]> git.proxmox.com Git - mirror_ubuntu-hirsute-kernel.git/blame - drivers/infiniband/hw/ipath/ipath_verbs.c
projects / mirror_ubuntu-hirsute-kernel.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
IB/ipath: Misc changes to prepare for IB7220 introduction
[mirror_ubuntu-hirsute-kernel.git] / drivers / infiniband / hw / ipath / ipath_verbs.c
CommitLineData
6522108f 1/*
87427da5 2 * Copyright (c) 2006, 2007 QLogic Corporation. All rights reserved.
6522108f
BS		 3 * Copyright (c) 2005, 2006 PathScale, Inc. All rights reserved.
4 *
5 * This software is available to you under a choice of one of two
6 * licenses. You may choose to be licensed under the terms of the GNU
7 * General Public License (GPL) Version 2, available from the file
8 * COPYING in the main directory of this source tree, or the
9 * OpenIB.org BSD license below:
10 *
11 * Redistribution and use in source and binary forms, with or
12 * without modification, are permitted provided that the following
13 * conditions are met:
14 *
15 * - Redistributions of source code must retain the above
16 * copyright notice, this list of conditions and the following
17 * disclaimer.
18 *
19 * - Redistributions in binary form must reproduce the above
20 * copyright notice, this list of conditions and the following
21 * disclaimer in the documentation and/or other materials
22 * provided with the distribution.
23 *
24 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
25 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
26 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
27 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
28 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
29 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
30 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
31 * SOFTWARE.
32 */
33
34#include <rdma/ib_mad.h>
35#include <rdma/ib_user_verbs.h>
34b2aafe 36#include <linux/io.h>
6522108f
BS		 37#include <linux/utsname.h>
38
39#include "ipath_kernel.h"
40#include "ipath_verbs.h"
27b678dd 41#include "ipath_common.h"
6522108f 42
ac2ae4c9 43static unsigned int ib_ipath_qp_table_size = 251;
6522108f
BS		 44module_param_named(qp_table_size, ib_ipath_qp_table_size, uint, S_IRUGO);
45MODULE_PARM_DESC(qp_table_size, "QP table size");
46
47unsigned int ib_ipath_lkey_table_size = 12;
48module_param_named(lkey_table_size, ib_ipath_lkey_table_size, uint,
49 S_IRUGO);
50MODULE_PARM_DESC(lkey_table_size,
51 "LKEY table size in bits (2^n, 1 <= n <= 23)");
52
fe62546a
BS		 53static unsigned int ib_ipath_max_pds = 0xFFFF;
54module_param_named(max_pds, ib_ipath_max_pds, uint, S_IWUSR | S_IRUGO);
55MODULE_PARM_DESC(max_pds,
56 "Maximum number of protection domains to support");
57
58static unsigned int ib_ipath_max_ahs = 0xFFFF;
59module_param_named(max_ahs, ib_ipath_max_ahs, uint, S_IWUSR | S_IRUGO);
60MODULE_PARM_DESC(max_ahs, "Maximum number of address handles to support");
61
62unsigned int ib_ipath_max_cqes = 0x2FFFF;
63module_param_named(max_cqes, ib_ipath_max_cqes, uint, S_IWUSR | S_IRUGO);
64MODULE_PARM_DESC(max_cqes,
65 "Maximum number of completion queue entries to support");
66
67unsigned int ib_ipath_max_cqs = 0x1FFFF;
68module_param_named(max_cqs, ib_ipath_max_cqs, uint, S_IWUSR | S_IRUGO);
69MODULE_PARM_DESC(max_cqs, "Maximum number of completion queues to support");
70
71unsigned int ib_ipath_max_qp_wrs = 0x3FFF;
72module_param_named(max_qp_wrs, ib_ipath_max_qp_wrs, uint,
73 S_IWUSR | S_IRUGO);
74MODULE_PARM_DESC(max_qp_wrs, "Maximum number of QP WRs to support");
75
0b81e4f7
BS		 76unsigned int ib_ipath_max_qps = 16384;
77module_param_named(max_qps, ib_ipath_max_qps, uint, S_IWUSR | S_IRUGO);
78MODULE_PARM_DESC(max_qps, "Maximum number of QPs to support");
79
fe62546a
BS		 80unsigned int ib_ipath_max_sges = 0x60;
81module_param_named(max_sges, ib_ipath_max_sges, uint, S_IWUSR | S_IRUGO);
82MODULE_PARM_DESC(max_sges, "Maximum number of SGEs to support");
83
84unsigned int ib_ipath_max_mcast_grps = 16384;
85module_param_named(max_mcast_grps, ib_ipath_max_mcast_grps, uint,
86 S_IWUSR | S_IRUGO);
87MODULE_PARM_DESC(max_mcast_grps,
88 "Maximum number of multicast groups to support");
89
90unsigned int ib_ipath_max_mcast_qp_attached = 16;
91module_param_named(max_mcast_qp_attached, ib_ipath_max_mcast_qp_attached,
92 uint, S_IWUSR | S_IRUGO);
93MODULE_PARM_DESC(max_mcast_qp_attached,
94 "Maximum number of attached QPs to support");
95
96unsigned int ib_ipath_max_srqs = 1024;
97module_param_named(max_srqs, ib_ipath_max_srqs, uint, S_IWUSR | S_IRUGO);
98MODULE_PARM_DESC(max_srqs, "Maximum number of SRQs to support");
99
100unsigned int ib_ipath_max_srq_sges = 128;
101module_param_named(max_srq_sges, ib_ipath_max_srq_sges,
102 uint, S_IWUSR | S_IRUGO);
103MODULE_PARM_DESC(max_srq_sges, "Maximum number of SRQ SGEs to support");
104
105unsigned int ib_ipath_max_srq_wrs = 0x1FFFF;
106module_param_named(max_srq_wrs, ib_ipath_max_srq_wrs,
107 uint, S_IWUSR | S_IRUGO);
108MODULE_PARM_DESC(max_srq_wrs, "Maximum number of SRQ WRs support");
109
d821f02a
BS		 110static unsigned int ib_ipath_disable_sma;
111module_param_named(disable_sma, ib_ipath_disable_sma, uint, S_IWUSR | S_IRUGO);
112MODULE_PARM_DESC(ib_ipath_disable_sma, "Disable the SMA");
113
6522108f
BS		 114const int ib_ipath_state_ops[IB_QPS_ERR + 1] = {
115 [IB_QPS_RESET] = 0,
116 [IB_QPS_INIT] = IPATH_POST_RECV_OK,
117 [IB_QPS_RTR] = IPATH_POST_RECV_OK | IPATH_PROCESS_RECV_OK,
118 [IB_QPS_RTS] = IPATH_POST_RECV_OK | IPATH_PROCESS_RECV_OK |
119 IPATH_POST_SEND_OK | IPATH_PROCESS_SEND_OK,
120 [IB_QPS_SQD] = IPATH_POST_RECV_OK | IPATH_PROCESS_RECV_OK |
121 IPATH_POST_SEND_OK,
122 [IB_QPS_SQE] = IPATH_POST_RECV_OK | IPATH_PROCESS_RECV_OK,
123 [IB_QPS_ERR] = 0,
124};
125
34b2aafe
BS		 126struct ipath_ucontext {
127 struct ib_ucontext ibucontext;
128};
129
130static inline struct ipath_ucontext *to_iucontext(struct ib_ucontext
131 *ibucontext)
132{
133 return container_of(ibucontext, struct ipath_ucontext, ibucontext);
134}
135
6522108f
BS		 136/*
137 * Translate ib_wr_opcode into ib_wc_opcode.
138 */
139const enum ib_wc_opcode ib_ipath_wc_opcode[] = {
140 [IB_WR_RDMA_WRITE] = IB_WC_RDMA_WRITE,
141 [IB_WR_RDMA_WRITE_WITH_IMM] = IB_WC_RDMA_WRITE,
142 [IB_WR_SEND] = IB_WC_SEND,
143 [IB_WR_SEND_WITH_IMM] = IB_WC_SEND,
144 [IB_WR_RDMA_READ] = IB_WC_RDMA_READ,
145 [IB_WR_ATOMIC_CMP_AND_SWP] = IB_WC_COMP_SWAP,
146 [IB_WR_ATOMIC_FETCH_AND_ADD] = IB_WC_FETCH_ADD
147};
148
149/*
150 * System image GUID.
151 */
ac2ae4c9 152static __be64 sys_image_guid;
6522108f
BS		 153
154/**
155 * ipath_copy_sge - copy data to SGE memory
156 * @ss: the SGE state
157 * @data: the data to copy
158 * @length: the length of the data
159 */
160void ipath_copy_sge(struct ipath_sge_state *ss, void *data, u32 length)
161{
162 struct ipath_sge *sge = &ss->sge;
163
164 while (length) {
165 u32 len = sge->length;
166
6522108f
BS		 167 if (len > length)
168 len = length;
30d149ab
RC		 169 if (len > sge->sge_length)
170 len = sge->sge_length;
171 BUG_ON(len == 0);
6522108f
BS		 172 memcpy(sge->vaddr, data, len);
173 sge->vaddr += len;
174 sge->length -= len;
175 sge->sge_length -= len;
176 if (sge->sge_length == 0) {
177 if (--ss->num_sge)
178 *sge = *ss->sg_list++;
179 } else if (sge->length == 0 && sge->mr != NULL) {
180 if (++sge->n >= IPATH_SEGSZ) {
181 if (++sge->m >= sge->mr->mapsz)
182 break;
183 sge->n = 0;
184 }
185 sge->vaddr =
186 sge->mr->map[sge->m]->segs[sge->n].vaddr;
187 sge->length =
188 sge->mr->map[sge->m]->segs[sge->n].length;
189 }
190 data += len;
191 length -= len;
192 }
193}
194
195/**
196 * ipath_skip_sge - skip over SGE memory - XXX almost dup of prev func
197 * @ss: the SGE state
198 * @length: the number of bytes to skip
199 */
200void ipath_skip_sge(struct ipath_sge_state *ss, u32 length)
201{
202 struct ipath_sge *sge = &ss->sge;
203
6522108f
BS		 204 while (length) {
205 u32 len = sge->length;
206
6522108f
BS		 207 if (len > length)
208 len = length;
30d149ab
RC		 209 if (len > sge->sge_length)
210 len = sge->sge_length;
211 BUG_ON(len == 0);
6522108f
BS		 212 sge->vaddr += len;
213 sge->length -= len;
214 sge->sge_length -= len;
215 if (sge->sge_length == 0) {
216 if (--ss->num_sge)
217 *sge = *ss->sg_list++;
218 } else if (sge->length == 0 && sge->mr != NULL) {
219 if (++sge->n >= IPATH_SEGSZ) {
220 if (++sge->m >= sge->mr->mapsz)
221 break;
222 sge->n = 0;
223 }
224 sge->vaddr =
225 sge->mr->map[sge->m]->segs[sge->n].vaddr;
226 sge->length =
227 sge->mr->map[sge->m]->segs[sge->n].length;
228 }
229 length -= len;
230 }
231}
232
c9cf7db2
RC		 233static void ipath_flush_wqe(struct ipath_qp *qp, struct ib_send_wr *wr)
234{
235 struct ib_wc wc;
236
237 memset(&wc, 0, sizeof(wc));
238 wc.wr_id = wr->wr_id;
239 wc.status = IB_WC_WR_FLUSH_ERR;
240 wc.opcode = ib_ipath_wc_opcode[wr->opcode];
241 wc.qp = &qp->ibqp;
242 ipath_cq_enter(to_icq(qp->ibqp.send_cq), &wc, 1);
243}
244
4ee97180
RC		 245/**
246 * ipath_post_one_send - post one RC, UC, or UD send work request
247 * @qp: the QP to post on
248 * @wr: the work request to send
249 */
250static int ipath_post_one_send(struct ipath_qp *qp, struct ib_send_wr *wr)
251{
252 struct ipath_swqe *wqe;
253 u32 next;
254 int i;
255 int j;
256 int acc;
257 int ret;
258 unsigned long flags;
259
260 spin_lock_irqsave(&qp->s_lock, flags);
261
262 /* Check that state is OK to post send. */
c9cf7db2
RC		 263 if (unlikely(!(ib_ipath_state_ops[qp->state] & IPATH_POST_SEND_OK))) {
264 if (qp->state != IB_QPS_SQE && qp->state != IB_QPS_ERR)
265 goto bail_inval;
266 /* C10-96 says generate a flushed completion entry. */
267 ipath_flush_wqe(qp, wr);
268 ret = 0;
269 goto bail;
270 }
4ee97180
RC		 271
272 /* IB spec says that num_sge == 0 is OK. */
273 if (wr->num_sge > qp->s_max_sge)
274 goto bail_inval;
275
276 /*
277 * Don't allow RDMA reads or atomic operations on UC or
278 * undefined operations.
279 * Make sure buffer is large enough to hold the result for atomics.
280 */
281 if (qp->ibqp.qp_type == IB_QPT_UC) {
282 if ((unsigned) wr->opcode >= IB_WR_RDMA_READ)
283 goto bail_inval;
284 } else if (qp->ibqp.qp_type == IB_QPT_UD) {
285 /* Check UD opcode */
286 if (wr->opcode != IB_WR_SEND &&
287 wr->opcode != IB_WR_SEND_WITH_IMM)
288 goto bail_inval;
289 /* Check UD destination address PD */
290 if (qp->ibqp.pd != wr->wr.ud.ah->pd)
291 goto bail_inval;
292 } else if ((unsigned) wr->opcode > IB_WR_ATOMIC_FETCH_AND_ADD)
293 goto bail_inval;
294 else if (wr->opcode >= IB_WR_ATOMIC_CMP_AND_SWP &&
295 (wr->num_sge == 0 ||
296 wr->sg_list[0].length < sizeof(u64) ||
297 wr->sg_list[0].addr & (sizeof(u64) - 1)))
298 goto bail_inval;
299 else if (wr->opcode >= IB_WR_RDMA_READ && !qp->s_max_rd_atomic)
300 goto bail_inval;
301
302 next = qp->s_head + 1;
303 if (next >= qp->s_size)
304 next = 0;
4187b915
RC		 305 if (next == qp->s_last) {
306 ret = -ENOMEM;
307 goto bail;
308 }
4ee97180
RC		 309
310 wqe = get_swqe_ptr(qp, qp->s_head);
311 wqe->wr = *wr;
312 wqe->ssn = qp->s_ssn++;
313 wqe->length = 0;
314 if (wr->num_sge) {
315 acc = wr->opcode >= IB_WR_RDMA_READ ?
316 IB_ACCESS_LOCAL_WRITE : 0;
317 for (i = 0, j = 0; i < wr->num_sge; i++) {
318 u32 length = wr->sg_list[i].length;
319 int ok;
320
321 if (length == 0)
322 continue;
323 ok = ipath_lkey_ok(qp, &wqe->sg_list[j],
324 &wr->sg_list[i], acc);
325 if (!ok)
326 goto bail_inval;
327 wqe->length += length;
328 j++;
329 }
330 wqe->wr.num_sge = j;
331 }
332 if (qp->ibqp.qp_type == IB_QPT_UC ||
333 qp->ibqp.qp_type == IB_QPT_RC) {
334 if (wqe->length > 0x80000000U)
335 goto bail_inval;
336 } else if (wqe->length > to_idev(qp->ibqp.device)->dd->ipath_ibmtu)
337 goto bail_inval;
338 qp->s_head = next;
339
340 ret = 0;
341 goto bail;
342
343bail_inval:
344 ret = -EINVAL;
345bail:
346 spin_unlock_irqrestore(&qp->s_lock, flags);
347 return ret;
348}
349
6522108f
BS		 350/**
351 * ipath_post_send - post a send on a QP
352 * @ibqp: the QP to post the send on
353 * @wr: the list of work requests to post
354 * @bad_wr: the first bad WR is put here
355 *
356 * This may be called from interrupt context.
357 */
358static int ipath_post_send(struct ib_qp *ibqp, struct ib_send_wr *wr,
359 struct ib_send_wr **bad_wr)
360{
361 struct ipath_qp *qp = to_iqp(ibqp);
362 int err = 0;
363
6522108f 364 for (; wr; wr = wr->next) {
4ee97180 365 err = ipath_post_one_send(qp, wr);
6522108f
BS		 366 if (err) {
367 *bad_wr = wr;
4ee97180 368 goto bail;
6522108f
BS		 369 }
370 }
371
4ee97180
RC		 372 /* Try to do the send work in the caller's context. */
373 ipath_do_send((unsigned long) qp);
374
6522108f
BS		 375bail:
376 return err;
377}
378
379/**
380 * ipath_post_receive - post a receive on a QP
381 * @ibqp: the QP to post the receive on
382 * @wr: the WR to post
383 * @bad_wr: the first bad WR is put here
384 *
385 * This may be called from interrupt context.
386 */
387static int ipath_post_receive(struct ib_qp *ibqp, struct ib_recv_wr *wr,
388 struct ib_recv_wr **bad_wr)
389{
390 struct ipath_qp *qp = to_iqp(ibqp);
373d9915 391 struct ipath_rwq *wq = qp->r_rq.wq;
6522108f
BS		 392 unsigned long flags;
393 int ret;
394
395 /* Check that state is OK to post receive. */
373d9915 396 if (!(ib_ipath_state_ops[qp->state] & IPATH_POST_RECV_OK) || !wq) {
6522108f
BS		 397 *bad_wr = wr;
398 ret = -EINVAL;
399 goto bail;
400 }
401
402 for (; wr; wr = wr->next) {
403 struct ipath_rwqe *wqe;
404 u32 next;
373d9915 405 int i;
6522108f 406
373d9915 407 if ((unsigned) wr->num_sge > qp->r_rq.max_sge) {
6522108f 408 *bad_wr = wr;
4187b915 409 ret = -EINVAL;
6522108f
BS		 410 goto bail;
411 }
412
413 spin_lock_irqsave(&qp->r_rq.lock, flags);
373d9915 414 next = wq->head + 1;
6522108f
BS		 415 if (next >= qp->r_rq.size)
416 next = 0;
373d9915 417 if (next == wq->tail) {
6522108f
BS		 418 spin_unlock_irqrestore(&qp->r_rq.lock, flags);
419 *bad_wr = wr;
420 ret = -ENOMEM;
421 goto bail;
422 }
423
373d9915 424 wqe = get_rwqe_ptr(&qp->r_rq, wq->head);
6522108f 425 wqe->wr_id = wr->wr_id;
373d9915
RC		 426 wqe->num_sge = wr->num_sge;
427 for (i = 0; i < wr->num_sge; i++)
428 wqe->sg_list[i] = wr->sg_list[i];
4fc570bc
RC		 429 /* Make sure queue entry is written before the head index. */
430 smp_wmb();
373d9915 431 wq->head = next;
6522108f
BS		 432 spin_unlock_irqrestore(&qp->r_rq.lock, flags);
433 }
434 ret = 0;
435
436bail:
437 return ret;
438}
439
440/**
441 * ipath_qp_rcv - processing an incoming packet on a QP
442 * @dev: the device the packet came on
443 * @hdr: the packet header
444 * @has_grh: true if the packet has a GRH
445 * @data: the packet data
446 * @tlen: the packet length
447 * @qp: the QP the packet came on
448 *
449 * This is called from ipath_ib_rcv() to process an incoming packet
450 * for the given QP.
451 * Called at interrupt level.
452 */
453static void ipath_qp_rcv(struct ipath_ibdev *dev,
454 struct ipath_ib_header *hdr, int has_grh,
455 void *data, u32 tlen, struct ipath_qp *qp)
456{
457 /* Check for valid receive state. */
458 if (!(ib_ipath_state_ops[qp->state] & IPATH_PROCESS_RECV_OK)) {
459 dev->n_pkt_drops++;
460 return;
461 }
462
463 switch (qp->ibqp.qp_type) {
464 case IB_QPT_SMI:
465 case IB_QPT_GSI:
d821f02a
BS		 466 if (ib_ipath_disable_sma)
467 break;
468 /* FALLTHROUGH */
6522108f
BS		 469 case IB_QPT_UD:
470 ipath_ud_rcv(dev, hdr, has_grh, data, tlen, qp);
471 break;
472
473 case IB_QPT_RC:
474 ipath_rc_rcv(dev, hdr, has_grh, data, tlen, qp);
475 break;
476
477 case IB_QPT_UC:
478 ipath_uc_rcv(dev, hdr, has_grh, data, tlen, qp);
479 break;
480
481 default:
482 break;
483 }
484}
485
486/**
b1c1b6a3 487 * ipath_ib_rcv - process an incoming packet
6522108f
BS		 488 * @arg: the device pointer
489 * @rhdr: the header of the packet
490 * @data: the packet data
491 * @tlen: the packet length
492 *
493 * This is called from ipath_kreceive() to process an incoming packet at
494 * interrupt level. Tlen is the length of the header + data + CRC in bytes.
495 */
b1c1b6a3
BS		 496void ipath_ib_rcv(struct ipath_ibdev *dev, void *rhdr, void *data,
497 u32 tlen)
6522108f 498{
6522108f
BS		 499 struct ipath_ib_header *hdr = rhdr;
500 struct ipath_other_headers *ohdr;
501 struct ipath_qp *qp;
502 u32 qp_num;
503 int lnh;
504 u8 opcode;
505 u16 lid;
506
507 if (unlikely(dev == NULL))
508 goto bail;
509
510 if (unlikely(tlen < 24)) { /* LRH+BTH+CRC */
511 dev->rcv_errors++;
512 goto bail;
513 }
514
515 /* Check for a valid destination LID (see ch. 7.11.1). */
516 lid = be16_to_cpu(hdr->lrh[1]);
27b678dd 517 if (lid < IPATH_MULTICAST_LID_BASE) {
542869a1 518 lid &= ~((1 << dev->dd->ipath_lmc) - 1);
34b2aafe 519 if (unlikely(lid != dev->dd->ipath_lid)) {
6522108f
BS		 520 dev->rcv_errors++;
521 goto bail;
522 }
523 }
524
525 /* Check for GRH */
526 lnh = be16_to_cpu(hdr->lrh[0]) & 3;
27b678dd 527 if (lnh == IPATH_LRH_BTH)
6522108f 528 ohdr = &hdr->u.oth;
27b678dd 529 else if (lnh == IPATH_LRH_GRH)
6522108f
BS		 530 ohdr = &hdr->u.l.oth;
531 else {
532 dev->rcv_errors++;
533 goto bail;
534 }
535
536 opcode = be32_to_cpu(ohdr->bth[0]) >> 24;
537 dev->opstats[opcode].n_bytes += tlen;
538 dev->opstats[opcode].n_packets++;
539
540 /* Get the destination QP number. */
27b678dd
BS		 541 qp_num = be32_to_cpu(ohdr->bth[1]) & IPATH_QPN_MASK;
542 if (qp_num == IPATH_MULTICAST_QPN) {
6522108f
BS		 543 struct ipath_mcast *mcast;
544 struct ipath_mcast_qp *p;
545
5a7d4eea
BS		 546 if (lnh != IPATH_LRH_GRH) {
547 dev->n_pkt_drops++;
548 goto bail;
549 }
6522108f
BS		 550 mcast = ipath_mcast_find(&hdr->u.l.grh.dgid);
551 if (mcast == NULL) {
552 dev->n_pkt_drops++;
553 goto bail;
554 }
555 dev->n_multicast_rcv++;
556 list_for_each_entry_rcu(p, &mcast->qp_list, list)
5a7d4eea 557 ipath_qp_rcv(dev, hdr, 1, data, tlen, p->qp);
6522108f
BS		 558 /*
559 * Notify ipath_multicast_detach() if it is waiting for us
560 * to finish.
561 */
562 if (atomic_dec_return(&mcast->refcount) <= 1)
563 wake_up(&mcast->wait);
564 } else {
565 qp = ipath_lookup_qpn(&dev->qp_table, qp_num);
566 if (qp) {
567 dev->n_unicast_rcv++;
27b678dd 568 ipath_qp_rcv(dev, hdr, lnh == IPATH_LRH_GRH, data,
6522108f
BS		 569 tlen, qp);
570 /*
571 * Notify ipath_destroy_qp() if it is waiting
572 * for us to finish.
573 */
574 if (atomic_dec_and_test(&qp->refcount))
575 wake_up(&qp->wait);
576 } else
577 dev->n_pkt_drops++;
578 }
579
580bail:;
581}
582
583/**
584 * ipath_ib_timer - verbs timer
585 * @arg: the device pointer
586 *
587 * This is called from ipath_do_rcv_timer() at interrupt level to check for
588 * QPs which need retransmits and to collect performance numbers.
589 */
da9aec7b 590static void ipath_ib_timer(struct ipath_ibdev *dev)
6522108f 591{
6522108f 592 struct ipath_qp *resend = NULL;
6522108f
BS		 593 struct list_head *last;
594 struct ipath_qp *qp;
595 unsigned long flags;
596
597 if (dev == NULL)
598 return;
599
600 spin_lock_irqsave(&dev->pending_lock, flags);
601 /* Start filling the next pending queue. */
602 if (++dev->pending_index >= ARRAY_SIZE(dev->pending))
603 dev->pending_index = 0;
604 /* Save any requests still in the new queue, they have timed out. */
605 last = &dev->pending[dev->pending_index];
606 while (!list_empty(last)) {
607 qp = list_entry(last->next, struct ipath_qp, timerwait);
94b8d9f9 608 list_del_init(&qp->timerwait);
9b2017f1
BS		 609 qp->timer_next = resend;
610 resend = qp;
611 atomic_inc(&qp->refcount);
6522108f
BS		 612 }
613 last = &dev->rnrwait;
614 if (!list_empty(last)) {
615 qp = list_entry(last->next, struct ipath_qp, timerwait);
616 if (--qp->s_rnr_timeout == 0) {
617 do {
94b8d9f9 618 list_del_init(&qp->timerwait);
9b2017f1 619 tasklet_hi_schedule(&qp->s_task);
6522108f
BS		 620 if (list_empty(last))
621 break;
622 qp = list_entry(last->next, struct ipath_qp,
623 timerwait);
624 } while (qp->s_rnr_timeout == 0);
625 }
626 }
627 /*
628 * We should only be in the started state if pma_sample_start != 0
629 */
630 if (dev->pma_sample_status == IB_PMA_SAMPLE_STATUS_STARTED &&
631 --dev->pma_sample_start == 0) {
632 dev->pma_sample_status = IB_PMA_SAMPLE_STATUS_RUNNING;
34b2aafe
BS		 633 ipath_snapshot_counters(dev->dd, &dev->ipath_sword,
634 &dev->ipath_rword,
635 &dev->ipath_spkts,
636 &dev->ipath_rpkts,
637 &dev->ipath_xmit_wait);
6522108f
BS		 638 }
639 if (dev->pma_sample_status == IB_PMA_SAMPLE_STATUS_RUNNING) {
640 if (dev->pma_sample_interval == 0) {
641 u64 ta, tb, tc, td, te;
642
643 dev->pma_sample_status = IB_PMA_SAMPLE_STATUS_DONE;
34b2aafe
BS		 644 ipath_snapshot_counters(dev->dd, &ta, &tb,
645 &tc, &td, &te);
6522108f
BS		 646
647 dev->ipath_sword = ta - dev->ipath_sword;
648 dev->ipath_rword = tb - dev->ipath_rword;
649 dev->ipath_spkts = tc - dev->ipath_spkts;
650 dev->ipath_rpkts = td - dev->ipath_rpkts;
651 dev->ipath_xmit_wait = te - dev->ipath_xmit_wait;
652 }
653 else
654 dev->pma_sample_interval--;
655 }
656 spin_unlock_irqrestore(&dev->pending_lock, flags);
657
658 /* XXX What if timer fires again while this is running? */
9b2017f1 659 for (qp = resend; qp != NULL; qp = qp->timer_next) {
6522108f
BS		 660 struct ib_wc wc;
661
662 spin_lock_irqsave(&qp->s_lock, flags);
663 if (qp->s_last != qp->s_tail && qp->state == IB_QPS_RTS) {
664 dev->n_timeouts++;
665 ipath_restart_rc(qp, qp->s_last_psn + 1, &wc);
666 }
667 spin_unlock_irqrestore(&qp->s_lock, flags);
668
669 /* Notify ipath_destroy_qp() if it is waiting. */
670 if (atomic_dec_and_test(&qp->refcount))
671 wake_up(&qp->wait);
672 }
6522108f
BS		 673}
674
34b2aafe
BS		 675static void update_sge(struct ipath_sge_state *ss, u32 length)
676{
677 struct ipath_sge *sge = &ss->sge;
678
679 sge->vaddr += length;
680 sge->length -= length;
681 sge->sge_length -= length;
682 if (sge->sge_length == 0) {
683 if (--ss->num_sge)
684 *sge = *ss->sg_list++;
685 } else if (sge->length == 0 && sge->mr != NULL) {
686 if (++sge->n >= IPATH_SEGSZ) {
687 if (++sge->m >= sge->mr->mapsz)
688 return;
689 sge->n = 0;
690 }
691 sge->vaddr = sge->mr->map[sge->m]->segs[sge->n].vaddr;
692 sge->length = sge->mr->map[sge->m]->segs[sge->n].length;
693 }
694}
695
696#ifdef __LITTLE_ENDIAN
697static inline u32 get_upper_bits(u32 data, u32 shift)
698{
699 return data >> shift;
700}
701
702static inline u32 set_upper_bits(u32 data, u32 shift)
703{
704 return data << shift;
705}
706
707static inline u32 clear_upper_bytes(u32 data, u32 n, u32 off)
708{
709 data <<= ((sizeof(u32) - n) * BITS_PER_BYTE);
710 data >>= ((sizeof(u32) - n - off) * BITS_PER_BYTE);
711 return data;
712}
713#else
714static inline u32 get_upper_bits(u32 data, u32 shift)
715{
716 return data << shift;
717}
718
719static inline u32 set_upper_bits(u32 data, u32 shift)
720{
721 return data >> shift;
722}
723
724static inline u32 clear_upper_bytes(u32 data, u32 n, u32 off)
725{
726 data >>= ((sizeof(u32) - n) * BITS_PER_BYTE);
727 data <<= ((sizeof(u32) - n - off) * BITS_PER_BYTE);
728 return data;
729}
730#endif
731
732static void copy_io(u32 __iomem *piobuf, struct ipath_sge_state *ss,
210d6ca3 733 u32 length, unsigned flush_wc)
34b2aafe
BS		 734{
735 u32 extra = 0;
736 u32 data = 0;
737 u32 last;
738
739 while (1) {
740 u32 len = ss->sge.length;
741 u32 off;
742
34b2aafe
BS		 743 if (len > length)
744 len = length;
745 if (len > ss->sge.sge_length)
746 len = ss->sge.sge_length;
4ee97180 747 BUG_ON(len == 0);
34b2aafe
BS		 748 /* If the source address is not aligned, try to align it. */
749 off = (unsigned long)ss->sge.vaddr & (sizeof(u32) - 1);
750 if (off) {
751 u32 *addr = (u32 *)((unsigned long)ss->sge.vaddr &
752 ~(sizeof(u32) - 1));
753 u32 v = get_upper_bits(*addr, off * BITS_PER_BYTE);
754 u32 y;
755
756 y = sizeof(u32) - off;
757 if (len > y)
758 len = y;
759 if (len + extra >= sizeof(u32)) {
760 data |= set_upper_bits(v, extra *
761 BITS_PER_BYTE);
762 len = sizeof(u32) - extra;
763 if (len == length) {
764 last = data;
765 break;
766 }
767 __raw_writel(data, piobuf);
768 piobuf++;
769 extra = 0;
770 data = 0;
771 } else {
772 /* Clear unused upper bytes */
773 data |= clear_upper_bytes(v, len, extra);
774 if (len == length) {
775 last = data;
776 break;
777 }
778 extra += len;
779 }
780 } else if (extra) {
781 /* Source address is aligned. */
782 u32 *addr = (u32 *) ss->sge.vaddr;
783 int shift = extra * BITS_PER_BYTE;
784 int ushift = 32 - shift;
785 u32 l = len;
786
787 while (l >= sizeof(u32)) {
788 u32 v = *addr;
789
790 data |= set_upper_bits(v, shift);
791 __raw_writel(data, piobuf);
792 data = get_upper_bits(v, ushift);
793 piobuf++;
794 addr++;
795 l -= sizeof(u32);
796 }
797 /*
798 * We still have 'extra' number of bytes leftover.
799 */
800 if (l) {
801 u32 v = *addr;
802
803 if (l + extra >= sizeof(u32)) {
804 data |= set_upper_bits(v, shift);
805 len -= l + extra - sizeof(u32);
806 if (len == length) {
807 last = data;
808 break;
809 }
810 __raw_writel(data, piobuf);
811 piobuf++;
812 extra = 0;
813 data = 0;
814 } else {
815 /* Clear unused upper bytes */
816 data |= clear_upper_bytes(v, l,
817 extra);
818 if (len == length) {
819 last = data;
820 break;
821 }
822 extra += l;
823 }
824 } else if (len == length) {
825 last = data;
826 break;
827 }
828 } else if (len == length) {
829 u32 w;
830
831 /*
832 * Need to round up for the last dword in the
833 * packet.
834 */
835 w = (len + 3) >> 2;
836 __iowrite32_copy(piobuf, ss->sge.vaddr, w - 1);
837 piobuf += w - 1;
838 last = ((u32 *) ss->sge.vaddr)[w - 1];
839 break;
840 } else {
841 u32 w = len >> 2;
842
843 __iowrite32_copy(piobuf, ss->sge.vaddr, w);
844 piobuf += w;
845
846 extra = len & (sizeof(u32) - 1);
847 if (extra) {
848 u32 v = ((u32 *) ss->sge.vaddr)[w];
849
850 /* Clear unused upper bytes */
851 data = clear_upper_bytes(v, extra, 0);
852 }
853 }
854 update_sge(ss, len);
855 length -= len;
856 }
857 /* Update address before sending packet. */
858 update_sge(ss, length);
210d6ca3
RC		 859 if (flush_wc) {
860 /* must flush early everything before trigger word */
861 ipath_flush_wc();
862 __raw_writel(last, piobuf);
863 /* be sure trigger word is written */
864 ipath_flush_wc();
865 } else
866 __raw_writel(last, piobuf);
34b2aafe
BS		 867}
868
4ee97180
RC		 869static int ipath_verbs_send_pio(struct ipath_qp *qp, u32 *hdr, u32 hdrwords,
870 struct ipath_sge_state *ss, u32 len,
871 u32 plen, u32 dwords)
34b2aafe 872{
4ee97180 873 struct ipath_devdata *dd = to_idev(qp->ibqp.device)->dd;
34b2aafe 874 u32 __iomem *piobuf;
210d6ca3 875 unsigned flush_wc;
34b2aafe
BS		 876 int ret;
877
c4b4d16e 878 piobuf = ipath_getpiobuf(dd, plen, NULL);
34b2aafe
BS		 879 if (unlikely(piobuf == NULL)) {
880 ret = -EBUSY;
881 goto bail;
882 }
883
884 /*
885 * Write len to control qword, no flags.
886 * We have to flush after the PBC for correctness on some cpus
887 * or WC buffer can be written out of order.
888 */
889 writeq(plen, piobuf);
34b2aafe 890 piobuf += 2;
210d6ca3
RC		 891
892 flush_wc = dd->ipath_flags & IPATH_PIO_FLUSH_WC;
34b2aafe
BS		 893 if (len == 0) {
894 /*
895 * If there is just the header portion, must flush before
896 * writing last word of header for correctness, and after
897 * the last header word (trigger word).
898 */
210d6ca3
RC		 899 if (flush_wc) {
900 ipath_flush_wc();
901 __iowrite32_copy(piobuf, hdr, hdrwords - 1);
902 ipath_flush_wc();
903 __raw_writel(hdr[hdrwords - 1], piobuf + hdrwords - 1);
904 ipath_flush_wc();
905 } else
906 __iowrite32_copy(piobuf, hdr, hdrwords);
907 goto done;
34b2aafe
BS		 908 }
909
210d6ca3
RC		 910 if (flush_wc)
911 ipath_flush_wc();
34b2aafe
BS		 912 __iowrite32_copy(piobuf, hdr, hdrwords);
913 piobuf += hdrwords;
914
915 /* The common case is aligned and contained in one segment. */
916 if (likely(ss->num_sge == 1 && len <= ss->sge.length &&
917 !((unsigned long)ss->sge.vaddr & (sizeof(u32) - 1)))) {
34b2aafe
BS		 918 u32 *addr = (u32 *) ss->sge.vaddr;
919
920 /* Update address before sending packet. */
921 update_sge(ss, len);
210d6ca3
RC		 922 if (flush_wc) {
923 __iowrite32_copy(piobuf, addr, dwords - 1);
924 /* must flush early everything before trigger word */
925 ipath_flush_wc();
926 __raw_writel(addr[dwords - 1], piobuf + dwords - 1);
927 /* be sure trigger word is written */
928 ipath_flush_wc();
929 } else
930 __iowrite32_copy(piobuf, addr, dwords);
931 goto done;
34b2aafe 932 }
210d6ca3
RC		 933 copy_io(piobuf, ss, len, flush_wc);
934done:
4ee97180
RC		 935 if (qp->s_wqe)
936 ipath_send_complete(qp, qp->s_wqe, IB_WC_SUCCESS);
34b2aafe 937 ret = 0;
34b2aafe
BS		 938bail:
939 return ret;
940}
941
4ee97180
RC		 942/**
943 * ipath_verbs_send - send a packet
944 * @qp: the QP to send on
945 * @hdr: the packet header
9e2ef36b 946 * @hdrwords: the number of 32-bit words in the header
4ee97180
RC		 947 * @ss: the SGE to send
948 * @len: the length of the packet in bytes
949 */
950int ipath_verbs_send(struct ipath_qp *qp, struct ipath_ib_header *hdr,
951 u32 hdrwords, struct ipath_sge_state *ss, u32 len)
952{
49739b3e 953 struct ipath_devdata *dd = to_idev(qp->ibqp.device)->dd;
4ee97180
RC		 954 u32 plen;
955 int ret;
956 u32 dwords = (len + 3) >> 2;
957
9e2ef36b
DO		 958 /*
959 * Calculate the send buffer trigger address.
960 * The +1 counts for the pbc control dword following the pbc length.
961 */
4ee97180
RC		 962 plen = hdrwords + dwords + 1;
963
49739b3e
RC		 964 /* Drop non-VL15 packets if we are not in the active state */
965 if (!(dd->ipath_flags & IPATH_LINKACTIVE) &&
966 qp->ibqp.qp_type != IB_QPT_SMI) {
967 if (qp->s_wqe)
968 ipath_send_complete(qp, qp->s_wqe, IB_WC_SUCCESS);
969 ret = 0;
970 } else
971 ret = ipath_verbs_send_pio(qp, (u32 *) hdr, hdrwords,
972 ss, len, plen, dwords);
4ee97180
RC		 973
974 return ret;
975}
976
34b2aafe
BS		 977int ipath_snapshot_counters(struct ipath_devdata *dd, u64 *swords,
978 u64 *rwords, u64 *spkts, u64 *rpkts,
979 u64 *xmit_wait)
980{
981 int ret;
982
983 if (!(dd->ipath_flags & IPATH_INITTED)) {
984 /* no hardware, freeze, etc. */
34b2aafe
BS		 985 ret = -EINVAL;
986 goto bail;
987 }
988 *swords = ipath_snap_cntr(dd, dd->ipath_cregs->cr_wordsendcnt);
989 *rwords = ipath_snap_cntr(dd, dd->ipath_cregs->cr_wordrcvcnt);
990 *spkts = ipath_snap_cntr(dd, dd->ipath_cregs->cr_pktsendcnt);
991 *rpkts = ipath_snap_cntr(dd, dd->ipath_cregs->cr_pktrcvcnt);
992 *xmit_wait = ipath_snap_cntr(dd, dd->ipath_cregs->cr_sendstallcnt);
993
994 ret = 0;
995
996bail:
997 return ret;
998}
999
1000/**
1001 * ipath_get_counters - get various chip counters
1002 * @dd: the infinipath device
1003 * @cntrs: counters are placed here
1004 *
1005 * Return the counters needed by recv_pma_get_portcounters().
1006 */
1007int ipath_get_counters(struct ipath_devdata *dd,
1008 struct ipath_verbs_counters *cntrs)
1009{
4ee97180 1010 struct ipath_cregs const *crp = dd->ipath_cregs;
34b2aafe
BS		 1011 int ret;
1012
1013 if (!(dd->ipath_flags & IPATH_INITTED)) {
1014 /* no hardware, freeze, etc. */
34b2aafe
BS		 1015 ret = -EINVAL;
1016 goto bail;
1017 }
1018 cntrs->symbol_error_counter =
4ee97180 1019 ipath_snap_cntr(dd, crp->cr_ibsymbolerrcnt);
34b2aafe 1020 cntrs->link_error_recovery_counter =
4ee97180 1021 ipath_snap_cntr(dd, crp->cr_iblinkerrrecovcnt);
34b2aafe
BS		 1022 /*
1023 * The link downed counter counts when the other side downs the
1024 * connection. We add in the number of times we downed the link
1025 * due to local link integrity errors to compensate.
1026 */
1027 cntrs->link_downed_counter =
4ee97180 1028 ipath_snap_cntr(dd, crp->cr_iblinkdowncnt);
34b2aafe 1029 cntrs->port_rcv_errors =
4ee97180
RC		 1030 ipath_snap_cntr(dd, crp->cr_rxdroppktcnt) +
1031 ipath_snap_cntr(dd, crp->cr_rcvovflcnt) +
1032 ipath_snap_cntr(dd, crp->cr_portovflcnt) +
1033 ipath_snap_cntr(dd, crp->cr_err_rlencnt) +
1034 ipath_snap_cntr(dd, crp->cr_invalidrlencnt) +
1035 ipath_snap_cntr(dd, crp->cr_errlinkcnt) +
1036 ipath_snap_cntr(dd, crp->cr_erricrccnt) +
1037 ipath_snap_cntr(dd, crp->cr_errvcrccnt) +
1038 ipath_snap_cntr(dd, crp->cr_errlpcrccnt) +
1039 ipath_snap_cntr(dd, crp->cr_badformatcnt) +
2c9446a1 1040 dd->ipath_rxfc_unsupvl_errs;
34b2aafe 1041 cntrs->port_rcv_remphys_errors =
4ee97180
RC		 1042 ipath_snap_cntr(dd, crp->cr_rcvebpcnt);
1043 cntrs->port_xmit_discards = ipath_snap_cntr(dd, crp->cr_unsupvlcnt);
1044 cntrs->port_xmit_data = ipath_snap_cntr(dd, crp->cr_wordsendcnt);
1045 cntrs->port_rcv_data = ipath_snap_cntr(dd, crp->cr_wordrcvcnt);
1046 cntrs->port_xmit_packets = ipath_snap_cntr(dd, crp->cr_pktsendcnt);
1047 cntrs->port_rcv_packets = ipath_snap_cntr(dd, crp->cr_pktrcvcnt);
2c9446a1
BS		 1048 cntrs->local_link_integrity_errors =
1049 (dd->ipath_flags & IPATH_GPIO_ERRINTRS) ?
1050 dd->ipath_lli_errs : dd->ipath_lli_errors;
1051 cntrs->excessive_buffer_overrun_errors = dd->ipath_overrun_thresh_errs;
34b2aafe
BS		 1052
1053 ret = 0;
1054
1055bail:
1056 return ret;
1057}
1058
6522108f
BS		 1059/**
1060 * ipath_ib_piobufavail - callback when a PIO buffer is available
1061 * @arg: the device pointer
1062 *
1063 * This is called from ipath_intr() at interrupt level when a PIO buffer is
1064 * available after ipath_verbs_send() returned an error that no buffers were
9b2017f1 1065 * available. Return 1 if we consumed all the PIO buffers and we still have
6522108f 1066 * QPs waiting for buffers (for now, just do a tasklet_hi_schedule and
9b2017f1 1067 * return zero).
6522108f 1068 */
b1c1b6a3 1069int ipath_ib_piobufavail(struct ipath_ibdev *dev)
6522108f 1070{
6522108f
BS		 1071 struct ipath_qp *qp;
1072 unsigned long flags;
1073
1074 if (dev == NULL)
1075 goto bail;
1076
1077 spin_lock_irqsave(&dev->pending_lock, flags);
1078 while (!list_empty(&dev->piowait)) {
1079 qp = list_entry(dev->piowait.next, struct ipath_qp,
1080 piowait);
94b8d9f9 1081 list_del_init(&qp->piowait);
db5518cd 1082 clear_bit(IPATH_S_BUSY, &qp->s_busy);
6522108f
BS		 1083 tasklet_hi_schedule(&qp->s_task);
1084 }
1085 spin_unlock_irqrestore(&dev->pending_lock, flags);
1086
1087bail:
9b2017f1 1088 return 0;
6522108f
BS		 1089}
1090
1091static int ipath_query_device(struct ib_device *ibdev,
1092 struct ib_device_attr *props)
1093{
1094 struct ipath_ibdev *dev = to_idev(ibdev);
6522108f
BS		 1095
1096 memset(props, 0, sizeof(*props));
1097
1098 props->device_cap_flags = IB_DEVICE_BAD_PKEY_CNTR |
1099 IB_DEVICE_BAD_QKEY_CNTR | IB_DEVICE_SHUTDOWN_PORT |
1100 IB_DEVICE_SYS_IMAGE_GUID;
c9f79bdc 1101 props->page_size_cap = PAGE_SIZE;
34b2aafe
BS		 1102 props->vendor_id = dev->dd->ipath_vendorid;
1103 props->vendor_part_id = dev->dd->ipath_deviceid;
1104 props->hw_ver = dev->dd->ipath_pcirev;
6522108f
BS		 1105
1106 props->sys_image_guid = dev->sys_image_guid;
1107
1108 props->max_mr_size = ~0ull;
0b81e4f7 1109 props->max_qp = ib_ipath_max_qps;
fe62546a
BS		 1110 props->max_qp_wr = ib_ipath_max_qp_wrs;
1111 props->max_sge = ib_ipath_max_sges;
1112 props->max_cq = ib_ipath_max_cqs;
1113 props->max_ah = ib_ipath_max_ahs;
1114 props->max_cqe = ib_ipath_max_cqes;
1115 props->max_mr = dev->lk_table.max;
fdc7215f
RW		 1116 props->max_fmr = dev->lk_table.max;
1117 props->max_map_per_fmr = 32767;
fe62546a 1118 props->max_pd = ib_ipath_max_pds;
3859e39d
RC		 1119 props->max_qp_rd_atom = IPATH_MAX_RDMA_ATOMIC;
1120 props->max_qp_init_rd_atom = 255;
6522108f 1121 /* props->max_res_rd_atom */
fe62546a
BS		 1122 props->max_srq = ib_ipath_max_srqs;
1123 props->max_srq_wr = ib_ipath_max_srq_wrs;
1124 props->max_srq_sge = ib_ipath_max_srq_sges;
6522108f 1125 /* props->local_ca_ack_delay */
3859e39d 1126 props->atomic_cap = IB_ATOMIC_GLOB;
34b2aafe 1127 props->max_pkeys = ipath_get_npkeys(dev->dd);
fe62546a
BS		 1128 props->max_mcast_grp = ib_ipath_max_mcast_grps;
1129 props->max_mcast_qp_attach = ib_ipath_max_mcast_qp_attached;
6522108f
BS		 1130 props->max_total_mcast_qp_attach = props->max_mcast_qp_attach *
1131 props->max_mcast_grp;
1132
1133 return 0;
1134}
1135
3d68ea32
RC		 1136const u8 ipath_cvt_physportstate[32] = {
1137 [INFINIPATH_IBCS_LT_STATE_DISABLED] = IB_PHYSPORTSTATE_DISABLED,
1138 [INFINIPATH_IBCS_LT_STATE_LINKUP] = IB_PHYSPORTSTATE_LINKUP,
1139 [INFINIPATH_IBCS_LT_STATE_POLLACTIVE] = IB_PHYSPORTSTATE_POLL,
1140 [INFINIPATH_IBCS_LT_STATE_POLLQUIET] = IB_PHYSPORTSTATE_POLL,
1141 [INFINIPATH_IBCS_LT_STATE_SLEEPDELAY] = IB_PHYSPORTSTATE_SLEEP,
1142 [INFINIPATH_IBCS_LT_STATE_SLEEPQUIET] = IB_PHYSPORTSTATE_SLEEP,
1143 [INFINIPATH_IBCS_LT_STATE_CFGDEBOUNCE] =
1144 IB_PHYSPORTSTATE_CFG_TRAIN,
1145 [INFINIPATH_IBCS_LT_STATE_CFGRCVFCFG] =
1146 IB_PHYSPORTSTATE_CFG_TRAIN,
1147 [INFINIPATH_IBCS_LT_STATE_CFGWAITRMT] =
1148 IB_PHYSPORTSTATE_CFG_TRAIN,
1149 [INFINIPATH_IBCS_LT_STATE_CFGIDLE] = IB_PHYSPORTSTATE_CFG_TRAIN,
1150 [INFINIPATH_IBCS_LT_STATE_RECOVERRETRAIN] =
1151 IB_PHYSPORTSTATE_LINK_ERR_RECOVER,
1152 [INFINIPATH_IBCS_LT_STATE_RECOVERWAITRMT] =
1153 IB_PHYSPORTSTATE_LINK_ERR_RECOVER,
1154 [INFINIPATH_IBCS_LT_STATE_RECOVERIDLE] =
1155 IB_PHYSPORTSTATE_LINK_ERR_RECOVER,
1156 [0x10] = IB_PHYSPORTSTATE_CFG_TRAIN,
1157 [0x11] = IB_PHYSPORTSTATE_CFG_TRAIN,
1158 [0x12] = IB_PHYSPORTSTATE_CFG_TRAIN,
1159 [0x13] = IB_PHYSPORTSTATE_CFG_TRAIN,
1160 [0x14] = IB_PHYSPORTSTATE_CFG_TRAIN,
1161 [0x15] = IB_PHYSPORTSTATE_CFG_TRAIN,
1162 [0x16] = IB_PHYSPORTSTATE_CFG_TRAIN,
1163 [0x17] = IB_PHYSPORTSTATE_CFG_TRAIN
6522108f
BS		 1164};
1165
34b2aafe
BS		 1166u32 ipath_get_cr_errpkey(struct ipath_devdata *dd)
1167{
1168 return ipath_read_creg32(dd, dd->ipath_cregs->cr_errpkey);
1169}
1170
6522108f
BS		 1171static int ipath_query_port(struct ib_device *ibdev,
1172 u8 port, struct ib_port_attr *props)
1173{
1174 struct ipath_ibdev *dev = to_idev(ibdev);
4ee97180 1175 struct ipath_devdata *dd = dev->dd;
6522108f 1176 enum ib_mtu mtu;
4ee97180 1177 u16 lid = dd->ipath_lid;
6522108f
BS		 1178 u64 ibcstat;
1179
1180 memset(props, 0, sizeof(*props));
1181 props->lid = lid ? lid : __constant_be16_to_cpu(IB_LID_PERMISSIVE);
542869a1 1182 props->lmc = dd->ipath_lmc;
6522108f
BS		 1183 props->sm_lid = dev->sm_lid;
1184 props->sm_sl = dev->sm_sl;
4ee97180 1185 ibcstat = dd->ipath_lastibcstat;
a51a2513
RC		 1186 /* map LinkState to IB portinfo values. */
1187 props->state = ipath_ib_linkstate(dd, ibcstat) + 1;
1188
6522108f 1189 /* See phys_state_show() */
3d68ea32
RC		 1190 props->phys_state = /* MEA: assumes shift == 0 */
1191 ipath_cvt_physportstate[dd->ipath_lastibcstat &
1192 dd->ibcs_lts_mask];
6522108f
BS		 1193 props->port_cap_flags = dev->port_cap_flags;
1194 props->gid_tbl_len = 1;
c100f622 1195 props->max_msg_sz = 0x80000000;
4ee97180
RC		 1196 props->pkey_tbl_len = ipath_get_npkeys(dd);
1197 props->bad_pkey_cntr = ipath_get_cr_errpkey(dd) -
443a64ab 1198 dev->z_pkey_violations;
6522108f 1199 props->qkey_viol_cntr = dev->qkey_violations;
a51a2513 1200 props->active_width = dd->ipath_link_width_active;
6522108f 1201 /* See rate_show() */
a51a2513 1202 props->active_speed = dd->ipath_link_speed_active;
6522108f
BS		 1203 props->max_vl_num = 1; /* VLCap = VL0 */
1204 props->init_type_reply = 0;
1205
826d8010 1206 props->max_mtu = ipath_mtu4096 ? IB_MTU_4096 : IB_MTU_2048;
4ee97180 1207 switch (dd->ipath_ibmtu) {
6522108f
BS		 1208 case 4096:
1209 mtu = IB_MTU_4096;
1210 break;
1211 case 2048:
1212 mtu = IB_MTU_2048;
1213 break;
1214 case 1024:
1215 mtu = IB_MTU_1024;
1216 break;
1217 case 512:
1218 mtu = IB_MTU_512;
1219 break;
1220 case 256:
1221 mtu = IB_MTU_256;
1222 break;
1223 default:
1224 mtu = IB_MTU_2048;
1225 }
1226 props->active_mtu = mtu;
1227 props->subnet_timeout = dev->subnet_timeout;
1228
1229 return 0;
1230}
1231
1232static int ipath_modify_device(struct ib_device *device,
1233 int device_modify_mask,
1234 struct ib_device_modify *device_modify)
1235{
1236 int ret;
1237
1238 if (device_modify_mask & ~(IB_DEVICE_MODIFY_SYS_IMAGE_GUID |
1239 IB_DEVICE_MODIFY_NODE_DESC)) {
1240 ret = -EOPNOTSUPP;
1241 goto bail;
1242 }
1243
1244 if (device_modify_mask & IB_DEVICE_MODIFY_NODE_DESC)
1245 memcpy(device->node_desc, device_modify->node_desc, 64);
1246
1247 if (device_modify_mask & IB_DEVICE_MODIFY_SYS_IMAGE_GUID)
1248 to_idev(device)->sys_image_guid =
1249 cpu_to_be64(device_modify->sys_image_guid);
1250
1251 ret = 0;
1252
1253bail:
1254 return ret;
1255}
1256
1257static int ipath_modify_port(struct ib_device *ibdev,
1258 u8 port, int port_modify_mask,
1259 struct ib_port_modify *props)
1260{
1261 struct ipath_ibdev *dev = to_idev(ibdev);
1262
1263 dev->port_cap_flags |= props->set_port_cap_mask;
1264 dev->port_cap_flags &= ~props->clr_port_cap_mask;
1265 if (port_modify_mask & IB_PORT_SHUTDOWN)
34b2aafe 1266 ipath_set_linkstate(dev->dd, IPATH_IB_LINKDOWN);
6522108f
BS		 1267 if (port_modify_mask & IB_PORT_RESET_QKEY_CNTR)
1268 dev->qkey_violations = 0;
1269 return 0;
1270}
1271
1272static int ipath_query_gid(struct ib_device *ibdev, u8 port,
1273 int index, union ib_gid *gid)
1274{
1275 struct ipath_ibdev *dev = to_idev(ibdev);
1276 int ret;
1277
1278 if (index >= 1) {
1279 ret = -EINVAL;
1280 goto bail;
1281 }
1282 gid->global.subnet_prefix = dev->gid_prefix;
34b2aafe 1283 gid->global.interface_id = dev->dd->ipath_guid;
6522108f
BS		 1284
1285 ret = 0;
1286
1287bail:
1288 return ret;
1289}
1290
1291static struct ib_pd *ipath_alloc_pd(struct ib_device *ibdev,
1292 struct ib_ucontext *context,
1293 struct ib_udata *udata)
1294{
fe62546a 1295 struct ipath_ibdev *dev = to_idev(ibdev);
6522108f
BS		 1296 struct ipath_pd *pd;
1297 struct ib_pd *ret;
1298
fe62546a
BS		 1299 /*
1300 * This is actually totally arbitrary. Some correctness tests
1301 * assume there's a maximum number of PDs that can be allocated.
1302 * We don't actually have this limit, but we fail the test if
1303 * we allow allocations of more than we report for this value.
1304 */
1305
c27fef26
BS		 1306 pd = kmalloc(sizeof *pd, GFP_KERNEL);
1307 if (!pd) {
fe62546a
BS		 1308 ret = ERR_PTR(-ENOMEM);
1309 goto bail;
1310 }
1311
c27fef26
BS		 1312 spin_lock(&dev->n_pds_lock);
1313 if (dev->n_pds_allocated == ib_ipath_max_pds) {
1314 spin_unlock(&dev->n_pds_lock);
1315 kfree(pd);
6522108f
BS		 1316 ret = ERR_PTR(-ENOMEM);
1317 goto bail;
1318 }
1319
fe62546a 1320 dev->n_pds_allocated++;
c27fef26 1321 spin_unlock(&dev->n_pds_lock);
fe62546a 1322
6522108f
BS		 1323 /* ib_alloc_pd() will initialize pd->ibpd. */
1324 pd->user = udata != NULL;
1325
1326 ret = &pd->ibpd;
1327
1328bail:
1329 return ret;
1330}
1331
1332static int ipath_dealloc_pd(struct ib_pd *ibpd)
1333{
1334 struct ipath_pd *pd = to_ipd(ibpd);
fe62546a
BS		 1335 struct ipath_ibdev *dev = to_idev(ibpd->device);
1336
c27fef26 1337 spin_lock(&dev->n_pds_lock);
fe62546a 1338 dev->n_pds_allocated--;
c27fef26 1339 spin_unlock(&dev->n_pds_lock);
6522108f
BS		 1340
1341 kfree(pd);
1342
1343 return 0;
1344}
1345
1346/**
1347 * ipath_create_ah - create an address handle
1348 * @pd: the protection domain
1349 * @ah_attr: the attributes of the AH
1350 *
1351 * This may be called from interrupt context.
1352 */
1353static struct ib_ah *ipath_create_ah(struct ib_pd *pd,
1354 struct ib_ah_attr *ah_attr)
1355{
1356 struct ipath_ah *ah;
1357 struct ib_ah *ret;
fe62546a 1358 struct ipath_ibdev *dev = to_idev(pd->device);
3d27b004 1359 unsigned long flags;
fe62546a 1360
6522108f 1361 /* A multicast address requires a GRH (see ch. 8.4.1). */
27b678dd
BS		 1362 if (ah_attr->dlid >= IPATH_MULTICAST_LID_BASE &&
1363 ah_attr->dlid != IPATH_PERMISSIVE_LID &&
6522108f
BS		 1364 !(ah_attr->ah_flags & IB_AH_GRH)) {
1365 ret = ERR_PTR(-EINVAL);
1366 goto bail;
1367 }
1368
4a45b7d4
BS		 1369 if (ah_attr->dlid == 0) {
1370 ret = ERR_PTR(-EINVAL);
1371 goto bail;
1372 }
1373
fe62546a 1374 if (ah_attr->port_num < 1 ||
4a45b7d4
BS		 1375 ah_attr->port_num > pd->device->phys_port_cnt) {
1376 ret = ERR_PTR(-EINVAL);
1377 goto bail;
1378 }
1379
6522108f
BS		 1380 ah = kmalloc(sizeof *ah, GFP_ATOMIC);
1381 if (!ah) {
1382 ret = ERR_PTR(-ENOMEM);
1383 goto bail;
1384 }
1385
3d27b004 1386 spin_lock_irqsave(&dev->n_ahs_lock, flags);
c27fef26 1387 if (dev->n_ahs_allocated == ib_ipath_max_ahs) {
3d27b004 1388 spin_unlock_irqrestore(&dev->n_ahs_lock, flags);
c27fef26
BS		 1389 kfree(ah);
1390 ret = ERR_PTR(-ENOMEM);
1391 goto bail;
1392 }
1393
fe62546a 1394 dev->n_ahs_allocated++;
3d27b004 1395 spin_unlock_irqrestore(&dev->n_ahs_lock, flags);
fe62546a 1396
6522108f
BS		 1397 /* ib_create_ah() will initialize ah->ibah. */
1398 ah->attr = *ah_attr;
1399
1400 ret = &ah->ibah;
1401
1402bail:
1403 return ret;
1404}
1405
1406/**
1407 * ipath_destroy_ah - destroy an address handle
1408 * @ibah: the AH to destroy
1409 *
1410 * This may be called from interrupt context.
1411 */
1412static int ipath_destroy_ah(struct ib_ah *ibah)
1413{
fe62546a 1414 struct ipath_ibdev *dev = to_idev(ibah->device);
6522108f 1415 struct ipath_ah *ah = to_iah(ibah);
3d27b004 1416 unsigned long flags;
6522108f 1417
3d27b004 1418 spin_lock_irqsave(&dev->n_ahs_lock, flags);
fe62546a 1419 dev->n_ahs_allocated--;
3d27b004 1420 spin_unlock_irqrestore(&dev->n_ahs_lock, flags);
fe62546a 1421
6522108f
BS		 1422 kfree(ah);
1423
1424 return 0;
1425}
1426
1427static int ipath_query_ah(struct ib_ah *ibah, struct ib_ah_attr *ah_attr)
1428{
1429 struct ipath_ah *ah = to_iah(ibah);
1430
1431 *ah_attr = ah->attr;
1432
1433 return 0;
1434}
1435
34b2aafe
BS		 1436/**
1437 * ipath_get_npkeys - return the size of the PKEY table for port 0
1438 * @dd: the infinipath device
1439 */
1440unsigned ipath_get_npkeys(struct ipath_devdata *dd)
1441{
1442 return ARRAY_SIZE(dd->ipath_pd[0]->port_pkeys);
1443}
1444
1445/**
1446 * ipath_get_pkey - return the indexed PKEY from the port 0 PKEY table
1447 * @dd: the infinipath device
1448 * @index: the PKEY index
1449 */
1450unsigned ipath_get_pkey(struct ipath_devdata *dd, unsigned index)
1451{
1452 unsigned ret;
1453
1454 if (index >= ARRAY_SIZE(dd->ipath_pd[0]->port_pkeys))
1455 ret = 0;
1456 else
1457 ret = dd->ipath_pd[0]->port_pkeys[index];
1458
1459 return ret;
1460}
1461
6522108f
BS		 1462static int ipath_query_pkey(struct ib_device *ibdev, u8 port, u16 index,
1463 u16 *pkey)
1464{
1465 struct ipath_ibdev *dev = to_idev(ibdev);
1466 int ret;
1467
34b2aafe 1468 if (index >= ipath_get_npkeys(dev->dd)) {
6522108f
BS		 1469 ret = -EINVAL;
1470 goto bail;
1471 }
1472
34b2aafe 1473 *pkey = ipath_get_pkey(dev->dd, index);
6522108f
BS		 1474 ret = 0;
1475
1476bail:
1477 return ret;
1478}
1479
6522108f
BS		 1480/**
1481 * ipath_alloc_ucontext - allocate a ucontest
1482 * @ibdev: the infiniband device
1483 * @udata: not used by the InfiniPath driver
1484 */
1485
1486static struct ib_ucontext *ipath_alloc_ucontext(struct ib_device *ibdev,
1487 struct ib_udata *udata)
1488{
1489 struct ipath_ucontext *context;
1490 struct ib_ucontext *ret;
1491
1492 context = kmalloc(sizeof *context, GFP_KERNEL);
1493 if (!context) {
1494 ret = ERR_PTR(-ENOMEM);
1495 goto bail;
1496 }
1497
1498 ret = &context->ibucontext;
1499
1500bail:
1501 return ret;
1502}
1503
1504static int ipath_dealloc_ucontext(struct ib_ucontext *context)
1505{
1506 kfree(to_iucontext(context));
1507 return 0;
1508}
1509
1510static int ipath_verbs_register_sysfs(struct ib_device *dev);
1511
34b2aafe
BS		 1512static void __verbs_timer(unsigned long arg)
1513{
1514 struct ipath_devdata *dd = (struct ipath_devdata *) arg;
1515
34b2aafe
BS		 1516 /* Handle verbs layer timeouts. */
1517 ipath_ib_timer(dd->verbs_dev);
1518
1519 mod_timer(&dd->verbs_timer, jiffies + 1);
1520}
1521
1522static int enable_timer(struct ipath_devdata *dd)
1523{
1524 /*
1525 * Early chips had a design flaw where the chip and kernel idea
1526 * of the tail register don't always agree, and therefore we won't
1527 * get an interrupt on the next packet received.
1528 * If the board supports per packet receive interrupts, use it.
1529 * Otherwise, the timer function periodically checks for packets
1530 * to cover this case.
1531 * Either way, the timer is needed for verbs layer related
1532 * processing.
1533 */
1534 if (dd->ipath_flags & IPATH_GPIO_INTR) {
1535 ipath_write_kreg(dd, dd->ipath_kregs->kr_debugportselect,
1536 0x2074076542310ULL);
1537 /* Enable GPIO bit 2 interrupt */
8f140b40
AJ		 1538 dd->ipath_gpio_mask |= (u64) (1 << IPATH_GPIO_PORT0_BIT);
1539 ipath_write_kreg(dd, dd->ipath_kregs->kr_gpio_mask,
1540 dd->ipath_gpio_mask);
34b2aafe
BS		 1541 }
1542
1543 init_timer(&dd->verbs_timer);
1544 dd->verbs_timer.function = __verbs_timer;
1545 dd->verbs_timer.data = (unsigned long)dd;
1546 dd->verbs_timer.expires = jiffies + 1;
1547 add_timer(&dd->verbs_timer);
1548
1549 return 0;
1550}
1551
1552static int disable_timer(struct ipath_devdata *dd)
1553{
1554 /* Disable GPIO bit 2 interrupt */
2c9446a1 1555 if (dd->ipath_flags & IPATH_GPIO_INTR) {
2c9446a1 1556 /* Disable GPIO bit 2 interrupt */
8f140b40
AJ		 1557 dd->ipath_gpio_mask &= ~((u64) (1 << IPATH_GPIO_PORT0_BIT));
1558 ipath_write_kreg(dd, dd->ipath_kregs->kr_gpio_mask,
1559 dd->ipath_gpio_mask);
2c9446a1
BS		 1560 /*
1561 * We might want to undo changes to debugportselect,
1562 * but how?
1563 */
1564 }
34b2aafe
BS		 1565
1566 del_timer_sync(&dd->verbs_timer);
1567
1568 return 0;
1569}
1570
6522108f
BS		 1571/**
1572 * ipath_register_ib_device - register our device with the infiniband core
6522108f
BS		 1573 * @dd: the device data structure
1574 * Return the allocated ipath_ibdev pointer or NULL on error.
1575 */
b1c1b6a3 1576int ipath_register_ib_device(struct ipath_devdata *dd)
6522108f 1577{
34b2aafe 1578 struct ipath_verbs_counters cntrs;
6522108f
BS		 1579 struct ipath_ibdev *idev;
1580 struct ib_device *dev;
1581 int ret;
1582
1583 idev = (struct ipath_ibdev *)ib_alloc_device(sizeof *idev);
b1c1b6a3
BS		 1584 if (idev == NULL) {
1585 ret = -ENOMEM;
6522108f 1586 goto bail;
b1c1b6a3 1587 }
6522108f
BS		 1588
1589 dev = &idev->ibdev;
1590
1591 /* Only need to initialize non-zero fields. */
c27fef26
BS		 1592 spin_lock_init(&idev->n_pds_lock);
1593 spin_lock_init(&idev->n_ahs_lock);
1594 spin_lock_init(&idev->n_cqs_lock);
0b81e4f7 1595 spin_lock_init(&idev->n_qps_lock);
c27fef26
BS		 1596 spin_lock_init(&idev->n_srqs_lock);
1597 spin_lock_init(&idev->n_mcast_grps_lock);
1598
6522108f
BS		 1599 spin_lock_init(&idev->qp_table.lock);
1600 spin_lock_init(&idev->lk_table.lock);
1601 idev->sm_lid = __constant_be16_to_cpu(IB_LID_PERMISSIVE);
1602 /* Set the prefix to the default value (see ch. 4.1.1) */
1603 idev->gid_prefix = __constant_cpu_to_be64(0xfe80000000000000ULL);
1604
1605 ret = ipath_init_qp_table(idev, ib_ipath_qp_table_size);
1606 if (ret)
1607 goto err_qp;
1608
1609 /*
1610 * The top ib_ipath_lkey_table_size bits are used to index the
1611 * table. The lower 8 bits can be owned by the user (copied from
1612 * the LKEY). The remaining bits act as a generation number or tag.
1613 */
1614 idev->lk_table.max = 1 << ib_ipath_lkey_table_size;
1615 idev->lk_table.table = kzalloc(idev->lk_table.max *
1616 sizeof(*idev->lk_table.table),
1617 GFP_KERNEL);
1618 if (idev->lk_table.table == NULL) {
1619 ret = -ENOMEM;
1620 goto err_lk;
1621 }
6b66b2da 1622 INIT_LIST_HEAD(&idev->pending_mmaps);
6522108f 1623 spin_lock_init(&idev->pending_lock);
6b66b2da
RW		 1624 idev->mmap_offset = PAGE_SIZE;
1625 spin_lock_init(&idev->mmap_offset_lock);
6522108f
BS		 1626 INIT_LIST_HEAD(&idev->pending[0]);
1627 INIT_LIST_HEAD(&idev->pending[1]);
1628 INIT_LIST_HEAD(&idev->pending[2]);
1629 INIT_LIST_HEAD(&idev->piowait);
1630 INIT_LIST_HEAD(&idev->rnrwait);
1631 idev->pending_index = 0;
1632 idev->port_cap_flags =
1633 IB_PORT_SYS_IMAGE_GUID_SUP | IB_PORT_CLIENT_REG_SUP;
a51a2513
RC		 1634 if (dd->ipath_flags & IPATH_HAS_LINK_LATENCY)
1635 idev->port_cap_flags |= IB_PORT_LINK_LATENCY_SUP;
6522108f
BS		 1636 idev->pma_counter_select[0] = IB_PMA_PORT_XMIT_DATA;
1637 idev->pma_counter_select[1] = IB_PMA_PORT_RCV_DATA;
1638 idev->pma_counter_select[2] = IB_PMA_PORT_XMIT_PKTS;
1639 idev->pma_counter_select[3] = IB_PMA_PORT_RCV_PKTS;
53533e16 1640 idev->pma_counter_select[4] = IB_PMA_PORT_XMIT_WAIT;
6522108f 1641
fba75200 1642 /* Snapshot current HW counters to "clear" them. */
34b2aafe 1643 ipath_get_counters(dd, &cntrs);
fba75200
BS		 1644 idev->z_symbol_error_counter = cntrs.symbol_error_counter;
1645 idev->z_link_error_recovery_counter =
1646 cntrs.link_error_recovery_counter;
1647 idev->z_link_downed_counter = cntrs.link_downed_counter;
1648 idev->z_port_rcv_errors = cntrs.port_rcv_errors;
1649 idev->z_port_rcv_remphys_errors =
1650 cntrs.port_rcv_remphys_errors;
1651 idev->z_port_xmit_discards = cntrs.port_xmit_discards;
1652 idev->z_port_xmit_data = cntrs.port_xmit_data;
1653 idev->z_port_rcv_data = cntrs.port_rcv_data;
1654 idev->z_port_xmit_packets = cntrs.port_xmit_packets;
1655 idev->z_port_rcv_packets = cntrs.port_rcv_packets;
1656 idev->z_local_link_integrity_errors =
1657 cntrs.local_link_integrity_errors;
1658 idev->z_excessive_buffer_overrun_errors =
1659 cntrs.excessive_buffer_overrun_errors;
6c719cae 1660 idev->z_vl15_dropped = cntrs.vl15_dropped;
fba75200 1661
6522108f
BS		 1662 /*
1663 * The system image GUID is supposed to be the same for all
1664 * IB HCAs in a single system but since there can be other
1665 * device types in the system, we can't be sure this is unique.
1666 */
1667 if (!sys_image_guid)
34b2aafe 1668 sys_image_guid = dd->ipath_guid;
6522108f 1669 idev->sys_image_guid = sys_image_guid;
b1c1b6a3 1670 idev->ib_unit = dd->ipath_unit;
6522108f
BS		 1671 idev->dd = dd;
1672
1673 strlcpy(dev->name, "ipath%d", IB_DEVICE_NAME_MAX);
41c75a19 1674 dev->owner = THIS_MODULE;
34b2aafe 1675 dev->node_guid = dd->ipath_guid;
6522108f
BS		 1676 dev->uverbs_abi_ver = IPATH_UVERBS_ABI_VERSION;
1677 dev->uverbs_cmd_mask =
1678 (1ull << IB_USER_VERBS_CMD_GET_CONTEXT) |
1679 (1ull << IB_USER_VERBS_CMD_QUERY_DEVICE) |
1680 (1ull << IB_USER_VERBS_CMD_QUERY_PORT) |
1681 (1ull << IB_USER_VERBS_CMD_ALLOC_PD) |
1682 (1ull << IB_USER_VERBS_CMD_DEALLOC_PD) |
1683 (1ull << IB_USER_VERBS_CMD_CREATE_AH) |
1684 (1ull << IB_USER_VERBS_CMD_DESTROY_AH) |
1685 (1ull << IB_USER_VERBS_CMD_QUERY_AH) |
1686 (1ull << IB_USER_VERBS_CMD_REG_MR) |
1687 (1ull << IB_USER_VERBS_CMD_DEREG_MR) |
1688 (1ull << IB_USER_VERBS_CMD_CREATE_COMP_CHANNEL) |
1689 (1ull << IB_USER_VERBS_CMD_CREATE_CQ) |
1690 (1ull << IB_USER_VERBS_CMD_RESIZE_CQ) |
1691 (1ull << IB_USER_VERBS_CMD_DESTROY_CQ) |
1692 (1ull << IB_USER_VERBS_CMD_POLL_CQ) |
1693 (1ull << IB_USER_VERBS_CMD_REQ_NOTIFY_CQ) |
1694 (1ull << IB_USER_VERBS_CMD_CREATE_QP) |
1695 (1ull << IB_USER_VERBS_CMD_QUERY_QP) |
1696 (1ull << IB_USER_VERBS_CMD_MODIFY_QP) |
1697 (1ull << IB_USER_VERBS_CMD_DESTROY_QP) |
1698 (1ull << IB_USER_VERBS_CMD_POST_SEND) |
1699 (1ull << IB_USER_VERBS_CMD_POST_RECV) |
1700 (1ull << IB_USER_VERBS_CMD_ATTACH_MCAST) |
1701 (1ull << IB_USER_VERBS_CMD_DETACH_MCAST) |
1702 (1ull << IB_USER_VERBS_CMD_CREATE_SRQ) |
1703 (1ull << IB_USER_VERBS_CMD_MODIFY_SRQ) |
1704 (1ull << IB_USER_VERBS_CMD_QUERY_SRQ) |
1705 (1ull << IB_USER_VERBS_CMD_DESTROY_SRQ) |
1706 (1ull << IB_USER_VERBS_CMD_POST_SRQ_RECV);
07ebafba 1707 dev->node_type = RDMA_NODE_IB_CA;
6522108f 1708 dev->phys_port_cnt = 1;
f4fd0b22 1709 dev->num_comp_vectors = 1;
34b2aafe 1710 dev->dma_device = &dd->pcidev->dev;
6522108f
BS		 1711 dev->query_device = ipath_query_device;
1712 dev->modify_device = ipath_modify_device;
1713 dev->query_port = ipath_query_port;
1714 dev->modify_port = ipath_modify_port;
1715 dev->query_pkey = ipath_query_pkey;
1716 dev->query_gid = ipath_query_gid;
1717 dev->alloc_ucontext = ipath_alloc_ucontext;
1718 dev->dealloc_ucontext = ipath_dealloc_ucontext;
1719 dev->alloc_pd = ipath_alloc_pd;
1720 dev->dealloc_pd = ipath_dealloc_pd;
1721 dev->create_ah = ipath_create_ah;
1722 dev->destroy_ah = ipath_destroy_ah;
1723 dev->query_ah = ipath_query_ah;
1724 dev->create_srq = ipath_create_srq;
1725 dev->modify_srq = ipath_modify_srq;
1726 dev->query_srq = ipath_query_srq;
1727 dev->destroy_srq = ipath_destroy_srq;
1728 dev->create_qp = ipath_create_qp;
1729 dev->modify_qp = ipath_modify_qp;
1730 dev->query_qp = ipath_query_qp;
1731 dev->destroy_qp = ipath_destroy_qp;
1732 dev->post_send = ipath_post_send;
1733 dev->post_recv = ipath_post_receive;
1734 dev->post_srq_recv = ipath_post_srq_receive;
1735 dev->create_cq = ipath_create_cq;
1736 dev->destroy_cq = ipath_destroy_cq;
1737 dev->resize_cq = ipath_resize_cq;
1738 dev->poll_cq = ipath_poll_cq;
1739 dev->req_notify_cq = ipath_req_notify_cq;
1740 dev->get_dma_mr = ipath_get_dma_mr;
1741 dev->reg_phys_mr = ipath_reg_phys_mr;
1742 dev->reg_user_mr = ipath_reg_user_mr;
1743 dev->dereg_mr = ipath_dereg_mr;
1744 dev->alloc_fmr = ipath_alloc_fmr;
1745 dev->map_phys_fmr = ipath_map_phys_fmr;
1746 dev->unmap_fmr = ipath_unmap_fmr;
1747 dev->dealloc_fmr = ipath_dealloc_fmr;
1748 dev->attach_mcast = ipath_multicast_attach;
1749 dev->detach_mcast = ipath_multicast_detach;
1750 dev->process_mad = ipath_process_mad;
373d9915 1751 dev->mmap = ipath_mmap;
f2cbb660 1752 dev->dma_ops = &ipath_dma_mapping_ops;
6522108f
BS		 1753
1754 snprintf(dev->node_desc, sizeof(dev->node_desc),
96b644bd 1755 IPATH_IDSTR " %s", init_utsname()->nodename);
6522108f
BS		 1756
1757 ret = ib_register_device(dev);
1758 if (ret)
1759 goto err_reg;
1760
1761 if (ipath_verbs_register_sysfs(dev))
1762 goto err_class;
1763
34b2aafe 1764 enable_timer(dd);
6522108f
BS		 1765
1766 goto bail;
1767
1768err_class:
1769 ib_unregister_device(dev);
1770err_reg:
1771 kfree(idev->lk_table.table);
1772err_lk:
1773 kfree(idev->qp_table.table);
1774err_qp:
1775 ib_dealloc_device(dev);
b55f4f06 1776 ipath_dev_err(dd, "cannot register verbs: %d!\n", -ret);
6522108f
BS		 1777 idev = NULL;
1778
1779bail:
b1c1b6a3
BS		 1780 dd->verbs_dev = idev;
1781 return ret;
6522108f
BS		 1782}
1783
b1c1b6a3 1784void ipath_unregister_ib_device(struct ipath_ibdev *dev)
6522108f 1785{
6522108f
BS		 1786 struct ib_device *ibdev = &dev->ibdev;
1787
34b2aafe 1788 disable_timer(dev->dd);
6522108f
BS		 1789
1790 ib_unregister_device(ibdev);
1791
1792 if (!list_empty(&dev->pending[0]) ||
1793 !list_empty(&dev->pending[1]) ||
1794 !list_empty(&dev->pending[2]))
b55f4f06 1795 ipath_dev_err(dev->dd, "pending list not empty!\n");
6522108f 1796 if (!list_empty(&dev->piowait))
b55f4f06 1797 ipath_dev_err(dev->dd, "piowait list not empty!\n");
6522108f 1798 if (!list_empty(&dev->rnrwait))
b55f4f06 1799 ipath_dev_err(dev->dd, "rnrwait list not empty!\n");
6522108f 1800 if (!ipath_mcast_tree_empty())
b55f4f06 1801 ipath_dev_err(dev->dd, "multicast table memory leak!\n");
6522108f
BS		 1802 /*
1803 * Note that ipath_unregister_ib_device() can be called before all
1804 * the QPs are destroyed!
1805 */
1806 ipath_free_all_qps(&dev->qp_table);
1807 kfree(dev->qp_table.table);
1808 kfree(dev->lk_table.table);
1809 ib_dealloc_device(ibdev);
1810}
1811
6522108f
BS		 1812static ssize_t show_rev(struct class_device *cdev, char *buf)
1813{
5494c22b
RD		 1814 struct ipath_ibdev *dev =
1815 container_of(cdev, struct ipath_ibdev, ibdev.class_dev);
6522108f 1816
34b2aafe 1817 return sprintf(buf, "%x\n", dev->dd->ipath_pcirev);
6522108f
BS		 1818}
1819
1820static ssize_t show_hca(struct class_device *cdev, char *buf)
1821{
5494c22b
RD		 1822 struct ipath_ibdev *dev =
1823 container_of(cdev, struct ipath_ibdev, ibdev.class_dev);
1824 int ret;
6522108f 1825
34b2aafe 1826 ret = dev->dd->ipath_f_get_boardname(dev->dd, buf, 128);
5494c22b
RD		 1827 if (ret < 0)
1828 goto bail;
1829 strcat(buf, "\n");
1830 ret = strlen(buf);
6522108f
BS		 1831
1832bail:
1833 return ret;
1834}
1835
1836static ssize_t show_stats(struct class_device *cdev, char *buf)
1837{
5494c22b
RD		 1838 struct ipath_ibdev *dev =
1839 container_of(cdev, struct ipath_ibdev, ibdev.class_dev);
1840 int i;
1841 int len;
1842
1843 len = sprintf(buf,
1844 "RC resends %d\n"
9b2017f1 1845 "RC no QACK %d\n"
5494c22b
RD		 1846 "RC ACKs %d\n"
1847 "RC SEQ NAKs %d\n"
1848 "RC RDMA seq %d\n"
1849 "RC RNR NAKs %d\n"
1850 "RC OTH NAKs %d\n"
1851 "RC timeouts %d\n"
1852 "RC RDMA dup %d\n"
6022943e 1853 "RC stalls %d\n"
5494c22b
RD		 1854 "piobuf wait %d\n"
1855 "no piobuf %d\n"
1856 "PKT drops %d\n"
1857 "WQE errs %d\n",
1858 dev->n_rc_resends, dev->n_rc_qacks, dev->n_rc_acks,
1859 dev->n_seq_naks, dev->n_rdma_seq, dev->n_rnr_naks,
1860 dev->n_other_naks, dev->n_timeouts,
6022943e 1861 dev->n_rdma_dup_busy, dev->n_rc_stalls, dev->n_piowait,
5494c22b
RD		 1862 dev->n_no_piobuf, dev->n_pkt_drops, dev->n_wqe_errs);
1863 for (i = 0; i < ARRAY_SIZE(dev->opstats); i++) {
6522108f
BS		 1864 const struct ipath_opcode_stats *si = &dev->opstats[i];
1865
5494c22b
RD		 1866 if (!si->n_packets && !si->n_bytes)
1867 continue;
1868 len += sprintf(buf + len, "%02x %llu/%llu\n", i,
6522108f 1869 (unsigned long long) si->n_packets,
5494c22b
RD		 1870 (unsigned long long) si->n_bytes);
1871 }
1872 return len;
6522108f
BS		 1873}
1874
1875static CLASS_DEVICE_ATTR(hw_rev, S_IRUGO, show_rev, NULL);
1876static CLASS_DEVICE_ATTR(hca_type, S_IRUGO, show_hca, NULL);
1877static CLASS_DEVICE_ATTR(board_id, S_IRUGO, show_hca, NULL);
1878static CLASS_DEVICE_ATTR(stats, S_IRUGO, show_stats, NULL);
1879
1880static struct class_device_attribute *ipath_class_attributes[] = {
5494c22b
RD		 1881 &class_device_attr_hw_rev,
1882 &class_device_attr_hca_type,
1883 &class_device_attr_board_id,
1884 &class_device_attr_stats
6522108f
BS		 1885};
1886
1887static int ipath_verbs_register_sysfs(struct ib_device *dev)
1888{
5494c22b 1889 int i;
6522108f
BS		 1890 int ret;
1891
5494c22b
RD		 1892 for (i = 0; i < ARRAY_SIZE(ipath_class_attributes); ++i)
1893 if (class_device_create_file(&dev->class_dev,
1894 ipath_class_attributes[i])) {
1895 ret = 1;
6522108f
BS		 1896 goto bail;
1897 }
1898
5494c22b 1899 ret = 0;
6522108f
BS		 1900
1901bail:
1902 return ret;
1903}
Ubuntu Hirsute Linux kernel mirror