Merge branch 'srp' of master.kernel.org:/pub/scm/linux/kernel/git/roland/infiniband
[linux-2.6] / arch / ia64 / sn / kernel / xpc_partition.c
1 /*
2  * This file is subject to the terms and conditions of the GNU General Public
3  * License.  See the file "COPYING" in the main directory of this archive
4  * for more details.
5  *
6  * Copyright (c) 2004-2005 Silicon Graphics, Inc.  All Rights Reserved.
7  */
8
9
10 /*
11  * Cross Partition Communication (XPC) partition support.
12  *
13  *      This is the part of XPC that detects the presence/absence of
14  *      other partitions. It provides a heartbeat and monitors the
15  *      heartbeats of other partitions.
16  *
17  */
18
19
20 #include <linux/kernel.h>
21 #include <linux/sysctl.h>
22 #include <linux/cache.h>
23 #include <linux/mmzone.h>
24 #include <linux/nodemask.h>
25 #include <asm/uncached.h>
26 #include <asm/sn/bte.h>
27 #include <asm/sn/intr.h>
28 #include <asm/sn/sn_sal.h>
29 #include <asm/sn/nodepda.h>
30 #include <asm/sn/addrs.h>
31 #include "xpc.h"
32
33
34 /* XPC is exiting flag */
35 int xpc_exiting;
36
37
38 /* SH_IPI_ACCESS shub register value on startup */
39 static u64 xpc_sh1_IPI_access;
40 static u64 xpc_sh2_IPI_access0;
41 static u64 xpc_sh2_IPI_access1;
42 static u64 xpc_sh2_IPI_access2;
43 static u64 xpc_sh2_IPI_access3;
44
45
46 /* original protection values for each node */
47 u64 xpc_prot_vec[MAX_NUMNODES];
48
49
50 /* this partition's reserved page pointers */
51 struct xpc_rsvd_page *xpc_rsvd_page;
52 static u64 *xpc_part_nasids;
53 static u64 *xpc_mach_nasids;
54 struct xpc_vars *xpc_vars;
55 struct xpc_vars_part *xpc_vars_part;
56
57 static int xp_nasid_mask_bytes; /* actual size in bytes of nasid mask */
58 static int xp_nasid_mask_words; /* actual size in words of nasid mask */
59
60
61 /*
62  * For performance reasons, each entry of xpc_partitions[] is cacheline
63  * aligned. And xpc_partitions[] is padded with an additional entry at the
64  * end so that the last legitimate entry doesn't share its cacheline with
65  * another variable.
66  */
67 struct xpc_partition xpc_partitions[XP_MAX_PARTITIONS + 1];
68
69
70 /*
71  * Generic buffer used to store a local copy of portions of a remote
72  * partition's reserved page (either its header and part_nasids mask,
73  * or its vars).
74  *
75  * xpc_discovery runs only once and is a seperate thread that is
76  * very likely going to be processing in parallel with receiving
77  * interrupts.
78  */
79 char ____cacheline_aligned xpc_remote_copy_buffer[XPC_RP_HEADER_SIZE +
80                                                         XP_NASID_MASK_BYTES];
81
82
83 /*
84  * Given a nasid, get the physical address of the  partition's reserved page
85  * for that nasid. This function returns 0 on any error.
86  */
87 static u64
88 xpc_get_rsvd_page_pa(int nasid)
89 {
90         bte_result_t bte_res;
91         s64 status;
92         u64 cookie = 0;
93         u64 rp_pa = nasid;      /* seed with nasid */
94         u64 len = 0;
95         u64 buf = buf;
96         u64 buf_len = 0;
97         void *buf_base = NULL;
98
99
100         while (1) {
101
102                 status = sn_partition_reserved_page_pa(buf, &cookie, &rp_pa,
103                                                                 &len);
104
105                 dev_dbg(xpc_part, "SAL returned with status=%li, cookie="
106                         "0x%016lx, address=0x%016lx, len=0x%016lx\n",
107                         status, cookie, rp_pa, len);
108
109                 if (status != SALRET_MORE_PASSES) {
110                         break;
111                 }
112
113                 if (L1_CACHE_ALIGN(len) > buf_len) {
114                         if (buf_base != NULL) {
115                                 kfree(buf_base);
116                         }
117                         buf_len = L1_CACHE_ALIGN(len);
118                         buf = (u64) xpc_kmalloc_cacheline_aligned(buf_len,
119                                                         GFP_KERNEL, &buf_base);
120                         if (buf_base == NULL) {
121                                 dev_err(xpc_part, "unable to kmalloc "
122                                         "len=0x%016lx\n", buf_len);
123                                 status = SALRET_ERROR;
124                                 break;
125                         }
126                 }
127
128                 bte_res = xp_bte_copy(rp_pa, ia64_tpa(buf), buf_len,
129                                         (BTE_NOTIFY | BTE_WACQUIRE), NULL);
130                 if (bte_res != BTE_SUCCESS) {
131                         dev_dbg(xpc_part, "xp_bte_copy failed %i\n", bte_res);
132                         status = SALRET_ERROR;
133                         break;
134                 }
135         }
136
137         if (buf_base != NULL) {
138                 kfree(buf_base);
139         }
140
141         if (status != SALRET_OK) {
142                 rp_pa = 0;
143         }
144         dev_dbg(xpc_part, "reserved page at phys address 0x%016lx\n", rp_pa);
145         return rp_pa;
146 }
147
148
149 /*
150  * Fill the partition reserved page with the information needed by
151  * other partitions to discover we are alive and establish initial
152  * communications.
153  */
154 struct xpc_rsvd_page *
155 xpc_rsvd_page_init(void)
156 {
157         struct xpc_rsvd_page *rp;
158         AMO_t *amos_page;
159         u64 rp_pa, nasid_array = 0;
160         int i, ret;
161
162
163         /* get the local reserved page's address */
164
165         preempt_disable();
166         rp_pa = xpc_get_rsvd_page_pa(cpuid_to_nasid(smp_processor_id()));
167         preempt_enable();
168         if (rp_pa == 0) {
169                 dev_err(xpc_part, "SAL failed to locate the reserved page\n");
170                 return NULL;
171         }
172         rp = (struct xpc_rsvd_page *) __va(rp_pa);
173
174         if (rp->partid != sn_partition_id) {
175                 dev_err(xpc_part, "the reserved page's partid of %d should be "
176                         "%d\n", rp->partid, sn_partition_id);
177                 return NULL;
178         }
179
180         rp->version = XPC_RP_VERSION;
181
182         /* establish the actual sizes of the nasid masks */
183         if (rp->SAL_version == 1) {
184                 /* SAL_version 1 didn't set the nasids_size field */
185                 rp->nasids_size = 128;
186         }
187         xp_nasid_mask_bytes = rp->nasids_size;
188         xp_nasid_mask_words = xp_nasid_mask_bytes / 8;
189
190         /* setup the pointers to the various items in the reserved page */
191         xpc_part_nasids = XPC_RP_PART_NASIDS(rp);
192         xpc_mach_nasids = XPC_RP_MACH_NASIDS(rp);
193         xpc_vars = XPC_RP_VARS(rp);
194         xpc_vars_part = XPC_RP_VARS_PART(rp);
195
196         /*
197          * Before clearing xpc_vars, see if a page of AMOs had been previously
198          * allocated. If not we'll need to allocate one and set permissions
199          * so that cross-partition AMOs are allowed.
200          *
201          * The allocated AMO page needs MCA reporting to remain disabled after
202          * XPC has unloaded.  To make this work, we keep a copy of the pointer
203          * to this page (i.e., amos_page) in the struct xpc_vars structure,
204          * which is pointed to by the reserved page, and re-use that saved copy
205          * on subsequent loads of XPC. This AMO page is never freed, and its
206          * memory protections are never restricted.
207          */
208         if ((amos_page = xpc_vars->amos_page) == NULL) {
209                 amos_page = (AMO_t *) TO_AMO(uncached_alloc_page(0));
210                 if (amos_page == NULL) {
211                         dev_err(xpc_part, "can't allocate page of AMOs\n");
212                         return NULL;
213                 }
214
215                 /*
216                  * Open up AMO-R/W to cpu.  This is done for Shub 1.1 systems
217                  * when xpc_allow_IPI_ops() is called via xpc_hb_init().
218                  */
219                 if (!enable_shub_wars_1_1()) {
220                         ret = sn_change_memprotect(ia64_tpa((u64) amos_page),
221                                         PAGE_SIZE, SN_MEMPROT_ACCESS_CLASS_1,
222                                         &nasid_array);
223                         if (ret != 0) {
224                                 dev_err(xpc_part, "can't change memory "
225                                         "protections\n");
226                                 uncached_free_page(__IA64_UNCACHED_OFFSET |
227                                                    TO_PHYS((u64) amos_page));
228                                 return NULL;
229                         }
230                 }
231         } else if (!IS_AMO_ADDRESS((u64) amos_page)) {
232                 /*
233                  * EFI's XPBOOT can also set amos_page in the reserved page,
234                  * but it happens to leave it as an uncached physical address
235                  * and we need it to be an uncached virtual, so we'll have to
236                  * convert it.
237                  */
238                 if (!IS_AMO_PHYS_ADDRESS((u64) amos_page)) {
239                         dev_err(xpc_part, "previously used amos_page address "
240                                 "is bad = 0x%p\n", (void *) amos_page);
241                         return NULL;
242                 }
243                 amos_page = (AMO_t *) TO_AMO((u64) amos_page);
244         }
245
246         /* clear xpc_vars */
247         memset(xpc_vars, 0, sizeof(struct xpc_vars));
248
249         xpc_vars->version = XPC_V_VERSION;
250         xpc_vars->act_nasid = cpuid_to_nasid(0);
251         xpc_vars->act_phys_cpuid = cpu_physical_id(0);
252         xpc_vars->vars_part_pa = __pa(xpc_vars_part);
253         xpc_vars->amos_page_pa = ia64_tpa((u64) amos_page);
254         xpc_vars->amos_page = amos_page;  /* save for next load of XPC */
255
256
257         /* clear xpc_vars_part */
258         memset((u64 *) xpc_vars_part, 0, sizeof(struct xpc_vars_part) *
259                                                         XP_MAX_PARTITIONS);
260
261         /* initialize the activate IRQ related AMO variables */
262         for (i = 0; i < xp_nasid_mask_words; i++) {
263                 (void) xpc_IPI_init(XPC_ACTIVATE_IRQ_AMOS + i);
264         }
265
266         /* initialize the engaged remote partitions related AMO variables */
267         (void) xpc_IPI_init(XPC_ENGAGED_PARTITIONS_AMO);
268         (void) xpc_IPI_init(XPC_DISENGAGE_REQUEST_AMO);
269
270         /* timestamp of when reserved page was setup by XPC */
271         rp->stamp = CURRENT_TIME;
272
273         /*
274          * This signifies to the remote partition that our reserved
275          * page is initialized.
276          */
277         rp->vars_pa = __pa(xpc_vars);
278
279         return rp;
280 }
281
282
283 /*
284  * Change protections to allow IPI operations (and AMO operations on
285  * Shub 1.1 systems).
286  */
287 void
288 xpc_allow_IPI_ops(void)
289 {
290         int node;
291         int nasid;
292
293
294         // >>> Change SH_IPI_ACCESS code to use SAL call once it is available.
295
296         if (is_shub2()) {
297                 xpc_sh2_IPI_access0 =
298                         (u64) HUB_L((u64 *) LOCAL_MMR_ADDR(SH2_IPI_ACCESS0));
299                 xpc_sh2_IPI_access1 =
300                         (u64) HUB_L((u64 *) LOCAL_MMR_ADDR(SH2_IPI_ACCESS1));
301                 xpc_sh2_IPI_access2 =
302                         (u64) HUB_L((u64 *) LOCAL_MMR_ADDR(SH2_IPI_ACCESS2));
303                 xpc_sh2_IPI_access3 =
304                         (u64) HUB_L((u64 *) LOCAL_MMR_ADDR(SH2_IPI_ACCESS3));
305
306                 for_each_online_node(node) {
307                         nasid = cnodeid_to_nasid(node);
308                         HUB_S((u64 *) GLOBAL_MMR_ADDR(nasid, SH2_IPI_ACCESS0),
309                                                                 -1UL);
310                         HUB_S((u64 *) GLOBAL_MMR_ADDR(nasid, SH2_IPI_ACCESS1),
311                                                                 -1UL);
312                         HUB_S((u64 *) GLOBAL_MMR_ADDR(nasid, SH2_IPI_ACCESS2),
313                                                                 -1UL);
314                         HUB_S((u64 *) GLOBAL_MMR_ADDR(nasid, SH2_IPI_ACCESS3),
315                                                                 -1UL);
316                 }
317
318         } else {
319                 xpc_sh1_IPI_access =
320                         (u64) HUB_L((u64 *) LOCAL_MMR_ADDR(SH1_IPI_ACCESS));
321
322                 for_each_online_node(node) {
323                         nasid = cnodeid_to_nasid(node);
324                         HUB_S((u64 *) GLOBAL_MMR_ADDR(nasid, SH1_IPI_ACCESS),
325                                                                 -1UL);
326
327                         /*
328                          * Since the BIST collides with memory operations on
329                          * SHUB 1.1 sn_change_memprotect() cannot be used.
330                          */
331                         if (enable_shub_wars_1_1()) {
332                                 /* open up everything */
333                                 xpc_prot_vec[node] = (u64) HUB_L((u64 *)
334                                                 GLOBAL_MMR_ADDR(nasid,
335                                                 SH1_MD_DQLP_MMR_DIR_PRIVEC0));
336                                 HUB_S((u64 *) GLOBAL_MMR_ADDR(nasid,
337                                                 SH1_MD_DQLP_MMR_DIR_PRIVEC0),
338                                                                 -1UL);
339                                 HUB_S((u64 *) GLOBAL_MMR_ADDR(nasid,
340                                                 SH1_MD_DQRP_MMR_DIR_PRIVEC0),
341                                                                 -1UL);
342                         }
343                 }
344         }
345 }
346
347
348 /*
349  * Restrict protections to disallow IPI operations (and AMO operations on
350  * Shub 1.1 systems).
351  */
352 void
353 xpc_restrict_IPI_ops(void)
354 {
355         int node;
356         int nasid;
357
358
359         // >>> Change SH_IPI_ACCESS code to use SAL call once it is available.
360
361         if (is_shub2()) {
362
363                 for_each_online_node(node) {
364                         nasid = cnodeid_to_nasid(node);
365                         HUB_S((u64 *) GLOBAL_MMR_ADDR(nasid, SH2_IPI_ACCESS0),
366                                                         xpc_sh2_IPI_access0);
367                         HUB_S((u64 *) GLOBAL_MMR_ADDR(nasid, SH2_IPI_ACCESS1),
368                                                         xpc_sh2_IPI_access1);
369                         HUB_S((u64 *) GLOBAL_MMR_ADDR(nasid, SH2_IPI_ACCESS2),
370                                                         xpc_sh2_IPI_access2);
371                         HUB_S((u64 *) GLOBAL_MMR_ADDR(nasid, SH2_IPI_ACCESS3),
372                                                         xpc_sh2_IPI_access3);
373                 }
374
375         } else {
376
377                 for_each_online_node(node) {
378                         nasid = cnodeid_to_nasid(node);
379                         HUB_S((u64 *) GLOBAL_MMR_ADDR(nasid, SH1_IPI_ACCESS),
380                                                         xpc_sh1_IPI_access);
381
382                         if (enable_shub_wars_1_1()) {
383                                 HUB_S((u64 *) GLOBAL_MMR_ADDR(nasid,
384                                                 SH1_MD_DQLP_MMR_DIR_PRIVEC0),
385                                                         xpc_prot_vec[node]);
386                                 HUB_S((u64 *) GLOBAL_MMR_ADDR(nasid,
387                                                 SH1_MD_DQRP_MMR_DIR_PRIVEC0),
388                                                         xpc_prot_vec[node]);
389                         }
390                 }
391         }
392 }
393
394
395 /*
396  * At periodic intervals, scan through all active partitions and ensure
397  * their heartbeat is still active.  If not, the partition is deactivated.
398  */
399 void
400 xpc_check_remote_hb(void)
401 {
402         struct xpc_vars *remote_vars;
403         struct xpc_partition *part;
404         partid_t partid;
405         bte_result_t bres;
406
407
408         remote_vars = (struct xpc_vars *) xpc_remote_copy_buffer;
409
410         for (partid = 1; partid < XP_MAX_PARTITIONS; partid++) {
411
412                 if (xpc_exiting) {
413                         break;
414                 }
415
416                 if (partid == sn_partition_id) {
417                         continue;
418                 }
419
420                 part = &xpc_partitions[partid];
421
422                 if (part->act_state == XPC_P_INACTIVE ||
423                                 part->act_state == XPC_P_DEACTIVATING) {
424                         continue;
425                 }
426
427                 /* pull the remote_hb cache line */
428                 bres = xp_bte_copy(part->remote_vars_pa,
429                                         ia64_tpa((u64) remote_vars),
430                                         XPC_RP_VARS_SIZE,
431                                         (BTE_NOTIFY | BTE_WACQUIRE), NULL);
432                 if (bres != BTE_SUCCESS) {
433                         XPC_DEACTIVATE_PARTITION(part,
434                                                 xpc_map_bte_errors(bres));
435                         continue;
436                 }
437
438                 dev_dbg(xpc_part, "partid = %d, heartbeat = %ld, last_heartbeat"
439                         " = %ld, kdb_status = %ld, HB_mask = 0x%lx\n", partid,
440                         remote_vars->heartbeat, part->last_heartbeat,
441                         remote_vars->kdb_status,
442                         remote_vars->heartbeating_to_mask);
443
444                 if (((remote_vars->heartbeat == part->last_heartbeat) &&
445                         (remote_vars->kdb_status == 0)) ||
446                              !xpc_hb_allowed(sn_partition_id, remote_vars)) {
447
448                         XPC_DEACTIVATE_PARTITION(part, xpcNoHeartbeat);
449                         continue;
450                 }
451
452                 part->last_heartbeat = remote_vars->heartbeat;
453         }
454 }
455
456
457 /*
458  * Get a copy of a portion of the remote partition's rsvd page.
459  *
460  * remote_rp points to a buffer that is cacheline aligned for BTE copies and
461  * is large enough to contain a copy of their reserved page header and
462  * part_nasids mask.
463  */
464 static enum xpc_retval
465 xpc_get_remote_rp(int nasid, u64 *discovered_nasids,
466                 struct xpc_rsvd_page *remote_rp, u64 *remote_rp_pa)
467 {
468         int bres, i;
469
470
471         /* get the reserved page's physical address */
472
473         *remote_rp_pa = xpc_get_rsvd_page_pa(nasid);
474         if (*remote_rp_pa == 0) {
475                 return xpcNoRsvdPageAddr;
476         }
477
478
479         /* pull over the reserved page header and part_nasids mask */
480
481         bres = xp_bte_copy(*remote_rp_pa, ia64_tpa((u64) remote_rp),
482                                 XPC_RP_HEADER_SIZE + xp_nasid_mask_bytes,
483                                 (BTE_NOTIFY | BTE_WACQUIRE), NULL);
484         if (bres != BTE_SUCCESS) {
485                 return xpc_map_bte_errors(bres);
486         }
487
488
489         if (discovered_nasids != NULL) {
490                 u64 *remote_part_nasids = XPC_RP_PART_NASIDS(remote_rp);
491
492
493                 for (i = 0; i < xp_nasid_mask_words; i++) {
494                         discovered_nasids[i] |= remote_part_nasids[i];
495                 }
496         }
497
498
499         /* check that the partid is for another partition */
500
501         if (remote_rp->partid < 1 ||
502                                 remote_rp->partid > (XP_MAX_PARTITIONS - 1)) {
503                 return xpcInvalidPartid;
504         }
505
506         if (remote_rp->partid == sn_partition_id) {
507                 return xpcLocalPartid;
508         }
509
510
511         if (XPC_VERSION_MAJOR(remote_rp->version) !=
512                                         XPC_VERSION_MAJOR(XPC_RP_VERSION)) {
513                 return xpcBadVersion;
514         }
515
516         return xpcSuccess;
517 }
518
519
520 /*
521  * Get a copy of the remote partition's XPC variables from the reserved page.
522  *
523  * remote_vars points to a buffer that is cacheline aligned for BTE copies and
524  * assumed to be of size XPC_RP_VARS_SIZE.
525  */
526 static enum xpc_retval
527 xpc_get_remote_vars(u64 remote_vars_pa, struct xpc_vars *remote_vars)
528 {
529         int bres;
530
531
532         if (remote_vars_pa == 0) {
533                 return xpcVarsNotSet;
534         }
535
536
537         /* pull over the cross partition variables */
538
539         bres = xp_bte_copy(remote_vars_pa, ia64_tpa((u64) remote_vars),
540                                 XPC_RP_VARS_SIZE,
541                                 (BTE_NOTIFY | BTE_WACQUIRE), NULL);
542         if (bres != BTE_SUCCESS) {
543                 return xpc_map_bte_errors(bres);
544         }
545
546         if (XPC_VERSION_MAJOR(remote_vars->version) !=
547                                         XPC_VERSION_MAJOR(XPC_V_VERSION)) {
548                 return xpcBadVersion;
549         }
550
551         return xpcSuccess;
552 }
553
554
555 /*
556  * Update the remote partition's info.
557  */
558 static void
559 xpc_update_partition_info(struct xpc_partition *part, u8 remote_rp_version,
560                 struct timespec *remote_rp_stamp, u64 remote_rp_pa,
561                 u64 remote_vars_pa, struct xpc_vars *remote_vars)
562 {
563         part->remote_rp_version = remote_rp_version;
564         dev_dbg(xpc_part, "  remote_rp_version = 0x%016lx\n",
565                 part->remote_rp_version);
566
567         part->remote_rp_stamp = *remote_rp_stamp;
568         dev_dbg(xpc_part, "  remote_rp_stamp (tv_sec = 0x%lx tv_nsec = 0x%lx\n",
569                 part->remote_rp_stamp.tv_sec, part->remote_rp_stamp.tv_nsec);
570
571         part->remote_rp_pa = remote_rp_pa;
572         dev_dbg(xpc_part, "  remote_rp_pa = 0x%016lx\n", part->remote_rp_pa);
573
574         part->remote_vars_pa = remote_vars_pa;
575         dev_dbg(xpc_part, "  remote_vars_pa = 0x%016lx\n",
576                 part->remote_vars_pa);
577
578         part->last_heartbeat = remote_vars->heartbeat;
579         dev_dbg(xpc_part, "  last_heartbeat = 0x%016lx\n",
580                 part->last_heartbeat);
581
582         part->remote_vars_part_pa = remote_vars->vars_part_pa;
583         dev_dbg(xpc_part, "  remote_vars_part_pa = 0x%016lx\n",
584                 part->remote_vars_part_pa);
585
586         part->remote_act_nasid = remote_vars->act_nasid;
587         dev_dbg(xpc_part, "  remote_act_nasid = 0x%x\n",
588                 part->remote_act_nasid);
589
590         part->remote_act_phys_cpuid = remote_vars->act_phys_cpuid;
591         dev_dbg(xpc_part, "  remote_act_phys_cpuid = 0x%x\n",
592                 part->remote_act_phys_cpuid);
593
594         part->remote_amos_page_pa = remote_vars->amos_page_pa;
595         dev_dbg(xpc_part, "  remote_amos_page_pa = 0x%lx\n",
596                 part->remote_amos_page_pa);
597
598         part->remote_vars_version = remote_vars->version;
599         dev_dbg(xpc_part, "  remote_vars_version = 0x%x\n",
600                 part->remote_vars_version);
601 }
602
603
604 /*
605  * Prior code has determined the nasid which generated an IPI.  Inspect
606  * that nasid to determine if its partition needs to be activated or
607  * deactivated.
608  *
609  * A partition is consider "awaiting activation" if our partition
610  * flags indicate it is not active and it has a heartbeat.  A
611  * partition is considered "awaiting deactivation" if our partition
612  * flags indicate it is active but it has no heartbeat or it is not
613  * sending its heartbeat to us.
614  *
615  * To determine the heartbeat, the remote nasid must have a properly
616  * initialized reserved page.
617  */
618 static void
619 xpc_identify_act_IRQ_req(int nasid)
620 {
621         struct xpc_rsvd_page *remote_rp;
622         struct xpc_vars *remote_vars;
623         u64 remote_rp_pa;
624         u64 remote_vars_pa;
625         int remote_rp_version;
626         int reactivate = 0;
627         int stamp_diff;
628         struct timespec remote_rp_stamp = { 0, 0 };
629         partid_t partid;
630         struct xpc_partition *part;
631         enum xpc_retval ret;
632
633
634         /* pull over the reserved page structure */
635
636         remote_rp = (struct xpc_rsvd_page *) xpc_remote_copy_buffer;
637
638         ret = xpc_get_remote_rp(nasid, NULL, remote_rp, &remote_rp_pa);
639         if (ret != xpcSuccess) {
640                 dev_warn(xpc_part, "unable to get reserved page from nasid %d, "
641                         "which sent interrupt, reason=%d\n", nasid, ret);
642                 return;
643         }
644
645         remote_vars_pa = remote_rp->vars_pa;
646         remote_rp_version = remote_rp->version;
647         if (XPC_SUPPORTS_RP_STAMP(remote_rp_version)) {
648                 remote_rp_stamp = remote_rp->stamp;
649         }
650         partid = remote_rp->partid;
651         part = &xpc_partitions[partid];
652
653
654         /* pull over the cross partition variables */
655
656         remote_vars = (struct xpc_vars *) xpc_remote_copy_buffer;
657
658         ret = xpc_get_remote_vars(remote_vars_pa, remote_vars);
659         if (ret != xpcSuccess) {
660
661                 dev_warn(xpc_part, "unable to get XPC variables from nasid %d, "
662                         "which sent interrupt, reason=%d\n", nasid, ret);
663
664                 XPC_DEACTIVATE_PARTITION(part, ret);
665                 return;
666         }
667
668
669         part->act_IRQ_rcvd++;
670
671         dev_dbg(xpc_part, "partid for nasid %d is %d; IRQs = %d; HB = "
672                 "%ld:0x%lx\n", (int) nasid, (int) partid, part->act_IRQ_rcvd,
673                 remote_vars->heartbeat, remote_vars->heartbeating_to_mask);
674
675         if (xpc_partition_disengaged(part) &&
676                                         part->act_state == XPC_P_INACTIVE) {
677
678                 xpc_update_partition_info(part, remote_rp_version,
679                                         &remote_rp_stamp, remote_rp_pa,
680                                         remote_vars_pa, remote_vars);
681
682                 if (XPC_SUPPORTS_DISENGAGE_REQUEST(part->remote_vars_version)) {
683                         if (xpc_partition_disengage_requested(1UL << partid)) {
684                                 /*
685                                  * Other side is waiting on us to disengage,
686                                  * even though we already have.
687                                  */
688                                 return;
689                         }
690                 } else {
691                         /* other side doesn't support disengage requests */
692                         xpc_clear_partition_disengage_request(1UL << partid);
693                 }
694
695                 xpc_activate_partition(part);
696                 return;
697         }
698
699         DBUG_ON(part->remote_rp_version == 0);
700         DBUG_ON(part->remote_vars_version == 0);
701
702         if (!XPC_SUPPORTS_RP_STAMP(part->remote_rp_version)) {
703                 DBUG_ON(XPC_SUPPORTS_DISENGAGE_REQUEST(part->
704                                                         remote_vars_version));
705
706                 if (!XPC_SUPPORTS_RP_STAMP(remote_rp_version)) {
707                         DBUG_ON(XPC_SUPPORTS_DISENGAGE_REQUEST(remote_vars->
708                                                                 version));
709                         /* see if the other side rebooted */
710                         if (part->remote_amos_page_pa ==
711                                 remote_vars->amos_page_pa &&
712                                         xpc_hb_allowed(sn_partition_id,
713                                                                 remote_vars)) {
714                                 /* doesn't look that way, so ignore the IPI */
715                                 return;
716                         }
717                 }
718
719                 /*
720                  * Other side rebooted and previous XPC didn't support the
721                  * disengage request, so we don't need to do anything special.
722                  */
723
724                 xpc_update_partition_info(part, remote_rp_version,
725                                                 &remote_rp_stamp, remote_rp_pa,
726                                                 remote_vars_pa, remote_vars);
727                 part->reactivate_nasid = nasid;
728                 XPC_DEACTIVATE_PARTITION(part, xpcReactivating);
729                 return;
730         }
731
732         DBUG_ON(!XPC_SUPPORTS_DISENGAGE_REQUEST(part->remote_vars_version));
733
734         if (!XPC_SUPPORTS_RP_STAMP(remote_rp_version)) {
735                 DBUG_ON(!XPC_SUPPORTS_DISENGAGE_REQUEST(remote_vars->version));
736
737                 /*
738                  * Other side rebooted and previous XPC did support the
739                  * disengage request, but the new one doesn't.
740                  */
741
742                 xpc_clear_partition_engaged(1UL << partid);
743                 xpc_clear_partition_disengage_request(1UL << partid);
744
745                 xpc_update_partition_info(part, remote_rp_version,
746                                                 &remote_rp_stamp, remote_rp_pa,
747                                                 remote_vars_pa, remote_vars);
748                 reactivate = 1;
749
750         } else {
751                 DBUG_ON(!XPC_SUPPORTS_DISENGAGE_REQUEST(remote_vars->version));
752
753                 stamp_diff = xpc_compare_stamps(&part->remote_rp_stamp,
754                                                         &remote_rp_stamp);
755                 if (stamp_diff != 0) {
756                         DBUG_ON(stamp_diff >= 0);
757
758                         /*
759                          * Other side rebooted and the previous XPC did support
760                          * the disengage request, as does the new one.
761                          */
762
763                         DBUG_ON(xpc_partition_engaged(1UL << partid));
764                         DBUG_ON(xpc_partition_disengage_requested(1UL <<
765                                                                 partid));
766
767                         xpc_update_partition_info(part, remote_rp_version,
768                                                 &remote_rp_stamp, remote_rp_pa,
769                                                 remote_vars_pa, remote_vars);
770                         reactivate = 1;
771                 }
772         }
773
774         if (!xpc_partition_disengaged(part)) {
775                 /* still waiting on other side to disengage from us */
776                 return;
777         }
778
779         if (reactivate) {
780                 part->reactivate_nasid = nasid;
781                 XPC_DEACTIVATE_PARTITION(part, xpcReactivating);
782
783         } else if (XPC_SUPPORTS_DISENGAGE_REQUEST(part->remote_vars_version) &&
784                         xpc_partition_disengage_requested(1UL << partid)) {
785                 XPC_DEACTIVATE_PARTITION(part, xpcOtherGoingDown);
786         }
787 }
788
789
790 /*
791  * Loop through the activation AMO variables and process any bits
792  * which are set.  Each bit indicates a nasid sending a partition
793  * activation or deactivation request.
794  *
795  * Return #of IRQs detected.
796  */
797 int
798 xpc_identify_act_IRQ_sender(void)
799 {
800         int word, bit;
801         u64 nasid_mask;
802         u64 nasid;                      /* remote nasid */
803         int n_IRQs_detected = 0;
804         AMO_t *act_amos;
805
806
807         act_amos = xpc_vars->amos_page + XPC_ACTIVATE_IRQ_AMOS;
808
809
810         /* scan through act AMO variable looking for non-zero entries */
811         for (word = 0; word < xp_nasid_mask_words; word++) {
812
813                 if (xpc_exiting) {
814                         break;
815                 }
816
817                 nasid_mask = xpc_IPI_receive(&act_amos[word]);
818                 if (nasid_mask == 0) {
819                         /* no IRQs from nasids in this variable */
820                         continue;
821                 }
822
823                 dev_dbg(xpc_part, "AMO[%d] gave back 0x%lx\n", word,
824                         nasid_mask);
825
826
827                 /*
828                  * If this nasid has been added to the machine since
829                  * our partition was reset, this will retain the
830                  * remote nasid in our reserved pages machine mask.
831                  * This is used in the event of module reload.
832                  */
833                 xpc_mach_nasids[word] |= nasid_mask;
834
835
836                 /* locate the nasid(s) which sent interrupts */
837
838                 for (bit = 0; bit < (8 * sizeof(u64)); bit++) {
839                         if (nasid_mask & (1UL << bit)) {
840                                 n_IRQs_detected++;
841                                 nasid = XPC_NASID_FROM_W_B(word, bit);
842                                 dev_dbg(xpc_part, "interrupt from nasid %ld\n",
843                                         nasid);
844                                 xpc_identify_act_IRQ_req(nasid);
845                         }
846                 }
847         }
848         return n_IRQs_detected;
849 }
850
851
852 /*
853  * See if the other side has responded to a partition disengage request
854  * from us.
855  */
856 int
857 xpc_partition_disengaged(struct xpc_partition *part)
858 {
859         partid_t partid = XPC_PARTID(part);
860         int disengaged;
861
862
863         disengaged = (xpc_partition_engaged(1UL << partid) == 0);
864         if (part->disengage_request_timeout) {
865                 if (!disengaged) {
866                         if (jiffies < part->disengage_request_timeout) {
867                                 /* timelimit hasn't been reached yet */
868                                 return 0;
869                         }
870
871                         /*
872                          * Other side hasn't responded to our disengage
873                          * request in a timely fashion, so assume it's dead.
874                          */
875
876                         xpc_clear_partition_engaged(1UL << partid);
877                         disengaged = 1;
878                 }
879                 part->disengage_request_timeout = 0;
880
881                 /* cancel the timer function, provided it's not us */
882                 if (!in_interrupt()) {
883                         del_singleshot_timer_sync(&part->
884                                                       disengage_request_timer);
885                 }
886
887                 DBUG_ON(part->act_state != XPC_P_DEACTIVATING &&
888                                         part->act_state != XPC_P_INACTIVE);
889                 if (part->act_state != XPC_P_INACTIVE) {
890                         xpc_wakeup_channel_mgr(part);
891                 }
892
893                 if (XPC_SUPPORTS_DISENGAGE_REQUEST(part->remote_vars_version)) {
894                         xpc_cancel_partition_disengage_request(part);
895                 }
896         }
897         return disengaged;
898 }
899
900
901 /*
902  * Mark specified partition as active.
903  */
904 enum xpc_retval
905 xpc_mark_partition_active(struct xpc_partition *part)
906 {
907         unsigned long irq_flags;
908         enum xpc_retval ret;
909
910
911         dev_dbg(xpc_part, "setting partition %d to ACTIVE\n", XPC_PARTID(part));
912
913         spin_lock_irqsave(&part->act_lock, irq_flags);
914         if (part->act_state == XPC_P_ACTIVATING) {
915                 part->act_state = XPC_P_ACTIVE;
916                 ret = xpcSuccess;
917         } else {
918                 DBUG_ON(part->reason == xpcSuccess);
919                 ret = part->reason;
920         }
921         spin_unlock_irqrestore(&part->act_lock, irq_flags);
922
923         return ret;
924 }
925
926
927 /*
928  * Notify XPC that the partition is down.
929  */
930 void
931 xpc_deactivate_partition(const int line, struct xpc_partition *part,
932                                 enum xpc_retval reason)
933 {
934         unsigned long irq_flags;
935
936
937         spin_lock_irqsave(&part->act_lock, irq_flags);
938
939         if (part->act_state == XPC_P_INACTIVE) {
940                 XPC_SET_REASON(part, reason, line);
941                 spin_unlock_irqrestore(&part->act_lock, irq_flags);
942                 if (reason == xpcReactivating) {
943                         /* we interrupt ourselves to reactivate partition */
944                         xpc_IPI_send_reactivate(part);
945                 }
946                 return;
947         }
948         if (part->act_state == XPC_P_DEACTIVATING) {
949                 if ((part->reason == xpcUnloading && reason != xpcUnloading) ||
950                                         reason == xpcReactivating) {
951                         XPC_SET_REASON(part, reason, line);
952                 }
953                 spin_unlock_irqrestore(&part->act_lock, irq_flags);
954                 return;
955         }
956
957         part->act_state = XPC_P_DEACTIVATING;
958         XPC_SET_REASON(part, reason, line);
959
960         spin_unlock_irqrestore(&part->act_lock, irq_flags);
961
962         if (XPC_SUPPORTS_DISENGAGE_REQUEST(part->remote_vars_version)) {
963                 xpc_request_partition_disengage(part);
964                 xpc_IPI_send_disengage(part);
965
966                 /* set a timelimit on the disengage request */
967                 part->disengage_request_timeout = jiffies +
968                                         (xpc_disengage_request_timelimit * HZ);
969                 part->disengage_request_timer.expires =
970                                         part->disengage_request_timeout;
971                 add_timer(&part->disengage_request_timer);
972         }
973
974         dev_dbg(xpc_part, "bringing partition %d down, reason = %d\n",
975                 XPC_PARTID(part), reason);
976
977         xpc_partition_going_down(part, reason);
978 }
979
980
981 /*
982  * Mark specified partition as inactive.
983  */
984 void
985 xpc_mark_partition_inactive(struct xpc_partition *part)
986 {
987         unsigned long irq_flags;
988
989
990         dev_dbg(xpc_part, "setting partition %d to INACTIVE\n",
991                 XPC_PARTID(part));
992
993         spin_lock_irqsave(&part->act_lock, irq_flags);
994         part->act_state = XPC_P_INACTIVE;
995         spin_unlock_irqrestore(&part->act_lock, irq_flags);
996         part->remote_rp_pa = 0;
997 }
998
999
1000 /*
1001  * SAL has provided a partition and machine mask.  The partition mask
1002  * contains a bit for each even nasid in our partition.  The machine
1003  * mask contains a bit for each even nasid in the entire machine.
1004  *
1005  * Using those two bit arrays, we can determine which nasids are
1006  * known in the machine.  Each should also have a reserved page
1007  * initialized if they are available for partitioning.
1008  */
1009 void
1010 xpc_discovery(void)
1011 {
1012         void *remote_rp_base;
1013         struct xpc_rsvd_page *remote_rp;
1014         struct xpc_vars *remote_vars;
1015         u64 remote_rp_pa;
1016         u64 remote_vars_pa;
1017         int region;
1018         int region_size;
1019         int max_regions;
1020         int nasid;
1021         struct xpc_rsvd_page *rp;
1022         partid_t partid;
1023         struct xpc_partition *part;
1024         u64 *discovered_nasids;
1025         enum xpc_retval ret;
1026
1027
1028         remote_rp = xpc_kmalloc_cacheline_aligned(XPC_RP_HEADER_SIZE +
1029                                                 xp_nasid_mask_bytes,
1030                                                 GFP_KERNEL, &remote_rp_base);
1031         if (remote_rp == NULL) {
1032                 return;
1033         }
1034         remote_vars = (struct xpc_vars *) remote_rp;
1035
1036
1037         discovered_nasids = kmalloc(sizeof(u64) * xp_nasid_mask_words,
1038                                                         GFP_KERNEL);
1039         if (discovered_nasids == NULL) {
1040                 kfree(remote_rp_base);
1041                 return;
1042         }
1043         memset(discovered_nasids, 0, sizeof(u64) * xp_nasid_mask_words);
1044
1045         rp = (struct xpc_rsvd_page *) xpc_rsvd_page;
1046
1047         /*
1048          * The term 'region' in this context refers to the minimum number of
1049          * nodes that can comprise an access protection grouping. The access
1050          * protection is in regards to memory, IOI and IPI.
1051          */
1052         max_regions = 64;
1053         region_size = sn_region_size;
1054
1055         switch (region_size) {
1056         case 128:
1057                 max_regions *= 2;
1058         case 64:
1059                 max_regions *= 2;
1060         case 32:
1061                 max_regions *= 2;
1062                 region_size = 16;
1063                 DBUG_ON(!is_shub2());
1064         }
1065
1066         for (region = 0; region < max_regions; region++) {
1067
1068                 if ((volatile int) xpc_exiting) {
1069                         break;
1070                 }
1071
1072                 dev_dbg(xpc_part, "searching region %d\n", region);
1073
1074                 for (nasid = (region * region_size * 2);
1075                      nasid < ((region + 1) * region_size * 2);
1076                      nasid += 2) {
1077
1078                         if ((volatile int) xpc_exiting) {
1079                                 break;
1080                         }
1081
1082                         dev_dbg(xpc_part, "checking nasid %d\n", nasid);
1083
1084
1085                         if (XPC_NASID_IN_ARRAY(nasid, xpc_part_nasids)) {
1086                                 dev_dbg(xpc_part, "PROM indicates Nasid %d is "
1087                                         "part of the local partition; skipping "
1088                                         "region\n", nasid);
1089                                 break;
1090                         }
1091
1092                         if (!(XPC_NASID_IN_ARRAY(nasid, xpc_mach_nasids))) {
1093                                 dev_dbg(xpc_part, "PROM indicates Nasid %d was "
1094                                         "not on Numa-Link network at reset\n",
1095                                         nasid);
1096                                 continue;
1097                         }
1098
1099                         if (XPC_NASID_IN_ARRAY(nasid, discovered_nasids)) {
1100                                 dev_dbg(xpc_part, "Nasid %d is part of a "
1101                                         "partition which was previously "
1102                                         "discovered\n", nasid);
1103                                 continue;
1104                         }
1105
1106
1107                         /* pull over the reserved page structure */
1108
1109                         ret = xpc_get_remote_rp(nasid, discovered_nasids,
1110                                               remote_rp, &remote_rp_pa);
1111                         if (ret != xpcSuccess) {
1112                                 dev_dbg(xpc_part, "unable to get reserved page "
1113                                         "from nasid %d, reason=%d\n", nasid,
1114                                         ret);
1115
1116                                 if (ret == xpcLocalPartid) {
1117                                         break;
1118                                 }
1119                                 continue;
1120                         }
1121
1122                         remote_vars_pa = remote_rp->vars_pa;
1123
1124                         partid = remote_rp->partid;
1125                         part = &xpc_partitions[partid];
1126
1127
1128                         /* pull over the cross partition variables */
1129
1130                         ret = xpc_get_remote_vars(remote_vars_pa, remote_vars);
1131                         if (ret != xpcSuccess) {
1132                                 dev_dbg(xpc_part, "unable to get XPC variables "
1133                                         "from nasid %d, reason=%d\n", nasid,
1134                                         ret);
1135
1136                                 XPC_DEACTIVATE_PARTITION(part, ret);
1137                                 continue;
1138                         }
1139
1140                         if (part->act_state != XPC_P_INACTIVE) {
1141                                 dev_dbg(xpc_part, "partition %d on nasid %d is "
1142                                         "already activating\n", partid, nasid);
1143                                 break;
1144                         }
1145
1146                         /*
1147                          * Register the remote partition's AMOs with SAL so it
1148                          * can handle and cleanup errors within that address
1149                          * range should the remote partition go down. We don't
1150                          * unregister this range because it is difficult to
1151                          * tell when outstanding writes to the remote partition
1152                          * are finished and thus when it is thus safe to
1153                          * unregister. This should not result in wasted space
1154                          * in the SAL xp_addr_region table because we should
1155                          * get the same page for remote_act_amos_pa after
1156                          * module reloads and system reboots.
1157                          */
1158                         if (sn_register_xp_addr_region(
1159                                             remote_vars->amos_page_pa,
1160                                                         PAGE_SIZE, 1) < 0) {
1161                                 dev_dbg(xpc_part, "partition %d failed to "
1162                                         "register xp_addr region 0x%016lx\n",
1163                                         partid, remote_vars->amos_page_pa);
1164
1165                                 XPC_SET_REASON(part, xpcPhysAddrRegFailed,
1166                                                 __LINE__);
1167                                 break;
1168                         }
1169
1170                         /*
1171                          * The remote nasid is valid and available.
1172                          * Send an interrupt to that nasid to notify
1173                          * it that we are ready to begin activation.
1174                          */
1175                         dev_dbg(xpc_part, "sending an interrupt to AMO 0x%lx, "
1176                                 "nasid %d, phys_cpuid 0x%x\n",
1177                                 remote_vars->amos_page_pa,
1178                                 remote_vars->act_nasid,
1179                                 remote_vars->act_phys_cpuid);
1180
1181                         if (XPC_SUPPORTS_DISENGAGE_REQUEST(remote_vars->
1182                                                                 version)) {
1183                                 part->remote_amos_page_pa =
1184                                                 remote_vars->amos_page_pa;
1185                                 xpc_mark_partition_disengaged(part);
1186                                 xpc_cancel_partition_disengage_request(part);
1187                         }
1188                         xpc_IPI_send_activate(remote_vars);
1189                 }
1190         }
1191
1192         kfree(discovered_nasids);
1193         kfree(remote_rp_base);
1194 }
1195
1196
1197 /*
1198  * Given a partid, get the nasids owned by that partition from the
1199  * remote partition's reserved page.
1200  */
1201 enum xpc_retval
1202 xpc_initiate_partid_to_nasids(partid_t partid, void *nasid_mask)
1203 {
1204         struct xpc_partition *part;
1205         u64 part_nasid_pa;
1206         int bte_res;
1207
1208
1209         part = &xpc_partitions[partid];
1210         if (part->remote_rp_pa == 0) {
1211                 return xpcPartitionDown;
1212         }
1213
1214         memset(nasid_mask, 0, XP_NASID_MASK_BYTES);
1215
1216         part_nasid_pa = (u64) XPC_RP_PART_NASIDS(part->remote_rp_pa);
1217
1218         bte_res = xp_bte_copy(part_nasid_pa, ia64_tpa((u64) nasid_mask),
1219                         xp_nasid_mask_bytes, (BTE_NOTIFY | BTE_WACQUIRE), NULL);
1220
1221         return xpc_map_bte_errors(bte_res);
1222 }
1223