Merge master.kernel.org:/pub/scm/linux/kernel/git/davej/agpgart
[linux-2.6] / drivers / infiniband / hw / ipath / ipath_mad.c
1 /*
2  * Copyright (c) 2006 QLogic, Inc. All rights reserved.
3  * Copyright (c) 2005, 2006 PathScale, Inc. All rights reserved.
4  *
5  * This software is available to you under a choice of one of two
6  * licenses.  You may choose to be licensed under the terms of the GNU
7  * General Public License (GPL) Version 2, available from the file
8  * COPYING in the main directory of this source tree, or the
9  * OpenIB.org BSD license below:
10  *
11  *     Redistribution and use in source and binary forms, with or
12  *     without modification, are permitted provided that the following
13  *     conditions are met:
14  *
15  *      - Redistributions of source code must retain the above
16  *        copyright notice, this list of conditions and the following
17  *        disclaimer.
18  *
19  *      - Redistributions in binary form must reproduce the above
20  *        copyright notice, this list of conditions and the following
21  *        disclaimer in the documentation and/or other materials
22  *        provided with the distribution.
23  *
24  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
25  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
26  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
27  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
28  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
29  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
30  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
31  * SOFTWARE.
32  */
33
34 #include <rdma/ib_smi.h>
35
36 #include "ipath_kernel.h"
37 #include "ipath_verbs.h"
38 #include "ipath_common.h"
39
40 #define IB_SMP_UNSUP_VERSION    __constant_htons(0x0004)
41 #define IB_SMP_UNSUP_METHOD     __constant_htons(0x0008)
42 #define IB_SMP_UNSUP_METH_ATTR  __constant_htons(0x000C)
43 #define IB_SMP_INVALID_FIELD    __constant_htons(0x001C)
44
45 static int reply(struct ib_smp *smp)
46 {
47         /*
48          * The verbs framework will handle the directed/LID route
49          * packet changes.
50          */
51         smp->method = IB_MGMT_METHOD_GET_RESP;
52         if (smp->mgmt_class == IB_MGMT_CLASS_SUBN_DIRECTED_ROUTE)
53                 smp->status |= IB_SMP_DIRECTION;
54         return IB_MAD_RESULT_SUCCESS | IB_MAD_RESULT_REPLY;
55 }
56
57 static int recv_subn_get_nodedescription(struct ib_smp *smp,
58                                          struct ib_device *ibdev)
59 {
60         if (smp->attr_mod)
61                 smp->status |= IB_SMP_INVALID_FIELD;
62
63         strncpy(smp->data, ibdev->node_desc, sizeof(smp->data));
64
65         return reply(smp);
66 }
67
68 struct nodeinfo {
69         u8 base_version;
70         u8 class_version;
71         u8 node_type;
72         u8 num_ports;
73         __be64 sys_guid;
74         __be64 node_guid;
75         __be64 port_guid;
76         __be16 partition_cap;
77         __be16 device_id;
78         __be32 revision;
79         u8 local_port_num;
80         u8 vendor_id[3];
81 } __attribute__ ((packed));
82
83 static int recv_subn_get_nodeinfo(struct ib_smp *smp,
84                                   struct ib_device *ibdev, u8 port)
85 {
86         struct nodeinfo *nip = (struct nodeinfo *)&smp->data;
87         struct ipath_devdata *dd = to_idev(ibdev)->dd;
88         u32 vendor, majrev, minrev;
89
90         /* GUID 0 is illegal */
91         if (smp->attr_mod || (dd->ipath_guid == 0))
92                 smp->status |= IB_SMP_INVALID_FIELD;
93
94         nip->base_version = 1;
95         nip->class_version = 1;
96         nip->node_type = 1;     /* channel adapter */
97         /*
98          * XXX The num_ports value will need a layer function to get
99          * the value if we ever have more than one IB port on a chip.
100          * We will also need to get the GUID for the port.
101          */
102         nip->num_ports = ibdev->phys_port_cnt;
103         /* This is already in network order */
104         nip->sys_guid = to_idev(ibdev)->sys_image_guid;
105         nip->node_guid = dd->ipath_guid;
106         nip->port_guid = nip->sys_guid;
107         nip->partition_cap = cpu_to_be16(ipath_get_npkeys(dd));
108         nip->device_id = cpu_to_be16(dd->ipath_deviceid);
109         majrev = dd->ipath_majrev;
110         minrev = dd->ipath_minrev;
111         nip->revision = cpu_to_be32((majrev << 16) | minrev);
112         nip->local_port_num = port;
113         vendor = dd->ipath_vendorid;
114         nip->vendor_id[0] = 0;
115         nip->vendor_id[1] = vendor >> 8;
116         nip->vendor_id[2] = vendor;
117
118         return reply(smp);
119 }
120
121 static int recv_subn_get_guidinfo(struct ib_smp *smp,
122                                   struct ib_device *ibdev)
123 {
124         u32 startgx = 8 * be32_to_cpu(smp->attr_mod);
125         __be64 *p = (__be64 *) smp->data;
126
127         /* 32 blocks of 8 64-bit GUIDs per block */
128
129         memset(smp->data, 0, sizeof(smp->data));
130
131         /*
132          * We only support one GUID for now.  If this changes, the
133          * portinfo.guid_cap field needs to be updated too.
134          */
135         if (startgx == 0) {
136                 __be64 g = to_idev(ibdev)->dd->ipath_guid;
137                 if (g == 0)
138                         /* GUID 0 is illegal */
139                         smp->status |= IB_SMP_INVALID_FIELD;
140                 else
141                         /* The first is a copy of the read-only HW GUID. */
142                         *p = g;
143         } else
144                 smp->status |= IB_SMP_INVALID_FIELD;
145
146         return reply(smp);
147 }
148
149
150 static int get_overrunthreshold(struct ipath_devdata *dd)
151 {
152         return (dd->ipath_ibcctrl >>
153                 INFINIPATH_IBCC_OVERRUNTHRESHOLD_SHIFT) &
154                 INFINIPATH_IBCC_OVERRUNTHRESHOLD_MASK;
155 }
156
157 /**
158  * set_overrunthreshold - set the overrun threshold
159  * @dd: the infinipath device
160  * @n: the new threshold
161  *
162  * Note that this will only take effect when the link state changes.
163  */
164 static int set_overrunthreshold(struct ipath_devdata *dd, unsigned n)
165 {
166         unsigned v;
167
168         v = (dd->ipath_ibcctrl >> INFINIPATH_IBCC_OVERRUNTHRESHOLD_SHIFT) &
169                 INFINIPATH_IBCC_OVERRUNTHRESHOLD_MASK;
170         if (v != n) {
171                 dd->ipath_ibcctrl &=
172                         ~(INFINIPATH_IBCC_OVERRUNTHRESHOLD_MASK <<
173                           INFINIPATH_IBCC_OVERRUNTHRESHOLD_SHIFT);
174                 dd->ipath_ibcctrl |=
175                         (u64) n << INFINIPATH_IBCC_OVERRUNTHRESHOLD_SHIFT;
176                 ipath_write_kreg(dd, dd->ipath_kregs->kr_ibcctrl,
177                                  dd->ipath_ibcctrl);
178         }
179         return 0;
180 }
181
182 static int get_phyerrthreshold(struct ipath_devdata *dd)
183 {
184         return (dd->ipath_ibcctrl >>
185                 INFINIPATH_IBCC_PHYERRTHRESHOLD_SHIFT) &
186                 INFINIPATH_IBCC_PHYERRTHRESHOLD_MASK;
187 }
188
189 /**
190  * set_phyerrthreshold - set the physical error threshold
191  * @dd: the infinipath device
192  * @n: the new threshold
193  *
194  * Note that this will only take effect when the link state changes.
195  */
196 static int set_phyerrthreshold(struct ipath_devdata *dd, unsigned n)
197 {
198         unsigned v;
199
200         v = (dd->ipath_ibcctrl >> INFINIPATH_IBCC_PHYERRTHRESHOLD_SHIFT) &
201                 INFINIPATH_IBCC_PHYERRTHRESHOLD_MASK;
202         if (v != n) {
203                 dd->ipath_ibcctrl &=
204                         ~(INFINIPATH_IBCC_PHYERRTHRESHOLD_MASK <<
205                           INFINIPATH_IBCC_PHYERRTHRESHOLD_SHIFT);
206                 dd->ipath_ibcctrl |=
207                         (u64) n << INFINIPATH_IBCC_PHYERRTHRESHOLD_SHIFT;
208                 ipath_write_kreg(dd, dd->ipath_kregs->kr_ibcctrl,
209                                  dd->ipath_ibcctrl);
210         }
211         return 0;
212 }
213
214 /**
215  * get_linkdowndefaultstate - get the default linkdown state
216  * @dd: the infinipath device
217  *
218  * Returns zero if the default is POLL, 1 if the default is SLEEP.
219  */
220 static int get_linkdowndefaultstate(struct ipath_devdata *dd)
221 {
222         return !!(dd->ipath_ibcctrl & INFINIPATH_IBCC_LINKDOWNDEFAULTSTATE);
223 }
224
225 static int recv_subn_get_portinfo(struct ib_smp *smp,
226                                   struct ib_device *ibdev, u8 port)
227 {
228         struct ipath_ibdev *dev;
229         struct ib_port_info *pip = (struct ib_port_info *)smp->data;
230         u16 lid;
231         u8 ibcstat;
232         u8 mtu;
233         int ret;
234
235         if (be32_to_cpu(smp->attr_mod) > ibdev->phys_port_cnt) {
236                 smp->status |= IB_SMP_INVALID_FIELD;
237                 ret = reply(smp);
238                 goto bail;
239         }
240
241         dev = to_idev(ibdev);
242
243         /* Clear all fields.  Only set the non-zero fields. */
244         memset(smp->data, 0, sizeof(smp->data));
245
246         /* Only return the mkey if the protection field allows it. */
247         if (smp->method == IB_MGMT_METHOD_SET || dev->mkey == smp->mkey ||
248             (dev->mkeyprot_resv_lmc >> 6) == 0)
249                 pip->mkey = dev->mkey;
250         pip->gid_prefix = dev->gid_prefix;
251         lid = dev->dd->ipath_lid;
252         pip->lid = lid ? cpu_to_be16(lid) : IB_LID_PERMISSIVE;
253         pip->sm_lid = cpu_to_be16(dev->sm_lid);
254         pip->cap_mask = cpu_to_be32(dev->port_cap_flags);
255         /* pip->diag_code; */
256         pip->mkey_lease_period = cpu_to_be16(dev->mkey_lease_period);
257         pip->local_port_num = port;
258         pip->link_width_enabled = dev->link_width_enabled;
259         pip->link_width_supported = 3;  /* 1x or 4x */
260         pip->link_width_active = 2;     /* 4x */
261         pip->linkspeed_portstate = 0x10;        /* 2.5Gbps */
262         ibcstat = dev->dd->ipath_lastibcstat;
263         pip->linkspeed_portstate |= ((ibcstat >> 4) & 0x3) + 1;
264         pip->portphysstate_linkdown =
265                 (ipath_cvt_physportstate[ibcstat & 0xf] << 4) |
266                 (get_linkdowndefaultstate(dev->dd) ? 1 : 2);
267         pip->mkeyprot_resv_lmc = dev->mkeyprot_resv_lmc;
268         pip->linkspeedactive_enabled = 0x11;    /* 2.5Gbps, 2.5Gbps */
269         switch (dev->dd->ipath_ibmtu) {
270         case 4096:
271                 mtu = IB_MTU_4096;
272                 break;
273         case 2048:
274                 mtu = IB_MTU_2048;
275                 break;
276         case 1024:
277                 mtu = IB_MTU_1024;
278                 break;
279         case 512:
280                 mtu = IB_MTU_512;
281                 break;
282         case 256:
283                 mtu = IB_MTU_256;
284                 break;
285         default:                /* oops, something is wrong */
286                 mtu = IB_MTU_2048;
287                 break;
288         }
289         pip->neighbormtu_mastersmsl = (mtu << 4) | dev->sm_sl;
290         pip->vlcap_inittype = 0x10;     /* VLCap = VL0, InitType = 0 */
291         pip->vl_high_limit = dev->vl_high_limit;
292         /* pip->vl_arb_high_cap; // only one VL */
293         /* pip->vl_arb_low_cap; // only one VL */
294         /* InitTypeReply = 0 */
295         pip->inittypereply_mtucap = IB_MTU_4096;
296         // HCAs ignore VLStallCount and HOQLife
297         /* pip->vlstallcnt_hoqlife; */
298         pip->operationalvl_pei_peo_fpi_fpo = 0x10;      /* OVLs = 1 */
299         pip->mkey_violations = cpu_to_be16(dev->mkey_violations);
300         /* P_KeyViolations are counted by hardware. */
301         pip->pkey_violations =
302                 cpu_to_be16((ipath_get_cr_errpkey(dev->dd) -
303                              dev->z_pkey_violations) & 0xFFFF);
304         pip->qkey_violations = cpu_to_be16(dev->qkey_violations);
305         /* Only the hardware GUID is supported for now */
306         pip->guid_cap = 1;
307         pip->clientrereg_resv_subnetto = dev->subnet_timeout;
308         /* 32.768 usec. response time (guessing) */
309         pip->resv_resptimevalue = 3;
310         pip->localphyerrors_overrunerrors =
311                 (get_phyerrthreshold(dev->dd) << 4) |
312                 get_overrunthreshold(dev->dd);
313         /* pip->max_credit_hint; */
314         /* pip->link_roundtrip_latency[3]; */
315
316         ret = reply(smp);
317
318 bail:
319         return ret;
320 }
321
322 /**
323  * get_pkeys - return the PKEY table for port 0
324  * @dd: the infinipath device
325  * @pkeys: the pkey table is placed here
326  */
327 static int get_pkeys(struct ipath_devdata *dd, u16 * pkeys)
328 {
329         struct ipath_portdata *pd = dd->ipath_pd[0];
330
331         memcpy(pkeys, pd->port_pkeys, sizeof(pd->port_pkeys));
332
333         return 0;
334 }
335
336 static int recv_subn_get_pkeytable(struct ib_smp *smp,
337                                    struct ib_device *ibdev)
338 {
339         u32 startpx = 32 * (be32_to_cpu(smp->attr_mod) & 0xffff);
340         u16 *p = (u16 *) smp->data;
341         __be16 *q = (__be16 *) smp->data;
342
343         /* 64 blocks of 32 16-bit P_Key entries */
344
345         memset(smp->data, 0, sizeof(smp->data));
346         if (startpx == 0) {
347                 struct ipath_ibdev *dev = to_idev(ibdev);
348                 unsigned i, n = ipath_get_npkeys(dev->dd);
349
350                 get_pkeys(dev->dd, p);
351
352                 for (i = 0; i < n; i++)
353                         q[i] = cpu_to_be16(p[i]);
354         } else
355                 smp->status |= IB_SMP_INVALID_FIELD;
356
357         return reply(smp);
358 }
359
360 static int recv_subn_set_guidinfo(struct ib_smp *smp,
361                                   struct ib_device *ibdev)
362 {
363         /* The only GUID we support is the first read-only entry. */
364         return recv_subn_get_guidinfo(smp, ibdev);
365 }
366
367 /**
368  * set_linkdowndefaultstate - set the default linkdown state
369  * @dd: the infinipath device
370  * @sleep: the new state
371  *
372  * Note that this will only take effect when the link state changes.
373  */
374 static int set_linkdowndefaultstate(struct ipath_devdata *dd, int sleep)
375 {
376         if (sleep)
377                 dd->ipath_ibcctrl |= INFINIPATH_IBCC_LINKDOWNDEFAULTSTATE;
378         else
379                 dd->ipath_ibcctrl &= ~INFINIPATH_IBCC_LINKDOWNDEFAULTSTATE;
380         ipath_write_kreg(dd, dd->ipath_kregs->kr_ibcctrl,
381                          dd->ipath_ibcctrl);
382         return 0;
383 }
384
385 /**
386  * recv_subn_set_portinfo - set port information
387  * @smp: the incoming SM packet
388  * @ibdev: the infiniband device
389  * @port: the port on the device
390  *
391  * Set Portinfo (see ch. 14.2.5.6).
392  */
393 static int recv_subn_set_portinfo(struct ib_smp *smp,
394                                   struct ib_device *ibdev, u8 port)
395 {
396         struct ib_port_info *pip = (struct ib_port_info *)smp->data;
397         struct ib_event event;
398         struct ipath_ibdev *dev;
399         u32 flags;
400         char clientrereg = 0;
401         u16 lid, smlid;
402         u8 lwe;
403         u8 lse;
404         u8 state;
405         u16 lstate;
406         u32 mtu;
407         int ret, ore;
408
409         if (be32_to_cpu(smp->attr_mod) > ibdev->phys_port_cnt)
410                 goto err;
411
412         dev = to_idev(ibdev);
413         event.device = ibdev;
414         event.element.port_num = port;
415
416         dev->mkey = pip->mkey;
417         dev->gid_prefix = pip->gid_prefix;
418         dev->mkey_lease_period = be16_to_cpu(pip->mkey_lease_period);
419
420         lid = be16_to_cpu(pip->lid);
421         if (lid != dev->dd->ipath_lid) {
422                 /* Must be a valid unicast LID address. */
423                 if (lid == 0 || lid >= IPATH_MULTICAST_LID_BASE)
424                         goto err;
425                 ipath_set_lid(dev->dd, lid, pip->mkeyprot_resv_lmc & 7);
426                 event.event = IB_EVENT_LID_CHANGE;
427                 ib_dispatch_event(&event);
428         }
429
430         smlid = be16_to_cpu(pip->sm_lid);
431         if (smlid != dev->sm_lid) {
432                 /* Must be a valid unicast LID address. */
433                 if (smlid == 0 || smlid >= IPATH_MULTICAST_LID_BASE)
434                         goto err;
435                 dev->sm_lid = smlid;
436                 event.event = IB_EVENT_SM_CHANGE;
437                 ib_dispatch_event(&event);
438         }
439
440         /* Only 4x supported but allow 1x or 4x to be set (see 14.2.6.6). */
441         lwe = pip->link_width_enabled;
442         if ((lwe >= 4 && lwe <= 8) || (lwe >= 0xC && lwe <= 0xFE))
443                 goto err;
444         if (lwe == 0xFF)
445                 dev->link_width_enabled = 3;    /* 1x or 4x */
446         else if (lwe)
447                 dev->link_width_enabled = lwe;
448
449         /* Only 2.5 Gbs supported. */
450         lse = pip->linkspeedactive_enabled & 0xF;
451         if (lse >= 2 && lse <= 0xE)
452                 goto err;
453
454         /* Set link down default state. */
455         switch (pip->portphysstate_linkdown & 0xF) {
456         case 0: /* NOP */
457                 break;
458         case 1: /* SLEEP */
459                 if (set_linkdowndefaultstate(dev->dd, 1))
460                         goto err;
461                 break;
462         case 2: /* POLL */
463                 if (set_linkdowndefaultstate(dev->dd, 0))
464                         goto err;
465                 break;
466         default:
467                 goto err;
468         }
469
470         dev->mkeyprot_resv_lmc = pip->mkeyprot_resv_lmc;
471         dev->vl_high_limit = pip->vl_high_limit;
472
473         switch ((pip->neighbormtu_mastersmsl >> 4) & 0xF) {
474         case IB_MTU_256:
475                 mtu = 256;
476                 break;
477         case IB_MTU_512:
478                 mtu = 512;
479                 break;
480         case IB_MTU_1024:
481                 mtu = 1024;
482                 break;
483         case IB_MTU_2048:
484                 mtu = 2048;
485                 break;
486         case IB_MTU_4096:
487                 mtu = 4096;
488                 break;
489         default:
490                 /* XXX We have already partially updated our state! */
491                 goto err;
492         }
493         ipath_set_mtu(dev->dd, mtu);
494
495         dev->sm_sl = pip->neighbormtu_mastersmsl & 0xF;
496
497         /* We only support VL0 */
498         if (((pip->operationalvl_pei_peo_fpi_fpo >> 4) & 0xF) > 1)
499                 goto err;
500
501         if (pip->mkey_violations == 0)
502                 dev->mkey_violations = 0;
503
504         /*
505          * Hardware counter can't be reset so snapshot and subtract
506          * later.
507          */
508         if (pip->pkey_violations == 0)
509                 dev->z_pkey_violations = ipath_get_cr_errpkey(dev->dd);
510
511         if (pip->qkey_violations == 0)
512                 dev->qkey_violations = 0;
513
514         ore = pip->localphyerrors_overrunerrors;
515         if (set_phyerrthreshold(dev->dd, (ore >> 4) & 0xF))
516                 goto err;
517
518         if (set_overrunthreshold(dev->dd, (ore & 0xF)))
519                 goto err;
520
521         dev->subnet_timeout = pip->clientrereg_resv_subnetto & 0x1F;
522
523         if (pip->clientrereg_resv_subnetto & 0x80) {
524                 clientrereg = 1;
525                 event.event = IB_EVENT_CLIENT_REREGISTER;
526                 ib_dispatch_event(&event);
527         }
528
529         /*
530          * Do the port state change now that the other link parameters
531          * have been set.
532          * Changing the port physical state only makes sense if the link
533          * is down or is being set to down.
534          */
535         state = pip->linkspeed_portstate & 0xF;
536         flags = dev->dd->ipath_flags;
537         lstate = (pip->portphysstate_linkdown >> 4) & 0xF;
538         if (lstate && !(state == IB_PORT_DOWN || state == IB_PORT_NOP))
539                 goto err;
540
541         /*
542          * Only state changes of DOWN, ARM, and ACTIVE are valid
543          * and must be in the correct state to take effect (see 7.2.6).
544          */
545         switch (state) {
546         case IB_PORT_NOP:
547                 if (lstate == 0)
548                         break;
549                 /* FALLTHROUGH */
550         case IB_PORT_DOWN:
551                 if (lstate == 0)
552                         if (get_linkdowndefaultstate(dev->dd))
553                                 lstate = IPATH_IB_LINKDOWN_SLEEP;
554                         else
555                                 lstate = IPATH_IB_LINKDOWN;
556                 else if (lstate == 1)
557                         lstate = IPATH_IB_LINKDOWN_SLEEP;
558                 else if (lstate == 2)
559                         lstate = IPATH_IB_LINKDOWN;
560                 else if (lstate == 3)
561                         lstate = IPATH_IB_LINKDOWN_DISABLE;
562                 else
563                         goto err;
564                 ipath_set_linkstate(dev->dd, lstate);
565                 if (flags & IPATH_LINKACTIVE) {
566                         event.event = IB_EVENT_PORT_ERR;
567                         ib_dispatch_event(&event);
568                 }
569                 break;
570         case IB_PORT_ARMED:
571                 if (!(flags & (IPATH_LINKINIT | IPATH_LINKACTIVE)))
572                         break;
573                 ipath_set_linkstate(dev->dd, IPATH_IB_LINKARM);
574                 if (flags & IPATH_LINKACTIVE) {
575                         event.event = IB_EVENT_PORT_ERR;
576                         ib_dispatch_event(&event);
577                 }
578                 break;
579         case IB_PORT_ACTIVE:
580                 if (!(flags & IPATH_LINKARMED))
581                         break;
582                 ipath_set_linkstate(dev->dd, IPATH_IB_LINKACTIVE);
583                 event.event = IB_EVENT_PORT_ACTIVE;
584                 ib_dispatch_event(&event);
585                 break;
586         default:
587                 /* XXX We have already partially updated our state! */
588                 goto err;
589         }
590
591         ret = recv_subn_get_portinfo(smp, ibdev, port);
592
593         if (clientrereg)
594                 pip->clientrereg_resv_subnetto |= 0x80;
595
596         goto done;
597
598 err:
599         smp->status |= IB_SMP_INVALID_FIELD;
600         ret = recv_subn_get_portinfo(smp, ibdev, port);
601
602 done:
603         return ret;
604 }
605
606 /**
607  * rm_pkey - decrecment the reference count for the given PKEY
608  * @dd: the infinipath device
609  * @key: the PKEY index
610  *
611  * Return true if this was the last reference and the hardware table entry
612  * needs to be changed.
613  */
614 static int rm_pkey(struct ipath_devdata *dd, u16 key)
615 {
616         int i;
617         int ret;
618
619         for (i = 0; i < ARRAY_SIZE(dd->ipath_pkeys); i++) {
620                 if (dd->ipath_pkeys[i] != key)
621                         continue;
622                 if (atomic_dec_and_test(&dd->ipath_pkeyrefs[i])) {
623                         dd->ipath_pkeys[i] = 0;
624                         ret = 1;
625                         goto bail;
626                 }
627                 break;
628         }
629
630         ret = 0;
631
632 bail:
633         return ret;
634 }
635
636 /**
637  * add_pkey - add the given PKEY to the hardware table
638  * @dd: the infinipath device
639  * @key: the PKEY
640  *
641  * Return an error code if unable to add the entry, zero if no change,
642  * or 1 if the hardware PKEY register needs to be updated.
643  */
644 static int add_pkey(struct ipath_devdata *dd, u16 key)
645 {
646         int i;
647         u16 lkey = key & 0x7FFF;
648         int any = 0;
649         int ret;
650
651         if (lkey == 0x7FFF) {
652                 ret = 0;
653                 goto bail;
654         }
655
656         /* Look for an empty slot or a matching PKEY. */
657         for (i = 0; i < ARRAY_SIZE(dd->ipath_pkeys); i++) {
658                 if (!dd->ipath_pkeys[i]) {
659                         any++;
660                         continue;
661                 }
662                 /* If it matches exactly, try to increment the ref count */
663                 if (dd->ipath_pkeys[i] == key) {
664                         if (atomic_inc_return(&dd->ipath_pkeyrefs[i]) > 1) {
665                                 ret = 0;
666                                 goto bail;
667                         }
668                         /* Lost the race. Look for an empty slot below. */
669                         atomic_dec(&dd->ipath_pkeyrefs[i]);
670                         any++;
671                 }
672                 /*
673                  * It makes no sense to have both the limited and unlimited
674                  * PKEY set at the same time since the unlimited one will
675                  * disable the limited one.
676                  */
677                 if ((dd->ipath_pkeys[i] & 0x7FFF) == lkey) {
678                         ret = -EEXIST;
679                         goto bail;
680                 }
681         }
682         if (!any) {
683                 ret = -EBUSY;
684                 goto bail;
685         }
686         for (i = 0; i < ARRAY_SIZE(dd->ipath_pkeys); i++) {
687                 if (!dd->ipath_pkeys[i] &&
688                     atomic_inc_return(&dd->ipath_pkeyrefs[i]) == 1) {
689                         /* for ipathstats, etc. */
690                         ipath_stats.sps_pkeys[i] = lkey;
691                         dd->ipath_pkeys[i] = key;
692                         ret = 1;
693                         goto bail;
694                 }
695         }
696         ret = -EBUSY;
697
698 bail:
699         return ret;
700 }
701
702 /**
703  * set_pkeys - set the PKEY table for port 0
704  * @dd: the infinipath device
705  * @pkeys: the PKEY table
706  */
707 static int set_pkeys(struct ipath_devdata *dd, u16 *pkeys)
708 {
709         struct ipath_portdata *pd;
710         int i;
711         int changed = 0;
712
713         pd = dd->ipath_pd[0];
714
715         for (i = 0; i < ARRAY_SIZE(pd->port_pkeys); i++) {
716                 u16 key = pkeys[i];
717                 u16 okey = pd->port_pkeys[i];
718
719                 if (key == okey)
720                         continue;
721                 /*
722                  * The value of this PKEY table entry is changing.
723                  * Remove the old entry in the hardware's array of PKEYs.
724                  */
725                 if (okey & 0x7FFF)
726                         changed |= rm_pkey(dd, okey);
727                 if (key & 0x7FFF) {
728                         int ret = add_pkey(dd, key);
729
730                         if (ret < 0)
731                                 key = 0;
732                         else
733                                 changed |= ret;
734                 }
735                 pd->port_pkeys[i] = key;
736         }
737         if (changed) {
738                 u64 pkey;
739
740                 pkey = (u64) dd->ipath_pkeys[0] |
741                         ((u64) dd->ipath_pkeys[1] << 16) |
742                         ((u64) dd->ipath_pkeys[2] << 32) |
743                         ((u64) dd->ipath_pkeys[3] << 48);
744                 ipath_cdbg(VERBOSE, "p0 new pkey reg %llx\n",
745                            (unsigned long long) pkey);
746                 ipath_write_kreg(dd, dd->ipath_kregs->kr_partitionkey,
747                                  pkey);
748         }
749         return 0;
750 }
751
752 static int recv_subn_set_pkeytable(struct ib_smp *smp,
753                                    struct ib_device *ibdev)
754 {
755         u32 startpx = 32 * (be32_to_cpu(smp->attr_mod) & 0xffff);
756         __be16 *p = (__be16 *) smp->data;
757         u16 *q = (u16 *) smp->data;
758         struct ipath_ibdev *dev = to_idev(ibdev);
759         unsigned i, n = ipath_get_npkeys(dev->dd);
760
761         for (i = 0; i < n; i++)
762                 q[i] = be16_to_cpu(p[i]);
763
764         if (startpx != 0 || set_pkeys(dev->dd, q) != 0)
765                 smp->status |= IB_SMP_INVALID_FIELD;
766
767         return recv_subn_get_pkeytable(smp, ibdev);
768 }
769
770 #define IB_PMA_CLASS_PORT_INFO          __constant_htons(0x0001)
771 #define IB_PMA_PORT_SAMPLES_CONTROL     __constant_htons(0x0010)
772 #define IB_PMA_PORT_SAMPLES_RESULT      __constant_htons(0x0011)
773 #define IB_PMA_PORT_COUNTERS            __constant_htons(0x0012)
774 #define IB_PMA_PORT_COUNTERS_EXT        __constant_htons(0x001D)
775 #define IB_PMA_PORT_SAMPLES_RESULT_EXT  __constant_htons(0x001E)
776
777 struct ib_perf {
778         u8 base_version;
779         u8 mgmt_class;
780         u8 class_version;
781         u8 method;
782         __be16 status;
783         __be16 unused;
784         __be64 tid;
785         __be16 attr_id;
786         __be16 resv;
787         __be32 attr_mod;
788         u8 reserved[40];
789         u8 data[192];
790 } __attribute__ ((packed));
791
792 struct ib_pma_classportinfo {
793         u8 base_version;
794         u8 class_version;
795         __be16 cap_mask;
796         u8 reserved[3];
797         u8 resp_time_value;     /* only lower 5 bits */
798         union ib_gid redirect_gid;
799         __be32 redirect_tc_sl_fl;       /* 8, 4, 20 bits respectively */
800         __be16 redirect_lid;
801         __be16 redirect_pkey;
802         __be32 redirect_qp;     /* only lower 24 bits */
803         __be32 redirect_qkey;
804         union ib_gid trap_gid;
805         __be32 trap_tc_sl_fl;   /* 8, 4, 20 bits respectively */
806         __be16 trap_lid;
807         __be16 trap_pkey;
808         __be32 trap_hl_qp;      /* 8, 24 bits respectively */
809         __be32 trap_qkey;
810 } __attribute__ ((packed));
811
812 struct ib_pma_portsamplescontrol {
813         u8 opcode;
814         u8 port_select;
815         u8 tick;
816         u8 counter_width;       /* only lower 3 bits */
817         __be32 counter_mask0_9; /* 2, 10 * 3, bits */
818         __be16 counter_mask10_14;       /* 1, 5 * 3, bits */
819         u8 sample_mechanisms;
820         u8 sample_status;       /* only lower 2 bits */
821         __be64 option_mask;
822         __be64 vendor_mask;
823         __be32 sample_start;
824         __be32 sample_interval;
825         __be16 tag;
826         __be16 counter_select[15];
827 } __attribute__ ((packed));
828
829 struct ib_pma_portsamplesresult {
830         __be16 tag;
831         __be16 sample_status;   /* only lower 2 bits */
832         __be32 counter[15];
833 } __attribute__ ((packed));
834
835 struct ib_pma_portsamplesresult_ext {
836         __be16 tag;
837         __be16 sample_status;   /* only lower 2 bits */
838         __be32 extended_width;  /* only upper 2 bits */
839         __be64 counter[15];
840 } __attribute__ ((packed));
841
842 struct ib_pma_portcounters {
843         u8 reserved;
844         u8 port_select;
845         __be16 counter_select;
846         __be16 symbol_error_counter;
847         u8 link_error_recovery_counter;
848         u8 link_downed_counter;
849         __be16 port_rcv_errors;
850         __be16 port_rcv_remphys_errors;
851         __be16 port_rcv_switch_relay_errors;
852         __be16 port_xmit_discards;
853         u8 port_xmit_constraint_errors;
854         u8 port_rcv_constraint_errors;
855         u8 reserved1;
856         u8 lli_ebor_errors;     /* 4, 4, bits */
857         __be16 reserved2;
858         __be16 vl15_dropped;
859         __be32 port_xmit_data;
860         __be32 port_rcv_data;
861         __be32 port_xmit_packets;
862         __be32 port_rcv_packets;
863 } __attribute__ ((packed));
864
865 #define IB_PMA_SEL_SYMBOL_ERROR                 __constant_htons(0x0001)
866 #define IB_PMA_SEL_LINK_ERROR_RECOVERY          __constant_htons(0x0002)
867 #define IB_PMA_SEL_LINK_DOWNED                  __constant_htons(0x0004)
868 #define IB_PMA_SEL_PORT_RCV_ERRORS              __constant_htons(0x0008)
869 #define IB_PMA_SEL_PORT_RCV_REMPHYS_ERRORS      __constant_htons(0x0010)
870 #define IB_PMA_SEL_PORT_XMIT_DISCARDS           __constant_htons(0x0040)
871 #define IB_PMA_SEL_LOCAL_LINK_INTEGRITY_ERRORS  __constant_htons(0x0200)
872 #define IB_PMA_SEL_EXCESSIVE_BUFFER_OVERRUNS    __constant_htons(0x0400)
873 #define IB_PMA_SEL_PORT_VL15_DROPPED            __constant_htons(0x0800)
874 #define IB_PMA_SEL_PORT_XMIT_DATA               __constant_htons(0x1000)
875 #define IB_PMA_SEL_PORT_RCV_DATA                __constant_htons(0x2000)
876 #define IB_PMA_SEL_PORT_XMIT_PACKETS            __constant_htons(0x4000)
877 #define IB_PMA_SEL_PORT_RCV_PACKETS             __constant_htons(0x8000)
878
879 struct ib_pma_portcounters_ext {
880         u8 reserved;
881         u8 port_select;
882         __be16 counter_select;
883         __be32 reserved1;
884         __be64 port_xmit_data;
885         __be64 port_rcv_data;
886         __be64 port_xmit_packets;
887         __be64 port_rcv_packets;
888         __be64 port_unicast_xmit_packets;
889         __be64 port_unicast_rcv_packets;
890         __be64 port_multicast_xmit_packets;
891         __be64 port_multicast_rcv_packets;
892 } __attribute__ ((packed));
893
894 #define IB_PMA_SELX_PORT_XMIT_DATA              __constant_htons(0x0001)
895 #define IB_PMA_SELX_PORT_RCV_DATA               __constant_htons(0x0002)
896 #define IB_PMA_SELX_PORT_XMIT_PACKETS           __constant_htons(0x0004)
897 #define IB_PMA_SELX_PORT_RCV_PACKETS            __constant_htons(0x0008)
898 #define IB_PMA_SELX_PORT_UNI_XMIT_PACKETS       __constant_htons(0x0010)
899 #define IB_PMA_SELX_PORT_UNI_RCV_PACKETS        __constant_htons(0x0020)
900 #define IB_PMA_SELX_PORT_MULTI_XMIT_PACKETS     __constant_htons(0x0040)
901 #define IB_PMA_SELX_PORT_MULTI_RCV_PACKETS      __constant_htons(0x0080)
902
903 static int recv_pma_get_classportinfo(struct ib_perf *pmp)
904 {
905         struct ib_pma_classportinfo *p =
906                 (struct ib_pma_classportinfo *)pmp->data;
907
908         memset(pmp->data, 0, sizeof(pmp->data));
909
910         if (pmp->attr_mod != 0)
911                 pmp->status |= IB_SMP_INVALID_FIELD;
912
913         /* Indicate AllPortSelect is valid (only one port anyway) */
914         p->cap_mask = __constant_cpu_to_be16(1 << 8);
915         p->base_version = 1;
916         p->class_version = 1;
917         /*
918          * Expected response time is 4.096 usec. * 2^18 == 1.073741824
919          * sec.
920          */
921         p->resp_time_value = 18;
922
923         return reply((struct ib_smp *) pmp);
924 }
925
926 /*
927  * The PortSamplesControl.CounterMasks field is an array of 3 bit fields
928  * which specify the N'th counter's capabilities. See ch. 16.1.3.2.
929  * We support 5 counters which only count the mandatory quantities.
930  */
931 #define COUNTER_MASK(q, n) (q << ((9 - n) * 3))
932 #define COUNTER_MASK0_9 \
933         __constant_cpu_to_be32(COUNTER_MASK(1, 0) | \
934                                COUNTER_MASK(1, 1) | \
935                                COUNTER_MASK(1, 2) | \
936                                COUNTER_MASK(1, 3) | \
937                                COUNTER_MASK(1, 4))
938
939 static int recv_pma_get_portsamplescontrol(struct ib_perf *pmp,
940                                            struct ib_device *ibdev, u8 port)
941 {
942         struct ib_pma_portsamplescontrol *p =
943                 (struct ib_pma_portsamplescontrol *)pmp->data;
944         struct ipath_ibdev *dev = to_idev(ibdev);
945         unsigned long flags;
946         u8 port_select = p->port_select;
947
948         memset(pmp->data, 0, sizeof(pmp->data));
949
950         p->port_select = port_select;
951         if (pmp->attr_mod != 0 ||
952             (port_select != port && port_select != 0xFF))
953                 pmp->status |= IB_SMP_INVALID_FIELD;
954         /*
955          * Ticks are 10x the link transfer period which for 2.5Gbs is 4
956          * nsec.  0 == 4 nsec., 1 == 8 nsec., ..., 255 == 1020 nsec.  Sample
957          * intervals are counted in ticks.  Since we use Linux timers, that
958          * count in jiffies, we can't sample for less than 1000 ticks if HZ
959          * == 1000 (4000 ticks if HZ is 250).
960          */
961         /* XXX This is WRONG. */
962         p->tick = 250;          /* 1 usec. */
963         p->counter_width = 4;   /* 32 bit counters */
964         p->counter_mask0_9 = COUNTER_MASK0_9;
965         spin_lock_irqsave(&dev->pending_lock, flags);
966         p->sample_status = dev->pma_sample_status;
967         p->sample_start = cpu_to_be32(dev->pma_sample_start);
968         p->sample_interval = cpu_to_be32(dev->pma_sample_interval);
969         p->tag = cpu_to_be16(dev->pma_tag);
970         p->counter_select[0] = dev->pma_counter_select[0];
971         p->counter_select[1] = dev->pma_counter_select[1];
972         p->counter_select[2] = dev->pma_counter_select[2];
973         p->counter_select[3] = dev->pma_counter_select[3];
974         p->counter_select[4] = dev->pma_counter_select[4];
975         spin_unlock_irqrestore(&dev->pending_lock, flags);
976
977         return reply((struct ib_smp *) pmp);
978 }
979
980 static int recv_pma_set_portsamplescontrol(struct ib_perf *pmp,
981                                            struct ib_device *ibdev, u8 port)
982 {
983         struct ib_pma_portsamplescontrol *p =
984                 (struct ib_pma_portsamplescontrol *)pmp->data;
985         struct ipath_ibdev *dev = to_idev(ibdev);
986         unsigned long flags;
987         u32 start;
988         int ret;
989
990         if (pmp->attr_mod != 0 ||
991             (p->port_select != port && p->port_select != 0xFF)) {
992                 pmp->status |= IB_SMP_INVALID_FIELD;
993                 ret = reply((struct ib_smp *) pmp);
994                 goto bail;
995         }
996
997         start = be32_to_cpu(p->sample_start);
998         if (start != 0) {
999                 spin_lock_irqsave(&dev->pending_lock, flags);
1000                 if (dev->pma_sample_status == IB_PMA_SAMPLE_STATUS_DONE) {
1001                         dev->pma_sample_status =
1002                                 IB_PMA_SAMPLE_STATUS_STARTED;
1003                         dev->pma_sample_start = start;
1004                         dev->pma_sample_interval =
1005                                 be32_to_cpu(p->sample_interval);
1006                         dev->pma_tag = be16_to_cpu(p->tag);
1007                         if (p->counter_select[0])
1008                                 dev->pma_counter_select[0] =
1009                                         p->counter_select[0];
1010                         if (p->counter_select[1])
1011                                 dev->pma_counter_select[1] =
1012                                         p->counter_select[1];
1013                         if (p->counter_select[2])
1014                                 dev->pma_counter_select[2] =
1015                                         p->counter_select[2];
1016                         if (p->counter_select[3])
1017                                 dev->pma_counter_select[3] =
1018                                         p->counter_select[3];
1019                         if (p->counter_select[4])
1020                                 dev->pma_counter_select[4] =
1021                                         p->counter_select[4];
1022                 }
1023                 spin_unlock_irqrestore(&dev->pending_lock, flags);
1024         }
1025         ret = recv_pma_get_portsamplescontrol(pmp, ibdev, port);
1026
1027 bail:
1028         return ret;
1029 }
1030
1031 static u64 get_counter(struct ipath_ibdev *dev, __be16 sel)
1032 {
1033         u64 ret;
1034
1035         switch (sel) {
1036         case IB_PMA_PORT_XMIT_DATA:
1037                 ret = dev->ipath_sword;
1038                 break;
1039         case IB_PMA_PORT_RCV_DATA:
1040                 ret = dev->ipath_rword;
1041                 break;
1042         case IB_PMA_PORT_XMIT_PKTS:
1043                 ret = dev->ipath_spkts;
1044                 break;
1045         case IB_PMA_PORT_RCV_PKTS:
1046                 ret = dev->ipath_rpkts;
1047                 break;
1048         case IB_PMA_PORT_XMIT_WAIT:
1049                 ret = dev->ipath_xmit_wait;
1050                 break;
1051         default:
1052                 ret = 0;
1053         }
1054
1055         return ret;
1056 }
1057
1058 static int recv_pma_get_portsamplesresult(struct ib_perf *pmp,
1059                                           struct ib_device *ibdev)
1060 {
1061         struct ib_pma_portsamplesresult *p =
1062                 (struct ib_pma_portsamplesresult *)pmp->data;
1063         struct ipath_ibdev *dev = to_idev(ibdev);
1064         int i;
1065
1066         memset(pmp->data, 0, sizeof(pmp->data));
1067         p->tag = cpu_to_be16(dev->pma_tag);
1068         p->sample_status = cpu_to_be16(dev->pma_sample_status);
1069         for (i = 0; i < ARRAY_SIZE(dev->pma_counter_select); i++)
1070                 p->counter[i] = cpu_to_be32(
1071                         get_counter(dev, dev->pma_counter_select[i]));
1072
1073         return reply((struct ib_smp *) pmp);
1074 }
1075
1076 static int recv_pma_get_portsamplesresult_ext(struct ib_perf *pmp,
1077                                               struct ib_device *ibdev)
1078 {
1079         struct ib_pma_portsamplesresult_ext *p =
1080                 (struct ib_pma_portsamplesresult_ext *)pmp->data;
1081         struct ipath_ibdev *dev = to_idev(ibdev);
1082         int i;
1083
1084         memset(pmp->data, 0, sizeof(pmp->data));
1085         p->tag = cpu_to_be16(dev->pma_tag);
1086         p->sample_status = cpu_to_be16(dev->pma_sample_status);
1087         /* 64 bits */
1088         p->extended_width = __constant_cpu_to_be32(0x80000000);
1089         for (i = 0; i < ARRAY_SIZE(dev->pma_counter_select); i++)
1090                 p->counter[i] = cpu_to_be64(
1091                         get_counter(dev, dev->pma_counter_select[i]));
1092
1093         return reply((struct ib_smp *) pmp);
1094 }
1095
1096 static int recv_pma_get_portcounters(struct ib_perf *pmp,
1097                                      struct ib_device *ibdev, u8 port)
1098 {
1099         struct ib_pma_portcounters *p = (struct ib_pma_portcounters *)
1100                 pmp->data;
1101         struct ipath_ibdev *dev = to_idev(ibdev);
1102         struct ipath_verbs_counters cntrs;
1103         u8 port_select = p->port_select;
1104
1105         ipath_get_counters(dev->dd, &cntrs);
1106
1107         /* Adjust counters for any resets done. */
1108         cntrs.symbol_error_counter -= dev->z_symbol_error_counter;
1109         cntrs.link_error_recovery_counter -=
1110                 dev->z_link_error_recovery_counter;
1111         cntrs.link_downed_counter -= dev->z_link_downed_counter;
1112         cntrs.port_rcv_errors += dev->rcv_errors;
1113         cntrs.port_rcv_errors -= dev->z_port_rcv_errors;
1114         cntrs.port_rcv_remphys_errors -= dev->z_port_rcv_remphys_errors;
1115         cntrs.port_xmit_discards -= dev->z_port_xmit_discards;
1116         cntrs.port_xmit_data -= dev->z_port_xmit_data;
1117         cntrs.port_rcv_data -= dev->z_port_rcv_data;
1118         cntrs.port_xmit_packets -= dev->z_port_xmit_packets;
1119         cntrs.port_rcv_packets -= dev->z_port_rcv_packets;
1120         cntrs.local_link_integrity_errors -=
1121                 dev->z_local_link_integrity_errors;
1122         cntrs.excessive_buffer_overrun_errors -=
1123                 dev->z_excessive_buffer_overrun_errors;
1124
1125         memset(pmp->data, 0, sizeof(pmp->data));
1126
1127         p->port_select = port_select;
1128         if (pmp->attr_mod != 0 ||
1129             (port_select != port && port_select != 0xFF))
1130                 pmp->status |= IB_SMP_INVALID_FIELD;
1131
1132         if (cntrs.symbol_error_counter > 0xFFFFUL)
1133                 p->symbol_error_counter = __constant_cpu_to_be16(0xFFFF);
1134         else
1135                 p->symbol_error_counter =
1136                         cpu_to_be16((u16)cntrs.symbol_error_counter);
1137         if (cntrs.link_error_recovery_counter > 0xFFUL)
1138                 p->link_error_recovery_counter = 0xFF;
1139         else
1140                 p->link_error_recovery_counter =
1141                         (u8)cntrs.link_error_recovery_counter;
1142         if (cntrs.link_downed_counter > 0xFFUL)
1143                 p->link_downed_counter = 0xFF;
1144         else
1145                 p->link_downed_counter = (u8)cntrs.link_downed_counter;
1146         if (cntrs.port_rcv_errors > 0xFFFFUL)
1147                 p->port_rcv_errors = __constant_cpu_to_be16(0xFFFF);
1148         else
1149                 p->port_rcv_errors =
1150                         cpu_to_be16((u16) cntrs.port_rcv_errors);
1151         if (cntrs.port_rcv_remphys_errors > 0xFFFFUL)
1152                 p->port_rcv_remphys_errors = __constant_cpu_to_be16(0xFFFF);
1153         else
1154                 p->port_rcv_remphys_errors =
1155                         cpu_to_be16((u16)cntrs.port_rcv_remphys_errors);
1156         if (cntrs.port_xmit_discards > 0xFFFFUL)
1157                 p->port_xmit_discards = __constant_cpu_to_be16(0xFFFF);
1158         else
1159                 p->port_xmit_discards =
1160                         cpu_to_be16((u16)cntrs.port_xmit_discards);
1161         if (cntrs.local_link_integrity_errors > 0xFUL)
1162                 cntrs.local_link_integrity_errors = 0xFUL;
1163         if (cntrs.excessive_buffer_overrun_errors > 0xFUL)
1164                 cntrs.excessive_buffer_overrun_errors = 0xFUL;
1165         p->lli_ebor_errors = (cntrs.local_link_integrity_errors << 4) |
1166                 cntrs.excessive_buffer_overrun_errors;
1167         if (dev->n_vl15_dropped > 0xFFFFUL)
1168                 p->vl15_dropped = __constant_cpu_to_be16(0xFFFF);
1169         else
1170                 p->vl15_dropped = cpu_to_be16((u16)dev->n_vl15_dropped);
1171         if (cntrs.port_xmit_data > 0xFFFFFFFFUL)
1172                 p->port_xmit_data = __constant_cpu_to_be32(0xFFFFFFFF);
1173         else
1174                 p->port_xmit_data = cpu_to_be32((u32)cntrs.port_xmit_data);
1175         if (cntrs.port_rcv_data > 0xFFFFFFFFUL)
1176                 p->port_rcv_data = __constant_cpu_to_be32(0xFFFFFFFF);
1177         else
1178                 p->port_rcv_data = cpu_to_be32((u32)cntrs.port_rcv_data);
1179         if (cntrs.port_xmit_packets > 0xFFFFFFFFUL)
1180                 p->port_xmit_packets = __constant_cpu_to_be32(0xFFFFFFFF);
1181         else
1182                 p->port_xmit_packets =
1183                         cpu_to_be32((u32)cntrs.port_xmit_packets);
1184         if (cntrs.port_rcv_packets > 0xFFFFFFFFUL)
1185                 p->port_rcv_packets = __constant_cpu_to_be32(0xFFFFFFFF);
1186         else
1187                 p->port_rcv_packets =
1188                         cpu_to_be32((u32) cntrs.port_rcv_packets);
1189
1190         return reply((struct ib_smp *) pmp);
1191 }
1192
1193 static int recv_pma_get_portcounters_ext(struct ib_perf *pmp,
1194                                          struct ib_device *ibdev, u8 port)
1195 {
1196         struct ib_pma_portcounters_ext *p =
1197                 (struct ib_pma_portcounters_ext *)pmp->data;
1198         struct ipath_ibdev *dev = to_idev(ibdev);
1199         u64 swords, rwords, spkts, rpkts, xwait;
1200         u8 port_select = p->port_select;
1201
1202         ipath_snapshot_counters(dev->dd, &swords, &rwords, &spkts,
1203                                 &rpkts, &xwait);
1204
1205         /* Adjust counters for any resets done. */
1206         swords -= dev->z_port_xmit_data;
1207         rwords -= dev->z_port_rcv_data;
1208         spkts -= dev->z_port_xmit_packets;
1209         rpkts -= dev->z_port_rcv_packets;
1210
1211         memset(pmp->data, 0, sizeof(pmp->data));
1212
1213         p->port_select = port_select;
1214         if (pmp->attr_mod != 0 ||
1215             (port_select != port && port_select != 0xFF))
1216                 pmp->status |= IB_SMP_INVALID_FIELD;
1217
1218         p->port_xmit_data = cpu_to_be64(swords);
1219         p->port_rcv_data = cpu_to_be64(rwords);
1220         p->port_xmit_packets = cpu_to_be64(spkts);
1221         p->port_rcv_packets = cpu_to_be64(rpkts);
1222         p->port_unicast_xmit_packets = cpu_to_be64(dev->n_unicast_xmit);
1223         p->port_unicast_rcv_packets = cpu_to_be64(dev->n_unicast_rcv);
1224         p->port_multicast_xmit_packets = cpu_to_be64(dev->n_multicast_xmit);
1225         p->port_multicast_rcv_packets = cpu_to_be64(dev->n_multicast_rcv);
1226
1227         return reply((struct ib_smp *) pmp);
1228 }
1229
1230 static int recv_pma_set_portcounters(struct ib_perf *pmp,
1231                                      struct ib_device *ibdev, u8 port)
1232 {
1233         struct ib_pma_portcounters *p = (struct ib_pma_portcounters *)
1234                 pmp->data;
1235         struct ipath_ibdev *dev = to_idev(ibdev);
1236         struct ipath_verbs_counters cntrs;
1237
1238         /*
1239          * Since the HW doesn't support clearing counters, we save the
1240          * current count and subtract it from future responses.
1241          */
1242         ipath_get_counters(dev->dd, &cntrs);
1243
1244         if (p->counter_select & IB_PMA_SEL_SYMBOL_ERROR)
1245                 dev->z_symbol_error_counter = cntrs.symbol_error_counter;
1246
1247         if (p->counter_select & IB_PMA_SEL_LINK_ERROR_RECOVERY)
1248                 dev->z_link_error_recovery_counter =
1249                         cntrs.link_error_recovery_counter;
1250
1251         if (p->counter_select & IB_PMA_SEL_LINK_DOWNED)
1252                 dev->z_link_downed_counter = cntrs.link_downed_counter;
1253
1254         if (p->counter_select & IB_PMA_SEL_PORT_RCV_ERRORS)
1255                 dev->z_port_rcv_errors =
1256                         cntrs.port_rcv_errors + dev->rcv_errors;
1257
1258         if (p->counter_select & IB_PMA_SEL_PORT_RCV_REMPHYS_ERRORS)
1259                 dev->z_port_rcv_remphys_errors =
1260                         cntrs.port_rcv_remphys_errors;
1261
1262         if (p->counter_select & IB_PMA_SEL_PORT_XMIT_DISCARDS)
1263                 dev->z_port_xmit_discards = cntrs.port_xmit_discards;
1264
1265         if (p->counter_select & IB_PMA_SEL_LOCAL_LINK_INTEGRITY_ERRORS)
1266                 dev->z_local_link_integrity_errors =
1267                         cntrs.local_link_integrity_errors;
1268
1269         if (p->counter_select & IB_PMA_SEL_EXCESSIVE_BUFFER_OVERRUNS)
1270                 dev->z_excessive_buffer_overrun_errors =
1271                         cntrs.excessive_buffer_overrun_errors;
1272
1273         if (p->counter_select & IB_PMA_SEL_PORT_VL15_DROPPED)
1274                 dev->n_vl15_dropped = 0;
1275
1276         if (p->counter_select & IB_PMA_SEL_PORT_XMIT_DATA)
1277                 dev->z_port_xmit_data = cntrs.port_xmit_data;
1278
1279         if (p->counter_select & IB_PMA_SEL_PORT_RCV_DATA)
1280                 dev->z_port_rcv_data = cntrs.port_rcv_data;
1281
1282         if (p->counter_select & IB_PMA_SEL_PORT_XMIT_PACKETS)
1283                 dev->z_port_xmit_packets = cntrs.port_xmit_packets;
1284
1285         if (p->counter_select & IB_PMA_SEL_PORT_RCV_PACKETS)
1286                 dev->z_port_rcv_packets = cntrs.port_rcv_packets;
1287
1288         return recv_pma_get_portcounters(pmp, ibdev, port);
1289 }
1290
1291 static int recv_pma_set_portcounters_ext(struct ib_perf *pmp,
1292                                          struct ib_device *ibdev, u8 port)
1293 {
1294         struct ib_pma_portcounters *p = (struct ib_pma_portcounters *)
1295                 pmp->data;
1296         struct ipath_ibdev *dev = to_idev(ibdev);
1297         u64 swords, rwords, spkts, rpkts, xwait;
1298
1299         ipath_snapshot_counters(dev->dd, &swords, &rwords, &spkts,
1300                                 &rpkts, &xwait);
1301
1302         if (p->counter_select & IB_PMA_SELX_PORT_XMIT_DATA)
1303                 dev->z_port_xmit_data = swords;
1304
1305         if (p->counter_select & IB_PMA_SELX_PORT_RCV_DATA)
1306                 dev->z_port_rcv_data = rwords;
1307
1308         if (p->counter_select & IB_PMA_SELX_PORT_XMIT_PACKETS)
1309                 dev->z_port_xmit_packets = spkts;
1310
1311         if (p->counter_select & IB_PMA_SELX_PORT_RCV_PACKETS)
1312                 dev->z_port_rcv_packets = rpkts;
1313
1314         if (p->counter_select & IB_PMA_SELX_PORT_UNI_XMIT_PACKETS)
1315                 dev->n_unicast_xmit = 0;
1316
1317         if (p->counter_select & IB_PMA_SELX_PORT_UNI_RCV_PACKETS)
1318                 dev->n_unicast_rcv = 0;
1319
1320         if (p->counter_select & IB_PMA_SELX_PORT_MULTI_XMIT_PACKETS)
1321                 dev->n_multicast_xmit = 0;
1322
1323         if (p->counter_select & IB_PMA_SELX_PORT_MULTI_RCV_PACKETS)
1324                 dev->n_multicast_rcv = 0;
1325
1326         return recv_pma_get_portcounters_ext(pmp, ibdev, port);
1327 }
1328
1329 static int process_subn(struct ib_device *ibdev, int mad_flags,
1330                         u8 port_num, struct ib_mad *in_mad,
1331                         struct ib_mad *out_mad)
1332 {
1333         struct ib_smp *smp = (struct ib_smp *)out_mad;
1334         struct ipath_ibdev *dev = to_idev(ibdev);
1335         int ret;
1336
1337         *out_mad = *in_mad;
1338         if (smp->class_version != 1) {
1339                 smp->status |= IB_SMP_UNSUP_VERSION;
1340                 ret = reply(smp);
1341                 goto bail;
1342         }
1343
1344         /* Is the mkey in the process of expiring? */
1345         if (dev->mkey_lease_timeout && jiffies >= dev->mkey_lease_timeout) {
1346                 /* Clear timeout and mkey protection field. */
1347                 dev->mkey_lease_timeout = 0;
1348                 dev->mkeyprot_resv_lmc &= 0x3F;
1349         }
1350
1351         /*
1352          * M_Key checking depends on
1353          * Portinfo:M_Key_protect_bits
1354          */
1355         if ((mad_flags & IB_MAD_IGNORE_MKEY) == 0 && dev->mkey != 0 &&
1356             dev->mkey != smp->mkey &&
1357             (smp->method == IB_MGMT_METHOD_SET ||
1358              (smp->method == IB_MGMT_METHOD_GET &&
1359               (dev->mkeyprot_resv_lmc >> 7) != 0))) {
1360                 if (dev->mkey_violations != 0xFFFF)
1361                         ++dev->mkey_violations;
1362                 if (dev->mkey_lease_timeout ||
1363                     dev->mkey_lease_period == 0) {
1364                         ret = IB_MAD_RESULT_SUCCESS |
1365                                 IB_MAD_RESULT_CONSUMED;
1366                         goto bail;
1367                 }
1368                 dev->mkey_lease_timeout = jiffies +
1369                         dev->mkey_lease_period * HZ;
1370                 /* Future: Generate a trap notice. */
1371                 ret = IB_MAD_RESULT_SUCCESS | IB_MAD_RESULT_CONSUMED;
1372                 goto bail;
1373         } else if (dev->mkey_lease_timeout)
1374                 dev->mkey_lease_timeout = 0;
1375
1376         switch (smp->method) {
1377         case IB_MGMT_METHOD_GET:
1378                 switch (smp->attr_id) {
1379                 case IB_SMP_ATTR_NODE_DESC:
1380                         ret = recv_subn_get_nodedescription(smp, ibdev);
1381                         goto bail;
1382                 case IB_SMP_ATTR_NODE_INFO:
1383                         ret = recv_subn_get_nodeinfo(smp, ibdev, port_num);
1384                         goto bail;
1385                 case IB_SMP_ATTR_GUID_INFO:
1386                         ret = recv_subn_get_guidinfo(smp, ibdev);
1387                         goto bail;
1388                 case IB_SMP_ATTR_PORT_INFO:
1389                         ret = recv_subn_get_portinfo(smp, ibdev, port_num);
1390                         goto bail;
1391                 case IB_SMP_ATTR_PKEY_TABLE:
1392                         ret = recv_subn_get_pkeytable(smp, ibdev);
1393                         goto bail;
1394                 case IB_SMP_ATTR_SM_INFO:
1395                         if (dev->port_cap_flags & IB_PORT_SM_DISABLED) {
1396                                 ret = IB_MAD_RESULT_SUCCESS |
1397                                         IB_MAD_RESULT_CONSUMED;
1398                                 goto bail;
1399                         }
1400                         if (dev->port_cap_flags & IB_PORT_SM) {
1401                                 ret = IB_MAD_RESULT_SUCCESS;
1402                                 goto bail;
1403                         }
1404                         /* FALLTHROUGH */
1405                 default:
1406                         smp->status |= IB_SMP_UNSUP_METH_ATTR;
1407                         ret = reply(smp);
1408                         goto bail;
1409                 }
1410
1411         case IB_MGMT_METHOD_SET:
1412                 switch (smp->attr_id) {
1413                 case IB_SMP_ATTR_GUID_INFO:
1414                         ret = recv_subn_set_guidinfo(smp, ibdev);
1415                         goto bail;
1416                 case IB_SMP_ATTR_PORT_INFO:
1417                         ret = recv_subn_set_portinfo(smp, ibdev, port_num);
1418                         goto bail;
1419                 case IB_SMP_ATTR_PKEY_TABLE:
1420                         ret = recv_subn_set_pkeytable(smp, ibdev);
1421                         goto bail;
1422                 case IB_SMP_ATTR_SM_INFO:
1423                         if (dev->port_cap_flags & IB_PORT_SM_DISABLED) {
1424                                 ret = IB_MAD_RESULT_SUCCESS |
1425                                         IB_MAD_RESULT_CONSUMED;
1426                                 goto bail;
1427                         }
1428                         if (dev->port_cap_flags & IB_PORT_SM) {
1429                                 ret = IB_MAD_RESULT_SUCCESS;
1430                                 goto bail;
1431                         }
1432                         /* FALLTHROUGH */
1433                 default:
1434                         smp->status |= IB_SMP_UNSUP_METH_ATTR;
1435                         ret = reply(smp);
1436                         goto bail;
1437                 }
1438
1439         case IB_MGMT_METHOD_GET_RESP:
1440                 /*
1441                  * The ib_mad module will call us to process responses
1442                  * before checking for other consumers.
1443                  * Just tell the caller to process it normally.
1444                  */
1445                 ret = IB_MAD_RESULT_FAILURE;
1446                 goto bail;
1447         default:
1448                 smp->status |= IB_SMP_UNSUP_METHOD;
1449                 ret = reply(smp);
1450         }
1451
1452 bail:
1453         return ret;
1454 }
1455
1456 static int process_perf(struct ib_device *ibdev, u8 port_num,
1457                         struct ib_mad *in_mad,
1458                         struct ib_mad *out_mad)
1459 {
1460         struct ib_perf *pmp = (struct ib_perf *)out_mad;
1461         int ret;
1462
1463         *out_mad = *in_mad;
1464         if (pmp->class_version != 1) {
1465                 pmp->status |= IB_SMP_UNSUP_VERSION;
1466                 ret = reply((struct ib_smp *) pmp);
1467                 goto bail;
1468         }
1469
1470         switch (pmp->method) {
1471         case IB_MGMT_METHOD_GET:
1472                 switch (pmp->attr_id) {
1473                 case IB_PMA_CLASS_PORT_INFO:
1474                         ret = recv_pma_get_classportinfo(pmp);
1475                         goto bail;
1476                 case IB_PMA_PORT_SAMPLES_CONTROL:
1477                         ret = recv_pma_get_portsamplescontrol(pmp, ibdev,
1478                                                               port_num);
1479                         goto bail;
1480                 case IB_PMA_PORT_SAMPLES_RESULT:
1481                         ret = recv_pma_get_portsamplesresult(pmp, ibdev);
1482                         goto bail;
1483                 case IB_PMA_PORT_SAMPLES_RESULT_EXT:
1484                         ret = recv_pma_get_portsamplesresult_ext(pmp,
1485                                                                  ibdev);
1486                         goto bail;
1487                 case IB_PMA_PORT_COUNTERS:
1488                         ret = recv_pma_get_portcounters(pmp, ibdev,
1489                                                         port_num);
1490                         goto bail;
1491                 case IB_PMA_PORT_COUNTERS_EXT:
1492                         ret = recv_pma_get_portcounters_ext(pmp, ibdev,
1493                                                             port_num);
1494                         goto bail;
1495                 default:
1496                         pmp->status |= IB_SMP_UNSUP_METH_ATTR;
1497                         ret = reply((struct ib_smp *) pmp);
1498                         goto bail;
1499                 }
1500
1501         case IB_MGMT_METHOD_SET:
1502                 switch (pmp->attr_id) {
1503                 case IB_PMA_PORT_SAMPLES_CONTROL:
1504                         ret = recv_pma_set_portsamplescontrol(pmp, ibdev,
1505                                                               port_num);
1506                         goto bail;
1507                 case IB_PMA_PORT_COUNTERS:
1508                         ret = recv_pma_set_portcounters(pmp, ibdev,
1509                                                         port_num);
1510                         goto bail;
1511                 case IB_PMA_PORT_COUNTERS_EXT:
1512                         ret = recv_pma_set_portcounters_ext(pmp, ibdev,
1513                                                             port_num);
1514                         goto bail;
1515                 default:
1516                         pmp->status |= IB_SMP_UNSUP_METH_ATTR;
1517                         ret = reply((struct ib_smp *) pmp);
1518                         goto bail;
1519                 }
1520
1521         case IB_MGMT_METHOD_GET_RESP:
1522                 /*
1523                  * The ib_mad module will call us to process responses
1524                  * before checking for other consumers.
1525                  * Just tell the caller to process it normally.
1526                  */
1527                 ret = IB_MAD_RESULT_FAILURE;
1528                 goto bail;
1529         default:
1530                 pmp->status |= IB_SMP_UNSUP_METHOD;
1531                 ret = reply((struct ib_smp *) pmp);
1532         }
1533
1534 bail:
1535         return ret;
1536 }
1537
1538 /**
1539  * ipath_process_mad - process an incoming MAD packet
1540  * @ibdev: the infiniband device this packet came in on
1541  * @mad_flags: MAD flags
1542  * @port_num: the port number this packet came in on
1543  * @in_wc: the work completion entry for this packet
1544  * @in_grh: the global route header for this packet
1545  * @in_mad: the incoming MAD
1546  * @out_mad: any outgoing MAD reply
1547  *
1548  * Returns IB_MAD_RESULT_SUCCESS if this is a MAD that we are not
1549  * interested in processing.
1550  *
1551  * Note that the verbs framework has already done the MAD sanity checks,
1552  * and hop count/pointer updating for IB_MGMT_CLASS_SUBN_DIRECTED_ROUTE
1553  * MADs.
1554  *
1555  * This is called by the ib_mad module.
1556  */
1557 int ipath_process_mad(struct ib_device *ibdev, int mad_flags, u8 port_num,
1558                       struct ib_wc *in_wc, struct ib_grh *in_grh,
1559                       struct ib_mad *in_mad, struct ib_mad *out_mad)
1560 {
1561         int ret;
1562
1563         switch (in_mad->mad_hdr.mgmt_class) {
1564         case IB_MGMT_CLASS_SUBN_DIRECTED_ROUTE:
1565         case IB_MGMT_CLASS_SUBN_LID_ROUTED:
1566                 ret = process_subn(ibdev, mad_flags, port_num,
1567                                    in_mad, out_mad);
1568                 goto bail;
1569         case IB_MGMT_CLASS_PERF_MGMT:
1570                 ret = process_perf(ibdev, port_num, in_mad, out_mad);
1571                 goto bail;
1572         default:
1573                 ret = IB_MAD_RESULT_SUCCESS;
1574         }
1575
1576 bail:
1577         return ret;
1578 }