IB/ipath: Really run work in ipath_release_user_pages_on_close()
[linux-3.10.git] / drivers / infiniband / hw / ipath / ipath_mad.c
1 /*
2  * Copyright (c) 2006, 2007, 2008 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] = IPATH_SRC_OUI_1;
115         nip->vendor_id[1] = IPATH_SRC_OUI_2;
116         nip->vendor_id[2] = IPATH_SRC_OUI_3;
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 static void set_link_width_enabled(struct ipath_devdata *dd, u32 w)
150 {
151         (void) dd->ipath_f_set_ib_cfg(dd, IPATH_IB_CFG_LWID_ENB, w);
152 }
153
154 static void set_link_speed_enabled(struct ipath_devdata *dd, u32 s)
155 {
156         (void) dd->ipath_f_set_ib_cfg(dd, IPATH_IB_CFG_SPD_ENB, s);
157 }
158
159 static int get_overrunthreshold(struct ipath_devdata *dd)
160 {
161         return (dd->ipath_ibcctrl >>
162                 INFINIPATH_IBCC_OVERRUNTHRESHOLD_SHIFT) &
163                 INFINIPATH_IBCC_OVERRUNTHRESHOLD_MASK;
164 }
165
166 /**
167  * set_overrunthreshold - set the overrun threshold
168  * @dd: the infinipath device
169  * @n: the new threshold
170  *
171  * Note that this will only take effect when the link state changes.
172  */
173 static int set_overrunthreshold(struct ipath_devdata *dd, unsigned n)
174 {
175         unsigned v;
176
177         v = (dd->ipath_ibcctrl >> INFINIPATH_IBCC_OVERRUNTHRESHOLD_SHIFT) &
178                 INFINIPATH_IBCC_OVERRUNTHRESHOLD_MASK;
179         if (v != n) {
180                 dd->ipath_ibcctrl &=
181                         ~(INFINIPATH_IBCC_OVERRUNTHRESHOLD_MASK <<
182                           INFINIPATH_IBCC_OVERRUNTHRESHOLD_SHIFT);
183                 dd->ipath_ibcctrl |=
184                         (u64) n << INFINIPATH_IBCC_OVERRUNTHRESHOLD_SHIFT;
185                 ipath_write_kreg(dd, dd->ipath_kregs->kr_ibcctrl,
186                                  dd->ipath_ibcctrl);
187         }
188         return 0;
189 }
190
191 static int get_phyerrthreshold(struct ipath_devdata *dd)
192 {
193         return (dd->ipath_ibcctrl >>
194                 INFINIPATH_IBCC_PHYERRTHRESHOLD_SHIFT) &
195                 INFINIPATH_IBCC_PHYERRTHRESHOLD_MASK;
196 }
197
198 /**
199  * set_phyerrthreshold - set the physical error threshold
200  * @dd: the infinipath device
201  * @n: the new threshold
202  *
203  * Note that this will only take effect when the link state changes.
204  */
205 static int set_phyerrthreshold(struct ipath_devdata *dd, unsigned n)
206 {
207         unsigned v;
208
209         v = (dd->ipath_ibcctrl >> INFINIPATH_IBCC_PHYERRTHRESHOLD_SHIFT) &
210                 INFINIPATH_IBCC_PHYERRTHRESHOLD_MASK;
211         if (v != n) {
212                 dd->ipath_ibcctrl &=
213                         ~(INFINIPATH_IBCC_PHYERRTHRESHOLD_MASK <<
214                           INFINIPATH_IBCC_PHYERRTHRESHOLD_SHIFT);
215                 dd->ipath_ibcctrl |=
216                         (u64) n << INFINIPATH_IBCC_PHYERRTHRESHOLD_SHIFT;
217                 ipath_write_kreg(dd, dd->ipath_kregs->kr_ibcctrl,
218                                  dd->ipath_ibcctrl);
219         }
220         return 0;
221 }
222
223 /**
224  * get_linkdowndefaultstate - get the default linkdown state
225  * @dd: the infinipath device
226  *
227  * Returns zero if the default is POLL, 1 if the default is SLEEP.
228  */
229 static int get_linkdowndefaultstate(struct ipath_devdata *dd)
230 {
231         return !!(dd->ipath_ibcctrl & INFINIPATH_IBCC_LINKDOWNDEFAULTSTATE);
232 }
233
234 static int recv_subn_get_portinfo(struct ib_smp *smp,
235                                   struct ib_device *ibdev, u8 port)
236 {
237         struct ipath_ibdev *dev;
238         struct ipath_devdata *dd;
239         struct ib_port_info *pip = (struct ib_port_info *)smp->data;
240         u16 lid;
241         u8 ibcstat;
242         u8 mtu;
243         int ret;
244
245         if (be32_to_cpu(smp->attr_mod) > ibdev->phys_port_cnt) {
246                 smp->status |= IB_SMP_INVALID_FIELD;
247                 ret = reply(smp);
248                 goto bail;
249         }
250
251         dev = to_idev(ibdev);
252         dd = dev->dd;
253
254         /* Clear all fields.  Only set the non-zero fields. */
255         memset(smp->data, 0, sizeof(smp->data));
256
257         /* Only return the mkey if the protection field allows it. */
258         if (smp->method == IB_MGMT_METHOD_SET || dev->mkey == smp->mkey ||
259             dev->mkeyprot == 0)
260                 pip->mkey = dev->mkey;
261         pip->gid_prefix = dev->gid_prefix;
262         lid = dd->ipath_lid;
263         pip->lid = lid ? cpu_to_be16(lid) : IB_LID_PERMISSIVE;
264         pip->sm_lid = cpu_to_be16(dev->sm_lid);
265         pip->cap_mask = cpu_to_be32(dev->port_cap_flags);
266         /* pip->diag_code; */
267         pip->mkey_lease_period = cpu_to_be16(dev->mkey_lease_period);
268         pip->local_port_num = port;
269         pip->link_width_enabled = dd->ipath_link_width_enabled;
270         pip->link_width_supported = dd->ipath_link_width_supported;
271         pip->link_width_active = dd->ipath_link_width_active;
272         pip->linkspeed_portstate = dd->ipath_link_speed_supported << 4;
273         ibcstat = dd->ipath_lastibcstat;
274         /* map LinkState to IB portinfo values.  */
275         pip->linkspeed_portstate |= ipath_ib_linkstate(dd, ibcstat) + 1;
276
277         pip->portphysstate_linkdown =
278                 (ipath_cvt_physportstate[ibcstat & dd->ibcs_lts_mask] << 4) |
279                 (get_linkdowndefaultstate(dd) ? 1 : 2);
280         pip->mkeyprot_resv_lmc = (dev->mkeyprot << 6) | dd->ipath_lmc;
281         pip->linkspeedactive_enabled = (dd->ipath_link_speed_active << 4) |
282                 dd->ipath_link_speed_enabled;
283         switch (dd->ipath_ibmtu) {
284         case 4096:
285                 mtu = IB_MTU_4096;
286                 break;
287         case 2048:
288                 mtu = IB_MTU_2048;
289                 break;
290         case 1024:
291                 mtu = IB_MTU_1024;
292                 break;
293         case 512:
294                 mtu = IB_MTU_512;
295                 break;
296         case 256:
297                 mtu = IB_MTU_256;
298                 break;
299         default:                /* oops, something is wrong */
300                 mtu = IB_MTU_2048;
301                 break;
302         }
303         pip->neighbormtu_mastersmsl = (mtu << 4) | dev->sm_sl;
304         pip->vlcap_inittype = 0x10;     /* VLCap = VL0, InitType = 0 */
305         pip->vl_high_limit = dev->vl_high_limit;
306         /* pip->vl_arb_high_cap; // only one VL */
307         /* pip->vl_arb_low_cap; // only one VL */
308         /* InitTypeReply = 0 */
309         /* our mtu cap depends on whether 4K MTU enabled or not */
310         pip->inittypereply_mtucap = ipath_mtu4096 ? IB_MTU_4096 : IB_MTU_2048;
311         /* HCAs ignore VLStallCount and HOQLife */
312         /* pip->vlstallcnt_hoqlife; */
313         pip->operationalvl_pei_peo_fpi_fpo = 0x10;      /* OVLs = 1 */
314         pip->mkey_violations = cpu_to_be16(dev->mkey_violations);
315         /* P_KeyViolations are counted by hardware. */
316         pip->pkey_violations =
317                 cpu_to_be16((ipath_get_cr_errpkey(dd) -
318                              dev->z_pkey_violations) & 0xFFFF);
319         pip->qkey_violations = cpu_to_be16(dev->qkey_violations);
320         /* Only the hardware GUID is supported for now */
321         pip->guid_cap = 1;
322         pip->clientrereg_resv_subnetto = dev->subnet_timeout;
323         /* 32.768 usec. response time (guessing) */
324         pip->resv_resptimevalue = 3;
325         pip->localphyerrors_overrunerrors =
326                 (get_phyerrthreshold(dd) << 4) |
327                 get_overrunthreshold(dd);
328         /* pip->max_credit_hint; */
329         if (dev->port_cap_flags & IB_PORT_LINK_LATENCY_SUP) {
330                 u32 v;
331
332                 v = dd->ipath_f_get_ib_cfg(dd, IPATH_IB_CFG_LINKLATENCY);
333                 pip->link_roundtrip_latency[0] = v >> 16;
334                 pip->link_roundtrip_latency[1] = v >> 8;
335                 pip->link_roundtrip_latency[2] = v;
336         }
337
338         ret = reply(smp);
339
340 bail:
341         return ret;
342 }
343
344 /**
345  * get_pkeys - return the PKEY table for port 0
346  * @dd: the infinipath device
347  * @pkeys: the pkey table is placed here
348  */
349 static int get_pkeys(struct ipath_devdata *dd, u16 * pkeys)
350 {
351         /* always a kernel port, no locking needed */
352         struct ipath_portdata *pd = dd->ipath_pd[0];
353
354         memcpy(pkeys, pd->port_pkeys, sizeof(pd->port_pkeys));
355
356         return 0;
357 }
358
359 static int recv_subn_get_pkeytable(struct ib_smp *smp,
360                                    struct ib_device *ibdev)
361 {
362         u32 startpx = 32 * (be32_to_cpu(smp->attr_mod) & 0xffff);
363         u16 *p = (u16 *) smp->data;
364         __be16 *q = (__be16 *) smp->data;
365
366         /* 64 blocks of 32 16-bit P_Key entries */
367
368         memset(smp->data, 0, sizeof(smp->data));
369         if (startpx == 0) {
370                 struct ipath_ibdev *dev = to_idev(ibdev);
371                 unsigned i, n = ipath_get_npkeys(dev->dd);
372
373                 get_pkeys(dev->dd, p);
374
375                 for (i = 0; i < n; i++)
376                         q[i] = cpu_to_be16(p[i]);
377         } else
378                 smp->status |= IB_SMP_INVALID_FIELD;
379
380         return reply(smp);
381 }
382
383 static int recv_subn_set_guidinfo(struct ib_smp *smp,
384                                   struct ib_device *ibdev)
385 {
386         /* The only GUID we support is the first read-only entry. */
387         return recv_subn_get_guidinfo(smp, ibdev);
388 }
389
390 /**
391  * set_linkdowndefaultstate - set the default linkdown state
392  * @dd: the infinipath device
393  * @sleep: the new state
394  *
395  * Note that this will only take effect when the link state changes.
396  */
397 static int set_linkdowndefaultstate(struct ipath_devdata *dd, int sleep)
398 {
399         if (sleep)
400                 dd->ipath_ibcctrl |= INFINIPATH_IBCC_LINKDOWNDEFAULTSTATE;
401         else
402                 dd->ipath_ibcctrl &= ~INFINIPATH_IBCC_LINKDOWNDEFAULTSTATE;
403         ipath_write_kreg(dd, dd->ipath_kregs->kr_ibcctrl,
404                          dd->ipath_ibcctrl);
405         return 0;
406 }
407
408 /**
409  * recv_subn_set_portinfo - set port information
410  * @smp: the incoming SM packet
411  * @ibdev: the infiniband device
412  * @port: the port on the device
413  *
414  * Set Portinfo (see ch. 14.2.5.6).
415  */
416 static int recv_subn_set_portinfo(struct ib_smp *smp,
417                                   struct ib_device *ibdev, u8 port)
418 {
419         struct ib_port_info *pip = (struct ib_port_info *)smp->data;
420         struct ib_event event;
421         struct ipath_ibdev *dev;
422         struct ipath_devdata *dd;
423         char clientrereg = 0;
424         u16 lid, smlid;
425         u8 lwe;
426         u8 lse;
427         u8 state;
428         u16 lstate;
429         u32 mtu;
430         int ret, ore;
431
432         if (be32_to_cpu(smp->attr_mod) > ibdev->phys_port_cnt)
433                 goto err;
434
435         dev = to_idev(ibdev);
436         dd = dev->dd;
437         event.device = ibdev;
438         event.element.port_num = port;
439
440         dev->mkey = pip->mkey;
441         dev->gid_prefix = pip->gid_prefix;
442         dev->mkey_lease_period = be16_to_cpu(pip->mkey_lease_period);
443
444         lid = be16_to_cpu(pip->lid);
445         if (dd->ipath_lid != lid ||
446             dd->ipath_lmc != (pip->mkeyprot_resv_lmc & 7)) {
447                 /* Must be a valid unicast LID address. */
448                 if (lid == 0 || lid >= IPATH_MULTICAST_LID_BASE)
449                         goto err;
450                 ipath_set_lid(dd, lid, pip->mkeyprot_resv_lmc & 7);
451                 event.event = IB_EVENT_LID_CHANGE;
452                 ib_dispatch_event(&event);
453         }
454
455         smlid = be16_to_cpu(pip->sm_lid);
456         if (smlid != dev->sm_lid) {
457                 /* Must be a valid unicast LID address. */
458                 if (smlid == 0 || smlid >= IPATH_MULTICAST_LID_BASE)
459                         goto err;
460                 dev->sm_lid = smlid;
461                 event.event = IB_EVENT_SM_CHANGE;
462                 ib_dispatch_event(&event);
463         }
464
465         /* Allow 1x or 4x to be set (see 14.2.6.6). */
466         lwe = pip->link_width_enabled;
467         if (lwe) {
468                 if (lwe == 0xFF)
469                         lwe = dd->ipath_link_width_supported;
470                 else if (lwe >= 16 || (lwe & ~dd->ipath_link_width_supported))
471                         goto err;
472                 set_link_width_enabled(dd, lwe);
473         }
474
475         /* Allow 2.5 or 5.0 Gbs. */
476         lse = pip->linkspeedactive_enabled & 0xF;
477         if (lse) {
478                 if (lse == 15)
479                         lse = dd->ipath_link_speed_supported;
480                 else if (lse >= 8 || (lse & ~dd->ipath_link_speed_supported))
481                         goto err;
482                 set_link_speed_enabled(dd, lse);
483         }
484
485         /* Set link down default state. */
486         switch (pip->portphysstate_linkdown & 0xF) {
487         case 0: /* NOP */
488                 break;
489         case 1: /* SLEEP */
490                 if (set_linkdowndefaultstate(dd, 1))
491                         goto err;
492                 break;
493         case 2: /* POLL */
494                 if (set_linkdowndefaultstate(dd, 0))
495                         goto err;
496                 break;
497         default:
498                 goto err;
499         }
500
501         dev->mkeyprot = pip->mkeyprot_resv_lmc >> 6;
502         dev->vl_high_limit = pip->vl_high_limit;
503
504         switch ((pip->neighbormtu_mastersmsl >> 4) & 0xF) {
505         case IB_MTU_256:
506                 mtu = 256;
507                 break;
508         case IB_MTU_512:
509                 mtu = 512;
510                 break;
511         case IB_MTU_1024:
512                 mtu = 1024;
513                 break;
514         case IB_MTU_2048:
515                 mtu = 2048;
516                 break;
517         case IB_MTU_4096:
518                 if (!ipath_mtu4096)
519                         goto err;
520                 mtu = 4096;
521                 break;
522         default:
523                 /* XXX We have already partially updated our state! */
524                 goto err;
525         }
526         ipath_set_mtu(dd, mtu);
527
528         dev->sm_sl = pip->neighbormtu_mastersmsl & 0xF;
529
530         /* We only support VL0 */
531         if (((pip->operationalvl_pei_peo_fpi_fpo >> 4) & 0xF) > 1)
532                 goto err;
533
534         if (pip->mkey_violations == 0)
535                 dev->mkey_violations = 0;
536
537         /*
538          * Hardware counter can't be reset so snapshot and subtract
539          * later.
540          */
541         if (pip->pkey_violations == 0)
542                 dev->z_pkey_violations = ipath_get_cr_errpkey(dd);
543
544         if (pip->qkey_violations == 0)
545                 dev->qkey_violations = 0;
546
547         ore = pip->localphyerrors_overrunerrors;
548         if (set_phyerrthreshold(dd, (ore >> 4) & 0xF))
549                 goto err;
550
551         if (set_overrunthreshold(dd, (ore & 0xF)))
552                 goto err;
553
554         dev->subnet_timeout = pip->clientrereg_resv_subnetto & 0x1F;
555
556         if (pip->clientrereg_resv_subnetto & 0x80) {
557                 clientrereg = 1;
558                 event.event = IB_EVENT_CLIENT_REREGISTER;
559                 ib_dispatch_event(&event);
560         }
561
562         /*
563          * Do the port state change now that the other link parameters
564          * have been set.
565          * Changing the port physical state only makes sense if the link
566          * is down or is being set to down.
567          */
568         state = pip->linkspeed_portstate & 0xF;
569         lstate = (pip->portphysstate_linkdown >> 4) & 0xF;
570         if (lstate && !(state == IB_PORT_DOWN || state == IB_PORT_NOP))
571                 goto err;
572
573         /*
574          * Only state changes of DOWN, ARM, and ACTIVE are valid
575          * and must be in the correct state to take effect (see 7.2.6).
576          */
577         switch (state) {
578         case IB_PORT_NOP:
579                 if (lstate == 0)
580                         break;
581                 /* FALLTHROUGH */
582         case IB_PORT_DOWN:
583                 if (lstate == 0)
584                         lstate = IPATH_IB_LINKDOWN_ONLY;
585                 else if (lstate == 1)
586                         lstate = IPATH_IB_LINKDOWN_SLEEP;
587                 else if (lstate == 2)
588                         lstate = IPATH_IB_LINKDOWN;
589                 else if (lstate == 3)
590                         lstate = IPATH_IB_LINKDOWN_DISABLE;
591                 else
592                         goto err;
593                 ipath_set_linkstate(dd, lstate);
594                 if (lstate == IPATH_IB_LINKDOWN_DISABLE) {
595                         ret = IB_MAD_RESULT_SUCCESS | IB_MAD_RESULT_CONSUMED;
596                         goto done;
597                 }
598                 ipath_wait_linkstate(dd, IPATH_LINKINIT | IPATH_LINKARMED |
599                                 IPATH_LINKACTIVE, 1000);
600                 break;
601         case IB_PORT_ARMED:
602                 ipath_set_linkstate(dd, IPATH_IB_LINKARM);
603                 break;
604         case IB_PORT_ACTIVE:
605                 ipath_set_linkstate(dd, IPATH_IB_LINKACTIVE);
606                 break;
607         default:
608                 /* XXX We have already partially updated our state! */
609                 goto err;
610         }
611
612         ret = recv_subn_get_portinfo(smp, ibdev, port);
613
614         if (clientrereg)
615                 pip->clientrereg_resv_subnetto |= 0x80;
616
617         goto done;
618
619 err:
620         smp->status |= IB_SMP_INVALID_FIELD;
621         ret = recv_subn_get_portinfo(smp, ibdev, port);
622
623 done:
624         return ret;
625 }
626
627 /**
628  * rm_pkey - decrecment the reference count for the given PKEY
629  * @dd: the infinipath device
630  * @key: the PKEY index
631  *
632  * Return true if this was the last reference and the hardware table entry
633  * needs to be changed.
634  */
635 static int rm_pkey(struct ipath_devdata *dd, u16 key)
636 {
637         int i;
638         int ret;
639
640         for (i = 0; i < ARRAY_SIZE(dd->ipath_pkeys); i++) {
641                 if (dd->ipath_pkeys[i] != key)
642                         continue;
643                 if (atomic_dec_and_test(&dd->ipath_pkeyrefs[i])) {
644                         dd->ipath_pkeys[i] = 0;
645                         ret = 1;
646                         goto bail;
647                 }
648                 break;
649         }
650
651         ret = 0;
652
653 bail:
654         return ret;
655 }
656
657 /**
658  * add_pkey - add the given PKEY to the hardware table
659  * @dd: the infinipath device
660  * @key: the PKEY
661  *
662  * Return an error code if unable to add the entry, zero if no change,
663  * or 1 if the hardware PKEY register needs to be updated.
664  */
665 static int add_pkey(struct ipath_devdata *dd, u16 key)
666 {
667         int i;
668         u16 lkey = key & 0x7FFF;
669         int any = 0;
670         int ret;
671
672         if (lkey == 0x7FFF) {
673                 ret = 0;
674                 goto bail;
675         }
676
677         /* Look for an empty slot or a matching PKEY. */
678         for (i = 0; i < ARRAY_SIZE(dd->ipath_pkeys); i++) {
679                 if (!dd->ipath_pkeys[i]) {
680                         any++;
681                         continue;
682                 }
683                 /* If it matches exactly, try to increment the ref count */
684                 if (dd->ipath_pkeys[i] == key) {
685                         if (atomic_inc_return(&dd->ipath_pkeyrefs[i]) > 1) {
686                                 ret = 0;
687                                 goto bail;
688                         }
689                         /* Lost the race. Look for an empty slot below. */
690                         atomic_dec(&dd->ipath_pkeyrefs[i]);
691                         any++;
692                 }
693                 /*
694                  * It makes no sense to have both the limited and unlimited
695                  * PKEY set at the same time since the unlimited one will
696                  * disable the limited one.
697                  */
698                 if ((dd->ipath_pkeys[i] & 0x7FFF) == lkey) {
699                         ret = -EEXIST;
700                         goto bail;
701                 }
702         }
703         if (!any) {
704                 ret = -EBUSY;
705                 goto bail;
706         }
707         for (i = 0; i < ARRAY_SIZE(dd->ipath_pkeys); i++) {
708                 if (!dd->ipath_pkeys[i] &&
709                     atomic_inc_return(&dd->ipath_pkeyrefs[i]) == 1) {
710                         /* for ipathstats, etc. */
711                         ipath_stats.sps_pkeys[i] = lkey;
712                         dd->ipath_pkeys[i] = key;
713                         ret = 1;
714                         goto bail;
715                 }
716         }
717         ret = -EBUSY;
718
719 bail:
720         return ret;
721 }
722
723 /**
724  * set_pkeys - set the PKEY table for port 0
725  * @dd: the infinipath device
726  * @pkeys: the PKEY table
727  */
728 static int set_pkeys(struct ipath_devdata *dd, u16 *pkeys)
729 {
730         struct ipath_portdata *pd;
731         int i;
732         int changed = 0;
733
734         /* always a kernel port, no locking needed */
735         pd = dd->ipath_pd[0];
736
737         for (i = 0; i < ARRAY_SIZE(pd->port_pkeys); i++) {
738                 u16 key = pkeys[i];
739                 u16 okey = pd->port_pkeys[i];
740
741                 if (key == okey)
742                         continue;
743                 /*
744                  * The value of this PKEY table entry is changing.
745                  * Remove the old entry in the hardware's array of PKEYs.
746                  */
747                 if (okey & 0x7FFF)
748                         changed |= rm_pkey(dd, okey);
749                 if (key & 0x7FFF) {
750                         int ret = add_pkey(dd, key);
751
752                         if (ret < 0)
753                                 key = 0;
754                         else
755                                 changed |= ret;
756                 }
757                 pd->port_pkeys[i] = key;
758         }
759         if (changed) {
760                 u64 pkey;
761
762                 pkey = (u64) dd->ipath_pkeys[0] |
763                         ((u64) dd->ipath_pkeys[1] << 16) |
764                         ((u64) dd->ipath_pkeys[2] << 32) |
765                         ((u64) dd->ipath_pkeys[3] << 48);
766                 ipath_cdbg(VERBOSE, "p0 new pkey reg %llx\n",
767                            (unsigned long long) pkey);
768                 ipath_write_kreg(dd, dd->ipath_kregs->kr_partitionkey,
769                                  pkey);
770         }
771         return 0;
772 }
773
774 static int recv_subn_set_pkeytable(struct ib_smp *smp,
775                                    struct ib_device *ibdev)
776 {
777         u32 startpx = 32 * (be32_to_cpu(smp->attr_mod) & 0xffff);
778         __be16 *p = (__be16 *) smp->data;
779         u16 *q = (u16 *) smp->data;
780         struct ipath_ibdev *dev = to_idev(ibdev);
781         unsigned i, n = ipath_get_npkeys(dev->dd);
782
783         for (i = 0; i < n; i++)
784                 q[i] = be16_to_cpu(p[i]);
785
786         if (startpx != 0 || set_pkeys(dev->dd, q) != 0)
787                 smp->status |= IB_SMP_INVALID_FIELD;
788
789         return recv_subn_get_pkeytable(smp, ibdev);
790 }
791
792 #define IB_PMA_CLASS_PORT_INFO          __constant_htons(0x0001)
793 #define IB_PMA_PORT_SAMPLES_CONTROL     __constant_htons(0x0010)
794 #define IB_PMA_PORT_SAMPLES_RESULT      __constant_htons(0x0011)
795 #define IB_PMA_PORT_COUNTERS            __constant_htons(0x0012)
796 #define IB_PMA_PORT_COUNTERS_EXT        __constant_htons(0x001D)
797 #define IB_PMA_PORT_SAMPLES_RESULT_EXT  __constant_htons(0x001E)
798
799 struct ib_perf {
800         u8 base_version;
801         u8 mgmt_class;
802         u8 class_version;
803         u8 method;
804         __be16 status;
805         __be16 unused;
806         __be64 tid;
807         __be16 attr_id;
808         __be16 resv;
809         __be32 attr_mod;
810         u8 reserved[40];
811         u8 data[192];
812 } __attribute__ ((packed));
813
814 struct ib_pma_classportinfo {
815         u8 base_version;
816         u8 class_version;
817         __be16 cap_mask;
818         u8 reserved[3];
819         u8 resp_time_value;     /* only lower 5 bits */
820         union ib_gid redirect_gid;
821         __be32 redirect_tc_sl_fl;       /* 8, 4, 20 bits respectively */
822         __be16 redirect_lid;
823         __be16 redirect_pkey;
824         __be32 redirect_qp;     /* only lower 24 bits */
825         __be32 redirect_qkey;
826         union ib_gid trap_gid;
827         __be32 trap_tc_sl_fl;   /* 8, 4, 20 bits respectively */
828         __be16 trap_lid;
829         __be16 trap_pkey;
830         __be32 trap_hl_qp;      /* 8, 24 bits respectively */
831         __be32 trap_qkey;
832 } __attribute__ ((packed));
833
834 struct ib_pma_portsamplescontrol {
835         u8 opcode;
836         u8 port_select;
837         u8 tick;
838         u8 counter_width;       /* only lower 3 bits */
839         __be32 counter_mask0_9; /* 2, 10 * 3, bits */
840         __be16 counter_mask10_14;       /* 1, 5 * 3, bits */
841         u8 sample_mechanisms;
842         u8 sample_status;       /* only lower 2 bits */
843         __be64 option_mask;
844         __be64 vendor_mask;
845         __be32 sample_start;
846         __be32 sample_interval;
847         __be16 tag;
848         __be16 counter_select[15];
849 } __attribute__ ((packed));
850
851 struct ib_pma_portsamplesresult {
852         __be16 tag;
853         __be16 sample_status;   /* only lower 2 bits */
854         __be32 counter[15];
855 } __attribute__ ((packed));
856
857 struct ib_pma_portsamplesresult_ext {
858         __be16 tag;
859         __be16 sample_status;   /* only lower 2 bits */
860         __be32 extended_width;  /* only upper 2 bits */
861         __be64 counter[15];
862 } __attribute__ ((packed));
863
864 struct ib_pma_portcounters {
865         u8 reserved;
866         u8 port_select;
867         __be16 counter_select;
868         __be16 symbol_error_counter;
869         u8 link_error_recovery_counter;
870         u8 link_downed_counter;
871         __be16 port_rcv_errors;
872         __be16 port_rcv_remphys_errors;
873         __be16 port_rcv_switch_relay_errors;
874         __be16 port_xmit_discards;
875         u8 port_xmit_constraint_errors;
876         u8 port_rcv_constraint_errors;
877         u8 reserved1;
878         u8 lli_ebor_errors;     /* 4, 4, bits */
879         __be16 reserved2;
880         __be16 vl15_dropped;
881         __be32 port_xmit_data;
882         __be32 port_rcv_data;
883         __be32 port_xmit_packets;
884         __be32 port_rcv_packets;
885 } __attribute__ ((packed));
886
887 #define IB_PMA_SEL_SYMBOL_ERROR                 __constant_htons(0x0001)
888 #define IB_PMA_SEL_LINK_ERROR_RECOVERY          __constant_htons(0x0002)
889 #define IB_PMA_SEL_LINK_DOWNED                  __constant_htons(0x0004)
890 #define IB_PMA_SEL_PORT_RCV_ERRORS              __constant_htons(0x0008)
891 #define IB_PMA_SEL_PORT_RCV_REMPHYS_ERRORS      __constant_htons(0x0010)
892 #define IB_PMA_SEL_PORT_XMIT_DISCARDS           __constant_htons(0x0040)
893 #define IB_PMA_SEL_LOCAL_LINK_INTEGRITY_ERRORS  __constant_htons(0x0200)
894 #define IB_PMA_SEL_EXCESSIVE_BUFFER_OVERRUNS    __constant_htons(0x0400)
895 #define IB_PMA_SEL_PORT_VL15_DROPPED            __constant_htons(0x0800)
896 #define IB_PMA_SEL_PORT_XMIT_DATA               __constant_htons(0x1000)
897 #define IB_PMA_SEL_PORT_RCV_DATA                __constant_htons(0x2000)
898 #define IB_PMA_SEL_PORT_XMIT_PACKETS            __constant_htons(0x4000)
899 #define IB_PMA_SEL_PORT_RCV_PACKETS             __constant_htons(0x8000)
900
901 struct ib_pma_portcounters_ext {
902         u8 reserved;
903         u8 port_select;
904         __be16 counter_select;
905         __be32 reserved1;
906         __be64 port_xmit_data;
907         __be64 port_rcv_data;
908         __be64 port_xmit_packets;
909         __be64 port_rcv_packets;
910         __be64 port_unicast_xmit_packets;
911         __be64 port_unicast_rcv_packets;
912         __be64 port_multicast_xmit_packets;
913         __be64 port_multicast_rcv_packets;
914 } __attribute__ ((packed));
915
916 #define IB_PMA_SELX_PORT_XMIT_DATA              __constant_htons(0x0001)
917 #define IB_PMA_SELX_PORT_RCV_DATA               __constant_htons(0x0002)
918 #define IB_PMA_SELX_PORT_XMIT_PACKETS           __constant_htons(0x0004)
919 #define IB_PMA_SELX_PORT_RCV_PACKETS            __constant_htons(0x0008)
920 #define IB_PMA_SELX_PORT_UNI_XMIT_PACKETS       __constant_htons(0x0010)
921 #define IB_PMA_SELX_PORT_UNI_RCV_PACKETS        __constant_htons(0x0020)
922 #define IB_PMA_SELX_PORT_MULTI_XMIT_PACKETS     __constant_htons(0x0040)
923 #define IB_PMA_SELX_PORT_MULTI_RCV_PACKETS      __constant_htons(0x0080)
924
925 static int recv_pma_get_classportinfo(struct ib_perf *pmp)
926 {
927         struct ib_pma_classportinfo *p =
928                 (struct ib_pma_classportinfo *)pmp->data;
929
930         memset(pmp->data, 0, sizeof(pmp->data));
931
932         if (pmp->attr_mod != 0)
933                 pmp->status |= IB_SMP_INVALID_FIELD;
934
935         /* Indicate AllPortSelect is valid (only one port anyway) */
936         p->cap_mask = __constant_cpu_to_be16(1 << 8);
937         p->base_version = 1;
938         p->class_version = 1;
939         /*
940          * Expected response time is 4.096 usec. * 2^18 == 1.073741824
941          * sec.
942          */
943         p->resp_time_value = 18;
944
945         return reply((struct ib_smp *) pmp);
946 }
947
948 /*
949  * The PortSamplesControl.CounterMasks field is an array of 3 bit fields
950  * which specify the N'th counter's capabilities. See ch. 16.1.3.2.
951  * We support 5 counters which only count the mandatory quantities.
952  */
953 #define COUNTER_MASK(q, n) (q << ((9 - n) * 3))
954 #define COUNTER_MASK0_9 \
955         __constant_cpu_to_be32(COUNTER_MASK(1, 0) | \
956                                COUNTER_MASK(1, 1) | \
957                                COUNTER_MASK(1, 2) | \
958                                COUNTER_MASK(1, 3) | \
959                                COUNTER_MASK(1, 4))
960
961 static int recv_pma_get_portsamplescontrol(struct ib_perf *pmp,
962                                            struct ib_device *ibdev, u8 port)
963 {
964         struct ib_pma_portsamplescontrol *p =
965                 (struct ib_pma_portsamplescontrol *)pmp->data;
966         struct ipath_ibdev *dev = to_idev(ibdev);
967         struct ipath_cregs const *crp = dev->dd->ipath_cregs;
968         unsigned long flags;
969         u8 port_select = p->port_select;
970
971         memset(pmp->data, 0, sizeof(pmp->data));
972
973         p->port_select = port_select;
974         if (pmp->attr_mod != 0 ||
975             (port_select != port && port_select != 0xFF))
976                 pmp->status |= IB_SMP_INVALID_FIELD;
977         /*
978          * Ticks are 10x the link transfer period which for 2.5Gbs is 4
979          * nsec.  0 == 4 nsec., 1 == 8 nsec., ..., 255 == 1020 nsec.  Sample
980          * intervals are counted in ticks.  Since we use Linux timers, that
981          * count in jiffies, we can't sample for less than 1000 ticks if HZ
982          * == 1000 (4000 ticks if HZ is 250).  link_speed_active returns 2 for
983          * DDR, 1 for SDR, set the tick to 1 for DDR, 0 for SDR on chips that
984          * have hardware support for delaying packets.
985          */
986         if (crp->cr_psstat)
987                 p->tick = dev->dd->ipath_link_speed_active - 1;
988         else
989                 p->tick = 250;          /* 1 usec. */
990         p->counter_width = 4;   /* 32 bit counters */
991         p->counter_mask0_9 = COUNTER_MASK0_9;
992         spin_lock_irqsave(&dev->pending_lock, flags);
993         if (crp->cr_psstat)
994                 p->sample_status = ipath_read_creg32(dev->dd, crp->cr_psstat);
995         else
996                 p->sample_status = dev->pma_sample_status;
997         p->sample_start = cpu_to_be32(dev->pma_sample_start);
998         p->sample_interval = cpu_to_be32(dev->pma_sample_interval);
999         p->tag = cpu_to_be16(dev->pma_tag);
1000         p->counter_select[0] = dev->pma_counter_select[0];
1001         p->counter_select[1] = dev->pma_counter_select[1];
1002         p->counter_select[2] = dev->pma_counter_select[2];
1003         p->counter_select[3] = dev->pma_counter_select[3];
1004         p->counter_select[4] = dev->pma_counter_select[4];
1005         spin_unlock_irqrestore(&dev->pending_lock, flags);
1006
1007         return reply((struct ib_smp *) pmp);
1008 }
1009
1010 static int recv_pma_set_portsamplescontrol(struct ib_perf *pmp,
1011                                            struct ib_device *ibdev, u8 port)
1012 {
1013         struct ib_pma_portsamplescontrol *p =
1014                 (struct ib_pma_portsamplescontrol *)pmp->data;
1015         struct ipath_ibdev *dev = to_idev(ibdev);
1016         struct ipath_cregs const *crp = dev->dd->ipath_cregs;
1017         unsigned long flags;
1018         u8 status;
1019         int ret;
1020
1021         if (pmp->attr_mod != 0 ||
1022             (p->port_select != port && p->port_select != 0xFF)) {
1023                 pmp->status |= IB_SMP_INVALID_FIELD;
1024                 ret = reply((struct ib_smp *) pmp);
1025                 goto bail;
1026         }
1027
1028         spin_lock_irqsave(&dev->pending_lock, flags);
1029         if (crp->cr_psstat)
1030                 status = ipath_read_creg32(dev->dd, crp->cr_psstat);
1031         else
1032                 status = dev->pma_sample_status;
1033         if (status == IB_PMA_SAMPLE_STATUS_DONE) {
1034                 dev->pma_sample_start = be32_to_cpu(p->sample_start);
1035                 dev->pma_sample_interval = be32_to_cpu(p->sample_interval);
1036                 dev->pma_tag = be16_to_cpu(p->tag);
1037                 dev->pma_counter_select[0] = p->counter_select[0];
1038                 dev->pma_counter_select[1] = p->counter_select[1];
1039                 dev->pma_counter_select[2] = p->counter_select[2];
1040                 dev->pma_counter_select[3] = p->counter_select[3];
1041                 dev->pma_counter_select[4] = p->counter_select[4];
1042                 if (crp->cr_psstat) {
1043                         ipath_write_creg(dev->dd, crp->cr_psinterval,
1044                                          dev->pma_sample_interval);
1045                         ipath_write_creg(dev->dd, crp->cr_psstart,
1046                                          dev->pma_sample_start);
1047                 } else
1048                         dev->pma_sample_status = IB_PMA_SAMPLE_STATUS_STARTED;
1049         }
1050         spin_unlock_irqrestore(&dev->pending_lock, flags);
1051
1052         ret = recv_pma_get_portsamplescontrol(pmp, ibdev, port);
1053
1054 bail:
1055         return ret;
1056 }
1057
1058 static u64 get_counter(struct ipath_ibdev *dev,
1059                        struct ipath_cregs const *crp,
1060                        __be16 sel)
1061 {
1062         u64 ret;
1063
1064         switch (sel) {
1065         case IB_PMA_PORT_XMIT_DATA:
1066                 ret = (crp->cr_psxmitdatacount) ?
1067                         ipath_read_creg32(dev->dd, crp->cr_psxmitdatacount) :
1068                         dev->ipath_sword;
1069                 break;
1070         case IB_PMA_PORT_RCV_DATA:
1071                 ret = (crp->cr_psrcvdatacount) ?
1072                         ipath_read_creg32(dev->dd, crp->cr_psrcvdatacount) :
1073                         dev->ipath_rword;
1074                 break;
1075         case IB_PMA_PORT_XMIT_PKTS:
1076                 ret = (crp->cr_psxmitpktscount) ?
1077                         ipath_read_creg32(dev->dd, crp->cr_psxmitpktscount) :
1078                         dev->ipath_spkts;
1079                 break;
1080         case IB_PMA_PORT_RCV_PKTS:
1081                 ret = (crp->cr_psrcvpktscount) ?
1082                         ipath_read_creg32(dev->dd, crp->cr_psrcvpktscount) :
1083                         dev->ipath_rpkts;
1084                 break;
1085         case IB_PMA_PORT_XMIT_WAIT:
1086                 ret = (crp->cr_psxmitwaitcount) ?
1087                         ipath_read_creg32(dev->dd, crp->cr_psxmitwaitcount) :
1088                         dev->ipath_xmit_wait;
1089                 break;
1090         default:
1091                 ret = 0;
1092         }
1093
1094         return ret;
1095 }
1096
1097 static int recv_pma_get_portsamplesresult(struct ib_perf *pmp,
1098                                           struct ib_device *ibdev)
1099 {
1100         struct ib_pma_portsamplesresult *p =
1101                 (struct ib_pma_portsamplesresult *)pmp->data;
1102         struct ipath_ibdev *dev = to_idev(ibdev);
1103         struct ipath_cregs const *crp = dev->dd->ipath_cregs;
1104         u8 status;
1105         int i;
1106
1107         memset(pmp->data, 0, sizeof(pmp->data));
1108         p->tag = cpu_to_be16(dev->pma_tag);
1109         if (crp->cr_psstat)
1110                 status = ipath_read_creg32(dev->dd, crp->cr_psstat);
1111         else
1112                 status = dev->pma_sample_status;
1113         p->sample_status = cpu_to_be16(status);
1114         for (i = 0; i < ARRAY_SIZE(dev->pma_counter_select); i++)
1115                 p->counter[i] = (status != IB_PMA_SAMPLE_STATUS_DONE) ? 0 :
1116                     cpu_to_be32(
1117                         get_counter(dev, crp, dev->pma_counter_select[i]));
1118
1119         return reply((struct ib_smp *) pmp);
1120 }
1121
1122 static int recv_pma_get_portsamplesresult_ext(struct ib_perf *pmp,
1123                                               struct ib_device *ibdev)
1124 {
1125         struct ib_pma_portsamplesresult_ext *p =
1126                 (struct ib_pma_portsamplesresult_ext *)pmp->data;
1127         struct ipath_ibdev *dev = to_idev(ibdev);
1128         struct ipath_cregs const *crp = dev->dd->ipath_cregs;
1129         u8 status;
1130         int i;
1131
1132         memset(pmp->data, 0, sizeof(pmp->data));
1133         p->tag = cpu_to_be16(dev->pma_tag);
1134         if (crp->cr_psstat)
1135                 status = ipath_read_creg32(dev->dd, crp->cr_psstat);
1136         else
1137                 status = dev->pma_sample_status;
1138         p->sample_status = cpu_to_be16(status);
1139         /* 64 bits */
1140         p->extended_width = __constant_cpu_to_be32(0x80000000);
1141         for (i = 0; i < ARRAY_SIZE(dev->pma_counter_select); i++)
1142                 p->counter[i] = (status != IB_PMA_SAMPLE_STATUS_DONE) ? 0 :
1143                     cpu_to_be64(
1144                         get_counter(dev, crp, dev->pma_counter_select[i]));
1145
1146         return reply((struct ib_smp *) pmp);
1147 }
1148
1149 static int recv_pma_get_portcounters(struct ib_perf *pmp,
1150                                      struct ib_device *ibdev, u8 port)
1151 {
1152         struct ib_pma_portcounters *p = (struct ib_pma_portcounters *)
1153                 pmp->data;
1154         struct ipath_ibdev *dev = to_idev(ibdev);
1155         struct ipath_verbs_counters cntrs;
1156         u8 port_select = p->port_select;
1157
1158         ipath_get_counters(dev->dd, &cntrs);
1159
1160         /* Adjust counters for any resets done. */
1161         cntrs.symbol_error_counter -= dev->z_symbol_error_counter;
1162         cntrs.link_error_recovery_counter -=
1163                 dev->z_link_error_recovery_counter;
1164         cntrs.link_downed_counter -= dev->z_link_downed_counter;
1165         cntrs.port_rcv_errors += dev->rcv_errors;
1166         cntrs.port_rcv_errors -= dev->z_port_rcv_errors;
1167         cntrs.port_rcv_remphys_errors -= dev->z_port_rcv_remphys_errors;
1168         cntrs.port_xmit_discards -= dev->z_port_xmit_discards;
1169         cntrs.port_xmit_data -= dev->z_port_xmit_data;
1170         cntrs.port_rcv_data -= dev->z_port_rcv_data;
1171         cntrs.port_xmit_packets -= dev->z_port_xmit_packets;
1172         cntrs.port_rcv_packets -= dev->z_port_rcv_packets;
1173         cntrs.local_link_integrity_errors -=
1174                 dev->z_local_link_integrity_errors;
1175         cntrs.excessive_buffer_overrun_errors -=
1176                 dev->z_excessive_buffer_overrun_errors;
1177         cntrs.vl15_dropped -= dev->z_vl15_dropped;
1178         cntrs.vl15_dropped += dev->n_vl15_dropped;
1179
1180         memset(pmp->data, 0, sizeof(pmp->data));
1181
1182         p->port_select = port_select;
1183         if (pmp->attr_mod != 0 ||
1184             (port_select != port && port_select != 0xFF))
1185                 pmp->status |= IB_SMP_INVALID_FIELD;
1186
1187         if (cntrs.symbol_error_counter > 0xFFFFUL)
1188                 p->symbol_error_counter = __constant_cpu_to_be16(0xFFFF);
1189         else
1190                 p->symbol_error_counter =
1191                         cpu_to_be16((u16)cntrs.symbol_error_counter);
1192         if (cntrs.link_error_recovery_counter > 0xFFUL)
1193                 p->link_error_recovery_counter = 0xFF;
1194         else
1195                 p->link_error_recovery_counter =
1196                         (u8)cntrs.link_error_recovery_counter;
1197         if (cntrs.link_downed_counter > 0xFFUL)
1198                 p->link_downed_counter = 0xFF;
1199         else
1200                 p->link_downed_counter = (u8)cntrs.link_downed_counter;
1201         if (cntrs.port_rcv_errors > 0xFFFFUL)
1202                 p->port_rcv_errors = __constant_cpu_to_be16(0xFFFF);
1203         else
1204                 p->port_rcv_errors =
1205                         cpu_to_be16((u16) cntrs.port_rcv_errors);
1206         if (cntrs.port_rcv_remphys_errors > 0xFFFFUL)
1207                 p->port_rcv_remphys_errors = __constant_cpu_to_be16(0xFFFF);
1208         else
1209                 p->port_rcv_remphys_errors =
1210                         cpu_to_be16((u16)cntrs.port_rcv_remphys_errors);
1211         if (cntrs.port_xmit_discards > 0xFFFFUL)
1212                 p->port_xmit_discards = __constant_cpu_to_be16(0xFFFF);
1213         else
1214                 p->port_xmit_discards =
1215                         cpu_to_be16((u16)cntrs.port_xmit_discards);
1216         if (cntrs.local_link_integrity_errors > 0xFUL)
1217                 cntrs.local_link_integrity_errors = 0xFUL;
1218         if (cntrs.excessive_buffer_overrun_errors > 0xFUL)
1219                 cntrs.excessive_buffer_overrun_errors = 0xFUL;
1220         p->lli_ebor_errors = (cntrs.local_link_integrity_errors << 4) |
1221                 cntrs.excessive_buffer_overrun_errors;
1222         if (cntrs.vl15_dropped > 0xFFFFUL)
1223                 p->vl15_dropped = __constant_cpu_to_be16(0xFFFF);
1224         else
1225                 p->vl15_dropped = cpu_to_be16((u16)cntrs.vl15_dropped);
1226         if (cntrs.port_xmit_data > 0xFFFFFFFFUL)
1227                 p->port_xmit_data = __constant_cpu_to_be32(0xFFFFFFFF);
1228         else
1229                 p->port_xmit_data = cpu_to_be32((u32)cntrs.port_xmit_data);
1230         if (cntrs.port_rcv_data > 0xFFFFFFFFUL)
1231                 p->port_rcv_data = __constant_cpu_to_be32(0xFFFFFFFF);
1232         else
1233                 p->port_rcv_data = cpu_to_be32((u32)cntrs.port_rcv_data);
1234         if (cntrs.port_xmit_packets > 0xFFFFFFFFUL)
1235                 p->port_xmit_packets = __constant_cpu_to_be32(0xFFFFFFFF);
1236         else
1237                 p->port_xmit_packets =
1238                         cpu_to_be32((u32)cntrs.port_xmit_packets);
1239         if (cntrs.port_rcv_packets > 0xFFFFFFFFUL)
1240                 p->port_rcv_packets = __constant_cpu_to_be32(0xFFFFFFFF);
1241         else
1242                 p->port_rcv_packets =
1243                         cpu_to_be32((u32) cntrs.port_rcv_packets);
1244
1245         return reply((struct ib_smp *) pmp);
1246 }
1247
1248 static int recv_pma_get_portcounters_ext(struct ib_perf *pmp,
1249                                          struct ib_device *ibdev, u8 port)
1250 {
1251         struct ib_pma_portcounters_ext *p =
1252                 (struct ib_pma_portcounters_ext *)pmp->data;
1253         struct ipath_ibdev *dev = to_idev(ibdev);
1254         u64 swords, rwords, spkts, rpkts, xwait;
1255         u8 port_select = p->port_select;
1256
1257         ipath_snapshot_counters(dev->dd, &swords, &rwords, &spkts,
1258                                 &rpkts, &xwait);
1259
1260         /* Adjust counters for any resets done. */
1261         swords -= dev->z_port_xmit_data;
1262         rwords -= dev->z_port_rcv_data;
1263         spkts -= dev->z_port_xmit_packets;
1264         rpkts -= dev->z_port_rcv_packets;
1265
1266         memset(pmp->data, 0, sizeof(pmp->data));
1267
1268         p->port_select = port_select;
1269         if (pmp->attr_mod != 0 ||
1270             (port_select != port && port_select != 0xFF))
1271                 pmp->status |= IB_SMP_INVALID_FIELD;
1272
1273         p->port_xmit_data = cpu_to_be64(swords);
1274         p->port_rcv_data = cpu_to_be64(rwords);
1275         p->port_xmit_packets = cpu_to_be64(spkts);
1276         p->port_rcv_packets = cpu_to_be64(rpkts);
1277         p->port_unicast_xmit_packets = cpu_to_be64(dev->n_unicast_xmit);
1278         p->port_unicast_rcv_packets = cpu_to_be64(dev->n_unicast_rcv);
1279         p->port_multicast_xmit_packets = cpu_to_be64(dev->n_multicast_xmit);
1280         p->port_multicast_rcv_packets = cpu_to_be64(dev->n_multicast_rcv);
1281
1282         return reply((struct ib_smp *) pmp);
1283 }
1284
1285 static int recv_pma_set_portcounters(struct ib_perf *pmp,
1286                                      struct ib_device *ibdev, u8 port)
1287 {
1288         struct ib_pma_portcounters *p = (struct ib_pma_portcounters *)
1289                 pmp->data;
1290         struct ipath_ibdev *dev = to_idev(ibdev);
1291         struct ipath_verbs_counters cntrs;
1292
1293         /*
1294          * Since the HW doesn't support clearing counters, we save the
1295          * current count and subtract it from future responses.
1296          */
1297         ipath_get_counters(dev->dd, &cntrs);
1298
1299         if (p->counter_select & IB_PMA_SEL_SYMBOL_ERROR)
1300                 dev->z_symbol_error_counter = cntrs.symbol_error_counter;
1301
1302         if (p->counter_select & IB_PMA_SEL_LINK_ERROR_RECOVERY)
1303                 dev->z_link_error_recovery_counter =
1304                         cntrs.link_error_recovery_counter;
1305
1306         if (p->counter_select & IB_PMA_SEL_LINK_DOWNED)
1307                 dev->z_link_downed_counter = cntrs.link_downed_counter;
1308
1309         if (p->counter_select & IB_PMA_SEL_PORT_RCV_ERRORS)
1310                 dev->z_port_rcv_errors =
1311                         cntrs.port_rcv_errors + dev->rcv_errors;
1312
1313         if (p->counter_select & IB_PMA_SEL_PORT_RCV_REMPHYS_ERRORS)
1314                 dev->z_port_rcv_remphys_errors =
1315                         cntrs.port_rcv_remphys_errors;
1316
1317         if (p->counter_select & IB_PMA_SEL_PORT_XMIT_DISCARDS)
1318                 dev->z_port_xmit_discards = cntrs.port_xmit_discards;
1319
1320         if (p->counter_select & IB_PMA_SEL_LOCAL_LINK_INTEGRITY_ERRORS)
1321                 dev->z_local_link_integrity_errors =
1322                         cntrs.local_link_integrity_errors;
1323
1324         if (p->counter_select & IB_PMA_SEL_EXCESSIVE_BUFFER_OVERRUNS)
1325                 dev->z_excessive_buffer_overrun_errors =
1326                         cntrs.excessive_buffer_overrun_errors;
1327
1328         if (p->counter_select & IB_PMA_SEL_PORT_VL15_DROPPED) {
1329                 dev->n_vl15_dropped = 0;
1330                 dev->z_vl15_dropped = cntrs.vl15_dropped;
1331         }
1332
1333         if (p->counter_select & IB_PMA_SEL_PORT_XMIT_DATA)
1334                 dev->z_port_xmit_data = cntrs.port_xmit_data;
1335
1336         if (p->counter_select & IB_PMA_SEL_PORT_RCV_DATA)
1337                 dev->z_port_rcv_data = cntrs.port_rcv_data;
1338
1339         if (p->counter_select & IB_PMA_SEL_PORT_XMIT_PACKETS)
1340                 dev->z_port_xmit_packets = cntrs.port_xmit_packets;
1341
1342         if (p->counter_select & IB_PMA_SEL_PORT_RCV_PACKETS)
1343                 dev->z_port_rcv_packets = cntrs.port_rcv_packets;
1344
1345         return recv_pma_get_portcounters(pmp, ibdev, port);
1346 }
1347
1348 static int recv_pma_set_portcounters_ext(struct ib_perf *pmp,
1349                                          struct ib_device *ibdev, u8 port)
1350 {
1351         struct ib_pma_portcounters *p = (struct ib_pma_portcounters *)
1352                 pmp->data;
1353         struct ipath_ibdev *dev = to_idev(ibdev);
1354         u64 swords, rwords, spkts, rpkts, xwait;
1355
1356         ipath_snapshot_counters(dev->dd, &swords, &rwords, &spkts,
1357                                 &rpkts, &xwait);
1358
1359         if (p->counter_select & IB_PMA_SELX_PORT_XMIT_DATA)
1360                 dev->z_port_xmit_data = swords;
1361
1362         if (p->counter_select & IB_PMA_SELX_PORT_RCV_DATA)
1363                 dev->z_port_rcv_data = rwords;
1364
1365         if (p->counter_select & IB_PMA_SELX_PORT_XMIT_PACKETS)
1366                 dev->z_port_xmit_packets = spkts;
1367
1368         if (p->counter_select & IB_PMA_SELX_PORT_RCV_PACKETS)
1369                 dev->z_port_rcv_packets = rpkts;
1370
1371         if (p->counter_select & IB_PMA_SELX_PORT_UNI_XMIT_PACKETS)
1372                 dev->n_unicast_xmit = 0;
1373
1374         if (p->counter_select & IB_PMA_SELX_PORT_UNI_RCV_PACKETS)
1375                 dev->n_unicast_rcv = 0;
1376
1377         if (p->counter_select & IB_PMA_SELX_PORT_MULTI_XMIT_PACKETS)
1378                 dev->n_multicast_xmit = 0;
1379
1380         if (p->counter_select & IB_PMA_SELX_PORT_MULTI_RCV_PACKETS)
1381                 dev->n_multicast_rcv = 0;
1382
1383         return recv_pma_get_portcounters_ext(pmp, ibdev, port);
1384 }
1385
1386 static int process_subn(struct ib_device *ibdev, int mad_flags,
1387                         u8 port_num, struct ib_mad *in_mad,
1388                         struct ib_mad *out_mad)
1389 {
1390         struct ib_smp *smp = (struct ib_smp *)out_mad;
1391         struct ipath_ibdev *dev = to_idev(ibdev);
1392         int ret;
1393
1394         *out_mad = *in_mad;
1395         if (smp->class_version != 1) {
1396                 smp->status |= IB_SMP_UNSUP_VERSION;
1397                 ret = reply(smp);
1398                 goto bail;
1399         }
1400
1401         /* Is the mkey in the process of expiring? */
1402         if (dev->mkey_lease_timeout &&
1403             time_after_eq(jiffies, dev->mkey_lease_timeout)) {
1404                 /* Clear timeout and mkey protection field. */
1405                 dev->mkey_lease_timeout = 0;
1406                 dev->mkeyprot = 0;
1407         }
1408
1409         /*
1410          * M_Key checking depends on
1411          * Portinfo:M_Key_protect_bits
1412          */
1413         if ((mad_flags & IB_MAD_IGNORE_MKEY) == 0 && dev->mkey != 0 &&
1414             dev->mkey != smp->mkey &&
1415             (smp->method == IB_MGMT_METHOD_SET ||
1416              (smp->method == IB_MGMT_METHOD_GET &&
1417               dev->mkeyprot >= 2))) {
1418                 if (dev->mkey_violations != 0xFFFF)
1419                         ++dev->mkey_violations;
1420                 if (dev->mkey_lease_timeout ||
1421                     dev->mkey_lease_period == 0) {
1422                         ret = IB_MAD_RESULT_SUCCESS |
1423                                 IB_MAD_RESULT_CONSUMED;
1424                         goto bail;
1425                 }
1426                 dev->mkey_lease_timeout = jiffies +
1427                         dev->mkey_lease_period * HZ;
1428                 /* Future: Generate a trap notice. */
1429                 ret = IB_MAD_RESULT_SUCCESS | IB_MAD_RESULT_CONSUMED;
1430                 goto bail;
1431         } else if (dev->mkey_lease_timeout)
1432                 dev->mkey_lease_timeout = 0;
1433
1434         switch (smp->method) {
1435         case IB_MGMT_METHOD_GET:
1436                 switch (smp->attr_id) {
1437                 case IB_SMP_ATTR_NODE_DESC:
1438                         ret = recv_subn_get_nodedescription(smp, ibdev);
1439                         goto bail;
1440                 case IB_SMP_ATTR_NODE_INFO:
1441                         ret = recv_subn_get_nodeinfo(smp, ibdev, port_num);
1442                         goto bail;
1443                 case IB_SMP_ATTR_GUID_INFO:
1444                         ret = recv_subn_get_guidinfo(smp, ibdev);
1445                         goto bail;
1446                 case IB_SMP_ATTR_PORT_INFO:
1447                         ret = recv_subn_get_portinfo(smp, ibdev, port_num);
1448                         goto bail;
1449                 case IB_SMP_ATTR_PKEY_TABLE:
1450                         ret = recv_subn_get_pkeytable(smp, ibdev);
1451                         goto bail;
1452                 case IB_SMP_ATTR_SM_INFO:
1453                         if (dev->port_cap_flags & IB_PORT_SM_DISABLED) {
1454                                 ret = IB_MAD_RESULT_SUCCESS |
1455                                         IB_MAD_RESULT_CONSUMED;
1456                                 goto bail;
1457                         }
1458                         if (dev->port_cap_flags & IB_PORT_SM) {
1459                                 ret = IB_MAD_RESULT_SUCCESS;
1460                                 goto bail;
1461                         }
1462                         /* FALLTHROUGH */
1463                 default:
1464                         smp->status |= IB_SMP_UNSUP_METH_ATTR;
1465                         ret = reply(smp);
1466                         goto bail;
1467                 }
1468
1469         case IB_MGMT_METHOD_SET:
1470                 switch (smp->attr_id) {
1471                 case IB_SMP_ATTR_GUID_INFO:
1472                         ret = recv_subn_set_guidinfo(smp, ibdev);
1473                         goto bail;
1474                 case IB_SMP_ATTR_PORT_INFO:
1475                         ret = recv_subn_set_portinfo(smp, ibdev, port_num);
1476                         goto bail;
1477                 case IB_SMP_ATTR_PKEY_TABLE:
1478                         ret = recv_subn_set_pkeytable(smp, ibdev);
1479                         goto bail;
1480                 case IB_SMP_ATTR_SM_INFO:
1481                         if (dev->port_cap_flags & IB_PORT_SM_DISABLED) {
1482                                 ret = IB_MAD_RESULT_SUCCESS |
1483                                         IB_MAD_RESULT_CONSUMED;
1484                                 goto bail;
1485                         }
1486                         if (dev->port_cap_flags & IB_PORT_SM) {
1487                                 ret = IB_MAD_RESULT_SUCCESS;
1488                                 goto bail;
1489                         }
1490                         /* FALLTHROUGH */
1491                 default:
1492                         smp->status |= IB_SMP_UNSUP_METH_ATTR;
1493                         ret = reply(smp);
1494                         goto bail;
1495                 }
1496
1497         case IB_MGMT_METHOD_TRAP:
1498         case IB_MGMT_METHOD_REPORT:
1499         case IB_MGMT_METHOD_REPORT_RESP:
1500         case IB_MGMT_METHOD_TRAP_REPRESS:
1501         case IB_MGMT_METHOD_GET_RESP:
1502                 /*
1503                  * The ib_mad module will call us to process responses
1504                  * before checking for other consumers.
1505                  * Just tell the caller to process it normally.
1506                  */
1507                 ret = IB_MAD_RESULT_SUCCESS;
1508                 goto bail;
1509         default:
1510                 smp->status |= IB_SMP_UNSUP_METHOD;
1511                 ret = reply(smp);
1512         }
1513
1514 bail:
1515         return ret;
1516 }
1517
1518 static int process_perf(struct ib_device *ibdev, u8 port_num,
1519                         struct ib_mad *in_mad,
1520                         struct ib_mad *out_mad)
1521 {
1522         struct ib_perf *pmp = (struct ib_perf *)out_mad;
1523         int ret;
1524
1525         *out_mad = *in_mad;
1526         if (pmp->class_version != 1) {
1527                 pmp->status |= IB_SMP_UNSUP_VERSION;
1528                 ret = reply((struct ib_smp *) pmp);
1529                 goto bail;
1530         }
1531
1532         switch (pmp->method) {
1533         case IB_MGMT_METHOD_GET:
1534                 switch (pmp->attr_id) {
1535                 case IB_PMA_CLASS_PORT_INFO:
1536                         ret = recv_pma_get_classportinfo(pmp);
1537                         goto bail;
1538                 case IB_PMA_PORT_SAMPLES_CONTROL:
1539                         ret = recv_pma_get_portsamplescontrol(pmp, ibdev,
1540                                                               port_num);
1541                         goto bail;
1542                 case IB_PMA_PORT_SAMPLES_RESULT:
1543                         ret = recv_pma_get_portsamplesresult(pmp, ibdev);
1544                         goto bail;
1545                 case IB_PMA_PORT_SAMPLES_RESULT_EXT:
1546                         ret = recv_pma_get_portsamplesresult_ext(pmp,
1547                                                                  ibdev);
1548                         goto bail;
1549                 case IB_PMA_PORT_COUNTERS:
1550                         ret = recv_pma_get_portcounters(pmp, ibdev,
1551                                                         port_num);
1552                         goto bail;
1553                 case IB_PMA_PORT_COUNTERS_EXT:
1554                         ret = recv_pma_get_portcounters_ext(pmp, ibdev,
1555                                                             port_num);
1556                         goto bail;
1557                 default:
1558                         pmp->status |= IB_SMP_UNSUP_METH_ATTR;
1559                         ret = reply((struct ib_smp *) pmp);
1560                         goto bail;
1561                 }
1562
1563         case IB_MGMT_METHOD_SET:
1564                 switch (pmp->attr_id) {
1565                 case IB_PMA_PORT_SAMPLES_CONTROL:
1566                         ret = recv_pma_set_portsamplescontrol(pmp, ibdev,
1567                                                               port_num);
1568                         goto bail;
1569                 case IB_PMA_PORT_COUNTERS:
1570                         ret = recv_pma_set_portcounters(pmp, ibdev,
1571                                                         port_num);
1572                         goto bail;
1573                 case IB_PMA_PORT_COUNTERS_EXT:
1574                         ret = recv_pma_set_portcounters_ext(pmp, ibdev,
1575                                                             port_num);
1576                         goto bail;
1577                 default:
1578                         pmp->status |= IB_SMP_UNSUP_METH_ATTR;
1579                         ret = reply((struct ib_smp *) pmp);
1580                         goto bail;
1581                 }
1582
1583         case IB_MGMT_METHOD_GET_RESP:
1584                 /*
1585                  * The ib_mad module will call us to process responses
1586                  * before checking for other consumers.
1587                  * Just tell the caller to process it normally.
1588                  */
1589                 ret = IB_MAD_RESULT_SUCCESS;
1590                 goto bail;
1591         default:
1592                 pmp->status |= IB_SMP_UNSUP_METHOD;
1593                 ret = reply((struct ib_smp *) pmp);
1594         }
1595
1596 bail:
1597         return ret;
1598 }
1599
1600 /**
1601  * ipath_process_mad - process an incoming MAD packet
1602  * @ibdev: the infiniband device this packet came in on
1603  * @mad_flags: MAD flags
1604  * @port_num: the port number this packet came in on
1605  * @in_wc: the work completion entry for this packet
1606  * @in_grh: the global route header for this packet
1607  * @in_mad: the incoming MAD
1608  * @out_mad: any outgoing MAD reply
1609  *
1610  * Returns IB_MAD_RESULT_SUCCESS if this is a MAD that we are not
1611  * interested in processing.
1612  *
1613  * Note that the verbs framework has already done the MAD sanity checks,
1614  * and hop count/pointer updating for IB_MGMT_CLASS_SUBN_DIRECTED_ROUTE
1615  * MADs.
1616  *
1617  * This is called by the ib_mad module.
1618  */
1619 int ipath_process_mad(struct ib_device *ibdev, int mad_flags, u8 port_num,
1620                       struct ib_wc *in_wc, struct ib_grh *in_grh,
1621                       struct ib_mad *in_mad, struct ib_mad *out_mad)
1622 {
1623         int ret;
1624
1625         switch (in_mad->mad_hdr.mgmt_class) {
1626         case IB_MGMT_CLASS_SUBN_DIRECTED_ROUTE:
1627         case IB_MGMT_CLASS_SUBN_LID_ROUTED:
1628                 ret = process_subn(ibdev, mad_flags, port_num,
1629                                    in_mad, out_mad);
1630                 goto bail;
1631         case IB_MGMT_CLASS_PERF_MGMT:
1632                 ret = process_perf(ibdev, port_num, in_mad, out_mad);
1633                 goto bail;
1634         default:
1635                 ret = IB_MAD_RESULT_SUCCESS;
1636         }
1637
1638 bail:
1639         return ret;
1640 }