projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
staging/rdma/hfi1: Support query gid in rdmavt
[linux-2.6-block.git] / drivers / staging / rdma / hfi1 / mad.c
CommitLineData
77241056
MM		 1/*
2 *
3 * This file is provided under a dual BSD/GPLv2 license. When using or
4 * redistributing this file, you may do so under either license.
5 *
6 * GPL LICENSE SUMMARY
7 *
91ab4ed3 8 * Copyright(c) 2015, 2016 Intel Corporation.
77241056
MM		 9 *
10 * This program is free software; you can redistribute it and/or modify
11 * it under the terms of version 2 of the GNU General Public License as
12 * published by the Free Software Foundation.
13 *
14 * This program is distributed in the hope that it will be useful, but
15 * WITHOUT ANY WARRANTY; without even the implied warranty of
16 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
17 * General Public License for more details.
18 *
19 * BSD LICENSE
20 *
91ab4ed3 21 * Copyright(c) 2015, 2016 Intel Corporation.
77241056
MM		 22 *
23 * Redistribution and use in source and binary forms, with or without
24 * modification, are permitted provided that the following conditions
25 * are met:
26 *
27 * - Redistributions of source code must retain the above copyright
28 * notice, this list of conditions and the following disclaimer.
29 * - Redistributions in binary form must reproduce the above copyright
30 * notice, this list of conditions and the following disclaimer in
31 * the documentation and/or other materials provided with the
32 * distribution.
33 * - Neither the name of Intel Corporation nor the names of its
34 * contributors may be used to endorse or promote products derived
35 * from this software without specific prior written permission.
36 *
37 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
38 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
39 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
40 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
41 * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
42 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
43 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
44 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
45 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
46 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
47 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
48 *
49 */
50
51#include <linux/net.h>
52#define OPA_NUM_PKEY_BLOCKS_PER_SMP (OPA_SMP_DR_DATA_SIZE \
53 / (OPA_PARTITION_TABLE_BLK_SIZE * sizeof(u16)))
54
55#include "hfi.h"
56#include "mad.h"
57#include "trace.h"
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
63static int reply(struct ib_mad_hdr *smp)
64{
65 /*
66 * The verbs framework will handle the directed/LID route
67 * packet changes.
68 */
69 smp->method = IB_MGMT_METHOD_GET_RESP;
70 if (smp->mgmt_class == IB_MGMT_CLASS_SUBN_DIRECTED_ROUTE)
71 smp->status |= IB_SMP_DIRECTION;
72 return IB_MAD_RESULT_SUCCESS | IB_MAD_RESULT_REPLY;
73}
74
75static inline void clear_opa_smp_data(struct opa_smp *smp)
76{
77 void *data = opa_get_smp_data(smp);
78 size_t size = opa_get_smp_data_size(smp);
79
80 memset(data, 0, size);
81}
82
83static void send_trap(struct hfi1_ibport *ibp, void *data, unsigned len)
84{
85 struct ib_mad_send_buf *send_buf;
86 struct ib_mad_agent *agent;
5cd24119 87 struct opa_smp *smp;
77241056
MM		 88 int ret;
89 unsigned long flags;
90 unsigned long timeout;
91 int pkey_idx;
92 u32 qpn = ppd_from_ibp(ibp)->sm_trap_qp;
93
4eb06882 94 agent = ibp->rvp.send_agent;
77241056
MM		 95 if (!agent)
96 return;
97
98 /* o14-3.2.1 */
99 if (ppd_from_ibp(ibp)->lstate != IB_PORT_ACTIVE)
100 return;
101
102 /* o14-2 */
4eb06882
DD		 103 if (ibp->rvp.trap_timeout && time_before(jiffies,
104 ibp->rvp.trap_timeout))
77241056
MM		 105 return;
106
107 pkey_idx = hfi1_lookup_pkey_idx(ibp, LIM_MGMT_P_KEY);
108 if (pkey_idx < 0) {
109 pr_warn("%s: failed to find limited mgmt pkey, defaulting 0x%x\n",
110 __func__, hfi1_get_pkey(ibp, 1));
111 pkey_idx = 1;
112 }
113
114 send_buf = ib_create_send_mad(agent, qpn, pkey_idx, 0,
115 IB_MGMT_MAD_HDR, IB_MGMT_MAD_DATA,
116 GFP_ATOMIC, IB_MGMT_BASE_VERSION);
117 if (IS_ERR(send_buf))
118 return;
119
120 smp = send_buf->mad;
5cd24119 121 smp->base_version = OPA_MGMT_BASE_VERSION;
77241056 122 smp->mgmt_class = IB_MGMT_CLASS_SUBN_LID_ROUTED;
5cd24119 123 smp->class_version = OPA_SMI_CLASS_VERSION;
77241056 124 smp->method = IB_MGMT_METHOD_TRAP;
4eb06882
DD		 125 ibp->rvp.tid++;
126 smp->tid = cpu_to_be64(ibp->rvp.tid);
77241056
MM		 127 smp->attr_id = IB_SMP_ATTR_NOTICE;
128 /* o14-1: smp->mkey = 0; */
5cd24119 129 memcpy(smp->route.lid.data, data, len);
77241056 130
4eb06882 131 spin_lock_irqsave(&ibp->rvp.lock, flags);
9c4a311e 132 if (!ibp->rvp.sm_ah) {
4eb06882 133 if (ibp->rvp.sm_lid != be16_to_cpu(IB_LID_PERMISSIVE)) {
77241056
MM		 134 struct ib_ah *ah;
135
4eb06882 136 ah = hfi1_create_qp0_ah(ibp, ibp->rvp.sm_lid);
77241056
MM		 137 if (IS_ERR(ah))
138 ret = PTR_ERR(ah);
139 else {
140 send_buf->ah = ah;
9c4a311e 141 ibp->rvp.sm_ah = ibah_to_rvtah(ah);
77241056
MM		 142 ret = 0;
143 }
144 } else
145 ret = -EINVAL;
146 } else {
9c4a311e 147 send_buf->ah = &ibp->rvp.sm_ah->ibah;
77241056
MM		 148 ret = 0;
149 }
4eb06882 150 spin_unlock_irqrestore(&ibp->rvp.lock, flags);
77241056
MM		 151
152 if (!ret)
153 ret = ib_post_send_mad(send_buf, NULL);
154 if (!ret) {
155 /* 4.096 usec. */
4eb06882
DD		 156 timeout = (4096 * (1UL << ibp->rvp.subnet_timeout)) / 1000;
157 ibp->rvp.trap_timeout = jiffies + usecs_to_jiffies(timeout);
77241056
MM		 158 } else {
159 ib_free_send_mad(send_buf);
4eb06882 160 ibp->rvp.trap_timeout = 0;
77241056
MM		 161 }
162}
163
164/*
165 * Send a bad [PQ]_Key trap (ch. 14.3.8).
166 */
167void hfi1_bad_pqkey(struct hfi1_ibport *ibp, __be16 trap_num, u32 key, u32 sl,
5cd24119 168 u32 qp1, u32 qp2, u16 lid1, u16 lid2)
77241056 169{
5cd24119
EK		 170 struct opa_mad_notice_attr data;
171 u32 lid = ppd_from_ibp(ibp)->lid;
172 u32 _lid1 = lid1;
173 u32 _lid2 = lid2;
77241056 174
5cd24119
EK		 175 memset(&data, 0, sizeof(data));
176
177 if (trap_num == OPA_TRAP_BAD_P_KEY)
4eb06882 178 ibp->rvp.pkey_violations++;
77241056 179 else
4eb06882
DD		 180 ibp->rvp.qkey_violations++;
181 ibp->rvp.n_pkt_drops++;
77241056
MM		 182
183 /* Send violation trap */
184 data.generic_type = IB_NOTICE_TYPE_SECURITY;
77241056
MM		 185 data.prod_type_lsb = IB_NOTICE_PROD_CA;
186 data.trap_num = trap_num;
5cd24119
EK		 187 data.issuer_lid = cpu_to_be32(lid);
188 data.ntc_257_258.lid1 = cpu_to_be32(_lid1);
189 data.ntc_257_258.lid2 = cpu_to_be32(_lid2);
190 data.ntc_257_258.key = cpu_to_be32(key);
191 data.ntc_257_258.sl = sl << 3;
192 data.ntc_257_258.qp1 = cpu_to_be32(qp1);
193 data.ntc_257_258.qp2 = cpu_to_be32(qp2);
77241056
MM		 194
195 send_trap(ibp, &data, sizeof(data));
196}
197
198/*
199 * Send a bad M_Key trap (ch. 14.3.9).
200 */
201static void bad_mkey(struct hfi1_ibport *ibp, struct ib_mad_hdr *mad,
202 __be64 mkey, __be32 dr_slid, u8 return_path[], u8 hop_cnt)
203{
5cd24119
EK		 204 struct opa_mad_notice_attr data;
205 u32 lid = ppd_from_ibp(ibp)->lid;
77241056 206
5cd24119 207 memset(&data, 0, sizeof(data));
77241056
MM		 208 /* Send violation trap */
209 data.generic_type = IB_NOTICE_TYPE_SECURITY;
77241056 210 data.prod_type_lsb = IB_NOTICE_PROD_CA;
5cd24119
EK		 211 data.trap_num = OPA_TRAP_BAD_M_KEY;
212 data.issuer_lid = cpu_to_be32(lid);
213 data.ntc_256.lid = data.issuer_lid;
214 data.ntc_256.method = mad->method;
215 data.ntc_256.attr_id = mad->attr_id;
216 data.ntc_256.attr_mod = mad->attr_mod;
217 data.ntc_256.mkey = mkey;
77241056 218 if (mad->mgmt_class == IB_MGMT_CLASS_SUBN_DIRECTED_ROUTE) {
5cd24119
EK		 219 data.ntc_256.dr_slid = dr_slid;
220 data.ntc_256.dr_trunc_hop = IB_NOTICE_TRAP_DR_NOTICE;
221 if (hop_cnt > ARRAY_SIZE(data.ntc_256.dr_rtn_path)) {
222 data.ntc_256.dr_trunc_hop |=
77241056 223 IB_NOTICE_TRAP_DR_TRUNC;
5cd24119 224 hop_cnt = ARRAY_SIZE(data.ntc_256.dr_rtn_path);
77241056 225 }
5cd24119
EK		 226 data.ntc_256.dr_trunc_hop |= hop_cnt;
227 memcpy(data.ntc_256.dr_rtn_path, return_path,
77241056
MM		 228 hop_cnt);
229 }
230
231 send_trap(ibp, &data, sizeof(data));
232}
233
234/*
235 * Send a Port Capability Mask Changed trap (ch. 14.3.11).
236 */
237void hfi1_cap_mask_chg(struct hfi1_ibport *ibp)
238{
5cd24119
EK		 239 struct opa_mad_notice_attr data;
240 u32 lid = ppd_from_ibp(ibp)->lid;
241
242 memset(&data, 0, sizeof(data));
77241056
MM		 243
244 data.generic_type = IB_NOTICE_TYPE_INFO;
77241056 245 data.prod_type_lsb = IB_NOTICE_PROD_CA;
5cd24119
EK		 246 data.trap_num = OPA_TRAP_CHANGE_CAPABILITY;
247 data.issuer_lid = cpu_to_be32(lid);
248 data.ntc_144.lid = data.issuer_lid;
4eb06882 249 data.ntc_144.new_cap_mask = cpu_to_be32(ibp->rvp.port_cap_flags);
77241056
MM		 250
251 send_trap(ibp, &data, sizeof(data));
252}
253
254/*
255 * Send a System Image GUID Changed trap (ch. 14.3.12).
256 */
257void hfi1_sys_guid_chg(struct hfi1_ibport *ibp)
258{
5cd24119
EK		 259 struct opa_mad_notice_attr data;
260 u32 lid = ppd_from_ibp(ibp)->lid;
261
262 memset(&data, 0, sizeof(data));
77241056
MM		 263
264 data.generic_type = IB_NOTICE_TYPE_INFO;
77241056 265 data.prod_type_lsb = IB_NOTICE_PROD_CA;
5cd24119
EK		 266 data.trap_num = OPA_TRAP_CHANGE_SYSGUID;
267 data.issuer_lid = cpu_to_be32(lid);
268 data.ntc_145.new_sys_guid = ib_hfi1_sys_image_guid;
269 data.ntc_145.lid = data.issuer_lid;
77241056
MM		 270
271 send_trap(ibp, &data, sizeof(data));
272}
273
274/*
275 * Send a Node Description Changed trap (ch. 14.3.13).
276 */
277void hfi1_node_desc_chg(struct hfi1_ibport *ibp)
278{
5cd24119
EK		 279 struct opa_mad_notice_attr data;
280 u32 lid = ppd_from_ibp(ibp)->lid;
281
282 memset(&data, 0, sizeof(data));
77241056
MM		 283
284 data.generic_type = IB_NOTICE_TYPE_INFO;
77241056 285 data.prod_type_lsb = IB_NOTICE_PROD_CA;
5cd24119
EK		 286 data.trap_num = OPA_TRAP_CHANGE_CAPABILITY;
287 data.issuer_lid = cpu_to_be32(lid);
288 data.ntc_144.lid = data.issuer_lid;
289 data.ntc_144.change_flags =
290 cpu_to_be16(OPA_NOTICE_TRAP_NODE_DESC_CHG);
77241056
MM		 291
292 send_trap(ibp, &data, sizeof(data));
293}
294
295static int __subn_get_opa_nodedesc(struct opa_smp *smp, u32 am,
296 u8 *data, struct ib_device *ibdev,
297 u8 port, u32 *resp_len)
298{
299 struct opa_node_description *nd;
300
301 if (am) {
302 smp->status |= IB_SMP_INVALID_FIELD;
303 return reply((struct ib_mad_hdr *)smp);
304 }
305
306 nd = (struct opa_node_description *)data;
307
308 memcpy(nd->data, ibdev->node_desc, sizeof(nd->data));
309
310 if (resp_len)
311 *resp_len += sizeof(*nd);
312
313 return reply((struct ib_mad_hdr *)smp);
314}
315
316static int __subn_get_opa_nodeinfo(struct opa_smp *smp, u32 am, u8 *data,
317 struct ib_device *ibdev, u8 port,
318 u32 *resp_len)
319{
320 struct opa_node_info *ni;
321 struct hfi1_devdata *dd = dd_from_ibdev(ibdev);
322 unsigned pidx = port - 1; /* IB number port from 1, hw from 0 */
323
324 ni = (struct opa_node_info *)data;
325
326 /* GUID 0 is illegal */
327 if (am || pidx >= dd->num_pports || dd->pport[pidx].guid == 0) {
328 smp->status |= IB_SMP_INVALID_FIELD;
329 return reply((struct ib_mad_hdr *)smp);
330 }
331
332 ni->port_guid = cpu_to_be64(dd->pport[pidx].guid);
333 ni->base_version = OPA_MGMT_BASE_VERSION;
334 ni->class_version = OPA_SMI_CLASS_VERSION;
335 ni->node_type = 1; /* channel adapter */
336 ni->num_ports = ibdev->phys_port_cnt;
337 /* This is already in network order */
338 ni->system_image_guid = ib_hfi1_sys_image_guid;
339 /* Use first-port GUID as node */
340 ni->node_guid = cpu_to_be64(dd->pport->guid);
341 ni->partition_cap = cpu_to_be16(hfi1_get_npkeys(dd));
342 ni->device_id = cpu_to_be16(dd->pcidev->device);
343 ni->revision = cpu_to_be32(dd->minrev);
344 ni->local_port_num = port;
345 ni->vendor_id[0] = dd->oui1;
346 ni->vendor_id[1] = dd->oui2;
347 ni->vendor_id[2] = dd->oui3;
348
349 if (resp_len)
350 *resp_len += sizeof(*ni);
351
352 return reply((struct ib_mad_hdr *)smp);
353}
354
355static int subn_get_nodeinfo(struct ib_smp *smp, struct ib_device *ibdev,
356 u8 port)
357{
358 struct ib_node_info *nip = (struct ib_node_info *)&smp->data;
359 struct hfi1_devdata *dd = dd_from_ibdev(ibdev);
360 unsigned pidx = port - 1; /* IB number port from 1, hw from 0 */
361
362 /* GUID 0 is illegal */
363 if (smp->attr_mod || pidx >= dd->num_pports ||
364 dd->pport[pidx].guid == 0)
365 smp->status |= IB_SMP_INVALID_FIELD;
366 else
367 nip->port_guid = cpu_to_be64(dd->pport[pidx].guid);
368
369 nip->base_version = OPA_MGMT_BASE_VERSION;
370 nip->class_version = OPA_SMI_CLASS_VERSION;
371 nip->node_type = 1; /* channel adapter */
372 nip->num_ports = ibdev->phys_port_cnt;
373 /* This is already in network order */
374 nip->sys_guid = ib_hfi1_sys_image_guid;
375 /* Use first-port GUID as node */
376 nip->node_guid = cpu_to_be64(dd->pport->guid);
377 nip->partition_cap = cpu_to_be16(hfi1_get_npkeys(dd));
378 nip->device_id = cpu_to_be16(dd->pcidev->device);
379 nip->revision = cpu_to_be32(dd->minrev);
380 nip->local_port_num = port;
381 nip->vendor_id[0] = dd->oui1;
382 nip->vendor_id[1] = dd->oui2;
383 nip->vendor_id[2] = dd->oui3;
384
385 return reply((struct ib_mad_hdr *)smp);
386}
387
388static void set_link_width_enabled(struct hfi1_pportdata *ppd, u32 w)
389{
390 (void)hfi1_set_ib_cfg(ppd, HFI1_IB_CFG_LWID_ENB, w);
391}
392
393static void set_link_width_downgrade_enabled(struct hfi1_pportdata *ppd, u32 w)
394{
395 (void)hfi1_set_ib_cfg(ppd, HFI1_IB_CFG_LWID_DG_ENB, w);
396}
397
398static void set_link_speed_enabled(struct hfi1_pportdata *ppd, u32 s)
399{
400 (void)hfi1_set_ib_cfg(ppd, HFI1_IB_CFG_SPD_ENB, s);
401}
402
403static int check_mkey(struct hfi1_ibport *ibp, struct ib_mad_hdr *mad,
404 int mad_flags, __be64 mkey, __be32 dr_slid,
405 u8 return_path[], u8 hop_cnt)
406{
407 int valid_mkey = 0;
408 int ret = 0;
409
410 /* Is the mkey in the process of expiring? */
4eb06882
DD		 411 if (ibp->rvp.mkey_lease_timeout &&
412 time_after_eq(jiffies, ibp->rvp.mkey_lease_timeout)) {
77241056 413 /* Clear timeout and mkey protection field. */
4eb06882
DD		 414 ibp->rvp.mkey_lease_timeout = 0;
415 ibp->rvp.mkeyprot = 0;
77241056
MM		 416 }
417
4eb06882
DD		 418 if ((mad_flags & IB_MAD_IGNORE_MKEY) || ibp->rvp.mkey == 0 ||
419 ibp->rvp.mkey == mkey)
77241056
MM		 420 valid_mkey = 1;
421
422 /* Unset lease timeout on any valid Get/Set/TrapRepress */
4eb06882 423 if (valid_mkey && ibp->rvp.mkey_lease_timeout &&
77241056
MM		 424 (mad->method == IB_MGMT_METHOD_GET ||
425 mad->method == IB_MGMT_METHOD_SET ||
426 mad->method == IB_MGMT_METHOD_TRAP_REPRESS))
4eb06882 427 ibp->rvp.mkey_lease_timeout = 0;
77241056
MM		 428
429 if (!valid_mkey) {
430 switch (mad->method) {
431 case IB_MGMT_METHOD_GET:
432 /* Bad mkey not a violation below level 2 */
4eb06882 433 if (ibp->rvp.mkeyprot < 2)
77241056
MM		 434 break;
435 case IB_MGMT_METHOD_SET:
436 case IB_MGMT_METHOD_TRAP_REPRESS:
4eb06882
DD		 437 if (ibp->rvp.mkey_violations != 0xFFFF)
438 ++ibp->rvp.mkey_violations;
439 if (!ibp->rvp.mkey_lease_timeout &&
440 ibp->rvp.mkey_lease_period)
441 ibp->rvp.mkey_lease_timeout = jiffies +
442 ibp->rvp.mkey_lease_period * HZ;
77241056
MM		 443 /* Generate a trap notice. */
444 bad_mkey(ibp, mad, mkey, dr_slid, return_path,
445 hop_cnt);
446 ret = 1;
447 }
448 }
449
450 return ret;
451}
452
453/*
454 * The SMA caches reads from LCB registers in case the LCB is unavailable.
455 * (The LCB is unavailable in certain link states, for example.)
456 */
457struct lcb_datum {
458 u32 off;
459 u64 val;
460};
461
462static struct lcb_datum lcb_cache[] = {
463 { DC_LCB_STS_ROUND_TRIP_LTP_CNT, 0 },
464};
465
466static int write_lcb_cache(u32 off, u64 val)
467{
468 int i;
469
470 for (i = 0; i < ARRAY_SIZE(lcb_cache); i++) {
471 if (lcb_cache[i].off == off) {
472 lcb_cache[i].val = val;
473 return 0;
474 }
475 }
476
477 pr_warn("%s bad offset 0x%x\n", __func__, off);
478 return -1;
479}
480
481static int read_lcb_cache(u32 off, u64 *val)
482{
483 int i;
484
485 for (i = 0; i < ARRAY_SIZE(lcb_cache); i++) {
486 if (lcb_cache[i].off == off) {
487 *val = lcb_cache[i].val;
488 return 0;
489 }
490 }
491
492 pr_warn("%s bad offset 0x%x\n", __func__, off);
493 return -1;
494}
495
496void read_ltp_rtt(struct hfi1_devdata *dd)
497{
498 u64 reg;
499
500 if (read_lcb_csr(dd, DC_LCB_STS_ROUND_TRIP_LTP_CNT, &reg))
501 dd_dev_err(dd, "%s: unable to read LTP RTT\n", __func__);
502 else
503 write_lcb_cache(DC_LCB_STS_ROUND_TRIP_LTP_CNT, reg);
504}
505
77241056
MM		 506static int __subn_get_opa_portinfo(struct opa_smp *smp, u32 am, u8 *data,
507 struct ib_device *ibdev, u8 port,
508 u32 *resp_len)
509{
510 int i;
511 struct hfi1_devdata *dd;
512 struct hfi1_pportdata *ppd;
513 struct hfi1_ibport *ibp;
514 struct opa_port_info *pi = (struct opa_port_info *)data;
515 u8 mtu;
516 u8 credit_rate;
517 u32 state;
518 u32 num_ports = OPA_AM_NPORT(am);
519 u32 start_of_sm_config = OPA_AM_START_SM_CFG(am);
520 u32 buffer_units;
521 u64 tmp = 0;
522
523 if (num_ports != 1) {
524 smp->status |= IB_SMP_INVALID_FIELD;
525 return reply((struct ib_mad_hdr *)smp);
526 }
527
528 dd = dd_from_ibdev(ibdev);
529 /* IB numbers ports from 1, hw from 0 */
530 ppd = dd->pport + (port - 1);
531 ibp = &ppd->ibport_data;
532
533 if (ppd->vls_supported/2 > ARRAY_SIZE(pi->neigh_mtu.pvlx_to_mtu) ||
534 ppd->vls_supported > ARRAY_SIZE(dd->vld)) {
535 smp->status |= IB_SMP_INVALID_FIELD;
536 return reply((struct ib_mad_hdr *)smp);
537 }
538
539 pi->lid = cpu_to_be32(ppd->lid);
540
541 /* Only return the mkey if the protection field allows it. */
542 if (!(smp->method == IB_MGMT_METHOD_GET &&
4eb06882
DD		 543 ibp->rvp.mkey != smp->mkey &&
544 ibp->rvp.mkeyprot == 1))
545 pi->mkey = ibp->rvp.mkey;
546
547 pi->subnet_prefix = ibp->rvp.gid_prefix;
548 pi->sm_lid = cpu_to_be32(ibp->rvp.sm_lid);
549 pi->ib_cap_mask = cpu_to_be32(ibp->rvp.port_cap_flags);
550 pi->mkey_lease_period = cpu_to_be16(ibp->rvp.mkey_lease_period);
77241056
MM		 551 pi->sm_trap_qp = cpu_to_be32(ppd->sm_trap_qp);
552 pi->sa_qp = cpu_to_be32(ppd->sa_qp);
553
554 pi->link_width.enabled = cpu_to_be16(ppd->link_width_enabled);
555 pi->link_width.supported = cpu_to_be16(ppd->link_width_supported);
556 pi->link_width.active = cpu_to_be16(ppd->link_width_active);
557
558 pi->link_width_downgrade.supported =
559 cpu_to_be16(ppd->link_width_downgrade_supported);
560 pi->link_width_downgrade.enabled =
561 cpu_to_be16(ppd->link_width_downgrade_enabled);
562 pi->link_width_downgrade.tx_active =
563 cpu_to_be16(ppd->link_width_downgrade_tx_active);
564 pi->link_width_downgrade.rx_active =
565 cpu_to_be16(ppd->link_width_downgrade_rx_active);
566
567 pi->link_speed.supported = cpu_to_be16(ppd->link_speed_supported);
568 pi->link_speed.active = cpu_to_be16(ppd->link_speed_active);
569 pi->link_speed.enabled = cpu_to_be16(ppd->link_speed_enabled);
570
571 state = driver_lstate(ppd);
572
573 if (start_of_sm_config && (state == IB_PORT_INIT))
574 ppd->is_sm_config_started = 1;
575
1d01cf33 576 pi->port_phys_conf = (ppd->port_type & 0xf);
77241056
MM		 577
578#if PI_LED_ENABLE_SUP
579 pi->port_states.ledenable_offlinereason = ppd->neighbor_normal << 4;
580 pi->port_states.ledenable_offlinereason |=
581 ppd->is_sm_config_started << 5;
582 pi->port_states.ledenable_offlinereason |=
a9c05e35 583 ppd->offline_disabled_reason;
77241056
MM		 584#else
585 pi->port_states.offline_reason = ppd->neighbor_normal << 4;
586 pi->port_states.offline_reason |= ppd->is_sm_config_started << 5;
a9c05e35 587 pi->port_states.offline_reason |= ppd->offline_disabled_reason;
77241056
MM		 588#endif /* PI_LED_ENABLE_SUP */
589
590 pi->port_states.portphysstate_portstate =
591 (hfi1_ibphys_portstate(ppd) << 4) | state;
592
4eb06882 593 pi->mkeyprotect_lmc = (ibp->rvp.mkeyprot << 6) | ppd->lmc;
77241056
MM		 594
595 memset(pi->neigh_mtu.pvlx_to_mtu, 0, sizeof(pi->neigh_mtu.pvlx_to_mtu));
596 for (i = 0; i < ppd->vls_supported; i++) {
597 mtu = mtu_to_enum(dd->vld[i].mtu, HFI1_DEFAULT_ACTIVE_MTU);
598 if ((i % 2) == 0)
599 pi->neigh_mtu.pvlx_to_mtu[i/2] |= (mtu << 4);
600 else
601 pi->neigh_mtu.pvlx_to_mtu[i/2] |= mtu;
602 }
603 /* don't forget VL 15 */
604 mtu = mtu_to_enum(dd->vld[15].mtu, 2048);
605 pi->neigh_mtu.pvlx_to_mtu[15/2] |= mtu;
4eb06882 606 pi->smsl = ibp->rvp.sm_sl & OPA_PI_MASK_SMSL;
77241056
MM		 607 pi->operational_vls = hfi1_get_ib_cfg(ppd, HFI1_IB_CFG_OP_VLS);
608 pi->partenforce_filterraw |=
609 (ppd->linkinit_reason & OPA_PI_MASK_LINKINIT_REASON);
610 if (ppd->part_enforce & HFI1_PART_ENFORCE_IN)
611 pi->partenforce_filterraw |= OPA_PI_MASK_PARTITION_ENFORCE_IN;
612 if (ppd->part_enforce & HFI1_PART_ENFORCE_OUT)
613 pi->partenforce_filterraw |= OPA_PI_MASK_PARTITION_ENFORCE_OUT;
4eb06882 614 pi->mkey_violations = cpu_to_be16(ibp->rvp.mkey_violations);
77241056 615 /* P_KeyViolations are counted by hardware. */
4eb06882
DD		 616 pi->pkey_violations = cpu_to_be16(ibp->rvp.pkey_violations);
617 pi->qkey_violations = cpu_to_be16(ibp->rvp.qkey_violations);
77241056
MM		 618
619 pi->vl.cap = ppd->vls_supported;
4eb06882 620 pi->vl.high_limit = cpu_to_be16(ibp->rvp.vl_high_limit);
77241056
MM		 621 pi->vl.arb_high_cap = (u8)hfi1_get_ib_cfg(ppd, HFI1_IB_CFG_VL_HIGH_CAP);
622 pi->vl.arb_low_cap = (u8)hfi1_get_ib_cfg(ppd, HFI1_IB_CFG_VL_LOW_CAP);
623
4eb06882 624 pi->clientrereg_subnettimeout = ibp->rvp.subnet_timeout;
77241056
MM		 625
626 pi->port_link_mode = cpu_to_be16(OPA_PORT_LINK_MODE_OPA << 10 |
627 OPA_PORT_LINK_MODE_OPA << 5 |
628 OPA_PORT_LINK_MODE_OPA);
629
630 pi->port_ltp_crc_mode = cpu_to_be16(ppd->port_ltp_crc_mode);
631
632 pi->port_mode = cpu_to_be16(
633 ppd->is_active_optimize_enabled ?
634 OPA_PI_MASK_PORT_ACTIVE_OPTOMIZE : 0);
635
636 pi->port_packet_format.supported =
637 cpu_to_be16(OPA_PORT_PACKET_FORMAT_9B);
638 pi->port_packet_format.enabled =
639 cpu_to_be16(OPA_PORT_PACKET_FORMAT_9B);
640
641 /* flit_control.interleave is (OPA V1, version .76):
642 * bits use
643 * ---- ---
644 * 2 res
645 * 2 DistanceSupported
646 * 2 DistanceEnabled
647 * 5 MaxNextLevelTxEnabled
648 * 5 MaxNestLevelRxSupported
649 *
650 * HFI supports only "distance mode 1" (see OPA V1, version .76,
651 * section 9.6.2), so set DistanceSupported, DistanceEnabled
652 * to 0x1.
653 */
654 pi->flit_control.interleave = cpu_to_be16(0x1400);
655
656 pi->link_down_reason = ppd->local_link_down_reason.sma;
657 pi->neigh_link_down_reason = ppd->neigh_link_down_reason.sma;
658 pi->port_error_action = cpu_to_be32(ppd->port_error_action);
659 pi->mtucap = mtu_to_enum(hfi1_max_mtu, IB_MTU_4096);
660
661 /* 32.768 usec. response time (guessing) */
662 pi->resptimevalue = 3;
663
664 pi->local_port_num = port;
665
666 /* buffer info for FM */
667 pi->overall_buffer_space = cpu_to_be16(dd->link_credits);
668
669 pi->neigh_node_guid = cpu_to_be64(ppd->neighbor_guid);
670 pi->neigh_port_num = ppd->neighbor_port_number;
671 pi->port_neigh_mode =
672 (ppd->neighbor_type & OPA_PI_MASK_NEIGH_NODE_TYPE) |
673 (ppd->mgmt_allowed ? OPA_PI_MASK_NEIGH_MGMT_ALLOWED : 0) |
674 (ppd->neighbor_fm_security ?
675 OPA_PI_MASK_NEIGH_FW_AUTH_BYPASS : 0);
676
677 /* HFIs shall always return VL15 credits to their
678 * neighbor in a timely manner, without any credit return pacing.
679 */
680 credit_rate = 0;
681 buffer_units = (dd->vau) & OPA_PI_MASK_BUF_UNIT_BUF_ALLOC;
682 buffer_units |= (dd->vcu << 3) & OPA_PI_MASK_BUF_UNIT_CREDIT_ACK;
683 buffer_units |= (credit_rate << 6) &
684 OPA_PI_MASK_BUF_UNIT_VL15_CREDIT_RATE;
685 buffer_units |= (dd->vl15_init << 11) & OPA_PI_MASK_BUF_UNIT_VL15_INIT;
686 pi->buffer_units = cpu_to_be32(buffer_units);
687
688 pi->opa_cap_mask = cpu_to_be16(OPA_CAP_MASK3_IsSharedSpaceSupported);
689
690 /* HFI supports a replay buffer 128 LTPs in size */
691 pi->replay_depth.buffer = 0x80;
692 /* read the cached value of DC_LCB_STS_ROUND_TRIP_LTP_CNT */
693 read_lcb_cache(DC_LCB_STS_ROUND_TRIP_LTP_CNT, &tmp);
694
695 /* this counter is 16 bits wide, but the replay_depth.wire
696 * variable is only 8 bits */
697 if (tmp > 0xff)
698 tmp = 0xff;
699 pi->replay_depth.wire = tmp;
700
701 if (resp_len)
702 *resp_len += sizeof(struct opa_port_info);
703
704 return reply((struct ib_mad_hdr *)smp);
705}
706
707/**
708 * get_pkeys - return the PKEY table
709 * @dd: the hfi1_ib device
710 * @port: the IB port number
711 * @pkeys: the pkey table is placed here
712 */
713static int get_pkeys(struct hfi1_devdata *dd, u8 port, u16 *pkeys)
714{
715 struct hfi1_pportdata *ppd = dd->pport + port - 1;
716
717 memcpy(pkeys, ppd->pkeys, sizeof(ppd->pkeys));
718
719 return 0;
720}
721
722static int __subn_get_opa_pkeytable(struct opa_smp *smp, u32 am, u8 *data,
723 struct ib_device *ibdev, u8 port,
724 u32 *resp_len)
725{
726 struct hfi1_devdata *dd = dd_from_ibdev(ibdev);
727 u32 n_blocks_req = OPA_AM_NBLK(am);
728 u32 start_block = am & 0x7ff;
729 __be16 *p;
730 u16 *q;
731 int i;
732 u16 n_blocks_avail;
733 unsigned npkeys = hfi1_get_npkeys(dd);
734 size_t size;
735
736 if (n_blocks_req == 0) {
737 pr_warn("OPA Get PKey AM Invalid : P = %d; B = 0x%x; N = 0x%x\n",
738 port, start_block, n_blocks_req);
739 smp->status |= IB_SMP_INVALID_FIELD;
740 return reply((struct ib_mad_hdr *)smp);
741 }
742
743 n_blocks_avail = (u16) (npkeys/OPA_PARTITION_TABLE_BLK_SIZE) + 1;
744
745 size = (n_blocks_req * OPA_PARTITION_TABLE_BLK_SIZE) * sizeof(u16);
746
747 if (start_block + n_blocks_req > n_blocks_avail ||
748 n_blocks_req > OPA_NUM_PKEY_BLOCKS_PER_SMP) {
749 pr_warn("OPA Get PKey AM Invalid : s 0x%x; req 0x%x; "
750 "avail 0x%x; blk/smp 0x%lx\n",
751 start_block, n_blocks_req, n_blocks_avail,
752 OPA_NUM_PKEY_BLOCKS_PER_SMP);
753 smp->status |= IB_SMP_INVALID_FIELD;
754 return reply((struct ib_mad_hdr *)smp);
755 }
756
757 p = (__be16 *) data;
758 q = (u16 *)data;
759 /* get the real pkeys if we are requesting the first block */
760 if (start_block == 0) {
761 get_pkeys(dd, port, q);
762 for (i = 0; i < npkeys; i++)
763 p[i] = cpu_to_be16(q[i]);
764 if (resp_len)
765 *resp_len += size;
766 } else
767 smp->status |= IB_SMP_INVALID_FIELD;
768
769 return reply((struct ib_mad_hdr *)smp);
770}
771
772enum {
773 HFI_TRANSITION_DISALLOWED,
774 HFI_TRANSITION_IGNORED,
775 HFI_TRANSITION_ALLOWED,
776 HFI_TRANSITION_UNDEFINED,
777};
778
779/*
780 * Use shortened names to improve readability of
781 * {logical,physical}_state_transitions
782 */
783enum {
784 __D = HFI_TRANSITION_DISALLOWED,
785 __I = HFI_TRANSITION_IGNORED,
786 __A = HFI_TRANSITION_ALLOWED,
787 __U = HFI_TRANSITION_UNDEFINED,
788};
789
790/*
791 * IB_PORTPHYSSTATE_POLLING (2) through OPA_PORTPHYSSTATE_MAX (11) are
792 * represented in physical_state_transitions.
793 */
794#define __N_PHYSTATES (OPA_PORTPHYSSTATE_MAX - IB_PORTPHYSSTATE_POLLING + 1)
795
796/*
797 * Within physical_state_transitions, rows represent "old" states,
798 * columns "new" states, and physical_state_transitions.allowed[old][new]
799 * indicates if the transition from old state to new state is legal (see
800 * OPAg1v1, Table 6-4).
801 */
802static const struct {
803 u8 allowed[__N_PHYSTATES][__N_PHYSTATES];
804} physical_state_transitions = {
805 {
806 /* 2 3 4 5 6 7 8 9 10 11 */
807 /* 2 */ { __A, __A, __D, __D, __D, __D, __D, __D, __D, __D },
808 /* 3 */ { __A, __I, __D, __D, __D, __D, __D, __D, __D, __A },
809 /* 4 */ { __U, __U, __U, __U, __U, __U, __U, __U, __U, __U },
810 /* 5 */ { __A, __A, __D, __I, __D, __D, __D, __D, __D, __D },
811 /* 6 */ { __U, __U, __U, __U, __U, __U, __U, __U, __U, __U },
812 /* 7 */ { __D, __A, __D, __D, __D, __I, __D, __D, __D, __D },
813 /* 8 */ { __U, __U, __U, __U, __U, __U, __U, __U, __U, __U },
814 /* 9 */ { __I, __A, __D, __D, __D, __D, __D, __I, __D, __D },
815 /*10 */ { __U, __U, __U, __U, __U, __U, __U, __U, __U, __U },
816 /*11 */ { __D, __A, __D, __D, __D, __D, __D, __D, __D, __I },
817 }
818};
819
820/*
821 * IB_PORT_DOWN (1) through IB_PORT_ACTIVE_DEFER (5) are represented
822 * logical_state_transitions
823 */
824
825#define __N_LOGICAL_STATES (IB_PORT_ACTIVE_DEFER - IB_PORT_DOWN + 1)
826
827/*
828 * Within logical_state_transitions rows represent "old" states,
829 * columns "new" states, and logical_state_transitions.allowed[old][new]
830 * indicates if the transition from old state to new state is legal (see
831 * OPAg1v1, Table 9-12).
832 */
833static const struct {
834 u8 allowed[__N_LOGICAL_STATES][__N_LOGICAL_STATES];
835} logical_state_transitions = {
836 {
837 /* 1 2 3 4 5 */
838 /* 1 */ { __I, __D, __D, __D, __U},
839 /* 2 */ { __D, __I, __A, __D, __U},
840 /* 3 */ { __D, __D, __I, __A, __U},
841 /* 4 */ { __D, __D, __I, __I, __U},
842 /* 5 */ { __U, __U, __U, __U, __U},
843 }
844};
845
846static int logical_transition_allowed(int old, int new)
847{
848 if (old < IB_PORT_NOP || old > IB_PORT_ACTIVE_DEFER ||
849 new < IB_PORT_NOP || new > IB_PORT_ACTIVE_DEFER) {
850 pr_warn("invalid logical state(s) (old %d new %d)\n",
851 old, new);
852 return HFI_TRANSITION_UNDEFINED;
853 }
854
855 if (new == IB_PORT_NOP)
856 return HFI_TRANSITION_ALLOWED; /* always allowed */
857
858 /* adjust states for indexing into logical_state_transitions */
859 old -= IB_PORT_DOWN;
860 new -= IB_PORT_DOWN;
861
862 if (old < 0 || new < 0)
863 return HFI_TRANSITION_UNDEFINED;
864 return logical_state_transitions.allowed[old][new];
865}
866
867static int physical_transition_allowed(int old, int new)
868{
869 if (old < IB_PORTPHYSSTATE_NOP || old > OPA_PORTPHYSSTATE_MAX ||
870 new < IB_PORTPHYSSTATE_NOP || new > OPA_PORTPHYSSTATE_MAX) {
871 pr_warn("invalid physical state(s) (old %d new %d)\n",
872 old, new);
873 return HFI_TRANSITION_UNDEFINED;
874 }
875
876 if (new == IB_PORTPHYSSTATE_NOP)
877 return HFI_TRANSITION_ALLOWED; /* always allowed */
878
879 /* adjust states for indexing into physical_state_transitions */
880 old -= IB_PORTPHYSSTATE_POLLING;
881 new -= IB_PORTPHYSSTATE_POLLING;
882
883 if (old < 0 || new < 0)
884 return HFI_TRANSITION_UNDEFINED;
885 return physical_state_transitions.allowed[old][new];
886}
887
888static int port_states_transition_allowed(struct hfi1_pportdata *ppd,
889 u32 logical_new, u32 physical_new)
890{
891 u32 physical_old = driver_physical_state(ppd);
892 u32 logical_old = driver_logical_state(ppd);
893 int ret, logical_allowed, physical_allowed;
894
895 logical_allowed = ret =
896 logical_transition_allowed(logical_old, logical_new);
897
898 if (ret == HFI_TRANSITION_DISALLOWED ||
899 ret == HFI_TRANSITION_UNDEFINED) {
900 pr_warn("invalid logical state transition %s -> %s\n",
901 opa_lstate_name(logical_old),
902 opa_lstate_name(logical_new));
903 return ret;
904 }
905
906 physical_allowed = ret =
907 physical_transition_allowed(physical_old, physical_new);
908
909 if (ret == HFI_TRANSITION_DISALLOWED ||
910 ret == HFI_TRANSITION_UNDEFINED) {
911 pr_warn("invalid physical state transition %s -> %s\n",
912 opa_pstate_name(physical_old),
913 opa_pstate_name(physical_new));
914 return ret;
915 }
916
917 if (logical_allowed == HFI_TRANSITION_IGNORED &&
918 physical_allowed == HFI_TRANSITION_IGNORED)
919 return HFI_TRANSITION_IGNORED;
920
a9c05e35
BM		 921 /*
922 * A change request of Physical Port State from
923 * 'Offline' to 'Polling' should be ignored.
924 */
925 if ((physical_old == OPA_PORTPHYSSTATE_OFFLINE) &&
926 (physical_new == IB_PORTPHYSSTATE_POLLING))
927 return HFI_TRANSITION_IGNORED;
928
77241056
MM		 929 /*
930 * Either physical_allowed or logical_allowed is
931 * HFI_TRANSITION_ALLOWED.
932 */
933 return HFI_TRANSITION_ALLOWED;
934}
935
936static int set_port_states(struct hfi1_pportdata *ppd, struct opa_smp *smp,
937 u32 logical_state, u32 phys_state,
938 int suppress_idle_sma)
939{
940 struct hfi1_devdata *dd = ppd->dd;
941 u32 link_state;
942 int ret;
943
944 ret = port_states_transition_allowed(ppd, logical_state, phys_state);
945 if (ret == HFI_TRANSITION_DISALLOWED ||
946 ret == HFI_TRANSITION_UNDEFINED) {
947 /* error message emitted above */
948 smp->status |= IB_SMP_INVALID_FIELD;
949 return 0;
950 }
951
952 if (ret == HFI_TRANSITION_IGNORED)
953 return 0;
954
955 if ((phys_state != IB_PORTPHYSSTATE_NOP) &&
956 !(logical_state == IB_PORT_DOWN ||
957 logical_state == IB_PORT_NOP)){
958 pr_warn("SubnSet(OPA_PortInfo) port state invalid: logical_state 0x%x physical_state 0x%x\n",
959 logical_state, phys_state);
960 smp->status |= IB_SMP_INVALID_FIELD;
961 }
962
963 /*
964 * Logical state changes are summarized in OPAv1g1 spec.,
965 * Table 9-12; physical state changes are summarized in
966 * OPAv1g1 spec., Table 6.4.
967 */
968 switch (logical_state) {
969 case IB_PORT_NOP:
970 if (phys_state == IB_PORTPHYSSTATE_NOP)
971 break;
972 /* FALLTHROUGH */
973 case IB_PORT_DOWN:
974 if (phys_state == IB_PORTPHYSSTATE_NOP)
975 link_state = HLS_DN_DOWNDEF;
976 else if (phys_state == IB_PORTPHYSSTATE_POLLING) {
977 link_state = HLS_DN_POLL;
978 set_link_down_reason(ppd,
979 OPA_LINKDOWN_REASON_FM_BOUNCE, 0,
980 OPA_LINKDOWN_REASON_FM_BOUNCE);
981 } else if (phys_state == IB_PORTPHYSSTATE_DISABLED)
982 link_state = HLS_DN_DISABLE;
983 else {
984 pr_warn("SubnSet(OPA_PortInfo) invalid physical state 0x%x\n",
985 phys_state);
986 smp->status |= IB_SMP_INVALID_FIELD;
987 break;
988 }
989
990 set_link_state(ppd, link_state);
991 if (link_state == HLS_DN_DISABLE &&
992 (ppd->offline_disabled_reason >
a9c05e35 993 HFI1_ODR_MASK(OPA_LINKDOWN_REASON_SMA_DISABLED) ||
77241056 994 ppd->offline_disabled_reason ==
a9c05e35 995 HFI1_ODR_MASK(OPA_LINKDOWN_REASON_NONE)))
77241056 996 ppd->offline_disabled_reason =
a9c05e35 997 HFI1_ODR_MASK(OPA_LINKDOWN_REASON_SMA_DISABLED);
77241056
MM		 998 /*
999 * Don't send a reply if the response would be sent
1000 * through the disabled port.
1001 */
1002 if (link_state == HLS_DN_DISABLE && smp->hop_cnt)
1003 return IB_MAD_RESULT_SUCCESS | IB_MAD_RESULT_CONSUMED;
1004 break;
1005 case IB_PORT_ARMED:
1006 ret = set_link_state(ppd, HLS_UP_ARMED);
1007 if ((ret == 0) && (suppress_idle_sma == 0))
1008 send_idle_sma(dd, SMA_IDLE_ARM);
1009 break;
1010 case IB_PORT_ACTIVE:
1011 if (ppd->neighbor_normal) {
1012 ret = set_link_state(ppd, HLS_UP_ACTIVE);
1013 if (ret == 0)
1014 send_idle_sma(dd, SMA_IDLE_ACTIVE);
1015 } else {
1016 pr_warn("SubnSet(OPA_PortInfo) Cannot move to Active with NeighborNormal 0\n");
1017 smp->status |= IB_SMP_INVALID_FIELD;
1018 }
1019 break;
1020 default:
1021 pr_warn("SubnSet(OPA_PortInfo) invalid logical state 0x%x\n",
1022 logical_state);
1023 smp->status |= IB_SMP_INVALID_FIELD;
1024 }
1025
1026 return 0;
1027}
1028
1029/**
1030 * subn_set_opa_portinfo - set port information
1031 * @smp: the incoming SM packet
1032 * @ibdev: the infiniband device
1033 * @port: the port on the device
1034 *
1035 */
1036static int __subn_set_opa_portinfo(struct opa_smp *smp, u32 am, u8 *data,
1037 struct ib_device *ibdev, u8 port,
1038 u32 *resp_len)
1039{
1040 struct opa_port_info *pi = (struct opa_port_info *)data;
1041 struct ib_event event;
1042 struct hfi1_devdata *dd;
1043 struct hfi1_pportdata *ppd;
1044 struct hfi1_ibport *ibp;
1045 u8 clientrereg;
1046 unsigned long flags;
1047 u32 smlid, opa_lid; /* tmp vars to hold LID values */
1048 u16 lid;
1049 u8 ls_old, ls_new, ps_new;
1050 u8 vls;
1051 u8 msl;
1052 u8 crc_enabled;
1053 u16 lse, lwe, mtu;
1054 u32 num_ports = OPA_AM_NPORT(am);
1055 u32 start_of_sm_config = OPA_AM_START_SM_CFG(am);
1056 int ret, i, invalid = 0, call_set_mtu = 0;
1057 int call_link_downgrade_policy = 0;
1058
1059 if (num_ports != 1) {
1060 smp->status |= IB_SMP_INVALID_FIELD;
1061 return reply((struct ib_mad_hdr *)smp);
1062 }
1063
1064 opa_lid = be32_to_cpu(pi->lid);
1065 if (opa_lid & 0xFFFF0000) {
1066 pr_warn("OPA_PortInfo lid out of range: %X\n", opa_lid);
1067 smp->status |= IB_SMP_INVALID_FIELD;
1068 goto get_only;
1069 }
1070
1071 lid = (u16)(opa_lid & 0x0000FFFF);
1072
1073 smlid = be32_to_cpu(pi->sm_lid);
1074 if (smlid & 0xFFFF0000) {
1075 pr_warn("OPA_PortInfo SM lid out of range: %X\n", smlid);
1076 smp->status |= IB_SMP_INVALID_FIELD;
1077 goto get_only;
1078 }
1079 smlid &= 0x0000FFFF;
1080
1081 clientrereg = (pi->clientrereg_subnettimeout &
1082 OPA_PI_MASK_CLIENT_REREGISTER);
1083
1084 dd = dd_from_ibdev(ibdev);
1085 /* IB numbers ports from 1, hw from 0 */
1086 ppd = dd->pport + (port - 1);
1087 ibp = &ppd->ibport_data;
1088 event.device = ibdev;
1089 event.element.port_num = port;
1090
1091 ls_old = driver_lstate(ppd);
1092
4eb06882
DD		 1093 ibp->rvp.mkey = pi->mkey;
1094 ibp->rvp.gid_prefix = pi->subnet_prefix;
1095 ibp->rvp.mkey_lease_period = be16_to_cpu(pi->mkey_lease_period);
77241056
MM		 1096
1097 /* Must be a valid unicast LID address. */
1098 if ((lid == 0 && ls_old > IB_PORT_INIT) ||
8859b4a6 1099 lid >= be16_to_cpu(IB_MULTICAST_LID_BASE)) {
77241056
MM		 1100 smp->status |= IB_SMP_INVALID_FIELD;
1101 pr_warn("SubnSet(OPA_PortInfo) lid invalid 0x%x\n",
1102 lid);
1103 } else if (ppd->lid != lid ||
1104 ppd->lmc != (pi->mkeyprotect_lmc & OPA_PI_MASK_LMC)) {
1105 if (ppd->lid != lid)
1106 hfi1_set_uevent_bits(ppd, _HFI1_EVENT_LID_CHANGE_BIT);
1107 if (ppd->lmc != (pi->mkeyprotect_lmc & OPA_PI_MASK_LMC))
1108 hfi1_set_uevent_bits(ppd, _HFI1_EVENT_LMC_CHANGE_BIT);
1109 hfi1_set_lid(ppd, lid, pi->mkeyprotect_lmc & OPA_PI_MASK_LMC);
1110 event.event = IB_EVENT_LID_CHANGE;
1111 ib_dispatch_event(&event);
1112 }
1113
1114 msl = pi->smsl & OPA_PI_MASK_SMSL;
1115 if (pi->partenforce_filterraw & OPA_PI_MASK_LINKINIT_REASON)
1116 ppd->linkinit_reason =
1117 (pi->partenforce_filterraw &
1118 OPA_PI_MASK_LINKINIT_REASON);
1119 /* enable/disable SW pkey checking as per FM control */
1120 if (pi->partenforce_filterraw & OPA_PI_MASK_PARTITION_ENFORCE_IN)
1121 ppd->part_enforce |= HFI1_PART_ENFORCE_IN;
1122 else
1123 ppd->part_enforce &= ~HFI1_PART_ENFORCE_IN;
1124
1125 if (pi->partenforce_filterraw & OPA_PI_MASK_PARTITION_ENFORCE_OUT)
1126 ppd->part_enforce |= HFI1_PART_ENFORCE_OUT;
1127 else
1128 ppd->part_enforce &= ~HFI1_PART_ENFORCE_OUT;
1129
1130 /* Must be a valid unicast LID address. */
1131 if ((smlid == 0 && ls_old > IB_PORT_INIT) ||
8859b4a6 1132 smlid >= be16_to_cpu(IB_MULTICAST_LID_BASE)) {
77241056
MM		 1133 smp->status |= IB_SMP_INVALID_FIELD;
1134 pr_warn("SubnSet(OPA_PortInfo) smlid invalid 0x%x\n", smlid);
4eb06882 1135 } else if (smlid != ibp->rvp.sm_lid || msl != ibp->rvp.sm_sl) {
77241056 1136 pr_warn("SubnSet(OPA_PortInfo) smlid 0x%x\n", smlid);
4eb06882 1137 spin_lock_irqsave(&ibp->rvp.lock, flags);
9c4a311e 1138 if (ibp->rvp.sm_ah) {
4eb06882 1139 if (smlid != ibp->rvp.sm_lid)
9c4a311e 1140 ibp->rvp.sm_ah->attr.dlid = smlid;
4eb06882 1141 if (msl != ibp->rvp.sm_sl)
9c4a311e 1142 ibp->rvp.sm_ah->attr.sl = msl;
77241056 1143 }
4eb06882
DD		 1144 spin_unlock_irqrestore(&ibp->rvp.lock, flags);
1145 if (smlid != ibp->rvp.sm_lid)
1146 ibp->rvp.sm_lid = smlid;
1147 if (msl != ibp->rvp.sm_sl)
1148 ibp->rvp.sm_sl = msl;
77241056
MM		 1149 event.event = IB_EVENT_SM_CHANGE;
1150 ib_dispatch_event(&event);
1151 }
1152
1153 if (pi->link_down_reason == 0) {
1154 ppd->local_link_down_reason.sma = 0;
1155 ppd->local_link_down_reason.latest = 0;
1156 }
1157
1158 if (pi->neigh_link_down_reason == 0) {
1159 ppd->neigh_link_down_reason.sma = 0;
1160 ppd->neigh_link_down_reason.latest = 0;
1161 }
1162
1163 ppd->sm_trap_qp = be32_to_cpu(pi->sm_trap_qp);
1164 ppd->sa_qp = be32_to_cpu(pi->sa_qp);
1165
1166 ppd->port_error_action = be32_to_cpu(pi->port_error_action);
1167 lwe = be16_to_cpu(pi->link_width.enabled);
1168 if (lwe) {
1169 if (lwe == OPA_LINK_WIDTH_RESET
1170 || lwe == OPA_LINK_WIDTH_RESET_OLD)
1171 set_link_width_enabled(ppd, ppd->link_width_supported);
1172 else if ((lwe & ~ppd->link_width_supported) == 0)
1173 set_link_width_enabled(ppd, lwe);
1174 else
1175 smp->status |= IB_SMP_INVALID_FIELD;
1176 }
1177 lwe = be16_to_cpu(pi->link_width_downgrade.enabled);
1178 /* LWD.E is always applied - 0 means "disabled" */
1179 if (lwe == OPA_LINK_WIDTH_RESET
1180 || lwe == OPA_LINK_WIDTH_RESET_OLD) {
1181 set_link_width_downgrade_enabled(ppd,
1182 ppd->link_width_downgrade_supported);
1183 } else if ((lwe & ~ppd->link_width_downgrade_supported) == 0) {
1184 /* only set and apply if something changed */
1185 if (lwe != ppd->link_width_downgrade_enabled) {
1186 set_link_width_downgrade_enabled(ppd, lwe);
1187 call_link_downgrade_policy = 1;
1188 }
1189 } else
1190 smp->status |= IB_SMP_INVALID_FIELD;
1191
1192 lse = be16_to_cpu(pi->link_speed.enabled);
1193 if (lse) {
1194 if (lse & be16_to_cpu(pi->link_speed.supported))
1195 set_link_speed_enabled(ppd, lse);
1196 else
1197 smp->status |= IB_SMP_INVALID_FIELD;
1198 }
1199
4eb06882
DD		 1200 ibp->rvp.mkeyprot =
1201 (pi->mkeyprotect_lmc & OPA_PI_MASK_MKEY_PROT_BIT) >> 6;
1202 ibp->rvp.vl_high_limit = be16_to_cpu(pi->vl.high_limit) & 0xFF;
77241056 1203 (void)hfi1_set_ib_cfg(ppd, HFI1_IB_CFG_VL_HIGH_LIMIT,
4eb06882 1204 ibp->rvp.vl_high_limit);
77241056
MM		 1205
1206 if (ppd->vls_supported/2 > ARRAY_SIZE(pi->neigh_mtu.pvlx_to_mtu) ||
1207 ppd->vls_supported > ARRAY_SIZE(dd->vld)) {
1208 smp->status |= IB_SMP_INVALID_FIELD;
1209 return reply((struct ib_mad_hdr *)smp);
1210 }
1211 for (i = 0; i < ppd->vls_supported; i++) {
1212 if ((i % 2) == 0)
1213 mtu = enum_to_mtu((pi->neigh_mtu.pvlx_to_mtu[i/2] >> 4)
1214 & 0xF);
1215 else
1216 mtu = enum_to_mtu(pi->neigh_mtu.pvlx_to_mtu[i/2] & 0xF);
1217 if (mtu == 0xffff) {
1218 pr_warn("SubnSet(OPA_PortInfo) mtu invalid %d (0x%x)\n",
1219 mtu,
1220 (pi->neigh_mtu.pvlx_to_mtu[0] >> 4) & 0xF);
1221 smp->status |= IB_SMP_INVALID_FIELD;
1222 mtu = hfi1_max_mtu; /* use a valid MTU */
1223 }
1224 if (dd->vld[i].mtu != mtu) {
1225 dd_dev_info(dd,
1226 "MTU change on vl %d from %d to %d\n",
1227 i, dd->vld[i].mtu, mtu);
1228 dd->vld[i].mtu = mtu;
1229 call_set_mtu++;
1230 }
1231 }
1232 /* As per OPAV1 spec: VL15 must support and be configured
1233 * for operation with a 2048 or larger MTU.
1234 */
1235 mtu = enum_to_mtu(pi->neigh_mtu.pvlx_to_mtu[15/2] & 0xF);
1236 if (mtu < 2048 || mtu == 0xffff)
1237 mtu = 2048;
1238 if (dd->vld[15].mtu != mtu) {
1239 dd_dev_info(dd,
1240 "MTU change on vl 15 from %d to %d\n",
1241 dd->vld[15].mtu, mtu);
1242 dd->vld[15].mtu = mtu;
1243 call_set_mtu++;
1244 }
1245 if (call_set_mtu)
1246 set_mtu(ppd);
1247
1248 /* Set operational VLs */
1249 vls = pi->operational_vls & OPA_PI_MASK_OPERATIONAL_VL;
1250 if (vls) {
1251 if (vls > ppd->vls_supported) {
1252 pr_warn("SubnSet(OPA_PortInfo) VL's supported invalid %d\n",
1253 pi->operational_vls);
1254 smp->status |= IB_SMP_INVALID_FIELD;
1255 } else {
1256 if (hfi1_set_ib_cfg(ppd, HFI1_IB_CFG_OP_VLS,
1257 vls) == -EINVAL)
1258 smp->status |= IB_SMP_INVALID_FIELD;
1259 }
1260 }
1261
1262 if (pi->mkey_violations == 0)
4eb06882 1263 ibp->rvp.mkey_violations = 0;
77241056
MM		 1264
1265 if (pi->pkey_violations == 0)
4eb06882 1266 ibp->rvp.pkey_violations = 0;
77241056
MM		 1267
1268 if (pi->qkey_violations == 0)
4eb06882 1269 ibp->rvp.qkey_violations = 0;
77241056 1270
4eb06882 1271 ibp->rvp.subnet_timeout =
77241056
MM		 1272 pi->clientrereg_subnettimeout & OPA_PI_MASK_SUBNET_TIMEOUT;
1273
1274 crc_enabled = be16_to_cpu(pi->port_ltp_crc_mode);
1275 crc_enabled >>= 4;
1276 crc_enabled &= 0xf;
1277
1278 if (crc_enabled != 0)
1279 ppd->port_crc_mode_enabled = port_ltp_to_cap(crc_enabled);
1280
1281 ppd->is_active_optimize_enabled =
1282 !!(be16_to_cpu(pi->port_mode)
1283 & OPA_PI_MASK_PORT_ACTIVE_OPTOMIZE);
1284
1285 ls_new = pi->port_states.portphysstate_portstate &
1286 OPA_PI_MASK_PORT_STATE;
1287 ps_new = (pi->port_states.portphysstate_portstate &
1288 OPA_PI_MASK_PORT_PHYSICAL_STATE) >> 4;
1289
1290 if (ls_old == IB_PORT_INIT) {
1291 if (start_of_sm_config) {
1292 if (ls_new == ls_old || (ls_new == IB_PORT_ARMED))
1293 ppd->is_sm_config_started = 1;
1294 } else if (ls_new == IB_PORT_ARMED) {
1295 if (ppd->is_sm_config_started == 0)
1296 invalid = 1;
1297 }
1298 }
1299
1300 /* Handle CLIENT_REREGISTER event b/c SM asked us for it */
1301 if (clientrereg) {
1302 event.event = IB_EVENT_CLIENT_REREGISTER;
1303 ib_dispatch_event(&event);
1304 }
1305
1306 /*
1307 * Do the port state change now that the other link parameters
1308 * have been set.
1309 * Changing the port physical state only makes sense if the link
1310 * is down or is being set to down.
1311 */
1312
1313 ret = set_port_states(ppd, smp, ls_new, ps_new, invalid);
1314 if (ret)
1315 return ret;
1316
1317 ret = __subn_get_opa_portinfo(smp, am, data, ibdev, port, resp_len);
1318
1319 /* restore re-reg bit per o14-12.2.1 */
1320 pi->clientrereg_subnettimeout |= clientrereg;
1321
1322 /*
1323 * Apply the new link downgrade policy. This may result in a link
1324 * bounce. Do this after everything else so things are settled.
1325 * Possible problem: if setting the port state above fails, then
1326 * the policy change is not applied.
1327 */
1328 if (call_link_downgrade_policy)
1329 apply_link_downgrade_policy(ppd, 0);
1330
1331 return ret;
1332
1333get_only:
1334 return __subn_get_opa_portinfo(smp, am, data, ibdev, port, resp_len);
1335}
1336
1337/**
1338 * set_pkeys - set the PKEY table for ctxt 0
1339 * @dd: the hfi1_ib device
1340 * @port: the IB port number
1341 * @pkeys: the PKEY table
1342 */
1343static int set_pkeys(struct hfi1_devdata *dd, u8 port, u16 *pkeys)
1344{
1345 struct hfi1_pportdata *ppd;
1346 int i;
1347 int changed = 0;
1348 int update_includes_mgmt_partition = 0;
1349
1350 /*
1351 * IB port one/two always maps to context zero/one,
1352 * always a kernel context, no locking needed
1353 * If we get here with ppd setup, no need to check
1354 * that rcd is valid.
1355 */
1356 ppd = dd->pport + (port - 1);
1357 /*
1358 * If the update does not include the management pkey, don't do it.
1359 */
1360 for (i = 0; i < ARRAY_SIZE(ppd->pkeys); i++) {
1361 if (pkeys[i] == LIM_MGMT_P_KEY) {
1362 update_includes_mgmt_partition = 1;
1363 break;
1364 }
1365 }
1366
1367 if (!update_includes_mgmt_partition)
1368 return 1;
1369
1370 for (i = 0; i < ARRAY_SIZE(ppd->pkeys); i++) {
1371 u16 key = pkeys[i];
1372 u16 okey = ppd->pkeys[i];
1373
1374 if (key == okey)
1375 continue;
1376 /*
1377 * The SM gives us the complete PKey table. We have
1378 * to ensure that we put the PKeys in the matching
1379 * slots.
1380 */
1381 ppd->pkeys[i] = key;
1382 changed = 1;
1383 }
1384
1385 if (changed) {
1386 struct ib_event event;
1387
1388 (void)hfi1_set_ib_cfg(ppd, HFI1_IB_CFG_PKEYS, 0);
1389
1390 event.event = IB_EVENT_PKEY_CHANGE;
ec3f2c12 1391 event.device = &dd->verbs_dev.rdi.ibdev;
77241056
MM		 1392 event.element.port_num = port;
1393 ib_dispatch_event(&event);
1394 }
1395 return 0;
1396}
1397
1398static int __subn_set_opa_pkeytable(struct opa_smp *smp, u32 am, u8 *data,
1399 struct ib_device *ibdev, u8 port,
1400 u32 *resp_len)
1401{
1402 struct hfi1_devdata *dd = dd_from_ibdev(ibdev);
1403 u32 n_blocks_sent = OPA_AM_NBLK(am);
1404 u32 start_block = am & 0x7ff;
1405 u16 *p = (u16 *) data;
1406 __be16 *q = (__be16 *)data;
1407 int i;
1408 u16 n_blocks_avail;
1409 unsigned npkeys = hfi1_get_npkeys(dd);
1410
1411 if (n_blocks_sent == 0) {
1412 pr_warn("OPA Get PKey AM Invalid : P = %d; B = 0x%x; N = 0x%x\n",
1413 port, start_block, n_blocks_sent);
1414 smp->status |= IB_SMP_INVALID_FIELD;
1415 return reply((struct ib_mad_hdr *)smp);
1416 }
1417
1418 n_blocks_avail = (u16)(npkeys/OPA_PARTITION_TABLE_BLK_SIZE) + 1;
1419
1420 if (start_block + n_blocks_sent > n_blocks_avail ||
1421 n_blocks_sent > OPA_NUM_PKEY_BLOCKS_PER_SMP) {
1422 pr_warn("OPA Set PKey AM Invalid : s 0x%x; req 0x%x; avail 0x%x; blk/smp 0x%lx\n",
1423 start_block, n_blocks_sent, n_blocks_avail,
1424 OPA_NUM_PKEY_BLOCKS_PER_SMP);
1425 smp->status |= IB_SMP_INVALID_FIELD;
1426 return reply((struct ib_mad_hdr *)smp);
1427 }
1428
1429 for (i = 0; i < n_blocks_sent * OPA_PARTITION_TABLE_BLK_SIZE; i++)
1430 p[i] = be16_to_cpu(q[i]);
1431
1432 if (start_block == 0 && set_pkeys(dd, port, p) != 0) {
1433 smp->status |= IB_SMP_INVALID_FIELD;
1434 return reply((struct ib_mad_hdr *)smp);
1435 }
1436
1437 return __subn_get_opa_pkeytable(smp, am, data, ibdev, port, resp_len);
1438}
1439
1440static int get_sc2vlt_tables(struct hfi1_devdata *dd, void *data)
1441{
a787bde8 1442 u64 *val = data;
77241056
MM		 1443
1444 *val++ = read_csr(dd, SEND_SC2VLT0);
1445 *val++ = read_csr(dd, SEND_SC2VLT1);
1446 *val++ = read_csr(dd, SEND_SC2VLT2);
1447 *val++ = read_csr(dd, SEND_SC2VLT3);
1448 return 0;
1449}
1450
1451#define ILLEGAL_VL 12
1452/*
1453 * filter_sc2vlt changes mappings to VL15 to ILLEGAL_VL (except
1454 * for SC15, which must map to VL15). If we don't remap things this
1455 * way it is possible for VL15 counters to increment when we try to
1456 * send on a SC which is mapped to an invalid VL.
1457 */
1458static void filter_sc2vlt(void *data)
1459{
1460 int i;
a787bde8 1461 u8 *pd = data;
77241056
MM		 1462
1463 for (i = 0; i < OPA_MAX_SCS; i++) {
1464 if (i == 15)
1465 continue;
1466 if ((pd[i] & 0x1f) == 0xf)
1467 pd[i] = ILLEGAL_VL;
1468 }
1469}
1470
1471static int set_sc2vlt_tables(struct hfi1_devdata *dd, void *data)
1472{
a787bde8 1473 u64 *val = data;
77241056
MM		 1474
1475 filter_sc2vlt(data);
1476
1477 write_csr(dd, SEND_SC2VLT0, *val++);
1478 write_csr(dd, SEND_SC2VLT1, *val++);
1479 write_csr(dd, SEND_SC2VLT2, *val++);
1480 write_csr(dd, SEND_SC2VLT3, *val++);
1481 write_seqlock_irq(&dd->sc2vl_lock);
a787bde8 1482 memcpy(dd->sc2vl, data, sizeof(dd->sc2vl));
77241056
MM		 1483 write_sequnlock_irq(&dd->sc2vl_lock);
1484 return 0;
1485}
1486
1487static int __subn_get_opa_sl_to_sc(struct opa_smp *smp, u32 am, u8 *data,
1488 struct ib_device *ibdev, u8 port,
1489 u32 *resp_len)
1490{
1491 struct hfi1_ibport *ibp = to_iport(ibdev, port);
6618c051 1492 u8 *p = data;
77241056
MM		 1493 size_t size = ARRAY_SIZE(ibp->sl_to_sc); /* == 32 */
1494 unsigned i;
1495
1496 if (am) {
1497 smp->status |= IB_SMP_INVALID_FIELD;
1498 return reply((struct ib_mad_hdr *)smp);
1499 }
1500
1501 for (i = 0; i < ARRAY_SIZE(ibp->sl_to_sc); i++)
1502 *p++ = ibp->sl_to_sc[i];
1503
1504 if (resp_len)
1505 *resp_len += size;
1506
1507 return reply((struct ib_mad_hdr *)smp);
1508}
1509
1510static int __subn_set_opa_sl_to_sc(struct opa_smp *smp, u32 am, u8 *data,
1511 struct ib_device *ibdev, u8 port,
1512 u32 *resp_len)
1513{
1514 struct hfi1_ibport *ibp = to_iport(ibdev, port);
6618c051 1515 u8 *p = data;
77241056
MM		 1516 int i;
1517
1518 if (am) {
1519 smp->status |= IB_SMP_INVALID_FIELD;
1520 return reply((struct ib_mad_hdr *)smp);
1521 }
1522
1523 for (i = 0; i < ARRAY_SIZE(ibp->sl_to_sc); i++)
1524 ibp->sl_to_sc[i] = *p++;
1525
1526 return __subn_get_opa_sl_to_sc(smp, am, data, ibdev, port, resp_len);
1527}
1528
1529static int __subn_get_opa_sc_to_sl(struct opa_smp *smp, u32 am, u8 *data,
1530 struct ib_device *ibdev, u8 port,
1531 u32 *resp_len)
1532{
1533 struct hfi1_ibport *ibp = to_iport(ibdev, port);
6618c051 1534 u8 *p = data;
77241056
MM		 1535 size_t size = ARRAY_SIZE(ibp->sc_to_sl); /* == 32 */
1536 unsigned i;
1537
1538 if (am) {
1539 smp->status |= IB_SMP_INVALID_FIELD;
1540 return reply((struct ib_mad_hdr *)smp);
1541 }
1542
1543 for (i = 0; i < ARRAY_SIZE(ibp->sc_to_sl); i++)
1544 *p++ = ibp->sc_to_sl[i];
1545
1546 if (resp_len)
1547 *resp_len += size;
1548
1549 return reply((struct ib_mad_hdr *)smp);
1550}
1551
1552static int __subn_set_opa_sc_to_sl(struct opa_smp *smp, u32 am, u8 *data,
1553 struct ib_device *ibdev, u8 port,
1554 u32 *resp_len)
1555{
1556 struct hfi1_ibport *ibp = to_iport(ibdev, port);
6618c051 1557 u8 *p = data;
77241056
MM		 1558 int i;
1559
1560 if (am) {
1561 smp->status |= IB_SMP_INVALID_FIELD;
1562 return reply((struct ib_mad_hdr *)smp);
1563 }
1564
1565 for (i = 0; i < ARRAY_SIZE(ibp->sc_to_sl); i++)
1566 ibp->sc_to_sl[i] = *p++;
1567
1568 return __subn_get_opa_sc_to_sl(smp, am, data, ibdev, port, resp_len);
1569}
1570
1571static int __subn_get_opa_sc_to_vlt(struct opa_smp *smp, u32 am, u8 *data,
1572 struct ib_device *ibdev, u8 port,
1573 u32 *resp_len)
1574{
1575 u32 n_blocks = OPA_AM_NBLK(am);
1576 struct hfi1_devdata *dd = dd_from_ibdev(ibdev);
1577 void *vp = (void *) data;
1578 size_t size = 4 * sizeof(u64);
1579
1580 if (n_blocks != 1) {
1581 smp->status |= IB_SMP_INVALID_FIELD;
1582 return reply((struct ib_mad_hdr *)smp);
1583 }
1584
1585 get_sc2vlt_tables(dd, vp);
1586
1587 if (resp_len)
1588 *resp_len += size;
1589
1590 return reply((struct ib_mad_hdr *)smp);
1591}
1592
1593static int __subn_set_opa_sc_to_vlt(struct opa_smp *smp, u32 am, u8 *data,
1594 struct ib_device *ibdev, u8 port,
1595 u32 *resp_len)
1596{
1597 u32 n_blocks = OPA_AM_NBLK(am);
1598 int async_update = OPA_AM_ASYNC(am);
1599 struct hfi1_devdata *dd = dd_from_ibdev(ibdev);
1600 void *vp = (void *) data;
1601 struct hfi1_pportdata *ppd;
1602 int lstate;
1603
1604 if (n_blocks != 1 || async_update) {
1605 smp->status |= IB_SMP_INVALID_FIELD;
1606 return reply((struct ib_mad_hdr *)smp);
1607 }
1608
1609 /* IB numbers ports from 1, hw from 0 */
1610 ppd = dd->pport + (port - 1);
1611 lstate = driver_lstate(ppd);
1612 /* it's known that async_update is 0 by this point, but include
1613 * the explicit check for clarity */
1614 if (!async_update &&
1615 (lstate == IB_PORT_ARMED || lstate == IB_PORT_ACTIVE)) {
1616 smp->status |= IB_SMP_INVALID_FIELD;
1617 return reply((struct ib_mad_hdr *)smp);
1618 }
1619
1620 set_sc2vlt_tables(dd, vp);
1621
1622 return __subn_get_opa_sc_to_vlt(smp, am, data, ibdev, port, resp_len);
1623}
1624
1625static int __subn_get_opa_sc_to_vlnt(struct opa_smp *smp, u32 am, u8 *data,
1626 struct ib_device *ibdev, u8 port,
1627 u32 *resp_len)
1628{
1629 u32 n_blocks = OPA_AM_NPORT(am);
1630 struct hfi1_devdata *dd = dd_from_ibdev(ibdev);
1631 struct hfi1_pportdata *ppd;
1632 void *vp = (void *) data;
1633 int size;
1634
1635 if (n_blocks != 1) {
1636 smp->status |= IB_SMP_INVALID_FIELD;
1637 return reply((struct ib_mad_hdr *)smp);
1638 }
1639
1640 ppd = dd->pport + (port - 1);
1641
1642 size = fm_get_table(ppd, FM_TBL_SC2VLNT, vp);
1643
1644 if (resp_len)
1645 *resp_len += size;
1646
1647 return reply((struct ib_mad_hdr *)smp);
1648}
1649
1650static int __subn_set_opa_sc_to_vlnt(struct opa_smp *smp, u32 am, u8 *data,
1651 struct ib_device *ibdev, u8 port,
1652 u32 *resp_len)
1653{
1654 u32 n_blocks = OPA_AM_NPORT(am);
1655 struct hfi1_devdata *dd = dd_from_ibdev(ibdev);
1656 struct hfi1_pportdata *ppd;
1657 void *vp = (void *) data;
1658 int lstate;
1659
1660 if (n_blocks != 1) {
1661 smp->status |= IB_SMP_INVALID_FIELD;
1662 return reply((struct ib_mad_hdr *)smp);
1663 }
1664
1665 /* IB numbers ports from 1, hw from 0 */
1666 ppd = dd->pport + (port - 1);
1667 lstate = driver_lstate(ppd);
1668 if (lstate == IB_PORT_ARMED || lstate == IB_PORT_ACTIVE) {
1669 smp->status |= IB_SMP_INVALID_FIELD;
1670 return reply((struct ib_mad_hdr *)smp);
1671 }
1672
1673 ppd = dd->pport + (port - 1);
1674
1675 fm_set_table(ppd, FM_TBL_SC2VLNT, vp);
1676
1677 return __subn_get_opa_sc_to_vlnt(smp, am, data, ibdev, port,
1678 resp_len);
1679}
1680
1681static int __subn_get_opa_psi(struct opa_smp *smp, u32 am, u8 *data,
1682 struct ib_device *ibdev, u8 port,
1683 u32 *resp_len)
1684{
1685 u32 nports = OPA_AM_NPORT(am);
1686 u32 start_of_sm_config = OPA_AM_START_SM_CFG(am);
1687 u32 lstate;
1688 struct hfi1_ibport *ibp;
1689 struct hfi1_pportdata *ppd;
1690 struct opa_port_state_info *psi = (struct opa_port_state_info *) data;
1691
1692 if (nports != 1) {
1693 smp->status |= IB_SMP_INVALID_FIELD;
1694 return reply((struct ib_mad_hdr *)smp);
1695 }
1696
1697 ibp = to_iport(ibdev, port);
1698 ppd = ppd_from_ibp(ibp);
1699
1700 lstate = driver_lstate(ppd);
1701
1702 if (start_of_sm_config && (lstate == IB_PORT_INIT))
1703 ppd->is_sm_config_started = 1;
1704
1705#if PI_LED_ENABLE_SUP
1706 psi->port_states.ledenable_offlinereason = ppd->neighbor_normal << 4;
1707 psi->port_states.ledenable_offlinereason |=
1708 ppd->is_sm_config_started << 5;
1709 psi->port_states.ledenable_offlinereason |=
a9c05e35 1710 ppd->offline_disabled_reason;
77241056
MM		 1711#else
1712 psi->port_states.offline_reason = ppd->neighbor_normal << 4;
1713 psi->port_states.offline_reason |= ppd->is_sm_config_started << 5;
a9c05e35 1714 psi->port_states.offline_reason |= ppd->offline_disabled_reason;
77241056
MM		 1715#endif /* PI_LED_ENABLE_SUP */
1716
1717 psi->port_states.portphysstate_portstate =
1718 (hfi1_ibphys_portstate(ppd) << 4) | (lstate & 0xf);
1719 psi->link_width_downgrade_tx_active =
aadfc3b2 1720 cpu_to_be16(ppd->link_width_downgrade_tx_active);
77241056 1721 psi->link_width_downgrade_rx_active =
aadfc3b2 1722 cpu_to_be16(ppd->link_width_downgrade_rx_active);
77241056
MM		 1723 if (resp_len)
1724 *resp_len += sizeof(struct opa_port_state_info);
1725
1726 return reply((struct ib_mad_hdr *)smp);
1727}
1728
1729static int __subn_set_opa_psi(struct opa_smp *smp, u32 am, u8 *data,
1730 struct ib_device *ibdev, u8 port,
1731 u32 *resp_len)
1732{
1733 u32 nports = OPA_AM_NPORT(am);
1734 u32 start_of_sm_config = OPA_AM_START_SM_CFG(am);
1735 u32 ls_old;
1736 u8 ls_new, ps_new;
1737 struct hfi1_ibport *ibp;
1738 struct hfi1_pportdata *ppd;
1739 struct opa_port_state_info *psi = (struct opa_port_state_info *) data;
1740 int ret, invalid = 0;
1741
1742 if (nports != 1) {
1743 smp->status |= IB_SMP_INVALID_FIELD;
1744 return reply((struct ib_mad_hdr *)smp);
1745 }
1746
1747 ibp = to_iport(ibdev, port);
1748 ppd = ppd_from_ibp(ibp);
1749
1750 ls_old = driver_lstate(ppd);
1751
1752 ls_new = port_states_to_logical_state(&psi->port_states);
1753 ps_new = port_states_to_phys_state(&psi->port_states);
1754
1755 if (ls_old == IB_PORT_INIT) {
1756 if (start_of_sm_config) {
1757 if (ls_new == ls_old || (ls_new == IB_PORT_ARMED))
1758 ppd->is_sm_config_started = 1;
1759 } else if (ls_new == IB_PORT_ARMED) {
1760 if (ppd->is_sm_config_started == 0)
1761 invalid = 1;
1762 }
1763 }
1764
1765 ret = set_port_states(ppd, smp, ls_new, ps_new, invalid);
1766 if (ret)
1767 return ret;
1768
1769 if (invalid)
1770 smp->status |= IB_SMP_INVALID_FIELD;
1771
1772 return __subn_get_opa_psi(smp, am, data, ibdev, port, resp_len);
1773}
1774
1775static int __subn_get_opa_cable_info(struct opa_smp *smp, u32 am, u8 *data,
1776 struct ib_device *ibdev, u8 port,
1777 u32 *resp_len)
1778{
1779 struct hfi1_devdata *dd = dd_from_ibdev(ibdev);
1780 u32 addr = OPA_AM_CI_ADDR(am);
1781 u32 len = OPA_AM_CI_LEN(am) + 1;
1782 int ret;
1783
349ac71f 1784#define __CI_PAGE_SIZE BIT(7) /* 128 bytes */
77241056
MM		 1785#define __CI_PAGE_MASK ~(__CI_PAGE_SIZE - 1)
1786#define __CI_PAGE_NUM(a) ((a) & __CI_PAGE_MASK)
1787
1788 /* check that addr is within spec, and
1789 * addr and (addr + len - 1) are on the same "page" */
1790 if (addr >= 4096 ||
1791 (__CI_PAGE_NUM(addr) != __CI_PAGE_NUM(addr + len - 1))) {
1792 smp->status |= IB_SMP_INVALID_FIELD;
1793 return reply((struct ib_mad_hdr *)smp);
1794 }
1795
1796 ret = get_cable_info(dd, port, addr, len, data);
1797
1798 if (ret == -ENODEV) {
1799 smp->status |= IB_SMP_UNSUP_METH_ATTR;
1800 return reply((struct ib_mad_hdr *)smp);
1801 }
1802
1803 /* The address range for the CableInfo SMA query is wider than the
1804 * memory available on the QSFP cable. We want to return a valid
1805 * response, albeit zeroed out, for address ranges beyond available
1806 * memory but that are within the CableInfo query spec
1807 */
1808 if (ret < 0 && ret != -ERANGE) {
1809 smp->status |= IB_SMP_INVALID_FIELD;
1810 return reply((struct ib_mad_hdr *)smp);
1811 }
1812
1813 if (resp_len)
1814 *resp_len += len;
1815
1816 return reply((struct ib_mad_hdr *)smp);
1817}
1818
1819static int __subn_get_opa_bct(struct opa_smp *smp, u32 am, u8 *data,
1820 struct ib_device *ibdev, u8 port, u32 *resp_len)
1821{
1822 u32 num_ports = OPA_AM_NPORT(am);
1823 struct hfi1_devdata *dd = dd_from_ibdev(ibdev);
1824 struct hfi1_pportdata *ppd;
1825 struct buffer_control *p = (struct buffer_control *) data;
1826 int size;
1827
1828 if (num_ports != 1) {
1829 smp->status |= IB_SMP_INVALID_FIELD;
1830 return reply((struct ib_mad_hdr *)smp);
1831 }
1832
1833 ppd = dd->pport + (port - 1);
1834 size = fm_get_table(ppd, FM_TBL_BUFFER_CONTROL, p);
1835 trace_bct_get(dd, p);
1836 if (resp_len)
1837 *resp_len += size;
1838
1839 return reply((struct ib_mad_hdr *)smp);
1840}
1841
1842static int __subn_set_opa_bct(struct opa_smp *smp, u32 am, u8 *data,
1843 struct ib_device *ibdev, u8 port, u32 *resp_len)
1844{
1845 u32 num_ports = OPA_AM_NPORT(am);
1846 struct hfi1_devdata *dd = dd_from_ibdev(ibdev);
1847 struct hfi1_pportdata *ppd;
1848 struct buffer_control *p = (struct buffer_control *) data;
1849
1850 if (num_ports != 1) {
1851 smp->status |= IB_SMP_INVALID_FIELD;
1852 return reply((struct ib_mad_hdr *)smp);
1853 }
1854 ppd = dd->pport + (port - 1);
1855 trace_bct_set(dd, p);
1856 if (fm_set_table(ppd, FM_TBL_BUFFER_CONTROL, p) < 0) {
1857 smp->status |= IB_SMP_INVALID_FIELD;
1858 return reply((struct ib_mad_hdr *)smp);
1859 }
1860
1861 return __subn_get_opa_bct(smp, am, data, ibdev, port, resp_len);
1862}
1863
1864static int __subn_get_opa_vl_arb(struct opa_smp *smp, u32 am, u8 *data,
1865 struct ib_device *ibdev, u8 port,
1866 u32 *resp_len)
1867{
1868 struct hfi1_pportdata *ppd = ppd_from_ibp(to_iport(ibdev, port));
1869 u32 num_ports = OPA_AM_NPORT(am);
1870 u8 section = (am & 0x00ff0000) >> 16;
1871 u8 *p = data;
1872 int size = 0;
1873
1874 if (num_ports != 1) {
1875 smp->status |= IB_SMP_INVALID_FIELD;
1876 return reply((struct ib_mad_hdr *)smp);
1877 }
1878
1879 switch (section) {
1880 case OPA_VLARB_LOW_ELEMENTS:
1881 size = fm_get_table(ppd, FM_TBL_VL_LOW_ARB, p);
1882 break;
1883 case OPA_VLARB_HIGH_ELEMENTS:
1884 size = fm_get_table(ppd, FM_TBL_VL_HIGH_ARB, p);
1885 break;
1886 case OPA_VLARB_PREEMPT_ELEMENTS:
1887 size = fm_get_table(ppd, FM_TBL_VL_PREEMPT_ELEMS, p);
1888 break;
1889 case OPA_VLARB_PREEMPT_MATRIX:
1890 size = fm_get_table(ppd, FM_TBL_VL_PREEMPT_MATRIX, p);
1891 break;
1892 default:
1893 pr_warn("OPA SubnGet(VL Arb) AM Invalid : 0x%x\n",
1894 be32_to_cpu(smp->attr_mod));
1895 smp->status |= IB_SMP_INVALID_FIELD;
1896 break;
1897 }
1898
1899 if (size > 0 && resp_len)
1900 *resp_len += size;
1901
1902 return reply((struct ib_mad_hdr *)smp);
1903}
1904
1905static int __subn_set_opa_vl_arb(struct opa_smp *smp, u32 am, u8 *data,
1906 struct ib_device *ibdev, u8 port,
1907 u32 *resp_len)
1908{
1909 struct hfi1_pportdata *ppd = ppd_from_ibp(to_iport(ibdev, port));
1910 u32 num_ports = OPA_AM_NPORT(am);
1911 u8 section = (am & 0x00ff0000) >> 16;
1912 u8 *p = data;
1913
1914 if (num_ports != 1) {
1915 smp->status |= IB_SMP_INVALID_FIELD;
1916 return reply((struct ib_mad_hdr *)smp);
1917 }
1918
1919 switch (section) {
1920 case OPA_VLARB_LOW_ELEMENTS:
1921 (void) fm_set_table(ppd, FM_TBL_VL_LOW_ARB, p);
1922 break;
1923 case OPA_VLARB_HIGH_ELEMENTS:
1924 (void) fm_set_table(ppd, FM_TBL_VL_HIGH_ARB, p);
1925 break;
1926 /* neither OPA_VLARB_PREEMPT_ELEMENTS, or OPA_VLARB_PREEMPT_MATRIX
1927 * can be changed from the default values */
1928 case OPA_VLARB_PREEMPT_ELEMENTS:
1929 /* FALLTHROUGH */
1930 case OPA_VLARB_PREEMPT_MATRIX:
1931 smp->status |= IB_SMP_UNSUP_METH_ATTR;
1932 break;
1933 default:
1934 pr_warn("OPA SubnSet(VL Arb) AM Invalid : 0x%x\n",
1935 be32_to_cpu(smp->attr_mod));
1936 smp->status |= IB_SMP_INVALID_FIELD;
1937 break;
1938 }
1939
1940 return __subn_get_opa_vl_arb(smp, am, data, ibdev, port, resp_len);
1941}
1942
1943struct opa_pma_mad {
1944 struct ib_mad_hdr mad_hdr;
1945 u8 data[2024];
1946} __packed;
1947
1948struct opa_class_port_info {
1949 u8 base_version;
1950 u8 class_version;
1951 __be16 cap_mask;
1952 __be32 cap_mask2_resp_time;
1953
1954 u8 redirect_gid[16];
1955 __be32 redirect_tc_fl;
1956 __be32 redirect_lid;
1957 __be32 redirect_sl_qp;
1958 __be32 redirect_qkey;
1959
1960 u8 trap_gid[16];
1961 __be32 trap_tc_fl;
1962 __be32 trap_lid;
1963 __be32 trap_hl_qp;
1964 __be32 trap_qkey;
1965
1966 __be16 trap_pkey;
1967 __be16 redirect_pkey;
1968
1969 u8 trap_sl_rsvd;
1970 u8 reserved[3];
1971} __packed;
1972
1973struct opa_port_status_req {
1974 __u8 port_num;
1975 __u8 reserved[3];
1976 __be32 vl_select_mask;
1977};
1978
1979#define VL_MASK_ALL 0x000080ff
1980
1981struct opa_port_status_rsp {
1982 __u8 port_num;
1983 __u8 reserved[3];
1984 __be32 vl_select_mask;
1985
1986 /* Data counters */
1987 __be64 port_xmit_data;
1988 __be64 port_rcv_data;
1989 __be64 port_xmit_pkts;
1990 __be64 port_rcv_pkts;
1991 __be64 port_multicast_xmit_pkts;
1992 __be64 port_multicast_rcv_pkts;
1993 __be64 port_xmit_wait;
1994 __be64 sw_port_congestion;
1995 __be64 port_rcv_fecn;
1996 __be64 port_rcv_becn;
1997 __be64 port_xmit_time_cong;
1998 __be64 port_xmit_wasted_bw;
1999 __be64 port_xmit_wait_data;
2000 __be64 port_rcv_bubble;
2001 __be64 port_mark_fecn;
2002 /* Error counters */
2003 __be64 port_rcv_constraint_errors;
2004 __be64 port_rcv_switch_relay_errors;
2005 __be64 port_xmit_discards;
2006 __be64 port_xmit_constraint_errors;
2007 __be64 port_rcv_remote_physical_errors;
2008 __be64 local_link_integrity_errors;
2009 __be64 port_rcv_errors;
2010 __be64 excessive_buffer_overruns;
2011 __be64 fm_config_errors;
2012 __be32 link_error_recovery;
2013 __be32 link_downed;
2014 u8 uncorrectable_errors;
2015
2016 u8 link_quality_indicator; /* 5res, 3bit */
2017 u8 res2[6];
2018 struct _vls_pctrs {
2019 /* per-VL Data counters */
2020 __be64 port_vl_xmit_data;
2021 __be64 port_vl_rcv_data;
2022 __be64 port_vl_xmit_pkts;
2023 __be64 port_vl_rcv_pkts;
2024 __be64 port_vl_xmit_wait;
2025 __be64 sw_port_vl_congestion;
2026 __be64 port_vl_rcv_fecn;
2027 __be64 port_vl_rcv_becn;
2028 __be64 port_xmit_time_cong;
2029 __be64 port_vl_xmit_wasted_bw;
2030 __be64 port_vl_xmit_wait_data;
2031 __be64 port_vl_rcv_bubble;
2032 __be64 port_vl_mark_fecn;
2033 __be64 port_vl_xmit_discards;
2034 } vls[0]; /* real array size defined by # bits set in vl_select_mask */
2035};
2036
2037enum counter_selects {
2038 CS_PORT_XMIT_DATA = (1 << 31),
2039 CS_PORT_RCV_DATA = (1 << 30),
2040 CS_PORT_XMIT_PKTS = (1 << 29),
2041 CS_PORT_RCV_PKTS = (1 << 28),
2042 CS_PORT_MCAST_XMIT_PKTS = (1 << 27),
2043 CS_PORT_MCAST_RCV_PKTS = (1 << 26),
2044 CS_PORT_XMIT_WAIT = (1 << 25),
2045 CS_SW_PORT_CONGESTION = (1 << 24),
2046 CS_PORT_RCV_FECN = (1 << 23),
2047 CS_PORT_RCV_BECN = (1 << 22),
2048 CS_PORT_XMIT_TIME_CONG = (1 << 21),
2049 CS_PORT_XMIT_WASTED_BW = (1 << 20),
2050 CS_PORT_XMIT_WAIT_DATA = (1 << 19),
2051 CS_PORT_RCV_BUBBLE = (1 << 18),
2052 CS_PORT_MARK_FECN = (1 << 17),
2053 CS_PORT_RCV_CONSTRAINT_ERRORS = (1 << 16),
2054 CS_PORT_RCV_SWITCH_RELAY_ERRORS = (1 << 15),
2055 CS_PORT_XMIT_DISCARDS = (1 << 14),
2056 CS_PORT_XMIT_CONSTRAINT_ERRORS = (1 << 13),
2057 CS_PORT_RCV_REMOTE_PHYSICAL_ERRORS = (1 << 12),
2058 CS_LOCAL_LINK_INTEGRITY_ERRORS = (1 << 11),
2059 CS_PORT_RCV_ERRORS = (1 << 10),
2060 CS_EXCESSIVE_BUFFER_OVERRUNS = (1 << 9),
2061 CS_FM_CONFIG_ERRORS = (1 << 8),
2062 CS_LINK_ERROR_RECOVERY = (1 << 7),
2063 CS_LINK_DOWNED = (1 << 6),
2064 CS_UNCORRECTABLE_ERRORS = (1 << 5),
2065};
2066
2067struct opa_clear_port_status {
2068 __be64 port_select_mask[4];
2069 __be32 counter_select_mask;
2070};
2071
2072struct opa_aggregate {
2073 __be16 attr_id;
2074 __be16 err_reqlength; /* 1 bit, 8 res, 7 bit */
2075 __be32 attr_mod;
2076 u8 data[0];
2077};
2078
f0852922
AL		 2079#define MSK_LLI 0x000000f0
2080#define MSK_LLI_SFT 4
2081#define MSK_LER 0x0000000f
2082#define MSK_LER_SFT 0
2083#define ADD_LLI 8
2084#define ADD_LER 2
2085
2086/* Request contains first three fields, response contains those plus the rest */
77241056
MM		 2087struct opa_port_data_counters_msg {
2088 __be64 port_select_mask[4];
2089 __be32 vl_select_mask;
f0852922 2090 __be32 resolution;
77241056
MM		 2091
2092 /* Response fields follow */
77241056
MM		 2093 struct _port_dctrs {
2094 u8 port_number;
2095 u8 reserved2[3];
2096 __be32 link_quality_indicator; /* 29res, 3bit */
2097
2098 /* Data counters */
2099 __be64 port_xmit_data;
2100 __be64 port_rcv_data;
2101 __be64 port_xmit_pkts;
2102 __be64 port_rcv_pkts;
2103 __be64 port_multicast_xmit_pkts;
2104 __be64 port_multicast_rcv_pkts;
2105 __be64 port_xmit_wait;
2106 __be64 sw_port_congestion;
2107 __be64 port_rcv_fecn;
2108 __be64 port_rcv_becn;
2109 __be64 port_xmit_time_cong;
2110 __be64 port_xmit_wasted_bw;
2111 __be64 port_xmit_wait_data;
2112 __be64 port_rcv_bubble;
2113 __be64 port_mark_fecn;
2114
2115 __be64 port_error_counter_summary;
2116 /* Sum of error counts/port */
2117
2118 struct _vls_dctrs {
2119 /* per-VL Data counters */
2120 __be64 port_vl_xmit_data;
2121 __be64 port_vl_rcv_data;
2122 __be64 port_vl_xmit_pkts;
2123 __be64 port_vl_rcv_pkts;
2124 __be64 port_vl_xmit_wait;
2125 __be64 sw_port_vl_congestion;
2126 __be64 port_vl_rcv_fecn;
2127 __be64 port_vl_rcv_becn;
2128 __be64 port_xmit_time_cong;
2129 __be64 port_vl_xmit_wasted_bw;
2130 __be64 port_vl_xmit_wait_data;
2131 __be64 port_vl_rcv_bubble;
2132 __be64 port_vl_mark_fecn;
2133 } vls[0];
2134 /* array size defined by #bits set in vl_select_mask*/
2135 } port[1]; /* array size defined by #ports in attribute modifier */
2136};
2137
2138struct opa_port_error_counters64_msg {
2139 /* Request contains first two fields, response contains the
2140 * whole magilla */
2141 __be64 port_select_mask[4];
2142 __be32 vl_select_mask;
2143
2144 /* Response-only fields follow */
2145 __be32 reserved1;
2146 struct _port_ectrs {
2147 u8 port_number;
2148 u8 reserved2[7];
2149 __be64 port_rcv_constraint_errors;
2150 __be64 port_rcv_switch_relay_errors;
2151 __be64 port_xmit_discards;
2152 __be64 port_xmit_constraint_errors;
2153 __be64 port_rcv_remote_physical_errors;
2154 __be64 local_link_integrity_errors;
2155 __be64 port_rcv_errors;
2156 __be64 excessive_buffer_overruns;
2157 __be64 fm_config_errors;
2158 __be32 link_error_recovery;
2159 __be32 link_downed;
2160 u8 uncorrectable_errors;
2161 u8 reserved3[7];
2162 struct _vls_ectrs {
2163 __be64 port_vl_xmit_discards;
2164 } vls[0];
2165 /* array size defined by #bits set in vl_select_mask */
2166 } port[1]; /* array size defined by #ports in attribute modifier */
2167};
2168
2169struct opa_port_error_info_msg {
2170 __be64 port_select_mask[4];
2171 __be32 error_info_select_mask;
2172 __be32 reserved1;
2173 struct _port_ei {
2174
2175 u8 port_number;
2176 u8 reserved2[7];
2177
2178 /* PortRcvErrorInfo */
2179 struct {
2180 u8 status_and_code;
2181 union {
2182 u8 raw[17];
2183 struct {
2184 /* EI1to12 format */
2185 u8 packet_flit1[8];
2186 u8 packet_flit2[8];
2187 u8 remaining_flit_bits12;
2188 } ei1to12;
2189 struct {
2190 u8 packet_bytes[8];
2191 u8 remaining_flit_bits;
2192 } ei13;
2193 } ei;
2194 u8 reserved3[6];
2195 } __packed port_rcv_ei;
2196
2197 /* ExcessiveBufferOverrunInfo */
2198 struct {
2199 u8 status_and_sc;
2200 u8 reserved4[7];
2201 } __packed excessive_buffer_overrun_ei;
2202
2203 /* PortXmitConstraintErrorInfo */
2204 struct {
2205 u8 status;
2206 u8 reserved5;
2207 __be16 pkey;
2208 __be32 slid;
2209 } __packed port_xmit_constraint_ei;
2210
2211 /* PortRcvConstraintErrorInfo */
2212 struct {
2213 u8 status;
2214 u8 reserved6;
2215 __be16 pkey;
2216 __be32 slid;
2217 } __packed port_rcv_constraint_ei;
2218
2219 /* PortRcvSwitchRelayErrorInfo */
2220 struct {
2221 u8 status_and_code;
2222 u8 reserved7[3];
2223 __u32 error_info;
2224 } __packed port_rcv_switch_relay_ei;
2225
2226 /* UncorrectableErrorInfo */
2227 struct {
2228 u8 status_and_code;
2229 u8 reserved8;
2230 } __packed uncorrectable_ei;
2231
2232 /* FMConfigErrorInfo */
2233 struct {
2234 u8 status_and_code;
2235 u8 error_info;
2236 } __packed fm_config_ei;
2237 __u32 reserved9;
2238 } port[1]; /* actual array size defined by #ports in attr modifier */
2239};
2240
2241/* opa_port_error_info_msg error_info_select_mask bit definitions */
2242enum error_info_selects {
2243 ES_PORT_RCV_ERROR_INFO = (1 << 31),
2244 ES_EXCESSIVE_BUFFER_OVERRUN_INFO = (1 << 30),
2245 ES_PORT_XMIT_CONSTRAINT_ERROR_INFO = (1 << 29),
2246 ES_PORT_RCV_CONSTRAINT_ERROR_INFO = (1 << 28),
2247 ES_PORT_RCV_SWITCH_RELAY_ERROR_INFO = (1 << 27),
2248 ES_UNCORRECTABLE_ERROR_INFO = (1 << 26),
2249 ES_FM_CONFIG_ERROR_INFO = (1 << 25)
2250};
2251
2252static int pma_get_opa_classportinfo(struct opa_pma_mad *pmp,
2253 struct ib_device *ibdev, u32 *resp_len)
2254{
2255 struct opa_class_port_info *p =
2256 (struct opa_class_port_info *)pmp->data;
2257
2258 memset(pmp->data, 0, sizeof(pmp->data));
2259
2260 if (pmp->mad_hdr.attr_mod != 0)
2261 pmp->mad_hdr.status |= IB_SMP_INVALID_FIELD;
2262
2263 p->base_version = OPA_MGMT_BASE_VERSION;
2264 p->class_version = OPA_SMI_CLASS_VERSION;
2265 /*
2266 * Expected response time is 4.096 usec. * 2^18 == 1.073741824 sec.
2267 */
2268 p->cap_mask2_resp_time = cpu_to_be32(18);
2269
2270 if (resp_len)
2271 *resp_len += sizeof(*p);
2272
2273 return reply((struct ib_mad_hdr *)pmp);
2274}
2275
2276static void a0_portstatus(struct hfi1_pportdata *ppd,
2277 struct opa_port_status_rsp *rsp, u32 vl_select_mask)
2278{
2279 if (!is_bx(ppd->dd)) {
2280 unsigned long vl;
f4ddedf4 2281 u64 sum_vl_xmit_wait = 0;
77241056 2282 u32 vl_all_mask = VL_MASK_ALL;
77241056
MM		 2283
2284 for_each_set_bit(vl, (unsigned long *)&(vl_all_mask),
2285 8 * sizeof(vl_all_mask)) {
f4ddedf4
IW		 2286 u64 tmp = sum_vl_xmit_wait +
2287 read_port_cntr(ppd, C_TX_WAIT_VL,
2288 idx_from_vl(vl));
2289 if (tmp < sum_vl_xmit_wait) {
2290 /* we wrapped */
2291 sum_vl_xmit_wait = (u64)~0;
2292 break;
2293 }
2294 sum_vl_xmit_wait = tmp;
77241056 2295 }
f4ddedf4
IW		 2296 if (be64_to_cpu(rsp->port_xmit_wait) > sum_vl_xmit_wait)
2297 rsp->port_xmit_wait = cpu_to_be64(sum_vl_xmit_wait);
77241056
MM		 2298 }
2299}
2300
2301
2302static int pma_get_opa_portstatus(struct opa_pma_mad *pmp,
2303 struct ib_device *ibdev, u8 port, u32 *resp_len)
2304{
2305 struct opa_port_status_req *req =
2306 (struct opa_port_status_req *)pmp->data;
2307 struct hfi1_devdata *dd = dd_from_ibdev(ibdev);
2308 struct opa_port_status_rsp *rsp;
2309 u32 vl_select_mask = be32_to_cpu(req->vl_select_mask);
2310 unsigned long vl;
2311 size_t response_data_size;
2312 u32 nports = be32_to_cpu(pmp->mad_hdr.attr_mod) >> 24;
2313 u8 port_num = req->port_num;
2314 u8 num_vls = hweight32(vl_select_mask);
2315 struct _vls_pctrs *vlinfo;
2316 struct hfi1_ibport *ibp = to_iport(ibdev, port);
2317 struct hfi1_pportdata *ppd = ppd_from_ibp(ibp);
2318 int vfi;
2319 u64 tmp, tmp2;
2320
2321 response_data_size = sizeof(struct opa_port_status_rsp) +
2322 num_vls * sizeof(struct _vls_pctrs);
2323 if (response_data_size > sizeof(pmp->data)) {
2324 pmp->mad_hdr.status |= OPA_PM_STATUS_REQUEST_TOO_LARGE;
2325 return reply((struct ib_mad_hdr *)pmp);
2326 }
2327
2328 if (nports != 1 || (port_num && port_num != port)
2329 || num_vls > OPA_MAX_VLS || (vl_select_mask & ~VL_MASK_ALL)) {
2330 pmp->mad_hdr.status |= IB_SMP_INVALID_FIELD;
2331 return reply((struct ib_mad_hdr *)pmp);
2332 }
2333
2334 memset(pmp->data, 0, sizeof(pmp->data));
2335
2336 rsp = (struct opa_port_status_rsp *)pmp->data;
2337 if (port_num)
2338 rsp->port_num = port_num;
2339 else
2340 rsp->port_num = port;
2341
2342 rsp->port_rcv_constraint_errors =
2343 cpu_to_be64(read_port_cntr(ppd, C_SW_RCV_CSTR_ERR,
2344 CNTR_INVALID_VL));
2345
2346 hfi1_read_link_quality(dd, &rsp->link_quality_indicator);
2347
2348 rsp->vl_select_mask = cpu_to_be32(vl_select_mask);
2349 rsp->port_xmit_data = cpu_to_be64(read_dev_cntr(dd, C_DC_XMIT_FLITS,
2350 CNTR_INVALID_VL));
2351 rsp->port_rcv_data = cpu_to_be64(read_dev_cntr(dd, C_DC_RCV_FLITS,
2352 CNTR_INVALID_VL));
77241056
MM		 2353 rsp->port_xmit_pkts = cpu_to_be64(read_dev_cntr(dd, C_DC_XMIT_PKTS,
2354 CNTR_INVALID_VL));
2355 rsp->port_rcv_pkts = cpu_to_be64(read_dev_cntr(dd, C_DC_RCV_PKTS,
2356 CNTR_INVALID_VL));
2357 rsp->port_multicast_xmit_pkts =
2358 cpu_to_be64(read_dev_cntr(dd, C_DC_MC_XMIT_PKTS,
2359 CNTR_INVALID_VL));
2360 rsp->port_multicast_rcv_pkts =
2361 cpu_to_be64(read_dev_cntr(dd, C_DC_MC_RCV_PKTS,
2362 CNTR_INVALID_VL));
2363 rsp->port_xmit_wait =
2364 cpu_to_be64(read_port_cntr(ppd, C_TX_WAIT, CNTR_INVALID_VL));
2365 rsp->port_rcv_fecn =
2366 cpu_to_be64(read_dev_cntr(dd, C_DC_RCV_FCN, CNTR_INVALID_VL));
2367 rsp->port_rcv_becn =
2368 cpu_to_be64(read_dev_cntr(dd, C_DC_RCV_BCN, CNTR_INVALID_VL));
2369 rsp->port_xmit_discards =
2370 cpu_to_be64(read_port_cntr(ppd, C_SW_XMIT_DSCD,
2371 CNTR_INVALID_VL));
2372 rsp->port_xmit_constraint_errors =
2373 cpu_to_be64(read_port_cntr(ppd, C_SW_XMIT_CSTR_ERR,
2374 CNTR_INVALID_VL));
2375 rsp->port_rcv_remote_physical_errors =
2376 cpu_to_be64(read_dev_cntr(dd, C_DC_RMT_PHY_ERR,
2377 CNTR_INVALID_VL));
2378 tmp = read_dev_cntr(dd, C_DC_RX_REPLAY, CNTR_INVALID_VL);
2379 tmp2 = tmp + read_dev_cntr(dd, C_DC_TX_REPLAY, CNTR_INVALID_VL);
2380 if (tmp2 < tmp) {
2381 /* overflow/wrapped */
2382 rsp->local_link_integrity_errors = cpu_to_be64(~0);
2383 } else {
2384 rsp->local_link_integrity_errors = cpu_to_be64(tmp2);
2385 }
2386 tmp = read_dev_cntr(dd, C_DC_SEQ_CRC_CNT, CNTR_INVALID_VL);
2387 tmp2 = tmp + read_dev_cntr(dd, C_DC_REINIT_FROM_PEER_CNT,
2388 CNTR_INVALID_VL);
2389 if (tmp2 > (u32)UINT_MAX || tmp2 < tmp) {
2390 /* overflow/wrapped */
2391 rsp->link_error_recovery = cpu_to_be32(~0);
2392 } else {
2393 rsp->link_error_recovery = cpu_to_be32(tmp2);
2394 }
2395 rsp->port_rcv_errors =
2396 cpu_to_be64(read_dev_cntr(dd, C_DC_RCV_ERR, CNTR_INVALID_VL));
2397 rsp->excessive_buffer_overruns =
2398 cpu_to_be64(read_dev_cntr(dd, C_RCV_OVF, CNTR_INVALID_VL));
2399 rsp->fm_config_errors =
2400 cpu_to_be64(read_dev_cntr(dd, C_DC_FM_CFG_ERR,
2401 CNTR_INVALID_VL));
2402 rsp->link_downed = cpu_to_be32(read_port_cntr(ppd, C_SW_LINK_DOWN,
2403 CNTR_INVALID_VL));
2404
2405 /* rsp->uncorrectable_errors is 8 bits wide, and it pegs at 0xff */
2406 tmp = read_dev_cntr(dd, C_DC_UNC_ERR, CNTR_INVALID_VL);
2407 rsp->uncorrectable_errors = tmp < 0x100 ? (tmp & 0xff) : 0xff;
2408
2409 vlinfo = &(rsp->vls[0]);
2410 vfi = 0;
2411 /* The vl_select_mask has been checked above, and we know
2412 * that it contains only entries which represent valid VLs.
2413 * So in the for_each_set_bit() loop below, we don't need
2414 * any additional checks for vl.
2415 */
2416 for_each_set_bit(vl, (unsigned long *)&(vl_select_mask),
2417 8 * sizeof(vl_select_mask)) {
2418 memset(vlinfo, 0, sizeof(*vlinfo));
2419
2420 tmp = read_dev_cntr(dd, C_DC_RX_FLIT_VL, idx_from_vl(vl));
2421 rsp->vls[vfi].port_vl_rcv_data = cpu_to_be64(tmp);
77241056
MM		 2422
2423 rsp->vls[vfi].port_vl_rcv_pkts =
2424 cpu_to_be64(read_dev_cntr(dd, C_DC_RX_PKT_VL,
2425 idx_from_vl(vl)));
2426
2427 rsp->vls[vfi].port_vl_xmit_data =
2428 cpu_to_be64(read_port_cntr(ppd, C_TX_FLIT_VL,
2429 idx_from_vl(vl)));
2430
2431 rsp->vls[vfi].port_vl_xmit_pkts =
2432 cpu_to_be64(read_port_cntr(ppd, C_TX_PKT_VL,
2433 idx_from_vl(vl)));
2434
2435 rsp->vls[vfi].port_vl_xmit_wait =
2436 cpu_to_be64(read_port_cntr(ppd, C_TX_WAIT_VL,
2437 idx_from_vl(vl)));
2438
2439 rsp->vls[vfi].port_vl_rcv_fecn =
2440 cpu_to_be64(read_dev_cntr(dd, C_DC_RCV_FCN_VL,
2441 idx_from_vl(vl)));
2442
2443 rsp->vls[vfi].port_vl_rcv_becn =
2444 cpu_to_be64(read_dev_cntr(dd, C_DC_RCV_BCN_VL,
2445 idx_from_vl(vl)));
2446
2447 vlinfo++;
2448 vfi++;
2449 }
2450
2451 a0_portstatus(ppd, rsp, vl_select_mask);
2452
2453 if (resp_len)
2454 *resp_len += response_data_size;
2455
2456 return reply((struct ib_mad_hdr *)pmp);
2457}
2458
f0852922
AL		 2459static u64 get_error_counter_summary(struct ib_device *ibdev, u8 port,
2460 u8 res_lli, u8 res_ler)
77241056
MM		 2461{
2462 struct hfi1_devdata *dd = dd_from_ibdev(ibdev);
2463 struct hfi1_ibport *ibp = to_iport(ibdev, port);
2464 struct hfi1_pportdata *ppd = ppd_from_ibp(ibp);
2465 u64 error_counter_summary = 0, tmp;
2466
2467 error_counter_summary += read_port_cntr(ppd, C_SW_RCV_CSTR_ERR,
2468 CNTR_INVALID_VL);
2469 /* port_rcv_switch_relay_errors is 0 for HFIs */
2470 error_counter_summary += read_port_cntr(ppd, C_SW_XMIT_DSCD,
2471 CNTR_INVALID_VL);
2472 error_counter_summary += read_port_cntr(ppd, C_SW_XMIT_CSTR_ERR,
2473 CNTR_INVALID_VL);
2474 error_counter_summary += read_dev_cntr(dd, C_DC_RMT_PHY_ERR,
2475 CNTR_INVALID_VL);
f0852922
AL		 2476 /* local link integrity must be right-shifted by the lli resolution */
2477 tmp = read_dev_cntr(dd, C_DC_RX_REPLAY, CNTR_INVALID_VL);
2478 tmp += read_dev_cntr(dd, C_DC_TX_REPLAY, CNTR_INVALID_VL);
2479 error_counter_summary += (tmp >> res_lli);
2480 /* link error recovery must b right-shifted by the ler resolution */
2481 tmp = read_dev_cntr(dd, C_DC_SEQ_CRC_CNT, CNTR_INVALID_VL);
2482 tmp += read_dev_cntr(dd, C_DC_REINIT_FROM_PEER_CNT, CNTR_INVALID_VL);
2483 error_counter_summary += (tmp >> res_ler);
77241056
MM		 2484 error_counter_summary += read_dev_cntr(dd, C_DC_RCV_ERR,
2485 CNTR_INVALID_VL);
2486 error_counter_summary += read_dev_cntr(dd, C_RCV_OVF, CNTR_INVALID_VL);
2487 error_counter_summary += read_dev_cntr(dd, C_DC_FM_CFG_ERR,
2488 CNTR_INVALID_VL);
2489 /* ppd->link_downed is a 32-bit value */
2490 error_counter_summary += read_port_cntr(ppd, C_SW_LINK_DOWN,
2491 CNTR_INVALID_VL);
2492 tmp = read_dev_cntr(dd, C_DC_UNC_ERR, CNTR_INVALID_VL);
2493 /* this is an 8-bit quantity */
2494 error_counter_summary += tmp < 0x100 ? (tmp & 0xff) : 0xff;
2495
2496 return error_counter_summary;
2497}
2498
f4ddedf4 2499static void a0_datacounters(struct hfi1_pportdata *ppd, struct _port_dctrs *rsp,
77241056
MM		 2500 u32 vl_select_mask)
2501{
f4ddedf4 2502 if (!is_bx(ppd->dd)) {
77241056 2503 unsigned long vl;
db00a055 2504 u64 sum_vl_xmit_wait = 0;
f4ddedf4 2505 u32 vl_all_mask = VL_MASK_ALL;
db00a055 2506
f4ddedf4
IW		 2507 for_each_set_bit(vl, (unsigned long *)&(vl_all_mask),
2508 8 * sizeof(vl_all_mask)) {
77241056 2509 u64 tmp = sum_vl_xmit_wait +
f4ddedf4
IW		 2510 read_port_cntr(ppd, C_TX_WAIT_VL,
2511 idx_from_vl(vl));
77241056
MM		 2512 if (tmp < sum_vl_xmit_wait) {
2513 /* we wrapped */
2514 sum_vl_xmit_wait = (u64) ~0;
2515 break;
2516 }
2517 sum_vl_xmit_wait = tmp;
2518 }
2519 if (be64_to_cpu(rsp->port_xmit_wait) > sum_vl_xmit_wait)
2520 rsp->port_xmit_wait = cpu_to_be64(sum_vl_xmit_wait);
2521 }
2522}
2523
2524static int pma_get_opa_datacounters(struct opa_pma_mad *pmp,
2525 struct ib_device *ibdev, u8 port, u32 *resp_len)
2526{
2527 struct opa_port_data_counters_msg *req =
2528 (struct opa_port_data_counters_msg *)pmp->data;
2529 struct hfi1_devdata *dd = dd_from_ibdev(ibdev);
2530 struct hfi1_ibport *ibp = to_iport(ibdev, port);
2531 struct hfi1_pportdata *ppd = ppd_from_ibp(ibp);
2532 struct _port_dctrs *rsp;
2533 struct _vls_dctrs *vlinfo;
2534 size_t response_data_size;
2535 u32 num_ports;
2536 u8 num_pslm;
2537 u8 lq, num_vls;
f0852922 2538 u8 res_lli, res_ler;
77241056
MM		 2539 u64 port_mask;
2540 unsigned long port_num;
2541 unsigned long vl;
2542 u32 vl_select_mask;
2543 int vfi;
2544
2545 num_ports = be32_to_cpu(pmp->mad_hdr.attr_mod) >> 24;
2546 num_pslm = hweight64(be64_to_cpu(req->port_select_mask[3]));
2547 num_vls = hweight32(be32_to_cpu(req->vl_select_mask));
2548 vl_select_mask = be32_to_cpu(req->vl_select_mask);
f0852922
AL		 2549 res_lli = (u8)(be32_to_cpu(req->resolution) & MSK_LLI) >> MSK_LLI_SFT;
2550 res_lli = res_lli ? res_lli + ADD_LLI : 0;
2551 res_ler = (u8)(be32_to_cpu(req->resolution) & MSK_LER) >> MSK_LER_SFT;
2552 res_ler = res_ler ? res_ler + ADD_LER : 0;
77241056
MM		 2553
2554 if (num_ports != 1 || (vl_select_mask & ~VL_MASK_ALL)) {
2555 pmp->mad_hdr.status |= IB_SMP_INVALID_FIELD;
2556 return reply((struct ib_mad_hdr *)pmp);
2557 }
2558
2559 /* Sanity check */
2560 response_data_size = sizeof(struct opa_port_data_counters_msg) +
2561 num_vls * sizeof(struct _vls_dctrs);
2562
2563 if (response_data_size > sizeof(pmp->data)) {
2564 pmp->mad_hdr.status |= IB_SMP_INVALID_FIELD;
2565 return reply((struct ib_mad_hdr *)pmp);
2566 }
2567
2568 /*
2569 * The bit set in the mask needs to be consistent with the
2570 * port the request came in on.
2571 */
2572 port_mask = be64_to_cpu(req->port_select_mask[3]);
2573 port_num = find_first_bit((unsigned long *)&port_mask,
2574 sizeof(port_mask));
2575
2576 if ((u8)port_num != port) {
2577 pmp->mad_hdr.status |= IB_SMP_INVALID_FIELD;
2578 return reply((struct ib_mad_hdr *)pmp);
2579 }
2580
2581 rsp = (struct _port_dctrs *)&(req->port[0]);
2582 memset(rsp, 0, sizeof(*rsp));
2583
2584 rsp->port_number = port;
2585 /*
2586 * Note that link_quality_indicator is a 32 bit quantity in
2587 * 'datacounters' queries (as opposed to 'portinfo' queries,
2588 * where it's a byte).
2589 */
2590 hfi1_read_link_quality(dd, &lq);
2591 rsp->link_quality_indicator = cpu_to_be32((u32)lq);
2592
2593 /* rsp->sw_port_congestion is 0 for HFIs */
2594 /* rsp->port_xmit_time_cong is 0 for HFIs */
2595 /* rsp->port_xmit_wasted_bw ??? */
2596 /* rsp->port_xmit_wait_data ??? */
2597 /* rsp->port_mark_fecn is 0 for HFIs */
2598
2599 rsp->port_xmit_data = cpu_to_be64(read_dev_cntr(dd, C_DC_XMIT_FLITS,
2600 CNTR_INVALID_VL));
2601 rsp->port_rcv_data = cpu_to_be64(read_dev_cntr(dd, C_DC_RCV_FLITS,
2602 CNTR_INVALID_VL));
77241056
MM		 2603 rsp->port_xmit_pkts = cpu_to_be64(read_dev_cntr(dd, C_DC_XMIT_PKTS,
2604 CNTR_INVALID_VL));
2605 rsp->port_rcv_pkts = cpu_to_be64(read_dev_cntr(dd, C_DC_RCV_PKTS,
2606 CNTR_INVALID_VL));
2607 rsp->port_multicast_xmit_pkts =
2608 cpu_to_be64(read_dev_cntr(dd, C_DC_MC_XMIT_PKTS,
2609 CNTR_INVALID_VL));
2610 rsp->port_multicast_rcv_pkts =
2611 cpu_to_be64(read_dev_cntr(dd, C_DC_MC_RCV_PKTS,
2612 CNTR_INVALID_VL));
2613 rsp->port_xmit_wait =
2614 cpu_to_be64(read_port_cntr(ppd, C_TX_WAIT, CNTR_INVALID_VL));
2615 rsp->port_rcv_fecn =
2616 cpu_to_be64(read_dev_cntr(dd, C_DC_RCV_FCN, CNTR_INVALID_VL));
2617 rsp->port_rcv_becn =
2618 cpu_to_be64(read_dev_cntr(dd, C_DC_RCV_BCN, CNTR_INVALID_VL));
2619
2620 rsp->port_error_counter_summary =
f0852922
AL		 2621 cpu_to_be64(get_error_counter_summary(ibdev, port,
2622 res_lli, res_ler));
77241056
MM		 2623
2624 vlinfo = &(rsp->vls[0]);
2625 vfi = 0;
2626 /* The vl_select_mask has been checked above, and we know
2627 * that it contains only entries which represent valid VLs.
2628 * So in the for_each_set_bit() loop below, we don't need
2629 * any additional checks for vl.
2630 */
2631 for_each_set_bit(vl, (unsigned long *)&(vl_select_mask),
2632 8 * sizeof(req->vl_select_mask)) {
2633 memset(vlinfo, 0, sizeof(*vlinfo));
2634
2635 rsp->vls[vfi].port_vl_xmit_data =
2636 cpu_to_be64(read_port_cntr(ppd, C_TX_FLIT_VL,
2637 idx_from_vl(vl)));
2638
2639 rsp->vls[vfi].port_vl_rcv_data =
2640 cpu_to_be64(read_dev_cntr(dd, C_DC_RX_FLIT_VL,
2641 idx_from_vl(vl)));
77241056
MM		 2642
2643 rsp->vls[vfi].port_vl_xmit_pkts =
2644 cpu_to_be64(read_port_cntr(ppd, C_TX_PKT_VL,
2645 idx_from_vl(vl)));
2646
2647 rsp->vls[vfi].port_vl_rcv_pkts =
2648 cpu_to_be64(read_dev_cntr(dd, C_DC_RX_PKT_VL,
2649 idx_from_vl(vl)));
2650
2651 rsp->vls[vfi].port_vl_xmit_wait =
2652 cpu_to_be64(read_port_cntr(ppd, C_TX_WAIT_VL,
2653 idx_from_vl(vl)));
2654
2655 rsp->vls[vfi].port_vl_rcv_fecn =
2656 cpu_to_be64(read_dev_cntr(dd, C_DC_RCV_FCN_VL,
2657 idx_from_vl(vl)));
2658 rsp->vls[vfi].port_vl_rcv_becn =
2659 cpu_to_be64(read_dev_cntr(dd, C_DC_RCV_BCN_VL,
2660 idx_from_vl(vl)));
2661
2662 /* rsp->port_vl_xmit_time_cong is 0 for HFIs */
2663 /* rsp->port_vl_xmit_wasted_bw ??? */
2664 /* port_vl_xmit_wait_data - TXE (table 13-9 HFI spec) ???
2665 * does this differ from rsp->vls[vfi].port_vl_xmit_wait */
2666 /*rsp->vls[vfi].port_vl_mark_fecn =
2667 cpu_to_be64(read_csr(dd, DCC_PRF_PORT_VL_MARK_FECN_CNT
2668 + offset));
2669 */
2670 vlinfo++;
2671 vfi++;
2672 }
2673
f4ddedf4 2674 a0_datacounters(ppd, rsp, vl_select_mask);
77241056
MM		 2675
2676 if (resp_len)
2677 *resp_len += response_data_size;
2678
2679 return reply((struct ib_mad_hdr *)pmp);
2680}
2681
2682static int pma_get_opa_porterrors(struct opa_pma_mad *pmp,
2683 struct ib_device *ibdev, u8 port, u32 *resp_len)
2684{
2685 size_t response_data_size;
2686 struct _port_ectrs *rsp;
2687 unsigned long port_num;
2688 struct opa_port_error_counters64_msg *req;
2689 struct hfi1_devdata *dd = dd_from_ibdev(ibdev);
2690 u32 num_ports;
2691 u8 num_pslm;
2692 u8 num_vls;
2693 struct hfi1_ibport *ibp;
2694 struct hfi1_pportdata *ppd;
2695 struct _vls_ectrs *vlinfo;
2696 unsigned long vl;
2697 u64 port_mask, tmp, tmp2;
2698 u32 vl_select_mask;
2699 int vfi;
2700
2701 req = (struct opa_port_error_counters64_msg *)pmp->data;
2702
2703 num_ports = be32_to_cpu(pmp->mad_hdr.attr_mod) >> 24;
2704
2705 num_pslm = hweight64(be64_to_cpu(req->port_select_mask[3]));
2706 num_vls = hweight32(be32_to_cpu(req->vl_select_mask));
2707
2708 if (num_ports != 1 || num_ports != num_pslm) {
2709 pmp->mad_hdr.status |= IB_SMP_INVALID_FIELD;
2710 return reply((struct ib_mad_hdr *)pmp);
2711 }
2712
2713 response_data_size = sizeof(struct opa_port_error_counters64_msg) +
2714 num_vls * sizeof(struct _vls_ectrs);
2715
2716 if (response_data_size > sizeof(pmp->data)) {
2717 pmp->mad_hdr.status |= IB_SMP_INVALID_FIELD;
2718 return reply((struct ib_mad_hdr *)pmp);
2719 }
2720 /*
2721 * The bit set in the mask needs to be consistent with the
2722 * port the request came in on.
2723 */
2724 port_mask = be64_to_cpu(req->port_select_mask[3]);
2725 port_num = find_first_bit((unsigned long *)&port_mask,
2726 sizeof(port_mask));
2727
2728 if ((u8)port_num != port) {
2729 pmp->mad_hdr.status |= IB_SMP_INVALID_FIELD;
2730 return reply((struct ib_mad_hdr *)pmp);
2731 }
2732
2733 rsp = (struct _port_ectrs *)&(req->port[0]);
2734
2735 ibp = to_iport(ibdev, port_num);
2736 ppd = ppd_from_ibp(ibp);
2737
2738 memset(rsp, 0, sizeof(*rsp));
2739 rsp->port_number = (u8)port_num;
2740
2741 rsp->port_rcv_constraint_errors =
2742 cpu_to_be64(read_port_cntr(ppd, C_SW_RCV_CSTR_ERR,
2743 CNTR_INVALID_VL));
2744 /* port_rcv_switch_relay_errors is 0 for HFIs */
2745 rsp->port_xmit_discards =
2746 cpu_to_be64(read_port_cntr(ppd, C_SW_XMIT_DSCD,
2747 CNTR_INVALID_VL));
2748 rsp->port_rcv_remote_physical_errors =
2749 cpu_to_be64(read_dev_cntr(dd, C_DC_RMT_PHY_ERR,
2750 CNTR_INVALID_VL));
2751 tmp = read_dev_cntr(dd, C_DC_RX_REPLAY, CNTR_INVALID_VL);
2752 tmp2 = tmp + read_dev_cntr(dd, C_DC_TX_REPLAY, CNTR_INVALID_VL);
2753 if (tmp2 < tmp) {
2754 /* overflow/wrapped */
2755 rsp->local_link_integrity_errors = cpu_to_be64(~0);
2756 } else {
2757 rsp->local_link_integrity_errors = cpu_to_be64(tmp2);
2758 }
2759 tmp = read_dev_cntr(dd, C_DC_SEQ_CRC_CNT, CNTR_INVALID_VL);
2760 tmp2 = tmp + read_dev_cntr(dd, C_DC_REINIT_FROM_PEER_CNT,
2761 CNTR_INVALID_VL);
2762 if (tmp2 > (u32)UINT_MAX || tmp2 < tmp) {
2763 /* overflow/wrapped */
2764 rsp->link_error_recovery = cpu_to_be32(~0);
2765 } else {
2766 rsp->link_error_recovery = cpu_to_be32(tmp2);
2767 }
2768 rsp->port_xmit_constraint_errors =
2769 cpu_to_be64(read_port_cntr(ppd, C_SW_XMIT_CSTR_ERR,
2770 CNTR_INVALID_VL));
2771 rsp->excessive_buffer_overruns =
2772 cpu_to_be64(read_dev_cntr(dd, C_RCV_OVF, CNTR_INVALID_VL));
2773 rsp->fm_config_errors =
2774 cpu_to_be64(read_dev_cntr(dd, C_DC_FM_CFG_ERR,
2775 CNTR_INVALID_VL));
2776 rsp->link_downed = cpu_to_be32(read_port_cntr(ppd, C_SW_LINK_DOWN,
2777 CNTR_INVALID_VL));
2778 tmp = read_dev_cntr(dd, C_DC_UNC_ERR, CNTR_INVALID_VL);
2779 rsp->uncorrectable_errors = tmp < 0x100 ? (tmp & 0xff) : 0xff;
2780
2781 vlinfo = (struct _vls_ectrs *)&(rsp->vls[0]);
2782 vfi = 0;
2783 vl_select_mask = be32_to_cpu(req->vl_select_mask);
2784 for_each_set_bit(vl, (unsigned long *)&(vl_select_mask),
2785 8 * sizeof(req->vl_select_mask)) {
2786 memset(vlinfo, 0, sizeof(*vlinfo));
2787 /* vlinfo->vls[vfi].port_vl_xmit_discards ??? */
2788 vlinfo += 1;
2789 vfi++;
2790 }
2791
2792 if (resp_len)
2793 *resp_len += response_data_size;
2794
2795 return reply((struct ib_mad_hdr *)pmp);
2796}
2797
2798static int pma_get_opa_errorinfo(struct opa_pma_mad *pmp,
2799 struct ib_device *ibdev, u8 port, u32 *resp_len)
2800{
2801 size_t response_data_size;
2802 struct _port_ei *rsp;
2803 struct opa_port_error_info_msg *req;
2804 struct hfi1_devdata *dd = dd_from_ibdev(ibdev);
2805 u64 port_mask;
2806 u32 num_ports;
2807 unsigned long port_num;
2808 u8 num_pslm;
2809 u64 reg;
2810
2811 req = (struct opa_port_error_info_msg *)pmp->data;
2812 rsp = (struct _port_ei *)&(req->port[0]);
2813
2814 num_ports = OPA_AM_NPORT(be32_to_cpu(pmp->mad_hdr.attr_mod));
2815 num_pslm = hweight64(be64_to_cpu(req->port_select_mask[3]));
2816
2817 memset(rsp, 0, sizeof(*rsp));
2818
2819 if (num_ports != 1 || num_ports != num_pslm) {
2820 pmp->mad_hdr.status |= IB_SMP_INVALID_FIELD;
2821 return reply((struct ib_mad_hdr *)pmp);
2822 }
2823
2824 /* Sanity check */
2825 response_data_size = sizeof(struct opa_port_error_info_msg);
2826
2827 if (response_data_size > sizeof(pmp->data)) {
2828 pmp->mad_hdr.status |= IB_SMP_INVALID_FIELD;
2829 return reply((struct ib_mad_hdr *)pmp);
2830 }
2831
2832 /*
2833 * The bit set in the mask needs to be consistent with the port
2834 * the request came in on.
2835 */
2836 port_mask = be64_to_cpu(req->port_select_mask[3]);
2837 port_num = find_first_bit((unsigned long *)&port_mask,
2838 sizeof(port_mask));
2839
2840 if ((u8)port_num != port) {
2841 pmp->mad_hdr.status |= IB_SMP_INVALID_FIELD;
2842 return reply((struct ib_mad_hdr *)pmp);
2843 }
2844
2845 /* PortRcvErrorInfo */
2846 rsp->port_rcv_ei.status_and_code =
2847 dd->err_info_rcvport.status_and_code;
2848 memcpy(&rsp->port_rcv_ei.ei.ei1to12.packet_flit1,
2849 &dd->err_info_rcvport.packet_flit1, sizeof(u64));
2850 memcpy(&rsp->port_rcv_ei.ei.ei1to12.packet_flit2,
2851 &dd->err_info_rcvport.packet_flit2, sizeof(u64));
2852
2853 /* ExcessiverBufferOverrunInfo */
2854 reg = read_csr(dd, RCV_ERR_INFO);
2855 if (reg & RCV_ERR_INFO_RCV_EXCESS_BUFFER_OVERRUN_SMASK) {
2856 /* if the RcvExcessBufferOverrun bit is set, save SC of
2857 * first pkt that encountered an excess buffer overrun */
2858 u8 tmp = (u8)reg;
2859
2860 tmp &= RCV_ERR_INFO_RCV_EXCESS_BUFFER_OVERRUN_SC_SMASK;
2861 tmp <<= 2;
2862 rsp->excessive_buffer_overrun_ei.status_and_sc = tmp;
2863 /* set the status bit */
2864 rsp->excessive_buffer_overrun_ei.status_and_sc |= 0x80;
2865 }
2866
2867 rsp->port_xmit_constraint_ei.status =
2868 dd->err_info_xmit_constraint.status;
2869 rsp->port_xmit_constraint_ei.pkey =
2870 cpu_to_be16(dd->err_info_xmit_constraint.pkey);
2871 rsp->port_xmit_constraint_ei.slid =
2872 cpu_to_be32(dd->err_info_xmit_constraint.slid);
2873
2874 rsp->port_rcv_constraint_ei.status =
2875 dd->err_info_rcv_constraint.status;
2876 rsp->port_rcv_constraint_ei.pkey =
2877 cpu_to_be16(dd->err_info_rcv_constraint.pkey);
2878 rsp->port_rcv_constraint_ei.slid =
2879 cpu_to_be32(dd->err_info_rcv_constraint.slid);
2880
2881 /* UncorrectableErrorInfo */
2882 rsp->uncorrectable_ei.status_and_code = dd->err_info_uncorrectable;
2883
2884 /* FMConfigErrorInfo */
2885 rsp->fm_config_ei.status_and_code = dd->err_info_fmconfig;
2886
2887 if (resp_len)
2888 *resp_len += response_data_size;
2889
2890 return reply((struct ib_mad_hdr *)pmp);
2891}
2892
2893static int pma_set_opa_portstatus(struct opa_pma_mad *pmp,
2894 struct ib_device *ibdev, u8 port, u32 *resp_len)
2895{
2896 struct opa_clear_port_status *req =
2897 (struct opa_clear_port_status *)pmp->data;
2898 struct hfi1_devdata *dd = dd_from_ibdev(ibdev);
2899 struct hfi1_ibport *ibp = to_iport(ibdev, port);
2900 struct hfi1_pportdata *ppd = ppd_from_ibp(ibp);
2901 u32 nports = be32_to_cpu(pmp->mad_hdr.attr_mod) >> 24;
2902 u64 portn = be64_to_cpu(req->port_select_mask[3]);
2903 u32 counter_select = be32_to_cpu(req->counter_select_mask);
2904 u32 vl_select_mask = VL_MASK_ALL; /* clear all per-vl cnts */
2905 unsigned long vl;
2906
2907 if ((nports != 1) || (portn != 1 << port)) {
2908 pmp->mad_hdr.status |= IB_SMP_INVALID_FIELD;
2909 return reply((struct ib_mad_hdr *)pmp);
2910 }
2911 /*
2912 * only counters returned by pma_get_opa_portstatus() are
2913 * handled, so when pma_get_opa_portstatus() gets a fix,
2914 * the corresponding change should be made here as well.
2915 */
2916
2917 if (counter_select & CS_PORT_XMIT_DATA)
2918 write_dev_cntr(dd, C_DC_XMIT_FLITS, CNTR_INVALID_VL, 0);
2919
2920 if (counter_select & CS_PORT_RCV_DATA)
2921 write_dev_cntr(dd, C_DC_RCV_FLITS, CNTR_INVALID_VL, 0);
2922
2923 if (counter_select & CS_PORT_XMIT_PKTS)
2924 write_dev_cntr(dd, C_DC_XMIT_PKTS, CNTR_INVALID_VL, 0);
2925
2926 if (counter_select & CS_PORT_RCV_PKTS)
2927 write_dev_cntr(dd, C_DC_RCV_PKTS, CNTR_INVALID_VL, 0);
2928
2929 if (counter_select & CS_PORT_MCAST_XMIT_PKTS)
2930 write_dev_cntr(dd, C_DC_MC_XMIT_PKTS, CNTR_INVALID_VL, 0);
2931
2932 if (counter_select & CS_PORT_MCAST_RCV_PKTS)
2933 write_dev_cntr(dd, C_DC_MC_RCV_PKTS, CNTR_INVALID_VL, 0);
2934
2935 if (counter_select & CS_PORT_XMIT_WAIT)
2936 write_port_cntr(ppd, C_TX_WAIT, CNTR_INVALID_VL, 0);
2937
2938 /* ignore cs_sw_portCongestion for HFIs */
2939
2940 if (counter_select & CS_PORT_RCV_FECN)
2941 write_dev_cntr(dd, C_DC_RCV_FCN, CNTR_INVALID_VL, 0);
2942
2943 if (counter_select & CS_PORT_RCV_BECN)
2944 write_dev_cntr(dd, C_DC_RCV_BCN, CNTR_INVALID_VL, 0);
2945
2946 /* ignore cs_port_xmit_time_cong for HFIs */
2947 /* ignore cs_port_xmit_wasted_bw for now */
2948 /* ignore cs_port_xmit_wait_data for now */
2949 if (counter_select & CS_PORT_RCV_BUBBLE)
2950 write_dev_cntr(dd, C_DC_RCV_BBL, CNTR_INVALID_VL, 0);
2951
2952 /* Only applicable for switch */
2953 /*if (counter_select & CS_PORT_MARK_FECN)
2954 write_csr(dd, DCC_PRF_PORT_MARK_FECN_CNT, 0);*/
2955
2956 if (counter_select & CS_PORT_RCV_CONSTRAINT_ERRORS)
2957 write_port_cntr(ppd, C_SW_RCV_CSTR_ERR, CNTR_INVALID_VL, 0);
2958
2959 /* ignore cs_port_rcv_switch_relay_errors for HFIs */
2960 if (counter_select & CS_PORT_XMIT_DISCARDS)
2961 write_port_cntr(ppd, C_SW_XMIT_DSCD, CNTR_INVALID_VL, 0);
2962
2963 if (counter_select & CS_PORT_XMIT_CONSTRAINT_ERRORS)
2964 write_port_cntr(ppd, C_SW_XMIT_CSTR_ERR, CNTR_INVALID_VL, 0);
2965
2966 if (counter_select & CS_PORT_RCV_REMOTE_PHYSICAL_ERRORS)
2967 write_dev_cntr(dd, C_DC_RMT_PHY_ERR, CNTR_INVALID_VL, 0);
2968
2969 if (counter_select & CS_LOCAL_LINK_INTEGRITY_ERRORS) {
2970 write_dev_cntr(dd, C_DC_TX_REPLAY, CNTR_INVALID_VL, 0);
2971 write_dev_cntr(dd, C_DC_RX_REPLAY, CNTR_INVALID_VL, 0);
2972 }
2973
2974 if (counter_select & CS_LINK_ERROR_RECOVERY) {
2975 write_dev_cntr(dd, C_DC_SEQ_CRC_CNT, CNTR_INVALID_VL, 0);
2976 write_dev_cntr(dd, C_DC_REINIT_FROM_PEER_CNT,
2977 CNTR_INVALID_VL, 0);
2978 }
2979
2980 if (counter_select & CS_PORT_RCV_ERRORS)
2981 write_dev_cntr(dd, C_DC_RCV_ERR, CNTR_INVALID_VL, 0);
2982
2983 if (counter_select & CS_EXCESSIVE_BUFFER_OVERRUNS) {
2984 write_dev_cntr(dd, C_RCV_OVF, CNTR_INVALID_VL, 0);
2985 dd->rcv_ovfl_cnt = 0;
2986 }
2987
2988 if (counter_select & CS_FM_CONFIG_ERRORS)
2989 write_dev_cntr(dd, C_DC_FM_CFG_ERR, CNTR_INVALID_VL, 0);
2990
2991 if (counter_select & CS_LINK_DOWNED)
2992 write_port_cntr(ppd, C_SW_LINK_DOWN, CNTR_INVALID_VL, 0);
2993
2994 if (counter_select & CS_UNCORRECTABLE_ERRORS)
2995 write_dev_cntr(dd, C_DC_UNC_ERR, CNTR_INVALID_VL, 0);
2996
2997 for_each_set_bit(vl, (unsigned long *)&(vl_select_mask),
2998 8 * sizeof(vl_select_mask)) {
2999
3000 if (counter_select & CS_PORT_XMIT_DATA)
3001 write_port_cntr(ppd, C_TX_FLIT_VL, idx_from_vl(vl), 0);
3002
3003 if (counter_select & CS_PORT_RCV_DATA)
3004 write_dev_cntr(dd, C_DC_RX_FLIT_VL, idx_from_vl(vl), 0);
3005
3006 if (counter_select & CS_PORT_XMIT_PKTS)
3007 write_port_cntr(ppd, C_TX_PKT_VL, idx_from_vl(vl), 0);
3008
3009 if (counter_select & CS_PORT_RCV_PKTS)
3010 write_dev_cntr(dd, C_DC_RX_PKT_VL, idx_from_vl(vl), 0);
3011
3012 if (counter_select & CS_PORT_XMIT_WAIT)
3013 write_port_cntr(ppd, C_TX_WAIT_VL, idx_from_vl(vl), 0);
3014
3015 /* sw_port_vl_congestion is 0 for HFIs */
3016 if (counter_select & CS_PORT_RCV_FECN)
3017 write_dev_cntr(dd, C_DC_RCV_FCN_VL, idx_from_vl(vl), 0);
3018
3019 if (counter_select & CS_PORT_RCV_BECN)
3020 write_dev_cntr(dd, C_DC_RCV_BCN_VL, idx_from_vl(vl), 0);
3021
3022 /* port_vl_xmit_time_cong is 0 for HFIs */
3023 /* port_vl_xmit_wasted_bw ??? */
3024 /* port_vl_xmit_wait_data - TXE (table 13-9 HFI spec) ??? */
3025 if (counter_select & CS_PORT_RCV_BUBBLE)
3026 write_dev_cntr(dd, C_DC_RCV_BBL_VL, idx_from_vl(vl), 0);
3027
3028 /*if (counter_select & CS_PORT_MARK_FECN)
3029 write_csr(dd, DCC_PRF_PORT_VL_MARK_FECN_CNT + offset, 0);
3030 */
3031 /* port_vl_xmit_discards ??? */
3032 }
3033
3034 if (resp_len)
3035 *resp_len += sizeof(*req);
3036
3037 return reply((struct ib_mad_hdr *)pmp);
3038}
3039
3040static int pma_set_opa_errorinfo(struct opa_pma_mad *pmp,
3041 struct ib_device *ibdev, u8 port, u32 *resp_len)
3042{
3043 struct _port_ei *rsp;
3044 struct opa_port_error_info_msg *req;
3045 struct hfi1_devdata *dd = dd_from_ibdev(ibdev);
3046 u64 port_mask;
3047 u32 num_ports;
3048 unsigned long port_num;
3049 u8 num_pslm;
3050 u32 error_info_select;
3051
3052 req = (struct opa_port_error_info_msg *)pmp->data;
3053 rsp = (struct _port_ei *)&(req->port[0]);
3054
3055 num_ports = OPA_AM_NPORT(be32_to_cpu(pmp->mad_hdr.attr_mod));
3056 num_pslm = hweight64(be64_to_cpu(req->port_select_mask[3]));
3057
3058 memset(rsp, 0, sizeof(*rsp));
3059
3060 if (num_ports != 1 || num_ports != num_pslm) {
3061 pmp->mad_hdr.status |= IB_SMP_INVALID_FIELD;
3062 return reply((struct ib_mad_hdr *)pmp);
3063 }
3064
3065 /*
3066 * The bit set in the mask needs to be consistent with the port
3067 * the request came in on.
3068 */
3069 port_mask = be64_to_cpu(req->port_select_mask[3]);
3070 port_num = find_first_bit((unsigned long *)&port_mask,
3071 sizeof(port_mask));
3072
3073 if ((u8)port_num != port) {
3074 pmp->mad_hdr.status |= IB_SMP_INVALID_FIELD;
3075 return reply((struct ib_mad_hdr *)pmp);
3076 }
3077
3078 error_info_select = be32_to_cpu(req->error_info_select_mask);
3079
3080 /* PortRcvErrorInfo */
3081 if (error_info_select & ES_PORT_RCV_ERROR_INFO)
3082 /* turn off status bit */
3083 dd->err_info_rcvport.status_and_code &= ~OPA_EI_STATUS_SMASK;
3084
3085 /* ExcessiverBufferOverrunInfo */
3086 if (error_info_select & ES_EXCESSIVE_BUFFER_OVERRUN_INFO)
3087 /* status bit is essentially kept in the h/w - bit 5 of
3088 * RCV_ERR_INFO */
3089 write_csr(dd, RCV_ERR_INFO,
3090 RCV_ERR_INFO_RCV_EXCESS_BUFFER_OVERRUN_SMASK);
3091
3092 if (error_info_select & ES_PORT_XMIT_CONSTRAINT_ERROR_INFO)
3093 dd->err_info_xmit_constraint.status &= ~OPA_EI_STATUS_SMASK;
3094
3095 if (error_info_select & ES_PORT_RCV_CONSTRAINT_ERROR_INFO)
3096 dd->err_info_rcv_constraint.status &= ~OPA_EI_STATUS_SMASK;
3097
3098 /* UncorrectableErrorInfo */
3099 if (error_info_select & ES_UNCORRECTABLE_ERROR_INFO)
3100 /* turn off status bit */
3101 dd->err_info_uncorrectable &= ~OPA_EI_STATUS_SMASK;
3102
3103 /* FMConfigErrorInfo */
3104 if (error_info_select & ES_FM_CONFIG_ERROR_INFO)
3105 /* turn off status bit */
3106 dd->err_info_fmconfig &= ~OPA_EI_STATUS_SMASK;
3107
3108 if (resp_len)
3109 *resp_len += sizeof(*req);
3110
3111 return reply((struct ib_mad_hdr *)pmp);
3112}
3113
3114struct opa_congestion_info_attr {
3115 __be16 congestion_info;
3116 u8 control_table_cap; /* Multiple of 64 entry unit CCTs */
3117 u8 congestion_log_length;
3118} __packed;
3119
3120static int __subn_get_opa_cong_info(struct opa_smp *smp, u32 am, u8 *data,
3121 struct ib_device *ibdev, u8 port,
3122 u32 *resp_len)
3123{
3124 struct opa_congestion_info_attr *p =
3125 (struct opa_congestion_info_attr *)data;
3126 struct hfi1_ibport *ibp = to_iport(ibdev, port);
3127 struct hfi1_pportdata *ppd = ppd_from_ibp(ibp);
3128
3129 p->congestion_info = 0;
3130 p->control_table_cap = ppd->cc_max_table_entries;
3131 p->congestion_log_length = OPA_CONG_LOG_ELEMS;
3132
3133 if (resp_len)
3134 *resp_len += sizeof(*p);
3135
3136 return reply((struct ib_mad_hdr *)smp);
3137}
3138
3139static int __subn_get_opa_cong_setting(struct opa_smp *smp, u32 am,
3140 u8 *data,
3141 struct ib_device *ibdev,
3142 u8 port, u32 *resp_len)
3143{
3144 int i;
3145 struct opa_congestion_setting_attr *p =
3146 (struct opa_congestion_setting_attr *) data;
3147 struct hfi1_ibport *ibp = to_iport(ibdev, port);
3148 struct hfi1_pportdata *ppd = ppd_from_ibp(ibp);
3149 struct opa_congestion_setting_entry_shadow *entries;
3150 struct cc_state *cc_state;
3151
3152 rcu_read_lock();
3153
3154 cc_state = get_cc_state(ppd);
3155
3156 if (cc_state == NULL) {
3157 rcu_read_unlock();
3158 return reply((struct ib_mad_hdr *)smp);
3159 }
3160
3161 entries = cc_state->cong_setting.entries;
3162 p->port_control = cpu_to_be16(cc_state->cong_setting.port_control);
3163 p->control_map = cpu_to_be32(cc_state->cong_setting.control_map);
3164 for (i = 0; i < OPA_MAX_SLS; i++) {
3165 p->entries[i].ccti_increase = entries[i].ccti_increase;
3166 p->entries[i].ccti_timer = cpu_to_be16(entries[i].ccti_timer);
3167 p->entries[i].trigger_threshold =
3168 entries[i].trigger_threshold;
3169 p->entries[i].ccti_min = entries[i].ccti_min;
3170 }
3171
3172 rcu_read_unlock();
3173
3174 if (resp_len)
3175 *resp_len += sizeof(*p);
3176
3177 return reply((struct ib_mad_hdr *)smp);
3178}
3179
3180static int __subn_set_opa_cong_setting(struct opa_smp *smp, u32 am, u8 *data,
3181 struct ib_device *ibdev, u8 port,
3182 u32 *resp_len)
3183{
3184 struct opa_congestion_setting_attr *p =
3185 (struct opa_congestion_setting_attr *) data;
3186 struct hfi1_ibport *ibp = to_iport(ibdev, port);
3187 struct hfi1_pportdata *ppd = ppd_from_ibp(ibp);
3188 struct opa_congestion_setting_entry_shadow *entries;
3189 int i;
3190
3191 ppd->cc_sl_control_map = be32_to_cpu(p->control_map);
3192
3193 entries = ppd->congestion_entries;
3194 for (i = 0; i < OPA_MAX_SLS; i++) {
3195 entries[i].ccti_increase = p->entries[i].ccti_increase;
3196 entries[i].ccti_timer = be16_to_cpu(p->entries[i].ccti_timer);
3197 entries[i].trigger_threshold =
3198 p->entries[i].trigger_threshold;
3199 entries[i].ccti_min = p->entries[i].ccti_min;
3200 }
3201
3202 return __subn_get_opa_cong_setting(smp, am, data, ibdev, port,
3203 resp_len);
3204}
3205
3206static int __subn_get_opa_hfi1_cong_log(struct opa_smp *smp, u32 am,
3207 u8 *data, struct ib_device *ibdev,
3208 u8 port, u32 *resp_len)
3209{
3210 struct hfi1_ibport *ibp = to_iport(ibdev, port);
3211 struct hfi1_pportdata *ppd = ppd_from_ibp(ibp);
3212 struct opa_hfi1_cong_log *cong_log = (struct opa_hfi1_cong_log *)data;
3213 s64 ts;
3214 int i;
3215
3216 if (am != 0) {
3217 smp->status |= IB_SMP_INVALID_FIELD;
3218 return reply((struct ib_mad_hdr *)smp);
3219 }
3220
b77d713a 3221 spin_lock_irq(&ppd->cc_log_lock);
77241056
MM		 3222
3223 cong_log->log_type = OPA_CC_LOG_TYPE_HFI;
3224 cong_log->congestion_flags = 0;
3225 cong_log->threshold_event_counter =
3226 cpu_to_be16(ppd->threshold_event_counter);
3227 memcpy(cong_log->threshold_cong_event_map,
3228 ppd->threshold_cong_event_map,
3229 sizeof(cong_log->threshold_cong_event_map));
3230 /* keep timestamp in units of 1.024 usec */
3231 ts = ktime_to_ns(ktime_get()) / 1024;
3232 cong_log->current_time_stamp = cpu_to_be32(ts);
3233 for (i = 0; i < OPA_CONG_LOG_ELEMS; i++) {
3234 struct opa_hfi1_cong_log_event_internal *cce =
3235 &ppd->cc_events[ppd->cc_mad_idx++];
3236 if (ppd->cc_mad_idx == OPA_CONG_LOG_ELEMS)
3237 ppd->cc_mad_idx = 0;
3238 /*
3239 * Entries which are older than twice the time
3240 * required to wrap the counter are supposed to
3241 * be zeroed (CA10-49 IBTA, release 1.2.1, V1).
3242 */
3243 if ((u64)(ts - cce->timestamp) > (2 * UINT_MAX))
3244 continue;
3245 memcpy(cong_log->events[i].local_qp_cn_entry, &cce->lqpn, 3);
3246 memcpy(cong_log->events[i].remote_qp_number_cn_entry,
3247 &cce->rqpn, 3);
3248 cong_log->events[i].sl_svc_type_cn_entry =
3249 ((cce->sl & 0x1f) << 3) | (cce->svc_type & 0x7);
3250 cong_log->events[i].remote_lid_cn_entry =
3251 cpu_to_be32(cce->rlid);
3252 cong_log->events[i].timestamp_cn_entry =
3253 cpu_to_be32(cce->timestamp);
3254 }
3255
3256 /*
3257 * Reset threshold_cong_event_map, and threshold_event_counter
3258 * to 0 when log is read.
3259 */
3260 memset(ppd->threshold_cong_event_map, 0x0,
3261 sizeof(ppd->threshold_cong_event_map));
3262 ppd->threshold_event_counter = 0;
3263
b77d713a 3264 spin_unlock_irq(&ppd->cc_log_lock);
77241056
MM		 3265
3266 if (resp_len)
3267 *resp_len += sizeof(struct opa_hfi1_cong_log);
3268
3269 return reply((struct ib_mad_hdr *)smp);
3270}
3271
3272static int __subn_get_opa_cc_table(struct opa_smp *smp, u32 am, u8 *data,
3273 struct ib_device *ibdev, u8 port,
3274 u32 *resp_len)
3275{
3276 struct ib_cc_table_attr *cc_table_attr =
3277 (struct ib_cc_table_attr *) data;
3278 struct hfi1_ibport *ibp = to_iport(ibdev, port);
3279 struct hfi1_pportdata *ppd = ppd_from_ibp(ibp);
3280 u32 start_block = OPA_AM_START_BLK(am);
3281 u32 n_blocks = OPA_AM_NBLK(am);
3282 struct ib_cc_table_entry_shadow *entries;
3283 int i, j;
3284 u32 sentry, eentry;
3285 struct cc_state *cc_state;
3286
3287 /* sanity check n_blocks, start_block */
3288 if (n_blocks == 0 ||
3289 start_block + n_blocks > ppd->cc_max_table_entries) {
3290 smp->status |= IB_SMP_INVALID_FIELD;
3291 return reply((struct ib_mad_hdr *)smp);
3292 }
3293
3294 rcu_read_lock();
3295
3296 cc_state = get_cc_state(ppd);
3297
3298 if (cc_state == NULL) {
3299 rcu_read_unlock();
3300 return reply((struct ib_mad_hdr *)smp);
3301 }
3302
3303 sentry = start_block * IB_CCT_ENTRIES;
3304 eentry = sentry + (IB_CCT_ENTRIES * n_blocks);
3305
3306 cc_table_attr->ccti_limit = cpu_to_be16(cc_state->cct.ccti_limit);
3307
3308 entries = cc_state->cct.entries;
3309
3310 /* return n_blocks, though the last block may not be full */
3311 for (j = 0, i = sentry; i < eentry; j++, i++)
3312 cc_table_attr->ccti_entries[j].entry =
3313 cpu_to_be16(entries[i].entry);
3314
3315 rcu_read_unlock();
3316
3317 if (resp_len)
3318 *resp_len += sizeof(u16)*(IB_CCT_ENTRIES * n_blocks + 1);
3319
3320 return reply((struct ib_mad_hdr *)smp);
3321}
3322
3323void cc_state_reclaim(struct rcu_head *rcu)
3324{
3325 struct cc_state *cc_state = container_of(rcu, struct cc_state, rcu);
3326
3327 kfree(cc_state);
3328}
3329
3330static int __subn_set_opa_cc_table(struct opa_smp *smp, u32 am, u8 *data,
3331 struct ib_device *ibdev, u8 port,
3332 u32 *resp_len)
3333{
3334 struct ib_cc_table_attr *p = (struct ib_cc_table_attr *) data;
3335 struct hfi1_ibport *ibp = to_iport(ibdev, port);
3336 struct hfi1_pportdata *ppd = ppd_from_ibp(ibp);
3337 u32 start_block = OPA_AM_START_BLK(am);
3338 u32 n_blocks = OPA_AM_NBLK(am);
3339 struct ib_cc_table_entry_shadow *entries;
3340 int i, j;
3341 u32 sentry, eentry;
3342 u16 ccti_limit;
3343 struct cc_state *old_cc_state, *new_cc_state;
3344
3345 /* sanity check n_blocks, start_block */
3346 if (n_blocks == 0 ||
3347 start_block + n_blocks > ppd->cc_max_table_entries) {
3348 smp->status |= IB_SMP_INVALID_FIELD;
3349 return reply((struct ib_mad_hdr *)smp);
3350 }
3351
3352 sentry = start_block * IB_CCT_ENTRIES;
3353 eentry = sentry + ((n_blocks - 1) * IB_CCT_ENTRIES) +
3354 (be16_to_cpu(p->ccti_limit)) % IB_CCT_ENTRIES + 1;
3355
3356 /* sanity check ccti_limit */
3357 ccti_limit = be16_to_cpu(p->ccti_limit);
3358 if (ccti_limit + 1 > eentry) {
3359 smp->status |= IB_SMP_INVALID_FIELD;
3360 return reply((struct ib_mad_hdr *)smp);
3361 }
3362
3363 new_cc_state = kzalloc(sizeof(*new_cc_state), GFP_KERNEL);
3364 if (new_cc_state == NULL)
3365 goto getit;
3366
3367 spin_lock(&ppd->cc_state_lock);
3368
3369 old_cc_state = get_cc_state(ppd);
3370
3371 if (old_cc_state == NULL) {
3372 spin_unlock(&ppd->cc_state_lock);
3373 kfree(new_cc_state);
3374 return reply((struct ib_mad_hdr *)smp);
3375 }
3376
3377 *new_cc_state = *old_cc_state;
3378
3379 new_cc_state->cct.ccti_limit = ccti_limit;
3380
3381 entries = ppd->ccti_entries;
3382 ppd->total_cct_entry = ccti_limit + 1;
3383
3384 for (j = 0, i = sentry; i < eentry; j++, i++)
3385 entries[i].entry = be16_to_cpu(p->ccti_entries[j].entry);
3386
3387 memcpy(new_cc_state->cct.entries, entries,
3388 eentry * sizeof(struct ib_cc_table_entry));
3389
3390 new_cc_state->cong_setting.port_control = IB_CC_CCS_PC_SL_BASED;
3391 new_cc_state->cong_setting.control_map = ppd->cc_sl_control_map;
3392 memcpy(new_cc_state->cong_setting.entries, ppd->congestion_entries,
3393 OPA_MAX_SLS * sizeof(struct opa_congestion_setting_entry));
3394
3395 rcu_assign_pointer(ppd->cc_state, new_cc_state);
3396
3397 spin_unlock(&ppd->cc_state_lock);
3398
3399 call_rcu(&old_cc_state->rcu, cc_state_reclaim);
3400
3401getit:
3402 return __subn_get_opa_cc_table(smp, am, data, ibdev, port, resp_len);
3403}
3404
3405struct opa_led_info {
3406 __be32 rsvd_led_mask;
3407 __be32 rsvd;
3408};
3409
3410#define OPA_LED_SHIFT 31
349ac71f 3411#define OPA_LED_MASK BIT(OPA_LED_SHIFT)
77241056
MM		 3412
3413static int __subn_get_opa_led_info(struct opa_smp *smp, u32 am, u8 *data,
3414 struct ib_device *ibdev, u8 port,
3415 u32 *resp_len)
3416{
3417 struct hfi1_devdata *dd = dd_from_ibdev(ibdev);
3418 struct opa_led_info *p = (struct opa_led_info *) data;
3419 u32 nport = OPA_AM_NPORT(am);
3420 u64 reg;
3421
801cfd6d 3422 if (nport != 1) {
77241056
MM		 3423 smp->status |= IB_SMP_INVALID_FIELD;
3424 return reply((struct ib_mad_hdr *)smp);
3425 }
3426
3427 reg = read_csr(dd, DCC_CFG_LED_CNTRL);
3428 if ((reg & DCC_CFG_LED_CNTRL_LED_CNTRL_SMASK) &&
3429 ((reg & DCC_CFG_LED_CNTRL_LED_SW_BLINK_RATE_SMASK) == 0xf))
3430 p->rsvd_led_mask = cpu_to_be32(OPA_LED_MASK);
3431
3432 if (resp_len)
3433 *resp_len += sizeof(struct opa_led_info);
3434
3435 return reply((struct ib_mad_hdr *)smp);
3436}
3437
3438static int __subn_set_opa_led_info(struct opa_smp *smp, u32 am, u8 *data,
3439 struct ib_device *ibdev, u8 port,
3440 u32 *resp_len)
3441{
3442 struct hfi1_devdata *dd = dd_from_ibdev(ibdev);
3443 struct opa_led_info *p = (struct opa_led_info *) data;
3444 u32 nport = OPA_AM_NPORT(am);
3445 int on = !!(be32_to_cpu(p->rsvd_led_mask) & OPA_LED_MASK);
3446
801cfd6d 3447 if (nport != 1) {
77241056
MM		 3448 smp->status |= IB_SMP_INVALID_FIELD;
3449 return reply((struct ib_mad_hdr *)smp);
3450 }
3451
91ab4ed3
EH		 3452 if (on)
3453 hfi1_set_led_override(dd->pport, 2000, 1500);
3454 else
3455 hfi1_set_led_override(dd->pport, 0, 0);
77241056
MM		 3456
3457 return __subn_get_opa_led_info(smp, am, data, ibdev, port, resp_len);
3458}
3459
3460static int subn_get_opa_sma(__be16 attr_id, struct opa_smp *smp, u32 am,
3461 u8 *data, struct ib_device *ibdev, u8 port,
3462 u32 *resp_len)
3463{
3464 int ret;
3465 struct hfi1_ibport *ibp = to_iport(ibdev, port);
3466
3467 switch (attr_id) {
3468 case IB_SMP_ATTR_NODE_DESC:
3469 ret = __subn_get_opa_nodedesc(smp, am, data, ibdev, port,
3470 resp_len);
3471 break;
3472 case IB_SMP_ATTR_NODE_INFO:
3473 ret = __subn_get_opa_nodeinfo(smp, am, data, ibdev, port,
3474 resp_len);
3475 break;
3476 case IB_SMP_ATTR_PORT_INFO:
3477 ret = __subn_get_opa_portinfo(smp, am, data, ibdev, port,
3478 resp_len);
3479 break;
3480 case IB_SMP_ATTR_PKEY_TABLE:
3481 ret = __subn_get_opa_pkeytable(smp, am, data, ibdev, port,
3482 resp_len);
3483 break;
3484 case OPA_ATTRIB_ID_SL_TO_SC_MAP:
3485 ret = __subn_get_opa_sl_to_sc(smp, am, data, ibdev, port,
3486 resp_len);
3487 break;
3488 case OPA_ATTRIB_ID_SC_TO_SL_MAP:
3489 ret = __subn_get_opa_sc_to_sl(smp, am, data, ibdev, port,
3490 resp_len);
3491 break;
3492 case OPA_ATTRIB_ID_SC_TO_VLT_MAP:
3493 ret = __subn_get_opa_sc_to_vlt(smp, am, data, ibdev, port,
3494 resp_len);
3495 break;
3496 case OPA_ATTRIB_ID_SC_TO_VLNT_MAP:
3497 ret = __subn_get_opa_sc_to_vlnt(smp, am, data, ibdev, port,
3498 resp_len);
3499 break;
3500 case OPA_ATTRIB_ID_PORT_STATE_INFO:
3501 ret = __subn_get_opa_psi(smp, am, data, ibdev, port,
3502 resp_len);
3503 break;
3504 case OPA_ATTRIB_ID_BUFFER_CONTROL_TABLE:
3505 ret = __subn_get_opa_bct(smp, am, data, ibdev, port,
3506 resp_len);
3507 break;
3508 case OPA_ATTRIB_ID_CABLE_INFO:
3509 ret = __subn_get_opa_cable_info(smp, am, data, ibdev, port,
3510 resp_len);
3511 break;
3512 case IB_SMP_ATTR_VL_ARB_TABLE:
3513 ret = __subn_get_opa_vl_arb(smp, am, data, ibdev, port,
3514 resp_len);
3515 break;
3516 case OPA_ATTRIB_ID_CONGESTION_INFO:
3517 ret = __subn_get_opa_cong_info(smp, am, data, ibdev, port,
3518 resp_len);
3519 break;
3520 case OPA_ATTRIB_ID_HFI_CONGESTION_SETTING:
3521 ret = __subn_get_opa_cong_setting(smp, am, data, ibdev,
3522 port, resp_len);
3523 break;
3524 case OPA_ATTRIB_ID_HFI_CONGESTION_LOG:
3525 ret = __subn_get_opa_hfi1_cong_log(smp, am, data, ibdev,
3526 port, resp_len);
3527 break;
3528 case OPA_ATTRIB_ID_CONGESTION_CONTROL_TABLE:
3529 ret = __subn_get_opa_cc_table(smp, am, data, ibdev, port,
3530 resp_len);
3531 break;
3532 case IB_SMP_ATTR_LED_INFO:
3533 ret = __subn_get_opa_led_info(smp, am, data, ibdev, port,
3534 resp_len);
3535 break;
3536 case IB_SMP_ATTR_SM_INFO:
4eb06882 3537 if (ibp->rvp.port_cap_flags & IB_PORT_SM_DISABLED)
77241056 3538 return IB_MAD_RESULT_SUCCESS | IB_MAD_RESULT_CONSUMED;
4eb06882 3539 if (ibp->rvp.port_cap_flags & IB_PORT_SM)
77241056
MM		 3540 return IB_MAD_RESULT_SUCCESS;
3541 /* FALLTHROUGH */
3542 default:
3543 smp->status |= IB_SMP_UNSUP_METH_ATTR;
3544 ret = reply((struct ib_mad_hdr *)smp);
3545 break;
3546 }
3547 return ret;
3548}
3549
3550static int subn_set_opa_sma(__be16 attr_id, struct opa_smp *smp, u32 am,
3551 u8 *data, struct ib_device *ibdev, u8 port,
3552 u32 *resp_len)
3553{
3554 int ret;
3555 struct hfi1_ibport *ibp = to_iport(ibdev, port);
3556
3557 switch (attr_id) {
3558 case IB_SMP_ATTR_PORT_INFO:
3559 ret = __subn_set_opa_portinfo(smp, am, data, ibdev, port,
3560 resp_len);
3561 break;
3562 case IB_SMP_ATTR_PKEY_TABLE:
3563 ret = __subn_set_opa_pkeytable(smp, am, data, ibdev, port,
3564 resp_len);
3565 break;
3566 case OPA_ATTRIB_ID_SL_TO_SC_MAP:
3567 ret = __subn_set_opa_sl_to_sc(smp, am, data, ibdev, port,
3568 resp_len);
3569 break;
3570 case OPA_ATTRIB_ID_SC_TO_SL_MAP:
3571 ret = __subn_set_opa_sc_to_sl(smp, am, data, ibdev, port,
3572 resp_len);
3573 break;
3574 case OPA_ATTRIB_ID_SC_TO_VLT_MAP:
3575 ret = __subn_set_opa_sc_to_vlt(smp, am, data, ibdev, port,
3576 resp_len);
3577 break;
3578 case OPA_ATTRIB_ID_SC_TO_VLNT_MAP:
3579 ret = __subn_set_opa_sc_to_vlnt(smp, am, data, ibdev, port,
3580 resp_len);
3581 break;
3582 case OPA_ATTRIB_ID_PORT_STATE_INFO:
3583 ret = __subn_set_opa_psi(smp, am, data, ibdev, port,
3584 resp_len);
3585 break;
3586 case OPA_ATTRIB_ID_BUFFER_CONTROL_TABLE:
3587 ret = __subn_set_opa_bct(smp, am, data, ibdev, port,
3588 resp_len);
3589 break;
3590 case IB_SMP_ATTR_VL_ARB_TABLE:
3591 ret = __subn_set_opa_vl_arb(smp, am, data, ibdev, port,
3592 resp_len);
3593 break;
3594 case OPA_ATTRIB_ID_HFI_CONGESTION_SETTING:
3595 ret = __subn_set_opa_cong_setting(smp, am, data, ibdev,
3596 port, resp_len);
3597 break;
3598 case OPA_ATTRIB_ID_CONGESTION_CONTROL_TABLE:
3599 ret = __subn_set_opa_cc_table(smp, am, data, ibdev, port,
3600 resp_len);
3601 break;
3602 case IB_SMP_ATTR_LED_INFO:
3603 ret = __subn_set_opa_led_info(smp, am, data, ibdev, port,
3604 resp_len);
3605 break;
3606 case IB_SMP_ATTR_SM_INFO:
4eb06882 3607 if (ibp->rvp.port_cap_flags & IB_PORT_SM_DISABLED)
77241056 3608 return IB_MAD_RESULT_SUCCESS | IB_MAD_RESULT_CONSUMED;
4eb06882 3609 if (ibp->rvp.port_cap_flags & IB_PORT_SM)
77241056
MM		 3610 return IB_MAD_RESULT_SUCCESS;
3611 /* FALLTHROUGH */
3612 default:
3613 smp->status |= IB_SMP_UNSUP_METH_ATTR;
3614 ret = reply((struct ib_mad_hdr *)smp);
3615 break;
3616 }
3617 return ret;
3618}
3619
3620static inline void set_aggr_error(struct opa_aggregate *ag)
3621{
3622 ag->err_reqlength |= cpu_to_be16(0x8000);
3623}
3624
3625static int subn_get_opa_aggregate(struct opa_smp *smp,
3626 struct ib_device *ibdev, u8 port,
3627 u32 *resp_len)
3628{
3629 int i;
3630 u32 num_attr = be32_to_cpu(smp->attr_mod) & 0x000000ff;
3631 u8 *next_smp = opa_get_smp_data(smp);
3632
3633 if (num_attr < 1 || num_attr > 117) {
3634 smp->status |= IB_SMP_INVALID_FIELD;
3635 return reply((struct ib_mad_hdr *)smp);
3636 }
3637
3638 for (i = 0; i < num_attr; i++) {
3639 struct opa_aggregate *agg;
3640 size_t agg_data_len;
3641 size_t agg_size;
3642 u32 am;
3643
3644 agg = (struct opa_aggregate *)next_smp;
3645 agg_data_len = (be16_to_cpu(agg->err_reqlength) & 0x007f) * 8;
3646 agg_size = sizeof(*agg) + agg_data_len;
3647 am = be32_to_cpu(agg->attr_mod);
3648
3649 *resp_len += agg_size;
3650
3651 if (next_smp + agg_size > ((u8 *)smp) + sizeof(*smp)) {
3652 smp->status |= IB_SMP_INVALID_FIELD;
3653 return reply((struct ib_mad_hdr *)smp);
3654 }
3655
3656 /* zero the payload for this segment */
3657 memset(next_smp + sizeof(*agg), 0, agg_data_len);
3658
3659 (void) subn_get_opa_sma(agg->attr_id, smp, am, agg->data,
3660 ibdev, port, NULL);
3661 if (smp->status & ~IB_SMP_DIRECTION) {
3662 set_aggr_error(agg);
3663 return reply((struct ib_mad_hdr *)smp);
3664 }
3665 next_smp += agg_size;
3666
3667 }
3668
3669 return reply((struct ib_mad_hdr *)smp);
3670}
3671
3672static int subn_set_opa_aggregate(struct opa_smp *smp,
3673 struct ib_device *ibdev, u8 port,
3674 u32 *resp_len)
3675{
3676 int i;
3677 u32 num_attr = be32_to_cpu(smp->attr_mod) & 0x000000ff;
3678 u8 *next_smp = opa_get_smp_data(smp);
3679
3680 if (num_attr < 1 || num_attr > 117) {
3681 smp->status |= IB_SMP_INVALID_FIELD;
3682 return reply((struct ib_mad_hdr *)smp);
3683 }
3684
3685 for (i = 0; i < num_attr; i++) {
3686 struct opa_aggregate *agg;
3687 size_t agg_data_len;
3688 size_t agg_size;
3689 u32 am;
3690
3691 agg = (struct opa_aggregate *)next_smp;
3692 agg_data_len = (be16_to_cpu(agg->err_reqlength) & 0x007f) * 8;
3693 agg_size = sizeof(*agg) + agg_data_len;
3694 am = be32_to_cpu(agg->attr_mod);
3695
3696 *resp_len += agg_size;
3697
3698 if (next_smp + agg_size > ((u8 *)smp) + sizeof(*smp)) {
3699 smp->status |= IB_SMP_INVALID_FIELD;
3700 return reply((struct ib_mad_hdr *)smp);
3701 }
3702
3703 (void) subn_set_opa_sma(agg->attr_id, smp, am, agg->data,
3704 ibdev, port, NULL);
3705 if (smp->status & ~IB_SMP_DIRECTION) {
3706 set_aggr_error(agg);
3707 return reply((struct ib_mad_hdr *)smp);
3708 }
3709 next_smp += agg_size;
3710
3711 }
3712
3713 return reply((struct ib_mad_hdr *)smp);
3714}
3715
3716/*
3717 * OPAv1 specifies that, on the transition to link up, these counters
3718 * are cleared:
3719 * PortRcvErrors [*]
3720 * LinkErrorRecovery
3721 * LocalLinkIntegrityErrors
3722 * ExcessiveBufferOverruns [*]
3723 *
3724 * [*] Error info associated with these counters is retained, but the
3725 * error info status is reset to 0.
3726 */
3727void clear_linkup_counters(struct hfi1_devdata *dd)
3728{
3729 /* PortRcvErrors */
3730 write_dev_cntr(dd, C_DC_RCV_ERR, CNTR_INVALID_VL, 0);
3731 dd->err_info_rcvport.status_and_code &= ~OPA_EI_STATUS_SMASK;
3732 /* LinkErrorRecovery */
3733 write_dev_cntr(dd, C_DC_SEQ_CRC_CNT, CNTR_INVALID_VL, 0);
3734 write_dev_cntr(dd, C_DC_REINIT_FROM_PEER_CNT, CNTR_INVALID_VL, 0);
3735 /* LocalLinkIntegrityErrors */
3736 write_dev_cntr(dd, C_DC_TX_REPLAY, CNTR_INVALID_VL, 0);
3737 write_dev_cntr(dd, C_DC_RX_REPLAY, CNTR_INVALID_VL, 0);
3738 /* ExcessiveBufferOverruns */
3739 write_dev_cntr(dd, C_RCV_OVF, CNTR_INVALID_VL, 0);
3740 dd->rcv_ovfl_cnt = 0;
3741 dd->err_info_xmit_constraint.status &= ~OPA_EI_STATUS_SMASK;
3742}
3743
3744/*
3745 * is_local_mad() returns 1 if 'mad' is sent from, and destined to the
3746 * local node, 0 otherwise.
3747 */
3748static int is_local_mad(struct hfi1_ibport *ibp, const struct opa_mad *mad,
3749 const struct ib_wc *in_wc)
3750{
3751 struct hfi1_pportdata *ppd = ppd_from_ibp(ibp);
3752 const struct opa_smp *smp = (const struct opa_smp *)mad;
3753
3754 if (smp->mgmt_class == IB_MGMT_CLASS_SUBN_DIRECTED_ROUTE) {
3755 return (smp->hop_cnt == 0 &&
3756 smp->route.dr.dr_slid == OPA_LID_PERMISSIVE &&
3757 smp->route.dr.dr_dlid == OPA_LID_PERMISSIVE);
3758 }
3759
3760 return (in_wc->slid == ppd->lid);
3761}
3762
3763/*
3764 * opa_local_smp_check() should only be called on MADs for which
3765 * is_local_mad() returns true. It applies the SMP checks that are
3766 * specific to SMPs which are sent from, and destined to this node.
3767 * opa_local_smp_check() returns 0 if the SMP passes its checks, 1
3768 * otherwise.
3769 *
3770 * SMPs which arrive from other nodes are instead checked by
3771 * opa_smp_check().
3772 */
3773static int opa_local_smp_check(struct hfi1_ibport *ibp,
3774 const struct ib_wc *in_wc)
3775{
3776 struct hfi1_pportdata *ppd = ppd_from_ibp(ibp);
3777 u16 slid = in_wc->slid;
3778 u16 pkey;
3779
3780 if (in_wc->pkey_index >= ARRAY_SIZE(ppd->pkeys))
3781 return 1;
3782
3783 pkey = ppd->pkeys[in_wc->pkey_index];
3784 /*
3785 * We need to do the "node-local" checks specified in OPAv1,
3786 * rev 0.90, section 9.10.26, which are:
3787 * - pkey is 0x7fff, or 0xffff
3788 * - Source QPN == 0 || Destination QPN == 0
3789 * - the MAD header's management class is either
3790 * IB_MGMT_CLASS_SUBN_DIRECTED_ROUTE or
3791 * IB_MGMT_CLASS_SUBN_LID_ROUTED
3792 * - SLID != 0
3793 *
3794 * However, we know (and so don't need to check again) that,
3795 * for local SMPs, the MAD stack passes MADs with:
3796 * - Source QPN of 0
3797 * - MAD mgmt_class is IB_MGMT_CLASS_SUBN_DIRECTED_ROUTE
3798 * - SLID is either: OPA_LID_PERMISSIVE (0xFFFFFFFF), or
3799 * our own port's lid
3800 *
3801 */
3802 if (pkey == LIM_MGMT_P_KEY || pkey == FULL_MGMT_P_KEY)
3803 return 0;
3804 ingress_pkey_table_fail(ppd, pkey, slid);
3805 return 1;
3806}
3807
3808static int process_subn_opa(struct ib_device *ibdev, int mad_flags,
3809 u8 port, const struct opa_mad *in_mad,
3810 struct opa_mad *out_mad,
3811 u32 *resp_len)
3812{
3813 struct opa_smp *smp = (struct opa_smp *)out_mad;
3814 struct hfi1_ibport *ibp = to_iport(ibdev, port);
3815 u8 *data;
3816 u32 am;
3817 __be16 attr_id;
3818 int ret;
3819
3820 *out_mad = *in_mad;
3821 data = opa_get_smp_data(smp);
3822
3823 am = be32_to_cpu(smp->attr_mod);
3824 attr_id = smp->attr_id;
3825 if (smp->class_version != OPA_SMI_CLASS_VERSION) {
3826 smp->status |= IB_SMP_UNSUP_VERSION;
3827 ret = reply((struct ib_mad_hdr *)smp);
3828 goto bail;
3829 }
3830 ret = check_mkey(ibp, (struct ib_mad_hdr *)smp, mad_flags, smp->mkey,
3831 smp->route.dr.dr_slid, smp->route.dr.return_path,
3832 smp->hop_cnt);
3833 if (ret) {
3834 u32 port_num = be32_to_cpu(smp->attr_mod);
3835
3836 /*
3837 * If this is a get/set portinfo, we already check the
3838 * M_Key if the MAD is for another port and the M_Key
3839 * is OK on the receiving port. This check is needed
3840 * to increment the error counters when the M_Key
3841 * fails to match on *both* ports.
3842 */
3843 if (attr_id == IB_SMP_ATTR_PORT_INFO &&
3844 (smp->method == IB_MGMT_METHOD_GET ||
3845 smp->method == IB_MGMT_METHOD_SET) &&
3846 port_num && port_num <= ibdev->phys_port_cnt &&
3847 port != port_num)
3848 (void) check_mkey(to_iport(ibdev, port_num),
3849 (struct ib_mad_hdr *)smp, 0,
3850 smp->mkey, smp->route.dr.dr_slid,
3851 smp->route.dr.return_path,
3852 smp->hop_cnt);
3853 ret = IB_MAD_RESULT_FAILURE;
3854 goto bail;
3855 }
3856
3857 *resp_len = opa_get_smp_header_size(smp);
3858
3859 switch (smp->method) {
3860 case IB_MGMT_METHOD_GET:
3861 switch (attr_id) {
3862 default:
3863 clear_opa_smp_data(smp);
3864 ret = subn_get_opa_sma(attr_id, smp, am, data,
3865 ibdev, port, resp_len);
3866 goto bail;
3867 case OPA_ATTRIB_ID_AGGREGATE:
3868 ret = subn_get_opa_aggregate(smp, ibdev, port,
3869 resp_len);
3870 goto bail;
3871 }
3872 case IB_MGMT_METHOD_SET:
3873 switch (attr_id) {
3874 default:
3875 ret = subn_set_opa_sma(attr_id, smp, am, data,
3876 ibdev, port, resp_len);
3877 goto bail;
3878 case OPA_ATTRIB_ID_AGGREGATE:
3879 ret = subn_set_opa_aggregate(smp, ibdev, port,
3880 resp_len);
3881 goto bail;
3882 }
3883 case IB_MGMT_METHOD_TRAP:
3884 case IB_MGMT_METHOD_REPORT:
3885 case IB_MGMT_METHOD_REPORT_RESP:
3886 case IB_MGMT_METHOD_GET_RESP:
3887 /*
3888 * The ib_mad module will call us to process responses
3889 * before checking for other consumers.
3890 * Just tell the caller to process it normally.
3891 */
3892 ret = IB_MAD_RESULT_SUCCESS;
3893 goto bail;
3894 default:
3895 smp->status |= IB_SMP_UNSUP_METHOD;
3896 ret = reply((struct ib_mad_hdr *)smp);
3897 }
3898
3899bail:
3900 return ret;
3901}
3902
3903static int process_subn(struct ib_device *ibdev, int mad_flags,
3904 u8 port, const struct ib_mad *in_mad,
3905 struct ib_mad *out_mad)
3906{
3907 struct ib_smp *smp = (struct ib_smp *)out_mad;
3908 struct hfi1_ibport *ibp = to_iport(ibdev, port);
3909 int ret;
3910
3911 *out_mad = *in_mad;
3912 if (smp->class_version != 1) {
3913 smp->status |= IB_SMP_UNSUP_VERSION;
3914 ret = reply((struct ib_mad_hdr *)smp);
3915 goto bail;
3916 }
3917
3918 ret = check_mkey(ibp, (struct ib_mad_hdr *)smp, mad_flags,
3919 smp->mkey, (__force __be32)smp->dr_slid,
3920 smp->return_path, smp->hop_cnt);
3921 if (ret) {
3922 u32 port_num = be32_to_cpu(smp->attr_mod);
3923
3924 /*
3925 * If this is a get/set portinfo, we already check the
3926 * M_Key if the MAD is for another port and the M_Key
3927 * is OK on the receiving port. This check is needed
3928 * to increment the error counters when the M_Key
3929 * fails to match on *both* ports.
3930 */
3931 if (in_mad->mad_hdr.attr_id == IB_SMP_ATTR_PORT_INFO &&
3932 (smp->method == IB_MGMT_METHOD_GET ||
3933 smp->method == IB_MGMT_METHOD_SET) &&
3934 port_num && port_num <= ibdev->phys_port_cnt &&
3935 port != port_num)
3936 (void) check_mkey(to_iport(ibdev, port_num),
3937 (struct ib_mad_hdr *)smp, 0,
3938 smp->mkey,
3939 (__force __be32)smp->dr_slid,
3940 smp->return_path, smp->hop_cnt);
3941 ret = IB_MAD_RESULT_FAILURE;
3942 goto bail;
3943 }
3944
3945 switch (smp->method) {
3946 case IB_MGMT_METHOD_GET:
3947 switch (smp->attr_id) {
3948 case IB_SMP_ATTR_NODE_INFO:
3949 ret = subn_get_nodeinfo(smp, ibdev, port);
3950 goto bail;
3951 default:
3952 smp->status |= IB_SMP_UNSUP_METH_ATTR;
3953 ret = reply((struct ib_mad_hdr *)smp);
3954 goto bail;
3955 }
3956 }
3957
3958bail:
3959 return ret;
3960}
3961
3962static int process_perf_opa(struct ib_device *ibdev, u8 port,
3963 const struct opa_mad *in_mad,
3964 struct opa_mad *out_mad, u32 *resp_len)
3965{
3966 struct opa_pma_mad *pmp = (struct opa_pma_mad *)out_mad;
3967 int ret;
3968
3969 *out_mad = *in_mad;
3970
3971 if (pmp->mad_hdr.class_version != OPA_SMI_CLASS_VERSION) {
3972 pmp->mad_hdr.status |= IB_SMP_UNSUP_VERSION;
3973 return reply((struct ib_mad_hdr *)pmp);
3974 }
3975
3976 *resp_len = sizeof(pmp->mad_hdr);
3977
3978 switch (pmp->mad_hdr.method) {
3979 case IB_MGMT_METHOD_GET:
3980 switch (pmp->mad_hdr.attr_id) {
3981 case IB_PMA_CLASS_PORT_INFO:
3982 ret = pma_get_opa_classportinfo(pmp, ibdev, resp_len);
3983 goto bail;
3984 case OPA_PM_ATTRIB_ID_PORT_STATUS:
3985 ret = pma_get_opa_portstatus(pmp, ibdev, port,
3986 resp_len);
3987 goto bail;
3988 case OPA_PM_ATTRIB_ID_DATA_PORT_COUNTERS:
3989 ret = pma_get_opa_datacounters(pmp, ibdev, port,
3990 resp_len);
3991 goto bail;
3992 case OPA_PM_ATTRIB_ID_ERROR_PORT_COUNTERS:
3993 ret = pma_get_opa_porterrors(pmp, ibdev, port,
3994 resp_len);
3995 goto bail;
3996 case OPA_PM_ATTRIB_ID_ERROR_INFO:
3997 ret = pma_get_opa_errorinfo(pmp, ibdev, port,
3998 resp_len);
3999 goto bail;
4000 default:
4001 pmp->mad_hdr.status |= IB_SMP_UNSUP_METH_ATTR;
4002 ret = reply((struct ib_mad_hdr *)pmp);
4003 goto bail;
4004 }
4005
4006 case IB_MGMT_METHOD_SET:
4007 switch (pmp->mad_hdr.attr_id) {
4008 case OPA_PM_ATTRIB_ID_CLEAR_PORT_STATUS:
4009 ret = pma_set_opa_portstatus(pmp, ibdev, port,
4010 resp_len);
4011 goto bail;
4012 case OPA_PM_ATTRIB_ID_ERROR_INFO:
4013 ret = pma_set_opa_errorinfo(pmp, ibdev, port,
4014 resp_len);
4015 goto bail;
4016 default:
4017 pmp->mad_hdr.status |= IB_SMP_UNSUP_METH_ATTR;
4018 ret = reply((struct ib_mad_hdr *)pmp);
4019 goto bail;
4020 }
4021
4022 case IB_MGMT_METHOD_TRAP:
4023 case IB_MGMT_METHOD_GET_RESP:
4024 /*
4025 * The ib_mad module will call us to process responses
4026 * before checking for other consumers.
4027 * Just tell the caller to process it normally.
4028 */
4029 ret = IB_MAD_RESULT_SUCCESS;
4030 goto bail;
4031
4032 default:
4033 pmp->mad_hdr.status |= IB_SMP_UNSUP_METHOD;
4034 ret = reply((struct ib_mad_hdr *)pmp);
4035 }
4036
4037bail:
4038 return ret;
4039}
4040
4041static int hfi1_process_opa_mad(struct ib_device *ibdev, int mad_flags,
a724648e
JB		 4042 u8 port, const struct ib_wc *in_wc,
4043 const struct ib_grh *in_grh,
4044 const struct opa_mad *in_mad,
4045 struct opa_mad *out_mad, size_t *out_mad_size,
4046 u16 *out_mad_pkey_index)
77241056
MM		 4047{
4048 int ret;
4049 int pkey_idx;
4050 u32 resp_len = 0;
4051 struct hfi1_ibport *ibp = to_iport(ibdev, port);
4052
4053 pkey_idx = hfi1_lookup_pkey_idx(ibp, LIM_MGMT_P_KEY);
4054 if (pkey_idx < 0) {
4055 pr_warn("failed to find limited mgmt pkey, defaulting 0x%x\n",
4056 hfi1_get_pkey(ibp, 1));
4057 pkey_idx = 1;
4058 }
4059 *out_mad_pkey_index = (u16)pkey_idx;
4060
4061 switch (in_mad->mad_hdr.mgmt_class) {
4062 case IB_MGMT_CLASS_SUBN_DIRECTED_ROUTE:
4063 case IB_MGMT_CLASS_SUBN_LID_ROUTED:
4064 if (is_local_mad(ibp, in_mad, in_wc)) {
4065 ret = opa_local_smp_check(ibp, in_wc);
4066 if (ret)
4067 return IB_MAD_RESULT_FAILURE;
4068 }
4069 ret = process_subn_opa(ibdev, mad_flags, port, in_mad,
4070 out_mad, &resp_len);
4071 goto bail;
4072 case IB_MGMT_CLASS_PERF_MGMT:
4073 ret = process_perf_opa(ibdev, port, in_mad, out_mad,
4074 &resp_len);
4075 goto bail;
4076
4077 default:
4078 ret = IB_MAD_RESULT_SUCCESS;
4079 }
4080
4081bail:
4082 if (ret & IB_MAD_RESULT_REPLY)
4083 *out_mad_size = round_up(resp_len, 8);
4084 else if (ret & IB_MAD_RESULT_SUCCESS)
4085 *out_mad_size = in_wc->byte_len - sizeof(struct ib_grh);
4086
4087 return ret;
4088}
4089
4090static int hfi1_process_ib_mad(struct ib_device *ibdev, int mad_flags, u8 port,
4091 const struct ib_wc *in_wc,
4092 const struct ib_grh *in_grh,
4093 const struct ib_mad *in_mad,
4094 struct ib_mad *out_mad)
4095{
4096 int ret;
4097
4098 switch (in_mad->mad_hdr.mgmt_class) {
4099 case IB_MGMT_CLASS_SUBN_DIRECTED_ROUTE:
4100 case IB_MGMT_CLASS_SUBN_LID_ROUTED:
4101 ret = process_subn(ibdev, mad_flags, port, in_mad, out_mad);
4102 goto bail;
4103 default:
4104 ret = IB_MAD_RESULT_SUCCESS;
4105 }
4106
4107bail:
4108 return ret;
4109}
4110
4111/**
4112 * hfi1_process_mad - process an incoming MAD packet
4113 * @ibdev: the infiniband device this packet came in on
4114 * @mad_flags: MAD flags
4115 * @port: the port number this packet came in on
4116 * @in_wc: the work completion entry for this packet
4117 * @in_grh: the global route header for this packet
4118 * @in_mad: the incoming MAD
4119 * @out_mad: any outgoing MAD reply
4120 *
4121 * Returns IB_MAD_RESULT_SUCCESS if this is a MAD that we are not
4122 * interested in processing.
4123 *
4124 * Note that the verbs framework has already done the MAD sanity checks,
4125 * and hop count/pointer updating for IB_MGMT_CLASS_SUBN_DIRECTED_ROUTE
4126 * MADs.
4127 *
4128 * This is called by the ib_mad module.
4129 */
4130int hfi1_process_mad(struct ib_device *ibdev, int mad_flags, u8 port,
4131 const struct ib_wc *in_wc, const struct ib_grh *in_grh,
4132 const struct ib_mad_hdr *in_mad, size_t in_mad_size,
4133 struct ib_mad_hdr *out_mad, size_t *out_mad_size,
4134 u16 *out_mad_pkey_index)
4135{
4136 switch (in_mad->base_version) {
4137 case OPA_MGMT_BASE_VERSION:
4138 if (unlikely(in_mad_size != sizeof(struct opa_mad))) {
4139 dev_err(ibdev->dma_device, "invalid in_mad_size\n");
4140 return IB_MAD_RESULT_FAILURE;
4141 }
4142 return hfi1_process_opa_mad(ibdev, mad_flags, port,
4143 in_wc, in_grh,
4144 (struct opa_mad *)in_mad,
4145 (struct opa_mad *)out_mad,
4146 out_mad_size,
4147 out_mad_pkey_index);
4148 case IB_MGMT_BASE_VERSION:
4149 return hfi1_process_ib_mad(ibdev, mad_flags, port,
4150 in_wc, in_grh,
4151 (const struct ib_mad *)in_mad,
4152 (struct ib_mad *)out_mad);
4153 default:
4154 break;
4155 }
4156
4157 return IB_MAD_RESULT_FAILURE;
4158}
Linux 6.x block layer and io_uring tree(s)