ALSA: hda - Add the top speaker pin config for HP Spectre x360
[linux/fpc-iii.git] / drivers / infiniband / hw / hfi1 / mad.c
blob1263abe01999e3e84306b5594921bc4b11019fca
1 /*
2 * Copyright(c) 2015, 2016 Intel Corporation.
4 * This file is provided under a dual BSD/GPLv2 license. When using or
5 * redistributing this file, you may do so under either license.
7 * GPL LICENSE SUMMARY
9 * This program is free software; you can redistribute it and/or modify
10 * it under the terms of version 2 of the GNU General Public License as
11 * published by the Free Software Foundation.
13 * This program is distributed in the hope that it will be useful, but
14 * WITHOUT ANY WARRANTY; without even the implied warranty of
15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
16 * General Public License for more details.
18 * BSD LICENSE
20 * Redistribution and use in source and binary forms, with or without
21 * modification, are permitted provided that the following conditions
22 * are met:
24 * - Redistributions of source code must retain the above copyright
25 * notice, this list of conditions and the following disclaimer.
26 * - Redistributions in binary form must reproduce the above copyright
27 * notice, this list of conditions and the following disclaimer in
28 * the documentation and/or other materials provided with the
29 * distribution.
30 * - Neither the name of Intel Corporation nor the names of its
31 * contributors may be used to endorse or promote products derived
32 * from this software without specific prior written permission.
34 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
35 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
36 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
37 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
38 * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
39 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
40 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
41 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
42 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
43 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
44 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
48 #include <linux/net.h>
49 #define OPA_NUM_PKEY_BLOCKS_PER_SMP (OPA_SMP_DR_DATA_SIZE \
50 / (OPA_PARTITION_TABLE_BLK_SIZE * sizeof(u16)))
52 #include "hfi.h"
53 #include "mad.h"
54 #include "trace.h"
55 #include "qp.h"
57 /* the reset value from the FM is supposed to be 0xffff, handle both */
58 #define OPA_LINK_WIDTH_RESET_OLD 0x0fff
59 #define OPA_LINK_WIDTH_RESET 0xffff
61 static int reply(struct ib_mad_hdr *smp)
64 * The verbs framework will handle the directed/LID route
65 * packet changes.
67 smp->method = IB_MGMT_METHOD_GET_RESP;
68 if (smp->mgmt_class == IB_MGMT_CLASS_SUBN_DIRECTED_ROUTE)
69 smp->status |= IB_SMP_DIRECTION;
70 return IB_MAD_RESULT_SUCCESS | IB_MAD_RESULT_REPLY;
73 static inline void clear_opa_smp_data(struct opa_smp *smp)
75 void *data = opa_get_smp_data(smp);
76 size_t size = opa_get_smp_data_size(smp);
78 memset(data, 0, size);
81 void hfi1_event_pkey_change(struct hfi1_devdata *dd, u8 port)
83 struct ib_event event;
85 event.event = IB_EVENT_PKEY_CHANGE;
86 event.device = &dd->verbs_dev.rdi.ibdev;
87 event.element.port_num = port;
88 ib_dispatch_event(&event);
91 static void send_trap(struct hfi1_ibport *ibp, void *data, unsigned len)
93 struct ib_mad_send_buf *send_buf;
94 struct ib_mad_agent *agent;
95 struct opa_smp *smp;
96 int ret;
97 unsigned long flags;
98 unsigned long timeout;
99 int pkey_idx;
100 u32 qpn = ppd_from_ibp(ibp)->sm_trap_qp;
102 agent = ibp->rvp.send_agent;
103 if (!agent)
104 return;
106 /* o14-3.2.1 */
107 if (ppd_from_ibp(ibp)->lstate != IB_PORT_ACTIVE)
108 return;
110 /* o14-2 */
111 if (ibp->rvp.trap_timeout && time_before(jiffies,
112 ibp->rvp.trap_timeout))
113 return;
115 pkey_idx = hfi1_lookup_pkey_idx(ibp, LIM_MGMT_P_KEY);
116 if (pkey_idx < 0) {
117 pr_warn("%s: failed to find limited mgmt pkey, defaulting 0x%x\n",
118 __func__, hfi1_get_pkey(ibp, 1));
119 pkey_idx = 1;
122 send_buf = ib_create_send_mad(agent, qpn, pkey_idx, 0,
123 IB_MGMT_MAD_HDR, IB_MGMT_MAD_DATA,
124 GFP_ATOMIC, IB_MGMT_BASE_VERSION);
125 if (IS_ERR(send_buf))
126 return;
128 smp = send_buf->mad;
129 smp->base_version = OPA_MGMT_BASE_VERSION;
130 smp->mgmt_class = IB_MGMT_CLASS_SUBN_LID_ROUTED;
131 smp->class_version = OPA_SMI_CLASS_VERSION;
132 smp->method = IB_MGMT_METHOD_TRAP;
133 ibp->rvp.tid++;
134 smp->tid = cpu_to_be64(ibp->rvp.tid);
135 smp->attr_id = IB_SMP_ATTR_NOTICE;
136 /* o14-1: smp->mkey = 0; */
137 memcpy(smp->route.lid.data, data, len);
139 spin_lock_irqsave(&ibp->rvp.lock, flags);
140 if (!ibp->rvp.sm_ah) {
141 if (ibp->rvp.sm_lid != be16_to_cpu(IB_LID_PERMISSIVE)) {
142 struct ib_ah *ah;
144 ah = hfi1_create_qp0_ah(ibp, ibp->rvp.sm_lid);
145 if (IS_ERR(ah)) {
146 ret = PTR_ERR(ah);
147 } else {
148 send_buf->ah = ah;
149 ibp->rvp.sm_ah = ibah_to_rvtah(ah);
150 ret = 0;
152 } else {
153 ret = -EINVAL;
155 } else {
156 send_buf->ah = &ibp->rvp.sm_ah->ibah;
157 ret = 0;
159 spin_unlock_irqrestore(&ibp->rvp.lock, flags);
161 if (!ret)
162 ret = ib_post_send_mad(send_buf, NULL);
163 if (!ret) {
164 /* 4.096 usec. */
165 timeout = (4096 * (1UL << ibp->rvp.subnet_timeout)) / 1000;
166 ibp->rvp.trap_timeout = jiffies + usecs_to_jiffies(timeout);
167 } else {
168 ib_free_send_mad(send_buf);
169 ibp->rvp.trap_timeout = 0;
174 * Send a bad [PQ]_Key trap (ch. 14.3.8).
176 void hfi1_bad_pqkey(struct hfi1_ibport *ibp, __be16 trap_num, u32 key, u32 sl,
177 u32 qp1, u32 qp2, u16 lid1, u16 lid2)
179 struct opa_mad_notice_attr data;
180 u32 lid = ppd_from_ibp(ibp)->lid;
181 u32 _lid1 = lid1;
182 u32 _lid2 = lid2;
184 memset(&data, 0, sizeof(data));
186 if (trap_num == OPA_TRAP_BAD_P_KEY)
187 ibp->rvp.pkey_violations++;
188 else
189 ibp->rvp.qkey_violations++;
190 ibp->rvp.n_pkt_drops++;
192 /* Send violation trap */
193 data.generic_type = IB_NOTICE_TYPE_SECURITY;
194 data.prod_type_lsb = IB_NOTICE_PROD_CA;
195 data.trap_num = trap_num;
196 data.issuer_lid = cpu_to_be32(lid);
197 data.ntc_257_258.lid1 = cpu_to_be32(_lid1);
198 data.ntc_257_258.lid2 = cpu_to_be32(_lid2);
199 data.ntc_257_258.key = cpu_to_be32(key);
200 data.ntc_257_258.sl = sl << 3;
201 data.ntc_257_258.qp1 = cpu_to_be32(qp1);
202 data.ntc_257_258.qp2 = cpu_to_be32(qp2);
204 send_trap(ibp, &data, sizeof(data));
208 * Send a bad M_Key trap (ch. 14.3.9).
210 static void bad_mkey(struct hfi1_ibport *ibp, struct ib_mad_hdr *mad,
211 __be64 mkey, __be32 dr_slid, u8 return_path[], u8 hop_cnt)
213 struct opa_mad_notice_attr data;
214 u32 lid = ppd_from_ibp(ibp)->lid;
216 memset(&data, 0, sizeof(data));
217 /* Send violation trap */
218 data.generic_type = IB_NOTICE_TYPE_SECURITY;
219 data.prod_type_lsb = IB_NOTICE_PROD_CA;
220 data.trap_num = OPA_TRAP_BAD_M_KEY;
221 data.issuer_lid = cpu_to_be32(lid);
222 data.ntc_256.lid = data.issuer_lid;
223 data.ntc_256.method = mad->method;
224 data.ntc_256.attr_id = mad->attr_id;
225 data.ntc_256.attr_mod = mad->attr_mod;
226 data.ntc_256.mkey = mkey;
227 if (mad->mgmt_class == IB_MGMT_CLASS_SUBN_DIRECTED_ROUTE) {
228 data.ntc_256.dr_slid = dr_slid;
229 data.ntc_256.dr_trunc_hop = IB_NOTICE_TRAP_DR_NOTICE;
230 if (hop_cnt > ARRAY_SIZE(data.ntc_256.dr_rtn_path)) {
231 data.ntc_256.dr_trunc_hop |=
232 IB_NOTICE_TRAP_DR_TRUNC;
233 hop_cnt = ARRAY_SIZE(data.ntc_256.dr_rtn_path);
235 data.ntc_256.dr_trunc_hop |= hop_cnt;
236 memcpy(data.ntc_256.dr_rtn_path, return_path,
237 hop_cnt);
240 send_trap(ibp, &data, sizeof(data));
244 * Send a Port Capability Mask Changed trap (ch. 14.3.11).
246 void hfi1_cap_mask_chg(struct rvt_dev_info *rdi, u8 port_num)
248 struct opa_mad_notice_attr data;
249 struct hfi1_ibdev *verbs_dev = dev_from_rdi(rdi);
250 struct hfi1_devdata *dd = dd_from_dev(verbs_dev);
251 struct hfi1_ibport *ibp = &dd->pport[port_num - 1].ibport_data;
252 u32 lid = ppd_from_ibp(ibp)->lid;
254 memset(&data, 0, sizeof(data));
256 data.generic_type = IB_NOTICE_TYPE_INFO;
257 data.prod_type_lsb = IB_NOTICE_PROD_CA;
258 data.trap_num = OPA_TRAP_CHANGE_CAPABILITY;
259 data.issuer_lid = cpu_to_be32(lid);
260 data.ntc_144.lid = data.issuer_lid;
261 data.ntc_144.new_cap_mask = cpu_to_be32(ibp->rvp.port_cap_flags);
263 send_trap(ibp, &data, sizeof(data));
267 * Send a System Image GUID Changed trap (ch. 14.3.12).
269 void hfi1_sys_guid_chg(struct hfi1_ibport *ibp)
271 struct opa_mad_notice_attr data;
272 u32 lid = ppd_from_ibp(ibp)->lid;
274 memset(&data, 0, sizeof(data));
276 data.generic_type = IB_NOTICE_TYPE_INFO;
277 data.prod_type_lsb = IB_NOTICE_PROD_CA;
278 data.trap_num = OPA_TRAP_CHANGE_SYSGUID;
279 data.issuer_lid = cpu_to_be32(lid);
280 data.ntc_145.new_sys_guid = ib_hfi1_sys_image_guid;
281 data.ntc_145.lid = data.issuer_lid;
283 send_trap(ibp, &data, sizeof(data));
287 * Send a Node Description Changed trap (ch. 14.3.13).
289 void hfi1_node_desc_chg(struct hfi1_ibport *ibp)
291 struct opa_mad_notice_attr data;
292 u32 lid = ppd_from_ibp(ibp)->lid;
294 memset(&data, 0, sizeof(data));
296 data.generic_type = IB_NOTICE_TYPE_INFO;
297 data.prod_type_lsb = IB_NOTICE_PROD_CA;
298 data.trap_num = OPA_TRAP_CHANGE_CAPABILITY;
299 data.issuer_lid = cpu_to_be32(lid);
300 data.ntc_144.lid = data.issuer_lid;
301 data.ntc_144.change_flags =
302 cpu_to_be16(OPA_NOTICE_TRAP_NODE_DESC_CHG);
304 send_trap(ibp, &data, sizeof(data));
307 static int __subn_get_opa_nodedesc(struct opa_smp *smp, u32 am,
308 u8 *data, struct ib_device *ibdev,
309 u8 port, u32 *resp_len)
311 struct opa_node_description *nd;
313 if (am) {
314 smp->status |= IB_SMP_INVALID_FIELD;
315 return reply((struct ib_mad_hdr *)smp);
318 nd = (struct opa_node_description *)data;
320 memcpy(nd->data, ibdev->node_desc, sizeof(nd->data));
322 if (resp_len)
323 *resp_len += sizeof(*nd);
325 return reply((struct ib_mad_hdr *)smp);
328 static int __subn_get_opa_nodeinfo(struct opa_smp *smp, u32 am, u8 *data,
329 struct ib_device *ibdev, u8 port,
330 u32 *resp_len)
332 struct opa_node_info *ni;
333 struct hfi1_devdata *dd = dd_from_ibdev(ibdev);
334 unsigned pidx = port - 1; /* IB number port from 1, hw from 0 */
336 ni = (struct opa_node_info *)data;
338 /* GUID 0 is illegal */
339 if (am || pidx >= dd->num_pports || dd->pport[pidx].guid == 0) {
340 smp->status |= IB_SMP_INVALID_FIELD;
341 return reply((struct ib_mad_hdr *)smp);
344 ni->port_guid = cpu_to_be64(dd->pport[pidx].guid);
345 ni->base_version = OPA_MGMT_BASE_VERSION;
346 ni->class_version = OPA_SMI_CLASS_VERSION;
347 ni->node_type = 1; /* channel adapter */
348 ni->num_ports = ibdev->phys_port_cnt;
349 /* This is already in network order */
350 ni->system_image_guid = ib_hfi1_sys_image_guid;
351 /* Use first-port GUID as node */
352 ni->node_guid = cpu_to_be64(dd->pport->guid);
353 ni->partition_cap = cpu_to_be16(hfi1_get_npkeys(dd));
354 ni->device_id = cpu_to_be16(dd->pcidev->device);
355 ni->revision = cpu_to_be32(dd->minrev);
356 ni->local_port_num = port;
357 ni->vendor_id[0] = dd->oui1;
358 ni->vendor_id[1] = dd->oui2;
359 ni->vendor_id[2] = dd->oui3;
361 if (resp_len)
362 *resp_len += sizeof(*ni);
364 return reply((struct ib_mad_hdr *)smp);
367 static int subn_get_nodeinfo(struct ib_smp *smp, struct ib_device *ibdev,
368 u8 port)
370 struct ib_node_info *nip = (struct ib_node_info *)&smp->data;
371 struct hfi1_devdata *dd = dd_from_ibdev(ibdev);
372 unsigned pidx = port - 1; /* IB number port from 1, hw from 0 */
374 /* GUID 0 is illegal */
375 if (smp->attr_mod || pidx >= dd->num_pports ||
376 dd->pport[pidx].guid == 0)
377 smp->status |= IB_SMP_INVALID_FIELD;
378 else
379 nip->port_guid = cpu_to_be64(dd->pport[pidx].guid);
381 nip->base_version = OPA_MGMT_BASE_VERSION;
382 nip->class_version = OPA_SMI_CLASS_VERSION;
383 nip->node_type = 1; /* channel adapter */
384 nip->num_ports = ibdev->phys_port_cnt;
385 /* This is already in network order */
386 nip->sys_guid = ib_hfi1_sys_image_guid;
387 /* Use first-port GUID as node */
388 nip->node_guid = cpu_to_be64(dd->pport->guid);
389 nip->partition_cap = cpu_to_be16(hfi1_get_npkeys(dd));
390 nip->device_id = cpu_to_be16(dd->pcidev->device);
391 nip->revision = cpu_to_be32(dd->minrev);
392 nip->local_port_num = port;
393 nip->vendor_id[0] = dd->oui1;
394 nip->vendor_id[1] = dd->oui2;
395 nip->vendor_id[2] = dd->oui3;
397 return reply((struct ib_mad_hdr *)smp);
400 static void set_link_width_enabled(struct hfi1_pportdata *ppd, u32 w)
402 (void)hfi1_set_ib_cfg(ppd, HFI1_IB_CFG_LWID_ENB, w);
405 static void set_link_width_downgrade_enabled(struct hfi1_pportdata *ppd, u32 w)
407 (void)hfi1_set_ib_cfg(ppd, HFI1_IB_CFG_LWID_DG_ENB, w);
410 static void set_link_speed_enabled(struct hfi1_pportdata *ppd, u32 s)
412 (void)hfi1_set_ib_cfg(ppd, HFI1_IB_CFG_SPD_ENB, s);
415 static int check_mkey(struct hfi1_ibport *ibp, struct ib_mad_hdr *mad,
416 int mad_flags, __be64 mkey, __be32 dr_slid,
417 u8 return_path[], u8 hop_cnt)
419 int valid_mkey = 0;
420 int ret = 0;
422 /* Is the mkey in the process of expiring? */
423 if (ibp->rvp.mkey_lease_timeout &&
424 time_after_eq(jiffies, ibp->rvp.mkey_lease_timeout)) {
425 /* Clear timeout and mkey protection field. */
426 ibp->rvp.mkey_lease_timeout = 0;
427 ibp->rvp.mkeyprot = 0;
430 if ((mad_flags & IB_MAD_IGNORE_MKEY) || ibp->rvp.mkey == 0 ||
431 ibp->rvp.mkey == mkey)
432 valid_mkey = 1;
434 /* Unset lease timeout on any valid Get/Set/TrapRepress */
435 if (valid_mkey && ibp->rvp.mkey_lease_timeout &&
436 (mad->method == IB_MGMT_METHOD_GET ||
437 mad->method == IB_MGMT_METHOD_SET ||
438 mad->method == IB_MGMT_METHOD_TRAP_REPRESS))
439 ibp->rvp.mkey_lease_timeout = 0;
441 if (!valid_mkey) {
442 switch (mad->method) {
443 case IB_MGMT_METHOD_GET:
444 /* Bad mkey not a violation below level 2 */
445 if (ibp->rvp.mkeyprot < 2)
446 break;
447 case IB_MGMT_METHOD_SET:
448 case IB_MGMT_METHOD_TRAP_REPRESS:
449 if (ibp->rvp.mkey_violations != 0xFFFF)
450 ++ibp->rvp.mkey_violations;
451 if (!ibp->rvp.mkey_lease_timeout &&
452 ibp->rvp.mkey_lease_period)
453 ibp->rvp.mkey_lease_timeout = jiffies +
454 ibp->rvp.mkey_lease_period * HZ;
455 /* Generate a trap notice. */
456 bad_mkey(ibp, mad, mkey, dr_slid, return_path,
457 hop_cnt);
458 ret = 1;
462 return ret;
466 * The SMA caches reads from LCB registers in case the LCB is unavailable.
467 * (The LCB is unavailable in certain link states, for example.)
469 struct lcb_datum {
470 u32 off;
471 u64 val;
474 static struct lcb_datum lcb_cache[] = {
475 { DC_LCB_STS_ROUND_TRIP_LTP_CNT, 0 },
478 static int write_lcb_cache(u32 off, u64 val)
480 int i;
482 for (i = 0; i < ARRAY_SIZE(lcb_cache); i++) {
483 if (lcb_cache[i].off == off) {
484 lcb_cache[i].val = val;
485 return 0;
489 pr_warn("%s bad offset 0x%x\n", __func__, off);
490 return -1;
493 static int read_lcb_cache(u32 off, u64 *val)
495 int i;
497 for (i = 0; i < ARRAY_SIZE(lcb_cache); i++) {
498 if (lcb_cache[i].off == off) {
499 *val = lcb_cache[i].val;
500 return 0;
504 pr_warn("%s bad offset 0x%x\n", __func__, off);
505 return -1;
508 void read_ltp_rtt(struct hfi1_devdata *dd)
510 u64 reg;
512 if (read_lcb_csr(dd, DC_LCB_STS_ROUND_TRIP_LTP_CNT, &reg))
513 dd_dev_err(dd, "%s: unable to read LTP RTT\n", __func__);
514 else
515 write_lcb_cache(DC_LCB_STS_ROUND_TRIP_LTP_CNT, reg);
518 static int __subn_get_opa_portinfo(struct opa_smp *smp, u32 am, u8 *data,
519 struct ib_device *ibdev, u8 port,
520 u32 *resp_len)
522 int i;
523 struct hfi1_devdata *dd;
524 struct hfi1_pportdata *ppd;
525 struct hfi1_ibport *ibp;
526 struct opa_port_info *pi = (struct opa_port_info *)data;
527 u8 mtu;
528 u8 credit_rate;
529 u8 is_beaconing_active;
530 u32 state;
531 u32 num_ports = OPA_AM_NPORT(am);
532 u32 start_of_sm_config = OPA_AM_START_SM_CFG(am);
533 u32 buffer_units;
534 u64 tmp = 0;
536 if (num_ports != 1) {
537 smp->status |= IB_SMP_INVALID_FIELD;
538 return reply((struct ib_mad_hdr *)smp);
541 dd = dd_from_ibdev(ibdev);
542 /* IB numbers ports from 1, hw from 0 */
543 ppd = dd->pport + (port - 1);
544 ibp = &ppd->ibport_data;
546 if (ppd->vls_supported / 2 > ARRAY_SIZE(pi->neigh_mtu.pvlx_to_mtu) ||
547 ppd->vls_supported > ARRAY_SIZE(dd->vld)) {
548 smp->status |= IB_SMP_INVALID_FIELD;
549 return reply((struct ib_mad_hdr *)smp);
552 pi->lid = cpu_to_be32(ppd->lid);
554 /* Only return the mkey if the protection field allows it. */
555 if (!(smp->method == IB_MGMT_METHOD_GET &&
556 ibp->rvp.mkey != smp->mkey &&
557 ibp->rvp.mkeyprot == 1))
558 pi->mkey = ibp->rvp.mkey;
560 pi->subnet_prefix = ibp->rvp.gid_prefix;
561 pi->sm_lid = cpu_to_be32(ibp->rvp.sm_lid);
562 pi->ib_cap_mask = cpu_to_be32(ibp->rvp.port_cap_flags);
563 pi->mkey_lease_period = cpu_to_be16(ibp->rvp.mkey_lease_period);
564 pi->sm_trap_qp = cpu_to_be32(ppd->sm_trap_qp);
565 pi->sa_qp = cpu_to_be32(ppd->sa_qp);
567 pi->link_width.enabled = cpu_to_be16(ppd->link_width_enabled);
568 pi->link_width.supported = cpu_to_be16(ppd->link_width_supported);
569 pi->link_width.active = cpu_to_be16(ppd->link_width_active);
571 pi->link_width_downgrade.supported =
572 cpu_to_be16(ppd->link_width_downgrade_supported);
573 pi->link_width_downgrade.enabled =
574 cpu_to_be16(ppd->link_width_downgrade_enabled);
575 pi->link_width_downgrade.tx_active =
576 cpu_to_be16(ppd->link_width_downgrade_tx_active);
577 pi->link_width_downgrade.rx_active =
578 cpu_to_be16(ppd->link_width_downgrade_rx_active);
580 pi->link_speed.supported = cpu_to_be16(ppd->link_speed_supported);
581 pi->link_speed.active = cpu_to_be16(ppd->link_speed_active);
582 pi->link_speed.enabled = cpu_to_be16(ppd->link_speed_enabled);
584 state = driver_lstate(ppd);
586 if (start_of_sm_config && (state == IB_PORT_INIT))
587 ppd->is_sm_config_started = 1;
589 pi->port_phys_conf = (ppd->port_type & 0xf);
591 pi->port_states.ledenable_offlinereason = ppd->neighbor_normal << 4;
592 pi->port_states.ledenable_offlinereason |=
593 ppd->is_sm_config_started << 5;
595 * This pairs with the memory barrier in hfi1_start_led_override to
596 * ensure that we read the correct state of LED beaconing represented
597 * by led_override_timer_active
599 smp_rmb();
600 is_beaconing_active = !!atomic_read(&ppd->led_override_timer_active);
601 pi->port_states.ledenable_offlinereason |= is_beaconing_active << 6;
602 pi->port_states.ledenable_offlinereason |=
603 ppd->offline_disabled_reason;
605 pi->port_states.portphysstate_portstate =
606 (hfi1_ibphys_portstate(ppd) << 4) | state;
608 pi->mkeyprotect_lmc = (ibp->rvp.mkeyprot << 6) | ppd->lmc;
610 memset(pi->neigh_mtu.pvlx_to_mtu, 0, sizeof(pi->neigh_mtu.pvlx_to_mtu));
611 for (i = 0; i < ppd->vls_supported; i++) {
612 mtu = mtu_to_enum(dd->vld[i].mtu, HFI1_DEFAULT_ACTIVE_MTU);
613 if ((i % 2) == 0)
614 pi->neigh_mtu.pvlx_to_mtu[i / 2] |= (mtu << 4);
615 else
616 pi->neigh_mtu.pvlx_to_mtu[i / 2] |= mtu;
618 /* don't forget VL 15 */
619 mtu = mtu_to_enum(dd->vld[15].mtu, 2048);
620 pi->neigh_mtu.pvlx_to_mtu[15 / 2] |= mtu;
621 pi->smsl = ibp->rvp.sm_sl & OPA_PI_MASK_SMSL;
622 pi->operational_vls = hfi1_get_ib_cfg(ppd, HFI1_IB_CFG_OP_VLS);
623 pi->partenforce_filterraw |=
624 (ppd->linkinit_reason & OPA_PI_MASK_LINKINIT_REASON);
625 if (ppd->part_enforce & HFI1_PART_ENFORCE_IN)
626 pi->partenforce_filterraw |= OPA_PI_MASK_PARTITION_ENFORCE_IN;
627 if (ppd->part_enforce & HFI1_PART_ENFORCE_OUT)
628 pi->partenforce_filterraw |= OPA_PI_MASK_PARTITION_ENFORCE_OUT;
629 pi->mkey_violations = cpu_to_be16(ibp->rvp.mkey_violations);
630 /* P_KeyViolations are counted by hardware. */
631 pi->pkey_violations = cpu_to_be16(ibp->rvp.pkey_violations);
632 pi->qkey_violations = cpu_to_be16(ibp->rvp.qkey_violations);
634 pi->vl.cap = ppd->vls_supported;
635 pi->vl.high_limit = cpu_to_be16(ibp->rvp.vl_high_limit);
636 pi->vl.arb_high_cap = (u8)hfi1_get_ib_cfg(ppd, HFI1_IB_CFG_VL_HIGH_CAP);
637 pi->vl.arb_low_cap = (u8)hfi1_get_ib_cfg(ppd, HFI1_IB_CFG_VL_LOW_CAP);
639 pi->clientrereg_subnettimeout = ibp->rvp.subnet_timeout;
641 pi->port_link_mode = cpu_to_be16(OPA_PORT_LINK_MODE_OPA << 10 |
642 OPA_PORT_LINK_MODE_OPA << 5 |
643 OPA_PORT_LINK_MODE_OPA);
645 pi->port_ltp_crc_mode = cpu_to_be16(ppd->port_ltp_crc_mode);
647 pi->port_mode = cpu_to_be16(
648 ppd->is_active_optimize_enabled ?
649 OPA_PI_MASK_PORT_ACTIVE_OPTOMIZE : 0);
651 pi->port_packet_format.supported =
652 cpu_to_be16(OPA_PORT_PACKET_FORMAT_9B);
653 pi->port_packet_format.enabled =
654 cpu_to_be16(OPA_PORT_PACKET_FORMAT_9B);
656 /* flit_control.interleave is (OPA V1, version .76):
657 * bits use
658 * ---- ---
659 * 2 res
660 * 2 DistanceSupported
661 * 2 DistanceEnabled
662 * 5 MaxNextLevelTxEnabled
663 * 5 MaxNestLevelRxSupported
665 * HFI supports only "distance mode 1" (see OPA V1, version .76,
666 * section 9.6.2), so set DistanceSupported, DistanceEnabled
667 * to 0x1.
669 pi->flit_control.interleave = cpu_to_be16(0x1400);
671 pi->link_down_reason = ppd->local_link_down_reason.sma;
672 pi->neigh_link_down_reason = ppd->neigh_link_down_reason.sma;
673 pi->port_error_action = cpu_to_be32(ppd->port_error_action);
674 pi->mtucap = mtu_to_enum(hfi1_max_mtu, IB_MTU_4096);
676 /* 32.768 usec. response time (guessing) */
677 pi->resptimevalue = 3;
679 pi->local_port_num = port;
681 /* buffer info for FM */
682 pi->overall_buffer_space = cpu_to_be16(dd->link_credits);
684 pi->neigh_node_guid = cpu_to_be64(ppd->neighbor_guid);
685 pi->neigh_port_num = ppd->neighbor_port_number;
686 pi->port_neigh_mode =
687 (ppd->neighbor_type & OPA_PI_MASK_NEIGH_NODE_TYPE) |
688 (ppd->mgmt_allowed ? OPA_PI_MASK_NEIGH_MGMT_ALLOWED : 0) |
689 (ppd->neighbor_fm_security ?
690 OPA_PI_MASK_NEIGH_FW_AUTH_BYPASS : 0);
692 /* HFIs shall always return VL15 credits to their
693 * neighbor in a timely manner, without any credit return pacing.
695 credit_rate = 0;
696 buffer_units = (dd->vau) & OPA_PI_MASK_BUF_UNIT_BUF_ALLOC;
697 buffer_units |= (dd->vcu << 3) & OPA_PI_MASK_BUF_UNIT_CREDIT_ACK;
698 buffer_units |= (credit_rate << 6) &
699 OPA_PI_MASK_BUF_UNIT_VL15_CREDIT_RATE;
700 buffer_units |= (dd->vl15_init << 11) & OPA_PI_MASK_BUF_UNIT_VL15_INIT;
701 pi->buffer_units = cpu_to_be32(buffer_units);
703 pi->opa_cap_mask = cpu_to_be16(OPA_CAP_MASK3_IsSharedSpaceSupported);
705 /* HFI supports a replay buffer 128 LTPs in size */
706 pi->replay_depth.buffer = 0x80;
707 /* read the cached value of DC_LCB_STS_ROUND_TRIP_LTP_CNT */
708 read_lcb_cache(DC_LCB_STS_ROUND_TRIP_LTP_CNT, &tmp);
711 * this counter is 16 bits wide, but the replay_depth.wire
712 * variable is only 8 bits
714 if (tmp > 0xff)
715 tmp = 0xff;
716 pi->replay_depth.wire = tmp;
718 if (resp_len)
719 *resp_len += sizeof(struct opa_port_info);
721 return reply((struct ib_mad_hdr *)smp);
725 * get_pkeys - return the PKEY table
726 * @dd: the hfi1_ib device
727 * @port: the IB port number
728 * @pkeys: the pkey table is placed here
730 static int get_pkeys(struct hfi1_devdata *dd, u8 port, u16 *pkeys)
732 struct hfi1_pportdata *ppd = dd->pport + port - 1;
734 memcpy(pkeys, ppd->pkeys, sizeof(ppd->pkeys));
736 return 0;
739 static int __subn_get_opa_pkeytable(struct opa_smp *smp, u32 am, u8 *data,
740 struct ib_device *ibdev, u8 port,
741 u32 *resp_len)
743 struct hfi1_devdata *dd = dd_from_ibdev(ibdev);
744 u32 n_blocks_req = OPA_AM_NBLK(am);
745 u32 start_block = am & 0x7ff;
746 __be16 *p;
747 u16 *q;
748 int i;
749 u16 n_blocks_avail;
750 unsigned npkeys = hfi1_get_npkeys(dd);
751 size_t size;
753 if (n_blocks_req == 0) {
754 pr_warn("OPA Get PKey AM Invalid : P = %d; B = 0x%x; N = 0x%x\n",
755 port, start_block, n_blocks_req);
756 smp->status |= IB_SMP_INVALID_FIELD;
757 return reply((struct ib_mad_hdr *)smp);
760 n_blocks_avail = (u16)(npkeys / OPA_PARTITION_TABLE_BLK_SIZE) + 1;
762 size = (n_blocks_req * OPA_PARTITION_TABLE_BLK_SIZE) * sizeof(u16);
764 if (start_block + n_blocks_req > n_blocks_avail ||
765 n_blocks_req > OPA_NUM_PKEY_BLOCKS_PER_SMP) {
766 pr_warn("OPA Get PKey AM Invalid : s 0x%x; req 0x%x; "
767 "avail 0x%x; blk/smp 0x%lx\n",
768 start_block, n_blocks_req, n_blocks_avail,
769 OPA_NUM_PKEY_BLOCKS_PER_SMP);
770 smp->status |= IB_SMP_INVALID_FIELD;
771 return reply((struct ib_mad_hdr *)smp);
774 p = (__be16 *)data;
775 q = (u16 *)data;
776 /* get the real pkeys if we are requesting the first block */
777 if (start_block == 0) {
778 get_pkeys(dd, port, q);
779 for (i = 0; i < npkeys; i++)
780 p[i] = cpu_to_be16(q[i]);
781 if (resp_len)
782 *resp_len += size;
783 } else {
784 smp->status |= IB_SMP_INVALID_FIELD;
786 return reply((struct ib_mad_hdr *)smp);
789 enum {
790 HFI_TRANSITION_DISALLOWED,
791 HFI_TRANSITION_IGNORED,
792 HFI_TRANSITION_ALLOWED,
793 HFI_TRANSITION_UNDEFINED,
797 * Use shortened names to improve readability of
798 * {logical,physical}_state_transitions
800 enum {
801 __D = HFI_TRANSITION_DISALLOWED,
802 __I = HFI_TRANSITION_IGNORED,
803 __A = HFI_TRANSITION_ALLOWED,
804 __U = HFI_TRANSITION_UNDEFINED,
808 * IB_PORTPHYSSTATE_POLLING (2) through OPA_PORTPHYSSTATE_MAX (11) are
809 * represented in physical_state_transitions.
811 #define __N_PHYSTATES (OPA_PORTPHYSSTATE_MAX - IB_PORTPHYSSTATE_POLLING + 1)
814 * Within physical_state_transitions, rows represent "old" states,
815 * columns "new" states, and physical_state_transitions.allowed[old][new]
816 * indicates if the transition from old state to new state is legal (see
817 * OPAg1v1, Table 6-4).
819 static const struct {
820 u8 allowed[__N_PHYSTATES][__N_PHYSTATES];
821 } physical_state_transitions = {
823 /* 2 3 4 5 6 7 8 9 10 11 */
824 /* 2 */ { __A, __A, __D, __D, __D, __D, __D, __D, __D, __D },
825 /* 3 */ { __A, __I, __D, __D, __D, __D, __D, __D, __D, __A },
826 /* 4 */ { __U, __U, __U, __U, __U, __U, __U, __U, __U, __U },
827 /* 5 */ { __A, __A, __D, __I, __D, __D, __D, __D, __D, __D },
828 /* 6 */ { __U, __U, __U, __U, __U, __U, __U, __U, __U, __U },
829 /* 7 */ { __D, __A, __D, __D, __D, __I, __D, __D, __D, __D },
830 /* 8 */ { __U, __U, __U, __U, __U, __U, __U, __U, __U, __U },
831 /* 9 */ { __I, __A, __D, __D, __D, __D, __D, __I, __D, __D },
832 /*10 */ { __U, __U, __U, __U, __U, __U, __U, __U, __U, __U },
833 /*11 */ { __D, __A, __D, __D, __D, __D, __D, __D, __D, __I },
838 * IB_PORT_DOWN (1) through IB_PORT_ACTIVE_DEFER (5) are represented
839 * logical_state_transitions
842 #define __N_LOGICAL_STATES (IB_PORT_ACTIVE_DEFER - IB_PORT_DOWN + 1)
845 * Within logical_state_transitions rows represent "old" states,
846 * columns "new" states, and logical_state_transitions.allowed[old][new]
847 * indicates if the transition from old state to new state is legal (see
848 * OPAg1v1, Table 9-12).
850 static const struct {
851 u8 allowed[__N_LOGICAL_STATES][__N_LOGICAL_STATES];
852 } logical_state_transitions = {
854 /* 1 2 3 4 5 */
855 /* 1 */ { __I, __D, __D, __D, __U},
856 /* 2 */ { __D, __I, __A, __D, __U},
857 /* 3 */ { __D, __D, __I, __A, __U},
858 /* 4 */ { __D, __D, __I, __I, __U},
859 /* 5 */ { __U, __U, __U, __U, __U},
863 static int logical_transition_allowed(int old, int new)
865 if (old < IB_PORT_NOP || old > IB_PORT_ACTIVE_DEFER ||
866 new < IB_PORT_NOP || new > IB_PORT_ACTIVE_DEFER) {
867 pr_warn("invalid logical state(s) (old %d new %d)\n",
868 old, new);
869 return HFI_TRANSITION_UNDEFINED;
872 if (new == IB_PORT_NOP)
873 return HFI_TRANSITION_ALLOWED; /* always allowed */
875 /* adjust states for indexing into logical_state_transitions */
876 old -= IB_PORT_DOWN;
877 new -= IB_PORT_DOWN;
879 if (old < 0 || new < 0)
880 return HFI_TRANSITION_UNDEFINED;
881 return logical_state_transitions.allowed[old][new];
884 static int physical_transition_allowed(int old, int new)
886 if (old < IB_PORTPHYSSTATE_NOP || old > OPA_PORTPHYSSTATE_MAX ||
887 new < IB_PORTPHYSSTATE_NOP || new > OPA_PORTPHYSSTATE_MAX) {
888 pr_warn("invalid physical state(s) (old %d new %d)\n",
889 old, new);
890 return HFI_TRANSITION_UNDEFINED;
893 if (new == IB_PORTPHYSSTATE_NOP)
894 return HFI_TRANSITION_ALLOWED; /* always allowed */
896 /* adjust states for indexing into physical_state_transitions */
897 old -= IB_PORTPHYSSTATE_POLLING;
898 new -= IB_PORTPHYSSTATE_POLLING;
900 if (old < 0 || new < 0)
901 return HFI_TRANSITION_UNDEFINED;
902 return physical_state_transitions.allowed[old][new];
905 static int port_states_transition_allowed(struct hfi1_pportdata *ppd,
906 u32 logical_new, u32 physical_new)
908 u32 physical_old = driver_physical_state(ppd);
909 u32 logical_old = driver_logical_state(ppd);
910 int ret, logical_allowed, physical_allowed;
912 ret = logical_transition_allowed(logical_old, logical_new);
913 logical_allowed = ret;
915 if (ret == HFI_TRANSITION_DISALLOWED ||
916 ret == HFI_TRANSITION_UNDEFINED) {
917 pr_warn("invalid logical state transition %s -> %s\n",
918 opa_lstate_name(logical_old),
919 opa_lstate_name(logical_new));
920 return ret;
923 ret = physical_transition_allowed(physical_old, physical_new);
924 physical_allowed = ret;
926 if (ret == HFI_TRANSITION_DISALLOWED ||
927 ret == HFI_TRANSITION_UNDEFINED) {
928 pr_warn("invalid physical state transition %s -> %s\n",
929 opa_pstate_name(physical_old),
930 opa_pstate_name(physical_new));
931 return ret;
934 if (logical_allowed == HFI_TRANSITION_IGNORED &&
935 physical_allowed == HFI_TRANSITION_IGNORED)
936 return HFI_TRANSITION_IGNORED;
939 * A change request of Physical Port State from
940 * 'Offline' to 'Polling' should be ignored.
942 if ((physical_old == OPA_PORTPHYSSTATE_OFFLINE) &&
943 (physical_new == IB_PORTPHYSSTATE_POLLING))
944 return HFI_TRANSITION_IGNORED;
947 * Either physical_allowed or logical_allowed is
948 * HFI_TRANSITION_ALLOWED.
950 return HFI_TRANSITION_ALLOWED;
953 static int set_port_states(struct hfi1_pportdata *ppd, struct opa_smp *smp,
954 u32 logical_state, u32 phys_state,
955 int suppress_idle_sma)
957 struct hfi1_devdata *dd = ppd->dd;
958 u32 link_state;
959 int ret;
961 ret = port_states_transition_allowed(ppd, logical_state, phys_state);
962 if (ret == HFI_TRANSITION_DISALLOWED ||
963 ret == HFI_TRANSITION_UNDEFINED) {
964 /* error message emitted above */
965 smp->status |= IB_SMP_INVALID_FIELD;
966 return 0;
969 if (ret == HFI_TRANSITION_IGNORED)
970 return 0;
972 if ((phys_state != IB_PORTPHYSSTATE_NOP) &&
973 !(logical_state == IB_PORT_DOWN ||
974 logical_state == IB_PORT_NOP)){
975 pr_warn("SubnSet(OPA_PortInfo) port state invalid: logical_state 0x%x physical_state 0x%x\n",
976 logical_state, phys_state);
977 smp->status |= IB_SMP_INVALID_FIELD;
981 * Logical state changes are summarized in OPAv1g1 spec.,
982 * Table 9-12; physical state changes are summarized in
983 * OPAv1g1 spec., Table 6.4.
985 switch (logical_state) {
986 case IB_PORT_NOP:
987 if (phys_state == IB_PORTPHYSSTATE_NOP)
988 break;
989 /* FALLTHROUGH */
990 case IB_PORT_DOWN:
991 if (phys_state == IB_PORTPHYSSTATE_NOP) {
992 link_state = HLS_DN_DOWNDEF;
993 } else if (phys_state == IB_PORTPHYSSTATE_POLLING) {
994 link_state = HLS_DN_POLL;
995 set_link_down_reason(ppd, OPA_LINKDOWN_REASON_FM_BOUNCE,
996 0, OPA_LINKDOWN_REASON_FM_BOUNCE);
997 } else if (phys_state == IB_PORTPHYSSTATE_DISABLED) {
998 link_state = HLS_DN_DISABLE;
999 } else {
1000 pr_warn("SubnSet(OPA_PortInfo) invalid physical state 0x%x\n",
1001 phys_state);
1002 smp->status |= IB_SMP_INVALID_FIELD;
1003 break;
1006 if ((link_state == HLS_DN_POLL ||
1007 link_state == HLS_DN_DOWNDEF)) {
1009 * Going to poll. No matter what the current state,
1010 * always move offline first, then tune and start the
1011 * link. This correctly handles a FM link bounce and
1012 * a link enable. Going offline is a no-op if already
1013 * offline.
1015 set_link_state(ppd, HLS_DN_OFFLINE);
1016 tune_serdes(ppd);
1017 start_link(ppd);
1018 } else {
1019 set_link_state(ppd, link_state);
1021 if (link_state == HLS_DN_DISABLE &&
1022 (ppd->offline_disabled_reason >
1023 HFI1_ODR_MASK(OPA_LINKDOWN_REASON_SMA_DISABLED) ||
1024 ppd->offline_disabled_reason ==
1025 HFI1_ODR_MASK(OPA_LINKDOWN_REASON_NONE)))
1026 ppd->offline_disabled_reason =
1027 HFI1_ODR_MASK(OPA_LINKDOWN_REASON_SMA_DISABLED);
1029 * Don't send a reply if the response would be sent
1030 * through the disabled port.
1032 if (link_state == HLS_DN_DISABLE && smp->hop_cnt)
1033 return IB_MAD_RESULT_SUCCESS | IB_MAD_RESULT_CONSUMED;
1034 break;
1035 case IB_PORT_ARMED:
1036 ret = set_link_state(ppd, HLS_UP_ARMED);
1037 if ((ret == 0) && (suppress_idle_sma == 0))
1038 send_idle_sma(dd, SMA_IDLE_ARM);
1039 break;
1040 case IB_PORT_ACTIVE:
1041 if (ppd->neighbor_normal) {
1042 ret = set_link_state(ppd, HLS_UP_ACTIVE);
1043 if (ret == 0)
1044 send_idle_sma(dd, SMA_IDLE_ACTIVE);
1045 } else {
1046 pr_warn("SubnSet(OPA_PortInfo) Cannot move to Active with NeighborNormal 0\n");
1047 smp->status |= IB_SMP_INVALID_FIELD;
1049 break;
1050 default:
1051 pr_warn("SubnSet(OPA_PortInfo) invalid logical state 0x%x\n",
1052 logical_state);
1053 smp->status |= IB_SMP_INVALID_FIELD;
1056 return 0;
1060 * subn_set_opa_portinfo - set port information
1061 * @smp: the incoming SM packet
1062 * @ibdev: the infiniband device
1063 * @port: the port on the device
1066 static int __subn_set_opa_portinfo(struct opa_smp *smp, u32 am, u8 *data,
1067 struct ib_device *ibdev, u8 port,
1068 u32 *resp_len)
1070 struct opa_port_info *pi = (struct opa_port_info *)data;
1071 struct ib_event event;
1072 struct hfi1_devdata *dd;
1073 struct hfi1_pportdata *ppd;
1074 struct hfi1_ibport *ibp;
1075 u8 clientrereg;
1076 unsigned long flags;
1077 u32 smlid, opa_lid; /* tmp vars to hold LID values */
1078 u16 lid;
1079 u8 ls_old, ls_new, ps_new;
1080 u8 vls;
1081 u8 msl;
1082 u8 crc_enabled;
1083 u16 lse, lwe, mtu;
1084 u32 num_ports = OPA_AM_NPORT(am);
1085 u32 start_of_sm_config = OPA_AM_START_SM_CFG(am);
1086 int ret, i, invalid = 0, call_set_mtu = 0;
1087 int call_link_downgrade_policy = 0;
1089 if (num_ports != 1) {
1090 smp->status |= IB_SMP_INVALID_FIELD;
1091 return reply((struct ib_mad_hdr *)smp);
1094 opa_lid = be32_to_cpu(pi->lid);
1095 if (opa_lid & 0xFFFF0000) {
1096 pr_warn("OPA_PortInfo lid out of range: %X\n", opa_lid);
1097 smp->status |= IB_SMP_INVALID_FIELD;
1098 goto get_only;
1101 lid = (u16)(opa_lid & 0x0000FFFF);
1103 smlid = be32_to_cpu(pi->sm_lid);
1104 if (smlid & 0xFFFF0000) {
1105 pr_warn("OPA_PortInfo SM lid out of range: %X\n", smlid);
1106 smp->status |= IB_SMP_INVALID_FIELD;
1107 goto get_only;
1109 smlid &= 0x0000FFFF;
1111 clientrereg = (pi->clientrereg_subnettimeout &
1112 OPA_PI_MASK_CLIENT_REREGISTER);
1114 dd = dd_from_ibdev(ibdev);
1115 /* IB numbers ports from 1, hw from 0 */
1116 ppd = dd->pport + (port - 1);
1117 ibp = &ppd->ibport_data;
1118 event.device = ibdev;
1119 event.element.port_num = port;
1121 ls_old = driver_lstate(ppd);
1123 ibp->rvp.mkey = pi->mkey;
1124 ibp->rvp.gid_prefix = pi->subnet_prefix;
1125 ibp->rvp.mkey_lease_period = be16_to_cpu(pi->mkey_lease_period);
1127 /* Must be a valid unicast LID address. */
1128 if ((lid == 0 && ls_old > IB_PORT_INIT) ||
1129 lid >= be16_to_cpu(IB_MULTICAST_LID_BASE)) {
1130 smp->status |= IB_SMP_INVALID_FIELD;
1131 pr_warn("SubnSet(OPA_PortInfo) lid invalid 0x%x\n",
1132 lid);
1133 } else if (ppd->lid != lid ||
1134 ppd->lmc != (pi->mkeyprotect_lmc & OPA_PI_MASK_LMC)) {
1135 if (ppd->lid != lid)
1136 hfi1_set_uevent_bits(ppd, _HFI1_EVENT_LID_CHANGE_BIT);
1137 if (ppd->lmc != (pi->mkeyprotect_lmc & OPA_PI_MASK_LMC))
1138 hfi1_set_uevent_bits(ppd, _HFI1_EVENT_LMC_CHANGE_BIT);
1139 hfi1_set_lid(ppd, lid, pi->mkeyprotect_lmc & OPA_PI_MASK_LMC);
1140 event.event = IB_EVENT_LID_CHANGE;
1141 ib_dispatch_event(&event);
1144 msl = pi->smsl & OPA_PI_MASK_SMSL;
1145 if (pi->partenforce_filterraw & OPA_PI_MASK_LINKINIT_REASON)
1146 ppd->linkinit_reason =
1147 (pi->partenforce_filterraw &
1148 OPA_PI_MASK_LINKINIT_REASON);
1149 /* enable/disable SW pkey checking as per FM control */
1150 if (pi->partenforce_filterraw & OPA_PI_MASK_PARTITION_ENFORCE_IN)
1151 ppd->part_enforce |= HFI1_PART_ENFORCE_IN;
1152 else
1153 ppd->part_enforce &= ~HFI1_PART_ENFORCE_IN;
1155 if (pi->partenforce_filterraw & OPA_PI_MASK_PARTITION_ENFORCE_OUT)
1156 ppd->part_enforce |= HFI1_PART_ENFORCE_OUT;
1157 else
1158 ppd->part_enforce &= ~HFI1_PART_ENFORCE_OUT;
1160 /* Must be a valid unicast LID address. */
1161 if ((smlid == 0 && ls_old > IB_PORT_INIT) ||
1162 smlid >= be16_to_cpu(IB_MULTICAST_LID_BASE)) {
1163 smp->status |= IB_SMP_INVALID_FIELD;
1164 pr_warn("SubnSet(OPA_PortInfo) smlid invalid 0x%x\n", smlid);
1165 } else if (smlid != ibp->rvp.sm_lid || msl != ibp->rvp.sm_sl) {
1166 pr_warn("SubnSet(OPA_PortInfo) smlid 0x%x\n", smlid);
1167 spin_lock_irqsave(&ibp->rvp.lock, flags);
1168 if (ibp->rvp.sm_ah) {
1169 if (smlid != ibp->rvp.sm_lid)
1170 ibp->rvp.sm_ah->attr.dlid = smlid;
1171 if (msl != ibp->rvp.sm_sl)
1172 ibp->rvp.sm_ah->attr.sl = msl;
1174 spin_unlock_irqrestore(&ibp->rvp.lock, flags);
1175 if (smlid != ibp->rvp.sm_lid)
1176 ibp->rvp.sm_lid = smlid;
1177 if (msl != ibp->rvp.sm_sl)
1178 ibp->rvp.sm_sl = msl;
1179 event.event = IB_EVENT_SM_CHANGE;
1180 ib_dispatch_event(&event);
1183 if (pi->link_down_reason == 0) {
1184 ppd->local_link_down_reason.sma = 0;
1185 ppd->local_link_down_reason.latest = 0;
1188 if (pi->neigh_link_down_reason == 0) {
1189 ppd->neigh_link_down_reason.sma = 0;
1190 ppd->neigh_link_down_reason.latest = 0;
1193 ppd->sm_trap_qp = be32_to_cpu(pi->sm_trap_qp);
1194 ppd->sa_qp = be32_to_cpu(pi->sa_qp);
1196 ppd->port_error_action = be32_to_cpu(pi->port_error_action);
1197 lwe = be16_to_cpu(pi->link_width.enabled);
1198 if (lwe) {
1199 if (lwe == OPA_LINK_WIDTH_RESET ||
1200 lwe == OPA_LINK_WIDTH_RESET_OLD)
1201 set_link_width_enabled(ppd, ppd->link_width_supported);
1202 else if ((lwe & ~ppd->link_width_supported) == 0)
1203 set_link_width_enabled(ppd, lwe);
1204 else
1205 smp->status |= IB_SMP_INVALID_FIELD;
1207 lwe = be16_to_cpu(pi->link_width_downgrade.enabled);
1208 /* LWD.E is always applied - 0 means "disabled" */
1209 if (lwe == OPA_LINK_WIDTH_RESET ||
1210 lwe == OPA_LINK_WIDTH_RESET_OLD) {
1211 set_link_width_downgrade_enabled(ppd,
1212 ppd->
1213 link_width_downgrade_supported
1215 } else if ((lwe & ~ppd->link_width_downgrade_supported) == 0) {
1216 /* only set and apply if something changed */
1217 if (lwe != ppd->link_width_downgrade_enabled) {
1218 set_link_width_downgrade_enabled(ppd, lwe);
1219 call_link_downgrade_policy = 1;
1221 } else {
1222 smp->status |= IB_SMP_INVALID_FIELD;
1224 lse = be16_to_cpu(pi->link_speed.enabled);
1225 if (lse) {
1226 if (lse & be16_to_cpu(pi->link_speed.supported))
1227 set_link_speed_enabled(ppd, lse);
1228 else
1229 smp->status |= IB_SMP_INVALID_FIELD;
1232 ibp->rvp.mkeyprot =
1233 (pi->mkeyprotect_lmc & OPA_PI_MASK_MKEY_PROT_BIT) >> 6;
1234 ibp->rvp.vl_high_limit = be16_to_cpu(pi->vl.high_limit) & 0xFF;
1235 (void)hfi1_set_ib_cfg(ppd, HFI1_IB_CFG_VL_HIGH_LIMIT,
1236 ibp->rvp.vl_high_limit);
1238 if (ppd->vls_supported / 2 > ARRAY_SIZE(pi->neigh_mtu.pvlx_to_mtu) ||
1239 ppd->vls_supported > ARRAY_SIZE(dd->vld)) {
1240 smp->status |= IB_SMP_INVALID_FIELD;
1241 return reply((struct ib_mad_hdr *)smp);
1243 for (i = 0; i < ppd->vls_supported; i++) {
1244 if ((i % 2) == 0)
1245 mtu = enum_to_mtu((pi->neigh_mtu.pvlx_to_mtu[i / 2] >>
1246 4) & 0xF);
1247 else
1248 mtu = enum_to_mtu(pi->neigh_mtu.pvlx_to_mtu[i / 2] &
1249 0xF);
1250 if (mtu == 0xffff) {
1251 pr_warn("SubnSet(OPA_PortInfo) mtu invalid %d (0x%x)\n",
1252 mtu,
1253 (pi->neigh_mtu.pvlx_to_mtu[0] >> 4) & 0xF);
1254 smp->status |= IB_SMP_INVALID_FIELD;
1255 mtu = hfi1_max_mtu; /* use a valid MTU */
1257 if (dd->vld[i].mtu != mtu) {
1258 dd_dev_info(dd,
1259 "MTU change on vl %d from %d to %d\n",
1260 i, dd->vld[i].mtu, mtu);
1261 dd->vld[i].mtu = mtu;
1262 call_set_mtu++;
1265 /* As per OPAV1 spec: VL15 must support and be configured
1266 * for operation with a 2048 or larger MTU.
1268 mtu = enum_to_mtu(pi->neigh_mtu.pvlx_to_mtu[15 / 2] & 0xF);
1269 if (mtu < 2048 || mtu == 0xffff)
1270 mtu = 2048;
1271 if (dd->vld[15].mtu != mtu) {
1272 dd_dev_info(dd,
1273 "MTU change on vl 15 from %d to %d\n",
1274 dd->vld[15].mtu, mtu);
1275 dd->vld[15].mtu = mtu;
1276 call_set_mtu++;
1278 if (call_set_mtu)
1279 set_mtu(ppd);
1281 /* Set operational VLs */
1282 vls = pi->operational_vls & OPA_PI_MASK_OPERATIONAL_VL;
1283 if (vls) {
1284 if (vls > ppd->vls_supported) {
1285 pr_warn("SubnSet(OPA_PortInfo) VL's supported invalid %d\n",
1286 pi->operational_vls);
1287 smp->status |= IB_SMP_INVALID_FIELD;
1288 } else {
1289 if (hfi1_set_ib_cfg(ppd, HFI1_IB_CFG_OP_VLS,
1290 vls) == -EINVAL)
1291 smp->status |= IB_SMP_INVALID_FIELD;
1295 if (pi->mkey_violations == 0)
1296 ibp->rvp.mkey_violations = 0;
1298 if (pi->pkey_violations == 0)
1299 ibp->rvp.pkey_violations = 0;
1301 if (pi->qkey_violations == 0)
1302 ibp->rvp.qkey_violations = 0;
1304 ibp->rvp.subnet_timeout =
1305 pi->clientrereg_subnettimeout & OPA_PI_MASK_SUBNET_TIMEOUT;
1307 crc_enabled = be16_to_cpu(pi->port_ltp_crc_mode);
1308 crc_enabled >>= 4;
1309 crc_enabled &= 0xf;
1311 if (crc_enabled != 0)
1312 ppd->port_crc_mode_enabled = port_ltp_to_cap(crc_enabled);
1314 ppd->is_active_optimize_enabled =
1315 !!(be16_to_cpu(pi->port_mode)
1316 & OPA_PI_MASK_PORT_ACTIVE_OPTOMIZE);
1318 ls_new = pi->port_states.portphysstate_portstate &
1319 OPA_PI_MASK_PORT_STATE;
1320 ps_new = (pi->port_states.portphysstate_portstate &
1321 OPA_PI_MASK_PORT_PHYSICAL_STATE) >> 4;
1323 if (ls_old == IB_PORT_INIT) {
1324 if (start_of_sm_config) {
1325 if (ls_new == ls_old || (ls_new == IB_PORT_ARMED))
1326 ppd->is_sm_config_started = 1;
1327 } else if (ls_new == IB_PORT_ARMED) {
1328 if (ppd->is_sm_config_started == 0)
1329 invalid = 1;
1333 /* Handle CLIENT_REREGISTER event b/c SM asked us for it */
1334 if (clientrereg) {
1335 event.event = IB_EVENT_CLIENT_REREGISTER;
1336 ib_dispatch_event(&event);
1340 * Do the port state change now that the other link parameters
1341 * have been set.
1342 * Changing the port physical state only makes sense if the link
1343 * is down or is being set to down.
1346 ret = set_port_states(ppd, smp, ls_new, ps_new, invalid);
1347 if (ret)
1348 return ret;
1350 ret = __subn_get_opa_portinfo(smp, am, data, ibdev, port, resp_len);
1352 /* restore re-reg bit per o14-12.2.1 */
1353 pi->clientrereg_subnettimeout |= clientrereg;
1356 * Apply the new link downgrade policy. This may result in a link
1357 * bounce. Do this after everything else so things are settled.
1358 * Possible problem: if setting the port state above fails, then
1359 * the policy change is not applied.
1361 if (call_link_downgrade_policy)
1362 apply_link_downgrade_policy(ppd, 0);
1364 return ret;
1366 get_only:
1367 return __subn_get_opa_portinfo(smp, am, data, ibdev, port, resp_len);
1371 * set_pkeys - set the PKEY table for ctxt 0
1372 * @dd: the hfi1_ib device
1373 * @port: the IB port number
1374 * @pkeys: the PKEY table
1376 static int set_pkeys(struct hfi1_devdata *dd, u8 port, u16 *pkeys)
1378 struct hfi1_pportdata *ppd;
1379 int i;
1380 int changed = 0;
1381 int update_includes_mgmt_partition = 0;
1384 * IB port one/two always maps to context zero/one,
1385 * always a kernel context, no locking needed
1386 * If we get here with ppd setup, no need to check
1387 * that rcd is valid.
1389 ppd = dd->pport + (port - 1);
1391 * If the update does not include the management pkey, don't do it.
1393 for (i = 0; i < ARRAY_SIZE(ppd->pkeys); i++) {
1394 if (pkeys[i] == LIM_MGMT_P_KEY) {
1395 update_includes_mgmt_partition = 1;
1396 break;
1400 if (!update_includes_mgmt_partition)
1401 return 1;
1403 for (i = 0; i < ARRAY_SIZE(ppd->pkeys); i++) {
1404 u16 key = pkeys[i];
1405 u16 okey = ppd->pkeys[i];
1407 if (key == okey)
1408 continue;
1410 * Don't update pkeys[2], if an HFI port without MgmtAllowed
1411 * by neighbor is a switch.
1413 if (i == 2 && !ppd->mgmt_allowed && ppd->neighbor_type == 1)
1414 continue;
1416 * The SM gives us the complete PKey table. We have
1417 * to ensure that we put the PKeys in the matching
1418 * slots.
1420 ppd->pkeys[i] = key;
1421 changed = 1;
1424 if (changed) {
1425 (void)hfi1_set_ib_cfg(ppd, HFI1_IB_CFG_PKEYS, 0);
1426 hfi1_event_pkey_change(dd, port);
1429 return 0;
1432 static int __subn_set_opa_pkeytable(struct opa_smp *smp, u32 am, u8 *data,
1433 struct ib_device *ibdev, u8 port,
1434 u32 *resp_len)
1436 struct hfi1_devdata *dd = dd_from_ibdev(ibdev);
1437 u32 n_blocks_sent = OPA_AM_NBLK(am);
1438 u32 start_block = am & 0x7ff;
1439 u16 *p = (u16 *)data;
1440 __be16 *q = (__be16 *)data;
1441 int i;
1442 u16 n_blocks_avail;
1443 unsigned npkeys = hfi1_get_npkeys(dd);
1445 if (n_blocks_sent == 0) {
1446 pr_warn("OPA Get PKey AM Invalid : P = %d; B = 0x%x; N = 0x%x\n",
1447 port, start_block, n_blocks_sent);
1448 smp->status |= IB_SMP_INVALID_FIELD;
1449 return reply((struct ib_mad_hdr *)smp);
1452 n_blocks_avail = (u16)(npkeys / OPA_PARTITION_TABLE_BLK_SIZE) + 1;
1454 if (start_block + n_blocks_sent > n_blocks_avail ||
1455 n_blocks_sent > OPA_NUM_PKEY_BLOCKS_PER_SMP) {
1456 pr_warn("OPA Set PKey AM Invalid : s 0x%x; req 0x%x; avail 0x%x; blk/smp 0x%lx\n",
1457 start_block, n_blocks_sent, n_blocks_avail,
1458 OPA_NUM_PKEY_BLOCKS_PER_SMP);
1459 smp->status |= IB_SMP_INVALID_FIELD;
1460 return reply((struct ib_mad_hdr *)smp);
1463 for (i = 0; i < n_blocks_sent * OPA_PARTITION_TABLE_BLK_SIZE; i++)
1464 p[i] = be16_to_cpu(q[i]);
1466 if (start_block == 0 && set_pkeys(dd, port, p) != 0) {
1467 smp->status |= IB_SMP_INVALID_FIELD;
1468 return reply((struct ib_mad_hdr *)smp);
1471 return __subn_get_opa_pkeytable(smp, am, data, ibdev, port, resp_len);
1474 static int get_sc2vlt_tables(struct hfi1_devdata *dd, void *data)
1476 u64 *val = data;
1478 *val++ = read_csr(dd, SEND_SC2VLT0);
1479 *val++ = read_csr(dd, SEND_SC2VLT1);
1480 *val++ = read_csr(dd, SEND_SC2VLT2);
1481 *val++ = read_csr(dd, SEND_SC2VLT3);
1482 return 0;
1485 #define ILLEGAL_VL 12
1487 * filter_sc2vlt changes mappings to VL15 to ILLEGAL_VL (except
1488 * for SC15, which must map to VL15). If we don't remap things this
1489 * way it is possible for VL15 counters to increment when we try to
1490 * send on a SC which is mapped to an invalid VL.
1492 static void filter_sc2vlt(void *data)
1494 int i;
1495 u8 *pd = data;
1497 for (i = 0; i < OPA_MAX_SCS; i++) {
1498 if (i == 15)
1499 continue;
1500 if ((pd[i] & 0x1f) == 0xf)
1501 pd[i] = ILLEGAL_VL;
1505 static int set_sc2vlt_tables(struct hfi1_devdata *dd, void *data)
1507 u64 *val = data;
1509 filter_sc2vlt(data);
1511 write_csr(dd, SEND_SC2VLT0, *val++);
1512 write_csr(dd, SEND_SC2VLT1, *val++);
1513 write_csr(dd, SEND_SC2VLT2, *val++);
1514 write_csr(dd, SEND_SC2VLT3, *val++);
1515 write_seqlock_irq(&dd->sc2vl_lock);
1516 memcpy(dd->sc2vl, data, sizeof(dd->sc2vl));
1517 write_sequnlock_irq(&dd->sc2vl_lock);
1518 return 0;
1521 static int __subn_get_opa_sl_to_sc(struct opa_smp *smp, u32 am, u8 *data,
1522 struct ib_device *ibdev, u8 port,
1523 u32 *resp_len)
1525 struct hfi1_ibport *ibp = to_iport(ibdev, port);
1526 u8 *p = data;
1527 size_t size = ARRAY_SIZE(ibp->sl_to_sc); /* == 32 */
1528 unsigned i;
1530 if (am) {
1531 smp->status |= IB_SMP_INVALID_FIELD;
1532 return reply((struct ib_mad_hdr *)smp);
1535 for (i = 0; i < ARRAY_SIZE(ibp->sl_to_sc); i++)
1536 *p++ = ibp->sl_to_sc[i];
1538 if (resp_len)
1539 *resp_len += size;
1541 return reply((struct ib_mad_hdr *)smp);
1544 static int __subn_set_opa_sl_to_sc(struct opa_smp *smp, u32 am, u8 *data,
1545 struct ib_device *ibdev, u8 port,
1546 u32 *resp_len)
1548 struct hfi1_ibport *ibp = to_iport(ibdev, port);
1549 u8 *p = data;
1550 int i;
1551 u8 sc;
1553 if (am) {
1554 smp->status |= IB_SMP_INVALID_FIELD;
1555 return reply((struct ib_mad_hdr *)smp);
1558 for (i = 0; i < ARRAY_SIZE(ibp->sl_to_sc); i++) {
1559 sc = *p++;
1560 if (ibp->sl_to_sc[i] != sc) {
1561 ibp->sl_to_sc[i] = sc;
1563 /* Put all stale qps into error state */
1564 hfi1_error_port_qps(ibp, i);
1568 return __subn_get_opa_sl_to_sc(smp, am, data, ibdev, port, resp_len);
1571 static int __subn_get_opa_sc_to_sl(struct opa_smp *smp, u32 am, u8 *data,
1572 struct ib_device *ibdev, u8 port,
1573 u32 *resp_len)
1575 struct hfi1_ibport *ibp = to_iport(ibdev, port);
1576 u8 *p = data;
1577 size_t size = ARRAY_SIZE(ibp->sc_to_sl); /* == 32 */
1578 unsigned i;
1580 if (am) {
1581 smp->status |= IB_SMP_INVALID_FIELD;
1582 return reply((struct ib_mad_hdr *)smp);
1585 for (i = 0; i < ARRAY_SIZE(ibp->sc_to_sl); i++)
1586 *p++ = ibp->sc_to_sl[i];
1588 if (resp_len)
1589 *resp_len += size;
1591 return reply((struct ib_mad_hdr *)smp);
1594 static int __subn_set_opa_sc_to_sl(struct opa_smp *smp, u32 am, u8 *data,
1595 struct ib_device *ibdev, u8 port,
1596 u32 *resp_len)
1598 struct hfi1_ibport *ibp = to_iport(ibdev, port);
1599 u8 *p = data;
1600 int i;
1602 if (am) {
1603 smp->status |= IB_SMP_INVALID_FIELD;
1604 return reply((struct ib_mad_hdr *)smp);
1607 for (i = 0; i < ARRAY_SIZE(ibp->sc_to_sl); i++)
1608 ibp->sc_to_sl[i] = *p++;
1610 return __subn_get_opa_sc_to_sl(smp, am, data, ibdev, port, resp_len);
1613 static int __subn_get_opa_sc_to_vlt(struct opa_smp *smp, u32 am, u8 *data,
1614 struct ib_device *ibdev, u8 port,
1615 u32 *resp_len)
1617 u32 n_blocks = OPA_AM_NBLK(am);
1618 struct hfi1_devdata *dd = dd_from_ibdev(ibdev);
1619 void *vp = (void *)data;
1620 size_t size = 4 * sizeof(u64);
1622 if (n_blocks != 1) {
1623 smp->status |= IB_SMP_INVALID_FIELD;
1624 return reply((struct ib_mad_hdr *)smp);
1627 get_sc2vlt_tables(dd, vp);
1629 if (resp_len)
1630 *resp_len += size;
1632 return reply((struct ib_mad_hdr *)smp);
1635 static int __subn_set_opa_sc_to_vlt(struct opa_smp *smp, u32 am, u8 *data,
1636 struct ib_device *ibdev, u8 port,
1637 u32 *resp_len)
1639 u32 n_blocks = OPA_AM_NBLK(am);
1640 int async_update = OPA_AM_ASYNC(am);
1641 struct hfi1_devdata *dd = dd_from_ibdev(ibdev);
1642 void *vp = (void *)data;
1643 struct hfi1_pportdata *ppd;
1644 int lstate;
1646 if (n_blocks != 1 || async_update) {
1647 smp->status |= IB_SMP_INVALID_FIELD;
1648 return reply((struct ib_mad_hdr *)smp);
1651 /* IB numbers ports from 1, hw from 0 */
1652 ppd = dd->pport + (port - 1);
1653 lstate = driver_lstate(ppd);
1655 * it's known that async_update is 0 by this point, but include
1656 * the explicit check for clarity
1658 if (!async_update &&
1659 (lstate == IB_PORT_ARMED || lstate == IB_PORT_ACTIVE)) {
1660 smp->status |= IB_SMP_INVALID_FIELD;
1661 return reply((struct ib_mad_hdr *)smp);
1664 set_sc2vlt_tables(dd, vp);
1666 return __subn_get_opa_sc_to_vlt(smp, am, data, ibdev, port, resp_len);
1669 static int __subn_get_opa_sc_to_vlnt(struct opa_smp *smp, u32 am, u8 *data,
1670 struct ib_device *ibdev, u8 port,
1671 u32 *resp_len)
1673 u32 n_blocks = OPA_AM_NPORT(am);
1674 struct hfi1_devdata *dd = dd_from_ibdev(ibdev);
1675 struct hfi1_pportdata *ppd;
1676 void *vp = (void *)data;
1677 int size;
1679 if (n_blocks != 1) {
1680 smp->status |= IB_SMP_INVALID_FIELD;
1681 return reply((struct ib_mad_hdr *)smp);
1684 ppd = dd->pport + (port - 1);
1686 size = fm_get_table(ppd, FM_TBL_SC2VLNT, vp);
1688 if (resp_len)
1689 *resp_len += size;
1691 return reply((struct ib_mad_hdr *)smp);
1694 static int __subn_set_opa_sc_to_vlnt(struct opa_smp *smp, u32 am, u8 *data,
1695 struct ib_device *ibdev, u8 port,
1696 u32 *resp_len)
1698 u32 n_blocks = OPA_AM_NPORT(am);
1699 struct hfi1_devdata *dd = dd_from_ibdev(ibdev);
1700 struct hfi1_pportdata *ppd;
1701 void *vp = (void *)data;
1702 int lstate;
1704 if (n_blocks != 1) {
1705 smp->status |= IB_SMP_INVALID_FIELD;
1706 return reply((struct ib_mad_hdr *)smp);
1709 /* IB numbers ports from 1, hw from 0 */
1710 ppd = dd->pport + (port - 1);
1711 lstate = driver_lstate(ppd);
1712 if (lstate == IB_PORT_ARMED || lstate == IB_PORT_ACTIVE) {
1713 smp->status |= IB_SMP_INVALID_FIELD;
1714 return reply((struct ib_mad_hdr *)smp);
1717 ppd = dd->pport + (port - 1);
1719 fm_set_table(ppd, FM_TBL_SC2VLNT, vp);
1721 return __subn_get_opa_sc_to_vlnt(smp, am, data, ibdev, port,
1722 resp_len);
1725 static int __subn_get_opa_psi(struct opa_smp *smp, u32 am, u8 *data,
1726 struct ib_device *ibdev, u8 port,
1727 u32 *resp_len)
1729 u32 nports = OPA_AM_NPORT(am);
1730 u32 start_of_sm_config = OPA_AM_START_SM_CFG(am);
1731 u32 lstate;
1732 struct hfi1_ibport *ibp;
1733 struct hfi1_pportdata *ppd;
1734 struct opa_port_state_info *psi = (struct opa_port_state_info *)data;
1736 if (nports != 1) {
1737 smp->status |= IB_SMP_INVALID_FIELD;
1738 return reply((struct ib_mad_hdr *)smp);
1741 ibp = to_iport(ibdev, port);
1742 ppd = ppd_from_ibp(ibp);
1744 lstate = driver_lstate(ppd);
1746 if (start_of_sm_config && (lstate == IB_PORT_INIT))
1747 ppd->is_sm_config_started = 1;
1749 psi->port_states.ledenable_offlinereason = ppd->neighbor_normal << 4;
1750 psi->port_states.ledenable_offlinereason |=
1751 ppd->is_sm_config_started << 5;
1752 psi->port_states.ledenable_offlinereason |=
1753 ppd->offline_disabled_reason;
1755 psi->port_states.portphysstate_portstate =
1756 (hfi1_ibphys_portstate(ppd) << 4) | (lstate & 0xf);
1757 psi->link_width_downgrade_tx_active =
1758 cpu_to_be16(ppd->link_width_downgrade_tx_active);
1759 psi->link_width_downgrade_rx_active =
1760 cpu_to_be16(ppd->link_width_downgrade_rx_active);
1761 if (resp_len)
1762 *resp_len += sizeof(struct opa_port_state_info);
1764 return reply((struct ib_mad_hdr *)smp);
1767 static int __subn_set_opa_psi(struct opa_smp *smp, u32 am, u8 *data,
1768 struct ib_device *ibdev, u8 port,
1769 u32 *resp_len)
1771 u32 nports = OPA_AM_NPORT(am);
1772 u32 start_of_sm_config = OPA_AM_START_SM_CFG(am);
1773 u32 ls_old;
1774 u8 ls_new, ps_new;
1775 struct hfi1_ibport *ibp;
1776 struct hfi1_pportdata *ppd;
1777 struct opa_port_state_info *psi = (struct opa_port_state_info *)data;
1778 int ret, invalid = 0;
1780 if (nports != 1) {
1781 smp->status |= IB_SMP_INVALID_FIELD;
1782 return reply((struct ib_mad_hdr *)smp);
1785 ibp = to_iport(ibdev, port);
1786 ppd = ppd_from_ibp(ibp);
1788 ls_old = driver_lstate(ppd);
1790 ls_new = port_states_to_logical_state(&psi->port_states);
1791 ps_new = port_states_to_phys_state(&psi->port_states);
1793 if (ls_old == IB_PORT_INIT) {
1794 if (start_of_sm_config) {
1795 if (ls_new == ls_old || (ls_new == IB_PORT_ARMED))
1796 ppd->is_sm_config_started = 1;
1797 } else if (ls_new == IB_PORT_ARMED) {
1798 if (ppd->is_sm_config_started == 0)
1799 invalid = 1;
1803 ret = set_port_states(ppd, smp, ls_new, ps_new, invalid);
1804 if (ret)
1805 return ret;
1807 if (invalid)
1808 smp->status |= IB_SMP_INVALID_FIELD;
1810 return __subn_get_opa_psi(smp, am, data, ibdev, port, resp_len);
1813 static int __subn_get_opa_cable_info(struct opa_smp *smp, u32 am, u8 *data,
1814 struct ib_device *ibdev, u8 port,
1815 u32 *resp_len)
1817 struct hfi1_devdata *dd = dd_from_ibdev(ibdev);
1818 u32 addr = OPA_AM_CI_ADDR(am);
1819 u32 len = OPA_AM_CI_LEN(am) + 1;
1820 int ret;
1822 #define __CI_PAGE_SIZE BIT(7) /* 128 bytes */
1823 #define __CI_PAGE_MASK ~(__CI_PAGE_SIZE - 1)
1824 #define __CI_PAGE_NUM(a) ((a) & __CI_PAGE_MASK)
1827 * check that addr is within spec, and
1828 * addr and (addr + len - 1) are on the same "page"
1830 if (addr >= 4096 ||
1831 (__CI_PAGE_NUM(addr) != __CI_PAGE_NUM(addr + len - 1))) {
1832 smp->status |= IB_SMP_INVALID_FIELD;
1833 return reply((struct ib_mad_hdr *)smp);
1836 ret = get_cable_info(dd, port, addr, len, data);
1838 if (ret == -ENODEV) {
1839 smp->status |= IB_SMP_UNSUP_METH_ATTR;
1840 return reply((struct ib_mad_hdr *)smp);
1843 /* The address range for the CableInfo SMA query is wider than the
1844 * memory available on the QSFP cable. We want to return a valid
1845 * response, albeit zeroed out, for address ranges beyond available
1846 * memory but that are within the CableInfo query spec
1848 if (ret < 0 && ret != -ERANGE) {
1849 smp->status |= IB_SMP_INVALID_FIELD;
1850 return reply((struct ib_mad_hdr *)smp);
1853 if (resp_len)
1854 *resp_len += len;
1856 return reply((struct ib_mad_hdr *)smp);
1859 static int __subn_get_opa_bct(struct opa_smp *smp, u32 am, u8 *data,
1860 struct ib_device *ibdev, u8 port, u32 *resp_len)
1862 u32 num_ports = OPA_AM_NPORT(am);
1863 struct hfi1_devdata *dd = dd_from_ibdev(ibdev);
1864 struct hfi1_pportdata *ppd;
1865 struct buffer_control *p = (struct buffer_control *)data;
1866 int size;
1868 if (num_ports != 1) {
1869 smp->status |= IB_SMP_INVALID_FIELD;
1870 return reply((struct ib_mad_hdr *)smp);
1873 ppd = dd->pport + (port - 1);
1874 size = fm_get_table(ppd, FM_TBL_BUFFER_CONTROL, p);
1875 trace_bct_get(dd, p);
1876 if (resp_len)
1877 *resp_len += size;
1879 return reply((struct ib_mad_hdr *)smp);
1882 static int __subn_set_opa_bct(struct opa_smp *smp, u32 am, u8 *data,
1883 struct ib_device *ibdev, u8 port, u32 *resp_len)
1885 u32 num_ports = OPA_AM_NPORT(am);
1886 struct hfi1_devdata *dd = dd_from_ibdev(ibdev);
1887 struct hfi1_pportdata *ppd;
1888 struct buffer_control *p = (struct buffer_control *)data;
1890 if (num_ports != 1) {
1891 smp->status |= IB_SMP_INVALID_FIELD;
1892 return reply((struct ib_mad_hdr *)smp);
1894 ppd = dd->pport + (port - 1);
1895 trace_bct_set(dd, p);
1896 if (fm_set_table(ppd, FM_TBL_BUFFER_CONTROL, p) < 0) {
1897 smp->status |= IB_SMP_INVALID_FIELD;
1898 return reply((struct ib_mad_hdr *)smp);
1901 return __subn_get_opa_bct(smp, am, data, ibdev, port, resp_len);
1904 static int __subn_get_opa_vl_arb(struct opa_smp *smp, u32 am, u8 *data,
1905 struct ib_device *ibdev, u8 port,
1906 u32 *resp_len)
1908 struct hfi1_pportdata *ppd = ppd_from_ibp(to_iport(ibdev, port));
1909 u32 num_ports = OPA_AM_NPORT(am);
1910 u8 section = (am & 0x00ff0000) >> 16;
1911 u8 *p = data;
1912 int size = 0;
1914 if (num_ports != 1) {
1915 smp->status |= IB_SMP_INVALID_FIELD;
1916 return reply((struct ib_mad_hdr *)smp);
1919 switch (section) {
1920 case OPA_VLARB_LOW_ELEMENTS:
1921 size = fm_get_table(ppd, FM_TBL_VL_LOW_ARB, p);
1922 break;
1923 case OPA_VLARB_HIGH_ELEMENTS:
1924 size = fm_get_table(ppd, FM_TBL_VL_HIGH_ARB, p);
1925 break;
1926 case OPA_VLARB_PREEMPT_ELEMENTS:
1927 size = fm_get_table(ppd, FM_TBL_VL_PREEMPT_ELEMS, p);
1928 break;
1929 case OPA_VLARB_PREEMPT_MATRIX:
1930 size = fm_get_table(ppd, FM_TBL_VL_PREEMPT_MATRIX, p);
1931 break;
1932 default:
1933 pr_warn("OPA SubnGet(VL Arb) AM Invalid : 0x%x\n",
1934 be32_to_cpu(smp->attr_mod));
1935 smp->status |= IB_SMP_INVALID_FIELD;
1936 break;
1939 if (size > 0 && resp_len)
1940 *resp_len += size;
1942 return reply((struct ib_mad_hdr *)smp);
1945 static int __subn_set_opa_vl_arb(struct opa_smp *smp, u32 am, u8 *data,
1946 struct ib_device *ibdev, u8 port,
1947 u32 *resp_len)
1949 struct hfi1_pportdata *ppd = ppd_from_ibp(to_iport(ibdev, port));
1950 u32 num_ports = OPA_AM_NPORT(am);
1951 u8 section = (am & 0x00ff0000) >> 16;
1952 u8 *p = data;
1954 if (num_ports != 1) {
1955 smp->status |= IB_SMP_INVALID_FIELD;
1956 return reply((struct ib_mad_hdr *)smp);
1959 switch (section) {
1960 case OPA_VLARB_LOW_ELEMENTS:
1961 (void)fm_set_table(ppd, FM_TBL_VL_LOW_ARB, p);
1962 break;
1963 case OPA_VLARB_HIGH_ELEMENTS:
1964 (void)fm_set_table(ppd, FM_TBL_VL_HIGH_ARB, p);
1965 break;
1967 * neither OPA_VLARB_PREEMPT_ELEMENTS, or OPA_VLARB_PREEMPT_MATRIX
1968 * can be changed from the default values
1970 case OPA_VLARB_PREEMPT_ELEMENTS:
1971 /* FALLTHROUGH */
1972 case OPA_VLARB_PREEMPT_MATRIX:
1973 smp->status |= IB_SMP_UNSUP_METH_ATTR;
1974 break;
1975 default:
1976 pr_warn("OPA SubnSet(VL Arb) AM Invalid : 0x%x\n",
1977 be32_to_cpu(smp->attr_mod));
1978 smp->status |= IB_SMP_INVALID_FIELD;
1979 break;
1982 return __subn_get_opa_vl_arb(smp, am, data, ibdev, port, resp_len);
1985 struct opa_pma_mad {
1986 struct ib_mad_hdr mad_hdr;
1987 u8 data[2024];
1988 } __packed;
1990 struct opa_class_port_info {
1991 u8 base_version;
1992 u8 class_version;
1993 __be16 cap_mask;
1994 __be32 cap_mask2_resp_time;
1996 u8 redirect_gid[16];
1997 __be32 redirect_tc_fl;
1998 __be32 redirect_lid;
1999 __be32 redirect_sl_qp;
2000 __be32 redirect_qkey;
2002 u8 trap_gid[16];
2003 __be32 trap_tc_fl;
2004 __be32 trap_lid;
2005 __be32 trap_hl_qp;
2006 __be32 trap_qkey;
2008 __be16 trap_pkey;
2009 __be16 redirect_pkey;
2011 u8 trap_sl_rsvd;
2012 u8 reserved[3];
2013 } __packed;
2015 struct opa_port_status_req {
2016 __u8 port_num;
2017 __u8 reserved[3];
2018 __be32 vl_select_mask;
2021 #define VL_MASK_ALL 0x000080ff
2023 struct opa_port_status_rsp {
2024 __u8 port_num;
2025 __u8 reserved[3];
2026 __be32 vl_select_mask;
2028 /* Data counters */
2029 __be64 port_xmit_data;
2030 __be64 port_rcv_data;
2031 __be64 port_xmit_pkts;
2032 __be64 port_rcv_pkts;
2033 __be64 port_multicast_xmit_pkts;
2034 __be64 port_multicast_rcv_pkts;
2035 __be64 port_xmit_wait;
2036 __be64 sw_port_congestion;
2037 __be64 port_rcv_fecn;
2038 __be64 port_rcv_becn;
2039 __be64 port_xmit_time_cong;
2040 __be64 port_xmit_wasted_bw;
2041 __be64 port_xmit_wait_data;
2042 __be64 port_rcv_bubble;
2043 __be64 port_mark_fecn;
2044 /* Error counters */
2045 __be64 port_rcv_constraint_errors;
2046 __be64 port_rcv_switch_relay_errors;
2047 __be64 port_xmit_discards;
2048 __be64 port_xmit_constraint_errors;
2049 __be64 port_rcv_remote_physical_errors;
2050 __be64 local_link_integrity_errors;
2051 __be64 port_rcv_errors;
2052 __be64 excessive_buffer_overruns;
2053 __be64 fm_config_errors;
2054 __be32 link_error_recovery;
2055 __be32 link_downed;
2056 u8 uncorrectable_errors;
2058 u8 link_quality_indicator; /* 5res, 3bit */
2059 u8 res2[6];
2060 struct _vls_pctrs {
2061 /* per-VL Data counters */
2062 __be64 port_vl_xmit_data;
2063 __be64 port_vl_rcv_data;
2064 __be64 port_vl_xmit_pkts;
2065 __be64 port_vl_rcv_pkts;
2066 __be64 port_vl_xmit_wait;
2067 __be64 sw_port_vl_congestion;
2068 __be64 port_vl_rcv_fecn;
2069 __be64 port_vl_rcv_becn;
2070 __be64 port_xmit_time_cong;
2071 __be64 port_vl_xmit_wasted_bw;
2072 __be64 port_vl_xmit_wait_data;
2073 __be64 port_vl_rcv_bubble;
2074 __be64 port_vl_mark_fecn;
2075 __be64 port_vl_xmit_discards;
2076 } vls[0]; /* real array size defined by # bits set in vl_select_mask */
2079 enum counter_selects {
2080 CS_PORT_XMIT_DATA = (1 << 31),
2081 CS_PORT_RCV_DATA = (1 << 30),
2082 CS_PORT_XMIT_PKTS = (1 << 29),
2083 CS_PORT_RCV_PKTS = (1 << 28),
2084 CS_PORT_MCAST_XMIT_PKTS = (1 << 27),
2085 CS_PORT_MCAST_RCV_PKTS = (1 << 26),
2086 CS_PORT_XMIT_WAIT = (1 << 25),
2087 CS_SW_PORT_CONGESTION = (1 << 24),
2088 CS_PORT_RCV_FECN = (1 << 23),
2089 CS_PORT_RCV_BECN = (1 << 22),
2090 CS_PORT_XMIT_TIME_CONG = (1 << 21),
2091 CS_PORT_XMIT_WASTED_BW = (1 << 20),
2092 CS_PORT_XMIT_WAIT_DATA = (1 << 19),
2093 CS_PORT_RCV_BUBBLE = (1 << 18),
2094 CS_PORT_MARK_FECN = (1 << 17),
2095 CS_PORT_RCV_CONSTRAINT_ERRORS = (1 << 16),
2096 CS_PORT_RCV_SWITCH_RELAY_ERRORS = (1 << 15),
2097 CS_PORT_XMIT_DISCARDS = (1 << 14),
2098 CS_PORT_XMIT_CONSTRAINT_ERRORS = (1 << 13),
2099 CS_PORT_RCV_REMOTE_PHYSICAL_ERRORS = (1 << 12),
2100 CS_LOCAL_LINK_INTEGRITY_ERRORS = (1 << 11),
2101 CS_PORT_RCV_ERRORS = (1 << 10),
2102 CS_EXCESSIVE_BUFFER_OVERRUNS = (1 << 9),
2103 CS_FM_CONFIG_ERRORS = (1 << 8),
2104 CS_LINK_ERROR_RECOVERY = (1 << 7),
2105 CS_LINK_DOWNED = (1 << 6),
2106 CS_UNCORRECTABLE_ERRORS = (1 << 5),
2109 struct opa_clear_port_status {
2110 __be64 port_select_mask[4];
2111 __be32 counter_select_mask;
2114 struct opa_aggregate {
2115 __be16 attr_id;
2116 __be16 err_reqlength; /* 1 bit, 8 res, 7 bit */
2117 __be32 attr_mod;
2118 u8 data[0];
2121 #define MSK_LLI 0x000000f0
2122 #define MSK_LLI_SFT 4
2123 #define MSK_LER 0x0000000f
2124 #define MSK_LER_SFT 0
2125 #define ADD_LLI 8
2126 #define ADD_LER 2
2128 /* Request contains first three fields, response contains those plus the rest */
2129 struct opa_port_data_counters_msg {
2130 __be64 port_select_mask[4];
2131 __be32 vl_select_mask;
2132 __be32 resolution;
2134 /* Response fields follow */
2135 struct _port_dctrs {
2136 u8 port_number;
2137 u8 reserved2[3];
2138 __be32 link_quality_indicator; /* 29res, 3bit */
2140 /* Data counters */
2141 __be64 port_xmit_data;
2142 __be64 port_rcv_data;
2143 __be64 port_xmit_pkts;
2144 __be64 port_rcv_pkts;
2145 __be64 port_multicast_xmit_pkts;
2146 __be64 port_multicast_rcv_pkts;
2147 __be64 port_xmit_wait;
2148 __be64 sw_port_congestion;
2149 __be64 port_rcv_fecn;
2150 __be64 port_rcv_becn;
2151 __be64 port_xmit_time_cong;
2152 __be64 port_xmit_wasted_bw;
2153 __be64 port_xmit_wait_data;
2154 __be64 port_rcv_bubble;
2155 __be64 port_mark_fecn;
2157 __be64 port_error_counter_summary;
2158 /* Sum of error counts/port */
2160 struct _vls_dctrs {
2161 /* per-VL Data counters */
2162 __be64 port_vl_xmit_data;
2163 __be64 port_vl_rcv_data;
2164 __be64 port_vl_xmit_pkts;
2165 __be64 port_vl_rcv_pkts;
2166 __be64 port_vl_xmit_wait;
2167 __be64 sw_port_vl_congestion;
2168 __be64 port_vl_rcv_fecn;
2169 __be64 port_vl_rcv_becn;
2170 __be64 port_xmit_time_cong;
2171 __be64 port_vl_xmit_wasted_bw;
2172 __be64 port_vl_xmit_wait_data;
2173 __be64 port_vl_rcv_bubble;
2174 __be64 port_vl_mark_fecn;
2175 } vls[0];
2176 /* array size defined by #bits set in vl_select_mask*/
2177 } port[1]; /* array size defined by #ports in attribute modifier */
2180 struct opa_port_error_counters64_msg {
2182 * Request contains first two fields, response contains the
2183 * whole magilla
2185 __be64 port_select_mask[4];
2186 __be32 vl_select_mask;
2188 /* Response-only fields follow */
2189 __be32 reserved1;
2190 struct _port_ectrs {
2191 u8 port_number;
2192 u8 reserved2[7];
2193 __be64 port_rcv_constraint_errors;
2194 __be64 port_rcv_switch_relay_errors;
2195 __be64 port_xmit_discards;
2196 __be64 port_xmit_constraint_errors;
2197 __be64 port_rcv_remote_physical_errors;
2198 __be64 local_link_integrity_errors;
2199 __be64 port_rcv_errors;
2200 __be64 excessive_buffer_overruns;
2201 __be64 fm_config_errors;
2202 __be32 link_error_recovery;
2203 __be32 link_downed;
2204 u8 uncorrectable_errors;
2205 u8 reserved3[7];
2206 struct _vls_ectrs {
2207 __be64 port_vl_xmit_discards;
2208 } vls[0];
2209 /* array size defined by #bits set in vl_select_mask */
2210 } port[1]; /* array size defined by #ports in attribute modifier */
2213 struct opa_port_error_info_msg {
2214 __be64 port_select_mask[4];
2215 __be32 error_info_select_mask;
2216 __be32 reserved1;
2217 struct _port_ei {
2218 u8 port_number;
2219 u8 reserved2[7];
2221 /* PortRcvErrorInfo */
2222 struct {
2223 u8 status_and_code;
2224 union {
2225 u8 raw[17];
2226 struct {
2227 /* EI1to12 format */
2228 u8 packet_flit1[8];
2229 u8 packet_flit2[8];
2230 u8 remaining_flit_bits12;
2231 } ei1to12;
2232 struct {
2233 u8 packet_bytes[8];
2234 u8 remaining_flit_bits;
2235 } ei13;
2236 } ei;
2237 u8 reserved3[6];
2238 } __packed port_rcv_ei;
2240 /* ExcessiveBufferOverrunInfo */
2241 struct {
2242 u8 status_and_sc;
2243 u8 reserved4[7];
2244 } __packed excessive_buffer_overrun_ei;
2246 /* PortXmitConstraintErrorInfo */
2247 struct {
2248 u8 status;
2249 u8 reserved5;
2250 __be16 pkey;
2251 __be32 slid;
2252 } __packed port_xmit_constraint_ei;
2254 /* PortRcvConstraintErrorInfo */
2255 struct {
2256 u8 status;
2257 u8 reserved6;
2258 __be16 pkey;
2259 __be32 slid;
2260 } __packed port_rcv_constraint_ei;
2262 /* PortRcvSwitchRelayErrorInfo */
2263 struct {
2264 u8 status_and_code;
2265 u8 reserved7[3];
2266 __u32 error_info;
2267 } __packed port_rcv_switch_relay_ei;
2269 /* UncorrectableErrorInfo */
2270 struct {
2271 u8 status_and_code;
2272 u8 reserved8;
2273 } __packed uncorrectable_ei;
2275 /* FMConfigErrorInfo */
2276 struct {
2277 u8 status_and_code;
2278 u8 error_info;
2279 } __packed fm_config_ei;
2280 __u32 reserved9;
2281 } port[1]; /* actual array size defined by #ports in attr modifier */
2284 /* opa_port_error_info_msg error_info_select_mask bit definitions */
2285 enum error_info_selects {
2286 ES_PORT_RCV_ERROR_INFO = (1 << 31),
2287 ES_EXCESSIVE_BUFFER_OVERRUN_INFO = (1 << 30),
2288 ES_PORT_XMIT_CONSTRAINT_ERROR_INFO = (1 << 29),
2289 ES_PORT_RCV_CONSTRAINT_ERROR_INFO = (1 << 28),
2290 ES_PORT_RCV_SWITCH_RELAY_ERROR_INFO = (1 << 27),
2291 ES_UNCORRECTABLE_ERROR_INFO = (1 << 26),
2292 ES_FM_CONFIG_ERROR_INFO = (1 << 25)
2295 static int pma_get_opa_classportinfo(struct opa_pma_mad *pmp,
2296 struct ib_device *ibdev, u32 *resp_len)
2298 struct opa_class_port_info *p =
2299 (struct opa_class_port_info *)pmp->data;
2301 memset(pmp->data, 0, sizeof(pmp->data));
2303 if (pmp->mad_hdr.attr_mod != 0)
2304 pmp->mad_hdr.status |= IB_SMP_INVALID_FIELD;
2306 p->base_version = OPA_MGMT_BASE_VERSION;
2307 p->class_version = OPA_SMI_CLASS_VERSION;
2309 * Expected response time is 4.096 usec. * 2^18 == 1.073741824 sec.
2311 p->cap_mask2_resp_time = cpu_to_be32(18);
2313 if (resp_len)
2314 *resp_len += sizeof(*p);
2316 return reply((struct ib_mad_hdr *)pmp);
2319 static void a0_portstatus(struct hfi1_pportdata *ppd,
2320 struct opa_port_status_rsp *rsp, u32 vl_select_mask)
2322 if (!is_bx(ppd->dd)) {
2323 unsigned long vl;
2324 u64 sum_vl_xmit_wait = 0;
2325 u32 vl_all_mask = VL_MASK_ALL;
2327 for_each_set_bit(vl, (unsigned long *)&(vl_all_mask),
2328 8 * sizeof(vl_all_mask)) {
2329 u64 tmp = sum_vl_xmit_wait +
2330 read_port_cntr(ppd, C_TX_WAIT_VL,
2331 idx_from_vl(vl));
2332 if (tmp < sum_vl_xmit_wait) {
2333 /* we wrapped */
2334 sum_vl_xmit_wait = (u64)~0;
2335 break;
2337 sum_vl_xmit_wait = tmp;
2339 if (be64_to_cpu(rsp->port_xmit_wait) > sum_vl_xmit_wait)
2340 rsp->port_xmit_wait = cpu_to_be64(sum_vl_xmit_wait);
2344 static int pma_get_opa_portstatus(struct opa_pma_mad *pmp,
2345 struct ib_device *ibdev,
2346 u8 port, u32 *resp_len)
2348 struct opa_port_status_req *req =
2349 (struct opa_port_status_req *)pmp->data;
2350 struct hfi1_devdata *dd = dd_from_ibdev(ibdev);
2351 struct opa_port_status_rsp *rsp;
2352 u32 vl_select_mask = be32_to_cpu(req->vl_select_mask);
2353 unsigned long vl;
2354 size_t response_data_size;
2355 u32 nports = be32_to_cpu(pmp->mad_hdr.attr_mod) >> 24;
2356 u8 port_num = req->port_num;
2357 u8 num_vls = hweight32(vl_select_mask);
2358 struct _vls_pctrs *vlinfo;
2359 struct hfi1_ibport *ibp = to_iport(ibdev, port);
2360 struct hfi1_pportdata *ppd = ppd_from_ibp(ibp);
2361 int vfi;
2362 u64 tmp, tmp2;
2364 response_data_size = sizeof(struct opa_port_status_rsp) +
2365 num_vls * sizeof(struct _vls_pctrs);
2366 if (response_data_size > sizeof(pmp->data)) {
2367 pmp->mad_hdr.status |= OPA_PM_STATUS_REQUEST_TOO_LARGE;
2368 return reply((struct ib_mad_hdr *)pmp);
2371 if (nports != 1 || (port_num && port_num != port) ||
2372 num_vls > OPA_MAX_VLS || (vl_select_mask & ~VL_MASK_ALL)) {
2373 pmp->mad_hdr.status |= IB_SMP_INVALID_FIELD;
2374 return reply((struct ib_mad_hdr *)pmp);
2377 memset(pmp->data, 0, sizeof(pmp->data));
2379 rsp = (struct opa_port_status_rsp *)pmp->data;
2380 if (port_num)
2381 rsp->port_num = port_num;
2382 else
2383 rsp->port_num = port;
2385 rsp->port_rcv_constraint_errors =
2386 cpu_to_be64(read_port_cntr(ppd, C_SW_RCV_CSTR_ERR,
2387 CNTR_INVALID_VL));
2389 hfi1_read_link_quality(dd, &rsp->link_quality_indicator);
2391 rsp->vl_select_mask = cpu_to_be32(vl_select_mask);
2392 rsp->port_xmit_data = cpu_to_be64(read_dev_cntr(dd, C_DC_XMIT_FLITS,
2393 CNTR_INVALID_VL));
2394 rsp->port_rcv_data = cpu_to_be64(read_dev_cntr(dd, C_DC_RCV_FLITS,
2395 CNTR_INVALID_VL));
2396 rsp->port_xmit_pkts = cpu_to_be64(read_dev_cntr(dd, C_DC_XMIT_PKTS,
2397 CNTR_INVALID_VL));
2398 rsp->port_rcv_pkts = cpu_to_be64(read_dev_cntr(dd, C_DC_RCV_PKTS,
2399 CNTR_INVALID_VL));
2400 rsp->port_multicast_xmit_pkts =
2401 cpu_to_be64(read_dev_cntr(dd, C_DC_MC_XMIT_PKTS,
2402 CNTR_INVALID_VL));
2403 rsp->port_multicast_rcv_pkts =
2404 cpu_to_be64(read_dev_cntr(dd, C_DC_MC_RCV_PKTS,
2405 CNTR_INVALID_VL));
2406 rsp->port_xmit_wait =
2407 cpu_to_be64(read_port_cntr(ppd, C_TX_WAIT, CNTR_INVALID_VL));
2408 rsp->port_rcv_fecn =
2409 cpu_to_be64(read_dev_cntr(dd, C_DC_RCV_FCN, CNTR_INVALID_VL));
2410 rsp->port_rcv_becn =
2411 cpu_to_be64(read_dev_cntr(dd, C_DC_RCV_BCN, CNTR_INVALID_VL));
2412 rsp->port_xmit_discards =
2413 cpu_to_be64(read_port_cntr(ppd, C_SW_XMIT_DSCD,
2414 CNTR_INVALID_VL));
2415 rsp->port_xmit_constraint_errors =
2416 cpu_to_be64(read_port_cntr(ppd, C_SW_XMIT_CSTR_ERR,
2417 CNTR_INVALID_VL));
2418 rsp->port_rcv_remote_physical_errors =
2419 cpu_to_be64(read_dev_cntr(dd, C_DC_RMT_PHY_ERR,
2420 CNTR_INVALID_VL));
2421 rsp->local_link_integrity_errors =
2422 cpu_to_be64(read_dev_cntr(dd, C_DC_RX_REPLAY,
2423 CNTR_INVALID_VL));
2424 tmp = read_dev_cntr(dd, C_DC_SEQ_CRC_CNT, CNTR_INVALID_VL);
2425 tmp2 = tmp + read_dev_cntr(dd, C_DC_REINIT_FROM_PEER_CNT,
2426 CNTR_INVALID_VL);
2427 if (tmp2 > (u32)UINT_MAX || tmp2 < tmp) {
2428 /* overflow/wrapped */
2429 rsp->link_error_recovery = cpu_to_be32(~0);
2430 } else {
2431 rsp->link_error_recovery = cpu_to_be32(tmp2);
2433 rsp->port_rcv_errors =
2434 cpu_to_be64(read_dev_cntr(dd, C_DC_RCV_ERR, CNTR_INVALID_VL));
2435 rsp->excessive_buffer_overruns =
2436 cpu_to_be64(read_dev_cntr(dd, C_RCV_OVF, CNTR_INVALID_VL));
2437 rsp->fm_config_errors =
2438 cpu_to_be64(read_dev_cntr(dd, C_DC_FM_CFG_ERR,
2439 CNTR_INVALID_VL));
2440 rsp->link_downed = cpu_to_be32(read_port_cntr(ppd, C_SW_LINK_DOWN,
2441 CNTR_INVALID_VL));
2443 /* rsp->uncorrectable_errors is 8 bits wide, and it pegs at 0xff */
2444 tmp = read_dev_cntr(dd, C_DC_UNC_ERR, CNTR_INVALID_VL);
2445 rsp->uncorrectable_errors = tmp < 0x100 ? (tmp & 0xff) : 0xff;
2447 vlinfo = &rsp->vls[0];
2448 vfi = 0;
2449 /* The vl_select_mask has been checked above, and we know
2450 * that it contains only entries which represent valid VLs.
2451 * So in the for_each_set_bit() loop below, we don't need
2452 * any additional checks for vl.
2454 for_each_set_bit(vl, (unsigned long *)&(vl_select_mask),
2455 8 * sizeof(vl_select_mask)) {
2456 memset(vlinfo, 0, sizeof(*vlinfo));
2458 tmp = read_dev_cntr(dd, C_DC_RX_FLIT_VL, idx_from_vl(vl));
2459 rsp->vls[vfi].port_vl_rcv_data = cpu_to_be64(tmp);
2461 rsp->vls[vfi].port_vl_rcv_pkts =
2462 cpu_to_be64(read_dev_cntr(dd, C_DC_RX_PKT_VL,
2463 idx_from_vl(vl)));
2465 rsp->vls[vfi].port_vl_xmit_data =
2466 cpu_to_be64(read_port_cntr(ppd, C_TX_FLIT_VL,
2467 idx_from_vl(vl)));
2469 rsp->vls[vfi].port_vl_xmit_pkts =
2470 cpu_to_be64(read_port_cntr(ppd, C_TX_PKT_VL,
2471 idx_from_vl(vl)));
2473 rsp->vls[vfi].port_vl_xmit_wait =
2474 cpu_to_be64(read_port_cntr(ppd, C_TX_WAIT_VL,
2475 idx_from_vl(vl)));
2477 rsp->vls[vfi].port_vl_rcv_fecn =
2478 cpu_to_be64(read_dev_cntr(dd, C_DC_RCV_FCN_VL,
2479 idx_from_vl(vl)));
2481 rsp->vls[vfi].port_vl_rcv_becn =
2482 cpu_to_be64(read_dev_cntr(dd, C_DC_RCV_BCN_VL,
2483 idx_from_vl(vl)));
2485 rsp->vls[vfi].port_vl_xmit_discards =
2486 cpu_to_be64(read_port_cntr(ppd, C_SW_XMIT_DSCD_VL,
2487 idx_from_vl(vl)));
2488 vlinfo++;
2489 vfi++;
2492 a0_portstatus(ppd, rsp, vl_select_mask);
2494 if (resp_len)
2495 *resp_len += response_data_size;
2497 return reply((struct ib_mad_hdr *)pmp);
2500 static u64 get_error_counter_summary(struct ib_device *ibdev, u8 port,
2501 u8 res_lli, u8 res_ler)
2503 struct hfi1_devdata *dd = dd_from_ibdev(ibdev);
2504 struct hfi1_ibport *ibp = to_iport(ibdev, port);
2505 struct hfi1_pportdata *ppd = ppd_from_ibp(ibp);
2506 u64 error_counter_summary = 0, tmp;
2508 error_counter_summary += read_port_cntr(ppd, C_SW_RCV_CSTR_ERR,
2509 CNTR_INVALID_VL);
2510 /* port_rcv_switch_relay_errors is 0 for HFIs */
2511 error_counter_summary += read_port_cntr(ppd, C_SW_XMIT_DSCD,
2512 CNTR_INVALID_VL);
2513 error_counter_summary += read_port_cntr(ppd, C_SW_XMIT_CSTR_ERR,
2514 CNTR_INVALID_VL);
2515 error_counter_summary += read_dev_cntr(dd, C_DC_RMT_PHY_ERR,
2516 CNTR_INVALID_VL);
2517 /* local link integrity must be right-shifted by the lli resolution */
2518 error_counter_summary += (read_dev_cntr(dd, C_DC_RX_REPLAY,
2519 CNTR_INVALID_VL) >> res_lli);
2520 /* link error recovery must b right-shifted by the ler resolution */
2521 tmp = read_dev_cntr(dd, C_DC_SEQ_CRC_CNT, CNTR_INVALID_VL);
2522 tmp += read_dev_cntr(dd, C_DC_REINIT_FROM_PEER_CNT, CNTR_INVALID_VL);
2523 error_counter_summary += (tmp >> res_ler);
2524 error_counter_summary += read_dev_cntr(dd, C_DC_RCV_ERR,
2525 CNTR_INVALID_VL);
2526 error_counter_summary += read_dev_cntr(dd, C_RCV_OVF, CNTR_INVALID_VL);
2527 error_counter_summary += read_dev_cntr(dd, C_DC_FM_CFG_ERR,
2528 CNTR_INVALID_VL);
2529 /* ppd->link_downed is a 32-bit value */
2530 error_counter_summary += read_port_cntr(ppd, C_SW_LINK_DOWN,
2531 CNTR_INVALID_VL);
2532 tmp = read_dev_cntr(dd, C_DC_UNC_ERR, CNTR_INVALID_VL);
2533 /* this is an 8-bit quantity */
2534 error_counter_summary += tmp < 0x100 ? (tmp & 0xff) : 0xff;
2536 return error_counter_summary;
2539 static void a0_datacounters(struct hfi1_pportdata *ppd, struct _port_dctrs *rsp,
2540 u32 vl_select_mask)
2542 if (!is_bx(ppd->dd)) {
2543 unsigned long vl;
2544 u64 sum_vl_xmit_wait = 0;
2545 u32 vl_all_mask = VL_MASK_ALL;
2547 for_each_set_bit(vl, (unsigned long *)&(vl_all_mask),
2548 8 * sizeof(vl_all_mask)) {
2549 u64 tmp = sum_vl_xmit_wait +
2550 read_port_cntr(ppd, C_TX_WAIT_VL,
2551 idx_from_vl(vl));
2552 if (tmp < sum_vl_xmit_wait) {
2553 /* we wrapped */
2554 sum_vl_xmit_wait = (u64)~0;
2555 break;
2557 sum_vl_xmit_wait = tmp;
2559 if (be64_to_cpu(rsp->port_xmit_wait) > sum_vl_xmit_wait)
2560 rsp->port_xmit_wait = cpu_to_be64(sum_vl_xmit_wait);
2564 static void pma_get_opa_port_dctrs(struct ib_device *ibdev,
2565 struct _port_dctrs *rsp)
2567 struct hfi1_devdata *dd = dd_from_ibdev(ibdev);
2569 rsp->port_xmit_data = cpu_to_be64(read_dev_cntr(dd, C_DC_XMIT_FLITS,
2570 CNTR_INVALID_VL));
2571 rsp->port_rcv_data = cpu_to_be64(read_dev_cntr(dd, C_DC_RCV_FLITS,
2572 CNTR_INVALID_VL));
2573 rsp->port_xmit_pkts = cpu_to_be64(read_dev_cntr(dd, C_DC_XMIT_PKTS,
2574 CNTR_INVALID_VL));
2575 rsp->port_rcv_pkts = cpu_to_be64(read_dev_cntr(dd, C_DC_RCV_PKTS,
2576 CNTR_INVALID_VL));
2577 rsp->port_multicast_xmit_pkts =
2578 cpu_to_be64(read_dev_cntr(dd, C_DC_MC_XMIT_PKTS,
2579 CNTR_INVALID_VL));
2580 rsp->port_multicast_rcv_pkts =
2581 cpu_to_be64(read_dev_cntr(dd, C_DC_MC_RCV_PKTS,
2582 CNTR_INVALID_VL));
2585 static int pma_get_opa_datacounters(struct opa_pma_mad *pmp,
2586 struct ib_device *ibdev,
2587 u8 port, u32 *resp_len)
2589 struct opa_port_data_counters_msg *req =
2590 (struct opa_port_data_counters_msg *)pmp->data;
2591 struct hfi1_devdata *dd = dd_from_ibdev(ibdev);
2592 struct hfi1_ibport *ibp = to_iport(ibdev, port);
2593 struct hfi1_pportdata *ppd = ppd_from_ibp(ibp);
2594 struct _port_dctrs *rsp;
2595 struct _vls_dctrs *vlinfo;
2596 size_t response_data_size;
2597 u32 num_ports;
2598 u8 num_pslm;
2599 u8 lq, num_vls;
2600 u8 res_lli, res_ler;
2601 u64 port_mask;
2602 unsigned long port_num;
2603 unsigned long vl;
2604 u32 vl_select_mask;
2605 int vfi;
2607 num_ports = be32_to_cpu(pmp->mad_hdr.attr_mod) >> 24;
2608 num_pslm = hweight64(be64_to_cpu(req->port_select_mask[3]));
2609 num_vls = hweight32(be32_to_cpu(req->vl_select_mask));
2610 vl_select_mask = be32_to_cpu(req->vl_select_mask);
2611 res_lli = (u8)(be32_to_cpu(req->resolution) & MSK_LLI) >> MSK_LLI_SFT;
2612 res_lli = res_lli ? res_lli + ADD_LLI : 0;
2613 res_ler = (u8)(be32_to_cpu(req->resolution) & MSK_LER) >> MSK_LER_SFT;
2614 res_ler = res_ler ? res_ler + ADD_LER : 0;
2616 if (num_ports != 1 || (vl_select_mask & ~VL_MASK_ALL)) {
2617 pmp->mad_hdr.status |= IB_SMP_INVALID_FIELD;
2618 return reply((struct ib_mad_hdr *)pmp);
2621 /* Sanity check */
2622 response_data_size = sizeof(struct opa_port_data_counters_msg) +
2623 num_vls * sizeof(struct _vls_dctrs);
2625 if (response_data_size > sizeof(pmp->data)) {
2626 pmp->mad_hdr.status |= IB_SMP_INVALID_FIELD;
2627 return reply((struct ib_mad_hdr *)pmp);
2631 * The bit set in the mask needs to be consistent with the
2632 * port the request came in on.
2634 port_mask = be64_to_cpu(req->port_select_mask[3]);
2635 port_num = find_first_bit((unsigned long *)&port_mask,
2636 sizeof(port_mask));
2638 if ((u8)port_num != port) {
2639 pmp->mad_hdr.status |= IB_SMP_INVALID_FIELD;
2640 return reply((struct ib_mad_hdr *)pmp);
2643 rsp = &req->port[0];
2644 memset(rsp, 0, sizeof(*rsp));
2646 rsp->port_number = port;
2648 * Note that link_quality_indicator is a 32 bit quantity in
2649 * 'datacounters' queries (as opposed to 'portinfo' queries,
2650 * where it's a byte).
2652 hfi1_read_link_quality(dd, &lq);
2653 rsp->link_quality_indicator = cpu_to_be32((u32)lq);
2654 pma_get_opa_port_dctrs(ibdev, rsp);
2656 rsp->port_xmit_wait =
2657 cpu_to_be64(read_port_cntr(ppd, C_TX_WAIT, CNTR_INVALID_VL));
2658 rsp->port_rcv_fecn =
2659 cpu_to_be64(read_dev_cntr(dd, C_DC_RCV_FCN, CNTR_INVALID_VL));
2660 rsp->port_rcv_becn =
2661 cpu_to_be64(read_dev_cntr(dd, C_DC_RCV_BCN, CNTR_INVALID_VL));
2662 rsp->port_error_counter_summary =
2663 cpu_to_be64(get_error_counter_summary(ibdev, port,
2664 res_lli, res_ler));
2666 vlinfo = &rsp->vls[0];
2667 vfi = 0;
2668 /* The vl_select_mask has been checked above, and we know
2669 * that it contains only entries which represent valid VLs.
2670 * So in the for_each_set_bit() loop below, we don't need
2671 * any additional checks for vl.
2673 for_each_set_bit(vl, (unsigned long *)&(vl_select_mask),
2674 8 * sizeof(req->vl_select_mask)) {
2675 memset(vlinfo, 0, sizeof(*vlinfo));
2677 rsp->vls[vfi].port_vl_xmit_data =
2678 cpu_to_be64(read_port_cntr(ppd, C_TX_FLIT_VL,
2679 idx_from_vl(vl)));
2681 rsp->vls[vfi].port_vl_rcv_data =
2682 cpu_to_be64(read_dev_cntr(dd, C_DC_RX_FLIT_VL,
2683 idx_from_vl(vl)));
2685 rsp->vls[vfi].port_vl_xmit_pkts =
2686 cpu_to_be64(read_port_cntr(ppd, C_TX_PKT_VL,
2687 idx_from_vl(vl)));
2689 rsp->vls[vfi].port_vl_rcv_pkts =
2690 cpu_to_be64(read_dev_cntr(dd, C_DC_RX_PKT_VL,
2691 idx_from_vl(vl)));
2693 rsp->vls[vfi].port_vl_xmit_wait =
2694 cpu_to_be64(read_port_cntr(ppd, C_TX_WAIT_VL,
2695 idx_from_vl(vl)));
2697 rsp->vls[vfi].port_vl_rcv_fecn =
2698 cpu_to_be64(read_dev_cntr(dd, C_DC_RCV_FCN_VL,
2699 idx_from_vl(vl)));
2700 rsp->vls[vfi].port_vl_rcv_becn =
2701 cpu_to_be64(read_dev_cntr(dd, C_DC_RCV_BCN_VL,
2702 idx_from_vl(vl)));
2704 /* rsp->port_vl_xmit_time_cong is 0 for HFIs */
2705 /* rsp->port_vl_xmit_wasted_bw ??? */
2706 /* port_vl_xmit_wait_data - TXE (table 13-9 HFI spec) ???
2707 * does this differ from rsp->vls[vfi].port_vl_xmit_wait
2709 /*rsp->vls[vfi].port_vl_mark_fecn =
2710 * cpu_to_be64(read_csr(dd, DCC_PRF_PORT_VL_MARK_FECN_CNT
2711 * + offset));
2713 vlinfo++;
2714 vfi++;
2717 a0_datacounters(ppd, rsp, vl_select_mask);
2719 if (resp_len)
2720 *resp_len += response_data_size;
2722 return reply((struct ib_mad_hdr *)pmp);
2725 static int pma_get_ib_portcounters_ext(struct ib_pma_mad *pmp,
2726 struct ib_device *ibdev, u8 port)
2728 struct ib_pma_portcounters_ext *p = (struct ib_pma_portcounters_ext *)
2729 pmp->data;
2730 struct _port_dctrs rsp;
2732 if (pmp->mad_hdr.attr_mod != 0 || p->port_select != port) {
2733 pmp->mad_hdr.status |= IB_SMP_INVALID_FIELD;
2734 goto bail;
2737 memset(&rsp, 0, sizeof(rsp));
2738 pma_get_opa_port_dctrs(ibdev, &rsp);
2740 p->port_xmit_data = rsp.port_xmit_data;
2741 p->port_rcv_data = rsp.port_rcv_data;
2742 p->port_xmit_packets = rsp.port_xmit_pkts;
2743 p->port_rcv_packets = rsp.port_rcv_pkts;
2744 p->port_unicast_xmit_packets = 0;
2745 p->port_unicast_rcv_packets = 0;
2746 p->port_multicast_xmit_packets = rsp.port_multicast_xmit_pkts;
2747 p->port_multicast_rcv_packets = rsp.port_multicast_rcv_pkts;
2749 bail:
2750 return reply((struct ib_mad_hdr *)pmp);
2753 static void pma_get_opa_port_ectrs(struct ib_device *ibdev,
2754 struct _port_ectrs *rsp, u8 port)
2756 u64 tmp, tmp2;
2757 struct hfi1_devdata *dd = dd_from_ibdev(ibdev);
2758 struct hfi1_ibport *ibp = to_iport(ibdev, port);
2759 struct hfi1_pportdata *ppd = ppd_from_ibp(ibp);
2761 tmp = read_dev_cntr(dd, C_DC_SEQ_CRC_CNT, CNTR_INVALID_VL);
2762 tmp2 = tmp + read_dev_cntr(dd, C_DC_REINIT_FROM_PEER_CNT,
2763 CNTR_INVALID_VL);
2764 if (tmp2 > (u32)UINT_MAX || tmp2 < tmp) {
2765 /* overflow/wrapped */
2766 rsp->link_error_recovery = cpu_to_be32(~0);
2767 } else {
2768 rsp->link_error_recovery = cpu_to_be32(tmp2);
2771 rsp->link_downed = cpu_to_be32(read_port_cntr(ppd, C_SW_LINK_DOWN,
2772 CNTR_INVALID_VL));
2773 rsp->port_rcv_errors =
2774 cpu_to_be64(read_dev_cntr(dd, C_DC_RCV_ERR, CNTR_INVALID_VL));
2775 rsp->port_rcv_remote_physical_errors =
2776 cpu_to_be64(read_dev_cntr(dd, C_DC_RMT_PHY_ERR,
2777 CNTR_INVALID_VL));
2778 rsp->port_rcv_switch_relay_errors = 0;
2779 rsp->port_xmit_discards =
2780 cpu_to_be64(read_port_cntr(ppd, C_SW_XMIT_DSCD,
2781 CNTR_INVALID_VL));
2782 rsp->port_xmit_constraint_errors =
2783 cpu_to_be64(read_port_cntr(ppd, C_SW_XMIT_CSTR_ERR,
2784 CNTR_INVALID_VL));
2785 rsp->port_rcv_constraint_errors =
2786 cpu_to_be64(read_port_cntr(ppd, C_SW_RCV_CSTR_ERR,
2787 CNTR_INVALID_VL));
2788 rsp->local_link_integrity_errors =
2789 cpu_to_be64(read_dev_cntr(dd, C_DC_RX_REPLAY,
2790 CNTR_INVALID_VL));
2791 rsp->excessive_buffer_overruns =
2792 cpu_to_be64(read_dev_cntr(dd, C_RCV_OVF, CNTR_INVALID_VL));
2795 static int pma_get_opa_porterrors(struct opa_pma_mad *pmp,
2796 struct ib_device *ibdev,
2797 u8 port, u32 *resp_len)
2799 size_t response_data_size;
2800 struct _port_ectrs *rsp;
2801 u8 port_num;
2802 struct opa_port_error_counters64_msg *req;
2803 struct hfi1_devdata *dd = dd_from_ibdev(ibdev);
2804 u32 num_ports;
2805 u8 num_pslm;
2806 u8 num_vls;
2807 struct hfi1_ibport *ibp;
2808 struct hfi1_pportdata *ppd;
2809 struct _vls_ectrs *vlinfo;
2810 unsigned long vl;
2811 u64 port_mask, tmp;
2812 u32 vl_select_mask;
2813 int vfi;
2815 req = (struct opa_port_error_counters64_msg *)pmp->data;
2817 num_ports = be32_to_cpu(pmp->mad_hdr.attr_mod) >> 24;
2819 num_pslm = hweight64(be64_to_cpu(req->port_select_mask[3]));
2820 num_vls = hweight32(be32_to_cpu(req->vl_select_mask));
2822 if (num_ports != 1 || num_ports != num_pslm) {
2823 pmp->mad_hdr.status |= IB_SMP_INVALID_FIELD;
2824 return reply((struct ib_mad_hdr *)pmp);
2827 response_data_size = sizeof(struct opa_port_error_counters64_msg) +
2828 num_vls * sizeof(struct _vls_ectrs);
2830 if (response_data_size > sizeof(pmp->data)) {
2831 pmp->mad_hdr.status |= IB_SMP_INVALID_FIELD;
2832 return reply((struct ib_mad_hdr *)pmp);
2835 * The bit set in the mask needs to be consistent with the
2836 * port the request came in on.
2838 port_mask = be64_to_cpu(req->port_select_mask[3]);
2839 port_num = find_first_bit((unsigned long *)&port_mask,
2840 sizeof(port_mask));
2842 if (port_num != port) {
2843 pmp->mad_hdr.status |= IB_SMP_INVALID_FIELD;
2844 return reply((struct ib_mad_hdr *)pmp);
2847 rsp = &req->port[0];
2849 ibp = to_iport(ibdev, port_num);
2850 ppd = ppd_from_ibp(ibp);
2852 memset(rsp, 0, sizeof(*rsp));
2853 rsp->port_number = port_num;
2855 pma_get_opa_port_ectrs(ibdev, rsp, port_num);
2857 rsp->port_rcv_remote_physical_errors =
2858 cpu_to_be64(read_dev_cntr(dd, C_DC_RMT_PHY_ERR,
2859 CNTR_INVALID_VL));
2860 rsp->fm_config_errors =
2861 cpu_to_be64(read_dev_cntr(dd, C_DC_FM_CFG_ERR,
2862 CNTR_INVALID_VL));
2863 tmp = read_dev_cntr(dd, C_DC_UNC_ERR, CNTR_INVALID_VL);
2865 rsp->uncorrectable_errors = tmp < 0x100 ? (tmp & 0xff) : 0xff;
2866 rsp->port_rcv_errors =
2867 cpu_to_be64(read_dev_cntr(dd, C_DC_RCV_ERR, CNTR_INVALID_VL));
2868 vlinfo = &rsp->vls[0];
2869 vfi = 0;
2870 vl_select_mask = be32_to_cpu(req->vl_select_mask);
2871 for_each_set_bit(vl, (unsigned long *)&(vl_select_mask),
2872 8 * sizeof(req->vl_select_mask)) {
2873 memset(vlinfo, 0, sizeof(*vlinfo));
2874 rsp->vls[vfi].port_vl_xmit_discards =
2875 cpu_to_be64(read_port_cntr(ppd, C_SW_XMIT_DSCD_VL,
2876 idx_from_vl(vl)));
2877 vlinfo += 1;
2878 vfi++;
2881 if (resp_len)
2882 *resp_len += response_data_size;
2884 return reply((struct ib_mad_hdr *)pmp);
2887 static int pma_get_ib_portcounters(struct ib_pma_mad *pmp,
2888 struct ib_device *ibdev, u8 port)
2890 struct ib_pma_portcounters *p = (struct ib_pma_portcounters *)
2891 pmp->data;
2892 struct _port_ectrs rsp;
2893 u64 temp_link_overrun_errors;
2894 u64 temp_64;
2895 u32 temp_32;
2897 memset(&rsp, 0, sizeof(rsp));
2898 pma_get_opa_port_ectrs(ibdev, &rsp, port);
2900 if (pmp->mad_hdr.attr_mod != 0 || p->port_select != port) {
2901 pmp->mad_hdr.status |= IB_SMP_INVALID_FIELD;
2902 goto bail;
2905 p->symbol_error_counter = 0; /* N/A for OPA */
2907 temp_32 = be32_to_cpu(rsp.link_error_recovery);
2908 if (temp_32 > 0xFFUL)
2909 p->link_error_recovery_counter = 0xFF;
2910 else
2911 p->link_error_recovery_counter = (u8)temp_32;
2913 temp_32 = be32_to_cpu(rsp.link_downed);
2914 if (temp_32 > 0xFFUL)
2915 p->link_downed_counter = 0xFF;
2916 else
2917 p->link_downed_counter = (u8)temp_32;
2919 temp_64 = be64_to_cpu(rsp.port_rcv_errors);
2920 if (temp_64 > 0xFFFFUL)
2921 p->port_rcv_errors = cpu_to_be16(0xFFFF);
2922 else
2923 p->port_rcv_errors = cpu_to_be16((u16)temp_64);
2925 temp_64 = be64_to_cpu(rsp.port_rcv_remote_physical_errors);
2926 if (temp_64 > 0xFFFFUL)
2927 p->port_rcv_remphys_errors = cpu_to_be16(0xFFFF);
2928 else
2929 p->port_rcv_remphys_errors = cpu_to_be16((u16)temp_64);
2931 temp_64 = be64_to_cpu(rsp.port_rcv_switch_relay_errors);
2932 p->port_rcv_switch_relay_errors = cpu_to_be16((u16)temp_64);
2934 temp_64 = be64_to_cpu(rsp.port_xmit_discards);
2935 if (temp_64 > 0xFFFFUL)
2936 p->port_xmit_discards = cpu_to_be16(0xFFFF);
2937 else
2938 p->port_xmit_discards = cpu_to_be16((u16)temp_64);
2940 temp_64 = be64_to_cpu(rsp.port_xmit_constraint_errors);
2941 if (temp_64 > 0xFFUL)
2942 p->port_xmit_constraint_errors = 0xFF;
2943 else
2944 p->port_xmit_constraint_errors = (u8)temp_64;
2946 temp_64 = be64_to_cpu(rsp.port_rcv_constraint_errors);
2947 if (temp_64 > 0xFFUL)
2948 p->port_rcv_constraint_errors = 0xFFUL;
2949 else
2950 p->port_rcv_constraint_errors = (u8)temp_64;
2952 /* LocalLink: 7:4, BufferOverrun: 3:0 */
2953 temp_64 = be64_to_cpu(rsp.local_link_integrity_errors);
2954 if (temp_64 > 0xFUL)
2955 temp_64 = 0xFUL;
2957 temp_link_overrun_errors = temp_64 << 4;
2959 temp_64 = be64_to_cpu(rsp.excessive_buffer_overruns);
2960 if (temp_64 > 0xFUL)
2961 temp_64 = 0xFUL;
2962 temp_link_overrun_errors |= temp_64;
2964 p->link_overrun_errors = (u8)temp_link_overrun_errors;
2966 p->vl15_dropped = 0; /* N/A for OPA */
2968 bail:
2969 return reply((struct ib_mad_hdr *)pmp);
2972 static int pma_get_opa_errorinfo(struct opa_pma_mad *pmp,
2973 struct ib_device *ibdev,
2974 u8 port, u32 *resp_len)
2976 size_t response_data_size;
2977 struct _port_ei *rsp;
2978 struct opa_port_error_info_msg *req;
2979 struct hfi1_devdata *dd = dd_from_ibdev(ibdev);
2980 u64 port_mask;
2981 u32 num_ports;
2982 u8 port_num;
2983 u8 num_pslm;
2984 u64 reg;
2986 req = (struct opa_port_error_info_msg *)pmp->data;
2987 rsp = &req->port[0];
2989 num_ports = OPA_AM_NPORT(be32_to_cpu(pmp->mad_hdr.attr_mod));
2990 num_pslm = hweight64(be64_to_cpu(req->port_select_mask[3]));
2992 memset(rsp, 0, sizeof(*rsp));
2994 if (num_ports != 1 || num_ports != num_pslm) {
2995 pmp->mad_hdr.status |= IB_SMP_INVALID_FIELD;
2996 return reply((struct ib_mad_hdr *)pmp);
2999 /* Sanity check */
3000 response_data_size = sizeof(struct opa_port_error_info_msg);
3002 if (response_data_size > sizeof(pmp->data)) {
3003 pmp->mad_hdr.status |= IB_SMP_INVALID_FIELD;
3004 return reply((struct ib_mad_hdr *)pmp);
3008 * The bit set in the mask needs to be consistent with the port
3009 * the request came in on.
3011 port_mask = be64_to_cpu(req->port_select_mask[3]);
3012 port_num = find_first_bit((unsigned long *)&port_mask,
3013 sizeof(port_mask));
3015 if (port_num != port) {
3016 pmp->mad_hdr.status |= IB_SMP_INVALID_FIELD;
3017 return reply((struct ib_mad_hdr *)pmp);
3020 /* PortRcvErrorInfo */
3021 rsp->port_rcv_ei.status_and_code =
3022 dd->err_info_rcvport.status_and_code;
3023 memcpy(&rsp->port_rcv_ei.ei.ei1to12.packet_flit1,
3024 &dd->err_info_rcvport.packet_flit1, sizeof(u64));
3025 memcpy(&rsp->port_rcv_ei.ei.ei1to12.packet_flit2,
3026 &dd->err_info_rcvport.packet_flit2, sizeof(u64));
3028 /* ExcessiverBufferOverrunInfo */
3029 reg = read_csr(dd, RCV_ERR_INFO);
3030 if (reg & RCV_ERR_INFO_RCV_EXCESS_BUFFER_OVERRUN_SMASK) {
3032 * if the RcvExcessBufferOverrun bit is set, save SC of
3033 * first pkt that encountered an excess buffer overrun
3035 u8 tmp = (u8)reg;
3037 tmp &= RCV_ERR_INFO_RCV_EXCESS_BUFFER_OVERRUN_SC_SMASK;
3038 tmp <<= 2;
3039 rsp->excessive_buffer_overrun_ei.status_and_sc = tmp;
3040 /* set the status bit */
3041 rsp->excessive_buffer_overrun_ei.status_and_sc |= 0x80;
3044 rsp->port_xmit_constraint_ei.status =
3045 dd->err_info_xmit_constraint.status;
3046 rsp->port_xmit_constraint_ei.pkey =
3047 cpu_to_be16(dd->err_info_xmit_constraint.pkey);
3048 rsp->port_xmit_constraint_ei.slid =
3049 cpu_to_be32(dd->err_info_xmit_constraint.slid);
3051 rsp->port_rcv_constraint_ei.status =
3052 dd->err_info_rcv_constraint.status;
3053 rsp->port_rcv_constraint_ei.pkey =
3054 cpu_to_be16(dd->err_info_rcv_constraint.pkey);
3055 rsp->port_rcv_constraint_ei.slid =
3056 cpu_to_be32(dd->err_info_rcv_constraint.slid);
3058 /* UncorrectableErrorInfo */
3059 rsp->uncorrectable_ei.status_and_code = dd->err_info_uncorrectable;
3061 /* FMConfigErrorInfo */
3062 rsp->fm_config_ei.status_and_code = dd->err_info_fmconfig;
3064 if (resp_len)
3065 *resp_len += response_data_size;
3067 return reply((struct ib_mad_hdr *)pmp);
3070 static int pma_set_opa_portstatus(struct opa_pma_mad *pmp,
3071 struct ib_device *ibdev,
3072 u8 port, u32 *resp_len)
3074 struct opa_clear_port_status *req =
3075 (struct opa_clear_port_status *)pmp->data;
3076 struct hfi1_devdata *dd = dd_from_ibdev(ibdev);
3077 struct hfi1_ibport *ibp = to_iport(ibdev, port);
3078 struct hfi1_pportdata *ppd = ppd_from_ibp(ibp);
3079 u32 nports = be32_to_cpu(pmp->mad_hdr.attr_mod) >> 24;
3080 u64 portn = be64_to_cpu(req->port_select_mask[3]);
3081 u32 counter_select = be32_to_cpu(req->counter_select_mask);
3082 u32 vl_select_mask = VL_MASK_ALL; /* clear all per-vl cnts */
3083 unsigned long vl;
3085 if ((nports != 1) || (portn != 1 << port)) {
3086 pmp->mad_hdr.status |= IB_SMP_INVALID_FIELD;
3087 return reply((struct ib_mad_hdr *)pmp);
3090 * only counters returned by pma_get_opa_portstatus() are
3091 * handled, so when pma_get_opa_portstatus() gets a fix,
3092 * the corresponding change should be made here as well.
3095 if (counter_select & CS_PORT_XMIT_DATA)
3096 write_dev_cntr(dd, C_DC_XMIT_FLITS, CNTR_INVALID_VL, 0);
3098 if (counter_select & CS_PORT_RCV_DATA)
3099 write_dev_cntr(dd, C_DC_RCV_FLITS, CNTR_INVALID_VL, 0);
3101 if (counter_select & CS_PORT_XMIT_PKTS)
3102 write_dev_cntr(dd, C_DC_XMIT_PKTS, CNTR_INVALID_VL, 0);
3104 if (counter_select & CS_PORT_RCV_PKTS)
3105 write_dev_cntr(dd, C_DC_RCV_PKTS, CNTR_INVALID_VL, 0);
3107 if (counter_select & CS_PORT_MCAST_XMIT_PKTS)
3108 write_dev_cntr(dd, C_DC_MC_XMIT_PKTS, CNTR_INVALID_VL, 0);
3110 if (counter_select & CS_PORT_MCAST_RCV_PKTS)
3111 write_dev_cntr(dd, C_DC_MC_RCV_PKTS, CNTR_INVALID_VL, 0);
3113 if (counter_select & CS_PORT_XMIT_WAIT)
3114 write_port_cntr(ppd, C_TX_WAIT, CNTR_INVALID_VL, 0);
3116 /* ignore cs_sw_portCongestion for HFIs */
3118 if (counter_select & CS_PORT_RCV_FECN)
3119 write_dev_cntr(dd, C_DC_RCV_FCN, CNTR_INVALID_VL, 0);
3121 if (counter_select & CS_PORT_RCV_BECN)
3122 write_dev_cntr(dd, C_DC_RCV_BCN, CNTR_INVALID_VL, 0);
3124 /* ignore cs_port_xmit_time_cong for HFIs */
3125 /* ignore cs_port_xmit_wasted_bw for now */
3126 /* ignore cs_port_xmit_wait_data for now */
3127 if (counter_select & CS_PORT_RCV_BUBBLE)
3128 write_dev_cntr(dd, C_DC_RCV_BBL, CNTR_INVALID_VL, 0);
3130 /* Only applicable for switch */
3131 /* if (counter_select & CS_PORT_MARK_FECN)
3132 * write_csr(dd, DCC_PRF_PORT_MARK_FECN_CNT, 0);
3135 if (counter_select & CS_PORT_RCV_CONSTRAINT_ERRORS)
3136 write_port_cntr(ppd, C_SW_RCV_CSTR_ERR, CNTR_INVALID_VL, 0);
3138 /* ignore cs_port_rcv_switch_relay_errors for HFIs */
3139 if (counter_select & CS_PORT_XMIT_DISCARDS)
3140 write_port_cntr(ppd, C_SW_XMIT_DSCD, CNTR_INVALID_VL, 0);
3142 if (counter_select & CS_PORT_XMIT_CONSTRAINT_ERRORS)
3143 write_port_cntr(ppd, C_SW_XMIT_CSTR_ERR, CNTR_INVALID_VL, 0);
3145 if (counter_select & CS_PORT_RCV_REMOTE_PHYSICAL_ERRORS)
3146 write_dev_cntr(dd, C_DC_RMT_PHY_ERR, CNTR_INVALID_VL, 0);
3148 if (counter_select & CS_LOCAL_LINK_INTEGRITY_ERRORS)
3149 write_dev_cntr(dd, C_DC_RX_REPLAY, CNTR_INVALID_VL, 0);
3151 if (counter_select & CS_LINK_ERROR_RECOVERY) {
3152 write_dev_cntr(dd, C_DC_SEQ_CRC_CNT, CNTR_INVALID_VL, 0);
3153 write_dev_cntr(dd, C_DC_REINIT_FROM_PEER_CNT,
3154 CNTR_INVALID_VL, 0);
3157 if (counter_select & CS_PORT_RCV_ERRORS)
3158 write_dev_cntr(dd, C_DC_RCV_ERR, CNTR_INVALID_VL, 0);
3160 if (counter_select & CS_EXCESSIVE_BUFFER_OVERRUNS) {
3161 write_dev_cntr(dd, C_RCV_OVF, CNTR_INVALID_VL, 0);
3162 dd->rcv_ovfl_cnt = 0;
3165 if (counter_select & CS_FM_CONFIG_ERRORS)
3166 write_dev_cntr(dd, C_DC_FM_CFG_ERR, CNTR_INVALID_VL, 0);
3168 if (counter_select & CS_LINK_DOWNED)
3169 write_port_cntr(ppd, C_SW_LINK_DOWN, CNTR_INVALID_VL, 0);
3171 if (counter_select & CS_UNCORRECTABLE_ERRORS)
3172 write_dev_cntr(dd, C_DC_UNC_ERR, CNTR_INVALID_VL, 0);
3174 for_each_set_bit(vl, (unsigned long *)&(vl_select_mask),
3175 8 * sizeof(vl_select_mask)) {
3176 if (counter_select & CS_PORT_XMIT_DATA)
3177 write_port_cntr(ppd, C_TX_FLIT_VL, idx_from_vl(vl), 0);
3179 if (counter_select & CS_PORT_RCV_DATA)
3180 write_dev_cntr(dd, C_DC_RX_FLIT_VL, idx_from_vl(vl), 0);
3182 if (counter_select & CS_PORT_XMIT_PKTS)
3183 write_port_cntr(ppd, C_TX_PKT_VL, idx_from_vl(vl), 0);
3185 if (counter_select & CS_PORT_RCV_PKTS)
3186 write_dev_cntr(dd, C_DC_RX_PKT_VL, idx_from_vl(vl), 0);
3188 if (counter_select & CS_PORT_XMIT_WAIT)
3189 write_port_cntr(ppd, C_TX_WAIT_VL, idx_from_vl(vl), 0);
3191 /* sw_port_vl_congestion is 0 for HFIs */
3192 if (counter_select & CS_PORT_RCV_FECN)
3193 write_dev_cntr(dd, C_DC_RCV_FCN_VL, idx_from_vl(vl), 0);
3195 if (counter_select & CS_PORT_RCV_BECN)
3196 write_dev_cntr(dd, C_DC_RCV_BCN_VL, idx_from_vl(vl), 0);
3198 /* port_vl_xmit_time_cong is 0 for HFIs */
3199 /* port_vl_xmit_wasted_bw ??? */
3200 /* port_vl_xmit_wait_data - TXE (table 13-9 HFI spec) ??? */
3201 if (counter_select & CS_PORT_RCV_BUBBLE)
3202 write_dev_cntr(dd, C_DC_RCV_BBL_VL, idx_from_vl(vl), 0);
3204 /* if (counter_select & CS_PORT_MARK_FECN)
3205 * write_csr(dd, DCC_PRF_PORT_VL_MARK_FECN_CNT + offset, 0);
3207 if (counter_select & C_SW_XMIT_DSCD_VL)
3208 write_port_cntr(ppd, C_SW_XMIT_DSCD_VL,
3209 idx_from_vl(vl), 0);
3212 if (resp_len)
3213 *resp_len += sizeof(*req);
3215 return reply((struct ib_mad_hdr *)pmp);
3218 static int pma_set_opa_errorinfo(struct opa_pma_mad *pmp,
3219 struct ib_device *ibdev,
3220 u8 port, u32 *resp_len)
3222 struct _port_ei *rsp;
3223 struct opa_port_error_info_msg *req;
3224 struct hfi1_devdata *dd = dd_from_ibdev(ibdev);
3225 u64 port_mask;
3226 u32 num_ports;
3227 u8 port_num;
3228 u8 num_pslm;
3229 u32 error_info_select;
3231 req = (struct opa_port_error_info_msg *)pmp->data;
3232 rsp = &req->port[0];
3234 num_ports = OPA_AM_NPORT(be32_to_cpu(pmp->mad_hdr.attr_mod));
3235 num_pslm = hweight64(be64_to_cpu(req->port_select_mask[3]));
3237 memset(rsp, 0, sizeof(*rsp));
3239 if (num_ports != 1 || num_ports != num_pslm) {
3240 pmp->mad_hdr.status |= IB_SMP_INVALID_FIELD;
3241 return reply((struct ib_mad_hdr *)pmp);
3245 * The bit set in the mask needs to be consistent with the port
3246 * the request came in on.
3248 port_mask = be64_to_cpu(req->port_select_mask[3]);
3249 port_num = find_first_bit((unsigned long *)&port_mask,
3250 sizeof(port_mask));
3252 if (port_num != port) {
3253 pmp->mad_hdr.status |= IB_SMP_INVALID_FIELD;
3254 return reply((struct ib_mad_hdr *)pmp);
3257 error_info_select = be32_to_cpu(req->error_info_select_mask);
3259 /* PortRcvErrorInfo */
3260 if (error_info_select & ES_PORT_RCV_ERROR_INFO)
3261 /* turn off status bit */
3262 dd->err_info_rcvport.status_and_code &= ~OPA_EI_STATUS_SMASK;
3264 /* ExcessiverBufferOverrunInfo */
3265 if (error_info_select & ES_EXCESSIVE_BUFFER_OVERRUN_INFO)
3267 * status bit is essentially kept in the h/w - bit 5 of
3268 * RCV_ERR_INFO
3270 write_csr(dd, RCV_ERR_INFO,
3271 RCV_ERR_INFO_RCV_EXCESS_BUFFER_OVERRUN_SMASK);
3273 if (error_info_select & ES_PORT_XMIT_CONSTRAINT_ERROR_INFO)
3274 dd->err_info_xmit_constraint.status &= ~OPA_EI_STATUS_SMASK;
3276 if (error_info_select & ES_PORT_RCV_CONSTRAINT_ERROR_INFO)
3277 dd->err_info_rcv_constraint.status &= ~OPA_EI_STATUS_SMASK;
3279 /* UncorrectableErrorInfo */
3280 if (error_info_select & ES_UNCORRECTABLE_ERROR_INFO)
3281 /* turn off status bit */
3282 dd->err_info_uncorrectable &= ~OPA_EI_STATUS_SMASK;
3284 /* FMConfigErrorInfo */
3285 if (error_info_select & ES_FM_CONFIG_ERROR_INFO)
3286 /* turn off status bit */
3287 dd->err_info_fmconfig &= ~OPA_EI_STATUS_SMASK;
3289 if (resp_len)
3290 *resp_len += sizeof(*req);
3292 return reply((struct ib_mad_hdr *)pmp);
3295 struct opa_congestion_info_attr {
3296 __be16 congestion_info;
3297 u8 control_table_cap; /* Multiple of 64 entry unit CCTs */
3298 u8 congestion_log_length;
3299 } __packed;
3301 static int __subn_get_opa_cong_info(struct opa_smp *smp, u32 am, u8 *data,
3302 struct ib_device *ibdev, u8 port,
3303 u32 *resp_len)
3305 struct opa_congestion_info_attr *p =
3306 (struct opa_congestion_info_attr *)data;
3307 struct hfi1_ibport *ibp = to_iport(ibdev, port);
3308 struct hfi1_pportdata *ppd = ppd_from_ibp(ibp);
3310 p->congestion_info = 0;
3311 p->control_table_cap = ppd->cc_max_table_entries;
3312 p->congestion_log_length = OPA_CONG_LOG_ELEMS;
3314 if (resp_len)
3315 *resp_len += sizeof(*p);
3317 return reply((struct ib_mad_hdr *)smp);
3320 static int __subn_get_opa_cong_setting(struct opa_smp *smp, u32 am,
3321 u8 *data, struct ib_device *ibdev,
3322 u8 port, u32 *resp_len)
3324 int i;
3325 struct opa_congestion_setting_attr *p =
3326 (struct opa_congestion_setting_attr *)data;
3327 struct hfi1_ibport *ibp = to_iport(ibdev, port);
3328 struct hfi1_pportdata *ppd = ppd_from_ibp(ibp);
3329 struct opa_congestion_setting_entry_shadow *entries;
3330 struct cc_state *cc_state;
3332 rcu_read_lock();
3334 cc_state = get_cc_state(ppd);
3336 if (!cc_state) {
3337 rcu_read_unlock();
3338 return reply((struct ib_mad_hdr *)smp);
3341 entries = cc_state->cong_setting.entries;
3342 p->port_control = cpu_to_be16(cc_state->cong_setting.port_control);
3343 p->control_map = cpu_to_be32(cc_state->cong_setting.control_map);
3344 for (i = 0; i < OPA_MAX_SLS; i++) {
3345 p->entries[i].ccti_increase = entries[i].ccti_increase;
3346 p->entries[i].ccti_timer = cpu_to_be16(entries[i].ccti_timer);
3347 p->entries[i].trigger_threshold =
3348 entries[i].trigger_threshold;
3349 p->entries[i].ccti_min = entries[i].ccti_min;
3352 rcu_read_unlock();
3354 if (resp_len)
3355 *resp_len += sizeof(*p);
3357 return reply((struct ib_mad_hdr *)smp);
3361 * Apply congestion control information stored in the ppd to the
3362 * active structure.
3364 static void apply_cc_state(struct hfi1_pportdata *ppd)
3366 struct cc_state *old_cc_state, *new_cc_state;
3368 new_cc_state = kzalloc(sizeof(*new_cc_state), GFP_KERNEL);
3369 if (!new_cc_state)
3370 return;
3373 * Hold the lock for updating *and* to prevent ppd information
3374 * from changing during the update.
3376 spin_lock(&ppd->cc_state_lock);
3378 old_cc_state = get_cc_state_protected(ppd);
3379 if (!old_cc_state) {
3380 /* never active, or shutting down */
3381 spin_unlock(&ppd->cc_state_lock);
3382 kfree(new_cc_state);
3383 return;
3386 *new_cc_state = *old_cc_state;
3388 new_cc_state->cct.ccti_limit = ppd->total_cct_entry - 1;
3389 memcpy(new_cc_state->cct.entries, ppd->ccti_entries,
3390 ppd->total_cct_entry * sizeof(struct ib_cc_table_entry));
3392 new_cc_state->cong_setting.port_control = IB_CC_CCS_PC_SL_BASED;
3393 new_cc_state->cong_setting.control_map = ppd->cc_sl_control_map;
3394 memcpy(new_cc_state->cong_setting.entries, ppd->congestion_entries,
3395 OPA_MAX_SLS * sizeof(struct opa_congestion_setting_entry));
3397 rcu_assign_pointer(ppd->cc_state, new_cc_state);
3399 spin_unlock(&ppd->cc_state_lock);
3401 call_rcu(&old_cc_state->rcu, cc_state_reclaim);
3404 static int __subn_set_opa_cong_setting(struct opa_smp *smp, u32 am, u8 *data,
3405 struct ib_device *ibdev, u8 port,
3406 u32 *resp_len)
3408 struct opa_congestion_setting_attr *p =
3409 (struct opa_congestion_setting_attr *)data;
3410 struct hfi1_ibport *ibp = to_iport(ibdev, port);
3411 struct hfi1_pportdata *ppd = ppd_from_ibp(ibp);
3412 struct opa_congestion_setting_entry_shadow *entries;
3413 int i;
3416 * Save details from packet into the ppd. Hold the cc_state_lock so
3417 * our information is consistent with anyone trying to apply the state.
3419 spin_lock(&ppd->cc_state_lock);
3420 ppd->cc_sl_control_map = be32_to_cpu(p->control_map);
3422 entries = ppd->congestion_entries;
3423 for (i = 0; i < OPA_MAX_SLS; i++) {
3424 entries[i].ccti_increase = p->entries[i].ccti_increase;
3425 entries[i].ccti_timer = be16_to_cpu(p->entries[i].ccti_timer);
3426 entries[i].trigger_threshold =
3427 p->entries[i].trigger_threshold;
3428 entries[i].ccti_min = p->entries[i].ccti_min;
3430 spin_unlock(&ppd->cc_state_lock);
3432 /* now apply the information */
3433 apply_cc_state(ppd);
3435 return __subn_get_opa_cong_setting(smp, am, data, ibdev, port,
3436 resp_len);
3439 static int __subn_get_opa_hfi1_cong_log(struct opa_smp *smp, u32 am,
3440 u8 *data, struct ib_device *ibdev,
3441 u8 port, u32 *resp_len)
3443 struct hfi1_ibport *ibp = to_iport(ibdev, port);
3444 struct hfi1_pportdata *ppd = ppd_from_ibp(ibp);
3445 struct opa_hfi1_cong_log *cong_log = (struct opa_hfi1_cong_log *)data;
3446 s64 ts;
3447 int i;
3449 if (am != 0) {
3450 smp->status |= IB_SMP_INVALID_FIELD;
3451 return reply((struct ib_mad_hdr *)smp);
3454 spin_lock_irq(&ppd->cc_log_lock);
3456 cong_log->log_type = OPA_CC_LOG_TYPE_HFI;
3457 cong_log->congestion_flags = 0;
3458 cong_log->threshold_event_counter =
3459 cpu_to_be16(ppd->threshold_event_counter);
3460 memcpy(cong_log->threshold_cong_event_map,
3461 ppd->threshold_cong_event_map,
3462 sizeof(cong_log->threshold_cong_event_map));
3463 /* keep timestamp in units of 1.024 usec */
3464 ts = ktime_to_ns(ktime_get()) / 1024;
3465 cong_log->current_time_stamp = cpu_to_be32(ts);
3466 for (i = 0; i < OPA_CONG_LOG_ELEMS; i++) {
3467 struct opa_hfi1_cong_log_event_internal *cce =
3468 &ppd->cc_events[ppd->cc_mad_idx++];
3469 if (ppd->cc_mad_idx == OPA_CONG_LOG_ELEMS)
3470 ppd->cc_mad_idx = 0;
3472 * Entries which are older than twice the time
3473 * required to wrap the counter are supposed to
3474 * be zeroed (CA10-49 IBTA, release 1.2.1, V1).
3476 if ((u64)(ts - cce->timestamp) > (2 * UINT_MAX))
3477 continue;
3478 memcpy(cong_log->events[i].local_qp_cn_entry, &cce->lqpn, 3);
3479 memcpy(cong_log->events[i].remote_qp_number_cn_entry,
3480 &cce->rqpn, 3);
3481 cong_log->events[i].sl_svc_type_cn_entry =
3482 ((cce->sl & 0x1f) << 3) | (cce->svc_type & 0x7);
3483 cong_log->events[i].remote_lid_cn_entry =
3484 cpu_to_be32(cce->rlid);
3485 cong_log->events[i].timestamp_cn_entry =
3486 cpu_to_be32(cce->timestamp);
3490 * Reset threshold_cong_event_map, and threshold_event_counter
3491 * to 0 when log is read.
3493 memset(ppd->threshold_cong_event_map, 0x0,
3494 sizeof(ppd->threshold_cong_event_map));
3495 ppd->threshold_event_counter = 0;
3497 spin_unlock_irq(&ppd->cc_log_lock);
3499 if (resp_len)
3500 *resp_len += sizeof(struct opa_hfi1_cong_log);
3502 return reply((struct ib_mad_hdr *)smp);
3505 static int __subn_get_opa_cc_table(struct opa_smp *smp, u32 am, u8 *data,
3506 struct ib_device *ibdev, u8 port,
3507 u32 *resp_len)
3509 struct ib_cc_table_attr *cc_table_attr =
3510 (struct ib_cc_table_attr *)data;
3511 struct hfi1_ibport *ibp = to_iport(ibdev, port);
3512 struct hfi1_pportdata *ppd = ppd_from_ibp(ibp);
3513 u32 start_block = OPA_AM_START_BLK(am);
3514 u32 n_blocks = OPA_AM_NBLK(am);
3515 struct ib_cc_table_entry_shadow *entries;
3516 int i, j;
3517 u32 sentry, eentry;
3518 struct cc_state *cc_state;
3520 /* sanity check n_blocks, start_block */
3521 if (n_blocks == 0 ||
3522 start_block + n_blocks > ppd->cc_max_table_entries) {
3523 smp->status |= IB_SMP_INVALID_FIELD;
3524 return reply((struct ib_mad_hdr *)smp);
3527 rcu_read_lock();
3529 cc_state = get_cc_state(ppd);
3531 if (!cc_state) {
3532 rcu_read_unlock();
3533 return reply((struct ib_mad_hdr *)smp);
3536 sentry = start_block * IB_CCT_ENTRIES;
3537 eentry = sentry + (IB_CCT_ENTRIES * n_blocks);
3539 cc_table_attr->ccti_limit = cpu_to_be16(cc_state->cct.ccti_limit);
3541 entries = cc_state->cct.entries;
3543 /* return n_blocks, though the last block may not be full */
3544 for (j = 0, i = sentry; i < eentry; j++, i++)
3545 cc_table_attr->ccti_entries[j].entry =
3546 cpu_to_be16(entries[i].entry);
3548 rcu_read_unlock();
3550 if (resp_len)
3551 *resp_len += sizeof(u16) * (IB_CCT_ENTRIES * n_blocks + 1);
3553 return reply((struct ib_mad_hdr *)smp);
3556 void cc_state_reclaim(struct rcu_head *rcu)
3558 struct cc_state *cc_state = container_of(rcu, struct cc_state, rcu);
3560 kfree(cc_state);
3563 static int __subn_set_opa_cc_table(struct opa_smp *smp, u32 am, u8 *data,
3564 struct ib_device *ibdev, u8 port,
3565 u32 *resp_len)
3567 struct ib_cc_table_attr *p = (struct ib_cc_table_attr *)data;
3568 struct hfi1_ibport *ibp = to_iport(ibdev, port);
3569 struct hfi1_pportdata *ppd = ppd_from_ibp(ibp);
3570 u32 start_block = OPA_AM_START_BLK(am);
3571 u32 n_blocks = OPA_AM_NBLK(am);
3572 struct ib_cc_table_entry_shadow *entries;
3573 int i, j;
3574 u32 sentry, eentry;
3575 u16 ccti_limit;
3577 /* sanity check n_blocks, start_block */
3578 if (n_blocks == 0 ||
3579 start_block + n_blocks > ppd->cc_max_table_entries) {
3580 smp->status |= IB_SMP_INVALID_FIELD;
3581 return reply((struct ib_mad_hdr *)smp);
3584 sentry = start_block * IB_CCT_ENTRIES;
3585 eentry = sentry + ((n_blocks - 1) * IB_CCT_ENTRIES) +
3586 (be16_to_cpu(p->ccti_limit)) % IB_CCT_ENTRIES + 1;
3588 /* sanity check ccti_limit */
3589 ccti_limit = be16_to_cpu(p->ccti_limit);
3590 if (ccti_limit + 1 > eentry) {
3591 smp->status |= IB_SMP_INVALID_FIELD;
3592 return reply((struct ib_mad_hdr *)smp);
3596 * Save details from packet into the ppd. Hold the cc_state_lock so
3597 * our information is consistent with anyone trying to apply the state.
3599 spin_lock(&ppd->cc_state_lock);
3600 ppd->total_cct_entry = ccti_limit + 1;
3601 entries = ppd->ccti_entries;
3602 for (j = 0, i = sentry; i < eentry; j++, i++)
3603 entries[i].entry = be16_to_cpu(p->ccti_entries[j].entry);
3604 spin_unlock(&ppd->cc_state_lock);
3606 /* now apply the information */
3607 apply_cc_state(ppd);
3609 return __subn_get_opa_cc_table(smp, am, data, ibdev, port, resp_len);
3612 struct opa_led_info {
3613 __be32 rsvd_led_mask;
3614 __be32 rsvd;
3617 #define OPA_LED_SHIFT 31
3618 #define OPA_LED_MASK BIT(OPA_LED_SHIFT)
3620 static int __subn_get_opa_led_info(struct opa_smp *smp, u32 am, u8 *data,
3621 struct ib_device *ibdev, u8 port,
3622 u32 *resp_len)
3624 struct hfi1_devdata *dd = dd_from_ibdev(ibdev);
3625 struct hfi1_pportdata *ppd = dd->pport;
3626 struct opa_led_info *p = (struct opa_led_info *)data;
3627 u32 nport = OPA_AM_NPORT(am);
3628 u32 is_beaconing_active;
3630 if (nport != 1) {
3631 smp->status |= IB_SMP_INVALID_FIELD;
3632 return reply((struct ib_mad_hdr *)smp);
3636 * This pairs with the memory barrier in hfi1_start_led_override to
3637 * ensure that we read the correct state of LED beaconing represented
3638 * by led_override_timer_active
3640 smp_rmb();
3641 is_beaconing_active = !!atomic_read(&ppd->led_override_timer_active);
3642 p->rsvd_led_mask = cpu_to_be32(is_beaconing_active << OPA_LED_SHIFT);
3644 if (resp_len)
3645 *resp_len += sizeof(struct opa_led_info);
3647 return reply((struct ib_mad_hdr *)smp);
3650 static int __subn_set_opa_led_info(struct opa_smp *smp, u32 am, u8 *data,
3651 struct ib_device *ibdev, u8 port,
3652 u32 *resp_len)
3654 struct hfi1_devdata *dd = dd_from_ibdev(ibdev);
3655 struct opa_led_info *p = (struct opa_led_info *)data;
3656 u32 nport = OPA_AM_NPORT(am);
3657 int on = !!(be32_to_cpu(p->rsvd_led_mask) & OPA_LED_MASK);
3659 if (nport != 1) {
3660 smp->status |= IB_SMP_INVALID_FIELD;
3661 return reply((struct ib_mad_hdr *)smp);
3664 if (on)
3665 hfi1_start_led_override(dd->pport, 2000, 1500);
3666 else
3667 shutdown_led_override(dd->pport);
3669 return __subn_get_opa_led_info(smp, am, data, ibdev, port, resp_len);
3672 static int subn_get_opa_sma(__be16 attr_id, struct opa_smp *smp, u32 am,
3673 u8 *data, struct ib_device *ibdev, u8 port,
3674 u32 *resp_len)
3676 int ret;
3677 struct hfi1_ibport *ibp = to_iport(ibdev, port);
3679 switch (attr_id) {
3680 case IB_SMP_ATTR_NODE_DESC:
3681 ret = __subn_get_opa_nodedesc(smp, am, data, ibdev, port,
3682 resp_len);
3683 break;
3684 case IB_SMP_ATTR_NODE_INFO:
3685 ret = __subn_get_opa_nodeinfo(smp, am, data, ibdev, port,
3686 resp_len);
3687 break;
3688 case IB_SMP_ATTR_PORT_INFO:
3689 ret = __subn_get_opa_portinfo(smp, am, data, ibdev, port,
3690 resp_len);
3691 break;
3692 case IB_SMP_ATTR_PKEY_TABLE:
3693 ret = __subn_get_opa_pkeytable(smp, am, data, ibdev, port,
3694 resp_len);
3695 break;
3696 case OPA_ATTRIB_ID_SL_TO_SC_MAP:
3697 ret = __subn_get_opa_sl_to_sc(smp, am, data, ibdev, port,
3698 resp_len);
3699 break;
3700 case OPA_ATTRIB_ID_SC_TO_SL_MAP:
3701 ret = __subn_get_opa_sc_to_sl(smp, am, data, ibdev, port,
3702 resp_len);
3703 break;
3704 case OPA_ATTRIB_ID_SC_TO_VLT_MAP:
3705 ret = __subn_get_opa_sc_to_vlt(smp, am, data, ibdev, port,
3706 resp_len);
3707 break;
3708 case OPA_ATTRIB_ID_SC_TO_VLNT_MAP:
3709 ret = __subn_get_opa_sc_to_vlnt(smp, am, data, ibdev, port,
3710 resp_len);
3711 break;
3712 case OPA_ATTRIB_ID_PORT_STATE_INFO:
3713 ret = __subn_get_opa_psi(smp, am, data, ibdev, port,
3714 resp_len);
3715 break;
3716 case OPA_ATTRIB_ID_BUFFER_CONTROL_TABLE:
3717 ret = __subn_get_opa_bct(smp, am, data, ibdev, port,
3718 resp_len);
3719 break;
3720 case OPA_ATTRIB_ID_CABLE_INFO:
3721 ret = __subn_get_opa_cable_info(smp, am, data, ibdev, port,
3722 resp_len);
3723 break;
3724 case IB_SMP_ATTR_VL_ARB_TABLE:
3725 ret = __subn_get_opa_vl_arb(smp, am, data, ibdev, port,
3726 resp_len);
3727 break;
3728 case OPA_ATTRIB_ID_CONGESTION_INFO:
3729 ret = __subn_get_opa_cong_info(smp, am, data, ibdev, port,
3730 resp_len);
3731 break;
3732 case OPA_ATTRIB_ID_HFI_CONGESTION_SETTING:
3733 ret = __subn_get_opa_cong_setting(smp, am, data, ibdev,
3734 port, resp_len);
3735 break;
3736 case OPA_ATTRIB_ID_HFI_CONGESTION_LOG:
3737 ret = __subn_get_opa_hfi1_cong_log(smp, am, data, ibdev,
3738 port, resp_len);
3739 break;
3740 case OPA_ATTRIB_ID_CONGESTION_CONTROL_TABLE:
3741 ret = __subn_get_opa_cc_table(smp, am, data, ibdev, port,
3742 resp_len);
3743 break;
3744 case IB_SMP_ATTR_LED_INFO:
3745 ret = __subn_get_opa_led_info(smp, am, data, ibdev, port,
3746 resp_len);
3747 break;
3748 case IB_SMP_ATTR_SM_INFO:
3749 if (ibp->rvp.port_cap_flags & IB_PORT_SM_DISABLED)
3750 return IB_MAD_RESULT_SUCCESS | IB_MAD_RESULT_CONSUMED;
3751 if (ibp->rvp.port_cap_flags & IB_PORT_SM)
3752 return IB_MAD_RESULT_SUCCESS;
3753 /* FALLTHROUGH */
3754 default:
3755 smp->status |= IB_SMP_UNSUP_METH_ATTR;
3756 ret = reply((struct ib_mad_hdr *)smp);
3757 break;
3759 return ret;
3762 static int subn_set_opa_sma(__be16 attr_id, struct opa_smp *smp, u32 am,
3763 u8 *data, struct ib_device *ibdev, u8 port,
3764 u32 *resp_len)
3766 int ret;
3767 struct hfi1_ibport *ibp = to_iport(ibdev, port);
3769 switch (attr_id) {
3770 case IB_SMP_ATTR_PORT_INFO:
3771 ret = __subn_set_opa_portinfo(smp, am, data, ibdev, port,
3772 resp_len);
3773 break;
3774 case IB_SMP_ATTR_PKEY_TABLE:
3775 ret = __subn_set_opa_pkeytable(smp, am, data, ibdev, port,
3776 resp_len);
3777 break;
3778 case OPA_ATTRIB_ID_SL_TO_SC_MAP:
3779 ret = __subn_set_opa_sl_to_sc(smp, am, data, ibdev, port,
3780 resp_len);
3781 break;
3782 case OPA_ATTRIB_ID_SC_TO_SL_MAP:
3783 ret = __subn_set_opa_sc_to_sl(smp, am, data, ibdev, port,
3784 resp_len);
3785 break;
3786 case OPA_ATTRIB_ID_SC_TO_VLT_MAP:
3787 ret = __subn_set_opa_sc_to_vlt(smp, am, data, ibdev, port,
3788 resp_len);
3789 break;
3790 case OPA_ATTRIB_ID_SC_TO_VLNT_MAP:
3791 ret = __subn_set_opa_sc_to_vlnt(smp, am, data, ibdev, port,
3792 resp_len);
3793 break;
3794 case OPA_ATTRIB_ID_PORT_STATE_INFO:
3795 ret = __subn_set_opa_psi(smp, am, data, ibdev, port,
3796 resp_len);
3797 break;
3798 case OPA_ATTRIB_ID_BUFFER_CONTROL_TABLE:
3799 ret = __subn_set_opa_bct(smp, am, data, ibdev, port,
3800 resp_len);
3801 break;
3802 case IB_SMP_ATTR_VL_ARB_TABLE:
3803 ret = __subn_set_opa_vl_arb(smp, am, data, ibdev, port,
3804 resp_len);
3805 break;
3806 case OPA_ATTRIB_ID_HFI_CONGESTION_SETTING:
3807 ret = __subn_set_opa_cong_setting(smp, am, data, ibdev,
3808 port, resp_len);
3809 break;
3810 case OPA_ATTRIB_ID_CONGESTION_CONTROL_TABLE:
3811 ret = __subn_set_opa_cc_table(smp, am, data, ibdev, port,
3812 resp_len);
3813 break;
3814 case IB_SMP_ATTR_LED_INFO:
3815 ret = __subn_set_opa_led_info(smp, am, data, ibdev, port,
3816 resp_len);
3817 break;
3818 case IB_SMP_ATTR_SM_INFO:
3819 if (ibp->rvp.port_cap_flags & IB_PORT_SM_DISABLED)
3820 return IB_MAD_RESULT_SUCCESS | IB_MAD_RESULT_CONSUMED;
3821 if (ibp->rvp.port_cap_flags & IB_PORT_SM)
3822 return IB_MAD_RESULT_SUCCESS;
3823 /* FALLTHROUGH */
3824 default:
3825 smp->status |= IB_SMP_UNSUP_METH_ATTR;
3826 ret = reply((struct ib_mad_hdr *)smp);
3827 break;
3829 return ret;
3832 static inline void set_aggr_error(struct opa_aggregate *ag)
3834 ag->err_reqlength |= cpu_to_be16(0x8000);
3837 static int subn_get_opa_aggregate(struct opa_smp *smp,
3838 struct ib_device *ibdev, u8 port,
3839 u32 *resp_len)
3841 int i;
3842 u32 num_attr = be32_to_cpu(smp->attr_mod) & 0x000000ff;
3843 u8 *next_smp = opa_get_smp_data(smp);
3845 if (num_attr < 1 || num_attr > 117) {
3846 smp->status |= IB_SMP_INVALID_FIELD;
3847 return reply((struct ib_mad_hdr *)smp);
3850 for (i = 0; i < num_attr; i++) {
3851 struct opa_aggregate *agg;
3852 size_t agg_data_len;
3853 size_t agg_size;
3854 u32 am;
3856 agg = (struct opa_aggregate *)next_smp;
3857 agg_data_len = (be16_to_cpu(agg->err_reqlength) & 0x007f) * 8;
3858 agg_size = sizeof(*agg) + agg_data_len;
3859 am = be32_to_cpu(agg->attr_mod);
3861 *resp_len += agg_size;
3863 if (next_smp + agg_size > ((u8 *)smp) + sizeof(*smp)) {
3864 smp->status |= IB_SMP_INVALID_FIELD;
3865 return reply((struct ib_mad_hdr *)smp);
3868 /* zero the payload for this segment */
3869 memset(next_smp + sizeof(*agg), 0, agg_data_len);
3871 (void)subn_get_opa_sma(agg->attr_id, smp, am, agg->data,
3872 ibdev, port, NULL);
3873 if (smp->status & ~IB_SMP_DIRECTION) {
3874 set_aggr_error(agg);
3875 return reply((struct ib_mad_hdr *)smp);
3877 next_smp += agg_size;
3880 return reply((struct ib_mad_hdr *)smp);
3883 static int subn_set_opa_aggregate(struct opa_smp *smp,
3884 struct ib_device *ibdev, u8 port,
3885 u32 *resp_len)
3887 int i;
3888 u32 num_attr = be32_to_cpu(smp->attr_mod) & 0x000000ff;
3889 u8 *next_smp = opa_get_smp_data(smp);
3891 if (num_attr < 1 || num_attr > 117) {
3892 smp->status |= IB_SMP_INVALID_FIELD;
3893 return reply((struct ib_mad_hdr *)smp);
3896 for (i = 0; i < num_attr; i++) {
3897 struct opa_aggregate *agg;
3898 size_t agg_data_len;
3899 size_t agg_size;
3900 u32 am;
3902 agg = (struct opa_aggregate *)next_smp;
3903 agg_data_len = (be16_to_cpu(agg->err_reqlength) & 0x007f) * 8;
3904 agg_size = sizeof(*agg) + agg_data_len;
3905 am = be32_to_cpu(agg->attr_mod);
3907 *resp_len += agg_size;
3909 if (next_smp + agg_size > ((u8 *)smp) + sizeof(*smp)) {
3910 smp->status |= IB_SMP_INVALID_FIELD;
3911 return reply((struct ib_mad_hdr *)smp);
3914 (void)subn_set_opa_sma(agg->attr_id, smp, am, agg->data,
3915 ibdev, port, NULL);
3916 if (smp->status & ~IB_SMP_DIRECTION) {
3917 set_aggr_error(agg);
3918 return reply((struct ib_mad_hdr *)smp);
3920 next_smp += agg_size;
3923 return reply((struct ib_mad_hdr *)smp);
3927 * OPAv1 specifies that, on the transition to link up, these counters
3928 * are cleared:
3929 * PortRcvErrors [*]
3930 * LinkErrorRecovery
3931 * LocalLinkIntegrityErrors
3932 * ExcessiveBufferOverruns [*]
3934 * [*] Error info associated with these counters is retained, but the
3935 * error info status is reset to 0.
3937 void clear_linkup_counters(struct hfi1_devdata *dd)
3939 /* PortRcvErrors */
3940 write_dev_cntr(dd, C_DC_RCV_ERR, CNTR_INVALID_VL, 0);
3941 dd->err_info_rcvport.status_and_code &= ~OPA_EI_STATUS_SMASK;
3942 /* LinkErrorRecovery */
3943 write_dev_cntr(dd, C_DC_SEQ_CRC_CNT, CNTR_INVALID_VL, 0);
3944 write_dev_cntr(dd, C_DC_REINIT_FROM_PEER_CNT, CNTR_INVALID_VL, 0);
3945 /* LocalLinkIntegrityErrors */
3946 write_dev_cntr(dd, C_DC_RX_REPLAY, CNTR_INVALID_VL, 0);
3947 /* ExcessiveBufferOverruns */
3948 write_dev_cntr(dd, C_RCV_OVF, CNTR_INVALID_VL, 0);
3949 dd->rcv_ovfl_cnt = 0;
3950 dd->err_info_xmit_constraint.status &= ~OPA_EI_STATUS_SMASK;
3954 * is_local_mad() returns 1 if 'mad' is sent from, and destined to the
3955 * local node, 0 otherwise.
3957 static int is_local_mad(struct hfi1_ibport *ibp, const struct opa_mad *mad,
3958 const struct ib_wc *in_wc)
3960 struct hfi1_pportdata *ppd = ppd_from_ibp(ibp);
3961 const struct opa_smp *smp = (const struct opa_smp *)mad;
3963 if (smp->mgmt_class == IB_MGMT_CLASS_SUBN_DIRECTED_ROUTE) {
3964 return (smp->hop_cnt == 0 &&
3965 smp->route.dr.dr_slid == OPA_LID_PERMISSIVE &&
3966 smp->route.dr.dr_dlid == OPA_LID_PERMISSIVE);
3969 return (in_wc->slid == ppd->lid);
3973 * opa_local_smp_check() should only be called on MADs for which
3974 * is_local_mad() returns true. It applies the SMP checks that are
3975 * specific to SMPs which are sent from, and destined to this node.
3976 * opa_local_smp_check() returns 0 if the SMP passes its checks, 1
3977 * otherwise.
3979 * SMPs which arrive from other nodes are instead checked by
3980 * opa_smp_check().
3982 static int opa_local_smp_check(struct hfi1_ibport *ibp,
3983 const struct ib_wc *in_wc)
3985 struct hfi1_pportdata *ppd = ppd_from_ibp(ibp);
3986 u16 slid = in_wc->slid;
3987 u16 pkey;
3989 if (in_wc->pkey_index >= ARRAY_SIZE(ppd->pkeys))
3990 return 1;
3992 pkey = ppd->pkeys[in_wc->pkey_index];
3994 * We need to do the "node-local" checks specified in OPAv1,
3995 * rev 0.90, section 9.10.26, which are:
3996 * - pkey is 0x7fff, or 0xffff
3997 * - Source QPN == 0 || Destination QPN == 0
3998 * - the MAD header's management class is either
3999 * IB_MGMT_CLASS_SUBN_DIRECTED_ROUTE or
4000 * IB_MGMT_CLASS_SUBN_LID_ROUTED
4001 * - SLID != 0
4003 * However, we know (and so don't need to check again) that,
4004 * for local SMPs, the MAD stack passes MADs with:
4005 * - Source QPN of 0
4006 * - MAD mgmt_class is IB_MGMT_CLASS_SUBN_DIRECTED_ROUTE
4007 * - SLID is either: OPA_LID_PERMISSIVE (0xFFFFFFFF), or
4008 * our own port's lid
4011 if (pkey == LIM_MGMT_P_KEY || pkey == FULL_MGMT_P_KEY)
4012 return 0;
4013 ingress_pkey_table_fail(ppd, pkey, slid);
4014 return 1;
4017 static int process_subn_opa(struct ib_device *ibdev, int mad_flags,
4018 u8 port, const struct opa_mad *in_mad,
4019 struct opa_mad *out_mad,
4020 u32 *resp_len)
4022 struct opa_smp *smp = (struct opa_smp *)out_mad;
4023 struct hfi1_ibport *ibp = to_iport(ibdev, port);
4024 u8 *data;
4025 u32 am;
4026 __be16 attr_id;
4027 int ret;
4029 *out_mad = *in_mad;
4030 data = opa_get_smp_data(smp);
4032 am = be32_to_cpu(smp->attr_mod);
4033 attr_id = smp->attr_id;
4034 if (smp->class_version != OPA_SMI_CLASS_VERSION) {
4035 smp->status |= IB_SMP_UNSUP_VERSION;
4036 ret = reply((struct ib_mad_hdr *)smp);
4037 return ret;
4039 ret = check_mkey(ibp, (struct ib_mad_hdr *)smp, mad_flags, smp->mkey,
4040 smp->route.dr.dr_slid, smp->route.dr.return_path,
4041 smp->hop_cnt);
4042 if (ret) {
4043 u32 port_num = be32_to_cpu(smp->attr_mod);
4046 * If this is a get/set portinfo, we already check the
4047 * M_Key if the MAD is for another port and the M_Key
4048 * is OK on the receiving port. This check is needed
4049 * to increment the error counters when the M_Key
4050 * fails to match on *both* ports.
4052 if (attr_id == IB_SMP_ATTR_PORT_INFO &&
4053 (smp->method == IB_MGMT_METHOD_GET ||
4054 smp->method == IB_MGMT_METHOD_SET) &&
4055 port_num && port_num <= ibdev->phys_port_cnt &&
4056 port != port_num)
4057 (void)check_mkey(to_iport(ibdev, port_num),
4058 (struct ib_mad_hdr *)smp, 0,
4059 smp->mkey, smp->route.dr.dr_slid,
4060 smp->route.dr.return_path,
4061 smp->hop_cnt);
4062 ret = IB_MAD_RESULT_FAILURE;
4063 return ret;
4066 *resp_len = opa_get_smp_header_size(smp);
4068 switch (smp->method) {
4069 case IB_MGMT_METHOD_GET:
4070 switch (attr_id) {
4071 default:
4072 clear_opa_smp_data(smp);
4073 ret = subn_get_opa_sma(attr_id, smp, am, data,
4074 ibdev, port, resp_len);
4075 break;
4076 case OPA_ATTRIB_ID_AGGREGATE:
4077 ret = subn_get_opa_aggregate(smp, ibdev, port,
4078 resp_len);
4079 break;
4081 break;
4082 case IB_MGMT_METHOD_SET:
4083 switch (attr_id) {
4084 default:
4085 ret = subn_set_opa_sma(attr_id, smp, am, data,
4086 ibdev, port, resp_len);
4087 break;
4088 case OPA_ATTRIB_ID_AGGREGATE:
4089 ret = subn_set_opa_aggregate(smp, ibdev, port,
4090 resp_len);
4091 break;
4093 break;
4094 case IB_MGMT_METHOD_TRAP:
4095 case IB_MGMT_METHOD_REPORT:
4096 case IB_MGMT_METHOD_REPORT_RESP:
4097 case IB_MGMT_METHOD_GET_RESP:
4099 * The ib_mad module will call us to process responses
4100 * before checking for other consumers.
4101 * Just tell the caller to process it normally.
4103 ret = IB_MAD_RESULT_SUCCESS;
4104 break;
4105 default:
4106 smp->status |= IB_SMP_UNSUP_METHOD;
4107 ret = reply((struct ib_mad_hdr *)smp);
4108 break;
4111 return ret;
4114 static int process_subn(struct ib_device *ibdev, int mad_flags,
4115 u8 port, const struct ib_mad *in_mad,
4116 struct ib_mad *out_mad)
4118 struct ib_smp *smp = (struct ib_smp *)out_mad;
4119 struct hfi1_ibport *ibp = to_iport(ibdev, port);
4120 int ret;
4122 *out_mad = *in_mad;
4123 if (smp->class_version != 1) {
4124 smp->status |= IB_SMP_UNSUP_VERSION;
4125 ret = reply((struct ib_mad_hdr *)smp);
4126 return ret;
4129 ret = check_mkey(ibp, (struct ib_mad_hdr *)smp, mad_flags,
4130 smp->mkey, (__force __be32)smp->dr_slid,
4131 smp->return_path, smp->hop_cnt);
4132 if (ret) {
4133 u32 port_num = be32_to_cpu(smp->attr_mod);
4136 * If this is a get/set portinfo, we already check the
4137 * M_Key if the MAD is for another port and the M_Key
4138 * is OK on the receiving port. This check is needed
4139 * to increment the error counters when the M_Key
4140 * fails to match on *both* ports.
4142 if (in_mad->mad_hdr.attr_id == IB_SMP_ATTR_PORT_INFO &&
4143 (smp->method == IB_MGMT_METHOD_GET ||
4144 smp->method == IB_MGMT_METHOD_SET) &&
4145 port_num && port_num <= ibdev->phys_port_cnt &&
4146 port != port_num)
4147 (void)check_mkey(to_iport(ibdev, port_num),
4148 (struct ib_mad_hdr *)smp, 0,
4149 smp->mkey,
4150 (__force __be32)smp->dr_slid,
4151 smp->return_path, smp->hop_cnt);
4152 ret = IB_MAD_RESULT_FAILURE;
4153 return ret;
4156 switch (smp->method) {
4157 case IB_MGMT_METHOD_GET:
4158 switch (smp->attr_id) {
4159 case IB_SMP_ATTR_NODE_INFO:
4160 ret = subn_get_nodeinfo(smp, ibdev, port);
4161 break;
4162 default:
4163 smp->status |= IB_SMP_UNSUP_METH_ATTR;
4164 ret = reply((struct ib_mad_hdr *)smp);
4165 break;
4167 break;
4170 return ret;
4173 static int process_perf(struct ib_device *ibdev, u8 port,
4174 const struct ib_mad *in_mad,
4175 struct ib_mad *out_mad)
4177 struct ib_pma_mad *pmp = (struct ib_pma_mad *)out_mad;
4178 struct ib_class_port_info *cpi = (struct ib_class_port_info *)
4179 &pmp->data;
4180 int ret = IB_MAD_RESULT_FAILURE;
4182 *out_mad = *in_mad;
4183 if (pmp->mad_hdr.class_version != 1) {
4184 pmp->mad_hdr.status |= IB_SMP_UNSUP_VERSION;
4185 ret = reply((struct ib_mad_hdr *)pmp);
4186 return ret;
4189 switch (pmp->mad_hdr.method) {
4190 case IB_MGMT_METHOD_GET:
4191 switch (pmp->mad_hdr.attr_id) {
4192 case IB_PMA_PORT_COUNTERS:
4193 ret = pma_get_ib_portcounters(pmp, ibdev, port);
4194 break;
4195 case IB_PMA_PORT_COUNTERS_EXT:
4196 ret = pma_get_ib_portcounters_ext(pmp, ibdev, port);
4197 break;
4198 case IB_PMA_CLASS_PORT_INFO:
4199 cpi->capability_mask = IB_PMA_CLASS_CAP_EXT_WIDTH;
4200 ret = reply((struct ib_mad_hdr *)pmp);
4201 break;
4202 default:
4203 pmp->mad_hdr.status |= IB_SMP_UNSUP_METH_ATTR;
4204 ret = reply((struct ib_mad_hdr *)pmp);
4205 break;
4207 break;
4209 case IB_MGMT_METHOD_SET:
4210 if (pmp->mad_hdr.attr_id) {
4211 pmp->mad_hdr.status |= IB_SMP_UNSUP_METH_ATTR;
4212 ret = reply((struct ib_mad_hdr *)pmp);
4214 break;
4216 case IB_MGMT_METHOD_TRAP:
4217 case IB_MGMT_METHOD_GET_RESP:
4219 * The ib_mad module will call us to process responses
4220 * before checking for other consumers.
4221 * Just tell the caller to process it normally.
4223 ret = IB_MAD_RESULT_SUCCESS;
4224 break;
4226 default:
4227 pmp->mad_hdr.status |= IB_SMP_UNSUP_METHOD;
4228 ret = reply((struct ib_mad_hdr *)pmp);
4229 break;
4232 return ret;
4235 static int process_perf_opa(struct ib_device *ibdev, u8 port,
4236 const struct opa_mad *in_mad,
4237 struct opa_mad *out_mad, u32 *resp_len)
4239 struct opa_pma_mad *pmp = (struct opa_pma_mad *)out_mad;
4240 int ret;
4242 *out_mad = *in_mad;
4244 if (pmp->mad_hdr.class_version != OPA_SMI_CLASS_VERSION) {
4245 pmp->mad_hdr.status |= IB_SMP_UNSUP_VERSION;
4246 return reply((struct ib_mad_hdr *)pmp);
4249 *resp_len = sizeof(pmp->mad_hdr);
4251 switch (pmp->mad_hdr.method) {
4252 case IB_MGMT_METHOD_GET:
4253 switch (pmp->mad_hdr.attr_id) {
4254 case IB_PMA_CLASS_PORT_INFO:
4255 ret = pma_get_opa_classportinfo(pmp, ibdev, resp_len);
4256 break;
4257 case OPA_PM_ATTRIB_ID_PORT_STATUS:
4258 ret = pma_get_opa_portstatus(pmp, ibdev, port,
4259 resp_len);
4260 break;
4261 case OPA_PM_ATTRIB_ID_DATA_PORT_COUNTERS:
4262 ret = pma_get_opa_datacounters(pmp, ibdev, port,
4263 resp_len);
4264 break;
4265 case OPA_PM_ATTRIB_ID_ERROR_PORT_COUNTERS:
4266 ret = pma_get_opa_porterrors(pmp, ibdev, port,
4267 resp_len);
4268 break;
4269 case OPA_PM_ATTRIB_ID_ERROR_INFO:
4270 ret = pma_get_opa_errorinfo(pmp, ibdev, port,
4271 resp_len);
4272 break;
4273 default:
4274 pmp->mad_hdr.status |= IB_SMP_UNSUP_METH_ATTR;
4275 ret = reply((struct ib_mad_hdr *)pmp);
4276 break;
4278 break;
4280 case IB_MGMT_METHOD_SET:
4281 switch (pmp->mad_hdr.attr_id) {
4282 case OPA_PM_ATTRIB_ID_CLEAR_PORT_STATUS:
4283 ret = pma_set_opa_portstatus(pmp, ibdev, port,
4284 resp_len);
4285 break;
4286 case OPA_PM_ATTRIB_ID_ERROR_INFO:
4287 ret = pma_set_opa_errorinfo(pmp, ibdev, port,
4288 resp_len);
4289 break;
4290 default:
4291 pmp->mad_hdr.status |= IB_SMP_UNSUP_METH_ATTR;
4292 ret = reply((struct ib_mad_hdr *)pmp);
4293 break;
4295 break;
4297 case IB_MGMT_METHOD_TRAP:
4298 case IB_MGMT_METHOD_GET_RESP:
4300 * The ib_mad module will call us to process responses
4301 * before checking for other consumers.
4302 * Just tell the caller to process it normally.
4304 ret = IB_MAD_RESULT_SUCCESS;
4305 break;
4307 default:
4308 pmp->mad_hdr.status |= IB_SMP_UNSUP_METHOD;
4309 ret = reply((struct ib_mad_hdr *)pmp);
4310 break;
4313 return ret;
4316 static int hfi1_process_opa_mad(struct ib_device *ibdev, int mad_flags,
4317 u8 port, const struct ib_wc *in_wc,
4318 const struct ib_grh *in_grh,
4319 const struct opa_mad *in_mad,
4320 struct opa_mad *out_mad, size_t *out_mad_size,
4321 u16 *out_mad_pkey_index)
4323 int ret;
4324 int pkey_idx;
4325 u32 resp_len = 0;
4326 struct hfi1_ibport *ibp = to_iport(ibdev, port);
4328 pkey_idx = hfi1_lookup_pkey_idx(ibp, LIM_MGMT_P_KEY);
4329 if (pkey_idx < 0) {
4330 pr_warn("failed to find limited mgmt pkey, defaulting 0x%x\n",
4331 hfi1_get_pkey(ibp, 1));
4332 pkey_idx = 1;
4334 *out_mad_pkey_index = (u16)pkey_idx;
4336 switch (in_mad->mad_hdr.mgmt_class) {
4337 case IB_MGMT_CLASS_SUBN_DIRECTED_ROUTE:
4338 case IB_MGMT_CLASS_SUBN_LID_ROUTED:
4339 if (is_local_mad(ibp, in_mad, in_wc)) {
4340 ret = opa_local_smp_check(ibp, in_wc);
4341 if (ret)
4342 return IB_MAD_RESULT_FAILURE;
4344 ret = process_subn_opa(ibdev, mad_flags, port, in_mad,
4345 out_mad, &resp_len);
4346 goto bail;
4347 case IB_MGMT_CLASS_PERF_MGMT:
4348 ret = process_perf_opa(ibdev, port, in_mad, out_mad,
4349 &resp_len);
4350 goto bail;
4352 default:
4353 ret = IB_MAD_RESULT_SUCCESS;
4356 bail:
4357 if (ret & IB_MAD_RESULT_REPLY)
4358 *out_mad_size = round_up(resp_len, 8);
4359 else if (ret & IB_MAD_RESULT_SUCCESS)
4360 *out_mad_size = in_wc->byte_len - sizeof(struct ib_grh);
4362 return ret;
4365 static int hfi1_process_ib_mad(struct ib_device *ibdev, int mad_flags, u8 port,
4366 const struct ib_wc *in_wc,
4367 const struct ib_grh *in_grh,
4368 const struct ib_mad *in_mad,
4369 struct ib_mad *out_mad)
4371 int ret;
4373 switch (in_mad->mad_hdr.mgmt_class) {
4374 case IB_MGMT_CLASS_SUBN_DIRECTED_ROUTE:
4375 case IB_MGMT_CLASS_SUBN_LID_ROUTED:
4376 ret = process_subn(ibdev, mad_flags, port, in_mad, out_mad);
4377 break;
4378 case IB_MGMT_CLASS_PERF_MGMT:
4379 ret = process_perf(ibdev, port, in_mad, out_mad);
4380 break;
4381 default:
4382 ret = IB_MAD_RESULT_SUCCESS;
4383 break;
4386 return ret;
4390 * hfi1_process_mad - process an incoming MAD packet
4391 * @ibdev: the infiniband device this packet came in on
4392 * @mad_flags: MAD flags
4393 * @port: the port number this packet came in on
4394 * @in_wc: the work completion entry for this packet
4395 * @in_grh: the global route header for this packet
4396 * @in_mad: the incoming MAD
4397 * @out_mad: any outgoing MAD reply
4399 * Returns IB_MAD_RESULT_SUCCESS if this is a MAD that we are not
4400 * interested in processing.
4402 * Note that the verbs framework has already done the MAD sanity checks,
4403 * and hop count/pointer updating for IB_MGMT_CLASS_SUBN_DIRECTED_ROUTE
4404 * MADs.
4406 * This is called by the ib_mad module.
4408 int hfi1_process_mad(struct ib_device *ibdev, int mad_flags, u8 port,
4409 const struct ib_wc *in_wc, const struct ib_grh *in_grh,
4410 const struct ib_mad_hdr *in_mad, size_t in_mad_size,
4411 struct ib_mad_hdr *out_mad, size_t *out_mad_size,
4412 u16 *out_mad_pkey_index)
4414 switch (in_mad->base_version) {
4415 case OPA_MGMT_BASE_VERSION:
4416 if (unlikely(in_mad_size != sizeof(struct opa_mad))) {
4417 dev_err(ibdev->dma_device, "invalid in_mad_size\n");
4418 return IB_MAD_RESULT_FAILURE;
4420 return hfi1_process_opa_mad(ibdev, mad_flags, port,
4421 in_wc, in_grh,
4422 (struct opa_mad *)in_mad,
4423 (struct opa_mad *)out_mad,
4424 out_mad_size,
4425 out_mad_pkey_index);
4426 case IB_MGMT_BASE_VERSION:
4427 return hfi1_process_ib_mad(ibdev, mad_flags, port,
4428 in_wc, in_grh,
4429 (const struct ib_mad *)in_mad,
4430 (struct ib_mad *)out_mad);
4431 default:
4432 break;
4435 return IB_MAD_RESULT_FAILURE;