1 /*
2 * Copyright (c) 2004, 2005 Topspin Communications. All rights reserved.
3 * Copyright (c) 2005 Sun Microsystems, Inc. All rights reserved.
4 * Copyright (c) 2005, 2006 Cisco Systems. All rights reserved.
5 * Copyright (c) 2005 Mellanox Technologies. All rights reserved.
6 * Copyright (c) 2004 Voltaire, Inc. All rights reserved.
7 *
8 * This software is available to you under a choice of one of two
9 * licenses. You may choose to be licensed under the terms of the GNU
10 * General Public License (GPL) Version 2, available from the file
11 * COPYING in the main directory of this source tree, or the
12 * OpenIB.org BSD license below:
13 *
14 * Redistribution and use in source and binary forms, with or
15 * without modification, are permitted provided that the following
16 * conditions are met:
17 *
18 * - Redistributions of source code must retain the above
19 * copyright notice, this list of conditions and the following
20 * disclaimer.
21 *
22 * - Redistributions in binary form must reproduce the above
23 * copyright notice, this list of conditions and the following
24 * disclaimer in the documentation and/or other materials
25 * provided with the distribution.
26 *
27 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
28 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
29 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
30 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
31 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
32 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
33 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
34 * SOFTWARE.
35 */
36
37 #include <rdma/ib_smi.h>
38 #include <rdma/ib_user_verbs.h>
39 #include <rdma/iter.h>
40 #include <rdma/uverbs_ioctl.h>
41
42 #include <linux/sched.h>
43 #include <linux/slab.h>
44 #include <linux/stat.h>
45 #include <linux/mm.h>
46 #include <linux/export.h>
47
48 #include "mthca_dev.h"
49 #include "mthca_cmd.h"
50 #include <rdma/mthca-abi.h>
51 #include "mthca_memfree.h"
52
mthca_query_device(struct ib_device * ibdev,struct ib_device_attr * props,struct ib_udata * uhw)53 static int mthca_query_device(struct ib_device *ibdev, struct ib_device_attr *props,
54 struct ib_udata *uhw)
55 {
56 struct ib_smp *in_mad;
57 struct ib_smp *out_mad;
58 int err = -ENOMEM;
59 struct mthca_dev *mdev = to_mdev(ibdev);
60
61 if (uhw->inlen || uhw->outlen)
62 return -EINVAL;
63
64 in_mad = kzalloc_obj(*in_mad);
65 out_mad = kmalloc_obj(*out_mad);
66 if (!in_mad || !out_mad)
67 goto out;
68
69 memset(props, 0, sizeof *props);
70
71 props->fw_ver = mdev->fw_ver;
72
73 ib_init_query_mad(in_mad);
74 in_mad->attr_id = IB_SMP_ATTR_NODE_INFO;
75
76 err = mthca_MAD_IFC(mdev, 1, 1,
77 1, NULL, NULL, in_mad, out_mad);
78 if (err)
79 goto out;
80
81 props->device_cap_flags = mdev->device_cap_flags;
82 props->vendor_id = be32_to_cpup((__be32 *) (out_mad->data + 36)) &
83 0xffffff;
84 props->vendor_part_id = be16_to_cpup((__be16 *) (out_mad->data + 30));
85 props->hw_ver = be32_to_cpup((__be32 *) (out_mad->data + 32));
86 memcpy(&props->sys_image_guid, out_mad->data + 4, 8);
87
88 props->max_mr_size = ~0ull;
89 props->page_size_cap = mdev->limits.page_size_cap;
90 props->max_qp = mdev->limits.num_qps - mdev->limits.reserved_qps;
91 props->max_qp_wr = mdev->limits.max_wqes;
92 props->max_send_sge = mdev->limits.max_sg;
93 props->max_recv_sge = mdev->limits.max_sg;
94 props->max_sge_rd = mdev->limits.max_sg;
95 props->max_cq = mdev->limits.num_cqs - mdev->limits.reserved_cqs;
96 props->max_cqe = mdev->limits.max_cqes;
97 props->max_mr = mdev->limits.num_mpts - mdev->limits.reserved_mrws;
98 props->max_pd = mdev->limits.num_pds - mdev->limits.reserved_pds;
99 props->max_qp_rd_atom = 1 << mdev->qp_table.rdb_shift;
100 props->max_qp_init_rd_atom = mdev->limits.max_qp_init_rdma;
101 props->max_res_rd_atom = props->max_qp_rd_atom * props->max_qp;
102 props->max_srq = mdev->limits.num_srqs - mdev->limits.reserved_srqs;
103 props->max_srq_wr = mdev->limits.max_srq_wqes;
104 props->max_srq_sge = mdev->limits.max_srq_sge;
105 props->local_ca_ack_delay = mdev->limits.local_ca_ack_delay;
106 props->atomic_cap = mdev->limits.flags & DEV_LIM_FLAG_ATOMIC ?
107 IB_ATOMIC_HCA : IB_ATOMIC_NONE;
108 props->max_pkeys = mdev->limits.pkey_table_len;
109 props->max_mcast_grp = mdev->limits.num_mgms + mdev->limits.num_amgms;
110 props->max_mcast_qp_attach = MTHCA_QP_PER_MGM;
111 props->max_total_mcast_qp_attach = props->max_mcast_qp_attach *
112 props->max_mcast_grp;
113
114 err = 0;
115 out:
116 kfree(in_mad);
117 kfree(out_mad);
118 return err;
119 }
120
mthca_query_port(struct ib_device * ibdev,u32 port,struct ib_port_attr * props)121 static int mthca_query_port(struct ib_device *ibdev,
122 u32 port, struct ib_port_attr *props)
123 {
124 struct ib_smp *in_mad;
125 struct ib_smp *out_mad;
126 int err = -ENOMEM;
127
128 in_mad = kzalloc_obj(*in_mad);
129 out_mad = kmalloc_obj(*out_mad);
130 if (!in_mad || !out_mad)
131 goto out;
132
133 /* props being zeroed by the caller, avoid zeroing it here */
134
135 ib_init_query_mad(in_mad);
136 in_mad->attr_id = IB_SMP_ATTR_PORT_INFO;
137 in_mad->attr_mod = cpu_to_be32(port);
138
139 err = mthca_MAD_IFC(to_mdev(ibdev), 1, 1,
140 port, NULL, NULL, in_mad, out_mad);
141 if (err)
142 goto out;
143
144 props->lid = be16_to_cpup((__be16 *) (out_mad->data + 16));
145 props->lmc = out_mad->data[34] & 0x7;
146 props->sm_lid = be16_to_cpup((__be16 *) (out_mad->data + 18));
147 props->sm_sl = out_mad->data[36] & 0xf;
148 props->state = out_mad->data[32] & 0xf;
149 props->phys_state = out_mad->data[33] >> 4;
150 props->port_cap_flags = be32_to_cpup((__be32 *) (out_mad->data + 20));
151 props->gid_tbl_len = to_mdev(ibdev)->limits.gid_table_len;
152 props->max_msg_sz = 0x80000000;
153 props->pkey_tbl_len = to_mdev(ibdev)->limits.pkey_table_len;
154 props->bad_pkey_cntr = be16_to_cpup((__be16 *) (out_mad->data + 46));
155 props->qkey_viol_cntr = be16_to_cpup((__be16 *) (out_mad->data + 48));
156 props->active_width = out_mad->data[31] & 0xf;
157 props->active_speed = out_mad->data[35] >> 4;
158 props->max_mtu = out_mad->data[41] & 0xf;
159 props->active_mtu = out_mad->data[36] >> 4;
160 props->subnet_timeout = out_mad->data[51] & 0x1f;
161 props->max_vl_num = out_mad->data[37] >> 4;
162 props->init_type_reply = out_mad->data[41] >> 4;
163
164 out:
165 kfree(in_mad);
166 kfree(out_mad);
167 return err;
168 }
169
mthca_modify_device(struct ib_device * ibdev,int mask,struct ib_device_modify * props)170 static int mthca_modify_device(struct ib_device *ibdev,
171 int mask,
172 struct ib_device_modify *props)
173 {
174 if (mask & ~IB_DEVICE_MODIFY_NODE_DESC)
175 return -EOPNOTSUPP;
176
177 if (mask & IB_DEVICE_MODIFY_NODE_DESC) {
178 if (mutex_lock_interruptible(&to_mdev(ibdev)->cap_mask_mutex))
179 return -ERESTARTSYS;
180 memcpy(ibdev->node_desc, props->node_desc,
181 IB_DEVICE_NODE_DESC_MAX);
182 mutex_unlock(&to_mdev(ibdev)->cap_mask_mutex);
183 }
184
185 return 0;
186 }
187
mthca_modify_port(struct ib_device * ibdev,u32 port,int port_modify_mask,struct ib_port_modify * props)188 static int mthca_modify_port(struct ib_device *ibdev,
189 u32 port, int port_modify_mask,
190 struct ib_port_modify *props)
191 {
192 struct mthca_set_ib_param set_ib;
193 struct ib_port_attr attr;
194 int err;
195
196 if (mutex_lock_interruptible(&to_mdev(ibdev)->cap_mask_mutex))
197 return -ERESTARTSYS;
198
199 err = ib_query_port(ibdev, port, &attr);
200 if (err)
201 goto out;
202
203 set_ib.set_si_guid = 0;
204 set_ib.reset_qkey_viol = !!(port_modify_mask & IB_PORT_RESET_QKEY_CNTR);
205
206 set_ib.cap_mask = (attr.port_cap_flags | props->set_port_cap_mask) &
207 ~props->clr_port_cap_mask;
208
209 err = mthca_SET_IB(to_mdev(ibdev), &set_ib, port);
210 if (err)
211 goto out;
212 out:
213 mutex_unlock(&to_mdev(ibdev)->cap_mask_mutex);
214 return err;
215 }
216
mthca_query_pkey(struct ib_device * ibdev,u32 port,u16 index,u16 * pkey)217 static int mthca_query_pkey(struct ib_device *ibdev,
218 u32 port, u16 index, u16 *pkey)
219 {
220 struct ib_smp *in_mad;
221 struct ib_smp *out_mad;
222 int err = -ENOMEM;
223
224 in_mad = kzalloc_obj(*in_mad);
225 out_mad = kmalloc_obj(*out_mad);
226 if (!in_mad || !out_mad)
227 goto out;
228
229 ib_init_query_mad(in_mad);
230 in_mad->attr_id = IB_SMP_ATTR_PKEY_TABLE;
231 in_mad->attr_mod = cpu_to_be32(index / 32);
232
233 err = mthca_MAD_IFC(to_mdev(ibdev), 1, 1,
234 port, NULL, NULL, in_mad, out_mad);
235 if (err)
236 goto out;
237
238 *pkey = be16_to_cpu(((__be16 *) out_mad->data)[index % 32]);
239
240 out:
241 kfree(in_mad);
242 kfree(out_mad);
243 return err;
244 }
245
mthca_query_gid(struct ib_device * ibdev,u32 port,int index,union ib_gid * gid)246 static int mthca_query_gid(struct ib_device *ibdev, u32 port,
247 int index, union ib_gid *gid)
248 {
249 struct ib_smp *in_mad;
250 struct ib_smp *out_mad;
251 int err = -ENOMEM;
252
253 in_mad = kzalloc_obj(*in_mad);
254 out_mad = kmalloc_obj(*out_mad);
255 if (!in_mad || !out_mad)
256 goto out;
257
258 ib_init_query_mad(in_mad);
259 in_mad->attr_id = IB_SMP_ATTR_PORT_INFO;
260 in_mad->attr_mod = cpu_to_be32(port);
261
262 err = mthca_MAD_IFC(to_mdev(ibdev), 1, 1,
263 port, NULL, NULL, in_mad, out_mad);
264 if (err)
265 goto out;
266
267 memcpy(gid->raw, out_mad->data + 8, 8);
268
269 ib_init_query_mad(in_mad);
270 in_mad->attr_id = IB_SMP_ATTR_GUID_INFO;
271 in_mad->attr_mod = cpu_to_be32(index / 8);
272
273 err = mthca_MAD_IFC(to_mdev(ibdev), 1, 1,
274 port, NULL, NULL, in_mad, out_mad);
275 if (err)
276 goto out;
277
278 memcpy(gid->raw + 8, out_mad->data + (index % 8) * 8, 8);
279
280 out:
281 kfree(in_mad);
282 kfree(out_mad);
283 return err;
284 }
285
mthca_alloc_ucontext(struct ib_ucontext * uctx,struct ib_udata * udata)286 static int mthca_alloc_ucontext(struct ib_ucontext *uctx,
287 struct ib_udata *udata)
288 {
289 struct ib_device *ibdev = uctx->device;
290 struct mthca_alloc_ucontext_resp uresp = {};
291 struct mthca_ucontext *context = to_mucontext(uctx);
292 int err;
293
294 if (!(to_mdev(ibdev)->active))
295 return -EAGAIN;
296
297 uresp.qp_tab_size = to_mdev(ibdev)->limits.num_qps;
298 if (mthca_is_memfree(to_mdev(ibdev)))
299 uresp.uarc_size = to_mdev(ibdev)->uar_table.uarc_size;
300 else
301 uresp.uarc_size = 0;
302
303 err = mthca_uar_alloc(to_mdev(ibdev), &context->uar);
304 if (err)
305 return err;
306
307 context->db_tab = mthca_init_user_db_tab(to_mdev(ibdev));
308 if (IS_ERR(context->db_tab)) {
309 err = PTR_ERR(context->db_tab);
310 mthca_uar_free(to_mdev(ibdev), &context->uar);
311 return err;
312 }
313
314 if (ib_copy_to_udata(udata, &uresp, sizeof(uresp))) {
315 mthca_cleanup_user_db_tab(to_mdev(ibdev), &context->uar, context->db_tab);
316 mthca_uar_free(to_mdev(ibdev), &context->uar);
317 return -EFAULT;
318 }
319
320 context->reg_mr_warned = 0;
321
322 return 0;
323 }
324
mthca_dealloc_ucontext(struct ib_ucontext * context)325 static void mthca_dealloc_ucontext(struct ib_ucontext *context)
326 {
327 mthca_cleanup_user_db_tab(to_mdev(context->device), &to_mucontext(context)->uar,
328 to_mucontext(context)->db_tab);
329 mthca_uar_free(to_mdev(context->device), &to_mucontext(context)->uar);
330 }
331
mthca_mmap_uar(struct ib_ucontext * context,struct vm_area_struct * vma)332 static int mthca_mmap_uar(struct ib_ucontext *context,
333 struct vm_area_struct *vma)
334 {
335 if (vma->vm_end - vma->vm_start != PAGE_SIZE)
336 return -EINVAL;
337
338 vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
339
340 if (io_remap_pfn_range(vma, vma->vm_start,
341 to_mucontext(context)->uar.pfn,
342 PAGE_SIZE, vma->vm_page_prot))
343 return -EAGAIN;
344
345 return 0;
346 }
347
mthca_alloc_pd(struct ib_pd * ibpd,struct ib_udata * udata)348 static int mthca_alloc_pd(struct ib_pd *ibpd, struct ib_udata *udata)
349 {
350 struct ib_device *ibdev = ibpd->device;
351 struct mthca_pd *pd = to_mpd(ibpd);
352 int err;
353
354 err = mthca_pd_alloc(to_mdev(ibdev), !udata, pd);
355 if (err)
356 return err;
357
358 if (udata) {
359 if (ib_copy_to_udata(udata, &pd->pd_num, sizeof (__u32))) {
360 mthca_pd_free(to_mdev(ibdev), pd);
361 return -EFAULT;
362 }
363 }
364
365 return 0;
366 }
367
mthca_dealloc_pd(struct ib_pd * pd,struct ib_udata * udata)368 static int mthca_dealloc_pd(struct ib_pd *pd, struct ib_udata *udata)
369 {
370 mthca_pd_free(to_mdev(pd->device), to_mpd(pd));
371 return 0;
372 }
373
mthca_ah_create(struct ib_ah * ibah,struct rdma_ah_init_attr * init_attr,struct ib_udata * udata)374 static int mthca_ah_create(struct ib_ah *ibah,
375 struct rdma_ah_init_attr *init_attr,
376 struct ib_udata *udata)
377
378 {
379 struct mthca_ah *ah = to_mah(ibah);
380
381 return mthca_create_ah(to_mdev(ibah->device), to_mpd(ibah->pd),
382 init_attr->ah_attr, ah);
383 }
384
mthca_ah_destroy(struct ib_ah * ah,u32 flags)385 static int mthca_ah_destroy(struct ib_ah *ah, u32 flags)
386 {
387 mthca_destroy_ah(to_mdev(ah->device), to_mah(ah));
388 return 0;
389 }
390
mthca_create_srq(struct ib_srq * ibsrq,struct ib_srq_init_attr * init_attr,struct ib_udata * udata)391 static int mthca_create_srq(struct ib_srq *ibsrq,
392 struct ib_srq_init_attr *init_attr,
393 struct ib_udata *udata)
394 {
395 struct mthca_create_srq ucmd;
396 struct mthca_ucontext *context = rdma_udata_to_drv_context(
397 udata, struct mthca_ucontext, ibucontext);
398 struct mthca_srq *srq = to_msrq(ibsrq);
399 int err;
400
401 if (init_attr->srq_type != IB_SRQT_BASIC)
402 return -EOPNOTSUPP;
403
404 if (udata) {
405 err = ib_copy_validate_udata_in(udata, ucmd, db_page);
406 if (err)
407 return err;
408
409 err = mthca_map_user_db(to_mdev(ibsrq->device), &context->uar,
410 context->db_tab, ucmd.db_index,
411 ucmd.db_page);
412
413 if (err)
414 return err;
415
416 srq->mr.ibmr.lkey = ucmd.lkey;
417 srq->db_index = ucmd.db_index;
418 }
419
420 err = mthca_alloc_srq(to_mdev(ibsrq->device), to_mpd(ibsrq->pd),
421 &init_attr->attr, srq, udata);
422
423 if (err && udata)
424 mthca_unmap_user_db(to_mdev(ibsrq->device), &context->uar,
425 context->db_tab, ucmd.db_index);
426
427 if (err)
428 return err;
429
430 if (context && ib_copy_to_udata(udata, &srq->srqn, sizeof(__u32))) {
431 mthca_free_srq(to_mdev(ibsrq->device), srq);
432 mthca_unmap_user_db(to_mdev(ibsrq->device), &context->uar,
433 context->db_tab, ucmd.db_index);
434 return -EFAULT;
435 }
436
437 return 0;
438 }
439
mthca_destroy_srq(struct ib_srq * srq,struct ib_udata * udata)440 static int mthca_destroy_srq(struct ib_srq *srq, struct ib_udata *udata)
441 {
442 mthca_free_srq(to_mdev(srq->device), to_msrq(srq));
443 if (udata) {
444 struct mthca_ucontext *context =
445 rdma_udata_to_drv_context(
446 udata,
447 struct mthca_ucontext,
448 ibucontext);
449
450 mthca_unmap_user_db(to_mdev(srq->device), &context->uar,
451 context->db_tab, to_msrq(srq)->db_index);
452 }
453 return 0;
454 }
455
mthca_create_qp(struct ib_qp * ibqp,struct ib_qp_init_attr * init_attr,struct ib_udata * udata)456 static int mthca_create_qp(struct ib_qp *ibqp,
457 struct ib_qp_init_attr *init_attr,
458 struct ib_udata *udata)
459 {
460 struct mthca_ucontext *context = rdma_udata_to_drv_context(
461 udata, struct mthca_ucontext, ibucontext);
462 struct mthca_create_qp ucmd;
463 struct mthca_qp *qp = to_mqp(ibqp);
464 struct mthca_dev *dev = to_mdev(ibqp->device);
465 int err;
466
467 if (init_attr->create_flags)
468 return -EOPNOTSUPP;
469
470 switch (init_attr->qp_type) {
471 case IB_QPT_RC:
472 case IB_QPT_UC:
473 case IB_QPT_UD:
474 {
475 if (udata) {
476 err = ib_copy_validate_udata_in(udata, ucmd, rq_db_index);
477 if (err)
478 return err;
479
480 err = mthca_map_user_db(dev, &context->uar,
481 context->db_tab,
482 ucmd.sq_db_index,
483 ucmd.sq_db_page);
484 if (err)
485 return err;
486
487 err = mthca_map_user_db(dev, &context->uar,
488 context->db_tab,
489 ucmd.rq_db_index,
490 ucmd.rq_db_page);
491 if (err) {
492 mthca_unmap_user_db(dev, &context->uar,
493 context->db_tab,
494 ucmd.sq_db_index);
495 return err;
496 }
497
498 qp->mr.ibmr.lkey = ucmd.lkey;
499 qp->sq.db_index = ucmd.sq_db_index;
500 qp->rq.db_index = ucmd.rq_db_index;
501 }
502
503 err = mthca_alloc_qp(dev, to_mpd(ibqp->pd),
504 to_mcq(init_attr->send_cq),
505 to_mcq(init_attr->recv_cq),
506 init_attr->qp_type, init_attr->sq_sig_type,
507 &init_attr->cap, qp, udata);
508
509 if (err && udata) {
510 mthca_unmap_user_db(dev, &context->uar, context->db_tab,
511 ucmd.sq_db_index);
512 mthca_unmap_user_db(dev, &context->uar, context->db_tab,
513 ucmd.rq_db_index);
514 }
515
516 qp->ibqp.qp_num = qp->qpn;
517 break;
518 }
519 case IB_QPT_SMI:
520 case IB_QPT_GSI:
521 {
522 qp->sqp = kzalloc_obj(struct mthca_sqp);
523 if (!qp->sqp)
524 return -ENOMEM;
525
526 qp->ibqp.qp_num = init_attr->qp_type == IB_QPT_SMI ? 0 : 1;
527
528 err = mthca_alloc_sqp(dev, to_mpd(ibqp->pd),
529 to_mcq(init_attr->send_cq),
530 to_mcq(init_attr->recv_cq),
531 init_attr->sq_sig_type, &init_attr->cap,
532 qp->ibqp.qp_num, init_attr->port_num, qp,
533 udata);
534 break;
535 }
536 default:
537 /* Don't support raw QPs */
538 return -EOPNOTSUPP;
539 }
540
541 if (err) {
542 kfree(qp->sqp);
543 return err;
544 }
545
546 init_attr->cap.max_send_wr = qp->sq.max;
547 init_attr->cap.max_recv_wr = qp->rq.max;
548 init_attr->cap.max_send_sge = qp->sq.max_gs;
549 init_attr->cap.max_recv_sge = qp->rq.max_gs;
550 init_attr->cap.max_inline_data = qp->max_inline_data;
551
552 return 0;
553 }
554
mthca_destroy_qp(struct ib_qp * qp,struct ib_udata * udata)555 static int mthca_destroy_qp(struct ib_qp *qp, struct ib_udata *udata)
556 {
557 if (udata) {
558 struct mthca_ucontext *context =
559 rdma_udata_to_drv_context(
560 udata,
561 struct mthca_ucontext,
562 ibucontext);
563
564 mthca_unmap_user_db(to_mdev(qp->device),
565 &context->uar,
566 context->db_tab,
567 to_mqp(qp)->sq.db_index);
568 mthca_unmap_user_db(to_mdev(qp->device),
569 &context->uar,
570 context->db_tab,
571 to_mqp(qp)->rq.db_index);
572 }
573 mthca_free_qp(to_mdev(qp->device), to_mqp(qp));
574 kfree(to_mqp(qp)->sqp);
575 return 0;
576 }
577
mthca_create_cq(struct ib_cq * ibcq,const struct ib_cq_init_attr * attr,struct uverbs_attr_bundle * attrs)578 static int mthca_create_cq(struct ib_cq *ibcq,
579 const struct ib_cq_init_attr *attr,
580 struct uverbs_attr_bundle *attrs)
581 {
582 struct ib_udata *udata = &attrs->driver_udata;
583 struct ib_device *ibdev = ibcq->device;
584 int entries = attr->cqe;
585 struct mthca_create_cq ucmd;
586 struct mthca_cq *cq;
587 int nent;
588 int err;
589 struct mthca_ucontext *context = rdma_udata_to_drv_context(
590 udata, struct mthca_ucontext, ibucontext);
591
592 if (attr->flags)
593 return -EOPNOTSUPP;
594
595 if (entries < 1 || entries > to_mdev(ibdev)->limits.max_cqes)
596 return -EINVAL;
597
598 if (udata) {
599 err = ib_copy_validate_udata_in(udata, ucmd, set_db_index);
600 if (err)
601 return err;
602
603 err = mthca_map_user_db(to_mdev(ibdev), &context->uar,
604 context->db_tab, ucmd.set_db_index,
605 ucmd.set_db_page);
606 if (err)
607 return err;
608
609 err = mthca_map_user_db(to_mdev(ibdev), &context->uar,
610 context->db_tab, ucmd.arm_db_index,
611 ucmd.arm_db_page);
612 if (err)
613 goto err_unmap_set;
614 }
615
616 cq = to_mcq(ibcq);
617
618 if (udata) {
619 cq->buf.mr.ibmr.lkey = ucmd.lkey;
620 cq->set_ci_db_index = ucmd.set_db_index;
621 cq->arm_db_index = ucmd.arm_db_index;
622 }
623
624 for (nent = 1; nent <= entries; nent <<= 1)
625 ; /* nothing */
626
627 err = mthca_init_cq(to_mdev(ibdev), nent, context,
628 udata ? ucmd.pdn : to_mdev(ibdev)->driver_pd.pd_num,
629 cq);
630 if (err)
631 goto err_unmap_arm;
632
633 if (udata && ib_copy_to_udata(udata, &cq->cqn, sizeof(__u32))) {
634 mthca_free_cq(to_mdev(ibdev), cq);
635 err = -EFAULT;
636 goto err_unmap_arm;
637 }
638
639 cq->resize_buf = NULL;
640
641 return 0;
642
643 err_unmap_arm:
644 if (udata)
645 mthca_unmap_user_db(to_mdev(ibdev), &context->uar,
646 context->db_tab, ucmd.arm_db_index);
647
648 err_unmap_set:
649 if (udata)
650 mthca_unmap_user_db(to_mdev(ibdev), &context->uar,
651 context->db_tab, ucmd.set_db_index);
652
653 return err;
654 }
655
mthca_alloc_resize_buf(struct mthca_dev * dev,struct mthca_cq * cq,int entries)656 static int mthca_alloc_resize_buf(struct mthca_dev *dev, struct mthca_cq *cq,
657 int entries)
658 {
659 int ret;
660
661 spin_lock_irq(&cq->lock);
662 if (cq->resize_buf) {
663 ret = -EBUSY;
664 goto unlock;
665 }
666
667 cq->resize_buf = kmalloc_obj(*cq->resize_buf, GFP_ATOMIC);
668 if (!cq->resize_buf) {
669 ret = -ENOMEM;
670 goto unlock;
671 }
672
673 cq->resize_buf->state = CQ_RESIZE_ALLOC;
674
675 ret = 0;
676
677 unlock:
678 spin_unlock_irq(&cq->lock);
679
680 if (ret)
681 return ret;
682
683 ret = mthca_alloc_cq_buf(dev, &cq->resize_buf->buf, entries);
684 if (ret) {
685 spin_lock_irq(&cq->lock);
686 kfree(cq->resize_buf);
687 cq->resize_buf = NULL;
688 spin_unlock_irq(&cq->lock);
689 return ret;
690 }
691
692 cq->resize_buf->cqe = entries - 1;
693
694 spin_lock_irq(&cq->lock);
695 cq->resize_buf->state = CQ_RESIZE_READY;
696 spin_unlock_irq(&cq->lock);
697
698 return 0;
699 }
700
mthca_resize_cq(struct ib_cq * ibcq,unsigned int entries,struct ib_udata * udata)701 static int mthca_resize_cq(struct ib_cq *ibcq, unsigned int entries,
702 struct ib_udata *udata)
703 {
704 struct mthca_dev *dev = to_mdev(ibcq->device);
705 struct mthca_cq *cq = to_mcq(ibcq);
706 struct mthca_resize_cq ucmd;
707 u32 lkey;
708 int ret;
709
710 if (entries > dev->limits.max_cqes)
711 return -EINVAL;
712
713 mutex_lock(&cq->mutex);
714
715 entries = roundup_pow_of_two(entries + 1);
716 if (entries == ibcq->cqe + 1) {
717 ret = 0;
718 goto out;
719 }
720
721 if (cq->is_kernel) {
722 ret = mthca_alloc_resize_buf(dev, cq, entries);
723 if (ret)
724 goto out;
725 lkey = cq->resize_buf->buf.mr.ibmr.lkey;
726 } else {
727 ret = ib_copy_validate_udata_in(udata, ucmd, reserved);
728 if (ret)
729 goto out;
730 lkey = ucmd.lkey;
731 }
732
733 ret = mthca_RESIZE_CQ(dev, cq->cqn, lkey, ilog2(entries));
734
735 if (ret) {
736 if (cq->resize_buf) {
737 mthca_free_cq_buf(dev, &cq->resize_buf->buf,
738 cq->resize_buf->cqe);
739 kfree(cq->resize_buf);
740 spin_lock_irq(&cq->lock);
741 cq->resize_buf = NULL;
742 spin_unlock_irq(&cq->lock);
743 }
744 goto out;
745 }
746
747 if (cq->is_kernel) {
748 struct mthca_cq_buf tbuf;
749 int tcqe;
750
751 spin_lock_irq(&cq->lock);
752 if (cq->resize_buf->state == CQ_RESIZE_READY) {
753 mthca_cq_resize_copy_cqes(cq);
754 tbuf = cq->buf;
755 tcqe = cq->ibcq.cqe;
756 cq->buf = cq->resize_buf->buf;
757 cq->ibcq.cqe = cq->resize_buf->cqe;
758 } else {
759 tbuf = cq->resize_buf->buf;
760 tcqe = cq->resize_buf->cqe;
761 }
762
763 kfree(cq->resize_buf);
764 cq->resize_buf = NULL;
765 spin_unlock_irq(&cq->lock);
766
767 mthca_free_cq_buf(dev, &tbuf, tcqe);
768 } else
769 ibcq->cqe = entries - 1;
770
771 out:
772 mutex_unlock(&cq->mutex);
773
774 return ret;
775 }
776
mthca_destroy_cq(struct ib_cq * cq,struct ib_udata * udata)777 static int mthca_destroy_cq(struct ib_cq *cq, struct ib_udata *udata)
778 {
779 if (udata) {
780 struct mthca_ucontext *context =
781 rdma_udata_to_drv_context(
782 udata,
783 struct mthca_ucontext,
784 ibucontext);
785
786 mthca_unmap_user_db(to_mdev(cq->device),
787 &context->uar,
788 context->db_tab,
789 to_mcq(cq)->arm_db_index);
790 mthca_unmap_user_db(to_mdev(cq->device),
791 &context->uar,
792 context->db_tab,
793 to_mcq(cq)->set_ci_db_index);
794 }
795 mthca_free_cq(to_mdev(cq->device), to_mcq(cq));
796 return 0;
797 }
798
convert_access(int acc)799 static inline u32 convert_access(int acc)
800 {
801 return (acc & IB_ACCESS_REMOTE_ATOMIC ? MTHCA_MPT_FLAG_ATOMIC : 0) |
802 (acc & IB_ACCESS_REMOTE_WRITE ? MTHCA_MPT_FLAG_REMOTE_WRITE : 0) |
803 (acc & IB_ACCESS_REMOTE_READ ? MTHCA_MPT_FLAG_REMOTE_READ : 0) |
804 (acc & IB_ACCESS_LOCAL_WRITE ? MTHCA_MPT_FLAG_LOCAL_WRITE : 0) |
805 MTHCA_MPT_FLAG_LOCAL_READ;
806 }
807
mthca_get_dma_mr(struct ib_pd * pd,int acc)808 static struct ib_mr *mthca_get_dma_mr(struct ib_pd *pd, int acc)
809 {
810 struct mthca_mr *mr;
811 int err;
812
813 mr = kmalloc_obj(*mr);
814 if (!mr)
815 return ERR_PTR(-ENOMEM);
816
817 err = mthca_mr_alloc_notrans(to_mdev(pd->device),
818 to_mpd(pd)->pd_num,
819 convert_access(acc), mr);
820
821 if (err) {
822 kfree(mr);
823 return ERR_PTR(err);
824 }
825
826 mr->umem = NULL;
827
828 return &mr->ibmr;
829 }
830
mthca_reg_user_mr(struct ib_pd * pd,u64 start,u64 length,u64 virt,int acc,struct ib_dmah * dmah,struct ib_udata * udata)831 static struct ib_mr *mthca_reg_user_mr(struct ib_pd *pd, u64 start, u64 length,
832 u64 virt, int acc, struct ib_dmah *dmah,
833 struct ib_udata *udata)
834 {
835 struct mthca_dev *dev = to_mdev(pd->device);
836 struct ib_block_iter biter;
837 struct mthca_ucontext *context = rdma_udata_to_drv_context(
838 udata, struct mthca_ucontext, ibucontext);
839 struct mthca_mr *mr;
840 struct mthca_reg_mr ucmd;
841 u64 *pages;
842 int n, i;
843 int err = 0;
844 int write_mtt_size;
845
846 if (dmah)
847 return ERR_PTR(-EOPNOTSUPP);
848
849 if (udata->inlen < sizeof ucmd) {
850 if (!context->reg_mr_warned) {
851 mthca_warn(dev, "Process '%s' did not pass in MR attrs.\n",
852 current->comm);
853 mthca_warn(dev, " Update libmthca to fix this.\n");
854 }
855 ++context->reg_mr_warned;
856 ucmd.mr_attrs = 0;
857 } else {
858 err = ib_copy_validate_udata_in(udata, ucmd, reserved);
859 if (err)
860 return ERR_PTR(err);
861 }
862
863 mr = kmalloc_obj(*mr);
864 if (!mr)
865 return ERR_PTR(-ENOMEM);
866
867 mr->umem = ib_umem_get(pd->device, start, length, acc);
868 if (IS_ERR(mr->umem)) {
869 err = PTR_ERR(mr->umem);
870 goto err;
871 }
872
873 n = ib_umem_num_dma_blocks(mr->umem, PAGE_SIZE);
874
875 mr->mtt = mthca_alloc_mtt(dev, n);
876 if (IS_ERR(mr->mtt)) {
877 err = PTR_ERR(mr->mtt);
878 goto err_umem;
879 }
880
881 pages = (u64 *) __get_free_page(GFP_KERNEL);
882 if (!pages) {
883 err = -ENOMEM;
884 goto err_mtt;
885 }
886
887 i = n = 0;
888
889 write_mtt_size = min(mthca_write_mtt_size(dev), (int) (PAGE_SIZE / sizeof *pages));
890
891 rdma_umem_for_each_dma_block(mr->umem, &biter, PAGE_SIZE) {
892 pages[i++] = rdma_block_iter_dma_address(&biter);
893
894 /*
895 * Be friendly to write_mtt and pass it chunks
896 * of appropriate size.
897 */
898 if (i == write_mtt_size) {
899 err = mthca_write_mtt(dev, mr->mtt, n, pages, i);
900 if (err)
901 goto mtt_done;
902 n += i;
903 i = 0;
904 }
905 }
906
907 if (i)
908 err = mthca_write_mtt(dev, mr->mtt, n, pages, i);
909 mtt_done:
910 free_page((unsigned long) pages);
911 if (err)
912 goto err_mtt;
913
914 err = mthca_mr_alloc(dev, to_mpd(pd)->pd_num, PAGE_SHIFT, virt, length,
915 convert_access(acc), mr);
916
917 if (err)
918 goto err_mtt;
919
920 return &mr->ibmr;
921
922 err_mtt:
923 mthca_free_mtt(dev, mr->mtt);
924
925 err_umem:
926 ib_umem_release(mr->umem);
927
928 err:
929 kfree(mr);
930 return ERR_PTR(err);
931 }
932
mthca_dereg_mr(struct ib_mr * mr,struct ib_udata * udata)933 static int mthca_dereg_mr(struct ib_mr *mr, struct ib_udata *udata)
934 {
935 struct mthca_mr *mmr = to_mmr(mr);
936
937 mthca_free_mr(to_mdev(mr->device), mmr);
938 ib_umem_release(mmr->umem);
939 kfree(mmr);
940
941 return 0;
942 }
943
hw_rev_show(struct device * device,struct device_attribute * attr,char * buf)944 static ssize_t hw_rev_show(struct device *device,
945 struct device_attribute *attr, char *buf)
946 {
947 struct mthca_dev *dev =
948 rdma_device_to_drv_device(device, struct mthca_dev, ib_dev);
949
950 return sysfs_emit(buf, "%x\n", dev->rev_id);
951 }
952 static DEVICE_ATTR_RO(hw_rev);
953
hca_type_string(int hca_type)954 static const char *hca_type_string(int hca_type)
955 {
956 switch (hca_type) {
957 case PCI_DEVICE_ID_MELLANOX_TAVOR:
958 return "MT23108";
959 case PCI_DEVICE_ID_MELLANOX_ARBEL_COMPAT:
960 return "MT25208 (MT23108 compat mode)";
961 case PCI_DEVICE_ID_MELLANOX_ARBEL:
962 return "MT25208";
963 case PCI_DEVICE_ID_MELLANOX_SINAI:
964 case PCI_DEVICE_ID_MELLANOX_SINAI_OLD:
965 return "MT25204";
966 }
967
968 return "unknown";
969 }
970
hca_type_show(struct device * device,struct device_attribute * attr,char * buf)971 static ssize_t hca_type_show(struct device *device,
972 struct device_attribute *attr, char *buf)
973 {
974 struct mthca_dev *dev =
975 rdma_device_to_drv_device(device, struct mthca_dev, ib_dev);
976
977 return sysfs_emit(buf, "%s\n", hca_type_string(dev->pdev->device));
978 }
979 static DEVICE_ATTR_RO(hca_type);
980
board_id_show(struct device * device,struct device_attribute * attr,char * buf)981 static ssize_t board_id_show(struct device *device,
982 struct device_attribute *attr, char *buf)
983 {
984 struct mthca_dev *dev =
985 rdma_device_to_drv_device(device, struct mthca_dev, ib_dev);
986
987 return sysfs_emit(buf, "%.*s\n", MTHCA_BOARD_ID_LEN, dev->board_id);
988 }
989 static DEVICE_ATTR_RO(board_id);
990
991 static struct attribute *mthca_dev_attributes[] = {
992 &dev_attr_hw_rev.attr,
993 &dev_attr_hca_type.attr,
994 &dev_attr_board_id.attr,
995 NULL
996 };
997
998 static const struct attribute_group mthca_attr_group = {
999 .attrs = mthca_dev_attributes,
1000 };
1001
mthca_init_node_data(struct mthca_dev * dev)1002 static int mthca_init_node_data(struct mthca_dev *dev)
1003 {
1004 struct ib_smp *in_mad;
1005 struct ib_smp *out_mad;
1006 int err = -ENOMEM;
1007
1008 in_mad = kzalloc_obj(*in_mad);
1009 out_mad = kmalloc_obj(*out_mad);
1010 if (!in_mad || !out_mad)
1011 goto out;
1012
1013 ib_init_query_mad(in_mad);
1014 in_mad->attr_id = IB_SMP_ATTR_NODE_DESC;
1015
1016 err = mthca_MAD_IFC(dev, 1, 1,
1017 1, NULL, NULL, in_mad, out_mad);
1018 if (err)
1019 goto out;
1020
1021 memcpy(dev->ib_dev.node_desc, out_mad->data, IB_DEVICE_NODE_DESC_MAX);
1022
1023 in_mad->attr_id = IB_SMP_ATTR_NODE_INFO;
1024
1025 err = mthca_MAD_IFC(dev, 1, 1,
1026 1, NULL, NULL, in_mad, out_mad);
1027 if (err)
1028 goto out;
1029
1030 if (mthca_is_memfree(dev))
1031 dev->rev_id = be32_to_cpup((__be32 *) (out_mad->data + 32));
1032 memcpy(&dev->ib_dev.node_guid, out_mad->data + 12, 8);
1033
1034 out:
1035 kfree(in_mad);
1036 kfree(out_mad);
1037 return err;
1038 }
1039
mthca_port_immutable(struct ib_device * ibdev,u32 port_num,struct ib_port_immutable * immutable)1040 static int mthca_port_immutable(struct ib_device *ibdev, u32 port_num,
1041 struct ib_port_immutable *immutable)
1042 {
1043 struct ib_port_attr attr;
1044 int err;
1045
1046 immutable->core_cap_flags = RDMA_CORE_PORT_IBA_IB;
1047
1048 err = ib_query_port(ibdev, port_num, &attr);
1049 if (err)
1050 return err;
1051
1052 immutable->pkey_tbl_len = attr.pkey_tbl_len;
1053 immutable->gid_tbl_len = attr.gid_tbl_len;
1054 immutable->max_mad_size = IB_MGMT_MAD_SIZE;
1055
1056 return 0;
1057 }
1058
get_dev_fw_str(struct ib_device * device,char * str)1059 static void get_dev_fw_str(struct ib_device *device, char *str)
1060 {
1061 struct mthca_dev *dev =
1062 container_of(device, struct mthca_dev, ib_dev);
1063 snprintf(str, IB_FW_VERSION_NAME_MAX, "%d.%d.%d",
1064 (int) (dev->fw_ver >> 32),
1065 (int) (dev->fw_ver >> 16) & 0xffff,
1066 (int) dev->fw_ver & 0xffff);
1067 }
1068
1069 static const struct ib_device_ops mthca_dev_ops = {
1070 .owner = THIS_MODULE,
1071 .driver_id = RDMA_DRIVER_MTHCA,
1072 .uverbs_abi_ver = MTHCA_UVERBS_ABI_VERSION,
1073 .uverbs_no_driver_id_binding = 1,
1074
1075 .alloc_pd = mthca_alloc_pd,
1076 .alloc_ucontext = mthca_alloc_ucontext,
1077 .attach_mcast = mthca_multicast_attach,
1078 .create_ah = mthca_ah_create,
1079 .create_cq = mthca_create_cq,
1080 .create_qp = mthca_create_qp,
1081 .dealloc_pd = mthca_dealloc_pd,
1082 .dealloc_ucontext = mthca_dealloc_ucontext,
1083 .dereg_mr = mthca_dereg_mr,
1084 .destroy_ah = mthca_ah_destroy,
1085 .destroy_cq = mthca_destroy_cq,
1086 .destroy_qp = mthca_destroy_qp,
1087 .detach_mcast = mthca_multicast_detach,
1088 .device_group = &mthca_attr_group,
1089 .get_dev_fw_str = get_dev_fw_str,
1090 .get_dma_mr = mthca_get_dma_mr,
1091 .get_port_immutable = mthca_port_immutable,
1092 .mmap = mthca_mmap_uar,
1093 .modify_device = mthca_modify_device,
1094 .modify_port = mthca_modify_port,
1095 .modify_qp = mthca_modify_qp,
1096 .poll_cq = mthca_poll_cq,
1097 .process_mad = mthca_process_mad,
1098 .query_ah = mthca_ah_query,
1099 .query_device = mthca_query_device,
1100 .query_gid = mthca_query_gid,
1101 .query_pkey = mthca_query_pkey,
1102 .query_port = mthca_query_port,
1103 .query_qp = mthca_query_qp,
1104 .reg_user_mr = mthca_reg_user_mr,
1105 .resize_user_cq = mthca_resize_cq,
1106
1107 INIT_RDMA_OBJ_SIZE(ib_ah, mthca_ah, ibah),
1108 INIT_RDMA_OBJ_SIZE(ib_cq, mthca_cq, ibcq),
1109 INIT_RDMA_OBJ_SIZE(ib_pd, mthca_pd, ibpd),
1110 INIT_RDMA_OBJ_SIZE(ib_qp, mthca_qp, ibqp),
1111 INIT_RDMA_OBJ_SIZE(ib_ucontext, mthca_ucontext, ibucontext),
1112 };
1113
1114 static const struct ib_device_ops mthca_dev_arbel_srq_ops = {
1115 .create_srq = mthca_create_srq,
1116 .destroy_srq = mthca_destroy_srq,
1117 .modify_srq = mthca_modify_srq,
1118 .post_srq_recv = mthca_arbel_post_srq_recv,
1119 .query_srq = mthca_query_srq,
1120
1121 INIT_RDMA_OBJ_SIZE(ib_srq, mthca_srq, ibsrq),
1122 };
1123
1124 static const struct ib_device_ops mthca_dev_tavor_srq_ops = {
1125 .create_srq = mthca_create_srq,
1126 .destroy_srq = mthca_destroy_srq,
1127 .modify_srq = mthca_modify_srq,
1128 .post_srq_recv = mthca_tavor_post_srq_recv,
1129 .query_srq = mthca_query_srq,
1130
1131 INIT_RDMA_OBJ_SIZE(ib_srq, mthca_srq, ibsrq),
1132 };
1133
1134 static const struct ib_device_ops mthca_dev_arbel_ops = {
1135 .post_recv = mthca_arbel_post_receive,
1136 .post_send = mthca_arbel_post_send,
1137 .req_notify_cq = mthca_arbel_arm_cq,
1138 };
1139
1140 static const struct ib_device_ops mthca_dev_tavor_ops = {
1141 .post_recv = mthca_tavor_post_receive,
1142 .post_send = mthca_tavor_post_send,
1143 .req_notify_cq = mthca_tavor_arm_cq,
1144 };
1145
mthca_register_device(struct mthca_dev * dev)1146 int mthca_register_device(struct mthca_dev *dev)
1147 {
1148 int ret;
1149
1150 ret = mthca_init_node_data(dev);
1151 if (ret)
1152 return ret;
1153
1154 dev->ib_dev.node_type = RDMA_NODE_IB_CA;
1155 dev->ib_dev.phys_port_cnt = dev->limits.num_ports;
1156 dev->ib_dev.num_comp_vectors = 1;
1157 dev->ib_dev.dev.parent = &dev->pdev->dev;
1158
1159 if (dev->mthca_flags & MTHCA_FLAG_SRQ) {
1160 if (mthca_is_memfree(dev))
1161 ib_set_device_ops(&dev->ib_dev,
1162 &mthca_dev_arbel_srq_ops);
1163 else
1164 ib_set_device_ops(&dev->ib_dev,
1165 &mthca_dev_tavor_srq_ops);
1166 }
1167
1168 ib_set_device_ops(&dev->ib_dev, &mthca_dev_ops);
1169
1170 if (mthca_is_memfree(dev))
1171 ib_set_device_ops(&dev->ib_dev, &mthca_dev_arbel_ops);
1172 else
1173 ib_set_device_ops(&dev->ib_dev, &mthca_dev_tavor_ops);
1174
1175 mutex_init(&dev->cap_mask_mutex);
1176
1177 ret = ib_register_device(&dev->ib_dev, "mthca%d", &dev->pdev->dev);
1178 if (ret)
1179 return ret;
1180
1181 mthca_start_catas_poll(dev);
1182
1183 return 0;
1184 }
1185
mthca_unregister_device(struct mthca_dev * dev)1186 void mthca_unregister_device(struct mthca_dev *dev)
1187 {
1188 mthca_stop_catas_poll(dev);
1189 ib_unregister_device(&dev->ib_dev);
1190 }
1191