Merge master.kernel.org:/pub/scm/linux/kernel/git/torvalds/linux-2.6
[linux-2.6] / drivers / infiniband / hw / ipath / ipath_mad.c
1 /*
2  * Copyright (c) 2006, 2007 QLogic Corporation. 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 = dd->ipath_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 == 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 << 6) | dev->dd->ipath_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         /*
296          * Note: the chips support a maximum MTU of 4096, but the driver
297          * hasn't implemented this feature yet, so set the maximum value
298          * to 2048.
299          */
300         pip->inittypereply_mtucap = IB_MTU_2048;
301         // HCAs ignore VLStallCount and HOQLife
302         /* pip->vlstallcnt_hoqlife; */
303         pip->operationalvl_pei_peo_fpi_fpo = 0x10;      /* OVLs = 1 */
304         pip->mkey_violations = cpu_to_be16(dev->mkey_violations);
305         /* P_KeyViolations are counted by hardware. */
306         pip->pkey_violations =
307                 cpu_to_be16((ipath_get_cr_errpkey(dev->dd) -
308                              dev->z_pkey_violations) & 0xFFFF);
309         pip->qkey_violations = cpu_to_be16(dev->qkey_violations);
310         /* Only the hardware GUID is supported for now */
311         pip->guid_cap = 1;
312         pip->clientrereg_resv_subnetto = dev->subnet_timeout;
313         /* 32.768 usec. response time (guessing) */
314         pip->resv_resptimevalue = 3;
315         pip->localphyerrors_overrunerrors =
316                 (get_phyerrthreshold(dev->dd) << 4) |
317                 get_overrunthreshold(dev->dd);
318         /* pip->max_credit_hint; */
319         /* pip->link_roundtrip_latency[3]; */
320
321         ret = reply(smp);
322
323 bail:
324         return ret;
325 }
326
327 /**
328  * get_pkeys - return the PKEY table for port 0
329  * @dd: the infinipath device
330  * @pkeys: the pkey table is placed here
331  */
332 static int get_pkeys(struct ipath_devdata *dd, u16 * pkeys)
333 {
334         struct ipath_portdata *pd = dd->ipath_pd[0];
335
336         memcpy(pkeys, pd->port_pkeys, sizeof(pd->port_pkeys));
337
338         return 0;
339 }
340
341 static int recv_subn_get_pkeytable(struct ib_smp *smp,
342                                    struct ib_device *ibdev)
343 {
344         u32 startpx = 32 * (be32_to_cpu(smp->attr_mod) & 0xffff);
345         u16 *p = (u16 *) smp->data;
346         __be16 *q = (__be16 *) smp->data;
347
348         /* 64 blocks of 32 16-bit P_Key entries */
349
350         memset(smp->data, 0, sizeof(smp->data));
351         if (startpx == 0) {
352                 struct ipath_ibdev *dev = to_idev(ibdev);
353                 unsigned i, n = ipath_get_npkeys(dev->dd);
354
355                 get_pkeys(dev->dd, p);
356
357                 for (i = 0; i < n; i++)
358                         q[i] = cpu_to_be16(p[i]);
359         } else
360                 smp->status |= IB_SMP_INVALID_FIELD;
361
362         return reply(smp);
363 }
364
365 static int recv_subn_set_guidinfo(struct ib_smp *smp,
366                                   struct ib_device *ibdev)
367 {
368         /* The only GUID we support is the first read-only entry. */
369         return recv_subn_get_guidinfo(smp, ibdev);
370 }
371
372 /**
373  * set_linkdowndefaultstate - set the default linkdown state
374  * @dd: the infinipath device
375  * @sleep: the new state
376  *
377  * Note that this will only take effect when the link state changes.
378  */
379 static int set_linkdowndefaultstate(struct ipath_devdata *dd, int sleep)
380 {
381         if (sleep)
382                 dd->ipath_ibcctrl |= INFINIPATH_IBCC_LINKDOWNDEFAULTSTATE;
383         else
384                 dd->ipath_ibcctrl &= ~INFINIPATH_IBCC_LINKDOWNDEFAULTSTATE;
385         ipath_write_kreg(dd, dd->ipath_kregs->kr_ibcctrl,
386                          dd->ipath_ibcctrl);
387         return 0;
388 }
389
390 /**
391  * recv_subn_set_portinfo - set port information
392  * @smp: the incoming SM packet
393  * @ibdev: the infiniband device
394  * @port: the port on the device
395  *
396  * Set Portinfo (see ch. 14.2.5.6).
397  */
398 static int recv_subn_set_portinfo(struct ib_smp *smp,
399                                   struct ib_device *ibdev, u8 port)
400 {
401         struct ib_port_info *pip = (struct ib_port_info *)smp->data;
402         struct ib_event event;
403         struct ipath_ibdev *dev;
404         struct ipath_devdata *dd;
405         char clientrereg = 0;
406         u16 lid, smlid;
407         u8 lwe;
408         u8 lse;
409         u8 state;
410         u16 lstate;
411         u32 mtu;
412         int ret, ore;
413
414         if (be32_to_cpu(smp->attr_mod) > ibdev->phys_port_cnt)
415                 goto err;
416
417         dev = to_idev(ibdev);
418         dd = dev->dd;
419         event.device = ibdev;
420         event.element.port_num = port;
421
422         dev->mkey = pip->mkey;
423         dev->gid_prefix = pip->gid_prefix;
424         dev->mkey_lease_period = be16_to_cpu(pip->mkey_lease_period);
425
426         lid = be16_to_cpu(pip->lid);
427         if (dd->ipath_lid != lid ||
428             dd->ipath_lmc != (pip->mkeyprot_resv_lmc & 7)) {
429                 /* Must be a valid unicast LID address. */
430                 if (lid == 0 || lid >= IPATH_MULTICAST_LID_BASE)
431                         goto err;
432                 ipath_set_lid(dd, lid, pip->mkeyprot_resv_lmc & 7);
433                 event.event = IB_EVENT_LID_CHANGE;
434                 ib_dispatch_event(&event);
435         }
436
437         smlid = be16_to_cpu(pip->sm_lid);
438         if (smlid != dev->sm_lid) {
439                 /* Must be a valid unicast LID address. */
440                 if (smlid == 0 || smlid >= IPATH_MULTICAST_LID_BASE)
441                         goto err;
442                 dev->sm_lid = smlid;
443                 event.event = IB_EVENT_SM_CHANGE;
444                 ib_dispatch_event(&event);
445         }
446
447         /* Only 4x supported but allow 1x or 4x to be set (see 14.2.6.6). */
448         lwe = pip->link_width_enabled;
449         if ((lwe >= 4 && lwe <= 8) || (lwe >= 0xC && lwe <= 0xFE))
450                 goto err;
451         if (lwe == 0xFF)
452                 dev->link_width_enabled = 3;    /* 1x or 4x */
453         else if (lwe)
454                 dev->link_width_enabled = lwe;
455
456         /* Only 2.5 Gbs supported. */
457         lse = pip->linkspeedactive_enabled & 0xF;
458         if (lse >= 2 && lse <= 0xE)
459                 goto err;
460
461         /* Set link down default state. */
462         switch (pip->portphysstate_linkdown & 0xF) {
463         case 0: /* NOP */
464                 break;
465         case 1: /* SLEEP */
466                 if (set_linkdowndefaultstate(dd, 1))
467                         goto err;
468                 break;
469         case 2: /* POLL */
470                 if (set_linkdowndefaultstate(dd, 0))
471                         goto err;
472                 break;
473         default:
474                 goto err;
475         }
476
477         dev->mkeyprot = pip->mkeyprot_resv_lmc >> 6;
478         dev->vl_high_limit = pip->vl_high_limit;
479
480         switch ((pip->neighbormtu_mastersmsl >> 4) & 0xF) {
481         case IB_MTU_256:
482                 mtu = 256;
483                 break;
484         case IB_MTU_512:
485                 mtu = 512;
486                 break;
487         case IB_MTU_1024:
488                 mtu = 1024;
489                 break;
490         case IB_MTU_2048:
491                 mtu = 2048;
492                 break;
493         case IB_MTU_4096:
494                 mtu = 4096;
495                 break;
496         default:
497                 /* XXX We have already partially updated our state! */
498                 goto err;
499         }
500         ipath_set_mtu(dd, mtu);
501
502         dev->sm_sl = pip->neighbormtu_mastersmsl & 0xF;
503
504         /* We only support VL0 */
505         if (((pip->operationalvl_pei_peo_fpi_fpo >> 4) & 0xF) > 1)
506                 goto err;
507
508         if (pip->mkey_violations == 0)
509                 dev->mkey_violations = 0;
510
511         /*
512          * Hardware counter can't be reset so snapshot and subtract
513          * later.
514          */
515         if (pip->pkey_violations == 0)
516                 dev->z_pkey_violations = ipath_get_cr_errpkey(dd);
517
518         if (pip->qkey_violations == 0)
519                 dev->qkey_violations = 0;
520
521         ore = pip->localphyerrors_overrunerrors;
522         if (set_phyerrthreshold(dd, (ore >> 4) & 0xF))
523                 goto err;
524
525         if (set_overrunthreshold(dd, (ore & 0xF)))
526                 goto err;
527
528         dev->subnet_timeout = pip->clientrereg_resv_subnetto & 0x1F;
529
530         if (pip->clientrereg_resv_subnetto & 0x80) {
531                 clientrereg = 1;
532                 event.event = IB_EVENT_CLIENT_REREGISTER;
533                 ib_dispatch_event(&event);
534         }
535
536         /*
537          * Do the port state change now that the other link parameters
538          * have been set.
539          * Changing the port physical state only makes sense if the link
540          * is down or is being set to down.
541          */
542         state = pip->linkspeed_portstate & 0xF;
543         lstate = (pip->portphysstate_linkdown >> 4) & 0xF;
544         if (lstate && !(state == IB_PORT_DOWN || state == IB_PORT_NOP))
545                 goto err;
546
547         /*
548          * Only state changes of DOWN, ARM, and ACTIVE are valid
549          * and must be in the correct state to take effect (see 7.2.6).
550          */
551         switch (state) {
552         case IB_PORT_NOP:
553                 if (lstate == 0)
554                         break;
555                 /* FALLTHROUGH */
556         case IB_PORT_DOWN:
557                 if (lstate == 0)
558                         if (get_linkdowndefaultstate(dd))
559                                 lstate = IPATH_IB_LINKDOWN_SLEEP;
560                         else
561                                 lstate = IPATH_IB_LINKDOWN;
562                 else if (lstate == 1)
563                         lstate = IPATH_IB_LINKDOWN_SLEEP;
564                 else if (lstate == 2)
565                         lstate = IPATH_IB_LINKDOWN;
566                 else if (lstate == 3)
567                         lstate = IPATH_IB_LINKDOWN_DISABLE;
568                 else
569                         goto err;
570                 ipath_set_linkstate(dd, lstate);
571                 break;
572         case IB_PORT_ARMED:
573                 ipath_set_linkstate(dd, IPATH_IB_LINKARM);
574                 break;
575         case IB_PORT_ACTIVE:
576                 ipath_set_linkstate(dd, IPATH_IB_LINKACTIVE);
577                 break;
578         default:
579                 /* XXX We have already partially updated our state! */
580                 goto err;
581         }
582
583         ret = recv_subn_get_portinfo(smp, ibdev, port);
584
585         if (clientrereg)
586                 pip->clientrereg_resv_subnetto |= 0x80;
587
588         goto done;
589
590 err:
591         smp->status |= IB_SMP_INVALID_FIELD;
592         ret = recv_subn_get_portinfo(smp, ibdev, port);
593
594 done:
595         return ret;
596 }
597
598 /**
599  * rm_pkey - decrecment the reference count for the given PKEY
600  * @dd: the infinipath device
601  * @key: the PKEY index
602  *
603  * Return true if this was the last reference and the hardware table entry
604  * needs to be changed.
605  */
606 static int rm_pkey(struct ipath_devdata *dd, u16 key)
607 {
608         int i;
609         int ret;
610
611         for (i = 0; i < ARRAY_SIZE(dd->ipath_pkeys); i++) {
612                 if (dd->ipath_pkeys[i] != key)
613                         continue;
614                 if (atomic_dec_and_test(&dd->ipath_pkeyrefs[i])) {
615                         dd->ipath_pkeys[i] = 0;
616                         ret = 1;
617                         goto bail;
618                 }
619                 break;
620         }
621
622         ret = 0;
623
624 bail:
625         return ret;
626 }
627
628 /**
629  * add_pkey - add the given PKEY to the hardware table
630  * @dd: the infinipath device
631  * @key: the PKEY
632  *
633  * Return an error code if unable to add the entry, zero if no change,
634  * or 1 if the hardware PKEY register needs to be updated.
635  */
636 static int add_pkey(struct ipath_devdata *dd, u16 key)
637 {
638         int i;
639         u16 lkey = key & 0x7FFF;
640         int any = 0;
641         int ret;
642
643         if (lkey == 0x7FFF) {
644                 ret = 0;
645                 goto bail;
646         }
647
648         /* Look for an empty slot or a matching PKEY. */
649         for (i = 0; i < ARRAY_SIZE(dd->ipath_pkeys); i++) {
650                 if (!dd->ipath_pkeys[i]) {
651                         any++;
652                         continue;
653                 }
654                 /* If it matches exactly, try to increment the ref count */
655                 if (dd->ipath_pkeys[i] == key) {
656                         if (atomic_inc_return(&dd->ipath_pkeyrefs[i]) > 1) {
657                                 ret = 0;
658                                 goto bail;
659                         }
660                         /* Lost the race. Look for an empty slot below. */
661                         atomic_dec(&dd->ipath_pkeyrefs[i]);
662                         any++;
663                 }
664                 /*
665                  * It makes no sense to have both the limited and unlimited
666                  * PKEY set at the same time since the unlimited one will
667                  * disable the limited one.
668                  */
669                 if ((dd->ipath_pkeys[i] & 0x7FFF) == lkey) {
670                         ret = -EEXIST;
671                         goto bail;
672                 }
673         }
674         if (!any) {
675                 ret = -EBUSY;
676                 goto bail;
677         }
678         for (i = 0; i < ARRAY_SIZE(dd->ipath_pkeys); i++) {
679                 if (!dd->ipath_pkeys[i] &&
680                     atomic_inc_return(&dd->ipath_pkeyrefs[i]) == 1) {
681                         /* for ipathstats, etc. */
682                         ipath_stats.sps_pkeys[i] = lkey;
683                         dd->ipath_pkeys[i] = key;
684                         ret = 1;
685                         goto bail;
686                 }
687         }
688         ret = -EBUSY;
689
690 bail:
691         return ret;
692 }
693
694 /**
695  * set_pkeys - set the PKEY table for port 0
696  * @dd: the infinipath device
697  * @pkeys: the PKEY table
698  */
699 static int set_pkeys(struct ipath_devdata *dd, u16 *pkeys)
700 {
701         struct ipath_portdata *pd;
702         int i;
703         int changed = 0;
704
705         pd = dd->ipath_pd[0];
706
707         for (i = 0; i < ARRAY_SIZE(pd->port_pkeys); i++) {
708                 u16 key = pkeys[i];
709                 u16 okey = pd->port_pkeys[i];
710
711                 if (key == okey)
712                         continue;
713                 /*
714                  * The value of this PKEY table entry is changing.
715                  * Remove the old entry in the hardware's array of PKEYs.
716                  */
717                 if (okey & 0x7FFF)
718                         changed |= rm_pkey(dd, okey);
719                 if (key & 0x7FFF) {
720                         int ret = add_pkey(dd, key);
721
722                         if (ret < 0)
723                                 key = 0;
724                         else
725                                 changed |= ret;
726                 }
727                 pd->port_pkeys[i] = key;
728         }
729         if (changed) {
730                 u64 pkey;
731
732                 pkey = (u64) dd->ipath_pkeys[0] |
733                         ((u64) dd->ipath_pkeys[1] << 16) |
734                         ((u64) dd->ipath_pkeys[2] << 32) |
735                         ((u64) dd->ipath_pkeys[3] << 48);
736                 ipath_cdbg(VERBOSE, "p0 new pkey reg %llx\n",
737                            (unsigned long long) pkey);
738                 ipath_write_kreg(dd, dd->ipath_kregs->kr_partitionkey,
739                                  pkey);
740         }
741         return 0;
742 }
743
744 static int recv_subn_set_pkeytable(struct ib_smp *smp,
745                                    struct ib_device *ibdev)
746 {
747         u32 startpx = 32 * (be32_to_cpu(smp->attr_mod) & 0xffff);
748         __be16 *p = (__be16 *) smp->data;
749         u16 *q = (u16 *) smp->data;
750         struct ipath_ibdev *dev = to_idev(ibdev);
751         unsigned i, n = ipath_get_npkeys(dev->dd);
752
753         for (i = 0; i < n; i++)
754                 q[i] = be16_to_cpu(p[i]);
755
756         if (startpx != 0 || set_pkeys(dev->dd, q) != 0)
757                 smp->status |= IB_SMP_INVALID_FIELD;
758
759         return recv_subn_get_pkeytable(smp, ibdev);
760 }
761
762 #define IB_PMA_CLASS_PORT_INFO          __constant_htons(0x0001)
763 #define IB_PMA_PORT_SAMPLES_CONTROL     __constant_htons(0x0010)
764 #define IB_PMA_PORT_SAMPLES_RESULT      __constant_htons(0x0011)
765 #define IB_PMA_PORT_COUNTERS            __constant_htons(0x0012)
766 #define IB_PMA_PORT_COUNTERS_EXT        __constant_htons(0x001D)
767 #define IB_PMA_PORT_SAMPLES_RESULT_EXT  __constant_htons(0x001E)
768
769 struct ib_perf {
770         u8 base_version;
771         u8 mgmt_class;
772         u8 class_version;
773         u8 method;
774         __be16 status;
775         __be16 unused;
776         __be64 tid;
777         __be16 attr_id;
778         __be16 resv;
779         __be32 attr_mod;
780         u8 reserved[40];
781         u8 data[192];
782 } __attribute__ ((packed));
783
784 struct ib_pma_classportinfo {
785         u8 base_version;
786         u8 class_version;
787         __be16 cap_mask;
788         u8 reserved[3];
789         u8 resp_time_value;     /* only lower 5 bits */
790         union ib_gid redirect_gid;
791         __be32 redirect_tc_sl_fl;       /* 8, 4, 20 bits respectively */
792         __be16 redirect_lid;
793         __be16 redirect_pkey;
794         __be32 redirect_qp;     /* only lower 24 bits */
795         __be32 redirect_qkey;
796         union ib_gid trap_gid;
797         __be32 trap_tc_sl_fl;   /* 8, 4, 20 bits respectively */
798         __be16 trap_lid;
799         __be16 trap_pkey;
800         __be32 trap_hl_qp;      /* 8, 24 bits respectively */
801         __be32 trap_qkey;
802 } __attribute__ ((packed));
803
804 struct ib_pma_portsamplescontrol {
805         u8 opcode;
806         u8 port_select;
807         u8 tick;
808         u8 counter_width;       /* only lower 3 bits */
809         __be32 counter_mask0_9; /* 2, 10 * 3, bits */
810         __be16 counter_mask10_14;       /* 1, 5 * 3, bits */
811         u8 sample_mechanisms;
812         u8 sample_status;       /* only lower 2 bits */
813         __be64 option_mask;
814         __be64 vendor_mask;
815         __be32 sample_start;
816         __be32 sample_interval;
817         __be16 tag;
818         __be16 counter_select[15];
819 } __attribute__ ((packed));
820
821 struct ib_pma_portsamplesresult {
822         __be16 tag;
823         __be16 sample_status;   /* only lower 2 bits */
824         __be32 counter[15];
825 } __attribute__ ((packed));
826
827 struct ib_pma_portsamplesresult_ext {
828         __be16 tag;
829         __be16 sample_status;   /* only lower 2 bits */
830         __be32 extended_width;  /* only upper 2 bits */
831         __be64 counter[15];
832 } __attribute__ ((packed));
833
834 struct ib_pma_portcounters {
835         u8 reserved;
836         u8 port_select;
837         __be16 counter_select;
838         __be16 symbol_error_counter;
839         u8 link_error_recovery_counter;
840         u8 link_downed_counter;
841         __be16 port_rcv_errors;
842         __be16 port_rcv_remphys_errors;
843         __be16 port_rcv_switch_relay_errors;
844         __be16 port_xmit_discards;
845         u8 port_xmit_constraint_errors;
846         u8 port_rcv_constraint_errors;
847         u8 reserved1;
848         u8 lli_ebor_errors;     /* 4, 4, bits */
849         __be16 reserved2;
850         __be16 vl15_dropped;
851         __be32 port_xmit_data;
852         __be32 port_rcv_data;
853         __be32 port_xmit_packets;
854         __be32 port_rcv_packets;
855 } __attribute__ ((packed));
856
857 #define IB_PMA_SEL_SYMBOL_ERROR                 __constant_htons(0x0001)
858 #define IB_PMA_SEL_LINK_ERROR_RECOVERY          __constant_htons(0x0002)
859 #define IB_PMA_SEL_LINK_DOWNED                  __constant_htons(0x0004)
860 #define IB_PMA_SEL_PORT_RCV_ERRORS              __constant_htons(0x0008)
861 #define IB_PMA_SEL_PORT_RCV_REMPHYS_ERRORS      __constant_htons(0x0010)
862 #define IB_PMA_SEL_PORT_XMIT_DISCARDS           __constant_htons(0x0040)
863 #define IB_PMA_SEL_LOCAL_LINK_INTEGRITY_ERRORS  __constant_htons(0x0200)
864 #define IB_PMA_SEL_EXCESSIVE_BUFFER_OVERRUNS    __constant_htons(0x0400)
865 #define IB_PMA_SEL_PORT_VL15_DROPPED            __constant_htons(0x0800)
866 #define IB_PMA_SEL_PORT_XMIT_DATA               __constant_htons(0x1000)
867 #define IB_PMA_SEL_PORT_RCV_DATA                __constant_htons(0x2000)
868 #define IB_PMA_SEL_PORT_XMIT_PACKETS            __constant_htons(0x4000)
869 #define IB_PMA_SEL_PORT_RCV_PACKETS             __constant_htons(0x8000)
870
871 struct ib_pma_portcounters_ext {
872         u8 reserved;
873         u8 port_select;
874         __be16 counter_select;
875         __be32 reserved1;
876         __be64 port_xmit_data;
877         __be64 port_rcv_data;
878         __be64 port_xmit_packets;
879         __be64 port_rcv_packets;
880         __be64 port_unicast_xmit_packets;
881         __be64 port_unicast_rcv_packets;
882         __be64 port_multicast_xmit_packets;
883         __be64 port_multicast_rcv_packets;
884 } __attribute__ ((packed));
885
886 #define IB_PMA_SELX_PORT_XMIT_DATA              __constant_htons(0x0001)
887 #define IB_PMA_SELX_PORT_RCV_DATA               __constant_htons(0x0002)
888 #define IB_PMA_SELX_PORT_XMIT_PACKETS           __constant_htons(0x0004)
889 #define IB_PMA_SELX_PORT_RCV_PACKETS            __constant_htons(0x0008)
890 #define IB_PMA_SELX_PORT_UNI_XMIT_PACKETS       __constant_htons(0x0010)
891 #define IB_PMA_SELX_PORT_UNI_RCV_PACKETS        __constant_htons(0x0020)
892 #define IB_PMA_SELX_PORT_MULTI_XMIT_PACKETS     __constant_htons(0x0040)
893 #define IB_PMA_SELX_PORT_MULTI_RCV_PACKETS      __constant_htons(0x0080)
894
895 static int recv_pma_get_classportinfo(struct ib_perf *pmp)
896 {
897         struct ib_pma_classportinfo *p =
898                 (struct ib_pma_classportinfo *)pmp->data;
899
900         memset(pmp->data, 0, sizeof(pmp->data));
901
902         if (pmp->attr_mod != 0)
903                 pmp->status |= IB_SMP_INVALID_FIELD;
904
905         /* Indicate AllPortSelect is valid (only one port anyway) */
906         p->cap_mask = __constant_cpu_to_be16(1 << 8);
907         p->base_version = 1;
908         p->class_version = 1;
909         /*
910          * Expected response time is 4.096 usec. * 2^18 == 1.073741824
911          * sec.
912          */
913         p->resp_time_value = 18;
914
915         return reply((struct ib_smp *) pmp);
916 }
917
918 /*
919  * The PortSamplesControl.CounterMasks field is an array of 3 bit fields
920  * which specify the N'th counter's capabilities. See ch. 16.1.3.2.
921  * We support 5 counters which only count the mandatory quantities.
922  */
923 #define COUNTER_MASK(q, n) (q << ((9 - n) * 3))
924 #define COUNTER_MASK0_9 \
925         __constant_cpu_to_be32(COUNTER_MASK(1, 0) | \
926                                COUNTER_MASK(1, 1) | \
927                                COUNTER_MASK(1, 2) | \
928                                COUNTER_MASK(1, 3) | \
929                                COUNTER_MASK(1, 4))
930
931 static int recv_pma_get_portsamplescontrol(struct ib_perf *pmp,
932                                            struct ib_device *ibdev, u8 port)
933 {
934         struct ib_pma_portsamplescontrol *p =
935                 (struct ib_pma_portsamplescontrol *)pmp->data;
936         struct ipath_ibdev *dev = to_idev(ibdev);
937         unsigned long flags;
938         u8 port_select = p->port_select;
939
940         memset(pmp->data, 0, sizeof(pmp->data));
941
942         p->port_select = port_select;
943         if (pmp->attr_mod != 0 ||
944             (port_select != port && port_select != 0xFF))
945                 pmp->status |= IB_SMP_INVALID_FIELD;
946         /*
947          * Ticks are 10x the link transfer period which for 2.5Gbs is 4
948          * nsec.  0 == 4 nsec., 1 == 8 nsec., ..., 255 == 1020 nsec.  Sample
949          * intervals are counted in ticks.  Since we use Linux timers, that
950          * count in jiffies, we can't sample for less than 1000 ticks if HZ
951          * == 1000 (4000 ticks if HZ is 250).
952          */
953         /* XXX This is WRONG. */
954         p->tick = 250;          /* 1 usec. */
955         p->counter_width = 4;   /* 32 bit counters */
956         p->counter_mask0_9 = COUNTER_MASK0_9;
957         spin_lock_irqsave(&dev->pending_lock, flags);
958         p->sample_status = dev->pma_sample_status;
959         p->sample_start = cpu_to_be32(dev->pma_sample_start);
960         p->sample_interval = cpu_to_be32(dev->pma_sample_interval);
961         p->tag = cpu_to_be16(dev->pma_tag);
962         p->counter_select[0] = dev->pma_counter_select[0];
963         p->counter_select[1] = dev->pma_counter_select[1];
964         p->counter_select[2] = dev->pma_counter_select[2];
965         p->counter_select[3] = dev->pma_counter_select[3];
966         p->counter_select[4] = dev->pma_counter_select[4];
967         spin_unlock_irqrestore(&dev->pending_lock, flags);
968
969         return reply((struct ib_smp *) pmp);
970 }
971
972 static int recv_pma_set_portsamplescontrol(struct ib_perf *pmp,
973                                            struct ib_device *ibdev, u8 port)
974 {
975         struct ib_pma_portsamplescontrol *p =
976                 (struct ib_pma_portsamplescontrol *)pmp->data;
977         struct ipath_ibdev *dev = to_idev(ibdev);
978         unsigned long flags;
979         u32 start;
980         int ret;
981
982         if (pmp->attr_mod != 0 ||
983             (p->port_select != port && p->port_select != 0xFF)) {
984                 pmp->status |= IB_SMP_INVALID_FIELD;
985                 ret = reply((struct ib_smp *) pmp);
986                 goto bail;
987         }
988
989         start = be32_to_cpu(p->sample_start);
990         if (start != 0) {
991                 spin_lock_irqsave(&dev->pending_lock, flags);
992                 if (dev->pma_sample_status == IB_PMA_SAMPLE_STATUS_DONE) {
993                         dev->pma_sample_status =
994                                 IB_PMA_SAMPLE_STATUS_STARTED;
995                         dev->pma_sample_start = start;
996                         dev->pma_sample_interval =
997                                 be32_to_cpu(p->sample_interval);
998                         dev->pma_tag = be16_to_cpu(p->tag);
999                         if (p->counter_select[0])
1000                                 dev->pma_counter_select[0] =
1001                                         p->counter_select[0];
1002                         if (p->counter_select[1])
1003                                 dev->pma_counter_select[1] =
1004                                         p->counter_select[1];
1005                         if (p->counter_select[2])
1006                                 dev->pma_counter_select[2] =
1007                                         p->counter_select[2];
1008                         if (p->counter_select[3])
1009                                 dev->pma_counter_select[3] =
1010                                         p->counter_select[3];
1011                         if (p->counter_select[4])
1012                                 dev->pma_counter_select[4] =
1013                                         p->counter_select[4];
1014                 }
1015                 spin_unlock_irqrestore(&dev->pending_lock, flags);
1016         }
1017         ret = recv_pma_get_portsamplescontrol(pmp, ibdev, port);
1018
1019 bail:
1020         return ret;
1021 }
1022
1023 static u64 get_counter(struct ipath_ibdev *dev, __be16 sel)
1024 {
1025         u64 ret;
1026
1027         switch (sel) {
1028         case IB_PMA_PORT_XMIT_DATA:
1029                 ret = dev->ipath_sword;
1030                 break;
1031         case IB_PMA_PORT_RCV_DATA:
1032                 ret = dev->ipath_rword;
1033                 break;
1034         case IB_PMA_PORT_XMIT_PKTS:
1035                 ret = dev->ipath_spkts;
1036                 break;
1037         case IB_PMA_PORT_RCV_PKTS:
1038                 ret = dev->ipath_rpkts;
1039                 break;
1040         case IB_PMA_PORT_XMIT_WAIT:
1041                 ret = dev->ipath_xmit_wait;
1042                 break;
1043         default:
1044                 ret = 0;
1045         }
1046
1047         return ret;
1048 }
1049
1050 static int recv_pma_get_portsamplesresult(struct ib_perf *pmp,
1051                                           struct ib_device *ibdev)
1052 {
1053         struct ib_pma_portsamplesresult *p =
1054                 (struct ib_pma_portsamplesresult *)pmp->data;
1055         struct ipath_ibdev *dev = to_idev(ibdev);
1056         int i;
1057
1058         memset(pmp->data, 0, sizeof(pmp->data));
1059         p->tag = cpu_to_be16(dev->pma_tag);
1060         p->sample_status = cpu_to_be16(dev->pma_sample_status);
1061         for (i = 0; i < ARRAY_SIZE(dev->pma_counter_select); i++)
1062                 p->counter[i] = cpu_to_be32(
1063                         get_counter(dev, dev->pma_counter_select[i]));
1064
1065         return reply((struct ib_smp *) pmp);
1066 }
1067
1068 static int recv_pma_get_portsamplesresult_ext(struct ib_perf *pmp,
1069                                               struct ib_device *ibdev)
1070 {
1071         struct ib_pma_portsamplesresult_ext *p =
1072                 (struct ib_pma_portsamplesresult_ext *)pmp->data;
1073         struct ipath_ibdev *dev = to_idev(ibdev);
1074         int i;
1075
1076         memset(pmp->data, 0, sizeof(pmp->data));
1077         p->tag = cpu_to_be16(dev->pma_tag);
1078         p->sample_status = cpu_to_be16(dev->pma_sample_status);
1079         /* 64 bits */
1080         p->extended_width = __constant_cpu_to_be32(0x80000000);
1081         for (i = 0; i < ARRAY_SIZE(dev->pma_counter_select); i++)
1082                 p->counter[i] = cpu_to_be64(
1083                         get_counter(dev, dev->pma_counter_select[i]));
1084
1085         return reply((struct ib_smp *) pmp);
1086 }
1087
1088 static int recv_pma_get_portcounters(struct ib_perf *pmp,
1089                                      struct ib_device *ibdev, u8 port)
1090 {
1091         struct ib_pma_portcounters *p = (struct ib_pma_portcounters *)
1092                 pmp->data;
1093         struct ipath_ibdev *dev = to_idev(ibdev);
1094         struct ipath_verbs_counters cntrs;
1095         u8 port_select = p->port_select;
1096
1097         ipath_get_counters(dev->dd, &cntrs);
1098
1099         /* Adjust counters for any resets done. */
1100         cntrs.symbol_error_counter -= dev->z_symbol_error_counter;
1101         cntrs.link_error_recovery_counter -=
1102                 dev->z_link_error_recovery_counter;
1103         cntrs.link_downed_counter -= dev->z_link_downed_counter;
1104         cntrs.port_rcv_errors += dev->rcv_errors;
1105         cntrs.port_rcv_errors -= dev->z_port_rcv_errors;
1106         cntrs.port_rcv_remphys_errors -= dev->z_port_rcv_remphys_errors;
1107         cntrs.port_xmit_discards -= dev->z_port_xmit_discards;
1108         cntrs.port_xmit_data -= dev->z_port_xmit_data;
1109         cntrs.port_rcv_data -= dev->z_port_rcv_data;
1110         cntrs.port_xmit_packets -= dev->z_port_xmit_packets;
1111         cntrs.port_rcv_packets -= dev->z_port_rcv_packets;
1112         cntrs.local_link_integrity_errors -=
1113                 dev->z_local_link_integrity_errors;
1114         cntrs.excessive_buffer_overrun_errors -=
1115                 dev->z_excessive_buffer_overrun_errors;
1116
1117         memset(pmp->data, 0, sizeof(pmp->data));
1118
1119         p->port_select = port_select;
1120         if (pmp->attr_mod != 0 ||
1121             (port_select != port && port_select != 0xFF))
1122                 pmp->status |= IB_SMP_INVALID_FIELD;
1123
1124         if (cntrs.symbol_error_counter > 0xFFFFUL)
1125                 p->symbol_error_counter = __constant_cpu_to_be16(0xFFFF);
1126         else
1127                 p->symbol_error_counter =
1128                         cpu_to_be16((u16)cntrs.symbol_error_counter);
1129         if (cntrs.link_error_recovery_counter > 0xFFUL)
1130                 p->link_error_recovery_counter = 0xFF;
1131         else
1132                 p->link_error_recovery_counter =
1133                         (u8)cntrs.link_error_recovery_counter;
1134         if (cntrs.link_downed_counter > 0xFFUL)
1135                 p->link_downed_counter = 0xFF;
1136         else
1137                 p->link_downed_counter = (u8)cntrs.link_downed_counter;
1138         if (cntrs.port_rcv_errors > 0xFFFFUL)
1139                 p->port_rcv_errors = __constant_cpu_to_be16(0xFFFF);
1140         else
1141                 p->port_rcv_errors =
1142                         cpu_to_be16((u16) cntrs.port_rcv_errors);
1143         if (cntrs.port_rcv_remphys_errors > 0xFFFFUL)
1144                 p->port_rcv_remphys_errors = __constant_cpu_to_be16(0xFFFF);
1145         else
1146                 p->port_rcv_remphys_errors =
1147                         cpu_to_be16((u16)cntrs.port_rcv_remphys_errors);
1148         if (cntrs.port_xmit_discards > 0xFFFFUL)
1149                 p->port_xmit_discards = __constant_cpu_to_be16(0xFFFF);
1150         else
1151                 p->port_xmit_discards =
1152                         cpu_to_be16((u16)cntrs.port_xmit_discards);
1153         if (cntrs.local_link_integrity_errors > 0xFUL)
1154                 cntrs.local_link_integrity_errors = 0xFUL;
1155         if (cntrs.excessive_buffer_overrun_errors > 0xFUL)
1156                 cntrs.excessive_buffer_overrun_errors = 0xFUL;
1157         p->lli_ebor_errors = (cntrs.local_link_integrity_errors << 4) |
1158                 cntrs.excessive_buffer_overrun_errors;
1159         if (dev->n_vl15_dropped > 0xFFFFUL)
1160                 p->vl15_dropped = __constant_cpu_to_be16(0xFFFF);
1161         else
1162                 p->vl15_dropped = cpu_to_be16((u16)dev->n_vl15_dropped);
1163         if (cntrs.port_xmit_data > 0xFFFFFFFFUL)
1164                 p->port_xmit_data = __constant_cpu_to_be32(0xFFFFFFFF);
1165         else
1166                 p->port_xmit_data = cpu_to_be32((u32)cntrs.port_xmit_data);
1167         if (cntrs.port_rcv_data > 0xFFFFFFFFUL)
1168                 p->port_rcv_data = __constant_cpu_to_be32(0xFFFFFFFF);
1169         else
1170                 p->port_rcv_data = cpu_to_be32((u32)cntrs.port_rcv_data);
1171         if (cntrs.port_xmit_packets > 0xFFFFFFFFUL)
1172                 p->port_xmit_packets = __constant_cpu_to_be32(0xFFFFFFFF);
1173         else
1174                 p->port_xmit_packets =
1175                         cpu_to_be32((u32)cntrs.port_xmit_packets);
1176         if (cntrs.port_rcv_packets > 0xFFFFFFFFUL)
1177                 p->port_rcv_packets = __constant_cpu_to_be32(0xFFFFFFFF);
1178         else
1179                 p->port_rcv_packets =
1180                         cpu_to_be32((u32) cntrs.port_rcv_packets);
1181
1182         return reply((struct ib_smp *) pmp);
1183 }
1184
1185 static int recv_pma_get_portcounters_ext(struct ib_perf *pmp,
1186                                          struct ib_device *ibdev, u8 port)
1187 {
1188         struct ib_pma_portcounters_ext *p =
1189                 (struct ib_pma_portcounters_ext *)pmp->data;
1190         struct ipath_ibdev *dev = to_idev(ibdev);
1191         u64 swords, rwords, spkts, rpkts, xwait;
1192         u8 port_select = p->port_select;
1193
1194         ipath_snapshot_counters(dev->dd, &swords, &rwords, &spkts,
1195                                 &rpkts, &xwait);
1196
1197         /* Adjust counters for any resets done. */
1198         swords -= dev->z_port_xmit_data;
1199         rwords -= dev->z_port_rcv_data;
1200         spkts -= dev->z_port_xmit_packets;
1201         rpkts -= dev->z_port_rcv_packets;
1202
1203         memset(pmp->data, 0, sizeof(pmp->data));
1204
1205         p->port_select = port_select;
1206         if (pmp->attr_mod != 0 ||
1207             (port_select != port && port_select != 0xFF))
1208                 pmp->status |= IB_SMP_INVALID_FIELD;
1209
1210         p->port_xmit_data = cpu_to_be64(swords);
1211         p->port_rcv_data = cpu_to_be64(rwords);
1212         p->port_xmit_packets = cpu_to_be64(spkts);
1213         p->port_rcv_packets = cpu_to_be64(rpkts);
1214         p->port_unicast_xmit_packets = cpu_to_be64(dev->n_unicast_xmit);
1215         p->port_unicast_rcv_packets = cpu_to_be64(dev->n_unicast_rcv);
1216         p->port_multicast_xmit_packets = cpu_to_be64(dev->n_multicast_xmit);
1217         p->port_multicast_rcv_packets = cpu_to_be64(dev->n_multicast_rcv);
1218
1219         return reply((struct ib_smp *) pmp);
1220 }
1221
1222 static int recv_pma_set_portcounters(struct ib_perf *pmp,
1223                                      struct ib_device *ibdev, u8 port)
1224 {
1225         struct ib_pma_portcounters *p = (struct ib_pma_portcounters *)
1226                 pmp->data;
1227         struct ipath_ibdev *dev = to_idev(ibdev);
1228         struct ipath_verbs_counters cntrs;
1229
1230         /*
1231          * Since the HW doesn't support clearing counters, we save the
1232          * current count and subtract it from future responses.
1233          */
1234         ipath_get_counters(dev->dd, &cntrs);
1235
1236         if (p->counter_select & IB_PMA_SEL_SYMBOL_ERROR)
1237                 dev->z_symbol_error_counter = cntrs.symbol_error_counter;
1238
1239         if (p->counter_select & IB_PMA_SEL_LINK_ERROR_RECOVERY)
1240                 dev->z_link_error_recovery_counter =
1241                         cntrs.link_error_recovery_counter;
1242
1243         if (p->counter_select & IB_PMA_SEL_LINK_DOWNED)
1244                 dev->z_link_downed_counter = cntrs.link_downed_counter;
1245
1246         if (p->counter_select & IB_PMA_SEL_PORT_RCV_ERRORS)
1247                 dev->z_port_rcv_errors =
1248                         cntrs.port_rcv_errors + dev->rcv_errors;
1249
1250         if (p->counter_select & IB_PMA_SEL_PORT_RCV_REMPHYS_ERRORS)
1251                 dev->z_port_rcv_remphys_errors =
1252                         cntrs.port_rcv_remphys_errors;
1253
1254         if (p->counter_select & IB_PMA_SEL_PORT_XMIT_DISCARDS)
1255                 dev->z_port_xmit_discards = cntrs.port_xmit_discards;
1256
1257         if (p->counter_select & IB_PMA_SEL_LOCAL_LINK_INTEGRITY_ERRORS)
1258                 dev->z_local_link_integrity_errors =
1259                         cntrs.local_link_integrity_errors;
1260
1261         if (p->counter_select & IB_PMA_SEL_EXCESSIVE_BUFFER_OVERRUNS)
1262                 dev->z_excessive_buffer_overrun_errors =
1263                         cntrs.excessive_buffer_overrun_errors;
1264
1265         if (p->counter_select & IB_PMA_SEL_PORT_VL15_DROPPED)
1266                 dev->n_vl15_dropped = 0;
1267
1268         if (p->counter_select & IB_PMA_SEL_PORT_XMIT_DATA)
1269                 dev->z_port_xmit_data = cntrs.port_xmit_data;
1270
1271         if (p->counter_select & IB_PMA_SEL_PORT_RCV_DATA)
1272                 dev->z_port_rcv_data = cntrs.port_rcv_data;
1273
1274         if (p->counter_select & IB_PMA_SEL_PORT_XMIT_PACKETS)
1275                 dev->z_port_xmit_packets = cntrs.port_xmit_packets;
1276
1277         if (p->counter_select & IB_PMA_SEL_PORT_RCV_PACKETS)
1278                 dev->z_port_rcv_packets = cntrs.port_rcv_packets;
1279
1280         return recv_pma_get_portcounters(pmp, ibdev, port);
1281 }
1282
1283 static int recv_pma_set_portcounters_ext(struct ib_perf *pmp,
1284                                          struct ib_device *ibdev, u8 port)
1285 {
1286         struct ib_pma_portcounters *p = (struct ib_pma_portcounters *)
1287                 pmp->data;
1288         struct ipath_ibdev *dev = to_idev(ibdev);
1289         u64 swords, rwords, spkts, rpkts, xwait;
1290
1291         ipath_snapshot_counters(dev->dd, &swords, &rwords, &spkts,
1292                                 &rpkts, &xwait);
1293
1294         if (p->counter_select & IB_PMA_SELX_PORT_XMIT_DATA)
1295                 dev->z_port_xmit_data = swords;
1296
1297         if (p->counter_select & IB_PMA_SELX_PORT_RCV_DATA)
1298                 dev->z_port_rcv_data = rwords;
1299
1300         if (p->counter_select & IB_PMA_SELX_PORT_XMIT_PACKETS)
1301                 dev->z_port_xmit_packets = spkts;
1302
1303         if (p->counter_select & IB_PMA_SELX_PORT_RCV_PACKETS)
1304                 dev->z_port_rcv_packets = rpkts;
1305
1306         if (p->counter_select & IB_PMA_SELX_PORT_UNI_XMIT_PACKETS)
1307                 dev->n_unicast_xmit = 0;
1308
1309         if (p->counter_select & IB_PMA_SELX_PORT_UNI_RCV_PACKETS)
1310                 dev->n_unicast_rcv = 0;
1311
1312         if (p->counter_select & IB_PMA_SELX_PORT_MULTI_XMIT_PACKETS)
1313                 dev->n_multicast_xmit = 0;
1314
1315         if (p->counter_select & IB_PMA_SELX_PORT_MULTI_RCV_PACKETS)
1316                 dev->n_multicast_rcv = 0;
1317
1318         return recv_pma_get_portcounters_ext(pmp, ibdev, port);
1319 }
1320
1321 static int process_subn(struct ib_device *ibdev, int mad_flags,
1322                         u8 port_num, struct ib_mad *in_mad,
1323                         struct ib_mad *out_mad)
1324 {
1325         struct ib_smp *smp = (struct ib_smp *)out_mad;
1326         struct ipath_ibdev *dev = to_idev(ibdev);
1327         int ret;
1328
1329         *out_mad = *in_mad;
1330         if (smp->class_version != 1) {
1331                 smp->status |= IB_SMP_UNSUP_VERSION;
1332                 ret = reply(smp);
1333                 goto bail;
1334         }
1335
1336         /* Is the mkey in the process of expiring? */
1337         if (dev->mkey_lease_timeout && jiffies >= dev->mkey_lease_timeout) {
1338                 /* Clear timeout and mkey protection field. */
1339                 dev->mkey_lease_timeout = 0;
1340                 dev->mkeyprot = 0;
1341         }
1342
1343         /*
1344          * M_Key checking depends on
1345          * Portinfo:M_Key_protect_bits
1346          */
1347         if ((mad_flags & IB_MAD_IGNORE_MKEY) == 0 && dev->mkey != 0 &&
1348             dev->mkey != smp->mkey &&
1349             (smp->method == IB_MGMT_METHOD_SET ||
1350              (smp->method == IB_MGMT_METHOD_GET &&
1351               dev->mkeyprot >= 2))) {
1352                 if (dev->mkey_violations != 0xFFFF)
1353                         ++dev->mkey_violations;
1354                 if (dev->mkey_lease_timeout ||
1355                     dev->mkey_lease_period == 0) {
1356                         ret = IB_MAD_RESULT_SUCCESS |
1357                                 IB_MAD_RESULT_CONSUMED;
1358                         goto bail;
1359                 }
1360                 dev->mkey_lease_timeout = jiffies +
1361                         dev->mkey_lease_period * HZ;
1362                 /* Future: Generate a trap notice. */
1363                 ret = IB_MAD_RESULT_SUCCESS | IB_MAD_RESULT_CONSUMED;
1364                 goto bail;
1365         } else if (dev->mkey_lease_timeout)
1366                 dev->mkey_lease_timeout = 0;
1367
1368         switch (smp->method) {
1369         case IB_MGMT_METHOD_GET:
1370                 switch (smp->attr_id) {
1371                 case IB_SMP_ATTR_NODE_DESC:
1372                         ret = recv_subn_get_nodedescription(smp, ibdev);
1373                         goto bail;
1374                 case IB_SMP_ATTR_NODE_INFO:
1375                         ret = recv_subn_get_nodeinfo(smp, ibdev, port_num);
1376                         goto bail;
1377                 case IB_SMP_ATTR_GUID_INFO:
1378                         ret = recv_subn_get_guidinfo(smp, ibdev);
1379                         goto bail;
1380                 case IB_SMP_ATTR_PORT_INFO:
1381                         ret = recv_subn_get_portinfo(smp, ibdev, port_num);
1382                         goto bail;
1383                 case IB_SMP_ATTR_PKEY_TABLE:
1384                         ret = recv_subn_get_pkeytable(smp, ibdev);
1385                         goto bail;
1386                 case IB_SMP_ATTR_SM_INFO:
1387                         if (dev->port_cap_flags & IB_PORT_SM_DISABLED) {
1388                                 ret = IB_MAD_RESULT_SUCCESS |
1389                                         IB_MAD_RESULT_CONSUMED;
1390                                 goto bail;
1391                         }
1392                         if (dev->port_cap_flags & IB_PORT_SM) {
1393                                 ret = IB_MAD_RESULT_SUCCESS;
1394                                 goto bail;
1395                         }
1396                         /* FALLTHROUGH */
1397                 default:
1398                         smp->status |= IB_SMP_UNSUP_METH_ATTR;
1399                         ret = reply(smp);
1400                         goto bail;
1401                 }
1402
1403         case IB_MGMT_METHOD_SET:
1404                 switch (smp->attr_id) {
1405                 case IB_SMP_ATTR_GUID_INFO:
1406                         ret = recv_subn_set_guidinfo(smp, ibdev);
1407                         goto bail;
1408                 case IB_SMP_ATTR_PORT_INFO:
1409                         ret = recv_subn_set_portinfo(smp, ibdev, port_num);
1410                         goto bail;
1411                 case IB_SMP_ATTR_PKEY_TABLE:
1412                         ret = recv_subn_set_pkeytable(smp, ibdev);
1413                         goto bail;
1414                 case IB_SMP_ATTR_SM_INFO:
1415                         if (dev->port_cap_flags & IB_PORT_SM_DISABLED) {
1416                                 ret = IB_MAD_RESULT_SUCCESS |
1417                                         IB_MAD_RESULT_CONSUMED;
1418                                 goto bail;
1419                         }
1420                         if (dev->port_cap_flags & IB_PORT_SM) {
1421                                 ret = IB_MAD_RESULT_SUCCESS;
1422                                 goto bail;
1423                         }
1424                         /* FALLTHROUGH */
1425                 default:
1426                         smp->status |= IB_SMP_UNSUP_METH_ATTR;
1427                         ret = reply(smp);
1428                         goto bail;
1429                 }
1430
1431         case IB_MGMT_METHOD_GET_RESP:
1432                 /*
1433                  * The ib_mad module will call us to process responses
1434                  * before checking for other consumers.
1435                  * Just tell the caller to process it normally.
1436                  */
1437                 ret = IB_MAD_RESULT_FAILURE;
1438                 goto bail;
1439         default:
1440                 smp->status |= IB_SMP_UNSUP_METHOD;
1441                 ret = reply(smp);
1442         }
1443
1444 bail:
1445         return ret;
1446 }
1447
1448 static int process_perf(struct ib_device *ibdev, u8 port_num,
1449                         struct ib_mad *in_mad,
1450                         struct ib_mad *out_mad)
1451 {
1452         struct ib_perf *pmp = (struct ib_perf *)out_mad;
1453         int ret;
1454
1455         *out_mad = *in_mad;
1456         if (pmp->class_version != 1) {
1457                 pmp->status |= IB_SMP_UNSUP_VERSION;
1458                 ret = reply((struct ib_smp *) pmp);
1459                 goto bail;
1460         }
1461
1462         switch (pmp->method) {
1463         case IB_MGMT_METHOD_GET:
1464                 switch (pmp->attr_id) {
1465                 case IB_PMA_CLASS_PORT_INFO:
1466                         ret = recv_pma_get_classportinfo(pmp);
1467                         goto bail;
1468                 case IB_PMA_PORT_SAMPLES_CONTROL:
1469                         ret = recv_pma_get_portsamplescontrol(pmp, ibdev,
1470                                                               port_num);
1471                         goto bail;
1472                 case IB_PMA_PORT_SAMPLES_RESULT:
1473                         ret = recv_pma_get_portsamplesresult(pmp, ibdev);
1474                         goto bail;
1475                 case IB_PMA_PORT_SAMPLES_RESULT_EXT:
1476                         ret = recv_pma_get_portsamplesresult_ext(pmp,
1477                                                                  ibdev);
1478                         goto bail;
1479                 case IB_PMA_PORT_COUNTERS:
1480                         ret = recv_pma_get_portcounters(pmp, ibdev,
1481                                                         port_num);
1482                         goto bail;
1483                 case IB_PMA_PORT_COUNTERS_EXT:
1484                         ret = recv_pma_get_portcounters_ext(pmp, ibdev,
1485                                                             port_num);
1486                         goto bail;
1487                 default:
1488                         pmp->status |= IB_SMP_UNSUP_METH_ATTR;
1489                         ret = reply((struct ib_smp *) pmp);
1490                         goto bail;
1491                 }
1492
1493         case IB_MGMT_METHOD_SET:
1494                 switch (pmp->attr_id) {
1495                 case IB_PMA_PORT_SAMPLES_CONTROL:
1496                         ret = recv_pma_set_portsamplescontrol(pmp, ibdev,
1497                                                               port_num);
1498                         goto bail;
1499                 case IB_PMA_PORT_COUNTERS:
1500                         ret = recv_pma_set_portcounters(pmp, ibdev,
1501                                                         port_num);
1502                         goto bail;
1503                 case IB_PMA_PORT_COUNTERS_EXT:
1504                         ret = recv_pma_set_portcounters_ext(pmp, ibdev,
1505                                                             port_num);
1506                         goto bail;
1507                 default:
1508                         pmp->status |= IB_SMP_UNSUP_METH_ATTR;
1509                         ret = reply((struct ib_smp *) pmp);
1510                         goto bail;
1511                 }
1512
1513         case IB_MGMT_METHOD_GET_RESP:
1514                 /*
1515                  * The ib_mad module will call us to process responses
1516                  * before checking for other consumers.
1517                  * Just tell the caller to process it normally.
1518                  */
1519                 ret = IB_MAD_RESULT_FAILURE;
1520                 goto bail;
1521         default:
1522                 pmp->status |= IB_SMP_UNSUP_METHOD;
1523                 ret = reply((struct ib_smp *) pmp);
1524         }
1525
1526 bail:
1527         return ret;
1528 }
1529
1530 /**
1531  * ipath_process_mad - process an incoming MAD packet
1532  * @ibdev: the infiniband device this packet came in on
1533  * @mad_flags: MAD flags
1534  * @port_num: the port number this packet came in on
1535  * @in_wc: the work completion entry for this packet
1536  * @in_grh: the global route header for this packet
1537  * @in_mad: the incoming MAD
1538  * @out_mad: any outgoing MAD reply
1539  *
1540  * Returns IB_MAD_RESULT_SUCCESS if this is a MAD that we are not
1541  * interested in processing.
1542  *
1543  * Note that the verbs framework has already done the MAD sanity checks,
1544  * and hop count/pointer updating for IB_MGMT_CLASS_SUBN_DIRECTED_ROUTE
1545  * MADs.
1546  *
1547  * This is called by the ib_mad module.
1548  */
1549 int ipath_process_mad(struct ib_device *ibdev, int mad_flags, u8 port_num,
1550                       struct ib_wc *in_wc, struct ib_grh *in_grh,
1551                       struct ib_mad *in_mad, struct ib_mad *out_mad)
1552 {
1553         int ret;
1554
1555         switch (in_mad->mad_hdr.mgmt_class) {
1556         case IB_MGMT_CLASS_SUBN_DIRECTED_ROUTE:
1557         case IB_MGMT_CLASS_SUBN_LID_ROUTED:
1558                 ret = process_subn(ibdev, mad_flags, port_num,
1559                                    in_mad, out_mad);
1560                 goto bail;
1561         case IB_MGMT_CLASS_PERF_MGMT:
1562                 ret = process_perf(ibdev, port_num, in_mad, out_mad);
1563                 goto bail;
1564         default:
1565                 ret = IB_MAD_RESULT_SUCCESS;
1566         }
1567
1568 bail:
1569         return ret;
1570 }