Merge git://git.kernel.org/pub/scm/linux/kernel/git/mingo/linux-2.6-sched
[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_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         /*
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         u32 flags;
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         event.device = ibdev;
419         event.element.port_num = port;
420
421         dev->mkey = pip->mkey;
422         dev->gid_prefix = pip->gid_prefix;
423         dev->mkey_lease_period = be16_to_cpu(pip->mkey_lease_period);
424
425         lid = be16_to_cpu(pip->lid);
426         if (lid != dev->dd->ipath_lid) {
427                 /* Must be a valid unicast LID address. */
428                 if (lid == 0 || lid >= IPATH_MULTICAST_LID_BASE)
429                         goto err;
430                 ipath_set_lid(dev->dd, lid, pip->mkeyprot_resv_lmc & 7);
431                 event.event = IB_EVENT_LID_CHANGE;
432                 ib_dispatch_event(&event);
433         }
434
435         smlid = be16_to_cpu(pip->sm_lid);
436         if (smlid != dev->sm_lid) {
437                 /* Must be a valid unicast LID address. */
438                 if (smlid == 0 || smlid >= IPATH_MULTICAST_LID_BASE)
439                         goto err;
440                 dev->sm_lid = smlid;
441                 event.event = IB_EVENT_SM_CHANGE;
442                 ib_dispatch_event(&event);
443         }
444
445         /* Only 4x supported but allow 1x or 4x to be set (see 14.2.6.6). */
446         lwe = pip->link_width_enabled;
447         if ((lwe >= 4 && lwe <= 8) || (lwe >= 0xC && lwe <= 0xFE))
448                 goto err;
449         if (lwe == 0xFF)
450                 dev->link_width_enabled = 3;    /* 1x or 4x */
451         else if (lwe)
452                 dev->link_width_enabled = lwe;
453
454         /* Only 2.5 Gbs supported. */
455         lse = pip->linkspeedactive_enabled & 0xF;
456         if (lse >= 2 && lse <= 0xE)
457                 goto err;
458
459         /* Set link down default state. */
460         switch (pip->portphysstate_linkdown & 0xF) {
461         case 0: /* NOP */
462                 break;
463         case 1: /* SLEEP */
464                 if (set_linkdowndefaultstate(dev->dd, 1))
465                         goto err;
466                 break;
467         case 2: /* POLL */
468                 if (set_linkdowndefaultstate(dev->dd, 0))
469                         goto err;
470                 break;
471         default:
472                 goto err;
473         }
474
475         dev->mkeyprot_resv_lmc = pip->mkeyprot_resv_lmc;
476         dev->vl_high_limit = pip->vl_high_limit;
477
478         switch ((pip->neighbormtu_mastersmsl >> 4) & 0xF) {
479         case IB_MTU_256:
480                 mtu = 256;
481                 break;
482         case IB_MTU_512:
483                 mtu = 512;
484                 break;
485         case IB_MTU_1024:
486                 mtu = 1024;
487                 break;
488         case IB_MTU_2048:
489                 mtu = 2048;
490                 break;
491         case IB_MTU_4096:
492                 mtu = 4096;
493                 break;
494         default:
495                 /* XXX We have already partially updated our state! */
496                 goto err;
497         }
498         ipath_set_mtu(dev->dd, mtu);
499
500         dev->sm_sl = pip->neighbormtu_mastersmsl & 0xF;
501
502         /* We only support VL0 */
503         if (((pip->operationalvl_pei_peo_fpi_fpo >> 4) & 0xF) > 1)
504                 goto err;
505
506         if (pip->mkey_violations == 0)
507                 dev->mkey_violations = 0;
508
509         /*
510          * Hardware counter can't be reset so snapshot and subtract
511          * later.
512          */
513         if (pip->pkey_violations == 0)
514                 dev->z_pkey_violations = ipath_get_cr_errpkey(dev->dd);
515
516         if (pip->qkey_violations == 0)
517                 dev->qkey_violations = 0;
518
519         ore = pip->localphyerrors_overrunerrors;
520         if (set_phyerrthreshold(dev->dd, (ore >> 4) & 0xF))
521                 goto err;
522
523         if (set_overrunthreshold(dev->dd, (ore & 0xF)))
524                 goto err;
525
526         dev->subnet_timeout = pip->clientrereg_resv_subnetto & 0x1F;
527
528         if (pip->clientrereg_resv_subnetto & 0x80) {
529                 clientrereg = 1;
530                 event.event = IB_EVENT_CLIENT_REREGISTER;
531                 ib_dispatch_event(&event);
532         }
533
534         /*
535          * Do the port state change now that the other link parameters
536          * have been set.
537          * Changing the port physical state only makes sense if the link
538          * is down or is being set to down.
539          */
540         state = pip->linkspeed_portstate & 0xF;
541         flags = dev->dd->ipath_flags;
542         lstate = (pip->portphysstate_linkdown >> 4) & 0xF;
543         if (lstate && !(state == IB_PORT_DOWN || state == IB_PORT_NOP))
544                 goto err;
545
546         /*
547          * Only state changes of DOWN, ARM, and ACTIVE are valid
548          * and must be in the correct state to take effect (see 7.2.6).
549          */
550         switch (state) {
551         case IB_PORT_NOP:
552                 if (lstate == 0)
553                         break;
554                 /* FALLTHROUGH */
555         case IB_PORT_DOWN:
556                 if (lstate == 0)
557                         if (get_linkdowndefaultstate(dev->dd))
558                                 lstate = IPATH_IB_LINKDOWN_SLEEP;
559                         else
560                                 lstate = IPATH_IB_LINKDOWN;
561                 else if (lstate == 1)
562                         lstate = IPATH_IB_LINKDOWN_SLEEP;
563                 else if (lstate == 2)
564                         lstate = IPATH_IB_LINKDOWN;
565                 else if (lstate == 3)
566                         lstate = IPATH_IB_LINKDOWN_DISABLE;
567                 else
568                         goto err;
569                 ipath_set_linkstate(dev->dd, lstate);
570                 if (flags & IPATH_LINKACTIVE) {
571                         event.event = IB_EVENT_PORT_ERR;
572                         ib_dispatch_event(&event);
573                 }
574                 break;
575         case IB_PORT_ARMED:
576                 if (!(flags & (IPATH_LINKINIT | IPATH_LINKACTIVE)))
577                         break;
578                 ipath_set_linkstate(dev->dd, IPATH_IB_LINKARM);
579                 if (flags & IPATH_LINKACTIVE) {
580                         event.event = IB_EVENT_PORT_ERR;
581                         ib_dispatch_event(&event);
582                 }
583                 break;
584         case IB_PORT_ACTIVE:
585                 if (!(flags & IPATH_LINKARMED))
586                         break;
587                 ipath_set_linkstate(dev->dd, IPATH_IB_LINKACTIVE);
588                 event.event = IB_EVENT_PORT_ACTIVE;
589                 ib_dispatch_event(&event);
590                 break;
591         default:
592                 /* XXX We have already partially updated our state! */
593                 goto err;
594         }
595
596         ret = recv_subn_get_portinfo(smp, ibdev, port);
597
598         if (clientrereg)
599                 pip->clientrereg_resv_subnetto |= 0x80;
600
601         goto done;
602
603 err:
604         smp->status |= IB_SMP_INVALID_FIELD;
605         ret = recv_subn_get_portinfo(smp, ibdev, port);
606
607 done:
608         return ret;
609 }
610
611 /**
612  * rm_pkey - decrecment the reference count for the given PKEY
613  * @dd: the infinipath device
614  * @key: the PKEY index
615  *
616  * Return true if this was the last reference and the hardware table entry
617  * needs to be changed.
618  */
619 static int rm_pkey(struct ipath_devdata *dd, u16 key)
620 {
621         int i;
622         int ret;
623
624         for (i = 0; i < ARRAY_SIZE(dd->ipath_pkeys); i++) {
625                 if (dd->ipath_pkeys[i] != key)
626                         continue;
627                 if (atomic_dec_and_test(&dd->ipath_pkeyrefs[i])) {
628                         dd->ipath_pkeys[i] = 0;
629                         ret = 1;
630                         goto bail;
631                 }
632                 break;
633         }
634
635         ret = 0;
636
637 bail:
638         return ret;
639 }
640
641 /**
642  * add_pkey - add the given PKEY to the hardware table
643  * @dd: the infinipath device
644  * @key: the PKEY
645  *
646  * Return an error code if unable to add the entry, zero if no change,
647  * or 1 if the hardware PKEY register needs to be updated.
648  */
649 static int add_pkey(struct ipath_devdata *dd, u16 key)
650 {
651         int i;
652         u16 lkey = key & 0x7FFF;
653         int any = 0;
654         int ret;
655
656         if (lkey == 0x7FFF) {
657                 ret = 0;
658                 goto bail;
659         }
660
661         /* Look for an empty slot or a matching PKEY. */
662         for (i = 0; i < ARRAY_SIZE(dd->ipath_pkeys); i++) {
663                 if (!dd->ipath_pkeys[i]) {
664                         any++;
665                         continue;
666                 }
667                 /* If it matches exactly, try to increment the ref count */
668                 if (dd->ipath_pkeys[i] == key) {
669                         if (atomic_inc_return(&dd->ipath_pkeyrefs[i]) > 1) {
670                                 ret = 0;
671                                 goto bail;
672                         }
673                         /* Lost the race. Look for an empty slot below. */
674                         atomic_dec(&dd->ipath_pkeyrefs[i]);
675                         any++;
676                 }
677                 /*
678                  * It makes no sense to have both the limited and unlimited
679                  * PKEY set at the same time since the unlimited one will
680                  * disable the limited one.
681                  */
682                 if ((dd->ipath_pkeys[i] & 0x7FFF) == lkey) {
683                         ret = -EEXIST;
684                         goto bail;
685                 }
686         }
687         if (!any) {
688                 ret = -EBUSY;
689                 goto bail;
690         }
691         for (i = 0; i < ARRAY_SIZE(dd->ipath_pkeys); i++) {
692                 if (!dd->ipath_pkeys[i] &&
693                     atomic_inc_return(&dd->ipath_pkeyrefs[i]) == 1) {
694                         /* for ipathstats, etc. */
695                         ipath_stats.sps_pkeys[i] = lkey;
696                         dd->ipath_pkeys[i] = key;
697                         ret = 1;
698                         goto bail;
699                 }
700         }
701         ret = -EBUSY;
702
703 bail:
704         return ret;
705 }
706
707 /**
708  * set_pkeys - set the PKEY table for port 0
709  * @dd: the infinipath device
710  * @pkeys: the PKEY table
711  */
712 static int set_pkeys(struct ipath_devdata *dd, u16 *pkeys)
713 {
714         struct ipath_portdata *pd;
715         int i;
716         int changed = 0;
717
718         pd = dd->ipath_pd[0];
719
720         for (i = 0; i < ARRAY_SIZE(pd->port_pkeys); i++) {
721                 u16 key = pkeys[i];
722                 u16 okey = pd->port_pkeys[i];
723
724                 if (key == okey)
725                         continue;
726                 /*
727                  * The value of this PKEY table entry is changing.
728                  * Remove the old entry in the hardware's array of PKEYs.
729                  */
730                 if (okey & 0x7FFF)
731                         changed |= rm_pkey(dd, okey);
732                 if (key & 0x7FFF) {
733                         int ret = add_pkey(dd, key);
734
735                         if (ret < 0)
736                                 key = 0;
737                         else
738                                 changed |= ret;
739                 }
740                 pd->port_pkeys[i] = key;
741         }
742         if (changed) {
743                 u64 pkey;
744
745                 pkey = (u64) dd->ipath_pkeys[0] |
746                         ((u64) dd->ipath_pkeys[1] << 16) |
747                         ((u64) dd->ipath_pkeys[2] << 32) |
748                         ((u64) dd->ipath_pkeys[3] << 48);
749                 ipath_cdbg(VERBOSE, "p0 new pkey reg %llx\n",
750                            (unsigned long long) pkey);
751                 ipath_write_kreg(dd, dd->ipath_kregs->kr_partitionkey,
752                                  pkey);
753         }
754         return 0;
755 }
756
757 static int recv_subn_set_pkeytable(struct ib_smp *smp,
758                                    struct ib_device *ibdev)
759 {
760         u32 startpx = 32 * (be32_to_cpu(smp->attr_mod) & 0xffff);
761         __be16 *p = (__be16 *) smp->data;
762         u16 *q = (u16 *) smp->data;
763         struct ipath_ibdev *dev = to_idev(ibdev);
764         unsigned i, n = ipath_get_npkeys(dev->dd);
765
766         for (i = 0; i < n; i++)
767                 q[i] = be16_to_cpu(p[i]);
768
769         if (startpx != 0 || set_pkeys(dev->dd, q) != 0)
770                 smp->status |= IB_SMP_INVALID_FIELD;
771
772         return recv_subn_get_pkeytable(smp, ibdev);
773 }
774
775 #define IB_PMA_CLASS_PORT_INFO          __constant_htons(0x0001)
776 #define IB_PMA_PORT_SAMPLES_CONTROL     __constant_htons(0x0010)
777 #define IB_PMA_PORT_SAMPLES_RESULT      __constant_htons(0x0011)
778 #define IB_PMA_PORT_COUNTERS            __constant_htons(0x0012)
779 #define IB_PMA_PORT_COUNTERS_EXT        __constant_htons(0x001D)
780 #define IB_PMA_PORT_SAMPLES_RESULT_EXT  __constant_htons(0x001E)
781
782 struct ib_perf {
783         u8 base_version;
784         u8 mgmt_class;
785         u8 class_version;
786         u8 method;
787         __be16 status;
788         __be16 unused;
789         __be64 tid;
790         __be16 attr_id;
791         __be16 resv;
792         __be32 attr_mod;
793         u8 reserved[40];
794         u8 data[192];
795 } __attribute__ ((packed));
796
797 struct ib_pma_classportinfo {
798         u8 base_version;
799         u8 class_version;
800         __be16 cap_mask;
801         u8 reserved[3];
802         u8 resp_time_value;     /* only lower 5 bits */
803         union ib_gid redirect_gid;
804         __be32 redirect_tc_sl_fl;       /* 8, 4, 20 bits respectively */
805         __be16 redirect_lid;
806         __be16 redirect_pkey;
807         __be32 redirect_qp;     /* only lower 24 bits */
808         __be32 redirect_qkey;
809         union ib_gid trap_gid;
810         __be32 trap_tc_sl_fl;   /* 8, 4, 20 bits respectively */
811         __be16 trap_lid;
812         __be16 trap_pkey;
813         __be32 trap_hl_qp;      /* 8, 24 bits respectively */
814         __be32 trap_qkey;
815 } __attribute__ ((packed));
816
817 struct ib_pma_portsamplescontrol {
818         u8 opcode;
819         u8 port_select;
820         u8 tick;
821         u8 counter_width;       /* only lower 3 bits */
822         __be32 counter_mask0_9; /* 2, 10 * 3, bits */
823         __be16 counter_mask10_14;       /* 1, 5 * 3, bits */
824         u8 sample_mechanisms;
825         u8 sample_status;       /* only lower 2 bits */
826         __be64 option_mask;
827         __be64 vendor_mask;
828         __be32 sample_start;
829         __be32 sample_interval;
830         __be16 tag;
831         __be16 counter_select[15];
832 } __attribute__ ((packed));
833
834 struct ib_pma_portsamplesresult {
835         __be16 tag;
836         __be16 sample_status;   /* only lower 2 bits */
837         __be32 counter[15];
838 } __attribute__ ((packed));
839
840 struct ib_pma_portsamplesresult_ext {
841         __be16 tag;
842         __be16 sample_status;   /* only lower 2 bits */
843         __be32 extended_width;  /* only upper 2 bits */
844         __be64 counter[15];
845 } __attribute__ ((packed));
846
847 struct ib_pma_portcounters {
848         u8 reserved;
849         u8 port_select;
850         __be16 counter_select;
851         __be16 symbol_error_counter;
852         u8 link_error_recovery_counter;
853         u8 link_downed_counter;
854         __be16 port_rcv_errors;
855         __be16 port_rcv_remphys_errors;
856         __be16 port_rcv_switch_relay_errors;
857         __be16 port_xmit_discards;
858         u8 port_xmit_constraint_errors;
859         u8 port_rcv_constraint_errors;
860         u8 reserved1;
861         u8 lli_ebor_errors;     /* 4, 4, bits */
862         __be16 reserved2;
863         __be16 vl15_dropped;
864         __be32 port_xmit_data;
865         __be32 port_rcv_data;
866         __be32 port_xmit_packets;
867         __be32 port_rcv_packets;
868 } __attribute__ ((packed));
869
870 #define IB_PMA_SEL_SYMBOL_ERROR                 __constant_htons(0x0001)
871 #define IB_PMA_SEL_LINK_ERROR_RECOVERY          __constant_htons(0x0002)
872 #define IB_PMA_SEL_LINK_DOWNED                  __constant_htons(0x0004)
873 #define IB_PMA_SEL_PORT_RCV_ERRORS              __constant_htons(0x0008)
874 #define IB_PMA_SEL_PORT_RCV_REMPHYS_ERRORS      __constant_htons(0x0010)
875 #define IB_PMA_SEL_PORT_XMIT_DISCARDS           __constant_htons(0x0040)
876 #define IB_PMA_SEL_LOCAL_LINK_INTEGRITY_ERRORS  __constant_htons(0x0200)
877 #define IB_PMA_SEL_EXCESSIVE_BUFFER_OVERRUNS    __constant_htons(0x0400)
878 #define IB_PMA_SEL_PORT_VL15_DROPPED            __constant_htons(0x0800)
879 #define IB_PMA_SEL_PORT_XMIT_DATA               __constant_htons(0x1000)
880 #define IB_PMA_SEL_PORT_RCV_DATA                __constant_htons(0x2000)
881 #define IB_PMA_SEL_PORT_XMIT_PACKETS            __constant_htons(0x4000)
882 #define IB_PMA_SEL_PORT_RCV_PACKETS             __constant_htons(0x8000)
883
884 struct ib_pma_portcounters_ext {
885         u8 reserved;
886         u8 port_select;
887         __be16 counter_select;
888         __be32 reserved1;
889         __be64 port_xmit_data;
890         __be64 port_rcv_data;
891         __be64 port_xmit_packets;
892         __be64 port_rcv_packets;
893         __be64 port_unicast_xmit_packets;
894         __be64 port_unicast_rcv_packets;
895         __be64 port_multicast_xmit_packets;
896         __be64 port_multicast_rcv_packets;
897 } __attribute__ ((packed));
898
899 #define IB_PMA_SELX_PORT_XMIT_DATA              __constant_htons(0x0001)
900 #define IB_PMA_SELX_PORT_RCV_DATA               __constant_htons(0x0002)
901 #define IB_PMA_SELX_PORT_XMIT_PACKETS           __constant_htons(0x0004)
902 #define IB_PMA_SELX_PORT_RCV_PACKETS            __constant_htons(0x0008)
903 #define IB_PMA_SELX_PORT_UNI_XMIT_PACKETS       __constant_htons(0x0010)
904 #define IB_PMA_SELX_PORT_UNI_RCV_PACKETS        __constant_htons(0x0020)
905 #define IB_PMA_SELX_PORT_MULTI_XMIT_PACKETS     __constant_htons(0x0040)
906 #define IB_PMA_SELX_PORT_MULTI_RCV_PACKETS      __constant_htons(0x0080)
907
908 static int recv_pma_get_classportinfo(struct ib_perf *pmp)
909 {
910         struct ib_pma_classportinfo *p =
911                 (struct ib_pma_classportinfo *)pmp->data;
912
913         memset(pmp->data, 0, sizeof(pmp->data));
914
915         if (pmp->attr_mod != 0)
916                 pmp->status |= IB_SMP_INVALID_FIELD;
917
918         /* Indicate AllPortSelect is valid (only one port anyway) */
919         p->cap_mask = __constant_cpu_to_be16(1 << 8);
920         p->base_version = 1;
921         p->class_version = 1;
922         /*
923          * Expected response time is 4.096 usec. * 2^18 == 1.073741824
924          * sec.
925          */
926         p->resp_time_value = 18;
927
928         return reply((struct ib_smp *) pmp);
929 }
930
931 /*
932  * The PortSamplesControl.CounterMasks field is an array of 3 bit fields
933  * which specify the N'th counter's capabilities. See ch. 16.1.3.2.
934  * We support 5 counters which only count the mandatory quantities.
935  */
936 #define COUNTER_MASK(q, n) (q << ((9 - n) * 3))
937 #define COUNTER_MASK0_9 \
938         __constant_cpu_to_be32(COUNTER_MASK(1, 0) | \
939                                COUNTER_MASK(1, 1) | \
940                                COUNTER_MASK(1, 2) | \
941                                COUNTER_MASK(1, 3) | \
942                                COUNTER_MASK(1, 4))
943
944 static int recv_pma_get_portsamplescontrol(struct ib_perf *pmp,
945                                            struct ib_device *ibdev, u8 port)
946 {
947         struct ib_pma_portsamplescontrol *p =
948                 (struct ib_pma_portsamplescontrol *)pmp->data;
949         struct ipath_ibdev *dev = to_idev(ibdev);
950         unsigned long flags;
951         u8 port_select = p->port_select;
952
953         memset(pmp->data, 0, sizeof(pmp->data));
954
955         p->port_select = port_select;
956         if (pmp->attr_mod != 0 ||
957             (port_select != port && port_select != 0xFF))
958                 pmp->status |= IB_SMP_INVALID_FIELD;
959         /*
960          * Ticks are 10x the link transfer period which for 2.5Gbs is 4
961          * nsec.  0 == 4 nsec., 1 == 8 nsec., ..., 255 == 1020 nsec.  Sample
962          * intervals are counted in ticks.  Since we use Linux timers, that
963          * count in jiffies, we can't sample for less than 1000 ticks if HZ
964          * == 1000 (4000 ticks if HZ is 250).
965          */
966         /* XXX This is WRONG. */
967         p->tick = 250;          /* 1 usec. */
968         p->counter_width = 4;   /* 32 bit counters */
969         p->counter_mask0_9 = COUNTER_MASK0_9;
970         spin_lock_irqsave(&dev->pending_lock, flags);
971         p->sample_status = dev->pma_sample_status;
972         p->sample_start = cpu_to_be32(dev->pma_sample_start);
973         p->sample_interval = cpu_to_be32(dev->pma_sample_interval);
974         p->tag = cpu_to_be16(dev->pma_tag);
975         p->counter_select[0] = dev->pma_counter_select[0];
976         p->counter_select[1] = dev->pma_counter_select[1];
977         p->counter_select[2] = dev->pma_counter_select[2];
978         p->counter_select[3] = dev->pma_counter_select[3];
979         p->counter_select[4] = dev->pma_counter_select[4];
980         spin_unlock_irqrestore(&dev->pending_lock, flags);
981
982         return reply((struct ib_smp *) pmp);
983 }
984
985 static int recv_pma_set_portsamplescontrol(struct ib_perf *pmp,
986                                            struct ib_device *ibdev, u8 port)
987 {
988         struct ib_pma_portsamplescontrol *p =
989                 (struct ib_pma_portsamplescontrol *)pmp->data;
990         struct ipath_ibdev *dev = to_idev(ibdev);
991         unsigned long flags;
992         u32 start;
993         int ret;
994
995         if (pmp->attr_mod != 0 ||
996             (p->port_select != port && p->port_select != 0xFF)) {
997                 pmp->status |= IB_SMP_INVALID_FIELD;
998                 ret = reply((struct ib_smp *) pmp);
999                 goto bail;
1000         }
1001
1002         start = be32_to_cpu(p->sample_start);
1003         if (start != 0) {
1004                 spin_lock_irqsave(&dev->pending_lock, flags);
1005                 if (dev->pma_sample_status == IB_PMA_SAMPLE_STATUS_DONE) {
1006                         dev->pma_sample_status =
1007                                 IB_PMA_SAMPLE_STATUS_STARTED;
1008                         dev->pma_sample_start = start;
1009                         dev->pma_sample_interval =
1010                                 be32_to_cpu(p->sample_interval);
1011                         dev->pma_tag = be16_to_cpu(p->tag);
1012                         if (p->counter_select[0])
1013                                 dev->pma_counter_select[0] =
1014                                         p->counter_select[0];
1015                         if (p->counter_select[1])
1016                                 dev->pma_counter_select[1] =
1017                                         p->counter_select[1];
1018                         if (p->counter_select[2])
1019                                 dev->pma_counter_select[2] =
1020                                         p->counter_select[2];
1021                         if (p->counter_select[3])
1022                                 dev->pma_counter_select[3] =
1023                                         p->counter_select[3];
1024                         if (p->counter_select[4])
1025                                 dev->pma_counter_select[4] =
1026                                         p->counter_select[4];
1027                 }
1028                 spin_unlock_irqrestore(&dev->pending_lock, flags);
1029         }
1030         ret = recv_pma_get_portsamplescontrol(pmp, ibdev, port);
1031
1032 bail:
1033         return ret;
1034 }
1035
1036 static u64 get_counter(struct ipath_ibdev *dev, __be16 sel)
1037 {
1038         u64 ret;
1039
1040         switch (sel) {
1041         case IB_PMA_PORT_XMIT_DATA:
1042                 ret = dev->ipath_sword;
1043                 break;
1044         case IB_PMA_PORT_RCV_DATA:
1045                 ret = dev->ipath_rword;
1046                 break;
1047         case IB_PMA_PORT_XMIT_PKTS:
1048                 ret = dev->ipath_spkts;
1049                 break;
1050         case IB_PMA_PORT_RCV_PKTS:
1051                 ret = dev->ipath_rpkts;
1052                 break;
1053         case IB_PMA_PORT_XMIT_WAIT:
1054                 ret = dev->ipath_xmit_wait;
1055                 break;
1056         default:
1057                 ret = 0;
1058         }
1059
1060         return ret;
1061 }
1062
1063 static int recv_pma_get_portsamplesresult(struct ib_perf *pmp,
1064                                           struct ib_device *ibdev)
1065 {
1066         struct ib_pma_portsamplesresult *p =
1067                 (struct ib_pma_portsamplesresult *)pmp->data;
1068         struct ipath_ibdev *dev = to_idev(ibdev);
1069         int i;
1070
1071         memset(pmp->data, 0, sizeof(pmp->data));
1072         p->tag = cpu_to_be16(dev->pma_tag);
1073         p->sample_status = cpu_to_be16(dev->pma_sample_status);
1074         for (i = 0; i < ARRAY_SIZE(dev->pma_counter_select); i++)
1075                 p->counter[i] = cpu_to_be32(
1076                         get_counter(dev, dev->pma_counter_select[i]));
1077
1078         return reply((struct ib_smp *) pmp);
1079 }
1080
1081 static int recv_pma_get_portsamplesresult_ext(struct ib_perf *pmp,
1082                                               struct ib_device *ibdev)
1083 {
1084         struct ib_pma_portsamplesresult_ext *p =
1085                 (struct ib_pma_portsamplesresult_ext *)pmp->data;
1086         struct ipath_ibdev *dev = to_idev(ibdev);
1087         int i;
1088
1089         memset(pmp->data, 0, sizeof(pmp->data));
1090         p->tag = cpu_to_be16(dev->pma_tag);
1091         p->sample_status = cpu_to_be16(dev->pma_sample_status);
1092         /* 64 bits */
1093         p->extended_width = __constant_cpu_to_be32(0x80000000);
1094         for (i = 0; i < ARRAY_SIZE(dev->pma_counter_select); i++)
1095                 p->counter[i] = cpu_to_be64(
1096                         get_counter(dev, dev->pma_counter_select[i]));
1097
1098         return reply((struct ib_smp *) pmp);
1099 }
1100
1101 static int recv_pma_get_portcounters(struct ib_perf *pmp,
1102                                      struct ib_device *ibdev, u8 port)
1103 {
1104         struct ib_pma_portcounters *p = (struct ib_pma_portcounters *)
1105                 pmp->data;
1106         struct ipath_ibdev *dev = to_idev(ibdev);
1107         struct ipath_verbs_counters cntrs;
1108         u8 port_select = p->port_select;
1109
1110         ipath_get_counters(dev->dd, &cntrs);
1111
1112         /* Adjust counters for any resets done. */
1113         cntrs.symbol_error_counter -= dev->z_symbol_error_counter;
1114         cntrs.link_error_recovery_counter -=
1115                 dev->z_link_error_recovery_counter;
1116         cntrs.link_downed_counter -= dev->z_link_downed_counter;
1117         cntrs.port_rcv_errors += dev->rcv_errors;
1118         cntrs.port_rcv_errors -= dev->z_port_rcv_errors;
1119         cntrs.port_rcv_remphys_errors -= dev->z_port_rcv_remphys_errors;
1120         cntrs.port_xmit_discards -= dev->z_port_xmit_discards;
1121         cntrs.port_xmit_data -= dev->z_port_xmit_data;
1122         cntrs.port_rcv_data -= dev->z_port_rcv_data;
1123         cntrs.port_xmit_packets -= dev->z_port_xmit_packets;
1124         cntrs.port_rcv_packets -= dev->z_port_rcv_packets;
1125         cntrs.local_link_integrity_errors -=
1126                 dev->z_local_link_integrity_errors;
1127         cntrs.excessive_buffer_overrun_errors -=
1128                 dev->z_excessive_buffer_overrun_errors;
1129
1130         memset(pmp->data, 0, sizeof(pmp->data));
1131
1132         p->port_select = port_select;
1133         if (pmp->attr_mod != 0 ||
1134             (port_select != port && port_select != 0xFF))
1135                 pmp->status |= IB_SMP_INVALID_FIELD;
1136
1137         if (cntrs.symbol_error_counter > 0xFFFFUL)
1138                 p->symbol_error_counter = __constant_cpu_to_be16(0xFFFF);
1139         else
1140                 p->symbol_error_counter =
1141                         cpu_to_be16((u16)cntrs.symbol_error_counter);
1142         if (cntrs.link_error_recovery_counter > 0xFFUL)
1143                 p->link_error_recovery_counter = 0xFF;
1144         else
1145                 p->link_error_recovery_counter =
1146                         (u8)cntrs.link_error_recovery_counter;
1147         if (cntrs.link_downed_counter > 0xFFUL)
1148                 p->link_downed_counter = 0xFF;
1149         else
1150                 p->link_downed_counter = (u8)cntrs.link_downed_counter;
1151         if (cntrs.port_rcv_errors > 0xFFFFUL)
1152                 p->port_rcv_errors = __constant_cpu_to_be16(0xFFFF);
1153         else
1154                 p->port_rcv_errors =
1155                         cpu_to_be16((u16) cntrs.port_rcv_errors);
1156         if (cntrs.port_rcv_remphys_errors > 0xFFFFUL)
1157                 p->port_rcv_remphys_errors = __constant_cpu_to_be16(0xFFFF);
1158         else
1159                 p->port_rcv_remphys_errors =
1160                         cpu_to_be16((u16)cntrs.port_rcv_remphys_errors);
1161         if (cntrs.port_xmit_discards > 0xFFFFUL)
1162                 p->port_xmit_discards = __constant_cpu_to_be16(0xFFFF);
1163         else
1164                 p->port_xmit_discards =
1165                         cpu_to_be16((u16)cntrs.port_xmit_discards);
1166         if (cntrs.local_link_integrity_errors > 0xFUL)
1167                 cntrs.local_link_integrity_errors = 0xFUL;
1168         if (cntrs.excessive_buffer_overrun_errors > 0xFUL)
1169                 cntrs.excessive_buffer_overrun_errors = 0xFUL;
1170         p->lli_ebor_errors = (cntrs.local_link_integrity_errors << 4) |
1171                 cntrs.excessive_buffer_overrun_errors;
1172         if (dev->n_vl15_dropped > 0xFFFFUL)
1173                 p->vl15_dropped = __constant_cpu_to_be16(0xFFFF);
1174         else
1175                 p->vl15_dropped = cpu_to_be16((u16)dev->n_vl15_dropped);
1176         if (cntrs.port_xmit_data > 0xFFFFFFFFUL)
1177                 p->port_xmit_data = __constant_cpu_to_be32(0xFFFFFFFF);
1178         else
1179                 p->port_xmit_data = cpu_to_be32((u32)cntrs.port_xmit_data);
1180         if (cntrs.port_rcv_data > 0xFFFFFFFFUL)
1181                 p->port_rcv_data = __constant_cpu_to_be32(0xFFFFFFFF);
1182         else
1183                 p->port_rcv_data = cpu_to_be32((u32)cntrs.port_rcv_data);
1184         if (cntrs.port_xmit_packets > 0xFFFFFFFFUL)
1185                 p->port_xmit_packets = __constant_cpu_to_be32(0xFFFFFFFF);
1186         else
1187                 p->port_xmit_packets =
1188                         cpu_to_be32((u32)cntrs.port_xmit_packets);
1189         if (cntrs.port_rcv_packets > 0xFFFFFFFFUL)
1190                 p->port_rcv_packets = __constant_cpu_to_be32(0xFFFFFFFF);
1191         else
1192                 p->port_rcv_packets =
1193                         cpu_to_be32((u32) cntrs.port_rcv_packets);
1194
1195         return reply((struct ib_smp *) pmp);
1196 }
1197
1198 static int recv_pma_get_portcounters_ext(struct ib_perf *pmp,
1199                                          struct ib_device *ibdev, u8 port)
1200 {
1201         struct ib_pma_portcounters_ext *p =
1202                 (struct ib_pma_portcounters_ext *)pmp->data;
1203         struct ipath_ibdev *dev = to_idev(ibdev);
1204         u64 swords, rwords, spkts, rpkts, xwait;
1205         u8 port_select = p->port_select;
1206
1207         ipath_snapshot_counters(dev->dd, &swords, &rwords, &spkts,
1208                                 &rpkts, &xwait);
1209
1210         /* Adjust counters for any resets done. */
1211         swords -= dev->z_port_xmit_data;
1212         rwords -= dev->z_port_rcv_data;
1213         spkts -= dev->z_port_xmit_packets;
1214         rpkts -= dev->z_port_rcv_packets;
1215
1216         memset(pmp->data, 0, sizeof(pmp->data));
1217
1218         p->port_select = port_select;
1219         if (pmp->attr_mod != 0 ||
1220             (port_select != port && port_select != 0xFF))
1221                 pmp->status |= IB_SMP_INVALID_FIELD;
1222
1223         p->port_xmit_data = cpu_to_be64(swords);
1224         p->port_rcv_data = cpu_to_be64(rwords);
1225         p->port_xmit_packets = cpu_to_be64(spkts);
1226         p->port_rcv_packets = cpu_to_be64(rpkts);
1227         p->port_unicast_xmit_packets = cpu_to_be64(dev->n_unicast_xmit);
1228         p->port_unicast_rcv_packets = cpu_to_be64(dev->n_unicast_rcv);
1229         p->port_multicast_xmit_packets = cpu_to_be64(dev->n_multicast_xmit);
1230         p->port_multicast_rcv_packets = cpu_to_be64(dev->n_multicast_rcv);
1231
1232         return reply((struct ib_smp *) pmp);
1233 }
1234
1235 static int recv_pma_set_portcounters(struct ib_perf *pmp,
1236                                      struct ib_device *ibdev, u8 port)
1237 {
1238         struct ib_pma_portcounters *p = (struct ib_pma_portcounters *)
1239                 pmp->data;
1240         struct ipath_ibdev *dev = to_idev(ibdev);
1241         struct ipath_verbs_counters cntrs;
1242
1243         /*
1244          * Since the HW doesn't support clearing counters, we save the
1245          * current count and subtract it from future responses.
1246          */
1247         ipath_get_counters(dev->dd, &cntrs);
1248
1249         if (p->counter_select & IB_PMA_SEL_SYMBOL_ERROR)
1250                 dev->z_symbol_error_counter = cntrs.symbol_error_counter;
1251
1252         if (p->counter_select & IB_PMA_SEL_LINK_ERROR_RECOVERY)
1253                 dev->z_link_error_recovery_counter =
1254                         cntrs.link_error_recovery_counter;
1255
1256         if (p->counter_select & IB_PMA_SEL_LINK_DOWNED)
1257                 dev->z_link_downed_counter = cntrs.link_downed_counter;
1258
1259         if (p->counter_select & IB_PMA_SEL_PORT_RCV_ERRORS)
1260                 dev->z_port_rcv_errors =
1261                         cntrs.port_rcv_errors + dev->rcv_errors;
1262
1263         if (p->counter_select & IB_PMA_SEL_PORT_RCV_REMPHYS_ERRORS)
1264                 dev->z_port_rcv_remphys_errors =
1265                         cntrs.port_rcv_remphys_errors;
1266
1267         if (p->counter_select & IB_PMA_SEL_PORT_XMIT_DISCARDS)
1268                 dev->z_port_xmit_discards = cntrs.port_xmit_discards;
1269
1270         if (p->counter_select & IB_PMA_SEL_LOCAL_LINK_INTEGRITY_ERRORS)
1271                 dev->z_local_link_integrity_errors =
1272                         cntrs.local_link_integrity_errors;
1273
1274         if (p->counter_select & IB_PMA_SEL_EXCESSIVE_BUFFER_OVERRUNS)
1275                 dev->z_excessive_buffer_overrun_errors =
1276                         cntrs.excessive_buffer_overrun_errors;
1277
1278         if (p->counter_select & IB_PMA_SEL_PORT_VL15_DROPPED)
1279                 dev->n_vl15_dropped = 0;
1280
1281         if (p->counter_select & IB_PMA_SEL_PORT_XMIT_DATA)
1282                 dev->z_port_xmit_data = cntrs.port_xmit_data;
1283
1284         if (p->counter_select & IB_PMA_SEL_PORT_RCV_DATA)
1285                 dev->z_port_rcv_data = cntrs.port_rcv_data;
1286
1287         if (p->counter_select & IB_PMA_SEL_PORT_XMIT_PACKETS)
1288                 dev->z_port_xmit_packets = cntrs.port_xmit_packets;
1289
1290         if (p->counter_select & IB_PMA_SEL_PORT_RCV_PACKETS)
1291                 dev->z_port_rcv_packets = cntrs.port_rcv_packets;
1292
1293         return recv_pma_get_portcounters(pmp, ibdev, port);
1294 }
1295
1296 static int recv_pma_set_portcounters_ext(struct ib_perf *pmp,
1297                                          struct ib_device *ibdev, u8 port)
1298 {
1299         struct ib_pma_portcounters *p = (struct ib_pma_portcounters *)
1300                 pmp->data;
1301         struct ipath_ibdev *dev = to_idev(ibdev);
1302         u64 swords, rwords, spkts, rpkts, xwait;
1303
1304         ipath_snapshot_counters(dev->dd, &swords, &rwords, &spkts,
1305                                 &rpkts, &xwait);
1306
1307         if (p->counter_select & IB_PMA_SELX_PORT_XMIT_DATA)
1308                 dev->z_port_xmit_data = swords;
1309
1310         if (p->counter_select & IB_PMA_SELX_PORT_RCV_DATA)
1311                 dev->z_port_rcv_data = rwords;
1312
1313         if (p->counter_select & IB_PMA_SELX_PORT_XMIT_PACKETS)
1314                 dev->z_port_xmit_packets = spkts;
1315
1316         if (p->counter_select & IB_PMA_SELX_PORT_RCV_PACKETS)
1317                 dev->z_port_rcv_packets = rpkts;
1318
1319         if (p->counter_select & IB_PMA_SELX_PORT_UNI_XMIT_PACKETS)
1320                 dev->n_unicast_xmit = 0;
1321
1322         if (p->counter_select & IB_PMA_SELX_PORT_UNI_RCV_PACKETS)
1323                 dev->n_unicast_rcv = 0;
1324
1325         if (p->counter_select & IB_PMA_SELX_PORT_MULTI_XMIT_PACKETS)
1326                 dev->n_multicast_xmit = 0;
1327
1328         if (p->counter_select & IB_PMA_SELX_PORT_MULTI_RCV_PACKETS)
1329                 dev->n_multicast_rcv = 0;
1330
1331         return recv_pma_get_portcounters_ext(pmp, ibdev, port);
1332 }
1333
1334 static int process_subn(struct ib_device *ibdev, int mad_flags,
1335                         u8 port_num, struct ib_mad *in_mad,
1336                         struct ib_mad *out_mad)
1337 {
1338         struct ib_smp *smp = (struct ib_smp *)out_mad;
1339         struct ipath_ibdev *dev = to_idev(ibdev);
1340         int ret;
1341
1342         *out_mad = *in_mad;
1343         if (smp->class_version != 1) {
1344                 smp->status |= IB_SMP_UNSUP_VERSION;
1345                 ret = reply(smp);
1346                 goto bail;
1347         }
1348
1349         /* Is the mkey in the process of expiring? */
1350         if (dev->mkey_lease_timeout && jiffies >= dev->mkey_lease_timeout) {
1351                 /* Clear timeout and mkey protection field. */
1352                 dev->mkey_lease_timeout = 0;
1353                 dev->mkeyprot_resv_lmc &= 0x3F;
1354         }
1355
1356         /*
1357          * M_Key checking depends on
1358          * Portinfo:M_Key_protect_bits
1359          */
1360         if ((mad_flags & IB_MAD_IGNORE_MKEY) == 0 && dev->mkey != 0 &&
1361             dev->mkey != smp->mkey &&
1362             (smp->method == IB_MGMT_METHOD_SET ||
1363              (smp->method == IB_MGMT_METHOD_GET &&
1364               (dev->mkeyprot_resv_lmc >> 7) != 0))) {
1365                 if (dev->mkey_violations != 0xFFFF)
1366                         ++dev->mkey_violations;
1367                 if (dev->mkey_lease_timeout ||
1368                     dev->mkey_lease_period == 0) {
1369                         ret = IB_MAD_RESULT_SUCCESS |
1370                                 IB_MAD_RESULT_CONSUMED;
1371                         goto bail;
1372                 }
1373                 dev->mkey_lease_timeout = jiffies +
1374                         dev->mkey_lease_period * HZ;
1375                 /* Future: Generate a trap notice. */
1376                 ret = IB_MAD_RESULT_SUCCESS | IB_MAD_RESULT_CONSUMED;
1377                 goto bail;
1378         } else if (dev->mkey_lease_timeout)
1379                 dev->mkey_lease_timeout = 0;
1380
1381         switch (smp->method) {
1382         case IB_MGMT_METHOD_GET:
1383                 switch (smp->attr_id) {
1384                 case IB_SMP_ATTR_NODE_DESC:
1385                         ret = recv_subn_get_nodedescription(smp, ibdev);
1386                         goto bail;
1387                 case IB_SMP_ATTR_NODE_INFO:
1388                         ret = recv_subn_get_nodeinfo(smp, ibdev, port_num);
1389                         goto bail;
1390                 case IB_SMP_ATTR_GUID_INFO:
1391                         ret = recv_subn_get_guidinfo(smp, ibdev);
1392                         goto bail;
1393                 case IB_SMP_ATTR_PORT_INFO:
1394                         ret = recv_subn_get_portinfo(smp, ibdev, port_num);
1395                         goto bail;
1396                 case IB_SMP_ATTR_PKEY_TABLE:
1397                         ret = recv_subn_get_pkeytable(smp, ibdev);
1398                         goto bail;
1399                 case IB_SMP_ATTR_SM_INFO:
1400                         if (dev->port_cap_flags & IB_PORT_SM_DISABLED) {
1401                                 ret = IB_MAD_RESULT_SUCCESS |
1402                                         IB_MAD_RESULT_CONSUMED;
1403                                 goto bail;
1404                         }
1405                         if (dev->port_cap_flags & IB_PORT_SM) {
1406                                 ret = IB_MAD_RESULT_SUCCESS;
1407                                 goto bail;
1408                         }
1409                         /* FALLTHROUGH */
1410                 default:
1411                         smp->status |= IB_SMP_UNSUP_METH_ATTR;
1412                         ret = reply(smp);
1413                         goto bail;
1414                 }
1415
1416         case IB_MGMT_METHOD_SET:
1417                 switch (smp->attr_id) {
1418                 case IB_SMP_ATTR_GUID_INFO:
1419                         ret = recv_subn_set_guidinfo(smp, ibdev);
1420                         goto bail;
1421                 case IB_SMP_ATTR_PORT_INFO:
1422                         ret = recv_subn_set_portinfo(smp, ibdev, port_num);
1423                         goto bail;
1424                 case IB_SMP_ATTR_PKEY_TABLE:
1425                         ret = recv_subn_set_pkeytable(smp, ibdev);
1426                         goto bail;
1427                 case IB_SMP_ATTR_SM_INFO:
1428                         if (dev->port_cap_flags & IB_PORT_SM_DISABLED) {
1429                                 ret = IB_MAD_RESULT_SUCCESS |
1430                                         IB_MAD_RESULT_CONSUMED;
1431                                 goto bail;
1432                         }
1433                         if (dev->port_cap_flags & IB_PORT_SM) {
1434                                 ret = IB_MAD_RESULT_SUCCESS;
1435                                 goto bail;
1436                         }
1437                         /* FALLTHROUGH */
1438                 default:
1439                         smp->status |= IB_SMP_UNSUP_METH_ATTR;
1440                         ret = reply(smp);
1441                         goto bail;
1442                 }
1443
1444         case IB_MGMT_METHOD_GET_RESP:
1445                 /*
1446                  * The ib_mad module will call us to process responses
1447                  * before checking for other consumers.
1448                  * Just tell the caller to process it normally.
1449                  */
1450                 ret = IB_MAD_RESULT_FAILURE;
1451                 goto bail;
1452         default:
1453                 smp->status |= IB_SMP_UNSUP_METHOD;
1454                 ret = reply(smp);
1455         }
1456
1457 bail:
1458         return ret;
1459 }
1460
1461 static int process_perf(struct ib_device *ibdev, u8 port_num,
1462                         struct ib_mad *in_mad,
1463                         struct ib_mad *out_mad)
1464 {
1465         struct ib_perf *pmp = (struct ib_perf *)out_mad;
1466         int ret;
1467
1468         *out_mad = *in_mad;
1469         if (pmp->class_version != 1) {
1470                 pmp->status |= IB_SMP_UNSUP_VERSION;
1471                 ret = reply((struct ib_smp *) pmp);
1472                 goto bail;
1473         }
1474
1475         switch (pmp->method) {
1476         case IB_MGMT_METHOD_GET:
1477                 switch (pmp->attr_id) {
1478                 case IB_PMA_CLASS_PORT_INFO:
1479                         ret = recv_pma_get_classportinfo(pmp);
1480                         goto bail;
1481                 case IB_PMA_PORT_SAMPLES_CONTROL:
1482                         ret = recv_pma_get_portsamplescontrol(pmp, ibdev,
1483                                                               port_num);
1484                         goto bail;
1485                 case IB_PMA_PORT_SAMPLES_RESULT:
1486                         ret = recv_pma_get_portsamplesresult(pmp, ibdev);
1487                         goto bail;
1488                 case IB_PMA_PORT_SAMPLES_RESULT_EXT:
1489                         ret = recv_pma_get_portsamplesresult_ext(pmp,
1490                                                                  ibdev);
1491                         goto bail;
1492                 case IB_PMA_PORT_COUNTERS:
1493                         ret = recv_pma_get_portcounters(pmp, ibdev,
1494                                                         port_num);
1495                         goto bail;
1496                 case IB_PMA_PORT_COUNTERS_EXT:
1497                         ret = recv_pma_get_portcounters_ext(pmp, ibdev,
1498                                                             port_num);
1499                         goto bail;
1500                 default:
1501                         pmp->status |= IB_SMP_UNSUP_METH_ATTR;
1502                         ret = reply((struct ib_smp *) pmp);
1503                         goto bail;
1504                 }
1505
1506         case IB_MGMT_METHOD_SET:
1507                 switch (pmp->attr_id) {
1508                 case IB_PMA_PORT_SAMPLES_CONTROL:
1509                         ret = recv_pma_set_portsamplescontrol(pmp, ibdev,
1510                                                               port_num);
1511                         goto bail;
1512                 case IB_PMA_PORT_COUNTERS:
1513                         ret = recv_pma_set_portcounters(pmp, ibdev,
1514                                                         port_num);
1515                         goto bail;
1516                 case IB_PMA_PORT_COUNTERS_EXT:
1517                         ret = recv_pma_set_portcounters_ext(pmp, ibdev,
1518                                                             port_num);
1519                         goto bail;
1520                 default:
1521                         pmp->status |= IB_SMP_UNSUP_METH_ATTR;
1522                         ret = reply((struct ib_smp *) pmp);
1523                         goto bail;
1524                 }
1525
1526         case IB_MGMT_METHOD_GET_RESP:
1527                 /*
1528                  * The ib_mad module will call us to process responses
1529                  * before checking for other consumers.
1530                  * Just tell the caller to process it normally.
1531                  */
1532                 ret = IB_MAD_RESULT_FAILURE;
1533                 goto bail;
1534         default:
1535                 pmp->status |= IB_SMP_UNSUP_METHOD;
1536                 ret = reply((struct ib_smp *) pmp);
1537         }
1538
1539 bail:
1540         return ret;
1541 }
1542
1543 /**
1544  * ipath_process_mad - process an incoming MAD packet
1545  * @ibdev: the infiniband device this packet came in on
1546  * @mad_flags: MAD flags
1547  * @port_num: the port number this packet came in on
1548  * @in_wc: the work completion entry for this packet
1549  * @in_grh: the global route header for this packet
1550  * @in_mad: the incoming MAD
1551  * @out_mad: any outgoing MAD reply
1552  *
1553  * Returns IB_MAD_RESULT_SUCCESS if this is a MAD that we are not
1554  * interested in processing.
1555  *
1556  * Note that the verbs framework has already done the MAD sanity checks,
1557  * and hop count/pointer updating for IB_MGMT_CLASS_SUBN_DIRECTED_ROUTE
1558  * MADs.
1559  *
1560  * This is called by the ib_mad module.
1561  */
1562 int ipath_process_mad(struct ib_device *ibdev, int mad_flags, u8 port_num,
1563                       struct ib_wc *in_wc, struct ib_grh *in_grh,
1564                       struct ib_mad *in_mad, struct ib_mad *out_mad)
1565 {
1566         int ret;
1567
1568         switch (in_mad->mad_hdr.mgmt_class) {
1569         case IB_MGMT_CLASS_SUBN_DIRECTED_ROUTE:
1570         case IB_MGMT_CLASS_SUBN_LID_ROUTED:
1571                 ret = process_subn(ibdev, mad_flags, port_num,
1572                                    in_mad, out_mad);
1573                 goto bail;
1574         case IB_MGMT_CLASS_PERF_MGMT:
1575                 ret = process_perf(ibdev, port_num, in_mad, out_mad);
1576                 goto bail;
1577         default:
1578                 ret = IB_MAD_RESULT_SUCCESS;
1579         }
1580
1581 bail:
1582         return ret;
1583 }