projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
IB/hfi1: Do not allocate PIO send contexts for VNIC
[linux-2.6-block.git] / drivers / infiniband / hw / hfi1 / mad.c
CommitLineData
77241056 1/*
3ca4fbc8 2 * Copyright(c) 2015-2017 Intel Corporation.
77241056
MM		 3 *
4 * This file is provided under a dual BSD/GPLv2 license. When using or
5 * redistributing this file, you may do so under either license.
6 *
7 * GPL LICENSE SUMMARY
8 *
77241056
MM		 9 * This program is free software; you can redistribute it and/or modify
10 * it under the terms of version 2 of the GNU General Public License as
11 * published by the Free Software Foundation.
12 *
13 * This program is distributed in the hope that it will be useful, but
14 * WITHOUT ANY WARRANTY; without even the implied warranty of
15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
16 * General Public License for more details.
17 *
18 * BSD LICENSE
19 *
77241056
MM		 20 * Redistribution and use in source and binary forms, with or without
21 * modification, are permitted provided that the following conditions
22 * are met:
23 *
24 * - Redistributions of source code must retain the above copyright
25 * notice, this list of conditions and the following disclaimer.
26 * - Redistributions in binary form must reproduce the above copyright
27 * notice, this list of conditions and the following disclaimer in
28 * the documentation and/or other materials provided with the
29 * distribution.
30 * - Neither the name of Intel Corporation nor the names of its
31 * contributors may be used to endorse or promote products derived
32 * from this software without specific prior written permission.
33 *
34 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
35 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
36 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
37 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
38 * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
39 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
40 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
41 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
42 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
43 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
44 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
45 *
46 */
47
48#include <linux/net.h>
13c19222 49#include <rdma/opa_addr.h>
77241056
MM		 50#define OPA_NUM_PKEY_BLOCKS_PER_SMP (OPA_SMP_DR_DATA_SIZE \
51 / (OPA_PARTITION_TABLE_BLK_SIZE * sizeof(u16)))
52
53#include "hfi.h"
54#include "mad.h"
55#include "trace.h"
0ec79e87 56#include "qp.h"
2280740f 57#include "vnic.h"
77241056
MM		 58
59/* the reset value from the FM is supposed to be 0xffff, handle both */
60#define OPA_LINK_WIDTH_RESET_OLD 0x0fff
61#define OPA_LINK_WIDTH_RESET 0xffff
62
bf90aadd
MR		 63struct trap_node {
64 struct list_head list;
65 struct opa_mad_notice_attr data;
66 __be64 tid;
67 int len;
68 u32 retry;
69 u8 in_use;
70 u8 repress;
71};
72
f1685179
ND		 73static int smp_length_check(u32 data_size, u32 request_len)
74{
75 if (unlikely(request_len < data_size))
76 return -EINVAL;
77
78 return 0;
79}
80
77241056
MM		 81static int reply(struct ib_mad_hdr *smp)
82{
83 /*
84 * The verbs framework will handle the directed/LID route
85 * packet changes.
86 */
87 smp->method = IB_MGMT_METHOD_GET_RESP;
88 if (smp->mgmt_class == IB_MGMT_CLASS_SUBN_DIRECTED_ROUTE)
89 smp->status |= IB_SMP_DIRECTION;
90 return IB_MAD_RESULT_SUCCESS | IB_MAD_RESULT_REPLY;
91}
92
93static inline void clear_opa_smp_data(struct opa_smp *smp)
94{
95 void *data = opa_get_smp_data(smp);
96 size_t size = opa_get_smp_data_size(smp);
97
98 memset(data, 0, size);
99}
100
406310c6
SS		 101static u16 hfi1_lookup_pkey_value(struct hfi1_ibport *ibp, int pkey_idx)
102{
103 struct hfi1_pportdata *ppd = ppd_from_ibp(ibp);
104
105 if (pkey_idx < ARRAY_SIZE(ppd->pkeys))
106 return ppd->pkeys[pkey_idx];
107
108 return 0;
109}
110
34d351f8
SS		 111void hfi1_event_pkey_change(struct hfi1_devdata *dd, u8 port)
112{
113 struct ib_event event;
114
115 event.event = IB_EVENT_PKEY_CHANGE;
116 event.device = &dd->verbs_dev.rdi.ibdev;
117 event.element.port_num = port;
118 ib_dispatch_event(&event);
119}
120
bf90aadd
MR		 121/*
122 * If the port is down, clean up all pending traps. We need to be careful
123 * with the given trap, because it may be queued.
124 */
125static void cleanup_traps(struct hfi1_ibport *ibp, struct trap_node *trap)
126{
127 struct trap_node *node, *q;
128 unsigned long flags;
129 struct list_head trap_list;
130 int i;
131
132 for (i = 0; i < RVT_MAX_TRAP_LISTS; i++) {
133 spin_lock_irqsave(&ibp->rvp.lock, flags);
134 list_replace_init(&ibp->rvp.trap_lists[i].list, &trap_list);
135 ibp->rvp.trap_lists[i].list_len = 0;
136 spin_unlock_irqrestore(&ibp->rvp.lock, flags);
137
138 /*
139 * Remove all items from the list, freeing all the non-given
140 * traps.
141 */
142 list_for_each_entry_safe(node, q, &trap_list, list) {
143 list_del(&node->list);
144 if (node != trap)
145 kfree(node);
146 }
147 }
148
149 /*
150 * If this wasn't on one of the lists it would not be freed. If it
151 * was on the list, it is now safe to free.
152 */
153 kfree(trap);
154}
155
156static struct trap_node *check_and_add_trap(struct hfi1_ibport *ibp,
157 struct trap_node *trap)
158{
159 struct trap_node *node;
160 struct trap_list *trap_list;
161 unsigned long flags;
162 unsigned long timeout;
163 int found = 0;
ec0d8b8a
KA		 164 unsigned int queue_id;
165 static int trap_count;
166
167 queue_id = trap->data.generic_type & 0x0F;
168 if (queue_id >= RVT_MAX_TRAP_LISTS) {
169 trap_count++;
170 pr_err_ratelimited("hfi1: Invalid trap 0x%0x dropped. Total dropped: %d\n",
d68e68e5 171 trap->data.generic_type, trap_count);
ec0d8b8a
KA		 172 kfree(trap);
173 return NULL;
174 }
bf90aadd
MR		 175
176 /*
177 * Since the retry (handle timeout) does not remove a trap request
178 * from the list, all we have to do is compare the node.
179 */
180 spin_lock_irqsave(&ibp->rvp.lock, flags);
ec0d8b8a 181 trap_list = &ibp->rvp.trap_lists[queue_id];
bf90aadd
MR		 182
183 list_for_each_entry(node, &trap_list->list, list) {
184 if (node == trap) {
185 node->retry++;
186 found = 1;
187 break;
188 }
189 }
190
191 /* If it is not on the list, add it, limited to RVT-MAX_TRAP_LEN. */
192 if (!found) {
193 if (trap_list->list_len < RVT_MAX_TRAP_LEN) {
194 trap_list->list_len++;
195 list_add_tail(&trap->list, &trap_list->list);
196 } else {
028e0a67 197 pr_warn_ratelimited("hfi1: Maximum trap limit reached for 0x%0x traps\n",
bf90aadd
MR		 198 trap->data.generic_type);
199 kfree(trap);
200 }
201 }
202
203 /*
204 * Next check to see if there is a timer pending. If not, set it up
205 * and get the first trap from the list.
206 */
207 node = NULL;
208 if (!timer_pending(&ibp->rvp.trap_timer)) {
209 /*
210 * o14-2
211 * If the time out is set we have to wait until it expires
212 * before the trap can be sent.
213 * This should be > RVT_TRAP_TIMEOUT
214 */
215 timeout = (RVT_TRAP_TIMEOUT *
216 (1UL << ibp->rvp.subnet_timeout)) / 1000;
217 mod_timer(&ibp->rvp.trap_timer,
218 jiffies + usecs_to_jiffies(timeout));
219 node = list_first_entry(&trap_list->list, struct trap_node,
220 list);
221 node->in_use = 1;
222 }
223 spin_unlock_irqrestore(&ibp->rvp.lock, flags);
224
225 return node;
226}
227
228static void subn_handle_opa_trap_repress(struct hfi1_ibport *ibp,
229 struct opa_smp *smp)
230{
231 struct trap_list *trap_list;
232 struct trap_node *trap;
233 unsigned long flags;
234 int i;
235
236 if (smp->attr_id != IB_SMP_ATTR_NOTICE)
237 return;
238
239 spin_lock_irqsave(&ibp->rvp.lock, flags);
240 for (i = 0; i < RVT_MAX_TRAP_LISTS; i++) {
241 trap_list = &ibp->rvp.trap_lists[i];
242 trap = list_first_entry_or_null(&trap_list->list,
243 struct trap_node, list);
244 if (trap && trap->tid == smp->tid) {
245 if (trap->in_use) {
246 trap->repress = 1;
247 } else {
248 trap_list->list_len--;
249 list_del(&trap->list);
250 kfree(trap);
251 }
252 break;
253 }
254 }
255 spin_unlock_irqrestore(&ibp->rvp.lock, flags);
256}
257
51e658f5
DC		 258static void hfi1_update_sm_ah_attr(struct hfi1_ibport *ibp,
259 struct rdma_ah_attr *attr, u32 dlid)
260{
261 rdma_ah_set_dlid(attr, dlid);
262 rdma_ah_set_port_num(attr, ppd_from_ibp(ibp)->port);
263 if (dlid >= be16_to_cpu(IB_MULTICAST_LID_BASE)) {
264 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
265
266 rdma_ah_set_ah_flags(attr, IB_AH_GRH);
267 grh->sgid_index = 0;
268 grh->hop_limit = 1;
269 grh->dgid.global.subnet_prefix =
270 ibp->rvp.gid_prefix;
271 grh->dgid.global.interface_id = OPA_MAKE_ID(dlid);
272 }
273}
274
275static int hfi1_modify_qp0_ah(struct hfi1_ibport *ibp,
276 struct rvt_ah *ah, u32 dlid)
277{
278 struct rdma_ah_attr attr;
279 struct rvt_qp *qp0;
280 int ret = -EINVAL;
281
282 memset(&attr, 0, sizeof(attr));
283 attr.type = ah->ibah.type;
284 hfi1_update_sm_ah_attr(ibp, &attr, dlid);
285 rcu_read_lock();
286 qp0 = rcu_dereference(ibp->rvp.qp[0]);
287 if (qp0)
288 ret = rdma_modify_ah(&ah->ibah, &attr);
289 rcu_read_unlock();
290 return ret;
291}
292
293static struct ib_ah *hfi1_create_qp0_ah(struct hfi1_ibport *ibp, u32 dlid)
294{
295 struct rdma_ah_attr attr;
296 struct ib_ah *ah = ERR_PTR(-EINVAL);
297 struct rvt_qp *qp0;
298 struct hfi1_pportdata *ppd = ppd_from_ibp(ibp);
299 struct hfi1_devdata *dd = dd_from_ppd(ppd);
300 u8 port_num = ppd->port;
301
302 memset(&attr, 0, sizeof(attr));
303 attr.type = rdma_ah_find_type(&dd->verbs_dev.rdi.ibdev, port_num);
304 hfi1_update_sm_ah_attr(ibp, &attr, dlid);
305 rcu_read_lock();
306 qp0 = rcu_dereference(ibp->rvp.qp[0]);
307 if (qp0)
308 ah = rdma_create_ah(qp0->ibqp.pd, &attr);
309 rcu_read_unlock();
310 return ah;
311}
312
bf90aadd 313static void send_trap(struct hfi1_ibport *ibp, struct trap_node *trap)
77241056
MM		 314{
315 struct ib_mad_send_buf *send_buf;
316 struct ib_mad_agent *agent;
5cd24119 317 struct opa_smp *smp;
77241056 318 unsigned long flags;
77241056
MM		 319 int pkey_idx;
320 u32 qpn = ppd_from_ibp(ibp)->sm_trap_qp;
321
4eb06882 322 agent = ibp->rvp.send_agent;
bf90aadd
MR		 323 if (!agent) {
324 cleanup_traps(ibp, trap);
77241056 325 return;
bf90aadd 326 }
77241056
MM		 327
328 /* o14-3.2.1 */
bf90aadd
MR		 329 if (driver_lstate(ppd_from_ibp(ibp)) != IB_PORT_ACTIVE) {
330 cleanup_traps(ibp, trap);
77241056 331 return;
bf90aadd 332 }
77241056 333
bf90aadd
MR		 334 /* Add the trap to the list if necessary and see if we can send it */
335 trap = check_and_add_trap(ibp, trap);
336 if (!trap)
77241056
MM		 337 return;
338
339 pkey_idx = hfi1_lookup_pkey_idx(ibp, LIM_MGMT_P_KEY);
340 if (pkey_idx < 0) {
341 pr_warn("%s: failed to find limited mgmt pkey, defaulting 0x%x\n",
342 __func__, hfi1_get_pkey(ibp, 1));
343 pkey_idx = 1;
344 }
345
346 send_buf = ib_create_send_mad(agent, qpn, pkey_idx, 0,
347 IB_MGMT_MAD_HDR, IB_MGMT_MAD_DATA,
348 GFP_ATOMIC, IB_MGMT_BASE_VERSION);
349 if (IS_ERR(send_buf))
350 return;
351
352 smp = send_buf->mad;
5cd24119 353 smp->base_version = OPA_MGMT_BASE_VERSION;
77241056 354 smp->mgmt_class = IB_MGMT_CLASS_SUBN_LID_ROUTED;
9fa240bb 355 smp->class_version = OPA_SM_CLASS_VERSION;
77241056 356 smp->method = IB_MGMT_METHOD_TRAP;
bf90aadd
MR		 357
358 /* Only update the transaction ID for new traps (o13-5). */
359 if (trap->tid == 0) {
360 ibp->rvp.tid++;
361 /* make sure that tid != 0 */
362 if (ibp->rvp.tid == 0)
363 ibp->rvp.tid++;
364 trap->tid = cpu_to_be64(ibp->rvp.tid);
365 }
366 smp->tid = trap->tid;
367
77241056
MM		 368 smp->attr_id = IB_SMP_ATTR_NOTICE;
369 /* o14-1: smp->mkey = 0; */
bf90aadd
MR		 370
371 memcpy(smp->route.lid.data, &trap->data, trap->len);
77241056 372
4eb06882 373 spin_lock_irqsave(&ibp->rvp.lock, flags);
9c4a311e 374 if (!ibp->rvp.sm_ah) {
4eb06882 375 if (ibp->rvp.sm_lid != be16_to_cpu(IB_LID_PERMISSIVE)) {
77241056
MM		 376 struct ib_ah *ah;
377
4eb06882 378 ah = hfi1_create_qp0_ah(ibp, ibp->rvp.sm_lid);
e490974e 379 if (IS_ERR(ah)) {
bf90aadd
MR		 380 spin_unlock_irqrestore(&ibp->rvp.lock, flags);
381 return;
77241056 382 }
bf90aadd
MR		 383 send_buf->ah = ah;
384 ibp->rvp.sm_ah = ibah_to_rvtah(ah);
e490974e 385 } else {
bf90aadd
MR		 386 spin_unlock_irqrestore(&ibp->rvp.lock, flags);
387 return;
e490974e 388 }
77241056 389 } else {
9c4a311e 390 send_buf->ah = &ibp->rvp.sm_ah->ibah;
77241056 391 }
bf90aadd
MR		 392
393 /*
394 * If the trap was repressed while things were getting set up, don't
395 * bother sending it. This could happen for a retry.
396 */
397 if (trap->repress) {
398 list_del(&trap->list);
399 spin_unlock_irqrestore(&ibp->rvp.lock, flags);
400 kfree(trap);
401 ib_free_send_mad(send_buf);
402 return;
403 }
404
405 trap->in_use = 0;
4eb06882 406 spin_unlock_irqrestore(&ibp->rvp.lock, flags);
77241056 407
bf90aadd 408 if (ib_post_send_mad(send_buf, NULL))
77241056 409 ib_free_send_mad(send_buf);
bf90aadd
MR		 410}
411
8064135e 412void hfi1_handle_trap_timer(struct timer_list *t)
bf90aadd 413{
8064135e 414 struct hfi1_ibport *ibp = from_timer(ibp, t, rvp.trap_timer);
bf90aadd
MR		 415 struct trap_node *trap = NULL;
416 unsigned long flags;
417 int i;
418
419 /* Find the trap with the highest priority */
420 spin_lock_irqsave(&ibp->rvp.lock, flags);
421 for (i = 0; !trap && i < RVT_MAX_TRAP_LISTS; i++) {
422 trap = list_first_entry_or_null(&ibp->rvp.trap_lists[i].list,
423 struct trap_node, list);
77241056 424 }
bf90aadd
MR		 425 spin_unlock_irqrestore(&ibp->rvp.lock, flags);
426
427 if (trap)
428 send_trap(ibp, trap);
429}
430
431static struct trap_node *create_trap_node(u8 type, __be16 trap_num, u32 lid)
432{
433 struct trap_node *trap;
434
435 trap = kzalloc(sizeof(*trap), GFP_ATOMIC);
436 if (!trap)
437 return NULL;
438
439 INIT_LIST_HEAD(&trap->list);
440 trap->data.generic_type = type;
441 trap->data.prod_type_lsb = IB_NOTICE_PROD_CA;
442 trap->data.trap_num = trap_num;
443 trap->data.issuer_lid = cpu_to_be32(lid);
444
445 return trap;
77241056
MM		 446}
447
448/*
13d84914 449 * Send a bad P_Key trap (ch. 14.3.8).
77241056 450 */
13d84914 451void hfi1_bad_pkey(struct hfi1_ibport *ibp, u32 key, u32 sl,
88733e3b 452 u32 qp1, u32 qp2, u32 lid1, u32 lid2)
77241056 453{
bf90aadd 454 struct trap_node *trap;
5cd24119 455 u32 lid = ppd_from_ibp(ibp)->lid;
77241056 456
4eb06882 457 ibp->rvp.n_pkt_drops++;
13d84914 458 ibp->rvp.pkey_violations++;
77241056 459
bf90aadd
MR		 460 trap = create_trap_node(IB_NOTICE_TYPE_SECURITY, OPA_TRAP_BAD_P_KEY,
461 lid);
462 if (!trap)
463 return;
464
77241056 465 /* Send violation trap */
88733e3b
DH		 466 trap->data.ntc_257_258.lid1 = cpu_to_be32(lid1);
467 trap->data.ntc_257_258.lid2 = cpu_to_be32(lid2);
bf90aadd
MR		 468 trap->data.ntc_257_258.key = cpu_to_be32(key);
469 trap->data.ntc_257_258.sl = sl << 3;
470 trap->data.ntc_257_258.qp1 = cpu_to_be32(qp1);
471 trap->data.ntc_257_258.qp2 = cpu_to_be32(qp2);
472
473 trap->len = sizeof(trap->data);
474 send_trap(ibp, trap);
77241056
MM		 475}
476
477/*
478 * Send a bad M_Key trap (ch. 14.3.9).
479 */
480static void bad_mkey(struct hfi1_ibport *ibp, struct ib_mad_hdr *mad,
481 __be64 mkey, __be32 dr_slid, u8 return_path[], u8 hop_cnt)
482{
bf90aadd 483 struct trap_node *trap;
5cd24119 484 u32 lid = ppd_from_ibp(ibp)->lid;
77241056 485
bf90aadd
MR		 486 trap = create_trap_node(IB_NOTICE_TYPE_SECURITY, OPA_TRAP_BAD_M_KEY,
487 lid);
488 if (!trap)
489 return;
490
77241056 491 /* Send violation trap */
bf90aadd
MR		 492 trap->data.ntc_256.lid = trap->data.issuer_lid;
493 trap->data.ntc_256.method = mad->method;
494 trap->data.ntc_256.attr_id = mad->attr_id;
495 trap->data.ntc_256.attr_mod = mad->attr_mod;
496 trap->data.ntc_256.mkey = mkey;
77241056 497 if (mad->mgmt_class == IB_MGMT_CLASS_SUBN_DIRECTED_ROUTE) {
bf90aadd
MR		 498 trap->data.ntc_256.dr_slid = dr_slid;
499 trap->data.ntc_256.dr_trunc_hop = IB_NOTICE_TRAP_DR_NOTICE;
500 if (hop_cnt > ARRAY_SIZE(trap->data.ntc_256.dr_rtn_path)) {
501 trap->data.ntc_256.dr_trunc_hop |=
77241056 502 IB_NOTICE_TRAP_DR_TRUNC;
bf90aadd 503 hop_cnt = ARRAY_SIZE(trap->data.ntc_256.dr_rtn_path);
77241056 504 }
bf90aadd
MR		 505 trap->data.ntc_256.dr_trunc_hop |= hop_cnt;
506 memcpy(trap->data.ntc_256.dr_rtn_path, return_path,
77241056
MM		 507 hop_cnt);
508 }
509
bf90aadd
MR		 510 trap->len = sizeof(trap->data);
511
512 send_trap(ibp, trap);
77241056
MM		 513}
514
515/*
516 * Send a Port Capability Mask Changed trap (ch. 14.3.11).
517 */
45b59eef 518void hfi1_cap_mask_chg(struct rvt_dev_info *rdi, u8 port_num)
77241056 519{
bf90aadd 520 struct trap_node *trap;
45b59eef
HC		 521 struct hfi1_ibdev *verbs_dev = dev_from_rdi(rdi);
522 struct hfi1_devdata *dd = dd_from_dev(verbs_dev);
523 struct hfi1_ibport *ibp = &dd->pport[port_num - 1].ibport_data;
5cd24119
EK		 524 u32 lid = ppd_from_ibp(ibp)->lid;
525
bf90aadd
MR		 526 trap = create_trap_node(IB_NOTICE_TYPE_INFO,
527 OPA_TRAP_CHANGE_CAPABILITY,
528 lid);
529 if (!trap)
530 return;
77241056 531
bf90aadd
MR		 532 trap->data.ntc_144.lid = trap->data.issuer_lid;
533 trap->data.ntc_144.new_cap_mask = cpu_to_be32(ibp->rvp.port_cap_flags);
534 trap->data.ntc_144.cap_mask3 = cpu_to_be16(ibp->rvp.port_cap3_flags);
77241056 535
bf90aadd
MR		 536 trap->len = sizeof(trap->data);
537 send_trap(ibp, trap);
77241056
MM		 538}
539
540/*
541 * Send a System Image GUID Changed trap (ch. 14.3.12).
542 */
543void hfi1_sys_guid_chg(struct hfi1_ibport *ibp)
544{
bf90aadd 545 struct trap_node *trap;
5cd24119
EK		 546 u32 lid = ppd_from_ibp(ibp)->lid;
547
bf90aadd
MR		 548 trap = create_trap_node(IB_NOTICE_TYPE_INFO, OPA_TRAP_CHANGE_SYSGUID,
549 lid);
550 if (!trap)
551 return;
77241056 552
bf90aadd
MR		 553 trap->data.ntc_145.new_sys_guid = ib_hfi1_sys_image_guid;
554 trap->data.ntc_145.lid = trap->data.issuer_lid;
77241056 555
bf90aadd
MR		 556 trap->len = sizeof(trap->data);
557 send_trap(ibp, trap);
77241056
MM		 558}
559
560/*
561 * Send a Node Description Changed trap (ch. 14.3.13).
562 */
563void hfi1_node_desc_chg(struct hfi1_ibport *ibp)
564{
bf90aadd 565 struct trap_node *trap;
5cd24119
EK		 566 u32 lid = ppd_from_ibp(ibp)->lid;
567
bf90aadd
MR		 568 trap = create_trap_node(IB_NOTICE_TYPE_INFO,
569 OPA_TRAP_CHANGE_CAPABILITY,
570 lid);
571 if (!trap)
572 return;
77241056 573
bf90aadd
MR		 574 trap->data.ntc_144.lid = trap->data.issuer_lid;
575 trap->data.ntc_144.change_flags =
5cd24119 576 cpu_to_be16(OPA_NOTICE_TRAP_NODE_DESC_CHG);
77241056 577
bf90aadd
MR		 578 trap->len = sizeof(trap->data);
579 send_trap(ibp, trap);
77241056
MM		 580}
581
582static int __subn_get_opa_nodedesc(struct opa_smp *smp, u32 am,
583 u8 *data, struct ib_device *ibdev,
f1685179 584 u8 port, u32 *resp_len, u32 max_len)
77241056
MM		 585{
586 struct opa_node_description *nd;
587
f1685179 588 if (am || smp_length_check(sizeof(*nd), max_len)) {
77241056
MM		 589 smp->status |= IB_SMP_INVALID_FIELD;
590 return reply((struct ib_mad_hdr *)smp);
591 }
592
593 nd = (struct opa_node_description *)data;
594
595 memcpy(nd->data, ibdev->node_desc, sizeof(nd->data));
596
597 if (resp_len)
598 *resp_len += sizeof(*nd);
599
600 return reply((struct ib_mad_hdr *)smp);
601}
602
603static int __subn_get_opa_nodeinfo(struct opa_smp *smp, u32 am, u8 *data,
604 struct ib_device *ibdev, u8 port,
f1685179 605 u32 *resp_len, u32 max_len)
77241056
MM		 606{
607 struct opa_node_info *ni;
608 struct hfi1_devdata *dd = dd_from_ibdev(ibdev);
609 unsigned pidx = port - 1; /* IB number port from 1, hw from 0 */
610
611 ni = (struct opa_node_info *)data;
612
613 /* GUID 0 is illegal */
a6cd5f08 614 if (am || pidx >= dd->num_pports || ibdev->node_guid == 0 ||
f1685179 615 smp_length_check(sizeof(*ni), max_len) ||
a6cd5f08 616 get_sguid(to_iport(ibdev, port), HFI1_PORT_GUID_INDEX) == 0) {
77241056
MM		 617 smp->status |= IB_SMP_INVALID_FIELD;
618 return reply((struct ib_mad_hdr *)smp);
619 }
620
a6cd5f08 621 ni->port_guid = get_sguid(to_iport(ibdev, port), HFI1_PORT_GUID_INDEX);
77241056 622 ni->base_version = OPA_MGMT_BASE_VERSION;
9fa240bb 623 ni->class_version = OPA_SM_CLASS_VERSION;
77241056
MM		 624 ni->node_type = 1; /* channel adapter */
625 ni->num_ports = ibdev->phys_port_cnt;
626 /* This is already in network order */
627 ni->system_image_guid = ib_hfi1_sys_image_guid;
a6cd5f08 628 ni->node_guid = ibdev->node_guid;
77241056
MM		 629 ni->partition_cap = cpu_to_be16(hfi1_get_npkeys(dd));
630 ni->device_id = cpu_to_be16(dd->pcidev->device);
631 ni->revision = cpu_to_be32(dd->minrev);
632 ni->local_port_num = port;
633 ni->vendor_id[0] = dd->oui1;
634 ni->vendor_id[1] = dd->oui2;
635 ni->vendor_id[2] = dd->oui3;
636
637 if (resp_len)
638 *resp_len += sizeof(*ni);
639
640 return reply((struct ib_mad_hdr *)smp);
641}
642
643static int subn_get_nodeinfo(struct ib_smp *smp, struct ib_device *ibdev,
644 u8 port)
645{
646 struct ib_node_info *nip = (struct ib_node_info *)&smp->data;
647 struct hfi1_devdata *dd = dd_from_ibdev(ibdev);
648 unsigned pidx = port - 1; /* IB number port from 1, hw from 0 */
649
650 /* GUID 0 is illegal */
651 if (smp->attr_mod || pidx >= dd->num_pports ||
a6cd5f08
JP		 652 ibdev->node_guid == 0 ||
653 get_sguid(to_iport(ibdev, port), HFI1_PORT_GUID_INDEX) == 0) {
77241056 654 smp->status |= IB_SMP_INVALID_FIELD;
a6cd5f08
JP		 655 return reply((struct ib_mad_hdr *)smp);
656 }
77241056 657
a6cd5f08 658 nip->port_guid = get_sguid(to_iport(ibdev, port), HFI1_PORT_GUID_INDEX);
77241056 659 nip->base_version = OPA_MGMT_BASE_VERSION;
9fa240bb 660 nip->class_version = OPA_SM_CLASS_VERSION;
77241056
MM		 661 nip->node_type = 1; /* channel adapter */
662 nip->num_ports = ibdev->phys_port_cnt;
663 /* This is already in network order */
664 nip->sys_guid = ib_hfi1_sys_image_guid;
a6cd5f08 665 nip->node_guid = ibdev->node_guid;
77241056
MM		 666 nip->partition_cap = cpu_to_be16(hfi1_get_npkeys(dd));
667 nip->device_id = cpu_to_be16(dd->pcidev->device);
668 nip->revision = cpu_to_be32(dd->minrev);
669 nip->local_port_num = port;
670 nip->vendor_id[0] = dd->oui1;
671 nip->vendor_id[1] = dd->oui2;
672 nip->vendor_id[2] = dd->oui3;
673
674 return reply((struct ib_mad_hdr *)smp);
675}
676
677static void set_link_width_enabled(struct hfi1_pportdata *ppd, u32 w)
678{
679 (void)hfi1_set_ib_cfg(ppd, HFI1_IB_CFG_LWID_ENB, w);
680}
681
682static void set_link_width_downgrade_enabled(struct hfi1_pportdata *ppd, u32 w)
683{
684 (void)hfi1_set_ib_cfg(ppd, HFI1_IB_CFG_LWID_DG_ENB, w);
685}
686
687static void set_link_speed_enabled(struct hfi1_pportdata *ppd, u32 s)
688{
689 (void)hfi1_set_ib_cfg(ppd, HFI1_IB_CFG_SPD_ENB, s);
690}
691
692static int check_mkey(struct hfi1_ibport *ibp, struct ib_mad_hdr *mad,
693 int mad_flags, __be64 mkey, __be32 dr_slid,
694 u8 return_path[], u8 hop_cnt)
695{
696 int valid_mkey = 0;
697 int ret = 0;
698
699 /* Is the mkey in the process of expiring? */
4eb06882
DD		 700 if (ibp->rvp.mkey_lease_timeout &&
701 time_after_eq(jiffies, ibp->rvp.mkey_lease_timeout)) {
77241056 702 /* Clear timeout and mkey protection field. */
4eb06882
DD		 703 ibp->rvp.mkey_lease_timeout = 0;
704 ibp->rvp.mkeyprot = 0;
77241056
MM		 705 }
706
4eb06882
DD		 707 if ((mad_flags & IB_MAD_IGNORE_MKEY) || ibp->rvp.mkey == 0 ||
708 ibp->rvp.mkey == mkey)
77241056
MM		 709 valid_mkey = 1;
710
711 /* Unset lease timeout on any valid Get/Set/TrapRepress */
4eb06882 712 if (valid_mkey && ibp->rvp.mkey_lease_timeout &&
77241056
MM		 713 (mad->method == IB_MGMT_METHOD_GET ||
714 mad->method == IB_MGMT_METHOD_SET ||
715 mad->method == IB_MGMT_METHOD_TRAP_REPRESS))
4eb06882 716 ibp->rvp.mkey_lease_timeout = 0;
77241056
MM		 717
718 if (!valid_mkey) {
719 switch (mad->method) {
720 case IB_MGMT_METHOD_GET:
721 /* Bad mkey not a violation below level 2 */
4eb06882 722 if (ibp->rvp.mkeyprot < 2)
77241056 723 break;
6ffeb21f 724 /* fall through */
77241056
MM		 725 case IB_MGMT_METHOD_SET:
726 case IB_MGMT_METHOD_TRAP_REPRESS:
4eb06882
DD		 727 if (ibp->rvp.mkey_violations != 0xFFFF)
728 ++ibp->rvp.mkey_violations;
729 if (!ibp->rvp.mkey_lease_timeout &&
730 ibp->rvp.mkey_lease_period)
731 ibp->rvp.mkey_lease_timeout = jiffies +
732 ibp->rvp.mkey_lease_period * HZ;
77241056
MM		 733 /* Generate a trap notice. */
734 bad_mkey(ibp, mad, mkey, dr_slid, return_path,
735 hop_cnt);
736 ret = 1;
737 }
738 }
739
740 return ret;
741}
742
743/*
744 * The SMA caches reads from LCB registers in case the LCB is unavailable.
745 * (The LCB is unavailable in certain link states, for example.)
746 */
747struct lcb_datum {
748 u32 off;
749 u64 val;
750};
751
752static struct lcb_datum lcb_cache[] = {
753 { DC_LCB_STS_ROUND_TRIP_LTP_CNT, 0 },
754};
755
756static int write_lcb_cache(u32 off, u64 val)
757{
758 int i;
759
760 for (i = 0; i < ARRAY_SIZE(lcb_cache); i++) {
761 if (lcb_cache[i].off == off) {
762 lcb_cache[i].val = val;
763 return 0;
764 }
765 }
766
767 pr_warn("%s bad offset 0x%x\n", __func__, off);
768 return -1;
769}
770
771static int read_lcb_cache(u32 off, u64 *val)
772{
773 int i;
774
775 for (i = 0; i < ARRAY_SIZE(lcb_cache); i++) {
776 if (lcb_cache[i].off == off) {
777 *val = lcb_cache[i].val;
778 return 0;
779 }
780 }
781
782 pr_warn("%s bad offset 0x%x\n", __func__, off);
783 return -1;
784}
785
786void read_ltp_rtt(struct hfi1_devdata *dd)
787{
788 u64 reg;
789
790 if (read_lcb_csr(dd, DC_LCB_STS_ROUND_TRIP_LTP_CNT, &reg))
791 dd_dev_err(dd, "%s: unable to read LTP RTT\n", __func__);
792 else
793 write_lcb_cache(DC_LCB_STS_ROUND_TRIP_LTP_CNT, reg);
794}
795
77241056
MM		 796static int __subn_get_opa_portinfo(struct opa_smp *smp, u32 am, u8 *data,
797 struct ib_device *ibdev, u8 port,
f1685179 798 u32 *resp_len, u32 max_len)
77241056
MM		 799{
800 int i;
801 struct hfi1_devdata *dd;
802 struct hfi1_pportdata *ppd;
803 struct hfi1_ibport *ibp;
804 struct opa_port_info *pi = (struct opa_port_info *)data;
805 u8 mtu;
806 u8 credit_rate;
409b1462 807 u8 is_beaconing_active;
77241056
MM		 808 u32 state;
809 u32 num_ports = OPA_AM_NPORT(am);
810 u32 start_of_sm_config = OPA_AM_START_SM_CFG(am);
811 u32 buffer_units;
812 u64 tmp = 0;
813
f1685179 814 if (num_ports != 1 || smp_length_check(sizeof(*pi), max_len)) {
77241056
MM		 815 smp->status |= IB_SMP_INVALID_FIELD;
816 return reply((struct ib_mad_hdr *)smp);
817 }
818
819 dd = dd_from_ibdev(ibdev);
820 /* IB numbers ports from 1, hw from 0 */
821 ppd = dd->pport + (port - 1);
822 ibp = &ppd->ibport_data;
823
8638b77f 824 if (ppd->vls_supported / 2 > ARRAY_SIZE(pi->neigh_mtu.pvlx_to_mtu) ||
17fb4f29 825 ppd->vls_supported > ARRAY_SIZE(dd->vld)) {
77241056
MM		 826 smp->status |= IB_SMP_INVALID_FIELD;
827 return reply((struct ib_mad_hdr *)smp);
828 }
829
830 pi->lid = cpu_to_be32(ppd->lid);
831
832 /* Only return the mkey if the protection field allows it. */
833 if (!(smp->method == IB_MGMT_METHOD_GET &&
4eb06882
DD		 834 ibp->rvp.mkey != smp->mkey &&
835 ibp->rvp.mkeyprot == 1))
836 pi->mkey = ibp->rvp.mkey;
837
838 pi->subnet_prefix = ibp->rvp.gid_prefix;
839 pi->sm_lid = cpu_to_be32(ibp->rvp.sm_lid);
840 pi->ib_cap_mask = cpu_to_be32(ibp->rvp.port_cap_flags);
841 pi->mkey_lease_period = cpu_to_be16(ibp->rvp.mkey_lease_period);
77241056
MM		 842 pi->sm_trap_qp = cpu_to_be32(ppd->sm_trap_qp);
843 pi->sa_qp = cpu_to_be32(ppd->sa_qp);
844
845 pi->link_width.enabled = cpu_to_be16(ppd->link_width_enabled);
846 pi->link_width.supported = cpu_to_be16(ppd->link_width_supported);
847 pi->link_width.active = cpu_to_be16(ppd->link_width_active);
848
849 pi->link_width_downgrade.supported =
850 cpu_to_be16(ppd->link_width_downgrade_supported);
851 pi->link_width_downgrade.enabled =
852 cpu_to_be16(ppd->link_width_downgrade_enabled);
853 pi->link_width_downgrade.tx_active =
854 cpu_to_be16(ppd->link_width_downgrade_tx_active);
855 pi->link_width_downgrade.rx_active =
856 cpu_to_be16(ppd->link_width_downgrade_rx_active);
857
858 pi->link_speed.supported = cpu_to_be16(ppd->link_speed_supported);
859 pi->link_speed.active = cpu_to_be16(ppd->link_speed_active);
860 pi->link_speed.enabled = cpu_to_be16(ppd->link_speed_enabled);
861
862 state = driver_lstate(ppd);
863
864 if (start_of_sm_config && (state == IB_PORT_INIT))
865 ppd->is_sm_config_started = 1;
866
1d01cf33 867 pi->port_phys_conf = (ppd->port_type & 0xf);
77241056 868
77241056
MM		 869 pi->port_states.ledenable_offlinereason = ppd->neighbor_normal << 4;
870 pi->port_states.ledenable_offlinereason |=
871 ppd->is_sm_config_started << 5;
409b1462 872 /*
2243472e
EH		 873 * This pairs with the memory barrier in hfi1_start_led_override to
874 * ensure that we read the correct state of LED beaconing represented
875 * by led_override_timer_active
409b1462 876 */
2243472e 877 smp_rmb();
409b1462
EH		 878 is_beaconing_active = !!atomic_read(&ppd->led_override_timer_active);
879 pi->port_states.ledenable_offlinereason |= is_beaconing_active << 6;
77241056 880 pi->port_states.ledenable_offlinereason |=
a9c05e35 881 ppd->offline_disabled_reason;
77241056
MM		 882
883 pi->port_states.portphysstate_portstate =
bec7c79c 884 (driver_pstate(ppd) << 4) | state;
77241056 885
4eb06882 886 pi->mkeyprotect_lmc = (ibp->rvp.mkeyprot << 6) | ppd->lmc;
77241056
MM		 887
888 memset(pi->neigh_mtu.pvlx_to_mtu, 0, sizeof(pi->neigh_mtu.pvlx_to_mtu));
889 for (i = 0; i < ppd->vls_supported; i++) {
890 mtu = mtu_to_enum(dd->vld[i].mtu, HFI1_DEFAULT_ACTIVE_MTU);
891 if ((i % 2) == 0)
8638b77f 892 pi->neigh_mtu.pvlx_to_mtu[i / 2] |= (mtu << 4);
77241056 893 else
8638b77f 894 pi->neigh_mtu.pvlx_to_mtu[i / 2] |= mtu;
77241056
MM		 895 }
896 /* don't forget VL 15 */
897 mtu = mtu_to_enum(dd->vld[15].mtu, 2048);
8638b77f 898 pi->neigh_mtu.pvlx_to_mtu[15 / 2] |= mtu;
4eb06882 899 pi->smsl = ibp->rvp.sm_sl & OPA_PI_MASK_SMSL;
77241056
MM		 900 pi->operational_vls = hfi1_get_ib_cfg(ppd, HFI1_IB_CFG_OP_VLS);
901 pi->partenforce_filterraw |=
902 (ppd->linkinit_reason & OPA_PI_MASK_LINKINIT_REASON);
903 if (ppd->part_enforce & HFI1_PART_ENFORCE_IN)
904 pi->partenforce_filterraw |= OPA_PI_MASK_PARTITION_ENFORCE_IN;
905 if (ppd->part_enforce & HFI1_PART_ENFORCE_OUT)
906 pi->partenforce_filterraw |= OPA_PI_MASK_PARTITION_ENFORCE_OUT;
4eb06882 907 pi->mkey_violations = cpu_to_be16(ibp->rvp.mkey_violations);
77241056 908 /* P_KeyViolations are counted by hardware. */
4eb06882
DD		 909 pi->pkey_violations = cpu_to_be16(ibp->rvp.pkey_violations);
910 pi->qkey_violations = cpu_to_be16(ibp->rvp.qkey_violations);
77241056
MM		 911
912 pi->vl.cap = ppd->vls_supported;
4eb06882 913 pi->vl.high_limit = cpu_to_be16(ibp->rvp.vl_high_limit);
77241056
MM		 914 pi->vl.arb_high_cap = (u8)hfi1_get_ib_cfg(ppd, HFI1_IB_CFG_VL_HIGH_CAP);
915 pi->vl.arb_low_cap = (u8)hfi1_get_ib_cfg(ppd, HFI1_IB_CFG_VL_LOW_CAP);
916
4eb06882 917 pi->clientrereg_subnettimeout = ibp->rvp.subnet_timeout;
77241056
MM		 918
919 pi->port_link_mode = cpu_to_be16(OPA_PORT_LINK_MODE_OPA << 10 |
920 OPA_PORT_LINK_MODE_OPA << 5 |
921 OPA_PORT_LINK_MODE_OPA);
922
923 pi->port_ltp_crc_mode = cpu_to_be16(ppd->port_ltp_crc_mode);
924
925 pi->port_mode = cpu_to_be16(
926 ppd->is_active_optimize_enabled ?
927 OPA_PI_MASK_PORT_ACTIVE_OPTOMIZE : 0);
928
929 pi->port_packet_format.supported =
2280740f
VN		 930 cpu_to_be16(OPA_PORT_PACKET_FORMAT_9B |
931 OPA_PORT_PACKET_FORMAT_16B);
77241056 932 pi->port_packet_format.enabled =
2280740f
VN		 933 cpu_to_be16(OPA_PORT_PACKET_FORMAT_9B |
934 OPA_PORT_PACKET_FORMAT_16B);
77241056
MM		 935
936 /* flit_control.interleave is (OPA V1, version .76):
937 * bits use
938 * ---- ---
939 * 2 res
940 * 2 DistanceSupported
941 * 2 DistanceEnabled
942 * 5 MaxNextLevelTxEnabled
943 * 5 MaxNestLevelRxSupported
944 *
945 * HFI supports only "distance mode 1" (see OPA V1, version .76,
946 * section 9.6.2), so set DistanceSupported, DistanceEnabled
947 * to 0x1.
948 */
949 pi->flit_control.interleave = cpu_to_be16(0x1400);
950
951 pi->link_down_reason = ppd->local_link_down_reason.sma;
952 pi->neigh_link_down_reason = ppd->neigh_link_down_reason.sma;
953 pi->port_error_action = cpu_to_be32(ppd->port_error_action);
954 pi->mtucap = mtu_to_enum(hfi1_max_mtu, IB_MTU_4096);
955
956 /* 32.768 usec. response time (guessing) */
957 pi->resptimevalue = 3;
958
959 pi->local_port_num = port;
960
961 /* buffer info for FM */
962 pi->overall_buffer_space = cpu_to_be16(dd->link_credits);
963
964 pi->neigh_node_guid = cpu_to_be64(ppd->neighbor_guid);
965 pi->neigh_port_num = ppd->neighbor_port_number;
966 pi->port_neigh_mode =
967 (ppd->neighbor_type & OPA_PI_MASK_NEIGH_NODE_TYPE) |
968 (ppd->mgmt_allowed ? OPA_PI_MASK_NEIGH_MGMT_ALLOWED : 0) |
969 (ppd->neighbor_fm_security ?
970 OPA_PI_MASK_NEIGH_FW_AUTH_BYPASS : 0);
971
972 /* HFIs shall always return VL15 credits to their
973 * neighbor in a timely manner, without any credit return pacing.
974 */
975 credit_rate = 0;
976 buffer_units = (dd->vau) & OPA_PI_MASK_BUF_UNIT_BUF_ALLOC;
977 buffer_units |= (dd->vcu << 3) & OPA_PI_MASK_BUF_UNIT_CREDIT_ACK;
978 buffer_units |= (credit_rate << 6) &
979 OPA_PI_MASK_BUF_UNIT_VL15_CREDIT_RATE;
980 buffer_units |= (dd->vl15_init << 11) & OPA_PI_MASK_BUF_UNIT_VL15_INIT;
981 pi->buffer_units = cpu_to_be32(buffer_units);
982
cb49366f 983 pi->opa_cap_mask = cpu_to_be16(ibp->rvp.port_cap3_flags);
13c19222
DH		 984 pi->collectivemask_multicastmask = ((OPA_COLLECTIVE_NR & 0x7)
985 << 3 | (OPA_MCAST_NR & 0x7));
77241056
MM		 986
987 /* HFI supports a replay buffer 128 LTPs in size */
988 pi->replay_depth.buffer = 0x80;
989 /* read the cached value of DC_LCB_STS_ROUND_TRIP_LTP_CNT */
990 read_lcb_cache(DC_LCB_STS_ROUND_TRIP_LTP_CNT, &tmp);
991
4d114fdd
JJ		 992 /*
993 * this counter is 16 bits wide, but the replay_depth.wire
994 * variable is only 8 bits
995 */
77241056
MM		 996 if (tmp > 0xff)
997 tmp = 0xff;
998 pi->replay_depth.wire = tmp;
999
1000 if (resp_len)
1001 *resp_len += sizeof(struct opa_port_info);
1002
1003 return reply((struct ib_mad_hdr *)smp);
1004}
1005
1006/**
1007 * get_pkeys - return the PKEY table
1008 * @dd: the hfi1_ib device
1009 * @port: the IB port number
1010 * @pkeys: the pkey table is placed here
1011 */
1012static int get_pkeys(struct hfi1_devdata *dd, u8 port, u16 *pkeys)
1013{
1014 struct hfi1_pportdata *ppd = dd->pport + port - 1;
1015
1016 memcpy(pkeys, ppd->pkeys, sizeof(ppd->pkeys));
1017
1018 return 0;
1019}
1020
1021static int __subn_get_opa_pkeytable(struct opa_smp *smp, u32 am, u8 *data,
1022 struct ib_device *ibdev, u8 port,
f1685179 1023 u32 *resp_len, u32 max_len)
77241056
MM		 1024{
1025 struct hfi1_devdata *dd = dd_from_ibdev(ibdev);
1026 u32 n_blocks_req = OPA_AM_NBLK(am);
1027 u32 start_block = am & 0x7ff;
1028 __be16 *p;
1029 u16 *q;
1030 int i;
1031 u16 n_blocks_avail;
1032 unsigned npkeys = hfi1_get_npkeys(dd);
1033 size_t size;
1034
1035 if (n_blocks_req == 0) {
1036 pr_warn("OPA Get PKey AM Invalid : P = %d; B = 0x%x; N = 0x%x\n",
1037 port, start_block, n_blocks_req);
1038 smp->status |= IB_SMP_INVALID_FIELD;
1039 return reply((struct ib_mad_hdr *)smp);
1040 }
1041
50e5dcbe 1042 n_blocks_avail = (u16)(npkeys / OPA_PARTITION_TABLE_BLK_SIZE) + 1;
77241056
MM		 1043
1044 size = (n_blocks_req * OPA_PARTITION_TABLE_BLK_SIZE) * sizeof(u16);
1045
f1685179
ND		 1046 if (smp_length_check(size, max_len)) {
1047 smp->status |= IB_SMP_INVALID_FIELD;
1048 return reply((struct ib_mad_hdr *)smp);
1049 }
1050
77241056
MM		 1051 if (start_block + n_blocks_req > n_blocks_avail ||
1052 n_blocks_req > OPA_NUM_PKEY_BLOCKS_PER_SMP) {
1053 pr_warn("OPA Get PKey AM Invalid : s 0x%x; req 0x%x; "
1054 "avail 0x%x; blk/smp 0x%lx\n",
1055 start_block, n_blocks_req, n_blocks_avail,
1056 OPA_NUM_PKEY_BLOCKS_PER_SMP);
1057 smp->status |= IB_SMP_INVALID_FIELD;
1058 return reply((struct ib_mad_hdr *)smp);
1059 }
1060
50e5dcbe 1061 p = (__be16 *)data;
77241056
MM		 1062 q = (u16 *)data;
1063 /* get the real pkeys if we are requesting the first block */
1064 if (start_block == 0) {
1065 get_pkeys(dd, port, q);
1066 for (i = 0; i < npkeys; i++)
1067 p[i] = cpu_to_be16(q[i]);
1068 if (resp_len)
1069 *resp_len += size;
e490974e 1070 } else {
77241056 1071 smp->status |= IB_SMP_INVALID_FIELD;
e490974e 1072 }
77241056
MM		 1073 return reply((struct ib_mad_hdr *)smp);
1074}
1075
1076enum {
1077 HFI_TRANSITION_DISALLOWED,
1078 HFI_TRANSITION_IGNORED,
1079 HFI_TRANSITION_ALLOWED,
1080 HFI_TRANSITION_UNDEFINED,
1081};
1082
1083/*
1084 * Use shortened names to improve readability of
1085 * {logical,physical}_state_transitions
1086 */
1087enum {
1088 __D = HFI_TRANSITION_DISALLOWED,
1089 __I = HFI_TRANSITION_IGNORED,
1090 __A = HFI_TRANSITION_ALLOWED,
1091 __U = HFI_TRANSITION_UNDEFINED,
1092};
1093
1094/*
1095 * IB_PORTPHYSSTATE_POLLING (2) through OPA_PORTPHYSSTATE_MAX (11) are
1096 * represented in physical_state_transitions.
1097 */
1098#define __N_PHYSTATES (OPA_PORTPHYSSTATE_MAX - IB_PORTPHYSSTATE_POLLING + 1)
1099
1100/*
1101 * Within physical_state_transitions, rows represent "old" states,
1102 * columns "new" states, and physical_state_transitions.allowed[old][new]
1103 * indicates if the transition from old state to new state is legal (see
1104 * OPAg1v1, Table 6-4).
1105 */
1106static const struct {
1107 u8 allowed[__N_PHYSTATES][__N_PHYSTATES];
1108} physical_state_transitions = {
1109 {
1110 /* 2 3 4 5 6 7 8 9 10 11 */
1111 /* 2 */ { __A, __A, __D, __D, __D, __D, __D, __D, __D, __D },
1112 /* 3 */ { __A, __I, __D, __D, __D, __D, __D, __D, __D, __A },
1113 /* 4 */ { __U, __U, __U, __U, __U, __U, __U, __U, __U, __U },
1114 /* 5 */ { __A, __A, __D, __I, __D, __D, __D, __D, __D, __D },
1115 /* 6 */ { __U, __U, __U, __U, __U, __U, __U, __U, __U, __U },
1116 /* 7 */ { __D, __A, __D, __D, __D, __I, __D, __D, __D, __D },
1117 /* 8 */ { __U, __U, __U, __U, __U, __U, __U, __U, __U, __U },
1118 /* 9 */ { __I, __A, __D, __D, __D, __D, __D, __I, __D, __D },
1119 /*10 */ { __U, __U, __U, __U, __U, __U, __U, __U, __U, __U },
1120 /*11 */ { __D, __A, __D, __D, __D, __D, __D, __D, __D, __I },
1121 }
1122};
1123
1124/*
1125 * IB_PORT_DOWN (1) through IB_PORT_ACTIVE_DEFER (5) are represented
1126 * logical_state_transitions
1127 */
1128
1129#define __N_LOGICAL_STATES (IB_PORT_ACTIVE_DEFER - IB_PORT_DOWN + 1)
1130
1131/*
1132 * Within logical_state_transitions rows represent "old" states,
1133 * columns "new" states, and logical_state_transitions.allowed[old][new]
1134 * indicates if the transition from old state to new state is legal (see
1135 * OPAg1v1, Table 9-12).
1136 */
1137static const struct {
1138 u8 allowed[__N_LOGICAL_STATES][__N_LOGICAL_STATES];
1139} logical_state_transitions = {
1140 {
1141 /* 1 2 3 4 5 */
1142 /* 1 */ { __I, __D, __D, __D, __U},
1143 /* 2 */ { __D, __I, __A, __D, __U},
1144 /* 3 */ { __D, __D, __I, __A, __U},
1145 /* 4 */ { __D, __D, __I, __I, __U},
1146 /* 5 */ { __U, __U, __U, __U, __U},
1147 }
1148};
1149
1150static int logical_transition_allowed(int old, int new)
1151{
1152 if (old < IB_PORT_NOP || old > IB_PORT_ACTIVE_DEFER ||
1153 new < IB_PORT_NOP || new > IB_PORT_ACTIVE_DEFER) {
1154 pr_warn("invalid logical state(s) (old %d new %d)\n",
1155 old, new);
1156 return HFI_TRANSITION_UNDEFINED;
1157 }
1158
1159 if (new == IB_PORT_NOP)
1160 return HFI_TRANSITION_ALLOWED; /* always allowed */
1161
1162 /* adjust states for indexing into logical_state_transitions */
1163 old -= IB_PORT_DOWN;
1164 new -= IB_PORT_DOWN;
1165
1166 if (old < 0 || new < 0)
1167 return HFI_TRANSITION_UNDEFINED;
1168 return logical_state_transitions.allowed[old][new];
1169}
1170
1171static int physical_transition_allowed(int old, int new)
1172{
1173 if (old < IB_PORTPHYSSTATE_NOP || old > OPA_PORTPHYSSTATE_MAX ||
1174 new < IB_PORTPHYSSTATE_NOP || new > OPA_PORTPHYSSTATE_MAX) {
1175 pr_warn("invalid physical state(s) (old %d new %d)\n",
1176 old, new);
1177 return HFI_TRANSITION_UNDEFINED;
1178 }
1179
1180 if (new == IB_PORTPHYSSTATE_NOP)
1181 return HFI_TRANSITION_ALLOWED; /* always allowed */
1182
1183 /* adjust states for indexing into physical_state_transitions */
1184 old -= IB_PORTPHYSSTATE_POLLING;
1185 new -= IB_PORTPHYSSTATE_POLLING;
1186
1187 if (old < 0 || new < 0)
1188 return HFI_TRANSITION_UNDEFINED;
1189 return physical_state_transitions.allowed[old][new];
1190}
1191
1192static int port_states_transition_allowed(struct hfi1_pportdata *ppd,
1193 u32 logical_new, u32 physical_new)
1194{
d392a673 1195 u32 physical_old = driver_pstate(ppd);
02a222c7 1196 u32 logical_old = driver_lstate(ppd);
77241056
MM		 1197 int ret, logical_allowed, physical_allowed;
1198
f3ff8189
JJ		 1199 ret = logical_transition_allowed(logical_old, logical_new);
1200 logical_allowed = ret;
77241056
MM		 1201
1202 if (ret == HFI_TRANSITION_DISALLOWED ||
1203 ret == HFI_TRANSITION_UNDEFINED) {
1204 pr_warn("invalid logical state transition %s -> %s\n",
1205 opa_lstate_name(logical_old),
1206 opa_lstate_name(logical_new));
1207 return ret;
1208 }
1209
f3ff8189
JJ		 1210 ret = physical_transition_allowed(physical_old, physical_new);
1211 physical_allowed = ret;
77241056
MM		 1212
1213 if (ret == HFI_TRANSITION_DISALLOWED ||
1214 ret == HFI_TRANSITION_UNDEFINED) {
1215 pr_warn("invalid physical state transition %s -> %s\n",
1216 opa_pstate_name(physical_old),
1217 opa_pstate_name(physical_new));
1218 return ret;
1219 }
1220
1221 if (logical_allowed == HFI_TRANSITION_IGNORED &&
1222 physical_allowed == HFI_TRANSITION_IGNORED)
1223 return HFI_TRANSITION_IGNORED;
1224
a9c05e35
BM		 1225 /*
1226 * A change request of Physical Port State from
1227 * 'Offline' to 'Polling' should be ignored.
1228 */
1229 if ((physical_old == OPA_PORTPHYSSTATE_OFFLINE) &&
1230 (physical_new == IB_PORTPHYSSTATE_POLLING))
1231 return HFI_TRANSITION_IGNORED;
1232
77241056
MM		 1233 /*
1234 * Either physical_allowed or logical_allowed is
1235 * HFI_TRANSITION_ALLOWED.
1236 */
1237 return HFI_TRANSITION_ALLOWED;
1238}
1239
1240static int set_port_states(struct hfi1_pportdata *ppd, struct opa_smp *smp,
1241 u32 logical_state, u32 phys_state,
1242 int suppress_idle_sma)
1243{
1244 struct hfi1_devdata *dd = ppd->dd;
1245 u32 link_state;
1246 int ret;
1247
1248 ret = port_states_transition_allowed(ppd, logical_state, phys_state);
1249 if (ret == HFI_TRANSITION_DISALLOWED ||
1250 ret == HFI_TRANSITION_UNDEFINED) {
1251 /* error message emitted above */
1252 smp->status |= IB_SMP_INVALID_FIELD;
1253 return 0;
1254 }
1255
1256 if (ret == HFI_TRANSITION_IGNORED)
1257 return 0;
1258
1259 if ((phys_state != IB_PORTPHYSSTATE_NOP) &&
1260 !(logical_state == IB_PORT_DOWN ||
1261 logical_state == IB_PORT_NOP)){
1262 pr_warn("SubnSet(OPA_PortInfo) port state invalid: logical_state 0x%x physical_state 0x%x\n",
1263 logical_state, phys_state);
1264 smp->status |= IB_SMP_INVALID_FIELD;
1265 }
1266
1267 /*
1268 * Logical state changes are summarized in OPAv1g1 spec.,
1269 * Table 9-12; physical state changes are summarized in
1270 * OPAv1g1 spec., Table 6.4.
1271 */
1272 switch (logical_state) {
1273 case IB_PORT_NOP:
1274 if (phys_state == IB_PORTPHYSSTATE_NOP)
1275 break;
1276 /* FALLTHROUGH */
1277 case IB_PORT_DOWN:
e490974e 1278 if (phys_state == IB_PORTPHYSSTATE_NOP) {
77241056 1279 link_state = HLS_DN_DOWNDEF;
e490974e 1280 } else if (phys_state == IB_PORTPHYSSTATE_POLLING) {
77241056 1281 link_state = HLS_DN_POLL;
17fb4f29
JJ		 1282 set_link_down_reason(ppd, OPA_LINKDOWN_REASON_FM_BOUNCE,
1283 0, OPA_LINKDOWN_REASON_FM_BOUNCE);
e490974e 1284 } else if (phys_state == IB_PORTPHYSSTATE_DISABLED) {
77241056 1285 link_state = HLS_DN_DISABLE;
e490974e 1286 } else {
77241056
MM		 1287 pr_warn("SubnSet(OPA_PortInfo) invalid physical state 0x%x\n",
1288 phys_state);
1289 smp->status |= IB_SMP_INVALID_FIELD;
1290 break;
1291 }
1292
1cbaa670
DL		 1293 if ((link_state == HLS_DN_POLL ||
1294 link_state == HLS_DN_DOWNDEF)) {
1295 /*
1296 * Going to poll. No matter what the current state,
1297 * always move offline first, then tune and start the
1298 * link. This correctly handles a FM link bounce and
1299 * a link enable. Going offline is a no-op if already
1300 * offline.
1301 */
1302 set_link_state(ppd, HLS_DN_OFFLINE);
1cbaa670
DL		 1303 start_link(ppd);
1304 } else {
1305 set_link_state(ppd, link_state);
1306 }
77241056
MM		 1307 if (link_state == HLS_DN_DISABLE &&
1308 (ppd->offline_disabled_reason >
a9c05e35 1309 HFI1_ODR_MASK(OPA_LINKDOWN_REASON_SMA_DISABLED) ||
77241056 1310 ppd->offline_disabled_reason ==
a9c05e35 1311 HFI1_ODR_MASK(OPA_LINKDOWN_REASON_NONE)))
77241056 1312 ppd->offline_disabled_reason =
a9c05e35 1313 HFI1_ODR_MASK(OPA_LINKDOWN_REASON_SMA_DISABLED);
77241056
MM		 1314 /*
1315 * Don't send a reply if the response would be sent
1316 * through the disabled port.
1317 */
1318 if (link_state == HLS_DN_DISABLE && smp->hop_cnt)
1319 return IB_MAD_RESULT_SUCCESS | IB_MAD_RESULT_CONSUMED;
1320 break;
1321 case IB_PORT_ARMED:
1322 ret = set_link_state(ppd, HLS_UP_ARMED);
1323 if ((ret == 0) && (suppress_idle_sma == 0))
1324 send_idle_sma(dd, SMA_IDLE_ARM);
1325 break;
1326 case IB_PORT_ACTIVE:
1327 if (ppd->neighbor_normal) {
1328 ret = set_link_state(ppd, HLS_UP_ACTIVE);
1329 if (ret == 0)
1330 send_idle_sma(dd, SMA_IDLE_ACTIVE);
1331 } else {
1332 pr_warn("SubnSet(OPA_PortInfo) Cannot move to Active with NeighborNormal 0\n");
1333 smp->status |= IB_SMP_INVALID_FIELD;
1334 }
1335 break;
1336 default:
1337 pr_warn("SubnSet(OPA_PortInfo) invalid logical state 0x%x\n",
1338 logical_state);
1339 smp->status |= IB_SMP_INVALID_FIELD;
1340 }
1341
1342 return 0;
1343}
1344
1345/**
1346 * subn_set_opa_portinfo - set port information
1347 * @smp: the incoming SM packet
1348 * @ibdev: the infiniband device
1349 * @port: the port on the device
1350 *
1351 */
1352static int __subn_set_opa_portinfo(struct opa_smp *smp, u32 am, u8 *data,
1353 struct ib_device *ibdev, u8 port,
f1685179 1354 u32 *resp_len, u32 max_len)
77241056
MM		 1355{
1356 struct opa_port_info *pi = (struct opa_port_info *)data;
1357 struct ib_event event;
1358 struct hfi1_devdata *dd;
1359 struct hfi1_pportdata *ppd;
1360 struct hfi1_ibport *ibp;
1361 u8 clientrereg;
1362 unsigned long flags;
51e658f5
DC		 1363 u32 smlid;
1364 u32 lid;
77241056
MM		 1365 u8 ls_old, ls_new, ps_new;
1366 u8 vls;
1367 u8 msl;
1368 u8 crc_enabled;
1369 u16 lse, lwe, mtu;
1370 u32 num_ports = OPA_AM_NPORT(am);
1371 u32 start_of_sm_config = OPA_AM_START_SM_CFG(am);
1372 int ret, i, invalid = 0, call_set_mtu = 0;
1373 int call_link_downgrade_policy = 0;
1374
f1685179
ND		 1375 if (num_ports != 1 ||
1376 smp_length_check(sizeof(*pi), max_len)) {
77241056
MM		 1377 smp->status |= IB_SMP_INVALID_FIELD;
1378 return reply((struct ib_mad_hdr *)smp);
1379 }
1380
51e658f5
DC		 1381 lid = be32_to_cpu(pi->lid);
1382 if (lid & 0xFF000000) {
1383 pr_warn("OPA_PortInfo lid out of range: %X\n", lid);
77241056
MM		 1384 smp->status |= IB_SMP_INVALID_FIELD;
1385 goto get_only;
1386 }
1387
77241056
MM		 1388
1389 smlid = be32_to_cpu(pi->sm_lid);
51e658f5 1390 if (smlid & 0xFF000000) {
77241056
MM		 1391 pr_warn("OPA_PortInfo SM lid out of range: %X\n", smlid);
1392 smp->status |= IB_SMP_INVALID_FIELD;
1393 goto get_only;
1394 }
77241056
MM		 1395
1396 clientrereg = (pi->clientrereg_subnettimeout &
1397 OPA_PI_MASK_CLIENT_REREGISTER);
1398
1399 dd = dd_from_ibdev(ibdev);
1400 /* IB numbers ports from 1, hw from 0 */
1401 ppd = dd->pport + (port - 1);
1402 ibp = &ppd->ibport_data;
1403 event.device = ibdev;
1404 event.element.port_num = port;
1405
1406 ls_old = driver_lstate(ppd);
1407
4eb06882 1408 ibp->rvp.mkey = pi->mkey;
51e658f5
DC		 1409 if (ibp->rvp.gid_prefix != pi->subnet_prefix) {
1410 ibp->rvp.gid_prefix = pi->subnet_prefix;
1411 event.event = IB_EVENT_GID_CHANGE;
1412 ib_dispatch_event(&event);
1413 }
4eb06882 1414 ibp->rvp.mkey_lease_period = be16_to_cpu(pi->mkey_lease_period);
77241056
MM		 1415
1416 /* Must be a valid unicast LID address. */
1417 if ((lid == 0 && ls_old > IB_PORT_INIT) ||
51e658f5 1418 (hfi1_is_16B_mcast(lid))) {
77241056
MM		 1419 smp->status |= IB_SMP_INVALID_FIELD;
1420 pr_warn("SubnSet(OPA_PortInfo) lid invalid 0x%x\n",
1421 lid);
1422 } else if (ppd->lid != lid ||
1423 ppd->lmc != (pi->mkeyprotect_lmc & OPA_PI_MASK_LMC)) {
1424 if (ppd->lid != lid)
1425 hfi1_set_uevent_bits(ppd, _HFI1_EVENT_LID_CHANGE_BIT);
1426 if (ppd->lmc != (pi->mkeyprotect_lmc & OPA_PI_MASK_LMC))
1427 hfi1_set_uevent_bits(ppd, _HFI1_EVENT_LMC_CHANGE_BIT);
1428 hfi1_set_lid(ppd, lid, pi->mkeyprotect_lmc & OPA_PI_MASK_LMC);
1429 event.event = IB_EVENT_LID_CHANGE;
1430 ib_dispatch_event(&event);
51e658f5
DC		 1431
1432 if (HFI1_PORT_GUID_INDEX + 1 < HFI1_GUIDS_PER_PORT) {
1433 /* Manufacture GID from LID to support extended
1434 * addresses
1435 */
1436 ppd->guids[HFI1_PORT_GUID_INDEX + 1] =
1437 be64_to_cpu(OPA_MAKE_ID(lid));
1438 event.event = IB_EVENT_GID_CHANGE;
1439 ib_dispatch_event(&event);
1440 }
77241056
MM		 1441 }
1442
1443 msl = pi->smsl & OPA_PI_MASK_SMSL;
1444 if (pi->partenforce_filterraw & OPA_PI_MASK_LINKINIT_REASON)
1445 ppd->linkinit_reason =
1446 (pi->partenforce_filterraw &
1447 OPA_PI_MASK_LINKINIT_REASON);
77241056
MM		 1448
1449 /* Must be a valid unicast LID address. */
1450 if ((smlid == 0 && ls_old > IB_PORT_INIT) ||
51e658f5 1451 (hfi1_is_16B_mcast(smlid))) {
77241056
MM		 1452 smp->status |= IB_SMP_INVALID_FIELD;
1453 pr_warn("SubnSet(OPA_PortInfo) smlid invalid 0x%x\n", smlid);
4eb06882 1454 } else if (smlid != ibp->rvp.sm_lid || msl != ibp->rvp.sm_sl) {
77241056 1455 pr_warn("SubnSet(OPA_PortInfo) smlid 0x%x\n", smlid);
4eb06882 1456 spin_lock_irqsave(&ibp->rvp.lock, flags);
9c4a311e 1457 if (ibp->rvp.sm_ah) {
4eb06882 1458 if (smlid != ibp->rvp.sm_lid)
51e658f5 1459 hfi1_modify_qp0_ah(ibp, ibp->rvp.sm_ah, smlid);
4eb06882 1460 if (msl != ibp->rvp.sm_sl)
d8966fcd 1461 rdma_ah_set_sl(&ibp->rvp.sm_ah->attr, msl);
77241056 1462 }
4eb06882
DD		 1463 spin_unlock_irqrestore(&ibp->rvp.lock, flags);
1464 if (smlid != ibp->rvp.sm_lid)
1465 ibp->rvp.sm_lid = smlid;
1466 if (msl != ibp->rvp.sm_sl)
1467 ibp->rvp.sm_sl = msl;
77241056
MM		 1468 event.event = IB_EVENT_SM_CHANGE;
1469 ib_dispatch_event(&event);
1470 }
1471
1472 if (pi->link_down_reason == 0) {
1473 ppd->local_link_down_reason.sma = 0;
1474 ppd->local_link_down_reason.latest = 0;
1475 }
1476
1477 if (pi->neigh_link_down_reason == 0) {
1478 ppd->neigh_link_down_reason.sma = 0;
1479 ppd->neigh_link_down_reason.latest = 0;
1480 }
1481
1482 ppd->sm_trap_qp = be32_to_cpu(pi->sm_trap_qp);
1483 ppd->sa_qp = be32_to_cpu(pi->sa_qp);
1484
1485 ppd->port_error_action = be32_to_cpu(pi->port_error_action);
1486 lwe = be16_to_cpu(pi->link_width.enabled);
1487 if (lwe) {
d0d236ea
JJ		 1488 if (lwe == OPA_LINK_WIDTH_RESET ||
1489 lwe == OPA_LINK_WIDTH_RESET_OLD)
77241056
MM		 1490 set_link_width_enabled(ppd, ppd->link_width_supported);
1491 else if ((lwe & ~ppd->link_width_supported) == 0)
1492 set_link_width_enabled(ppd, lwe);
1493 else
1494 smp->status |= IB_SMP_INVALID_FIELD;
1495 }
1496 lwe = be16_to_cpu(pi->link_width_downgrade.enabled);
1497 /* LWD.E is always applied - 0 means "disabled" */
d0d236ea
JJ		 1498 if (lwe == OPA_LINK_WIDTH_RESET ||
1499 lwe == OPA_LINK_WIDTH_RESET_OLD) {
77241056 1500 set_link_width_downgrade_enabled(ppd,
17fb4f29
JJ		 1501 ppd->
1502 link_width_downgrade_supported
1503 );
77241056
MM		 1504 } else if ((lwe & ~ppd->link_width_downgrade_supported) == 0) {
1505 /* only set and apply if something changed */
1506 if (lwe != ppd->link_width_downgrade_enabled) {
1507 set_link_width_downgrade_enabled(ppd, lwe);
1508 call_link_downgrade_policy = 1;
1509 }
e490974e 1510 } else {
77241056 1511 smp->status |= IB_SMP_INVALID_FIELD;
e490974e 1512 }
77241056
MM		 1513 lse = be16_to_cpu(pi->link_speed.enabled);
1514 if (lse) {
1515 if (lse & be16_to_cpu(pi->link_speed.supported))
1516 set_link_speed_enabled(ppd, lse);
1517 else
1518 smp->status |= IB_SMP_INVALID_FIELD;
1519 }
1520
4eb06882
DD		 1521 ibp->rvp.mkeyprot =
1522 (pi->mkeyprotect_lmc & OPA_PI_MASK_MKEY_PROT_BIT) >> 6;
1523 ibp->rvp.vl_high_limit = be16_to_cpu(pi->vl.high_limit) & 0xFF;
77241056 1524 (void)hfi1_set_ib_cfg(ppd, HFI1_IB_CFG_VL_HIGH_LIMIT,
4eb06882 1525 ibp->rvp.vl_high_limit);
77241056 1526
8638b77f 1527 if (ppd->vls_supported / 2 > ARRAY_SIZE(pi->neigh_mtu.pvlx_to_mtu) ||
17fb4f29 1528 ppd->vls_supported > ARRAY_SIZE(dd->vld)) {
77241056
MM		 1529 smp->status |= IB_SMP_INVALID_FIELD;
1530 return reply((struct ib_mad_hdr *)smp);
1531 }
1532 for (i = 0; i < ppd->vls_supported; i++) {
1533 if ((i % 2) == 0)
17fb4f29
JJ		 1534 mtu = enum_to_mtu((pi->neigh_mtu.pvlx_to_mtu[i / 2] >>
1535 4) & 0xF);
77241056 1536 else
17fb4f29
JJ		 1537 mtu = enum_to_mtu(pi->neigh_mtu.pvlx_to_mtu[i / 2] &
1538 0xF);
77241056
MM		 1539 if (mtu == 0xffff) {
1540 pr_warn("SubnSet(OPA_PortInfo) mtu invalid %d (0x%x)\n",
1541 mtu,
1542 (pi->neigh_mtu.pvlx_to_mtu[0] >> 4) & 0xF);
1543 smp->status |= IB_SMP_INVALID_FIELD;
1544 mtu = hfi1_max_mtu; /* use a valid MTU */
1545 }
1546 if (dd->vld[i].mtu != mtu) {
1547 dd_dev_info(dd,
17fb4f29
JJ		 1548 "MTU change on vl %d from %d to %d\n",
1549 i, dd->vld[i].mtu, mtu);
77241056
MM		 1550 dd->vld[i].mtu = mtu;
1551 call_set_mtu++;
1552 }
1553 }
1554 /* As per OPAV1 spec: VL15 must support and be configured
1555 * for operation with a 2048 or larger MTU.
1556 */
8638b77f 1557 mtu = enum_to_mtu(pi->neigh_mtu.pvlx_to_mtu[15 / 2] & 0xF);
77241056
MM		 1558 if (mtu < 2048 || mtu == 0xffff)
1559 mtu = 2048;
1560 if (dd->vld[15].mtu != mtu) {
1561 dd_dev_info(dd,
17fb4f29
JJ		 1562 "MTU change on vl 15 from %d to %d\n",
1563 dd->vld[15].mtu, mtu);
77241056
MM		 1564 dd->vld[15].mtu = mtu;
1565 call_set_mtu++;
1566 }
1567 if (call_set_mtu)
1568 set_mtu(ppd);
1569
1570 /* Set operational VLs */
1571 vls = pi->operational_vls & OPA_PI_MASK_OPERATIONAL_VL;
1572 if (vls) {
1573 if (vls > ppd->vls_supported) {
1574 pr_warn("SubnSet(OPA_PortInfo) VL's supported invalid %d\n",
1575 pi->operational_vls);
1576 smp->status |= IB_SMP_INVALID_FIELD;
1577 } else {
1578 if (hfi1_set_ib_cfg(ppd, HFI1_IB_CFG_OP_VLS,
17fb4f29 1579 vls) == -EINVAL)
77241056
MM		 1580 smp->status |= IB_SMP_INVALID_FIELD;
1581 }
1582 }
1583
1584 if (pi->mkey_violations == 0)
4eb06882 1585 ibp->rvp.mkey_violations = 0;
77241056
MM		 1586
1587 if (pi->pkey_violations == 0)
4eb06882 1588 ibp->rvp.pkey_violations = 0;
77241056
MM		 1589
1590 if (pi->qkey_violations == 0)
4eb06882 1591 ibp->rvp.qkey_violations = 0;
77241056 1592
4eb06882 1593 ibp->rvp.subnet_timeout =
77241056
MM		 1594 pi->clientrereg_subnettimeout & OPA_PI_MASK_SUBNET_TIMEOUT;
1595
1596 crc_enabled = be16_to_cpu(pi->port_ltp_crc_mode);
1597 crc_enabled >>= 4;
1598 crc_enabled &= 0xf;
1599
1600 if (crc_enabled != 0)
1601 ppd->port_crc_mode_enabled = port_ltp_to_cap(crc_enabled);
1602
1603 ppd->is_active_optimize_enabled =
1604 !!(be16_to_cpu(pi->port_mode)
1605 & OPA_PI_MASK_PORT_ACTIVE_OPTOMIZE);
1606
1607 ls_new = pi->port_states.portphysstate_portstate &
1608 OPA_PI_MASK_PORT_STATE;
1609 ps_new = (pi->port_states.portphysstate_portstate &
1610 OPA_PI_MASK_PORT_PHYSICAL_STATE) >> 4;
1611
1612 if (ls_old == IB_PORT_INIT) {
1613 if (start_of_sm_config) {
1614 if (ls_new == ls_old || (ls_new == IB_PORT_ARMED))
1615 ppd->is_sm_config_started = 1;
1616 } else if (ls_new == IB_PORT_ARMED) {
1617 if (ppd->is_sm_config_started == 0)
1618 invalid = 1;
1619 }
1620 }
1621
1622 /* Handle CLIENT_REREGISTER event b/c SM asked us for it */
1623 if (clientrereg) {
1624 event.event = IB_EVENT_CLIENT_REREGISTER;
1625 ib_dispatch_event(&event);
1626 }
1627
1628 /*
1629 * Do the port state change now that the other link parameters
1630 * have been set.
1631 * Changing the port physical state only makes sense if the link
1632 * is down or is being set to down.
1633 */
1634
1635 ret = set_port_states(ppd, smp, ls_new, ps_new, invalid);
1636 if (ret)
1637 return ret;
1638
f1685179
ND		 1639 ret = __subn_get_opa_portinfo(smp, am, data, ibdev, port, resp_len,
1640 max_len);
77241056
MM		 1641
1642 /* restore re-reg bit per o14-12.2.1 */
1643 pi->clientrereg_subnettimeout |= clientrereg;
1644
1645 /*
1646 * Apply the new link downgrade policy. This may result in a link
1647 * bounce. Do this after everything else so things are settled.
1648 * Possible problem: if setting the port state above fails, then
1649 * the policy change is not applied.
1650 */
1651 if (call_link_downgrade_policy)
1652 apply_link_downgrade_policy(ppd, 0);
1653
1654 return ret;
1655
1656get_only:
f1685179
ND		 1657 return __subn_get_opa_portinfo(smp, am, data, ibdev, port, resp_len,
1658 max_len);
77241056
MM		 1659}
1660
1661/**
1662 * set_pkeys - set the PKEY table for ctxt 0
1663 * @dd: the hfi1_ib device
1664 * @port: the IB port number
1665 * @pkeys: the PKEY table
1666 */
1667static int set_pkeys(struct hfi1_devdata *dd, u8 port, u16 *pkeys)
1668{
1669 struct hfi1_pportdata *ppd;
1670 int i;
1671 int changed = 0;
1672 int update_includes_mgmt_partition = 0;
1673
1674 /*
1675 * IB port one/two always maps to context zero/one,
1676 * always a kernel context, no locking needed
1677 * If we get here with ppd setup, no need to check
1678 * that rcd is valid.
1679 */
1680 ppd = dd->pport + (port - 1);
1681 /*
1682 * If the update does not include the management pkey, don't do it.
1683 */
1684 for (i = 0; i < ARRAY_SIZE(ppd->pkeys); i++) {
1685 if (pkeys[i] == LIM_MGMT_P_KEY) {
1686 update_includes_mgmt_partition = 1;
1687 break;
1688 }
1689 }
1690
1691 if (!update_includes_mgmt_partition)
1692 return 1;
1693
1694 for (i = 0; i < ARRAY_SIZE(ppd->pkeys); i++) {
1695 u16 key = pkeys[i];
1696 u16 okey = ppd->pkeys[i];
1697
1698 if (key == okey)
1699 continue;
1700 /*
1701 * The SM gives us the complete PKey table. We have
1702 * to ensure that we put the PKeys in the matching
1703 * slots.
1704 */
1705 ppd->pkeys[i] = key;
1706 changed = 1;
1707 }
1708
1709 if (changed) {
77241056 1710 (void)hfi1_set_ib_cfg(ppd, HFI1_IB_CFG_PKEYS, 0);
34d351f8 1711 hfi1_event_pkey_change(dd, port);
77241056 1712 }
34d351f8 1713
77241056
MM		 1714 return 0;
1715}
1716
1717static int __subn_set_opa_pkeytable(struct opa_smp *smp, u32 am, u8 *data,
1718 struct ib_device *ibdev, u8 port,
f1685179 1719 u32 *resp_len, u32 max_len)
77241056
MM		 1720{
1721 struct hfi1_devdata *dd = dd_from_ibdev(ibdev);
1722 u32 n_blocks_sent = OPA_AM_NBLK(am);
1723 u32 start_block = am & 0x7ff;
50e5dcbe 1724 u16 *p = (u16 *)data;
77241056
MM		 1725 __be16 *q = (__be16 *)data;
1726 int i;
1727 u16 n_blocks_avail;
1728 unsigned npkeys = hfi1_get_npkeys(dd);
f1685179 1729 u32 size = 0;
77241056
MM		 1730
1731 if (n_blocks_sent == 0) {
1732 pr_warn("OPA Get PKey AM Invalid : P = %d; B = 0x%x; N = 0x%x\n",
1733 port, start_block, n_blocks_sent);
1734 smp->status |= IB_SMP_INVALID_FIELD;
1735 return reply((struct ib_mad_hdr *)smp);
1736 }
1737
8638b77f 1738 n_blocks_avail = (u16)(npkeys / OPA_PARTITION_TABLE_BLK_SIZE) + 1;
77241056 1739
f1685179
ND		 1740 size = sizeof(u16) * (n_blocks_sent * OPA_PARTITION_TABLE_BLK_SIZE);
1741
1742 if (smp_length_check(size, max_len)) {
1743 smp->status |= IB_SMP_INVALID_FIELD;
1744 return reply((struct ib_mad_hdr *)smp);
1745 }
1746
77241056
MM		 1747 if (start_block + n_blocks_sent > n_blocks_avail ||
1748 n_blocks_sent > OPA_NUM_PKEY_BLOCKS_PER_SMP) {
1749 pr_warn("OPA Set PKey AM Invalid : s 0x%x; req 0x%x; avail 0x%x; blk/smp 0x%lx\n",
1750 start_block, n_blocks_sent, n_blocks_avail,
1751 OPA_NUM_PKEY_BLOCKS_PER_SMP);
1752 smp->status |= IB_SMP_INVALID_FIELD;
1753 return reply((struct ib_mad_hdr *)smp);
1754 }
1755
1756 for (i = 0; i < n_blocks_sent * OPA_PARTITION_TABLE_BLK_SIZE; i++)
1757 p[i] = be16_to_cpu(q[i]);
1758
1759 if (start_block == 0 && set_pkeys(dd, port, p) != 0) {
1760 smp->status |= IB_SMP_INVALID_FIELD;
1761 return reply((struct ib_mad_hdr *)smp);
1762 }
1763
f1685179
ND		 1764 return __subn_get_opa_pkeytable(smp, am, data, ibdev, port, resp_len,
1765 max_len);
77241056
MM		 1766}
1767
77241056
MM		 1768#define ILLEGAL_VL 12
1769/*
1770 * filter_sc2vlt changes mappings to VL15 to ILLEGAL_VL (except
1771 * for SC15, which must map to VL15). If we don't remap things this
1772 * way it is possible for VL15 counters to increment when we try to
1773 * send on a SC which is mapped to an invalid VL.
3ca4fbc8 1774 * When getting the table convert ILLEGAL_VL back to VL15.
77241056 1775 */
3ca4fbc8 1776static void filter_sc2vlt(void *data, bool set)
77241056
MM		 1777{
1778 int i;
a787bde8 1779 u8 *pd = data;
77241056
MM		 1780
1781 for (i = 0; i < OPA_MAX_SCS; i++) {
1782 if (i == 15)
1783 continue;
3ca4fbc8
SS		 1784
1785 if (set) {
1786 if ((pd[i] & 0x1f) == 0xf)
1787 pd[i] = ILLEGAL_VL;
1788 } else {
1789 if ((pd[i] & 0x1f) == ILLEGAL_VL)
1790 pd[i] = 0xf;
1791 }
77241056
MM		 1792 }
1793}
1794
1795static int set_sc2vlt_tables(struct hfi1_devdata *dd, void *data)
1796{
a787bde8 1797 u64 *val = data;
77241056 1798
3ca4fbc8 1799 filter_sc2vlt(data, true);
77241056
MM		 1800
1801 write_csr(dd, SEND_SC2VLT0, *val++);
1802 write_csr(dd, SEND_SC2VLT1, *val++);
1803 write_csr(dd, SEND_SC2VLT2, *val++);
1804 write_csr(dd, SEND_SC2VLT3, *val++);
1805 write_seqlock_irq(&dd->sc2vl_lock);
a787bde8 1806 memcpy(dd->sc2vl, data, sizeof(dd->sc2vl));
77241056
MM		 1807 write_sequnlock_irq(&dd->sc2vl_lock);
1808 return 0;
1809}
1810
3ca4fbc8
SS		 1811static int get_sc2vlt_tables(struct hfi1_devdata *dd, void *data)
1812{
1813 u64 *val = (u64 *)data;
1814
1815 *val++ = read_csr(dd, SEND_SC2VLT0);
1816 *val++ = read_csr(dd, SEND_SC2VLT1);
1817 *val++ = read_csr(dd, SEND_SC2VLT2);
1818 *val++ = read_csr(dd, SEND_SC2VLT3);
1819
1820 filter_sc2vlt((u64 *)data, false);
1821 return 0;
1822}
1823
77241056
MM		 1824static int __subn_get_opa_sl_to_sc(struct opa_smp *smp, u32 am, u8 *data,
1825 struct ib_device *ibdev, u8 port,
f1685179 1826 u32 *resp_len, u32 max_len)
77241056
MM		 1827{
1828 struct hfi1_ibport *ibp = to_iport(ibdev, port);
6618c051 1829 u8 *p = data;
77241056
MM		 1830 size_t size = ARRAY_SIZE(ibp->sl_to_sc); /* == 32 */
1831 unsigned i;
1832
f1685179 1833 if (am || smp_length_check(size, max_len)) {
77241056
MM		 1834 smp->status |= IB_SMP_INVALID_FIELD;
1835 return reply((struct ib_mad_hdr *)smp);
1836 }
1837
1838 for (i = 0; i < ARRAY_SIZE(ibp->sl_to_sc); i++)
1839 *p++ = ibp->sl_to_sc[i];
1840
1841 if (resp_len)
1842 *resp_len += size;
1843
1844 return reply((struct ib_mad_hdr *)smp);
1845}
1846
1847static int __subn_set_opa_sl_to_sc(struct opa_smp *smp, u32 am, u8 *data,
1848 struct ib_device *ibdev, u8 port,
f1685179 1849 u32 *resp_len, u32 max_len)
77241056
MM		 1850{
1851 struct hfi1_ibport *ibp = to_iport(ibdev, port);
6618c051 1852 u8 *p = data;
f1685179 1853 size_t size = ARRAY_SIZE(ibp->sl_to_sc);
77241056 1854 int i;
0ec79e87 1855 u8 sc;
77241056 1856
f1685179 1857 if (am || smp_length_check(size, max_len)) {
77241056
MM		 1858 smp->status |= IB_SMP_INVALID_FIELD;
1859 return reply((struct ib_mad_hdr *)smp);
1860 }
1861
0ec79e87
KW		 1862 for (i = 0; i < ARRAY_SIZE(ibp->sl_to_sc); i++) {
1863 sc = *p++;
1864 if (ibp->sl_to_sc[i] != sc) {
1865 ibp->sl_to_sc[i] = sc;
1866
1867 /* Put all stale qps into error state */
1868 hfi1_error_port_qps(ibp, i);
1869 }
1870 }
77241056 1871
f1685179
ND		 1872 return __subn_get_opa_sl_to_sc(smp, am, data, ibdev, port, resp_len,
1873 max_len);
77241056
MM		 1874}
1875
1876static int __subn_get_opa_sc_to_sl(struct opa_smp *smp, u32 am, u8 *data,
1877 struct ib_device *ibdev, u8 port,
f1685179 1878 u32 *resp_len, u32 max_len)
77241056
MM		 1879{
1880 struct hfi1_ibport *ibp = to_iport(ibdev, port);
6618c051 1881 u8 *p = data;
77241056
MM		 1882 size_t size = ARRAY_SIZE(ibp->sc_to_sl); /* == 32 */
1883 unsigned i;
1884
f1685179 1885 if (am || smp_length_check(size, max_len)) {
77241056
MM		 1886 smp->status |= IB_SMP_INVALID_FIELD;
1887 return reply((struct ib_mad_hdr *)smp);
1888 }
1889
1890 for (i = 0; i < ARRAY_SIZE(ibp->sc_to_sl); i++)
1891 *p++ = ibp->sc_to_sl[i];
1892
1893 if (resp_len)
1894 *resp_len += size;
1895
1896 return reply((struct ib_mad_hdr *)smp);
1897}
1898
1899static int __subn_set_opa_sc_to_sl(struct opa_smp *smp, u32 am, u8 *data,
1900 struct ib_device *ibdev, u8 port,
f1685179 1901 u32 *resp_len, u32 max_len)
77241056
MM		 1902{
1903 struct hfi1_ibport *ibp = to_iport(ibdev, port);
f1685179 1904 size_t size = ARRAY_SIZE(ibp->sc_to_sl);
6618c051 1905 u8 *p = data;
77241056
MM		 1906 int i;
1907
f1685179 1908 if (am || smp_length_check(size, max_len)) {
77241056
MM		 1909 smp->status |= IB_SMP_INVALID_FIELD;
1910 return reply((struct ib_mad_hdr *)smp);
1911 }
1912
1913 for (i = 0; i < ARRAY_SIZE(ibp->sc_to_sl); i++)
1914 ibp->sc_to_sl[i] = *p++;
1915
f1685179
ND		 1916 return __subn_get_opa_sc_to_sl(smp, am, data, ibdev, port, resp_len,
1917 max_len);
77241056
MM		 1918}
1919
1920static int __subn_get_opa_sc_to_vlt(struct opa_smp *smp, u32 am, u8 *data,
1921 struct ib_device *ibdev, u8 port,
f1685179 1922 u32 *resp_len, u32 max_len)
77241056
MM		 1923{
1924 u32 n_blocks = OPA_AM_NBLK(am);
1925 struct hfi1_devdata *dd = dd_from_ibdev(ibdev);
50e5dcbe 1926 void *vp = (void *)data;
77241056
MM		 1927 size_t size = 4 * sizeof(u64);
1928
f1685179 1929 if (n_blocks != 1 || smp_length_check(size, max_len)) {
77241056
MM		 1930 smp->status |= IB_SMP_INVALID_FIELD;
1931 return reply((struct ib_mad_hdr *)smp);
1932 }
1933
1934 get_sc2vlt_tables(dd, vp);
1935
1936 if (resp_len)
1937 *resp_len += size;
1938
1939 return reply((struct ib_mad_hdr *)smp);
1940}
1941
1942static int __subn_set_opa_sc_to_vlt(struct opa_smp *smp, u32 am, u8 *data,
1943 struct ib_device *ibdev, u8 port,
f1685179 1944 u32 *resp_len, u32 max_len)
77241056
MM		 1945{
1946 u32 n_blocks = OPA_AM_NBLK(am);
1947 int async_update = OPA_AM_ASYNC(am);
1948 struct hfi1_devdata *dd = dd_from_ibdev(ibdev);
50e5dcbe 1949 void *vp = (void *)data;
77241056
MM		 1950 struct hfi1_pportdata *ppd;
1951 int lstate;
f1685179
ND		 1952 /*
1953 * set_sc2vlt_tables writes the information contained in *data
1954 * to four 64-bit registers SendSC2VLt[0-3]. We need to make
1955 * sure *max_len is not greater than the total size of the four
1956 * SendSC2VLt[0-3] registers.
1957 */
1958 size_t size = 4 * sizeof(u64);
77241056 1959
f1685179 1960 if (n_blocks != 1 || async_update || smp_length_check(size, max_len)) {
77241056
MM		 1961 smp->status |= IB_SMP_INVALID_FIELD;
1962 return reply((struct ib_mad_hdr *)smp);
1963 }
1964
1965 /* IB numbers ports from 1, hw from 0 */
1966 ppd = dd->pport + (port - 1);
1967 lstate = driver_lstate(ppd);
4d114fdd
JJ		 1968 /*
1969 * it's known that async_update is 0 by this point, but include
1970 * the explicit check for clarity
1971 */
77241056
MM		 1972 if (!async_update &&
1973 (lstate == IB_PORT_ARMED || lstate == IB_PORT_ACTIVE)) {
1974 smp->status |= IB_SMP_INVALID_FIELD;
1975 return reply((struct ib_mad_hdr *)smp);
1976 }
1977
1978 set_sc2vlt_tables(dd, vp);
1979
f1685179
ND		 1980 return __subn_get_opa_sc_to_vlt(smp, am, data, ibdev, port, resp_len,
1981 max_len);
77241056
MM		 1982}
1983
1984static int __subn_get_opa_sc_to_vlnt(struct opa_smp *smp, u32 am, u8 *data,
1985 struct ib_device *ibdev, u8 port,
f1685179 1986 u32 *resp_len, u32 max_len)
77241056
MM		 1987{
1988 u32 n_blocks = OPA_AM_NPORT(am);
1989 struct hfi1_devdata *dd = dd_from_ibdev(ibdev);
1990 struct hfi1_pportdata *ppd;
50e5dcbe 1991 void *vp = (void *)data;
f1685179 1992 int size = sizeof(struct sc2vlnt);
77241056 1993
f1685179 1994 if (n_blocks != 1 || smp_length_check(size, max_len)) {
77241056
MM		 1995 smp->status |= IB_SMP_INVALID_FIELD;
1996 return reply((struct ib_mad_hdr *)smp);
1997 }
1998
1999 ppd = dd->pport + (port - 1);
2000
f1685179 2001 fm_get_table(ppd, FM_TBL_SC2VLNT, vp);
77241056
MM		 2002
2003 if (resp_len)
2004 *resp_len += size;
2005
2006 return reply((struct ib_mad_hdr *)smp);
2007}
2008
2009static int __subn_set_opa_sc_to_vlnt(struct opa_smp *smp, u32 am, u8 *data,
2010 struct ib_device *ibdev, u8 port,
f1685179 2011 u32 *resp_len, u32 max_len)
77241056
MM		 2012{
2013 u32 n_blocks = OPA_AM_NPORT(am);
2014 struct hfi1_devdata *dd = dd_from_ibdev(ibdev);
2015 struct hfi1_pportdata *ppd;
50e5dcbe 2016 void *vp = (void *)data;
77241056 2017 int lstate;
f1685179 2018 int size = sizeof(struct sc2vlnt);
77241056 2019
f1685179 2020 if (n_blocks != 1 || smp_length_check(size, max_len)) {
77241056
MM		 2021 smp->status |= IB_SMP_INVALID_FIELD;
2022 return reply((struct ib_mad_hdr *)smp);
2023 }
2024
2025 /* IB numbers ports from 1, hw from 0 */
2026 ppd = dd->pport + (port - 1);
2027 lstate = driver_lstate(ppd);
2028 if (lstate == IB_PORT_ARMED || lstate == IB_PORT_ACTIVE) {
2029 smp->status |= IB_SMP_INVALID_FIELD;
2030 return reply((struct ib_mad_hdr *)smp);
2031 }
2032
2033 ppd = dd->pport + (port - 1);
2034
2035 fm_set_table(ppd, FM_TBL_SC2VLNT, vp);
2036
2037 return __subn_get_opa_sc_to_vlnt(smp, am, data, ibdev, port,
f1685179 2038 resp_len, max_len);
77241056
MM		 2039}
2040
2041static int __subn_get_opa_psi(struct opa_smp *smp, u32 am, u8 *data,
2042 struct ib_device *ibdev, u8 port,
f1685179 2043 u32 *resp_len, u32 max_len)
77241056
MM		 2044{
2045 u32 nports = OPA_AM_NPORT(am);
2046 u32 start_of_sm_config = OPA_AM_START_SM_CFG(am);
2047 u32 lstate;
2048 struct hfi1_ibport *ibp;
2049 struct hfi1_pportdata *ppd;
50e5dcbe 2050 struct opa_port_state_info *psi = (struct opa_port_state_info *)data;
77241056 2051
f1685179 2052 if (nports != 1 || smp_length_check(sizeof(*psi), max_len)) {
77241056
MM		 2053 smp->status |= IB_SMP_INVALID_FIELD;
2054 return reply((struct ib_mad_hdr *)smp);
2055 }
2056
2057 ibp = to_iport(ibdev, port);
2058 ppd = ppd_from_ibp(ibp);
2059
2060 lstate = driver_lstate(ppd);
2061
2062 if (start_of_sm_config && (lstate == IB_PORT_INIT))
2063 ppd->is_sm_config_started = 1;
2064
77241056
MM		 2065 psi->port_states.ledenable_offlinereason = ppd->neighbor_normal << 4;
2066 psi->port_states.ledenable_offlinereason |=
2067 ppd->is_sm_config_started << 5;
2068 psi->port_states.ledenable_offlinereason |=
a9c05e35 2069 ppd->offline_disabled_reason;
77241056
MM		 2070
2071 psi->port_states.portphysstate_portstate =
bec7c79c 2072 (driver_pstate(ppd) << 4) | (lstate & 0xf);
77241056 2073 psi->link_width_downgrade_tx_active =
aadfc3b2 2074 cpu_to_be16(ppd->link_width_downgrade_tx_active);
77241056 2075 psi->link_width_downgrade_rx_active =
aadfc3b2 2076 cpu_to_be16(ppd->link_width_downgrade_rx_active);
77241056
MM		 2077 if (resp_len)
2078 *resp_len += sizeof(struct opa_port_state_info);
2079
2080 return reply((struct ib_mad_hdr *)smp);
2081}
2082
2083static int __subn_set_opa_psi(struct opa_smp *smp, u32 am, u8 *data,
2084 struct ib_device *ibdev, u8 port,
f1685179 2085 u32 *resp_len, u32 max_len)
77241056
MM		 2086{
2087 u32 nports = OPA_AM_NPORT(am);
2088 u32 start_of_sm_config = OPA_AM_START_SM_CFG(am);
2089 u32 ls_old;
2090 u8 ls_new, ps_new;
2091 struct hfi1_ibport *ibp;
2092 struct hfi1_pportdata *ppd;
50e5dcbe 2093 struct opa_port_state_info *psi = (struct opa_port_state_info *)data;
77241056
MM		 2094 int ret, invalid = 0;
2095
f1685179 2096 if (nports != 1 || smp_length_check(sizeof(*psi), max_len)) {
77241056
MM		 2097 smp->status |= IB_SMP_INVALID_FIELD;
2098 return reply((struct ib_mad_hdr *)smp);
2099 }
2100
2101 ibp = to_iport(ibdev, port);
2102 ppd = ppd_from_ibp(ibp);
2103
2104 ls_old = driver_lstate(ppd);
2105
2106 ls_new = port_states_to_logical_state(&psi->port_states);
2107 ps_new = port_states_to_phys_state(&psi->port_states);
2108
2109 if (ls_old == IB_PORT_INIT) {
2110 if (start_of_sm_config) {
2111 if (ls_new == ls_old || (ls_new == IB_PORT_ARMED))
2112 ppd->is_sm_config_started = 1;
2113 } else if (ls_new == IB_PORT_ARMED) {
2114 if (ppd->is_sm_config_started == 0)
2115 invalid = 1;
2116 }
2117 }
2118
2119 ret = set_port_states(ppd, smp, ls_new, ps_new, invalid);
2120 if (ret)
2121 return ret;
2122
2123 if (invalid)
2124 smp->status |= IB_SMP_INVALID_FIELD;
2125
f1685179
ND		 2126 return __subn_get_opa_psi(smp, am, data, ibdev, port, resp_len,
2127 max_len);
77241056
MM		 2128}
2129
2130static int __subn_get_opa_cable_info(struct opa_smp *smp, u32 am, u8 *data,
2131 struct ib_device *ibdev, u8 port,
f1685179 2132 u32 *resp_len, u32 max_len)
77241056
MM		 2133{
2134 struct hfi1_devdata *dd = dd_from_ibdev(ibdev);
2135 u32 addr = OPA_AM_CI_ADDR(am);
2136 u32 len = OPA_AM_CI_LEN(am) + 1;
2137 int ret;
2138
f1685179
ND		 2139 if (dd->pport->port_type != PORT_TYPE_QSFP ||
2140 smp_length_check(len, max_len)) {
f29a08dc
EH		 2141 smp->status |= IB_SMP_INVALID_FIELD;
2142 return reply((struct ib_mad_hdr *)smp);
2143 }
2144
349ac71f 2145#define __CI_PAGE_SIZE BIT(7) /* 128 bytes */
77241056
MM		 2146#define __CI_PAGE_MASK ~(__CI_PAGE_SIZE - 1)
2147#define __CI_PAGE_NUM(a) ((a) & __CI_PAGE_MASK)
2148
4d114fdd
JJ		 2149 /*
2150 * check that addr is within spec, and
2151 * addr and (addr + len - 1) are on the same "page"
2152 */
77241056 2153 if (addr >= 4096 ||
17fb4f29 2154 (__CI_PAGE_NUM(addr) != __CI_PAGE_NUM(addr + len - 1))) {
77241056
MM		 2155 smp->status |= IB_SMP_INVALID_FIELD;
2156 return reply((struct ib_mad_hdr *)smp);
2157 }
2158
2159 ret = get_cable_info(dd, port, addr, len, data);
2160
2161 if (ret == -ENODEV) {
2162 smp->status |= IB_SMP_UNSUP_METH_ATTR;
2163 return reply((struct ib_mad_hdr *)smp);
2164 }
2165
2166 /* The address range for the CableInfo SMA query is wider than the
2167 * memory available on the QSFP cable. We want to return a valid
2168 * response, albeit zeroed out, for address ranges beyond available
2169 * memory but that are within the CableInfo query spec
2170 */
2171 if (ret < 0 && ret != -ERANGE) {
2172 smp->status |= IB_SMP_INVALID_FIELD;
2173 return reply((struct ib_mad_hdr *)smp);
2174 }
2175
2176 if (resp_len)
2177 *resp_len += len;
2178
2179 return reply((struct ib_mad_hdr *)smp);
2180}
2181
2182static int __subn_get_opa_bct(struct opa_smp *smp, u32 am, u8 *data,
f1685179
ND		 2183 struct ib_device *ibdev, u8 port, u32 *resp_len,
2184 u32 max_len)
77241056
MM		 2185{
2186 u32 num_ports = OPA_AM_NPORT(am);
2187 struct hfi1_devdata *dd = dd_from_ibdev(ibdev);
2188 struct hfi1_pportdata *ppd;
50e5dcbe 2189 struct buffer_control *p = (struct buffer_control *)data;
f1685179 2190 int size = sizeof(struct buffer_control);
77241056 2191
f1685179 2192 if (num_ports != 1 || smp_length_check(size, max_len)) {
77241056
MM		 2193 smp->status |= IB_SMP_INVALID_FIELD;
2194 return reply((struct ib_mad_hdr *)smp);
2195 }
2196
2197 ppd = dd->pport + (port - 1);
f1685179 2198 fm_get_table(ppd, FM_TBL_BUFFER_CONTROL, p);
77241056
MM		 2199 trace_bct_get(dd, p);
2200 if (resp_len)
2201 *resp_len += size;
2202
2203 return reply((struct ib_mad_hdr *)smp);
2204}
2205
2206static int __subn_set_opa_bct(struct opa_smp *smp, u32 am, u8 *data,
f1685179
ND		 2207 struct ib_device *ibdev, u8 port, u32 *resp_len,
2208 u32 max_len)
77241056
MM		 2209{
2210 u32 num_ports = OPA_AM_NPORT(am);
2211 struct hfi1_devdata *dd = dd_from_ibdev(ibdev);
2212 struct hfi1_pportdata *ppd;
50e5dcbe 2213 struct buffer_control *p = (struct buffer_control *)data;
77241056 2214
f1685179 2215 if (num_ports != 1 || smp_length_check(sizeof(*p), max_len)) {
77241056
MM		 2216 smp->status |= IB_SMP_INVALID_FIELD;
2217 return reply((struct ib_mad_hdr *)smp);
2218 }
2219 ppd = dd->pport + (port - 1);
2220 trace_bct_set(dd, p);
2221 if (fm_set_table(ppd, FM_TBL_BUFFER_CONTROL, p) < 0) {
2222 smp->status |= IB_SMP_INVALID_FIELD;
2223 return reply((struct ib_mad_hdr *)smp);
2224 }
2225
f1685179
ND		 2226 return __subn_get_opa_bct(smp, am, data, ibdev, port, resp_len,
2227 max_len);
77241056
MM		 2228}
2229
2230static int __subn_get_opa_vl_arb(struct opa_smp *smp, u32 am, u8 *data,
2231 struct ib_device *ibdev, u8 port,
f1685179 2232 u32 *resp_len, u32 max_len)
77241056
MM		 2233{
2234 struct hfi1_pportdata *ppd = ppd_from_ibp(to_iport(ibdev, port));
2235 u32 num_ports = OPA_AM_NPORT(am);
2236 u8 section = (am & 0x00ff0000) >> 16;
2237 u8 *p = data;
f1685179 2238 int size = 256;
77241056 2239
f1685179 2240 if (num_ports != 1 || smp_length_check(size, max_len)) {
77241056
MM		 2241 smp->status |= IB_SMP_INVALID_FIELD;
2242 return reply((struct ib_mad_hdr *)smp);
2243 }
2244
2245 switch (section) {
2246 case OPA_VLARB_LOW_ELEMENTS:
f1685179 2247 fm_get_table(ppd, FM_TBL_VL_LOW_ARB, p);
77241056
MM		 2248 break;
2249 case OPA_VLARB_HIGH_ELEMENTS:
f1685179 2250 fm_get_table(ppd, FM_TBL_VL_HIGH_ARB, p);
77241056
MM		 2251 break;
2252 case OPA_VLARB_PREEMPT_ELEMENTS:
f1685179 2253 fm_get_table(ppd, FM_TBL_VL_PREEMPT_ELEMS, p);
77241056
MM		 2254 break;
2255 case OPA_VLARB_PREEMPT_MATRIX:
f1685179 2256 fm_get_table(ppd, FM_TBL_VL_PREEMPT_MATRIX, p);
77241056
MM		 2257 break;
2258 default:
2259 pr_warn("OPA SubnGet(VL Arb) AM Invalid : 0x%x\n",
2260 be32_to_cpu(smp->attr_mod));
2261 smp->status |= IB_SMP_INVALID_FIELD;
f1685179 2262 size = 0;
77241056
MM		 2263 break;
2264 }
2265
2266 if (size > 0 && resp_len)
2267 *resp_len += size;
2268
2269 return reply((struct ib_mad_hdr *)smp);
2270}
2271
2272static int __subn_set_opa_vl_arb(struct opa_smp *smp, u32 am, u8 *data,
2273 struct ib_device *ibdev, u8 port,
f1685179 2274 u32 *resp_len, u32 max_len)
77241056
MM		 2275{
2276 struct hfi1_pportdata *ppd = ppd_from_ibp(to_iport(ibdev, port));
2277 u32 num_ports = OPA_AM_NPORT(am);
2278 u8 section = (am & 0x00ff0000) >> 16;
2279 u8 *p = data;
f1685179 2280 int size = 256;
77241056 2281
f1685179 2282 if (num_ports != 1 || smp_length_check(size, max_len)) {
77241056
MM		 2283 smp->status |= IB_SMP_INVALID_FIELD;
2284 return reply((struct ib_mad_hdr *)smp);
2285 }
2286
2287 switch (section) {
2288 case OPA_VLARB_LOW_ELEMENTS:
50e5dcbe 2289 (void)fm_set_table(ppd, FM_TBL_VL_LOW_ARB, p);
77241056
MM		 2290 break;
2291 case OPA_VLARB_HIGH_ELEMENTS:
50e5dcbe 2292 (void)fm_set_table(ppd, FM_TBL_VL_HIGH_ARB, p);
77241056 2293 break;
4d114fdd
JJ		 2294 /*
2295 * neither OPA_VLARB_PREEMPT_ELEMENTS, or OPA_VLARB_PREEMPT_MATRIX
2296 * can be changed from the default values
2297 */
77241056
MM		 2298 case OPA_VLARB_PREEMPT_ELEMENTS:
2299 /* FALLTHROUGH */
2300 case OPA_VLARB_PREEMPT_MATRIX:
2301 smp->status |= IB_SMP_UNSUP_METH_ATTR;
2302 break;
2303 default:
2304 pr_warn("OPA SubnSet(VL Arb) AM Invalid : 0x%x\n",
2305 be32_to_cpu(smp->attr_mod));
2306 smp->status |= IB_SMP_INVALID_FIELD;
2307 break;
2308 }
2309
f1685179
ND		 2310 return __subn_get_opa_vl_arb(smp, am, data, ibdev, port, resp_len,
2311 max_len);
77241056
MM		 2312}
2313
2314struct opa_pma_mad {
2315 struct ib_mad_hdr mad_hdr;
2316 u8 data[2024];
2317} __packed;
2318
77241056
MM		 2319struct opa_port_status_req {
2320 __u8 port_num;
2321 __u8 reserved[3];
2322 __be32 vl_select_mask;
2323};
2324
2325#define VL_MASK_ALL 0x000080ff
2326
2327struct opa_port_status_rsp {
2328 __u8 port_num;
2329 __u8 reserved[3];
2330 __be32 vl_select_mask;
2331
2332 /* Data counters */
2333 __be64 port_xmit_data;
2334 __be64 port_rcv_data;
2335 __be64 port_xmit_pkts;
2336 __be64 port_rcv_pkts;
2337 __be64 port_multicast_xmit_pkts;
2338 __be64 port_multicast_rcv_pkts;
2339 __be64 port_xmit_wait;
2340 __be64 sw_port_congestion;
2341 __be64 port_rcv_fecn;
2342 __be64 port_rcv_becn;
2343 __be64 port_xmit_time_cong;
2344 __be64 port_xmit_wasted_bw;
2345 __be64 port_xmit_wait_data;
2346 __be64 port_rcv_bubble;
2347 __be64 port_mark_fecn;
2348 /* Error counters */
2349 __be64 port_rcv_constraint_errors;
2350 __be64 port_rcv_switch_relay_errors;
2351 __be64 port_xmit_discards;
2352 __be64 port_xmit_constraint_errors;
2353 __be64 port_rcv_remote_physical_errors;
2354 __be64 local_link_integrity_errors;
2355 __be64 port_rcv_errors;
2356 __be64 excessive_buffer_overruns;
2357 __be64 fm_config_errors;
2358 __be32 link_error_recovery;
2359 __be32 link_downed;
2360 u8 uncorrectable_errors;
2361
2362 u8 link_quality_indicator; /* 5res, 3bit */
2363 u8 res2[6];
2364 struct _vls_pctrs {
2365 /* per-VL Data counters */
2366 __be64 port_vl_xmit_data;
2367 __be64 port_vl_rcv_data;
2368 __be64 port_vl_xmit_pkts;
2369 __be64 port_vl_rcv_pkts;
2370 __be64 port_vl_xmit_wait;
2371 __be64 sw_port_vl_congestion;
2372 __be64 port_vl_rcv_fecn;
2373 __be64 port_vl_rcv_becn;
2374 __be64 port_xmit_time_cong;
2375 __be64 port_vl_xmit_wasted_bw;
2376 __be64 port_vl_xmit_wait_data;
2377 __be64 port_vl_rcv_bubble;
2378 __be64 port_vl_mark_fecn;
2379 __be64 port_vl_xmit_discards;
2380 } vls[0]; /* real array size defined by # bits set in vl_select_mask */
2381};
2382
2383enum counter_selects {
2384 CS_PORT_XMIT_DATA = (1 << 31),
2385 CS_PORT_RCV_DATA = (1 << 30),
2386 CS_PORT_XMIT_PKTS = (1 << 29),
2387 CS_PORT_RCV_PKTS = (1 << 28),
2388 CS_PORT_MCAST_XMIT_PKTS = (1 << 27),
2389 CS_PORT_MCAST_RCV_PKTS = (1 << 26),
2390 CS_PORT_XMIT_WAIT = (1 << 25),
2391 CS_SW_PORT_CONGESTION = (1 << 24),
2392 CS_PORT_RCV_FECN = (1 << 23),
2393 CS_PORT_RCV_BECN = (1 << 22),
2394 CS_PORT_XMIT_TIME_CONG = (1 << 21),
2395 CS_PORT_XMIT_WASTED_BW = (1 << 20),
2396 CS_PORT_XMIT_WAIT_DATA = (1 << 19),
2397 CS_PORT_RCV_BUBBLE = (1 << 18),
2398 CS_PORT_MARK_FECN = (1 << 17),
2399 CS_PORT_RCV_CONSTRAINT_ERRORS = (1 << 16),
2400 CS_PORT_RCV_SWITCH_RELAY_ERRORS = (1 << 15),
2401 CS_PORT_XMIT_DISCARDS = (1 << 14),
2402 CS_PORT_XMIT_CONSTRAINT_ERRORS = (1 << 13),
2403 CS_PORT_RCV_REMOTE_PHYSICAL_ERRORS = (1 << 12),
2404 CS_LOCAL_LINK_INTEGRITY_ERRORS = (1 << 11),
2405 CS_PORT_RCV_ERRORS = (1 << 10),
2406 CS_EXCESSIVE_BUFFER_OVERRUNS = (1 << 9),
2407 CS_FM_CONFIG_ERRORS = (1 << 8),
2408 CS_LINK_ERROR_RECOVERY = (1 << 7),
2409 CS_LINK_DOWNED = (1 << 6),
2410 CS_UNCORRECTABLE_ERRORS = (1 << 5),
2411};
2412
2413struct opa_clear_port_status {
2414 __be64 port_select_mask[4];
2415 __be32 counter_select_mask;
2416};
2417
2418struct opa_aggregate {
2419 __be16 attr_id;
2420 __be16 err_reqlength; /* 1 bit, 8 res, 7 bit */
2421 __be32 attr_mod;
2422 u8 data[0];
2423};
2424
f0852922
AL		 2425#define MSK_LLI 0x000000f0
2426#define MSK_LLI_SFT 4
2427#define MSK_LER 0x0000000f
2428#define MSK_LER_SFT 0
2429#define ADD_LLI 8
2430#define ADD_LER 2
2431
2432/* Request contains first three fields, response contains those plus the rest */
77241056
MM		 2433struct opa_port_data_counters_msg {
2434 __be64 port_select_mask[4];
2435 __be32 vl_select_mask;
f0852922 2436 __be32 resolution;
77241056
MM		 2437
2438 /* Response fields follow */
77241056
MM		 2439 struct _port_dctrs {
2440 u8 port_number;
2441 u8 reserved2[3];
2442 __be32 link_quality_indicator; /* 29res, 3bit */
2443
2444 /* Data counters */
2445 __be64 port_xmit_data;
2446 __be64 port_rcv_data;
2447 __be64 port_xmit_pkts;
2448 __be64 port_rcv_pkts;
2449 __be64 port_multicast_xmit_pkts;
2450 __be64 port_multicast_rcv_pkts;
2451 __be64 port_xmit_wait;
2452 __be64 sw_port_congestion;
2453 __be64 port_rcv_fecn;
2454 __be64 port_rcv_becn;
2455 __be64 port_xmit_time_cong;
2456 __be64 port_xmit_wasted_bw;
2457 __be64 port_xmit_wait_data;
2458 __be64 port_rcv_bubble;
2459 __be64 port_mark_fecn;
2460
2461 __be64 port_error_counter_summary;
2462 /* Sum of error counts/port */
2463
2464 struct _vls_dctrs {
2465 /* per-VL Data counters */
2466 __be64 port_vl_xmit_data;
2467 __be64 port_vl_rcv_data;
2468 __be64 port_vl_xmit_pkts;
2469 __be64 port_vl_rcv_pkts;
2470 __be64 port_vl_xmit_wait;
2471 __be64 sw_port_vl_congestion;
2472 __be64 port_vl_rcv_fecn;
2473 __be64 port_vl_rcv_becn;
2474 __be64 port_xmit_time_cong;
2475 __be64 port_vl_xmit_wasted_bw;
2476 __be64 port_vl_xmit_wait_data;
2477 __be64 port_vl_rcv_bubble;
2478 __be64 port_vl_mark_fecn;
2479 } vls[0];
2480 /* array size defined by #bits set in vl_select_mask*/
2481 } port[1]; /* array size defined by #ports in attribute modifier */
2482};
2483
2484struct opa_port_error_counters64_msg {
4d114fdd
JJ		 2485 /*
2486 * Request contains first two fields, response contains the
2487 * whole magilla
2488 */
77241056
MM		 2489 __be64 port_select_mask[4];
2490 __be32 vl_select_mask;
2491
2492 /* Response-only fields follow */
2493 __be32 reserved1;
2494 struct _port_ectrs {
2495 u8 port_number;
2496 u8 reserved2[7];
2497 __be64 port_rcv_constraint_errors;
2498 __be64 port_rcv_switch_relay_errors;
2499 __be64 port_xmit_discards;
2500 __be64 port_xmit_constraint_errors;
2501 __be64 port_rcv_remote_physical_errors;
2502 __be64 local_link_integrity_errors;
2503 __be64 port_rcv_errors;
2504 __be64 excessive_buffer_overruns;
2505 __be64 fm_config_errors;
2506 __be32 link_error_recovery;
2507 __be32 link_downed;
2508 u8 uncorrectable_errors;
2509 u8 reserved3[7];
2510 struct _vls_ectrs {
2511 __be64 port_vl_xmit_discards;
2512 } vls[0];
2513 /* array size defined by #bits set in vl_select_mask */
2514 } port[1]; /* array size defined by #ports in attribute modifier */
2515};
2516
2517struct opa_port_error_info_msg {
2518 __be64 port_select_mask[4];
2519 __be32 error_info_select_mask;
2520 __be32 reserved1;
2521 struct _port_ei {
77241056
MM		 2522 u8 port_number;
2523 u8 reserved2[7];
2524
2525 /* PortRcvErrorInfo */
2526 struct {
2527 u8 status_and_code;
2528 union {
2529 u8 raw[17];
2530 struct {
2531 /* EI1to12 format */
2532 u8 packet_flit1[8];
2533 u8 packet_flit2[8];
2534 u8 remaining_flit_bits12;
2535 } ei1to12;
2536 struct {
2537 u8 packet_bytes[8];
2538 u8 remaining_flit_bits;
2539 } ei13;
2540 } ei;
2541 u8 reserved3[6];
2542 } __packed port_rcv_ei;
2543
2544 /* ExcessiveBufferOverrunInfo */
2545 struct {
2546 u8 status_and_sc;
2547 u8 reserved4[7];
2548 } __packed excessive_buffer_overrun_ei;
2549
2550 /* PortXmitConstraintErrorInfo */
2551 struct {
2552 u8 status;
2553 u8 reserved5;
2554 __be16 pkey;
2555 __be32 slid;
2556 } __packed port_xmit_constraint_ei;
2557
2558 /* PortRcvConstraintErrorInfo */
2559 struct {
2560 u8 status;
2561 u8 reserved6;
2562 __be16 pkey;
2563 __be32 slid;
2564 } __packed port_rcv_constraint_ei;
2565
2566 /* PortRcvSwitchRelayErrorInfo */
2567 struct {
2568 u8 status_and_code;
2569 u8 reserved7[3];
2570 __u32 error_info;
2571 } __packed port_rcv_switch_relay_ei;
2572
2573 /* UncorrectableErrorInfo */
2574 struct {
2575 u8 status_and_code;
2576 u8 reserved8;
2577 } __packed uncorrectable_ei;
2578
2579 /* FMConfigErrorInfo */
2580 struct {
2581 u8 status_and_code;
2582 u8 error_info;
2583 } __packed fm_config_ei;
2584 __u32 reserved9;
2585 } port[1]; /* actual array size defined by #ports in attr modifier */
2586};
2587
2588/* opa_port_error_info_msg error_info_select_mask bit definitions */
2589enum error_info_selects {
2590 ES_PORT_RCV_ERROR_INFO = (1 << 31),
2591 ES_EXCESSIVE_BUFFER_OVERRUN_INFO = (1 << 30),
2592 ES_PORT_XMIT_CONSTRAINT_ERROR_INFO = (1 << 29),
2593 ES_PORT_RCV_CONSTRAINT_ERROR_INFO = (1 << 28),
2594 ES_PORT_RCV_SWITCH_RELAY_ERROR_INFO = (1 << 27),
2595 ES_UNCORRECTABLE_ERROR_INFO = (1 << 26),
2596 ES_FM_CONFIG_ERROR_INFO = (1 << 25)
2597};
2598
2599static int pma_get_opa_classportinfo(struct opa_pma_mad *pmp,
17fb4f29 2600 struct ib_device *ibdev, u32 *resp_len)
77241056
MM		 2601{
2602 struct opa_class_port_info *p =
2603 (struct opa_class_port_info *)pmp->data;
2604
2605 memset(pmp->data, 0, sizeof(pmp->data));
2606
2607 if (pmp->mad_hdr.attr_mod != 0)
2608 pmp->mad_hdr.status |= IB_SMP_INVALID_FIELD;
2609
2610 p->base_version = OPA_MGMT_BASE_VERSION;
9fa240bb 2611 p->class_version = OPA_SM_CLASS_VERSION;
77241056
MM		 2612 /*
2613 * Expected response time is 4.096 usec. * 2^18 == 1.073741824 sec.
2614 */
2615 p->cap_mask2_resp_time = cpu_to_be32(18);
2616
2617 if (resp_len)
2618 *resp_len += sizeof(*p);
2619
2620 return reply((struct ib_mad_hdr *)pmp);
2621}
2622
2623static void a0_portstatus(struct hfi1_pportdata *ppd,
2624 struct opa_port_status_rsp *rsp, u32 vl_select_mask)
2625{
2626 if (!is_bx(ppd->dd)) {
2627 unsigned long vl;
d9d3e025 2628 u64 sum_vl_xmit_wait = 0;
77241056 2629 u32 vl_all_mask = VL_MASK_ALL;
77241056
MM		 2630
2631 for_each_set_bit(vl, (unsigned long *)&(vl_all_mask),
2632 8 * sizeof(vl_all_mask)) {
d9d3e025
IW		 2633 u64 tmp = sum_vl_xmit_wait +
2634 read_port_cntr(ppd, C_TX_WAIT_VL,
2635 idx_from_vl(vl));
2636 if (tmp < sum_vl_xmit_wait) {
2637 /* we wrapped */
2638 sum_vl_xmit_wait = (u64)~0;
2639 break;
2640 }
2641 sum_vl_xmit_wait = tmp;
77241056 2642 }
d9d3e025
IW		 2643 if (be64_to_cpu(rsp->port_xmit_wait) > sum_vl_xmit_wait)
2644 rsp->port_xmit_wait = cpu_to_be64(sum_vl_xmit_wait);
77241056
MM		 2645 }
2646}
2647
77241056 2648static int pma_get_opa_portstatus(struct opa_pma_mad *pmp,
17fb4f29
JJ		 2649 struct ib_device *ibdev,
2650 u8 port, u32 *resp_len)
77241056
MM		 2651{
2652 struct opa_port_status_req *req =
2653 (struct opa_port_status_req *)pmp->data;
2654 struct hfi1_devdata *dd = dd_from_ibdev(ibdev);
2655 struct opa_port_status_rsp *rsp;
2656 u32 vl_select_mask = be32_to_cpu(req->vl_select_mask);
2657 unsigned long vl;
2658 size_t response_data_size;
2659 u32 nports = be32_to_cpu(pmp->mad_hdr.attr_mod) >> 24;
2660 u8 port_num = req->port_num;
2661 u8 num_vls = hweight32(vl_select_mask);
2662 struct _vls_pctrs *vlinfo;
2663 struct hfi1_ibport *ibp = to_iport(ibdev, port);
2664 struct hfi1_pportdata *ppd = ppd_from_ibp(ibp);
2665 int vfi;
2666 u64 tmp, tmp2;
2667
2668 response_data_size = sizeof(struct opa_port_status_rsp) +
2669 num_vls * sizeof(struct _vls_pctrs);
2670 if (response_data_size > sizeof(pmp->data)) {
2671 pmp->mad_hdr.status |= OPA_PM_STATUS_REQUEST_TOO_LARGE;
2672 return reply((struct ib_mad_hdr *)pmp);
2673 }
2674
d0d236ea
JJ		 2675 if (nports != 1 || (port_num && port_num != port) ||
2676 num_vls > OPA_MAX_VLS || (vl_select_mask & ~VL_MASK_ALL)) {
77241056
MM		 2677 pmp->mad_hdr.status |= IB_SMP_INVALID_FIELD;
2678 return reply((struct ib_mad_hdr *)pmp);
2679 }
2680
2681 memset(pmp->data, 0, sizeof(pmp->data));
2682
2683 rsp = (struct opa_port_status_rsp *)pmp->data;
2684 if (port_num)
2685 rsp->port_num = port_num;
2686 else
2687 rsp->port_num = port;
2688
2689 rsp->port_rcv_constraint_errors =
2690 cpu_to_be64(read_port_cntr(ppd, C_SW_RCV_CSTR_ERR,
2691 CNTR_INVALID_VL));
2692
2693 hfi1_read_link_quality(dd, &rsp->link_quality_indicator);
2694
2695 rsp->vl_select_mask = cpu_to_be32(vl_select_mask);
2696 rsp->port_xmit_data = cpu_to_be64(read_dev_cntr(dd, C_DC_XMIT_FLITS,
2697 CNTR_INVALID_VL));
2698 rsp->port_rcv_data = cpu_to_be64(read_dev_cntr(dd, C_DC_RCV_FLITS,
2699 CNTR_INVALID_VL));
77241056
MM		 2700 rsp->port_xmit_pkts = cpu_to_be64(read_dev_cntr(dd, C_DC_XMIT_PKTS,
2701 CNTR_INVALID_VL));
2702 rsp->port_rcv_pkts = cpu_to_be64(read_dev_cntr(dd, C_DC_RCV_PKTS,
2703 CNTR_INVALID_VL));
2704 rsp->port_multicast_xmit_pkts =
2705 cpu_to_be64(read_dev_cntr(dd, C_DC_MC_XMIT_PKTS,
17fb4f29 2706 CNTR_INVALID_VL));
77241056
MM		 2707 rsp->port_multicast_rcv_pkts =
2708 cpu_to_be64(read_dev_cntr(dd, C_DC_MC_RCV_PKTS,
2709 CNTR_INVALID_VL));
2710 rsp->port_xmit_wait =
2711 cpu_to_be64(read_port_cntr(ppd, C_TX_WAIT, CNTR_INVALID_VL));
2712 rsp->port_rcv_fecn =
2713 cpu_to_be64(read_dev_cntr(dd, C_DC_RCV_FCN, CNTR_INVALID_VL));
2714 rsp->port_rcv_becn =
2715 cpu_to_be64(read_dev_cntr(dd, C_DC_RCV_BCN, CNTR_INVALID_VL));
2716 rsp->port_xmit_discards =
2717 cpu_to_be64(read_port_cntr(ppd, C_SW_XMIT_DSCD,
2718 CNTR_INVALID_VL));
2719 rsp->port_xmit_constraint_errors =
2720 cpu_to_be64(read_port_cntr(ppd, C_SW_XMIT_CSTR_ERR,
2721 CNTR_INVALID_VL));
2722 rsp->port_rcv_remote_physical_errors =
2723 cpu_to_be64(read_dev_cntr(dd, C_DC_RMT_PHY_ERR,
2724 CNTR_INVALID_VL));
3210314a
JP		 2725 rsp->local_link_integrity_errors =
2726 cpu_to_be64(read_dev_cntr(dd, C_DC_RX_REPLAY,
2727 CNTR_INVALID_VL));
77241056
MM		 2728 tmp = read_dev_cntr(dd, C_DC_SEQ_CRC_CNT, CNTR_INVALID_VL);
2729 tmp2 = tmp + read_dev_cntr(dd, C_DC_REINIT_FROM_PEER_CNT,
17fb4f29 2730 CNTR_INVALID_VL);
77241056
MM		 2731 if (tmp2 > (u32)UINT_MAX || tmp2 < tmp) {
2732 /* overflow/wrapped */
2733 rsp->link_error_recovery = cpu_to_be32(~0);
2734 } else {
2735 rsp->link_error_recovery = cpu_to_be32(tmp2);
2736 }
2737 rsp->port_rcv_errors =
2738 cpu_to_be64(read_dev_cntr(dd, C_DC_RCV_ERR, CNTR_INVALID_VL));
2739 rsp->excessive_buffer_overruns =
2740 cpu_to_be64(read_dev_cntr(dd, C_RCV_OVF, CNTR_INVALID_VL));
2741 rsp->fm_config_errors =
2742 cpu_to_be64(read_dev_cntr(dd, C_DC_FM_CFG_ERR,
2743 CNTR_INVALID_VL));
2744 rsp->link_downed = cpu_to_be32(read_port_cntr(ppd, C_SW_LINK_DOWN,
17fb4f29 2745 CNTR_INVALID_VL));
77241056
MM		 2746
2747 /* rsp->uncorrectable_errors is 8 bits wide, and it pegs at 0xff */
2748 tmp = read_dev_cntr(dd, C_DC_UNC_ERR, CNTR_INVALID_VL);
2749 rsp->uncorrectable_errors = tmp < 0x100 ? (tmp & 0xff) : 0xff;
2750
58721b8f 2751 vlinfo = &rsp->vls[0];
77241056
MM		 2752 vfi = 0;
2753 /* The vl_select_mask has been checked above, and we know
2754 * that it contains only entries which represent valid VLs.
2755 * So in the for_each_set_bit() loop below, we don't need
2756 * any additional checks for vl.
2757 */
2758 for_each_set_bit(vl, (unsigned long *)&(vl_select_mask),
2759 8 * sizeof(vl_select_mask)) {
2760 memset(vlinfo, 0, sizeof(*vlinfo));
2761
2762 tmp = read_dev_cntr(dd, C_DC_RX_FLIT_VL, idx_from_vl(vl));
2763 rsp->vls[vfi].port_vl_rcv_data = cpu_to_be64(tmp);
77241056
MM		 2764
2765 rsp->vls[vfi].port_vl_rcv_pkts =
2766 cpu_to_be64(read_dev_cntr(dd, C_DC_RX_PKT_VL,
17fb4f29 2767 idx_from_vl(vl)));
77241056
MM		 2768
2769 rsp->vls[vfi].port_vl_xmit_data =
2770 cpu_to_be64(read_port_cntr(ppd, C_TX_FLIT_VL,
17fb4f29 2771 idx_from_vl(vl)));
77241056
MM		 2772
2773 rsp->vls[vfi].port_vl_xmit_pkts =
2774 cpu_to_be64(read_port_cntr(ppd, C_TX_PKT_VL,
17fb4f29 2775 idx_from_vl(vl)));
77241056
MM		 2776
2777 rsp->vls[vfi].port_vl_xmit_wait =
2778 cpu_to_be64(read_port_cntr(ppd, C_TX_WAIT_VL,
17fb4f29 2779 idx_from_vl(vl)));
77241056
MM		 2780
2781 rsp->vls[vfi].port_vl_rcv_fecn =
2782 cpu_to_be64(read_dev_cntr(dd, C_DC_RCV_FCN_VL,
17fb4f29 2783 idx_from_vl(vl)));
77241056
MM		 2784
2785 rsp->vls[vfi].port_vl_rcv_becn =
2786 cpu_to_be64(read_dev_cntr(dd, C_DC_RCV_BCN_VL,
17fb4f29 2787 idx_from_vl(vl)));
77241056 2788
583eb8b8
JP		 2789 rsp->vls[vfi].port_vl_xmit_discards =
2790 cpu_to_be64(read_port_cntr(ppd, C_SW_XMIT_DSCD_VL,
2791 idx_from_vl(vl)));
77241056
MM		 2792 vlinfo++;
2793 vfi++;
2794 }
2795
2796 a0_portstatus(ppd, rsp, vl_select_mask);
2797
2798 if (resp_len)
2799 *resp_len += response_data_size;
2800
2801 return reply((struct ib_mad_hdr *)pmp);
2802}
2803
f0852922
AL		 2804static u64 get_error_counter_summary(struct ib_device *ibdev, u8 port,
2805 u8 res_lli, u8 res_ler)
77241056
MM		 2806{
2807 struct hfi1_devdata *dd = dd_from_ibdev(ibdev);
2808 struct hfi1_ibport *ibp = to_iport(ibdev, port);
2809 struct hfi1_pportdata *ppd = ppd_from_ibp(ibp);
2810 u64 error_counter_summary = 0, tmp;
2811
2812 error_counter_summary += read_port_cntr(ppd, C_SW_RCV_CSTR_ERR,
2813 CNTR_INVALID_VL);
2814 /* port_rcv_switch_relay_errors is 0 for HFIs */
2815 error_counter_summary += read_port_cntr(ppd, C_SW_XMIT_DSCD,
2816 CNTR_INVALID_VL);
2817 error_counter_summary += read_port_cntr(ppd, C_SW_XMIT_CSTR_ERR,
2818 CNTR_INVALID_VL);
2819 error_counter_summary += read_dev_cntr(dd, C_DC_RMT_PHY_ERR,
17fb4f29 2820 CNTR_INVALID_VL);
f0852922 2821 /* local link integrity must be right-shifted by the lli resolution */
3210314a
JP		 2822 error_counter_summary += (read_dev_cntr(dd, C_DC_RX_REPLAY,
2823 CNTR_INVALID_VL) >> res_lli);
f0852922
AL		 2824 /* link error recovery must b right-shifted by the ler resolution */
2825 tmp = read_dev_cntr(dd, C_DC_SEQ_CRC_CNT, CNTR_INVALID_VL);
2826 tmp += read_dev_cntr(dd, C_DC_REINIT_FROM_PEER_CNT, CNTR_INVALID_VL);
2827 error_counter_summary += (tmp >> res_ler);
77241056 2828 error_counter_summary += read_dev_cntr(dd, C_DC_RCV_ERR,
17fb4f29 2829 CNTR_INVALID_VL);
77241056
MM		 2830 error_counter_summary += read_dev_cntr(dd, C_RCV_OVF, CNTR_INVALID_VL);
2831 error_counter_summary += read_dev_cntr(dd, C_DC_FM_CFG_ERR,
17fb4f29 2832 CNTR_INVALID_VL);
77241056
MM		 2833 /* ppd->link_downed is a 32-bit value */
2834 error_counter_summary += read_port_cntr(ppd, C_SW_LINK_DOWN,
2835 CNTR_INVALID_VL);
2836 tmp = read_dev_cntr(dd, C_DC_UNC_ERR, CNTR_INVALID_VL);
2837 /* this is an 8-bit quantity */
2838 error_counter_summary += tmp < 0x100 ? (tmp & 0xff) : 0xff;
2839
2840 return error_counter_summary;
2841}
2842
d9d3e025 2843static void a0_datacounters(struct hfi1_pportdata *ppd, struct _port_dctrs *rsp,
77241056
MM		 2844 u32 vl_select_mask)
2845{
d9d3e025 2846 if (!is_bx(ppd->dd)) {
77241056 2847 unsigned long vl;
db00a055 2848 u64 sum_vl_xmit_wait = 0;
d9d3e025 2849 u32 vl_all_mask = VL_MASK_ALL;
db00a055 2850
d9d3e025
IW		 2851 for_each_set_bit(vl, (unsigned long *)&(vl_all_mask),
2852 8 * sizeof(vl_all_mask)) {
77241056 2853 u64 tmp = sum_vl_xmit_wait +
d9d3e025
IW		 2854 read_port_cntr(ppd, C_TX_WAIT_VL,
2855 idx_from_vl(vl));
77241056
MM		 2856 if (tmp < sum_vl_xmit_wait) {
2857 /* we wrapped */
50e5dcbe 2858 sum_vl_xmit_wait = (u64)~0;
77241056
MM		 2859 break;
2860 }
2861 sum_vl_xmit_wait = tmp;
2862 }
2863 if (be64_to_cpu(rsp->port_xmit_wait) > sum_vl_xmit_wait)
2864 rsp->port_xmit_wait = cpu_to_be64(sum_vl_xmit_wait);
2865 }
2866}
2867
b8d114eb
SS		 2868static void pma_get_opa_port_dctrs(struct ib_device *ibdev,
2869 struct _port_dctrs *rsp)
2870{
2871 struct hfi1_devdata *dd = dd_from_ibdev(ibdev);
2872
2873 rsp->port_xmit_data = cpu_to_be64(read_dev_cntr(dd, C_DC_XMIT_FLITS,
2874 CNTR_INVALID_VL));
2875 rsp->port_rcv_data = cpu_to_be64(read_dev_cntr(dd, C_DC_RCV_FLITS,
2876 CNTR_INVALID_VL));
2877 rsp->port_xmit_pkts = cpu_to_be64(read_dev_cntr(dd, C_DC_XMIT_PKTS,
2878 CNTR_INVALID_VL));
2879 rsp->port_rcv_pkts = cpu_to_be64(read_dev_cntr(dd, C_DC_RCV_PKTS,
2880 CNTR_INVALID_VL));
2881 rsp->port_multicast_xmit_pkts =
2882 cpu_to_be64(read_dev_cntr(dd, C_DC_MC_XMIT_PKTS,
2883 CNTR_INVALID_VL));
2884 rsp->port_multicast_rcv_pkts =
2885 cpu_to_be64(read_dev_cntr(dd, C_DC_MC_RCV_PKTS,
2886 CNTR_INVALID_VL));
2887}
2888
77241056 2889static int pma_get_opa_datacounters(struct opa_pma_mad *pmp,
17fb4f29
JJ		 2890 struct ib_device *ibdev,
2891 u8 port, u32 *resp_len)
77241056
MM		 2892{
2893 struct opa_port_data_counters_msg *req =
2894 (struct opa_port_data_counters_msg *)pmp->data;
2895 struct hfi1_devdata *dd = dd_from_ibdev(ibdev);
2896 struct hfi1_ibport *ibp = to_iport(ibdev, port);
2897 struct hfi1_pportdata *ppd = ppd_from_ibp(ibp);
2898 struct _port_dctrs *rsp;
2899 struct _vls_dctrs *vlinfo;
2900 size_t response_data_size;
2901 u32 num_ports;
77241056 2902 u8 lq, num_vls;
f0852922 2903 u8 res_lli, res_ler;
77241056 2904 u64 port_mask;
61a28d2b 2905 u8 port_num;
77241056
MM		 2906 unsigned long vl;
2907 u32 vl_select_mask;
2908 int vfi;
2909
2910 num_ports = be32_to_cpu(pmp->mad_hdr.attr_mod) >> 24;
77241056
MM		 2911 num_vls = hweight32(be32_to_cpu(req->vl_select_mask));
2912 vl_select_mask = be32_to_cpu(req->vl_select_mask);
f0852922
AL		 2913 res_lli = (u8)(be32_to_cpu(req->resolution) & MSK_LLI) >> MSK_LLI_SFT;
2914 res_lli = res_lli ? res_lli + ADD_LLI : 0;
2915 res_ler = (u8)(be32_to_cpu(req->resolution) & MSK_LER) >> MSK_LER_SFT;
2916 res_ler = res_ler ? res_ler + ADD_LER : 0;
77241056
MM		 2917
2918 if (num_ports != 1 || (vl_select_mask & ~VL_MASK_ALL)) {
2919 pmp->mad_hdr.status |= IB_SMP_INVALID_FIELD;
2920 return reply((struct ib_mad_hdr *)pmp);
2921 }
2922
2923 /* Sanity check */
2924 response_data_size = sizeof(struct opa_port_data_counters_msg) +
2925 num_vls * sizeof(struct _vls_dctrs);
2926
2927 if (response_data_size > sizeof(pmp->data)) {
2928 pmp->mad_hdr.status |= IB_SMP_INVALID_FIELD;
2929 return reply((struct ib_mad_hdr *)pmp);
2930 }
2931
2932 /*
2933 * The bit set in the mask needs to be consistent with the
2934 * port the request came in on.
2935 */
2936 port_mask = be64_to_cpu(req->port_select_mask[3]);
2937 port_num = find_first_bit((unsigned long *)&port_mask,
6aaa382f 2938 sizeof(port_mask) * 8);
77241056 2939
61a28d2b 2940 if (port_num != port) {
77241056
MM		 2941 pmp->mad_hdr.status |= IB_SMP_INVALID_FIELD;
2942 return reply((struct ib_mad_hdr *)pmp);
2943 }
2944
acc17d67 2945 rsp = &req->port[0];
77241056
MM		 2946 memset(rsp, 0, sizeof(*rsp));
2947
2948 rsp->port_number = port;
2949 /*
2950 * Note that link_quality_indicator is a 32 bit quantity in
2951 * 'datacounters' queries (as opposed to 'portinfo' queries,
2952 * where it's a byte).
2953 */
2954 hfi1_read_link_quality(dd, &lq);
2955 rsp->link_quality_indicator = cpu_to_be32((u32)lq);
b8d114eb 2956 pma_get_opa_port_dctrs(ibdev, rsp);
77241056 2957
77241056
MM		 2958 rsp->port_xmit_wait =
2959 cpu_to_be64(read_port_cntr(ppd, C_TX_WAIT, CNTR_INVALID_VL));
2960 rsp->port_rcv_fecn =
2961 cpu_to_be64(read_dev_cntr(dd, C_DC_RCV_FCN, CNTR_INVALID_VL));
2962 rsp->port_rcv_becn =
2963 cpu_to_be64(read_dev_cntr(dd, C_DC_RCV_BCN, CNTR_INVALID_VL));
77241056 2964 rsp->port_error_counter_summary =
f0852922
AL		 2965 cpu_to_be64(get_error_counter_summary(ibdev, port,
2966 res_lli, res_ler));
77241056 2967
58721b8f 2968 vlinfo = &rsp->vls[0];
77241056
MM		 2969 vfi = 0;
2970 /* The vl_select_mask has been checked above, and we know
2971 * that it contains only entries which represent valid VLs.
2972 * So in the for_each_set_bit() loop below, we don't need
2973 * any additional checks for vl.
2974 */
2975 for_each_set_bit(vl, (unsigned long *)&(vl_select_mask),
17fb4f29 2976 8 * sizeof(req->vl_select_mask)) {
77241056
MM		 2977 memset(vlinfo, 0, sizeof(*vlinfo));
2978
2979 rsp->vls[vfi].port_vl_xmit_data =
2980 cpu_to_be64(read_port_cntr(ppd, C_TX_FLIT_VL,
17fb4f29 2981 idx_from_vl(vl)));
77241056
MM		 2982
2983 rsp->vls[vfi].port_vl_rcv_data =
2984 cpu_to_be64(read_dev_cntr(dd, C_DC_RX_FLIT_VL,
17fb4f29 2985 idx_from_vl(vl)));
77241056
MM		 2986
2987 rsp->vls[vfi].port_vl_xmit_pkts =
2988 cpu_to_be64(read_port_cntr(ppd, C_TX_PKT_VL,
17fb4f29 2989 idx_from_vl(vl)));
77241056
MM		 2990
2991 rsp->vls[vfi].port_vl_rcv_pkts =
2992 cpu_to_be64(read_dev_cntr(dd, C_DC_RX_PKT_VL,
17fb4f29 2993 idx_from_vl(vl)));
77241056
MM		 2994
2995 rsp->vls[vfi].port_vl_xmit_wait =
2996 cpu_to_be64(read_port_cntr(ppd, C_TX_WAIT_VL,
17fb4f29 2997 idx_from_vl(vl)));
77241056
MM		 2998
2999 rsp->vls[vfi].port_vl_rcv_fecn =
3000 cpu_to_be64(read_dev_cntr(dd, C_DC_RCV_FCN_VL,
17fb4f29 3001 idx_from_vl(vl)));
77241056
MM		 3002 rsp->vls[vfi].port_vl_rcv_becn =
3003 cpu_to_be64(read_dev_cntr(dd, C_DC_RCV_BCN_VL,
17fb4f29 3004 idx_from_vl(vl)));
77241056
MM		 3005
3006 /* rsp->port_vl_xmit_time_cong is 0 for HFIs */
3007 /* rsp->port_vl_xmit_wasted_bw ??? */
3008 /* port_vl_xmit_wait_data - TXE (table 13-9 HFI spec) ???
4d114fdd
JJ		 3009 * does this differ from rsp->vls[vfi].port_vl_xmit_wait
3010 */
77241056 3011 /*rsp->vls[vfi].port_vl_mark_fecn =
4d114fdd
JJ		 3012 * cpu_to_be64(read_csr(dd, DCC_PRF_PORT_VL_MARK_FECN_CNT
3013 * + offset));
3014 */
77241056
MM		 3015 vlinfo++;
3016 vfi++;
3017 }
3018
d9d3e025 3019 a0_datacounters(ppd, rsp, vl_select_mask);
77241056
MM		 3020
3021 if (resp_len)
3022 *resp_len += response_data_size;
3023
3024 return reply((struct ib_mad_hdr *)pmp);
3025}
3026
b8d114eb
SS		 3027static int pma_get_ib_portcounters_ext(struct ib_pma_mad *pmp,
3028 struct ib_device *ibdev, u8 port)
3029{
3030 struct ib_pma_portcounters_ext *p = (struct ib_pma_portcounters_ext *)
3031 pmp->data;
3032 struct _port_dctrs rsp;
3033
3034 if (pmp->mad_hdr.attr_mod != 0 || p->port_select != port) {
3035 pmp->mad_hdr.status |= IB_SMP_INVALID_FIELD;
3036 goto bail;
3037 }
3038
3039 memset(&rsp, 0, sizeof(rsp));
3040 pma_get_opa_port_dctrs(ibdev, &rsp);
3041
3042 p->port_xmit_data = rsp.port_xmit_data;
3043 p->port_rcv_data = rsp.port_rcv_data;
3044 p->port_xmit_packets = rsp.port_xmit_pkts;
3045 p->port_rcv_packets = rsp.port_rcv_pkts;
3046 p->port_unicast_xmit_packets = 0;
3047 p->port_unicast_rcv_packets = 0;
3048 p->port_multicast_xmit_packets = rsp.port_multicast_xmit_pkts;
3049 p->port_multicast_rcv_packets = rsp.port_multicast_rcv_pkts;
3050
3051bail:
3052 return reply((struct ib_mad_hdr *)pmp);
3053}
3054
3055static void pma_get_opa_port_ectrs(struct ib_device *ibdev,
3056 struct _port_ectrs *rsp, u8 port)
3057{
3058 u64 tmp, tmp2;
3059 struct hfi1_devdata *dd = dd_from_ibdev(ibdev);
3060 struct hfi1_ibport *ibp = to_iport(ibdev, port);
3061 struct hfi1_pportdata *ppd = ppd_from_ibp(ibp);
3062
3063 tmp = read_dev_cntr(dd, C_DC_SEQ_CRC_CNT, CNTR_INVALID_VL);
3064 tmp2 = tmp + read_dev_cntr(dd, C_DC_REINIT_FROM_PEER_CNT,
3065 CNTR_INVALID_VL);
3066 if (tmp2 > (u32)UINT_MAX || tmp2 < tmp) {
3067 /* overflow/wrapped */
3068 rsp->link_error_recovery = cpu_to_be32(~0);
3069 } else {
3070 rsp->link_error_recovery = cpu_to_be32(tmp2);
3071 }
3072
3073 rsp->link_downed = cpu_to_be32(read_port_cntr(ppd, C_SW_LINK_DOWN,
3074 CNTR_INVALID_VL));
3075 rsp->port_rcv_errors =
3076 cpu_to_be64(read_dev_cntr(dd, C_DC_RCV_ERR, CNTR_INVALID_VL));
3077 rsp->port_rcv_remote_physical_errors =
3078 cpu_to_be64(read_dev_cntr(dd, C_DC_RMT_PHY_ERR,
3079 CNTR_INVALID_VL));
3080 rsp->port_rcv_switch_relay_errors = 0;
3081 rsp->port_xmit_discards =
3082 cpu_to_be64(read_port_cntr(ppd, C_SW_XMIT_DSCD,
3083 CNTR_INVALID_VL));
3084 rsp->port_xmit_constraint_errors =
3085 cpu_to_be64(read_port_cntr(ppd, C_SW_XMIT_CSTR_ERR,
3086 CNTR_INVALID_VL));
3087 rsp->port_rcv_constraint_errors =
3088 cpu_to_be64(read_port_cntr(ppd, C_SW_RCV_CSTR_ERR,
3089 CNTR_INVALID_VL));
3210314a
JP		 3090 rsp->local_link_integrity_errors =
3091 cpu_to_be64(read_dev_cntr(dd, C_DC_RX_REPLAY,
3092 CNTR_INVALID_VL));
b8d114eb
SS		 3093 rsp->excessive_buffer_overruns =
3094 cpu_to_be64(read_dev_cntr(dd, C_RCV_OVF, CNTR_INVALID_VL));
3095}
3096
77241056 3097static int pma_get_opa_porterrors(struct opa_pma_mad *pmp,
17fb4f29
JJ		 3098 struct ib_device *ibdev,
3099 u8 port, u32 *resp_len)
77241056
MM		 3100{
3101 size_t response_data_size;
3102 struct _port_ectrs *rsp;
eb2e557c 3103 u8 port_num;
77241056
MM		 3104 struct opa_port_error_counters64_msg *req;
3105 struct hfi1_devdata *dd = dd_from_ibdev(ibdev);
3106 u32 num_ports;
3107 u8 num_pslm;
3108 u8 num_vls;
3109 struct hfi1_ibport *ibp;
3110 struct hfi1_pportdata *ppd;
3111 struct _vls_ectrs *vlinfo;
3112 unsigned long vl;
b8d114eb 3113 u64 port_mask, tmp;
77241056
MM		 3114 u32 vl_select_mask;
3115 int vfi;
3116
3117 req = (struct opa_port_error_counters64_msg *)pmp->data;
3118
3119 num_ports = be32_to_cpu(pmp->mad_hdr.attr_mod) >> 24;
3120
3121 num_pslm = hweight64(be64_to_cpu(req->port_select_mask[3]));
3122 num_vls = hweight32(be32_to_cpu(req->vl_select_mask));
3123
3124 if (num_ports != 1 || num_ports != num_pslm) {
3125 pmp->mad_hdr.status |= IB_SMP_INVALID_FIELD;
3126 return reply((struct ib_mad_hdr *)pmp);
3127 }
3128
3129 response_data_size = sizeof(struct opa_port_error_counters64_msg) +
3130 num_vls * sizeof(struct _vls_ectrs);
3131
3132 if (response_data_size > sizeof(pmp->data)) {
3133 pmp->mad_hdr.status |= IB_SMP_INVALID_FIELD;
3134 return reply((struct ib_mad_hdr *)pmp);
3135 }
3136 /*
3137 * The bit set in the mask needs to be consistent with the
3138 * port the request came in on.
3139 */
3140 port_mask = be64_to_cpu(req->port_select_mask[3]);
3141 port_num = find_first_bit((unsigned long *)&port_mask,
6aaa382f 3142 sizeof(port_mask) * 8);
77241056 3143
eb2e557c 3144 if (port_num != port) {
77241056
MM		 3145 pmp->mad_hdr.status |= IB_SMP_INVALID_FIELD;
3146 return reply((struct ib_mad_hdr *)pmp);
3147 }
3148
acc17d67 3149 rsp = &req->port[0];
77241056
MM		 3150
3151 ibp = to_iport(ibdev, port_num);
3152 ppd = ppd_from_ibp(ibp);
3153
3154 memset(rsp, 0, sizeof(*rsp));
eb2e557c 3155 rsp->port_number = port_num;
77241056 3156
b8d114eb 3157 pma_get_opa_port_ectrs(ibdev, rsp, port_num);
77241056 3158
77241056
MM		 3159 rsp->port_rcv_remote_physical_errors =
3160 cpu_to_be64(read_dev_cntr(dd, C_DC_RMT_PHY_ERR,
b8d114eb 3161 CNTR_INVALID_VL));
77241056
MM		 3162 rsp->fm_config_errors =
3163 cpu_to_be64(read_dev_cntr(dd, C_DC_FM_CFG_ERR,
17fb4f29 3164 CNTR_INVALID_VL));
77241056 3165 tmp = read_dev_cntr(dd, C_DC_UNC_ERR, CNTR_INVALID_VL);
b8d114eb 3166
77241056 3167 rsp->uncorrectable_errors = tmp < 0x100 ? (tmp & 0xff) : 0xff;
2b719046
JP		 3168 rsp->port_rcv_errors =
3169 cpu_to_be64(read_dev_cntr(dd, C_DC_RCV_ERR, CNTR_INVALID_VL));
acc17d67 3170 vlinfo = &rsp->vls[0];
77241056
MM		 3171 vfi = 0;
3172 vl_select_mask = be32_to_cpu(req->vl_select_mask);
3173 for_each_set_bit(vl, (unsigned long *)&(vl_select_mask),
3174 8 * sizeof(req->vl_select_mask)) {
3175 memset(vlinfo, 0, sizeof(*vlinfo));
583eb8b8
JP		 3176 rsp->vls[vfi].port_vl_xmit_discards =
3177 cpu_to_be64(read_port_cntr(ppd, C_SW_XMIT_DSCD_VL,
3178 idx_from_vl(vl)));
77241056
MM		 3179 vlinfo += 1;
3180 vfi++;
3181 }
3182
3183 if (resp_len)
3184 *resp_len += response_data_size;
3185
3186 return reply((struct ib_mad_hdr *)pmp);
3187}
3188
b8d114eb
SS		 3189static int pma_get_ib_portcounters(struct ib_pma_mad *pmp,
3190 struct ib_device *ibdev, u8 port)
3191{
3192 struct ib_pma_portcounters *p = (struct ib_pma_portcounters *)
3193 pmp->data;
3194 struct _port_ectrs rsp;
3195 u64 temp_link_overrun_errors;
3196 u64 temp_64;
3197 u32 temp_32;
3198
3199 memset(&rsp, 0, sizeof(rsp));
3200 pma_get_opa_port_ectrs(ibdev, &rsp, port);
3201
3202 if (pmp->mad_hdr.attr_mod != 0 || p->port_select != port) {
3203 pmp->mad_hdr.status |= IB_SMP_INVALID_FIELD;
3204 goto bail;
3205 }
3206
3207 p->symbol_error_counter = 0; /* N/A for OPA */
3208
3209 temp_32 = be32_to_cpu(rsp.link_error_recovery);
3210 if (temp_32 > 0xFFUL)
3211 p->link_error_recovery_counter = 0xFF;
3212 else
3213 p->link_error_recovery_counter = (u8)temp_32;
3214
3215 temp_32 = be32_to_cpu(rsp.link_downed);
3216 if (temp_32 > 0xFFUL)
3217 p->link_downed_counter = 0xFF;
3218 else
3219 p->link_downed_counter = (u8)temp_32;
3220
3221 temp_64 = be64_to_cpu(rsp.port_rcv_errors);
3222 if (temp_64 > 0xFFFFUL)
3223 p->port_rcv_errors = cpu_to_be16(0xFFFF);
3224 else
3225 p->port_rcv_errors = cpu_to_be16((u16)temp_64);
3226
3227 temp_64 = be64_to_cpu(rsp.port_rcv_remote_physical_errors);
3228 if (temp_64 > 0xFFFFUL)
3229 p->port_rcv_remphys_errors = cpu_to_be16(0xFFFF);
3230 else
3231 p->port_rcv_remphys_errors = cpu_to_be16((u16)temp_64);
3232
3233 temp_64 = be64_to_cpu(rsp.port_rcv_switch_relay_errors);
3234 p->port_rcv_switch_relay_errors = cpu_to_be16((u16)temp_64);
3235
3236 temp_64 = be64_to_cpu(rsp.port_xmit_discards);
3237 if (temp_64 > 0xFFFFUL)
3238 p->port_xmit_discards = cpu_to_be16(0xFFFF);
3239 else
3240 p->port_xmit_discards = cpu_to_be16((u16)temp_64);
3241
3242 temp_64 = be64_to_cpu(rsp.port_xmit_constraint_errors);
3243 if (temp_64 > 0xFFUL)
3244 p->port_xmit_constraint_errors = 0xFF;
3245 else
3246 p->port_xmit_constraint_errors = (u8)temp_64;
3247
3248 temp_64 = be64_to_cpu(rsp.port_rcv_constraint_errors);
3249 if (temp_64 > 0xFFUL)
3250 p->port_rcv_constraint_errors = 0xFFUL;
3251 else
3252 p->port_rcv_constraint_errors = (u8)temp_64;
3253
3254 /* LocalLink: 7:4, BufferOverrun: 3:0 */
3255 temp_64 = be64_to_cpu(rsp.local_link_integrity_errors);
3256 if (temp_64 > 0xFUL)
3257 temp_64 = 0xFUL;
3258
3259 temp_link_overrun_errors = temp_64 << 4;
3260
3261 temp_64 = be64_to_cpu(rsp.excessive_buffer_overruns);
3262 if (temp_64 > 0xFUL)
3263 temp_64 = 0xFUL;
3264 temp_link_overrun_errors |= temp_64;
3265
3266 p->link_overrun_errors = (u8)temp_link_overrun_errors;
3267
3268 p->vl15_dropped = 0; /* N/A for OPA */
3269
3270bail:
3271 return reply((struct ib_mad_hdr *)pmp);
3272}
3273
77241056 3274static int pma_get_opa_errorinfo(struct opa_pma_mad *pmp,
17fb4f29
JJ		 3275 struct ib_device *ibdev,
3276 u8 port, u32 *resp_len)
77241056
MM		 3277{
3278 size_t response_data_size;
3279 struct _port_ei *rsp;
3280 struct opa_port_error_info_msg *req;
3281 struct hfi1_devdata *dd = dd_from_ibdev(ibdev);
3282 u64 port_mask;
3283 u32 num_ports;
eb2e557c 3284 u8 port_num;
77241056
MM		 3285 u8 num_pslm;
3286 u64 reg;
3287
3288 req = (struct opa_port_error_info_msg *)pmp->data;
acc17d67 3289 rsp = &req->port[0];
77241056
MM		 3290
3291 num_ports = OPA_AM_NPORT(be32_to_cpu(pmp->mad_hdr.attr_mod));
3292 num_pslm = hweight64(be64_to_cpu(req->port_select_mask[3]));
3293
3294 memset(rsp, 0, sizeof(*rsp));
3295
3296 if (num_ports != 1 || num_ports != num_pslm) {
3297 pmp->mad_hdr.status |= IB_SMP_INVALID_FIELD;
3298 return reply((struct ib_mad_hdr *)pmp);
3299 }
3300
3301 /* Sanity check */
3302 response_data_size = sizeof(struct opa_port_error_info_msg);
3303
3304 if (response_data_size > sizeof(pmp->data)) {
3305 pmp->mad_hdr.status |= IB_SMP_INVALID_FIELD;
3306 return reply((struct ib_mad_hdr *)pmp);
3307 }
3308
3309 /*
3310 * The bit set in the mask needs to be consistent with the port
3311 * the request came in on.
3312 */
3313 port_mask = be64_to_cpu(req->port_select_mask[3]);
3314 port_num = find_first_bit((unsigned long *)&port_mask,
6aaa382f 3315 sizeof(port_mask) * 8);
77241056 3316
eb2e557c 3317 if (port_num != port) {
77241056
MM		 3318 pmp->mad_hdr.status |= IB_SMP_INVALID_FIELD;
3319 return reply((struct ib_mad_hdr *)pmp);
3320 }
3321
3322 /* PortRcvErrorInfo */
3323 rsp->port_rcv_ei.status_and_code =
3324 dd->err_info_rcvport.status_and_code;
3325 memcpy(&rsp->port_rcv_ei.ei.ei1to12.packet_flit1,
17fb4f29 3326 &dd->err_info_rcvport.packet_flit1, sizeof(u64));
77241056 3327 memcpy(&rsp->port_rcv_ei.ei.ei1to12.packet_flit2,
17fb4f29 3328 &dd->err_info_rcvport.packet_flit2, sizeof(u64));
77241056
MM		 3329
3330 /* ExcessiverBufferOverrunInfo */
3331 reg = read_csr(dd, RCV_ERR_INFO);
3332 if (reg & RCV_ERR_INFO_RCV_EXCESS_BUFFER_OVERRUN_SMASK) {
4d114fdd
JJ		 3333 /*
3334 * if the RcvExcessBufferOverrun bit is set, save SC of
3335 * first pkt that encountered an excess buffer overrun
3336 */
77241056
MM		 3337 u8 tmp = (u8)reg;
3338
3339 tmp &= RCV_ERR_INFO_RCV_EXCESS_BUFFER_OVERRUN_SC_SMASK;
3340 tmp <<= 2;
3341 rsp->excessive_buffer_overrun_ei.status_and_sc = tmp;
3342 /* set the status bit */
3343 rsp->excessive_buffer_overrun_ei.status_and_sc |= 0x80;
3344 }
3345
3346 rsp->port_xmit_constraint_ei.status =
3347 dd->err_info_xmit_constraint.status;
3348 rsp->port_xmit_constraint_ei.pkey =
3349 cpu_to_be16(dd->err_info_xmit_constraint.pkey);
3350 rsp->port_xmit_constraint_ei.slid =
3351 cpu_to_be32(dd->err_info_xmit_constraint.slid);
3352
3353 rsp->port_rcv_constraint_ei.status =
3354 dd->err_info_rcv_constraint.status;
3355 rsp->port_rcv_constraint_ei.pkey =
3356 cpu_to_be16(dd->err_info_rcv_constraint.pkey);
3357 rsp->port_rcv_constraint_ei.slid =
3358 cpu_to_be32(dd->err_info_rcv_constraint.slid);
3359
3360 /* UncorrectableErrorInfo */
3361 rsp->uncorrectable_ei.status_and_code = dd->err_info_uncorrectable;
3362
3363 /* FMConfigErrorInfo */
3364 rsp->fm_config_ei.status_and_code = dd->err_info_fmconfig;
3365
3366 if (resp_len)
3367 *resp_len += response_data_size;
3368
3369 return reply((struct ib_mad_hdr *)pmp);
3370}
3371
3372static int pma_set_opa_portstatus(struct opa_pma_mad *pmp,
17fb4f29
JJ		 3373 struct ib_device *ibdev,
3374 u8 port, u32 *resp_len)
77241056
MM		 3375{
3376 struct opa_clear_port_status *req =
3377 (struct opa_clear_port_status *)pmp->data;
3378 struct hfi1_devdata *dd = dd_from_ibdev(ibdev);
3379 struct hfi1_ibport *ibp = to_iport(ibdev, port);
3380 struct hfi1_pportdata *ppd = ppd_from_ibp(ibp);
3381 u32 nports = be32_to_cpu(pmp->mad_hdr.attr_mod) >> 24;
3382 u64 portn = be64_to_cpu(req->port_select_mask[3]);
3383 u32 counter_select = be32_to_cpu(req->counter_select_mask);
3384 u32 vl_select_mask = VL_MASK_ALL; /* clear all per-vl cnts */
3385 unsigned long vl;
3386
3387 if ((nports != 1) || (portn != 1 << port)) {
3388 pmp->mad_hdr.status |= IB_SMP_INVALID_FIELD;
3389 return reply((struct ib_mad_hdr *)pmp);
3390 }
3391 /*
3392 * only counters returned by pma_get_opa_portstatus() are
3393 * handled, so when pma_get_opa_portstatus() gets a fix,
3394 * the corresponding change should be made here as well.
3395 */
3396
3397 if (counter_select & CS_PORT_XMIT_DATA)
3398 write_dev_cntr(dd, C_DC_XMIT_FLITS, CNTR_INVALID_VL, 0);
3399
3400 if (counter_select & CS_PORT_RCV_DATA)
3401 write_dev_cntr(dd, C_DC_RCV_FLITS, CNTR_INVALID_VL, 0);
3402
3403 if (counter_select & CS_PORT_XMIT_PKTS)
3404 write_dev_cntr(dd, C_DC_XMIT_PKTS, CNTR_INVALID_VL, 0);
3405
3406 if (counter_select & CS_PORT_RCV_PKTS)
3407 write_dev_cntr(dd, C_DC_RCV_PKTS, CNTR_INVALID_VL, 0);
3408
3409 if (counter_select & CS_PORT_MCAST_XMIT_PKTS)
3410 write_dev_cntr(dd, C_DC_MC_XMIT_PKTS, CNTR_INVALID_VL, 0);
3411
3412 if (counter_select & CS_PORT_MCAST_RCV_PKTS)
3413 write_dev_cntr(dd, C_DC_MC_RCV_PKTS, CNTR_INVALID_VL, 0);
3414
3415 if (counter_select & CS_PORT_XMIT_WAIT)
3416 write_port_cntr(ppd, C_TX_WAIT, CNTR_INVALID_VL, 0);
3417
3418 /* ignore cs_sw_portCongestion for HFIs */
3419
3420 if (counter_select & CS_PORT_RCV_FECN)
3421 write_dev_cntr(dd, C_DC_RCV_FCN, CNTR_INVALID_VL, 0);
3422
3423 if (counter_select & CS_PORT_RCV_BECN)
3424 write_dev_cntr(dd, C_DC_RCV_BCN, CNTR_INVALID_VL, 0);
3425
3426 /* ignore cs_port_xmit_time_cong for HFIs */
3427 /* ignore cs_port_xmit_wasted_bw for now */
3428 /* ignore cs_port_xmit_wait_data for now */
3429 if (counter_select & CS_PORT_RCV_BUBBLE)
3430 write_dev_cntr(dd, C_DC_RCV_BBL, CNTR_INVALID_VL, 0);
3431
3432 /* Only applicable for switch */
4d114fdd
JJ		 3433 /* if (counter_select & CS_PORT_MARK_FECN)
3434 * write_csr(dd, DCC_PRF_PORT_MARK_FECN_CNT, 0);
3435 */
77241056
MM		 3436
3437 if (counter_select & CS_PORT_RCV_CONSTRAINT_ERRORS)
3438 write_port_cntr(ppd, C_SW_RCV_CSTR_ERR, CNTR_INVALID_VL, 0);
3439
3440 /* ignore cs_port_rcv_switch_relay_errors for HFIs */
3441 if (counter_select & CS_PORT_XMIT_DISCARDS)
3442 write_port_cntr(ppd, C_SW_XMIT_DSCD, CNTR_INVALID_VL, 0);
3443
3444 if (counter_select & CS_PORT_XMIT_CONSTRAINT_ERRORS)
3445 write_port_cntr(ppd, C_SW_XMIT_CSTR_ERR, CNTR_INVALID_VL, 0);
3446
3447 if (counter_select & CS_PORT_RCV_REMOTE_PHYSICAL_ERRORS)
3448 write_dev_cntr(dd, C_DC_RMT_PHY_ERR, CNTR_INVALID_VL, 0);
3449
3210314a 3450 if (counter_select & CS_LOCAL_LINK_INTEGRITY_ERRORS)
77241056 3451 write_dev_cntr(dd, C_DC_RX_REPLAY, CNTR_INVALID_VL, 0);
77241056
MM		 3452
3453 if (counter_select & CS_LINK_ERROR_RECOVERY) {
3454 write_dev_cntr(dd, C_DC_SEQ_CRC_CNT, CNTR_INVALID_VL, 0);
3455 write_dev_cntr(dd, C_DC_REINIT_FROM_PEER_CNT,
17fb4f29 3456 CNTR_INVALID_VL, 0);
77241056
MM		 3457 }
3458
3459 if (counter_select & CS_PORT_RCV_ERRORS)
3460 write_dev_cntr(dd, C_DC_RCV_ERR, CNTR_INVALID_VL, 0);
3461
3462 if (counter_select & CS_EXCESSIVE_BUFFER_OVERRUNS) {
3463 write_dev_cntr(dd, C_RCV_OVF, CNTR_INVALID_VL, 0);
3464 dd->rcv_ovfl_cnt = 0;
3465 }
3466
3467 if (counter_select & CS_FM_CONFIG_ERRORS)
3468 write_dev_cntr(dd, C_DC_FM_CFG_ERR, CNTR_INVALID_VL, 0);
3469
3470 if (counter_select & CS_LINK_DOWNED)
3471 write_port_cntr(ppd, C_SW_LINK_DOWN, CNTR_INVALID_VL, 0);
3472
3473 if (counter_select & CS_UNCORRECTABLE_ERRORS)
3474 write_dev_cntr(dd, C_DC_UNC_ERR, CNTR_INVALID_VL, 0);
3475
3476 for_each_set_bit(vl, (unsigned long *)&(vl_select_mask),
3477 8 * sizeof(vl_select_mask)) {
77241056
MM		 3478 if (counter_select & CS_PORT_XMIT_DATA)
3479 write_port_cntr(ppd, C_TX_FLIT_VL, idx_from_vl(vl), 0);
3480
3481 if (counter_select & CS_PORT_RCV_DATA)
3482 write_dev_cntr(dd, C_DC_RX_FLIT_VL, idx_from_vl(vl), 0);
3483
3484 if (counter_select & CS_PORT_XMIT_PKTS)
3485 write_port_cntr(ppd, C_TX_PKT_VL, idx_from_vl(vl), 0);
3486
3487 if (counter_select & CS_PORT_RCV_PKTS)
3488 write_dev_cntr(dd, C_DC_RX_PKT_VL, idx_from_vl(vl), 0);
3489
3490 if (counter_select & CS_PORT_XMIT_WAIT)
3491 write_port_cntr(ppd, C_TX_WAIT_VL, idx_from_vl(vl), 0);
3492
3493 /* sw_port_vl_congestion is 0 for HFIs */
3494 if (counter_select & CS_PORT_RCV_FECN)
3495 write_dev_cntr(dd, C_DC_RCV_FCN_VL, idx_from_vl(vl), 0);
3496
3497 if (counter_select & CS_PORT_RCV_BECN)
3498 write_dev_cntr(dd, C_DC_RCV_BCN_VL, idx_from_vl(vl), 0);
3499
3500 /* port_vl_xmit_time_cong is 0 for HFIs */
3501 /* port_vl_xmit_wasted_bw ??? */
3502 /* port_vl_xmit_wait_data - TXE (table 13-9 HFI spec) ??? */
3503 if (counter_select & CS_PORT_RCV_BUBBLE)
3504 write_dev_cntr(dd, C_DC_RCV_BBL_VL, idx_from_vl(vl), 0);
3505
4d114fdd
JJ		 3506 /* if (counter_select & CS_PORT_MARK_FECN)
3507 * write_csr(dd, DCC_PRF_PORT_VL_MARK_FECN_CNT + offset, 0);
3508 */
583eb8b8
JP		 3509 if (counter_select & C_SW_XMIT_DSCD_VL)
3510 write_port_cntr(ppd, C_SW_XMIT_DSCD_VL,
3511 idx_from_vl(vl), 0);
77241056
MM		 3512 }
3513
3514 if (resp_len)
3515 *resp_len += sizeof(*req);
3516
3517 return reply((struct ib_mad_hdr *)pmp);
3518}
3519
3520static int pma_set_opa_errorinfo(struct opa_pma_mad *pmp,
17fb4f29
JJ		 3521 struct ib_device *ibdev,
3522 u8 port, u32 *resp_len)
77241056
MM		 3523{
3524 struct _port_ei *rsp;
3525 struct opa_port_error_info_msg *req;
3526 struct hfi1_devdata *dd = dd_from_ibdev(ibdev);
3527 u64 port_mask;
3528 u32 num_ports;
eb2e557c 3529 u8 port_num;
77241056
MM		 3530 u8 num_pslm;
3531 u32 error_info_select;
3532
3533 req = (struct opa_port_error_info_msg *)pmp->data;
acc17d67 3534 rsp = &req->port[0];
77241056
MM		 3535
3536 num_ports = OPA_AM_NPORT(be32_to_cpu(pmp->mad_hdr.attr_mod));
3537 num_pslm = hweight64(be64_to_cpu(req->port_select_mask[3]));
3538
3539 memset(rsp, 0, sizeof(*rsp));
3540
3541 if (num_ports != 1 || num_ports != num_pslm) {
3542 pmp->mad_hdr.status |= IB_SMP_INVALID_FIELD;
3543 return reply((struct ib_mad_hdr *)pmp);
3544 }
3545
3546 /*
3547 * The bit set in the mask needs to be consistent with the port
3548 * the request came in on.
3549 */
3550 port_mask = be64_to_cpu(req->port_select_mask[3]);
3551 port_num = find_first_bit((unsigned long *)&port_mask,
6aaa382f 3552 sizeof(port_mask) * 8);
77241056 3553
eb2e557c 3554 if (port_num != port) {
77241056
MM		 3555 pmp->mad_hdr.status |= IB_SMP_INVALID_FIELD;
3556 return reply((struct ib_mad_hdr *)pmp);
3557 }
3558
3559 error_info_select = be32_to_cpu(req->error_info_select_mask);
3560
3561 /* PortRcvErrorInfo */
3562 if (error_info_select & ES_PORT_RCV_ERROR_INFO)
3563 /* turn off status bit */
3564 dd->err_info_rcvport.status_and_code &= ~OPA_EI_STATUS_SMASK;
3565
3566 /* ExcessiverBufferOverrunInfo */
3567 if (error_info_select & ES_EXCESSIVE_BUFFER_OVERRUN_INFO)
4d114fdd
JJ		 3568 /*
3569 * status bit is essentially kept in the h/w - bit 5 of
3570 * RCV_ERR_INFO
3571 */
77241056
MM		 3572 write_csr(dd, RCV_ERR_INFO,
3573 RCV_ERR_INFO_RCV_EXCESS_BUFFER_OVERRUN_SMASK);
3574
3575 if (error_info_select & ES_PORT_XMIT_CONSTRAINT_ERROR_INFO)
3576 dd->err_info_xmit_constraint.status &= ~OPA_EI_STATUS_SMASK;
3577
3578 if (error_info_select & ES_PORT_RCV_CONSTRAINT_ERROR_INFO)
3579 dd->err_info_rcv_constraint.status &= ~OPA_EI_STATUS_SMASK;
3580
3581 /* UncorrectableErrorInfo */
3582 if (error_info_select & ES_UNCORRECTABLE_ERROR_INFO)
3583 /* turn off status bit */
3584 dd->err_info_uncorrectable &= ~OPA_EI_STATUS_SMASK;
3585
3586 /* FMConfigErrorInfo */
3587 if (error_info_select & ES_FM_CONFIG_ERROR_INFO)
3588 /* turn off status bit */
3589 dd->err_info_fmconfig &= ~OPA_EI_STATUS_SMASK;
3590
3591 if (resp_len)
3592 *resp_len += sizeof(*req);
3593
3594 return reply((struct ib_mad_hdr *)pmp);
3595}
3596
3597struct opa_congestion_info_attr {
3598 __be16 congestion_info;
3599 u8 control_table_cap; /* Multiple of 64 entry unit CCTs */
3600 u8 congestion_log_length;
3601} __packed;
3602
3603static int __subn_get_opa_cong_info(struct opa_smp *smp, u32 am, u8 *data,
3604 struct ib_device *ibdev, u8 port,
f1685179 3605 u32 *resp_len, u32 max_len)
77241056
MM		 3606{
3607 struct opa_congestion_info_attr *p =
3608 (struct opa_congestion_info_attr *)data;
3609 struct hfi1_ibport *ibp = to_iport(ibdev, port);
3610 struct hfi1_pportdata *ppd = ppd_from_ibp(ibp);
3611
f1685179
ND		 3612 if (smp_length_check(sizeof(*p), max_len)) {
3613 smp->status |= IB_SMP_INVALID_FIELD;
3614 return reply((struct ib_mad_hdr *)smp);
3615 }
3616
77241056
MM		 3617 p->congestion_info = 0;
3618 p->control_table_cap = ppd->cc_max_table_entries;
3619 p->congestion_log_length = OPA_CONG_LOG_ELEMS;
3620
3621 if (resp_len)
3622 *resp_len += sizeof(*p);
3623
3624 return reply((struct ib_mad_hdr *)smp);
3625}
3626
3627static int __subn_get_opa_cong_setting(struct opa_smp *smp, u32 am,
17fb4f29 3628 u8 *data, struct ib_device *ibdev,
f1685179 3629 u8 port, u32 *resp_len, u32 max_len)
77241056
MM		 3630{
3631 int i;
3632 struct opa_congestion_setting_attr *p =
50e5dcbe 3633 (struct opa_congestion_setting_attr *)data;
77241056
MM		 3634 struct hfi1_ibport *ibp = to_iport(ibdev, port);
3635 struct hfi1_pportdata *ppd = ppd_from_ibp(ibp);
3636 struct opa_congestion_setting_entry_shadow *entries;
3637 struct cc_state *cc_state;
3638
f1685179
ND		 3639 if (smp_length_check(sizeof(*p), max_len)) {
3640 smp->status |= IB_SMP_INVALID_FIELD;
3641 return reply((struct ib_mad_hdr *)smp);
3642 }
3643
77241056
MM		 3644 rcu_read_lock();
3645
3646 cc_state = get_cc_state(ppd);
3647
d125a6c6 3648 if (!cc_state) {
77241056
MM		 3649 rcu_read_unlock();
3650 return reply((struct ib_mad_hdr *)smp);
3651 }
3652
3653 entries = cc_state->cong_setting.entries;
3654 p->port_control = cpu_to_be16(cc_state->cong_setting.port_control);
3655 p->control_map = cpu_to_be32(cc_state->cong_setting.control_map);
3656 for (i = 0; i < OPA_MAX_SLS; i++) {
3657 p->entries[i].ccti_increase = entries[i].ccti_increase;
3658 p->entries[i].ccti_timer = cpu_to_be16(entries[i].ccti_timer);
3659 p->entries[i].trigger_threshold =
3660 entries[i].trigger_threshold;
3661 p->entries[i].ccti_min = entries[i].ccti_min;
3662 }
3663
3664 rcu_read_unlock();
3665
3666 if (resp_len)
3667 *resp_len += sizeof(*p);
3668
3669 return reply((struct ib_mad_hdr *)smp);
3670}
3671
f036780b
DL		 3672/*
3673 * Apply congestion control information stored in the ppd to the
3674 * active structure.
3675 */
3676static void apply_cc_state(struct hfi1_pportdata *ppd)
3677{
3678 struct cc_state *old_cc_state, *new_cc_state;
3679
3680 new_cc_state = kzalloc(sizeof(*new_cc_state), GFP_KERNEL);
3681 if (!new_cc_state)
3682 return;
3683
3684 /*
3685 * Hold the lock for updating *and* to prevent ppd information
3686 * from changing during the update.
3687 */
3688 spin_lock(&ppd->cc_state_lock);
3689
8adf71fa 3690 old_cc_state = get_cc_state_protected(ppd);
f036780b
DL		 3691 if (!old_cc_state) {
3692 /* never active, or shutting down */
3693 spin_unlock(&ppd->cc_state_lock);
3694 kfree(new_cc_state);
3695 return;
3696 }
3697
3698 *new_cc_state = *old_cc_state;
3699
3700 new_cc_state->cct.ccti_limit = ppd->total_cct_entry - 1;
3701 memcpy(new_cc_state->cct.entries, ppd->ccti_entries,
3702 ppd->total_cct_entry * sizeof(struct ib_cc_table_entry));
3703
3704 new_cc_state->cong_setting.port_control = IB_CC_CCS_PC_SL_BASED;
3705 new_cc_state->cong_setting.control_map = ppd->cc_sl_control_map;
3706 memcpy(new_cc_state->cong_setting.entries, ppd->congestion_entries,
3707 OPA_MAX_SLS * sizeof(struct opa_congestion_setting_entry));
3708
3709 rcu_assign_pointer(ppd->cc_state, new_cc_state);
3710
3711 spin_unlock(&ppd->cc_state_lock);
3712
476d95bd 3713 kfree_rcu(old_cc_state, rcu);
f036780b
DL		 3714}
3715
77241056
MM		 3716static int __subn_set_opa_cong_setting(struct opa_smp *smp, u32 am, u8 *data,
3717 struct ib_device *ibdev, u8 port,
f1685179 3718 u32 *resp_len, u32 max_len)
77241056
MM		 3719{
3720 struct opa_congestion_setting_attr *p =
50e5dcbe 3721 (struct opa_congestion_setting_attr *)data;
77241056
MM		 3722 struct hfi1_ibport *ibp = to_iport(ibdev, port);
3723 struct hfi1_pportdata *ppd = ppd_from_ibp(ibp);
3724 struct opa_congestion_setting_entry_shadow *entries;
3725 int i;
3726
f1685179
ND		 3727 if (smp_length_check(sizeof(*p), max_len)) {
3728 smp->status |= IB_SMP_INVALID_FIELD;
3729 return reply((struct ib_mad_hdr *)smp);
3730 }
3731
f036780b
DL		 3732 /*
3733 * Save details from packet into the ppd. Hold the cc_state_lock so
3734 * our information is consistent with anyone trying to apply the state.
3735 */
3736 spin_lock(&ppd->cc_state_lock);
77241056
MM		 3737 ppd->cc_sl_control_map = be32_to_cpu(p->control_map);
3738
3739 entries = ppd->congestion_entries;
3740 for (i = 0; i < OPA_MAX_SLS; i++) {
3741 entries[i].ccti_increase = p->entries[i].ccti_increase;
3742 entries[i].ccti_timer = be16_to_cpu(p->entries[i].ccti_timer);
3743 entries[i].trigger_threshold =
3744 p->entries[i].trigger_threshold;
3745 entries[i].ccti_min = p->entries[i].ccti_min;
3746 }
f036780b
DL		 3747 spin_unlock(&ppd->cc_state_lock);
3748
3749 /* now apply the information */
3750 apply_cc_state(ppd);
77241056
MM		 3751
3752 return __subn_get_opa_cong_setting(smp, am, data, ibdev, port,
f1685179 3753 resp_len, max_len);
77241056
MM		 3754}
3755
3756static int __subn_get_opa_hfi1_cong_log(struct opa_smp *smp, u32 am,
3757 u8 *data, struct ib_device *ibdev,
f1685179 3758 u8 port, u32 *resp_len, u32 max_len)
77241056
MM		 3759{
3760 struct hfi1_ibport *ibp = to_iport(ibdev, port);
3761 struct hfi1_pportdata *ppd = ppd_from_ibp(ibp);
3762 struct opa_hfi1_cong_log *cong_log = (struct opa_hfi1_cong_log *)data;
d61ea075 3763 u64 ts;
77241056
MM		 3764 int i;
3765
f1685179 3766 if (am || smp_length_check(sizeof(*cong_log), max_len)) {
77241056
MM		 3767 smp->status |= IB_SMP_INVALID_FIELD;
3768 return reply((struct ib_mad_hdr *)smp);
3769 }
3770
b77d713a 3771 spin_lock_irq(&ppd->cc_log_lock);
77241056
MM		 3772
3773 cong_log->log_type = OPA_CC_LOG_TYPE_HFI;
3774 cong_log->congestion_flags = 0;
3775 cong_log->threshold_event_counter =
3776 cpu_to_be16(ppd->threshold_event_counter);
3777 memcpy(cong_log->threshold_cong_event_map,
3778 ppd->threshold_cong_event_map,
3779 sizeof(cong_log->threshold_cong_event_map));
3780 /* keep timestamp in units of 1.024 usec */
d61ea075 3781 ts = ktime_get_ns() / 1024;
77241056
MM		 3782 cong_log->current_time_stamp = cpu_to_be32(ts);
3783 for (i = 0; i < OPA_CONG_LOG_ELEMS; i++) {
3784 struct opa_hfi1_cong_log_event_internal *cce =
3785 &ppd->cc_events[ppd->cc_mad_idx++];
3786 if (ppd->cc_mad_idx == OPA_CONG_LOG_ELEMS)
3787 ppd->cc_mad_idx = 0;
3788 /*
3789 * Entries which are older than twice the time
3790 * required to wrap the counter are supposed to
3791 * be zeroed (CA10-49 IBTA, release 1.2.1, V1).
3792 */
d61ea075 3793 if ((ts - cce->timestamp) / 2 > U32_MAX)
77241056
MM		 3794 continue;
3795 memcpy(cong_log->events[i].local_qp_cn_entry, &cce->lqpn, 3);
3796 memcpy(cong_log->events[i].remote_qp_number_cn_entry,
17fb4f29 3797 &cce->rqpn, 3);
77241056
MM		 3798 cong_log->events[i].sl_svc_type_cn_entry =
3799 ((cce->sl & 0x1f) << 3) | (cce->svc_type & 0x7);
3800 cong_log->events[i].remote_lid_cn_entry =
3801 cpu_to_be32(cce->rlid);
3802 cong_log->events[i].timestamp_cn_entry =
3803 cpu_to_be32(cce->timestamp);
3804 }
3805
3806 /*
3807 * Reset threshold_cong_event_map, and threshold_event_counter
3808 * to 0 when log is read.
3809 */
3810 memset(ppd->threshold_cong_event_map, 0x0,
3811 sizeof(ppd->threshold_cong_event_map));
3812 ppd->threshold_event_counter = 0;
3813
b77d713a 3814 spin_unlock_irq(&ppd->cc_log_lock);
77241056
MM		 3815
3816 if (resp_len)
3817 *resp_len += sizeof(struct opa_hfi1_cong_log);
3818
3819 return reply((struct ib_mad_hdr *)smp);
3820}
3821
3822static int __subn_get_opa_cc_table(struct opa_smp *smp, u32 am, u8 *data,
3823 struct ib_device *ibdev, u8 port,
f1685179 3824 u32 *resp_len, u32 max_len)
77241056
MM		 3825{
3826 struct ib_cc_table_attr *cc_table_attr =
50e5dcbe 3827 (struct ib_cc_table_attr *)data;
77241056
MM		 3828 struct hfi1_ibport *ibp = to_iport(ibdev, port);
3829 struct hfi1_pportdata *ppd = ppd_from_ibp(ibp);
3830 u32 start_block = OPA_AM_START_BLK(am);
3831 u32 n_blocks = OPA_AM_NBLK(am);
3832 struct ib_cc_table_entry_shadow *entries;
3833 int i, j;
3834 u32 sentry, eentry;
3835 struct cc_state *cc_state;
f1685179 3836 u32 size = sizeof(u16) * (IB_CCT_ENTRIES * n_blocks + 1);
77241056
MM		 3837
3838 /* sanity check n_blocks, start_block */
f1685179 3839 if (n_blocks == 0 || smp_length_check(size, max_len) ||
77241056
MM		 3840 start_block + n_blocks > ppd->cc_max_table_entries) {
3841 smp->status |= IB_SMP_INVALID_FIELD;
3842 return reply((struct ib_mad_hdr *)smp);
3843 }
3844
3845 rcu_read_lock();
3846
3847 cc_state = get_cc_state(ppd);
3848
d125a6c6 3849 if (!cc_state) {
77241056
MM		 3850 rcu_read_unlock();
3851 return reply((struct ib_mad_hdr *)smp);
3852 }
3853
3854 sentry = start_block * IB_CCT_ENTRIES;
3855 eentry = sentry + (IB_CCT_ENTRIES * n_blocks);
3856
3857 cc_table_attr->ccti_limit = cpu_to_be16(cc_state->cct.ccti_limit);
3858
3859 entries = cc_state->cct.entries;
3860
3861 /* return n_blocks, though the last block may not be full */
3862 for (j = 0, i = sentry; i < eentry; j++, i++)
3863 cc_table_attr->ccti_entries[j].entry =
3864 cpu_to_be16(entries[i].entry);
3865
3866 rcu_read_unlock();
3867
3868 if (resp_len)
f1685179 3869 *resp_len += size;
77241056
MM		 3870
3871 return reply((struct ib_mad_hdr *)smp);
3872}
3873
77241056
MM		 3874static int __subn_set_opa_cc_table(struct opa_smp *smp, u32 am, u8 *data,
3875 struct ib_device *ibdev, u8 port,
f1685179 3876 u32 *resp_len, u32 max_len)
77241056 3877{
50e5dcbe 3878 struct ib_cc_table_attr *p = (struct ib_cc_table_attr *)data;
77241056
MM		 3879 struct hfi1_ibport *ibp = to_iport(ibdev, port);
3880 struct hfi1_pportdata *ppd = ppd_from_ibp(ibp);
3881 u32 start_block = OPA_AM_START_BLK(am);
3882 u32 n_blocks = OPA_AM_NBLK(am);
3883 struct ib_cc_table_entry_shadow *entries;
3884 int i, j;
3885 u32 sentry, eentry;
3886 u16 ccti_limit;
f1685179 3887 u32 size = sizeof(u16) * (IB_CCT_ENTRIES * n_blocks + 1);
77241056
MM		 3888
3889 /* sanity check n_blocks, start_block */
f1685179 3890 if (n_blocks == 0 || smp_length_check(size, max_len) ||
77241056
MM		 3891 start_block + n_blocks > ppd->cc_max_table_entries) {
3892 smp->status |= IB_SMP_INVALID_FIELD;
3893 return reply((struct ib_mad_hdr *)smp);
3894 }
3895
3896 sentry = start_block * IB_CCT_ENTRIES;
3897 eentry = sentry + ((n_blocks - 1) * IB_CCT_ENTRIES) +
3898 (be16_to_cpu(p->ccti_limit)) % IB_CCT_ENTRIES + 1;
3899
3900 /* sanity check ccti_limit */
3901 ccti_limit = be16_to_cpu(p->ccti_limit);
3902 if (ccti_limit + 1 > eentry) {
3903 smp->status |= IB_SMP_INVALID_FIELD;
3904 return reply((struct ib_mad_hdr *)smp);
3905 }
3906
f036780b
DL		 3907 /*
3908 * Save details from packet into the ppd. Hold the cc_state_lock so
3909 * our information is consistent with anyone trying to apply the state.
3910 */
77241056 3911 spin_lock(&ppd->cc_state_lock);
77241056 3912 ppd->total_cct_entry = ccti_limit + 1;
f036780b 3913 entries = ppd->ccti_entries;
77241056
MM		 3914 for (j = 0, i = sentry; i < eentry; j++, i++)
3915 entries[i].entry = be16_to_cpu(p->ccti_entries[j].entry);
77241056
MM		 3916 spin_unlock(&ppd->cc_state_lock);
3917
f036780b
DL		 3918 /* now apply the information */
3919 apply_cc_state(ppd);
77241056 3920
f1685179
ND		 3921 return __subn_get_opa_cc_table(smp, am, data, ibdev, port, resp_len,
3922 max_len);
77241056
MM		 3923}
3924
3925struct opa_led_info {
3926 __be32 rsvd_led_mask;
3927 __be32 rsvd;
3928};
3929
3930#define OPA_LED_SHIFT 31
349ac71f 3931#define OPA_LED_MASK BIT(OPA_LED_SHIFT)
77241056
MM		 3932
3933static int __subn_get_opa_led_info(struct opa_smp *smp, u32 am, u8 *data,
3934 struct ib_device *ibdev, u8 port,
f1685179 3935 u32 *resp_len, u32 max_len)
77241056
MM		 3936{
3937 struct hfi1_devdata *dd = dd_from_ibdev(ibdev);
409b1462 3938 struct hfi1_pportdata *ppd = dd->pport;
50e5dcbe 3939 struct opa_led_info *p = (struct opa_led_info *)data;
77241056 3940 u32 nport = OPA_AM_NPORT(am);
409b1462 3941 u32 is_beaconing_active;
77241056 3942
f1685179 3943 if (nport != 1 || smp_length_check(sizeof(*p), max_len)) {
77241056
MM		 3944 smp->status |= IB_SMP_INVALID_FIELD;
3945 return reply((struct ib_mad_hdr *)smp);
3946 }
3947
409b1462 3948 /*
2243472e
EH		 3949 * This pairs with the memory barrier in hfi1_start_led_override to
3950 * ensure that we read the correct state of LED beaconing represented
3951 * by led_override_timer_active
409b1462 3952 */
2243472e 3953 smp_rmb();
409b1462
EH		 3954 is_beaconing_active = !!atomic_read(&ppd->led_override_timer_active);
3955 p->rsvd_led_mask = cpu_to_be32(is_beaconing_active << OPA_LED_SHIFT);
77241056
MM		 3956
3957 if (resp_len)
3958 *resp_len += sizeof(struct opa_led_info);
3959
3960 return reply((struct ib_mad_hdr *)smp);
3961}
3962
3963static int __subn_set_opa_led_info(struct opa_smp *smp, u32 am, u8 *data,
3964 struct ib_device *ibdev, u8 port,
f1685179 3965 u32 *resp_len, u32 max_len)
77241056
MM		 3966{
3967 struct hfi1_devdata *dd = dd_from_ibdev(ibdev);
50e5dcbe 3968 struct opa_led_info *p = (struct opa_led_info *)data;
77241056
MM		 3969 u32 nport = OPA_AM_NPORT(am);
3970 int on = !!(be32_to_cpu(p->rsvd_led_mask) & OPA_LED_MASK);
3971
f1685179 3972 if (nport != 1 || smp_length_check(sizeof(*p), max_len)) {
77241056
MM		 3973 smp->status |= IB_SMP_INVALID_FIELD;
3974 return reply((struct ib_mad_hdr *)smp);
3975 }
3976
91ab4ed3 3977 if (on)
2243472e 3978 hfi1_start_led_override(dd->pport, 2000, 1500);
91ab4ed3 3979 else
2243472e 3980 shutdown_led_override(dd->pport);
77241056 3981
f1685179
ND		 3982 return __subn_get_opa_led_info(smp, am, data, ibdev, port, resp_len,
3983 max_len);
77241056
MM		 3984}
3985
3986static int subn_get_opa_sma(__be16 attr_id, struct opa_smp *smp, u32 am,
3987 u8 *data, struct ib_device *ibdev, u8 port,
f1685179 3988 u32 *resp_len, u32 max_len)
77241056
MM		 3989{
3990 int ret;
3991 struct hfi1_ibport *ibp = to_iport(ibdev, port);
3992
3993 switch (attr_id) {
3994 case IB_SMP_ATTR_NODE_DESC:
3995 ret = __subn_get_opa_nodedesc(smp, am, data, ibdev, port,
f1685179 3996 resp_len, max_len);
77241056
MM		 3997 break;
3998 case IB_SMP_ATTR_NODE_INFO:
3999 ret = __subn_get_opa_nodeinfo(smp, am, data, ibdev, port,
f1685179 4000 resp_len, max_len);
77241056
MM		 4001 break;
4002 case IB_SMP_ATTR_PORT_INFO:
4003 ret = __subn_get_opa_portinfo(smp, am, data, ibdev, port,
f1685179 4004 resp_len, max_len);
77241056
MM		 4005 break;
4006 case IB_SMP_ATTR_PKEY_TABLE:
4007 ret = __subn_get_opa_pkeytable(smp, am, data, ibdev, port,
f1685179 4008 resp_len, max_len);
77241056
MM		 4009 break;
4010 case OPA_ATTRIB_ID_SL_TO_SC_MAP:
4011 ret = __subn_get_opa_sl_to_sc(smp, am, data, ibdev, port,
f1685179 4012 resp_len, max_len);
77241056
MM		 4013 break;
4014 case OPA_ATTRIB_ID_SC_TO_SL_MAP:
4015 ret = __subn_get_opa_sc_to_sl(smp, am, data, ibdev, port,
f1685179 4016 resp_len, max_len);
77241056
MM		 4017 break;
4018 case OPA_ATTRIB_ID_SC_TO_VLT_MAP:
4019 ret = __subn_get_opa_sc_to_vlt(smp, am, data, ibdev, port,
f1685179 4020 resp_len, max_len);
77241056
MM		 4021 break;
4022 case OPA_ATTRIB_ID_SC_TO_VLNT_MAP:
4023 ret = __subn_get_opa_sc_to_vlnt(smp, am, data, ibdev, port,
f1685179 4024 resp_len, max_len);
77241056
MM		 4025 break;
4026 case OPA_ATTRIB_ID_PORT_STATE_INFO:
4027 ret = __subn_get_opa_psi(smp, am, data, ibdev, port,
f1685179 4028 resp_len, max_len);
77241056
MM		 4029 break;
4030 case OPA_ATTRIB_ID_BUFFER_CONTROL_TABLE:
4031 ret = __subn_get_opa_bct(smp, am, data, ibdev, port,
f1685179 4032 resp_len, max_len);
77241056
MM		 4033 break;
4034 case OPA_ATTRIB_ID_CABLE_INFO:
4035 ret = __subn_get_opa_cable_info(smp, am, data, ibdev, port,
f1685179 4036 resp_len, max_len);
77241056
MM		 4037 break;
4038 case IB_SMP_ATTR_VL_ARB_TABLE:
4039 ret = __subn_get_opa_vl_arb(smp, am, data, ibdev, port,
f1685179 4040 resp_len, max_len);
77241056
MM		 4041 break;
4042 case OPA_ATTRIB_ID_CONGESTION_INFO:
4043 ret = __subn_get_opa_cong_info(smp, am, data, ibdev, port,
f1685179 4044 resp_len, max_len);
77241056
MM		 4045 break;
4046 case OPA_ATTRIB_ID_HFI_CONGESTION_SETTING:
4047 ret = __subn_get_opa_cong_setting(smp, am, data, ibdev,
f1685179 4048 port, resp_len, max_len);
77241056
MM		 4049 break;
4050 case OPA_ATTRIB_ID_HFI_CONGESTION_LOG:
4051 ret = __subn_get_opa_hfi1_cong_log(smp, am, data, ibdev,
f1685179 4052 port, resp_len, max_len);
77241056
MM		 4053 break;
4054 case OPA_ATTRIB_ID_CONGESTION_CONTROL_TABLE:
4055 ret = __subn_get_opa_cc_table(smp, am, data, ibdev, port,
f1685179 4056 resp_len, max_len);
77241056
MM		 4057 break;
4058 case IB_SMP_ATTR_LED_INFO:
4059 ret = __subn_get_opa_led_info(smp, am, data, ibdev, port,
f1685179 4060 resp_len, max_len);
77241056
MM		 4061 break;
4062 case IB_SMP_ATTR_SM_INFO:
4eb06882 4063 if (ibp->rvp.port_cap_flags & IB_PORT_SM_DISABLED)
77241056 4064 return IB_MAD_RESULT_SUCCESS | IB_MAD_RESULT_CONSUMED;
4eb06882 4065 if (ibp->rvp.port_cap_flags & IB_PORT_SM)
77241056
MM		 4066 return IB_MAD_RESULT_SUCCESS;
4067 /* FALLTHROUGH */
4068 default:
4069 smp->status |= IB_SMP_UNSUP_METH_ATTR;
4070 ret = reply((struct ib_mad_hdr *)smp);
4071 break;
4072 }
4073 return ret;
4074}
4075
4076static int subn_set_opa_sma(__be16 attr_id, struct opa_smp *smp, u32 am,
4077 u8 *data, struct ib_device *ibdev, u8 port,
f1685179 4078 u32 *resp_len, u32 max_len)
77241056
MM		 4079{
4080 int ret;
4081 struct hfi1_ibport *ibp = to_iport(ibdev, port);
4082
4083 switch (attr_id) {
4084 case IB_SMP_ATTR_PORT_INFO:
4085 ret = __subn_set_opa_portinfo(smp, am, data, ibdev, port,
f1685179 4086 resp_len, max_len);
77241056
MM		 4087 break;
4088 case IB_SMP_ATTR_PKEY_TABLE:
4089 ret = __subn_set_opa_pkeytable(smp, am, data, ibdev, port,
f1685179 4090 resp_len, max_len);
77241056
MM		 4091 break;
4092 case OPA_ATTRIB_ID_SL_TO_SC_MAP:
4093 ret = __subn_set_opa_sl_to_sc(smp, am, data, ibdev, port,
f1685179 4094 resp_len, max_len);
77241056
MM		 4095 break;
4096 case OPA_ATTRIB_ID_SC_TO_SL_MAP:
4097 ret = __subn_set_opa_sc_to_sl(smp, am, data, ibdev, port,
f1685179 4098 resp_len, max_len);
77241056
MM		 4099 break;
4100 case OPA_ATTRIB_ID_SC_TO_VLT_MAP:
4101 ret = __subn_set_opa_sc_to_vlt(smp, am, data, ibdev, port,
f1685179 4102 resp_len, max_len);
77241056
MM		 4103 break;
4104 case OPA_ATTRIB_ID_SC_TO_VLNT_MAP:
4105 ret = __subn_set_opa_sc_to_vlnt(smp, am, data, ibdev, port,
f1685179 4106 resp_len, max_len);
77241056
MM		 4107 break;
4108 case OPA_ATTRIB_ID_PORT_STATE_INFO:
4109 ret = __subn_set_opa_psi(smp, am, data, ibdev, port,
f1685179 4110 resp_len, max_len);
77241056
MM		 4111 break;
4112 case OPA_ATTRIB_ID_BUFFER_CONTROL_TABLE:
4113 ret = __subn_set_opa_bct(smp, am, data, ibdev, port,
f1685179 4114 resp_len, max_len);
77241056
MM		 4115 break;
4116 case IB_SMP_ATTR_VL_ARB_TABLE:
4117 ret = __subn_set_opa_vl_arb(smp, am, data, ibdev, port,
f1685179 4118 resp_len, max_len);
77241056
MM		 4119 break;
4120 case OPA_ATTRIB_ID_HFI_CONGESTION_SETTING:
4121 ret = __subn_set_opa_cong_setting(smp, am, data, ibdev,
f1685179 4122 port, resp_len, max_len);
77241056
MM		 4123 break;
4124 case OPA_ATTRIB_ID_CONGESTION_CONTROL_TABLE:
4125 ret = __subn_set_opa_cc_table(smp, am, data, ibdev, port,
f1685179 4126 resp_len, max_len);
77241056
MM		 4127 break;
4128 case IB_SMP_ATTR_LED_INFO:
4129 ret = __subn_set_opa_led_info(smp, am, data, ibdev, port,
f1685179 4130 resp_len, max_len);
77241056
MM		 4131 break;
4132 case IB_SMP_ATTR_SM_INFO:
4eb06882 4133 if (ibp->rvp.port_cap_flags & IB_PORT_SM_DISABLED)
77241056 4134 return IB_MAD_RESULT_SUCCESS | IB_MAD_RESULT_CONSUMED;
4eb06882 4135 if (ibp->rvp.port_cap_flags & IB_PORT_SM)
77241056
MM		 4136 return IB_MAD_RESULT_SUCCESS;
4137 /* FALLTHROUGH */
4138 default:
4139 smp->status |= IB_SMP_UNSUP_METH_ATTR;
4140 ret = reply((struct ib_mad_hdr *)smp);
4141 break;
4142 }
4143 return ret;
4144}
4145
4146static inline void set_aggr_error(struct opa_aggregate *ag)
4147{
4148 ag->err_reqlength |= cpu_to_be16(0x8000);
4149}
4150
4151static int subn_get_opa_aggregate(struct opa_smp *smp,
4152 struct ib_device *ibdev, u8 port,
4153 u32 *resp_len)
4154{
4155 int i;
4156 u32 num_attr = be32_to_cpu(smp->attr_mod) & 0x000000ff;
4157 u8 *next_smp = opa_get_smp_data(smp);
4158
4159 if (num_attr < 1 || num_attr > 117) {
4160 smp->status |= IB_SMP_INVALID_FIELD;
4161 return reply((struct ib_mad_hdr *)smp);
4162 }
4163
4164 for (i = 0; i < num_attr; i++) {
4165 struct opa_aggregate *agg;
4166 size_t agg_data_len;
4167 size_t agg_size;
4168 u32 am;
4169
4170 agg = (struct opa_aggregate *)next_smp;
4171 agg_data_len = (be16_to_cpu(agg->err_reqlength) & 0x007f) * 8;
4172 agg_size = sizeof(*agg) + agg_data_len;
4173 am = be32_to_cpu(agg->attr_mod);
4174
4175 *resp_len += agg_size;
4176
4177 if (next_smp + agg_size > ((u8 *)smp) + sizeof(*smp)) {
4178 smp->status |= IB_SMP_INVALID_FIELD;
4179 return reply((struct ib_mad_hdr *)smp);
4180 }
4181
4182 /* zero the payload for this segment */
4183 memset(next_smp + sizeof(*agg), 0, agg_data_len);
4184
50e5dcbe 4185 (void)subn_get_opa_sma(agg->attr_id, smp, am, agg->data,
f1685179
ND		 4186 ibdev, port, NULL, (u32)agg_data_len);
4187
4188 if (smp->status & IB_SMP_INVALID_FIELD)
4189 break;
77241056
MM		 4190 if (smp->status & ~IB_SMP_DIRECTION) {
4191 set_aggr_error(agg);
4192 return reply((struct ib_mad_hdr *)smp);
4193 }
4194 next_smp += agg_size;
77241056
MM		 4195 }
4196
4197 return reply((struct ib_mad_hdr *)smp);
4198}
4199
4200static int subn_set_opa_aggregate(struct opa_smp *smp,
4201 struct ib_device *ibdev, u8 port,
4202 u32 *resp_len)
4203{
4204 int i;
4205 u32 num_attr = be32_to_cpu(smp->attr_mod) & 0x000000ff;
4206 u8 *next_smp = opa_get_smp_data(smp);
4207
4208 if (num_attr < 1 || num_attr > 117) {
4209 smp->status |= IB_SMP_INVALID_FIELD;
4210 return reply((struct ib_mad_hdr *)smp);
4211 }
4212
4213 for (i = 0; i < num_attr; i++) {
4214 struct opa_aggregate *agg;
4215 size_t agg_data_len;
4216 size_t agg_size;
4217 u32 am;
4218
4219 agg = (struct opa_aggregate *)next_smp;
4220 agg_data_len = (be16_to_cpu(agg->err_reqlength) & 0x007f) * 8;
4221 agg_size = sizeof(*agg) + agg_data_len;
4222 am = be32_to_cpu(agg->attr_mod);
4223
4224 *resp_len += agg_size;
4225
4226 if (next_smp + agg_size > ((u8 *)smp) + sizeof(*smp)) {
4227 smp->status |= IB_SMP_INVALID_FIELD;
4228 return reply((struct ib_mad_hdr *)smp);
4229 }
4230
50e5dcbe 4231 (void)subn_set_opa_sma(agg->attr_id, smp, am, agg->data,
f1685179
ND		 4232 ibdev, port, NULL, (u32)agg_data_len);
4233 if (smp->status & IB_SMP_INVALID_FIELD)
4234 break;
77241056
MM		 4235 if (smp->status & ~IB_SMP_DIRECTION) {
4236 set_aggr_error(agg);
4237 return reply((struct ib_mad_hdr *)smp);
4238 }
4239 next_smp += agg_size;
77241056
MM		 4240 }
4241
4242 return reply((struct ib_mad_hdr *)smp);
4243}
4244
4245/*
4246 * OPAv1 specifies that, on the transition to link up, these counters
4247 * are cleared:
4248 * PortRcvErrors [*]
4249 * LinkErrorRecovery
4250 * LocalLinkIntegrityErrors
4251 * ExcessiveBufferOverruns [*]
4252 *
4253 * [*] Error info associated with these counters is retained, but the
4254 * error info status is reset to 0.
4255 */
4256void clear_linkup_counters(struct hfi1_devdata *dd)
4257{
4258 /* PortRcvErrors */
4259 write_dev_cntr(dd, C_DC_RCV_ERR, CNTR_INVALID_VL, 0);
4260 dd->err_info_rcvport.status_and_code &= ~OPA_EI_STATUS_SMASK;
4261 /* LinkErrorRecovery */
4262 write_dev_cntr(dd, C_DC_SEQ_CRC_CNT, CNTR_INVALID_VL, 0);
4263 write_dev_cntr(dd, C_DC_REINIT_FROM_PEER_CNT, CNTR_INVALID_VL, 0);
4264 /* LocalLinkIntegrityErrors */
77241056
MM		 4265 write_dev_cntr(dd, C_DC_RX_REPLAY, CNTR_INVALID_VL, 0);
4266 /* ExcessiveBufferOverruns */
4267 write_dev_cntr(dd, C_RCV_OVF, CNTR_INVALID_VL, 0);
4268 dd->rcv_ovfl_cnt = 0;
4269 dd->err_info_xmit_constraint.status &= ~OPA_EI_STATUS_SMASK;
4270}
4271
406310c6
SS		 4272static int is_full_mgmt_pkey_in_table(struct hfi1_ibport *ibp)
4273{
4274 unsigned int i;
4275 struct hfi1_pportdata *ppd = ppd_from_ibp(ibp);
4276
4277 for (i = 0; i < ARRAY_SIZE(ppd->pkeys); ++i)
4278 if (ppd->pkeys[i] == FULL_MGMT_P_KEY)
4279 return 1;
4280
4281 return 0;
4282}
4283
77241056
MM		 4284/*
4285 * is_local_mad() returns 1 if 'mad' is sent from, and destined to the
4286 * local node, 0 otherwise.
4287 */
4288static int is_local_mad(struct hfi1_ibport *ibp, const struct opa_mad *mad,
4289 const struct ib_wc *in_wc)
4290{
4291 struct hfi1_pportdata *ppd = ppd_from_ibp(ibp);
4292 const struct opa_smp *smp = (const struct opa_smp *)mad;
4293
4294 if (smp->mgmt_class == IB_MGMT_CLASS_SUBN_DIRECTED_ROUTE) {
4295 return (smp->hop_cnt == 0 &&
4296 smp->route.dr.dr_slid == OPA_LID_PERMISSIVE &&
4297 smp->route.dr.dr_dlid == OPA_LID_PERMISSIVE);
4298 }
4299
4300 return (in_wc->slid == ppd->lid);
4301}
4302
4303/*
4304 * opa_local_smp_check() should only be called on MADs for which
4305 * is_local_mad() returns true. It applies the SMP checks that are
4306 * specific to SMPs which are sent from, and destined to this node.
4307 * opa_local_smp_check() returns 0 if the SMP passes its checks, 1
4308 * otherwise.
4309 *
4310 * SMPs which arrive from other nodes are instead checked by
4311 * opa_smp_check().
4312 */
4313static int opa_local_smp_check(struct hfi1_ibport *ibp,
4314 const struct ib_wc *in_wc)
4315{
4316 struct hfi1_pportdata *ppd = ppd_from_ibp(ibp);
77241056
MM		 4317 u16 pkey;
4318
4319 if (in_wc->pkey_index >= ARRAY_SIZE(ppd->pkeys))
4320 return 1;
4321
4322 pkey = ppd->pkeys[in_wc->pkey_index];
4323 /*
4324 * We need to do the "node-local" checks specified in OPAv1,
4325 * rev 0.90, section 9.10.26, which are:
4326 * - pkey is 0x7fff, or 0xffff
4327 * - Source QPN == 0 || Destination QPN == 0
4328 * - the MAD header's management class is either
4329 * IB_MGMT_CLASS_SUBN_DIRECTED_ROUTE or
4330 * IB_MGMT_CLASS_SUBN_LID_ROUTED
4331 * - SLID != 0
4332 *
4333 * However, we know (and so don't need to check again) that,
4334 * for local SMPs, the MAD stack passes MADs with:
4335 * - Source QPN of 0
4336 * - MAD mgmt_class is IB_MGMT_CLASS_SUBN_DIRECTED_ROUTE
4337 * - SLID is either: OPA_LID_PERMISSIVE (0xFFFFFFFF), or
4338 * our own port's lid
4339 *
4340 */
4341 if (pkey == LIM_MGMT_P_KEY || pkey == FULL_MGMT_P_KEY)
4342 return 0;
4988be58
DH		 4343 /*
4344 * On OPA devices it is okay to lose the upper 16 bits of LID as this
4345 * information is obtained elsewhere. Mask off the upper 16 bits.
4346 */
4347 ingress_pkey_table_fail(ppd, pkey, ib_lid_cpu16(0xFFFF & in_wc->slid));
77241056
MM		 4348 return 1;
4349}
4350
406310c6
SS		 4351/**
4352 * hfi1_pkey_validation_pma - It validates PKEYs for incoming PMA MAD packets.
4353 * @ibp: IB port data
4354 * @in_mad: MAD packet with header and data
4355 * @in_wc: Work completion data such as source LID, port number, etc.
4356 *
4357 * These are all the possible logic rules for validating a pkey:
4358 *
4359 * a) If pkey neither FULL_MGMT_P_KEY nor LIM_MGMT_P_KEY,
4360 * and NOT self-originated packet:
4361 * Drop MAD packet as it should always be part of the
4362 * management partition unless it's a self-originated packet.
4363 *
4364 * b) If pkey_index -> FULL_MGMT_P_KEY, and LIM_MGMT_P_KEY in pkey table:
4365 * The packet is coming from a management node and the receiving node
4366 * is also a management node, so it is safe for the packet to go through.
4367 *
4368 * c) If pkey_index -> FULL_MGMT_P_KEY, and LIM_MGMT_P_KEY is NOT in pkey table:
4369 * Drop the packet as LIM_MGMT_P_KEY should always be in the pkey table.
4370 * It could be an FM misconfiguration.
4371 *
4372 * d) If pkey_index -> LIM_MGMT_P_KEY and FULL_MGMT_P_KEY is NOT in pkey table:
4373 * It is safe for the packet to go through since a non-management node is
4374 * talking to another non-management node.
4375 *
4376 * e) If pkey_index -> LIM_MGMT_P_KEY and FULL_MGMT_P_KEY in pkey table:
4377 * Drop the packet because a non-management node is talking to a
4378 * management node, and it could be an attack.
4379 *
4380 * For the implementation, these rules can be simplied to only checking
4381 * for (a) and (e). There's no need to check for rule (b) as
4382 * the packet doesn't need to be dropped. Rule (c) is not possible in
4383 * the driver as LIM_MGMT_P_KEY is always in the pkey table.
4384 *
4385 * Return:
4386 * 0 - pkey is okay, -EINVAL it's a bad pkey
4387 */
4388static int hfi1_pkey_validation_pma(struct hfi1_ibport *ibp,
4389 const struct opa_mad *in_mad,
4390 const struct ib_wc *in_wc)
4391{
4392 u16 pkey_value = hfi1_lookup_pkey_value(ibp, in_wc->pkey_index);
4393
4394 /* Rule (a) from above */
4395 if (!is_local_mad(ibp, in_mad, in_wc) &&
4396 pkey_value != LIM_MGMT_P_KEY &&
4397 pkey_value != FULL_MGMT_P_KEY)
4398 return -EINVAL;
4399
4400 /* Rule (e) from above */
4401 if (pkey_value == LIM_MGMT_P_KEY &&
4402 is_full_mgmt_pkey_in_table(ibp))
4403 return -EINVAL;
4404
4405 return 0;
4406}
4407
77241056
MM		 4408static int process_subn_opa(struct ib_device *ibdev, int mad_flags,
4409 u8 port, const struct opa_mad *in_mad,
4410 struct opa_mad *out_mad,
4411 u32 *resp_len)
4412{
4413 struct opa_smp *smp = (struct opa_smp *)out_mad;
4414 struct hfi1_ibport *ibp = to_iport(ibdev, port);
4415 u8 *data;
f1685179 4416 u32 am, data_size;
77241056
MM		 4417 __be16 attr_id;
4418 int ret;
4419
4420 *out_mad = *in_mad;
4421 data = opa_get_smp_data(smp);
f1685179 4422 data_size = (u32)opa_get_smp_data_size(smp);
77241056
MM		 4423
4424 am = be32_to_cpu(smp->attr_mod);
4425 attr_id = smp->attr_id;
9fa240bb 4426 if (smp->class_version != OPA_SM_CLASS_VERSION) {
77241056
MM		 4427 smp->status |= IB_SMP_UNSUP_VERSION;
4428 ret = reply((struct ib_mad_hdr *)smp);
5950e9b1 4429 return ret;
77241056
MM		 4430 }
4431 ret = check_mkey(ibp, (struct ib_mad_hdr *)smp, mad_flags, smp->mkey,
4432 smp->route.dr.dr_slid, smp->route.dr.return_path,
4433 smp->hop_cnt);
4434 if (ret) {
4435 u32 port_num = be32_to_cpu(smp->attr_mod);
4436
4437 /*
4438 * If this is a get/set portinfo, we already check the
4439 * M_Key if the MAD is for another port and the M_Key
4440 * is OK on the receiving port. This check is needed
4441 * to increment the error counters when the M_Key
4442 * fails to match on *both* ports.
4443 */
4444 if (attr_id == IB_SMP_ATTR_PORT_INFO &&
4445 (smp->method == IB_MGMT_METHOD_GET ||
4446 smp->method == IB_MGMT_METHOD_SET) &&
4447 port_num && port_num <= ibdev->phys_port_cnt &&
4448 port != port_num)
50e5dcbe 4449 (void)check_mkey(to_iport(ibdev, port_num),
77241056
MM		 4450 (struct ib_mad_hdr *)smp, 0,
4451 smp->mkey, smp->route.dr.dr_slid,
4452 smp->route.dr.return_path,
4453 smp->hop_cnt);
4454 ret = IB_MAD_RESULT_FAILURE;
5950e9b1 4455 return ret;
77241056
MM		 4456 }
4457
4458 *resp_len = opa_get_smp_header_size(smp);
4459
4460 switch (smp->method) {
4461 case IB_MGMT_METHOD_GET:
4462 switch (attr_id) {
4463 default:
4464 clear_opa_smp_data(smp);
4465 ret = subn_get_opa_sma(attr_id, smp, am, data,
f1685179
ND		 4466 ibdev, port, resp_len,
4467 data_size);
5950e9b1 4468 break;
77241056
MM		 4469 case OPA_ATTRIB_ID_AGGREGATE:
4470 ret = subn_get_opa_aggregate(smp, ibdev, port,
4471 resp_len);
5950e9b1 4472 break;
77241056 4473 }
5950e9b1 4474 break;
77241056
MM		 4475 case IB_MGMT_METHOD_SET:
4476 switch (attr_id) {
4477 default:
4478 ret = subn_set_opa_sma(attr_id, smp, am, data,
f1685179
ND		 4479 ibdev, port, resp_len,
4480 data_size);
5950e9b1 4481 break;
77241056
MM		 4482 case OPA_ATTRIB_ID_AGGREGATE:
4483 ret = subn_set_opa_aggregate(smp, ibdev, port,
4484 resp_len);
5950e9b1 4485 break;
77241056 4486 }
5950e9b1 4487 break;
77241056
MM		 4488 case IB_MGMT_METHOD_TRAP:
4489 case IB_MGMT_METHOD_REPORT:
4490 case IB_MGMT_METHOD_REPORT_RESP:
4491 case IB_MGMT_METHOD_GET_RESP:
4492 /*
4493 * The ib_mad module will call us to process responses
4494 * before checking for other consumers.
4495 * Just tell the caller to process it normally.
4496 */
4497 ret = IB_MAD_RESULT_SUCCESS;
5950e9b1 4498 break;
bf90aadd
MR		 4499 case IB_MGMT_METHOD_TRAP_REPRESS:
4500 subn_handle_opa_trap_repress(ibp, smp);
4501 /* Always successful */
4502 ret = IB_MAD_RESULT_SUCCESS;
4503 break;
77241056
MM		 4504 default:
4505 smp->status |= IB_SMP_UNSUP_METHOD;
4506 ret = reply((struct ib_mad_hdr *)smp);
5950e9b1 4507 break;
77241056
MM		 4508 }
4509
77241056
MM		 4510 return ret;
4511}
4512
4513static int process_subn(struct ib_device *ibdev, int mad_flags,
4514 u8 port, const struct ib_mad *in_mad,
4515 struct ib_mad *out_mad)
4516{
4517 struct ib_smp *smp = (struct ib_smp *)out_mad;
4518 struct hfi1_ibport *ibp = to_iport(ibdev, port);
4519 int ret;
4520
4521 *out_mad = *in_mad;
4522 if (smp->class_version != 1) {
4523 smp->status |= IB_SMP_UNSUP_VERSION;
4524 ret = reply((struct ib_mad_hdr *)smp);
5950e9b1 4525 return ret;
77241056
MM		 4526 }
4527
4528 ret = check_mkey(ibp, (struct ib_mad_hdr *)smp, mad_flags,
4529 smp->mkey, (__force __be32)smp->dr_slid,
4530 smp->return_path, smp->hop_cnt);
4531 if (ret) {
4532 u32 port_num = be32_to_cpu(smp->attr_mod);
4533
4534 /*
4535 * If this is a get/set portinfo, we already check the
4536 * M_Key if the MAD is for another port and the M_Key
4537 * is OK on the receiving port. This check is needed
4538 * to increment the error counters when the M_Key
4539 * fails to match on *both* ports.
4540 */
4541 if (in_mad->mad_hdr.attr_id == IB_SMP_ATTR_PORT_INFO &&
4542 (smp->method == IB_MGMT_METHOD_GET ||
4543 smp->method == IB_MGMT_METHOD_SET) &&
4544 port_num && port_num <= ibdev->phys_port_cnt &&
4545 port != port_num)
50e5dcbe 4546 (void)check_mkey(to_iport(ibdev, port_num),
17fb4f29
JJ		 4547 (struct ib_mad_hdr *)smp, 0,
4548 smp->mkey,
4549 (__force __be32)smp->dr_slid,
4550 smp->return_path, smp->hop_cnt);
77241056 4551 ret = IB_MAD_RESULT_FAILURE;
5950e9b1 4552 return ret;
77241056
MM		 4553 }
4554
4555 switch (smp->method) {
4556 case IB_MGMT_METHOD_GET:
4557 switch (smp->attr_id) {
4558 case IB_SMP_ATTR_NODE_INFO:
4559 ret = subn_get_nodeinfo(smp, ibdev, port);
5950e9b1 4560 break;
77241056
MM		 4561 default:
4562 smp->status |= IB_SMP_UNSUP_METH_ATTR;
4563 ret = reply((struct ib_mad_hdr *)smp);
5950e9b1 4564 break;
77241056 4565 }
5950e9b1 4566 break;
77241056
MM		 4567 }
4568
77241056
MM		 4569 return ret;
4570}
4571
b8d114eb
SS		 4572static int process_perf(struct ib_device *ibdev, u8 port,
4573 const struct ib_mad *in_mad,
4574 struct ib_mad *out_mad)
4575{
4576 struct ib_pma_mad *pmp = (struct ib_pma_mad *)out_mad;
4577 struct ib_class_port_info *cpi = (struct ib_class_port_info *)
4578 &pmp->data;
4579 int ret = IB_MAD_RESULT_FAILURE;
4580
4581 *out_mad = *in_mad;
4582 if (pmp->mad_hdr.class_version != 1) {
4583 pmp->mad_hdr.status |= IB_SMP_UNSUP_VERSION;
4584 ret = reply((struct ib_mad_hdr *)pmp);
4585 return ret;
4586 }
4587
4588 switch (pmp->mad_hdr.method) {
4589 case IB_MGMT_METHOD_GET:
4590 switch (pmp->mad_hdr.attr_id) {
4591 case IB_PMA_PORT_COUNTERS:
4592 ret = pma_get_ib_portcounters(pmp, ibdev, port);
4593 break;
4594 case IB_PMA_PORT_COUNTERS_EXT:
4595 ret = pma_get_ib_portcounters_ext(pmp, ibdev, port);
4596 break;
4597 case IB_PMA_CLASS_PORT_INFO:
4598 cpi->capability_mask = IB_PMA_CLASS_CAP_EXT_WIDTH;
4599 ret = reply((struct ib_mad_hdr *)pmp);
4600 break;
4601 default:
4602 pmp->mad_hdr.status |= IB_SMP_UNSUP_METH_ATTR;
4603 ret = reply((struct ib_mad_hdr *)pmp);
4604 break;
4605 }
4606 break;
4607
4608 case IB_MGMT_METHOD_SET:
4609 if (pmp->mad_hdr.attr_id) {
4610 pmp->mad_hdr.status |= IB_SMP_UNSUP_METH_ATTR;
4611 ret = reply((struct ib_mad_hdr *)pmp);
4612 }
4613 break;
4614
4615 case IB_MGMT_METHOD_TRAP:
4616 case IB_MGMT_METHOD_GET_RESP:
4617 /*
4618 * The ib_mad module will call us to process responses
4619 * before checking for other consumers.
4620 * Just tell the caller to process it normally.
4621 */
4622 ret = IB_MAD_RESULT_SUCCESS;
4623 break;
4624
4625 default:
4626 pmp->mad_hdr.status |= IB_SMP_UNSUP_METHOD;
4627 ret = reply((struct ib_mad_hdr *)pmp);
4628 break;
77241056
MM		 4629 }
4630
77241056
MM		 4631 return ret;
4632}
4633
4634static int process_perf_opa(struct ib_device *ibdev, u8 port,
4635 const struct opa_mad *in_mad,
4636 struct opa_mad *out_mad, u32 *resp_len)
4637{
4638 struct opa_pma_mad *pmp = (struct opa_pma_mad *)out_mad;
4639 int ret;
4640
4641 *out_mad = *in_mad;
4642
9fa240bb 4643 if (pmp->mad_hdr.class_version != OPA_SM_CLASS_VERSION) {
77241056
MM		 4644 pmp->mad_hdr.status |= IB_SMP_UNSUP_VERSION;
4645 return reply((struct ib_mad_hdr *)pmp);
4646 }
4647
4648 *resp_len = sizeof(pmp->mad_hdr);
4649
4650 switch (pmp->mad_hdr.method) {
4651 case IB_MGMT_METHOD_GET:
4652 switch (pmp->mad_hdr.attr_id) {
4653 case IB_PMA_CLASS_PORT_INFO:
4654 ret = pma_get_opa_classportinfo(pmp, ibdev, resp_len);
5950e9b1 4655 break;
77241056
MM		 4656 case OPA_PM_ATTRIB_ID_PORT_STATUS:
4657 ret = pma_get_opa_portstatus(pmp, ibdev, port,
17fb4f29 4658 resp_len);
5950e9b1 4659 break;
77241056
MM		 4660 case OPA_PM_ATTRIB_ID_DATA_PORT_COUNTERS:
4661 ret = pma_get_opa_datacounters(pmp, ibdev, port,
17fb4f29 4662 resp_len);
5950e9b1 4663 break;
77241056
MM		 4664 case OPA_PM_ATTRIB_ID_ERROR_PORT_COUNTERS:
4665 ret = pma_get_opa_porterrors(pmp, ibdev, port,
17fb4f29 4666 resp_len);
5950e9b1 4667 break;
77241056
MM		 4668 case OPA_PM_ATTRIB_ID_ERROR_INFO:
4669 ret = pma_get_opa_errorinfo(pmp, ibdev, port,
17fb4f29 4670 resp_len);
5950e9b1 4671 break;
77241056
MM		 4672 default:
4673 pmp->mad_hdr.status |= IB_SMP_UNSUP_METH_ATTR;
4674 ret = reply((struct ib_mad_hdr *)pmp);
5950e9b1 4675 break;
77241056 4676 }
5950e9b1 4677 break;
77241056
MM		 4678
4679 case IB_MGMT_METHOD_SET:
4680 switch (pmp->mad_hdr.attr_id) {
4681 case OPA_PM_ATTRIB_ID_CLEAR_PORT_STATUS:
4682 ret = pma_set_opa_portstatus(pmp, ibdev, port,
17fb4f29 4683 resp_len);
5950e9b1 4684 break;
77241056
MM		 4685 case OPA_PM_ATTRIB_ID_ERROR_INFO:
4686 ret = pma_set_opa_errorinfo(pmp, ibdev, port,
17fb4f29 4687 resp_len);
5950e9b1 4688 break;
77241056
MM		 4689 default:
4690 pmp->mad_hdr.status |= IB_SMP_UNSUP_METH_ATTR;
4691 ret = reply((struct ib_mad_hdr *)pmp);
5950e9b1 4692 break;
77241056 4693 }
5950e9b1 4694 break;
77241056
MM		 4695
4696 case IB_MGMT_METHOD_TRAP:
4697 case IB_MGMT_METHOD_GET_RESP:
4698 /*
4699 * The ib_mad module will call us to process responses
4700 * before checking for other consumers.
4701 * Just tell the caller to process it normally.
4702 */
4703 ret = IB_MAD_RESULT_SUCCESS;
5950e9b1 4704 break;
77241056
MM		 4705
4706 default:
4707 pmp->mad_hdr.status |= IB_SMP_UNSUP_METHOD;
4708 ret = reply((struct ib_mad_hdr *)pmp);
5950e9b1 4709 break;
77241056
MM		 4710 }
4711
77241056
MM		 4712 return ret;
4713}
4714
4715static int hfi1_process_opa_mad(struct ib_device *ibdev, int mad_flags,
a724648e
JB		 4716 u8 port, const struct ib_wc *in_wc,
4717 const struct ib_grh *in_grh,
4718 const struct opa_mad *in_mad,
4719 struct opa_mad *out_mad, size_t *out_mad_size,
4720 u16 *out_mad_pkey_index)
77241056
MM		 4721{
4722 int ret;
4723 int pkey_idx;
4724 u32 resp_len = 0;
4725 struct hfi1_ibport *ibp = to_iport(ibdev, port);
4726
4727 pkey_idx = hfi1_lookup_pkey_idx(ibp, LIM_MGMT_P_KEY);
4728 if (pkey_idx < 0) {
4729 pr_warn("failed to find limited mgmt pkey, defaulting 0x%x\n",
4730 hfi1_get_pkey(ibp, 1));
4731 pkey_idx = 1;
4732 }
4733 *out_mad_pkey_index = (u16)pkey_idx;
4734
4735 switch (in_mad->mad_hdr.mgmt_class) {
4736 case IB_MGMT_CLASS_SUBN_DIRECTED_ROUTE:
4737 case IB_MGMT_CLASS_SUBN_LID_ROUTED:
4738 if (is_local_mad(ibp, in_mad, in_wc)) {
4739 ret = opa_local_smp_check(ibp, in_wc);
4740 if (ret)
4741 return IB_MAD_RESULT_FAILURE;
4742 }
4743 ret = process_subn_opa(ibdev, mad_flags, port, in_mad,
4744 out_mad, &resp_len);
4745 goto bail;
4746 case IB_MGMT_CLASS_PERF_MGMT:
406310c6
SS		 4747 ret = hfi1_pkey_validation_pma(ibp, in_mad, in_wc);
4748 if (ret)
4749 return IB_MAD_RESULT_FAILURE;
4750
4751 ret = process_perf_opa(ibdev, port, in_mad, out_mad, &resp_len);
77241056
MM		 4752 goto bail;
4753
4754 default:
4755 ret = IB_MAD_RESULT_SUCCESS;
4756 }
4757
4758bail:
4759 if (ret & IB_MAD_RESULT_REPLY)
4760 *out_mad_size = round_up(resp_len, 8);
4761 else if (ret & IB_MAD_RESULT_SUCCESS)
4762 *out_mad_size = in_wc->byte_len - sizeof(struct ib_grh);
4763
4764 return ret;
4765}
4766
4767static int hfi1_process_ib_mad(struct ib_device *ibdev, int mad_flags, u8 port,
4768 const struct ib_wc *in_wc,
4769 const struct ib_grh *in_grh,
4770 const struct ib_mad *in_mad,
4771 struct ib_mad *out_mad)
4772{
4773 int ret;
4774
4775 switch (in_mad->mad_hdr.mgmt_class) {
4776 case IB_MGMT_CLASS_SUBN_DIRECTED_ROUTE:
4777 case IB_MGMT_CLASS_SUBN_LID_ROUTED:
4778 ret = process_subn(ibdev, mad_flags, port, in_mad, out_mad);
5950e9b1 4779 break;
b8d114eb
SS		 4780 case IB_MGMT_CLASS_PERF_MGMT:
4781 ret = process_perf(ibdev, port, in_mad, out_mad);
4782 break;
77241056
MM		 4783 default:
4784 ret = IB_MAD_RESULT_SUCCESS;
5950e9b1 4785 break;
77241056
MM		 4786 }
4787
77241056
MM		 4788 return ret;
4789}
4790
4791/**
4792 * hfi1_process_mad - process an incoming MAD packet
4793 * @ibdev: the infiniband device this packet came in on
4794 * @mad_flags: MAD flags
4795 * @port: the port number this packet came in on
4796 * @in_wc: the work completion entry for this packet
4797 * @in_grh: the global route header for this packet
4798 * @in_mad: the incoming MAD
4799 * @out_mad: any outgoing MAD reply
4800 *
4801 * Returns IB_MAD_RESULT_SUCCESS if this is a MAD that we are not
4802 * interested in processing.
4803 *
4804 * Note that the verbs framework has already done the MAD sanity checks,
4805 * and hop count/pointer updating for IB_MGMT_CLASS_SUBN_DIRECTED_ROUTE
4806 * MADs.
4807 *
4808 * This is called by the ib_mad module.
4809 */
4810int hfi1_process_mad(struct ib_device *ibdev, int mad_flags, u8 port,
4811 const struct ib_wc *in_wc, const struct ib_grh *in_grh,
4812 const struct ib_mad_hdr *in_mad, size_t in_mad_size,
4813 struct ib_mad_hdr *out_mad, size_t *out_mad_size,
4814 u16 *out_mad_pkey_index)
4815{
4816 switch (in_mad->base_version) {
4817 case OPA_MGMT_BASE_VERSION:
4818 if (unlikely(in_mad_size != sizeof(struct opa_mad))) {
3067771c 4819 dev_err(ibdev->dev.parent, "invalid in_mad_size\n");
77241056
MM		 4820 return IB_MAD_RESULT_FAILURE;
4821 }
4822 return hfi1_process_opa_mad(ibdev, mad_flags, port,
4823 in_wc, in_grh,
4824 (struct opa_mad *)in_mad,
4825 (struct opa_mad *)out_mad,
4826 out_mad_size,
4827 out_mad_pkey_index);
4828 case IB_MGMT_BASE_VERSION:
4829 return hfi1_process_ib_mad(ibdev, mad_flags, port,
4830 in_wc, in_grh,
4831 (const struct ib_mad *)in_mad,
4832 (struct ib_mad *)out_mad);
4833 default:
4834 break;
4835 }
4836
4837 return IB_MAD_RESULT_FAILURE;
4838}
Linux 6.x block layer and io_uring tree(s)