1 /*-
2 * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
3 *
4 * Copyright (c) 2009-2013 Chelsio, Inc. All rights reserved.
5 *
6 * This software is available to you under a choice of one of two
7 * licenses. You may choose to be licensed under the terms of the GNU
8 * General Public License (GPL) Version 2, available from the file
9 * COPYING in the main directory of this source tree, or the
10 * OpenIB.org BSD license below:
11 *
12 * Redistribution and use in source and binary forms, with or
13 * without modification, are permitted provided that the following
14 * conditions are met:
15 *
16 * - Redistributions of source code must retain the above
17 * copyright notice, this list of conditions and the following
18 * disclaimer.
19 *
20 * - Redistributions in binary form must reproduce the above
21 * copyright notice, this list of conditions and the following
22 * disclaimer in the documentation and/or other materials
23 * provided with the distribution.
24 *
25 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
26 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
27 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
28 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
29 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
30 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
31 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
32 * SOFTWARE.
33 */
34 #include <sys/cdefs.h>
35 __FBSDID("$FreeBSD$");
36
37 #include "opt_inet.h"
38
39 #include <sys/ktr.h>
40
41 #include <linux/module.h>
42 #include <linux/moduleparam.h>
43
44 #include <rdma/ib_verbs.h>
45 #include <linux/idr.h>
46
47 #ifdef TCP_OFFLOAD
48 #include "iw_cxgbe.h"
49
50 void
c4iw_release_dev_ucontext(struct c4iw_rdev * rdev,struct c4iw_dev_ucontext * uctx)51 c4iw_release_dev_ucontext(struct c4iw_rdev *rdev,
52 struct c4iw_dev_ucontext *uctx)
53 {
54 struct list_head *pos, *nxt;
55 struct c4iw_qid_list *entry;
56
57 mutex_lock(&uctx->lock);
58 list_for_each_safe(pos, nxt, &uctx->qpids) {
59 entry = list_entry(pos, struct c4iw_qid_list, entry);
60 list_del_init(&entry->entry);
61 if (!(entry->qid & rdev->qpmask)) {
62 c4iw_put_resource(&rdev->resource.qid_table,
63 entry->qid);
64 mutex_lock(&rdev->stats.lock);
65 rdev->stats.qid.cur -= rdev->qpmask + 1;
66 mutex_unlock(&rdev->stats.lock);
67 }
68 kfree(entry);
69 }
70
71 list_for_each_safe(pos, nxt, &uctx->cqids) {
72 entry = list_entry(pos, struct c4iw_qid_list, entry);
73 list_del_init(&entry->entry);
74 kfree(entry);
75 }
76 mutex_unlock(&uctx->lock);
77 }
78
79 void
c4iw_init_dev_ucontext(struct c4iw_rdev * rdev,struct c4iw_dev_ucontext * uctx)80 c4iw_init_dev_ucontext(struct c4iw_rdev *rdev, struct c4iw_dev_ucontext *uctx)
81 {
82
83 INIT_LIST_HEAD(&uctx->qpids);
84 INIT_LIST_HEAD(&uctx->cqids);
85 mutex_init(&uctx->lock);
86 }
87
88 static int
c4iw_rdev_open(struct c4iw_rdev * rdev)89 c4iw_rdev_open(struct c4iw_rdev *rdev)
90 {
91 struct adapter *sc = rdev->adap;
92 struct sge_params *sp = &sc->params.sge;
93 int rc;
94 unsigned short ucq_density = 1 << sp->iq_s_qpp; /* # of user CQs/page */
95 unsigned short udb_density = 1 << sp->eq_s_qpp; /* # of user DB/page */
96
97
98 c4iw_init_dev_ucontext(rdev, &rdev->uctx);
99
100 /*
101 * This implementation assumes udb_density == ucq_density! Eventually
102 * we might need to support this but for now fail the open. Also the
103 * cqid and qpid range must match for now.
104 */
105 if (udb_density != ucq_density) {
106 device_printf(sc->dev, "unsupported udb/ucq densities %u/%u\n",
107 udb_density, ucq_density);
108 rc = -EINVAL;
109 goto err1;
110 }
111 if (sc->vres.qp.start != sc->vres.cq.start ||
112 sc->vres.qp.size != sc->vres.cq.size) {
113 device_printf(sc->dev, "%s: unsupported qp and cq id ranges "
114 "qp start %u size %u cq start %u size %u\n", __func__,
115 sc->vres.qp.start, sc->vres.qp.size, sc->vres.cq.start,
116 sc->vres.cq.size);
117 rc = -EINVAL;
118 goto err1;
119 }
120
121 rdev->qpshift = PAGE_SHIFT - sp->eq_s_qpp;
122 rdev->qpmask = udb_density - 1;
123 rdev->cqshift = PAGE_SHIFT - sp->iq_s_qpp;
124 rdev->cqmask = ucq_density - 1;
125
126 if (c4iw_num_stags(rdev) == 0) {
127 rc = -EINVAL;
128 goto err1;
129 }
130
131 rdev->stats.pd.total = T4_MAX_NUM_PD;
132 rdev->stats.stag.total = sc->vres.stag.size;
133 rdev->stats.pbl.total = sc->vres.pbl.size;
134 rdev->stats.rqt.total = sc->vres.rq.size;
135 rdev->stats.qid.total = sc->vres.qp.size;
136
137 rc = c4iw_init_resource(rdev, c4iw_num_stags(rdev), T4_MAX_NUM_PD);
138 if (rc) {
139 device_printf(sc->dev, "error %d initializing resources\n", rc);
140 goto err1;
141 }
142 rc = c4iw_pblpool_create(rdev);
143 if (rc) {
144 device_printf(sc->dev, "error %d initializing pbl pool\n", rc);
145 goto err2;
146 }
147 rc = c4iw_rqtpool_create(rdev);
148 if (rc) {
149 device_printf(sc->dev, "error %d initializing rqt pool\n", rc);
150 goto err3;
151 }
152 rdev->status_page = (struct t4_dev_status_page *)
153 __get_free_page(GFP_KERNEL);
154 if (!rdev->status_page) {
155 rc = -ENOMEM;
156 goto err4;
157 }
158 rdev->status_page->qp_start = sc->vres.qp.start;
159 rdev->status_page->qp_size = sc->vres.qp.size;
160 rdev->status_page->cq_start = sc->vres.cq.start;
161 rdev->status_page->cq_size = sc->vres.cq.size;
162
163 /* T5 and above devices don't need Doorbell recovery logic,
164 * so db_off is always set to '0'.
165 */
166 rdev->status_page->db_off = 0;
167
168 rdev->status_page->wc_supported = rdev->adap->iwt.wc_en;
169
170 rdev->free_workq = create_singlethread_workqueue("iw_cxgb4_free");
171 if (!rdev->free_workq) {
172 rc = -ENOMEM;
173 goto err5;
174 }
175 return (0);
176 err5:
177 free_page((unsigned long)rdev->status_page);
178 err4:
179 c4iw_rqtpool_destroy(rdev);
180 err3:
181 c4iw_pblpool_destroy(rdev);
182 err2:
183 c4iw_destroy_resource(&rdev->resource);
184 err1:
185 return (rc);
186 }
187
c4iw_rdev_close(struct c4iw_rdev * rdev)188 static void c4iw_rdev_close(struct c4iw_rdev *rdev)
189 {
190 free_page((unsigned long)rdev->status_page);
191 c4iw_pblpool_destroy(rdev);
192 c4iw_rqtpool_destroy(rdev);
193 c4iw_destroy_resource(&rdev->resource);
194 }
195
196 static void
c4iw_dealloc(struct c4iw_dev * iwsc)197 c4iw_dealloc(struct c4iw_dev *iwsc)
198 {
199
200 c4iw_rdev_close(&iwsc->rdev);
201 idr_destroy(&iwsc->cqidr);
202 idr_destroy(&iwsc->qpidr);
203 idr_destroy(&iwsc->mmidr);
204 ib_dealloc_device(&iwsc->ibdev);
205 }
206
207 static struct c4iw_dev *
c4iw_alloc(struct adapter * sc)208 c4iw_alloc(struct adapter *sc)
209 {
210 struct c4iw_dev *iwsc;
211 int rc;
212
213 iwsc = (struct c4iw_dev *)ib_alloc_device(sizeof(*iwsc));
214 if (iwsc == NULL) {
215 device_printf(sc->dev, "Cannot allocate ib device.\n");
216 return (ERR_PTR(-ENOMEM));
217 }
218 iwsc->rdev.adap = sc;
219
220 /* init various hw-queue params based on lld info */
221 iwsc->rdev.hw_queue.t4_eq_status_entries =
222 sc->params.sge.spg_len / EQ_ESIZE;
223 iwsc->rdev.hw_queue.t4_max_eq_size = 65520;
224 iwsc->rdev.hw_queue.t4_max_iq_size = 65520;
225 iwsc->rdev.hw_queue.t4_max_rq_size = 8192 -
226 iwsc->rdev.hw_queue.t4_eq_status_entries - 1;
227 iwsc->rdev.hw_queue.t4_max_sq_size =
228 iwsc->rdev.hw_queue.t4_max_eq_size -
229 iwsc->rdev.hw_queue.t4_eq_status_entries - 1;
230 iwsc->rdev.hw_queue.t4_max_qp_depth =
231 iwsc->rdev.hw_queue.t4_max_rq_size;
232 iwsc->rdev.hw_queue.t4_max_cq_depth =
233 iwsc->rdev.hw_queue.t4_max_iq_size - 2;
234 iwsc->rdev.hw_queue.t4_stat_len = iwsc->rdev.adap->params.sge.spg_len;
235
236 /* As T5 and above devices support BAR2 kernel doorbells & WC, we map
237 * all of BAR2, for both User and Kernel Doorbells-GTS.
238 */
239 iwsc->rdev.bar2_kva = (void __iomem *)((u64)iwsc->rdev.adap->udbs_base);
240 iwsc->rdev.bar2_pa = vtophys(iwsc->rdev.adap->udbs_base);
241 iwsc->rdev.bar2_len = rman_get_size(iwsc->rdev.adap->udbs_res);
242
243 rc = c4iw_rdev_open(&iwsc->rdev);
244 if (rc != 0) {
245 device_printf(sc->dev, "Unable to open CXIO rdev (%d)\n", rc);
246 ib_dealloc_device(&iwsc->ibdev);
247 return (ERR_PTR(rc));
248 }
249
250 idr_init(&iwsc->cqidr);
251 idr_init(&iwsc->qpidr);
252 idr_init(&iwsc->mmidr);
253 spin_lock_init(&iwsc->lock);
254 mutex_init(&iwsc->rdev.stats.lock);
255 iwsc->avail_ird = iwsc->rdev.adap->params.max_ird_adapter;
256
257 return (iwsc);
258 }
259
260 static int c4iw_mod_load(void);
261 static int c4iw_mod_unload(void);
262 static int c4iw_activate(struct adapter *);
263 static int c4iw_deactivate(struct adapter *);
264
265 static struct uld_info c4iw_uld_info = {
266 .uld_id = ULD_IWARP,
267 .activate = c4iw_activate,
268 .deactivate = c4iw_deactivate,
269 };
270
271 static int
c4iw_activate(struct adapter * sc)272 c4iw_activate(struct adapter *sc)
273 {
274 struct c4iw_dev *iwsc;
275 int rc;
276
277 ASSERT_SYNCHRONIZED_OP(sc);
278
279 if (is_t4(sc)) {
280 device_printf(sc->dev, "No iWARP support for T4 devices, "
281 "please install T5 or above devices.\n");
282 return (ENOSYS);
283 }
284
285 if (uld_active(sc, ULD_IWARP)) {
286 KASSERT(0, ("%s: RDMA already eanbled on sc %p", __func__, sc));
287 return (0);
288 }
289
290 if (sc->rdmacaps == 0) {
291 device_printf(sc->dev,
292 "RDMA not supported or RDMA cap is not enabled.\n");
293 return (ENOSYS);
294 }
295
296 iwsc = c4iw_alloc(sc);
297 if (IS_ERR(iwsc)) {
298 rc = -PTR_ERR(iwsc);
299 device_printf(sc->dev, "initialization failed: %d\n", rc);
300 return (rc);
301 }
302
303 sc->iwarp_softc = iwsc;
304
305 rc = -c4iw_register_device(iwsc);
306 if (rc) {
307 device_printf(sc->dev, "RDMA registration failed: %d\n", rc);
308 c4iw_dealloc(iwsc);
309 sc->iwarp_softc = NULL;
310 }
311
312 return (rc);
313 }
314
315 static int
c4iw_deactivate(struct adapter * sc)316 c4iw_deactivate(struct adapter *sc)
317 {
318 struct c4iw_dev *iwsc = sc->iwarp_softc;
319
320 ASSERT_SYNCHRONIZED_OP(sc);
321
322 c4iw_unregister_device(iwsc);
323 c4iw_dealloc(iwsc);
324 sc->iwarp_softc = NULL;
325
326 return (0);
327 }
328
329 static void
c4iw_activate_all(struct adapter * sc,void * arg __unused)330 c4iw_activate_all(struct adapter *sc, void *arg __unused)
331 {
332
333 if (begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4iwact") != 0)
334 return;
335
336 /* Activate iWARP if any port on this adapter has IFCAP_TOE enabled. */
337 if (sc->offload_map && !uld_active(sc, ULD_IWARP))
338 (void) t4_activate_uld(sc, ULD_IWARP);
339
340 end_synchronized_op(sc, 0);
341 }
342
343 static void
c4iw_deactivate_all(struct adapter * sc,void * arg __unused)344 c4iw_deactivate_all(struct adapter *sc, void *arg __unused)
345 {
346
347 if (begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4iwdea") != 0)
348 return;
349
350 if (uld_active(sc, ULD_IWARP))
351 (void) t4_deactivate_uld(sc, ULD_IWARP);
352
353 end_synchronized_op(sc, 0);
354 }
355
356 static int
c4iw_mod_load(void)357 c4iw_mod_load(void)
358 {
359 int rc;
360
361 rc = -c4iw_cm_init();
362 if (rc != 0)
363 return (rc);
364
365 rc = t4_register_uld(&c4iw_uld_info);
366 if (rc != 0) {
367 c4iw_cm_term();
368 return (rc);
369 }
370
371 t4_iterate(c4iw_activate_all, NULL);
372
373 return (rc);
374 }
375
376 static int
c4iw_mod_unload(void)377 c4iw_mod_unload(void)
378 {
379
380 t4_iterate(c4iw_deactivate_all, NULL);
381
382 c4iw_cm_term();
383
384 if (t4_unregister_uld(&c4iw_uld_info) == EBUSY)
385 return (EBUSY);
386
387 return (0);
388 }
389
390 #endif
391
392 /*
393 * t4_tom won't load on kernels without TCP_OFFLOAD and this module's dependency
394 * on t4_tom ensures that it won't either. So we don't directly check for
395 * TCP_OFFLOAD here.
396 */
397 static int
c4iw_modevent(module_t mod,int cmd,void * arg)398 c4iw_modevent(module_t mod, int cmd, void *arg)
399 {
400 int rc = 0;
401
402 #ifdef TCP_OFFLOAD
403 switch (cmd) {
404 case MOD_LOAD:
405 rc = c4iw_mod_load();
406 if (rc == 0)
407 printf("iw_cxgbe: Chelsio T5/T6 RDMA driver loaded.\n");
408 break;
409
410 case MOD_UNLOAD:
411 rc = c4iw_mod_unload();
412 break;
413
414 default:
415 rc = EINVAL;
416 }
417 #else
418 printf("t4_tom: compiled without TCP_OFFLOAD support.\n");
419 rc = EOPNOTSUPP;
420 #endif
421 return (rc);
422 }
423
424 static moduledata_t c4iw_mod_data = {
425 "iw_cxgbe",
426 c4iw_modevent,
427 0
428 };
429
430 MODULE_VERSION(iw_cxgbe, 1);
431 MODULE_DEPEND(iw_cxgbe, t4nex, 1, 1, 1);
432 MODULE_DEPEND(iw_cxgbe, t4_tom, 1, 1, 1);
433 MODULE_DEPEND(iw_cxgbe, ibcore, 1, 1, 1);
434 MODULE_DEPEND(iw_cxgbe, linuxkpi, 1, 1, 1);
435 DECLARE_MODULE(iw_cxgbe, c4iw_mod_data, SI_SUB_EXEC, SI_ORDER_ANY);
436