ctx)
1272 {
1273 	spin_lock_irq(&ctx->dev->lock);
1274 	if (ctx->dev->db_state != STOPPED)
1275 		goto out;
1276 	ctx->dev->db_state = FLOW_CONTROL;
1277 	while (1) {
1278 		if (list_empty(&ctx->dev->db_fc_list)) {
1279 			WARN_ON(ctx->dev->db_state != FLOW_CONTROL);
1280 			ctx->dev->db_state = NORMAL;
1281 			ctx->dev->rdev.stats.db_state_transitions++;
1282 			if (ctx->dev->rdev.flags & T4_STATUS_PAGE_DISABLED) {
1283 				idr_for_each(&ctx->dev->qpidr, enable_qp_db,
1284 					     NULL);
1285 			} else {
1286 				ctx->dev->rdev.status_page->db_off = 0;
1287 			}
1288 			break;
1289 		} else {
1290 			if (cxgb4_dbfifo_count(ctx->dev->rdev.lldi.ports[0], 1)
1291 			    < (ctx->dev->rdev.lldi.dbfifo_int_thresh <<
1292 			       DB_FC_DRAIN_THRESH)) {
1293 				resume_a_chunk(ctx);
1294 			}
1295 			if (!list_empty(&ctx->dev->db_fc_list)) {
1296 				spin_unlock_irq(&ctx->dev->lock);
1297 				if (DB_FC_RESUME_DELAY) {
1298 					set_current_state(TASK_UNINTERRUPTIBLE);
1299 					schedule_timeout(DB_FC_RESUME_DELAY);
1300 				}
1301 				spin_lock_irq(&ctx->dev->lock);
1302 				if (ctx->dev->db_state != FLOW_CONTROL)
1303 					break;
1304 			}
1305 		}
1306 	}
1307 out:
1308 	if (ctx->dev->db_state != NORMAL)
1309 		ctx->dev->rdev.stats.db_fc_interruptions++;
1310 	spin_unlock_irq(&ctx->dev->lock);
1311 }
1312 
1313 struct qp_list {
1314 	unsigned idx;
1315 	struct c4iw_qp **qps;
1316 };
1317 
1318 static int add_and_ref_qp(int id, void *p, void *data)
1319 {
1320 	struct qp_list *qp_listp = data;
1321 	struct c4iw_qp *qp = p;
1322 
1323 	c4iw_qp_add_ref(&qp->ibqp);
1324 	qp_listp->qps[qp_listp->idx++] = qp;
1325 	return 0;
1326 }
1327 
1328 static int count_qps(int id, void *p, void *data)
1329 {
1330 	unsigned *countp = data;
1331 	(*countp)++;
1332 	return 0;
1333 }
1334 
1335 static void deref_qps(struct qp_list *qp_list)
1336 {
1337 	int idx;
1338 
1339 	for (idx = 0; idx < qp_list->idx; idx++)
1340 		c4iw_qp_rem_ref(&qp_list->qps[idx]->ibqp);
1341 }
1342 
1343 static void recover_lost_dbs(struct uld_ctx *ctx, struct qp_list *qp_list)
1344 {
1345 	int idx;
1346 	int ret;
1347 
1348 	for (idx = 0; idx < qp_list->idx; idx++) {
1349 		struct c4iw_qp *qp = qp_list->qps[idx];
1350 
1351 		spin_lock_irq(&qp->rhp->lock);
1352 		spin_lock(&qp->lock);
1353 		ret = cxgb4_sync_txq_pidx(qp->rhp->rdev.lldi.ports[0],
1354 					  qp->wq.sq.qid,
1355 					  t4_sq_host_wq_pidx(&qp->wq),
1356 					  t4_sq_wq_size(&qp->wq));
1357 		if (ret) {
1358 			pr_err(KERN_ERR MOD "%s: Fatal error - "
1359 			       "DB overflow recovery failed - "
1360 			       "error syncing SQ qid %u\n",
1361 			       pci_name(ctx->lldi.pdev), qp->wq.sq.qid);
1362 			spin_unlock(&qp->lock);
1363 			spin_unlock_irq(&qp->rhp->lock);
1364 			return;
1365 		}
1366 		qp->wq.sq.wq_pidx_inc = 0;
1367 
1368 		ret = cxgb4_sync_txq_pidx(qp->rhp->rdev.lldi.ports[0],
1369 					  qp->wq.rq.qid,
1370 					  t4_rq_host_wq_pidx(&qp->wq),
1371 					  t4_rq_wq_size(&qp->wq));
1372 
1373 		if (ret) {
1374 			pr_err(KERN_ERR MOD "%s: Fatal error - "
1375 			       "DB overflow recovery failed - "
1376 			       "error syncing RQ qid %u\n",
1377 			       pci_name(ctx->lldi.pdev), qp->wq.rq.qid);
1378 			spin_unlock(&qp->lock);
1379 			spin_unlock_irq(&qp->rhp->lock);
1380 			return;
1381 		}
1382 		qp->wq.rq.wq_pidx_inc = 0;
1383 		spin_unlock(&qp->lock);
1384 		spin_unlock_irq(&qp->rhp->lock);
1385 
1386 		/* Wait for the dbfifo to drain */
1387 		while (cxgb4_dbfifo_count(qp->rhp->rdev.lldi.ports[0], 1) > 0) {
1388 			set_current_state(TASK_UNINTERRUPTIBLE);
1389 			schedule_timeout(usecs_to_jiffies(10));
1390 		}
1391 	}
1392 }
1393 
1394 static void recover_queues(struct uld_ctx *ctx)
1395 {
1396 	int count = 0;
1397 	struct qp_list qp_list;
1398 	int ret;
1399 
1400 	/* slow everybody down */
1401 	set_current_state(TASK_UNINTERRUPTIBLE);
1402 	schedule_timeout(usecs_to_jiffies(1000));
1403 
1404 	/* flush the SGE contexts */
1405 	ret = cxgb4_flush_eq_cache(ctx->dev->rdev.lldi.ports[0]);
1406 	if (ret) {
1407 		printk(KERN_ERR MOD "%s: Fatal error - DB overflow recovery failed\n",
1408 		       pci_name(ctx->lldi.pdev));
1409 		return;
1410 	}
1411 
1412 	/* Count active queues so we can build a list of queues to recover */
1413 	spin_lock_irq(&ctx->dev->lock);
1414 	WARN_ON(ctx->dev->db_state != STOPPED);
1415 	ctx->dev->db_state = RECOVERY;
1416 	idr_for_each(&ctx->dev->qpidr, count_qps, &count);
1417 
1418 	qp_list.qps = kzalloc(count * sizeof *qp_list.qps, GFP_ATOMIC);
1419 	if (!qp_list.qps) {
1420 		printk(KERN_ERR MOD "%s: Fatal error - DB overflow recovery failed\n",
1421 		       pci_name(ctx->lldi.pdev));
1422 		spin_unlock_irq(&ctx->dev->lock);
1423 		return;
1424 	}
1425 	qp_list.idx = 0;
1426 
1427 	/* add and ref each qp so it doesn't get freed */
1428 	idr_for_each(&ctx->dev->qpidr, add_and_ref_qp, &qp_list);
1429 
1430 	spin_unlock_irq(&ctx->dev->lock);
1431 
1432 	/* now traverse the list in a safe context to recover the db state*/
1433 	recover_lost_dbs(ctx, &qp_list);
1434 
1435 	/* we're almost done!  deref the qps and clean up */
1436 	deref_qps(&qp_list);
1437 	kfree(qp_list.qps);
1438 
1439 	spin_lock_irq(&ctx->dev->lock);
1440 	WARN_ON(ctx->dev->db_state != RECOVERY);
1441 	ctx->dev->db_state = STOPPED;
1442 	spin_unlock_irq(&ctx->dev->lock);
1443 }
1444 
1445 static int c4iw_uld_control(void *handle, enum cxgb4_control control, ...)
1446 {
1447 	struct uld_ctx *ctx = handle;
1448 
1449 	switch (control) {
1450 	case CXGB4_CONTROL_DB_FULL:
1451 		stop_queues(ctx);
1452 		ctx->dev->rdev.stats.db_full++;
1453 		break;
1454 	case CXGB4_CONTROL_DB_EMPTY:
1455 		resume_queues(ctx);
1456 		mutex_lock(&ctx->dev->rdev.stats.lock);
1457 		ctx->dev->rdev.stats.db_empty++;
1458 		mutex_unlock(&ctx->dev->rdev.stats.lock);
1459 		break;
1460 	case CXGB4_CONTROL_DB_DROP:
1461 		recover_queues(ctx);
1462 		mutex_lock(&ctx->dev->rdev.stats.lock);
1463 		ctx->dev->rdev.stats.db_drop++;
1464 		mutex_unlock(&ctx->dev->rdev.stats.lock);
1465 		break;
1466 	default:
1467 		printk(KERN_WARNING MOD "%s: unknown control cmd %u\n",
1468 		       pci_name(ctx->lldi.pdev), control);
1469 		break;
1470 	}
1471 	return 0;
1472 }
1473 
1474 static struct cxgb4_uld_info c4iw_uld_info = {
1475 	.name = DRV_NAME,
1476 	.add = c4iw_uld_add,
1477 	.rx_handler = c4iw_uld_rx_handler,
1478 	.state_change = c4iw_uld_state_change,
1479 	.control = c4iw_uld_control,
1480 };
1481 
1482 static int __init c4iw_init_module(void)
1483 {
1484 	int err;
1485 
1486 	err = c4iw_cm_init();
1487 	if (err)
1488 		return err;
1489 
1490 	c4iw_debugfs_root = debugfs_create_dir(DRV_NAME, NULL);
1491 	if (!c4iw_debugfs_root)
1492 		printk(KERN_WARNING MOD
1493 		       "could not create debugfs entry, continuing\n");
1494 
1495 	if (ibnl_add_client(RDMA_NL_C4IW, RDMA_NL_IWPM_NUM_OPS,
1496 			    c4iw_nl_cb_table))
1497 		pr_err("%s[%u]: Failed to add netlink callback\n"
1498 		       , __func__, __LINE__);
1499 
1500 	err = iwpm_init(RDMA_NL_C4IW);
1501 	if (err) {
1502 		pr_err("port mapper initialization failed with %d\n", err);
1503 		ibnl_remove_client(RDMA_NL_C4IW);
1504 		c4iw_cm_term();
1505 		debugfs_remove_recursive(c4iw_debugfs_root);
1506 		return err;
1507 	}
1508 
1509 	cxgb4_register_uld(CXGB4_ULD_RDMA, &c4iw_uld_info);
1510 
1511 	return 0;
1512 }
1513 
1514 static void __exit c4iw_exit_module(void)
1515 {
1516 	struct uld_ctx *ctx, *tmp;
1517 
1518 	mutex_lock(&dev_mutex);
1519 	list_for_each_entry_safe(ctx, tmp, &uld_ctx_list, entry) {
1520 		if (ctx->dev)
1521 			c4iw_remove(ctx);
1522 		kfree(ctx);
1523 	}
1524 	mutex_unlock(&dev_mutex);
1525 	cxgb4_unregister_uld(CXGB4_ULD_RDMA);
1526 	iwpm_exit(RDMA_NL_C4IW);
1527 	ibnl_remove_client(RDMA_NL_C4IW);
1528 	c4iw_cm_term();
1529 	debugfs_remove_recursive(c4iw_debugfs_root);
1530 }
1531 
1532 module_init(c4iw_init_module);
1533 module_exit(c4iw_exit_module);
1534