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
44MODULE_AUTHOR("Steve Wise");
45MODULE_DESCRIPTION("Chelsio T4/T5 RDMA Driver");
46MODULE_LICENSE("Dual BSD/GPL");
47MODULE_VERSION(DRV_VERSION);
48
49static int allow_db_fc_on_t5;
50module_param(allow_db_fc_on_t5, int, 0644);
51MODULE_PARM_DESC(allow_db_fc_on_t5,
52		 "Allow DB Flow Control on T5 (default = 0)");
53
54static int allow_db_coalescing_on_t5;
55module_param(allow_db_coalescing_on_t5, int, 0644);
56MODULE_PARM_DESC(allow_db_coalescing_on_t5,
57		 "Allow DB Coalescing on T5 (default = 0)");
58
59int c4iw_wr_log = 0;
60module_param(c4iw_wr_log, int, 0444);
61MODULE_PARM_DESC(c4iw_wr_log, "Enables logging of work request timing data.");
62
63static int c4iw_wr_log_size_order = 12;
64module_param(c4iw_wr_log_size_order, int, 0444);
65MODULE_PARM_DESC(c4iw_wr_log_size_order,
66		 "Number of entries (log2) in the work request timing log.");
67
68struct uld_ctx {
69	struct list_head entry;
70	struct cxgb4_lld_info lldi;
71	struct c4iw_dev *dev;
72};
73
74static LIST_HEAD(uld_ctx_list);
75static 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
81static struct dentry *c4iw_debugfs_root;
82
83struct c4iw_debugfs_data {
84	struct c4iw_dev *devp;
85	char *buf;
86	int bufsize;
87	int pos;
88};
89
90/* registered cxgb4 netlink callbacks */
91static 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_REMOTE_INFO] = {.dump = iwpm_remote_info_cb},
97	[RDMA_NL_IWPM_MAPINFO] = {.dump = iwpm_mapping_info_cb},
98	[RDMA_NL_IWPM_MAPINFO_NUM] = {.dump = iwpm_ack_mapping_info_cb}
99};
100
101static int count_idrs(int id, void *p, void *data)
102{
103	int *countp = data;
104
105	*countp = *countp + 1;
106	return 0;
107}
108
109static ssize_t debugfs_read(struct file *file, char __user *buf, size_t count,
110			    loff_t *ppos)
111{
112	struct c4iw_debugfs_data *d = file->private_data;
113
114	return simple_read_from_buffer(buf, count, ppos, d->buf, d->pos);
115}
116
117void c4iw_log_wr_stats(struct t4_wq *wq, struct t4_cqe *cqe)
118{
119	struct wr_log_entry le;
120	int idx;
121
122	if (!wq->rdev->wr_log)
123		return;
124
125	idx = (atomic_inc_return(&wq->rdev->wr_log_idx) - 1) &
126		(wq->rdev->wr_log_size - 1);
127	le.poll_sge_ts = cxgb4_read_sge_timestamp(wq->rdev->lldi.ports[0]);
128	getnstimeofday(&le.poll_host_ts);
129	le.valid = 1;
130	le.cqe_sge_ts = CQE_TS(cqe);
131	if (SQ_TYPE(cqe)) {
132		le.qid = wq->sq.qid;
133		le.opcode = CQE_OPCODE(cqe);
134		le.post_host_ts = wq->sq.sw_sq[wq->sq.cidx].host_ts;
135		le.post_sge_ts = wq->sq.sw_sq[wq->sq.cidx].sge_ts;
136		le.wr_id = CQE_WRID_SQ_IDX(cqe);
137	} else {
138		le.qid = wq->rq.qid;
139		le.opcode = FW_RI_RECEIVE;
140		le.post_host_ts = wq->rq.sw_rq[wq->rq.cidx].host_ts;
141		le.post_sge_ts = wq->rq.sw_rq[wq->rq.cidx].sge_ts;
142		le.wr_id = CQE_WRID_MSN(cqe);
143	}
144	wq->rdev->wr_log[idx] = le;
145}
146
147static int wr_log_show(struct seq_file *seq, void *v)
148{
149	struct c4iw_dev *dev = seq->private;
150	struct timespec prev_ts = {0, 0};
151	struct wr_log_entry *lep;
152	int prev_ts_set = 0;
153	int idx, end;
154
155#define ts2ns(ts) div64_u64((ts) * dev->rdev.lldi.cclk_ps, 1000)
156
157	idx = atomic_read(&dev->rdev.wr_log_idx) &
158		(dev->rdev.wr_log_size - 1);
159	end = idx - 1;
160	if (end < 0)
161		end = dev->rdev.wr_log_size - 1;
162	lep = &dev->rdev.wr_log[idx];
163	while (idx != end) {
164		if (lep->valid) {
165			if (!prev_ts_set) {
166				prev_ts_set = 1;
167				prev_ts = lep->poll_host_ts;
168			}
169			seq_printf(seq, "%04u: sec %lu nsec %lu qid %u opcode "
170				   "%u %s 0x%x host_wr_delta sec %lu nsec %lu "
171				   "post_sge_ts 0x%llx cqe_sge_ts 0x%llx "
172				   "poll_sge_ts 0x%llx post_poll_delta_ns %llu "
173				   "cqe_poll_delta_ns %llu\n",
174				   idx,
175				   timespec_sub(lep->poll_host_ts,
176						prev_ts).tv_sec,
177				   timespec_sub(lep->poll_host_ts,
178						prev_ts).tv_nsec,
179				   lep->qid, lep->opcode,
180				   lep->opcode == FW_RI_RECEIVE ?
181							"msn" : "wrid",
182				   lep->wr_id,
183				   timespec_sub(lep->poll_host_ts,
184						lep->post_host_ts).tv_sec,
185				   timespec_sub(lep->poll_host_ts,
186						lep->post_host_ts).tv_nsec,
187				   lep->post_sge_ts, lep->cqe_sge_ts,
188				   lep->poll_sge_ts,
189				   ts2ns(lep->poll_sge_ts - lep->post_sge_ts),
190				   ts2ns(lep->poll_sge_ts - lep->cqe_sge_ts));
191			prev_ts = lep->poll_host_ts;
192		}
193		idx++;
194		if (idx > (dev->rdev.wr_log_size - 1))
195			idx = 0;
196		lep = &dev->rdev.wr_log[idx];
197	}
198#undef ts2ns
199	return 0;
200}
201
202static int wr_log_open(struct inode *inode, struct file *file)
203{
204	return single_open(file, wr_log_show, inode->i_private);
205}
206
207static ssize_t wr_log_clear(struct file *file, const char __user *buf,
208			    size_t count, loff_t *pos)
209{
210	struct c4iw_dev *dev = ((struct seq_file *)file->private_data)->private;
211	int i;
212
213	if (dev->rdev.wr_log)
214		for (i = 0; i < dev->rdev.wr_log_size; i++)
215			dev->rdev.wr_log[i].valid = 0;
216	return count;
217}
218
219static const struct file_operations wr_log_debugfs_fops = {
220	.owner   = THIS_MODULE,
221	.open    = wr_log_open,
222	.release = single_release,
223	.read    = seq_read,
224	.llseek  = seq_lseek,
225	.write   = wr_log_clear,
226};
227
228static int dump_qp(int id, void *p, void *data)
229{
230	struct c4iw_qp *qp = p;
231	struct c4iw_debugfs_data *qpd = data;
232	int space;
233	int cc;
234
235	if (id != qp->wq.sq.qid)
236		return 0;
237
238	space = qpd->bufsize - qpd->pos - 1;
239	if (space == 0)
240		return 1;
241
242	if (qp->ep) {
243		if (qp->ep->com.local_addr.ss_family == AF_INET) {
244			struct sockaddr_in *lsin = (struct sockaddr_in *)
245				&qp->ep->com.local_addr;
246			struct sockaddr_in *rsin = (struct sockaddr_in *)
247				&qp->ep->com.remote_addr;
248			struct sockaddr_in *mapped_lsin = (struct sockaddr_in *)
249				&qp->ep->com.mapped_local_addr;
250			struct sockaddr_in *mapped_rsin = (struct sockaddr_in *)
251				&qp->ep->com.mapped_remote_addr;
252
253			cc = snprintf(qpd->buf + qpd->pos, space,
254				      "rc qp sq id %u rq id %u state %u "
255				      "onchip %u ep tid %u state %u "
256				      "%pI4:%u/%u->%pI4:%u/%u\n",
257				      qp->wq.sq.qid, qp->wq.rq.qid,
258				      (int)qp->attr.state,
259				      qp->wq.sq.flags & T4_SQ_ONCHIP,
260				      qp->ep->hwtid, (int)qp->ep->com.state,
261				      &lsin->sin_addr, ntohs(lsin->sin_port),
262				      ntohs(mapped_lsin->sin_port),
263				      &rsin->sin_addr, ntohs(rsin->sin_port),
264				      ntohs(mapped_rsin->sin_port));
265		} else {
266			struct sockaddr_in6 *lsin6 = (struct sockaddr_in6 *)
267				&qp->ep->com.local_addr;
268			struct sockaddr_in6 *rsin6 = (struct sockaddr_in6 *)
269				&qp->ep->com.remote_addr;
270			struct sockaddr_in6 *mapped_lsin6 =
271				(struct sockaddr_in6 *)
272				&qp->ep->com.mapped_local_addr;
273			struct sockaddr_in6 *mapped_rsin6 =
274				(struct sockaddr_in6 *)
275				&qp->ep->com.mapped_remote_addr;
276
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				      "%pI6:%u/%u->%pI6:%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				      qp->ep->hwtid, (int)qp->ep->com.state,
285				      &lsin6->sin6_addr,
286				      ntohs(lsin6->sin6_port),
287				      ntohs(mapped_lsin6->sin6_port),
288				      &rsin6->sin6_addr,
289				      ntohs(rsin6->sin6_port),
290				      ntohs(mapped_rsin6->sin6_port));
291		}
292	} else
293		cc = snprintf(qpd->buf + qpd->pos, space,
294			     "qp sq id %u rq id %u state %u onchip %u\n",
295			      qp->wq.sq.qid, qp->wq.rq.qid,
296			      (int)qp->attr.state,
297			      qp->wq.sq.flags & T4_SQ_ONCHIP);
298	if (cc < space)
299		qpd->pos += cc;
300	return 0;
301}
302
303static int qp_release(struct inode *inode, struct file *file)
304{
305	struct c4iw_debugfs_data *qpd = file->private_data;
306	if (!qpd) {
307		printk(KERN_INFO "%s null qpd?\n", __func__);
308		return 0;
309	}
310	vfree(qpd->buf);
311	kfree(qpd);
312	return 0;
313}
314
315static int qp_open(struct inode *inode, struct file *file)
316{
317	struct c4iw_debugfs_data *qpd;
318	int ret = 0;
319	int count = 1;
320
321	qpd = kmalloc(sizeof *qpd, GFP_KERNEL);
322	if (!qpd) {
323		ret = -ENOMEM;
324		goto out;
325	}
326	qpd->devp = inode->i_private;
327	qpd->pos = 0;
328
329	spin_lock_irq(&qpd->devp->lock);
330	idr_for_each(&qpd->devp->qpidr, count_idrs, &count);
331	spin_unlock_irq(&qpd->devp->lock);
332
333	qpd->bufsize = count * 128;
334	qpd->buf = vmalloc(qpd->bufsize);
335	if (!qpd->buf) {
336		ret = -ENOMEM;
337		goto err1;
338	}
339
340	spin_lock_irq(&qpd->devp->lock);
341	idr_for_each(&qpd->devp->qpidr, dump_qp, qpd);
342	spin_unlock_irq(&qpd->devp->lock);
343
344	qpd->buf[qpd->pos++] = 0;
345	file->private_data = qpd;
346	goto out;
347err1:
348	kfree(qpd);
349out:
350	return ret;
351}
352
353static const struct file_operations qp_debugfs_fops = {
354	.owner   = THIS_MODULE,
355	.open    = qp_open,
356	.release = qp_release,
357	.read    = debugfs_read,
358	.llseek  = default_llseek,
359};
360
361static int dump_stag(int id, void *p, void *data)
362{
363	struct c4iw_debugfs_data *stagd = data;
364	int space;
365	int cc;
366	struct fw_ri_tpte tpte;
367	int ret;
368
369	space = stagd->bufsize - stagd->pos - 1;
370	if (space == 0)
371		return 1;
372
373	ret = cxgb4_read_tpte(stagd->devp->rdev.lldi.ports[0], (u32)id<<8,
374			      (__be32 *)&tpte);
375	if (ret) {
376		dev_err(&stagd->devp->rdev.lldi.pdev->dev,
377			"%s cxgb4_read_tpte err %d\n", __func__, ret);
378		return ret;
379	}
380	cc = snprintf(stagd->buf + stagd->pos, space,
381		      "stag: idx 0x%x valid %d key 0x%x state %d pdid %d "
382		      "perm 0x%x ps %d len 0x%llx va 0x%llx\n",
383		      (u32)id<<8,
384		      FW_RI_TPTE_VALID_G(ntohl(tpte.valid_to_pdid)),
385		      FW_RI_TPTE_STAGKEY_G(ntohl(tpte.valid_to_pdid)),
386		      FW_RI_TPTE_STAGSTATE_G(ntohl(tpte.valid_to_pdid)),
387		      FW_RI_TPTE_PDID_G(ntohl(tpte.valid_to_pdid)),
388		      FW_RI_TPTE_PERM_G(ntohl(tpte.locread_to_qpid)),
389		      FW_RI_TPTE_PS_G(ntohl(tpte.locread_to_qpid)),
390		      ((u64)ntohl(tpte.len_hi) << 32) | ntohl(tpte.len_lo),
391		      ((u64)ntohl(tpte.va_hi) << 32) | ntohl(tpte.va_lo_fbo));
392	if (cc < space)
393		stagd->pos += cc;
394	return 0;
395}
396
397static int stag_release(struct inode *inode, struct file *file)
398{
399	struct c4iw_debugfs_data *stagd = file->private_data;
400	if (!stagd) {
401		printk(KERN_INFO "%s null stagd?\n", __func__);
402		return 0;
403	}
404	vfree(stagd->buf);
405	kfree(stagd);
406	return 0;
407}
408
409static int stag_open(struct inode *inode, struct file *file)
410{
411	struct c4iw_debugfs_data *stagd;
412	int ret = 0;
413	int count = 1;
414
415	stagd = kmalloc(sizeof *stagd, GFP_KERNEL);
416	if (!stagd) {
417		ret = -ENOMEM;
418		goto out;
419	}
420	stagd->devp = inode->i_private;
421	stagd->pos = 0;
422
423	spin_lock_irq(&stagd->devp->lock);
424	idr_for_each(&stagd->devp->mmidr, count_idrs, &count);
425	spin_unlock_irq(&stagd->devp->lock);
426
427	stagd->bufsize = count * 256;
428	stagd->buf = vmalloc(stagd->bufsize);
429	if (!stagd->buf) {
430		ret = -ENOMEM;
431		goto err1;
432	}
433
434	spin_lock_irq(&stagd->devp->lock);
435	idr_for_each(&stagd->devp->mmidr, dump_stag, stagd);
436	spin_unlock_irq(&stagd->devp->lock);
437
438	stagd->buf[stagd->pos++] = 0;
439	file->private_data = stagd;
440	goto out;
441err1:
442	kfree(stagd);
443out:
444	return ret;
445}
446
447static const struct file_operations stag_debugfs_fops = {
448	.owner   = THIS_MODULE,
449	.open    = stag_open,
450	.release = stag_release,
451	.read    = debugfs_read,
452	.llseek  = default_llseek,
453};
454
455static char *db_state_str[] = {"NORMAL", "FLOW_CONTROL", "RECOVERY", "STOPPED"};
456
457static int stats_show(struct seq_file *seq, void *v)
458{
459	struct c4iw_dev *dev = seq->private;
460
461	seq_printf(seq, "   Object: %10s %10s %10s %10s\n", "Total", "Current",
462		   "Max", "Fail");
463	seq_printf(seq, "     PDID: %10llu %10llu %10llu %10llu\n",
464			dev->rdev.stats.pd.total, dev->rdev.stats.pd.cur,
465			dev->rdev.stats.pd.max, dev->rdev.stats.pd.fail);
466	seq_printf(seq, "      QID: %10llu %10llu %10llu %10llu\n",
467			dev->rdev.stats.qid.total, dev->rdev.stats.qid.cur,
468			dev->rdev.stats.qid.max, dev->rdev.stats.qid.fail);
469	seq_printf(seq, "   TPTMEM: %10llu %10llu %10llu %10llu\n",
470			dev->rdev.stats.stag.total, dev->rdev.stats.stag.cur,
471			dev->rdev.stats.stag.max, dev->rdev.stats.stag.fail);
472	seq_printf(seq, "   PBLMEM: %10llu %10llu %10llu %10llu\n",
473			dev->rdev.stats.pbl.total, dev->rdev.stats.pbl.cur,
474			dev->rdev.stats.pbl.max, dev->rdev.stats.pbl.fail);
475	seq_printf(seq, "   RQTMEM: %10llu %10llu %10llu %10llu\n",
476			dev->rdev.stats.rqt.total, dev->rdev.stats.rqt.cur,
477			dev->rdev.stats.rqt.max, dev->rdev.stats.rqt.fail);
478	seq_printf(seq, "  OCQPMEM: %10llu %10llu %10llu %10llu\n",
479			dev->rdev.stats.ocqp.total, dev->rdev.stats.ocqp.cur,
480			dev->rdev.stats.ocqp.max, dev->rdev.stats.ocqp.fail);
481	seq_printf(seq, "  DB FULL: %10llu\n", dev->rdev.stats.db_full);
482	seq_printf(seq, " DB EMPTY: %10llu\n", dev->rdev.stats.db_empty);
483	seq_printf(seq, "  DB DROP: %10llu\n", dev->rdev.stats.db_drop);
484	seq_printf(seq, " DB State: %s Transitions %llu FC Interruptions %llu\n",
485		   db_state_str[dev->db_state],
486		   dev->rdev.stats.db_state_transitions,
487		   dev->rdev.stats.db_fc_interruptions);
488	seq_printf(seq, "TCAM_FULL: %10llu\n", dev->rdev.stats.tcam_full);
489	seq_printf(seq, "ACT_OFLD_CONN_FAILS: %10llu\n",
490		   dev->rdev.stats.act_ofld_conn_fails);
491	seq_printf(seq, "PAS_OFLD_CONN_FAILS: %10llu\n",
492		   dev->rdev.stats.pas_ofld_conn_fails);
493	seq_printf(seq, "NEG_ADV_RCVD: %10llu\n", dev->rdev.stats.neg_adv);
494	seq_printf(seq, "AVAILABLE IRD: %10u\n", dev->avail_ird);
495	return 0;
496}
497
498static int stats_open(struct inode *inode, struct file *file)
499{
500	return single_open(file, stats_show, inode->i_private);
501}
502
503static ssize_t stats_clear(struct file *file, const char __user *buf,
504		size_t count, loff_t *pos)
505{
506	struct c4iw_dev *dev = ((struct seq_file *)file->private_data)->private;
507
508	mutex_lock(&dev->rdev.stats.lock);
509	dev->rdev.stats.pd.max = 0;
510	dev->rdev.stats.pd.fail = 0;
511	dev->rdev.stats.qid.max = 0;
512	dev->rdev.stats.qid.fail = 0;
513	dev->rdev.stats.stag.max = 0;
514	dev->rdev.stats.stag.fail = 0;
515	dev->rdev.stats.pbl.max = 0;
516	dev->rdev.stats.pbl.fail = 0;
517	dev->rdev.stats.rqt.max = 0;
518	dev->rdev.stats.rqt.fail = 0;
519	dev->rdev.stats.ocqp.max = 0;
520	dev->rdev.stats.ocqp.fail = 0;
521	dev->rdev.stats.db_full = 0;
522	dev->rdev.stats.db_empty = 0;
523	dev->rdev.stats.db_drop = 0;
524	dev->rdev.stats.db_state_transitions = 0;
525	dev->rdev.stats.tcam_full = 0;
526	dev->rdev.stats.act_ofld_conn_fails = 0;
527	dev->rdev.stats.pas_ofld_conn_fails = 0;
528	mutex_unlock(&dev->rdev.stats.lock);
529	return count;
530}
531
532static const struct file_operations stats_debugfs_fops = {
533	.owner   = THIS_MODULE,
534	.open    = stats_open,
535	.release = single_release,
536	.read    = seq_read,
537	.llseek  = seq_lseek,
538	.write   = stats_clear,
539};
540
541static int dump_ep(int id, void *p, void *data)
542{
543	struct c4iw_ep *ep = p;
544	struct c4iw_debugfs_data *epd = data;
545	int space;
546	int cc;
547
548	space = epd->bufsize - epd->pos - 1;
549	if (space == 0)
550		return 1;
551
552	if (ep->com.local_addr.ss_family == AF_INET) {
553		struct sockaddr_in *lsin = (struct sockaddr_in *)
554			&ep->com.local_addr;
555		struct sockaddr_in *rsin = (struct sockaddr_in *)
556			&ep->com.remote_addr;
557		struct sockaddr_in *mapped_lsin = (struct sockaddr_in *)
558			&ep->com.mapped_local_addr;
559		struct sockaddr_in *mapped_rsin = (struct sockaddr_in *)
560			&ep->com.mapped_remote_addr;
561
562		cc = snprintf(epd->buf + epd->pos, space,
563			      "ep %p cm_id %p qp %p state %d flags 0x%lx "
564			      "history 0x%lx hwtid %d atid %d "
565			      "conn_na %u abort_na %u "
566			      "%pI4:%d/%d <-> %pI4:%d/%d\n",
567			      ep, ep->com.cm_id, ep->com.qp,
568			      (int)ep->com.state, ep->com.flags,
569			      ep->com.history, ep->hwtid, ep->atid,
570			      ep->stats.connect_neg_adv,
571			      ep->stats.abort_neg_adv,
572			      &lsin->sin_addr, ntohs(lsin->sin_port),
573			      ntohs(mapped_lsin->sin_port),
574			      &rsin->sin_addr, ntohs(rsin->sin_port),
575			      ntohs(mapped_rsin->sin_port));
576	} else {
577		struct sockaddr_in6 *lsin6 = (struct sockaddr_in6 *)
578			&ep->com.local_addr;
579		struct sockaddr_in6 *rsin6 = (struct sockaddr_in6 *)
580			&ep->com.remote_addr;
581		struct sockaddr_in6 *mapped_lsin6 = (struct sockaddr_in6 *)
582			&ep->com.mapped_local_addr;
583		struct sockaddr_in6 *mapped_rsin6 = (struct sockaddr_in6 *)
584			&ep->com.mapped_remote_addr;
585
586		cc = snprintf(epd->buf + epd->pos, space,
587			      "ep %p cm_id %p qp %p state %d flags 0x%lx "
588			      "history 0x%lx hwtid %d atid %d "
589			      "conn_na %u abort_na %u "
590			      "%pI6:%d/%d <-> %pI6:%d/%d\n",
591			      ep, ep->com.cm_id, ep->com.qp,
592			      (int)ep->com.state, ep->com.flags,
593			      ep->com.history, ep->hwtid, ep->atid,
594			      ep->stats.connect_neg_adv,
595			      ep->stats.abort_neg_adv,
596			      &lsin6->sin6_addr, ntohs(lsin6->sin6_port),
597			      ntohs(mapped_lsin6->sin6_port),
598			      &rsin6->sin6_addr, ntohs(rsin6->sin6_port),
599			      ntohs(mapped_rsin6->sin6_port));
600	}
601	if (cc < space)
602		epd->pos += cc;
603	return 0;
604}
605
606static int dump_listen_ep(int id, void *p, void *data)
607{
608	struct c4iw_listen_ep *ep = p;
609	struct c4iw_debugfs_data *epd = data;
610	int space;
611	int cc;
612
613	space = epd->bufsize - epd->pos - 1;
614	if (space == 0)
615		return 1;
616
617	if (ep->com.local_addr.ss_family == AF_INET) {
618		struct sockaddr_in *lsin = (struct sockaddr_in *)
619			&ep->com.local_addr;
620		struct sockaddr_in *mapped_lsin = (struct sockaddr_in *)
621			&ep->com.mapped_local_addr;
622
623		cc = snprintf(epd->buf + epd->pos, space,
624			      "ep %p cm_id %p state %d flags 0x%lx stid %d "
625			      "backlog %d %pI4:%d/%d\n",
626			      ep, ep->com.cm_id, (int)ep->com.state,
627			      ep->com.flags, ep->stid, ep->backlog,
628			      &lsin->sin_addr, ntohs(lsin->sin_port),
629			      ntohs(mapped_lsin->sin_port));
630	} else {
631		struct sockaddr_in6 *lsin6 = (struct sockaddr_in6 *)
632			&ep->com.local_addr;
633		struct sockaddr_in6 *mapped_lsin6 = (struct sockaddr_in6 *)
634			&ep->com.mapped_local_addr;
635
636		cc = snprintf(epd->buf + epd->pos, space,
637			      "ep %p cm_id %p state %d flags 0x%lx stid %d "
638			      "backlog %d %pI6:%d/%d\n",
639			      ep, ep->com.cm_id, (int)ep->com.state,
640			      ep->com.flags, ep->stid, ep->backlog,
641			      &lsin6->sin6_addr, ntohs(lsin6->sin6_port),
642			      ntohs(mapped_lsin6->sin6_port));
643	}
644	if (cc < space)
645		epd->pos += cc;
646	return 0;
647}
648
649static int ep_release(struct inode *inode, struct file *file)
650{
651	struct c4iw_debugfs_data *epd = file->private_data;
652	if (!epd) {
653		pr_info("%s null qpd?\n", __func__);
654		return 0;
655	}
656	vfree(epd->buf);
657	kfree(epd);
658	return 0;
659}
660
661static int ep_open(struct inode *inode, struct file *file)
662{
663	struct c4iw_debugfs_data *epd;
664	int ret = 0;
665	int count = 1;
666
667	epd = kmalloc(sizeof(*epd), GFP_KERNEL);
668	if (!epd) {
669		ret = -ENOMEM;
670		goto out;
671	}
672	epd->devp = inode->i_private;
673	epd->pos = 0;
674
675	spin_lock_irq(&epd->devp->lock);
676	idr_for_each(&epd->devp->hwtid_idr, count_idrs, &count);
677	idr_for_each(&epd->devp->atid_idr, count_idrs, &count);
678	idr_for_each(&epd->devp->stid_idr, count_idrs, &count);
679	spin_unlock_irq(&epd->devp->lock);
680
681	epd->bufsize = count * 240;
682	epd->buf = vmalloc(epd->bufsize);
683	if (!epd->buf) {
684		ret = -ENOMEM;
685		goto err1;
686	}
687
688	spin_lock_irq(&epd->devp->lock);
689	idr_for_each(&epd->devp->hwtid_idr, dump_ep, epd);
690	idr_for_each(&epd->devp->atid_idr, dump_ep, epd);
691	idr_for_each(&epd->devp->stid_idr, dump_listen_ep, epd);
692	spin_unlock_irq(&epd->devp->lock);
693
694	file->private_data = epd;
695	goto out;
696err1:
697	kfree(epd);
698out:
699	return ret;
700}
701
702static const struct file_operations ep_debugfs_fops = {
703	.owner   = THIS_MODULE,
704	.open    = ep_open,
705	.release = ep_release,
706	.read    = debugfs_read,
707};
708
709static int setup_debugfs(struct c4iw_dev *devp)
710{
711	if (!devp->debugfs_root)
712		return -1;
713
714	debugfs_create_file_size("qps", S_IWUSR, devp->debugfs_root,
715				 (void *)devp, &qp_debugfs_fops, 4096);
716
717	debugfs_create_file_size("stags", S_IWUSR, devp->debugfs_root,
718				 (void *)devp, &stag_debugfs_fops, 4096);
719
720	debugfs_create_file_size("stats", S_IWUSR, devp->debugfs_root,
721				 (void *)devp, &stats_debugfs_fops, 4096);
722
723	debugfs_create_file_size("eps", S_IWUSR, devp->debugfs_root,
724				 (void *)devp, &ep_debugfs_fops, 4096);
725
726	if (c4iw_wr_log)
727		debugfs_create_file_size("wr_log", S_IWUSR, devp->debugfs_root,
728					 (void *)devp, &wr_log_debugfs_fops, 4096);
729	return 0;
730}
731
732void c4iw_release_dev_ucontext(struct c4iw_rdev *rdev,
733			       struct c4iw_dev_ucontext *uctx)
734{
735	struct list_head *pos, *nxt;
736	struct c4iw_qid_list *entry;
737
738	mutex_lock(&uctx->lock);
739	list_for_each_safe(pos, nxt, &uctx->qpids) {
740		entry = list_entry(pos, struct c4iw_qid_list, entry);
741		list_del_init(&entry->entry);
742		if (!(entry->qid & rdev->qpmask)) {
743			c4iw_put_resource(&rdev->resource.qid_table,
744					  entry->qid);
745			mutex_lock(&rdev->stats.lock);
746			rdev->stats.qid.cur -= rdev->qpmask + 1;
747			mutex_unlock(&rdev->stats.lock);
748		}
749		kfree(entry);
750	}
751
752	list_for_each_safe(pos, nxt, &uctx->qpids) {
753		entry = list_entry(pos, struct c4iw_qid_list, entry);
754		list_del_init(&entry->entry);
755		kfree(entry);
756	}
757	mutex_unlock(&uctx->lock);
758}
759
760void c4iw_init_dev_ucontext(struct c4iw_rdev *rdev,
761			    struct c4iw_dev_ucontext *uctx)
762{
763	INIT_LIST_HEAD(&uctx->qpids);
764	INIT_LIST_HEAD(&uctx->cqids);
765	mutex_init(&uctx->lock);
766}
767
768/* Caller takes care of locking if needed */
769static int c4iw_rdev_open(struct c4iw_rdev *rdev)
770{
771	int err;
772
773	c4iw_init_dev_ucontext(rdev, &rdev->uctx);
774
775	/*
776	 * This implementation assumes udb_density == ucq_density!  Eventually
777	 * we might need to support this but for now fail the open. Also the
778	 * cqid and qpid range must match for now.
779	 */
780	if (rdev->lldi.udb_density != rdev->lldi.ucq_density) {
781		pr_err(MOD "%s: unsupported udb/ucq densities %u/%u\n",
782		       pci_name(rdev->lldi.pdev), rdev->lldi.udb_density,
783		       rdev->lldi.ucq_density);
784		err = -EINVAL;
785		goto err1;
786	}
787	if (rdev->lldi.vr->qp.start != rdev->lldi.vr->cq.start ||
788	    rdev->lldi.vr->qp.size != rdev->lldi.vr->cq.size) {
789		pr_err(MOD "%s: unsupported qp and cq id ranges "
790		       "qp start %u size %u cq start %u size %u\n",
791		       pci_name(rdev->lldi.pdev), rdev->lldi.vr->qp.start,
792		       rdev->lldi.vr->qp.size, rdev->lldi.vr->cq.size,
793		       rdev->lldi.vr->cq.size);
794		err = -EINVAL;
795		goto err1;
796	}
797
798	/*
799	 * qpshift is the number of bits to shift the qpid left in order
800	 * to get the correct address of the doorbell for that qp.
801	 */
802	rdev->qpshift = PAGE_SHIFT - ilog2(rdev->lldi.udb_density);
803	rdev->qpmask = rdev->lldi.udb_density - 1;
804	rdev->cqshift = PAGE_SHIFT - ilog2(rdev->lldi.ucq_density);
805	rdev->cqmask = rdev->lldi.ucq_density - 1;
806	PDBG("%s dev %s stag start 0x%0x size 0x%0x num stags %d "
807	     "pbl start 0x%0x size 0x%0x rq start 0x%0x size 0x%0x "
808	     "qp qid start %u size %u cq qid start %u size %u\n",
809	     __func__, pci_name(rdev->lldi.pdev), rdev->lldi.vr->stag.start,
810	     rdev->lldi.vr->stag.size, c4iw_num_stags(rdev),
811	     rdev->lldi.vr->pbl.start,
812	     rdev->lldi.vr->pbl.size, rdev->lldi.vr->rq.start,
813	     rdev->lldi.vr->rq.size,
814	     rdev->lldi.vr->qp.start,
815	     rdev->lldi.vr->qp.size,
816	     rdev->lldi.vr->cq.start,
817	     rdev->lldi.vr->cq.size);
818	PDBG("udb len 0x%x udb base %p db_reg %p gts_reg %p qpshift %lu "
819	     "qpmask 0x%x cqshift %lu cqmask 0x%x\n",
820	     (unsigned)pci_resource_len(rdev->lldi.pdev, 2),
821	     (void *)pci_resource_start(rdev->lldi.pdev, 2),
822	     rdev->lldi.db_reg,
823	     rdev->lldi.gts_reg,
824	     rdev->qpshift, rdev->qpmask,
825	     rdev->cqshift, rdev->cqmask);
826
827	if (c4iw_num_stags(rdev) == 0) {
828		err = -EINVAL;
829		goto err1;
830	}
831
832	rdev->stats.pd.total = T4_MAX_NUM_PD;
833	rdev->stats.stag.total = rdev->lldi.vr->stag.size;
834	rdev->stats.pbl.total = rdev->lldi.vr->pbl.size;
835	rdev->stats.rqt.total = rdev->lldi.vr->rq.size;
836	rdev->stats.ocqp.total = rdev->lldi.vr->ocq.size;
837	rdev->stats.qid.total = rdev->lldi.vr->qp.size;
838
839	err = c4iw_init_resource(rdev, c4iw_num_stags(rdev), T4_MAX_NUM_PD);
840	if (err) {
841		printk(KERN_ERR MOD "error %d initializing resources\n", err);
842		goto err1;
843	}
844	err = c4iw_pblpool_create(rdev);
845	if (err) {
846		printk(KERN_ERR MOD "error %d initializing pbl pool\n", err);
847		goto err2;
848	}
849	err = c4iw_rqtpool_create(rdev);
850	if (err) {
851		printk(KERN_ERR MOD "error %d initializing rqt pool\n", err);
852		goto err3;
853	}
854	err = c4iw_ocqp_pool_create(rdev);
855	if (err) {
856		printk(KERN_ERR MOD "error %d initializing ocqp pool\n", err);
857		goto err4;
858	}
859	rdev->status_page = (struct t4_dev_status_page *)
860			    __get_free_page(GFP_KERNEL);
861	if (!rdev->status_page) {
862		pr_err(MOD "error allocating status page\n");
863		goto err4;
864	}
865
866	if (c4iw_wr_log) {
867		rdev->wr_log = kzalloc((1 << c4iw_wr_log_size_order) *
868				       sizeof(*rdev->wr_log), GFP_KERNEL);
869		if (rdev->wr_log) {
870			rdev->wr_log_size = 1 << c4iw_wr_log_size_order;
871			atomic_set(&rdev->wr_log_idx, 0);
872		} else {
873			pr_err(MOD "error allocating wr_log. Logging disabled\n");
874		}
875	}
876
877	rdev->status_page->db_off = 0;
878
879	return 0;
880err4:
881	c4iw_rqtpool_destroy(rdev);
882err3:
883	c4iw_pblpool_destroy(rdev);
884err2:
885	c4iw_destroy_resource(&rdev->resource);
886err1:
887	return err;
888}
889
890static void c4iw_rdev_close(struct c4iw_rdev *rdev)
891{
892	kfree(rdev->wr_log);
893	free_page((unsigned long)rdev->status_page);
894	c4iw_pblpool_destroy(rdev);
895	c4iw_rqtpool_destroy(rdev);
896	c4iw_destroy_resource(&rdev->resource);
897}
898
899static void c4iw_dealloc(struct uld_ctx *ctx)
900{
901	c4iw_rdev_close(&ctx->dev->rdev);
902	idr_destroy(&ctx->dev->cqidr);
903	idr_destroy(&ctx->dev->qpidr);
904	idr_destroy(&ctx->dev->mmidr);
905	idr_destroy(&ctx->dev->hwtid_idr);
906	idr_destroy(&ctx->dev->stid_idr);
907	idr_destroy(&ctx->dev->atid_idr);
908	if (ctx->dev->rdev.bar2_kva)
909		iounmap(ctx->dev->rdev.bar2_kva);
910	if (ctx->dev->rdev.oc_mw_kva)
911		iounmap(ctx->dev->rdev.oc_mw_kva);
912	ib_dealloc_device(&ctx->dev->ibdev);
913	ctx->dev = NULL;
914}
915
916static void c4iw_remove(struct uld_ctx *ctx)
917{
918	PDBG("%s c4iw_dev %p\n", __func__,  ctx->dev);
919	c4iw_unregister_device(ctx->dev);
920	c4iw_dealloc(ctx);
921}
922
923static int rdma_supported(const struct cxgb4_lld_info *infop)
924{
925	return infop->vr->stag.size > 0 && infop->vr->pbl.size > 0 &&
926	       infop->vr->rq.size > 0 && infop->vr->qp.size > 0 &&
927	       infop->vr->cq.size > 0;
928}
929
930static struct c4iw_dev *c4iw_alloc(const struct cxgb4_lld_info *infop)
931{
932	struct c4iw_dev *devp;
933	int ret;
934
935	if (!rdma_supported(infop)) {
936		printk(KERN_INFO MOD "%s: RDMA not supported on this device.\n",
937		       pci_name(infop->pdev));
938		return ERR_PTR(-ENOSYS);
939	}
940	if (!ocqp_supported(infop))
941		pr_info("%s: On-Chip Queues not supported on this device.\n",
942			pci_name(infop->pdev));
943
944	devp = (struct c4iw_dev *)ib_alloc_device(sizeof(*devp));
945	if (!devp) {
946		printk(KERN_ERR MOD "Cannot allocate ib device\n");
947		return ERR_PTR(-ENOMEM);
948	}
949	devp->rdev.lldi = *infop;
950
951	/* init various hw-queue params based on lld info */
952	PDBG("%s: Ing. padding boundary is %d, egrsstatuspagesize = %d\n",
953	     __func__, devp->rdev.lldi.sge_ingpadboundary,
954	     devp->rdev.lldi.sge_egrstatuspagesize);
955
956	devp->rdev.hw_queue.t4_eq_status_entries =
957		devp->rdev.lldi.sge_ingpadboundary > 64 ? 2 : 1;
958	devp->rdev.hw_queue.t4_max_eq_size = 65520;
959	devp->rdev.hw_queue.t4_max_iq_size = 65520;
960	devp->rdev.hw_queue.t4_max_rq_size = 8192 -
961		devp->rdev.hw_queue.t4_eq_status_entries - 1;
962	devp->rdev.hw_queue.t4_max_sq_size =
963		devp->rdev.hw_queue.t4_max_eq_size -
964		devp->rdev.hw_queue.t4_eq_status_entries - 1;
965	devp->rdev.hw_queue.t4_max_qp_depth =
966		devp->rdev.hw_queue.t4_max_rq_size;
967	devp->rdev.hw_queue.t4_max_cq_depth =
968		devp->rdev.hw_queue.t4_max_iq_size - 2;
969	devp->rdev.hw_queue.t4_stat_len =
970		devp->rdev.lldi.sge_egrstatuspagesize;
971
972	/*
973	 * For T5 devices, we map all of BAR2 with WC.
974	 * For T4 devices with onchip qp mem, we map only that part
975	 * of BAR2 with WC.
976	 */
977	devp->rdev.bar2_pa = pci_resource_start(devp->rdev.lldi.pdev, 2);
978	if (is_t5(devp->rdev.lldi.adapter_type)) {
979		devp->rdev.bar2_kva = ioremap_wc(devp->rdev.bar2_pa,
980			pci_resource_len(devp->rdev.lldi.pdev, 2));
981		if (!devp->rdev.bar2_kva) {
982			pr_err(MOD "Unable to ioremap BAR2\n");
983			ib_dealloc_device(&devp->ibdev);
984			return ERR_PTR(-EINVAL);
985		}
986	} else if (ocqp_supported(infop)) {
987		devp->rdev.oc_mw_pa =
988			pci_resource_start(devp->rdev.lldi.pdev, 2) +
989			pci_resource_len(devp->rdev.lldi.pdev, 2) -
990			roundup_pow_of_two(devp->rdev.lldi.vr->ocq.size);
991		devp->rdev.oc_mw_kva = ioremap_wc(devp->rdev.oc_mw_pa,
992			devp->rdev.lldi.vr->ocq.size);
993		if (!devp->rdev.oc_mw_kva) {
994			pr_err(MOD "Unable to ioremap onchip mem\n");
995			ib_dealloc_device(&devp->ibdev);
996			return ERR_PTR(-EINVAL);
997		}
998	}
999
1000	PDBG(KERN_INFO MOD "ocq memory: "
1001	       "hw_start 0x%x size %u mw_pa 0x%lx mw_kva %p\n",
1002	       devp->rdev.lldi.vr->ocq.start, devp->rdev.lldi.vr->ocq.size,
1003	       devp->rdev.oc_mw_pa, devp->rdev.oc_mw_kva);
1004
1005	ret = c4iw_rdev_open(&devp->rdev);
1006	if (ret) {
1007		printk(KERN_ERR MOD "Unable to open CXIO rdev err %d\n", ret);
1008		ib_dealloc_device(&devp->ibdev);
1009		return ERR_PTR(ret);
1010	}
1011
1012	idr_init(&devp->cqidr);
1013	idr_init(&devp->qpidr);
1014	idr_init(&devp->mmidr);
1015	idr_init(&devp->hwtid_idr);
1016	idr_init(&devp->stid_idr);
1017	idr_init(&devp->atid_idr);
1018	spin_lock_init(&devp->lock);
1019	mutex_init(&devp->rdev.stats.lock);
1020	mutex_init(&devp->db_mutex);
1021	INIT_LIST_HEAD(&devp->db_fc_list);
1022	devp->avail_ird = devp->rdev.lldi.max_ird_adapter;
1023
1024	if (c4iw_debugfs_root) {
1025		devp->debugfs_root = debugfs_create_dir(
1026					pci_name(devp->rdev.lldi.pdev),
1027					c4iw_debugfs_root);
1028		setup_debugfs(devp);
1029	}
1030
1031
1032	return devp;
1033}
1034
1035static void *c4iw_uld_add(const struct cxgb4_lld_info *infop)
1036{
1037	struct uld_ctx *ctx;
1038	static int vers_printed;
1039	int i;
1040
1041	if (!vers_printed++)
1042		pr_info("Chelsio T4/T5 RDMA Driver - version %s\n",
1043			DRV_VERSION);
1044
1045	ctx = kzalloc(sizeof *ctx, GFP_KERNEL);
1046	if (!ctx) {
1047		ctx = ERR_PTR(-ENOMEM);
1048		goto out;
1049	}
1050	ctx->lldi = *infop;
1051
1052	PDBG("%s found device %s nchan %u nrxq %u ntxq %u nports %u\n",
1053	     __func__, pci_name(ctx->lldi.pdev),
1054	     ctx->lldi.nchan, ctx->lldi.nrxq,
1055	     ctx->lldi.ntxq, ctx->lldi.nports);
1056
1057	mutex_lock(&dev_mutex);
1058	list_add_tail(&ctx->entry, &uld_ctx_list);
1059	mutex_unlock(&dev_mutex);
1060
1061	for (i = 0; i < ctx->lldi.nrxq; i++)
1062		PDBG("rxqid[%u] %u\n", i, ctx->lldi.rxq_ids[i]);
1063out:
1064	return ctx;
1065}
1066
1067static inline struct sk_buff *copy_gl_to_skb_pkt(const struct pkt_gl *gl,
1068						 const __be64 *rsp,
1069						 u32 pktshift)
1070{
1071	struct sk_buff *skb;
1072
1073	/*
1074	 * Allocate space for cpl_pass_accept_req which will be synthesized by
1075	 * driver. Once the driver synthesizes the request the skb will go
1076	 * through the regular cpl_pass_accept_req processing.
1077	 * The math here assumes sizeof cpl_pass_accept_req >= sizeof
1078	 * cpl_rx_pkt.
1079	 */
1080	skb = alloc_skb(gl->tot_len + sizeof(struct cpl_pass_accept_req) +
1081			sizeof(struct rss_header) - pktshift, GFP_ATOMIC);
1082	if (unlikely(!skb))
1083		return NULL;
1084
1085	 __skb_put(skb, gl->tot_len + sizeof(struct cpl_pass_accept_req) +
1086		   sizeof(struct rss_header) - pktshift);
1087
1088	/*
1089	 * This skb will contain:
1090	 *   rss_header from the rspq descriptor (1 flit)
1091	 *   cpl_rx_pkt struct from the rspq descriptor (2 flits)
1092	 *   space for the difference between the size of an
1093	 *      rx_pkt and pass_accept_req cpl (1 flit)
1094	 *   the packet data from the gl
1095	 */
1096	skb_copy_to_linear_data(skb, rsp, sizeof(struct cpl_pass_accept_req) +
1097				sizeof(struct rss_header));
1098	skb_copy_to_linear_data_offset(skb, sizeof(struct rss_header) +
1099				       sizeof(struct cpl_pass_accept_req),
1100				       gl->va + pktshift,
1101				       gl->tot_len - pktshift);
1102	return skb;
1103}
1104
1105static inline int recv_rx_pkt(struct c4iw_dev *dev, const struct pkt_gl *gl,
1106			   const __be64 *rsp)
1107{
1108	unsigned int opcode = *(u8 *)rsp;
1109	struct sk_buff *skb;
1110
1111	if (opcode != CPL_RX_PKT)
1112		goto out;
1113
1114	skb = copy_gl_to_skb_pkt(gl , rsp, dev->rdev.lldi.sge_pktshift);
1115	if (skb == NULL)
1116		goto out;
1117
1118	if (c4iw_handlers[opcode] == NULL) {
1119		pr_info("%s no handler opcode 0x%x...\n", __func__,
1120		       opcode);
1121		kfree_skb(skb);
1122		goto out;
1123	}
1124	c4iw_handlers[opcode](dev, skb);
1125	return 1;
1126out:
1127	return 0;
1128}
1129
1130static int c4iw_uld_rx_handler(void *handle, const __be64 *rsp,
1131			const struct pkt_gl *gl)
1132{
1133	struct uld_ctx *ctx = handle;
1134	struct c4iw_dev *dev = ctx->dev;
1135	struct sk_buff *skb;
1136	u8 opcode;
1137
1138	if (gl == NULL) {
1139		/* omit RSS and rsp_ctrl at end of descriptor */
1140		unsigned int len = 64 - sizeof(struct rsp_ctrl) - 8;
1141
1142		skb = alloc_skb(256, GFP_ATOMIC);
1143		if (!skb)
1144			goto nomem;
1145		__skb_put(skb, len);
1146		skb_copy_to_linear_data(skb, &rsp[1], len);
1147	} else if (gl == CXGB4_MSG_AN) {
1148		const struct rsp_ctrl *rc = (void *)rsp;
1149
1150		u32 qid = be32_to_cpu(rc->pldbuflen_qid);
1151		c4iw_ev_handler(dev, qid);
1152		return 0;
1153	} else if (unlikely(*(u8 *)rsp != *(u8 *)gl->va)) {
1154		if (recv_rx_pkt(dev, gl, rsp))
1155			return 0;
1156
1157		pr_info("%s: unexpected FL contents at %p, " \
1158		       "RSS %#llx, FL %#llx, len %u\n",
1159		       pci_name(ctx->lldi.pdev), gl->va,
1160		       (unsigned long long)be64_to_cpu(*rsp),
1161		       (unsigned long long)be64_to_cpu(
1162		       *(__force __be64 *)gl->va),
1163		       gl->tot_len);
1164
1165		return 0;
1166	} else {
1167		skb = cxgb4_pktgl_to_skb(gl, 128, 128);
1168		if (unlikely(!skb))
1169			goto nomem;
1170	}
1171
1172	opcode = *(u8 *)rsp;
1173	if (c4iw_handlers[opcode]) {
1174		c4iw_handlers[opcode](dev, skb);
1175	} else {
1176		pr_info("%s no handler opcode 0x%x...\n", __func__,
1177		       opcode);
1178		kfree_skb(skb);
1179	}
1180
1181	return 0;
1182nomem:
1183	return -1;
1184}
1185
1186static int c4iw_uld_state_change(void *handle, enum cxgb4_state new_state)
1187{
1188	struct uld_ctx *ctx = handle;
1189
1190	PDBG("%s new_state %u\n", __func__, new_state);
1191	switch (new_state) {
1192	case CXGB4_STATE_UP:
1193		printk(KERN_INFO MOD "%s: Up\n", pci_name(ctx->lldi.pdev));
1194		if (!ctx->dev) {
1195			int ret;
1196
1197			ctx->dev = c4iw_alloc(&ctx->lldi);
1198			if (IS_ERR(ctx->dev)) {
1199				printk(KERN_ERR MOD
1200				       "%s: initialization failed: %ld\n",
1201				       pci_name(ctx->lldi.pdev),
1202				       PTR_ERR(ctx->dev));
1203				ctx->dev = NULL;
1204				break;
1205			}
1206			ret = c4iw_register_device(ctx->dev);
1207			if (ret) {
1208				printk(KERN_ERR MOD
1209				       "%s: RDMA registration failed: %d\n",
1210				       pci_name(ctx->lldi.pdev), ret);
1211				c4iw_dealloc(ctx);
1212			}
1213		}
1214		break;
1215	case CXGB4_STATE_DOWN:
1216		printk(KERN_INFO MOD "%s: Down\n",
1217		       pci_name(ctx->lldi.pdev));
1218		if (ctx->dev)
1219			c4iw_remove(ctx);
1220		break;
1221	case CXGB4_STATE_START_RECOVERY:
1222		printk(KERN_INFO MOD "%s: Fatal Error\n",
1223		       pci_name(ctx->lldi.pdev));
1224		if (ctx->dev) {
1225			struct ib_event event;
1226
1227			ctx->dev->rdev.flags |= T4_FATAL_ERROR;
1228			memset(&event, 0, sizeof event);
1229			event.event  = IB_EVENT_DEVICE_FATAL;
1230			event.device = &ctx->dev->ibdev;
1231			ib_dispatch_event(&event);
1232			c4iw_remove(ctx);
1233		}
1234		break;
1235	case CXGB4_STATE_DETACH:
1236		printk(KERN_INFO MOD "%s: Detach\n",
1237		       pci_name(ctx->lldi.pdev));
1238		if (ctx->dev)
1239			c4iw_remove(ctx);
1240		break;
1241	}
1242	return 0;
1243}
1244
1245static int disable_qp_db(int id, void *p, void *data)
1246{
1247	struct c4iw_qp *qp = p;
1248
1249	t4_disable_wq_db(&qp->wq);
1250	return 0;
1251}
1252
1253static void stop_queues(struct uld_ctx *ctx)
1254{
1255	unsigned long flags;
1256
1257	spin_lock_irqsave(&ctx->dev->lock, flags);
1258	ctx->dev->rdev.stats.db_state_transitions++;
1259	ctx->dev->db_state = STOPPED;
1260	if (ctx->dev->rdev.flags & T4_STATUS_PAGE_DISABLED)
1261		idr_for_each(&ctx->dev->qpidr, disable_qp_db, NULL);
1262	else
1263		ctx->dev->rdev.status_page->db_off = 1;
1264	spin_unlock_irqrestore(&ctx->dev->lock, flags);
1265}
1266
1267static int enable_qp_db(int id, void *p, void *data)
1268{
1269	struct c4iw_qp *qp = p;
1270
1271	t4_enable_wq_db(&qp->wq);
1272	return 0;
1273}
1274
1275static void resume_rc_qp(struct c4iw_qp *qp)
1276{
1277	spin_lock(&qp->lock);
1278	t4_ring_sq_db(&qp->wq, qp->wq.sq.wq_pidx_inc,
1279		      is_t5(qp->rhp->rdev.lldi.adapter_type), NULL);
1280	qp->wq.sq.wq_pidx_inc = 0;
1281	t4_ring_rq_db(&qp->wq, qp->wq.rq.wq_pidx_inc,
1282		      is_t5(qp->rhp->rdev.lldi.adapter_type), NULL);
1283	qp->wq.rq.wq_pidx_inc = 0;
1284	spin_unlock(&qp->lock);
1285}
1286
1287static void resume_a_chunk(struct uld_ctx *ctx)
1288{
1289	int i;
1290	struct c4iw_qp *qp;
1291
1292	for (i = 0; i < DB_FC_RESUME_SIZE; i++) {
1293		qp = list_first_entry(&ctx->dev->db_fc_list, struct c4iw_qp,
1294				      db_fc_entry);
1295		list_del_init(&qp->db_fc_entry);
1296		resume_rc_qp(qp);
1297		if (list_empty(&ctx->dev->db_fc_list))
1298			break;
1299	}
1300}
1301
1302static void resume_queues(struct uld_ctx *ctx)
1303{
1304	spin_lock_irq(&ctx->dev->lock);
1305	if (ctx->dev->db_state != STOPPED)
1306		goto out;
1307	ctx->dev->db_state = FLOW_CONTROL;
1308	while (1) {
1309		if (list_empty(&ctx->dev->db_fc_list)) {
1310			WARN_ON(ctx->dev->db_state != FLOW_CONTROL);
1311			ctx->dev->db_state = NORMAL;
1312			ctx->dev->rdev.stats.db_state_transitions++;
1313			if (ctx->dev->rdev.flags & T4_STATUS_PAGE_DISABLED) {
1314				idr_for_each(&ctx->dev->qpidr, enable_qp_db,
1315					     NULL);
1316			} else {
1317				ctx->dev->rdev.status_page->db_off = 0;
1318			}
1319			break;
1320		} else {
1321			if (cxgb4_dbfifo_count(ctx->dev->rdev.lldi.ports[0], 1)
1322			    < (ctx->dev->rdev.lldi.dbfifo_int_thresh <<
1323			       DB_FC_DRAIN_THRESH)) {
1324				resume_a_chunk(ctx);
1325			}
1326			if (!list_empty(&ctx->dev->db_fc_list)) {
1327				spin_unlock_irq(&ctx->dev->lock);
1328				if (DB_FC_RESUME_DELAY) {
1329					set_current_state(TASK_UNINTERRUPTIBLE);
1330					schedule_timeout(DB_FC_RESUME_DELAY);
1331				}
1332				spin_lock_irq(&ctx->dev->lock);
1333				if (ctx->dev->db_state != FLOW_CONTROL)
1334					break;
1335			}
1336		}
1337	}
1338out:
1339	if (ctx->dev->db_state != NORMAL)
1340		ctx->dev->rdev.stats.db_fc_interruptions++;
1341	spin_unlock_irq(&ctx->dev->lock);
1342}
1343
1344struct qp_list {
1345	unsigned idx;
1346	struct c4iw_qp **qps;
1347};
1348
1349static int add_and_ref_qp(int id, void *p, void *data)
1350{
1351	struct qp_list *qp_listp = data;
1352	struct c4iw_qp *qp = p;
1353
1354	c4iw_qp_add_ref(&qp->ibqp);
1355	qp_listp->qps[qp_listp->idx++] = qp;
1356	return 0;
1357}
1358
1359static int count_qps(int id, void *p, void *data)
1360{
1361	unsigned *countp = data;
1362	(*countp)++;
1363	return 0;
1364}
1365
1366static void deref_qps(struct qp_list *qp_list)
1367{
1368	int idx;
1369
1370	for (idx = 0; idx < qp_list->idx; idx++)
1371		c4iw_qp_rem_ref(&qp_list->qps[idx]->ibqp);
1372}
1373
1374static void recover_lost_dbs(struct uld_ctx *ctx, struct qp_list *qp_list)
1375{
1376	int idx;
1377	int ret;
1378
1379	for (idx = 0; idx < qp_list->idx; idx++) {
1380		struct c4iw_qp *qp = qp_list->qps[idx];
1381
1382		spin_lock_irq(&qp->rhp->lock);
1383		spin_lock(&qp->lock);
1384		ret = cxgb4_sync_txq_pidx(qp->rhp->rdev.lldi.ports[0],
1385					  qp->wq.sq.qid,
1386					  t4_sq_host_wq_pidx(&qp->wq),
1387					  t4_sq_wq_size(&qp->wq));
1388		if (ret) {
1389			pr_err(MOD "%s: Fatal error - "
1390			       "DB overflow recovery failed - "
1391			       "error syncing SQ qid %u\n",
1392			       pci_name(ctx->lldi.pdev), qp->wq.sq.qid);
1393			spin_unlock(&qp->lock);
1394			spin_unlock_irq(&qp->rhp->lock);
1395			return;
1396		}
1397		qp->wq.sq.wq_pidx_inc = 0;
1398
1399		ret = cxgb4_sync_txq_pidx(qp->rhp->rdev.lldi.ports[0],
1400					  qp->wq.rq.qid,
1401					  t4_rq_host_wq_pidx(&qp->wq),
1402					  t4_rq_wq_size(&qp->wq));
1403
1404		if (ret) {
1405			pr_err(MOD "%s: Fatal error - "
1406			       "DB overflow recovery failed - "
1407			       "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
1425static 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		printk(KERN_ERR MOD "%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 = kzalloc(count * sizeof *qp_list.qps, GFP_ATOMIC);
1450	if (!qp_list.qps) {
1451		printk(KERN_ERR MOD "%s: Fatal error - DB overflow recovery failed\n",
1452		       pci_name(ctx->lldi.pdev));
1453		spin_unlock_irq(&ctx->dev->lock);
1454		return;
1455	}
1456	qp_list.idx = 0;
1457
1458	/* add and ref each qp so it doesn't get freed */
1459	idr_for_each(&ctx->dev->qpidr, add_and_ref_qp, &qp_list);
1460
1461	spin_unlock_irq(&ctx->dev->lock);
1462
1463	/* now traverse the list in a safe context to recover the db state*/
1464	recover_lost_dbs(ctx, &qp_list);
1465
1466	/* we're almost done!  deref the qps and clean up */
1467	deref_qps(&qp_list);
1468	kfree(qp_list.qps);
1469
1470	spin_lock_irq(&ctx->dev->lock);
1471	WARN_ON(ctx->dev->db_state != RECOVERY);
1472	ctx->dev->db_state = STOPPED;
1473	spin_unlock_irq(&ctx->dev->lock);
1474}
1475
1476static int c4iw_uld_control(void *handle, enum cxgb4_control control, ...)
1477{
1478	struct uld_ctx *ctx = handle;
1479
1480	switch (control) {
1481	case CXGB4_CONTROL_DB_FULL:
1482		stop_queues(ctx);
1483		ctx->dev->rdev.stats.db_full++;
1484		break;
1485	case CXGB4_CONTROL_DB_EMPTY:
1486		resume_queues(ctx);
1487		mutex_lock(&ctx->dev->rdev.stats.lock);
1488		ctx->dev->rdev.stats.db_empty++;
1489		mutex_unlock(&ctx->dev->rdev.stats.lock);
1490		break;
1491	case CXGB4_CONTROL_DB_DROP:
1492		recover_queues(ctx);
1493		mutex_lock(&ctx->dev->rdev.stats.lock);
1494		ctx->dev->rdev.stats.db_drop++;
1495		mutex_unlock(&ctx->dev->rdev.stats.lock);
1496		break;
1497	default:
1498		printk(KERN_WARNING MOD "%s: unknown control cmd %u\n",
1499		       pci_name(ctx->lldi.pdev), control);
1500		break;
1501	}
1502	return 0;
1503}
1504
1505static struct cxgb4_uld_info c4iw_uld_info = {
1506	.name = DRV_NAME,
1507	.add = c4iw_uld_add,
1508	.rx_handler = c4iw_uld_rx_handler,
1509	.state_change = c4iw_uld_state_change,
1510	.control = c4iw_uld_control,
1511};
1512
1513static int __init c4iw_init_module(void)
1514{
1515	int err;
1516
1517	err = c4iw_cm_init();
1518	if (err)
1519		return err;
1520
1521	c4iw_debugfs_root = debugfs_create_dir(DRV_NAME, NULL);
1522	if (!c4iw_debugfs_root)
1523		printk(KERN_WARNING MOD
1524		       "could not create debugfs entry, continuing\n");
1525
1526	if (ibnl_add_client(RDMA_NL_C4IW, RDMA_NL_IWPM_NUM_OPS,
1527			    c4iw_nl_cb_table))
1528		pr_err("%s[%u]: Failed to add netlink callback\n"
1529		       , __func__, __LINE__);
1530
1531	err = iwpm_init(RDMA_NL_C4IW);
1532	if (err) {
1533		pr_err("port mapper initialization failed with %d\n", err);
1534		ibnl_remove_client(RDMA_NL_C4IW);
1535		c4iw_cm_term();
1536		debugfs_remove_recursive(c4iw_debugfs_root);
1537		return err;
1538	}
1539
1540	cxgb4_register_uld(CXGB4_ULD_RDMA, &c4iw_uld_info);
1541
1542	return 0;
1543}
1544
1545static void __exit c4iw_exit_module(void)
1546{
1547	struct uld_ctx *ctx, *tmp;
1548
1549	mutex_lock(&dev_mutex);
1550	list_for_each_entry_safe(ctx, tmp, &uld_ctx_list, entry) {
1551		if (ctx->dev)
1552			c4iw_remove(ctx);
1553		kfree(ctx);
1554	}
1555	mutex_unlock(&dev_mutex);
1556	cxgb4_unregister_uld(CXGB4_ULD_RDMA);
1557	iwpm_exit(RDMA_NL_C4IW);
1558	ibnl_remove_client(RDMA_NL_C4IW);
1559	c4iw_cm_term();
1560	debugfs_remove_recursive(c4iw_debugfs_root);
1561}
1562
1563module_init(c4iw_init_module);
1564module_exit(c4iw_exit_module);
1565