xref: /linux/drivers/infiniband/hw/qib/qib_mad.c (revision 95298d63c67673c654c08952672d016212b26054)
1 /*
2  * Copyright (c) 2012 Intel Corporation.  All rights reserved.
3  * Copyright (c) 2006 - 2012 QLogic Corporation. All rights reserved.
4  * Copyright (c) 2005, 2006 PathScale, Inc. All rights reserved.
5  *
6  * This software is available to you under a choice of one of two
7  * licenses.  You may choose to be licensed under the terms of the GNU
8  * General Public License (GPL) Version 2, available from the file
9  * COPYING in the main directory of this source tree, or the
10  * OpenIB.org BSD license below:
11  *
12  *     Redistribution and use in source and binary forms, with or
13  *     without modification, are permitted provided that the following
14  *     conditions are met:
15  *
16  *      - Redistributions of source code must retain the above
17  *        copyright notice, this list of conditions and the following
18  *        disclaimer.
19  *
20  *      - Redistributions in binary form must reproduce the above
21  *        copyright notice, this list of conditions and the following
22  *        disclaimer in the documentation and/or other materials
23  *        provided with the distribution.
24  *
25  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
26  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
27  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
28  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
29  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
30  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
31  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
32  * SOFTWARE.
33  */
34 
35 #include <rdma/ib_smi.h>
36 
37 #include "qib.h"
38 #include "qib_mad.h"
39 
40 static int reply(struct ib_smp *smp)
41 {
42 	/*
43 	 * The verbs framework will handle the directed/LID route
44 	 * packet changes.
45 	 */
46 	smp->method = IB_MGMT_METHOD_GET_RESP;
47 	if (smp->mgmt_class == IB_MGMT_CLASS_SUBN_DIRECTED_ROUTE)
48 		smp->status |= IB_SMP_DIRECTION;
49 	return IB_MAD_RESULT_SUCCESS | IB_MAD_RESULT_REPLY;
50 }
51 
52 static int reply_failure(struct ib_smp *smp)
53 {
54 	/*
55 	 * The verbs framework will handle the directed/LID route
56 	 * packet changes.
57 	 */
58 	smp->method = IB_MGMT_METHOD_GET_RESP;
59 	if (smp->mgmt_class == IB_MGMT_CLASS_SUBN_DIRECTED_ROUTE)
60 		smp->status |= IB_SMP_DIRECTION;
61 	return IB_MAD_RESULT_FAILURE | IB_MAD_RESULT_REPLY;
62 }
63 
64 static void qib_send_trap(struct qib_ibport *ibp, void *data, unsigned len)
65 {
66 	struct ib_mad_send_buf *send_buf;
67 	struct ib_mad_agent *agent;
68 	struct ib_smp *smp;
69 	int ret;
70 	unsigned long flags;
71 	unsigned long timeout;
72 
73 	agent = ibp->rvp.send_agent;
74 	if (!agent)
75 		return;
76 
77 	/* o14-3.2.1 */
78 	if (!(ppd_from_ibp(ibp)->lflags & QIBL_LINKACTIVE))
79 		return;
80 
81 	/* o14-2 */
82 	if (ibp->rvp.trap_timeout &&
83 	    time_before(jiffies, ibp->rvp.trap_timeout))
84 		return;
85 
86 	send_buf = ib_create_send_mad(agent, 0, 0, 0, IB_MGMT_MAD_HDR,
87 				      IB_MGMT_MAD_DATA, GFP_ATOMIC,
88 				      IB_MGMT_BASE_VERSION);
89 	if (IS_ERR(send_buf))
90 		return;
91 
92 	smp = send_buf->mad;
93 	smp->base_version = IB_MGMT_BASE_VERSION;
94 	smp->mgmt_class = IB_MGMT_CLASS_SUBN_LID_ROUTED;
95 	smp->class_version = 1;
96 	smp->method = IB_MGMT_METHOD_TRAP;
97 	ibp->rvp.tid++;
98 	smp->tid = cpu_to_be64(ibp->rvp.tid);
99 	smp->attr_id = IB_SMP_ATTR_NOTICE;
100 	/* o14-1: smp->mkey = 0; */
101 	memcpy(smp->data, data, len);
102 
103 	spin_lock_irqsave(&ibp->rvp.lock, flags);
104 	if (!ibp->rvp.sm_ah) {
105 		if (ibp->rvp.sm_lid != be16_to_cpu(IB_LID_PERMISSIVE)) {
106 			struct ib_ah *ah;
107 
108 			ah = qib_create_qp0_ah(ibp, (u16)ibp->rvp.sm_lid);
109 			if (IS_ERR(ah))
110 				ret = PTR_ERR(ah);
111 			else {
112 				send_buf->ah = ah;
113 				ibp->rvp.sm_ah = ibah_to_rvtah(ah);
114 				ret = 0;
115 			}
116 		} else
117 			ret = -EINVAL;
118 	} else {
119 		send_buf->ah = &ibp->rvp.sm_ah->ibah;
120 		ret = 0;
121 	}
122 	spin_unlock_irqrestore(&ibp->rvp.lock, flags);
123 
124 	if (!ret)
125 		ret = ib_post_send_mad(send_buf, NULL);
126 	if (!ret) {
127 		/* 4.096 usec. */
128 		timeout = (4096 * (1UL << ibp->rvp.subnet_timeout)) / 1000;
129 		ibp->rvp.trap_timeout = jiffies + usecs_to_jiffies(timeout);
130 	} else {
131 		ib_free_send_mad(send_buf);
132 		ibp->rvp.trap_timeout = 0;
133 	}
134 }
135 
136 /*
137  * Send a bad P_Key trap (ch. 14.3.8).
138  */
139 void qib_bad_pkey(struct qib_ibport *ibp, u32 key, u32 sl,
140 		  u32 qp1, u32 qp2, __be16 lid1, __be16 lid2)
141 {
142 	struct ib_mad_notice_attr data;
143 
144 	ibp->rvp.n_pkt_drops++;
145 	ibp->rvp.pkey_violations++;
146 
147 	/* Send violation trap */
148 	data.generic_type = IB_NOTICE_TYPE_SECURITY;
149 	data.prod_type_msb = 0;
150 	data.prod_type_lsb = IB_NOTICE_PROD_CA;
151 	data.trap_num = IB_NOTICE_TRAP_BAD_PKEY;
152 	data.issuer_lid = cpu_to_be16(ppd_from_ibp(ibp)->lid);
153 	data.toggle_count = 0;
154 	memset(&data.details, 0, sizeof(data.details));
155 	data.details.ntc_257_258.lid1 = lid1;
156 	data.details.ntc_257_258.lid2 = lid2;
157 	data.details.ntc_257_258.key = cpu_to_be32(key);
158 	data.details.ntc_257_258.sl_qp1 = cpu_to_be32((sl << 28) | qp1);
159 	data.details.ntc_257_258.qp2 = cpu_to_be32(qp2);
160 
161 	qib_send_trap(ibp, &data, sizeof(data));
162 }
163 
164 /*
165  * Send a bad M_Key trap (ch. 14.3.9).
166  */
167 static void qib_bad_mkey(struct qib_ibport *ibp, struct ib_smp *smp)
168 {
169 	struct ib_mad_notice_attr data;
170 
171 	/* Send violation trap */
172 	data.generic_type = IB_NOTICE_TYPE_SECURITY;
173 	data.prod_type_msb = 0;
174 	data.prod_type_lsb = IB_NOTICE_PROD_CA;
175 	data.trap_num = IB_NOTICE_TRAP_BAD_MKEY;
176 	data.issuer_lid = cpu_to_be16(ppd_from_ibp(ibp)->lid);
177 	data.toggle_count = 0;
178 	memset(&data.details, 0, sizeof(data.details));
179 	data.details.ntc_256.lid = data.issuer_lid;
180 	data.details.ntc_256.method = smp->method;
181 	data.details.ntc_256.attr_id = smp->attr_id;
182 	data.details.ntc_256.attr_mod = smp->attr_mod;
183 	data.details.ntc_256.mkey = smp->mkey;
184 	if (smp->mgmt_class == IB_MGMT_CLASS_SUBN_DIRECTED_ROUTE) {
185 		u8 hop_cnt;
186 
187 		data.details.ntc_256.dr_slid = smp->dr_slid;
188 		data.details.ntc_256.dr_trunc_hop = IB_NOTICE_TRAP_DR_NOTICE;
189 		hop_cnt = smp->hop_cnt;
190 		if (hop_cnt > ARRAY_SIZE(data.details.ntc_256.dr_rtn_path)) {
191 			data.details.ntc_256.dr_trunc_hop |=
192 				IB_NOTICE_TRAP_DR_TRUNC;
193 			hop_cnt = ARRAY_SIZE(data.details.ntc_256.dr_rtn_path);
194 		}
195 		data.details.ntc_256.dr_trunc_hop |= hop_cnt;
196 		memcpy(data.details.ntc_256.dr_rtn_path, smp->return_path,
197 		       hop_cnt);
198 	}
199 
200 	qib_send_trap(ibp, &data, sizeof(data));
201 }
202 
203 /*
204  * Send a Port Capability Mask Changed trap (ch. 14.3.11).
205  */
206 void qib_cap_mask_chg(struct rvt_dev_info *rdi, u8 port_num)
207 {
208 	struct qib_ibdev *ibdev = container_of(rdi, struct qib_ibdev, rdi);
209 	struct qib_devdata *dd = dd_from_dev(ibdev);
210 	struct qib_ibport *ibp = &dd->pport[port_num - 1].ibport_data;
211 	struct ib_mad_notice_attr data;
212 
213 	data.generic_type = IB_NOTICE_TYPE_INFO;
214 	data.prod_type_msb = 0;
215 	data.prod_type_lsb = IB_NOTICE_PROD_CA;
216 	data.trap_num = IB_NOTICE_TRAP_CAP_MASK_CHG;
217 	data.issuer_lid = cpu_to_be16(ppd_from_ibp(ibp)->lid);
218 	data.toggle_count = 0;
219 	memset(&data.details, 0, sizeof(data.details));
220 	data.details.ntc_144.lid = data.issuer_lid;
221 	data.details.ntc_144.new_cap_mask =
222 					cpu_to_be32(ibp->rvp.port_cap_flags);
223 	qib_send_trap(ibp, &data, sizeof(data));
224 }
225 
226 /*
227  * Send a System Image GUID Changed trap (ch. 14.3.12).
228  */
229 void qib_sys_guid_chg(struct qib_ibport *ibp)
230 {
231 	struct ib_mad_notice_attr data;
232 
233 	data.generic_type = IB_NOTICE_TYPE_INFO;
234 	data.prod_type_msb = 0;
235 	data.prod_type_lsb = IB_NOTICE_PROD_CA;
236 	data.trap_num = IB_NOTICE_TRAP_SYS_GUID_CHG;
237 	data.issuer_lid = cpu_to_be16(ppd_from_ibp(ibp)->lid);
238 	data.toggle_count = 0;
239 	memset(&data.details, 0, sizeof(data.details));
240 	data.details.ntc_145.lid = data.issuer_lid;
241 	data.details.ntc_145.new_sys_guid = ib_qib_sys_image_guid;
242 
243 	qib_send_trap(ibp, &data, sizeof(data));
244 }
245 
246 /*
247  * Send a Node Description Changed trap (ch. 14.3.13).
248  */
249 void qib_node_desc_chg(struct qib_ibport *ibp)
250 {
251 	struct ib_mad_notice_attr data;
252 
253 	data.generic_type = IB_NOTICE_TYPE_INFO;
254 	data.prod_type_msb = 0;
255 	data.prod_type_lsb = IB_NOTICE_PROD_CA;
256 	data.trap_num = IB_NOTICE_TRAP_CAP_MASK_CHG;
257 	data.issuer_lid = cpu_to_be16(ppd_from_ibp(ibp)->lid);
258 	data.toggle_count = 0;
259 	memset(&data.details, 0, sizeof(data.details));
260 	data.details.ntc_144.lid = data.issuer_lid;
261 	data.details.ntc_144.local_changes = 1;
262 	data.details.ntc_144.change_flags = IB_NOTICE_TRAP_NODE_DESC_CHG;
263 
264 	qib_send_trap(ibp, &data, sizeof(data));
265 }
266 
267 static int subn_get_nodedescription(struct ib_smp *smp,
268 				    struct ib_device *ibdev)
269 {
270 	if (smp->attr_mod)
271 		smp->status |= IB_SMP_INVALID_FIELD;
272 
273 	memcpy(smp->data, ibdev->node_desc, sizeof(smp->data));
274 
275 	return reply(smp);
276 }
277 
278 static int subn_get_nodeinfo(struct ib_smp *smp, struct ib_device *ibdev,
279 			     u8 port)
280 {
281 	struct ib_node_info *nip = (struct ib_node_info *)&smp->data;
282 	struct qib_devdata *dd = dd_from_ibdev(ibdev);
283 	u32 majrev, minrev;
284 	unsigned pidx = port - 1; /* IB number port from 1, hdw from 0 */
285 
286 	/* GUID 0 is illegal */
287 	if (smp->attr_mod || pidx >= dd->num_pports ||
288 	    dd->pport[pidx].guid == 0)
289 		smp->status |= IB_SMP_INVALID_FIELD;
290 	else
291 		nip->port_guid = dd->pport[pidx].guid;
292 
293 	nip->base_version = 1;
294 	nip->class_version = 1;
295 	nip->node_type = 1;     /* channel adapter */
296 	nip->num_ports = ibdev->phys_port_cnt;
297 	/* This is already in network order */
298 	nip->sys_guid = ib_qib_sys_image_guid;
299 	nip->node_guid = dd->pport->guid; /* Use first-port GUID as node */
300 	nip->partition_cap = cpu_to_be16(qib_get_npkeys(dd));
301 	nip->device_id = cpu_to_be16(dd->deviceid);
302 	majrev = dd->majrev;
303 	minrev = dd->minrev;
304 	nip->revision = cpu_to_be32((majrev << 16) | minrev);
305 	nip->local_port_num = port;
306 	nip->vendor_id[0] = QIB_SRC_OUI_1;
307 	nip->vendor_id[1] = QIB_SRC_OUI_2;
308 	nip->vendor_id[2] = QIB_SRC_OUI_3;
309 
310 	return reply(smp);
311 }
312 
313 static int subn_get_guidinfo(struct ib_smp *smp, struct ib_device *ibdev,
314 			     u8 port)
315 {
316 	struct qib_devdata *dd = dd_from_ibdev(ibdev);
317 	u32 startgx = 8 * be32_to_cpu(smp->attr_mod);
318 	__be64 *p = (__be64 *) smp->data;
319 	unsigned pidx = port - 1; /* IB number port from 1, hdw from 0 */
320 
321 	/* 32 blocks of 8 64-bit GUIDs per block */
322 
323 	memset(smp->data, 0, sizeof(smp->data));
324 
325 	if (startgx == 0 && pidx < dd->num_pports) {
326 		struct qib_pportdata *ppd = dd->pport + pidx;
327 		struct qib_ibport *ibp = &ppd->ibport_data;
328 		__be64 g = ppd->guid;
329 		unsigned i;
330 
331 		/* GUID 0 is illegal */
332 		if (g == 0)
333 			smp->status |= IB_SMP_INVALID_FIELD;
334 		else {
335 			/* The first is a copy of the read-only HW GUID. */
336 			p[0] = g;
337 			for (i = 1; i < QIB_GUIDS_PER_PORT; i++)
338 				p[i] = ibp->guids[i - 1];
339 		}
340 	} else
341 		smp->status |= IB_SMP_INVALID_FIELD;
342 
343 	return reply(smp);
344 }
345 
346 static void set_link_width_enabled(struct qib_pportdata *ppd, u32 w)
347 {
348 	(void) ppd->dd->f_set_ib_cfg(ppd, QIB_IB_CFG_LWID_ENB, w);
349 }
350 
351 static void set_link_speed_enabled(struct qib_pportdata *ppd, u32 s)
352 {
353 	(void) ppd->dd->f_set_ib_cfg(ppd, QIB_IB_CFG_SPD_ENB, s);
354 }
355 
356 static int get_overrunthreshold(struct qib_pportdata *ppd)
357 {
358 	return ppd->dd->f_get_ib_cfg(ppd, QIB_IB_CFG_OVERRUN_THRESH);
359 }
360 
361 /**
362  * set_overrunthreshold - set the overrun threshold
363  * @ppd: the physical port data
364  * @n: the new threshold
365  *
366  * Note that this will only take effect when the link state changes.
367  */
368 static int set_overrunthreshold(struct qib_pportdata *ppd, unsigned n)
369 {
370 	(void) ppd->dd->f_set_ib_cfg(ppd, QIB_IB_CFG_OVERRUN_THRESH,
371 					 (u32)n);
372 	return 0;
373 }
374 
375 static int get_phyerrthreshold(struct qib_pportdata *ppd)
376 {
377 	return ppd->dd->f_get_ib_cfg(ppd, QIB_IB_CFG_PHYERR_THRESH);
378 }
379 
380 /**
381  * set_phyerrthreshold - set the physical error threshold
382  * @ppd: the physical port data
383  * @n: the new threshold
384  *
385  * Note that this will only take effect when the link state changes.
386  */
387 static int set_phyerrthreshold(struct qib_pportdata *ppd, unsigned n)
388 {
389 	(void) ppd->dd->f_set_ib_cfg(ppd, QIB_IB_CFG_PHYERR_THRESH,
390 					 (u32)n);
391 	return 0;
392 }
393 
394 /**
395  * get_linkdowndefaultstate - get the default linkdown state
396  * @ppd: the physical port data
397  *
398  * Returns zero if the default is POLL, 1 if the default is SLEEP.
399  */
400 static int get_linkdowndefaultstate(struct qib_pportdata *ppd)
401 {
402 	return ppd->dd->f_get_ib_cfg(ppd, QIB_IB_CFG_LINKDEFAULT) ==
403 		IB_LINKINITCMD_SLEEP;
404 }
405 
406 static int check_mkey(struct qib_ibport *ibp, struct ib_smp *smp, int mad_flags)
407 {
408 	int valid_mkey = 0;
409 	int ret = 0;
410 
411 	/* Is the mkey in the process of expiring? */
412 	if (ibp->rvp.mkey_lease_timeout &&
413 	    time_after_eq(jiffies, ibp->rvp.mkey_lease_timeout)) {
414 		/* Clear timeout and mkey protection field. */
415 		ibp->rvp.mkey_lease_timeout = 0;
416 		ibp->rvp.mkeyprot = 0;
417 	}
418 
419 	if ((mad_flags & IB_MAD_IGNORE_MKEY) ||  ibp->rvp.mkey == 0 ||
420 	    ibp->rvp.mkey == smp->mkey)
421 		valid_mkey = 1;
422 
423 	/* Unset lease timeout on any valid Get/Set/TrapRepress */
424 	if (valid_mkey && ibp->rvp.mkey_lease_timeout &&
425 	    (smp->method == IB_MGMT_METHOD_GET ||
426 	     smp->method == IB_MGMT_METHOD_SET ||
427 	     smp->method == IB_MGMT_METHOD_TRAP_REPRESS))
428 		ibp->rvp.mkey_lease_timeout = 0;
429 
430 	if (!valid_mkey) {
431 		switch (smp->method) {
432 		case IB_MGMT_METHOD_GET:
433 			/* Bad mkey not a violation below level 2 */
434 			if (ibp->rvp.mkeyprot < 2)
435 				break;
436 			/* fall through */
437 		case IB_MGMT_METHOD_SET:
438 		case IB_MGMT_METHOD_TRAP_REPRESS:
439 			if (ibp->rvp.mkey_violations != 0xFFFF)
440 				++ibp->rvp.mkey_violations;
441 			if (!ibp->rvp.mkey_lease_timeout &&
442 			    ibp->rvp.mkey_lease_period)
443 				ibp->rvp.mkey_lease_timeout = jiffies +
444 					ibp->rvp.mkey_lease_period * HZ;
445 			/* Generate a trap notice. */
446 			qib_bad_mkey(ibp, smp);
447 			ret = 1;
448 		}
449 	}
450 
451 	return ret;
452 }
453 
454 static int subn_get_portinfo(struct ib_smp *smp, struct ib_device *ibdev,
455 			     u8 port)
456 {
457 	struct qib_devdata *dd;
458 	struct qib_pportdata *ppd;
459 	struct qib_ibport *ibp;
460 	struct ib_port_info *pip = (struct ib_port_info *)smp->data;
461 	u8 mtu;
462 	int ret;
463 	u32 state;
464 	u32 port_num = be32_to_cpu(smp->attr_mod);
465 
466 	if (port_num == 0)
467 		port_num = port;
468 	else {
469 		if (port_num > ibdev->phys_port_cnt) {
470 			smp->status |= IB_SMP_INVALID_FIELD;
471 			ret = reply(smp);
472 			goto bail;
473 		}
474 		if (port_num != port) {
475 			ibp = to_iport(ibdev, port_num);
476 			ret = check_mkey(ibp, smp, 0);
477 			if (ret) {
478 				ret = IB_MAD_RESULT_FAILURE;
479 				goto bail;
480 			}
481 		}
482 	}
483 
484 	dd = dd_from_ibdev(ibdev);
485 	/* IB numbers ports from 1, hdw from 0 */
486 	ppd = dd->pport + (port_num - 1);
487 	ibp = &ppd->ibport_data;
488 
489 	/* Clear all fields.  Only set the non-zero fields. */
490 	memset(smp->data, 0, sizeof(smp->data));
491 
492 	/* Only return the mkey if the protection field allows it. */
493 	if (!(smp->method == IB_MGMT_METHOD_GET &&
494 	      ibp->rvp.mkey != smp->mkey &&
495 	      ibp->rvp.mkeyprot == 1))
496 		pip->mkey = ibp->rvp.mkey;
497 	pip->gid_prefix = ibp->rvp.gid_prefix;
498 	pip->lid = cpu_to_be16(ppd->lid);
499 	pip->sm_lid = cpu_to_be16((u16)ibp->rvp.sm_lid);
500 	pip->cap_mask = cpu_to_be32(ibp->rvp.port_cap_flags);
501 	/* pip->diag_code; */
502 	pip->mkey_lease_period = cpu_to_be16(ibp->rvp.mkey_lease_period);
503 	pip->local_port_num = port;
504 	pip->link_width_enabled = ppd->link_width_enabled;
505 	pip->link_width_supported = ppd->link_width_supported;
506 	pip->link_width_active = ppd->link_width_active;
507 	state = dd->f_iblink_state(ppd->lastibcstat);
508 	pip->linkspeed_portstate = ppd->link_speed_supported << 4 | state;
509 
510 	pip->portphysstate_linkdown =
511 		(dd->f_ibphys_portstate(ppd->lastibcstat) << 4) |
512 		(get_linkdowndefaultstate(ppd) ? 1 : 2);
513 	pip->mkeyprot_resv_lmc = (ibp->rvp.mkeyprot << 6) | ppd->lmc;
514 	pip->linkspeedactive_enabled = (ppd->link_speed_active << 4) |
515 		ppd->link_speed_enabled;
516 	switch (ppd->ibmtu) {
517 	default: /* something is wrong; fall through */
518 	case 4096:
519 		mtu = IB_MTU_4096;
520 		break;
521 	case 2048:
522 		mtu = IB_MTU_2048;
523 		break;
524 	case 1024:
525 		mtu = IB_MTU_1024;
526 		break;
527 	case 512:
528 		mtu = IB_MTU_512;
529 		break;
530 	case 256:
531 		mtu = IB_MTU_256;
532 		break;
533 	}
534 	pip->neighbormtu_mastersmsl = (mtu << 4) | ibp->rvp.sm_sl;
535 	pip->vlcap_inittype = ppd->vls_supported << 4;  /* InitType = 0 */
536 	pip->vl_high_limit = ibp->rvp.vl_high_limit;
537 	pip->vl_arb_high_cap =
538 		dd->f_get_ib_cfg(ppd, QIB_IB_CFG_VL_HIGH_CAP);
539 	pip->vl_arb_low_cap =
540 		dd->f_get_ib_cfg(ppd, QIB_IB_CFG_VL_LOW_CAP);
541 	/* InitTypeReply = 0 */
542 	pip->inittypereply_mtucap = qib_ibmtu ? qib_ibmtu : IB_MTU_4096;
543 	/* HCAs ignore VLStallCount and HOQLife */
544 	/* pip->vlstallcnt_hoqlife; */
545 	pip->operationalvl_pei_peo_fpi_fpo =
546 		dd->f_get_ib_cfg(ppd, QIB_IB_CFG_OP_VLS) << 4;
547 	pip->mkey_violations = cpu_to_be16(ibp->rvp.mkey_violations);
548 	/* P_KeyViolations are counted by hardware. */
549 	pip->pkey_violations = cpu_to_be16(ibp->rvp.pkey_violations);
550 	pip->qkey_violations = cpu_to_be16(ibp->rvp.qkey_violations);
551 	/* Only the hardware GUID is supported for now */
552 	pip->guid_cap = QIB_GUIDS_PER_PORT;
553 	pip->clientrereg_resv_subnetto = ibp->rvp.subnet_timeout;
554 	/* 32.768 usec. response time (guessing) */
555 	pip->resv_resptimevalue = 3;
556 	pip->localphyerrors_overrunerrors =
557 		(get_phyerrthreshold(ppd) << 4) |
558 		get_overrunthreshold(ppd);
559 	/* pip->max_credit_hint; */
560 	if (ibp->rvp.port_cap_flags & IB_PORT_LINK_LATENCY_SUP) {
561 		u32 v;
562 
563 		v = dd->f_get_ib_cfg(ppd, QIB_IB_CFG_LINKLATENCY);
564 		pip->link_roundtrip_latency[0] = v >> 16;
565 		pip->link_roundtrip_latency[1] = v >> 8;
566 		pip->link_roundtrip_latency[2] = v;
567 	}
568 
569 	ret = reply(smp);
570 
571 bail:
572 	return ret;
573 }
574 
575 /**
576  * get_pkeys - return the PKEY table
577  * @dd: the qlogic_ib device
578  * @port: the IB port number
579  * @pkeys: the pkey table is placed here
580  */
581 static int get_pkeys(struct qib_devdata *dd, u8 port, u16 *pkeys)
582 {
583 	struct qib_pportdata *ppd = dd->pport + port - 1;
584 	/*
585 	 * always a kernel context, no locking needed.
586 	 * If we get here with ppd setup, no need to check
587 	 * that pd is valid.
588 	 */
589 	struct qib_ctxtdata *rcd = dd->rcd[ppd->hw_pidx];
590 
591 	memcpy(pkeys, rcd->pkeys, sizeof(rcd->pkeys));
592 
593 	return 0;
594 }
595 
596 static int subn_get_pkeytable(struct ib_smp *smp, struct ib_device *ibdev,
597 			      u8 port)
598 {
599 	u32 startpx = 32 * (be32_to_cpu(smp->attr_mod) & 0xffff);
600 	u16 *p = (u16 *) smp->data;
601 	__be16 *q = (__be16 *) smp->data;
602 
603 	/* 64 blocks of 32 16-bit P_Key entries */
604 
605 	memset(smp->data, 0, sizeof(smp->data));
606 	if (startpx == 0) {
607 		struct qib_devdata *dd = dd_from_ibdev(ibdev);
608 		unsigned i, n = qib_get_npkeys(dd);
609 
610 		get_pkeys(dd, port, p);
611 
612 		for (i = 0; i < n; i++)
613 			q[i] = cpu_to_be16(p[i]);
614 	} else
615 		smp->status |= IB_SMP_INVALID_FIELD;
616 
617 	return reply(smp);
618 }
619 
620 static int subn_set_guidinfo(struct ib_smp *smp, struct ib_device *ibdev,
621 			     u8 port)
622 {
623 	struct qib_devdata *dd = dd_from_ibdev(ibdev);
624 	u32 startgx = 8 * be32_to_cpu(smp->attr_mod);
625 	__be64 *p = (__be64 *) smp->data;
626 	unsigned pidx = port - 1; /* IB number port from 1, hdw from 0 */
627 
628 	/* 32 blocks of 8 64-bit GUIDs per block */
629 
630 	if (startgx == 0 && pidx < dd->num_pports) {
631 		struct qib_pportdata *ppd = dd->pport + pidx;
632 		struct qib_ibport *ibp = &ppd->ibport_data;
633 		unsigned i;
634 
635 		/* The first entry is read-only. */
636 		for (i = 1; i < QIB_GUIDS_PER_PORT; i++)
637 			ibp->guids[i - 1] = p[i];
638 	} else
639 		smp->status |= IB_SMP_INVALID_FIELD;
640 
641 	/* The only GUID we support is the first read-only entry. */
642 	return subn_get_guidinfo(smp, ibdev, port);
643 }
644 
645 /**
646  * subn_set_portinfo - set port information
647  * @smp: the incoming SM packet
648  * @ibdev: the infiniband device
649  * @port: the port on the device
650  *
651  * Set Portinfo (see ch. 14.2.5.6).
652  */
653 static int subn_set_portinfo(struct ib_smp *smp, struct ib_device *ibdev,
654 			     u8 port)
655 {
656 	struct ib_port_info *pip = (struct ib_port_info *)smp->data;
657 	struct ib_event event;
658 	struct qib_devdata *dd;
659 	struct qib_pportdata *ppd;
660 	struct qib_ibport *ibp;
661 	u8 clientrereg = (pip->clientrereg_resv_subnetto & 0x80);
662 	unsigned long flags;
663 	u16 lid, smlid;
664 	u8 lwe;
665 	u8 lse;
666 	u8 state;
667 	u8 vls;
668 	u8 msl;
669 	u16 lstate;
670 	int ret, ore, mtu;
671 	u32 port_num = be32_to_cpu(smp->attr_mod);
672 
673 	if (port_num == 0)
674 		port_num = port;
675 	else {
676 		if (port_num > ibdev->phys_port_cnt)
677 			goto err;
678 		/* Port attributes can only be set on the receiving port */
679 		if (port_num != port)
680 			goto get_only;
681 	}
682 
683 	dd = dd_from_ibdev(ibdev);
684 	/* IB numbers ports from 1, hdw from 0 */
685 	ppd = dd->pport + (port_num - 1);
686 	ibp = &ppd->ibport_data;
687 	event.device = ibdev;
688 	event.element.port_num = port;
689 
690 	ibp->rvp.mkey = pip->mkey;
691 	ibp->rvp.gid_prefix = pip->gid_prefix;
692 	ibp->rvp.mkey_lease_period = be16_to_cpu(pip->mkey_lease_period);
693 
694 	lid = be16_to_cpu(pip->lid);
695 	/* Must be a valid unicast LID address. */
696 	if (lid == 0 || lid >= be16_to_cpu(IB_MULTICAST_LID_BASE))
697 		smp->status |= IB_SMP_INVALID_FIELD;
698 	else if (ppd->lid != lid || ppd->lmc != (pip->mkeyprot_resv_lmc & 7)) {
699 		if (ppd->lid != lid)
700 			qib_set_uevent_bits(ppd, _QIB_EVENT_LID_CHANGE_BIT);
701 		if (ppd->lmc != (pip->mkeyprot_resv_lmc & 7))
702 			qib_set_uevent_bits(ppd, _QIB_EVENT_LMC_CHANGE_BIT);
703 		qib_set_lid(ppd, lid, pip->mkeyprot_resv_lmc & 7);
704 		event.event = IB_EVENT_LID_CHANGE;
705 		ib_dispatch_event(&event);
706 	}
707 
708 	smlid = be16_to_cpu(pip->sm_lid);
709 	msl = pip->neighbormtu_mastersmsl & 0xF;
710 	/* Must be a valid unicast LID address. */
711 	if (smlid == 0 || smlid >= be16_to_cpu(IB_MULTICAST_LID_BASE))
712 		smp->status |= IB_SMP_INVALID_FIELD;
713 	else if (smlid != ibp->rvp.sm_lid || msl != ibp->rvp.sm_sl) {
714 		spin_lock_irqsave(&ibp->rvp.lock, flags);
715 		if (ibp->rvp.sm_ah) {
716 			if (smlid != ibp->rvp.sm_lid)
717 				rdma_ah_set_dlid(&ibp->rvp.sm_ah->attr,
718 						 smlid);
719 			if (msl != ibp->rvp.sm_sl)
720 				rdma_ah_set_sl(&ibp->rvp.sm_ah->attr, msl);
721 		}
722 		spin_unlock_irqrestore(&ibp->rvp.lock, flags);
723 		if (smlid != ibp->rvp.sm_lid)
724 			ibp->rvp.sm_lid = smlid;
725 		if (msl != ibp->rvp.sm_sl)
726 			ibp->rvp.sm_sl = msl;
727 		event.event = IB_EVENT_SM_CHANGE;
728 		ib_dispatch_event(&event);
729 	}
730 
731 	/* Allow 1x or 4x to be set (see 14.2.6.6). */
732 	lwe = pip->link_width_enabled;
733 	if (lwe) {
734 		if (lwe == 0xFF)
735 			set_link_width_enabled(ppd, ppd->link_width_supported);
736 		else if (lwe >= 16 || (lwe & ~ppd->link_width_supported))
737 			smp->status |= IB_SMP_INVALID_FIELD;
738 		else if (lwe != ppd->link_width_enabled)
739 			set_link_width_enabled(ppd, lwe);
740 	}
741 
742 	lse = pip->linkspeedactive_enabled & 0xF;
743 	if (lse) {
744 		/*
745 		 * The IB 1.2 spec. only allows link speed values
746 		 * 1, 3, 5, 7, 15.  1.2.1 extended to allow specific
747 		 * speeds.
748 		 */
749 		if (lse == 15)
750 			set_link_speed_enabled(ppd,
751 					       ppd->link_speed_supported);
752 		else if (lse >= 8 || (lse & ~ppd->link_speed_supported))
753 			smp->status |= IB_SMP_INVALID_FIELD;
754 		else if (lse != ppd->link_speed_enabled)
755 			set_link_speed_enabled(ppd, lse);
756 	}
757 
758 	/* Set link down default state. */
759 	switch (pip->portphysstate_linkdown & 0xF) {
760 	case 0: /* NOP */
761 		break;
762 	case 1: /* SLEEP */
763 		(void) dd->f_set_ib_cfg(ppd, QIB_IB_CFG_LINKDEFAULT,
764 					IB_LINKINITCMD_SLEEP);
765 		break;
766 	case 2: /* POLL */
767 		(void) dd->f_set_ib_cfg(ppd, QIB_IB_CFG_LINKDEFAULT,
768 					IB_LINKINITCMD_POLL);
769 		break;
770 	default:
771 		smp->status |= IB_SMP_INVALID_FIELD;
772 	}
773 
774 	ibp->rvp.mkeyprot = pip->mkeyprot_resv_lmc >> 6;
775 	ibp->rvp.vl_high_limit = pip->vl_high_limit;
776 	(void) dd->f_set_ib_cfg(ppd, QIB_IB_CFG_VL_HIGH_LIMIT,
777 				    ibp->rvp.vl_high_limit);
778 
779 	mtu = ib_mtu_enum_to_int((pip->neighbormtu_mastersmsl >> 4) & 0xF);
780 	if (mtu == -1)
781 		smp->status |= IB_SMP_INVALID_FIELD;
782 	else
783 		qib_set_mtu(ppd, mtu);
784 
785 	/* Set operational VLs */
786 	vls = (pip->operationalvl_pei_peo_fpi_fpo >> 4) & 0xF;
787 	if (vls) {
788 		if (vls > ppd->vls_supported)
789 			smp->status |= IB_SMP_INVALID_FIELD;
790 		else
791 			(void) dd->f_set_ib_cfg(ppd, QIB_IB_CFG_OP_VLS, vls);
792 	}
793 
794 	if (pip->mkey_violations == 0)
795 		ibp->rvp.mkey_violations = 0;
796 
797 	if (pip->pkey_violations == 0)
798 		ibp->rvp.pkey_violations = 0;
799 
800 	if (pip->qkey_violations == 0)
801 		ibp->rvp.qkey_violations = 0;
802 
803 	ore = pip->localphyerrors_overrunerrors;
804 	if (set_phyerrthreshold(ppd, (ore >> 4) & 0xF))
805 		smp->status |= IB_SMP_INVALID_FIELD;
806 
807 	if (set_overrunthreshold(ppd, (ore & 0xF)))
808 		smp->status |= IB_SMP_INVALID_FIELD;
809 
810 	ibp->rvp.subnet_timeout = pip->clientrereg_resv_subnetto & 0x1F;
811 
812 	/*
813 	 * Do the port state change now that the other link parameters
814 	 * have been set.
815 	 * Changing the port physical state only makes sense if the link
816 	 * is down or is being set to down.
817 	 */
818 	state = pip->linkspeed_portstate & 0xF;
819 	lstate = (pip->portphysstate_linkdown >> 4) & 0xF;
820 	if (lstate && !(state == IB_PORT_DOWN || state == IB_PORT_NOP))
821 		smp->status |= IB_SMP_INVALID_FIELD;
822 
823 	/*
824 	 * Only state changes of DOWN, ARM, and ACTIVE are valid
825 	 * and must be in the correct state to take effect (see 7.2.6).
826 	 */
827 	switch (state) {
828 	case IB_PORT_NOP:
829 		if (lstate == 0)
830 			break;
831 		/* FALLTHROUGH */
832 	case IB_PORT_DOWN:
833 		if (lstate == 0)
834 			lstate = QIB_IB_LINKDOWN_ONLY;
835 		else if (lstate == 1)
836 			lstate = QIB_IB_LINKDOWN_SLEEP;
837 		else if (lstate == 2)
838 			lstate = QIB_IB_LINKDOWN;
839 		else if (lstate == 3)
840 			lstate = QIB_IB_LINKDOWN_DISABLE;
841 		else {
842 			smp->status |= IB_SMP_INVALID_FIELD;
843 			break;
844 		}
845 		spin_lock_irqsave(&ppd->lflags_lock, flags);
846 		ppd->lflags &= ~QIBL_LINKV;
847 		spin_unlock_irqrestore(&ppd->lflags_lock, flags);
848 		qib_set_linkstate(ppd, lstate);
849 		/*
850 		 * Don't send a reply if the response would be sent
851 		 * through the disabled port.
852 		 */
853 		if (lstate == QIB_IB_LINKDOWN_DISABLE && smp->hop_cnt) {
854 			ret = IB_MAD_RESULT_SUCCESS | IB_MAD_RESULT_CONSUMED;
855 			goto done;
856 		}
857 		qib_wait_linkstate(ppd, QIBL_LINKV, 10);
858 		break;
859 	case IB_PORT_ARMED:
860 		qib_set_linkstate(ppd, QIB_IB_LINKARM);
861 		break;
862 	case IB_PORT_ACTIVE:
863 		qib_set_linkstate(ppd, QIB_IB_LINKACTIVE);
864 		break;
865 	default:
866 		smp->status |= IB_SMP_INVALID_FIELD;
867 	}
868 
869 	if (clientrereg) {
870 		event.event = IB_EVENT_CLIENT_REREGISTER;
871 		ib_dispatch_event(&event);
872 	}
873 
874 	/* restore re-reg bit per o14-12.2.1 */
875 	pip->clientrereg_resv_subnetto |= clientrereg;
876 
877 	goto get_only;
878 
879 err:
880 	smp->status |= IB_SMP_INVALID_FIELD;
881 get_only:
882 	ret = subn_get_portinfo(smp, ibdev, port);
883 done:
884 	return ret;
885 }
886 
887 /**
888  * rm_pkey - decrecment the reference count for the given PKEY
889  * @dd: the qlogic_ib device
890  * @key: the PKEY index
891  *
892  * Return true if this was the last reference and the hardware table entry
893  * needs to be changed.
894  */
895 static int rm_pkey(struct qib_pportdata *ppd, u16 key)
896 {
897 	int i;
898 	int ret;
899 
900 	for (i = 0; i < ARRAY_SIZE(ppd->pkeys); i++) {
901 		if (ppd->pkeys[i] != key)
902 			continue;
903 		if (atomic_dec_and_test(&ppd->pkeyrefs[i])) {
904 			ppd->pkeys[i] = 0;
905 			ret = 1;
906 			goto bail;
907 		}
908 		break;
909 	}
910 
911 	ret = 0;
912 
913 bail:
914 	return ret;
915 }
916 
917 /**
918  * add_pkey - add the given PKEY to the hardware table
919  * @dd: the qlogic_ib device
920  * @key: the PKEY
921  *
922  * Return an error code if unable to add the entry, zero if no change,
923  * or 1 if the hardware PKEY register needs to be updated.
924  */
925 static int add_pkey(struct qib_pportdata *ppd, u16 key)
926 {
927 	int i;
928 	u16 lkey = key & 0x7FFF;
929 	int any = 0;
930 	int ret;
931 
932 	if (lkey == 0x7FFF) {
933 		ret = 0;
934 		goto bail;
935 	}
936 
937 	/* Look for an empty slot or a matching PKEY. */
938 	for (i = 0; i < ARRAY_SIZE(ppd->pkeys); i++) {
939 		if (!ppd->pkeys[i]) {
940 			any++;
941 			continue;
942 		}
943 		/* If it matches exactly, try to increment the ref count */
944 		if (ppd->pkeys[i] == key) {
945 			if (atomic_inc_return(&ppd->pkeyrefs[i]) > 1) {
946 				ret = 0;
947 				goto bail;
948 			}
949 			/* Lost the race. Look for an empty slot below. */
950 			atomic_dec(&ppd->pkeyrefs[i]);
951 			any++;
952 		}
953 		/*
954 		 * It makes no sense to have both the limited and unlimited
955 		 * PKEY set at the same time since the unlimited one will
956 		 * disable the limited one.
957 		 */
958 		if ((ppd->pkeys[i] & 0x7FFF) == lkey) {
959 			ret = -EEXIST;
960 			goto bail;
961 		}
962 	}
963 	if (!any) {
964 		ret = -EBUSY;
965 		goto bail;
966 	}
967 	for (i = 0; i < ARRAY_SIZE(ppd->pkeys); i++) {
968 		if (!ppd->pkeys[i] &&
969 		    atomic_inc_return(&ppd->pkeyrefs[i]) == 1) {
970 			/* for qibstats, etc. */
971 			ppd->pkeys[i] = key;
972 			ret = 1;
973 			goto bail;
974 		}
975 	}
976 	ret = -EBUSY;
977 
978 bail:
979 	return ret;
980 }
981 
982 /**
983  * set_pkeys - set the PKEY table for ctxt 0
984  * @dd: the qlogic_ib device
985  * @port: the IB port number
986  * @pkeys: the PKEY table
987  */
988 static int set_pkeys(struct qib_devdata *dd, u8 port, u16 *pkeys)
989 {
990 	struct qib_pportdata *ppd;
991 	struct qib_ctxtdata *rcd;
992 	int i;
993 	int changed = 0;
994 
995 	/*
996 	 * IB port one/two always maps to context zero/one,
997 	 * always a kernel context, no locking needed
998 	 * If we get here with ppd setup, no need to check
999 	 * that rcd is valid.
1000 	 */
1001 	ppd = dd->pport + (port - 1);
1002 	rcd = dd->rcd[ppd->hw_pidx];
1003 
1004 	for (i = 0; i < ARRAY_SIZE(rcd->pkeys); i++) {
1005 		u16 key = pkeys[i];
1006 		u16 okey = rcd->pkeys[i];
1007 
1008 		if (key == okey)
1009 			continue;
1010 		/*
1011 		 * The value of this PKEY table entry is changing.
1012 		 * Remove the old entry in the hardware's array of PKEYs.
1013 		 */
1014 		if (okey & 0x7FFF)
1015 			changed |= rm_pkey(ppd, okey);
1016 		if (key & 0x7FFF) {
1017 			int ret = add_pkey(ppd, key);
1018 
1019 			if (ret < 0)
1020 				key = 0;
1021 			else
1022 				changed |= ret;
1023 		}
1024 		rcd->pkeys[i] = key;
1025 	}
1026 	if (changed) {
1027 		struct ib_event event;
1028 
1029 		(void) dd->f_set_ib_cfg(ppd, QIB_IB_CFG_PKEYS, 0);
1030 
1031 		event.event = IB_EVENT_PKEY_CHANGE;
1032 		event.device = &dd->verbs_dev.rdi.ibdev;
1033 		event.element.port_num = port;
1034 		ib_dispatch_event(&event);
1035 	}
1036 	return 0;
1037 }
1038 
1039 static int subn_set_pkeytable(struct ib_smp *smp, struct ib_device *ibdev,
1040 			      u8 port)
1041 {
1042 	u32 startpx = 32 * (be32_to_cpu(smp->attr_mod) & 0xffff);
1043 	__be16 *p = (__be16 *) smp->data;
1044 	u16 *q = (u16 *) smp->data;
1045 	struct qib_devdata *dd = dd_from_ibdev(ibdev);
1046 	unsigned i, n = qib_get_npkeys(dd);
1047 
1048 	for (i = 0; i < n; i++)
1049 		q[i] = be16_to_cpu(p[i]);
1050 
1051 	if (startpx != 0 || set_pkeys(dd, port, q) != 0)
1052 		smp->status |= IB_SMP_INVALID_FIELD;
1053 
1054 	return subn_get_pkeytable(smp, ibdev, port);
1055 }
1056 
1057 static int subn_get_sl_to_vl(struct ib_smp *smp, struct ib_device *ibdev,
1058 			     u8 port)
1059 {
1060 	struct qib_ibport *ibp = to_iport(ibdev, port);
1061 	u8 *p = (u8 *) smp->data;
1062 	unsigned i;
1063 
1064 	memset(smp->data, 0, sizeof(smp->data));
1065 
1066 	if (!(ibp->rvp.port_cap_flags & IB_PORT_SL_MAP_SUP))
1067 		smp->status |= IB_SMP_UNSUP_METHOD;
1068 	else
1069 		for (i = 0; i < ARRAY_SIZE(ibp->sl_to_vl); i += 2)
1070 			*p++ = (ibp->sl_to_vl[i] << 4) | ibp->sl_to_vl[i + 1];
1071 
1072 	return reply(smp);
1073 }
1074 
1075 static int subn_set_sl_to_vl(struct ib_smp *smp, struct ib_device *ibdev,
1076 			     u8 port)
1077 {
1078 	struct qib_ibport *ibp = to_iport(ibdev, port);
1079 	u8 *p = (u8 *) smp->data;
1080 	unsigned i;
1081 
1082 	if (!(ibp->rvp.port_cap_flags & IB_PORT_SL_MAP_SUP)) {
1083 		smp->status |= IB_SMP_UNSUP_METHOD;
1084 		return reply(smp);
1085 	}
1086 
1087 	for (i = 0; i < ARRAY_SIZE(ibp->sl_to_vl); i += 2, p++) {
1088 		ibp->sl_to_vl[i] = *p >> 4;
1089 		ibp->sl_to_vl[i + 1] = *p & 0xF;
1090 	}
1091 	qib_set_uevent_bits(ppd_from_ibp(to_iport(ibdev, port)),
1092 			    _QIB_EVENT_SL2VL_CHANGE_BIT);
1093 
1094 	return subn_get_sl_to_vl(smp, ibdev, port);
1095 }
1096 
1097 static int subn_get_vl_arb(struct ib_smp *smp, struct ib_device *ibdev,
1098 			   u8 port)
1099 {
1100 	unsigned which = be32_to_cpu(smp->attr_mod) >> 16;
1101 	struct qib_pportdata *ppd = ppd_from_ibp(to_iport(ibdev, port));
1102 
1103 	memset(smp->data, 0, sizeof(smp->data));
1104 
1105 	if (ppd->vls_supported == IB_VL_VL0)
1106 		smp->status |= IB_SMP_UNSUP_METHOD;
1107 	else if (which == IB_VLARB_LOWPRI_0_31)
1108 		(void) ppd->dd->f_get_ib_table(ppd, QIB_IB_TBL_VL_LOW_ARB,
1109 						   smp->data);
1110 	else if (which == IB_VLARB_HIGHPRI_0_31)
1111 		(void) ppd->dd->f_get_ib_table(ppd, QIB_IB_TBL_VL_HIGH_ARB,
1112 						   smp->data);
1113 	else
1114 		smp->status |= IB_SMP_INVALID_FIELD;
1115 
1116 	return reply(smp);
1117 }
1118 
1119 static int subn_set_vl_arb(struct ib_smp *smp, struct ib_device *ibdev,
1120 			   u8 port)
1121 {
1122 	unsigned which = be32_to_cpu(smp->attr_mod) >> 16;
1123 	struct qib_pportdata *ppd = ppd_from_ibp(to_iport(ibdev, port));
1124 
1125 	if (ppd->vls_supported == IB_VL_VL0)
1126 		smp->status |= IB_SMP_UNSUP_METHOD;
1127 	else if (which == IB_VLARB_LOWPRI_0_31)
1128 		(void) ppd->dd->f_set_ib_table(ppd, QIB_IB_TBL_VL_LOW_ARB,
1129 						   smp->data);
1130 	else if (which == IB_VLARB_HIGHPRI_0_31)
1131 		(void) ppd->dd->f_set_ib_table(ppd, QIB_IB_TBL_VL_HIGH_ARB,
1132 						   smp->data);
1133 	else
1134 		smp->status |= IB_SMP_INVALID_FIELD;
1135 
1136 	return subn_get_vl_arb(smp, ibdev, port);
1137 }
1138 
1139 static int subn_trap_repress(struct ib_smp *smp, struct ib_device *ibdev,
1140 			     u8 port)
1141 {
1142 	/*
1143 	 * For now, we only send the trap once so no need to process this.
1144 	 * o13-6, o13-7,
1145 	 * o14-3.a4 The SMA shall not send any message in response to a valid
1146 	 * SubnTrapRepress() message.
1147 	 */
1148 	return IB_MAD_RESULT_SUCCESS | IB_MAD_RESULT_CONSUMED;
1149 }
1150 
1151 static int pma_get_classportinfo(struct ib_pma_mad *pmp,
1152 				 struct ib_device *ibdev)
1153 {
1154 	struct ib_class_port_info *p =
1155 		(struct ib_class_port_info *)pmp->data;
1156 	struct qib_devdata *dd = dd_from_ibdev(ibdev);
1157 
1158 	memset(pmp->data, 0, sizeof(pmp->data));
1159 
1160 	if (pmp->mad_hdr.attr_mod != 0)
1161 		pmp->mad_hdr.status |= IB_SMP_INVALID_FIELD;
1162 
1163 	/* Note that AllPortSelect is not valid */
1164 	p->base_version = 1;
1165 	p->class_version = 1;
1166 	p->capability_mask = IB_PMA_CLASS_CAP_EXT_WIDTH;
1167 	/*
1168 	 * Set the most significant bit of CM2 to indicate support for
1169 	 * congestion statistics
1170 	 */
1171 	ib_set_cpi_capmask2(p,
1172 			    dd->psxmitwait_supported <<
1173 			    (31 - IB_CLASS_PORT_INFO_RESP_TIME_FIELD_SIZE));
1174 	/*
1175 	 * Expected response time is 4.096 usec. * 2^18 == 1.073741824 sec.
1176 	 */
1177 	ib_set_cpi_resp_time(p, 18);
1178 
1179 	return reply((struct ib_smp *) pmp);
1180 }
1181 
1182 static int pma_get_portsamplescontrol(struct ib_pma_mad *pmp,
1183 				      struct ib_device *ibdev, u8 port)
1184 {
1185 	struct ib_pma_portsamplescontrol *p =
1186 		(struct ib_pma_portsamplescontrol *)pmp->data;
1187 	struct qib_ibdev *dev = to_idev(ibdev);
1188 	struct qib_devdata *dd = dd_from_dev(dev);
1189 	struct qib_ibport *ibp = to_iport(ibdev, port);
1190 	struct qib_pportdata *ppd = ppd_from_ibp(ibp);
1191 	unsigned long flags;
1192 	u8 port_select = p->port_select;
1193 
1194 	memset(pmp->data, 0, sizeof(pmp->data));
1195 
1196 	p->port_select = port_select;
1197 	if (pmp->mad_hdr.attr_mod != 0 || port_select != port) {
1198 		pmp->mad_hdr.status |= IB_SMP_INVALID_FIELD;
1199 		goto bail;
1200 	}
1201 	spin_lock_irqsave(&ibp->rvp.lock, flags);
1202 	p->tick = dd->f_get_ib_cfg(ppd, QIB_IB_CFG_PMA_TICKS);
1203 	p->sample_status = dd->f_portcntr(ppd, QIBPORTCNTR_PSSTAT);
1204 	p->counter_width = 4;   /* 32 bit counters */
1205 	p->counter_mask0_9 = COUNTER_MASK0_9;
1206 	p->sample_start = cpu_to_be32(ibp->rvp.pma_sample_start);
1207 	p->sample_interval = cpu_to_be32(ibp->rvp.pma_sample_interval);
1208 	p->tag = cpu_to_be16(ibp->rvp.pma_tag);
1209 	p->counter_select[0] = ibp->rvp.pma_counter_select[0];
1210 	p->counter_select[1] = ibp->rvp.pma_counter_select[1];
1211 	p->counter_select[2] = ibp->rvp.pma_counter_select[2];
1212 	p->counter_select[3] = ibp->rvp.pma_counter_select[3];
1213 	p->counter_select[4] = ibp->rvp.pma_counter_select[4];
1214 	spin_unlock_irqrestore(&ibp->rvp.lock, flags);
1215 
1216 bail:
1217 	return reply((struct ib_smp *) pmp);
1218 }
1219 
1220 static int pma_set_portsamplescontrol(struct ib_pma_mad *pmp,
1221 				      struct ib_device *ibdev, u8 port)
1222 {
1223 	struct ib_pma_portsamplescontrol *p =
1224 		(struct ib_pma_portsamplescontrol *)pmp->data;
1225 	struct qib_ibdev *dev = to_idev(ibdev);
1226 	struct qib_devdata *dd = dd_from_dev(dev);
1227 	struct qib_ibport *ibp = to_iport(ibdev, port);
1228 	struct qib_pportdata *ppd = ppd_from_ibp(ibp);
1229 	unsigned long flags;
1230 	u8 status, xmit_flags;
1231 	int ret;
1232 
1233 	if (pmp->mad_hdr.attr_mod != 0 || p->port_select != port) {
1234 		pmp->mad_hdr.status |= IB_SMP_INVALID_FIELD;
1235 		ret = reply((struct ib_smp *) pmp);
1236 		goto bail;
1237 	}
1238 
1239 	spin_lock_irqsave(&ibp->rvp.lock, flags);
1240 
1241 	/* Port Sampling code owns the PS* HW counters */
1242 	xmit_flags = ppd->cong_stats.flags;
1243 	ppd->cong_stats.flags = IB_PMA_CONG_HW_CONTROL_SAMPLE;
1244 	status = dd->f_portcntr(ppd, QIBPORTCNTR_PSSTAT);
1245 	if (status == IB_PMA_SAMPLE_STATUS_DONE ||
1246 	    (status == IB_PMA_SAMPLE_STATUS_RUNNING &&
1247 	     xmit_flags == IB_PMA_CONG_HW_CONTROL_TIMER)) {
1248 		ibp->rvp.pma_sample_start = be32_to_cpu(p->sample_start);
1249 		ibp->rvp.pma_sample_interval = be32_to_cpu(p->sample_interval);
1250 		ibp->rvp.pma_tag = be16_to_cpu(p->tag);
1251 		ibp->rvp.pma_counter_select[0] = p->counter_select[0];
1252 		ibp->rvp.pma_counter_select[1] = p->counter_select[1];
1253 		ibp->rvp.pma_counter_select[2] = p->counter_select[2];
1254 		ibp->rvp.pma_counter_select[3] = p->counter_select[3];
1255 		ibp->rvp.pma_counter_select[4] = p->counter_select[4];
1256 		dd->f_set_cntr_sample(ppd, ibp->rvp.pma_sample_interval,
1257 				      ibp->rvp.pma_sample_start);
1258 	}
1259 	spin_unlock_irqrestore(&ibp->rvp.lock, flags);
1260 
1261 	ret = pma_get_portsamplescontrol(pmp, ibdev, port);
1262 
1263 bail:
1264 	return ret;
1265 }
1266 
1267 static u64 get_counter(struct qib_ibport *ibp, struct qib_pportdata *ppd,
1268 		       __be16 sel)
1269 {
1270 	u64 ret;
1271 
1272 	switch (sel) {
1273 	case IB_PMA_PORT_XMIT_DATA:
1274 		ret = ppd->dd->f_portcntr(ppd, QIBPORTCNTR_PSXMITDATA);
1275 		break;
1276 	case IB_PMA_PORT_RCV_DATA:
1277 		ret = ppd->dd->f_portcntr(ppd, QIBPORTCNTR_PSRCVDATA);
1278 		break;
1279 	case IB_PMA_PORT_XMIT_PKTS:
1280 		ret = ppd->dd->f_portcntr(ppd, QIBPORTCNTR_PSXMITPKTS);
1281 		break;
1282 	case IB_PMA_PORT_RCV_PKTS:
1283 		ret = ppd->dd->f_portcntr(ppd, QIBPORTCNTR_PSRCVPKTS);
1284 		break;
1285 	case IB_PMA_PORT_XMIT_WAIT:
1286 		ret = ppd->dd->f_portcntr(ppd, QIBPORTCNTR_PSXMITWAIT);
1287 		break;
1288 	default:
1289 		ret = 0;
1290 	}
1291 
1292 	return ret;
1293 }
1294 
1295 /* This function assumes that the xmit_wait lock is already held */
1296 static u64 xmit_wait_get_value_delta(struct qib_pportdata *ppd)
1297 {
1298 	u32 delta;
1299 
1300 	delta = get_counter(&ppd->ibport_data, ppd,
1301 			    IB_PMA_PORT_XMIT_WAIT);
1302 	return ppd->cong_stats.counter + delta;
1303 }
1304 
1305 static void cache_hw_sample_counters(struct qib_pportdata *ppd)
1306 {
1307 	struct qib_ibport *ibp = &ppd->ibport_data;
1308 
1309 	ppd->cong_stats.counter_cache.psxmitdata =
1310 		get_counter(ibp, ppd, IB_PMA_PORT_XMIT_DATA);
1311 	ppd->cong_stats.counter_cache.psrcvdata =
1312 		get_counter(ibp, ppd, IB_PMA_PORT_RCV_DATA);
1313 	ppd->cong_stats.counter_cache.psxmitpkts =
1314 		get_counter(ibp, ppd, IB_PMA_PORT_XMIT_PKTS);
1315 	ppd->cong_stats.counter_cache.psrcvpkts =
1316 		get_counter(ibp, ppd, IB_PMA_PORT_RCV_PKTS);
1317 	ppd->cong_stats.counter_cache.psxmitwait =
1318 		get_counter(ibp, ppd, IB_PMA_PORT_XMIT_WAIT);
1319 }
1320 
1321 static u64 get_cache_hw_sample_counters(struct qib_pportdata *ppd,
1322 					__be16 sel)
1323 {
1324 	u64 ret;
1325 
1326 	switch (sel) {
1327 	case IB_PMA_PORT_XMIT_DATA:
1328 		ret = ppd->cong_stats.counter_cache.psxmitdata;
1329 		break;
1330 	case IB_PMA_PORT_RCV_DATA:
1331 		ret = ppd->cong_stats.counter_cache.psrcvdata;
1332 		break;
1333 	case IB_PMA_PORT_XMIT_PKTS:
1334 		ret = ppd->cong_stats.counter_cache.psxmitpkts;
1335 		break;
1336 	case IB_PMA_PORT_RCV_PKTS:
1337 		ret = ppd->cong_stats.counter_cache.psrcvpkts;
1338 		break;
1339 	case IB_PMA_PORT_XMIT_WAIT:
1340 		ret = ppd->cong_stats.counter_cache.psxmitwait;
1341 		break;
1342 	default:
1343 		ret = 0;
1344 	}
1345 
1346 	return ret;
1347 }
1348 
1349 static int pma_get_portsamplesresult(struct ib_pma_mad *pmp,
1350 				     struct ib_device *ibdev, u8 port)
1351 {
1352 	struct ib_pma_portsamplesresult *p =
1353 		(struct ib_pma_portsamplesresult *)pmp->data;
1354 	struct qib_ibdev *dev = to_idev(ibdev);
1355 	struct qib_devdata *dd = dd_from_dev(dev);
1356 	struct qib_ibport *ibp = to_iport(ibdev, port);
1357 	struct qib_pportdata *ppd = ppd_from_ibp(ibp);
1358 	unsigned long flags;
1359 	u8 status;
1360 	int i;
1361 
1362 	memset(pmp->data, 0, sizeof(pmp->data));
1363 	spin_lock_irqsave(&ibp->rvp.lock, flags);
1364 	p->tag = cpu_to_be16(ibp->rvp.pma_tag);
1365 	if (ppd->cong_stats.flags == IB_PMA_CONG_HW_CONTROL_TIMER)
1366 		p->sample_status = IB_PMA_SAMPLE_STATUS_DONE;
1367 	else {
1368 		status = dd->f_portcntr(ppd, QIBPORTCNTR_PSSTAT);
1369 		p->sample_status = cpu_to_be16(status);
1370 		if (status == IB_PMA_SAMPLE_STATUS_DONE) {
1371 			cache_hw_sample_counters(ppd);
1372 			ppd->cong_stats.counter =
1373 				xmit_wait_get_value_delta(ppd);
1374 			dd->f_set_cntr_sample(ppd,
1375 					      QIB_CONG_TIMER_PSINTERVAL, 0);
1376 			ppd->cong_stats.flags = IB_PMA_CONG_HW_CONTROL_TIMER;
1377 		}
1378 	}
1379 	for (i = 0; i < ARRAY_SIZE(ibp->rvp.pma_counter_select); i++)
1380 		p->counter[i] = cpu_to_be32(
1381 			get_cache_hw_sample_counters(
1382 				ppd, ibp->rvp.pma_counter_select[i]));
1383 	spin_unlock_irqrestore(&ibp->rvp.lock, flags);
1384 
1385 	return reply((struct ib_smp *) pmp);
1386 }
1387 
1388 static int pma_get_portsamplesresult_ext(struct ib_pma_mad *pmp,
1389 					 struct ib_device *ibdev, u8 port)
1390 {
1391 	struct ib_pma_portsamplesresult_ext *p =
1392 		(struct ib_pma_portsamplesresult_ext *)pmp->data;
1393 	struct qib_ibdev *dev = to_idev(ibdev);
1394 	struct qib_devdata *dd = dd_from_dev(dev);
1395 	struct qib_ibport *ibp = to_iport(ibdev, port);
1396 	struct qib_pportdata *ppd = ppd_from_ibp(ibp);
1397 	unsigned long flags;
1398 	u8 status;
1399 	int i;
1400 
1401 	/* Port Sampling code owns the PS* HW counters */
1402 	memset(pmp->data, 0, sizeof(pmp->data));
1403 	spin_lock_irqsave(&ibp->rvp.lock, flags);
1404 	p->tag = cpu_to_be16(ibp->rvp.pma_tag);
1405 	if (ppd->cong_stats.flags == IB_PMA_CONG_HW_CONTROL_TIMER)
1406 		p->sample_status = IB_PMA_SAMPLE_STATUS_DONE;
1407 	else {
1408 		status = dd->f_portcntr(ppd, QIBPORTCNTR_PSSTAT);
1409 		p->sample_status = cpu_to_be16(status);
1410 		/* 64 bits */
1411 		p->extended_width = cpu_to_be32(0x80000000);
1412 		if (status == IB_PMA_SAMPLE_STATUS_DONE) {
1413 			cache_hw_sample_counters(ppd);
1414 			ppd->cong_stats.counter =
1415 				xmit_wait_get_value_delta(ppd);
1416 			dd->f_set_cntr_sample(ppd,
1417 					      QIB_CONG_TIMER_PSINTERVAL, 0);
1418 			ppd->cong_stats.flags = IB_PMA_CONG_HW_CONTROL_TIMER;
1419 		}
1420 	}
1421 	for (i = 0; i < ARRAY_SIZE(ibp->rvp.pma_counter_select); i++)
1422 		p->counter[i] = cpu_to_be64(
1423 			get_cache_hw_sample_counters(
1424 				ppd, ibp->rvp.pma_counter_select[i]));
1425 	spin_unlock_irqrestore(&ibp->rvp.lock, flags);
1426 
1427 	return reply((struct ib_smp *) pmp);
1428 }
1429 
1430 static int pma_get_portcounters(struct ib_pma_mad *pmp,
1431 				struct ib_device *ibdev, u8 port)
1432 {
1433 	struct ib_pma_portcounters *p = (struct ib_pma_portcounters *)
1434 		pmp->data;
1435 	struct qib_ibport *ibp = to_iport(ibdev, port);
1436 	struct qib_pportdata *ppd = ppd_from_ibp(ibp);
1437 	struct qib_verbs_counters cntrs;
1438 	u8 port_select = p->port_select;
1439 
1440 	qib_get_counters(ppd, &cntrs);
1441 
1442 	/* Adjust counters for any resets done. */
1443 	cntrs.symbol_error_counter -= ibp->z_symbol_error_counter;
1444 	cntrs.link_error_recovery_counter -=
1445 		ibp->z_link_error_recovery_counter;
1446 	cntrs.link_downed_counter -= ibp->z_link_downed_counter;
1447 	cntrs.port_rcv_errors -= ibp->z_port_rcv_errors;
1448 	cntrs.port_rcv_remphys_errors -= ibp->z_port_rcv_remphys_errors;
1449 	cntrs.port_xmit_discards -= ibp->z_port_xmit_discards;
1450 	cntrs.port_xmit_data -= ibp->z_port_xmit_data;
1451 	cntrs.port_rcv_data -= ibp->z_port_rcv_data;
1452 	cntrs.port_xmit_packets -= ibp->z_port_xmit_packets;
1453 	cntrs.port_rcv_packets -= ibp->z_port_rcv_packets;
1454 	cntrs.local_link_integrity_errors -=
1455 		ibp->z_local_link_integrity_errors;
1456 	cntrs.excessive_buffer_overrun_errors -=
1457 		ibp->z_excessive_buffer_overrun_errors;
1458 	cntrs.vl15_dropped -= ibp->z_vl15_dropped;
1459 	cntrs.vl15_dropped += ibp->rvp.n_vl15_dropped;
1460 
1461 	memset(pmp->data, 0, sizeof(pmp->data));
1462 
1463 	p->port_select = port_select;
1464 	if (pmp->mad_hdr.attr_mod != 0 || port_select != port)
1465 		pmp->mad_hdr.status |= IB_SMP_INVALID_FIELD;
1466 
1467 	if (cntrs.symbol_error_counter > 0xFFFFUL)
1468 		p->symbol_error_counter = cpu_to_be16(0xFFFF);
1469 	else
1470 		p->symbol_error_counter =
1471 			cpu_to_be16((u16)cntrs.symbol_error_counter);
1472 	if (cntrs.link_error_recovery_counter > 0xFFUL)
1473 		p->link_error_recovery_counter = 0xFF;
1474 	else
1475 		p->link_error_recovery_counter =
1476 			(u8)cntrs.link_error_recovery_counter;
1477 	if (cntrs.link_downed_counter > 0xFFUL)
1478 		p->link_downed_counter = 0xFF;
1479 	else
1480 		p->link_downed_counter = (u8)cntrs.link_downed_counter;
1481 	if (cntrs.port_rcv_errors > 0xFFFFUL)
1482 		p->port_rcv_errors = cpu_to_be16(0xFFFF);
1483 	else
1484 		p->port_rcv_errors =
1485 			cpu_to_be16((u16) cntrs.port_rcv_errors);
1486 	if (cntrs.port_rcv_remphys_errors > 0xFFFFUL)
1487 		p->port_rcv_remphys_errors = cpu_to_be16(0xFFFF);
1488 	else
1489 		p->port_rcv_remphys_errors =
1490 			cpu_to_be16((u16)cntrs.port_rcv_remphys_errors);
1491 	if (cntrs.port_xmit_discards > 0xFFFFUL)
1492 		p->port_xmit_discards = cpu_to_be16(0xFFFF);
1493 	else
1494 		p->port_xmit_discards =
1495 			cpu_to_be16((u16)cntrs.port_xmit_discards);
1496 	if (cntrs.local_link_integrity_errors > 0xFUL)
1497 		cntrs.local_link_integrity_errors = 0xFUL;
1498 	if (cntrs.excessive_buffer_overrun_errors > 0xFUL)
1499 		cntrs.excessive_buffer_overrun_errors = 0xFUL;
1500 	p->link_overrun_errors = (cntrs.local_link_integrity_errors << 4) |
1501 		cntrs.excessive_buffer_overrun_errors;
1502 	if (cntrs.vl15_dropped > 0xFFFFUL)
1503 		p->vl15_dropped = cpu_to_be16(0xFFFF);
1504 	else
1505 		p->vl15_dropped = cpu_to_be16((u16)cntrs.vl15_dropped);
1506 	if (cntrs.port_xmit_data > 0xFFFFFFFFUL)
1507 		p->port_xmit_data = cpu_to_be32(0xFFFFFFFF);
1508 	else
1509 		p->port_xmit_data = cpu_to_be32((u32)cntrs.port_xmit_data);
1510 	if (cntrs.port_rcv_data > 0xFFFFFFFFUL)
1511 		p->port_rcv_data = cpu_to_be32(0xFFFFFFFF);
1512 	else
1513 		p->port_rcv_data = cpu_to_be32((u32)cntrs.port_rcv_data);
1514 	if (cntrs.port_xmit_packets > 0xFFFFFFFFUL)
1515 		p->port_xmit_packets = cpu_to_be32(0xFFFFFFFF);
1516 	else
1517 		p->port_xmit_packets =
1518 			cpu_to_be32((u32)cntrs.port_xmit_packets);
1519 	if (cntrs.port_rcv_packets > 0xFFFFFFFFUL)
1520 		p->port_rcv_packets = cpu_to_be32(0xFFFFFFFF);
1521 	else
1522 		p->port_rcv_packets =
1523 			cpu_to_be32((u32) cntrs.port_rcv_packets);
1524 
1525 	return reply((struct ib_smp *) pmp);
1526 }
1527 
1528 static int pma_get_portcounters_cong(struct ib_pma_mad *pmp,
1529 				     struct ib_device *ibdev, u8 port)
1530 {
1531 	/* Congestion PMA packets start at offset 24 not 64 */
1532 	struct ib_pma_portcounters_cong *p =
1533 		(struct ib_pma_portcounters_cong *)pmp->reserved;
1534 	struct qib_verbs_counters cntrs;
1535 	struct qib_ibport *ibp = to_iport(ibdev, port);
1536 	struct qib_pportdata *ppd = ppd_from_ibp(ibp);
1537 	struct qib_devdata *dd = dd_from_ppd(ppd);
1538 	u32 port_select = be32_to_cpu(pmp->mad_hdr.attr_mod) & 0xFF;
1539 	u64 xmit_wait_counter;
1540 	unsigned long flags;
1541 
1542 	/*
1543 	 * This check is performed only in the GET method because the
1544 	 * SET method ends up calling this anyway.
1545 	 */
1546 	if (!dd->psxmitwait_supported)
1547 		pmp->mad_hdr.status |= IB_SMP_UNSUP_METH_ATTR;
1548 	if (port_select != port)
1549 		pmp->mad_hdr.status |= IB_SMP_INVALID_FIELD;
1550 
1551 	qib_get_counters(ppd, &cntrs);
1552 	spin_lock_irqsave(&ppd->ibport_data.rvp.lock, flags);
1553 	xmit_wait_counter = xmit_wait_get_value_delta(ppd);
1554 	spin_unlock_irqrestore(&ppd->ibport_data.rvp.lock, flags);
1555 
1556 	/* Adjust counters for any resets done. */
1557 	cntrs.symbol_error_counter -= ibp->z_symbol_error_counter;
1558 	cntrs.link_error_recovery_counter -=
1559 		ibp->z_link_error_recovery_counter;
1560 	cntrs.link_downed_counter -= ibp->z_link_downed_counter;
1561 	cntrs.port_rcv_errors -= ibp->z_port_rcv_errors;
1562 	cntrs.port_rcv_remphys_errors -=
1563 		ibp->z_port_rcv_remphys_errors;
1564 	cntrs.port_xmit_discards -= ibp->z_port_xmit_discards;
1565 	cntrs.local_link_integrity_errors -=
1566 		ibp->z_local_link_integrity_errors;
1567 	cntrs.excessive_buffer_overrun_errors -=
1568 		ibp->z_excessive_buffer_overrun_errors;
1569 	cntrs.vl15_dropped -= ibp->z_vl15_dropped;
1570 	cntrs.vl15_dropped += ibp->rvp.n_vl15_dropped;
1571 	cntrs.port_xmit_data -= ibp->z_port_xmit_data;
1572 	cntrs.port_rcv_data -= ibp->z_port_rcv_data;
1573 	cntrs.port_xmit_packets -= ibp->z_port_xmit_packets;
1574 	cntrs.port_rcv_packets -= ibp->z_port_rcv_packets;
1575 
1576 	memset(pmp->reserved, 0, sizeof(pmp->reserved));
1577 	memset(pmp->data, 0, sizeof(pmp->data));
1578 
1579 	/*
1580 	 * Set top 3 bits to indicate interval in picoseconds in
1581 	 * remaining bits.
1582 	 */
1583 	p->port_check_rate =
1584 		cpu_to_be16((QIB_XMIT_RATE_PICO << 13) |
1585 			    (dd->psxmitwait_check_rate &
1586 			     ~(QIB_XMIT_RATE_PICO << 13)));
1587 	p->port_adr_events = cpu_to_be64(0);
1588 	p->port_xmit_wait = cpu_to_be64(xmit_wait_counter);
1589 	p->port_xmit_data = cpu_to_be64(cntrs.port_xmit_data);
1590 	p->port_rcv_data = cpu_to_be64(cntrs.port_rcv_data);
1591 	p->port_xmit_packets =
1592 		cpu_to_be64(cntrs.port_xmit_packets);
1593 	p->port_rcv_packets =
1594 		cpu_to_be64(cntrs.port_rcv_packets);
1595 	if (cntrs.symbol_error_counter > 0xFFFFUL)
1596 		p->symbol_error_counter = cpu_to_be16(0xFFFF);
1597 	else
1598 		p->symbol_error_counter =
1599 			cpu_to_be16(
1600 				(u16)cntrs.symbol_error_counter);
1601 	if (cntrs.link_error_recovery_counter > 0xFFUL)
1602 		p->link_error_recovery_counter = 0xFF;
1603 	else
1604 		p->link_error_recovery_counter =
1605 			(u8)cntrs.link_error_recovery_counter;
1606 	if (cntrs.link_downed_counter > 0xFFUL)
1607 		p->link_downed_counter = 0xFF;
1608 	else
1609 		p->link_downed_counter =
1610 			(u8)cntrs.link_downed_counter;
1611 	if (cntrs.port_rcv_errors > 0xFFFFUL)
1612 		p->port_rcv_errors = cpu_to_be16(0xFFFF);
1613 	else
1614 		p->port_rcv_errors =
1615 			cpu_to_be16((u16) cntrs.port_rcv_errors);
1616 	if (cntrs.port_rcv_remphys_errors > 0xFFFFUL)
1617 		p->port_rcv_remphys_errors = cpu_to_be16(0xFFFF);
1618 	else
1619 		p->port_rcv_remphys_errors =
1620 			cpu_to_be16(
1621 				(u16)cntrs.port_rcv_remphys_errors);
1622 	if (cntrs.port_xmit_discards > 0xFFFFUL)
1623 		p->port_xmit_discards = cpu_to_be16(0xFFFF);
1624 	else
1625 		p->port_xmit_discards =
1626 			cpu_to_be16((u16)cntrs.port_xmit_discards);
1627 	if (cntrs.local_link_integrity_errors > 0xFUL)
1628 		cntrs.local_link_integrity_errors = 0xFUL;
1629 	if (cntrs.excessive_buffer_overrun_errors > 0xFUL)
1630 		cntrs.excessive_buffer_overrun_errors = 0xFUL;
1631 	p->link_overrun_errors = (cntrs.local_link_integrity_errors << 4) |
1632 		cntrs.excessive_buffer_overrun_errors;
1633 	if (cntrs.vl15_dropped > 0xFFFFUL)
1634 		p->vl15_dropped = cpu_to_be16(0xFFFF);
1635 	else
1636 		p->vl15_dropped = cpu_to_be16((u16)cntrs.vl15_dropped);
1637 
1638 	return reply((struct ib_smp *)pmp);
1639 }
1640 
1641 static void qib_snapshot_pmacounters(
1642 	struct qib_ibport *ibp,
1643 	struct qib_pma_counters *pmacounters)
1644 {
1645 	struct qib_pma_counters *p;
1646 	int cpu;
1647 
1648 	memset(pmacounters, 0, sizeof(*pmacounters));
1649 	for_each_possible_cpu(cpu) {
1650 		p = per_cpu_ptr(ibp->pmastats, cpu);
1651 		pmacounters->n_unicast_xmit += p->n_unicast_xmit;
1652 		pmacounters->n_unicast_rcv += p->n_unicast_rcv;
1653 		pmacounters->n_multicast_xmit += p->n_multicast_xmit;
1654 		pmacounters->n_multicast_rcv += p->n_multicast_rcv;
1655 	}
1656 }
1657 
1658 static int pma_get_portcounters_ext(struct ib_pma_mad *pmp,
1659 				    struct ib_device *ibdev, u8 port)
1660 {
1661 	struct ib_pma_portcounters_ext *p =
1662 		(struct ib_pma_portcounters_ext *)pmp->data;
1663 	struct qib_ibport *ibp = to_iport(ibdev, port);
1664 	struct qib_pportdata *ppd = ppd_from_ibp(ibp);
1665 	u64 swords, rwords, spkts, rpkts, xwait;
1666 	struct qib_pma_counters pma;
1667 	u8 port_select = p->port_select;
1668 
1669 	memset(pmp->data, 0, sizeof(pmp->data));
1670 
1671 	p->port_select = port_select;
1672 	if (pmp->mad_hdr.attr_mod != 0 || port_select != port) {
1673 		pmp->mad_hdr.status |= IB_SMP_INVALID_FIELD;
1674 		goto bail;
1675 	}
1676 
1677 	qib_snapshot_counters(ppd, &swords, &rwords, &spkts, &rpkts, &xwait);
1678 
1679 	/* Adjust counters for any resets done. */
1680 	swords -= ibp->z_port_xmit_data;
1681 	rwords -= ibp->z_port_rcv_data;
1682 	spkts -= ibp->z_port_xmit_packets;
1683 	rpkts -= ibp->z_port_rcv_packets;
1684 
1685 	p->port_xmit_data = cpu_to_be64(swords);
1686 	p->port_rcv_data = cpu_to_be64(rwords);
1687 	p->port_xmit_packets = cpu_to_be64(spkts);
1688 	p->port_rcv_packets = cpu_to_be64(rpkts);
1689 
1690 	qib_snapshot_pmacounters(ibp, &pma);
1691 
1692 	p->port_unicast_xmit_packets = cpu_to_be64(pma.n_unicast_xmit
1693 		- ibp->z_unicast_xmit);
1694 	p->port_unicast_rcv_packets = cpu_to_be64(pma.n_unicast_rcv
1695 		- ibp->z_unicast_rcv);
1696 	p->port_multicast_xmit_packets = cpu_to_be64(pma.n_multicast_xmit
1697 		- ibp->z_multicast_xmit);
1698 	p->port_multicast_rcv_packets = cpu_to_be64(pma.n_multicast_rcv
1699 		- ibp->z_multicast_rcv);
1700 
1701 bail:
1702 	return reply((struct ib_smp *) pmp);
1703 }
1704 
1705 static int pma_set_portcounters(struct ib_pma_mad *pmp,
1706 				struct ib_device *ibdev, u8 port)
1707 {
1708 	struct ib_pma_portcounters *p = (struct ib_pma_portcounters *)
1709 		pmp->data;
1710 	struct qib_ibport *ibp = to_iport(ibdev, port);
1711 	struct qib_pportdata *ppd = ppd_from_ibp(ibp);
1712 	struct qib_verbs_counters cntrs;
1713 
1714 	/*
1715 	 * Since the HW doesn't support clearing counters, we save the
1716 	 * current count and subtract it from future responses.
1717 	 */
1718 	qib_get_counters(ppd, &cntrs);
1719 
1720 	if (p->counter_select & IB_PMA_SEL_SYMBOL_ERROR)
1721 		ibp->z_symbol_error_counter = cntrs.symbol_error_counter;
1722 
1723 	if (p->counter_select & IB_PMA_SEL_LINK_ERROR_RECOVERY)
1724 		ibp->z_link_error_recovery_counter =
1725 			cntrs.link_error_recovery_counter;
1726 
1727 	if (p->counter_select & IB_PMA_SEL_LINK_DOWNED)
1728 		ibp->z_link_downed_counter = cntrs.link_downed_counter;
1729 
1730 	if (p->counter_select & IB_PMA_SEL_PORT_RCV_ERRORS)
1731 		ibp->z_port_rcv_errors = cntrs.port_rcv_errors;
1732 
1733 	if (p->counter_select & IB_PMA_SEL_PORT_RCV_REMPHYS_ERRORS)
1734 		ibp->z_port_rcv_remphys_errors =
1735 			cntrs.port_rcv_remphys_errors;
1736 
1737 	if (p->counter_select & IB_PMA_SEL_PORT_XMIT_DISCARDS)
1738 		ibp->z_port_xmit_discards = cntrs.port_xmit_discards;
1739 
1740 	if (p->counter_select & IB_PMA_SEL_LOCAL_LINK_INTEGRITY_ERRORS)
1741 		ibp->z_local_link_integrity_errors =
1742 			cntrs.local_link_integrity_errors;
1743 
1744 	if (p->counter_select & IB_PMA_SEL_EXCESSIVE_BUFFER_OVERRUNS)
1745 		ibp->z_excessive_buffer_overrun_errors =
1746 			cntrs.excessive_buffer_overrun_errors;
1747 
1748 	if (p->counter_select & IB_PMA_SEL_PORT_VL15_DROPPED) {
1749 		ibp->rvp.n_vl15_dropped = 0;
1750 		ibp->z_vl15_dropped = cntrs.vl15_dropped;
1751 	}
1752 
1753 	if (p->counter_select & IB_PMA_SEL_PORT_XMIT_DATA)
1754 		ibp->z_port_xmit_data = cntrs.port_xmit_data;
1755 
1756 	if (p->counter_select & IB_PMA_SEL_PORT_RCV_DATA)
1757 		ibp->z_port_rcv_data = cntrs.port_rcv_data;
1758 
1759 	if (p->counter_select & IB_PMA_SEL_PORT_XMIT_PACKETS)
1760 		ibp->z_port_xmit_packets = cntrs.port_xmit_packets;
1761 
1762 	if (p->counter_select & IB_PMA_SEL_PORT_RCV_PACKETS)
1763 		ibp->z_port_rcv_packets = cntrs.port_rcv_packets;
1764 
1765 	return pma_get_portcounters(pmp, ibdev, port);
1766 }
1767 
1768 static int pma_set_portcounters_cong(struct ib_pma_mad *pmp,
1769 				     struct ib_device *ibdev, u8 port)
1770 {
1771 	struct qib_ibport *ibp = to_iport(ibdev, port);
1772 	struct qib_pportdata *ppd = ppd_from_ibp(ibp);
1773 	struct qib_devdata *dd = dd_from_ppd(ppd);
1774 	struct qib_verbs_counters cntrs;
1775 	u32 counter_select = (be32_to_cpu(pmp->mad_hdr.attr_mod) >> 24) & 0xFF;
1776 	int ret = 0;
1777 	unsigned long flags;
1778 
1779 	qib_get_counters(ppd, &cntrs);
1780 	/* Get counter values before we save them */
1781 	ret = pma_get_portcounters_cong(pmp, ibdev, port);
1782 
1783 	if (counter_select & IB_PMA_SEL_CONG_XMIT) {
1784 		spin_lock_irqsave(&ppd->ibport_data.rvp.lock, flags);
1785 		ppd->cong_stats.counter = 0;
1786 		dd->f_set_cntr_sample(ppd, QIB_CONG_TIMER_PSINTERVAL,
1787 				      0x0);
1788 		spin_unlock_irqrestore(&ppd->ibport_data.rvp.lock, flags);
1789 	}
1790 	if (counter_select & IB_PMA_SEL_CONG_PORT_DATA) {
1791 		ibp->z_port_xmit_data = cntrs.port_xmit_data;
1792 		ibp->z_port_rcv_data = cntrs.port_rcv_data;
1793 		ibp->z_port_xmit_packets = cntrs.port_xmit_packets;
1794 		ibp->z_port_rcv_packets = cntrs.port_rcv_packets;
1795 	}
1796 	if (counter_select & IB_PMA_SEL_CONG_ALL) {
1797 		ibp->z_symbol_error_counter =
1798 			cntrs.symbol_error_counter;
1799 		ibp->z_link_error_recovery_counter =
1800 			cntrs.link_error_recovery_counter;
1801 		ibp->z_link_downed_counter =
1802 			cntrs.link_downed_counter;
1803 		ibp->z_port_rcv_errors = cntrs.port_rcv_errors;
1804 		ibp->z_port_rcv_remphys_errors =
1805 			cntrs.port_rcv_remphys_errors;
1806 		ibp->z_port_xmit_discards =
1807 			cntrs.port_xmit_discards;
1808 		ibp->z_local_link_integrity_errors =
1809 			cntrs.local_link_integrity_errors;
1810 		ibp->z_excessive_buffer_overrun_errors =
1811 			cntrs.excessive_buffer_overrun_errors;
1812 		ibp->rvp.n_vl15_dropped = 0;
1813 		ibp->z_vl15_dropped = cntrs.vl15_dropped;
1814 	}
1815 
1816 	return ret;
1817 }
1818 
1819 static int pma_set_portcounters_ext(struct ib_pma_mad *pmp,
1820 				    struct ib_device *ibdev, u8 port)
1821 {
1822 	struct ib_pma_portcounters *p = (struct ib_pma_portcounters *)
1823 		pmp->data;
1824 	struct qib_ibport *ibp = to_iport(ibdev, port);
1825 	struct qib_pportdata *ppd = ppd_from_ibp(ibp);
1826 	u64 swords, rwords, spkts, rpkts, xwait;
1827 	struct qib_pma_counters pma;
1828 
1829 	qib_snapshot_counters(ppd, &swords, &rwords, &spkts, &rpkts, &xwait);
1830 
1831 	if (p->counter_select & IB_PMA_SELX_PORT_XMIT_DATA)
1832 		ibp->z_port_xmit_data = swords;
1833 
1834 	if (p->counter_select & IB_PMA_SELX_PORT_RCV_DATA)
1835 		ibp->z_port_rcv_data = rwords;
1836 
1837 	if (p->counter_select & IB_PMA_SELX_PORT_XMIT_PACKETS)
1838 		ibp->z_port_xmit_packets = spkts;
1839 
1840 	if (p->counter_select & IB_PMA_SELX_PORT_RCV_PACKETS)
1841 		ibp->z_port_rcv_packets = rpkts;
1842 
1843 	qib_snapshot_pmacounters(ibp, &pma);
1844 
1845 	if (p->counter_select & IB_PMA_SELX_PORT_UNI_XMIT_PACKETS)
1846 		ibp->z_unicast_xmit = pma.n_unicast_xmit;
1847 
1848 	if (p->counter_select & IB_PMA_SELX_PORT_UNI_RCV_PACKETS)
1849 		ibp->z_unicast_rcv = pma.n_unicast_rcv;
1850 
1851 	if (p->counter_select & IB_PMA_SELX_PORT_MULTI_XMIT_PACKETS)
1852 		ibp->z_multicast_xmit = pma.n_multicast_xmit;
1853 
1854 	if (p->counter_select & IB_PMA_SELX_PORT_MULTI_RCV_PACKETS)
1855 		ibp->z_multicast_rcv = pma.n_multicast_rcv;
1856 
1857 	return pma_get_portcounters_ext(pmp, ibdev, port);
1858 }
1859 
1860 static int process_subn(struct ib_device *ibdev, int mad_flags,
1861 			u8 port, const struct ib_mad *in_mad,
1862 			struct ib_mad *out_mad)
1863 {
1864 	struct ib_smp *smp = (struct ib_smp *)out_mad;
1865 	struct qib_ibport *ibp = to_iport(ibdev, port);
1866 	struct qib_pportdata *ppd = ppd_from_ibp(ibp);
1867 	int ret;
1868 
1869 	*out_mad = *in_mad;
1870 	if (smp->class_version != 1) {
1871 		smp->status |= IB_SMP_UNSUP_VERSION;
1872 		ret = reply(smp);
1873 		goto bail;
1874 	}
1875 
1876 	ret = check_mkey(ibp, smp, mad_flags);
1877 	if (ret) {
1878 		u32 port_num = be32_to_cpu(smp->attr_mod);
1879 
1880 		/*
1881 		 * If this is a get/set portinfo, we already check the
1882 		 * M_Key if the MAD is for another port and the M_Key
1883 		 * is OK on the receiving port. This check is needed
1884 		 * to increment the error counters when the M_Key
1885 		 * fails to match on *both* ports.
1886 		 */
1887 		if (in_mad->mad_hdr.attr_id == IB_SMP_ATTR_PORT_INFO &&
1888 		    (smp->method == IB_MGMT_METHOD_GET ||
1889 		     smp->method == IB_MGMT_METHOD_SET) &&
1890 		    port_num && port_num <= ibdev->phys_port_cnt &&
1891 		    port != port_num)
1892 			(void) check_mkey(to_iport(ibdev, port_num), smp, 0);
1893 		ret = IB_MAD_RESULT_FAILURE;
1894 		goto bail;
1895 	}
1896 
1897 	switch (smp->method) {
1898 	case IB_MGMT_METHOD_GET:
1899 		switch (smp->attr_id) {
1900 		case IB_SMP_ATTR_NODE_DESC:
1901 			ret = subn_get_nodedescription(smp, ibdev);
1902 			goto bail;
1903 		case IB_SMP_ATTR_NODE_INFO:
1904 			ret = subn_get_nodeinfo(smp, ibdev, port);
1905 			goto bail;
1906 		case IB_SMP_ATTR_GUID_INFO:
1907 			ret = subn_get_guidinfo(smp, ibdev, port);
1908 			goto bail;
1909 		case IB_SMP_ATTR_PORT_INFO:
1910 			ret = subn_get_portinfo(smp, ibdev, port);
1911 			goto bail;
1912 		case IB_SMP_ATTR_PKEY_TABLE:
1913 			ret = subn_get_pkeytable(smp, ibdev, port);
1914 			goto bail;
1915 		case IB_SMP_ATTR_SL_TO_VL_TABLE:
1916 			ret = subn_get_sl_to_vl(smp, ibdev, port);
1917 			goto bail;
1918 		case IB_SMP_ATTR_VL_ARB_TABLE:
1919 			ret = subn_get_vl_arb(smp, ibdev, port);
1920 			goto bail;
1921 		case IB_SMP_ATTR_SM_INFO:
1922 			if (ibp->rvp.port_cap_flags & IB_PORT_SM_DISABLED) {
1923 				ret = IB_MAD_RESULT_SUCCESS |
1924 					IB_MAD_RESULT_CONSUMED;
1925 				goto bail;
1926 			}
1927 			if (ibp->rvp.port_cap_flags & IB_PORT_SM) {
1928 				ret = IB_MAD_RESULT_SUCCESS;
1929 				goto bail;
1930 			}
1931 			/* FALLTHROUGH */
1932 		default:
1933 			smp->status |= IB_SMP_UNSUP_METH_ATTR;
1934 			ret = reply(smp);
1935 			goto bail;
1936 		}
1937 
1938 	case IB_MGMT_METHOD_SET:
1939 		switch (smp->attr_id) {
1940 		case IB_SMP_ATTR_GUID_INFO:
1941 			ret = subn_set_guidinfo(smp, ibdev, port);
1942 			goto bail;
1943 		case IB_SMP_ATTR_PORT_INFO:
1944 			ret = subn_set_portinfo(smp, ibdev, port);
1945 			goto bail;
1946 		case IB_SMP_ATTR_PKEY_TABLE:
1947 			ret = subn_set_pkeytable(smp, ibdev, port);
1948 			goto bail;
1949 		case IB_SMP_ATTR_SL_TO_VL_TABLE:
1950 			ret = subn_set_sl_to_vl(smp, ibdev, port);
1951 			goto bail;
1952 		case IB_SMP_ATTR_VL_ARB_TABLE:
1953 			ret = subn_set_vl_arb(smp, ibdev, port);
1954 			goto bail;
1955 		case IB_SMP_ATTR_SM_INFO:
1956 			if (ibp->rvp.port_cap_flags & IB_PORT_SM_DISABLED) {
1957 				ret = IB_MAD_RESULT_SUCCESS |
1958 					IB_MAD_RESULT_CONSUMED;
1959 				goto bail;
1960 			}
1961 			if (ibp->rvp.port_cap_flags & IB_PORT_SM) {
1962 				ret = IB_MAD_RESULT_SUCCESS;
1963 				goto bail;
1964 			}
1965 			/* FALLTHROUGH */
1966 		default:
1967 			smp->status |= IB_SMP_UNSUP_METH_ATTR;
1968 			ret = reply(smp);
1969 			goto bail;
1970 		}
1971 
1972 	case IB_MGMT_METHOD_TRAP_REPRESS:
1973 		if (smp->attr_id == IB_SMP_ATTR_NOTICE)
1974 			ret = subn_trap_repress(smp, ibdev, port);
1975 		else {
1976 			smp->status |= IB_SMP_UNSUP_METH_ATTR;
1977 			ret = reply(smp);
1978 		}
1979 		goto bail;
1980 
1981 	case IB_MGMT_METHOD_TRAP:
1982 	case IB_MGMT_METHOD_REPORT:
1983 	case IB_MGMT_METHOD_REPORT_RESP:
1984 	case IB_MGMT_METHOD_GET_RESP:
1985 		/*
1986 		 * The ib_mad module will call us to process responses
1987 		 * before checking for other consumers.
1988 		 * Just tell the caller to process it normally.
1989 		 */
1990 		ret = IB_MAD_RESULT_SUCCESS;
1991 		goto bail;
1992 
1993 	case IB_MGMT_METHOD_SEND:
1994 		if (ib_get_smp_direction(smp) &&
1995 		    smp->attr_id == QIB_VENDOR_IPG) {
1996 			ppd->dd->f_set_ib_cfg(ppd, QIB_IB_CFG_PORT,
1997 					      smp->data[0]);
1998 			ret = IB_MAD_RESULT_SUCCESS | IB_MAD_RESULT_CONSUMED;
1999 		} else
2000 			ret = IB_MAD_RESULT_SUCCESS;
2001 		goto bail;
2002 
2003 	default:
2004 		smp->status |= IB_SMP_UNSUP_METHOD;
2005 		ret = reply(smp);
2006 	}
2007 
2008 bail:
2009 	return ret;
2010 }
2011 
2012 static int process_perf(struct ib_device *ibdev, u8 port,
2013 			const struct ib_mad *in_mad,
2014 			struct ib_mad *out_mad)
2015 {
2016 	struct ib_pma_mad *pmp = (struct ib_pma_mad *)out_mad;
2017 	int ret;
2018 
2019 	*out_mad = *in_mad;
2020 	if (pmp->mad_hdr.class_version != 1) {
2021 		pmp->mad_hdr.status |= IB_SMP_UNSUP_VERSION;
2022 		ret = reply((struct ib_smp *) pmp);
2023 		goto bail;
2024 	}
2025 
2026 	switch (pmp->mad_hdr.method) {
2027 	case IB_MGMT_METHOD_GET:
2028 		switch (pmp->mad_hdr.attr_id) {
2029 		case IB_PMA_CLASS_PORT_INFO:
2030 			ret = pma_get_classportinfo(pmp, ibdev);
2031 			goto bail;
2032 		case IB_PMA_PORT_SAMPLES_CONTROL:
2033 			ret = pma_get_portsamplescontrol(pmp, ibdev, port);
2034 			goto bail;
2035 		case IB_PMA_PORT_SAMPLES_RESULT:
2036 			ret = pma_get_portsamplesresult(pmp, ibdev, port);
2037 			goto bail;
2038 		case IB_PMA_PORT_SAMPLES_RESULT_EXT:
2039 			ret = pma_get_portsamplesresult_ext(pmp, ibdev, port);
2040 			goto bail;
2041 		case IB_PMA_PORT_COUNTERS:
2042 			ret = pma_get_portcounters(pmp, ibdev, port);
2043 			goto bail;
2044 		case IB_PMA_PORT_COUNTERS_EXT:
2045 			ret = pma_get_portcounters_ext(pmp, ibdev, port);
2046 			goto bail;
2047 		case IB_PMA_PORT_COUNTERS_CONG:
2048 			ret = pma_get_portcounters_cong(pmp, ibdev, port);
2049 			goto bail;
2050 		default:
2051 			pmp->mad_hdr.status |= IB_SMP_UNSUP_METH_ATTR;
2052 			ret = reply((struct ib_smp *) pmp);
2053 			goto bail;
2054 		}
2055 
2056 	case IB_MGMT_METHOD_SET:
2057 		switch (pmp->mad_hdr.attr_id) {
2058 		case IB_PMA_PORT_SAMPLES_CONTROL:
2059 			ret = pma_set_portsamplescontrol(pmp, ibdev, port);
2060 			goto bail;
2061 		case IB_PMA_PORT_COUNTERS:
2062 			ret = pma_set_portcounters(pmp, ibdev, port);
2063 			goto bail;
2064 		case IB_PMA_PORT_COUNTERS_EXT:
2065 			ret = pma_set_portcounters_ext(pmp, ibdev, port);
2066 			goto bail;
2067 		case IB_PMA_PORT_COUNTERS_CONG:
2068 			ret = pma_set_portcounters_cong(pmp, ibdev, port);
2069 			goto bail;
2070 		default:
2071 			pmp->mad_hdr.status |= IB_SMP_UNSUP_METH_ATTR;
2072 			ret = reply((struct ib_smp *) pmp);
2073 			goto bail;
2074 		}
2075 
2076 	case IB_MGMT_METHOD_TRAP:
2077 	case IB_MGMT_METHOD_GET_RESP:
2078 		/*
2079 		 * The ib_mad module will call us to process responses
2080 		 * before checking for other consumers.
2081 		 * Just tell the caller to process it normally.
2082 		 */
2083 		ret = IB_MAD_RESULT_SUCCESS;
2084 		goto bail;
2085 
2086 	default:
2087 		pmp->mad_hdr.status |= IB_SMP_UNSUP_METHOD;
2088 		ret = reply((struct ib_smp *) pmp);
2089 	}
2090 
2091 bail:
2092 	return ret;
2093 }
2094 
2095 static int cc_get_classportinfo(struct ib_cc_mad *ccp,
2096 				struct ib_device *ibdev)
2097 {
2098 	struct ib_cc_classportinfo_attr *p =
2099 		(struct ib_cc_classportinfo_attr *)ccp->mgmt_data;
2100 
2101 	p->base_version = 1;
2102 	p->class_version = 1;
2103 	p->cap_mask = 0;
2104 
2105 	/*
2106 	 * Expected response time is 4.096 usec. * 2^18 == 1.073741824 sec.
2107 	 */
2108 	p->resp_time_value = 18;
2109 
2110 	return reply((struct ib_smp *) ccp);
2111 }
2112 
2113 static int cc_get_congestion_info(struct ib_cc_mad *ccp,
2114 				struct ib_device *ibdev, u8 port)
2115 {
2116 	struct ib_cc_info_attr *p =
2117 		(struct ib_cc_info_attr *)ccp->mgmt_data;
2118 	struct qib_ibport *ibp = to_iport(ibdev, port);
2119 	struct qib_pportdata *ppd = ppd_from_ibp(ibp);
2120 
2121 	p->congestion_info = 0;
2122 	p->control_table_cap = ppd->cc_max_table_entries;
2123 
2124 	return reply((struct ib_smp *) ccp);
2125 }
2126 
2127 static int cc_get_congestion_setting(struct ib_cc_mad *ccp,
2128 				struct ib_device *ibdev, u8 port)
2129 {
2130 	int i;
2131 	struct ib_cc_congestion_setting_attr *p =
2132 		(struct ib_cc_congestion_setting_attr *)ccp->mgmt_data;
2133 	struct qib_ibport *ibp = to_iport(ibdev, port);
2134 	struct qib_pportdata *ppd = ppd_from_ibp(ibp);
2135 	struct ib_cc_congestion_entry_shadow *entries;
2136 
2137 	spin_lock(&ppd->cc_shadow_lock);
2138 
2139 	entries = ppd->congestion_entries_shadow->entries;
2140 	p->port_control = cpu_to_be16(
2141 		ppd->congestion_entries_shadow->port_control);
2142 	p->control_map = cpu_to_be16(
2143 		ppd->congestion_entries_shadow->control_map);
2144 	for (i = 0; i < IB_CC_CCS_ENTRIES; i++) {
2145 		p->entries[i].ccti_increase = entries[i].ccti_increase;
2146 		p->entries[i].ccti_timer = cpu_to_be16(entries[i].ccti_timer);
2147 		p->entries[i].trigger_threshold = entries[i].trigger_threshold;
2148 		p->entries[i].ccti_min = entries[i].ccti_min;
2149 	}
2150 
2151 	spin_unlock(&ppd->cc_shadow_lock);
2152 
2153 	return reply((struct ib_smp *) ccp);
2154 }
2155 
2156 static int cc_get_congestion_control_table(struct ib_cc_mad *ccp,
2157 				struct ib_device *ibdev, u8 port)
2158 {
2159 	struct ib_cc_table_attr *p =
2160 		(struct ib_cc_table_attr *)ccp->mgmt_data;
2161 	struct qib_ibport *ibp = to_iport(ibdev, port);
2162 	struct qib_pportdata *ppd = ppd_from_ibp(ibp);
2163 	u32 cct_block_index = be32_to_cpu(ccp->attr_mod);
2164 	u32 max_cct_block;
2165 	u32 cct_entry;
2166 	struct ib_cc_table_entry_shadow *entries;
2167 	int i;
2168 
2169 	/* Is the table index more than what is supported? */
2170 	if (cct_block_index > IB_CC_TABLE_CAP_DEFAULT - 1)
2171 		goto bail;
2172 
2173 	spin_lock(&ppd->cc_shadow_lock);
2174 
2175 	max_cct_block =
2176 		(ppd->ccti_entries_shadow->ccti_last_entry + 1)/IB_CCT_ENTRIES;
2177 	max_cct_block = max_cct_block ? max_cct_block - 1 : 0;
2178 
2179 	if (cct_block_index > max_cct_block) {
2180 		spin_unlock(&ppd->cc_shadow_lock);
2181 		goto bail;
2182 	}
2183 
2184 	ccp->attr_mod = cpu_to_be32(cct_block_index);
2185 
2186 	cct_entry = IB_CCT_ENTRIES * (cct_block_index + 1);
2187 
2188 	cct_entry--;
2189 
2190 	p->ccti_limit = cpu_to_be16(cct_entry);
2191 
2192 	entries = &ppd->ccti_entries_shadow->
2193 			entries[IB_CCT_ENTRIES * cct_block_index];
2194 	cct_entry %= IB_CCT_ENTRIES;
2195 
2196 	for (i = 0; i <= cct_entry; i++)
2197 		p->ccti_entries[i].entry = cpu_to_be16(entries[i].entry);
2198 
2199 	spin_unlock(&ppd->cc_shadow_lock);
2200 
2201 	return reply((struct ib_smp *) ccp);
2202 
2203 bail:
2204 	return reply_failure((struct ib_smp *) ccp);
2205 }
2206 
2207 static int cc_set_congestion_setting(struct ib_cc_mad *ccp,
2208 				struct ib_device *ibdev, u8 port)
2209 {
2210 	struct ib_cc_congestion_setting_attr *p =
2211 		(struct ib_cc_congestion_setting_attr *)ccp->mgmt_data;
2212 	struct qib_ibport *ibp = to_iport(ibdev, port);
2213 	struct qib_pportdata *ppd = ppd_from_ibp(ibp);
2214 	int i;
2215 
2216 	ppd->cc_sl_control_map = be16_to_cpu(p->control_map);
2217 
2218 	for (i = 0; i < IB_CC_CCS_ENTRIES; i++) {
2219 		ppd->congestion_entries[i].ccti_increase =
2220 			p->entries[i].ccti_increase;
2221 
2222 		ppd->congestion_entries[i].ccti_timer =
2223 			be16_to_cpu(p->entries[i].ccti_timer);
2224 
2225 		ppd->congestion_entries[i].trigger_threshold =
2226 			p->entries[i].trigger_threshold;
2227 
2228 		ppd->congestion_entries[i].ccti_min =
2229 			p->entries[i].ccti_min;
2230 	}
2231 
2232 	return reply((struct ib_smp *) ccp);
2233 }
2234 
2235 static int cc_set_congestion_control_table(struct ib_cc_mad *ccp,
2236 				struct ib_device *ibdev, u8 port)
2237 {
2238 	struct ib_cc_table_attr *p =
2239 		(struct ib_cc_table_attr *)ccp->mgmt_data;
2240 	struct qib_ibport *ibp = to_iport(ibdev, port);
2241 	struct qib_pportdata *ppd = ppd_from_ibp(ibp);
2242 	u32 cct_block_index = be32_to_cpu(ccp->attr_mod);
2243 	u32 cct_entry;
2244 	struct ib_cc_table_entry_shadow *entries;
2245 	int i;
2246 
2247 	/* Is the table index more than what is supported? */
2248 	if (cct_block_index > IB_CC_TABLE_CAP_DEFAULT - 1)
2249 		goto bail;
2250 
2251 	/* If this packet is the first in the sequence then
2252 	 * zero the total table entry count.
2253 	 */
2254 	if (be16_to_cpu(p->ccti_limit) < IB_CCT_ENTRIES)
2255 		ppd->total_cct_entry = 0;
2256 
2257 	cct_entry = (be16_to_cpu(p->ccti_limit))%IB_CCT_ENTRIES;
2258 
2259 	/* ccti_limit is 0 to 63 */
2260 	ppd->total_cct_entry += (cct_entry + 1);
2261 
2262 	if (ppd->total_cct_entry > ppd->cc_supported_table_entries)
2263 		goto bail;
2264 
2265 	ppd->ccti_limit = be16_to_cpu(p->ccti_limit);
2266 
2267 	entries = ppd->ccti_entries + (IB_CCT_ENTRIES * cct_block_index);
2268 
2269 	for (i = 0; i <= cct_entry; i++)
2270 		entries[i].entry = be16_to_cpu(p->ccti_entries[i].entry);
2271 
2272 	spin_lock(&ppd->cc_shadow_lock);
2273 
2274 	ppd->ccti_entries_shadow->ccti_last_entry = ppd->total_cct_entry - 1;
2275 	memcpy(ppd->ccti_entries_shadow->entries, ppd->ccti_entries,
2276 		(ppd->total_cct_entry * sizeof(struct ib_cc_table_entry)));
2277 
2278 	ppd->congestion_entries_shadow->port_control = IB_CC_CCS_PC_SL_BASED;
2279 	ppd->congestion_entries_shadow->control_map = ppd->cc_sl_control_map;
2280 	memcpy(ppd->congestion_entries_shadow->entries, ppd->congestion_entries,
2281 		IB_CC_CCS_ENTRIES * sizeof(struct ib_cc_congestion_entry));
2282 
2283 	spin_unlock(&ppd->cc_shadow_lock);
2284 
2285 	return reply((struct ib_smp *) ccp);
2286 
2287 bail:
2288 	return reply_failure((struct ib_smp *) ccp);
2289 }
2290 
2291 static int process_cc(struct ib_device *ibdev, int mad_flags,
2292 			u8 port, const struct ib_mad *in_mad,
2293 			struct ib_mad *out_mad)
2294 {
2295 	struct ib_cc_mad *ccp = (struct ib_cc_mad *)out_mad;
2296 	int ret;
2297 
2298 	*out_mad = *in_mad;
2299 
2300 	if (ccp->class_version != 2) {
2301 		ccp->status |= IB_SMP_UNSUP_VERSION;
2302 		ret = reply((struct ib_smp *)ccp);
2303 		goto bail;
2304 	}
2305 
2306 	switch (ccp->method) {
2307 	case IB_MGMT_METHOD_GET:
2308 		switch (ccp->attr_id) {
2309 		case IB_CC_ATTR_CLASSPORTINFO:
2310 			ret = cc_get_classportinfo(ccp, ibdev);
2311 			goto bail;
2312 
2313 		case IB_CC_ATTR_CONGESTION_INFO:
2314 			ret = cc_get_congestion_info(ccp, ibdev, port);
2315 			goto bail;
2316 
2317 		case IB_CC_ATTR_CA_CONGESTION_SETTING:
2318 			ret = cc_get_congestion_setting(ccp, ibdev, port);
2319 			goto bail;
2320 
2321 		case IB_CC_ATTR_CONGESTION_CONTROL_TABLE:
2322 			ret = cc_get_congestion_control_table(ccp, ibdev, port);
2323 			goto bail;
2324 
2325 			/* FALLTHROUGH */
2326 		default:
2327 			ccp->status |= IB_SMP_UNSUP_METH_ATTR;
2328 			ret = reply((struct ib_smp *) ccp);
2329 			goto bail;
2330 		}
2331 
2332 	case IB_MGMT_METHOD_SET:
2333 		switch (ccp->attr_id) {
2334 		case IB_CC_ATTR_CA_CONGESTION_SETTING:
2335 			ret = cc_set_congestion_setting(ccp, ibdev, port);
2336 			goto bail;
2337 
2338 		case IB_CC_ATTR_CONGESTION_CONTROL_TABLE:
2339 			ret = cc_set_congestion_control_table(ccp, ibdev, port);
2340 			goto bail;
2341 
2342 			/* FALLTHROUGH */
2343 		default:
2344 			ccp->status |= IB_SMP_UNSUP_METH_ATTR;
2345 			ret = reply((struct ib_smp *) ccp);
2346 			goto bail;
2347 		}
2348 
2349 	case IB_MGMT_METHOD_GET_RESP:
2350 		/*
2351 		 * The ib_mad module will call us to process responses
2352 		 * before checking for other consumers.
2353 		 * Just tell the caller to process it normally.
2354 		 */
2355 		ret = IB_MAD_RESULT_SUCCESS;
2356 		goto bail;
2357 
2358 	case IB_MGMT_METHOD_TRAP:
2359 	default:
2360 		ccp->status |= IB_SMP_UNSUP_METHOD;
2361 		ret = reply((struct ib_smp *) ccp);
2362 	}
2363 
2364 bail:
2365 	return ret;
2366 }
2367 
2368 /**
2369  * qib_process_mad - process an incoming MAD packet
2370  * @ibdev: the infiniband device this packet came in on
2371  * @mad_flags: MAD flags
2372  * @port: the port number this packet came in on
2373  * @in_wc: the work completion entry for this packet
2374  * @in_grh: the global route header for this packet
2375  * @in_mad: the incoming MAD
2376  * @out_mad: any outgoing MAD reply
2377  *
2378  * Returns IB_MAD_RESULT_SUCCESS if this is a MAD that we are not
2379  * interested in processing.
2380  *
2381  * Note that the verbs framework has already done the MAD sanity checks,
2382  * and hop count/pointer updating for IB_MGMT_CLASS_SUBN_DIRECTED_ROUTE
2383  * MADs.
2384  *
2385  * This is called by the ib_mad module.
2386  */
2387 int qib_process_mad(struct ib_device *ibdev, int mad_flags, u8 port,
2388 		    const struct ib_wc *in_wc, const struct ib_grh *in_grh,
2389 		    const struct ib_mad *in, struct ib_mad *out,
2390 		    size_t *out_mad_size, u16 *out_mad_pkey_index)
2391 {
2392 	int ret;
2393 	struct qib_ibport *ibp = to_iport(ibdev, port);
2394 	struct qib_pportdata *ppd = ppd_from_ibp(ibp);
2395 
2396 	switch (in->mad_hdr.mgmt_class) {
2397 	case IB_MGMT_CLASS_SUBN_DIRECTED_ROUTE:
2398 	case IB_MGMT_CLASS_SUBN_LID_ROUTED:
2399 		ret = process_subn(ibdev, mad_flags, port, in, out);
2400 		goto bail;
2401 
2402 	case IB_MGMT_CLASS_PERF_MGMT:
2403 		ret = process_perf(ibdev, port, in, out);
2404 		goto bail;
2405 
2406 	case IB_MGMT_CLASS_CONG_MGMT:
2407 		if (!ppd->congestion_entries_shadow ||
2408 			 !qib_cc_table_size) {
2409 			ret = IB_MAD_RESULT_SUCCESS;
2410 			goto bail;
2411 		}
2412 		ret = process_cc(ibdev, mad_flags, port, in, out);
2413 		goto bail;
2414 
2415 	default:
2416 		ret = IB_MAD_RESULT_SUCCESS;
2417 	}
2418 
2419 bail:
2420 	return ret;
2421 }
2422 
2423 static void xmit_wait_timer_func(struct timer_list *t)
2424 {
2425 	struct qib_pportdata *ppd = from_timer(ppd, t, cong_stats.timer);
2426 	struct qib_devdata *dd = dd_from_ppd(ppd);
2427 	unsigned long flags;
2428 	u8 status;
2429 
2430 	spin_lock_irqsave(&ppd->ibport_data.rvp.lock, flags);
2431 	if (ppd->cong_stats.flags == IB_PMA_CONG_HW_CONTROL_SAMPLE) {
2432 		status = dd->f_portcntr(ppd, QIBPORTCNTR_PSSTAT);
2433 		if (status == IB_PMA_SAMPLE_STATUS_DONE) {
2434 			/* save counter cache */
2435 			cache_hw_sample_counters(ppd);
2436 			ppd->cong_stats.flags = IB_PMA_CONG_HW_CONTROL_TIMER;
2437 		} else
2438 			goto done;
2439 	}
2440 	ppd->cong_stats.counter = xmit_wait_get_value_delta(ppd);
2441 	dd->f_set_cntr_sample(ppd, QIB_CONG_TIMER_PSINTERVAL, 0x0);
2442 done:
2443 	spin_unlock_irqrestore(&ppd->ibport_data.rvp.lock, flags);
2444 	mod_timer(&ppd->cong_stats.timer, jiffies + HZ);
2445 }
2446 
2447 void qib_notify_create_mad_agent(struct rvt_dev_info *rdi, int port_idx)
2448 {
2449 	struct qib_ibdev *ibdev = container_of(rdi, struct qib_ibdev, rdi);
2450 	struct qib_devdata *dd = container_of(ibdev,
2451 					      struct qib_devdata, verbs_dev);
2452 
2453 	/* Initialize xmit_wait structure */
2454 	dd->pport[port_idx].cong_stats.counter = 0;
2455 	timer_setup(&dd->pport[port_idx].cong_stats.timer,
2456 		    xmit_wait_timer_func, 0);
2457 	dd->pport[port_idx].cong_stats.timer.expires = 0;
2458 	add_timer(&dd->pport[port_idx].cong_stats.timer);
2459 }
2460 
2461 void qib_notify_free_mad_agent(struct rvt_dev_info *rdi, int port_idx)
2462 {
2463 	struct qib_ibdev *ibdev = container_of(rdi, struct qib_ibdev, rdi);
2464 	struct qib_devdata *dd = container_of(ibdev,
2465 					      struct qib_devdata, verbs_dev);
2466 
2467 	if (dd->pport[port_idx].cong_stats.timer.function)
2468 		del_timer_sync(&dd->pport[port_idx].cong_stats.timer);
2469 
2470 	if (dd->pport[port_idx].ibport_data.smi_ah)
2471 		rdma_destroy_ah(&dd->pport[port_idx].ibport_data.smi_ah->ibah,
2472 				RDMA_DESTROY_AH_SLEEPABLE);
2473 }
2474